diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000000000000000000000000000000000000..e43b0f988953ae3a84b00331d0ccf5f7d51cb3cf --- /dev/null +++ b/.gitignore @@ -0,0 +1 @@ +.DS_Store diff --git a/LICENSE b/LICENSE deleted file mode 100644 index e69de29bb2d1d6434b8b29ae775ad8c2e48c5391..0000000000000000000000000000000000000000 diff --git a/LICENSE_NOTICE.txt b/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/README.md b/README.md index 0bbe02ac6f493881ad316a54eba35ee8ceebbfa9..237fa4aba683451ab0e91ac140ff630b0226934c 100644 --- a/README.md +++ b/README.md @@ -1,5 +1,24 @@ ---- -license: other -license_name: argmax-fmod-license -license_link: LICENSE ---- +--- +license: other +license_name: argmax-fmod-license +license_link: https://huggingface.co/argmaxinc/whisperkit-pro/blob/main/LICENSE_NOTICE.txt +pretty_name: "WhisperKit" +viewer: false +library_name: whisperkit +tags: +- whisper +- whisperkit +- coreml +- asr +- quantized +- automatic-speech-recognition +extra_gated_heading: "WhisperKit Pro is now in early access!" +extra_gated_description: "WhisperKit Pro is the commercial tier of [WhisperKit](https://github.com/argmaxinc/WhisperKit). Please submit your information below to request early access or directly send an email to [earlyaccess@argmaxinc.com](mailto:earlyaccess@argmaxinc.com)" +extra_gated_fields: + Company: text + Work email: text + I acknowledge the license notice: checkbox +extra_gated_button_content: "Submit" +--- + +# WhisperKit Pro \ No newline at end of file diff --git a/config.json b/config.json new file mode 100644 index 0000000000000000000000000000000000000000..1e85966a4ad04d743dcc03ef8813959b4acb1604 --- /dev/null +++ b/config.json @@ -0,0 +1,112 @@ +{ + "name": "whisperkit-coreml", + "version": "0.2", + "device_support": [ + { + "identifiers": ["iPhone11", "iPhone12", "Watch7", "Watch8"], + "models": { + "default": "openai_whisper-tiny", + "supported": [ + "openai_whisper-tiny", + "openai_whisper-tiny.en", + "openai_whisper-base", + "openai_whisper-base.en" + ] + } + }, + { + "identifiers": ["iPhone13", "iPad13,18", "iPad13,1"], + "models": { + "default": "openai_whisper-base", + "supported": [ + "openai_whisper-tiny", + "openai_whisper-tiny.en", + "openai_whisper-base", + "openai_whisper-base.en", + "openai_whisper-small", + "openai_whisper-small.en" + ] + } + }, + { + "identifiers": [ + "iPhone14", + "iPhone15", + "iPhone16", + "iPhone17", + "iPad14,1", + "iPad14,2" + ], + "models": { + "default": "openai_whisper-base", + "supported": [ + "openai_whisper-tiny", + "openai_whisper-tiny.en", + "openai_whisper-base", + "openai_whisper-base.en", + "openai_whisper-small", + "openai_whisper-small.en", + ] + } + }, + { + "identifiers": [ + "Mac13", + "iMac21", + "MacBookAir10,1", + "MacBookPro17", + "MacBookPro18", + "Macmini9", + "iPad13,16", + "iPad13,4", + "iPad13,8" + ], + "models": { + "default": "openai_whisper-large-v3-v20240930", + "supported": [ + "openai_whisper-tiny", + "openai_whisper-tiny.en", + "openai_whisper-base", + "openai_whisper-base.en", + "openai_whisper-small", + "openai_whisper-small.en", + "distil-whisper_distil-large-v3", + "openai_whisper-large-v3-v20240930", + ] + } + }, + { + "identifiers": [ + "Mac14", + "Mac15", + "Mac16", + "iPad14,3", + "iPad14,4", + "iPad14,5", + "iPad14,6", + "iPad14,8", + "iPad14,9", + "iPad14,10", + "iPad14,11", + "iPad16" + ], + "models": { + "default": "openai_whisper-large-v3-v20240930", + "supported": [ + "openai_whisper-tiny", + "openai_whisper-tiny.en", + "openai_whisper-base", + "openai_whisper-base.en", + "openai_whisper-small", + "openai_whisper-small.en", + "distil-whisper_distil-large-v3", + "distil-whisper_distil-large-v3_turbo", + "openai_whisper-large-v3-v20240930", + "openai_whisper-large-v3-v20240930_turbo", + ] + } + } + ], + "model_checksums": { + } +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/analytics/coremldata.bin b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..54e0b944cc74967f8a343f43956b5809aa828c3c --- /dev/null +++ b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1e272b852871322d9a085dfa95707110046e749f525a0db0d8f277709fcde3f +size 243 diff --git a/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/coremldata.bin b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..5754f6415d17f592683fbed877c4faa41698d605 --- /dev/null +++ b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49b36ed70ce2688383176bec4bd74a1a7716b505ddffd95d5f8eda4063ce6ae3 +size 434 diff --git a/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/metadata.json b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..ed880cd91f3496915cdc9e84d38ce6102bff51e2 --- /dev/null +++ b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,90 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1280 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 1280, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.reshape" : 128, + "Ios18.batchNorm" : 65, + "Ios18.softmax" : 32, + "Pad" : 2, + "Ios18.concat" : 2, + "Ios18.gelu" : 34, + "Ios18.layerNorm" : 65, + "Ios18.matmul" : 64, + "Ios18.conv" : 198, + "Ios18.mul" : 32, + "Ios18.add" : 65 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.version" : "8.0", + "com.github.apple.coremltools.source" : "torch==2.5.1" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/model.mil b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..762acbe0cd4a56482973460db42d052dd9e90fdc --- /dev/null +++ b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,2737 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_110_pad_type_0 = const()[name = string("op_110_pad_type_0"), val = string("custom")]; + tensor var_110_pad_0 = const()[name = string("op_110_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_110_strides_0 = const()[name = string("op_110_strides_0"), val = tensor([1, 1])]; + tensor var_110_dilations_0 = const()[name = string("op_110_dilations_0"), val = tensor([1, 1])]; + int32 var_110_groups_0 = const()[name = string("op_110_groups_0"), val = int32(1)]; + tensor var_85_to_fp16 = const()[name = string("op_85_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_91_to_fp16 = const()[name = string("op_91_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(983168)))]; + tensor var_110_cast_fp16 = conv(bias = var_91_to_fp16, dilations = var_110_dilations_0, groups = var_110_groups_0, pad = var_110_pad_0, pad_type = var_110_pad_type_0, strides = var_110_strides_0, weight = var_85_to_fp16, x = melspectrogram_features)[name = string("op_110_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_110_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_150_pad_type_0 = const()[name = string("op_150_pad_type_0"), val = string("custom")]; + tensor var_150_pad_0 = const()[name = string("op_150_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_150_strides_0 = const()[name = string("op_150_strides_0"), val = tensor([2, 2])]; + tensor var_150_dilations_0 = const()[name = string("op_150_dilations_0"), val = tensor([1, 1])]; + int32 var_150_groups_0 = const()[name = string("op_150_groups_0"), val = int32(1)]; + tensor var_125_to_fp16 = const()[name = string("op_125_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985792)))]; + tensor var_131_to_fp16 = const()[name = string("op_131_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10816256)))]; + tensor var_150_cast_fp16 = conv(bias = var_131_to_fp16, dilations = var_150_dilations_0, groups = var_150_groups_0, pad = var_150_pad_0, pad_type = var_150_pad_type_0, strides = var_150_strides_0, weight = var_125_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_150_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_150_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_168_to_fp16 = const()[name = string("op_168_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_168_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_182 = const()[name = string("op_182"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_201_to_fp16 = const()[name = string("op_201_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_201_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14658944)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14661568)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14664192)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14666816)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14669440)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17946304)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17948928)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21225792)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24502656)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_236 = const()[name = string("op_236"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_236, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_238_to_fp16 = const()[name = string("op_238_to_fp16"), val = fp16(0x1p-3)]; + tensor var_239_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_238_to_fp16)[name = string("op_239_cast_fp16")]; + tensor var_240 = const()[name = string("op_240"), val = tensor([1, 20, 64, -1])]; + tensor var_241_cast_fp16 = reshape(shape = var_240, x = key_1_cast_fp16)[name = string("op_241_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_239_cast_fp16, y = var_241_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_244_cast_fp16 = softmax(axis = var_182, x = mh_w_1_cast_fp16)[name = string("op_244_cast_fp16")]; + tensor var_245 = const()[name = string("op_245"), val = tensor([1, 20, 64, -1])]; + tensor var_246_cast_fp16 = reshape(shape = var_245, x = value_1_cast_fp16)[name = string("op_246_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_246_cast_fp16, y = var_244_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_249 = const()[name = string("op_249"), val = tensor([1, 1280, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_249, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24505280)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27782144)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_267_to_fp16 = const()[name = string("op_267_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_267_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27784768)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27787392)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27790016)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40897280)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40907584)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54014848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_300 = const()[name = string("op_300"), val = int32(3)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_319_to_fp16 = const()[name = string("op_319_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_319_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54017472)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54020096)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54022720)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57299584)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57302208)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60579072)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63855936)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_354 = const()[name = string("op_354"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_354, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_356_to_fp16 = const()[name = string("op_356_to_fp16"), val = fp16(0x1p-3)]; + tensor var_357_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_356_to_fp16)[name = string("op_357_cast_fp16")]; + tensor var_358 = const()[name = string("op_358"), val = tensor([1, 20, 64, -1])]; + tensor var_359_cast_fp16 = reshape(shape = var_358, x = key_3_cast_fp16)[name = string("op_359_cast_fp16")]; + bool mh_w_3_transpose_x_0 = const()[name = string("mh_w_3_transpose_x_0"), val = bool(true)]; + bool mh_w_3_transpose_y_0 = const()[name = string("mh_w_3_transpose_y_0"), val = bool(false)]; + tensor mh_w_3_cast_fp16 = matmul(transpose_x = mh_w_3_transpose_x_0, transpose_y = mh_w_3_transpose_y_0, x = var_357_cast_fp16, y = var_359_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_362_cast_fp16 = softmax(axis = var_300, x = mh_w_3_cast_fp16)[name = string("op_362_cast_fp16")]; + tensor var_363 = const()[name = string("op_363"), val = tensor([1, 20, 64, -1])]; + tensor var_364_cast_fp16 = reshape(shape = var_363, x = value_3_cast_fp16)[name = string("op_364_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_364_cast_fp16, y = var_362_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([1, 1280, 1, -1])]; + tensor input_9_cast_fp16 = reshape(shape = var_367, x = attn_3_cast_fp16)[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63858560)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67135424)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_385_to_fp16 = const()[name = string("op_385_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_385_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67138048)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67140672)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67143296)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80250560)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80260864)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93368128)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_418 = const()[name = string("op_418"), val = int32(3)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_437_to_fp16 = const()[name = string("op_437_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_437_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93370752)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93373376)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93376000)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96652864)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96655488)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99932352)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103209216)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_472 = const()[name = string("op_472"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_472, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_474_to_fp16 = const()[name = string("op_474_to_fp16"), val = fp16(0x1p-3)]; + tensor var_475_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_474_to_fp16)[name = string("op_475_cast_fp16")]; + tensor var_476 = const()[name = string("op_476"), val = tensor([1, 20, 64, -1])]; + tensor var_477_cast_fp16 = reshape(shape = var_476, x = key_5_cast_fp16)[name = string("op_477_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_475_cast_fp16, y = var_477_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor var_480_cast_fp16 = softmax(axis = var_418, x = mh_w_5_cast_fp16)[name = string("op_480_cast_fp16")]; + tensor var_481 = const()[name = string("op_481"), val = tensor([1, 20, 64, -1])]; + tensor var_482_cast_fp16 = reshape(shape = var_481, x = value_5_cast_fp16)[name = string("op_482_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_482_cast_fp16, y = var_480_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_485 = const()[name = string("op_485"), val = tensor([1, 1280, 1, -1])]; + tensor input_17_cast_fp16 = reshape(shape = var_485, x = attn_5_cast_fp16)[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103211840)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106488704)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_503_to_fp16 = const()[name = string("op_503_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_503_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106491328)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106493952)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106496576)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119603840)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119614144)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132721408)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_536 = const()[name = string("op_536"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_555_to_fp16 = const()[name = string("op_555_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_555_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132724032)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132726656)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132729280)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136006144)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136008768)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139285632)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142562496)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_590 = const()[name = string("op_590"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_590, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_592_to_fp16 = const()[name = string("op_592_to_fp16"), val = fp16(0x1p-3)]; + tensor var_593_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_592_to_fp16)[name = string("op_593_cast_fp16")]; + tensor var_594 = const()[name = string("op_594"), val = tensor([1, 20, 64, -1])]; + tensor var_595_cast_fp16 = reshape(shape = var_594, x = key_7_cast_fp16)[name = string("op_595_cast_fp16")]; + bool mh_w_7_transpose_x_0 = const()[name = string("mh_w_7_transpose_x_0"), val = bool(true)]; + bool mh_w_7_transpose_y_0 = const()[name = string("mh_w_7_transpose_y_0"), val = bool(false)]; + tensor mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_593_cast_fp16, y = var_595_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor var_598_cast_fp16 = softmax(axis = var_536, x = mh_w_7_cast_fp16)[name = string("op_598_cast_fp16")]; + tensor var_599 = const()[name = string("op_599"), val = tensor([1, 20, 64, -1])]; + tensor var_600_cast_fp16 = reshape(shape = var_599, x = value_7_cast_fp16)[name = string("op_600_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_600_cast_fp16, y = var_598_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_603 = const()[name = string("op_603"), val = tensor([1, 1280, 1, -1])]; + tensor input_25_cast_fp16 = reshape(shape = var_603, x = attn_7_cast_fp16)[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142565120)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145841984)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_621_to_fp16 = const()[name = string("op_621_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_621_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145844608)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145847232)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145849856)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158957120)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158967424)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172074688)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_654 = const()[name = string("op_654"), val = int32(3)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_673_to_fp16 = const()[name = string("op_673_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_673_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172077312)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172079936)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172082560)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175359424)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175362048)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178638912)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181915776)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_708 = const()[name = string("op_708"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_708, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_710_to_fp16 = const()[name = string("op_710_to_fp16"), val = fp16(0x1p-3)]; + tensor var_711_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_710_to_fp16)[name = string("op_711_cast_fp16")]; + tensor var_712 = const()[name = string("op_712"), val = tensor([1, 20, 64, -1])]; + tensor var_713_cast_fp16 = reshape(shape = var_712, x = key_9_cast_fp16)[name = string("op_713_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_711_cast_fp16, y = var_713_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor var_716_cast_fp16 = softmax(axis = var_654, x = mh_w_9_cast_fp16)[name = string("op_716_cast_fp16")]; + tensor var_717 = const()[name = string("op_717"), val = tensor([1, 20, 64, -1])]; + tensor var_718_cast_fp16 = reshape(shape = var_717, x = value_9_cast_fp16)[name = string("op_718_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_718_cast_fp16, y = var_716_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_721 = const()[name = string("op_721"), val = tensor([1, 1280, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_721, x = attn_9_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181918400)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185195264)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_739_to_fp16 = const()[name = string("op_739_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_739_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185197888)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185200512)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185203136)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198310400)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198320704)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211427968)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_772 = const()[name = string("op_772"), val = int32(3)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_791_to_fp16 = const()[name = string("op_791_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_791_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211430592)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211433216)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211435840)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214712704)))]; + tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; + string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; + tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; + tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; + int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214715328)))]; + tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; + string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; + tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; + tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; + int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217992192)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221269056)))]; + tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_826 = const()[name = string("op_826"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_826, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_828_to_fp16 = const()[name = string("op_828_to_fp16"), val = fp16(0x1p-3)]; + tensor var_829_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_828_to_fp16)[name = string("op_829_cast_fp16")]; + tensor var_830 = const()[name = string("op_830"), val = tensor([1, 20, 64, -1])]; + tensor var_831_cast_fp16 = reshape(shape = var_830, x = key_11_cast_fp16)[name = string("op_831_cast_fp16")]; + bool mh_w_11_transpose_x_0 = const()[name = string("mh_w_11_transpose_x_0"), val = bool(true)]; + bool mh_w_11_transpose_y_0 = const()[name = string("mh_w_11_transpose_y_0"), val = bool(false)]; + tensor mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_829_cast_fp16, y = var_831_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_834_cast_fp16 = softmax(axis = var_772, x = mh_w_11_cast_fp16)[name = string("op_834_cast_fp16")]; + tensor var_835 = const()[name = string("op_835"), val = tensor([1, 20, 64, -1])]; + tensor var_836_cast_fp16 = reshape(shape = var_835, x = value_11_cast_fp16)[name = string("op_836_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_836_cast_fp16, y = var_834_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 1280, 1, -1])]; + tensor input_41_cast_fp16 = reshape(shape = var_839, x = attn_11_cast_fp16)[name = string("input_41_cast_fp16")]; + string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; + tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; + tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; + int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221271680)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224548544)))]; + tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_857_to_fp16 = const()[name = string("op_857_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_857_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224551168)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224553792)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224556416)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237663680)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237673984)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250781248)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; + int32 var_890 = const()[name = string("op_890"), val = int32(3)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_909_to_fp16 = const()[name = string("op_909_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_909_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250783872)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250786496)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250789120)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254065984)))]; + tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; + string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; + tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; + tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; + int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254068608)))]; + tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; + string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; + tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; + tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; + int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257345472)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260622336)))]; + tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_944 = const()[name = string("op_944"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_944, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_946_to_fp16 = const()[name = string("op_946_to_fp16"), val = fp16(0x1p-3)]; + tensor var_947_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_946_to_fp16)[name = string("op_947_cast_fp16")]; + tensor var_948 = const()[name = string("op_948"), val = tensor([1, 20, 64, -1])]; + tensor var_949_cast_fp16 = reshape(shape = var_948, x = key_13_cast_fp16)[name = string("op_949_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_947_cast_fp16, y = var_949_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor var_952_cast_fp16 = softmax(axis = var_890, x = mh_w_13_cast_fp16)[name = string("op_952_cast_fp16")]; + tensor var_953 = const()[name = string("op_953"), val = tensor([1, 20, 64, -1])]; + tensor var_954_cast_fp16 = reshape(shape = var_953, x = value_13_cast_fp16)[name = string("op_954_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_954_cast_fp16, y = var_952_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_957 = const()[name = string("op_957"), val = tensor([1, 1280, 1, -1])]; + tensor input_49_cast_fp16 = reshape(shape = var_957, x = attn_13_cast_fp16)[name = string("input_49_cast_fp16")]; + string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; + tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; + tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; + int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260624960)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263901824)))]; + tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_975_to_fp16 = const()[name = string("op_975_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_975_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263904448)))]; + tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263907072)))]; + fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; + string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; + tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; + tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; + int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263909696)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277016960)))]; + tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; + string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; + tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277027264)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290134528)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; + int32 var_1008 = const()[name = string("op_1008"), val = int32(3)]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_1027_to_fp16 = const()[name = string("op_1027_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1027_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290137152)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290139776)))]; + fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290142400)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293419264)))]; + tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; + string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; + tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; + tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; + int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293421888)))]; + tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; + string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; + tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; + tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; + int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296698752)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299975616)))]; + tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_1062 = const()[name = string("op_1062"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_1062, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; + fp16 var_1064_to_fp16 = const()[name = string("op_1064_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1065_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_1064_to_fp16)[name = string("op_1065_cast_fp16")]; + tensor var_1066 = const()[name = string("op_1066"), val = tensor([1, 20, 64, -1])]; + tensor var_1067_cast_fp16 = reshape(shape = var_1066, x = key_15_cast_fp16)[name = string("op_1067_cast_fp16")]; + bool mh_w_15_transpose_x_0 = const()[name = string("mh_w_15_transpose_x_0"), val = bool(true)]; + bool mh_w_15_transpose_y_0 = const()[name = string("mh_w_15_transpose_y_0"), val = bool(false)]; + tensor mh_w_15_cast_fp16 = matmul(transpose_x = mh_w_15_transpose_x_0, transpose_y = mh_w_15_transpose_y_0, x = var_1065_cast_fp16, y = var_1067_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor var_1070_cast_fp16 = softmax(axis = var_1008, x = mh_w_15_cast_fp16)[name = string("op_1070_cast_fp16")]; + tensor var_1071 = const()[name = string("op_1071"), val = tensor([1, 20, 64, -1])]; + tensor var_1072_cast_fp16 = reshape(shape = var_1071, x = value_15_cast_fp16)[name = string("op_1072_cast_fp16")]; + bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; + bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_1072_cast_fp16, y = var_1070_cast_fp16)[name = string("attn_15_cast_fp16")]; + tensor var_1075 = const()[name = string("op_1075"), val = tensor([1, 1280, 1, -1])]; + tensor input_57_cast_fp16 = reshape(shape = var_1075, x = attn_15_cast_fp16)[name = string("input_57_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299978240)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303255104)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_1093_to_fp16 = const()[name = string("op_1093_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1093_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303257728)))]; + tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303260352)))]; + fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303262976)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316370240)))]; + tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; + tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316380544)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329487808)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; + int32 var_1126 = const()[name = string("op_1126"), val = int32(3)]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_1145_to_fp16 = const()[name = string("op_1145_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1145_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329490432)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329493056)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329495680)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332772544)))]; + tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; + string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; + tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; + tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; + int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332775168)))]; + tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; + string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; + tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; + tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; + int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336052032)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339328896)))]; + tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_1180 = const()[name = string("op_1180"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_1180, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; + fp16 var_1182_to_fp16 = const()[name = string("op_1182_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1183_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1182_to_fp16)[name = string("op_1183_cast_fp16")]; + tensor var_1184 = const()[name = string("op_1184"), val = tensor([1, 20, 64, -1])]; + tensor var_1185_cast_fp16 = reshape(shape = var_1184, x = key_17_cast_fp16)[name = string("op_1185_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1183_cast_fp16, y = var_1185_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor var_1188_cast_fp16 = softmax(axis = var_1126, x = mh_w_17_cast_fp16)[name = string("op_1188_cast_fp16")]; + tensor var_1189 = const()[name = string("op_1189"), val = tensor([1, 20, 64, -1])]; + tensor var_1190_cast_fp16 = reshape(shape = var_1189, x = value_17_cast_fp16)[name = string("op_1190_cast_fp16")]; + bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; + bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1190_cast_fp16, y = var_1188_cast_fp16)[name = string("attn_17_cast_fp16")]; + tensor var_1193 = const()[name = string("op_1193"), val = tensor([1, 1280, 1, -1])]; + tensor input_65_cast_fp16 = reshape(shape = var_1193, x = attn_17_cast_fp16)[name = string("input_65_cast_fp16")]; + string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; + tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; + tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; + int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339331520)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342608384)))]; + tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_1211_to_fp16 = const()[name = string("op_1211_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1211_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342611008)))]; + tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342613632)))]; + fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; + tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; + tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; + int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342616256)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355723520)))]; + tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; + tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355733824)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368841088)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; + int32 var_1244 = const()[name = string("op_1244"), val = int32(3)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_1263_to_fp16 = const()[name = string("op_1263_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1263_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368843712)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368846336)))]; + fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368848960)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372125824)))]; + tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; + string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; + tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; + tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; + int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372128448)))]; + tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; + string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; + tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; + tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; + int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375405312)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378682176)))]; + tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_1298 = const()[name = string("op_1298"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_1298, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; + fp16 var_1300_to_fp16 = const()[name = string("op_1300_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1301_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1300_to_fp16)[name = string("op_1301_cast_fp16")]; + tensor var_1302 = const()[name = string("op_1302"), val = tensor([1, 20, 64, -1])]; + tensor var_1303_cast_fp16 = reshape(shape = var_1302, x = key_19_cast_fp16)[name = string("op_1303_cast_fp16")]; + bool mh_w_19_transpose_x_0 = const()[name = string("mh_w_19_transpose_x_0"), val = bool(true)]; + bool mh_w_19_transpose_y_0 = const()[name = string("mh_w_19_transpose_y_0"), val = bool(false)]; + tensor mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_1301_cast_fp16, y = var_1303_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_1306_cast_fp16 = softmax(axis = var_1244, x = mh_w_19_cast_fp16)[name = string("op_1306_cast_fp16")]; + tensor var_1307 = const()[name = string("op_1307"), val = tensor([1, 20, 64, -1])]; + tensor var_1308_cast_fp16 = reshape(shape = var_1307, x = value_19_cast_fp16)[name = string("op_1308_cast_fp16")]; + bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; + bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1308_cast_fp16, y = var_1306_cast_fp16)[name = string("attn_19_cast_fp16")]; + tensor var_1311 = const()[name = string("op_1311"), val = tensor([1, 1280, 1, -1])]; + tensor input_73_cast_fp16 = reshape(shape = var_1311, x = attn_19_cast_fp16)[name = string("input_73_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378684800)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381961664)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_1329_to_fp16 = const()[name = string("op_1329_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1329_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381964288)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381966912)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381969536)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395076800)))]; + tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395087104)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408194368)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; + int32 var_1362 = const()[name = string("op_1362"), val = int32(3)]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_1381_to_fp16 = const()[name = string("op_1381_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1381_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408196992)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408199616)))]; + fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408202240)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411479104)))]; + tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; + string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; + tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; + tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; + int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411481728)))]; + tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; + string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; + tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; + tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; + int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414758592)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418035456)))]; + tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_1416 = const()[name = string("op_1416"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_1416, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; + fp16 var_1418_to_fp16 = const()[name = string("op_1418_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1419_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1418_to_fp16)[name = string("op_1419_cast_fp16")]; + tensor var_1420 = const()[name = string("op_1420"), val = tensor([1, 20, 64, -1])]; + tensor var_1421_cast_fp16 = reshape(shape = var_1420, x = key_21_cast_fp16)[name = string("op_1421_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_1419_cast_fp16, y = var_1421_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor var_1424_cast_fp16 = softmax(axis = var_1362, x = mh_w_21_cast_fp16)[name = string("op_1424_cast_fp16")]; + tensor var_1425 = const()[name = string("op_1425"), val = tensor([1, 20, 64, -1])]; + tensor var_1426_cast_fp16 = reshape(shape = var_1425, x = value_21_cast_fp16)[name = string("op_1426_cast_fp16")]; + bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; + bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1426_cast_fp16, y = var_1424_cast_fp16)[name = string("attn_21_cast_fp16")]; + tensor var_1429 = const()[name = string("op_1429"), val = tensor([1, 1280, 1, -1])]; + tensor input_81_cast_fp16 = reshape(shape = var_1429, x = attn_21_cast_fp16)[name = string("input_81_cast_fp16")]; + string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; + tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; + tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; + int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418038080)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421314944)))]; + tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_1447_to_fp16 = const()[name = string("op_1447_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1447_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421317568)))]; + tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421320192)))]; + fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; + string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; + tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; + tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; + int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421322816)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434430080)))]; + tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; + tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434440384)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447547648)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; + int32 var_1480 = const()[name = string("op_1480"), val = int32(3)]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_1499_to_fp16 = const()[name = string("op_1499_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1499_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447550272)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447552896)))]; + fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; + string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; + tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; + tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; + int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447555520)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450832384)))]; + tensor query_23_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_23_cast_fp16")]; + string key_23_pad_type_0 = const()[name = string("key_23_pad_type_0"), val = string("valid")]; + tensor key_23_strides_0 = const()[name = string("key_23_strides_0"), val = tensor([1, 1])]; + tensor key_23_pad_0 = const()[name = string("key_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_23_dilations_0 = const()[name = string("key_23_dilations_0"), val = tensor([1, 1])]; + int32 key_23_groups_0 = const()[name = string("key_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450835008)))]; + tensor key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_23_cast_fp16")]; + string value_23_pad_type_0 = const()[name = string("value_23_pad_type_0"), val = string("valid")]; + tensor value_23_strides_0 = const()[name = string("value_23_strides_0"), val = tensor([1, 1])]; + tensor value_23_pad_0 = const()[name = string("value_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_23_dilations_0 = const()[name = string("value_23_dilations_0"), val = tensor([1, 1])]; + int32 value_23_groups_0 = const()[name = string("value_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454111872)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457388736)))]; + tensor value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_23_cast_fp16")]; + tensor var_1534 = const()[name = string("op_1534"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_23_cast_fp16 = reshape(shape = var_1534, x = query_23_cast_fp16)[name = string("mh_q_23_cast_fp16")]; + fp16 var_1536_to_fp16 = const()[name = string("op_1536_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1537_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1536_to_fp16)[name = string("op_1537_cast_fp16")]; + tensor var_1538 = const()[name = string("op_1538"), val = tensor([1, 20, 64, -1])]; + tensor var_1539_cast_fp16 = reshape(shape = var_1538, x = key_23_cast_fp16)[name = string("op_1539_cast_fp16")]; + bool mh_w_23_transpose_x_0 = const()[name = string("mh_w_23_transpose_x_0"), val = bool(true)]; + bool mh_w_23_transpose_y_0 = const()[name = string("mh_w_23_transpose_y_0"), val = bool(false)]; + tensor mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_1537_cast_fp16, y = var_1539_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor var_1542_cast_fp16 = softmax(axis = var_1480, x = mh_w_23_cast_fp16)[name = string("op_1542_cast_fp16")]; + tensor var_1543 = const()[name = string("op_1543"), val = tensor([1, 20, 64, -1])]; + tensor var_1544_cast_fp16 = reshape(shape = var_1543, x = value_23_cast_fp16)[name = string("op_1544_cast_fp16")]; + bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)]; + bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)]; + tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1544_cast_fp16, y = var_1542_cast_fp16)[name = string("attn_23_cast_fp16")]; + tensor var_1547 = const()[name = string("op_1547"), val = tensor([1, 1280, 1, -1])]; + tensor input_89_cast_fp16 = reshape(shape = var_1547, x = attn_23_cast_fp16)[name = string("input_89_cast_fp16")]; + string obj_47_pad_type_0 = const()[name = string("obj_47_pad_type_0"), val = string("valid")]; + tensor obj_47_strides_0 = const()[name = string("obj_47_strides_0"), val = tensor([1, 1])]; + tensor obj_47_pad_0 = const()[name = string("obj_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_47_dilations_0 = const()[name = string("obj_47_dilations_0"), val = tensor([1, 1])]; + int32 obj_47_groups_0 = const()[name = string("obj_47_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457391360)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460668224)))]; + tensor obj_47_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_47_dilations_0, groups = obj_47_groups_0, pad = obj_47_pad_0, pad_type = obj_47_pad_type_0, strides = obj_47_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_47_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_47_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_1565_to_fp16 = const()[name = string("op_1565_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1565_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460670848)))]; + tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460673472)))]; + fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460676096)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473783360)))]; + tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; + tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + string hidden_states_27_pad_type_0 = const()[name = string("hidden_states_27_pad_type_0"), val = string("valid")]; + tensor hidden_states_27_strides_0 = const()[name = string("hidden_states_27_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_27_pad_0 = const()[name = string("hidden_states_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_27_dilations_0 = const()[name = string("hidden_states_27_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_27_groups_0 = const()[name = string("hidden_states_27_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473793664)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486900928)))]; + tensor hidden_states_27_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_27_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_27_cast_fp16)[name = string("inputs_49_cast_fp16")]; + int32 var_1598 = const()[name = string("op_1598"), val = int32(3)]; + tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; + fp16 var_1617_to_fp16 = const()[name = string("op_1617_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1617_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; + tensor obj_49_gamma_0_to_fp16 = const()[name = string("obj_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486903552)))]; + tensor obj_49_beta_0_to_fp16 = const()[name = string("obj_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486906176)))]; + fp16 obj_49_epsilon_0_to_fp16 = const()[name = string("obj_49_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_49_cast_fp16 = batch_norm(beta = obj_49_beta_0_to_fp16, epsilon = obj_49_epsilon_0_to_fp16, gamma = obj_49_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_49_cast_fp16")]; + string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; + tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; + tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; + int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486908800)))]; + tensor layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490185664)))]; + tensor query_25_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("query_25_cast_fp16")]; + string key_25_pad_type_0 = const()[name = string("key_25_pad_type_0"), val = string("valid")]; + tensor key_25_strides_0 = const()[name = string("key_25_strides_0"), val = tensor([1, 1])]; + tensor key_25_pad_0 = const()[name = string("key_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_25_dilations_0 = const()[name = string("key_25_dilations_0"), val = tensor([1, 1])]; + int32 key_25_groups_0 = const()[name = string("key_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490188288)))]; + tensor key_25_cast_fp16 = conv(dilations = key_25_dilations_0, groups = key_25_groups_0, pad = key_25_pad_0, pad_type = key_25_pad_type_0, strides = key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("key_25_cast_fp16")]; + string value_25_pad_type_0 = const()[name = string("value_25_pad_type_0"), val = string("valid")]; + tensor value_25_strides_0 = const()[name = string("value_25_strides_0"), val = tensor([1, 1])]; + tensor value_25_pad_0 = const()[name = string("value_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_25_dilations_0 = const()[name = string("value_25_dilations_0"), val = tensor([1, 1])]; + int32 value_25_groups_0 = const()[name = string("value_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(493465152)))]; + tensor layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496742016)))]; + tensor value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = value_25_dilations_0, groups = value_25_groups_0, pad = value_25_pad_0, pad_type = value_25_pad_type_0, strides = value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("value_25_cast_fp16")]; + tensor var_1652 = const()[name = string("op_1652"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_25_cast_fp16 = reshape(shape = var_1652, x = query_25_cast_fp16)[name = string("mh_q_25_cast_fp16")]; + fp16 var_1654_to_fp16 = const()[name = string("op_1654_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1655_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1654_to_fp16)[name = string("op_1655_cast_fp16")]; + tensor var_1656 = const()[name = string("op_1656"), val = tensor([1, 20, 64, -1])]; + tensor var_1657_cast_fp16 = reshape(shape = var_1656, x = key_25_cast_fp16)[name = string("op_1657_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1655_cast_fp16, y = var_1657_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor var_1660_cast_fp16 = softmax(axis = var_1598, x = mh_w_25_cast_fp16)[name = string("op_1660_cast_fp16")]; + tensor var_1661 = const()[name = string("op_1661"), val = tensor([1, 20, 64, -1])]; + tensor var_1662_cast_fp16 = reshape(shape = var_1661, x = value_25_cast_fp16)[name = string("op_1662_cast_fp16")]; + bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)]; + bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)]; + tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1662_cast_fp16, y = var_1660_cast_fp16)[name = string("attn_25_cast_fp16")]; + tensor var_1665 = const()[name = string("op_1665"), val = tensor([1, 1280, 1, -1])]; + tensor input_97_cast_fp16 = reshape(shape = var_1665, x = attn_25_cast_fp16)[name = string("input_97_cast_fp16")]; + string obj_51_pad_type_0 = const()[name = string("obj_51_pad_type_0"), val = string("valid")]; + tensor obj_51_strides_0 = const()[name = string("obj_51_strides_0"), val = tensor([1, 1])]; + tensor obj_51_pad_0 = const()[name = string("obj_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_51_dilations_0 = const()[name = string("obj_51_dilations_0"), val = tensor([1, 1])]; + int32 obj_51_groups_0 = const()[name = string("obj_51_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496744640)))]; + tensor layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500021504)))]; + tensor obj_51_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_51_dilations_0, groups = obj_51_groups_0, pad = obj_51_pad_0, pad_type = obj_51_pad_type_0, strides = obj_51_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("obj_51_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_51_cast_fp16)[name = string("inputs_51_cast_fp16")]; + tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; + fp16 var_1683_to_fp16 = const()[name = string("op_1683_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1683_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; + tensor input_99_gamma_0_to_fp16 = const()[name = string("input_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500024128)))]; + tensor input_99_beta_0_to_fp16 = const()[name = string("input_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500026752)))]; + fp16 input_99_epsilon_0_to_fp16 = const()[name = string("input_99_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_99_cast_fp16 = batch_norm(beta = input_99_beta_0_to_fp16, epsilon = input_99_epsilon_0_to_fp16, gamma = input_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("input_99_cast_fp16")]; + string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")]; + tensor input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor([1, 1])]; + tensor input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor([1, 1])]; + int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1)]; + tensor layers_12_fc1_weight_to_fp16 = const()[name = string("layers_12_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500029376)))]; + tensor layers_12_fc1_bias_to_fp16 = const()[name = string("layers_12_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513136640)))]; + tensor input_101_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = layers_12_fc1_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")]; + string input_103_mode_0 = const()[name = string("input_103_mode_0"), val = string("EXACT")]; + tensor input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")]; + string hidden_states_29_pad_type_0 = const()[name = string("hidden_states_29_pad_type_0"), val = string("valid")]; + tensor hidden_states_29_strides_0 = const()[name = string("hidden_states_29_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_29_pad_0 = const()[name = string("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_29_dilations_0 = const()[name = string("hidden_states_29_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_29_groups_0 = const()[name = string("hidden_states_29_groups_0"), val = int32(1)]; + tensor layers_12_fc2_weight_to_fp16 = const()[name = string("layers_12_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513146944)))]; + tensor layers_12_fc2_bias_to_fp16 = const()[name = string("layers_12_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526254208)))]; + tensor hidden_states_29_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_12_fc2_weight_to_fp16, x = input_103_cast_fp16)[name = string("hidden_states_29_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = hidden_states_29_cast_fp16)[name = string("inputs_53_cast_fp16")]; + int32 var_1716 = const()[name = string("op_1716"), val = int32(3)]; + tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; + fp16 var_1735_to_fp16 = const()[name = string("op_1735_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_1735_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; + tensor obj_53_gamma_0_to_fp16 = const()[name = string("obj_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526256832)))]; + tensor obj_53_beta_0_to_fp16 = const()[name = string("obj_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526259456)))]; + fp16 obj_53_epsilon_0_to_fp16 = const()[name = string("obj_53_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_53_cast_fp16 = batch_norm(beta = obj_53_beta_0_to_fp16, epsilon = obj_53_epsilon_0_to_fp16, gamma = obj_53_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("obj_53_cast_fp16")]; + string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; + tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; + tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; + int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526262080)))]; + tensor layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529538944)))]; + tensor query_27_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("query_27_cast_fp16")]; + string key_27_pad_type_0 = const()[name = string("key_27_pad_type_0"), val = string("valid")]; + tensor key_27_strides_0 = const()[name = string("key_27_strides_0"), val = tensor([1, 1])]; + tensor key_27_pad_0 = const()[name = string("key_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_27_dilations_0 = const()[name = string("key_27_dilations_0"), val = tensor([1, 1])]; + int32 key_27_groups_0 = const()[name = string("key_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529541568)))]; + tensor key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("key_27_cast_fp16")]; + string value_27_pad_type_0 = const()[name = string("value_27_pad_type_0"), val = string("valid")]; + tensor value_27_strides_0 = const()[name = string("value_27_strides_0"), val = tensor([1, 1])]; + tensor value_27_pad_0 = const()[name = string("value_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_27_dilations_0 = const()[name = string("value_27_dilations_0"), val = tensor([1, 1])]; + int32 value_27_groups_0 = const()[name = string("value_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(532818432)))]; + tensor layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536095296)))]; + tensor value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("value_27_cast_fp16")]; + tensor var_1770 = const()[name = string("op_1770"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_27_cast_fp16 = reshape(shape = var_1770, x = query_27_cast_fp16)[name = string("mh_q_27_cast_fp16")]; + fp16 var_1772_to_fp16 = const()[name = string("op_1772_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1773_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1772_to_fp16)[name = string("op_1773_cast_fp16")]; + tensor var_1774 = const()[name = string("op_1774"), val = tensor([1, 20, 64, -1])]; + tensor var_1775_cast_fp16 = reshape(shape = var_1774, x = key_27_cast_fp16)[name = string("op_1775_cast_fp16")]; + bool mh_w_27_transpose_x_0 = const()[name = string("mh_w_27_transpose_x_0"), val = bool(true)]; + bool mh_w_27_transpose_y_0 = const()[name = string("mh_w_27_transpose_y_0"), val = bool(false)]; + tensor mh_w_27_cast_fp16 = matmul(transpose_x = mh_w_27_transpose_x_0, transpose_y = mh_w_27_transpose_y_0, x = var_1773_cast_fp16, y = var_1775_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_1778_cast_fp16 = softmax(axis = var_1716, x = mh_w_27_cast_fp16)[name = string("op_1778_cast_fp16")]; + tensor var_1779 = const()[name = string("op_1779"), val = tensor([1, 20, 64, -1])]; + tensor var_1780_cast_fp16 = reshape(shape = var_1779, x = value_27_cast_fp16)[name = string("op_1780_cast_fp16")]; + bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)]; + bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)]; + tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1780_cast_fp16, y = var_1778_cast_fp16)[name = string("attn_27_cast_fp16")]; + tensor var_1783 = const()[name = string("op_1783"), val = tensor([1, 1280, 1, -1])]; + tensor input_105_cast_fp16 = reshape(shape = var_1783, x = attn_27_cast_fp16)[name = string("input_105_cast_fp16")]; + string obj_55_pad_type_0 = const()[name = string("obj_55_pad_type_0"), val = string("valid")]; + tensor obj_55_strides_0 = const()[name = string("obj_55_strides_0"), val = tensor([1, 1])]; + tensor obj_55_pad_0 = const()[name = string("obj_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_55_dilations_0 = const()[name = string("obj_55_dilations_0"), val = tensor([1, 1])]; + int32 obj_55_groups_0 = const()[name = string("obj_55_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536097920)))]; + tensor layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539374784)))]; + tensor obj_55_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_55_dilations_0, groups = obj_55_groups_0, pad = obj_55_pad_0, pad_type = obj_55_pad_type_0, strides = obj_55_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("obj_55_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_55_cast_fp16)[name = string("inputs_55_cast_fp16")]; + tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; + fp16 var_1801_to_fp16 = const()[name = string("op_1801_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_1801_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; + tensor input_107_gamma_0_to_fp16 = const()[name = string("input_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539377408)))]; + tensor input_107_beta_0_to_fp16 = const()[name = string("input_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539380032)))]; + fp16 input_107_epsilon_0_to_fp16 = const()[name = string("input_107_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_107_cast_fp16 = batch_norm(beta = input_107_beta_0_to_fp16, epsilon = input_107_epsilon_0_to_fp16, gamma = input_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("input_107_cast_fp16")]; + string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("valid")]; + tensor input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor([1, 1])]; + tensor input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor([1, 1])]; + int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1)]; + tensor layers_13_fc1_weight_to_fp16 = const()[name = string("layers_13_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539382656)))]; + tensor layers_13_fc1_bias_to_fp16 = const()[name = string("layers_13_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552489920)))]; + tensor input_109_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = layers_13_fc1_weight_to_fp16, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")]; + string input_111_mode_0 = const()[name = string("input_111_mode_0"), val = string("EXACT")]; + tensor input_111_cast_fp16 = gelu(mode = input_111_mode_0, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")]; + string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")]; + tensor hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)]; + tensor layers_13_fc2_weight_to_fp16 = const()[name = string("layers_13_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552500224)))]; + tensor layers_13_fc2_bias_to_fp16 = const()[name = string("layers_13_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565607488)))]; + tensor hidden_states_31_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_13_fc2_weight_to_fp16, x = input_111_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = hidden_states_31_cast_fp16)[name = string("inputs_57_cast_fp16")]; + int32 var_1834 = const()[name = string("op_1834"), val = int32(3)]; + tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; + fp16 var_1853_to_fp16 = const()[name = string("op_1853_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_1853_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; + tensor obj_57_gamma_0_to_fp16 = const()[name = string("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565610112)))]; + tensor obj_57_beta_0_to_fp16 = const()[name = string("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565612736)))]; + fp16 obj_57_epsilon_0_to_fp16 = const()[name = string("obj_57_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_57_cast_fp16")]; + string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; + tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; + tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; + int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565615360)))]; + tensor layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568892224)))]; + tensor query_29_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("query_29_cast_fp16")]; + string key_29_pad_type_0 = const()[name = string("key_29_pad_type_0"), val = string("valid")]; + tensor key_29_strides_0 = const()[name = string("key_29_strides_0"), val = tensor([1, 1])]; + tensor key_29_pad_0 = const()[name = string("key_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_29_dilations_0 = const()[name = string("key_29_dilations_0"), val = tensor([1, 1])]; + int32 key_29_groups_0 = const()[name = string("key_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568894848)))]; + tensor key_29_cast_fp16 = conv(dilations = key_29_dilations_0, groups = key_29_groups_0, pad = key_29_pad_0, pad_type = key_29_pad_type_0, strides = key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("key_29_cast_fp16")]; + string value_29_pad_type_0 = const()[name = string("value_29_pad_type_0"), val = string("valid")]; + tensor value_29_strides_0 = const()[name = string("value_29_strides_0"), val = tensor([1, 1])]; + tensor value_29_pad_0 = const()[name = string("value_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_29_dilations_0 = const()[name = string("value_29_dilations_0"), val = tensor([1, 1])]; + int32 value_29_groups_0 = const()[name = string("value_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572171712)))]; + tensor layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575448576)))]; + tensor value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = value_29_dilations_0, groups = value_29_groups_0, pad = value_29_pad_0, pad_type = value_29_pad_type_0, strides = value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("value_29_cast_fp16")]; + tensor var_1888 = const()[name = string("op_1888"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_29_cast_fp16 = reshape(shape = var_1888, x = query_29_cast_fp16)[name = string("mh_q_29_cast_fp16")]; + fp16 var_1890_to_fp16 = const()[name = string("op_1890_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1891_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1890_to_fp16)[name = string("op_1891_cast_fp16")]; + tensor var_1892 = const()[name = string("op_1892"), val = tensor([1, 20, 64, -1])]; + tensor var_1893_cast_fp16 = reshape(shape = var_1892, x = key_29_cast_fp16)[name = string("op_1893_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1891_cast_fp16, y = var_1893_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor var_1896_cast_fp16 = softmax(axis = var_1834, x = mh_w_29_cast_fp16)[name = string("op_1896_cast_fp16")]; + tensor var_1897 = const()[name = string("op_1897"), val = tensor([1, 20, 64, -1])]; + tensor var_1898_cast_fp16 = reshape(shape = var_1897, x = value_29_cast_fp16)[name = string("op_1898_cast_fp16")]; + bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)]; + bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)]; + tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1898_cast_fp16, y = var_1896_cast_fp16)[name = string("attn_29_cast_fp16")]; + tensor var_1901 = const()[name = string("op_1901"), val = tensor([1, 1280, 1, -1])]; + tensor input_113_cast_fp16 = reshape(shape = var_1901, x = attn_29_cast_fp16)[name = string("input_113_cast_fp16")]; + string obj_59_pad_type_0 = const()[name = string("obj_59_pad_type_0"), val = string("valid")]; + tensor obj_59_strides_0 = const()[name = string("obj_59_strides_0"), val = tensor([1, 1])]; + tensor obj_59_pad_0 = const()[name = string("obj_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_59_dilations_0 = const()[name = string("obj_59_dilations_0"), val = tensor([1, 1])]; + int32 obj_59_groups_0 = const()[name = string("obj_59_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575451200)))]; + tensor layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578728064)))]; + tensor obj_59_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_59_dilations_0, groups = obj_59_groups_0, pad = obj_59_pad_0, pad_type = obj_59_pad_type_0, strides = obj_59_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_59_cast_fp16)[name = string("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; + fp16 var_1919_to_fp16 = const()[name = string("op_1919_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_1919_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578730688)))]; + tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578733312)))]; + fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor layers_14_fc1_weight_to_fp16 = const()[name = string("layers_14_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578735936)))]; + tensor layers_14_fc1_bias_to_fp16 = const()[name = string("layers_14_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591843200)))]; + tensor input_117_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_14_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")]; + tensor input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")]; + string hidden_states_33_pad_type_0 = const()[name = string("hidden_states_33_pad_type_0"), val = string("valid")]; + tensor hidden_states_33_strides_0 = const()[name = string("hidden_states_33_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_33_pad_0 = const()[name = string("hidden_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_33_dilations_0 = const()[name = string("hidden_states_33_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_33_groups_0 = const()[name = string("hidden_states_33_groups_0"), val = int32(1)]; + tensor layers_14_fc2_weight_to_fp16 = const()[name = string("layers_14_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591853504)))]; + tensor layers_14_fc2_bias_to_fp16 = const()[name = string("layers_14_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604960768)))]; + tensor hidden_states_33_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_14_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = string("hidden_states_33_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_33_cast_fp16)[name = string("inputs_61_cast_fp16")]; + int32 var_1952 = const()[name = string("op_1952"), val = int32(3)]; + tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; + fp16 var_1971_to_fp16 = const()[name = string("op_1971_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_1971_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604963392)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604966016)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; + tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; + tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; + int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604968640)))]; + tensor layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608245504)))]; + tensor query_31_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_31_cast_fp16")]; + string key_31_pad_type_0 = const()[name = string("key_31_pad_type_0"), val = string("valid")]; + tensor key_31_strides_0 = const()[name = string("key_31_strides_0"), val = tensor([1, 1])]; + tensor key_31_pad_0 = const()[name = string("key_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_31_dilations_0 = const()[name = string("key_31_dilations_0"), val = tensor([1, 1])]; + int32 key_31_groups_0 = const()[name = string("key_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608248128)))]; + tensor key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("key_31_cast_fp16")]; + string value_31_pad_type_0 = const()[name = string("value_31_pad_type_0"), val = string("valid")]; + tensor value_31_strides_0 = const()[name = string("value_31_strides_0"), val = tensor([1, 1])]; + tensor value_31_pad_0 = const()[name = string("value_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_31_dilations_0 = const()[name = string("value_31_dilations_0"), val = tensor([1, 1])]; + int32 value_31_groups_0 = const()[name = string("value_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(611524992)))]; + tensor layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614801856)))]; + tensor value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("value_31_cast_fp16")]; + tensor var_2006 = const()[name = string("op_2006"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_31_cast_fp16 = reshape(shape = var_2006, x = query_31_cast_fp16)[name = string("mh_q_31_cast_fp16")]; + fp16 var_2008_to_fp16 = const()[name = string("op_2008_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2009_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_2008_to_fp16)[name = string("op_2009_cast_fp16")]; + tensor var_2010 = const()[name = string("op_2010"), val = tensor([1, 20, 64, -1])]; + tensor var_2011_cast_fp16 = reshape(shape = var_2010, x = key_31_cast_fp16)[name = string("op_2011_cast_fp16")]; + bool mh_w_31_transpose_x_0 = const()[name = string("mh_w_31_transpose_x_0"), val = bool(true)]; + bool mh_w_31_transpose_y_0 = const()[name = string("mh_w_31_transpose_y_0"), val = bool(false)]; + tensor mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_2009_cast_fp16, y = var_2011_cast_fp16)[name = string("mh_w_31_cast_fp16")]; + tensor var_2014_cast_fp16 = softmax(axis = var_1952, x = mh_w_31_cast_fp16)[name = string("op_2014_cast_fp16")]; + tensor var_2015 = const()[name = string("op_2015"), val = tensor([1, 20, 64, -1])]; + tensor var_2016_cast_fp16 = reshape(shape = var_2015, x = value_31_cast_fp16)[name = string("op_2016_cast_fp16")]; + bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)]; + bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)]; + tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_2016_cast_fp16, y = var_2014_cast_fp16)[name = string("attn_31_cast_fp16")]; + tensor var_2019 = const()[name = string("op_2019"), val = tensor([1, 1280, 1, -1])]; + tensor input_121_cast_fp16 = reshape(shape = var_2019, x = attn_31_cast_fp16)[name = string("input_121_cast_fp16")]; + string obj_63_pad_type_0 = const()[name = string("obj_63_pad_type_0"), val = string("valid")]; + tensor obj_63_strides_0 = const()[name = string("obj_63_strides_0"), val = tensor([1, 1])]; + tensor obj_63_pad_0 = const()[name = string("obj_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_63_dilations_0 = const()[name = string("obj_63_dilations_0"), val = tensor([1, 1])]; + int32 obj_63_groups_0 = const()[name = string("obj_63_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614804480)))]; + tensor layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618081344)))]; + tensor obj_63_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_121_cast_fp16)[name = string("obj_63_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_63_cast_fp16)[name = string("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; + fp16 var_2037_to_fp16 = const()[name = string("op_2037_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2037_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; + tensor input_123_gamma_0_to_fp16 = const()[name = string("input_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618083968)))]; + tensor input_123_beta_0_to_fp16 = const()[name = string("input_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618086592)))]; + fp16 input_123_epsilon_0_to_fp16 = const()[name = string("input_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_123_cast_fp16 = batch_norm(beta = input_123_beta_0_to_fp16, epsilon = input_123_epsilon_0_to_fp16, gamma = input_123_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("input_123_cast_fp16")]; + string input_125_pad_type_0 = const()[name = string("input_125_pad_type_0"), val = string("valid")]; + tensor input_125_strides_0 = const()[name = string("input_125_strides_0"), val = tensor([1, 1])]; + tensor input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_125_dilations_0 = const()[name = string("input_125_dilations_0"), val = tensor([1, 1])]; + int32 input_125_groups_0 = const()[name = string("input_125_groups_0"), val = int32(1)]; + tensor layers_15_fc1_weight_to_fp16 = const()[name = string("layers_15_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618089216)))]; + tensor layers_15_fc1_bias_to_fp16 = const()[name = string("layers_15_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631196480)))]; + tensor input_125_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = layers_15_fc1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")]; + string input_127_mode_0 = const()[name = string("input_127_mode_0"), val = string("EXACT")]; + tensor input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")]; + string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")]; + tensor hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)]; + tensor layers_15_fc2_weight_to_fp16 = const()[name = string("layers_15_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631206784)))]; + tensor layers_15_fc2_bias_to_fp16 = const()[name = string("layers_15_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644314048)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_15_fc2_weight_to_fp16, x = input_127_cast_fp16)[name = string("hidden_states_35_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = hidden_states_35_cast_fp16)[name = string("inputs_65_cast_fp16")]; + int32 var_2070 = const()[name = string("op_2070"), val = int32(3)]; + tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; + fp16 var_2089_to_fp16 = const()[name = string("op_2089_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2089_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; + tensor obj_65_gamma_0_to_fp16 = const()[name = string("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644316672)))]; + tensor obj_65_beta_0_to_fp16 = const()[name = string("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644319296)))]; + fp16 obj_65_epsilon_0_to_fp16 = const()[name = string("obj_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("obj_65_cast_fp16")]; + string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; + tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; + tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; + int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644321920)))]; + tensor layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647598784)))]; + tensor query_33_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("query_33_cast_fp16")]; + string key_33_pad_type_0 = const()[name = string("key_33_pad_type_0"), val = string("valid")]; + tensor key_33_strides_0 = const()[name = string("key_33_strides_0"), val = tensor([1, 1])]; + tensor key_33_pad_0 = const()[name = string("key_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_33_dilations_0 = const()[name = string("key_33_dilations_0"), val = tensor([1, 1])]; + int32 key_33_groups_0 = const()[name = string("key_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647601408)))]; + tensor key_33_cast_fp16 = conv(dilations = key_33_dilations_0, groups = key_33_groups_0, pad = key_33_pad_0, pad_type = key_33_pad_type_0, strides = key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("key_33_cast_fp16")]; + string value_33_pad_type_0 = const()[name = string("value_33_pad_type_0"), val = string("valid")]; + tensor value_33_strides_0 = const()[name = string("value_33_strides_0"), val = tensor([1, 1])]; + tensor value_33_pad_0 = const()[name = string("value_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_33_dilations_0 = const()[name = string("value_33_dilations_0"), val = tensor([1, 1])]; + int32 value_33_groups_0 = const()[name = string("value_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(650878272)))]; + tensor layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654155136)))]; + tensor value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = value_33_dilations_0, groups = value_33_groups_0, pad = value_33_pad_0, pad_type = value_33_pad_type_0, strides = value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("value_33_cast_fp16")]; + tensor var_2124 = const()[name = string("op_2124"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_33_cast_fp16 = reshape(shape = var_2124, x = query_33_cast_fp16)[name = string("mh_q_33_cast_fp16")]; + fp16 var_2126_to_fp16 = const()[name = string("op_2126_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2127_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_2126_to_fp16)[name = string("op_2127_cast_fp16")]; + tensor var_2128 = const()[name = string("op_2128"), val = tensor([1, 20, 64, -1])]; + tensor var_2129_cast_fp16 = reshape(shape = var_2128, x = key_33_cast_fp16)[name = string("op_2129_cast_fp16")]; + bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; + bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_2127_cast_fp16, y = var_2129_cast_fp16)[name = string("mh_w_33_cast_fp16")]; + tensor var_2132_cast_fp16 = softmax(axis = var_2070, x = mh_w_33_cast_fp16)[name = string("op_2132_cast_fp16")]; + tensor var_2133 = const()[name = string("op_2133"), val = tensor([1, 20, 64, -1])]; + tensor var_2134_cast_fp16 = reshape(shape = var_2133, x = value_33_cast_fp16)[name = string("op_2134_cast_fp16")]; + bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)]; + bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)]; + tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_2134_cast_fp16, y = var_2132_cast_fp16)[name = string("attn_33_cast_fp16")]; + tensor var_2137 = const()[name = string("op_2137"), val = tensor([1, 1280, 1, -1])]; + tensor input_129_cast_fp16 = reshape(shape = var_2137, x = attn_33_cast_fp16)[name = string("input_129_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654157760)))]; + tensor layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657434624)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_129_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_67_cast_fp16")]; + tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; + fp16 var_2155_to_fp16 = const()[name = string("op_2155_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2155_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; + tensor input_131_gamma_0_to_fp16 = const()[name = string("input_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657437248)))]; + tensor input_131_beta_0_to_fp16 = const()[name = string("input_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657439872)))]; + fp16 input_131_epsilon_0_to_fp16 = const()[name = string("input_131_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("input_131_cast_fp16")]; + string input_133_pad_type_0 = const()[name = string("input_133_pad_type_0"), val = string("valid")]; + tensor input_133_strides_0 = const()[name = string("input_133_strides_0"), val = tensor([1, 1])]; + tensor input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_133_dilations_0 = const()[name = string("input_133_dilations_0"), val = tensor([1, 1])]; + int32 input_133_groups_0 = const()[name = string("input_133_groups_0"), val = int32(1)]; + tensor layers_16_fc1_weight_to_fp16 = const()[name = string("layers_16_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657442496)))]; + tensor layers_16_fc1_bias_to_fp16 = const()[name = string("layers_16_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670549760)))]; + tensor input_133_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = layers_16_fc1_weight_to_fp16, x = input_131_cast_fp16)[name = string("input_133_cast_fp16")]; + string input_135_mode_0 = const()[name = string("input_135_mode_0"), val = string("EXACT")]; + tensor input_135_cast_fp16 = gelu(mode = input_135_mode_0, x = input_133_cast_fp16)[name = string("input_135_cast_fp16")]; + string hidden_states_37_pad_type_0 = const()[name = string("hidden_states_37_pad_type_0"), val = string("valid")]; + tensor hidden_states_37_strides_0 = const()[name = string("hidden_states_37_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_37_pad_0 = const()[name = string("hidden_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_37_dilations_0 = const()[name = string("hidden_states_37_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_37_groups_0 = const()[name = string("hidden_states_37_groups_0"), val = int32(1)]; + tensor layers_16_fc2_weight_to_fp16 = const()[name = string("layers_16_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670560064)))]; + tensor layers_16_fc2_bias_to_fp16 = const()[name = string("layers_16_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683667328)))]; + tensor hidden_states_37_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_16_fc2_weight_to_fp16, x = input_135_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = hidden_states_37_cast_fp16)[name = string("inputs_69_cast_fp16")]; + int32 var_2188 = const()[name = string("op_2188"), val = int32(3)]; + tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; + fp16 var_2207_to_fp16 = const()[name = string("op_2207_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2207_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683669952)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683672576)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; + tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; + tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; + int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683675200)))]; + tensor layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686952064)))]; + tensor query_35_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_35_cast_fp16")]; + string key_35_pad_type_0 = const()[name = string("key_35_pad_type_0"), val = string("valid")]; + tensor key_35_strides_0 = const()[name = string("key_35_strides_0"), val = tensor([1, 1])]; + tensor key_35_pad_0 = const()[name = string("key_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_35_dilations_0 = const()[name = string("key_35_dilations_0"), val = tensor([1, 1])]; + int32 key_35_groups_0 = const()[name = string("key_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686954688)))]; + tensor key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("key_35_cast_fp16")]; + string value_35_pad_type_0 = const()[name = string("value_35_pad_type_0"), val = string("valid")]; + tensor value_35_strides_0 = const()[name = string("value_35_strides_0"), val = tensor([1, 1])]; + tensor value_35_pad_0 = const()[name = string("value_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_35_dilations_0 = const()[name = string("value_35_dilations_0"), val = tensor([1, 1])]; + int32 value_35_groups_0 = const()[name = string("value_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(690231552)))]; + tensor layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693508416)))]; + tensor value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("value_35_cast_fp16")]; + tensor var_2242 = const()[name = string("op_2242"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_35_cast_fp16 = reshape(shape = var_2242, x = query_35_cast_fp16)[name = string("mh_q_35_cast_fp16")]; + fp16 var_2244_to_fp16 = const()[name = string("op_2244_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2245_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2244_to_fp16)[name = string("op_2245_cast_fp16")]; + tensor var_2246 = const()[name = string("op_2246"), val = tensor([1, 20, 64, -1])]; + tensor var_2247_cast_fp16 = reshape(shape = var_2246, x = key_35_cast_fp16)[name = string("op_2247_cast_fp16")]; + bool mh_w_35_transpose_x_0 = const()[name = string("mh_w_35_transpose_x_0"), val = bool(true)]; + bool mh_w_35_transpose_y_0 = const()[name = string("mh_w_35_transpose_y_0"), val = bool(false)]; + tensor mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_2245_cast_fp16, y = var_2247_cast_fp16)[name = string("mh_w_35_cast_fp16")]; + tensor var_2250_cast_fp16 = softmax(axis = var_2188, x = mh_w_35_cast_fp16)[name = string("op_2250_cast_fp16")]; + tensor var_2251 = const()[name = string("op_2251"), val = tensor([1, 20, 64, -1])]; + tensor var_2252_cast_fp16 = reshape(shape = var_2251, x = value_35_cast_fp16)[name = string("op_2252_cast_fp16")]; + bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)]; + bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)]; + tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2252_cast_fp16, y = var_2250_cast_fp16)[name = string("attn_35_cast_fp16")]; + tensor var_2255 = const()[name = string("op_2255"), val = tensor([1, 1280, 1, -1])]; + tensor input_137_cast_fp16 = reshape(shape = var_2255, x = attn_35_cast_fp16)[name = string("input_137_cast_fp16")]; + string obj_71_pad_type_0 = const()[name = string("obj_71_pad_type_0"), val = string("valid")]; + tensor obj_71_strides_0 = const()[name = string("obj_71_strides_0"), val = tensor([1, 1])]; + tensor obj_71_pad_0 = const()[name = string("obj_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_71_dilations_0 = const()[name = string("obj_71_dilations_0"), val = tensor([1, 1])]; + int32 obj_71_groups_0 = const()[name = string("obj_71_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693511040)))]; + tensor layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696787904)))]; + tensor obj_71_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_71_dilations_0, groups = obj_71_groups_0, pad = obj_71_pad_0, pad_type = obj_71_pad_type_0, strides = obj_71_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("obj_71_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_71_cast_fp16)[name = string("inputs_71_cast_fp16")]; + tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; + fp16 var_2273_to_fp16 = const()[name = string("op_2273_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2273_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; + tensor input_139_gamma_0_to_fp16 = const()[name = string("input_139_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696790528)))]; + tensor input_139_beta_0_to_fp16 = const()[name = string("input_139_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696793152)))]; + fp16 input_139_epsilon_0_to_fp16 = const()[name = string("input_139_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_139_cast_fp16 = batch_norm(beta = input_139_beta_0_to_fp16, epsilon = input_139_epsilon_0_to_fp16, gamma = input_139_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_139_cast_fp16")]; + string input_141_pad_type_0 = const()[name = string("input_141_pad_type_0"), val = string("valid")]; + tensor input_141_strides_0 = const()[name = string("input_141_strides_0"), val = tensor([1, 1])]; + tensor input_141_pad_0 = const()[name = string("input_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_141_dilations_0 = const()[name = string("input_141_dilations_0"), val = tensor([1, 1])]; + int32 input_141_groups_0 = const()[name = string("input_141_groups_0"), val = int32(1)]; + tensor layers_17_fc1_weight_to_fp16 = const()[name = string("layers_17_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696795776)))]; + tensor layers_17_fc1_bias_to_fp16 = const()[name = string("layers_17_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709903040)))]; + tensor input_141_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_141_dilations_0, groups = input_141_groups_0, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = input_141_strides_0, weight = layers_17_fc1_weight_to_fp16, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")]; + string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("EXACT")]; + tensor input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")]; + string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")]; + tensor hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)]; + tensor layers_17_fc2_weight_to_fp16 = const()[name = string("layers_17_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709913344)))]; + tensor layers_17_fc2_bias_to_fp16 = const()[name = string("layers_17_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723020608)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_17_fc2_weight_to_fp16, x = input_143_cast_fp16)[name = string("hidden_states_39_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("inputs_73_cast_fp16")]; + int32 var_2306 = const()[name = string("op_2306"), val = int32(3)]; + tensor out_73_axes_0 = const()[name = string("out_73_axes_0"), val = tensor([1])]; + fp16 var_2325_to_fp16 = const()[name = string("op_2325_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2325_to_fp16, x = inputs_73_cast_fp16)[name = string("out_73_cast_fp16")]; + tensor obj_73_gamma_0_to_fp16 = const()[name = string("obj_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723023232)))]; + tensor obj_73_beta_0_to_fp16 = const()[name = string("obj_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723025856)))]; + fp16 obj_73_epsilon_0_to_fp16 = const()[name = string("obj_73_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_73_cast_fp16 = batch_norm(beta = obj_73_beta_0_to_fp16, epsilon = obj_73_epsilon_0_to_fp16, gamma = obj_73_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = string("obj_73_cast_fp16")]; + string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; + tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; + tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; + int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723028480)))]; + tensor layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726305344)))]; + tensor query_37_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("query_37_cast_fp16")]; + string key_37_pad_type_0 = const()[name = string("key_37_pad_type_0"), val = string("valid")]; + tensor key_37_strides_0 = const()[name = string("key_37_strides_0"), val = tensor([1, 1])]; + tensor key_37_pad_0 = const()[name = string("key_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_37_dilations_0 = const()[name = string("key_37_dilations_0"), val = tensor([1, 1])]; + int32 key_37_groups_0 = const()[name = string("key_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726307968)))]; + tensor key_37_cast_fp16 = conv(dilations = key_37_dilations_0, groups = key_37_groups_0, pad = key_37_pad_0, pad_type = key_37_pad_type_0, strides = key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("key_37_cast_fp16")]; + string value_37_pad_type_0 = const()[name = string("value_37_pad_type_0"), val = string("valid")]; + tensor value_37_strides_0 = const()[name = string("value_37_strides_0"), val = tensor([1, 1])]; + tensor value_37_pad_0 = const()[name = string("value_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_37_dilations_0 = const()[name = string("value_37_dilations_0"), val = tensor([1, 1])]; + int32 value_37_groups_0 = const()[name = string("value_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(729584832)))]; + tensor layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732861696)))]; + tensor value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = value_37_dilations_0, groups = value_37_groups_0, pad = value_37_pad_0, pad_type = value_37_pad_type_0, strides = value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("value_37_cast_fp16")]; + tensor var_2360 = const()[name = string("op_2360"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_37_cast_fp16 = reshape(shape = var_2360, x = query_37_cast_fp16)[name = string("mh_q_37_cast_fp16")]; + fp16 var_2362_to_fp16 = const()[name = string("op_2362_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2363_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2362_to_fp16)[name = string("op_2363_cast_fp16")]; + tensor var_2364 = const()[name = string("op_2364"), val = tensor([1, 20, 64, -1])]; + tensor var_2365_cast_fp16 = reshape(shape = var_2364, x = key_37_cast_fp16)[name = string("op_2365_cast_fp16")]; + bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; + bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_2363_cast_fp16, y = var_2365_cast_fp16)[name = string("mh_w_37_cast_fp16")]; + tensor var_2368_cast_fp16 = softmax(axis = var_2306, x = mh_w_37_cast_fp16)[name = string("op_2368_cast_fp16")]; + tensor var_2369 = const()[name = string("op_2369"), val = tensor([1, 20, 64, -1])]; + tensor var_2370_cast_fp16 = reshape(shape = var_2369, x = value_37_cast_fp16)[name = string("op_2370_cast_fp16")]; + bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)]; + bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)]; + tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2370_cast_fp16, y = var_2368_cast_fp16)[name = string("attn_37_cast_fp16")]; + tensor var_2373 = const()[name = string("op_2373"), val = tensor([1, 1280, 1, -1])]; + tensor input_145_cast_fp16 = reshape(shape = var_2373, x = attn_37_cast_fp16)[name = string("input_145_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732864320)))]; + tensor layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736141184)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_75_cast_fp16")]; + tensor out_75_axes_0 = const()[name = string("out_75_axes_0"), val = tensor([1])]; + fp16 var_2391_to_fp16 = const()[name = string("op_2391_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_2391_to_fp16, x = inputs_75_cast_fp16)[name = string("out_75_cast_fp16")]; + tensor input_147_gamma_0_to_fp16 = const()[name = string("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736143808)))]; + tensor input_147_beta_0_to_fp16 = const()[name = string("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736146432)))]; + fp16 input_147_epsilon_0_to_fp16 = const()[name = string("input_147_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = string("input_147_cast_fp16")]; + string input_149_pad_type_0 = const()[name = string("input_149_pad_type_0"), val = string("valid")]; + tensor input_149_strides_0 = const()[name = string("input_149_strides_0"), val = tensor([1, 1])]; + tensor input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_149_dilations_0 = const()[name = string("input_149_dilations_0"), val = tensor([1, 1])]; + int32 input_149_groups_0 = const()[name = string("input_149_groups_0"), val = int32(1)]; + tensor layers_18_fc1_weight_to_fp16 = const()[name = string("layers_18_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736149056)))]; + tensor layers_18_fc1_bias_to_fp16 = const()[name = string("layers_18_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749256320)))]; + tensor input_149_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_149_dilations_0, groups = input_149_groups_0, pad = input_149_pad_0, pad_type = input_149_pad_type_0, strides = input_149_strides_0, weight = layers_18_fc1_weight_to_fp16, x = input_147_cast_fp16)[name = string("input_149_cast_fp16")]; + string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("EXACT")]; + tensor input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")]; + string hidden_states_41_pad_type_0 = const()[name = string("hidden_states_41_pad_type_0"), val = string("valid")]; + tensor hidden_states_41_strides_0 = const()[name = string("hidden_states_41_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_41_pad_0 = const()[name = string("hidden_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_41_dilations_0 = const()[name = string("hidden_states_41_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_41_groups_0 = const()[name = string("hidden_states_41_groups_0"), val = int32(1)]; + tensor layers_18_fc2_weight_to_fp16 = const()[name = string("layers_18_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749266624)))]; + tensor layers_18_fc2_bias_to_fp16 = const()[name = string("layers_18_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762373888)))]; + tensor hidden_states_41_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_18_fc2_weight_to_fp16, x = input_151_cast_fp16)[name = string("hidden_states_41_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = hidden_states_41_cast_fp16)[name = string("inputs_77_cast_fp16")]; + int32 var_2424 = const()[name = string("op_2424"), val = int32(3)]; + tensor out_77_axes_0 = const()[name = string("out_77_axes_0"), val = tensor([1])]; + fp16 var_2443_to_fp16 = const()[name = string("op_2443_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_2443_to_fp16, x = inputs_77_cast_fp16)[name = string("out_77_cast_fp16")]; + tensor obj_77_gamma_0_to_fp16 = const()[name = string("obj_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762376512)))]; + tensor obj_77_beta_0_to_fp16 = const()[name = string("obj_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762379136)))]; + fp16 obj_77_epsilon_0_to_fp16 = const()[name = string("obj_77_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_77_cast_fp16 = batch_norm(beta = obj_77_beta_0_to_fp16, epsilon = obj_77_epsilon_0_to_fp16, gamma = obj_77_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = string("obj_77_cast_fp16")]; + string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; + tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; + tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; + int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762381760)))]; + tensor layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765658624)))]; + tensor query_39_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("query_39_cast_fp16")]; + string key_39_pad_type_0 = const()[name = string("key_39_pad_type_0"), val = string("valid")]; + tensor key_39_strides_0 = const()[name = string("key_39_strides_0"), val = tensor([1, 1])]; + tensor key_39_pad_0 = const()[name = string("key_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_39_dilations_0 = const()[name = string("key_39_dilations_0"), val = tensor([1, 1])]; + int32 key_39_groups_0 = const()[name = string("key_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765661248)))]; + tensor key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("key_39_cast_fp16")]; + string value_39_pad_type_0 = const()[name = string("value_39_pad_type_0"), val = string("valid")]; + tensor value_39_strides_0 = const()[name = string("value_39_strides_0"), val = tensor([1, 1])]; + tensor value_39_pad_0 = const()[name = string("value_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_39_dilations_0 = const()[name = string("value_39_dilations_0"), val = tensor([1, 1])]; + int32 value_39_groups_0 = const()[name = string("value_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(768938112)))]; + tensor layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772214976)))]; + tensor value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("value_39_cast_fp16")]; + tensor var_2478 = const()[name = string("op_2478"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_39_cast_fp16 = reshape(shape = var_2478, x = query_39_cast_fp16)[name = string("mh_q_39_cast_fp16")]; + fp16 var_2480_to_fp16 = const()[name = string("op_2480_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2481_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2480_to_fp16)[name = string("op_2481_cast_fp16")]; + tensor var_2482 = const()[name = string("op_2482"), val = tensor([1, 20, 64, -1])]; + tensor var_2483_cast_fp16 = reshape(shape = var_2482, x = key_39_cast_fp16)[name = string("op_2483_cast_fp16")]; + bool mh_w_39_transpose_x_0 = const()[name = string("mh_w_39_transpose_x_0"), val = bool(true)]; + bool mh_w_39_transpose_y_0 = const()[name = string("mh_w_39_transpose_y_0"), val = bool(false)]; + tensor mh_w_39_cast_fp16 = matmul(transpose_x = mh_w_39_transpose_x_0, transpose_y = mh_w_39_transpose_y_0, x = var_2481_cast_fp16, y = var_2483_cast_fp16)[name = string("mh_w_39_cast_fp16")]; + tensor var_2486_cast_fp16 = softmax(axis = var_2424, x = mh_w_39_cast_fp16)[name = string("op_2486_cast_fp16")]; + tensor var_2487 = const()[name = string("op_2487"), val = tensor([1, 20, 64, -1])]; + tensor var_2488_cast_fp16 = reshape(shape = var_2487, x = value_39_cast_fp16)[name = string("op_2488_cast_fp16")]; + bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)]; + bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)]; + tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2488_cast_fp16, y = var_2486_cast_fp16)[name = string("attn_39_cast_fp16")]; + tensor var_2491 = const()[name = string("op_2491"), val = tensor([1, 1280, 1, -1])]; + tensor input_153_cast_fp16 = reshape(shape = var_2491, x = attn_39_cast_fp16)[name = string("input_153_cast_fp16")]; + string obj_79_pad_type_0 = const()[name = string("obj_79_pad_type_0"), val = string("valid")]; + tensor obj_79_strides_0 = const()[name = string("obj_79_strides_0"), val = tensor([1, 1])]; + tensor obj_79_pad_0 = const()[name = string("obj_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_79_dilations_0 = const()[name = string("obj_79_dilations_0"), val = tensor([1, 1])]; + int32 obj_79_groups_0 = const()[name = string("obj_79_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772217600)))]; + tensor layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775494464)))]; + tensor obj_79_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_79_dilations_0, groups = obj_79_groups_0, pad = obj_79_pad_0, pad_type = obj_79_pad_type_0, strides = obj_79_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = string("obj_79_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = obj_79_cast_fp16)[name = string("inputs_79_cast_fp16")]; + tensor out_79_axes_0 = const()[name = string("out_79_axes_0"), val = tensor([1])]; + fp16 var_2509_to_fp16 = const()[name = string("op_2509_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_2509_to_fp16, x = inputs_79_cast_fp16)[name = string("out_79_cast_fp16")]; + tensor input_155_gamma_0_to_fp16 = const()[name = string("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775497088)))]; + tensor input_155_beta_0_to_fp16 = const()[name = string("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775499712)))]; + fp16 input_155_epsilon_0_to_fp16 = const()[name = string("input_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = string("input_155_cast_fp16")]; + string input_157_pad_type_0 = const()[name = string("input_157_pad_type_0"), val = string("valid")]; + tensor input_157_strides_0 = const()[name = string("input_157_strides_0"), val = tensor([1, 1])]; + tensor input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_157_dilations_0 = const()[name = string("input_157_dilations_0"), val = tensor([1, 1])]; + int32 input_157_groups_0 = const()[name = string("input_157_groups_0"), val = int32(1)]; + tensor layers_19_fc1_weight_to_fp16 = const()[name = string("layers_19_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775502336)))]; + tensor layers_19_fc1_bias_to_fp16 = const()[name = string("layers_19_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788609600)))]; + tensor input_157_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_19_fc1_weight_to_fp16, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")]; + string input_159_mode_0 = const()[name = string("input_159_mode_0"), val = string("EXACT")]; + tensor input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")]; + string hidden_states_43_pad_type_0 = const()[name = string("hidden_states_43_pad_type_0"), val = string("valid")]; + tensor hidden_states_43_strides_0 = const()[name = string("hidden_states_43_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_43_pad_0 = const()[name = string("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_43_dilations_0 = const()[name = string("hidden_states_43_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_43_groups_0 = const()[name = string("hidden_states_43_groups_0"), val = int32(1)]; + tensor layers_19_fc2_weight_to_fp16 = const()[name = string("layers_19_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788619904)))]; + tensor layers_19_fc2_bias_to_fp16 = const()[name = string("layers_19_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801727168)))]; + tensor hidden_states_43_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_19_fc2_weight_to_fp16, x = input_159_cast_fp16)[name = string("hidden_states_43_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = hidden_states_43_cast_fp16)[name = string("inputs_81_cast_fp16")]; + int32 var_2542 = const()[name = string("op_2542"), val = int32(3)]; + tensor out_81_axes_0 = const()[name = string("out_81_axes_0"), val = tensor([1])]; + fp16 var_2561_to_fp16 = const()[name = string("op_2561_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_2561_to_fp16, x = inputs_81_cast_fp16)[name = string("out_81_cast_fp16")]; + tensor obj_81_gamma_0_to_fp16 = const()[name = string("obj_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801729792)))]; + tensor obj_81_beta_0_to_fp16 = const()[name = string("obj_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801732416)))]; + fp16 obj_81_epsilon_0_to_fp16 = const()[name = string("obj_81_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_81_cast_fp16 = batch_norm(beta = obj_81_beta_0_to_fp16, epsilon = obj_81_epsilon_0_to_fp16, gamma = obj_81_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = string("obj_81_cast_fp16")]; + string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; + tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; + tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; + int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801735040)))]; + tensor layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805011904)))]; + tensor query_41_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("query_41_cast_fp16")]; + string key_41_pad_type_0 = const()[name = string("key_41_pad_type_0"), val = string("valid")]; + tensor key_41_strides_0 = const()[name = string("key_41_strides_0"), val = tensor([1, 1])]; + tensor key_41_pad_0 = const()[name = string("key_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_41_dilations_0 = const()[name = string("key_41_dilations_0"), val = tensor([1, 1])]; + int32 key_41_groups_0 = const()[name = string("key_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805014528)))]; + tensor key_41_cast_fp16 = conv(dilations = key_41_dilations_0, groups = key_41_groups_0, pad = key_41_pad_0, pad_type = key_41_pad_type_0, strides = key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("key_41_cast_fp16")]; + string value_41_pad_type_0 = const()[name = string("value_41_pad_type_0"), val = string("valid")]; + tensor value_41_strides_0 = const()[name = string("value_41_strides_0"), val = tensor([1, 1])]; + tensor value_41_pad_0 = const()[name = string("value_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_41_dilations_0 = const()[name = string("value_41_dilations_0"), val = tensor([1, 1])]; + int32 value_41_groups_0 = const()[name = string("value_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(808291392)))]; + tensor layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811568256)))]; + tensor value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = value_41_dilations_0, groups = value_41_groups_0, pad = value_41_pad_0, pad_type = value_41_pad_type_0, strides = value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("value_41_cast_fp16")]; + tensor var_2596 = const()[name = string("op_2596"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_41_cast_fp16 = reshape(shape = var_2596, x = query_41_cast_fp16)[name = string("mh_q_41_cast_fp16")]; + fp16 var_2598_to_fp16 = const()[name = string("op_2598_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2599_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2598_to_fp16)[name = string("op_2599_cast_fp16")]; + tensor var_2600 = const()[name = string("op_2600"), val = tensor([1, 20, 64, -1])]; + tensor var_2601_cast_fp16 = reshape(shape = var_2600, x = key_41_cast_fp16)[name = string("op_2601_cast_fp16")]; + bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; + bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_2599_cast_fp16, y = var_2601_cast_fp16)[name = string("mh_w_41_cast_fp16")]; + tensor var_2604_cast_fp16 = softmax(axis = var_2542, x = mh_w_41_cast_fp16)[name = string("op_2604_cast_fp16")]; + tensor var_2605 = const()[name = string("op_2605"), val = tensor([1, 20, 64, -1])]; + tensor var_2606_cast_fp16 = reshape(shape = var_2605, x = value_41_cast_fp16)[name = string("op_2606_cast_fp16")]; + bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)]; + bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)]; + tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2606_cast_fp16, y = var_2604_cast_fp16)[name = string("attn_41_cast_fp16")]; + tensor var_2609 = const()[name = string("op_2609"), val = tensor([1, 1280, 1, -1])]; + tensor input_161_cast_fp16 = reshape(shape = var_2609, x = attn_41_cast_fp16)[name = string("input_161_cast_fp16")]; + string obj_83_pad_type_0 = const()[name = string("obj_83_pad_type_0"), val = string("valid")]; + tensor obj_83_strides_0 = const()[name = string("obj_83_strides_0"), val = tensor([1, 1])]; + tensor obj_83_pad_0 = const()[name = string("obj_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_83_dilations_0 = const()[name = string("obj_83_dilations_0"), val = tensor([1, 1])]; + int32 obj_83_groups_0 = const()[name = string("obj_83_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811570880)))]; + tensor layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814847744)))]; + tensor obj_83_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_83_dilations_0, groups = obj_83_groups_0, pad = obj_83_pad_0, pad_type = obj_83_pad_type_0, strides = obj_83_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_161_cast_fp16)[name = string("obj_83_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_83_cast_fp16)[name = string("inputs_83_cast_fp16")]; + tensor out_83_axes_0 = const()[name = string("out_83_axes_0"), val = tensor([1])]; + fp16 var_2627_to_fp16 = const()[name = string("op_2627_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_2627_to_fp16, x = inputs_83_cast_fp16)[name = string("out_83_cast_fp16")]; + tensor input_163_gamma_0_to_fp16 = const()[name = string("input_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814850368)))]; + tensor input_163_beta_0_to_fp16 = const()[name = string("input_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814852992)))]; + fp16 input_163_epsilon_0_to_fp16 = const()[name = string("input_163_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_163_cast_fp16 = batch_norm(beta = input_163_beta_0_to_fp16, epsilon = input_163_epsilon_0_to_fp16, gamma = input_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = string("input_163_cast_fp16")]; + string input_165_pad_type_0 = const()[name = string("input_165_pad_type_0"), val = string("valid")]; + tensor input_165_strides_0 = const()[name = string("input_165_strides_0"), val = tensor([1, 1])]; + tensor input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_165_dilations_0 = const()[name = string("input_165_dilations_0"), val = tensor([1, 1])]; + int32 input_165_groups_0 = const()[name = string("input_165_groups_0"), val = int32(1)]; + tensor layers_20_fc1_weight_to_fp16 = const()[name = string("layers_20_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814855616)))]; + tensor layers_20_fc1_bias_to_fp16 = const()[name = string("layers_20_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827962880)))]; + tensor input_165_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_165_dilations_0, groups = input_165_groups_0, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = input_165_strides_0, weight = layers_20_fc1_weight_to_fp16, x = input_163_cast_fp16)[name = string("input_165_cast_fp16")]; + string input_167_mode_0 = const()[name = string("input_167_mode_0"), val = string("EXACT")]; + tensor input_167_cast_fp16 = gelu(mode = input_167_mode_0, x = input_165_cast_fp16)[name = string("input_167_cast_fp16")]; + string hidden_states_45_pad_type_0 = const()[name = string("hidden_states_45_pad_type_0"), val = string("valid")]; + tensor hidden_states_45_strides_0 = const()[name = string("hidden_states_45_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_45_pad_0 = const()[name = string("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_45_dilations_0 = const()[name = string("hidden_states_45_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_45_groups_0 = const()[name = string("hidden_states_45_groups_0"), val = int32(1)]; + tensor layers_20_fc2_weight_to_fp16 = const()[name = string("layers_20_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827973184)))]; + tensor layers_20_fc2_bias_to_fp16 = const()[name = string("layers_20_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841080448)))]; + tensor hidden_states_45_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_20_fc2_weight_to_fp16, x = input_167_cast_fp16)[name = string("hidden_states_45_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_45_cast_fp16)[name = string("inputs_85_cast_fp16")]; + int32 var_2660 = const()[name = string("op_2660"), val = int32(3)]; + tensor out_85_axes_0 = const()[name = string("out_85_axes_0"), val = tensor([1])]; + fp16 var_2679_to_fp16 = const()[name = string("op_2679_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_2679_to_fp16, x = inputs_85_cast_fp16)[name = string("out_85_cast_fp16")]; + tensor obj_85_gamma_0_to_fp16 = const()[name = string("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841083072)))]; + tensor obj_85_beta_0_to_fp16 = const()[name = string("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841085696)))]; + fp16 obj_85_epsilon_0_to_fp16 = const()[name = string("obj_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = string("obj_85_cast_fp16")]; + string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; + tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; + tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; + int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841088320)))]; + tensor layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844365184)))]; + tensor query_43_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("query_43_cast_fp16")]; + string key_43_pad_type_0 = const()[name = string("key_43_pad_type_0"), val = string("valid")]; + tensor key_43_strides_0 = const()[name = string("key_43_strides_0"), val = tensor([1, 1])]; + tensor key_43_pad_0 = const()[name = string("key_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_43_dilations_0 = const()[name = string("key_43_dilations_0"), val = tensor([1, 1])]; + int32 key_43_groups_0 = const()[name = string("key_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844367808)))]; + tensor key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("key_43_cast_fp16")]; + string value_43_pad_type_0 = const()[name = string("value_43_pad_type_0"), val = string("valid")]; + tensor value_43_strides_0 = const()[name = string("value_43_strides_0"), val = tensor([1, 1])]; + tensor value_43_pad_0 = const()[name = string("value_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_43_dilations_0 = const()[name = string("value_43_dilations_0"), val = tensor([1, 1])]; + int32 value_43_groups_0 = const()[name = string("value_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(847644672)))]; + tensor layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850921536)))]; + tensor value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("value_43_cast_fp16")]; + tensor var_2714 = const()[name = string("op_2714"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_43_cast_fp16 = reshape(shape = var_2714, x = query_43_cast_fp16)[name = string("mh_q_43_cast_fp16")]; + fp16 var_2716_to_fp16 = const()[name = string("op_2716_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2717_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2716_to_fp16)[name = string("op_2717_cast_fp16")]; + tensor var_2718 = const()[name = string("op_2718"), val = tensor([1, 20, 64, -1])]; + tensor var_2719_cast_fp16 = reshape(shape = var_2718, x = key_43_cast_fp16)[name = string("op_2719_cast_fp16")]; + bool mh_w_43_transpose_x_0 = const()[name = string("mh_w_43_transpose_x_0"), val = bool(true)]; + bool mh_w_43_transpose_y_0 = const()[name = string("mh_w_43_transpose_y_0"), val = bool(false)]; + tensor mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_2717_cast_fp16, y = var_2719_cast_fp16)[name = string("mh_w_43_cast_fp16")]; + tensor var_2722_cast_fp16 = softmax(axis = var_2660, x = mh_w_43_cast_fp16)[name = string("op_2722_cast_fp16")]; + tensor var_2723 = const()[name = string("op_2723"), val = tensor([1, 20, 64, -1])]; + tensor var_2724_cast_fp16 = reshape(shape = var_2723, x = value_43_cast_fp16)[name = string("op_2724_cast_fp16")]; + bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)]; + bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)]; + tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2724_cast_fp16, y = var_2722_cast_fp16)[name = string("attn_43_cast_fp16")]; + tensor var_2727 = const()[name = string("op_2727"), val = tensor([1, 1280, 1, -1])]; + tensor input_169_cast_fp16 = reshape(shape = var_2727, x = attn_43_cast_fp16)[name = string("input_169_cast_fp16")]; + string obj_87_pad_type_0 = const()[name = string("obj_87_pad_type_0"), val = string("valid")]; + tensor obj_87_strides_0 = const()[name = string("obj_87_strides_0"), val = tensor([1, 1])]; + tensor obj_87_pad_0 = const()[name = string("obj_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_87_dilations_0 = const()[name = string("obj_87_dilations_0"), val = tensor([1, 1])]; + int32 obj_87_groups_0 = const()[name = string("obj_87_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850924160)))]; + tensor layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854201024)))]; + tensor obj_87_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_87_dilations_0, groups = obj_87_groups_0, pad = obj_87_pad_0, pad_type = obj_87_pad_type_0, strides = obj_87_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_169_cast_fp16)[name = string("obj_87_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_87_cast_fp16)[name = string("inputs_87_cast_fp16")]; + tensor out_87_axes_0 = const()[name = string("out_87_axes_0"), val = tensor([1])]; + fp16 var_2745_to_fp16 = const()[name = string("op_2745_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_2745_to_fp16, x = inputs_87_cast_fp16)[name = string("out_87_cast_fp16")]; + tensor input_171_gamma_0_to_fp16 = const()[name = string("input_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854203648)))]; + tensor input_171_beta_0_to_fp16 = const()[name = string("input_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854206272)))]; + fp16 input_171_epsilon_0_to_fp16 = const()[name = string("input_171_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_171_cast_fp16 = batch_norm(beta = input_171_beta_0_to_fp16, epsilon = input_171_epsilon_0_to_fp16, gamma = input_171_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = string("input_171_cast_fp16")]; + string input_173_pad_type_0 = const()[name = string("input_173_pad_type_0"), val = string("valid")]; + tensor input_173_strides_0 = const()[name = string("input_173_strides_0"), val = tensor([1, 1])]; + tensor input_173_pad_0 = const()[name = string("input_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_173_dilations_0 = const()[name = string("input_173_dilations_0"), val = tensor([1, 1])]; + int32 input_173_groups_0 = const()[name = string("input_173_groups_0"), val = int32(1)]; + tensor layers_21_fc1_weight_to_fp16 = const()[name = string("layers_21_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854208896)))]; + tensor layers_21_fc1_bias_to_fp16 = const()[name = string("layers_21_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867316160)))]; + tensor input_173_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_173_dilations_0, groups = input_173_groups_0, pad = input_173_pad_0, pad_type = input_173_pad_type_0, strides = input_173_strides_0, weight = layers_21_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = string("input_173_cast_fp16")]; + string input_175_mode_0 = const()[name = string("input_175_mode_0"), val = string("EXACT")]; + tensor input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")]; + string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")]; + tensor hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)]; + tensor layers_21_fc2_weight_to_fp16 = const()[name = string("layers_21_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867326464)))]; + tensor layers_21_fc2_bias_to_fp16 = const()[name = string("layers_21_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880433728)))]; + tensor hidden_states_47_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_21_fc2_weight_to_fp16, x = input_175_cast_fp16)[name = string("hidden_states_47_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = hidden_states_47_cast_fp16)[name = string("inputs_89_cast_fp16")]; + int32 var_2778 = const()[name = string("op_2778"), val = int32(3)]; + tensor out_89_axes_0 = const()[name = string("out_89_axes_0"), val = tensor([1])]; + fp16 var_2797_to_fp16 = const()[name = string("op_2797_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_2797_to_fp16, x = inputs_89_cast_fp16)[name = string("out_89_cast_fp16")]; + tensor obj_89_gamma_0_to_fp16 = const()[name = string("obj_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880436352)))]; + tensor obj_89_beta_0_to_fp16 = const()[name = string("obj_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880438976)))]; + fp16 obj_89_epsilon_0_to_fp16 = const()[name = string("obj_89_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_89_cast_fp16 = batch_norm(beta = obj_89_beta_0_to_fp16, epsilon = obj_89_epsilon_0_to_fp16, gamma = obj_89_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = string("obj_89_cast_fp16")]; + string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; + tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; + tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; + int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880441600)))]; + tensor layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883718464)))]; + tensor query_45_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("query_45_cast_fp16")]; + string key_45_pad_type_0 = const()[name = string("key_45_pad_type_0"), val = string("valid")]; + tensor key_45_strides_0 = const()[name = string("key_45_strides_0"), val = tensor([1, 1])]; + tensor key_45_pad_0 = const()[name = string("key_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_45_dilations_0 = const()[name = string("key_45_dilations_0"), val = tensor([1, 1])]; + int32 key_45_groups_0 = const()[name = string("key_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883721088)))]; + tensor key_45_cast_fp16 = conv(dilations = key_45_dilations_0, groups = key_45_groups_0, pad = key_45_pad_0, pad_type = key_45_pad_type_0, strides = key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("key_45_cast_fp16")]; + string value_45_pad_type_0 = const()[name = string("value_45_pad_type_0"), val = string("valid")]; + tensor value_45_strides_0 = const()[name = string("value_45_strides_0"), val = tensor([1, 1])]; + tensor value_45_pad_0 = const()[name = string("value_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_45_dilations_0 = const()[name = string("value_45_dilations_0"), val = tensor([1, 1])]; + int32 value_45_groups_0 = const()[name = string("value_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(886997952)))]; + tensor layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890274816)))]; + tensor value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = value_45_dilations_0, groups = value_45_groups_0, pad = value_45_pad_0, pad_type = value_45_pad_type_0, strides = value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("value_45_cast_fp16")]; + tensor var_2832 = const()[name = string("op_2832"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_45_cast_fp16 = reshape(shape = var_2832, x = query_45_cast_fp16)[name = string("mh_q_45_cast_fp16")]; + fp16 var_2834_to_fp16 = const()[name = string("op_2834_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2835_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2834_to_fp16)[name = string("op_2835_cast_fp16")]; + tensor var_2836 = const()[name = string("op_2836"), val = tensor([1, 20, 64, -1])]; + tensor var_2837_cast_fp16 = reshape(shape = var_2836, x = key_45_cast_fp16)[name = string("op_2837_cast_fp16")]; + bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; + bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_2835_cast_fp16, y = var_2837_cast_fp16)[name = string("mh_w_45_cast_fp16")]; + tensor var_2840_cast_fp16 = softmax(axis = var_2778, x = mh_w_45_cast_fp16)[name = string("op_2840_cast_fp16")]; + tensor var_2841 = const()[name = string("op_2841"), val = tensor([1, 20, 64, -1])]; + tensor var_2842_cast_fp16 = reshape(shape = var_2841, x = value_45_cast_fp16)[name = string("op_2842_cast_fp16")]; + bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)]; + bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)]; + tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2842_cast_fp16, y = var_2840_cast_fp16)[name = string("attn_45_cast_fp16")]; + tensor var_2845 = const()[name = string("op_2845"), val = tensor([1, 1280, 1, -1])]; + tensor input_177_cast_fp16 = reshape(shape = var_2845, x = attn_45_cast_fp16)[name = string("input_177_cast_fp16")]; + string obj_91_pad_type_0 = const()[name = string("obj_91_pad_type_0"), val = string("valid")]; + tensor obj_91_strides_0 = const()[name = string("obj_91_strides_0"), val = tensor([1, 1])]; + tensor obj_91_pad_0 = const()[name = string("obj_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_91_dilations_0 = const()[name = string("obj_91_dilations_0"), val = tensor([1, 1])]; + int32 obj_91_groups_0 = const()[name = string("obj_91_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890277440)))]; + tensor layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893554304)))]; + tensor obj_91_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_177_cast_fp16)[name = string("obj_91_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = obj_91_cast_fp16)[name = string("inputs_91_cast_fp16")]; + tensor out_91_axes_0 = const()[name = string("out_91_axes_0"), val = tensor([1])]; + fp16 var_2863_to_fp16 = const()[name = string("op_2863_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_2863_to_fp16, x = inputs_91_cast_fp16)[name = string("out_91_cast_fp16")]; + tensor input_179_gamma_0_to_fp16 = const()[name = string("input_179_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893556928)))]; + tensor input_179_beta_0_to_fp16 = const()[name = string("input_179_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893559552)))]; + fp16 input_179_epsilon_0_to_fp16 = const()[name = string("input_179_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_179_cast_fp16 = batch_norm(beta = input_179_beta_0_to_fp16, epsilon = input_179_epsilon_0_to_fp16, gamma = input_179_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = string("input_179_cast_fp16")]; + string input_181_pad_type_0 = const()[name = string("input_181_pad_type_0"), val = string("valid")]; + tensor input_181_strides_0 = const()[name = string("input_181_strides_0"), val = tensor([1, 1])]; + tensor input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_181_dilations_0 = const()[name = string("input_181_dilations_0"), val = tensor([1, 1])]; + int32 input_181_groups_0 = const()[name = string("input_181_groups_0"), val = int32(1)]; + tensor layers_22_fc1_weight_to_fp16 = const()[name = string("layers_22_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893562176)))]; + tensor layers_22_fc1_bias_to_fp16 = const()[name = string("layers_22_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906669440)))]; + tensor input_181_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_181_dilations_0, groups = input_181_groups_0, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = input_181_strides_0, weight = layers_22_fc1_weight_to_fp16, x = input_179_cast_fp16)[name = string("input_181_cast_fp16")]; + string input_183_mode_0 = const()[name = string("input_183_mode_0"), val = string("EXACT")]; + tensor input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = input_181_cast_fp16)[name = string("input_183_cast_fp16")]; + string hidden_states_49_pad_type_0 = const()[name = string("hidden_states_49_pad_type_0"), val = string("valid")]; + tensor hidden_states_49_strides_0 = const()[name = string("hidden_states_49_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_49_pad_0 = const()[name = string("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_49_dilations_0 = const()[name = string("hidden_states_49_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_49_groups_0 = const()[name = string("hidden_states_49_groups_0"), val = int32(1)]; + tensor layers_22_fc2_weight_to_fp16 = const()[name = string("layers_22_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906679744)))]; + tensor layers_22_fc2_bias_to_fp16 = const()[name = string("layers_22_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919787008)))]; + tensor hidden_states_49_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_22_fc2_weight_to_fp16, x = input_183_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("inputs_93_cast_fp16")]; + int32 var_2896 = const()[name = string("op_2896"), val = int32(3)]; + tensor out_93_axes_0 = const()[name = string("out_93_axes_0"), val = tensor([1])]; + fp16 var_2915_to_fp16 = const()[name = string("op_2915_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_2915_to_fp16, x = inputs_93_cast_fp16)[name = string("out_93_cast_fp16")]; + tensor obj_93_gamma_0_to_fp16 = const()[name = string("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919789632)))]; + tensor obj_93_beta_0_to_fp16 = const()[name = string("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919792256)))]; + fp16 obj_93_epsilon_0_to_fp16 = const()[name = string("obj_93_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = string("obj_93_cast_fp16")]; + string query_47_pad_type_0 = const()[name = string("query_47_pad_type_0"), val = string("valid")]; + tensor query_47_strides_0 = const()[name = string("query_47_strides_0"), val = tensor([1, 1])]; + tensor query_47_pad_0 = const()[name = string("query_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_47_dilations_0 = const()[name = string("query_47_dilations_0"), val = tensor([1, 1])]; + int32 query_47_groups_0 = const()[name = string("query_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919794880)))]; + tensor layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923071744)))]; + tensor query_47_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("query_47_cast_fp16")]; + string key_47_pad_type_0 = const()[name = string("key_47_pad_type_0"), val = string("valid")]; + tensor key_47_strides_0 = const()[name = string("key_47_strides_0"), val = tensor([1, 1])]; + tensor key_47_pad_0 = const()[name = string("key_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_47_dilations_0 = const()[name = string("key_47_dilations_0"), val = tensor([1, 1])]; + int32 key_47_groups_0 = const()[name = string("key_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923074368)))]; + tensor key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("key_47_cast_fp16")]; + string value_47_pad_type_0 = const()[name = string("value_47_pad_type_0"), val = string("valid")]; + tensor value_47_strides_0 = const()[name = string("value_47_strides_0"), val = tensor([1, 1])]; + tensor value_47_pad_0 = const()[name = string("value_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_47_dilations_0 = const()[name = string("value_47_dilations_0"), val = tensor([1, 1])]; + int32 value_47_groups_0 = const()[name = string("value_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(926351232)))]; + tensor layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929628096)))]; + tensor value_47_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("value_47_cast_fp16")]; + tensor var_2950 = const()[name = string("op_2950"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_47_cast_fp16 = reshape(shape = var_2950, x = query_47_cast_fp16)[name = string("mh_q_47_cast_fp16")]; + fp16 var_2952_to_fp16 = const()[name = string("op_2952_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2953_cast_fp16 = mul(x = mh_q_47_cast_fp16, y = var_2952_to_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2954 = const()[name = string("op_2954"), val = tensor([1, 20, 64, -1])]; + tensor var_2955_cast_fp16 = reshape(shape = var_2954, x = key_47_cast_fp16)[name = string("op_2955_cast_fp16")]; + bool mh_w_47_transpose_x_0 = const()[name = string("mh_w_47_transpose_x_0"), val = bool(true)]; + bool mh_w_47_transpose_y_0 = const()[name = string("mh_w_47_transpose_y_0"), val = bool(false)]; + tensor mh_w_47_cast_fp16 = matmul(transpose_x = mh_w_47_transpose_x_0, transpose_y = mh_w_47_transpose_y_0, x = var_2953_cast_fp16, y = var_2955_cast_fp16)[name = string("mh_w_47_cast_fp16")]; + tensor var_2958_cast_fp16 = softmax(axis = var_2896, x = mh_w_47_cast_fp16)[name = string("op_2958_cast_fp16")]; + tensor var_2959 = const()[name = string("op_2959"), val = tensor([1, 20, 64, -1])]; + tensor var_2960_cast_fp16 = reshape(shape = var_2959, x = value_47_cast_fp16)[name = string("op_2960_cast_fp16")]; + bool attn_47_transpose_x_0 = const()[name = string("attn_47_transpose_x_0"), val = bool(false)]; + bool attn_47_transpose_y_0 = const()[name = string("attn_47_transpose_y_0"), val = bool(true)]; + tensor attn_47_cast_fp16 = matmul(transpose_x = attn_47_transpose_x_0, transpose_y = attn_47_transpose_y_0, x = var_2960_cast_fp16, y = var_2958_cast_fp16)[name = string("attn_47_cast_fp16")]; + tensor var_2963 = const()[name = string("op_2963"), val = tensor([1, 1280, 1, -1])]; + tensor input_185_cast_fp16 = reshape(shape = var_2963, x = attn_47_cast_fp16)[name = string("input_185_cast_fp16")]; + string obj_95_pad_type_0 = const()[name = string("obj_95_pad_type_0"), val = string("valid")]; + tensor obj_95_strides_0 = const()[name = string("obj_95_strides_0"), val = tensor([1, 1])]; + tensor obj_95_pad_0 = const()[name = string("obj_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_95_dilations_0 = const()[name = string("obj_95_dilations_0"), val = tensor([1, 1])]; + int32 obj_95_groups_0 = const()[name = string("obj_95_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929630720)))]; + tensor layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932907584)))]; + tensor obj_95_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_95_cast_fp16)[name = string("inputs_95_cast_fp16")]; + tensor out_95_axes_0 = const()[name = string("out_95_axes_0"), val = tensor([1])]; + fp16 var_2981_to_fp16 = const()[name = string("op_2981_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_2981_to_fp16, x = inputs_95_cast_fp16)[name = string("out_95_cast_fp16")]; + tensor input_187_gamma_0_to_fp16 = const()[name = string("input_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932910208)))]; + tensor input_187_beta_0_to_fp16 = const()[name = string("input_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932912832)))]; + fp16 input_187_epsilon_0_to_fp16 = const()[name = string("input_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_187_cast_fp16 = batch_norm(beta = input_187_beta_0_to_fp16, epsilon = input_187_epsilon_0_to_fp16, gamma = input_187_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = string("input_187_cast_fp16")]; + string input_189_pad_type_0 = const()[name = string("input_189_pad_type_0"), val = string("valid")]; + tensor input_189_strides_0 = const()[name = string("input_189_strides_0"), val = tensor([1, 1])]; + tensor input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_189_dilations_0 = const()[name = string("input_189_dilations_0"), val = tensor([1, 1])]; + int32 input_189_groups_0 = const()[name = string("input_189_groups_0"), val = int32(1)]; + tensor layers_23_fc1_weight_to_fp16 = const()[name = string("layers_23_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932915456)))]; + tensor layers_23_fc1_bias_to_fp16 = const()[name = string("layers_23_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946022720)))]; + tensor input_189_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = layers_23_fc1_weight_to_fp16, x = input_187_cast_fp16)[name = string("input_189_cast_fp16")]; + string input_191_mode_0 = const()[name = string("input_191_mode_0"), val = string("EXACT")]; + tensor input_191_cast_fp16 = gelu(mode = input_191_mode_0, x = input_189_cast_fp16)[name = string("input_191_cast_fp16")]; + string hidden_states_51_pad_type_0 = const()[name = string("hidden_states_51_pad_type_0"), val = string("valid")]; + tensor hidden_states_51_strides_0 = const()[name = string("hidden_states_51_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_51_pad_0 = const()[name = string("hidden_states_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_51_dilations_0 = const()[name = string("hidden_states_51_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_51_groups_0 = const()[name = string("hidden_states_51_groups_0"), val = int32(1)]; + tensor layers_23_fc2_weight_to_fp16 = const()[name = string("layers_23_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946033024)))]; + tensor layers_23_fc2_bias_to_fp16 = const()[name = string("layers_23_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959140288)))]; + tensor hidden_states_51_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_51_dilations_0, groups = hidden_states_51_groups_0, pad = hidden_states_51_pad_0, pad_type = hidden_states_51_pad_type_0, strides = hidden_states_51_strides_0, weight = layers_23_fc2_weight_to_fp16, x = input_191_cast_fp16)[name = string("hidden_states_51_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_51_cast_fp16)[name = string("inputs_97_cast_fp16")]; + int32 var_3014 = const()[name = string("op_3014"), val = int32(3)]; + tensor out_97_axes_0 = const()[name = string("out_97_axes_0"), val = tensor([1])]; + fp16 var_3033_to_fp16 = const()[name = string("op_3033_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3033_to_fp16, x = inputs_97_cast_fp16)[name = string("out_97_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959142912)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959145536)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_49_pad_type_0 = const()[name = string("query_49_pad_type_0"), val = string("valid")]; + tensor query_49_strides_0 = const()[name = string("query_49_strides_0"), val = tensor([1, 1])]; + tensor query_49_pad_0 = const()[name = string("query_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_49_dilations_0 = const()[name = string("query_49_dilations_0"), val = tensor([1, 1])]; + int32 query_49_groups_0 = const()[name = string("query_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959148160)))]; + tensor layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962425024)))]; + tensor query_49_cast_fp16 = conv(bias = layers_24_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_24_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_49_cast_fp16")]; + string key_49_pad_type_0 = const()[name = string("key_49_pad_type_0"), val = string("valid")]; + tensor key_49_strides_0 = const()[name = string("key_49_strides_0"), val = tensor([1, 1])]; + tensor key_49_pad_0 = const()[name = string("key_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_49_dilations_0 = const()[name = string("key_49_dilations_0"), val = tensor([1, 1])]; + int32 key_49_groups_0 = const()[name = string("key_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962427648)))]; + tensor key_49_cast_fp16 = conv(dilations = key_49_dilations_0, groups = key_49_groups_0, pad = key_49_pad_0, pad_type = key_49_pad_type_0, strides = key_49_strides_0, weight = layers_24_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("key_49_cast_fp16")]; + string value_49_pad_type_0 = const()[name = string("value_49_pad_type_0"), val = string("valid")]; + tensor value_49_strides_0 = const()[name = string("value_49_strides_0"), val = tensor([1, 1])]; + tensor value_49_pad_0 = const()[name = string("value_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_49_dilations_0 = const()[name = string("value_49_dilations_0"), val = tensor([1, 1])]; + int32 value_49_groups_0 = const()[name = string("value_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(965704512)))]; + tensor layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968981376)))]; + tensor value_49_cast_fp16 = conv(bias = layers_24_self_attn_v_proj_bias_to_fp16, dilations = value_49_dilations_0, groups = value_49_groups_0, pad = value_49_pad_0, pad_type = value_49_pad_type_0, strides = value_49_strides_0, weight = layers_24_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("value_49_cast_fp16")]; + tensor var_3068 = const()[name = string("op_3068"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_49_cast_fp16 = reshape(shape = var_3068, x = query_49_cast_fp16)[name = string("mh_q_49_cast_fp16")]; + fp16 var_3070_to_fp16 = const()[name = string("op_3070_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3071_cast_fp16 = mul(x = mh_q_49_cast_fp16, y = var_3070_to_fp16)[name = string("op_3071_cast_fp16")]; + tensor var_3072 = const()[name = string("op_3072"), val = tensor([1, 20, 64, -1])]; + tensor var_3073_cast_fp16 = reshape(shape = var_3072, x = key_49_cast_fp16)[name = string("op_3073_cast_fp16")]; + bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)]; + bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)]; + tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_3071_cast_fp16, y = var_3073_cast_fp16)[name = string("mh_w_49_cast_fp16")]; + tensor var_3076_cast_fp16 = softmax(axis = var_3014, x = mh_w_49_cast_fp16)[name = string("op_3076_cast_fp16")]; + tensor var_3077 = const()[name = string("op_3077"), val = tensor([1, 20, 64, -1])]; + tensor var_3078_cast_fp16 = reshape(shape = var_3077, x = value_49_cast_fp16)[name = string("op_3078_cast_fp16")]; + bool attn_49_transpose_x_0 = const()[name = string("attn_49_transpose_x_0"), val = bool(false)]; + bool attn_49_transpose_y_0 = const()[name = string("attn_49_transpose_y_0"), val = bool(true)]; + tensor attn_49_cast_fp16 = matmul(transpose_x = attn_49_transpose_x_0, transpose_y = attn_49_transpose_y_0, x = var_3078_cast_fp16, y = var_3076_cast_fp16)[name = string("attn_49_cast_fp16")]; + tensor var_3081 = const()[name = string("op_3081"), val = tensor([1, 1280, 1, -1])]; + tensor input_193_cast_fp16 = reshape(shape = var_3081, x = attn_49_cast_fp16)[name = string("input_193_cast_fp16")]; + string obj_99_pad_type_0 = const()[name = string("obj_99_pad_type_0"), val = string("valid")]; + tensor obj_99_strides_0 = const()[name = string("obj_99_strides_0"), val = tensor([1, 1])]; + tensor obj_99_pad_0 = const()[name = string("obj_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_99_dilations_0 = const()[name = string("obj_99_dilations_0"), val = tensor([1, 1])]; + int32 obj_99_groups_0 = const()[name = string("obj_99_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968984000)))]; + tensor layers_24_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972260864)))]; + tensor obj_99_cast_fp16 = conv(bias = layers_24_self_attn_o_proj_bias_to_fp16, dilations = obj_99_dilations_0, groups = obj_99_groups_0, pad = obj_99_pad_0, pad_type = obj_99_pad_type_0, strides = obj_99_strides_0, weight = layers_24_self_attn_o_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("obj_99_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_99_cast_fp16)[name = string("inputs_99_cast_fp16")]; + tensor out_99_axes_0 = const()[name = string("out_99_axes_0"), val = tensor([1])]; + fp16 var_3099_to_fp16 = const()[name = string("op_3099_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3099_to_fp16, x = inputs_99_cast_fp16)[name = string("out_99_cast_fp16")]; + tensor input_195_gamma_0_to_fp16 = const()[name = string("input_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972263488)))]; + tensor input_195_beta_0_to_fp16 = const()[name = string("input_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972266112)))]; + fp16 input_195_epsilon_0_to_fp16 = const()[name = string("input_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = string("input_195_cast_fp16")]; + string input_197_pad_type_0 = const()[name = string("input_197_pad_type_0"), val = string("valid")]; + tensor input_197_strides_0 = const()[name = string("input_197_strides_0"), val = tensor([1, 1])]; + tensor input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_197_dilations_0 = const()[name = string("input_197_dilations_0"), val = tensor([1, 1])]; + int32 input_197_groups_0 = const()[name = string("input_197_groups_0"), val = int32(1)]; + tensor layers_24_fc1_weight_to_fp16 = const()[name = string("layers_24_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972268736)))]; + tensor layers_24_fc1_bias_to_fp16 = const()[name = string("layers_24_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985376000)))]; + tensor input_197_cast_fp16 = conv(bias = layers_24_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_24_fc1_weight_to_fp16, x = input_195_cast_fp16)[name = string("input_197_cast_fp16")]; + string input_199_mode_0 = const()[name = string("input_199_mode_0"), val = string("EXACT")]; + tensor input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")]; + string hidden_states_53_pad_type_0 = const()[name = string("hidden_states_53_pad_type_0"), val = string("valid")]; + tensor hidden_states_53_strides_0 = const()[name = string("hidden_states_53_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_53_pad_0 = const()[name = string("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_53_dilations_0 = const()[name = string("hidden_states_53_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_53_groups_0 = const()[name = string("hidden_states_53_groups_0"), val = int32(1)]; + tensor layers_24_fc2_weight_to_fp16 = const()[name = string("layers_24_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985386304)))]; + tensor layers_24_fc2_bias_to_fp16 = const()[name = string("layers_24_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998493568)))]; + tensor hidden_states_53_cast_fp16 = conv(bias = layers_24_fc2_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = layers_24_fc2_weight_to_fp16, x = input_199_cast_fp16)[name = string("hidden_states_53_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = hidden_states_53_cast_fp16)[name = string("inputs_101_cast_fp16")]; + int32 var_3132 = const()[name = string("op_3132"), val = int32(3)]; + tensor out_101_axes_0 = const()[name = string("out_101_axes_0"), val = tensor([1])]; + fp16 var_3151_to_fp16 = const()[name = string("op_3151_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3151_to_fp16, x = inputs_101_cast_fp16)[name = string("out_101_cast_fp16")]; + tensor obj_101_gamma_0_to_fp16 = const()[name = string("obj_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998496192)))]; + tensor obj_101_beta_0_to_fp16 = const()[name = string("obj_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998498816)))]; + fp16 obj_101_epsilon_0_to_fp16 = const()[name = string("obj_101_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_101_cast_fp16 = batch_norm(beta = obj_101_beta_0_to_fp16, epsilon = obj_101_epsilon_0_to_fp16, gamma = obj_101_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = string("obj_101_cast_fp16")]; + string query_51_pad_type_0 = const()[name = string("query_51_pad_type_0"), val = string("valid")]; + tensor query_51_strides_0 = const()[name = string("query_51_strides_0"), val = tensor([1, 1])]; + tensor query_51_pad_0 = const()[name = string("query_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_51_dilations_0 = const()[name = string("query_51_dilations_0"), val = tensor([1, 1])]; + int32 query_51_groups_0 = const()[name = string("query_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998501440)))]; + tensor layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001778304)))]; + tensor query_51_cast_fp16 = conv(bias = layers_25_self_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_25_self_attn_q_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("query_51_cast_fp16")]; + string key_51_pad_type_0 = const()[name = string("key_51_pad_type_0"), val = string("valid")]; + tensor key_51_strides_0 = const()[name = string("key_51_strides_0"), val = tensor([1, 1])]; + tensor key_51_pad_0 = const()[name = string("key_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_51_dilations_0 = const()[name = string("key_51_dilations_0"), val = tensor([1, 1])]; + int32 key_51_groups_0 = const()[name = string("key_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001780928)))]; + tensor key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_25_self_attn_k_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("key_51_cast_fp16")]; + string value_51_pad_type_0 = const()[name = string("value_51_pad_type_0"), val = string("valid")]; + tensor value_51_strides_0 = const()[name = string("value_51_strides_0"), val = tensor([1, 1])]; + tensor value_51_pad_0 = const()[name = string("value_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_51_dilations_0 = const()[name = string("value_51_dilations_0"), val = tensor([1, 1])]; + int32 value_51_groups_0 = const()[name = string("value_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1005057792)))]; + tensor layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008334656)))]; + tensor value_51_cast_fp16 = conv(bias = layers_25_self_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_25_self_attn_v_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("value_51_cast_fp16")]; + tensor var_3186 = const()[name = string("op_3186"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_51_cast_fp16 = reshape(shape = var_3186, x = query_51_cast_fp16)[name = string("mh_q_51_cast_fp16")]; + fp16 var_3188_to_fp16 = const()[name = string("op_3188_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3189_cast_fp16 = mul(x = mh_q_51_cast_fp16, y = var_3188_to_fp16)[name = string("op_3189_cast_fp16")]; + tensor var_3190 = const()[name = string("op_3190"), val = tensor([1, 20, 64, -1])]; + tensor var_3191_cast_fp16 = reshape(shape = var_3190, x = key_51_cast_fp16)[name = string("op_3191_cast_fp16")]; + bool mh_w_51_transpose_x_0 = const()[name = string("mh_w_51_transpose_x_0"), val = bool(true)]; + bool mh_w_51_transpose_y_0 = const()[name = string("mh_w_51_transpose_y_0"), val = bool(false)]; + tensor mh_w_51_cast_fp16 = matmul(transpose_x = mh_w_51_transpose_x_0, transpose_y = mh_w_51_transpose_y_0, x = var_3189_cast_fp16, y = var_3191_cast_fp16)[name = string("mh_w_51_cast_fp16")]; + tensor var_3194_cast_fp16 = softmax(axis = var_3132, x = mh_w_51_cast_fp16)[name = string("op_3194_cast_fp16")]; + tensor var_3195 = const()[name = string("op_3195"), val = tensor([1, 20, 64, -1])]; + tensor var_3196_cast_fp16 = reshape(shape = var_3195, x = value_51_cast_fp16)[name = string("op_3196_cast_fp16")]; + bool attn_51_transpose_x_0 = const()[name = string("attn_51_transpose_x_0"), val = bool(false)]; + bool attn_51_transpose_y_0 = const()[name = string("attn_51_transpose_y_0"), val = bool(true)]; + tensor attn_51_cast_fp16 = matmul(transpose_x = attn_51_transpose_x_0, transpose_y = attn_51_transpose_y_0, x = var_3196_cast_fp16, y = var_3194_cast_fp16)[name = string("attn_51_cast_fp16")]; + tensor var_3199 = const()[name = string("op_3199"), val = tensor([1, 1280, 1, -1])]; + tensor input_201_cast_fp16 = reshape(shape = var_3199, x = attn_51_cast_fp16)[name = string("input_201_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008337280)))]; + tensor layers_25_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011614144)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_25_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_25_self_attn_o_proj_weight_to_fp16, x = input_201_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_103_cast_fp16")]; + tensor out_103_axes_0 = const()[name = string("out_103_axes_0"), val = tensor([1])]; + fp16 var_3217_to_fp16 = const()[name = string("op_3217_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_3217_to_fp16, x = inputs_103_cast_fp16)[name = string("out_103_cast_fp16")]; + tensor input_203_gamma_0_to_fp16 = const()[name = string("input_203_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011616768)))]; + tensor input_203_beta_0_to_fp16 = const()[name = string("input_203_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011619392)))]; + fp16 input_203_epsilon_0_to_fp16 = const()[name = string("input_203_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_203_cast_fp16 = batch_norm(beta = input_203_beta_0_to_fp16, epsilon = input_203_epsilon_0_to_fp16, gamma = input_203_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = string("input_203_cast_fp16")]; + string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")]; + tensor input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor([1, 1])]; + tensor input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor([1, 1])]; + int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1)]; + tensor layers_25_fc1_weight_to_fp16 = const()[name = string("layers_25_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011622016)))]; + tensor layers_25_fc1_bias_to_fp16 = const()[name = string("layers_25_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024729280)))]; + tensor input_205_cast_fp16 = conv(bias = layers_25_fc1_bias_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = layers_25_fc1_weight_to_fp16, x = input_203_cast_fp16)[name = string("input_205_cast_fp16")]; + string input_207_mode_0 = const()[name = string("input_207_mode_0"), val = string("EXACT")]; + tensor input_207_cast_fp16 = gelu(mode = input_207_mode_0, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")]; + string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")]; + tensor hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)]; + tensor layers_25_fc2_weight_to_fp16 = const()[name = string("layers_25_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024739584)))]; + tensor layers_25_fc2_bias_to_fp16 = const()[name = string("layers_25_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037846848)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = layers_25_fc2_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = layers_25_fc2_weight_to_fp16, x = input_207_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = hidden_states_55_cast_fp16)[name = string("inputs_105_cast_fp16")]; + int32 var_3250 = const()[name = string("op_3250"), val = int32(3)]; + tensor out_105_axes_0 = const()[name = string("out_105_axes_0"), val = tensor([1])]; + fp16 var_3269_to_fp16 = const()[name = string("op_3269_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_3269_to_fp16, x = inputs_105_cast_fp16)[name = string("out_105_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037849472)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037852096)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_53_pad_type_0 = const()[name = string("query_53_pad_type_0"), val = string("valid")]; + tensor query_53_strides_0 = const()[name = string("query_53_strides_0"), val = tensor([1, 1])]; + tensor query_53_pad_0 = const()[name = string("query_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_53_dilations_0 = const()[name = string("query_53_dilations_0"), val = tensor([1, 1])]; + int32 query_53_groups_0 = const()[name = string("query_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037854720)))]; + tensor layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041131584)))]; + tensor query_53_cast_fp16 = conv(bias = layers_26_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_26_self_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_53_cast_fp16")]; + string key_53_pad_type_0 = const()[name = string("key_53_pad_type_0"), val = string("valid")]; + tensor key_53_strides_0 = const()[name = string("key_53_strides_0"), val = tensor([1, 1])]; + tensor key_53_pad_0 = const()[name = string("key_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_53_dilations_0 = const()[name = string("key_53_dilations_0"), val = tensor([1, 1])]; + int32 key_53_groups_0 = const()[name = string("key_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041134208)))]; + tensor key_53_cast_fp16 = conv(dilations = key_53_dilations_0, groups = key_53_groups_0, pad = key_53_pad_0, pad_type = key_53_pad_type_0, strides = key_53_strides_0, weight = layers_26_self_attn_k_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("key_53_cast_fp16")]; + string value_53_pad_type_0 = const()[name = string("value_53_pad_type_0"), val = string("valid")]; + tensor value_53_strides_0 = const()[name = string("value_53_strides_0"), val = tensor([1, 1])]; + tensor value_53_pad_0 = const()[name = string("value_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_53_dilations_0 = const()[name = string("value_53_dilations_0"), val = tensor([1, 1])]; + int32 value_53_groups_0 = const()[name = string("value_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1044411072)))]; + tensor layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047687936)))]; + tensor value_53_cast_fp16 = conv(bias = layers_26_self_attn_v_proj_bias_to_fp16, dilations = value_53_dilations_0, groups = value_53_groups_0, pad = value_53_pad_0, pad_type = value_53_pad_type_0, strides = value_53_strides_0, weight = layers_26_self_attn_v_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("value_53_cast_fp16")]; + tensor var_3304 = const()[name = string("op_3304"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_53_cast_fp16 = reshape(shape = var_3304, x = query_53_cast_fp16)[name = string("mh_q_53_cast_fp16")]; + fp16 var_3306_to_fp16 = const()[name = string("op_3306_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3307_cast_fp16 = mul(x = mh_q_53_cast_fp16, y = var_3306_to_fp16)[name = string("op_3307_cast_fp16")]; + tensor var_3308 = const()[name = string("op_3308"), val = tensor([1, 20, 64, -1])]; + tensor var_3309_cast_fp16 = reshape(shape = var_3308, x = key_53_cast_fp16)[name = string("op_3309_cast_fp16")]; + bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)]; + bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)]; + tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_3307_cast_fp16, y = var_3309_cast_fp16)[name = string("mh_w_53_cast_fp16")]; + tensor var_3312_cast_fp16 = softmax(axis = var_3250, x = mh_w_53_cast_fp16)[name = string("op_3312_cast_fp16")]; + tensor var_3313 = const()[name = string("op_3313"), val = tensor([1, 20, 64, -1])]; + tensor var_3314_cast_fp16 = reshape(shape = var_3313, x = value_53_cast_fp16)[name = string("op_3314_cast_fp16")]; + bool attn_53_transpose_x_0 = const()[name = string("attn_53_transpose_x_0"), val = bool(false)]; + bool attn_53_transpose_y_0 = const()[name = string("attn_53_transpose_y_0"), val = bool(true)]; + tensor attn_53_cast_fp16 = matmul(transpose_x = attn_53_transpose_x_0, transpose_y = attn_53_transpose_y_0, x = var_3314_cast_fp16, y = var_3312_cast_fp16)[name = string("attn_53_cast_fp16")]; + tensor var_3317 = const()[name = string("op_3317"), val = tensor([1, 1280, 1, -1])]; + tensor input_209_cast_fp16 = reshape(shape = var_3317, x = attn_53_cast_fp16)[name = string("input_209_cast_fp16")]; + string obj_107_pad_type_0 = const()[name = string("obj_107_pad_type_0"), val = string("valid")]; + tensor obj_107_strides_0 = const()[name = string("obj_107_strides_0"), val = tensor([1, 1])]; + tensor obj_107_pad_0 = const()[name = string("obj_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_107_dilations_0 = const()[name = string("obj_107_dilations_0"), val = tensor([1, 1])]; + int32 obj_107_groups_0 = const()[name = string("obj_107_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047690560)))]; + tensor layers_26_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050967424)))]; + tensor obj_107_cast_fp16 = conv(bias = layers_26_self_attn_o_proj_bias_to_fp16, dilations = obj_107_dilations_0, groups = obj_107_groups_0, pad = obj_107_pad_0, pad_type = obj_107_pad_type_0, strides = obj_107_strides_0, weight = layers_26_self_attn_o_proj_weight_to_fp16, x = input_209_cast_fp16)[name = string("obj_107_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_107_cast_fp16)[name = string("inputs_107_cast_fp16")]; + tensor out_107_axes_0 = const()[name = string("out_107_axes_0"), val = tensor([1])]; + fp16 var_3335_to_fp16 = const()[name = string("op_3335_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_3335_to_fp16, x = inputs_107_cast_fp16)[name = string("out_107_cast_fp16")]; + tensor input_211_gamma_0_to_fp16 = const()[name = string("input_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050970048)))]; + tensor input_211_beta_0_to_fp16 = const()[name = string("input_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050972672)))]; + fp16 input_211_epsilon_0_to_fp16 = const()[name = string("input_211_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_211_cast_fp16 = batch_norm(beta = input_211_beta_0_to_fp16, epsilon = input_211_epsilon_0_to_fp16, gamma = input_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = string("input_211_cast_fp16")]; + string input_213_pad_type_0 = const()[name = string("input_213_pad_type_0"), val = string("valid")]; + tensor input_213_strides_0 = const()[name = string("input_213_strides_0"), val = tensor([1, 1])]; + tensor input_213_pad_0 = const()[name = string("input_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_213_dilations_0 = const()[name = string("input_213_dilations_0"), val = tensor([1, 1])]; + int32 input_213_groups_0 = const()[name = string("input_213_groups_0"), val = int32(1)]; + tensor layers_26_fc1_weight_to_fp16 = const()[name = string("layers_26_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050975296)))]; + tensor layers_26_fc1_bias_to_fp16 = const()[name = string("layers_26_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064082560)))]; + tensor input_213_cast_fp16 = conv(bias = layers_26_fc1_bias_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = layers_26_fc1_weight_to_fp16, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")]; + string input_215_mode_0 = const()[name = string("input_215_mode_0"), val = string("EXACT")]; + tensor input_215_cast_fp16 = gelu(mode = input_215_mode_0, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")]; + string hidden_states_57_pad_type_0 = const()[name = string("hidden_states_57_pad_type_0"), val = string("valid")]; + tensor hidden_states_57_strides_0 = const()[name = string("hidden_states_57_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_57_pad_0 = const()[name = string("hidden_states_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_57_dilations_0 = const()[name = string("hidden_states_57_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_57_groups_0 = const()[name = string("hidden_states_57_groups_0"), val = int32(1)]; + tensor layers_26_fc2_weight_to_fp16 = const()[name = string("layers_26_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064092864)))]; + tensor layers_26_fc2_bias_to_fp16 = const()[name = string("layers_26_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077200128)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = layers_26_fc2_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = layers_26_fc2_weight_to_fp16, x = input_215_cast_fp16)[name = string("hidden_states_57_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_57_cast_fp16)[name = string("inputs_109_cast_fp16")]; + int32 var_3368 = const()[name = string("op_3368"), val = int32(3)]; + tensor out_109_axes_0 = const()[name = string("out_109_axes_0"), val = tensor([1])]; + fp16 var_3387_to_fp16 = const()[name = string("op_3387_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_3387_to_fp16, x = inputs_109_cast_fp16)[name = string("out_109_cast_fp16")]; + tensor obj_109_gamma_0_to_fp16 = const()[name = string("obj_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077202752)))]; + tensor obj_109_beta_0_to_fp16 = const()[name = string("obj_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077205376)))]; + fp16 obj_109_epsilon_0_to_fp16 = const()[name = string("obj_109_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_109_cast_fp16 = batch_norm(beta = obj_109_beta_0_to_fp16, epsilon = obj_109_epsilon_0_to_fp16, gamma = obj_109_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = string("obj_109_cast_fp16")]; + string query_55_pad_type_0 = const()[name = string("query_55_pad_type_0"), val = string("valid")]; + tensor query_55_strides_0 = const()[name = string("query_55_strides_0"), val = tensor([1, 1])]; + tensor query_55_pad_0 = const()[name = string("query_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_55_dilations_0 = const()[name = string("query_55_dilations_0"), val = tensor([1, 1])]; + int32 query_55_groups_0 = const()[name = string("query_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077208000)))]; + tensor layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080484864)))]; + tensor query_55_cast_fp16 = conv(bias = layers_27_self_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_27_self_attn_q_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("query_55_cast_fp16")]; + string key_55_pad_type_0 = const()[name = string("key_55_pad_type_0"), val = string("valid")]; + tensor key_55_strides_0 = const()[name = string("key_55_strides_0"), val = tensor([1, 1])]; + tensor key_55_pad_0 = const()[name = string("key_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_55_dilations_0 = const()[name = string("key_55_dilations_0"), val = tensor([1, 1])]; + int32 key_55_groups_0 = const()[name = string("key_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080487488)))]; + tensor key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_27_self_attn_k_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("key_55_cast_fp16")]; + string value_55_pad_type_0 = const()[name = string("value_55_pad_type_0"), val = string("valid")]; + tensor value_55_strides_0 = const()[name = string("value_55_strides_0"), val = tensor([1, 1])]; + tensor value_55_pad_0 = const()[name = string("value_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_55_dilations_0 = const()[name = string("value_55_dilations_0"), val = tensor([1, 1])]; + int32 value_55_groups_0 = const()[name = string("value_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1083764352)))]; + tensor layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087041216)))]; + tensor value_55_cast_fp16 = conv(bias = layers_27_self_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_27_self_attn_v_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("value_55_cast_fp16")]; + tensor var_3422 = const()[name = string("op_3422"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_55_cast_fp16 = reshape(shape = var_3422, x = query_55_cast_fp16)[name = string("mh_q_55_cast_fp16")]; + fp16 var_3424_to_fp16 = const()[name = string("op_3424_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3425_cast_fp16 = mul(x = mh_q_55_cast_fp16, y = var_3424_to_fp16)[name = string("op_3425_cast_fp16")]; + tensor var_3426 = const()[name = string("op_3426"), val = tensor([1, 20, 64, -1])]; + tensor var_3427_cast_fp16 = reshape(shape = var_3426, x = key_55_cast_fp16)[name = string("op_3427_cast_fp16")]; + bool mh_w_55_transpose_x_0 = const()[name = string("mh_w_55_transpose_x_0"), val = bool(true)]; + bool mh_w_55_transpose_y_0 = const()[name = string("mh_w_55_transpose_y_0"), val = bool(false)]; + tensor mh_w_55_cast_fp16 = matmul(transpose_x = mh_w_55_transpose_x_0, transpose_y = mh_w_55_transpose_y_0, x = var_3425_cast_fp16, y = var_3427_cast_fp16)[name = string("mh_w_55_cast_fp16")]; + tensor var_3430_cast_fp16 = softmax(axis = var_3368, x = mh_w_55_cast_fp16)[name = string("op_3430_cast_fp16")]; + tensor var_3431 = const()[name = string("op_3431"), val = tensor([1, 20, 64, -1])]; + tensor var_3432_cast_fp16 = reshape(shape = var_3431, x = value_55_cast_fp16)[name = string("op_3432_cast_fp16")]; + bool attn_55_transpose_x_0 = const()[name = string("attn_55_transpose_x_0"), val = bool(false)]; + bool attn_55_transpose_y_0 = const()[name = string("attn_55_transpose_y_0"), val = bool(true)]; + tensor attn_55_cast_fp16 = matmul(transpose_x = attn_55_transpose_x_0, transpose_y = attn_55_transpose_y_0, x = var_3432_cast_fp16, y = var_3430_cast_fp16)[name = string("attn_55_cast_fp16")]; + tensor var_3435 = const()[name = string("op_3435"), val = tensor([1, 1280, 1, -1])]; + tensor input_217_cast_fp16 = reshape(shape = var_3435, x = attn_55_cast_fp16)[name = string("input_217_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087043840)))]; + tensor layers_27_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090320704)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_27_self_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_27_self_attn_o_proj_weight_to_fp16, x = input_217_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_111_cast_fp16")]; + tensor out_111_axes_0 = const()[name = string("out_111_axes_0"), val = tensor([1])]; + fp16 var_3453_to_fp16 = const()[name = string("op_3453_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_3453_to_fp16, x = inputs_111_cast_fp16)[name = string("out_111_cast_fp16")]; + tensor input_219_gamma_0_to_fp16 = const()[name = string("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090323328)))]; + tensor input_219_beta_0_to_fp16 = const()[name = string("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090325952)))]; + fp16 input_219_epsilon_0_to_fp16 = const()[name = string("input_219_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = string("input_219_cast_fp16")]; + string input_221_pad_type_0 = const()[name = string("input_221_pad_type_0"), val = string("valid")]; + tensor input_221_strides_0 = const()[name = string("input_221_strides_0"), val = tensor([1, 1])]; + tensor input_221_pad_0 = const()[name = string("input_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_221_dilations_0 = const()[name = string("input_221_dilations_0"), val = tensor([1, 1])]; + int32 input_221_groups_0 = const()[name = string("input_221_groups_0"), val = int32(1)]; + tensor layers_27_fc1_weight_to_fp16 = const()[name = string("layers_27_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090328576)))]; + tensor layers_27_fc1_bias_to_fp16 = const()[name = string("layers_27_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103435840)))]; + tensor input_221_cast_fp16 = conv(bias = layers_27_fc1_bias_to_fp16, dilations = input_221_dilations_0, groups = input_221_groups_0, pad = input_221_pad_0, pad_type = input_221_pad_type_0, strides = input_221_strides_0, weight = layers_27_fc1_weight_to_fp16, x = input_219_cast_fp16)[name = string("input_221_cast_fp16")]; + string input_223_mode_0 = const()[name = string("input_223_mode_0"), val = string("EXACT")]; + tensor input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")]; + string hidden_states_59_pad_type_0 = const()[name = string("hidden_states_59_pad_type_0"), val = string("valid")]; + tensor hidden_states_59_strides_0 = const()[name = string("hidden_states_59_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_59_pad_0 = const()[name = string("hidden_states_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_59_dilations_0 = const()[name = string("hidden_states_59_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_59_groups_0 = const()[name = string("hidden_states_59_groups_0"), val = int32(1)]; + tensor layers_27_fc2_weight_to_fp16 = const()[name = string("layers_27_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103446144)))]; + tensor layers_27_fc2_bias_to_fp16 = const()[name = string("layers_27_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116553408)))]; + tensor hidden_states_59_cast_fp16 = conv(bias = layers_27_fc2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = layers_27_fc2_weight_to_fp16, x = input_223_cast_fp16)[name = string("hidden_states_59_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = hidden_states_59_cast_fp16)[name = string("inputs_113_cast_fp16")]; + int32 var_3486 = const()[name = string("op_3486"), val = int32(3)]; + tensor out_113_axes_0 = const()[name = string("out_113_axes_0"), val = tensor([1])]; + fp16 var_3505_to_fp16 = const()[name = string("op_3505_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_3505_to_fp16, x = inputs_113_cast_fp16)[name = string("out_113_cast_fp16")]; + tensor obj_113_gamma_0_to_fp16 = const()[name = string("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116556032)))]; + tensor obj_113_beta_0_to_fp16 = const()[name = string("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116558656)))]; + fp16 obj_113_epsilon_0_to_fp16 = const()[name = string("obj_113_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = string("obj_113_cast_fp16")]; + string query_57_pad_type_0 = const()[name = string("query_57_pad_type_0"), val = string("valid")]; + tensor query_57_strides_0 = const()[name = string("query_57_strides_0"), val = tensor([1, 1])]; + tensor query_57_pad_0 = const()[name = string("query_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_57_dilations_0 = const()[name = string("query_57_dilations_0"), val = tensor([1, 1])]; + int32 query_57_groups_0 = const()[name = string("query_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116561280)))]; + tensor layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119838144)))]; + tensor query_57_cast_fp16 = conv(bias = layers_28_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_28_self_attn_q_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("query_57_cast_fp16")]; + string key_57_pad_type_0 = const()[name = string("key_57_pad_type_0"), val = string("valid")]; + tensor key_57_strides_0 = const()[name = string("key_57_strides_0"), val = tensor([1, 1])]; + tensor key_57_pad_0 = const()[name = string("key_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_57_dilations_0 = const()[name = string("key_57_dilations_0"), val = tensor([1, 1])]; + int32 key_57_groups_0 = const()[name = string("key_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119840768)))]; + tensor key_57_cast_fp16 = conv(dilations = key_57_dilations_0, groups = key_57_groups_0, pad = key_57_pad_0, pad_type = key_57_pad_type_0, strides = key_57_strides_0, weight = layers_28_self_attn_k_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("key_57_cast_fp16")]; + string value_57_pad_type_0 = const()[name = string("value_57_pad_type_0"), val = string("valid")]; + tensor value_57_strides_0 = const()[name = string("value_57_strides_0"), val = tensor([1, 1])]; + tensor value_57_pad_0 = const()[name = string("value_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_57_dilations_0 = const()[name = string("value_57_dilations_0"), val = tensor([1, 1])]; + int32 value_57_groups_0 = const()[name = string("value_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1123117632)))]; + tensor layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126394496)))]; + tensor value_57_cast_fp16 = conv(bias = layers_28_self_attn_v_proj_bias_to_fp16, dilations = value_57_dilations_0, groups = value_57_groups_0, pad = value_57_pad_0, pad_type = value_57_pad_type_0, strides = value_57_strides_0, weight = layers_28_self_attn_v_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("value_57_cast_fp16")]; + tensor var_3540 = const()[name = string("op_3540"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_57_cast_fp16 = reshape(shape = var_3540, x = query_57_cast_fp16)[name = string("mh_q_57_cast_fp16")]; + fp16 var_3542_to_fp16 = const()[name = string("op_3542_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3543_cast_fp16 = mul(x = mh_q_57_cast_fp16, y = var_3542_to_fp16)[name = string("op_3543_cast_fp16")]; + tensor var_3544 = const()[name = string("op_3544"), val = tensor([1, 20, 64, -1])]; + tensor var_3545_cast_fp16 = reshape(shape = var_3544, x = key_57_cast_fp16)[name = string("op_3545_cast_fp16")]; + bool mh_w_57_transpose_x_0 = const()[name = string("mh_w_57_transpose_x_0"), val = bool(true)]; + bool mh_w_57_transpose_y_0 = const()[name = string("mh_w_57_transpose_y_0"), val = bool(false)]; + tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_3543_cast_fp16, y = var_3545_cast_fp16)[name = string("mh_w_57_cast_fp16")]; + tensor var_3548_cast_fp16 = softmax(axis = var_3486, x = mh_w_57_cast_fp16)[name = string("op_3548_cast_fp16")]; + tensor var_3549 = const()[name = string("op_3549"), val = tensor([1, 20, 64, -1])]; + tensor var_3550_cast_fp16 = reshape(shape = var_3549, x = value_57_cast_fp16)[name = string("op_3550_cast_fp16")]; + bool attn_57_transpose_x_0 = const()[name = string("attn_57_transpose_x_0"), val = bool(false)]; + bool attn_57_transpose_y_0 = const()[name = string("attn_57_transpose_y_0"), val = bool(true)]; + tensor attn_57_cast_fp16 = matmul(transpose_x = attn_57_transpose_x_0, transpose_y = attn_57_transpose_y_0, x = var_3550_cast_fp16, y = var_3548_cast_fp16)[name = string("attn_57_cast_fp16")]; + tensor var_3553 = const()[name = string("op_3553"), val = tensor([1, 1280, 1, -1])]; + tensor input_225_cast_fp16 = reshape(shape = var_3553, x = attn_57_cast_fp16)[name = string("input_225_cast_fp16")]; + string obj_115_pad_type_0 = const()[name = string("obj_115_pad_type_0"), val = string("valid")]; + tensor obj_115_strides_0 = const()[name = string("obj_115_strides_0"), val = tensor([1, 1])]; + tensor obj_115_pad_0 = const()[name = string("obj_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_115_dilations_0 = const()[name = string("obj_115_dilations_0"), val = tensor([1, 1])]; + int32 obj_115_groups_0 = const()[name = string("obj_115_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126397120)))]; + tensor layers_28_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129673984)))]; + tensor obj_115_cast_fp16 = conv(bias = layers_28_self_attn_o_proj_bias_to_fp16, dilations = obj_115_dilations_0, groups = obj_115_groups_0, pad = obj_115_pad_0, pad_type = obj_115_pad_type_0, strides = obj_115_strides_0, weight = layers_28_self_attn_o_proj_weight_to_fp16, x = input_225_cast_fp16)[name = string("obj_115_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_115_cast_fp16)[name = string("inputs_115_cast_fp16")]; + tensor out_115_axes_0 = const()[name = string("out_115_axes_0"), val = tensor([1])]; + fp16 var_3571_to_fp16 = const()[name = string("op_3571_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_3571_to_fp16, x = inputs_115_cast_fp16)[name = string("out_115_cast_fp16")]; + tensor input_227_gamma_0_to_fp16 = const()[name = string("input_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129676608)))]; + tensor input_227_beta_0_to_fp16 = const()[name = string("input_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129679232)))]; + fp16 input_227_epsilon_0_to_fp16 = const()[name = string("input_227_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_227_cast_fp16 = batch_norm(beta = input_227_beta_0_to_fp16, epsilon = input_227_epsilon_0_to_fp16, gamma = input_227_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = string("input_227_cast_fp16")]; + string input_229_pad_type_0 = const()[name = string("input_229_pad_type_0"), val = string("valid")]; + tensor input_229_strides_0 = const()[name = string("input_229_strides_0"), val = tensor([1, 1])]; + tensor input_229_pad_0 = const()[name = string("input_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_229_dilations_0 = const()[name = string("input_229_dilations_0"), val = tensor([1, 1])]; + int32 input_229_groups_0 = const()[name = string("input_229_groups_0"), val = int32(1)]; + tensor layers_28_fc1_weight_to_fp16 = const()[name = string("layers_28_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129681856)))]; + tensor layers_28_fc1_bias_to_fp16 = const()[name = string("layers_28_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142789120)))]; + tensor input_229_cast_fp16 = conv(bias = layers_28_fc1_bias_to_fp16, dilations = input_229_dilations_0, groups = input_229_groups_0, pad = input_229_pad_0, pad_type = input_229_pad_type_0, strides = input_229_strides_0, weight = layers_28_fc1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")]; + string input_231_mode_0 = const()[name = string("input_231_mode_0"), val = string("EXACT")]; + tensor input_231_cast_fp16 = gelu(mode = input_231_mode_0, x = input_229_cast_fp16)[name = string("input_231_cast_fp16")]; + string hidden_states_61_pad_type_0 = const()[name = string("hidden_states_61_pad_type_0"), val = string("valid")]; + tensor hidden_states_61_strides_0 = const()[name = string("hidden_states_61_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_61_pad_0 = const()[name = string("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_61_dilations_0 = const()[name = string("hidden_states_61_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_61_groups_0 = const()[name = string("hidden_states_61_groups_0"), val = int32(1)]; + tensor layers_28_fc2_weight_to_fp16 = const()[name = string("layers_28_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142799424)))]; + tensor layers_28_fc2_bias_to_fp16 = const()[name = string("layers_28_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155906688)))]; + tensor hidden_states_61_cast_fp16 = conv(bias = layers_28_fc2_bias_to_fp16, dilations = hidden_states_61_dilations_0, groups = hidden_states_61_groups_0, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = hidden_states_61_strides_0, weight = layers_28_fc2_weight_to_fp16, x = input_231_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = hidden_states_61_cast_fp16)[name = string("inputs_117_cast_fp16")]; + int32 var_3604 = const()[name = string("op_3604"), val = int32(3)]; + tensor out_117_axes_0 = const()[name = string("out_117_axes_0"), val = tensor([1])]; + fp16 var_3623_to_fp16 = const()[name = string("op_3623_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_3623_to_fp16, x = inputs_117_cast_fp16)[name = string("out_117_cast_fp16")]; + tensor obj_117_gamma_0_to_fp16 = const()[name = string("obj_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155909312)))]; + tensor obj_117_beta_0_to_fp16 = const()[name = string("obj_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155911936)))]; + fp16 obj_117_epsilon_0_to_fp16 = const()[name = string("obj_117_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_117_cast_fp16 = batch_norm(beta = obj_117_beta_0_to_fp16, epsilon = obj_117_epsilon_0_to_fp16, gamma = obj_117_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = string("obj_117_cast_fp16")]; + string query_59_pad_type_0 = const()[name = string("query_59_pad_type_0"), val = string("valid")]; + tensor query_59_strides_0 = const()[name = string("query_59_strides_0"), val = tensor([1, 1])]; + tensor query_59_pad_0 = const()[name = string("query_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_59_dilations_0 = const()[name = string("query_59_dilations_0"), val = tensor([1, 1])]; + int32 query_59_groups_0 = const()[name = string("query_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155914560)))]; + tensor layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159191424)))]; + tensor query_59_cast_fp16 = conv(bias = layers_29_self_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_29_self_attn_q_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("query_59_cast_fp16")]; + string key_59_pad_type_0 = const()[name = string("key_59_pad_type_0"), val = string("valid")]; + tensor key_59_strides_0 = const()[name = string("key_59_strides_0"), val = tensor([1, 1])]; + tensor key_59_pad_0 = const()[name = string("key_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_59_dilations_0 = const()[name = string("key_59_dilations_0"), val = tensor([1, 1])]; + int32 key_59_groups_0 = const()[name = string("key_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159194048)))]; + tensor key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_29_self_attn_k_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("key_59_cast_fp16")]; + string value_59_pad_type_0 = const()[name = string("value_59_pad_type_0"), val = string("valid")]; + tensor value_59_strides_0 = const()[name = string("value_59_strides_0"), val = tensor([1, 1])]; + tensor value_59_pad_0 = const()[name = string("value_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_59_dilations_0 = const()[name = string("value_59_dilations_0"), val = tensor([1, 1])]; + int32 value_59_groups_0 = const()[name = string("value_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1162470912)))]; + tensor layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165747776)))]; + tensor value_59_cast_fp16 = conv(bias = layers_29_self_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_29_self_attn_v_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("value_59_cast_fp16")]; + tensor var_3658 = const()[name = string("op_3658"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_59_cast_fp16 = reshape(shape = var_3658, x = query_59_cast_fp16)[name = string("mh_q_59_cast_fp16")]; + fp16 var_3660_to_fp16 = const()[name = string("op_3660_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3661_cast_fp16 = mul(x = mh_q_59_cast_fp16, y = var_3660_to_fp16)[name = string("op_3661_cast_fp16")]; + tensor var_3662 = const()[name = string("op_3662"), val = tensor([1, 20, 64, -1])]; + tensor var_3663_cast_fp16 = reshape(shape = var_3662, x = key_59_cast_fp16)[name = string("op_3663_cast_fp16")]; + bool mh_w_59_transpose_x_0 = const()[name = string("mh_w_59_transpose_x_0"), val = bool(true)]; + bool mh_w_59_transpose_y_0 = const()[name = string("mh_w_59_transpose_y_0"), val = bool(false)]; + tensor mh_w_59_cast_fp16 = matmul(transpose_x = mh_w_59_transpose_x_0, transpose_y = mh_w_59_transpose_y_0, x = var_3661_cast_fp16, y = var_3663_cast_fp16)[name = string("mh_w_59_cast_fp16")]; + tensor var_3666_cast_fp16 = softmax(axis = var_3604, x = mh_w_59_cast_fp16)[name = string("op_3666_cast_fp16")]; + tensor var_3667 = const()[name = string("op_3667"), val = tensor([1, 20, 64, -1])]; + tensor var_3668_cast_fp16 = reshape(shape = var_3667, x = value_59_cast_fp16)[name = string("op_3668_cast_fp16")]; + bool attn_59_transpose_x_0 = const()[name = string("attn_59_transpose_x_0"), val = bool(false)]; + bool attn_59_transpose_y_0 = const()[name = string("attn_59_transpose_y_0"), val = bool(true)]; + tensor attn_59_cast_fp16 = matmul(transpose_x = attn_59_transpose_x_0, transpose_y = attn_59_transpose_y_0, x = var_3668_cast_fp16, y = var_3666_cast_fp16)[name = string("attn_59_cast_fp16")]; + tensor var_3671 = const()[name = string("op_3671"), val = tensor([1, 1280, 1, -1])]; + tensor input_233_cast_fp16 = reshape(shape = var_3671, x = attn_59_cast_fp16)[name = string("input_233_cast_fp16")]; + string obj_119_pad_type_0 = const()[name = string("obj_119_pad_type_0"), val = string("valid")]; + tensor obj_119_strides_0 = const()[name = string("obj_119_strides_0"), val = tensor([1, 1])]; + tensor obj_119_pad_0 = const()[name = string("obj_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_119_dilations_0 = const()[name = string("obj_119_dilations_0"), val = tensor([1, 1])]; + int32 obj_119_groups_0 = const()[name = string("obj_119_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165750400)))]; + tensor layers_29_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169027264)))]; + tensor obj_119_cast_fp16 = conv(bias = layers_29_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_29_self_attn_o_proj_weight_to_fp16, x = input_233_cast_fp16)[name = string("obj_119_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_119_cast_fp16)[name = string("inputs_119_cast_fp16")]; + tensor out_119_axes_0 = const()[name = string("out_119_axes_0"), val = tensor([1])]; + fp16 var_3689_to_fp16 = const()[name = string("op_3689_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_3689_to_fp16, x = inputs_119_cast_fp16)[name = string("out_119_cast_fp16")]; + tensor input_235_gamma_0_to_fp16 = const()[name = string("input_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169029888)))]; + tensor input_235_beta_0_to_fp16 = const()[name = string("input_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169032512)))]; + fp16 input_235_epsilon_0_to_fp16 = const()[name = string("input_235_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = string("input_235_cast_fp16")]; + string input_237_pad_type_0 = const()[name = string("input_237_pad_type_0"), val = string("valid")]; + tensor input_237_strides_0 = const()[name = string("input_237_strides_0"), val = tensor([1, 1])]; + tensor input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_237_dilations_0 = const()[name = string("input_237_dilations_0"), val = tensor([1, 1])]; + int32 input_237_groups_0 = const()[name = string("input_237_groups_0"), val = int32(1)]; + tensor layers_29_fc1_weight_to_fp16 = const()[name = string("layers_29_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169035136)))]; + tensor layers_29_fc1_bias_to_fp16 = const()[name = string("layers_29_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182142400)))]; + tensor input_237_cast_fp16 = conv(bias = layers_29_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_29_fc1_weight_to_fp16, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")]; + string input_239_mode_0 = const()[name = string("input_239_mode_0"), val = string("EXACT")]; + tensor input_239_cast_fp16 = gelu(mode = input_239_mode_0, x = input_237_cast_fp16)[name = string("input_239_cast_fp16")]; + string hidden_states_63_pad_type_0 = const()[name = string("hidden_states_63_pad_type_0"), val = string("valid")]; + tensor hidden_states_63_strides_0 = const()[name = string("hidden_states_63_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_63_pad_0 = const()[name = string("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_63_dilations_0 = const()[name = string("hidden_states_63_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_63_groups_0 = const()[name = string("hidden_states_63_groups_0"), val = int32(1)]; + tensor layers_29_fc2_weight_to_fp16 = const()[name = string("layers_29_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182152704)))]; + tensor layers_29_fc2_bias_to_fp16 = const()[name = string("layers_29_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195259968)))]; + tensor hidden_states_63_cast_fp16 = conv(bias = layers_29_fc2_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = layers_29_fc2_weight_to_fp16, x = input_239_cast_fp16)[name = string("hidden_states_63_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_63_cast_fp16)[name = string("inputs_121_cast_fp16")]; + int32 var_3722 = const()[name = string("op_3722"), val = int32(3)]; + tensor out_121_axes_0 = const()[name = string("out_121_axes_0"), val = tensor([1])]; + fp16 var_3741_to_fp16 = const()[name = string("op_3741_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_3741_to_fp16, x = inputs_121_cast_fp16)[name = string("out_121_cast_fp16")]; + tensor obj_121_gamma_0_to_fp16 = const()[name = string("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195262592)))]; + tensor obj_121_beta_0_to_fp16 = const()[name = string("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195265216)))]; + fp16 obj_121_epsilon_0_to_fp16 = const()[name = string("obj_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = string("obj_121_cast_fp16")]; + string query_61_pad_type_0 = const()[name = string("query_61_pad_type_0"), val = string("valid")]; + tensor query_61_strides_0 = const()[name = string("query_61_strides_0"), val = tensor([1, 1])]; + tensor query_61_pad_0 = const()[name = string("query_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_61_dilations_0 = const()[name = string("query_61_dilations_0"), val = tensor([1, 1])]; + int32 query_61_groups_0 = const()[name = string("query_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195267840)))]; + tensor layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198544704)))]; + tensor query_61_cast_fp16 = conv(bias = layers_30_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_30_self_attn_q_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("query_61_cast_fp16")]; + string key_61_pad_type_0 = const()[name = string("key_61_pad_type_0"), val = string("valid")]; + tensor key_61_strides_0 = const()[name = string("key_61_strides_0"), val = tensor([1, 1])]; + tensor key_61_pad_0 = const()[name = string("key_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_61_dilations_0 = const()[name = string("key_61_dilations_0"), val = tensor([1, 1])]; + int32 key_61_groups_0 = const()[name = string("key_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198547328)))]; + tensor key_61_cast_fp16 = conv(dilations = key_61_dilations_0, groups = key_61_groups_0, pad = key_61_pad_0, pad_type = key_61_pad_type_0, strides = key_61_strides_0, weight = layers_30_self_attn_k_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("key_61_cast_fp16")]; + string value_61_pad_type_0 = const()[name = string("value_61_pad_type_0"), val = string("valid")]; + tensor value_61_strides_0 = const()[name = string("value_61_strides_0"), val = tensor([1, 1])]; + tensor value_61_pad_0 = const()[name = string("value_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_61_dilations_0 = const()[name = string("value_61_dilations_0"), val = tensor([1, 1])]; + int32 value_61_groups_0 = const()[name = string("value_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1201824192)))]; + tensor layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205101056)))]; + tensor value_61_cast_fp16 = conv(bias = layers_30_self_attn_v_proj_bias_to_fp16, dilations = value_61_dilations_0, groups = value_61_groups_0, pad = value_61_pad_0, pad_type = value_61_pad_type_0, strides = value_61_strides_0, weight = layers_30_self_attn_v_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("value_61_cast_fp16")]; + tensor var_3776 = const()[name = string("op_3776"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_61_cast_fp16 = reshape(shape = var_3776, x = query_61_cast_fp16)[name = string("mh_q_61_cast_fp16")]; + fp16 var_3778_to_fp16 = const()[name = string("op_3778_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3779_cast_fp16 = mul(x = mh_q_61_cast_fp16, y = var_3778_to_fp16)[name = string("op_3779_cast_fp16")]; + tensor var_3780 = const()[name = string("op_3780"), val = tensor([1, 20, 64, -1])]; + tensor var_3781_cast_fp16 = reshape(shape = var_3780, x = key_61_cast_fp16)[name = string("op_3781_cast_fp16")]; + bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)]; + bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)]; + tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_3779_cast_fp16, y = var_3781_cast_fp16)[name = string("mh_w_61_cast_fp16")]; + tensor var_3784_cast_fp16 = softmax(axis = var_3722, x = mh_w_61_cast_fp16)[name = string("op_3784_cast_fp16")]; + tensor var_3785 = const()[name = string("op_3785"), val = tensor([1, 20, 64, -1])]; + tensor var_3786_cast_fp16 = reshape(shape = var_3785, x = value_61_cast_fp16)[name = string("op_3786_cast_fp16")]; + bool attn_61_transpose_x_0 = const()[name = string("attn_61_transpose_x_0"), val = bool(false)]; + bool attn_61_transpose_y_0 = const()[name = string("attn_61_transpose_y_0"), val = bool(true)]; + tensor attn_61_cast_fp16 = matmul(transpose_x = attn_61_transpose_x_0, transpose_y = attn_61_transpose_y_0, x = var_3786_cast_fp16, y = var_3784_cast_fp16)[name = string("attn_61_cast_fp16")]; + tensor var_3789 = const()[name = string("op_3789"), val = tensor([1, 1280, 1, -1])]; + tensor input_241_cast_fp16 = reshape(shape = var_3789, x = attn_61_cast_fp16)[name = string("input_241_cast_fp16")]; + string obj_123_pad_type_0 = const()[name = string("obj_123_pad_type_0"), val = string("valid")]; + tensor obj_123_strides_0 = const()[name = string("obj_123_strides_0"), val = tensor([1, 1])]; + tensor obj_123_pad_0 = const()[name = string("obj_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_123_dilations_0 = const()[name = string("obj_123_dilations_0"), val = tensor([1, 1])]; + int32 obj_123_groups_0 = const()[name = string("obj_123_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205103680)))]; + tensor layers_30_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208380544)))]; + tensor obj_123_cast_fp16 = conv(bias = layers_30_self_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_30_self_attn_o_proj_weight_to_fp16, x = input_241_cast_fp16)[name = string("obj_123_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_123_cast_fp16)[name = string("inputs_123_cast_fp16")]; + tensor out_123_axes_0 = const()[name = string("out_123_axes_0"), val = tensor([1])]; + fp16 var_3807_to_fp16 = const()[name = string("op_3807_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_3807_to_fp16, x = inputs_123_cast_fp16)[name = string("out_123_cast_fp16")]; + tensor input_243_gamma_0_to_fp16 = const()[name = string("input_243_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208383168)))]; + tensor input_243_beta_0_to_fp16 = const()[name = string("input_243_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208385792)))]; + fp16 input_243_epsilon_0_to_fp16 = const()[name = string("input_243_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_243_cast_fp16 = batch_norm(beta = input_243_beta_0_to_fp16, epsilon = input_243_epsilon_0_to_fp16, gamma = input_243_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = string("input_243_cast_fp16")]; + string input_245_pad_type_0 = const()[name = string("input_245_pad_type_0"), val = string("valid")]; + tensor input_245_strides_0 = const()[name = string("input_245_strides_0"), val = tensor([1, 1])]; + tensor input_245_pad_0 = const()[name = string("input_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_245_dilations_0 = const()[name = string("input_245_dilations_0"), val = tensor([1, 1])]; + int32 input_245_groups_0 = const()[name = string("input_245_groups_0"), val = int32(1)]; + tensor layers_30_fc1_weight_to_fp16 = const()[name = string("layers_30_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208388416)))]; + tensor layers_30_fc1_bias_to_fp16 = const()[name = string("layers_30_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221495680)))]; + tensor input_245_cast_fp16 = conv(bias = layers_30_fc1_bias_to_fp16, dilations = input_245_dilations_0, groups = input_245_groups_0, pad = input_245_pad_0, pad_type = input_245_pad_type_0, strides = input_245_strides_0, weight = layers_30_fc1_weight_to_fp16, x = input_243_cast_fp16)[name = string("input_245_cast_fp16")]; + string input_247_mode_0 = const()[name = string("input_247_mode_0"), val = string("EXACT")]; + tensor input_247_cast_fp16 = gelu(mode = input_247_mode_0, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")]; + string hidden_states_65_pad_type_0 = const()[name = string("hidden_states_65_pad_type_0"), val = string("valid")]; + tensor hidden_states_65_strides_0 = const()[name = string("hidden_states_65_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_65_pad_0 = const()[name = string("hidden_states_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_65_dilations_0 = const()[name = string("hidden_states_65_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_65_groups_0 = const()[name = string("hidden_states_65_groups_0"), val = int32(1)]; + tensor layers_30_fc2_weight_to_fp16 = const()[name = string("layers_30_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221505984)))]; + tensor layers_30_fc2_bias_to_fp16 = const()[name = string("layers_30_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234613248)))]; + tensor hidden_states_65_cast_fp16 = conv(bias = layers_30_fc2_bias_to_fp16, dilations = hidden_states_65_dilations_0, groups = hidden_states_65_groups_0, pad = hidden_states_65_pad_0, pad_type = hidden_states_65_pad_type_0, strides = hidden_states_65_strides_0, weight = layers_30_fc2_weight_to_fp16, x = input_247_cast_fp16)[name = string("hidden_states_65_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = hidden_states_65_cast_fp16)[name = string("inputs_125_cast_fp16")]; + int32 var_3840 = const()[name = string("op_3840"), val = int32(3)]; + tensor out_125_axes_0 = const()[name = string("out_125_axes_0"), val = tensor([1])]; + fp16 var_3859_to_fp16 = const()[name = string("op_3859_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_3859_to_fp16, x = inputs_125_cast_fp16)[name = string("out_125_cast_fp16")]; + tensor obj_125_gamma_0_to_fp16 = const()[name = string("obj_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234615872)))]; + tensor obj_125_beta_0_to_fp16 = const()[name = string("obj_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234618496)))]; + fp16 obj_125_epsilon_0_to_fp16 = const()[name = string("obj_125_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_125_cast_fp16 = batch_norm(beta = obj_125_beta_0_to_fp16, epsilon = obj_125_epsilon_0_to_fp16, gamma = obj_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = string("obj_125_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234621120)))]; + tensor layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237897984)))]; + tensor query_cast_fp16 = conv(bias = layers_31_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_31_self_attn_q_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237900608)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_31_self_attn_k_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1241177472)))]; + tensor layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244454336)))]; + tensor value_cast_fp16 = conv(bias = layers_31_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_31_self_attn_v_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_3894 = const()[name = string("op_3894"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_3894, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_3896_to_fp16 = const()[name = string("op_3896_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3897_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_3896_to_fp16)[name = string("op_3897_cast_fp16")]; + tensor var_3898 = const()[name = string("op_3898"), val = tensor([1, 20, 64, -1])]; + tensor var_3899_cast_fp16 = reshape(shape = var_3898, x = key_cast_fp16)[name = string("op_3899_cast_fp16")]; + bool mh_w_transpose_x_0 = const()[name = string("mh_w_transpose_x_0"), val = bool(true)]; + bool mh_w_transpose_y_0 = const()[name = string("mh_w_transpose_y_0"), val = bool(false)]; + tensor mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_3897_cast_fp16, y = var_3899_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor var_3902_cast_fp16 = softmax(axis = var_3840, x = mh_w_cast_fp16)[name = string("op_3902_cast_fp16")]; + tensor var_3903 = const()[name = string("op_3903"), val = tensor([1, 20, 64, -1])]; + tensor var_3904_cast_fp16 = reshape(shape = var_3903, x = value_cast_fp16)[name = string("op_3904_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_3904_cast_fp16, y = var_3902_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_3907 = const()[name = string("op_3907"), val = tensor([1, 1280, 1, -1])]; + tensor input_249_cast_fp16 = reshape(shape = var_3907, x = attn_cast_fp16)[name = string("input_249_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244456960)))]; + tensor layers_31_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247733824)))]; + tensor obj_cast_fp16 = conv(bias = layers_31_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_31_self_attn_o_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = obj_cast_fp16)[name = string("inputs_127_cast_fp16")]; + tensor out_127_axes_0 = const()[name = string("out_127_axes_0"), val = tensor([1])]; + fp16 var_3925_to_fp16 = const()[name = string("op_3925_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_3925_to_fp16, x = inputs_127_cast_fp16)[name = string("out_127_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = string("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247736448)))]; + tensor input_251_beta_0_to_fp16 = const()[name = string("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247739072)))]; + fp16 input_251_epsilon_0_to_fp16 = const()[name = string("input_251_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = string("input_251_cast_fp16")]; + string input_253_pad_type_0 = const()[name = string("input_253_pad_type_0"), val = string("valid")]; + tensor input_253_strides_0 = const()[name = string("input_253_strides_0"), val = tensor([1, 1])]; + tensor input_253_pad_0 = const()[name = string("input_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_253_dilations_0 = const()[name = string("input_253_dilations_0"), val = tensor([1, 1])]; + int32 input_253_groups_0 = const()[name = string("input_253_groups_0"), val = int32(1)]; + tensor layers_31_fc1_weight_to_fp16 = const()[name = string("layers_31_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247741696)))]; + tensor layers_31_fc1_bias_to_fp16 = const()[name = string("layers_31_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260848960)))]; + tensor input_253_cast_fp16 = conv(bias = layers_31_fc1_bias_to_fp16, dilations = input_253_dilations_0, groups = input_253_groups_0, pad = input_253_pad_0, pad_type = input_253_pad_type_0, strides = input_253_strides_0, weight = layers_31_fc1_weight_to_fp16, x = input_251_cast_fp16)[name = string("input_253_cast_fp16")]; + string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("EXACT")]; + tensor input_255_cast_fp16 = gelu(mode = input_255_mode_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_31_fc2_weight_to_fp16 = const()[name = string("layers_31_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260859264)))]; + tensor layers_31_fc2_bias_to_fp16 = const()[name = string("layers_31_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273966528)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_31_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_31_fc2_weight_to_fp16, x = input_255_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_127_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_3963_to_fp16 = const()[name = string("op_3963_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_3963_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273969152)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273971776)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_3987_pad_type_0 = const()[name = string("op_3987_pad_type_0"), val = string("valid")]; + tensor var_3987_strides_0 = const()[name = string("op_3987_strides_0"), val = tensor([1, 1])]; + tensor var_3987_pad_0 = const()[name = string("op_3987_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3987_dilations_0 = const()[name = string("op_3987_dilations_0"), val = tensor([1, 1])]; + int32 var_3987_groups_0 = const()[name = string("op_3987_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273974400)))]; + tensor var_3987_cast_fp16 = conv(dilations = var_3987_dilations_0, groups = var_3987_groups_0, pad = var_3987_pad_0, pad_type = var_3987_pad_type_0, strides = var_3987_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_3987_cast_fp16")]; + string var_3994_pad_type_0 = const()[name = string("op_3994_pad_type_0"), val = string("valid")]; + tensor var_3994_strides_0 = const()[name = string("op_3994_strides_0"), val = tensor([1, 1])]; + tensor var_3994_pad_0 = const()[name = string("op_3994_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3994_dilations_0 = const()[name = string("op_3994_dilations_0"), val = tensor([1, 1])]; + int32 var_3994_groups_0 = const()[name = string("op_3994_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1277251264)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280528128)))]; + tensor var_3994_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_3994_dilations_0, groups = var_3994_groups_0, pad = var_3994_pad_0, pad_type = var_3994_pad_type_0, strides = var_3994_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_3994_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280530752)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1283807616)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1287084480)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_4024 = const()[name = string("op_4024"), val = int32(0)]; + bool input_259_interleave_0 = const()[name = string("input_259_interleave_0"), val = bool(false)]; + tensor input_259_cast_fp16 = concat(axis = var_4024, interleave = input_259_interleave_0, values = (var_3987_cast_fp16, k_cast_fp16))[name = string("input_259_cast_fp16")]; + int32 var_4027 = const()[name = string("op_4027"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_4027, interleave = input_interleave_0, values = (var_3994_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_4034_pad_0 = const()[name = string("op_4034_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4034_mode_0 = const()[name = string("op_4034_mode_0"), val = string("constant")]; + fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_33_to_fp16, mode = var_4034_mode_0, pad = var_4034_pad_0, x = input_259_cast_fp16)[name = string("op_4034_cast_fp16")]; + tensor var_4040_pad_0 = const()[name = string("op_4040_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4040_mode_0 = const()[name = string("op_4040_mode_0"), val = string("constant")]; + fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_34_to_fp16, mode = var_4040_mode_0, pad = var_4040_pad_0, x = input_cast_fp16)[name = string("op_4040_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/weights/weight.bin b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..d381c5cf6c7d4bbd79f9efc62d1372cd5cf6e67a --- /dev/null +++ b/distil-whisper_distil-large-v3/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b43a5d9e21e95067e0af8cf4b8fcbd16cc8e6f99993084f5e67cdf81bde16e79 +size 1287087104 diff --git a/distil-whisper_distil-large-v3/LICENSE_NOTICE.txt b/distil-whisper_distil-large-v3/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/distil-whisper_distil-large-v3/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..3ba3246801c85f92f79ac029f59b94e7fb646f85 --- /dev/null +++ b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0980462db89a546e1e90888ea38e0a5ddf1f1fec84608802cdbb12f8a5cc7215 +size 243 diff --git a/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/coremldata.bin b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..0ee276e2ae9c22b14311ffa08cf3ec21fe5c60c2 --- /dev/null +++ b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:451a5796f1dafb1969fef7bac32cd7fcf51fc763d1e1826ee6211dd046ede15a +size 329 diff --git a/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/metadata.json b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..65be90aad1d0e5f73a1f50b19705ccad3c0da822 --- /dev/null +++ b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/model.mil b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..6cf57d7dbf15af35e56636caf15aff60353296f0 --- /dev/null +++ b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/weights/weight.bin b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..2ae170c9000db89326cc2600450001654bb10f7f --- /dev/null +++ b/distil-whisper_distil-large-v3/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:009d9fb8f6b589accfa08cebf1c712ef07c3405229ce3cfb3a57ee033c9d8a49 +size 373376 diff --git a/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/analytics/coremldata.bin b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..99e84c4a1fe76264a47fffd68e9bb0b795e5458f --- /dev/null +++ b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77cb1b565a336e7fc01586698e50aa32d9a2a8f1ca5c439172564f4af0515f5d +size 243 diff --git a/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/coremldata.bin b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..bd5b35d879f406171df9765f277e32012e734dac --- /dev/null +++ b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a5e6f62b5ae897c8f846e22cacbe7d4f7d6bdbeb5f46366e2387f1082676b62 +size 754 diff --git a/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/metadata.json b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..d916f4d101c50c85e7cd58cb4f3baca787cacd86 --- /dev/null +++ b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51866)", + "shortDescription" : "", + "shape" : "[1, 1, 51866]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 2560 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 2560, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 2560 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 2560, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 4, + "Ios18.mul" : 8, + "Ios18.matmul" : 8, + "Ios18.batchNorm" : 7, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 15, + "Ios18.layerNorm" : 7, + "Ios18.reshape" : 16, + "Ios18.linear" : 1, + "Ios18.conv" : 16, + "Ios18.gelu" : 2, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 44, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/model.mil b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..5be4fa08fa43ad78d542009812ac7702c5a5d9fc --- /dev/null +++ b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,529 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_22_axis_0 = const()[name = string("op_22_axis_0"), val = int32(0)]; + int32 var_22_batch_dims_0 = const()[name = string("op_22_batch_dims_0"), val = int32(0)]; + bool var_22_validate_indices_0 = const()[name = string("op_22_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_22_cast_fp16 = gather(axis = var_22_axis_0, batch_dims = var_22_batch_dims_0, indices = input_ids, validate_indices = var_22_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_22_cast_fp16")]; + int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; + int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; + bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132777088)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_43")]; + tensor var_26_cast_fp16_cast_uint16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_26_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_26_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_22_cast_fp16, y = var_26_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_40_axes_0 = const()[name = string("op_40_axes_0"), val = tensor([2])]; + tensor var_40_cast_fp16 = expand_dims(axes = var_40_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_40_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_40_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1])]; + int32 var_45_axis_0 = const()[name = string("op_45_axis_0"), val = int32(0)]; + tensor var_45_cast_fp16_0, tensor var_45_cast_fp16_1 = split(axis = var_45_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_45_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1])]; + int32 var_50_axis_0 = const()[name = string("op_50_axis_0"), val = int32(0)]; + tensor var_50_cast_fp16_0, tensor var_50_cast_fp16_1 = split(axis = var_50_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_50_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_68 = const()[name = string("op_68"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_93_to_fp16 = const()[name = string("op_93_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_93_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133924032)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133926656)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133929280)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133931904)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133934528)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137211392)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137214016)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140490880)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143767744)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_128_axes_0 = const()[name = string("op_128_axes_0"), val = tensor([1])]; + tensor var_128_cast_fp16 = expand_dims(axes = var_128_axes_0, x = kv_cache_update_mask)[name = string("op_128_cast_fp16")]; + tensor var_129_axes_0 = const()[name = string("op_129_axes_0"), val = tensor([2])]; + tensor var_129_cast_fp16 = expand_dims(axes = var_129_axes_0, x = var_128_cast_fp16)[name = string("op_129_cast_fp16")]; + tensor var_131_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_129_cast_fp16)[name = string("op_131_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_45_cast_fp16_0, y = var_131_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_133_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_129_cast_fp16)[name = string("op_133_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_50_cast_fp16_0, y = var_133_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_136 = const()[name = string("op_136"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_136, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_138_to_fp16 = const()[name = string("op_138_to_fp16"), val = fp16(0x1p-3)]; + tensor var_139_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_138_to_fp16)[name = string("op_139_cast_fp16")]; + tensor var_140 = const()[name = string("op_140"), val = tensor([1, 20, 64, -1])]; + tensor var_141_cast_fp16 = reshape(shape = var_140, x = key_1_cast_fp16)[name = string("op_141_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_139_cast_fp16, y = var_141_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_145_axes_0 = const()[name = string("op_145_axes_0"), val = tensor([1])]; + tensor var_145_cast_fp16 = expand_dims(axes = var_145_axes_0, x = decoder_key_padding_mask)[name = string("op_145_cast_fp16")]; + tensor var_146_axes_0 = const()[name = string("op_146_axes_0"), val = tensor([2])]; + tensor var_146_cast_fp16 = expand_dims(axes = var_146_axes_0, x = var_145_cast_fp16)[name = string("op_146_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_146_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_149_cast_fp16 = softmax(axis = var_68, x = mh_w_3_cast_fp16)[name = string("op_149_cast_fp16")]; + tensor var_150 = const()[name = string("op_150"), val = tensor([1, 20, 64, -1])]; + tensor var_151_cast_fp16 = reshape(shape = var_150, x = value_1_cast_fp16)[name = string("op_151_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_151_cast_fp16, y = var_149_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_154 = const()[name = string("op_154"), val = tensor([1, 1280, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_154, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143770368)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147047232)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_176_to_fp16 = const()[name = string("op_176_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_176_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147049856)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147052480)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147055104)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331968)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_196 = const()[name = string("op_196"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_196, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_198_to_fp16 = const()[name = string("op_198_to_fp16"), val = fp16(0x1p-3)]; + tensor var_199_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_198_to_fp16)[name = string("op_199_cast_fp16")]; + tensor var_200 = const()[name = string("op_200"), val = tensor([1, 20, 64, -1])]; + tensor var_201_cast_fp16 = reshape(shape = var_200, x = obj_17_cast_fp16)[name = string("op_201_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_199_cast_fp16, y = var_201_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_205_axes_0 = const()[name = string("op_205_axes_0"), val = tensor([1])]; + tensor var_205_cast_fp16 = expand_dims(axes = var_205_axes_0, x = read_state_4)[name = string("op_205_cast_fp16")]; + tensor var_206_axes_0 = const()[name = string("op_206_axes_0"), val = tensor([2])]; + tensor var_206_cast_fp16 = expand_dims(axes = var_206_axes_0, x = var_205_cast_fp16)[name = string("op_206_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_206_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_68, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_210 = const()[name = string("op_210"), val = tensor([1, 20, 64, -1])]; + tensor var_211_cast_fp16 = reshape(shape = var_210, x = obj_19_cast_fp16)[name = string("op_211_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_211_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_214 = const()[name = string("op_214"), val = tensor([1, 1280, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_214, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334592)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_232_to_fp16 = const()[name = string("op_232_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_232_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614080)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616704)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153619328)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166726592)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166736896)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179844160)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_277 = const()[name = string("op_277"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_302_to_fp16 = const()[name = string("op_302_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_302_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179846784)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179849408)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179852032)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128896)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183131520)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186408384)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189685248)))]; + tensor current_value_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_340_cast_fp16 = mul(x = current_key_cast_fp16, y = var_129_cast_fp16)[name = string("op_340_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_45_cast_fp16_1, y = var_340_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_342_cast_fp16 = mul(x = current_value_cast_fp16, y = var_129_cast_fp16)[name = string("op_342_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_50_cast_fp16_1, y = var_342_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_345 = const()[name = string("op_345"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_345, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_347_to_fp16 = const()[name = string("op_347_to_fp16"), val = fp16(0x1p-3)]; + tensor var_348_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_347_to_fp16)[name = string("op_348_cast_fp16")]; + tensor var_349 = const()[name = string("op_349"), val = tensor([1, 20, 64, -1])]; + tensor var_350_cast_fp16 = reshape(shape = var_349, x = key_cast_fp16)[name = string("op_350_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_348_cast_fp16, y = var_350_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_146_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_358_cast_fp16 = softmax(axis = var_277, x = mh_w_11_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor var_359 = const()[name = string("op_359"), val = tensor([1, 20, 64, -1])]; + tensor var_360_cast_fp16 = reshape(shape = var_359, x = value_cast_fp16)[name = string("op_360_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_360_cast_fp16, y = var_358_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_363 = const()[name = string("op_363"), val = tensor([1, 1280, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_363, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189687872)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192964736)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_385_to_fp16 = const()[name = string("op_385_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_385_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192967360)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192969984)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192972608)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196249472)))]; + tensor query_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_405 = const()[name = string("op_405"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_405, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_407_to_fp16 = const()[name = string("op_407_to_fp16"), val = fp16(0x1p-3)]; + tensor var_408_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_407_to_fp16)[name = string("op_408_cast_fp16")]; + tensor var_409 = const()[name = string("op_409"), val = tensor([1, 20, 64, -1])]; + tensor var_410_cast_fp16 = reshape(shape = var_409, x = obj_35_cast_fp16)[name = string("op_410_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_408_cast_fp16, y = var_410_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_206_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_277, x = mh_w_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_419 = const()[name = string("op_419"), val = tensor([1, 20, 64, -1])]; + tensor var_420_cast_fp16 = reshape(shape = var_419, x = obj_37_cast_fp16)[name = string("op_420_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_420_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_423 = const()[name = string("op_423"), val = tensor([1, 1280, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_423, x = attn_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196252096)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199528960)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_444_to_fp16 = const()[name = string("op_444_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_444_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199531584)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199534208)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199536832)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212644096)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_17_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212654400)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225761664)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_487_to_fp16 = const()[name = string("op_487_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_487_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225764288)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225766912)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_498_axes_0 = const()[name = string("op_498_axes_0"), val = tensor([2])]; + tensor var_498_cast_fp16 = squeeze(axes = var_498_axes_0, x = hidden_states_cast_fp16)[name = string("op_498_cast_fp16")]; + tensor var_501_perm_0 = const()[name = string("op_501_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225769536)))]; + tensor var_501_cast_fp16 = transpose(perm = var_501_perm_0, x = var_498_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_501_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_505 = const()[name = string("op_505"), val = int32(1)]; + bool obj_45_interleave_0 = const()[name = string("obj_45_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_505, interleave = obj_45_interleave_0, values = (current_key_1_cast_fp16, current_key_cast_fp16))[name = string("obj_45_cast_fp16")]; + int32 var_508 = const()[name = string("op_508"), val = int32(1)]; + bool obj_47_interleave_0 = const()[name = string("obj_47_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_508, interleave = obj_47_interleave_0, values = (current_value_1_cast_fp16, current_value_cast_fp16))[name = string("obj_47_cast_fp16")]; + tensor var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = obj_41_cast_fp16)[name = string("op_519_cast_fp16")]; + tensor var_522_begin_0 = const()[name = string("op_522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_522_end_0 = const()[name = string("op_522_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_522_end_mask_0 = const()[name = string("op_522_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_522_squeeze_mask_0 = const()[name = string("op_522_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_522_cast_fp16 = slice_by_index(begin = var_522_begin_0, end = var_522_end_0, end_mask = var_522_end_mask_0, squeeze_mask = var_522_squeeze_mask_0, x = var_519_cast_fp16)[name = string("op_522_cast_fp16")]; + tensor var_537_begin_0 = const()[name = string("op_537_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_537_end_0 = const()[name = string("op_537_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_537_end_mask_0 = const()[name = string("op_537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_537_cast_fp16 = slice_by_index(begin = var_537_begin_0, end = var_537_end_0, end_mask = var_537_end_mask_0, x = obj_41_cast_fp16)[name = string("op_537_cast_fp16")]; + tensor var_540_begin_0 = const()[name = string("op_540_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_540_end_0 = const()[name = string("op_540_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_540_end_mask_0 = const()[name = string("op_540_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_540_squeeze_mask_0 = const()[name = string("op_540_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_540_cast_fp16 = slice_by_index(begin = var_540_begin_0, end = var_540_end_0, end_mask = var_540_end_mask_0, squeeze_mask = var_540_squeeze_mask_0, x = var_537_cast_fp16)[name = string("op_540_cast_fp16")]; + tensor var_555_begin_0 = const()[name = string("op_555_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_555_end_0 = const()[name = string("op_555_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_555_end_mask_0 = const()[name = string("op_555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_555_cast_fp16 = slice_by_index(begin = var_555_begin_0, end = var_555_end_0, end_mask = var_555_end_mask_0, x = obj_41_cast_fp16)[name = string("op_555_cast_fp16")]; + tensor var_558_begin_0 = const()[name = string("op_558_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_558_end_0 = const()[name = string("op_558_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_558_end_mask_0 = const()[name = string("op_558_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_558_squeeze_mask_0 = const()[name = string("op_558_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_558_cast_fp16 = slice_by_index(begin = var_558_begin_0, end = var_558_end_0, end_mask = var_558_end_mask_0, squeeze_mask = var_558_squeeze_mask_0, x = var_555_cast_fp16)[name = string("op_558_cast_fp16")]; + tensor var_573_begin_0 = const()[name = string("op_573_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_573_end_0 = const()[name = string("op_573_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_573_end_mask_0 = const()[name = string("op_573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_573_cast_fp16 = slice_by_index(begin = var_573_begin_0, end = var_573_end_0, end_mask = var_573_end_mask_0, x = obj_41_cast_fp16)[name = string("op_573_cast_fp16")]; + tensor var_576_begin_0 = const()[name = string("op_576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_576_end_0 = const()[name = string("op_576_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_576_end_mask_0 = const()[name = string("op_576_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_576_squeeze_mask_0 = const()[name = string("op_576_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_576_cast_fp16 = slice_by_index(begin = var_576_begin_0, end = var_576_end_0, end_mask = var_576_end_mask_0, squeeze_mask = var_576_squeeze_mask_0, x = var_573_cast_fp16)[name = string("op_576_cast_fp16")]; + tensor var_591_begin_0 = const()[name = string("op_591_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_591_end_0 = const()[name = string("op_591_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_591_end_mask_0 = const()[name = string("op_591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_591_cast_fp16 = slice_by_index(begin = var_591_begin_0, end = var_591_end_0, end_mask = var_591_end_mask_0, x = obj_41_cast_fp16)[name = string("op_591_cast_fp16")]; + tensor var_594_begin_0 = const()[name = string("op_594_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_594_end_0 = const()[name = string("op_594_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_594_end_mask_0 = const()[name = string("op_594_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_594_squeeze_mask_0 = const()[name = string("op_594_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_594_cast_fp16 = slice_by_index(begin = var_594_begin_0, end = var_594_end_0, end_mask = var_594_end_mask_0, squeeze_mask = var_594_squeeze_mask_0, x = var_591_cast_fp16)[name = string("op_594_cast_fp16")]; + tensor var_609_begin_0 = const()[name = string("op_609_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_609_end_0 = const()[name = string("op_609_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_609_end_mask_0 = const()[name = string("op_609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_609_cast_fp16 = slice_by_index(begin = var_609_begin_0, end = var_609_end_0, end_mask = var_609_end_mask_0, x = obj_41_cast_fp16)[name = string("op_609_cast_fp16")]; + tensor var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_612_squeeze_mask_0 = const()[name = string("op_612_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, squeeze_mask = var_612_squeeze_mask_0, x = var_609_cast_fp16)[name = string("op_612_cast_fp16")]; + tensor var_627_begin_0 = const()[name = string("op_627_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_627_end_0 = const()[name = string("op_627_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_627_end_mask_0 = const()[name = string("op_627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_627_cast_fp16 = slice_by_index(begin = var_627_begin_0, end = var_627_end_0, end_mask = var_627_end_mask_0, x = obj_41_cast_fp16)[name = string("op_627_cast_fp16")]; + tensor var_630_begin_0 = const()[name = string("op_630_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_630_end_0 = const()[name = string("op_630_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_630_end_mask_0 = const()[name = string("op_630_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_630_squeeze_mask_0 = const()[name = string("op_630_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_630_cast_fp16 = slice_by_index(begin = var_630_begin_0, end = var_630_end_0, end_mask = var_630_end_mask_0, squeeze_mask = var_630_squeeze_mask_0, x = var_627_cast_fp16)[name = string("op_630_cast_fp16")]; + tensor var_645_begin_0 = const()[name = string("op_645_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_645_end_0 = const()[name = string("op_645_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_645_end_mask_0 = const()[name = string("op_645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = obj_41_cast_fp16)[name = string("op_645_cast_fp16")]; + tensor var_648_begin_0 = const()[name = string("op_648_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_648_end_0 = const()[name = string("op_648_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_648_end_mask_0 = const()[name = string("op_648_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_648_squeeze_mask_0 = const()[name = string("op_648_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_648_cast_fp16 = slice_by_index(begin = var_648_begin_0, end = var_648_end_0, end_mask = var_648_end_mask_0, squeeze_mask = var_648_squeeze_mask_0, x = var_645_cast_fp16)[name = string("op_648_cast_fp16")]; + tensor var_663_begin_0 = const()[name = string("op_663_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_663_end_0 = const()[name = string("op_663_end_0"), val = tensor([1, 9, 1, 1536])]; + tensor var_663_end_mask_0 = const()[name = string("op_663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = obj_41_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_666_begin_0 = const()[name = string("op_666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_666_end_0 = const()[name = string("op_666_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_666_end_mask_0 = const()[name = string("op_666_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_666_squeeze_mask_0 = const()[name = string("op_666_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_666_cast_fp16 = slice_by_index(begin = var_666_begin_0, end = var_666_end_0, end_mask = var_666_end_mask_0, squeeze_mask = var_666_squeeze_mask_0, x = var_663_cast_fp16)[name = string("op_666_cast_fp16")]; + tensor var_681_begin_0 = const()[name = string("op_681_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_681_end_0 = const()[name = string("op_681_end_0"), val = tensor([1, 10, 1, 1536])]; + tensor var_681_end_mask_0 = const()[name = string("op_681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_681_cast_fp16 = slice_by_index(begin = var_681_begin_0, end = var_681_end_0, end_mask = var_681_end_mask_0, x = obj_41_cast_fp16)[name = string("op_681_cast_fp16")]; + tensor var_684_begin_0 = const()[name = string("op_684_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_684_end_0 = const()[name = string("op_684_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_684_end_mask_0 = const()[name = string("op_684_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_684_squeeze_mask_0 = const()[name = string("op_684_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_684_cast_fp16 = slice_by_index(begin = var_684_begin_0, end = var_684_end_0, end_mask = var_684_end_mask_0, squeeze_mask = var_684_squeeze_mask_0, x = var_681_cast_fp16)[name = string("op_684_cast_fp16")]; + tensor var_699_begin_0 = const()[name = string("op_699_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_699_end_0 = const()[name = string("op_699_end_0"), val = tensor([1, 11, 1, 1536])]; + tensor var_699_end_mask_0 = const()[name = string("op_699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_699_cast_fp16 = slice_by_index(begin = var_699_begin_0, end = var_699_end_0, end_mask = var_699_end_mask_0, x = obj_41_cast_fp16)[name = string("op_699_cast_fp16")]; + tensor var_702_begin_0 = const()[name = string("op_702_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_702_end_0 = const()[name = string("op_702_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_702_end_mask_0 = const()[name = string("op_702_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_702_squeeze_mask_0 = const()[name = string("op_702_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_702_cast_fp16 = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, squeeze_mask = var_702_squeeze_mask_0, x = var_699_cast_fp16)[name = string("op_702_cast_fp16")]; + tensor var_717_begin_0 = const()[name = string("op_717_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_717_end_0 = const()[name = string("op_717_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_717_end_mask_0 = const()[name = string("op_717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_717_cast_fp16 = slice_by_index(begin = var_717_begin_0, end = var_717_end_0, end_mask = var_717_end_mask_0, x = obj_41_cast_fp16)[name = string("op_717_cast_fp16")]; + tensor var_720_begin_0 = const()[name = string("op_720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_720_end_0 = const()[name = string("op_720_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_720_end_mask_0 = const()[name = string("op_720_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_720_squeeze_mask_0 = const()[name = string("op_720_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_720_cast_fp16 = slice_by_index(begin = var_720_begin_0, end = var_720_end_0, end_mask = var_720_end_mask_0, squeeze_mask = var_720_squeeze_mask_0, x = var_717_cast_fp16)[name = string("op_720_cast_fp16")]; + tensor var_735_begin_0 = const()[name = string("op_735_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_735_end_0 = const()[name = string("op_735_end_0"), val = tensor([1, 13, 1, 1536])]; + tensor var_735_end_mask_0 = const()[name = string("op_735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_735_cast_fp16 = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = obj_41_cast_fp16)[name = string("op_735_cast_fp16")]; + tensor var_738_begin_0 = const()[name = string("op_738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_738_end_0 = const()[name = string("op_738_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_738_end_mask_0 = const()[name = string("op_738_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_738_squeeze_mask_0 = const()[name = string("op_738_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_738_cast_fp16 = slice_by_index(begin = var_738_begin_0, end = var_738_end_0, end_mask = var_738_end_mask_0, squeeze_mask = var_738_squeeze_mask_0, x = var_735_cast_fp16)[name = string("op_738_cast_fp16")]; + tensor var_753_begin_0 = const()[name = string("op_753_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_753_end_0 = const()[name = string("op_753_end_0"), val = tensor([1, 14, 1, 1536])]; + tensor var_753_end_mask_0 = const()[name = string("op_753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = obj_41_cast_fp16)[name = string("op_753_cast_fp16")]; + tensor var_756_begin_0 = const()[name = string("op_756_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_756_end_0 = const()[name = string("op_756_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_756_end_mask_0 = const()[name = string("op_756_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_756_squeeze_mask_0 = const()[name = string("op_756_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_756_cast_fp16 = slice_by_index(begin = var_756_begin_0, end = var_756_end_0, end_mask = var_756_end_mask_0, squeeze_mask = var_756_squeeze_mask_0, x = var_753_cast_fp16)[name = string("op_756_cast_fp16")]; + tensor var_771_begin_0 = const()[name = string("op_771_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_771_end_0 = const()[name = string("op_771_end_0"), val = tensor([1, 15, 1, 1536])]; + tensor var_771_end_mask_0 = const()[name = string("op_771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_771_cast_fp16 = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = obj_41_cast_fp16)[name = string("op_771_cast_fp16")]; + tensor var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_774_squeeze_mask_0 = const()[name = string("op_774_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, squeeze_mask = var_774_squeeze_mask_0, x = var_771_cast_fp16)[name = string("op_774_cast_fp16")]; + tensor var_789_begin_0 = const()[name = string("op_789_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_789_end_0 = const()[name = string("op_789_end_0"), val = tensor([1, 16, 1, 1536])]; + tensor var_789_end_mask_0 = const()[name = string("op_789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_789_cast_fp16 = slice_by_index(begin = var_789_begin_0, end = var_789_end_0, end_mask = var_789_end_mask_0, x = obj_41_cast_fp16)[name = string("op_789_cast_fp16")]; + tensor var_792_begin_0 = const()[name = string("op_792_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_792_end_0 = const()[name = string("op_792_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_792_end_mask_0 = const()[name = string("op_792_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_792_squeeze_mask_0 = const()[name = string("op_792_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_792_cast_fp16 = slice_by_index(begin = var_792_begin_0, end = var_792_end_0, end_mask = var_792_end_mask_0, squeeze_mask = var_792_squeeze_mask_0, x = var_789_cast_fp16)[name = string("op_792_cast_fp16")]; + tensor var_807_begin_0 = const()[name = string("op_807_begin_0"), val = tensor([0, 16, 0, 0])]; + tensor var_807_end_0 = const()[name = string("op_807_end_0"), val = tensor([1, 17, 1, 1536])]; + tensor var_807_end_mask_0 = const()[name = string("op_807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_807_cast_fp16 = slice_by_index(begin = var_807_begin_0, end = var_807_end_0, end_mask = var_807_end_mask_0, x = obj_41_cast_fp16)[name = string("op_807_cast_fp16")]; + tensor var_810_begin_0 = const()[name = string("op_810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_810_end_0 = const()[name = string("op_810_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_810_end_mask_0 = const()[name = string("op_810_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_810_squeeze_mask_0 = const()[name = string("op_810_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_810_cast_fp16 = slice_by_index(begin = var_810_begin_0, end = var_810_end_0, end_mask = var_810_end_mask_0, squeeze_mask = var_810_squeeze_mask_0, x = var_807_cast_fp16)[name = string("op_810_cast_fp16")]; + tensor var_825_begin_0 = const()[name = string("op_825_begin_0"), val = tensor([0, 17, 0, 0])]; + tensor var_825_end_0 = const()[name = string("op_825_end_0"), val = tensor([1, 18, 1, 1536])]; + tensor var_825_end_mask_0 = const()[name = string("op_825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_825_cast_fp16 = slice_by_index(begin = var_825_begin_0, end = var_825_end_0, end_mask = var_825_end_mask_0, x = obj_41_cast_fp16)[name = string("op_825_cast_fp16")]; + tensor var_828_begin_0 = const()[name = string("op_828_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_828_end_0 = const()[name = string("op_828_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_828_end_mask_0 = const()[name = string("op_828_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_828_squeeze_mask_0 = const()[name = string("op_828_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_828_cast_fp16 = slice_by_index(begin = var_828_begin_0, end = var_828_end_0, end_mask = var_828_end_mask_0, squeeze_mask = var_828_squeeze_mask_0, x = var_825_cast_fp16)[name = string("op_828_cast_fp16")]; + tensor var_843_begin_0 = const()[name = string("op_843_begin_0"), val = tensor([0, 18, 0, 0])]; + tensor var_843_end_0 = const()[name = string("op_843_end_0"), val = tensor([1, 19, 1, 1536])]; + tensor var_843_end_mask_0 = const()[name = string("op_843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_843_cast_fp16 = slice_by_index(begin = var_843_begin_0, end = var_843_end_0, end_mask = var_843_end_mask_0, x = obj_41_cast_fp16)[name = string("op_843_cast_fp16")]; + tensor var_846_begin_0 = const()[name = string("op_846_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_846_end_0 = const()[name = string("op_846_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_846_end_mask_0 = const()[name = string("op_846_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_846_squeeze_mask_0 = const()[name = string("op_846_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_846_cast_fp16 = slice_by_index(begin = var_846_begin_0, end = var_846_end_0, end_mask = var_846_end_mask_0, squeeze_mask = var_846_squeeze_mask_0, x = var_843_cast_fp16)[name = string("op_846_cast_fp16")]; + tensor var_861_begin_0 = const()[name = string("op_861_begin_0"), val = tensor([0, 19, 0, 0])]; + tensor var_861_end_0 = const()[name = string("op_861_end_0"), val = tensor([1, 20, 1, 1536])]; + tensor var_861_end_mask_0 = const()[name = string("op_861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_861_cast_fp16 = slice_by_index(begin = var_861_begin_0, end = var_861_end_0, end_mask = var_861_end_mask_0, x = obj_41_cast_fp16)[name = string("op_861_cast_fp16")]; + tensor var_864_begin_0 = const()[name = string("op_864_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_864_end_0 = const()[name = string("op_864_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_864_end_mask_0 = const()[name = string("op_864_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_864_squeeze_mask_0 = const()[name = string("op_864_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_864_cast_fp16 = slice_by_index(begin = var_864_begin_0, end = var_864_end_0, end_mask = var_864_end_mask_0, squeeze_mask = var_864_squeeze_mask_0, x = var_861_cast_fp16)[name = string("op_864_cast_fp16")]; + int32 var_871 = const()[name = string("op_871"), val = int32(1)]; + bool var_872_interleave_0 = const()[name = string("op_872_interleave_0"), val = bool(false)]; + tensor var_872_cast_fp16 = concat(axis = var_871, interleave = var_872_interleave_0, values = (var_522_cast_fp16, var_540_cast_fp16, var_558_cast_fp16, var_576_cast_fp16, var_594_cast_fp16, var_612_cast_fp16, var_630_cast_fp16, var_648_cast_fp16, var_666_cast_fp16, var_684_cast_fp16, var_702_cast_fp16, var_720_cast_fp16, var_738_cast_fp16, var_756_cast_fp16, var_774_cast_fp16, var_792_cast_fp16, var_810_cast_fp16, var_828_cast_fp16, var_846_cast_fp16, var_864_cast_fp16))[name = string("op_872_cast_fp16")]; + bool var_875 = const()[name = string("op_875"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_875, x = var_872_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/weights/weight.bin b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..9929465fd5f57561fddd5514a2227d4c6c31a160 --- /dev/null +++ b/distil-whisper_distil-large-v3/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c1afaacaec2fac64e8867d758742347e10c849fdbf81c8761344b5c56a55b5d +size 225873332 diff --git a/distil-whisper_distil-large-v3/config.json b/distil-whisper_distil-large-v3/config.json new file mode 100644 index 0000000000000000000000000000000000000000..35fb9504486c4c3df76f5f84aad0bdffd3de00f6 --- /dev/null +++ b/distil-whisper_distil-large-v3/config.json @@ -0,0 +1 @@ +{"_name_or_path": "./distil-large-v3", "activation_dropout": 0.0, "activation_function": "gelu", "apply_spec_augment": false, "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "classifier_proj_size": 256, "d_model": 1280, "decoder_attention_heads": 20, "decoder_ffn_dim": 5120, "decoder_layerdrop": 0.0, "decoder_layers": 2, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 20, "encoder_ffn_dim": 5120, "encoder_layerdrop": 0.0, "encoder_layers": 32, "eos_token_id": 50257, "init_std": 0.02, "is_encoder_decoder": true, "mask_feature_length": 10, "mask_feature_min_masks": 0, "mask_feature_prob": 0.0, "mask_time_length": 10, "mask_time_min_masks": 2, "mask_time_prob": 0.05, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "median_filter_width": 7, "model_type": "whisper", "num_hidden_layers": 32, "num_mel_bins": 128, "pad_token_id": 50256, "scale_embedding": false, "torch_dtype": "float16", "transformers_version": "4.38.0.dev0", "use_cache": true, "use_weighted_layer_sum": false, "vocab_size": 51866} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3/generation_config.json b/distil-whisper_distil-large-v3/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..8d1525665be3225ab0cbd947863a3f75fe5a5b2b --- /dev/null +++ b/distil-whisper_distil-large-v3/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[7, 0], [10, 17], [12, 18], [13, 12], [16, 1], [17, 14], [19, 11], [21, 4], [24, 1], [25, 6]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50360]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|yue|>": 50358, "<|zh|>": 50260}, "language": "<|en|>", "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50364, "pad_token_id": 50257, "prev_sot_token_id": 50362, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50359, 50360, 50361, 50362, 50363], "task": "transcribe", "task_to_id": {"transcribe": 50360, "translate": 50359}, "transformers_version": "4.38.0.dev0"} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/analytics/coremldata.bin b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..f10ee26bb022ef2d896fc2b45f4dd6d83b7bdba1 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5916646e39691156fca37ff36b96f162e80acce84cd8ee2e971115edf412a87a +size 243 diff --git a/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/coremldata.bin b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..9d9ee61ab4ec55f649de3498eb4269653f592b80 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dbd63faaf82dd7f50bad861be88605fbdce6f59a2ced9954ef1a54a51f1e26ac +size 434 diff --git a/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/metadata.json b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..76c530437d48ee45df0b264d68509a026c6da9e1 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1280 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 1280, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 65, + "Ios18.conv" : 198, + "Ios18.gelu" : 34, + "Ios18.concat" : 674, + "Ios16.einsum" : 5120, + "Ios18.add" : 65, + "Ios18.softmax" : 2560, + "Ios18.sliceByIndex" : 4480, + "Ios18.layerNorm" : 65, + "Ios18.transpose" : 32, + "Ios18.mul" : 2560 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/model.mil b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..87874ed6c3eff34b543734b0bdd46f4669aad6f7 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,39473 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_110_pad_type_0 = const()[name = string("op_110_pad_type_0"), val = string("custom")]; + tensor var_110_pad_0 = const()[name = string("op_110_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_110_strides_0 = const()[name = string("op_110_strides_0"), val = tensor([1, 1])]; + tensor var_110_dilations_0 = const()[name = string("op_110_dilations_0"), val = tensor([1, 1])]; + int32 var_110_groups_0 = const()[name = string("op_110_groups_0"), val = int32(1)]; + tensor var_85_to_fp16 = const()[name = string("op_85_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_91_to_fp16 = const()[name = string("op_91_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(983168)))]; + tensor var_110_cast_fp16 = conv(bias = var_91_to_fp16, dilations = var_110_dilations_0, groups = var_110_groups_0, pad = var_110_pad_0, pad_type = var_110_pad_type_0, strides = var_110_strides_0, weight = var_85_to_fp16, x = melspectrogram_features)[name = string("op_110_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_110_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_150_pad_type_0 = const()[name = string("op_150_pad_type_0"), val = string("custom")]; + tensor var_150_pad_0 = const()[name = string("op_150_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_150_strides_0 = const()[name = string("op_150_strides_0"), val = tensor([2, 2])]; + tensor var_150_dilations_0 = const()[name = string("op_150_dilations_0"), val = tensor([1, 1])]; + int32 var_150_groups_0 = const()[name = string("op_150_groups_0"), val = int32(1)]; + tensor var_125_to_fp16 = const()[name = string("op_125_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985792)))]; + tensor var_131_to_fp16 = const()[name = string("op_131_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10816256)))]; + tensor var_150_cast_fp16 = conv(bias = var_131_to_fp16, dilations = var_150_dilations_0, groups = var_150_groups_0, pad = var_150_pad_0, pad_type = var_150_pad_type_0, strides = var_150_strides_0, weight = var_125_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_150_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_150_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_168_to_fp16 = const()[name = string("op_168_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_168_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_178 = const()[name = string("op_178"), val = int32(3)]; + int32 var_203 = const()[name = string("op_203"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_220_to_fp16 = const()[name = string("op_220_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_220_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14658944)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14661568)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14664192)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14666816)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14669440)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17946304)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17948928)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21225792)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24502656)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_258_begin_0 = const()[name = string("op_258_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_258_end_0 = const()[name = string("op_258_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_258_end_mask_0 = const()[name = string("op_258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_258_cast_fp16 = slice_by_index(begin = var_258_begin_0, end = var_258_end_0, end_mask = var_258_end_mask_0, x = query_1_cast_fp16)[name = string("op_258_cast_fp16")]; + tensor var_262_begin_0 = const()[name = string("op_262_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_262_end_0 = const()[name = string("op_262_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_262_end_mask_0 = const()[name = string("op_262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_262_cast_fp16 = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = query_1_cast_fp16)[name = string("op_262_cast_fp16")]; + tensor var_266_begin_0 = const()[name = string("op_266_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_266_end_0 = const()[name = string("op_266_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_266_end_mask_0 = const()[name = string("op_266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = query_1_cast_fp16)[name = string("op_266_cast_fp16")]; + tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = query_1_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = query_1_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_278_begin_0 = const()[name = string("op_278_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_278_end_0 = const()[name = string("op_278_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_278_end_mask_0 = const()[name = string("op_278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_278_cast_fp16 = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = query_1_cast_fp16)[name = string("op_278_cast_fp16")]; + tensor var_282_begin_0 = const()[name = string("op_282_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_282_end_0 = const()[name = string("op_282_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_282_end_mask_0 = const()[name = string("op_282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_282_cast_fp16 = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, x = query_1_cast_fp16)[name = string("op_282_cast_fp16")]; + tensor var_286_begin_0 = const()[name = string("op_286_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_286_end_0 = const()[name = string("op_286_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_286_end_mask_0 = const()[name = string("op_286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_286_cast_fp16 = slice_by_index(begin = var_286_begin_0, end = var_286_end_0, end_mask = var_286_end_mask_0, x = query_1_cast_fp16)[name = string("op_286_cast_fp16")]; + tensor var_290_begin_0 = const()[name = string("op_290_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_290_end_0 = const()[name = string("op_290_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_290_end_mask_0 = const()[name = string("op_290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = query_1_cast_fp16)[name = string("op_290_cast_fp16")]; + tensor var_294_begin_0 = const()[name = string("op_294_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_294_end_0 = const()[name = string("op_294_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_294_end_mask_0 = const()[name = string("op_294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_294_cast_fp16 = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = query_1_cast_fp16)[name = string("op_294_cast_fp16")]; + tensor var_298_begin_0 = const()[name = string("op_298_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_298_end_0 = const()[name = string("op_298_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_298_end_mask_0 = const()[name = string("op_298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = query_1_cast_fp16)[name = string("op_298_cast_fp16")]; + tensor var_302_begin_0 = const()[name = string("op_302_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_302_end_0 = const()[name = string("op_302_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_302_end_mask_0 = const()[name = string("op_302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = query_1_cast_fp16)[name = string("op_302_cast_fp16")]; + tensor var_306_begin_0 = const()[name = string("op_306_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_306_end_0 = const()[name = string("op_306_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_306_end_mask_0 = const()[name = string("op_306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_306_cast_fp16 = slice_by_index(begin = var_306_begin_0, end = var_306_end_0, end_mask = var_306_end_mask_0, x = query_1_cast_fp16)[name = string("op_306_cast_fp16")]; + tensor var_310_begin_0 = const()[name = string("op_310_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_310_end_0 = const()[name = string("op_310_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_310_end_mask_0 = const()[name = string("op_310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_310_cast_fp16 = slice_by_index(begin = var_310_begin_0, end = var_310_end_0, end_mask = var_310_end_mask_0, x = query_1_cast_fp16)[name = string("op_310_cast_fp16")]; + tensor var_314_begin_0 = const()[name = string("op_314_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_314_end_0 = const()[name = string("op_314_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_314_end_mask_0 = const()[name = string("op_314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_314_cast_fp16 = slice_by_index(begin = var_314_begin_0, end = var_314_end_0, end_mask = var_314_end_mask_0, x = query_1_cast_fp16)[name = string("op_314_cast_fp16")]; + tensor var_318_begin_0 = const()[name = string("op_318_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_318_end_0 = const()[name = string("op_318_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_318_end_mask_0 = const()[name = string("op_318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = query_1_cast_fp16)[name = string("op_318_cast_fp16")]; + tensor var_322_begin_0 = const()[name = string("op_322_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_322_end_0 = const()[name = string("op_322_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_322_end_mask_0 = const()[name = string("op_322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_322_cast_fp16 = slice_by_index(begin = var_322_begin_0, end = var_322_end_0, end_mask = var_322_end_mask_0, x = query_1_cast_fp16)[name = string("op_322_cast_fp16")]; + tensor var_326_begin_0 = const()[name = string("op_326_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_326_end_0 = const()[name = string("op_326_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_326_end_mask_0 = const()[name = string("op_326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = query_1_cast_fp16)[name = string("op_326_cast_fp16")]; + tensor var_330_begin_0 = const()[name = string("op_330_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_330_end_0 = const()[name = string("op_330_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_330_end_mask_0 = const()[name = string("op_330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_330_cast_fp16 = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = query_1_cast_fp16)[name = string("op_330_cast_fp16")]; + tensor var_334_begin_0 = const()[name = string("op_334_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_334_end_0 = const()[name = string("op_334_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_334_end_mask_0 = const()[name = string("op_334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_334_cast_fp16 = slice_by_index(begin = var_334_begin_0, end = var_334_end_0, end_mask = var_334_end_mask_0, x = query_1_cast_fp16)[name = string("op_334_cast_fp16")]; + tensor var_343_begin_0 = const()[name = string("op_343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_343_end_0 = const()[name = string("op_343_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_343_end_mask_0 = const()[name = string("op_343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_343_cast_fp16 = slice_by_index(begin = var_343_begin_0, end = var_343_end_0, end_mask = var_343_end_mask_0, x = var_258_cast_fp16)[name = string("op_343_cast_fp16")]; + tensor var_350_begin_0 = const()[name = string("op_350_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_350_end_0 = const()[name = string("op_350_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_350_end_mask_0 = const()[name = string("op_350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_350_cast_fp16 = slice_by_index(begin = var_350_begin_0, end = var_350_end_0, end_mask = var_350_end_mask_0, x = var_258_cast_fp16)[name = string("op_350_cast_fp16")]; + tensor var_357_begin_0 = const()[name = string("op_357_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_357_end_0 = const()[name = string("op_357_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_357_end_mask_0 = const()[name = string("op_357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_357_cast_fp16 = slice_by_index(begin = var_357_begin_0, end = var_357_end_0, end_mask = var_357_end_mask_0, x = var_258_cast_fp16)[name = string("op_357_cast_fp16")]; + tensor var_364_begin_0 = const()[name = string("op_364_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_364_end_0 = const()[name = string("op_364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_364_end_mask_0 = const()[name = string("op_364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_364_cast_fp16 = slice_by_index(begin = var_364_begin_0, end = var_364_end_0, end_mask = var_364_end_mask_0, x = var_258_cast_fp16)[name = string("op_364_cast_fp16")]; + tensor var_371_begin_0 = const()[name = string("op_371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_371_end_0 = const()[name = string("op_371_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_371_end_mask_0 = const()[name = string("op_371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_371_cast_fp16 = slice_by_index(begin = var_371_begin_0, end = var_371_end_0, end_mask = var_371_end_mask_0, x = var_262_cast_fp16)[name = string("op_371_cast_fp16")]; + tensor var_378_begin_0 = const()[name = string("op_378_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_378_end_0 = const()[name = string("op_378_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_378_end_mask_0 = const()[name = string("op_378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_378_cast_fp16 = slice_by_index(begin = var_378_begin_0, end = var_378_end_0, end_mask = var_378_end_mask_0, x = var_262_cast_fp16)[name = string("op_378_cast_fp16")]; + tensor var_385_begin_0 = const()[name = string("op_385_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_385_end_0 = const()[name = string("op_385_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_385_end_mask_0 = const()[name = string("op_385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_385_cast_fp16 = slice_by_index(begin = var_385_begin_0, end = var_385_end_0, end_mask = var_385_end_mask_0, x = var_262_cast_fp16)[name = string("op_385_cast_fp16")]; + tensor var_392_begin_0 = const()[name = string("op_392_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_392_end_0 = const()[name = string("op_392_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_392_end_mask_0 = const()[name = string("op_392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_392_cast_fp16 = slice_by_index(begin = var_392_begin_0, end = var_392_end_0, end_mask = var_392_end_mask_0, x = var_262_cast_fp16)[name = string("op_392_cast_fp16")]; + tensor var_399_begin_0 = const()[name = string("op_399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_399_end_0 = const()[name = string("op_399_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_399_end_mask_0 = const()[name = string("op_399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_399_cast_fp16 = slice_by_index(begin = var_399_begin_0, end = var_399_end_0, end_mask = var_399_end_mask_0, x = var_266_cast_fp16)[name = string("op_399_cast_fp16")]; + tensor var_406_begin_0 = const()[name = string("op_406_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_406_end_0 = const()[name = string("op_406_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_406_end_mask_0 = const()[name = string("op_406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_406_cast_fp16 = slice_by_index(begin = var_406_begin_0, end = var_406_end_0, end_mask = var_406_end_mask_0, x = var_266_cast_fp16)[name = string("op_406_cast_fp16")]; + tensor var_413_begin_0 = const()[name = string("op_413_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_413_end_0 = const()[name = string("op_413_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_413_end_mask_0 = const()[name = string("op_413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, x = var_266_cast_fp16)[name = string("op_413_cast_fp16")]; + tensor var_420_begin_0 = const()[name = string("op_420_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_420_end_0 = const()[name = string("op_420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_420_end_mask_0 = const()[name = string("op_420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_420_cast_fp16 = slice_by_index(begin = var_420_begin_0, end = var_420_end_0, end_mask = var_420_end_mask_0, x = var_266_cast_fp16)[name = string("op_420_cast_fp16")]; + tensor var_427_begin_0 = const()[name = string("op_427_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_427_end_0 = const()[name = string("op_427_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_427_end_mask_0 = const()[name = string("op_427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_427_cast_fp16 = slice_by_index(begin = var_427_begin_0, end = var_427_end_0, end_mask = var_427_end_mask_0, x = var_270_cast_fp16)[name = string("op_427_cast_fp16")]; + tensor var_434_begin_0 = const()[name = string("op_434_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_434_end_0 = const()[name = string("op_434_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_434_end_mask_0 = const()[name = string("op_434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_434_cast_fp16 = slice_by_index(begin = var_434_begin_0, end = var_434_end_0, end_mask = var_434_end_mask_0, x = var_270_cast_fp16)[name = string("op_434_cast_fp16")]; + tensor var_441_begin_0 = const()[name = string("op_441_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_441_end_0 = const()[name = string("op_441_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_441_end_mask_0 = const()[name = string("op_441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_441_cast_fp16 = slice_by_index(begin = var_441_begin_0, end = var_441_end_0, end_mask = var_441_end_mask_0, x = var_270_cast_fp16)[name = string("op_441_cast_fp16")]; + tensor var_448_begin_0 = const()[name = string("op_448_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_448_end_0 = const()[name = string("op_448_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_448_end_mask_0 = const()[name = string("op_448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = var_270_cast_fp16)[name = string("op_448_cast_fp16")]; + tensor var_455_begin_0 = const()[name = string("op_455_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_455_end_0 = const()[name = string("op_455_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_455_end_mask_0 = const()[name = string("op_455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_455_cast_fp16 = slice_by_index(begin = var_455_begin_0, end = var_455_end_0, end_mask = var_455_end_mask_0, x = var_274_cast_fp16)[name = string("op_455_cast_fp16")]; + tensor var_462_begin_0 = const()[name = string("op_462_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_462_end_0 = const()[name = string("op_462_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_462_end_mask_0 = const()[name = string("op_462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_462_cast_fp16 = slice_by_index(begin = var_462_begin_0, end = var_462_end_0, end_mask = var_462_end_mask_0, x = var_274_cast_fp16)[name = string("op_462_cast_fp16")]; + tensor var_469_begin_0 = const()[name = string("op_469_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_469_end_0 = const()[name = string("op_469_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_469_end_mask_0 = const()[name = string("op_469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = var_274_cast_fp16)[name = string("op_469_cast_fp16")]; + tensor var_476_begin_0 = const()[name = string("op_476_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_476_end_0 = const()[name = string("op_476_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_476_end_mask_0 = const()[name = string("op_476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_476_cast_fp16 = slice_by_index(begin = var_476_begin_0, end = var_476_end_0, end_mask = var_476_end_mask_0, x = var_274_cast_fp16)[name = string("op_476_cast_fp16")]; + tensor var_483_begin_0 = const()[name = string("op_483_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_483_end_0 = const()[name = string("op_483_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_483_end_mask_0 = const()[name = string("op_483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_483_cast_fp16 = slice_by_index(begin = var_483_begin_0, end = var_483_end_0, end_mask = var_483_end_mask_0, x = var_278_cast_fp16)[name = string("op_483_cast_fp16")]; + tensor var_490_begin_0 = const()[name = string("op_490_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_490_end_0 = const()[name = string("op_490_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_490_end_mask_0 = const()[name = string("op_490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_490_cast_fp16 = slice_by_index(begin = var_490_begin_0, end = var_490_end_0, end_mask = var_490_end_mask_0, x = var_278_cast_fp16)[name = string("op_490_cast_fp16")]; + tensor var_497_begin_0 = const()[name = string("op_497_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_497_end_0 = const()[name = string("op_497_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_497_end_mask_0 = const()[name = string("op_497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_497_cast_fp16 = slice_by_index(begin = var_497_begin_0, end = var_497_end_0, end_mask = var_497_end_mask_0, x = var_278_cast_fp16)[name = string("op_497_cast_fp16")]; + tensor var_504_begin_0 = const()[name = string("op_504_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_504_end_0 = const()[name = string("op_504_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_504_end_mask_0 = const()[name = string("op_504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_504_cast_fp16 = slice_by_index(begin = var_504_begin_0, end = var_504_end_0, end_mask = var_504_end_mask_0, x = var_278_cast_fp16)[name = string("op_504_cast_fp16")]; + tensor var_511_begin_0 = const()[name = string("op_511_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_511_end_0 = const()[name = string("op_511_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_511_end_mask_0 = const()[name = string("op_511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = var_282_cast_fp16)[name = string("op_511_cast_fp16")]; + tensor var_518_begin_0 = const()[name = string("op_518_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_518_end_0 = const()[name = string("op_518_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_518_end_mask_0 = const()[name = string("op_518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_518_cast_fp16 = slice_by_index(begin = var_518_begin_0, end = var_518_end_0, end_mask = var_518_end_mask_0, x = var_282_cast_fp16)[name = string("op_518_cast_fp16")]; + tensor var_525_begin_0 = const()[name = string("op_525_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_525_end_0 = const()[name = string("op_525_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_525_end_mask_0 = const()[name = string("op_525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_525_cast_fp16 = slice_by_index(begin = var_525_begin_0, end = var_525_end_0, end_mask = var_525_end_mask_0, x = var_282_cast_fp16)[name = string("op_525_cast_fp16")]; + tensor var_532_begin_0 = const()[name = string("op_532_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_532_end_0 = const()[name = string("op_532_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_532_end_mask_0 = const()[name = string("op_532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_532_cast_fp16 = slice_by_index(begin = var_532_begin_0, end = var_532_end_0, end_mask = var_532_end_mask_0, x = var_282_cast_fp16)[name = string("op_532_cast_fp16")]; + tensor var_539_begin_0 = const()[name = string("op_539_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_539_end_0 = const()[name = string("op_539_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_539_end_mask_0 = const()[name = string("op_539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_539_cast_fp16 = slice_by_index(begin = var_539_begin_0, end = var_539_end_0, end_mask = var_539_end_mask_0, x = var_286_cast_fp16)[name = string("op_539_cast_fp16")]; + tensor var_546_begin_0 = const()[name = string("op_546_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_546_end_0 = const()[name = string("op_546_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_546_end_mask_0 = const()[name = string("op_546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_546_cast_fp16 = slice_by_index(begin = var_546_begin_0, end = var_546_end_0, end_mask = var_546_end_mask_0, x = var_286_cast_fp16)[name = string("op_546_cast_fp16")]; + tensor var_553_begin_0 = const()[name = string("op_553_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_553_end_0 = const()[name = string("op_553_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_553_end_mask_0 = const()[name = string("op_553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_553_cast_fp16 = slice_by_index(begin = var_553_begin_0, end = var_553_end_0, end_mask = var_553_end_mask_0, x = var_286_cast_fp16)[name = string("op_553_cast_fp16")]; + tensor var_560_begin_0 = const()[name = string("op_560_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_560_end_0 = const()[name = string("op_560_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_560_end_mask_0 = const()[name = string("op_560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_560_cast_fp16 = slice_by_index(begin = var_560_begin_0, end = var_560_end_0, end_mask = var_560_end_mask_0, x = var_286_cast_fp16)[name = string("op_560_cast_fp16")]; + tensor var_567_begin_0 = const()[name = string("op_567_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_567_end_0 = const()[name = string("op_567_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_567_end_mask_0 = const()[name = string("op_567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_567_cast_fp16 = slice_by_index(begin = var_567_begin_0, end = var_567_end_0, end_mask = var_567_end_mask_0, x = var_290_cast_fp16)[name = string("op_567_cast_fp16")]; + tensor var_574_begin_0 = const()[name = string("op_574_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_574_end_0 = const()[name = string("op_574_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_574_end_mask_0 = const()[name = string("op_574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_574_cast_fp16 = slice_by_index(begin = var_574_begin_0, end = var_574_end_0, end_mask = var_574_end_mask_0, x = var_290_cast_fp16)[name = string("op_574_cast_fp16")]; + tensor var_581_begin_0 = const()[name = string("op_581_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_581_end_0 = const()[name = string("op_581_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_581_end_mask_0 = const()[name = string("op_581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_581_cast_fp16 = slice_by_index(begin = var_581_begin_0, end = var_581_end_0, end_mask = var_581_end_mask_0, x = var_290_cast_fp16)[name = string("op_581_cast_fp16")]; + tensor var_588_begin_0 = const()[name = string("op_588_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_588_end_0 = const()[name = string("op_588_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_588_end_mask_0 = const()[name = string("op_588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_588_cast_fp16 = slice_by_index(begin = var_588_begin_0, end = var_588_end_0, end_mask = var_588_end_mask_0, x = var_290_cast_fp16)[name = string("op_588_cast_fp16")]; + tensor var_595_begin_0 = const()[name = string("op_595_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_595_end_0 = const()[name = string("op_595_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_595_end_mask_0 = const()[name = string("op_595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_595_cast_fp16 = slice_by_index(begin = var_595_begin_0, end = var_595_end_0, end_mask = var_595_end_mask_0, x = var_294_cast_fp16)[name = string("op_595_cast_fp16")]; + tensor var_602_begin_0 = const()[name = string("op_602_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_602_end_0 = const()[name = string("op_602_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_602_end_mask_0 = const()[name = string("op_602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_602_cast_fp16 = slice_by_index(begin = var_602_begin_0, end = var_602_end_0, end_mask = var_602_end_mask_0, x = var_294_cast_fp16)[name = string("op_602_cast_fp16")]; + tensor var_609_begin_0 = const()[name = string("op_609_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_609_end_0 = const()[name = string("op_609_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_609_end_mask_0 = const()[name = string("op_609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_609_cast_fp16 = slice_by_index(begin = var_609_begin_0, end = var_609_end_0, end_mask = var_609_end_mask_0, x = var_294_cast_fp16)[name = string("op_609_cast_fp16")]; + tensor var_616_begin_0 = const()[name = string("op_616_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_616_end_0 = const()[name = string("op_616_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_616_end_mask_0 = const()[name = string("op_616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_616_cast_fp16 = slice_by_index(begin = var_616_begin_0, end = var_616_end_0, end_mask = var_616_end_mask_0, x = var_294_cast_fp16)[name = string("op_616_cast_fp16")]; + tensor var_623_begin_0 = const()[name = string("op_623_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_623_end_0 = const()[name = string("op_623_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_623_end_mask_0 = const()[name = string("op_623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_623_cast_fp16 = slice_by_index(begin = var_623_begin_0, end = var_623_end_0, end_mask = var_623_end_mask_0, x = var_298_cast_fp16)[name = string("op_623_cast_fp16")]; + tensor var_630_begin_0 = const()[name = string("op_630_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_630_end_0 = const()[name = string("op_630_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_630_end_mask_0 = const()[name = string("op_630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_630_cast_fp16 = slice_by_index(begin = var_630_begin_0, end = var_630_end_0, end_mask = var_630_end_mask_0, x = var_298_cast_fp16)[name = string("op_630_cast_fp16")]; + tensor var_637_begin_0 = const()[name = string("op_637_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_637_end_0 = const()[name = string("op_637_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_637_end_mask_0 = const()[name = string("op_637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_637_cast_fp16 = slice_by_index(begin = var_637_begin_0, end = var_637_end_0, end_mask = var_637_end_mask_0, x = var_298_cast_fp16)[name = string("op_637_cast_fp16")]; + tensor var_644_begin_0 = const()[name = string("op_644_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_644_end_0 = const()[name = string("op_644_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_644_end_mask_0 = const()[name = string("op_644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_644_cast_fp16 = slice_by_index(begin = var_644_begin_0, end = var_644_end_0, end_mask = var_644_end_mask_0, x = var_298_cast_fp16)[name = string("op_644_cast_fp16")]; + tensor var_651_begin_0 = const()[name = string("op_651_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_651_end_0 = const()[name = string("op_651_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_651_end_mask_0 = const()[name = string("op_651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_651_cast_fp16 = slice_by_index(begin = var_651_begin_0, end = var_651_end_0, end_mask = var_651_end_mask_0, x = var_302_cast_fp16)[name = string("op_651_cast_fp16")]; + tensor var_658_begin_0 = const()[name = string("op_658_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_658_end_0 = const()[name = string("op_658_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_658_end_mask_0 = const()[name = string("op_658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_658_cast_fp16 = slice_by_index(begin = var_658_begin_0, end = var_658_end_0, end_mask = var_658_end_mask_0, x = var_302_cast_fp16)[name = string("op_658_cast_fp16")]; + tensor var_665_begin_0 = const()[name = string("op_665_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_665_end_0 = const()[name = string("op_665_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_665_end_mask_0 = const()[name = string("op_665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_665_cast_fp16 = slice_by_index(begin = var_665_begin_0, end = var_665_end_0, end_mask = var_665_end_mask_0, x = var_302_cast_fp16)[name = string("op_665_cast_fp16")]; + tensor var_672_begin_0 = const()[name = string("op_672_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_672_end_0 = const()[name = string("op_672_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_672_end_mask_0 = const()[name = string("op_672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_672_cast_fp16 = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, x = var_302_cast_fp16)[name = string("op_672_cast_fp16")]; + tensor var_679_begin_0 = const()[name = string("op_679_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_679_end_0 = const()[name = string("op_679_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_679_end_mask_0 = const()[name = string("op_679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_679_cast_fp16 = slice_by_index(begin = var_679_begin_0, end = var_679_end_0, end_mask = var_679_end_mask_0, x = var_306_cast_fp16)[name = string("op_679_cast_fp16")]; + tensor var_686_begin_0 = const()[name = string("op_686_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_686_end_0 = const()[name = string("op_686_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_686_end_mask_0 = const()[name = string("op_686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_686_cast_fp16 = slice_by_index(begin = var_686_begin_0, end = var_686_end_0, end_mask = var_686_end_mask_0, x = var_306_cast_fp16)[name = string("op_686_cast_fp16")]; + tensor var_693_begin_0 = const()[name = string("op_693_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_693_end_0 = const()[name = string("op_693_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_693_end_mask_0 = const()[name = string("op_693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_693_cast_fp16 = slice_by_index(begin = var_693_begin_0, end = var_693_end_0, end_mask = var_693_end_mask_0, x = var_306_cast_fp16)[name = string("op_693_cast_fp16")]; + tensor var_700_begin_0 = const()[name = string("op_700_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_700_end_0 = const()[name = string("op_700_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_700_end_mask_0 = const()[name = string("op_700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_700_cast_fp16 = slice_by_index(begin = var_700_begin_0, end = var_700_end_0, end_mask = var_700_end_mask_0, x = var_306_cast_fp16)[name = string("op_700_cast_fp16")]; + tensor var_707_begin_0 = const()[name = string("op_707_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_707_end_0 = const()[name = string("op_707_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_707_end_mask_0 = const()[name = string("op_707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = var_310_cast_fp16)[name = string("op_707_cast_fp16")]; + tensor var_714_begin_0 = const()[name = string("op_714_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_714_end_0 = const()[name = string("op_714_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_714_end_mask_0 = const()[name = string("op_714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_714_cast_fp16 = slice_by_index(begin = var_714_begin_0, end = var_714_end_0, end_mask = var_714_end_mask_0, x = var_310_cast_fp16)[name = string("op_714_cast_fp16")]; + tensor var_721_begin_0 = const()[name = string("op_721_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_721_end_0 = const()[name = string("op_721_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_721_end_mask_0 = const()[name = string("op_721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_721_cast_fp16 = slice_by_index(begin = var_721_begin_0, end = var_721_end_0, end_mask = var_721_end_mask_0, x = var_310_cast_fp16)[name = string("op_721_cast_fp16")]; + tensor var_728_begin_0 = const()[name = string("op_728_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_728_end_0 = const()[name = string("op_728_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_728_end_mask_0 = const()[name = string("op_728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_728_cast_fp16 = slice_by_index(begin = var_728_begin_0, end = var_728_end_0, end_mask = var_728_end_mask_0, x = var_310_cast_fp16)[name = string("op_728_cast_fp16")]; + tensor var_735_begin_0 = const()[name = string("op_735_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_735_end_0 = const()[name = string("op_735_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_735_end_mask_0 = const()[name = string("op_735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_735_cast_fp16 = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = var_314_cast_fp16)[name = string("op_735_cast_fp16")]; + tensor var_742_begin_0 = const()[name = string("op_742_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_742_end_0 = const()[name = string("op_742_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_742_end_mask_0 = const()[name = string("op_742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_742_cast_fp16 = slice_by_index(begin = var_742_begin_0, end = var_742_end_0, end_mask = var_742_end_mask_0, x = var_314_cast_fp16)[name = string("op_742_cast_fp16")]; + tensor var_749_begin_0 = const()[name = string("op_749_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_749_end_0 = const()[name = string("op_749_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_749_end_mask_0 = const()[name = string("op_749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_749_cast_fp16 = slice_by_index(begin = var_749_begin_0, end = var_749_end_0, end_mask = var_749_end_mask_0, x = var_314_cast_fp16)[name = string("op_749_cast_fp16")]; + tensor var_756_begin_0 = const()[name = string("op_756_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_756_end_0 = const()[name = string("op_756_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_756_end_mask_0 = const()[name = string("op_756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_756_cast_fp16 = slice_by_index(begin = var_756_begin_0, end = var_756_end_0, end_mask = var_756_end_mask_0, x = var_314_cast_fp16)[name = string("op_756_cast_fp16")]; + tensor var_763_begin_0 = const()[name = string("op_763_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_763_end_0 = const()[name = string("op_763_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_763_end_mask_0 = const()[name = string("op_763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_763_cast_fp16 = slice_by_index(begin = var_763_begin_0, end = var_763_end_0, end_mask = var_763_end_mask_0, x = var_318_cast_fp16)[name = string("op_763_cast_fp16")]; + tensor var_770_begin_0 = const()[name = string("op_770_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_770_end_0 = const()[name = string("op_770_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_770_end_mask_0 = const()[name = string("op_770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_770_cast_fp16 = slice_by_index(begin = var_770_begin_0, end = var_770_end_0, end_mask = var_770_end_mask_0, x = var_318_cast_fp16)[name = string("op_770_cast_fp16")]; + tensor var_777_begin_0 = const()[name = string("op_777_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_777_end_0 = const()[name = string("op_777_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_777_end_mask_0 = const()[name = string("op_777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_777_cast_fp16 = slice_by_index(begin = var_777_begin_0, end = var_777_end_0, end_mask = var_777_end_mask_0, x = var_318_cast_fp16)[name = string("op_777_cast_fp16")]; + tensor var_784_begin_0 = const()[name = string("op_784_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_784_end_0 = const()[name = string("op_784_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_784_end_mask_0 = const()[name = string("op_784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_784_cast_fp16 = slice_by_index(begin = var_784_begin_0, end = var_784_end_0, end_mask = var_784_end_mask_0, x = var_318_cast_fp16)[name = string("op_784_cast_fp16")]; + tensor var_791_begin_0 = const()[name = string("op_791_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_791_end_0 = const()[name = string("op_791_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_791_end_mask_0 = const()[name = string("op_791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_791_cast_fp16 = slice_by_index(begin = var_791_begin_0, end = var_791_end_0, end_mask = var_791_end_mask_0, x = var_322_cast_fp16)[name = string("op_791_cast_fp16")]; + tensor var_798_begin_0 = const()[name = string("op_798_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_798_end_0 = const()[name = string("op_798_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_798_end_mask_0 = const()[name = string("op_798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_798_cast_fp16 = slice_by_index(begin = var_798_begin_0, end = var_798_end_0, end_mask = var_798_end_mask_0, x = var_322_cast_fp16)[name = string("op_798_cast_fp16")]; + tensor var_805_begin_0 = const()[name = string("op_805_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_805_end_0 = const()[name = string("op_805_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_805_end_mask_0 = const()[name = string("op_805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_805_cast_fp16 = slice_by_index(begin = var_805_begin_0, end = var_805_end_0, end_mask = var_805_end_mask_0, x = var_322_cast_fp16)[name = string("op_805_cast_fp16")]; + tensor var_812_begin_0 = const()[name = string("op_812_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_812_end_0 = const()[name = string("op_812_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_812_end_mask_0 = const()[name = string("op_812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_812_cast_fp16 = slice_by_index(begin = var_812_begin_0, end = var_812_end_0, end_mask = var_812_end_mask_0, x = var_322_cast_fp16)[name = string("op_812_cast_fp16")]; + tensor var_819_begin_0 = const()[name = string("op_819_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_819_end_0 = const()[name = string("op_819_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_819_end_mask_0 = const()[name = string("op_819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_819_cast_fp16 = slice_by_index(begin = var_819_begin_0, end = var_819_end_0, end_mask = var_819_end_mask_0, x = var_326_cast_fp16)[name = string("op_819_cast_fp16")]; + tensor var_826_begin_0 = const()[name = string("op_826_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_826_end_0 = const()[name = string("op_826_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_826_end_mask_0 = const()[name = string("op_826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_826_cast_fp16 = slice_by_index(begin = var_826_begin_0, end = var_826_end_0, end_mask = var_826_end_mask_0, x = var_326_cast_fp16)[name = string("op_826_cast_fp16")]; + tensor var_833_begin_0 = const()[name = string("op_833_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_833_end_0 = const()[name = string("op_833_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_833_end_mask_0 = const()[name = string("op_833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_833_cast_fp16 = slice_by_index(begin = var_833_begin_0, end = var_833_end_0, end_mask = var_833_end_mask_0, x = var_326_cast_fp16)[name = string("op_833_cast_fp16")]; + tensor var_840_begin_0 = const()[name = string("op_840_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_840_end_0 = const()[name = string("op_840_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_840_end_mask_0 = const()[name = string("op_840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_840_cast_fp16 = slice_by_index(begin = var_840_begin_0, end = var_840_end_0, end_mask = var_840_end_mask_0, x = var_326_cast_fp16)[name = string("op_840_cast_fp16")]; + tensor var_847_begin_0 = const()[name = string("op_847_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_847_end_0 = const()[name = string("op_847_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_847_end_mask_0 = const()[name = string("op_847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_847_cast_fp16 = slice_by_index(begin = var_847_begin_0, end = var_847_end_0, end_mask = var_847_end_mask_0, x = var_330_cast_fp16)[name = string("op_847_cast_fp16")]; + tensor var_854_begin_0 = const()[name = string("op_854_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_854_end_0 = const()[name = string("op_854_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_854_end_mask_0 = const()[name = string("op_854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_854_cast_fp16 = slice_by_index(begin = var_854_begin_0, end = var_854_end_0, end_mask = var_854_end_mask_0, x = var_330_cast_fp16)[name = string("op_854_cast_fp16")]; + tensor var_861_begin_0 = const()[name = string("op_861_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_861_end_0 = const()[name = string("op_861_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_861_end_mask_0 = const()[name = string("op_861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_861_cast_fp16 = slice_by_index(begin = var_861_begin_0, end = var_861_end_0, end_mask = var_861_end_mask_0, x = var_330_cast_fp16)[name = string("op_861_cast_fp16")]; + tensor var_868_begin_0 = const()[name = string("op_868_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_868_end_0 = const()[name = string("op_868_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_868_end_mask_0 = const()[name = string("op_868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_868_cast_fp16 = slice_by_index(begin = var_868_begin_0, end = var_868_end_0, end_mask = var_868_end_mask_0, x = var_330_cast_fp16)[name = string("op_868_cast_fp16")]; + tensor var_875_begin_0 = const()[name = string("op_875_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_875_end_0 = const()[name = string("op_875_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_875_end_mask_0 = const()[name = string("op_875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_875_cast_fp16 = slice_by_index(begin = var_875_begin_0, end = var_875_end_0, end_mask = var_875_end_mask_0, x = var_334_cast_fp16)[name = string("op_875_cast_fp16")]; + tensor var_882_begin_0 = const()[name = string("op_882_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_882_end_0 = const()[name = string("op_882_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_882_end_mask_0 = const()[name = string("op_882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_882_cast_fp16 = slice_by_index(begin = var_882_begin_0, end = var_882_end_0, end_mask = var_882_end_mask_0, x = var_334_cast_fp16)[name = string("op_882_cast_fp16")]; + tensor var_889_begin_0 = const()[name = string("op_889_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_889_end_0 = const()[name = string("op_889_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_889_end_mask_0 = const()[name = string("op_889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_889_cast_fp16 = slice_by_index(begin = var_889_begin_0, end = var_889_end_0, end_mask = var_889_end_mask_0, x = var_334_cast_fp16)[name = string("op_889_cast_fp16")]; + tensor var_896_begin_0 = const()[name = string("op_896_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_896_end_0 = const()[name = string("op_896_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_896_end_mask_0 = const()[name = string("op_896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_896_cast_fp16 = slice_by_index(begin = var_896_begin_0, end = var_896_end_0, end_mask = var_896_end_mask_0, x = var_334_cast_fp16)[name = string("op_896_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_31")]; + tensor var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = k_1_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_905_begin_0 = const()[name = string("op_905_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_905_end_0 = const()[name = string("op_905_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_905_end_mask_0 = const()[name = string("op_905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_905_cast_fp16 = slice_by_index(begin = var_905_begin_0, end = var_905_end_0, end_mask = var_905_end_mask_0, x = k_1_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_909_begin_0 = const()[name = string("op_909_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_909_end_0 = const()[name = string("op_909_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_909_end_mask_0 = const()[name = string("op_909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_909_cast_fp16 = slice_by_index(begin = var_909_begin_0, end = var_909_end_0, end_mask = var_909_end_mask_0, x = k_1_cast_fp16)[name = string("op_909_cast_fp16")]; + tensor var_913_begin_0 = const()[name = string("op_913_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_913_end_0 = const()[name = string("op_913_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_913_end_mask_0 = const()[name = string("op_913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_913_cast_fp16 = slice_by_index(begin = var_913_begin_0, end = var_913_end_0, end_mask = var_913_end_mask_0, x = k_1_cast_fp16)[name = string("op_913_cast_fp16")]; + tensor var_917_begin_0 = const()[name = string("op_917_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_917_end_0 = const()[name = string("op_917_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_917_end_mask_0 = const()[name = string("op_917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_917_cast_fp16 = slice_by_index(begin = var_917_begin_0, end = var_917_end_0, end_mask = var_917_end_mask_0, x = k_1_cast_fp16)[name = string("op_917_cast_fp16")]; + tensor var_921_begin_0 = const()[name = string("op_921_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_921_end_0 = const()[name = string("op_921_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_921_end_mask_0 = const()[name = string("op_921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_921_cast_fp16 = slice_by_index(begin = var_921_begin_0, end = var_921_end_0, end_mask = var_921_end_mask_0, x = k_1_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_925_begin_0 = const()[name = string("op_925_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_925_end_0 = const()[name = string("op_925_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_925_end_mask_0 = const()[name = string("op_925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_925_cast_fp16 = slice_by_index(begin = var_925_begin_0, end = var_925_end_0, end_mask = var_925_end_mask_0, x = k_1_cast_fp16)[name = string("op_925_cast_fp16")]; + tensor var_929_begin_0 = const()[name = string("op_929_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_929_end_0 = const()[name = string("op_929_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_929_end_mask_0 = const()[name = string("op_929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_929_cast_fp16 = slice_by_index(begin = var_929_begin_0, end = var_929_end_0, end_mask = var_929_end_mask_0, x = k_1_cast_fp16)[name = string("op_929_cast_fp16")]; + tensor var_933_begin_0 = const()[name = string("op_933_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_933_end_0 = const()[name = string("op_933_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_933_end_mask_0 = const()[name = string("op_933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_933_cast_fp16 = slice_by_index(begin = var_933_begin_0, end = var_933_end_0, end_mask = var_933_end_mask_0, x = k_1_cast_fp16)[name = string("op_933_cast_fp16")]; + tensor var_937_begin_0 = const()[name = string("op_937_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_937_end_0 = const()[name = string("op_937_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_937_end_mask_0 = const()[name = string("op_937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_937_cast_fp16 = slice_by_index(begin = var_937_begin_0, end = var_937_end_0, end_mask = var_937_end_mask_0, x = k_1_cast_fp16)[name = string("op_937_cast_fp16")]; + tensor var_941_begin_0 = const()[name = string("op_941_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_941_end_0 = const()[name = string("op_941_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_941_end_mask_0 = const()[name = string("op_941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_941_cast_fp16 = slice_by_index(begin = var_941_begin_0, end = var_941_end_0, end_mask = var_941_end_mask_0, x = k_1_cast_fp16)[name = string("op_941_cast_fp16")]; + tensor var_945_begin_0 = const()[name = string("op_945_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_945_end_0 = const()[name = string("op_945_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_945_end_mask_0 = const()[name = string("op_945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_945_cast_fp16 = slice_by_index(begin = var_945_begin_0, end = var_945_end_0, end_mask = var_945_end_mask_0, x = k_1_cast_fp16)[name = string("op_945_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = k_1_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_953_begin_0 = const()[name = string("op_953_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_953_end_0 = const()[name = string("op_953_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_953_end_mask_0 = const()[name = string("op_953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = k_1_cast_fp16)[name = string("op_953_cast_fp16")]; + tensor var_957_begin_0 = const()[name = string("op_957_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_957_end_0 = const()[name = string("op_957_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_957_end_mask_0 = const()[name = string("op_957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_957_cast_fp16 = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = k_1_cast_fp16)[name = string("op_957_cast_fp16")]; + tensor var_961_begin_0 = const()[name = string("op_961_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_961_end_0 = const()[name = string("op_961_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_961_end_mask_0 = const()[name = string("op_961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_961_cast_fp16 = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = k_1_cast_fp16)[name = string("op_961_cast_fp16")]; + tensor var_965_begin_0 = const()[name = string("op_965_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_965_end_0 = const()[name = string("op_965_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_965_end_mask_0 = const()[name = string("op_965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_965_cast_fp16 = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = k_1_cast_fp16)[name = string("op_965_cast_fp16")]; + tensor var_969_begin_0 = const()[name = string("op_969_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_969_end_0 = const()[name = string("op_969_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_969_end_mask_0 = const()[name = string("op_969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_969_cast_fp16 = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = k_1_cast_fp16)[name = string("op_969_cast_fp16")]; + tensor var_973_begin_0 = const()[name = string("op_973_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_973_end_0 = const()[name = string("op_973_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_973_end_mask_0 = const()[name = string("op_973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = k_1_cast_fp16)[name = string("op_973_cast_fp16")]; + tensor var_977_begin_0 = const()[name = string("op_977_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_977_end_0 = const()[name = string("op_977_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_977_end_mask_0 = const()[name = string("op_977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = k_1_cast_fp16)[name = string("op_977_cast_fp16")]; + tensor var_979_begin_0 = const()[name = string("op_979_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_979_end_0 = const()[name = string("op_979_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_979_end_mask_0 = const()[name = string("op_979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_979_cast_fp16 = slice_by_index(begin = var_979_begin_0, end = var_979_end_0, end_mask = var_979_end_mask_0, x = value_1_cast_fp16)[name = string("op_979_cast_fp16")]; + tensor var_983_begin_0 = const()[name = string("op_983_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_983_end_0 = const()[name = string("op_983_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_983_end_mask_0 = const()[name = string("op_983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_983_cast_fp16 = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = value_1_cast_fp16)[name = string("op_983_cast_fp16")]; + tensor var_987_begin_0 = const()[name = string("op_987_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_987_end_0 = const()[name = string("op_987_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_987_end_mask_0 = const()[name = string("op_987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_987_cast_fp16 = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = value_1_cast_fp16)[name = string("op_987_cast_fp16")]; + tensor var_991_begin_0 = const()[name = string("op_991_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_991_end_0 = const()[name = string("op_991_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_991_end_mask_0 = const()[name = string("op_991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = value_1_cast_fp16)[name = string("op_991_cast_fp16")]; + tensor var_995_begin_0 = const()[name = string("op_995_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_995_end_0 = const()[name = string("op_995_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_995_end_mask_0 = const()[name = string("op_995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_995_cast_fp16 = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = value_1_cast_fp16)[name = string("op_995_cast_fp16")]; + tensor var_999_begin_0 = const()[name = string("op_999_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_999_end_0 = const()[name = string("op_999_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_999_end_mask_0 = const()[name = string("op_999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_999_cast_fp16 = slice_by_index(begin = var_999_begin_0, end = var_999_end_0, end_mask = var_999_end_mask_0, x = value_1_cast_fp16)[name = string("op_999_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = string("op_1003_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1003_end_0 = const()[name = string("op_1003_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1003_end_mask_0 = const()[name = string("op_1003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = value_1_cast_fp16)[name = string("op_1003_cast_fp16")]; + tensor var_1007_begin_0 = const()[name = string("op_1007_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1007_end_0 = const()[name = string("op_1007_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1007_end_mask_0 = const()[name = string("op_1007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1007_cast_fp16 = slice_by_index(begin = var_1007_begin_0, end = var_1007_end_0, end_mask = var_1007_end_mask_0, x = value_1_cast_fp16)[name = string("op_1007_cast_fp16")]; + tensor var_1011_begin_0 = const()[name = string("op_1011_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1011_end_0 = const()[name = string("op_1011_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1011_end_mask_0 = const()[name = string("op_1011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1011_cast_fp16 = slice_by_index(begin = var_1011_begin_0, end = var_1011_end_0, end_mask = var_1011_end_mask_0, x = value_1_cast_fp16)[name = string("op_1011_cast_fp16")]; + tensor var_1015_begin_0 = const()[name = string("op_1015_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1015_end_0 = const()[name = string("op_1015_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1015_end_mask_0 = const()[name = string("op_1015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1015_cast_fp16 = slice_by_index(begin = var_1015_begin_0, end = var_1015_end_0, end_mask = var_1015_end_mask_0, x = value_1_cast_fp16)[name = string("op_1015_cast_fp16")]; + tensor var_1019_begin_0 = const()[name = string("op_1019_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1019_end_0 = const()[name = string("op_1019_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1019_end_mask_0 = const()[name = string("op_1019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = var_1019_end_0, end_mask = var_1019_end_mask_0, x = value_1_cast_fp16)[name = string("op_1019_cast_fp16")]; + tensor var_1023_begin_0 = const()[name = string("op_1023_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1023_end_0 = const()[name = string("op_1023_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1023_end_mask_0 = const()[name = string("op_1023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1023_cast_fp16 = slice_by_index(begin = var_1023_begin_0, end = var_1023_end_0, end_mask = var_1023_end_mask_0, x = value_1_cast_fp16)[name = string("op_1023_cast_fp16")]; + tensor var_1027_begin_0 = const()[name = string("op_1027_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_1027_end_0 = const()[name = string("op_1027_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_1027_end_mask_0 = const()[name = string("op_1027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1027_cast_fp16 = slice_by_index(begin = var_1027_begin_0, end = var_1027_end_0, end_mask = var_1027_end_mask_0, x = value_1_cast_fp16)[name = string("op_1027_cast_fp16")]; + tensor var_1031_begin_0 = const()[name = string("op_1031_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_1031_end_0 = const()[name = string("op_1031_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_1031_end_mask_0 = const()[name = string("op_1031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1031_cast_fp16 = slice_by_index(begin = var_1031_begin_0, end = var_1031_end_0, end_mask = var_1031_end_mask_0, x = value_1_cast_fp16)[name = string("op_1031_cast_fp16")]; + tensor var_1035_begin_0 = const()[name = string("op_1035_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_1035_end_0 = const()[name = string("op_1035_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_1035_end_mask_0 = const()[name = string("op_1035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1035_cast_fp16 = slice_by_index(begin = var_1035_begin_0, end = var_1035_end_0, end_mask = var_1035_end_mask_0, x = value_1_cast_fp16)[name = string("op_1035_cast_fp16")]; + tensor var_1039_begin_0 = const()[name = string("op_1039_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_1039_end_0 = const()[name = string("op_1039_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_1039_end_mask_0 = const()[name = string("op_1039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = value_1_cast_fp16)[name = string("op_1039_cast_fp16")]; + tensor var_1043_begin_0 = const()[name = string("op_1043_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_1043_end_0 = const()[name = string("op_1043_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_1043_end_mask_0 = const()[name = string("op_1043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1043_cast_fp16 = slice_by_index(begin = var_1043_begin_0, end = var_1043_end_0, end_mask = var_1043_end_mask_0, x = value_1_cast_fp16)[name = string("op_1043_cast_fp16")]; + tensor var_1047_begin_0 = const()[name = string("op_1047_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_1047_end_0 = const()[name = string("op_1047_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_1047_end_mask_0 = const()[name = string("op_1047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = value_1_cast_fp16)[name = string("op_1047_cast_fp16")]; + tensor var_1051_begin_0 = const()[name = string("op_1051_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_1051_end_0 = const()[name = string("op_1051_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_1051_end_mask_0 = const()[name = string("op_1051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1051_cast_fp16 = slice_by_index(begin = var_1051_begin_0, end = var_1051_end_0, end_mask = var_1051_end_mask_0, x = value_1_cast_fp16)[name = string("op_1051_cast_fp16")]; + tensor var_1055_begin_0 = const()[name = string("op_1055_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_1055_end_0 = const()[name = string("op_1055_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_1055_end_mask_0 = const()[name = string("op_1055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1055_cast_fp16 = slice_by_index(begin = var_1055_begin_0, end = var_1055_end_0, end_mask = var_1055_end_mask_0, x = value_1_cast_fp16)[name = string("op_1055_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_901_cast_fp16, var_343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_901_cast_fp16, var_350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_901_cast_fp16, var_357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_901_cast_fp16, var_364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_905_cast_fp16, var_371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_905_cast_fp16, var_378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_905_cast_fp16, var_385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_905_cast_fp16, var_392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_909_cast_fp16, var_399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_909_cast_fp16, var_406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_909_cast_fp16, var_413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_909_cast_fp16, var_420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_913_cast_fp16, var_427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_913_cast_fp16, var_434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_913_cast_fp16, var_441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_913_cast_fp16, var_448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_917_cast_fp16, var_455_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_917_cast_fp16, var_462_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_917_cast_fp16, var_469_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_917_cast_fp16, var_476_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_921_cast_fp16, var_483_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_921_cast_fp16, var_490_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_921_cast_fp16, var_497_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_921_cast_fp16, var_504_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_925_cast_fp16, var_511_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_925_cast_fp16, var_518_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_925_cast_fp16, var_525_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_925_cast_fp16, var_532_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_929_cast_fp16, var_539_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_929_cast_fp16, var_546_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_929_cast_fp16, var_553_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_929_cast_fp16, var_560_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_933_cast_fp16, var_567_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_933_cast_fp16, var_574_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_933_cast_fp16, var_581_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_933_cast_fp16, var_588_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_937_cast_fp16, var_595_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_937_cast_fp16, var_602_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_937_cast_fp16, var_609_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_937_cast_fp16, var_616_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_941_cast_fp16, var_623_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_941_cast_fp16, var_630_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_941_cast_fp16, var_637_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_941_cast_fp16, var_644_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_945_cast_fp16, var_651_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_945_cast_fp16, var_658_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_945_cast_fp16, var_665_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_945_cast_fp16, var_672_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_949_cast_fp16, var_679_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_949_cast_fp16, var_686_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_949_cast_fp16, var_693_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_949_cast_fp16, var_700_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_953_cast_fp16, var_707_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_953_cast_fp16, var_714_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_953_cast_fp16, var_721_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_953_cast_fp16, var_728_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_957_cast_fp16, var_735_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_957_cast_fp16, var_742_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_957_cast_fp16, var_749_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_957_cast_fp16, var_756_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_961_cast_fp16, var_763_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_961_cast_fp16, var_770_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_961_cast_fp16, var_777_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_961_cast_fp16, var_784_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_965_cast_fp16, var_791_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_965_cast_fp16, var_798_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_965_cast_fp16, var_805_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_965_cast_fp16, var_812_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_969_cast_fp16, var_819_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_969_cast_fp16, var_826_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_969_cast_fp16, var_833_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_969_cast_fp16, var_840_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_973_cast_fp16, var_847_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_973_cast_fp16, var_854_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_973_cast_fp16, var_861_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_973_cast_fp16, var_868_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_977_cast_fp16, var_875_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_977_cast_fp16, var_882_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_977_cast_fp16, var_889_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_977_cast_fp16, var_896_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + fp16 var_1218_to_fp16 = const()[name = string("op_1218_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_1218_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_1220_to_fp16 = const()[name = string("op_1220_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_1220_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_1222_to_fp16 = const()[name = string("op_1222_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_1222_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_1224_to_fp16 = const()[name = string("op_1224_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_1224_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_1226_to_fp16 = const()[name = string("op_1226_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_1226_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_1228_to_fp16 = const()[name = string("op_1228_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_1228_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_1230_to_fp16 = const()[name = string("op_1230_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_1230_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_1232_to_fp16 = const()[name = string("op_1232_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_1232_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_1234_to_fp16 = const()[name = string("op_1234_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_1234_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_1236_to_fp16 = const()[name = string("op_1236_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_1236_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_1238_to_fp16 = const()[name = string("op_1238_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_1238_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_1240_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_1242_to_fp16 = const()[name = string("op_1242_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_1242_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_1244_to_fp16 = const()[name = string("op_1244_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_1244_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_1246_to_fp16 = const()[name = string("op_1246_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_1246_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_1248_to_fp16 = const()[name = string("op_1248_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_1248_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_1250_to_fp16 = const()[name = string("op_1250_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_1250_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_1252_to_fp16 = const()[name = string("op_1252_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_1252_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_1254_to_fp16 = const()[name = string("op_1254_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_1254_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_1256_to_fp16 = const()[name = string("op_1256_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_1256_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_1258_to_fp16 = const()[name = string("op_1258_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_1258_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_1260_to_fp16 = const()[name = string("op_1260_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_1260_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_1262_to_fp16 = const()[name = string("op_1262_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_1262_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_1264_to_fp16 = const()[name = string("op_1264_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_1264_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_1266_to_fp16 = const()[name = string("op_1266_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_1266_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_1268_to_fp16 = const()[name = string("op_1268_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_1268_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_1270_to_fp16 = const()[name = string("op_1270_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_1270_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_1272_to_fp16 = const()[name = string("op_1272_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_1272_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_1274_to_fp16 = const()[name = string("op_1274_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_1274_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_1276_to_fp16 = const()[name = string("op_1276_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_1276_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_1278_to_fp16 = const()[name = string("op_1278_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_1278_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_1280_to_fp16 = const()[name = string("op_1280_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_1280_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + fp16 var_1282_to_fp16 = const()[name = string("op_1282_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1282_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1284_to_fp16 = const()[name = string("op_1284_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1284_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1286_to_fp16 = const()[name = string("op_1286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1286_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1288_to_fp16 = const()[name = string("op_1288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1288_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1290_to_fp16 = const()[name = string("op_1290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1290_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1292_to_fp16 = const()[name = string("op_1292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1292_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1294_to_fp16 = const()[name = string("op_1294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1294_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1296_to_fp16 = const()[name = string("op_1296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1296_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1298_to_fp16 = const()[name = string("op_1298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1298_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1300_to_fp16 = const()[name = string("op_1300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1300_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1302_to_fp16 = const()[name = string("op_1302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1302_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1304_to_fp16 = const()[name = string("op_1304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1304_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1306_to_fp16 = const()[name = string("op_1306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1306_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1308_to_fp16 = const()[name = string("op_1308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1308_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1310_to_fp16 = const()[name = string("op_1310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1310_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1312_to_fp16 = const()[name = string("op_1312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1312_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + fp16 var_1314_to_fp16 = const()[name = string("op_1314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1314_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1316_to_fp16 = const()[name = string("op_1316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1316_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1318_to_fp16 = const()[name = string("op_1318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1318_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1320_to_fp16 = const()[name = string("op_1320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1320_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1322_to_fp16 = const()[name = string("op_1322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1322_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1324_to_fp16 = const()[name = string("op_1324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1324_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1326_to_fp16 = const()[name = string("op_1326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1326_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1328_to_fp16 = const()[name = string("op_1328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1328_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1330_to_fp16 = const()[name = string("op_1330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1330_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1332_to_fp16 = const()[name = string("op_1332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1332_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1334_to_fp16 = const()[name = string("op_1334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1334_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1336_to_fp16 = const()[name = string("op_1336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1336_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1338_to_fp16 = const()[name = string("op_1338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1338_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1340_to_fp16 = const()[name = string("op_1340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1340_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1342_to_fp16 = const()[name = string("op_1342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1342_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1344_to_fp16 = const()[name = string("op_1344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1344_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + fp16 var_1346_to_fp16 = const()[name = string("op_1346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1346_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1348_to_fp16 = const()[name = string("op_1348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1348_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1350_to_fp16 = const()[name = string("op_1350_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1350_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1352_to_fp16 = const()[name = string("op_1352_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1352_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1354_to_fp16 = const()[name = string("op_1354_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1354_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1356_to_fp16 = const()[name = string("op_1356_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1356_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1358_to_fp16 = const()[name = string("op_1358_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1358_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1360_to_fp16 = const()[name = string("op_1360_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1360_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1362_to_fp16 = const()[name = string("op_1362_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1362_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1364_to_fp16 = const()[name = string("op_1364_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1364_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1366_to_fp16 = const()[name = string("op_1366_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1366_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1368_to_fp16 = const()[name = string("op_1368_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1368_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1370_to_fp16 = const()[name = string("op_1370_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1370_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1372_to_fp16 = const()[name = string("op_1372_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1372_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1374_to_fp16 = const()[name = string("op_1374_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1374_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1376_to_fp16 = const()[name = string("op_1376_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1376_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + tensor var_1378_cast_fp16 = softmax(axis = var_203, x = aw_chunk_1_cast_fp16)[name = string("op_1378_cast_fp16")]; + tensor var_1379_cast_fp16 = softmax(axis = var_203, x = aw_chunk_3_cast_fp16)[name = string("op_1379_cast_fp16")]; + tensor var_1380_cast_fp16 = softmax(axis = var_203, x = aw_chunk_5_cast_fp16)[name = string("op_1380_cast_fp16")]; + tensor var_1381_cast_fp16 = softmax(axis = var_203, x = aw_chunk_7_cast_fp16)[name = string("op_1381_cast_fp16")]; + tensor var_1382_cast_fp16 = softmax(axis = var_203, x = aw_chunk_9_cast_fp16)[name = string("op_1382_cast_fp16")]; + tensor var_1383_cast_fp16 = softmax(axis = var_203, x = aw_chunk_11_cast_fp16)[name = string("op_1383_cast_fp16")]; + tensor var_1384_cast_fp16 = softmax(axis = var_203, x = aw_chunk_13_cast_fp16)[name = string("op_1384_cast_fp16")]; + tensor var_1385_cast_fp16 = softmax(axis = var_203, x = aw_chunk_15_cast_fp16)[name = string("op_1385_cast_fp16")]; + tensor var_1386_cast_fp16 = softmax(axis = var_203, x = aw_chunk_17_cast_fp16)[name = string("op_1386_cast_fp16")]; + tensor var_1387_cast_fp16 = softmax(axis = var_203, x = aw_chunk_19_cast_fp16)[name = string("op_1387_cast_fp16")]; + tensor var_1388_cast_fp16 = softmax(axis = var_203, x = aw_chunk_21_cast_fp16)[name = string("op_1388_cast_fp16")]; + tensor var_1389_cast_fp16 = softmax(axis = var_203, x = aw_chunk_23_cast_fp16)[name = string("op_1389_cast_fp16")]; + tensor var_1390_cast_fp16 = softmax(axis = var_203, x = aw_chunk_25_cast_fp16)[name = string("op_1390_cast_fp16")]; + tensor var_1391_cast_fp16 = softmax(axis = var_203, x = aw_chunk_27_cast_fp16)[name = string("op_1391_cast_fp16")]; + tensor var_1392_cast_fp16 = softmax(axis = var_203, x = aw_chunk_29_cast_fp16)[name = string("op_1392_cast_fp16")]; + tensor var_1393_cast_fp16 = softmax(axis = var_203, x = aw_chunk_31_cast_fp16)[name = string("op_1393_cast_fp16")]; + tensor var_1394_cast_fp16 = softmax(axis = var_203, x = aw_chunk_33_cast_fp16)[name = string("op_1394_cast_fp16")]; + tensor var_1395_cast_fp16 = softmax(axis = var_203, x = aw_chunk_35_cast_fp16)[name = string("op_1395_cast_fp16")]; + tensor var_1396_cast_fp16 = softmax(axis = var_203, x = aw_chunk_37_cast_fp16)[name = string("op_1396_cast_fp16")]; + tensor var_1397_cast_fp16 = softmax(axis = var_203, x = aw_chunk_39_cast_fp16)[name = string("op_1397_cast_fp16")]; + tensor var_1398_cast_fp16 = softmax(axis = var_203, x = aw_chunk_41_cast_fp16)[name = string("op_1398_cast_fp16")]; + tensor var_1399_cast_fp16 = softmax(axis = var_203, x = aw_chunk_43_cast_fp16)[name = string("op_1399_cast_fp16")]; + tensor var_1400_cast_fp16 = softmax(axis = var_203, x = aw_chunk_45_cast_fp16)[name = string("op_1400_cast_fp16")]; + tensor var_1401_cast_fp16 = softmax(axis = var_203, x = aw_chunk_47_cast_fp16)[name = string("op_1401_cast_fp16")]; + tensor var_1402_cast_fp16 = softmax(axis = var_203, x = aw_chunk_49_cast_fp16)[name = string("op_1402_cast_fp16")]; + tensor var_1403_cast_fp16 = softmax(axis = var_203, x = aw_chunk_51_cast_fp16)[name = string("op_1403_cast_fp16")]; + tensor var_1404_cast_fp16 = softmax(axis = var_203, x = aw_chunk_53_cast_fp16)[name = string("op_1404_cast_fp16")]; + tensor var_1405_cast_fp16 = softmax(axis = var_203, x = aw_chunk_55_cast_fp16)[name = string("op_1405_cast_fp16")]; + tensor var_1406_cast_fp16 = softmax(axis = var_203, x = aw_chunk_57_cast_fp16)[name = string("op_1406_cast_fp16")]; + tensor var_1407_cast_fp16 = softmax(axis = var_203, x = aw_chunk_59_cast_fp16)[name = string("op_1407_cast_fp16")]; + tensor var_1408_cast_fp16 = softmax(axis = var_203, x = aw_chunk_61_cast_fp16)[name = string("op_1408_cast_fp16")]; + tensor var_1409_cast_fp16 = softmax(axis = var_203, x = aw_chunk_63_cast_fp16)[name = string("op_1409_cast_fp16")]; + tensor var_1410_cast_fp16 = softmax(axis = var_203, x = aw_chunk_65_cast_fp16)[name = string("op_1410_cast_fp16")]; + tensor var_1411_cast_fp16 = softmax(axis = var_203, x = aw_chunk_67_cast_fp16)[name = string("op_1411_cast_fp16")]; + tensor var_1412_cast_fp16 = softmax(axis = var_203, x = aw_chunk_69_cast_fp16)[name = string("op_1412_cast_fp16")]; + tensor var_1413_cast_fp16 = softmax(axis = var_203, x = aw_chunk_71_cast_fp16)[name = string("op_1413_cast_fp16")]; + tensor var_1414_cast_fp16 = softmax(axis = var_203, x = aw_chunk_73_cast_fp16)[name = string("op_1414_cast_fp16")]; + tensor var_1415_cast_fp16 = softmax(axis = var_203, x = aw_chunk_75_cast_fp16)[name = string("op_1415_cast_fp16")]; + tensor var_1416_cast_fp16 = softmax(axis = var_203, x = aw_chunk_77_cast_fp16)[name = string("op_1416_cast_fp16")]; + tensor var_1417_cast_fp16 = softmax(axis = var_203, x = aw_chunk_79_cast_fp16)[name = string("op_1417_cast_fp16")]; + tensor var_1418_cast_fp16 = softmax(axis = var_203, x = aw_chunk_81_cast_fp16)[name = string("op_1418_cast_fp16")]; + tensor var_1419_cast_fp16 = softmax(axis = var_203, x = aw_chunk_83_cast_fp16)[name = string("op_1419_cast_fp16")]; + tensor var_1420_cast_fp16 = softmax(axis = var_203, x = aw_chunk_85_cast_fp16)[name = string("op_1420_cast_fp16")]; + tensor var_1421_cast_fp16 = softmax(axis = var_203, x = aw_chunk_87_cast_fp16)[name = string("op_1421_cast_fp16")]; + tensor var_1422_cast_fp16 = softmax(axis = var_203, x = aw_chunk_89_cast_fp16)[name = string("op_1422_cast_fp16")]; + tensor var_1423_cast_fp16 = softmax(axis = var_203, x = aw_chunk_91_cast_fp16)[name = string("op_1423_cast_fp16")]; + tensor var_1424_cast_fp16 = softmax(axis = var_203, x = aw_chunk_93_cast_fp16)[name = string("op_1424_cast_fp16")]; + tensor var_1425_cast_fp16 = softmax(axis = var_203, x = aw_chunk_95_cast_fp16)[name = string("op_1425_cast_fp16")]; + tensor var_1426_cast_fp16 = softmax(axis = var_203, x = aw_chunk_97_cast_fp16)[name = string("op_1426_cast_fp16")]; + tensor var_1427_cast_fp16 = softmax(axis = var_203, x = aw_chunk_99_cast_fp16)[name = string("op_1427_cast_fp16")]; + tensor var_1428_cast_fp16 = softmax(axis = var_203, x = aw_chunk_101_cast_fp16)[name = string("op_1428_cast_fp16")]; + tensor var_1429_cast_fp16 = softmax(axis = var_203, x = aw_chunk_103_cast_fp16)[name = string("op_1429_cast_fp16")]; + tensor var_1430_cast_fp16 = softmax(axis = var_203, x = aw_chunk_105_cast_fp16)[name = string("op_1430_cast_fp16")]; + tensor var_1431_cast_fp16 = softmax(axis = var_203, x = aw_chunk_107_cast_fp16)[name = string("op_1431_cast_fp16")]; + tensor var_1432_cast_fp16 = softmax(axis = var_203, x = aw_chunk_109_cast_fp16)[name = string("op_1432_cast_fp16")]; + tensor var_1433_cast_fp16 = softmax(axis = var_203, x = aw_chunk_111_cast_fp16)[name = string("op_1433_cast_fp16")]; + tensor var_1434_cast_fp16 = softmax(axis = var_203, x = aw_chunk_113_cast_fp16)[name = string("op_1434_cast_fp16")]; + tensor var_1435_cast_fp16 = softmax(axis = var_203, x = aw_chunk_115_cast_fp16)[name = string("op_1435_cast_fp16")]; + tensor var_1436_cast_fp16 = softmax(axis = var_203, x = aw_chunk_117_cast_fp16)[name = string("op_1436_cast_fp16")]; + tensor var_1437_cast_fp16 = softmax(axis = var_203, x = aw_chunk_119_cast_fp16)[name = string("op_1437_cast_fp16")]; + tensor var_1438_cast_fp16 = softmax(axis = var_203, x = aw_chunk_121_cast_fp16)[name = string("op_1438_cast_fp16")]; + tensor var_1439_cast_fp16 = softmax(axis = var_203, x = aw_chunk_123_cast_fp16)[name = string("op_1439_cast_fp16")]; + tensor var_1440_cast_fp16 = softmax(axis = var_203, x = aw_chunk_125_cast_fp16)[name = string("op_1440_cast_fp16")]; + tensor var_1441_cast_fp16 = softmax(axis = var_203, x = aw_chunk_127_cast_fp16)[name = string("op_1441_cast_fp16")]; + tensor var_1442_cast_fp16 = softmax(axis = var_203, x = aw_chunk_129_cast_fp16)[name = string("op_1442_cast_fp16")]; + tensor var_1443_cast_fp16 = softmax(axis = var_203, x = aw_chunk_131_cast_fp16)[name = string("op_1443_cast_fp16")]; + tensor var_1444_cast_fp16 = softmax(axis = var_203, x = aw_chunk_133_cast_fp16)[name = string("op_1444_cast_fp16")]; + tensor var_1445_cast_fp16 = softmax(axis = var_203, x = aw_chunk_135_cast_fp16)[name = string("op_1445_cast_fp16")]; + tensor var_1446_cast_fp16 = softmax(axis = var_203, x = aw_chunk_137_cast_fp16)[name = string("op_1446_cast_fp16")]; + tensor var_1447_cast_fp16 = softmax(axis = var_203, x = aw_chunk_139_cast_fp16)[name = string("op_1447_cast_fp16")]; + tensor var_1448_cast_fp16 = softmax(axis = var_203, x = aw_chunk_141_cast_fp16)[name = string("op_1448_cast_fp16")]; + tensor var_1449_cast_fp16 = softmax(axis = var_203, x = aw_chunk_143_cast_fp16)[name = string("op_1449_cast_fp16")]; + tensor var_1450_cast_fp16 = softmax(axis = var_203, x = aw_chunk_145_cast_fp16)[name = string("op_1450_cast_fp16")]; + tensor var_1451_cast_fp16 = softmax(axis = var_203, x = aw_chunk_147_cast_fp16)[name = string("op_1451_cast_fp16")]; + tensor var_1452_cast_fp16 = softmax(axis = var_203, x = aw_chunk_149_cast_fp16)[name = string("op_1452_cast_fp16")]; + tensor var_1453_cast_fp16 = softmax(axis = var_203, x = aw_chunk_151_cast_fp16)[name = string("op_1453_cast_fp16")]; + tensor var_1454_cast_fp16 = softmax(axis = var_203, x = aw_chunk_153_cast_fp16)[name = string("op_1454_cast_fp16")]; + tensor var_1455_cast_fp16 = softmax(axis = var_203, x = aw_chunk_155_cast_fp16)[name = string("op_1455_cast_fp16")]; + tensor var_1456_cast_fp16 = softmax(axis = var_203, x = aw_chunk_157_cast_fp16)[name = string("op_1456_cast_fp16")]; + tensor var_1457_cast_fp16 = softmax(axis = var_203, x = aw_chunk_159_cast_fp16)[name = string("op_1457_cast_fp16")]; + string var_1459_equation_0 = const()[name = string("op_1459_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1459_cast_fp16 = einsum(equation = var_1459_equation_0, values = (var_979_cast_fp16, var_1378_cast_fp16))[name = string("op_1459_cast_fp16")]; + string var_1461_equation_0 = const()[name = string("op_1461_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1461_cast_fp16 = einsum(equation = var_1461_equation_0, values = (var_979_cast_fp16, var_1379_cast_fp16))[name = string("op_1461_cast_fp16")]; + string var_1463_equation_0 = const()[name = string("op_1463_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1463_cast_fp16 = einsum(equation = var_1463_equation_0, values = (var_979_cast_fp16, var_1380_cast_fp16))[name = string("op_1463_cast_fp16")]; + string var_1465_equation_0 = const()[name = string("op_1465_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1465_cast_fp16 = einsum(equation = var_1465_equation_0, values = (var_979_cast_fp16, var_1381_cast_fp16))[name = string("op_1465_cast_fp16")]; + string var_1467_equation_0 = const()[name = string("op_1467_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1467_cast_fp16 = einsum(equation = var_1467_equation_0, values = (var_983_cast_fp16, var_1382_cast_fp16))[name = string("op_1467_cast_fp16")]; + string var_1469_equation_0 = const()[name = string("op_1469_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1469_cast_fp16 = einsum(equation = var_1469_equation_0, values = (var_983_cast_fp16, var_1383_cast_fp16))[name = string("op_1469_cast_fp16")]; + string var_1471_equation_0 = const()[name = string("op_1471_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1471_cast_fp16 = einsum(equation = var_1471_equation_0, values = (var_983_cast_fp16, var_1384_cast_fp16))[name = string("op_1471_cast_fp16")]; + string var_1473_equation_0 = const()[name = string("op_1473_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1473_cast_fp16 = einsum(equation = var_1473_equation_0, values = (var_983_cast_fp16, var_1385_cast_fp16))[name = string("op_1473_cast_fp16")]; + string var_1475_equation_0 = const()[name = string("op_1475_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1475_cast_fp16 = einsum(equation = var_1475_equation_0, values = (var_987_cast_fp16, var_1386_cast_fp16))[name = string("op_1475_cast_fp16")]; + string var_1477_equation_0 = const()[name = string("op_1477_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1477_cast_fp16 = einsum(equation = var_1477_equation_0, values = (var_987_cast_fp16, var_1387_cast_fp16))[name = string("op_1477_cast_fp16")]; + string var_1479_equation_0 = const()[name = string("op_1479_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1479_cast_fp16 = einsum(equation = var_1479_equation_0, values = (var_987_cast_fp16, var_1388_cast_fp16))[name = string("op_1479_cast_fp16")]; + string var_1481_equation_0 = const()[name = string("op_1481_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1481_cast_fp16 = einsum(equation = var_1481_equation_0, values = (var_987_cast_fp16, var_1389_cast_fp16))[name = string("op_1481_cast_fp16")]; + string var_1483_equation_0 = const()[name = string("op_1483_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1483_cast_fp16 = einsum(equation = var_1483_equation_0, values = (var_991_cast_fp16, var_1390_cast_fp16))[name = string("op_1483_cast_fp16")]; + string var_1485_equation_0 = const()[name = string("op_1485_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1485_cast_fp16 = einsum(equation = var_1485_equation_0, values = (var_991_cast_fp16, var_1391_cast_fp16))[name = string("op_1485_cast_fp16")]; + string var_1487_equation_0 = const()[name = string("op_1487_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1487_cast_fp16 = einsum(equation = var_1487_equation_0, values = (var_991_cast_fp16, var_1392_cast_fp16))[name = string("op_1487_cast_fp16")]; + string var_1489_equation_0 = const()[name = string("op_1489_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1489_cast_fp16 = einsum(equation = var_1489_equation_0, values = (var_991_cast_fp16, var_1393_cast_fp16))[name = string("op_1489_cast_fp16")]; + string var_1491_equation_0 = const()[name = string("op_1491_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1491_cast_fp16 = einsum(equation = var_1491_equation_0, values = (var_995_cast_fp16, var_1394_cast_fp16))[name = string("op_1491_cast_fp16")]; + string var_1493_equation_0 = const()[name = string("op_1493_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1493_cast_fp16 = einsum(equation = var_1493_equation_0, values = (var_995_cast_fp16, var_1395_cast_fp16))[name = string("op_1493_cast_fp16")]; + string var_1495_equation_0 = const()[name = string("op_1495_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1495_cast_fp16 = einsum(equation = var_1495_equation_0, values = (var_995_cast_fp16, var_1396_cast_fp16))[name = string("op_1495_cast_fp16")]; + string var_1497_equation_0 = const()[name = string("op_1497_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1497_cast_fp16 = einsum(equation = var_1497_equation_0, values = (var_995_cast_fp16, var_1397_cast_fp16))[name = string("op_1497_cast_fp16")]; + string var_1499_equation_0 = const()[name = string("op_1499_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1499_cast_fp16 = einsum(equation = var_1499_equation_0, values = (var_999_cast_fp16, var_1398_cast_fp16))[name = string("op_1499_cast_fp16")]; + string var_1501_equation_0 = const()[name = string("op_1501_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1501_cast_fp16 = einsum(equation = var_1501_equation_0, values = (var_999_cast_fp16, var_1399_cast_fp16))[name = string("op_1501_cast_fp16")]; + string var_1503_equation_0 = const()[name = string("op_1503_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1503_cast_fp16 = einsum(equation = var_1503_equation_0, values = (var_999_cast_fp16, var_1400_cast_fp16))[name = string("op_1503_cast_fp16")]; + string var_1505_equation_0 = const()[name = string("op_1505_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1505_cast_fp16 = einsum(equation = var_1505_equation_0, values = (var_999_cast_fp16, var_1401_cast_fp16))[name = string("op_1505_cast_fp16")]; + string var_1507_equation_0 = const()[name = string("op_1507_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1507_cast_fp16 = einsum(equation = var_1507_equation_0, values = (var_1003_cast_fp16, var_1402_cast_fp16))[name = string("op_1507_cast_fp16")]; + string var_1509_equation_0 = const()[name = string("op_1509_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1509_cast_fp16 = einsum(equation = var_1509_equation_0, values = (var_1003_cast_fp16, var_1403_cast_fp16))[name = string("op_1509_cast_fp16")]; + string var_1511_equation_0 = const()[name = string("op_1511_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1511_cast_fp16 = einsum(equation = var_1511_equation_0, values = (var_1003_cast_fp16, var_1404_cast_fp16))[name = string("op_1511_cast_fp16")]; + string var_1513_equation_0 = const()[name = string("op_1513_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1513_cast_fp16 = einsum(equation = var_1513_equation_0, values = (var_1003_cast_fp16, var_1405_cast_fp16))[name = string("op_1513_cast_fp16")]; + string var_1515_equation_0 = const()[name = string("op_1515_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1515_cast_fp16 = einsum(equation = var_1515_equation_0, values = (var_1007_cast_fp16, var_1406_cast_fp16))[name = string("op_1515_cast_fp16")]; + string var_1517_equation_0 = const()[name = string("op_1517_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1517_cast_fp16 = einsum(equation = var_1517_equation_0, values = (var_1007_cast_fp16, var_1407_cast_fp16))[name = string("op_1517_cast_fp16")]; + string var_1519_equation_0 = const()[name = string("op_1519_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1519_cast_fp16 = einsum(equation = var_1519_equation_0, values = (var_1007_cast_fp16, var_1408_cast_fp16))[name = string("op_1519_cast_fp16")]; + string var_1521_equation_0 = const()[name = string("op_1521_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1521_cast_fp16 = einsum(equation = var_1521_equation_0, values = (var_1007_cast_fp16, var_1409_cast_fp16))[name = string("op_1521_cast_fp16")]; + string var_1523_equation_0 = const()[name = string("op_1523_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1523_cast_fp16 = einsum(equation = var_1523_equation_0, values = (var_1011_cast_fp16, var_1410_cast_fp16))[name = string("op_1523_cast_fp16")]; + string var_1525_equation_0 = const()[name = string("op_1525_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1525_cast_fp16 = einsum(equation = var_1525_equation_0, values = (var_1011_cast_fp16, var_1411_cast_fp16))[name = string("op_1525_cast_fp16")]; + string var_1527_equation_0 = const()[name = string("op_1527_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1527_cast_fp16 = einsum(equation = var_1527_equation_0, values = (var_1011_cast_fp16, var_1412_cast_fp16))[name = string("op_1527_cast_fp16")]; + string var_1529_equation_0 = const()[name = string("op_1529_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1529_cast_fp16 = einsum(equation = var_1529_equation_0, values = (var_1011_cast_fp16, var_1413_cast_fp16))[name = string("op_1529_cast_fp16")]; + string var_1531_equation_0 = const()[name = string("op_1531_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1531_cast_fp16 = einsum(equation = var_1531_equation_0, values = (var_1015_cast_fp16, var_1414_cast_fp16))[name = string("op_1531_cast_fp16")]; + string var_1533_equation_0 = const()[name = string("op_1533_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1533_cast_fp16 = einsum(equation = var_1533_equation_0, values = (var_1015_cast_fp16, var_1415_cast_fp16))[name = string("op_1533_cast_fp16")]; + string var_1535_equation_0 = const()[name = string("op_1535_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1535_cast_fp16 = einsum(equation = var_1535_equation_0, values = (var_1015_cast_fp16, var_1416_cast_fp16))[name = string("op_1535_cast_fp16")]; + string var_1537_equation_0 = const()[name = string("op_1537_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1537_cast_fp16 = einsum(equation = var_1537_equation_0, values = (var_1015_cast_fp16, var_1417_cast_fp16))[name = string("op_1537_cast_fp16")]; + string var_1539_equation_0 = const()[name = string("op_1539_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1539_cast_fp16 = einsum(equation = var_1539_equation_0, values = (var_1019_cast_fp16, var_1418_cast_fp16))[name = string("op_1539_cast_fp16")]; + string var_1541_equation_0 = const()[name = string("op_1541_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1541_cast_fp16 = einsum(equation = var_1541_equation_0, values = (var_1019_cast_fp16, var_1419_cast_fp16))[name = string("op_1541_cast_fp16")]; + string var_1543_equation_0 = const()[name = string("op_1543_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1543_cast_fp16 = einsum(equation = var_1543_equation_0, values = (var_1019_cast_fp16, var_1420_cast_fp16))[name = string("op_1543_cast_fp16")]; + string var_1545_equation_0 = const()[name = string("op_1545_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1545_cast_fp16 = einsum(equation = var_1545_equation_0, values = (var_1019_cast_fp16, var_1421_cast_fp16))[name = string("op_1545_cast_fp16")]; + string var_1547_equation_0 = const()[name = string("op_1547_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1547_cast_fp16 = einsum(equation = var_1547_equation_0, values = (var_1023_cast_fp16, var_1422_cast_fp16))[name = string("op_1547_cast_fp16")]; + string var_1549_equation_0 = const()[name = string("op_1549_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1549_cast_fp16 = einsum(equation = var_1549_equation_0, values = (var_1023_cast_fp16, var_1423_cast_fp16))[name = string("op_1549_cast_fp16")]; + string var_1551_equation_0 = const()[name = string("op_1551_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1551_cast_fp16 = einsum(equation = var_1551_equation_0, values = (var_1023_cast_fp16, var_1424_cast_fp16))[name = string("op_1551_cast_fp16")]; + string var_1553_equation_0 = const()[name = string("op_1553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1553_cast_fp16 = einsum(equation = var_1553_equation_0, values = (var_1023_cast_fp16, var_1425_cast_fp16))[name = string("op_1553_cast_fp16")]; + string var_1555_equation_0 = const()[name = string("op_1555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1555_cast_fp16 = einsum(equation = var_1555_equation_0, values = (var_1027_cast_fp16, var_1426_cast_fp16))[name = string("op_1555_cast_fp16")]; + string var_1557_equation_0 = const()[name = string("op_1557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1557_cast_fp16 = einsum(equation = var_1557_equation_0, values = (var_1027_cast_fp16, var_1427_cast_fp16))[name = string("op_1557_cast_fp16")]; + string var_1559_equation_0 = const()[name = string("op_1559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1559_cast_fp16 = einsum(equation = var_1559_equation_0, values = (var_1027_cast_fp16, var_1428_cast_fp16))[name = string("op_1559_cast_fp16")]; + string var_1561_equation_0 = const()[name = string("op_1561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1561_cast_fp16 = einsum(equation = var_1561_equation_0, values = (var_1027_cast_fp16, var_1429_cast_fp16))[name = string("op_1561_cast_fp16")]; + string var_1563_equation_0 = const()[name = string("op_1563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1563_cast_fp16 = einsum(equation = var_1563_equation_0, values = (var_1031_cast_fp16, var_1430_cast_fp16))[name = string("op_1563_cast_fp16")]; + string var_1565_equation_0 = const()[name = string("op_1565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1565_cast_fp16 = einsum(equation = var_1565_equation_0, values = (var_1031_cast_fp16, var_1431_cast_fp16))[name = string("op_1565_cast_fp16")]; + string var_1567_equation_0 = const()[name = string("op_1567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1567_cast_fp16 = einsum(equation = var_1567_equation_0, values = (var_1031_cast_fp16, var_1432_cast_fp16))[name = string("op_1567_cast_fp16")]; + string var_1569_equation_0 = const()[name = string("op_1569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1569_cast_fp16 = einsum(equation = var_1569_equation_0, values = (var_1031_cast_fp16, var_1433_cast_fp16))[name = string("op_1569_cast_fp16")]; + string var_1571_equation_0 = const()[name = string("op_1571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1571_cast_fp16 = einsum(equation = var_1571_equation_0, values = (var_1035_cast_fp16, var_1434_cast_fp16))[name = string("op_1571_cast_fp16")]; + string var_1573_equation_0 = const()[name = string("op_1573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1573_cast_fp16 = einsum(equation = var_1573_equation_0, values = (var_1035_cast_fp16, var_1435_cast_fp16))[name = string("op_1573_cast_fp16")]; + string var_1575_equation_0 = const()[name = string("op_1575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1575_cast_fp16 = einsum(equation = var_1575_equation_0, values = (var_1035_cast_fp16, var_1436_cast_fp16))[name = string("op_1575_cast_fp16")]; + string var_1577_equation_0 = const()[name = string("op_1577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1577_cast_fp16 = einsum(equation = var_1577_equation_0, values = (var_1035_cast_fp16, var_1437_cast_fp16))[name = string("op_1577_cast_fp16")]; + string var_1579_equation_0 = const()[name = string("op_1579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1579_cast_fp16 = einsum(equation = var_1579_equation_0, values = (var_1039_cast_fp16, var_1438_cast_fp16))[name = string("op_1579_cast_fp16")]; + string var_1581_equation_0 = const()[name = string("op_1581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1581_cast_fp16 = einsum(equation = var_1581_equation_0, values = (var_1039_cast_fp16, var_1439_cast_fp16))[name = string("op_1581_cast_fp16")]; + string var_1583_equation_0 = const()[name = string("op_1583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1583_cast_fp16 = einsum(equation = var_1583_equation_0, values = (var_1039_cast_fp16, var_1440_cast_fp16))[name = string("op_1583_cast_fp16")]; + string var_1585_equation_0 = const()[name = string("op_1585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1585_cast_fp16 = einsum(equation = var_1585_equation_0, values = (var_1039_cast_fp16, var_1441_cast_fp16))[name = string("op_1585_cast_fp16")]; + string var_1587_equation_0 = const()[name = string("op_1587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1587_cast_fp16 = einsum(equation = var_1587_equation_0, values = (var_1043_cast_fp16, var_1442_cast_fp16))[name = string("op_1587_cast_fp16")]; + string var_1589_equation_0 = const()[name = string("op_1589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1589_cast_fp16 = einsum(equation = var_1589_equation_0, values = (var_1043_cast_fp16, var_1443_cast_fp16))[name = string("op_1589_cast_fp16")]; + string var_1591_equation_0 = const()[name = string("op_1591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1591_cast_fp16 = einsum(equation = var_1591_equation_0, values = (var_1043_cast_fp16, var_1444_cast_fp16))[name = string("op_1591_cast_fp16")]; + string var_1593_equation_0 = const()[name = string("op_1593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1593_cast_fp16 = einsum(equation = var_1593_equation_0, values = (var_1043_cast_fp16, var_1445_cast_fp16))[name = string("op_1593_cast_fp16")]; + string var_1595_equation_0 = const()[name = string("op_1595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1595_cast_fp16 = einsum(equation = var_1595_equation_0, values = (var_1047_cast_fp16, var_1446_cast_fp16))[name = string("op_1595_cast_fp16")]; + string var_1597_equation_0 = const()[name = string("op_1597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1597_cast_fp16 = einsum(equation = var_1597_equation_0, values = (var_1047_cast_fp16, var_1447_cast_fp16))[name = string("op_1597_cast_fp16")]; + string var_1599_equation_0 = const()[name = string("op_1599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1599_cast_fp16 = einsum(equation = var_1599_equation_0, values = (var_1047_cast_fp16, var_1448_cast_fp16))[name = string("op_1599_cast_fp16")]; + string var_1601_equation_0 = const()[name = string("op_1601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1601_cast_fp16 = einsum(equation = var_1601_equation_0, values = (var_1047_cast_fp16, var_1449_cast_fp16))[name = string("op_1601_cast_fp16")]; + string var_1603_equation_0 = const()[name = string("op_1603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1603_cast_fp16 = einsum(equation = var_1603_equation_0, values = (var_1051_cast_fp16, var_1450_cast_fp16))[name = string("op_1603_cast_fp16")]; + string var_1605_equation_0 = const()[name = string("op_1605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1605_cast_fp16 = einsum(equation = var_1605_equation_0, values = (var_1051_cast_fp16, var_1451_cast_fp16))[name = string("op_1605_cast_fp16")]; + string var_1607_equation_0 = const()[name = string("op_1607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1607_cast_fp16 = einsum(equation = var_1607_equation_0, values = (var_1051_cast_fp16, var_1452_cast_fp16))[name = string("op_1607_cast_fp16")]; + string var_1609_equation_0 = const()[name = string("op_1609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1609_cast_fp16 = einsum(equation = var_1609_equation_0, values = (var_1051_cast_fp16, var_1453_cast_fp16))[name = string("op_1609_cast_fp16")]; + string var_1611_equation_0 = const()[name = string("op_1611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1611_cast_fp16 = einsum(equation = var_1611_equation_0, values = (var_1055_cast_fp16, var_1454_cast_fp16))[name = string("op_1611_cast_fp16")]; + string var_1613_equation_0 = const()[name = string("op_1613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1613_cast_fp16 = einsum(equation = var_1613_equation_0, values = (var_1055_cast_fp16, var_1455_cast_fp16))[name = string("op_1613_cast_fp16")]; + string var_1615_equation_0 = const()[name = string("op_1615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1615_cast_fp16 = einsum(equation = var_1615_equation_0, values = (var_1055_cast_fp16, var_1456_cast_fp16))[name = string("op_1615_cast_fp16")]; + string var_1617_equation_0 = const()[name = string("op_1617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1617_cast_fp16 = einsum(equation = var_1617_equation_0, values = (var_1055_cast_fp16, var_1457_cast_fp16))[name = string("op_1617_cast_fp16")]; + bool var_1619_interleave_0 = const()[name = string("op_1619_interleave_0"), val = bool(false)]; + tensor var_1619_cast_fp16 = concat(axis = var_178, interleave = var_1619_interleave_0, values = (var_1459_cast_fp16, var_1461_cast_fp16, var_1463_cast_fp16, var_1465_cast_fp16))[name = string("op_1619_cast_fp16")]; + bool var_1621_interleave_0 = const()[name = string("op_1621_interleave_0"), val = bool(false)]; + tensor var_1621_cast_fp16 = concat(axis = var_178, interleave = var_1621_interleave_0, values = (var_1467_cast_fp16, var_1469_cast_fp16, var_1471_cast_fp16, var_1473_cast_fp16))[name = string("op_1621_cast_fp16")]; + bool var_1623_interleave_0 = const()[name = string("op_1623_interleave_0"), val = bool(false)]; + tensor var_1623_cast_fp16 = concat(axis = var_178, interleave = var_1623_interleave_0, values = (var_1475_cast_fp16, var_1477_cast_fp16, var_1479_cast_fp16, var_1481_cast_fp16))[name = string("op_1623_cast_fp16")]; + bool var_1625_interleave_0 = const()[name = string("op_1625_interleave_0"), val = bool(false)]; + tensor var_1625_cast_fp16 = concat(axis = var_178, interleave = var_1625_interleave_0, values = (var_1483_cast_fp16, var_1485_cast_fp16, var_1487_cast_fp16, var_1489_cast_fp16))[name = string("op_1625_cast_fp16")]; + bool var_1627_interleave_0 = const()[name = string("op_1627_interleave_0"), val = bool(false)]; + tensor var_1627_cast_fp16 = concat(axis = var_178, interleave = var_1627_interleave_0, values = (var_1491_cast_fp16, var_1493_cast_fp16, var_1495_cast_fp16, var_1497_cast_fp16))[name = string("op_1627_cast_fp16")]; + bool var_1629_interleave_0 = const()[name = string("op_1629_interleave_0"), val = bool(false)]; + tensor var_1629_cast_fp16 = concat(axis = var_178, interleave = var_1629_interleave_0, values = (var_1499_cast_fp16, var_1501_cast_fp16, var_1503_cast_fp16, var_1505_cast_fp16))[name = string("op_1629_cast_fp16")]; + bool var_1631_interleave_0 = const()[name = string("op_1631_interleave_0"), val = bool(false)]; + tensor var_1631_cast_fp16 = concat(axis = var_178, interleave = var_1631_interleave_0, values = (var_1507_cast_fp16, var_1509_cast_fp16, var_1511_cast_fp16, var_1513_cast_fp16))[name = string("op_1631_cast_fp16")]; + bool var_1633_interleave_0 = const()[name = string("op_1633_interleave_0"), val = bool(false)]; + tensor var_1633_cast_fp16 = concat(axis = var_178, interleave = var_1633_interleave_0, values = (var_1515_cast_fp16, var_1517_cast_fp16, var_1519_cast_fp16, var_1521_cast_fp16))[name = string("op_1633_cast_fp16")]; + bool var_1635_interleave_0 = const()[name = string("op_1635_interleave_0"), val = bool(false)]; + tensor var_1635_cast_fp16 = concat(axis = var_178, interleave = var_1635_interleave_0, values = (var_1523_cast_fp16, var_1525_cast_fp16, var_1527_cast_fp16, var_1529_cast_fp16))[name = string("op_1635_cast_fp16")]; + bool var_1637_interleave_0 = const()[name = string("op_1637_interleave_0"), val = bool(false)]; + tensor var_1637_cast_fp16 = concat(axis = var_178, interleave = var_1637_interleave_0, values = (var_1531_cast_fp16, var_1533_cast_fp16, var_1535_cast_fp16, var_1537_cast_fp16))[name = string("op_1637_cast_fp16")]; + bool var_1639_interleave_0 = const()[name = string("op_1639_interleave_0"), val = bool(false)]; + tensor var_1639_cast_fp16 = concat(axis = var_178, interleave = var_1639_interleave_0, values = (var_1539_cast_fp16, var_1541_cast_fp16, var_1543_cast_fp16, var_1545_cast_fp16))[name = string("op_1639_cast_fp16")]; + bool var_1641_interleave_0 = const()[name = string("op_1641_interleave_0"), val = bool(false)]; + tensor var_1641_cast_fp16 = concat(axis = var_178, interleave = var_1641_interleave_0, values = (var_1547_cast_fp16, var_1549_cast_fp16, var_1551_cast_fp16, var_1553_cast_fp16))[name = string("op_1641_cast_fp16")]; + bool var_1643_interleave_0 = const()[name = string("op_1643_interleave_0"), val = bool(false)]; + tensor var_1643_cast_fp16 = concat(axis = var_178, interleave = var_1643_interleave_0, values = (var_1555_cast_fp16, var_1557_cast_fp16, var_1559_cast_fp16, var_1561_cast_fp16))[name = string("op_1643_cast_fp16")]; + bool var_1645_interleave_0 = const()[name = string("op_1645_interleave_0"), val = bool(false)]; + tensor var_1645_cast_fp16 = concat(axis = var_178, interleave = var_1645_interleave_0, values = (var_1563_cast_fp16, var_1565_cast_fp16, var_1567_cast_fp16, var_1569_cast_fp16))[name = string("op_1645_cast_fp16")]; + bool var_1647_interleave_0 = const()[name = string("op_1647_interleave_0"), val = bool(false)]; + tensor var_1647_cast_fp16 = concat(axis = var_178, interleave = var_1647_interleave_0, values = (var_1571_cast_fp16, var_1573_cast_fp16, var_1575_cast_fp16, var_1577_cast_fp16))[name = string("op_1647_cast_fp16")]; + bool var_1649_interleave_0 = const()[name = string("op_1649_interleave_0"), val = bool(false)]; + tensor var_1649_cast_fp16 = concat(axis = var_178, interleave = var_1649_interleave_0, values = (var_1579_cast_fp16, var_1581_cast_fp16, var_1583_cast_fp16, var_1585_cast_fp16))[name = string("op_1649_cast_fp16")]; + bool var_1651_interleave_0 = const()[name = string("op_1651_interleave_0"), val = bool(false)]; + tensor var_1651_cast_fp16 = concat(axis = var_178, interleave = var_1651_interleave_0, values = (var_1587_cast_fp16, var_1589_cast_fp16, var_1591_cast_fp16, var_1593_cast_fp16))[name = string("op_1651_cast_fp16")]; + bool var_1653_interleave_0 = const()[name = string("op_1653_interleave_0"), val = bool(false)]; + tensor var_1653_cast_fp16 = concat(axis = var_178, interleave = var_1653_interleave_0, values = (var_1595_cast_fp16, var_1597_cast_fp16, var_1599_cast_fp16, var_1601_cast_fp16))[name = string("op_1653_cast_fp16")]; + bool var_1655_interleave_0 = const()[name = string("op_1655_interleave_0"), val = bool(false)]; + tensor var_1655_cast_fp16 = concat(axis = var_178, interleave = var_1655_interleave_0, values = (var_1603_cast_fp16, var_1605_cast_fp16, var_1607_cast_fp16, var_1609_cast_fp16))[name = string("op_1655_cast_fp16")]; + bool var_1657_interleave_0 = const()[name = string("op_1657_interleave_0"), val = bool(false)]; + tensor var_1657_cast_fp16 = concat(axis = var_178, interleave = var_1657_interleave_0, values = (var_1611_cast_fp16, var_1613_cast_fp16, var_1615_cast_fp16, var_1617_cast_fp16))[name = string("op_1657_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_203, interleave = input_1_interleave_0, values = (var_1619_cast_fp16, var_1621_cast_fp16, var_1623_cast_fp16, var_1625_cast_fp16, var_1627_cast_fp16, var_1629_cast_fp16, var_1631_cast_fp16, var_1633_cast_fp16, var_1635_cast_fp16, var_1637_cast_fp16, var_1639_cast_fp16, var_1641_cast_fp16, var_1643_cast_fp16, var_1645_cast_fp16, var_1647_cast_fp16, var_1649_cast_fp16, var_1651_cast_fp16, var_1653_cast_fp16, var_1655_cast_fp16, var_1657_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24505280)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27782144)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_1676_to_fp16 = const()[name = string("op_1676_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_1676_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27784768)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27787392)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27790016)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40897280)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40907584)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54014848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_1705 = const()[name = string("op_1705"), val = int32(3)]; + int32 var_1730 = const()[name = string("op_1730"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_1747_to_fp16 = const()[name = string("op_1747_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_1747_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54017472)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54020096)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54022720)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57299584)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57302208)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60579072)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63855936)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_1785_begin_0 = const()[name = string("op_1785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1785_end_0 = const()[name = string("op_1785_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1785_end_mask_0 = const()[name = string("op_1785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1785_cast_fp16 = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = query_3_cast_fp16)[name = string("op_1785_cast_fp16")]; + tensor var_1789_begin_0 = const()[name = string("op_1789_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1789_end_0 = const()[name = string("op_1789_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1789_end_mask_0 = const()[name = string("op_1789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1789_cast_fp16 = slice_by_index(begin = var_1789_begin_0, end = var_1789_end_0, end_mask = var_1789_end_mask_0, x = query_3_cast_fp16)[name = string("op_1789_cast_fp16")]; + tensor var_1793_begin_0 = const()[name = string("op_1793_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1793_end_0 = const()[name = string("op_1793_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1793_end_mask_0 = const()[name = string("op_1793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1793_cast_fp16 = slice_by_index(begin = var_1793_begin_0, end = var_1793_end_0, end_mask = var_1793_end_mask_0, x = query_3_cast_fp16)[name = string("op_1793_cast_fp16")]; + tensor var_1797_begin_0 = const()[name = string("op_1797_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1797_end_0 = const()[name = string("op_1797_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1797_end_mask_0 = const()[name = string("op_1797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1797_cast_fp16 = slice_by_index(begin = var_1797_begin_0, end = var_1797_end_0, end_mask = var_1797_end_mask_0, x = query_3_cast_fp16)[name = string("op_1797_cast_fp16")]; + tensor var_1801_begin_0 = const()[name = string("op_1801_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1801_end_0 = const()[name = string("op_1801_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1801_end_mask_0 = const()[name = string("op_1801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1801_cast_fp16 = slice_by_index(begin = var_1801_begin_0, end = var_1801_end_0, end_mask = var_1801_end_mask_0, x = query_3_cast_fp16)[name = string("op_1801_cast_fp16")]; + tensor var_1805_begin_0 = const()[name = string("op_1805_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1805_end_0 = const()[name = string("op_1805_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1805_end_mask_0 = const()[name = string("op_1805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1805_cast_fp16 = slice_by_index(begin = var_1805_begin_0, end = var_1805_end_0, end_mask = var_1805_end_mask_0, x = query_3_cast_fp16)[name = string("op_1805_cast_fp16")]; + tensor var_1809_begin_0 = const()[name = string("op_1809_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1809_end_0 = const()[name = string("op_1809_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1809_end_mask_0 = const()[name = string("op_1809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1809_cast_fp16 = slice_by_index(begin = var_1809_begin_0, end = var_1809_end_0, end_mask = var_1809_end_mask_0, x = query_3_cast_fp16)[name = string("op_1809_cast_fp16")]; + tensor var_1813_begin_0 = const()[name = string("op_1813_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1813_end_0 = const()[name = string("op_1813_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1813_end_mask_0 = const()[name = string("op_1813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1813_cast_fp16 = slice_by_index(begin = var_1813_begin_0, end = var_1813_end_0, end_mask = var_1813_end_mask_0, x = query_3_cast_fp16)[name = string("op_1813_cast_fp16")]; + tensor var_1817_begin_0 = const()[name = string("op_1817_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1817_end_0 = const()[name = string("op_1817_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1817_end_mask_0 = const()[name = string("op_1817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1817_cast_fp16 = slice_by_index(begin = var_1817_begin_0, end = var_1817_end_0, end_mask = var_1817_end_mask_0, x = query_3_cast_fp16)[name = string("op_1817_cast_fp16")]; + tensor var_1821_begin_0 = const()[name = string("op_1821_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1821_end_0 = const()[name = string("op_1821_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1821_end_mask_0 = const()[name = string("op_1821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1821_cast_fp16 = slice_by_index(begin = var_1821_begin_0, end = var_1821_end_0, end_mask = var_1821_end_mask_0, x = query_3_cast_fp16)[name = string("op_1821_cast_fp16")]; + tensor var_1825_begin_0 = const()[name = string("op_1825_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1825_end_0 = const()[name = string("op_1825_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1825_end_mask_0 = const()[name = string("op_1825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1825_cast_fp16 = slice_by_index(begin = var_1825_begin_0, end = var_1825_end_0, end_mask = var_1825_end_mask_0, x = query_3_cast_fp16)[name = string("op_1825_cast_fp16")]; + tensor var_1829_begin_0 = const()[name = string("op_1829_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1829_end_0 = const()[name = string("op_1829_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1829_end_mask_0 = const()[name = string("op_1829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1829_cast_fp16 = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = query_3_cast_fp16)[name = string("op_1829_cast_fp16")]; + tensor var_1833_begin_0 = const()[name = string("op_1833_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_1833_end_0 = const()[name = string("op_1833_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_1833_end_mask_0 = const()[name = string("op_1833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1833_cast_fp16 = slice_by_index(begin = var_1833_begin_0, end = var_1833_end_0, end_mask = var_1833_end_mask_0, x = query_3_cast_fp16)[name = string("op_1833_cast_fp16")]; + tensor var_1837_begin_0 = const()[name = string("op_1837_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_1837_end_0 = const()[name = string("op_1837_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_1837_end_mask_0 = const()[name = string("op_1837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1837_cast_fp16 = slice_by_index(begin = var_1837_begin_0, end = var_1837_end_0, end_mask = var_1837_end_mask_0, x = query_3_cast_fp16)[name = string("op_1837_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = query_3_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1845_begin_0 = const()[name = string("op_1845_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_1845_end_0 = const()[name = string("op_1845_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_1845_end_mask_0 = const()[name = string("op_1845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1845_cast_fp16 = slice_by_index(begin = var_1845_begin_0, end = var_1845_end_0, end_mask = var_1845_end_mask_0, x = query_3_cast_fp16)[name = string("op_1845_cast_fp16")]; + tensor var_1849_begin_0 = const()[name = string("op_1849_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_1849_end_0 = const()[name = string("op_1849_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_1849_end_mask_0 = const()[name = string("op_1849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1849_cast_fp16 = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, x = query_3_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor var_1853_begin_0 = const()[name = string("op_1853_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_1853_end_0 = const()[name = string("op_1853_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_1853_end_mask_0 = const()[name = string("op_1853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1853_cast_fp16 = slice_by_index(begin = var_1853_begin_0, end = var_1853_end_0, end_mask = var_1853_end_mask_0, x = query_3_cast_fp16)[name = string("op_1853_cast_fp16")]; + tensor var_1857_begin_0 = const()[name = string("op_1857_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_1857_end_0 = const()[name = string("op_1857_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_1857_end_mask_0 = const()[name = string("op_1857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1857_cast_fp16 = slice_by_index(begin = var_1857_begin_0, end = var_1857_end_0, end_mask = var_1857_end_mask_0, x = query_3_cast_fp16)[name = string("op_1857_cast_fp16")]; + tensor var_1861_begin_0 = const()[name = string("op_1861_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_1861_end_0 = const()[name = string("op_1861_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_1861_end_mask_0 = const()[name = string("op_1861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = query_3_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1870_begin_0 = const()[name = string("op_1870_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1870_end_0 = const()[name = string("op_1870_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1870_end_mask_0 = const()[name = string("op_1870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1870_cast_fp16 = slice_by_index(begin = var_1870_begin_0, end = var_1870_end_0, end_mask = var_1870_end_mask_0, x = var_1785_cast_fp16)[name = string("op_1870_cast_fp16")]; + tensor var_1877_begin_0 = const()[name = string("op_1877_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1877_end_0 = const()[name = string("op_1877_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1877_end_mask_0 = const()[name = string("op_1877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1877_cast_fp16 = slice_by_index(begin = var_1877_begin_0, end = var_1877_end_0, end_mask = var_1877_end_mask_0, x = var_1785_cast_fp16)[name = string("op_1877_cast_fp16")]; + tensor var_1884_begin_0 = const()[name = string("op_1884_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1884_end_0 = const()[name = string("op_1884_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1884_end_mask_0 = const()[name = string("op_1884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1884_cast_fp16 = slice_by_index(begin = var_1884_begin_0, end = var_1884_end_0, end_mask = var_1884_end_mask_0, x = var_1785_cast_fp16)[name = string("op_1884_cast_fp16")]; + tensor var_1891_begin_0 = const()[name = string("op_1891_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1891_end_0 = const()[name = string("op_1891_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1891_end_mask_0 = const()[name = string("op_1891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1891_cast_fp16 = slice_by_index(begin = var_1891_begin_0, end = var_1891_end_0, end_mask = var_1891_end_mask_0, x = var_1785_cast_fp16)[name = string("op_1891_cast_fp16")]; + tensor var_1898_begin_0 = const()[name = string("op_1898_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1898_end_0 = const()[name = string("op_1898_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1898_end_mask_0 = const()[name = string("op_1898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1898_cast_fp16 = slice_by_index(begin = var_1898_begin_0, end = var_1898_end_0, end_mask = var_1898_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1898_cast_fp16")]; + tensor var_1905_begin_0 = const()[name = string("op_1905_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1905_end_0 = const()[name = string("op_1905_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1905_end_mask_0 = const()[name = string("op_1905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1905_cast_fp16 = slice_by_index(begin = var_1905_begin_0, end = var_1905_end_0, end_mask = var_1905_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1905_cast_fp16")]; + tensor var_1912_begin_0 = const()[name = string("op_1912_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1912_end_0 = const()[name = string("op_1912_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1912_end_mask_0 = const()[name = string("op_1912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1912_cast_fp16 = slice_by_index(begin = var_1912_begin_0, end = var_1912_end_0, end_mask = var_1912_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1912_cast_fp16")]; + tensor var_1919_begin_0 = const()[name = string("op_1919_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1919_end_0 = const()[name = string("op_1919_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1919_end_mask_0 = const()[name = string("op_1919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1919_cast_fp16 = slice_by_index(begin = var_1919_begin_0, end = var_1919_end_0, end_mask = var_1919_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1919_cast_fp16")]; + tensor var_1926_begin_0 = const()[name = string("op_1926_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1926_end_0 = const()[name = string("op_1926_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1926_end_mask_0 = const()[name = string("op_1926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1926_cast_fp16 = slice_by_index(begin = var_1926_begin_0, end = var_1926_end_0, end_mask = var_1926_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1926_cast_fp16")]; + tensor var_1933_begin_0 = const()[name = string("op_1933_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1933_end_0 = const()[name = string("op_1933_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1933_end_mask_0 = const()[name = string("op_1933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1933_cast_fp16 = slice_by_index(begin = var_1933_begin_0, end = var_1933_end_0, end_mask = var_1933_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1933_cast_fp16")]; + tensor var_1940_begin_0 = const()[name = string("op_1940_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1940_end_0 = const()[name = string("op_1940_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1940_end_mask_0 = const()[name = string("op_1940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1940_cast_fp16 = slice_by_index(begin = var_1940_begin_0, end = var_1940_end_0, end_mask = var_1940_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1940_cast_fp16")]; + tensor var_1947_begin_0 = const()[name = string("op_1947_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1947_end_0 = const()[name = string("op_1947_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1947_end_mask_0 = const()[name = string("op_1947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1947_cast_fp16 = slice_by_index(begin = var_1947_begin_0, end = var_1947_end_0, end_mask = var_1947_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1947_cast_fp16")]; + tensor var_1954_begin_0 = const()[name = string("op_1954_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1954_end_0 = const()[name = string("op_1954_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1954_end_mask_0 = const()[name = string("op_1954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1954_cast_fp16 = slice_by_index(begin = var_1954_begin_0, end = var_1954_end_0, end_mask = var_1954_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1954_cast_fp16")]; + tensor var_1961_begin_0 = const()[name = string("op_1961_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1961_end_0 = const()[name = string("op_1961_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1961_end_mask_0 = const()[name = string("op_1961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1961_cast_fp16 = slice_by_index(begin = var_1961_begin_0, end = var_1961_end_0, end_mask = var_1961_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1961_cast_fp16")]; + tensor var_1968_begin_0 = const()[name = string("op_1968_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1968_end_0 = const()[name = string("op_1968_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1968_end_mask_0 = const()[name = string("op_1968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1968_cast_fp16 = slice_by_index(begin = var_1968_begin_0, end = var_1968_end_0, end_mask = var_1968_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1968_cast_fp16")]; + tensor var_1975_begin_0 = const()[name = string("op_1975_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1975_end_0 = const()[name = string("op_1975_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1975_end_mask_0 = const()[name = string("op_1975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1975_cast_fp16 = slice_by_index(begin = var_1975_begin_0, end = var_1975_end_0, end_mask = var_1975_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1975_cast_fp16")]; + tensor var_1982_begin_0 = const()[name = string("op_1982_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1982_end_0 = const()[name = string("op_1982_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1982_end_mask_0 = const()[name = string("op_1982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1982_cast_fp16 = slice_by_index(begin = var_1982_begin_0, end = var_1982_end_0, end_mask = var_1982_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1982_cast_fp16")]; + tensor var_1989_begin_0 = const()[name = string("op_1989_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1989_end_0 = const()[name = string("op_1989_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1989_end_mask_0 = const()[name = string("op_1989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1989_cast_fp16 = slice_by_index(begin = var_1989_begin_0, end = var_1989_end_0, end_mask = var_1989_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1989_cast_fp16")]; + tensor var_1996_begin_0 = const()[name = string("op_1996_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1996_end_0 = const()[name = string("op_1996_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1996_end_mask_0 = const()[name = string("op_1996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1996_cast_fp16 = slice_by_index(begin = var_1996_begin_0, end = var_1996_end_0, end_mask = var_1996_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1996_cast_fp16")]; + tensor var_2003_begin_0 = const()[name = string("op_2003_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2003_end_0 = const()[name = string("op_2003_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2003_end_mask_0 = const()[name = string("op_2003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2003_cast_fp16 = slice_by_index(begin = var_2003_begin_0, end = var_2003_end_0, end_mask = var_2003_end_mask_0, x = var_1801_cast_fp16)[name = string("op_2003_cast_fp16")]; + tensor var_2010_begin_0 = const()[name = string("op_2010_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2010_end_0 = const()[name = string("op_2010_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2010_end_mask_0 = const()[name = string("op_2010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2010_cast_fp16 = slice_by_index(begin = var_2010_begin_0, end = var_2010_end_0, end_mask = var_2010_end_mask_0, x = var_1805_cast_fp16)[name = string("op_2010_cast_fp16")]; + tensor var_2017_begin_0 = const()[name = string("op_2017_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2017_end_0 = const()[name = string("op_2017_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2017_end_mask_0 = const()[name = string("op_2017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2017_cast_fp16 = slice_by_index(begin = var_2017_begin_0, end = var_2017_end_0, end_mask = var_2017_end_mask_0, x = var_1805_cast_fp16)[name = string("op_2017_cast_fp16")]; + tensor var_2024_begin_0 = const()[name = string("op_2024_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2024_end_0 = const()[name = string("op_2024_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2024_end_mask_0 = const()[name = string("op_2024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2024_cast_fp16 = slice_by_index(begin = var_2024_begin_0, end = var_2024_end_0, end_mask = var_2024_end_mask_0, x = var_1805_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2031_begin_0 = const()[name = string("op_2031_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2031_end_0 = const()[name = string("op_2031_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2031_end_mask_0 = const()[name = string("op_2031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2031_cast_fp16 = slice_by_index(begin = var_2031_begin_0, end = var_2031_end_0, end_mask = var_2031_end_mask_0, x = var_1805_cast_fp16)[name = string("op_2031_cast_fp16")]; + tensor var_2038_begin_0 = const()[name = string("op_2038_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2038_end_0 = const()[name = string("op_2038_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2038_end_mask_0 = const()[name = string("op_2038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2038_cast_fp16 = slice_by_index(begin = var_2038_begin_0, end = var_2038_end_0, end_mask = var_2038_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2038_cast_fp16")]; + tensor var_2045_begin_0 = const()[name = string("op_2045_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2045_end_0 = const()[name = string("op_2045_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2045_end_mask_0 = const()[name = string("op_2045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2045_cast_fp16 = slice_by_index(begin = var_2045_begin_0, end = var_2045_end_0, end_mask = var_2045_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2045_cast_fp16")]; + tensor var_2052_begin_0 = const()[name = string("op_2052_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2052_end_0 = const()[name = string("op_2052_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2052_end_mask_0 = const()[name = string("op_2052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2052_cast_fp16 = slice_by_index(begin = var_2052_begin_0, end = var_2052_end_0, end_mask = var_2052_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2052_cast_fp16")]; + tensor var_2059_begin_0 = const()[name = string("op_2059_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2059_end_0 = const()[name = string("op_2059_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2059_end_mask_0 = const()[name = string("op_2059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2059_cast_fp16 = slice_by_index(begin = var_2059_begin_0, end = var_2059_end_0, end_mask = var_2059_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2059_cast_fp16")]; + tensor var_2066_begin_0 = const()[name = string("op_2066_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2066_end_0 = const()[name = string("op_2066_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2066_end_mask_0 = const()[name = string("op_2066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2066_cast_fp16 = slice_by_index(begin = var_2066_begin_0, end = var_2066_end_0, end_mask = var_2066_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2066_cast_fp16")]; + tensor var_2073_begin_0 = const()[name = string("op_2073_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2073_end_0 = const()[name = string("op_2073_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2073_end_mask_0 = const()[name = string("op_2073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2073_cast_fp16 = slice_by_index(begin = var_2073_begin_0, end = var_2073_end_0, end_mask = var_2073_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2073_cast_fp16")]; + tensor var_2080_begin_0 = const()[name = string("op_2080_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2080_end_0 = const()[name = string("op_2080_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2080_end_mask_0 = const()[name = string("op_2080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2080_cast_fp16 = slice_by_index(begin = var_2080_begin_0, end = var_2080_end_0, end_mask = var_2080_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2080_cast_fp16")]; + tensor var_2087_begin_0 = const()[name = string("op_2087_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2087_end_0 = const()[name = string("op_2087_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2087_end_mask_0 = const()[name = string("op_2087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2087_cast_fp16 = slice_by_index(begin = var_2087_begin_0, end = var_2087_end_0, end_mask = var_2087_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2087_cast_fp16")]; + tensor var_2094_begin_0 = const()[name = string("op_2094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2094_end_0 = const()[name = string("op_2094_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2094_end_mask_0 = const()[name = string("op_2094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2094_cast_fp16 = slice_by_index(begin = var_2094_begin_0, end = var_2094_end_0, end_mask = var_2094_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2094_cast_fp16")]; + tensor var_2101_begin_0 = const()[name = string("op_2101_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2101_end_0 = const()[name = string("op_2101_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2101_end_mask_0 = const()[name = string("op_2101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2101_cast_fp16 = slice_by_index(begin = var_2101_begin_0, end = var_2101_end_0, end_mask = var_2101_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2101_cast_fp16")]; + tensor var_2108_begin_0 = const()[name = string("op_2108_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2108_end_0 = const()[name = string("op_2108_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2108_end_mask_0 = const()[name = string("op_2108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2108_cast_fp16 = slice_by_index(begin = var_2108_begin_0, end = var_2108_end_0, end_mask = var_2108_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2108_cast_fp16")]; + tensor var_2115_begin_0 = const()[name = string("op_2115_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2115_end_0 = const()[name = string("op_2115_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2115_end_mask_0 = const()[name = string("op_2115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2115_cast_fp16 = slice_by_index(begin = var_2115_begin_0, end = var_2115_end_0, end_mask = var_2115_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2115_cast_fp16")]; + tensor var_2122_begin_0 = const()[name = string("op_2122_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2122_end_0 = const()[name = string("op_2122_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2122_end_mask_0 = const()[name = string("op_2122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2122_cast_fp16 = slice_by_index(begin = var_2122_begin_0, end = var_2122_end_0, end_mask = var_2122_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2122_cast_fp16")]; + tensor var_2129_begin_0 = const()[name = string("op_2129_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2129_end_0 = const()[name = string("op_2129_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2129_end_mask_0 = const()[name = string("op_2129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2129_cast_fp16 = slice_by_index(begin = var_2129_begin_0, end = var_2129_end_0, end_mask = var_2129_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2129_cast_fp16")]; + tensor var_2136_begin_0 = const()[name = string("op_2136_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2136_end_0 = const()[name = string("op_2136_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2136_end_mask_0 = const()[name = string("op_2136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2136_cast_fp16 = slice_by_index(begin = var_2136_begin_0, end = var_2136_end_0, end_mask = var_2136_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2136_cast_fp16")]; + tensor var_2143_begin_0 = const()[name = string("op_2143_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2143_end_0 = const()[name = string("op_2143_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2143_end_mask_0 = const()[name = string("op_2143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2143_cast_fp16 = slice_by_index(begin = var_2143_begin_0, end = var_2143_end_0, end_mask = var_2143_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2143_cast_fp16")]; + tensor var_2150_begin_0 = const()[name = string("op_2150_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2150_end_0 = const()[name = string("op_2150_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2150_end_mask_0 = const()[name = string("op_2150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2150_cast_fp16 = slice_by_index(begin = var_2150_begin_0, end = var_2150_end_0, end_mask = var_2150_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2150_cast_fp16")]; + tensor var_2157_begin_0 = const()[name = string("op_2157_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2157_end_0 = const()[name = string("op_2157_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2157_end_mask_0 = const()[name = string("op_2157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2157_cast_fp16 = slice_by_index(begin = var_2157_begin_0, end = var_2157_end_0, end_mask = var_2157_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2157_cast_fp16")]; + tensor var_2164_begin_0 = const()[name = string("op_2164_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2164_end_0 = const()[name = string("op_2164_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2164_end_mask_0 = const()[name = string("op_2164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2164_cast_fp16")]; + tensor var_2171_begin_0 = const()[name = string("op_2171_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2171_end_0 = const()[name = string("op_2171_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2171_end_mask_0 = const()[name = string("op_2171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2171_cast_fp16 = slice_by_index(begin = var_2171_begin_0, end = var_2171_end_0, end_mask = var_2171_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2171_cast_fp16")]; + tensor var_2178_begin_0 = const()[name = string("op_2178_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2178_end_0 = const()[name = string("op_2178_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2178_end_mask_0 = const()[name = string("op_2178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2178_cast_fp16 = slice_by_index(begin = var_2178_begin_0, end = var_2178_end_0, end_mask = var_2178_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2178_cast_fp16")]; + tensor var_2185_begin_0 = const()[name = string("op_2185_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2185_end_0 = const()[name = string("op_2185_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2185_end_mask_0 = const()[name = string("op_2185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2185_cast_fp16 = slice_by_index(begin = var_2185_begin_0, end = var_2185_end_0, end_mask = var_2185_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2185_cast_fp16")]; + tensor var_2192_begin_0 = const()[name = string("op_2192_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2192_end_0 = const()[name = string("op_2192_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2192_end_mask_0 = const()[name = string("op_2192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2192_cast_fp16 = slice_by_index(begin = var_2192_begin_0, end = var_2192_end_0, end_mask = var_2192_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2192_cast_fp16")]; + tensor var_2199_begin_0 = const()[name = string("op_2199_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2199_end_0 = const()[name = string("op_2199_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2199_end_mask_0 = const()[name = string("op_2199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2199_cast_fp16 = slice_by_index(begin = var_2199_begin_0, end = var_2199_end_0, end_mask = var_2199_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2199_cast_fp16")]; + tensor var_2206_begin_0 = const()[name = string("op_2206_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2206_end_0 = const()[name = string("op_2206_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2206_end_mask_0 = const()[name = string("op_2206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2206_cast_fp16 = slice_by_index(begin = var_2206_begin_0, end = var_2206_end_0, end_mask = var_2206_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2206_cast_fp16")]; + tensor var_2213_begin_0 = const()[name = string("op_2213_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2213_end_0 = const()[name = string("op_2213_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2213_end_mask_0 = const()[name = string("op_2213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2213_cast_fp16 = slice_by_index(begin = var_2213_begin_0, end = var_2213_end_0, end_mask = var_2213_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2213_cast_fp16")]; + tensor var_2220_begin_0 = const()[name = string("op_2220_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2220_end_0 = const()[name = string("op_2220_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2220_end_mask_0 = const()[name = string("op_2220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2220_cast_fp16 = slice_by_index(begin = var_2220_begin_0, end = var_2220_end_0, end_mask = var_2220_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2220_cast_fp16")]; + tensor var_2227_begin_0 = const()[name = string("op_2227_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2227_end_0 = const()[name = string("op_2227_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2227_end_mask_0 = const()[name = string("op_2227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2227_cast_fp16 = slice_by_index(begin = var_2227_begin_0, end = var_2227_end_0, end_mask = var_2227_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2227_cast_fp16")]; + tensor var_2234_begin_0 = const()[name = string("op_2234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2234_end_0 = const()[name = string("op_2234_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2234_end_mask_0 = const()[name = string("op_2234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2234_cast_fp16 = slice_by_index(begin = var_2234_begin_0, end = var_2234_end_0, end_mask = var_2234_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2234_cast_fp16")]; + tensor var_2241_begin_0 = const()[name = string("op_2241_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2241_end_0 = const()[name = string("op_2241_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2241_end_mask_0 = const()[name = string("op_2241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2241_cast_fp16 = slice_by_index(begin = var_2241_begin_0, end = var_2241_end_0, end_mask = var_2241_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2241_cast_fp16")]; + tensor var_2248_begin_0 = const()[name = string("op_2248_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2248_end_0 = const()[name = string("op_2248_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2248_end_mask_0 = const()[name = string("op_2248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2248_cast_fp16 = slice_by_index(begin = var_2248_begin_0, end = var_2248_end_0, end_mask = var_2248_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2248_cast_fp16")]; + tensor var_2255_begin_0 = const()[name = string("op_2255_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2255_end_0 = const()[name = string("op_2255_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2255_end_mask_0 = const()[name = string("op_2255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2255_cast_fp16 = slice_by_index(begin = var_2255_begin_0, end = var_2255_end_0, end_mask = var_2255_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2255_cast_fp16")]; + tensor var_2262_begin_0 = const()[name = string("op_2262_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2262_end_0 = const()[name = string("op_2262_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2262_end_mask_0 = const()[name = string("op_2262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2262_cast_fp16 = slice_by_index(begin = var_2262_begin_0, end = var_2262_end_0, end_mask = var_2262_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2262_cast_fp16")]; + tensor var_2269_begin_0 = const()[name = string("op_2269_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2269_end_0 = const()[name = string("op_2269_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2269_end_mask_0 = const()[name = string("op_2269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2269_cast_fp16 = slice_by_index(begin = var_2269_begin_0, end = var_2269_end_0, end_mask = var_2269_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2269_cast_fp16")]; + tensor var_2276_begin_0 = const()[name = string("op_2276_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2276_end_0 = const()[name = string("op_2276_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2276_end_mask_0 = const()[name = string("op_2276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2276_cast_fp16 = slice_by_index(begin = var_2276_begin_0, end = var_2276_end_0, end_mask = var_2276_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2276_cast_fp16")]; + tensor var_2283_begin_0 = const()[name = string("op_2283_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2283_end_0 = const()[name = string("op_2283_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2283_end_mask_0 = const()[name = string("op_2283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2283_cast_fp16 = slice_by_index(begin = var_2283_begin_0, end = var_2283_end_0, end_mask = var_2283_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2283_cast_fp16")]; + tensor var_2290_begin_0 = const()[name = string("op_2290_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2290_end_0 = const()[name = string("op_2290_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2290_end_mask_0 = const()[name = string("op_2290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2290_cast_fp16 = slice_by_index(begin = var_2290_begin_0, end = var_2290_end_0, end_mask = var_2290_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2290_cast_fp16")]; + tensor var_2297_begin_0 = const()[name = string("op_2297_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2297_end_0 = const()[name = string("op_2297_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2297_end_mask_0 = const()[name = string("op_2297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2297_cast_fp16 = slice_by_index(begin = var_2297_begin_0, end = var_2297_end_0, end_mask = var_2297_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2297_cast_fp16")]; + tensor var_2304_begin_0 = const()[name = string("op_2304_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2304_end_0 = const()[name = string("op_2304_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2304_end_mask_0 = const()[name = string("op_2304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2304_cast_fp16 = slice_by_index(begin = var_2304_begin_0, end = var_2304_end_0, end_mask = var_2304_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2304_cast_fp16")]; + tensor var_2311_begin_0 = const()[name = string("op_2311_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2311_end_0 = const()[name = string("op_2311_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2311_end_mask_0 = const()[name = string("op_2311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2311_cast_fp16 = slice_by_index(begin = var_2311_begin_0, end = var_2311_end_0, end_mask = var_2311_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2311_cast_fp16")]; + tensor var_2318_begin_0 = const()[name = string("op_2318_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2318_end_0 = const()[name = string("op_2318_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2318_end_mask_0 = const()[name = string("op_2318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2318_cast_fp16 = slice_by_index(begin = var_2318_begin_0, end = var_2318_end_0, end_mask = var_2318_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2318_cast_fp16")]; + tensor var_2325_begin_0 = const()[name = string("op_2325_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2325_end_0 = const()[name = string("op_2325_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2325_end_mask_0 = const()[name = string("op_2325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2325_cast_fp16 = slice_by_index(begin = var_2325_begin_0, end = var_2325_end_0, end_mask = var_2325_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2325_cast_fp16")]; + tensor var_2332_begin_0 = const()[name = string("op_2332_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2332_end_0 = const()[name = string("op_2332_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2332_end_mask_0 = const()[name = string("op_2332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2332_cast_fp16 = slice_by_index(begin = var_2332_begin_0, end = var_2332_end_0, end_mask = var_2332_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2332_cast_fp16")]; + tensor var_2339_begin_0 = const()[name = string("op_2339_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2339_end_0 = const()[name = string("op_2339_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2339_end_mask_0 = const()[name = string("op_2339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2339_cast_fp16 = slice_by_index(begin = var_2339_begin_0, end = var_2339_end_0, end_mask = var_2339_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2339_cast_fp16")]; + tensor var_2346_begin_0 = const()[name = string("op_2346_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2346_end_0 = const()[name = string("op_2346_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2346_end_mask_0 = const()[name = string("op_2346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2346_cast_fp16 = slice_by_index(begin = var_2346_begin_0, end = var_2346_end_0, end_mask = var_2346_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2346_cast_fp16")]; + tensor var_2353_begin_0 = const()[name = string("op_2353_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2353_end_0 = const()[name = string("op_2353_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2353_end_mask_0 = const()[name = string("op_2353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2353_cast_fp16 = slice_by_index(begin = var_2353_begin_0, end = var_2353_end_0, end_mask = var_2353_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2353_cast_fp16")]; + tensor var_2360_begin_0 = const()[name = string("op_2360_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2360_end_0 = const()[name = string("op_2360_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2360_end_mask_0 = const()[name = string("op_2360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2360_cast_fp16 = slice_by_index(begin = var_2360_begin_0, end = var_2360_end_0, end_mask = var_2360_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2360_cast_fp16")]; + tensor var_2367_begin_0 = const()[name = string("op_2367_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2367_end_0 = const()[name = string("op_2367_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2367_end_mask_0 = const()[name = string("op_2367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2367_cast_fp16 = slice_by_index(begin = var_2367_begin_0, end = var_2367_end_0, end_mask = var_2367_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2367_cast_fp16")]; + tensor var_2374_begin_0 = const()[name = string("op_2374_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2374_end_0 = const()[name = string("op_2374_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2374_end_mask_0 = const()[name = string("op_2374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2374_cast_fp16 = slice_by_index(begin = var_2374_begin_0, end = var_2374_end_0, end_mask = var_2374_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2374_cast_fp16")]; + tensor var_2381_begin_0 = const()[name = string("op_2381_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2381_end_0 = const()[name = string("op_2381_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2381_end_mask_0 = const()[name = string("op_2381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2381_cast_fp16 = slice_by_index(begin = var_2381_begin_0, end = var_2381_end_0, end_mask = var_2381_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2381_cast_fp16")]; + tensor var_2388_begin_0 = const()[name = string("op_2388_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2388_end_0 = const()[name = string("op_2388_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2388_end_mask_0 = const()[name = string("op_2388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2388_cast_fp16 = slice_by_index(begin = var_2388_begin_0, end = var_2388_end_0, end_mask = var_2388_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2388_cast_fp16")]; + tensor var_2395_begin_0 = const()[name = string("op_2395_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2395_end_0 = const()[name = string("op_2395_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2395_end_mask_0 = const()[name = string("op_2395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2395_cast_fp16 = slice_by_index(begin = var_2395_begin_0, end = var_2395_end_0, end_mask = var_2395_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2395_cast_fp16")]; + tensor var_2402_begin_0 = const()[name = string("op_2402_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2402_end_0 = const()[name = string("op_2402_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2402_end_mask_0 = const()[name = string("op_2402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2402_cast_fp16 = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2402_cast_fp16")]; + tensor var_2409_begin_0 = const()[name = string("op_2409_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2409_end_0 = const()[name = string("op_2409_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2409_end_mask_0 = const()[name = string("op_2409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2409_cast_fp16 = slice_by_index(begin = var_2409_begin_0, end = var_2409_end_0, end_mask = var_2409_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2409_cast_fp16")]; + tensor var_2416_begin_0 = const()[name = string("op_2416_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2416_end_0 = const()[name = string("op_2416_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2416_end_mask_0 = const()[name = string("op_2416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2416_cast_fp16 = slice_by_index(begin = var_2416_begin_0, end = var_2416_end_0, end_mask = var_2416_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2416_cast_fp16")]; + tensor var_2423_begin_0 = const()[name = string("op_2423_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2423_end_0 = const()[name = string("op_2423_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2423_end_mask_0 = const()[name = string("op_2423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2423_cast_fp16 = slice_by_index(begin = var_2423_begin_0, end = var_2423_end_0, end_mask = var_2423_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2423_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2428_begin_0 = const()[name = string("op_2428_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2428_end_0 = const()[name = string("op_2428_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2428_end_mask_0 = const()[name = string("op_2428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_30")]; + tensor var_2428_cast_fp16 = slice_by_index(begin = var_2428_begin_0, end = var_2428_end_0, end_mask = var_2428_end_mask_0, x = k_3_cast_fp16)[name = string("op_2428_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = k_3_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2436_begin_0 = const()[name = string("op_2436_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2436_end_0 = const()[name = string("op_2436_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2436_end_mask_0 = const()[name = string("op_2436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2436_cast_fp16 = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = k_3_cast_fp16)[name = string("op_2436_cast_fp16")]; + tensor var_2440_begin_0 = const()[name = string("op_2440_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2440_end_0 = const()[name = string("op_2440_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2440_end_mask_0 = const()[name = string("op_2440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2440_cast_fp16 = slice_by_index(begin = var_2440_begin_0, end = var_2440_end_0, end_mask = var_2440_end_mask_0, x = k_3_cast_fp16)[name = string("op_2440_cast_fp16")]; + tensor var_2444_begin_0 = const()[name = string("op_2444_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2444_end_0 = const()[name = string("op_2444_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2444_end_mask_0 = const()[name = string("op_2444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2444_cast_fp16 = slice_by_index(begin = var_2444_begin_0, end = var_2444_end_0, end_mask = var_2444_end_mask_0, x = k_3_cast_fp16)[name = string("op_2444_cast_fp16")]; + tensor var_2448_begin_0 = const()[name = string("op_2448_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2448_end_0 = const()[name = string("op_2448_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2448_end_mask_0 = const()[name = string("op_2448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2448_cast_fp16 = slice_by_index(begin = var_2448_begin_0, end = var_2448_end_0, end_mask = var_2448_end_mask_0, x = k_3_cast_fp16)[name = string("op_2448_cast_fp16")]; + tensor var_2452_begin_0 = const()[name = string("op_2452_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2452_end_0 = const()[name = string("op_2452_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2452_end_mask_0 = const()[name = string("op_2452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2452_cast_fp16 = slice_by_index(begin = var_2452_begin_0, end = var_2452_end_0, end_mask = var_2452_end_mask_0, x = k_3_cast_fp16)[name = string("op_2452_cast_fp16")]; + tensor var_2456_begin_0 = const()[name = string("op_2456_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2456_end_0 = const()[name = string("op_2456_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2456_end_mask_0 = const()[name = string("op_2456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2456_cast_fp16 = slice_by_index(begin = var_2456_begin_0, end = var_2456_end_0, end_mask = var_2456_end_mask_0, x = k_3_cast_fp16)[name = string("op_2456_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = k_3_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2464_begin_0 = const()[name = string("op_2464_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2464_end_0 = const()[name = string("op_2464_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_2464_end_mask_0 = const()[name = string("op_2464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2464_cast_fp16 = slice_by_index(begin = var_2464_begin_0, end = var_2464_end_0, end_mask = var_2464_end_mask_0, x = k_3_cast_fp16)[name = string("op_2464_cast_fp16")]; + tensor var_2468_begin_0 = const()[name = string("op_2468_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2468_end_0 = const()[name = string("op_2468_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_2468_end_mask_0 = const()[name = string("op_2468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2468_cast_fp16 = slice_by_index(begin = var_2468_begin_0, end = var_2468_end_0, end_mask = var_2468_end_mask_0, x = k_3_cast_fp16)[name = string("op_2468_cast_fp16")]; + tensor var_2472_begin_0 = const()[name = string("op_2472_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2472_end_0 = const()[name = string("op_2472_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_2472_end_mask_0 = const()[name = string("op_2472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2472_cast_fp16 = slice_by_index(begin = var_2472_begin_0, end = var_2472_end_0, end_mask = var_2472_end_mask_0, x = k_3_cast_fp16)[name = string("op_2472_cast_fp16")]; + tensor var_2476_begin_0 = const()[name = string("op_2476_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_2476_end_0 = const()[name = string("op_2476_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_2476_end_mask_0 = const()[name = string("op_2476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2476_cast_fp16 = slice_by_index(begin = var_2476_begin_0, end = var_2476_end_0, end_mask = var_2476_end_mask_0, x = k_3_cast_fp16)[name = string("op_2476_cast_fp16")]; + tensor var_2480_begin_0 = const()[name = string("op_2480_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_2480_end_0 = const()[name = string("op_2480_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_2480_end_mask_0 = const()[name = string("op_2480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2480_cast_fp16 = slice_by_index(begin = var_2480_begin_0, end = var_2480_end_0, end_mask = var_2480_end_mask_0, x = k_3_cast_fp16)[name = string("op_2480_cast_fp16")]; + tensor var_2484_begin_0 = const()[name = string("op_2484_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_2484_end_0 = const()[name = string("op_2484_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_2484_end_mask_0 = const()[name = string("op_2484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2484_cast_fp16 = slice_by_index(begin = var_2484_begin_0, end = var_2484_end_0, end_mask = var_2484_end_mask_0, x = k_3_cast_fp16)[name = string("op_2484_cast_fp16")]; + tensor var_2488_begin_0 = const()[name = string("op_2488_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_2488_end_0 = const()[name = string("op_2488_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_2488_end_mask_0 = const()[name = string("op_2488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2488_cast_fp16 = slice_by_index(begin = var_2488_begin_0, end = var_2488_end_0, end_mask = var_2488_end_mask_0, x = k_3_cast_fp16)[name = string("op_2488_cast_fp16")]; + tensor var_2492_begin_0 = const()[name = string("op_2492_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2492_end_0 = const()[name = string("op_2492_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_2492_end_mask_0 = const()[name = string("op_2492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2492_cast_fp16 = slice_by_index(begin = var_2492_begin_0, end = var_2492_end_0, end_mask = var_2492_end_mask_0, x = k_3_cast_fp16)[name = string("op_2492_cast_fp16")]; + tensor var_2496_begin_0 = const()[name = string("op_2496_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_2496_end_0 = const()[name = string("op_2496_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_2496_end_mask_0 = const()[name = string("op_2496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2496_cast_fp16 = slice_by_index(begin = var_2496_begin_0, end = var_2496_end_0, end_mask = var_2496_end_mask_0, x = k_3_cast_fp16)[name = string("op_2496_cast_fp16")]; + tensor var_2500_begin_0 = const()[name = string("op_2500_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_2500_end_0 = const()[name = string("op_2500_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_2500_end_mask_0 = const()[name = string("op_2500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2500_cast_fp16 = slice_by_index(begin = var_2500_begin_0, end = var_2500_end_0, end_mask = var_2500_end_mask_0, x = k_3_cast_fp16)[name = string("op_2500_cast_fp16")]; + tensor var_2504_begin_0 = const()[name = string("op_2504_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_2504_end_0 = const()[name = string("op_2504_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_2504_end_mask_0 = const()[name = string("op_2504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2504_cast_fp16 = slice_by_index(begin = var_2504_begin_0, end = var_2504_end_0, end_mask = var_2504_end_mask_0, x = k_3_cast_fp16)[name = string("op_2504_cast_fp16")]; + tensor var_2506_begin_0 = const()[name = string("op_2506_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2506_end_0 = const()[name = string("op_2506_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2506_end_mask_0 = const()[name = string("op_2506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2506_cast_fp16 = slice_by_index(begin = var_2506_begin_0, end = var_2506_end_0, end_mask = var_2506_end_mask_0, x = value_3_cast_fp16)[name = string("op_2506_cast_fp16")]; + tensor var_2510_begin_0 = const()[name = string("op_2510_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2510_end_0 = const()[name = string("op_2510_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2510_end_mask_0 = const()[name = string("op_2510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2510_cast_fp16 = slice_by_index(begin = var_2510_begin_0, end = var_2510_end_0, end_mask = var_2510_end_mask_0, x = value_3_cast_fp16)[name = string("op_2510_cast_fp16")]; + tensor var_2514_begin_0 = const()[name = string("op_2514_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2514_end_0 = const()[name = string("op_2514_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2514_end_mask_0 = const()[name = string("op_2514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2514_cast_fp16 = slice_by_index(begin = var_2514_begin_0, end = var_2514_end_0, end_mask = var_2514_end_mask_0, x = value_3_cast_fp16)[name = string("op_2514_cast_fp16")]; + tensor var_2518_begin_0 = const()[name = string("op_2518_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2518_end_0 = const()[name = string("op_2518_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2518_end_mask_0 = const()[name = string("op_2518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2518_cast_fp16 = slice_by_index(begin = var_2518_begin_0, end = var_2518_end_0, end_mask = var_2518_end_mask_0, x = value_3_cast_fp16)[name = string("op_2518_cast_fp16")]; + tensor var_2522_begin_0 = const()[name = string("op_2522_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2522_end_0 = const()[name = string("op_2522_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2522_end_mask_0 = const()[name = string("op_2522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2522_cast_fp16 = slice_by_index(begin = var_2522_begin_0, end = var_2522_end_0, end_mask = var_2522_end_mask_0, x = value_3_cast_fp16)[name = string("op_2522_cast_fp16")]; + tensor var_2526_begin_0 = const()[name = string("op_2526_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2526_end_0 = const()[name = string("op_2526_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2526_end_mask_0 = const()[name = string("op_2526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2526_cast_fp16 = slice_by_index(begin = var_2526_begin_0, end = var_2526_end_0, end_mask = var_2526_end_mask_0, x = value_3_cast_fp16)[name = string("op_2526_cast_fp16")]; + tensor var_2530_begin_0 = const()[name = string("op_2530_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2530_end_0 = const()[name = string("op_2530_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2530_end_mask_0 = const()[name = string("op_2530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2530_cast_fp16 = slice_by_index(begin = var_2530_begin_0, end = var_2530_end_0, end_mask = var_2530_end_mask_0, x = value_3_cast_fp16)[name = string("op_2530_cast_fp16")]; + tensor var_2534_begin_0 = const()[name = string("op_2534_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2534_end_0 = const()[name = string("op_2534_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2534_end_mask_0 = const()[name = string("op_2534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2534_cast_fp16 = slice_by_index(begin = var_2534_begin_0, end = var_2534_end_0, end_mask = var_2534_end_mask_0, x = value_3_cast_fp16)[name = string("op_2534_cast_fp16")]; + tensor var_2538_begin_0 = const()[name = string("op_2538_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2538_end_0 = const()[name = string("op_2538_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2538_end_mask_0 = const()[name = string("op_2538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2538_cast_fp16 = slice_by_index(begin = var_2538_begin_0, end = var_2538_end_0, end_mask = var_2538_end_mask_0, x = value_3_cast_fp16)[name = string("op_2538_cast_fp16")]; + tensor var_2542_begin_0 = const()[name = string("op_2542_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2542_end_0 = const()[name = string("op_2542_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2542_end_mask_0 = const()[name = string("op_2542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2542_cast_fp16 = slice_by_index(begin = var_2542_begin_0, end = var_2542_end_0, end_mask = var_2542_end_mask_0, x = value_3_cast_fp16)[name = string("op_2542_cast_fp16")]; + tensor var_2546_begin_0 = const()[name = string("op_2546_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2546_end_0 = const()[name = string("op_2546_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2546_end_mask_0 = const()[name = string("op_2546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2546_cast_fp16 = slice_by_index(begin = var_2546_begin_0, end = var_2546_end_0, end_mask = var_2546_end_mask_0, x = value_3_cast_fp16)[name = string("op_2546_cast_fp16")]; + tensor var_2550_begin_0 = const()[name = string("op_2550_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2550_end_0 = const()[name = string("op_2550_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2550_end_mask_0 = const()[name = string("op_2550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2550_cast_fp16 = slice_by_index(begin = var_2550_begin_0, end = var_2550_end_0, end_mask = var_2550_end_mask_0, x = value_3_cast_fp16)[name = string("op_2550_cast_fp16")]; + tensor var_2554_begin_0 = const()[name = string("op_2554_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2554_end_0 = const()[name = string("op_2554_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_2554_end_mask_0 = const()[name = string("op_2554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2554_cast_fp16 = slice_by_index(begin = var_2554_begin_0, end = var_2554_end_0, end_mask = var_2554_end_mask_0, x = value_3_cast_fp16)[name = string("op_2554_cast_fp16")]; + tensor var_2558_begin_0 = const()[name = string("op_2558_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2558_end_0 = const()[name = string("op_2558_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_2558_end_mask_0 = const()[name = string("op_2558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2558_cast_fp16 = slice_by_index(begin = var_2558_begin_0, end = var_2558_end_0, end_mask = var_2558_end_mask_0, x = value_3_cast_fp16)[name = string("op_2558_cast_fp16")]; + tensor var_2562_begin_0 = const()[name = string("op_2562_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2562_end_0 = const()[name = string("op_2562_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_2562_end_mask_0 = const()[name = string("op_2562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2562_cast_fp16 = slice_by_index(begin = var_2562_begin_0, end = var_2562_end_0, end_mask = var_2562_end_mask_0, x = value_3_cast_fp16)[name = string("op_2562_cast_fp16")]; + tensor var_2566_begin_0 = const()[name = string("op_2566_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2566_end_0 = const()[name = string("op_2566_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_2566_end_mask_0 = const()[name = string("op_2566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2566_cast_fp16 = slice_by_index(begin = var_2566_begin_0, end = var_2566_end_0, end_mask = var_2566_end_mask_0, x = value_3_cast_fp16)[name = string("op_2566_cast_fp16")]; + tensor var_2570_begin_0 = const()[name = string("op_2570_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2570_end_0 = const()[name = string("op_2570_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_2570_end_mask_0 = const()[name = string("op_2570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2570_cast_fp16 = slice_by_index(begin = var_2570_begin_0, end = var_2570_end_0, end_mask = var_2570_end_mask_0, x = value_3_cast_fp16)[name = string("op_2570_cast_fp16")]; + tensor var_2574_begin_0 = const()[name = string("op_2574_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2574_end_0 = const()[name = string("op_2574_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_2574_end_mask_0 = const()[name = string("op_2574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2574_cast_fp16 = slice_by_index(begin = var_2574_begin_0, end = var_2574_end_0, end_mask = var_2574_end_mask_0, x = value_3_cast_fp16)[name = string("op_2574_cast_fp16")]; + tensor var_2578_begin_0 = const()[name = string("op_2578_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_2578_end_0 = const()[name = string("op_2578_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_2578_end_mask_0 = const()[name = string("op_2578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2578_cast_fp16 = slice_by_index(begin = var_2578_begin_0, end = var_2578_end_0, end_mask = var_2578_end_mask_0, x = value_3_cast_fp16)[name = string("op_2578_cast_fp16")]; + tensor var_2582_begin_0 = const()[name = string("op_2582_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_2582_end_0 = const()[name = string("op_2582_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_2582_end_mask_0 = const()[name = string("op_2582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2582_cast_fp16 = slice_by_index(begin = var_2582_begin_0, end = var_2582_end_0, end_mask = var_2582_end_mask_0, x = value_3_cast_fp16)[name = string("op_2582_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_2428_cast_fp16, var_1870_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_2428_cast_fp16, var_1877_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_2428_cast_fp16, var_1884_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_2428_cast_fp16, var_1891_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_2432_cast_fp16, var_1898_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_2432_cast_fp16, var_1905_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_2432_cast_fp16, var_1912_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_2432_cast_fp16, var_1919_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_2436_cast_fp16, var_1926_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_2436_cast_fp16, var_1933_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_2436_cast_fp16, var_1940_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_2436_cast_fp16, var_1947_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_2440_cast_fp16, var_1954_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_2440_cast_fp16, var_1961_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_2440_cast_fp16, var_1968_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_2440_cast_fp16, var_1975_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2444_cast_fp16, var_1982_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2444_cast_fp16, var_1989_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2444_cast_fp16, var_1996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2444_cast_fp16, var_2003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2448_cast_fp16, var_2010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2448_cast_fp16, var_2017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2448_cast_fp16, var_2024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2448_cast_fp16, var_2031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2452_cast_fp16, var_2038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2452_cast_fp16, var_2045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2452_cast_fp16, var_2052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2452_cast_fp16, var_2059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2456_cast_fp16, var_2066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2456_cast_fp16, var_2073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2456_cast_fp16, var_2080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2456_cast_fp16, var_2087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2460_cast_fp16, var_2094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2460_cast_fp16, var_2101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2460_cast_fp16, var_2108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2460_cast_fp16, var_2115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2464_cast_fp16, var_2122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2464_cast_fp16, var_2129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2464_cast_fp16, var_2136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2464_cast_fp16, var_2143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2468_cast_fp16, var_2150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2468_cast_fp16, var_2157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2468_cast_fp16, var_2164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2468_cast_fp16, var_2171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2472_cast_fp16, var_2178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2472_cast_fp16, var_2185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2472_cast_fp16, var_2192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2472_cast_fp16, var_2199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_2476_cast_fp16, var_2206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_2476_cast_fp16, var_2213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_2476_cast_fp16, var_2220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_2476_cast_fp16, var_2227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_2480_cast_fp16, var_2234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_2480_cast_fp16, var_2241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_2480_cast_fp16, var_2248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_2480_cast_fp16, var_2255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_2484_cast_fp16, var_2262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_2484_cast_fp16, var_2269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_2484_cast_fp16, var_2276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_2484_cast_fp16, var_2283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_2488_cast_fp16, var_2290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_2488_cast_fp16, var_2297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_2488_cast_fp16, var_2304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_2488_cast_fp16, var_2311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_2492_cast_fp16, var_2318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_2492_cast_fp16, var_2325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_2492_cast_fp16, var_2332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_2492_cast_fp16, var_2339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_2496_cast_fp16, var_2346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_2496_cast_fp16, var_2353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_2496_cast_fp16, var_2360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_2496_cast_fp16, var_2367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_2500_cast_fp16, var_2374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_2500_cast_fp16, var_2381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_2500_cast_fp16, var_2388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_2500_cast_fp16, var_2395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_2504_cast_fp16, var_2402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_2504_cast_fp16, var_2409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_2504_cast_fp16, var_2416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_2504_cast_fp16, var_2423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + fp16 var_2745_to_fp16 = const()[name = string("op_2745_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_2745_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_2747_to_fp16 = const()[name = string("op_2747_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_2747_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_2749_to_fp16 = const()[name = string("op_2749_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_2749_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_2751_to_fp16 = const()[name = string("op_2751_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_2751_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_2753_to_fp16 = const()[name = string("op_2753_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_2753_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_2755_to_fp16 = const()[name = string("op_2755_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_2755_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_2757_to_fp16 = const()[name = string("op_2757_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_2757_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_2759_to_fp16 = const()[name = string("op_2759_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_2759_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_2761_to_fp16 = const()[name = string("op_2761_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_2761_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_2763_to_fp16 = const()[name = string("op_2763_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_2763_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_2765_to_fp16 = const()[name = string("op_2765_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_2765_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_2767_to_fp16 = const()[name = string("op_2767_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_2767_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_2769_to_fp16 = const()[name = string("op_2769_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_2769_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_2771_to_fp16 = const()[name = string("op_2771_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_2771_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_2773_to_fp16 = const()[name = string("op_2773_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_2773_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_2775_to_fp16 = const()[name = string("op_2775_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_2775_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + fp16 var_2777_to_fp16 = const()[name = string("op_2777_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2777_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2779_to_fp16 = const()[name = string("op_2779_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2779_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2781_to_fp16 = const()[name = string("op_2781_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2781_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2783_to_fp16 = const()[name = string("op_2783_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2783_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2785_to_fp16 = const()[name = string("op_2785_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2785_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2787_to_fp16 = const()[name = string("op_2787_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2787_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2789_to_fp16 = const()[name = string("op_2789_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2789_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2791_to_fp16 = const()[name = string("op_2791_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2791_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2793_to_fp16 = const()[name = string("op_2793_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2793_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2795_to_fp16 = const()[name = string("op_2795_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2795_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2797_to_fp16 = const()[name = string("op_2797_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2797_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2799_to_fp16 = const()[name = string("op_2799_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2799_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2801_to_fp16 = const()[name = string("op_2801_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2801_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2803_to_fp16 = const()[name = string("op_2803_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2803_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2805_to_fp16 = const()[name = string("op_2805_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2805_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2807_to_fp16 = const()[name = string("op_2807_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2807_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2809_to_fp16 = const()[name = string("op_2809_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2809_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2811_to_fp16 = const()[name = string("op_2811_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2811_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2813_to_fp16 = const()[name = string("op_2813_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2813_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2815_to_fp16 = const()[name = string("op_2815_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2815_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2817_to_fp16 = const()[name = string("op_2817_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2817_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2819_to_fp16 = const()[name = string("op_2819_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2819_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2821_to_fp16 = const()[name = string("op_2821_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2821_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2823_to_fp16 = const()[name = string("op_2823_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2823_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2825_to_fp16 = const()[name = string("op_2825_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2825_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2827_to_fp16 = const()[name = string("op_2827_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2827_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2829_to_fp16 = const()[name = string("op_2829_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2829_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2831_to_fp16 = const()[name = string("op_2831_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2831_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2833_to_fp16 = const()[name = string("op_2833_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2833_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2835_to_fp16 = const()[name = string("op_2835_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2835_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2837_to_fp16 = const()[name = string("op_2837_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2837_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2839_to_fp16 = const()[name = string("op_2839_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2839_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + fp16 var_2841_to_fp16 = const()[name = string("op_2841_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_2841_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_2843_to_fp16 = const()[name = string("op_2843_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_2843_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_2845_to_fp16 = const()[name = string("op_2845_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_2845_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_2847_to_fp16 = const()[name = string("op_2847_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_2847_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_2849_to_fp16 = const()[name = string("op_2849_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_2849_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_2851_to_fp16 = const()[name = string("op_2851_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_2851_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_2853_to_fp16 = const()[name = string("op_2853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_2853_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_2855_to_fp16 = const()[name = string("op_2855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_2855_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_2857_to_fp16 = const()[name = string("op_2857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_2857_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_2859_to_fp16 = const()[name = string("op_2859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_2859_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_2861_to_fp16 = const()[name = string("op_2861_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_2861_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_2863_to_fp16 = const()[name = string("op_2863_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_2863_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_2865_to_fp16 = const()[name = string("op_2865_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_2865_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_2867_to_fp16 = const()[name = string("op_2867_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_2867_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_2869_to_fp16 = const()[name = string("op_2869_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_2869_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_2871_to_fp16 = const()[name = string("op_2871_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_2871_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + fp16 var_2873_to_fp16 = const()[name = string("op_2873_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_2873_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_2875_to_fp16 = const()[name = string("op_2875_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_2875_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_2877_to_fp16 = const()[name = string("op_2877_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_2877_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_2879_to_fp16 = const()[name = string("op_2879_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_2879_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_2881_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_2883_to_fp16 = const()[name = string("op_2883_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_2883_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_2885_to_fp16 = const()[name = string("op_2885_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_2885_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_2887_to_fp16 = const()[name = string("op_2887_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_2887_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_2889_to_fp16 = const()[name = string("op_2889_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_2889_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_2891_to_fp16 = const()[name = string("op_2891_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_2891_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_2893_to_fp16 = const()[name = string("op_2893_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_2893_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_2895_to_fp16 = const()[name = string("op_2895_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_2895_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_2897_to_fp16 = const()[name = string("op_2897_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_2897_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_2899_to_fp16 = const()[name = string("op_2899_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_2899_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_2901_to_fp16 = const()[name = string("op_2901_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_2901_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_2903_to_fp16 = const()[name = string("op_2903_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_2903_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + tensor var_2905_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_161_cast_fp16)[name = string("op_2905_cast_fp16")]; + tensor var_2906_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_163_cast_fp16)[name = string("op_2906_cast_fp16")]; + tensor var_2907_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_165_cast_fp16)[name = string("op_2907_cast_fp16")]; + tensor var_2908_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_167_cast_fp16)[name = string("op_2908_cast_fp16")]; + tensor var_2909_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_169_cast_fp16)[name = string("op_2909_cast_fp16")]; + tensor var_2910_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_171_cast_fp16)[name = string("op_2910_cast_fp16")]; + tensor var_2911_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_173_cast_fp16)[name = string("op_2911_cast_fp16")]; + tensor var_2912_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_175_cast_fp16)[name = string("op_2912_cast_fp16")]; + tensor var_2913_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_177_cast_fp16)[name = string("op_2913_cast_fp16")]; + tensor var_2914_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_179_cast_fp16)[name = string("op_2914_cast_fp16")]; + tensor var_2915_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_181_cast_fp16)[name = string("op_2915_cast_fp16")]; + tensor var_2916_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_183_cast_fp16)[name = string("op_2916_cast_fp16")]; + tensor var_2917_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_185_cast_fp16)[name = string("op_2917_cast_fp16")]; + tensor var_2918_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_187_cast_fp16)[name = string("op_2918_cast_fp16")]; + tensor var_2919_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_189_cast_fp16)[name = string("op_2919_cast_fp16")]; + tensor var_2920_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_191_cast_fp16)[name = string("op_2920_cast_fp16")]; + tensor var_2921_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_193_cast_fp16)[name = string("op_2921_cast_fp16")]; + tensor var_2922_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_195_cast_fp16)[name = string("op_2922_cast_fp16")]; + tensor var_2923_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_197_cast_fp16)[name = string("op_2923_cast_fp16")]; + tensor var_2924_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_199_cast_fp16)[name = string("op_2924_cast_fp16")]; + tensor var_2925_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_201_cast_fp16)[name = string("op_2925_cast_fp16")]; + tensor var_2926_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_203_cast_fp16)[name = string("op_2926_cast_fp16")]; + tensor var_2927_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_205_cast_fp16)[name = string("op_2927_cast_fp16")]; + tensor var_2928_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_207_cast_fp16)[name = string("op_2928_cast_fp16")]; + tensor var_2929_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_209_cast_fp16)[name = string("op_2929_cast_fp16")]; + tensor var_2930_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_211_cast_fp16)[name = string("op_2930_cast_fp16")]; + tensor var_2931_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_213_cast_fp16)[name = string("op_2931_cast_fp16")]; + tensor var_2932_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_215_cast_fp16)[name = string("op_2932_cast_fp16")]; + tensor var_2933_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_217_cast_fp16)[name = string("op_2933_cast_fp16")]; + tensor var_2934_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_219_cast_fp16)[name = string("op_2934_cast_fp16")]; + tensor var_2935_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_221_cast_fp16)[name = string("op_2935_cast_fp16")]; + tensor var_2936_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_223_cast_fp16)[name = string("op_2936_cast_fp16")]; + tensor var_2937_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_225_cast_fp16)[name = string("op_2937_cast_fp16")]; + tensor var_2938_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_227_cast_fp16)[name = string("op_2938_cast_fp16")]; + tensor var_2939_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_229_cast_fp16)[name = string("op_2939_cast_fp16")]; + tensor var_2940_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_231_cast_fp16)[name = string("op_2940_cast_fp16")]; + tensor var_2941_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_233_cast_fp16)[name = string("op_2941_cast_fp16")]; + tensor var_2942_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_235_cast_fp16)[name = string("op_2942_cast_fp16")]; + tensor var_2943_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_237_cast_fp16)[name = string("op_2943_cast_fp16")]; + tensor var_2944_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_239_cast_fp16)[name = string("op_2944_cast_fp16")]; + tensor var_2945_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_241_cast_fp16)[name = string("op_2945_cast_fp16")]; + tensor var_2946_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_243_cast_fp16)[name = string("op_2946_cast_fp16")]; + tensor var_2947_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_245_cast_fp16)[name = string("op_2947_cast_fp16")]; + tensor var_2948_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_247_cast_fp16)[name = string("op_2948_cast_fp16")]; + tensor var_2949_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_249_cast_fp16)[name = string("op_2949_cast_fp16")]; + tensor var_2950_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_251_cast_fp16)[name = string("op_2950_cast_fp16")]; + tensor var_2951_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_253_cast_fp16)[name = string("op_2951_cast_fp16")]; + tensor var_2952_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_255_cast_fp16)[name = string("op_2952_cast_fp16")]; + tensor var_2953_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_257_cast_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2954_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_259_cast_fp16)[name = string("op_2954_cast_fp16")]; + tensor var_2955_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_261_cast_fp16)[name = string("op_2955_cast_fp16")]; + tensor var_2956_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_263_cast_fp16)[name = string("op_2956_cast_fp16")]; + tensor var_2957_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_265_cast_fp16)[name = string("op_2957_cast_fp16")]; + tensor var_2958_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_267_cast_fp16)[name = string("op_2958_cast_fp16")]; + tensor var_2959_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_269_cast_fp16)[name = string("op_2959_cast_fp16")]; + tensor var_2960_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_271_cast_fp16)[name = string("op_2960_cast_fp16")]; + tensor var_2961_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_273_cast_fp16)[name = string("op_2961_cast_fp16")]; + tensor var_2962_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_275_cast_fp16)[name = string("op_2962_cast_fp16")]; + tensor var_2963_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_277_cast_fp16)[name = string("op_2963_cast_fp16")]; + tensor var_2964_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_279_cast_fp16)[name = string("op_2964_cast_fp16")]; + tensor var_2965_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_281_cast_fp16)[name = string("op_2965_cast_fp16")]; + tensor var_2966_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_283_cast_fp16)[name = string("op_2966_cast_fp16")]; + tensor var_2967_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_285_cast_fp16)[name = string("op_2967_cast_fp16")]; + tensor var_2968_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_287_cast_fp16)[name = string("op_2968_cast_fp16")]; + tensor var_2969_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_289_cast_fp16)[name = string("op_2969_cast_fp16")]; + tensor var_2970_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_291_cast_fp16)[name = string("op_2970_cast_fp16")]; + tensor var_2971_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_293_cast_fp16)[name = string("op_2971_cast_fp16")]; + tensor var_2972_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_295_cast_fp16)[name = string("op_2972_cast_fp16")]; + tensor var_2973_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_297_cast_fp16)[name = string("op_2973_cast_fp16")]; + tensor var_2974_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_299_cast_fp16)[name = string("op_2974_cast_fp16")]; + tensor var_2975_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_301_cast_fp16)[name = string("op_2975_cast_fp16")]; + tensor var_2976_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_303_cast_fp16)[name = string("op_2976_cast_fp16")]; + tensor var_2977_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_305_cast_fp16)[name = string("op_2977_cast_fp16")]; + tensor var_2978_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_307_cast_fp16)[name = string("op_2978_cast_fp16")]; + tensor var_2979_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_309_cast_fp16)[name = string("op_2979_cast_fp16")]; + tensor var_2980_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_311_cast_fp16)[name = string("op_2980_cast_fp16")]; + tensor var_2981_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_313_cast_fp16)[name = string("op_2981_cast_fp16")]; + tensor var_2982_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_315_cast_fp16)[name = string("op_2982_cast_fp16")]; + tensor var_2983_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_317_cast_fp16)[name = string("op_2983_cast_fp16")]; + tensor var_2984_cast_fp16 = softmax(axis = var_1730, x = aw_chunk_319_cast_fp16)[name = string("op_2984_cast_fp16")]; + string var_2986_equation_0 = const()[name = string("op_2986_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2986_cast_fp16 = einsum(equation = var_2986_equation_0, values = (var_2506_cast_fp16, var_2905_cast_fp16))[name = string("op_2986_cast_fp16")]; + string var_2988_equation_0 = const()[name = string("op_2988_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2988_cast_fp16 = einsum(equation = var_2988_equation_0, values = (var_2506_cast_fp16, var_2906_cast_fp16))[name = string("op_2988_cast_fp16")]; + string var_2990_equation_0 = const()[name = string("op_2990_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2990_cast_fp16 = einsum(equation = var_2990_equation_0, values = (var_2506_cast_fp16, var_2907_cast_fp16))[name = string("op_2990_cast_fp16")]; + string var_2992_equation_0 = const()[name = string("op_2992_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2992_cast_fp16 = einsum(equation = var_2992_equation_0, values = (var_2506_cast_fp16, var_2908_cast_fp16))[name = string("op_2992_cast_fp16")]; + string var_2994_equation_0 = const()[name = string("op_2994_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2994_cast_fp16 = einsum(equation = var_2994_equation_0, values = (var_2510_cast_fp16, var_2909_cast_fp16))[name = string("op_2994_cast_fp16")]; + string var_2996_equation_0 = const()[name = string("op_2996_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2996_cast_fp16 = einsum(equation = var_2996_equation_0, values = (var_2510_cast_fp16, var_2910_cast_fp16))[name = string("op_2996_cast_fp16")]; + string var_2998_equation_0 = const()[name = string("op_2998_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2998_cast_fp16 = einsum(equation = var_2998_equation_0, values = (var_2510_cast_fp16, var_2911_cast_fp16))[name = string("op_2998_cast_fp16")]; + string var_3000_equation_0 = const()[name = string("op_3000_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3000_cast_fp16 = einsum(equation = var_3000_equation_0, values = (var_2510_cast_fp16, var_2912_cast_fp16))[name = string("op_3000_cast_fp16")]; + string var_3002_equation_0 = const()[name = string("op_3002_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3002_cast_fp16 = einsum(equation = var_3002_equation_0, values = (var_2514_cast_fp16, var_2913_cast_fp16))[name = string("op_3002_cast_fp16")]; + string var_3004_equation_0 = const()[name = string("op_3004_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3004_cast_fp16 = einsum(equation = var_3004_equation_0, values = (var_2514_cast_fp16, var_2914_cast_fp16))[name = string("op_3004_cast_fp16")]; + string var_3006_equation_0 = const()[name = string("op_3006_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3006_cast_fp16 = einsum(equation = var_3006_equation_0, values = (var_2514_cast_fp16, var_2915_cast_fp16))[name = string("op_3006_cast_fp16")]; + string var_3008_equation_0 = const()[name = string("op_3008_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3008_cast_fp16 = einsum(equation = var_3008_equation_0, values = (var_2514_cast_fp16, var_2916_cast_fp16))[name = string("op_3008_cast_fp16")]; + string var_3010_equation_0 = const()[name = string("op_3010_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3010_cast_fp16 = einsum(equation = var_3010_equation_0, values = (var_2518_cast_fp16, var_2917_cast_fp16))[name = string("op_3010_cast_fp16")]; + string var_3012_equation_0 = const()[name = string("op_3012_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3012_cast_fp16 = einsum(equation = var_3012_equation_0, values = (var_2518_cast_fp16, var_2918_cast_fp16))[name = string("op_3012_cast_fp16")]; + string var_3014_equation_0 = const()[name = string("op_3014_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3014_cast_fp16 = einsum(equation = var_3014_equation_0, values = (var_2518_cast_fp16, var_2919_cast_fp16))[name = string("op_3014_cast_fp16")]; + string var_3016_equation_0 = const()[name = string("op_3016_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3016_cast_fp16 = einsum(equation = var_3016_equation_0, values = (var_2518_cast_fp16, var_2920_cast_fp16))[name = string("op_3016_cast_fp16")]; + string var_3018_equation_0 = const()[name = string("op_3018_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3018_cast_fp16 = einsum(equation = var_3018_equation_0, values = (var_2522_cast_fp16, var_2921_cast_fp16))[name = string("op_3018_cast_fp16")]; + string var_3020_equation_0 = const()[name = string("op_3020_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3020_cast_fp16 = einsum(equation = var_3020_equation_0, values = (var_2522_cast_fp16, var_2922_cast_fp16))[name = string("op_3020_cast_fp16")]; + string var_3022_equation_0 = const()[name = string("op_3022_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3022_cast_fp16 = einsum(equation = var_3022_equation_0, values = (var_2522_cast_fp16, var_2923_cast_fp16))[name = string("op_3022_cast_fp16")]; + string var_3024_equation_0 = const()[name = string("op_3024_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3024_cast_fp16 = einsum(equation = var_3024_equation_0, values = (var_2522_cast_fp16, var_2924_cast_fp16))[name = string("op_3024_cast_fp16")]; + string var_3026_equation_0 = const()[name = string("op_3026_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3026_cast_fp16 = einsum(equation = var_3026_equation_0, values = (var_2526_cast_fp16, var_2925_cast_fp16))[name = string("op_3026_cast_fp16")]; + string var_3028_equation_0 = const()[name = string("op_3028_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3028_cast_fp16 = einsum(equation = var_3028_equation_0, values = (var_2526_cast_fp16, var_2926_cast_fp16))[name = string("op_3028_cast_fp16")]; + string var_3030_equation_0 = const()[name = string("op_3030_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3030_cast_fp16 = einsum(equation = var_3030_equation_0, values = (var_2526_cast_fp16, var_2927_cast_fp16))[name = string("op_3030_cast_fp16")]; + string var_3032_equation_0 = const()[name = string("op_3032_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3032_cast_fp16 = einsum(equation = var_3032_equation_0, values = (var_2526_cast_fp16, var_2928_cast_fp16))[name = string("op_3032_cast_fp16")]; + string var_3034_equation_0 = const()[name = string("op_3034_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3034_cast_fp16 = einsum(equation = var_3034_equation_0, values = (var_2530_cast_fp16, var_2929_cast_fp16))[name = string("op_3034_cast_fp16")]; + string var_3036_equation_0 = const()[name = string("op_3036_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3036_cast_fp16 = einsum(equation = var_3036_equation_0, values = (var_2530_cast_fp16, var_2930_cast_fp16))[name = string("op_3036_cast_fp16")]; + string var_3038_equation_0 = const()[name = string("op_3038_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3038_cast_fp16 = einsum(equation = var_3038_equation_0, values = (var_2530_cast_fp16, var_2931_cast_fp16))[name = string("op_3038_cast_fp16")]; + string var_3040_equation_0 = const()[name = string("op_3040_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3040_cast_fp16 = einsum(equation = var_3040_equation_0, values = (var_2530_cast_fp16, var_2932_cast_fp16))[name = string("op_3040_cast_fp16")]; + string var_3042_equation_0 = const()[name = string("op_3042_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3042_cast_fp16 = einsum(equation = var_3042_equation_0, values = (var_2534_cast_fp16, var_2933_cast_fp16))[name = string("op_3042_cast_fp16")]; + string var_3044_equation_0 = const()[name = string("op_3044_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3044_cast_fp16 = einsum(equation = var_3044_equation_0, values = (var_2534_cast_fp16, var_2934_cast_fp16))[name = string("op_3044_cast_fp16")]; + string var_3046_equation_0 = const()[name = string("op_3046_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3046_cast_fp16 = einsum(equation = var_3046_equation_0, values = (var_2534_cast_fp16, var_2935_cast_fp16))[name = string("op_3046_cast_fp16")]; + string var_3048_equation_0 = const()[name = string("op_3048_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3048_cast_fp16 = einsum(equation = var_3048_equation_0, values = (var_2534_cast_fp16, var_2936_cast_fp16))[name = string("op_3048_cast_fp16")]; + string var_3050_equation_0 = const()[name = string("op_3050_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3050_cast_fp16 = einsum(equation = var_3050_equation_0, values = (var_2538_cast_fp16, var_2937_cast_fp16))[name = string("op_3050_cast_fp16")]; + string var_3052_equation_0 = const()[name = string("op_3052_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3052_cast_fp16 = einsum(equation = var_3052_equation_0, values = (var_2538_cast_fp16, var_2938_cast_fp16))[name = string("op_3052_cast_fp16")]; + string var_3054_equation_0 = const()[name = string("op_3054_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3054_cast_fp16 = einsum(equation = var_3054_equation_0, values = (var_2538_cast_fp16, var_2939_cast_fp16))[name = string("op_3054_cast_fp16")]; + string var_3056_equation_0 = const()[name = string("op_3056_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3056_cast_fp16 = einsum(equation = var_3056_equation_0, values = (var_2538_cast_fp16, var_2940_cast_fp16))[name = string("op_3056_cast_fp16")]; + string var_3058_equation_0 = const()[name = string("op_3058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3058_cast_fp16 = einsum(equation = var_3058_equation_0, values = (var_2542_cast_fp16, var_2941_cast_fp16))[name = string("op_3058_cast_fp16")]; + string var_3060_equation_0 = const()[name = string("op_3060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3060_cast_fp16 = einsum(equation = var_3060_equation_0, values = (var_2542_cast_fp16, var_2942_cast_fp16))[name = string("op_3060_cast_fp16")]; + string var_3062_equation_0 = const()[name = string("op_3062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3062_cast_fp16 = einsum(equation = var_3062_equation_0, values = (var_2542_cast_fp16, var_2943_cast_fp16))[name = string("op_3062_cast_fp16")]; + string var_3064_equation_0 = const()[name = string("op_3064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3064_cast_fp16 = einsum(equation = var_3064_equation_0, values = (var_2542_cast_fp16, var_2944_cast_fp16))[name = string("op_3064_cast_fp16")]; + string var_3066_equation_0 = const()[name = string("op_3066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3066_cast_fp16 = einsum(equation = var_3066_equation_0, values = (var_2546_cast_fp16, var_2945_cast_fp16))[name = string("op_3066_cast_fp16")]; + string var_3068_equation_0 = const()[name = string("op_3068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3068_cast_fp16 = einsum(equation = var_3068_equation_0, values = (var_2546_cast_fp16, var_2946_cast_fp16))[name = string("op_3068_cast_fp16")]; + string var_3070_equation_0 = const()[name = string("op_3070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3070_cast_fp16 = einsum(equation = var_3070_equation_0, values = (var_2546_cast_fp16, var_2947_cast_fp16))[name = string("op_3070_cast_fp16")]; + string var_3072_equation_0 = const()[name = string("op_3072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3072_cast_fp16 = einsum(equation = var_3072_equation_0, values = (var_2546_cast_fp16, var_2948_cast_fp16))[name = string("op_3072_cast_fp16")]; + string var_3074_equation_0 = const()[name = string("op_3074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3074_cast_fp16 = einsum(equation = var_3074_equation_0, values = (var_2550_cast_fp16, var_2949_cast_fp16))[name = string("op_3074_cast_fp16")]; + string var_3076_equation_0 = const()[name = string("op_3076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3076_cast_fp16 = einsum(equation = var_3076_equation_0, values = (var_2550_cast_fp16, var_2950_cast_fp16))[name = string("op_3076_cast_fp16")]; + string var_3078_equation_0 = const()[name = string("op_3078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3078_cast_fp16 = einsum(equation = var_3078_equation_0, values = (var_2550_cast_fp16, var_2951_cast_fp16))[name = string("op_3078_cast_fp16")]; + string var_3080_equation_0 = const()[name = string("op_3080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3080_cast_fp16 = einsum(equation = var_3080_equation_0, values = (var_2550_cast_fp16, var_2952_cast_fp16))[name = string("op_3080_cast_fp16")]; + string var_3082_equation_0 = const()[name = string("op_3082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3082_cast_fp16 = einsum(equation = var_3082_equation_0, values = (var_2554_cast_fp16, var_2953_cast_fp16))[name = string("op_3082_cast_fp16")]; + string var_3084_equation_0 = const()[name = string("op_3084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3084_cast_fp16 = einsum(equation = var_3084_equation_0, values = (var_2554_cast_fp16, var_2954_cast_fp16))[name = string("op_3084_cast_fp16")]; + string var_3086_equation_0 = const()[name = string("op_3086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3086_cast_fp16 = einsum(equation = var_3086_equation_0, values = (var_2554_cast_fp16, var_2955_cast_fp16))[name = string("op_3086_cast_fp16")]; + string var_3088_equation_0 = const()[name = string("op_3088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3088_cast_fp16 = einsum(equation = var_3088_equation_0, values = (var_2554_cast_fp16, var_2956_cast_fp16))[name = string("op_3088_cast_fp16")]; + string var_3090_equation_0 = const()[name = string("op_3090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3090_cast_fp16 = einsum(equation = var_3090_equation_0, values = (var_2558_cast_fp16, var_2957_cast_fp16))[name = string("op_3090_cast_fp16")]; + string var_3092_equation_0 = const()[name = string("op_3092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3092_cast_fp16 = einsum(equation = var_3092_equation_0, values = (var_2558_cast_fp16, var_2958_cast_fp16))[name = string("op_3092_cast_fp16")]; + string var_3094_equation_0 = const()[name = string("op_3094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3094_cast_fp16 = einsum(equation = var_3094_equation_0, values = (var_2558_cast_fp16, var_2959_cast_fp16))[name = string("op_3094_cast_fp16")]; + string var_3096_equation_0 = const()[name = string("op_3096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3096_cast_fp16 = einsum(equation = var_3096_equation_0, values = (var_2558_cast_fp16, var_2960_cast_fp16))[name = string("op_3096_cast_fp16")]; + string var_3098_equation_0 = const()[name = string("op_3098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3098_cast_fp16 = einsum(equation = var_3098_equation_0, values = (var_2562_cast_fp16, var_2961_cast_fp16))[name = string("op_3098_cast_fp16")]; + string var_3100_equation_0 = const()[name = string("op_3100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3100_cast_fp16 = einsum(equation = var_3100_equation_0, values = (var_2562_cast_fp16, var_2962_cast_fp16))[name = string("op_3100_cast_fp16")]; + string var_3102_equation_0 = const()[name = string("op_3102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3102_cast_fp16 = einsum(equation = var_3102_equation_0, values = (var_2562_cast_fp16, var_2963_cast_fp16))[name = string("op_3102_cast_fp16")]; + string var_3104_equation_0 = const()[name = string("op_3104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3104_cast_fp16 = einsum(equation = var_3104_equation_0, values = (var_2562_cast_fp16, var_2964_cast_fp16))[name = string("op_3104_cast_fp16")]; + string var_3106_equation_0 = const()[name = string("op_3106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3106_cast_fp16 = einsum(equation = var_3106_equation_0, values = (var_2566_cast_fp16, var_2965_cast_fp16))[name = string("op_3106_cast_fp16")]; + string var_3108_equation_0 = const()[name = string("op_3108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3108_cast_fp16 = einsum(equation = var_3108_equation_0, values = (var_2566_cast_fp16, var_2966_cast_fp16))[name = string("op_3108_cast_fp16")]; + string var_3110_equation_0 = const()[name = string("op_3110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3110_cast_fp16 = einsum(equation = var_3110_equation_0, values = (var_2566_cast_fp16, var_2967_cast_fp16))[name = string("op_3110_cast_fp16")]; + string var_3112_equation_0 = const()[name = string("op_3112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3112_cast_fp16 = einsum(equation = var_3112_equation_0, values = (var_2566_cast_fp16, var_2968_cast_fp16))[name = string("op_3112_cast_fp16")]; + string var_3114_equation_0 = const()[name = string("op_3114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3114_cast_fp16 = einsum(equation = var_3114_equation_0, values = (var_2570_cast_fp16, var_2969_cast_fp16))[name = string("op_3114_cast_fp16")]; + string var_3116_equation_0 = const()[name = string("op_3116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3116_cast_fp16 = einsum(equation = var_3116_equation_0, values = (var_2570_cast_fp16, var_2970_cast_fp16))[name = string("op_3116_cast_fp16")]; + string var_3118_equation_0 = const()[name = string("op_3118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3118_cast_fp16 = einsum(equation = var_3118_equation_0, values = (var_2570_cast_fp16, var_2971_cast_fp16))[name = string("op_3118_cast_fp16")]; + string var_3120_equation_0 = const()[name = string("op_3120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3120_cast_fp16 = einsum(equation = var_3120_equation_0, values = (var_2570_cast_fp16, var_2972_cast_fp16))[name = string("op_3120_cast_fp16")]; + string var_3122_equation_0 = const()[name = string("op_3122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3122_cast_fp16 = einsum(equation = var_3122_equation_0, values = (var_2574_cast_fp16, var_2973_cast_fp16))[name = string("op_3122_cast_fp16")]; + string var_3124_equation_0 = const()[name = string("op_3124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3124_cast_fp16 = einsum(equation = var_3124_equation_0, values = (var_2574_cast_fp16, var_2974_cast_fp16))[name = string("op_3124_cast_fp16")]; + string var_3126_equation_0 = const()[name = string("op_3126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3126_cast_fp16 = einsum(equation = var_3126_equation_0, values = (var_2574_cast_fp16, var_2975_cast_fp16))[name = string("op_3126_cast_fp16")]; + string var_3128_equation_0 = const()[name = string("op_3128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3128_cast_fp16 = einsum(equation = var_3128_equation_0, values = (var_2574_cast_fp16, var_2976_cast_fp16))[name = string("op_3128_cast_fp16")]; + string var_3130_equation_0 = const()[name = string("op_3130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3130_cast_fp16 = einsum(equation = var_3130_equation_0, values = (var_2578_cast_fp16, var_2977_cast_fp16))[name = string("op_3130_cast_fp16")]; + string var_3132_equation_0 = const()[name = string("op_3132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3132_cast_fp16 = einsum(equation = var_3132_equation_0, values = (var_2578_cast_fp16, var_2978_cast_fp16))[name = string("op_3132_cast_fp16")]; + string var_3134_equation_0 = const()[name = string("op_3134_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3134_cast_fp16 = einsum(equation = var_3134_equation_0, values = (var_2578_cast_fp16, var_2979_cast_fp16))[name = string("op_3134_cast_fp16")]; + string var_3136_equation_0 = const()[name = string("op_3136_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3136_cast_fp16 = einsum(equation = var_3136_equation_0, values = (var_2578_cast_fp16, var_2980_cast_fp16))[name = string("op_3136_cast_fp16")]; + string var_3138_equation_0 = const()[name = string("op_3138_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3138_cast_fp16 = einsum(equation = var_3138_equation_0, values = (var_2582_cast_fp16, var_2981_cast_fp16))[name = string("op_3138_cast_fp16")]; + string var_3140_equation_0 = const()[name = string("op_3140_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3140_cast_fp16 = einsum(equation = var_3140_equation_0, values = (var_2582_cast_fp16, var_2982_cast_fp16))[name = string("op_3140_cast_fp16")]; + string var_3142_equation_0 = const()[name = string("op_3142_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3142_cast_fp16 = einsum(equation = var_3142_equation_0, values = (var_2582_cast_fp16, var_2983_cast_fp16))[name = string("op_3142_cast_fp16")]; + string var_3144_equation_0 = const()[name = string("op_3144_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3144_cast_fp16 = einsum(equation = var_3144_equation_0, values = (var_2582_cast_fp16, var_2984_cast_fp16))[name = string("op_3144_cast_fp16")]; + bool var_3146_interleave_0 = const()[name = string("op_3146_interleave_0"), val = bool(false)]; + tensor var_3146_cast_fp16 = concat(axis = var_1705, interleave = var_3146_interleave_0, values = (var_2986_cast_fp16, var_2988_cast_fp16, var_2990_cast_fp16, var_2992_cast_fp16))[name = string("op_3146_cast_fp16")]; + bool var_3148_interleave_0 = const()[name = string("op_3148_interleave_0"), val = bool(false)]; + tensor var_3148_cast_fp16 = concat(axis = var_1705, interleave = var_3148_interleave_0, values = (var_2994_cast_fp16, var_2996_cast_fp16, var_2998_cast_fp16, var_3000_cast_fp16))[name = string("op_3148_cast_fp16")]; + bool var_3150_interleave_0 = const()[name = string("op_3150_interleave_0"), val = bool(false)]; + tensor var_3150_cast_fp16 = concat(axis = var_1705, interleave = var_3150_interleave_0, values = (var_3002_cast_fp16, var_3004_cast_fp16, var_3006_cast_fp16, var_3008_cast_fp16))[name = string("op_3150_cast_fp16")]; + bool var_3152_interleave_0 = const()[name = string("op_3152_interleave_0"), val = bool(false)]; + tensor var_3152_cast_fp16 = concat(axis = var_1705, interleave = var_3152_interleave_0, values = (var_3010_cast_fp16, var_3012_cast_fp16, var_3014_cast_fp16, var_3016_cast_fp16))[name = string("op_3152_cast_fp16")]; + bool var_3154_interleave_0 = const()[name = string("op_3154_interleave_0"), val = bool(false)]; + tensor var_3154_cast_fp16 = concat(axis = var_1705, interleave = var_3154_interleave_0, values = (var_3018_cast_fp16, var_3020_cast_fp16, var_3022_cast_fp16, var_3024_cast_fp16))[name = string("op_3154_cast_fp16")]; + bool var_3156_interleave_0 = const()[name = string("op_3156_interleave_0"), val = bool(false)]; + tensor var_3156_cast_fp16 = concat(axis = var_1705, interleave = var_3156_interleave_0, values = (var_3026_cast_fp16, var_3028_cast_fp16, var_3030_cast_fp16, var_3032_cast_fp16))[name = string("op_3156_cast_fp16")]; + bool var_3158_interleave_0 = const()[name = string("op_3158_interleave_0"), val = bool(false)]; + tensor var_3158_cast_fp16 = concat(axis = var_1705, interleave = var_3158_interleave_0, values = (var_3034_cast_fp16, var_3036_cast_fp16, var_3038_cast_fp16, var_3040_cast_fp16))[name = string("op_3158_cast_fp16")]; + bool var_3160_interleave_0 = const()[name = string("op_3160_interleave_0"), val = bool(false)]; + tensor var_3160_cast_fp16 = concat(axis = var_1705, interleave = var_3160_interleave_0, values = (var_3042_cast_fp16, var_3044_cast_fp16, var_3046_cast_fp16, var_3048_cast_fp16))[name = string("op_3160_cast_fp16")]; + bool var_3162_interleave_0 = const()[name = string("op_3162_interleave_0"), val = bool(false)]; + tensor var_3162_cast_fp16 = concat(axis = var_1705, interleave = var_3162_interleave_0, values = (var_3050_cast_fp16, var_3052_cast_fp16, var_3054_cast_fp16, var_3056_cast_fp16))[name = string("op_3162_cast_fp16")]; + bool var_3164_interleave_0 = const()[name = string("op_3164_interleave_0"), val = bool(false)]; + tensor var_3164_cast_fp16 = concat(axis = var_1705, interleave = var_3164_interleave_0, values = (var_3058_cast_fp16, var_3060_cast_fp16, var_3062_cast_fp16, var_3064_cast_fp16))[name = string("op_3164_cast_fp16")]; + bool var_3166_interleave_0 = const()[name = string("op_3166_interleave_0"), val = bool(false)]; + tensor var_3166_cast_fp16 = concat(axis = var_1705, interleave = var_3166_interleave_0, values = (var_3066_cast_fp16, var_3068_cast_fp16, var_3070_cast_fp16, var_3072_cast_fp16))[name = string("op_3166_cast_fp16")]; + bool var_3168_interleave_0 = const()[name = string("op_3168_interleave_0"), val = bool(false)]; + tensor var_3168_cast_fp16 = concat(axis = var_1705, interleave = var_3168_interleave_0, values = (var_3074_cast_fp16, var_3076_cast_fp16, var_3078_cast_fp16, var_3080_cast_fp16))[name = string("op_3168_cast_fp16")]; + bool var_3170_interleave_0 = const()[name = string("op_3170_interleave_0"), val = bool(false)]; + tensor var_3170_cast_fp16 = concat(axis = var_1705, interleave = var_3170_interleave_0, values = (var_3082_cast_fp16, var_3084_cast_fp16, var_3086_cast_fp16, var_3088_cast_fp16))[name = string("op_3170_cast_fp16")]; + bool var_3172_interleave_0 = const()[name = string("op_3172_interleave_0"), val = bool(false)]; + tensor var_3172_cast_fp16 = concat(axis = var_1705, interleave = var_3172_interleave_0, values = (var_3090_cast_fp16, var_3092_cast_fp16, var_3094_cast_fp16, var_3096_cast_fp16))[name = string("op_3172_cast_fp16")]; + bool var_3174_interleave_0 = const()[name = string("op_3174_interleave_0"), val = bool(false)]; + tensor var_3174_cast_fp16 = concat(axis = var_1705, interleave = var_3174_interleave_0, values = (var_3098_cast_fp16, var_3100_cast_fp16, var_3102_cast_fp16, var_3104_cast_fp16))[name = string("op_3174_cast_fp16")]; + bool var_3176_interleave_0 = const()[name = string("op_3176_interleave_0"), val = bool(false)]; + tensor var_3176_cast_fp16 = concat(axis = var_1705, interleave = var_3176_interleave_0, values = (var_3106_cast_fp16, var_3108_cast_fp16, var_3110_cast_fp16, var_3112_cast_fp16))[name = string("op_3176_cast_fp16")]; + bool var_3178_interleave_0 = const()[name = string("op_3178_interleave_0"), val = bool(false)]; + tensor var_3178_cast_fp16 = concat(axis = var_1705, interleave = var_3178_interleave_0, values = (var_3114_cast_fp16, var_3116_cast_fp16, var_3118_cast_fp16, var_3120_cast_fp16))[name = string("op_3178_cast_fp16")]; + bool var_3180_interleave_0 = const()[name = string("op_3180_interleave_0"), val = bool(false)]; + tensor var_3180_cast_fp16 = concat(axis = var_1705, interleave = var_3180_interleave_0, values = (var_3122_cast_fp16, var_3124_cast_fp16, var_3126_cast_fp16, var_3128_cast_fp16))[name = string("op_3180_cast_fp16")]; + bool var_3182_interleave_0 = const()[name = string("op_3182_interleave_0"), val = bool(false)]; + tensor var_3182_cast_fp16 = concat(axis = var_1705, interleave = var_3182_interleave_0, values = (var_3130_cast_fp16, var_3132_cast_fp16, var_3134_cast_fp16, var_3136_cast_fp16))[name = string("op_3182_cast_fp16")]; + bool var_3184_interleave_0 = const()[name = string("op_3184_interleave_0"), val = bool(false)]; + tensor var_3184_cast_fp16 = concat(axis = var_1705, interleave = var_3184_interleave_0, values = (var_3138_cast_fp16, var_3140_cast_fp16, var_3142_cast_fp16, var_3144_cast_fp16))[name = string("op_3184_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_1730, interleave = input_9_interleave_0, values = (var_3146_cast_fp16, var_3148_cast_fp16, var_3150_cast_fp16, var_3152_cast_fp16, var_3154_cast_fp16, var_3156_cast_fp16, var_3158_cast_fp16, var_3160_cast_fp16, var_3162_cast_fp16, var_3164_cast_fp16, var_3166_cast_fp16, var_3168_cast_fp16, var_3170_cast_fp16, var_3172_cast_fp16, var_3174_cast_fp16, var_3176_cast_fp16, var_3178_cast_fp16, var_3180_cast_fp16, var_3182_cast_fp16, var_3184_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63858560)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67135424)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_3203_to_fp16 = const()[name = string("op_3203_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_3203_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67138048)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67140672)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67143296)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80250560)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80260864)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93368128)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_3232 = const()[name = string("op_3232"), val = int32(3)]; + int32 var_3257 = const()[name = string("op_3257"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_3274_to_fp16 = const()[name = string("op_3274_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_3274_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93370752)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93373376)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93376000)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96652864)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96655488)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99932352)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103209216)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_3312_begin_0 = const()[name = string("op_3312_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3312_end_0 = const()[name = string("op_3312_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3312_end_mask_0 = const()[name = string("op_3312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3312_cast_fp16 = slice_by_index(begin = var_3312_begin_0, end = var_3312_end_0, end_mask = var_3312_end_mask_0, x = query_5_cast_fp16)[name = string("op_3312_cast_fp16")]; + tensor var_3316_begin_0 = const()[name = string("op_3316_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3316_end_0 = const()[name = string("op_3316_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3316_end_mask_0 = const()[name = string("op_3316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3316_cast_fp16 = slice_by_index(begin = var_3316_begin_0, end = var_3316_end_0, end_mask = var_3316_end_mask_0, x = query_5_cast_fp16)[name = string("op_3316_cast_fp16")]; + tensor var_3320_begin_0 = const()[name = string("op_3320_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3320_end_0 = const()[name = string("op_3320_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3320_end_mask_0 = const()[name = string("op_3320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3320_cast_fp16 = slice_by_index(begin = var_3320_begin_0, end = var_3320_end_0, end_mask = var_3320_end_mask_0, x = query_5_cast_fp16)[name = string("op_3320_cast_fp16")]; + tensor var_3324_begin_0 = const()[name = string("op_3324_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3324_end_0 = const()[name = string("op_3324_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3324_end_mask_0 = const()[name = string("op_3324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3324_cast_fp16 = slice_by_index(begin = var_3324_begin_0, end = var_3324_end_0, end_mask = var_3324_end_mask_0, x = query_5_cast_fp16)[name = string("op_3324_cast_fp16")]; + tensor var_3328_begin_0 = const()[name = string("op_3328_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3328_end_0 = const()[name = string("op_3328_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3328_end_mask_0 = const()[name = string("op_3328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3328_cast_fp16 = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = query_5_cast_fp16)[name = string("op_3328_cast_fp16")]; + tensor var_3332_begin_0 = const()[name = string("op_3332_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3332_end_0 = const()[name = string("op_3332_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3332_end_mask_0 = const()[name = string("op_3332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3332_cast_fp16 = slice_by_index(begin = var_3332_begin_0, end = var_3332_end_0, end_mask = var_3332_end_mask_0, x = query_5_cast_fp16)[name = string("op_3332_cast_fp16")]; + tensor var_3336_begin_0 = const()[name = string("op_3336_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3336_end_0 = const()[name = string("op_3336_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3336_end_mask_0 = const()[name = string("op_3336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3336_cast_fp16 = slice_by_index(begin = var_3336_begin_0, end = var_3336_end_0, end_mask = var_3336_end_mask_0, x = query_5_cast_fp16)[name = string("op_3336_cast_fp16")]; + tensor var_3340_begin_0 = const()[name = string("op_3340_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3340_end_0 = const()[name = string("op_3340_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3340_end_mask_0 = const()[name = string("op_3340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3340_cast_fp16 = slice_by_index(begin = var_3340_begin_0, end = var_3340_end_0, end_mask = var_3340_end_mask_0, x = query_5_cast_fp16)[name = string("op_3340_cast_fp16")]; + tensor var_3344_begin_0 = const()[name = string("op_3344_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3344_end_0 = const()[name = string("op_3344_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3344_end_mask_0 = const()[name = string("op_3344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3344_cast_fp16 = slice_by_index(begin = var_3344_begin_0, end = var_3344_end_0, end_mask = var_3344_end_mask_0, x = query_5_cast_fp16)[name = string("op_3344_cast_fp16")]; + tensor var_3348_begin_0 = const()[name = string("op_3348_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3348_end_0 = const()[name = string("op_3348_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3348_end_mask_0 = const()[name = string("op_3348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3348_cast_fp16 = slice_by_index(begin = var_3348_begin_0, end = var_3348_end_0, end_mask = var_3348_end_mask_0, x = query_5_cast_fp16)[name = string("op_3348_cast_fp16")]; + tensor var_3352_begin_0 = const()[name = string("op_3352_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3352_end_0 = const()[name = string("op_3352_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3352_end_mask_0 = const()[name = string("op_3352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3352_cast_fp16 = slice_by_index(begin = var_3352_begin_0, end = var_3352_end_0, end_mask = var_3352_end_mask_0, x = query_5_cast_fp16)[name = string("op_3352_cast_fp16")]; + tensor var_3356_begin_0 = const()[name = string("op_3356_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3356_end_0 = const()[name = string("op_3356_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3356_end_mask_0 = const()[name = string("op_3356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3356_cast_fp16 = slice_by_index(begin = var_3356_begin_0, end = var_3356_end_0, end_mask = var_3356_end_mask_0, x = query_5_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3360_begin_0 = const()[name = string("op_3360_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3360_end_0 = const()[name = string("op_3360_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_3360_end_mask_0 = const()[name = string("op_3360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3360_cast_fp16 = slice_by_index(begin = var_3360_begin_0, end = var_3360_end_0, end_mask = var_3360_end_mask_0, x = query_5_cast_fp16)[name = string("op_3360_cast_fp16")]; + tensor var_3364_begin_0 = const()[name = string("op_3364_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3364_end_0 = const()[name = string("op_3364_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_3364_end_mask_0 = const()[name = string("op_3364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3364_cast_fp16 = slice_by_index(begin = var_3364_begin_0, end = var_3364_end_0, end_mask = var_3364_end_mask_0, x = query_5_cast_fp16)[name = string("op_3364_cast_fp16")]; + tensor var_3368_begin_0 = const()[name = string("op_3368_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3368_end_0 = const()[name = string("op_3368_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_3368_end_mask_0 = const()[name = string("op_3368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3368_cast_fp16 = slice_by_index(begin = var_3368_begin_0, end = var_3368_end_0, end_mask = var_3368_end_mask_0, x = query_5_cast_fp16)[name = string("op_3368_cast_fp16")]; + tensor var_3372_begin_0 = const()[name = string("op_3372_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3372_end_0 = const()[name = string("op_3372_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_3372_end_mask_0 = const()[name = string("op_3372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3372_cast_fp16 = slice_by_index(begin = var_3372_begin_0, end = var_3372_end_0, end_mask = var_3372_end_mask_0, x = query_5_cast_fp16)[name = string("op_3372_cast_fp16")]; + tensor var_3376_begin_0 = const()[name = string("op_3376_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3376_end_0 = const()[name = string("op_3376_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_3376_end_mask_0 = const()[name = string("op_3376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3376_cast_fp16 = slice_by_index(begin = var_3376_begin_0, end = var_3376_end_0, end_mask = var_3376_end_mask_0, x = query_5_cast_fp16)[name = string("op_3376_cast_fp16")]; + tensor var_3380_begin_0 = const()[name = string("op_3380_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3380_end_0 = const()[name = string("op_3380_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_3380_end_mask_0 = const()[name = string("op_3380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3380_cast_fp16 = slice_by_index(begin = var_3380_begin_0, end = var_3380_end_0, end_mask = var_3380_end_mask_0, x = query_5_cast_fp16)[name = string("op_3380_cast_fp16")]; + tensor var_3384_begin_0 = const()[name = string("op_3384_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3384_end_0 = const()[name = string("op_3384_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_3384_end_mask_0 = const()[name = string("op_3384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3384_cast_fp16 = slice_by_index(begin = var_3384_begin_0, end = var_3384_end_0, end_mask = var_3384_end_mask_0, x = query_5_cast_fp16)[name = string("op_3384_cast_fp16")]; + tensor var_3388_begin_0 = const()[name = string("op_3388_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3388_end_0 = const()[name = string("op_3388_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_3388_end_mask_0 = const()[name = string("op_3388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3388_cast_fp16 = slice_by_index(begin = var_3388_begin_0, end = var_3388_end_0, end_mask = var_3388_end_mask_0, x = query_5_cast_fp16)[name = string("op_3388_cast_fp16")]; + tensor var_3397_begin_0 = const()[name = string("op_3397_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3397_end_0 = const()[name = string("op_3397_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3397_end_mask_0 = const()[name = string("op_3397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3397_cast_fp16 = slice_by_index(begin = var_3397_begin_0, end = var_3397_end_0, end_mask = var_3397_end_mask_0, x = var_3312_cast_fp16)[name = string("op_3397_cast_fp16")]; + tensor var_3404_begin_0 = const()[name = string("op_3404_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3404_end_0 = const()[name = string("op_3404_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3404_end_mask_0 = const()[name = string("op_3404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3404_cast_fp16 = slice_by_index(begin = var_3404_begin_0, end = var_3404_end_0, end_mask = var_3404_end_mask_0, x = var_3312_cast_fp16)[name = string("op_3404_cast_fp16")]; + tensor var_3411_begin_0 = const()[name = string("op_3411_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3411_end_0 = const()[name = string("op_3411_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3411_end_mask_0 = const()[name = string("op_3411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3411_cast_fp16 = slice_by_index(begin = var_3411_begin_0, end = var_3411_end_0, end_mask = var_3411_end_mask_0, x = var_3312_cast_fp16)[name = string("op_3411_cast_fp16")]; + tensor var_3418_begin_0 = const()[name = string("op_3418_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3418_end_0 = const()[name = string("op_3418_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3418_end_mask_0 = const()[name = string("op_3418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3418_cast_fp16 = slice_by_index(begin = var_3418_begin_0, end = var_3418_end_0, end_mask = var_3418_end_mask_0, x = var_3312_cast_fp16)[name = string("op_3418_cast_fp16")]; + tensor var_3425_begin_0 = const()[name = string("op_3425_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3425_end_0 = const()[name = string("op_3425_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3425_end_mask_0 = const()[name = string("op_3425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3425_cast_fp16 = slice_by_index(begin = var_3425_begin_0, end = var_3425_end_0, end_mask = var_3425_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3425_cast_fp16")]; + tensor var_3432_begin_0 = const()[name = string("op_3432_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3432_end_0 = const()[name = string("op_3432_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3432_end_mask_0 = const()[name = string("op_3432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3432_cast_fp16 = slice_by_index(begin = var_3432_begin_0, end = var_3432_end_0, end_mask = var_3432_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3432_cast_fp16")]; + tensor var_3439_begin_0 = const()[name = string("op_3439_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3439_end_0 = const()[name = string("op_3439_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3439_end_mask_0 = const()[name = string("op_3439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3439_cast_fp16 = slice_by_index(begin = var_3439_begin_0, end = var_3439_end_0, end_mask = var_3439_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3439_cast_fp16")]; + tensor var_3446_begin_0 = const()[name = string("op_3446_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3446_end_0 = const()[name = string("op_3446_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3446_end_mask_0 = const()[name = string("op_3446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3446_cast_fp16 = slice_by_index(begin = var_3446_begin_0, end = var_3446_end_0, end_mask = var_3446_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3446_cast_fp16")]; + tensor var_3453_begin_0 = const()[name = string("op_3453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3453_end_0 = const()[name = string("op_3453_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3453_end_mask_0 = const()[name = string("op_3453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3453_cast_fp16 = slice_by_index(begin = var_3453_begin_0, end = var_3453_end_0, end_mask = var_3453_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3453_cast_fp16")]; + tensor var_3460_begin_0 = const()[name = string("op_3460_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3460_end_0 = const()[name = string("op_3460_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3460_end_mask_0 = const()[name = string("op_3460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3460_cast_fp16 = slice_by_index(begin = var_3460_begin_0, end = var_3460_end_0, end_mask = var_3460_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3460_cast_fp16")]; + tensor var_3467_begin_0 = const()[name = string("op_3467_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3467_end_0 = const()[name = string("op_3467_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3467_end_mask_0 = const()[name = string("op_3467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3467_cast_fp16 = slice_by_index(begin = var_3467_begin_0, end = var_3467_end_0, end_mask = var_3467_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3467_cast_fp16")]; + tensor var_3474_begin_0 = const()[name = string("op_3474_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3474_end_0 = const()[name = string("op_3474_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3474_end_mask_0 = const()[name = string("op_3474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3474_cast_fp16 = slice_by_index(begin = var_3474_begin_0, end = var_3474_end_0, end_mask = var_3474_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3474_cast_fp16")]; + tensor var_3481_begin_0 = const()[name = string("op_3481_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3481_end_0 = const()[name = string("op_3481_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3481_end_mask_0 = const()[name = string("op_3481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3481_cast_fp16 = slice_by_index(begin = var_3481_begin_0, end = var_3481_end_0, end_mask = var_3481_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3481_cast_fp16")]; + tensor var_3488_begin_0 = const()[name = string("op_3488_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3488_end_0 = const()[name = string("op_3488_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3488_end_mask_0 = const()[name = string("op_3488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3488_cast_fp16 = slice_by_index(begin = var_3488_begin_0, end = var_3488_end_0, end_mask = var_3488_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3488_cast_fp16")]; + tensor var_3495_begin_0 = const()[name = string("op_3495_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3495_end_0 = const()[name = string("op_3495_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3495_end_mask_0 = const()[name = string("op_3495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3495_cast_fp16 = slice_by_index(begin = var_3495_begin_0, end = var_3495_end_0, end_mask = var_3495_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3495_cast_fp16")]; + tensor var_3502_begin_0 = const()[name = string("op_3502_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3502_end_0 = const()[name = string("op_3502_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3502_end_mask_0 = const()[name = string("op_3502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3502_cast_fp16 = slice_by_index(begin = var_3502_begin_0, end = var_3502_end_0, end_mask = var_3502_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3502_cast_fp16")]; + tensor var_3509_begin_0 = const()[name = string("op_3509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3509_end_0 = const()[name = string("op_3509_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3509_end_mask_0 = const()[name = string("op_3509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3509_cast_fp16 = slice_by_index(begin = var_3509_begin_0, end = var_3509_end_0, end_mask = var_3509_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3509_cast_fp16")]; + tensor var_3516_begin_0 = const()[name = string("op_3516_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3516_end_0 = const()[name = string("op_3516_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3516_end_mask_0 = const()[name = string("op_3516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3516_cast_fp16 = slice_by_index(begin = var_3516_begin_0, end = var_3516_end_0, end_mask = var_3516_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3516_cast_fp16")]; + tensor var_3523_begin_0 = const()[name = string("op_3523_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3523_end_0 = const()[name = string("op_3523_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3523_end_mask_0 = const()[name = string("op_3523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3523_cast_fp16 = slice_by_index(begin = var_3523_begin_0, end = var_3523_end_0, end_mask = var_3523_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3523_cast_fp16")]; + tensor var_3530_begin_0 = const()[name = string("op_3530_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3530_end_0 = const()[name = string("op_3530_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3530_end_mask_0 = const()[name = string("op_3530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3530_cast_fp16 = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3530_cast_fp16")]; + tensor var_3537_begin_0 = const()[name = string("op_3537_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3537_end_0 = const()[name = string("op_3537_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3537_end_mask_0 = const()[name = string("op_3537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3537_cast_fp16 = slice_by_index(begin = var_3537_begin_0, end = var_3537_end_0, end_mask = var_3537_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3537_cast_fp16")]; + tensor var_3544_begin_0 = const()[name = string("op_3544_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3544_end_0 = const()[name = string("op_3544_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3544_end_mask_0 = const()[name = string("op_3544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3544_cast_fp16 = slice_by_index(begin = var_3544_begin_0, end = var_3544_end_0, end_mask = var_3544_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3544_cast_fp16")]; + tensor var_3551_begin_0 = const()[name = string("op_3551_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3551_end_0 = const()[name = string("op_3551_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3551_end_mask_0 = const()[name = string("op_3551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3551_cast_fp16 = slice_by_index(begin = var_3551_begin_0, end = var_3551_end_0, end_mask = var_3551_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3551_cast_fp16")]; + tensor var_3558_begin_0 = const()[name = string("op_3558_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3558_end_0 = const()[name = string("op_3558_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3558_end_mask_0 = const()[name = string("op_3558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3558_cast_fp16 = slice_by_index(begin = var_3558_begin_0, end = var_3558_end_0, end_mask = var_3558_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3558_cast_fp16")]; + tensor var_3565_begin_0 = const()[name = string("op_3565_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3565_end_0 = const()[name = string("op_3565_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3565_end_mask_0 = const()[name = string("op_3565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3565_cast_fp16 = slice_by_index(begin = var_3565_begin_0, end = var_3565_end_0, end_mask = var_3565_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3565_cast_fp16")]; + tensor var_3572_begin_0 = const()[name = string("op_3572_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3572_end_0 = const()[name = string("op_3572_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3572_end_mask_0 = const()[name = string("op_3572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3572_cast_fp16 = slice_by_index(begin = var_3572_begin_0, end = var_3572_end_0, end_mask = var_3572_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3572_cast_fp16")]; + tensor var_3579_begin_0 = const()[name = string("op_3579_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3579_end_0 = const()[name = string("op_3579_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3579_end_mask_0 = const()[name = string("op_3579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3579_cast_fp16 = slice_by_index(begin = var_3579_begin_0, end = var_3579_end_0, end_mask = var_3579_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3579_cast_fp16")]; + tensor var_3586_begin_0 = const()[name = string("op_3586_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3586_end_0 = const()[name = string("op_3586_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3586_end_mask_0 = const()[name = string("op_3586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3586_cast_fp16 = slice_by_index(begin = var_3586_begin_0, end = var_3586_end_0, end_mask = var_3586_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3586_cast_fp16")]; + tensor var_3593_begin_0 = const()[name = string("op_3593_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3593_end_0 = const()[name = string("op_3593_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3593_end_mask_0 = const()[name = string("op_3593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3593_cast_fp16 = slice_by_index(begin = var_3593_begin_0, end = var_3593_end_0, end_mask = var_3593_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3593_cast_fp16")]; + tensor var_3600_begin_0 = const()[name = string("op_3600_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3600_end_0 = const()[name = string("op_3600_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3600_end_mask_0 = const()[name = string("op_3600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3600_cast_fp16 = slice_by_index(begin = var_3600_begin_0, end = var_3600_end_0, end_mask = var_3600_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3600_cast_fp16")]; + tensor var_3607_begin_0 = const()[name = string("op_3607_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3607_end_0 = const()[name = string("op_3607_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3607_end_mask_0 = const()[name = string("op_3607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3607_cast_fp16 = slice_by_index(begin = var_3607_begin_0, end = var_3607_end_0, end_mask = var_3607_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3607_cast_fp16")]; + tensor var_3614_begin_0 = const()[name = string("op_3614_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3614_end_0 = const()[name = string("op_3614_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3614_end_mask_0 = const()[name = string("op_3614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3614_cast_fp16 = slice_by_index(begin = var_3614_begin_0, end = var_3614_end_0, end_mask = var_3614_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3614_cast_fp16")]; + tensor var_3621_begin_0 = const()[name = string("op_3621_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3621_end_0 = const()[name = string("op_3621_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3621_end_mask_0 = const()[name = string("op_3621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3621_cast_fp16 = slice_by_index(begin = var_3621_begin_0, end = var_3621_end_0, end_mask = var_3621_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3621_cast_fp16")]; + tensor var_3628_begin_0 = const()[name = string("op_3628_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3628_end_0 = const()[name = string("op_3628_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3628_end_mask_0 = const()[name = string("op_3628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3628_cast_fp16 = slice_by_index(begin = var_3628_begin_0, end = var_3628_end_0, end_mask = var_3628_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3628_cast_fp16")]; + tensor var_3635_begin_0 = const()[name = string("op_3635_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3635_end_0 = const()[name = string("op_3635_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3635_end_mask_0 = const()[name = string("op_3635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3635_cast_fp16 = slice_by_index(begin = var_3635_begin_0, end = var_3635_end_0, end_mask = var_3635_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3635_cast_fp16")]; + tensor var_3642_begin_0 = const()[name = string("op_3642_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3642_end_0 = const()[name = string("op_3642_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3642_end_mask_0 = const()[name = string("op_3642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3642_cast_fp16 = slice_by_index(begin = var_3642_begin_0, end = var_3642_end_0, end_mask = var_3642_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3642_cast_fp16")]; + tensor var_3649_begin_0 = const()[name = string("op_3649_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3649_end_0 = const()[name = string("op_3649_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3649_end_mask_0 = const()[name = string("op_3649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3649_cast_fp16 = slice_by_index(begin = var_3649_begin_0, end = var_3649_end_0, end_mask = var_3649_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3649_cast_fp16")]; + tensor var_3656_begin_0 = const()[name = string("op_3656_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3656_end_0 = const()[name = string("op_3656_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3656_end_mask_0 = const()[name = string("op_3656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3656_cast_fp16 = slice_by_index(begin = var_3656_begin_0, end = var_3656_end_0, end_mask = var_3656_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3656_cast_fp16")]; + tensor var_3663_begin_0 = const()[name = string("op_3663_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3663_end_0 = const()[name = string("op_3663_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3663_end_mask_0 = const()[name = string("op_3663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3663_cast_fp16 = slice_by_index(begin = var_3663_begin_0, end = var_3663_end_0, end_mask = var_3663_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3663_cast_fp16")]; + tensor var_3670_begin_0 = const()[name = string("op_3670_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3670_end_0 = const()[name = string("op_3670_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3670_end_mask_0 = const()[name = string("op_3670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3670_cast_fp16 = slice_by_index(begin = var_3670_begin_0, end = var_3670_end_0, end_mask = var_3670_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3670_cast_fp16")]; + tensor var_3677_begin_0 = const()[name = string("op_3677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3677_end_0 = const()[name = string("op_3677_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3677_end_mask_0 = const()[name = string("op_3677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3677_cast_fp16 = slice_by_index(begin = var_3677_begin_0, end = var_3677_end_0, end_mask = var_3677_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3677_cast_fp16")]; + tensor var_3684_begin_0 = const()[name = string("op_3684_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3684_end_0 = const()[name = string("op_3684_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3684_end_mask_0 = const()[name = string("op_3684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3684_cast_fp16 = slice_by_index(begin = var_3684_begin_0, end = var_3684_end_0, end_mask = var_3684_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3684_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = string("op_3691_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3691_end_0 = const()[name = string("op_3691_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3691_end_mask_0 = const()[name = string("op_3691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3691_cast_fp16")]; + tensor var_3698_begin_0 = const()[name = string("op_3698_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3698_end_0 = const()[name = string("op_3698_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3698_end_mask_0 = const()[name = string("op_3698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3698_cast_fp16 = slice_by_index(begin = var_3698_begin_0, end = var_3698_end_0, end_mask = var_3698_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3698_cast_fp16")]; + tensor var_3705_begin_0 = const()[name = string("op_3705_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3705_end_0 = const()[name = string("op_3705_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3705_end_mask_0 = const()[name = string("op_3705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3705_cast_fp16 = slice_by_index(begin = var_3705_begin_0, end = var_3705_end_0, end_mask = var_3705_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3705_cast_fp16")]; + tensor var_3712_begin_0 = const()[name = string("op_3712_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3712_end_0 = const()[name = string("op_3712_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3712_end_mask_0 = const()[name = string("op_3712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3712_cast_fp16 = slice_by_index(begin = var_3712_begin_0, end = var_3712_end_0, end_mask = var_3712_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3712_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = string("op_3719_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3719_end_0 = const()[name = string("op_3719_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3719_end_mask_0 = const()[name = string("op_3719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3719_cast_fp16")]; + tensor var_3726_begin_0 = const()[name = string("op_3726_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3726_end_0 = const()[name = string("op_3726_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3726_end_mask_0 = const()[name = string("op_3726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3726_cast_fp16 = slice_by_index(begin = var_3726_begin_0, end = var_3726_end_0, end_mask = var_3726_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3726_cast_fp16")]; + tensor var_3733_begin_0 = const()[name = string("op_3733_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3733_end_0 = const()[name = string("op_3733_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3733_end_mask_0 = const()[name = string("op_3733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3733_cast_fp16 = slice_by_index(begin = var_3733_begin_0, end = var_3733_end_0, end_mask = var_3733_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3733_cast_fp16")]; + tensor var_3740_begin_0 = const()[name = string("op_3740_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3740_end_0 = const()[name = string("op_3740_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3740_end_mask_0 = const()[name = string("op_3740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3740_cast_fp16 = slice_by_index(begin = var_3740_begin_0, end = var_3740_end_0, end_mask = var_3740_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3740_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = string("op_3747_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3747_end_0 = const()[name = string("op_3747_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3747_end_mask_0 = const()[name = string("op_3747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3747_cast_fp16")]; + tensor var_3754_begin_0 = const()[name = string("op_3754_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3754_end_0 = const()[name = string("op_3754_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3754_end_mask_0 = const()[name = string("op_3754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3754_cast_fp16 = slice_by_index(begin = var_3754_begin_0, end = var_3754_end_0, end_mask = var_3754_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3754_cast_fp16")]; + tensor var_3761_begin_0 = const()[name = string("op_3761_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3761_end_0 = const()[name = string("op_3761_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3761_end_mask_0 = const()[name = string("op_3761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3761_cast_fp16 = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3761_cast_fp16")]; + tensor var_3768_begin_0 = const()[name = string("op_3768_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3768_end_0 = const()[name = string("op_3768_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3768_end_mask_0 = const()[name = string("op_3768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3768_cast_fp16 = slice_by_index(begin = var_3768_begin_0, end = var_3768_end_0, end_mask = var_3768_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3768_cast_fp16")]; + tensor var_3775_begin_0 = const()[name = string("op_3775_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3775_end_0 = const()[name = string("op_3775_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3775_end_mask_0 = const()[name = string("op_3775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3775_cast_fp16 = slice_by_index(begin = var_3775_begin_0, end = var_3775_end_0, end_mask = var_3775_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3775_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = string("op_3782_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3782_end_0 = const()[name = string("op_3782_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3782_end_mask_0 = const()[name = string("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3782_cast_fp16")]; + tensor var_3789_begin_0 = const()[name = string("op_3789_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3789_end_0 = const()[name = string("op_3789_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3789_end_mask_0 = const()[name = string("op_3789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3789_cast_fp16 = slice_by_index(begin = var_3789_begin_0, end = var_3789_end_0, end_mask = var_3789_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3789_cast_fp16")]; + tensor var_3796_begin_0 = const()[name = string("op_3796_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3796_end_0 = const()[name = string("op_3796_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3796_end_mask_0 = const()[name = string("op_3796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3796_cast_fp16 = slice_by_index(begin = var_3796_begin_0, end = var_3796_end_0, end_mask = var_3796_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3796_cast_fp16")]; + tensor var_3803_begin_0 = const()[name = string("op_3803_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3803_end_0 = const()[name = string("op_3803_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3803_end_mask_0 = const()[name = string("op_3803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3803_cast_fp16 = slice_by_index(begin = var_3803_begin_0, end = var_3803_end_0, end_mask = var_3803_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3803_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = string("op_3810_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3810_end_0 = const()[name = string("op_3810_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3810_end_mask_0 = const()[name = string("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3810_cast_fp16")]; + tensor var_3817_begin_0 = const()[name = string("op_3817_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3817_end_0 = const()[name = string("op_3817_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3817_end_mask_0 = const()[name = string("op_3817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3817_cast_fp16 = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3817_cast_fp16")]; + tensor var_3824_begin_0 = const()[name = string("op_3824_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3824_end_0 = const()[name = string("op_3824_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3824_end_mask_0 = const()[name = string("op_3824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3824_cast_fp16 = slice_by_index(begin = var_3824_begin_0, end = var_3824_end_0, end_mask = var_3824_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3824_cast_fp16")]; + tensor var_3831_begin_0 = const()[name = string("op_3831_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3831_end_0 = const()[name = string("op_3831_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3831_end_mask_0 = const()[name = string("op_3831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3831_cast_fp16 = slice_by_index(begin = var_3831_begin_0, end = var_3831_end_0, end_mask = var_3831_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3831_cast_fp16")]; + tensor var_3838_begin_0 = const()[name = string("op_3838_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3838_end_0 = const()[name = string("op_3838_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3838_end_mask_0 = const()[name = string("op_3838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3838_cast_fp16 = slice_by_index(begin = var_3838_begin_0, end = var_3838_end_0, end_mask = var_3838_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3838_cast_fp16")]; + tensor var_3845_begin_0 = const()[name = string("op_3845_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3845_end_0 = const()[name = string("op_3845_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3845_end_mask_0 = const()[name = string("op_3845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3845_cast_fp16 = slice_by_index(begin = var_3845_begin_0, end = var_3845_end_0, end_mask = var_3845_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3845_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = string("op_3852_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3852_end_0 = const()[name = string("op_3852_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3852_end_mask_0 = const()[name = string("op_3852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3852_cast_fp16")]; + tensor var_3859_begin_0 = const()[name = string("op_3859_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3859_end_0 = const()[name = string("op_3859_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3859_end_mask_0 = const()[name = string("op_3859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3859_cast_fp16 = slice_by_index(begin = var_3859_begin_0, end = var_3859_end_0, end_mask = var_3859_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3859_cast_fp16")]; + tensor var_3866_begin_0 = const()[name = string("op_3866_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3866_end_0 = const()[name = string("op_3866_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3866_end_mask_0 = const()[name = string("op_3866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3866_cast_fp16 = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3866_cast_fp16")]; + tensor var_3873_begin_0 = const()[name = string("op_3873_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3873_end_0 = const()[name = string("op_3873_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3873_end_mask_0 = const()[name = string("op_3873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3873_cast_fp16 = slice_by_index(begin = var_3873_begin_0, end = var_3873_end_0, end_mask = var_3873_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3873_cast_fp16")]; + tensor var_3880_begin_0 = const()[name = string("op_3880_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3880_end_0 = const()[name = string("op_3880_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3880_end_mask_0 = const()[name = string("op_3880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3880_cast_fp16 = slice_by_index(begin = var_3880_begin_0, end = var_3880_end_0, end_mask = var_3880_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3880_cast_fp16")]; + tensor var_3887_begin_0 = const()[name = string("op_3887_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3887_end_0 = const()[name = string("op_3887_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3887_end_mask_0 = const()[name = string("op_3887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3887_cast_fp16 = slice_by_index(begin = var_3887_begin_0, end = var_3887_end_0, end_mask = var_3887_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3887_cast_fp16")]; + tensor var_3894_begin_0 = const()[name = string("op_3894_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3894_end_0 = const()[name = string("op_3894_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3894_end_mask_0 = const()[name = string("op_3894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3894_cast_fp16 = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3894_cast_fp16")]; + tensor var_3901_begin_0 = const()[name = string("op_3901_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3901_end_0 = const()[name = string("op_3901_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3901_end_mask_0 = const()[name = string("op_3901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3901_cast_fp16 = slice_by_index(begin = var_3901_begin_0, end = var_3901_end_0, end_mask = var_3901_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3901_cast_fp16")]; + tensor var_3908_begin_0 = const()[name = string("op_3908_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3908_end_0 = const()[name = string("op_3908_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3908_end_mask_0 = const()[name = string("op_3908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3908_cast_fp16 = slice_by_index(begin = var_3908_begin_0, end = var_3908_end_0, end_mask = var_3908_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3908_cast_fp16")]; + tensor var_3915_begin_0 = const()[name = string("op_3915_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3915_end_0 = const()[name = string("op_3915_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3915_end_mask_0 = const()[name = string("op_3915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3915_cast_fp16 = slice_by_index(begin = var_3915_begin_0, end = var_3915_end_0, end_mask = var_3915_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3915_cast_fp16")]; + tensor var_3922_begin_0 = const()[name = string("op_3922_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3922_end_0 = const()[name = string("op_3922_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3922_end_mask_0 = const()[name = string("op_3922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3922_cast_fp16 = slice_by_index(begin = var_3922_begin_0, end = var_3922_end_0, end_mask = var_3922_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3922_cast_fp16")]; + tensor var_3929_begin_0 = const()[name = string("op_3929_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3929_end_0 = const()[name = string("op_3929_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3929_end_mask_0 = const()[name = string("op_3929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3929_cast_fp16 = slice_by_index(begin = var_3929_begin_0, end = var_3929_end_0, end_mask = var_3929_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3929_cast_fp16")]; + tensor var_3936_begin_0 = const()[name = string("op_3936_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3936_end_0 = const()[name = string("op_3936_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3936_end_mask_0 = const()[name = string("op_3936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3936_cast_fp16 = slice_by_index(begin = var_3936_begin_0, end = var_3936_end_0, end_mask = var_3936_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3936_cast_fp16")]; + tensor var_3943_begin_0 = const()[name = string("op_3943_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3943_end_0 = const()[name = string("op_3943_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3943_end_mask_0 = const()[name = string("op_3943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3943_cast_fp16 = slice_by_index(begin = var_3943_begin_0, end = var_3943_end_0, end_mask = var_3943_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3943_cast_fp16")]; + tensor var_3950_begin_0 = const()[name = string("op_3950_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3950_end_0 = const()[name = string("op_3950_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3950_end_mask_0 = const()[name = string("op_3950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3950_cast_fp16 = slice_by_index(begin = var_3950_begin_0, end = var_3950_end_0, end_mask = var_3950_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3950_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3955_begin_0 = const()[name = string("op_3955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3955_end_0 = const()[name = string("op_3955_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3955_end_mask_0 = const()[name = string("op_3955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_29")]; + tensor var_3955_cast_fp16 = slice_by_index(begin = var_3955_begin_0, end = var_3955_end_0, end_mask = var_3955_end_mask_0, x = k_5_cast_fp16)[name = string("op_3955_cast_fp16")]; + tensor var_3959_begin_0 = const()[name = string("op_3959_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3959_end_0 = const()[name = string("op_3959_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3959_end_mask_0 = const()[name = string("op_3959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3959_cast_fp16 = slice_by_index(begin = var_3959_begin_0, end = var_3959_end_0, end_mask = var_3959_end_mask_0, x = k_5_cast_fp16)[name = string("op_3959_cast_fp16")]; + tensor var_3963_begin_0 = const()[name = string("op_3963_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3963_end_0 = const()[name = string("op_3963_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3963_end_mask_0 = const()[name = string("op_3963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3963_cast_fp16 = slice_by_index(begin = var_3963_begin_0, end = var_3963_end_0, end_mask = var_3963_end_mask_0, x = k_5_cast_fp16)[name = string("op_3963_cast_fp16")]; + tensor var_3967_begin_0 = const()[name = string("op_3967_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3967_end_0 = const()[name = string("op_3967_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3967_end_mask_0 = const()[name = string("op_3967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3967_cast_fp16 = slice_by_index(begin = var_3967_begin_0, end = var_3967_end_0, end_mask = var_3967_end_mask_0, x = k_5_cast_fp16)[name = string("op_3967_cast_fp16")]; + tensor var_3971_begin_0 = const()[name = string("op_3971_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3971_end_0 = const()[name = string("op_3971_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3971_end_mask_0 = const()[name = string("op_3971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3971_cast_fp16 = slice_by_index(begin = var_3971_begin_0, end = var_3971_end_0, end_mask = var_3971_end_mask_0, x = k_5_cast_fp16)[name = string("op_3971_cast_fp16")]; + tensor var_3975_begin_0 = const()[name = string("op_3975_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3975_end_0 = const()[name = string("op_3975_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3975_end_mask_0 = const()[name = string("op_3975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3975_cast_fp16 = slice_by_index(begin = var_3975_begin_0, end = var_3975_end_0, end_mask = var_3975_end_mask_0, x = k_5_cast_fp16)[name = string("op_3975_cast_fp16")]; + tensor var_3979_begin_0 = const()[name = string("op_3979_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3979_end_0 = const()[name = string("op_3979_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3979_end_mask_0 = const()[name = string("op_3979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3979_cast_fp16 = slice_by_index(begin = var_3979_begin_0, end = var_3979_end_0, end_mask = var_3979_end_mask_0, x = k_5_cast_fp16)[name = string("op_3979_cast_fp16")]; + tensor var_3983_begin_0 = const()[name = string("op_3983_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3983_end_0 = const()[name = string("op_3983_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3983_end_mask_0 = const()[name = string("op_3983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3983_cast_fp16 = slice_by_index(begin = var_3983_begin_0, end = var_3983_end_0, end_mask = var_3983_end_mask_0, x = k_5_cast_fp16)[name = string("op_3983_cast_fp16")]; + tensor var_3987_begin_0 = const()[name = string("op_3987_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3987_end_0 = const()[name = string("op_3987_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_3987_end_mask_0 = const()[name = string("op_3987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3987_cast_fp16 = slice_by_index(begin = var_3987_begin_0, end = var_3987_end_0, end_mask = var_3987_end_mask_0, x = k_5_cast_fp16)[name = string("op_3987_cast_fp16")]; + tensor var_3991_begin_0 = const()[name = string("op_3991_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3991_end_0 = const()[name = string("op_3991_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_3991_end_mask_0 = const()[name = string("op_3991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3991_cast_fp16 = slice_by_index(begin = var_3991_begin_0, end = var_3991_end_0, end_mask = var_3991_end_mask_0, x = k_5_cast_fp16)[name = string("op_3991_cast_fp16")]; + tensor var_3995_begin_0 = const()[name = string("op_3995_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3995_end_0 = const()[name = string("op_3995_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_3995_end_mask_0 = const()[name = string("op_3995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3995_cast_fp16 = slice_by_index(begin = var_3995_begin_0, end = var_3995_end_0, end_mask = var_3995_end_mask_0, x = k_5_cast_fp16)[name = string("op_3995_cast_fp16")]; + tensor var_3999_begin_0 = const()[name = string("op_3999_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3999_end_0 = const()[name = string("op_3999_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_3999_end_mask_0 = const()[name = string("op_3999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3999_cast_fp16 = slice_by_index(begin = var_3999_begin_0, end = var_3999_end_0, end_mask = var_3999_end_mask_0, x = k_5_cast_fp16)[name = string("op_3999_cast_fp16")]; + tensor var_4003_begin_0 = const()[name = string("op_4003_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4003_end_0 = const()[name = string("op_4003_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_4003_end_mask_0 = const()[name = string("op_4003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4003_cast_fp16 = slice_by_index(begin = var_4003_begin_0, end = var_4003_end_0, end_mask = var_4003_end_mask_0, x = k_5_cast_fp16)[name = string("op_4003_cast_fp16")]; + tensor var_4007_begin_0 = const()[name = string("op_4007_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4007_end_0 = const()[name = string("op_4007_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_4007_end_mask_0 = const()[name = string("op_4007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4007_cast_fp16 = slice_by_index(begin = var_4007_begin_0, end = var_4007_end_0, end_mask = var_4007_end_mask_0, x = k_5_cast_fp16)[name = string("op_4007_cast_fp16")]; + tensor var_4011_begin_0 = const()[name = string("op_4011_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4011_end_0 = const()[name = string("op_4011_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_4011_end_mask_0 = const()[name = string("op_4011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4011_cast_fp16 = slice_by_index(begin = var_4011_begin_0, end = var_4011_end_0, end_mask = var_4011_end_mask_0, x = k_5_cast_fp16)[name = string("op_4011_cast_fp16")]; + tensor var_4015_begin_0 = const()[name = string("op_4015_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4015_end_0 = const()[name = string("op_4015_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_4015_end_mask_0 = const()[name = string("op_4015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4015_cast_fp16 = slice_by_index(begin = var_4015_begin_0, end = var_4015_end_0, end_mask = var_4015_end_mask_0, x = k_5_cast_fp16)[name = string("op_4015_cast_fp16")]; + tensor var_4019_begin_0 = const()[name = string("op_4019_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4019_end_0 = const()[name = string("op_4019_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_4019_end_mask_0 = const()[name = string("op_4019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4019_cast_fp16 = slice_by_index(begin = var_4019_begin_0, end = var_4019_end_0, end_mask = var_4019_end_mask_0, x = k_5_cast_fp16)[name = string("op_4019_cast_fp16")]; + tensor var_4023_begin_0 = const()[name = string("op_4023_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4023_end_0 = const()[name = string("op_4023_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_4023_end_mask_0 = const()[name = string("op_4023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4023_cast_fp16 = slice_by_index(begin = var_4023_begin_0, end = var_4023_end_0, end_mask = var_4023_end_mask_0, x = k_5_cast_fp16)[name = string("op_4023_cast_fp16")]; + tensor var_4027_begin_0 = const()[name = string("op_4027_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4027_end_0 = const()[name = string("op_4027_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_4027_end_mask_0 = const()[name = string("op_4027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4027_cast_fp16 = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = k_5_cast_fp16)[name = string("op_4027_cast_fp16")]; + tensor var_4031_begin_0 = const()[name = string("op_4031_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4031_end_0 = const()[name = string("op_4031_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_4031_end_mask_0 = const()[name = string("op_4031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4031_cast_fp16 = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = k_5_cast_fp16)[name = string("op_4031_cast_fp16")]; + tensor var_4033_begin_0 = const()[name = string("op_4033_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4033_end_0 = const()[name = string("op_4033_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4033_end_mask_0 = const()[name = string("op_4033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4033_cast_fp16 = slice_by_index(begin = var_4033_begin_0, end = var_4033_end_0, end_mask = var_4033_end_mask_0, x = value_5_cast_fp16)[name = string("op_4033_cast_fp16")]; + tensor var_4037_begin_0 = const()[name = string("op_4037_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4037_end_0 = const()[name = string("op_4037_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4037_end_mask_0 = const()[name = string("op_4037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4037_cast_fp16 = slice_by_index(begin = var_4037_begin_0, end = var_4037_end_0, end_mask = var_4037_end_mask_0, x = value_5_cast_fp16)[name = string("op_4037_cast_fp16")]; + tensor var_4041_begin_0 = const()[name = string("op_4041_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4041_end_0 = const()[name = string("op_4041_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4041_end_mask_0 = const()[name = string("op_4041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4041_cast_fp16 = slice_by_index(begin = var_4041_begin_0, end = var_4041_end_0, end_mask = var_4041_end_mask_0, x = value_5_cast_fp16)[name = string("op_4041_cast_fp16")]; + tensor var_4045_begin_0 = const()[name = string("op_4045_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4045_end_0 = const()[name = string("op_4045_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4045_end_mask_0 = const()[name = string("op_4045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4045_cast_fp16 = slice_by_index(begin = var_4045_begin_0, end = var_4045_end_0, end_mask = var_4045_end_mask_0, x = value_5_cast_fp16)[name = string("op_4045_cast_fp16")]; + tensor var_4049_begin_0 = const()[name = string("op_4049_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4049_end_0 = const()[name = string("op_4049_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4049_end_mask_0 = const()[name = string("op_4049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4049_cast_fp16 = slice_by_index(begin = var_4049_begin_0, end = var_4049_end_0, end_mask = var_4049_end_mask_0, x = value_5_cast_fp16)[name = string("op_4049_cast_fp16")]; + tensor var_4053_begin_0 = const()[name = string("op_4053_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4053_end_0 = const()[name = string("op_4053_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4053_end_mask_0 = const()[name = string("op_4053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4053_cast_fp16 = slice_by_index(begin = var_4053_begin_0, end = var_4053_end_0, end_mask = var_4053_end_mask_0, x = value_5_cast_fp16)[name = string("op_4053_cast_fp16")]; + tensor var_4057_begin_0 = const()[name = string("op_4057_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4057_end_0 = const()[name = string("op_4057_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4057_end_mask_0 = const()[name = string("op_4057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4057_cast_fp16 = slice_by_index(begin = var_4057_begin_0, end = var_4057_end_0, end_mask = var_4057_end_mask_0, x = value_5_cast_fp16)[name = string("op_4057_cast_fp16")]; + tensor var_4061_begin_0 = const()[name = string("op_4061_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4061_end_0 = const()[name = string("op_4061_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4061_end_mask_0 = const()[name = string("op_4061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4061_cast_fp16 = slice_by_index(begin = var_4061_begin_0, end = var_4061_end_0, end_mask = var_4061_end_mask_0, x = value_5_cast_fp16)[name = string("op_4061_cast_fp16")]; + tensor var_4065_begin_0 = const()[name = string("op_4065_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4065_end_0 = const()[name = string("op_4065_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4065_end_mask_0 = const()[name = string("op_4065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4065_cast_fp16 = slice_by_index(begin = var_4065_begin_0, end = var_4065_end_0, end_mask = var_4065_end_mask_0, x = value_5_cast_fp16)[name = string("op_4065_cast_fp16")]; + tensor var_4069_begin_0 = const()[name = string("op_4069_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4069_end_0 = const()[name = string("op_4069_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4069_end_mask_0 = const()[name = string("op_4069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4069_cast_fp16 = slice_by_index(begin = var_4069_begin_0, end = var_4069_end_0, end_mask = var_4069_end_mask_0, x = value_5_cast_fp16)[name = string("op_4069_cast_fp16")]; + tensor var_4073_begin_0 = const()[name = string("op_4073_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4073_end_0 = const()[name = string("op_4073_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4073_end_mask_0 = const()[name = string("op_4073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4073_cast_fp16 = slice_by_index(begin = var_4073_begin_0, end = var_4073_end_0, end_mask = var_4073_end_mask_0, x = value_5_cast_fp16)[name = string("op_4073_cast_fp16")]; + tensor var_4077_begin_0 = const()[name = string("op_4077_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4077_end_0 = const()[name = string("op_4077_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4077_end_mask_0 = const()[name = string("op_4077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4077_cast_fp16 = slice_by_index(begin = var_4077_begin_0, end = var_4077_end_0, end_mask = var_4077_end_mask_0, x = value_5_cast_fp16)[name = string("op_4077_cast_fp16")]; + tensor var_4081_begin_0 = const()[name = string("op_4081_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4081_end_0 = const()[name = string("op_4081_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_4081_end_mask_0 = const()[name = string("op_4081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4081_cast_fp16 = slice_by_index(begin = var_4081_begin_0, end = var_4081_end_0, end_mask = var_4081_end_mask_0, x = value_5_cast_fp16)[name = string("op_4081_cast_fp16")]; + tensor var_4085_begin_0 = const()[name = string("op_4085_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4085_end_0 = const()[name = string("op_4085_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_4085_end_mask_0 = const()[name = string("op_4085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4085_cast_fp16 = slice_by_index(begin = var_4085_begin_0, end = var_4085_end_0, end_mask = var_4085_end_mask_0, x = value_5_cast_fp16)[name = string("op_4085_cast_fp16")]; + tensor var_4089_begin_0 = const()[name = string("op_4089_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4089_end_0 = const()[name = string("op_4089_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_4089_end_mask_0 = const()[name = string("op_4089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4089_cast_fp16 = slice_by_index(begin = var_4089_begin_0, end = var_4089_end_0, end_mask = var_4089_end_mask_0, x = value_5_cast_fp16)[name = string("op_4089_cast_fp16")]; + tensor var_4093_begin_0 = const()[name = string("op_4093_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4093_end_0 = const()[name = string("op_4093_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_4093_end_mask_0 = const()[name = string("op_4093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4093_cast_fp16 = slice_by_index(begin = var_4093_begin_0, end = var_4093_end_0, end_mask = var_4093_end_mask_0, x = value_5_cast_fp16)[name = string("op_4093_cast_fp16")]; + tensor var_4097_begin_0 = const()[name = string("op_4097_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4097_end_0 = const()[name = string("op_4097_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_4097_end_mask_0 = const()[name = string("op_4097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4097_cast_fp16 = slice_by_index(begin = var_4097_begin_0, end = var_4097_end_0, end_mask = var_4097_end_mask_0, x = value_5_cast_fp16)[name = string("op_4097_cast_fp16")]; + tensor var_4101_begin_0 = const()[name = string("op_4101_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4101_end_0 = const()[name = string("op_4101_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_4101_end_mask_0 = const()[name = string("op_4101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4101_cast_fp16 = slice_by_index(begin = var_4101_begin_0, end = var_4101_end_0, end_mask = var_4101_end_mask_0, x = value_5_cast_fp16)[name = string("op_4101_cast_fp16")]; + tensor var_4105_begin_0 = const()[name = string("op_4105_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4105_end_0 = const()[name = string("op_4105_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_4105_end_mask_0 = const()[name = string("op_4105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4105_cast_fp16 = slice_by_index(begin = var_4105_begin_0, end = var_4105_end_0, end_mask = var_4105_end_mask_0, x = value_5_cast_fp16)[name = string("op_4105_cast_fp16")]; + tensor var_4109_begin_0 = const()[name = string("op_4109_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4109_end_0 = const()[name = string("op_4109_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_4109_end_mask_0 = const()[name = string("op_4109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4109_cast_fp16 = slice_by_index(begin = var_4109_begin_0, end = var_4109_end_0, end_mask = var_4109_end_mask_0, x = value_5_cast_fp16)[name = string("op_4109_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3955_cast_fp16, var_3397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3955_cast_fp16, var_3404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3955_cast_fp16, var_3411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3955_cast_fp16, var_3418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3959_cast_fp16, var_3425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3959_cast_fp16, var_3432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3959_cast_fp16, var_3439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3959_cast_fp16, var_3446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3963_cast_fp16, var_3453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3963_cast_fp16, var_3460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3963_cast_fp16, var_3467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3963_cast_fp16, var_3474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3967_cast_fp16, var_3481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3967_cast_fp16, var_3488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3967_cast_fp16, var_3495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3967_cast_fp16, var_3502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3971_cast_fp16, var_3509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3971_cast_fp16, var_3516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3971_cast_fp16, var_3523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3971_cast_fp16, var_3530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3975_cast_fp16, var_3537_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3975_cast_fp16, var_3544_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3975_cast_fp16, var_3551_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3975_cast_fp16, var_3558_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3979_cast_fp16, var_3565_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3979_cast_fp16, var_3572_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3979_cast_fp16, var_3579_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3979_cast_fp16, var_3586_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3983_cast_fp16, var_3593_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3983_cast_fp16, var_3600_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3983_cast_fp16, var_3607_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_383_equation_0, values = (var_3983_cast_fp16, var_3614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_383_cast_fp16")]; + string _SplitHeadsQ__mh_w_385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_385_equation_0, values = (var_3987_cast_fp16, var_3621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_385_cast_fp16")]; + string _SplitHeadsQ__mh_w_387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_387_equation_0, values = (var_3987_cast_fp16, var_3628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_387_cast_fp16")]; + string _SplitHeadsQ__mh_w_389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_389_equation_0, values = (var_3987_cast_fp16, var_3635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_389_cast_fp16")]; + string _SplitHeadsQ__mh_w_391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_391_equation_0, values = (var_3987_cast_fp16, var_3642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_391_cast_fp16")]; + string _SplitHeadsQ__mh_w_393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_393_equation_0, values = (var_3991_cast_fp16, var_3649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_393_cast_fp16")]; + string _SplitHeadsQ__mh_w_395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_395_equation_0, values = (var_3991_cast_fp16, var_3656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_395_cast_fp16")]; + string _SplitHeadsQ__mh_w_397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_397_equation_0, values = (var_3991_cast_fp16, var_3663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_397_cast_fp16")]; + string _SplitHeadsQ__mh_w_399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_399_equation_0, values = (var_3991_cast_fp16, var_3670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_399_cast_fp16")]; + string _SplitHeadsQ__mh_w_401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_401_equation_0, values = (var_3995_cast_fp16, var_3677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_401_cast_fp16")]; + string _SplitHeadsQ__mh_w_403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_403_equation_0, values = (var_3995_cast_fp16, var_3684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_403_cast_fp16")]; + string _SplitHeadsQ__mh_w_405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_405_equation_0, values = (var_3995_cast_fp16, var_3691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_405_cast_fp16")]; + string _SplitHeadsQ__mh_w_407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_407_equation_0, values = (var_3995_cast_fp16, var_3698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_407_cast_fp16")]; + string _SplitHeadsQ__mh_w_409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_409_equation_0, values = (var_3999_cast_fp16, var_3705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_409_cast_fp16")]; + string _SplitHeadsQ__mh_w_411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_411_equation_0, values = (var_3999_cast_fp16, var_3712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_411_cast_fp16")]; + string _SplitHeadsQ__mh_w_413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_413_equation_0, values = (var_3999_cast_fp16, var_3719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_413_cast_fp16")]; + string _SplitHeadsQ__mh_w_415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_415_equation_0, values = (var_3999_cast_fp16, var_3726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_415_cast_fp16")]; + string _SplitHeadsQ__mh_w_417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_417_equation_0, values = (var_4003_cast_fp16, var_3733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_417_cast_fp16")]; + string _SplitHeadsQ__mh_w_419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_419_equation_0, values = (var_4003_cast_fp16, var_3740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_419_cast_fp16")]; + string _SplitHeadsQ__mh_w_421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_421_equation_0, values = (var_4003_cast_fp16, var_3747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_421_cast_fp16")]; + string _SplitHeadsQ__mh_w_423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_423_equation_0, values = (var_4003_cast_fp16, var_3754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_423_cast_fp16")]; + string _SplitHeadsQ__mh_w_425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_425_equation_0, values = (var_4007_cast_fp16, var_3761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_425_cast_fp16")]; + string _SplitHeadsQ__mh_w_427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_427_equation_0, values = (var_4007_cast_fp16, var_3768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_427_cast_fp16")]; + string _SplitHeadsQ__mh_w_429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_429_equation_0, values = (var_4007_cast_fp16, var_3775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_429_cast_fp16")]; + string _SplitHeadsQ__mh_w_431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_431_equation_0, values = (var_4007_cast_fp16, var_3782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_431_cast_fp16")]; + string _SplitHeadsQ__mh_w_433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_433_equation_0, values = (var_4011_cast_fp16, var_3789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_433_cast_fp16")]; + string _SplitHeadsQ__mh_w_435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_435_equation_0, values = (var_4011_cast_fp16, var_3796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_435_cast_fp16")]; + string _SplitHeadsQ__mh_w_437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_437_equation_0, values = (var_4011_cast_fp16, var_3803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_437_cast_fp16")]; + string _SplitHeadsQ__mh_w_439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_439_equation_0, values = (var_4011_cast_fp16, var_3810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_439_cast_fp16")]; + string _SplitHeadsQ__mh_w_441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_441_equation_0, values = (var_4015_cast_fp16, var_3817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_441_cast_fp16")]; + string _SplitHeadsQ__mh_w_443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_443_equation_0, values = (var_4015_cast_fp16, var_3824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_443_cast_fp16")]; + string _SplitHeadsQ__mh_w_445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_445_equation_0, values = (var_4015_cast_fp16, var_3831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_445_cast_fp16")]; + string _SplitHeadsQ__mh_w_447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_447_equation_0, values = (var_4015_cast_fp16, var_3838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_447_cast_fp16")]; + string _SplitHeadsQ__mh_w_449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_449_equation_0, values = (var_4019_cast_fp16, var_3845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_449_cast_fp16")]; + string _SplitHeadsQ__mh_w_451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_451_equation_0, values = (var_4019_cast_fp16, var_3852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_451_cast_fp16")]; + string _SplitHeadsQ__mh_w_453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_453_equation_0, values = (var_4019_cast_fp16, var_3859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_453_cast_fp16")]; + string _SplitHeadsQ__mh_w_455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_455_equation_0, values = (var_4019_cast_fp16, var_3866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_455_cast_fp16")]; + string _SplitHeadsQ__mh_w_457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_457_equation_0, values = (var_4023_cast_fp16, var_3873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_457_cast_fp16")]; + string _SplitHeadsQ__mh_w_459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_459_equation_0, values = (var_4023_cast_fp16, var_3880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_459_cast_fp16")]; + string _SplitHeadsQ__mh_w_461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_461_equation_0, values = (var_4023_cast_fp16, var_3887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_461_cast_fp16")]; + string _SplitHeadsQ__mh_w_463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_463_equation_0, values = (var_4023_cast_fp16, var_3894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_463_cast_fp16")]; + string _SplitHeadsQ__mh_w_465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_465_equation_0, values = (var_4027_cast_fp16, var_3901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_465_cast_fp16")]; + string _SplitHeadsQ__mh_w_467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_467_equation_0, values = (var_4027_cast_fp16, var_3908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_467_cast_fp16")]; + string _SplitHeadsQ__mh_w_469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_469_equation_0, values = (var_4027_cast_fp16, var_3915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_469_cast_fp16")]; + string _SplitHeadsQ__mh_w_471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_471_equation_0, values = (var_4027_cast_fp16, var_3922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_471_cast_fp16")]; + string _SplitHeadsQ__mh_w_473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_473_equation_0, values = (var_4031_cast_fp16, var_3929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_473_cast_fp16")]; + string _SplitHeadsQ__mh_w_475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_475_equation_0, values = (var_4031_cast_fp16, var_3936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_475_cast_fp16")]; + string _SplitHeadsQ__mh_w_477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_477_equation_0, values = (var_4031_cast_fp16, var_3943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_477_cast_fp16")]; + string _SplitHeadsQ__mh_w_479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_479_equation_0, values = (var_4031_cast_fp16, var_3950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_479_cast_fp16")]; + fp16 var_4272_to_fp16 = const()[name = string("op_4272_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_4272_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_4274_to_fp16 = const()[name = string("op_4274_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_4274_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_4276_to_fp16 = const()[name = string("op_4276_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_4276_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_4278_to_fp16 = const()[name = string("op_4278_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_4278_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_4280_to_fp16 = const()[name = string("op_4280_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_4280_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_4282_to_fp16 = const()[name = string("op_4282_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_4282_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_4284_to_fp16 = const()[name = string("op_4284_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_4284_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_4286_to_fp16 = const()[name = string("op_4286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_4286_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_4288_to_fp16 = const()[name = string("op_4288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_4288_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_4290_to_fp16 = const()[name = string("op_4290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_4290_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_4292_to_fp16 = const()[name = string("op_4292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_4292_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_4294_to_fp16 = const()[name = string("op_4294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_4294_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_4296_to_fp16 = const()[name = string("op_4296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_4296_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_4298_to_fp16 = const()[name = string("op_4298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_4298_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_4300_to_fp16 = const()[name = string("op_4300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_4300_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_4302_to_fp16 = const()[name = string("op_4302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_4302_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_4304_to_fp16 = const()[name = string("op_4304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_4304_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_4306_to_fp16 = const()[name = string("op_4306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_4306_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_4308_to_fp16 = const()[name = string("op_4308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_4308_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_4310_to_fp16 = const()[name = string("op_4310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_4310_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_4312_to_fp16 = const()[name = string("op_4312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_4312_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_4314_to_fp16 = const()[name = string("op_4314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_4314_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_4316_to_fp16 = const()[name = string("op_4316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_4316_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_4318_to_fp16 = const()[name = string("op_4318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_4318_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_4320_to_fp16 = const()[name = string("op_4320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_4320_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_4322_to_fp16 = const()[name = string("op_4322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_4322_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_4324_to_fp16 = const()[name = string("op_4324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_4324_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_4326_to_fp16 = const()[name = string("op_4326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_4326_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_4328_to_fp16 = const()[name = string("op_4328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_4328_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_4330_to_fp16 = const()[name = string("op_4330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_4330_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_4332_to_fp16 = const()[name = string("op_4332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_4332_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_4334_to_fp16 = const()[name = string("op_4334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_383_cast_fp16, y = var_4334_to_fp16)[name = string("aw_chunk_383_cast_fp16")]; + fp16 var_4336_to_fp16 = const()[name = string("op_4336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_385_cast_fp16, y = var_4336_to_fp16)[name = string("aw_chunk_385_cast_fp16")]; + fp16 var_4338_to_fp16 = const()[name = string("op_4338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_387_cast_fp16, y = var_4338_to_fp16)[name = string("aw_chunk_387_cast_fp16")]; + fp16 var_4340_to_fp16 = const()[name = string("op_4340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_389_cast_fp16, y = var_4340_to_fp16)[name = string("aw_chunk_389_cast_fp16")]; + fp16 var_4342_to_fp16 = const()[name = string("op_4342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_391_cast_fp16, y = var_4342_to_fp16)[name = string("aw_chunk_391_cast_fp16")]; + fp16 var_4344_to_fp16 = const()[name = string("op_4344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_393_cast_fp16, y = var_4344_to_fp16)[name = string("aw_chunk_393_cast_fp16")]; + fp16 var_4346_to_fp16 = const()[name = string("op_4346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_395_cast_fp16, y = var_4346_to_fp16)[name = string("aw_chunk_395_cast_fp16")]; + fp16 var_4348_to_fp16 = const()[name = string("op_4348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_397_cast_fp16, y = var_4348_to_fp16)[name = string("aw_chunk_397_cast_fp16")]; + fp16 var_4350_to_fp16 = const()[name = string("op_4350_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_399_cast_fp16, y = var_4350_to_fp16)[name = string("aw_chunk_399_cast_fp16")]; + fp16 var_4352_to_fp16 = const()[name = string("op_4352_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_401_cast_fp16, y = var_4352_to_fp16)[name = string("aw_chunk_401_cast_fp16")]; + fp16 var_4354_to_fp16 = const()[name = string("op_4354_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_403_cast_fp16, y = var_4354_to_fp16)[name = string("aw_chunk_403_cast_fp16")]; + fp16 var_4356_to_fp16 = const()[name = string("op_4356_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_405_cast_fp16, y = var_4356_to_fp16)[name = string("aw_chunk_405_cast_fp16")]; + fp16 var_4358_to_fp16 = const()[name = string("op_4358_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_407_cast_fp16, y = var_4358_to_fp16)[name = string("aw_chunk_407_cast_fp16")]; + fp16 var_4360_to_fp16 = const()[name = string("op_4360_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_409_cast_fp16, y = var_4360_to_fp16)[name = string("aw_chunk_409_cast_fp16")]; + fp16 var_4362_to_fp16 = const()[name = string("op_4362_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_411_cast_fp16, y = var_4362_to_fp16)[name = string("aw_chunk_411_cast_fp16")]; + fp16 var_4364_to_fp16 = const()[name = string("op_4364_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_413_cast_fp16, y = var_4364_to_fp16)[name = string("aw_chunk_413_cast_fp16")]; + fp16 var_4366_to_fp16 = const()[name = string("op_4366_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_415_cast_fp16, y = var_4366_to_fp16)[name = string("aw_chunk_415_cast_fp16")]; + fp16 var_4368_to_fp16 = const()[name = string("op_4368_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_417_cast_fp16, y = var_4368_to_fp16)[name = string("aw_chunk_417_cast_fp16")]; + fp16 var_4370_to_fp16 = const()[name = string("op_4370_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_419_cast_fp16, y = var_4370_to_fp16)[name = string("aw_chunk_419_cast_fp16")]; + fp16 var_4372_to_fp16 = const()[name = string("op_4372_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_421_cast_fp16, y = var_4372_to_fp16)[name = string("aw_chunk_421_cast_fp16")]; + fp16 var_4374_to_fp16 = const()[name = string("op_4374_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_423_cast_fp16, y = var_4374_to_fp16)[name = string("aw_chunk_423_cast_fp16")]; + fp16 var_4376_to_fp16 = const()[name = string("op_4376_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_425_cast_fp16, y = var_4376_to_fp16)[name = string("aw_chunk_425_cast_fp16")]; + fp16 var_4378_to_fp16 = const()[name = string("op_4378_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_427_cast_fp16, y = var_4378_to_fp16)[name = string("aw_chunk_427_cast_fp16")]; + fp16 var_4380_to_fp16 = const()[name = string("op_4380_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_429_cast_fp16, y = var_4380_to_fp16)[name = string("aw_chunk_429_cast_fp16")]; + fp16 var_4382_to_fp16 = const()[name = string("op_4382_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_431_cast_fp16, y = var_4382_to_fp16)[name = string("aw_chunk_431_cast_fp16")]; + fp16 var_4384_to_fp16 = const()[name = string("op_4384_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_433_cast_fp16, y = var_4384_to_fp16)[name = string("aw_chunk_433_cast_fp16")]; + fp16 var_4386_to_fp16 = const()[name = string("op_4386_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_435_cast_fp16, y = var_4386_to_fp16)[name = string("aw_chunk_435_cast_fp16")]; + fp16 var_4388_to_fp16 = const()[name = string("op_4388_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_437_cast_fp16, y = var_4388_to_fp16)[name = string("aw_chunk_437_cast_fp16")]; + fp16 var_4390_to_fp16 = const()[name = string("op_4390_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_439_cast_fp16, y = var_4390_to_fp16)[name = string("aw_chunk_439_cast_fp16")]; + fp16 var_4392_to_fp16 = const()[name = string("op_4392_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_441_cast_fp16, y = var_4392_to_fp16)[name = string("aw_chunk_441_cast_fp16")]; + fp16 var_4394_to_fp16 = const()[name = string("op_4394_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_443_cast_fp16, y = var_4394_to_fp16)[name = string("aw_chunk_443_cast_fp16")]; + fp16 var_4396_to_fp16 = const()[name = string("op_4396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_445_cast_fp16, y = var_4396_to_fp16)[name = string("aw_chunk_445_cast_fp16")]; + fp16 var_4398_to_fp16 = const()[name = string("op_4398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_447_cast_fp16, y = var_4398_to_fp16)[name = string("aw_chunk_447_cast_fp16")]; + fp16 var_4400_to_fp16 = const()[name = string("op_4400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_449_cast_fp16, y = var_4400_to_fp16)[name = string("aw_chunk_449_cast_fp16")]; + fp16 var_4402_to_fp16 = const()[name = string("op_4402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_451_cast_fp16, y = var_4402_to_fp16)[name = string("aw_chunk_451_cast_fp16")]; + fp16 var_4404_to_fp16 = const()[name = string("op_4404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_453_cast_fp16, y = var_4404_to_fp16)[name = string("aw_chunk_453_cast_fp16")]; + fp16 var_4406_to_fp16 = const()[name = string("op_4406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_455_cast_fp16, y = var_4406_to_fp16)[name = string("aw_chunk_455_cast_fp16")]; + fp16 var_4408_to_fp16 = const()[name = string("op_4408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_457_cast_fp16, y = var_4408_to_fp16)[name = string("aw_chunk_457_cast_fp16")]; + fp16 var_4410_to_fp16 = const()[name = string("op_4410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_459_cast_fp16, y = var_4410_to_fp16)[name = string("aw_chunk_459_cast_fp16")]; + fp16 var_4412_to_fp16 = const()[name = string("op_4412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_461_cast_fp16, y = var_4412_to_fp16)[name = string("aw_chunk_461_cast_fp16")]; + fp16 var_4414_to_fp16 = const()[name = string("op_4414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_463_cast_fp16, y = var_4414_to_fp16)[name = string("aw_chunk_463_cast_fp16")]; + fp16 var_4416_to_fp16 = const()[name = string("op_4416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_465_cast_fp16, y = var_4416_to_fp16)[name = string("aw_chunk_465_cast_fp16")]; + fp16 var_4418_to_fp16 = const()[name = string("op_4418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_467_cast_fp16, y = var_4418_to_fp16)[name = string("aw_chunk_467_cast_fp16")]; + fp16 var_4420_to_fp16 = const()[name = string("op_4420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_469_cast_fp16, y = var_4420_to_fp16)[name = string("aw_chunk_469_cast_fp16")]; + fp16 var_4422_to_fp16 = const()[name = string("op_4422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_471_cast_fp16, y = var_4422_to_fp16)[name = string("aw_chunk_471_cast_fp16")]; + fp16 var_4424_to_fp16 = const()[name = string("op_4424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_473_cast_fp16, y = var_4424_to_fp16)[name = string("aw_chunk_473_cast_fp16")]; + fp16 var_4426_to_fp16 = const()[name = string("op_4426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_475_cast_fp16, y = var_4426_to_fp16)[name = string("aw_chunk_475_cast_fp16")]; + fp16 var_4428_to_fp16 = const()[name = string("op_4428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_477_cast_fp16, y = var_4428_to_fp16)[name = string("aw_chunk_477_cast_fp16")]; + fp16 var_4430_to_fp16 = const()[name = string("op_4430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_479_cast_fp16, y = var_4430_to_fp16)[name = string("aw_chunk_479_cast_fp16")]; + tensor var_4432_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_321_cast_fp16)[name = string("op_4432_cast_fp16")]; + tensor var_4433_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_323_cast_fp16)[name = string("op_4433_cast_fp16")]; + tensor var_4434_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_325_cast_fp16)[name = string("op_4434_cast_fp16")]; + tensor var_4435_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_327_cast_fp16)[name = string("op_4435_cast_fp16")]; + tensor var_4436_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_329_cast_fp16)[name = string("op_4436_cast_fp16")]; + tensor var_4437_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_331_cast_fp16)[name = string("op_4437_cast_fp16")]; + tensor var_4438_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_333_cast_fp16)[name = string("op_4438_cast_fp16")]; + tensor var_4439_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_335_cast_fp16)[name = string("op_4439_cast_fp16")]; + tensor var_4440_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_337_cast_fp16)[name = string("op_4440_cast_fp16")]; + tensor var_4441_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_339_cast_fp16)[name = string("op_4441_cast_fp16")]; + tensor var_4442_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_341_cast_fp16)[name = string("op_4442_cast_fp16")]; + tensor var_4443_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_343_cast_fp16)[name = string("op_4443_cast_fp16")]; + tensor var_4444_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_345_cast_fp16)[name = string("op_4444_cast_fp16")]; + tensor var_4445_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_347_cast_fp16)[name = string("op_4445_cast_fp16")]; + tensor var_4446_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_349_cast_fp16)[name = string("op_4446_cast_fp16")]; + tensor var_4447_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_351_cast_fp16)[name = string("op_4447_cast_fp16")]; + tensor var_4448_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_353_cast_fp16)[name = string("op_4448_cast_fp16")]; + tensor var_4449_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_355_cast_fp16)[name = string("op_4449_cast_fp16")]; + tensor var_4450_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_357_cast_fp16)[name = string("op_4450_cast_fp16")]; + tensor var_4451_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_359_cast_fp16)[name = string("op_4451_cast_fp16")]; + tensor var_4452_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_361_cast_fp16)[name = string("op_4452_cast_fp16")]; + tensor var_4453_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_363_cast_fp16)[name = string("op_4453_cast_fp16")]; + tensor var_4454_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_365_cast_fp16)[name = string("op_4454_cast_fp16")]; + tensor var_4455_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_367_cast_fp16)[name = string("op_4455_cast_fp16")]; + tensor var_4456_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_369_cast_fp16)[name = string("op_4456_cast_fp16")]; + tensor var_4457_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_371_cast_fp16)[name = string("op_4457_cast_fp16")]; + tensor var_4458_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_373_cast_fp16)[name = string("op_4458_cast_fp16")]; + tensor var_4459_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_375_cast_fp16)[name = string("op_4459_cast_fp16")]; + tensor var_4460_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_377_cast_fp16)[name = string("op_4460_cast_fp16")]; + tensor var_4461_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_379_cast_fp16)[name = string("op_4461_cast_fp16")]; + tensor var_4462_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_381_cast_fp16)[name = string("op_4462_cast_fp16")]; + tensor var_4463_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_383_cast_fp16)[name = string("op_4463_cast_fp16")]; + tensor var_4464_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_385_cast_fp16)[name = string("op_4464_cast_fp16")]; + tensor var_4465_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_387_cast_fp16)[name = string("op_4465_cast_fp16")]; + tensor var_4466_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_389_cast_fp16)[name = string("op_4466_cast_fp16")]; + tensor var_4467_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_391_cast_fp16)[name = string("op_4467_cast_fp16")]; + tensor var_4468_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_393_cast_fp16)[name = string("op_4468_cast_fp16")]; + tensor var_4469_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_395_cast_fp16)[name = string("op_4469_cast_fp16")]; + tensor var_4470_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_397_cast_fp16)[name = string("op_4470_cast_fp16")]; + tensor var_4471_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_399_cast_fp16)[name = string("op_4471_cast_fp16")]; + tensor var_4472_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_401_cast_fp16)[name = string("op_4472_cast_fp16")]; + tensor var_4473_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_403_cast_fp16)[name = string("op_4473_cast_fp16")]; + tensor var_4474_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_405_cast_fp16)[name = string("op_4474_cast_fp16")]; + tensor var_4475_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_407_cast_fp16)[name = string("op_4475_cast_fp16")]; + tensor var_4476_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_409_cast_fp16)[name = string("op_4476_cast_fp16")]; + tensor var_4477_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_411_cast_fp16)[name = string("op_4477_cast_fp16")]; + tensor var_4478_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_413_cast_fp16)[name = string("op_4478_cast_fp16")]; + tensor var_4479_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_415_cast_fp16)[name = string("op_4479_cast_fp16")]; + tensor var_4480_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_417_cast_fp16)[name = string("op_4480_cast_fp16")]; + tensor var_4481_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_419_cast_fp16)[name = string("op_4481_cast_fp16")]; + tensor var_4482_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_421_cast_fp16)[name = string("op_4482_cast_fp16")]; + tensor var_4483_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_423_cast_fp16)[name = string("op_4483_cast_fp16")]; + tensor var_4484_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_425_cast_fp16)[name = string("op_4484_cast_fp16")]; + tensor var_4485_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_427_cast_fp16)[name = string("op_4485_cast_fp16")]; + tensor var_4486_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_429_cast_fp16)[name = string("op_4486_cast_fp16")]; + tensor var_4487_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_431_cast_fp16)[name = string("op_4487_cast_fp16")]; + tensor var_4488_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_433_cast_fp16)[name = string("op_4488_cast_fp16")]; + tensor var_4489_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_435_cast_fp16)[name = string("op_4489_cast_fp16")]; + tensor var_4490_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_437_cast_fp16)[name = string("op_4490_cast_fp16")]; + tensor var_4491_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_439_cast_fp16)[name = string("op_4491_cast_fp16")]; + tensor var_4492_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_441_cast_fp16)[name = string("op_4492_cast_fp16")]; + tensor var_4493_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_443_cast_fp16)[name = string("op_4493_cast_fp16")]; + tensor var_4494_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_445_cast_fp16)[name = string("op_4494_cast_fp16")]; + tensor var_4495_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_447_cast_fp16)[name = string("op_4495_cast_fp16")]; + tensor var_4496_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_449_cast_fp16)[name = string("op_4496_cast_fp16")]; + tensor var_4497_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_451_cast_fp16)[name = string("op_4497_cast_fp16")]; + tensor var_4498_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_453_cast_fp16)[name = string("op_4498_cast_fp16")]; + tensor var_4499_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_455_cast_fp16)[name = string("op_4499_cast_fp16")]; + tensor var_4500_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_457_cast_fp16)[name = string("op_4500_cast_fp16")]; + tensor var_4501_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_459_cast_fp16)[name = string("op_4501_cast_fp16")]; + tensor var_4502_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_461_cast_fp16)[name = string("op_4502_cast_fp16")]; + tensor var_4503_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_463_cast_fp16)[name = string("op_4503_cast_fp16")]; + tensor var_4504_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_465_cast_fp16)[name = string("op_4504_cast_fp16")]; + tensor var_4505_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_467_cast_fp16)[name = string("op_4505_cast_fp16")]; + tensor var_4506_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_469_cast_fp16)[name = string("op_4506_cast_fp16")]; + tensor var_4507_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_471_cast_fp16)[name = string("op_4507_cast_fp16")]; + tensor var_4508_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_473_cast_fp16)[name = string("op_4508_cast_fp16")]; + tensor var_4509_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_475_cast_fp16)[name = string("op_4509_cast_fp16")]; + tensor var_4510_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_477_cast_fp16)[name = string("op_4510_cast_fp16")]; + tensor var_4511_cast_fp16 = softmax(axis = var_3257, x = aw_chunk_479_cast_fp16)[name = string("op_4511_cast_fp16")]; + string var_4513_equation_0 = const()[name = string("op_4513_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4513_cast_fp16 = einsum(equation = var_4513_equation_0, values = (var_4033_cast_fp16, var_4432_cast_fp16))[name = string("op_4513_cast_fp16")]; + string var_4515_equation_0 = const()[name = string("op_4515_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4515_cast_fp16 = einsum(equation = var_4515_equation_0, values = (var_4033_cast_fp16, var_4433_cast_fp16))[name = string("op_4515_cast_fp16")]; + string var_4517_equation_0 = const()[name = string("op_4517_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4517_cast_fp16 = einsum(equation = var_4517_equation_0, values = (var_4033_cast_fp16, var_4434_cast_fp16))[name = string("op_4517_cast_fp16")]; + string var_4519_equation_0 = const()[name = string("op_4519_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4519_cast_fp16 = einsum(equation = var_4519_equation_0, values = (var_4033_cast_fp16, var_4435_cast_fp16))[name = string("op_4519_cast_fp16")]; + string var_4521_equation_0 = const()[name = string("op_4521_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4521_cast_fp16 = einsum(equation = var_4521_equation_0, values = (var_4037_cast_fp16, var_4436_cast_fp16))[name = string("op_4521_cast_fp16")]; + string var_4523_equation_0 = const()[name = string("op_4523_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4523_cast_fp16 = einsum(equation = var_4523_equation_0, values = (var_4037_cast_fp16, var_4437_cast_fp16))[name = string("op_4523_cast_fp16")]; + string var_4525_equation_0 = const()[name = string("op_4525_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4525_cast_fp16 = einsum(equation = var_4525_equation_0, values = (var_4037_cast_fp16, var_4438_cast_fp16))[name = string("op_4525_cast_fp16")]; + string var_4527_equation_0 = const()[name = string("op_4527_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4527_cast_fp16 = einsum(equation = var_4527_equation_0, values = (var_4037_cast_fp16, var_4439_cast_fp16))[name = string("op_4527_cast_fp16")]; + string var_4529_equation_0 = const()[name = string("op_4529_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4529_cast_fp16 = einsum(equation = var_4529_equation_0, values = (var_4041_cast_fp16, var_4440_cast_fp16))[name = string("op_4529_cast_fp16")]; + string var_4531_equation_0 = const()[name = string("op_4531_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4531_cast_fp16 = einsum(equation = var_4531_equation_0, values = (var_4041_cast_fp16, var_4441_cast_fp16))[name = string("op_4531_cast_fp16")]; + string var_4533_equation_0 = const()[name = string("op_4533_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4533_cast_fp16 = einsum(equation = var_4533_equation_0, values = (var_4041_cast_fp16, var_4442_cast_fp16))[name = string("op_4533_cast_fp16")]; + string var_4535_equation_0 = const()[name = string("op_4535_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4535_cast_fp16 = einsum(equation = var_4535_equation_0, values = (var_4041_cast_fp16, var_4443_cast_fp16))[name = string("op_4535_cast_fp16")]; + string var_4537_equation_0 = const()[name = string("op_4537_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4537_cast_fp16 = einsum(equation = var_4537_equation_0, values = (var_4045_cast_fp16, var_4444_cast_fp16))[name = string("op_4537_cast_fp16")]; + string var_4539_equation_0 = const()[name = string("op_4539_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4539_cast_fp16 = einsum(equation = var_4539_equation_0, values = (var_4045_cast_fp16, var_4445_cast_fp16))[name = string("op_4539_cast_fp16")]; + string var_4541_equation_0 = const()[name = string("op_4541_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4541_cast_fp16 = einsum(equation = var_4541_equation_0, values = (var_4045_cast_fp16, var_4446_cast_fp16))[name = string("op_4541_cast_fp16")]; + string var_4543_equation_0 = const()[name = string("op_4543_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4543_cast_fp16 = einsum(equation = var_4543_equation_0, values = (var_4045_cast_fp16, var_4447_cast_fp16))[name = string("op_4543_cast_fp16")]; + string var_4545_equation_0 = const()[name = string("op_4545_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4545_cast_fp16 = einsum(equation = var_4545_equation_0, values = (var_4049_cast_fp16, var_4448_cast_fp16))[name = string("op_4545_cast_fp16")]; + string var_4547_equation_0 = const()[name = string("op_4547_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4547_cast_fp16 = einsum(equation = var_4547_equation_0, values = (var_4049_cast_fp16, var_4449_cast_fp16))[name = string("op_4547_cast_fp16")]; + string var_4549_equation_0 = const()[name = string("op_4549_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4549_cast_fp16 = einsum(equation = var_4549_equation_0, values = (var_4049_cast_fp16, var_4450_cast_fp16))[name = string("op_4549_cast_fp16")]; + string var_4551_equation_0 = const()[name = string("op_4551_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4551_cast_fp16 = einsum(equation = var_4551_equation_0, values = (var_4049_cast_fp16, var_4451_cast_fp16))[name = string("op_4551_cast_fp16")]; + string var_4553_equation_0 = const()[name = string("op_4553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4553_cast_fp16 = einsum(equation = var_4553_equation_0, values = (var_4053_cast_fp16, var_4452_cast_fp16))[name = string("op_4553_cast_fp16")]; + string var_4555_equation_0 = const()[name = string("op_4555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4555_cast_fp16 = einsum(equation = var_4555_equation_0, values = (var_4053_cast_fp16, var_4453_cast_fp16))[name = string("op_4555_cast_fp16")]; + string var_4557_equation_0 = const()[name = string("op_4557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4557_cast_fp16 = einsum(equation = var_4557_equation_0, values = (var_4053_cast_fp16, var_4454_cast_fp16))[name = string("op_4557_cast_fp16")]; + string var_4559_equation_0 = const()[name = string("op_4559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4559_cast_fp16 = einsum(equation = var_4559_equation_0, values = (var_4053_cast_fp16, var_4455_cast_fp16))[name = string("op_4559_cast_fp16")]; + string var_4561_equation_0 = const()[name = string("op_4561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4561_cast_fp16 = einsum(equation = var_4561_equation_0, values = (var_4057_cast_fp16, var_4456_cast_fp16))[name = string("op_4561_cast_fp16")]; + string var_4563_equation_0 = const()[name = string("op_4563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4563_cast_fp16 = einsum(equation = var_4563_equation_0, values = (var_4057_cast_fp16, var_4457_cast_fp16))[name = string("op_4563_cast_fp16")]; + string var_4565_equation_0 = const()[name = string("op_4565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4565_cast_fp16 = einsum(equation = var_4565_equation_0, values = (var_4057_cast_fp16, var_4458_cast_fp16))[name = string("op_4565_cast_fp16")]; + string var_4567_equation_0 = const()[name = string("op_4567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4567_cast_fp16 = einsum(equation = var_4567_equation_0, values = (var_4057_cast_fp16, var_4459_cast_fp16))[name = string("op_4567_cast_fp16")]; + string var_4569_equation_0 = const()[name = string("op_4569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4569_cast_fp16 = einsum(equation = var_4569_equation_0, values = (var_4061_cast_fp16, var_4460_cast_fp16))[name = string("op_4569_cast_fp16")]; + string var_4571_equation_0 = const()[name = string("op_4571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4571_cast_fp16 = einsum(equation = var_4571_equation_0, values = (var_4061_cast_fp16, var_4461_cast_fp16))[name = string("op_4571_cast_fp16")]; + string var_4573_equation_0 = const()[name = string("op_4573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4573_cast_fp16 = einsum(equation = var_4573_equation_0, values = (var_4061_cast_fp16, var_4462_cast_fp16))[name = string("op_4573_cast_fp16")]; + string var_4575_equation_0 = const()[name = string("op_4575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4575_cast_fp16 = einsum(equation = var_4575_equation_0, values = (var_4061_cast_fp16, var_4463_cast_fp16))[name = string("op_4575_cast_fp16")]; + string var_4577_equation_0 = const()[name = string("op_4577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4577_cast_fp16 = einsum(equation = var_4577_equation_0, values = (var_4065_cast_fp16, var_4464_cast_fp16))[name = string("op_4577_cast_fp16")]; + string var_4579_equation_0 = const()[name = string("op_4579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4579_cast_fp16 = einsum(equation = var_4579_equation_0, values = (var_4065_cast_fp16, var_4465_cast_fp16))[name = string("op_4579_cast_fp16")]; + string var_4581_equation_0 = const()[name = string("op_4581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4581_cast_fp16 = einsum(equation = var_4581_equation_0, values = (var_4065_cast_fp16, var_4466_cast_fp16))[name = string("op_4581_cast_fp16")]; + string var_4583_equation_0 = const()[name = string("op_4583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4583_cast_fp16 = einsum(equation = var_4583_equation_0, values = (var_4065_cast_fp16, var_4467_cast_fp16))[name = string("op_4583_cast_fp16")]; + string var_4585_equation_0 = const()[name = string("op_4585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4585_cast_fp16 = einsum(equation = var_4585_equation_0, values = (var_4069_cast_fp16, var_4468_cast_fp16))[name = string("op_4585_cast_fp16")]; + string var_4587_equation_0 = const()[name = string("op_4587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4587_cast_fp16 = einsum(equation = var_4587_equation_0, values = (var_4069_cast_fp16, var_4469_cast_fp16))[name = string("op_4587_cast_fp16")]; + string var_4589_equation_0 = const()[name = string("op_4589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4589_cast_fp16 = einsum(equation = var_4589_equation_0, values = (var_4069_cast_fp16, var_4470_cast_fp16))[name = string("op_4589_cast_fp16")]; + string var_4591_equation_0 = const()[name = string("op_4591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4591_cast_fp16 = einsum(equation = var_4591_equation_0, values = (var_4069_cast_fp16, var_4471_cast_fp16))[name = string("op_4591_cast_fp16")]; + string var_4593_equation_0 = const()[name = string("op_4593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4593_cast_fp16 = einsum(equation = var_4593_equation_0, values = (var_4073_cast_fp16, var_4472_cast_fp16))[name = string("op_4593_cast_fp16")]; + string var_4595_equation_0 = const()[name = string("op_4595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4595_cast_fp16 = einsum(equation = var_4595_equation_0, values = (var_4073_cast_fp16, var_4473_cast_fp16))[name = string("op_4595_cast_fp16")]; + string var_4597_equation_0 = const()[name = string("op_4597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4597_cast_fp16 = einsum(equation = var_4597_equation_0, values = (var_4073_cast_fp16, var_4474_cast_fp16))[name = string("op_4597_cast_fp16")]; + string var_4599_equation_0 = const()[name = string("op_4599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4599_cast_fp16 = einsum(equation = var_4599_equation_0, values = (var_4073_cast_fp16, var_4475_cast_fp16))[name = string("op_4599_cast_fp16")]; + string var_4601_equation_0 = const()[name = string("op_4601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4601_cast_fp16 = einsum(equation = var_4601_equation_0, values = (var_4077_cast_fp16, var_4476_cast_fp16))[name = string("op_4601_cast_fp16")]; + string var_4603_equation_0 = const()[name = string("op_4603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4603_cast_fp16 = einsum(equation = var_4603_equation_0, values = (var_4077_cast_fp16, var_4477_cast_fp16))[name = string("op_4603_cast_fp16")]; + string var_4605_equation_0 = const()[name = string("op_4605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4605_cast_fp16 = einsum(equation = var_4605_equation_0, values = (var_4077_cast_fp16, var_4478_cast_fp16))[name = string("op_4605_cast_fp16")]; + string var_4607_equation_0 = const()[name = string("op_4607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4607_cast_fp16 = einsum(equation = var_4607_equation_0, values = (var_4077_cast_fp16, var_4479_cast_fp16))[name = string("op_4607_cast_fp16")]; + string var_4609_equation_0 = const()[name = string("op_4609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4609_cast_fp16 = einsum(equation = var_4609_equation_0, values = (var_4081_cast_fp16, var_4480_cast_fp16))[name = string("op_4609_cast_fp16")]; + string var_4611_equation_0 = const()[name = string("op_4611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4611_cast_fp16 = einsum(equation = var_4611_equation_0, values = (var_4081_cast_fp16, var_4481_cast_fp16))[name = string("op_4611_cast_fp16")]; + string var_4613_equation_0 = const()[name = string("op_4613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4613_cast_fp16 = einsum(equation = var_4613_equation_0, values = (var_4081_cast_fp16, var_4482_cast_fp16))[name = string("op_4613_cast_fp16")]; + string var_4615_equation_0 = const()[name = string("op_4615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4615_cast_fp16 = einsum(equation = var_4615_equation_0, values = (var_4081_cast_fp16, var_4483_cast_fp16))[name = string("op_4615_cast_fp16")]; + string var_4617_equation_0 = const()[name = string("op_4617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4617_cast_fp16 = einsum(equation = var_4617_equation_0, values = (var_4085_cast_fp16, var_4484_cast_fp16))[name = string("op_4617_cast_fp16")]; + string var_4619_equation_0 = const()[name = string("op_4619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4619_cast_fp16 = einsum(equation = var_4619_equation_0, values = (var_4085_cast_fp16, var_4485_cast_fp16))[name = string("op_4619_cast_fp16")]; + string var_4621_equation_0 = const()[name = string("op_4621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4621_cast_fp16 = einsum(equation = var_4621_equation_0, values = (var_4085_cast_fp16, var_4486_cast_fp16))[name = string("op_4621_cast_fp16")]; + string var_4623_equation_0 = const()[name = string("op_4623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4623_cast_fp16 = einsum(equation = var_4623_equation_0, values = (var_4085_cast_fp16, var_4487_cast_fp16))[name = string("op_4623_cast_fp16")]; + string var_4625_equation_0 = const()[name = string("op_4625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4625_cast_fp16 = einsum(equation = var_4625_equation_0, values = (var_4089_cast_fp16, var_4488_cast_fp16))[name = string("op_4625_cast_fp16")]; + string var_4627_equation_0 = const()[name = string("op_4627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4627_cast_fp16 = einsum(equation = var_4627_equation_0, values = (var_4089_cast_fp16, var_4489_cast_fp16))[name = string("op_4627_cast_fp16")]; + string var_4629_equation_0 = const()[name = string("op_4629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4629_cast_fp16 = einsum(equation = var_4629_equation_0, values = (var_4089_cast_fp16, var_4490_cast_fp16))[name = string("op_4629_cast_fp16")]; + string var_4631_equation_0 = const()[name = string("op_4631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4631_cast_fp16 = einsum(equation = var_4631_equation_0, values = (var_4089_cast_fp16, var_4491_cast_fp16))[name = string("op_4631_cast_fp16")]; + string var_4633_equation_0 = const()[name = string("op_4633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4633_cast_fp16 = einsum(equation = var_4633_equation_0, values = (var_4093_cast_fp16, var_4492_cast_fp16))[name = string("op_4633_cast_fp16")]; + string var_4635_equation_0 = const()[name = string("op_4635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4635_cast_fp16 = einsum(equation = var_4635_equation_0, values = (var_4093_cast_fp16, var_4493_cast_fp16))[name = string("op_4635_cast_fp16")]; + string var_4637_equation_0 = const()[name = string("op_4637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4637_cast_fp16 = einsum(equation = var_4637_equation_0, values = (var_4093_cast_fp16, var_4494_cast_fp16))[name = string("op_4637_cast_fp16")]; + string var_4639_equation_0 = const()[name = string("op_4639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4639_cast_fp16 = einsum(equation = var_4639_equation_0, values = (var_4093_cast_fp16, var_4495_cast_fp16))[name = string("op_4639_cast_fp16")]; + string var_4641_equation_0 = const()[name = string("op_4641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4641_cast_fp16 = einsum(equation = var_4641_equation_0, values = (var_4097_cast_fp16, var_4496_cast_fp16))[name = string("op_4641_cast_fp16")]; + string var_4643_equation_0 = const()[name = string("op_4643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4643_cast_fp16 = einsum(equation = var_4643_equation_0, values = (var_4097_cast_fp16, var_4497_cast_fp16))[name = string("op_4643_cast_fp16")]; + string var_4645_equation_0 = const()[name = string("op_4645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4645_cast_fp16 = einsum(equation = var_4645_equation_0, values = (var_4097_cast_fp16, var_4498_cast_fp16))[name = string("op_4645_cast_fp16")]; + string var_4647_equation_0 = const()[name = string("op_4647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4647_cast_fp16 = einsum(equation = var_4647_equation_0, values = (var_4097_cast_fp16, var_4499_cast_fp16))[name = string("op_4647_cast_fp16")]; + string var_4649_equation_0 = const()[name = string("op_4649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4649_cast_fp16 = einsum(equation = var_4649_equation_0, values = (var_4101_cast_fp16, var_4500_cast_fp16))[name = string("op_4649_cast_fp16")]; + string var_4651_equation_0 = const()[name = string("op_4651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4651_cast_fp16 = einsum(equation = var_4651_equation_0, values = (var_4101_cast_fp16, var_4501_cast_fp16))[name = string("op_4651_cast_fp16")]; + string var_4653_equation_0 = const()[name = string("op_4653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4653_cast_fp16 = einsum(equation = var_4653_equation_0, values = (var_4101_cast_fp16, var_4502_cast_fp16))[name = string("op_4653_cast_fp16")]; + string var_4655_equation_0 = const()[name = string("op_4655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4655_cast_fp16 = einsum(equation = var_4655_equation_0, values = (var_4101_cast_fp16, var_4503_cast_fp16))[name = string("op_4655_cast_fp16")]; + string var_4657_equation_0 = const()[name = string("op_4657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4657_cast_fp16 = einsum(equation = var_4657_equation_0, values = (var_4105_cast_fp16, var_4504_cast_fp16))[name = string("op_4657_cast_fp16")]; + string var_4659_equation_0 = const()[name = string("op_4659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4659_cast_fp16 = einsum(equation = var_4659_equation_0, values = (var_4105_cast_fp16, var_4505_cast_fp16))[name = string("op_4659_cast_fp16")]; + string var_4661_equation_0 = const()[name = string("op_4661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4661_cast_fp16 = einsum(equation = var_4661_equation_0, values = (var_4105_cast_fp16, var_4506_cast_fp16))[name = string("op_4661_cast_fp16")]; + string var_4663_equation_0 = const()[name = string("op_4663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4663_cast_fp16 = einsum(equation = var_4663_equation_0, values = (var_4105_cast_fp16, var_4507_cast_fp16))[name = string("op_4663_cast_fp16")]; + string var_4665_equation_0 = const()[name = string("op_4665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4665_cast_fp16 = einsum(equation = var_4665_equation_0, values = (var_4109_cast_fp16, var_4508_cast_fp16))[name = string("op_4665_cast_fp16")]; + string var_4667_equation_0 = const()[name = string("op_4667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4667_cast_fp16 = einsum(equation = var_4667_equation_0, values = (var_4109_cast_fp16, var_4509_cast_fp16))[name = string("op_4667_cast_fp16")]; + string var_4669_equation_0 = const()[name = string("op_4669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4669_cast_fp16 = einsum(equation = var_4669_equation_0, values = (var_4109_cast_fp16, var_4510_cast_fp16))[name = string("op_4669_cast_fp16")]; + string var_4671_equation_0 = const()[name = string("op_4671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4671_cast_fp16 = einsum(equation = var_4671_equation_0, values = (var_4109_cast_fp16, var_4511_cast_fp16))[name = string("op_4671_cast_fp16")]; + bool var_4673_interleave_0 = const()[name = string("op_4673_interleave_0"), val = bool(false)]; + tensor var_4673_cast_fp16 = concat(axis = var_3232, interleave = var_4673_interleave_0, values = (var_4513_cast_fp16, var_4515_cast_fp16, var_4517_cast_fp16, var_4519_cast_fp16))[name = string("op_4673_cast_fp16")]; + bool var_4675_interleave_0 = const()[name = string("op_4675_interleave_0"), val = bool(false)]; + tensor var_4675_cast_fp16 = concat(axis = var_3232, interleave = var_4675_interleave_0, values = (var_4521_cast_fp16, var_4523_cast_fp16, var_4525_cast_fp16, var_4527_cast_fp16))[name = string("op_4675_cast_fp16")]; + bool var_4677_interleave_0 = const()[name = string("op_4677_interleave_0"), val = bool(false)]; + tensor var_4677_cast_fp16 = concat(axis = var_3232, interleave = var_4677_interleave_0, values = (var_4529_cast_fp16, var_4531_cast_fp16, var_4533_cast_fp16, var_4535_cast_fp16))[name = string("op_4677_cast_fp16")]; + bool var_4679_interleave_0 = const()[name = string("op_4679_interleave_0"), val = bool(false)]; + tensor var_4679_cast_fp16 = concat(axis = var_3232, interleave = var_4679_interleave_0, values = (var_4537_cast_fp16, var_4539_cast_fp16, var_4541_cast_fp16, var_4543_cast_fp16))[name = string("op_4679_cast_fp16")]; + bool var_4681_interleave_0 = const()[name = string("op_4681_interleave_0"), val = bool(false)]; + tensor var_4681_cast_fp16 = concat(axis = var_3232, interleave = var_4681_interleave_0, values = (var_4545_cast_fp16, var_4547_cast_fp16, var_4549_cast_fp16, var_4551_cast_fp16))[name = string("op_4681_cast_fp16")]; + bool var_4683_interleave_0 = const()[name = string("op_4683_interleave_0"), val = bool(false)]; + tensor var_4683_cast_fp16 = concat(axis = var_3232, interleave = var_4683_interleave_0, values = (var_4553_cast_fp16, var_4555_cast_fp16, var_4557_cast_fp16, var_4559_cast_fp16))[name = string("op_4683_cast_fp16")]; + bool var_4685_interleave_0 = const()[name = string("op_4685_interleave_0"), val = bool(false)]; + tensor var_4685_cast_fp16 = concat(axis = var_3232, interleave = var_4685_interleave_0, values = (var_4561_cast_fp16, var_4563_cast_fp16, var_4565_cast_fp16, var_4567_cast_fp16))[name = string("op_4685_cast_fp16")]; + bool var_4687_interleave_0 = const()[name = string("op_4687_interleave_0"), val = bool(false)]; + tensor var_4687_cast_fp16 = concat(axis = var_3232, interleave = var_4687_interleave_0, values = (var_4569_cast_fp16, var_4571_cast_fp16, var_4573_cast_fp16, var_4575_cast_fp16))[name = string("op_4687_cast_fp16")]; + bool var_4689_interleave_0 = const()[name = string("op_4689_interleave_0"), val = bool(false)]; + tensor var_4689_cast_fp16 = concat(axis = var_3232, interleave = var_4689_interleave_0, values = (var_4577_cast_fp16, var_4579_cast_fp16, var_4581_cast_fp16, var_4583_cast_fp16))[name = string("op_4689_cast_fp16")]; + bool var_4691_interleave_0 = const()[name = string("op_4691_interleave_0"), val = bool(false)]; + tensor var_4691_cast_fp16 = concat(axis = var_3232, interleave = var_4691_interleave_0, values = (var_4585_cast_fp16, var_4587_cast_fp16, var_4589_cast_fp16, var_4591_cast_fp16))[name = string("op_4691_cast_fp16")]; + bool var_4693_interleave_0 = const()[name = string("op_4693_interleave_0"), val = bool(false)]; + tensor var_4693_cast_fp16 = concat(axis = var_3232, interleave = var_4693_interleave_0, values = (var_4593_cast_fp16, var_4595_cast_fp16, var_4597_cast_fp16, var_4599_cast_fp16))[name = string("op_4693_cast_fp16")]; + bool var_4695_interleave_0 = const()[name = string("op_4695_interleave_0"), val = bool(false)]; + tensor var_4695_cast_fp16 = concat(axis = var_3232, interleave = var_4695_interleave_0, values = (var_4601_cast_fp16, var_4603_cast_fp16, var_4605_cast_fp16, var_4607_cast_fp16))[name = string("op_4695_cast_fp16")]; + bool var_4697_interleave_0 = const()[name = string("op_4697_interleave_0"), val = bool(false)]; + tensor var_4697_cast_fp16 = concat(axis = var_3232, interleave = var_4697_interleave_0, values = (var_4609_cast_fp16, var_4611_cast_fp16, var_4613_cast_fp16, var_4615_cast_fp16))[name = string("op_4697_cast_fp16")]; + bool var_4699_interleave_0 = const()[name = string("op_4699_interleave_0"), val = bool(false)]; + tensor var_4699_cast_fp16 = concat(axis = var_3232, interleave = var_4699_interleave_0, values = (var_4617_cast_fp16, var_4619_cast_fp16, var_4621_cast_fp16, var_4623_cast_fp16))[name = string("op_4699_cast_fp16")]; + bool var_4701_interleave_0 = const()[name = string("op_4701_interleave_0"), val = bool(false)]; + tensor var_4701_cast_fp16 = concat(axis = var_3232, interleave = var_4701_interleave_0, values = (var_4625_cast_fp16, var_4627_cast_fp16, var_4629_cast_fp16, var_4631_cast_fp16))[name = string("op_4701_cast_fp16")]; + bool var_4703_interleave_0 = const()[name = string("op_4703_interleave_0"), val = bool(false)]; + tensor var_4703_cast_fp16 = concat(axis = var_3232, interleave = var_4703_interleave_0, values = (var_4633_cast_fp16, var_4635_cast_fp16, var_4637_cast_fp16, var_4639_cast_fp16))[name = string("op_4703_cast_fp16")]; + bool var_4705_interleave_0 = const()[name = string("op_4705_interleave_0"), val = bool(false)]; + tensor var_4705_cast_fp16 = concat(axis = var_3232, interleave = var_4705_interleave_0, values = (var_4641_cast_fp16, var_4643_cast_fp16, var_4645_cast_fp16, var_4647_cast_fp16))[name = string("op_4705_cast_fp16")]; + bool var_4707_interleave_0 = const()[name = string("op_4707_interleave_0"), val = bool(false)]; + tensor var_4707_cast_fp16 = concat(axis = var_3232, interleave = var_4707_interleave_0, values = (var_4649_cast_fp16, var_4651_cast_fp16, var_4653_cast_fp16, var_4655_cast_fp16))[name = string("op_4707_cast_fp16")]; + bool var_4709_interleave_0 = const()[name = string("op_4709_interleave_0"), val = bool(false)]; + tensor var_4709_cast_fp16 = concat(axis = var_3232, interleave = var_4709_interleave_0, values = (var_4657_cast_fp16, var_4659_cast_fp16, var_4661_cast_fp16, var_4663_cast_fp16))[name = string("op_4709_cast_fp16")]; + bool var_4711_interleave_0 = const()[name = string("op_4711_interleave_0"), val = bool(false)]; + tensor var_4711_cast_fp16 = concat(axis = var_3232, interleave = var_4711_interleave_0, values = (var_4665_cast_fp16, var_4667_cast_fp16, var_4669_cast_fp16, var_4671_cast_fp16))[name = string("op_4711_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_3257, interleave = input_17_interleave_0, values = (var_4673_cast_fp16, var_4675_cast_fp16, var_4677_cast_fp16, var_4679_cast_fp16, var_4681_cast_fp16, var_4683_cast_fp16, var_4685_cast_fp16, var_4687_cast_fp16, var_4689_cast_fp16, var_4691_cast_fp16, var_4693_cast_fp16, var_4695_cast_fp16, var_4697_cast_fp16, var_4699_cast_fp16, var_4701_cast_fp16, var_4703_cast_fp16, var_4705_cast_fp16, var_4707_cast_fp16, var_4709_cast_fp16, var_4711_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103211840)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106488704)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_4730_to_fp16 = const()[name = string("op_4730_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_4730_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106491328)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106493952)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106496576)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119603840)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119614144)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132721408)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_4759 = const()[name = string("op_4759"), val = int32(3)]; + int32 var_4784 = const()[name = string("op_4784"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_4801_to_fp16 = const()[name = string("op_4801_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_4801_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132724032)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132726656)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132729280)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136006144)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136008768)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139285632)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142562496)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_4839_begin_0 = const()[name = string("op_4839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4839_end_0 = const()[name = string("op_4839_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4839_end_mask_0 = const()[name = string("op_4839_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4839_cast_fp16 = slice_by_index(begin = var_4839_begin_0, end = var_4839_end_0, end_mask = var_4839_end_mask_0, x = query_7_cast_fp16)[name = string("op_4839_cast_fp16")]; + tensor var_4843_begin_0 = const()[name = string("op_4843_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4843_end_0 = const()[name = string("op_4843_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4843_end_mask_0 = const()[name = string("op_4843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4843_cast_fp16 = slice_by_index(begin = var_4843_begin_0, end = var_4843_end_0, end_mask = var_4843_end_mask_0, x = query_7_cast_fp16)[name = string("op_4843_cast_fp16")]; + tensor var_4847_begin_0 = const()[name = string("op_4847_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4847_end_0 = const()[name = string("op_4847_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4847_end_mask_0 = const()[name = string("op_4847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4847_cast_fp16 = slice_by_index(begin = var_4847_begin_0, end = var_4847_end_0, end_mask = var_4847_end_mask_0, x = query_7_cast_fp16)[name = string("op_4847_cast_fp16")]; + tensor var_4851_begin_0 = const()[name = string("op_4851_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4851_end_0 = const()[name = string("op_4851_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4851_end_mask_0 = const()[name = string("op_4851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4851_cast_fp16 = slice_by_index(begin = var_4851_begin_0, end = var_4851_end_0, end_mask = var_4851_end_mask_0, x = query_7_cast_fp16)[name = string("op_4851_cast_fp16")]; + tensor var_4855_begin_0 = const()[name = string("op_4855_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4855_end_0 = const()[name = string("op_4855_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4855_end_mask_0 = const()[name = string("op_4855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4855_cast_fp16 = slice_by_index(begin = var_4855_begin_0, end = var_4855_end_0, end_mask = var_4855_end_mask_0, x = query_7_cast_fp16)[name = string("op_4855_cast_fp16")]; + tensor var_4859_begin_0 = const()[name = string("op_4859_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4859_end_0 = const()[name = string("op_4859_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4859_end_mask_0 = const()[name = string("op_4859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4859_cast_fp16 = slice_by_index(begin = var_4859_begin_0, end = var_4859_end_0, end_mask = var_4859_end_mask_0, x = query_7_cast_fp16)[name = string("op_4859_cast_fp16")]; + tensor var_4863_begin_0 = const()[name = string("op_4863_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4863_end_0 = const()[name = string("op_4863_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4863_end_mask_0 = const()[name = string("op_4863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4863_cast_fp16 = slice_by_index(begin = var_4863_begin_0, end = var_4863_end_0, end_mask = var_4863_end_mask_0, x = query_7_cast_fp16)[name = string("op_4863_cast_fp16")]; + tensor var_4867_begin_0 = const()[name = string("op_4867_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4867_end_0 = const()[name = string("op_4867_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4867_end_mask_0 = const()[name = string("op_4867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4867_cast_fp16 = slice_by_index(begin = var_4867_begin_0, end = var_4867_end_0, end_mask = var_4867_end_mask_0, x = query_7_cast_fp16)[name = string("op_4867_cast_fp16")]; + tensor var_4871_begin_0 = const()[name = string("op_4871_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4871_end_0 = const()[name = string("op_4871_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4871_end_mask_0 = const()[name = string("op_4871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4871_cast_fp16 = slice_by_index(begin = var_4871_begin_0, end = var_4871_end_0, end_mask = var_4871_end_mask_0, x = query_7_cast_fp16)[name = string("op_4871_cast_fp16")]; + tensor var_4875_begin_0 = const()[name = string("op_4875_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4875_end_0 = const()[name = string("op_4875_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4875_end_mask_0 = const()[name = string("op_4875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4875_cast_fp16 = slice_by_index(begin = var_4875_begin_0, end = var_4875_end_0, end_mask = var_4875_end_mask_0, x = query_7_cast_fp16)[name = string("op_4875_cast_fp16")]; + tensor var_4879_begin_0 = const()[name = string("op_4879_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4879_end_0 = const()[name = string("op_4879_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4879_end_mask_0 = const()[name = string("op_4879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4879_cast_fp16 = slice_by_index(begin = var_4879_begin_0, end = var_4879_end_0, end_mask = var_4879_end_mask_0, x = query_7_cast_fp16)[name = string("op_4879_cast_fp16")]; + tensor var_4883_begin_0 = const()[name = string("op_4883_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4883_end_0 = const()[name = string("op_4883_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4883_end_mask_0 = const()[name = string("op_4883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4883_cast_fp16 = slice_by_index(begin = var_4883_begin_0, end = var_4883_end_0, end_mask = var_4883_end_mask_0, x = query_7_cast_fp16)[name = string("op_4883_cast_fp16")]; + tensor var_4887_begin_0 = const()[name = string("op_4887_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4887_end_0 = const()[name = string("op_4887_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_4887_end_mask_0 = const()[name = string("op_4887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4887_cast_fp16 = slice_by_index(begin = var_4887_begin_0, end = var_4887_end_0, end_mask = var_4887_end_mask_0, x = query_7_cast_fp16)[name = string("op_4887_cast_fp16")]; + tensor var_4891_begin_0 = const()[name = string("op_4891_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4891_end_0 = const()[name = string("op_4891_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_4891_end_mask_0 = const()[name = string("op_4891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4891_cast_fp16 = slice_by_index(begin = var_4891_begin_0, end = var_4891_end_0, end_mask = var_4891_end_mask_0, x = query_7_cast_fp16)[name = string("op_4891_cast_fp16")]; + tensor var_4895_begin_0 = const()[name = string("op_4895_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4895_end_0 = const()[name = string("op_4895_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_4895_end_mask_0 = const()[name = string("op_4895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4895_cast_fp16 = slice_by_index(begin = var_4895_begin_0, end = var_4895_end_0, end_mask = var_4895_end_mask_0, x = query_7_cast_fp16)[name = string("op_4895_cast_fp16")]; + tensor var_4899_begin_0 = const()[name = string("op_4899_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4899_end_0 = const()[name = string("op_4899_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_4899_end_mask_0 = const()[name = string("op_4899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4899_cast_fp16 = slice_by_index(begin = var_4899_begin_0, end = var_4899_end_0, end_mask = var_4899_end_mask_0, x = query_7_cast_fp16)[name = string("op_4899_cast_fp16")]; + tensor var_4903_begin_0 = const()[name = string("op_4903_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4903_end_0 = const()[name = string("op_4903_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_4903_end_mask_0 = const()[name = string("op_4903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4903_cast_fp16 = slice_by_index(begin = var_4903_begin_0, end = var_4903_end_0, end_mask = var_4903_end_mask_0, x = query_7_cast_fp16)[name = string("op_4903_cast_fp16")]; + tensor var_4907_begin_0 = const()[name = string("op_4907_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4907_end_0 = const()[name = string("op_4907_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_4907_end_mask_0 = const()[name = string("op_4907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4907_cast_fp16 = slice_by_index(begin = var_4907_begin_0, end = var_4907_end_0, end_mask = var_4907_end_mask_0, x = query_7_cast_fp16)[name = string("op_4907_cast_fp16")]; + tensor var_4911_begin_0 = const()[name = string("op_4911_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4911_end_0 = const()[name = string("op_4911_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_4911_end_mask_0 = const()[name = string("op_4911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4911_cast_fp16 = slice_by_index(begin = var_4911_begin_0, end = var_4911_end_0, end_mask = var_4911_end_mask_0, x = query_7_cast_fp16)[name = string("op_4911_cast_fp16")]; + tensor var_4915_begin_0 = const()[name = string("op_4915_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4915_end_0 = const()[name = string("op_4915_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_4915_end_mask_0 = const()[name = string("op_4915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4915_cast_fp16 = slice_by_index(begin = var_4915_begin_0, end = var_4915_end_0, end_mask = var_4915_end_mask_0, x = query_7_cast_fp16)[name = string("op_4915_cast_fp16")]; + tensor var_4924_begin_0 = const()[name = string("op_4924_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4924_end_0 = const()[name = string("op_4924_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4924_end_mask_0 = const()[name = string("op_4924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4924_cast_fp16 = slice_by_index(begin = var_4924_begin_0, end = var_4924_end_0, end_mask = var_4924_end_mask_0, x = var_4839_cast_fp16)[name = string("op_4924_cast_fp16")]; + tensor var_4931_begin_0 = const()[name = string("op_4931_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4931_end_0 = const()[name = string("op_4931_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4931_end_mask_0 = const()[name = string("op_4931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4931_cast_fp16 = slice_by_index(begin = var_4931_begin_0, end = var_4931_end_0, end_mask = var_4931_end_mask_0, x = var_4839_cast_fp16)[name = string("op_4931_cast_fp16")]; + tensor var_4938_begin_0 = const()[name = string("op_4938_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4938_end_0 = const()[name = string("op_4938_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4938_end_mask_0 = const()[name = string("op_4938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4938_cast_fp16 = slice_by_index(begin = var_4938_begin_0, end = var_4938_end_0, end_mask = var_4938_end_mask_0, x = var_4839_cast_fp16)[name = string("op_4938_cast_fp16")]; + tensor var_4945_begin_0 = const()[name = string("op_4945_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4945_end_0 = const()[name = string("op_4945_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4945_end_mask_0 = const()[name = string("op_4945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4945_cast_fp16 = slice_by_index(begin = var_4945_begin_0, end = var_4945_end_0, end_mask = var_4945_end_mask_0, x = var_4839_cast_fp16)[name = string("op_4945_cast_fp16")]; + tensor var_4952_begin_0 = const()[name = string("op_4952_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4952_end_0 = const()[name = string("op_4952_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4952_end_mask_0 = const()[name = string("op_4952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4952_cast_fp16 = slice_by_index(begin = var_4952_begin_0, end = var_4952_end_0, end_mask = var_4952_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4952_cast_fp16")]; + tensor var_4959_begin_0 = const()[name = string("op_4959_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4959_end_0 = const()[name = string("op_4959_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4959_end_mask_0 = const()[name = string("op_4959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4959_cast_fp16 = slice_by_index(begin = var_4959_begin_0, end = var_4959_end_0, end_mask = var_4959_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4959_cast_fp16")]; + tensor var_4966_begin_0 = const()[name = string("op_4966_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4966_end_0 = const()[name = string("op_4966_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4966_end_mask_0 = const()[name = string("op_4966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4966_cast_fp16 = slice_by_index(begin = var_4966_begin_0, end = var_4966_end_0, end_mask = var_4966_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4966_cast_fp16")]; + tensor var_4973_begin_0 = const()[name = string("op_4973_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4973_end_0 = const()[name = string("op_4973_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4973_end_mask_0 = const()[name = string("op_4973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4973_cast_fp16 = slice_by_index(begin = var_4973_begin_0, end = var_4973_end_0, end_mask = var_4973_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4973_cast_fp16")]; + tensor var_4980_begin_0 = const()[name = string("op_4980_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4980_end_0 = const()[name = string("op_4980_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4980_end_mask_0 = const()[name = string("op_4980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4980_cast_fp16 = slice_by_index(begin = var_4980_begin_0, end = var_4980_end_0, end_mask = var_4980_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4980_cast_fp16")]; + tensor var_4987_begin_0 = const()[name = string("op_4987_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4987_end_0 = const()[name = string("op_4987_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4987_end_mask_0 = const()[name = string("op_4987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4987_cast_fp16 = slice_by_index(begin = var_4987_begin_0, end = var_4987_end_0, end_mask = var_4987_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4987_cast_fp16")]; + tensor var_4994_begin_0 = const()[name = string("op_4994_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4994_end_0 = const()[name = string("op_4994_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4994_end_mask_0 = const()[name = string("op_4994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4994_cast_fp16 = slice_by_index(begin = var_4994_begin_0, end = var_4994_end_0, end_mask = var_4994_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4994_cast_fp16")]; + tensor var_5001_begin_0 = const()[name = string("op_5001_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5001_end_0 = const()[name = string("op_5001_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5001_end_mask_0 = const()[name = string("op_5001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5001_cast_fp16 = slice_by_index(begin = var_5001_begin_0, end = var_5001_end_0, end_mask = var_5001_end_mask_0, x = var_4847_cast_fp16)[name = string("op_5001_cast_fp16")]; + tensor var_5008_begin_0 = const()[name = string("op_5008_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5008_end_0 = const()[name = string("op_5008_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5008_end_mask_0 = const()[name = string("op_5008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5008_cast_fp16 = slice_by_index(begin = var_5008_begin_0, end = var_5008_end_0, end_mask = var_5008_end_mask_0, x = var_4851_cast_fp16)[name = string("op_5008_cast_fp16")]; + tensor var_5015_begin_0 = const()[name = string("op_5015_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5015_end_0 = const()[name = string("op_5015_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5015_end_mask_0 = const()[name = string("op_5015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5015_cast_fp16 = slice_by_index(begin = var_5015_begin_0, end = var_5015_end_0, end_mask = var_5015_end_mask_0, x = var_4851_cast_fp16)[name = string("op_5015_cast_fp16")]; + tensor var_5022_begin_0 = const()[name = string("op_5022_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5022_end_0 = const()[name = string("op_5022_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5022_end_mask_0 = const()[name = string("op_5022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5022_cast_fp16 = slice_by_index(begin = var_5022_begin_0, end = var_5022_end_0, end_mask = var_5022_end_mask_0, x = var_4851_cast_fp16)[name = string("op_5022_cast_fp16")]; + tensor var_5029_begin_0 = const()[name = string("op_5029_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5029_end_0 = const()[name = string("op_5029_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5029_end_mask_0 = const()[name = string("op_5029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5029_cast_fp16 = slice_by_index(begin = var_5029_begin_0, end = var_5029_end_0, end_mask = var_5029_end_mask_0, x = var_4851_cast_fp16)[name = string("op_5029_cast_fp16")]; + tensor var_5036_begin_0 = const()[name = string("op_5036_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5036_end_0 = const()[name = string("op_5036_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5036_end_mask_0 = const()[name = string("op_5036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5036_cast_fp16 = slice_by_index(begin = var_5036_begin_0, end = var_5036_end_0, end_mask = var_5036_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5036_cast_fp16")]; + tensor var_5043_begin_0 = const()[name = string("op_5043_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5043_end_0 = const()[name = string("op_5043_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5043_end_mask_0 = const()[name = string("op_5043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5043_cast_fp16 = slice_by_index(begin = var_5043_begin_0, end = var_5043_end_0, end_mask = var_5043_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5043_cast_fp16")]; + tensor var_5050_begin_0 = const()[name = string("op_5050_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5050_end_0 = const()[name = string("op_5050_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5050_end_mask_0 = const()[name = string("op_5050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5050_cast_fp16 = slice_by_index(begin = var_5050_begin_0, end = var_5050_end_0, end_mask = var_5050_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5050_cast_fp16")]; + tensor var_5057_begin_0 = const()[name = string("op_5057_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5057_end_0 = const()[name = string("op_5057_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5057_end_mask_0 = const()[name = string("op_5057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5057_cast_fp16 = slice_by_index(begin = var_5057_begin_0, end = var_5057_end_0, end_mask = var_5057_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5057_cast_fp16")]; + tensor var_5064_begin_0 = const()[name = string("op_5064_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5064_end_0 = const()[name = string("op_5064_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5064_end_mask_0 = const()[name = string("op_5064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5064_cast_fp16 = slice_by_index(begin = var_5064_begin_0, end = var_5064_end_0, end_mask = var_5064_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5064_cast_fp16")]; + tensor var_5071_begin_0 = const()[name = string("op_5071_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5071_end_0 = const()[name = string("op_5071_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5071_end_mask_0 = const()[name = string("op_5071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5071_cast_fp16 = slice_by_index(begin = var_5071_begin_0, end = var_5071_end_0, end_mask = var_5071_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5071_cast_fp16")]; + tensor var_5078_begin_0 = const()[name = string("op_5078_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5078_end_0 = const()[name = string("op_5078_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5078_end_mask_0 = const()[name = string("op_5078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5078_cast_fp16 = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5078_cast_fp16")]; + tensor var_5085_begin_0 = const()[name = string("op_5085_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5085_end_0 = const()[name = string("op_5085_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5085_end_mask_0 = const()[name = string("op_5085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5085_cast_fp16 = slice_by_index(begin = var_5085_begin_0, end = var_5085_end_0, end_mask = var_5085_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5085_cast_fp16")]; + tensor var_5092_begin_0 = const()[name = string("op_5092_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5092_end_0 = const()[name = string("op_5092_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5092_end_mask_0 = const()[name = string("op_5092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5092_cast_fp16 = slice_by_index(begin = var_5092_begin_0, end = var_5092_end_0, end_mask = var_5092_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5092_cast_fp16")]; + tensor var_5099_begin_0 = const()[name = string("op_5099_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5099_end_0 = const()[name = string("op_5099_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5099_end_mask_0 = const()[name = string("op_5099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5099_cast_fp16 = slice_by_index(begin = var_5099_begin_0, end = var_5099_end_0, end_mask = var_5099_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5099_cast_fp16")]; + tensor var_5106_begin_0 = const()[name = string("op_5106_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5106_end_0 = const()[name = string("op_5106_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5106_end_mask_0 = const()[name = string("op_5106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5106_cast_fp16 = slice_by_index(begin = var_5106_begin_0, end = var_5106_end_0, end_mask = var_5106_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5106_cast_fp16")]; + tensor var_5113_begin_0 = const()[name = string("op_5113_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5113_end_0 = const()[name = string("op_5113_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5113_end_mask_0 = const()[name = string("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5113_cast_fp16 = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5113_cast_fp16")]; + tensor var_5120_begin_0 = const()[name = string("op_5120_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5120_end_0 = const()[name = string("op_5120_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5120_end_mask_0 = const()[name = string("op_5120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5120_cast_fp16 = slice_by_index(begin = var_5120_begin_0, end = var_5120_end_0, end_mask = var_5120_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5120_cast_fp16")]; + tensor var_5127_begin_0 = const()[name = string("op_5127_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5127_end_0 = const()[name = string("op_5127_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5127_end_mask_0 = const()[name = string("op_5127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5127_cast_fp16 = slice_by_index(begin = var_5127_begin_0, end = var_5127_end_0, end_mask = var_5127_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5127_cast_fp16")]; + tensor var_5134_begin_0 = const()[name = string("op_5134_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5134_end_0 = const()[name = string("op_5134_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5134_end_mask_0 = const()[name = string("op_5134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5134_cast_fp16 = slice_by_index(begin = var_5134_begin_0, end = var_5134_end_0, end_mask = var_5134_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5134_cast_fp16")]; + tensor var_5141_begin_0 = const()[name = string("op_5141_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5141_end_0 = const()[name = string("op_5141_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5141_end_mask_0 = const()[name = string("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5141_cast_fp16 = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5141_cast_fp16")]; + tensor var_5148_begin_0 = const()[name = string("op_5148_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5148_end_0 = const()[name = string("op_5148_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5148_end_mask_0 = const()[name = string("op_5148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5148_cast_fp16 = slice_by_index(begin = var_5148_begin_0, end = var_5148_end_0, end_mask = var_5148_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5148_cast_fp16")]; + tensor var_5155_begin_0 = const()[name = string("op_5155_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5155_end_0 = const()[name = string("op_5155_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5155_end_mask_0 = const()[name = string("op_5155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5155_cast_fp16 = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5155_cast_fp16")]; + tensor var_5162_begin_0 = const()[name = string("op_5162_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5162_end_0 = const()[name = string("op_5162_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5162_end_mask_0 = const()[name = string("op_5162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5162_cast_fp16 = slice_by_index(begin = var_5162_begin_0, end = var_5162_end_0, end_mask = var_5162_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5162_cast_fp16")]; + tensor var_5169_begin_0 = const()[name = string("op_5169_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5169_end_0 = const()[name = string("op_5169_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5169_end_mask_0 = const()[name = string("op_5169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5169_cast_fp16 = slice_by_index(begin = var_5169_begin_0, end = var_5169_end_0, end_mask = var_5169_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5169_cast_fp16")]; + tensor var_5176_begin_0 = const()[name = string("op_5176_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5176_end_0 = const()[name = string("op_5176_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5176_end_mask_0 = const()[name = string("op_5176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5176_cast_fp16 = slice_by_index(begin = var_5176_begin_0, end = var_5176_end_0, end_mask = var_5176_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5176_cast_fp16")]; + tensor var_5183_begin_0 = const()[name = string("op_5183_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5183_end_0 = const()[name = string("op_5183_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5183_end_mask_0 = const()[name = string("op_5183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5183_cast_fp16 = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5183_cast_fp16")]; + tensor var_5190_begin_0 = const()[name = string("op_5190_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5190_end_0 = const()[name = string("op_5190_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5190_end_mask_0 = const()[name = string("op_5190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5190_cast_fp16 = slice_by_index(begin = var_5190_begin_0, end = var_5190_end_0, end_mask = var_5190_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5190_cast_fp16")]; + tensor var_5197_begin_0 = const()[name = string("op_5197_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5197_end_0 = const()[name = string("op_5197_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5197_end_mask_0 = const()[name = string("op_5197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5197_cast_fp16 = slice_by_index(begin = var_5197_begin_0, end = var_5197_end_0, end_mask = var_5197_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5197_cast_fp16")]; + tensor var_5204_begin_0 = const()[name = string("op_5204_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5204_end_0 = const()[name = string("op_5204_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5204_end_mask_0 = const()[name = string("op_5204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5204_cast_fp16 = slice_by_index(begin = var_5204_begin_0, end = var_5204_end_0, end_mask = var_5204_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5204_cast_fp16")]; + tensor var_5211_begin_0 = const()[name = string("op_5211_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5211_end_0 = const()[name = string("op_5211_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5211_end_mask_0 = const()[name = string("op_5211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5211_cast_fp16 = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5211_cast_fp16")]; + tensor var_5218_begin_0 = const()[name = string("op_5218_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5218_end_0 = const()[name = string("op_5218_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5218_end_mask_0 = const()[name = string("op_5218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5218_cast_fp16 = slice_by_index(begin = var_5218_begin_0, end = var_5218_end_0, end_mask = var_5218_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5218_cast_fp16")]; + tensor var_5225_begin_0 = const()[name = string("op_5225_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5225_end_0 = const()[name = string("op_5225_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5225_end_mask_0 = const()[name = string("op_5225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5225_cast_fp16 = slice_by_index(begin = var_5225_begin_0, end = var_5225_end_0, end_mask = var_5225_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5225_cast_fp16")]; + tensor var_5232_begin_0 = const()[name = string("op_5232_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5232_end_0 = const()[name = string("op_5232_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5232_end_mask_0 = const()[name = string("op_5232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5232_cast_fp16 = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5232_cast_fp16")]; + tensor var_5239_begin_0 = const()[name = string("op_5239_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5239_end_0 = const()[name = string("op_5239_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5239_end_mask_0 = const()[name = string("op_5239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5239_cast_fp16")]; + tensor var_5246_begin_0 = const()[name = string("op_5246_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5246_end_0 = const()[name = string("op_5246_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5246_end_mask_0 = const()[name = string("op_5246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5246_cast_fp16 = slice_by_index(begin = var_5246_begin_0, end = var_5246_end_0, end_mask = var_5246_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5246_cast_fp16")]; + tensor var_5253_begin_0 = const()[name = string("op_5253_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5253_end_0 = const()[name = string("op_5253_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5253_end_mask_0 = const()[name = string("op_5253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5253_cast_fp16 = slice_by_index(begin = var_5253_begin_0, end = var_5253_end_0, end_mask = var_5253_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5253_cast_fp16")]; + tensor var_5260_begin_0 = const()[name = string("op_5260_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5260_end_0 = const()[name = string("op_5260_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5260_end_mask_0 = const()[name = string("op_5260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5260_cast_fp16 = slice_by_index(begin = var_5260_begin_0, end = var_5260_end_0, end_mask = var_5260_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5260_cast_fp16")]; + tensor var_5267_begin_0 = const()[name = string("op_5267_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5267_end_0 = const()[name = string("op_5267_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5267_end_mask_0 = const()[name = string("op_5267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5267_cast_fp16 = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5267_cast_fp16")]; + tensor var_5274_begin_0 = const()[name = string("op_5274_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5274_end_0 = const()[name = string("op_5274_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5274_end_mask_0 = const()[name = string("op_5274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5274_cast_fp16 = slice_by_index(begin = var_5274_begin_0, end = var_5274_end_0, end_mask = var_5274_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5274_cast_fp16")]; + tensor var_5281_begin_0 = const()[name = string("op_5281_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5281_end_0 = const()[name = string("op_5281_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5281_end_mask_0 = const()[name = string("op_5281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5281_cast_fp16")]; + tensor var_5288_begin_0 = const()[name = string("op_5288_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5288_end_0 = const()[name = string("op_5288_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5288_end_mask_0 = const()[name = string("op_5288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5288_cast_fp16 = slice_by_index(begin = var_5288_begin_0, end = var_5288_end_0, end_mask = var_5288_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5288_cast_fp16")]; + tensor var_5295_begin_0 = const()[name = string("op_5295_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5295_end_0 = const()[name = string("op_5295_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5295_end_mask_0 = const()[name = string("op_5295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5295_cast_fp16 = slice_by_index(begin = var_5295_begin_0, end = var_5295_end_0, end_mask = var_5295_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5295_cast_fp16")]; + tensor var_5302_begin_0 = const()[name = string("op_5302_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5302_end_0 = const()[name = string("op_5302_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5302_end_mask_0 = const()[name = string("op_5302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5302_cast_fp16 = slice_by_index(begin = var_5302_begin_0, end = var_5302_end_0, end_mask = var_5302_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5302_cast_fp16")]; + tensor var_5309_begin_0 = const()[name = string("op_5309_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5309_end_0 = const()[name = string("op_5309_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5309_end_mask_0 = const()[name = string("op_5309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5309_cast_fp16 = slice_by_index(begin = var_5309_begin_0, end = var_5309_end_0, end_mask = var_5309_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5309_cast_fp16")]; + tensor var_5316_begin_0 = const()[name = string("op_5316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5316_end_0 = const()[name = string("op_5316_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5316_end_mask_0 = const()[name = string("op_5316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5316_cast_fp16 = slice_by_index(begin = var_5316_begin_0, end = var_5316_end_0, end_mask = var_5316_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5316_cast_fp16")]; + tensor var_5323_begin_0 = const()[name = string("op_5323_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5323_end_0 = const()[name = string("op_5323_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5323_end_mask_0 = const()[name = string("op_5323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5323_cast_fp16 = slice_by_index(begin = var_5323_begin_0, end = var_5323_end_0, end_mask = var_5323_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5323_cast_fp16")]; + tensor var_5330_begin_0 = const()[name = string("op_5330_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5330_end_0 = const()[name = string("op_5330_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5330_end_mask_0 = const()[name = string("op_5330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5330_cast_fp16 = slice_by_index(begin = var_5330_begin_0, end = var_5330_end_0, end_mask = var_5330_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5330_cast_fp16")]; + tensor var_5337_begin_0 = const()[name = string("op_5337_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5337_end_0 = const()[name = string("op_5337_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5337_end_mask_0 = const()[name = string("op_5337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5337_cast_fp16 = slice_by_index(begin = var_5337_begin_0, end = var_5337_end_0, end_mask = var_5337_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5337_cast_fp16")]; + tensor var_5344_begin_0 = const()[name = string("op_5344_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5344_end_0 = const()[name = string("op_5344_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5344_end_mask_0 = const()[name = string("op_5344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5344_cast_fp16 = slice_by_index(begin = var_5344_begin_0, end = var_5344_end_0, end_mask = var_5344_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5344_cast_fp16")]; + tensor var_5351_begin_0 = const()[name = string("op_5351_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5351_end_0 = const()[name = string("op_5351_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5351_end_mask_0 = const()[name = string("op_5351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5351_cast_fp16 = slice_by_index(begin = var_5351_begin_0, end = var_5351_end_0, end_mask = var_5351_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5351_cast_fp16")]; + tensor var_5358_begin_0 = const()[name = string("op_5358_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5358_end_0 = const()[name = string("op_5358_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5358_end_mask_0 = const()[name = string("op_5358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5358_cast_fp16 = slice_by_index(begin = var_5358_begin_0, end = var_5358_end_0, end_mask = var_5358_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5358_cast_fp16")]; + tensor var_5365_begin_0 = const()[name = string("op_5365_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5365_end_0 = const()[name = string("op_5365_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5365_end_mask_0 = const()[name = string("op_5365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5365_cast_fp16 = slice_by_index(begin = var_5365_begin_0, end = var_5365_end_0, end_mask = var_5365_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5365_cast_fp16")]; + tensor var_5372_begin_0 = const()[name = string("op_5372_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5372_end_0 = const()[name = string("op_5372_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5372_end_mask_0 = const()[name = string("op_5372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5372_cast_fp16 = slice_by_index(begin = var_5372_begin_0, end = var_5372_end_0, end_mask = var_5372_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5372_cast_fp16")]; + tensor var_5379_begin_0 = const()[name = string("op_5379_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5379_end_0 = const()[name = string("op_5379_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5379_end_mask_0 = const()[name = string("op_5379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5379_cast_fp16 = slice_by_index(begin = var_5379_begin_0, end = var_5379_end_0, end_mask = var_5379_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5379_cast_fp16")]; + tensor var_5386_begin_0 = const()[name = string("op_5386_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5386_end_0 = const()[name = string("op_5386_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5386_end_mask_0 = const()[name = string("op_5386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5386_cast_fp16 = slice_by_index(begin = var_5386_begin_0, end = var_5386_end_0, end_mask = var_5386_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5386_cast_fp16")]; + tensor var_5393_begin_0 = const()[name = string("op_5393_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5393_end_0 = const()[name = string("op_5393_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5393_end_mask_0 = const()[name = string("op_5393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5393_cast_fp16 = slice_by_index(begin = var_5393_begin_0, end = var_5393_end_0, end_mask = var_5393_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5393_cast_fp16")]; + tensor var_5400_begin_0 = const()[name = string("op_5400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5400_end_0 = const()[name = string("op_5400_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5400_end_mask_0 = const()[name = string("op_5400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5400_cast_fp16 = slice_by_index(begin = var_5400_begin_0, end = var_5400_end_0, end_mask = var_5400_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5400_cast_fp16")]; + tensor var_5407_begin_0 = const()[name = string("op_5407_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5407_end_0 = const()[name = string("op_5407_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5407_end_mask_0 = const()[name = string("op_5407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5407_cast_fp16 = slice_by_index(begin = var_5407_begin_0, end = var_5407_end_0, end_mask = var_5407_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5407_cast_fp16")]; + tensor var_5414_begin_0 = const()[name = string("op_5414_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5414_end_0 = const()[name = string("op_5414_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5414_end_mask_0 = const()[name = string("op_5414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5414_cast_fp16 = slice_by_index(begin = var_5414_begin_0, end = var_5414_end_0, end_mask = var_5414_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5414_cast_fp16")]; + tensor var_5421_begin_0 = const()[name = string("op_5421_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5421_end_0 = const()[name = string("op_5421_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5421_end_mask_0 = const()[name = string("op_5421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5421_cast_fp16 = slice_by_index(begin = var_5421_begin_0, end = var_5421_end_0, end_mask = var_5421_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5421_cast_fp16")]; + tensor var_5428_begin_0 = const()[name = string("op_5428_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5428_end_0 = const()[name = string("op_5428_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5428_end_mask_0 = const()[name = string("op_5428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5428_cast_fp16 = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5428_cast_fp16")]; + tensor var_5435_begin_0 = const()[name = string("op_5435_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5435_end_0 = const()[name = string("op_5435_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5435_end_mask_0 = const()[name = string("op_5435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5435_cast_fp16 = slice_by_index(begin = var_5435_begin_0, end = var_5435_end_0, end_mask = var_5435_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5435_cast_fp16")]; + tensor var_5442_begin_0 = const()[name = string("op_5442_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5442_end_0 = const()[name = string("op_5442_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5442_end_mask_0 = const()[name = string("op_5442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5442_cast_fp16 = slice_by_index(begin = var_5442_begin_0, end = var_5442_end_0, end_mask = var_5442_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5442_cast_fp16")]; + tensor var_5449_begin_0 = const()[name = string("op_5449_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5449_end_0 = const()[name = string("op_5449_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5449_end_mask_0 = const()[name = string("op_5449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5449_cast_fp16 = slice_by_index(begin = var_5449_begin_0, end = var_5449_end_0, end_mask = var_5449_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5449_cast_fp16")]; + tensor var_5456_begin_0 = const()[name = string("op_5456_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5456_end_0 = const()[name = string("op_5456_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5456_end_mask_0 = const()[name = string("op_5456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5456_cast_fp16 = slice_by_index(begin = var_5456_begin_0, end = var_5456_end_0, end_mask = var_5456_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5456_cast_fp16")]; + tensor var_5463_begin_0 = const()[name = string("op_5463_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5463_end_0 = const()[name = string("op_5463_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5463_end_mask_0 = const()[name = string("op_5463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5463_cast_fp16 = slice_by_index(begin = var_5463_begin_0, end = var_5463_end_0, end_mask = var_5463_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5463_cast_fp16")]; + tensor var_5470_begin_0 = const()[name = string("op_5470_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5470_end_0 = const()[name = string("op_5470_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5470_end_mask_0 = const()[name = string("op_5470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5470_cast_fp16 = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5470_cast_fp16")]; + tensor var_5477_begin_0 = const()[name = string("op_5477_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5477_end_0 = const()[name = string("op_5477_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5477_end_mask_0 = const()[name = string("op_5477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5477_cast_fp16 = slice_by_index(begin = var_5477_begin_0, end = var_5477_end_0, end_mask = var_5477_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5477_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5482_begin_0 = const()[name = string("op_5482_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5482_end_0 = const()[name = string("op_5482_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_5482_end_mask_0 = const()[name = string("op_5482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_28")]; + tensor var_5482_cast_fp16 = slice_by_index(begin = var_5482_begin_0, end = var_5482_end_0, end_mask = var_5482_end_mask_0, x = k_7_cast_fp16)[name = string("op_5482_cast_fp16")]; + tensor var_5486_begin_0 = const()[name = string("op_5486_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5486_end_0 = const()[name = string("op_5486_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_5486_end_mask_0 = const()[name = string("op_5486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5486_cast_fp16 = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = k_7_cast_fp16)[name = string("op_5486_cast_fp16")]; + tensor var_5490_begin_0 = const()[name = string("op_5490_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5490_end_0 = const()[name = string("op_5490_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_5490_end_mask_0 = const()[name = string("op_5490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5490_cast_fp16 = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = k_7_cast_fp16)[name = string("op_5490_cast_fp16")]; + tensor var_5494_begin_0 = const()[name = string("op_5494_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5494_end_0 = const()[name = string("op_5494_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_5494_end_mask_0 = const()[name = string("op_5494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5494_cast_fp16 = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = k_7_cast_fp16)[name = string("op_5494_cast_fp16")]; + tensor var_5498_begin_0 = const()[name = string("op_5498_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5498_end_0 = const()[name = string("op_5498_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_5498_end_mask_0 = const()[name = string("op_5498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = k_7_cast_fp16)[name = string("op_5498_cast_fp16")]; + tensor var_5502_begin_0 = const()[name = string("op_5502_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5502_end_0 = const()[name = string("op_5502_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_5502_end_mask_0 = const()[name = string("op_5502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5502_cast_fp16 = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = k_7_cast_fp16)[name = string("op_5502_cast_fp16")]; + tensor var_5506_begin_0 = const()[name = string("op_5506_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5506_end_0 = const()[name = string("op_5506_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_5506_end_mask_0 = const()[name = string("op_5506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5506_cast_fp16 = slice_by_index(begin = var_5506_begin_0, end = var_5506_end_0, end_mask = var_5506_end_mask_0, x = k_7_cast_fp16)[name = string("op_5506_cast_fp16")]; + tensor var_5510_begin_0 = const()[name = string("op_5510_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5510_end_0 = const()[name = string("op_5510_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_5510_end_mask_0 = const()[name = string("op_5510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5510_cast_fp16 = slice_by_index(begin = var_5510_begin_0, end = var_5510_end_0, end_mask = var_5510_end_mask_0, x = k_7_cast_fp16)[name = string("op_5510_cast_fp16")]; + tensor var_5514_begin_0 = const()[name = string("op_5514_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5514_end_0 = const()[name = string("op_5514_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_5514_end_mask_0 = const()[name = string("op_5514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5514_cast_fp16 = slice_by_index(begin = var_5514_begin_0, end = var_5514_end_0, end_mask = var_5514_end_mask_0, x = k_7_cast_fp16)[name = string("op_5514_cast_fp16")]; + tensor var_5518_begin_0 = const()[name = string("op_5518_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5518_end_0 = const()[name = string("op_5518_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_5518_end_mask_0 = const()[name = string("op_5518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5518_cast_fp16 = slice_by_index(begin = var_5518_begin_0, end = var_5518_end_0, end_mask = var_5518_end_mask_0, x = k_7_cast_fp16)[name = string("op_5518_cast_fp16")]; + tensor var_5522_begin_0 = const()[name = string("op_5522_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5522_end_0 = const()[name = string("op_5522_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_5522_end_mask_0 = const()[name = string("op_5522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5522_cast_fp16 = slice_by_index(begin = var_5522_begin_0, end = var_5522_end_0, end_mask = var_5522_end_mask_0, x = k_7_cast_fp16)[name = string("op_5522_cast_fp16")]; + tensor var_5526_begin_0 = const()[name = string("op_5526_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5526_end_0 = const()[name = string("op_5526_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_5526_end_mask_0 = const()[name = string("op_5526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5526_cast_fp16 = slice_by_index(begin = var_5526_begin_0, end = var_5526_end_0, end_mask = var_5526_end_mask_0, x = k_7_cast_fp16)[name = string("op_5526_cast_fp16")]; + tensor var_5530_begin_0 = const()[name = string("op_5530_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5530_end_0 = const()[name = string("op_5530_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_5530_end_mask_0 = const()[name = string("op_5530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5530_cast_fp16 = slice_by_index(begin = var_5530_begin_0, end = var_5530_end_0, end_mask = var_5530_end_mask_0, x = k_7_cast_fp16)[name = string("op_5530_cast_fp16")]; + tensor var_5534_begin_0 = const()[name = string("op_5534_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5534_end_0 = const()[name = string("op_5534_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_5534_end_mask_0 = const()[name = string("op_5534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5534_cast_fp16 = slice_by_index(begin = var_5534_begin_0, end = var_5534_end_0, end_mask = var_5534_end_mask_0, x = k_7_cast_fp16)[name = string("op_5534_cast_fp16")]; + tensor var_5538_begin_0 = const()[name = string("op_5538_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5538_end_0 = const()[name = string("op_5538_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_5538_end_mask_0 = const()[name = string("op_5538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5538_cast_fp16 = slice_by_index(begin = var_5538_begin_0, end = var_5538_end_0, end_mask = var_5538_end_mask_0, x = k_7_cast_fp16)[name = string("op_5538_cast_fp16")]; + tensor var_5542_begin_0 = const()[name = string("op_5542_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5542_end_0 = const()[name = string("op_5542_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_5542_end_mask_0 = const()[name = string("op_5542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5542_cast_fp16 = slice_by_index(begin = var_5542_begin_0, end = var_5542_end_0, end_mask = var_5542_end_mask_0, x = k_7_cast_fp16)[name = string("op_5542_cast_fp16")]; + tensor var_5546_begin_0 = const()[name = string("op_5546_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5546_end_0 = const()[name = string("op_5546_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_5546_end_mask_0 = const()[name = string("op_5546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5546_cast_fp16 = slice_by_index(begin = var_5546_begin_0, end = var_5546_end_0, end_mask = var_5546_end_mask_0, x = k_7_cast_fp16)[name = string("op_5546_cast_fp16")]; + tensor var_5550_begin_0 = const()[name = string("op_5550_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5550_end_0 = const()[name = string("op_5550_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_5550_end_mask_0 = const()[name = string("op_5550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5550_cast_fp16 = slice_by_index(begin = var_5550_begin_0, end = var_5550_end_0, end_mask = var_5550_end_mask_0, x = k_7_cast_fp16)[name = string("op_5550_cast_fp16")]; + tensor var_5554_begin_0 = const()[name = string("op_5554_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5554_end_0 = const()[name = string("op_5554_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_5554_end_mask_0 = const()[name = string("op_5554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5554_cast_fp16 = slice_by_index(begin = var_5554_begin_0, end = var_5554_end_0, end_mask = var_5554_end_mask_0, x = k_7_cast_fp16)[name = string("op_5554_cast_fp16")]; + tensor var_5558_begin_0 = const()[name = string("op_5558_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5558_end_0 = const()[name = string("op_5558_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_5558_end_mask_0 = const()[name = string("op_5558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5558_cast_fp16 = slice_by_index(begin = var_5558_begin_0, end = var_5558_end_0, end_mask = var_5558_end_mask_0, x = k_7_cast_fp16)[name = string("op_5558_cast_fp16")]; + tensor var_5560_begin_0 = const()[name = string("op_5560_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5560_end_0 = const()[name = string("op_5560_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5560_end_mask_0 = const()[name = string("op_5560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5560_cast_fp16 = slice_by_index(begin = var_5560_begin_0, end = var_5560_end_0, end_mask = var_5560_end_mask_0, x = value_7_cast_fp16)[name = string("op_5560_cast_fp16")]; + tensor var_5564_begin_0 = const()[name = string("op_5564_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5564_end_0 = const()[name = string("op_5564_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5564_end_mask_0 = const()[name = string("op_5564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5564_cast_fp16 = slice_by_index(begin = var_5564_begin_0, end = var_5564_end_0, end_mask = var_5564_end_mask_0, x = value_7_cast_fp16)[name = string("op_5564_cast_fp16")]; + tensor var_5568_begin_0 = const()[name = string("op_5568_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5568_end_0 = const()[name = string("op_5568_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5568_end_mask_0 = const()[name = string("op_5568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5568_cast_fp16 = slice_by_index(begin = var_5568_begin_0, end = var_5568_end_0, end_mask = var_5568_end_mask_0, x = value_7_cast_fp16)[name = string("op_5568_cast_fp16")]; + tensor var_5572_begin_0 = const()[name = string("op_5572_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5572_end_0 = const()[name = string("op_5572_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5572_end_mask_0 = const()[name = string("op_5572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5572_cast_fp16 = slice_by_index(begin = var_5572_begin_0, end = var_5572_end_0, end_mask = var_5572_end_mask_0, x = value_7_cast_fp16)[name = string("op_5572_cast_fp16")]; + tensor var_5576_begin_0 = const()[name = string("op_5576_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5576_end_0 = const()[name = string("op_5576_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5576_end_mask_0 = const()[name = string("op_5576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5576_cast_fp16 = slice_by_index(begin = var_5576_begin_0, end = var_5576_end_0, end_mask = var_5576_end_mask_0, x = value_7_cast_fp16)[name = string("op_5576_cast_fp16")]; + tensor var_5580_begin_0 = const()[name = string("op_5580_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5580_end_0 = const()[name = string("op_5580_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5580_end_mask_0 = const()[name = string("op_5580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5580_cast_fp16 = slice_by_index(begin = var_5580_begin_0, end = var_5580_end_0, end_mask = var_5580_end_mask_0, x = value_7_cast_fp16)[name = string("op_5580_cast_fp16")]; + tensor var_5584_begin_0 = const()[name = string("op_5584_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5584_end_0 = const()[name = string("op_5584_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5584_end_mask_0 = const()[name = string("op_5584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5584_cast_fp16 = slice_by_index(begin = var_5584_begin_0, end = var_5584_end_0, end_mask = var_5584_end_mask_0, x = value_7_cast_fp16)[name = string("op_5584_cast_fp16")]; + tensor var_5588_begin_0 = const()[name = string("op_5588_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5588_end_0 = const()[name = string("op_5588_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5588_end_mask_0 = const()[name = string("op_5588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5588_cast_fp16 = slice_by_index(begin = var_5588_begin_0, end = var_5588_end_0, end_mask = var_5588_end_mask_0, x = value_7_cast_fp16)[name = string("op_5588_cast_fp16")]; + tensor var_5592_begin_0 = const()[name = string("op_5592_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5592_end_0 = const()[name = string("op_5592_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5592_end_mask_0 = const()[name = string("op_5592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5592_cast_fp16 = slice_by_index(begin = var_5592_begin_0, end = var_5592_end_0, end_mask = var_5592_end_mask_0, x = value_7_cast_fp16)[name = string("op_5592_cast_fp16")]; + tensor var_5596_begin_0 = const()[name = string("op_5596_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5596_end_0 = const()[name = string("op_5596_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5596_end_mask_0 = const()[name = string("op_5596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5596_cast_fp16 = slice_by_index(begin = var_5596_begin_0, end = var_5596_end_0, end_mask = var_5596_end_mask_0, x = value_7_cast_fp16)[name = string("op_5596_cast_fp16")]; + tensor var_5600_begin_0 = const()[name = string("op_5600_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5600_end_0 = const()[name = string("op_5600_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5600_end_mask_0 = const()[name = string("op_5600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5600_cast_fp16 = slice_by_index(begin = var_5600_begin_0, end = var_5600_end_0, end_mask = var_5600_end_mask_0, x = value_7_cast_fp16)[name = string("op_5600_cast_fp16")]; + tensor var_5604_begin_0 = const()[name = string("op_5604_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5604_end_0 = const()[name = string("op_5604_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5604_end_mask_0 = const()[name = string("op_5604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5604_cast_fp16 = slice_by_index(begin = var_5604_begin_0, end = var_5604_end_0, end_mask = var_5604_end_mask_0, x = value_7_cast_fp16)[name = string("op_5604_cast_fp16")]; + tensor var_5608_begin_0 = const()[name = string("op_5608_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5608_end_0 = const()[name = string("op_5608_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_5608_end_mask_0 = const()[name = string("op_5608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5608_cast_fp16 = slice_by_index(begin = var_5608_begin_0, end = var_5608_end_0, end_mask = var_5608_end_mask_0, x = value_7_cast_fp16)[name = string("op_5608_cast_fp16")]; + tensor var_5612_begin_0 = const()[name = string("op_5612_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5612_end_0 = const()[name = string("op_5612_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_5612_end_mask_0 = const()[name = string("op_5612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5612_cast_fp16 = slice_by_index(begin = var_5612_begin_0, end = var_5612_end_0, end_mask = var_5612_end_mask_0, x = value_7_cast_fp16)[name = string("op_5612_cast_fp16")]; + tensor var_5616_begin_0 = const()[name = string("op_5616_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5616_end_0 = const()[name = string("op_5616_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_5616_end_mask_0 = const()[name = string("op_5616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5616_cast_fp16 = slice_by_index(begin = var_5616_begin_0, end = var_5616_end_0, end_mask = var_5616_end_mask_0, x = value_7_cast_fp16)[name = string("op_5616_cast_fp16")]; + tensor var_5620_begin_0 = const()[name = string("op_5620_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5620_end_0 = const()[name = string("op_5620_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_5620_end_mask_0 = const()[name = string("op_5620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5620_cast_fp16 = slice_by_index(begin = var_5620_begin_0, end = var_5620_end_0, end_mask = var_5620_end_mask_0, x = value_7_cast_fp16)[name = string("op_5620_cast_fp16")]; + tensor var_5624_begin_0 = const()[name = string("op_5624_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5624_end_0 = const()[name = string("op_5624_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_5624_end_mask_0 = const()[name = string("op_5624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5624_cast_fp16 = slice_by_index(begin = var_5624_begin_0, end = var_5624_end_0, end_mask = var_5624_end_mask_0, x = value_7_cast_fp16)[name = string("op_5624_cast_fp16")]; + tensor var_5628_begin_0 = const()[name = string("op_5628_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5628_end_0 = const()[name = string("op_5628_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_5628_end_mask_0 = const()[name = string("op_5628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5628_cast_fp16 = slice_by_index(begin = var_5628_begin_0, end = var_5628_end_0, end_mask = var_5628_end_mask_0, x = value_7_cast_fp16)[name = string("op_5628_cast_fp16")]; + tensor var_5632_begin_0 = const()[name = string("op_5632_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5632_end_0 = const()[name = string("op_5632_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_5632_end_mask_0 = const()[name = string("op_5632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5632_cast_fp16 = slice_by_index(begin = var_5632_begin_0, end = var_5632_end_0, end_mask = var_5632_end_mask_0, x = value_7_cast_fp16)[name = string("op_5632_cast_fp16")]; + tensor var_5636_begin_0 = const()[name = string("op_5636_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5636_end_0 = const()[name = string("op_5636_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_5636_end_mask_0 = const()[name = string("op_5636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5636_cast_fp16 = slice_by_index(begin = var_5636_begin_0, end = var_5636_end_0, end_mask = var_5636_end_mask_0, x = value_7_cast_fp16)[name = string("op_5636_cast_fp16")]; + string _SplitHeadsQ__mh_w_481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_481_equation_0, values = (var_5482_cast_fp16, var_4924_cast_fp16))[name = string("_SplitHeadsQ__mh_w_481_cast_fp16")]; + string _SplitHeadsQ__mh_w_483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_483_equation_0, values = (var_5482_cast_fp16, var_4931_cast_fp16))[name = string("_SplitHeadsQ__mh_w_483_cast_fp16")]; + string _SplitHeadsQ__mh_w_485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_485_equation_0, values = (var_5482_cast_fp16, var_4938_cast_fp16))[name = string("_SplitHeadsQ__mh_w_485_cast_fp16")]; + string _SplitHeadsQ__mh_w_487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_487_equation_0, values = (var_5482_cast_fp16, var_4945_cast_fp16))[name = string("_SplitHeadsQ__mh_w_487_cast_fp16")]; + string _SplitHeadsQ__mh_w_489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_489_equation_0, values = (var_5486_cast_fp16, var_4952_cast_fp16))[name = string("_SplitHeadsQ__mh_w_489_cast_fp16")]; + string _SplitHeadsQ__mh_w_491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_491_equation_0, values = (var_5486_cast_fp16, var_4959_cast_fp16))[name = string("_SplitHeadsQ__mh_w_491_cast_fp16")]; + string _SplitHeadsQ__mh_w_493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_493_equation_0, values = (var_5486_cast_fp16, var_4966_cast_fp16))[name = string("_SplitHeadsQ__mh_w_493_cast_fp16")]; + string _SplitHeadsQ__mh_w_495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_495_equation_0, values = (var_5486_cast_fp16, var_4973_cast_fp16))[name = string("_SplitHeadsQ__mh_w_495_cast_fp16")]; + string _SplitHeadsQ__mh_w_497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_497_equation_0, values = (var_5490_cast_fp16, var_4980_cast_fp16))[name = string("_SplitHeadsQ__mh_w_497_cast_fp16")]; + string _SplitHeadsQ__mh_w_499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_499_equation_0, values = (var_5490_cast_fp16, var_4987_cast_fp16))[name = string("_SplitHeadsQ__mh_w_499_cast_fp16")]; + string _SplitHeadsQ__mh_w_501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_501_equation_0, values = (var_5490_cast_fp16, var_4994_cast_fp16))[name = string("_SplitHeadsQ__mh_w_501_cast_fp16")]; + string _SplitHeadsQ__mh_w_503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_503_equation_0, values = (var_5490_cast_fp16, var_5001_cast_fp16))[name = string("_SplitHeadsQ__mh_w_503_cast_fp16")]; + string _SplitHeadsQ__mh_w_505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_505_equation_0, values = (var_5494_cast_fp16, var_5008_cast_fp16))[name = string("_SplitHeadsQ__mh_w_505_cast_fp16")]; + string _SplitHeadsQ__mh_w_507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_507_equation_0, values = (var_5494_cast_fp16, var_5015_cast_fp16))[name = string("_SplitHeadsQ__mh_w_507_cast_fp16")]; + string _SplitHeadsQ__mh_w_509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_509_equation_0, values = (var_5494_cast_fp16, var_5022_cast_fp16))[name = string("_SplitHeadsQ__mh_w_509_cast_fp16")]; + string _SplitHeadsQ__mh_w_511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_511_equation_0, values = (var_5494_cast_fp16, var_5029_cast_fp16))[name = string("_SplitHeadsQ__mh_w_511_cast_fp16")]; + string _SplitHeadsQ__mh_w_513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_513_equation_0, values = (var_5498_cast_fp16, var_5036_cast_fp16))[name = string("_SplitHeadsQ__mh_w_513_cast_fp16")]; + string _SplitHeadsQ__mh_w_515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_515_equation_0, values = (var_5498_cast_fp16, var_5043_cast_fp16))[name = string("_SplitHeadsQ__mh_w_515_cast_fp16")]; + string _SplitHeadsQ__mh_w_517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_517_equation_0, values = (var_5498_cast_fp16, var_5050_cast_fp16))[name = string("_SplitHeadsQ__mh_w_517_cast_fp16")]; + string _SplitHeadsQ__mh_w_519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_519_equation_0, values = (var_5498_cast_fp16, var_5057_cast_fp16))[name = string("_SplitHeadsQ__mh_w_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_521_equation_0, values = (var_5502_cast_fp16, var_5064_cast_fp16))[name = string("_SplitHeadsQ__mh_w_521_cast_fp16")]; + string _SplitHeadsQ__mh_w_523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_523_equation_0, values = (var_5502_cast_fp16, var_5071_cast_fp16))[name = string("_SplitHeadsQ__mh_w_523_cast_fp16")]; + string _SplitHeadsQ__mh_w_525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_525_equation_0, values = (var_5502_cast_fp16, var_5078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_525_cast_fp16")]; + string _SplitHeadsQ__mh_w_527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_527_equation_0, values = (var_5502_cast_fp16, var_5085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_527_cast_fp16")]; + string _SplitHeadsQ__mh_w_529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_529_equation_0, values = (var_5506_cast_fp16, var_5092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_529_cast_fp16")]; + string _SplitHeadsQ__mh_w_531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_531_equation_0, values = (var_5506_cast_fp16, var_5099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_531_cast_fp16")]; + string _SplitHeadsQ__mh_w_533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_533_equation_0, values = (var_5506_cast_fp16, var_5106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_533_cast_fp16")]; + string _SplitHeadsQ__mh_w_535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_535_equation_0, values = (var_5506_cast_fp16, var_5113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_535_cast_fp16")]; + string _SplitHeadsQ__mh_w_537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_537_equation_0, values = (var_5510_cast_fp16, var_5120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_537_cast_fp16")]; + string _SplitHeadsQ__mh_w_539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_539_equation_0, values = (var_5510_cast_fp16, var_5127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_539_cast_fp16")]; + string _SplitHeadsQ__mh_w_541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_541_equation_0, values = (var_5510_cast_fp16, var_5134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_541_cast_fp16")]; + string _SplitHeadsQ__mh_w_543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_543_equation_0, values = (var_5510_cast_fp16, var_5141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_543_cast_fp16")]; + string _SplitHeadsQ__mh_w_545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_545_equation_0, values = (var_5514_cast_fp16, var_5148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_545_cast_fp16")]; + string _SplitHeadsQ__mh_w_547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_547_equation_0, values = (var_5514_cast_fp16, var_5155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_547_cast_fp16")]; + string _SplitHeadsQ__mh_w_549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_549_equation_0, values = (var_5514_cast_fp16, var_5162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_549_cast_fp16")]; + string _SplitHeadsQ__mh_w_551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_551_equation_0, values = (var_5514_cast_fp16, var_5169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_551_cast_fp16")]; + string _SplitHeadsQ__mh_w_553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_553_equation_0, values = (var_5518_cast_fp16, var_5176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_553_cast_fp16")]; + string _SplitHeadsQ__mh_w_555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_555_equation_0, values = (var_5518_cast_fp16, var_5183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_555_cast_fp16")]; + string _SplitHeadsQ__mh_w_557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_557_equation_0, values = (var_5518_cast_fp16, var_5190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_557_cast_fp16")]; + string _SplitHeadsQ__mh_w_559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_559_equation_0, values = (var_5518_cast_fp16, var_5197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_559_cast_fp16")]; + string _SplitHeadsQ__mh_w_561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_561_equation_0, values = (var_5522_cast_fp16, var_5204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_561_cast_fp16")]; + string _SplitHeadsQ__mh_w_563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_563_equation_0, values = (var_5522_cast_fp16, var_5211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_563_cast_fp16")]; + string _SplitHeadsQ__mh_w_565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_565_equation_0, values = (var_5522_cast_fp16, var_5218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_565_cast_fp16")]; + string _SplitHeadsQ__mh_w_567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_567_equation_0, values = (var_5522_cast_fp16, var_5225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_567_cast_fp16")]; + string _SplitHeadsQ__mh_w_569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_569_equation_0, values = (var_5526_cast_fp16, var_5232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_569_cast_fp16")]; + string _SplitHeadsQ__mh_w_571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_571_equation_0, values = (var_5526_cast_fp16, var_5239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_571_cast_fp16")]; + string _SplitHeadsQ__mh_w_573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_573_equation_0, values = (var_5526_cast_fp16, var_5246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_573_cast_fp16")]; + string _SplitHeadsQ__mh_w_575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_575_equation_0, values = (var_5526_cast_fp16, var_5253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_575_cast_fp16")]; + string _SplitHeadsQ__mh_w_577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_577_equation_0, values = (var_5530_cast_fp16, var_5260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_577_cast_fp16")]; + string _SplitHeadsQ__mh_w_579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_579_equation_0, values = (var_5530_cast_fp16, var_5267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_579_cast_fp16")]; + string _SplitHeadsQ__mh_w_581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_581_equation_0, values = (var_5530_cast_fp16, var_5274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_581_cast_fp16")]; + string _SplitHeadsQ__mh_w_583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_583_equation_0, values = (var_5530_cast_fp16, var_5281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_583_cast_fp16")]; + string _SplitHeadsQ__mh_w_585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_585_equation_0, values = (var_5534_cast_fp16, var_5288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_585_cast_fp16")]; + string _SplitHeadsQ__mh_w_587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_587_equation_0, values = (var_5534_cast_fp16, var_5295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_587_cast_fp16")]; + string _SplitHeadsQ__mh_w_589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_589_equation_0, values = (var_5534_cast_fp16, var_5302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_589_cast_fp16")]; + string _SplitHeadsQ__mh_w_591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_591_equation_0, values = (var_5534_cast_fp16, var_5309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_591_cast_fp16")]; + string _SplitHeadsQ__mh_w_593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_593_equation_0, values = (var_5538_cast_fp16, var_5316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_593_cast_fp16")]; + string _SplitHeadsQ__mh_w_595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_595_equation_0, values = (var_5538_cast_fp16, var_5323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_595_cast_fp16")]; + string _SplitHeadsQ__mh_w_597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_597_equation_0, values = (var_5538_cast_fp16, var_5330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_597_cast_fp16")]; + string _SplitHeadsQ__mh_w_599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_599_equation_0, values = (var_5538_cast_fp16, var_5337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_599_cast_fp16")]; + string _SplitHeadsQ__mh_w_601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_601_equation_0, values = (var_5542_cast_fp16, var_5344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_601_cast_fp16")]; + string _SplitHeadsQ__mh_w_603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_603_equation_0, values = (var_5542_cast_fp16, var_5351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_603_cast_fp16")]; + string _SplitHeadsQ__mh_w_605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_605_equation_0, values = (var_5542_cast_fp16, var_5358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_605_cast_fp16")]; + string _SplitHeadsQ__mh_w_607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_607_equation_0, values = (var_5542_cast_fp16, var_5365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_607_cast_fp16")]; + string _SplitHeadsQ__mh_w_609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_609_equation_0, values = (var_5546_cast_fp16, var_5372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_609_cast_fp16")]; + string _SplitHeadsQ__mh_w_611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_611_equation_0, values = (var_5546_cast_fp16, var_5379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_611_cast_fp16")]; + string _SplitHeadsQ__mh_w_613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_613_equation_0, values = (var_5546_cast_fp16, var_5386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_613_cast_fp16")]; + string _SplitHeadsQ__mh_w_615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_615_equation_0, values = (var_5546_cast_fp16, var_5393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_615_cast_fp16")]; + string _SplitHeadsQ__mh_w_617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_617_equation_0, values = (var_5550_cast_fp16, var_5400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_617_cast_fp16")]; + string _SplitHeadsQ__mh_w_619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_619_equation_0, values = (var_5550_cast_fp16, var_5407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_619_cast_fp16")]; + string _SplitHeadsQ__mh_w_621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_621_equation_0, values = (var_5550_cast_fp16, var_5414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_621_cast_fp16")]; + string _SplitHeadsQ__mh_w_623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_623_equation_0, values = (var_5550_cast_fp16, var_5421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_623_cast_fp16")]; + string _SplitHeadsQ__mh_w_625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_625_equation_0, values = (var_5554_cast_fp16, var_5428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_625_cast_fp16")]; + string _SplitHeadsQ__mh_w_627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_627_equation_0, values = (var_5554_cast_fp16, var_5435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_627_cast_fp16")]; + string _SplitHeadsQ__mh_w_629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_629_equation_0, values = (var_5554_cast_fp16, var_5442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_629_cast_fp16")]; + string _SplitHeadsQ__mh_w_631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_631_equation_0, values = (var_5554_cast_fp16, var_5449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_631_cast_fp16")]; + string _SplitHeadsQ__mh_w_633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_633_equation_0, values = (var_5558_cast_fp16, var_5456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_633_cast_fp16")]; + string _SplitHeadsQ__mh_w_635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_635_equation_0, values = (var_5558_cast_fp16, var_5463_cast_fp16))[name = string("_SplitHeadsQ__mh_w_635_cast_fp16")]; + string _SplitHeadsQ__mh_w_637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_637_equation_0, values = (var_5558_cast_fp16, var_5470_cast_fp16))[name = string("_SplitHeadsQ__mh_w_637_cast_fp16")]; + string _SplitHeadsQ__mh_w_639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_639_equation_0, values = (var_5558_cast_fp16, var_5477_cast_fp16))[name = string("_SplitHeadsQ__mh_w_639_cast_fp16")]; + fp16 var_5799_to_fp16 = const()[name = string("op_5799_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_481_cast_fp16, y = var_5799_to_fp16)[name = string("aw_chunk_481_cast_fp16")]; + fp16 var_5801_to_fp16 = const()[name = string("op_5801_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_483_cast_fp16, y = var_5801_to_fp16)[name = string("aw_chunk_483_cast_fp16")]; + fp16 var_5803_to_fp16 = const()[name = string("op_5803_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_485_cast_fp16, y = var_5803_to_fp16)[name = string("aw_chunk_485_cast_fp16")]; + fp16 var_5805_to_fp16 = const()[name = string("op_5805_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_487_cast_fp16, y = var_5805_to_fp16)[name = string("aw_chunk_487_cast_fp16")]; + fp16 var_5807_to_fp16 = const()[name = string("op_5807_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_489_cast_fp16, y = var_5807_to_fp16)[name = string("aw_chunk_489_cast_fp16")]; + fp16 var_5809_to_fp16 = const()[name = string("op_5809_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_491_cast_fp16, y = var_5809_to_fp16)[name = string("aw_chunk_491_cast_fp16")]; + fp16 var_5811_to_fp16 = const()[name = string("op_5811_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_493_cast_fp16, y = var_5811_to_fp16)[name = string("aw_chunk_493_cast_fp16")]; + fp16 var_5813_to_fp16 = const()[name = string("op_5813_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_495_cast_fp16, y = var_5813_to_fp16)[name = string("aw_chunk_495_cast_fp16")]; + fp16 var_5815_to_fp16 = const()[name = string("op_5815_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_497_cast_fp16, y = var_5815_to_fp16)[name = string("aw_chunk_497_cast_fp16")]; + fp16 var_5817_to_fp16 = const()[name = string("op_5817_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_499_cast_fp16, y = var_5817_to_fp16)[name = string("aw_chunk_499_cast_fp16")]; + fp16 var_5819_to_fp16 = const()[name = string("op_5819_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_501_cast_fp16, y = var_5819_to_fp16)[name = string("aw_chunk_501_cast_fp16")]; + fp16 var_5821_to_fp16 = const()[name = string("op_5821_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_503_cast_fp16, y = var_5821_to_fp16)[name = string("aw_chunk_503_cast_fp16")]; + fp16 var_5823_to_fp16 = const()[name = string("op_5823_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_505_cast_fp16, y = var_5823_to_fp16)[name = string("aw_chunk_505_cast_fp16")]; + fp16 var_5825_to_fp16 = const()[name = string("op_5825_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_507_cast_fp16, y = var_5825_to_fp16)[name = string("aw_chunk_507_cast_fp16")]; + fp16 var_5827_to_fp16 = const()[name = string("op_5827_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_509_cast_fp16, y = var_5827_to_fp16)[name = string("aw_chunk_509_cast_fp16")]; + fp16 var_5829_to_fp16 = const()[name = string("op_5829_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_511_cast_fp16, y = var_5829_to_fp16)[name = string("aw_chunk_511_cast_fp16")]; + fp16 var_5831_to_fp16 = const()[name = string("op_5831_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_513_cast_fp16, y = var_5831_to_fp16)[name = string("aw_chunk_513_cast_fp16")]; + fp16 var_5833_to_fp16 = const()[name = string("op_5833_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_515_cast_fp16, y = var_5833_to_fp16)[name = string("aw_chunk_515_cast_fp16")]; + fp16 var_5835_to_fp16 = const()[name = string("op_5835_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_517_cast_fp16, y = var_5835_to_fp16)[name = string("aw_chunk_517_cast_fp16")]; + fp16 var_5837_to_fp16 = const()[name = string("op_5837_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_519_cast_fp16, y = var_5837_to_fp16)[name = string("aw_chunk_519_cast_fp16")]; + fp16 var_5839_to_fp16 = const()[name = string("op_5839_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_521_cast_fp16, y = var_5839_to_fp16)[name = string("aw_chunk_521_cast_fp16")]; + fp16 var_5841_to_fp16 = const()[name = string("op_5841_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_523_cast_fp16, y = var_5841_to_fp16)[name = string("aw_chunk_523_cast_fp16")]; + fp16 var_5843_to_fp16 = const()[name = string("op_5843_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_525_cast_fp16, y = var_5843_to_fp16)[name = string("aw_chunk_525_cast_fp16")]; + fp16 var_5845_to_fp16 = const()[name = string("op_5845_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_527_cast_fp16, y = var_5845_to_fp16)[name = string("aw_chunk_527_cast_fp16")]; + fp16 var_5847_to_fp16 = const()[name = string("op_5847_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_529_cast_fp16, y = var_5847_to_fp16)[name = string("aw_chunk_529_cast_fp16")]; + fp16 var_5849_to_fp16 = const()[name = string("op_5849_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_531_cast_fp16, y = var_5849_to_fp16)[name = string("aw_chunk_531_cast_fp16")]; + fp16 var_5851_to_fp16 = const()[name = string("op_5851_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_533_cast_fp16, y = var_5851_to_fp16)[name = string("aw_chunk_533_cast_fp16")]; + fp16 var_5853_to_fp16 = const()[name = string("op_5853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_535_cast_fp16, y = var_5853_to_fp16)[name = string("aw_chunk_535_cast_fp16")]; + fp16 var_5855_to_fp16 = const()[name = string("op_5855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_537_cast_fp16, y = var_5855_to_fp16)[name = string("aw_chunk_537_cast_fp16")]; + fp16 var_5857_to_fp16 = const()[name = string("op_5857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_539_cast_fp16, y = var_5857_to_fp16)[name = string("aw_chunk_539_cast_fp16")]; + fp16 var_5859_to_fp16 = const()[name = string("op_5859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_541_cast_fp16, y = var_5859_to_fp16)[name = string("aw_chunk_541_cast_fp16")]; + fp16 var_5861_to_fp16 = const()[name = string("op_5861_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_543_cast_fp16, y = var_5861_to_fp16)[name = string("aw_chunk_543_cast_fp16")]; + fp16 var_5863_to_fp16 = const()[name = string("op_5863_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_545_cast_fp16, y = var_5863_to_fp16)[name = string("aw_chunk_545_cast_fp16")]; + fp16 var_5865_to_fp16 = const()[name = string("op_5865_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_547_cast_fp16, y = var_5865_to_fp16)[name = string("aw_chunk_547_cast_fp16")]; + fp16 var_5867_to_fp16 = const()[name = string("op_5867_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_549_cast_fp16, y = var_5867_to_fp16)[name = string("aw_chunk_549_cast_fp16")]; + fp16 var_5869_to_fp16 = const()[name = string("op_5869_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_551_cast_fp16, y = var_5869_to_fp16)[name = string("aw_chunk_551_cast_fp16")]; + fp16 var_5871_to_fp16 = const()[name = string("op_5871_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_553_cast_fp16, y = var_5871_to_fp16)[name = string("aw_chunk_553_cast_fp16")]; + fp16 var_5873_to_fp16 = const()[name = string("op_5873_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_555_cast_fp16, y = var_5873_to_fp16)[name = string("aw_chunk_555_cast_fp16")]; + fp16 var_5875_to_fp16 = const()[name = string("op_5875_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_557_cast_fp16, y = var_5875_to_fp16)[name = string("aw_chunk_557_cast_fp16")]; + fp16 var_5877_to_fp16 = const()[name = string("op_5877_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_559_cast_fp16, y = var_5877_to_fp16)[name = string("aw_chunk_559_cast_fp16")]; + fp16 var_5879_to_fp16 = const()[name = string("op_5879_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_561_cast_fp16, y = var_5879_to_fp16)[name = string("aw_chunk_561_cast_fp16")]; + fp16 var_5881_to_fp16 = const()[name = string("op_5881_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_563_cast_fp16, y = var_5881_to_fp16)[name = string("aw_chunk_563_cast_fp16")]; + fp16 var_5883_to_fp16 = const()[name = string("op_5883_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_565_cast_fp16, y = var_5883_to_fp16)[name = string("aw_chunk_565_cast_fp16")]; + fp16 var_5885_to_fp16 = const()[name = string("op_5885_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_567_cast_fp16, y = var_5885_to_fp16)[name = string("aw_chunk_567_cast_fp16")]; + fp16 var_5887_to_fp16 = const()[name = string("op_5887_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_569_cast_fp16, y = var_5887_to_fp16)[name = string("aw_chunk_569_cast_fp16")]; + fp16 var_5889_to_fp16 = const()[name = string("op_5889_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_571_cast_fp16, y = var_5889_to_fp16)[name = string("aw_chunk_571_cast_fp16")]; + fp16 var_5891_to_fp16 = const()[name = string("op_5891_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_573_cast_fp16, y = var_5891_to_fp16)[name = string("aw_chunk_573_cast_fp16")]; + fp16 var_5893_to_fp16 = const()[name = string("op_5893_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_575_cast_fp16, y = var_5893_to_fp16)[name = string("aw_chunk_575_cast_fp16")]; + fp16 var_5895_to_fp16 = const()[name = string("op_5895_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_577_cast_fp16, y = var_5895_to_fp16)[name = string("aw_chunk_577_cast_fp16")]; + fp16 var_5897_to_fp16 = const()[name = string("op_5897_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_579_cast_fp16, y = var_5897_to_fp16)[name = string("aw_chunk_579_cast_fp16")]; + fp16 var_5899_to_fp16 = const()[name = string("op_5899_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_581_cast_fp16, y = var_5899_to_fp16)[name = string("aw_chunk_581_cast_fp16")]; + fp16 var_5901_to_fp16 = const()[name = string("op_5901_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_583_cast_fp16, y = var_5901_to_fp16)[name = string("aw_chunk_583_cast_fp16")]; + fp16 var_5903_to_fp16 = const()[name = string("op_5903_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_585_cast_fp16, y = var_5903_to_fp16)[name = string("aw_chunk_585_cast_fp16")]; + fp16 var_5905_to_fp16 = const()[name = string("op_5905_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_587_cast_fp16, y = var_5905_to_fp16)[name = string("aw_chunk_587_cast_fp16")]; + fp16 var_5907_to_fp16 = const()[name = string("op_5907_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_589_cast_fp16, y = var_5907_to_fp16)[name = string("aw_chunk_589_cast_fp16")]; + fp16 var_5909_to_fp16 = const()[name = string("op_5909_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_591_cast_fp16, y = var_5909_to_fp16)[name = string("aw_chunk_591_cast_fp16")]; + fp16 var_5911_to_fp16 = const()[name = string("op_5911_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_593_cast_fp16, y = var_5911_to_fp16)[name = string("aw_chunk_593_cast_fp16")]; + fp16 var_5913_to_fp16 = const()[name = string("op_5913_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_595_cast_fp16, y = var_5913_to_fp16)[name = string("aw_chunk_595_cast_fp16")]; + fp16 var_5915_to_fp16 = const()[name = string("op_5915_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_597_cast_fp16, y = var_5915_to_fp16)[name = string("aw_chunk_597_cast_fp16")]; + fp16 var_5917_to_fp16 = const()[name = string("op_5917_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_599_cast_fp16, y = var_5917_to_fp16)[name = string("aw_chunk_599_cast_fp16")]; + fp16 var_5919_to_fp16 = const()[name = string("op_5919_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_601_cast_fp16, y = var_5919_to_fp16)[name = string("aw_chunk_601_cast_fp16")]; + fp16 var_5921_to_fp16 = const()[name = string("op_5921_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_603_cast_fp16, y = var_5921_to_fp16)[name = string("aw_chunk_603_cast_fp16")]; + fp16 var_5923_to_fp16 = const()[name = string("op_5923_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_605_cast_fp16, y = var_5923_to_fp16)[name = string("aw_chunk_605_cast_fp16")]; + fp16 var_5925_to_fp16 = const()[name = string("op_5925_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_607_cast_fp16, y = var_5925_to_fp16)[name = string("aw_chunk_607_cast_fp16")]; + fp16 var_5927_to_fp16 = const()[name = string("op_5927_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_609_cast_fp16, y = var_5927_to_fp16)[name = string("aw_chunk_609_cast_fp16")]; + fp16 var_5929_to_fp16 = const()[name = string("op_5929_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_611_cast_fp16, y = var_5929_to_fp16)[name = string("aw_chunk_611_cast_fp16")]; + fp16 var_5931_to_fp16 = const()[name = string("op_5931_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_613_cast_fp16, y = var_5931_to_fp16)[name = string("aw_chunk_613_cast_fp16")]; + fp16 var_5933_to_fp16 = const()[name = string("op_5933_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_615_cast_fp16, y = var_5933_to_fp16)[name = string("aw_chunk_615_cast_fp16")]; + fp16 var_5935_to_fp16 = const()[name = string("op_5935_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_617_cast_fp16, y = var_5935_to_fp16)[name = string("aw_chunk_617_cast_fp16")]; + fp16 var_5937_to_fp16 = const()[name = string("op_5937_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_619_cast_fp16, y = var_5937_to_fp16)[name = string("aw_chunk_619_cast_fp16")]; + fp16 var_5939_to_fp16 = const()[name = string("op_5939_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_621_cast_fp16, y = var_5939_to_fp16)[name = string("aw_chunk_621_cast_fp16")]; + fp16 var_5941_to_fp16 = const()[name = string("op_5941_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_623_cast_fp16, y = var_5941_to_fp16)[name = string("aw_chunk_623_cast_fp16")]; + fp16 var_5943_to_fp16 = const()[name = string("op_5943_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_625_cast_fp16, y = var_5943_to_fp16)[name = string("aw_chunk_625_cast_fp16")]; + fp16 var_5945_to_fp16 = const()[name = string("op_5945_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_627_cast_fp16, y = var_5945_to_fp16)[name = string("aw_chunk_627_cast_fp16")]; + fp16 var_5947_to_fp16 = const()[name = string("op_5947_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_629_cast_fp16, y = var_5947_to_fp16)[name = string("aw_chunk_629_cast_fp16")]; + fp16 var_5949_to_fp16 = const()[name = string("op_5949_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_631_cast_fp16, y = var_5949_to_fp16)[name = string("aw_chunk_631_cast_fp16")]; + fp16 var_5951_to_fp16 = const()[name = string("op_5951_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_633_cast_fp16, y = var_5951_to_fp16)[name = string("aw_chunk_633_cast_fp16")]; + fp16 var_5953_to_fp16 = const()[name = string("op_5953_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_635_cast_fp16, y = var_5953_to_fp16)[name = string("aw_chunk_635_cast_fp16")]; + fp16 var_5955_to_fp16 = const()[name = string("op_5955_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_637_cast_fp16, y = var_5955_to_fp16)[name = string("aw_chunk_637_cast_fp16")]; + fp16 var_5957_to_fp16 = const()[name = string("op_5957_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_639_cast_fp16, y = var_5957_to_fp16)[name = string("aw_chunk_639_cast_fp16")]; + tensor var_5959_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_481_cast_fp16)[name = string("op_5959_cast_fp16")]; + tensor var_5960_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_483_cast_fp16)[name = string("op_5960_cast_fp16")]; + tensor var_5961_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_485_cast_fp16)[name = string("op_5961_cast_fp16")]; + tensor var_5962_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_487_cast_fp16)[name = string("op_5962_cast_fp16")]; + tensor var_5963_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_489_cast_fp16)[name = string("op_5963_cast_fp16")]; + tensor var_5964_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_491_cast_fp16)[name = string("op_5964_cast_fp16")]; + tensor var_5965_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_493_cast_fp16)[name = string("op_5965_cast_fp16")]; + tensor var_5966_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_495_cast_fp16)[name = string("op_5966_cast_fp16")]; + tensor var_5967_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_497_cast_fp16)[name = string("op_5967_cast_fp16")]; + tensor var_5968_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_499_cast_fp16)[name = string("op_5968_cast_fp16")]; + tensor var_5969_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_501_cast_fp16)[name = string("op_5969_cast_fp16")]; + tensor var_5970_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_503_cast_fp16)[name = string("op_5970_cast_fp16")]; + tensor var_5971_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_505_cast_fp16)[name = string("op_5971_cast_fp16")]; + tensor var_5972_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_507_cast_fp16)[name = string("op_5972_cast_fp16")]; + tensor var_5973_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_509_cast_fp16)[name = string("op_5973_cast_fp16")]; + tensor var_5974_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_511_cast_fp16)[name = string("op_5974_cast_fp16")]; + tensor var_5975_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_513_cast_fp16)[name = string("op_5975_cast_fp16")]; + tensor var_5976_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_515_cast_fp16)[name = string("op_5976_cast_fp16")]; + tensor var_5977_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_517_cast_fp16)[name = string("op_5977_cast_fp16")]; + tensor var_5978_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_519_cast_fp16)[name = string("op_5978_cast_fp16")]; + tensor var_5979_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_521_cast_fp16)[name = string("op_5979_cast_fp16")]; + tensor var_5980_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_523_cast_fp16)[name = string("op_5980_cast_fp16")]; + tensor var_5981_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_525_cast_fp16)[name = string("op_5981_cast_fp16")]; + tensor var_5982_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_527_cast_fp16)[name = string("op_5982_cast_fp16")]; + tensor var_5983_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_529_cast_fp16)[name = string("op_5983_cast_fp16")]; + tensor var_5984_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_531_cast_fp16)[name = string("op_5984_cast_fp16")]; + tensor var_5985_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_533_cast_fp16)[name = string("op_5985_cast_fp16")]; + tensor var_5986_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_535_cast_fp16)[name = string("op_5986_cast_fp16")]; + tensor var_5987_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_537_cast_fp16)[name = string("op_5987_cast_fp16")]; + tensor var_5988_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_539_cast_fp16)[name = string("op_5988_cast_fp16")]; + tensor var_5989_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_541_cast_fp16)[name = string("op_5989_cast_fp16")]; + tensor var_5990_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_543_cast_fp16)[name = string("op_5990_cast_fp16")]; + tensor var_5991_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_545_cast_fp16)[name = string("op_5991_cast_fp16")]; + tensor var_5992_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_547_cast_fp16)[name = string("op_5992_cast_fp16")]; + tensor var_5993_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_549_cast_fp16)[name = string("op_5993_cast_fp16")]; + tensor var_5994_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_551_cast_fp16)[name = string("op_5994_cast_fp16")]; + tensor var_5995_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_553_cast_fp16)[name = string("op_5995_cast_fp16")]; + tensor var_5996_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_555_cast_fp16)[name = string("op_5996_cast_fp16")]; + tensor var_5997_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_557_cast_fp16)[name = string("op_5997_cast_fp16")]; + tensor var_5998_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_559_cast_fp16)[name = string("op_5998_cast_fp16")]; + tensor var_5999_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_561_cast_fp16)[name = string("op_5999_cast_fp16")]; + tensor var_6000_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_563_cast_fp16)[name = string("op_6000_cast_fp16")]; + tensor var_6001_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_565_cast_fp16)[name = string("op_6001_cast_fp16")]; + tensor var_6002_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_567_cast_fp16)[name = string("op_6002_cast_fp16")]; + tensor var_6003_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_569_cast_fp16)[name = string("op_6003_cast_fp16")]; + tensor var_6004_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_571_cast_fp16)[name = string("op_6004_cast_fp16")]; + tensor var_6005_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_573_cast_fp16)[name = string("op_6005_cast_fp16")]; + tensor var_6006_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_575_cast_fp16)[name = string("op_6006_cast_fp16")]; + tensor var_6007_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_577_cast_fp16)[name = string("op_6007_cast_fp16")]; + tensor var_6008_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_579_cast_fp16)[name = string("op_6008_cast_fp16")]; + tensor var_6009_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_581_cast_fp16)[name = string("op_6009_cast_fp16")]; + tensor var_6010_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_583_cast_fp16)[name = string("op_6010_cast_fp16")]; + tensor var_6011_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_585_cast_fp16)[name = string("op_6011_cast_fp16")]; + tensor var_6012_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_587_cast_fp16)[name = string("op_6012_cast_fp16")]; + tensor var_6013_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_589_cast_fp16)[name = string("op_6013_cast_fp16")]; + tensor var_6014_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_591_cast_fp16)[name = string("op_6014_cast_fp16")]; + tensor var_6015_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_593_cast_fp16)[name = string("op_6015_cast_fp16")]; + tensor var_6016_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_595_cast_fp16)[name = string("op_6016_cast_fp16")]; + tensor var_6017_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_597_cast_fp16)[name = string("op_6017_cast_fp16")]; + tensor var_6018_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_599_cast_fp16)[name = string("op_6018_cast_fp16")]; + tensor var_6019_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_601_cast_fp16)[name = string("op_6019_cast_fp16")]; + tensor var_6020_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_603_cast_fp16)[name = string("op_6020_cast_fp16")]; + tensor var_6021_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_605_cast_fp16)[name = string("op_6021_cast_fp16")]; + tensor var_6022_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_607_cast_fp16)[name = string("op_6022_cast_fp16")]; + tensor var_6023_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_609_cast_fp16)[name = string("op_6023_cast_fp16")]; + tensor var_6024_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_611_cast_fp16)[name = string("op_6024_cast_fp16")]; + tensor var_6025_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_613_cast_fp16)[name = string("op_6025_cast_fp16")]; + tensor var_6026_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_615_cast_fp16)[name = string("op_6026_cast_fp16")]; + tensor var_6027_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_617_cast_fp16)[name = string("op_6027_cast_fp16")]; + tensor var_6028_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_619_cast_fp16)[name = string("op_6028_cast_fp16")]; + tensor var_6029_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_621_cast_fp16)[name = string("op_6029_cast_fp16")]; + tensor var_6030_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_623_cast_fp16)[name = string("op_6030_cast_fp16")]; + tensor var_6031_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_625_cast_fp16)[name = string("op_6031_cast_fp16")]; + tensor var_6032_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_627_cast_fp16)[name = string("op_6032_cast_fp16")]; + tensor var_6033_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_629_cast_fp16)[name = string("op_6033_cast_fp16")]; + tensor var_6034_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_631_cast_fp16)[name = string("op_6034_cast_fp16")]; + tensor var_6035_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_633_cast_fp16)[name = string("op_6035_cast_fp16")]; + tensor var_6036_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_635_cast_fp16)[name = string("op_6036_cast_fp16")]; + tensor var_6037_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_637_cast_fp16)[name = string("op_6037_cast_fp16")]; + tensor var_6038_cast_fp16 = softmax(axis = var_4784, x = aw_chunk_639_cast_fp16)[name = string("op_6038_cast_fp16")]; + string var_6040_equation_0 = const()[name = string("op_6040_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6040_cast_fp16 = einsum(equation = var_6040_equation_0, values = (var_5560_cast_fp16, var_5959_cast_fp16))[name = string("op_6040_cast_fp16")]; + string var_6042_equation_0 = const()[name = string("op_6042_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6042_cast_fp16 = einsum(equation = var_6042_equation_0, values = (var_5560_cast_fp16, var_5960_cast_fp16))[name = string("op_6042_cast_fp16")]; + string var_6044_equation_0 = const()[name = string("op_6044_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6044_cast_fp16 = einsum(equation = var_6044_equation_0, values = (var_5560_cast_fp16, var_5961_cast_fp16))[name = string("op_6044_cast_fp16")]; + string var_6046_equation_0 = const()[name = string("op_6046_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6046_cast_fp16 = einsum(equation = var_6046_equation_0, values = (var_5560_cast_fp16, var_5962_cast_fp16))[name = string("op_6046_cast_fp16")]; + string var_6048_equation_0 = const()[name = string("op_6048_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6048_cast_fp16 = einsum(equation = var_6048_equation_0, values = (var_5564_cast_fp16, var_5963_cast_fp16))[name = string("op_6048_cast_fp16")]; + string var_6050_equation_0 = const()[name = string("op_6050_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6050_cast_fp16 = einsum(equation = var_6050_equation_0, values = (var_5564_cast_fp16, var_5964_cast_fp16))[name = string("op_6050_cast_fp16")]; + string var_6052_equation_0 = const()[name = string("op_6052_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6052_cast_fp16 = einsum(equation = var_6052_equation_0, values = (var_5564_cast_fp16, var_5965_cast_fp16))[name = string("op_6052_cast_fp16")]; + string var_6054_equation_0 = const()[name = string("op_6054_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6054_cast_fp16 = einsum(equation = var_6054_equation_0, values = (var_5564_cast_fp16, var_5966_cast_fp16))[name = string("op_6054_cast_fp16")]; + string var_6056_equation_0 = const()[name = string("op_6056_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6056_cast_fp16 = einsum(equation = var_6056_equation_0, values = (var_5568_cast_fp16, var_5967_cast_fp16))[name = string("op_6056_cast_fp16")]; + string var_6058_equation_0 = const()[name = string("op_6058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6058_cast_fp16 = einsum(equation = var_6058_equation_0, values = (var_5568_cast_fp16, var_5968_cast_fp16))[name = string("op_6058_cast_fp16")]; + string var_6060_equation_0 = const()[name = string("op_6060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6060_cast_fp16 = einsum(equation = var_6060_equation_0, values = (var_5568_cast_fp16, var_5969_cast_fp16))[name = string("op_6060_cast_fp16")]; + string var_6062_equation_0 = const()[name = string("op_6062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6062_cast_fp16 = einsum(equation = var_6062_equation_0, values = (var_5568_cast_fp16, var_5970_cast_fp16))[name = string("op_6062_cast_fp16")]; + string var_6064_equation_0 = const()[name = string("op_6064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6064_cast_fp16 = einsum(equation = var_6064_equation_0, values = (var_5572_cast_fp16, var_5971_cast_fp16))[name = string("op_6064_cast_fp16")]; + string var_6066_equation_0 = const()[name = string("op_6066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6066_cast_fp16 = einsum(equation = var_6066_equation_0, values = (var_5572_cast_fp16, var_5972_cast_fp16))[name = string("op_6066_cast_fp16")]; + string var_6068_equation_0 = const()[name = string("op_6068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6068_cast_fp16 = einsum(equation = var_6068_equation_0, values = (var_5572_cast_fp16, var_5973_cast_fp16))[name = string("op_6068_cast_fp16")]; + string var_6070_equation_0 = const()[name = string("op_6070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6070_cast_fp16 = einsum(equation = var_6070_equation_0, values = (var_5572_cast_fp16, var_5974_cast_fp16))[name = string("op_6070_cast_fp16")]; + string var_6072_equation_0 = const()[name = string("op_6072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6072_cast_fp16 = einsum(equation = var_6072_equation_0, values = (var_5576_cast_fp16, var_5975_cast_fp16))[name = string("op_6072_cast_fp16")]; + string var_6074_equation_0 = const()[name = string("op_6074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6074_cast_fp16 = einsum(equation = var_6074_equation_0, values = (var_5576_cast_fp16, var_5976_cast_fp16))[name = string("op_6074_cast_fp16")]; + string var_6076_equation_0 = const()[name = string("op_6076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6076_cast_fp16 = einsum(equation = var_6076_equation_0, values = (var_5576_cast_fp16, var_5977_cast_fp16))[name = string("op_6076_cast_fp16")]; + string var_6078_equation_0 = const()[name = string("op_6078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6078_cast_fp16 = einsum(equation = var_6078_equation_0, values = (var_5576_cast_fp16, var_5978_cast_fp16))[name = string("op_6078_cast_fp16")]; + string var_6080_equation_0 = const()[name = string("op_6080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6080_cast_fp16 = einsum(equation = var_6080_equation_0, values = (var_5580_cast_fp16, var_5979_cast_fp16))[name = string("op_6080_cast_fp16")]; + string var_6082_equation_0 = const()[name = string("op_6082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6082_cast_fp16 = einsum(equation = var_6082_equation_0, values = (var_5580_cast_fp16, var_5980_cast_fp16))[name = string("op_6082_cast_fp16")]; + string var_6084_equation_0 = const()[name = string("op_6084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6084_cast_fp16 = einsum(equation = var_6084_equation_0, values = (var_5580_cast_fp16, var_5981_cast_fp16))[name = string("op_6084_cast_fp16")]; + string var_6086_equation_0 = const()[name = string("op_6086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6086_cast_fp16 = einsum(equation = var_6086_equation_0, values = (var_5580_cast_fp16, var_5982_cast_fp16))[name = string("op_6086_cast_fp16")]; + string var_6088_equation_0 = const()[name = string("op_6088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6088_cast_fp16 = einsum(equation = var_6088_equation_0, values = (var_5584_cast_fp16, var_5983_cast_fp16))[name = string("op_6088_cast_fp16")]; + string var_6090_equation_0 = const()[name = string("op_6090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6090_cast_fp16 = einsum(equation = var_6090_equation_0, values = (var_5584_cast_fp16, var_5984_cast_fp16))[name = string("op_6090_cast_fp16")]; + string var_6092_equation_0 = const()[name = string("op_6092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6092_cast_fp16 = einsum(equation = var_6092_equation_0, values = (var_5584_cast_fp16, var_5985_cast_fp16))[name = string("op_6092_cast_fp16")]; + string var_6094_equation_0 = const()[name = string("op_6094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6094_cast_fp16 = einsum(equation = var_6094_equation_0, values = (var_5584_cast_fp16, var_5986_cast_fp16))[name = string("op_6094_cast_fp16")]; + string var_6096_equation_0 = const()[name = string("op_6096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6096_cast_fp16 = einsum(equation = var_6096_equation_0, values = (var_5588_cast_fp16, var_5987_cast_fp16))[name = string("op_6096_cast_fp16")]; + string var_6098_equation_0 = const()[name = string("op_6098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6098_cast_fp16 = einsum(equation = var_6098_equation_0, values = (var_5588_cast_fp16, var_5988_cast_fp16))[name = string("op_6098_cast_fp16")]; + string var_6100_equation_0 = const()[name = string("op_6100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6100_cast_fp16 = einsum(equation = var_6100_equation_0, values = (var_5588_cast_fp16, var_5989_cast_fp16))[name = string("op_6100_cast_fp16")]; + string var_6102_equation_0 = const()[name = string("op_6102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6102_cast_fp16 = einsum(equation = var_6102_equation_0, values = (var_5588_cast_fp16, var_5990_cast_fp16))[name = string("op_6102_cast_fp16")]; + string var_6104_equation_0 = const()[name = string("op_6104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6104_cast_fp16 = einsum(equation = var_6104_equation_0, values = (var_5592_cast_fp16, var_5991_cast_fp16))[name = string("op_6104_cast_fp16")]; + string var_6106_equation_0 = const()[name = string("op_6106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6106_cast_fp16 = einsum(equation = var_6106_equation_0, values = (var_5592_cast_fp16, var_5992_cast_fp16))[name = string("op_6106_cast_fp16")]; + string var_6108_equation_0 = const()[name = string("op_6108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6108_cast_fp16 = einsum(equation = var_6108_equation_0, values = (var_5592_cast_fp16, var_5993_cast_fp16))[name = string("op_6108_cast_fp16")]; + string var_6110_equation_0 = const()[name = string("op_6110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6110_cast_fp16 = einsum(equation = var_6110_equation_0, values = (var_5592_cast_fp16, var_5994_cast_fp16))[name = string("op_6110_cast_fp16")]; + string var_6112_equation_0 = const()[name = string("op_6112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6112_cast_fp16 = einsum(equation = var_6112_equation_0, values = (var_5596_cast_fp16, var_5995_cast_fp16))[name = string("op_6112_cast_fp16")]; + string var_6114_equation_0 = const()[name = string("op_6114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6114_cast_fp16 = einsum(equation = var_6114_equation_0, values = (var_5596_cast_fp16, var_5996_cast_fp16))[name = string("op_6114_cast_fp16")]; + string var_6116_equation_0 = const()[name = string("op_6116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6116_cast_fp16 = einsum(equation = var_6116_equation_0, values = (var_5596_cast_fp16, var_5997_cast_fp16))[name = string("op_6116_cast_fp16")]; + string var_6118_equation_0 = const()[name = string("op_6118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6118_cast_fp16 = einsum(equation = var_6118_equation_0, values = (var_5596_cast_fp16, var_5998_cast_fp16))[name = string("op_6118_cast_fp16")]; + string var_6120_equation_0 = const()[name = string("op_6120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6120_cast_fp16 = einsum(equation = var_6120_equation_0, values = (var_5600_cast_fp16, var_5999_cast_fp16))[name = string("op_6120_cast_fp16")]; + string var_6122_equation_0 = const()[name = string("op_6122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6122_cast_fp16 = einsum(equation = var_6122_equation_0, values = (var_5600_cast_fp16, var_6000_cast_fp16))[name = string("op_6122_cast_fp16")]; + string var_6124_equation_0 = const()[name = string("op_6124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6124_cast_fp16 = einsum(equation = var_6124_equation_0, values = (var_5600_cast_fp16, var_6001_cast_fp16))[name = string("op_6124_cast_fp16")]; + string var_6126_equation_0 = const()[name = string("op_6126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6126_cast_fp16 = einsum(equation = var_6126_equation_0, values = (var_5600_cast_fp16, var_6002_cast_fp16))[name = string("op_6126_cast_fp16")]; + string var_6128_equation_0 = const()[name = string("op_6128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6128_cast_fp16 = einsum(equation = var_6128_equation_0, values = (var_5604_cast_fp16, var_6003_cast_fp16))[name = string("op_6128_cast_fp16")]; + string var_6130_equation_0 = const()[name = string("op_6130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6130_cast_fp16 = einsum(equation = var_6130_equation_0, values = (var_5604_cast_fp16, var_6004_cast_fp16))[name = string("op_6130_cast_fp16")]; + string var_6132_equation_0 = const()[name = string("op_6132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6132_cast_fp16 = einsum(equation = var_6132_equation_0, values = (var_5604_cast_fp16, var_6005_cast_fp16))[name = string("op_6132_cast_fp16")]; + string var_6134_equation_0 = const()[name = string("op_6134_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6134_cast_fp16 = einsum(equation = var_6134_equation_0, values = (var_5604_cast_fp16, var_6006_cast_fp16))[name = string("op_6134_cast_fp16")]; + string var_6136_equation_0 = const()[name = string("op_6136_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6136_cast_fp16 = einsum(equation = var_6136_equation_0, values = (var_5608_cast_fp16, var_6007_cast_fp16))[name = string("op_6136_cast_fp16")]; + string var_6138_equation_0 = const()[name = string("op_6138_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6138_cast_fp16 = einsum(equation = var_6138_equation_0, values = (var_5608_cast_fp16, var_6008_cast_fp16))[name = string("op_6138_cast_fp16")]; + string var_6140_equation_0 = const()[name = string("op_6140_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6140_cast_fp16 = einsum(equation = var_6140_equation_0, values = (var_5608_cast_fp16, var_6009_cast_fp16))[name = string("op_6140_cast_fp16")]; + string var_6142_equation_0 = const()[name = string("op_6142_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6142_cast_fp16 = einsum(equation = var_6142_equation_0, values = (var_5608_cast_fp16, var_6010_cast_fp16))[name = string("op_6142_cast_fp16")]; + string var_6144_equation_0 = const()[name = string("op_6144_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6144_cast_fp16 = einsum(equation = var_6144_equation_0, values = (var_5612_cast_fp16, var_6011_cast_fp16))[name = string("op_6144_cast_fp16")]; + string var_6146_equation_0 = const()[name = string("op_6146_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6146_cast_fp16 = einsum(equation = var_6146_equation_0, values = (var_5612_cast_fp16, var_6012_cast_fp16))[name = string("op_6146_cast_fp16")]; + string var_6148_equation_0 = const()[name = string("op_6148_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6148_cast_fp16 = einsum(equation = var_6148_equation_0, values = (var_5612_cast_fp16, var_6013_cast_fp16))[name = string("op_6148_cast_fp16")]; + string var_6150_equation_0 = const()[name = string("op_6150_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6150_cast_fp16 = einsum(equation = var_6150_equation_0, values = (var_5612_cast_fp16, var_6014_cast_fp16))[name = string("op_6150_cast_fp16")]; + string var_6152_equation_0 = const()[name = string("op_6152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6152_cast_fp16 = einsum(equation = var_6152_equation_0, values = (var_5616_cast_fp16, var_6015_cast_fp16))[name = string("op_6152_cast_fp16")]; + string var_6154_equation_0 = const()[name = string("op_6154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6154_cast_fp16 = einsum(equation = var_6154_equation_0, values = (var_5616_cast_fp16, var_6016_cast_fp16))[name = string("op_6154_cast_fp16")]; + string var_6156_equation_0 = const()[name = string("op_6156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6156_cast_fp16 = einsum(equation = var_6156_equation_0, values = (var_5616_cast_fp16, var_6017_cast_fp16))[name = string("op_6156_cast_fp16")]; + string var_6158_equation_0 = const()[name = string("op_6158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6158_cast_fp16 = einsum(equation = var_6158_equation_0, values = (var_5616_cast_fp16, var_6018_cast_fp16))[name = string("op_6158_cast_fp16")]; + string var_6160_equation_0 = const()[name = string("op_6160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6160_cast_fp16 = einsum(equation = var_6160_equation_0, values = (var_5620_cast_fp16, var_6019_cast_fp16))[name = string("op_6160_cast_fp16")]; + string var_6162_equation_0 = const()[name = string("op_6162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6162_cast_fp16 = einsum(equation = var_6162_equation_0, values = (var_5620_cast_fp16, var_6020_cast_fp16))[name = string("op_6162_cast_fp16")]; + string var_6164_equation_0 = const()[name = string("op_6164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6164_cast_fp16 = einsum(equation = var_6164_equation_0, values = (var_5620_cast_fp16, var_6021_cast_fp16))[name = string("op_6164_cast_fp16")]; + string var_6166_equation_0 = const()[name = string("op_6166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6166_cast_fp16 = einsum(equation = var_6166_equation_0, values = (var_5620_cast_fp16, var_6022_cast_fp16))[name = string("op_6166_cast_fp16")]; + string var_6168_equation_0 = const()[name = string("op_6168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6168_cast_fp16 = einsum(equation = var_6168_equation_0, values = (var_5624_cast_fp16, var_6023_cast_fp16))[name = string("op_6168_cast_fp16")]; + string var_6170_equation_0 = const()[name = string("op_6170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6170_cast_fp16 = einsum(equation = var_6170_equation_0, values = (var_5624_cast_fp16, var_6024_cast_fp16))[name = string("op_6170_cast_fp16")]; + string var_6172_equation_0 = const()[name = string("op_6172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6172_cast_fp16 = einsum(equation = var_6172_equation_0, values = (var_5624_cast_fp16, var_6025_cast_fp16))[name = string("op_6172_cast_fp16")]; + string var_6174_equation_0 = const()[name = string("op_6174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6174_cast_fp16 = einsum(equation = var_6174_equation_0, values = (var_5624_cast_fp16, var_6026_cast_fp16))[name = string("op_6174_cast_fp16")]; + string var_6176_equation_0 = const()[name = string("op_6176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6176_cast_fp16 = einsum(equation = var_6176_equation_0, values = (var_5628_cast_fp16, var_6027_cast_fp16))[name = string("op_6176_cast_fp16")]; + string var_6178_equation_0 = const()[name = string("op_6178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6178_cast_fp16 = einsum(equation = var_6178_equation_0, values = (var_5628_cast_fp16, var_6028_cast_fp16))[name = string("op_6178_cast_fp16")]; + string var_6180_equation_0 = const()[name = string("op_6180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6180_cast_fp16 = einsum(equation = var_6180_equation_0, values = (var_5628_cast_fp16, var_6029_cast_fp16))[name = string("op_6180_cast_fp16")]; + string var_6182_equation_0 = const()[name = string("op_6182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6182_cast_fp16 = einsum(equation = var_6182_equation_0, values = (var_5628_cast_fp16, var_6030_cast_fp16))[name = string("op_6182_cast_fp16")]; + string var_6184_equation_0 = const()[name = string("op_6184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6184_cast_fp16 = einsum(equation = var_6184_equation_0, values = (var_5632_cast_fp16, var_6031_cast_fp16))[name = string("op_6184_cast_fp16")]; + string var_6186_equation_0 = const()[name = string("op_6186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6186_cast_fp16 = einsum(equation = var_6186_equation_0, values = (var_5632_cast_fp16, var_6032_cast_fp16))[name = string("op_6186_cast_fp16")]; + string var_6188_equation_0 = const()[name = string("op_6188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6188_cast_fp16 = einsum(equation = var_6188_equation_0, values = (var_5632_cast_fp16, var_6033_cast_fp16))[name = string("op_6188_cast_fp16")]; + string var_6190_equation_0 = const()[name = string("op_6190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6190_cast_fp16 = einsum(equation = var_6190_equation_0, values = (var_5632_cast_fp16, var_6034_cast_fp16))[name = string("op_6190_cast_fp16")]; + string var_6192_equation_0 = const()[name = string("op_6192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6192_cast_fp16 = einsum(equation = var_6192_equation_0, values = (var_5636_cast_fp16, var_6035_cast_fp16))[name = string("op_6192_cast_fp16")]; + string var_6194_equation_0 = const()[name = string("op_6194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6194_cast_fp16 = einsum(equation = var_6194_equation_0, values = (var_5636_cast_fp16, var_6036_cast_fp16))[name = string("op_6194_cast_fp16")]; + string var_6196_equation_0 = const()[name = string("op_6196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6196_cast_fp16 = einsum(equation = var_6196_equation_0, values = (var_5636_cast_fp16, var_6037_cast_fp16))[name = string("op_6196_cast_fp16")]; + string var_6198_equation_0 = const()[name = string("op_6198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6198_cast_fp16 = einsum(equation = var_6198_equation_0, values = (var_5636_cast_fp16, var_6038_cast_fp16))[name = string("op_6198_cast_fp16")]; + bool var_6200_interleave_0 = const()[name = string("op_6200_interleave_0"), val = bool(false)]; + tensor var_6200_cast_fp16 = concat(axis = var_4759, interleave = var_6200_interleave_0, values = (var_6040_cast_fp16, var_6042_cast_fp16, var_6044_cast_fp16, var_6046_cast_fp16))[name = string("op_6200_cast_fp16")]; + bool var_6202_interleave_0 = const()[name = string("op_6202_interleave_0"), val = bool(false)]; + tensor var_6202_cast_fp16 = concat(axis = var_4759, interleave = var_6202_interleave_0, values = (var_6048_cast_fp16, var_6050_cast_fp16, var_6052_cast_fp16, var_6054_cast_fp16))[name = string("op_6202_cast_fp16")]; + bool var_6204_interleave_0 = const()[name = string("op_6204_interleave_0"), val = bool(false)]; + tensor var_6204_cast_fp16 = concat(axis = var_4759, interleave = var_6204_interleave_0, values = (var_6056_cast_fp16, var_6058_cast_fp16, var_6060_cast_fp16, var_6062_cast_fp16))[name = string("op_6204_cast_fp16")]; + bool var_6206_interleave_0 = const()[name = string("op_6206_interleave_0"), val = bool(false)]; + tensor var_6206_cast_fp16 = concat(axis = var_4759, interleave = var_6206_interleave_0, values = (var_6064_cast_fp16, var_6066_cast_fp16, var_6068_cast_fp16, var_6070_cast_fp16))[name = string("op_6206_cast_fp16")]; + bool var_6208_interleave_0 = const()[name = string("op_6208_interleave_0"), val = bool(false)]; + tensor var_6208_cast_fp16 = concat(axis = var_4759, interleave = var_6208_interleave_0, values = (var_6072_cast_fp16, var_6074_cast_fp16, var_6076_cast_fp16, var_6078_cast_fp16))[name = string("op_6208_cast_fp16")]; + bool var_6210_interleave_0 = const()[name = string("op_6210_interleave_0"), val = bool(false)]; + tensor var_6210_cast_fp16 = concat(axis = var_4759, interleave = var_6210_interleave_0, values = (var_6080_cast_fp16, var_6082_cast_fp16, var_6084_cast_fp16, var_6086_cast_fp16))[name = string("op_6210_cast_fp16")]; + bool var_6212_interleave_0 = const()[name = string("op_6212_interleave_0"), val = bool(false)]; + tensor var_6212_cast_fp16 = concat(axis = var_4759, interleave = var_6212_interleave_0, values = (var_6088_cast_fp16, var_6090_cast_fp16, var_6092_cast_fp16, var_6094_cast_fp16))[name = string("op_6212_cast_fp16")]; + bool var_6214_interleave_0 = const()[name = string("op_6214_interleave_0"), val = bool(false)]; + tensor var_6214_cast_fp16 = concat(axis = var_4759, interleave = var_6214_interleave_0, values = (var_6096_cast_fp16, var_6098_cast_fp16, var_6100_cast_fp16, var_6102_cast_fp16))[name = string("op_6214_cast_fp16")]; + bool var_6216_interleave_0 = const()[name = string("op_6216_interleave_0"), val = bool(false)]; + tensor var_6216_cast_fp16 = concat(axis = var_4759, interleave = var_6216_interleave_0, values = (var_6104_cast_fp16, var_6106_cast_fp16, var_6108_cast_fp16, var_6110_cast_fp16))[name = string("op_6216_cast_fp16")]; + bool var_6218_interleave_0 = const()[name = string("op_6218_interleave_0"), val = bool(false)]; + tensor var_6218_cast_fp16 = concat(axis = var_4759, interleave = var_6218_interleave_0, values = (var_6112_cast_fp16, var_6114_cast_fp16, var_6116_cast_fp16, var_6118_cast_fp16))[name = string("op_6218_cast_fp16")]; + bool var_6220_interleave_0 = const()[name = string("op_6220_interleave_0"), val = bool(false)]; + tensor var_6220_cast_fp16 = concat(axis = var_4759, interleave = var_6220_interleave_0, values = (var_6120_cast_fp16, var_6122_cast_fp16, var_6124_cast_fp16, var_6126_cast_fp16))[name = string("op_6220_cast_fp16")]; + bool var_6222_interleave_0 = const()[name = string("op_6222_interleave_0"), val = bool(false)]; + tensor var_6222_cast_fp16 = concat(axis = var_4759, interleave = var_6222_interleave_0, values = (var_6128_cast_fp16, var_6130_cast_fp16, var_6132_cast_fp16, var_6134_cast_fp16))[name = string("op_6222_cast_fp16")]; + bool var_6224_interleave_0 = const()[name = string("op_6224_interleave_0"), val = bool(false)]; + tensor var_6224_cast_fp16 = concat(axis = var_4759, interleave = var_6224_interleave_0, values = (var_6136_cast_fp16, var_6138_cast_fp16, var_6140_cast_fp16, var_6142_cast_fp16))[name = string("op_6224_cast_fp16")]; + bool var_6226_interleave_0 = const()[name = string("op_6226_interleave_0"), val = bool(false)]; + tensor var_6226_cast_fp16 = concat(axis = var_4759, interleave = var_6226_interleave_0, values = (var_6144_cast_fp16, var_6146_cast_fp16, var_6148_cast_fp16, var_6150_cast_fp16))[name = string("op_6226_cast_fp16")]; + bool var_6228_interleave_0 = const()[name = string("op_6228_interleave_0"), val = bool(false)]; + tensor var_6228_cast_fp16 = concat(axis = var_4759, interleave = var_6228_interleave_0, values = (var_6152_cast_fp16, var_6154_cast_fp16, var_6156_cast_fp16, var_6158_cast_fp16))[name = string("op_6228_cast_fp16")]; + bool var_6230_interleave_0 = const()[name = string("op_6230_interleave_0"), val = bool(false)]; + tensor var_6230_cast_fp16 = concat(axis = var_4759, interleave = var_6230_interleave_0, values = (var_6160_cast_fp16, var_6162_cast_fp16, var_6164_cast_fp16, var_6166_cast_fp16))[name = string("op_6230_cast_fp16")]; + bool var_6232_interleave_0 = const()[name = string("op_6232_interleave_0"), val = bool(false)]; + tensor var_6232_cast_fp16 = concat(axis = var_4759, interleave = var_6232_interleave_0, values = (var_6168_cast_fp16, var_6170_cast_fp16, var_6172_cast_fp16, var_6174_cast_fp16))[name = string("op_6232_cast_fp16")]; + bool var_6234_interleave_0 = const()[name = string("op_6234_interleave_0"), val = bool(false)]; + tensor var_6234_cast_fp16 = concat(axis = var_4759, interleave = var_6234_interleave_0, values = (var_6176_cast_fp16, var_6178_cast_fp16, var_6180_cast_fp16, var_6182_cast_fp16))[name = string("op_6234_cast_fp16")]; + bool var_6236_interleave_0 = const()[name = string("op_6236_interleave_0"), val = bool(false)]; + tensor var_6236_cast_fp16 = concat(axis = var_4759, interleave = var_6236_interleave_0, values = (var_6184_cast_fp16, var_6186_cast_fp16, var_6188_cast_fp16, var_6190_cast_fp16))[name = string("op_6236_cast_fp16")]; + bool var_6238_interleave_0 = const()[name = string("op_6238_interleave_0"), val = bool(false)]; + tensor var_6238_cast_fp16 = concat(axis = var_4759, interleave = var_6238_interleave_0, values = (var_6192_cast_fp16, var_6194_cast_fp16, var_6196_cast_fp16, var_6198_cast_fp16))[name = string("op_6238_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_4784, interleave = input_25_interleave_0, values = (var_6200_cast_fp16, var_6202_cast_fp16, var_6204_cast_fp16, var_6206_cast_fp16, var_6208_cast_fp16, var_6210_cast_fp16, var_6212_cast_fp16, var_6214_cast_fp16, var_6216_cast_fp16, var_6218_cast_fp16, var_6220_cast_fp16, var_6222_cast_fp16, var_6224_cast_fp16, var_6226_cast_fp16, var_6228_cast_fp16, var_6230_cast_fp16, var_6232_cast_fp16, var_6234_cast_fp16, var_6236_cast_fp16, var_6238_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142565120)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145841984)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_6257_to_fp16 = const()[name = string("op_6257_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_6257_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145844608)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145847232)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145849856)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158957120)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158967424)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172074688)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_6286 = const()[name = string("op_6286"), val = int32(3)]; + int32 var_6311 = const()[name = string("op_6311"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_6328_to_fp16 = const()[name = string("op_6328_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_6328_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172077312)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172079936)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172082560)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175359424)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175362048)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178638912)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181915776)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_6366_begin_0 = const()[name = string("op_6366_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6366_end_0 = const()[name = string("op_6366_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6366_end_mask_0 = const()[name = string("op_6366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6366_cast_fp16 = slice_by_index(begin = var_6366_begin_0, end = var_6366_end_0, end_mask = var_6366_end_mask_0, x = query_9_cast_fp16)[name = string("op_6366_cast_fp16")]; + tensor var_6370_begin_0 = const()[name = string("op_6370_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6370_end_0 = const()[name = string("op_6370_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6370_end_mask_0 = const()[name = string("op_6370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6370_cast_fp16 = slice_by_index(begin = var_6370_begin_0, end = var_6370_end_0, end_mask = var_6370_end_mask_0, x = query_9_cast_fp16)[name = string("op_6370_cast_fp16")]; + tensor var_6374_begin_0 = const()[name = string("op_6374_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6374_end_0 = const()[name = string("op_6374_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6374_end_mask_0 = const()[name = string("op_6374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6374_cast_fp16 = slice_by_index(begin = var_6374_begin_0, end = var_6374_end_0, end_mask = var_6374_end_mask_0, x = query_9_cast_fp16)[name = string("op_6374_cast_fp16")]; + tensor var_6378_begin_0 = const()[name = string("op_6378_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6378_end_0 = const()[name = string("op_6378_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6378_end_mask_0 = const()[name = string("op_6378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6378_cast_fp16 = slice_by_index(begin = var_6378_begin_0, end = var_6378_end_0, end_mask = var_6378_end_mask_0, x = query_9_cast_fp16)[name = string("op_6378_cast_fp16")]; + tensor var_6382_begin_0 = const()[name = string("op_6382_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6382_end_0 = const()[name = string("op_6382_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6382_end_mask_0 = const()[name = string("op_6382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6382_cast_fp16 = slice_by_index(begin = var_6382_begin_0, end = var_6382_end_0, end_mask = var_6382_end_mask_0, x = query_9_cast_fp16)[name = string("op_6382_cast_fp16")]; + tensor var_6386_begin_0 = const()[name = string("op_6386_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6386_end_0 = const()[name = string("op_6386_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6386_end_mask_0 = const()[name = string("op_6386_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6386_cast_fp16 = slice_by_index(begin = var_6386_begin_0, end = var_6386_end_0, end_mask = var_6386_end_mask_0, x = query_9_cast_fp16)[name = string("op_6386_cast_fp16")]; + tensor var_6390_begin_0 = const()[name = string("op_6390_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6390_end_0 = const()[name = string("op_6390_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6390_end_mask_0 = const()[name = string("op_6390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6390_cast_fp16 = slice_by_index(begin = var_6390_begin_0, end = var_6390_end_0, end_mask = var_6390_end_mask_0, x = query_9_cast_fp16)[name = string("op_6390_cast_fp16")]; + tensor var_6394_begin_0 = const()[name = string("op_6394_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6394_end_0 = const()[name = string("op_6394_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6394_end_mask_0 = const()[name = string("op_6394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6394_cast_fp16 = slice_by_index(begin = var_6394_begin_0, end = var_6394_end_0, end_mask = var_6394_end_mask_0, x = query_9_cast_fp16)[name = string("op_6394_cast_fp16")]; + tensor var_6398_begin_0 = const()[name = string("op_6398_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6398_end_0 = const()[name = string("op_6398_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6398_end_mask_0 = const()[name = string("op_6398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6398_cast_fp16 = slice_by_index(begin = var_6398_begin_0, end = var_6398_end_0, end_mask = var_6398_end_mask_0, x = query_9_cast_fp16)[name = string("op_6398_cast_fp16")]; + tensor var_6402_begin_0 = const()[name = string("op_6402_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6402_end_0 = const()[name = string("op_6402_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6402_end_mask_0 = const()[name = string("op_6402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6402_cast_fp16 = slice_by_index(begin = var_6402_begin_0, end = var_6402_end_0, end_mask = var_6402_end_mask_0, x = query_9_cast_fp16)[name = string("op_6402_cast_fp16")]; + tensor var_6406_begin_0 = const()[name = string("op_6406_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6406_end_0 = const()[name = string("op_6406_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6406_end_mask_0 = const()[name = string("op_6406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6406_cast_fp16 = slice_by_index(begin = var_6406_begin_0, end = var_6406_end_0, end_mask = var_6406_end_mask_0, x = query_9_cast_fp16)[name = string("op_6406_cast_fp16")]; + tensor var_6410_begin_0 = const()[name = string("op_6410_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6410_end_0 = const()[name = string("op_6410_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6410_end_mask_0 = const()[name = string("op_6410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6410_cast_fp16 = slice_by_index(begin = var_6410_begin_0, end = var_6410_end_0, end_mask = var_6410_end_mask_0, x = query_9_cast_fp16)[name = string("op_6410_cast_fp16")]; + tensor var_6414_begin_0 = const()[name = string("op_6414_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6414_end_0 = const()[name = string("op_6414_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_6414_end_mask_0 = const()[name = string("op_6414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6414_cast_fp16 = slice_by_index(begin = var_6414_begin_0, end = var_6414_end_0, end_mask = var_6414_end_mask_0, x = query_9_cast_fp16)[name = string("op_6414_cast_fp16")]; + tensor var_6418_begin_0 = const()[name = string("op_6418_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6418_end_0 = const()[name = string("op_6418_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_6418_end_mask_0 = const()[name = string("op_6418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6418_cast_fp16 = slice_by_index(begin = var_6418_begin_0, end = var_6418_end_0, end_mask = var_6418_end_mask_0, x = query_9_cast_fp16)[name = string("op_6418_cast_fp16")]; + tensor var_6422_begin_0 = const()[name = string("op_6422_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6422_end_0 = const()[name = string("op_6422_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_6422_end_mask_0 = const()[name = string("op_6422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6422_cast_fp16 = slice_by_index(begin = var_6422_begin_0, end = var_6422_end_0, end_mask = var_6422_end_mask_0, x = query_9_cast_fp16)[name = string("op_6422_cast_fp16")]; + tensor var_6426_begin_0 = const()[name = string("op_6426_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6426_end_0 = const()[name = string("op_6426_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_6426_end_mask_0 = const()[name = string("op_6426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6426_cast_fp16 = slice_by_index(begin = var_6426_begin_0, end = var_6426_end_0, end_mask = var_6426_end_mask_0, x = query_9_cast_fp16)[name = string("op_6426_cast_fp16")]; + tensor var_6430_begin_0 = const()[name = string("op_6430_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6430_end_0 = const()[name = string("op_6430_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_6430_end_mask_0 = const()[name = string("op_6430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6430_cast_fp16 = slice_by_index(begin = var_6430_begin_0, end = var_6430_end_0, end_mask = var_6430_end_mask_0, x = query_9_cast_fp16)[name = string("op_6430_cast_fp16")]; + tensor var_6434_begin_0 = const()[name = string("op_6434_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6434_end_0 = const()[name = string("op_6434_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_6434_end_mask_0 = const()[name = string("op_6434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6434_cast_fp16 = slice_by_index(begin = var_6434_begin_0, end = var_6434_end_0, end_mask = var_6434_end_mask_0, x = query_9_cast_fp16)[name = string("op_6434_cast_fp16")]; + tensor var_6438_begin_0 = const()[name = string("op_6438_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6438_end_0 = const()[name = string("op_6438_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_6438_end_mask_0 = const()[name = string("op_6438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6438_cast_fp16 = slice_by_index(begin = var_6438_begin_0, end = var_6438_end_0, end_mask = var_6438_end_mask_0, x = query_9_cast_fp16)[name = string("op_6438_cast_fp16")]; + tensor var_6442_begin_0 = const()[name = string("op_6442_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6442_end_0 = const()[name = string("op_6442_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_6442_end_mask_0 = const()[name = string("op_6442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6442_cast_fp16 = slice_by_index(begin = var_6442_begin_0, end = var_6442_end_0, end_mask = var_6442_end_mask_0, x = query_9_cast_fp16)[name = string("op_6442_cast_fp16")]; + tensor var_6451_begin_0 = const()[name = string("op_6451_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6451_end_0 = const()[name = string("op_6451_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6451_end_mask_0 = const()[name = string("op_6451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6451_cast_fp16 = slice_by_index(begin = var_6451_begin_0, end = var_6451_end_0, end_mask = var_6451_end_mask_0, x = var_6366_cast_fp16)[name = string("op_6451_cast_fp16")]; + tensor var_6458_begin_0 = const()[name = string("op_6458_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6458_end_0 = const()[name = string("op_6458_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6458_end_mask_0 = const()[name = string("op_6458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6458_cast_fp16 = slice_by_index(begin = var_6458_begin_0, end = var_6458_end_0, end_mask = var_6458_end_mask_0, x = var_6366_cast_fp16)[name = string("op_6458_cast_fp16")]; + tensor var_6465_begin_0 = const()[name = string("op_6465_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6465_end_0 = const()[name = string("op_6465_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6465_end_mask_0 = const()[name = string("op_6465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6465_cast_fp16 = slice_by_index(begin = var_6465_begin_0, end = var_6465_end_0, end_mask = var_6465_end_mask_0, x = var_6366_cast_fp16)[name = string("op_6465_cast_fp16")]; + tensor var_6472_begin_0 = const()[name = string("op_6472_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6472_end_0 = const()[name = string("op_6472_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6472_end_mask_0 = const()[name = string("op_6472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6472_cast_fp16 = slice_by_index(begin = var_6472_begin_0, end = var_6472_end_0, end_mask = var_6472_end_mask_0, x = var_6366_cast_fp16)[name = string("op_6472_cast_fp16")]; + tensor var_6479_begin_0 = const()[name = string("op_6479_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6479_end_0 = const()[name = string("op_6479_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6479_end_mask_0 = const()[name = string("op_6479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6479_cast_fp16 = slice_by_index(begin = var_6479_begin_0, end = var_6479_end_0, end_mask = var_6479_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6479_cast_fp16")]; + tensor var_6486_begin_0 = const()[name = string("op_6486_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6486_end_0 = const()[name = string("op_6486_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6486_end_mask_0 = const()[name = string("op_6486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6486_cast_fp16 = slice_by_index(begin = var_6486_begin_0, end = var_6486_end_0, end_mask = var_6486_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6486_cast_fp16")]; + tensor var_6493_begin_0 = const()[name = string("op_6493_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6493_end_0 = const()[name = string("op_6493_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6493_end_mask_0 = const()[name = string("op_6493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6493_cast_fp16 = slice_by_index(begin = var_6493_begin_0, end = var_6493_end_0, end_mask = var_6493_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6493_cast_fp16")]; + tensor var_6500_begin_0 = const()[name = string("op_6500_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6500_end_0 = const()[name = string("op_6500_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6500_end_mask_0 = const()[name = string("op_6500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6500_cast_fp16 = slice_by_index(begin = var_6500_begin_0, end = var_6500_end_0, end_mask = var_6500_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6500_cast_fp16")]; + tensor var_6507_begin_0 = const()[name = string("op_6507_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6507_end_0 = const()[name = string("op_6507_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6507_end_mask_0 = const()[name = string("op_6507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6507_cast_fp16 = slice_by_index(begin = var_6507_begin_0, end = var_6507_end_0, end_mask = var_6507_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6507_cast_fp16")]; + tensor var_6514_begin_0 = const()[name = string("op_6514_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6514_end_0 = const()[name = string("op_6514_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6514_end_mask_0 = const()[name = string("op_6514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6514_cast_fp16 = slice_by_index(begin = var_6514_begin_0, end = var_6514_end_0, end_mask = var_6514_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6514_cast_fp16")]; + tensor var_6521_begin_0 = const()[name = string("op_6521_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6521_end_0 = const()[name = string("op_6521_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6521_end_mask_0 = const()[name = string("op_6521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6521_cast_fp16 = slice_by_index(begin = var_6521_begin_0, end = var_6521_end_0, end_mask = var_6521_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6521_cast_fp16")]; + tensor var_6528_begin_0 = const()[name = string("op_6528_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6528_end_0 = const()[name = string("op_6528_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6528_end_mask_0 = const()[name = string("op_6528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6528_cast_fp16 = slice_by_index(begin = var_6528_begin_0, end = var_6528_end_0, end_mask = var_6528_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6528_cast_fp16")]; + tensor var_6535_begin_0 = const()[name = string("op_6535_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6535_end_0 = const()[name = string("op_6535_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6535_end_mask_0 = const()[name = string("op_6535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6535_cast_fp16 = slice_by_index(begin = var_6535_begin_0, end = var_6535_end_0, end_mask = var_6535_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6535_cast_fp16")]; + tensor var_6542_begin_0 = const()[name = string("op_6542_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6542_end_0 = const()[name = string("op_6542_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6542_end_mask_0 = const()[name = string("op_6542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6542_cast_fp16 = slice_by_index(begin = var_6542_begin_0, end = var_6542_end_0, end_mask = var_6542_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6542_cast_fp16")]; + tensor var_6549_begin_0 = const()[name = string("op_6549_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6549_end_0 = const()[name = string("op_6549_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6549_end_mask_0 = const()[name = string("op_6549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6549_cast_fp16 = slice_by_index(begin = var_6549_begin_0, end = var_6549_end_0, end_mask = var_6549_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6549_cast_fp16")]; + tensor var_6556_begin_0 = const()[name = string("op_6556_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6556_end_0 = const()[name = string("op_6556_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6556_end_mask_0 = const()[name = string("op_6556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6556_cast_fp16 = slice_by_index(begin = var_6556_begin_0, end = var_6556_end_0, end_mask = var_6556_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6556_cast_fp16")]; + tensor var_6563_begin_0 = const()[name = string("op_6563_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6563_end_0 = const()[name = string("op_6563_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6563_end_mask_0 = const()[name = string("op_6563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6563_cast_fp16 = slice_by_index(begin = var_6563_begin_0, end = var_6563_end_0, end_mask = var_6563_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6563_cast_fp16")]; + tensor var_6570_begin_0 = const()[name = string("op_6570_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6570_end_0 = const()[name = string("op_6570_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6570_end_mask_0 = const()[name = string("op_6570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6570_cast_fp16 = slice_by_index(begin = var_6570_begin_0, end = var_6570_end_0, end_mask = var_6570_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6570_cast_fp16")]; + tensor var_6577_begin_0 = const()[name = string("op_6577_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6577_end_0 = const()[name = string("op_6577_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6577_end_mask_0 = const()[name = string("op_6577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6577_cast_fp16 = slice_by_index(begin = var_6577_begin_0, end = var_6577_end_0, end_mask = var_6577_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6577_cast_fp16")]; + tensor var_6584_begin_0 = const()[name = string("op_6584_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6584_end_0 = const()[name = string("op_6584_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6584_end_mask_0 = const()[name = string("op_6584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6584_cast_fp16 = slice_by_index(begin = var_6584_begin_0, end = var_6584_end_0, end_mask = var_6584_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6584_cast_fp16")]; + tensor var_6591_begin_0 = const()[name = string("op_6591_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6591_end_0 = const()[name = string("op_6591_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6591_end_mask_0 = const()[name = string("op_6591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6591_cast_fp16 = slice_by_index(begin = var_6591_begin_0, end = var_6591_end_0, end_mask = var_6591_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6591_cast_fp16")]; + tensor var_6598_begin_0 = const()[name = string("op_6598_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6598_end_0 = const()[name = string("op_6598_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6598_end_mask_0 = const()[name = string("op_6598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6598_cast_fp16 = slice_by_index(begin = var_6598_begin_0, end = var_6598_end_0, end_mask = var_6598_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6598_cast_fp16")]; + tensor var_6605_begin_0 = const()[name = string("op_6605_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6605_end_0 = const()[name = string("op_6605_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6605_end_mask_0 = const()[name = string("op_6605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6605_cast_fp16 = slice_by_index(begin = var_6605_begin_0, end = var_6605_end_0, end_mask = var_6605_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6605_cast_fp16")]; + tensor var_6612_begin_0 = const()[name = string("op_6612_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6612_end_0 = const()[name = string("op_6612_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6612_end_mask_0 = const()[name = string("op_6612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6612_cast_fp16 = slice_by_index(begin = var_6612_begin_0, end = var_6612_end_0, end_mask = var_6612_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6612_cast_fp16")]; + tensor var_6619_begin_0 = const()[name = string("op_6619_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6619_end_0 = const()[name = string("op_6619_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6619_end_mask_0 = const()[name = string("op_6619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6619_cast_fp16 = slice_by_index(begin = var_6619_begin_0, end = var_6619_end_0, end_mask = var_6619_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6619_cast_fp16")]; + tensor var_6626_begin_0 = const()[name = string("op_6626_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6626_end_0 = const()[name = string("op_6626_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6626_end_mask_0 = const()[name = string("op_6626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6626_cast_fp16 = slice_by_index(begin = var_6626_begin_0, end = var_6626_end_0, end_mask = var_6626_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6626_cast_fp16")]; + tensor var_6633_begin_0 = const()[name = string("op_6633_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6633_end_0 = const()[name = string("op_6633_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6633_end_mask_0 = const()[name = string("op_6633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6633_cast_fp16 = slice_by_index(begin = var_6633_begin_0, end = var_6633_end_0, end_mask = var_6633_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6633_cast_fp16")]; + tensor var_6640_begin_0 = const()[name = string("op_6640_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6640_end_0 = const()[name = string("op_6640_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6640_end_mask_0 = const()[name = string("op_6640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6640_cast_fp16 = slice_by_index(begin = var_6640_begin_0, end = var_6640_end_0, end_mask = var_6640_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6640_cast_fp16")]; + tensor var_6647_begin_0 = const()[name = string("op_6647_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6647_end_0 = const()[name = string("op_6647_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6647_end_mask_0 = const()[name = string("op_6647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6647_cast_fp16 = slice_by_index(begin = var_6647_begin_0, end = var_6647_end_0, end_mask = var_6647_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6647_cast_fp16")]; + tensor var_6654_begin_0 = const()[name = string("op_6654_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6654_end_0 = const()[name = string("op_6654_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6654_end_mask_0 = const()[name = string("op_6654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6654_cast_fp16 = slice_by_index(begin = var_6654_begin_0, end = var_6654_end_0, end_mask = var_6654_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6654_cast_fp16")]; + tensor var_6661_begin_0 = const()[name = string("op_6661_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6661_end_0 = const()[name = string("op_6661_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6661_end_mask_0 = const()[name = string("op_6661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6661_cast_fp16 = slice_by_index(begin = var_6661_begin_0, end = var_6661_end_0, end_mask = var_6661_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6661_cast_fp16")]; + tensor var_6668_begin_0 = const()[name = string("op_6668_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6668_end_0 = const()[name = string("op_6668_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6668_end_mask_0 = const()[name = string("op_6668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6668_cast_fp16 = slice_by_index(begin = var_6668_begin_0, end = var_6668_end_0, end_mask = var_6668_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6668_cast_fp16")]; + tensor var_6675_begin_0 = const()[name = string("op_6675_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6675_end_0 = const()[name = string("op_6675_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6675_end_mask_0 = const()[name = string("op_6675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6675_cast_fp16 = slice_by_index(begin = var_6675_begin_0, end = var_6675_end_0, end_mask = var_6675_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6675_cast_fp16")]; + tensor var_6682_begin_0 = const()[name = string("op_6682_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6682_end_0 = const()[name = string("op_6682_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6682_end_mask_0 = const()[name = string("op_6682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6682_cast_fp16 = slice_by_index(begin = var_6682_begin_0, end = var_6682_end_0, end_mask = var_6682_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6682_cast_fp16")]; + tensor var_6689_begin_0 = const()[name = string("op_6689_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6689_end_0 = const()[name = string("op_6689_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6689_end_mask_0 = const()[name = string("op_6689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6689_cast_fp16 = slice_by_index(begin = var_6689_begin_0, end = var_6689_end_0, end_mask = var_6689_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6689_cast_fp16")]; + tensor var_6696_begin_0 = const()[name = string("op_6696_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6696_end_0 = const()[name = string("op_6696_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6696_end_mask_0 = const()[name = string("op_6696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6696_cast_fp16 = slice_by_index(begin = var_6696_begin_0, end = var_6696_end_0, end_mask = var_6696_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6696_cast_fp16")]; + tensor var_6703_begin_0 = const()[name = string("op_6703_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6703_end_0 = const()[name = string("op_6703_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6703_end_mask_0 = const()[name = string("op_6703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6703_cast_fp16 = slice_by_index(begin = var_6703_begin_0, end = var_6703_end_0, end_mask = var_6703_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6703_cast_fp16")]; + tensor var_6710_begin_0 = const()[name = string("op_6710_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6710_end_0 = const()[name = string("op_6710_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6710_end_mask_0 = const()[name = string("op_6710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6710_cast_fp16 = slice_by_index(begin = var_6710_begin_0, end = var_6710_end_0, end_mask = var_6710_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6710_cast_fp16")]; + tensor var_6717_begin_0 = const()[name = string("op_6717_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6717_end_0 = const()[name = string("op_6717_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6717_end_mask_0 = const()[name = string("op_6717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6717_cast_fp16 = slice_by_index(begin = var_6717_begin_0, end = var_6717_end_0, end_mask = var_6717_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6717_cast_fp16")]; + tensor var_6724_begin_0 = const()[name = string("op_6724_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6724_end_0 = const()[name = string("op_6724_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6724_end_mask_0 = const()[name = string("op_6724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6724_cast_fp16 = slice_by_index(begin = var_6724_begin_0, end = var_6724_end_0, end_mask = var_6724_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6724_cast_fp16")]; + tensor var_6731_begin_0 = const()[name = string("op_6731_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6731_end_0 = const()[name = string("op_6731_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6731_end_mask_0 = const()[name = string("op_6731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6731_cast_fp16 = slice_by_index(begin = var_6731_begin_0, end = var_6731_end_0, end_mask = var_6731_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6731_cast_fp16")]; + tensor var_6738_begin_0 = const()[name = string("op_6738_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6738_end_0 = const()[name = string("op_6738_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6738_end_mask_0 = const()[name = string("op_6738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6738_cast_fp16 = slice_by_index(begin = var_6738_begin_0, end = var_6738_end_0, end_mask = var_6738_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6738_cast_fp16")]; + tensor var_6745_begin_0 = const()[name = string("op_6745_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6745_end_0 = const()[name = string("op_6745_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6745_end_mask_0 = const()[name = string("op_6745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6745_cast_fp16 = slice_by_index(begin = var_6745_begin_0, end = var_6745_end_0, end_mask = var_6745_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6745_cast_fp16")]; + tensor var_6752_begin_0 = const()[name = string("op_6752_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6752_end_0 = const()[name = string("op_6752_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6752_end_mask_0 = const()[name = string("op_6752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6752_cast_fp16 = slice_by_index(begin = var_6752_begin_0, end = var_6752_end_0, end_mask = var_6752_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6752_cast_fp16")]; + tensor var_6759_begin_0 = const()[name = string("op_6759_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6759_end_0 = const()[name = string("op_6759_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6759_end_mask_0 = const()[name = string("op_6759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6759_cast_fp16 = slice_by_index(begin = var_6759_begin_0, end = var_6759_end_0, end_mask = var_6759_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6759_cast_fp16")]; + tensor var_6766_begin_0 = const()[name = string("op_6766_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6766_end_0 = const()[name = string("op_6766_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6766_end_mask_0 = const()[name = string("op_6766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6766_cast_fp16 = slice_by_index(begin = var_6766_begin_0, end = var_6766_end_0, end_mask = var_6766_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6766_cast_fp16")]; + tensor var_6773_begin_0 = const()[name = string("op_6773_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6773_end_0 = const()[name = string("op_6773_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6773_end_mask_0 = const()[name = string("op_6773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6773_cast_fp16 = slice_by_index(begin = var_6773_begin_0, end = var_6773_end_0, end_mask = var_6773_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6773_cast_fp16")]; + tensor var_6780_begin_0 = const()[name = string("op_6780_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6780_end_0 = const()[name = string("op_6780_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6780_end_mask_0 = const()[name = string("op_6780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6780_cast_fp16 = slice_by_index(begin = var_6780_begin_0, end = var_6780_end_0, end_mask = var_6780_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6780_cast_fp16")]; + tensor var_6787_begin_0 = const()[name = string("op_6787_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6787_end_0 = const()[name = string("op_6787_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6787_end_mask_0 = const()[name = string("op_6787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6787_cast_fp16 = slice_by_index(begin = var_6787_begin_0, end = var_6787_end_0, end_mask = var_6787_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6787_cast_fp16")]; + tensor var_6794_begin_0 = const()[name = string("op_6794_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6794_end_0 = const()[name = string("op_6794_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6794_end_mask_0 = const()[name = string("op_6794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6794_cast_fp16 = slice_by_index(begin = var_6794_begin_0, end = var_6794_end_0, end_mask = var_6794_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6794_cast_fp16")]; + tensor var_6801_begin_0 = const()[name = string("op_6801_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6801_end_0 = const()[name = string("op_6801_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6801_end_mask_0 = const()[name = string("op_6801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6801_cast_fp16 = slice_by_index(begin = var_6801_begin_0, end = var_6801_end_0, end_mask = var_6801_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6801_cast_fp16")]; + tensor var_6808_begin_0 = const()[name = string("op_6808_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6808_end_0 = const()[name = string("op_6808_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6808_end_mask_0 = const()[name = string("op_6808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6808_cast_fp16 = slice_by_index(begin = var_6808_begin_0, end = var_6808_end_0, end_mask = var_6808_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6808_cast_fp16")]; + tensor var_6815_begin_0 = const()[name = string("op_6815_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6815_end_0 = const()[name = string("op_6815_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6815_end_mask_0 = const()[name = string("op_6815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6815_cast_fp16 = slice_by_index(begin = var_6815_begin_0, end = var_6815_end_0, end_mask = var_6815_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6815_cast_fp16")]; + tensor var_6822_begin_0 = const()[name = string("op_6822_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6822_end_0 = const()[name = string("op_6822_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6822_end_mask_0 = const()[name = string("op_6822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6822_cast_fp16 = slice_by_index(begin = var_6822_begin_0, end = var_6822_end_0, end_mask = var_6822_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6822_cast_fp16")]; + tensor var_6829_begin_0 = const()[name = string("op_6829_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6829_end_0 = const()[name = string("op_6829_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6829_end_mask_0 = const()[name = string("op_6829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6829_cast_fp16 = slice_by_index(begin = var_6829_begin_0, end = var_6829_end_0, end_mask = var_6829_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6829_cast_fp16")]; + tensor var_6836_begin_0 = const()[name = string("op_6836_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6836_end_0 = const()[name = string("op_6836_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6836_end_mask_0 = const()[name = string("op_6836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6836_cast_fp16 = slice_by_index(begin = var_6836_begin_0, end = var_6836_end_0, end_mask = var_6836_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6836_cast_fp16")]; + tensor var_6843_begin_0 = const()[name = string("op_6843_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6843_end_0 = const()[name = string("op_6843_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6843_end_mask_0 = const()[name = string("op_6843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6843_cast_fp16 = slice_by_index(begin = var_6843_begin_0, end = var_6843_end_0, end_mask = var_6843_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6843_cast_fp16")]; + tensor var_6850_begin_0 = const()[name = string("op_6850_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6850_end_0 = const()[name = string("op_6850_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6850_end_mask_0 = const()[name = string("op_6850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6850_cast_fp16 = slice_by_index(begin = var_6850_begin_0, end = var_6850_end_0, end_mask = var_6850_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6850_cast_fp16")]; + tensor var_6857_begin_0 = const()[name = string("op_6857_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6857_end_0 = const()[name = string("op_6857_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6857_end_mask_0 = const()[name = string("op_6857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6857_cast_fp16 = slice_by_index(begin = var_6857_begin_0, end = var_6857_end_0, end_mask = var_6857_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6857_cast_fp16")]; + tensor var_6864_begin_0 = const()[name = string("op_6864_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6864_end_0 = const()[name = string("op_6864_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6864_end_mask_0 = const()[name = string("op_6864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6864_cast_fp16 = slice_by_index(begin = var_6864_begin_0, end = var_6864_end_0, end_mask = var_6864_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6864_cast_fp16")]; + tensor var_6871_begin_0 = const()[name = string("op_6871_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6871_end_0 = const()[name = string("op_6871_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6871_end_mask_0 = const()[name = string("op_6871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6871_cast_fp16 = slice_by_index(begin = var_6871_begin_0, end = var_6871_end_0, end_mask = var_6871_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6871_cast_fp16")]; + tensor var_6878_begin_0 = const()[name = string("op_6878_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6878_end_0 = const()[name = string("op_6878_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6878_end_mask_0 = const()[name = string("op_6878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6878_cast_fp16 = slice_by_index(begin = var_6878_begin_0, end = var_6878_end_0, end_mask = var_6878_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6878_cast_fp16")]; + tensor var_6885_begin_0 = const()[name = string("op_6885_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6885_end_0 = const()[name = string("op_6885_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6885_end_mask_0 = const()[name = string("op_6885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6885_cast_fp16 = slice_by_index(begin = var_6885_begin_0, end = var_6885_end_0, end_mask = var_6885_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6885_cast_fp16")]; + tensor var_6892_begin_0 = const()[name = string("op_6892_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6892_end_0 = const()[name = string("op_6892_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6892_end_mask_0 = const()[name = string("op_6892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6892_cast_fp16 = slice_by_index(begin = var_6892_begin_0, end = var_6892_end_0, end_mask = var_6892_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6892_cast_fp16")]; + tensor var_6899_begin_0 = const()[name = string("op_6899_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6899_end_0 = const()[name = string("op_6899_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6899_end_mask_0 = const()[name = string("op_6899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6899_cast_fp16 = slice_by_index(begin = var_6899_begin_0, end = var_6899_end_0, end_mask = var_6899_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6899_cast_fp16")]; + tensor var_6906_begin_0 = const()[name = string("op_6906_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6906_end_0 = const()[name = string("op_6906_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6906_end_mask_0 = const()[name = string("op_6906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6906_cast_fp16 = slice_by_index(begin = var_6906_begin_0, end = var_6906_end_0, end_mask = var_6906_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6906_cast_fp16")]; + tensor var_6913_begin_0 = const()[name = string("op_6913_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6913_end_0 = const()[name = string("op_6913_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6913_end_mask_0 = const()[name = string("op_6913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6913_cast_fp16 = slice_by_index(begin = var_6913_begin_0, end = var_6913_end_0, end_mask = var_6913_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6913_cast_fp16")]; + tensor var_6920_begin_0 = const()[name = string("op_6920_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6920_end_0 = const()[name = string("op_6920_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6920_end_mask_0 = const()[name = string("op_6920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6920_cast_fp16 = slice_by_index(begin = var_6920_begin_0, end = var_6920_end_0, end_mask = var_6920_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6920_cast_fp16")]; + tensor var_6927_begin_0 = const()[name = string("op_6927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6927_end_0 = const()[name = string("op_6927_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6927_end_mask_0 = const()[name = string("op_6927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6927_cast_fp16 = slice_by_index(begin = var_6927_begin_0, end = var_6927_end_0, end_mask = var_6927_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6927_cast_fp16")]; + tensor var_6934_begin_0 = const()[name = string("op_6934_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6934_end_0 = const()[name = string("op_6934_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6934_end_mask_0 = const()[name = string("op_6934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6934_cast_fp16 = slice_by_index(begin = var_6934_begin_0, end = var_6934_end_0, end_mask = var_6934_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6934_cast_fp16")]; + tensor var_6941_begin_0 = const()[name = string("op_6941_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6941_end_0 = const()[name = string("op_6941_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6941_end_mask_0 = const()[name = string("op_6941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6941_cast_fp16 = slice_by_index(begin = var_6941_begin_0, end = var_6941_end_0, end_mask = var_6941_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6941_cast_fp16")]; + tensor var_6948_begin_0 = const()[name = string("op_6948_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6948_end_0 = const()[name = string("op_6948_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6948_end_mask_0 = const()[name = string("op_6948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6948_cast_fp16 = slice_by_index(begin = var_6948_begin_0, end = var_6948_end_0, end_mask = var_6948_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6948_cast_fp16")]; + tensor var_6955_begin_0 = const()[name = string("op_6955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6955_end_0 = const()[name = string("op_6955_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6955_end_mask_0 = const()[name = string("op_6955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6955_cast_fp16")]; + tensor var_6962_begin_0 = const()[name = string("op_6962_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6962_end_0 = const()[name = string("op_6962_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6962_end_mask_0 = const()[name = string("op_6962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6962_cast_fp16 = slice_by_index(begin = var_6962_begin_0, end = var_6962_end_0, end_mask = var_6962_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6962_cast_fp16")]; + tensor var_6969_begin_0 = const()[name = string("op_6969_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6969_end_0 = const()[name = string("op_6969_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6969_end_mask_0 = const()[name = string("op_6969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6969_cast_fp16 = slice_by_index(begin = var_6969_begin_0, end = var_6969_end_0, end_mask = var_6969_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6969_cast_fp16")]; + tensor var_6976_begin_0 = const()[name = string("op_6976_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6976_end_0 = const()[name = string("op_6976_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6976_end_mask_0 = const()[name = string("op_6976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6976_cast_fp16 = slice_by_index(begin = var_6976_begin_0, end = var_6976_end_0, end_mask = var_6976_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6976_cast_fp16")]; + tensor var_6983_begin_0 = const()[name = string("op_6983_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6983_end_0 = const()[name = string("op_6983_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6983_end_mask_0 = const()[name = string("op_6983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6983_cast_fp16 = slice_by_index(begin = var_6983_begin_0, end = var_6983_end_0, end_mask = var_6983_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6983_cast_fp16")]; + tensor var_6990_begin_0 = const()[name = string("op_6990_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6990_end_0 = const()[name = string("op_6990_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6990_end_mask_0 = const()[name = string("op_6990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6990_cast_fp16 = slice_by_index(begin = var_6990_begin_0, end = var_6990_end_0, end_mask = var_6990_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6990_cast_fp16")]; + tensor var_6997_begin_0 = const()[name = string("op_6997_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6997_end_0 = const()[name = string("op_6997_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6997_end_mask_0 = const()[name = string("op_6997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6997_cast_fp16 = slice_by_index(begin = var_6997_begin_0, end = var_6997_end_0, end_mask = var_6997_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6997_cast_fp16")]; + tensor var_7004_begin_0 = const()[name = string("op_7004_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7004_end_0 = const()[name = string("op_7004_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7004_end_mask_0 = const()[name = string("op_7004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7004_cast_fp16 = slice_by_index(begin = var_7004_begin_0, end = var_7004_end_0, end_mask = var_7004_end_mask_0, x = var_6442_cast_fp16)[name = string("op_7004_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7009_begin_0 = const()[name = string("op_7009_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7009_end_0 = const()[name = string("op_7009_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_7009_end_mask_0 = const()[name = string("op_7009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_27")]; + tensor var_7009_cast_fp16 = slice_by_index(begin = var_7009_begin_0, end = var_7009_end_0, end_mask = var_7009_end_mask_0, x = k_9_cast_fp16)[name = string("op_7009_cast_fp16")]; + tensor var_7013_begin_0 = const()[name = string("op_7013_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7013_end_0 = const()[name = string("op_7013_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_7013_end_mask_0 = const()[name = string("op_7013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7013_cast_fp16 = slice_by_index(begin = var_7013_begin_0, end = var_7013_end_0, end_mask = var_7013_end_mask_0, x = k_9_cast_fp16)[name = string("op_7013_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = string("op_7017_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7017_end_0 = const()[name = string("op_7017_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_7017_end_mask_0 = const()[name = string("op_7017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = k_9_cast_fp16)[name = string("op_7017_cast_fp16")]; + tensor var_7021_begin_0 = const()[name = string("op_7021_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7021_end_0 = const()[name = string("op_7021_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_7021_end_mask_0 = const()[name = string("op_7021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7021_cast_fp16 = slice_by_index(begin = var_7021_begin_0, end = var_7021_end_0, end_mask = var_7021_end_mask_0, x = k_9_cast_fp16)[name = string("op_7021_cast_fp16")]; + tensor var_7025_begin_0 = const()[name = string("op_7025_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7025_end_0 = const()[name = string("op_7025_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_7025_end_mask_0 = const()[name = string("op_7025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7025_cast_fp16 = slice_by_index(begin = var_7025_begin_0, end = var_7025_end_0, end_mask = var_7025_end_mask_0, x = k_9_cast_fp16)[name = string("op_7025_cast_fp16")]; + tensor var_7029_begin_0 = const()[name = string("op_7029_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7029_end_0 = const()[name = string("op_7029_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_7029_end_mask_0 = const()[name = string("op_7029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7029_cast_fp16 = slice_by_index(begin = var_7029_begin_0, end = var_7029_end_0, end_mask = var_7029_end_mask_0, x = k_9_cast_fp16)[name = string("op_7029_cast_fp16")]; + tensor var_7033_begin_0 = const()[name = string("op_7033_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7033_end_0 = const()[name = string("op_7033_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_7033_end_mask_0 = const()[name = string("op_7033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7033_cast_fp16 = slice_by_index(begin = var_7033_begin_0, end = var_7033_end_0, end_mask = var_7033_end_mask_0, x = k_9_cast_fp16)[name = string("op_7033_cast_fp16")]; + tensor var_7037_begin_0 = const()[name = string("op_7037_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7037_end_0 = const()[name = string("op_7037_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_7037_end_mask_0 = const()[name = string("op_7037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7037_cast_fp16 = slice_by_index(begin = var_7037_begin_0, end = var_7037_end_0, end_mask = var_7037_end_mask_0, x = k_9_cast_fp16)[name = string("op_7037_cast_fp16")]; + tensor var_7041_begin_0 = const()[name = string("op_7041_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7041_end_0 = const()[name = string("op_7041_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_7041_end_mask_0 = const()[name = string("op_7041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7041_cast_fp16 = slice_by_index(begin = var_7041_begin_0, end = var_7041_end_0, end_mask = var_7041_end_mask_0, x = k_9_cast_fp16)[name = string("op_7041_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = string("op_7045_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7045_end_0 = const()[name = string("op_7045_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_7045_end_mask_0 = const()[name = string("op_7045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = k_9_cast_fp16)[name = string("op_7045_cast_fp16")]; + tensor var_7049_begin_0 = const()[name = string("op_7049_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7049_end_0 = const()[name = string("op_7049_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_7049_end_mask_0 = const()[name = string("op_7049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7049_cast_fp16 = slice_by_index(begin = var_7049_begin_0, end = var_7049_end_0, end_mask = var_7049_end_mask_0, x = k_9_cast_fp16)[name = string("op_7049_cast_fp16")]; + tensor var_7053_begin_0 = const()[name = string("op_7053_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7053_end_0 = const()[name = string("op_7053_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_7053_end_mask_0 = const()[name = string("op_7053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7053_cast_fp16 = slice_by_index(begin = var_7053_begin_0, end = var_7053_end_0, end_mask = var_7053_end_mask_0, x = k_9_cast_fp16)[name = string("op_7053_cast_fp16")]; + tensor var_7057_begin_0 = const()[name = string("op_7057_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7057_end_0 = const()[name = string("op_7057_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_7057_end_mask_0 = const()[name = string("op_7057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7057_cast_fp16 = slice_by_index(begin = var_7057_begin_0, end = var_7057_end_0, end_mask = var_7057_end_mask_0, x = k_9_cast_fp16)[name = string("op_7057_cast_fp16")]; + tensor var_7061_begin_0 = const()[name = string("op_7061_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7061_end_0 = const()[name = string("op_7061_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_7061_end_mask_0 = const()[name = string("op_7061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7061_cast_fp16 = slice_by_index(begin = var_7061_begin_0, end = var_7061_end_0, end_mask = var_7061_end_mask_0, x = k_9_cast_fp16)[name = string("op_7061_cast_fp16")]; + tensor var_7065_begin_0 = const()[name = string("op_7065_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7065_end_0 = const()[name = string("op_7065_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_7065_end_mask_0 = const()[name = string("op_7065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7065_cast_fp16 = slice_by_index(begin = var_7065_begin_0, end = var_7065_end_0, end_mask = var_7065_end_mask_0, x = k_9_cast_fp16)[name = string("op_7065_cast_fp16")]; + tensor var_7069_begin_0 = const()[name = string("op_7069_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7069_end_0 = const()[name = string("op_7069_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_7069_end_mask_0 = const()[name = string("op_7069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7069_cast_fp16 = slice_by_index(begin = var_7069_begin_0, end = var_7069_end_0, end_mask = var_7069_end_mask_0, x = k_9_cast_fp16)[name = string("op_7069_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = string("op_7073_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7073_end_0 = const()[name = string("op_7073_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_7073_end_mask_0 = const()[name = string("op_7073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = k_9_cast_fp16)[name = string("op_7073_cast_fp16")]; + tensor var_7077_begin_0 = const()[name = string("op_7077_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7077_end_0 = const()[name = string("op_7077_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_7077_end_mask_0 = const()[name = string("op_7077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7077_cast_fp16 = slice_by_index(begin = var_7077_begin_0, end = var_7077_end_0, end_mask = var_7077_end_mask_0, x = k_9_cast_fp16)[name = string("op_7077_cast_fp16")]; + tensor var_7081_begin_0 = const()[name = string("op_7081_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7081_end_0 = const()[name = string("op_7081_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_7081_end_mask_0 = const()[name = string("op_7081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7081_cast_fp16 = slice_by_index(begin = var_7081_begin_0, end = var_7081_end_0, end_mask = var_7081_end_mask_0, x = k_9_cast_fp16)[name = string("op_7081_cast_fp16")]; + tensor var_7085_begin_0 = const()[name = string("op_7085_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7085_end_0 = const()[name = string("op_7085_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_7085_end_mask_0 = const()[name = string("op_7085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7085_cast_fp16 = slice_by_index(begin = var_7085_begin_0, end = var_7085_end_0, end_mask = var_7085_end_mask_0, x = k_9_cast_fp16)[name = string("op_7085_cast_fp16")]; + tensor var_7087_begin_0 = const()[name = string("op_7087_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7087_end_0 = const()[name = string("op_7087_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7087_end_mask_0 = const()[name = string("op_7087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7087_cast_fp16 = slice_by_index(begin = var_7087_begin_0, end = var_7087_end_0, end_mask = var_7087_end_mask_0, x = value_9_cast_fp16)[name = string("op_7087_cast_fp16")]; + tensor var_7091_begin_0 = const()[name = string("op_7091_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7091_end_0 = const()[name = string("op_7091_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7091_end_mask_0 = const()[name = string("op_7091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7091_cast_fp16 = slice_by_index(begin = var_7091_begin_0, end = var_7091_end_0, end_mask = var_7091_end_mask_0, x = value_9_cast_fp16)[name = string("op_7091_cast_fp16")]; + tensor var_7095_begin_0 = const()[name = string("op_7095_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7095_end_0 = const()[name = string("op_7095_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7095_end_mask_0 = const()[name = string("op_7095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7095_cast_fp16 = slice_by_index(begin = var_7095_begin_0, end = var_7095_end_0, end_mask = var_7095_end_mask_0, x = value_9_cast_fp16)[name = string("op_7095_cast_fp16")]; + tensor var_7099_begin_0 = const()[name = string("op_7099_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7099_end_0 = const()[name = string("op_7099_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7099_end_mask_0 = const()[name = string("op_7099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7099_cast_fp16 = slice_by_index(begin = var_7099_begin_0, end = var_7099_end_0, end_mask = var_7099_end_mask_0, x = value_9_cast_fp16)[name = string("op_7099_cast_fp16")]; + tensor var_7103_begin_0 = const()[name = string("op_7103_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7103_end_0 = const()[name = string("op_7103_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7103_end_mask_0 = const()[name = string("op_7103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7103_cast_fp16 = slice_by_index(begin = var_7103_begin_0, end = var_7103_end_0, end_mask = var_7103_end_mask_0, x = value_9_cast_fp16)[name = string("op_7103_cast_fp16")]; + tensor var_7107_begin_0 = const()[name = string("op_7107_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7107_end_0 = const()[name = string("op_7107_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7107_end_mask_0 = const()[name = string("op_7107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7107_cast_fp16 = slice_by_index(begin = var_7107_begin_0, end = var_7107_end_0, end_mask = var_7107_end_mask_0, x = value_9_cast_fp16)[name = string("op_7107_cast_fp16")]; + tensor var_7111_begin_0 = const()[name = string("op_7111_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7111_end_0 = const()[name = string("op_7111_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7111_end_mask_0 = const()[name = string("op_7111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7111_cast_fp16 = slice_by_index(begin = var_7111_begin_0, end = var_7111_end_0, end_mask = var_7111_end_mask_0, x = value_9_cast_fp16)[name = string("op_7111_cast_fp16")]; + tensor var_7115_begin_0 = const()[name = string("op_7115_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7115_end_0 = const()[name = string("op_7115_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7115_end_mask_0 = const()[name = string("op_7115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7115_cast_fp16 = slice_by_index(begin = var_7115_begin_0, end = var_7115_end_0, end_mask = var_7115_end_mask_0, x = value_9_cast_fp16)[name = string("op_7115_cast_fp16")]; + tensor var_7119_begin_0 = const()[name = string("op_7119_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7119_end_0 = const()[name = string("op_7119_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7119_end_mask_0 = const()[name = string("op_7119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7119_cast_fp16 = slice_by_index(begin = var_7119_begin_0, end = var_7119_end_0, end_mask = var_7119_end_mask_0, x = value_9_cast_fp16)[name = string("op_7119_cast_fp16")]; + tensor var_7123_begin_0 = const()[name = string("op_7123_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7123_end_0 = const()[name = string("op_7123_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7123_end_mask_0 = const()[name = string("op_7123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7123_cast_fp16 = slice_by_index(begin = var_7123_begin_0, end = var_7123_end_0, end_mask = var_7123_end_mask_0, x = value_9_cast_fp16)[name = string("op_7123_cast_fp16")]; + tensor var_7127_begin_0 = const()[name = string("op_7127_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7127_end_0 = const()[name = string("op_7127_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7127_end_mask_0 = const()[name = string("op_7127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7127_cast_fp16 = slice_by_index(begin = var_7127_begin_0, end = var_7127_end_0, end_mask = var_7127_end_mask_0, x = value_9_cast_fp16)[name = string("op_7127_cast_fp16")]; + tensor var_7131_begin_0 = const()[name = string("op_7131_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7131_end_0 = const()[name = string("op_7131_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7131_end_mask_0 = const()[name = string("op_7131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7131_cast_fp16 = slice_by_index(begin = var_7131_begin_0, end = var_7131_end_0, end_mask = var_7131_end_mask_0, x = value_9_cast_fp16)[name = string("op_7131_cast_fp16")]; + tensor var_7135_begin_0 = const()[name = string("op_7135_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7135_end_0 = const()[name = string("op_7135_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_7135_end_mask_0 = const()[name = string("op_7135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7135_cast_fp16 = slice_by_index(begin = var_7135_begin_0, end = var_7135_end_0, end_mask = var_7135_end_mask_0, x = value_9_cast_fp16)[name = string("op_7135_cast_fp16")]; + tensor var_7139_begin_0 = const()[name = string("op_7139_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7139_end_0 = const()[name = string("op_7139_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_7139_end_mask_0 = const()[name = string("op_7139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7139_cast_fp16 = slice_by_index(begin = var_7139_begin_0, end = var_7139_end_0, end_mask = var_7139_end_mask_0, x = value_9_cast_fp16)[name = string("op_7139_cast_fp16")]; + tensor var_7143_begin_0 = const()[name = string("op_7143_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7143_end_0 = const()[name = string("op_7143_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_7143_end_mask_0 = const()[name = string("op_7143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7143_cast_fp16 = slice_by_index(begin = var_7143_begin_0, end = var_7143_end_0, end_mask = var_7143_end_mask_0, x = value_9_cast_fp16)[name = string("op_7143_cast_fp16")]; + tensor var_7147_begin_0 = const()[name = string("op_7147_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7147_end_0 = const()[name = string("op_7147_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_7147_end_mask_0 = const()[name = string("op_7147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7147_cast_fp16 = slice_by_index(begin = var_7147_begin_0, end = var_7147_end_0, end_mask = var_7147_end_mask_0, x = value_9_cast_fp16)[name = string("op_7147_cast_fp16")]; + tensor var_7151_begin_0 = const()[name = string("op_7151_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7151_end_0 = const()[name = string("op_7151_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_7151_end_mask_0 = const()[name = string("op_7151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7151_cast_fp16 = slice_by_index(begin = var_7151_begin_0, end = var_7151_end_0, end_mask = var_7151_end_mask_0, x = value_9_cast_fp16)[name = string("op_7151_cast_fp16")]; + tensor var_7155_begin_0 = const()[name = string("op_7155_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7155_end_0 = const()[name = string("op_7155_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_7155_end_mask_0 = const()[name = string("op_7155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7155_cast_fp16 = slice_by_index(begin = var_7155_begin_0, end = var_7155_end_0, end_mask = var_7155_end_mask_0, x = value_9_cast_fp16)[name = string("op_7155_cast_fp16")]; + tensor var_7159_begin_0 = const()[name = string("op_7159_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7159_end_0 = const()[name = string("op_7159_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_7159_end_mask_0 = const()[name = string("op_7159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7159_cast_fp16 = slice_by_index(begin = var_7159_begin_0, end = var_7159_end_0, end_mask = var_7159_end_mask_0, x = value_9_cast_fp16)[name = string("op_7159_cast_fp16")]; + tensor var_7163_begin_0 = const()[name = string("op_7163_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7163_end_0 = const()[name = string("op_7163_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_7163_end_mask_0 = const()[name = string("op_7163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7163_cast_fp16 = slice_by_index(begin = var_7163_begin_0, end = var_7163_end_0, end_mask = var_7163_end_mask_0, x = value_9_cast_fp16)[name = string("op_7163_cast_fp16")]; + string _SplitHeadsQ__mh_w_641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_641_equation_0, values = (var_7009_cast_fp16, var_6451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_641_cast_fp16")]; + string _SplitHeadsQ__mh_w_643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_643_equation_0, values = (var_7009_cast_fp16, var_6458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_643_cast_fp16")]; + string _SplitHeadsQ__mh_w_645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_645_equation_0, values = (var_7009_cast_fp16, var_6465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_645_cast_fp16")]; + string _SplitHeadsQ__mh_w_647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_647_equation_0, values = (var_7009_cast_fp16, var_6472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_647_cast_fp16")]; + string _SplitHeadsQ__mh_w_649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_649_equation_0, values = (var_7013_cast_fp16, var_6479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_649_cast_fp16")]; + string _SplitHeadsQ__mh_w_651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_651_equation_0, values = (var_7013_cast_fp16, var_6486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_651_cast_fp16")]; + string _SplitHeadsQ__mh_w_653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_653_equation_0, values = (var_7013_cast_fp16, var_6493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_653_cast_fp16")]; + string _SplitHeadsQ__mh_w_655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_655_equation_0, values = (var_7013_cast_fp16, var_6500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_655_cast_fp16")]; + string _SplitHeadsQ__mh_w_657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_657_equation_0, values = (var_7017_cast_fp16, var_6507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_657_cast_fp16")]; + string _SplitHeadsQ__mh_w_659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_659_equation_0, values = (var_7017_cast_fp16, var_6514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_659_cast_fp16")]; + string _SplitHeadsQ__mh_w_661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_661_equation_0, values = (var_7017_cast_fp16, var_6521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_661_cast_fp16")]; + string _SplitHeadsQ__mh_w_663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_663_equation_0, values = (var_7017_cast_fp16, var_6528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_663_cast_fp16")]; + string _SplitHeadsQ__mh_w_665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_665_equation_0, values = (var_7021_cast_fp16, var_6535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_665_cast_fp16")]; + string _SplitHeadsQ__mh_w_667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_667_equation_0, values = (var_7021_cast_fp16, var_6542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_667_cast_fp16")]; + string _SplitHeadsQ__mh_w_669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_669_equation_0, values = (var_7021_cast_fp16, var_6549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_669_cast_fp16")]; + string _SplitHeadsQ__mh_w_671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_671_equation_0, values = (var_7021_cast_fp16, var_6556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_671_cast_fp16")]; + string _SplitHeadsQ__mh_w_673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_673_equation_0, values = (var_7025_cast_fp16, var_6563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_673_cast_fp16")]; + string _SplitHeadsQ__mh_w_675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_675_equation_0, values = (var_7025_cast_fp16, var_6570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_675_cast_fp16")]; + string _SplitHeadsQ__mh_w_677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_677_equation_0, values = (var_7025_cast_fp16, var_6577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_677_cast_fp16")]; + string _SplitHeadsQ__mh_w_679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_679_equation_0, values = (var_7025_cast_fp16, var_6584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_679_cast_fp16")]; + string _SplitHeadsQ__mh_w_681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_681_equation_0, values = (var_7029_cast_fp16, var_6591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_681_cast_fp16")]; + string _SplitHeadsQ__mh_w_683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_683_equation_0, values = (var_7029_cast_fp16, var_6598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_683_cast_fp16")]; + string _SplitHeadsQ__mh_w_685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_685_equation_0, values = (var_7029_cast_fp16, var_6605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_685_cast_fp16")]; + string _SplitHeadsQ__mh_w_687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_687_equation_0, values = (var_7029_cast_fp16, var_6612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_687_cast_fp16")]; + string _SplitHeadsQ__mh_w_689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_689_equation_0, values = (var_7033_cast_fp16, var_6619_cast_fp16))[name = string("_SplitHeadsQ__mh_w_689_cast_fp16")]; + string _SplitHeadsQ__mh_w_691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_691_equation_0, values = (var_7033_cast_fp16, var_6626_cast_fp16))[name = string("_SplitHeadsQ__mh_w_691_cast_fp16")]; + string _SplitHeadsQ__mh_w_693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_693_equation_0, values = (var_7033_cast_fp16, var_6633_cast_fp16))[name = string("_SplitHeadsQ__mh_w_693_cast_fp16")]; + string _SplitHeadsQ__mh_w_695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_695_equation_0, values = (var_7033_cast_fp16, var_6640_cast_fp16))[name = string("_SplitHeadsQ__mh_w_695_cast_fp16")]; + string _SplitHeadsQ__mh_w_697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_697_equation_0, values = (var_7037_cast_fp16, var_6647_cast_fp16))[name = string("_SplitHeadsQ__mh_w_697_cast_fp16")]; + string _SplitHeadsQ__mh_w_699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_699_equation_0, values = (var_7037_cast_fp16, var_6654_cast_fp16))[name = string("_SplitHeadsQ__mh_w_699_cast_fp16")]; + string _SplitHeadsQ__mh_w_701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_701_equation_0, values = (var_7037_cast_fp16, var_6661_cast_fp16))[name = string("_SplitHeadsQ__mh_w_701_cast_fp16")]; + string _SplitHeadsQ__mh_w_703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_703_equation_0, values = (var_7037_cast_fp16, var_6668_cast_fp16))[name = string("_SplitHeadsQ__mh_w_703_cast_fp16")]; + string _SplitHeadsQ__mh_w_705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_705_equation_0, values = (var_7041_cast_fp16, var_6675_cast_fp16))[name = string("_SplitHeadsQ__mh_w_705_cast_fp16")]; + string _SplitHeadsQ__mh_w_707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_707_equation_0, values = (var_7041_cast_fp16, var_6682_cast_fp16))[name = string("_SplitHeadsQ__mh_w_707_cast_fp16")]; + string _SplitHeadsQ__mh_w_709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_709_equation_0, values = (var_7041_cast_fp16, var_6689_cast_fp16))[name = string("_SplitHeadsQ__mh_w_709_cast_fp16")]; + string _SplitHeadsQ__mh_w_711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_711_equation_0, values = (var_7041_cast_fp16, var_6696_cast_fp16))[name = string("_SplitHeadsQ__mh_w_711_cast_fp16")]; + string _SplitHeadsQ__mh_w_713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_713_equation_0, values = (var_7045_cast_fp16, var_6703_cast_fp16))[name = string("_SplitHeadsQ__mh_w_713_cast_fp16")]; + string _SplitHeadsQ__mh_w_715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_715_equation_0, values = (var_7045_cast_fp16, var_6710_cast_fp16))[name = string("_SplitHeadsQ__mh_w_715_cast_fp16")]; + string _SplitHeadsQ__mh_w_717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_717_equation_0, values = (var_7045_cast_fp16, var_6717_cast_fp16))[name = string("_SplitHeadsQ__mh_w_717_cast_fp16")]; + string _SplitHeadsQ__mh_w_719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_719_equation_0, values = (var_7045_cast_fp16, var_6724_cast_fp16))[name = string("_SplitHeadsQ__mh_w_719_cast_fp16")]; + string _SplitHeadsQ__mh_w_721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_721_equation_0, values = (var_7049_cast_fp16, var_6731_cast_fp16))[name = string("_SplitHeadsQ__mh_w_721_cast_fp16")]; + string _SplitHeadsQ__mh_w_723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_723_equation_0, values = (var_7049_cast_fp16, var_6738_cast_fp16))[name = string("_SplitHeadsQ__mh_w_723_cast_fp16")]; + string _SplitHeadsQ__mh_w_725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_725_equation_0, values = (var_7049_cast_fp16, var_6745_cast_fp16))[name = string("_SplitHeadsQ__mh_w_725_cast_fp16")]; + string _SplitHeadsQ__mh_w_727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_727_equation_0, values = (var_7049_cast_fp16, var_6752_cast_fp16))[name = string("_SplitHeadsQ__mh_w_727_cast_fp16")]; + string _SplitHeadsQ__mh_w_729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_729_equation_0, values = (var_7053_cast_fp16, var_6759_cast_fp16))[name = string("_SplitHeadsQ__mh_w_729_cast_fp16")]; + string _SplitHeadsQ__mh_w_731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_731_equation_0, values = (var_7053_cast_fp16, var_6766_cast_fp16))[name = string("_SplitHeadsQ__mh_w_731_cast_fp16")]; + string _SplitHeadsQ__mh_w_733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_733_equation_0, values = (var_7053_cast_fp16, var_6773_cast_fp16))[name = string("_SplitHeadsQ__mh_w_733_cast_fp16")]; + string _SplitHeadsQ__mh_w_735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_735_equation_0, values = (var_7053_cast_fp16, var_6780_cast_fp16))[name = string("_SplitHeadsQ__mh_w_735_cast_fp16")]; + string _SplitHeadsQ__mh_w_737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_737_equation_0, values = (var_7057_cast_fp16, var_6787_cast_fp16))[name = string("_SplitHeadsQ__mh_w_737_cast_fp16")]; + string _SplitHeadsQ__mh_w_739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_739_equation_0, values = (var_7057_cast_fp16, var_6794_cast_fp16))[name = string("_SplitHeadsQ__mh_w_739_cast_fp16")]; + string _SplitHeadsQ__mh_w_741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_741_equation_0, values = (var_7057_cast_fp16, var_6801_cast_fp16))[name = string("_SplitHeadsQ__mh_w_741_cast_fp16")]; + string _SplitHeadsQ__mh_w_743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_743_equation_0, values = (var_7057_cast_fp16, var_6808_cast_fp16))[name = string("_SplitHeadsQ__mh_w_743_cast_fp16")]; + string _SplitHeadsQ__mh_w_745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_745_equation_0, values = (var_7061_cast_fp16, var_6815_cast_fp16))[name = string("_SplitHeadsQ__mh_w_745_cast_fp16")]; + string _SplitHeadsQ__mh_w_747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_747_equation_0, values = (var_7061_cast_fp16, var_6822_cast_fp16))[name = string("_SplitHeadsQ__mh_w_747_cast_fp16")]; + string _SplitHeadsQ__mh_w_749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_749_equation_0, values = (var_7061_cast_fp16, var_6829_cast_fp16))[name = string("_SplitHeadsQ__mh_w_749_cast_fp16")]; + string _SplitHeadsQ__mh_w_751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_751_equation_0, values = (var_7061_cast_fp16, var_6836_cast_fp16))[name = string("_SplitHeadsQ__mh_w_751_cast_fp16")]; + string _SplitHeadsQ__mh_w_753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_753_equation_0, values = (var_7065_cast_fp16, var_6843_cast_fp16))[name = string("_SplitHeadsQ__mh_w_753_cast_fp16")]; + string _SplitHeadsQ__mh_w_755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_755_equation_0, values = (var_7065_cast_fp16, var_6850_cast_fp16))[name = string("_SplitHeadsQ__mh_w_755_cast_fp16")]; + string _SplitHeadsQ__mh_w_757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_757_equation_0, values = (var_7065_cast_fp16, var_6857_cast_fp16))[name = string("_SplitHeadsQ__mh_w_757_cast_fp16")]; + string _SplitHeadsQ__mh_w_759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_759_equation_0, values = (var_7065_cast_fp16, var_6864_cast_fp16))[name = string("_SplitHeadsQ__mh_w_759_cast_fp16")]; + string _SplitHeadsQ__mh_w_761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_761_equation_0, values = (var_7069_cast_fp16, var_6871_cast_fp16))[name = string("_SplitHeadsQ__mh_w_761_cast_fp16")]; + string _SplitHeadsQ__mh_w_763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_763_equation_0, values = (var_7069_cast_fp16, var_6878_cast_fp16))[name = string("_SplitHeadsQ__mh_w_763_cast_fp16")]; + string _SplitHeadsQ__mh_w_765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_765_equation_0, values = (var_7069_cast_fp16, var_6885_cast_fp16))[name = string("_SplitHeadsQ__mh_w_765_cast_fp16")]; + string _SplitHeadsQ__mh_w_767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_767_equation_0, values = (var_7069_cast_fp16, var_6892_cast_fp16))[name = string("_SplitHeadsQ__mh_w_767_cast_fp16")]; + string _SplitHeadsQ__mh_w_769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_769_equation_0, values = (var_7073_cast_fp16, var_6899_cast_fp16))[name = string("_SplitHeadsQ__mh_w_769_cast_fp16")]; + string _SplitHeadsQ__mh_w_771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_771_equation_0, values = (var_7073_cast_fp16, var_6906_cast_fp16))[name = string("_SplitHeadsQ__mh_w_771_cast_fp16")]; + string _SplitHeadsQ__mh_w_773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_773_equation_0, values = (var_7073_cast_fp16, var_6913_cast_fp16))[name = string("_SplitHeadsQ__mh_w_773_cast_fp16")]; + string _SplitHeadsQ__mh_w_775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_775_equation_0, values = (var_7073_cast_fp16, var_6920_cast_fp16))[name = string("_SplitHeadsQ__mh_w_775_cast_fp16")]; + string _SplitHeadsQ__mh_w_777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_777_equation_0, values = (var_7077_cast_fp16, var_6927_cast_fp16))[name = string("_SplitHeadsQ__mh_w_777_cast_fp16")]; + string _SplitHeadsQ__mh_w_779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_779_equation_0, values = (var_7077_cast_fp16, var_6934_cast_fp16))[name = string("_SplitHeadsQ__mh_w_779_cast_fp16")]; + string _SplitHeadsQ__mh_w_781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_781_equation_0, values = (var_7077_cast_fp16, var_6941_cast_fp16))[name = string("_SplitHeadsQ__mh_w_781_cast_fp16")]; + string _SplitHeadsQ__mh_w_783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_783_equation_0, values = (var_7077_cast_fp16, var_6948_cast_fp16))[name = string("_SplitHeadsQ__mh_w_783_cast_fp16")]; + string _SplitHeadsQ__mh_w_785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_785_equation_0, values = (var_7081_cast_fp16, var_6955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_785_cast_fp16")]; + string _SplitHeadsQ__mh_w_787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_787_equation_0, values = (var_7081_cast_fp16, var_6962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_787_cast_fp16")]; + string _SplitHeadsQ__mh_w_789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_789_equation_0, values = (var_7081_cast_fp16, var_6969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_789_cast_fp16")]; + string _SplitHeadsQ__mh_w_791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_791_equation_0, values = (var_7081_cast_fp16, var_6976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_791_cast_fp16")]; + string _SplitHeadsQ__mh_w_793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_793_equation_0, values = (var_7085_cast_fp16, var_6983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_793_cast_fp16")]; + string _SplitHeadsQ__mh_w_795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_795_equation_0, values = (var_7085_cast_fp16, var_6990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_795_cast_fp16")]; + string _SplitHeadsQ__mh_w_797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_797_equation_0, values = (var_7085_cast_fp16, var_6997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_797_cast_fp16")]; + string _SplitHeadsQ__mh_w_799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_799_equation_0, values = (var_7085_cast_fp16, var_7004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_799_cast_fp16")]; + fp16 var_7326_to_fp16 = const()[name = string("op_7326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_641_cast_fp16, y = var_7326_to_fp16)[name = string("aw_chunk_641_cast_fp16")]; + fp16 var_7328_to_fp16 = const()[name = string("op_7328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_643_cast_fp16, y = var_7328_to_fp16)[name = string("aw_chunk_643_cast_fp16")]; + fp16 var_7330_to_fp16 = const()[name = string("op_7330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_645_cast_fp16, y = var_7330_to_fp16)[name = string("aw_chunk_645_cast_fp16")]; + fp16 var_7332_to_fp16 = const()[name = string("op_7332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_647_cast_fp16, y = var_7332_to_fp16)[name = string("aw_chunk_647_cast_fp16")]; + fp16 var_7334_to_fp16 = const()[name = string("op_7334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_649_cast_fp16, y = var_7334_to_fp16)[name = string("aw_chunk_649_cast_fp16")]; + fp16 var_7336_to_fp16 = const()[name = string("op_7336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_651_cast_fp16, y = var_7336_to_fp16)[name = string("aw_chunk_651_cast_fp16")]; + fp16 var_7338_to_fp16 = const()[name = string("op_7338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_653_cast_fp16, y = var_7338_to_fp16)[name = string("aw_chunk_653_cast_fp16")]; + fp16 var_7340_to_fp16 = const()[name = string("op_7340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_655_cast_fp16, y = var_7340_to_fp16)[name = string("aw_chunk_655_cast_fp16")]; + fp16 var_7342_to_fp16 = const()[name = string("op_7342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_657_cast_fp16, y = var_7342_to_fp16)[name = string("aw_chunk_657_cast_fp16")]; + fp16 var_7344_to_fp16 = const()[name = string("op_7344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_659_cast_fp16, y = var_7344_to_fp16)[name = string("aw_chunk_659_cast_fp16")]; + fp16 var_7346_to_fp16 = const()[name = string("op_7346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_661_cast_fp16, y = var_7346_to_fp16)[name = string("aw_chunk_661_cast_fp16")]; + fp16 var_7348_to_fp16 = const()[name = string("op_7348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_663_cast_fp16, y = var_7348_to_fp16)[name = string("aw_chunk_663_cast_fp16")]; + fp16 var_7350_to_fp16 = const()[name = string("op_7350_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_665_cast_fp16, y = var_7350_to_fp16)[name = string("aw_chunk_665_cast_fp16")]; + fp16 var_7352_to_fp16 = const()[name = string("op_7352_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_667_cast_fp16, y = var_7352_to_fp16)[name = string("aw_chunk_667_cast_fp16")]; + fp16 var_7354_to_fp16 = const()[name = string("op_7354_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_669_cast_fp16, y = var_7354_to_fp16)[name = string("aw_chunk_669_cast_fp16")]; + fp16 var_7356_to_fp16 = const()[name = string("op_7356_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_671_cast_fp16, y = var_7356_to_fp16)[name = string("aw_chunk_671_cast_fp16")]; + fp16 var_7358_to_fp16 = const()[name = string("op_7358_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_673_cast_fp16, y = var_7358_to_fp16)[name = string("aw_chunk_673_cast_fp16")]; + fp16 var_7360_to_fp16 = const()[name = string("op_7360_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_675_cast_fp16, y = var_7360_to_fp16)[name = string("aw_chunk_675_cast_fp16")]; + fp16 var_7362_to_fp16 = const()[name = string("op_7362_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_677_cast_fp16, y = var_7362_to_fp16)[name = string("aw_chunk_677_cast_fp16")]; + fp16 var_7364_to_fp16 = const()[name = string("op_7364_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_679_cast_fp16, y = var_7364_to_fp16)[name = string("aw_chunk_679_cast_fp16")]; + fp16 var_7366_to_fp16 = const()[name = string("op_7366_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_681_cast_fp16, y = var_7366_to_fp16)[name = string("aw_chunk_681_cast_fp16")]; + fp16 var_7368_to_fp16 = const()[name = string("op_7368_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_683_cast_fp16, y = var_7368_to_fp16)[name = string("aw_chunk_683_cast_fp16")]; + fp16 var_7370_to_fp16 = const()[name = string("op_7370_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_685_cast_fp16, y = var_7370_to_fp16)[name = string("aw_chunk_685_cast_fp16")]; + fp16 var_7372_to_fp16 = const()[name = string("op_7372_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_687_cast_fp16, y = var_7372_to_fp16)[name = string("aw_chunk_687_cast_fp16")]; + fp16 var_7374_to_fp16 = const()[name = string("op_7374_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_689_cast_fp16, y = var_7374_to_fp16)[name = string("aw_chunk_689_cast_fp16")]; + fp16 var_7376_to_fp16 = const()[name = string("op_7376_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_691_cast_fp16, y = var_7376_to_fp16)[name = string("aw_chunk_691_cast_fp16")]; + fp16 var_7378_to_fp16 = const()[name = string("op_7378_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_693_cast_fp16, y = var_7378_to_fp16)[name = string("aw_chunk_693_cast_fp16")]; + fp16 var_7380_to_fp16 = const()[name = string("op_7380_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_695_cast_fp16, y = var_7380_to_fp16)[name = string("aw_chunk_695_cast_fp16")]; + fp16 var_7382_to_fp16 = const()[name = string("op_7382_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_697_cast_fp16, y = var_7382_to_fp16)[name = string("aw_chunk_697_cast_fp16")]; + fp16 var_7384_to_fp16 = const()[name = string("op_7384_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_699_cast_fp16, y = var_7384_to_fp16)[name = string("aw_chunk_699_cast_fp16")]; + fp16 var_7386_to_fp16 = const()[name = string("op_7386_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_701_cast_fp16, y = var_7386_to_fp16)[name = string("aw_chunk_701_cast_fp16")]; + fp16 var_7388_to_fp16 = const()[name = string("op_7388_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_703_cast_fp16, y = var_7388_to_fp16)[name = string("aw_chunk_703_cast_fp16")]; + fp16 var_7390_to_fp16 = const()[name = string("op_7390_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_705_cast_fp16, y = var_7390_to_fp16)[name = string("aw_chunk_705_cast_fp16")]; + fp16 var_7392_to_fp16 = const()[name = string("op_7392_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_707_cast_fp16, y = var_7392_to_fp16)[name = string("aw_chunk_707_cast_fp16")]; + fp16 var_7394_to_fp16 = const()[name = string("op_7394_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_709_cast_fp16, y = var_7394_to_fp16)[name = string("aw_chunk_709_cast_fp16")]; + fp16 var_7396_to_fp16 = const()[name = string("op_7396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_711_cast_fp16, y = var_7396_to_fp16)[name = string("aw_chunk_711_cast_fp16")]; + fp16 var_7398_to_fp16 = const()[name = string("op_7398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_713_cast_fp16, y = var_7398_to_fp16)[name = string("aw_chunk_713_cast_fp16")]; + fp16 var_7400_to_fp16 = const()[name = string("op_7400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_715_cast_fp16, y = var_7400_to_fp16)[name = string("aw_chunk_715_cast_fp16")]; + fp16 var_7402_to_fp16 = const()[name = string("op_7402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_717_cast_fp16, y = var_7402_to_fp16)[name = string("aw_chunk_717_cast_fp16")]; + fp16 var_7404_to_fp16 = const()[name = string("op_7404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_719_cast_fp16, y = var_7404_to_fp16)[name = string("aw_chunk_719_cast_fp16")]; + fp16 var_7406_to_fp16 = const()[name = string("op_7406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_721_cast_fp16, y = var_7406_to_fp16)[name = string("aw_chunk_721_cast_fp16")]; + fp16 var_7408_to_fp16 = const()[name = string("op_7408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_723_cast_fp16, y = var_7408_to_fp16)[name = string("aw_chunk_723_cast_fp16")]; + fp16 var_7410_to_fp16 = const()[name = string("op_7410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_725_cast_fp16, y = var_7410_to_fp16)[name = string("aw_chunk_725_cast_fp16")]; + fp16 var_7412_to_fp16 = const()[name = string("op_7412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_727_cast_fp16, y = var_7412_to_fp16)[name = string("aw_chunk_727_cast_fp16")]; + fp16 var_7414_to_fp16 = const()[name = string("op_7414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_729_cast_fp16, y = var_7414_to_fp16)[name = string("aw_chunk_729_cast_fp16")]; + fp16 var_7416_to_fp16 = const()[name = string("op_7416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_731_cast_fp16, y = var_7416_to_fp16)[name = string("aw_chunk_731_cast_fp16")]; + fp16 var_7418_to_fp16 = const()[name = string("op_7418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_733_cast_fp16, y = var_7418_to_fp16)[name = string("aw_chunk_733_cast_fp16")]; + fp16 var_7420_to_fp16 = const()[name = string("op_7420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_735_cast_fp16, y = var_7420_to_fp16)[name = string("aw_chunk_735_cast_fp16")]; + fp16 var_7422_to_fp16 = const()[name = string("op_7422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_737_cast_fp16, y = var_7422_to_fp16)[name = string("aw_chunk_737_cast_fp16")]; + fp16 var_7424_to_fp16 = const()[name = string("op_7424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_739_cast_fp16, y = var_7424_to_fp16)[name = string("aw_chunk_739_cast_fp16")]; + fp16 var_7426_to_fp16 = const()[name = string("op_7426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_741_cast_fp16, y = var_7426_to_fp16)[name = string("aw_chunk_741_cast_fp16")]; + fp16 var_7428_to_fp16 = const()[name = string("op_7428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_743_cast_fp16, y = var_7428_to_fp16)[name = string("aw_chunk_743_cast_fp16")]; + fp16 var_7430_to_fp16 = const()[name = string("op_7430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_745_cast_fp16, y = var_7430_to_fp16)[name = string("aw_chunk_745_cast_fp16")]; + fp16 var_7432_to_fp16 = const()[name = string("op_7432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_747_cast_fp16, y = var_7432_to_fp16)[name = string("aw_chunk_747_cast_fp16")]; + fp16 var_7434_to_fp16 = const()[name = string("op_7434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_749_cast_fp16, y = var_7434_to_fp16)[name = string("aw_chunk_749_cast_fp16")]; + fp16 var_7436_to_fp16 = const()[name = string("op_7436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_751_cast_fp16, y = var_7436_to_fp16)[name = string("aw_chunk_751_cast_fp16")]; + fp16 var_7438_to_fp16 = const()[name = string("op_7438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_753_cast_fp16, y = var_7438_to_fp16)[name = string("aw_chunk_753_cast_fp16")]; + fp16 var_7440_to_fp16 = const()[name = string("op_7440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_755_cast_fp16, y = var_7440_to_fp16)[name = string("aw_chunk_755_cast_fp16")]; + fp16 var_7442_to_fp16 = const()[name = string("op_7442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_757_cast_fp16, y = var_7442_to_fp16)[name = string("aw_chunk_757_cast_fp16")]; + fp16 var_7444_to_fp16 = const()[name = string("op_7444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_759_cast_fp16, y = var_7444_to_fp16)[name = string("aw_chunk_759_cast_fp16")]; + fp16 var_7446_to_fp16 = const()[name = string("op_7446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_761_cast_fp16, y = var_7446_to_fp16)[name = string("aw_chunk_761_cast_fp16")]; + fp16 var_7448_to_fp16 = const()[name = string("op_7448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_763_cast_fp16, y = var_7448_to_fp16)[name = string("aw_chunk_763_cast_fp16")]; + fp16 var_7450_to_fp16 = const()[name = string("op_7450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_765_cast_fp16, y = var_7450_to_fp16)[name = string("aw_chunk_765_cast_fp16")]; + fp16 var_7452_to_fp16 = const()[name = string("op_7452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_767_cast_fp16, y = var_7452_to_fp16)[name = string("aw_chunk_767_cast_fp16")]; + fp16 var_7454_to_fp16 = const()[name = string("op_7454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_769_cast_fp16, y = var_7454_to_fp16)[name = string("aw_chunk_769_cast_fp16")]; + fp16 var_7456_to_fp16 = const()[name = string("op_7456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_771_cast_fp16, y = var_7456_to_fp16)[name = string("aw_chunk_771_cast_fp16")]; + fp16 var_7458_to_fp16 = const()[name = string("op_7458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_773_cast_fp16, y = var_7458_to_fp16)[name = string("aw_chunk_773_cast_fp16")]; + fp16 var_7460_to_fp16 = const()[name = string("op_7460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_775_cast_fp16, y = var_7460_to_fp16)[name = string("aw_chunk_775_cast_fp16")]; + fp16 var_7462_to_fp16 = const()[name = string("op_7462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_777_cast_fp16, y = var_7462_to_fp16)[name = string("aw_chunk_777_cast_fp16")]; + fp16 var_7464_to_fp16 = const()[name = string("op_7464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_779_cast_fp16, y = var_7464_to_fp16)[name = string("aw_chunk_779_cast_fp16")]; + fp16 var_7466_to_fp16 = const()[name = string("op_7466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_781_cast_fp16, y = var_7466_to_fp16)[name = string("aw_chunk_781_cast_fp16")]; + fp16 var_7468_to_fp16 = const()[name = string("op_7468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_783_cast_fp16, y = var_7468_to_fp16)[name = string("aw_chunk_783_cast_fp16")]; + fp16 var_7470_to_fp16 = const()[name = string("op_7470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_785_cast_fp16, y = var_7470_to_fp16)[name = string("aw_chunk_785_cast_fp16")]; + fp16 var_7472_to_fp16 = const()[name = string("op_7472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_787_cast_fp16, y = var_7472_to_fp16)[name = string("aw_chunk_787_cast_fp16")]; + fp16 var_7474_to_fp16 = const()[name = string("op_7474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_789_cast_fp16, y = var_7474_to_fp16)[name = string("aw_chunk_789_cast_fp16")]; + fp16 var_7476_to_fp16 = const()[name = string("op_7476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_791_cast_fp16, y = var_7476_to_fp16)[name = string("aw_chunk_791_cast_fp16")]; + fp16 var_7478_to_fp16 = const()[name = string("op_7478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_793_cast_fp16, y = var_7478_to_fp16)[name = string("aw_chunk_793_cast_fp16")]; + fp16 var_7480_to_fp16 = const()[name = string("op_7480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_795_cast_fp16, y = var_7480_to_fp16)[name = string("aw_chunk_795_cast_fp16")]; + fp16 var_7482_to_fp16 = const()[name = string("op_7482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_797_cast_fp16, y = var_7482_to_fp16)[name = string("aw_chunk_797_cast_fp16")]; + fp16 var_7484_to_fp16 = const()[name = string("op_7484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_799_cast_fp16, y = var_7484_to_fp16)[name = string("aw_chunk_799_cast_fp16")]; + tensor var_7486_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_641_cast_fp16)[name = string("op_7486_cast_fp16")]; + tensor var_7487_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_643_cast_fp16)[name = string("op_7487_cast_fp16")]; + tensor var_7488_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_645_cast_fp16)[name = string("op_7488_cast_fp16")]; + tensor var_7489_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_647_cast_fp16)[name = string("op_7489_cast_fp16")]; + tensor var_7490_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_649_cast_fp16)[name = string("op_7490_cast_fp16")]; + tensor var_7491_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_651_cast_fp16)[name = string("op_7491_cast_fp16")]; + tensor var_7492_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_653_cast_fp16)[name = string("op_7492_cast_fp16")]; + tensor var_7493_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_655_cast_fp16)[name = string("op_7493_cast_fp16")]; + tensor var_7494_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_657_cast_fp16)[name = string("op_7494_cast_fp16")]; + tensor var_7495_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_659_cast_fp16)[name = string("op_7495_cast_fp16")]; + tensor var_7496_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_661_cast_fp16)[name = string("op_7496_cast_fp16")]; + tensor var_7497_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_663_cast_fp16)[name = string("op_7497_cast_fp16")]; + tensor var_7498_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_665_cast_fp16)[name = string("op_7498_cast_fp16")]; + tensor var_7499_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_667_cast_fp16)[name = string("op_7499_cast_fp16")]; + tensor var_7500_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_669_cast_fp16)[name = string("op_7500_cast_fp16")]; + tensor var_7501_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_671_cast_fp16)[name = string("op_7501_cast_fp16")]; + tensor var_7502_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_673_cast_fp16)[name = string("op_7502_cast_fp16")]; + tensor var_7503_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_675_cast_fp16)[name = string("op_7503_cast_fp16")]; + tensor var_7504_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_677_cast_fp16)[name = string("op_7504_cast_fp16")]; + tensor var_7505_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_679_cast_fp16)[name = string("op_7505_cast_fp16")]; + tensor var_7506_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_681_cast_fp16)[name = string("op_7506_cast_fp16")]; + tensor var_7507_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_683_cast_fp16)[name = string("op_7507_cast_fp16")]; + tensor var_7508_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_685_cast_fp16)[name = string("op_7508_cast_fp16")]; + tensor var_7509_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_687_cast_fp16)[name = string("op_7509_cast_fp16")]; + tensor var_7510_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_689_cast_fp16)[name = string("op_7510_cast_fp16")]; + tensor var_7511_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_691_cast_fp16)[name = string("op_7511_cast_fp16")]; + tensor var_7512_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_693_cast_fp16)[name = string("op_7512_cast_fp16")]; + tensor var_7513_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_695_cast_fp16)[name = string("op_7513_cast_fp16")]; + tensor var_7514_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_697_cast_fp16)[name = string("op_7514_cast_fp16")]; + tensor var_7515_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_699_cast_fp16)[name = string("op_7515_cast_fp16")]; + tensor var_7516_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_701_cast_fp16)[name = string("op_7516_cast_fp16")]; + tensor var_7517_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_703_cast_fp16)[name = string("op_7517_cast_fp16")]; + tensor var_7518_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_705_cast_fp16)[name = string("op_7518_cast_fp16")]; + tensor var_7519_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_707_cast_fp16)[name = string("op_7519_cast_fp16")]; + tensor var_7520_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_709_cast_fp16)[name = string("op_7520_cast_fp16")]; + tensor var_7521_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_711_cast_fp16)[name = string("op_7521_cast_fp16")]; + tensor var_7522_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_713_cast_fp16)[name = string("op_7522_cast_fp16")]; + tensor var_7523_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_715_cast_fp16)[name = string("op_7523_cast_fp16")]; + tensor var_7524_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_717_cast_fp16)[name = string("op_7524_cast_fp16")]; + tensor var_7525_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_719_cast_fp16)[name = string("op_7525_cast_fp16")]; + tensor var_7526_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_721_cast_fp16)[name = string("op_7526_cast_fp16")]; + tensor var_7527_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_723_cast_fp16)[name = string("op_7527_cast_fp16")]; + tensor var_7528_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_725_cast_fp16)[name = string("op_7528_cast_fp16")]; + tensor var_7529_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_727_cast_fp16)[name = string("op_7529_cast_fp16")]; + tensor var_7530_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_729_cast_fp16)[name = string("op_7530_cast_fp16")]; + tensor var_7531_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_731_cast_fp16)[name = string("op_7531_cast_fp16")]; + tensor var_7532_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_733_cast_fp16)[name = string("op_7532_cast_fp16")]; + tensor var_7533_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_735_cast_fp16)[name = string("op_7533_cast_fp16")]; + tensor var_7534_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_737_cast_fp16)[name = string("op_7534_cast_fp16")]; + tensor var_7535_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_739_cast_fp16)[name = string("op_7535_cast_fp16")]; + tensor var_7536_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_741_cast_fp16)[name = string("op_7536_cast_fp16")]; + tensor var_7537_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_743_cast_fp16)[name = string("op_7537_cast_fp16")]; + tensor var_7538_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_745_cast_fp16)[name = string("op_7538_cast_fp16")]; + tensor var_7539_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_747_cast_fp16)[name = string("op_7539_cast_fp16")]; + tensor var_7540_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_749_cast_fp16)[name = string("op_7540_cast_fp16")]; + tensor var_7541_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_751_cast_fp16)[name = string("op_7541_cast_fp16")]; + tensor var_7542_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_753_cast_fp16)[name = string("op_7542_cast_fp16")]; + tensor var_7543_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_755_cast_fp16)[name = string("op_7543_cast_fp16")]; + tensor var_7544_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_757_cast_fp16)[name = string("op_7544_cast_fp16")]; + tensor var_7545_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_759_cast_fp16)[name = string("op_7545_cast_fp16")]; + tensor var_7546_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_761_cast_fp16)[name = string("op_7546_cast_fp16")]; + tensor var_7547_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_763_cast_fp16)[name = string("op_7547_cast_fp16")]; + tensor var_7548_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_765_cast_fp16)[name = string("op_7548_cast_fp16")]; + tensor var_7549_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_767_cast_fp16)[name = string("op_7549_cast_fp16")]; + tensor var_7550_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_769_cast_fp16)[name = string("op_7550_cast_fp16")]; + tensor var_7551_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_771_cast_fp16)[name = string("op_7551_cast_fp16")]; + tensor var_7552_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_773_cast_fp16)[name = string("op_7552_cast_fp16")]; + tensor var_7553_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_775_cast_fp16)[name = string("op_7553_cast_fp16")]; + tensor var_7554_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_777_cast_fp16)[name = string("op_7554_cast_fp16")]; + tensor var_7555_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_779_cast_fp16)[name = string("op_7555_cast_fp16")]; + tensor var_7556_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_781_cast_fp16)[name = string("op_7556_cast_fp16")]; + tensor var_7557_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_783_cast_fp16)[name = string("op_7557_cast_fp16")]; + tensor var_7558_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_785_cast_fp16)[name = string("op_7558_cast_fp16")]; + tensor var_7559_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_787_cast_fp16)[name = string("op_7559_cast_fp16")]; + tensor var_7560_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_789_cast_fp16)[name = string("op_7560_cast_fp16")]; + tensor var_7561_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_791_cast_fp16)[name = string("op_7561_cast_fp16")]; + tensor var_7562_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_793_cast_fp16)[name = string("op_7562_cast_fp16")]; + tensor var_7563_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_795_cast_fp16)[name = string("op_7563_cast_fp16")]; + tensor var_7564_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_797_cast_fp16)[name = string("op_7564_cast_fp16")]; + tensor var_7565_cast_fp16 = softmax(axis = var_6311, x = aw_chunk_799_cast_fp16)[name = string("op_7565_cast_fp16")]; + string var_7567_equation_0 = const()[name = string("op_7567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7567_cast_fp16 = einsum(equation = var_7567_equation_0, values = (var_7087_cast_fp16, var_7486_cast_fp16))[name = string("op_7567_cast_fp16")]; + string var_7569_equation_0 = const()[name = string("op_7569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7569_cast_fp16 = einsum(equation = var_7569_equation_0, values = (var_7087_cast_fp16, var_7487_cast_fp16))[name = string("op_7569_cast_fp16")]; + string var_7571_equation_0 = const()[name = string("op_7571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7571_cast_fp16 = einsum(equation = var_7571_equation_0, values = (var_7087_cast_fp16, var_7488_cast_fp16))[name = string("op_7571_cast_fp16")]; + string var_7573_equation_0 = const()[name = string("op_7573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7573_cast_fp16 = einsum(equation = var_7573_equation_0, values = (var_7087_cast_fp16, var_7489_cast_fp16))[name = string("op_7573_cast_fp16")]; + string var_7575_equation_0 = const()[name = string("op_7575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7575_cast_fp16 = einsum(equation = var_7575_equation_0, values = (var_7091_cast_fp16, var_7490_cast_fp16))[name = string("op_7575_cast_fp16")]; + string var_7577_equation_0 = const()[name = string("op_7577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7577_cast_fp16 = einsum(equation = var_7577_equation_0, values = (var_7091_cast_fp16, var_7491_cast_fp16))[name = string("op_7577_cast_fp16")]; + string var_7579_equation_0 = const()[name = string("op_7579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7579_cast_fp16 = einsum(equation = var_7579_equation_0, values = (var_7091_cast_fp16, var_7492_cast_fp16))[name = string("op_7579_cast_fp16")]; + string var_7581_equation_0 = const()[name = string("op_7581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7581_cast_fp16 = einsum(equation = var_7581_equation_0, values = (var_7091_cast_fp16, var_7493_cast_fp16))[name = string("op_7581_cast_fp16")]; + string var_7583_equation_0 = const()[name = string("op_7583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7583_cast_fp16 = einsum(equation = var_7583_equation_0, values = (var_7095_cast_fp16, var_7494_cast_fp16))[name = string("op_7583_cast_fp16")]; + string var_7585_equation_0 = const()[name = string("op_7585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7585_cast_fp16 = einsum(equation = var_7585_equation_0, values = (var_7095_cast_fp16, var_7495_cast_fp16))[name = string("op_7585_cast_fp16")]; + string var_7587_equation_0 = const()[name = string("op_7587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7587_cast_fp16 = einsum(equation = var_7587_equation_0, values = (var_7095_cast_fp16, var_7496_cast_fp16))[name = string("op_7587_cast_fp16")]; + string var_7589_equation_0 = const()[name = string("op_7589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7589_cast_fp16 = einsum(equation = var_7589_equation_0, values = (var_7095_cast_fp16, var_7497_cast_fp16))[name = string("op_7589_cast_fp16")]; + string var_7591_equation_0 = const()[name = string("op_7591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7591_cast_fp16 = einsum(equation = var_7591_equation_0, values = (var_7099_cast_fp16, var_7498_cast_fp16))[name = string("op_7591_cast_fp16")]; + string var_7593_equation_0 = const()[name = string("op_7593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7593_cast_fp16 = einsum(equation = var_7593_equation_0, values = (var_7099_cast_fp16, var_7499_cast_fp16))[name = string("op_7593_cast_fp16")]; + string var_7595_equation_0 = const()[name = string("op_7595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7595_cast_fp16 = einsum(equation = var_7595_equation_0, values = (var_7099_cast_fp16, var_7500_cast_fp16))[name = string("op_7595_cast_fp16")]; + string var_7597_equation_0 = const()[name = string("op_7597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7597_cast_fp16 = einsum(equation = var_7597_equation_0, values = (var_7099_cast_fp16, var_7501_cast_fp16))[name = string("op_7597_cast_fp16")]; + string var_7599_equation_0 = const()[name = string("op_7599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7599_cast_fp16 = einsum(equation = var_7599_equation_0, values = (var_7103_cast_fp16, var_7502_cast_fp16))[name = string("op_7599_cast_fp16")]; + string var_7601_equation_0 = const()[name = string("op_7601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7601_cast_fp16 = einsum(equation = var_7601_equation_0, values = (var_7103_cast_fp16, var_7503_cast_fp16))[name = string("op_7601_cast_fp16")]; + string var_7603_equation_0 = const()[name = string("op_7603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7603_cast_fp16 = einsum(equation = var_7603_equation_0, values = (var_7103_cast_fp16, var_7504_cast_fp16))[name = string("op_7603_cast_fp16")]; + string var_7605_equation_0 = const()[name = string("op_7605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7605_cast_fp16 = einsum(equation = var_7605_equation_0, values = (var_7103_cast_fp16, var_7505_cast_fp16))[name = string("op_7605_cast_fp16")]; + string var_7607_equation_0 = const()[name = string("op_7607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7607_cast_fp16 = einsum(equation = var_7607_equation_0, values = (var_7107_cast_fp16, var_7506_cast_fp16))[name = string("op_7607_cast_fp16")]; + string var_7609_equation_0 = const()[name = string("op_7609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7609_cast_fp16 = einsum(equation = var_7609_equation_0, values = (var_7107_cast_fp16, var_7507_cast_fp16))[name = string("op_7609_cast_fp16")]; + string var_7611_equation_0 = const()[name = string("op_7611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7611_cast_fp16 = einsum(equation = var_7611_equation_0, values = (var_7107_cast_fp16, var_7508_cast_fp16))[name = string("op_7611_cast_fp16")]; + string var_7613_equation_0 = const()[name = string("op_7613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7613_cast_fp16 = einsum(equation = var_7613_equation_0, values = (var_7107_cast_fp16, var_7509_cast_fp16))[name = string("op_7613_cast_fp16")]; + string var_7615_equation_0 = const()[name = string("op_7615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7615_cast_fp16 = einsum(equation = var_7615_equation_0, values = (var_7111_cast_fp16, var_7510_cast_fp16))[name = string("op_7615_cast_fp16")]; + string var_7617_equation_0 = const()[name = string("op_7617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7617_cast_fp16 = einsum(equation = var_7617_equation_0, values = (var_7111_cast_fp16, var_7511_cast_fp16))[name = string("op_7617_cast_fp16")]; + string var_7619_equation_0 = const()[name = string("op_7619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7619_cast_fp16 = einsum(equation = var_7619_equation_0, values = (var_7111_cast_fp16, var_7512_cast_fp16))[name = string("op_7619_cast_fp16")]; + string var_7621_equation_0 = const()[name = string("op_7621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7621_cast_fp16 = einsum(equation = var_7621_equation_0, values = (var_7111_cast_fp16, var_7513_cast_fp16))[name = string("op_7621_cast_fp16")]; + string var_7623_equation_0 = const()[name = string("op_7623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7623_cast_fp16 = einsum(equation = var_7623_equation_0, values = (var_7115_cast_fp16, var_7514_cast_fp16))[name = string("op_7623_cast_fp16")]; + string var_7625_equation_0 = const()[name = string("op_7625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7625_cast_fp16 = einsum(equation = var_7625_equation_0, values = (var_7115_cast_fp16, var_7515_cast_fp16))[name = string("op_7625_cast_fp16")]; + string var_7627_equation_0 = const()[name = string("op_7627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7627_cast_fp16 = einsum(equation = var_7627_equation_0, values = (var_7115_cast_fp16, var_7516_cast_fp16))[name = string("op_7627_cast_fp16")]; + string var_7629_equation_0 = const()[name = string("op_7629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7629_cast_fp16 = einsum(equation = var_7629_equation_0, values = (var_7115_cast_fp16, var_7517_cast_fp16))[name = string("op_7629_cast_fp16")]; + string var_7631_equation_0 = const()[name = string("op_7631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7631_cast_fp16 = einsum(equation = var_7631_equation_0, values = (var_7119_cast_fp16, var_7518_cast_fp16))[name = string("op_7631_cast_fp16")]; + string var_7633_equation_0 = const()[name = string("op_7633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7633_cast_fp16 = einsum(equation = var_7633_equation_0, values = (var_7119_cast_fp16, var_7519_cast_fp16))[name = string("op_7633_cast_fp16")]; + string var_7635_equation_0 = const()[name = string("op_7635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7635_cast_fp16 = einsum(equation = var_7635_equation_0, values = (var_7119_cast_fp16, var_7520_cast_fp16))[name = string("op_7635_cast_fp16")]; + string var_7637_equation_0 = const()[name = string("op_7637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7637_cast_fp16 = einsum(equation = var_7637_equation_0, values = (var_7119_cast_fp16, var_7521_cast_fp16))[name = string("op_7637_cast_fp16")]; + string var_7639_equation_0 = const()[name = string("op_7639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7639_cast_fp16 = einsum(equation = var_7639_equation_0, values = (var_7123_cast_fp16, var_7522_cast_fp16))[name = string("op_7639_cast_fp16")]; + string var_7641_equation_0 = const()[name = string("op_7641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7641_cast_fp16 = einsum(equation = var_7641_equation_0, values = (var_7123_cast_fp16, var_7523_cast_fp16))[name = string("op_7641_cast_fp16")]; + string var_7643_equation_0 = const()[name = string("op_7643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7643_cast_fp16 = einsum(equation = var_7643_equation_0, values = (var_7123_cast_fp16, var_7524_cast_fp16))[name = string("op_7643_cast_fp16")]; + string var_7645_equation_0 = const()[name = string("op_7645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7645_cast_fp16 = einsum(equation = var_7645_equation_0, values = (var_7123_cast_fp16, var_7525_cast_fp16))[name = string("op_7645_cast_fp16")]; + string var_7647_equation_0 = const()[name = string("op_7647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7647_cast_fp16 = einsum(equation = var_7647_equation_0, values = (var_7127_cast_fp16, var_7526_cast_fp16))[name = string("op_7647_cast_fp16")]; + string var_7649_equation_0 = const()[name = string("op_7649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7649_cast_fp16 = einsum(equation = var_7649_equation_0, values = (var_7127_cast_fp16, var_7527_cast_fp16))[name = string("op_7649_cast_fp16")]; + string var_7651_equation_0 = const()[name = string("op_7651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7651_cast_fp16 = einsum(equation = var_7651_equation_0, values = (var_7127_cast_fp16, var_7528_cast_fp16))[name = string("op_7651_cast_fp16")]; + string var_7653_equation_0 = const()[name = string("op_7653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7653_cast_fp16 = einsum(equation = var_7653_equation_0, values = (var_7127_cast_fp16, var_7529_cast_fp16))[name = string("op_7653_cast_fp16")]; + string var_7655_equation_0 = const()[name = string("op_7655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7655_cast_fp16 = einsum(equation = var_7655_equation_0, values = (var_7131_cast_fp16, var_7530_cast_fp16))[name = string("op_7655_cast_fp16")]; + string var_7657_equation_0 = const()[name = string("op_7657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7657_cast_fp16 = einsum(equation = var_7657_equation_0, values = (var_7131_cast_fp16, var_7531_cast_fp16))[name = string("op_7657_cast_fp16")]; + string var_7659_equation_0 = const()[name = string("op_7659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7659_cast_fp16 = einsum(equation = var_7659_equation_0, values = (var_7131_cast_fp16, var_7532_cast_fp16))[name = string("op_7659_cast_fp16")]; + string var_7661_equation_0 = const()[name = string("op_7661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7661_cast_fp16 = einsum(equation = var_7661_equation_0, values = (var_7131_cast_fp16, var_7533_cast_fp16))[name = string("op_7661_cast_fp16")]; + string var_7663_equation_0 = const()[name = string("op_7663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7663_cast_fp16 = einsum(equation = var_7663_equation_0, values = (var_7135_cast_fp16, var_7534_cast_fp16))[name = string("op_7663_cast_fp16")]; + string var_7665_equation_0 = const()[name = string("op_7665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7665_cast_fp16 = einsum(equation = var_7665_equation_0, values = (var_7135_cast_fp16, var_7535_cast_fp16))[name = string("op_7665_cast_fp16")]; + string var_7667_equation_0 = const()[name = string("op_7667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7667_cast_fp16 = einsum(equation = var_7667_equation_0, values = (var_7135_cast_fp16, var_7536_cast_fp16))[name = string("op_7667_cast_fp16")]; + string var_7669_equation_0 = const()[name = string("op_7669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7669_cast_fp16 = einsum(equation = var_7669_equation_0, values = (var_7135_cast_fp16, var_7537_cast_fp16))[name = string("op_7669_cast_fp16")]; + string var_7671_equation_0 = const()[name = string("op_7671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7671_cast_fp16 = einsum(equation = var_7671_equation_0, values = (var_7139_cast_fp16, var_7538_cast_fp16))[name = string("op_7671_cast_fp16")]; + string var_7673_equation_0 = const()[name = string("op_7673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7673_cast_fp16 = einsum(equation = var_7673_equation_0, values = (var_7139_cast_fp16, var_7539_cast_fp16))[name = string("op_7673_cast_fp16")]; + string var_7675_equation_0 = const()[name = string("op_7675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7675_cast_fp16 = einsum(equation = var_7675_equation_0, values = (var_7139_cast_fp16, var_7540_cast_fp16))[name = string("op_7675_cast_fp16")]; + string var_7677_equation_0 = const()[name = string("op_7677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7677_cast_fp16 = einsum(equation = var_7677_equation_0, values = (var_7139_cast_fp16, var_7541_cast_fp16))[name = string("op_7677_cast_fp16")]; + string var_7679_equation_0 = const()[name = string("op_7679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7679_cast_fp16 = einsum(equation = var_7679_equation_0, values = (var_7143_cast_fp16, var_7542_cast_fp16))[name = string("op_7679_cast_fp16")]; + string var_7681_equation_0 = const()[name = string("op_7681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7681_cast_fp16 = einsum(equation = var_7681_equation_0, values = (var_7143_cast_fp16, var_7543_cast_fp16))[name = string("op_7681_cast_fp16")]; + string var_7683_equation_0 = const()[name = string("op_7683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7683_cast_fp16 = einsum(equation = var_7683_equation_0, values = (var_7143_cast_fp16, var_7544_cast_fp16))[name = string("op_7683_cast_fp16")]; + string var_7685_equation_0 = const()[name = string("op_7685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7685_cast_fp16 = einsum(equation = var_7685_equation_0, values = (var_7143_cast_fp16, var_7545_cast_fp16))[name = string("op_7685_cast_fp16")]; + string var_7687_equation_0 = const()[name = string("op_7687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7687_cast_fp16 = einsum(equation = var_7687_equation_0, values = (var_7147_cast_fp16, var_7546_cast_fp16))[name = string("op_7687_cast_fp16")]; + string var_7689_equation_0 = const()[name = string("op_7689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7689_cast_fp16 = einsum(equation = var_7689_equation_0, values = (var_7147_cast_fp16, var_7547_cast_fp16))[name = string("op_7689_cast_fp16")]; + string var_7691_equation_0 = const()[name = string("op_7691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7691_cast_fp16 = einsum(equation = var_7691_equation_0, values = (var_7147_cast_fp16, var_7548_cast_fp16))[name = string("op_7691_cast_fp16")]; + string var_7693_equation_0 = const()[name = string("op_7693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7693_cast_fp16 = einsum(equation = var_7693_equation_0, values = (var_7147_cast_fp16, var_7549_cast_fp16))[name = string("op_7693_cast_fp16")]; + string var_7695_equation_0 = const()[name = string("op_7695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7695_cast_fp16 = einsum(equation = var_7695_equation_0, values = (var_7151_cast_fp16, var_7550_cast_fp16))[name = string("op_7695_cast_fp16")]; + string var_7697_equation_0 = const()[name = string("op_7697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7697_cast_fp16 = einsum(equation = var_7697_equation_0, values = (var_7151_cast_fp16, var_7551_cast_fp16))[name = string("op_7697_cast_fp16")]; + string var_7699_equation_0 = const()[name = string("op_7699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7699_cast_fp16 = einsum(equation = var_7699_equation_0, values = (var_7151_cast_fp16, var_7552_cast_fp16))[name = string("op_7699_cast_fp16")]; + string var_7701_equation_0 = const()[name = string("op_7701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7701_cast_fp16 = einsum(equation = var_7701_equation_0, values = (var_7151_cast_fp16, var_7553_cast_fp16))[name = string("op_7701_cast_fp16")]; + string var_7703_equation_0 = const()[name = string("op_7703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7703_cast_fp16 = einsum(equation = var_7703_equation_0, values = (var_7155_cast_fp16, var_7554_cast_fp16))[name = string("op_7703_cast_fp16")]; + string var_7705_equation_0 = const()[name = string("op_7705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7705_cast_fp16 = einsum(equation = var_7705_equation_0, values = (var_7155_cast_fp16, var_7555_cast_fp16))[name = string("op_7705_cast_fp16")]; + string var_7707_equation_0 = const()[name = string("op_7707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7707_cast_fp16 = einsum(equation = var_7707_equation_0, values = (var_7155_cast_fp16, var_7556_cast_fp16))[name = string("op_7707_cast_fp16")]; + string var_7709_equation_0 = const()[name = string("op_7709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7709_cast_fp16 = einsum(equation = var_7709_equation_0, values = (var_7155_cast_fp16, var_7557_cast_fp16))[name = string("op_7709_cast_fp16")]; + string var_7711_equation_0 = const()[name = string("op_7711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7711_cast_fp16 = einsum(equation = var_7711_equation_0, values = (var_7159_cast_fp16, var_7558_cast_fp16))[name = string("op_7711_cast_fp16")]; + string var_7713_equation_0 = const()[name = string("op_7713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7713_cast_fp16 = einsum(equation = var_7713_equation_0, values = (var_7159_cast_fp16, var_7559_cast_fp16))[name = string("op_7713_cast_fp16")]; + string var_7715_equation_0 = const()[name = string("op_7715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7715_cast_fp16 = einsum(equation = var_7715_equation_0, values = (var_7159_cast_fp16, var_7560_cast_fp16))[name = string("op_7715_cast_fp16")]; + string var_7717_equation_0 = const()[name = string("op_7717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7717_cast_fp16 = einsum(equation = var_7717_equation_0, values = (var_7159_cast_fp16, var_7561_cast_fp16))[name = string("op_7717_cast_fp16")]; + string var_7719_equation_0 = const()[name = string("op_7719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7719_cast_fp16 = einsum(equation = var_7719_equation_0, values = (var_7163_cast_fp16, var_7562_cast_fp16))[name = string("op_7719_cast_fp16")]; + string var_7721_equation_0 = const()[name = string("op_7721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7721_cast_fp16 = einsum(equation = var_7721_equation_0, values = (var_7163_cast_fp16, var_7563_cast_fp16))[name = string("op_7721_cast_fp16")]; + string var_7723_equation_0 = const()[name = string("op_7723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7723_cast_fp16 = einsum(equation = var_7723_equation_0, values = (var_7163_cast_fp16, var_7564_cast_fp16))[name = string("op_7723_cast_fp16")]; + string var_7725_equation_0 = const()[name = string("op_7725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7725_cast_fp16 = einsum(equation = var_7725_equation_0, values = (var_7163_cast_fp16, var_7565_cast_fp16))[name = string("op_7725_cast_fp16")]; + bool var_7727_interleave_0 = const()[name = string("op_7727_interleave_0"), val = bool(false)]; + tensor var_7727_cast_fp16 = concat(axis = var_6286, interleave = var_7727_interleave_0, values = (var_7567_cast_fp16, var_7569_cast_fp16, var_7571_cast_fp16, var_7573_cast_fp16))[name = string("op_7727_cast_fp16")]; + bool var_7729_interleave_0 = const()[name = string("op_7729_interleave_0"), val = bool(false)]; + tensor var_7729_cast_fp16 = concat(axis = var_6286, interleave = var_7729_interleave_0, values = (var_7575_cast_fp16, var_7577_cast_fp16, var_7579_cast_fp16, var_7581_cast_fp16))[name = string("op_7729_cast_fp16")]; + bool var_7731_interleave_0 = const()[name = string("op_7731_interleave_0"), val = bool(false)]; + tensor var_7731_cast_fp16 = concat(axis = var_6286, interleave = var_7731_interleave_0, values = (var_7583_cast_fp16, var_7585_cast_fp16, var_7587_cast_fp16, var_7589_cast_fp16))[name = string("op_7731_cast_fp16")]; + bool var_7733_interleave_0 = const()[name = string("op_7733_interleave_0"), val = bool(false)]; + tensor var_7733_cast_fp16 = concat(axis = var_6286, interleave = var_7733_interleave_0, values = (var_7591_cast_fp16, var_7593_cast_fp16, var_7595_cast_fp16, var_7597_cast_fp16))[name = string("op_7733_cast_fp16")]; + bool var_7735_interleave_0 = const()[name = string("op_7735_interleave_0"), val = bool(false)]; + tensor var_7735_cast_fp16 = concat(axis = var_6286, interleave = var_7735_interleave_0, values = (var_7599_cast_fp16, var_7601_cast_fp16, var_7603_cast_fp16, var_7605_cast_fp16))[name = string("op_7735_cast_fp16")]; + bool var_7737_interleave_0 = const()[name = string("op_7737_interleave_0"), val = bool(false)]; + tensor var_7737_cast_fp16 = concat(axis = var_6286, interleave = var_7737_interleave_0, values = (var_7607_cast_fp16, var_7609_cast_fp16, var_7611_cast_fp16, var_7613_cast_fp16))[name = string("op_7737_cast_fp16")]; + bool var_7739_interleave_0 = const()[name = string("op_7739_interleave_0"), val = bool(false)]; + tensor var_7739_cast_fp16 = concat(axis = var_6286, interleave = var_7739_interleave_0, values = (var_7615_cast_fp16, var_7617_cast_fp16, var_7619_cast_fp16, var_7621_cast_fp16))[name = string("op_7739_cast_fp16")]; + bool var_7741_interleave_0 = const()[name = string("op_7741_interleave_0"), val = bool(false)]; + tensor var_7741_cast_fp16 = concat(axis = var_6286, interleave = var_7741_interleave_0, values = (var_7623_cast_fp16, var_7625_cast_fp16, var_7627_cast_fp16, var_7629_cast_fp16))[name = string("op_7741_cast_fp16")]; + bool var_7743_interleave_0 = const()[name = string("op_7743_interleave_0"), val = bool(false)]; + tensor var_7743_cast_fp16 = concat(axis = var_6286, interleave = var_7743_interleave_0, values = (var_7631_cast_fp16, var_7633_cast_fp16, var_7635_cast_fp16, var_7637_cast_fp16))[name = string("op_7743_cast_fp16")]; + bool var_7745_interleave_0 = const()[name = string("op_7745_interleave_0"), val = bool(false)]; + tensor var_7745_cast_fp16 = concat(axis = var_6286, interleave = var_7745_interleave_0, values = (var_7639_cast_fp16, var_7641_cast_fp16, var_7643_cast_fp16, var_7645_cast_fp16))[name = string("op_7745_cast_fp16")]; + bool var_7747_interleave_0 = const()[name = string("op_7747_interleave_0"), val = bool(false)]; + tensor var_7747_cast_fp16 = concat(axis = var_6286, interleave = var_7747_interleave_0, values = (var_7647_cast_fp16, var_7649_cast_fp16, var_7651_cast_fp16, var_7653_cast_fp16))[name = string("op_7747_cast_fp16")]; + bool var_7749_interleave_0 = const()[name = string("op_7749_interleave_0"), val = bool(false)]; + tensor var_7749_cast_fp16 = concat(axis = var_6286, interleave = var_7749_interleave_0, values = (var_7655_cast_fp16, var_7657_cast_fp16, var_7659_cast_fp16, var_7661_cast_fp16))[name = string("op_7749_cast_fp16")]; + bool var_7751_interleave_0 = const()[name = string("op_7751_interleave_0"), val = bool(false)]; + tensor var_7751_cast_fp16 = concat(axis = var_6286, interleave = var_7751_interleave_0, values = (var_7663_cast_fp16, var_7665_cast_fp16, var_7667_cast_fp16, var_7669_cast_fp16))[name = string("op_7751_cast_fp16")]; + bool var_7753_interleave_0 = const()[name = string("op_7753_interleave_0"), val = bool(false)]; + tensor var_7753_cast_fp16 = concat(axis = var_6286, interleave = var_7753_interleave_0, values = (var_7671_cast_fp16, var_7673_cast_fp16, var_7675_cast_fp16, var_7677_cast_fp16))[name = string("op_7753_cast_fp16")]; + bool var_7755_interleave_0 = const()[name = string("op_7755_interleave_0"), val = bool(false)]; + tensor var_7755_cast_fp16 = concat(axis = var_6286, interleave = var_7755_interleave_0, values = (var_7679_cast_fp16, var_7681_cast_fp16, var_7683_cast_fp16, var_7685_cast_fp16))[name = string("op_7755_cast_fp16")]; + bool var_7757_interleave_0 = const()[name = string("op_7757_interleave_0"), val = bool(false)]; + tensor var_7757_cast_fp16 = concat(axis = var_6286, interleave = var_7757_interleave_0, values = (var_7687_cast_fp16, var_7689_cast_fp16, var_7691_cast_fp16, var_7693_cast_fp16))[name = string("op_7757_cast_fp16")]; + bool var_7759_interleave_0 = const()[name = string("op_7759_interleave_0"), val = bool(false)]; + tensor var_7759_cast_fp16 = concat(axis = var_6286, interleave = var_7759_interleave_0, values = (var_7695_cast_fp16, var_7697_cast_fp16, var_7699_cast_fp16, var_7701_cast_fp16))[name = string("op_7759_cast_fp16")]; + bool var_7761_interleave_0 = const()[name = string("op_7761_interleave_0"), val = bool(false)]; + tensor var_7761_cast_fp16 = concat(axis = var_6286, interleave = var_7761_interleave_0, values = (var_7703_cast_fp16, var_7705_cast_fp16, var_7707_cast_fp16, var_7709_cast_fp16))[name = string("op_7761_cast_fp16")]; + bool var_7763_interleave_0 = const()[name = string("op_7763_interleave_0"), val = bool(false)]; + tensor var_7763_cast_fp16 = concat(axis = var_6286, interleave = var_7763_interleave_0, values = (var_7711_cast_fp16, var_7713_cast_fp16, var_7715_cast_fp16, var_7717_cast_fp16))[name = string("op_7763_cast_fp16")]; + bool var_7765_interleave_0 = const()[name = string("op_7765_interleave_0"), val = bool(false)]; + tensor var_7765_cast_fp16 = concat(axis = var_6286, interleave = var_7765_interleave_0, values = (var_7719_cast_fp16, var_7721_cast_fp16, var_7723_cast_fp16, var_7725_cast_fp16))[name = string("op_7765_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_6311, interleave = input_33_interleave_0, values = (var_7727_cast_fp16, var_7729_cast_fp16, var_7731_cast_fp16, var_7733_cast_fp16, var_7735_cast_fp16, var_7737_cast_fp16, var_7739_cast_fp16, var_7741_cast_fp16, var_7743_cast_fp16, var_7745_cast_fp16, var_7747_cast_fp16, var_7749_cast_fp16, var_7751_cast_fp16, var_7753_cast_fp16, var_7755_cast_fp16, var_7757_cast_fp16, var_7759_cast_fp16, var_7761_cast_fp16, var_7763_cast_fp16, var_7765_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181918400)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185195264)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_7784_to_fp16 = const()[name = string("op_7784_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_7784_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185197888)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185200512)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185203136)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198310400)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198320704)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211427968)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_7813 = const()[name = string("op_7813"), val = int32(3)]; + int32 var_7838 = const()[name = string("op_7838"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_7855_to_fp16 = const()[name = string("op_7855_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_7855_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211430592)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211433216)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211435840)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214712704)))]; + tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; + string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; + tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; + tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; + int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214715328)))]; + tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; + string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; + tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; + tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; + int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217992192)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221269056)))]; + tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_7893_begin_0 = const()[name = string("op_7893_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7893_end_0 = const()[name = string("op_7893_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7893_end_mask_0 = const()[name = string("op_7893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7893_cast_fp16 = slice_by_index(begin = var_7893_begin_0, end = var_7893_end_0, end_mask = var_7893_end_mask_0, x = query_11_cast_fp16)[name = string("op_7893_cast_fp16")]; + tensor var_7897_begin_0 = const()[name = string("op_7897_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7897_end_0 = const()[name = string("op_7897_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7897_end_mask_0 = const()[name = string("op_7897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7897_cast_fp16 = slice_by_index(begin = var_7897_begin_0, end = var_7897_end_0, end_mask = var_7897_end_mask_0, x = query_11_cast_fp16)[name = string("op_7897_cast_fp16")]; + tensor var_7901_begin_0 = const()[name = string("op_7901_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7901_end_0 = const()[name = string("op_7901_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7901_end_mask_0 = const()[name = string("op_7901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7901_cast_fp16 = slice_by_index(begin = var_7901_begin_0, end = var_7901_end_0, end_mask = var_7901_end_mask_0, x = query_11_cast_fp16)[name = string("op_7901_cast_fp16")]; + tensor var_7905_begin_0 = const()[name = string("op_7905_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7905_end_0 = const()[name = string("op_7905_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7905_end_mask_0 = const()[name = string("op_7905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7905_cast_fp16 = slice_by_index(begin = var_7905_begin_0, end = var_7905_end_0, end_mask = var_7905_end_mask_0, x = query_11_cast_fp16)[name = string("op_7905_cast_fp16")]; + tensor var_7909_begin_0 = const()[name = string("op_7909_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7909_end_0 = const()[name = string("op_7909_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7909_end_mask_0 = const()[name = string("op_7909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7909_cast_fp16 = slice_by_index(begin = var_7909_begin_0, end = var_7909_end_0, end_mask = var_7909_end_mask_0, x = query_11_cast_fp16)[name = string("op_7909_cast_fp16")]; + tensor var_7913_begin_0 = const()[name = string("op_7913_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7913_end_0 = const()[name = string("op_7913_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7913_end_mask_0 = const()[name = string("op_7913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7913_cast_fp16 = slice_by_index(begin = var_7913_begin_0, end = var_7913_end_0, end_mask = var_7913_end_mask_0, x = query_11_cast_fp16)[name = string("op_7913_cast_fp16")]; + tensor var_7917_begin_0 = const()[name = string("op_7917_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7917_end_0 = const()[name = string("op_7917_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7917_end_mask_0 = const()[name = string("op_7917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7917_cast_fp16 = slice_by_index(begin = var_7917_begin_0, end = var_7917_end_0, end_mask = var_7917_end_mask_0, x = query_11_cast_fp16)[name = string("op_7917_cast_fp16")]; + tensor var_7921_begin_0 = const()[name = string("op_7921_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7921_end_0 = const()[name = string("op_7921_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7921_end_mask_0 = const()[name = string("op_7921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7921_cast_fp16 = slice_by_index(begin = var_7921_begin_0, end = var_7921_end_0, end_mask = var_7921_end_mask_0, x = query_11_cast_fp16)[name = string("op_7921_cast_fp16")]; + tensor var_7925_begin_0 = const()[name = string("op_7925_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7925_end_0 = const()[name = string("op_7925_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7925_end_mask_0 = const()[name = string("op_7925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7925_cast_fp16 = slice_by_index(begin = var_7925_begin_0, end = var_7925_end_0, end_mask = var_7925_end_mask_0, x = query_11_cast_fp16)[name = string("op_7925_cast_fp16")]; + tensor var_7929_begin_0 = const()[name = string("op_7929_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7929_end_0 = const()[name = string("op_7929_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7929_end_mask_0 = const()[name = string("op_7929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7929_cast_fp16 = slice_by_index(begin = var_7929_begin_0, end = var_7929_end_0, end_mask = var_7929_end_mask_0, x = query_11_cast_fp16)[name = string("op_7929_cast_fp16")]; + tensor var_7933_begin_0 = const()[name = string("op_7933_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7933_end_0 = const()[name = string("op_7933_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7933_end_mask_0 = const()[name = string("op_7933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7933_cast_fp16 = slice_by_index(begin = var_7933_begin_0, end = var_7933_end_0, end_mask = var_7933_end_mask_0, x = query_11_cast_fp16)[name = string("op_7933_cast_fp16")]; + tensor var_7937_begin_0 = const()[name = string("op_7937_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7937_end_0 = const()[name = string("op_7937_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7937_end_mask_0 = const()[name = string("op_7937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7937_cast_fp16 = slice_by_index(begin = var_7937_begin_0, end = var_7937_end_0, end_mask = var_7937_end_mask_0, x = query_11_cast_fp16)[name = string("op_7937_cast_fp16")]; + tensor var_7941_begin_0 = const()[name = string("op_7941_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7941_end_0 = const()[name = string("op_7941_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_7941_end_mask_0 = const()[name = string("op_7941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7941_cast_fp16 = slice_by_index(begin = var_7941_begin_0, end = var_7941_end_0, end_mask = var_7941_end_mask_0, x = query_11_cast_fp16)[name = string("op_7941_cast_fp16")]; + tensor var_7945_begin_0 = const()[name = string("op_7945_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7945_end_0 = const()[name = string("op_7945_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_7945_end_mask_0 = const()[name = string("op_7945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7945_cast_fp16 = slice_by_index(begin = var_7945_begin_0, end = var_7945_end_0, end_mask = var_7945_end_mask_0, x = query_11_cast_fp16)[name = string("op_7945_cast_fp16")]; + tensor var_7949_begin_0 = const()[name = string("op_7949_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7949_end_0 = const()[name = string("op_7949_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_7949_end_mask_0 = const()[name = string("op_7949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7949_cast_fp16 = slice_by_index(begin = var_7949_begin_0, end = var_7949_end_0, end_mask = var_7949_end_mask_0, x = query_11_cast_fp16)[name = string("op_7949_cast_fp16")]; + tensor var_7953_begin_0 = const()[name = string("op_7953_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7953_end_0 = const()[name = string("op_7953_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_7953_end_mask_0 = const()[name = string("op_7953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7953_cast_fp16 = slice_by_index(begin = var_7953_begin_0, end = var_7953_end_0, end_mask = var_7953_end_mask_0, x = query_11_cast_fp16)[name = string("op_7953_cast_fp16")]; + tensor var_7957_begin_0 = const()[name = string("op_7957_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7957_end_0 = const()[name = string("op_7957_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_7957_end_mask_0 = const()[name = string("op_7957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7957_cast_fp16 = slice_by_index(begin = var_7957_begin_0, end = var_7957_end_0, end_mask = var_7957_end_mask_0, x = query_11_cast_fp16)[name = string("op_7957_cast_fp16")]; + tensor var_7961_begin_0 = const()[name = string("op_7961_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7961_end_0 = const()[name = string("op_7961_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_7961_end_mask_0 = const()[name = string("op_7961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7961_cast_fp16 = slice_by_index(begin = var_7961_begin_0, end = var_7961_end_0, end_mask = var_7961_end_mask_0, x = query_11_cast_fp16)[name = string("op_7961_cast_fp16")]; + tensor var_7965_begin_0 = const()[name = string("op_7965_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7965_end_0 = const()[name = string("op_7965_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_7965_end_mask_0 = const()[name = string("op_7965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7965_cast_fp16 = slice_by_index(begin = var_7965_begin_0, end = var_7965_end_0, end_mask = var_7965_end_mask_0, x = query_11_cast_fp16)[name = string("op_7965_cast_fp16")]; + tensor var_7969_begin_0 = const()[name = string("op_7969_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7969_end_0 = const()[name = string("op_7969_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_7969_end_mask_0 = const()[name = string("op_7969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7969_cast_fp16 = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = query_11_cast_fp16)[name = string("op_7969_cast_fp16")]; + tensor var_7978_begin_0 = const()[name = string("op_7978_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7978_end_0 = const()[name = string("op_7978_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7978_end_mask_0 = const()[name = string("op_7978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7978_cast_fp16 = slice_by_index(begin = var_7978_begin_0, end = var_7978_end_0, end_mask = var_7978_end_mask_0, x = var_7893_cast_fp16)[name = string("op_7978_cast_fp16")]; + tensor var_7985_begin_0 = const()[name = string("op_7985_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7985_end_0 = const()[name = string("op_7985_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7985_end_mask_0 = const()[name = string("op_7985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7985_cast_fp16 = slice_by_index(begin = var_7985_begin_0, end = var_7985_end_0, end_mask = var_7985_end_mask_0, x = var_7893_cast_fp16)[name = string("op_7985_cast_fp16")]; + tensor var_7992_begin_0 = const()[name = string("op_7992_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7992_end_0 = const()[name = string("op_7992_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7992_end_mask_0 = const()[name = string("op_7992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7992_cast_fp16 = slice_by_index(begin = var_7992_begin_0, end = var_7992_end_0, end_mask = var_7992_end_mask_0, x = var_7893_cast_fp16)[name = string("op_7992_cast_fp16")]; + tensor var_7999_begin_0 = const()[name = string("op_7999_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7999_end_0 = const()[name = string("op_7999_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7999_end_mask_0 = const()[name = string("op_7999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7999_cast_fp16 = slice_by_index(begin = var_7999_begin_0, end = var_7999_end_0, end_mask = var_7999_end_mask_0, x = var_7893_cast_fp16)[name = string("op_7999_cast_fp16")]; + tensor var_8006_begin_0 = const()[name = string("op_8006_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8006_end_0 = const()[name = string("op_8006_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8006_end_mask_0 = const()[name = string("op_8006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8006_cast_fp16 = slice_by_index(begin = var_8006_begin_0, end = var_8006_end_0, end_mask = var_8006_end_mask_0, x = var_7897_cast_fp16)[name = string("op_8006_cast_fp16")]; + tensor var_8013_begin_0 = const()[name = string("op_8013_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8013_end_0 = const()[name = string("op_8013_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8013_end_mask_0 = const()[name = string("op_8013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8013_cast_fp16 = slice_by_index(begin = var_8013_begin_0, end = var_8013_end_0, end_mask = var_8013_end_mask_0, x = var_7897_cast_fp16)[name = string("op_8013_cast_fp16")]; + tensor var_8020_begin_0 = const()[name = string("op_8020_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8020_end_0 = const()[name = string("op_8020_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8020_end_mask_0 = const()[name = string("op_8020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8020_cast_fp16 = slice_by_index(begin = var_8020_begin_0, end = var_8020_end_0, end_mask = var_8020_end_mask_0, x = var_7897_cast_fp16)[name = string("op_8020_cast_fp16")]; + tensor var_8027_begin_0 = const()[name = string("op_8027_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8027_end_0 = const()[name = string("op_8027_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8027_end_mask_0 = const()[name = string("op_8027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8027_cast_fp16 = slice_by_index(begin = var_8027_begin_0, end = var_8027_end_0, end_mask = var_8027_end_mask_0, x = var_7897_cast_fp16)[name = string("op_8027_cast_fp16")]; + tensor var_8034_begin_0 = const()[name = string("op_8034_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8034_end_0 = const()[name = string("op_8034_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8034_end_mask_0 = const()[name = string("op_8034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8034_cast_fp16 = slice_by_index(begin = var_8034_begin_0, end = var_8034_end_0, end_mask = var_8034_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8034_cast_fp16")]; + tensor var_8041_begin_0 = const()[name = string("op_8041_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8041_end_0 = const()[name = string("op_8041_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8041_end_mask_0 = const()[name = string("op_8041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8041_cast_fp16 = slice_by_index(begin = var_8041_begin_0, end = var_8041_end_0, end_mask = var_8041_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8041_cast_fp16")]; + tensor var_8048_begin_0 = const()[name = string("op_8048_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8048_end_0 = const()[name = string("op_8048_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8048_end_mask_0 = const()[name = string("op_8048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8048_cast_fp16 = slice_by_index(begin = var_8048_begin_0, end = var_8048_end_0, end_mask = var_8048_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8048_cast_fp16")]; + tensor var_8055_begin_0 = const()[name = string("op_8055_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8055_end_0 = const()[name = string("op_8055_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8055_end_mask_0 = const()[name = string("op_8055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8055_cast_fp16 = slice_by_index(begin = var_8055_begin_0, end = var_8055_end_0, end_mask = var_8055_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8055_cast_fp16")]; + tensor var_8062_begin_0 = const()[name = string("op_8062_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8062_end_0 = const()[name = string("op_8062_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8062_end_mask_0 = const()[name = string("op_8062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8062_cast_fp16 = slice_by_index(begin = var_8062_begin_0, end = var_8062_end_0, end_mask = var_8062_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8062_cast_fp16")]; + tensor var_8069_begin_0 = const()[name = string("op_8069_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8069_end_0 = const()[name = string("op_8069_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8069_end_mask_0 = const()[name = string("op_8069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8069_cast_fp16 = slice_by_index(begin = var_8069_begin_0, end = var_8069_end_0, end_mask = var_8069_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8069_cast_fp16")]; + tensor var_8076_begin_0 = const()[name = string("op_8076_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8076_end_0 = const()[name = string("op_8076_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8076_end_mask_0 = const()[name = string("op_8076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8076_cast_fp16 = slice_by_index(begin = var_8076_begin_0, end = var_8076_end_0, end_mask = var_8076_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8076_cast_fp16")]; + tensor var_8083_begin_0 = const()[name = string("op_8083_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8083_end_0 = const()[name = string("op_8083_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8083_end_mask_0 = const()[name = string("op_8083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8083_cast_fp16 = slice_by_index(begin = var_8083_begin_0, end = var_8083_end_0, end_mask = var_8083_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8083_cast_fp16")]; + tensor var_8090_begin_0 = const()[name = string("op_8090_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8090_end_0 = const()[name = string("op_8090_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8090_end_mask_0 = const()[name = string("op_8090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8090_cast_fp16 = slice_by_index(begin = var_8090_begin_0, end = var_8090_end_0, end_mask = var_8090_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8090_cast_fp16")]; + tensor var_8097_begin_0 = const()[name = string("op_8097_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8097_end_0 = const()[name = string("op_8097_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8097_end_mask_0 = const()[name = string("op_8097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8097_cast_fp16 = slice_by_index(begin = var_8097_begin_0, end = var_8097_end_0, end_mask = var_8097_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8097_cast_fp16")]; + tensor var_8104_begin_0 = const()[name = string("op_8104_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8104_end_0 = const()[name = string("op_8104_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8104_end_mask_0 = const()[name = string("op_8104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8104_cast_fp16 = slice_by_index(begin = var_8104_begin_0, end = var_8104_end_0, end_mask = var_8104_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8104_cast_fp16")]; + tensor var_8111_begin_0 = const()[name = string("op_8111_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8111_end_0 = const()[name = string("op_8111_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8111_end_mask_0 = const()[name = string("op_8111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8111_cast_fp16 = slice_by_index(begin = var_8111_begin_0, end = var_8111_end_0, end_mask = var_8111_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8111_cast_fp16")]; + tensor var_8118_begin_0 = const()[name = string("op_8118_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8118_end_0 = const()[name = string("op_8118_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8118_end_mask_0 = const()[name = string("op_8118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8118_cast_fp16 = slice_by_index(begin = var_8118_begin_0, end = var_8118_end_0, end_mask = var_8118_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8118_cast_fp16")]; + tensor var_8125_begin_0 = const()[name = string("op_8125_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8125_end_0 = const()[name = string("op_8125_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8125_end_mask_0 = const()[name = string("op_8125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8125_cast_fp16 = slice_by_index(begin = var_8125_begin_0, end = var_8125_end_0, end_mask = var_8125_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8125_cast_fp16")]; + tensor var_8132_begin_0 = const()[name = string("op_8132_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8132_end_0 = const()[name = string("op_8132_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8132_end_mask_0 = const()[name = string("op_8132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8132_cast_fp16 = slice_by_index(begin = var_8132_begin_0, end = var_8132_end_0, end_mask = var_8132_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8132_cast_fp16")]; + tensor var_8139_begin_0 = const()[name = string("op_8139_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8139_end_0 = const()[name = string("op_8139_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8139_end_mask_0 = const()[name = string("op_8139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8139_cast_fp16 = slice_by_index(begin = var_8139_begin_0, end = var_8139_end_0, end_mask = var_8139_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8139_cast_fp16")]; + tensor var_8146_begin_0 = const()[name = string("op_8146_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8146_end_0 = const()[name = string("op_8146_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8146_end_mask_0 = const()[name = string("op_8146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8146_cast_fp16 = slice_by_index(begin = var_8146_begin_0, end = var_8146_end_0, end_mask = var_8146_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8146_cast_fp16")]; + tensor var_8153_begin_0 = const()[name = string("op_8153_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8153_end_0 = const()[name = string("op_8153_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8153_end_mask_0 = const()[name = string("op_8153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8153_cast_fp16 = slice_by_index(begin = var_8153_begin_0, end = var_8153_end_0, end_mask = var_8153_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8153_cast_fp16")]; + tensor var_8160_begin_0 = const()[name = string("op_8160_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8160_end_0 = const()[name = string("op_8160_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8160_end_mask_0 = const()[name = string("op_8160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8160_cast_fp16 = slice_by_index(begin = var_8160_begin_0, end = var_8160_end_0, end_mask = var_8160_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8160_cast_fp16")]; + tensor var_8167_begin_0 = const()[name = string("op_8167_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8167_end_0 = const()[name = string("op_8167_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8167_end_mask_0 = const()[name = string("op_8167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8167_cast_fp16 = slice_by_index(begin = var_8167_begin_0, end = var_8167_end_0, end_mask = var_8167_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8167_cast_fp16")]; + tensor var_8174_begin_0 = const()[name = string("op_8174_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8174_end_0 = const()[name = string("op_8174_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8174_end_mask_0 = const()[name = string("op_8174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8174_cast_fp16 = slice_by_index(begin = var_8174_begin_0, end = var_8174_end_0, end_mask = var_8174_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8174_cast_fp16")]; + tensor var_8181_begin_0 = const()[name = string("op_8181_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8181_end_0 = const()[name = string("op_8181_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8181_end_mask_0 = const()[name = string("op_8181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8181_cast_fp16 = slice_by_index(begin = var_8181_begin_0, end = var_8181_end_0, end_mask = var_8181_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8181_cast_fp16")]; + tensor var_8188_begin_0 = const()[name = string("op_8188_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8188_end_0 = const()[name = string("op_8188_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8188_end_mask_0 = const()[name = string("op_8188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8188_cast_fp16 = slice_by_index(begin = var_8188_begin_0, end = var_8188_end_0, end_mask = var_8188_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8188_cast_fp16")]; + tensor var_8195_begin_0 = const()[name = string("op_8195_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8195_end_0 = const()[name = string("op_8195_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8195_end_mask_0 = const()[name = string("op_8195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8195_cast_fp16 = slice_by_index(begin = var_8195_begin_0, end = var_8195_end_0, end_mask = var_8195_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8195_cast_fp16")]; + tensor var_8202_begin_0 = const()[name = string("op_8202_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8202_end_0 = const()[name = string("op_8202_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8202_end_mask_0 = const()[name = string("op_8202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8202_cast_fp16 = slice_by_index(begin = var_8202_begin_0, end = var_8202_end_0, end_mask = var_8202_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8202_cast_fp16")]; + tensor var_8209_begin_0 = const()[name = string("op_8209_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8209_end_0 = const()[name = string("op_8209_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8209_end_mask_0 = const()[name = string("op_8209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8209_cast_fp16 = slice_by_index(begin = var_8209_begin_0, end = var_8209_end_0, end_mask = var_8209_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8209_cast_fp16")]; + tensor var_8216_begin_0 = const()[name = string("op_8216_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8216_end_0 = const()[name = string("op_8216_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8216_end_mask_0 = const()[name = string("op_8216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8216_cast_fp16 = slice_by_index(begin = var_8216_begin_0, end = var_8216_end_0, end_mask = var_8216_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8216_cast_fp16")]; + tensor var_8223_begin_0 = const()[name = string("op_8223_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8223_end_0 = const()[name = string("op_8223_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8223_end_mask_0 = const()[name = string("op_8223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8223_cast_fp16 = slice_by_index(begin = var_8223_begin_0, end = var_8223_end_0, end_mask = var_8223_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8223_cast_fp16")]; + tensor var_8230_begin_0 = const()[name = string("op_8230_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8230_end_0 = const()[name = string("op_8230_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8230_end_mask_0 = const()[name = string("op_8230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8230_cast_fp16 = slice_by_index(begin = var_8230_begin_0, end = var_8230_end_0, end_mask = var_8230_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8230_cast_fp16")]; + tensor var_8237_begin_0 = const()[name = string("op_8237_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8237_end_0 = const()[name = string("op_8237_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8237_end_mask_0 = const()[name = string("op_8237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8237_cast_fp16 = slice_by_index(begin = var_8237_begin_0, end = var_8237_end_0, end_mask = var_8237_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8237_cast_fp16")]; + tensor var_8244_begin_0 = const()[name = string("op_8244_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8244_end_0 = const()[name = string("op_8244_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8244_end_mask_0 = const()[name = string("op_8244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8244_cast_fp16 = slice_by_index(begin = var_8244_begin_0, end = var_8244_end_0, end_mask = var_8244_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8244_cast_fp16")]; + tensor var_8251_begin_0 = const()[name = string("op_8251_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8251_end_0 = const()[name = string("op_8251_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8251_end_mask_0 = const()[name = string("op_8251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8251_cast_fp16 = slice_by_index(begin = var_8251_begin_0, end = var_8251_end_0, end_mask = var_8251_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8251_cast_fp16")]; + tensor var_8258_begin_0 = const()[name = string("op_8258_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8258_end_0 = const()[name = string("op_8258_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8258_end_mask_0 = const()[name = string("op_8258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8258_cast_fp16 = slice_by_index(begin = var_8258_begin_0, end = var_8258_end_0, end_mask = var_8258_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8258_cast_fp16")]; + tensor var_8265_begin_0 = const()[name = string("op_8265_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8265_end_0 = const()[name = string("op_8265_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8265_end_mask_0 = const()[name = string("op_8265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8265_cast_fp16 = slice_by_index(begin = var_8265_begin_0, end = var_8265_end_0, end_mask = var_8265_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8265_cast_fp16")]; + tensor var_8272_begin_0 = const()[name = string("op_8272_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8272_end_0 = const()[name = string("op_8272_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8272_end_mask_0 = const()[name = string("op_8272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8272_cast_fp16 = slice_by_index(begin = var_8272_begin_0, end = var_8272_end_0, end_mask = var_8272_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8272_cast_fp16")]; + tensor var_8279_begin_0 = const()[name = string("op_8279_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8279_end_0 = const()[name = string("op_8279_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8279_end_mask_0 = const()[name = string("op_8279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8279_cast_fp16 = slice_by_index(begin = var_8279_begin_0, end = var_8279_end_0, end_mask = var_8279_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8279_cast_fp16")]; + tensor var_8286_begin_0 = const()[name = string("op_8286_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8286_end_0 = const()[name = string("op_8286_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8286_end_mask_0 = const()[name = string("op_8286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8286_cast_fp16 = slice_by_index(begin = var_8286_begin_0, end = var_8286_end_0, end_mask = var_8286_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8286_cast_fp16")]; + tensor var_8293_begin_0 = const()[name = string("op_8293_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8293_end_0 = const()[name = string("op_8293_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8293_end_mask_0 = const()[name = string("op_8293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8293_cast_fp16 = slice_by_index(begin = var_8293_begin_0, end = var_8293_end_0, end_mask = var_8293_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8293_cast_fp16")]; + tensor var_8300_begin_0 = const()[name = string("op_8300_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8300_end_0 = const()[name = string("op_8300_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8300_end_mask_0 = const()[name = string("op_8300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8300_cast_fp16 = slice_by_index(begin = var_8300_begin_0, end = var_8300_end_0, end_mask = var_8300_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8300_cast_fp16")]; + tensor var_8307_begin_0 = const()[name = string("op_8307_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8307_end_0 = const()[name = string("op_8307_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8307_end_mask_0 = const()[name = string("op_8307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8307_cast_fp16 = slice_by_index(begin = var_8307_begin_0, end = var_8307_end_0, end_mask = var_8307_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8307_cast_fp16")]; + tensor var_8314_begin_0 = const()[name = string("op_8314_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8314_end_0 = const()[name = string("op_8314_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8314_end_mask_0 = const()[name = string("op_8314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8314_cast_fp16 = slice_by_index(begin = var_8314_begin_0, end = var_8314_end_0, end_mask = var_8314_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8314_cast_fp16")]; + tensor var_8321_begin_0 = const()[name = string("op_8321_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8321_end_0 = const()[name = string("op_8321_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8321_end_mask_0 = const()[name = string("op_8321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8321_cast_fp16 = slice_by_index(begin = var_8321_begin_0, end = var_8321_end_0, end_mask = var_8321_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8321_cast_fp16")]; + tensor var_8328_begin_0 = const()[name = string("op_8328_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8328_end_0 = const()[name = string("op_8328_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8328_end_mask_0 = const()[name = string("op_8328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8328_cast_fp16 = slice_by_index(begin = var_8328_begin_0, end = var_8328_end_0, end_mask = var_8328_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8328_cast_fp16")]; + tensor var_8335_begin_0 = const()[name = string("op_8335_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8335_end_0 = const()[name = string("op_8335_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8335_end_mask_0 = const()[name = string("op_8335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8335_cast_fp16 = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8335_cast_fp16")]; + tensor var_8342_begin_0 = const()[name = string("op_8342_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8342_end_0 = const()[name = string("op_8342_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8342_end_mask_0 = const()[name = string("op_8342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8342_cast_fp16 = slice_by_index(begin = var_8342_begin_0, end = var_8342_end_0, end_mask = var_8342_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8342_cast_fp16")]; + tensor var_8349_begin_0 = const()[name = string("op_8349_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8349_end_0 = const()[name = string("op_8349_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8349_end_mask_0 = const()[name = string("op_8349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8349_cast_fp16 = slice_by_index(begin = var_8349_begin_0, end = var_8349_end_0, end_mask = var_8349_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8349_cast_fp16")]; + tensor var_8356_begin_0 = const()[name = string("op_8356_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8356_end_0 = const()[name = string("op_8356_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8356_end_mask_0 = const()[name = string("op_8356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8356_cast_fp16 = slice_by_index(begin = var_8356_begin_0, end = var_8356_end_0, end_mask = var_8356_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8356_cast_fp16")]; + tensor var_8363_begin_0 = const()[name = string("op_8363_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8363_end_0 = const()[name = string("op_8363_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8363_end_mask_0 = const()[name = string("op_8363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8363_cast_fp16 = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8363_cast_fp16")]; + tensor var_8370_begin_0 = const()[name = string("op_8370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8370_end_0 = const()[name = string("op_8370_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8370_end_mask_0 = const()[name = string("op_8370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8370_cast_fp16 = slice_by_index(begin = var_8370_begin_0, end = var_8370_end_0, end_mask = var_8370_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8370_cast_fp16")]; + tensor var_8377_begin_0 = const()[name = string("op_8377_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8377_end_0 = const()[name = string("op_8377_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8377_end_mask_0 = const()[name = string("op_8377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8377_cast_fp16 = slice_by_index(begin = var_8377_begin_0, end = var_8377_end_0, end_mask = var_8377_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8377_cast_fp16")]; + tensor var_8384_begin_0 = const()[name = string("op_8384_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8384_end_0 = const()[name = string("op_8384_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8384_end_mask_0 = const()[name = string("op_8384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8384_cast_fp16 = slice_by_index(begin = var_8384_begin_0, end = var_8384_end_0, end_mask = var_8384_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8384_cast_fp16")]; + tensor var_8391_begin_0 = const()[name = string("op_8391_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8391_end_0 = const()[name = string("op_8391_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8391_end_mask_0 = const()[name = string("op_8391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8391_cast_fp16 = slice_by_index(begin = var_8391_begin_0, end = var_8391_end_0, end_mask = var_8391_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8391_cast_fp16")]; + tensor var_8398_begin_0 = const()[name = string("op_8398_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8398_end_0 = const()[name = string("op_8398_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8398_end_mask_0 = const()[name = string("op_8398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8398_cast_fp16 = slice_by_index(begin = var_8398_begin_0, end = var_8398_end_0, end_mask = var_8398_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8398_cast_fp16")]; + tensor var_8405_begin_0 = const()[name = string("op_8405_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8405_end_0 = const()[name = string("op_8405_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8405_end_mask_0 = const()[name = string("op_8405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8405_cast_fp16 = slice_by_index(begin = var_8405_begin_0, end = var_8405_end_0, end_mask = var_8405_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8405_cast_fp16")]; + tensor var_8412_begin_0 = const()[name = string("op_8412_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8412_end_0 = const()[name = string("op_8412_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8412_end_mask_0 = const()[name = string("op_8412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8412_cast_fp16 = slice_by_index(begin = var_8412_begin_0, end = var_8412_end_0, end_mask = var_8412_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8412_cast_fp16")]; + tensor var_8419_begin_0 = const()[name = string("op_8419_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8419_end_0 = const()[name = string("op_8419_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8419_end_mask_0 = const()[name = string("op_8419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8419_cast_fp16 = slice_by_index(begin = var_8419_begin_0, end = var_8419_end_0, end_mask = var_8419_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8419_cast_fp16")]; + tensor var_8426_begin_0 = const()[name = string("op_8426_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8426_end_0 = const()[name = string("op_8426_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8426_end_mask_0 = const()[name = string("op_8426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8426_cast_fp16 = slice_by_index(begin = var_8426_begin_0, end = var_8426_end_0, end_mask = var_8426_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8426_cast_fp16")]; + tensor var_8433_begin_0 = const()[name = string("op_8433_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8433_end_0 = const()[name = string("op_8433_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8433_end_mask_0 = const()[name = string("op_8433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8433_cast_fp16 = slice_by_index(begin = var_8433_begin_0, end = var_8433_end_0, end_mask = var_8433_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8433_cast_fp16")]; + tensor var_8440_begin_0 = const()[name = string("op_8440_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8440_end_0 = const()[name = string("op_8440_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8440_end_mask_0 = const()[name = string("op_8440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8440_cast_fp16 = slice_by_index(begin = var_8440_begin_0, end = var_8440_end_0, end_mask = var_8440_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8440_cast_fp16")]; + tensor var_8447_begin_0 = const()[name = string("op_8447_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8447_end_0 = const()[name = string("op_8447_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8447_end_mask_0 = const()[name = string("op_8447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8447_cast_fp16 = slice_by_index(begin = var_8447_begin_0, end = var_8447_end_0, end_mask = var_8447_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8447_cast_fp16")]; + tensor var_8454_begin_0 = const()[name = string("op_8454_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8454_end_0 = const()[name = string("op_8454_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8454_end_mask_0 = const()[name = string("op_8454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8454_cast_fp16 = slice_by_index(begin = var_8454_begin_0, end = var_8454_end_0, end_mask = var_8454_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8454_cast_fp16")]; + tensor var_8461_begin_0 = const()[name = string("op_8461_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8461_end_0 = const()[name = string("op_8461_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8461_end_mask_0 = const()[name = string("op_8461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8461_cast_fp16 = slice_by_index(begin = var_8461_begin_0, end = var_8461_end_0, end_mask = var_8461_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8461_cast_fp16")]; + tensor var_8468_begin_0 = const()[name = string("op_8468_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8468_end_0 = const()[name = string("op_8468_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8468_end_mask_0 = const()[name = string("op_8468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8468_cast_fp16 = slice_by_index(begin = var_8468_begin_0, end = var_8468_end_0, end_mask = var_8468_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8468_cast_fp16")]; + tensor var_8475_begin_0 = const()[name = string("op_8475_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8475_end_0 = const()[name = string("op_8475_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8475_end_mask_0 = const()[name = string("op_8475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8475_cast_fp16 = slice_by_index(begin = var_8475_begin_0, end = var_8475_end_0, end_mask = var_8475_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8475_cast_fp16")]; + tensor var_8482_begin_0 = const()[name = string("op_8482_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8482_end_0 = const()[name = string("op_8482_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8482_end_mask_0 = const()[name = string("op_8482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8482_cast_fp16 = slice_by_index(begin = var_8482_begin_0, end = var_8482_end_0, end_mask = var_8482_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8482_cast_fp16")]; + tensor var_8489_begin_0 = const()[name = string("op_8489_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8489_end_0 = const()[name = string("op_8489_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8489_end_mask_0 = const()[name = string("op_8489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8489_cast_fp16 = slice_by_index(begin = var_8489_begin_0, end = var_8489_end_0, end_mask = var_8489_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8489_cast_fp16")]; + tensor var_8496_begin_0 = const()[name = string("op_8496_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8496_end_0 = const()[name = string("op_8496_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8496_end_mask_0 = const()[name = string("op_8496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8496_cast_fp16 = slice_by_index(begin = var_8496_begin_0, end = var_8496_end_0, end_mask = var_8496_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8496_cast_fp16")]; + tensor var_8503_begin_0 = const()[name = string("op_8503_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8503_end_0 = const()[name = string("op_8503_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8503_end_mask_0 = const()[name = string("op_8503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8503_cast_fp16 = slice_by_index(begin = var_8503_begin_0, end = var_8503_end_0, end_mask = var_8503_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8503_cast_fp16")]; + tensor var_8510_begin_0 = const()[name = string("op_8510_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8510_end_0 = const()[name = string("op_8510_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8510_end_mask_0 = const()[name = string("op_8510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8510_cast_fp16 = slice_by_index(begin = var_8510_begin_0, end = var_8510_end_0, end_mask = var_8510_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8510_cast_fp16")]; + tensor var_8517_begin_0 = const()[name = string("op_8517_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8517_end_0 = const()[name = string("op_8517_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8517_end_mask_0 = const()[name = string("op_8517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8517_cast_fp16 = slice_by_index(begin = var_8517_begin_0, end = var_8517_end_0, end_mask = var_8517_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8517_cast_fp16")]; + tensor var_8524_begin_0 = const()[name = string("op_8524_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8524_end_0 = const()[name = string("op_8524_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8524_end_mask_0 = const()[name = string("op_8524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8524_cast_fp16 = slice_by_index(begin = var_8524_begin_0, end = var_8524_end_0, end_mask = var_8524_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8524_cast_fp16")]; + tensor var_8531_begin_0 = const()[name = string("op_8531_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8531_end_0 = const()[name = string("op_8531_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8531_end_mask_0 = const()[name = string("op_8531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8531_cast_fp16 = slice_by_index(begin = var_8531_begin_0, end = var_8531_end_0, end_mask = var_8531_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8531_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8536_begin_0 = const()[name = string("op_8536_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8536_end_0 = const()[name = string("op_8536_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_8536_end_mask_0 = const()[name = string("op_8536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = string("transpose_26")]; + tensor var_8536_cast_fp16 = slice_by_index(begin = var_8536_begin_0, end = var_8536_end_0, end_mask = var_8536_end_mask_0, x = k_11_cast_fp16)[name = string("op_8536_cast_fp16")]; + tensor var_8540_begin_0 = const()[name = string("op_8540_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8540_end_0 = const()[name = string("op_8540_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_8540_end_mask_0 = const()[name = string("op_8540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8540_cast_fp16 = slice_by_index(begin = var_8540_begin_0, end = var_8540_end_0, end_mask = var_8540_end_mask_0, x = k_11_cast_fp16)[name = string("op_8540_cast_fp16")]; + tensor var_8544_begin_0 = const()[name = string("op_8544_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8544_end_0 = const()[name = string("op_8544_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_8544_end_mask_0 = const()[name = string("op_8544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8544_cast_fp16 = slice_by_index(begin = var_8544_begin_0, end = var_8544_end_0, end_mask = var_8544_end_mask_0, x = k_11_cast_fp16)[name = string("op_8544_cast_fp16")]; + tensor var_8548_begin_0 = const()[name = string("op_8548_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8548_end_0 = const()[name = string("op_8548_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_8548_end_mask_0 = const()[name = string("op_8548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8548_cast_fp16 = slice_by_index(begin = var_8548_begin_0, end = var_8548_end_0, end_mask = var_8548_end_mask_0, x = k_11_cast_fp16)[name = string("op_8548_cast_fp16")]; + tensor var_8552_begin_0 = const()[name = string("op_8552_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8552_end_0 = const()[name = string("op_8552_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_8552_end_mask_0 = const()[name = string("op_8552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8552_cast_fp16 = slice_by_index(begin = var_8552_begin_0, end = var_8552_end_0, end_mask = var_8552_end_mask_0, x = k_11_cast_fp16)[name = string("op_8552_cast_fp16")]; + tensor var_8556_begin_0 = const()[name = string("op_8556_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8556_end_0 = const()[name = string("op_8556_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_8556_end_mask_0 = const()[name = string("op_8556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8556_cast_fp16 = slice_by_index(begin = var_8556_begin_0, end = var_8556_end_0, end_mask = var_8556_end_mask_0, x = k_11_cast_fp16)[name = string("op_8556_cast_fp16")]; + tensor var_8560_begin_0 = const()[name = string("op_8560_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8560_end_0 = const()[name = string("op_8560_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_8560_end_mask_0 = const()[name = string("op_8560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8560_cast_fp16 = slice_by_index(begin = var_8560_begin_0, end = var_8560_end_0, end_mask = var_8560_end_mask_0, x = k_11_cast_fp16)[name = string("op_8560_cast_fp16")]; + tensor var_8564_begin_0 = const()[name = string("op_8564_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8564_end_0 = const()[name = string("op_8564_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_8564_end_mask_0 = const()[name = string("op_8564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8564_cast_fp16 = slice_by_index(begin = var_8564_begin_0, end = var_8564_end_0, end_mask = var_8564_end_mask_0, x = k_11_cast_fp16)[name = string("op_8564_cast_fp16")]; + tensor var_8568_begin_0 = const()[name = string("op_8568_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8568_end_0 = const()[name = string("op_8568_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_8568_end_mask_0 = const()[name = string("op_8568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8568_cast_fp16 = slice_by_index(begin = var_8568_begin_0, end = var_8568_end_0, end_mask = var_8568_end_mask_0, x = k_11_cast_fp16)[name = string("op_8568_cast_fp16")]; + tensor var_8572_begin_0 = const()[name = string("op_8572_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8572_end_0 = const()[name = string("op_8572_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_8572_end_mask_0 = const()[name = string("op_8572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8572_cast_fp16 = slice_by_index(begin = var_8572_begin_0, end = var_8572_end_0, end_mask = var_8572_end_mask_0, x = k_11_cast_fp16)[name = string("op_8572_cast_fp16")]; + tensor var_8576_begin_0 = const()[name = string("op_8576_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8576_end_0 = const()[name = string("op_8576_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_8576_end_mask_0 = const()[name = string("op_8576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8576_cast_fp16 = slice_by_index(begin = var_8576_begin_0, end = var_8576_end_0, end_mask = var_8576_end_mask_0, x = k_11_cast_fp16)[name = string("op_8576_cast_fp16")]; + tensor var_8580_begin_0 = const()[name = string("op_8580_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8580_end_0 = const()[name = string("op_8580_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_8580_end_mask_0 = const()[name = string("op_8580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8580_cast_fp16 = slice_by_index(begin = var_8580_begin_0, end = var_8580_end_0, end_mask = var_8580_end_mask_0, x = k_11_cast_fp16)[name = string("op_8580_cast_fp16")]; + tensor var_8584_begin_0 = const()[name = string("op_8584_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8584_end_0 = const()[name = string("op_8584_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_8584_end_mask_0 = const()[name = string("op_8584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8584_cast_fp16 = slice_by_index(begin = var_8584_begin_0, end = var_8584_end_0, end_mask = var_8584_end_mask_0, x = k_11_cast_fp16)[name = string("op_8584_cast_fp16")]; + tensor var_8588_begin_0 = const()[name = string("op_8588_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8588_end_0 = const()[name = string("op_8588_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_8588_end_mask_0 = const()[name = string("op_8588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8588_cast_fp16 = slice_by_index(begin = var_8588_begin_0, end = var_8588_end_0, end_mask = var_8588_end_mask_0, x = k_11_cast_fp16)[name = string("op_8588_cast_fp16")]; + tensor var_8592_begin_0 = const()[name = string("op_8592_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8592_end_0 = const()[name = string("op_8592_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_8592_end_mask_0 = const()[name = string("op_8592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8592_cast_fp16 = slice_by_index(begin = var_8592_begin_0, end = var_8592_end_0, end_mask = var_8592_end_mask_0, x = k_11_cast_fp16)[name = string("op_8592_cast_fp16")]; + tensor var_8596_begin_0 = const()[name = string("op_8596_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8596_end_0 = const()[name = string("op_8596_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_8596_end_mask_0 = const()[name = string("op_8596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8596_cast_fp16 = slice_by_index(begin = var_8596_begin_0, end = var_8596_end_0, end_mask = var_8596_end_mask_0, x = k_11_cast_fp16)[name = string("op_8596_cast_fp16")]; + tensor var_8600_begin_0 = const()[name = string("op_8600_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8600_end_0 = const()[name = string("op_8600_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_8600_end_mask_0 = const()[name = string("op_8600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8600_cast_fp16 = slice_by_index(begin = var_8600_begin_0, end = var_8600_end_0, end_mask = var_8600_end_mask_0, x = k_11_cast_fp16)[name = string("op_8600_cast_fp16")]; + tensor var_8604_begin_0 = const()[name = string("op_8604_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8604_end_0 = const()[name = string("op_8604_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_8604_end_mask_0 = const()[name = string("op_8604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8604_cast_fp16 = slice_by_index(begin = var_8604_begin_0, end = var_8604_end_0, end_mask = var_8604_end_mask_0, x = k_11_cast_fp16)[name = string("op_8604_cast_fp16")]; + tensor var_8608_begin_0 = const()[name = string("op_8608_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8608_end_0 = const()[name = string("op_8608_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_8608_end_mask_0 = const()[name = string("op_8608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8608_cast_fp16 = slice_by_index(begin = var_8608_begin_0, end = var_8608_end_0, end_mask = var_8608_end_mask_0, x = k_11_cast_fp16)[name = string("op_8608_cast_fp16")]; + tensor var_8612_begin_0 = const()[name = string("op_8612_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8612_end_0 = const()[name = string("op_8612_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_8612_end_mask_0 = const()[name = string("op_8612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8612_cast_fp16 = slice_by_index(begin = var_8612_begin_0, end = var_8612_end_0, end_mask = var_8612_end_mask_0, x = k_11_cast_fp16)[name = string("op_8612_cast_fp16")]; + tensor var_8614_begin_0 = const()[name = string("op_8614_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8614_end_0 = const()[name = string("op_8614_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8614_end_mask_0 = const()[name = string("op_8614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8614_cast_fp16 = slice_by_index(begin = var_8614_begin_0, end = var_8614_end_0, end_mask = var_8614_end_mask_0, x = value_11_cast_fp16)[name = string("op_8614_cast_fp16")]; + tensor var_8618_begin_0 = const()[name = string("op_8618_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8618_end_0 = const()[name = string("op_8618_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8618_end_mask_0 = const()[name = string("op_8618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8618_cast_fp16 = slice_by_index(begin = var_8618_begin_0, end = var_8618_end_0, end_mask = var_8618_end_mask_0, x = value_11_cast_fp16)[name = string("op_8618_cast_fp16")]; + tensor var_8622_begin_0 = const()[name = string("op_8622_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8622_end_0 = const()[name = string("op_8622_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8622_end_mask_0 = const()[name = string("op_8622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8622_cast_fp16 = slice_by_index(begin = var_8622_begin_0, end = var_8622_end_0, end_mask = var_8622_end_mask_0, x = value_11_cast_fp16)[name = string("op_8622_cast_fp16")]; + tensor var_8626_begin_0 = const()[name = string("op_8626_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8626_end_0 = const()[name = string("op_8626_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8626_end_mask_0 = const()[name = string("op_8626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8626_cast_fp16 = slice_by_index(begin = var_8626_begin_0, end = var_8626_end_0, end_mask = var_8626_end_mask_0, x = value_11_cast_fp16)[name = string("op_8626_cast_fp16")]; + tensor var_8630_begin_0 = const()[name = string("op_8630_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8630_end_0 = const()[name = string("op_8630_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8630_end_mask_0 = const()[name = string("op_8630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8630_cast_fp16 = slice_by_index(begin = var_8630_begin_0, end = var_8630_end_0, end_mask = var_8630_end_mask_0, x = value_11_cast_fp16)[name = string("op_8630_cast_fp16")]; + tensor var_8634_begin_0 = const()[name = string("op_8634_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8634_end_0 = const()[name = string("op_8634_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8634_end_mask_0 = const()[name = string("op_8634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8634_cast_fp16 = slice_by_index(begin = var_8634_begin_0, end = var_8634_end_0, end_mask = var_8634_end_mask_0, x = value_11_cast_fp16)[name = string("op_8634_cast_fp16")]; + tensor var_8638_begin_0 = const()[name = string("op_8638_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8638_end_0 = const()[name = string("op_8638_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8638_end_mask_0 = const()[name = string("op_8638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8638_cast_fp16 = slice_by_index(begin = var_8638_begin_0, end = var_8638_end_0, end_mask = var_8638_end_mask_0, x = value_11_cast_fp16)[name = string("op_8638_cast_fp16")]; + tensor var_8642_begin_0 = const()[name = string("op_8642_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8642_end_0 = const()[name = string("op_8642_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8642_end_mask_0 = const()[name = string("op_8642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8642_cast_fp16 = slice_by_index(begin = var_8642_begin_0, end = var_8642_end_0, end_mask = var_8642_end_mask_0, x = value_11_cast_fp16)[name = string("op_8642_cast_fp16")]; + tensor var_8646_begin_0 = const()[name = string("op_8646_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8646_end_0 = const()[name = string("op_8646_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8646_end_mask_0 = const()[name = string("op_8646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8646_cast_fp16 = slice_by_index(begin = var_8646_begin_0, end = var_8646_end_0, end_mask = var_8646_end_mask_0, x = value_11_cast_fp16)[name = string("op_8646_cast_fp16")]; + tensor var_8650_begin_0 = const()[name = string("op_8650_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8650_end_0 = const()[name = string("op_8650_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8650_end_mask_0 = const()[name = string("op_8650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8650_cast_fp16 = slice_by_index(begin = var_8650_begin_0, end = var_8650_end_0, end_mask = var_8650_end_mask_0, x = value_11_cast_fp16)[name = string("op_8650_cast_fp16")]; + tensor var_8654_begin_0 = const()[name = string("op_8654_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8654_end_0 = const()[name = string("op_8654_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8654_end_mask_0 = const()[name = string("op_8654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8654_cast_fp16 = slice_by_index(begin = var_8654_begin_0, end = var_8654_end_0, end_mask = var_8654_end_mask_0, x = value_11_cast_fp16)[name = string("op_8654_cast_fp16")]; + tensor var_8658_begin_0 = const()[name = string("op_8658_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8658_end_0 = const()[name = string("op_8658_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8658_end_mask_0 = const()[name = string("op_8658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8658_cast_fp16 = slice_by_index(begin = var_8658_begin_0, end = var_8658_end_0, end_mask = var_8658_end_mask_0, x = value_11_cast_fp16)[name = string("op_8658_cast_fp16")]; + tensor var_8662_begin_0 = const()[name = string("op_8662_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8662_end_0 = const()[name = string("op_8662_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_8662_end_mask_0 = const()[name = string("op_8662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8662_cast_fp16 = slice_by_index(begin = var_8662_begin_0, end = var_8662_end_0, end_mask = var_8662_end_mask_0, x = value_11_cast_fp16)[name = string("op_8662_cast_fp16")]; + tensor var_8666_begin_0 = const()[name = string("op_8666_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8666_end_0 = const()[name = string("op_8666_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_8666_end_mask_0 = const()[name = string("op_8666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8666_cast_fp16 = slice_by_index(begin = var_8666_begin_0, end = var_8666_end_0, end_mask = var_8666_end_mask_0, x = value_11_cast_fp16)[name = string("op_8666_cast_fp16")]; + tensor var_8670_begin_0 = const()[name = string("op_8670_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8670_end_0 = const()[name = string("op_8670_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_8670_end_mask_0 = const()[name = string("op_8670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8670_cast_fp16 = slice_by_index(begin = var_8670_begin_0, end = var_8670_end_0, end_mask = var_8670_end_mask_0, x = value_11_cast_fp16)[name = string("op_8670_cast_fp16")]; + tensor var_8674_begin_0 = const()[name = string("op_8674_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8674_end_0 = const()[name = string("op_8674_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_8674_end_mask_0 = const()[name = string("op_8674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8674_cast_fp16 = slice_by_index(begin = var_8674_begin_0, end = var_8674_end_0, end_mask = var_8674_end_mask_0, x = value_11_cast_fp16)[name = string("op_8674_cast_fp16")]; + tensor var_8678_begin_0 = const()[name = string("op_8678_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8678_end_0 = const()[name = string("op_8678_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_8678_end_mask_0 = const()[name = string("op_8678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8678_cast_fp16 = slice_by_index(begin = var_8678_begin_0, end = var_8678_end_0, end_mask = var_8678_end_mask_0, x = value_11_cast_fp16)[name = string("op_8678_cast_fp16")]; + tensor var_8682_begin_0 = const()[name = string("op_8682_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8682_end_0 = const()[name = string("op_8682_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_8682_end_mask_0 = const()[name = string("op_8682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8682_cast_fp16 = slice_by_index(begin = var_8682_begin_0, end = var_8682_end_0, end_mask = var_8682_end_mask_0, x = value_11_cast_fp16)[name = string("op_8682_cast_fp16")]; + tensor var_8686_begin_0 = const()[name = string("op_8686_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8686_end_0 = const()[name = string("op_8686_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_8686_end_mask_0 = const()[name = string("op_8686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8686_cast_fp16 = slice_by_index(begin = var_8686_begin_0, end = var_8686_end_0, end_mask = var_8686_end_mask_0, x = value_11_cast_fp16)[name = string("op_8686_cast_fp16")]; + tensor var_8690_begin_0 = const()[name = string("op_8690_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8690_end_0 = const()[name = string("op_8690_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_8690_end_mask_0 = const()[name = string("op_8690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8690_cast_fp16 = slice_by_index(begin = var_8690_begin_0, end = var_8690_end_0, end_mask = var_8690_end_mask_0, x = value_11_cast_fp16)[name = string("op_8690_cast_fp16")]; + string _SplitHeadsQ__mh_w_801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_801_equation_0, values = (var_8536_cast_fp16, var_7978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_801_cast_fp16")]; + string _SplitHeadsQ__mh_w_803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_803_equation_0, values = (var_8536_cast_fp16, var_7985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_803_cast_fp16")]; + string _SplitHeadsQ__mh_w_805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_805_equation_0, values = (var_8536_cast_fp16, var_7992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_805_cast_fp16")]; + string _SplitHeadsQ__mh_w_807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_807_equation_0, values = (var_8536_cast_fp16, var_7999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_807_cast_fp16")]; + string _SplitHeadsQ__mh_w_809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_809_equation_0, values = (var_8540_cast_fp16, var_8006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_809_cast_fp16")]; + string _SplitHeadsQ__mh_w_811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_811_equation_0, values = (var_8540_cast_fp16, var_8013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_811_cast_fp16")]; + string _SplitHeadsQ__mh_w_813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_813_equation_0, values = (var_8540_cast_fp16, var_8020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_813_cast_fp16")]; + string _SplitHeadsQ__mh_w_815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_815_equation_0, values = (var_8540_cast_fp16, var_8027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_815_cast_fp16")]; + string _SplitHeadsQ__mh_w_817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_817_equation_0, values = (var_8544_cast_fp16, var_8034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_817_cast_fp16")]; + string _SplitHeadsQ__mh_w_819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_819_equation_0, values = (var_8544_cast_fp16, var_8041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_819_cast_fp16")]; + string _SplitHeadsQ__mh_w_821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_821_equation_0, values = (var_8544_cast_fp16, var_8048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_821_cast_fp16")]; + string _SplitHeadsQ__mh_w_823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_823_equation_0, values = (var_8544_cast_fp16, var_8055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_823_cast_fp16")]; + string _SplitHeadsQ__mh_w_825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_825_equation_0, values = (var_8548_cast_fp16, var_8062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_825_cast_fp16")]; + string _SplitHeadsQ__mh_w_827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_827_equation_0, values = (var_8548_cast_fp16, var_8069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_827_cast_fp16")]; + string _SplitHeadsQ__mh_w_829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_829_equation_0, values = (var_8548_cast_fp16, var_8076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_829_cast_fp16")]; + string _SplitHeadsQ__mh_w_831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_831_equation_0, values = (var_8548_cast_fp16, var_8083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_831_cast_fp16")]; + string _SplitHeadsQ__mh_w_833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_833_equation_0, values = (var_8552_cast_fp16, var_8090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_833_cast_fp16")]; + string _SplitHeadsQ__mh_w_835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_835_equation_0, values = (var_8552_cast_fp16, var_8097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_835_cast_fp16")]; + string _SplitHeadsQ__mh_w_837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_837_equation_0, values = (var_8552_cast_fp16, var_8104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_837_cast_fp16")]; + string _SplitHeadsQ__mh_w_839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_839_equation_0, values = (var_8552_cast_fp16, var_8111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_839_cast_fp16")]; + string _SplitHeadsQ__mh_w_841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_841_equation_0, values = (var_8556_cast_fp16, var_8118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_841_cast_fp16")]; + string _SplitHeadsQ__mh_w_843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_843_equation_0, values = (var_8556_cast_fp16, var_8125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_843_cast_fp16")]; + string _SplitHeadsQ__mh_w_845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_845_equation_0, values = (var_8556_cast_fp16, var_8132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_845_cast_fp16")]; + string _SplitHeadsQ__mh_w_847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_847_equation_0, values = (var_8556_cast_fp16, var_8139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_847_cast_fp16")]; + string _SplitHeadsQ__mh_w_849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_849_equation_0, values = (var_8560_cast_fp16, var_8146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_849_cast_fp16")]; + string _SplitHeadsQ__mh_w_851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_851_equation_0, values = (var_8560_cast_fp16, var_8153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_851_cast_fp16")]; + string _SplitHeadsQ__mh_w_853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_853_equation_0, values = (var_8560_cast_fp16, var_8160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_853_cast_fp16")]; + string _SplitHeadsQ__mh_w_855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_855_equation_0, values = (var_8560_cast_fp16, var_8167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_855_cast_fp16")]; + string _SplitHeadsQ__mh_w_857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_857_equation_0, values = (var_8564_cast_fp16, var_8174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_857_cast_fp16")]; + string _SplitHeadsQ__mh_w_859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_859_equation_0, values = (var_8564_cast_fp16, var_8181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_859_cast_fp16")]; + string _SplitHeadsQ__mh_w_861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_861_equation_0, values = (var_8564_cast_fp16, var_8188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_861_cast_fp16")]; + string _SplitHeadsQ__mh_w_863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_863_equation_0, values = (var_8564_cast_fp16, var_8195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_863_cast_fp16")]; + string _SplitHeadsQ__mh_w_865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_865_equation_0, values = (var_8568_cast_fp16, var_8202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_865_cast_fp16")]; + string _SplitHeadsQ__mh_w_867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_867_equation_0, values = (var_8568_cast_fp16, var_8209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_867_cast_fp16")]; + string _SplitHeadsQ__mh_w_869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_869_equation_0, values = (var_8568_cast_fp16, var_8216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_869_cast_fp16")]; + string _SplitHeadsQ__mh_w_871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_871_equation_0, values = (var_8568_cast_fp16, var_8223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_871_cast_fp16")]; + string _SplitHeadsQ__mh_w_873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_873_equation_0, values = (var_8572_cast_fp16, var_8230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_873_cast_fp16")]; + string _SplitHeadsQ__mh_w_875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_875_equation_0, values = (var_8572_cast_fp16, var_8237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_875_cast_fp16")]; + string _SplitHeadsQ__mh_w_877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_877_equation_0, values = (var_8572_cast_fp16, var_8244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_877_cast_fp16")]; + string _SplitHeadsQ__mh_w_879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_879_equation_0, values = (var_8572_cast_fp16, var_8251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_879_cast_fp16")]; + string _SplitHeadsQ__mh_w_881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_881_equation_0, values = (var_8576_cast_fp16, var_8258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_881_cast_fp16")]; + string _SplitHeadsQ__mh_w_883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_883_equation_0, values = (var_8576_cast_fp16, var_8265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_883_cast_fp16")]; + string _SplitHeadsQ__mh_w_885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_885_equation_0, values = (var_8576_cast_fp16, var_8272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_885_cast_fp16")]; + string _SplitHeadsQ__mh_w_887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_887_equation_0, values = (var_8576_cast_fp16, var_8279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_887_cast_fp16")]; + string _SplitHeadsQ__mh_w_889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_889_equation_0, values = (var_8580_cast_fp16, var_8286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_889_cast_fp16")]; + string _SplitHeadsQ__mh_w_891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_891_equation_0, values = (var_8580_cast_fp16, var_8293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_891_cast_fp16")]; + string _SplitHeadsQ__mh_w_893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_893_equation_0, values = (var_8580_cast_fp16, var_8300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_893_cast_fp16")]; + string _SplitHeadsQ__mh_w_895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_895_equation_0, values = (var_8580_cast_fp16, var_8307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_895_cast_fp16")]; + string _SplitHeadsQ__mh_w_897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_897_equation_0, values = (var_8584_cast_fp16, var_8314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_897_cast_fp16")]; + string _SplitHeadsQ__mh_w_899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_899_equation_0, values = (var_8584_cast_fp16, var_8321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_899_cast_fp16")]; + string _SplitHeadsQ__mh_w_901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_901_equation_0, values = (var_8584_cast_fp16, var_8328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_901_cast_fp16")]; + string _SplitHeadsQ__mh_w_903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_903_equation_0, values = (var_8584_cast_fp16, var_8335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_903_cast_fp16")]; + string _SplitHeadsQ__mh_w_905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_905_equation_0, values = (var_8588_cast_fp16, var_8342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_905_cast_fp16")]; + string _SplitHeadsQ__mh_w_907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_907_equation_0, values = (var_8588_cast_fp16, var_8349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_907_cast_fp16")]; + string _SplitHeadsQ__mh_w_909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_909_equation_0, values = (var_8588_cast_fp16, var_8356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_909_cast_fp16")]; + string _SplitHeadsQ__mh_w_911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_911_equation_0, values = (var_8588_cast_fp16, var_8363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_911_cast_fp16")]; + string _SplitHeadsQ__mh_w_913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_913_equation_0, values = (var_8592_cast_fp16, var_8370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_913_cast_fp16")]; + string _SplitHeadsQ__mh_w_915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_915_equation_0, values = (var_8592_cast_fp16, var_8377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_915_cast_fp16")]; + string _SplitHeadsQ__mh_w_917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_917_equation_0, values = (var_8592_cast_fp16, var_8384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_917_cast_fp16")]; + string _SplitHeadsQ__mh_w_919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_919_equation_0, values = (var_8592_cast_fp16, var_8391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_919_cast_fp16")]; + string _SplitHeadsQ__mh_w_921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_921_equation_0, values = (var_8596_cast_fp16, var_8398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_921_cast_fp16")]; + string _SplitHeadsQ__mh_w_923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_923_equation_0, values = (var_8596_cast_fp16, var_8405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_923_cast_fp16")]; + string _SplitHeadsQ__mh_w_925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_925_equation_0, values = (var_8596_cast_fp16, var_8412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_925_cast_fp16")]; + string _SplitHeadsQ__mh_w_927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_927_equation_0, values = (var_8596_cast_fp16, var_8419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_927_cast_fp16")]; + string _SplitHeadsQ__mh_w_929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_929_equation_0, values = (var_8600_cast_fp16, var_8426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_929_cast_fp16")]; + string _SplitHeadsQ__mh_w_931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_931_equation_0, values = (var_8600_cast_fp16, var_8433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_931_cast_fp16")]; + string _SplitHeadsQ__mh_w_933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_933_equation_0, values = (var_8600_cast_fp16, var_8440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_933_cast_fp16")]; + string _SplitHeadsQ__mh_w_935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_935_equation_0, values = (var_8600_cast_fp16, var_8447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_935_cast_fp16")]; + string _SplitHeadsQ__mh_w_937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_937_equation_0, values = (var_8604_cast_fp16, var_8454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_937_cast_fp16")]; + string _SplitHeadsQ__mh_w_939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_939_equation_0, values = (var_8604_cast_fp16, var_8461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_939_cast_fp16")]; + string _SplitHeadsQ__mh_w_941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_941_equation_0, values = (var_8604_cast_fp16, var_8468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_941_cast_fp16")]; + string _SplitHeadsQ__mh_w_943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_943_equation_0, values = (var_8604_cast_fp16, var_8475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_943_cast_fp16")]; + string _SplitHeadsQ__mh_w_945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_945_equation_0, values = (var_8608_cast_fp16, var_8482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_945_cast_fp16")]; + string _SplitHeadsQ__mh_w_947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_947_equation_0, values = (var_8608_cast_fp16, var_8489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_947_cast_fp16")]; + string _SplitHeadsQ__mh_w_949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_949_equation_0, values = (var_8608_cast_fp16, var_8496_cast_fp16))[name = string("_SplitHeadsQ__mh_w_949_cast_fp16")]; + string _SplitHeadsQ__mh_w_951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_951_equation_0, values = (var_8608_cast_fp16, var_8503_cast_fp16))[name = string("_SplitHeadsQ__mh_w_951_cast_fp16")]; + string _SplitHeadsQ__mh_w_953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_953_equation_0, values = (var_8612_cast_fp16, var_8510_cast_fp16))[name = string("_SplitHeadsQ__mh_w_953_cast_fp16")]; + string _SplitHeadsQ__mh_w_955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_955_equation_0, values = (var_8612_cast_fp16, var_8517_cast_fp16))[name = string("_SplitHeadsQ__mh_w_955_cast_fp16")]; + string _SplitHeadsQ__mh_w_957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_957_equation_0, values = (var_8612_cast_fp16, var_8524_cast_fp16))[name = string("_SplitHeadsQ__mh_w_957_cast_fp16")]; + string _SplitHeadsQ__mh_w_959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_959_equation_0, values = (var_8612_cast_fp16, var_8531_cast_fp16))[name = string("_SplitHeadsQ__mh_w_959_cast_fp16")]; + fp16 var_8853_to_fp16 = const()[name = string("op_8853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_801_cast_fp16, y = var_8853_to_fp16)[name = string("aw_chunk_801_cast_fp16")]; + fp16 var_8855_to_fp16 = const()[name = string("op_8855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_803_cast_fp16, y = var_8855_to_fp16)[name = string("aw_chunk_803_cast_fp16")]; + fp16 var_8857_to_fp16 = const()[name = string("op_8857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_805_cast_fp16, y = var_8857_to_fp16)[name = string("aw_chunk_805_cast_fp16")]; + fp16 var_8859_to_fp16 = const()[name = string("op_8859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_807_cast_fp16, y = var_8859_to_fp16)[name = string("aw_chunk_807_cast_fp16")]; + fp16 var_8861_to_fp16 = const()[name = string("op_8861_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_809_cast_fp16, y = var_8861_to_fp16)[name = string("aw_chunk_809_cast_fp16")]; + fp16 var_8863_to_fp16 = const()[name = string("op_8863_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_811_cast_fp16, y = var_8863_to_fp16)[name = string("aw_chunk_811_cast_fp16")]; + fp16 var_8865_to_fp16 = const()[name = string("op_8865_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_813_cast_fp16, y = var_8865_to_fp16)[name = string("aw_chunk_813_cast_fp16")]; + fp16 var_8867_to_fp16 = const()[name = string("op_8867_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_815_cast_fp16, y = var_8867_to_fp16)[name = string("aw_chunk_815_cast_fp16")]; + fp16 var_8869_to_fp16 = const()[name = string("op_8869_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_817_cast_fp16, y = var_8869_to_fp16)[name = string("aw_chunk_817_cast_fp16")]; + fp16 var_8871_to_fp16 = const()[name = string("op_8871_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_819_cast_fp16, y = var_8871_to_fp16)[name = string("aw_chunk_819_cast_fp16")]; + fp16 var_8873_to_fp16 = const()[name = string("op_8873_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_821_cast_fp16, y = var_8873_to_fp16)[name = string("aw_chunk_821_cast_fp16")]; + fp16 var_8875_to_fp16 = const()[name = string("op_8875_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_823_cast_fp16, y = var_8875_to_fp16)[name = string("aw_chunk_823_cast_fp16")]; + fp16 var_8877_to_fp16 = const()[name = string("op_8877_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_825_cast_fp16, y = var_8877_to_fp16)[name = string("aw_chunk_825_cast_fp16")]; + fp16 var_8879_to_fp16 = const()[name = string("op_8879_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_827_cast_fp16, y = var_8879_to_fp16)[name = string("aw_chunk_827_cast_fp16")]; + fp16 var_8881_to_fp16 = const()[name = string("op_8881_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_829_cast_fp16, y = var_8881_to_fp16)[name = string("aw_chunk_829_cast_fp16")]; + fp16 var_8883_to_fp16 = const()[name = string("op_8883_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_831_cast_fp16, y = var_8883_to_fp16)[name = string("aw_chunk_831_cast_fp16")]; + fp16 var_8885_to_fp16 = const()[name = string("op_8885_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_833_cast_fp16, y = var_8885_to_fp16)[name = string("aw_chunk_833_cast_fp16")]; + fp16 var_8887_to_fp16 = const()[name = string("op_8887_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_835_cast_fp16, y = var_8887_to_fp16)[name = string("aw_chunk_835_cast_fp16")]; + fp16 var_8889_to_fp16 = const()[name = string("op_8889_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_837_cast_fp16, y = var_8889_to_fp16)[name = string("aw_chunk_837_cast_fp16")]; + fp16 var_8891_to_fp16 = const()[name = string("op_8891_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_839_cast_fp16, y = var_8891_to_fp16)[name = string("aw_chunk_839_cast_fp16")]; + fp16 var_8893_to_fp16 = const()[name = string("op_8893_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_841_cast_fp16, y = var_8893_to_fp16)[name = string("aw_chunk_841_cast_fp16")]; + fp16 var_8895_to_fp16 = const()[name = string("op_8895_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_843_cast_fp16, y = var_8895_to_fp16)[name = string("aw_chunk_843_cast_fp16")]; + fp16 var_8897_to_fp16 = const()[name = string("op_8897_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_845_cast_fp16, y = var_8897_to_fp16)[name = string("aw_chunk_845_cast_fp16")]; + fp16 var_8899_to_fp16 = const()[name = string("op_8899_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_847_cast_fp16, y = var_8899_to_fp16)[name = string("aw_chunk_847_cast_fp16")]; + fp16 var_8901_to_fp16 = const()[name = string("op_8901_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_849_cast_fp16, y = var_8901_to_fp16)[name = string("aw_chunk_849_cast_fp16")]; + fp16 var_8903_to_fp16 = const()[name = string("op_8903_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_851_cast_fp16, y = var_8903_to_fp16)[name = string("aw_chunk_851_cast_fp16")]; + fp16 var_8905_to_fp16 = const()[name = string("op_8905_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_853_cast_fp16, y = var_8905_to_fp16)[name = string("aw_chunk_853_cast_fp16")]; + fp16 var_8907_to_fp16 = const()[name = string("op_8907_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_855_cast_fp16, y = var_8907_to_fp16)[name = string("aw_chunk_855_cast_fp16")]; + fp16 var_8909_to_fp16 = const()[name = string("op_8909_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_857_cast_fp16, y = var_8909_to_fp16)[name = string("aw_chunk_857_cast_fp16")]; + fp16 var_8911_to_fp16 = const()[name = string("op_8911_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_859_cast_fp16, y = var_8911_to_fp16)[name = string("aw_chunk_859_cast_fp16")]; + fp16 var_8913_to_fp16 = const()[name = string("op_8913_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_861_cast_fp16, y = var_8913_to_fp16)[name = string("aw_chunk_861_cast_fp16")]; + fp16 var_8915_to_fp16 = const()[name = string("op_8915_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_863_cast_fp16, y = var_8915_to_fp16)[name = string("aw_chunk_863_cast_fp16")]; + fp16 var_8917_to_fp16 = const()[name = string("op_8917_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_865_cast_fp16, y = var_8917_to_fp16)[name = string("aw_chunk_865_cast_fp16")]; + fp16 var_8919_to_fp16 = const()[name = string("op_8919_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_867_cast_fp16, y = var_8919_to_fp16)[name = string("aw_chunk_867_cast_fp16")]; + fp16 var_8921_to_fp16 = const()[name = string("op_8921_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_869_cast_fp16, y = var_8921_to_fp16)[name = string("aw_chunk_869_cast_fp16")]; + fp16 var_8923_to_fp16 = const()[name = string("op_8923_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_871_cast_fp16, y = var_8923_to_fp16)[name = string("aw_chunk_871_cast_fp16")]; + fp16 var_8925_to_fp16 = const()[name = string("op_8925_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_873_cast_fp16, y = var_8925_to_fp16)[name = string("aw_chunk_873_cast_fp16")]; + fp16 var_8927_to_fp16 = const()[name = string("op_8927_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_875_cast_fp16, y = var_8927_to_fp16)[name = string("aw_chunk_875_cast_fp16")]; + fp16 var_8929_to_fp16 = const()[name = string("op_8929_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_877_cast_fp16, y = var_8929_to_fp16)[name = string("aw_chunk_877_cast_fp16")]; + fp16 var_8931_to_fp16 = const()[name = string("op_8931_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_879_cast_fp16, y = var_8931_to_fp16)[name = string("aw_chunk_879_cast_fp16")]; + fp16 var_8933_to_fp16 = const()[name = string("op_8933_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_881_cast_fp16, y = var_8933_to_fp16)[name = string("aw_chunk_881_cast_fp16")]; + fp16 var_8935_to_fp16 = const()[name = string("op_8935_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_883_cast_fp16, y = var_8935_to_fp16)[name = string("aw_chunk_883_cast_fp16")]; + fp16 var_8937_to_fp16 = const()[name = string("op_8937_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_885_cast_fp16, y = var_8937_to_fp16)[name = string("aw_chunk_885_cast_fp16")]; + fp16 var_8939_to_fp16 = const()[name = string("op_8939_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_887_cast_fp16, y = var_8939_to_fp16)[name = string("aw_chunk_887_cast_fp16")]; + fp16 var_8941_to_fp16 = const()[name = string("op_8941_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_889_cast_fp16, y = var_8941_to_fp16)[name = string("aw_chunk_889_cast_fp16")]; + fp16 var_8943_to_fp16 = const()[name = string("op_8943_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_891_cast_fp16, y = var_8943_to_fp16)[name = string("aw_chunk_891_cast_fp16")]; + fp16 var_8945_to_fp16 = const()[name = string("op_8945_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_893_cast_fp16, y = var_8945_to_fp16)[name = string("aw_chunk_893_cast_fp16")]; + fp16 var_8947_to_fp16 = const()[name = string("op_8947_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_895_cast_fp16, y = var_8947_to_fp16)[name = string("aw_chunk_895_cast_fp16")]; + fp16 var_8949_to_fp16 = const()[name = string("op_8949_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_897_cast_fp16, y = var_8949_to_fp16)[name = string("aw_chunk_897_cast_fp16")]; + fp16 var_8951_to_fp16 = const()[name = string("op_8951_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_899_cast_fp16, y = var_8951_to_fp16)[name = string("aw_chunk_899_cast_fp16")]; + fp16 var_8953_to_fp16 = const()[name = string("op_8953_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_901_cast_fp16, y = var_8953_to_fp16)[name = string("aw_chunk_901_cast_fp16")]; + fp16 var_8955_to_fp16 = const()[name = string("op_8955_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_903_cast_fp16, y = var_8955_to_fp16)[name = string("aw_chunk_903_cast_fp16")]; + fp16 var_8957_to_fp16 = const()[name = string("op_8957_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_905_cast_fp16, y = var_8957_to_fp16)[name = string("aw_chunk_905_cast_fp16")]; + fp16 var_8959_to_fp16 = const()[name = string("op_8959_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_907_cast_fp16, y = var_8959_to_fp16)[name = string("aw_chunk_907_cast_fp16")]; + fp16 var_8961_to_fp16 = const()[name = string("op_8961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_909_cast_fp16, y = var_8961_to_fp16)[name = string("aw_chunk_909_cast_fp16")]; + fp16 var_8963_to_fp16 = const()[name = string("op_8963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_911_cast_fp16, y = var_8963_to_fp16)[name = string("aw_chunk_911_cast_fp16")]; + fp16 var_8965_to_fp16 = const()[name = string("op_8965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_913_cast_fp16, y = var_8965_to_fp16)[name = string("aw_chunk_913_cast_fp16")]; + fp16 var_8967_to_fp16 = const()[name = string("op_8967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_915_cast_fp16, y = var_8967_to_fp16)[name = string("aw_chunk_915_cast_fp16")]; + fp16 var_8969_to_fp16 = const()[name = string("op_8969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_917_cast_fp16, y = var_8969_to_fp16)[name = string("aw_chunk_917_cast_fp16")]; + fp16 var_8971_to_fp16 = const()[name = string("op_8971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_919_cast_fp16, y = var_8971_to_fp16)[name = string("aw_chunk_919_cast_fp16")]; + fp16 var_8973_to_fp16 = const()[name = string("op_8973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_921_cast_fp16, y = var_8973_to_fp16)[name = string("aw_chunk_921_cast_fp16")]; + fp16 var_8975_to_fp16 = const()[name = string("op_8975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_923_cast_fp16, y = var_8975_to_fp16)[name = string("aw_chunk_923_cast_fp16")]; + fp16 var_8977_to_fp16 = const()[name = string("op_8977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_925_cast_fp16, y = var_8977_to_fp16)[name = string("aw_chunk_925_cast_fp16")]; + fp16 var_8979_to_fp16 = const()[name = string("op_8979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_927_cast_fp16, y = var_8979_to_fp16)[name = string("aw_chunk_927_cast_fp16")]; + fp16 var_8981_to_fp16 = const()[name = string("op_8981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_929_cast_fp16, y = var_8981_to_fp16)[name = string("aw_chunk_929_cast_fp16")]; + fp16 var_8983_to_fp16 = const()[name = string("op_8983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_931_cast_fp16, y = var_8983_to_fp16)[name = string("aw_chunk_931_cast_fp16")]; + fp16 var_8985_to_fp16 = const()[name = string("op_8985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_933_cast_fp16, y = var_8985_to_fp16)[name = string("aw_chunk_933_cast_fp16")]; + fp16 var_8987_to_fp16 = const()[name = string("op_8987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_935_cast_fp16, y = var_8987_to_fp16)[name = string("aw_chunk_935_cast_fp16")]; + fp16 var_8989_to_fp16 = const()[name = string("op_8989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_937_cast_fp16, y = var_8989_to_fp16)[name = string("aw_chunk_937_cast_fp16")]; + fp16 var_8991_to_fp16 = const()[name = string("op_8991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_939_cast_fp16, y = var_8991_to_fp16)[name = string("aw_chunk_939_cast_fp16")]; + fp16 var_8993_to_fp16 = const()[name = string("op_8993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_941_cast_fp16, y = var_8993_to_fp16)[name = string("aw_chunk_941_cast_fp16")]; + fp16 var_8995_to_fp16 = const()[name = string("op_8995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_943_cast_fp16, y = var_8995_to_fp16)[name = string("aw_chunk_943_cast_fp16")]; + fp16 var_8997_to_fp16 = const()[name = string("op_8997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_945_cast_fp16, y = var_8997_to_fp16)[name = string("aw_chunk_945_cast_fp16")]; + fp16 var_8999_to_fp16 = const()[name = string("op_8999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_947_cast_fp16, y = var_8999_to_fp16)[name = string("aw_chunk_947_cast_fp16")]; + fp16 var_9001_to_fp16 = const()[name = string("op_9001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_949_cast_fp16, y = var_9001_to_fp16)[name = string("aw_chunk_949_cast_fp16")]; + fp16 var_9003_to_fp16 = const()[name = string("op_9003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_951_cast_fp16, y = var_9003_to_fp16)[name = string("aw_chunk_951_cast_fp16")]; + fp16 var_9005_to_fp16 = const()[name = string("op_9005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_953_cast_fp16, y = var_9005_to_fp16)[name = string("aw_chunk_953_cast_fp16")]; + fp16 var_9007_to_fp16 = const()[name = string("op_9007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_955_cast_fp16, y = var_9007_to_fp16)[name = string("aw_chunk_955_cast_fp16")]; + fp16 var_9009_to_fp16 = const()[name = string("op_9009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_957_cast_fp16, y = var_9009_to_fp16)[name = string("aw_chunk_957_cast_fp16")]; + fp16 var_9011_to_fp16 = const()[name = string("op_9011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_959_cast_fp16, y = var_9011_to_fp16)[name = string("aw_chunk_959_cast_fp16")]; + tensor var_9013_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_801_cast_fp16)[name = string("op_9013_cast_fp16")]; + tensor var_9014_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_803_cast_fp16)[name = string("op_9014_cast_fp16")]; + tensor var_9015_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_805_cast_fp16)[name = string("op_9015_cast_fp16")]; + tensor var_9016_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_807_cast_fp16)[name = string("op_9016_cast_fp16")]; + tensor var_9017_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_809_cast_fp16)[name = string("op_9017_cast_fp16")]; + tensor var_9018_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_811_cast_fp16)[name = string("op_9018_cast_fp16")]; + tensor var_9019_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_813_cast_fp16)[name = string("op_9019_cast_fp16")]; + tensor var_9020_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_815_cast_fp16)[name = string("op_9020_cast_fp16")]; + tensor var_9021_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_817_cast_fp16)[name = string("op_9021_cast_fp16")]; + tensor var_9022_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_819_cast_fp16)[name = string("op_9022_cast_fp16")]; + tensor var_9023_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_821_cast_fp16)[name = string("op_9023_cast_fp16")]; + tensor var_9024_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_823_cast_fp16)[name = string("op_9024_cast_fp16")]; + tensor var_9025_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_825_cast_fp16)[name = string("op_9025_cast_fp16")]; + tensor var_9026_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_827_cast_fp16)[name = string("op_9026_cast_fp16")]; + tensor var_9027_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_829_cast_fp16)[name = string("op_9027_cast_fp16")]; + tensor var_9028_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_831_cast_fp16)[name = string("op_9028_cast_fp16")]; + tensor var_9029_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_833_cast_fp16)[name = string("op_9029_cast_fp16")]; + tensor var_9030_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_835_cast_fp16)[name = string("op_9030_cast_fp16")]; + tensor var_9031_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_837_cast_fp16)[name = string("op_9031_cast_fp16")]; + tensor var_9032_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_839_cast_fp16)[name = string("op_9032_cast_fp16")]; + tensor var_9033_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_841_cast_fp16)[name = string("op_9033_cast_fp16")]; + tensor var_9034_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_843_cast_fp16)[name = string("op_9034_cast_fp16")]; + tensor var_9035_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_845_cast_fp16)[name = string("op_9035_cast_fp16")]; + tensor var_9036_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_847_cast_fp16)[name = string("op_9036_cast_fp16")]; + tensor var_9037_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_849_cast_fp16)[name = string("op_9037_cast_fp16")]; + tensor var_9038_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_851_cast_fp16)[name = string("op_9038_cast_fp16")]; + tensor var_9039_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_853_cast_fp16)[name = string("op_9039_cast_fp16")]; + tensor var_9040_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_855_cast_fp16)[name = string("op_9040_cast_fp16")]; + tensor var_9041_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_857_cast_fp16)[name = string("op_9041_cast_fp16")]; + tensor var_9042_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_859_cast_fp16)[name = string("op_9042_cast_fp16")]; + tensor var_9043_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_861_cast_fp16)[name = string("op_9043_cast_fp16")]; + tensor var_9044_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_863_cast_fp16)[name = string("op_9044_cast_fp16")]; + tensor var_9045_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_865_cast_fp16)[name = string("op_9045_cast_fp16")]; + tensor var_9046_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_867_cast_fp16)[name = string("op_9046_cast_fp16")]; + tensor var_9047_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_869_cast_fp16)[name = string("op_9047_cast_fp16")]; + tensor var_9048_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_871_cast_fp16)[name = string("op_9048_cast_fp16")]; + tensor var_9049_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_873_cast_fp16)[name = string("op_9049_cast_fp16")]; + tensor var_9050_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_875_cast_fp16)[name = string("op_9050_cast_fp16")]; + tensor var_9051_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_877_cast_fp16)[name = string("op_9051_cast_fp16")]; + tensor var_9052_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_879_cast_fp16)[name = string("op_9052_cast_fp16")]; + tensor var_9053_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_881_cast_fp16)[name = string("op_9053_cast_fp16")]; + tensor var_9054_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_883_cast_fp16)[name = string("op_9054_cast_fp16")]; + tensor var_9055_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_885_cast_fp16)[name = string("op_9055_cast_fp16")]; + tensor var_9056_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_887_cast_fp16)[name = string("op_9056_cast_fp16")]; + tensor var_9057_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_889_cast_fp16)[name = string("op_9057_cast_fp16")]; + tensor var_9058_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_891_cast_fp16)[name = string("op_9058_cast_fp16")]; + tensor var_9059_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_893_cast_fp16)[name = string("op_9059_cast_fp16")]; + tensor var_9060_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_895_cast_fp16)[name = string("op_9060_cast_fp16")]; + tensor var_9061_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_897_cast_fp16)[name = string("op_9061_cast_fp16")]; + tensor var_9062_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_899_cast_fp16)[name = string("op_9062_cast_fp16")]; + tensor var_9063_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_901_cast_fp16)[name = string("op_9063_cast_fp16")]; + tensor var_9064_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_903_cast_fp16)[name = string("op_9064_cast_fp16")]; + tensor var_9065_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_905_cast_fp16)[name = string("op_9065_cast_fp16")]; + tensor var_9066_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_907_cast_fp16)[name = string("op_9066_cast_fp16")]; + tensor var_9067_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_909_cast_fp16)[name = string("op_9067_cast_fp16")]; + tensor var_9068_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_911_cast_fp16)[name = string("op_9068_cast_fp16")]; + tensor var_9069_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_913_cast_fp16)[name = string("op_9069_cast_fp16")]; + tensor var_9070_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_915_cast_fp16)[name = string("op_9070_cast_fp16")]; + tensor var_9071_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_917_cast_fp16)[name = string("op_9071_cast_fp16")]; + tensor var_9072_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_919_cast_fp16)[name = string("op_9072_cast_fp16")]; + tensor var_9073_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_921_cast_fp16)[name = string("op_9073_cast_fp16")]; + tensor var_9074_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_923_cast_fp16)[name = string("op_9074_cast_fp16")]; + tensor var_9075_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_925_cast_fp16)[name = string("op_9075_cast_fp16")]; + tensor var_9076_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_927_cast_fp16)[name = string("op_9076_cast_fp16")]; + tensor var_9077_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_929_cast_fp16)[name = string("op_9077_cast_fp16")]; + tensor var_9078_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_931_cast_fp16)[name = string("op_9078_cast_fp16")]; + tensor var_9079_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_933_cast_fp16)[name = string("op_9079_cast_fp16")]; + tensor var_9080_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_935_cast_fp16)[name = string("op_9080_cast_fp16")]; + tensor var_9081_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_937_cast_fp16)[name = string("op_9081_cast_fp16")]; + tensor var_9082_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_939_cast_fp16)[name = string("op_9082_cast_fp16")]; + tensor var_9083_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_941_cast_fp16)[name = string("op_9083_cast_fp16")]; + tensor var_9084_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_943_cast_fp16)[name = string("op_9084_cast_fp16")]; + tensor var_9085_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_945_cast_fp16)[name = string("op_9085_cast_fp16")]; + tensor var_9086_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_947_cast_fp16)[name = string("op_9086_cast_fp16")]; + tensor var_9087_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_949_cast_fp16)[name = string("op_9087_cast_fp16")]; + tensor var_9088_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_951_cast_fp16)[name = string("op_9088_cast_fp16")]; + tensor var_9089_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_953_cast_fp16)[name = string("op_9089_cast_fp16")]; + tensor var_9090_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_955_cast_fp16)[name = string("op_9090_cast_fp16")]; + tensor var_9091_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_957_cast_fp16)[name = string("op_9091_cast_fp16")]; + tensor var_9092_cast_fp16 = softmax(axis = var_7838, x = aw_chunk_959_cast_fp16)[name = string("op_9092_cast_fp16")]; + string var_9094_equation_0 = const()[name = string("op_9094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9094_cast_fp16 = einsum(equation = var_9094_equation_0, values = (var_8614_cast_fp16, var_9013_cast_fp16))[name = string("op_9094_cast_fp16")]; + string var_9096_equation_0 = const()[name = string("op_9096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9096_cast_fp16 = einsum(equation = var_9096_equation_0, values = (var_8614_cast_fp16, var_9014_cast_fp16))[name = string("op_9096_cast_fp16")]; + string var_9098_equation_0 = const()[name = string("op_9098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9098_cast_fp16 = einsum(equation = var_9098_equation_0, values = (var_8614_cast_fp16, var_9015_cast_fp16))[name = string("op_9098_cast_fp16")]; + string var_9100_equation_0 = const()[name = string("op_9100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9100_cast_fp16 = einsum(equation = var_9100_equation_0, values = (var_8614_cast_fp16, var_9016_cast_fp16))[name = string("op_9100_cast_fp16")]; + string var_9102_equation_0 = const()[name = string("op_9102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9102_cast_fp16 = einsum(equation = var_9102_equation_0, values = (var_8618_cast_fp16, var_9017_cast_fp16))[name = string("op_9102_cast_fp16")]; + string var_9104_equation_0 = const()[name = string("op_9104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9104_cast_fp16 = einsum(equation = var_9104_equation_0, values = (var_8618_cast_fp16, var_9018_cast_fp16))[name = string("op_9104_cast_fp16")]; + string var_9106_equation_0 = const()[name = string("op_9106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9106_cast_fp16 = einsum(equation = var_9106_equation_0, values = (var_8618_cast_fp16, var_9019_cast_fp16))[name = string("op_9106_cast_fp16")]; + string var_9108_equation_0 = const()[name = string("op_9108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9108_cast_fp16 = einsum(equation = var_9108_equation_0, values = (var_8618_cast_fp16, var_9020_cast_fp16))[name = string("op_9108_cast_fp16")]; + string var_9110_equation_0 = const()[name = string("op_9110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9110_cast_fp16 = einsum(equation = var_9110_equation_0, values = (var_8622_cast_fp16, var_9021_cast_fp16))[name = string("op_9110_cast_fp16")]; + string var_9112_equation_0 = const()[name = string("op_9112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9112_cast_fp16 = einsum(equation = var_9112_equation_0, values = (var_8622_cast_fp16, var_9022_cast_fp16))[name = string("op_9112_cast_fp16")]; + string var_9114_equation_0 = const()[name = string("op_9114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9114_cast_fp16 = einsum(equation = var_9114_equation_0, values = (var_8622_cast_fp16, var_9023_cast_fp16))[name = string("op_9114_cast_fp16")]; + string var_9116_equation_0 = const()[name = string("op_9116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9116_cast_fp16 = einsum(equation = var_9116_equation_0, values = (var_8622_cast_fp16, var_9024_cast_fp16))[name = string("op_9116_cast_fp16")]; + string var_9118_equation_0 = const()[name = string("op_9118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9118_cast_fp16 = einsum(equation = var_9118_equation_0, values = (var_8626_cast_fp16, var_9025_cast_fp16))[name = string("op_9118_cast_fp16")]; + string var_9120_equation_0 = const()[name = string("op_9120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9120_cast_fp16 = einsum(equation = var_9120_equation_0, values = (var_8626_cast_fp16, var_9026_cast_fp16))[name = string("op_9120_cast_fp16")]; + string var_9122_equation_0 = const()[name = string("op_9122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9122_cast_fp16 = einsum(equation = var_9122_equation_0, values = (var_8626_cast_fp16, var_9027_cast_fp16))[name = string("op_9122_cast_fp16")]; + string var_9124_equation_0 = const()[name = string("op_9124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9124_cast_fp16 = einsum(equation = var_9124_equation_0, values = (var_8626_cast_fp16, var_9028_cast_fp16))[name = string("op_9124_cast_fp16")]; + string var_9126_equation_0 = const()[name = string("op_9126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9126_cast_fp16 = einsum(equation = var_9126_equation_0, values = (var_8630_cast_fp16, var_9029_cast_fp16))[name = string("op_9126_cast_fp16")]; + string var_9128_equation_0 = const()[name = string("op_9128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9128_cast_fp16 = einsum(equation = var_9128_equation_0, values = (var_8630_cast_fp16, var_9030_cast_fp16))[name = string("op_9128_cast_fp16")]; + string var_9130_equation_0 = const()[name = string("op_9130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9130_cast_fp16 = einsum(equation = var_9130_equation_0, values = (var_8630_cast_fp16, var_9031_cast_fp16))[name = string("op_9130_cast_fp16")]; + string var_9132_equation_0 = const()[name = string("op_9132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9132_cast_fp16 = einsum(equation = var_9132_equation_0, values = (var_8630_cast_fp16, var_9032_cast_fp16))[name = string("op_9132_cast_fp16")]; + string var_9134_equation_0 = const()[name = string("op_9134_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9134_cast_fp16 = einsum(equation = var_9134_equation_0, values = (var_8634_cast_fp16, var_9033_cast_fp16))[name = string("op_9134_cast_fp16")]; + string var_9136_equation_0 = const()[name = string("op_9136_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9136_cast_fp16 = einsum(equation = var_9136_equation_0, values = (var_8634_cast_fp16, var_9034_cast_fp16))[name = string("op_9136_cast_fp16")]; + string var_9138_equation_0 = const()[name = string("op_9138_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9138_cast_fp16 = einsum(equation = var_9138_equation_0, values = (var_8634_cast_fp16, var_9035_cast_fp16))[name = string("op_9138_cast_fp16")]; + string var_9140_equation_0 = const()[name = string("op_9140_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9140_cast_fp16 = einsum(equation = var_9140_equation_0, values = (var_8634_cast_fp16, var_9036_cast_fp16))[name = string("op_9140_cast_fp16")]; + string var_9142_equation_0 = const()[name = string("op_9142_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9142_cast_fp16 = einsum(equation = var_9142_equation_0, values = (var_8638_cast_fp16, var_9037_cast_fp16))[name = string("op_9142_cast_fp16")]; + string var_9144_equation_0 = const()[name = string("op_9144_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9144_cast_fp16 = einsum(equation = var_9144_equation_0, values = (var_8638_cast_fp16, var_9038_cast_fp16))[name = string("op_9144_cast_fp16")]; + string var_9146_equation_0 = const()[name = string("op_9146_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9146_cast_fp16 = einsum(equation = var_9146_equation_0, values = (var_8638_cast_fp16, var_9039_cast_fp16))[name = string("op_9146_cast_fp16")]; + string var_9148_equation_0 = const()[name = string("op_9148_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9148_cast_fp16 = einsum(equation = var_9148_equation_0, values = (var_8638_cast_fp16, var_9040_cast_fp16))[name = string("op_9148_cast_fp16")]; + string var_9150_equation_0 = const()[name = string("op_9150_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9150_cast_fp16 = einsum(equation = var_9150_equation_0, values = (var_8642_cast_fp16, var_9041_cast_fp16))[name = string("op_9150_cast_fp16")]; + string var_9152_equation_0 = const()[name = string("op_9152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9152_cast_fp16 = einsum(equation = var_9152_equation_0, values = (var_8642_cast_fp16, var_9042_cast_fp16))[name = string("op_9152_cast_fp16")]; + string var_9154_equation_0 = const()[name = string("op_9154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9154_cast_fp16 = einsum(equation = var_9154_equation_0, values = (var_8642_cast_fp16, var_9043_cast_fp16))[name = string("op_9154_cast_fp16")]; + string var_9156_equation_0 = const()[name = string("op_9156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9156_cast_fp16 = einsum(equation = var_9156_equation_0, values = (var_8642_cast_fp16, var_9044_cast_fp16))[name = string("op_9156_cast_fp16")]; + string var_9158_equation_0 = const()[name = string("op_9158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9158_cast_fp16 = einsum(equation = var_9158_equation_0, values = (var_8646_cast_fp16, var_9045_cast_fp16))[name = string("op_9158_cast_fp16")]; + string var_9160_equation_0 = const()[name = string("op_9160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9160_cast_fp16 = einsum(equation = var_9160_equation_0, values = (var_8646_cast_fp16, var_9046_cast_fp16))[name = string("op_9160_cast_fp16")]; + string var_9162_equation_0 = const()[name = string("op_9162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9162_cast_fp16 = einsum(equation = var_9162_equation_0, values = (var_8646_cast_fp16, var_9047_cast_fp16))[name = string("op_9162_cast_fp16")]; + string var_9164_equation_0 = const()[name = string("op_9164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9164_cast_fp16 = einsum(equation = var_9164_equation_0, values = (var_8646_cast_fp16, var_9048_cast_fp16))[name = string("op_9164_cast_fp16")]; + string var_9166_equation_0 = const()[name = string("op_9166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9166_cast_fp16 = einsum(equation = var_9166_equation_0, values = (var_8650_cast_fp16, var_9049_cast_fp16))[name = string("op_9166_cast_fp16")]; + string var_9168_equation_0 = const()[name = string("op_9168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9168_cast_fp16 = einsum(equation = var_9168_equation_0, values = (var_8650_cast_fp16, var_9050_cast_fp16))[name = string("op_9168_cast_fp16")]; + string var_9170_equation_0 = const()[name = string("op_9170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9170_cast_fp16 = einsum(equation = var_9170_equation_0, values = (var_8650_cast_fp16, var_9051_cast_fp16))[name = string("op_9170_cast_fp16")]; + string var_9172_equation_0 = const()[name = string("op_9172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9172_cast_fp16 = einsum(equation = var_9172_equation_0, values = (var_8650_cast_fp16, var_9052_cast_fp16))[name = string("op_9172_cast_fp16")]; + string var_9174_equation_0 = const()[name = string("op_9174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9174_cast_fp16 = einsum(equation = var_9174_equation_0, values = (var_8654_cast_fp16, var_9053_cast_fp16))[name = string("op_9174_cast_fp16")]; + string var_9176_equation_0 = const()[name = string("op_9176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9176_cast_fp16 = einsum(equation = var_9176_equation_0, values = (var_8654_cast_fp16, var_9054_cast_fp16))[name = string("op_9176_cast_fp16")]; + string var_9178_equation_0 = const()[name = string("op_9178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9178_cast_fp16 = einsum(equation = var_9178_equation_0, values = (var_8654_cast_fp16, var_9055_cast_fp16))[name = string("op_9178_cast_fp16")]; + string var_9180_equation_0 = const()[name = string("op_9180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9180_cast_fp16 = einsum(equation = var_9180_equation_0, values = (var_8654_cast_fp16, var_9056_cast_fp16))[name = string("op_9180_cast_fp16")]; + string var_9182_equation_0 = const()[name = string("op_9182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9182_cast_fp16 = einsum(equation = var_9182_equation_0, values = (var_8658_cast_fp16, var_9057_cast_fp16))[name = string("op_9182_cast_fp16")]; + string var_9184_equation_0 = const()[name = string("op_9184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9184_cast_fp16 = einsum(equation = var_9184_equation_0, values = (var_8658_cast_fp16, var_9058_cast_fp16))[name = string("op_9184_cast_fp16")]; + string var_9186_equation_0 = const()[name = string("op_9186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9186_cast_fp16 = einsum(equation = var_9186_equation_0, values = (var_8658_cast_fp16, var_9059_cast_fp16))[name = string("op_9186_cast_fp16")]; + string var_9188_equation_0 = const()[name = string("op_9188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9188_cast_fp16 = einsum(equation = var_9188_equation_0, values = (var_8658_cast_fp16, var_9060_cast_fp16))[name = string("op_9188_cast_fp16")]; + string var_9190_equation_0 = const()[name = string("op_9190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9190_cast_fp16 = einsum(equation = var_9190_equation_0, values = (var_8662_cast_fp16, var_9061_cast_fp16))[name = string("op_9190_cast_fp16")]; + string var_9192_equation_0 = const()[name = string("op_9192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9192_cast_fp16 = einsum(equation = var_9192_equation_0, values = (var_8662_cast_fp16, var_9062_cast_fp16))[name = string("op_9192_cast_fp16")]; + string var_9194_equation_0 = const()[name = string("op_9194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9194_cast_fp16 = einsum(equation = var_9194_equation_0, values = (var_8662_cast_fp16, var_9063_cast_fp16))[name = string("op_9194_cast_fp16")]; + string var_9196_equation_0 = const()[name = string("op_9196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9196_cast_fp16 = einsum(equation = var_9196_equation_0, values = (var_8662_cast_fp16, var_9064_cast_fp16))[name = string("op_9196_cast_fp16")]; + string var_9198_equation_0 = const()[name = string("op_9198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9198_cast_fp16 = einsum(equation = var_9198_equation_0, values = (var_8666_cast_fp16, var_9065_cast_fp16))[name = string("op_9198_cast_fp16")]; + string var_9200_equation_0 = const()[name = string("op_9200_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9200_cast_fp16 = einsum(equation = var_9200_equation_0, values = (var_8666_cast_fp16, var_9066_cast_fp16))[name = string("op_9200_cast_fp16")]; + string var_9202_equation_0 = const()[name = string("op_9202_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9202_cast_fp16 = einsum(equation = var_9202_equation_0, values = (var_8666_cast_fp16, var_9067_cast_fp16))[name = string("op_9202_cast_fp16")]; + string var_9204_equation_0 = const()[name = string("op_9204_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9204_cast_fp16 = einsum(equation = var_9204_equation_0, values = (var_8666_cast_fp16, var_9068_cast_fp16))[name = string("op_9204_cast_fp16")]; + string var_9206_equation_0 = const()[name = string("op_9206_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9206_cast_fp16 = einsum(equation = var_9206_equation_0, values = (var_8670_cast_fp16, var_9069_cast_fp16))[name = string("op_9206_cast_fp16")]; + string var_9208_equation_0 = const()[name = string("op_9208_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9208_cast_fp16 = einsum(equation = var_9208_equation_0, values = (var_8670_cast_fp16, var_9070_cast_fp16))[name = string("op_9208_cast_fp16")]; + string var_9210_equation_0 = const()[name = string("op_9210_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9210_cast_fp16 = einsum(equation = var_9210_equation_0, values = (var_8670_cast_fp16, var_9071_cast_fp16))[name = string("op_9210_cast_fp16")]; + string var_9212_equation_0 = const()[name = string("op_9212_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9212_cast_fp16 = einsum(equation = var_9212_equation_0, values = (var_8670_cast_fp16, var_9072_cast_fp16))[name = string("op_9212_cast_fp16")]; + string var_9214_equation_0 = const()[name = string("op_9214_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9214_cast_fp16 = einsum(equation = var_9214_equation_0, values = (var_8674_cast_fp16, var_9073_cast_fp16))[name = string("op_9214_cast_fp16")]; + string var_9216_equation_0 = const()[name = string("op_9216_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9216_cast_fp16 = einsum(equation = var_9216_equation_0, values = (var_8674_cast_fp16, var_9074_cast_fp16))[name = string("op_9216_cast_fp16")]; + string var_9218_equation_0 = const()[name = string("op_9218_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9218_cast_fp16 = einsum(equation = var_9218_equation_0, values = (var_8674_cast_fp16, var_9075_cast_fp16))[name = string("op_9218_cast_fp16")]; + string var_9220_equation_0 = const()[name = string("op_9220_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9220_cast_fp16 = einsum(equation = var_9220_equation_0, values = (var_8674_cast_fp16, var_9076_cast_fp16))[name = string("op_9220_cast_fp16")]; + string var_9222_equation_0 = const()[name = string("op_9222_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9222_cast_fp16 = einsum(equation = var_9222_equation_0, values = (var_8678_cast_fp16, var_9077_cast_fp16))[name = string("op_9222_cast_fp16")]; + string var_9224_equation_0 = const()[name = string("op_9224_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9224_cast_fp16 = einsum(equation = var_9224_equation_0, values = (var_8678_cast_fp16, var_9078_cast_fp16))[name = string("op_9224_cast_fp16")]; + string var_9226_equation_0 = const()[name = string("op_9226_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9226_cast_fp16 = einsum(equation = var_9226_equation_0, values = (var_8678_cast_fp16, var_9079_cast_fp16))[name = string("op_9226_cast_fp16")]; + string var_9228_equation_0 = const()[name = string("op_9228_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9228_cast_fp16 = einsum(equation = var_9228_equation_0, values = (var_8678_cast_fp16, var_9080_cast_fp16))[name = string("op_9228_cast_fp16")]; + string var_9230_equation_0 = const()[name = string("op_9230_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9230_cast_fp16 = einsum(equation = var_9230_equation_0, values = (var_8682_cast_fp16, var_9081_cast_fp16))[name = string("op_9230_cast_fp16")]; + string var_9232_equation_0 = const()[name = string("op_9232_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9232_cast_fp16 = einsum(equation = var_9232_equation_0, values = (var_8682_cast_fp16, var_9082_cast_fp16))[name = string("op_9232_cast_fp16")]; + string var_9234_equation_0 = const()[name = string("op_9234_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9234_cast_fp16 = einsum(equation = var_9234_equation_0, values = (var_8682_cast_fp16, var_9083_cast_fp16))[name = string("op_9234_cast_fp16")]; + string var_9236_equation_0 = const()[name = string("op_9236_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9236_cast_fp16 = einsum(equation = var_9236_equation_0, values = (var_8682_cast_fp16, var_9084_cast_fp16))[name = string("op_9236_cast_fp16")]; + string var_9238_equation_0 = const()[name = string("op_9238_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9238_cast_fp16 = einsum(equation = var_9238_equation_0, values = (var_8686_cast_fp16, var_9085_cast_fp16))[name = string("op_9238_cast_fp16")]; + string var_9240_equation_0 = const()[name = string("op_9240_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9240_cast_fp16 = einsum(equation = var_9240_equation_0, values = (var_8686_cast_fp16, var_9086_cast_fp16))[name = string("op_9240_cast_fp16")]; + string var_9242_equation_0 = const()[name = string("op_9242_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9242_cast_fp16 = einsum(equation = var_9242_equation_0, values = (var_8686_cast_fp16, var_9087_cast_fp16))[name = string("op_9242_cast_fp16")]; + string var_9244_equation_0 = const()[name = string("op_9244_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9244_cast_fp16 = einsum(equation = var_9244_equation_0, values = (var_8686_cast_fp16, var_9088_cast_fp16))[name = string("op_9244_cast_fp16")]; + string var_9246_equation_0 = const()[name = string("op_9246_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9246_cast_fp16 = einsum(equation = var_9246_equation_0, values = (var_8690_cast_fp16, var_9089_cast_fp16))[name = string("op_9246_cast_fp16")]; + string var_9248_equation_0 = const()[name = string("op_9248_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9248_cast_fp16 = einsum(equation = var_9248_equation_0, values = (var_8690_cast_fp16, var_9090_cast_fp16))[name = string("op_9248_cast_fp16")]; + string var_9250_equation_0 = const()[name = string("op_9250_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9250_cast_fp16 = einsum(equation = var_9250_equation_0, values = (var_8690_cast_fp16, var_9091_cast_fp16))[name = string("op_9250_cast_fp16")]; + string var_9252_equation_0 = const()[name = string("op_9252_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9252_cast_fp16 = einsum(equation = var_9252_equation_0, values = (var_8690_cast_fp16, var_9092_cast_fp16))[name = string("op_9252_cast_fp16")]; + bool var_9254_interleave_0 = const()[name = string("op_9254_interleave_0"), val = bool(false)]; + tensor var_9254_cast_fp16 = concat(axis = var_7813, interleave = var_9254_interleave_0, values = (var_9094_cast_fp16, var_9096_cast_fp16, var_9098_cast_fp16, var_9100_cast_fp16))[name = string("op_9254_cast_fp16")]; + bool var_9256_interleave_0 = const()[name = string("op_9256_interleave_0"), val = bool(false)]; + tensor var_9256_cast_fp16 = concat(axis = var_7813, interleave = var_9256_interleave_0, values = (var_9102_cast_fp16, var_9104_cast_fp16, var_9106_cast_fp16, var_9108_cast_fp16))[name = string("op_9256_cast_fp16")]; + bool var_9258_interleave_0 = const()[name = string("op_9258_interleave_0"), val = bool(false)]; + tensor var_9258_cast_fp16 = concat(axis = var_7813, interleave = var_9258_interleave_0, values = (var_9110_cast_fp16, var_9112_cast_fp16, var_9114_cast_fp16, var_9116_cast_fp16))[name = string("op_9258_cast_fp16")]; + bool var_9260_interleave_0 = const()[name = string("op_9260_interleave_0"), val = bool(false)]; + tensor var_9260_cast_fp16 = concat(axis = var_7813, interleave = var_9260_interleave_0, values = (var_9118_cast_fp16, var_9120_cast_fp16, var_9122_cast_fp16, var_9124_cast_fp16))[name = string("op_9260_cast_fp16")]; + bool var_9262_interleave_0 = const()[name = string("op_9262_interleave_0"), val = bool(false)]; + tensor var_9262_cast_fp16 = concat(axis = var_7813, interleave = var_9262_interleave_0, values = (var_9126_cast_fp16, var_9128_cast_fp16, var_9130_cast_fp16, var_9132_cast_fp16))[name = string("op_9262_cast_fp16")]; + bool var_9264_interleave_0 = const()[name = string("op_9264_interleave_0"), val = bool(false)]; + tensor var_9264_cast_fp16 = concat(axis = var_7813, interleave = var_9264_interleave_0, values = (var_9134_cast_fp16, var_9136_cast_fp16, var_9138_cast_fp16, var_9140_cast_fp16))[name = string("op_9264_cast_fp16")]; + bool var_9266_interleave_0 = const()[name = string("op_9266_interleave_0"), val = bool(false)]; + tensor var_9266_cast_fp16 = concat(axis = var_7813, interleave = var_9266_interleave_0, values = (var_9142_cast_fp16, var_9144_cast_fp16, var_9146_cast_fp16, var_9148_cast_fp16))[name = string("op_9266_cast_fp16")]; + bool var_9268_interleave_0 = const()[name = string("op_9268_interleave_0"), val = bool(false)]; + tensor var_9268_cast_fp16 = concat(axis = var_7813, interleave = var_9268_interleave_0, values = (var_9150_cast_fp16, var_9152_cast_fp16, var_9154_cast_fp16, var_9156_cast_fp16))[name = string("op_9268_cast_fp16")]; + bool var_9270_interleave_0 = const()[name = string("op_9270_interleave_0"), val = bool(false)]; + tensor var_9270_cast_fp16 = concat(axis = var_7813, interleave = var_9270_interleave_0, values = (var_9158_cast_fp16, var_9160_cast_fp16, var_9162_cast_fp16, var_9164_cast_fp16))[name = string("op_9270_cast_fp16")]; + bool var_9272_interleave_0 = const()[name = string("op_9272_interleave_0"), val = bool(false)]; + tensor var_9272_cast_fp16 = concat(axis = var_7813, interleave = var_9272_interleave_0, values = (var_9166_cast_fp16, var_9168_cast_fp16, var_9170_cast_fp16, var_9172_cast_fp16))[name = string("op_9272_cast_fp16")]; + bool var_9274_interleave_0 = const()[name = string("op_9274_interleave_0"), val = bool(false)]; + tensor var_9274_cast_fp16 = concat(axis = var_7813, interleave = var_9274_interleave_0, values = (var_9174_cast_fp16, var_9176_cast_fp16, var_9178_cast_fp16, var_9180_cast_fp16))[name = string("op_9274_cast_fp16")]; + bool var_9276_interleave_0 = const()[name = string("op_9276_interleave_0"), val = bool(false)]; + tensor var_9276_cast_fp16 = concat(axis = var_7813, interleave = var_9276_interleave_0, values = (var_9182_cast_fp16, var_9184_cast_fp16, var_9186_cast_fp16, var_9188_cast_fp16))[name = string("op_9276_cast_fp16")]; + bool var_9278_interleave_0 = const()[name = string("op_9278_interleave_0"), val = bool(false)]; + tensor var_9278_cast_fp16 = concat(axis = var_7813, interleave = var_9278_interleave_0, values = (var_9190_cast_fp16, var_9192_cast_fp16, var_9194_cast_fp16, var_9196_cast_fp16))[name = string("op_9278_cast_fp16")]; + bool var_9280_interleave_0 = const()[name = string("op_9280_interleave_0"), val = bool(false)]; + tensor var_9280_cast_fp16 = concat(axis = var_7813, interleave = var_9280_interleave_0, values = (var_9198_cast_fp16, var_9200_cast_fp16, var_9202_cast_fp16, var_9204_cast_fp16))[name = string("op_9280_cast_fp16")]; + bool var_9282_interleave_0 = const()[name = string("op_9282_interleave_0"), val = bool(false)]; + tensor var_9282_cast_fp16 = concat(axis = var_7813, interleave = var_9282_interleave_0, values = (var_9206_cast_fp16, var_9208_cast_fp16, var_9210_cast_fp16, var_9212_cast_fp16))[name = string("op_9282_cast_fp16")]; + bool var_9284_interleave_0 = const()[name = string("op_9284_interleave_0"), val = bool(false)]; + tensor var_9284_cast_fp16 = concat(axis = var_7813, interleave = var_9284_interleave_0, values = (var_9214_cast_fp16, var_9216_cast_fp16, var_9218_cast_fp16, var_9220_cast_fp16))[name = string("op_9284_cast_fp16")]; + bool var_9286_interleave_0 = const()[name = string("op_9286_interleave_0"), val = bool(false)]; + tensor var_9286_cast_fp16 = concat(axis = var_7813, interleave = var_9286_interleave_0, values = (var_9222_cast_fp16, var_9224_cast_fp16, var_9226_cast_fp16, var_9228_cast_fp16))[name = string("op_9286_cast_fp16")]; + bool var_9288_interleave_0 = const()[name = string("op_9288_interleave_0"), val = bool(false)]; + tensor var_9288_cast_fp16 = concat(axis = var_7813, interleave = var_9288_interleave_0, values = (var_9230_cast_fp16, var_9232_cast_fp16, var_9234_cast_fp16, var_9236_cast_fp16))[name = string("op_9288_cast_fp16")]; + bool var_9290_interleave_0 = const()[name = string("op_9290_interleave_0"), val = bool(false)]; + tensor var_9290_cast_fp16 = concat(axis = var_7813, interleave = var_9290_interleave_0, values = (var_9238_cast_fp16, var_9240_cast_fp16, var_9242_cast_fp16, var_9244_cast_fp16))[name = string("op_9290_cast_fp16")]; + bool var_9292_interleave_0 = const()[name = string("op_9292_interleave_0"), val = bool(false)]; + tensor var_9292_cast_fp16 = concat(axis = var_7813, interleave = var_9292_interleave_0, values = (var_9246_cast_fp16, var_9248_cast_fp16, var_9250_cast_fp16, var_9252_cast_fp16))[name = string("op_9292_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_7838, interleave = input_41_interleave_0, values = (var_9254_cast_fp16, var_9256_cast_fp16, var_9258_cast_fp16, var_9260_cast_fp16, var_9262_cast_fp16, var_9264_cast_fp16, var_9266_cast_fp16, var_9268_cast_fp16, var_9270_cast_fp16, var_9272_cast_fp16, var_9274_cast_fp16, var_9276_cast_fp16, var_9278_cast_fp16, var_9280_cast_fp16, var_9282_cast_fp16, var_9284_cast_fp16, var_9286_cast_fp16, var_9288_cast_fp16, var_9290_cast_fp16, var_9292_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; + tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; + tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; + int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221271680)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224548544)))]; + tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_9311_to_fp16 = const()[name = string("op_9311_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_9311_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224551168)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224553792)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224556416)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237663680)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237673984)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250781248)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; + int32 var_9340 = const()[name = string("op_9340"), val = int32(3)]; + int32 var_9365 = const()[name = string("op_9365"), val = int32(1)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_9382_to_fp16 = const()[name = string("op_9382_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_9382_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250783872)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250786496)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250789120)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254065984)))]; + tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; + string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; + tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; + tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; + int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254068608)))]; + tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; + string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; + tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; + tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; + int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257345472)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260622336)))]; + tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_9420_begin_0 = const()[name = string("op_9420_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9420_end_0 = const()[name = string("op_9420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9420_end_mask_0 = const()[name = string("op_9420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9420_cast_fp16 = slice_by_index(begin = var_9420_begin_0, end = var_9420_end_0, end_mask = var_9420_end_mask_0, x = query_13_cast_fp16)[name = string("op_9420_cast_fp16")]; + tensor var_9424_begin_0 = const()[name = string("op_9424_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9424_end_0 = const()[name = string("op_9424_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9424_end_mask_0 = const()[name = string("op_9424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9424_cast_fp16 = slice_by_index(begin = var_9424_begin_0, end = var_9424_end_0, end_mask = var_9424_end_mask_0, x = query_13_cast_fp16)[name = string("op_9424_cast_fp16")]; + tensor var_9428_begin_0 = const()[name = string("op_9428_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9428_end_0 = const()[name = string("op_9428_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9428_end_mask_0 = const()[name = string("op_9428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9428_cast_fp16 = slice_by_index(begin = var_9428_begin_0, end = var_9428_end_0, end_mask = var_9428_end_mask_0, x = query_13_cast_fp16)[name = string("op_9428_cast_fp16")]; + tensor var_9432_begin_0 = const()[name = string("op_9432_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9432_end_0 = const()[name = string("op_9432_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9432_end_mask_0 = const()[name = string("op_9432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9432_cast_fp16 = slice_by_index(begin = var_9432_begin_0, end = var_9432_end_0, end_mask = var_9432_end_mask_0, x = query_13_cast_fp16)[name = string("op_9432_cast_fp16")]; + tensor var_9436_begin_0 = const()[name = string("op_9436_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9436_end_0 = const()[name = string("op_9436_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9436_end_mask_0 = const()[name = string("op_9436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9436_cast_fp16 = slice_by_index(begin = var_9436_begin_0, end = var_9436_end_0, end_mask = var_9436_end_mask_0, x = query_13_cast_fp16)[name = string("op_9436_cast_fp16")]; + tensor var_9440_begin_0 = const()[name = string("op_9440_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9440_end_0 = const()[name = string("op_9440_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9440_end_mask_0 = const()[name = string("op_9440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9440_cast_fp16 = slice_by_index(begin = var_9440_begin_0, end = var_9440_end_0, end_mask = var_9440_end_mask_0, x = query_13_cast_fp16)[name = string("op_9440_cast_fp16")]; + tensor var_9444_begin_0 = const()[name = string("op_9444_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9444_end_0 = const()[name = string("op_9444_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9444_end_mask_0 = const()[name = string("op_9444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9444_cast_fp16 = slice_by_index(begin = var_9444_begin_0, end = var_9444_end_0, end_mask = var_9444_end_mask_0, x = query_13_cast_fp16)[name = string("op_9444_cast_fp16")]; + tensor var_9448_begin_0 = const()[name = string("op_9448_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9448_end_0 = const()[name = string("op_9448_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9448_end_mask_0 = const()[name = string("op_9448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9448_cast_fp16 = slice_by_index(begin = var_9448_begin_0, end = var_9448_end_0, end_mask = var_9448_end_mask_0, x = query_13_cast_fp16)[name = string("op_9448_cast_fp16")]; + tensor var_9452_begin_0 = const()[name = string("op_9452_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9452_end_0 = const()[name = string("op_9452_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9452_end_mask_0 = const()[name = string("op_9452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9452_cast_fp16 = slice_by_index(begin = var_9452_begin_0, end = var_9452_end_0, end_mask = var_9452_end_mask_0, x = query_13_cast_fp16)[name = string("op_9452_cast_fp16")]; + tensor var_9456_begin_0 = const()[name = string("op_9456_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9456_end_0 = const()[name = string("op_9456_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9456_end_mask_0 = const()[name = string("op_9456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9456_cast_fp16 = slice_by_index(begin = var_9456_begin_0, end = var_9456_end_0, end_mask = var_9456_end_mask_0, x = query_13_cast_fp16)[name = string("op_9456_cast_fp16")]; + tensor var_9460_begin_0 = const()[name = string("op_9460_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9460_end_0 = const()[name = string("op_9460_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9460_end_mask_0 = const()[name = string("op_9460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9460_cast_fp16 = slice_by_index(begin = var_9460_begin_0, end = var_9460_end_0, end_mask = var_9460_end_mask_0, x = query_13_cast_fp16)[name = string("op_9460_cast_fp16")]; + tensor var_9464_begin_0 = const()[name = string("op_9464_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9464_end_0 = const()[name = string("op_9464_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9464_end_mask_0 = const()[name = string("op_9464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9464_cast_fp16 = slice_by_index(begin = var_9464_begin_0, end = var_9464_end_0, end_mask = var_9464_end_mask_0, x = query_13_cast_fp16)[name = string("op_9464_cast_fp16")]; + tensor var_9468_begin_0 = const()[name = string("op_9468_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9468_end_0 = const()[name = string("op_9468_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_9468_end_mask_0 = const()[name = string("op_9468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9468_cast_fp16 = slice_by_index(begin = var_9468_begin_0, end = var_9468_end_0, end_mask = var_9468_end_mask_0, x = query_13_cast_fp16)[name = string("op_9468_cast_fp16")]; + tensor var_9472_begin_0 = const()[name = string("op_9472_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9472_end_0 = const()[name = string("op_9472_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_9472_end_mask_0 = const()[name = string("op_9472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9472_cast_fp16 = slice_by_index(begin = var_9472_begin_0, end = var_9472_end_0, end_mask = var_9472_end_mask_0, x = query_13_cast_fp16)[name = string("op_9472_cast_fp16")]; + tensor var_9476_begin_0 = const()[name = string("op_9476_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9476_end_0 = const()[name = string("op_9476_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_9476_end_mask_0 = const()[name = string("op_9476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9476_cast_fp16 = slice_by_index(begin = var_9476_begin_0, end = var_9476_end_0, end_mask = var_9476_end_mask_0, x = query_13_cast_fp16)[name = string("op_9476_cast_fp16")]; + tensor var_9480_begin_0 = const()[name = string("op_9480_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9480_end_0 = const()[name = string("op_9480_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_9480_end_mask_0 = const()[name = string("op_9480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9480_cast_fp16 = slice_by_index(begin = var_9480_begin_0, end = var_9480_end_0, end_mask = var_9480_end_mask_0, x = query_13_cast_fp16)[name = string("op_9480_cast_fp16")]; + tensor var_9484_begin_0 = const()[name = string("op_9484_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9484_end_0 = const()[name = string("op_9484_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_9484_end_mask_0 = const()[name = string("op_9484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9484_cast_fp16 = slice_by_index(begin = var_9484_begin_0, end = var_9484_end_0, end_mask = var_9484_end_mask_0, x = query_13_cast_fp16)[name = string("op_9484_cast_fp16")]; + tensor var_9488_begin_0 = const()[name = string("op_9488_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9488_end_0 = const()[name = string("op_9488_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_9488_end_mask_0 = const()[name = string("op_9488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9488_cast_fp16 = slice_by_index(begin = var_9488_begin_0, end = var_9488_end_0, end_mask = var_9488_end_mask_0, x = query_13_cast_fp16)[name = string("op_9488_cast_fp16")]; + tensor var_9492_begin_0 = const()[name = string("op_9492_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9492_end_0 = const()[name = string("op_9492_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_9492_end_mask_0 = const()[name = string("op_9492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9492_cast_fp16 = slice_by_index(begin = var_9492_begin_0, end = var_9492_end_0, end_mask = var_9492_end_mask_0, x = query_13_cast_fp16)[name = string("op_9492_cast_fp16")]; + tensor var_9496_begin_0 = const()[name = string("op_9496_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9496_end_0 = const()[name = string("op_9496_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_9496_end_mask_0 = const()[name = string("op_9496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9496_cast_fp16 = slice_by_index(begin = var_9496_begin_0, end = var_9496_end_0, end_mask = var_9496_end_mask_0, x = query_13_cast_fp16)[name = string("op_9496_cast_fp16")]; + tensor var_9505_begin_0 = const()[name = string("op_9505_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9505_end_0 = const()[name = string("op_9505_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9505_end_mask_0 = const()[name = string("op_9505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9505_cast_fp16 = slice_by_index(begin = var_9505_begin_0, end = var_9505_end_0, end_mask = var_9505_end_mask_0, x = var_9420_cast_fp16)[name = string("op_9505_cast_fp16")]; + tensor var_9512_begin_0 = const()[name = string("op_9512_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9512_end_0 = const()[name = string("op_9512_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9512_end_mask_0 = const()[name = string("op_9512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9512_cast_fp16 = slice_by_index(begin = var_9512_begin_0, end = var_9512_end_0, end_mask = var_9512_end_mask_0, x = var_9420_cast_fp16)[name = string("op_9512_cast_fp16")]; + tensor var_9519_begin_0 = const()[name = string("op_9519_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9519_end_0 = const()[name = string("op_9519_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9519_end_mask_0 = const()[name = string("op_9519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9519_cast_fp16 = slice_by_index(begin = var_9519_begin_0, end = var_9519_end_0, end_mask = var_9519_end_mask_0, x = var_9420_cast_fp16)[name = string("op_9519_cast_fp16")]; + tensor var_9526_begin_0 = const()[name = string("op_9526_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9526_end_0 = const()[name = string("op_9526_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9526_end_mask_0 = const()[name = string("op_9526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9526_cast_fp16 = slice_by_index(begin = var_9526_begin_0, end = var_9526_end_0, end_mask = var_9526_end_mask_0, x = var_9420_cast_fp16)[name = string("op_9526_cast_fp16")]; + tensor var_9533_begin_0 = const()[name = string("op_9533_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9533_end_0 = const()[name = string("op_9533_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9533_end_mask_0 = const()[name = string("op_9533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9533_cast_fp16 = slice_by_index(begin = var_9533_begin_0, end = var_9533_end_0, end_mask = var_9533_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9533_cast_fp16")]; + tensor var_9540_begin_0 = const()[name = string("op_9540_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9540_end_0 = const()[name = string("op_9540_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9540_end_mask_0 = const()[name = string("op_9540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9540_cast_fp16 = slice_by_index(begin = var_9540_begin_0, end = var_9540_end_0, end_mask = var_9540_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9540_cast_fp16")]; + tensor var_9547_begin_0 = const()[name = string("op_9547_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9547_end_0 = const()[name = string("op_9547_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9547_end_mask_0 = const()[name = string("op_9547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9547_cast_fp16 = slice_by_index(begin = var_9547_begin_0, end = var_9547_end_0, end_mask = var_9547_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9547_cast_fp16")]; + tensor var_9554_begin_0 = const()[name = string("op_9554_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9554_end_0 = const()[name = string("op_9554_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9554_end_mask_0 = const()[name = string("op_9554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9554_cast_fp16 = slice_by_index(begin = var_9554_begin_0, end = var_9554_end_0, end_mask = var_9554_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9554_cast_fp16")]; + tensor var_9561_begin_0 = const()[name = string("op_9561_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9561_end_0 = const()[name = string("op_9561_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9561_end_mask_0 = const()[name = string("op_9561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9561_cast_fp16 = slice_by_index(begin = var_9561_begin_0, end = var_9561_end_0, end_mask = var_9561_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9561_cast_fp16")]; + tensor var_9568_begin_0 = const()[name = string("op_9568_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9568_end_0 = const()[name = string("op_9568_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9568_end_mask_0 = const()[name = string("op_9568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9568_cast_fp16 = slice_by_index(begin = var_9568_begin_0, end = var_9568_end_0, end_mask = var_9568_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9568_cast_fp16")]; + tensor var_9575_begin_0 = const()[name = string("op_9575_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9575_end_0 = const()[name = string("op_9575_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9575_end_mask_0 = const()[name = string("op_9575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9575_cast_fp16 = slice_by_index(begin = var_9575_begin_0, end = var_9575_end_0, end_mask = var_9575_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9575_cast_fp16")]; + tensor var_9582_begin_0 = const()[name = string("op_9582_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9582_end_0 = const()[name = string("op_9582_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9582_end_mask_0 = const()[name = string("op_9582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9582_cast_fp16 = slice_by_index(begin = var_9582_begin_0, end = var_9582_end_0, end_mask = var_9582_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9582_cast_fp16")]; + tensor var_9589_begin_0 = const()[name = string("op_9589_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9589_end_0 = const()[name = string("op_9589_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9589_end_mask_0 = const()[name = string("op_9589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9589_cast_fp16 = slice_by_index(begin = var_9589_begin_0, end = var_9589_end_0, end_mask = var_9589_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9589_cast_fp16")]; + tensor var_9596_begin_0 = const()[name = string("op_9596_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9596_end_0 = const()[name = string("op_9596_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9596_end_mask_0 = const()[name = string("op_9596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9596_cast_fp16 = slice_by_index(begin = var_9596_begin_0, end = var_9596_end_0, end_mask = var_9596_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9596_cast_fp16")]; + tensor var_9603_begin_0 = const()[name = string("op_9603_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9603_end_0 = const()[name = string("op_9603_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9603_end_mask_0 = const()[name = string("op_9603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9603_cast_fp16 = slice_by_index(begin = var_9603_begin_0, end = var_9603_end_0, end_mask = var_9603_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9603_cast_fp16")]; + tensor var_9610_begin_0 = const()[name = string("op_9610_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9610_end_0 = const()[name = string("op_9610_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9610_end_mask_0 = const()[name = string("op_9610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9610_cast_fp16 = slice_by_index(begin = var_9610_begin_0, end = var_9610_end_0, end_mask = var_9610_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9610_cast_fp16")]; + tensor var_9617_begin_0 = const()[name = string("op_9617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9617_end_0 = const()[name = string("op_9617_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9617_end_mask_0 = const()[name = string("op_9617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9617_cast_fp16 = slice_by_index(begin = var_9617_begin_0, end = var_9617_end_0, end_mask = var_9617_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9617_cast_fp16")]; + tensor var_9624_begin_0 = const()[name = string("op_9624_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9624_end_0 = const()[name = string("op_9624_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9624_end_mask_0 = const()[name = string("op_9624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9624_cast_fp16 = slice_by_index(begin = var_9624_begin_0, end = var_9624_end_0, end_mask = var_9624_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9624_cast_fp16")]; + tensor var_9631_begin_0 = const()[name = string("op_9631_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9631_end_0 = const()[name = string("op_9631_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9631_end_mask_0 = const()[name = string("op_9631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9631_cast_fp16 = slice_by_index(begin = var_9631_begin_0, end = var_9631_end_0, end_mask = var_9631_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9631_cast_fp16")]; + tensor var_9638_begin_0 = const()[name = string("op_9638_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9638_end_0 = const()[name = string("op_9638_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9638_end_mask_0 = const()[name = string("op_9638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9638_cast_fp16 = slice_by_index(begin = var_9638_begin_0, end = var_9638_end_0, end_mask = var_9638_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9638_cast_fp16")]; + tensor var_9645_begin_0 = const()[name = string("op_9645_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9645_end_0 = const()[name = string("op_9645_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9645_end_mask_0 = const()[name = string("op_9645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9645_cast_fp16 = slice_by_index(begin = var_9645_begin_0, end = var_9645_end_0, end_mask = var_9645_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9645_cast_fp16")]; + tensor var_9652_begin_0 = const()[name = string("op_9652_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9652_end_0 = const()[name = string("op_9652_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9652_end_mask_0 = const()[name = string("op_9652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9652_cast_fp16 = slice_by_index(begin = var_9652_begin_0, end = var_9652_end_0, end_mask = var_9652_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9652_cast_fp16")]; + tensor var_9659_begin_0 = const()[name = string("op_9659_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9659_end_0 = const()[name = string("op_9659_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9659_end_mask_0 = const()[name = string("op_9659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9659_cast_fp16 = slice_by_index(begin = var_9659_begin_0, end = var_9659_end_0, end_mask = var_9659_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9659_cast_fp16")]; + tensor var_9666_begin_0 = const()[name = string("op_9666_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9666_end_0 = const()[name = string("op_9666_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9666_end_mask_0 = const()[name = string("op_9666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9666_cast_fp16 = slice_by_index(begin = var_9666_begin_0, end = var_9666_end_0, end_mask = var_9666_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9666_cast_fp16")]; + tensor var_9673_begin_0 = const()[name = string("op_9673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9673_end_0 = const()[name = string("op_9673_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9673_end_mask_0 = const()[name = string("op_9673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9673_cast_fp16 = slice_by_index(begin = var_9673_begin_0, end = var_9673_end_0, end_mask = var_9673_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9673_cast_fp16")]; + tensor var_9680_begin_0 = const()[name = string("op_9680_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9680_end_0 = const()[name = string("op_9680_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9680_end_mask_0 = const()[name = string("op_9680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9680_cast_fp16 = slice_by_index(begin = var_9680_begin_0, end = var_9680_end_0, end_mask = var_9680_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9680_cast_fp16")]; + tensor var_9687_begin_0 = const()[name = string("op_9687_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9687_end_0 = const()[name = string("op_9687_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9687_end_mask_0 = const()[name = string("op_9687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9687_cast_fp16 = slice_by_index(begin = var_9687_begin_0, end = var_9687_end_0, end_mask = var_9687_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9687_cast_fp16")]; + tensor var_9694_begin_0 = const()[name = string("op_9694_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9694_end_0 = const()[name = string("op_9694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9694_end_mask_0 = const()[name = string("op_9694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9694_cast_fp16 = slice_by_index(begin = var_9694_begin_0, end = var_9694_end_0, end_mask = var_9694_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9694_cast_fp16")]; + tensor var_9701_begin_0 = const()[name = string("op_9701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9701_end_0 = const()[name = string("op_9701_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9701_end_mask_0 = const()[name = string("op_9701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9701_cast_fp16 = slice_by_index(begin = var_9701_begin_0, end = var_9701_end_0, end_mask = var_9701_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9701_cast_fp16")]; + tensor var_9708_begin_0 = const()[name = string("op_9708_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9708_end_0 = const()[name = string("op_9708_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9708_end_mask_0 = const()[name = string("op_9708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9708_cast_fp16 = slice_by_index(begin = var_9708_begin_0, end = var_9708_end_0, end_mask = var_9708_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9708_cast_fp16")]; + tensor var_9715_begin_0 = const()[name = string("op_9715_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9715_end_0 = const()[name = string("op_9715_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9715_end_mask_0 = const()[name = string("op_9715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9715_cast_fp16 = slice_by_index(begin = var_9715_begin_0, end = var_9715_end_0, end_mask = var_9715_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9715_cast_fp16")]; + tensor var_9722_begin_0 = const()[name = string("op_9722_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9722_end_0 = const()[name = string("op_9722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9722_end_mask_0 = const()[name = string("op_9722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9722_cast_fp16 = slice_by_index(begin = var_9722_begin_0, end = var_9722_end_0, end_mask = var_9722_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9722_cast_fp16")]; + tensor var_9729_begin_0 = const()[name = string("op_9729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9729_end_0 = const()[name = string("op_9729_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9729_end_mask_0 = const()[name = string("op_9729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9729_cast_fp16 = slice_by_index(begin = var_9729_begin_0, end = var_9729_end_0, end_mask = var_9729_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9729_cast_fp16")]; + tensor var_9736_begin_0 = const()[name = string("op_9736_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9736_end_0 = const()[name = string("op_9736_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9736_end_mask_0 = const()[name = string("op_9736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9736_cast_fp16 = slice_by_index(begin = var_9736_begin_0, end = var_9736_end_0, end_mask = var_9736_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9736_cast_fp16")]; + tensor var_9743_begin_0 = const()[name = string("op_9743_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9743_end_0 = const()[name = string("op_9743_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9743_end_mask_0 = const()[name = string("op_9743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9743_cast_fp16 = slice_by_index(begin = var_9743_begin_0, end = var_9743_end_0, end_mask = var_9743_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9743_cast_fp16")]; + tensor var_9750_begin_0 = const()[name = string("op_9750_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9750_end_0 = const()[name = string("op_9750_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9750_end_mask_0 = const()[name = string("op_9750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9750_cast_fp16 = slice_by_index(begin = var_9750_begin_0, end = var_9750_end_0, end_mask = var_9750_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9750_cast_fp16")]; + tensor var_9757_begin_0 = const()[name = string("op_9757_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9757_end_0 = const()[name = string("op_9757_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9757_end_mask_0 = const()[name = string("op_9757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9757_cast_fp16 = slice_by_index(begin = var_9757_begin_0, end = var_9757_end_0, end_mask = var_9757_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9757_cast_fp16")]; + tensor var_9764_begin_0 = const()[name = string("op_9764_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9764_end_0 = const()[name = string("op_9764_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9764_end_mask_0 = const()[name = string("op_9764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9764_cast_fp16 = slice_by_index(begin = var_9764_begin_0, end = var_9764_end_0, end_mask = var_9764_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9764_cast_fp16")]; + tensor var_9771_begin_0 = const()[name = string("op_9771_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9771_end_0 = const()[name = string("op_9771_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9771_end_mask_0 = const()[name = string("op_9771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9771_cast_fp16 = slice_by_index(begin = var_9771_begin_0, end = var_9771_end_0, end_mask = var_9771_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9771_cast_fp16")]; + tensor var_9778_begin_0 = const()[name = string("op_9778_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9778_end_0 = const()[name = string("op_9778_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9778_end_mask_0 = const()[name = string("op_9778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9778_cast_fp16 = slice_by_index(begin = var_9778_begin_0, end = var_9778_end_0, end_mask = var_9778_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9778_cast_fp16")]; + tensor var_9785_begin_0 = const()[name = string("op_9785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9785_end_0 = const()[name = string("op_9785_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9785_end_mask_0 = const()[name = string("op_9785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9785_cast_fp16 = slice_by_index(begin = var_9785_begin_0, end = var_9785_end_0, end_mask = var_9785_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9785_cast_fp16")]; + tensor var_9792_begin_0 = const()[name = string("op_9792_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9792_end_0 = const()[name = string("op_9792_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9792_end_mask_0 = const()[name = string("op_9792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9792_cast_fp16 = slice_by_index(begin = var_9792_begin_0, end = var_9792_end_0, end_mask = var_9792_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9792_cast_fp16")]; + tensor var_9799_begin_0 = const()[name = string("op_9799_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9799_end_0 = const()[name = string("op_9799_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9799_end_mask_0 = const()[name = string("op_9799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9799_cast_fp16 = slice_by_index(begin = var_9799_begin_0, end = var_9799_end_0, end_mask = var_9799_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9799_cast_fp16")]; + tensor var_9806_begin_0 = const()[name = string("op_9806_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9806_end_0 = const()[name = string("op_9806_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9806_end_mask_0 = const()[name = string("op_9806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9806_cast_fp16 = slice_by_index(begin = var_9806_begin_0, end = var_9806_end_0, end_mask = var_9806_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9806_cast_fp16")]; + tensor var_9813_begin_0 = const()[name = string("op_9813_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9813_end_0 = const()[name = string("op_9813_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9813_end_mask_0 = const()[name = string("op_9813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9813_cast_fp16 = slice_by_index(begin = var_9813_begin_0, end = var_9813_end_0, end_mask = var_9813_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9813_cast_fp16")]; + tensor var_9820_begin_0 = const()[name = string("op_9820_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9820_end_0 = const()[name = string("op_9820_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9820_end_mask_0 = const()[name = string("op_9820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9820_cast_fp16 = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9820_cast_fp16")]; + tensor var_9827_begin_0 = const()[name = string("op_9827_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9827_end_0 = const()[name = string("op_9827_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9827_end_mask_0 = const()[name = string("op_9827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9827_cast_fp16 = slice_by_index(begin = var_9827_begin_0, end = var_9827_end_0, end_mask = var_9827_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9827_cast_fp16")]; + tensor var_9834_begin_0 = const()[name = string("op_9834_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9834_end_0 = const()[name = string("op_9834_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9834_end_mask_0 = const()[name = string("op_9834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9834_cast_fp16 = slice_by_index(begin = var_9834_begin_0, end = var_9834_end_0, end_mask = var_9834_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9834_cast_fp16")]; + tensor var_9841_begin_0 = const()[name = string("op_9841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9841_end_0 = const()[name = string("op_9841_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9841_end_mask_0 = const()[name = string("op_9841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9841_cast_fp16 = slice_by_index(begin = var_9841_begin_0, end = var_9841_end_0, end_mask = var_9841_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9841_cast_fp16")]; + tensor var_9848_begin_0 = const()[name = string("op_9848_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9848_end_0 = const()[name = string("op_9848_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9848_end_mask_0 = const()[name = string("op_9848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9848_cast_fp16 = slice_by_index(begin = var_9848_begin_0, end = var_9848_end_0, end_mask = var_9848_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9848_cast_fp16")]; + tensor var_9855_begin_0 = const()[name = string("op_9855_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9855_end_0 = const()[name = string("op_9855_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9855_end_mask_0 = const()[name = string("op_9855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9855_cast_fp16 = slice_by_index(begin = var_9855_begin_0, end = var_9855_end_0, end_mask = var_9855_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9855_cast_fp16")]; + tensor var_9862_begin_0 = const()[name = string("op_9862_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9862_end_0 = const()[name = string("op_9862_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9862_end_mask_0 = const()[name = string("op_9862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9862_cast_fp16 = slice_by_index(begin = var_9862_begin_0, end = var_9862_end_0, end_mask = var_9862_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9862_cast_fp16")]; + tensor var_9869_begin_0 = const()[name = string("op_9869_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9869_end_0 = const()[name = string("op_9869_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9869_end_mask_0 = const()[name = string("op_9869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9869_cast_fp16 = slice_by_index(begin = var_9869_begin_0, end = var_9869_end_0, end_mask = var_9869_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9869_cast_fp16")]; + tensor var_9876_begin_0 = const()[name = string("op_9876_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9876_end_0 = const()[name = string("op_9876_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9876_end_mask_0 = const()[name = string("op_9876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9876_cast_fp16 = slice_by_index(begin = var_9876_begin_0, end = var_9876_end_0, end_mask = var_9876_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9876_cast_fp16")]; + tensor var_9883_begin_0 = const()[name = string("op_9883_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9883_end_0 = const()[name = string("op_9883_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9883_end_mask_0 = const()[name = string("op_9883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9883_cast_fp16 = slice_by_index(begin = var_9883_begin_0, end = var_9883_end_0, end_mask = var_9883_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9883_cast_fp16")]; + tensor var_9890_begin_0 = const()[name = string("op_9890_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9890_end_0 = const()[name = string("op_9890_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9890_end_mask_0 = const()[name = string("op_9890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9890_cast_fp16 = slice_by_index(begin = var_9890_begin_0, end = var_9890_end_0, end_mask = var_9890_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9890_cast_fp16")]; + tensor var_9897_begin_0 = const()[name = string("op_9897_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9897_end_0 = const()[name = string("op_9897_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9897_end_mask_0 = const()[name = string("op_9897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9897_cast_fp16 = slice_by_index(begin = var_9897_begin_0, end = var_9897_end_0, end_mask = var_9897_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9897_cast_fp16")]; + tensor var_9904_begin_0 = const()[name = string("op_9904_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9904_end_0 = const()[name = string("op_9904_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9904_end_mask_0 = const()[name = string("op_9904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9904_cast_fp16 = slice_by_index(begin = var_9904_begin_0, end = var_9904_end_0, end_mask = var_9904_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9904_cast_fp16")]; + tensor var_9911_begin_0 = const()[name = string("op_9911_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9911_end_0 = const()[name = string("op_9911_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9911_end_mask_0 = const()[name = string("op_9911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9911_cast_fp16 = slice_by_index(begin = var_9911_begin_0, end = var_9911_end_0, end_mask = var_9911_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9911_cast_fp16")]; + tensor var_9918_begin_0 = const()[name = string("op_9918_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9918_end_0 = const()[name = string("op_9918_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9918_end_mask_0 = const()[name = string("op_9918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9918_cast_fp16 = slice_by_index(begin = var_9918_begin_0, end = var_9918_end_0, end_mask = var_9918_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9918_cast_fp16")]; + tensor var_9925_begin_0 = const()[name = string("op_9925_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9925_end_0 = const()[name = string("op_9925_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9925_end_mask_0 = const()[name = string("op_9925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9925_cast_fp16 = slice_by_index(begin = var_9925_begin_0, end = var_9925_end_0, end_mask = var_9925_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9925_cast_fp16")]; + tensor var_9932_begin_0 = const()[name = string("op_9932_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9932_end_0 = const()[name = string("op_9932_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9932_end_mask_0 = const()[name = string("op_9932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9932_cast_fp16 = slice_by_index(begin = var_9932_begin_0, end = var_9932_end_0, end_mask = var_9932_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9932_cast_fp16")]; + tensor var_9939_begin_0 = const()[name = string("op_9939_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9939_end_0 = const()[name = string("op_9939_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9939_end_mask_0 = const()[name = string("op_9939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9939_cast_fp16 = slice_by_index(begin = var_9939_begin_0, end = var_9939_end_0, end_mask = var_9939_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9939_cast_fp16")]; + tensor var_9946_begin_0 = const()[name = string("op_9946_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9946_end_0 = const()[name = string("op_9946_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9946_end_mask_0 = const()[name = string("op_9946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9946_cast_fp16 = slice_by_index(begin = var_9946_begin_0, end = var_9946_end_0, end_mask = var_9946_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9946_cast_fp16")]; + tensor var_9953_begin_0 = const()[name = string("op_9953_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9953_end_0 = const()[name = string("op_9953_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9953_end_mask_0 = const()[name = string("op_9953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9953_cast_fp16 = slice_by_index(begin = var_9953_begin_0, end = var_9953_end_0, end_mask = var_9953_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9953_cast_fp16")]; + tensor var_9960_begin_0 = const()[name = string("op_9960_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9960_end_0 = const()[name = string("op_9960_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9960_end_mask_0 = const()[name = string("op_9960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9960_cast_fp16 = slice_by_index(begin = var_9960_begin_0, end = var_9960_end_0, end_mask = var_9960_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9960_cast_fp16")]; + tensor var_9967_begin_0 = const()[name = string("op_9967_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9967_end_0 = const()[name = string("op_9967_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9967_end_mask_0 = const()[name = string("op_9967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9967_cast_fp16 = slice_by_index(begin = var_9967_begin_0, end = var_9967_end_0, end_mask = var_9967_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9967_cast_fp16")]; + tensor var_9974_begin_0 = const()[name = string("op_9974_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9974_end_0 = const()[name = string("op_9974_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9974_end_mask_0 = const()[name = string("op_9974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9974_cast_fp16 = slice_by_index(begin = var_9974_begin_0, end = var_9974_end_0, end_mask = var_9974_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9974_cast_fp16")]; + tensor var_9981_begin_0 = const()[name = string("op_9981_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9981_end_0 = const()[name = string("op_9981_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9981_end_mask_0 = const()[name = string("op_9981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9981_cast_fp16 = slice_by_index(begin = var_9981_begin_0, end = var_9981_end_0, end_mask = var_9981_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9981_cast_fp16")]; + tensor var_9988_begin_0 = const()[name = string("op_9988_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9988_end_0 = const()[name = string("op_9988_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9988_end_mask_0 = const()[name = string("op_9988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9988_cast_fp16 = slice_by_index(begin = var_9988_begin_0, end = var_9988_end_0, end_mask = var_9988_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9988_cast_fp16")]; + tensor var_9995_begin_0 = const()[name = string("op_9995_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9995_end_0 = const()[name = string("op_9995_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9995_end_mask_0 = const()[name = string("op_9995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9995_cast_fp16 = slice_by_index(begin = var_9995_begin_0, end = var_9995_end_0, end_mask = var_9995_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9995_cast_fp16")]; + tensor var_10002_begin_0 = const()[name = string("op_10002_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10002_end_0 = const()[name = string("op_10002_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10002_end_mask_0 = const()[name = string("op_10002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10002_cast_fp16 = slice_by_index(begin = var_10002_begin_0, end = var_10002_end_0, end_mask = var_10002_end_mask_0, x = var_9488_cast_fp16)[name = string("op_10002_cast_fp16")]; + tensor var_10009_begin_0 = const()[name = string("op_10009_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10009_end_0 = const()[name = string("op_10009_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10009_end_mask_0 = const()[name = string("op_10009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10009_cast_fp16 = slice_by_index(begin = var_10009_begin_0, end = var_10009_end_0, end_mask = var_10009_end_mask_0, x = var_9492_cast_fp16)[name = string("op_10009_cast_fp16")]; + tensor var_10016_begin_0 = const()[name = string("op_10016_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10016_end_0 = const()[name = string("op_10016_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10016_end_mask_0 = const()[name = string("op_10016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10016_cast_fp16 = slice_by_index(begin = var_10016_begin_0, end = var_10016_end_0, end_mask = var_10016_end_mask_0, x = var_9492_cast_fp16)[name = string("op_10016_cast_fp16")]; + tensor var_10023_begin_0 = const()[name = string("op_10023_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10023_end_0 = const()[name = string("op_10023_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10023_end_mask_0 = const()[name = string("op_10023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10023_cast_fp16 = slice_by_index(begin = var_10023_begin_0, end = var_10023_end_0, end_mask = var_10023_end_mask_0, x = var_9492_cast_fp16)[name = string("op_10023_cast_fp16")]; + tensor var_10030_begin_0 = const()[name = string("op_10030_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10030_end_0 = const()[name = string("op_10030_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10030_end_mask_0 = const()[name = string("op_10030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10030_cast_fp16 = slice_by_index(begin = var_10030_begin_0, end = var_10030_end_0, end_mask = var_10030_end_mask_0, x = var_9492_cast_fp16)[name = string("op_10030_cast_fp16")]; + tensor var_10037_begin_0 = const()[name = string("op_10037_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10037_end_0 = const()[name = string("op_10037_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10037_end_mask_0 = const()[name = string("op_10037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10037_cast_fp16 = slice_by_index(begin = var_10037_begin_0, end = var_10037_end_0, end_mask = var_10037_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10037_cast_fp16")]; + tensor var_10044_begin_0 = const()[name = string("op_10044_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10044_end_0 = const()[name = string("op_10044_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10044_end_mask_0 = const()[name = string("op_10044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10044_cast_fp16 = slice_by_index(begin = var_10044_begin_0, end = var_10044_end_0, end_mask = var_10044_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10044_cast_fp16")]; + tensor var_10051_begin_0 = const()[name = string("op_10051_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10051_end_0 = const()[name = string("op_10051_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10051_end_mask_0 = const()[name = string("op_10051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10051_cast_fp16 = slice_by_index(begin = var_10051_begin_0, end = var_10051_end_0, end_mask = var_10051_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10051_cast_fp16")]; + tensor var_10058_begin_0 = const()[name = string("op_10058_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10058_end_0 = const()[name = string("op_10058_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10058_end_mask_0 = const()[name = string("op_10058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10058_cast_fp16 = slice_by_index(begin = var_10058_begin_0, end = var_10058_end_0, end_mask = var_10058_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10058_cast_fp16")]; + tensor k_13_perm_0 = const()[name = string("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10063_begin_0 = const()[name = string("op_10063_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10063_end_0 = const()[name = string("op_10063_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_10063_end_mask_0 = const()[name = string("op_10063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_13_cast_fp16 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = string("transpose_25")]; + tensor var_10063_cast_fp16 = slice_by_index(begin = var_10063_begin_0, end = var_10063_end_0, end_mask = var_10063_end_mask_0, x = k_13_cast_fp16)[name = string("op_10063_cast_fp16")]; + tensor var_10067_begin_0 = const()[name = string("op_10067_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10067_end_0 = const()[name = string("op_10067_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_10067_end_mask_0 = const()[name = string("op_10067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10067_cast_fp16 = slice_by_index(begin = var_10067_begin_0, end = var_10067_end_0, end_mask = var_10067_end_mask_0, x = k_13_cast_fp16)[name = string("op_10067_cast_fp16")]; + tensor var_10071_begin_0 = const()[name = string("op_10071_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10071_end_0 = const()[name = string("op_10071_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_10071_end_mask_0 = const()[name = string("op_10071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10071_cast_fp16 = slice_by_index(begin = var_10071_begin_0, end = var_10071_end_0, end_mask = var_10071_end_mask_0, x = k_13_cast_fp16)[name = string("op_10071_cast_fp16")]; + tensor var_10075_begin_0 = const()[name = string("op_10075_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10075_end_0 = const()[name = string("op_10075_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_10075_end_mask_0 = const()[name = string("op_10075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10075_cast_fp16 = slice_by_index(begin = var_10075_begin_0, end = var_10075_end_0, end_mask = var_10075_end_mask_0, x = k_13_cast_fp16)[name = string("op_10075_cast_fp16")]; + tensor var_10079_begin_0 = const()[name = string("op_10079_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10079_end_0 = const()[name = string("op_10079_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_10079_end_mask_0 = const()[name = string("op_10079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10079_cast_fp16 = slice_by_index(begin = var_10079_begin_0, end = var_10079_end_0, end_mask = var_10079_end_mask_0, x = k_13_cast_fp16)[name = string("op_10079_cast_fp16")]; + tensor var_10083_begin_0 = const()[name = string("op_10083_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10083_end_0 = const()[name = string("op_10083_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_10083_end_mask_0 = const()[name = string("op_10083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10083_cast_fp16 = slice_by_index(begin = var_10083_begin_0, end = var_10083_end_0, end_mask = var_10083_end_mask_0, x = k_13_cast_fp16)[name = string("op_10083_cast_fp16")]; + tensor var_10087_begin_0 = const()[name = string("op_10087_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10087_end_0 = const()[name = string("op_10087_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_10087_end_mask_0 = const()[name = string("op_10087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10087_cast_fp16 = slice_by_index(begin = var_10087_begin_0, end = var_10087_end_0, end_mask = var_10087_end_mask_0, x = k_13_cast_fp16)[name = string("op_10087_cast_fp16")]; + tensor var_10091_begin_0 = const()[name = string("op_10091_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10091_end_0 = const()[name = string("op_10091_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_10091_end_mask_0 = const()[name = string("op_10091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10091_cast_fp16 = slice_by_index(begin = var_10091_begin_0, end = var_10091_end_0, end_mask = var_10091_end_mask_0, x = k_13_cast_fp16)[name = string("op_10091_cast_fp16")]; + tensor var_10095_begin_0 = const()[name = string("op_10095_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10095_end_0 = const()[name = string("op_10095_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_10095_end_mask_0 = const()[name = string("op_10095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10095_cast_fp16 = slice_by_index(begin = var_10095_begin_0, end = var_10095_end_0, end_mask = var_10095_end_mask_0, x = k_13_cast_fp16)[name = string("op_10095_cast_fp16")]; + tensor var_10099_begin_0 = const()[name = string("op_10099_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10099_end_0 = const()[name = string("op_10099_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_10099_end_mask_0 = const()[name = string("op_10099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10099_cast_fp16 = slice_by_index(begin = var_10099_begin_0, end = var_10099_end_0, end_mask = var_10099_end_mask_0, x = k_13_cast_fp16)[name = string("op_10099_cast_fp16")]; + tensor var_10103_begin_0 = const()[name = string("op_10103_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10103_end_0 = const()[name = string("op_10103_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_10103_end_mask_0 = const()[name = string("op_10103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10103_cast_fp16 = slice_by_index(begin = var_10103_begin_0, end = var_10103_end_0, end_mask = var_10103_end_mask_0, x = k_13_cast_fp16)[name = string("op_10103_cast_fp16")]; + tensor var_10107_begin_0 = const()[name = string("op_10107_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10107_end_0 = const()[name = string("op_10107_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_10107_end_mask_0 = const()[name = string("op_10107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10107_cast_fp16 = slice_by_index(begin = var_10107_begin_0, end = var_10107_end_0, end_mask = var_10107_end_mask_0, x = k_13_cast_fp16)[name = string("op_10107_cast_fp16")]; + tensor var_10111_begin_0 = const()[name = string("op_10111_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10111_end_0 = const()[name = string("op_10111_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_10111_end_mask_0 = const()[name = string("op_10111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10111_cast_fp16 = slice_by_index(begin = var_10111_begin_0, end = var_10111_end_0, end_mask = var_10111_end_mask_0, x = k_13_cast_fp16)[name = string("op_10111_cast_fp16")]; + tensor var_10115_begin_0 = const()[name = string("op_10115_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10115_end_0 = const()[name = string("op_10115_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_10115_end_mask_0 = const()[name = string("op_10115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10115_cast_fp16 = slice_by_index(begin = var_10115_begin_0, end = var_10115_end_0, end_mask = var_10115_end_mask_0, x = k_13_cast_fp16)[name = string("op_10115_cast_fp16")]; + tensor var_10119_begin_0 = const()[name = string("op_10119_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10119_end_0 = const()[name = string("op_10119_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_10119_end_mask_0 = const()[name = string("op_10119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10119_cast_fp16 = slice_by_index(begin = var_10119_begin_0, end = var_10119_end_0, end_mask = var_10119_end_mask_0, x = k_13_cast_fp16)[name = string("op_10119_cast_fp16")]; + tensor var_10123_begin_0 = const()[name = string("op_10123_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10123_end_0 = const()[name = string("op_10123_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_10123_end_mask_0 = const()[name = string("op_10123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10123_cast_fp16 = slice_by_index(begin = var_10123_begin_0, end = var_10123_end_0, end_mask = var_10123_end_mask_0, x = k_13_cast_fp16)[name = string("op_10123_cast_fp16")]; + tensor var_10127_begin_0 = const()[name = string("op_10127_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10127_end_0 = const()[name = string("op_10127_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_10127_end_mask_0 = const()[name = string("op_10127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10127_cast_fp16 = slice_by_index(begin = var_10127_begin_0, end = var_10127_end_0, end_mask = var_10127_end_mask_0, x = k_13_cast_fp16)[name = string("op_10127_cast_fp16")]; + tensor var_10131_begin_0 = const()[name = string("op_10131_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10131_end_0 = const()[name = string("op_10131_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_10131_end_mask_0 = const()[name = string("op_10131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10131_cast_fp16 = slice_by_index(begin = var_10131_begin_0, end = var_10131_end_0, end_mask = var_10131_end_mask_0, x = k_13_cast_fp16)[name = string("op_10131_cast_fp16")]; + tensor var_10135_begin_0 = const()[name = string("op_10135_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10135_end_0 = const()[name = string("op_10135_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_10135_end_mask_0 = const()[name = string("op_10135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10135_cast_fp16 = slice_by_index(begin = var_10135_begin_0, end = var_10135_end_0, end_mask = var_10135_end_mask_0, x = k_13_cast_fp16)[name = string("op_10135_cast_fp16")]; + tensor var_10139_begin_0 = const()[name = string("op_10139_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10139_end_0 = const()[name = string("op_10139_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_10139_end_mask_0 = const()[name = string("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = k_13_cast_fp16)[name = string("op_10139_cast_fp16")]; + tensor var_10141_begin_0 = const()[name = string("op_10141_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10141_end_0 = const()[name = string("op_10141_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10141_end_mask_0 = const()[name = string("op_10141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10141_cast_fp16 = slice_by_index(begin = var_10141_begin_0, end = var_10141_end_0, end_mask = var_10141_end_mask_0, x = value_13_cast_fp16)[name = string("op_10141_cast_fp16")]; + tensor var_10145_begin_0 = const()[name = string("op_10145_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10145_end_0 = const()[name = string("op_10145_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10145_end_mask_0 = const()[name = string("op_10145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10145_cast_fp16 = slice_by_index(begin = var_10145_begin_0, end = var_10145_end_0, end_mask = var_10145_end_mask_0, x = value_13_cast_fp16)[name = string("op_10145_cast_fp16")]; + tensor var_10149_begin_0 = const()[name = string("op_10149_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10149_end_0 = const()[name = string("op_10149_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10149_end_mask_0 = const()[name = string("op_10149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10149_cast_fp16 = slice_by_index(begin = var_10149_begin_0, end = var_10149_end_0, end_mask = var_10149_end_mask_0, x = value_13_cast_fp16)[name = string("op_10149_cast_fp16")]; + tensor var_10153_begin_0 = const()[name = string("op_10153_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10153_end_0 = const()[name = string("op_10153_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10153_end_mask_0 = const()[name = string("op_10153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10153_cast_fp16 = slice_by_index(begin = var_10153_begin_0, end = var_10153_end_0, end_mask = var_10153_end_mask_0, x = value_13_cast_fp16)[name = string("op_10153_cast_fp16")]; + tensor var_10157_begin_0 = const()[name = string("op_10157_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10157_end_0 = const()[name = string("op_10157_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10157_end_mask_0 = const()[name = string("op_10157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10157_cast_fp16 = slice_by_index(begin = var_10157_begin_0, end = var_10157_end_0, end_mask = var_10157_end_mask_0, x = value_13_cast_fp16)[name = string("op_10157_cast_fp16")]; + tensor var_10161_begin_0 = const()[name = string("op_10161_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10161_end_0 = const()[name = string("op_10161_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10161_end_mask_0 = const()[name = string("op_10161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10161_cast_fp16 = slice_by_index(begin = var_10161_begin_0, end = var_10161_end_0, end_mask = var_10161_end_mask_0, x = value_13_cast_fp16)[name = string("op_10161_cast_fp16")]; + tensor var_10165_begin_0 = const()[name = string("op_10165_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10165_end_0 = const()[name = string("op_10165_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10165_end_mask_0 = const()[name = string("op_10165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10165_cast_fp16 = slice_by_index(begin = var_10165_begin_0, end = var_10165_end_0, end_mask = var_10165_end_mask_0, x = value_13_cast_fp16)[name = string("op_10165_cast_fp16")]; + tensor var_10169_begin_0 = const()[name = string("op_10169_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10169_end_0 = const()[name = string("op_10169_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10169_end_mask_0 = const()[name = string("op_10169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10169_cast_fp16 = slice_by_index(begin = var_10169_begin_0, end = var_10169_end_0, end_mask = var_10169_end_mask_0, x = value_13_cast_fp16)[name = string("op_10169_cast_fp16")]; + tensor var_10173_begin_0 = const()[name = string("op_10173_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10173_end_0 = const()[name = string("op_10173_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10173_end_mask_0 = const()[name = string("op_10173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10173_cast_fp16 = slice_by_index(begin = var_10173_begin_0, end = var_10173_end_0, end_mask = var_10173_end_mask_0, x = value_13_cast_fp16)[name = string("op_10173_cast_fp16")]; + tensor var_10177_begin_0 = const()[name = string("op_10177_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10177_end_0 = const()[name = string("op_10177_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10177_end_mask_0 = const()[name = string("op_10177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10177_cast_fp16 = slice_by_index(begin = var_10177_begin_0, end = var_10177_end_0, end_mask = var_10177_end_mask_0, x = value_13_cast_fp16)[name = string("op_10177_cast_fp16")]; + tensor var_10181_begin_0 = const()[name = string("op_10181_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10181_end_0 = const()[name = string("op_10181_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10181_end_mask_0 = const()[name = string("op_10181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10181_cast_fp16 = slice_by_index(begin = var_10181_begin_0, end = var_10181_end_0, end_mask = var_10181_end_mask_0, x = value_13_cast_fp16)[name = string("op_10181_cast_fp16")]; + tensor var_10185_begin_0 = const()[name = string("op_10185_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10185_end_0 = const()[name = string("op_10185_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10185_end_mask_0 = const()[name = string("op_10185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10185_cast_fp16 = slice_by_index(begin = var_10185_begin_0, end = var_10185_end_0, end_mask = var_10185_end_mask_0, x = value_13_cast_fp16)[name = string("op_10185_cast_fp16")]; + tensor var_10189_begin_0 = const()[name = string("op_10189_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10189_end_0 = const()[name = string("op_10189_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_10189_end_mask_0 = const()[name = string("op_10189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10189_cast_fp16 = slice_by_index(begin = var_10189_begin_0, end = var_10189_end_0, end_mask = var_10189_end_mask_0, x = value_13_cast_fp16)[name = string("op_10189_cast_fp16")]; + tensor var_10193_begin_0 = const()[name = string("op_10193_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10193_end_0 = const()[name = string("op_10193_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_10193_end_mask_0 = const()[name = string("op_10193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10193_cast_fp16 = slice_by_index(begin = var_10193_begin_0, end = var_10193_end_0, end_mask = var_10193_end_mask_0, x = value_13_cast_fp16)[name = string("op_10193_cast_fp16")]; + tensor var_10197_begin_0 = const()[name = string("op_10197_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10197_end_0 = const()[name = string("op_10197_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_10197_end_mask_0 = const()[name = string("op_10197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10197_cast_fp16 = slice_by_index(begin = var_10197_begin_0, end = var_10197_end_0, end_mask = var_10197_end_mask_0, x = value_13_cast_fp16)[name = string("op_10197_cast_fp16")]; + tensor var_10201_begin_0 = const()[name = string("op_10201_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10201_end_0 = const()[name = string("op_10201_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_10201_end_mask_0 = const()[name = string("op_10201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10201_cast_fp16 = slice_by_index(begin = var_10201_begin_0, end = var_10201_end_0, end_mask = var_10201_end_mask_0, x = value_13_cast_fp16)[name = string("op_10201_cast_fp16")]; + tensor var_10205_begin_0 = const()[name = string("op_10205_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10205_end_0 = const()[name = string("op_10205_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_10205_end_mask_0 = const()[name = string("op_10205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10205_cast_fp16 = slice_by_index(begin = var_10205_begin_0, end = var_10205_end_0, end_mask = var_10205_end_mask_0, x = value_13_cast_fp16)[name = string("op_10205_cast_fp16")]; + tensor var_10209_begin_0 = const()[name = string("op_10209_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10209_end_0 = const()[name = string("op_10209_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_10209_end_mask_0 = const()[name = string("op_10209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10209_cast_fp16 = slice_by_index(begin = var_10209_begin_0, end = var_10209_end_0, end_mask = var_10209_end_mask_0, x = value_13_cast_fp16)[name = string("op_10209_cast_fp16")]; + tensor var_10213_begin_0 = const()[name = string("op_10213_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10213_end_0 = const()[name = string("op_10213_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_10213_end_mask_0 = const()[name = string("op_10213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10213_cast_fp16 = slice_by_index(begin = var_10213_begin_0, end = var_10213_end_0, end_mask = var_10213_end_mask_0, x = value_13_cast_fp16)[name = string("op_10213_cast_fp16")]; + tensor var_10217_begin_0 = const()[name = string("op_10217_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10217_end_0 = const()[name = string("op_10217_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_10217_end_mask_0 = const()[name = string("op_10217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10217_cast_fp16 = slice_by_index(begin = var_10217_begin_0, end = var_10217_end_0, end_mask = var_10217_end_mask_0, x = value_13_cast_fp16)[name = string("op_10217_cast_fp16")]; + string _SplitHeadsQ__mh_w_961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_961_equation_0, values = (var_10063_cast_fp16, var_9505_cast_fp16))[name = string("_SplitHeadsQ__mh_w_961_cast_fp16")]; + string _SplitHeadsQ__mh_w_963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_963_equation_0, values = (var_10063_cast_fp16, var_9512_cast_fp16))[name = string("_SplitHeadsQ__mh_w_963_cast_fp16")]; + string _SplitHeadsQ__mh_w_965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_965_equation_0, values = (var_10063_cast_fp16, var_9519_cast_fp16))[name = string("_SplitHeadsQ__mh_w_965_cast_fp16")]; + string _SplitHeadsQ__mh_w_967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_967_equation_0, values = (var_10063_cast_fp16, var_9526_cast_fp16))[name = string("_SplitHeadsQ__mh_w_967_cast_fp16")]; + string _SplitHeadsQ__mh_w_969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_969_equation_0, values = (var_10067_cast_fp16, var_9533_cast_fp16))[name = string("_SplitHeadsQ__mh_w_969_cast_fp16")]; + string _SplitHeadsQ__mh_w_971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_971_equation_0, values = (var_10067_cast_fp16, var_9540_cast_fp16))[name = string("_SplitHeadsQ__mh_w_971_cast_fp16")]; + string _SplitHeadsQ__mh_w_973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_973_equation_0, values = (var_10067_cast_fp16, var_9547_cast_fp16))[name = string("_SplitHeadsQ__mh_w_973_cast_fp16")]; + string _SplitHeadsQ__mh_w_975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_975_equation_0, values = (var_10067_cast_fp16, var_9554_cast_fp16))[name = string("_SplitHeadsQ__mh_w_975_cast_fp16")]; + string _SplitHeadsQ__mh_w_977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_977_equation_0, values = (var_10071_cast_fp16, var_9561_cast_fp16))[name = string("_SplitHeadsQ__mh_w_977_cast_fp16")]; + string _SplitHeadsQ__mh_w_979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_979_equation_0, values = (var_10071_cast_fp16, var_9568_cast_fp16))[name = string("_SplitHeadsQ__mh_w_979_cast_fp16")]; + string _SplitHeadsQ__mh_w_981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_981_equation_0, values = (var_10071_cast_fp16, var_9575_cast_fp16))[name = string("_SplitHeadsQ__mh_w_981_cast_fp16")]; + string _SplitHeadsQ__mh_w_983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_983_equation_0, values = (var_10071_cast_fp16, var_9582_cast_fp16))[name = string("_SplitHeadsQ__mh_w_983_cast_fp16")]; + string _SplitHeadsQ__mh_w_985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_985_equation_0, values = (var_10075_cast_fp16, var_9589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_985_cast_fp16")]; + string _SplitHeadsQ__mh_w_987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_987_equation_0, values = (var_10075_cast_fp16, var_9596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_987_cast_fp16")]; + string _SplitHeadsQ__mh_w_989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_989_equation_0, values = (var_10075_cast_fp16, var_9603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_989_cast_fp16")]; + string _SplitHeadsQ__mh_w_991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_991_equation_0, values = (var_10075_cast_fp16, var_9610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_991_cast_fp16")]; + string _SplitHeadsQ__mh_w_993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_993_equation_0, values = (var_10079_cast_fp16, var_9617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_993_cast_fp16")]; + string _SplitHeadsQ__mh_w_995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_995_equation_0, values = (var_10079_cast_fp16, var_9624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_995_cast_fp16")]; + string _SplitHeadsQ__mh_w_997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_997_equation_0, values = (var_10079_cast_fp16, var_9631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_997_cast_fp16")]; + string _SplitHeadsQ__mh_w_999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_999_equation_0, values = (var_10079_cast_fp16, var_9638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_999_cast_fp16")]; + string _SplitHeadsQ__mh_w_1001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1001_equation_0, values = (var_10083_cast_fp16, var_9645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1001_cast_fp16")]; + string _SplitHeadsQ__mh_w_1003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1003_equation_0, values = (var_10083_cast_fp16, var_9652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1003_cast_fp16")]; + string _SplitHeadsQ__mh_w_1005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1005_equation_0, values = (var_10083_cast_fp16, var_9659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1005_cast_fp16")]; + string _SplitHeadsQ__mh_w_1007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1007_equation_0, values = (var_10083_cast_fp16, var_9666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1007_cast_fp16")]; + string _SplitHeadsQ__mh_w_1009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1009_equation_0, values = (var_10087_cast_fp16, var_9673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1009_cast_fp16")]; + string _SplitHeadsQ__mh_w_1011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1011_equation_0, values = (var_10087_cast_fp16, var_9680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1011_cast_fp16")]; + string _SplitHeadsQ__mh_w_1013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1013_equation_0, values = (var_10087_cast_fp16, var_9687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1013_cast_fp16")]; + string _SplitHeadsQ__mh_w_1015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1015_equation_0, values = (var_10087_cast_fp16, var_9694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1015_cast_fp16")]; + string _SplitHeadsQ__mh_w_1017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1017_equation_0, values = (var_10091_cast_fp16, var_9701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1017_cast_fp16")]; + string _SplitHeadsQ__mh_w_1019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1019_equation_0, values = (var_10091_cast_fp16, var_9708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1019_cast_fp16")]; + string _SplitHeadsQ__mh_w_1021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1021_equation_0, values = (var_10091_cast_fp16, var_9715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1021_cast_fp16")]; + string _SplitHeadsQ__mh_w_1023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1023_equation_0, values = (var_10091_cast_fp16, var_9722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1023_cast_fp16")]; + string _SplitHeadsQ__mh_w_1025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1025_equation_0, values = (var_10095_cast_fp16, var_9729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1025_cast_fp16")]; + string _SplitHeadsQ__mh_w_1027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1027_equation_0, values = (var_10095_cast_fp16, var_9736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1027_cast_fp16")]; + string _SplitHeadsQ__mh_w_1029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1029_equation_0, values = (var_10095_cast_fp16, var_9743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1029_cast_fp16")]; + string _SplitHeadsQ__mh_w_1031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1031_equation_0, values = (var_10095_cast_fp16, var_9750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1031_cast_fp16")]; + string _SplitHeadsQ__mh_w_1033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1033_equation_0, values = (var_10099_cast_fp16, var_9757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1033_cast_fp16")]; + string _SplitHeadsQ__mh_w_1035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1035_equation_0, values = (var_10099_cast_fp16, var_9764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1035_cast_fp16")]; + string _SplitHeadsQ__mh_w_1037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1037_equation_0, values = (var_10099_cast_fp16, var_9771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1037_cast_fp16")]; + string _SplitHeadsQ__mh_w_1039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1039_equation_0, values = (var_10099_cast_fp16, var_9778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1039_cast_fp16")]; + string _SplitHeadsQ__mh_w_1041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1041_equation_0, values = (var_10103_cast_fp16, var_9785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1041_cast_fp16")]; + string _SplitHeadsQ__mh_w_1043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1043_equation_0, values = (var_10103_cast_fp16, var_9792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1043_cast_fp16")]; + string _SplitHeadsQ__mh_w_1045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1045_equation_0, values = (var_10103_cast_fp16, var_9799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1045_cast_fp16")]; + string _SplitHeadsQ__mh_w_1047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1047_equation_0, values = (var_10103_cast_fp16, var_9806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1047_cast_fp16")]; + string _SplitHeadsQ__mh_w_1049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1049_equation_0, values = (var_10107_cast_fp16, var_9813_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1049_cast_fp16")]; + string _SplitHeadsQ__mh_w_1051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1051_equation_0, values = (var_10107_cast_fp16, var_9820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1051_cast_fp16")]; + string _SplitHeadsQ__mh_w_1053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1053_equation_0, values = (var_10107_cast_fp16, var_9827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1053_cast_fp16")]; + string _SplitHeadsQ__mh_w_1055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1055_equation_0, values = (var_10107_cast_fp16, var_9834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1055_cast_fp16")]; + string _SplitHeadsQ__mh_w_1057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1057_equation_0, values = (var_10111_cast_fp16, var_9841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1057_cast_fp16")]; + string _SplitHeadsQ__mh_w_1059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1059_equation_0, values = (var_10111_cast_fp16, var_9848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1059_cast_fp16")]; + string _SplitHeadsQ__mh_w_1061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1061_equation_0, values = (var_10111_cast_fp16, var_9855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1061_cast_fp16")]; + string _SplitHeadsQ__mh_w_1063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1063_equation_0, values = (var_10111_cast_fp16, var_9862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1063_cast_fp16")]; + string _SplitHeadsQ__mh_w_1065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1065_equation_0, values = (var_10115_cast_fp16, var_9869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1065_cast_fp16")]; + string _SplitHeadsQ__mh_w_1067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1067_equation_0, values = (var_10115_cast_fp16, var_9876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1067_cast_fp16")]; + string _SplitHeadsQ__mh_w_1069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1069_equation_0, values = (var_10115_cast_fp16, var_9883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1069_cast_fp16")]; + string _SplitHeadsQ__mh_w_1071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1071_equation_0, values = (var_10115_cast_fp16, var_9890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1071_cast_fp16")]; + string _SplitHeadsQ__mh_w_1073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1073_equation_0, values = (var_10119_cast_fp16, var_9897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1073_cast_fp16")]; + string _SplitHeadsQ__mh_w_1075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1075_equation_0, values = (var_10119_cast_fp16, var_9904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1075_cast_fp16")]; + string _SplitHeadsQ__mh_w_1077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1077_equation_0, values = (var_10119_cast_fp16, var_9911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1077_cast_fp16")]; + string _SplitHeadsQ__mh_w_1079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1079_equation_0, values = (var_10119_cast_fp16, var_9918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1079_cast_fp16")]; + string _SplitHeadsQ__mh_w_1081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1081_equation_0, values = (var_10123_cast_fp16, var_9925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1081_cast_fp16")]; + string _SplitHeadsQ__mh_w_1083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1083_equation_0, values = (var_10123_cast_fp16, var_9932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1083_cast_fp16")]; + string _SplitHeadsQ__mh_w_1085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1085_equation_0, values = (var_10123_cast_fp16, var_9939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1085_cast_fp16")]; + string _SplitHeadsQ__mh_w_1087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1087_equation_0, values = (var_10123_cast_fp16, var_9946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1087_cast_fp16")]; + string _SplitHeadsQ__mh_w_1089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1089_equation_0, values = (var_10127_cast_fp16, var_9953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1089_cast_fp16")]; + string _SplitHeadsQ__mh_w_1091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1091_equation_0, values = (var_10127_cast_fp16, var_9960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1091_cast_fp16")]; + string _SplitHeadsQ__mh_w_1093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1093_equation_0, values = (var_10127_cast_fp16, var_9967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1093_cast_fp16")]; + string _SplitHeadsQ__mh_w_1095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1095_equation_0, values = (var_10127_cast_fp16, var_9974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1095_cast_fp16")]; + string _SplitHeadsQ__mh_w_1097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1097_equation_0, values = (var_10131_cast_fp16, var_9981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1097_cast_fp16")]; + string _SplitHeadsQ__mh_w_1099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1099_equation_0, values = (var_10131_cast_fp16, var_9988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1099_cast_fp16")]; + string _SplitHeadsQ__mh_w_1101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1101_equation_0, values = (var_10131_cast_fp16, var_9995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1101_cast_fp16")]; + string _SplitHeadsQ__mh_w_1103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1103_equation_0, values = (var_10131_cast_fp16, var_10002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1103_cast_fp16")]; + string _SplitHeadsQ__mh_w_1105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1105_equation_0, values = (var_10135_cast_fp16, var_10009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1105_cast_fp16")]; + string _SplitHeadsQ__mh_w_1107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1107_equation_0, values = (var_10135_cast_fp16, var_10016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1107_cast_fp16")]; + string _SplitHeadsQ__mh_w_1109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1109_equation_0, values = (var_10135_cast_fp16, var_10023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1109_cast_fp16")]; + string _SplitHeadsQ__mh_w_1111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1111_equation_0, values = (var_10135_cast_fp16, var_10030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1111_cast_fp16")]; + string _SplitHeadsQ__mh_w_1113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1113_equation_0, values = (var_10139_cast_fp16, var_10037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1113_cast_fp16")]; + string _SplitHeadsQ__mh_w_1115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1115_equation_0, values = (var_10139_cast_fp16, var_10044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1115_cast_fp16")]; + string _SplitHeadsQ__mh_w_1117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1117_equation_0, values = (var_10139_cast_fp16, var_10051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1117_cast_fp16")]; + string _SplitHeadsQ__mh_w_1119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1119_equation_0, values = (var_10139_cast_fp16, var_10058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1119_cast_fp16")]; + fp16 var_10380_to_fp16 = const()[name = string("op_10380_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_961_cast_fp16, y = var_10380_to_fp16)[name = string("aw_chunk_961_cast_fp16")]; + fp16 var_10382_to_fp16 = const()[name = string("op_10382_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_963_cast_fp16, y = var_10382_to_fp16)[name = string("aw_chunk_963_cast_fp16")]; + fp16 var_10384_to_fp16 = const()[name = string("op_10384_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_965_cast_fp16, y = var_10384_to_fp16)[name = string("aw_chunk_965_cast_fp16")]; + fp16 var_10386_to_fp16 = const()[name = string("op_10386_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_967_cast_fp16, y = var_10386_to_fp16)[name = string("aw_chunk_967_cast_fp16")]; + fp16 var_10388_to_fp16 = const()[name = string("op_10388_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_969_cast_fp16, y = var_10388_to_fp16)[name = string("aw_chunk_969_cast_fp16")]; + fp16 var_10390_to_fp16 = const()[name = string("op_10390_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_971_cast_fp16, y = var_10390_to_fp16)[name = string("aw_chunk_971_cast_fp16")]; + fp16 var_10392_to_fp16 = const()[name = string("op_10392_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_973_cast_fp16, y = var_10392_to_fp16)[name = string("aw_chunk_973_cast_fp16")]; + fp16 var_10394_to_fp16 = const()[name = string("op_10394_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_975_cast_fp16, y = var_10394_to_fp16)[name = string("aw_chunk_975_cast_fp16")]; + fp16 var_10396_to_fp16 = const()[name = string("op_10396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_977_cast_fp16, y = var_10396_to_fp16)[name = string("aw_chunk_977_cast_fp16")]; + fp16 var_10398_to_fp16 = const()[name = string("op_10398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_979_cast_fp16, y = var_10398_to_fp16)[name = string("aw_chunk_979_cast_fp16")]; + fp16 var_10400_to_fp16 = const()[name = string("op_10400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_981_cast_fp16, y = var_10400_to_fp16)[name = string("aw_chunk_981_cast_fp16")]; + fp16 var_10402_to_fp16 = const()[name = string("op_10402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_983_cast_fp16, y = var_10402_to_fp16)[name = string("aw_chunk_983_cast_fp16")]; + fp16 var_10404_to_fp16 = const()[name = string("op_10404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_985_cast_fp16, y = var_10404_to_fp16)[name = string("aw_chunk_985_cast_fp16")]; + fp16 var_10406_to_fp16 = const()[name = string("op_10406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_987_cast_fp16, y = var_10406_to_fp16)[name = string("aw_chunk_987_cast_fp16")]; + fp16 var_10408_to_fp16 = const()[name = string("op_10408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_989_cast_fp16, y = var_10408_to_fp16)[name = string("aw_chunk_989_cast_fp16")]; + fp16 var_10410_to_fp16 = const()[name = string("op_10410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_991_cast_fp16, y = var_10410_to_fp16)[name = string("aw_chunk_991_cast_fp16")]; + fp16 var_10412_to_fp16 = const()[name = string("op_10412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_993_cast_fp16, y = var_10412_to_fp16)[name = string("aw_chunk_993_cast_fp16")]; + fp16 var_10414_to_fp16 = const()[name = string("op_10414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_995_cast_fp16, y = var_10414_to_fp16)[name = string("aw_chunk_995_cast_fp16")]; + fp16 var_10416_to_fp16 = const()[name = string("op_10416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_997_cast_fp16, y = var_10416_to_fp16)[name = string("aw_chunk_997_cast_fp16")]; + fp16 var_10418_to_fp16 = const()[name = string("op_10418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_999_cast_fp16, y = var_10418_to_fp16)[name = string("aw_chunk_999_cast_fp16")]; + fp16 var_10420_to_fp16 = const()[name = string("op_10420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1001_cast_fp16, y = var_10420_to_fp16)[name = string("aw_chunk_1001_cast_fp16")]; + fp16 var_10422_to_fp16 = const()[name = string("op_10422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1003_cast_fp16, y = var_10422_to_fp16)[name = string("aw_chunk_1003_cast_fp16")]; + fp16 var_10424_to_fp16 = const()[name = string("op_10424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1005_cast_fp16, y = var_10424_to_fp16)[name = string("aw_chunk_1005_cast_fp16")]; + fp16 var_10426_to_fp16 = const()[name = string("op_10426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1007_cast_fp16, y = var_10426_to_fp16)[name = string("aw_chunk_1007_cast_fp16")]; + fp16 var_10428_to_fp16 = const()[name = string("op_10428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1009_cast_fp16, y = var_10428_to_fp16)[name = string("aw_chunk_1009_cast_fp16")]; + fp16 var_10430_to_fp16 = const()[name = string("op_10430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1011_cast_fp16, y = var_10430_to_fp16)[name = string("aw_chunk_1011_cast_fp16")]; + fp16 var_10432_to_fp16 = const()[name = string("op_10432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1013_cast_fp16, y = var_10432_to_fp16)[name = string("aw_chunk_1013_cast_fp16")]; + fp16 var_10434_to_fp16 = const()[name = string("op_10434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1015_cast_fp16, y = var_10434_to_fp16)[name = string("aw_chunk_1015_cast_fp16")]; + fp16 var_10436_to_fp16 = const()[name = string("op_10436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1017_cast_fp16, y = var_10436_to_fp16)[name = string("aw_chunk_1017_cast_fp16")]; + fp16 var_10438_to_fp16 = const()[name = string("op_10438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1019_cast_fp16, y = var_10438_to_fp16)[name = string("aw_chunk_1019_cast_fp16")]; + fp16 var_10440_to_fp16 = const()[name = string("op_10440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1021_cast_fp16, y = var_10440_to_fp16)[name = string("aw_chunk_1021_cast_fp16")]; + fp16 var_10442_to_fp16 = const()[name = string("op_10442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1023_cast_fp16, y = var_10442_to_fp16)[name = string("aw_chunk_1023_cast_fp16")]; + fp16 var_10444_to_fp16 = const()[name = string("op_10444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1025_cast_fp16, y = var_10444_to_fp16)[name = string("aw_chunk_1025_cast_fp16")]; + fp16 var_10446_to_fp16 = const()[name = string("op_10446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1027_cast_fp16, y = var_10446_to_fp16)[name = string("aw_chunk_1027_cast_fp16")]; + fp16 var_10448_to_fp16 = const()[name = string("op_10448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1029_cast_fp16, y = var_10448_to_fp16)[name = string("aw_chunk_1029_cast_fp16")]; + fp16 var_10450_to_fp16 = const()[name = string("op_10450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1031_cast_fp16, y = var_10450_to_fp16)[name = string("aw_chunk_1031_cast_fp16")]; + fp16 var_10452_to_fp16 = const()[name = string("op_10452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1033_cast_fp16, y = var_10452_to_fp16)[name = string("aw_chunk_1033_cast_fp16")]; + fp16 var_10454_to_fp16 = const()[name = string("op_10454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1035_cast_fp16, y = var_10454_to_fp16)[name = string("aw_chunk_1035_cast_fp16")]; + fp16 var_10456_to_fp16 = const()[name = string("op_10456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1037_cast_fp16, y = var_10456_to_fp16)[name = string("aw_chunk_1037_cast_fp16")]; + fp16 var_10458_to_fp16 = const()[name = string("op_10458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1039_cast_fp16, y = var_10458_to_fp16)[name = string("aw_chunk_1039_cast_fp16")]; + fp16 var_10460_to_fp16 = const()[name = string("op_10460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1041_cast_fp16, y = var_10460_to_fp16)[name = string("aw_chunk_1041_cast_fp16")]; + fp16 var_10462_to_fp16 = const()[name = string("op_10462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1043_cast_fp16, y = var_10462_to_fp16)[name = string("aw_chunk_1043_cast_fp16")]; + fp16 var_10464_to_fp16 = const()[name = string("op_10464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1045_cast_fp16, y = var_10464_to_fp16)[name = string("aw_chunk_1045_cast_fp16")]; + fp16 var_10466_to_fp16 = const()[name = string("op_10466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1047_cast_fp16, y = var_10466_to_fp16)[name = string("aw_chunk_1047_cast_fp16")]; + fp16 var_10468_to_fp16 = const()[name = string("op_10468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1049_cast_fp16, y = var_10468_to_fp16)[name = string("aw_chunk_1049_cast_fp16")]; + fp16 var_10470_to_fp16 = const()[name = string("op_10470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1051_cast_fp16, y = var_10470_to_fp16)[name = string("aw_chunk_1051_cast_fp16")]; + fp16 var_10472_to_fp16 = const()[name = string("op_10472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1053_cast_fp16, y = var_10472_to_fp16)[name = string("aw_chunk_1053_cast_fp16")]; + fp16 var_10474_to_fp16 = const()[name = string("op_10474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1055_cast_fp16, y = var_10474_to_fp16)[name = string("aw_chunk_1055_cast_fp16")]; + fp16 var_10476_to_fp16 = const()[name = string("op_10476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1057_cast_fp16, y = var_10476_to_fp16)[name = string("aw_chunk_1057_cast_fp16")]; + fp16 var_10478_to_fp16 = const()[name = string("op_10478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1059_cast_fp16, y = var_10478_to_fp16)[name = string("aw_chunk_1059_cast_fp16")]; + fp16 var_10480_to_fp16 = const()[name = string("op_10480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1061_cast_fp16, y = var_10480_to_fp16)[name = string("aw_chunk_1061_cast_fp16")]; + fp16 var_10482_to_fp16 = const()[name = string("op_10482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1063_cast_fp16, y = var_10482_to_fp16)[name = string("aw_chunk_1063_cast_fp16")]; + fp16 var_10484_to_fp16 = const()[name = string("op_10484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1065_cast_fp16, y = var_10484_to_fp16)[name = string("aw_chunk_1065_cast_fp16")]; + fp16 var_10486_to_fp16 = const()[name = string("op_10486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1067_cast_fp16, y = var_10486_to_fp16)[name = string("aw_chunk_1067_cast_fp16")]; + fp16 var_10488_to_fp16 = const()[name = string("op_10488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1069_cast_fp16, y = var_10488_to_fp16)[name = string("aw_chunk_1069_cast_fp16")]; + fp16 var_10490_to_fp16 = const()[name = string("op_10490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1071_cast_fp16, y = var_10490_to_fp16)[name = string("aw_chunk_1071_cast_fp16")]; + fp16 var_10492_to_fp16 = const()[name = string("op_10492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1073_cast_fp16, y = var_10492_to_fp16)[name = string("aw_chunk_1073_cast_fp16")]; + fp16 var_10494_to_fp16 = const()[name = string("op_10494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1075_cast_fp16, y = var_10494_to_fp16)[name = string("aw_chunk_1075_cast_fp16")]; + fp16 var_10496_to_fp16 = const()[name = string("op_10496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1077_cast_fp16, y = var_10496_to_fp16)[name = string("aw_chunk_1077_cast_fp16")]; + fp16 var_10498_to_fp16 = const()[name = string("op_10498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1079_cast_fp16, y = var_10498_to_fp16)[name = string("aw_chunk_1079_cast_fp16")]; + fp16 var_10500_to_fp16 = const()[name = string("op_10500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1081_cast_fp16, y = var_10500_to_fp16)[name = string("aw_chunk_1081_cast_fp16")]; + fp16 var_10502_to_fp16 = const()[name = string("op_10502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1083_cast_fp16, y = var_10502_to_fp16)[name = string("aw_chunk_1083_cast_fp16")]; + fp16 var_10504_to_fp16 = const()[name = string("op_10504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1085_cast_fp16, y = var_10504_to_fp16)[name = string("aw_chunk_1085_cast_fp16")]; + fp16 var_10506_to_fp16 = const()[name = string("op_10506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1087_cast_fp16, y = var_10506_to_fp16)[name = string("aw_chunk_1087_cast_fp16")]; + fp16 var_10508_to_fp16 = const()[name = string("op_10508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1089_cast_fp16, y = var_10508_to_fp16)[name = string("aw_chunk_1089_cast_fp16")]; + fp16 var_10510_to_fp16 = const()[name = string("op_10510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1091_cast_fp16, y = var_10510_to_fp16)[name = string("aw_chunk_1091_cast_fp16")]; + fp16 var_10512_to_fp16 = const()[name = string("op_10512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1093_cast_fp16, y = var_10512_to_fp16)[name = string("aw_chunk_1093_cast_fp16")]; + fp16 var_10514_to_fp16 = const()[name = string("op_10514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1095_cast_fp16, y = var_10514_to_fp16)[name = string("aw_chunk_1095_cast_fp16")]; + fp16 var_10516_to_fp16 = const()[name = string("op_10516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1097_cast_fp16, y = var_10516_to_fp16)[name = string("aw_chunk_1097_cast_fp16")]; + fp16 var_10518_to_fp16 = const()[name = string("op_10518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1099_cast_fp16, y = var_10518_to_fp16)[name = string("aw_chunk_1099_cast_fp16")]; + fp16 var_10520_to_fp16 = const()[name = string("op_10520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1101_cast_fp16, y = var_10520_to_fp16)[name = string("aw_chunk_1101_cast_fp16")]; + fp16 var_10522_to_fp16 = const()[name = string("op_10522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1103_cast_fp16, y = var_10522_to_fp16)[name = string("aw_chunk_1103_cast_fp16")]; + fp16 var_10524_to_fp16 = const()[name = string("op_10524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1105_cast_fp16, y = var_10524_to_fp16)[name = string("aw_chunk_1105_cast_fp16")]; + fp16 var_10526_to_fp16 = const()[name = string("op_10526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1107_cast_fp16, y = var_10526_to_fp16)[name = string("aw_chunk_1107_cast_fp16")]; + fp16 var_10528_to_fp16 = const()[name = string("op_10528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1109_cast_fp16, y = var_10528_to_fp16)[name = string("aw_chunk_1109_cast_fp16")]; + fp16 var_10530_to_fp16 = const()[name = string("op_10530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1111_cast_fp16, y = var_10530_to_fp16)[name = string("aw_chunk_1111_cast_fp16")]; + fp16 var_10532_to_fp16 = const()[name = string("op_10532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1113_cast_fp16, y = var_10532_to_fp16)[name = string("aw_chunk_1113_cast_fp16")]; + fp16 var_10534_to_fp16 = const()[name = string("op_10534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1115_cast_fp16, y = var_10534_to_fp16)[name = string("aw_chunk_1115_cast_fp16")]; + fp16 var_10536_to_fp16 = const()[name = string("op_10536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1117_cast_fp16, y = var_10536_to_fp16)[name = string("aw_chunk_1117_cast_fp16")]; + fp16 var_10538_to_fp16 = const()[name = string("op_10538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1119_cast_fp16, y = var_10538_to_fp16)[name = string("aw_chunk_1119_cast_fp16")]; + tensor var_10540_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_961_cast_fp16)[name = string("op_10540_cast_fp16")]; + tensor var_10541_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_963_cast_fp16)[name = string("op_10541_cast_fp16")]; + tensor var_10542_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_965_cast_fp16)[name = string("op_10542_cast_fp16")]; + tensor var_10543_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_967_cast_fp16)[name = string("op_10543_cast_fp16")]; + tensor var_10544_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_969_cast_fp16)[name = string("op_10544_cast_fp16")]; + tensor var_10545_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_971_cast_fp16)[name = string("op_10545_cast_fp16")]; + tensor var_10546_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_973_cast_fp16)[name = string("op_10546_cast_fp16")]; + tensor var_10547_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_975_cast_fp16)[name = string("op_10547_cast_fp16")]; + tensor var_10548_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_977_cast_fp16)[name = string("op_10548_cast_fp16")]; + tensor var_10549_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_979_cast_fp16)[name = string("op_10549_cast_fp16")]; + tensor var_10550_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_981_cast_fp16)[name = string("op_10550_cast_fp16")]; + tensor var_10551_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_983_cast_fp16)[name = string("op_10551_cast_fp16")]; + tensor var_10552_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_985_cast_fp16)[name = string("op_10552_cast_fp16")]; + tensor var_10553_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_987_cast_fp16)[name = string("op_10553_cast_fp16")]; + tensor var_10554_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_989_cast_fp16)[name = string("op_10554_cast_fp16")]; + tensor var_10555_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_991_cast_fp16)[name = string("op_10555_cast_fp16")]; + tensor var_10556_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_993_cast_fp16)[name = string("op_10556_cast_fp16")]; + tensor var_10557_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_995_cast_fp16)[name = string("op_10557_cast_fp16")]; + tensor var_10558_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_997_cast_fp16)[name = string("op_10558_cast_fp16")]; + tensor var_10559_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_999_cast_fp16)[name = string("op_10559_cast_fp16")]; + tensor var_10560_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1001_cast_fp16)[name = string("op_10560_cast_fp16")]; + tensor var_10561_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1003_cast_fp16)[name = string("op_10561_cast_fp16")]; + tensor var_10562_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1005_cast_fp16)[name = string("op_10562_cast_fp16")]; + tensor var_10563_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1007_cast_fp16)[name = string("op_10563_cast_fp16")]; + tensor var_10564_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1009_cast_fp16)[name = string("op_10564_cast_fp16")]; + tensor var_10565_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1011_cast_fp16)[name = string("op_10565_cast_fp16")]; + tensor var_10566_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1013_cast_fp16)[name = string("op_10566_cast_fp16")]; + tensor var_10567_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1015_cast_fp16)[name = string("op_10567_cast_fp16")]; + tensor var_10568_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1017_cast_fp16)[name = string("op_10568_cast_fp16")]; + tensor var_10569_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1019_cast_fp16)[name = string("op_10569_cast_fp16")]; + tensor var_10570_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1021_cast_fp16)[name = string("op_10570_cast_fp16")]; + tensor var_10571_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1023_cast_fp16)[name = string("op_10571_cast_fp16")]; + tensor var_10572_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1025_cast_fp16)[name = string("op_10572_cast_fp16")]; + tensor var_10573_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1027_cast_fp16)[name = string("op_10573_cast_fp16")]; + tensor var_10574_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1029_cast_fp16)[name = string("op_10574_cast_fp16")]; + tensor var_10575_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1031_cast_fp16)[name = string("op_10575_cast_fp16")]; + tensor var_10576_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1033_cast_fp16)[name = string("op_10576_cast_fp16")]; + tensor var_10577_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1035_cast_fp16)[name = string("op_10577_cast_fp16")]; + tensor var_10578_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1037_cast_fp16)[name = string("op_10578_cast_fp16")]; + tensor var_10579_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1039_cast_fp16)[name = string("op_10579_cast_fp16")]; + tensor var_10580_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1041_cast_fp16)[name = string("op_10580_cast_fp16")]; + tensor var_10581_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1043_cast_fp16)[name = string("op_10581_cast_fp16")]; + tensor var_10582_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1045_cast_fp16)[name = string("op_10582_cast_fp16")]; + tensor var_10583_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1047_cast_fp16)[name = string("op_10583_cast_fp16")]; + tensor var_10584_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1049_cast_fp16)[name = string("op_10584_cast_fp16")]; + tensor var_10585_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1051_cast_fp16)[name = string("op_10585_cast_fp16")]; + tensor var_10586_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1053_cast_fp16)[name = string("op_10586_cast_fp16")]; + tensor var_10587_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1055_cast_fp16)[name = string("op_10587_cast_fp16")]; + tensor var_10588_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1057_cast_fp16)[name = string("op_10588_cast_fp16")]; + tensor var_10589_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1059_cast_fp16)[name = string("op_10589_cast_fp16")]; + tensor var_10590_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1061_cast_fp16)[name = string("op_10590_cast_fp16")]; + tensor var_10591_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1063_cast_fp16)[name = string("op_10591_cast_fp16")]; + tensor var_10592_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1065_cast_fp16)[name = string("op_10592_cast_fp16")]; + tensor var_10593_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1067_cast_fp16)[name = string("op_10593_cast_fp16")]; + tensor var_10594_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1069_cast_fp16)[name = string("op_10594_cast_fp16")]; + tensor var_10595_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1071_cast_fp16)[name = string("op_10595_cast_fp16")]; + tensor var_10596_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1073_cast_fp16)[name = string("op_10596_cast_fp16")]; + tensor var_10597_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1075_cast_fp16)[name = string("op_10597_cast_fp16")]; + tensor var_10598_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1077_cast_fp16)[name = string("op_10598_cast_fp16")]; + tensor var_10599_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1079_cast_fp16)[name = string("op_10599_cast_fp16")]; + tensor var_10600_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1081_cast_fp16)[name = string("op_10600_cast_fp16")]; + tensor var_10601_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1083_cast_fp16)[name = string("op_10601_cast_fp16")]; + tensor var_10602_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1085_cast_fp16)[name = string("op_10602_cast_fp16")]; + tensor var_10603_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1087_cast_fp16)[name = string("op_10603_cast_fp16")]; + tensor var_10604_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1089_cast_fp16)[name = string("op_10604_cast_fp16")]; + tensor var_10605_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1091_cast_fp16)[name = string("op_10605_cast_fp16")]; + tensor var_10606_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1093_cast_fp16)[name = string("op_10606_cast_fp16")]; + tensor var_10607_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1095_cast_fp16)[name = string("op_10607_cast_fp16")]; + tensor var_10608_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1097_cast_fp16)[name = string("op_10608_cast_fp16")]; + tensor var_10609_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1099_cast_fp16)[name = string("op_10609_cast_fp16")]; + tensor var_10610_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1101_cast_fp16)[name = string("op_10610_cast_fp16")]; + tensor var_10611_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1103_cast_fp16)[name = string("op_10611_cast_fp16")]; + tensor var_10612_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1105_cast_fp16)[name = string("op_10612_cast_fp16")]; + tensor var_10613_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1107_cast_fp16)[name = string("op_10613_cast_fp16")]; + tensor var_10614_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1109_cast_fp16)[name = string("op_10614_cast_fp16")]; + tensor var_10615_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1111_cast_fp16)[name = string("op_10615_cast_fp16")]; + tensor var_10616_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1113_cast_fp16)[name = string("op_10616_cast_fp16")]; + tensor var_10617_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1115_cast_fp16)[name = string("op_10617_cast_fp16")]; + tensor var_10618_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1117_cast_fp16)[name = string("op_10618_cast_fp16")]; + tensor var_10619_cast_fp16 = softmax(axis = var_9365, x = aw_chunk_1119_cast_fp16)[name = string("op_10619_cast_fp16")]; + string var_10621_equation_0 = const()[name = string("op_10621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10621_cast_fp16 = einsum(equation = var_10621_equation_0, values = (var_10141_cast_fp16, var_10540_cast_fp16))[name = string("op_10621_cast_fp16")]; + string var_10623_equation_0 = const()[name = string("op_10623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10623_cast_fp16 = einsum(equation = var_10623_equation_0, values = (var_10141_cast_fp16, var_10541_cast_fp16))[name = string("op_10623_cast_fp16")]; + string var_10625_equation_0 = const()[name = string("op_10625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10625_cast_fp16 = einsum(equation = var_10625_equation_0, values = (var_10141_cast_fp16, var_10542_cast_fp16))[name = string("op_10625_cast_fp16")]; + string var_10627_equation_0 = const()[name = string("op_10627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10627_cast_fp16 = einsum(equation = var_10627_equation_0, values = (var_10141_cast_fp16, var_10543_cast_fp16))[name = string("op_10627_cast_fp16")]; + string var_10629_equation_0 = const()[name = string("op_10629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10629_cast_fp16 = einsum(equation = var_10629_equation_0, values = (var_10145_cast_fp16, var_10544_cast_fp16))[name = string("op_10629_cast_fp16")]; + string var_10631_equation_0 = const()[name = string("op_10631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10631_cast_fp16 = einsum(equation = var_10631_equation_0, values = (var_10145_cast_fp16, var_10545_cast_fp16))[name = string("op_10631_cast_fp16")]; + string var_10633_equation_0 = const()[name = string("op_10633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10633_cast_fp16 = einsum(equation = var_10633_equation_0, values = (var_10145_cast_fp16, var_10546_cast_fp16))[name = string("op_10633_cast_fp16")]; + string var_10635_equation_0 = const()[name = string("op_10635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10635_cast_fp16 = einsum(equation = var_10635_equation_0, values = (var_10145_cast_fp16, var_10547_cast_fp16))[name = string("op_10635_cast_fp16")]; + string var_10637_equation_0 = const()[name = string("op_10637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10637_cast_fp16 = einsum(equation = var_10637_equation_0, values = (var_10149_cast_fp16, var_10548_cast_fp16))[name = string("op_10637_cast_fp16")]; + string var_10639_equation_0 = const()[name = string("op_10639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10639_cast_fp16 = einsum(equation = var_10639_equation_0, values = (var_10149_cast_fp16, var_10549_cast_fp16))[name = string("op_10639_cast_fp16")]; + string var_10641_equation_0 = const()[name = string("op_10641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10641_cast_fp16 = einsum(equation = var_10641_equation_0, values = (var_10149_cast_fp16, var_10550_cast_fp16))[name = string("op_10641_cast_fp16")]; + string var_10643_equation_0 = const()[name = string("op_10643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10643_cast_fp16 = einsum(equation = var_10643_equation_0, values = (var_10149_cast_fp16, var_10551_cast_fp16))[name = string("op_10643_cast_fp16")]; + string var_10645_equation_0 = const()[name = string("op_10645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10645_cast_fp16 = einsum(equation = var_10645_equation_0, values = (var_10153_cast_fp16, var_10552_cast_fp16))[name = string("op_10645_cast_fp16")]; + string var_10647_equation_0 = const()[name = string("op_10647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10647_cast_fp16 = einsum(equation = var_10647_equation_0, values = (var_10153_cast_fp16, var_10553_cast_fp16))[name = string("op_10647_cast_fp16")]; + string var_10649_equation_0 = const()[name = string("op_10649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10649_cast_fp16 = einsum(equation = var_10649_equation_0, values = (var_10153_cast_fp16, var_10554_cast_fp16))[name = string("op_10649_cast_fp16")]; + string var_10651_equation_0 = const()[name = string("op_10651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10651_cast_fp16 = einsum(equation = var_10651_equation_0, values = (var_10153_cast_fp16, var_10555_cast_fp16))[name = string("op_10651_cast_fp16")]; + string var_10653_equation_0 = const()[name = string("op_10653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10653_cast_fp16 = einsum(equation = var_10653_equation_0, values = (var_10157_cast_fp16, var_10556_cast_fp16))[name = string("op_10653_cast_fp16")]; + string var_10655_equation_0 = const()[name = string("op_10655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10655_cast_fp16 = einsum(equation = var_10655_equation_0, values = (var_10157_cast_fp16, var_10557_cast_fp16))[name = string("op_10655_cast_fp16")]; + string var_10657_equation_0 = const()[name = string("op_10657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10657_cast_fp16 = einsum(equation = var_10657_equation_0, values = (var_10157_cast_fp16, var_10558_cast_fp16))[name = string("op_10657_cast_fp16")]; + string var_10659_equation_0 = const()[name = string("op_10659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10659_cast_fp16 = einsum(equation = var_10659_equation_0, values = (var_10157_cast_fp16, var_10559_cast_fp16))[name = string("op_10659_cast_fp16")]; + string var_10661_equation_0 = const()[name = string("op_10661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10661_cast_fp16 = einsum(equation = var_10661_equation_0, values = (var_10161_cast_fp16, var_10560_cast_fp16))[name = string("op_10661_cast_fp16")]; + string var_10663_equation_0 = const()[name = string("op_10663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10663_cast_fp16 = einsum(equation = var_10663_equation_0, values = (var_10161_cast_fp16, var_10561_cast_fp16))[name = string("op_10663_cast_fp16")]; + string var_10665_equation_0 = const()[name = string("op_10665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10665_cast_fp16 = einsum(equation = var_10665_equation_0, values = (var_10161_cast_fp16, var_10562_cast_fp16))[name = string("op_10665_cast_fp16")]; + string var_10667_equation_0 = const()[name = string("op_10667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10667_cast_fp16 = einsum(equation = var_10667_equation_0, values = (var_10161_cast_fp16, var_10563_cast_fp16))[name = string("op_10667_cast_fp16")]; + string var_10669_equation_0 = const()[name = string("op_10669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10669_cast_fp16 = einsum(equation = var_10669_equation_0, values = (var_10165_cast_fp16, var_10564_cast_fp16))[name = string("op_10669_cast_fp16")]; + string var_10671_equation_0 = const()[name = string("op_10671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10671_cast_fp16 = einsum(equation = var_10671_equation_0, values = (var_10165_cast_fp16, var_10565_cast_fp16))[name = string("op_10671_cast_fp16")]; + string var_10673_equation_0 = const()[name = string("op_10673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10673_cast_fp16 = einsum(equation = var_10673_equation_0, values = (var_10165_cast_fp16, var_10566_cast_fp16))[name = string("op_10673_cast_fp16")]; + string var_10675_equation_0 = const()[name = string("op_10675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10675_cast_fp16 = einsum(equation = var_10675_equation_0, values = (var_10165_cast_fp16, var_10567_cast_fp16))[name = string("op_10675_cast_fp16")]; + string var_10677_equation_0 = const()[name = string("op_10677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10677_cast_fp16 = einsum(equation = var_10677_equation_0, values = (var_10169_cast_fp16, var_10568_cast_fp16))[name = string("op_10677_cast_fp16")]; + string var_10679_equation_0 = const()[name = string("op_10679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10679_cast_fp16 = einsum(equation = var_10679_equation_0, values = (var_10169_cast_fp16, var_10569_cast_fp16))[name = string("op_10679_cast_fp16")]; + string var_10681_equation_0 = const()[name = string("op_10681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10681_cast_fp16 = einsum(equation = var_10681_equation_0, values = (var_10169_cast_fp16, var_10570_cast_fp16))[name = string("op_10681_cast_fp16")]; + string var_10683_equation_0 = const()[name = string("op_10683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10683_cast_fp16 = einsum(equation = var_10683_equation_0, values = (var_10169_cast_fp16, var_10571_cast_fp16))[name = string("op_10683_cast_fp16")]; + string var_10685_equation_0 = const()[name = string("op_10685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10685_cast_fp16 = einsum(equation = var_10685_equation_0, values = (var_10173_cast_fp16, var_10572_cast_fp16))[name = string("op_10685_cast_fp16")]; + string var_10687_equation_0 = const()[name = string("op_10687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10687_cast_fp16 = einsum(equation = var_10687_equation_0, values = (var_10173_cast_fp16, var_10573_cast_fp16))[name = string("op_10687_cast_fp16")]; + string var_10689_equation_0 = const()[name = string("op_10689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10689_cast_fp16 = einsum(equation = var_10689_equation_0, values = (var_10173_cast_fp16, var_10574_cast_fp16))[name = string("op_10689_cast_fp16")]; + string var_10691_equation_0 = const()[name = string("op_10691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10691_cast_fp16 = einsum(equation = var_10691_equation_0, values = (var_10173_cast_fp16, var_10575_cast_fp16))[name = string("op_10691_cast_fp16")]; + string var_10693_equation_0 = const()[name = string("op_10693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10693_cast_fp16 = einsum(equation = var_10693_equation_0, values = (var_10177_cast_fp16, var_10576_cast_fp16))[name = string("op_10693_cast_fp16")]; + string var_10695_equation_0 = const()[name = string("op_10695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10695_cast_fp16 = einsum(equation = var_10695_equation_0, values = (var_10177_cast_fp16, var_10577_cast_fp16))[name = string("op_10695_cast_fp16")]; + string var_10697_equation_0 = const()[name = string("op_10697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10697_cast_fp16 = einsum(equation = var_10697_equation_0, values = (var_10177_cast_fp16, var_10578_cast_fp16))[name = string("op_10697_cast_fp16")]; + string var_10699_equation_0 = const()[name = string("op_10699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10699_cast_fp16 = einsum(equation = var_10699_equation_0, values = (var_10177_cast_fp16, var_10579_cast_fp16))[name = string("op_10699_cast_fp16")]; + string var_10701_equation_0 = const()[name = string("op_10701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10701_cast_fp16 = einsum(equation = var_10701_equation_0, values = (var_10181_cast_fp16, var_10580_cast_fp16))[name = string("op_10701_cast_fp16")]; + string var_10703_equation_0 = const()[name = string("op_10703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10703_cast_fp16 = einsum(equation = var_10703_equation_0, values = (var_10181_cast_fp16, var_10581_cast_fp16))[name = string("op_10703_cast_fp16")]; + string var_10705_equation_0 = const()[name = string("op_10705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10705_cast_fp16 = einsum(equation = var_10705_equation_0, values = (var_10181_cast_fp16, var_10582_cast_fp16))[name = string("op_10705_cast_fp16")]; + string var_10707_equation_0 = const()[name = string("op_10707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10707_cast_fp16 = einsum(equation = var_10707_equation_0, values = (var_10181_cast_fp16, var_10583_cast_fp16))[name = string("op_10707_cast_fp16")]; + string var_10709_equation_0 = const()[name = string("op_10709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10709_cast_fp16 = einsum(equation = var_10709_equation_0, values = (var_10185_cast_fp16, var_10584_cast_fp16))[name = string("op_10709_cast_fp16")]; + string var_10711_equation_0 = const()[name = string("op_10711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10711_cast_fp16 = einsum(equation = var_10711_equation_0, values = (var_10185_cast_fp16, var_10585_cast_fp16))[name = string("op_10711_cast_fp16")]; + string var_10713_equation_0 = const()[name = string("op_10713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10713_cast_fp16 = einsum(equation = var_10713_equation_0, values = (var_10185_cast_fp16, var_10586_cast_fp16))[name = string("op_10713_cast_fp16")]; + string var_10715_equation_0 = const()[name = string("op_10715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10715_cast_fp16 = einsum(equation = var_10715_equation_0, values = (var_10185_cast_fp16, var_10587_cast_fp16))[name = string("op_10715_cast_fp16")]; + string var_10717_equation_0 = const()[name = string("op_10717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10717_cast_fp16 = einsum(equation = var_10717_equation_0, values = (var_10189_cast_fp16, var_10588_cast_fp16))[name = string("op_10717_cast_fp16")]; + string var_10719_equation_0 = const()[name = string("op_10719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10719_cast_fp16 = einsum(equation = var_10719_equation_0, values = (var_10189_cast_fp16, var_10589_cast_fp16))[name = string("op_10719_cast_fp16")]; + string var_10721_equation_0 = const()[name = string("op_10721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10721_cast_fp16 = einsum(equation = var_10721_equation_0, values = (var_10189_cast_fp16, var_10590_cast_fp16))[name = string("op_10721_cast_fp16")]; + string var_10723_equation_0 = const()[name = string("op_10723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10723_cast_fp16 = einsum(equation = var_10723_equation_0, values = (var_10189_cast_fp16, var_10591_cast_fp16))[name = string("op_10723_cast_fp16")]; + string var_10725_equation_0 = const()[name = string("op_10725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10725_cast_fp16 = einsum(equation = var_10725_equation_0, values = (var_10193_cast_fp16, var_10592_cast_fp16))[name = string("op_10725_cast_fp16")]; + string var_10727_equation_0 = const()[name = string("op_10727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10727_cast_fp16 = einsum(equation = var_10727_equation_0, values = (var_10193_cast_fp16, var_10593_cast_fp16))[name = string("op_10727_cast_fp16")]; + string var_10729_equation_0 = const()[name = string("op_10729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10729_cast_fp16 = einsum(equation = var_10729_equation_0, values = (var_10193_cast_fp16, var_10594_cast_fp16))[name = string("op_10729_cast_fp16")]; + string var_10731_equation_0 = const()[name = string("op_10731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10731_cast_fp16 = einsum(equation = var_10731_equation_0, values = (var_10193_cast_fp16, var_10595_cast_fp16))[name = string("op_10731_cast_fp16")]; + string var_10733_equation_0 = const()[name = string("op_10733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10733_cast_fp16 = einsum(equation = var_10733_equation_0, values = (var_10197_cast_fp16, var_10596_cast_fp16))[name = string("op_10733_cast_fp16")]; + string var_10735_equation_0 = const()[name = string("op_10735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10735_cast_fp16 = einsum(equation = var_10735_equation_0, values = (var_10197_cast_fp16, var_10597_cast_fp16))[name = string("op_10735_cast_fp16")]; + string var_10737_equation_0 = const()[name = string("op_10737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10737_cast_fp16 = einsum(equation = var_10737_equation_0, values = (var_10197_cast_fp16, var_10598_cast_fp16))[name = string("op_10737_cast_fp16")]; + string var_10739_equation_0 = const()[name = string("op_10739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10739_cast_fp16 = einsum(equation = var_10739_equation_0, values = (var_10197_cast_fp16, var_10599_cast_fp16))[name = string("op_10739_cast_fp16")]; + string var_10741_equation_0 = const()[name = string("op_10741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10741_cast_fp16 = einsum(equation = var_10741_equation_0, values = (var_10201_cast_fp16, var_10600_cast_fp16))[name = string("op_10741_cast_fp16")]; + string var_10743_equation_0 = const()[name = string("op_10743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10743_cast_fp16 = einsum(equation = var_10743_equation_0, values = (var_10201_cast_fp16, var_10601_cast_fp16))[name = string("op_10743_cast_fp16")]; + string var_10745_equation_0 = const()[name = string("op_10745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10745_cast_fp16 = einsum(equation = var_10745_equation_0, values = (var_10201_cast_fp16, var_10602_cast_fp16))[name = string("op_10745_cast_fp16")]; + string var_10747_equation_0 = const()[name = string("op_10747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10747_cast_fp16 = einsum(equation = var_10747_equation_0, values = (var_10201_cast_fp16, var_10603_cast_fp16))[name = string("op_10747_cast_fp16")]; + string var_10749_equation_0 = const()[name = string("op_10749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10749_cast_fp16 = einsum(equation = var_10749_equation_0, values = (var_10205_cast_fp16, var_10604_cast_fp16))[name = string("op_10749_cast_fp16")]; + string var_10751_equation_0 = const()[name = string("op_10751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10751_cast_fp16 = einsum(equation = var_10751_equation_0, values = (var_10205_cast_fp16, var_10605_cast_fp16))[name = string("op_10751_cast_fp16")]; + string var_10753_equation_0 = const()[name = string("op_10753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10753_cast_fp16 = einsum(equation = var_10753_equation_0, values = (var_10205_cast_fp16, var_10606_cast_fp16))[name = string("op_10753_cast_fp16")]; + string var_10755_equation_0 = const()[name = string("op_10755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10755_cast_fp16 = einsum(equation = var_10755_equation_0, values = (var_10205_cast_fp16, var_10607_cast_fp16))[name = string("op_10755_cast_fp16")]; + string var_10757_equation_0 = const()[name = string("op_10757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10757_cast_fp16 = einsum(equation = var_10757_equation_0, values = (var_10209_cast_fp16, var_10608_cast_fp16))[name = string("op_10757_cast_fp16")]; + string var_10759_equation_0 = const()[name = string("op_10759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10759_cast_fp16 = einsum(equation = var_10759_equation_0, values = (var_10209_cast_fp16, var_10609_cast_fp16))[name = string("op_10759_cast_fp16")]; + string var_10761_equation_0 = const()[name = string("op_10761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10761_cast_fp16 = einsum(equation = var_10761_equation_0, values = (var_10209_cast_fp16, var_10610_cast_fp16))[name = string("op_10761_cast_fp16")]; + string var_10763_equation_0 = const()[name = string("op_10763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10763_cast_fp16 = einsum(equation = var_10763_equation_0, values = (var_10209_cast_fp16, var_10611_cast_fp16))[name = string("op_10763_cast_fp16")]; + string var_10765_equation_0 = const()[name = string("op_10765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10765_cast_fp16 = einsum(equation = var_10765_equation_0, values = (var_10213_cast_fp16, var_10612_cast_fp16))[name = string("op_10765_cast_fp16")]; + string var_10767_equation_0 = const()[name = string("op_10767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10767_cast_fp16 = einsum(equation = var_10767_equation_0, values = (var_10213_cast_fp16, var_10613_cast_fp16))[name = string("op_10767_cast_fp16")]; + string var_10769_equation_0 = const()[name = string("op_10769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10769_cast_fp16 = einsum(equation = var_10769_equation_0, values = (var_10213_cast_fp16, var_10614_cast_fp16))[name = string("op_10769_cast_fp16")]; + string var_10771_equation_0 = const()[name = string("op_10771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10771_cast_fp16 = einsum(equation = var_10771_equation_0, values = (var_10213_cast_fp16, var_10615_cast_fp16))[name = string("op_10771_cast_fp16")]; + string var_10773_equation_0 = const()[name = string("op_10773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10773_cast_fp16 = einsum(equation = var_10773_equation_0, values = (var_10217_cast_fp16, var_10616_cast_fp16))[name = string("op_10773_cast_fp16")]; + string var_10775_equation_0 = const()[name = string("op_10775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10775_cast_fp16 = einsum(equation = var_10775_equation_0, values = (var_10217_cast_fp16, var_10617_cast_fp16))[name = string("op_10775_cast_fp16")]; + string var_10777_equation_0 = const()[name = string("op_10777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10777_cast_fp16 = einsum(equation = var_10777_equation_0, values = (var_10217_cast_fp16, var_10618_cast_fp16))[name = string("op_10777_cast_fp16")]; + string var_10779_equation_0 = const()[name = string("op_10779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10779_cast_fp16 = einsum(equation = var_10779_equation_0, values = (var_10217_cast_fp16, var_10619_cast_fp16))[name = string("op_10779_cast_fp16")]; + bool var_10781_interleave_0 = const()[name = string("op_10781_interleave_0"), val = bool(false)]; + tensor var_10781_cast_fp16 = concat(axis = var_9340, interleave = var_10781_interleave_0, values = (var_10621_cast_fp16, var_10623_cast_fp16, var_10625_cast_fp16, var_10627_cast_fp16))[name = string("op_10781_cast_fp16")]; + bool var_10783_interleave_0 = const()[name = string("op_10783_interleave_0"), val = bool(false)]; + tensor var_10783_cast_fp16 = concat(axis = var_9340, interleave = var_10783_interleave_0, values = (var_10629_cast_fp16, var_10631_cast_fp16, var_10633_cast_fp16, var_10635_cast_fp16))[name = string("op_10783_cast_fp16")]; + bool var_10785_interleave_0 = const()[name = string("op_10785_interleave_0"), val = bool(false)]; + tensor var_10785_cast_fp16 = concat(axis = var_9340, interleave = var_10785_interleave_0, values = (var_10637_cast_fp16, var_10639_cast_fp16, var_10641_cast_fp16, var_10643_cast_fp16))[name = string("op_10785_cast_fp16")]; + bool var_10787_interleave_0 = const()[name = string("op_10787_interleave_0"), val = bool(false)]; + tensor var_10787_cast_fp16 = concat(axis = var_9340, interleave = var_10787_interleave_0, values = (var_10645_cast_fp16, var_10647_cast_fp16, var_10649_cast_fp16, var_10651_cast_fp16))[name = string("op_10787_cast_fp16")]; + bool var_10789_interleave_0 = const()[name = string("op_10789_interleave_0"), val = bool(false)]; + tensor var_10789_cast_fp16 = concat(axis = var_9340, interleave = var_10789_interleave_0, values = (var_10653_cast_fp16, var_10655_cast_fp16, var_10657_cast_fp16, var_10659_cast_fp16))[name = string("op_10789_cast_fp16")]; + bool var_10791_interleave_0 = const()[name = string("op_10791_interleave_0"), val = bool(false)]; + tensor var_10791_cast_fp16 = concat(axis = var_9340, interleave = var_10791_interleave_0, values = (var_10661_cast_fp16, var_10663_cast_fp16, var_10665_cast_fp16, var_10667_cast_fp16))[name = string("op_10791_cast_fp16")]; + bool var_10793_interleave_0 = const()[name = string("op_10793_interleave_0"), val = bool(false)]; + tensor var_10793_cast_fp16 = concat(axis = var_9340, interleave = var_10793_interleave_0, values = (var_10669_cast_fp16, var_10671_cast_fp16, var_10673_cast_fp16, var_10675_cast_fp16))[name = string("op_10793_cast_fp16")]; + bool var_10795_interleave_0 = const()[name = string("op_10795_interleave_0"), val = bool(false)]; + tensor var_10795_cast_fp16 = concat(axis = var_9340, interleave = var_10795_interleave_0, values = (var_10677_cast_fp16, var_10679_cast_fp16, var_10681_cast_fp16, var_10683_cast_fp16))[name = string("op_10795_cast_fp16")]; + bool var_10797_interleave_0 = const()[name = string("op_10797_interleave_0"), val = bool(false)]; + tensor var_10797_cast_fp16 = concat(axis = var_9340, interleave = var_10797_interleave_0, values = (var_10685_cast_fp16, var_10687_cast_fp16, var_10689_cast_fp16, var_10691_cast_fp16))[name = string("op_10797_cast_fp16")]; + bool var_10799_interleave_0 = const()[name = string("op_10799_interleave_0"), val = bool(false)]; + tensor var_10799_cast_fp16 = concat(axis = var_9340, interleave = var_10799_interleave_0, values = (var_10693_cast_fp16, var_10695_cast_fp16, var_10697_cast_fp16, var_10699_cast_fp16))[name = string("op_10799_cast_fp16")]; + bool var_10801_interleave_0 = const()[name = string("op_10801_interleave_0"), val = bool(false)]; + tensor var_10801_cast_fp16 = concat(axis = var_9340, interleave = var_10801_interleave_0, values = (var_10701_cast_fp16, var_10703_cast_fp16, var_10705_cast_fp16, var_10707_cast_fp16))[name = string("op_10801_cast_fp16")]; + bool var_10803_interleave_0 = const()[name = string("op_10803_interleave_0"), val = bool(false)]; + tensor var_10803_cast_fp16 = concat(axis = var_9340, interleave = var_10803_interleave_0, values = (var_10709_cast_fp16, var_10711_cast_fp16, var_10713_cast_fp16, var_10715_cast_fp16))[name = string("op_10803_cast_fp16")]; + bool var_10805_interleave_0 = const()[name = string("op_10805_interleave_0"), val = bool(false)]; + tensor var_10805_cast_fp16 = concat(axis = var_9340, interleave = var_10805_interleave_0, values = (var_10717_cast_fp16, var_10719_cast_fp16, var_10721_cast_fp16, var_10723_cast_fp16))[name = string("op_10805_cast_fp16")]; + bool var_10807_interleave_0 = const()[name = string("op_10807_interleave_0"), val = bool(false)]; + tensor var_10807_cast_fp16 = concat(axis = var_9340, interleave = var_10807_interleave_0, values = (var_10725_cast_fp16, var_10727_cast_fp16, var_10729_cast_fp16, var_10731_cast_fp16))[name = string("op_10807_cast_fp16")]; + bool var_10809_interleave_0 = const()[name = string("op_10809_interleave_0"), val = bool(false)]; + tensor var_10809_cast_fp16 = concat(axis = var_9340, interleave = var_10809_interleave_0, values = (var_10733_cast_fp16, var_10735_cast_fp16, var_10737_cast_fp16, var_10739_cast_fp16))[name = string("op_10809_cast_fp16")]; + bool var_10811_interleave_0 = const()[name = string("op_10811_interleave_0"), val = bool(false)]; + tensor var_10811_cast_fp16 = concat(axis = var_9340, interleave = var_10811_interleave_0, values = (var_10741_cast_fp16, var_10743_cast_fp16, var_10745_cast_fp16, var_10747_cast_fp16))[name = string("op_10811_cast_fp16")]; + bool var_10813_interleave_0 = const()[name = string("op_10813_interleave_0"), val = bool(false)]; + tensor var_10813_cast_fp16 = concat(axis = var_9340, interleave = var_10813_interleave_0, values = (var_10749_cast_fp16, var_10751_cast_fp16, var_10753_cast_fp16, var_10755_cast_fp16))[name = string("op_10813_cast_fp16")]; + bool var_10815_interleave_0 = const()[name = string("op_10815_interleave_0"), val = bool(false)]; + tensor var_10815_cast_fp16 = concat(axis = var_9340, interleave = var_10815_interleave_0, values = (var_10757_cast_fp16, var_10759_cast_fp16, var_10761_cast_fp16, var_10763_cast_fp16))[name = string("op_10815_cast_fp16")]; + bool var_10817_interleave_0 = const()[name = string("op_10817_interleave_0"), val = bool(false)]; + tensor var_10817_cast_fp16 = concat(axis = var_9340, interleave = var_10817_interleave_0, values = (var_10765_cast_fp16, var_10767_cast_fp16, var_10769_cast_fp16, var_10771_cast_fp16))[name = string("op_10817_cast_fp16")]; + bool var_10819_interleave_0 = const()[name = string("op_10819_interleave_0"), val = bool(false)]; + tensor var_10819_cast_fp16 = concat(axis = var_9340, interleave = var_10819_interleave_0, values = (var_10773_cast_fp16, var_10775_cast_fp16, var_10777_cast_fp16, var_10779_cast_fp16))[name = string("op_10819_cast_fp16")]; + bool input_49_interleave_0 = const()[name = string("input_49_interleave_0"), val = bool(false)]; + tensor input_49_cast_fp16 = concat(axis = var_9365, interleave = input_49_interleave_0, values = (var_10781_cast_fp16, var_10783_cast_fp16, var_10785_cast_fp16, var_10787_cast_fp16, var_10789_cast_fp16, var_10791_cast_fp16, var_10793_cast_fp16, var_10795_cast_fp16, var_10797_cast_fp16, var_10799_cast_fp16, var_10801_cast_fp16, var_10803_cast_fp16, var_10805_cast_fp16, var_10807_cast_fp16, var_10809_cast_fp16, var_10811_cast_fp16, var_10813_cast_fp16, var_10815_cast_fp16, var_10817_cast_fp16, var_10819_cast_fp16))[name = string("input_49_cast_fp16")]; + string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; + tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; + tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; + int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260624960)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263901824)))]; + tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_10838_to_fp16 = const()[name = string("op_10838_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_10838_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263904448)))]; + tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263907072)))]; + fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; + string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; + tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; + tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; + int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263909696)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277016960)))]; + tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; + string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; + tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277027264)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290134528)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; + int32 var_10867 = const()[name = string("op_10867"), val = int32(3)]; + int32 var_10892 = const()[name = string("op_10892"), val = int32(1)]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_10909_to_fp16 = const()[name = string("op_10909_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_10909_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290137152)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290139776)))]; + fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290142400)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293419264)))]; + tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; + string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; + tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; + tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; + int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293421888)))]; + tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; + string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; + tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; + tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; + int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296698752)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299975616)))]; + tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_10947_begin_0 = const()[name = string("op_10947_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10947_end_0 = const()[name = string("op_10947_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10947_end_mask_0 = const()[name = string("op_10947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10947_cast_fp16 = slice_by_index(begin = var_10947_begin_0, end = var_10947_end_0, end_mask = var_10947_end_mask_0, x = query_15_cast_fp16)[name = string("op_10947_cast_fp16")]; + tensor var_10951_begin_0 = const()[name = string("op_10951_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10951_end_0 = const()[name = string("op_10951_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10951_end_mask_0 = const()[name = string("op_10951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10951_cast_fp16 = slice_by_index(begin = var_10951_begin_0, end = var_10951_end_0, end_mask = var_10951_end_mask_0, x = query_15_cast_fp16)[name = string("op_10951_cast_fp16")]; + tensor var_10955_begin_0 = const()[name = string("op_10955_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10955_end_0 = const()[name = string("op_10955_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10955_end_mask_0 = const()[name = string("op_10955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10955_cast_fp16 = slice_by_index(begin = var_10955_begin_0, end = var_10955_end_0, end_mask = var_10955_end_mask_0, x = query_15_cast_fp16)[name = string("op_10955_cast_fp16")]; + tensor var_10959_begin_0 = const()[name = string("op_10959_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10959_end_0 = const()[name = string("op_10959_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10959_end_mask_0 = const()[name = string("op_10959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10959_cast_fp16 = slice_by_index(begin = var_10959_begin_0, end = var_10959_end_0, end_mask = var_10959_end_mask_0, x = query_15_cast_fp16)[name = string("op_10959_cast_fp16")]; + tensor var_10963_begin_0 = const()[name = string("op_10963_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10963_end_0 = const()[name = string("op_10963_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10963_end_mask_0 = const()[name = string("op_10963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10963_cast_fp16 = slice_by_index(begin = var_10963_begin_0, end = var_10963_end_0, end_mask = var_10963_end_mask_0, x = query_15_cast_fp16)[name = string("op_10963_cast_fp16")]; + tensor var_10967_begin_0 = const()[name = string("op_10967_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10967_end_0 = const()[name = string("op_10967_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10967_end_mask_0 = const()[name = string("op_10967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10967_cast_fp16 = slice_by_index(begin = var_10967_begin_0, end = var_10967_end_0, end_mask = var_10967_end_mask_0, x = query_15_cast_fp16)[name = string("op_10967_cast_fp16")]; + tensor var_10971_begin_0 = const()[name = string("op_10971_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10971_end_0 = const()[name = string("op_10971_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10971_end_mask_0 = const()[name = string("op_10971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10971_cast_fp16 = slice_by_index(begin = var_10971_begin_0, end = var_10971_end_0, end_mask = var_10971_end_mask_0, x = query_15_cast_fp16)[name = string("op_10971_cast_fp16")]; + tensor var_10975_begin_0 = const()[name = string("op_10975_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10975_end_0 = const()[name = string("op_10975_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10975_end_mask_0 = const()[name = string("op_10975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10975_cast_fp16 = slice_by_index(begin = var_10975_begin_0, end = var_10975_end_0, end_mask = var_10975_end_mask_0, x = query_15_cast_fp16)[name = string("op_10975_cast_fp16")]; + tensor var_10979_begin_0 = const()[name = string("op_10979_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10979_end_0 = const()[name = string("op_10979_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10979_end_mask_0 = const()[name = string("op_10979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = query_15_cast_fp16)[name = string("op_10979_cast_fp16")]; + tensor var_10983_begin_0 = const()[name = string("op_10983_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10983_end_0 = const()[name = string("op_10983_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10983_end_mask_0 = const()[name = string("op_10983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10983_cast_fp16 = slice_by_index(begin = var_10983_begin_0, end = var_10983_end_0, end_mask = var_10983_end_mask_0, x = query_15_cast_fp16)[name = string("op_10983_cast_fp16")]; + tensor var_10987_begin_0 = const()[name = string("op_10987_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10987_end_0 = const()[name = string("op_10987_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10987_end_mask_0 = const()[name = string("op_10987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10987_cast_fp16 = slice_by_index(begin = var_10987_begin_0, end = var_10987_end_0, end_mask = var_10987_end_mask_0, x = query_15_cast_fp16)[name = string("op_10987_cast_fp16")]; + tensor var_10991_begin_0 = const()[name = string("op_10991_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10991_end_0 = const()[name = string("op_10991_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10991_end_mask_0 = const()[name = string("op_10991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10991_cast_fp16 = slice_by_index(begin = var_10991_begin_0, end = var_10991_end_0, end_mask = var_10991_end_mask_0, x = query_15_cast_fp16)[name = string("op_10991_cast_fp16")]; + tensor var_10995_begin_0 = const()[name = string("op_10995_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10995_end_0 = const()[name = string("op_10995_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_10995_end_mask_0 = const()[name = string("op_10995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10995_cast_fp16 = slice_by_index(begin = var_10995_begin_0, end = var_10995_end_0, end_mask = var_10995_end_mask_0, x = query_15_cast_fp16)[name = string("op_10995_cast_fp16")]; + tensor var_10999_begin_0 = const()[name = string("op_10999_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10999_end_0 = const()[name = string("op_10999_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_10999_end_mask_0 = const()[name = string("op_10999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10999_cast_fp16 = slice_by_index(begin = var_10999_begin_0, end = var_10999_end_0, end_mask = var_10999_end_mask_0, x = query_15_cast_fp16)[name = string("op_10999_cast_fp16")]; + tensor var_11003_begin_0 = const()[name = string("op_11003_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11003_end_0 = const()[name = string("op_11003_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_11003_end_mask_0 = const()[name = string("op_11003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11003_cast_fp16 = slice_by_index(begin = var_11003_begin_0, end = var_11003_end_0, end_mask = var_11003_end_mask_0, x = query_15_cast_fp16)[name = string("op_11003_cast_fp16")]; + tensor var_11007_begin_0 = const()[name = string("op_11007_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11007_end_0 = const()[name = string("op_11007_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_11007_end_mask_0 = const()[name = string("op_11007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = query_15_cast_fp16)[name = string("op_11007_cast_fp16")]; + tensor var_11011_begin_0 = const()[name = string("op_11011_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11011_end_0 = const()[name = string("op_11011_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_11011_end_mask_0 = const()[name = string("op_11011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11011_cast_fp16 = slice_by_index(begin = var_11011_begin_0, end = var_11011_end_0, end_mask = var_11011_end_mask_0, x = query_15_cast_fp16)[name = string("op_11011_cast_fp16")]; + tensor var_11015_begin_0 = const()[name = string("op_11015_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11015_end_0 = const()[name = string("op_11015_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_11015_end_mask_0 = const()[name = string("op_11015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11015_cast_fp16 = slice_by_index(begin = var_11015_begin_0, end = var_11015_end_0, end_mask = var_11015_end_mask_0, x = query_15_cast_fp16)[name = string("op_11015_cast_fp16")]; + tensor var_11019_begin_0 = const()[name = string("op_11019_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11019_end_0 = const()[name = string("op_11019_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_11019_end_mask_0 = const()[name = string("op_11019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11019_cast_fp16 = slice_by_index(begin = var_11019_begin_0, end = var_11019_end_0, end_mask = var_11019_end_mask_0, x = query_15_cast_fp16)[name = string("op_11019_cast_fp16")]; + tensor var_11023_begin_0 = const()[name = string("op_11023_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11023_end_0 = const()[name = string("op_11023_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_11023_end_mask_0 = const()[name = string("op_11023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11023_cast_fp16 = slice_by_index(begin = var_11023_begin_0, end = var_11023_end_0, end_mask = var_11023_end_mask_0, x = query_15_cast_fp16)[name = string("op_11023_cast_fp16")]; + tensor var_11032_begin_0 = const()[name = string("op_11032_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11032_end_0 = const()[name = string("op_11032_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11032_end_mask_0 = const()[name = string("op_11032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11032_cast_fp16 = slice_by_index(begin = var_11032_begin_0, end = var_11032_end_0, end_mask = var_11032_end_mask_0, x = var_10947_cast_fp16)[name = string("op_11032_cast_fp16")]; + tensor var_11039_begin_0 = const()[name = string("op_11039_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11039_end_0 = const()[name = string("op_11039_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11039_end_mask_0 = const()[name = string("op_11039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11039_cast_fp16 = slice_by_index(begin = var_11039_begin_0, end = var_11039_end_0, end_mask = var_11039_end_mask_0, x = var_10947_cast_fp16)[name = string("op_11039_cast_fp16")]; + tensor var_11046_begin_0 = const()[name = string("op_11046_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11046_end_0 = const()[name = string("op_11046_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11046_end_mask_0 = const()[name = string("op_11046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11046_cast_fp16 = slice_by_index(begin = var_11046_begin_0, end = var_11046_end_0, end_mask = var_11046_end_mask_0, x = var_10947_cast_fp16)[name = string("op_11046_cast_fp16")]; + tensor var_11053_begin_0 = const()[name = string("op_11053_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11053_end_0 = const()[name = string("op_11053_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11053_end_mask_0 = const()[name = string("op_11053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11053_cast_fp16 = slice_by_index(begin = var_11053_begin_0, end = var_11053_end_0, end_mask = var_11053_end_mask_0, x = var_10947_cast_fp16)[name = string("op_11053_cast_fp16")]; + tensor var_11060_begin_0 = const()[name = string("op_11060_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11060_end_0 = const()[name = string("op_11060_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11060_end_mask_0 = const()[name = string("op_11060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11060_cast_fp16 = slice_by_index(begin = var_11060_begin_0, end = var_11060_end_0, end_mask = var_11060_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11060_cast_fp16")]; + tensor var_11067_begin_0 = const()[name = string("op_11067_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11067_end_0 = const()[name = string("op_11067_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11067_end_mask_0 = const()[name = string("op_11067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11067_cast_fp16 = slice_by_index(begin = var_11067_begin_0, end = var_11067_end_0, end_mask = var_11067_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11067_cast_fp16")]; + tensor var_11074_begin_0 = const()[name = string("op_11074_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11074_end_0 = const()[name = string("op_11074_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11074_end_mask_0 = const()[name = string("op_11074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11074_cast_fp16 = slice_by_index(begin = var_11074_begin_0, end = var_11074_end_0, end_mask = var_11074_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11074_cast_fp16")]; + tensor var_11081_begin_0 = const()[name = string("op_11081_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11081_end_0 = const()[name = string("op_11081_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11081_end_mask_0 = const()[name = string("op_11081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11081_cast_fp16 = slice_by_index(begin = var_11081_begin_0, end = var_11081_end_0, end_mask = var_11081_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11081_cast_fp16")]; + tensor var_11088_begin_0 = const()[name = string("op_11088_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11088_end_0 = const()[name = string("op_11088_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11088_end_mask_0 = const()[name = string("op_11088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11088_cast_fp16 = slice_by_index(begin = var_11088_begin_0, end = var_11088_end_0, end_mask = var_11088_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11088_cast_fp16")]; + tensor var_11095_begin_0 = const()[name = string("op_11095_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11095_end_0 = const()[name = string("op_11095_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11095_end_mask_0 = const()[name = string("op_11095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11095_cast_fp16 = slice_by_index(begin = var_11095_begin_0, end = var_11095_end_0, end_mask = var_11095_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11095_cast_fp16")]; + tensor var_11102_begin_0 = const()[name = string("op_11102_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11102_end_0 = const()[name = string("op_11102_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11102_end_mask_0 = const()[name = string("op_11102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11102_cast_fp16 = slice_by_index(begin = var_11102_begin_0, end = var_11102_end_0, end_mask = var_11102_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11102_cast_fp16")]; + tensor var_11109_begin_0 = const()[name = string("op_11109_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11109_end_0 = const()[name = string("op_11109_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11109_end_mask_0 = const()[name = string("op_11109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11109_cast_fp16 = slice_by_index(begin = var_11109_begin_0, end = var_11109_end_0, end_mask = var_11109_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11109_cast_fp16")]; + tensor var_11116_begin_0 = const()[name = string("op_11116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11116_end_0 = const()[name = string("op_11116_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11116_end_mask_0 = const()[name = string("op_11116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11116_cast_fp16 = slice_by_index(begin = var_11116_begin_0, end = var_11116_end_0, end_mask = var_11116_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11116_cast_fp16")]; + tensor var_11123_begin_0 = const()[name = string("op_11123_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11123_end_0 = const()[name = string("op_11123_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11123_end_mask_0 = const()[name = string("op_11123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11123_cast_fp16 = slice_by_index(begin = var_11123_begin_0, end = var_11123_end_0, end_mask = var_11123_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11123_cast_fp16")]; + tensor var_11130_begin_0 = const()[name = string("op_11130_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11130_end_0 = const()[name = string("op_11130_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11130_end_mask_0 = const()[name = string("op_11130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11130_cast_fp16 = slice_by_index(begin = var_11130_begin_0, end = var_11130_end_0, end_mask = var_11130_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11130_cast_fp16")]; + tensor var_11137_begin_0 = const()[name = string("op_11137_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11137_end_0 = const()[name = string("op_11137_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11137_end_mask_0 = const()[name = string("op_11137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11137_cast_fp16 = slice_by_index(begin = var_11137_begin_0, end = var_11137_end_0, end_mask = var_11137_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11137_cast_fp16")]; + tensor var_11144_begin_0 = const()[name = string("op_11144_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11144_end_0 = const()[name = string("op_11144_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11144_end_mask_0 = const()[name = string("op_11144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11144_cast_fp16 = slice_by_index(begin = var_11144_begin_0, end = var_11144_end_0, end_mask = var_11144_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11144_cast_fp16")]; + tensor var_11151_begin_0 = const()[name = string("op_11151_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11151_end_0 = const()[name = string("op_11151_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11151_end_mask_0 = const()[name = string("op_11151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11151_cast_fp16 = slice_by_index(begin = var_11151_begin_0, end = var_11151_end_0, end_mask = var_11151_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11151_cast_fp16")]; + tensor var_11158_begin_0 = const()[name = string("op_11158_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11158_end_0 = const()[name = string("op_11158_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11158_end_mask_0 = const()[name = string("op_11158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11158_cast_fp16 = slice_by_index(begin = var_11158_begin_0, end = var_11158_end_0, end_mask = var_11158_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11158_cast_fp16")]; + tensor var_11165_begin_0 = const()[name = string("op_11165_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11165_end_0 = const()[name = string("op_11165_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11165_end_mask_0 = const()[name = string("op_11165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11165_cast_fp16 = slice_by_index(begin = var_11165_begin_0, end = var_11165_end_0, end_mask = var_11165_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11165_cast_fp16")]; + tensor var_11172_begin_0 = const()[name = string("op_11172_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11172_end_0 = const()[name = string("op_11172_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11172_end_mask_0 = const()[name = string("op_11172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11172_cast_fp16 = slice_by_index(begin = var_11172_begin_0, end = var_11172_end_0, end_mask = var_11172_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11172_cast_fp16")]; + tensor var_11179_begin_0 = const()[name = string("op_11179_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11179_end_0 = const()[name = string("op_11179_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11179_end_mask_0 = const()[name = string("op_11179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11179_cast_fp16 = slice_by_index(begin = var_11179_begin_0, end = var_11179_end_0, end_mask = var_11179_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11179_cast_fp16")]; + tensor var_11186_begin_0 = const()[name = string("op_11186_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11186_end_0 = const()[name = string("op_11186_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11186_end_mask_0 = const()[name = string("op_11186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11186_cast_fp16 = slice_by_index(begin = var_11186_begin_0, end = var_11186_end_0, end_mask = var_11186_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11186_cast_fp16")]; + tensor var_11193_begin_0 = const()[name = string("op_11193_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11193_end_0 = const()[name = string("op_11193_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11193_end_mask_0 = const()[name = string("op_11193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11193_cast_fp16 = slice_by_index(begin = var_11193_begin_0, end = var_11193_end_0, end_mask = var_11193_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11193_cast_fp16")]; + tensor var_11200_begin_0 = const()[name = string("op_11200_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11200_end_0 = const()[name = string("op_11200_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11200_end_mask_0 = const()[name = string("op_11200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11200_cast_fp16 = slice_by_index(begin = var_11200_begin_0, end = var_11200_end_0, end_mask = var_11200_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11200_cast_fp16")]; + tensor var_11207_begin_0 = const()[name = string("op_11207_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11207_end_0 = const()[name = string("op_11207_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11207_end_mask_0 = const()[name = string("op_11207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11207_cast_fp16 = slice_by_index(begin = var_11207_begin_0, end = var_11207_end_0, end_mask = var_11207_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11207_cast_fp16")]; + tensor var_11214_begin_0 = const()[name = string("op_11214_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11214_end_0 = const()[name = string("op_11214_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11214_end_mask_0 = const()[name = string("op_11214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11214_cast_fp16 = slice_by_index(begin = var_11214_begin_0, end = var_11214_end_0, end_mask = var_11214_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11214_cast_fp16")]; + tensor var_11221_begin_0 = const()[name = string("op_11221_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11221_end_0 = const()[name = string("op_11221_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11221_end_mask_0 = const()[name = string("op_11221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11221_cast_fp16 = slice_by_index(begin = var_11221_begin_0, end = var_11221_end_0, end_mask = var_11221_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11221_cast_fp16")]; + tensor var_11228_begin_0 = const()[name = string("op_11228_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11228_end_0 = const()[name = string("op_11228_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11228_end_mask_0 = const()[name = string("op_11228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11228_cast_fp16 = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11228_cast_fp16")]; + tensor var_11235_begin_0 = const()[name = string("op_11235_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11235_end_0 = const()[name = string("op_11235_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11235_end_mask_0 = const()[name = string("op_11235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11235_cast_fp16 = slice_by_index(begin = var_11235_begin_0, end = var_11235_end_0, end_mask = var_11235_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11235_cast_fp16")]; + tensor var_11242_begin_0 = const()[name = string("op_11242_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11242_end_0 = const()[name = string("op_11242_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11242_end_mask_0 = const()[name = string("op_11242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11242_cast_fp16 = slice_by_index(begin = var_11242_begin_0, end = var_11242_end_0, end_mask = var_11242_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11242_cast_fp16")]; + tensor var_11249_begin_0 = const()[name = string("op_11249_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11249_end_0 = const()[name = string("op_11249_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11249_end_mask_0 = const()[name = string("op_11249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11249_cast_fp16 = slice_by_index(begin = var_11249_begin_0, end = var_11249_end_0, end_mask = var_11249_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11249_cast_fp16")]; + tensor var_11256_begin_0 = const()[name = string("op_11256_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11256_end_0 = const()[name = string("op_11256_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11256_end_mask_0 = const()[name = string("op_11256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11256_cast_fp16 = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11256_cast_fp16")]; + tensor var_11263_begin_0 = const()[name = string("op_11263_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11263_end_0 = const()[name = string("op_11263_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11263_end_mask_0 = const()[name = string("op_11263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11263_cast_fp16 = slice_by_index(begin = var_11263_begin_0, end = var_11263_end_0, end_mask = var_11263_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11263_cast_fp16")]; + tensor var_11270_begin_0 = const()[name = string("op_11270_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11270_end_0 = const()[name = string("op_11270_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11270_end_mask_0 = const()[name = string("op_11270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11270_cast_fp16 = slice_by_index(begin = var_11270_begin_0, end = var_11270_end_0, end_mask = var_11270_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11270_cast_fp16")]; + tensor var_11277_begin_0 = const()[name = string("op_11277_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11277_end_0 = const()[name = string("op_11277_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11277_end_mask_0 = const()[name = string("op_11277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11277_cast_fp16 = slice_by_index(begin = var_11277_begin_0, end = var_11277_end_0, end_mask = var_11277_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11277_cast_fp16")]; + tensor var_11284_begin_0 = const()[name = string("op_11284_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11284_end_0 = const()[name = string("op_11284_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11284_end_mask_0 = const()[name = string("op_11284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11284_cast_fp16 = slice_by_index(begin = var_11284_begin_0, end = var_11284_end_0, end_mask = var_11284_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11284_cast_fp16")]; + tensor var_11291_begin_0 = const()[name = string("op_11291_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11291_end_0 = const()[name = string("op_11291_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11291_end_mask_0 = const()[name = string("op_11291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11291_cast_fp16 = slice_by_index(begin = var_11291_begin_0, end = var_11291_end_0, end_mask = var_11291_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11291_cast_fp16")]; + tensor var_11298_begin_0 = const()[name = string("op_11298_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11298_end_0 = const()[name = string("op_11298_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11298_end_mask_0 = const()[name = string("op_11298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11298_cast_fp16 = slice_by_index(begin = var_11298_begin_0, end = var_11298_end_0, end_mask = var_11298_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11298_cast_fp16")]; + tensor var_11305_begin_0 = const()[name = string("op_11305_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11305_end_0 = const()[name = string("op_11305_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11305_end_mask_0 = const()[name = string("op_11305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11305_cast_fp16 = slice_by_index(begin = var_11305_begin_0, end = var_11305_end_0, end_mask = var_11305_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11305_cast_fp16")]; + tensor var_11312_begin_0 = const()[name = string("op_11312_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11312_end_0 = const()[name = string("op_11312_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11312_end_mask_0 = const()[name = string("op_11312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11312_cast_fp16 = slice_by_index(begin = var_11312_begin_0, end = var_11312_end_0, end_mask = var_11312_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11312_cast_fp16")]; + tensor var_11319_begin_0 = const()[name = string("op_11319_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11319_end_0 = const()[name = string("op_11319_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11319_end_mask_0 = const()[name = string("op_11319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11319_cast_fp16 = slice_by_index(begin = var_11319_begin_0, end = var_11319_end_0, end_mask = var_11319_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11319_cast_fp16")]; + tensor var_11326_begin_0 = const()[name = string("op_11326_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11326_end_0 = const()[name = string("op_11326_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11326_end_mask_0 = const()[name = string("op_11326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11326_cast_fp16 = slice_by_index(begin = var_11326_begin_0, end = var_11326_end_0, end_mask = var_11326_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11326_cast_fp16")]; + tensor var_11333_begin_0 = const()[name = string("op_11333_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11333_end_0 = const()[name = string("op_11333_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11333_end_mask_0 = const()[name = string("op_11333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11333_cast_fp16 = slice_by_index(begin = var_11333_begin_0, end = var_11333_end_0, end_mask = var_11333_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11333_cast_fp16")]; + tensor var_11340_begin_0 = const()[name = string("op_11340_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11340_end_0 = const()[name = string("op_11340_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11340_end_mask_0 = const()[name = string("op_11340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11340_cast_fp16 = slice_by_index(begin = var_11340_begin_0, end = var_11340_end_0, end_mask = var_11340_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11340_cast_fp16")]; + tensor var_11347_begin_0 = const()[name = string("op_11347_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11347_end_0 = const()[name = string("op_11347_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11347_end_mask_0 = const()[name = string("op_11347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11347_cast_fp16 = slice_by_index(begin = var_11347_begin_0, end = var_11347_end_0, end_mask = var_11347_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11347_cast_fp16")]; + tensor var_11354_begin_0 = const()[name = string("op_11354_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11354_end_0 = const()[name = string("op_11354_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11354_end_mask_0 = const()[name = string("op_11354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11354_cast_fp16 = slice_by_index(begin = var_11354_begin_0, end = var_11354_end_0, end_mask = var_11354_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11354_cast_fp16")]; + tensor var_11361_begin_0 = const()[name = string("op_11361_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11361_end_0 = const()[name = string("op_11361_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11361_end_mask_0 = const()[name = string("op_11361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11361_cast_fp16 = slice_by_index(begin = var_11361_begin_0, end = var_11361_end_0, end_mask = var_11361_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11361_cast_fp16")]; + tensor var_11368_begin_0 = const()[name = string("op_11368_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11368_end_0 = const()[name = string("op_11368_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11368_end_mask_0 = const()[name = string("op_11368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11368_cast_fp16 = slice_by_index(begin = var_11368_begin_0, end = var_11368_end_0, end_mask = var_11368_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11368_cast_fp16")]; + tensor var_11375_begin_0 = const()[name = string("op_11375_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11375_end_0 = const()[name = string("op_11375_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11375_end_mask_0 = const()[name = string("op_11375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11375_cast_fp16 = slice_by_index(begin = var_11375_begin_0, end = var_11375_end_0, end_mask = var_11375_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11375_cast_fp16")]; + tensor var_11382_begin_0 = const()[name = string("op_11382_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11382_end_0 = const()[name = string("op_11382_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11382_end_mask_0 = const()[name = string("op_11382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11382_cast_fp16 = slice_by_index(begin = var_11382_begin_0, end = var_11382_end_0, end_mask = var_11382_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11382_cast_fp16")]; + tensor var_11389_begin_0 = const()[name = string("op_11389_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11389_end_0 = const()[name = string("op_11389_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11389_end_mask_0 = const()[name = string("op_11389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11389_cast_fp16 = slice_by_index(begin = var_11389_begin_0, end = var_11389_end_0, end_mask = var_11389_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11389_cast_fp16")]; + tensor var_11396_begin_0 = const()[name = string("op_11396_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11396_end_0 = const()[name = string("op_11396_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11396_end_mask_0 = const()[name = string("op_11396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11396_cast_fp16 = slice_by_index(begin = var_11396_begin_0, end = var_11396_end_0, end_mask = var_11396_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11396_cast_fp16")]; + tensor var_11403_begin_0 = const()[name = string("op_11403_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11403_end_0 = const()[name = string("op_11403_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11403_end_mask_0 = const()[name = string("op_11403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11403_cast_fp16 = slice_by_index(begin = var_11403_begin_0, end = var_11403_end_0, end_mask = var_11403_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11403_cast_fp16")]; + tensor var_11410_begin_0 = const()[name = string("op_11410_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11410_end_0 = const()[name = string("op_11410_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11410_end_mask_0 = const()[name = string("op_11410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11410_cast_fp16 = slice_by_index(begin = var_11410_begin_0, end = var_11410_end_0, end_mask = var_11410_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11410_cast_fp16")]; + tensor var_11417_begin_0 = const()[name = string("op_11417_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11417_end_0 = const()[name = string("op_11417_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11417_end_mask_0 = const()[name = string("op_11417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11417_cast_fp16 = slice_by_index(begin = var_11417_begin_0, end = var_11417_end_0, end_mask = var_11417_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11417_cast_fp16")]; + tensor var_11424_begin_0 = const()[name = string("op_11424_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11424_end_0 = const()[name = string("op_11424_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11424_end_mask_0 = const()[name = string("op_11424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11424_cast_fp16 = slice_by_index(begin = var_11424_begin_0, end = var_11424_end_0, end_mask = var_11424_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11424_cast_fp16")]; + tensor var_11431_begin_0 = const()[name = string("op_11431_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11431_end_0 = const()[name = string("op_11431_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11431_end_mask_0 = const()[name = string("op_11431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11431_cast_fp16 = slice_by_index(begin = var_11431_begin_0, end = var_11431_end_0, end_mask = var_11431_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11431_cast_fp16")]; + tensor var_11438_begin_0 = const()[name = string("op_11438_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11438_end_0 = const()[name = string("op_11438_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11438_end_mask_0 = const()[name = string("op_11438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11438_cast_fp16 = slice_by_index(begin = var_11438_begin_0, end = var_11438_end_0, end_mask = var_11438_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11438_cast_fp16")]; + tensor var_11445_begin_0 = const()[name = string("op_11445_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11445_end_0 = const()[name = string("op_11445_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11445_end_mask_0 = const()[name = string("op_11445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11445_cast_fp16 = slice_by_index(begin = var_11445_begin_0, end = var_11445_end_0, end_mask = var_11445_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11445_cast_fp16")]; + tensor var_11452_begin_0 = const()[name = string("op_11452_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11452_end_0 = const()[name = string("op_11452_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11452_end_mask_0 = const()[name = string("op_11452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11452_cast_fp16 = slice_by_index(begin = var_11452_begin_0, end = var_11452_end_0, end_mask = var_11452_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11452_cast_fp16")]; + tensor var_11459_begin_0 = const()[name = string("op_11459_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11459_end_0 = const()[name = string("op_11459_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11459_end_mask_0 = const()[name = string("op_11459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11459_cast_fp16 = slice_by_index(begin = var_11459_begin_0, end = var_11459_end_0, end_mask = var_11459_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11459_cast_fp16")]; + tensor var_11466_begin_0 = const()[name = string("op_11466_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11466_end_0 = const()[name = string("op_11466_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11466_end_mask_0 = const()[name = string("op_11466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11466_cast_fp16 = slice_by_index(begin = var_11466_begin_0, end = var_11466_end_0, end_mask = var_11466_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11466_cast_fp16")]; + tensor var_11473_begin_0 = const()[name = string("op_11473_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11473_end_0 = const()[name = string("op_11473_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11473_end_mask_0 = const()[name = string("op_11473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11473_cast_fp16 = slice_by_index(begin = var_11473_begin_0, end = var_11473_end_0, end_mask = var_11473_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11473_cast_fp16")]; + tensor var_11480_begin_0 = const()[name = string("op_11480_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11480_end_0 = const()[name = string("op_11480_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11480_end_mask_0 = const()[name = string("op_11480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11480_cast_fp16 = slice_by_index(begin = var_11480_begin_0, end = var_11480_end_0, end_mask = var_11480_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11480_cast_fp16")]; + tensor var_11487_begin_0 = const()[name = string("op_11487_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11487_end_0 = const()[name = string("op_11487_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11487_end_mask_0 = const()[name = string("op_11487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11487_cast_fp16 = slice_by_index(begin = var_11487_begin_0, end = var_11487_end_0, end_mask = var_11487_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11487_cast_fp16")]; + tensor var_11494_begin_0 = const()[name = string("op_11494_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11494_end_0 = const()[name = string("op_11494_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11494_end_mask_0 = const()[name = string("op_11494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11494_cast_fp16 = slice_by_index(begin = var_11494_begin_0, end = var_11494_end_0, end_mask = var_11494_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11494_cast_fp16")]; + tensor var_11501_begin_0 = const()[name = string("op_11501_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11501_end_0 = const()[name = string("op_11501_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11501_end_mask_0 = const()[name = string("op_11501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11501_cast_fp16 = slice_by_index(begin = var_11501_begin_0, end = var_11501_end_0, end_mask = var_11501_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11501_cast_fp16")]; + tensor var_11508_begin_0 = const()[name = string("op_11508_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11508_end_0 = const()[name = string("op_11508_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11508_end_mask_0 = const()[name = string("op_11508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11508_cast_fp16 = slice_by_index(begin = var_11508_begin_0, end = var_11508_end_0, end_mask = var_11508_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11508_cast_fp16")]; + tensor var_11515_begin_0 = const()[name = string("op_11515_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11515_end_0 = const()[name = string("op_11515_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11515_end_mask_0 = const()[name = string("op_11515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11515_cast_fp16 = slice_by_index(begin = var_11515_begin_0, end = var_11515_end_0, end_mask = var_11515_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11515_cast_fp16")]; + tensor var_11522_begin_0 = const()[name = string("op_11522_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11522_end_0 = const()[name = string("op_11522_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11522_end_mask_0 = const()[name = string("op_11522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11522_cast_fp16 = slice_by_index(begin = var_11522_begin_0, end = var_11522_end_0, end_mask = var_11522_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11522_cast_fp16")]; + tensor var_11529_begin_0 = const()[name = string("op_11529_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11529_end_0 = const()[name = string("op_11529_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11529_end_mask_0 = const()[name = string("op_11529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11529_cast_fp16 = slice_by_index(begin = var_11529_begin_0, end = var_11529_end_0, end_mask = var_11529_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11529_cast_fp16")]; + tensor var_11536_begin_0 = const()[name = string("op_11536_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11536_end_0 = const()[name = string("op_11536_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11536_end_mask_0 = const()[name = string("op_11536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11536_cast_fp16 = slice_by_index(begin = var_11536_begin_0, end = var_11536_end_0, end_mask = var_11536_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11536_cast_fp16")]; + tensor var_11543_begin_0 = const()[name = string("op_11543_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11543_end_0 = const()[name = string("op_11543_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11543_end_mask_0 = const()[name = string("op_11543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11543_cast_fp16 = slice_by_index(begin = var_11543_begin_0, end = var_11543_end_0, end_mask = var_11543_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11543_cast_fp16")]; + tensor var_11550_begin_0 = const()[name = string("op_11550_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11550_end_0 = const()[name = string("op_11550_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11550_end_mask_0 = const()[name = string("op_11550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11550_cast_fp16 = slice_by_index(begin = var_11550_begin_0, end = var_11550_end_0, end_mask = var_11550_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11550_cast_fp16")]; + tensor var_11557_begin_0 = const()[name = string("op_11557_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11557_end_0 = const()[name = string("op_11557_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11557_end_mask_0 = const()[name = string("op_11557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11557_cast_fp16 = slice_by_index(begin = var_11557_begin_0, end = var_11557_end_0, end_mask = var_11557_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11557_cast_fp16")]; + tensor var_11564_begin_0 = const()[name = string("op_11564_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11564_end_0 = const()[name = string("op_11564_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11564_end_mask_0 = const()[name = string("op_11564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11564_cast_fp16 = slice_by_index(begin = var_11564_begin_0, end = var_11564_end_0, end_mask = var_11564_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11564_cast_fp16")]; + tensor var_11571_begin_0 = const()[name = string("op_11571_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11571_end_0 = const()[name = string("op_11571_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11571_end_mask_0 = const()[name = string("op_11571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11571_cast_fp16 = slice_by_index(begin = var_11571_begin_0, end = var_11571_end_0, end_mask = var_11571_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11571_cast_fp16")]; + tensor var_11578_begin_0 = const()[name = string("op_11578_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11578_end_0 = const()[name = string("op_11578_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11578_end_mask_0 = const()[name = string("op_11578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11578_cast_fp16 = slice_by_index(begin = var_11578_begin_0, end = var_11578_end_0, end_mask = var_11578_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11578_cast_fp16")]; + tensor var_11585_begin_0 = const()[name = string("op_11585_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11585_end_0 = const()[name = string("op_11585_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11585_end_mask_0 = const()[name = string("op_11585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11585_cast_fp16 = slice_by_index(begin = var_11585_begin_0, end = var_11585_end_0, end_mask = var_11585_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11585_cast_fp16")]; + tensor k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11590_begin_0 = const()[name = string("op_11590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11590_end_0 = const()[name = string("op_11590_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_11590_end_mask_0 = const()[name = string("op_11590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = string("transpose_24")]; + tensor var_11590_cast_fp16 = slice_by_index(begin = var_11590_begin_0, end = var_11590_end_0, end_mask = var_11590_end_mask_0, x = k_15_cast_fp16)[name = string("op_11590_cast_fp16")]; + tensor var_11594_begin_0 = const()[name = string("op_11594_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11594_end_0 = const()[name = string("op_11594_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_11594_end_mask_0 = const()[name = string("op_11594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11594_cast_fp16 = slice_by_index(begin = var_11594_begin_0, end = var_11594_end_0, end_mask = var_11594_end_mask_0, x = k_15_cast_fp16)[name = string("op_11594_cast_fp16")]; + tensor var_11598_begin_0 = const()[name = string("op_11598_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11598_end_0 = const()[name = string("op_11598_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_11598_end_mask_0 = const()[name = string("op_11598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11598_cast_fp16 = slice_by_index(begin = var_11598_begin_0, end = var_11598_end_0, end_mask = var_11598_end_mask_0, x = k_15_cast_fp16)[name = string("op_11598_cast_fp16")]; + tensor var_11602_begin_0 = const()[name = string("op_11602_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11602_end_0 = const()[name = string("op_11602_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_11602_end_mask_0 = const()[name = string("op_11602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11602_cast_fp16 = slice_by_index(begin = var_11602_begin_0, end = var_11602_end_0, end_mask = var_11602_end_mask_0, x = k_15_cast_fp16)[name = string("op_11602_cast_fp16")]; + tensor var_11606_begin_0 = const()[name = string("op_11606_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11606_end_0 = const()[name = string("op_11606_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_11606_end_mask_0 = const()[name = string("op_11606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11606_cast_fp16 = slice_by_index(begin = var_11606_begin_0, end = var_11606_end_0, end_mask = var_11606_end_mask_0, x = k_15_cast_fp16)[name = string("op_11606_cast_fp16")]; + tensor var_11610_begin_0 = const()[name = string("op_11610_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11610_end_0 = const()[name = string("op_11610_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_11610_end_mask_0 = const()[name = string("op_11610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11610_cast_fp16 = slice_by_index(begin = var_11610_begin_0, end = var_11610_end_0, end_mask = var_11610_end_mask_0, x = k_15_cast_fp16)[name = string("op_11610_cast_fp16")]; + tensor var_11614_begin_0 = const()[name = string("op_11614_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11614_end_0 = const()[name = string("op_11614_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_11614_end_mask_0 = const()[name = string("op_11614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11614_cast_fp16 = slice_by_index(begin = var_11614_begin_0, end = var_11614_end_0, end_mask = var_11614_end_mask_0, x = k_15_cast_fp16)[name = string("op_11614_cast_fp16")]; + tensor var_11618_begin_0 = const()[name = string("op_11618_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11618_end_0 = const()[name = string("op_11618_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_11618_end_mask_0 = const()[name = string("op_11618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11618_cast_fp16 = slice_by_index(begin = var_11618_begin_0, end = var_11618_end_0, end_mask = var_11618_end_mask_0, x = k_15_cast_fp16)[name = string("op_11618_cast_fp16")]; + tensor var_11622_begin_0 = const()[name = string("op_11622_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11622_end_0 = const()[name = string("op_11622_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_11622_end_mask_0 = const()[name = string("op_11622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11622_cast_fp16 = slice_by_index(begin = var_11622_begin_0, end = var_11622_end_0, end_mask = var_11622_end_mask_0, x = k_15_cast_fp16)[name = string("op_11622_cast_fp16")]; + tensor var_11626_begin_0 = const()[name = string("op_11626_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11626_end_0 = const()[name = string("op_11626_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_11626_end_mask_0 = const()[name = string("op_11626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11626_cast_fp16 = slice_by_index(begin = var_11626_begin_0, end = var_11626_end_0, end_mask = var_11626_end_mask_0, x = k_15_cast_fp16)[name = string("op_11626_cast_fp16")]; + tensor var_11630_begin_0 = const()[name = string("op_11630_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11630_end_0 = const()[name = string("op_11630_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_11630_end_mask_0 = const()[name = string("op_11630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11630_cast_fp16 = slice_by_index(begin = var_11630_begin_0, end = var_11630_end_0, end_mask = var_11630_end_mask_0, x = k_15_cast_fp16)[name = string("op_11630_cast_fp16")]; + tensor var_11634_begin_0 = const()[name = string("op_11634_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11634_end_0 = const()[name = string("op_11634_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_11634_end_mask_0 = const()[name = string("op_11634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11634_cast_fp16 = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = k_15_cast_fp16)[name = string("op_11634_cast_fp16")]; + tensor var_11638_begin_0 = const()[name = string("op_11638_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_11638_end_0 = const()[name = string("op_11638_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_11638_end_mask_0 = const()[name = string("op_11638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11638_cast_fp16 = slice_by_index(begin = var_11638_begin_0, end = var_11638_end_0, end_mask = var_11638_end_mask_0, x = k_15_cast_fp16)[name = string("op_11638_cast_fp16")]; + tensor var_11642_begin_0 = const()[name = string("op_11642_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_11642_end_0 = const()[name = string("op_11642_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_11642_end_mask_0 = const()[name = string("op_11642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11642_cast_fp16 = slice_by_index(begin = var_11642_begin_0, end = var_11642_end_0, end_mask = var_11642_end_mask_0, x = k_15_cast_fp16)[name = string("op_11642_cast_fp16")]; + tensor var_11646_begin_0 = const()[name = string("op_11646_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_11646_end_0 = const()[name = string("op_11646_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_11646_end_mask_0 = const()[name = string("op_11646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11646_cast_fp16 = slice_by_index(begin = var_11646_begin_0, end = var_11646_end_0, end_mask = var_11646_end_mask_0, x = k_15_cast_fp16)[name = string("op_11646_cast_fp16")]; + tensor var_11650_begin_0 = const()[name = string("op_11650_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_11650_end_0 = const()[name = string("op_11650_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_11650_end_mask_0 = const()[name = string("op_11650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11650_cast_fp16 = slice_by_index(begin = var_11650_begin_0, end = var_11650_end_0, end_mask = var_11650_end_mask_0, x = k_15_cast_fp16)[name = string("op_11650_cast_fp16")]; + tensor var_11654_begin_0 = const()[name = string("op_11654_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_11654_end_0 = const()[name = string("op_11654_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_11654_end_mask_0 = const()[name = string("op_11654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11654_cast_fp16 = slice_by_index(begin = var_11654_begin_0, end = var_11654_end_0, end_mask = var_11654_end_mask_0, x = k_15_cast_fp16)[name = string("op_11654_cast_fp16")]; + tensor var_11658_begin_0 = const()[name = string("op_11658_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_11658_end_0 = const()[name = string("op_11658_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_11658_end_mask_0 = const()[name = string("op_11658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11658_cast_fp16 = slice_by_index(begin = var_11658_begin_0, end = var_11658_end_0, end_mask = var_11658_end_mask_0, x = k_15_cast_fp16)[name = string("op_11658_cast_fp16")]; + tensor var_11662_begin_0 = const()[name = string("op_11662_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_11662_end_0 = const()[name = string("op_11662_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_11662_end_mask_0 = const()[name = string("op_11662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11662_cast_fp16 = slice_by_index(begin = var_11662_begin_0, end = var_11662_end_0, end_mask = var_11662_end_mask_0, x = k_15_cast_fp16)[name = string("op_11662_cast_fp16")]; + tensor var_11666_begin_0 = const()[name = string("op_11666_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_11666_end_0 = const()[name = string("op_11666_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_11666_end_mask_0 = const()[name = string("op_11666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11666_cast_fp16 = slice_by_index(begin = var_11666_begin_0, end = var_11666_end_0, end_mask = var_11666_end_mask_0, x = k_15_cast_fp16)[name = string("op_11666_cast_fp16")]; + tensor var_11668_begin_0 = const()[name = string("op_11668_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11668_end_0 = const()[name = string("op_11668_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11668_end_mask_0 = const()[name = string("op_11668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11668_cast_fp16 = slice_by_index(begin = var_11668_begin_0, end = var_11668_end_0, end_mask = var_11668_end_mask_0, x = value_15_cast_fp16)[name = string("op_11668_cast_fp16")]; + tensor var_11672_begin_0 = const()[name = string("op_11672_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11672_end_0 = const()[name = string("op_11672_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_11672_end_mask_0 = const()[name = string("op_11672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11672_cast_fp16 = slice_by_index(begin = var_11672_begin_0, end = var_11672_end_0, end_mask = var_11672_end_mask_0, x = value_15_cast_fp16)[name = string("op_11672_cast_fp16")]; + tensor var_11676_begin_0 = const()[name = string("op_11676_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11676_end_0 = const()[name = string("op_11676_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_11676_end_mask_0 = const()[name = string("op_11676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11676_cast_fp16 = slice_by_index(begin = var_11676_begin_0, end = var_11676_end_0, end_mask = var_11676_end_mask_0, x = value_15_cast_fp16)[name = string("op_11676_cast_fp16")]; + tensor var_11680_begin_0 = const()[name = string("op_11680_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11680_end_0 = const()[name = string("op_11680_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_11680_end_mask_0 = const()[name = string("op_11680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11680_cast_fp16 = slice_by_index(begin = var_11680_begin_0, end = var_11680_end_0, end_mask = var_11680_end_mask_0, x = value_15_cast_fp16)[name = string("op_11680_cast_fp16")]; + tensor var_11684_begin_0 = const()[name = string("op_11684_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11684_end_0 = const()[name = string("op_11684_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_11684_end_mask_0 = const()[name = string("op_11684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11684_cast_fp16 = slice_by_index(begin = var_11684_begin_0, end = var_11684_end_0, end_mask = var_11684_end_mask_0, x = value_15_cast_fp16)[name = string("op_11684_cast_fp16")]; + tensor var_11688_begin_0 = const()[name = string("op_11688_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11688_end_0 = const()[name = string("op_11688_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_11688_end_mask_0 = const()[name = string("op_11688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11688_cast_fp16 = slice_by_index(begin = var_11688_begin_0, end = var_11688_end_0, end_mask = var_11688_end_mask_0, x = value_15_cast_fp16)[name = string("op_11688_cast_fp16")]; + tensor var_11692_begin_0 = const()[name = string("op_11692_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11692_end_0 = const()[name = string("op_11692_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_11692_end_mask_0 = const()[name = string("op_11692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11692_cast_fp16 = slice_by_index(begin = var_11692_begin_0, end = var_11692_end_0, end_mask = var_11692_end_mask_0, x = value_15_cast_fp16)[name = string("op_11692_cast_fp16")]; + tensor var_11696_begin_0 = const()[name = string("op_11696_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11696_end_0 = const()[name = string("op_11696_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_11696_end_mask_0 = const()[name = string("op_11696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11696_cast_fp16 = slice_by_index(begin = var_11696_begin_0, end = var_11696_end_0, end_mask = var_11696_end_mask_0, x = value_15_cast_fp16)[name = string("op_11696_cast_fp16")]; + tensor var_11700_begin_0 = const()[name = string("op_11700_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11700_end_0 = const()[name = string("op_11700_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_11700_end_mask_0 = const()[name = string("op_11700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11700_cast_fp16 = slice_by_index(begin = var_11700_begin_0, end = var_11700_end_0, end_mask = var_11700_end_mask_0, x = value_15_cast_fp16)[name = string("op_11700_cast_fp16")]; + tensor var_11704_begin_0 = const()[name = string("op_11704_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11704_end_0 = const()[name = string("op_11704_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_11704_end_mask_0 = const()[name = string("op_11704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11704_cast_fp16 = slice_by_index(begin = var_11704_begin_0, end = var_11704_end_0, end_mask = var_11704_end_mask_0, x = value_15_cast_fp16)[name = string("op_11704_cast_fp16")]; + tensor var_11708_begin_0 = const()[name = string("op_11708_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11708_end_0 = const()[name = string("op_11708_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_11708_end_mask_0 = const()[name = string("op_11708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11708_cast_fp16 = slice_by_index(begin = var_11708_begin_0, end = var_11708_end_0, end_mask = var_11708_end_mask_0, x = value_15_cast_fp16)[name = string("op_11708_cast_fp16")]; + tensor var_11712_begin_0 = const()[name = string("op_11712_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11712_end_0 = const()[name = string("op_11712_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_11712_end_mask_0 = const()[name = string("op_11712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11712_cast_fp16 = slice_by_index(begin = var_11712_begin_0, end = var_11712_end_0, end_mask = var_11712_end_mask_0, x = value_15_cast_fp16)[name = string("op_11712_cast_fp16")]; + tensor var_11716_begin_0 = const()[name = string("op_11716_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11716_end_0 = const()[name = string("op_11716_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_11716_end_mask_0 = const()[name = string("op_11716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11716_cast_fp16 = slice_by_index(begin = var_11716_begin_0, end = var_11716_end_0, end_mask = var_11716_end_mask_0, x = value_15_cast_fp16)[name = string("op_11716_cast_fp16")]; + tensor var_11720_begin_0 = const()[name = string("op_11720_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11720_end_0 = const()[name = string("op_11720_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_11720_end_mask_0 = const()[name = string("op_11720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11720_cast_fp16 = slice_by_index(begin = var_11720_begin_0, end = var_11720_end_0, end_mask = var_11720_end_mask_0, x = value_15_cast_fp16)[name = string("op_11720_cast_fp16")]; + tensor var_11724_begin_0 = const()[name = string("op_11724_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11724_end_0 = const()[name = string("op_11724_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_11724_end_mask_0 = const()[name = string("op_11724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11724_cast_fp16 = slice_by_index(begin = var_11724_begin_0, end = var_11724_end_0, end_mask = var_11724_end_mask_0, x = value_15_cast_fp16)[name = string("op_11724_cast_fp16")]; + tensor var_11728_begin_0 = const()[name = string("op_11728_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11728_end_0 = const()[name = string("op_11728_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_11728_end_mask_0 = const()[name = string("op_11728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11728_cast_fp16 = slice_by_index(begin = var_11728_begin_0, end = var_11728_end_0, end_mask = var_11728_end_mask_0, x = value_15_cast_fp16)[name = string("op_11728_cast_fp16")]; + tensor var_11732_begin_0 = const()[name = string("op_11732_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11732_end_0 = const()[name = string("op_11732_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_11732_end_mask_0 = const()[name = string("op_11732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11732_cast_fp16 = slice_by_index(begin = var_11732_begin_0, end = var_11732_end_0, end_mask = var_11732_end_mask_0, x = value_15_cast_fp16)[name = string("op_11732_cast_fp16")]; + tensor var_11736_begin_0 = const()[name = string("op_11736_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11736_end_0 = const()[name = string("op_11736_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_11736_end_mask_0 = const()[name = string("op_11736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11736_cast_fp16 = slice_by_index(begin = var_11736_begin_0, end = var_11736_end_0, end_mask = var_11736_end_mask_0, x = value_15_cast_fp16)[name = string("op_11736_cast_fp16")]; + tensor var_11740_begin_0 = const()[name = string("op_11740_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11740_end_0 = const()[name = string("op_11740_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_11740_end_mask_0 = const()[name = string("op_11740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11740_cast_fp16 = slice_by_index(begin = var_11740_begin_0, end = var_11740_end_0, end_mask = var_11740_end_mask_0, x = value_15_cast_fp16)[name = string("op_11740_cast_fp16")]; + tensor var_11744_begin_0 = const()[name = string("op_11744_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11744_end_0 = const()[name = string("op_11744_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_11744_end_mask_0 = const()[name = string("op_11744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11744_cast_fp16 = slice_by_index(begin = var_11744_begin_0, end = var_11744_end_0, end_mask = var_11744_end_mask_0, x = value_15_cast_fp16)[name = string("op_11744_cast_fp16")]; + string _SplitHeadsQ__mh_w_1121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1121_equation_0, values = (var_11590_cast_fp16, var_11032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1121_cast_fp16")]; + string _SplitHeadsQ__mh_w_1123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1123_equation_0, values = (var_11590_cast_fp16, var_11039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1123_cast_fp16")]; + string _SplitHeadsQ__mh_w_1125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1125_equation_0, values = (var_11590_cast_fp16, var_11046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1125_cast_fp16")]; + string _SplitHeadsQ__mh_w_1127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1127_equation_0, values = (var_11590_cast_fp16, var_11053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1127_cast_fp16")]; + string _SplitHeadsQ__mh_w_1129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1129_equation_0, values = (var_11594_cast_fp16, var_11060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1129_cast_fp16")]; + string _SplitHeadsQ__mh_w_1131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1131_equation_0, values = (var_11594_cast_fp16, var_11067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1131_cast_fp16")]; + string _SplitHeadsQ__mh_w_1133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1133_equation_0, values = (var_11594_cast_fp16, var_11074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1133_cast_fp16")]; + string _SplitHeadsQ__mh_w_1135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1135_equation_0, values = (var_11594_cast_fp16, var_11081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1135_cast_fp16")]; + string _SplitHeadsQ__mh_w_1137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1137_equation_0, values = (var_11598_cast_fp16, var_11088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1137_cast_fp16")]; + string _SplitHeadsQ__mh_w_1139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1139_equation_0, values = (var_11598_cast_fp16, var_11095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1139_cast_fp16")]; + string _SplitHeadsQ__mh_w_1141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1141_equation_0, values = (var_11598_cast_fp16, var_11102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1141_cast_fp16")]; + string _SplitHeadsQ__mh_w_1143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1143_equation_0, values = (var_11598_cast_fp16, var_11109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1143_cast_fp16")]; + string _SplitHeadsQ__mh_w_1145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1145_equation_0, values = (var_11602_cast_fp16, var_11116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1145_cast_fp16")]; + string _SplitHeadsQ__mh_w_1147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1147_equation_0, values = (var_11602_cast_fp16, var_11123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1147_cast_fp16")]; + string _SplitHeadsQ__mh_w_1149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1149_equation_0, values = (var_11602_cast_fp16, var_11130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1149_cast_fp16")]; + string _SplitHeadsQ__mh_w_1151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1151_equation_0, values = (var_11602_cast_fp16, var_11137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1151_cast_fp16")]; + string _SplitHeadsQ__mh_w_1153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1153_equation_0, values = (var_11606_cast_fp16, var_11144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1153_cast_fp16")]; + string _SplitHeadsQ__mh_w_1155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1155_equation_0, values = (var_11606_cast_fp16, var_11151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1155_cast_fp16")]; + string _SplitHeadsQ__mh_w_1157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1157_equation_0, values = (var_11606_cast_fp16, var_11158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1157_cast_fp16")]; + string _SplitHeadsQ__mh_w_1159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1159_equation_0, values = (var_11606_cast_fp16, var_11165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1159_cast_fp16")]; + string _SplitHeadsQ__mh_w_1161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1161_equation_0, values = (var_11610_cast_fp16, var_11172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1161_cast_fp16")]; + string _SplitHeadsQ__mh_w_1163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1163_equation_0, values = (var_11610_cast_fp16, var_11179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1163_cast_fp16")]; + string _SplitHeadsQ__mh_w_1165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1165_equation_0, values = (var_11610_cast_fp16, var_11186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1165_cast_fp16")]; + string _SplitHeadsQ__mh_w_1167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1167_equation_0, values = (var_11610_cast_fp16, var_11193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1167_cast_fp16")]; + string _SplitHeadsQ__mh_w_1169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1169_equation_0, values = (var_11614_cast_fp16, var_11200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1169_cast_fp16")]; + string _SplitHeadsQ__mh_w_1171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1171_equation_0, values = (var_11614_cast_fp16, var_11207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1171_cast_fp16")]; + string _SplitHeadsQ__mh_w_1173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1173_equation_0, values = (var_11614_cast_fp16, var_11214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1173_cast_fp16")]; + string _SplitHeadsQ__mh_w_1175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1175_equation_0, values = (var_11614_cast_fp16, var_11221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1175_cast_fp16")]; + string _SplitHeadsQ__mh_w_1177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1177_equation_0, values = (var_11618_cast_fp16, var_11228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1177_cast_fp16")]; + string _SplitHeadsQ__mh_w_1179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1179_equation_0, values = (var_11618_cast_fp16, var_11235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1179_cast_fp16")]; + string _SplitHeadsQ__mh_w_1181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1181_equation_0, values = (var_11618_cast_fp16, var_11242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1181_cast_fp16")]; + string _SplitHeadsQ__mh_w_1183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1183_equation_0, values = (var_11618_cast_fp16, var_11249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1183_cast_fp16")]; + string _SplitHeadsQ__mh_w_1185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1185_equation_0, values = (var_11622_cast_fp16, var_11256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1185_cast_fp16")]; + string _SplitHeadsQ__mh_w_1187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1187_equation_0, values = (var_11622_cast_fp16, var_11263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1187_cast_fp16")]; + string _SplitHeadsQ__mh_w_1189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1189_equation_0, values = (var_11622_cast_fp16, var_11270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1189_cast_fp16")]; + string _SplitHeadsQ__mh_w_1191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1191_equation_0, values = (var_11622_cast_fp16, var_11277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1191_cast_fp16")]; + string _SplitHeadsQ__mh_w_1193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1193_equation_0, values = (var_11626_cast_fp16, var_11284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1193_cast_fp16")]; + string _SplitHeadsQ__mh_w_1195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1195_equation_0, values = (var_11626_cast_fp16, var_11291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1195_cast_fp16")]; + string _SplitHeadsQ__mh_w_1197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1197_equation_0, values = (var_11626_cast_fp16, var_11298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1197_cast_fp16")]; + string _SplitHeadsQ__mh_w_1199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1199_equation_0, values = (var_11626_cast_fp16, var_11305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1199_cast_fp16")]; + string _SplitHeadsQ__mh_w_1201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1201_equation_0, values = (var_11630_cast_fp16, var_11312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1201_cast_fp16")]; + string _SplitHeadsQ__mh_w_1203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1203_equation_0, values = (var_11630_cast_fp16, var_11319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1203_cast_fp16")]; + string _SplitHeadsQ__mh_w_1205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1205_equation_0, values = (var_11630_cast_fp16, var_11326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1205_cast_fp16")]; + string _SplitHeadsQ__mh_w_1207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1207_equation_0, values = (var_11630_cast_fp16, var_11333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1207_cast_fp16")]; + string _SplitHeadsQ__mh_w_1209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1209_equation_0, values = (var_11634_cast_fp16, var_11340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1209_cast_fp16")]; + string _SplitHeadsQ__mh_w_1211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1211_equation_0, values = (var_11634_cast_fp16, var_11347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1211_cast_fp16")]; + string _SplitHeadsQ__mh_w_1213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1213_equation_0, values = (var_11634_cast_fp16, var_11354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1213_cast_fp16")]; + string _SplitHeadsQ__mh_w_1215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1215_equation_0, values = (var_11634_cast_fp16, var_11361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1215_cast_fp16")]; + string _SplitHeadsQ__mh_w_1217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1217_equation_0, values = (var_11638_cast_fp16, var_11368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1217_cast_fp16")]; + string _SplitHeadsQ__mh_w_1219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1219_equation_0, values = (var_11638_cast_fp16, var_11375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1219_cast_fp16")]; + string _SplitHeadsQ__mh_w_1221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1221_equation_0, values = (var_11638_cast_fp16, var_11382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1221_cast_fp16")]; + string _SplitHeadsQ__mh_w_1223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1223_equation_0, values = (var_11638_cast_fp16, var_11389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1223_cast_fp16")]; + string _SplitHeadsQ__mh_w_1225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1225_equation_0, values = (var_11642_cast_fp16, var_11396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1225_cast_fp16")]; + string _SplitHeadsQ__mh_w_1227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1227_equation_0, values = (var_11642_cast_fp16, var_11403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1227_cast_fp16")]; + string _SplitHeadsQ__mh_w_1229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1229_equation_0, values = (var_11642_cast_fp16, var_11410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1229_cast_fp16")]; + string _SplitHeadsQ__mh_w_1231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1231_equation_0, values = (var_11642_cast_fp16, var_11417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1231_cast_fp16")]; + string _SplitHeadsQ__mh_w_1233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1233_equation_0, values = (var_11646_cast_fp16, var_11424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1233_cast_fp16")]; + string _SplitHeadsQ__mh_w_1235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1235_equation_0, values = (var_11646_cast_fp16, var_11431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1235_cast_fp16")]; + string _SplitHeadsQ__mh_w_1237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1237_equation_0, values = (var_11646_cast_fp16, var_11438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1237_cast_fp16")]; + string _SplitHeadsQ__mh_w_1239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1239_equation_0, values = (var_11646_cast_fp16, var_11445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1239_cast_fp16")]; + string _SplitHeadsQ__mh_w_1241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1241_equation_0, values = (var_11650_cast_fp16, var_11452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1241_cast_fp16")]; + string _SplitHeadsQ__mh_w_1243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1243_equation_0, values = (var_11650_cast_fp16, var_11459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1243_cast_fp16")]; + string _SplitHeadsQ__mh_w_1245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1245_equation_0, values = (var_11650_cast_fp16, var_11466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1245_cast_fp16")]; + string _SplitHeadsQ__mh_w_1247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1247_equation_0, values = (var_11650_cast_fp16, var_11473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1247_cast_fp16")]; + string _SplitHeadsQ__mh_w_1249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1249_equation_0, values = (var_11654_cast_fp16, var_11480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1249_cast_fp16")]; + string _SplitHeadsQ__mh_w_1251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1251_equation_0, values = (var_11654_cast_fp16, var_11487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1251_cast_fp16")]; + string _SplitHeadsQ__mh_w_1253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1253_equation_0, values = (var_11654_cast_fp16, var_11494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1253_cast_fp16")]; + string _SplitHeadsQ__mh_w_1255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1255_equation_0, values = (var_11654_cast_fp16, var_11501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1255_cast_fp16")]; + string _SplitHeadsQ__mh_w_1257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1257_equation_0, values = (var_11658_cast_fp16, var_11508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1257_cast_fp16")]; + string _SplitHeadsQ__mh_w_1259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1259_equation_0, values = (var_11658_cast_fp16, var_11515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1259_cast_fp16")]; + string _SplitHeadsQ__mh_w_1261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1261_equation_0, values = (var_11658_cast_fp16, var_11522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1261_cast_fp16")]; + string _SplitHeadsQ__mh_w_1263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1263_equation_0, values = (var_11658_cast_fp16, var_11529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1263_cast_fp16")]; + string _SplitHeadsQ__mh_w_1265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1265_equation_0, values = (var_11662_cast_fp16, var_11536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1265_cast_fp16")]; + string _SplitHeadsQ__mh_w_1267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1267_equation_0, values = (var_11662_cast_fp16, var_11543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1267_cast_fp16")]; + string _SplitHeadsQ__mh_w_1269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1269_equation_0, values = (var_11662_cast_fp16, var_11550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1269_cast_fp16")]; + string _SplitHeadsQ__mh_w_1271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1271_equation_0, values = (var_11662_cast_fp16, var_11557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1271_cast_fp16")]; + string _SplitHeadsQ__mh_w_1273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1273_equation_0, values = (var_11666_cast_fp16, var_11564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1273_cast_fp16")]; + string _SplitHeadsQ__mh_w_1275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1275_equation_0, values = (var_11666_cast_fp16, var_11571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1275_cast_fp16")]; + string _SplitHeadsQ__mh_w_1277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1277_equation_0, values = (var_11666_cast_fp16, var_11578_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1277_cast_fp16")]; + string _SplitHeadsQ__mh_w_1279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1279_equation_0, values = (var_11666_cast_fp16, var_11585_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1279_cast_fp16")]; + fp16 var_11907_to_fp16 = const()[name = string("op_11907_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1121_cast_fp16, y = var_11907_to_fp16)[name = string("aw_chunk_1121_cast_fp16")]; + fp16 var_11909_to_fp16 = const()[name = string("op_11909_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1123_cast_fp16, y = var_11909_to_fp16)[name = string("aw_chunk_1123_cast_fp16")]; + fp16 var_11911_to_fp16 = const()[name = string("op_11911_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1125_cast_fp16, y = var_11911_to_fp16)[name = string("aw_chunk_1125_cast_fp16")]; + fp16 var_11913_to_fp16 = const()[name = string("op_11913_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1127_cast_fp16, y = var_11913_to_fp16)[name = string("aw_chunk_1127_cast_fp16")]; + fp16 var_11915_to_fp16 = const()[name = string("op_11915_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1129_cast_fp16, y = var_11915_to_fp16)[name = string("aw_chunk_1129_cast_fp16")]; + fp16 var_11917_to_fp16 = const()[name = string("op_11917_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1131_cast_fp16, y = var_11917_to_fp16)[name = string("aw_chunk_1131_cast_fp16")]; + fp16 var_11919_to_fp16 = const()[name = string("op_11919_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1133_cast_fp16, y = var_11919_to_fp16)[name = string("aw_chunk_1133_cast_fp16")]; + fp16 var_11921_to_fp16 = const()[name = string("op_11921_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1135_cast_fp16, y = var_11921_to_fp16)[name = string("aw_chunk_1135_cast_fp16")]; + fp16 var_11923_to_fp16 = const()[name = string("op_11923_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1137_cast_fp16, y = var_11923_to_fp16)[name = string("aw_chunk_1137_cast_fp16")]; + fp16 var_11925_to_fp16 = const()[name = string("op_11925_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1139_cast_fp16, y = var_11925_to_fp16)[name = string("aw_chunk_1139_cast_fp16")]; + fp16 var_11927_to_fp16 = const()[name = string("op_11927_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1141_cast_fp16, y = var_11927_to_fp16)[name = string("aw_chunk_1141_cast_fp16")]; + fp16 var_11929_to_fp16 = const()[name = string("op_11929_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1143_cast_fp16, y = var_11929_to_fp16)[name = string("aw_chunk_1143_cast_fp16")]; + fp16 var_11931_to_fp16 = const()[name = string("op_11931_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1145_cast_fp16, y = var_11931_to_fp16)[name = string("aw_chunk_1145_cast_fp16")]; + fp16 var_11933_to_fp16 = const()[name = string("op_11933_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1147_cast_fp16, y = var_11933_to_fp16)[name = string("aw_chunk_1147_cast_fp16")]; + fp16 var_11935_to_fp16 = const()[name = string("op_11935_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1149_cast_fp16, y = var_11935_to_fp16)[name = string("aw_chunk_1149_cast_fp16")]; + fp16 var_11937_to_fp16 = const()[name = string("op_11937_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1151_cast_fp16, y = var_11937_to_fp16)[name = string("aw_chunk_1151_cast_fp16")]; + fp16 var_11939_to_fp16 = const()[name = string("op_11939_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1153_cast_fp16, y = var_11939_to_fp16)[name = string("aw_chunk_1153_cast_fp16")]; + fp16 var_11941_to_fp16 = const()[name = string("op_11941_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1155_cast_fp16, y = var_11941_to_fp16)[name = string("aw_chunk_1155_cast_fp16")]; + fp16 var_11943_to_fp16 = const()[name = string("op_11943_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1157_cast_fp16, y = var_11943_to_fp16)[name = string("aw_chunk_1157_cast_fp16")]; + fp16 var_11945_to_fp16 = const()[name = string("op_11945_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1159_cast_fp16, y = var_11945_to_fp16)[name = string("aw_chunk_1159_cast_fp16")]; + fp16 var_11947_to_fp16 = const()[name = string("op_11947_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1161_cast_fp16, y = var_11947_to_fp16)[name = string("aw_chunk_1161_cast_fp16")]; + fp16 var_11949_to_fp16 = const()[name = string("op_11949_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1163_cast_fp16, y = var_11949_to_fp16)[name = string("aw_chunk_1163_cast_fp16")]; + fp16 var_11951_to_fp16 = const()[name = string("op_11951_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1165_cast_fp16, y = var_11951_to_fp16)[name = string("aw_chunk_1165_cast_fp16")]; + fp16 var_11953_to_fp16 = const()[name = string("op_11953_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1167_cast_fp16, y = var_11953_to_fp16)[name = string("aw_chunk_1167_cast_fp16")]; + fp16 var_11955_to_fp16 = const()[name = string("op_11955_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1169_cast_fp16, y = var_11955_to_fp16)[name = string("aw_chunk_1169_cast_fp16")]; + fp16 var_11957_to_fp16 = const()[name = string("op_11957_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1171_cast_fp16, y = var_11957_to_fp16)[name = string("aw_chunk_1171_cast_fp16")]; + fp16 var_11959_to_fp16 = const()[name = string("op_11959_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1173_cast_fp16, y = var_11959_to_fp16)[name = string("aw_chunk_1173_cast_fp16")]; + fp16 var_11961_to_fp16 = const()[name = string("op_11961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1175_cast_fp16, y = var_11961_to_fp16)[name = string("aw_chunk_1175_cast_fp16")]; + fp16 var_11963_to_fp16 = const()[name = string("op_11963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1177_cast_fp16, y = var_11963_to_fp16)[name = string("aw_chunk_1177_cast_fp16")]; + fp16 var_11965_to_fp16 = const()[name = string("op_11965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1179_cast_fp16, y = var_11965_to_fp16)[name = string("aw_chunk_1179_cast_fp16")]; + fp16 var_11967_to_fp16 = const()[name = string("op_11967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1181_cast_fp16, y = var_11967_to_fp16)[name = string("aw_chunk_1181_cast_fp16")]; + fp16 var_11969_to_fp16 = const()[name = string("op_11969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1183_cast_fp16, y = var_11969_to_fp16)[name = string("aw_chunk_1183_cast_fp16")]; + fp16 var_11971_to_fp16 = const()[name = string("op_11971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1185_cast_fp16, y = var_11971_to_fp16)[name = string("aw_chunk_1185_cast_fp16")]; + fp16 var_11973_to_fp16 = const()[name = string("op_11973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1187_cast_fp16, y = var_11973_to_fp16)[name = string("aw_chunk_1187_cast_fp16")]; + fp16 var_11975_to_fp16 = const()[name = string("op_11975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1189_cast_fp16, y = var_11975_to_fp16)[name = string("aw_chunk_1189_cast_fp16")]; + fp16 var_11977_to_fp16 = const()[name = string("op_11977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1191_cast_fp16, y = var_11977_to_fp16)[name = string("aw_chunk_1191_cast_fp16")]; + fp16 var_11979_to_fp16 = const()[name = string("op_11979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1193_cast_fp16, y = var_11979_to_fp16)[name = string("aw_chunk_1193_cast_fp16")]; + fp16 var_11981_to_fp16 = const()[name = string("op_11981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1195_cast_fp16, y = var_11981_to_fp16)[name = string("aw_chunk_1195_cast_fp16")]; + fp16 var_11983_to_fp16 = const()[name = string("op_11983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1197_cast_fp16, y = var_11983_to_fp16)[name = string("aw_chunk_1197_cast_fp16")]; + fp16 var_11985_to_fp16 = const()[name = string("op_11985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1199_cast_fp16, y = var_11985_to_fp16)[name = string("aw_chunk_1199_cast_fp16")]; + fp16 var_11987_to_fp16 = const()[name = string("op_11987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1201_cast_fp16, y = var_11987_to_fp16)[name = string("aw_chunk_1201_cast_fp16")]; + fp16 var_11989_to_fp16 = const()[name = string("op_11989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1203_cast_fp16, y = var_11989_to_fp16)[name = string("aw_chunk_1203_cast_fp16")]; + fp16 var_11991_to_fp16 = const()[name = string("op_11991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1205_cast_fp16, y = var_11991_to_fp16)[name = string("aw_chunk_1205_cast_fp16")]; + fp16 var_11993_to_fp16 = const()[name = string("op_11993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1207_cast_fp16, y = var_11993_to_fp16)[name = string("aw_chunk_1207_cast_fp16")]; + fp16 var_11995_to_fp16 = const()[name = string("op_11995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1209_cast_fp16, y = var_11995_to_fp16)[name = string("aw_chunk_1209_cast_fp16")]; + fp16 var_11997_to_fp16 = const()[name = string("op_11997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1211_cast_fp16, y = var_11997_to_fp16)[name = string("aw_chunk_1211_cast_fp16")]; + fp16 var_11999_to_fp16 = const()[name = string("op_11999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1213_cast_fp16, y = var_11999_to_fp16)[name = string("aw_chunk_1213_cast_fp16")]; + fp16 var_12001_to_fp16 = const()[name = string("op_12001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1215_cast_fp16, y = var_12001_to_fp16)[name = string("aw_chunk_1215_cast_fp16")]; + fp16 var_12003_to_fp16 = const()[name = string("op_12003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1217_cast_fp16, y = var_12003_to_fp16)[name = string("aw_chunk_1217_cast_fp16")]; + fp16 var_12005_to_fp16 = const()[name = string("op_12005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1219_cast_fp16, y = var_12005_to_fp16)[name = string("aw_chunk_1219_cast_fp16")]; + fp16 var_12007_to_fp16 = const()[name = string("op_12007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1221_cast_fp16, y = var_12007_to_fp16)[name = string("aw_chunk_1221_cast_fp16")]; + fp16 var_12009_to_fp16 = const()[name = string("op_12009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1223_cast_fp16, y = var_12009_to_fp16)[name = string("aw_chunk_1223_cast_fp16")]; + fp16 var_12011_to_fp16 = const()[name = string("op_12011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1225_cast_fp16, y = var_12011_to_fp16)[name = string("aw_chunk_1225_cast_fp16")]; + fp16 var_12013_to_fp16 = const()[name = string("op_12013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1227_cast_fp16, y = var_12013_to_fp16)[name = string("aw_chunk_1227_cast_fp16")]; + fp16 var_12015_to_fp16 = const()[name = string("op_12015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1229_cast_fp16, y = var_12015_to_fp16)[name = string("aw_chunk_1229_cast_fp16")]; + fp16 var_12017_to_fp16 = const()[name = string("op_12017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1231_cast_fp16, y = var_12017_to_fp16)[name = string("aw_chunk_1231_cast_fp16")]; + fp16 var_12019_to_fp16 = const()[name = string("op_12019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1233_cast_fp16, y = var_12019_to_fp16)[name = string("aw_chunk_1233_cast_fp16")]; + fp16 var_12021_to_fp16 = const()[name = string("op_12021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1235_cast_fp16, y = var_12021_to_fp16)[name = string("aw_chunk_1235_cast_fp16")]; + fp16 var_12023_to_fp16 = const()[name = string("op_12023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1237_cast_fp16, y = var_12023_to_fp16)[name = string("aw_chunk_1237_cast_fp16")]; + fp16 var_12025_to_fp16 = const()[name = string("op_12025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1239_cast_fp16, y = var_12025_to_fp16)[name = string("aw_chunk_1239_cast_fp16")]; + fp16 var_12027_to_fp16 = const()[name = string("op_12027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1241_cast_fp16, y = var_12027_to_fp16)[name = string("aw_chunk_1241_cast_fp16")]; + fp16 var_12029_to_fp16 = const()[name = string("op_12029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1243_cast_fp16, y = var_12029_to_fp16)[name = string("aw_chunk_1243_cast_fp16")]; + fp16 var_12031_to_fp16 = const()[name = string("op_12031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1245_cast_fp16, y = var_12031_to_fp16)[name = string("aw_chunk_1245_cast_fp16")]; + fp16 var_12033_to_fp16 = const()[name = string("op_12033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1247_cast_fp16, y = var_12033_to_fp16)[name = string("aw_chunk_1247_cast_fp16")]; + fp16 var_12035_to_fp16 = const()[name = string("op_12035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1249_cast_fp16, y = var_12035_to_fp16)[name = string("aw_chunk_1249_cast_fp16")]; + fp16 var_12037_to_fp16 = const()[name = string("op_12037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1251_cast_fp16, y = var_12037_to_fp16)[name = string("aw_chunk_1251_cast_fp16")]; + fp16 var_12039_to_fp16 = const()[name = string("op_12039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1253_cast_fp16, y = var_12039_to_fp16)[name = string("aw_chunk_1253_cast_fp16")]; + fp16 var_12041_to_fp16 = const()[name = string("op_12041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1255_cast_fp16, y = var_12041_to_fp16)[name = string("aw_chunk_1255_cast_fp16")]; + fp16 var_12043_to_fp16 = const()[name = string("op_12043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1257_cast_fp16, y = var_12043_to_fp16)[name = string("aw_chunk_1257_cast_fp16")]; + fp16 var_12045_to_fp16 = const()[name = string("op_12045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1259_cast_fp16, y = var_12045_to_fp16)[name = string("aw_chunk_1259_cast_fp16")]; + fp16 var_12047_to_fp16 = const()[name = string("op_12047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1261_cast_fp16, y = var_12047_to_fp16)[name = string("aw_chunk_1261_cast_fp16")]; + fp16 var_12049_to_fp16 = const()[name = string("op_12049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1263_cast_fp16, y = var_12049_to_fp16)[name = string("aw_chunk_1263_cast_fp16")]; + fp16 var_12051_to_fp16 = const()[name = string("op_12051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1265_cast_fp16, y = var_12051_to_fp16)[name = string("aw_chunk_1265_cast_fp16")]; + fp16 var_12053_to_fp16 = const()[name = string("op_12053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1267_cast_fp16, y = var_12053_to_fp16)[name = string("aw_chunk_1267_cast_fp16")]; + fp16 var_12055_to_fp16 = const()[name = string("op_12055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1269_cast_fp16, y = var_12055_to_fp16)[name = string("aw_chunk_1269_cast_fp16")]; + fp16 var_12057_to_fp16 = const()[name = string("op_12057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1271_cast_fp16, y = var_12057_to_fp16)[name = string("aw_chunk_1271_cast_fp16")]; + fp16 var_12059_to_fp16 = const()[name = string("op_12059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1273_cast_fp16, y = var_12059_to_fp16)[name = string("aw_chunk_1273_cast_fp16")]; + fp16 var_12061_to_fp16 = const()[name = string("op_12061_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1275_cast_fp16, y = var_12061_to_fp16)[name = string("aw_chunk_1275_cast_fp16")]; + fp16 var_12063_to_fp16 = const()[name = string("op_12063_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1277_cast_fp16, y = var_12063_to_fp16)[name = string("aw_chunk_1277_cast_fp16")]; + fp16 var_12065_to_fp16 = const()[name = string("op_12065_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1279_cast_fp16, y = var_12065_to_fp16)[name = string("aw_chunk_1279_cast_fp16")]; + tensor var_12067_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1121_cast_fp16)[name = string("op_12067_cast_fp16")]; + tensor var_12068_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1123_cast_fp16)[name = string("op_12068_cast_fp16")]; + tensor var_12069_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1125_cast_fp16)[name = string("op_12069_cast_fp16")]; + tensor var_12070_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1127_cast_fp16)[name = string("op_12070_cast_fp16")]; + tensor var_12071_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1129_cast_fp16)[name = string("op_12071_cast_fp16")]; + tensor var_12072_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1131_cast_fp16)[name = string("op_12072_cast_fp16")]; + tensor var_12073_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1133_cast_fp16)[name = string("op_12073_cast_fp16")]; + tensor var_12074_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1135_cast_fp16)[name = string("op_12074_cast_fp16")]; + tensor var_12075_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1137_cast_fp16)[name = string("op_12075_cast_fp16")]; + tensor var_12076_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1139_cast_fp16)[name = string("op_12076_cast_fp16")]; + tensor var_12077_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1141_cast_fp16)[name = string("op_12077_cast_fp16")]; + tensor var_12078_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1143_cast_fp16)[name = string("op_12078_cast_fp16")]; + tensor var_12079_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1145_cast_fp16)[name = string("op_12079_cast_fp16")]; + tensor var_12080_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1147_cast_fp16)[name = string("op_12080_cast_fp16")]; + tensor var_12081_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1149_cast_fp16)[name = string("op_12081_cast_fp16")]; + tensor var_12082_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1151_cast_fp16)[name = string("op_12082_cast_fp16")]; + tensor var_12083_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1153_cast_fp16)[name = string("op_12083_cast_fp16")]; + tensor var_12084_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1155_cast_fp16)[name = string("op_12084_cast_fp16")]; + tensor var_12085_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1157_cast_fp16)[name = string("op_12085_cast_fp16")]; + tensor var_12086_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1159_cast_fp16)[name = string("op_12086_cast_fp16")]; + tensor var_12087_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1161_cast_fp16)[name = string("op_12087_cast_fp16")]; + tensor var_12088_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1163_cast_fp16)[name = string("op_12088_cast_fp16")]; + tensor var_12089_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1165_cast_fp16)[name = string("op_12089_cast_fp16")]; + tensor var_12090_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1167_cast_fp16)[name = string("op_12090_cast_fp16")]; + tensor var_12091_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1169_cast_fp16)[name = string("op_12091_cast_fp16")]; + tensor var_12092_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1171_cast_fp16)[name = string("op_12092_cast_fp16")]; + tensor var_12093_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1173_cast_fp16)[name = string("op_12093_cast_fp16")]; + tensor var_12094_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1175_cast_fp16)[name = string("op_12094_cast_fp16")]; + tensor var_12095_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1177_cast_fp16)[name = string("op_12095_cast_fp16")]; + tensor var_12096_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1179_cast_fp16)[name = string("op_12096_cast_fp16")]; + tensor var_12097_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1181_cast_fp16)[name = string("op_12097_cast_fp16")]; + tensor var_12098_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1183_cast_fp16)[name = string("op_12098_cast_fp16")]; + tensor var_12099_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1185_cast_fp16)[name = string("op_12099_cast_fp16")]; + tensor var_12100_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1187_cast_fp16)[name = string("op_12100_cast_fp16")]; + tensor var_12101_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1189_cast_fp16)[name = string("op_12101_cast_fp16")]; + tensor var_12102_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1191_cast_fp16)[name = string("op_12102_cast_fp16")]; + tensor var_12103_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1193_cast_fp16)[name = string("op_12103_cast_fp16")]; + tensor var_12104_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1195_cast_fp16)[name = string("op_12104_cast_fp16")]; + tensor var_12105_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1197_cast_fp16)[name = string("op_12105_cast_fp16")]; + tensor var_12106_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1199_cast_fp16)[name = string("op_12106_cast_fp16")]; + tensor var_12107_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1201_cast_fp16)[name = string("op_12107_cast_fp16")]; + tensor var_12108_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1203_cast_fp16)[name = string("op_12108_cast_fp16")]; + tensor var_12109_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1205_cast_fp16)[name = string("op_12109_cast_fp16")]; + tensor var_12110_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1207_cast_fp16)[name = string("op_12110_cast_fp16")]; + tensor var_12111_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1209_cast_fp16)[name = string("op_12111_cast_fp16")]; + tensor var_12112_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1211_cast_fp16)[name = string("op_12112_cast_fp16")]; + tensor var_12113_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1213_cast_fp16)[name = string("op_12113_cast_fp16")]; + tensor var_12114_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1215_cast_fp16)[name = string("op_12114_cast_fp16")]; + tensor var_12115_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1217_cast_fp16)[name = string("op_12115_cast_fp16")]; + tensor var_12116_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1219_cast_fp16)[name = string("op_12116_cast_fp16")]; + tensor var_12117_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1221_cast_fp16)[name = string("op_12117_cast_fp16")]; + tensor var_12118_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1223_cast_fp16)[name = string("op_12118_cast_fp16")]; + tensor var_12119_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1225_cast_fp16)[name = string("op_12119_cast_fp16")]; + tensor var_12120_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1227_cast_fp16)[name = string("op_12120_cast_fp16")]; + tensor var_12121_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1229_cast_fp16)[name = string("op_12121_cast_fp16")]; + tensor var_12122_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1231_cast_fp16)[name = string("op_12122_cast_fp16")]; + tensor var_12123_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1233_cast_fp16)[name = string("op_12123_cast_fp16")]; + tensor var_12124_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1235_cast_fp16)[name = string("op_12124_cast_fp16")]; + tensor var_12125_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1237_cast_fp16)[name = string("op_12125_cast_fp16")]; + tensor var_12126_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1239_cast_fp16)[name = string("op_12126_cast_fp16")]; + tensor var_12127_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1241_cast_fp16)[name = string("op_12127_cast_fp16")]; + tensor var_12128_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1243_cast_fp16)[name = string("op_12128_cast_fp16")]; + tensor var_12129_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1245_cast_fp16)[name = string("op_12129_cast_fp16")]; + tensor var_12130_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1247_cast_fp16)[name = string("op_12130_cast_fp16")]; + tensor var_12131_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1249_cast_fp16)[name = string("op_12131_cast_fp16")]; + tensor var_12132_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1251_cast_fp16)[name = string("op_12132_cast_fp16")]; + tensor var_12133_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1253_cast_fp16)[name = string("op_12133_cast_fp16")]; + tensor var_12134_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1255_cast_fp16)[name = string("op_12134_cast_fp16")]; + tensor var_12135_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1257_cast_fp16)[name = string("op_12135_cast_fp16")]; + tensor var_12136_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1259_cast_fp16)[name = string("op_12136_cast_fp16")]; + tensor var_12137_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1261_cast_fp16)[name = string("op_12137_cast_fp16")]; + tensor var_12138_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1263_cast_fp16)[name = string("op_12138_cast_fp16")]; + tensor var_12139_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1265_cast_fp16)[name = string("op_12139_cast_fp16")]; + tensor var_12140_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1267_cast_fp16)[name = string("op_12140_cast_fp16")]; + tensor var_12141_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1269_cast_fp16)[name = string("op_12141_cast_fp16")]; + tensor var_12142_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1271_cast_fp16)[name = string("op_12142_cast_fp16")]; + tensor var_12143_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1273_cast_fp16)[name = string("op_12143_cast_fp16")]; + tensor var_12144_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1275_cast_fp16)[name = string("op_12144_cast_fp16")]; + tensor var_12145_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1277_cast_fp16)[name = string("op_12145_cast_fp16")]; + tensor var_12146_cast_fp16 = softmax(axis = var_10892, x = aw_chunk_1279_cast_fp16)[name = string("op_12146_cast_fp16")]; + string var_12148_equation_0 = const()[name = string("op_12148_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12148_cast_fp16 = einsum(equation = var_12148_equation_0, values = (var_11668_cast_fp16, var_12067_cast_fp16))[name = string("op_12148_cast_fp16")]; + string var_12150_equation_0 = const()[name = string("op_12150_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12150_cast_fp16 = einsum(equation = var_12150_equation_0, values = (var_11668_cast_fp16, var_12068_cast_fp16))[name = string("op_12150_cast_fp16")]; + string var_12152_equation_0 = const()[name = string("op_12152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12152_cast_fp16 = einsum(equation = var_12152_equation_0, values = (var_11668_cast_fp16, var_12069_cast_fp16))[name = string("op_12152_cast_fp16")]; + string var_12154_equation_0 = const()[name = string("op_12154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12154_cast_fp16 = einsum(equation = var_12154_equation_0, values = (var_11668_cast_fp16, var_12070_cast_fp16))[name = string("op_12154_cast_fp16")]; + string var_12156_equation_0 = const()[name = string("op_12156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12156_cast_fp16 = einsum(equation = var_12156_equation_0, values = (var_11672_cast_fp16, var_12071_cast_fp16))[name = string("op_12156_cast_fp16")]; + string var_12158_equation_0 = const()[name = string("op_12158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12158_cast_fp16 = einsum(equation = var_12158_equation_0, values = (var_11672_cast_fp16, var_12072_cast_fp16))[name = string("op_12158_cast_fp16")]; + string var_12160_equation_0 = const()[name = string("op_12160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12160_cast_fp16 = einsum(equation = var_12160_equation_0, values = (var_11672_cast_fp16, var_12073_cast_fp16))[name = string("op_12160_cast_fp16")]; + string var_12162_equation_0 = const()[name = string("op_12162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12162_cast_fp16 = einsum(equation = var_12162_equation_0, values = (var_11672_cast_fp16, var_12074_cast_fp16))[name = string("op_12162_cast_fp16")]; + string var_12164_equation_0 = const()[name = string("op_12164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12164_cast_fp16 = einsum(equation = var_12164_equation_0, values = (var_11676_cast_fp16, var_12075_cast_fp16))[name = string("op_12164_cast_fp16")]; + string var_12166_equation_0 = const()[name = string("op_12166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12166_cast_fp16 = einsum(equation = var_12166_equation_0, values = (var_11676_cast_fp16, var_12076_cast_fp16))[name = string("op_12166_cast_fp16")]; + string var_12168_equation_0 = const()[name = string("op_12168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12168_cast_fp16 = einsum(equation = var_12168_equation_0, values = (var_11676_cast_fp16, var_12077_cast_fp16))[name = string("op_12168_cast_fp16")]; + string var_12170_equation_0 = const()[name = string("op_12170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12170_cast_fp16 = einsum(equation = var_12170_equation_0, values = (var_11676_cast_fp16, var_12078_cast_fp16))[name = string("op_12170_cast_fp16")]; + string var_12172_equation_0 = const()[name = string("op_12172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12172_cast_fp16 = einsum(equation = var_12172_equation_0, values = (var_11680_cast_fp16, var_12079_cast_fp16))[name = string("op_12172_cast_fp16")]; + string var_12174_equation_0 = const()[name = string("op_12174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12174_cast_fp16 = einsum(equation = var_12174_equation_0, values = (var_11680_cast_fp16, var_12080_cast_fp16))[name = string("op_12174_cast_fp16")]; + string var_12176_equation_0 = const()[name = string("op_12176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12176_cast_fp16 = einsum(equation = var_12176_equation_0, values = (var_11680_cast_fp16, var_12081_cast_fp16))[name = string("op_12176_cast_fp16")]; + string var_12178_equation_0 = const()[name = string("op_12178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12178_cast_fp16 = einsum(equation = var_12178_equation_0, values = (var_11680_cast_fp16, var_12082_cast_fp16))[name = string("op_12178_cast_fp16")]; + string var_12180_equation_0 = const()[name = string("op_12180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12180_cast_fp16 = einsum(equation = var_12180_equation_0, values = (var_11684_cast_fp16, var_12083_cast_fp16))[name = string("op_12180_cast_fp16")]; + string var_12182_equation_0 = const()[name = string("op_12182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12182_cast_fp16 = einsum(equation = var_12182_equation_0, values = (var_11684_cast_fp16, var_12084_cast_fp16))[name = string("op_12182_cast_fp16")]; + string var_12184_equation_0 = const()[name = string("op_12184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12184_cast_fp16 = einsum(equation = var_12184_equation_0, values = (var_11684_cast_fp16, var_12085_cast_fp16))[name = string("op_12184_cast_fp16")]; + string var_12186_equation_0 = const()[name = string("op_12186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12186_cast_fp16 = einsum(equation = var_12186_equation_0, values = (var_11684_cast_fp16, var_12086_cast_fp16))[name = string("op_12186_cast_fp16")]; + string var_12188_equation_0 = const()[name = string("op_12188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12188_cast_fp16 = einsum(equation = var_12188_equation_0, values = (var_11688_cast_fp16, var_12087_cast_fp16))[name = string("op_12188_cast_fp16")]; + string var_12190_equation_0 = const()[name = string("op_12190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12190_cast_fp16 = einsum(equation = var_12190_equation_0, values = (var_11688_cast_fp16, var_12088_cast_fp16))[name = string("op_12190_cast_fp16")]; + string var_12192_equation_0 = const()[name = string("op_12192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12192_cast_fp16 = einsum(equation = var_12192_equation_0, values = (var_11688_cast_fp16, var_12089_cast_fp16))[name = string("op_12192_cast_fp16")]; + string var_12194_equation_0 = const()[name = string("op_12194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12194_cast_fp16 = einsum(equation = var_12194_equation_0, values = (var_11688_cast_fp16, var_12090_cast_fp16))[name = string("op_12194_cast_fp16")]; + string var_12196_equation_0 = const()[name = string("op_12196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12196_cast_fp16 = einsum(equation = var_12196_equation_0, values = (var_11692_cast_fp16, var_12091_cast_fp16))[name = string("op_12196_cast_fp16")]; + string var_12198_equation_0 = const()[name = string("op_12198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12198_cast_fp16 = einsum(equation = var_12198_equation_0, values = (var_11692_cast_fp16, var_12092_cast_fp16))[name = string("op_12198_cast_fp16")]; + string var_12200_equation_0 = const()[name = string("op_12200_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12200_cast_fp16 = einsum(equation = var_12200_equation_0, values = (var_11692_cast_fp16, var_12093_cast_fp16))[name = string("op_12200_cast_fp16")]; + string var_12202_equation_0 = const()[name = string("op_12202_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12202_cast_fp16 = einsum(equation = var_12202_equation_0, values = (var_11692_cast_fp16, var_12094_cast_fp16))[name = string("op_12202_cast_fp16")]; + string var_12204_equation_0 = const()[name = string("op_12204_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12204_cast_fp16 = einsum(equation = var_12204_equation_0, values = (var_11696_cast_fp16, var_12095_cast_fp16))[name = string("op_12204_cast_fp16")]; + string var_12206_equation_0 = const()[name = string("op_12206_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12206_cast_fp16 = einsum(equation = var_12206_equation_0, values = (var_11696_cast_fp16, var_12096_cast_fp16))[name = string("op_12206_cast_fp16")]; + string var_12208_equation_0 = const()[name = string("op_12208_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12208_cast_fp16 = einsum(equation = var_12208_equation_0, values = (var_11696_cast_fp16, var_12097_cast_fp16))[name = string("op_12208_cast_fp16")]; + string var_12210_equation_0 = const()[name = string("op_12210_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12210_cast_fp16 = einsum(equation = var_12210_equation_0, values = (var_11696_cast_fp16, var_12098_cast_fp16))[name = string("op_12210_cast_fp16")]; + string var_12212_equation_0 = const()[name = string("op_12212_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12212_cast_fp16 = einsum(equation = var_12212_equation_0, values = (var_11700_cast_fp16, var_12099_cast_fp16))[name = string("op_12212_cast_fp16")]; + string var_12214_equation_0 = const()[name = string("op_12214_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12214_cast_fp16 = einsum(equation = var_12214_equation_0, values = (var_11700_cast_fp16, var_12100_cast_fp16))[name = string("op_12214_cast_fp16")]; + string var_12216_equation_0 = const()[name = string("op_12216_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12216_cast_fp16 = einsum(equation = var_12216_equation_0, values = (var_11700_cast_fp16, var_12101_cast_fp16))[name = string("op_12216_cast_fp16")]; + string var_12218_equation_0 = const()[name = string("op_12218_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12218_cast_fp16 = einsum(equation = var_12218_equation_0, values = (var_11700_cast_fp16, var_12102_cast_fp16))[name = string("op_12218_cast_fp16")]; + string var_12220_equation_0 = const()[name = string("op_12220_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12220_cast_fp16 = einsum(equation = var_12220_equation_0, values = (var_11704_cast_fp16, var_12103_cast_fp16))[name = string("op_12220_cast_fp16")]; + string var_12222_equation_0 = const()[name = string("op_12222_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12222_cast_fp16 = einsum(equation = var_12222_equation_0, values = (var_11704_cast_fp16, var_12104_cast_fp16))[name = string("op_12222_cast_fp16")]; + string var_12224_equation_0 = const()[name = string("op_12224_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12224_cast_fp16 = einsum(equation = var_12224_equation_0, values = (var_11704_cast_fp16, var_12105_cast_fp16))[name = string("op_12224_cast_fp16")]; + string var_12226_equation_0 = const()[name = string("op_12226_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12226_cast_fp16 = einsum(equation = var_12226_equation_0, values = (var_11704_cast_fp16, var_12106_cast_fp16))[name = string("op_12226_cast_fp16")]; + string var_12228_equation_0 = const()[name = string("op_12228_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12228_cast_fp16 = einsum(equation = var_12228_equation_0, values = (var_11708_cast_fp16, var_12107_cast_fp16))[name = string("op_12228_cast_fp16")]; + string var_12230_equation_0 = const()[name = string("op_12230_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12230_cast_fp16 = einsum(equation = var_12230_equation_0, values = (var_11708_cast_fp16, var_12108_cast_fp16))[name = string("op_12230_cast_fp16")]; + string var_12232_equation_0 = const()[name = string("op_12232_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12232_cast_fp16 = einsum(equation = var_12232_equation_0, values = (var_11708_cast_fp16, var_12109_cast_fp16))[name = string("op_12232_cast_fp16")]; + string var_12234_equation_0 = const()[name = string("op_12234_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12234_cast_fp16 = einsum(equation = var_12234_equation_0, values = (var_11708_cast_fp16, var_12110_cast_fp16))[name = string("op_12234_cast_fp16")]; + string var_12236_equation_0 = const()[name = string("op_12236_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12236_cast_fp16 = einsum(equation = var_12236_equation_0, values = (var_11712_cast_fp16, var_12111_cast_fp16))[name = string("op_12236_cast_fp16")]; + string var_12238_equation_0 = const()[name = string("op_12238_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12238_cast_fp16 = einsum(equation = var_12238_equation_0, values = (var_11712_cast_fp16, var_12112_cast_fp16))[name = string("op_12238_cast_fp16")]; + string var_12240_equation_0 = const()[name = string("op_12240_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12240_cast_fp16 = einsum(equation = var_12240_equation_0, values = (var_11712_cast_fp16, var_12113_cast_fp16))[name = string("op_12240_cast_fp16")]; + string var_12242_equation_0 = const()[name = string("op_12242_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12242_cast_fp16 = einsum(equation = var_12242_equation_0, values = (var_11712_cast_fp16, var_12114_cast_fp16))[name = string("op_12242_cast_fp16")]; + string var_12244_equation_0 = const()[name = string("op_12244_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12244_cast_fp16 = einsum(equation = var_12244_equation_0, values = (var_11716_cast_fp16, var_12115_cast_fp16))[name = string("op_12244_cast_fp16")]; + string var_12246_equation_0 = const()[name = string("op_12246_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12246_cast_fp16 = einsum(equation = var_12246_equation_0, values = (var_11716_cast_fp16, var_12116_cast_fp16))[name = string("op_12246_cast_fp16")]; + string var_12248_equation_0 = const()[name = string("op_12248_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12248_cast_fp16 = einsum(equation = var_12248_equation_0, values = (var_11716_cast_fp16, var_12117_cast_fp16))[name = string("op_12248_cast_fp16")]; + string var_12250_equation_0 = const()[name = string("op_12250_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12250_cast_fp16 = einsum(equation = var_12250_equation_0, values = (var_11716_cast_fp16, var_12118_cast_fp16))[name = string("op_12250_cast_fp16")]; + string var_12252_equation_0 = const()[name = string("op_12252_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12252_cast_fp16 = einsum(equation = var_12252_equation_0, values = (var_11720_cast_fp16, var_12119_cast_fp16))[name = string("op_12252_cast_fp16")]; + string var_12254_equation_0 = const()[name = string("op_12254_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12254_cast_fp16 = einsum(equation = var_12254_equation_0, values = (var_11720_cast_fp16, var_12120_cast_fp16))[name = string("op_12254_cast_fp16")]; + string var_12256_equation_0 = const()[name = string("op_12256_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12256_cast_fp16 = einsum(equation = var_12256_equation_0, values = (var_11720_cast_fp16, var_12121_cast_fp16))[name = string("op_12256_cast_fp16")]; + string var_12258_equation_0 = const()[name = string("op_12258_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12258_cast_fp16 = einsum(equation = var_12258_equation_0, values = (var_11720_cast_fp16, var_12122_cast_fp16))[name = string("op_12258_cast_fp16")]; + string var_12260_equation_0 = const()[name = string("op_12260_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12260_cast_fp16 = einsum(equation = var_12260_equation_0, values = (var_11724_cast_fp16, var_12123_cast_fp16))[name = string("op_12260_cast_fp16")]; + string var_12262_equation_0 = const()[name = string("op_12262_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12262_cast_fp16 = einsum(equation = var_12262_equation_0, values = (var_11724_cast_fp16, var_12124_cast_fp16))[name = string("op_12262_cast_fp16")]; + string var_12264_equation_0 = const()[name = string("op_12264_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12264_cast_fp16 = einsum(equation = var_12264_equation_0, values = (var_11724_cast_fp16, var_12125_cast_fp16))[name = string("op_12264_cast_fp16")]; + string var_12266_equation_0 = const()[name = string("op_12266_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12266_cast_fp16 = einsum(equation = var_12266_equation_0, values = (var_11724_cast_fp16, var_12126_cast_fp16))[name = string("op_12266_cast_fp16")]; + string var_12268_equation_0 = const()[name = string("op_12268_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12268_cast_fp16 = einsum(equation = var_12268_equation_0, values = (var_11728_cast_fp16, var_12127_cast_fp16))[name = string("op_12268_cast_fp16")]; + string var_12270_equation_0 = const()[name = string("op_12270_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12270_cast_fp16 = einsum(equation = var_12270_equation_0, values = (var_11728_cast_fp16, var_12128_cast_fp16))[name = string("op_12270_cast_fp16")]; + string var_12272_equation_0 = const()[name = string("op_12272_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12272_cast_fp16 = einsum(equation = var_12272_equation_0, values = (var_11728_cast_fp16, var_12129_cast_fp16))[name = string("op_12272_cast_fp16")]; + string var_12274_equation_0 = const()[name = string("op_12274_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12274_cast_fp16 = einsum(equation = var_12274_equation_0, values = (var_11728_cast_fp16, var_12130_cast_fp16))[name = string("op_12274_cast_fp16")]; + string var_12276_equation_0 = const()[name = string("op_12276_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12276_cast_fp16 = einsum(equation = var_12276_equation_0, values = (var_11732_cast_fp16, var_12131_cast_fp16))[name = string("op_12276_cast_fp16")]; + string var_12278_equation_0 = const()[name = string("op_12278_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12278_cast_fp16 = einsum(equation = var_12278_equation_0, values = (var_11732_cast_fp16, var_12132_cast_fp16))[name = string("op_12278_cast_fp16")]; + string var_12280_equation_0 = const()[name = string("op_12280_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12280_cast_fp16 = einsum(equation = var_12280_equation_0, values = (var_11732_cast_fp16, var_12133_cast_fp16))[name = string("op_12280_cast_fp16")]; + string var_12282_equation_0 = const()[name = string("op_12282_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12282_cast_fp16 = einsum(equation = var_12282_equation_0, values = (var_11732_cast_fp16, var_12134_cast_fp16))[name = string("op_12282_cast_fp16")]; + string var_12284_equation_0 = const()[name = string("op_12284_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12284_cast_fp16 = einsum(equation = var_12284_equation_0, values = (var_11736_cast_fp16, var_12135_cast_fp16))[name = string("op_12284_cast_fp16")]; + string var_12286_equation_0 = const()[name = string("op_12286_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12286_cast_fp16 = einsum(equation = var_12286_equation_0, values = (var_11736_cast_fp16, var_12136_cast_fp16))[name = string("op_12286_cast_fp16")]; + string var_12288_equation_0 = const()[name = string("op_12288_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12288_cast_fp16 = einsum(equation = var_12288_equation_0, values = (var_11736_cast_fp16, var_12137_cast_fp16))[name = string("op_12288_cast_fp16")]; + string var_12290_equation_0 = const()[name = string("op_12290_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12290_cast_fp16 = einsum(equation = var_12290_equation_0, values = (var_11736_cast_fp16, var_12138_cast_fp16))[name = string("op_12290_cast_fp16")]; + string var_12292_equation_0 = const()[name = string("op_12292_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12292_cast_fp16 = einsum(equation = var_12292_equation_0, values = (var_11740_cast_fp16, var_12139_cast_fp16))[name = string("op_12292_cast_fp16")]; + string var_12294_equation_0 = const()[name = string("op_12294_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12294_cast_fp16 = einsum(equation = var_12294_equation_0, values = (var_11740_cast_fp16, var_12140_cast_fp16))[name = string("op_12294_cast_fp16")]; + string var_12296_equation_0 = const()[name = string("op_12296_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12296_cast_fp16 = einsum(equation = var_12296_equation_0, values = (var_11740_cast_fp16, var_12141_cast_fp16))[name = string("op_12296_cast_fp16")]; + string var_12298_equation_0 = const()[name = string("op_12298_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12298_cast_fp16 = einsum(equation = var_12298_equation_0, values = (var_11740_cast_fp16, var_12142_cast_fp16))[name = string("op_12298_cast_fp16")]; + string var_12300_equation_0 = const()[name = string("op_12300_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12300_cast_fp16 = einsum(equation = var_12300_equation_0, values = (var_11744_cast_fp16, var_12143_cast_fp16))[name = string("op_12300_cast_fp16")]; + string var_12302_equation_0 = const()[name = string("op_12302_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12302_cast_fp16 = einsum(equation = var_12302_equation_0, values = (var_11744_cast_fp16, var_12144_cast_fp16))[name = string("op_12302_cast_fp16")]; + string var_12304_equation_0 = const()[name = string("op_12304_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12304_cast_fp16 = einsum(equation = var_12304_equation_0, values = (var_11744_cast_fp16, var_12145_cast_fp16))[name = string("op_12304_cast_fp16")]; + string var_12306_equation_0 = const()[name = string("op_12306_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12306_cast_fp16 = einsum(equation = var_12306_equation_0, values = (var_11744_cast_fp16, var_12146_cast_fp16))[name = string("op_12306_cast_fp16")]; + bool var_12308_interleave_0 = const()[name = string("op_12308_interleave_0"), val = bool(false)]; + tensor var_12308_cast_fp16 = concat(axis = var_10867, interleave = var_12308_interleave_0, values = (var_12148_cast_fp16, var_12150_cast_fp16, var_12152_cast_fp16, var_12154_cast_fp16))[name = string("op_12308_cast_fp16")]; + bool var_12310_interleave_0 = const()[name = string("op_12310_interleave_0"), val = bool(false)]; + tensor var_12310_cast_fp16 = concat(axis = var_10867, interleave = var_12310_interleave_0, values = (var_12156_cast_fp16, var_12158_cast_fp16, var_12160_cast_fp16, var_12162_cast_fp16))[name = string("op_12310_cast_fp16")]; + bool var_12312_interleave_0 = const()[name = string("op_12312_interleave_0"), val = bool(false)]; + tensor var_12312_cast_fp16 = concat(axis = var_10867, interleave = var_12312_interleave_0, values = (var_12164_cast_fp16, var_12166_cast_fp16, var_12168_cast_fp16, var_12170_cast_fp16))[name = string("op_12312_cast_fp16")]; + bool var_12314_interleave_0 = const()[name = string("op_12314_interleave_0"), val = bool(false)]; + tensor var_12314_cast_fp16 = concat(axis = var_10867, interleave = var_12314_interleave_0, values = (var_12172_cast_fp16, var_12174_cast_fp16, var_12176_cast_fp16, var_12178_cast_fp16))[name = string("op_12314_cast_fp16")]; + bool var_12316_interleave_0 = const()[name = string("op_12316_interleave_0"), val = bool(false)]; + tensor var_12316_cast_fp16 = concat(axis = var_10867, interleave = var_12316_interleave_0, values = (var_12180_cast_fp16, var_12182_cast_fp16, var_12184_cast_fp16, var_12186_cast_fp16))[name = string("op_12316_cast_fp16")]; + bool var_12318_interleave_0 = const()[name = string("op_12318_interleave_0"), val = bool(false)]; + tensor var_12318_cast_fp16 = concat(axis = var_10867, interleave = var_12318_interleave_0, values = (var_12188_cast_fp16, var_12190_cast_fp16, var_12192_cast_fp16, var_12194_cast_fp16))[name = string("op_12318_cast_fp16")]; + bool var_12320_interleave_0 = const()[name = string("op_12320_interleave_0"), val = bool(false)]; + tensor var_12320_cast_fp16 = concat(axis = var_10867, interleave = var_12320_interleave_0, values = (var_12196_cast_fp16, var_12198_cast_fp16, var_12200_cast_fp16, var_12202_cast_fp16))[name = string("op_12320_cast_fp16")]; + bool var_12322_interleave_0 = const()[name = string("op_12322_interleave_0"), val = bool(false)]; + tensor var_12322_cast_fp16 = concat(axis = var_10867, interleave = var_12322_interleave_0, values = (var_12204_cast_fp16, var_12206_cast_fp16, var_12208_cast_fp16, var_12210_cast_fp16))[name = string("op_12322_cast_fp16")]; + bool var_12324_interleave_0 = const()[name = string("op_12324_interleave_0"), val = bool(false)]; + tensor var_12324_cast_fp16 = concat(axis = var_10867, interleave = var_12324_interleave_0, values = (var_12212_cast_fp16, var_12214_cast_fp16, var_12216_cast_fp16, var_12218_cast_fp16))[name = string("op_12324_cast_fp16")]; + bool var_12326_interleave_0 = const()[name = string("op_12326_interleave_0"), val = bool(false)]; + tensor var_12326_cast_fp16 = concat(axis = var_10867, interleave = var_12326_interleave_0, values = (var_12220_cast_fp16, var_12222_cast_fp16, var_12224_cast_fp16, var_12226_cast_fp16))[name = string("op_12326_cast_fp16")]; + bool var_12328_interleave_0 = const()[name = string("op_12328_interleave_0"), val = bool(false)]; + tensor var_12328_cast_fp16 = concat(axis = var_10867, interleave = var_12328_interleave_0, values = (var_12228_cast_fp16, var_12230_cast_fp16, var_12232_cast_fp16, var_12234_cast_fp16))[name = string("op_12328_cast_fp16")]; + bool var_12330_interleave_0 = const()[name = string("op_12330_interleave_0"), val = bool(false)]; + tensor var_12330_cast_fp16 = concat(axis = var_10867, interleave = var_12330_interleave_0, values = (var_12236_cast_fp16, var_12238_cast_fp16, var_12240_cast_fp16, var_12242_cast_fp16))[name = string("op_12330_cast_fp16")]; + bool var_12332_interleave_0 = const()[name = string("op_12332_interleave_0"), val = bool(false)]; + tensor var_12332_cast_fp16 = concat(axis = var_10867, interleave = var_12332_interleave_0, values = (var_12244_cast_fp16, var_12246_cast_fp16, var_12248_cast_fp16, var_12250_cast_fp16))[name = string("op_12332_cast_fp16")]; + bool var_12334_interleave_0 = const()[name = string("op_12334_interleave_0"), val = bool(false)]; + tensor var_12334_cast_fp16 = concat(axis = var_10867, interleave = var_12334_interleave_0, values = (var_12252_cast_fp16, var_12254_cast_fp16, var_12256_cast_fp16, var_12258_cast_fp16))[name = string("op_12334_cast_fp16")]; + bool var_12336_interleave_0 = const()[name = string("op_12336_interleave_0"), val = bool(false)]; + tensor var_12336_cast_fp16 = concat(axis = var_10867, interleave = var_12336_interleave_0, values = (var_12260_cast_fp16, var_12262_cast_fp16, var_12264_cast_fp16, var_12266_cast_fp16))[name = string("op_12336_cast_fp16")]; + bool var_12338_interleave_0 = const()[name = string("op_12338_interleave_0"), val = bool(false)]; + tensor var_12338_cast_fp16 = concat(axis = var_10867, interleave = var_12338_interleave_0, values = (var_12268_cast_fp16, var_12270_cast_fp16, var_12272_cast_fp16, var_12274_cast_fp16))[name = string("op_12338_cast_fp16")]; + bool var_12340_interleave_0 = const()[name = string("op_12340_interleave_0"), val = bool(false)]; + tensor var_12340_cast_fp16 = concat(axis = var_10867, interleave = var_12340_interleave_0, values = (var_12276_cast_fp16, var_12278_cast_fp16, var_12280_cast_fp16, var_12282_cast_fp16))[name = string("op_12340_cast_fp16")]; + bool var_12342_interleave_0 = const()[name = string("op_12342_interleave_0"), val = bool(false)]; + tensor var_12342_cast_fp16 = concat(axis = var_10867, interleave = var_12342_interleave_0, values = (var_12284_cast_fp16, var_12286_cast_fp16, var_12288_cast_fp16, var_12290_cast_fp16))[name = string("op_12342_cast_fp16")]; + bool var_12344_interleave_0 = const()[name = string("op_12344_interleave_0"), val = bool(false)]; + tensor var_12344_cast_fp16 = concat(axis = var_10867, interleave = var_12344_interleave_0, values = (var_12292_cast_fp16, var_12294_cast_fp16, var_12296_cast_fp16, var_12298_cast_fp16))[name = string("op_12344_cast_fp16")]; + bool var_12346_interleave_0 = const()[name = string("op_12346_interleave_0"), val = bool(false)]; + tensor var_12346_cast_fp16 = concat(axis = var_10867, interleave = var_12346_interleave_0, values = (var_12300_cast_fp16, var_12302_cast_fp16, var_12304_cast_fp16, var_12306_cast_fp16))[name = string("op_12346_cast_fp16")]; + bool input_57_interleave_0 = const()[name = string("input_57_interleave_0"), val = bool(false)]; + tensor input_57_cast_fp16 = concat(axis = var_10892, interleave = input_57_interleave_0, values = (var_12308_cast_fp16, var_12310_cast_fp16, var_12312_cast_fp16, var_12314_cast_fp16, var_12316_cast_fp16, var_12318_cast_fp16, var_12320_cast_fp16, var_12322_cast_fp16, var_12324_cast_fp16, var_12326_cast_fp16, var_12328_cast_fp16, var_12330_cast_fp16, var_12332_cast_fp16, var_12334_cast_fp16, var_12336_cast_fp16, var_12338_cast_fp16, var_12340_cast_fp16, var_12342_cast_fp16, var_12344_cast_fp16, var_12346_cast_fp16))[name = string("input_57_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299978240)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303255104)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_12365_to_fp16 = const()[name = string("op_12365_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_12365_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303257728)))]; + tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303260352)))]; + fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303262976)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316370240)))]; + tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; + tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316380544)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329487808)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; + int32 var_12394 = const()[name = string("op_12394"), val = int32(3)]; + int32 var_12419 = const()[name = string("op_12419"), val = int32(1)]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_12436_to_fp16 = const()[name = string("op_12436_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_12436_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329490432)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329493056)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329495680)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332772544)))]; + tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; + string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; + tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; + tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; + int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332775168)))]; + tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; + string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; + tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; + tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; + int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336052032)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339328896)))]; + tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_12474_begin_0 = const()[name = string("op_12474_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12474_end_0 = const()[name = string("op_12474_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12474_end_mask_0 = const()[name = string("op_12474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12474_cast_fp16 = slice_by_index(begin = var_12474_begin_0, end = var_12474_end_0, end_mask = var_12474_end_mask_0, x = query_17_cast_fp16)[name = string("op_12474_cast_fp16")]; + tensor var_12478_begin_0 = const()[name = string("op_12478_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12478_end_0 = const()[name = string("op_12478_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_12478_end_mask_0 = const()[name = string("op_12478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12478_cast_fp16 = slice_by_index(begin = var_12478_begin_0, end = var_12478_end_0, end_mask = var_12478_end_mask_0, x = query_17_cast_fp16)[name = string("op_12478_cast_fp16")]; + tensor var_12482_begin_0 = const()[name = string("op_12482_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12482_end_0 = const()[name = string("op_12482_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_12482_end_mask_0 = const()[name = string("op_12482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12482_cast_fp16 = slice_by_index(begin = var_12482_begin_0, end = var_12482_end_0, end_mask = var_12482_end_mask_0, x = query_17_cast_fp16)[name = string("op_12482_cast_fp16")]; + tensor var_12486_begin_0 = const()[name = string("op_12486_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12486_end_0 = const()[name = string("op_12486_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_12486_end_mask_0 = const()[name = string("op_12486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12486_cast_fp16 = slice_by_index(begin = var_12486_begin_0, end = var_12486_end_0, end_mask = var_12486_end_mask_0, x = query_17_cast_fp16)[name = string("op_12486_cast_fp16")]; + tensor var_12490_begin_0 = const()[name = string("op_12490_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12490_end_0 = const()[name = string("op_12490_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_12490_end_mask_0 = const()[name = string("op_12490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12490_cast_fp16 = slice_by_index(begin = var_12490_begin_0, end = var_12490_end_0, end_mask = var_12490_end_mask_0, x = query_17_cast_fp16)[name = string("op_12490_cast_fp16")]; + tensor var_12494_begin_0 = const()[name = string("op_12494_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12494_end_0 = const()[name = string("op_12494_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_12494_end_mask_0 = const()[name = string("op_12494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12494_cast_fp16 = slice_by_index(begin = var_12494_begin_0, end = var_12494_end_0, end_mask = var_12494_end_mask_0, x = query_17_cast_fp16)[name = string("op_12494_cast_fp16")]; + tensor var_12498_begin_0 = const()[name = string("op_12498_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12498_end_0 = const()[name = string("op_12498_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_12498_end_mask_0 = const()[name = string("op_12498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12498_cast_fp16 = slice_by_index(begin = var_12498_begin_0, end = var_12498_end_0, end_mask = var_12498_end_mask_0, x = query_17_cast_fp16)[name = string("op_12498_cast_fp16")]; + tensor var_12502_begin_0 = const()[name = string("op_12502_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12502_end_0 = const()[name = string("op_12502_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_12502_end_mask_0 = const()[name = string("op_12502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12502_cast_fp16 = slice_by_index(begin = var_12502_begin_0, end = var_12502_end_0, end_mask = var_12502_end_mask_0, x = query_17_cast_fp16)[name = string("op_12502_cast_fp16")]; + tensor var_12506_begin_0 = const()[name = string("op_12506_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12506_end_0 = const()[name = string("op_12506_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_12506_end_mask_0 = const()[name = string("op_12506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12506_cast_fp16 = slice_by_index(begin = var_12506_begin_0, end = var_12506_end_0, end_mask = var_12506_end_mask_0, x = query_17_cast_fp16)[name = string("op_12506_cast_fp16")]; + tensor var_12510_begin_0 = const()[name = string("op_12510_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12510_end_0 = const()[name = string("op_12510_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_12510_end_mask_0 = const()[name = string("op_12510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12510_cast_fp16 = slice_by_index(begin = var_12510_begin_0, end = var_12510_end_0, end_mask = var_12510_end_mask_0, x = query_17_cast_fp16)[name = string("op_12510_cast_fp16")]; + tensor var_12514_begin_0 = const()[name = string("op_12514_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12514_end_0 = const()[name = string("op_12514_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_12514_end_mask_0 = const()[name = string("op_12514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12514_cast_fp16 = slice_by_index(begin = var_12514_begin_0, end = var_12514_end_0, end_mask = var_12514_end_mask_0, x = query_17_cast_fp16)[name = string("op_12514_cast_fp16")]; + tensor var_12518_begin_0 = const()[name = string("op_12518_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12518_end_0 = const()[name = string("op_12518_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_12518_end_mask_0 = const()[name = string("op_12518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12518_cast_fp16 = slice_by_index(begin = var_12518_begin_0, end = var_12518_end_0, end_mask = var_12518_end_mask_0, x = query_17_cast_fp16)[name = string("op_12518_cast_fp16")]; + tensor var_12522_begin_0 = const()[name = string("op_12522_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12522_end_0 = const()[name = string("op_12522_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_12522_end_mask_0 = const()[name = string("op_12522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12522_cast_fp16 = slice_by_index(begin = var_12522_begin_0, end = var_12522_end_0, end_mask = var_12522_end_mask_0, x = query_17_cast_fp16)[name = string("op_12522_cast_fp16")]; + tensor var_12526_begin_0 = const()[name = string("op_12526_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12526_end_0 = const()[name = string("op_12526_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_12526_end_mask_0 = const()[name = string("op_12526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12526_cast_fp16 = slice_by_index(begin = var_12526_begin_0, end = var_12526_end_0, end_mask = var_12526_end_mask_0, x = query_17_cast_fp16)[name = string("op_12526_cast_fp16")]; + tensor var_12530_begin_0 = const()[name = string("op_12530_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12530_end_0 = const()[name = string("op_12530_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_12530_end_mask_0 = const()[name = string("op_12530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12530_cast_fp16 = slice_by_index(begin = var_12530_begin_0, end = var_12530_end_0, end_mask = var_12530_end_mask_0, x = query_17_cast_fp16)[name = string("op_12530_cast_fp16")]; + tensor var_12534_begin_0 = const()[name = string("op_12534_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12534_end_0 = const()[name = string("op_12534_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_12534_end_mask_0 = const()[name = string("op_12534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12534_cast_fp16 = slice_by_index(begin = var_12534_begin_0, end = var_12534_end_0, end_mask = var_12534_end_mask_0, x = query_17_cast_fp16)[name = string("op_12534_cast_fp16")]; + tensor var_12538_begin_0 = const()[name = string("op_12538_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12538_end_0 = const()[name = string("op_12538_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_12538_end_mask_0 = const()[name = string("op_12538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12538_cast_fp16 = slice_by_index(begin = var_12538_begin_0, end = var_12538_end_0, end_mask = var_12538_end_mask_0, x = query_17_cast_fp16)[name = string("op_12538_cast_fp16")]; + tensor var_12542_begin_0 = const()[name = string("op_12542_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12542_end_0 = const()[name = string("op_12542_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_12542_end_mask_0 = const()[name = string("op_12542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12542_cast_fp16 = slice_by_index(begin = var_12542_begin_0, end = var_12542_end_0, end_mask = var_12542_end_mask_0, x = query_17_cast_fp16)[name = string("op_12542_cast_fp16")]; + tensor var_12546_begin_0 = const()[name = string("op_12546_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12546_end_0 = const()[name = string("op_12546_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_12546_end_mask_0 = const()[name = string("op_12546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12546_cast_fp16 = slice_by_index(begin = var_12546_begin_0, end = var_12546_end_0, end_mask = var_12546_end_mask_0, x = query_17_cast_fp16)[name = string("op_12546_cast_fp16")]; + tensor var_12550_begin_0 = const()[name = string("op_12550_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12550_end_0 = const()[name = string("op_12550_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_12550_end_mask_0 = const()[name = string("op_12550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12550_cast_fp16 = slice_by_index(begin = var_12550_begin_0, end = var_12550_end_0, end_mask = var_12550_end_mask_0, x = query_17_cast_fp16)[name = string("op_12550_cast_fp16")]; + tensor var_12559_begin_0 = const()[name = string("op_12559_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12559_end_0 = const()[name = string("op_12559_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12559_end_mask_0 = const()[name = string("op_12559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12559_cast_fp16 = slice_by_index(begin = var_12559_begin_0, end = var_12559_end_0, end_mask = var_12559_end_mask_0, x = var_12474_cast_fp16)[name = string("op_12559_cast_fp16")]; + tensor var_12566_begin_0 = const()[name = string("op_12566_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12566_end_0 = const()[name = string("op_12566_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12566_end_mask_0 = const()[name = string("op_12566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12566_cast_fp16 = slice_by_index(begin = var_12566_begin_0, end = var_12566_end_0, end_mask = var_12566_end_mask_0, x = var_12474_cast_fp16)[name = string("op_12566_cast_fp16")]; + tensor var_12573_begin_0 = const()[name = string("op_12573_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12573_end_0 = const()[name = string("op_12573_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12573_end_mask_0 = const()[name = string("op_12573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12573_cast_fp16 = slice_by_index(begin = var_12573_begin_0, end = var_12573_end_0, end_mask = var_12573_end_mask_0, x = var_12474_cast_fp16)[name = string("op_12573_cast_fp16")]; + tensor var_12580_begin_0 = const()[name = string("op_12580_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12580_end_0 = const()[name = string("op_12580_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12580_end_mask_0 = const()[name = string("op_12580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12580_cast_fp16 = slice_by_index(begin = var_12580_begin_0, end = var_12580_end_0, end_mask = var_12580_end_mask_0, x = var_12474_cast_fp16)[name = string("op_12580_cast_fp16")]; + tensor var_12587_begin_0 = const()[name = string("op_12587_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12587_end_0 = const()[name = string("op_12587_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12587_end_mask_0 = const()[name = string("op_12587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12587_cast_fp16 = slice_by_index(begin = var_12587_begin_0, end = var_12587_end_0, end_mask = var_12587_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12587_cast_fp16")]; + tensor var_12594_begin_0 = const()[name = string("op_12594_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12594_end_0 = const()[name = string("op_12594_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12594_end_mask_0 = const()[name = string("op_12594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12594_cast_fp16 = slice_by_index(begin = var_12594_begin_0, end = var_12594_end_0, end_mask = var_12594_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12594_cast_fp16")]; + tensor var_12601_begin_0 = const()[name = string("op_12601_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12601_end_0 = const()[name = string("op_12601_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12601_end_mask_0 = const()[name = string("op_12601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12601_cast_fp16 = slice_by_index(begin = var_12601_begin_0, end = var_12601_end_0, end_mask = var_12601_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12601_cast_fp16")]; + tensor var_12608_begin_0 = const()[name = string("op_12608_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12608_end_0 = const()[name = string("op_12608_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12608_end_mask_0 = const()[name = string("op_12608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12608_cast_fp16 = slice_by_index(begin = var_12608_begin_0, end = var_12608_end_0, end_mask = var_12608_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12608_cast_fp16")]; + tensor var_12615_begin_0 = const()[name = string("op_12615_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12615_end_0 = const()[name = string("op_12615_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12615_end_mask_0 = const()[name = string("op_12615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12615_cast_fp16 = slice_by_index(begin = var_12615_begin_0, end = var_12615_end_0, end_mask = var_12615_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12615_cast_fp16")]; + tensor var_12622_begin_0 = const()[name = string("op_12622_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12622_end_0 = const()[name = string("op_12622_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12622_end_mask_0 = const()[name = string("op_12622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12622_cast_fp16 = slice_by_index(begin = var_12622_begin_0, end = var_12622_end_0, end_mask = var_12622_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12622_cast_fp16")]; + tensor var_12629_begin_0 = const()[name = string("op_12629_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12629_end_0 = const()[name = string("op_12629_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12629_end_mask_0 = const()[name = string("op_12629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12629_cast_fp16 = slice_by_index(begin = var_12629_begin_0, end = var_12629_end_0, end_mask = var_12629_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12629_cast_fp16")]; + tensor var_12636_begin_0 = const()[name = string("op_12636_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12636_end_0 = const()[name = string("op_12636_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12636_end_mask_0 = const()[name = string("op_12636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12636_cast_fp16 = slice_by_index(begin = var_12636_begin_0, end = var_12636_end_0, end_mask = var_12636_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12636_cast_fp16")]; + tensor var_12643_begin_0 = const()[name = string("op_12643_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12643_end_0 = const()[name = string("op_12643_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12643_end_mask_0 = const()[name = string("op_12643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12643_cast_fp16 = slice_by_index(begin = var_12643_begin_0, end = var_12643_end_0, end_mask = var_12643_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12643_cast_fp16")]; + tensor var_12650_begin_0 = const()[name = string("op_12650_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12650_end_0 = const()[name = string("op_12650_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12650_end_mask_0 = const()[name = string("op_12650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12650_cast_fp16 = slice_by_index(begin = var_12650_begin_0, end = var_12650_end_0, end_mask = var_12650_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12650_cast_fp16")]; + tensor var_12657_begin_0 = const()[name = string("op_12657_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12657_end_0 = const()[name = string("op_12657_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12657_end_mask_0 = const()[name = string("op_12657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12657_cast_fp16 = slice_by_index(begin = var_12657_begin_0, end = var_12657_end_0, end_mask = var_12657_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12657_cast_fp16")]; + tensor var_12664_begin_0 = const()[name = string("op_12664_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12664_end_0 = const()[name = string("op_12664_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12664_end_mask_0 = const()[name = string("op_12664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12664_cast_fp16 = slice_by_index(begin = var_12664_begin_0, end = var_12664_end_0, end_mask = var_12664_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12664_cast_fp16")]; + tensor var_12671_begin_0 = const()[name = string("op_12671_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12671_end_0 = const()[name = string("op_12671_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12671_end_mask_0 = const()[name = string("op_12671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12671_cast_fp16 = slice_by_index(begin = var_12671_begin_0, end = var_12671_end_0, end_mask = var_12671_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12671_cast_fp16")]; + tensor var_12678_begin_0 = const()[name = string("op_12678_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12678_end_0 = const()[name = string("op_12678_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12678_end_mask_0 = const()[name = string("op_12678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12678_cast_fp16 = slice_by_index(begin = var_12678_begin_0, end = var_12678_end_0, end_mask = var_12678_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12678_cast_fp16")]; + tensor var_12685_begin_0 = const()[name = string("op_12685_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12685_end_0 = const()[name = string("op_12685_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12685_end_mask_0 = const()[name = string("op_12685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12685_cast_fp16 = slice_by_index(begin = var_12685_begin_0, end = var_12685_end_0, end_mask = var_12685_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12685_cast_fp16")]; + tensor var_12692_begin_0 = const()[name = string("op_12692_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12692_end_0 = const()[name = string("op_12692_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12692_end_mask_0 = const()[name = string("op_12692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12692_cast_fp16 = slice_by_index(begin = var_12692_begin_0, end = var_12692_end_0, end_mask = var_12692_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12692_cast_fp16")]; + tensor var_12699_begin_0 = const()[name = string("op_12699_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12699_end_0 = const()[name = string("op_12699_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12699_end_mask_0 = const()[name = string("op_12699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12699_cast_fp16 = slice_by_index(begin = var_12699_begin_0, end = var_12699_end_0, end_mask = var_12699_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12699_cast_fp16")]; + tensor var_12706_begin_0 = const()[name = string("op_12706_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12706_end_0 = const()[name = string("op_12706_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12706_end_mask_0 = const()[name = string("op_12706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12706_cast_fp16 = slice_by_index(begin = var_12706_begin_0, end = var_12706_end_0, end_mask = var_12706_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12706_cast_fp16")]; + tensor var_12713_begin_0 = const()[name = string("op_12713_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12713_end_0 = const()[name = string("op_12713_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12713_end_mask_0 = const()[name = string("op_12713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12713_cast_fp16 = slice_by_index(begin = var_12713_begin_0, end = var_12713_end_0, end_mask = var_12713_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12713_cast_fp16")]; + tensor var_12720_begin_0 = const()[name = string("op_12720_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12720_end_0 = const()[name = string("op_12720_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12720_end_mask_0 = const()[name = string("op_12720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12720_cast_fp16 = slice_by_index(begin = var_12720_begin_0, end = var_12720_end_0, end_mask = var_12720_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12720_cast_fp16")]; + tensor var_12727_begin_0 = const()[name = string("op_12727_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12727_end_0 = const()[name = string("op_12727_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12727_end_mask_0 = const()[name = string("op_12727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12727_cast_fp16 = slice_by_index(begin = var_12727_begin_0, end = var_12727_end_0, end_mask = var_12727_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12727_cast_fp16")]; + tensor var_12734_begin_0 = const()[name = string("op_12734_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12734_end_0 = const()[name = string("op_12734_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12734_end_mask_0 = const()[name = string("op_12734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12734_cast_fp16 = slice_by_index(begin = var_12734_begin_0, end = var_12734_end_0, end_mask = var_12734_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12734_cast_fp16")]; + tensor var_12741_begin_0 = const()[name = string("op_12741_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12741_end_0 = const()[name = string("op_12741_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12741_end_mask_0 = const()[name = string("op_12741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12741_cast_fp16 = slice_by_index(begin = var_12741_begin_0, end = var_12741_end_0, end_mask = var_12741_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12741_cast_fp16")]; + tensor var_12748_begin_0 = const()[name = string("op_12748_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12748_end_0 = const()[name = string("op_12748_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12748_end_mask_0 = const()[name = string("op_12748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12748_cast_fp16 = slice_by_index(begin = var_12748_begin_0, end = var_12748_end_0, end_mask = var_12748_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12748_cast_fp16")]; + tensor var_12755_begin_0 = const()[name = string("op_12755_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12755_end_0 = const()[name = string("op_12755_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12755_end_mask_0 = const()[name = string("op_12755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12755_cast_fp16 = slice_by_index(begin = var_12755_begin_0, end = var_12755_end_0, end_mask = var_12755_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12755_cast_fp16")]; + tensor var_12762_begin_0 = const()[name = string("op_12762_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12762_end_0 = const()[name = string("op_12762_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12762_end_mask_0 = const()[name = string("op_12762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12762_cast_fp16 = slice_by_index(begin = var_12762_begin_0, end = var_12762_end_0, end_mask = var_12762_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12762_cast_fp16")]; + tensor var_12769_begin_0 = const()[name = string("op_12769_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12769_end_0 = const()[name = string("op_12769_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12769_end_mask_0 = const()[name = string("op_12769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12769_cast_fp16 = slice_by_index(begin = var_12769_begin_0, end = var_12769_end_0, end_mask = var_12769_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12769_cast_fp16")]; + tensor var_12776_begin_0 = const()[name = string("op_12776_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12776_end_0 = const()[name = string("op_12776_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12776_end_mask_0 = const()[name = string("op_12776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12776_cast_fp16 = slice_by_index(begin = var_12776_begin_0, end = var_12776_end_0, end_mask = var_12776_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12776_cast_fp16")]; + tensor var_12783_begin_0 = const()[name = string("op_12783_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12783_end_0 = const()[name = string("op_12783_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12783_end_mask_0 = const()[name = string("op_12783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12783_cast_fp16 = slice_by_index(begin = var_12783_begin_0, end = var_12783_end_0, end_mask = var_12783_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12783_cast_fp16")]; + tensor var_12790_begin_0 = const()[name = string("op_12790_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12790_end_0 = const()[name = string("op_12790_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12790_end_mask_0 = const()[name = string("op_12790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12790_cast_fp16 = slice_by_index(begin = var_12790_begin_0, end = var_12790_end_0, end_mask = var_12790_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12790_cast_fp16")]; + tensor var_12797_begin_0 = const()[name = string("op_12797_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12797_end_0 = const()[name = string("op_12797_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12797_end_mask_0 = const()[name = string("op_12797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12797_cast_fp16 = slice_by_index(begin = var_12797_begin_0, end = var_12797_end_0, end_mask = var_12797_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12797_cast_fp16")]; + tensor var_12804_begin_0 = const()[name = string("op_12804_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12804_end_0 = const()[name = string("op_12804_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12804_end_mask_0 = const()[name = string("op_12804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12804_cast_fp16 = slice_by_index(begin = var_12804_begin_0, end = var_12804_end_0, end_mask = var_12804_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12804_cast_fp16")]; + tensor var_12811_begin_0 = const()[name = string("op_12811_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12811_end_0 = const()[name = string("op_12811_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12811_end_mask_0 = const()[name = string("op_12811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12811_cast_fp16 = slice_by_index(begin = var_12811_begin_0, end = var_12811_end_0, end_mask = var_12811_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12811_cast_fp16")]; + tensor var_12818_begin_0 = const()[name = string("op_12818_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12818_end_0 = const()[name = string("op_12818_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12818_end_mask_0 = const()[name = string("op_12818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12818_cast_fp16 = slice_by_index(begin = var_12818_begin_0, end = var_12818_end_0, end_mask = var_12818_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12818_cast_fp16")]; + tensor var_12825_begin_0 = const()[name = string("op_12825_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12825_end_0 = const()[name = string("op_12825_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12825_end_mask_0 = const()[name = string("op_12825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12825_cast_fp16 = slice_by_index(begin = var_12825_begin_0, end = var_12825_end_0, end_mask = var_12825_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12825_cast_fp16")]; + tensor var_12832_begin_0 = const()[name = string("op_12832_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12832_end_0 = const()[name = string("op_12832_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12832_end_mask_0 = const()[name = string("op_12832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12832_cast_fp16 = slice_by_index(begin = var_12832_begin_0, end = var_12832_end_0, end_mask = var_12832_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12832_cast_fp16")]; + tensor var_12839_begin_0 = const()[name = string("op_12839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12839_end_0 = const()[name = string("op_12839_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12839_end_mask_0 = const()[name = string("op_12839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12839_cast_fp16 = slice_by_index(begin = var_12839_begin_0, end = var_12839_end_0, end_mask = var_12839_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12839_cast_fp16")]; + tensor var_12846_begin_0 = const()[name = string("op_12846_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12846_end_0 = const()[name = string("op_12846_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12846_end_mask_0 = const()[name = string("op_12846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12846_cast_fp16 = slice_by_index(begin = var_12846_begin_0, end = var_12846_end_0, end_mask = var_12846_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12846_cast_fp16")]; + tensor var_12853_begin_0 = const()[name = string("op_12853_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12853_end_0 = const()[name = string("op_12853_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12853_end_mask_0 = const()[name = string("op_12853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12853_cast_fp16 = slice_by_index(begin = var_12853_begin_0, end = var_12853_end_0, end_mask = var_12853_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12853_cast_fp16")]; + tensor var_12860_begin_0 = const()[name = string("op_12860_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12860_end_0 = const()[name = string("op_12860_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12860_end_mask_0 = const()[name = string("op_12860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12860_cast_fp16 = slice_by_index(begin = var_12860_begin_0, end = var_12860_end_0, end_mask = var_12860_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12860_cast_fp16")]; + tensor var_12867_begin_0 = const()[name = string("op_12867_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12867_end_0 = const()[name = string("op_12867_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12867_end_mask_0 = const()[name = string("op_12867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12867_cast_fp16 = slice_by_index(begin = var_12867_begin_0, end = var_12867_end_0, end_mask = var_12867_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12867_cast_fp16")]; + tensor var_12874_begin_0 = const()[name = string("op_12874_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12874_end_0 = const()[name = string("op_12874_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12874_end_mask_0 = const()[name = string("op_12874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12874_cast_fp16 = slice_by_index(begin = var_12874_begin_0, end = var_12874_end_0, end_mask = var_12874_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12874_cast_fp16")]; + tensor var_12881_begin_0 = const()[name = string("op_12881_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12881_end_0 = const()[name = string("op_12881_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12881_end_mask_0 = const()[name = string("op_12881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12881_cast_fp16 = slice_by_index(begin = var_12881_begin_0, end = var_12881_end_0, end_mask = var_12881_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12881_cast_fp16")]; + tensor var_12888_begin_0 = const()[name = string("op_12888_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12888_end_0 = const()[name = string("op_12888_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12888_end_mask_0 = const()[name = string("op_12888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12888_cast_fp16 = slice_by_index(begin = var_12888_begin_0, end = var_12888_end_0, end_mask = var_12888_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12888_cast_fp16")]; + tensor var_12895_begin_0 = const()[name = string("op_12895_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12895_end_0 = const()[name = string("op_12895_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12895_end_mask_0 = const()[name = string("op_12895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12895_cast_fp16 = slice_by_index(begin = var_12895_begin_0, end = var_12895_end_0, end_mask = var_12895_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12895_cast_fp16")]; + tensor var_12902_begin_0 = const()[name = string("op_12902_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12902_end_0 = const()[name = string("op_12902_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12902_end_mask_0 = const()[name = string("op_12902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12902_cast_fp16 = slice_by_index(begin = var_12902_begin_0, end = var_12902_end_0, end_mask = var_12902_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12902_cast_fp16")]; + tensor var_12909_begin_0 = const()[name = string("op_12909_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12909_end_0 = const()[name = string("op_12909_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12909_end_mask_0 = const()[name = string("op_12909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12909_cast_fp16 = slice_by_index(begin = var_12909_begin_0, end = var_12909_end_0, end_mask = var_12909_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12909_cast_fp16")]; + tensor var_12916_begin_0 = const()[name = string("op_12916_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12916_end_0 = const()[name = string("op_12916_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12916_end_mask_0 = const()[name = string("op_12916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12916_cast_fp16 = slice_by_index(begin = var_12916_begin_0, end = var_12916_end_0, end_mask = var_12916_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12916_cast_fp16")]; + tensor var_12923_begin_0 = const()[name = string("op_12923_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12923_end_0 = const()[name = string("op_12923_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12923_end_mask_0 = const()[name = string("op_12923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12923_cast_fp16 = slice_by_index(begin = var_12923_begin_0, end = var_12923_end_0, end_mask = var_12923_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12923_cast_fp16")]; + tensor var_12930_begin_0 = const()[name = string("op_12930_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12930_end_0 = const()[name = string("op_12930_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12930_end_mask_0 = const()[name = string("op_12930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12930_cast_fp16 = slice_by_index(begin = var_12930_begin_0, end = var_12930_end_0, end_mask = var_12930_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12930_cast_fp16")]; + tensor var_12937_begin_0 = const()[name = string("op_12937_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12937_end_0 = const()[name = string("op_12937_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12937_end_mask_0 = const()[name = string("op_12937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12937_cast_fp16 = slice_by_index(begin = var_12937_begin_0, end = var_12937_end_0, end_mask = var_12937_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12937_cast_fp16")]; + tensor var_12944_begin_0 = const()[name = string("op_12944_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12944_end_0 = const()[name = string("op_12944_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12944_end_mask_0 = const()[name = string("op_12944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12944_cast_fp16 = slice_by_index(begin = var_12944_begin_0, end = var_12944_end_0, end_mask = var_12944_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12944_cast_fp16")]; + tensor var_12951_begin_0 = const()[name = string("op_12951_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12951_end_0 = const()[name = string("op_12951_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12951_end_mask_0 = const()[name = string("op_12951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12951_cast_fp16 = slice_by_index(begin = var_12951_begin_0, end = var_12951_end_0, end_mask = var_12951_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12951_cast_fp16")]; + tensor var_12958_begin_0 = const()[name = string("op_12958_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12958_end_0 = const()[name = string("op_12958_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12958_end_mask_0 = const()[name = string("op_12958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12958_cast_fp16 = slice_by_index(begin = var_12958_begin_0, end = var_12958_end_0, end_mask = var_12958_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12958_cast_fp16")]; + tensor var_12965_begin_0 = const()[name = string("op_12965_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12965_end_0 = const()[name = string("op_12965_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12965_end_mask_0 = const()[name = string("op_12965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12965_cast_fp16 = slice_by_index(begin = var_12965_begin_0, end = var_12965_end_0, end_mask = var_12965_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12965_cast_fp16")]; + tensor var_12972_begin_0 = const()[name = string("op_12972_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12972_end_0 = const()[name = string("op_12972_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12972_end_mask_0 = const()[name = string("op_12972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12972_cast_fp16 = slice_by_index(begin = var_12972_begin_0, end = var_12972_end_0, end_mask = var_12972_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12972_cast_fp16")]; + tensor var_12979_begin_0 = const()[name = string("op_12979_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12979_end_0 = const()[name = string("op_12979_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12979_end_mask_0 = const()[name = string("op_12979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12979_cast_fp16 = slice_by_index(begin = var_12979_begin_0, end = var_12979_end_0, end_mask = var_12979_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12979_cast_fp16")]; + tensor var_12986_begin_0 = const()[name = string("op_12986_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12986_end_0 = const()[name = string("op_12986_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12986_end_mask_0 = const()[name = string("op_12986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12986_cast_fp16 = slice_by_index(begin = var_12986_begin_0, end = var_12986_end_0, end_mask = var_12986_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12986_cast_fp16")]; + tensor var_12993_begin_0 = const()[name = string("op_12993_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12993_end_0 = const()[name = string("op_12993_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12993_end_mask_0 = const()[name = string("op_12993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12993_cast_fp16 = slice_by_index(begin = var_12993_begin_0, end = var_12993_end_0, end_mask = var_12993_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12993_cast_fp16")]; + tensor var_13000_begin_0 = const()[name = string("op_13000_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13000_end_0 = const()[name = string("op_13000_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13000_end_mask_0 = const()[name = string("op_13000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13000_cast_fp16 = slice_by_index(begin = var_13000_begin_0, end = var_13000_end_0, end_mask = var_13000_end_mask_0, x = var_12534_cast_fp16)[name = string("op_13000_cast_fp16")]; + tensor var_13007_begin_0 = const()[name = string("op_13007_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13007_end_0 = const()[name = string("op_13007_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13007_end_mask_0 = const()[name = string("op_13007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13007_cast_fp16 = slice_by_index(begin = var_13007_begin_0, end = var_13007_end_0, end_mask = var_13007_end_mask_0, x = var_12538_cast_fp16)[name = string("op_13007_cast_fp16")]; + tensor var_13014_begin_0 = const()[name = string("op_13014_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13014_end_0 = const()[name = string("op_13014_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13014_end_mask_0 = const()[name = string("op_13014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13014_cast_fp16 = slice_by_index(begin = var_13014_begin_0, end = var_13014_end_0, end_mask = var_13014_end_mask_0, x = var_12538_cast_fp16)[name = string("op_13014_cast_fp16")]; + tensor var_13021_begin_0 = const()[name = string("op_13021_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13021_end_0 = const()[name = string("op_13021_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13021_end_mask_0 = const()[name = string("op_13021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13021_cast_fp16 = slice_by_index(begin = var_13021_begin_0, end = var_13021_end_0, end_mask = var_13021_end_mask_0, x = var_12538_cast_fp16)[name = string("op_13021_cast_fp16")]; + tensor var_13028_begin_0 = const()[name = string("op_13028_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13028_end_0 = const()[name = string("op_13028_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13028_end_mask_0 = const()[name = string("op_13028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13028_cast_fp16 = slice_by_index(begin = var_13028_begin_0, end = var_13028_end_0, end_mask = var_13028_end_mask_0, x = var_12538_cast_fp16)[name = string("op_13028_cast_fp16")]; + tensor var_13035_begin_0 = const()[name = string("op_13035_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13035_end_0 = const()[name = string("op_13035_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13035_end_mask_0 = const()[name = string("op_13035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13035_cast_fp16 = slice_by_index(begin = var_13035_begin_0, end = var_13035_end_0, end_mask = var_13035_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13035_cast_fp16")]; + tensor var_13042_begin_0 = const()[name = string("op_13042_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13042_end_0 = const()[name = string("op_13042_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13042_end_mask_0 = const()[name = string("op_13042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13042_cast_fp16 = slice_by_index(begin = var_13042_begin_0, end = var_13042_end_0, end_mask = var_13042_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13042_cast_fp16")]; + tensor var_13049_begin_0 = const()[name = string("op_13049_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13049_end_0 = const()[name = string("op_13049_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13049_end_mask_0 = const()[name = string("op_13049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13049_cast_fp16 = slice_by_index(begin = var_13049_begin_0, end = var_13049_end_0, end_mask = var_13049_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13049_cast_fp16")]; + tensor var_13056_begin_0 = const()[name = string("op_13056_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13056_end_0 = const()[name = string("op_13056_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13056_end_mask_0 = const()[name = string("op_13056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13056_cast_fp16 = slice_by_index(begin = var_13056_begin_0, end = var_13056_end_0, end_mask = var_13056_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13056_cast_fp16")]; + tensor var_13063_begin_0 = const()[name = string("op_13063_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13063_end_0 = const()[name = string("op_13063_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13063_end_mask_0 = const()[name = string("op_13063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13063_cast_fp16 = slice_by_index(begin = var_13063_begin_0, end = var_13063_end_0, end_mask = var_13063_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13063_cast_fp16")]; + tensor var_13070_begin_0 = const()[name = string("op_13070_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13070_end_0 = const()[name = string("op_13070_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13070_end_mask_0 = const()[name = string("op_13070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13070_cast_fp16 = slice_by_index(begin = var_13070_begin_0, end = var_13070_end_0, end_mask = var_13070_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13070_cast_fp16")]; + tensor var_13077_begin_0 = const()[name = string("op_13077_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13077_end_0 = const()[name = string("op_13077_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13077_end_mask_0 = const()[name = string("op_13077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13077_cast_fp16 = slice_by_index(begin = var_13077_begin_0, end = var_13077_end_0, end_mask = var_13077_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13077_cast_fp16")]; + tensor var_13084_begin_0 = const()[name = string("op_13084_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13084_end_0 = const()[name = string("op_13084_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13084_end_mask_0 = const()[name = string("op_13084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13084_cast_fp16 = slice_by_index(begin = var_13084_begin_0, end = var_13084_end_0, end_mask = var_13084_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13084_cast_fp16")]; + tensor var_13091_begin_0 = const()[name = string("op_13091_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13091_end_0 = const()[name = string("op_13091_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13091_end_mask_0 = const()[name = string("op_13091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13091_cast_fp16 = slice_by_index(begin = var_13091_begin_0, end = var_13091_end_0, end_mask = var_13091_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13091_cast_fp16")]; + tensor var_13098_begin_0 = const()[name = string("op_13098_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13098_end_0 = const()[name = string("op_13098_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13098_end_mask_0 = const()[name = string("op_13098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13098_cast_fp16 = slice_by_index(begin = var_13098_begin_0, end = var_13098_end_0, end_mask = var_13098_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13098_cast_fp16")]; + tensor var_13105_begin_0 = const()[name = string("op_13105_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13105_end_0 = const()[name = string("op_13105_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13105_end_mask_0 = const()[name = string("op_13105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13105_cast_fp16 = slice_by_index(begin = var_13105_begin_0, end = var_13105_end_0, end_mask = var_13105_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13105_cast_fp16")]; + tensor var_13112_begin_0 = const()[name = string("op_13112_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13112_end_0 = const()[name = string("op_13112_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13112_end_mask_0 = const()[name = string("op_13112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13112_cast_fp16 = slice_by_index(begin = var_13112_begin_0, end = var_13112_end_0, end_mask = var_13112_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13112_cast_fp16")]; + tensor k_17_perm_0 = const()[name = string("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13117_begin_0 = const()[name = string("op_13117_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13117_end_0 = const()[name = string("op_13117_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_13117_end_mask_0 = const()[name = string("op_13117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_17_cast_fp16 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = string("transpose_23")]; + tensor var_13117_cast_fp16 = slice_by_index(begin = var_13117_begin_0, end = var_13117_end_0, end_mask = var_13117_end_mask_0, x = k_17_cast_fp16)[name = string("op_13117_cast_fp16")]; + tensor var_13121_begin_0 = const()[name = string("op_13121_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13121_end_0 = const()[name = string("op_13121_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_13121_end_mask_0 = const()[name = string("op_13121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13121_cast_fp16 = slice_by_index(begin = var_13121_begin_0, end = var_13121_end_0, end_mask = var_13121_end_mask_0, x = k_17_cast_fp16)[name = string("op_13121_cast_fp16")]; + tensor var_13125_begin_0 = const()[name = string("op_13125_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13125_end_0 = const()[name = string("op_13125_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_13125_end_mask_0 = const()[name = string("op_13125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13125_cast_fp16 = slice_by_index(begin = var_13125_begin_0, end = var_13125_end_0, end_mask = var_13125_end_mask_0, x = k_17_cast_fp16)[name = string("op_13125_cast_fp16")]; + tensor var_13129_begin_0 = const()[name = string("op_13129_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13129_end_0 = const()[name = string("op_13129_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_13129_end_mask_0 = const()[name = string("op_13129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13129_cast_fp16 = slice_by_index(begin = var_13129_begin_0, end = var_13129_end_0, end_mask = var_13129_end_mask_0, x = k_17_cast_fp16)[name = string("op_13129_cast_fp16")]; + tensor var_13133_begin_0 = const()[name = string("op_13133_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13133_end_0 = const()[name = string("op_13133_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_13133_end_mask_0 = const()[name = string("op_13133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13133_cast_fp16 = slice_by_index(begin = var_13133_begin_0, end = var_13133_end_0, end_mask = var_13133_end_mask_0, x = k_17_cast_fp16)[name = string("op_13133_cast_fp16")]; + tensor var_13137_begin_0 = const()[name = string("op_13137_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_13137_end_0 = const()[name = string("op_13137_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_13137_end_mask_0 = const()[name = string("op_13137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13137_cast_fp16 = slice_by_index(begin = var_13137_begin_0, end = var_13137_end_0, end_mask = var_13137_end_mask_0, x = k_17_cast_fp16)[name = string("op_13137_cast_fp16")]; + tensor var_13141_begin_0 = const()[name = string("op_13141_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_13141_end_0 = const()[name = string("op_13141_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_13141_end_mask_0 = const()[name = string("op_13141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13141_cast_fp16 = slice_by_index(begin = var_13141_begin_0, end = var_13141_end_0, end_mask = var_13141_end_mask_0, x = k_17_cast_fp16)[name = string("op_13141_cast_fp16")]; + tensor var_13145_begin_0 = const()[name = string("op_13145_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_13145_end_0 = const()[name = string("op_13145_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_13145_end_mask_0 = const()[name = string("op_13145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13145_cast_fp16 = slice_by_index(begin = var_13145_begin_0, end = var_13145_end_0, end_mask = var_13145_end_mask_0, x = k_17_cast_fp16)[name = string("op_13145_cast_fp16")]; + tensor var_13149_begin_0 = const()[name = string("op_13149_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13149_end_0 = const()[name = string("op_13149_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_13149_end_mask_0 = const()[name = string("op_13149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13149_cast_fp16 = slice_by_index(begin = var_13149_begin_0, end = var_13149_end_0, end_mask = var_13149_end_mask_0, x = k_17_cast_fp16)[name = string("op_13149_cast_fp16")]; + tensor var_13153_begin_0 = const()[name = string("op_13153_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_13153_end_0 = const()[name = string("op_13153_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_13153_end_mask_0 = const()[name = string("op_13153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13153_cast_fp16 = slice_by_index(begin = var_13153_begin_0, end = var_13153_end_0, end_mask = var_13153_end_mask_0, x = k_17_cast_fp16)[name = string("op_13153_cast_fp16")]; + tensor var_13157_begin_0 = const()[name = string("op_13157_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13157_end_0 = const()[name = string("op_13157_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_13157_end_mask_0 = const()[name = string("op_13157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13157_cast_fp16 = slice_by_index(begin = var_13157_begin_0, end = var_13157_end_0, end_mask = var_13157_end_mask_0, x = k_17_cast_fp16)[name = string("op_13157_cast_fp16")]; + tensor var_13161_begin_0 = const()[name = string("op_13161_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13161_end_0 = const()[name = string("op_13161_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_13161_end_mask_0 = const()[name = string("op_13161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13161_cast_fp16 = slice_by_index(begin = var_13161_begin_0, end = var_13161_end_0, end_mask = var_13161_end_mask_0, x = k_17_cast_fp16)[name = string("op_13161_cast_fp16")]; + tensor var_13165_begin_0 = const()[name = string("op_13165_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13165_end_0 = const()[name = string("op_13165_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_13165_end_mask_0 = const()[name = string("op_13165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13165_cast_fp16 = slice_by_index(begin = var_13165_begin_0, end = var_13165_end_0, end_mask = var_13165_end_mask_0, x = k_17_cast_fp16)[name = string("op_13165_cast_fp16")]; + tensor var_13169_begin_0 = const()[name = string("op_13169_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13169_end_0 = const()[name = string("op_13169_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_13169_end_mask_0 = const()[name = string("op_13169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13169_cast_fp16 = slice_by_index(begin = var_13169_begin_0, end = var_13169_end_0, end_mask = var_13169_end_mask_0, x = k_17_cast_fp16)[name = string("op_13169_cast_fp16")]; + tensor var_13173_begin_0 = const()[name = string("op_13173_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13173_end_0 = const()[name = string("op_13173_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_13173_end_mask_0 = const()[name = string("op_13173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13173_cast_fp16 = slice_by_index(begin = var_13173_begin_0, end = var_13173_end_0, end_mask = var_13173_end_mask_0, x = k_17_cast_fp16)[name = string("op_13173_cast_fp16")]; + tensor var_13177_begin_0 = const()[name = string("op_13177_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13177_end_0 = const()[name = string("op_13177_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_13177_end_mask_0 = const()[name = string("op_13177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13177_cast_fp16 = slice_by_index(begin = var_13177_begin_0, end = var_13177_end_0, end_mask = var_13177_end_mask_0, x = k_17_cast_fp16)[name = string("op_13177_cast_fp16")]; + tensor var_13181_begin_0 = const()[name = string("op_13181_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13181_end_0 = const()[name = string("op_13181_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_13181_end_mask_0 = const()[name = string("op_13181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13181_cast_fp16 = slice_by_index(begin = var_13181_begin_0, end = var_13181_end_0, end_mask = var_13181_end_mask_0, x = k_17_cast_fp16)[name = string("op_13181_cast_fp16")]; + tensor var_13185_begin_0 = const()[name = string("op_13185_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13185_end_0 = const()[name = string("op_13185_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_13185_end_mask_0 = const()[name = string("op_13185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13185_cast_fp16 = slice_by_index(begin = var_13185_begin_0, end = var_13185_end_0, end_mask = var_13185_end_mask_0, x = k_17_cast_fp16)[name = string("op_13185_cast_fp16")]; + tensor var_13189_begin_0 = const()[name = string("op_13189_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13189_end_0 = const()[name = string("op_13189_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_13189_end_mask_0 = const()[name = string("op_13189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13189_cast_fp16 = slice_by_index(begin = var_13189_begin_0, end = var_13189_end_0, end_mask = var_13189_end_mask_0, x = k_17_cast_fp16)[name = string("op_13189_cast_fp16")]; + tensor var_13193_begin_0 = const()[name = string("op_13193_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13193_end_0 = const()[name = string("op_13193_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_13193_end_mask_0 = const()[name = string("op_13193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13193_cast_fp16 = slice_by_index(begin = var_13193_begin_0, end = var_13193_end_0, end_mask = var_13193_end_mask_0, x = k_17_cast_fp16)[name = string("op_13193_cast_fp16")]; + tensor var_13195_begin_0 = const()[name = string("op_13195_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13195_end_0 = const()[name = string("op_13195_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13195_end_mask_0 = const()[name = string("op_13195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13195_cast_fp16 = slice_by_index(begin = var_13195_begin_0, end = var_13195_end_0, end_mask = var_13195_end_mask_0, x = value_17_cast_fp16)[name = string("op_13195_cast_fp16")]; + tensor var_13199_begin_0 = const()[name = string("op_13199_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13199_end_0 = const()[name = string("op_13199_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_13199_end_mask_0 = const()[name = string("op_13199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13199_cast_fp16 = slice_by_index(begin = var_13199_begin_0, end = var_13199_end_0, end_mask = var_13199_end_mask_0, x = value_17_cast_fp16)[name = string("op_13199_cast_fp16")]; + tensor var_13203_begin_0 = const()[name = string("op_13203_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13203_end_0 = const()[name = string("op_13203_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_13203_end_mask_0 = const()[name = string("op_13203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13203_cast_fp16 = slice_by_index(begin = var_13203_begin_0, end = var_13203_end_0, end_mask = var_13203_end_mask_0, x = value_17_cast_fp16)[name = string("op_13203_cast_fp16")]; + tensor var_13207_begin_0 = const()[name = string("op_13207_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13207_end_0 = const()[name = string("op_13207_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_13207_end_mask_0 = const()[name = string("op_13207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13207_cast_fp16 = slice_by_index(begin = var_13207_begin_0, end = var_13207_end_0, end_mask = var_13207_end_mask_0, x = value_17_cast_fp16)[name = string("op_13207_cast_fp16")]; + tensor var_13211_begin_0 = const()[name = string("op_13211_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13211_end_0 = const()[name = string("op_13211_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_13211_end_mask_0 = const()[name = string("op_13211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13211_cast_fp16 = slice_by_index(begin = var_13211_begin_0, end = var_13211_end_0, end_mask = var_13211_end_mask_0, x = value_17_cast_fp16)[name = string("op_13211_cast_fp16")]; + tensor var_13215_begin_0 = const()[name = string("op_13215_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13215_end_0 = const()[name = string("op_13215_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_13215_end_mask_0 = const()[name = string("op_13215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13215_cast_fp16 = slice_by_index(begin = var_13215_begin_0, end = var_13215_end_0, end_mask = var_13215_end_mask_0, x = value_17_cast_fp16)[name = string("op_13215_cast_fp16")]; + tensor var_13219_begin_0 = const()[name = string("op_13219_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13219_end_0 = const()[name = string("op_13219_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_13219_end_mask_0 = const()[name = string("op_13219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13219_cast_fp16 = slice_by_index(begin = var_13219_begin_0, end = var_13219_end_0, end_mask = var_13219_end_mask_0, x = value_17_cast_fp16)[name = string("op_13219_cast_fp16")]; + tensor var_13223_begin_0 = const()[name = string("op_13223_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13223_end_0 = const()[name = string("op_13223_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_13223_end_mask_0 = const()[name = string("op_13223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13223_cast_fp16 = slice_by_index(begin = var_13223_begin_0, end = var_13223_end_0, end_mask = var_13223_end_mask_0, x = value_17_cast_fp16)[name = string("op_13223_cast_fp16")]; + tensor var_13227_begin_0 = const()[name = string("op_13227_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13227_end_0 = const()[name = string("op_13227_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_13227_end_mask_0 = const()[name = string("op_13227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13227_cast_fp16 = slice_by_index(begin = var_13227_begin_0, end = var_13227_end_0, end_mask = var_13227_end_mask_0, x = value_17_cast_fp16)[name = string("op_13227_cast_fp16")]; + tensor var_13231_begin_0 = const()[name = string("op_13231_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13231_end_0 = const()[name = string("op_13231_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_13231_end_mask_0 = const()[name = string("op_13231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13231_cast_fp16 = slice_by_index(begin = var_13231_begin_0, end = var_13231_end_0, end_mask = var_13231_end_mask_0, x = value_17_cast_fp16)[name = string("op_13231_cast_fp16")]; + tensor var_13235_begin_0 = const()[name = string("op_13235_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13235_end_0 = const()[name = string("op_13235_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_13235_end_mask_0 = const()[name = string("op_13235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13235_cast_fp16 = slice_by_index(begin = var_13235_begin_0, end = var_13235_end_0, end_mask = var_13235_end_mask_0, x = value_17_cast_fp16)[name = string("op_13235_cast_fp16")]; + tensor var_13239_begin_0 = const()[name = string("op_13239_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13239_end_0 = const()[name = string("op_13239_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_13239_end_mask_0 = const()[name = string("op_13239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13239_cast_fp16 = slice_by_index(begin = var_13239_begin_0, end = var_13239_end_0, end_mask = var_13239_end_mask_0, x = value_17_cast_fp16)[name = string("op_13239_cast_fp16")]; + tensor var_13243_begin_0 = const()[name = string("op_13243_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13243_end_0 = const()[name = string("op_13243_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_13243_end_mask_0 = const()[name = string("op_13243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13243_cast_fp16 = slice_by_index(begin = var_13243_begin_0, end = var_13243_end_0, end_mask = var_13243_end_mask_0, x = value_17_cast_fp16)[name = string("op_13243_cast_fp16")]; + tensor var_13247_begin_0 = const()[name = string("op_13247_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13247_end_0 = const()[name = string("op_13247_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_13247_end_mask_0 = const()[name = string("op_13247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13247_cast_fp16 = slice_by_index(begin = var_13247_begin_0, end = var_13247_end_0, end_mask = var_13247_end_mask_0, x = value_17_cast_fp16)[name = string("op_13247_cast_fp16")]; + tensor var_13251_begin_0 = const()[name = string("op_13251_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13251_end_0 = const()[name = string("op_13251_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_13251_end_mask_0 = const()[name = string("op_13251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13251_cast_fp16 = slice_by_index(begin = var_13251_begin_0, end = var_13251_end_0, end_mask = var_13251_end_mask_0, x = value_17_cast_fp16)[name = string("op_13251_cast_fp16")]; + tensor var_13255_begin_0 = const()[name = string("op_13255_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13255_end_0 = const()[name = string("op_13255_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_13255_end_mask_0 = const()[name = string("op_13255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13255_cast_fp16 = slice_by_index(begin = var_13255_begin_0, end = var_13255_end_0, end_mask = var_13255_end_mask_0, x = value_17_cast_fp16)[name = string("op_13255_cast_fp16")]; + tensor var_13259_begin_0 = const()[name = string("op_13259_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13259_end_0 = const()[name = string("op_13259_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_13259_end_mask_0 = const()[name = string("op_13259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13259_cast_fp16 = slice_by_index(begin = var_13259_begin_0, end = var_13259_end_0, end_mask = var_13259_end_mask_0, x = value_17_cast_fp16)[name = string("op_13259_cast_fp16")]; + tensor var_13263_begin_0 = const()[name = string("op_13263_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13263_end_0 = const()[name = string("op_13263_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_13263_end_mask_0 = const()[name = string("op_13263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13263_cast_fp16 = slice_by_index(begin = var_13263_begin_0, end = var_13263_end_0, end_mask = var_13263_end_mask_0, x = value_17_cast_fp16)[name = string("op_13263_cast_fp16")]; + tensor var_13267_begin_0 = const()[name = string("op_13267_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13267_end_0 = const()[name = string("op_13267_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_13267_end_mask_0 = const()[name = string("op_13267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13267_cast_fp16 = slice_by_index(begin = var_13267_begin_0, end = var_13267_end_0, end_mask = var_13267_end_mask_0, x = value_17_cast_fp16)[name = string("op_13267_cast_fp16")]; + tensor var_13271_begin_0 = const()[name = string("op_13271_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13271_end_0 = const()[name = string("op_13271_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_13271_end_mask_0 = const()[name = string("op_13271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13271_cast_fp16 = slice_by_index(begin = var_13271_begin_0, end = var_13271_end_0, end_mask = var_13271_end_mask_0, x = value_17_cast_fp16)[name = string("op_13271_cast_fp16")]; + string _SplitHeadsQ__mh_w_1281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1281_equation_0, values = (var_13117_cast_fp16, var_12559_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1281_cast_fp16")]; + string _SplitHeadsQ__mh_w_1283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1283_equation_0, values = (var_13117_cast_fp16, var_12566_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1283_cast_fp16")]; + string _SplitHeadsQ__mh_w_1285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1285_equation_0, values = (var_13117_cast_fp16, var_12573_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1285_cast_fp16")]; + string _SplitHeadsQ__mh_w_1287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1287_equation_0, values = (var_13117_cast_fp16, var_12580_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1287_cast_fp16")]; + string _SplitHeadsQ__mh_w_1289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1289_equation_0, values = (var_13121_cast_fp16, var_12587_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1289_cast_fp16")]; + string _SplitHeadsQ__mh_w_1291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1291_equation_0, values = (var_13121_cast_fp16, var_12594_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1291_cast_fp16")]; + string _SplitHeadsQ__mh_w_1293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1293_equation_0, values = (var_13121_cast_fp16, var_12601_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1293_cast_fp16")]; + string _SplitHeadsQ__mh_w_1295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1295_equation_0, values = (var_13121_cast_fp16, var_12608_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1295_cast_fp16")]; + string _SplitHeadsQ__mh_w_1297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1297_equation_0, values = (var_13125_cast_fp16, var_12615_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1297_cast_fp16")]; + string _SplitHeadsQ__mh_w_1299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1299_equation_0, values = (var_13125_cast_fp16, var_12622_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1299_cast_fp16")]; + string _SplitHeadsQ__mh_w_1301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1301_equation_0, values = (var_13125_cast_fp16, var_12629_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1301_cast_fp16")]; + string _SplitHeadsQ__mh_w_1303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1303_equation_0, values = (var_13125_cast_fp16, var_12636_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1303_cast_fp16")]; + string _SplitHeadsQ__mh_w_1305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1305_equation_0, values = (var_13129_cast_fp16, var_12643_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1305_cast_fp16")]; + string _SplitHeadsQ__mh_w_1307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1307_equation_0, values = (var_13129_cast_fp16, var_12650_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1307_cast_fp16")]; + string _SplitHeadsQ__mh_w_1309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1309_equation_0, values = (var_13129_cast_fp16, var_12657_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1309_cast_fp16")]; + string _SplitHeadsQ__mh_w_1311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1311_equation_0, values = (var_13129_cast_fp16, var_12664_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1311_cast_fp16")]; + string _SplitHeadsQ__mh_w_1313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1313_equation_0, values = (var_13133_cast_fp16, var_12671_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1313_cast_fp16")]; + string _SplitHeadsQ__mh_w_1315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1315_equation_0, values = (var_13133_cast_fp16, var_12678_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1315_cast_fp16")]; + string _SplitHeadsQ__mh_w_1317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1317_equation_0, values = (var_13133_cast_fp16, var_12685_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1317_cast_fp16")]; + string _SplitHeadsQ__mh_w_1319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1319_equation_0, values = (var_13133_cast_fp16, var_12692_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1319_cast_fp16")]; + string _SplitHeadsQ__mh_w_1321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1321_equation_0, values = (var_13137_cast_fp16, var_12699_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1321_cast_fp16")]; + string _SplitHeadsQ__mh_w_1323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1323_equation_0, values = (var_13137_cast_fp16, var_12706_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1323_cast_fp16")]; + string _SplitHeadsQ__mh_w_1325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1325_equation_0, values = (var_13137_cast_fp16, var_12713_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1325_cast_fp16")]; + string _SplitHeadsQ__mh_w_1327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1327_equation_0, values = (var_13137_cast_fp16, var_12720_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1327_cast_fp16")]; + string _SplitHeadsQ__mh_w_1329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1329_equation_0, values = (var_13141_cast_fp16, var_12727_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1329_cast_fp16")]; + string _SplitHeadsQ__mh_w_1331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1331_equation_0, values = (var_13141_cast_fp16, var_12734_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1331_cast_fp16")]; + string _SplitHeadsQ__mh_w_1333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1333_equation_0, values = (var_13141_cast_fp16, var_12741_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1333_cast_fp16")]; + string _SplitHeadsQ__mh_w_1335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1335_equation_0, values = (var_13141_cast_fp16, var_12748_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1335_cast_fp16")]; + string _SplitHeadsQ__mh_w_1337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1337_equation_0, values = (var_13145_cast_fp16, var_12755_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1337_cast_fp16")]; + string _SplitHeadsQ__mh_w_1339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1339_equation_0, values = (var_13145_cast_fp16, var_12762_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1339_cast_fp16")]; + string _SplitHeadsQ__mh_w_1341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1341_equation_0, values = (var_13145_cast_fp16, var_12769_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1341_cast_fp16")]; + string _SplitHeadsQ__mh_w_1343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1343_equation_0, values = (var_13145_cast_fp16, var_12776_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1343_cast_fp16")]; + string _SplitHeadsQ__mh_w_1345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1345_equation_0, values = (var_13149_cast_fp16, var_12783_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1345_cast_fp16")]; + string _SplitHeadsQ__mh_w_1347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1347_equation_0, values = (var_13149_cast_fp16, var_12790_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1347_cast_fp16")]; + string _SplitHeadsQ__mh_w_1349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1349_equation_0, values = (var_13149_cast_fp16, var_12797_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1349_cast_fp16")]; + string _SplitHeadsQ__mh_w_1351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1351_equation_0, values = (var_13149_cast_fp16, var_12804_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1351_cast_fp16")]; + string _SplitHeadsQ__mh_w_1353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1353_equation_0, values = (var_13153_cast_fp16, var_12811_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1353_cast_fp16")]; + string _SplitHeadsQ__mh_w_1355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1355_equation_0, values = (var_13153_cast_fp16, var_12818_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1355_cast_fp16")]; + string _SplitHeadsQ__mh_w_1357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1357_equation_0, values = (var_13153_cast_fp16, var_12825_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1357_cast_fp16")]; + string _SplitHeadsQ__mh_w_1359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1359_equation_0, values = (var_13153_cast_fp16, var_12832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1359_cast_fp16")]; + string _SplitHeadsQ__mh_w_1361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1361_equation_0, values = (var_13157_cast_fp16, var_12839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1361_cast_fp16")]; + string _SplitHeadsQ__mh_w_1363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1363_equation_0, values = (var_13157_cast_fp16, var_12846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1363_cast_fp16")]; + string _SplitHeadsQ__mh_w_1365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1365_equation_0, values = (var_13157_cast_fp16, var_12853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1365_cast_fp16")]; + string _SplitHeadsQ__mh_w_1367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1367_equation_0, values = (var_13157_cast_fp16, var_12860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1367_cast_fp16")]; + string _SplitHeadsQ__mh_w_1369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1369_equation_0, values = (var_13161_cast_fp16, var_12867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1369_cast_fp16")]; + string _SplitHeadsQ__mh_w_1371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1371_equation_0, values = (var_13161_cast_fp16, var_12874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1371_cast_fp16")]; + string _SplitHeadsQ__mh_w_1373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1373_equation_0, values = (var_13161_cast_fp16, var_12881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1373_cast_fp16")]; + string _SplitHeadsQ__mh_w_1375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1375_equation_0, values = (var_13161_cast_fp16, var_12888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1375_cast_fp16")]; + string _SplitHeadsQ__mh_w_1377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1377_equation_0, values = (var_13165_cast_fp16, var_12895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1377_cast_fp16")]; + string _SplitHeadsQ__mh_w_1379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1379_equation_0, values = (var_13165_cast_fp16, var_12902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1379_cast_fp16")]; + string _SplitHeadsQ__mh_w_1381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1381_equation_0, values = (var_13165_cast_fp16, var_12909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1381_cast_fp16")]; + string _SplitHeadsQ__mh_w_1383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1383_equation_0, values = (var_13165_cast_fp16, var_12916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1383_cast_fp16")]; + string _SplitHeadsQ__mh_w_1385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1385_equation_0, values = (var_13169_cast_fp16, var_12923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1385_cast_fp16")]; + string _SplitHeadsQ__mh_w_1387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1387_equation_0, values = (var_13169_cast_fp16, var_12930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1387_cast_fp16")]; + string _SplitHeadsQ__mh_w_1389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1389_equation_0, values = (var_13169_cast_fp16, var_12937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1389_cast_fp16")]; + string _SplitHeadsQ__mh_w_1391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1391_equation_0, values = (var_13169_cast_fp16, var_12944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1391_cast_fp16")]; + string _SplitHeadsQ__mh_w_1393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1393_equation_0, values = (var_13173_cast_fp16, var_12951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1393_cast_fp16")]; + string _SplitHeadsQ__mh_w_1395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1395_equation_0, values = (var_13173_cast_fp16, var_12958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1395_cast_fp16")]; + string _SplitHeadsQ__mh_w_1397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1397_equation_0, values = (var_13173_cast_fp16, var_12965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1397_cast_fp16")]; + string _SplitHeadsQ__mh_w_1399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1399_equation_0, values = (var_13173_cast_fp16, var_12972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1399_cast_fp16")]; + string _SplitHeadsQ__mh_w_1401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1401_equation_0, values = (var_13177_cast_fp16, var_12979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1401_cast_fp16")]; + string _SplitHeadsQ__mh_w_1403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1403_equation_0, values = (var_13177_cast_fp16, var_12986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1403_cast_fp16")]; + string _SplitHeadsQ__mh_w_1405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1405_equation_0, values = (var_13177_cast_fp16, var_12993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1405_cast_fp16")]; + string _SplitHeadsQ__mh_w_1407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1407_equation_0, values = (var_13177_cast_fp16, var_13000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1407_cast_fp16")]; + string _SplitHeadsQ__mh_w_1409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1409_equation_0, values = (var_13181_cast_fp16, var_13007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1409_cast_fp16")]; + string _SplitHeadsQ__mh_w_1411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1411_equation_0, values = (var_13181_cast_fp16, var_13014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1411_cast_fp16")]; + string _SplitHeadsQ__mh_w_1413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1413_equation_0, values = (var_13181_cast_fp16, var_13021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1413_cast_fp16")]; + string _SplitHeadsQ__mh_w_1415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1415_equation_0, values = (var_13181_cast_fp16, var_13028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1415_cast_fp16")]; + string _SplitHeadsQ__mh_w_1417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1417_equation_0, values = (var_13185_cast_fp16, var_13035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1417_cast_fp16")]; + string _SplitHeadsQ__mh_w_1419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1419_equation_0, values = (var_13185_cast_fp16, var_13042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1419_cast_fp16")]; + string _SplitHeadsQ__mh_w_1421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1421_equation_0, values = (var_13185_cast_fp16, var_13049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1421_cast_fp16")]; + string _SplitHeadsQ__mh_w_1423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1423_equation_0, values = (var_13185_cast_fp16, var_13056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1423_cast_fp16")]; + string _SplitHeadsQ__mh_w_1425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1425_equation_0, values = (var_13189_cast_fp16, var_13063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1425_cast_fp16")]; + string _SplitHeadsQ__mh_w_1427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1427_equation_0, values = (var_13189_cast_fp16, var_13070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1427_cast_fp16")]; + string _SplitHeadsQ__mh_w_1429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1429_equation_0, values = (var_13189_cast_fp16, var_13077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1429_cast_fp16")]; + string _SplitHeadsQ__mh_w_1431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1431_equation_0, values = (var_13189_cast_fp16, var_13084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1431_cast_fp16")]; + string _SplitHeadsQ__mh_w_1433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1433_equation_0, values = (var_13193_cast_fp16, var_13091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1433_cast_fp16")]; + string _SplitHeadsQ__mh_w_1435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1435_equation_0, values = (var_13193_cast_fp16, var_13098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1435_cast_fp16")]; + string _SplitHeadsQ__mh_w_1437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1437_equation_0, values = (var_13193_cast_fp16, var_13105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1437_cast_fp16")]; + string _SplitHeadsQ__mh_w_1439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1439_equation_0, values = (var_13193_cast_fp16, var_13112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1439_cast_fp16")]; + fp16 var_13434_to_fp16 = const()[name = string("op_13434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1281_cast_fp16, y = var_13434_to_fp16)[name = string("aw_chunk_1281_cast_fp16")]; + fp16 var_13436_to_fp16 = const()[name = string("op_13436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1283_cast_fp16, y = var_13436_to_fp16)[name = string("aw_chunk_1283_cast_fp16")]; + fp16 var_13438_to_fp16 = const()[name = string("op_13438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1285_cast_fp16, y = var_13438_to_fp16)[name = string("aw_chunk_1285_cast_fp16")]; + fp16 var_13440_to_fp16 = const()[name = string("op_13440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1287_cast_fp16, y = var_13440_to_fp16)[name = string("aw_chunk_1287_cast_fp16")]; + fp16 var_13442_to_fp16 = const()[name = string("op_13442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1289_cast_fp16, y = var_13442_to_fp16)[name = string("aw_chunk_1289_cast_fp16")]; + fp16 var_13444_to_fp16 = const()[name = string("op_13444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1291_cast_fp16, y = var_13444_to_fp16)[name = string("aw_chunk_1291_cast_fp16")]; + fp16 var_13446_to_fp16 = const()[name = string("op_13446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1293_cast_fp16, y = var_13446_to_fp16)[name = string("aw_chunk_1293_cast_fp16")]; + fp16 var_13448_to_fp16 = const()[name = string("op_13448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1295_cast_fp16, y = var_13448_to_fp16)[name = string("aw_chunk_1295_cast_fp16")]; + fp16 var_13450_to_fp16 = const()[name = string("op_13450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1297_cast_fp16, y = var_13450_to_fp16)[name = string("aw_chunk_1297_cast_fp16")]; + fp16 var_13452_to_fp16 = const()[name = string("op_13452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1299_cast_fp16, y = var_13452_to_fp16)[name = string("aw_chunk_1299_cast_fp16")]; + fp16 var_13454_to_fp16 = const()[name = string("op_13454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1301_cast_fp16, y = var_13454_to_fp16)[name = string("aw_chunk_1301_cast_fp16")]; + fp16 var_13456_to_fp16 = const()[name = string("op_13456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1303_cast_fp16, y = var_13456_to_fp16)[name = string("aw_chunk_1303_cast_fp16")]; + fp16 var_13458_to_fp16 = const()[name = string("op_13458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1305_cast_fp16, y = var_13458_to_fp16)[name = string("aw_chunk_1305_cast_fp16")]; + fp16 var_13460_to_fp16 = const()[name = string("op_13460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1307_cast_fp16, y = var_13460_to_fp16)[name = string("aw_chunk_1307_cast_fp16")]; + fp16 var_13462_to_fp16 = const()[name = string("op_13462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1309_cast_fp16, y = var_13462_to_fp16)[name = string("aw_chunk_1309_cast_fp16")]; + fp16 var_13464_to_fp16 = const()[name = string("op_13464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1311_cast_fp16, y = var_13464_to_fp16)[name = string("aw_chunk_1311_cast_fp16")]; + fp16 var_13466_to_fp16 = const()[name = string("op_13466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1313_cast_fp16, y = var_13466_to_fp16)[name = string("aw_chunk_1313_cast_fp16")]; + fp16 var_13468_to_fp16 = const()[name = string("op_13468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1315_cast_fp16, y = var_13468_to_fp16)[name = string("aw_chunk_1315_cast_fp16")]; + fp16 var_13470_to_fp16 = const()[name = string("op_13470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1317_cast_fp16, y = var_13470_to_fp16)[name = string("aw_chunk_1317_cast_fp16")]; + fp16 var_13472_to_fp16 = const()[name = string("op_13472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1319_cast_fp16, y = var_13472_to_fp16)[name = string("aw_chunk_1319_cast_fp16")]; + fp16 var_13474_to_fp16 = const()[name = string("op_13474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1321_cast_fp16, y = var_13474_to_fp16)[name = string("aw_chunk_1321_cast_fp16")]; + fp16 var_13476_to_fp16 = const()[name = string("op_13476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1323_cast_fp16, y = var_13476_to_fp16)[name = string("aw_chunk_1323_cast_fp16")]; + fp16 var_13478_to_fp16 = const()[name = string("op_13478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1325_cast_fp16, y = var_13478_to_fp16)[name = string("aw_chunk_1325_cast_fp16")]; + fp16 var_13480_to_fp16 = const()[name = string("op_13480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1327_cast_fp16, y = var_13480_to_fp16)[name = string("aw_chunk_1327_cast_fp16")]; + fp16 var_13482_to_fp16 = const()[name = string("op_13482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1329_cast_fp16, y = var_13482_to_fp16)[name = string("aw_chunk_1329_cast_fp16")]; + fp16 var_13484_to_fp16 = const()[name = string("op_13484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1331_cast_fp16, y = var_13484_to_fp16)[name = string("aw_chunk_1331_cast_fp16")]; + fp16 var_13486_to_fp16 = const()[name = string("op_13486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1333_cast_fp16, y = var_13486_to_fp16)[name = string("aw_chunk_1333_cast_fp16")]; + fp16 var_13488_to_fp16 = const()[name = string("op_13488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1335_cast_fp16, y = var_13488_to_fp16)[name = string("aw_chunk_1335_cast_fp16")]; + fp16 var_13490_to_fp16 = const()[name = string("op_13490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1337_cast_fp16, y = var_13490_to_fp16)[name = string("aw_chunk_1337_cast_fp16")]; + fp16 var_13492_to_fp16 = const()[name = string("op_13492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1339_cast_fp16, y = var_13492_to_fp16)[name = string("aw_chunk_1339_cast_fp16")]; + fp16 var_13494_to_fp16 = const()[name = string("op_13494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1341_cast_fp16, y = var_13494_to_fp16)[name = string("aw_chunk_1341_cast_fp16")]; + fp16 var_13496_to_fp16 = const()[name = string("op_13496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1343_cast_fp16, y = var_13496_to_fp16)[name = string("aw_chunk_1343_cast_fp16")]; + fp16 var_13498_to_fp16 = const()[name = string("op_13498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1345_cast_fp16, y = var_13498_to_fp16)[name = string("aw_chunk_1345_cast_fp16")]; + fp16 var_13500_to_fp16 = const()[name = string("op_13500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1347_cast_fp16, y = var_13500_to_fp16)[name = string("aw_chunk_1347_cast_fp16")]; + fp16 var_13502_to_fp16 = const()[name = string("op_13502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1349_cast_fp16, y = var_13502_to_fp16)[name = string("aw_chunk_1349_cast_fp16")]; + fp16 var_13504_to_fp16 = const()[name = string("op_13504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1351_cast_fp16, y = var_13504_to_fp16)[name = string("aw_chunk_1351_cast_fp16")]; + fp16 var_13506_to_fp16 = const()[name = string("op_13506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1353_cast_fp16, y = var_13506_to_fp16)[name = string("aw_chunk_1353_cast_fp16")]; + fp16 var_13508_to_fp16 = const()[name = string("op_13508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1355_cast_fp16, y = var_13508_to_fp16)[name = string("aw_chunk_1355_cast_fp16")]; + fp16 var_13510_to_fp16 = const()[name = string("op_13510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1357_cast_fp16, y = var_13510_to_fp16)[name = string("aw_chunk_1357_cast_fp16")]; + fp16 var_13512_to_fp16 = const()[name = string("op_13512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1359_cast_fp16, y = var_13512_to_fp16)[name = string("aw_chunk_1359_cast_fp16")]; + fp16 var_13514_to_fp16 = const()[name = string("op_13514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1361_cast_fp16, y = var_13514_to_fp16)[name = string("aw_chunk_1361_cast_fp16")]; + fp16 var_13516_to_fp16 = const()[name = string("op_13516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1363_cast_fp16, y = var_13516_to_fp16)[name = string("aw_chunk_1363_cast_fp16")]; + fp16 var_13518_to_fp16 = const()[name = string("op_13518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1365_cast_fp16, y = var_13518_to_fp16)[name = string("aw_chunk_1365_cast_fp16")]; + fp16 var_13520_to_fp16 = const()[name = string("op_13520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1367_cast_fp16, y = var_13520_to_fp16)[name = string("aw_chunk_1367_cast_fp16")]; + fp16 var_13522_to_fp16 = const()[name = string("op_13522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1369_cast_fp16, y = var_13522_to_fp16)[name = string("aw_chunk_1369_cast_fp16")]; + fp16 var_13524_to_fp16 = const()[name = string("op_13524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1371_cast_fp16, y = var_13524_to_fp16)[name = string("aw_chunk_1371_cast_fp16")]; + fp16 var_13526_to_fp16 = const()[name = string("op_13526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1373_cast_fp16, y = var_13526_to_fp16)[name = string("aw_chunk_1373_cast_fp16")]; + fp16 var_13528_to_fp16 = const()[name = string("op_13528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1375_cast_fp16, y = var_13528_to_fp16)[name = string("aw_chunk_1375_cast_fp16")]; + fp16 var_13530_to_fp16 = const()[name = string("op_13530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1377_cast_fp16, y = var_13530_to_fp16)[name = string("aw_chunk_1377_cast_fp16")]; + fp16 var_13532_to_fp16 = const()[name = string("op_13532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1379_cast_fp16, y = var_13532_to_fp16)[name = string("aw_chunk_1379_cast_fp16")]; + fp16 var_13534_to_fp16 = const()[name = string("op_13534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1381_cast_fp16, y = var_13534_to_fp16)[name = string("aw_chunk_1381_cast_fp16")]; + fp16 var_13536_to_fp16 = const()[name = string("op_13536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1383_cast_fp16, y = var_13536_to_fp16)[name = string("aw_chunk_1383_cast_fp16")]; + fp16 var_13538_to_fp16 = const()[name = string("op_13538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1385_cast_fp16, y = var_13538_to_fp16)[name = string("aw_chunk_1385_cast_fp16")]; + fp16 var_13540_to_fp16 = const()[name = string("op_13540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1387_cast_fp16, y = var_13540_to_fp16)[name = string("aw_chunk_1387_cast_fp16")]; + fp16 var_13542_to_fp16 = const()[name = string("op_13542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1389_cast_fp16, y = var_13542_to_fp16)[name = string("aw_chunk_1389_cast_fp16")]; + fp16 var_13544_to_fp16 = const()[name = string("op_13544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1391_cast_fp16, y = var_13544_to_fp16)[name = string("aw_chunk_1391_cast_fp16")]; + fp16 var_13546_to_fp16 = const()[name = string("op_13546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1393_cast_fp16, y = var_13546_to_fp16)[name = string("aw_chunk_1393_cast_fp16")]; + fp16 var_13548_to_fp16 = const()[name = string("op_13548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1395_cast_fp16, y = var_13548_to_fp16)[name = string("aw_chunk_1395_cast_fp16")]; + fp16 var_13550_to_fp16 = const()[name = string("op_13550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1397_cast_fp16, y = var_13550_to_fp16)[name = string("aw_chunk_1397_cast_fp16")]; + fp16 var_13552_to_fp16 = const()[name = string("op_13552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1399_cast_fp16, y = var_13552_to_fp16)[name = string("aw_chunk_1399_cast_fp16")]; + fp16 var_13554_to_fp16 = const()[name = string("op_13554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1401_cast_fp16, y = var_13554_to_fp16)[name = string("aw_chunk_1401_cast_fp16")]; + fp16 var_13556_to_fp16 = const()[name = string("op_13556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1403_cast_fp16, y = var_13556_to_fp16)[name = string("aw_chunk_1403_cast_fp16")]; + fp16 var_13558_to_fp16 = const()[name = string("op_13558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1405_cast_fp16, y = var_13558_to_fp16)[name = string("aw_chunk_1405_cast_fp16")]; + fp16 var_13560_to_fp16 = const()[name = string("op_13560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1407_cast_fp16, y = var_13560_to_fp16)[name = string("aw_chunk_1407_cast_fp16")]; + fp16 var_13562_to_fp16 = const()[name = string("op_13562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1409_cast_fp16, y = var_13562_to_fp16)[name = string("aw_chunk_1409_cast_fp16")]; + fp16 var_13564_to_fp16 = const()[name = string("op_13564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1411_cast_fp16, y = var_13564_to_fp16)[name = string("aw_chunk_1411_cast_fp16")]; + fp16 var_13566_to_fp16 = const()[name = string("op_13566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1413_cast_fp16, y = var_13566_to_fp16)[name = string("aw_chunk_1413_cast_fp16")]; + fp16 var_13568_to_fp16 = const()[name = string("op_13568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1415_cast_fp16, y = var_13568_to_fp16)[name = string("aw_chunk_1415_cast_fp16")]; + fp16 var_13570_to_fp16 = const()[name = string("op_13570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1417_cast_fp16, y = var_13570_to_fp16)[name = string("aw_chunk_1417_cast_fp16")]; + fp16 var_13572_to_fp16 = const()[name = string("op_13572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1419_cast_fp16, y = var_13572_to_fp16)[name = string("aw_chunk_1419_cast_fp16")]; + fp16 var_13574_to_fp16 = const()[name = string("op_13574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1421_cast_fp16, y = var_13574_to_fp16)[name = string("aw_chunk_1421_cast_fp16")]; + fp16 var_13576_to_fp16 = const()[name = string("op_13576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1423_cast_fp16, y = var_13576_to_fp16)[name = string("aw_chunk_1423_cast_fp16")]; + fp16 var_13578_to_fp16 = const()[name = string("op_13578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1425_cast_fp16, y = var_13578_to_fp16)[name = string("aw_chunk_1425_cast_fp16")]; + fp16 var_13580_to_fp16 = const()[name = string("op_13580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1427_cast_fp16, y = var_13580_to_fp16)[name = string("aw_chunk_1427_cast_fp16")]; + fp16 var_13582_to_fp16 = const()[name = string("op_13582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1429_cast_fp16, y = var_13582_to_fp16)[name = string("aw_chunk_1429_cast_fp16")]; + fp16 var_13584_to_fp16 = const()[name = string("op_13584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1431_cast_fp16, y = var_13584_to_fp16)[name = string("aw_chunk_1431_cast_fp16")]; + fp16 var_13586_to_fp16 = const()[name = string("op_13586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1433_cast_fp16, y = var_13586_to_fp16)[name = string("aw_chunk_1433_cast_fp16")]; + fp16 var_13588_to_fp16 = const()[name = string("op_13588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1435_cast_fp16, y = var_13588_to_fp16)[name = string("aw_chunk_1435_cast_fp16")]; + fp16 var_13590_to_fp16 = const()[name = string("op_13590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1437_cast_fp16, y = var_13590_to_fp16)[name = string("aw_chunk_1437_cast_fp16")]; + fp16 var_13592_to_fp16 = const()[name = string("op_13592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1439_cast_fp16, y = var_13592_to_fp16)[name = string("aw_chunk_1439_cast_fp16")]; + tensor var_13594_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1281_cast_fp16)[name = string("op_13594_cast_fp16")]; + tensor var_13595_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1283_cast_fp16)[name = string("op_13595_cast_fp16")]; + tensor var_13596_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1285_cast_fp16)[name = string("op_13596_cast_fp16")]; + tensor var_13597_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1287_cast_fp16)[name = string("op_13597_cast_fp16")]; + tensor var_13598_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1289_cast_fp16)[name = string("op_13598_cast_fp16")]; + tensor var_13599_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1291_cast_fp16)[name = string("op_13599_cast_fp16")]; + tensor var_13600_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1293_cast_fp16)[name = string("op_13600_cast_fp16")]; + tensor var_13601_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1295_cast_fp16)[name = string("op_13601_cast_fp16")]; + tensor var_13602_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1297_cast_fp16)[name = string("op_13602_cast_fp16")]; + tensor var_13603_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1299_cast_fp16)[name = string("op_13603_cast_fp16")]; + tensor var_13604_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1301_cast_fp16)[name = string("op_13604_cast_fp16")]; + tensor var_13605_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1303_cast_fp16)[name = string("op_13605_cast_fp16")]; + tensor var_13606_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1305_cast_fp16)[name = string("op_13606_cast_fp16")]; + tensor var_13607_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1307_cast_fp16)[name = string("op_13607_cast_fp16")]; + tensor var_13608_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1309_cast_fp16)[name = string("op_13608_cast_fp16")]; + tensor var_13609_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1311_cast_fp16)[name = string("op_13609_cast_fp16")]; + tensor var_13610_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1313_cast_fp16)[name = string("op_13610_cast_fp16")]; + tensor var_13611_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1315_cast_fp16)[name = string("op_13611_cast_fp16")]; + tensor var_13612_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1317_cast_fp16)[name = string("op_13612_cast_fp16")]; + tensor var_13613_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1319_cast_fp16)[name = string("op_13613_cast_fp16")]; + tensor var_13614_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1321_cast_fp16)[name = string("op_13614_cast_fp16")]; + tensor var_13615_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1323_cast_fp16)[name = string("op_13615_cast_fp16")]; + tensor var_13616_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1325_cast_fp16)[name = string("op_13616_cast_fp16")]; + tensor var_13617_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1327_cast_fp16)[name = string("op_13617_cast_fp16")]; + tensor var_13618_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1329_cast_fp16)[name = string("op_13618_cast_fp16")]; + tensor var_13619_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1331_cast_fp16)[name = string("op_13619_cast_fp16")]; + tensor var_13620_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1333_cast_fp16)[name = string("op_13620_cast_fp16")]; + tensor var_13621_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1335_cast_fp16)[name = string("op_13621_cast_fp16")]; + tensor var_13622_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1337_cast_fp16)[name = string("op_13622_cast_fp16")]; + tensor var_13623_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1339_cast_fp16)[name = string("op_13623_cast_fp16")]; + tensor var_13624_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1341_cast_fp16)[name = string("op_13624_cast_fp16")]; + tensor var_13625_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1343_cast_fp16)[name = string("op_13625_cast_fp16")]; + tensor var_13626_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1345_cast_fp16)[name = string("op_13626_cast_fp16")]; + tensor var_13627_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1347_cast_fp16)[name = string("op_13627_cast_fp16")]; + tensor var_13628_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1349_cast_fp16)[name = string("op_13628_cast_fp16")]; + tensor var_13629_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1351_cast_fp16)[name = string("op_13629_cast_fp16")]; + tensor var_13630_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1353_cast_fp16)[name = string("op_13630_cast_fp16")]; + tensor var_13631_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1355_cast_fp16)[name = string("op_13631_cast_fp16")]; + tensor var_13632_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1357_cast_fp16)[name = string("op_13632_cast_fp16")]; + tensor var_13633_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1359_cast_fp16)[name = string("op_13633_cast_fp16")]; + tensor var_13634_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1361_cast_fp16)[name = string("op_13634_cast_fp16")]; + tensor var_13635_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1363_cast_fp16)[name = string("op_13635_cast_fp16")]; + tensor var_13636_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1365_cast_fp16)[name = string("op_13636_cast_fp16")]; + tensor var_13637_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1367_cast_fp16)[name = string("op_13637_cast_fp16")]; + tensor var_13638_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1369_cast_fp16)[name = string("op_13638_cast_fp16")]; + tensor var_13639_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1371_cast_fp16)[name = string("op_13639_cast_fp16")]; + tensor var_13640_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1373_cast_fp16)[name = string("op_13640_cast_fp16")]; + tensor var_13641_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1375_cast_fp16)[name = string("op_13641_cast_fp16")]; + tensor var_13642_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1377_cast_fp16)[name = string("op_13642_cast_fp16")]; + tensor var_13643_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1379_cast_fp16)[name = string("op_13643_cast_fp16")]; + tensor var_13644_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1381_cast_fp16)[name = string("op_13644_cast_fp16")]; + tensor var_13645_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1383_cast_fp16)[name = string("op_13645_cast_fp16")]; + tensor var_13646_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1385_cast_fp16)[name = string("op_13646_cast_fp16")]; + tensor var_13647_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1387_cast_fp16)[name = string("op_13647_cast_fp16")]; + tensor var_13648_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1389_cast_fp16)[name = string("op_13648_cast_fp16")]; + tensor var_13649_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1391_cast_fp16)[name = string("op_13649_cast_fp16")]; + tensor var_13650_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1393_cast_fp16)[name = string("op_13650_cast_fp16")]; + tensor var_13651_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1395_cast_fp16)[name = string("op_13651_cast_fp16")]; + tensor var_13652_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1397_cast_fp16)[name = string("op_13652_cast_fp16")]; + tensor var_13653_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1399_cast_fp16)[name = string("op_13653_cast_fp16")]; + tensor var_13654_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1401_cast_fp16)[name = string("op_13654_cast_fp16")]; + tensor var_13655_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1403_cast_fp16)[name = string("op_13655_cast_fp16")]; + tensor var_13656_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1405_cast_fp16)[name = string("op_13656_cast_fp16")]; + tensor var_13657_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1407_cast_fp16)[name = string("op_13657_cast_fp16")]; + tensor var_13658_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1409_cast_fp16)[name = string("op_13658_cast_fp16")]; + tensor var_13659_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1411_cast_fp16)[name = string("op_13659_cast_fp16")]; + tensor var_13660_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1413_cast_fp16)[name = string("op_13660_cast_fp16")]; + tensor var_13661_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1415_cast_fp16)[name = string("op_13661_cast_fp16")]; + tensor var_13662_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1417_cast_fp16)[name = string("op_13662_cast_fp16")]; + tensor var_13663_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1419_cast_fp16)[name = string("op_13663_cast_fp16")]; + tensor var_13664_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1421_cast_fp16)[name = string("op_13664_cast_fp16")]; + tensor var_13665_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1423_cast_fp16)[name = string("op_13665_cast_fp16")]; + tensor var_13666_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1425_cast_fp16)[name = string("op_13666_cast_fp16")]; + tensor var_13667_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1427_cast_fp16)[name = string("op_13667_cast_fp16")]; + tensor var_13668_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1429_cast_fp16)[name = string("op_13668_cast_fp16")]; + tensor var_13669_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1431_cast_fp16)[name = string("op_13669_cast_fp16")]; + tensor var_13670_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1433_cast_fp16)[name = string("op_13670_cast_fp16")]; + tensor var_13671_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1435_cast_fp16)[name = string("op_13671_cast_fp16")]; + tensor var_13672_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1437_cast_fp16)[name = string("op_13672_cast_fp16")]; + tensor var_13673_cast_fp16 = softmax(axis = var_12419, x = aw_chunk_1439_cast_fp16)[name = string("op_13673_cast_fp16")]; + string var_13675_equation_0 = const()[name = string("op_13675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13675_cast_fp16 = einsum(equation = var_13675_equation_0, values = (var_13195_cast_fp16, var_13594_cast_fp16))[name = string("op_13675_cast_fp16")]; + string var_13677_equation_0 = const()[name = string("op_13677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13677_cast_fp16 = einsum(equation = var_13677_equation_0, values = (var_13195_cast_fp16, var_13595_cast_fp16))[name = string("op_13677_cast_fp16")]; + string var_13679_equation_0 = const()[name = string("op_13679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13679_cast_fp16 = einsum(equation = var_13679_equation_0, values = (var_13195_cast_fp16, var_13596_cast_fp16))[name = string("op_13679_cast_fp16")]; + string var_13681_equation_0 = const()[name = string("op_13681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13681_cast_fp16 = einsum(equation = var_13681_equation_0, values = (var_13195_cast_fp16, var_13597_cast_fp16))[name = string("op_13681_cast_fp16")]; + string var_13683_equation_0 = const()[name = string("op_13683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13683_cast_fp16 = einsum(equation = var_13683_equation_0, values = (var_13199_cast_fp16, var_13598_cast_fp16))[name = string("op_13683_cast_fp16")]; + string var_13685_equation_0 = const()[name = string("op_13685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13685_cast_fp16 = einsum(equation = var_13685_equation_0, values = (var_13199_cast_fp16, var_13599_cast_fp16))[name = string("op_13685_cast_fp16")]; + string var_13687_equation_0 = const()[name = string("op_13687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13687_cast_fp16 = einsum(equation = var_13687_equation_0, values = (var_13199_cast_fp16, var_13600_cast_fp16))[name = string("op_13687_cast_fp16")]; + string var_13689_equation_0 = const()[name = string("op_13689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13689_cast_fp16 = einsum(equation = var_13689_equation_0, values = (var_13199_cast_fp16, var_13601_cast_fp16))[name = string("op_13689_cast_fp16")]; + string var_13691_equation_0 = const()[name = string("op_13691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13691_cast_fp16 = einsum(equation = var_13691_equation_0, values = (var_13203_cast_fp16, var_13602_cast_fp16))[name = string("op_13691_cast_fp16")]; + string var_13693_equation_0 = const()[name = string("op_13693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13693_cast_fp16 = einsum(equation = var_13693_equation_0, values = (var_13203_cast_fp16, var_13603_cast_fp16))[name = string("op_13693_cast_fp16")]; + string var_13695_equation_0 = const()[name = string("op_13695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13695_cast_fp16 = einsum(equation = var_13695_equation_0, values = (var_13203_cast_fp16, var_13604_cast_fp16))[name = string("op_13695_cast_fp16")]; + string var_13697_equation_0 = const()[name = string("op_13697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13697_cast_fp16 = einsum(equation = var_13697_equation_0, values = (var_13203_cast_fp16, var_13605_cast_fp16))[name = string("op_13697_cast_fp16")]; + string var_13699_equation_0 = const()[name = string("op_13699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13699_cast_fp16 = einsum(equation = var_13699_equation_0, values = (var_13207_cast_fp16, var_13606_cast_fp16))[name = string("op_13699_cast_fp16")]; + string var_13701_equation_0 = const()[name = string("op_13701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13701_cast_fp16 = einsum(equation = var_13701_equation_0, values = (var_13207_cast_fp16, var_13607_cast_fp16))[name = string("op_13701_cast_fp16")]; + string var_13703_equation_0 = const()[name = string("op_13703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13703_cast_fp16 = einsum(equation = var_13703_equation_0, values = (var_13207_cast_fp16, var_13608_cast_fp16))[name = string("op_13703_cast_fp16")]; + string var_13705_equation_0 = const()[name = string("op_13705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13705_cast_fp16 = einsum(equation = var_13705_equation_0, values = (var_13207_cast_fp16, var_13609_cast_fp16))[name = string("op_13705_cast_fp16")]; + string var_13707_equation_0 = const()[name = string("op_13707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13707_cast_fp16 = einsum(equation = var_13707_equation_0, values = (var_13211_cast_fp16, var_13610_cast_fp16))[name = string("op_13707_cast_fp16")]; + string var_13709_equation_0 = const()[name = string("op_13709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13709_cast_fp16 = einsum(equation = var_13709_equation_0, values = (var_13211_cast_fp16, var_13611_cast_fp16))[name = string("op_13709_cast_fp16")]; + string var_13711_equation_0 = const()[name = string("op_13711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13711_cast_fp16 = einsum(equation = var_13711_equation_0, values = (var_13211_cast_fp16, var_13612_cast_fp16))[name = string("op_13711_cast_fp16")]; + string var_13713_equation_0 = const()[name = string("op_13713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13713_cast_fp16 = einsum(equation = var_13713_equation_0, values = (var_13211_cast_fp16, var_13613_cast_fp16))[name = string("op_13713_cast_fp16")]; + string var_13715_equation_0 = const()[name = string("op_13715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13715_cast_fp16 = einsum(equation = var_13715_equation_0, values = (var_13215_cast_fp16, var_13614_cast_fp16))[name = string("op_13715_cast_fp16")]; + string var_13717_equation_0 = const()[name = string("op_13717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13717_cast_fp16 = einsum(equation = var_13717_equation_0, values = (var_13215_cast_fp16, var_13615_cast_fp16))[name = string("op_13717_cast_fp16")]; + string var_13719_equation_0 = const()[name = string("op_13719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13719_cast_fp16 = einsum(equation = var_13719_equation_0, values = (var_13215_cast_fp16, var_13616_cast_fp16))[name = string("op_13719_cast_fp16")]; + string var_13721_equation_0 = const()[name = string("op_13721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13721_cast_fp16 = einsum(equation = var_13721_equation_0, values = (var_13215_cast_fp16, var_13617_cast_fp16))[name = string("op_13721_cast_fp16")]; + string var_13723_equation_0 = const()[name = string("op_13723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13723_cast_fp16 = einsum(equation = var_13723_equation_0, values = (var_13219_cast_fp16, var_13618_cast_fp16))[name = string("op_13723_cast_fp16")]; + string var_13725_equation_0 = const()[name = string("op_13725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13725_cast_fp16 = einsum(equation = var_13725_equation_0, values = (var_13219_cast_fp16, var_13619_cast_fp16))[name = string("op_13725_cast_fp16")]; + string var_13727_equation_0 = const()[name = string("op_13727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13727_cast_fp16 = einsum(equation = var_13727_equation_0, values = (var_13219_cast_fp16, var_13620_cast_fp16))[name = string("op_13727_cast_fp16")]; + string var_13729_equation_0 = const()[name = string("op_13729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13729_cast_fp16 = einsum(equation = var_13729_equation_0, values = (var_13219_cast_fp16, var_13621_cast_fp16))[name = string("op_13729_cast_fp16")]; + string var_13731_equation_0 = const()[name = string("op_13731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13731_cast_fp16 = einsum(equation = var_13731_equation_0, values = (var_13223_cast_fp16, var_13622_cast_fp16))[name = string("op_13731_cast_fp16")]; + string var_13733_equation_0 = const()[name = string("op_13733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13733_cast_fp16 = einsum(equation = var_13733_equation_0, values = (var_13223_cast_fp16, var_13623_cast_fp16))[name = string("op_13733_cast_fp16")]; + string var_13735_equation_0 = const()[name = string("op_13735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13735_cast_fp16 = einsum(equation = var_13735_equation_0, values = (var_13223_cast_fp16, var_13624_cast_fp16))[name = string("op_13735_cast_fp16")]; + string var_13737_equation_0 = const()[name = string("op_13737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13737_cast_fp16 = einsum(equation = var_13737_equation_0, values = (var_13223_cast_fp16, var_13625_cast_fp16))[name = string("op_13737_cast_fp16")]; + string var_13739_equation_0 = const()[name = string("op_13739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13739_cast_fp16 = einsum(equation = var_13739_equation_0, values = (var_13227_cast_fp16, var_13626_cast_fp16))[name = string("op_13739_cast_fp16")]; + string var_13741_equation_0 = const()[name = string("op_13741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13741_cast_fp16 = einsum(equation = var_13741_equation_0, values = (var_13227_cast_fp16, var_13627_cast_fp16))[name = string("op_13741_cast_fp16")]; + string var_13743_equation_0 = const()[name = string("op_13743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13743_cast_fp16 = einsum(equation = var_13743_equation_0, values = (var_13227_cast_fp16, var_13628_cast_fp16))[name = string("op_13743_cast_fp16")]; + string var_13745_equation_0 = const()[name = string("op_13745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13745_cast_fp16 = einsum(equation = var_13745_equation_0, values = (var_13227_cast_fp16, var_13629_cast_fp16))[name = string("op_13745_cast_fp16")]; + string var_13747_equation_0 = const()[name = string("op_13747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13747_cast_fp16 = einsum(equation = var_13747_equation_0, values = (var_13231_cast_fp16, var_13630_cast_fp16))[name = string("op_13747_cast_fp16")]; + string var_13749_equation_0 = const()[name = string("op_13749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13749_cast_fp16 = einsum(equation = var_13749_equation_0, values = (var_13231_cast_fp16, var_13631_cast_fp16))[name = string("op_13749_cast_fp16")]; + string var_13751_equation_0 = const()[name = string("op_13751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13751_cast_fp16 = einsum(equation = var_13751_equation_0, values = (var_13231_cast_fp16, var_13632_cast_fp16))[name = string("op_13751_cast_fp16")]; + string var_13753_equation_0 = const()[name = string("op_13753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13753_cast_fp16 = einsum(equation = var_13753_equation_0, values = (var_13231_cast_fp16, var_13633_cast_fp16))[name = string("op_13753_cast_fp16")]; + string var_13755_equation_0 = const()[name = string("op_13755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13755_cast_fp16 = einsum(equation = var_13755_equation_0, values = (var_13235_cast_fp16, var_13634_cast_fp16))[name = string("op_13755_cast_fp16")]; + string var_13757_equation_0 = const()[name = string("op_13757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13757_cast_fp16 = einsum(equation = var_13757_equation_0, values = (var_13235_cast_fp16, var_13635_cast_fp16))[name = string("op_13757_cast_fp16")]; + string var_13759_equation_0 = const()[name = string("op_13759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13759_cast_fp16 = einsum(equation = var_13759_equation_0, values = (var_13235_cast_fp16, var_13636_cast_fp16))[name = string("op_13759_cast_fp16")]; + string var_13761_equation_0 = const()[name = string("op_13761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13761_cast_fp16 = einsum(equation = var_13761_equation_0, values = (var_13235_cast_fp16, var_13637_cast_fp16))[name = string("op_13761_cast_fp16")]; + string var_13763_equation_0 = const()[name = string("op_13763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13763_cast_fp16 = einsum(equation = var_13763_equation_0, values = (var_13239_cast_fp16, var_13638_cast_fp16))[name = string("op_13763_cast_fp16")]; + string var_13765_equation_0 = const()[name = string("op_13765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13765_cast_fp16 = einsum(equation = var_13765_equation_0, values = (var_13239_cast_fp16, var_13639_cast_fp16))[name = string("op_13765_cast_fp16")]; + string var_13767_equation_0 = const()[name = string("op_13767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13767_cast_fp16 = einsum(equation = var_13767_equation_0, values = (var_13239_cast_fp16, var_13640_cast_fp16))[name = string("op_13767_cast_fp16")]; + string var_13769_equation_0 = const()[name = string("op_13769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13769_cast_fp16 = einsum(equation = var_13769_equation_0, values = (var_13239_cast_fp16, var_13641_cast_fp16))[name = string("op_13769_cast_fp16")]; + string var_13771_equation_0 = const()[name = string("op_13771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13771_cast_fp16 = einsum(equation = var_13771_equation_0, values = (var_13243_cast_fp16, var_13642_cast_fp16))[name = string("op_13771_cast_fp16")]; + string var_13773_equation_0 = const()[name = string("op_13773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13773_cast_fp16 = einsum(equation = var_13773_equation_0, values = (var_13243_cast_fp16, var_13643_cast_fp16))[name = string("op_13773_cast_fp16")]; + string var_13775_equation_0 = const()[name = string("op_13775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13775_cast_fp16 = einsum(equation = var_13775_equation_0, values = (var_13243_cast_fp16, var_13644_cast_fp16))[name = string("op_13775_cast_fp16")]; + string var_13777_equation_0 = const()[name = string("op_13777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13777_cast_fp16 = einsum(equation = var_13777_equation_0, values = (var_13243_cast_fp16, var_13645_cast_fp16))[name = string("op_13777_cast_fp16")]; + string var_13779_equation_0 = const()[name = string("op_13779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13779_cast_fp16 = einsum(equation = var_13779_equation_0, values = (var_13247_cast_fp16, var_13646_cast_fp16))[name = string("op_13779_cast_fp16")]; + string var_13781_equation_0 = const()[name = string("op_13781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13781_cast_fp16 = einsum(equation = var_13781_equation_0, values = (var_13247_cast_fp16, var_13647_cast_fp16))[name = string("op_13781_cast_fp16")]; + string var_13783_equation_0 = const()[name = string("op_13783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13783_cast_fp16 = einsum(equation = var_13783_equation_0, values = (var_13247_cast_fp16, var_13648_cast_fp16))[name = string("op_13783_cast_fp16")]; + string var_13785_equation_0 = const()[name = string("op_13785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13785_cast_fp16 = einsum(equation = var_13785_equation_0, values = (var_13247_cast_fp16, var_13649_cast_fp16))[name = string("op_13785_cast_fp16")]; + string var_13787_equation_0 = const()[name = string("op_13787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13787_cast_fp16 = einsum(equation = var_13787_equation_0, values = (var_13251_cast_fp16, var_13650_cast_fp16))[name = string("op_13787_cast_fp16")]; + string var_13789_equation_0 = const()[name = string("op_13789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13789_cast_fp16 = einsum(equation = var_13789_equation_0, values = (var_13251_cast_fp16, var_13651_cast_fp16))[name = string("op_13789_cast_fp16")]; + string var_13791_equation_0 = const()[name = string("op_13791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13791_cast_fp16 = einsum(equation = var_13791_equation_0, values = (var_13251_cast_fp16, var_13652_cast_fp16))[name = string("op_13791_cast_fp16")]; + string var_13793_equation_0 = const()[name = string("op_13793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13793_cast_fp16 = einsum(equation = var_13793_equation_0, values = (var_13251_cast_fp16, var_13653_cast_fp16))[name = string("op_13793_cast_fp16")]; + string var_13795_equation_0 = const()[name = string("op_13795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13795_cast_fp16 = einsum(equation = var_13795_equation_0, values = (var_13255_cast_fp16, var_13654_cast_fp16))[name = string("op_13795_cast_fp16")]; + string var_13797_equation_0 = const()[name = string("op_13797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13797_cast_fp16 = einsum(equation = var_13797_equation_0, values = (var_13255_cast_fp16, var_13655_cast_fp16))[name = string("op_13797_cast_fp16")]; + string var_13799_equation_0 = const()[name = string("op_13799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13799_cast_fp16 = einsum(equation = var_13799_equation_0, values = (var_13255_cast_fp16, var_13656_cast_fp16))[name = string("op_13799_cast_fp16")]; + string var_13801_equation_0 = const()[name = string("op_13801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13801_cast_fp16 = einsum(equation = var_13801_equation_0, values = (var_13255_cast_fp16, var_13657_cast_fp16))[name = string("op_13801_cast_fp16")]; + string var_13803_equation_0 = const()[name = string("op_13803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13803_cast_fp16 = einsum(equation = var_13803_equation_0, values = (var_13259_cast_fp16, var_13658_cast_fp16))[name = string("op_13803_cast_fp16")]; + string var_13805_equation_0 = const()[name = string("op_13805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13805_cast_fp16 = einsum(equation = var_13805_equation_0, values = (var_13259_cast_fp16, var_13659_cast_fp16))[name = string("op_13805_cast_fp16")]; + string var_13807_equation_0 = const()[name = string("op_13807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13807_cast_fp16 = einsum(equation = var_13807_equation_0, values = (var_13259_cast_fp16, var_13660_cast_fp16))[name = string("op_13807_cast_fp16")]; + string var_13809_equation_0 = const()[name = string("op_13809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13809_cast_fp16 = einsum(equation = var_13809_equation_0, values = (var_13259_cast_fp16, var_13661_cast_fp16))[name = string("op_13809_cast_fp16")]; + string var_13811_equation_0 = const()[name = string("op_13811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13811_cast_fp16 = einsum(equation = var_13811_equation_0, values = (var_13263_cast_fp16, var_13662_cast_fp16))[name = string("op_13811_cast_fp16")]; + string var_13813_equation_0 = const()[name = string("op_13813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13813_cast_fp16 = einsum(equation = var_13813_equation_0, values = (var_13263_cast_fp16, var_13663_cast_fp16))[name = string("op_13813_cast_fp16")]; + string var_13815_equation_0 = const()[name = string("op_13815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13815_cast_fp16 = einsum(equation = var_13815_equation_0, values = (var_13263_cast_fp16, var_13664_cast_fp16))[name = string("op_13815_cast_fp16")]; + string var_13817_equation_0 = const()[name = string("op_13817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13817_cast_fp16 = einsum(equation = var_13817_equation_0, values = (var_13263_cast_fp16, var_13665_cast_fp16))[name = string("op_13817_cast_fp16")]; + string var_13819_equation_0 = const()[name = string("op_13819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13819_cast_fp16 = einsum(equation = var_13819_equation_0, values = (var_13267_cast_fp16, var_13666_cast_fp16))[name = string("op_13819_cast_fp16")]; + string var_13821_equation_0 = const()[name = string("op_13821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13821_cast_fp16 = einsum(equation = var_13821_equation_0, values = (var_13267_cast_fp16, var_13667_cast_fp16))[name = string("op_13821_cast_fp16")]; + string var_13823_equation_0 = const()[name = string("op_13823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13823_cast_fp16 = einsum(equation = var_13823_equation_0, values = (var_13267_cast_fp16, var_13668_cast_fp16))[name = string("op_13823_cast_fp16")]; + string var_13825_equation_0 = const()[name = string("op_13825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13825_cast_fp16 = einsum(equation = var_13825_equation_0, values = (var_13267_cast_fp16, var_13669_cast_fp16))[name = string("op_13825_cast_fp16")]; + string var_13827_equation_0 = const()[name = string("op_13827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13827_cast_fp16 = einsum(equation = var_13827_equation_0, values = (var_13271_cast_fp16, var_13670_cast_fp16))[name = string("op_13827_cast_fp16")]; + string var_13829_equation_0 = const()[name = string("op_13829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13829_cast_fp16 = einsum(equation = var_13829_equation_0, values = (var_13271_cast_fp16, var_13671_cast_fp16))[name = string("op_13829_cast_fp16")]; + string var_13831_equation_0 = const()[name = string("op_13831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13831_cast_fp16 = einsum(equation = var_13831_equation_0, values = (var_13271_cast_fp16, var_13672_cast_fp16))[name = string("op_13831_cast_fp16")]; + string var_13833_equation_0 = const()[name = string("op_13833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13833_cast_fp16 = einsum(equation = var_13833_equation_0, values = (var_13271_cast_fp16, var_13673_cast_fp16))[name = string("op_13833_cast_fp16")]; + bool var_13835_interleave_0 = const()[name = string("op_13835_interleave_0"), val = bool(false)]; + tensor var_13835_cast_fp16 = concat(axis = var_12394, interleave = var_13835_interleave_0, values = (var_13675_cast_fp16, var_13677_cast_fp16, var_13679_cast_fp16, var_13681_cast_fp16))[name = string("op_13835_cast_fp16")]; + bool var_13837_interleave_0 = const()[name = string("op_13837_interleave_0"), val = bool(false)]; + tensor var_13837_cast_fp16 = concat(axis = var_12394, interleave = var_13837_interleave_0, values = (var_13683_cast_fp16, var_13685_cast_fp16, var_13687_cast_fp16, var_13689_cast_fp16))[name = string("op_13837_cast_fp16")]; + bool var_13839_interleave_0 = const()[name = string("op_13839_interleave_0"), val = bool(false)]; + tensor var_13839_cast_fp16 = concat(axis = var_12394, interleave = var_13839_interleave_0, values = (var_13691_cast_fp16, var_13693_cast_fp16, var_13695_cast_fp16, var_13697_cast_fp16))[name = string("op_13839_cast_fp16")]; + bool var_13841_interleave_0 = const()[name = string("op_13841_interleave_0"), val = bool(false)]; + tensor var_13841_cast_fp16 = concat(axis = var_12394, interleave = var_13841_interleave_0, values = (var_13699_cast_fp16, var_13701_cast_fp16, var_13703_cast_fp16, var_13705_cast_fp16))[name = string("op_13841_cast_fp16")]; + bool var_13843_interleave_0 = const()[name = string("op_13843_interleave_0"), val = bool(false)]; + tensor var_13843_cast_fp16 = concat(axis = var_12394, interleave = var_13843_interleave_0, values = (var_13707_cast_fp16, var_13709_cast_fp16, var_13711_cast_fp16, var_13713_cast_fp16))[name = string("op_13843_cast_fp16")]; + bool var_13845_interleave_0 = const()[name = string("op_13845_interleave_0"), val = bool(false)]; + tensor var_13845_cast_fp16 = concat(axis = var_12394, interleave = var_13845_interleave_0, values = (var_13715_cast_fp16, var_13717_cast_fp16, var_13719_cast_fp16, var_13721_cast_fp16))[name = string("op_13845_cast_fp16")]; + bool var_13847_interleave_0 = const()[name = string("op_13847_interleave_0"), val = bool(false)]; + tensor var_13847_cast_fp16 = concat(axis = var_12394, interleave = var_13847_interleave_0, values = (var_13723_cast_fp16, var_13725_cast_fp16, var_13727_cast_fp16, var_13729_cast_fp16))[name = string("op_13847_cast_fp16")]; + bool var_13849_interleave_0 = const()[name = string("op_13849_interleave_0"), val = bool(false)]; + tensor var_13849_cast_fp16 = concat(axis = var_12394, interleave = var_13849_interleave_0, values = (var_13731_cast_fp16, var_13733_cast_fp16, var_13735_cast_fp16, var_13737_cast_fp16))[name = string("op_13849_cast_fp16")]; + bool var_13851_interleave_0 = const()[name = string("op_13851_interleave_0"), val = bool(false)]; + tensor var_13851_cast_fp16 = concat(axis = var_12394, interleave = var_13851_interleave_0, values = (var_13739_cast_fp16, var_13741_cast_fp16, var_13743_cast_fp16, var_13745_cast_fp16))[name = string("op_13851_cast_fp16")]; + bool var_13853_interleave_0 = const()[name = string("op_13853_interleave_0"), val = bool(false)]; + tensor var_13853_cast_fp16 = concat(axis = var_12394, interleave = var_13853_interleave_0, values = (var_13747_cast_fp16, var_13749_cast_fp16, var_13751_cast_fp16, var_13753_cast_fp16))[name = string("op_13853_cast_fp16")]; + bool var_13855_interleave_0 = const()[name = string("op_13855_interleave_0"), val = bool(false)]; + tensor var_13855_cast_fp16 = concat(axis = var_12394, interleave = var_13855_interleave_0, values = (var_13755_cast_fp16, var_13757_cast_fp16, var_13759_cast_fp16, var_13761_cast_fp16))[name = string("op_13855_cast_fp16")]; + bool var_13857_interleave_0 = const()[name = string("op_13857_interleave_0"), val = bool(false)]; + tensor var_13857_cast_fp16 = concat(axis = var_12394, interleave = var_13857_interleave_0, values = (var_13763_cast_fp16, var_13765_cast_fp16, var_13767_cast_fp16, var_13769_cast_fp16))[name = string("op_13857_cast_fp16")]; + bool var_13859_interleave_0 = const()[name = string("op_13859_interleave_0"), val = bool(false)]; + tensor var_13859_cast_fp16 = concat(axis = var_12394, interleave = var_13859_interleave_0, values = (var_13771_cast_fp16, var_13773_cast_fp16, var_13775_cast_fp16, var_13777_cast_fp16))[name = string("op_13859_cast_fp16")]; + bool var_13861_interleave_0 = const()[name = string("op_13861_interleave_0"), val = bool(false)]; + tensor var_13861_cast_fp16 = concat(axis = var_12394, interleave = var_13861_interleave_0, values = (var_13779_cast_fp16, var_13781_cast_fp16, var_13783_cast_fp16, var_13785_cast_fp16))[name = string("op_13861_cast_fp16")]; + bool var_13863_interleave_0 = const()[name = string("op_13863_interleave_0"), val = bool(false)]; + tensor var_13863_cast_fp16 = concat(axis = var_12394, interleave = var_13863_interleave_0, values = (var_13787_cast_fp16, var_13789_cast_fp16, var_13791_cast_fp16, var_13793_cast_fp16))[name = string("op_13863_cast_fp16")]; + bool var_13865_interleave_0 = const()[name = string("op_13865_interleave_0"), val = bool(false)]; + tensor var_13865_cast_fp16 = concat(axis = var_12394, interleave = var_13865_interleave_0, values = (var_13795_cast_fp16, var_13797_cast_fp16, var_13799_cast_fp16, var_13801_cast_fp16))[name = string("op_13865_cast_fp16")]; + bool var_13867_interleave_0 = const()[name = string("op_13867_interleave_0"), val = bool(false)]; + tensor var_13867_cast_fp16 = concat(axis = var_12394, interleave = var_13867_interleave_0, values = (var_13803_cast_fp16, var_13805_cast_fp16, var_13807_cast_fp16, var_13809_cast_fp16))[name = string("op_13867_cast_fp16")]; + bool var_13869_interleave_0 = const()[name = string("op_13869_interleave_0"), val = bool(false)]; + tensor var_13869_cast_fp16 = concat(axis = var_12394, interleave = var_13869_interleave_0, values = (var_13811_cast_fp16, var_13813_cast_fp16, var_13815_cast_fp16, var_13817_cast_fp16))[name = string("op_13869_cast_fp16")]; + bool var_13871_interleave_0 = const()[name = string("op_13871_interleave_0"), val = bool(false)]; + tensor var_13871_cast_fp16 = concat(axis = var_12394, interleave = var_13871_interleave_0, values = (var_13819_cast_fp16, var_13821_cast_fp16, var_13823_cast_fp16, var_13825_cast_fp16))[name = string("op_13871_cast_fp16")]; + bool var_13873_interleave_0 = const()[name = string("op_13873_interleave_0"), val = bool(false)]; + tensor var_13873_cast_fp16 = concat(axis = var_12394, interleave = var_13873_interleave_0, values = (var_13827_cast_fp16, var_13829_cast_fp16, var_13831_cast_fp16, var_13833_cast_fp16))[name = string("op_13873_cast_fp16")]; + bool input_65_interleave_0 = const()[name = string("input_65_interleave_0"), val = bool(false)]; + tensor input_65_cast_fp16 = concat(axis = var_12419, interleave = input_65_interleave_0, values = (var_13835_cast_fp16, var_13837_cast_fp16, var_13839_cast_fp16, var_13841_cast_fp16, var_13843_cast_fp16, var_13845_cast_fp16, var_13847_cast_fp16, var_13849_cast_fp16, var_13851_cast_fp16, var_13853_cast_fp16, var_13855_cast_fp16, var_13857_cast_fp16, var_13859_cast_fp16, var_13861_cast_fp16, var_13863_cast_fp16, var_13865_cast_fp16, var_13867_cast_fp16, var_13869_cast_fp16, var_13871_cast_fp16, var_13873_cast_fp16))[name = string("input_65_cast_fp16")]; + string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; + tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; + tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; + int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339331520)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342608384)))]; + tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_13892_to_fp16 = const()[name = string("op_13892_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_13892_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342611008)))]; + tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342613632)))]; + fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; + tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; + tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; + int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342616256)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355723520)))]; + tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; + tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355733824)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368841088)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; + int32 var_13921 = const()[name = string("op_13921"), val = int32(3)]; + int32 var_13946 = const()[name = string("op_13946"), val = int32(1)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_13963_to_fp16 = const()[name = string("op_13963_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_13963_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368843712)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368846336)))]; + fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368848960)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372125824)))]; + tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; + string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; + tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; + tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; + int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372128448)))]; + tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; + string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; + tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; + tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; + int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375405312)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378682176)))]; + tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_14001_begin_0 = const()[name = string("op_14001_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14001_end_0 = const()[name = string("op_14001_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14001_end_mask_0 = const()[name = string("op_14001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14001_cast_fp16 = slice_by_index(begin = var_14001_begin_0, end = var_14001_end_0, end_mask = var_14001_end_mask_0, x = query_19_cast_fp16)[name = string("op_14001_cast_fp16")]; + tensor var_14005_begin_0 = const()[name = string("op_14005_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14005_end_0 = const()[name = string("op_14005_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_14005_end_mask_0 = const()[name = string("op_14005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14005_cast_fp16 = slice_by_index(begin = var_14005_begin_0, end = var_14005_end_0, end_mask = var_14005_end_mask_0, x = query_19_cast_fp16)[name = string("op_14005_cast_fp16")]; + tensor var_14009_begin_0 = const()[name = string("op_14009_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14009_end_0 = const()[name = string("op_14009_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_14009_end_mask_0 = const()[name = string("op_14009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14009_cast_fp16 = slice_by_index(begin = var_14009_begin_0, end = var_14009_end_0, end_mask = var_14009_end_mask_0, x = query_19_cast_fp16)[name = string("op_14009_cast_fp16")]; + tensor var_14013_begin_0 = const()[name = string("op_14013_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14013_end_0 = const()[name = string("op_14013_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_14013_end_mask_0 = const()[name = string("op_14013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14013_cast_fp16 = slice_by_index(begin = var_14013_begin_0, end = var_14013_end_0, end_mask = var_14013_end_mask_0, x = query_19_cast_fp16)[name = string("op_14013_cast_fp16")]; + tensor var_14017_begin_0 = const()[name = string("op_14017_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14017_end_0 = const()[name = string("op_14017_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_14017_end_mask_0 = const()[name = string("op_14017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14017_cast_fp16 = slice_by_index(begin = var_14017_begin_0, end = var_14017_end_0, end_mask = var_14017_end_mask_0, x = query_19_cast_fp16)[name = string("op_14017_cast_fp16")]; + tensor var_14021_begin_0 = const()[name = string("op_14021_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14021_end_0 = const()[name = string("op_14021_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_14021_end_mask_0 = const()[name = string("op_14021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14021_cast_fp16 = slice_by_index(begin = var_14021_begin_0, end = var_14021_end_0, end_mask = var_14021_end_mask_0, x = query_19_cast_fp16)[name = string("op_14021_cast_fp16")]; + tensor var_14025_begin_0 = const()[name = string("op_14025_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14025_end_0 = const()[name = string("op_14025_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_14025_end_mask_0 = const()[name = string("op_14025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14025_cast_fp16 = slice_by_index(begin = var_14025_begin_0, end = var_14025_end_0, end_mask = var_14025_end_mask_0, x = query_19_cast_fp16)[name = string("op_14025_cast_fp16")]; + tensor var_14029_begin_0 = const()[name = string("op_14029_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14029_end_0 = const()[name = string("op_14029_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_14029_end_mask_0 = const()[name = string("op_14029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14029_cast_fp16 = slice_by_index(begin = var_14029_begin_0, end = var_14029_end_0, end_mask = var_14029_end_mask_0, x = query_19_cast_fp16)[name = string("op_14029_cast_fp16")]; + tensor var_14033_begin_0 = const()[name = string("op_14033_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14033_end_0 = const()[name = string("op_14033_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_14033_end_mask_0 = const()[name = string("op_14033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14033_cast_fp16 = slice_by_index(begin = var_14033_begin_0, end = var_14033_end_0, end_mask = var_14033_end_mask_0, x = query_19_cast_fp16)[name = string("op_14033_cast_fp16")]; + tensor var_14037_begin_0 = const()[name = string("op_14037_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14037_end_0 = const()[name = string("op_14037_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_14037_end_mask_0 = const()[name = string("op_14037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14037_cast_fp16 = slice_by_index(begin = var_14037_begin_0, end = var_14037_end_0, end_mask = var_14037_end_mask_0, x = query_19_cast_fp16)[name = string("op_14037_cast_fp16")]; + tensor var_14041_begin_0 = const()[name = string("op_14041_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14041_end_0 = const()[name = string("op_14041_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_14041_end_mask_0 = const()[name = string("op_14041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14041_cast_fp16 = slice_by_index(begin = var_14041_begin_0, end = var_14041_end_0, end_mask = var_14041_end_mask_0, x = query_19_cast_fp16)[name = string("op_14041_cast_fp16")]; + tensor var_14045_begin_0 = const()[name = string("op_14045_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14045_end_0 = const()[name = string("op_14045_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_14045_end_mask_0 = const()[name = string("op_14045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14045_cast_fp16 = slice_by_index(begin = var_14045_begin_0, end = var_14045_end_0, end_mask = var_14045_end_mask_0, x = query_19_cast_fp16)[name = string("op_14045_cast_fp16")]; + tensor var_14049_begin_0 = const()[name = string("op_14049_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14049_end_0 = const()[name = string("op_14049_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_14049_end_mask_0 = const()[name = string("op_14049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14049_cast_fp16 = slice_by_index(begin = var_14049_begin_0, end = var_14049_end_0, end_mask = var_14049_end_mask_0, x = query_19_cast_fp16)[name = string("op_14049_cast_fp16")]; + tensor var_14053_begin_0 = const()[name = string("op_14053_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14053_end_0 = const()[name = string("op_14053_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_14053_end_mask_0 = const()[name = string("op_14053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14053_cast_fp16 = slice_by_index(begin = var_14053_begin_0, end = var_14053_end_0, end_mask = var_14053_end_mask_0, x = query_19_cast_fp16)[name = string("op_14053_cast_fp16")]; + tensor var_14057_begin_0 = const()[name = string("op_14057_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14057_end_0 = const()[name = string("op_14057_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_14057_end_mask_0 = const()[name = string("op_14057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14057_cast_fp16 = slice_by_index(begin = var_14057_begin_0, end = var_14057_end_0, end_mask = var_14057_end_mask_0, x = query_19_cast_fp16)[name = string("op_14057_cast_fp16")]; + tensor var_14061_begin_0 = const()[name = string("op_14061_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14061_end_0 = const()[name = string("op_14061_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_14061_end_mask_0 = const()[name = string("op_14061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14061_cast_fp16 = slice_by_index(begin = var_14061_begin_0, end = var_14061_end_0, end_mask = var_14061_end_mask_0, x = query_19_cast_fp16)[name = string("op_14061_cast_fp16")]; + tensor var_14065_begin_0 = const()[name = string("op_14065_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14065_end_0 = const()[name = string("op_14065_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_14065_end_mask_0 = const()[name = string("op_14065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14065_cast_fp16 = slice_by_index(begin = var_14065_begin_0, end = var_14065_end_0, end_mask = var_14065_end_mask_0, x = query_19_cast_fp16)[name = string("op_14065_cast_fp16")]; + tensor var_14069_begin_0 = const()[name = string("op_14069_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14069_end_0 = const()[name = string("op_14069_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_14069_end_mask_0 = const()[name = string("op_14069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14069_cast_fp16 = slice_by_index(begin = var_14069_begin_0, end = var_14069_end_0, end_mask = var_14069_end_mask_0, x = query_19_cast_fp16)[name = string("op_14069_cast_fp16")]; + tensor var_14073_begin_0 = const()[name = string("op_14073_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14073_end_0 = const()[name = string("op_14073_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_14073_end_mask_0 = const()[name = string("op_14073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14073_cast_fp16 = slice_by_index(begin = var_14073_begin_0, end = var_14073_end_0, end_mask = var_14073_end_mask_0, x = query_19_cast_fp16)[name = string("op_14073_cast_fp16")]; + tensor var_14077_begin_0 = const()[name = string("op_14077_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14077_end_0 = const()[name = string("op_14077_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_14077_end_mask_0 = const()[name = string("op_14077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14077_cast_fp16 = slice_by_index(begin = var_14077_begin_0, end = var_14077_end_0, end_mask = var_14077_end_mask_0, x = query_19_cast_fp16)[name = string("op_14077_cast_fp16")]; + tensor var_14086_begin_0 = const()[name = string("op_14086_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14086_end_0 = const()[name = string("op_14086_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14086_end_mask_0 = const()[name = string("op_14086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14086_cast_fp16 = slice_by_index(begin = var_14086_begin_0, end = var_14086_end_0, end_mask = var_14086_end_mask_0, x = var_14001_cast_fp16)[name = string("op_14086_cast_fp16")]; + tensor var_14093_begin_0 = const()[name = string("op_14093_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14093_end_0 = const()[name = string("op_14093_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14093_end_mask_0 = const()[name = string("op_14093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14093_cast_fp16 = slice_by_index(begin = var_14093_begin_0, end = var_14093_end_0, end_mask = var_14093_end_mask_0, x = var_14001_cast_fp16)[name = string("op_14093_cast_fp16")]; + tensor var_14100_begin_0 = const()[name = string("op_14100_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14100_end_0 = const()[name = string("op_14100_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14100_end_mask_0 = const()[name = string("op_14100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14100_cast_fp16 = slice_by_index(begin = var_14100_begin_0, end = var_14100_end_0, end_mask = var_14100_end_mask_0, x = var_14001_cast_fp16)[name = string("op_14100_cast_fp16")]; + tensor var_14107_begin_0 = const()[name = string("op_14107_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14107_end_0 = const()[name = string("op_14107_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14107_end_mask_0 = const()[name = string("op_14107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14107_cast_fp16 = slice_by_index(begin = var_14107_begin_0, end = var_14107_end_0, end_mask = var_14107_end_mask_0, x = var_14001_cast_fp16)[name = string("op_14107_cast_fp16")]; + tensor var_14114_begin_0 = const()[name = string("op_14114_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14114_end_0 = const()[name = string("op_14114_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14114_end_mask_0 = const()[name = string("op_14114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14114_cast_fp16 = slice_by_index(begin = var_14114_begin_0, end = var_14114_end_0, end_mask = var_14114_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14114_cast_fp16")]; + tensor var_14121_begin_0 = const()[name = string("op_14121_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14121_end_0 = const()[name = string("op_14121_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14121_end_mask_0 = const()[name = string("op_14121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14121_cast_fp16 = slice_by_index(begin = var_14121_begin_0, end = var_14121_end_0, end_mask = var_14121_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14121_cast_fp16")]; + tensor var_14128_begin_0 = const()[name = string("op_14128_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14128_end_0 = const()[name = string("op_14128_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14128_end_mask_0 = const()[name = string("op_14128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14128_cast_fp16 = slice_by_index(begin = var_14128_begin_0, end = var_14128_end_0, end_mask = var_14128_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14128_cast_fp16")]; + tensor var_14135_begin_0 = const()[name = string("op_14135_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14135_end_0 = const()[name = string("op_14135_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14135_end_mask_0 = const()[name = string("op_14135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14135_cast_fp16 = slice_by_index(begin = var_14135_begin_0, end = var_14135_end_0, end_mask = var_14135_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14135_cast_fp16")]; + tensor var_14142_begin_0 = const()[name = string("op_14142_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14142_end_0 = const()[name = string("op_14142_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14142_end_mask_0 = const()[name = string("op_14142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14142_cast_fp16 = slice_by_index(begin = var_14142_begin_0, end = var_14142_end_0, end_mask = var_14142_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14142_cast_fp16")]; + tensor var_14149_begin_0 = const()[name = string("op_14149_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14149_end_0 = const()[name = string("op_14149_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14149_end_mask_0 = const()[name = string("op_14149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14149_cast_fp16 = slice_by_index(begin = var_14149_begin_0, end = var_14149_end_0, end_mask = var_14149_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14149_cast_fp16")]; + tensor var_14156_begin_0 = const()[name = string("op_14156_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14156_end_0 = const()[name = string("op_14156_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14156_end_mask_0 = const()[name = string("op_14156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14156_cast_fp16 = slice_by_index(begin = var_14156_begin_0, end = var_14156_end_0, end_mask = var_14156_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14156_cast_fp16")]; + tensor var_14163_begin_0 = const()[name = string("op_14163_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14163_end_0 = const()[name = string("op_14163_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14163_end_mask_0 = const()[name = string("op_14163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14163_cast_fp16 = slice_by_index(begin = var_14163_begin_0, end = var_14163_end_0, end_mask = var_14163_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14163_cast_fp16")]; + tensor var_14170_begin_0 = const()[name = string("op_14170_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14170_end_0 = const()[name = string("op_14170_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14170_end_mask_0 = const()[name = string("op_14170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14170_cast_fp16 = slice_by_index(begin = var_14170_begin_0, end = var_14170_end_0, end_mask = var_14170_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14170_cast_fp16")]; + tensor var_14177_begin_0 = const()[name = string("op_14177_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14177_end_0 = const()[name = string("op_14177_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14177_end_mask_0 = const()[name = string("op_14177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14177_cast_fp16 = slice_by_index(begin = var_14177_begin_0, end = var_14177_end_0, end_mask = var_14177_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14177_cast_fp16")]; + tensor var_14184_begin_0 = const()[name = string("op_14184_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14184_end_0 = const()[name = string("op_14184_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14184_end_mask_0 = const()[name = string("op_14184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14184_cast_fp16 = slice_by_index(begin = var_14184_begin_0, end = var_14184_end_0, end_mask = var_14184_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14184_cast_fp16")]; + tensor var_14191_begin_0 = const()[name = string("op_14191_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14191_end_0 = const()[name = string("op_14191_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14191_end_mask_0 = const()[name = string("op_14191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14191_cast_fp16 = slice_by_index(begin = var_14191_begin_0, end = var_14191_end_0, end_mask = var_14191_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14191_cast_fp16")]; + tensor var_14198_begin_0 = const()[name = string("op_14198_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14198_end_0 = const()[name = string("op_14198_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14198_end_mask_0 = const()[name = string("op_14198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14198_cast_fp16 = slice_by_index(begin = var_14198_begin_0, end = var_14198_end_0, end_mask = var_14198_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14198_cast_fp16")]; + tensor var_14205_begin_0 = const()[name = string("op_14205_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14205_end_0 = const()[name = string("op_14205_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14205_end_mask_0 = const()[name = string("op_14205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14205_cast_fp16 = slice_by_index(begin = var_14205_begin_0, end = var_14205_end_0, end_mask = var_14205_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14205_cast_fp16")]; + tensor var_14212_begin_0 = const()[name = string("op_14212_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14212_end_0 = const()[name = string("op_14212_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14212_end_mask_0 = const()[name = string("op_14212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14212_cast_fp16 = slice_by_index(begin = var_14212_begin_0, end = var_14212_end_0, end_mask = var_14212_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14212_cast_fp16")]; + tensor var_14219_begin_0 = const()[name = string("op_14219_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14219_end_0 = const()[name = string("op_14219_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14219_end_mask_0 = const()[name = string("op_14219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14219_cast_fp16 = slice_by_index(begin = var_14219_begin_0, end = var_14219_end_0, end_mask = var_14219_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14219_cast_fp16")]; + tensor var_14226_begin_0 = const()[name = string("op_14226_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14226_end_0 = const()[name = string("op_14226_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14226_end_mask_0 = const()[name = string("op_14226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14226_cast_fp16 = slice_by_index(begin = var_14226_begin_0, end = var_14226_end_0, end_mask = var_14226_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14226_cast_fp16")]; + tensor var_14233_begin_0 = const()[name = string("op_14233_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14233_end_0 = const()[name = string("op_14233_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14233_end_mask_0 = const()[name = string("op_14233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14233_cast_fp16 = slice_by_index(begin = var_14233_begin_0, end = var_14233_end_0, end_mask = var_14233_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14233_cast_fp16")]; + tensor var_14240_begin_0 = const()[name = string("op_14240_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14240_end_0 = const()[name = string("op_14240_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14240_end_mask_0 = const()[name = string("op_14240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14240_cast_fp16 = slice_by_index(begin = var_14240_begin_0, end = var_14240_end_0, end_mask = var_14240_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14240_cast_fp16")]; + tensor var_14247_begin_0 = const()[name = string("op_14247_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14247_end_0 = const()[name = string("op_14247_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14247_end_mask_0 = const()[name = string("op_14247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14247_cast_fp16 = slice_by_index(begin = var_14247_begin_0, end = var_14247_end_0, end_mask = var_14247_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14247_cast_fp16")]; + tensor var_14254_begin_0 = const()[name = string("op_14254_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14254_end_0 = const()[name = string("op_14254_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14254_end_mask_0 = const()[name = string("op_14254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14254_cast_fp16 = slice_by_index(begin = var_14254_begin_0, end = var_14254_end_0, end_mask = var_14254_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14254_cast_fp16")]; + tensor var_14261_begin_0 = const()[name = string("op_14261_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14261_end_0 = const()[name = string("op_14261_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14261_end_mask_0 = const()[name = string("op_14261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14261_cast_fp16 = slice_by_index(begin = var_14261_begin_0, end = var_14261_end_0, end_mask = var_14261_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14261_cast_fp16")]; + tensor var_14268_begin_0 = const()[name = string("op_14268_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14268_end_0 = const()[name = string("op_14268_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14268_end_mask_0 = const()[name = string("op_14268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14268_cast_fp16 = slice_by_index(begin = var_14268_begin_0, end = var_14268_end_0, end_mask = var_14268_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14268_cast_fp16")]; + tensor var_14275_begin_0 = const()[name = string("op_14275_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14275_end_0 = const()[name = string("op_14275_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14275_end_mask_0 = const()[name = string("op_14275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14275_cast_fp16 = slice_by_index(begin = var_14275_begin_0, end = var_14275_end_0, end_mask = var_14275_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14275_cast_fp16")]; + tensor var_14282_begin_0 = const()[name = string("op_14282_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14282_end_0 = const()[name = string("op_14282_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14282_end_mask_0 = const()[name = string("op_14282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14282_cast_fp16 = slice_by_index(begin = var_14282_begin_0, end = var_14282_end_0, end_mask = var_14282_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14282_cast_fp16")]; + tensor var_14289_begin_0 = const()[name = string("op_14289_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14289_end_0 = const()[name = string("op_14289_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14289_end_mask_0 = const()[name = string("op_14289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14289_cast_fp16 = slice_by_index(begin = var_14289_begin_0, end = var_14289_end_0, end_mask = var_14289_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14289_cast_fp16")]; + tensor var_14296_begin_0 = const()[name = string("op_14296_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14296_end_0 = const()[name = string("op_14296_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14296_end_mask_0 = const()[name = string("op_14296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14296_cast_fp16 = slice_by_index(begin = var_14296_begin_0, end = var_14296_end_0, end_mask = var_14296_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14296_cast_fp16")]; + tensor var_14303_begin_0 = const()[name = string("op_14303_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14303_end_0 = const()[name = string("op_14303_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14303_end_mask_0 = const()[name = string("op_14303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14303_cast_fp16 = slice_by_index(begin = var_14303_begin_0, end = var_14303_end_0, end_mask = var_14303_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14303_cast_fp16")]; + tensor var_14310_begin_0 = const()[name = string("op_14310_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14310_end_0 = const()[name = string("op_14310_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14310_end_mask_0 = const()[name = string("op_14310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14310_cast_fp16 = slice_by_index(begin = var_14310_begin_0, end = var_14310_end_0, end_mask = var_14310_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14310_cast_fp16")]; + tensor var_14317_begin_0 = const()[name = string("op_14317_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14317_end_0 = const()[name = string("op_14317_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14317_end_mask_0 = const()[name = string("op_14317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14317_cast_fp16 = slice_by_index(begin = var_14317_begin_0, end = var_14317_end_0, end_mask = var_14317_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14317_cast_fp16")]; + tensor var_14324_begin_0 = const()[name = string("op_14324_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14324_end_0 = const()[name = string("op_14324_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14324_end_mask_0 = const()[name = string("op_14324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14324_cast_fp16 = slice_by_index(begin = var_14324_begin_0, end = var_14324_end_0, end_mask = var_14324_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14324_cast_fp16")]; + tensor var_14331_begin_0 = const()[name = string("op_14331_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14331_end_0 = const()[name = string("op_14331_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14331_end_mask_0 = const()[name = string("op_14331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14331_cast_fp16 = slice_by_index(begin = var_14331_begin_0, end = var_14331_end_0, end_mask = var_14331_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14331_cast_fp16")]; + tensor var_14338_begin_0 = const()[name = string("op_14338_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14338_end_0 = const()[name = string("op_14338_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14338_end_mask_0 = const()[name = string("op_14338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14338_cast_fp16 = slice_by_index(begin = var_14338_begin_0, end = var_14338_end_0, end_mask = var_14338_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14338_cast_fp16")]; + tensor var_14345_begin_0 = const()[name = string("op_14345_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14345_end_0 = const()[name = string("op_14345_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14345_end_mask_0 = const()[name = string("op_14345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14345_cast_fp16 = slice_by_index(begin = var_14345_begin_0, end = var_14345_end_0, end_mask = var_14345_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14345_cast_fp16")]; + tensor var_14352_begin_0 = const()[name = string("op_14352_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14352_end_0 = const()[name = string("op_14352_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14352_end_mask_0 = const()[name = string("op_14352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14352_cast_fp16 = slice_by_index(begin = var_14352_begin_0, end = var_14352_end_0, end_mask = var_14352_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14352_cast_fp16")]; + tensor var_14359_begin_0 = const()[name = string("op_14359_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14359_end_0 = const()[name = string("op_14359_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14359_end_mask_0 = const()[name = string("op_14359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14359_cast_fp16 = slice_by_index(begin = var_14359_begin_0, end = var_14359_end_0, end_mask = var_14359_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14359_cast_fp16")]; + tensor var_14366_begin_0 = const()[name = string("op_14366_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14366_end_0 = const()[name = string("op_14366_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14366_end_mask_0 = const()[name = string("op_14366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14366_cast_fp16 = slice_by_index(begin = var_14366_begin_0, end = var_14366_end_0, end_mask = var_14366_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14366_cast_fp16")]; + tensor var_14373_begin_0 = const()[name = string("op_14373_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14373_end_0 = const()[name = string("op_14373_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14373_end_mask_0 = const()[name = string("op_14373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14373_cast_fp16 = slice_by_index(begin = var_14373_begin_0, end = var_14373_end_0, end_mask = var_14373_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14373_cast_fp16")]; + tensor var_14380_begin_0 = const()[name = string("op_14380_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14380_end_0 = const()[name = string("op_14380_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14380_end_mask_0 = const()[name = string("op_14380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14380_cast_fp16 = slice_by_index(begin = var_14380_begin_0, end = var_14380_end_0, end_mask = var_14380_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14380_cast_fp16")]; + tensor var_14387_begin_0 = const()[name = string("op_14387_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14387_end_0 = const()[name = string("op_14387_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14387_end_mask_0 = const()[name = string("op_14387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14387_cast_fp16 = slice_by_index(begin = var_14387_begin_0, end = var_14387_end_0, end_mask = var_14387_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14387_cast_fp16")]; + tensor var_14394_begin_0 = const()[name = string("op_14394_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14394_end_0 = const()[name = string("op_14394_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14394_end_mask_0 = const()[name = string("op_14394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14394_cast_fp16 = slice_by_index(begin = var_14394_begin_0, end = var_14394_end_0, end_mask = var_14394_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14394_cast_fp16")]; + tensor var_14401_begin_0 = const()[name = string("op_14401_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14401_end_0 = const()[name = string("op_14401_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14401_end_mask_0 = const()[name = string("op_14401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14401_cast_fp16 = slice_by_index(begin = var_14401_begin_0, end = var_14401_end_0, end_mask = var_14401_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14401_cast_fp16")]; + tensor var_14408_begin_0 = const()[name = string("op_14408_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14408_end_0 = const()[name = string("op_14408_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14408_end_mask_0 = const()[name = string("op_14408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14408_cast_fp16 = slice_by_index(begin = var_14408_begin_0, end = var_14408_end_0, end_mask = var_14408_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14408_cast_fp16")]; + tensor var_14415_begin_0 = const()[name = string("op_14415_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14415_end_0 = const()[name = string("op_14415_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14415_end_mask_0 = const()[name = string("op_14415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14415_cast_fp16 = slice_by_index(begin = var_14415_begin_0, end = var_14415_end_0, end_mask = var_14415_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14415_cast_fp16")]; + tensor var_14422_begin_0 = const()[name = string("op_14422_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14422_end_0 = const()[name = string("op_14422_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14422_end_mask_0 = const()[name = string("op_14422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14422_cast_fp16 = slice_by_index(begin = var_14422_begin_0, end = var_14422_end_0, end_mask = var_14422_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14422_cast_fp16")]; + tensor var_14429_begin_0 = const()[name = string("op_14429_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14429_end_0 = const()[name = string("op_14429_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14429_end_mask_0 = const()[name = string("op_14429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14429_cast_fp16 = slice_by_index(begin = var_14429_begin_0, end = var_14429_end_0, end_mask = var_14429_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14429_cast_fp16")]; + tensor var_14436_begin_0 = const()[name = string("op_14436_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14436_end_0 = const()[name = string("op_14436_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14436_end_mask_0 = const()[name = string("op_14436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14436_cast_fp16 = slice_by_index(begin = var_14436_begin_0, end = var_14436_end_0, end_mask = var_14436_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14436_cast_fp16")]; + tensor var_14443_begin_0 = const()[name = string("op_14443_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14443_end_0 = const()[name = string("op_14443_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14443_end_mask_0 = const()[name = string("op_14443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14443_cast_fp16 = slice_by_index(begin = var_14443_begin_0, end = var_14443_end_0, end_mask = var_14443_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14443_cast_fp16")]; + tensor var_14450_begin_0 = const()[name = string("op_14450_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14450_end_0 = const()[name = string("op_14450_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14450_end_mask_0 = const()[name = string("op_14450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14450_cast_fp16 = slice_by_index(begin = var_14450_begin_0, end = var_14450_end_0, end_mask = var_14450_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14450_cast_fp16")]; + tensor var_14457_begin_0 = const()[name = string("op_14457_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14457_end_0 = const()[name = string("op_14457_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14457_end_mask_0 = const()[name = string("op_14457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14457_cast_fp16 = slice_by_index(begin = var_14457_begin_0, end = var_14457_end_0, end_mask = var_14457_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14457_cast_fp16")]; + tensor var_14464_begin_0 = const()[name = string("op_14464_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14464_end_0 = const()[name = string("op_14464_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14464_end_mask_0 = const()[name = string("op_14464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14464_cast_fp16 = slice_by_index(begin = var_14464_begin_0, end = var_14464_end_0, end_mask = var_14464_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14464_cast_fp16")]; + tensor var_14471_begin_0 = const()[name = string("op_14471_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14471_end_0 = const()[name = string("op_14471_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14471_end_mask_0 = const()[name = string("op_14471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14471_cast_fp16 = slice_by_index(begin = var_14471_begin_0, end = var_14471_end_0, end_mask = var_14471_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14471_cast_fp16")]; + tensor var_14478_begin_0 = const()[name = string("op_14478_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14478_end_0 = const()[name = string("op_14478_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14478_end_mask_0 = const()[name = string("op_14478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14478_cast_fp16 = slice_by_index(begin = var_14478_begin_0, end = var_14478_end_0, end_mask = var_14478_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14478_cast_fp16")]; + tensor var_14485_begin_0 = const()[name = string("op_14485_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14485_end_0 = const()[name = string("op_14485_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14485_end_mask_0 = const()[name = string("op_14485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14485_cast_fp16 = slice_by_index(begin = var_14485_begin_0, end = var_14485_end_0, end_mask = var_14485_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14485_cast_fp16")]; + tensor var_14492_begin_0 = const()[name = string("op_14492_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14492_end_0 = const()[name = string("op_14492_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14492_end_mask_0 = const()[name = string("op_14492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14492_cast_fp16 = slice_by_index(begin = var_14492_begin_0, end = var_14492_end_0, end_mask = var_14492_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14492_cast_fp16")]; + tensor var_14499_begin_0 = const()[name = string("op_14499_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14499_end_0 = const()[name = string("op_14499_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14499_end_mask_0 = const()[name = string("op_14499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14499_cast_fp16 = slice_by_index(begin = var_14499_begin_0, end = var_14499_end_0, end_mask = var_14499_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14499_cast_fp16")]; + tensor var_14506_begin_0 = const()[name = string("op_14506_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14506_end_0 = const()[name = string("op_14506_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14506_end_mask_0 = const()[name = string("op_14506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14506_cast_fp16 = slice_by_index(begin = var_14506_begin_0, end = var_14506_end_0, end_mask = var_14506_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14506_cast_fp16")]; + tensor var_14513_begin_0 = const()[name = string("op_14513_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14513_end_0 = const()[name = string("op_14513_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14513_end_mask_0 = const()[name = string("op_14513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14513_cast_fp16 = slice_by_index(begin = var_14513_begin_0, end = var_14513_end_0, end_mask = var_14513_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14513_cast_fp16")]; + tensor var_14520_begin_0 = const()[name = string("op_14520_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14520_end_0 = const()[name = string("op_14520_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14520_end_mask_0 = const()[name = string("op_14520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14520_cast_fp16 = slice_by_index(begin = var_14520_begin_0, end = var_14520_end_0, end_mask = var_14520_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14520_cast_fp16")]; + tensor var_14527_begin_0 = const()[name = string("op_14527_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14527_end_0 = const()[name = string("op_14527_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14527_end_mask_0 = const()[name = string("op_14527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14527_cast_fp16 = slice_by_index(begin = var_14527_begin_0, end = var_14527_end_0, end_mask = var_14527_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14527_cast_fp16")]; + tensor var_14534_begin_0 = const()[name = string("op_14534_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14534_end_0 = const()[name = string("op_14534_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14534_end_mask_0 = const()[name = string("op_14534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14534_cast_fp16 = slice_by_index(begin = var_14534_begin_0, end = var_14534_end_0, end_mask = var_14534_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14534_cast_fp16")]; + tensor var_14541_begin_0 = const()[name = string("op_14541_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14541_end_0 = const()[name = string("op_14541_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14541_end_mask_0 = const()[name = string("op_14541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14541_cast_fp16 = slice_by_index(begin = var_14541_begin_0, end = var_14541_end_0, end_mask = var_14541_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14541_cast_fp16")]; + tensor var_14548_begin_0 = const()[name = string("op_14548_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14548_end_0 = const()[name = string("op_14548_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14548_end_mask_0 = const()[name = string("op_14548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14548_cast_fp16 = slice_by_index(begin = var_14548_begin_0, end = var_14548_end_0, end_mask = var_14548_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14548_cast_fp16")]; + tensor var_14555_begin_0 = const()[name = string("op_14555_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14555_end_0 = const()[name = string("op_14555_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14555_end_mask_0 = const()[name = string("op_14555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14555_cast_fp16 = slice_by_index(begin = var_14555_begin_0, end = var_14555_end_0, end_mask = var_14555_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14555_cast_fp16")]; + tensor var_14562_begin_0 = const()[name = string("op_14562_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14562_end_0 = const()[name = string("op_14562_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14562_end_mask_0 = const()[name = string("op_14562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14562_cast_fp16 = slice_by_index(begin = var_14562_begin_0, end = var_14562_end_0, end_mask = var_14562_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14562_cast_fp16")]; + tensor var_14569_begin_0 = const()[name = string("op_14569_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14569_end_0 = const()[name = string("op_14569_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14569_end_mask_0 = const()[name = string("op_14569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14569_cast_fp16 = slice_by_index(begin = var_14569_begin_0, end = var_14569_end_0, end_mask = var_14569_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14569_cast_fp16")]; + tensor var_14576_begin_0 = const()[name = string("op_14576_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14576_end_0 = const()[name = string("op_14576_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14576_end_mask_0 = const()[name = string("op_14576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14576_cast_fp16 = slice_by_index(begin = var_14576_begin_0, end = var_14576_end_0, end_mask = var_14576_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14576_cast_fp16")]; + tensor var_14583_begin_0 = const()[name = string("op_14583_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14583_end_0 = const()[name = string("op_14583_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14583_end_mask_0 = const()[name = string("op_14583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14583_cast_fp16 = slice_by_index(begin = var_14583_begin_0, end = var_14583_end_0, end_mask = var_14583_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14583_cast_fp16")]; + tensor var_14590_begin_0 = const()[name = string("op_14590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14590_end_0 = const()[name = string("op_14590_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14590_end_mask_0 = const()[name = string("op_14590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14590_cast_fp16 = slice_by_index(begin = var_14590_begin_0, end = var_14590_end_0, end_mask = var_14590_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14590_cast_fp16")]; + tensor var_14597_begin_0 = const()[name = string("op_14597_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14597_end_0 = const()[name = string("op_14597_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14597_end_mask_0 = const()[name = string("op_14597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14597_cast_fp16 = slice_by_index(begin = var_14597_begin_0, end = var_14597_end_0, end_mask = var_14597_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14597_cast_fp16")]; + tensor var_14604_begin_0 = const()[name = string("op_14604_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14604_end_0 = const()[name = string("op_14604_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14604_end_mask_0 = const()[name = string("op_14604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14604_cast_fp16 = slice_by_index(begin = var_14604_begin_0, end = var_14604_end_0, end_mask = var_14604_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14604_cast_fp16")]; + tensor var_14611_begin_0 = const()[name = string("op_14611_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14611_end_0 = const()[name = string("op_14611_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14611_end_mask_0 = const()[name = string("op_14611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14611_cast_fp16 = slice_by_index(begin = var_14611_begin_0, end = var_14611_end_0, end_mask = var_14611_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14611_cast_fp16")]; + tensor var_14618_begin_0 = const()[name = string("op_14618_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14618_end_0 = const()[name = string("op_14618_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14618_end_mask_0 = const()[name = string("op_14618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14618_cast_fp16 = slice_by_index(begin = var_14618_begin_0, end = var_14618_end_0, end_mask = var_14618_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14618_cast_fp16")]; + tensor var_14625_begin_0 = const()[name = string("op_14625_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14625_end_0 = const()[name = string("op_14625_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14625_end_mask_0 = const()[name = string("op_14625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14625_cast_fp16 = slice_by_index(begin = var_14625_begin_0, end = var_14625_end_0, end_mask = var_14625_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14625_cast_fp16")]; + tensor var_14632_begin_0 = const()[name = string("op_14632_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14632_end_0 = const()[name = string("op_14632_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14632_end_mask_0 = const()[name = string("op_14632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14632_cast_fp16 = slice_by_index(begin = var_14632_begin_0, end = var_14632_end_0, end_mask = var_14632_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14632_cast_fp16")]; + tensor var_14639_begin_0 = const()[name = string("op_14639_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14639_end_0 = const()[name = string("op_14639_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14639_end_mask_0 = const()[name = string("op_14639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14639_cast_fp16 = slice_by_index(begin = var_14639_begin_0, end = var_14639_end_0, end_mask = var_14639_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14639_cast_fp16")]; + tensor k_19_perm_0 = const()[name = string("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14644_begin_0 = const()[name = string("op_14644_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14644_end_0 = const()[name = string("op_14644_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_14644_end_mask_0 = const()[name = string("op_14644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = string("transpose_22")]; + tensor var_14644_cast_fp16 = slice_by_index(begin = var_14644_begin_0, end = var_14644_end_0, end_mask = var_14644_end_mask_0, x = k_19_cast_fp16)[name = string("op_14644_cast_fp16")]; + tensor var_14648_begin_0 = const()[name = string("op_14648_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14648_end_0 = const()[name = string("op_14648_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_14648_end_mask_0 = const()[name = string("op_14648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14648_cast_fp16 = slice_by_index(begin = var_14648_begin_0, end = var_14648_end_0, end_mask = var_14648_end_mask_0, x = k_19_cast_fp16)[name = string("op_14648_cast_fp16")]; + tensor var_14652_begin_0 = const()[name = string("op_14652_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14652_end_0 = const()[name = string("op_14652_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_14652_end_mask_0 = const()[name = string("op_14652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14652_cast_fp16 = slice_by_index(begin = var_14652_begin_0, end = var_14652_end_0, end_mask = var_14652_end_mask_0, x = k_19_cast_fp16)[name = string("op_14652_cast_fp16")]; + tensor var_14656_begin_0 = const()[name = string("op_14656_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14656_end_0 = const()[name = string("op_14656_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_14656_end_mask_0 = const()[name = string("op_14656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14656_cast_fp16 = slice_by_index(begin = var_14656_begin_0, end = var_14656_end_0, end_mask = var_14656_end_mask_0, x = k_19_cast_fp16)[name = string("op_14656_cast_fp16")]; + tensor var_14660_begin_0 = const()[name = string("op_14660_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14660_end_0 = const()[name = string("op_14660_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_14660_end_mask_0 = const()[name = string("op_14660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14660_cast_fp16 = slice_by_index(begin = var_14660_begin_0, end = var_14660_end_0, end_mask = var_14660_end_mask_0, x = k_19_cast_fp16)[name = string("op_14660_cast_fp16")]; + tensor var_14664_begin_0 = const()[name = string("op_14664_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14664_end_0 = const()[name = string("op_14664_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_14664_end_mask_0 = const()[name = string("op_14664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14664_cast_fp16 = slice_by_index(begin = var_14664_begin_0, end = var_14664_end_0, end_mask = var_14664_end_mask_0, x = k_19_cast_fp16)[name = string("op_14664_cast_fp16")]; + tensor var_14668_begin_0 = const()[name = string("op_14668_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14668_end_0 = const()[name = string("op_14668_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_14668_end_mask_0 = const()[name = string("op_14668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14668_cast_fp16 = slice_by_index(begin = var_14668_begin_0, end = var_14668_end_0, end_mask = var_14668_end_mask_0, x = k_19_cast_fp16)[name = string("op_14668_cast_fp16")]; + tensor var_14672_begin_0 = const()[name = string("op_14672_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14672_end_0 = const()[name = string("op_14672_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_14672_end_mask_0 = const()[name = string("op_14672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14672_cast_fp16 = slice_by_index(begin = var_14672_begin_0, end = var_14672_end_0, end_mask = var_14672_end_mask_0, x = k_19_cast_fp16)[name = string("op_14672_cast_fp16")]; + tensor var_14676_begin_0 = const()[name = string("op_14676_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14676_end_0 = const()[name = string("op_14676_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_14676_end_mask_0 = const()[name = string("op_14676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14676_cast_fp16 = slice_by_index(begin = var_14676_begin_0, end = var_14676_end_0, end_mask = var_14676_end_mask_0, x = k_19_cast_fp16)[name = string("op_14676_cast_fp16")]; + tensor var_14680_begin_0 = const()[name = string("op_14680_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14680_end_0 = const()[name = string("op_14680_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_14680_end_mask_0 = const()[name = string("op_14680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14680_cast_fp16 = slice_by_index(begin = var_14680_begin_0, end = var_14680_end_0, end_mask = var_14680_end_mask_0, x = k_19_cast_fp16)[name = string("op_14680_cast_fp16")]; + tensor var_14684_begin_0 = const()[name = string("op_14684_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14684_end_0 = const()[name = string("op_14684_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_14684_end_mask_0 = const()[name = string("op_14684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14684_cast_fp16 = slice_by_index(begin = var_14684_begin_0, end = var_14684_end_0, end_mask = var_14684_end_mask_0, x = k_19_cast_fp16)[name = string("op_14684_cast_fp16")]; + tensor var_14688_begin_0 = const()[name = string("op_14688_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14688_end_0 = const()[name = string("op_14688_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_14688_end_mask_0 = const()[name = string("op_14688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14688_cast_fp16 = slice_by_index(begin = var_14688_begin_0, end = var_14688_end_0, end_mask = var_14688_end_mask_0, x = k_19_cast_fp16)[name = string("op_14688_cast_fp16")]; + tensor var_14692_begin_0 = const()[name = string("op_14692_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14692_end_0 = const()[name = string("op_14692_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_14692_end_mask_0 = const()[name = string("op_14692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14692_cast_fp16 = slice_by_index(begin = var_14692_begin_0, end = var_14692_end_0, end_mask = var_14692_end_mask_0, x = k_19_cast_fp16)[name = string("op_14692_cast_fp16")]; + tensor var_14696_begin_0 = const()[name = string("op_14696_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14696_end_0 = const()[name = string("op_14696_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_14696_end_mask_0 = const()[name = string("op_14696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14696_cast_fp16 = slice_by_index(begin = var_14696_begin_0, end = var_14696_end_0, end_mask = var_14696_end_mask_0, x = k_19_cast_fp16)[name = string("op_14696_cast_fp16")]; + tensor var_14700_begin_0 = const()[name = string("op_14700_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14700_end_0 = const()[name = string("op_14700_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_14700_end_mask_0 = const()[name = string("op_14700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14700_cast_fp16 = slice_by_index(begin = var_14700_begin_0, end = var_14700_end_0, end_mask = var_14700_end_mask_0, x = k_19_cast_fp16)[name = string("op_14700_cast_fp16")]; + tensor var_14704_begin_0 = const()[name = string("op_14704_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14704_end_0 = const()[name = string("op_14704_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_14704_end_mask_0 = const()[name = string("op_14704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14704_cast_fp16 = slice_by_index(begin = var_14704_begin_0, end = var_14704_end_0, end_mask = var_14704_end_mask_0, x = k_19_cast_fp16)[name = string("op_14704_cast_fp16")]; + tensor var_14708_begin_0 = const()[name = string("op_14708_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14708_end_0 = const()[name = string("op_14708_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_14708_end_mask_0 = const()[name = string("op_14708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14708_cast_fp16 = slice_by_index(begin = var_14708_begin_0, end = var_14708_end_0, end_mask = var_14708_end_mask_0, x = k_19_cast_fp16)[name = string("op_14708_cast_fp16")]; + tensor var_14712_begin_0 = const()[name = string("op_14712_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14712_end_0 = const()[name = string("op_14712_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_14712_end_mask_0 = const()[name = string("op_14712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14712_cast_fp16 = slice_by_index(begin = var_14712_begin_0, end = var_14712_end_0, end_mask = var_14712_end_mask_0, x = k_19_cast_fp16)[name = string("op_14712_cast_fp16")]; + tensor var_14716_begin_0 = const()[name = string("op_14716_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14716_end_0 = const()[name = string("op_14716_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_14716_end_mask_0 = const()[name = string("op_14716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14716_cast_fp16 = slice_by_index(begin = var_14716_begin_0, end = var_14716_end_0, end_mask = var_14716_end_mask_0, x = k_19_cast_fp16)[name = string("op_14716_cast_fp16")]; + tensor var_14720_begin_0 = const()[name = string("op_14720_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14720_end_0 = const()[name = string("op_14720_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_14720_end_mask_0 = const()[name = string("op_14720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14720_cast_fp16 = slice_by_index(begin = var_14720_begin_0, end = var_14720_end_0, end_mask = var_14720_end_mask_0, x = k_19_cast_fp16)[name = string("op_14720_cast_fp16")]; + tensor var_14722_begin_0 = const()[name = string("op_14722_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14722_end_0 = const()[name = string("op_14722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14722_end_mask_0 = const()[name = string("op_14722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14722_cast_fp16 = slice_by_index(begin = var_14722_begin_0, end = var_14722_end_0, end_mask = var_14722_end_mask_0, x = value_19_cast_fp16)[name = string("op_14722_cast_fp16")]; + tensor var_14726_begin_0 = const()[name = string("op_14726_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14726_end_0 = const()[name = string("op_14726_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_14726_end_mask_0 = const()[name = string("op_14726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14726_cast_fp16 = slice_by_index(begin = var_14726_begin_0, end = var_14726_end_0, end_mask = var_14726_end_mask_0, x = value_19_cast_fp16)[name = string("op_14726_cast_fp16")]; + tensor var_14730_begin_0 = const()[name = string("op_14730_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14730_end_0 = const()[name = string("op_14730_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_14730_end_mask_0 = const()[name = string("op_14730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14730_cast_fp16 = slice_by_index(begin = var_14730_begin_0, end = var_14730_end_0, end_mask = var_14730_end_mask_0, x = value_19_cast_fp16)[name = string("op_14730_cast_fp16")]; + tensor var_14734_begin_0 = const()[name = string("op_14734_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14734_end_0 = const()[name = string("op_14734_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_14734_end_mask_0 = const()[name = string("op_14734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14734_cast_fp16 = slice_by_index(begin = var_14734_begin_0, end = var_14734_end_0, end_mask = var_14734_end_mask_0, x = value_19_cast_fp16)[name = string("op_14734_cast_fp16")]; + tensor var_14738_begin_0 = const()[name = string("op_14738_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14738_end_0 = const()[name = string("op_14738_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_14738_end_mask_0 = const()[name = string("op_14738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14738_cast_fp16 = slice_by_index(begin = var_14738_begin_0, end = var_14738_end_0, end_mask = var_14738_end_mask_0, x = value_19_cast_fp16)[name = string("op_14738_cast_fp16")]; + tensor var_14742_begin_0 = const()[name = string("op_14742_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14742_end_0 = const()[name = string("op_14742_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_14742_end_mask_0 = const()[name = string("op_14742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14742_cast_fp16 = slice_by_index(begin = var_14742_begin_0, end = var_14742_end_0, end_mask = var_14742_end_mask_0, x = value_19_cast_fp16)[name = string("op_14742_cast_fp16")]; + tensor var_14746_begin_0 = const()[name = string("op_14746_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14746_end_0 = const()[name = string("op_14746_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_14746_end_mask_0 = const()[name = string("op_14746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14746_cast_fp16 = slice_by_index(begin = var_14746_begin_0, end = var_14746_end_0, end_mask = var_14746_end_mask_0, x = value_19_cast_fp16)[name = string("op_14746_cast_fp16")]; + tensor var_14750_begin_0 = const()[name = string("op_14750_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14750_end_0 = const()[name = string("op_14750_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_14750_end_mask_0 = const()[name = string("op_14750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14750_cast_fp16 = slice_by_index(begin = var_14750_begin_0, end = var_14750_end_0, end_mask = var_14750_end_mask_0, x = value_19_cast_fp16)[name = string("op_14750_cast_fp16")]; + tensor var_14754_begin_0 = const()[name = string("op_14754_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14754_end_0 = const()[name = string("op_14754_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_14754_end_mask_0 = const()[name = string("op_14754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14754_cast_fp16 = slice_by_index(begin = var_14754_begin_0, end = var_14754_end_0, end_mask = var_14754_end_mask_0, x = value_19_cast_fp16)[name = string("op_14754_cast_fp16")]; + tensor var_14758_begin_0 = const()[name = string("op_14758_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14758_end_0 = const()[name = string("op_14758_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_14758_end_mask_0 = const()[name = string("op_14758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14758_cast_fp16 = slice_by_index(begin = var_14758_begin_0, end = var_14758_end_0, end_mask = var_14758_end_mask_0, x = value_19_cast_fp16)[name = string("op_14758_cast_fp16")]; + tensor var_14762_begin_0 = const()[name = string("op_14762_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14762_end_0 = const()[name = string("op_14762_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_14762_end_mask_0 = const()[name = string("op_14762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14762_cast_fp16 = slice_by_index(begin = var_14762_begin_0, end = var_14762_end_0, end_mask = var_14762_end_mask_0, x = value_19_cast_fp16)[name = string("op_14762_cast_fp16")]; + tensor var_14766_begin_0 = const()[name = string("op_14766_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14766_end_0 = const()[name = string("op_14766_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_14766_end_mask_0 = const()[name = string("op_14766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14766_cast_fp16 = slice_by_index(begin = var_14766_begin_0, end = var_14766_end_0, end_mask = var_14766_end_mask_0, x = value_19_cast_fp16)[name = string("op_14766_cast_fp16")]; + tensor var_14770_begin_0 = const()[name = string("op_14770_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14770_end_0 = const()[name = string("op_14770_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_14770_end_mask_0 = const()[name = string("op_14770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14770_cast_fp16 = slice_by_index(begin = var_14770_begin_0, end = var_14770_end_0, end_mask = var_14770_end_mask_0, x = value_19_cast_fp16)[name = string("op_14770_cast_fp16")]; + tensor var_14774_begin_0 = const()[name = string("op_14774_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14774_end_0 = const()[name = string("op_14774_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_14774_end_mask_0 = const()[name = string("op_14774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14774_cast_fp16 = slice_by_index(begin = var_14774_begin_0, end = var_14774_end_0, end_mask = var_14774_end_mask_0, x = value_19_cast_fp16)[name = string("op_14774_cast_fp16")]; + tensor var_14778_begin_0 = const()[name = string("op_14778_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14778_end_0 = const()[name = string("op_14778_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_14778_end_mask_0 = const()[name = string("op_14778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14778_cast_fp16 = slice_by_index(begin = var_14778_begin_0, end = var_14778_end_0, end_mask = var_14778_end_mask_0, x = value_19_cast_fp16)[name = string("op_14778_cast_fp16")]; + tensor var_14782_begin_0 = const()[name = string("op_14782_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14782_end_0 = const()[name = string("op_14782_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_14782_end_mask_0 = const()[name = string("op_14782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14782_cast_fp16 = slice_by_index(begin = var_14782_begin_0, end = var_14782_end_0, end_mask = var_14782_end_mask_0, x = value_19_cast_fp16)[name = string("op_14782_cast_fp16")]; + tensor var_14786_begin_0 = const()[name = string("op_14786_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14786_end_0 = const()[name = string("op_14786_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_14786_end_mask_0 = const()[name = string("op_14786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14786_cast_fp16 = slice_by_index(begin = var_14786_begin_0, end = var_14786_end_0, end_mask = var_14786_end_mask_0, x = value_19_cast_fp16)[name = string("op_14786_cast_fp16")]; + tensor var_14790_begin_0 = const()[name = string("op_14790_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14790_end_0 = const()[name = string("op_14790_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_14790_end_mask_0 = const()[name = string("op_14790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14790_cast_fp16 = slice_by_index(begin = var_14790_begin_0, end = var_14790_end_0, end_mask = var_14790_end_mask_0, x = value_19_cast_fp16)[name = string("op_14790_cast_fp16")]; + tensor var_14794_begin_0 = const()[name = string("op_14794_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14794_end_0 = const()[name = string("op_14794_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_14794_end_mask_0 = const()[name = string("op_14794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14794_cast_fp16 = slice_by_index(begin = var_14794_begin_0, end = var_14794_end_0, end_mask = var_14794_end_mask_0, x = value_19_cast_fp16)[name = string("op_14794_cast_fp16")]; + tensor var_14798_begin_0 = const()[name = string("op_14798_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14798_end_0 = const()[name = string("op_14798_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_14798_end_mask_0 = const()[name = string("op_14798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14798_cast_fp16 = slice_by_index(begin = var_14798_begin_0, end = var_14798_end_0, end_mask = var_14798_end_mask_0, x = value_19_cast_fp16)[name = string("op_14798_cast_fp16")]; + string _SplitHeadsQ__mh_w_1441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1441_equation_0, values = (var_14644_cast_fp16, var_14086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1441_cast_fp16")]; + string _SplitHeadsQ__mh_w_1443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1443_equation_0, values = (var_14644_cast_fp16, var_14093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1443_cast_fp16")]; + string _SplitHeadsQ__mh_w_1445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1445_equation_0, values = (var_14644_cast_fp16, var_14100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1445_cast_fp16")]; + string _SplitHeadsQ__mh_w_1447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1447_equation_0, values = (var_14644_cast_fp16, var_14107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1447_cast_fp16")]; + string _SplitHeadsQ__mh_w_1449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1449_equation_0, values = (var_14648_cast_fp16, var_14114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1449_cast_fp16")]; + string _SplitHeadsQ__mh_w_1451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1451_equation_0, values = (var_14648_cast_fp16, var_14121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1451_cast_fp16")]; + string _SplitHeadsQ__mh_w_1453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1453_equation_0, values = (var_14648_cast_fp16, var_14128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1453_cast_fp16")]; + string _SplitHeadsQ__mh_w_1455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1455_equation_0, values = (var_14648_cast_fp16, var_14135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1455_cast_fp16")]; + string _SplitHeadsQ__mh_w_1457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1457_equation_0, values = (var_14652_cast_fp16, var_14142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1457_cast_fp16")]; + string _SplitHeadsQ__mh_w_1459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1459_equation_0, values = (var_14652_cast_fp16, var_14149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1459_cast_fp16")]; + string _SplitHeadsQ__mh_w_1461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1461_equation_0, values = (var_14652_cast_fp16, var_14156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1461_cast_fp16")]; + string _SplitHeadsQ__mh_w_1463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1463_equation_0, values = (var_14652_cast_fp16, var_14163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1463_cast_fp16")]; + string _SplitHeadsQ__mh_w_1465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1465_equation_0, values = (var_14656_cast_fp16, var_14170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1465_cast_fp16")]; + string _SplitHeadsQ__mh_w_1467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1467_equation_0, values = (var_14656_cast_fp16, var_14177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1467_cast_fp16")]; + string _SplitHeadsQ__mh_w_1469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1469_equation_0, values = (var_14656_cast_fp16, var_14184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1469_cast_fp16")]; + string _SplitHeadsQ__mh_w_1471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1471_equation_0, values = (var_14656_cast_fp16, var_14191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1471_cast_fp16")]; + string _SplitHeadsQ__mh_w_1473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1473_equation_0, values = (var_14660_cast_fp16, var_14198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1473_cast_fp16")]; + string _SplitHeadsQ__mh_w_1475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1475_equation_0, values = (var_14660_cast_fp16, var_14205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1475_cast_fp16")]; + string _SplitHeadsQ__mh_w_1477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1477_equation_0, values = (var_14660_cast_fp16, var_14212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1477_cast_fp16")]; + string _SplitHeadsQ__mh_w_1479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1479_equation_0, values = (var_14660_cast_fp16, var_14219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1479_cast_fp16")]; + string _SplitHeadsQ__mh_w_1481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1481_equation_0, values = (var_14664_cast_fp16, var_14226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1481_cast_fp16")]; + string _SplitHeadsQ__mh_w_1483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1483_equation_0, values = (var_14664_cast_fp16, var_14233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1483_cast_fp16")]; + string _SplitHeadsQ__mh_w_1485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1485_equation_0, values = (var_14664_cast_fp16, var_14240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1485_cast_fp16")]; + string _SplitHeadsQ__mh_w_1487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1487_equation_0, values = (var_14664_cast_fp16, var_14247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1487_cast_fp16")]; + string _SplitHeadsQ__mh_w_1489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1489_equation_0, values = (var_14668_cast_fp16, var_14254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1489_cast_fp16")]; + string _SplitHeadsQ__mh_w_1491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1491_equation_0, values = (var_14668_cast_fp16, var_14261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1491_cast_fp16")]; + string _SplitHeadsQ__mh_w_1493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1493_equation_0, values = (var_14668_cast_fp16, var_14268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1493_cast_fp16")]; + string _SplitHeadsQ__mh_w_1495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1495_equation_0, values = (var_14668_cast_fp16, var_14275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1495_cast_fp16")]; + string _SplitHeadsQ__mh_w_1497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1497_equation_0, values = (var_14672_cast_fp16, var_14282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1497_cast_fp16")]; + string _SplitHeadsQ__mh_w_1499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1499_equation_0, values = (var_14672_cast_fp16, var_14289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1499_cast_fp16")]; + string _SplitHeadsQ__mh_w_1501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1501_equation_0, values = (var_14672_cast_fp16, var_14296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1501_cast_fp16")]; + string _SplitHeadsQ__mh_w_1503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1503_equation_0, values = (var_14672_cast_fp16, var_14303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1503_cast_fp16")]; + string _SplitHeadsQ__mh_w_1505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1505_equation_0, values = (var_14676_cast_fp16, var_14310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1505_cast_fp16")]; + string _SplitHeadsQ__mh_w_1507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1507_equation_0, values = (var_14676_cast_fp16, var_14317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1507_cast_fp16")]; + string _SplitHeadsQ__mh_w_1509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1509_equation_0, values = (var_14676_cast_fp16, var_14324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1509_cast_fp16")]; + string _SplitHeadsQ__mh_w_1511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1511_equation_0, values = (var_14676_cast_fp16, var_14331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1511_cast_fp16")]; + string _SplitHeadsQ__mh_w_1513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1513_equation_0, values = (var_14680_cast_fp16, var_14338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1513_cast_fp16")]; + string _SplitHeadsQ__mh_w_1515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1515_equation_0, values = (var_14680_cast_fp16, var_14345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1515_cast_fp16")]; + string _SplitHeadsQ__mh_w_1517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1517_equation_0, values = (var_14680_cast_fp16, var_14352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1517_cast_fp16")]; + string _SplitHeadsQ__mh_w_1519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1519_equation_0, values = (var_14680_cast_fp16, var_14359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1519_cast_fp16")]; + string _SplitHeadsQ__mh_w_1521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1521_equation_0, values = (var_14684_cast_fp16, var_14366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1521_cast_fp16")]; + string _SplitHeadsQ__mh_w_1523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1523_equation_0, values = (var_14684_cast_fp16, var_14373_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1523_cast_fp16")]; + string _SplitHeadsQ__mh_w_1525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1525_equation_0, values = (var_14684_cast_fp16, var_14380_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1525_cast_fp16")]; + string _SplitHeadsQ__mh_w_1527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1527_equation_0, values = (var_14684_cast_fp16, var_14387_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1527_cast_fp16")]; + string _SplitHeadsQ__mh_w_1529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1529_equation_0, values = (var_14688_cast_fp16, var_14394_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1529_cast_fp16")]; + string _SplitHeadsQ__mh_w_1531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1531_equation_0, values = (var_14688_cast_fp16, var_14401_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1531_cast_fp16")]; + string _SplitHeadsQ__mh_w_1533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1533_equation_0, values = (var_14688_cast_fp16, var_14408_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1533_cast_fp16")]; + string _SplitHeadsQ__mh_w_1535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1535_equation_0, values = (var_14688_cast_fp16, var_14415_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1535_cast_fp16")]; + string _SplitHeadsQ__mh_w_1537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1537_equation_0, values = (var_14692_cast_fp16, var_14422_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1537_cast_fp16")]; + string _SplitHeadsQ__mh_w_1539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1539_equation_0, values = (var_14692_cast_fp16, var_14429_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1539_cast_fp16")]; + string _SplitHeadsQ__mh_w_1541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1541_equation_0, values = (var_14692_cast_fp16, var_14436_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1541_cast_fp16")]; + string _SplitHeadsQ__mh_w_1543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1543_equation_0, values = (var_14692_cast_fp16, var_14443_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1543_cast_fp16")]; + string _SplitHeadsQ__mh_w_1545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1545_equation_0, values = (var_14696_cast_fp16, var_14450_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1545_cast_fp16")]; + string _SplitHeadsQ__mh_w_1547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1547_equation_0, values = (var_14696_cast_fp16, var_14457_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1547_cast_fp16")]; + string _SplitHeadsQ__mh_w_1549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1549_equation_0, values = (var_14696_cast_fp16, var_14464_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1549_cast_fp16")]; + string _SplitHeadsQ__mh_w_1551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1551_equation_0, values = (var_14696_cast_fp16, var_14471_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1551_cast_fp16")]; + string _SplitHeadsQ__mh_w_1553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1553_equation_0, values = (var_14700_cast_fp16, var_14478_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1553_cast_fp16")]; + string _SplitHeadsQ__mh_w_1555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1555_equation_0, values = (var_14700_cast_fp16, var_14485_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1555_cast_fp16")]; + string _SplitHeadsQ__mh_w_1557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1557_equation_0, values = (var_14700_cast_fp16, var_14492_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1557_cast_fp16")]; + string _SplitHeadsQ__mh_w_1559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1559_equation_0, values = (var_14700_cast_fp16, var_14499_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1559_cast_fp16")]; + string _SplitHeadsQ__mh_w_1561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1561_equation_0, values = (var_14704_cast_fp16, var_14506_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1561_cast_fp16")]; + string _SplitHeadsQ__mh_w_1563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1563_equation_0, values = (var_14704_cast_fp16, var_14513_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1563_cast_fp16")]; + string _SplitHeadsQ__mh_w_1565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1565_equation_0, values = (var_14704_cast_fp16, var_14520_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1565_cast_fp16")]; + string _SplitHeadsQ__mh_w_1567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1567_equation_0, values = (var_14704_cast_fp16, var_14527_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1567_cast_fp16")]; + string _SplitHeadsQ__mh_w_1569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1569_equation_0, values = (var_14708_cast_fp16, var_14534_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1569_cast_fp16")]; + string _SplitHeadsQ__mh_w_1571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1571_equation_0, values = (var_14708_cast_fp16, var_14541_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1571_cast_fp16")]; + string _SplitHeadsQ__mh_w_1573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1573_equation_0, values = (var_14708_cast_fp16, var_14548_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1573_cast_fp16")]; + string _SplitHeadsQ__mh_w_1575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1575_equation_0, values = (var_14708_cast_fp16, var_14555_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1575_cast_fp16")]; + string _SplitHeadsQ__mh_w_1577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1577_equation_0, values = (var_14712_cast_fp16, var_14562_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1577_cast_fp16")]; + string _SplitHeadsQ__mh_w_1579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1579_equation_0, values = (var_14712_cast_fp16, var_14569_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1579_cast_fp16")]; + string _SplitHeadsQ__mh_w_1581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1581_equation_0, values = (var_14712_cast_fp16, var_14576_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1581_cast_fp16")]; + string _SplitHeadsQ__mh_w_1583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1583_equation_0, values = (var_14712_cast_fp16, var_14583_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1583_cast_fp16")]; + string _SplitHeadsQ__mh_w_1585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1585_equation_0, values = (var_14716_cast_fp16, var_14590_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1585_cast_fp16")]; + string _SplitHeadsQ__mh_w_1587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1587_equation_0, values = (var_14716_cast_fp16, var_14597_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1587_cast_fp16")]; + string _SplitHeadsQ__mh_w_1589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1589_equation_0, values = (var_14716_cast_fp16, var_14604_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1589_cast_fp16")]; + string _SplitHeadsQ__mh_w_1591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1591_equation_0, values = (var_14716_cast_fp16, var_14611_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1591_cast_fp16")]; + string _SplitHeadsQ__mh_w_1593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1593_equation_0, values = (var_14720_cast_fp16, var_14618_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1593_cast_fp16")]; + string _SplitHeadsQ__mh_w_1595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1595_equation_0, values = (var_14720_cast_fp16, var_14625_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1595_cast_fp16")]; + string _SplitHeadsQ__mh_w_1597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1597_equation_0, values = (var_14720_cast_fp16, var_14632_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1597_cast_fp16")]; + string _SplitHeadsQ__mh_w_1599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1599_equation_0, values = (var_14720_cast_fp16, var_14639_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1599_cast_fp16")]; + fp16 var_14961_to_fp16 = const()[name = string("op_14961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1441_cast_fp16, y = var_14961_to_fp16)[name = string("aw_chunk_1441_cast_fp16")]; + fp16 var_14963_to_fp16 = const()[name = string("op_14963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1443_cast_fp16, y = var_14963_to_fp16)[name = string("aw_chunk_1443_cast_fp16")]; + fp16 var_14965_to_fp16 = const()[name = string("op_14965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1445_cast_fp16, y = var_14965_to_fp16)[name = string("aw_chunk_1445_cast_fp16")]; + fp16 var_14967_to_fp16 = const()[name = string("op_14967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1447_cast_fp16, y = var_14967_to_fp16)[name = string("aw_chunk_1447_cast_fp16")]; + fp16 var_14969_to_fp16 = const()[name = string("op_14969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1449_cast_fp16, y = var_14969_to_fp16)[name = string("aw_chunk_1449_cast_fp16")]; + fp16 var_14971_to_fp16 = const()[name = string("op_14971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1451_cast_fp16, y = var_14971_to_fp16)[name = string("aw_chunk_1451_cast_fp16")]; + fp16 var_14973_to_fp16 = const()[name = string("op_14973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1453_cast_fp16, y = var_14973_to_fp16)[name = string("aw_chunk_1453_cast_fp16")]; + fp16 var_14975_to_fp16 = const()[name = string("op_14975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1455_cast_fp16, y = var_14975_to_fp16)[name = string("aw_chunk_1455_cast_fp16")]; + fp16 var_14977_to_fp16 = const()[name = string("op_14977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1457_cast_fp16, y = var_14977_to_fp16)[name = string("aw_chunk_1457_cast_fp16")]; + fp16 var_14979_to_fp16 = const()[name = string("op_14979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1459_cast_fp16, y = var_14979_to_fp16)[name = string("aw_chunk_1459_cast_fp16")]; + fp16 var_14981_to_fp16 = const()[name = string("op_14981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1461_cast_fp16, y = var_14981_to_fp16)[name = string("aw_chunk_1461_cast_fp16")]; + fp16 var_14983_to_fp16 = const()[name = string("op_14983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1463_cast_fp16, y = var_14983_to_fp16)[name = string("aw_chunk_1463_cast_fp16")]; + fp16 var_14985_to_fp16 = const()[name = string("op_14985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1465_cast_fp16, y = var_14985_to_fp16)[name = string("aw_chunk_1465_cast_fp16")]; + fp16 var_14987_to_fp16 = const()[name = string("op_14987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1467_cast_fp16, y = var_14987_to_fp16)[name = string("aw_chunk_1467_cast_fp16")]; + fp16 var_14989_to_fp16 = const()[name = string("op_14989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1469_cast_fp16, y = var_14989_to_fp16)[name = string("aw_chunk_1469_cast_fp16")]; + fp16 var_14991_to_fp16 = const()[name = string("op_14991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1471_cast_fp16, y = var_14991_to_fp16)[name = string("aw_chunk_1471_cast_fp16")]; + fp16 var_14993_to_fp16 = const()[name = string("op_14993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1473_cast_fp16, y = var_14993_to_fp16)[name = string("aw_chunk_1473_cast_fp16")]; + fp16 var_14995_to_fp16 = const()[name = string("op_14995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1475_cast_fp16, y = var_14995_to_fp16)[name = string("aw_chunk_1475_cast_fp16")]; + fp16 var_14997_to_fp16 = const()[name = string("op_14997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1477_cast_fp16, y = var_14997_to_fp16)[name = string("aw_chunk_1477_cast_fp16")]; + fp16 var_14999_to_fp16 = const()[name = string("op_14999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1479_cast_fp16, y = var_14999_to_fp16)[name = string("aw_chunk_1479_cast_fp16")]; + fp16 var_15001_to_fp16 = const()[name = string("op_15001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1481_cast_fp16, y = var_15001_to_fp16)[name = string("aw_chunk_1481_cast_fp16")]; + fp16 var_15003_to_fp16 = const()[name = string("op_15003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1483_cast_fp16, y = var_15003_to_fp16)[name = string("aw_chunk_1483_cast_fp16")]; + fp16 var_15005_to_fp16 = const()[name = string("op_15005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1485_cast_fp16, y = var_15005_to_fp16)[name = string("aw_chunk_1485_cast_fp16")]; + fp16 var_15007_to_fp16 = const()[name = string("op_15007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1487_cast_fp16, y = var_15007_to_fp16)[name = string("aw_chunk_1487_cast_fp16")]; + fp16 var_15009_to_fp16 = const()[name = string("op_15009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1489_cast_fp16, y = var_15009_to_fp16)[name = string("aw_chunk_1489_cast_fp16")]; + fp16 var_15011_to_fp16 = const()[name = string("op_15011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1491_cast_fp16, y = var_15011_to_fp16)[name = string("aw_chunk_1491_cast_fp16")]; + fp16 var_15013_to_fp16 = const()[name = string("op_15013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1493_cast_fp16, y = var_15013_to_fp16)[name = string("aw_chunk_1493_cast_fp16")]; + fp16 var_15015_to_fp16 = const()[name = string("op_15015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1495_cast_fp16, y = var_15015_to_fp16)[name = string("aw_chunk_1495_cast_fp16")]; + fp16 var_15017_to_fp16 = const()[name = string("op_15017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1497_cast_fp16, y = var_15017_to_fp16)[name = string("aw_chunk_1497_cast_fp16")]; + fp16 var_15019_to_fp16 = const()[name = string("op_15019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1499_cast_fp16, y = var_15019_to_fp16)[name = string("aw_chunk_1499_cast_fp16")]; + fp16 var_15021_to_fp16 = const()[name = string("op_15021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1501_cast_fp16, y = var_15021_to_fp16)[name = string("aw_chunk_1501_cast_fp16")]; + fp16 var_15023_to_fp16 = const()[name = string("op_15023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1503_cast_fp16, y = var_15023_to_fp16)[name = string("aw_chunk_1503_cast_fp16")]; + fp16 var_15025_to_fp16 = const()[name = string("op_15025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1505_cast_fp16, y = var_15025_to_fp16)[name = string("aw_chunk_1505_cast_fp16")]; + fp16 var_15027_to_fp16 = const()[name = string("op_15027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1507_cast_fp16, y = var_15027_to_fp16)[name = string("aw_chunk_1507_cast_fp16")]; + fp16 var_15029_to_fp16 = const()[name = string("op_15029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1509_cast_fp16, y = var_15029_to_fp16)[name = string("aw_chunk_1509_cast_fp16")]; + fp16 var_15031_to_fp16 = const()[name = string("op_15031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1511_cast_fp16, y = var_15031_to_fp16)[name = string("aw_chunk_1511_cast_fp16")]; + fp16 var_15033_to_fp16 = const()[name = string("op_15033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1513_cast_fp16, y = var_15033_to_fp16)[name = string("aw_chunk_1513_cast_fp16")]; + fp16 var_15035_to_fp16 = const()[name = string("op_15035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1515_cast_fp16, y = var_15035_to_fp16)[name = string("aw_chunk_1515_cast_fp16")]; + fp16 var_15037_to_fp16 = const()[name = string("op_15037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1517_cast_fp16, y = var_15037_to_fp16)[name = string("aw_chunk_1517_cast_fp16")]; + fp16 var_15039_to_fp16 = const()[name = string("op_15039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1519_cast_fp16, y = var_15039_to_fp16)[name = string("aw_chunk_1519_cast_fp16")]; + fp16 var_15041_to_fp16 = const()[name = string("op_15041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1521_cast_fp16, y = var_15041_to_fp16)[name = string("aw_chunk_1521_cast_fp16")]; + fp16 var_15043_to_fp16 = const()[name = string("op_15043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1523_cast_fp16, y = var_15043_to_fp16)[name = string("aw_chunk_1523_cast_fp16")]; + fp16 var_15045_to_fp16 = const()[name = string("op_15045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1525_cast_fp16, y = var_15045_to_fp16)[name = string("aw_chunk_1525_cast_fp16")]; + fp16 var_15047_to_fp16 = const()[name = string("op_15047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1527_cast_fp16, y = var_15047_to_fp16)[name = string("aw_chunk_1527_cast_fp16")]; + fp16 var_15049_to_fp16 = const()[name = string("op_15049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1529_cast_fp16, y = var_15049_to_fp16)[name = string("aw_chunk_1529_cast_fp16")]; + fp16 var_15051_to_fp16 = const()[name = string("op_15051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1531_cast_fp16, y = var_15051_to_fp16)[name = string("aw_chunk_1531_cast_fp16")]; + fp16 var_15053_to_fp16 = const()[name = string("op_15053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1533_cast_fp16, y = var_15053_to_fp16)[name = string("aw_chunk_1533_cast_fp16")]; + fp16 var_15055_to_fp16 = const()[name = string("op_15055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1535_cast_fp16, y = var_15055_to_fp16)[name = string("aw_chunk_1535_cast_fp16")]; + fp16 var_15057_to_fp16 = const()[name = string("op_15057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1537_cast_fp16, y = var_15057_to_fp16)[name = string("aw_chunk_1537_cast_fp16")]; + fp16 var_15059_to_fp16 = const()[name = string("op_15059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1539_cast_fp16, y = var_15059_to_fp16)[name = string("aw_chunk_1539_cast_fp16")]; + fp16 var_15061_to_fp16 = const()[name = string("op_15061_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1541_cast_fp16, y = var_15061_to_fp16)[name = string("aw_chunk_1541_cast_fp16")]; + fp16 var_15063_to_fp16 = const()[name = string("op_15063_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1543_cast_fp16, y = var_15063_to_fp16)[name = string("aw_chunk_1543_cast_fp16")]; + fp16 var_15065_to_fp16 = const()[name = string("op_15065_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1545_cast_fp16, y = var_15065_to_fp16)[name = string("aw_chunk_1545_cast_fp16")]; + fp16 var_15067_to_fp16 = const()[name = string("op_15067_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1547_cast_fp16, y = var_15067_to_fp16)[name = string("aw_chunk_1547_cast_fp16")]; + fp16 var_15069_to_fp16 = const()[name = string("op_15069_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1549_cast_fp16, y = var_15069_to_fp16)[name = string("aw_chunk_1549_cast_fp16")]; + fp16 var_15071_to_fp16 = const()[name = string("op_15071_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1551_cast_fp16, y = var_15071_to_fp16)[name = string("aw_chunk_1551_cast_fp16")]; + fp16 var_15073_to_fp16 = const()[name = string("op_15073_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1553_cast_fp16, y = var_15073_to_fp16)[name = string("aw_chunk_1553_cast_fp16")]; + fp16 var_15075_to_fp16 = const()[name = string("op_15075_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1555_cast_fp16, y = var_15075_to_fp16)[name = string("aw_chunk_1555_cast_fp16")]; + fp16 var_15077_to_fp16 = const()[name = string("op_15077_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1557_cast_fp16, y = var_15077_to_fp16)[name = string("aw_chunk_1557_cast_fp16")]; + fp16 var_15079_to_fp16 = const()[name = string("op_15079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1559_cast_fp16, y = var_15079_to_fp16)[name = string("aw_chunk_1559_cast_fp16")]; + fp16 var_15081_to_fp16 = const()[name = string("op_15081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1561_cast_fp16, y = var_15081_to_fp16)[name = string("aw_chunk_1561_cast_fp16")]; + fp16 var_15083_to_fp16 = const()[name = string("op_15083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1563_cast_fp16, y = var_15083_to_fp16)[name = string("aw_chunk_1563_cast_fp16")]; + fp16 var_15085_to_fp16 = const()[name = string("op_15085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1565_cast_fp16, y = var_15085_to_fp16)[name = string("aw_chunk_1565_cast_fp16")]; + fp16 var_15087_to_fp16 = const()[name = string("op_15087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1567_cast_fp16, y = var_15087_to_fp16)[name = string("aw_chunk_1567_cast_fp16")]; + fp16 var_15089_to_fp16 = const()[name = string("op_15089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1569_cast_fp16, y = var_15089_to_fp16)[name = string("aw_chunk_1569_cast_fp16")]; + fp16 var_15091_to_fp16 = const()[name = string("op_15091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1571_cast_fp16, y = var_15091_to_fp16)[name = string("aw_chunk_1571_cast_fp16")]; + fp16 var_15093_to_fp16 = const()[name = string("op_15093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1573_cast_fp16, y = var_15093_to_fp16)[name = string("aw_chunk_1573_cast_fp16")]; + fp16 var_15095_to_fp16 = const()[name = string("op_15095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1575_cast_fp16, y = var_15095_to_fp16)[name = string("aw_chunk_1575_cast_fp16")]; + fp16 var_15097_to_fp16 = const()[name = string("op_15097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1577_cast_fp16, y = var_15097_to_fp16)[name = string("aw_chunk_1577_cast_fp16")]; + fp16 var_15099_to_fp16 = const()[name = string("op_15099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1579_cast_fp16, y = var_15099_to_fp16)[name = string("aw_chunk_1579_cast_fp16")]; + fp16 var_15101_to_fp16 = const()[name = string("op_15101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1581_cast_fp16, y = var_15101_to_fp16)[name = string("aw_chunk_1581_cast_fp16")]; + fp16 var_15103_to_fp16 = const()[name = string("op_15103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1583_cast_fp16, y = var_15103_to_fp16)[name = string("aw_chunk_1583_cast_fp16")]; + fp16 var_15105_to_fp16 = const()[name = string("op_15105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1585_cast_fp16, y = var_15105_to_fp16)[name = string("aw_chunk_1585_cast_fp16")]; + fp16 var_15107_to_fp16 = const()[name = string("op_15107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1587_cast_fp16, y = var_15107_to_fp16)[name = string("aw_chunk_1587_cast_fp16")]; + fp16 var_15109_to_fp16 = const()[name = string("op_15109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1589_cast_fp16, y = var_15109_to_fp16)[name = string("aw_chunk_1589_cast_fp16")]; + fp16 var_15111_to_fp16 = const()[name = string("op_15111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1591_cast_fp16, y = var_15111_to_fp16)[name = string("aw_chunk_1591_cast_fp16")]; + fp16 var_15113_to_fp16 = const()[name = string("op_15113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1593_cast_fp16, y = var_15113_to_fp16)[name = string("aw_chunk_1593_cast_fp16")]; + fp16 var_15115_to_fp16 = const()[name = string("op_15115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1595_cast_fp16, y = var_15115_to_fp16)[name = string("aw_chunk_1595_cast_fp16")]; + fp16 var_15117_to_fp16 = const()[name = string("op_15117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1597_cast_fp16, y = var_15117_to_fp16)[name = string("aw_chunk_1597_cast_fp16")]; + fp16 var_15119_to_fp16 = const()[name = string("op_15119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1599_cast_fp16, y = var_15119_to_fp16)[name = string("aw_chunk_1599_cast_fp16")]; + tensor var_15121_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1441_cast_fp16)[name = string("op_15121_cast_fp16")]; + tensor var_15122_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1443_cast_fp16)[name = string("op_15122_cast_fp16")]; + tensor var_15123_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1445_cast_fp16)[name = string("op_15123_cast_fp16")]; + tensor var_15124_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1447_cast_fp16)[name = string("op_15124_cast_fp16")]; + tensor var_15125_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1449_cast_fp16)[name = string("op_15125_cast_fp16")]; + tensor var_15126_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1451_cast_fp16)[name = string("op_15126_cast_fp16")]; + tensor var_15127_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1453_cast_fp16)[name = string("op_15127_cast_fp16")]; + tensor var_15128_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1455_cast_fp16)[name = string("op_15128_cast_fp16")]; + tensor var_15129_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1457_cast_fp16)[name = string("op_15129_cast_fp16")]; + tensor var_15130_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1459_cast_fp16)[name = string("op_15130_cast_fp16")]; + tensor var_15131_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1461_cast_fp16)[name = string("op_15131_cast_fp16")]; + tensor var_15132_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1463_cast_fp16)[name = string("op_15132_cast_fp16")]; + tensor var_15133_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1465_cast_fp16)[name = string("op_15133_cast_fp16")]; + tensor var_15134_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1467_cast_fp16)[name = string("op_15134_cast_fp16")]; + tensor var_15135_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1469_cast_fp16)[name = string("op_15135_cast_fp16")]; + tensor var_15136_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1471_cast_fp16)[name = string("op_15136_cast_fp16")]; + tensor var_15137_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1473_cast_fp16)[name = string("op_15137_cast_fp16")]; + tensor var_15138_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1475_cast_fp16)[name = string("op_15138_cast_fp16")]; + tensor var_15139_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1477_cast_fp16)[name = string("op_15139_cast_fp16")]; + tensor var_15140_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1479_cast_fp16)[name = string("op_15140_cast_fp16")]; + tensor var_15141_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1481_cast_fp16)[name = string("op_15141_cast_fp16")]; + tensor var_15142_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1483_cast_fp16)[name = string("op_15142_cast_fp16")]; + tensor var_15143_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1485_cast_fp16)[name = string("op_15143_cast_fp16")]; + tensor var_15144_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1487_cast_fp16)[name = string("op_15144_cast_fp16")]; + tensor var_15145_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1489_cast_fp16)[name = string("op_15145_cast_fp16")]; + tensor var_15146_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1491_cast_fp16)[name = string("op_15146_cast_fp16")]; + tensor var_15147_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1493_cast_fp16)[name = string("op_15147_cast_fp16")]; + tensor var_15148_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1495_cast_fp16)[name = string("op_15148_cast_fp16")]; + tensor var_15149_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1497_cast_fp16)[name = string("op_15149_cast_fp16")]; + tensor var_15150_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1499_cast_fp16)[name = string("op_15150_cast_fp16")]; + tensor var_15151_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1501_cast_fp16)[name = string("op_15151_cast_fp16")]; + tensor var_15152_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1503_cast_fp16)[name = string("op_15152_cast_fp16")]; + tensor var_15153_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1505_cast_fp16)[name = string("op_15153_cast_fp16")]; + tensor var_15154_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1507_cast_fp16)[name = string("op_15154_cast_fp16")]; + tensor var_15155_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1509_cast_fp16)[name = string("op_15155_cast_fp16")]; + tensor var_15156_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1511_cast_fp16)[name = string("op_15156_cast_fp16")]; + tensor var_15157_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1513_cast_fp16)[name = string("op_15157_cast_fp16")]; + tensor var_15158_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1515_cast_fp16)[name = string("op_15158_cast_fp16")]; + tensor var_15159_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1517_cast_fp16)[name = string("op_15159_cast_fp16")]; + tensor var_15160_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1519_cast_fp16)[name = string("op_15160_cast_fp16")]; + tensor var_15161_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1521_cast_fp16)[name = string("op_15161_cast_fp16")]; + tensor var_15162_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1523_cast_fp16)[name = string("op_15162_cast_fp16")]; + tensor var_15163_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1525_cast_fp16)[name = string("op_15163_cast_fp16")]; + tensor var_15164_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1527_cast_fp16)[name = string("op_15164_cast_fp16")]; + tensor var_15165_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1529_cast_fp16)[name = string("op_15165_cast_fp16")]; + tensor var_15166_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1531_cast_fp16)[name = string("op_15166_cast_fp16")]; + tensor var_15167_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1533_cast_fp16)[name = string("op_15167_cast_fp16")]; + tensor var_15168_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1535_cast_fp16)[name = string("op_15168_cast_fp16")]; + tensor var_15169_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1537_cast_fp16)[name = string("op_15169_cast_fp16")]; + tensor var_15170_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1539_cast_fp16)[name = string("op_15170_cast_fp16")]; + tensor var_15171_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1541_cast_fp16)[name = string("op_15171_cast_fp16")]; + tensor var_15172_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1543_cast_fp16)[name = string("op_15172_cast_fp16")]; + tensor var_15173_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1545_cast_fp16)[name = string("op_15173_cast_fp16")]; + tensor var_15174_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1547_cast_fp16)[name = string("op_15174_cast_fp16")]; + tensor var_15175_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1549_cast_fp16)[name = string("op_15175_cast_fp16")]; + tensor var_15176_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1551_cast_fp16)[name = string("op_15176_cast_fp16")]; + tensor var_15177_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1553_cast_fp16)[name = string("op_15177_cast_fp16")]; + tensor var_15178_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1555_cast_fp16)[name = string("op_15178_cast_fp16")]; + tensor var_15179_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1557_cast_fp16)[name = string("op_15179_cast_fp16")]; + tensor var_15180_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1559_cast_fp16)[name = string("op_15180_cast_fp16")]; + tensor var_15181_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1561_cast_fp16)[name = string("op_15181_cast_fp16")]; + tensor var_15182_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1563_cast_fp16)[name = string("op_15182_cast_fp16")]; + tensor var_15183_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1565_cast_fp16)[name = string("op_15183_cast_fp16")]; + tensor var_15184_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1567_cast_fp16)[name = string("op_15184_cast_fp16")]; + tensor var_15185_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1569_cast_fp16)[name = string("op_15185_cast_fp16")]; + tensor var_15186_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1571_cast_fp16)[name = string("op_15186_cast_fp16")]; + tensor var_15187_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1573_cast_fp16)[name = string("op_15187_cast_fp16")]; + tensor var_15188_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1575_cast_fp16)[name = string("op_15188_cast_fp16")]; + tensor var_15189_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1577_cast_fp16)[name = string("op_15189_cast_fp16")]; + tensor var_15190_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1579_cast_fp16)[name = string("op_15190_cast_fp16")]; + tensor var_15191_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1581_cast_fp16)[name = string("op_15191_cast_fp16")]; + tensor var_15192_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1583_cast_fp16)[name = string("op_15192_cast_fp16")]; + tensor var_15193_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1585_cast_fp16)[name = string("op_15193_cast_fp16")]; + tensor var_15194_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1587_cast_fp16)[name = string("op_15194_cast_fp16")]; + tensor var_15195_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1589_cast_fp16)[name = string("op_15195_cast_fp16")]; + tensor var_15196_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1591_cast_fp16)[name = string("op_15196_cast_fp16")]; + tensor var_15197_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1593_cast_fp16)[name = string("op_15197_cast_fp16")]; + tensor var_15198_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1595_cast_fp16)[name = string("op_15198_cast_fp16")]; + tensor var_15199_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1597_cast_fp16)[name = string("op_15199_cast_fp16")]; + tensor var_15200_cast_fp16 = softmax(axis = var_13946, x = aw_chunk_1599_cast_fp16)[name = string("op_15200_cast_fp16")]; + string var_15202_equation_0 = const()[name = string("op_15202_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15202_cast_fp16 = einsum(equation = var_15202_equation_0, values = (var_14722_cast_fp16, var_15121_cast_fp16))[name = string("op_15202_cast_fp16")]; + string var_15204_equation_0 = const()[name = string("op_15204_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15204_cast_fp16 = einsum(equation = var_15204_equation_0, values = (var_14722_cast_fp16, var_15122_cast_fp16))[name = string("op_15204_cast_fp16")]; + string var_15206_equation_0 = const()[name = string("op_15206_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15206_cast_fp16 = einsum(equation = var_15206_equation_0, values = (var_14722_cast_fp16, var_15123_cast_fp16))[name = string("op_15206_cast_fp16")]; + string var_15208_equation_0 = const()[name = string("op_15208_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15208_cast_fp16 = einsum(equation = var_15208_equation_0, values = (var_14722_cast_fp16, var_15124_cast_fp16))[name = string("op_15208_cast_fp16")]; + string var_15210_equation_0 = const()[name = string("op_15210_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15210_cast_fp16 = einsum(equation = var_15210_equation_0, values = (var_14726_cast_fp16, var_15125_cast_fp16))[name = string("op_15210_cast_fp16")]; + string var_15212_equation_0 = const()[name = string("op_15212_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15212_cast_fp16 = einsum(equation = var_15212_equation_0, values = (var_14726_cast_fp16, var_15126_cast_fp16))[name = string("op_15212_cast_fp16")]; + string var_15214_equation_0 = const()[name = string("op_15214_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15214_cast_fp16 = einsum(equation = var_15214_equation_0, values = (var_14726_cast_fp16, var_15127_cast_fp16))[name = string("op_15214_cast_fp16")]; + string var_15216_equation_0 = const()[name = string("op_15216_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15216_cast_fp16 = einsum(equation = var_15216_equation_0, values = (var_14726_cast_fp16, var_15128_cast_fp16))[name = string("op_15216_cast_fp16")]; + string var_15218_equation_0 = const()[name = string("op_15218_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15218_cast_fp16 = einsum(equation = var_15218_equation_0, values = (var_14730_cast_fp16, var_15129_cast_fp16))[name = string("op_15218_cast_fp16")]; + string var_15220_equation_0 = const()[name = string("op_15220_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15220_cast_fp16 = einsum(equation = var_15220_equation_0, values = (var_14730_cast_fp16, var_15130_cast_fp16))[name = string("op_15220_cast_fp16")]; + string var_15222_equation_0 = const()[name = string("op_15222_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15222_cast_fp16 = einsum(equation = var_15222_equation_0, values = (var_14730_cast_fp16, var_15131_cast_fp16))[name = string("op_15222_cast_fp16")]; + string var_15224_equation_0 = const()[name = string("op_15224_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15224_cast_fp16 = einsum(equation = var_15224_equation_0, values = (var_14730_cast_fp16, var_15132_cast_fp16))[name = string("op_15224_cast_fp16")]; + string var_15226_equation_0 = const()[name = string("op_15226_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15226_cast_fp16 = einsum(equation = var_15226_equation_0, values = (var_14734_cast_fp16, var_15133_cast_fp16))[name = string("op_15226_cast_fp16")]; + string var_15228_equation_0 = const()[name = string("op_15228_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15228_cast_fp16 = einsum(equation = var_15228_equation_0, values = (var_14734_cast_fp16, var_15134_cast_fp16))[name = string("op_15228_cast_fp16")]; + string var_15230_equation_0 = const()[name = string("op_15230_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15230_cast_fp16 = einsum(equation = var_15230_equation_0, values = (var_14734_cast_fp16, var_15135_cast_fp16))[name = string("op_15230_cast_fp16")]; + string var_15232_equation_0 = const()[name = string("op_15232_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15232_cast_fp16 = einsum(equation = var_15232_equation_0, values = (var_14734_cast_fp16, var_15136_cast_fp16))[name = string("op_15232_cast_fp16")]; + string var_15234_equation_0 = const()[name = string("op_15234_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15234_cast_fp16 = einsum(equation = var_15234_equation_0, values = (var_14738_cast_fp16, var_15137_cast_fp16))[name = string("op_15234_cast_fp16")]; + string var_15236_equation_0 = const()[name = string("op_15236_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15236_cast_fp16 = einsum(equation = var_15236_equation_0, values = (var_14738_cast_fp16, var_15138_cast_fp16))[name = string("op_15236_cast_fp16")]; + string var_15238_equation_0 = const()[name = string("op_15238_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15238_cast_fp16 = einsum(equation = var_15238_equation_0, values = (var_14738_cast_fp16, var_15139_cast_fp16))[name = string("op_15238_cast_fp16")]; + string var_15240_equation_0 = const()[name = string("op_15240_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15240_cast_fp16 = einsum(equation = var_15240_equation_0, values = (var_14738_cast_fp16, var_15140_cast_fp16))[name = string("op_15240_cast_fp16")]; + string var_15242_equation_0 = const()[name = string("op_15242_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15242_cast_fp16 = einsum(equation = var_15242_equation_0, values = (var_14742_cast_fp16, var_15141_cast_fp16))[name = string("op_15242_cast_fp16")]; + string var_15244_equation_0 = const()[name = string("op_15244_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15244_cast_fp16 = einsum(equation = var_15244_equation_0, values = (var_14742_cast_fp16, var_15142_cast_fp16))[name = string("op_15244_cast_fp16")]; + string var_15246_equation_0 = const()[name = string("op_15246_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15246_cast_fp16 = einsum(equation = var_15246_equation_0, values = (var_14742_cast_fp16, var_15143_cast_fp16))[name = string("op_15246_cast_fp16")]; + string var_15248_equation_0 = const()[name = string("op_15248_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15248_cast_fp16 = einsum(equation = var_15248_equation_0, values = (var_14742_cast_fp16, var_15144_cast_fp16))[name = string("op_15248_cast_fp16")]; + string var_15250_equation_0 = const()[name = string("op_15250_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15250_cast_fp16 = einsum(equation = var_15250_equation_0, values = (var_14746_cast_fp16, var_15145_cast_fp16))[name = string("op_15250_cast_fp16")]; + string var_15252_equation_0 = const()[name = string("op_15252_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15252_cast_fp16 = einsum(equation = var_15252_equation_0, values = (var_14746_cast_fp16, var_15146_cast_fp16))[name = string("op_15252_cast_fp16")]; + string var_15254_equation_0 = const()[name = string("op_15254_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15254_cast_fp16 = einsum(equation = var_15254_equation_0, values = (var_14746_cast_fp16, var_15147_cast_fp16))[name = string("op_15254_cast_fp16")]; + string var_15256_equation_0 = const()[name = string("op_15256_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15256_cast_fp16 = einsum(equation = var_15256_equation_0, values = (var_14746_cast_fp16, var_15148_cast_fp16))[name = string("op_15256_cast_fp16")]; + string var_15258_equation_0 = const()[name = string("op_15258_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15258_cast_fp16 = einsum(equation = var_15258_equation_0, values = (var_14750_cast_fp16, var_15149_cast_fp16))[name = string("op_15258_cast_fp16")]; + string var_15260_equation_0 = const()[name = string("op_15260_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15260_cast_fp16 = einsum(equation = var_15260_equation_0, values = (var_14750_cast_fp16, var_15150_cast_fp16))[name = string("op_15260_cast_fp16")]; + string var_15262_equation_0 = const()[name = string("op_15262_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15262_cast_fp16 = einsum(equation = var_15262_equation_0, values = (var_14750_cast_fp16, var_15151_cast_fp16))[name = string("op_15262_cast_fp16")]; + string var_15264_equation_0 = const()[name = string("op_15264_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15264_cast_fp16 = einsum(equation = var_15264_equation_0, values = (var_14750_cast_fp16, var_15152_cast_fp16))[name = string("op_15264_cast_fp16")]; + string var_15266_equation_0 = const()[name = string("op_15266_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15266_cast_fp16 = einsum(equation = var_15266_equation_0, values = (var_14754_cast_fp16, var_15153_cast_fp16))[name = string("op_15266_cast_fp16")]; + string var_15268_equation_0 = const()[name = string("op_15268_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15268_cast_fp16 = einsum(equation = var_15268_equation_0, values = (var_14754_cast_fp16, var_15154_cast_fp16))[name = string("op_15268_cast_fp16")]; + string var_15270_equation_0 = const()[name = string("op_15270_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15270_cast_fp16 = einsum(equation = var_15270_equation_0, values = (var_14754_cast_fp16, var_15155_cast_fp16))[name = string("op_15270_cast_fp16")]; + string var_15272_equation_0 = const()[name = string("op_15272_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15272_cast_fp16 = einsum(equation = var_15272_equation_0, values = (var_14754_cast_fp16, var_15156_cast_fp16))[name = string("op_15272_cast_fp16")]; + string var_15274_equation_0 = const()[name = string("op_15274_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15274_cast_fp16 = einsum(equation = var_15274_equation_0, values = (var_14758_cast_fp16, var_15157_cast_fp16))[name = string("op_15274_cast_fp16")]; + string var_15276_equation_0 = const()[name = string("op_15276_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15276_cast_fp16 = einsum(equation = var_15276_equation_0, values = (var_14758_cast_fp16, var_15158_cast_fp16))[name = string("op_15276_cast_fp16")]; + string var_15278_equation_0 = const()[name = string("op_15278_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15278_cast_fp16 = einsum(equation = var_15278_equation_0, values = (var_14758_cast_fp16, var_15159_cast_fp16))[name = string("op_15278_cast_fp16")]; + string var_15280_equation_0 = const()[name = string("op_15280_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15280_cast_fp16 = einsum(equation = var_15280_equation_0, values = (var_14758_cast_fp16, var_15160_cast_fp16))[name = string("op_15280_cast_fp16")]; + string var_15282_equation_0 = const()[name = string("op_15282_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15282_cast_fp16 = einsum(equation = var_15282_equation_0, values = (var_14762_cast_fp16, var_15161_cast_fp16))[name = string("op_15282_cast_fp16")]; + string var_15284_equation_0 = const()[name = string("op_15284_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15284_cast_fp16 = einsum(equation = var_15284_equation_0, values = (var_14762_cast_fp16, var_15162_cast_fp16))[name = string("op_15284_cast_fp16")]; + string var_15286_equation_0 = const()[name = string("op_15286_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15286_cast_fp16 = einsum(equation = var_15286_equation_0, values = (var_14762_cast_fp16, var_15163_cast_fp16))[name = string("op_15286_cast_fp16")]; + string var_15288_equation_0 = const()[name = string("op_15288_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15288_cast_fp16 = einsum(equation = var_15288_equation_0, values = (var_14762_cast_fp16, var_15164_cast_fp16))[name = string("op_15288_cast_fp16")]; + string var_15290_equation_0 = const()[name = string("op_15290_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15290_cast_fp16 = einsum(equation = var_15290_equation_0, values = (var_14766_cast_fp16, var_15165_cast_fp16))[name = string("op_15290_cast_fp16")]; + string var_15292_equation_0 = const()[name = string("op_15292_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15292_cast_fp16 = einsum(equation = var_15292_equation_0, values = (var_14766_cast_fp16, var_15166_cast_fp16))[name = string("op_15292_cast_fp16")]; + string var_15294_equation_0 = const()[name = string("op_15294_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15294_cast_fp16 = einsum(equation = var_15294_equation_0, values = (var_14766_cast_fp16, var_15167_cast_fp16))[name = string("op_15294_cast_fp16")]; + string var_15296_equation_0 = const()[name = string("op_15296_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15296_cast_fp16 = einsum(equation = var_15296_equation_0, values = (var_14766_cast_fp16, var_15168_cast_fp16))[name = string("op_15296_cast_fp16")]; + string var_15298_equation_0 = const()[name = string("op_15298_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15298_cast_fp16 = einsum(equation = var_15298_equation_0, values = (var_14770_cast_fp16, var_15169_cast_fp16))[name = string("op_15298_cast_fp16")]; + string var_15300_equation_0 = const()[name = string("op_15300_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15300_cast_fp16 = einsum(equation = var_15300_equation_0, values = (var_14770_cast_fp16, var_15170_cast_fp16))[name = string("op_15300_cast_fp16")]; + string var_15302_equation_0 = const()[name = string("op_15302_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15302_cast_fp16 = einsum(equation = var_15302_equation_0, values = (var_14770_cast_fp16, var_15171_cast_fp16))[name = string("op_15302_cast_fp16")]; + string var_15304_equation_0 = const()[name = string("op_15304_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15304_cast_fp16 = einsum(equation = var_15304_equation_0, values = (var_14770_cast_fp16, var_15172_cast_fp16))[name = string("op_15304_cast_fp16")]; + string var_15306_equation_0 = const()[name = string("op_15306_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15306_cast_fp16 = einsum(equation = var_15306_equation_0, values = (var_14774_cast_fp16, var_15173_cast_fp16))[name = string("op_15306_cast_fp16")]; + string var_15308_equation_0 = const()[name = string("op_15308_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15308_cast_fp16 = einsum(equation = var_15308_equation_0, values = (var_14774_cast_fp16, var_15174_cast_fp16))[name = string("op_15308_cast_fp16")]; + string var_15310_equation_0 = const()[name = string("op_15310_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15310_cast_fp16 = einsum(equation = var_15310_equation_0, values = (var_14774_cast_fp16, var_15175_cast_fp16))[name = string("op_15310_cast_fp16")]; + string var_15312_equation_0 = const()[name = string("op_15312_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15312_cast_fp16 = einsum(equation = var_15312_equation_0, values = (var_14774_cast_fp16, var_15176_cast_fp16))[name = string("op_15312_cast_fp16")]; + string var_15314_equation_0 = const()[name = string("op_15314_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15314_cast_fp16 = einsum(equation = var_15314_equation_0, values = (var_14778_cast_fp16, var_15177_cast_fp16))[name = string("op_15314_cast_fp16")]; + string var_15316_equation_0 = const()[name = string("op_15316_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15316_cast_fp16 = einsum(equation = var_15316_equation_0, values = (var_14778_cast_fp16, var_15178_cast_fp16))[name = string("op_15316_cast_fp16")]; + string var_15318_equation_0 = const()[name = string("op_15318_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15318_cast_fp16 = einsum(equation = var_15318_equation_0, values = (var_14778_cast_fp16, var_15179_cast_fp16))[name = string("op_15318_cast_fp16")]; + string var_15320_equation_0 = const()[name = string("op_15320_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15320_cast_fp16 = einsum(equation = var_15320_equation_0, values = (var_14778_cast_fp16, var_15180_cast_fp16))[name = string("op_15320_cast_fp16")]; + string var_15322_equation_0 = const()[name = string("op_15322_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15322_cast_fp16 = einsum(equation = var_15322_equation_0, values = (var_14782_cast_fp16, var_15181_cast_fp16))[name = string("op_15322_cast_fp16")]; + string var_15324_equation_0 = const()[name = string("op_15324_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15324_cast_fp16 = einsum(equation = var_15324_equation_0, values = (var_14782_cast_fp16, var_15182_cast_fp16))[name = string("op_15324_cast_fp16")]; + string var_15326_equation_0 = const()[name = string("op_15326_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15326_cast_fp16 = einsum(equation = var_15326_equation_0, values = (var_14782_cast_fp16, var_15183_cast_fp16))[name = string("op_15326_cast_fp16")]; + string var_15328_equation_0 = const()[name = string("op_15328_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15328_cast_fp16 = einsum(equation = var_15328_equation_0, values = (var_14782_cast_fp16, var_15184_cast_fp16))[name = string("op_15328_cast_fp16")]; + string var_15330_equation_0 = const()[name = string("op_15330_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15330_cast_fp16 = einsum(equation = var_15330_equation_0, values = (var_14786_cast_fp16, var_15185_cast_fp16))[name = string("op_15330_cast_fp16")]; + string var_15332_equation_0 = const()[name = string("op_15332_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15332_cast_fp16 = einsum(equation = var_15332_equation_0, values = (var_14786_cast_fp16, var_15186_cast_fp16))[name = string("op_15332_cast_fp16")]; + string var_15334_equation_0 = const()[name = string("op_15334_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15334_cast_fp16 = einsum(equation = var_15334_equation_0, values = (var_14786_cast_fp16, var_15187_cast_fp16))[name = string("op_15334_cast_fp16")]; + string var_15336_equation_0 = const()[name = string("op_15336_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15336_cast_fp16 = einsum(equation = var_15336_equation_0, values = (var_14786_cast_fp16, var_15188_cast_fp16))[name = string("op_15336_cast_fp16")]; + string var_15338_equation_0 = const()[name = string("op_15338_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15338_cast_fp16 = einsum(equation = var_15338_equation_0, values = (var_14790_cast_fp16, var_15189_cast_fp16))[name = string("op_15338_cast_fp16")]; + string var_15340_equation_0 = const()[name = string("op_15340_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15340_cast_fp16 = einsum(equation = var_15340_equation_0, values = (var_14790_cast_fp16, var_15190_cast_fp16))[name = string("op_15340_cast_fp16")]; + string var_15342_equation_0 = const()[name = string("op_15342_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15342_cast_fp16 = einsum(equation = var_15342_equation_0, values = (var_14790_cast_fp16, var_15191_cast_fp16))[name = string("op_15342_cast_fp16")]; + string var_15344_equation_0 = const()[name = string("op_15344_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15344_cast_fp16 = einsum(equation = var_15344_equation_0, values = (var_14790_cast_fp16, var_15192_cast_fp16))[name = string("op_15344_cast_fp16")]; + string var_15346_equation_0 = const()[name = string("op_15346_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15346_cast_fp16 = einsum(equation = var_15346_equation_0, values = (var_14794_cast_fp16, var_15193_cast_fp16))[name = string("op_15346_cast_fp16")]; + string var_15348_equation_0 = const()[name = string("op_15348_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15348_cast_fp16 = einsum(equation = var_15348_equation_0, values = (var_14794_cast_fp16, var_15194_cast_fp16))[name = string("op_15348_cast_fp16")]; + string var_15350_equation_0 = const()[name = string("op_15350_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15350_cast_fp16 = einsum(equation = var_15350_equation_0, values = (var_14794_cast_fp16, var_15195_cast_fp16))[name = string("op_15350_cast_fp16")]; + string var_15352_equation_0 = const()[name = string("op_15352_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15352_cast_fp16 = einsum(equation = var_15352_equation_0, values = (var_14794_cast_fp16, var_15196_cast_fp16))[name = string("op_15352_cast_fp16")]; + string var_15354_equation_0 = const()[name = string("op_15354_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15354_cast_fp16 = einsum(equation = var_15354_equation_0, values = (var_14798_cast_fp16, var_15197_cast_fp16))[name = string("op_15354_cast_fp16")]; + string var_15356_equation_0 = const()[name = string("op_15356_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15356_cast_fp16 = einsum(equation = var_15356_equation_0, values = (var_14798_cast_fp16, var_15198_cast_fp16))[name = string("op_15356_cast_fp16")]; + string var_15358_equation_0 = const()[name = string("op_15358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15358_cast_fp16 = einsum(equation = var_15358_equation_0, values = (var_14798_cast_fp16, var_15199_cast_fp16))[name = string("op_15358_cast_fp16")]; + string var_15360_equation_0 = const()[name = string("op_15360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15360_cast_fp16 = einsum(equation = var_15360_equation_0, values = (var_14798_cast_fp16, var_15200_cast_fp16))[name = string("op_15360_cast_fp16")]; + bool var_15362_interleave_0 = const()[name = string("op_15362_interleave_0"), val = bool(false)]; + tensor var_15362_cast_fp16 = concat(axis = var_13921, interleave = var_15362_interleave_0, values = (var_15202_cast_fp16, var_15204_cast_fp16, var_15206_cast_fp16, var_15208_cast_fp16))[name = string("op_15362_cast_fp16")]; + bool var_15364_interleave_0 = const()[name = string("op_15364_interleave_0"), val = bool(false)]; + tensor var_15364_cast_fp16 = concat(axis = var_13921, interleave = var_15364_interleave_0, values = (var_15210_cast_fp16, var_15212_cast_fp16, var_15214_cast_fp16, var_15216_cast_fp16))[name = string("op_15364_cast_fp16")]; + bool var_15366_interleave_0 = const()[name = string("op_15366_interleave_0"), val = bool(false)]; + tensor var_15366_cast_fp16 = concat(axis = var_13921, interleave = var_15366_interleave_0, values = (var_15218_cast_fp16, var_15220_cast_fp16, var_15222_cast_fp16, var_15224_cast_fp16))[name = string("op_15366_cast_fp16")]; + bool var_15368_interleave_0 = const()[name = string("op_15368_interleave_0"), val = bool(false)]; + tensor var_15368_cast_fp16 = concat(axis = var_13921, interleave = var_15368_interleave_0, values = (var_15226_cast_fp16, var_15228_cast_fp16, var_15230_cast_fp16, var_15232_cast_fp16))[name = string("op_15368_cast_fp16")]; + bool var_15370_interleave_0 = const()[name = string("op_15370_interleave_0"), val = bool(false)]; + tensor var_15370_cast_fp16 = concat(axis = var_13921, interleave = var_15370_interleave_0, values = (var_15234_cast_fp16, var_15236_cast_fp16, var_15238_cast_fp16, var_15240_cast_fp16))[name = string("op_15370_cast_fp16")]; + bool var_15372_interleave_0 = const()[name = string("op_15372_interleave_0"), val = bool(false)]; + tensor var_15372_cast_fp16 = concat(axis = var_13921, interleave = var_15372_interleave_0, values = (var_15242_cast_fp16, var_15244_cast_fp16, var_15246_cast_fp16, var_15248_cast_fp16))[name = string("op_15372_cast_fp16")]; + bool var_15374_interleave_0 = const()[name = string("op_15374_interleave_0"), val = bool(false)]; + tensor var_15374_cast_fp16 = concat(axis = var_13921, interleave = var_15374_interleave_0, values = (var_15250_cast_fp16, var_15252_cast_fp16, var_15254_cast_fp16, var_15256_cast_fp16))[name = string("op_15374_cast_fp16")]; + bool var_15376_interleave_0 = const()[name = string("op_15376_interleave_0"), val = bool(false)]; + tensor var_15376_cast_fp16 = concat(axis = var_13921, interleave = var_15376_interleave_0, values = (var_15258_cast_fp16, var_15260_cast_fp16, var_15262_cast_fp16, var_15264_cast_fp16))[name = string("op_15376_cast_fp16")]; + bool var_15378_interleave_0 = const()[name = string("op_15378_interleave_0"), val = bool(false)]; + tensor var_15378_cast_fp16 = concat(axis = var_13921, interleave = var_15378_interleave_0, values = (var_15266_cast_fp16, var_15268_cast_fp16, var_15270_cast_fp16, var_15272_cast_fp16))[name = string("op_15378_cast_fp16")]; + bool var_15380_interleave_0 = const()[name = string("op_15380_interleave_0"), val = bool(false)]; + tensor var_15380_cast_fp16 = concat(axis = var_13921, interleave = var_15380_interleave_0, values = (var_15274_cast_fp16, var_15276_cast_fp16, var_15278_cast_fp16, var_15280_cast_fp16))[name = string("op_15380_cast_fp16")]; + bool var_15382_interleave_0 = const()[name = string("op_15382_interleave_0"), val = bool(false)]; + tensor var_15382_cast_fp16 = concat(axis = var_13921, interleave = var_15382_interleave_0, values = (var_15282_cast_fp16, var_15284_cast_fp16, var_15286_cast_fp16, var_15288_cast_fp16))[name = string("op_15382_cast_fp16")]; + bool var_15384_interleave_0 = const()[name = string("op_15384_interleave_0"), val = bool(false)]; + tensor var_15384_cast_fp16 = concat(axis = var_13921, interleave = var_15384_interleave_0, values = (var_15290_cast_fp16, var_15292_cast_fp16, var_15294_cast_fp16, var_15296_cast_fp16))[name = string("op_15384_cast_fp16")]; + bool var_15386_interleave_0 = const()[name = string("op_15386_interleave_0"), val = bool(false)]; + tensor var_15386_cast_fp16 = concat(axis = var_13921, interleave = var_15386_interleave_0, values = (var_15298_cast_fp16, var_15300_cast_fp16, var_15302_cast_fp16, var_15304_cast_fp16))[name = string("op_15386_cast_fp16")]; + bool var_15388_interleave_0 = const()[name = string("op_15388_interleave_0"), val = bool(false)]; + tensor var_15388_cast_fp16 = concat(axis = var_13921, interleave = var_15388_interleave_0, values = (var_15306_cast_fp16, var_15308_cast_fp16, var_15310_cast_fp16, var_15312_cast_fp16))[name = string("op_15388_cast_fp16")]; + bool var_15390_interleave_0 = const()[name = string("op_15390_interleave_0"), val = bool(false)]; + tensor var_15390_cast_fp16 = concat(axis = var_13921, interleave = var_15390_interleave_0, values = (var_15314_cast_fp16, var_15316_cast_fp16, var_15318_cast_fp16, var_15320_cast_fp16))[name = string("op_15390_cast_fp16")]; + bool var_15392_interleave_0 = const()[name = string("op_15392_interleave_0"), val = bool(false)]; + tensor var_15392_cast_fp16 = concat(axis = var_13921, interleave = var_15392_interleave_0, values = (var_15322_cast_fp16, var_15324_cast_fp16, var_15326_cast_fp16, var_15328_cast_fp16))[name = string("op_15392_cast_fp16")]; + bool var_15394_interleave_0 = const()[name = string("op_15394_interleave_0"), val = bool(false)]; + tensor var_15394_cast_fp16 = concat(axis = var_13921, interleave = var_15394_interleave_0, values = (var_15330_cast_fp16, var_15332_cast_fp16, var_15334_cast_fp16, var_15336_cast_fp16))[name = string("op_15394_cast_fp16")]; + bool var_15396_interleave_0 = const()[name = string("op_15396_interleave_0"), val = bool(false)]; + tensor var_15396_cast_fp16 = concat(axis = var_13921, interleave = var_15396_interleave_0, values = (var_15338_cast_fp16, var_15340_cast_fp16, var_15342_cast_fp16, var_15344_cast_fp16))[name = string("op_15396_cast_fp16")]; + bool var_15398_interleave_0 = const()[name = string("op_15398_interleave_0"), val = bool(false)]; + tensor var_15398_cast_fp16 = concat(axis = var_13921, interleave = var_15398_interleave_0, values = (var_15346_cast_fp16, var_15348_cast_fp16, var_15350_cast_fp16, var_15352_cast_fp16))[name = string("op_15398_cast_fp16")]; + bool var_15400_interleave_0 = const()[name = string("op_15400_interleave_0"), val = bool(false)]; + tensor var_15400_cast_fp16 = concat(axis = var_13921, interleave = var_15400_interleave_0, values = (var_15354_cast_fp16, var_15356_cast_fp16, var_15358_cast_fp16, var_15360_cast_fp16))[name = string("op_15400_cast_fp16")]; + bool input_73_interleave_0 = const()[name = string("input_73_interleave_0"), val = bool(false)]; + tensor input_73_cast_fp16 = concat(axis = var_13946, interleave = input_73_interleave_0, values = (var_15362_cast_fp16, var_15364_cast_fp16, var_15366_cast_fp16, var_15368_cast_fp16, var_15370_cast_fp16, var_15372_cast_fp16, var_15374_cast_fp16, var_15376_cast_fp16, var_15378_cast_fp16, var_15380_cast_fp16, var_15382_cast_fp16, var_15384_cast_fp16, var_15386_cast_fp16, var_15388_cast_fp16, var_15390_cast_fp16, var_15392_cast_fp16, var_15394_cast_fp16, var_15396_cast_fp16, var_15398_cast_fp16, var_15400_cast_fp16))[name = string("input_73_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378684800)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381961664)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_15419_to_fp16 = const()[name = string("op_15419_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_15419_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381964288)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381966912)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381969536)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395076800)))]; + tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395087104)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408194368)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; + int32 var_15448 = const()[name = string("op_15448"), val = int32(3)]; + int32 var_15473 = const()[name = string("op_15473"), val = int32(1)]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_15490_to_fp16 = const()[name = string("op_15490_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_15490_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408196992)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408199616)))]; + fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408202240)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411479104)))]; + tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; + string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; + tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; + tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; + int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411481728)))]; + tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; + string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; + tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; + tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; + int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414758592)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418035456)))]; + tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_15528_begin_0 = const()[name = string("op_15528_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15528_end_0 = const()[name = string("op_15528_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15528_end_mask_0 = const()[name = string("op_15528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15528_cast_fp16 = slice_by_index(begin = var_15528_begin_0, end = var_15528_end_0, end_mask = var_15528_end_mask_0, x = query_21_cast_fp16)[name = string("op_15528_cast_fp16")]; + tensor var_15532_begin_0 = const()[name = string("op_15532_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15532_end_0 = const()[name = string("op_15532_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_15532_end_mask_0 = const()[name = string("op_15532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15532_cast_fp16 = slice_by_index(begin = var_15532_begin_0, end = var_15532_end_0, end_mask = var_15532_end_mask_0, x = query_21_cast_fp16)[name = string("op_15532_cast_fp16")]; + tensor var_15536_begin_0 = const()[name = string("op_15536_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15536_end_0 = const()[name = string("op_15536_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_15536_end_mask_0 = const()[name = string("op_15536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15536_cast_fp16 = slice_by_index(begin = var_15536_begin_0, end = var_15536_end_0, end_mask = var_15536_end_mask_0, x = query_21_cast_fp16)[name = string("op_15536_cast_fp16")]; + tensor var_15540_begin_0 = const()[name = string("op_15540_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15540_end_0 = const()[name = string("op_15540_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_15540_end_mask_0 = const()[name = string("op_15540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15540_cast_fp16 = slice_by_index(begin = var_15540_begin_0, end = var_15540_end_0, end_mask = var_15540_end_mask_0, x = query_21_cast_fp16)[name = string("op_15540_cast_fp16")]; + tensor var_15544_begin_0 = const()[name = string("op_15544_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15544_end_0 = const()[name = string("op_15544_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_15544_end_mask_0 = const()[name = string("op_15544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15544_cast_fp16 = slice_by_index(begin = var_15544_begin_0, end = var_15544_end_0, end_mask = var_15544_end_mask_0, x = query_21_cast_fp16)[name = string("op_15544_cast_fp16")]; + tensor var_15548_begin_0 = const()[name = string("op_15548_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15548_end_0 = const()[name = string("op_15548_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_15548_end_mask_0 = const()[name = string("op_15548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15548_cast_fp16 = slice_by_index(begin = var_15548_begin_0, end = var_15548_end_0, end_mask = var_15548_end_mask_0, x = query_21_cast_fp16)[name = string("op_15548_cast_fp16")]; + tensor var_15552_begin_0 = const()[name = string("op_15552_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15552_end_0 = const()[name = string("op_15552_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_15552_end_mask_0 = const()[name = string("op_15552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15552_cast_fp16 = slice_by_index(begin = var_15552_begin_0, end = var_15552_end_0, end_mask = var_15552_end_mask_0, x = query_21_cast_fp16)[name = string("op_15552_cast_fp16")]; + tensor var_15556_begin_0 = const()[name = string("op_15556_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15556_end_0 = const()[name = string("op_15556_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_15556_end_mask_0 = const()[name = string("op_15556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15556_cast_fp16 = slice_by_index(begin = var_15556_begin_0, end = var_15556_end_0, end_mask = var_15556_end_mask_0, x = query_21_cast_fp16)[name = string("op_15556_cast_fp16")]; + tensor var_15560_begin_0 = const()[name = string("op_15560_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15560_end_0 = const()[name = string("op_15560_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_15560_end_mask_0 = const()[name = string("op_15560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15560_cast_fp16 = slice_by_index(begin = var_15560_begin_0, end = var_15560_end_0, end_mask = var_15560_end_mask_0, x = query_21_cast_fp16)[name = string("op_15560_cast_fp16")]; + tensor var_15564_begin_0 = const()[name = string("op_15564_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15564_end_0 = const()[name = string("op_15564_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_15564_end_mask_0 = const()[name = string("op_15564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15564_cast_fp16 = slice_by_index(begin = var_15564_begin_0, end = var_15564_end_0, end_mask = var_15564_end_mask_0, x = query_21_cast_fp16)[name = string("op_15564_cast_fp16")]; + tensor var_15568_begin_0 = const()[name = string("op_15568_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15568_end_0 = const()[name = string("op_15568_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_15568_end_mask_0 = const()[name = string("op_15568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15568_cast_fp16 = slice_by_index(begin = var_15568_begin_0, end = var_15568_end_0, end_mask = var_15568_end_mask_0, x = query_21_cast_fp16)[name = string("op_15568_cast_fp16")]; + tensor var_15572_begin_0 = const()[name = string("op_15572_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15572_end_0 = const()[name = string("op_15572_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_15572_end_mask_0 = const()[name = string("op_15572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15572_cast_fp16 = slice_by_index(begin = var_15572_begin_0, end = var_15572_end_0, end_mask = var_15572_end_mask_0, x = query_21_cast_fp16)[name = string("op_15572_cast_fp16")]; + tensor var_15576_begin_0 = const()[name = string("op_15576_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15576_end_0 = const()[name = string("op_15576_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_15576_end_mask_0 = const()[name = string("op_15576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15576_cast_fp16 = slice_by_index(begin = var_15576_begin_0, end = var_15576_end_0, end_mask = var_15576_end_mask_0, x = query_21_cast_fp16)[name = string("op_15576_cast_fp16")]; + tensor var_15580_begin_0 = const()[name = string("op_15580_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15580_end_0 = const()[name = string("op_15580_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_15580_end_mask_0 = const()[name = string("op_15580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15580_cast_fp16 = slice_by_index(begin = var_15580_begin_0, end = var_15580_end_0, end_mask = var_15580_end_mask_0, x = query_21_cast_fp16)[name = string("op_15580_cast_fp16")]; + tensor var_15584_begin_0 = const()[name = string("op_15584_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15584_end_0 = const()[name = string("op_15584_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_15584_end_mask_0 = const()[name = string("op_15584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15584_cast_fp16 = slice_by_index(begin = var_15584_begin_0, end = var_15584_end_0, end_mask = var_15584_end_mask_0, x = query_21_cast_fp16)[name = string("op_15584_cast_fp16")]; + tensor var_15588_begin_0 = const()[name = string("op_15588_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15588_end_0 = const()[name = string("op_15588_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_15588_end_mask_0 = const()[name = string("op_15588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15588_cast_fp16 = slice_by_index(begin = var_15588_begin_0, end = var_15588_end_0, end_mask = var_15588_end_mask_0, x = query_21_cast_fp16)[name = string("op_15588_cast_fp16")]; + tensor var_15592_begin_0 = const()[name = string("op_15592_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15592_end_0 = const()[name = string("op_15592_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_15592_end_mask_0 = const()[name = string("op_15592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15592_cast_fp16 = slice_by_index(begin = var_15592_begin_0, end = var_15592_end_0, end_mask = var_15592_end_mask_0, x = query_21_cast_fp16)[name = string("op_15592_cast_fp16")]; + tensor var_15596_begin_0 = const()[name = string("op_15596_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15596_end_0 = const()[name = string("op_15596_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_15596_end_mask_0 = const()[name = string("op_15596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15596_cast_fp16 = slice_by_index(begin = var_15596_begin_0, end = var_15596_end_0, end_mask = var_15596_end_mask_0, x = query_21_cast_fp16)[name = string("op_15596_cast_fp16")]; + tensor var_15600_begin_0 = const()[name = string("op_15600_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15600_end_0 = const()[name = string("op_15600_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_15600_end_mask_0 = const()[name = string("op_15600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15600_cast_fp16 = slice_by_index(begin = var_15600_begin_0, end = var_15600_end_0, end_mask = var_15600_end_mask_0, x = query_21_cast_fp16)[name = string("op_15600_cast_fp16")]; + tensor var_15604_begin_0 = const()[name = string("op_15604_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15604_end_0 = const()[name = string("op_15604_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_15604_end_mask_0 = const()[name = string("op_15604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15604_cast_fp16 = slice_by_index(begin = var_15604_begin_0, end = var_15604_end_0, end_mask = var_15604_end_mask_0, x = query_21_cast_fp16)[name = string("op_15604_cast_fp16")]; + tensor var_15613_begin_0 = const()[name = string("op_15613_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15613_end_0 = const()[name = string("op_15613_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15613_end_mask_0 = const()[name = string("op_15613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15613_cast_fp16 = slice_by_index(begin = var_15613_begin_0, end = var_15613_end_0, end_mask = var_15613_end_mask_0, x = var_15528_cast_fp16)[name = string("op_15613_cast_fp16")]; + tensor var_15620_begin_0 = const()[name = string("op_15620_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15620_end_0 = const()[name = string("op_15620_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15620_end_mask_0 = const()[name = string("op_15620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15620_cast_fp16 = slice_by_index(begin = var_15620_begin_0, end = var_15620_end_0, end_mask = var_15620_end_mask_0, x = var_15528_cast_fp16)[name = string("op_15620_cast_fp16")]; + tensor var_15627_begin_0 = const()[name = string("op_15627_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15627_end_0 = const()[name = string("op_15627_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15627_end_mask_0 = const()[name = string("op_15627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15627_cast_fp16 = slice_by_index(begin = var_15627_begin_0, end = var_15627_end_0, end_mask = var_15627_end_mask_0, x = var_15528_cast_fp16)[name = string("op_15627_cast_fp16")]; + tensor var_15634_begin_0 = const()[name = string("op_15634_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15634_end_0 = const()[name = string("op_15634_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15634_end_mask_0 = const()[name = string("op_15634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15634_cast_fp16 = slice_by_index(begin = var_15634_begin_0, end = var_15634_end_0, end_mask = var_15634_end_mask_0, x = var_15528_cast_fp16)[name = string("op_15634_cast_fp16")]; + tensor var_15641_begin_0 = const()[name = string("op_15641_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15641_end_0 = const()[name = string("op_15641_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15641_end_mask_0 = const()[name = string("op_15641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15641_cast_fp16 = slice_by_index(begin = var_15641_begin_0, end = var_15641_end_0, end_mask = var_15641_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15641_cast_fp16")]; + tensor var_15648_begin_0 = const()[name = string("op_15648_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15648_end_0 = const()[name = string("op_15648_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15648_end_mask_0 = const()[name = string("op_15648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15648_cast_fp16 = slice_by_index(begin = var_15648_begin_0, end = var_15648_end_0, end_mask = var_15648_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15648_cast_fp16")]; + tensor var_15655_begin_0 = const()[name = string("op_15655_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15655_end_0 = const()[name = string("op_15655_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15655_end_mask_0 = const()[name = string("op_15655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15655_cast_fp16 = slice_by_index(begin = var_15655_begin_0, end = var_15655_end_0, end_mask = var_15655_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15655_cast_fp16")]; + tensor var_15662_begin_0 = const()[name = string("op_15662_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15662_end_0 = const()[name = string("op_15662_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15662_end_mask_0 = const()[name = string("op_15662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15662_cast_fp16 = slice_by_index(begin = var_15662_begin_0, end = var_15662_end_0, end_mask = var_15662_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15662_cast_fp16")]; + tensor var_15669_begin_0 = const()[name = string("op_15669_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15669_end_0 = const()[name = string("op_15669_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15669_end_mask_0 = const()[name = string("op_15669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15669_cast_fp16 = slice_by_index(begin = var_15669_begin_0, end = var_15669_end_0, end_mask = var_15669_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15669_cast_fp16")]; + tensor var_15676_begin_0 = const()[name = string("op_15676_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15676_end_0 = const()[name = string("op_15676_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15676_end_mask_0 = const()[name = string("op_15676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15676_cast_fp16 = slice_by_index(begin = var_15676_begin_0, end = var_15676_end_0, end_mask = var_15676_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15676_cast_fp16")]; + tensor var_15683_begin_0 = const()[name = string("op_15683_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15683_end_0 = const()[name = string("op_15683_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15683_end_mask_0 = const()[name = string("op_15683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15683_cast_fp16 = slice_by_index(begin = var_15683_begin_0, end = var_15683_end_0, end_mask = var_15683_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15683_cast_fp16")]; + tensor var_15690_begin_0 = const()[name = string("op_15690_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15690_end_0 = const()[name = string("op_15690_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15690_end_mask_0 = const()[name = string("op_15690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15690_cast_fp16 = slice_by_index(begin = var_15690_begin_0, end = var_15690_end_0, end_mask = var_15690_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15690_cast_fp16")]; + tensor var_15697_begin_0 = const()[name = string("op_15697_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15697_end_0 = const()[name = string("op_15697_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15697_end_mask_0 = const()[name = string("op_15697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15697_cast_fp16 = slice_by_index(begin = var_15697_begin_0, end = var_15697_end_0, end_mask = var_15697_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15697_cast_fp16")]; + tensor var_15704_begin_0 = const()[name = string("op_15704_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15704_end_0 = const()[name = string("op_15704_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15704_end_mask_0 = const()[name = string("op_15704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15704_cast_fp16 = slice_by_index(begin = var_15704_begin_0, end = var_15704_end_0, end_mask = var_15704_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15704_cast_fp16")]; + tensor var_15711_begin_0 = const()[name = string("op_15711_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15711_end_0 = const()[name = string("op_15711_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15711_end_mask_0 = const()[name = string("op_15711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15711_cast_fp16 = slice_by_index(begin = var_15711_begin_0, end = var_15711_end_0, end_mask = var_15711_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15711_cast_fp16")]; + tensor var_15718_begin_0 = const()[name = string("op_15718_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15718_end_0 = const()[name = string("op_15718_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15718_end_mask_0 = const()[name = string("op_15718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15718_cast_fp16 = slice_by_index(begin = var_15718_begin_0, end = var_15718_end_0, end_mask = var_15718_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15718_cast_fp16")]; + tensor var_15725_begin_0 = const()[name = string("op_15725_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15725_end_0 = const()[name = string("op_15725_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15725_end_mask_0 = const()[name = string("op_15725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15725_cast_fp16 = slice_by_index(begin = var_15725_begin_0, end = var_15725_end_0, end_mask = var_15725_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15725_cast_fp16")]; + tensor var_15732_begin_0 = const()[name = string("op_15732_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15732_end_0 = const()[name = string("op_15732_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15732_end_mask_0 = const()[name = string("op_15732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15732_cast_fp16 = slice_by_index(begin = var_15732_begin_0, end = var_15732_end_0, end_mask = var_15732_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15732_cast_fp16")]; + tensor var_15739_begin_0 = const()[name = string("op_15739_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15739_end_0 = const()[name = string("op_15739_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15739_end_mask_0 = const()[name = string("op_15739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15739_cast_fp16 = slice_by_index(begin = var_15739_begin_0, end = var_15739_end_0, end_mask = var_15739_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15739_cast_fp16")]; + tensor var_15746_begin_0 = const()[name = string("op_15746_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15746_end_0 = const()[name = string("op_15746_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15746_end_mask_0 = const()[name = string("op_15746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15746_cast_fp16 = slice_by_index(begin = var_15746_begin_0, end = var_15746_end_0, end_mask = var_15746_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15746_cast_fp16")]; + tensor var_15753_begin_0 = const()[name = string("op_15753_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15753_end_0 = const()[name = string("op_15753_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15753_end_mask_0 = const()[name = string("op_15753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15753_cast_fp16 = slice_by_index(begin = var_15753_begin_0, end = var_15753_end_0, end_mask = var_15753_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15753_cast_fp16")]; + tensor var_15760_begin_0 = const()[name = string("op_15760_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15760_end_0 = const()[name = string("op_15760_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15760_end_mask_0 = const()[name = string("op_15760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15760_cast_fp16 = slice_by_index(begin = var_15760_begin_0, end = var_15760_end_0, end_mask = var_15760_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15760_cast_fp16")]; + tensor var_15767_begin_0 = const()[name = string("op_15767_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15767_end_0 = const()[name = string("op_15767_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15767_end_mask_0 = const()[name = string("op_15767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15767_cast_fp16 = slice_by_index(begin = var_15767_begin_0, end = var_15767_end_0, end_mask = var_15767_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15767_cast_fp16")]; + tensor var_15774_begin_0 = const()[name = string("op_15774_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15774_end_0 = const()[name = string("op_15774_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15774_end_mask_0 = const()[name = string("op_15774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15774_cast_fp16 = slice_by_index(begin = var_15774_begin_0, end = var_15774_end_0, end_mask = var_15774_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15774_cast_fp16")]; + tensor var_15781_begin_0 = const()[name = string("op_15781_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15781_end_0 = const()[name = string("op_15781_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15781_end_mask_0 = const()[name = string("op_15781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15781_cast_fp16 = slice_by_index(begin = var_15781_begin_0, end = var_15781_end_0, end_mask = var_15781_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15781_cast_fp16")]; + tensor var_15788_begin_0 = const()[name = string("op_15788_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15788_end_0 = const()[name = string("op_15788_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15788_end_mask_0 = const()[name = string("op_15788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15788_cast_fp16 = slice_by_index(begin = var_15788_begin_0, end = var_15788_end_0, end_mask = var_15788_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15788_cast_fp16")]; + tensor var_15795_begin_0 = const()[name = string("op_15795_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15795_end_0 = const()[name = string("op_15795_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15795_end_mask_0 = const()[name = string("op_15795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15795_cast_fp16 = slice_by_index(begin = var_15795_begin_0, end = var_15795_end_0, end_mask = var_15795_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15795_cast_fp16")]; + tensor var_15802_begin_0 = const()[name = string("op_15802_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15802_end_0 = const()[name = string("op_15802_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15802_end_mask_0 = const()[name = string("op_15802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15802_cast_fp16 = slice_by_index(begin = var_15802_begin_0, end = var_15802_end_0, end_mask = var_15802_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15802_cast_fp16")]; + tensor var_15809_begin_0 = const()[name = string("op_15809_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15809_end_0 = const()[name = string("op_15809_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15809_end_mask_0 = const()[name = string("op_15809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15809_cast_fp16 = slice_by_index(begin = var_15809_begin_0, end = var_15809_end_0, end_mask = var_15809_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15809_cast_fp16")]; + tensor var_15816_begin_0 = const()[name = string("op_15816_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15816_end_0 = const()[name = string("op_15816_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15816_end_mask_0 = const()[name = string("op_15816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15816_cast_fp16 = slice_by_index(begin = var_15816_begin_0, end = var_15816_end_0, end_mask = var_15816_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15816_cast_fp16")]; + tensor var_15823_begin_0 = const()[name = string("op_15823_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15823_end_0 = const()[name = string("op_15823_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15823_end_mask_0 = const()[name = string("op_15823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15823_cast_fp16 = slice_by_index(begin = var_15823_begin_0, end = var_15823_end_0, end_mask = var_15823_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15823_cast_fp16")]; + tensor var_15830_begin_0 = const()[name = string("op_15830_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15830_end_0 = const()[name = string("op_15830_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15830_end_mask_0 = const()[name = string("op_15830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15830_cast_fp16 = slice_by_index(begin = var_15830_begin_0, end = var_15830_end_0, end_mask = var_15830_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15830_cast_fp16")]; + tensor var_15837_begin_0 = const()[name = string("op_15837_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15837_end_0 = const()[name = string("op_15837_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15837_end_mask_0 = const()[name = string("op_15837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15837_cast_fp16 = slice_by_index(begin = var_15837_begin_0, end = var_15837_end_0, end_mask = var_15837_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15837_cast_fp16")]; + tensor var_15844_begin_0 = const()[name = string("op_15844_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15844_end_0 = const()[name = string("op_15844_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15844_end_mask_0 = const()[name = string("op_15844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15844_cast_fp16 = slice_by_index(begin = var_15844_begin_0, end = var_15844_end_0, end_mask = var_15844_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15844_cast_fp16")]; + tensor var_15851_begin_0 = const()[name = string("op_15851_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15851_end_0 = const()[name = string("op_15851_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15851_end_mask_0 = const()[name = string("op_15851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15851_cast_fp16 = slice_by_index(begin = var_15851_begin_0, end = var_15851_end_0, end_mask = var_15851_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15851_cast_fp16")]; + tensor var_15858_begin_0 = const()[name = string("op_15858_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15858_end_0 = const()[name = string("op_15858_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15858_end_mask_0 = const()[name = string("op_15858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15858_cast_fp16 = slice_by_index(begin = var_15858_begin_0, end = var_15858_end_0, end_mask = var_15858_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15858_cast_fp16")]; + tensor var_15865_begin_0 = const()[name = string("op_15865_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15865_end_0 = const()[name = string("op_15865_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15865_end_mask_0 = const()[name = string("op_15865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15865_cast_fp16 = slice_by_index(begin = var_15865_begin_0, end = var_15865_end_0, end_mask = var_15865_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15865_cast_fp16")]; + tensor var_15872_begin_0 = const()[name = string("op_15872_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15872_end_0 = const()[name = string("op_15872_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15872_end_mask_0 = const()[name = string("op_15872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15872_cast_fp16 = slice_by_index(begin = var_15872_begin_0, end = var_15872_end_0, end_mask = var_15872_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15872_cast_fp16")]; + tensor var_15879_begin_0 = const()[name = string("op_15879_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15879_end_0 = const()[name = string("op_15879_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15879_end_mask_0 = const()[name = string("op_15879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15879_cast_fp16 = slice_by_index(begin = var_15879_begin_0, end = var_15879_end_0, end_mask = var_15879_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15879_cast_fp16")]; + tensor var_15886_begin_0 = const()[name = string("op_15886_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15886_end_0 = const()[name = string("op_15886_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15886_end_mask_0 = const()[name = string("op_15886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15886_cast_fp16 = slice_by_index(begin = var_15886_begin_0, end = var_15886_end_0, end_mask = var_15886_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15886_cast_fp16")]; + tensor var_15893_begin_0 = const()[name = string("op_15893_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15893_end_0 = const()[name = string("op_15893_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15893_end_mask_0 = const()[name = string("op_15893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15893_cast_fp16 = slice_by_index(begin = var_15893_begin_0, end = var_15893_end_0, end_mask = var_15893_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15893_cast_fp16")]; + tensor var_15900_begin_0 = const()[name = string("op_15900_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15900_end_0 = const()[name = string("op_15900_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15900_end_mask_0 = const()[name = string("op_15900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15900_cast_fp16 = slice_by_index(begin = var_15900_begin_0, end = var_15900_end_0, end_mask = var_15900_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15900_cast_fp16")]; + tensor var_15907_begin_0 = const()[name = string("op_15907_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15907_end_0 = const()[name = string("op_15907_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15907_end_mask_0 = const()[name = string("op_15907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15907_cast_fp16 = slice_by_index(begin = var_15907_begin_0, end = var_15907_end_0, end_mask = var_15907_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15907_cast_fp16")]; + tensor var_15914_begin_0 = const()[name = string("op_15914_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15914_end_0 = const()[name = string("op_15914_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15914_end_mask_0 = const()[name = string("op_15914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15914_cast_fp16 = slice_by_index(begin = var_15914_begin_0, end = var_15914_end_0, end_mask = var_15914_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15914_cast_fp16")]; + tensor var_15921_begin_0 = const()[name = string("op_15921_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15921_end_0 = const()[name = string("op_15921_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15921_end_mask_0 = const()[name = string("op_15921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15921_cast_fp16 = slice_by_index(begin = var_15921_begin_0, end = var_15921_end_0, end_mask = var_15921_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15921_cast_fp16")]; + tensor var_15928_begin_0 = const()[name = string("op_15928_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15928_end_0 = const()[name = string("op_15928_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15928_end_mask_0 = const()[name = string("op_15928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15928_cast_fp16 = slice_by_index(begin = var_15928_begin_0, end = var_15928_end_0, end_mask = var_15928_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15928_cast_fp16")]; + tensor var_15935_begin_0 = const()[name = string("op_15935_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15935_end_0 = const()[name = string("op_15935_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15935_end_mask_0 = const()[name = string("op_15935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15935_cast_fp16 = slice_by_index(begin = var_15935_begin_0, end = var_15935_end_0, end_mask = var_15935_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15935_cast_fp16")]; + tensor var_15942_begin_0 = const()[name = string("op_15942_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15942_end_0 = const()[name = string("op_15942_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15942_end_mask_0 = const()[name = string("op_15942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15942_cast_fp16 = slice_by_index(begin = var_15942_begin_0, end = var_15942_end_0, end_mask = var_15942_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15942_cast_fp16")]; + tensor var_15949_begin_0 = const()[name = string("op_15949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15949_end_0 = const()[name = string("op_15949_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15949_end_mask_0 = const()[name = string("op_15949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15949_cast_fp16 = slice_by_index(begin = var_15949_begin_0, end = var_15949_end_0, end_mask = var_15949_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15949_cast_fp16")]; + tensor var_15956_begin_0 = const()[name = string("op_15956_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15956_end_0 = const()[name = string("op_15956_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15956_end_mask_0 = const()[name = string("op_15956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15956_cast_fp16 = slice_by_index(begin = var_15956_begin_0, end = var_15956_end_0, end_mask = var_15956_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15956_cast_fp16")]; + tensor var_15963_begin_0 = const()[name = string("op_15963_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15963_end_0 = const()[name = string("op_15963_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15963_end_mask_0 = const()[name = string("op_15963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15963_cast_fp16 = slice_by_index(begin = var_15963_begin_0, end = var_15963_end_0, end_mask = var_15963_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15963_cast_fp16")]; + tensor var_15970_begin_0 = const()[name = string("op_15970_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15970_end_0 = const()[name = string("op_15970_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15970_end_mask_0 = const()[name = string("op_15970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15970_cast_fp16 = slice_by_index(begin = var_15970_begin_0, end = var_15970_end_0, end_mask = var_15970_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15970_cast_fp16")]; + tensor var_15977_begin_0 = const()[name = string("op_15977_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15977_end_0 = const()[name = string("op_15977_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15977_end_mask_0 = const()[name = string("op_15977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15977_cast_fp16 = slice_by_index(begin = var_15977_begin_0, end = var_15977_end_0, end_mask = var_15977_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15977_cast_fp16")]; + tensor var_15984_begin_0 = const()[name = string("op_15984_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15984_end_0 = const()[name = string("op_15984_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15984_end_mask_0 = const()[name = string("op_15984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15984_cast_fp16 = slice_by_index(begin = var_15984_begin_0, end = var_15984_end_0, end_mask = var_15984_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15984_cast_fp16")]; + tensor var_15991_begin_0 = const()[name = string("op_15991_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15991_end_0 = const()[name = string("op_15991_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15991_end_mask_0 = const()[name = string("op_15991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15991_cast_fp16 = slice_by_index(begin = var_15991_begin_0, end = var_15991_end_0, end_mask = var_15991_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15991_cast_fp16")]; + tensor var_15998_begin_0 = const()[name = string("op_15998_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15998_end_0 = const()[name = string("op_15998_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15998_end_mask_0 = const()[name = string("op_15998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15998_cast_fp16 = slice_by_index(begin = var_15998_begin_0, end = var_15998_end_0, end_mask = var_15998_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15998_cast_fp16")]; + tensor var_16005_begin_0 = const()[name = string("op_16005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16005_end_0 = const()[name = string("op_16005_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16005_end_mask_0 = const()[name = string("op_16005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16005_cast_fp16 = slice_by_index(begin = var_16005_begin_0, end = var_16005_end_0, end_mask = var_16005_end_mask_0, x = var_15584_cast_fp16)[name = string("op_16005_cast_fp16")]; + tensor var_16012_begin_0 = const()[name = string("op_16012_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16012_end_0 = const()[name = string("op_16012_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16012_end_mask_0 = const()[name = string("op_16012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16012_cast_fp16 = slice_by_index(begin = var_16012_begin_0, end = var_16012_end_0, end_mask = var_16012_end_mask_0, x = var_15584_cast_fp16)[name = string("op_16012_cast_fp16")]; + tensor var_16019_begin_0 = const()[name = string("op_16019_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16019_end_0 = const()[name = string("op_16019_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16019_end_mask_0 = const()[name = string("op_16019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16019_cast_fp16 = slice_by_index(begin = var_16019_begin_0, end = var_16019_end_0, end_mask = var_16019_end_mask_0, x = var_15584_cast_fp16)[name = string("op_16019_cast_fp16")]; + tensor var_16026_begin_0 = const()[name = string("op_16026_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16026_end_0 = const()[name = string("op_16026_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16026_end_mask_0 = const()[name = string("op_16026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16026_cast_fp16 = slice_by_index(begin = var_16026_begin_0, end = var_16026_end_0, end_mask = var_16026_end_mask_0, x = var_15584_cast_fp16)[name = string("op_16026_cast_fp16")]; + tensor var_16033_begin_0 = const()[name = string("op_16033_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16033_end_0 = const()[name = string("op_16033_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16033_end_mask_0 = const()[name = string("op_16033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16033_cast_fp16 = slice_by_index(begin = var_16033_begin_0, end = var_16033_end_0, end_mask = var_16033_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16033_cast_fp16")]; + tensor var_16040_begin_0 = const()[name = string("op_16040_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16040_end_0 = const()[name = string("op_16040_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16040_end_mask_0 = const()[name = string("op_16040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16040_cast_fp16 = slice_by_index(begin = var_16040_begin_0, end = var_16040_end_0, end_mask = var_16040_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16040_cast_fp16")]; + tensor var_16047_begin_0 = const()[name = string("op_16047_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16047_end_0 = const()[name = string("op_16047_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16047_end_mask_0 = const()[name = string("op_16047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16047_cast_fp16 = slice_by_index(begin = var_16047_begin_0, end = var_16047_end_0, end_mask = var_16047_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16047_cast_fp16")]; + tensor var_16054_begin_0 = const()[name = string("op_16054_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16054_end_0 = const()[name = string("op_16054_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16054_end_mask_0 = const()[name = string("op_16054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16054_cast_fp16 = slice_by_index(begin = var_16054_begin_0, end = var_16054_end_0, end_mask = var_16054_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16054_cast_fp16")]; + tensor var_16061_begin_0 = const()[name = string("op_16061_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16061_end_0 = const()[name = string("op_16061_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16061_end_mask_0 = const()[name = string("op_16061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16061_cast_fp16 = slice_by_index(begin = var_16061_begin_0, end = var_16061_end_0, end_mask = var_16061_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16061_cast_fp16")]; + tensor var_16068_begin_0 = const()[name = string("op_16068_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16068_end_0 = const()[name = string("op_16068_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16068_end_mask_0 = const()[name = string("op_16068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16068_cast_fp16 = slice_by_index(begin = var_16068_begin_0, end = var_16068_end_0, end_mask = var_16068_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16068_cast_fp16")]; + tensor var_16075_begin_0 = const()[name = string("op_16075_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16075_end_0 = const()[name = string("op_16075_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16075_end_mask_0 = const()[name = string("op_16075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16075_cast_fp16 = slice_by_index(begin = var_16075_begin_0, end = var_16075_end_0, end_mask = var_16075_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16075_cast_fp16")]; + tensor var_16082_begin_0 = const()[name = string("op_16082_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16082_end_0 = const()[name = string("op_16082_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16082_end_mask_0 = const()[name = string("op_16082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16082_cast_fp16 = slice_by_index(begin = var_16082_begin_0, end = var_16082_end_0, end_mask = var_16082_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16082_cast_fp16")]; + tensor var_16089_begin_0 = const()[name = string("op_16089_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16089_end_0 = const()[name = string("op_16089_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16089_end_mask_0 = const()[name = string("op_16089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16089_cast_fp16 = slice_by_index(begin = var_16089_begin_0, end = var_16089_end_0, end_mask = var_16089_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16089_cast_fp16")]; + tensor var_16096_begin_0 = const()[name = string("op_16096_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16096_end_0 = const()[name = string("op_16096_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16096_end_mask_0 = const()[name = string("op_16096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16096_cast_fp16 = slice_by_index(begin = var_16096_begin_0, end = var_16096_end_0, end_mask = var_16096_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16096_cast_fp16")]; + tensor var_16103_begin_0 = const()[name = string("op_16103_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16103_end_0 = const()[name = string("op_16103_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16103_end_mask_0 = const()[name = string("op_16103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16103_cast_fp16 = slice_by_index(begin = var_16103_begin_0, end = var_16103_end_0, end_mask = var_16103_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16103_cast_fp16")]; + tensor var_16110_begin_0 = const()[name = string("op_16110_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16110_end_0 = const()[name = string("op_16110_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16110_end_mask_0 = const()[name = string("op_16110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16110_cast_fp16 = slice_by_index(begin = var_16110_begin_0, end = var_16110_end_0, end_mask = var_16110_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16110_cast_fp16")]; + tensor var_16117_begin_0 = const()[name = string("op_16117_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16117_end_0 = const()[name = string("op_16117_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16117_end_mask_0 = const()[name = string("op_16117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16117_cast_fp16 = slice_by_index(begin = var_16117_begin_0, end = var_16117_end_0, end_mask = var_16117_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16117_cast_fp16")]; + tensor var_16124_begin_0 = const()[name = string("op_16124_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16124_end_0 = const()[name = string("op_16124_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16124_end_mask_0 = const()[name = string("op_16124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16124_cast_fp16 = slice_by_index(begin = var_16124_begin_0, end = var_16124_end_0, end_mask = var_16124_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16124_cast_fp16")]; + tensor var_16131_begin_0 = const()[name = string("op_16131_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16131_end_0 = const()[name = string("op_16131_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16131_end_mask_0 = const()[name = string("op_16131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16131_cast_fp16 = slice_by_index(begin = var_16131_begin_0, end = var_16131_end_0, end_mask = var_16131_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16131_cast_fp16")]; + tensor var_16138_begin_0 = const()[name = string("op_16138_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16138_end_0 = const()[name = string("op_16138_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16138_end_mask_0 = const()[name = string("op_16138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16138_cast_fp16 = slice_by_index(begin = var_16138_begin_0, end = var_16138_end_0, end_mask = var_16138_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16138_cast_fp16")]; + tensor var_16145_begin_0 = const()[name = string("op_16145_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16145_end_0 = const()[name = string("op_16145_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16145_end_mask_0 = const()[name = string("op_16145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16145_cast_fp16 = slice_by_index(begin = var_16145_begin_0, end = var_16145_end_0, end_mask = var_16145_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16145_cast_fp16")]; + tensor var_16152_begin_0 = const()[name = string("op_16152_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16152_end_0 = const()[name = string("op_16152_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16152_end_mask_0 = const()[name = string("op_16152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16152_cast_fp16 = slice_by_index(begin = var_16152_begin_0, end = var_16152_end_0, end_mask = var_16152_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16152_cast_fp16")]; + tensor var_16159_begin_0 = const()[name = string("op_16159_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16159_end_0 = const()[name = string("op_16159_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16159_end_mask_0 = const()[name = string("op_16159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16159_cast_fp16 = slice_by_index(begin = var_16159_begin_0, end = var_16159_end_0, end_mask = var_16159_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16159_cast_fp16")]; + tensor var_16166_begin_0 = const()[name = string("op_16166_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16166_end_0 = const()[name = string("op_16166_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16166_end_mask_0 = const()[name = string("op_16166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16166_cast_fp16 = slice_by_index(begin = var_16166_begin_0, end = var_16166_end_0, end_mask = var_16166_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16166_cast_fp16")]; + tensor k_21_perm_0 = const()[name = string("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_16171_begin_0 = const()[name = string("op_16171_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16171_end_0 = const()[name = string("op_16171_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_16171_end_mask_0 = const()[name = string("op_16171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = string("transpose_21")]; + tensor var_16171_cast_fp16 = slice_by_index(begin = var_16171_begin_0, end = var_16171_end_0, end_mask = var_16171_end_mask_0, x = k_21_cast_fp16)[name = string("op_16171_cast_fp16")]; + tensor var_16175_begin_0 = const()[name = string("op_16175_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_16175_end_0 = const()[name = string("op_16175_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_16175_end_mask_0 = const()[name = string("op_16175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16175_cast_fp16 = slice_by_index(begin = var_16175_begin_0, end = var_16175_end_0, end_mask = var_16175_end_mask_0, x = k_21_cast_fp16)[name = string("op_16175_cast_fp16")]; + tensor var_16179_begin_0 = const()[name = string("op_16179_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_16179_end_0 = const()[name = string("op_16179_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_16179_end_mask_0 = const()[name = string("op_16179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16179_cast_fp16 = slice_by_index(begin = var_16179_begin_0, end = var_16179_end_0, end_mask = var_16179_end_mask_0, x = k_21_cast_fp16)[name = string("op_16179_cast_fp16")]; + tensor var_16183_begin_0 = const()[name = string("op_16183_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_16183_end_0 = const()[name = string("op_16183_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_16183_end_mask_0 = const()[name = string("op_16183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16183_cast_fp16 = slice_by_index(begin = var_16183_begin_0, end = var_16183_end_0, end_mask = var_16183_end_mask_0, x = k_21_cast_fp16)[name = string("op_16183_cast_fp16")]; + tensor var_16187_begin_0 = const()[name = string("op_16187_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_16187_end_0 = const()[name = string("op_16187_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_16187_end_mask_0 = const()[name = string("op_16187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16187_cast_fp16 = slice_by_index(begin = var_16187_begin_0, end = var_16187_end_0, end_mask = var_16187_end_mask_0, x = k_21_cast_fp16)[name = string("op_16187_cast_fp16")]; + tensor var_16191_begin_0 = const()[name = string("op_16191_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_16191_end_0 = const()[name = string("op_16191_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_16191_end_mask_0 = const()[name = string("op_16191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16191_cast_fp16 = slice_by_index(begin = var_16191_begin_0, end = var_16191_end_0, end_mask = var_16191_end_mask_0, x = k_21_cast_fp16)[name = string("op_16191_cast_fp16")]; + tensor var_16195_begin_0 = const()[name = string("op_16195_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_16195_end_0 = const()[name = string("op_16195_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_16195_end_mask_0 = const()[name = string("op_16195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16195_cast_fp16 = slice_by_index(begin = var_16195_begin_0, end = var_16195_end_0, end_mask = var_16195_end_mask_0, x = k_21_cast_fp16)[name = string("op_16195_cast_fp16")]; + tensor var_16199_begin_0 = const()[name = string("op_16199_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_16199_end_0 = const()[name = string("op_16199_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_16199_end_mask_0 = const()[name = string("op_16199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16199_cast_fp16 = slice_by_index(begin = var_16199_begin_0, end = var_16199_end_0, end_mask = var_16199_end_mask_0, x = k_21_cast_fp16)[name = string("op_16199_cast_fp16")]; + tensor var_16203_begin_0 = const()[name = string("op_16203_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_16203_end_0 = const()[name = string("op_16203_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_16203_end_mask_0 = const()[name = string("op_16203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16203_cast_fp16 = slice_by_index(begin = var_16203_begin_0, end = var_16203_end_0, end_mask = var_16203_end_mask_0, x = k_21_cast_fp16)[name = string("op_16203_cast_fp16")]; + tensor var_16207_begin_0 = const()[name = string("op_16207_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_16207_end_0 = const()[name = string("op_16207_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_16207_end_mask_0 = const()[name = string("op_16207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16207_cast_fp16 = slice_by_index(begin = var_16207_begin_0, end = var_16207_end_0, end_mask = var_16207_end_mask_0, x = k_21_cast_fp16)[name = string("op_16207_cast_fp16")]; + tensor var_16211_begin_0 = const()[name = string("op_16211_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_16211_end_0 = const()[name = string("op_16211_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_16211_end_mask_0 = const()[name = string("op_16211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16211_cast_fp16 = slice_by_index(begin = var_16211_begin_0, end = var_16211_end_0, end_mask = var_16211_end_mask_0, x = k_21_cast_fp16)[name = string("op_16211_cast_fp16")]; + tensor var_16215_begin_0 = const()[name = string("op_16215_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_16215_end_0 = const()[name = string("op_16215_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_16215_end_mask_0 = const()[name = string("op_16215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16215_cast_fp16 = slice_by_index(begin = var_16215_begin_0, end = var_16215_end_0, end_mask = var_16215_end_mask_0, x = k_21_cast_fp16)[name = string("op_16215_cast_fp16")]; + tensor var_16219_begin_0 = const()[name = string("op_16219_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_16219_end_0 = const()[name = string("op_16219_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_16219_end_mask_0 = const()[name = string("op_16219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16219_cast_fp16 = slice_by_index(begin = var_16219_begin_0, end = var_16219_end_0, end_mask = var_16219_end_mask_0, x = k_21_cast_fp16)[name = string("op_16219_cast_fp16")]; + tensor var_16223_begin_0 = const()[name = string("op_16223_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_16223_end_0 = const()[name = string("op_16223_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_16223_end_mask_0 = const()[name = string("op_16223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16223_cast_fp16 = slice_by_index(begin = var_16223_begin_0, end = var_16223_end_0, end_mask = var_16223_end_mask_0, x = k_21_cast_fp16)[name = string("op_16223_cast_fp16")]; + tensor var_16227_begin_0 = const()[name = string("op_16227_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_16227_end_0 = const()[name = string("op_16227_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_16227_end_mask_0 = const()[name = string("op_16227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16227_cast_fp16 = slice_by_index(begin = var_16227_begin_0, end = var_16227_end_0, end_mask = var_16227_end_mask_0, x = k_21_cast_fp16)[name = string("op_16227_cast_fp16")]; + tensor var_16231_begin_0 = const()[name = string("op_16231_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_16231_end_0 = const()[name = string("op_16231_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_16231_end_mask_0 = const()[name = string("op_16231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16231_cast_fp16 = slice_by_index(begin = var_16231_begin_0, end = var_16231_end_0, end_mask = var_16231_end_mask_0, x = k_21_cast_fp16)[name = string("op_16231_cast_fp16")]; + tensor var_16235_begin_0 = const()[name = string("op_16235_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_16235_end_0 = const()[name = string("op_16235_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_16235_end_mask_0 = const()[name = string("op_16235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16235_cast_fp16 = slice_by_index(begin = var_16235_begin_0, end = var_16235_end_0, end_mask = var_16235_end_mask_0, x = k_21_cast_fp16)[name = string("op_16235_cast_fp16")]; + tensor var_16239_begin_0 = const()[name = string("op_16239_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_16239_end_0 = const()[name = string("op_16239_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_16239_end_mask_0 = const()[name = string("op_16239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16239_cast_fp16 = slice_by_index(begin = var_16239_begin_0, end = var_16239_end_0, end_mask = var_16239_end_mask_0, x = k_21_cast_fp16)[name = string("op_16239_cast_fp16")]; + tensor var_16243_begin_0 = const()[name = string("op_16243_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_16243_end_0 = const()[name = string("op_16243_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_16243_end_mask_0 = const()[name = string("op_16243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16243_cast_fp16 = slice_by_index(begin = var_16243_begin_0, end = var_16243_end_0, end_mask = var_16243_end_mask_0, x = k_21_cast_fp16)[name = string("op_16243_cast_fp16")]; + tensor var_16247_begin_0 = const()[name = string("op_16247_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_16247_end_0 = const()[name = string("op_16247_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_16247_end_mask_0 = const()[name = string("op_16247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16247_cast_fp16 = slice_by_index(begin = var_16247_begin_0, end = var_16247_end_0, end_mask = var_16247_end_mask_0, x = k_21_cast_fp16)[name = string("op_16247_cast_fp16")]; + tensor var_16249_begin_0 = const()[name = string("op_16249_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16249_end_0 = const()[name = string("op_16249_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16249_end_mask_0 = const()[name = string("op_16249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16249_cast_fp16 = slice_by_index(begin = var_16249_begin_0, end = var_16249_end_0, end_mask = var_16249_end_mask_0, x = value_21_cast_fp16)[name = string("op_16249_cast_fp16")]; + tensor var_16253_begin_0 = const()[name = string("op_16253_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16253_end_0 = const()[name = string("op_16253_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_16253_end_mask_0 = const()[name = string("op_16253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16253_cast_fp16 = slice_by_index(begin = var_16253_begin_0, end = var_16253_end_0, end_mask = var_16253_end_mask_0, x = value_21_cast_fp16)[name = string("op_16253_cast_fp16")]; + tensor var_16257_begin_0 = const()[name = string("op_16257_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16257_end_0 = const()[name = string("op_16257_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_16257_end_mask_0 = const()[name = string("op_16257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16257_cast_fp16 = slice_by_index(begin = var_16257_begin_0, end = var_16257_end_0, end_mask = var_16257_end_mask_0, x = value_21_cast_fp16)[name = string("op_16257_cast_fp16")]; + tensor var_16261_begin_0 = const()[name = string("op_16261_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16261_end_0 = const()[name = string("op_16261_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_16261_end_mask_0 = const()[name = string("op_16261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16261_cast_fp16 = slice_by_index(begin = var_16261_begin_0, end = var_16261_end_0, end_mask = var_16261_end_mask_0, x = value_21_cast_fp16)[name = string("op_16261_cast_fp16")]; + tensor var_16265_begin_0 = const()[name = string("op_16265_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16265_end_0 = const()[name = string("op_16265_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_16265_end_mask_0 = const()[name = string("op_16265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16265_cast_fp16 = slice_by_index(begin = var_16265_begin_0, end = var_16265_end_0, end_mask = var_16265_end_mask_0, x = value_21_cast_fp16)[name = string("op_16265_cast_fp16")]; + tensor var_16269_begin_0 = const()[name = string("op_16269_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16269_end_0 = const()[name = string("op_16269_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_16269_end_mask_0 = const()[name = string("op_16269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16269_cast_fp16 = slice_by_index(begin = var_16269_begin_0, end = var_16269_end_0, end_mask = var_16269_end_mask_0, x = value_21_cast_fp16)[name = string("op_16269_cast_fp16")]; + tensor var_16273_begin_0 = const()[name = string("op_16273_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16273_end_0 = const()[name = string("op_16273_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_16273_end_mask_0 = const()[name = string("op_16273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16273_cast_fp16 = slice_by_index(begin = var_16273_begin_0, end = var_16273_end_0, end_mask = var_16273_end_mask_0, x = value_21_cast_fp16)[name = string("op_16273_cast_fp16")]; + tensor var_16277_begin_0 = const()[name = string("op_16277_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16277_end_0 = const()[name = string("op_16277_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_16277_end_mask_0 = const()[name = string("op_16277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16277_cast_fp16 = slice_by_index(begin = var_16277_begin_0, end = var_16277_end_0, end_mask = var_16277_end_mask_0, x = value_21_cast_fp16)[name = string("op_16277_cast_fp16")]; + tensor var_16281_begin_0 = const()[name = string("op_16281_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16281_end_0 = const()[name = string("op_16281_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_16281_end_mask_0 = const()[name = string("op_16281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16281_cast_fp16 = slice_by_index(begin = var_16281_begin_0, end = var_16281_end_0, end_mask = var_16281_end_mask_0, x = value_21_cast_fp16)[name = string("op_16281_cast_fp16")]; + tensor var_16285_begin_0 = const()[name = string("op_16285_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16285_end_0 = const()[name = string("op_16285_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_16285_end_mask_0 = const()[name = string("op_16285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16285_cast_fp16 = slice_by_index(begin = var_16285_begin_0, end = var_16285_end_0, end_mask = var_16285_end_mask_0, x = value_21_cast_fp16)[name = string("op_16285_cast_fp16")]; + tensor var_16289_begin_0 = const()[name = string("op_16289_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16289_end_0 = const()[name = string("op_16289_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_16289_end_mask_0 = const()[name = string("op_16289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16289_cast_fp16 = slice_by_index(begin = var_16289_begin_0, end = var_16289_end_0, end_mask = var_16289_end_mask_0, x = value_21_cast_fp16)[name = string("op_16289_cast_fp16")]; + tensor var_16293_begin_0 = const()[name = string("op_16293_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16293_end_0 = const()[name = string("op_16293_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_16293_end_mask_0 = const()[name = string("op_16293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16293_cast_fp16 = slice_by_index(begin = var_16293_begin_0, end = var_16293_end_0, end_mask = var_16293_end_mask_0, x = value_21_cast_fp16)[name = string("op_16293_cast_fp16")]; + tensor var_16297_begin_0 = const()[name = string("op_16297_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16297_end_0 = const()[name = string("op_16297_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_16297_end_mask_0 = const()[name = string("op_16297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16297_cast_fp16 = slice_by_index(begin = var_16297_begin_0, end = var_16297_end_0, end_mask = var_16297_end_mask_0, x = value_21_cast_fp16)[name = string("op_16297_cast_fp16")]; + tensor var_16301_begin_0 = const()[name = string("op_16301_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16301_end_0 = const()[name = string("op_16301_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_16301_end_mask_0 = const()[name = string("op_16301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16301_cast_fp16 = slice_by_index(begin = var_16301_begin_0, end = var_16301_end_0, end_mask = var_16301_end_mask_0, x = value_21_cast_fp16)[name = string("op_16301_cast_fp16")]; + tensor var_16305_begin_0 = const()[name = string("op_16305_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16305_end_0 = const()[name = string("op_16305_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_16305_end_mask_0 = const()[name = string("op_16305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16305_cast_fp16 = slice_by_index(begin = var_16305_begin_0, end = var_16305_end_0, end_mask = var_16305_end_mask_0, x = value_21_cast_fp16)[name = string("op_16305_cast_fp16")]; + tensor var_16309_begin_0 = const()[name = string("op_16309_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16309_end_0 = const()[name = string("op_16309_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_16309_end_mask_0 = const()[name = string("op_16309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16309_cast_fp16 = slice_by_index(begin = var_16309_begin_0, end = var_16309_end_0, end_mask = var_16309_end_mask_0, x = value_21_cast_fp16)[name = string("op_16309_cast_fp16")]; + tensor var_16313_begin_0 = const()[name = string("op_16313_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16313_end_0 = const()[name = string("op_16313_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_16313_end_mask_0 = const()[name = string("op_16313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16313_cast_fp16 = slice_by_index(begin = var_16313_begin_0, end = var_16313_end_0, end_mask = var_16313_end_mask_0, x = value_21_cast_fp16)[name = string("op_16313_cast_fp16")]; + tensor var_16317_begin_0 = const()[name = string("op_16317_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16317_end_0 = const()[name = string("op_16317_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_16317_end_mask_0 = const()[name = string("op_16317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16317_cast_fp16 = slice_by_index(begin = var_16317_begin_0, end = var_16317_end_0, end_mask = var_16317_end_mask_0, x = value_21_cast_fp16)[name = string("op_16317_cast_fp16")]; + tensor var_16321_begin_0 = const()[name = string("op_16321_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16321_end_0 = const()[name = string("op_16321_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_16321_end_mask_0 = const()[name = string("op_16321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16321_cast_fp16 = slice_by_index(begin = var_16321_begin_0, end = var_16321_end_0, end_mask = var_16321_end_mask_0, x = value_21_cast_fp16)[name = string("op_16321_cast_fp16")]; + tensor var_16325_begin_0 = const()[name = string("op_16325_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16325_end_0 = const()[name = string("op_16325_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_16325_end_mask_0 = const()[name = string("op_16325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16325_cast_fp16 = slice_by_index(begin = var_16325_begin_0, end = var_16325_end_0, end_mask = var_16325_end_mask_0, x = value_21_cast_fp16)[name = string("op_16325_cast_fp16")]; + string _SplitHeadsQ__mh_w_1601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1601_equation_0, values = (var_16171_cast_fp16, var_15613_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1601_cast_fp16")]; + string _SplitHeadsQ__mh_w_1603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1603_equation_0, values = (var_16171_cast_fp16, var_15620_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1603_cast_fp16")]; + string _SplitHeadsQ__mh_w_1605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1605_equation_0, values = (var_16171_cast_fp16, var_15627_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1605_cast_fp16")]; + string _SplitHeadsQ__mh_w_1607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1607_equation_0, values = (var_16171_cast_fp16, var_15634_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1607_cast_fp16")]; + string _SplitHeadsQ__mh_w_1609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1609_equation_0, values = (var_16175_cast_fp16, var_15641_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1609_cast_fp16")]; + string _SplitHeadsQ__mh_w_1611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1611_equation_0, values = (var_16175_cast_fp16, var_15648_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1611_cast_fp16")]; + string _SplitHeadsQ__mh_w_1613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1613_equation_0, values = (var_16175_cast_fp16, var_15655_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1613_cast_fp16")]; + string _SplitHeadsQ__mh_w_1615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1615_equation_0, values = (var_16175_cast_fp16, var_15662_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1615_cast_fp16")]; + string _SplitHeadsQ__mh_w_1617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1617_equation_0, values = (var_16179_cast_fp16, var_15669_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1617_cast_fp16")]; + string _SplitHeadsQ__mh_w_1619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1619_equation_0, values = (var_16179_cast_fp16, var_15676_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1619_cast_fp16")]; + string _SplitHeadsQ__mh_w_1621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1621_equation_0, values = (var_16179_cast_fp16, var_15683_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1621_cast_fp16")]; + string _SplitHeadsQ__mh_w_1623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1623_equation_0, values = (var_16179_cast_fp16, var_15690_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1623_cast_fp16")]; + string _SplitHeadsQ__mh_w_1625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1625_equation_0, values = (var_16183_cast_fp16, var_15697_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1625_cast_fp16")]; + string _SplitHeadsQ__mh_w_1627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1627_equation_0, values = (var_16183_cast_fp16, var_15704_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1627_cast_fp16")]; + string _SplitHeadsQ__mh_w_1629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1629_equation_0, values = (var_16183_cast_fp16, var_15711_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1629_cast_fp16")]; + string _SplitHeadsQ__mh_w_1631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1631_equation_0, values = (var_16183_cast_fp16, var_15718_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1631_cast_fp16")]; + string _SplitHeadsQ__mh_w_1633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1633_equation_0, values = (var_16187_cast_fp16, var_15725_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1633_cast_fp16")]; + string _SplitHeadsQ__mh_w_1635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1635_equation_0, values = (var_16187_cast_fp16, var_15732_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1635_cast_fp16")]; + string _SplitHeadsQ__mh_w_1637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1637_equation_0, values = (var_16187_cast_fp16, var_15739_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1637_cast_fp16")]; + string _SplitHeadsQ__mh_w_1639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1639_equation_0, values = (var_16187_cast_fp16, var_15746_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1639_cast_fp16")]; + string _SplitHeadsQ__mh_w_1641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1641_equation_0, values = (var_16191_cast_fp16, var_15753_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1641_cast_fp16")]; + string _SplitHeadsQ__mh_w_1643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1643_equation_0, values = (var_16191_cast_fp16, var_15760_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1643_cast_fp16")]; + string _SplitHeadsQ__mh_w_1645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1645_equation_0, values = (var_16191_cast_fp16, var_15767_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1645_cast_fp16")]; + string _SplitHeadsQ__mh_w_1647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1647_equation_0, values = (var_16191_cast_fp16, var_15774_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1647_cast_fp16")]; + string _SplitHeadsQ__mh_w_1649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1649_equation_0, values = (var_16195_cast_fp16, var_15781_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1649_cast_fp16")]; + string _SplitHeadsQ__mh_w_1651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1651_equation_0, values = (var_16195_cast_fp16, var_15788_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1651_cast_fp16")]; + string _SplitHeadsQ__mh_w_1653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1653_equation_0, values = (var_16195_cast_fp16, var_15795_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1653_cast_fp16")]; + string _SplitHeadsQ__mh_w_1655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1655_equation_0, values = (var_16195_cast_fp16, var_15802_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1655_cast_fp16")]; + string _SplitHeadsQ__mh_w_1657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1657_equation_0, values = (var_16199_cast_fp16, var_15809_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1657_cast_fp16")]; + string _SplitHeadsQ__mh_w_1659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1659_equation_0, values = (var_16199_cast_fp16, var_15816_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1659_cast_fp16")]; + string _SplitHeadsQ__mh_w_1661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1661_equation_0, values = (var_16199_cast_fp16, var_15823_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1661_cast_fp16")]; + string _SplitHeadsQ__mh_w_1663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1663_equation_0, values = (var_16199_cast_fp16, var_15830_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1663_cast_fp16")]; + string _SplitHeadsQ__mh_w_1665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1665_equation_0, values = (var_16203_cast_fp16, var_15837_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1665_cast_fp16")]; + string _SplitHeadsQ__mh_w_1667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1667_equation_0, values = (var_16203_cast_fp16, var_15844_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1667_cast_fp16")]; + string _SplitHeadsQ__mh_w_1669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1669_equation_0, values = (var_16203_cast_fp16, var_15851_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1669_cast_fp16")]; + string _SplitHeadsQ__mh_w_1671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1671_equation_0, values = (var_16203_cast_fp16, var_15858_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1671_cast_fp16")]; + string _SplitHeadsQ__mh_w_1673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1673_equation_0, values = (var_16207_cast_fp16, var_15865_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1673_cast_fp16")]; + string _SplitHeadsQ__mh_w_1675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1675_equation_0, values = (var_16207_cast_fp16, var_15872_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1675_cast_fp16")]; + string _SplitHeadsQ__mh_w_1677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1677_equation_0, values = (var_16207_cast_fp16, var_15879_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1677_cast_fp16")]; + string _SplitHeadsQ__mh_w_1679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1679_equation_0, values = (var_16207_cast_fp16, var_15886_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1679_cast_fp16")]; + string _SplitHeadsQ__mh_w_1681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1681_equation_0, values = (var_16211_cast_fp16, var_15893_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1681_cast_fp16")]; + string _SplitHeadsQ__mh_w_1683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1683_equation_0, values = (var_16211_cast_fp16, var_15900_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1683_cast_fp16")]; + string _SplitHeadsQ__mh_w_1685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1685_equation_0, values = (var_16211_cast_fp16, var_15907_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1685_cast_fp16")]; + string _SplitHeadsQ__mh_w_1687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1687_equation_0, values = (var_16211_cast_fp16, var_15914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1687_cast_fp16")]; + string _SplitHeadsQ__mh_w_1689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1689_equation_0, values = (var_16215_cast_fp16, var_15921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1689_cast_fp16")]; + string _SplitHeadsQ__mh_w_1691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1691_equation_0, values = (var_16215_cast_fp16, var_15928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1691_cast_fp16")]; + string _SplitHeadsQ__mh_w_1693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1693_equation_0, values = (var_16215_cast_fp16, var_15935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1693_cast_fp16")]; + string _SplitHeadsQ__mh_w_1695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1695_equation_0, values = (var_16215_cast_fp16, var_15942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1695_cast_fp16")]; + string _SplitHeadsQ__mh_w_1697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1697_equation_0, values = (var_16219_cast_fp16, var_15949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1697_cast_fp16")]; + string _SplitHeadsQ__mh_w_1699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1699_equation_0, values = (var_16219_cast_fp16, var_15956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1699_cast_fp16")]; + string _SplitHeadsQ__mh_w_1701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1701_equation_0, values = (var_16219_cast_fp16, var_15963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1701_cast_fp16")]; + string _SplitHeadsQ__mh_w_1703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1703_equation_0, values = (var_16219_cast_fp16, var_15970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1703_cast_fp16")]; + string _SplitHeadsQ__mh_w_1705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1705_equation_0, values = (var_16223_cast_fp16, var_15977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1705_cast_fp16")]; + string _SplitHeadsQ__mh_w_1707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1707_equation_0, values = (var_16223_cast_fp16, var_15984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1707_cast_fp16")]; + string _SplitHeadsQ__mh_w_1709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1709_equation_0, values = (var_16223_cast_fp16, var_15991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1709_cast_fp16")]; + string _SplitHeadsQ__mh_w_1711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1711_equation_0, values = (var_16223_cast_fp16, var_15998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1711_cast_fp16")]; + string _SplitHeadsQ__mh_w_1713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1713_equation_0, values = (var_16227_cast_fp16, var_16005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1713_cast_fp16")]; + string _SplitHeadsQ__mh_w_1715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1715_equation_0, values = (var_16227_cast_fp16, var_16012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1715_cast_fp16")]; + string _SplitHeadsQ__mh_w_1717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1717_equation_0, values = (var_16227_cast_fp16, var_16019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1717_cast_fp16")]; + string _SplitHeadsQ__mh_w_1719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1719_equation_0, values = (var_16227_cast_fp16, var_16026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1719_cast_fp16")]; + string _SplitHeadsQ__mh_w_1721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1721_equation_0, values = (var_16231_cast_fp16, var_16033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1721_cast_fp16")]; + string _SplitHeadsQ__mh_w_1723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1723_equation_0, values = (var_16231_cast_fp16, var_16040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1723_cast_fp16")]; + string _SplitHeadsQ__mh_w_1725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1725_equation_0, values = (var_16231_cast_fp16, var_16047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1725_cast_fp16")]; + string _SplitHeadsQ__mh_w_1727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1727_equation_0, values = (var_16231_cast_fp16, var_16054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1727_cast_fp16")]; + string _SplitHeadsQ__mh_w_1729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1729_equation_0, values = (var_16235_cast_fp16, var_16061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1729_cast_fp16")]; + string _SplitHeadsQ__mh_w_1731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1731_equation_0, values = (var_16235_cast_fp16, var_16068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1731_cast_fp16")]; + string _SplitHeadsQ__mh_w_1733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1733_equation_0, values = (var_16235_cast_fp16, var_16075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1733_cast_fp16")]; + string _SplitHeadsQ__mh_w_1735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1735_equation_0, values = (var_16235_cast_fp16, var_16082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1735_cast_fp16")]; + string _SplitHeadsQ__mh_w_1737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1737_equation_0, values = (var_16239_cast_fp16, var_16089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1737_cast_fp16")]; + string _SplitHeadsQ__mh_w_1739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1739_equation_0, values = (var_16239_cast_fp16, var_16096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1739_cast_fp16")]; + string _SplitHeadsQ__mh_w_1741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1741_equation_0, values = (var_16239_cast_fp16, var_16103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1741_cast_fp16")]; + string _SplitHeadsQ__mh_w_1743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1743_equation_0, values = (var_16239_cast_fp16, var_16110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1743_cast_fp16")]; + string _SplitHeadsQ__mh_w_1745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1745_equation_0, values = (var_16243_cast_fp16, var_16117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1745_cast_fp16")]; + string _SplitHeadsQ__mh_w_1747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1747_equation_0, values = (var_16243_cast_fp16, var_16124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1747_cast_fp16")]; + string _SplitHeadsQ__mh_w_1749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1749_equation_0, values = (var_16243_cast_fp16, var_16131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1749_cast_fp16")]; + string _SplitHeadsQ__mh_w_1751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1751_equation_0, values = (var_16243_cast_fp16, var_16138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1751_cast_fp16")]; + string _SplitHeadsQ__mh_w_1753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1753_equation_0, values = (var_16247_cast_fp16, var_16145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1753_cast_fp16")]; + string _SplitHeadsQ__mh_w_1755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1755_equation_0, values = (var_16247_cast_fp16, var_16152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1755_cast_fp16")]; + string _SplitHeadsQ__mh_w_1757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1757_equation_0, values = (var_16247_cast_fp16, var_16159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1757_cast_fp16")]; + string _SplitHeadsQ__mh_w_1759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1759_equation_0, values = (var_16247_cast_fp16, var_16166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1759_cast_fp16")]; + fp16 var_16488_to_fp16 = const()[name = string("op_16488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1601_cast_fp16, y = var_16488_to_fp16)[name = string("aw_chunk_1601_cast_fp16")]; + fp16 var_16490_to_fp16 = const()[name = string("op_16490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1603_cast_fp16, y = var_16490_to_fp16)[name = string("aw_chunk_1603_cast_fp16")]; + fp16 var_16492_to_fp16 = const()[name = string("op_16492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1605_cast_fp16, y = var_16492_to_fp16)[name = string("aw_chunk_1605_cast_fp16")]; + fp16 var_16494_to_fp16 = const()[name = string("op_16494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1607_cast_fp16, y = var_16494_to_fp16)[name = string("aw_chunk_1607_cast_fp16")]; + fp16 var_16496_to_fp16 = const()[name = string("op_16496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1609_cast_fp16, y = var_16496_to_fp16)[name = string("aw_chunk_1609_cast_fp16")]; + fp16 var_16498_to_fp16 = const()[name = string("op_16498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1611_cast_fp16, y = var_16498_to_fp16)[name = string("aw_chunk_1611_cast_fp16")]; + fp16 var_16500_to_fp16 = const()[name = string("op_16500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1613_cast_fp16, y = var_16500_to_fp16)[name = string("aw_chunk_1613_cast_fp16")]; + fp16 var_16502_to_fp16 = const()[name = string("op_16502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1615_cast_fp16, y = var_16502_to_fp16)[name = string("aw_chunk_1615_cast_fp16")]; + fp16 var_16504_to_fp16 = const()[name = string("op_16504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1617_cast_fp16, y = var_16504_to_fp16)[name = string("aw_chunk_1617_cast_fp16")]; + fp16 var_16506_to_fp16 = const()[name = string("op_16506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1619_cast_fp16, y = var_16506_to_fp16)[name = string("aw_chunk_1619_cast_fp16")]; + fp16 var_16508_to_fp16 = const()[name = string("op_16508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1621_cast_fp16, y = var_16508_to_fp16)[name = string("aw_chunk_1621_cast_fp16")]; + fp16 var_16510_to_fp16 = const()[name = string("op_16510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1623_cast_fp16, y = var_16510_to_fp16)[name = string("aw_chunk_1623_cast_fp16")]; + fp16 var_16512_to_fp16 = const()[name = string("op_16512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1625_cast_fp16, y = var_16512_to_fp16)[name = string("aw_chunk_1625_cast_fp16")]; + fp16 var_16514_to_fp16 = const()[name = string("op_16514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1627_cast_fp16, y = var_16514_to_fp16)[name = string("aw_chunk_1627_cast_fp16")]; + fp16 var_16516_to_fp16 = const()[name = string("op_16516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1629_cast_fp16, y = var_16516_to_fp16)[name = string("aw_chunk_1629_cast_fp16")]; + fp16 var_16518_to_fp16 = const()[name = string("op_16518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1631_cast_fp16, y = var_16518_to_fp16)[name = string("aw_chunk_1631_cast_fp16")]; + fp16 var_16520_to_fp16 = const()[name = string("op_16520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1633_cast_fp16, y = var_16520_to_fp16)[name = string("aw_chunk_1633_cast_fp16")]; + fp16 var_16522_to_fp16 = const()[name = string("op_16522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1635_cast_fp16, y = var_16522_to_fp16)[name = string("aw_chunk_1635_cast_fp16")]; + fp16 var_16524_to_fp16 = const()[name = string("op_16524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1637_cast_fp16, y = var_16524_to_fp16)[name = string("aw_chunk_1637_cast_fp16")]; + fp16 var_16526_to_fp16 = const()[name = string("op_16526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1639_cast_fp16, y = var_16526_to_fp16)[name = string("aw_chunk_1639_cast_fp16")]; + fp16 var_16528_to_fp16 = const()[name = string("op_16528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1641_cast_fp16, y = var_16528_to_fp16)[name = string("aw_chunk_1641_cast_fp16")]; + fp16 var_16530_to_fp16 = const()[name = string("op_16530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1643_cast_fp16, y = var_16530_to_fp16)[name = string("aw_chunk_1643_cast_fp16")]; + fp16 var_16532_to_fp16 = const()[name = string("op_16532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1645_cast_fp16, y = var_16532_to_fp16)[name = string("aw_chunk_1645_cast_fp16")]; + fp16 var_16534_to_fp16 = const()[name = string("op_16534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1647_cast_fp16, y = var_16534_to_fp16)[name = string("aw_chunk_1647_cast_fp16")]; + fp16 var_16536_to_fp16 = const()[name = string("op_16536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1649_cast_fp16, y = var_16536_to_fp16)[name = string("aw_chunk_1649_cast_fp16")]; + fp16 var_16538_to_fp16 = const()[name = string("op_16538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1651_cast_fp16, y = var_16538_to_fp16)[name = string("aw_chunk_1651_cast_fp16")]; + fp16 var_16540_to_fp16 = const()[name = string("op_16540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1653_cast_fp16, y = var_16540_to_fp16)[name = string("aw_chunk_1653_cast_fp16")]; + fp16 var_16542_to_fp16 = const()[name = string("op_16542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1655_cast_fp16, y = var_16542_to_fp16)[name = string("aw_chunk_1655_cast_fp16")]; + fp16 var_16544_to_fp16 = const()[name = string("op_16544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1657_cast_fp16, y = var_16544_to_fp16)[name = string("aw_chunk_1657_cast_fp16")]; + fp16 var_16546_to_fp16 = const()[name = string("op_16546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1659_cast_fp16, y = var_16546_to_fp16)[name = string("aw_chunk_1659_cast_fp16")]; + fp16 var_16548_to_fp16 = const()[name = string("op_16548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1661_cast_fp16, y = var_16548_to_fp16)[name = string("aw_chunk_1661_cast_fp16")]; + fp16 var_16550_to_fp16 = const()[name = string("op_16550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1663_cast_fp16, y = var_16550_to_fp16)[name = string("aw_chunk_1663_cast_fp16")]; + fp16 var_16552_to_fp16 = const()[name = string("op_16552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1665_cast_fp16, y = var_16552_to_fp16)[name = string("aw_chunk_1665_cast_fp16")]; + fp16 var_16554_to_fp16 = const()[name = string("op_16554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1667_cast_fp16, y = var_16554_to_fp16)[name = string("aw_chunk_1667_cast_fp16")]; + fp16 var_16556_to_fp16 = const()[name = string("op_16556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1669_cast_fp16, y = var_16556_to_fp16)[name = string("aw_chunk_1669_cast_fp16")]; + fp16 var_16558_to_fp16 = const()[name = string("op_16558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1671_cast_fp16, y = var_16558_to_fp16)[name = string("aw_chunk_1671_cast_fp16")]; + fp16 var_16560_to_fp16 = const()[name = string("op_16560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1673_cast_fp16, y = var_16560_to_fp16)[name = string("aw_chunk_1673_cast_fp16")]; + fp16 var_16562_to_fp16 = const()[name = string("op_16562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1675_cast_fp16, y = var_16562_to_fp16)[name = string("aw_chunk_1675_cast_fp16")]; + fp16 var_16564_to_fp16 = const()[name = string("op_16564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1677_cast_fp16, y = var_16564_to_fp16)[name = string("aw_chunk_1677_cast_fp16")]; + fp16 var_16566_to_fp16 = const()[name = string("op_16566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1679_cast_fp16, y = var_16566_to_fp16)[name = string("aw_chunk_1679_cast_fp16")]; + fp16 var_16568_to_fp16 = const()[name = string("op_16568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1681_cast_fp16, y = var_16568_to_fp16)[name = string("aw_chunk_1681_cast_fp16")]; + fp16 var_16570_to_fp16 = const()[name = string("op_16570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1683_cast_fp16, y = var_16570_to_fp16)[name = string("aw_chunk_1683_cast_fp16")]; + fp16 var_16572_to_fp16 = const()[name = string("op_16572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1685_cast_fp16, y = var_16572_to_fp16)[name = string("aw_chunk_1685_cast_fp16")]; + fp16 var_16574_to_fp16 = const()[name = string("op_16574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1687_cast_fp16, y = var_16574_to_fp16)[name = string("aw_chunk_1687_cast_fp16")]; + fp16 var_16576_to_fp16 = const()[name = string("op_16576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1689_cast_fp16, y = var_16576_to_fp16)[name = string("aw_chunk_1689_cast_fp16")]; + fp16 var_16578_to_fp16 = const()[name = string("op_16578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1691_cast_fp16, y = var_16578_to_fp16)[name = string("aw_chunk_1691_cast_fp16")]; + fp16 var_16580_to_fp16 = const()[name = string("op_16580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1693_cast_fp16, y = var_16580_to_fp16)[name = string("aw_chunk_1693_cast_fp16")]; + fp16 var_16582_to_fp16 = const()[name = string("op_16582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1695_cast_fp16, y = var_16582_to_fp16)[name = string("aw_chunk_1695_cast_fp16")]; + fp16 var_16584_to_fp16 = const()[name = string("op_16584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1697_cast_fp16, y = var_16584_to_fp16)[name = string("aw_chunk_1697_cast_fp16")]; + fp16 var_16586_to_fp16 = const()[name = string("op_16586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1699_cast_fp16, y = var_16586_to_fp16)[name = string("aw_chunk_1699_cast_fp16")]; + fp16 var_16588_to_fp16 = const()[name = string("op_16588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1701_cast_fp16, y = var_16588_to_fp16)[name = string("aw_chunk_1701_cast_fp16")]; + fp16 var_16590_to_fp16 = const()[name = string("op_16590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1703_cast_fp16, y = var_16590_to_fp16)[name = string("aw_chunk_1703_cast_fp16")]; + fp16 var_16592_to_fp16 = const()[name = string("op_16592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1705_cast_fp16, y = var_16592_to_fp16)[name = string("aw_chunk_1705_cast_fp16")]; + fp16 var_16594_to_fp16 = const()[name = string("op_16594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1707_cast_fp16, y = var_16594_to_fp16)[name = string("aw_chunk_1707_cast_fp16")]; + fp16 var_16596_to_fp16 = const()[name = string("op_16596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1709_cast_fp16, y = var_16596_to_fp16)[name = string("aw_chunk_1709_cast_fp16")]; + fp16 var_16598_to_fp16 = const()[name = string("op_16598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1711_cast_fp16, y = var_16598_to_fp16)[name = string("aw_chunk_1711_cast_fp16")]; + fp16 var_16600_to_fp16 = const()[name = string("op_16600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1713_cast_fp16, y = var_16600_to_fp16)[name = string("aw_chunk_1713_cast_fp16")]; + fp16 var_16602_to_fp16 = const()[name = string("op_16602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1715_cast_fp16, y = var_16602_to_fp16)[name = string("aw_chunk_1715_cast_fp16")]; + fp16 var_16604_to_fp16 = const()[name = string("op_16604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1717_cast_fp16, y = var_16604_to_fp16)[name = string("aw_chunk_1717_cast_fp16")]; + fp16 var_16606_to_fp16 = const()[name = string("op_16606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1719_cast_fp16, y = var_16606_to_fp16)[name = string("aw_chunk_1719_cast_fp16")]; + fp16 var_16608_to_fp16 = const()[name = string("op_16608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1721_cast_fp16, y = var_16608_to_fp16)[name = string("aw_chunk_1721_cast_fp16")]; + fp16 var_16610_to_fp16 = const()[name = string("op_16610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1723_cast_fp16, y = var_16610_to_fp16)[name = string("aw_chunk_1723_cast_fp16")]; + fp16 var_16612_to_fp16 = const()[name = string("op_16612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1725_cast_fp16, y = var_16612_to_fp16)[name = string("aw_chunk_1725_cast_fp16")]; + fp16 var_16614_to_fp16 = const()[name = string("op_16614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1727_cast_fp16, y = var_16614_to_fp16)[name = string("aw_chunk_1727_cast_fp16")]; + fp16 var_16616_to_fp16 = const()[name = string("op_16616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1729_cast_fp16, y = var_16616_to_fp16)[name = string("aw_chunk_1729_cast_fp16")]; + fp16 var_16618_to_fp16 = const()[name = string("op_16618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1731_cast_fp16, y = var_16618_to_fp16)[name = string("aw_chunk_1731_cast_fp16")]; + fp16 var_16620_to_fp16 = const()[name = string("op_16620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1733_cast_fp16, y = var_16620_to_fp16)[name = string("aw_chunk_1733_cast_fp16")]; + fp16 var_16622_to_fp16 = const()[name = string("op_16622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1735_cast_fp16, y = var_16622_to_fp16)[name = string("aw_chunk_1735_cast_fp16")]; + fp16 var_16624_to_fp16 = const()[name = string("op_16624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1737_cast_fp16, y = var_16624_to_fp16)[name = string("aw_chunk_1737_cast_fp16")]; + fp16 var_16626_to_fp16 = const()[name = string("op_16626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1739_cast_fp16, y = var_16626_to_fp16)[name = string("aw_chunk_1739_cast_fp16")]; + fp16 var_16628_to_fp16 = const()[name = string("op_16628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1741_cast_fp16, y = var_16628_to_fp16)[name = string("aw_chunk_1741_cast_fp16")]; + fp16 var_16630_to_fp16 = const()[name = string("op_16630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1743_cast_fp16, y = var_16630_to_fp16)[name = string("aw_chunk_1743_cast_fp16")]; + fp16 var_16632_to_fp16 = const()[name = string("op_16632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1745_cast_fp16, y = var_16632_to_fp16)[name = string("aw_chunk_1745_cast_fp16")]; + fp16 var_16634_to_fp16 = const()[name = string("op_16634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1747_cast_fp16, y = var_16634_to_fp16)[name = string("aw_chunk_1747_cast_fp16")]; + fp16 var_16636_to_fp16 = const()[name = string("op_16636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1749_cast_fp16, y = var_16636_to_fp16)[name = string("aw_chunk_1749_cast_fp16")]; + fp16 var_16638_to_fp16 = const()[name = string("op_16638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1751_cast_fp16, y = var_16638_to_fp16)[name = string("aw_chunk_1751_cast_fp16")]; + fp16 var_16640_to_fp16 = const()[name = string("op_16640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1753_cast_fp16, y = var_16640_to_fp16)[name = string("aw_chunk_1753_cast_fp16")]; + fp16 var_16642_to_fp16 = const()[name = string("op_16642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1755_cast_fp16, y = var_16642_to_fp16)[name = string("aw_chunk_1755_cast_fp16")]; + fp16 var_16644_to_fp16 = const()[name = string("op_16644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1757_cast_fp16, y = var_16644_to_fp16)[name = string("aw_chunk_1757_cast_fp16")]; + fp16 var_16646_to_fp16 = const()[name = string("op_16646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1759_cast_fp16, y = var_16646_to_fp16)[name = string("aw_chunk_1759_cast_fp16")]; + tensor var_16648_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1601_cast_fp16)[name = string("op_16648_cast_fp16")]; + tensor var_16649_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1603_cast_fp16)[name = string("op_16649_cast_fp16")]; + tensor var_16650_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1605_cast_fp16)[name = string("op_16650_cast_fp16")]; + tensor var_16651_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1607_cast_fp16)[name = string("op_16651_cast_fp16")]; + tensor var_16652_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1609_cast_fp16)[name = string("op_16652_cast_fp16")]; + tensor var_16653_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1611_cast_fp16)[name = string("op_16653_cast_fp16")]; + tensor var_16654_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1613_cast_fp16)[name = string("op_16654_cast_fp16")]; + tensor var_16655_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1615_cast_fp16)[name = string("op_16655_cast_fp16")]; + tensor var_16656_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1617_cast_fp16)[name = string("op_16656_cast_fp16")]; + tensor var_16657_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1619_cast_fp16)[name = string("op_16657_cast_fp16")]; + tensor var_16658_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1621_cast_fp16)[name = string("op_16658_cast_fp16")]; + tensor var_16659_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1623_cast_fp16)[name = string("op_16659_cast_fp16")]; + tensor var_16660_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1625_cast_fp16)[name = string("op_16660_cast_fp16")]; + tensor var_16661_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1627_cast_fp16)[name = string("op_16661_cast_fp16")]; + tensor var_16662_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1629_cast_fp16)[name = string("op_16662_cast_fp16")]; + tensor var_16663_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1631_cast_fp16)[name = string("op_16663_cast_fp16")]; + tensor var_16664_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1633_cast_fp16)[name = string("op_16664_cast_fp16")]; + tensor var_16665_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1635_cast_fp16)[name = string("op_16665_cast_fp16")]; + tensor var_16666_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1637_cast_fp16)[name = string("op_16666_cast_fp16")]; + tensor var_16667_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1639_cast_fp16)[name = string("op_16667_cast_fp16")]; + tensor var_16668_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1641_cast_fp16)[name = string("op_16668_cast_fp16")]; + tensor var_16669_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1643_cast_fp16)[name = string("op_16669_cast_fp16")]; + tensor var_16670_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1645_cast_fp16)[name = string("op_16670_cast_fp16")]; + tensor var_16671_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1647_cast_fp16)[name = string("op_16671_cast_fp16")]; + tensor var_16672_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1649_cast_fp16)[name = string("op_16672_cast_fp16")]; + tensor var_16673_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1651_cast_fp16)[name = string("op_16673_cast_fp16")]; + tensor var_16674_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1653_cast_fp16)[name = string("op_16674_cast_fp16")]; + tensor var_16675_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1655_cast_fp16)[name = string("op_16675_cast_fp16")]; + tensor var_16676_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1657_cast_fp16)[name = string("op_16676_cast_fp16")]; + tensor var_16677_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1659_cast_fp16)[name = string("op_16677_cast_fp16")]; + tensor var_16678_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1661_cast_fp16)[name = string("op_16678_cast_fp16")]; + tensor var_16679_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1663_cast_fp16)[name = string("op_16679_cast_fp16")]; + tensor var_16680_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1665_cast_fp16)[name = string("op_16680_cast_fp16")]; + tensor var_16681_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1667_cast_fp16)[name = string("op_16681_cast_fp16")]; + tensor var_16682_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1669_cast_fp16)[name = string("op_16682_cast_fp16")]; + tensor var_16683_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1671_cast_fp16)[name = string("op_16683_cast_fp16")]; + tensor var_16684_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1673_cast_fp16)[name = string("op_16684_cast_fp16")]; + tensor var_16685_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1675_cast_fp16)[name = string("op_16685_cast_fp16")]; + tensor var_16686_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1677_cast_fp16)[name = string("op_16686_cast_fp16")]; + tensor var_16687_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1679_cast_fp16)[name = string("op_16687_cast_fp16")]; + tensor var_16688_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1681_cast_fp16)[name = string("op_16688_cast_fp16")]; + tensor var_16689_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1683_cast_fp16)[name = string("op_16689_cast_fp16")]; + tensor var_16690_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1685_cast_fp16)[name = string("op_16690_cast_fp16")]; + tensor var_16691_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1687_cast_fp16)[name = string("op_16691_cast_fp16")]; + tensor var_16692_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1689_cast_fp16)[name = string("op_16692_cast_fp16")]; + tensor var_16693_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1691_cast_fp16)[name = string("op_16693_cast_fp16")]; + tensor var_16694_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1693_cast_fp16)[name = string("op_16694_cast_fp16")]; + tensor var_16695_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1695_cast_fp16)[name = string("op_16695_cast_fp16")]; + tensor var_16696_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1697_cast_fp16)[name = string("op_16696_cast_fp16")]; + tensor var_16697_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1699_cast_fp16)[name = string("op_16697_cast_fp16")]; + tensor var_16698_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1701_cast_fp16)[name = string("op_16698_cast_fp16")]; + tensor var_16699_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1703_cast_fp16)[name = string("op_16699_cast_fp16")]; + tensor var_16700_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1705_cast_fp16)[name = string("op_16700_cast_fp16")]; + tensor var_16701_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1707_cast_fp16)[name = string("op_16701_cast_fp16")]; + tensor var_16702_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1709_cast_fp16)[name = string("op_16702_cast_fp16")]; + tensor var_16703_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1711_cast_fp16)[name = string("op_16703_cast_fp16")]; + tensor var_16704_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1713_cast_fp16)[name = string("op_16704_cast_fp16")]; + tensor var_16705_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1715_cast_fp16)[name = string("op_16705_cast_fp16")]; + tensor var_16706_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1717_cast_fp16)[name = string("op_16706_cast_fp16")]; + tensor var_16707_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1719_cast_fp16)[name = string("op_16707_cast_fp16")]; + tensor var_16708_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1721_cast_fp16)[name = string("op_16708_cast_fp16")]; + tensor var_16709_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1723_cast_fp16)[name = string("op_16709_cast_fp16")]; + tensor var_16710_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1725_cast_fp16)[name = string("op_16710_cast_fp16")]; + tensor var_16711_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1727_cast_fp16)[name = string("op_16711_cast_fp16")]; + tensor var_16712_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1729_cast_fp16)[name = string("op_16712_cast_fp16")]; + tensor var_16713_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1731_cast_fp16)[name = string("op_16713_cast_fp16")]; + tensor var_16714_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1733_cast_fp16)[name = string("op_16714_cast_fp16")]; + tensor var_16715_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1735_cast_fp16)[name = string("op_16715_cast_fp16")]; + tensor var_16716_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1737_cast_fp16)[name = string("op_16716_cast_fp16")]; + tensor var_16717_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1739_cast_fp16)[name = string("op_16717_cast_fp16")]; + tensor var_16718_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1741_cast_fp16)[name = string("op_16718_cast_fp16")]; + tensor var_16719_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1743_cast_fp16)[name = string("op_16719_cast_fp16")]; + tensor var_16720_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1745_cast_fp16)[name = string("op_16720_cast_fp16")]; + tensor var_16721_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1747_cast_fp16)[name = string("op_16721_cast_fp16")]; + tensor var_16722_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1749_cast_fp16)[name = string("op_16722_cast_fp16")]; + tensor var_16723_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1751_cast_fp16)[name = string("op_16723_cast_fp16")]; + tensor var_16724_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1753_cast_fp16)[name = string("op_16724_cast_fp16")]; + tensor var_16725_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1755_cast_fp16)[name = string("op_16725_cast_fp16")]; + tensor var_16726_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1757_cast_fp16)[name = string("op_16726_cast_fp16")]; + tensor var_16727_cast_fp16 = softmax(axis = var_15473, x = aw_chunk_1759_cast_fp16)[name = string("op_16727_cast_fp16")]; + string var_16729_equation_0 = const()[name = string("op_16729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16729_cast_fp16 = einsum(equation = var_16729_equation_0, values = (var_16249_cast_fp16, var_16648_cast_fp16))[name = string("op_16729_cast_fp16")]; + string var_16731_equation_0 = const()[name = string("op_16731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16731_cast_fp16 = einsum(equation = var_16731_equation_0, values = (var_16249_cast_fp16, var_16649_cast_fp16))[name = string("op_16731_cast_fp16")]; + string var_16733_equation_0 = const()[name = string("op_16733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16733_cast_fp16 = einsum(equation = var_16733_equation_0, values = (var_16249_cast_fp16, var_16650_cast_fp16))[name = string("op_16733_cast_fp16")]; + string var_16735_equation_0 = const()[name = string("op_16735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16735_cast_fp16 = einsum(equation = var_16735_equation_0, values = (var_16249_cast_fp16, var_16651_cast_fp16))[name = string("op_16735_cast_fp16")]; + string var_16737_equation_0 = const()[name = string("op_16737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16737_cast_fp16 = einsum(equation = var_16737_equation_0, values = (var_16253_cast_fp16, var_16652_cast_fp16))[name = string("op_16737_cast_fp16")]; + string var_16739_equation_0 = const()[name = string("op_16739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16739_cast_fp16 = einsum(equation = var_16739_equation_0, values = (var_16253_cast_fp16, var_16653_cast_fp16))[name = string("op_16739_cast_fp16")]; + string var_16741_equation_0 = const()[name = string("op_16741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16741_cast_fp16 = einsum(equation = var_16741_equation_0, values = (var_16253_cast_fp16, var_16654_cast_fp16))[name = string("op_16741_cast_fp16")]; + string var_16743_equation_0 = const()[name = string("op_16743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16743_cast_fp16 = einsum(equation = var_16743_equation_0, values = (var_16253_cast_fp16, var_16655_cast_fp16))[name = string("op_16743_cast_fp16")]; + string var_16745_equation_0 = const()[name = string("op_16745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16745_cast_fp16 = einsum(equation = var_16745_equation_0, values = (var_16257_cast_fp16, var_16656_cast_fp16))[name = string("op_16745_cast_fp16")]; + string var_16747_equation_0 = const()[name = string("op_16747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16747_cast_fp16 = einsum(equation = var_16747_equation_0, values = (var_16257_cast_fp16, var_16657_cast_fp16))[name = string("op_16747_cast_fp16")]; + string var_16749_equation_0 = const()[name = string("op_16749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16749_cast_fp16 = einsum(equation = var_16749_equation_0, values = (var_16257_cast_fp16, var_16658_cast_fp16))[name = string("op_16749_cast_fp16")]; + string var_16751_equation_0 = const()[name = string("op_16751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16751_cast_fp16 = einsum(equation = var_16751_equation_0, values = (var_16257_cast_fp16, var_16659_cast_fp16))[name = string("op_16751_cast_fp16")]; + string var_16753_equation_0 = const()[name = string("op_16753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16753_cast_fp16 = einsum(equation = var_16753_equation_0, values = (var_16261_cast_fp16, var_16660_cast_fp16))[name = string("op_16753_cast_fp16")]; + string var_16755_equation_0 = const()[name = string("op_16755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16755_cast_fp16 = einsum(equation = var_16755_equation_0, values = (var_16261_cast_fp16, var_16661_cast_fp16))[name = string("op_16755_cast_fp16")]; + string var_16757_equation_0 = const()[name = string("op_16757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16757_cast_fp16 = einsum(equation = var_16757_equation_0, values = (var_16261_cast_fp16, var_16662_cast_fp16))[name = string("op_16757_cast_fp16")]; + string var_16759_equation_0 = const()[name = string("op_16759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16759_cast_fp16 = einsum(equation = var_16759_equation_0, values = (var_16261_cast_fp16, var_16663_cast_fp16))[name = string("op_16759_cast_fp16")]; + string var_16761_equation_0 = const()[name = string("op_16761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16761_cast_fp16 = einsum(equation = var_16761_equation_0, values = (var_16265_cast_fp16, var_16664_cast_fp16))[name = string("op_16761_cast_fp16")]; + string var_16763_equation_0 = const()[name = string("op_16763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16763_cast_fp16 = einsum(equation = var_16763_equation_0, values = (var_16265_cast_fp16, var_16665_cast_fp16))[name = string("op_16763_cast_fp16")]; + string var_16765_equation_0 = const()[name = string("op_16765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16765_cast_fp16 = einsum(equation = var_16765_equation_0, values = (var_16265_cast_fp16, var_16666_cast_fp16))[name = string("op_16765_cast_fp16")]; + string var_16767_equation_0 = const()[name = string("op_16767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16767_cast_fp16 = einsum(equation = var_16767_equation_0, values = (var_16265_cast_fp16, var_16667_cast_fp16))[name = string("op_16767_cast_fp16")]; + string var_16769_equation_0 = const()[name = string("op_16769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16769_cast_fp16 = einsum(equation = var_16769_equation_0, values = (var_16269_cast_fp16, var_16668_cast_fp16))[name = string("op_16769_cast_fp16")]; + string var_16771_equation_0 = const()[name = string("op_16771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16771_cast_fp16 = einsum(equation = var_16771_equation_0, values = (var_16269_cast_fp16, var_16669_cast_fp16))[name = string("op_16771_cast_fp16")]; + string var_16773_equation_0 = const()[name = string("op_16773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16773_cast_fp16 = einsum(equation = var_16773_equation_0, values = (var_16269_cast_fp16, var_16670_cast_fp16))[name = string("op_16773_cast_fp16")]; + string var_16775_equation_0 = const()[name = string("op_16775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16775_cast_fp16 = einsum(equation = var_16775_equation_0, values = (var_16269_cast_fp16, var_16671_cast_fp16))[name = string("op_16775_cast_fp16")]; + string var_16777_equation_0 = const()[name = string("op_16777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16777_cast_fp16 = einsum(equation = var_16777_equation_0, values = (var_16273_cast_fp16, var_16672_cast_fp16))[name = string("op_16777_cast_fp16")]; + string var_16779_equation_0 = const()[name = string("op_16779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16779_cast_fp16 = einsum(equation = var_16779_equation_0, values = (var_16273_cast_fp16, var_16673_cast_fp16))[name = string("op_16779_cast_fp16")]; + string var_16781_equation_0 = const()[name = string("op_16781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16781_cast_fp16 = einsum(equation = var_16781_equation_0, values = (var_16273_cast_fp16, var_16674_cast_fp16))[name = string("op_16781_cast_fp16")]; + string var_16783_equation_0 = const()[name = string("op_16783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16783_cast_fp16 = einsum(equation = var_16783_equation_0, values = (var_16273_cast_fp16, var_16675_cast_fp16))[name = string("op_16783_cast_fp16")]; + string var_16785_equation_0 = const()[name = string("op_16785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16785_cast_fp16 = einsum(equation = var_16785_equation_0, values = (var_16277_cast_fp16, var_16676_cast_fp16))[name = string("op_16785_cast_fp16")]; + string var_16787_equation_0 = const()[name = string("op_16787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16787_cast_fp16 = einsum(equation = var_16787_equation_0, values = (var_16277_cast_fp16, var_16677_cast_fp16))[name = string("op_16787_cast_fp16")]; + string var_16789_equation_0 = const()[name = string("op_16789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16789_cast_fp16 = einsum(equation = var_16789_equation_0, values = (var_16277_cast_fp16, var_16678_cast_fp16))[name = string("op_16789_cast_fp16")]; + string var_16791_equation_0 = const()[name = string("op_16791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16791_cast_fp16 = einsum(equation = var_16791_equation_0, values = (var_16277_cast_fp16, var_16679_cast_fp16))[name = string("op_16791_cast_fp16")]; + string var_16793_equation_0 = const()[name = string("op_16793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16793_cast_fp16 = einsum(equation = var_16793_equation_0, values = (var_16281_cast_fp16, var_16680_cast_fp16))[name = string("op_16793_cast_fp16")]; + string var_16795_equation_0 = const()[name = string("op_16795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16795_cast_fp16 = einsum(equation = var_16795_equation_0, values = (var_16281_cast_fp16, var_16681_cast_fp16))[name = string("op_16795_cast_fp16")]; + string var_16797_equation_0 = const()[name = string("op_16797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16797_cast_fp16 = einsum(equation = var_16797_equation_0, values = (var_16281_cast_fp16, var_16682_cast_fp16))[name = string("op_16797_cast_fp16")]; + string var_16799_equation_0 = const()[name = string("op_16799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16799_cast_fp16 = einsum(equation = var_16799_equation_0, values = (var_16281_cast_fp16, var_16683_cast_fp16))[name = string("op_16799_cast_fp16")]; + string var_16801_equation_0 = const()[name = string("op_16801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16801_cast_fp16 = einsum(equation = var_16801_equation_0, values = (var_16285_cast_fp16, var_16684_cast_fp16))[name = string("op_16801_cast_fp16")]; + string var_16803_equation_0 = const()[name = string("op_16803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16803_cast_fp16 = einsum(equation = var_16803_equation_0, values = (var_16285_cast_fp16, var_16685_cast_fp16))[name = string("op_16803_cast_fp16")]; + string var_16805_equation_0 = const()[name = string("op_16805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16805_cast_fp16 = einsum(equation = var_16805_equation_0, values = (var_16285_cast_fp16, var_16686_cast_fp16))[name = string("op_16805_cast_fp16")]; + string var_16807_equation_0 = const()[name = string("op_16807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16807_cast_fp16 = einsum(equation = var_16807_equation_0, values = (var_16285_cast_fp16, var_16687_cast_fp16))[name = string("op_16807_cast_fp16")]; + string var_16809_equation_0 = const()[name = string("op_16809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16809_cast_fp16 = einsum(equation = var_16809_equation_0, values = (var_16289_cast_fp16, var_16688_cast_fp16))[name = string("op_16809_cast_fp16")]; + string var_16811_equation_0 = const()[name = string("op_16811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16811_cast_fp16 = einsum(equation = var_16811_equation_0, values = (var_16289_cast_fp16, var_16689_cast_fp16))[name = string("op_16811_cast_fp16")]; + string var_16813_equation_0 = const()[name = string("op_16813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16813_cast_fp16 = einsum(equation = var_16813_equation_0, values = (var_16289_cast_fp16, var_16690_cast_fp16))[name = string("op_16813_cast_fp16")]; + string var_16815_equation_0 = const()[name = string("op_16815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16815_cast_fp16 = einsum(equation = var_16815_equation_0, values = (var_16289_cast_fp16, var_16691_cast_fp16))[name = string("op_16815_cast_fp16")]; + string var_16817_equation_0 = const()[name = string("op_16817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16817_cast_fp16 = einsum(equation = var_16817_equation_0, values = (var_16293_cast_fp16, var_16692_cast_fp16))[name = string("op_16817_cast_fp16")]; + string var_16819_equation_0 = const()[name = string("op_16819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16819_cast_fp16 = einsum(equation = var_16819_equation_0, values = (var_16293_cast_fp16, var_16693_cast_fp16))[name = string("op_16819_cast_fp16")]; + string var_16821_equation_0 = const()[name = string("op_16821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16821_cast_fp16 = einsum(equation = var_16821_equation_0, values = (var_16293_cast_fp16, var_16694_cast_fp16))[name = string("op_16821_cast_fp16")]; + string var_16823_equation_0 = const()[name = string("op_16823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16823_cast_fp16 = einsum(equation = var_16823_equation_0, values = (var_16293_cast_fp16, var_16695_cast_fp16))[name = string("op_16823_cast_fp16")]; + string var_16825_equation_0 = const()[name = string("op_16825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16825_cast_fp16 = einsum(equation = var_16825_equation_0, values = (var_16297_cast_fp16, var_16696_cast_fp16))[name = string("op_16825_cast_fp16")]; + string var_16827_equation_0 = const()[name = string("op_16827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16827_cast_fp16 = einsum(equation = var_16827_equation_0, values = (var_16297_cast_fp16, var_16697_cast_fp16))[name = string("op_16827_cast_fp16")]; + string var_16829_equation_0 = const()[name = string("op_16829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16829_cast_fp16 = einsum(equation = var_16829_equation_0, values = (var_16297_cast_fp16, var_16698_cast_fp16))[name = string("op_16829_cast_fp16")]; + string var_16831_equation_0 = const()[name = string("op_16831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16831_cast_fp16 = einsum(equation = var_16831_equation_0, values = (var_16297_cast_fp16, var_16699_cast_fp16))[name = string("op_16831_cast_fp16")]; + string var_16833_equation_0 = const()[name = string("op_16833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16833_cast_fp16 = einsum(equation = var_16833_equation_0, values = (var_16301_cast_fp16, var_16700_cast_fp16))[name = string("op_16833_cast_fp16")]; + string var_16835_equation_0 = const()[name = string("op_16835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16835_cast_fp16 = einsum(equation = var_16835_equation_0, values = (var_16301_cast_fp16, var_16701_cast_fp16))[name = string("op_16835_cast_fp16")]; + string var_16837_equation_0 = const()[name = string("op_16837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16837_cast_fp16 = einsum(equation = var_16837_equation_0, values = (var_16301_cast_fp16, var_16702_cast_fp16))[name = string("op_16837_cast_fp16")]; + string var_16839_equation_0 = const()[name = string("op_16839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16839_cast_fp16 = einsum(equation = var_16839_equation_0, values = (var_16301_cast_fp16, var_16703_cast_fp16))[name = string("op_16839_cast_fp16")]; + string var_16841_equation_0 = const()[name = string("op_16841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16841_cast_fp16 = einsum(equation = var_16841_equation_0, values = (var_16305_cast_fp16, var_16704_cast_fp16))[name = string("op_16841_cast_fp16")]; + string var_16843_equation_0 = const()[name = string("op_16843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16843_cast_fp16 = einsum(equation = var_16843_equation_0, values = (var_16305_cast_fp16, var_16705_cast_fp16))[name = string("op_16843_cast_fp16")]; + string var_16845_equation_0 = const()[name = string("op_16845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16845_cast_fp16 = einsum(equation = var_16845_equation_0, values = (var_16305_cast_fp16, var_16706_cast_fp16))[name = string("op_16845_cast_fp16")]; + string var_16847_equation_0 = const()[name = string("op_16847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16847_cast_fp16 = einsum(equation = var_16847_equation_0, values = (var_16305_cast_fp16, var_16707_cast_fp16))[name = string("op_16847_cast_fp16")]; + string var_16849_equation_0 = const()[name = string("op_16849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16849_cast_fp16 = einsum(equation = var_16849_equation_0, values = (var_16309_cast_fp16, var_16708_cast_fp16))[name = string("op_16849_cast_fp16")]; + string var_16851_equation_0 = const()[name = string("op_16851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16851_cast_fp16 = einsum(equation = var_16851_equation_0, values = (var_16309_cast_fp16, var_16709_cast_fp16))[name = string("op_16851_cast_fp16")]; + string var_16853_equation_0 = const()[name = string("op_16853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16853_cast_fp16 = einsum(equation = var_16853_equation_0, values = (var_16309_cast_fp16, var_16710_cast_fp16))[name = string("op_16853_cast_fp16")]; + string var_16855_equation_0 = const()[name = string("op_16855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16855_cast_fp16 = einsum(equation = var_16855_equation_0, values = (var_16309_cast_fp16, var_16711_cast_fp16))[name = string("op_16855_cast_fp16")]; + string var_16857_equation_0 = const()[name = string("op_16857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16857_cast_fp16 = einsum(equation = var_16857_equation_0, values = (var_16313_cast_fp16, var_16712_cast_fp16))[name = string("op_16857_cast_fp16")]; + string var_16859_equation_0 = const()[name = string("op_16859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16859_cast_fp16 = einsum(equation = var_16859_equation_0, values = (var_16313_cast_fp16, var_16713_cast_fp16))[name = string("op_16859_cast_fp16")]; + string var_16861_equation_0 = const()[name = string("op_16861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16861_cast_fp16 = einsum(equation = var_16861_equation_0, values = (var_16313_cast_fp16, var_16714_cast_fp16))[name = string("op_16861_cast_fp16")]; + string var_16863_equation_0 = const()[name = string("op_16863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16863_cast_fp16 = einsum(equation = var_16863_equation_0, values = (var_16313_cast_fp16, var_16715_cast_fp16))[name = string("op_16863_cast_fp16")]; + string var_16865_equation_0 = const()[name = string("op_16865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16865_cast_fp16 = einsum(equation = var_16865_equation_0, values = (var_16317_cast_fp16, var_16716_cast_fp16))[name = string("op_16865_cast_fp16")]; + string var_16867_equation_0 = const()[name = string("op_16867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16867_cast_fp16 = einsum(equation = var_16867_equation_0, values = (var_16317_cast_fp16, var_16717_cast_fp16))[name = string("op_16867_cast_fp16")]; + string var_16869_equation_0 = const()[name = string("op_16869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16869_cast_fp16 = einsum(equation = var_16869_equation_0, values = (var_16317_cast_fp16, var_16718_cast_fp16))[name = string("op_16869_cast_fp16")]; + string var_16871_equation_0 = const()[name = string("op_16871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16871_cast_fp16 = einsum(equation = var_16871_equation_0, values = (var_16317_cast_fp16, var_16719_cast_fp16))[name = string("op_16871_cast_fp16")]; + string var_16873_equation_0 = const()[name = string("op_16873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16873_cast_fp16 = einsum(equation = var_16873_equation_0, values = (var_16321_cast_fp16, var_16720_cast_fp16))[name = string("op_16873_cast_fp16")]; + string var_16875_equation_0 = const()[name = string("op_16875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16875_cast_fp16 = einsum(equation = var_16875_equation_0, values = (var_16321_cast_fp16, var_16721_cast_fp16))[name = string("op_16875_cast_fp16")]; + string var_16877_equation_0 = const()[name = string("op_16877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16877_cast_fp16 = einsum(equation = var_16877_equation_0, values = (var_16321_cast_fp16, var_16722_cast_fp16))[name = string("op_16877_cast_fp16")]; + string var_16879_equation_0 = const()[name = string("op_16879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16879_cast_fp16 = einsum(equation = var_16879_equation_0, values = (var_16321_cast_fp16, var_16723_cast_fp16))[name = string("op_16879_cast_fp16")]; + string var_16881_equation_0 = const()[name = string("op_16881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16881_cast_fp16 = einsum(equation = var_16881_equation_0, values = (var_16325_cast_fp16, var_16724_cast_fp16))[name = string("op_16881_cast_fp16")]; + string var_16883_equation_0 = const()[name = string("op_16883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16883_cast_fp16 = einsum(equation = var_16883_equation_0, values = (var_16325_cast_fp16, var_16725_cast_fp16))[name = string("op_16883_cast_fp16")]; + string var_16885_equation_0 = const()[name = string("op_16885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16885_cast_fp16 = einsum(equation = var_16885_equation_0, values = (var_16325_cast_fp16, var_16726_cast_fp16))[name = string("op_16885_cast_fp16")]; + string var_16887_equation_0 = const()[name = string("op_16887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16887_cast_fp16 = einsum(equation = var_16887_equation_0, values = (var_16325_cast_fp16, var_16727_cast_fp16))[name = string("op_16887_cast_fp16")]; + bool var_16889_interleave_0 = const()[name = string("op_16889_interleave_0"), val = bool(false)]; + tensor var_16889_cast_fp16 = concat(axis = var_15448, interleave = var_16889_interleave_0, values = (var_16729_cast_fp16, var_16731_cast_fp16, var_16733_cast_fp16, var_16735_cast_fp16))[name = string("op_16889_cast_fp16")]; + bool var_16891_interleave_0 = const()[name = string("op_16891_interleave_0"), val = bool(false)]; + tensor var_16891_cast_fp16 = concat(axis = var_15448, interleave = var_16891_interleave_0, values = (var_16737_cast_fp16, var_16739_cast_fp16, var_16741_cast_fp16, var_16743_cast_fp16))[name = string("op_16891_cast_fp16")]; + bool var_16893_interleave_0 = const()[name = string("op_16893_interleave_0"), val = bool(false)]; + tensor var_16893_cast_fp16 = concat(axis = var_15448, interleave = var_16893_interleave_0, values = (var_16745_cast_fp16, var_16747_cast_fp16, var_16749_cast_fp16, var_16751_cast_fp16))[name = string("op_16893_cast_fp16")]; + bool var_16895_interleave_0 = const()[name = string("op_16895_interleave_0"), val = bool(false)]; + tensor var_16895_cast_fp16 = concat(axis = var_15448, interleave = var_16895_interleave_0, values = (var_16753_cast_fp16, var_16755_cast_fp16, var_16757_cast_fp16, var_16759_cast_fp16))[name = string("op_16895_cast_fp16")]; + bool var_16897_interleave_0 = const()[name = string("op_16897_interleave_0"), val = bool(false)]; + tensor var_16897_cast_fp16 = concat(axis = var_15448, interleave = var_16897_interleave_0, values = (var_16761_cast_fp16, var_16763_cast_fp16, var_16765_cast_fp16, var_16767_cast_fp16))[name = string("op_16897_cast_fp16")]; + bool var_16899_interleave_0 = const()[name = string("op_16899_interleave_0"), val = bool(false)]; + tensor var_16899_cast_fp16 = concat(axis = var_15448, interleave = var_16899_interleave_0, values = (var_16769_cast_fp16, var_16771_cast_fp16, var_16773_cast_fp16, var_16775_cast_fp16))[name = string("op_16899_cast_fp16")]; + bool var_16901_interleave_0 = const()[name = string("op_16901_interleave_0"), val = bool(false)]; + tensor var_16901_cast_fp16 = concat(axis = var_15448, interleave = var_16901_interleave_0, values = (var_16777_cast_fp16, var_16779_cast_fp16, var_16781_cast_fp16, var_16783_cast_fp16))[name = string("op_16901_cast_fp16")]; + bool var_16903_interleave_0 = const()[name = string("op_16903_interleave_0"), val = bool(false)]; + tensor var_16903_cast_fp16 = concat(axis = var_15448, interleave = var_16903_interleave_0, values = (var_16785_cast_fp16, var_16787_cast_fp16, var_16789_cast_fp16, var_16791_cast_fp16))[name = string("op_16903_cast_fp16")]; + bool var_16905_interleave_0 = const()[name = string("op_16905_interleave_0"), val = bool(false)]; + tensor var_16905_cast_fp16 = concat(axis = var_15448, interleave = var_16905_interleave_0, values = (var_16793_cast_fp16, var_16795_cast_fp16, var_16797_cast_fp16, var_16799_cast_fp16))[name = string("op_16905_cast_fp16")]; + bool var_16907_interleave_0 = const()[name = string("op_16907_interleave_0"), val = bool(false)]; + tensor var_16907_cast_fp16 = concat(axis = var_15448, interleave = var_16907_interleave_0, values = (var_16801_cast_fp16, var_16803_cast_fp16, var_16805_cast_fp16, var_16807_cast_fp16))[name = string("op_16907_cast_fp16")]; + bool var_16909_interleave_0 = const()[name = string("op_16909_interleave_0"), val = bool(false)]; + tensor var_16909_cast_fp16 = concat(axis = var_15448, interleave = var_16909_interleave_0, values = (var_16809_cast_fp16, var_16811_cast_fp16, var_16813_cast_fp16, var_16815_cast_fp16))[name = string("op_16909_cast_fp16")]; + bool var_16911_interleave_0 = const()[name = string("op_16911_interleave_0"), val = bool(false)]; + tensor var_16911_cast_fp16 = concat(axis = var_15448, interleave = var_16911_interleave_0, values = (var_16817_cast_fp16, var_16819_cast_fp16, var_16821_cast_fp16, var_16823_cast_fp16))[name = string("op_16911_cast_fp16")]; + bool var_16913_interleave_0 = const()[name = string("op_16913_interleave_0"), val = bool(false)]; + tensor var_16913_cast_fp16 = concat(axis = var_15448, interleave = var_16913_interleave_0, values = (var_16825_cast_fp16, var_16827_cast_fp16, var_16829_cast_fp16, var_16831_cast_fp16))[name = string("op_16913_cast_fp16")]; + bool var_16915_interleave_0 = const()[name = string("op_16915_interleave_0"), val = bool(false)]; + tensor var_16915_cast_fp16 = concat(axis = var_15448, interleave = var_16915_interleave_0, values = (var_16833_cast_fp16, var_16835_cast_fp16, var_16837_cast_fp16, var_16839_cast_fp16))[name = string("op_16915_cast_fp16")]; + bool var_16917_interleave_0 = const()[name = string("op_16917_interleave_0"), val = bool(false)]; + tensor var_16917_cast_fp16 = concat(axis = var_15448, interleave = var_16917_interleave_0, values = (var_16841_cast_fp16, var_16843_cast_fp16, var_16845_cast_fp16, var_16847_cast_fp16))[name = string("op_16917_cast_fp16")]; + bool var_16919_interleave_0 = const()[name = string("op_16919_interleave_0"), val = bool(false)]; + tensor var_16919_cast_fp16 = concat(axis = var_15448, interleave = var_16919_interleave_0, values = (var_16849_cast_fp16, var_16851_cast_fp16, var_16853_cast_fp16, var_16855_cast_fp16))[name = string("op_16919_cast_fp16")]; + bool var_16921_interleave_0 = const()[name = string("op_16921_interleave_0"), val = bool(false)]; + tensor var_16921_cast_fp16 = concat(axis = var_15448, interleave = var_16921_interleave_0, values = (var_16857_cast_fp16, var_16859_cast_fp16, var_16861_cast_fp16, var_16863_cast_fp16))[name = string("op_16921_cast_fp16")]; + bool var_16923_interleave_0 = const()[name = string("op_16923_interleave_0"), val = bool(false)]; + tensor var_16923_cast_fp16 = concat(axis = var_15448, interleave = var_16923_interleave_0, values = (var_16865_cast_fp16, var_16867_cast_fp16, var_16869_cast_fp16, var_16871_cast_fp16))[name = string("op_16923_cast_fp16")]; + bool var_16925_interleave_0 = const()[name = string("op_16925_interleave_0"), val = bool(false)]; + tensor var_16925_cast_fp16 = concat(axis = var_15448, interleave = var_16925_interleave_0, values = (var_16873_cast_fp16, var_16875_cast_fp16, var_16877_cast_fp16, var_16879_cast_fp16))[name = string("op_16925_cast_fp16")]; + bool var_16927_interleave_0 = const()[name = string("op_16927_interleave_0"), val = bool(false)]; + tensor var_16927_cast_fp16 = concat(axis = var_15448, interleave = var_16927_interleave_0, values = (var_16881_cast_fp16, var_16883_cast_fp16, var_16885_cast_fp16, var_16887_cast_fp16))[name = string("op_16927_cast_fp16")]; + bool input_81_interleave_0 = const()[name = string("input_81_interleave_0"), val = bool(false)]; + tensor input_81_cast_fp16 = concat(axis = var_15473, interleave = input_81_interleave_0, values = (var_16889_cast_fp16, var_16891_cast_fp16, var_16893_cast_fp16, var_16895_cast_fp16, var_16897_cast_fp16, var_16899_cast_fp16, var_16901_cast_fp16, var_16903_cast_fp16, var_16905_cast_fp16, var_16907_cast_fp16, var_16909_cast_fp16, var_16911_cast_fp16, var_16913_cast_fp16, var_16915_cast_fp16, var_16917_cast_fp16, var_16919_cast_fp16, var_16921_cast_fp16, var_16923_cast_fp16, var_16925_cast_fp16, var_16927_cast_fp16))[name = string("input_81_cast_fp16")]; + string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; + tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; + tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; + int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418038080)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421314944)))]; + tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_16946_to_fp16 = const()[name = string("op_16946_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_16946_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421317568)))]; + tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421320192)))]; + fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; + string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; + tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; + tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; + int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421322816)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434430080)))]; + tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; + tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434440384)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447547648)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; + int32 var_16975 = const()[name = string("op_16975"), val = int32(3)]; + int32 var_17000 = const()[name = string("op_17000"), val = int32(1)]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_17017_to_fp16 = const()[name = string("op_17017_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_17017_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447550272)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447552896)))]; + fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; + string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; + tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; + tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; + int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447555520)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450832384)))]; + tensor query_23_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_23_cast_fp16")]; + string key_23_pad_type_0 = const()[name = string("key_23_pad_type_0"), val = string("valid")]; + tensor key_23_strides_0 = const()[name = string("key_23_strides_0"), val = tensor([1, 1])]; + tensor key_23_pad_0 = const()[name = string("key_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_23_dilations_0 = const()[name = string("key_23_dilations_0"), val = tensor([1, 1])]; + int32 key_23_groups_0 = const()[name = string("key_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450835008)))]; + tensor key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_23_cast_fp16")]; + string value_23_pad_type_0 = const()[name = string("value_23_pad_type_0"), val = string("valid")]; + tensor value_23_strides_0 = const()[name = string("value_23_strides_0"), val = tensor([1, 1])]; + tensor value_23_pad_0 = const()[name = string("value_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_23_dilations_0 = const()[name = string("value_23_dilations_0"), val = tensor([1, 1])]; + int32 value_23_groups_0 = const()[name = string("value_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454111872)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457388736)))]; + tensor value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_23_cast_fp16")]; + tensor var_17055_begin_0 = const()[name = string("op_17055_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17055_end_0 = const()[name = string("op_17055_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17055_end_mask_0 = const()[name = string("op_17055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17055_cast_fp16 = slice_by_index(begin = var_17055_begin_0, end = var_17055_end_0, end_mask = var_17055_end_mask_0, x = query_23_cast_fp16)[name = string("op_17055_cast_fp16")]; + tensor var_17059_begin_0 = const()[name = string("op_17059_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17059_end_0 = const()[name = string("op_17059_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_17059_end_mask_0 = const()[name = string("op_17059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17059_cast_fp16 = slice_by_index(begin = var_17059_begin_0, end = var_17059_end_0, end_mask = var_17059_end_mask_0, x = query_23_cast_fp16)[name = string("op_17059_cast_fp16")]; + tensor var_17063_begin_0 = const()[name = string("op_17063_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17063_end_0 = const()[name = string("op_17063_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_17063_end_mask_0 = const()[name = string("op_17063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17063_cast_fp16 = slice_by_index(begin = var_17063_begin_0, end = var_17063_end_0, end_mask = var_17063_end_mask_0, x = query_23_cast_fp16)[name = string("op_17063_cast_fp16")]; + tensor var_17067_begin_0 = const()[name = string("op_17067_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17067_end_0 = const()[name = string("op_17067_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_17067_end_mask_0 = const()[name = string("op_17067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17067_cast_fp16 = slice_by_index(begin = var_17067_begin_0, end = var_17067_end_0, end_mask = var_17067_end_mask_0, x = query_23_cast_fp16)[name = string("op_17067_cast_fp16")]; + tensor var_17071_begin_0 = const()[name = string("op_17071_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17071_end_0 = const()[name = string("op_17071_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_17071_end_mask_0 = const()[name = string("op_17071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17071_cast_fp16 = slice_by_index(begin = var_17071_begin_0, end = var_17071_end_0, end_mask = var_17071_end_mask_0, x = query_23_cast_fp16)[name = string("op_17071_cast_fp16")]; + tensor var_17075_begin_0 = const()[name = string("op_17075_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17075_end_0 = const()[name = string("op_17075_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_17075_end_mask_0 = const()[name = string("op_17075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17075_cast_fp16 = slice_by_index(begin = var_17075_begin_0, end = var_17075_end_0, end_mask = var_17075_end_mask_0, x = query_23_cast_fp16)[name = string("op_17075_cast_fp16")]; + tensor var_17079_begin_0 = const()[name = string("op_17079_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17079_end_0 = const()[name = string("op_17079_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_17079_end_mask_0 = const()[name = string("op_17079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17079_cast_fp16 = slice_by_index(begin = var_17079_begin_0, end = var_17079_end_0, end_mask = var_17079_end_mask_0, x = query_23_cast_fp16)[name = string("op_17079_cast_fp16")]; + tensor var_17083_begin_0 = const()[name = string("op_17083_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17083_end_0 = const()[name = string("op_17083_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_17083_end_mask_0 = const()[name = string("op_17083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17083_cast_fp16 = slice_by_index(begin = var_17083_begin_0, end = var_17083_end_0, end_mask = var_17083_end_mask_0, x = query_23_cast_fp16)[name = string("op_17083_cast_fp16")]; + tensor var_17087_begin_0 = const()[name = string("op_17087_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17087_end_0 = const()[name = string("op_17087_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_17087_end_mask_0 = const()[name = string("op_17087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17087_cast_fp16 = slice_by_index(begin = var_17087_begin_0, end = var_17087_end_0, end_mask = var_17087_end_mask_0, x = query_23_cast_fp16)[name = string("op_17087_cast_fp16")]; + tensor var_17091_begin_0 = const()[name = string("op_17091_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17091_end_0 = const()[name = string("op_17091_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_17091_end_mask_0 = const()[name = string("op_17091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17091_cast_fp16 = slice_by_index(begin = var_17091_begin_0, end = var_17091_end_0, end_mask = var_17091_end_mask_0, x = query_23_cast_fp16)[name = string("op_17091_cast_fp16")]; + tensor var_17095_begin_0 = const()[name = string("op_17095_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17095_end_0 = const()[name = string("op_17095_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_17095_end_mask_0 = const()[name = string("op_17095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17095_cast_fp16 = slice_by_index(begin = var_17095_begin_0, end = var_17095_end_0, end_mask = var_17095_end_mask_0, x = query_23_cast_fp16)[name = string("op_17095_cast_fp16")]; + tensor var_17099_begin_0 = const()[name = string("op_17099_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17099_end_0 = const()[name = string("op_17099_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_17099_end_mask_0 = const()[name = string("op_17099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17099_cast_fp16 = slice_by_index(begin = var_17099_begin_0, end = var_17099_end_0, end_mask = var_17099_end_mask_0, x = query_23_cast_fp16)[name = string("op_17099_cast_fp16")]; + tensor var_17103_begin_0 = const()[name = string("op_17103_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17103_end_0 = const()[name = string("op_17103_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_17103_end_mask_0 = const()[name = string("op_17103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17103_cast_fp16 = slice_by_index(begin = var_17103_begin_0, end = var_17103_end_0, end_mask = var_17103_end_mask_0, x = query_23_cast_fp16)[name = string("op_17103_cast_fp16")]; + tensor var_17107_begin_0 = const()[name = string("op_17107_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17107_end_0 = const()[name = string("op_17107_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_17107_end_mask_0 = const()[name = string("op_17107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17107_cast_fp16 = slice_by_index(begin = var_17107_begin_0, end = var_17107_end_0, end_mask = var_17107_end_mask_0, x = query_23_cast_fp16)[name = string("op_17107_cast_fp16")]; + tensor var_17111_begin_0 = const()[name = string("op_17111_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17111_end_0 = const()[name = string("op_17111_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_17111_end_mask_0 = const()[name = string("op_17111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17111_cast_fp16 = slice_by_index(begin = var_17111_begin_0, end = var_17111_end_0, end_mask = var_17111_end_mask_0, x = query_23_cast_fp16)[name = string("op_17111_cast_fp16")]; + tensor var_17115_begin_0 = const()[name = string("op_17115_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17115_end_0 = const()[name = string("op_17115_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_17115_end_mask_0 = const()[name = string("op_17115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17115_cast_fp16 = slice_by_index(begin = var_17115_begin_0, end = var_17115_end_0, end_mask = var_17115_end_mask_0, x = query_23_cast_fp16)[name = string("op_17115_cast_fp16")]; + tensor var_17119_begin_0 = const()[name = string("op_17119_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17119_end_0 = const()[name = string("op_17119_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_17119_end_mask_0 = const()[name = string("op_17119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17119_cast_fp16 = slice_by_index(begin = var_17119_begin_0, end = var_17119_end_0, end_mask = var_17119_end_mask_0, x = query_23_cast_fp16)[name = string("op_17119_cast_fp16")]; + tensor var_17123_begin_0 = const()[name = string("op_17123_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17123_end_0 = const()[name = string("op_17123_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_17123_end_mask_0 = const()[name = string("op_17123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17123_cast_fp16 = slice_by_index(begin = var_17123_begin_0, end = var_17123_end_0, end_mask = var_17123_end_mask_0, x = query_23_cast_fp16)[name = string("op_17123_cast_fp16")]; + tensor var_17127_begin_0 = const()[name = string("op_17127_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17127_end_0 = const()[name = string("op_17127_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_17127_end_mask_0 = const()[name = string("op_17127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17127_cast_fp16 = slice_by_index(begin = var_17127_begin_0, end = var_17127_end_0, end_mask = var_17127_end_mask_0, x = query_23_cast_fp16)[name = string("op_17127_cast_fp16")]; + tensor var_17131_begin_0 = const()[name = string("op_17131_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17131_end_0 = const()[name = string("op_17131_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_17131_end_mask_0 = const()[name = string("op_17131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17131_cast_fp16 = slice_by_index(begin = var_17131_begin_0, end = var_17131_end_0, end_mask = var_17131_end_mask_0, x = query_23_cast_fp16)[name = string("op_17131_cast_fp16")]; + tensor var_17140_begin_0 = const()[name = string("op_17140_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17140_end_0 = const()[name = string("op_17140_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17140_end_mask_0 = const()[name = string("op_17140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17140_cast_fp16 = slice_by_index(begin = var_17140_begin_0, end = var_17140_end_0, end_mask = var_17140_end_mask_0, x = var_17055_cast_fp16)[name = string("op_17140_cast_fp16")]; + tensor var_17147_begin_0 = const()[name = string("op_17147_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17147_end_0 = const()[name = string("op_17147_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17147_end_mask_0 = const()[name = string("op_17147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17147_cast_fp16 = slice_by_index(begin = var_17147_begin_0, end = var_17147_end_0, end_mask = var_17147_end_mask_0, x = var_17055_cast_fp16)[name = string("op_17147_cast_fp16")]; + tensor var_17154_begin_0 = const()[name = string("op_17154_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17154_end_0 = const()[name = string("op_17154_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17154_end_mask_0 = const()[name = string("op_17154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17154_cast_fp16 = slice_by_index(begin = var_17154_begin_0, end = var_17154_end_0, end_mask = var_17154_end_mask_0, x = var_17055_cast_fp16)[name = string("op_17154_cast_fp16")]; + tensor var_17161_begin_0 = const()[name = string("op_17161_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17161_end_0 = const()[name = string("op_17161_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17161_end_mask_0 = const()[name = string("op_17161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17161_cast_fp16 = slice_by_index(begin = var_17161_begin_0, end = var_17161_end_0, end_mask = var_17161_end_mask_0, x = var_17055_cast_fp16)[name = string("op_17161_cast_fp16")]; + tensor var_17168_begin_0 = const()[name = string("op_17168_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17168_end_0 = const()[name = string("op_17168_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17168_end_mask_0 = const()[name = string("op_17168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17168_cast_fp16 = slice_by_index(begin = var_17168_begin_0, end = var_17168_end_0, end_mask = var_17168_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17168_cast_fp16")]; + tensor var_17175_begin_0 = const()[name = string("op_17175_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17175_end_0 = const()[name = string("op_17175_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17175_end_mask_0 = const()[name = string("op_17175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17175_cast_fp16 = slice_by_index(begin = var_17175_begin_0, end = var_17175_end_0, end_mask = var_17175_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17175_cast_fp16")]; + tensor var_17182_begin_0 = const()[name = string("op_17182_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17182_end_0 = const()[name = string("op_17182_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17182_end_mask_0 = const()[name = string("op_17182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17182_cast_fp16 = slice_by_index(begin = var_17182_begin_0, end = var_17182_end_0, end_mask = var_17182_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17182_cast_fp16")]; + tensor var_17189_begin_0 = const()[name = string("op_17189_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17189_end_0 = const()[name = string("op_17189_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17189_end_mask_0 = const()[name = string("op_17189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17189_cast_fp16 = slice_by_index(begin = var_17189_begin_0, end = var_17189_end_0, end_mask = var_17189_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17189_cast_fp16")]; + tensor var_17196_begin_0 = const()[name = string("op_17196_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17196_end_0 = const()[name = string("op_17196_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17196_end_mask_0 = const()[name = string("op_17196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17196_cast_fp16 = slice_by_index(begin = var_17196_begin_0, end = var_17196_end_0, end_mask = var_17196_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17196_cast_fp16")]; + tensor var_17203_begin_0 = const()[name = string("op_17203_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17203_end_0 = const()[name = string("op_17203_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17203_end_mask_0 = const()[name = string("op_17203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17203_cast_fp16 = slice_by_index(begin = var_17203_begin_0, end = var_17203_end_0, end_mask = var_17203_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17203_cast_fp16")]; + tensor var_17210_begin_0 = const()[name = string("op_17210_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17210_end_0 = const()[name = string("op_17210_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17210_end_mask_0 = const()[name = string("op_17210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17210_cast_fp16 = slice_by_index(begin = var_17210_begin_0, end = var_17210_end_0, end_mask = var_17210_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17210_cast_fp16")]; + tensor var_17217_begin_0 = const()[name = string("op_17217_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17217_end_0 = const()[name = string("op_17217_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17217_end_mask_0 = const()[name = string("op_17217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17217_cast_fp16 = slice_by_index(begin = var_17217_begin_0, end = var_17217_end_0, end_mask = var_17217_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17217_cast_fp16")]; + tensor var_17224_begin_0 = const()[name = string("op_17224_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17224_end_0 = const()[name = string("op_17224_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17224_end_mask_0 = const()[name = string("op_17224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17224_cast_fp16 = slice_by_index(begin = var_17224_begin_0, end = var_17224_end_0, end_mask = var_17224_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17224_cast_fp16")]; + tensor var_17231_begin_0 = const()[name = string("op_17231_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17231_end_0 = const()[name = string("op_17231_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17231_end_mask_0 = const()[name = string("op_17231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17231_cast_fp16 = slice_by_index(begin = var_17231_begin_0, end = var_17231_end_0, end_mask = var_17231_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17231_cast_fp16")]; + tensor var_17238_begin_0 = const()[name = string("op_17238_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17238_end_0 = const()[name = string("op_17238_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17238_end_mask_0 = const()[name = string("op_17238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17238_cast_fp16 = slice_by_index(begin = var_17238_begin_0, end = var_17238_end_0, end_mask = var_17238_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17238_cast_fp16")]; + tensor var_17245_begin_0 = const()[name = string("op_17245_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17245_end_0 = const()[name = string("op_17245_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17245_end_mask_0 = const()[name = string("op_17245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17245_cast_fp16 = slice_by_index(begin = var_17245_begin_0, end = var_17245_end_0, end_mask = var_17245_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17245_cast_fp16")]; + tensor var_17252_begin_0 = const()[name = string("op_17252_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17252_end_0 = const()[name = string("op_17252_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17252_end_mask_0 = const()[name = string("op_17252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17252_cast_fp16 = slice_by_index(begin = var_17252_begin_0, end = var_17252_end_0, end_mask = var_17252_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17252_cast_fp16")]; + tensor var_17259_begin_0 = const()[name = string("op_17259_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17259_end_0 = const()[name = string("op_17259_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17259_end_mask_0 = const()[name = string("op_17259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17259_cast_fp16 = slice_by_index(begin = var_17259_begin_0, end = var_17259_end_0, end_mask = var_17259_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17259_cast_fp16")]; + tensor var_17266_begin_0 = const()[name = string("op_17266_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17266_end_0 = const()[name = string("op_17266_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17266_end_mask_0 = const()[name = string("op_17266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17266_cast_fp16 = slice_by_index(begin = var_17266_begin_0, end = var_17266_end_0, end_mask = var_17266_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17266_cast_fp16")]; + tensor var_17273_begin_0 = const()[name = string("op_17273_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17273_end_0 = const()[name = string("op_17273_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17273_end_mask_0 = const()[name = string("op_17273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17273_cast_fp16 = slice_by_index(begin = var_17273_begin_0, end = var_17273_end_0, end_mask = var_17273_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17273_cast_fp16")]; + tensor var_17280_begin_0 = const()[name = string("op_17280_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17280_end_0 = const()[name = string("op_17280_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17280_end_mask_0 = const()[name = string("op_17280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17280_cast_fp16 = slice_by_index(begin = var_17280_begin_0, end = var_17280_end_0, end_mask = var_17280_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17280_cast_fp16")]; + tensor var_17287_begin_0 = const()[name = string("op_17287_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17287_end_0 = const()[name = string("op_17287_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17287_end_mask_0 = const()[name = string("op_17287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17287_cast_fp16 = slice_by_index(begin = var_17287_begin_0, end = var_17287_end_0, end_mask = var_17287_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17287_cast_fp16")]; + tensor var_17294_begin_0 = const()[name = string("op_17294_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17294_end_0 = const()[name = string("op_17294_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17294_end_mask_0 = const()[name = string("op_17294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17294_cast_fp16 = slice_by_index(begin = var_17294_begin_0, end = var_17294_end_0, end_mask = var_17294_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17294_cast_fp16")]; + tensor var_17301_begin_0 = const()[name = string("op_17301_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17301_end_0 = const()[name = string("op_17301_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17301_end_mask_0 = const()[name = string("op_17301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17301_cast_fp16 = slice_by_index(begin = var_17301_begin_0, end = var_17301_end_0, end_mask = var_17301_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17301_cast_fp16")]; + tensor var_17308_begin_0 = const()[name = string("op_17308_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17308_end_0 = const()[name = string("op_17308_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17308_end_mask_0 = const()[name = string("op_17308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17308_cast_fp16 = slice_by_index(begin = var_17308_begin_0, end = var_17308_end_0, end_mask = var_17308_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17308_cast_fp16")]; + tensor var_17315_begin_0 = const()[name = string("op_17315_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17315_end_0 = const()[name = string("op_17315_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17315_end_mask_0 = const()[name = string("op_17315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17315_cast_fp16 = slice_by_index(begin = var_17315_begin_0, end = var_17315_end_0, end_mask = var_17315_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17315_cast_fp16")]; + tensor var_17322_begin_0 = const()[name = string("op_17322_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17322_end_0 = const()[name = string("op_17322_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17322_end_mask_0 = const()[name = string("op_17322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17322_cast_fp16 = slice_by_index(begin = var_17322_begin_0, end = var_17322_end_0, end_mask = var_17322_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17322_cast_fp16")]; + tensor var_17329_begin_0 = const()[name = string("op_17329_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17329_end_0 = const()[name = string("op_17329_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17329_end_mask_0 = const()[name = string("op_17329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17329_cast_fp16 = slice_by_index(begin = var_17329_begin_0, end = var_17329_end_0, end_mask = var_17329_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17329_cast_fp16")]; + tensor var_17336_begin_0 = const()[name = string("op_17336_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17336_end_0 = const()[name = string("op_17336_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17336_end_mask_0 = const()[name = string("op_17336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17336_cast_fp16 = slice_by_index(begin = var_17336_begin_0, end = var_17336_end_0, end_mask = var_17336_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17336_cast_fp16")]; + tensor var_17343_begin_0 = const()[name = string("op_17343_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17343_end_0 = const()[name = string("op_17343_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17343_end_mask_0 = const()[name = string("op_17343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17343_cast_fp16 = slice_by_index(begin = var_17343_begin_0, end = var_17343_end_0, end_mask = var_17343_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17343_cast_fp16")]; + tensor var_17350_begin_0 = const()[name = string("op_17350_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17350_end_0 = const()[name = string("op_17350_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17350_end_mask_0 = const()[name = string("op_17350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17350_cast_fp16 = slice_by_index(begin = var_17350_begin_0, end = var_17350_end_0, end_mask = var_17350_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17350_cast_fp16")]; + tensor var_17357_begin_0 = const()[name = string("op_17357_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17357_end_0 = const()[name = string("op_17357_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17357_end_mask_0 = const()[name = string("op_17357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17357_cast_fp16 = slice_by_index(begin = var_17357_begin_0, end = var_17357_end_0, end_mask = var_17357_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17357_cast_fp16")]; + tensor var_17364_begin_0 = const()[name = string("op_17364_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17364_end_0 = const()[name = string("op_17364_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17364_end_mask_0 = const()[name = string("op_17364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17364_cast_fp16 = slice_by_index(begin = var_17364_begin_0, end = var_17364_end_0, end_mask = var_17364_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17364_cast_fp16")]; + tensor var_17371_begin_0 = const()[name = string("op_17371_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17371_end_0 = const()[name = string("op_17371_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17371_end_mask_0 = const()[name = string("op_17371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17371_cast_fp16 = slice_by_index(begin = var_17371_begin_0, end = var_17371_end_0, end_mask = var_17371_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17371_cast_fp16")]; + tensor var_17378_begin_0 = const()[name = string("op_17378_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17378_end_0 = const()[name = string("op_17378_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17378_end_mask_0 = const()[name = string("op_17378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17378_cast_fp16 = slice_by_index(begin = var_17378_begin_0, end = var_17378_end_0, end_mask = var_17378_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17378_cast_fp16")]; + tensor var_17385_begin_0 = const()[name = string("op_17385_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17385_end_0 = const()[name = string("op_17385_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17385_end_mask_0 = const()[name = string("op_17385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17385_cast_fp16 = slice_by_index(begin = var_17385_begin_0, end = var_17385_end_0, end_mask = var_17385_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17385_cast_fp16")]; + tensor var_17392_begin_0 = const()[name = string("op_17392_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17392_end_0 = const()[name = string("op_17392_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17392_end_mask_0 = const()[name = string("op_17392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17392_cast_fp16 = slice_by_index(begin = var_17392_begin_0, end = var_17392_end_0, end_mask = var_17392_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17392_cast_fp16")]; + tensor var_17399_begin_0 = const()[name = string("op_17399_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17399_end_0 = const()[name = string("op_17399_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17399_end_mask_0 = const()[name = string("op_17399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17399_cast_fp16 = slice_by_index(begin = var_17399_begin_0, end = var_17399_end_0, end_mask = var_17399_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17399_cast_fp16")]; + tensor var_17406_begin_0 = const()[name = string("op_17406_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17406_end_0 = const()[name = string("op_17406_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17406_end_mask_0 = const()[name = string("op_17406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17406_cast_fp16 = slice_by_index(begin = var_17406_begin_0, end = var_17406_end_0, end_mask = var_17406_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17406_cast_fp16")]; + tensor var_17413_begin_0 = const()[name = string("op_17413_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17413_end_0 = const()[name = string("op_17413_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17413_end_mask_0 = const()[name = string("op_17413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17413_cast_fp16 = slice_by_index(begin = var_17413_begin_0, end = var_17413_end_0, end_mask = var_17413_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17413_cast_fp16")]; + tensor var_17420_begin_0 = const()[name = string("op_17420_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17420_end_0 = const()[name = string("op_17420_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17420_end_mask_0 = const()[name = string("op_17420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17420_cast_fp16 = slice_by_index(begin = var_17420_begin_0, end = var_17420_end_0, end_mask = var_17420_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17420_cast_fp16")]; + tensor var_17427_begin_0 = const()[name = string("op_17427_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17427_end_0 = const()[name = string("op_17427_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17427_end_mask_0 = const()[name = string("op_17427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17427_cast_fp16 = slice_by_index(begin = var_17427_begin_0, end = var_17427_end_0, end_mask = var_17427_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17427_cast_fp16")]; + tensor var_17434_begin_0 = const()[name = string("op_17434_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17434_end_0 = const()[name = string("op_17434_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17434_end_mask_0 = const()[name = string("op_17434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17434_cast_fp16 = slice_by_index(begin = var_17434_begin_0, end = var_17434_end_0, end_mask = var_17434_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17434_cast_fp16")]; + tensor var_17441_begin_0 = const()[name = string("op_17441_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17441_end_0 = const()[name = string("op_17441_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17441_end_mask_0 = const()[name = string("op_17441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17441_cast_fp16 = slice_by_index(begin = var_17441_begin_0, end = var_17441_end_0, end_mask = var_17441_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17441_cast_fp16")]; + tensor var_17448_begin_0 = const()[name = string("op_17448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17448_end_0 = const()[name = string("op_17448_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17448_end_mask_0 = const()[name = string("op_17448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17448_cast_fp16 = slice_by_index(begin = var_17448_begin_0, end = var_17448_end_0, end_mask = var_17448_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17448_cast_fp16")]; + tensor var_17455_begin_0 = const()[name = string("op_17455_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17455_end_0 = const()[name = string("op_17455_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17455_end_mask_0 = const()[name = string("op_17455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17455_cast_fp16 = slice_by_index(begin = var_17455_begin_0, end = var_17455_end_0, end_mask = var_17455_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17455_cast_fp16")]; + tensor var_17462_begin_0 = const()[name = string("op_17462_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17462_end_0 = const()[name = string("op_17462_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17462_end_mask_0 = const()[name = string("op_17462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17462_cast_fp16 = slice_by_index(begin = var_17462_begin_0, end = var_17462_end_0, end_mask = var_17462_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17462_cast_fp16")]; + tensor var_17469_begin_0 = const()[name = string("op_17469_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17469_end_0 = const()[name = string("op_17469_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17469_end_mask_0 = const()[name = string("op_17469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17469_cast_fp16 = slice_by_index(begin = var_17469_begin_0, end = var_17469_end_0, end_mask = var_17469_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17469_cast_fp16")]; + tensor var_17476_begin_0 = const()[name = string("op_17476_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17476_end_0 = const()[name = string("op_17476_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17476_end_mask_0 = const()[name = string("op_17476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17476_cast_fp16 = slice_by_index(begin = var_17476_begin_0, end = var_17476_end_0, end_mask = var_17476_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17476_cast_fp16")]; + tensor var_17483_begin_0 = const()[name = string("op_17483_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17483_end_0 = const()[name = string("op_17483_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17483_end_mask_0 = const()[name = string("op_17483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17483_cast_fp16 = slice_by_index(begin = var_17483_begin_0, end = var_17483_end_0, end_mask = var_17483_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17483_cast_fp16")]; + tensor var_17490_begin_0 = const()[name = string("op_17490_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17490_end_0 = const()[name = string("op_17490_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17490_end_mask_0 = const()[name = string("op_17490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17490_cast_fp16 = slice_by_index(begin = var_17490_begin_0, end = var_17490_end_0, end_mask = var_17490_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17490_cast_fp16")]; + tensor var_17497_begin_0 = const()[name = string("op_17497_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17497_end_0 = const()[name = string("op_17497_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17497_end_mask_0 = const()[name = string("op_17497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17497_cast_fp16 = slice_by_index(begin = var_17497_begin_0, end = var_17497_end_0, end_mask = var_17497_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17497_cast_fp16")]; + tensor var_17504_begin_0 = const()[name = string("op_17504_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17504_end_0 = const()[name = string("op_17504_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17504_end_mask_0 = const()[name = string("op_17504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17504_cast_fp16 = slice_by_index(begin = var_17504_begin_0, end = var_17504_end_0, end_mask = var_17504_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17504_cast_fp16")]; + tensor var_17511_begin_0 = const()[name = string("op_17511_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17511_end_0 = const()[name = string("op_17511_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17511_end_mask_0 = const()[name = string("op_17511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17511_cast_fp16 = slice_by_index(begin = var_17511_begin_0, end = var_17511_end_0, end_mask = var_17511_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17511_cast_fp16")]; + tensor var_17518_begin_0 = const()[name = string("op_17518_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17518_end_0 = const()[name = string("op_17518_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17518_end_mask_0 = const()[name = string("op_17518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17518_cast_fp16 = slice_by_index(begin = var_17518_begin_0, end = var_17518_end_0, end_mask = var_17518_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17518_cast_fp16")]; + tensor var_17525_begin_0 = const()[name = string("op_17525_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17525_end_0 = const()[name = string("op_17525_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17525_end_mask_0 = const()[name = string("op_17525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17525_cast_fp16 = slice_by_index(begin = var_17525_begin_0, end = var_17525_end_0, end_mask = var_17525_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17525_cast_fp16")]; + tensor var_17532_begin_0 = const()[name = string("op_17532_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17532_end_0 = const()[name = string("op_17532_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17532_end_mask_0 = const()[name = string("op_17532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17532_cast_fp16 = slice_by_index(begin = var_17532_begin_0, end = var_17532_end_0, end_mask = var_17532_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17532_cast_fp16")]; + tensor var_17539_begin_0 = const()[name = string("op_17539_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17539_end_0 = const()[name = string("op_17539_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17539_end_mask_0 = const()[name = string("op_17539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17539_cast_fp16 = slice_by_index(begin = var_17539_begin_0, end = var_17539_end_0, end_mask = var_17539_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17539_cast_fp16")]; + tensor var_17546_begin_0 = const()[name = string("op_17546_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17546_end_0 = const()[name = string("op_17546_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17546_end_mask_0 = const()[name = string("op_17546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17546_cast_fp16 = slice_by_index(begin = var_17546_begin_0, end = var_17546_end_0, end_mask = var_17546_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17546_cast_fp16")]; + tensor var_17553_begin_0 = const()[name = string("op_17553_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17553_end_0 = const()[name = string("op_17553_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17553_end_mask_0 = const()[name = string("op_17553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17553_cast_fp16 = slice_by_index(begin = var_17553_begin_0, end = var_17553_end_0, end_mask = var_17553_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17553_cast_fp16")]; + tensor var_17560_begin_0 = const()[name = string("op_17560_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17560_end_0 = const()[name = string("op_17560_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17560_end_mask_0 = const()[name = string("op_17560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17560_cast_fp16 = slice_by_index(begin = var_17560_begin_0, end = var_17560_end_0, end_mask = var_17560_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17560_cast_fp16")]; + tensor var_17567_begin_0 = const()[name = string("op_17567_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17567_end_0 = const()[name = string("op_17567_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17567_end_mask_0 = const()[name = string("op_17567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17567_cast_fp16 = slice_by_index(begin = var_17567_begin_0, end = var_17567_end_0, end_mask = var_17567_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17567_cast_fp16")]; + tensor var_17574_begin_0 = const()[name = string("op_17574_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17574_end_0 = const()[name = string("op_17574_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17574_end_mask_0 = const()[name = string("op_17574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17574_cast_fp16 = slice_by_index(begin = var_17574_begin_0, end = var_17574_end_0, end_mask = var_17574_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17574_cast_fp16")]; + tensor var_17581_begin_0 = const()[name = string("op_17581_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17581_end_0 = const()[name = string("op_17581_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17581_end_mask_0 = const()[name = string("op_17581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17581_cast_fp16 = slice_by_index(begin = var_17581_begin_0, end = var_17581_end_0, end_mask = var_17581_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17581_cast_fp16")]; + tensor var_17588_begin_0 = const()[name = string("op_17588_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17588_end_0 = const()[name = string("op_17588_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17588_end_mask_0 = const()[name = string("op_17588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17588_cast_fp16 = slice_by_index(begin = var_17588_begin_0, end = var_17588_end_0, end_mask = var_17588_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17588_cast_fp16")]; + tensor var_17595_begin_0 = const()[name = string("op_17595_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17595_end_0 = const()[name = string("op_17595_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17595_end_mask_0 = const()[name = string("op_17595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17595_cast_fp16 = slice_by_index(begin = var_17595_begin_0, end = var_17595_end_0, end_mask = var_17595_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17595_cast_fp16")]; + tensor var_17602_begin_0 = const()[name = string("op_17602_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17602_end_0 = const()[name = string("op_17602_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17602_end_mask_0 = const()[name = string("op_17602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17602_cast_fp16 = slice_by_index(begin = var_17602_begin_0, end = var_17602_end_0, end_mask = var_17602_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17602_cast_fp16")]; + tensor var_17609_begin_0 = const()[name = string("op_17609_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17609_end_0 = const()[name = string("op_17609_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17609_end_mask_0 = const()[name = string("op_17609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17609_cast_fp16 = slice_by_index(begin = var_17609_begin_0, end = var_17609_end_0, end_mask = var_17609_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17609_cast_fp16")]; + tensor var_17616_begin_0 = const()[name = string("op_17616_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17616_end_0 = const()[name = string("op_17616_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17616_end_mask_0 = const()[name = string("op_17616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17616_cast_fp16 = slice_by_index(begin = var_17616_begin_0, end = var_17616_end_0, end_mask = var_17616_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17616_cast_fp16")]; + tensor var_17623_begin_0 = const()[name = string("op_17623_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17623_end_0 = const()[name = string("op_17623_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17623_end_mask_0 = const()[name = string("op_17623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17623_cast_fp16 = slice_by_index(begin = var_17623_begin_0, end = var_17623_end_0, end_mask = var_17623_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17623_cast_fp16")]; + tensor var_17630_begin_0 = const()[name = string("op_17630_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17630_end_0 = const()[name = string("op_17630_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17630_end_mask_0 = const()[name = string("op_17630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17630_cast_fp16 = slice_by_index(begin = var_17630_begin_0, end = var_17630_end_0, end_mask = var_17630_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17630_cast_fp16")]; + tensor var_17637_begin_0 = const()[name = string("op_17637_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17637_end_0 = const()[name = string("op_17637_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17637_end_mask_0 = const()[name = string("op_17637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17637_cast_fp16 = slice_by_index(begin = var_17637_begin_0, end = var_17637_end_0, end_mask = var_17637_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17637_cast_fp16")]; + tensor var_17644_begin_0 = const()[name = string("op_17644_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17644_end_0 = const()[name = string("op_17644_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17644_end_mask_0 = const()[name = string("op_17644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17644_cast_fp16 = slice_by_index(begin = var_17644_begin_0, end = var_17644_end_0, end_mask = var_17644_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17644_cast_fp16")]; + tensor var_17651_begin_0 = const()[name = string("op_17651_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17651_end_0 = const()[name = string("op_17651_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17651_end_mask_0 = const()[name = string("op_17651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17651_cast_fp16 = slice_by_index(begin = var_17651_begin_0, end = var_17651_end_0, end_mask = var_17651_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17651_cast_fp16")]; + tensor var_17658_begin_0 = const()[name = string("op_17658_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17658_end_0 = const()[name = string("op_17658_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17658_end_mask_0 = const()[name = string("op_17658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17658_cast_fp16 = slice_by_index(begin = var_17658_begin_0, end = var_17658_end_0, end_mask = var_17658_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17658_cast_fp16")]; + tensor var_17665_begin_0 = const()[name = string("op_17665_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17665_end_0 = const()[name = string("op_17665_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17665_end_mask_0 = const()[name = string("op_17665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17665_cast_fp16 = slice_by_index(begin = var_17665_begin_0, end = var_17665_end_0, end_mask = var_17665_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17665_cast_fp16")]; + tensor var_17672_begin_0 = const()[name = string("op_17672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17672_end_0 = const()[name = string("op_17672_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17672_end_mask_0 = const()[name = string("op_17672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17672_cast_fp16 = slice_by_index(begin = var_17672_begin_0, end = var_17672_end_0, end_mask = var_17672_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17672_cast_fp16")]; + tensor var_17679_begin_0 = const()[name = string("op_17679_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17679_end_0 = const()[name = string("op_17679_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17679_end_mask_0 = const()[name = string("op_17679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17679_cast_fp16 = slice_by_index(begin = var_17679_begin_0, end = var_17679_end_0, end_mask = var_17679_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17679_cast_fp16")]; + tensor var_17686_begin_0 = const()[name = string("op_17686_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17686_end_0 = const()[name = string("op_17686_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17686_end_mask_0 = const()[name = string("op_17686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17686_cast_fp16 = slice_by_index(begin = var_17686_begin_0, end = var_17686_end_0, end_mask = var_17686_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17686_cast_fp16")]; + tensor var_17693_begin_0 = const()[name = string("op_17693_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17693_end_0 = const()[name = string("op_17693_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17693_end_mask_0 = const()[name = string("op_17693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17693_cast_fp16 = slice_by_index(begin = var_17693_begin_0, end = var_17693_end_0, end_mask = var_17693_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17693_cast_fp16")]; + tensor k_23_perm_0 = const()[name = string("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17698_begin_0 = const()[name = string("op_17698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17698_end_0 = const()[name = string("op_17698_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_17698_end_mask_0 = const()[name = string("op_17698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = key_23_cast_fp16)[name = string("transpose_20")]; + tensor var_17698_cast_fp16 = slice_by_index(begin = var_17698_begin_0, end = var_17698_end_0, end_mask = var_17698_end_mask_0, x = k_23_cast_fp16)[name = string("op_17698_cast_fp16")]; + tensor var_17702_begin_0 = const()[name = string("op_17702_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17702_end_0 = const()[name = string("op_17702_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_17702_end_mask_0 = const()[name = string("op_17702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17702_cast_fp16 = slice_by_index(begin = var_17702_begin_0, end = var_17702_end_0, end_mask = var_17702_end_mask_0, x = k_23_cast_fp16)[name = string("op_17702_cast_fp16")]; + tensor var_17706_begin_0 = const()[name = string("op_17706_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17706_end_0 = const()[name = string("op_17706_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_17706_end_mask_0 = const()[name = string("op_17706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17706_cast_fp16 = slice_by_index(begin = var_17706_begin_0, end = var_17706_end_0, end_mask = var_17706_end_mask_0, x = k_23_cast_fp16)[name = string("op_17706_cast_fp16")]; + tensor var_17710_begin_0 = const()[name = string("op_17710_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17710_end_0 = const()[name = string("op_17710_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_17710_end_mask_0 = const()[name = string("op_17710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17710_cast_fp16 = slice_by_index(begin = var_17710_begin_0, end = var_17710_end_0, end_mask = var_17710_end_mask_0, x = k_23_cast_fp16)[name = string("op_17710_cast_fp16")]; + tensor var_17714_begin_0 = const()[name = string("op_17714_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17714_end_0 = const()[name = string("op_17714_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_17714_end_mask_0 = const()[name = string("op_17714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17714_cast_fp16 = slice_by_index(begin = var_17714_begin_0, end = var_17714_end_0, end_mask = var_17714_end_mask_0, x = k_23_cast_fp16)[name = string("op_17714_cast_fp16")]; + tensor var_17718_begin_0 = const()[name = string("op_17718_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17718_end_0 = const()[name = string("op_17718_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_17718_end_mask_0 = const()[name = string("op_17718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17718_cast_fp16 = slice_by_index(begin = var_17718_begin_0, end = var_17718_end_0, end_mask = var_17718_end_mask_0, x = k_23_cast_fp16)[name = string("op_17718_cast_fp16")]; + tensor var_17722_begin_0 = const()[name = string("op_17722_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17722_end_0 = const()[name = string("op_17722_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_17722_end_mask_0 = const()[name = string("op_17722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17722_cast_fp16 = slice_by_index(begin = var_17722_begin_0, end = var_17722_end_0, end_mask = var_17722_end_mask_0, x = k_23_cast_fp16)[name = string("op_17722_cast_fp16")]; + tensor var_17726_begin_0 = const()[name = string("op_17726_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17726_end_0 = const()[name = string("op_17726_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_17726_end_mask_0 = const()[name = string("op_17726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17726_cast_fp16 = slice_by_index(begin = var_17726_begin_0, end = var_17726_end_0, end_mask = var_17726_end_mask_0, x = k_23_cast_fp16)[name = string("op_17726_cast_fp16")]; + tensor var_17730_begin_0 = const()[name = string("op_17730_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17730_end_0 = const()[name = string("op_17730_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_17730_end_mask_0 = const()[name = string("op_17730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17730_cast_fp16 = slice_by_index(begin = var_17730_begin_0, end = var_17730_end_0, end_mask = var_17730_end_mask_0, x = k_23_cast_fp16)[name = string("op_17730_cast_fp16")]; + tensor var_17734_begin_0 = const()[name = string("op_17734_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17734_end_0 = const()[name = string("op_17734_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_17734_end_mask_0 = const()[name = string("op_17734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17734_cast_fp16 = slice_by_index(begin = var_17734_begin_0, end = var_17734_end_0, end_mask = var_17734_end_mask_0, x = k_23_cast_fp16)[name = string("op_17734_cast_fp16")]; + tensor var_17738_begin_0 = const()[name = string("op_17738_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17738_end_0 = const()[name = string("op_17738_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_17738_end_mask_0 = const()[name = string("op_17738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17738_cast_fp16 = slice_by_index(begin = var_17738_begin_0, end = var_17738_end_0, end_mask = var_17738_end_mask_0, x = k_23_cast_fp16)[name = string("op_17738_cast_fp16")]; + tensor var_17742_begin_0 = const()[name = string("op_17742_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17742_end_0 = const()[name = string("op_17742_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_17742_end_mask_0 = const()[name = string("op_17742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17742_cast_fp16 = slice_by_index(begin = var_17742_begin_0, end = var_17742_end_0, end_mask = var_17742_end_mask_0, x = k_23_cast_fp16)[name = string("op_17742_cast_fp16")]; + tensor var_17746_begin_0 = const()[name = string("op_17746_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17746_end_0 = const()[name = string("op_17746_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_17746_end_mask_0 = const()[name = string("op_17746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17746_cast_fp16 = slice_by_index(begin = var_17746_begin_0, end = var_17746_end_0, end_mask = var_17746_end_mask_0, x = k_23_cast_fp16)[name = string("op_17746_cast_fp16")]; + tensor var_17750_begin_0 = const()[name = string("op_17750_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17750_end_0 = const()[name = string("op_17750_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_17750_end_mask_0 = const()[name = string("op_17750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17750_cast_fp16 = slice_by_index(begin = var_17750_begin_0, end = var_17750_end_0, end_mask = var_17750_end_mask_0, x = k_23_cast_fp16)[name = string("op_17750_cast_fp16")]; + tensor var_17754_begin_0 = const()[name = string("op_17754_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17754_end_0 = const()[name = string("op_17754_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_17754_end_mask_0 = const()[name = string("op_17754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17754_cast_fp16 = slice_by_index(begin = var_17754_begin_0, end = var_17754_end_0, end_mask = var_17754_end_mask_0, x = k_23_cast_fp16)[name = string("op_17754_cast_fp16")]; + tensor var_17758_begin_0 = const()[name = string("op_17758_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17758_end_0 = const()[name = string("op_17758_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_17758_end_mask_0 = const()[name = string("op_17758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17758_cast_fp16 = slice_by_index(begin = var_17758_begin_0, end = var_17758_end_0, end_mask = var_17758_end_mask_0, x = k_23_cast_fp16)[name = string("op_17758_cast_fp16")]; + tensor var_17762_begin_0 = const()[name = string("op_17762_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17762_end_0 = const()[name = string("op_17762_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_17762_end_mask_0 = const()[name = string("op_17762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17762_cast_fp16 = slice_by_index(begin = var_17762_begin_0, end = var_17762_end_0, end_mask = var_17762_end_mask_0, x = k_23_cast_fp16)[name = string("op_17762_cast_fp16")]; + tensor var_17766_begin_0 = const()[name = string("op_17766_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17766_end_0 = const()[name = string("op_17766_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_17766_end_mask_0 = const()[name = string("op_17766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17766_cast_fp16 = slice_by_index(begin = var_17766_begin_0, end = var_17766_end_0, end_mask = var_17766_end_mask_0, x = k_23_cast_fp16)[name = string("op_17766_cast_fp16")]; + tensor var_17770_begin_0 = const()[name = string("op_17770_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17770_end_0 = const()[name = string("op_17770_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_17770_end_mask_0 = const()[name = string("op_17770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17770_cast_fp16 = slice_by_index(begin = var_17770_begin_0, end = var_17770_end_0, end_mask = var_17770_end_mask_0, x = k_23_cast_fp16)[name = string("op_17770_cast_fp16")]; + tensor var_17774_begin_0 = const()[name = string("op_17774_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17774_end_0 = const()[name = string("op_17774_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_17774_end_mask_0 = const()[name = string("op_17774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17774_cast_fp16 = slice_by_index(begin = var_17774_begin_0, end = var_17774_end_0, end_mask = var_17774_end_mask_0, x = k_23_cast_fp16)[name = string("op_17774_cast_fp16")]; + tensor var_17776_begin_0 = const()[name = string("op_17776_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17776_end_0 = const()[name = string("op_17776_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17776_end_mask_0 = const()[name = string("op_17776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17776_cast_fp16 = slice_by_index(begin = var_17776_begin_0, end = var_17776_end_0, end_mask = var_17776_end_mask_0, x = value_23_cast_fp16)[name = string("op_17776_cast_fp16")]; + tensor var_17780_begin_0 = const()[name = string("op_17780_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17780_end_0 = const()[name = string("op_17780_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_17780_end_mask_0 = const()[name = string("op_17780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17780_cast_fp16 = slice_by_index(begin = var_17780_begin_0, end = var_17780_end_0, end_mask = var_17780_end_mask_0, x = value_23_cast_fp16)[name = string("op_17780_cast_fp16")]; + tensor var_17784_begin_0 = const()[name = string("op_17784_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17784_end_0 = const()[name = string("op_17784_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_17784_end_mask_0 = const()[name = string("op_17784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17784_cast_fp16 = slice_by_index(begin = var_17784_begin_0, end = var_17784_end_0, end_mask = var_17784_end_mask_0, x = value_23_cast_fp16)[name = string("op_17784_cast_fp16")]; + tensor var_17788_begin_0 = const()[name = string("op_17788_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17788_end_0 = const()[name = string("op_17788_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_17788_end_mask_0 = const()[name = string("op_17788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17788_cast_fp16 = slice_by_index(begin = var_17788_begin_0, end = var_17788_end_0, end_mask = var_17788_end_mask_0, x = value_23_cast_fp16)[name = string("op_17788_cast_fp16")]; + tensor var_17792_begin_0 = const()[name = string("op_17792_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17792_end_0 = const()[name = string("op_17792_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_17792_end_mask_0 = const()[name = string("op_17792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17792_cast_fp16 = slice_by_index(begin = var_17792_begin_0, end = var_17792_end_0, end_mask = var_17792_end_mask_0, x = value_23_cast_fp16)[name = string("op_17792_cast_fp16")]; + tensor var_17796_begin_0 = const()[name = string("op_17796_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17796_end_0 = const()[name = string("op_17796_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_17796_end_mask_0 = const()[name = string("op_17796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17796_cast_fp16 = slice_by_index(begin = var_17796_begin_0, end = var_17796_end_0, end_mask = var_17796_end_mask_0, x = value_23_cast_fp16)[name = string("op_17796_cast_fp16")]; + tensor var_17800_begin_0 = const()[name = string("op_17800_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17800_end_0 = const()[name = string("op_17800_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_17800_end_mask_0 = const()[name = string("op_17800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17800_cast_fp16 = slice_by_index(begin = var_17800_begin_0, end = var_17800_end_0, end_mask = var_17800_end_mask_0, x = value_23_cast_fp16)[name = string("op_17800_cast_fp16")]; + tensor var_17804_begin_0 = const()[name = string("op_17804_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17804_end_0 = const()[name = string("op_17804_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_17804_end_mask_0 = const()[name = string("op_17804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17804_cast_fp16 = slice_by_index(begin = var_17804_begin_0, end = var_17804_end_0, end_mask = var_17804_end_mask_0, x = value_23_cast_fp16)[name = string("op_17804_cast_fp16")]; + tensor var_17808_begin_0 = const()[name = string("op_17808_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17808_end_0 = const()[name = string("op_17808_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_17808_end_mask_0 = const()[name = string("op_17808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17808_cast_fp16 = slice_by_index(begin = var_17808_begin_0, end = var_17808_end_0, end_mask = var_17808_end_mask_0, x = value_23_cast_fp16)[name = string("op_17808_cast_fp16")]; + tensor var_17812_begin_0 = const()[name = string("op_17812_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17812_end_0 = const()[name = string("op_17812_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_17812_end_mask_0 = const()[name = string("op_17812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17812_cast_fp16 = slice_by_index(begin = var_17812_begin_0, end = var_17812_end_0, end_mask = var_17812_end_mask_0, x = value_23_cast_fp16)[name = string("op_17812_cast_fp16")]; + tensor var_17816_begin_0 = const()[name = string("op_17816_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17816_end_0 = const()[name = string("op_17816_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_17816_end_mask_0 = const()[name = string("op_17816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17816_cast_fp16 = slice_by_index(begin = var_17816_begin_0, end = var_17816_end_0, end_mask = var_17816_end_mask_0, x = value_23_cast_fp16)[name = string("op_17816_cast_fp16")]; + tensor var_17820_begin_0 = const()[name = string("op_17820_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17820_end_0 = const()[name = string("op_17820_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_17820_end_mask_0 = const()[name = string("op_17820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17820_cast_fp16 = slice_by_index(begin = var_17820_begin_0, end = var_17820_end_0, end_mask = var_17820_end_mask_0, x = value_23_cast_fp16)[name = string("op_17820_cast_fp16")]; + tensor var_17824_begin_0 = const()[name = string("op_17824_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17824_end_0 = const()[name = string("op_17824_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_17824_end_mask_0 = const()[name = string("op_17824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17824_cast_fp16 = slice_by_index(begin = var_17824_begin_0, end = var_17824_end_0, end_mask = var_17824_end_mask_0, x = value_23_cast_fp16)[name = string("op_17824_cast_fp16")]; + tensor var_17828_begin_0 = const()[name = string("op_17828_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17828_end_0 = const()[name = string("op_17828_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_17828_end_mask_0 = const()[name = string("op_17828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17828_cast_fp16 = slice_by_index(begin = var_17828_begin_0, end = var_17828_end_0, end_mask = var_17828_end_mask_0, x = value_23_cast_fp16)[name = string("op_17828_cast_fp16")]; + tensor var_17832_begin_0 = const()[name = string("op_17832_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17832_end_0 = const()[name = string("op_17832_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_17832_end_mask_0 = const()[name = string("op_17832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17832_cast_fp16 = slice_by_index(begin = var_17832_begin_0, end = var_17832_end_0, end_mask = var_17832_end_mask_0, x = value_23_cast_fp16)[name = string("op_17832_cast_fp16")]; + tensor var_17836_begin_0 = const()[name = string("op_17836_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17836_end_0 = const()[name = string("op_17836_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_17836_end_mask_0 = const()[name = string("op_17836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17836_cast_fp16 = slice_by_index(begin = var_17836_begin_0, end = var_17836_end_0, end_mask = var_17836_end_mask_0, x = value_23_cast_fp16)[name = string("op_17836_cast_fp16")]; + tensor var_17840_begin_0 = const()[name = string("op_17840_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17840_end_0 = const()[name = string("op_17840_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_17840_end_mask_0 = const()[name = string("op_17840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17840_cast_fp16 = slice_by_index(begin = var_17840_begin_0, end = var_17840_end_0, end_mask = var_17840_end_mask_0, x = value_23_cast_fp16)[name = string("op_17840_cast_fp16")]; + tensor var_17844_begin_0 = const()[name = string("op_17844_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17844_end_0 = const()[name = string("op_17844_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_17844_end_mask_0 = const()[name = string("op_17844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17844_cast_fp16 = slice_by_index(begin = var_17844_begin_0, end = var_17844_end_0, end_mask = var_17844_end_mask_0, x = value_23_cast_fp16)[name = string("op_17844_cast_fp16")]; + tensor var_17848_begin_0 = const()[name = string("op_17848_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17848_end_0 = const()[name = string("op_17848_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_17848_end_mask_0 = const()[name = string("op_17848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17848_cast_fp16 = slice_by_index(begin = var_17848_begin_0, end = var_17848_end_0, end_mask = var_17848_end_mask_0, x = value_23_cast_fp16)[name = string("op_17848_cast_fp16")]; + tensor var_17852_begin_0 = const()[name = string("op_17852_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17852_end_0 = const()[name = string("op_17852_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_17852_end_mask_0 = const()[name = string("op_17852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17852_cast_fp16 = slice_by_index(begin = var_17852_begin_0, end = var_17852_end_0, end_mask = var_17852_end_mask_0, x = value_23_cast_fp16)[name = string("op_17852_cast_fp16")]; + string _SplitHeadsQ__mh_w_1761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1761_equation_0, values = (var_17698_cast_fp16, var_17140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1761_cast_fp16")]; + string _SplitHeadsQ__mh_w_1763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1763_equation_0, values = (var_17698_cast_fp16, var_17147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1763_cast_fp16")]; + string _SplitHeadsQ__mh_w_1765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1765_equation_0, values = (var_17698_cast_fp16, var_17154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1765_cast_fp16")]; + string _SplitHeadsQ__mh_w_1767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1767_equation_0, values = (var_17698_cast_fp16, var_17161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1767_cast_fp16")]; + string _SplitHeadsQ__mh_w_1769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1769_equation_0, values = (var_17702_cast_fp16, var_17168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1769_cast_fp16")]; + string _SplitHeadsQ__mh_w_1771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1771_equation_0, values = (var_17702_cast_fp16, var_17175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1771_cast_fp16")]; + string _SplitHeadsQ__mh_w_1773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1773_equation_0, values = (var_17702_cast_fp16, var_17182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1773_cast_fp16")]; + string _SplitHeadsQ__mh_w_1775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1775_equation_0, values = (var_17702_cast_fp16, var_17189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1775_cast_fp16")]; + string _SplitHeadsQ__mh_w_1777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1777_equation_0, values = (var_17706_cast_fp16, var_17196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1777_cast_fp16")]; + string _SplitHeadsQ__mh_w_1779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1779_equation_0, values = (var_17706_cast_fp16, var_17203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1779_cast_fp16")]; + string _SplitHeadsQ__mh_w_1781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1781_equation_0, values = (var_17706_cast_fp16, var_17210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1781_cast_fp16")]; + string _SplitHeadsQ__mh_w_1783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1783_equation_0, values = (var_17706_cast_fp16, var_17217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1783_cast_fp16")]; + string _SplitHeadsQ__mh_w_1785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1785_equation_0, values = (var_17710_cast_fp16, var_17224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1785_cast_fp16")]; + string _SplitHeadsQ__mh_w_1787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1787_equation_0, values = (var_17710_cast_fp16, var_17231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1787_cast_fp16")]; + string _SplitHeadsQ__mh_w_1789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1789_equation_0, values = (var_17710_cast_fp16, var_17238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1789_cast_fp16")]; + string _SplitHeadsQ__mh_w_1791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1791_equation_0, values = (var_17710_cast_fp16, var_17245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1791_cast_fp16")]; + string _SplitHeadsQ__mh_w_1793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1793_equation_0, values = (var_17714_cast_fp16, var_17252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1793_cast_fp16")]; + string _SplitHeadsQ__mh_w_1795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1795_equation_0, values = (var_17714_cast_fp16, var_17259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1795_cast_fp16")]; + string _SplitHeadsQ__mh_w_1797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1797_equation_0, values = (var_17714_cast_fp16, var_17266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1797_cast_fp16")]; + string _SplitHeadsQ__mh_w_1799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1799_equation_0, values = (var_17714_cast_fp16, var_17273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1799_cast_fp16")]; + string _SplitHeadsQ__mh_w_1801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1801_equation_0, values = (var_17718_cast_fp16, var_17280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1801_cast_fp16")]; + string _SplitHeadsQ__mh_w_1803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1803_equation_0, values = (var_17718_cast_fp16, var_17287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1803_cast_fp16")]; + string _SplitHeadsQ__mh_w_1805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1805_equation_0, values = (var_17718_cast_fp16, var_17294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1805_cast_fp16")]; + string _SplitHeadsQ__mh_w_1807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1807_equation_0, values = (var_17718_cast_fp16, var_17301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1807_cast_fp16")]; + string _SplitHeadsQ__mh_w_1809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1809_equation_0, values = (var_17722_cast_fp16, var_17308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1809_cast_fp16")]; + string _SplitHeadsQ__mh_w_1811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1811_equation_0, values = (var_17722_cast_fp16, var_17315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1811_cast_fp16")]; + string _SplitHeadsQ__mh_w_1813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1813_equation_0, values = (var_17722_cast_fp16, var_17322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1813_cast_fp16")]; + string _SplitHeadsQ__mh_w_1815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1815_equation_0, values = (var_17722_cast_fp16, var_17329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1815_cast_fp16")]; + string _SplitHeadsQ__mh_w_1817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1817_equation_0, values = (var_17726_cast_fp16, var_17336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1817_cast_fp16")]; + string _SplitHeadsQ__mh_w_1819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1819_equation_0, values = (var_17726_cast_fp16, var_17343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1819_cast_fp16")]; + string _SplitHeadsQ__mh_w_1821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1821_equation_0, values = (var_17726_cast_fp16, var_17350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1821_cast_fp16")]; + string _SplitHeadsQ__mh_w_1823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1823_equation_0, values = (var_17726_cast_fp16, var_17357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1823_cast_fp16")]; + string _SplitHeadsQ__mh_w_1825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1825_equation_0, values = (var_17730_cast_fp16, var_17364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1825_cast_fp16")]; + string _SplitHeadsQ__mh_w_1827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1827_equation_0, values = (var_17730_cast_fp16, var_17371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1827_cast_fp16")]; + string _SplitHeadsQ__mh_w_1829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1829_equation_0, values = (var_17730_cast_fp16, var_17378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1829_cast_fp16")]; + string _SplitHeadsQ__mh_w_1831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1831_equation_0, values = (var_17730_cast_fp16, var_17385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1831_cast_fp16")]; + string _SplitHeadsQ__mh_w_1833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1833_equation_0, values = (var_17734_cast_fp16, var_17392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1833_cast_fp16")]; + string _SplitHeadsQ__mh_w_1835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1835_equation_0, values = (var_17734_cast_fp16, var_17399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1835_cast_fp16")]; + string _SplitHeadsQ__mh_w_1837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1837_equation_0, values = (var_17734_cast_fp16, var_17406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1837_cast_fp16")]; + string _SplitHeadsQ__mh_w_1839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1839_equation_0, values = (var_17734_cast_fp16, var_17413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1839_cast_fp16")]; + string _SplitHeadsQ__mh_w_1841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1841_equation_0, values = (var_17738_cast_fp16, var_17420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1841_cast_fp16")]; + string _SplitHeadsQ__mh_w_1843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1843_equation_0, values = (var_17738_cast_fp16, var_17427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1843_cast_fp16")]; + string _SplitHeadsQ__mh_w_1845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1845_equation_0, values = (var_17738_cast_fp16, var_17434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1845_cast_fp16")]; + string _SplitHeadsQ__mh_w_1847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1847_equation_0, values = (var_17738_cast_fp16, var_17441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1847_cast_fp16")]; + string _SplitHeadsQ__mh_w_1849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1849_equation_0, values = (var_17742_cast_fp16, var_17448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1849_cast_fp16")]; + string _SplitHeadsQ__mh_w_1851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1851_equation_0, values = (var_17742_cast_fp16, var_17455_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1851_cast_fp16")]; + string _SplitHeadsQ__mh_w_1853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1853_equation_0, values = (var_17742_cast_fp16, var_17462_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1853_cast_fp16")]; + string _SplitHeadsQ__mh_w_1855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1855_equation_0, values = (var_17742_cast_fp16, var_17469_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1855_cast_fp16")]; + string _SplitHeadsQ__mh_w_1857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1857_equation_0, values = (var_17746_cast_fp16, var_17476_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1857_cast_fp16")]; + string _SplitHeadsQ__mh_w_1859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1859_equation_0, values = (var_17746_cast_fp16, var_17483_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1859_cast_fp16")]; + string _SplitHeadsQ__mh_w_1861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1861_equation_0, values = (var_17746_cast_fp16, var_17490_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1861_cast_fp16")]; + string _SplitHeadsQ__mh_w_1863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1863_equation_0, values = (var_17746_cast_fp16, var_17497_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1863_cast_fp16")]; + string _SplitHeadsQ__mh_w_1865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1865_equation_0, values = (var_17750_cast_fp16, var_17504_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1865_cast_fp16")]; + string _SplitHeadsQ__mh_w_1867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1867_equation_0, values = (var_17750_cast_fp16, var_17511_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1867_cast_fp16")]; + string _SplitHeadsQ__mh_w_1869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1869_equation_0, values = (var_17750_cast_fp16, var_17518_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1869_cast_fp16")]; + string _SplitHeadsQ__mh_w_1871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1871_equation_0, values = (var_17750_cast_fp16, var_17525_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1871_cast_fp16")]; + string _SplitHeadsQ__mh_w_1873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1873_equation_0, values = (var_17754_cast_fp16, var_17532_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1873_cast_fp16")]; + string _SplitHeadsQ__mh_w_1875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1875_equation_0, values = (var_17754_cast_fp16, var_17539_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1875_cast_fp16")]; + string _SplitHeadsQ__mh_w_1877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1877_equation_0, values = (var_17754_cast_fp16, var_17546_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1877_cast_fp16")]; + string _SplitHeadsQ__mh_w_1879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1879_equation_0, values = (var_17754_cast_fp16, var_17553_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1879_cast_fp16")]; + string _SplitHeadsQ__mh_w_1881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1881_equation_0, values = (var_17758_cast_fp16, var_17560_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1881_cast_fp16")]; + string _SplitHeadsQ__mh_w_1883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1883_equation_0, values = (var_17758_cast_fp16, var_17567_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1883_cast_fp16")]; + string _SplitHeadsQ__mh_w_1885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1885_equation_0, values = (var_17758_cast_fp16, var_17574_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1885_cast_fp16")]; + string _SplitHeadsQ__mh_w_1887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1887_equation_0, values = (var_17758_cast_fp16, var_17581_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1887_cast_fp16")]; + string _SplitHeadsQ__mh_w_1889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1889_equation_0, values = (var_17762_cast_fp16, var_17588_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1889_cast_fp16")]; + string _SplitHeadsQ__mh_w_1891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1891_equation_0, values = (var_17762_cast_fp16, var_17595_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1891_cast_fp16")]; + string _SplitHeadsQ__mh_w_1893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1893_equation_0, values = (var_17762_cast_fp16, var_17602_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1893_cast_fp16")]; + string _SplitHeadsQ__mh_w_1895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1895_equation_0, values = (var_17762_cast_fp16, var_17609_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1895_cast_fp16")]; + string _SplitHeadsQ__mh_w_1897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1897_equation_0, values = (var_17766_cast_fp16, var_17616_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1897_cast_fp16")]; + string _SplitHeadsQ__mh_w_1899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1899_equation_0, values = (var_17766_cast_fp16, var_17623_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1899_cast_fp16")]; + string _SplitHeadsQ__mh_w_1901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1901_equation_0, values = (var_17766_cast_fp16, var_17630_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1901_cast_fp16")]; + string _SplitHeadsQ__mh_w_1903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1903_equation_0, values = (var_17766_cast_fp16, var_17637_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1903_cast_fp16")]; + string _SplitHeadsQ__mh_w_1905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1905_equation_0, values = (var_17770_cast_fp16, var_17644_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1905_cast_fp16")]; + string _SplitHeadsQ__mh_w_1907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1907_equation_0, values = (var_17770_cast_fp16, var_17651_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1907_cast_fp16")]; + string _SplitHeadsQ__mh_w_1909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1909_equation_0, values = (var_17770_cast_fp16, var_17658_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1909_cast_fp16")]; + string _SplitHeadsQ__mh_w_1911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1911_equation_0, values = (var_17770_cast_fp16, var_17665_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1911_cast_fp16")]; + string _SplitHeadsQ__mh_w_1913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1913_equation_0, values = (var_17774_cast_fp16, var_17672_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1913_cast_fp16")]; + string _SplitHeadsQ__mh_w_1915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1915_equation_0, values = (var_17774_cast_fp16, var_17679_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1915_cast_fp16")]; + string _SplitHeadsQ__mh_w_1917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1917_equation_0, values = (var_17774_cast_fp16, var_17686_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1917_cast_fp16")]; + string _SplitHeadsQ__mh_w_1919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1919_equation_0, values = (var_17774_cast_fp16, var_17693_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1919_cast_fp16")]; + fp16 var_18015_to_fp16 = const()[name = string("op_18015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1761_cast_fp16, y = var_18015_to_fp16)[name = string("aw_chunk_1761_cast_fp16")]; + fp16 var_18017_to_fp16 = const()[name = string("op_18017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1763_cast_fp16, y = var_18017_to_fp16)[name = string("aw_chunk_1763_cast_fp16")]; + fp16 var_18019_to_fp16 = const()[name = string("op_18019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1765_cast_fp16, y = var_18019_to_fp16)[name = string("aw_chunk_1765_cast_fp16")]; + fp16 var_18021_to_fp16 = const()[name = string("op_18021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1767_cast_fp16, y = var_18021_to_fp16)[name = string("aw_chunk_1767_cast_fp16")]; + fp16 var_18023_to_fp16 = const()[name = string("op_18023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1769_cast_fp16, y = var_18023_to_fp16)[name = string("aw_chunk_1769_cast_fp16")]; + fp16 var_18025_to_fp16 = const()[name = string("op_18025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1771_cast_fp16, y = var_18025_to_fp16)[name = string("aw_chunk_1771_cast_fp16")]; + fp16 var_18027_to_fp16 = const()[name = string("op_18027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1773_cast_fp16, y = var_18027_to_fp16)[name = string("aw_chunk_1773_cast_fp16")]; + fp16 var_18029_to_fp16 = const()[name = string("op_18029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1775_cast_fp16, y = var_18029_to_fp16)[name = string("aw_chunk_1775_cast_fp16")]; + fp16 var_18031_to_fp16 = const()[name = string("op_18031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1777_cast_fp16, y = var_18031_to_fp16)[name = string("aw_chunk_1777_cast_fp16")]; + fp16 var_18033_to_fp16 = const()[name = string("op_18033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1779_cast_fp16, y = var_18033_to_fp16)[name = string("aw_chunk_1779_cast_fp16")]; + fp16 var_18035_to_fp16 = const()[name = string("op_18035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1781_cast_fp16, y = var_18035_to_fp16)[name = string("aw_chunk_1781_cast_fp16")]; + fp16 var_18037_to_fp16 = const()[name = string("op_18037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1783_cast_fp16, y = var_18037_to_fp16)[name = string("aw_chunk_1783_cast_fp16")]; + fp16 var_18039_to_fp16 = const()[name = string("op_18039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1785_cast_fp16, y = var_18039_to_fp16)[name = string("aw_chunk_1785_cast_fp16")]; + fp16 var_18041_to_fp16 = const()[name = string("op_18041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1787_cast_fp16, y = var_18041_to_fp16)[name = string("aw_chunk_1787_cast_fp16")]; + fp16 var_18043_to_fp16 = const()[name = string("op_18043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1789_cast_fp16, y = var_18043_to_fp16)[name = string("aw_chunk_1789_cast_fp16")]; + fp16 var_18045_to_fp16 = const()[name = string("op_18045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1791_cast_fp16, y = var_18045_to_fp16)[name = string("aw_chunk_1791_cast_fp16")]; + fp16 var_18047_to_fp16 = const()[name = string("op_18047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1793_cast_fp16, y = var_18047_to_fp16)[name = string("aw_chunk_1793_cast_fp16")]; + fp16 var_18049_to_fp16 = const()[name = string("op_18049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1795_cast_fp16, y = var_18049_to_fp16)[name = string("aw_chunk_1795_cast_fp16")]; + fp16 var_18051_to_fp16 = const()[name = string("op_18051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1797_cast_fp16, y = var_18051_to_fp16)[name = string("aw_chunk_1797_cast_fp16")]; + fp16 var_18053_to_fp16 = const()[name = string("op_18053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1799_cast_fp16, y = var_18053_to_fp16)[name = string("aw_chunk_1799_cast_fp16")]; + fp16 var_18055_to_fp16 = const()[name = string("op_18055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1801_cast_fp16, y = var_18055_to_fp16)[name = string("aw_chunk_1801_cast_fp16")]; + fp16 var_18057_to_fp16 = const()[name = string("op_18057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1803_cast_fp16, y = var_18057_to_fp16)[name = string("aw_chunk_1803_cast_fp16")]; + fp16 var_18059_to_fp16 = const()[name = string("op_18059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1805_cast_fp16, y = var_18059_to_fp16)[name = string("aw_chunk_1805_cast_fp16")]; + fp16 var_18061_to_fp16 = const()[name = string("op_18061_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1807_cast_fp16, y = var_18061_to_fp16)[name = string("aw_chunk_1807_cast_fp16")]; + fp16 var_18063_to_fp16 = const()[name = string("op_18063_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1809_cast_fp16, y = var_18063_to_fp16)[name = string("aw_chunk_1809_cast_fp16")]; + fp16 var_18065_to_fp16 = const()[name = string("op_18065_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1811_cast_fp16, y = var_18065_to_fp16)[name = string("aw_chunk_1811_cast_fp16")]; + fp16 var_18067_to_fp16 = const()[name = string("op_18067_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1813_cast_fp16, y = var_18067_to_fp16)[name = string("aw_chunk_1813_cast_fp16")]; + fp16 var_18069_to_fp16 = const()[name = string("op_18069_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1815_cast_fp16, y = var_18069_to_fp16)[name = string("aw_chunk_1815_cast_fp16")]; + fp16 var_18071_to_fp16 = const()[name = string("op_18071_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1817_cast_fp16, y = var_18071_to_fp16)[name = string("aw_chunk_1817_cast_fp16")]; + fp16 var_18073_to_fp16 = const()[name = string("op_18073_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1819_cast_fp16, y = var_18073_to_fp16)[name = string("aw_chunk_1819_cast_fp16")]; + fp16 var_18075_to_fp16 = const()[name = string("op_18075_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1821_cast_fp16, y = var_18075_to_fp16)[name = string("aw_chunk_1821_cast_fp16")]; + fp16 var_18077_to_fp16 = const()[name = string("op_18077_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1823_cast_fp16, y = var_18077_to_fp16)[name = string("aw_chunk_1823_cast_fp16")]; + fp16 var_18079_to_fp16 = const()[name = string("op_18079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1825_cast_fp16, y = var_18079_to_fp16)[name = string("aw_chunk_1825_cast_fp16")]; + fp16 var_18081_to_fp16 = const()[name = string("op_18081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1827_cast_fp16, y = var_18081_to_fp16)[name = string("aw_chunk_1827_cast_fp16")]; + fp16 var_18083_to_fp16 = const()[name = string("op_18083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1829_cast_fp16, y = var_18083_to_fp16)[name = string("aw_chunk_1829_cast_fp16")]; + fp16 var_18085_to_fp16 = const()[name = string("op_18085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1831_cast_fp16, y = var_18085_to_fp16)[name = string("aw_chunk_1831_cast_fp16")]; + fp16 var_18087_to_fp16 = const()[name = string("op_18087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1833_cast_fp16, y = var_18087_to_fp16)[name = string("aw_chunk_1833_cast_fp16")]; + fp16 var_18089_to_fp16 = const()[name = string("op_18089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1835_cast_fp16, y = var_18089_to_fp16)[name = string("aw_chunk_1835_cast_fp16")]; + fp16 var_18091_to_fp16 = const()[name = string("op_18091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1837_cast_fp16, y = var_18091_to_fp16)[name = string("aw_chunk_1837_cast_fp16")]; + fp16 var_18093_to_fp16 = const()[name = string("op_18093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1839_cast_fp16, y = var_18093_to_fp16)[name = string("aw_chunk_1839_cast_fp16")]; + fp16 var_18095_to_fp16 = const()[name = string("op_18095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1841_cast_fp16, y = var_18095_to_fp16)[name = string("aw_chunk_1841_cast_fp16")]; + fp16 var_18097_to_fp16 = const()[name = string("op_18097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1843_cast_fp16, y = var_18097_to_fp16)[name = string("aw_chunk_1843_cast_fp16")]; + fp16 var_18099_to_fp16 = const()[name = string("op_18099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1845_cast_fp16, y = var_18099_to_fp16)[name = string("aw_chunk_1845_cast_fp16")]; + fp16 var_18101_to_fp16 = const()[name = string("op_18101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1847_cast_fp16, y = var_18101_to_fp16)[name = string("aw_chunk_1847_cast_fp16")]; + fp16 var_18103_to_fp16 = const()[name = string("op_18103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1849_cast_fp16, y = var_18103_to_fp16)[name = string("aw_chunk_1849_cast_fp16")]; + fp16 var_18105_to_fp16 = const()[name = string("op_18105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1851_cast_fp16, y = var_18105_to_fp16)[name = string("aw_chunk_1851_cast_fp16")]; + fp16 var_18107_to_fp16 = const()[name = string("op_18107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1853_cast_fp16, y = var_18107_to_fp16)[name = string("aw_chunk_1853_cast_fp16")]; + fp16 var_18109_to_fp16 = const()[name = string("op_18109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1855_cast_fp16, y = var_18109_to_fp16)[name = string("aw_chunk_1855_cast_fp16")]; + fp16 var_18111_to_fp16 = const()[name = string("op_18111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1857_cast_fp16, y = var_18111_to_fp16)[name = string("aw_chunk_1857_cast_fp16")]; + fp16 var_18113_to_fp16 = const()[name = string("op_18113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1859_cast_fp16, y = var_18113_to_fp16)[name = string("aw_chunk_1859_cast_fp16")]; + fp16 var_18115_to_fp16 = const()[name = string("op_18115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1861_cast_fp16, y = var_18115_to_fp16)[name = string("aw_chunk_1861_cast_fp16")]; + fp16 var_18117_to_fp16 = const()[name = string("op_18117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1863_cast_fp16, y = var_18117_to_fp16)[name = string("aw_chunk_1863_cast_fp16")]; + fp16 var_18119_to_fp16 = const()[name = string("op_18119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1865_cast_fp16, y = var_18119_to_fp16)[name = string("aw_chunk_1865_cast_fp16")]; + fp16 var_18121_to_fp16 = const()[name = string("op_18121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1867_cast_fp16, y = var_18121_to_fp16)[name = string("aw_chunk_1867_cast_fp16")]; + fp16 var_18123_to_fp16 = const()[name = string("op_18123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1869_cast_fp16, y = var_18123_to_fp16)[name = string("aw_chunk_1869_cast_fp16")]; + fp16 var_18125_to_fp16 = const()[name = string("op_18125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1871_cast_fp16, y = var_18125_to_fp16)[name = string("aw_chunk_1871_cast_fp16")]; + fp16 var_18127_to_fp16 = const()[name = string("op_18127_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1873_cast_fp16, y = var_18127_to_fp16)[name = string("aw_chunk_1873_cast_fp16")]; + fp16 var_18129_to_fp16 = const()[name = string("op_18129_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1875_cast_fp16, y = var_18129_to_fp16)[name = string("aw_chunk_1875_cast_fp16")]; + fp16 var_18131_to_fp16 = const()[name = string("op_18131_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1877_cast_fp16, y = var_18131_to_fp16)[name = string("aw_chunk_1877_cast_fp16")]; + fp16 var_18133_to_fp16 = const()[name = string("op_18133_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1879_cast_fp16, y = var_18133_to_fp16)[name = string("aw_chunk_1879_cast_fp16")]; + fp16 var_18135_to_fp16 = const()[name = string("op_18135_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1881_cast_fp16, y = var_18135_to_fp16)[name = string("aw_chunk_1881_cast_fp16")]; + fp16 var_18137_to_fp16 = const()[name = string("op_18137_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1883_cast_fp16, y = var_18137_to_fp16)[name = string("aw_chunk_1883_cast_fp16")]; + fp16 var_18139_to_fp16 = const()[name = string("op_18139_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1885_cast_fp16, y = var_18139_to_fp16)[name = string("aw_chunk_1885_cast_fp16")]; + fp16 var_18141_to_fp16 = const()[name = string("op_18141_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1887_cast_fp16, y = var_18141_to_fp16)[name = string("aw_chunk_1887_cast_fp16")]; + fp16 var_18143_to_fp16 = const()[name = string("op_18143_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1889_cast_fp16, y = var_18143_to_fp16)[name = string("aw_chunk_1889_cast_fp16")]; + fp16 var_18145_to_fp16 = const()[name = string("op_18145_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1891_cast_fp16, y = var_18145_to_fp16)[name = string("aw_chunk_1891_cast_fp16")]; + fp16 var_18147_to_fp16 = const()[name = string("op_18147_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1893_cast_fp16, y = var_18147_to_fp16)[name = string("aw_chunk_1893_cast_fp16")]; + fp16 var_18149_to_fp16 = const()[name = string("op_18149_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1895_cast_fp16, y = var_18149_to_fp16)[name = string("aw_chunk_1895_cast_fp16")]; + fp16 var_18151_to_fp16 = const()[name = string("op_18151_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1897_cast_fp16, y = var_18151_to_fp16)[name = string("aw_chunk_1897_cast_fp16")]; + fp16 var_18153_to_fp16 = const()[name = string("op_18153_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1899_cast_fp16, y = var_18153_to_fp16)[name = string("aw_chunk_1899_cast_fp16")]; + fp16 var_18155_to_fp16 = const()[name = string("op_18155_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1901_cast_fp16, y = var_18155_to_fp16)[name = string("aw_chunk_1901_cast_fp16")]; + fp16 var_18157_to_fp16 = const()[name = string("op_18157_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1903_cast_fp16, y = var_18157_to_fp16)[name = string("aw_chunk_1903_cast_fp16")]; + fp16 var_18159_to_fp16 = const()[name = string("op_18159_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1905_cast_fp16, y = var_18159_to_fp16)[name = string("aw_chunk_1905_cast_fp16")]; + fp16 var_18161_to_fp16 = const()[name = string("op_18161_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1907_cast_fp16, y = var_18161_to_fp16)[name = string("aw_chunk_1907_cast_fp16")]; + fp16 var_18163_to_fp16 = const()[name = string("op_18163_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1909_cast_fp16, y = var_18163_to_fp16)[name = string("aw_chunk_1909_cast_fp16")]; + fp16 var_18165_to_fp16 = const()[name = string("op_18165_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1911_cast_fp16, y = var_18165_to_fp16)[name = string("aw_chunk_1911_cast_fp16")]; + fp16 var_18167_to_fp16 = const()[name = string("op_18167_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1913_cast_fp16, y = var_18167_to_fp16)[name = string("aw_chunk_1913_cast_fp16")]; + fp16 var_18169_to_fp16 = const()[name = string("op_18169_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1915_cast_fp16, y = var_18169_to_fp16)[name = string("aw_chunk_1915_cast_fp16")]; + fp16 var_18171_to_fp16 = const()[name = string("op_18171_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1917_cast_fp16, y = var_18171_to_fp16)[name = string("aw_chunk_1917_cast_fp16")]; + fp16 var_18173_to_fp16 = const()[name = string("op_18173_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1919_cast_fp16, y = var_18173_to_fp16)[name = string("aw_chunk_1919_cast_fp16")]; + tensor var_18175_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1761_cast_fp16)[name = string("op_18175_cast_fp16")]; + tensor var_18176_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1763_cast_fp16)[name = string("op_18176_cast_fp16")]; + tensor var_18177_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1765_cast_fp16)[name = string("op_18177_cast_fp16")]; + tensor var_18178_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1767_cast_fp16)[name = string("op_18178_cast_fp16")]; + tensor var_18179_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1769_cast_fp16)[name = string("op_18179_cast_fp16")]; + tensor var_18180_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1771_cast_fp16)[name = string("op_18180_cast_fp16")]; + tensor var_18181_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1773_cast_fp16)[name = string("op_18181_cast_fp16")]; + tensor var_18182_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1775_cast_fp16)[name = string("op_18182_cast_fp16")]; + tensor var_18183_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1777_cast_fp16)[name = string("op_18183_cast_fp16")]; + tensor var_18184_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1779_cast_fp16)[name = string("op_18184_cast_fp16")]; + tensor var_18185_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1781_cast_fp16)[name = string("op_18185_cast_fp16")]; + tensor var_18186_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1783_cast_fp16)[name = string("op_18186_cast_fp16")]; + tensor var_18187_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1785_cast_fp16)[name = string("op_18187_cast_fp16")]; + tensor var_18188_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1787_cast_fp16)[name = string("op_18188_cast_fp16")]; + tensor var_18189_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1789_cast_fp16)[name = string("op_18189_cast_fp16")]; + tensor var_18190_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1791_cast_fp16)[name = string("op_18190_cast_fp16")]; + tensor var_18191_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1793_cast_fp16)[name = string("op_18191_cast_fp16")]; + tensor var_18192_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1795_cast_fp16)[name = string("op_18192_cast_fp16")]; + tensor var_18193_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1797_cast_fp16)[name = string("op_18193_cast_fp16")]; + tensor var_18194_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1799_cast_fp16)[name = string("op_18194_cast_fp16")]; + tensor var_18195_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1801_cast_fp16)[name = string("op_18195_cast_fp16")]; + tensor var_18196_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1803_cast_fp16)[name = string("op_18196_cast_fp16")]; + tensor var_18197_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1805_cast_fp16)[name = string("op_18197_cast_fp16")]; + tensor var_18198_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1807_cast_fp16)[name = string("op_18198_cast_fp16")]; + tensor var_18199_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1809_cast_fp16)[name = string("op_18199_cast_fp16")]; + tensor var_18200_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1811_cast_fp16)[name = string("op_18200_cast_fp16")]; + tensor var_18201_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1813_cast_fp16)[name = string("op_18201_cast_fp16")]; + tensor var_18202_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1815_cast_fp16)[name = string("op_18202_cast_fp16")]; + tensor var_18203_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1817_cast_fp16)[name = string("op_18203_cast_fp16")]; + tensor var_18204_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1819_cast_fp16)[name = string("op_18204_cast_fp16")]; + tensor var_18205_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1821_cast_fp16)[name = string("op_18205_cast_fp16")]; + tensor var_18206_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1823_cast_fp16)[name = string("op_18206_cast_fp16")]; + tensor var_18207_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1825_cast_fp16)[name = string("op_18207_cast_fp16")]; + tensor var_18208_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1827_cast_fp16)[name = string("op_18208_cast_fp16")]; + tensor var_18209_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1829_cast_fp16)[name = string("op_18209_cast_fp16")]; + tensor var_18210_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1831_cast_fp16)[name = string("op_18210_cast_fp16")]; + tensor var_18211_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1833_cast_fp16)[name = string("op_18211_cast_fp16")]; + tensor var_18212_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1835_cast_fp16)[name = string("op_18212_cast_fp16")]; + tensor var_18213_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1837_cast_fp16)[name = string("op_18213_cast_fp16")]; + tensor var_18214_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1839_cast_fp16)[name = string("op_18214_cast_fp16")]; + tensor var_18215_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1841_cast_fp16)[name = string("op_18215_cast_fp16")]; + tensor var_18216_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1843_cast_fp16)[name = string("op_18216_cast_fp16")]; + tensor var_18217_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1845_cast_fp16)[name = string("op_18217_cast_fp16")]; + tensor var_18218_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1847_cast_fp16)[name = string("op_18218_cast_fp16")]; + tensor var_18219_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1849_cast_fp16)[name = string("op_18219_cast_fp16")]; + tensor var_18220_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1851_cast_fp16)[name = string("op_18220_cast_fp16")]; + tensor var_18221_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1853_cast_fp16)[name = string("op_18221_cast_fp16")]; + tensor var_18222_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1855_cast_fp16)[name = string("op_18222_cast_fp16")]; + tensor var_18223_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1857_cast_fp16)[name = string("op_18223_cast_fp16")]; + tensor var_18224_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1859_cast_fp16)[name = string("op_18224_cast_fp16")]; + tensor var_18225_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1861_cast_fp16)[name = string("op_18225_cast_fp16")]; + tensor var_18226_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1863_cast_fp16)[name = string("op_18226_cast_fp16")]; + tensor var_18227_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1865_cast_fp16)[name = string("op_18227_cast_fp16")]; + tensor var_18228_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1867_cast_fp16)[name = string("op_18228_cast_fp16")]; + tensor var_18229_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1869_cast_fp16)[name = string("op_18229_cast_fp16")]; + tensor var_18230_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1871_cast_fp16)[name = string("op_18230_cast_fp16")]; + tensor var_18231_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1873_cast_fp16)[name = string("op_18231_cast_fp16")]; + tensor var_18232_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1875_cast_fp16)[name = string("op_18232_cast_fp16")]; + tensor var_18233_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1877_cast_fp16)[name = string("op_18233_cast_fp16")]; + tensor var_18234_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1879_cast_fp16)[name = string("op_18234_cast_fp16")]; + tensor var_18235_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1881_cast_fp16)[name = string("op_18235_cast_fp16")]; + tensor var_18236_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1883_cast_fp16)[name = string("op_18236_cast_fp16")]; + tensor var_18237_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1885_cast_fp16)[name = string("op_18237_cast_fp16")]; + tensor var_18238_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1887_cast_fp16)[name = string("op_18238_cast_fp16")]; + tensor var_18239_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1889_cast_fp16)[name = string("op_18239_cast_fp16")]; + tensor var_18240_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1891_cast_fp16)[name = string("op_18240_cast_fp16")]; + tensor var_18241_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1893_cast_fp16)[name = string("op_18241_cast_fp16")]; + tensor var_18242_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1895_cast_fp16)[name = string("op_18242_cast_fp16")]; + tensor var_18243_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1897_cast_fp16)[name = string("op_18243_cast_fp16")]; + tensor var_18244_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1899_cast_fp16)[name = string("op_18244_cast_fp16")]; + tensor var_18245_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1901_cast_fp16)[name = string("op_18245_cast_fp16")]; + tensor var_18246_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1903_cast_fp16)[name = string("op_18246_cast_fp16")]; + tensor var_18247_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1905_cast_fp16)[name = string("op_18247_cast_fp16")]; + tensor var_18248_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1907_cast_fp16)[name = string("op_18248_cast_fp16")]; + tensor var_18249_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1909_cast_fp16)[name = string("op_18249_cast_fp16")]; + tensor var_18250_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1911_cast_fp16)[name = string("op_18250_cast_fp16")]; + tensor var_18251_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1913_cast_fp16)[name = string("op_18251_cast_fp16")]; + tensor var_18252_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1915_cast_fp16)[name = string("op_18252_cast_fp16")]; + tensor var_18253_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1917_cast_fp16)[name = string("op_18253_cast_fp16")]; + tensor var_18254_cast_fp16 = softmax(axis = var_17000, x = aw_chunk_1919_cast_fp16)[name = string("op_18254_cast_fp16")]; + string var_18256_equation_0 = const()[name = string("op_18256_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18256_cast_fp16 = einsum(equation = var_18256_equation_0, values = (var_17776_cast_fp16, var_18175_cast_fp16))[name = string("op_18256_cast_fp16")]; + string var_18258_equation_0 = const()[name = string("op_18258_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18258_cast_fp16 = einsum(equation = var_18258_equation_0, values = (var_17776_cast_fp16, var_18176_cast_fp16))[name = string("op_18258_cast_fp16")]; + string var_18260_equation_0 = const()[name = string("op_18260_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18260_cast_fp16 = einsum(equation = var_18260_equation_0, values = (var_17776_cast_fp16, var_18177_cast_fp16))[name = string("op_18260_cast_fp16")]; + string var_18262_equation_0 = const()[name = string("op_18262_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18262_cast_fp16 = einsum(equation = var_18262_equation_0, values = (var_17776_cast_fp16, var_18178_cast_fp16))[name = string("op_18262_cast_fp16")]; + string var_18264_equation_0 = const()[name = string("op_18264_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18264_cast_fp16 = einsum(equation = var_18264_equation_0, values = (var_17780_cast_fp16, var_18179_cast_fp16))[name = string("op_18264_cast_fp16")]; + string var_18266_equation_0 = const()[name = string("op_18266_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18266_cast_fp16 = einsum(equation = var_18266_equation_0, values = (var_17780_cast_fp16, var_18180_cast_fp16))[name = string("op_18266_cast_fp16")]; + string var_18268_equation_0 = const()[name = string("op_18268_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18268_cast_fp16 = einsum(equation = var_18268_equation_0, values = (var_17780_cast_fp16, var_18181_cast_fp16))[name = string("op_18268_cast_fp16")]; + string var_18270_equation_0 = const()[name = string("op_18270_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18270_cast_fp16 = einsum(equation = var_18270_equation_0, values = (var_17780_cast_fp16, var_18182_cast_fp16))[name = string("op_18270_cast_fp16")]; + string var_18272_equation_0 = const()[name = string("op_18272_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18272_cast_fp16 = einsum(equation = var_18272_equation_0, values = (var_17784_cast_fp16, var_18183_cast_fp16))[name = string("op_18272_cast_fp16")]; + string var_18274_equation_0 = const()[name = string("op_18274_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18274_cast_fp16 = einsum(equation = var_18274_equation_0, values = (var_17784_cast_fp16, var_18184_cast_fp16))[name = string("op_18274_cast_fp16")]; + string var_18276_equation_0 = const()[name = string("op_18276_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18276_cast_fp16 = einsum(equation = var_18276_equation_0, values = (var_17784_cast_fp16, var_18185_cast_fp16))[name = string("op_18276_cast_fp16")]; + string var_18278_equation_0 = const()[name = string("op_18278_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18278_cast_fp16 = einsum(equation = var_18278_equation_0, values = (var_17784_cast_fp16, var_18186_cast_fp16))[name = string("op_18278_cast_fp16")]; + string var_18280_equation_0 = const()[name = string("op_18280_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18280_cast_fp16 = einsum(equation = var_18280_equation_0, values = (var_17788_cast_fp16, var_18187_cast_fp16))[name = string("op_18280_cast_fp16")]; + string var_18282_equation_0 = const()[name = string("op_18282_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18282_cast_fp16 = einsum(equation = var_18282_equation_0, values = (var_17788_cast_fp16, var_18188_cast_fp16))[name = string("op_18282_cast_fp16")]; + string var_18284_equation_0 = const()[name = string("op_18284_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18284_cast_fp16 = einsum(equation = var_18284_equation_0, values = (var_17788_cast_fp16, var_18189_cast_fp16))[name = string("op_18284_cast_fp16")]; + string var_18286_equation_0 = const()[name = string("op_18286_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18286_cast_fp16 = einsum(equation = var_18286_equation_0, values = (var_17788_cast_fp16, var_18190_cast_fp16))[name = string("op_18286_cast_fp16")]; + string var_18288_equation_0 = const()[name = string("op_18288_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18288_cast_fp16 = einsum(equation = var_18288_equation_0, values = (var_17792_cast_fp16, var_18191_cast_fp16))[name = string("op_18288_cast_fp16")]; + string var_18290_equation_0 = const()[name = string("op_18290_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18290_cast_fp16 = einsum(equation = var_18290_equation_0, values = (var_17792_cast_fp16, var_18192_cast_fp16))[name = string("op_18290_cast_fp16")]; + string var_18292_equation_0 = const()[name = string("op_18292_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18292_cast_fp16 = einsum(equation = var_18292_equation_0, values = (var_17792_cast_fp16, var_18193_cast_fp16))[name = string("op_18292_cast_fp16")]; + string var_18294_equation_0 = const()[name = string("op_18294_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18294_cast_fp16 = einsum(equation = var_18294_equation_0, values = (var_17792_cast_fp16, var_18194_cast_fp16))[name = string("op_18294_cast_fp16")]; + string var_18296_equation_0 = const()[name = string("op_18296_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18296_cast_fp16 = einsum(equation = var_18296_equation_0, values = (var_17796_cast_fp16, var_18195_cast_fp16))[name = string("op_18296_cast_fp16")]; + string var_18298_equation_0 = const()[name = string("op_18298_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18298_cast_fp16 = einsum(equation = var_18298_equation_0, values = (var_17796_cast_fp16, var_18196_cast_fp16))[name = string("op_18298_cast_fp16")]; + string var_18300_equation_0 = const()[name = string("op_18300_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18300_cast_fp16 = einsum(equation = var_18300_equation_0, values = (var_17796_cast_fp16, var_18197_cast_fp16))[name = string("op_18300_cast_fp16")]; + string var_18302_equation_0 = const()[name = string("op_18302_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18302_cast_fp16 = einsum(equation = var_18302_equation_0, values = (var_17796_cast_fp16, var_18198_cast_fp16))[name = string("op_18302_cast_fp16")]; + string var_18304_equation_0 = const()[name = string("op_18304_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18304_cast_fp16 = einsum(equation = var_18304_equation_0, values = (var_17800_cast_fp16, var_18199_cast_fp16))[name = string("op_18304_cast_fp16")]; + string var_18306_equation_0 = const()[name = string("op_18306_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18306_cast_fp16 = einsum(equation = var_18306_equation_0, values = (var_17800_cast_fp16, var_18200_cast_fp16))[name = string("op_18306_cast_fp16")]; + string var_18308_equation_0 = const()[name = string("op_18308_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18308_cast_fp16 = einsum(equation = var_18308_equation_0, values = (var_17800_cast_fp16, var_18201_cast_fp16))[name = string("op_18308_cast_fp16")]; + string var_18310_equation_0 = const()[name = string("op_18310_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18310_cast_fp16 = einsum(equation = var_18310_equation_0, values = (var_17800_cast_fp16, var_18202_cast_fp16))[name = string("op_18310_cast_fp16")]; + string var_18312_equation_0 = const()[name = string("op_18312_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18312_cast_fp16 = einsum(equation = var_18312_equation_0, values = (var_17804_cast_fp16, var_18203_cast_fp16))[name = string("op_18312_cast_fp16")]; + string var_18314_equation_0 = const()[name = string("op_18314_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18314_cast_fp16 = einsum(equation = var_18314_equation_0, values = (var_17804_cast_fp16, var_18204_cast_fp16))[name = string("op_18314_cast_fp16")]; + string var_18316_equation_0 = const()[name = string("op_18316_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18316_cast_fp16 = einsum(equation = var_18316_equation_0, values = (var_17804_cast_fp16, var_18205_cast_fp16))[name = string("op_18316_cast_fp16")]; + string var_18318_equation_0 = const()[name = string("op_18318_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18318_cast_fp16 = einsum(equation = var_18318_equation_0, values = (var_17804_cast_fp16, var_18206_cast_fp16))[name = string("op_18318_cast_fp16")]; + string var_18320_equation_0 = const()[name = string("op_18320_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18320_cast_fp16 = einsum(equation = var_18320_equation_0, values = (var_17808_cast_fp16, var_18207_cast_fp16))[name = string("op_18320_cast_fp16")]; + string var_18322_equation_0 = const()[name = string("op_18322_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18322_cast_fp16 = einsum(equation = var_18322_equation_0, values = (var_17808_cast_fp16, var_18208_cast_fp16))[name = string("op_18322_cast_fp16")]; + string var_18324_equation_0 = const()[name = string("op_18324_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18324_cast_fp16 = einsum(equation = var_18324_equation_0, values = (var_17808_cast_fp16, var_18209_cast_fp16))[name = string("op_18324_cast_fp16")]; + string var_18326_equation_0 = const()[name = string("op_18326_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18326_cast_fp16 = einsum(equation = var_18326_equation_0, values = (var_17808_cast_fp16, var_18210_cast_fp16))[name = string("op_18326_cast_fp16")]; + string var_18328_equation_0 = const()[name = string("op_18328_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18328_cast_fp16 = einsum(equation = var_18328_equation_0, values = (var_17812_cast_fp16, var_18211_cast_fp16))[name = string("op_18328_cast_fp16")]; + string var_18330_equation_0 = const()[name = string("op_18330_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18330_cast_fp16 = einsum(equation = var_18330_equation_0, values = (var_17812_cast_fp16, var_18212_cast_fp16))[name = string("op_18330_cast_fp16")]; + string var_18332_equation_0 = const()[name = string("op_18332_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18332_cast_fp16 = einsum(equation = var_18332_equation_0, values = (var_17812_cast_fp16, var_18213_cast_fp16))[name = string("op_18332_cast_fp16")]; + string var_18334_equation_0 = const()[name = string("op_18334_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18334_cast_fp16 = einsum(equation = var_18334_equation_0, values = (var_17812_cast_fp16, var_18214_cast_fp16))[name = string("op_18334_cast_fp16")]; + string var_18336_equation_0 = const()[name = string("op_18336_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18336_cast_fp16 = einsum(equation = var_18336_equation_0, values = (var_17816_cast_fp16, var_18215_cast_fp16))[name = string("op_18336_cast_fp16")]; + string var_18338_equation_0 = const()[name = string("op_18338_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18338_cast_fp16 = einsum(equation = var_18338_equation_0, values = (var_17816_cast_fp16, var_18216_cast_fp16))[name = string("op_18338_cast_fp16")]; + string var_18340_equation_0 = const()[name = string("op_18340_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18340_cast_fp16 = einsum(equation = var_18340_equation_0, values = (var_17816_cast_fp16, var_18217_cast_fp16))[name = string("op_18340_cast_fp16")]; + string var_18342_equation_0 = const()[name = string("op_18342_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18342_cast_fp16 = einsum(equation = var_18342_equation_0, values = (var_17816_cast_fp16, var_18218_cast_fp16))[name = string("op_18342_cast_fp16")]; + string var_18344_equation_0 = const()[name = string("op_18344_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18344_cast_fp16 = einsum(equation = var_18344_equation_0, values = (var_17820_cast_fp16, var_18219_cast_fp16))[name = string("op_18344_cast_fp16")]; + string var_18346_equation_0 = const()[name = string("op_18346_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18346_cast_fp16 = einsum(equation = var_18346_equation_0, values = (var_17820_cast_fp16, var_18220_cast_fp16))[name = string("op_18346_cast_fp16")]; + string var_18348_equation_0 = const()[name = string("op_18348_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18348_cast_fp16 = einsum(equation = var_18348_equation_0, values = (var_17820_cast_fp16, var_18221_cast_fp16))[name = string("op_18348_cast_fp16")]; + string var_18350_equation_0 = const()[name = string("op_18350_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18350_cast_fp16 = einsum(equation = var_18350_equation_0, values = (var_17820_cast_fp16, var_18222_cast_fp16))[name = string("op_18350_cast_fp16")]; + string var_18352_equation_0 = const()[name = string("op_18352_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18352_cast_fp16 = einsum(equation = var_18352_equation_0, values = (var_17824_cast_fp16, var_18223_cast_fp16))[name = string("op_18352_cast_fp16")]; + string var_18354_equation_0 = const()[name = string("op_18354_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18354_cast_fp16 = einsum(equation = var_18354_equation_0, values = (var_17824_cast_fp16, var_18224_cast_fp16))[name = string("op_18354_cast_fp16")]; + string var_18356_equation_0 = const()[name = string("op_18356_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18356_cast_fp16 = einsum(equation = var_18356_equation_0, values = (var_17824_cast_fp16, var_18225_cast_fp16))[name = string("op_18356_cast_fp16")]; + string var_18358_equation_0 = const()[name = string("op_18358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18358_cast_fp16 = einsum(equation = var_18358_equation_0, values = (var_17824_cast_fp16, var_18226_cast_fp16))[name = string("op_18358_cast_fp16")]; + string var_18360_equation_0 = const()[name = string("op_18360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18360_cast_fp16 = einsum(equation = var_18360_equation_0, values = (var_17828_cast_fp16, var_18227_cast_fp16))[name = string("op_18360_cast_fp16")]; + string var_18362_equation_0 = const()[name = string("op_18362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18362_cast_fp16 = einsum(equation = var_18362_equation_0, values = (var_17828_cast_fp16, var_18228_cast_fp16))[name = string("op_18362_cast_fp16")]; + string var_18364_equation_0 = const()[name = string("op_18364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18364_cast_fp16 = einsum(equation = var_18364_equation_0, values = (var_17828_cast_fp16, var_18229_cast_fp16))[name = string("op_18364_cast_fp16")]; + string var_18366_equation_0 = const()[name = string("op_18366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18366_cast_fp16 = einsum(equation = var_18366_equation_0, values = (var_17828_cast_fp16, var_18230_cast_fp16))[name = string("op_18366_cast_fp16")]; + string var_18368_equation_0 = const()[name = string("op_18368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18368_cast_fp16 = einsum(equation = var_18368_equation_0, values = (var_17832_cast_fp16, var_18231_cast_fp16))[name = string("op_18368_cast_fp16")]; + string var_18370_equation_0 = const()[name = string("op_18370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18370_cast_fp16 = einsum(equation = var_18370_equation_0, values = (var_17832_cast_fp16, var_18232_cast_fp16))[name = string("op_18370_cast_fp16")]; + string var_18372_equation_0 = const()[name = string("op_18372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18372_cast_fp16 = einsum(equation = var_18372_equation_0, values = (var_17832_cast_fp16, var_18233_cast_fp16))[name = string("op_18372_cast_fp16")]; + string var_18374_equation_0 = const()[name = string("op_18374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18374_cast_fp16 = einsum(equation = var_18374_equation_0, values = (var_17832_cast_fp16, var_18234_cast_fp16))[name = string("op_18374_cast_fp16")]; + string var_18376_equation_0 = const()[name = string("op_18376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18376_cast_fp16 = einsum(equation = var_18376_equation_0, values = (var_17836_cast_fp16, var_18235_cast_fp16))[name = string("op_18376_cast_fp16")]; + string var_18378_equation_0 = const()[name = string("op_18378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18378_cast_fp16 = einsum(equation = var_18378_equation_0, values = (var_17836_cast_fp16, var_18236_cast_fp16))[name = string("op_18378_cast_fp16")]; + string var_18380_equation_0 = const()[name = string("op_18380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18380_cast_fp16 = einsum(equation = var_18380_equation_0, values = (var_17836_cast_fp16, var_18237_cast_fp16))[name = string("op_18380_cast_fp16")]; + string var_18382_equation_0 = const()[name = string("op_18382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18382_cast_fp16 = einsum(equation = var_18382_equation_0, values = (var_17836_cast_fp16, var_18238_cast_fp16))[name = string("op_18382_cast_fp16")]; + string var_18384_equation_0 = const()[name = string("op_18384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18384_cast_fp16 = einsum(equation = var_18384_equation_0, values = (var_17840_cast_fp16, var_18239_cast_fp16))[name = string("op_18384_cast_fp16")]; + string var_18386_equation_0 = const()[name = string("op_18386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18386_cast_fp16 = einsum(equation = var_18386_equation_0, values = (var_17840_cast_fp16, var_18240_cast_fp16))[name = string("op_18386_cast_fp16")]; + string var_18388_equation_0 = const()[name = string("op_18388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18388_cast_fp16 = einsum(equation = var_18388_equation_0, values = (var_17840_cast_fp16, var_18241_cast_fp16))[name = string("op_18388_cast_fp16")]; + string var_18390_equation_0 = const()[name = string("op_18390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18390_cast_fp16 = einsum(equation = var_18390_equation_0, values = (var_17840_cast_fp16, var_18242_cast_fp16))[name = string("op_18390_cast_fp16")]; + string var_18392_equation_0 = const()[name = string("op_18392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18392_cast_fp16 = einsum(equation = var_18392_equation_0, values = (var_17844_cast_fp16, var_18243_cast_fp16))[name = string("op_18392_cast_fp16")]; + string var_18394_equation_0 = const()[name = string("op_18394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18394_cast_fp16 = einsum(equation = var_18394_equation_0, values = (var_17844_cast_fp16, var_18244_cast_fp16))[name = string("op_18394_cast_fp16")]; + string var_18396_equation_0 = const()[name = string("op_18396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18396_cast_fp16 = einsum(equation = var_18396_equation_0, values = (var_17844_cast_fp16, var_18245_cast_fp16))[name = string("op_18396_cast_fp16")]; + string var_18398_equation_0 = const()[name = string("op_18398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18398_cast_fp16 = einsum(equation = var_18398_equation_0, values = (var_17844_cast_fp16, var_18246_cast_fp16))[name = string("op_18398_cast_fp16")]; + string var_18400_equation_0 = const()[name = string("op_18400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18400_cast_fp16 = einsum(equation = var_18400_equation_0, values = (var_17848_cast_fp16, var_18247_cast_fp16))[name = string("op_18400_cast_fp16")]; + string var_18402_equation_0 = const()[name = string("op_18402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18402_cast_fp16 = einsum(equation = var_18402_equation_0, values = (var_17848_cast_fp16, var_18248_cast_fp16))[name = string("op_18402_cast_fp16")]; + string var_18404_equation_0 = const()[name = string("op_18404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18404_cast_fp16 = einsum(equation = var_18404_equation_0, values = (var_17848_cast_fp16, var_18249_cast_fp16))[name = string("op_18404_cast_fp16")]; + string var_18406_equation_0 = const()[name = string("op_18406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18406_cast_fp16 = einsum(equation = var_18406_equation_0, values = (var_17848_cast_fp16, var_18250_cast_fp16))[name = string("op_18406_cast_fp16")]; + string var_18408_equation_0 = const()[name = string("op_18408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18408_cast_fp16 = einsum(equation = var_18408_equation_0, values = (var_17852_cast_fp16, var_18251_cast_fp16))[name = string("op_18408_cast_fp16")]; + string var_18410_equation_0 = const()[name = string("op_18410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18410_cast_fp16 = einsum(equation = var_18410_equation_0, values = (var_17852_cast_fp16, var_18252_cast_fp16))[name = string("op_18410_cast_fp16")]; + string var_18412_equation_0 = const()[name = string("op_18412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18412_cast_fp16 = einsum(equation = var_18412_equation_0, values = (var_17852_cast_fp16, var_18253_cast_fp16))[name = string("op_18412_cast_fp16")]; + string var_18414_equation_0 = const()[name = string("op_18414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18414_cast_fp16 = einsum(equation = var_18414_equation_0, values = (var_17852_cast_fp16, var_18254_cast_fp16))[name = string("op_18414_cast_fp16")]; + bool var_18416_interleave_0 = const()[name = string("op_18416_interleave_0"), val = bool(false)]; + tensor var_18416_cast_fp16 = concat(axis = var_16975, interleave = var_18416_interleave_0, values = (var_18256_cast_fp16, var_18258_cast_fp16, var_18260_cast_fp16, var_18262_cast_fp16))[name = string("op_18416_cast_fp16")]; + bool var_18418_interleave_0 = const()[name = string("op_18418_interleave_0"), val = bool(false)]; + tensor var_18418_cast_fp16 = concat(axis = var_16975, interleave = var_18418_interleave_0, values = (var_18264_cast_fp16, var_18266_cast_fp16, var_18268_cast_fp16, var_18270_cast_fp16))[name = string("op_18418_cast_fp16")]; + bool var_18420_interleave_0 = const()[name = string("op_18420_interleave_0"), val = bool(false)]; + tensor var_18420_cast_fp16 = concat(axis = var_16975, interleave = var_18420_interleave_0, values = (var_18272_cast_fp16, var_18274_cast_fp16, var_18276_cast_fp16, var_18278_cast_fp16))[name = string("op_18420_cast_fp16")]; + bool var_18422_interleave_0 = const()[name = string("op_18422_interleave_0"), val = bool(false)]; + tensor var_18422_cast_fp16 = concat(axis = var_16975, interleave = var_18422_interleave_0, values = (var_18280_cast_fp16, var_18282_cast_fp16, var_18284_cast_fp16, var_18286_cast_fp16))[name = string("op_18422_cast_fp16")]; + bool var_18424_interleave_0 = const()[name = string("op_18424_interleave_0"), val = bool(false)]; + tensor var_18424_cast_fp16 = concat(axis = var_16975, interleave = var_18424_interleave_0, values = (var_18288_cast_fp16, var_18290_cast_fp16, var_18292_cast_fp16, var_18294_cast_fp16))[name = string("op_18424_cast_fp16")]; + bool var_18426_interleave_0 = const()[name = string("op_18426_interleave_0"), val = bool(false)]; + tensor var_18426_cast_fp16 = concat(axis = var_16975, interleave = var_18426_interleave_0, values = (var_18296_cast_fp16, var_18298_cast_fp16, var_18300_cast_fp16, var_18302_cast_fp16))[name = string("op_18426_cast_fp16")]; + bool var_18428_interleave_0 = const()[name = string("op_18428_interleave_0"), val = bool(false)]; + tensor var_18428_cast_fp16 = concat(axis = var_16975, interleave = var_18428_interleave_0, values = (var_18304_cast_fp16, var_18306_cast_fp16, var_18308_cast_fp16, var_18310_cast_fp16))[name = string("op_18428_cast_fp16")]; + bool var_18430_interleave_0 = const()[name = string("op_18430_interleave_0"), val = bool(false)]; + tensor var_18430_cast_fp16 = concat(axis = var_16975, interleave = var_18430_interleave_0, values = (var_18312_cast_fp16, var_18314_cast_fp16, var_18316_cast_fp16, var_18318_cast_fp16))[name = string("op_18430_cast_fp16")]; + bool var_18432_interleave_0 = const()[name = string("op_18432_interleave_0"), val = bool(false)]; + tensor var_18432_cast_fp16 = concat(axis = var_16975, interleave = var_18432_interleave_0, values = (var_18320_cast_fp16, var_18322_cast_fp16, var_18324_cast_fp16, var_18326_cast_fp16))[name = string("op_18432_cast_fp16")]; + bool var_18434_interleave_0 = const()[name = string("op_18434_interleave_0"), val = bool(false)]; + tensor var_18434_cast_fp16 = concat(axis = var_16975, interleave = var_18434_interleave_0, values = (var_18328_cast_fp16, var_18330_cast_fp16, var_18332_cast_fp16, var_18334_cast_fp16))[name = string("op_18434_cast_fp16")]; + bool var_18436_interleave_0 = const()[name = string("op_18436_interleave_0"), val = bool(false)]; + tensor var_18436_cast_fp16 = concat(axis = var_16975, interleave = var_18436_interleave_0, values = (var_18336_cast_fp16, var_18338_cast_fp16, var_18340_cast_fp16, var_18342_cast_fp16))[name = string("op_18436_cast_fp16")]; + bool var_18438_interleave_0 = const()[name = string("op_18438_interleave_0"), val = bool(false)]; + tensor var_18438_cast_fp16 = concat(axis = var_16975, interleave = var_18438_interleave_0, values = (var_18344_cast_fp16, var_18346_cast_fp16, var_18348_cast_fp16, var_18350_cast_fp16))[name = string("op_18438_cast_fp16")]; + bool var_18440_interleave_0 = const()[name = string("op_18440_interleave_0"), val = bool(false)]; + tensor var_18440_cast_fp16 = concat(axis = var_16975, interleave = var_18440_interleave_0, values = (var_18352_cast_fp16, var_18354_cast_fp16, var_18356_cast_fp16, var_18358_cast_fp16))[name = string("op_18440_cast_fp16")]; + bool var_18442_interleave_0 = const()[name = string("op_18442_interleave_0"), val = bool(false)]; + tensor var_18442_cast_fp16 = concat(axis = var_16975, interleave = var_18442_interleave_0, values = (var_18360_cast_fp16, var_18362_cast_fp16, var_18364_cast_fp16, var_18366_cast_fp16))[name = string("op_18442_cast_fp16")]; + bool var_18444_interleave_0 = const()[name = string("op_18444_interleave_0"), val = bool(false)]; + tensor var_18444_cast_fp16 = concat(axis = var_16975, interleave = var_18444_interleave_0, values = (var_18368_cast_fp16, var_18370_cast_fp16, var_18372_cast_fp16, var_18374_cast_fp16))[name = string("op_18444_cast_fp16")]; + bool var_18446_interleave_0 = const()[name = string("op_18446_interleave_0"), val = bool(false)]; + tensor var_18446_cast_fp16 = concat(axis = var_16975, interleave = var_18446_interleave_0, values = (var_18376_cast_fp16, var_18378_cast_fp16, var_18380_cast_fp16, var_18382_cast_fp16))[name = string("op_18446_cast_fp16")]; + bool var_18448_interleave_0 = const()[name = string("op_18448_interleave_0"), val = bool(false)]; + tensor var_18448_cast_fp16 = concat(axis = var_16975, interleave = var_18448_interleave_0, values = (var_18384_cast_fp16, var_18386_cast_fp16, var_18388_cast_fp16, var_18390_cast_fp16))[name = string("op_18448_cast_fp16")]; + bool var_18450_interleave_0 = const()[name = string("op_18450_interleave_0"), val = bool(false)]; + tensor var_18450_cast_fp16 = concat(axis = var_16975, interleave = var_18450_interleave_0, values = (var_18392_cast_fp16, var_18394_cast_fp16, var_18396_cast_fp16, var_18398_cast_fp16))[name = string("op_18450_cast_fp16")]; + bool var_18452_interleave_0 = const()[name = string("op_18452_interleave_0"), val = bool(false)]; + tensor var_18452_cast_fp16 = concat(axis = var_16975, interleave = var_18452_interleave_0, values = (var_18400_cast_fp16, var_18402_cast_fp16, var_18404_cast_fp16, var_18406_cast_fp16))[name = string("op_18452_cast_fp16")]; + bool var_18454_interleave_0 = const()[name = string("op_18454_interleave_0"), val = bool(false)]; + tensor var_18454_cast_fp16 = concat(axis = var_16975, interleave = var_18454_interleave_0, values = (var_18408_cast_fp16, var_18410_cast_fp16, var_18412_cast_fp16, var_18414_cast_fp16))[name = string("op_18454_cast_fp16")]; + bool input_89_interleave_0 = const()[name = string("input_89_interleave_0"), val = bool(false)]; + tensor input_89_cast_fp16 = concat(axis = var_17000, interleave = input_89_interleave_0, values = (var_18416_cast_fp16, var_18418_cast_fp16, var_18420_cast_fp16, var_18422_cast_fp16, var_18424_cast_fp16, var_18426_cast_fp16, var_18428_cast_fp16, var_18430_cast_fp16, var_18432_cast_fp16, var_18434_cast_fp16, var_18436_cast_fp16, var_18438_cast_fp16, var_18440_cast_fp16, var_18442_cast_fp16, var_18444_cast_fp16, var_18446_cast_fp16, var_18448_cast_fp16, var_18450_cast_fp16, var_18452_cast_fp16, var_18454_cast_fp16))[name = string("input_89_cast_fp16")]; + string obj_47_pad_type_0 = const()[name = string("obj_47_pad_type_0"), val = string("valid")]; + tensor obj_47_strides_0 = const()[name = string("obj_47_strides_0"), val = tensor([1, 1])]; + tensor obj_47_pad_0 = const()[name = string("obj_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_47_dilations_0 = const()[name = string("obj_47_dilations_0"), val = tensor([1, 1])]; + int32 obj_47_groups_0 = const()[name = string("obj_47_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457391360)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460668224)))]; + tensor obj_47_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_47_dilations_0, groups = obj_47_groups_0, pad = obj_47_pad_0, pad_type = obj_47_pad_type_0, strides = obj_47_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_47_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_47_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_18473_to_fp16 = const()[name = string("op_18473_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_18473_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460670848)))]; + tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460673472)))]; + fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460676096)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473783360)))]; + tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; + tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + string hidden_states_27_pad_type_0 = const()[name = string("hidden_states_27_pad_type_0"), val = string("valid")]; + tensor hidden_states_27_strides_0 = const()[name = string("hidden_states_27_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_27_pad_0 = const()[name = string("hidden_states_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_27_dilations_0 = const()[name = string("hidden_states_27_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_27_groups_0 = const()[name = string("hidden_states_27_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473793664)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486900928)))]; + tensor hidden_states_27_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_27_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_27_cast_fp16)[name = string("inputs_49_cast_fp16")]; + int32 var_18502 = const()[name = string("op_18502"), val = int32(3)]; + int32 var_18527 = const()[name = string("op_18527"), val = int32(1)]; + tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; + fp16 var_18544_to_fp16 = const()[name = string("op_18544_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_18544_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; + tensor obj_49_gamma_0_to_fp16 = const()[name = string("obj_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486903552)))]; + tensor obj_49_beta_0_to_fp16 = const()[name = string("obj_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486906176)))]; + fp16 obj_49_epsilon_0_to_fp16 = const()[name = string("obj_49_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_49_cast_fp16 = batch_norm(beta = obj_49_beta_0_to_fp16, epsilon = obj_49_epsilon_0_to_fp16, gamma = obj_49_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_49_cast_fp16")]; + string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; + tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; + tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; + int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486908800)))]; + tensor layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490185664)))]; + tensor query_25_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("query_25_cast_fp16")]; + string key_25_pad_type_0 = const()[name = string("key_25_pad_type_0"), val = string("valid")]; + tensor key_25_strides_0 = const()[name = string("key_25_strides_0"), val = tensor([1, 1])]; + tensor key_25_pad_0 = const()[name = string("key_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_25_dilations_0 = const()[name = string("key_25_dilations_0"), val = tensor([1, 1])]; + int32 key_25_groups_0 = const()[name = string("key_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490188288)))]; + tensor key_25_cast_fp16 = conv(dilations = key_25_dilations_0, groups = key_25_groups_0, pad = key_25_pad_0, pad_type = key_25_pad_type_0, strides = key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("key_25_cast_fp16")]; + string value_25_pad_type_0 = const()[name = string("value_25_pad_type_0"), val = string("valid")]; + tensor value_25_strides_0 = const()[name = string("value_25_strides_0"), val = tensor([1, 1])]; + tensor value_25_pad_0 = const()[name = string("value_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_25_dilations_0 = const()[name = string("value_25_dilations_0"), val = tensor([1, 1])]; + int32 value_25_groups_0 = const()[name = string("value_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(493465152)))]; + tensor layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496742016)))]; + tensor value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = value_25_dilations_0, groups = value_25_groups_0, pad = value_25_pad_0, pad_type = value_25_pad_type_0, strides = value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("value_25_cast_fp16")]; + tensor var_18582_begin_0 = const()[name = string("op_18582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18582_end_0 = const()[name = string("op_18582_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18582_end_mask_0 = const()[name = string("op_18582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18582_cast_fp16 = slice_by_index(begin = var_18582_begin_0, end = var_18582_end_0, end_mask = var_18582_end_mask_0, x = query_25_cast_fp16)[name = string("op_18582_cast_fp16")]; + tensor var_18586_begin_0 = const()[name = string("op_18586_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18586_end_0 = const()[name = string("op_18586_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_18586_end_mask_0 = const()[name = string("op_18586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18586_cast_fp16 = slice_by_index(begin = var_18586_begin_0, end = var_18586_end_0, end_mask = var_18586_end_mask_0, x = query_25_cast_fp16)[name = string("op_18586_cast_fp16")]; + tensor var_18590_begin_0 = const()[name = string("op_18590_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18590_end_0 = const()[name = string("op_18590_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_18590_end_mask_0 = const()[name = string("op_18590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18590_cast_fp16 = slice_by_index(begin = var_18590_begin_0, end = var_18590_end_0, end_mask = var_18590_end_mask_0, x = query_25_cast_fp16)[name = string("op_18590_cast_fp16")]; + tensor var_18594_begin_0 = const()[name = string("op_18594_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18594_end_0 = const()[name = string("op_18594_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_18594_end_mask_0 = const()[name = string("op_18594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18594_cast_fp16 = slice_by_index(begin = var_18594_begin_0, end = var_18594_end_0, end_mask = var_18594_end_mask_0, x = query_25_cast_fp16)[name = string("op_18594_cast_fp16")]; + tensor var_18598_begin_0 = const()[name = string("op_18598_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18598_end_0 = const()[name = string("op_18598_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_18598_end_mask_0 = const()[name = string("op_18598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18598_cast_fp16 = slice_by_index(begin = var_18598_begin_0, end = var_18598_end_0, end_mask = var_18598_end_mask_0, x = query_25_cast_fp16)[name = string("op_18598_cast_fp16")]; + tensor var_18602_begin_0 = const()[name = string("op_18602_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18602_end_0 = const()[name = string("op_18602_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_18602_end_mask_0 = const()[name = string("op_18602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18602_cast_fp16 = slice_by_index(begin = var_18602_begin_0, end = var_18602_end_0, end_mask = var_18602_end_mask_0, x = query_25_cast_fp16)[name = string("op_18602_cast_fp16")]; + tensor var_18606_begin_0 = const()[name = string("op_18606_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18606_end_0 = const()[name = string("op_18606_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_18606_end_mask_0 = const()[name = string("op_18606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18606_cast_fp16 = slice_by_index(begin = var_18606_begin_0, end = var_18606_end_0, end_mask = var_18606_end_mask_0, x = query_25_cast_fp16)[name = string("op_18606_cast_fp16")]; + tensor var_18610_begin_0 = const()[name = string("op_18610_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18610_end_0 = const()[name = string("op_18610_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_18610_end_mask_0 = const()[name = string("op_18610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18610_cast_fp16 = slice_by_index(begin = var_18610_begin_0, end = var_18610_end_0, end_mask = var_18610_end_mask_0, x = query_25_cast_fp16)[name = string("op_18610_cast_fp16")]; + tensor var_18614_begin_0 = const()[name = string("op_18614_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18614_end_0 = const()[name = string("op_18614_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_18614_end_mask_0 = const()[name = string("op_18614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18614_cast_fp16 = slice_by_index(begin = var_18614_begin_0, end = var_18614_end_0, end_mask = var_18614_end_mask_0, x = query_25_cast_fp16)[name = string("op_18614_cast_fp16")]; + tensor var_18618_begin_0 = const()[name = string("op_18618_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18618_end_0 = const()[name = string("op_18618_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_18618_end_mask_0 = const()[name = string("op_18618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18618_cast_fp16 = slice_by_index(begin = var_18618_begin_0, end = var_18618_end_0, end_mask = var_18618_end_mask_0, x = query_25_cast_fp16)[name = string("op_18618_cast_fp16")]; + tensor var_18622_begin_0 = const()[name = string("op_18622_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18622_end_0 = const()[name = string("op_18622_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_18622_end_mask_0 = const()[name = string("op_18622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18622_cast_fp16 = slice_by_index(begin = var_18622_begin_0, end = var_18622_end_0, end_mask = var_18622_end_mask_0, x = query_25_cast_fp16)[name = string("op_18622_cast_fp16")]; + tensor var_18626_begin_0 = const()[name = string("op_18626_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18626_end_0 = const()[name = string("op_18626_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_18626_end_mask_0 = const()[name = string("op_18626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18626_cast_fp16 = slice_by_index(begin = var_18626_begin_0, end = var_18626_end_0, end_mask = var_18626_end_mask_0, x = query_25_cast_fp16)[name = string("op_18626_cast_fp16")]; + tensor var_18630_begin_0 = const()[name = string("op_18630_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18630_end_0 = const()[name = string("op_18630_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_18630_end_mask_0 = const()[name = string("op_18630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18630_cast_fp16 = slice_by_index(begin = var_18630_begin_0, end = var_18630_end_0, end_mask = var_18630_end_mask_0, x = query_25_cast_fp16)[name = string("op_18630_cast_fp16")]; + tensor var_18634_begin_0 = const()[name = string("op_18634_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18634_end_0 = const()[name = string("op_18634_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_18634_end_mask_0 = const()[name = string("op_18634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18634_cast_fp16 = slice_by_index(begin = var_18634_begin_0, end = var_18634_end_0, end_mask = var_18634_end_mask_0, x = query_25_cast_fp16)[name = string("op_18634_cast_fp16")]; + tensor var_18638_begin_0 = const()[name = string("op_18638_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18638_end_0 = const()[name = string("op_18638_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_18638_end_mask_0 = const()[name = string("op_18638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18638_cast_fp16 = slice_by_index(begin = var_18638_begin_0, end = var_18638_end_0, end_mask = var_18638_end_mask_0, x = query_25_cast_fp16)[name = string("op_18638_cast_fp16")]; + tensor var_18642_begin_0 = const()[name = string("op_18642_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18642_end_0 = const()[name = string("op_18642_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_18642_end_mask_0 = const()[name = string("op_18642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18642_cast_fp16 = slice_by_index(begin = var_18642_begin_0, end = var_18642_end_0, end_mask = var_18642_end_mask_0, x = query_25_cast_fp16)[name = string("op_18642_cast_fp16")]; + tensor var_18646_begin_0 = const()[name = string("op_18646_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18646_end_0 = const()[name = string("op_18646_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_18646_end_mask_0 = const()[name = string("op_18646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18646_cast_fp16 = slice_by_index(begin = var_18646_begin_0, end = var_18646_end_0, end_mask = var_18646_end_mask_0, x = query_25_cast_fp16)[name = string("op_18646_cast_fp16")]; + tensor var_18650_begin_0 = const()[name = string("op_18650_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18650_end_0 = const()[name = string("op_18650_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_18650_end_mask_0 = const()[name = string("op_18650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18650_cast_fp16 = slice_by_index(begin = var_18650_begin_0, end = var_18650_end_0, end_mask = var_18650_end_mask_0, x = query_25_cast_fp16)[name = string("op_18650_cast_fp16")]; + tensor var_18654_begin_0 = const()[name = string("op_18654_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18654_end_0 = const()[name = string("op_18654_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_18654_end_mask_0 = const()[name = string("op_18654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18654_cast_fp16 = slice_by_index(begin = var_18654_begin_0, end = var_18654_end_0, end_mask = var_18654_end_mask_0, x = query_25_cast_fp16)[name = string("op_18654_cast_fp16")]; + tensor var_18658_begin_0 = const()[name = string("op_18658_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18658_end_0 = const()[name = string("op_18658_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_18658_end_mask_0 = const()[name = string("op_18658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18658_cast_fp16 = slice_by_index(begin = var_18658_begin_0, end = var_18658_end_0, end_mask = var_18658_end_mask_0, x = query_25_cast_fp16)[name = string("op_18658_cast_fp16")]; + tensor var_18667_begin_0 = const()[name = string("op_18667_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18667_end_0 = const()[name = string("op_18667_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18667_end_mask_0 = const()[name = string("op_18667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18667_cast_fp16 = slice_by_index(begin = var_18667_begin_0, end = var_18667_end_0, end_mask = var_18667_end_mask_0, x = var_18582_cast_fp16)[name = string("op_18667_cast_fp16")]; + tensor var_18674_begin_0 = const()[name = string("op_18674_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18674_end_0 = const()[name = string("op_18674_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18674_end_mask_0 = const()[name = string("op_18674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18674_cast_fp16 = slice_by_index(begin = var_18674_begin_0, end = var_18674_end_0, end_mask = var_18674_end_mask_0, x = var_18582_cast_fp16)[name = string("op_18674_cast_fp16")]; + tensor var_18681_begin_0 = const()[name = string("op_18681_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18681_end_0 = const()[name = string("op_18681_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18681_end_mask_0 = const()[name = string("op_18681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18681_cast_fp16 = slice_by_index(begin = var_18681_begin_0, end = var_18681_end_0, end_mask = var_18681_end_mask_0, x = var_18582_cast_fp16)[name = string("op_18681_cast_fp16")]; + tensor var_18688_begin_0 = const()[name = string("op_18688_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18688_end_0 = const()[name = string("op_18688_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18688_end_mask_0 = const()[name = string("op_18688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18688_cast_fp16 = slice_by_index(begin = var_18688_begin_0, end = var_18688_end_0, end_mask = var_18688_end_mask_0, x = var_18582_cast_fp16)[name = string("op_18688_cast_fp16")]; + tensor var_18695_begin_0 = const()[name = string("op_18695_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18695_end_0 = const()[name = string("op_18695_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18695_end_mask_0 = const()[name = string("op_18695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18695_cast_fp16 = slice_by_index(begin = var_18695_begin_0, end = var_18695_end_0, end_mask = var_18695_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18695_cast_fp16")]; + tensor var_18702_begin_0 = const()[name = string("op_18702_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18702_end_0 = const()[name = string("op_18702_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18702_end_mask_0 = const()[name = string("op_18702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18702_cast_fp16 = slice_by_index(begin = var_18702_begin_0, end = var_18702_end_0, end_mask = var_18702_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18702_cast_fp16")]; + tensor var_18709_begin_0 = const()[name = string("op_18709_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18709_end_0 = const()[name = string("op_18709_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18709_end_mask_0 = const()[name = string("op_18709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18709_cast_fp16 = slice_by_index(begin = var_18709_begin_0, end = var_18709_end_0, end_mask = var_18709_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18709_cast_fp16")]; + tensor var_18716_begin_0 = const()[name = string("op_18716_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18716_end_0 = const()[name = string("op_18716_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18716_end_mask_0 = const()[name = string("op_18716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18716_cast_fp16 = slice_by_index(begin = var_18716_begin_0, end = var_18716_end_0, end_mask = var_18716_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18716_cast_fp16")]; + tensor var_18723_begin_0 = const()[name = string("op_18723_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18723_end_0 = const()[name = string("op_18723_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18723_end_mask_0 = const()[name = string("op_18723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18723_cast_fp16 = slice_by_index(begin = var_18723_begin_0, end = var_18723_end_0, end_mask = var_18723_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18723_cast_fp16")]; + tensor var_18730_begin_0 = const()[name = string("op_18730_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18730_end_0 = const()[name = string("op_18730_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18730_end_mask_0 = const()[name = string("op_18730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18730_cast_fp16 = slice_by_index(begin = var_18730_begin_0, end = var_18730_end_0, end_mask = var_18730_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18730_cast_fp16")]; + tensor var_18737_begin_0 = const()[name = string("op_18737_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18737_end_0 = const()[name = string("op_18737_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18737_end_mask_0 = const()[name = string("op_18737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18737_cast_fp16 = slice_by_index(begin = var_18737_begin_0, end = var_18737_end_0, end_mask = var_18737_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18737_cast_fp16")]; + tensor var_18744_begin_0 = const()[name = string("op_18744_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18744_end_0 = const()[name = string("op_18744_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18744_end_mask_0 = const()[name = string("op_18744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18744_cast_fp16 = slice_by_index(begin = var_18744_begin_0, end = var_18744_end_0, end_mask = var_18744_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18744_cast_fp16")]; + tensor var_18751_begin_0 = const()[name = string("op_18751_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18751_end_0 = const()[name = string("op_18751_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18751_end_mask_0 = const()[name = string("op_18751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18751_cast_fp16 = slice_by_index(begin = var_18751_begin_0, end = var_18751_end_0, end_mask = var_18751_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18751_cast_fp16")]; + tensor var_18758_begin_0 = const()[name = string("op_18758_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18758_end_0 = const()[name = string("op_18758_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18758_end_mask_0 = const()[name = string("op_18758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18758_cast_fp16 = slice_by_index(begin = var_18758_begin_0, end = var_18758_end_0, end_mask = var_18758_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18758_cast_fp16")]; + tensor var_18765_begin_0 = const()[name = string("op_18765_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18765_end_0 = const()[name = string("op_18765_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18765_end_mask_0 = const()[name = string("op_18765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18765_cast_fp16 = slice_by_index(begin = var_18765_begin_0, end = var_18765_end_0, end_mask = var_18765_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18765_cast_fp16")]; + tensor var_18772_begin_0 = const()[name = string("op_18772_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18772_end_0 = const()[name = string("op_18772_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18772_end_mask_0 = const()[name = string("op_18772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18772_cast_fp16 = slice_by_index(begin = var_18772_begin_0, end = var_18772_end_0, end_mask = var_18772_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18772_cast_fp16")]; + tensor var_18779_begin_0 = const()[name = string("op_18779_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18779_end_0 = const()[name = string("op_18779_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18779_end_mask_0 = const()[name = string("op_18779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18779_cast_fp16 = slice_by_index(begin = var_18779_begin_0, end = var_18779_end_0, end_mask = var_18779_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18779_cast_fp16")]; + tensor var_18786_begin_0 = const()[name = string("op_18786_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18786_end_0 = const()[name = string("op_18786_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18786_end_mask_0 = const()[name = string("op_18786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18786_cast_fp16 = slice_by_index(begin = var_18786_begin_0, end = var_18786_end_0, end_mask = var_18786_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18786_cast_fp16")]; + tensor var_18793_begin_0 = const()[name = string("op_18793_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18793_end_0 = const()[name = string("op_18793_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18793_end_mask_0 = const()[name = string("op_18793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18793_cast_fp16 = slice_by_index(begin = var_18793_begin_0, end = var_18793_end_0, end_mask = var_18793_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18793_cast_fp16")]; + tensor var_18800_begin_0 = const()[name = string("op_18800_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18800_end_0 = const()[name = string("op_18800_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18800_end_mask_0 = const()[name = string("op_18800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18800_cast_fp16 = slice_by_index(begin = var_18800_begin_0, end = var_18800_end_0, end_mask = var_18800_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18800_cast_fp16")]; + tensor var_18807_begin_0 = const()[name = string("op_18807_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18807_end_0 = const()[name = string("op_18807_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18807_end_mask_0 = const()[name = string("op_18807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18807_cast_fp16 = slice_by_index(begin = var_18807_begin_0, end = var_18807_end_0, end_mask = var_18807_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18807_cast_fp16")]; + tensor var_18814_begin_0 = const()[name = string("op_18814_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18814_end_0 = const()[name = string("op_18814_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18814_end_mask_0 = const()[name = string("op_18814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18814_cast_fp16 = slice_by_index(begin = var_18814_begin_0, end = var_18814_end_0, end_mask = var_18814_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18814_cast_fp16")]; + tensor var_18821_begin_0 = const()[name = string("op_18821_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18821_end_0 = const()[name = string("op_18821_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18821_end_mask_0 = const()[name = string("op_18821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18821_cast_fp16 = slice_by_index(begin = var_18821_begin_0, end = var_18821_end_0, end_mask = var_18821_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18821_cast_fp16")]; + tensor var_18828_begin_0 = const()[name = string("op_18828_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18828_end_0 = const()[name = string("op_18828_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18828_end_mask_0 = const()[name = string("op_18828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18828_cast_fp16 = slice_by_index(begin = var_18828_begin_0, end = var_18828_end_0, end_mask = var_18828_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18828_cast_fp16")]; + tensor var_18835_begin_0 = const()[name = string("op_18835_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18835_end_0 = const()[name = string("op_18835_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18835_end_mask_0 = const()[name = string("op_18835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18835_cast_fp16 = slice_by_index(begin = var_18835_begin_0, end = var_18835_end_0, end_mask = var_18835_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18835_cast_fp16")]; + tensor var_18842_begin_0 = const()[name = string("op_18842_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18842_end_0 = const()[name = string("op_18842_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18842_end_mask_0 = const()[name = string("op_18842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18842_cast_fp16 = slice_by_index(begin = var_18842_begin_0, end = var_18842_end_0, end_mask = var_18842_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18842_cast_fp16")]; + tensor var_18849_begin_0 = const()[name = string("op_18849_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18849_end_0 = const()[name = string("op_18849_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18849_end_mask_0 = const()[name = string("op_18849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18849_cast_fp16 = slice_by_index(begin = var_18849_begin_0, end = var_18849_end_0, end_mask = var_18849_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18849_cast_fp16")]; + tensor var_18856_begin_0 = const()[name = string("op_18856_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18856_end_0 = const()[name = string("op_18856_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18856_end_mask_0 = const()[name = string("op_18856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18856_cast_fp16 = slice_by_index(begin = var_18856_begin_0, end = var_18856_end_0, end_mask = var_18856_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18856_cast_fp16")]; + tensor var_18863_begin_0 = const()[name = string("op_18863_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18863_end_0 = const()[name = string("op_18863_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18863_end_mask_0 = const()[name = string("op_18863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18863_cast_fp16 = slice_by_index(begin = var_18863_begin_0, end = var_18863_end_0, end_mask = var_18863_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18863_cast_fp16")]; + tensor var_18870_begin_0 = const()[name = string("op_18870_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18870_end_0 = const()[name = string("op_18870_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18870_end_mask_0 = const()[name = string("op_18870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18870_cast_fp16 = slice_by_index(begin = var_18870_begin_0, end = var_18870_end_0, end_mask = var_18870_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18870_cast_fp16")]; + tensor var_18877_begin_0 = const()[name = string("op_18877_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18877_end_0 = const()[name = string("op_18877_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18877_end_mask_0 = const()[name = string("op_18877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18877_cast_fp16 = slice_by_index(begin = var_18877_begin_0, end = var_18877_end_0, end_mask = var_18877_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18877_cast_fp16")]; + tensor var_18884_begin_0 = const()[name = string("op_18884_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18884_end_0 = const()[name = string("op_18884_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18884_end_mask_0 = const()[name = string("op_18884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18884_cast_fp16 = slice_by_index(begin = var_18884_begin_0, end = var_18884_end_0, end_mask = var_18884_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18884_cast_fp16")]; + tensor var_18891_begin_0 = const()[name = string("op_18891_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18891_end_0 = const()[name = string("op_18891_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18891_end_mask_0 = const()[name = string("op_18891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18891_cast_fp16 = slice_by_index(begin = var_18891_begin_0, end = var_18891_end_0, end_mask = var_18891_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18891_cast_fp16")]; + tensor var_18898_begin_0 = const()[name = string("op_18898_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18898_end_0 = const()[name = string("op_18898_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18898_end_mask_0 = const()[name = string("op_18898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18898_cast_fp16 = slice_by_index(begin = var_18898_begin_0, end = var_18898_end_0, end_mask = var_18898_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18898_cast_fp16")]; + tensor var_18905_begin_0 = const()[name = string("op_18905_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18905_end_0 = const()[name = string("op_18905_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18905_end_mask_0 = const()[name = string("op_18905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18905_cast_fp16 = slice_by_index(begin = var_18905_begin_0, end = var_18905_end_0, end_mask = var_18905_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18905_cast_fp16")]; + tensor var_18912_begin_0 = const()[name = string("op_18912_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18912_end_0 = const()[name = string("op_18912_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18912_end_mask_0 = const()[name = string("op_18912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18912_cast_fp16 = slice_by_index(begin = var_18912_begin_0, end = var_18912_end_0, end_mask = var_18912_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18912_cast_fp16")]; + tensor var_18919_begin_0 = const()[name = string("op_18919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18919_end_0 = const()[name = string("op_18919_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18919_end_mask_0 = const()[name = string("op_18919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18919_cast_fp16 = slice_by_index(begin = var_18919_begin_0, end = var_18919_end_0, end_mask = var_18919_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18919_cast_fp16")]; + tensor var_18926_begin_0 = const()[name = string("op_18926_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18926_end_0 = const()[name = string("op_18926_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18926_end_mask_0 = const()[name = string("op_18926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18926_cast_fp16 = slice_by_index(begin = var_18926_begin_0, end = var_18926_end_0, end_mask = var_18926_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18926_cast_fp16")]; + tensor var_18933_begin_0 = const()[name = string("op_18933_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18933_end_0 = const()[name = string("op_18933_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18933_end_mask_0 = const()[name = string("op_18933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18933_cast_fp16 = slice_by_index(begin = var_18933_begin_0, end = var_18933_end_0, end_mask = var_18933_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18933_cast_fp16")]; + tensor var_18940_begin_0 = const()[name = string("op_18940_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18940_end_0 = const()[name = string("op_18940_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18940_end_mask_0 = const()[name = string("op_18940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18940_cast_fp16 = slice_by_index(begin = var_18940_begin_0, end = var_18940_end_0, end_mask = var_18940_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18940_cast_fp16")]; + tensor var_18947_begin_0 = const()[name = string("op_18947_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18947_end_0 = const()[name = string("op_18947_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18947_end_mask_0 = const()[name = string("op_18947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18947_cast_fp16 = slice_by_index(begin = var_18947_begin_0, end = var_18947_end_0, end_mask = var_18947_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18947_cast_fp16")]; + tensor var_18954_begin_0 = const()[name = string("op_18954_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18954_end_0 = const()[name = string("op_18954_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18954_end_mask_0 = const()[name = string("op_18954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18954_cast_fp16 = slice_by_index(begin = var_18954_begin_0, end = var_18954_end_0, end_mask = var_18954_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18954_cast_fp16")]; + tensor var_18961_begin_0 = const()[name = string("op_18961_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18961_end_0 = const()[name = string("op_18961_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18961_end_mask_0 = const()[name = string("op_18961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18961_cast_fp16 = slice_by_index(begin = var_18961_begin_0, end = var_18961_end_0, end_mask = var_18961_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18961_cast_fp16")]; + tensor var_18968_begin_0 = const()[name = string("op_18968_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18968_end_0 = const()[name = string("op_18968_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18968_end_mask_0 = const()[name = string("op_18968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18968_cast_fp16 = slice_by_index(begin = var_18968_begin_0, end = var_18968_end_0, end_mask = var_18968_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18968_cast_fp16")]; + tensor var_18975_begin_0 = const()[name = string("op_18975_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18975_end_0 = const()[name = string("op_18975_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18975_end_mask_0 = const()[name = string("op_18975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18975_cast_fp16 = slice_by_index(begin = var_18975_begin_0, end = var_18975_end_0, end_mask = var_18975_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18975_cast_fp16")]; + tensor var_18982_begin_0 = const()[name = string("op_18982_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18982_end_0 = const()[name = string("op_18982_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18982_end_mask_0 = const()[name = string("op_18982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18982_cast_fp16 = slice_by_index(begin = var_18982_begin_0, end = var_18982_end_0, end_mask = var_18982_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18982_cast_fp16")]; + tensor var_18989_begin_0 = const()[name = string("op_18989_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18989_end_0 = const()[name = string("op_18989_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18989_end_mask_0 = const()[name = string("op_18989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18989_cast_fp16 = slice_by_index(begin = var_18989_begin_0, end = var_18989_end_0, end_mask = var_18989_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18989_cast_fp16")]; + tensor var_18996_begin_0 = const()[name = string("op_18996_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18996_end_0 = const()[name = string("op_18996_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18996_end_mask_0 = const()[name = string("op_18996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18996_cast_fp16 = slice_by_index(begin = var_18996_begin_0, end = var_18996_end_0, end_mask = var_18996_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18996_cast_fp16")]; + tensor var_19003_begin_0 = const()[name = string("op_19003_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19003_end_0 = const()[name = string("op_19003_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19003_end_mask_0 = const()[name = string("op_19003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19003_cast_fp16 = slice_by_index(begin = var_19003_begin_0, end = var_19003_end_0, end_mask = var_19003_end_mask_0, x = var_18630_cast_fp16)[name = string("op_19003_cast_fp16")]; + tensor var_19010_begin_0 = const()[name = string("op_19010_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19010_end_0 = const()[name = string("op_19010_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19010_end_mask_0 = const()[name = string("op_19010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19010_cast_fp16 = slice_by_index(begin = var_19010_begin_0, end = var_19010_end_0, end_mask = var_19010_end_mask_0, x = var_18630_cast_fp16)[name = string("op_19010_cast_fp16")]; + tensor var_19017_begin_0 = const()[name = string("op_19017_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19017_end_0 = const()[name = string("op_19017_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19017_end_mask_0 = const()[name = string("op_19017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19017_cast_fp16 = slice_by_index(begin = var_19017_begin_0, end = var_19017_end_0, end_mask = var_19017_end_mask_0, x = var_18630_cast_fp16)[name = string("op_19017_cast_fp16")]; + tensor var_19024_begin_0 = const()[name = string("op_19024_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19024_end_0 = const()[name = string("op_19024_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19024_end_mask_0 = const()[name = string("op_19024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19024_cast_fp16 = slice_by_index(begin = var_19024_begin_0, end = var_19024_end_0, end_mask = var_19024_end_mask_0, x = var_18630_cast_fp16)[name = string("op_19024_cast_fp16")]; + tensor var_19031_begin_0 = const()[name = string("op_19031_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19031_end_0 = const()[name = string("op_19031_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19031_end_mask_0 = const()[name = string("op_19031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19031_cast_fp16 = slice_by_index(begin = var_19031_begin_0, end = var_19031_end_0, end_mask = var_19031_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19031_cast_fp16")]; + tensor var_19038_begin_0 = const()[name = string("op_19038_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19038_end_0 = const()[name = string("op_19038_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19038_end_mask_0 = const()[name = string("op_19038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19038_cast_fp16 = slice_by_index(begin = var_19038_begin_0, end = var_19038_end_0, end_mask = var_19038_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19038_cast_fp16")]; + tensor var_19045_begin_0 = const()[name = string("op_19045_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19045_end_0 = const()[name = string("op_19045_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19045_end_mask_0 = const()[name = string("op_19045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19045_cast_fp16 = slice_by_index(begin = var_19045_begin_0, end = var_19045_end_0, end_mask = var_19045_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19045_cast_fp16")]; + tensor var_19052_begin_0 = const()[name = string("op_19052_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19052_end_0 = const()[name = string("op_19052_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19052_end_mask_0 = const()[name = string("op_19052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19052_cast_fp16 = slice_by_index(begin = var_19052_begin_0, end = var_19052_end_0, end_mask = var_19052_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19052_cast_fp16")]; + tensor var_19059_begin_0 = const()[name = string("op_19059_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19059_end_0 = const()[name = string("op_19059_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19059_end_mask_0 = const()[name = string("op_19059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19059_cast_fp16 = slice_by_index(begin = var_19059_begin_0, end = var_19059_end_0, end_mask = var_19059_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19059_cast_fp16")]; + tensor var_19066_begin_0 = const()[name = string("op_19066_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19066_end_0 = const()[name = string("op_19066_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19066_end_mask_0 = const()[name = string("op_19066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19066_cast_fp16 = slice_by_index(begin = var_19066_begin_0, end = var_19066_end_0, end_mask = var_19066_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19066_cast_fp16")]; + tensor var_19073_begin_0 = const()[name = string("op_19073_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19073_end_0 = const()[name = string("op_19073_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19073_end_mask_0 = const()[name = string("op_19073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19073_cast_fp16 = slice_by_index(begin = var_19073_begin_0, end = var_19073_end_0, end_mask = var_19073_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19073_cast_fp16")]; + tensor var_19080_begin_0 = const()[name = string("op_19080_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19080_end_0 = const()[name = string("op_19080_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19080_end_mask_0 = const()[name = string("op_19080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19080_cast_fp16 = slice_by_index(begin = var_19080_begin_0, end = var_19080_end_0, end_mask = var_19080_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19080_cast_fp16")]; + tensor var_19087_begin_0 = const()[name = string("op_19087_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19087_end_0 = const()[name = string("op_19087_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19087_end_mask_0 = const()[name = string("op_19087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19087_cast_fp16 = slice_by_index(begin = var_19087_begin_0, end = var_19087_end_0, end_mask = var_19087_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19087_cast_fp16")]; + tensor var_19094_begin_0 = const()[name = string("op_19094_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19094_end_0 = const()[name = string("op_19094_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19094_end_mask_0 = const()[name = string("op_19094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19094_cast_fp16 = slice_by_index(begin = var_19094_begin_0, end = var_19094_end_0, end_mask = var_19094_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19094_cast_fp16")]; + tensor var_19101_begin_0 = const()[name = string("op_19101_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19101_end_0 = const()[name = string("op_19101_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19101_end_mask_0 = const()[name = string("op_19101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19101_cast_fp16 = slice_by_index(begin = var_19101_begin_0, end = var_19101_end_0, end_mask = var_19101_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19101_cast_fp16")]; + tensor var_19108_begin_0 = const()[name = string("op_19108_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19108_end_0 = const()[name = string("op_19108_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19108_end_mask_0 = const()[name = string("op_19108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19108_cast_fp16 = slice_by_index(begin = var_19108_begin_0, end = var_19108_end_0, end_mask = var_19108_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19108_cast_fp16")]; + tensor var_19115_begin_0 = const()[name = string("op_19115_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19115_end_0 = const()[name = string("op_19115_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19115_end_mask_0 = const()[name = string("op_19115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19115_cast_fp16 = slice_by_index(begin = var_19115_begin_0, end = var_19115_end_0, end_mask = var_19115_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19115_cast_fp16")]; + tensor var_19122_begin_0 = const()[name = string("op_19122_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19122_end_0 = const()[name = string("op_19122_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19122_end_mask_0 = const()[name = string("op_19122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19122_cast_fp16 = slice_by_index(begin = var_19122_begin_0, end = var_19122_end_0, end_mask = var_19122_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19122_cast_fp16")]; + tensor var_19129_begin_0 = const()[name = string("op_19129_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19129_end_0 = const()[name = string("op_19129_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19129_end_mask_0 = const()[name = string("op_19129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19129_cast_fp16 = slice_by_index(begin = var_19129_begin_0, end = var_19129_end_0, end_mask = var_19129_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19129_cast_fp16")]; + tensor var_19136_begin_0 = const()[name = string("op_19136_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19136_end_0 = const()[name = string("op_19136_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19136_end_mask_0 = const()[name = string("op_19136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19136_cast_fp16 = slice_by_index(begin = var_19136_begin_0, end = var_19136_end_0, end_mask = var_19136_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19136_cast_fp16")]; + tensor var_19143_begin_0 = const()[name = string("op_19143_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19143_end_0 = const()[name = string("op_19143_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19143_end_mask_0 = const()[name = string("op_19143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19143_cast_fp16 = slice_by_index(begin = var_19143_begin_0, end = var_19143_end_0, end_mask = var_19143_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19143_cast_fp16")]; + tensor var_19150_begin_0 = const()[name = string("op_19150_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19150_end_0 = const()[name = string("op_19150_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19150_end_mask_0 = const()[name = string("op_19150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19150_cast_fp16 = slice_by_index(begin = var_19150_begin_0, end = var_19150_end_0, end_mask = var_19150_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19150_cast_fp16")]; + tensor var_19157_begin_0 = const()[name = string("op_19157_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19157_end_0 = const()[name = string("op_19157_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19157_end_mask_0 = const()[name = string("op_19157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19157_cast_fp16 = slice_by_index(begin = var_19157_begin_0, end = var_19157_end_0, end_mask = var_19157_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19157_cast_fp16")]; + tensor var_19164_begin_0 = const()[name = string("op_19164_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19164_end_0 = const()[name = string("op_19164_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19164_end_mask_0 = const()[name = string("op_19164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19164_cast_fp16 = slice_by_index(begin = var_19164_begin_0, end = var_19164_end_0, end_mask = var_19164_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19164_cast_fp16")]; + tensor var_19171_begin_0 = const()[name = string("op_19171_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19171_end_0 = const()[name = string("op_19171_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19171_end_mask_0 = const()[name = string("op_19171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19171_cast_fp16 = slice_by_index(begin = var_19171_begin_0, end = var_19171_end_0, end_mask = var_19171_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19171_cast_fp16")]; + tensor var_19178_begin_0 = const()[name = string("op_19178_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19178_end_0 = const()[name = string("op_19178_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19178_end_mask_0 = const()[name = string("op_19178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19178_cast_fp16 = slice_by_index(begin = var_19178_begin_0, end = var_19178_end_0, end_mask = var_19178_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19178_cast_fp16")]; + tensor var_19185_begin_0 = const()[name = string("op_19185_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19185_end_0 = const()[name = string("op_19185_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19185_end_mask_0 = const()[name = string("op_19185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19185_cast_fp16 = slice_by_index(begin = var_19185_begin_0, end = var_19185_end_0, end_mask = var_19185_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19185_cast_fp16")]; + tensor var_19192_begin_0 = const()[name = string("op_19192_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19192_end_0 = const()[name = string("op_19192_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19192_end_mask_0 = const()[name = string("op_19192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19192_cast_fp16 = slice_by_index(begin = var_19192_begin_0, end = var_19192_end_0, end_mask = var_19192_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19192_cast_fp16")]; + tensor var_19199_begin_0 = const()[name = string("op_19199_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19199_end_0 = const()[name = string("op_19199_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19199_end_mask_0 = const()[name = string("op_19199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19199_cast_fp16 = slice_by_index(begin = var_19199_begin_0, end = var_19199_end_0, end_mask = var_19199_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19199_cast_fp16")]; + tensor var_19206_begin_0 = const()[name = string("op_19206_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19206_end_0 = const()[name = string("op_19206_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19206_end_mask_0 = const()[name = string("op_19206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19206_cast_fp16 = slice_by_index(begin = var_19206_begin_0, end = var_19206_end_0, end_mask = var_19206_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19206_cast_fp16")]; + tensor var_19213_begin_0 = const()[name = string("op_19213_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19213_end_0 = const()[name = string("op_19213_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19213_end_mask_0 = const()[name = string("op_19213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19213_cast_fp16 = slice_by_index(begin = var_19213_begin_0, end = var_19213_end_0, end_mask = var_19213_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19213_cast_fp16")]; + tensor var_19220_begin_0 = const()[name = string("op_19220_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19220_end_0 = const()[name = string("op_19220_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19220_end_mask_0 = const()[name = string("op_19220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19220_cast_fp16 = slice_by_index(begin = var_19220_begin_0, end = var_19220_end_0, end_mask = var_19220_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19220_cast_fp16")]; + tensor k_25_perm_0 = const()[name = string("k_25_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_19225_begin_0 = const()[name = string("op_19225_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19225_end_0 = const()[name = string("op_19225_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_19225_end_mask_0 = const()[name = string("op_19225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_25_cast_fp16 = transpose(perm = k_25_perm_0, x = key_25_cast_fp16)[name = string("transpose_19")]; + tensor var_19225_cast_fp16 = slice_by_index(begin = var_19225_begin_0, end = var_19225_end_0, end_mask = var_19225_end_mask_0, x = k_25_cast_fp16)[name = string("op_19225_cast_fp16")]; + tensor var_19229_begin_0 = const()[name = string("op_19229_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_19229_end_0 = const()[name = string("op_19229_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_19229_end_mask_0 = const()[name = string("op_19229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19229_cast_fp16 = slice_by_index(begin = var_19229_begin_0, end = var_19229_end_0, end_mask = var_19229_end_mask_0, x = k_25_cast_fp16)[name = string("op_19229_cast_fp16")]; + tensor var_19233_begin_0 = const()[name = string("op_19233_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_19233_end_0 = const()[name = string("op_19233_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_19233_end_mask_0 = const()[name = string("op_19233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19233_cast_fp16 = slice_by_index(begin = var_19233_begin_0, end = var_19233_end_0, end_mask = var_19233_end_mask_0, x = k_25_cast_fp16)[name = string("op_19233_cast_fp16")]; + tensor var_19237_begin_0 = const()[name = string("op_19237_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_19237_end_0 = const()[name = string("op_19237_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_19237_end_mask_0 = const()[name = string("op_19237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19237_cast_fp16 = slice_by_index(begin = var_19237_begin_0, end = var_19237_end_0, end_mask = var_19237_end_mask_0, x = k_25_cast_fp16)[name = string("op_19237_cast_fp16")]; + tensor var_19241_begin_0 = const()[name = string("op_19241_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_19241_end_0 = const()[name = string("op_19241_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_19241_end_mask_0 = const()[name = string("op_19241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19241_cast_fp16 = slice_by_index(begin = var_19241_begin_0, end = var_19241_end_0, end_mask = var_19241_end_mask_0, x = k_25_cast_fp16)[name = string("op_19241_cast_fp16")]; + tensor var_19245_begin_0 = const()[name = string("op_19245_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_19245_end_0 = const()[name = string("op_19245_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_19245_end_mask_0 = const()[name = string("op_19245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19245_cast_fp16 = slice_by_index(begin = var_19245_begin_0, end = var_19245_end_0, end_mask = var_19245_end_mask_0, x = k_25_cast_fp16)[name = string("op_19245_cast_fp16")]; + tensor var_19249_begin_0 = const()[name = string("op_19249_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_19249_end_0 = const()[name = string("op_19249_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_19249_end_mask_0 = const()[name = string("op_19249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19249_cast_fp16 = slice_by_index(begin = var_19249_begin_0, end = var_19249_end_0, end_mask = var_19249_end_mask_0, x = k_25_cast_fp16)[name = string("op_19249_cast_fp16")]; + tensor var_19253_begin_0 = const()[name = string("op_19253_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_19253_end_0 = const()[name = string("op_19253_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_19253_end_mask_0 = const()[name = string("op_19253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19253_cast_fp16 = slice_by_index(begin = var_19253_begin_0, end = var_19253_end_0, end_mask = var_19253_end_mask_0, x = k_25_cast_fp16)[name = string("op_19253_cast_fp16")]; + tensor var_19257_begin_0 = const()[name = string("op_19257_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_19257_end_0 = const()[name = string("op_19257_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_19257_end_mask_0 = const()[name = string("op_19257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19257_cast_fp16 = slice_by_index(begin = var_19257_begin_0, end = var_19257_end_0, end_mask = var_19257_end_mask_0, x = k_25_cast_fp16)[name = string("op_19257_cast_fp16")]; + tensor var_19261_begin_0 = const()[name = string("op_19261_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_19261_end_0 = const()[name = string("op_19261_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_19261_end_mask_0 = const()[name = string("op_19261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19261_cast_fp16 = slice_by_index(begin = var_19261_begin_0, end = var_19261_end_0, end_mask = var_19261_end_mask_0, x = k_25_cast_fp16)[name = string("op_19261_cast_fp16")]; + tensor var_19265_begin_0 = const()[name = string("op_19265_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_19265_end_0 = const()[name = string("op_19265_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_19265_end_mask_0 = const()[name = string("op_19265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19265_cast_fp16 = slice_by_index(begin = var_19265_begin_0, end = var_19265_end_0, end_mask = var_19265_end_mask_0, x = k_25_cast_fp16)[name = string("op_19265_cast_fp16")]; + tensor var_19269_begin_0 = const()[name = string("op_19269_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_19269_end_0 = const()[name = string("op_19269_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_19269_end_mask_0 = const()[name = string("op_19269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19269_cast_fp16 = slice_by_index(begin = var_19269_begin_0, end = var_19269_end_0, end_mask = var_19269_end_mask_0, x = k_25_cast_fp16)[name = string("op_19269_cast_fp16")]; + tensor var_19273_begin_0 = const()[name = string("op_19273_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_19273_end_0 = const()[name = string("op_19273_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_19273_end_mask_0 = const()[name = string("op_19273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19273_cast_fp16 = slice_by_index(begin = var_19273_begin_0, end = var_19273_end_0, end_mask = var_19273_end_mask_0, x = k_25_cast_fp16)[name = string("op_19273_cast_fp16")]; + tensor var_19277_begin_0 = const()[name = string("op_19277_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_19277_end_0 = const()[name = string("op_19277_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_19277_end_mask_0 = const()[name = string("op_19277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19277_cast_fp16 = slice_by_index(begin = var_19277_begin_0, end = var_19277_end_0, end_mask = var_19277_end_mask_0, x = k_25_cast_fp16)[name = string("op_19277_cast_fp16")]; + tensor var_19281_begin_0 = const()[name = string("op_19281_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_19281_end_0 = const()[name = string("op_19281_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_19281_end_mask_0 = const()[name = string("op_19281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19281_cast_fp16 = slice_by_index(begin = var_19281_begin_0, end = var_19281_end_0, end_mask = var_19281_end_mask_0, x = k_25_cast_fp16)[name = string("op_19281_cast_fp16")]; + tensor var_19285_begin_0 = const()[name = string("op_19285_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_19285_end_0 = const()[name = string("op_19285_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_19285_end_mask_0 = const()[name = string("op_19285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19285_cast_fp16 = slice_by_index(begin = var_19285_begin_0, end = var_19285_end_0, end_mask = var_19285_end_mask_0, x = k_25_cast_fp16)[name = string("op_19285_cast_fp16")]; + tensor var_19289_begin_0 = const()[name = string("op_19289_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19289_end_0 = const()[name = string("op_19289_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_19289_end_mask_0 = const()[name = string("op_19289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19289_cast_fp16 = slice_by_index(begin = var_19289_begin_0, end = var_19289_end_0, end_mask = var_19289_end_mask_0, x = k_25_cast_fp16)[name = string("op_19289_cast_fp16")]; + tensor var_19293_begin_0 = const()[name = string("op_19293_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19293_end_0 = const()[name = string("op_19293_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_19293_end_mask_0 = const()[name = string("op_19293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19293_cast_fp16 = slice_by_index(begin = var_19293_begin_0, end = var_19293_end_0, end_mask = var_19293_end_mask_0, x = k_25_cast_fp16)[name = string("op_19293_cast_fp16")]; + tensor var_19297_begin_0 = const()[name = string("op_19297_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19297_end_0 = const()[name = string("op_19297_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_19297_end_mask_0 = const()[name = string("op_19297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19297_cast_fp16 = slice_by_index(begin = var_19297_begin_0, end = var_19297_end_0, end_mask = var_19297_end_mask_0, x = k_25_cast_fp16)[name = string("op_19297_cast_fp16")]; + tensor var_19301_begin_0 = const()[name = string("op_19301_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19301_end_0 = const()[name = string("op_19301_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_19301_end_mask_0 = const()[name = string("op_19301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19301_cast_fp16 = slice_by_index(begin = var_19301_begin_0, end = var_19301_end_0, end_mask = var_19301_end_mask_0, x = k_25_cast_fp16)[name = string("op_19301_cast_fp16")]; + tensor var_19303_begin_0 = const()[name = string("op_19303_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19303_end_0 = const()[name = string("op_19303_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19303_end_mask_0 = const()[name = string("op_19303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19303_cast_fp16 = slice_by_index(begin = var_19303_begin_0, end = var_19303_end_0, end_mask = var_19303_end_mask_0, x = value_25_cast_fp16)[name = string("op_19303_cast_fp16")]; + tensor var_19307_begin_0 = const()[name = string("op_19307_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19307_end_0 = const()[name = string("op_19307_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_19307_end_mask_0 = const()[name = string("op_19307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19307_cast_fp16 = slice_by_index(begin = var_19307_begin_0, end = var_19307_end_0, end_mask = var_19307_end_mask_0, x = value_25_cast_fp16)[name = string("op_19307_cast_fp16")]; + tensor var_19311_begin_0 = const()[name = string("op_19311_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19311_end_0 = const()[name = string("op_19311_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_19311_end_mask_0 = const()[name = string("op_19311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19311_cast_fp16 = slice_by_index(begin = var_19311_begin_0, end = var_19311_end_0, end_mask = var_19311_end_mask_0, x = value_25_cast_fp16)[name = string("op_19311_cast_fp16")]; + tensor var_19315_begin_0 = const()[name = string("op_19315_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19315_end_0 = const()[name = string("op_19315_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_19315_end_mask_0 = const()[name = string("op_19315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19315_cast_fp16 = slice_by_index(begin = var_19315_begin_0, end = var_19315_end_0, end_mask = var_19315_end_mask_0, x = value_25_cast_fp16)[name = string("op_19315_cast_fp16")]; + tensor var_19319_begin_0 = const()[name = string("op_19319_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19319_end_0 = const()[name = string("op_19319_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_19319_end_mask_0 = const()[name = string("op_19319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19319_cast_fp16 = slice_by_index(begin = var_19319_begin_0, end = var_19319_end_0, end_mask = var_19319_end_mask_0, x = value_25_cast_fp16)[name = string("op_19319_cast_fp16")]; + tensor var_19323_begin_0 = const()[name = string("op_19323_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19323_end_0 = const()[name = string("op_19323_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_19323_end_mask_0 = const()[name = string("op_19323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19323_cast_fp16 = slice_by_index(begin = var_19323_begin_0, end = var_19323_end_0, end_mask = var_19323_end_mask_0, x = value_25_cast_fp16)[name = string("op_19323_cast_fp16")]; + tensor var_19327_begin_0 = const()[name = string("op_19327_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19327_end_0 = const()[name = string("op_19327_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_19327_end_mask_0 = const()[name = string("op_19327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19327_cast_fp16 = slice_by_index(begin = var_19327_begin_0, end = var_19327_end_0, end_mask = var_19327_end_mask_0, x = value_25_cast_fp16)[name = string("op_19327_cast_fp16")]; + tensor var_19331_begin_0 = const()[name = string("op_19331_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19331_end_0 = const()[name = string("op_19331_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_19331_end_mask_0 = const()[name = string("op_19331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19331_cast_fp16 = slice_by_index(begin = var_19331_begin_0, end = var_19331_end_0, end_mask = var_19331_end_mask_0, x = value_25_cast_fp16)[name = string("op_19331_cast_fp16")]; + tensor var_19335_begin_0 = const()[name = string("op_19335_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19335_end_0 = const()[name = string("op_19335_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_19335_end_mask_0 = const()[name = string("op_19335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19335_cast_fp16 = slice_by_index(begin = var_19335_begin_0, end = var_19335_end_0, end_mask = var_19335_end_mask_0, x = value_25_cast_fp16)[name = string("op_19335_cast_fp16")]; + tensor var_19339_begin_0 = const()[name = string("op_19339_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19339_end_0 = const()[name = string("op_19339_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_19339_end_mask_0 = const()[name = string("op_19339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19339_cast_fp16 = slice_by_index(begin = var_19339_begin_0, end = var_19339_end_0, end_mask = var_19339_end_mask_0, x = value_25_cast_fp16)[name = string("op_19339_cast_fp16")]; + tensor var_19343_begin_0 = const()[name = string("op_19343_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19343_end_0 = const()[name = string("op_19343_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_19343_end_mask_0 = const()[name = string("op_19343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19343_cast_fp16 = slice_by_index(begin = var_19343_begin_0, end = var_19343_end_0, end_mask = var_19343_end_mask_0, x = value_25_cast_fp16)[name = string("op_19343_cast_fp16")]; + tensor var_19347_begin_0 = const()[name = string("op_19347_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19347_end_0 = const()[name = string("op_19347_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_19347_end_mask_0 = const()[name = string("op_19347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19347_cast_fp16 = slice_by_index(begin = var_19347_begin_0, end = var_19347_end_0, end_mask = var_19347_end_mask_0, x = value_25_cast_fp16)[name = string("op_19347_cast_fp16")]; + tensor var_19351_begin_0 = const()[name = string("op_19351_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19351_end_0 = const()[name = string("op_19351_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_19351_end_mask_0 = const()[name = string("op_19351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19351_cast_fp16 = slice_by_index(begin = var_19351_begin_0, end = var_19351_end_0, end_mask = var_19351_end_mask_0, x = value_25_cast_fp16)[name = string("op_19351_cast_fp16")]; + tensor var_19355_begin_0 = const()[name = string("op_19355_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19355_end_0 = const()[name = string("op_19355_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_19355_end_mask_0 = const()[name = string("op_19355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19355_cast_fp16 = slice_by_index(begin = var_19355_begin_0, end = var_19355_end_0, end_mask = var_19355_end_mask_0, x = value_25_cast_fp16)[name = string("op_19355_cast_fp16")]; + tensor var_19359_begin_0 = const()[name = string("op_19359_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19359_end_0 = const()[name = string("op_19359_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_19359_end_mask_0 = const()[name = string("op_19359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19359_cast_fp16 = slice_by_index(begin = var_19359_begin_0, end = var_19359_end_0, end_mask = var_19359_end_mask_0, x = value_25_cast_fp16)[name = string("op_19359_cast_fp16")]; + tensor var_19363_begin_0 = const()[name = string("op_19363_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19363_end_0 = const()[name = string("op_19363_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_19363_end_mask_0 = const()[name = string("op_19363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19363_cast_fp16 = slice_by_index(begin = var_19363_begin_0, end = var_19363_end_0, end_mask = var_19363_end_mask_0, x = value_25_cast_fp16)[name = string("op_19363_cast_fp16")]; + tensor var_19367_begin_0 = const()[name = string("op_19367_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19367_end_0 = const()[name = string("op_19367_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_19367_end_mask_0 = const()[name = string("op_19367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19367_cast_fp16 = slice_by_index(begin = var_19367_begin_0, end = var_19367_end_0, end_mask = var_19367_end_mask_0, x = value_25_cast_fp16)[name = string("op_19367_cast_fp16")]; + tensor var_19371_begin_0 = const()[name = string("op_19371_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19371_end_0 = const()[name = string("op_19371_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_19371_end_mask_0 = const()[name = string("op_19371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19371_cast_fp16 = slice_by_index(begin = var_19371_begin_0, end = var_19371_end_0, end_mask = var_19371_end_mask_0, x = value_25_cast_fp16)[name = string("op_19371_cast_fp16")]; + tensor var_19375_begin_0 = const()[name = string("op_19375_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19375_end_0 = const()[name = string("op_19375_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_19375_end_mask_0 = const()[name = string("op_19375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19375_cast_fp16 = slice_by_index(begin = var_19375_begin_0, end = var_19375_end_0, end_mask = var_19375_end_mask_0, x = value_25_cast_fp16)[name = string("op_19375_cast_fp16")]; + tensor var_19379_begin_0 = const()[name = string("op_19379_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19379_end_0 = const()[name = string("op_19379_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_19379_end_mask_0 = const()[name = string("op_19379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19379_cast_fp16 = slice_by_index(begin = var_19379_begin_0, end = var_19379_end_0, end_mask = var_19379_end_mask_0, x = value_25_cast_fp16)[name = string("op_19379_cast_fp16")]; + string _SplitHeadsQ__mh_w_1921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1921_equation_0, values = (var_19225_cast_fp16, var_18667_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1921_cast_fp16")]; + string _SplitHeadsQ__mh_w_1923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1923_equation_0, values = (var_19225_cast_fp16, var_18674_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1923_cast_fp16")]; + string _SplitHeadsQ__mh_w_1925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1925_equation_0, values = (var_19225_cast_fp16, var_18681_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1925_cast_fp16")]; + string _SplitHeadsQ__mh_w_1927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1927_equation_0, values = (var_19225_cast_fp16, var_18688_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1927_cast_fp16")]; + string _SplitHeadsQ__mh_w_1929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1929_equation_0, values = (var_19229_cast_fp16, var_18695_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1929_cast_fp16")]; + string _SplitHeadsQ__mh_w_1931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1931_equation_0, values = (var_19229_cast_fp16, var_18702_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1931_cast_fp16")]; + string _SplitHeadsQ__mh_w_1933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1933_equation_0, values = (var_19229_cast_fp16, var_18709_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1933_cast_fp16")]; + string _SplitHeadsQ__mh_w_1935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1935_equation_0, values = (var_19229_cast_fp16, var_18716_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1935_cast_fp16")]; + string _SplitHeadsQ__mh_w_1937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1937_equation_0, values = (var_19233_cast_fp16, var_18723_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1937_cast_fp16")]; + string _SplitHeadsQ__mh_w_1939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1939_equation_0, values = (var_19233_cast_fp16, var_18730_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1939_cast_fp16")]; + string _SplitHeadsQ__mh_w_1941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1941_equation_0, values = (var_19233_cast_fp16, var_18737_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1941_cast_fp16")]; + string _SplitHeadsQ__mh_w_1943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1943_equation_0, values = (var_19233_cast_fp16, var_18744_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1943_cast_fp16")]; + string _SplitHeadsQ__mh_w_1945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1945_equation_0, values = (var_19237_cast_fp16, var_18751_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1945_cast_fp16")]; + string _SplitHeadsQ__mh_w_1947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1947_equation_0, values = (var_19237_cast_fp16, var_18758_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1947_cast_fp16")]; + string _SplitHeadsQ__mh_w_1949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1949_equation_0, values = (var_19237_cast_fp16, var_18765_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1949_cast_fp16")]; + string _SplitHeadsQ__mh_w_1951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1951_equation_0, values = (var_19237_cast_fp16, var_18772_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1951_cast_fp16")]; + string _SplitHeadsQ__mh_w_1953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1953_equation_0, values = (var_19241_cast_fp16, var_18779_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1953_cast_fp16")]; + string _SplitHeadsQ__mh_w_1955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1955_equation_0, values = (var_19241_cast_fp16, var_18786_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1955_cast_fp16")]; + string _SplitHeadsQ__mh_w_1957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1957_equation_0, values = (var_19241_cast_fp16, var_18793_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1957_cast_fp16")]; + string _SplitHeadsQ__mh_w_1959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1959_equation_0, values = (var_19241_cast_fp16, var_18800_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1959_cast_fp16")]; + string _SplitHeadsQ__mh_w_1961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1961_equation_0, values = (var_19245_cast_fp16, var_18807_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1961_cast_fp16")]; + string _SplitHeadsQ__mh_w_1963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1963_equation_0, values = (var_19245_cast_fp16, var_18814_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1963_cast_fp16")]; + string _SplitHeadsQ__mh_w_1965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1965_equation_0, values = (var_19245_cast_fp16, var_18821_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1965_cast_fp16")]; + string _SplitHeadsQ__mh_w_1967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1967_equation_0, values = (var_19245_cast_fp16, var_18828_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1967_cast_fp16")]; + string _SplitHeadsQ__mh_w_1969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1969_equation_0, values = (var_19249_cast_fp16, var_18835_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1969_cast_fp16")]; + string _SplitHeadsQ__mh_w_1971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1971_equation_0, values = (var_19249_cast_fp16, var_18842_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1971_cast_fp16")]; + string _SplitHeadsQ__mh_w_1973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1973_equation_0, values = (var_19249_cast_fp16, var_18849_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1973_cast_fp16")]; + string _SplitHeadsQ__mh_w_1975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1975_equation_0, values = (var_19249_cast_fp16, var_18856_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1975_cast_fp16")]; + string _SplitHeadsQ__mh_w_1977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1977_equation_0, values = (var_19253_cast_fp16, var_18863_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1977_cast_fp16")]; + string _SplitHeadsQ__mh_w_1979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1979_equation_0, values = (var_19253_cast_fp16, var_18870_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1979_cast_fp16")]; + string _SplitHeadsQ__mh_w_1981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1981_equation_0, values = (var_19253_cast_fp16, var_18877_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1981_cast_fp16")]; + string _SplitHeadsQ__mh_w_1983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1983_equation_0, values = (var_19253_cast_fp16, var_18884_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1983_cast_fp16")]; + string _SplitHeadsQ__mh_w_1985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1985_equation_0, values = (var_19257_cast_fp16, var_18891_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1985_cast_fp16")]; + string _SplitHeadsQ__mh_w_1987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1987_equation_0, values = (var_19257_cast_fp16, var_18898_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1987_cast_fp16")]; + string _SplitHeadsQ__mh_w_1989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1989_equation_0, values = (var_19257_cast_fp16, var_18905_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1989_cast_fp16")]; + string _SplitHeadsQ__mh_w_1991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1991_equation_0, values = (var_19257_cast_fp16, var_18912_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1991_cast_fp16")]; + string _SplitHeadsQ__mh_w_1993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1993_equation_0, values = (var_19261_cast_fp16, var_18919_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1993_cast_fp16")]; + string _SplitHeadsQ__mh_w_1995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1995_equation_0, values = (var_19261_cast_fp16, var_18926_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1995_cast_fp16")]; + string _SplitHeadsQ__mh_w_1997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1997_equation_0, values = (var_19261_cast_fp16, var_18933_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1997_cast_fp16")]; + string _SplitHeadsQ__mh_w_1999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1999_equation_0, values = (var_19261_cast_fp16, var_18940_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1999_cast_fp16")]; + string _SplitHeadsQ__mh_w_2001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2001_equation_0, values = (var_19265_cast_fp16, var_18947_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2001_cast_fp16")]; + string _SplitHeadsQ__mh_w_2003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2003_equation_0, values = (var_19265_cast_fp16, var_18954_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2003_cast_fp16")]; + string _SplitHeadsQ__mh_w_2005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2005_equation_0, values = (var_19265_cast_fp16, var_18961_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2005_cast_fp16")]; + string _SplitHeadsQ__mh_w_2007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2007_equation_0, values = (var_19265_cast_fp16, var_18968_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2007_cast_fp16")]; + string _SplitHeadsQ__mh_w_2009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2009_equation_0, values = (var_19269_cast_fp16, var_18975_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2009_cast_fp16")]; + string _SplitHeadsQ__mh_w_2011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2011_equation_0, values = (var_19269_cast_fp16, var_18982_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2011_cast_fp16")]; + string _SplitHeadsQ__mh_w_2013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2013_equation_0, values = (var_19269_cast_fp16, var_18989_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2013_cast_fp16")]; + string _SplitHeadsQ__mh_w_2015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2015_equation_0, values = (var_19269_cast_fp16, var_18996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2015_cast_fp16")]; + string _SplitHeadsQ__mh_w_2017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2017_equation_0, values = (var_19273_cast_fp16, var_19003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2017_cast_fp16")]; + string _SplitHeadsQ__mh_w_2019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2019_equation_0, values = (var_19273_cast_fp16, var_19010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2019_cast_fp16")]; + string _SplitHeadsQ__mh_w_2021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2021_equation_0, values = (var_19273_cast_fp16, var_19017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2021_cast_fp16")]; + string _SplitHeadsQ__mh_w_2023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2023_equation_0, values = (var_19273_cast_fp16, var_19024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2023_cast_fp16")]; + string _SplitHeadsQ__mh_w_2025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2025_equation_0, values = (var_19277_cast_fp16, var_19031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2025_cast_fp16")]; + string _SplitHeadsQ__mh_w_2027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2027_equation_0, values = (var_19277_cast_fp16, var_19038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2027_cast_fp16")]; + string _SplitHeadsQ__mh_w_2029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2029_equation_0, values = (var_19277_cast_fp16, var_19045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2029_cast_fp16")]; + string _SplitHeadsQ__mh_w_2031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2031_equation_0, values = (var_19277_cast_fp16, var_19052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2031_cast_fp16")]; + string _SplitHeadsQ__mh_w_2033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2033_equation_0, values = (var_19281_cast_fp16, var_19059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2033_cast_fp16")]; + string _SplitHeadsQ__mh_w_2035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2035_equation_0, values = (var_19281_cast_fp16, var_19066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2035_cast_fp16")]; + string _SplitHeadsQ__mh_w_2037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2037_equation_0, values = (var_19281_cast_fp16, var_19073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2037_cast_fp16")]; + string _SplitHeadsQ__mh_w_2039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2039_equation_0, values = (var_19281_cast_fp16, var_19080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2039_cast_fp16")]; + string _SplitHeadsQ__mh_w_2041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2041_equation_0, values = (var_19285_cast_fp16, var_19087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2041_cast_fp16")]; + string _SplitHeadsQ__mh_w_2043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2043_equation_0, values = (var_19285_cast_fp16, var_19094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2043_cast_fp16")]; + string _SplitHeadsQ__mh_w_2045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2045_equation_0, values = (var_19285_cast_fp16, var_19101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2045_cast_fp16")]; + string _SplitHeadsQ__mh_w_2047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2047_equation_0, values = (var_19285_cast_fp16, var_19108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2047_cast_fp16")]; + string _SplitHeadsQ__mh_w_2049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2049_equation_0, values = (var_19289_cast_fp16, var_19115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2049_cast_fp16")]; + string _SplitHeadsQ__mh_w_2051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2051_equation_0, values = (var_19289_cast_fp16, var_19122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2051_cast_fp16")]; + string _SplitHeadsQ__mh_w_2053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2053_equation_0, values = (var_19289_cast_fp16, var_19129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2053_cast_fp16")]; + string _SplitHeadsQ__mh_w_2055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2055_equation_0, values = (var_19289_cast_fp16, var_19136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2055_cast_fp16")]; + string _SplitHeadsQ__mh_w_2057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2057_equation_0, values = (var_19293_cast_fp16, var_19143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2057_cast_fp16")]; + string _SplitHeadsQ__mh_w_2059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2059_equation_0, values = (var_19293_cast_fp16, var_19150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2059_cast_fp16")]; + string _SplitHeadsQ__mh_w_2061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2061_equation_0, values = (var_19293_cast_fp16, var_19157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2061_cast_fp16")]; + string _SplitHeadsQ__mh_w_2063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2063_equation_0, values = (var_19293_cast_fp16, var_19164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2063_cast_fp16")]; + string _SplitHeadsQ__mh_w_2065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2065_equation_0, values = (var_19297_cast_fp16, var_19171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2065_cast_fp16")]; + string _SplitHeadsQ__mh_w_2067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2067_equation_0, values = (var_19297_cast_fp16, var_19178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2067_cast_fp16")]; + string _SplitHeadsQ__mh_w_2069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2069_equation_0, values = (var_19297_cast_fp16, var_19185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2069_cast_fp16")]; + string _SplitHeadsQ__mh_w_2071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2071_equation_0, values = (var_19297_cast_fp16, var_19192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2071_cast_fp16")]; + string _SplitHeadsQ__mh_w_2073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2073_equation_0, values = (var_19301_cast_fp16, var_19199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2073_cast_fp16")]; + string _SplitHeadsQ__mh_w_2075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2075_equation_0, values = (var_19301_cast_fp16, var_19206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2075_cast_fp16")]; + string _SplitHeadsQ__mh_w_2077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2077_equation_0, values = (var_19301_cast_fp16, var_19213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2077_cast_fp16")]; + string _SplitHeadsQ__mh_w_2079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2079_equation_0, values = (var_19301_cast_fp16, var_19220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2079_cast_fp16")]; + fp16 var_19542_to_fp16 = const()[name = string("op_19542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1921_cast_fp16, y = var_19542_to_fp16)[name = string("aw_chunk_1921_cast_fp16")]; + fp16 var_19544_to_fp16 = const()[name = string("op_19544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1923_cast_fp16, y = var_19544_to_fp16)[name = string("aw_chunk_1923_cast_fp16")]; + fp16 var_19546_to_fp16 = const()[name = string("op_19546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1925_cast_fp16, y = var_19546_to_fp16)[name = string("aw_chunk_1925_cast_fp16")]; + fp16 var_19548_to_fp16 = const()[name = string("op_19548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1927_cast_fp16, y = var_19548_to_fp16)[name = string("aw_chunk_1927_cast_fp16")]; + fp16 var_19550_to_fp16 = const()[name = string("op_19550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1929_cast_fp16, y = var_19550_to_fp16)[name = string("aw_chunk_1929_cast_fp16")]; + fp16 var_19552_to_fp16 = const()[name = string("op_19552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1931_cast_fp16, y = var_19552_to_fp16)[name = string("aw_chunk_1931_cast_fp16")]; + fp16 var_19554_to_fp16 = const()[name = string("op_19554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1933_cast_fp16, y = var_19554_to_fp16)[name = string("aw_chunk_1933_cast_fp16")]; + fp16 var_19556_to_fp16 = const()[name = string("op_19556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1935_cast_fp16, y = var_19556_to_fp16)[name = string("aw_chunk_1935_cast_fp16")]; + fp16 var_19558_to_fp16 = const()[name = string("op_19558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1937_cast_fp16, y = var_19558_to_fp16)[name = string("aw_chunk_1937_cast_fp16")]; + fp16 var_19560_to_fp16 = const()[name = string("op_19560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1939_cast_fp16, y = var_19560_to_fp16)[name = string("aw_chunk_1939_cast_fp16")]; + fp16 var_19562_to_fp16 = const()[name = string("op_19562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1941_cast_fp16, y = var_19562_to_fp16)[name = string("aw_chunk_1941_cast_fp16")]; + fp16 var_19564_to_fp16 = const()[name = string("op_19564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1943_cast_fp16, y = var_19564_to_fp16)[name = string("aw_chunk_1943_cast_fp16")]; + fp16 var_19566_to_fp16 = const()[name = string("op_19566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1945_cast_fp16, y = var_19566_to_fp16)[name = string("aw_chunk_1945_cast_fp16")]; + fp16 var_19568_to_fp16 = const()[name = string("op_19568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1947_cast_fp16, y = var_19568_to_fp16)[name = string("aw_chunk_1947_cast_fp16")]; + fp16 var_19570_to_fp16 = const()[name = string("op_19570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1949_cast_fp16, y = var_19570_to_fp16)[name = string("aw_chunk_1949_cast_fp16")]; + fp16 var_19572_to_fp16 = const()[name = string("op_19572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1951_cast_fp16, y = var_19572_to_fp16)[name = string("aw_chunk_1951_cast_fp16")]; + fp16 var_19574_to_fp16 = const()[name = string("op_19574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1953_cast_fp16, y = var_19574_to_fp16)[name = string("aw_chunk_1953_cast_fp16")]; + fp16 var_19576_to_fp16 = const()[name = string("op_19576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1955_cast_fp16, y = var_19576_to_fp16)[name = string("aw_chunk_1955_cast_fp16")]; + fp16 var_19578_to_fp16 = const()[name = string("op_19578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1957_cast_fp16, y = var_19578_to_fp16)[name = string("aw_chunk_1957_cast_fp16")]; + fp16 var_19580_to_fp16 = const()[name = string("op_19580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1959_cast_fp16, y = var_19580_to_fp16)[name = string("aw_chunk_1959_cast_fp16")]; + fp16 var_19582_to_fp16 = const()[name = string("op_19582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1961_cast_fp16, y = var_19582_to_fp16)[name = string("aw_chunk_1961_cast_fp16")]; + fp16 var_19584_to_fp16 = const()[name = string("op_19584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1963_cast_fp16, y = var_19584_to_fp16)[name = string("aw_chunk_1963_cast_fp16")]; + fp16 var_19586_to_fp16 = const()[name = string("op_19586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1965_cast_fp16, y = var_19586_to_fp16)[name = string("aw_chunk_1965_cast_fp16")]; + fp16 var_19588_to_fp16 = const()[name = string("op_19588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1967_cast_fp16, y = var_19588_to_fp16)[name = string("aw_chunk_1967_cast_fp16")]; + fp16 var_19590_to_fp16 = const()[name = string("op_19590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1969_cast_fp16, y = var_19590_to_fp16)[name = string("aw_chunk_1969_cast_fp16")]; + fp16 var_19592_to_fp16 = const()[name = string("op_19592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1971_cast_fp16, y = var_19592_to_fp16)[name = string("aw_chunk_1971_cast_fp16")]; + fp16 var_19594_to_fp16 = const()[name = string("op_19594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1973_cast_fp16, y = var_19594_to_fp16)[name = string("aw_chunk_1973_cast_fp16")]; + fp16 var_19596_to_fp16 = const()[name = string("op_19596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1975_cast_fp16, y = var_19596_to_fp16)[name = string("aw_chunk_1975_cast_fp16")]; + fp16 var_19598_to_fp16 = const()[name = string("op_19598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1977_cast_fp16, y = var_19598_to_fp16)[name = string("aw_chunk_1977_cast_fp16")]; + fp16 var_19600_to_fp16 = const()[name = string("op_19600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1979_cast_fp16, y = var_19600_to_fp16)[name = string("aw_chunk_1979_cast_fp16")]; + fp16 var_19602_to_fp16 = const()[name = string("op_19602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1981_cast_fp16, y = var_19602_to_fp16)[name = string("aw_chunk_1981_cast_fp16")]; + fp16 var_19604_to_fp16 = const()[name = string("op_19604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1983_cast_fp16, y = var_19604_to_fp16)[name = string("aw_chunk_1983_cast_fp16")]; + fp16 var_19606_to_fp16 = const()[name = string("op_19606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1985_cast_fp16, y = var_19606_to_fp16)[name = string("aw_chunk_1985_cast_fp16")]; + fp16 var_19608_to_fp16 = const()[name = string("op_19608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1987_cast_fp16, y = var_19608_to_fp16)[name = string("aw_chunk_1987_cast_fp16")]; + fp16 var_19610_to_fp16 = const()[name = string("op_19610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1989_cast_fp16, y = var_19610_to_fp16)[name = string("aw_chunk_1989_cast_fp16")]; + fp16 var_19612_to_fp16 = const()[name = string("op_19612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1991_cast_fp16, y = var_19612_to_fp16)[name = string("aw_chunk_1991_cast_fp16")]; + fp16 var_19614_to_fp16 = const()[name = string("op_19614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1993_cast_fp16, y = var_19614_to_fp16)[name = string("aw_chunk_1993_cast_fp16")]; + fp16 var_19616_to_fp16 = const()[name = string("op_19616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1995_cast_fp16, y = var_19616_to_fp16)[name = string("aw_chunk_1995_cast_fp16")]; + fp16 var_19618_to_fp16 = const()[name = string("op_19618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1997_cast_fp16, y = var_19618_to_fp16)[name = string("aw_chunk_1997_cast_fp16")]; + fp16 var_19620_to_fp16 = const()[name = string("op_19620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1999_cast_fp16, y = var_19620_to_fp16)[name = string("aw_chunk_1999_cast_fp16")]; + fp16 var_19622_to_fp16 = const()[name = string("op_19622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2001_cast_fp16, y = var_19622_to_fp16)[name = string("aw_chunk_2001_cast_fp16")]; + fp16 var_19624_to_fp16 = const()[name = string("op_19624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2003_cast_fp16, y = var_19624_to_fp16)[name = string("aw_chunk_2003_cast_fp16")]; + fp16 var_19626_to_fp16 = const()[name = string("op_19626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2005_cast_fp16, y = var_19626_to_fp16)[name = string("aw_chunk_2005_cast_fp16")]; + fp16 var_19628_to_fp16 = const()[name = string("op_19628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2007_cast_fp16, y = var_19628_to_fp16)[name = string("aw_chunk_2007_cast_fp16")]; + fp16 var_19630_to_fp16 = const()[name = string("op_19630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2009_cast_fp16, y = var_19630_to_fp16)[name = string("aw_chunk_2009_cast_fp16")]; + fp16 var_19632_to_fp16 = const()[name = string("op_19632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2011_cast_fp16, y = var_19632_to_fp16)[name = string("aw_chunk_2011_cast_fp16")]; + fp16 var_19634_to_fp16 = const()[name = string("op_19634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2013_cast_fp16, y = var_19634_to_fp16)[name = string("aw_chunk_2013_cast_fp16")]; + fp16 var_19636_to_fp16 = const()[name = string("op_19636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2015_cast_fp16, y = var_19636_to_fp16)[name = string("aw_chunk_2015_cast_fp16")]; + fp16 var_19638_to_fp16 = const()[name = string("op_19638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2017_cast_fp16, y = var_19638_to_fp16)[name = string("aw_chunk_2017_cast_fp16")]; + fp16 var_19640_to_fp16 = const()[name = string("op_19640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2019_cast_fp16, y = var_19640_to_fp16)[name = string("aw_chunk_2019_cast_fp16")]; + fp16 var_19642_to_fp16 = const()[name = string("op_19642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2021_cast_fp16, y = var_19642_to_fp16)[name = string("aw_chunk_2021_cast_fp16")]; + fp16 var_19644_to_fp16 = const()[name = string("op_19644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2023_cast_fp16, y = var_19644_to_fp16)[name = string("aw_chunk_2023_cast_fp16")]; + fp16 var_19646_to_fp16 = const()[name = string("op_19646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2025_cast_fp16, y = var_19646_to_fp16)[name = string("aw_chunk_2025_cast_fp16")]; + fp16 var_19648_to_fp16 = const()[name = string("op_19648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2027_cast_fp16, y = var_19648_to_fp16)[name = string("aw_chunk_2027_cast_fp16")]; + fp16 var_19650_to_fp16 = const()[name = string("op_19650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2029_cast_fp16, y = var_19650_to_fp16)[name = string("aw_chunk_2029_cast_fp16")]; + fp16 var_19652_to_fp16 = const()[name = string("op_19652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2031_cast_fp16, y = var_19652_to_fp16)[name = string("aw_chunk_2031_cast_fp16")]; + fp16 var_19654_to_fp16 = const()[name = string("op_19654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2033_cast_fp16, y = var_19654_to_fp16)[name = string("aw_chunk_2033_cast_fp16")]; + fp16 var_19656_to_fp16 = const()[name = string("op_19656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2035_cast_fp16, y = var_19656_to_fp16)[name = string("aw_chunk_2035_cast_fp16")]; + fp16 var_19658_to_fp16 = const()[name = string("op_19658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2037_cast_fp16, y = var_19658_to_fp16)[name = string("aw_chunk_2037_cast_fp16")]; + fp16 var_19660_to_fp16 = const()[name = string("op_19660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2039_cast_fp16, y = var_19660_to_fp16)[name = string("aw_chunk_2039_cast_fp16")]; + fp16 var_19662_to_fp16 = const()[name = string("op_19662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2041_cast_fp16, y = var_19662_to_fp16)[name = string("aw_chunk_2041_cast_fp16")]; + fp16 var_19664_to_fp16 = const()[name = string("op_19664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2043_cast_fp16, y = var_19664_to_fp16)[name = string("aw_chunk_2043_cast_fp16")]; + fp16 var_19666_to_fp16 = const()[name = string("op_19666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2045_cast_fp16, y = var_19666_to_fp16)[name = string("aw_chunk_2045_cast_fp16")]; + fp16 var_19668_to_fp16 = const()[name = string("op_19668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2047_cast_fp16, y = var_19668_to_fp16)[name = string("aw_chunk_2047_cast_fp16")]; + fp16 var_19670_to_fp16 = const()[name = string("op_19670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2049_cast_fp16, y = var_19670_to_fp16)[name = string("aw_chunk_2049_cast_fp16")]; + fp16 var_19672_to_fp16 = const()[name = string("op_19672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2051_cast_fp16, y = var_19672_to_fp16)[name = string("aw_chunk_2051_cast_fp16")]; + fp16 var_19674_to_fp16 = const()[name = string("op_19674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2053_cast_fp16, y = var_19674_to_fp16)[name = string("aw_chunk_2053_cast_fp16")]; + fp16 var_19676_to_fp16 = const()[name = string("op_19676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2055_cast_fp16, y = var_19676_to_fp16)[name = string("aw_chunk_2055_cast_fp16")]; + fp16 var_19678_to_fp16 = const()[name = string("op_19678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2057_cast_fp16, y = var_19678_to_fp16)[name = string("aw_chunk_2057_cast_fp16")]; + fp16 var_19680_to_fp16 = const()[name = string("op_19680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2059_cast_fp16, y = var_19680_to_fp16)[name = string("aw_chunk_2059_cast_fp16")]; + fp16 var_19682_to_fp16 = const()[name = string("op_19682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2061_cast_fp16, y = var_19682_to_fp16)[name = string("aw_chunk_2061_cast_fp16")]; + fp16 var_19684_to_fp16 = const()[name = string("op_19684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2063_cast_fp16, y = var_19684_to_fp16)[name = string("aw_chunk_2063_cast_fp16")]; + fp16 var_19686_to_fp16 = const()[name = string("op_19686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2065_cast_fp16, y = var_19686_to_fp16)[name = string("aw_chunk_2065_cast_fp16")]; + fp16 var_19688_to_fp16 = const()[name = string("op_19688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2067_cast_fp16, y = var_19688_to_fp16)[name = string("aw_chunk_2067_cast_fp16")]; + fp16 var_19690_to_fp16 = const()[name = string("op_19690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2069_cast_fp16, y = var_19690_to_fp16)[name = string("aw_chunk_2069_cast_fp16")]; + fp16 var_19692_to_fp16 = const()[name = string("op_19692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2071_cast_fp16, y = var_19692_to_fp16)[name = string("aw_chunk_2071_cast_fp16")]; + fp16 var_19694_to_fp16 = const()[name = string("op_19694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2073_cast_fp16, y = var_19694_to_fp16)[name = string("aw_chunk_2073_cast_fp16")]; + fp16 var_19696_to_fp16 = const()[name = string("op_19696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2075_cast_fp16, y = var_19696_to_fp16)[name = string("aw_chunk_2075_cast_fp16")]; + fp16 var_19698_to_fp16 = const()[name = string("op_19698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2077_cast_fp16, y = var_19698_to_fp16)[name = string("aw_chunk_2077_cast_fp16")]; + fp16 var_19700_to_fp16 = const()[name = string("op_19700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2079_cast_fp16, y = var_19700_to_fp16)[name = string("aw_chunk_2079_cast_fp16")]; + tensor var_19702_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1921_cast_fp16)[name = string("op_19702_cast_fp16")]; + tensor var_19703_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1923_cast_fp16)[name = string("op_19703_cast_fp16")]; + tensor var_19704_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1925_cast_fp16)[name = string("op_19704_cast_fp16")]; + tensor var_19705_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1927_cast_fp16)[name = string("op_19705_cast_fp16")]; + tensor var_19706_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1929_cast_fp16)[name = string("op_19706_cast_fp16")]; + tensor var_19707_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1931_cast_fp16)[name = string("op_19707_cast_fp16")]; + tensor var_19708_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1933_cast_fp16)[name = string("op_19708_cast_fp16")]; + tensor var_19709_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1935_cast_fp16)[name = string("op_19709_cast_fp16")]; + tensor var_19710_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1937_cast_fp16)[name = string("op_19710_cast_fp16")]; + tensor var_19711_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1939_cast_fp16)[name = string("op_19711_cast_fp16")]; + tensor var_19712_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1941_cast_fp16)[name = string("op_19712_cast_fp16")]; + tensor var_19713_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1943_cast_fp16)[name = string("op_19713_cast_fp16")]; + tensor var_19714_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1945_cast_fp16)[name = string("op_19714_cast_fp16")]; + tensor var_19715_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1947_cast_fp16)[name = string("op_19715_cast_fp16")]; + tensor var_19716_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1949_cast_fp16)[name = string("op_19716_cast_fp16")]; + tensor var_19717_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1951_cast_fp16)[name = string("op_19717_cast_fp16")]; + tensor var_19718_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1953_cast_fp16)[name = string("op_19718_cast_fp16")]; + tensor var_19719_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1955_cast_fp16)[name = string("op_19719_cast_fp16")]; + tensor var_19720_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1957_cast_fp16)[name = string("op_19720_cast_fp16")]; + tensor var_19721_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1959_cast_fp16)[name = string("op_19721_cast_fp16")]; + tensor var_19722_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1961_cast_fp16)[name = string("op_19722_cast_fp16")]; + tensor var_19723_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1963_cast_fp16)[name = string("op_19723_cast_fp16")]; + tensor var_19724_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1965_cast_fp16)[name = string("op_19724_cast_fp16")]; + tensor var_19725_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1967_cast_fp16)[name = string("op_19725_cast_fp16")]; + tensor var_19726_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1969_cast_fp16)[name = string("op_19726_cast_fp16")]; + tensor var_19727_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1971_cast_fp16)[name = string("op_19727_cast_fp16")]; + tensor var_19728_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1973_cast_fp16)[name = string("op_19728_cast_fp16")]; + tensor var_19729_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1975_cast_fp16)[name = string("op_19729_cast_fp16")]; + tensor var_19730_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1977_cast_fp16)[name = string("op_19730_cast_fp16")]; + tensor var_19731_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1979_cast_fp16)[name = string("op_19731_cast_fp16")]; + tensor var_19732_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1981_cast_fp16)[name = string("op_19732_cast_fp16")]; + tensor var_19733_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1983_cast_fp16)[name = string("op_19733_cast_fp16")]; + tensor var_19734_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1985_cast_fp16)[name = string("op_19734_cast_fp16")]; + tensor var_19735_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1987_cast_fp16)[name = string("op_19735_cast_fp16")]; + tensor var_19736_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1989_cast_fp16)[name = string("op_19736_cast_fp16")]; + tensor var_19737_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1991_cast_fp16)[name = string("op_19737_cast_fp16")]; + tensor var_19738_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1993_cast_fp16)[name = string("op_19738_cast_fp16")]; + tensor var_19739_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1995_cast_fp16)[name = string("op_19739_cast_fp16")]; + tensor var_19740_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1997_cast_fp16)[name = string("op_19740_cast_fp16")]; + tensor var_19741_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_1999_cast_fp16)[name = string("op_19741_cast_fp16")]; + tensor var_19742_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2001_cast_fp16)[name = string("op_19742_cast_fp16")]; + tensor var_19743_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2003_cast_fp16)[name = string("op_19743_cast_fp16")]; + tensor var_19744_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2005_cast_fp16)[name = string("op_19744_cast_fp16")]; + tensor var_19745_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2007_cast_fp16)[name = string("op_19745_cast_fp16")]; + tensor var_19746_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2009_cast_fp16)[name = string("op_19746_cast_fp16")]; + tensor var_19747_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2011_cast_fp16)[name = string("op_19747_cast_fp16")]; + tensor var_19748_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2013_cast_fp16)[name = string("op_19748_cast_fp16")]; + tensor var_19749_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2015_cast_fp16)[name = string("op_19749_cast_fp16")]; + tensor var_19750_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2017_cast_fp16)[name = string("op_19750_cast_fp16")]; + tensor var_19751_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2019_cast_fp16)[name = string("op_19751_cast_fp16")]; + tensor var_19752_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2021_cast_fp16)[name = string("op_19752_cast_fp16")]; + tensor var_19753_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2023_cast_fp16)[name = string("op_19753_cast_fp16")]; + tensor var_19754_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2025_cast_fp16)[name = string("op_19754_cast_fp16")]; + tensor var_19755_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2027_cast_fp16)[name = string("op_19755_cast_fp16")]; + tensor var_19756_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2029_cast_fp16)[name = string("op_19756_cast_fp16")]; + tensor var_19757_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2031_cast_fp16)[name = string("op_19757_cast_fp16")]; + tensor var_19758_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2033_cast_fp16)[name = string("op_19758_cast_fp16")]; + tensor var_19759_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2035_cast_fp16)[name = string("op_19759_cast_fp16")]; + tensor var_19760_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2037_cast_fp16)[name = string("op_19760_cast_fp16")]; + tensor var_19761_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2039_cast_fp16)[name = string("op_19761_cast_fp16")]; + tensor var_19762_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2041_cast_fp16)[name = string("op_19762_cast_fp16")]; + tensor var_19763_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2043_cast_fp16)[name = string("op_19763_cast_fp16")]; + tensor var_19764_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2045_cast_fp16)[name = string("op_19764_cast_fp16")]; + tensor var_19765_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2047_cast_fp16)[name = string("op_19765_cast_fp16")]; + tensor var_19766_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2049_cast_fp16)[name = string("op_19766_cast_fp16")]; + tensor var_19767_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2051_cast_fp16)[name = string("op_19767_cast_fp16")]; + tensor var_19768_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2053_cast_fp16)[name = string("op_19768_cast_fp16")]; + tensor var_19769_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2055_cast_fp16)[name = string("op_19769_cast_fp16")]; + tensor var_19770_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2057_cast_fp16)[name = string("op_19770_cast_fp16")]; + tensor var_19771_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2059_cast_fp16)[name = string("op_19771_cast_fp16")]; + tensor var_19772_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2061_cast_fp16)[name = string("op_19772_cast_fp16")]; + tensor var_19773_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2063_cast_fp16)[name = string("op_19773_cast_fp16")]; + tensor var_19774_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2065_cast_fp16)[name = string("op_19774_cast_fp16")]; + tensor var_19775_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2067_cast_fp16)[name = string("op_19775_cast_fp16")]; + tensor var_19776_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2069_cast_fp16)[name = string("op_19776_cast_fp16")]; + tensor var_19777_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2071_cast_fp16)[name = string("op_19777_cast_fp16")]; + tensor var_19778_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2073_cast_fp16)[name = string("op_19778_cast_fp16")]; + tensor var_19779_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2075_cast_fp16)[name = string("op_19779_cast_fp16")]; + tensor var_19780_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2077_cast_fp16)[name = string("op_19780_cast_fp16")]; + tensor var_19781_cast_fp16 = softmax(axis = var_18527, x = aw_chunk_2079_cast_fp16)[name = string("op_19781_cast_fp16")]; + string var_19783_equation_0 = const()[name = string("op_19783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19783_cast_fp16 = einsum(equation = var_19783_equation_0, values = (var_19303_cast_fp16, var_19702_cast_fp16))[name = string("op_19783_cast_fp16")]; + string var_19785_equation_0 = const()[name = string("op_19785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19785_cast_fp16 = einsum(equation = var_19785_equation_0, values = (var_19303_cast_fp16, var_19703_cast_fp16))[name = string("op_19785_cast_fp16")]; + string var_19787_equation_0 = const()[name = string("op_19787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19787_cast_fp16 = einsum(equation = var_19787_equation_0, values = (var_19303_cast_fp16, var_19704_cast_fp16))[name = string("op_19787_cast_fp16")]; + string var_19789_equation_0 = const()[name = string("op_19789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19789_cast_fp16 = einsum(equation = var_19789_equation_0, values = (var_19303_cast_fp16, var_19705_cast_fp16))[name = string("op_19789_cast_fp16")]; + string var_19791_equation_0 = const()[name = string("op_19791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19791_cast_fp16 = einsum(equation = var_19791_equation_0, values = (var_19307_cast_fp16, var_19706_cast_fp16))[name = string("op_19791_cast_fp16")]; + string var_19793_equation_0 = const()[name = string("op_19793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19793_cast_fp16 = einsum(equation = var_19793_equation_0, values = (var_19307_cast_fp16, var_19707_cast_fp16))[name = string("op_19793_cast_fp16")]; + string var_19795_equation_0 = const()[name = string("op_19795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19795_cast_fp16 = einsum(equation = var_19795_equation_0, values = (var_19307_cast_fp16, var_19708_cast_fp16))[name = string("op_19795_cast_fp16")]; + string var_19797_equation_0 = const()[name = string("op_19797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19797_cast_fp16 = einsum(equation = var_19797_equation_0, values = (var_19307_cast_fp16, var_19709_cast_fp16))[name = string("op_19797_cast_fp16")]; + string var_19799_equation_0 = const()[name = string("op_19799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19799_cast_fp16 = einsum(equation = var_19799_equation_0, values = (var_19311_cast_fp16, var_19710_cast_fp16))[name = string("op_19799_cast_fp16")]; + string var_19801_equation_0 = const()[name = string("op_19801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19801_cast_fp16 = einsum(equation = var_19801_equation_0, values = (var_19311_cast_fp16, var_19711_cast_fp16))[name = string("op_19801_cast_fp16")]; + string var_19803_equation_0 = const()[name = string("op_19803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19803_cast_fp16 = einsum(equation = var_19803_equation_0, values = (var_19311_cast_fp16, var_19712_cast_fp16))[name = string("op_19803_cast_fp16")]; + string var_19805_equation_0 = const()[name = string("op_19805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19805_cast_fp16 = einsum(equation = var_19805_equation_0, values = (var_19311_cast_fp16, var_19713_cast_fp16))[name = string("op_19805_cast_fp16")]; + string var_19807_equation_0 = const()[name = string("op_19807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19807_cast_fp16 = einsum(equation = var_19807_equation_0, values = (var_19315_cast_fp16, var_19714_cast_fp16))[name = string("op_19807_cast_fp16")]; + string var_19809_equation_0 = const()[name = string("op_19809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19809_cast_fp16 = einsum(equation = var_19809_equation_0, values = (var_19315_cast_fp16, var_19715_cast_fp16))[name = string("op_19809_cast_fp16")]; + string var_19811_equation_0 = const()[name = string("op_19811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19811_cast_fp16 = einsum(equation = var_19811_equation_0, values = (var_19315_cast_fp16, var_19716_cast_fp16))[name = string("op_19811_cast_fp16")]; + string var_19813_equation_0 = const()[name = string("op_19813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19813_cast_fp16 = einsum(equation = var_19813_equation_0, values = (var_19315_cast_fp16, var_19717_cast_fp16))[name = string("op_19813_cast_fp16")]; + string var_19815_equation_0 = const()[name = string("op_19815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19815_cast_fp16 = einsum(equation = var_19815_equation_0, values = (var_19319_cast_fp16, var_19718_cast_fp16))[name = string("op_19815_cast_fp16")]; + string var_19817_equation_0 = const()[name = string("op_19817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19817_cast_fp16 = einsum(equation = var_19817_equation_0, values = (var_19319_cast_fp16, var_19719_cast_fp16))[name = string("op_19817_cast_fp16")]; + string var_19819_equation_0 = const()[name = string("op_19819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19819_cast_fp16 = einsum(equation = var_19819_equation_0, values = (var_19319_cast_fp16, var_19720_cast_fp16))[name = string("op_19819_cast_fp16")]; + string var_19821_equation_0 = const()[name = string("op_19821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19821_cast_fp16 = einsum(equation = var_19821_equation_0, values = (var_19319_cast_fp16, var_19721_cast_fp16))[name = string("op_19821_cast_fp16")]; + string var_19823_equation_0 = const()[name = string("op_19823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19823_cast_fp16 = einsum(equation = var_19823_equation_0, values = (var_19323_cast_fp16, var_19722_cast_fp16))[name = string("op_19823_cast_fp16")]; + string var_19825_equation_0 = const()[name = string("op_19825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19825_cast_fp16 = einsum(equation = var_19825_equation_0, values = (var_19323_cast_fp16, var_19723_cast_fp16))[name = string("op_19825_cast_fp16")]; + string var_19827_equation_0 = const()[name = string("op_19827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19827_cast_fp16 = einsum(equation = var_19827_equation_0, values = (var_19323_cast_fp16, var_19724_cast_fp16))[name = string("op_19827_cast_fp16")]; + string var_19829_equation_0 = const()[name = string("op_19829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19829_cast_fp16 = einsum(equation = var_19829_equation_0, values = (var_19323_cast_fp16, var_19725_cast_fp16))[name = string("op_19829_cast_fp16")]; + string var_19831_equation_0 = const()[name = string("op_19831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19831_cast_fp16 = einsum(equation = var_19831_equation_0, values = (var_19327_cast_fp16, var_19726_cast_fp16))[name = string("op_19831_cast_fp16")]; + string var_19833_equation_0 = const()[name = string("op_19833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19833_cast_fp16 = einsum(equation = var_19833_equation_0, values = (var_19327_cast_fp16, var_19727_cast_fp16))[name = string("op_19833_cast_fp16")]; + string var_19835_equation_0 = const()[name = string("op_19835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19835_cast_fp16 = einsum(equation = var_19835_equation_0, values = (var_19327_cast_fp16, var_19728_cast_fp16))[name = string("op_19835_cast_fp16")]; + string var_19837_equation_0 = const()[name = string("op_19837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19837_cast_fp16 = einsum(equation = var_19837_equation_0, values = (var_19327_cast_fp16, var_19729_cast_fp16))[name = string("op_19837_cast_fp16")]; + string var_19839_equation_0 = const()[name = string("op_19839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19839_cast_fp16 = einsum(equation = var_19839_equation_0, values = (var_19331_cast_fp16, var_19730_cast_fp16))[name = string("op_19839_cast_fp16")]; + string var_19841_equation_0 = const()[name = string("op_19841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19841_cast_fp16 = einsum(equation = var_19841_equation_0, values = (var_19331_cast_fp16, var_19731_cast_fp16))[name = string("op_19841_cast_fp16")]; + string var_19843_equation_0 = const()[name = string("op_19843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19843_cast_fp16 = einsum(equation = var_19843_equation_0, values = (var_19331_cast_fp16, var_19732_cast_fp16))[name = string("op_19843_cast_fp16")]; + string var_19845_equation_0 = const()[name = string("op_19845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19845_cast_fp16 = einsum(equation = var_19845_equation_0, values = (var_19331_cast_fp16, var_19733_cast_fp16))[name = string("op_19845_cast_fp16")]; + string var_19847_equation_0 = const()[name = string("op_19847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19847_cast_fp16 = einsum(equation = var_19847_equation_0, values = (var_19335_cast_fp16, var_19734_cast_fp16))[name = string("op_19847_cast_fp16")]; + string var_19849_equation_0 = const()[name = string("op_19849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19849_cast_fp16 = einsum(equation = var_19849_equation_0, values = (var_19335_cast_fp16, var_19735_cast_fp16))[name = string("op_19849_cast_fp16")]; + string var_19851_equation_0 = const()[name = string("op_19851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19851_cast_fp16 = einsum(equation = var_19851_equation_0, values = (var_19335_cast_fp16, var_19736_cast_fp16))[name = string("op_19851_cast_fp16")]; + string var_19853_equation_0 = const()[name = string("op_19853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19853_cast_fp16 = einsum(equation = var_19853_equation_0, values = (var_19335_cast_fp16, var_19737_cast_fp16))[name = string("op_19853_cast_fp16")]; + string var_19855_equation_0 = const()[name = string("op_19855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19855_cast_fp16 = einsum(equation = var_19855_equation_0, values = (var_19339_cast_fp16, var_19738_cast_fp16))[name = string("op_19855_cast_fp16")]; + string var_19857_equation_0 = const()[name = string("op_19857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19857_cast_fp16 = einsum(equation = var_19857_equation_0, values = (var_19339_cast_fp16, var_19739_cast_fp16))[name = string("op_19857_cast_fp16")]; + string var_19859_equation_0 = const()[name = string("op_19859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19859_cast_fp16 = einsum(equation = var_19859_equation_0, values = (var_19339_cast_fp16, var_19740_cast_fp16))[name = string("op_19859_cast_fp16")]; + string var_19861_equation_0 = const()[name = string("op_19861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19861_cast_fp16 = einsum(equation = var_19861_equation_0, values = (var_19339_cast_fp16, var_19741_cast_fp16))[name = string("op_19861_cast_fp16")]; + string var_19863_equation_0 = const()[name = string("op_19863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19863_cast_fp16 = einsum(equation = var_19863_equation_0, values = (var_19343_cast_fp16, var_19742_cast_fp16))[name = string("op_19863_cast_fp16")]; + string var_19865_equation_0 = const()[name = string("op_19865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19865_cast_fp16 = einsum(equation = var_19865_equation_0, values = (var_19343_cast_fp16, var_19743_cast_fp16))[name = string("op_19865_cast_fp16")]; + string var_19867_equation_0 = const()[name = string("op_19867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19867_cast_fp16 = einsum(equation = var_19867_equation_0, values = (var_19343_cast_fp16, var_19744_cast_fp16))[name = string("op_19867_cast_fp16")]; + string var_19869_equation_0 = const()[name = string("op_19869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19869_cast_fp16 = einsum(equation = var_19869_equation_0, values = (var_19343_cast_fp16, var_19745_cast_fp16))[name = string("op_19869_cast_fp16")]; + string var_19871_equation_0 = const()[name = string("op_19871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19871_cast_fp16 = einsum(equation = var_19871_equation_0, values = (var_19347_cast_fp16, var_19746_cast_fp16))[name = string("op_19871_cast_fp16")]; + string var_19873_equation_0 = const()[name = string("op_19873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19873_cast_fp16 = einsum(equation = var_19873_equation_0, values = (var_19347_cast_fp16, var_19747_cast_fp16))[name = string("op_19873_cast_fp16")]; + string var_19875_equation_0 = const()[name = string("op_19875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19875_cast_fp16 = einsum(equation = var_19875_equation_0, values = (var_19347_cast_fp16, var_19748_cast_fp16))[name = string("op_19875_cast_fp16")]; + string var_19877_equation_0 = const()[name = string("op_19877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19877_cast_fp16 = einsum(equation = var_19877_equation_0, values = (var_19347_cast_fp16, var_19749_cast_fp16))[name = string("op_19877_cast_fp16")]; + string var_19879_equation_0 = const()[name = string("op_19879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19879_cast_fp16 = einsum(equation = var_19879_equation_0, values = (var_19351_cast_fp16, var_19750_cast_fp16))[name = string("op_19879_cast_fp16")]; + string var_19881_equation_0 = const()[name = string("op_19881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19881_cast_fp16 = einsum(equation = var_19881_equation_0, values = (var_19351_cast_fp16, var_19751_cast_fp16))[name = string("op_19881_cast_fp16")]; + string var_19883_equation_0 = const()[name = string("op_19883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19883_cast_fp16 = einsum(equation = var_19883_equation_0, values = (var_19351_cast_fp16, var_19752_cast_fp16))[name = string("op_19883_cast_fp16")]; + string var_19885_equation_0 = const()[name = string("op_19885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19885_cast_fp16 = einsum(equation = var_19885_equation_0, values = (var_19351_cast_fp16, var_19753_cast_fp16))[name = string("op_19885_cast_fp16")]; + string var_19887_equation_0 = const()[name = string("op_19887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19887_cast_fp16 = einsum(equation = var_19887_equation_0, values = (var_19355_cast_fp16, var_19754_cast_fp16))[name = string("op_19887_cast_fp16")]; + string var_19889_equation_0 = const()[name = string("op_19889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19889_cast_fp16 = einsum(equation = var_19889_equation_0, values = (var_19355_cast_fp16, var_19755_cast_fp16))[name = string("op_19889_cast_fp16")]; + string var_19891_equation_0 = const()[name = string("op_19891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19891_cast_fp16 = einsum(equation = var_19891_equation_0, values = (var_19355_cast_fp16, var_19756_cast_fp16))[name = string("op_19891_cast_fp16")]; + string var_19893_equation_0 = const()[name = string("op_19893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19893_cast_fp16 = einsum(equation = var_19893_equation_0, values = (var_19355_cast_fp16, var_19757_cast_fp16))[name = string("op_19893_cast_fp16")]; + string var_19895_equation_0 = const()[name = string("op_19895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19895_cast_fp16 = einsum(equation = var_19895_equation_0, values = (var_19359_cast_fp16, var_19758_cast_fp16))[name = string("op_19895_cast_fp16")]; + string var_19897_equation_0 = const()[name = string("op_19897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19897_cast_fp16 = einsum(equation = var_19897_equation_0, values = (var_19359_cast_fp16, var_19759_cast_fp16))[name = string("op_19897_cast_fp16")]; + string var_19899_equation_0 = const()[name = string("op_19899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19899_cast_fp16 = einsum(equation = var_19899_equation_0, values = (var_19359_cast_fp16, var_19760_cast_fp16))[name = string("op_19899_cast_fp16")]; + string var_19901_equation_0 = const()[name = string("op_19901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19901_cast_fp16 = einsum(equation = var_19901_equation_0, values = (var_19359_cast_fp16, var_19761_cast_fp16))[name = string("op_19901_cast_fp16")]; + string var_19903_equation_0 = const()[name = string("op_19903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19903_cast_fp16 = einsum(equation = var_19903_equation_0, values = (var_19363_cast_fp16, var_19762_cast_fp16))[name = string("op_19903_cast_fp16")]; + string var_19905_equation_0 = const()[name = string("op_19905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19905_cast_fp16 = einsum(equation = var_19905_equation_0, values = (var_19363_cast_fp16, var_19763_cast_fp16))[name = string("op_19905_cast_fp16")]; + string var_19907_equation_0 = const()[name = string("op_19907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19907_cast_fp16 = einsum(equation = var_19907_equation_0, values = (var_19363_cast_fp16, var_19764_cast_fp16))[name = string("op_19907_cast_fp16")]; + string var_19909_equation_0 = const()[name = string("op_19909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19909_cast_fp16 = einsum(equation = var_19909_equation_0, values = (var_19363_cast_fp16, var_19765_cast_fp16))[name = string("op_19909_cast_fp16")]; + string var_19911_equation_0 = const()[name = string("op_19911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19911_cast_fp16 = einsum(equation = var_19911_equation_0, values = (var_19367_cast_fp16, var_19766_cast_fp16))[name = string("op_19911_cast_fp16")]; + string var_19913_equation_0 = const()[name = string("op_19913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19913_cast_fp16 = einsum(equation = var_19913_equation_0, values = (var_19367_cast_fp16, var_19767_cast_fp16))[name = string("op_19913_cast_fp16")]; + string var_19915_equation_0 = const()[name = string("op_19915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19915_cast_fp16 = einsum(equation = var_19915_equation_0, values = (var_19367_cast_fp16, var_19768_cast_fp16))[name = string("op_19915_cast_fp16")]; + string var_19917_equation_0 = const()[name = string("op_19917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19917_cast_fp16 = einsum(equation = var_19917_equation_0, values = (var_19367_cast_fp16, var_19769_cast_fp16))[name = string("op_19917_cast_fp16")]; + string var_19919_equation_0 = const()[name = string("op_19919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19919_cast_fp16 = einsum(equation = var_19919_equation_0, values = (var_19371_cast_fp16, var_19770_cast_fp16))[name = string("op_19919_cast_fp16")]; + string var_19921_equation_0 = const()[name = string("op_19921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19921_cast_fp16 = einsum(equation = var_19921_equation_0, values = (var_19371_cast_fp16, var_19771_cast_fp16))[name = string("op_19921_cast_fp16")]; + string var_19923_equation_0 = const()[name = string("op_19923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19923_cast_fp16 = einsum(equation = var_19923_equation_0, values = (var_19371_cast_fp16, var_19772_cast_fp16))[name = string("op_19923_cast_fp16")]; + string var_19925_equation_0 = const()[name = string("op_19925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19925_cast_fp16 = einsum(equation = var_19925_equation_0, values = (var_19371_cast_fp16, var_19773_cast_fp16))[name = string("op_19925_cast_fp16")]; + string var_19927_equation_0 = const()[name = string("op_19927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19927_cast_fp16 = einsum(equation = var_19927_equation_0, values = (var_19375_cast_fp16, var_19774_cast_fp16))[name = string("op_19927_cast_fp16")]; + string var_19929_equation_0 = const()[name = string("op_19929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19929_cast_fp16 = einsum(equation = var_19929_equation_0, values = (var_19375_cast_fp16, var_19775_cast_fp16))[name = string("op_19929_cast_fp16")]; + string var_19931_equation_0 = const()[name = string("op_19931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19931_cast_fp16 = einsum(equation = var_19931_equation_0, values = (var_19375_cast_fp16, var_19776_cast_fp16))[name = string("op_19931_cast_fp16")]; + string var_19933_equation_0 = const()[name = string("op_19933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19933_cast_fp16 = einsum(equation = var_19933_equation_0, values = (var_19375_cast_fp16, var_19777_cast_fp16))[name = string("op_19933_cast_fp16")]; + string var_19935_equation_0 = const()[name = string("op_19935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19935_cast_fp16 = einsum(equation = var_19935_equation_0, values = (var_19379_cast_fp16, var_19778_cast_fp16))[name = string("op_19935_cast_fp16")]; + string var_19937_equation_0 = const()[name = string("op_19937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19937_cast_fp16 = einsum(equation = var_19937_equation_0, values = (var_19379_cast_fp16, var_19779_cast_fp16))[name = string("op_19937_cast_fp16")]; + string var_19939_equation_0 = const()[name = string("op_19939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19939_cast_fp16 = einsum(equation = var_19939_equation_0, values = (var_19379_cast_fp16, var_19780_cast_fp16))[name = string("op_19939_cast_fp16")]; + string var_19941_equation_0 = const()[name = string("op_19941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19941_cast_fp16 = einsum(equation = var_19941_equation_0, values = (var_19379_cast_fp16, var_19781_cast_fp16))[name = string("op_19941_cast_fp16")]; + bool var_19943_interleave_0 = const()[name = string("op_19943_interleave_0"), val = bool(false)]; + tensor var_19943_cast_fp16 = concat(axis = var_18502, interleave = var_19943_interleave_0, values = (var_19783_cast_fp16, var_19785_cast_fp16, var_19787_cast_fp16, var_19789_cast_fp16))[name = string("op_19943_cast_fp16")]; + bool var_19945_interleave_0 = const()[name = string("op_19945_interleave_0"), val = bool(false)]; + tensor var_19945_cast_fp16 = concat(axis = var_18502, interleave = var_19945_interleave_0, values = (var_19791_cast_fp16, var_19793_cast_fp16, var_19795_cast_fp16, var_19797_cast_fp16))[name = string("op_19945_cast_fp16")]; + bool var_19947_interleave_0 = const()[name = string("op_19947_interleave_0"), val = bool(false)]; + tensor var_19947_cast_fp16 = concat(axis = var_18502, interleave = var_19947_interleave_0, values = (var_19799_cast_fp16, var_19801_cast_fp16, var_19803_cast_fp16, var_19805_cast_fp16))[name = string("op_19947_cast_fp16")]; + bool var_19949_interleave_0 = const()[name = string("op_19949_interleave_0"), val = bool(false)]; + tensor var_19949_cast_fp16 = concat(axis = var_18502, interleave = var_19949_interleave_0, values = (var_19807_cast_fp16, var_19809_cast_fp16, var_19811_cast_fp16, var_19813_cast_fp16))[name = string("op_19949_cast_fp16")]; + bool var_19951_interleave_0 = const()[name = string("op_19951_interleave_0"), val = bool(false)]; + tensor var_19951_cast_fp16 = concat(axis = var_18502, interleave = var_19951_interleave_0, values = (var_19815_cast_fp16, var_19817_cast_fp16, var_19819_cast_fp16, var_19821_cast_fp16))[name = string("op_19951_cast_fp16")]; + bool var_19953_interleave_0 = const()[name = string("op_19953_interleave_0"), val = bool(false)]; + tensor var_19953_cast_fp16 = concat(axis = var_18502, interleave = var_19953_interleave_0, values = (var_19823_cast_fp16, var_19825_cast_fp16, var_19827_cast_fp16, var_19829_cast_fp16))[name = string("op_19953_cast_fp16")]; + bool var_19955_interleave_0 = const()[name = string("op_19955_interleave_0"), val = bool(false)]; + tensor var_19955_cast_fp16 = concat(axis = var_18502, interleave = var_19955_interleave_0, values = (var_19831_cast_fp16, var_19833_cast_fp16, var_19835_cast_fp16, var_19837_cast_fp16))[name = string("op_19955_cast_fp16")]; + bool var_19957_interleave_0 = const()[name = string("op_19957_interleave_0"), val = bool(false)]; + tensor var_19957_cast_fp16 = concat(axis = var_18502, interleave = var_19957_interleave_0, values = (var_19839_cast_fp16, var_19841_cast_fp16, var_19843_cast_fp16, var_19845_cast_fp16))[name = string("op_19957_cast_fp16")]; + bool var_19959_interleave_0 = const()[name = string("op_19959_interleave_0"), val = bool(false)]; + tensor var_19959_cast_fp16 = concat(axis = var_18502, interleave = var_19959_interleave_0, values = (var_19847_cast_fp16, var_19849_cast_fp16, var_19851_cast_fp16, var_19853_cast_fp16))[name = string("op_19959_cast_fp16")]; + bool var_19961_interleave_0 = const()[name = string("op_19961_interleave_0"), val = bool(false)]; + tensor var_19961_cast_fp16 = concat(axis = var_18502, interleave = var_19961_interleave_0, values = (var_19855_cast_fp16, var_19857_cast_fp16, var_19859_cast_fp16, var_19861_cast_fp16))[name = string("op_19961_cast_fp16")]; + bool var_19963_interleave_0 = const()[name = string("op_19963_interleave_0"), val = bool(false)]; + tensor var_19963_cast_fp16 = concat(axis = var_18502, interleave = var_19963_interleave_0, values = (var_19863_cast_fp16, var_19865_cast_fp16, var_19867_cast_fp16, var_19869_cast_fp16))[name = string("op_19963_cast_fp16")]; + bool var_19965_interleave_0 = const()[name = string("op_19965_interleave_0"), val = bool(false)]; + tensor var_19965_cast_fp16 = concat(axis = var_18502, interleave = var_19965_interleave_0, values = (var_19871_cast_fp16, var_19873_cast_fp16, var_19875_cast_fp16, var_19877_cast_fp16))[name = string("op_19965_cast_fp16")]; + bool var_19967_interleave_0 = const()[name = string("op_19967_interleave_0"), val = bool(false)]; + tensor var_19967_cast_fp16 = concat(axis = var_18502, interleave = var_19967_interleave_0, values = (var_19879_cast_fp16, var_19881_cast_fp16, var_19883_cast_fp16, var_19885_cast_fp16))[name = string("op_19967_cast_fp16")]; + bool var_19969_interleave_0 = const()[name = string("op_19969_interleave_0"), val = bool(false)]; + tensor var_19969_cast_fp16 = concat(axis = var_18502, interleave = var_19969_interleave_0, values = (var_19887_cast_fp16, var_19889_cast_fp16, var_19891_cast_fp16, var_19893_cast_fp16))[name = string("op_19969_cast_fp16")]; + bool var_19971_interleave_0 = const()[name = string("op_19971_interleave_0"), val = bool(false)]; + tensor var_19971_cast_fp16 = concat(axis = var_18502, interleave = var_19971_interleave_0, values = (var_19895_cast_fp16, var_19897_cast_fp16, var_19899_cast_fp16, var_19901_cast_fp16))[name = string("op_19971_cast_fp16")]; + bool var_19973_interleave_0 = const()[name = string("op_19973_interleave_0"), val = bool(false)]; + tensor var_19973_cast_fp16 = concat(axis = var_18502, interleave = var_19973_interleave_0, values = (var_19903_cast_fp16, var_19905_cast_fp16, var_19907_cast_fp16, var_19909_cast_fp16))[name = string("op_19973_cast_fp16")]; + bool var_19975_interleave_0 = const()[name = string("op_19975_interleave_0"), val = bool(false)]; + tensor var_19975_cast_fp16 = concat(axis = var_18502, interleave = var_19975_interleave_0, values = (var_19911_cast_fp16, var_19913_cast_fp16, var_19915_cast_fp16, var_19917_cast_fp16))[name = string("op_19975_cast_fp16")]; + bool var_19977_interleave_0 = const()[name = string("op_19977_interleave_0"), val = bool(false)]; + tensor var_19977_cast_fp16 = concat(axis = var_18502, interleave = var_19977_interleave_0, values = (var_19919_cast_fp16, var_19921_cast_fp16, var_19923_cast_fp16, var_19925_cast_fp16))[name = string("op_19977_cast_fp16")]; + bool var_19979_interleave_0 = const()[name = string("op_19979_interleave_0"), val = bool(false)]; + tensor var_19979_cast_fp16 = concat(axis = var_18502, interleave = var_19979_interleave_0, values = (var_19927_cast_fp16, var_19929_cast_fp16, var_19931_cast_fp16, var_19933_cast_fp16))[name = string("op_19979_cast_fp16")]; + bool var_19981_interleave_0 = const()[name = string("op_19981_interleave_0"), val = bool(false)]; + tensor var_19981_cast_fp16 = concat(axis = var_18502, interleave = var_19981_interleave_0, values = (var_19935_cast_fp16, var_19937_cast_fp16, var_19939_cast_fp16, var_19941_cast_fp16))[name = string("op_19981_cast_fp16")]; + bool input_97_interleave_0 = const()[name = string("input_97_interleave_0"), val = bool(false)]; + tensor input_97_cast_fp16 = concat(axis = var_18527, interleave = input_97_interleave_0, values = (var_19943_cast_fp16, var_19945_cast_fp16, var_19947_cast_fp16, var_19949_cast_fp16, var_19951_cast_fp16, var_19953_cast_fp16, var_19955_cast_fp16, var_19957_cast_fp16, var_19959_cast_fp16, var_19961_cast_fp16, var_19963_cast_fp16, var_19965_cast_fp16, var_19967_cast_fp16, var_19969_cast_fp16, var_19971_cast_fp16, var_19973_cast_fp16, var_19975_cast_fp16, var_19977_cast_fp16, var_19979_cast_fp16, var_19981_cast_fp16))[name = string("input_97_cast_fp16")]; + string obj_51_pad_type_0 = const()[name = string("obj_51_pad_type_0"), val = string("valid")]; + tensor obj_51_strides_0 = const()[name = string("obj_51_strides_0"), val = tensor([1, 1])]; + tensor obj_51_pad_0 = const()[name = string("obj_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_51_dilations_0 = const()[name = string("obj_51_dilations_0"), val = tensor([1, 1])]; + int32 obj_51_groups_0 = const()[name = string("obj_51_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496744640)))]; + tensor layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500021504)))]; + tensor obj_51_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_51_dilations_0, groups = obj_51_groups_0, pad = obj_51_pad_0, pad_type = obj_51_pad_type_0, strides = obj_51_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("obj_51_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_51_cast_fp16)[name = string("inputs_51_cast_fp16")]; + tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; + fp16 var_20000_to_fp16 = const()[name = string("op_20000_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_20000_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; + tensor input_99_gamma_0_to_fp16 = const()[name = string("input_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500024128)))]; + tensor input_99_beta_0_to_fp16 = const()[name = string("input_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500026752)))]; + fp16 input_99_epsilon_0_to_fp16 = const()[name = string("input_99_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_99_cast_fp16 = batch_norm(beta = input_99_beta_0_to_fp16, epsilon = input_99_epsilon_0_to_fp16, gamma = input_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("input_99_cast_fp16")]; + string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")]; + tensor input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor([1, 1])]; + tensor input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor([1, 1])]; + int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1)]; + tensor layers_12_fc1_weight_to_fp16 = const()[name = string("layers_12_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500029376)))]; + tensor layers_12_fc1_bias_to_fp16 = const()[name = string("layers_12_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513136640)))]; + tensor input_101_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = layers_12_fc1_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")]; + string input_103_mode_0 = const()[name = string("input_103_mode_0"), val = string("EXACT")]; + tensor input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")]; + string hidden_states_29_pad_type_0 = const()[name = string("hidden_states_29_pad_type_0"), val = string("valid")]; + tensor hidden_states_29_strides_0 = const()[name = string("hidden_states_29_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_29_pad_0 = const()[name = string("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_29_dilations_0 = const()[name = string("hidden_states_29_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_29_groups_0 = const()[name = string("hidden_states_29_groups_0"), val = int32(1)]; + tensor layers_12_fc2_weight_to_fp16 = const()[name = string("layers_12_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513146944)))]; + tensor layers_12_fc2_bias_to_fp16 = const()[name = string("layers_12_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526254208)))]; + tensor hidden_states_29_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_12_fc2_weight_to_fp16, x = input_103_cast_fp16)[name = string("hidden_states_29_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = hidden_states_29_cast_fp16)[name = string("inputs_53_cast_fp16")]; + int32 var_20029 = const()[name = string("op_20029"), val = int32(3)]; + int32 var_20054 = const()[name = string("op_20054"), val = int32(1)]; + tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; + fp16 var_20071_to_fp16 = const()[name = string("op_20071_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_20071_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; + tensor obj_53_gamma_0_to_fp16 = const()[name = string("obj_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526256832)))]; + tensor obj_53_beta_0_to_fp16 = const()[name = string("obj_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526259456)))]; + fp16 obj_53_epsilon_0_to_fp16 = const()[name = string("obj_53_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_53_cast_fp16 = batch_norm(beta = obj_53_beta_0_to_fp16, epsilon = obj_53_epsilon_0_to_fp16, gamma = obj_53_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("obj_53_cast_fp16")]; + string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; + tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; + tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; + int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526262080)))]; + tensor layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529538944)))]; + tensor query_27_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("query_27_cast_fp16")]; + string key_27_pad_type_0 = const()[name = string("key_27_pad_type_0"), val = string("valid")]; + tensor key_27_strides_0 = const()[name = string("key_27_strides_0"), val = tensor([1, 1])]; + tensor key_27_pad_0 = const()[name = string("key_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_27_dilations_0 = const()[name = string("key_27_dilations_0"), val = tensor([1, 1])]; + int32 key_27_groups_0 = const()[name = string("key_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529541568)))]; + tensor key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("key_27_cast_fp16")]; + string value_27_pad_type_0 = const()[name = string("value_27_pad_type_0"), val = string("valid")]; + tensor value_27_strides_0 = const()[name = string("value_27_strides_0"), val = tensor([1, 1])]; + tensor value_27_pad_0 = const()[name = string("value_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_27_dilations_0 = const()[name = string("value_27_dilations_0"), val = tensor([1, 1])]; + int32 value_27_groups_0 = const()[name = string("value_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(532818432)))]; + tensor layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536095296)))]; + tensor value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("value_27_cast_fp16")]; + tensor var_20109_begin_0 = const()[name = string("op_20109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20109_end_0 = const()[name = string("op_20109_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20109_end_mask_0 = const()[name = string("op_20109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20109_cast_fp16 = slice_by_index(begin = var_20109_begin_0, end = var_20109_end_0, end_mask = var_20109_end_mask_0, x = query_27_cast_fp16)[name = string("op_20109_cast_fp16")]; + tensor var_20113_begin_0 = const()[name = string("op_20113_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20113_end_0 = const()[name = string("op_20113_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_20113_end_mask_0 = const()[name = string("op_20113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20113_cast_fp16 = slice_by_index(begin = var_20113_begin_0, end = var_20113_end_0, end_mask = var_20113_end_mask_0, x = query_27_cast_fp16)[name = string("op_20113_cast_fp16")]; + tensor var_20117_begin_0 = const()[name = string("op_20117_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20117_end_0 = const()[name = string("op_20117_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_20117_end_mask_0 = const()[name = string("op_20117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20117_cast_fp16 = slice_by_index(begin = var_20117_begin_0, end = var_20117_end_0, end_mask = var_20117_end_mask_0, x = query_27_cast_fp16)[name = string("op_20117_cast_fp16")]; + tensor var_20121_begin_0 = const()[name = string("op_20121_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20121_end_0 = const()[name = string("op_20121_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_20121_end_mask_0 = const()[name = string("op_20121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20121_cast_fp16 = slice_by_index(begin = var_20121_begin_0, end = var_20121_end_0, end_mask = var_20121_end_mask_0, x = query_27_cast_fp16)[name = string("op_20121_cast_fp16")]; + tensor var_20125_begin_0 = const()[name = string("op_20125_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20125_end_0 = const()[name = string("op_20125_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_20125_end_mask_0 = const()[name = string("op_20125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20125_cast_fp16 = slice_by_index(begin = var_20125_begin_0, end = var_20125_end_0, end_mask = var_20125_end_mask_0, x = query_27_cast_fp16)[name = string("op_20125_cast_fp16")]; + tensor var_20129_begin_0 = const()[name = string("op_20129_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20129_end_0 = const()[name = string("op_20129_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_20129_end_mask_0 = const()[name = string("op_20129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20129_cast_fp16 = slice_by_index(begin = var_20129_begin_0, end = var_20129_end_0, end_mask = var_20129_end_mask_0, x = query_27_cast_fp16)[name = string("op_20129_cast_fp16")]; + tensor var_20133_begin_0 = const()[name = string("op_20133_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20133_end_0 = const()[name = string("op_20133_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_20133_end_mask_0 = const()[name = string("op_20133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20133_cast_fp16 = slice_by_index(begin = var_20133_begin_0, end = var_20133_end_0, end_mask = var_20133_end_mask_0, x = query_27_cast_fp16)[name = string("op_20133_cast_fp16")]; + tensor var_20137_begin_0 = const()[name = string("op_20137_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20137_end_0 = const()[name = string("op_20137_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_20137_end_mask_0 = const()[name = string("op_20137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20137_cast_fp16 = slice_by_index(begin = var_20137_begin_0, end = var_20137_end_0, end_mask = var_20137_end_mask_0, x = query_27_cast_fp16)[name = string("op_20137_cast_fp16")]; + tensor var_20141_begin_0 = const()[name = string("op_20141_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20141_end_0 = const()[name = string("op_20141_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_20141_end_mask_0 = const()[name = string("op_20141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20141_cast_fp16 = slice_by_index(begin = var_20141_begin_0, end = var_20141_end_0, end_mask = var_20141_end_mask_0, x = query_27_cast_fp16)[name = string("op_20141_cast_fp16")]; + tensor var_20145_begin_0 = const()[name = string("op_20145_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20145_end_0 = const()[name = string("op_20145_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_20145_end_mask_0 = const()[name = string("op_20145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20145_cast_fp16 = slice_by_index(begin = var_20145_begin_0, end = var_20145_end_0, end_mask = var_20145_end_mask_0, x = query_27_cast_fp16)[name = string("op_20145_cast_fp16")]; + tensor var_20149_begin_0 = const()[name = string("op_20149_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20149_end_0 = const()[name = string("op_20149_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_20149_end_mask_0 = const()[name = string("op_20149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20149_cast_fp16 = slice_by_index(begin = var_20149_begin_0, end = var_20149_end_0, end_mask = var_20149_end_mask_0, x = query_27_cast_fp16)[name = string("op_20149_cast_fp16")]; + tensor var_20153_begin_0 = const()[name = string("op_20153_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20153_end_0 = const()[name = string("op_20153_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_20153_end_mask_0 = const()[name = string("op_20153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20153_cast_fp16 = slice_by_index(begin = var_20153_begin_0, end = var_20153_end_0, end_mask = var_20153_end_mask_0, x = query_27_cast_fp16)[name = string("op_20153_cast_fp16")]; + tensor var_20157_begin_0 = const()[name = string("op_20157_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20157_end_0 = const()[name = string("op_20157_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_20157_end_mask_0 = const()[name = string("op_20157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20157_cast_fp16 = slice_by_index(begin = var_20157_begin_0, end = var_20157_end_0, end_mask = var_20157_end_mask_0, x = query_27_cast_fp16)[name = string("op_20157_cast_fp16")]; + tensor var_20161_begin_0 = const()[name = string("op_20161_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20161_end_0 = const()[name = string("op_20161_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_20161_end_mask_0 = const()[name = string("op_20161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20161_cast_fp16 = slice_by_index(begin = var_20161_begin_0, end = var_20161_end_0, end_mask = var_20161_end_mask_0, x = query_27_cast_fp16)[name = string("op_20161_cast_fp16")]; + tensor var_20165_begin_0 = const()[name = string("op_20165_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20165_end_0 = const()[name = string("op_20165_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_20165_end_mask_0 = const()[name = string("op_20165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20165_cast_fp16 = slice_by_index(begin = var_20165_begin_0, end = var_20165_end_0, end_mask = var_20165_end_mask_0, x = query_27_cast_fp16)[name = string("op_20165_cast_fp16")]; + tensor var_20169_begin_0 = const()[name = string("op_20169_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20169_end_0 = const()[name = string("op_20169_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_20169_end_mask_0 = const()[name = string("op_20169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20169_cast_fp16 = slice_by_index(begin = var_20169_begin_0, end = var_20169_end_0, end_mask = var_20169_end_mask_0, x = query_27_cast_fp16)[name = string("op_20169_cast_fp16")]; + tensor var_20173_begin_0 = const()[name = string("op_20173_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20173_end_0 = const()[name = string("op_20173_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_20173_end_mask_0 = const()[name = string("op_20173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20173_cast_fp16 = slice_by_index(begin = var_20173_begin_0, end = var_20173_end_0, end_mask = var_20173_end_mask_0, x = query_27_cast_fp16)[name = string("op_20173_cast_fp16")]; + tensor var_20177_begin_0 = const()[name = string("op_20177_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20177_end_0 = const()[name = string("op_20177_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_20177_end_mask_0 = const()[name = string("op_20177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20177_cast_fp16 = slice_by_index(begin = var_20177_begin_0, end = var_20177_end_0, end_mask = var_20177_end_mask_0, x = query_27_cast_fp16)[name = string("op_20177_cast_fp16")]; + tensor var_20181_begin_0 = const()[name = string("op_20181_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20181_end_0 = const()[name = string("op_20181_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_20181_end_mask_0 = const()[name = string("op_20181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20181_cast_fp16 = slice_by_index(begin = var_20181_begin_0, end = var_20181_end_0, end_mask = var_20181_end_mask_0, x = query_27_cast_fp16)[name = string("op_20181_cast_fp16")]; + tensor var_20185_begin_0 = const()[name = string("op_20185_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20185_end_0 = const()[name = string("op_20185_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_20185_end_mask_0 = const()[name = string("op_20185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20185_cast_fp16 = slice_by_index(begin = var_20185_begin_0, end = var_20185_end_0, end_mask = var_20185_end_mask_0, x = query_27_cast_fp16)[name = string("op_20185_cast_fp16")]; + tensor var_20194_begin_0 = const()[name = string("op_20194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20194_end_0 = const()[name = string("op_20194_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20194_end_mask_0 = const()[name = string("op_20194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20194_cast_fp16 = slice_by_index(begin = var_20194_begin_0, end = var_20194_end_0, end_mask = var_20194_end_mask_0, x = var_20109_cast_fp16)[name = string("op_20194_cast_fp16")]; + tensor var_20201_begin_0 = const()[name = string("op_20201_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20201_end_0 = const()[name = string("op_20201_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20201_end_mask_0 = const()[name = string("op_20201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20201_cast_fp16 = slice_by_index(begin = var_20201_begin_0, end = var_20201_end_0, end_mask = var_20201_end_mask_0, x = var_20109_cast_fp16)[name = string("op_20201_cast_fp16")]; + tensor var_20208_begin_0 = const()[name = string("op_20208_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20208_end_0 = const()[name = string("op_20208_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20208_end_mask_0 = const()[name = string("op_20208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20208_cast_fp16 = slice_by_index(begin = var_20208_begin_0, end = var_20208_end_0, end_mask = var_20208_end_mask_0, x = var_20109_cast_fp16)[name = string("op_20208_cast_fp16")]; + tensor var_20215_begin_0 = const()[name = string("op_20215_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20215_end_0 = const()[name = string("op_20215_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20215_end_mask_0 = const()[name = string("op_20215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20215_cast_fp16 = slice_by_index(begin = var_20215_begin_0, end = var_20215_end_0, end_mask = var_20215_end_mask_0, x = var_20109_cast_fp16)[name = string("op_20215_cast_fp16")]; + tensor var_20222_begin_0 = const()[name = string("op_20222_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20222_end_0 = const()[name = string("op_20222_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20222_end_mask_0 = const()[name = string("op_20222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20222_cast_fp16 = slice_by_index(begin = var_20222_begin_0, end = var_20222_end_0, end_mask = var_20222_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20222_cast_fp16")]; + tensor var_20229_begin_0 = const()[name = string("op_20229_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20229_end_0 = const()[name = string("op_20229_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20229_end_mask_0 = const()[name = string("op_20229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20229_cast_fp16 = slice_by_index(begin = var_20229_begin_0, end = var_20229_end_0, end_mask = var_20229_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20229_cast_fp16")]; + tensor var_20236_begin_0 = const()[name = string("op_20236_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20236_end_0 = const()[name = string("op_20236_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20236_end_mask_0 = const()[name = string("op_20236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20236_cast_fp16 = slice_by_index(begin = var_20236_begin_0, end = var_20236_end_0, end_mask = var_20236_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20236_cast_fp16")]; + tensor var_20243_begin_0 = const()[name = string("op_20243_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20243_end_0 = const()[name = string("op_20243_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20243_end_mask_0 = const()[name = string("op_20243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20243_cast_fp16 = slice_by_index(begin = var_20243_begin_0, end = var_20243_end_0, end_mask = var_20243_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20243_cast_fp16")]; + tensor var_20250_begin_0 = const()[name = string("op_20250_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20250_end_0 = const()[name = string("op_20250_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20250_end_mask_0 = const()[name = string("op_20250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20250_cast_fp16 = slice_by_index(begin = var_20250_begin_0, end = var_20250_end_0, end_mask = var_20250_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20250_cast_fp16")]; + tensor var_20257_begin_0 = const()[name = string("op_20257_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20257_end_0 = const()[name = string("op_20257_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20257_end_mask_0 = const()[name = string("op_20257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20257_cast_fp16 = slice_by_index(begin = var_20257_begin_0, end = var_20257_end_0, end_mask = var_20257_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20257_cast_fp16")]; + tensor var_20264_begin_0 = const()[name = string("op_20264_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20264_end_0 = const()[name = string("op_20264_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20264_end_mask_0 = const()[name = string("op_20264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20264_cast_fp16 = slice_by_index(begin = var_20264_begin_0, end = var_20264_end_0, end_mask = var_20264_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20264_cast_fp16")]; + tensor var_20271_begin_0 = const()[name = string("op_20271_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20271_end_0 = const()[name = string("op_20271_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20271_end_mask_0 = const()[name = string("op_20271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20271_cast_fp16 = slice_by_index(begin = var_20271_begin_0, end = var_20271_end_0, end_mask = var_20271_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20271_cast_fp16")]; + tensor var_20278_begin_0 = const()[name = string("op_20278_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20278_end_0 = const()[name = string("op_20278_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20278_end_mask_0 = const()[name = string("op_20278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20278_cast_fp16 = slice_by_index(begin = var_20278_begin_0, end = var_20278_end_0, end_mask = var_20278_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20278_cast_fp16")]; + tensor var_20285_begin_0 = const()[name = string("op_20285_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20285_end_0 = const()[name = string("op_20285_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20285_end_mask_0 = const()[name = string("op_20285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20285_cast_fp16 = slice_by_index(begin = var_20285_begin_0, end = var_20285_end_0, end_mask = var_20285_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20285_cast_fp16")]; + tensor var_20292_begin_0 = const()[name = string("op_20292_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20292_end_0 = const()[name = string("op_20292_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20292_end_mask_0 = const()[name = string("op_20292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20292_cast_fp16 = slice_by_index(begin = var_20292_begin_0, end = var_20292_end_0, end_mask = var_20292_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20292_cast_fp16")]; + tensor var_20299_begin_0 = const()[name = string("op_20299_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20299_end_0 = const()[name = string("op_20299_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20299_end_mask_0 = const()[name = string("op_20299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20299_cast_fp16 = slice_by_index(begin = var_20299_begin_0, end = var_20299_end_0, end_mask = var_20299_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20299_cast_fp16")]; + tensor var_20306_begin_0 = const()[name = string("op_20306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20306_end_0 = const()[name = string("op_20306_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20306_end_mask_0 = const()[name = string("op_20306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20306_cast_fp16 = slice_by_index(begin = var_20306_begin_0, end = var_20306_end_0, end_mask = var_20306_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20306_cast_fp16")]; + tensor var_20313_begin_0 = const()[name = string("op_20313_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20313_end_0 = const()[name = string("op_20313_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20313_end_mask_0 = const()[name = string("op_20313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20313_cast_fp16 = slice_by_index(begin = var_20313_begin_0, end = var_20313_end_0, end_mask = var_20313_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20313_cast_fp16")]; + tensor var_20320_begin_0 = const()[name = string("op_20320_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20320_end_0 = const()[name = string("op_20320_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20320_end_mask_0 = const()[name = string("op_20320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20320_cast_fp16 = slice_by_index(begin = var_20320_begin_0, end = var_20320_end_0, end_mask = var_20320_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20320_cast_fp16")]; + tensor var_20327_begin_0 = const()[name = string("op_20327_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20327_end_0 = const()[name = string("op_20327_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20327_end_mask_0 = const()[name = string("op_20327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20327_cast_fp16 = slice_by_index(begin = var_20327_begin_0, end = var_20327_end_0, end_mask = var_20327_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20327_cast_fp16")]; + tensor var_20334_begin_0 = const()[name = string("op_20334_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20334_end_0 = const()[name = string("op_20334_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20334_end_mask_0 = const()[name = string("op_20334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20334_cast_fp16 = slice_by_index(begin = var_20334_begin_0, end = var_20334_end_0, end_mask = var_20334_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20334_cast_fp16")]; + tensor var_20341_begin_0 = const()[name = string("op_20341_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20341_end_0 = const()[name = string("op_20341_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20341_end_mask_0 = const()[name = string("op_20341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20341_cast_fp16 = slice_by_index(begin = var_20341_begin_0, end = var_20341_end_0, end_mask = var_20341_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20341_cast_fp16")]; + tensor var_20348_begin_0 = const()[name = string("op_20348_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20348_end_0 = const()[name = string("op_20348_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20348_end_mask_0 = const()[name = string("op_20348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20348_cast_fp16 = slice_by_index(begin = var_20348_begin_0, end = var_20348_end_0, end_mask = var_20348_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20348_cast_fp16")]; + tensor var_20355_begin_0 = const()[name = string("op_20355_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20355_end_0 = const()[name = string("op_20355_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20355_end_mask_0 = const()[name = string("op_20355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20355_cast_fp16 = slice_by_index(begin = var_20355_begin_0, end = var_20355_end_0, end_mask = var_20355_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20355_cast_fp16")]; + tensor var_20362_begin_0 = const()[name = string("op_20362_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20362_end_0 = const()[name = string("op_20362_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20362_end_mask_0 = const()[name = string("op_20362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20362_cast_fp16 = slice_by_index(begin = var_20362_begin_0, end = var_20362_end_0, end_mask = var_20362_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20362_cast_fp16")]; + tensor var_20369_begin_0 = const()[name = string("op_20369_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20369_end_0 = const()[name = string("op_20369_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20369_end_mask_0 = const()[name = string("op_20369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20369_cast_fp16 = slice_by_index(begin = var_20369_begin_0, end = var_20369_end_0, end_mask = var_20369_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20369_cast_fp16")]; + tensor var_20376_begin_0 = const()[name = string("op_20376_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20376_end_0 = const()[name = string("op_20376_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20376_end_mask_0 = const()[name = string("op_20376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20376_cast_fp16 = slice_by_index(begin = var_20376_begin_0, end = var_20376_end_0, end_mask = var_20376_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20376_cast_fp16")]; + tensor var_20383_begin_0 = const()[name = string("op_20383_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20383_end_0 = const()[name = string("op_20383_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20383_end_mask_0 = const()[name = string("op_20383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20383_cast_fp16 = slice_by_index(begin = var_20383_begin_0, end = var_20383_end_0, end_mask = var_20383_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20383_cast_fp16")]; + tensor var_20390_begin_0 = const()[name = string("op_20390_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20390_end_0 = const()[name = string("op_20390_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20390_end_mask_0 = const()[name = string("op_20390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20390_cast_fp16 = slice_by_index(begin = var_20390_begin_0, end = var_20390_end_0, end_mask = var_20390_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20390_cast_fp16")]; + tensor var_20397_begin_0 = const()[name = string("op_20397_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20397_end_0 = const()[name = string("op_20397_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20397_end_mask_0 = const()[name = string("op_20397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20397_cast_fp16 = slice_by_index(begin = var_20397_begin_0, end = var_20397_end_0, end_mask = var_20397_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20397_cast_fp16")]; + tensor var_20404_begin_0 = const()[name = string("op_20404_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20404_end_0 = const()[name = string("op_20404_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20404_end_mask_0 = const()[name = string("op_20404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20404_cast_fp16 = slice_by_index(begin = var_20404_begin_0, end = var_20404_end_0, end_mask = var_20404_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20404_cast_fp16")]; + tensor var_20411_begin_0 = const()[name = string("op_20411_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20411_end_0 = const()[name = string("op_20411_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20411_end_mask_0 = const()[name = string("op_20411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20411_cast_fp16 = slice_by_index(begin = var_20411_begin_0, end = var_20411_end_0, end_mask = var_20411_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20411_cast_fp16")]; + tensor var_20418_begin_0 = const()[name = string("op_20418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20418_end_0 = const()[name = string("op_20418_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20418_end_mask_0 = const()[name = string("op_20418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20418_cast_fp16 = slice_by_index(begin = var_20418_begin_0, end = var_20418_end_0, end_mask = var_20418_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20418_cast_fp16")]; + tensor var_20425_begin_0 = const()[name = string("op_20425_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20425_end_0 = const()[name = string("op_20425_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20425_end_mask_0 = const()[name = string("op_20425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20425_cast_fp16 = slice_by_index(begin = var_20425_begin_0, end = var_20425_end_0, end_mask = var_20425_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20425_cast_fp16")]; + tensor var_20432_begin_0 = const()[name = string("op_20432_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20432_end_0 = const()[name = string("op_20432_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20432_end_mask_0 = const()[name = string("op_20432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20432_cast_fp16 = slice_by_index(begin = var_20432_begin_0, end = var_20432_end_0, end_mask = var_20432_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20432_cast_fp16")]; + tensor var_20439_begin_0 = const()[name = string("op_20439_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20439_end_0 = const()[name = string("op_20439_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20439_end_mask_0 = const()[name = string("op_20439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20439_cast_fp16 = slice_by_index(begin = var_20439_begin_0, end = var_20439_end_0, end_mask = var_20439_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20439_cast_fp16")]; + tensor var_20446_begin_0 = const()[name = string("op_20446_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20446_end_0 = const()[name = string("op_20446_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20446_end_mask_0 = const()[name = string("op_20446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20446_cast_fp16 = slice_by_index(begin = var_20446_begin_0, end = var_20446_end_0, end_mask = var_20446_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20446_cast_fp16")]; + tensor var_20453_begin_0 = const()[name = string("op_20453_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20453_end_0 = const()[name = string("op_20453_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20453_end_mask_0 = const()[name = string("op_20453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20453_cast_fp16 = slice_by_index(begin = var_20453_begin_0, end = var_20453_end_0, end_mask = var_20453_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20453_cast_fp16")]; + tensor var_20460_begin_0 = const()[name = string("op_20460_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20460_end_0 = const()[name = string("op_20460_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20460_end_mask_0 = const()[name = string("op_20460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20460_cast_fp16 = slice_by_index(begin = var_20460_begin_0, end = var_20460_end_0, end_mask = var_20460_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20460_cast_fp16")]; + tensor var_20467_begin_0 = const()[name = string("op_20467_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20467_end_0 = const()[name = string("op_20467_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20467_end_mask_0 = const()[name = string("op_20467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20467_cast_fp16 = slice_by_index(begin = var_20467_begin_0, end = var_20467_end_0, end_mask = var_20467_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20467_cast_fp16")]; + tensor var_20474_begin_0 = const()[name = string("op_20474_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20474_end_0 = const()[name = string("op_20474_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20474_end_mask_0 = const()[name = string("op_20474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20474_cast_fp16 = slice_by_index(begin = var_20474_begin_0, end = var_20474_end_0, end_mask = var_20474_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20474_cast_fp16")]; + tensor var_20481_begin_0 = const()[name = string("op_20481_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20481_end_0 = const()[name = string("op_20481_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20481_end_mask_0 = const()[name = string("op_20481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20481_cast_fp16 = slice_by_index(begin = var_20481_begin_0, end = var_20481_end_0, end_mask = var_20481_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20481_cast_fp16")]; + tensor var_20488_begin_0 = const()[name = string("op_20488_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20488_end_0 = const()[name = string("op_20488_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20488_end_mask_0 = const()[name = string("op_20488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20488_cast_fp16 = slice_by_index(begin = var_20488_begin_0, end = var_20488_end_0, end_mask = var_20488_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20488_cast_fp16")]; + tensor var_20495_begin_0 = const()[name = string("op_20495_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20495_end_0 = const()[name = string("op_20495_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20495_end_mask_0 = const()[name = string("op_20495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20495_cast_fp16 = slice_by_index(begin = var_20495_begin_0, end = var_20495_end_0, end_mask = var_20495_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20495_cast_fp16")]; + tensor var_20502_begin_0 = const()[name = string("op_20502_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20502_end_0 = const()[name = string("op_20502_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20502_end_mask_0 = const()[name = string("op_20502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20502_cast_fp16 = slice_by_index(begin = var_20502_begin_0, end = var_20502_end_0, end_mask = var_20502_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20502_cast_fp16")]; + tensor var_20509_begin_0 = const()[name = string("op_20509_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20509_end_0 = const()[name = string("op_20509_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20509_end_mask_0 = const()[name = string("op_20509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20509_cast_fp16 = slice_by_index(begin = var_20509_begin_0, end = var_20509_end_0, end_mask = var_20509_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20509_cast_fp16")]; + tensor var_20516_begin_0 = const()[name = string("op_20516_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20516_end_0 = const()[name = string("op_20516_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20516_end_mask_0 = const()[name = string("op_20516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20516_cast_fp16 = slice_by_index(begin = var_20516_begin_0, end = var_20516_end_0, end_mask = var_20516_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20516_cast_fp16")]; + tensor var_20523_begin_0 = const()[name = string("op_20523_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20523_end_0 = const()[name = string("op_20523_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20523_end_mask_0 = const()[name = string("op_20523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20523_cast_fp16 = slice_by_index(begin = var_20523_begin_0, end = var_20523_end_0, end_mask = var_20523_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20523_cast_fp16")]; + tensor var_20530_begin_0 = const()[name = string("op_20530_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20530_end_0 = const()[name = string("op_20530_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20530_end_mask_0 = const()[name = string("op_20530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20530_cast_fp16 = slice_by_index(begin = var_20530_begin_0, end = var_20530_end_0, end_mask = var_20530_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20530_cast_fp16")]; + tensor var_20537_begin_0 = const()[name = string("op_20537_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20537_end_0 = const()[name = string("op_20537_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20537_end_mask_0 = const()[name = string("op_20537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20537_cast_fp16 = slice_by_index(begin = var_20537_begin_0, end = var_20537_end_0, end_mask = var_20537_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20537_cast_fp16")]; + tensor var_20544_begin_0 = const()[name = string("op_20544_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20544_end_0 = const()[name = string("op_20544_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20544_end_mask_0 = const()[name = string("op_20544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20544_cast_fp16 = slice_by_index(begin = var_20544_begin_0, end = var_20544_end_0, end_mask = var_20544_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20544_cast_fp16")]; + tensor var_20551_begin_0 = const()[name = string("op_20551_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20551_end_0 = const()[name = string("op_20551_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20551_end_mask_0 = const()[name = string("op_20551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20551_cast_fp16 = slice_by_index(begin = var_20551_begin_0, end = var_20551_end_0, end_mask = var_20551_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20551_cast_fp16")]; + tensor var_20558_begin_0 = const()[name = string("op_20558_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20558_end_0 = const()[name = string("op_20558_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20558_end_mask_0 = const()[name = string("op_20558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20558_cast_fp16 = slice_by_index(begin = var_20558_begin_0, end = var_20558_end_0, end_mask = var_20558_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20558_cast_fp16")]; + tensor var_20565_begin_0 = const()[name = string("op_20565_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20565_end_0 = const()[name = string("op_20565_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20565_end_mask_0 = const()[name = string("op_20565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20565_cast_fp16 = slice_by_index(begin = var_20565_begin_0, end = var_20565_end_0, end_mask = var_20565_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20565_cast_fp16")]; + tensor var_20572_begin_0 = const()[name = string("op_20572_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20572_end_0 = const()[name = string("op_20572_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20572_end_mask_0 = const()[name = string("op_20572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20572_cast_fp16 = slice_by_index(begin = var_20572_begin_0, end = var_20572_end_0, end_mask = var_20572_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20572_cast_fp16")]; + tensor var_20579_begin_0 = const()[name = string("op_20579_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20579_end_0 = const()[name = string("op_20579_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20579_end_mask_0 = const()[name = string("op_20579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20579_cast_fp16 = slice_by_index(begin = var_20579_begin_0, end = var_20579_end_0, end_mask = var_20579_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20579_cast_fp16")]; + tensor var_20586_begin_0 = const()[name = string("op_20586_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20586_end_0 = const()[name = string("op_20586_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20586_end_mask_0 = const()[name = string("op_20586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20586_cast_fp16 = slice_by_index(begin = var_20586_begin_0, end = var_20586_end_0, end_mask = var_20586_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20586_cast_fp16")]; + tensor var_20593_begin_0 = const()[name = string("op_20593_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20593_end_0 = const()[name = string("op_20593_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20593_end_mask_0 = const()[name = string("op_20593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20593_cast_fp16 = slice_by_index(begin = var_20593_begin_0, end = var_20593_end_0, end_mask = var_20593_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20593_cast_fp16")]; + tensor var_20600_begin_0 = const()[name = string("op_20600_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20600_end_0 = const()[name = string("op_20600_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20600_end_mask_0 = const()[name = string("op_20600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20600_cast_fp16 = slice_by_index(begin = var_20600_begin_0, end = var_20600_end_0, end_mask = var_20600_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20600_cast_fp16")]; + tensor var_20607_begin_0 = const()[name = string("op_20607_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20607_end_0 = const()[name = string("op_20607_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20607_end_mask_0 = const()[name = string("op_20607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20607_cast_fp16 = slice_by_index(begin = var_20607_begin_0, end = var_20607_end_0, end_mask = var_20607_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20607_cast_fp16")]; + tensor var_20614_begin_0 = const()[name = string("op_20614_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20614_end_0 = const()[name = string("op_20614_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20614_end_mask_0 = const()[name = string("op_20614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20614_cast_fp16 = slice_by_index(begin = var_20614_begin_0, end = var_20614_end_0, end_mask = var_20614_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20614_cast_fp16")]; + tensor var_20621_begin_0 = const()[name = string("op_20621_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20621_end_0 = const()[name = string("op_20621_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20621_end_mask_0 = const()[name = string("op_20621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20621_cast_fp16 = slice_by_index(begin = var_20621_begin_0, end = var_20621_end_0, end_mask = var_20621_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20621_cast_fp16")]; + tensor var_20628_begin_0 = const()[name = string("op_20628_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20628_end_0 = const()[name = string("op_20628_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20628_end_mask_0 = const()[name = string("op_20628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20628_cast_fp16 = slice_by_index(begin = var_20628_begin_0, end = var_20628_end_0, end_mask = var_20628_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20628_cast_fp16")]; + tensor var_20635_begin_0 = const()[name = string("op_20635_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20635_end_0 = const()[name = string("op_20635_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20635_end_mask_0 = const()[name = string("op_20635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20635_cast_fp16 = slice_by_index(begin = var_20635_begin_0, end = var_20635_end_0, end_mask = var_20635_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20635_cast_fp16")]; + tensor var_20642_begin_0 = const()[name = string("op_20642_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20642_end_0 = const()[name = string("op_20642_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20642_end_mask_0 = const()[name = string("op_20642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20642_cast_fp16 = slice_by_index(begin = var_20642_begin_0, end = var_20642_end_0, end_mask = var_20642_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20642_cast_fp16")]; + tensor var_20649_begin_0 = const()[name = string("op_20649_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20649_end_0 = const()[name = string("op_20649_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20649_end_mask_0 = const()[name = string("op_20649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20649_cast_fp16 = slice_by_index(begin = var_20649_begin_0, end = var_20649_end_0, end_mask = var_20649_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20649_cast_fp16")]; + tensor var_20656_begin_0 = const()[name = string("op_20656_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20656_end_0 = const()[name = string("op_20656_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20656_end_mask_0 = const()[name = string("op_20656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20656_cast_fp16 = slice_by_index(begin = var_20656_begin_0, end = var_20656_end_0, end_mask = var_20656_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20656_cast_fp16")]; + tensor var_20663_begin_0 = const()[name = string("op_20663_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20663_end_0 = const()[name = string("op_20663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20663_end_mask_0 = const()[name = string("op_20663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20663_cast_fp16 = slice_by_index(begin = var_20663_begin_0, end = var_20663_end_0, end_mask = var_20663_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20663_cast_fp16")]; + tensor var_20670_begin_0 = const()[name = string("op_20670_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20670_end_0 = const()[name = string("op_20670_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20670_end_mask_0 = const()[name = string("op_20670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20670_cast_fp16 = slice_by_index(begin = var_20670_begin_0, end = var_20670_end_0, end_mask = var_20670_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20670_cast_fp16")]; + tensor var_20677_begin_0 = const()[name = string("op_20677_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20677_end_0 = const()[name = string("op_20677_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20677_end_mask_0 = const()[name = string("op_20677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20677_cast_fp16 = slice_by_index(begin = var_20677_begin_0, end = var_20677_end_0, end_mask = var_20677_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20677_cast_fp16")]; + tensor var_20684_begin_0 = const()[name = string("op_20684_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20684_end_0 = const()[name = string("op_20684_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20684_end_mask_0 = const()[name = string("op_20684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20684_cast_fp16 = slice_by_index(begin = var_20684_begin_0, end = var_20684_end_0, end_mask = var_20684_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20684_cast_fp16")]; + tensor var_20691_begin_0 = const()[name = string("op_20691_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20691_end_0 = const()[name = string("op_20691_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20691_end_mask_0 = const()[name = string("op_20691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20691_cast_fp16 = slice_by_index(begin = var_20691_begin_0, end = var_20691_end_0, end_mask = var_20691_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20691_cast_fp16")]; + tensor var_20698_begin_0 = const()[name = string("op_20698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20698_end_0 = const()[name = string("op_20698_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20698_end_mask_0 = const()[name = string("op_20698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20698_cast_fp16 = slice_by_index(begin = var_20698_begin_0, end = var_20698_end_0, end_mask = var_20698_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20698_cast_fp16")]; + tensor var_20705_begin_0 = const()[name = string("op_20705_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20705_end_0 = const()[name = string("op_20705_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20705_end_mask_0 = const()[name = string("op_20705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20705_cast_fp16 = slice_by_index(begin = var_20705_begin_0, end = var_20705_end_0, end_mask = var_20705_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20705_cast_fp16")]; + tensor var_20712_begin_0 = const()[name = string("op_20712_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20712_end_0 = const()[name = string("op_20712_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20712_end_mask_0 = const()[name = string("op_20712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20712_cast_fp16 = slice_by_index(begin = var_20712_begin_0, end = var_20712_end_0, end_mask = var_20712_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20712_cast_fp16")]; + tensor var_20719_begin_0 = const()[name = string("op_20719_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20719_end_0 = const()[name = string("op_20719_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20719_end_mask_0 = const()[name = string("op_20719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20719_cast_fp16 = slice_by_index(begin = var_20719_begin_0, end = var_20719_end_0, end_mask = var_20719_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20719_cast_fp16")]; + tensor var_20726_begin_0 = const()[name = string("op_20726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20726_end_0 = const()[name = string("op_20726_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20726_end_mask_0 = const()[name = string("op_20726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20726_cast_fp16 = slice_by_index(begin = var_20726_begin_0, end = var_20726_end_0, end_mask = var_20726_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20726_cast_fp16")]; + tensor var_20733_begin_0 = const()[name = string("op_20733_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20733_end_0 = const()[name = string("op_20733_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20733_end_mask_0 = const()[name = string("op_20733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20733_cast_fp16 = slice_by_index(begin = var_20733_begin_0, end = var_20733_end_0, end_mask = var_20733_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20733_cast_fp16")]; + tensor var_20740_begin_0 = const()[name = string("op_20740_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20740_end_0 = const()[name = string("op_20740_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20740_end_mask_0 = const()[name = string("op_20740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20740_cast_fp16 = slice_by_index(begin = var_20740_begin_0, end = var_20740_end_0, end_mask = var_20740_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20740_cast_fp16")]; + tensor var_20747_begin_0 = const()[name = string("op_20747_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20747_end_0 = const()[name = string("op_20747_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20747_end_mask_0 = const()[name = string("op_20747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20747_cast_fp16 = slice_by_index(begin = var_20747_begin_0, end = var_20747_end_0, end_mask = var_20747_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20747_cast_fp16")]; + tensor k_27_perm_0 = const()[name = string("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20752_begin_0 = const()[name = string("op_20752_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20752_end_0 = const()[name = string("op_20752_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_20752_end_mask_0 = const()[name = string("op_20752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = key_27_cast_fp16)[name = string("transpose_18")]; + tensor var_20752_cast_fp16 = slice_by_index(begin = var_20752_begin_0, end = var_20752_end_0, end_mask = var_20752_end_mask_0, x = k_27_cast_fp16)[name = string("op_20752_cast_fp16")]; + tensor var_20756_begin_0 = const()[name = string("op_20756_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20756_end_0 = const()[name = string("op_20756_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_20756_end_mask_0 = const()[name = string("op_20756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20756_cast_fp16 = slice_by_index(begin = var_20756_begin_0, end = var_20756_end_0, end_mask = var_20756_end_mask_0, x = k_27_cast_fp16)[name = string("op_20756_cast_fp16")]; + tensor var_20760_begin_0 = const()[name = string("op_20760_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20760_end_0 = const()[name = string("op_20760_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_20760_end_mask_0 = const()[name = string("op_20760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20760_cast_fp16 = slice_by_index(begin = var_20760_begin_0, end = var_20760_end_0, end_mask = var_20760_end_mask_0, x = k_27_cast_fp16)[name = string("op_20760_cast_fp16")]; + tensor var_20764_begin_0 = const()[name = string("op_20764_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20764_end_0 = const()[name = string("op_20764_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_20764_end_mask_0 = const()[name = string("op_20764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20764_cast_fp16 = slice_by_index(begin = var_20764_begin_0, end = var_20764_end_0, end_mask = var_20764_end_mask_0, x = k_27_cast_fp16)[name = string("op_20764_cast_fp16")]; + tensor var_20768_begin_0 = const()[name = string("op_20768_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20768_end_0 = const()[name = string("op_20768_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_20768_end_mask_0 = const()[name = string("op_20768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20768_cast_fp16 = slice_by_index(begin = var_20768_begin_0, end = var_20768_end_0, end_mask = var_20768_end_mask_0, x = k_27_cast_fp16)[name = string("op_20768_cast_fp16")]; + tensor var_20772_begin_0 = const()[name = string("op_20772_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20772_end_0 = const()[name = string("op_20772_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_20772_end_mask_0 = const()[name = string("op_20772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20772_cast_fp16 = slice_by_index(begin = var_20772_begin_0, end = var_20772_end_0, end_mask = var_20772_end_mask_0, x = k_27_cast_fp16)[name = string("op_20772_cast_fp16")]; + tensor var_20776_begin_0 = const()[name = string("op_20776_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20776_end_0 = const()[name = string("op_20776_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_20776_end_mask_0 = const()[name = string("op_20776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20776_cast_fp16 = slice_by_index(begin = var_20776_begin_0, end = var_20776_end_0, end_mask = var_20776_end_mask_0, x = k_27_cast_fp16)[name = string("op_20776_cast_fp16")]; + tensor var_20780_begin_0 = const()[name = string("op_20780_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20780_end_0 = const()[name = string("op_20780_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_20780_end_mask_0 = const()[name = string("op_20780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20780_cast_fp16 = slice_by_index(begin = var_20780_begin_0, end = var_20780_end_0, end_mask = var_20780_end_mask_0, x = k_27_cast_fp16)[name = string("op_20780_cast_fp16")]; + tensor var_20784_begin_0 = const()[name = string("op_20784_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20784_end_0 = const()[name = string("op_20784_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_20784_end_mask_0 = const()[name = string("op_20784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20784_cast_fp16 = slice_by_index(begin = var_20784_begin_0, end = var_20784_end_0, end_mask = var_20784_end_mask_0, x = k_27_cast_fp16)[name = string("op_20784_cast_fp16")]; + tensor var_20788_begin_0 = const()[name = string("op_20788_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20788_end_0 = const()[name = string("op_20788_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_20788_end_mask_0 = const()[name = string("op_20788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20788_cast_fp16 = slice_by_index(begin = var_20788_begin_0, end = var_20788_end_0, end_mask = var_20788_end_mask_0, x = k_27_cast_fp16)[name = string("op_20788_cast_fp16")]; + tensor var_20792_begin_0 = const()[name = string("op_20792_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20792_end_0 = const()[name = string("op_20792_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_20792_end_mask_0 = const()[name = string("op_20792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20792_cast_fp16 = slice_by_index(begin = var_20792_begin_0, end = var_20792_end_0, end_mask = var_20792_end_mask_0, x = k_27_cast_fp16)[name = string("op_20792_cast_fp16")]; + tensor var_20796_begin_0 = const()[name = string("op_20796_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20796_end_0 = const()[name = string("op_20796_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_20796_end_mask_0 = const()[name = string("op_20796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20796_cast_fp16 = slice_by_index(begin = var_20796_begin_0, end = var_20796_end_0, end_mask = var_20796_end_mask_0, x = k_27_cast_fp16)[name = string("op_20796_cast_fp16")]; + tensor var_20800_begin_0 = const()[name = string("op_20800_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20800_end_0 = const()[name = string("op_20800_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_20800_end_mask_0 = const()[name = string("op_20800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20800_cast_fp16 = slice_by_index(begin = var_20800_begin_0, end = var_20800_end_0, end_mask = var_20800_end_mask_0, x = k_27_cast_fp16)[name = string("op_20800_cast_fp16")]; + tensor var_20804_begin_0 = const()[name = string("op_20804_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20804_end_0 = const()[name = string("op_20804_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_20804_end_mask_0 = const()[name = string("op_20804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20804_cast_fp16 = slice_by_index(begin = var_20804_begin_0, end = var_20804_end_0, end_mask = var_20804_end_mask_0, x = k_27_cast_fp16)[name = string("op_20804_cast_fp16")]; + tensor var_20808_begin_0 = const()[name = string("op_20808_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20808_end_0 = const()[name = string("op_20808_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_20808_end_mask_0 = const()[name = string("op_20808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20808_cast_fp16 = slice_by_index(begin = var_20808_begin_0, end = var_20808_end_0, end_mask = var_20808_end_mask_0, x = k_27_cast_fp16)[name = string("op_20808_cast_fp16")]; + tensor var_20812_begin_0 = const()[name = string("op_20812_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20812_end_0 = const()[name = string("op_20812_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_20812_end_mask_0 = const()[name = string("op_20812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20812_cast_fp16 = slice_by_index(begin = var_20812_begin_0, end = var_20812_end_0, end_mask = var_20812_end_mask_0, x = k_27_cast_fp16)[name = string("op_20812_cast_fp16")]; + tensor var_20816_begin_0 = const()[name = string("op_20816_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20816_end_0 = const()[name = string("op_20816_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_20816_end_mask_0 = const()[name = string("op_20816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20816_cast_fp16 = slice_by_index(begin = var_20816_begin_0, end = var_20816_end_0, end_mask = var_20816_end_mask_0, x = k_27_cast_fp16)[name = string("op_20816_cast_fp16")]; + tensor var_20820_begin_0 = const()[name = string("op_20820_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20820_end_0 = const()[name = string("op_20820_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_20820_end_mask_0 = const()[name = string("op_20820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20820_cast_fp16 = slice_by_index(begin = var_20820_begin_0, end = var_20820_end_0, end_mask = var_20820_end_mask_0, x = k_27_cast_fp16)[name = string("op_20820_cast_fp16")]; + tensor var_20824_begin_0 = const()[name = string("op_20824_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20824_end_0 = const()[name = string("op_20824_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_20824_end_mask_0 = const()[name = string("op_20824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20824_cast_fp16 = slice_by_index(begin = var_20824_begin_0, end = var_20824_end_0, end_mask = var_20824_end_mask_0, x = k_27_cast_fp16)[name = string("op_20824_cast_fp16")]; + tensor var_20828_begin_0 = const()[name = string("op_20828_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20828_end_0 = const()[name = string("op_20828_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_20828_end_mask_0 = const()[name = string("op_20828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20828_cast_fp16 = slice_by_index(begin = var_20828_begin_0, end = var_20828_end_0, end_mask = var_20828_end_mask_0, x = k_27_cast_fp16)[name = string("op_20828_cast_fp16")]; + tensor var_20830_begin_0 = const()[name = string("op_20830_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20830_end_0 = const()[name = string("op_20830_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20830_end_mask_0 = const()[name = string("op_20830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20830_cast_fp16 = slice_by_index(begin = var_20830_begin_0, end = var_20830_end_0, end_mask = var_20830_end_mask_0, x = value_27_cast_fp16)[name = string("op_20830_cast_fp16")]; + tensor var_20834_begin_0 = const()[name = string("op_20834_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20834_end_0 = const()[name = string("op_20834_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_20834_end_mask_0 = const()[name = string("op_20834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20834_cast_fp16 = slice_by_index(begin = var_20834_begin_0, end = var_20834_end_0, end_mask = var_20834_end_mask_0, x = value_27_cast_fp16)[name = string("op_20834_cast_fp16")]; + tensor var_20838_begin_0 = const()[name = string("op_20838_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20838_end_0 = const()[name = string("op_20838_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_20838_end_mask_0 = const()[name = string("op_20838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20838_cast_fp16 = slice_by_index(begin = var_20838_begin_0, end = var_20838_end_0, end_mask = var_20838_end_mask_0, x = value_27_cast_fp16)[name = string("op_20838_cast_fp16")]; + tensor var_20842_begin_0 = const()[name = string("op_20842_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20842_end_0 = const()[name = string("op_20842_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_20842_end_mask_0 = const()[name = string("op_20842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20842_cast_fp16 = slice_by_index(begin = var_20842_begin_0, end = var_20842_end_0, end_mask = var_20842_end_mask_0, x = value_27_cast_fp16)[name = string("op_20842_cast_fp16")]; + tensor var_20846_begin_0 = const()[name = string("op_20846_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20846_end_0 = const()[name = string("op_20846_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_20846_end_mask_0 = const()[name = string("op_20846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20846_cast_fp16 = slice_by_index(begin = var_20846_begin_0, end = var_20846_end_0, end_mask = var_20846_end_mask_0, x = value_27_cast_fp16)[name = string("op_20846_cast_fp16")]; + tensor var_20850_begin_0 = const()[name = string("op_20850_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20850_end_0 = const()[name = string("op_20850_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_20850_end_mask_0 = const()[name = string("op_20850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20850_cast_fp16 = slice_by_index(begin = var_20850_begin_0, end = var_20850_end_0, end_mask = var_20850_end_mask_0, x = value_27_cast_fp16)[name = string("op_20850_cast_fp16")]; + tensor var_20854_begin_0 = const()[name = string("op_20854_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20854_end_0 = const()[name = string("op_20854_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_20854_end_mask_0 = const()[name = string("op_20854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20854_cast_fp16 = slice_by_index(begin = var_20854_begin_0, end = var_20854_end_0, end_mask = var_20854_end_mask_0, x = value_27_cast_fp16)[name = string("op_20854_cast_fp16")]; + tensor var_20858_begin_0 = const()[name = string("op_20858_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20858_end_0 = const()[name = string("op_20858_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_20858_end_mask_0 = const()[name = string("op_20858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20858_cast_fp16 = slice_by_index(begin = var_20858_begin_0, end = var_20858_end_0, end_mask = var_20858_end_mask_0, x = value_27_cast_fp16)[name = string("op_20858_cast_fp16")]; + tensor var_20862_begin_0 = const()[name = string("op_20862_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20862_end_0 = const()[name = string("op_20862_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_20862_end_mask_0 = const()[name = string("op_20862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20862_cast_fp16 = slice_by_index(begin = var_20862_begin_0, end = var_20862_end_0, end_mask = var_20862_end_mask_0, x = value_27_cast_fp16)[name = string("op_20862_cast_fp16")]; + tensor var_20866_begin_0 = const()[name = string("op_20866_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20866_end_0 = const()[name = string("op_20866_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_20866_end_mask_0 = const()[name = string("op_20866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20866_cast_fp16 = slice_by_index(begin = var_20866_begin_0, end = var_20866_end_0, end_mask = var_20866_end_mask_0, x = value_27_cast_fp16)[name = string("op_20866_cast_fp16")]; + tensor var_20870_begin_0 = const()[name = string("op_20870_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20870_end_0 = const()[name = string("op_20870_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_20870_end_mask_0 = const()[name = string("op_20870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20870_cast_fp16 = slice_by_index(begin = var_20870_begin_0, end = var_20870_end_0, end_mask = var_20870_end_mask_0, x = value_27_cast_fp16)[name = string("op_20870_cast_fp16")]; + tensor var_20874_begin_0 = const()[name = string("op_20874_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20874_end_0 = const()[name = string("op_20874_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_20874_end_mask_0 = const()[name = string("op_20874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20874_cast_fp16 = slice_by_index(begin = var_20874_begin_0, end = var_20874_end_0, end_mask = var_20874_end_mask_0, x = value_27_cast_fp16)[name = string("op_20874_cast_fp16")]; + tensor var_20878_begin_0 = const()[name = string("op_20878_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20878_end_0 = const()[name = string("op_20878_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_20878_end_mask_0 = const()[name = string("op_20878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20878_cast_fp16 = slice_by_index(begin = var_20878_begin_0, end = var_20878_end_0, end_mask = var_20878_end_mask_0, x = value_27_cast_fp16)[name = string("op_20878_cast_fp16")]; + tensor var_20882_begin_0 = const()[name = string("op_20882_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20882_end_0 = const()[name = string("op_20882_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_20882_end_mask_0 = const()[name = string("op_20882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20882_cast_fp16 = slice_by_index(begin = var_20882_begin_0, end = var_20882_end_0, end_mask = var_20882_end_mask_0, x = value_27_cast_fp16)[name = string("op_20882_cast_fp16")]; + tensor var_20886_begin_0 = const()[name = string("op_20886_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20886_end_0 = const()[name = string("op_20886_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_20886_end_mask_0 = const()[name = string("op_20886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20886_cast_fp16 = slice_by_index(begin = var_20886_begin_0, end = var_20886_end_0, end_mask = var_20886_end_mask_0, x = value_27_cast_fp16)[name = string("op_20886_cast_fp16")]; + tensor var_20890_begin_0 = const()[name = string("op_20890_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20890_end_0 = const()[name = string("op_20890_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_20890_end_mask_0 = const()[name = string("op_20890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20890_cast_fp16 = slice_by_index(begin = var_20890_begin_0, end = var_20890_end_0, end_mask = var_20890_end_mask_0, x = value_27_cast_fp16)[name = string("op_20890_cast_fp16")]; + tensor var_20894_begin_0 = const()[name = string("op_20894_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20894_end_0 = const()[name = string("op_20894_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_20894_end_mask_0 = const()[name = string("op_20894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20894_cast_fp16 = slice_by_index(begin = var_20894_begin_0, end = var_20894_end_0, end_mask = var_20894_end_mask_0, x = value_27_cast_fp16)[name = string("op_20894_cast_fp16")]; + tensor var_20898_begin_0 = const()[name = string("op_20898_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20898_end_0 = const()[name = string("op_20898_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_20898_end_mask_0 = const()[name = string("op_20898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20898_cast_fp16 = slice_by_index(begin = var_20898_begin_0, end = var_20898_end_0, end_mask = var_20898_end_mask_0, x = value_27_cast_fp16)[name = string("op_20898_cast_fp16")]; + tensor var_20902_begin_0 = const()[name = string("op_20902_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20902_end_0 = const()[name = string("op_20902_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_20902_end_mask_0 = const()[name = string("op_20902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20902_cast_fp16 = slice_by_index(begin = var_20902_begin_0, end = var_20902_end_0, end_mask = var_20902_end_mask_0, x = value_27_cast_fp16)[name = string("op_20902_cast_fp16")]; + tensor var_20906_begin_0 = const()[name = string("op_20906_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20906_end_0 = const()[name = string("op_20906_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_20906_end_mask_0 = const()[name = string("op_20906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20906_cast_fp16 = slice_by_index(begin = var_20906_begin_0, end = var_20906_end_0, end_mask = var_20906_end_mask_0, x = value_27_cast_fp16)[name = string("op_20906_cast_fp16")]; + string _SplitHeadsQ__mh_w_2081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2081_equation_0, values = (var_20752_cast_fp16, var_20194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2081_cast_fp16")]; + string _SplitHeadsQ__mh_w_2083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2083_equation_0, values = (var_20752_cast_fp16, var_20201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2083_cast_fp16")]; + string _SplitHeadsQ__mh_w_2085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2085_equation_0, values = (var_20752_cast_fp16, var_20208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2085_cast_fp16")]; + string _SplitHeadsQ__mh_w_2087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2087_equation_0, values = (var_20752_cast_fp16, var_20215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2087_cast_fp16")]; + string _SplitHeadsQ__mh_w_2089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2089_equation_0, values = (var_20756_cast_fp16, var_20222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2089_cast_fp16")]; + string _SplitHeadsQ__mh_w_2091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2091_equation_0, values = (var_20756_cast_fp16, var_20229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2091_cast_fp16")]; + string _SplitHeadsQ__mh_w_2093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2093_equation_0, values = (var_20756_cast_fp16, var_20236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2093_cast_fp16")]; + string _SplitHeadsQ__mh_w_2095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2095_equation_0, values = (var_20756_cast_fp16, var_20243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2095_cast_fp16")]; + string _SplitHeadsQ__mh_w_2097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2097_equation_0, values = (var_20760_cast_fp16, var_20250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2097_cast_fp16")]; + string _SplitHeadsQ__mh_w_2099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2099_equation_0, values = (var_20760_cast_fp16, var_20257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2099_cast_fp16")]; + string _SplitHeadsQ__mh_w_2101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2101_equation_0, values = (var_20760_cast_fp16, var_20264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2101_cast_fp16")]; + string _SplitHeadsQ__mh_w_2103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2103_equation_0, values = (var_20760_cast_fp16, var_20271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2103_cast_fp16")]; + string _SplitHeadsQ__mh_w_2105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2105_equation_0, values = (var_20764_cast_fp16, var_20278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2105_cast_fp16")]; + string _SplitHeadsQ__mh_w_2107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2107_equation_0, values = (var_20764_cast_fp16, var_20285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2107_cast_fp16")]; + string _SplitHeadsQ__mh_w_2109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2109_equation_0, values = (var_20764_cast_fp16, var_20292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2109_cast_fp16")]; + string _SplitHeadsQ__mh_w_2111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2111_equation_0, values = (var_20764_cast_fp16, var_20299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2111_cast_fp16")]; + string _SplitHeadsQ__mh_w_2113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2113_equation_0, values = (var_20768_cast_fp16, var_20306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2113_cast_fp16")]; + string _SplitHeadsQ__mh_w_2115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2115_equation_0, values = (var_20768_cast_fp16, var_20313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2115_cast_fp16")]; + string _SplitHeadsQ__mh_w_2117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2117_equation_0, values = (var_20768_cast_fp16, var_20320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2117_cast_fp16")]; + string _SplitHeadsQ__mh_w_2119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2119_equation_0, values = (var_20768_cast_fp16, var_20327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2119_cast_fp16")]; + string _SplitHeadsQ__mh_w_2121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2121_equation_0, values = (var_20772_cast_fp16, var_20334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2121_cast_fp16")]; + string _SplitHeadsQ__mh_w_2123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2123_equation_0, values = (var_20772_cast_fp16, var_20341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2123_cast_fp16")]; + string _SplitHeadsQ__mh_w_2125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2125_equation_0, values = (var_20772_cast_fp16, var_20348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2125_cast_fp16")]; + string _SplitHeadsQ__mh_w_2127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2127_equation_0, values = (var_20772_cast_fp16, var_20355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2127_cast_fp16")]; + string _SplitHeadsQ__mh_w_2129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2129_equation_0, values = (var_20776_cast_fp16, var_20362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2129_cast_fp16")]; + string _SplitHeadsQ__mh_w_2131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2131_equation_0, values = (var_20776_cast_fp16, var_20369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2131_cast_fp16")]; + string _SplitHeadsQ__mh_w_2133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2133_equation_0, values = (var_20776_cast_fp16, var_20376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2133_cast_fp16")]; + string _SplitHeadsQ__mh_w_2135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2135_equation_0, values = (var_20776_cast_fp16, var_20383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2135_cast_fp16")]; + string _SplitHeadsQ__mh_w_2137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2137_equation_0, values = (var_20780_cast_fp16, var_20390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2137_cast_fp16")]; + string _SplitHeadsQ__mh_w_2139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2139_equation_0, values = (var_20780_cast_fp16, var_20397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2139_cast_fp16")]; + string _SplitHeadsQ__mh_w_2141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2141_equation_0, values = (var_20780_cast_fp16, var_20404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2141_cast_fp16")]; + string _SplitHeadsQ__mh_w_2143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2143_equation_0, values = (var_20780_cast_fp16, var_20411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2143_cast_fp16")]; + string _SplitHeadsQ__mh_w_2145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2145_equation_0, values = (var_20784_cast_fp16, var_20418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2145_cast_fp16")]; + string _SplitHeadsQ__mh_w_2147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2147_equation_0, values = (var_20784_cast_fp16, var_20425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2147_cast_fp16")]; + string _SplitHeadsQ__mh_w_2149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2149_equation_0, values = (var_20784_cast_fp16, var_20432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2149_cast_fp16")]; + string _SplitHeadsQ__mh_w_2151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2151_equation_0, values = (var_20784_cast_fp16, var_20439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2151_cast_fp16")]; + string _SplitHeadsQ__mh_w_2153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2153_equation_0, values = (var_20788_cast_fp16, var_20446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2153_cast_fp16")]; + string _SplitHeadsQ__mh_w_2155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2155_equation_0, values = (var_20788_cast_fp16, var_20453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2155_cast_fp16")]; + string _SplitHeadsQ__mh_w_2157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2157_equation_0, values = (var_20788_cast_fp16, var_20460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2157_cast_fp16")]; + string _SplitHeadsQ__mh_w_2159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2159_equation_0, values = (var_20788_cast_fp16, var_20467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2159_cast_fp16")]; + string _SplitHeadsQ__mh_w_2161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2161_equation_0, values = (var_20792_cast_fp16, var_20474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2161_cast_fp16")]; + string _SplitHeadsQ__mh_w_2163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2163_equation_0, values = (var_20792_cast_fp16, var_20481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2163_cast_fp16")]; + string _SplitHeadsQ__mh_w_2165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2165_equation_0, values = (var_20792_cast_fp16, var_20488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2165_cast_fp16")]; + string _SplitHeadsQ__mh_w_2167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2167_equation_0, values = (var_20792_cast_fp16, var_20495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2167_cast_fp16")]; + string _SplitHeadsQ__mh_w_2169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2169_equation_0, values = (var_20796_cast_fp16, var_20502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2169_cast_fp16")]; + string _SplitHeadsQ__mh_w_2171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2171_equation_0, values = (var_20796_cast_fp16, var_20509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2171_cast_fp16")]; + string _SplitHeadsQ__mh_w_2173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2173_equation_0, values = (var_20796_cast_fp16, var_20516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2173_cast_fp16")]; + string _SplitHeadsQ__mh_w_2175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2175_equation_0, values = (var_20796_cast_fp16, var_20523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2175_cast_fp16")]; + string _SplitHeadsQ__mh_w_2177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2177_equation_0, values = (var_20800_cast_fp16, var_20530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2177_cast_fp16")]; + string _SplitHeadsQ__mh_w_2179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2179_equation_0, values = (var_20800_cast_fp16, var_20537_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2179_cast_fp16")]; + string _SplitHeadsQ__mh_w_2181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2181_equation_0, values = (var_20800_cast_fp16, var_20544_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2181_cast_fp16")]; + string _SplitHeadsQ__mh_w_2183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2183_equation_0, values = (var_20800_cast_fp16, var_20551_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2183_cast_fp16")]; + string _SplitHeadsQ__mh_w_2185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2185_equation_0, values = (var_20804_cast_fp16, var_20558_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2185_cast_fp16")]; + string _SplitHeadsQ__mh_w_2187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2187_equation_0, values = (var_20804_cast_fp16, var_20565_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2187_cast_fp16")]; + string _SplitHeadsQ__mh_w_2189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2189_equation_0, values = (var_20804_cast_fp16, var_20572_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2189_cast_fp16")]; + string _SplitHeadsQ__mh_w_2191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2191_equation_0, values = (var_20804_cast_fp16, var_20579_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2191_cast_fp16")]; + string _SplitHeadsQ__mh_w_2193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2193_equation_0, values = (var_20808_cast_fp16, var_20586_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2193_cast_fp16")]; + string _SplitHeadsQ__mh_w_2195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2195_equation_0, values = (var_20808_cast_fp16, var_20593_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2195_cast_fp16")]; + string _SplitHeadsQ__mh_w_2197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2197_equation_0, values = (var_20808_cast_fp16, var_20600_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2197_cast_fp16")]; + string _SplitHeadsQ__mh_w_2199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2199_equation_0, values = (var_20808_cast_fp16, var_20607_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2199_cast_fp16")]; + string _SplitHeadsQ__mh_w_2201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2201_equation_0, values = (var_20812_cast_fp16, var_20614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2201_cast_fp16")]; + string _SplitHeadsQ__mh_w_2203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2203_equation_0, values = (var_20812_cast_fp16, var_20621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2203_cast_fp16")]; + string _SplitHeadsQ__mh_w_2205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2205_equation_0, values = (var_20812_cast_fp16, var_20628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2205_cast_fp16")]; + string _SplitHeadsQ__mh_w_2207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2207_equation_0, values = (var_20812_cast_fp16, var_20635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2207_cast_fp16")]; + string _SplitHeadsQ__mh_w_2209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2209_equation_0, values = (var_20816_cast_fp16, var_20642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2209_cast_fp16")]; + string _SplitHeadsQ__mh_w_2211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2211_equation_0, values = (var_20816_cast_fp16, var_20649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2211_cast_fp16")]; + string _SplitHeadsQ__mh_w_2213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2213_equation_0, values = (var_20816_cast_fp16, var_20656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2213_cast_fp16")]; + string _SplitHeadsQ__mh_w_2215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2215_equation_0, values = (var_20816_cast_fp16, var_20663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2215_cast_fp16")]; + string _SplitHeadsQ__mh_w_2217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2217_equation_0, values = (var_20820_cast_fp16, var_20670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2217_cast_fp16")]; + string _SplitHeadsQ__mh_w_2219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2219_equation_0, values = (var_20820_cast_fp16, var_20677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2219_cast_fp16")]; + string _SplitHeadsQ__mh_w_2221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2221_equation_0, values = (var_20820_cast_fp16, var_20684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2221_cast_fp16")]; + string _SplitHeadsQ__mh_w_2223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2223_equation_0, values = (var_20820_cast_fp16, var_20691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2223_cast_fp16")]; + string _SplitHeadsQ__mh_w_2225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2225_equation_0, values = (var_20824_cast_fp16, var_20698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2225_cast_fp16")]; + string _SplitHeadsQ__mh_w_2227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2227_equation_0, values = (var_20824_cast_fp16, var_20705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2227_cast_fp16")]; + string _SplitHeadsQ__mh_w_2229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2229_equation_0, values = (var_20824_cast_fp16, var_20712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2229_cast_fp16")]; + string _SplitHeadsQ__mh_w_2231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2231_equation_0, values = (var_20824_cast_fp16, var_20719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2231_cast_fp16")]; + string _SplitHeadsQ__mh_w_2233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2233_equation_0, values = (var_20828_cast_fp16, var_20726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2233_cast_fp16")]; + string _SplitHeadsQ__mh_w_2235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2235_equation_0, values = (var_20828_cast_fp16, var_20733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2235_cast_fp16")]; + string _SplitHeadsQ__mh_w_2237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2237_equation_0, values = (var_20828_cast_fp16, var_20740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2237_cast_fp16")]; + string _SplitHeadsQ__mh_w_2239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2239_equation_0, values = (var_20828_cast_fp16, var_20747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2239_cast_fp16")]; + fp16 var_21069_to_fp16 = const()[name = string("op_21069_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2081_cast_fp16, y = var_21069_to_fp16)[name = string("aw_chunk_2081_cast_fp16")]; + fp16 var_21071_to_fp16 = const()[name = string("op_21071_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2083_cast_fp16, y = var_21071_to_fp16)[name = string("aw_chunk_2083_cast_fp16")]; + fp16 var_21073_to_fp16 = const()[name = string("op_21073_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2085_cast_fp16, y = var_21073_to_fp16)[name = string("aw_chunk_2085_cast_fp16")]; + fp16 var_21075_to_fp16 = const()[name = string("op_21075_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2087_cast_fp16, y = var_21075_to_fp16)[name = string("aw_chunk_2087_cast_fp16")]; + fp16 var_21077_to_fp16 = const()[name = string("op_21077_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2089_cast_fp16, y = var_21077_to_fp16)[name = string("aw_chunk_2089_cast_fp16")]; + fp16 var_21079_to_fp16 = const()[name = string("op_21079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2091_cast_fp16, y = var_21079_to_fp16)[name = string("aw_chunk_2091_cast_fp16")]; + fp16 var_21081_to_fp16 = const()[name = string("op_21081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2093_cast_fp16, y = var_21081_to_fp16)[name = string("aw_chunk_2093_cast_fp16")]; + fp16 var_21083_to_fp16 = const()[name = string("op_21083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2095_cast_fp16, y = var_21083_to_fp16)[name = string("aw_chunk_2095_cast_fp16")]; + fp16 var_21085_to_fp16 = const()[name = string("op_21085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2097_cast_fp16, y = var_21085_to_fp16)[name = string("aw_chunk_2097_cast_fp16")]; + fp16 var_21087_to_fp16 = const()[name = string("op_21087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2099_cast_fp16, y = var_21087_to_fp16)[name = string("aw_chunk_2099_cast_fp16")]; + fp16 var_21089_to_fp16 = const()[name = string("op_21089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2101_cast_fp16, y = var_21089_to_fp16)[name = string("aw_chunk_2101_cast_fp16")]; + fp16 var_21091_to_fp16 = const()[name = string("op_21091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2103_cast_fp16, y = var_21091_to_fp16)[name = string("aw_chunk_2103_cast_fp16")]; + fp16 var_21093_to_fp16 = const()[name = string("op_21093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2105_cast_fp16, y = var_21093_to_fp16)[name = string("aw_chunk_2105_cast_fp16")]; + fp16 var_21095_to_fp16 = const()[name = string("op_21095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2107_cast_fp16, y = var_21095_to_fp16)[name = string("aw_chunk_2107_cast_fp16")]; + fp16 var_21097_to_fp16 = const()[name = string("op_21097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2109_cast_fp16, y = var_21097_to_fp16)[name = string("aw_chunk_2109_cast_fp16")]; + fp16 var_21099_to_fp16 = const()[name = string("op_21099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2111_cast_fp16, y = var_21099_to_fp16)[name = string("aw_chunk_2111_cast_fp16")]; + fp16 var_21101_to_fp16 = const()[name = string("op_21101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2113_cast_fp16, y = var_21101_to_fp16)[name = string("aw_chunk_2113_cast_fp16")]; + fp16 var_21103_to_fp16 = const()[name = string("op_21103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2115_cast_fp16, y = var_21103_to_fp16)[name = string("aw_chunk_2115_cast_fp16")]; + fp16 var_21105_to_fp16 = const()[name = string("op_21105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2117_cast_fp16, y = var_21105_to_fp16)[name = string("aw_chunk_2117_cast_fp16")]; + fp16 var_21107_to_fp16 = const()[name = string("op_21107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2119_cast_fp16, y = var_21107_to_fp16)[name = string("aw_chunk_2119_cast_fp16")]; + fp16 var_21109_to_fp16 = const()[name = string("op_21109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2121_cast_fp16, y = var_21109_to_fp16)[name = string("aw_chunk_2121_cast_fp16")]; + fp16 var_21111_to_fp16 = const()[name = string("op_21111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2123_cast_fp16, y = var_21111_to_fp16)[name = string("aw_chunk_2123_cast_fp16")]; + fp16 var_21113_to_fp16 = const()[name = string("op_21113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2125_cast_fp16, y = var_21113_to_fp16)[name = string("aw_chunk_2125_cast_fp16")]; + fp16 var_21115_to_fp16 = const()[name = string("op_21115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2127_cast_fp16, y = var_21115_to_fp16)[name = string("aw_chunk_2127_cast_fp16")]; + fp16 var_21117_to_fp16 = const()[name = string("op_21117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2129_cast_fp16, y = var_21117_to_fp16)[name = string("aw_chunk_2129_cast_fp16")]; + fp16 var_21119_to_fp16 = const()[name = string("op_21119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2131_cast_fp16, y = var_21119_to_fp16)[name = string("aw_chunk_2131_cast_fp16")]; + fp16 var_21121_to_fp16 = const()[name = string("op_21121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2133_cast_fp16, y = var_21121_to_fp16)[name = string("aw_chunk_2133_cast_fp16")]; + fp16 var_21123_to_fp16 = const()[name = string("op_21123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2135_cast_fp16, y = var_21123_to_fp16)[name = string("aw_chunk_2135_cast_fp16")]; + fp16 var_21125_to_fp16 = const()[name = string("op_21125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2137_cast_fp16, y = var_21125_to_fp16)[name = string("aw_chunk_2137_cast_fp16")]; + fp16 var_21127_to_fp16 = const()[name = string("op_21127_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2139_cast_fp16, y = var_21127_to_fp16)[name = string("aw_chunk_2139_cast_fp16")]; + fp16 var_21129_to_fp16 = const()[name = string("op_21129_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2141_cast_fp16, y = var_21129_to_fp16)[name = string("aw_chunk_2141_cast_fp16")]; + fp16 var_21131_to_fp16 = const()[name = string("op_21131_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2143_cast_fp16, y = var_21131_to_fp16)[name = string("aw_chunk_2143_cast_fp16")]; + fp16 var_21133_to_fp16 = const()[name = string("op_21133_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2145_cast_fp16, y = var_21133_to_fp16)[name = string("aw_chunk_2145_cast_fp16")]; + fp16 var_21135_to_fp16 = const()[name = string("op_21135_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2147_cast_fp16, y = var_21135_to_fp16)[name = string("aw_chunk_2147_cast_fp16")]; + fp16 var_21137_to_fp16 = const()[name = string("op_21137_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2149_cast_fp16, y = var_21137_to_fp16)[name = string("aw_chunk_2149_cast_fp16")]; + fp16 var_21139_to_fp16 = const()[name = string("op_21139_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2151_cast_fp16, y = var_21139_to_fp16)[name = string("aw_chunk_2151_cast_fp16")]; + fp16 var_21141_to_fp16 = const()[name = string("op_21141_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2153_cast_fp16, y = var_21141_to_fp16)[name = string("aw_chunk_2153_cast_fp16")]; + fp16 var_21143_to_fp16 = const()[name = string("op_21143_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2155_cast_fp16, y = var_21143_to_fp16)[name = string("aw_chunk_2155_cast_fp16")]; + fp16 var_21145_to_fp16 = const()[name = string("op_21145_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2157_cast_fp16, y = var_21145_to_fp16)[name = string("aw_chunk_2157_cast_fp16")]; + fp16 var_21147_to_fp16 = const()[name = string("op_21147_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2159_cast_fp16, y = var_21147_to_fp16)[name = string("aw_chunk_2159_cast_fp16")]; + fp16 var_21149_to_fp16 = const()[name = string("op_21149_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2161_cast_fp16, y = var_21149_to_fp16)[name = string("aw_chunk_2161_cast_fp16")]; + fp16 var_21151_to_fp16 = const()[name = string("op_21151_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2163_cast_fp16, y = var_21151_to_fp16)[name = string("aw_chunk_2163_cast_fp16")]; + fp16 var_21153_to_fp16 = const()[name = string("op_21153_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2165_cast_fp16, y = var_21153_to_fp16)[name = string("aw_chunk_2165_cast_fp16")]; + fp16 var_21155_to_fp16 = const()[name = string("op_21155_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2167_cast_fp16, y = var_21155_to_fp16)[name = string("aw_chunk_2167_cast_fp16")]; + fp16 var_21157_to_fp16 = const()[name = string("op_21157_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2169_cast_fp16, y = var_21157_to_fp16)[name = string("aw_chunk_2169_cast_fp16")]; + fp16 var_21159_to_fp16 = const()[name = string("op_21159_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2171_cast_fp16, y = var_21159_to_fp16)[name = string("aw_chunk_2171_cast_fp16")]; + fp16 var_21161_to_fp16 = const()[name = string("op_21161_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2173_cast_fp16, y = var_21161_to_fp16)[name = string("aw_chunk_2173_cast_fp16")]; + fp16 var_21163_to_fp16 = const()[name = string("op_21163_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2175_cast_fp16, y = var_21163_to_fp16)[name = string("aw_chunk_2175_cast_fp16")]; + fp16 var_21165_to_fp16 = const()[name = string("op_21165_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2177_cast_fp16, y = var_21165_to_fp16)[name = string("aw_chunk_2177_cast_fp16")]; + fp16 var_21167_to_fp16 = const()[name = string("op_21167_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2179_cast_fp16, y = var_21167_to_fp16)[name = string("aw_chunk_2179_cast_fp16")]; + fp16 var_21169_to_fp16 = const()[name = string("op_21169_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2181_cast_fp16, y = var_21169_to_fp16)[name = string("aw_chunk_2181_cast_fp16")]; + fp16 var_21171_to_fp16 = const()[name = string("op_21171_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2183_cast_fp16, y = var_21171_to_fp16)[name = string("aw_chunk_2183_cast_fp16")]; + fp16 var_21173_to_fp16 = const()[name = string("op_21173_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2185_cast_fp16, y = var_21173_to_fp16)[name = string("aw_chunk_2185_cast_fp16")]; + fp16 var_21175_to_fp16 = const()[name = string("op_21175_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2187_cast_fp16, y = var_21175_to_fp16)[name = string("aw_chunk_2187_cast_fp16")]; + fp16 var_21177_to_fp16 = const()[name = string("op_21177_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2189_cast_fp16, y = var_21177_to_fp16)[name = string("aw_chunk_2189_cast_fp16")]; + fp16 var_21179_to_fp16 = const()[name = string("op_21179_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2191_cast_fp16, y = var_21179_to_fp16)[name = string("aw_chunk_2191_cast_fp16")]; + fp16 var_21181_to_fp16 = const()[name = string("op_21181_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2193_cast_fp16, y = var_21181_to_fp16)[name = string("aw_chunk_2193_cast_fp16")]; + fp16 var_21183_to_fp16 = const()[name = string("op_21183_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2195_cast_fp16, y = var_21183_to_fp16)[name = string("aw_chunk_2195_cast_fp16")]; + fp16 var_21185_to_fp16 = const()[name = string("op_21185_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2197_cast_fp16, y = var_21185_to_fp16)[name = string("aw_chunk_2197_cast_fp16")]; + fp16 var_21187_to_fp16 = const()[name = string("op_21187_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2199_cast_fp16, y = var_21187_to_fp16)[name = string("aw_chunk_2199_cast_fp16")]; + fp16 var_21189_to_fp16 = const()[name = string("op_21189_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2201_cast_fp16, y = var_21189_to_fp16)[name = string("aw_chunk_2201_cast_fp16")]; + fp16 var_21191_to_fp16 = const()[name = string("op_21191_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2203_cast_fp16, y = var_21191_to_fp16)[name = string("aw_chunk_2203_cast_fp16")]; + fp16 var_21193_to_fp16 = const()[name = string("op_21193_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2205_cast_fp16, y = var_21193_to_fp16)[name = string("aw_chunk_2205_cast_fp16")]; + fp16 var_21195_to_fp16 = const()[name = string("op_21195_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2207_cast_fp16, y = var_21195_to_fp16)[name = string("aw_chunk_2207_cast_fp16")]; + fp16 var_21197_to_fp16 = const()[name = string("op_21197_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2209_cast_fp16, y = var_21197_to_fp16)[name = string("aw_chunk_2209_cast_fp16")]; + fp16 var_21199_to_fp16 = const()[name = string("op_21199_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2211_cast_fp16, y = var_21199_to_fp16)[name = string("aw_chunk_2211_cast_fp16")]; + fp16 var_21201_to_fp16 = const()[name = string("op_21201_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2213_cast_fp16, y = var_21201_to_fp16)[name = string("aw_chunk_2213_cast_fp16")]; + fp16 var_21203_to_fp16 = const()[name = string("op_21203_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2215_cast_fp16, y = var_21203_to_fp16)[name = string("aw_chunk_2215_cast_fp16")]; + fp16 var_21205_to_fp16 = const()[name = string("op_21205_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2217_cast_fp16, y = var_21205_to_fp16)[name = string("aw_chunk_2217_cast_fp16")]; + fp16 var_21207_to_fp16 = const()[name = string("op_21207_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2219_cast_fp16, y = var_21207_to_fp16)[name = string("aw_chunk_2219_cast_fp16")]; + fp16 var_21209_to_fp16 = const()[name = string("op_21209_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2221_cast_fp16, y = var_21209_to_fp16)[name = string("aw_chunk_2221_cast_fp16")]; + fp16 var_21211_to_fp16 = const()[name = string("op_21211_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2223_cast_fp16, y = var_21211_to_fp16)[name = string("aw_chunk_2223_cast_fp16")]; + fp16 var_21213_to_fp16 = const()[name = string("op_21213_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2225_cast_fp16, y = var_21213_to_fp16)[name = string("aw_chunk_2225_cast_fp16")]; + fp16 var_21215_to_fp16 = const()[name = string("op_21215_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2227_cast_fp16, y = var_21215_to_fp16)[name = string("aw_chunk_2227_cast_fp16")]; + fp16 var_21217_to_fp16 = const()[name = string("op_21217_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2229_cast_fp16, y = var_21217_to_fp16)[name = string("aw_chunk_2229_cast_fp16")]; + fp16 var_21219_to_fp16 = const()[name = string("op_21219_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2231_cast_fp16, y = var_21219_to_fp16)[name = string("aw_chunk_2231_cast_fp16")]; + fp16 var_21221_to_fp16 = const()[name = string("op_21221_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2233_cast_fp16, y = var_21221_to_fp16)[name = string("aw_chunk_2233_cast_fp16")]; + fp16 var_21223_to_fp16 = const()[name = string("op_21223_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2235_cast_fp16, y = var_21223_to_fp16)[name = string("aw_chunk_2235_cast_fp16")]; + fp16 var_21225_to_fp16 = const()[name = string("op_21225_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2237_cast_fp16, y = var_21225_to_fp16)[name = string("aw_chunk_2237_cast_fp16")]; + fp16 var_21227_to_fp16 = const()[name = string("op_21227_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2239_cast_fp16, y = var_21227_to_fp16)[name = string("aw_chunk_2239_cast_fp16")]; + tensor var_21229_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2081_cast_fp16)[name = string("op_21229_cast_fp16")]; + tensor var_21230_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2083_cast_fp16)[name = string("op_21230_cast_fp16")]; + tensor var_21231_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2085_cast_fp16)[name = string("op_21231_cast_fp16")]; + tensor var_21232_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2087_cast_fp16)[name = string("op_21232_cast_fp16")]; + tensor var_21233_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2089_cast_fp16)[name = string("op_21233_cast_fp16")]; + tensor var_21234_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2091_cast_fp16)[name = string("op_21234_cast_fp16")]; + tensor var_21235_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2093_cast_fp16)[name = string("op_21235_cast_fp16")]; + tensor var_21236_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2095_cast_fp16)[name = string("op_21236_cast_fp16")]; + tensor var_21237_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2097_cast_fp16)[name = string("op_21237_cast_fp16")]; + tensor var_21238_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2099_cast_fp16)[name = string("op_21238_cast_fp16")]; + tensor var_21239_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2101_cast_fp16)[name = string("op_21239_cast_fp16")]; + tensor var_21240_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2103_cast_fp16)[name = string("op_21240_cast_fp16")]; + tensor var_21241_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2105_cast_fp16)[name = string("op_21241_cast_fp16")]; + tensor var_21242_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2107_cast_fp16)[name = string("op_21242_cast_fp16")]; + tensor var_21243_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2109_cast_fp16)[name = string("op_21243_cast_fp16")]; + tensor var_21244_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2111_cast_fp16)[name = string("op_21244_cast_fp16")]; + tensor var_21245_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2113_cast_fp16)[name = string("op_21245_cast_fp16")]; + tensor var_21246_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2115_cast_fp16)[name = string("op_21246_cast_fp16")]; + tensor var_21247_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2117_cast_fp16)[name = string("op_21247_cast_fp16")]; + tensor var_21248_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2119_cast_fp16)[name = string("op_21248_cast_fp16")]; + tensor var_21249_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2121_cast_fp16)[name = string("op_21249_cast_fp16")]; + tensor var_21250_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2123_cast_fp16)[name = string("op_21250_cast_fp16")]; + tensor var_21251_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2125_cast_fp16)[name = string("op_21251_cast_fp16")]; + tensor var_21252_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2127_cast_fp16)[name = string("op_21252_cast_fp16")]; + tensor var_21253_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2129_cast_fp16)[name = string("op_21253_cast_fp16")]; + tensor var_21254_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2131_cast_fp16)[name = string("op_21254_cast_fp16")]; + tensor var_21255_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2133_cast_fp16)[name = string("op_21255_cast_fp16")]; + tensor var_21256_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2135_cast_fp16)[name = string("op_21256_cast_fp16")]; + tensor var_21257_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2137_cast_fp16)[name = string("op_21257_cast_fp16")]; + tensor var_21258_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2139_cast_fp16)[name = string("op_21258_cast_fp16")]; + tensor var_21259_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2141_cast_fp16)[name = string("op_21259_cast_fp16")]; + tensor var_21260_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2143_cast_fp16)[name = string("op_21260_cast_fp16")]; + tensor var_21261_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2145_cast_fp16)[name = string("op_21261_cast_fp16")]; + tensor var_21262_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2147_cast_fp16)[name = string("op_21262_cast_fp16")]; + tensor var_21263_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2149_cast_fp16)[name = string("op_21263_cast_fp16")]; + tensor var_21264_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2151_cast_fp16)[name = string("op_21264_cast_fp16")]; + tensor var_21265_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2153_cast_fp16)[name = string("op_21265_cast_fp16")]; + tensor var_21266_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2155_cast_fp16)[name = string("op_21266_cast_fp16")]; + tensor var_21267_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2157_cast_fp16)[name = string("op_21267_cast_fp16")]; + tensor var_21268_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2159_cast_fp16)[name = string("op_21268_cast_fp16")]; + tensor var_21269_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2161_cast_fp16)[name = string("op_21269_cast_fp16")]; + tensor var_21270_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2163_cast_fp16)[name = string("op_21270_cast_fp16")]; + tensor var_21271_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2165_cast_fp16)[name = string("op_21271_cast_fp16")]; + tensor var_21272_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2167_cast_fp16)[name = string("op_21272_cast_fp16")]; + tensor var_21273_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2169_cast_fp16)[name = string("op_21273_cast_fp16")]; + tensor var_21274_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2171_cast_fp16)[name = string("op_21274_cast_fp16")]; + tensor var_21275_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2173_cast_fp16)[name = string("op_21275_cast_fp16")]; + tensor var_21276_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2175_cast_fp16)[name = string("op_21276_cast_fp16")]; + tensor var_21277_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2177_cast_fp16)[name = string("op_21277_cast_fp16")]; + tensor var_21278_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2179_cast_fp16)[name = string("op_21278_cast_fp16")]; + tensor var_21279_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2181_cast_fp16)[name = string("op_21279_cast_fp16")]; + tensor var_21280_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2183_cast_fp16)[name = string("op_21280_cast_fp16")]; + tensor var_21281_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2185_cast_fp16)[name = string("op_21281_cast_fp16")]; + tensor var_21282_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2187_cast_fp16)[name = string("op_21282_cast_fp16")]; + tensor var_21283_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2189_cast_fp16)[name = string("op_21283_cast_fp16")]; + tensor var_21284_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2191_cast_fp16)[name = string("op_21284_cast_fp16")]; + tensor var_21285_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2193_cast_fp16)[name = string("op_21285_cast_fp16")]; + tensor var_21286_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2195_cast_fp16)[name = string("op_21286_cast_fp16")]; + tensor var_21287_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2197_cast_fp16)[name = string("op_21287_cast_fp16")]; + tensor var_21288_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2199_cast_fp16)[name = string("op_21288_cast_fp16")]; + tensor var_21289_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2201_cast_fp16)[name = string("op_21289_cast_fp16")]; + tensor var_21290_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2203_cast_fp16)[name = string("op_21290_cast_fp16")]; + tensor var_21291_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2205_cast_fp16)[name = string("op_21291_cast_fp16")]; + tensor var_21292_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2207_cast_fp16)[name = string("op_21292_cast_fp16")]; + tensor var_21293_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2209_cast_fp16)[name = string("op_21293_cast_fp16")]; + tensor var_21294_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2211_cast_fp16)[name = string("op_21294_cast_fp16")]; + tensor var_21295_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2213_cast_fp16)[name = string("op_21295_cast_fp16")]; + tensor var_21296_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2215_cast_fp16)[name = string("op_21296_cast_fp16")]; + tensor var_21297_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2217_cast_fp16)[name = string("op_21297_cast_fp16")]; + tensor var_21298_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2219_cast_fp16)[name = string("op_21298_cast_fp16")]; + tensor var_21299_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2221_cast_fp16)[name = string("op_21299_cast_fp16")]; + tensor var_21300_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2223_cast_fp16)[name = string("op_21300_cast_fp16")]; + tensor var_21301_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2225_cast_fp16)[name = string("op_21301_cast_fp16")]; + tensor var_21302_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2227_cast_fp16)[name = string("op_21302_cast_fp16")]; + tensor var_21303_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2229_cast_fp16)[name = string("op_21303_cast_fp16")]; + tensor var_21304_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2231_cast_fp16)[name = string("op_21304_cast_fp16")]; + tensor var_21305_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2233_cast_fp16)[name = string("op_21305_cast_fp16")]; + tensor var_21306_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2235_cast_fp16)[name = string("op_21306_cast_fp16")]; + tensor var_21307_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2237_cast_fp16)[name = string("op_21307_cast_fp16")]; + tensor var_21308_cast_fp16 = softmax(axis = var_20054, x = aw_chunk_2239_cast_fp16)[name = string("op_21308_cast_fp16")]; + string var_21310_equation_0 = const()[name = string("op_21310_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21310_cast_fp16 = einsum(equation = var_21310_equation_0, values = (var_20830_cast_fp16, var_21229_cast_fp16))[name = string("op_21310_cast_fp16")]; + string var_21312_equation_0 = const()[name = string("op_21312_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21312_cast_fp16 = einsum(equation = var_21312_equation_0, values = (var_20830_cast_fp16, var_21230_cast_fp16))[name = string("op_21312_cast_fp16")]; + string var_21314_equation_0 = const()[name = string("op_21314_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21314_cast_fp16 = einsum(equation = var_21314_equation_0, values = (var_20830_cast_fp16, var_21231_cast_fp16))[name = string("op_21314_cast_fp16")]; + string var_21316_equation_0 = const()[name = string("op_21316_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21316_cast_fp16 = einsum(equation = var_21316_equation_0, values = (var_20830_cast_fp16, var_21232_cast_fp16))[name = string("op_21316_cast_fp16")]; + string var_21318_equation_0 = const()[name = string("op_21318_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21318_cast_fp16 = einsum(equation = var_21318_equation_0, values = (var_20834_cast_fp16, var_21233_cast_fp16))[name = string("op_21318_cast_fp16")]; + string var_21320_equation_0 = const()[name = string("op_21320_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21320_cast_fp16 = einsum(equation = var_21320_equation_0, values = (var_20834_cast_fp16, var_21234_cast_fp16))[name = string("op_21320_cast_fp16")]; + string var_21322_equation_0 = const()[name = string("op_21322_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21322_cast_fp16 = einsum(equation = var_21322_equation_0, values = (var_20834_cast_fp16, var_21235_cast_fp16))[name = string("op_21322_cast_fp16")]; + string var_21324_equation_0 = const()[name = string("op_21324_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21324_cast_fp16 = einsum(equation = var_21324_equation_0, values = (var_20834_cast_fp16, var_21236_cast_fp16))[name = string("op_21324_cast_fp16")]; + string var_21326_equation_0 = const()[name = string("op_21326_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21326_cast_fp16 = einsum(equation = var_21326_equation_0, values = (var_20838_cast_fp16, var_21237_cast_fp16))[name = string("op_21326_cast_fp16")]; + string var_21328_equation_0 = const()[name = string("op_21328_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21328_cast_fp16 = einsum(equation = var_21328_equation_0, values = (var_20838_cast_fp16, var_21238_cast_fp16))[name = string("op_21328_cast_fp16")]; + string var_21330_equation_0 = const()[name = string("op_21330_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21330_cast_fp16 = einsum(equation = var_21330_equation_0, values = (var_20838_cast_fp16, var_21239_cast_fp16))[name = string("op_21330_cast_fp16")]; + string var_21332_equation_0 = const()[name = string("op_21332_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21332_cast_fp16 = einsum(equation = var_21332_equation_0, values = (var_20838_cast_fp16, var_21240_cast_fp16))[name = string("op_21332_cast_fp16")]; + string var_21334_equation_0 = const()[name = string("op_21334_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21334_cast_fp16 = einsum(equation = var_21334_equation_0, values = (var_20842_cast_fp16, var_21241_cast_fp16))[name = string("op_21334_cast_fp16")]; + string var_21336_equation_0 = const()[name = string("op_21336_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21336_cast_fp16 = einsum(equation = var_21336_equation_0, values = (var_20842_cast_fp16, var_21242_cast_fp16))[name = string("op_21336_cast_fp16")]; + string var_21338_equation_0 = const()[name = string("op_21338_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21338_cast_fp16 = einsum(equation = var_21338_equation_0, values = (var_20842_cast_fp16, var_21243_cast_fp16))[name = string("op_21338_cast_fp16")]; + string var_21340_equation_0 = const()[name = string("op_21340_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21340_cast_fp16 = einsum(equation = var_21340_equation_0, values = (var_20842_cast_fp16, var_21244_cast_fp16))[name = string("op_21340_cast_fp16")]; + string var_21342_equation_0 = const()[name = string("op_21342_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21342_cast_fp16 = einsum(equation = var_21342_equation_0, values = (var_20846_cast_fp16, var_21245_cast_fp16))[name = string("op_21342_cast_fp16")]; + string var_21344_equation_0 = const()[name = string("op_21344_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21344_cast_fp16 = einsum(equation = var_21344_equation_0, values = (var_20846_cast_fp16, var_21246_cast_fp16))[name = string("op_21344_cast_fp16")]; + string var_21346_equation_0 = const()[name = string("op_21346_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21346_cast_fp16 = einsum(equation = var_21346_equation_0, values = (var_20846_cast_fp16, var_21247_cast_fp16))[name = string("op_21346_cast_fp16")]; + string var_21348_equation_0 = const()[name = string("op_21348_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21348_cast_fp16 = einsum(equation = var_21348_equation_0, values = (var_20846_cast_fp16, var_21248_cast_fp16))[name = string("op_21348_cast_fp16")]; + string var_21350_equation_0 = const()[name = string("op_21350_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21350_cast_fp16 = einsum(equation = var_21350_equation_0, values = (var_20850_cast_fp16, var_21249_cast_fp16))[name = string("op_21350_cast_fp16")]; + string var_21352_equation_0 = const()[name = string("op_21352_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21352_cast_fp16 = einsum(equation = var_21352_equation_0, values = (var_20850_cast_fp16, var_21250_cast_fp16))[name = string("op_21352_cast_fp16")]; + string var_21354_equation_0 = const()[name = string("op_21354_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21354_cast_fp16 = einsum(equation = var_21354_equation_0, values = (var_20850_cast_fp16, var_21251_cast_fp16))[name = string("op_21354_cast_fp16")]; + string var_21356_equation_0 = const()[name = string("op_21356_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21356_cast_fp16 = einsum(equation = var_21356_equation_0, values = (var_20850_cast_fp16, var_21252_cast_fp16))[name = string("op_21356_cast_fp16")]; + string var_21358_equation_0 = const()[name = string("op_21358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21358_cast_fp16 = einsum(equation = var_21358_equation_0, values = (var_20854_cast_fp16, var_21253_cast_fp16))[name = string("op_21358_cast_fp16")]; + string var_21360_equation_0 = const()[name = string("op_21360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21360_cast_fp16 = einsum(equation = var_21360_equation_0, values = (var_20854_cast_fp16, var_21254_cast_fp16))[name = string("op_21360_cast_fp16")]; + string var_21362_equation_0 = const()[name = string("op_21362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21362_cast_fp16 = einsum(equation = var_21362_equation_0, values = (var_20854_cast_fp16, var_21255_cast_fp16))[name = string("op_21362_cast_fp16")]; + string var_21364_equation_0 = const()[name = string("op_21364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21364_cast_fp16 = einsum(equation = var_21364_equation_0, values = (var_20854_cast_fp16, var_21256_cast_fp16))[name = string("op_21364_cast_fp16")]; + string var_21366_equation_0 = const()[name = string("op_21366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21366_cast_fp16 = einsum(equation = var_21366_equation_0, values = (var_20858_cast_fp16, var_21257_cast_fp16))[name = string("op_21366_cast_fp16")]; + string var_21368_equation_0 = const()[name = string("op_21368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21368_cast_fp16 = einsum(equation = var_21368_equation_0, values = (var_20858_cast_fp16, var_21258_cast_fp16))[name = string("op_21368_cast_fp16")]; + string var_21370_equation_0 = const()[name = string("op_21370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21370_cast_fp16 = einsum(equation = var_21370_equation_0, values = (var_20858_cast_fp16, var_21259_cast_fp16))[name = string("op_21370_cast_fp16")]; + string var_21372_equation_0 = const()[name = string("op_21372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21372_cast_fp16 = einsum(equation = var_21372_equation_0, values = (var_20858_cast_fp16, var_21260_cast_fp16))[name = string("op_21372_cast_fp16")]; + string var_21374_equation_0 = const()[name = string("op_21374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21374_cast_fp16 = einsum(equation = var_21374_equation_0, values = (var_20862_cast_fp16, var_21261_cast_fp16))[name = string("op_21374_cast_fp16")]; + string var_21376_equation_0 = const()[name = string("op_21376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21376_cast_fp16 = einsum(equation = var_21376_equation_0, values = (var_20862_cast_fp16, var_21262_cast_fp16))[name = string("op_21376_cast_fp16")]; + string var_21378_equation_0 = const()[name = string("op_21378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21378_cast_fp16 = einsum(equation = var_21378_equation_0, values = (var_20862_cast_fp16, var_21263_cast_fp16))[name = string("op_21378_cast_fp16")]; + string var_21380_equation_0 = const()[name = string("op_21380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21380_cast_fp16 = einsum(equation = var_21380_equation_0, values = (var_20862_cast_fp16, var_21264_cast_fp16))[name = string("op_21380_cast_fp16")]; + string var_21382_equation_0 = const()[name = string("op_21382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21382_cast_fp16 = einsum(equation = var_21382_equation_0, values = (var_20866_cast_fp16, var_21265_cast_fp16))[name = string("op_21382_cast_fp16")]; + string var_21384_equation_0 = const()[name = string("op_21384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21384_cast_fp16 = einsum(equation = var_21384_equation_0, values = (var_20866_cast_fp16, var_21266_cast_fp16))[name = string("op_21384_cast_fp16")]; + string var_21386_equation_0 = const()[name = string("op_21386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21386_cast_fp16 = einsum(equation = var_21386_equation_0, values = (var_20866_cast_fp16, var_21267_cast_fp16))[name = string("op_21386_cast_fp16")]; + string var_21388_equation_0 = const()[name = string("op_21388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21388_cast_fp16 = einsum(equation = var_21388_equation_0, values = (var_20866_cast_fp16, var_21268_cast_fp16))[name = string("op_21388_cast_fp16")]; + string var_21390_equation_0 = const()[name = string("op_21390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21390_cast_fp16 = einsum(equation = var_21390_equation_0, values = (var_20870_cast_fp16, var_21269_cast_fp16))[name = string("op_21390_cast_fp16")]; + string var_21392_equation_0 = const()[name = string("op_21392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21392_cast_fp16 = einsum(equation = var_21392_equation_0, values = (var_20870_cast_fp16, var_21270_cast_fp16))[name = string("op_21392_cast_fp16")]; + string var_21394_equation_0 = const()[name = string("op_21394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21394_cast_fp16 = einsum(equation = var_21394_equation_0, values = (var_20870_cast_fp16, var_21271_cast_fp16))[name = string("op_21394_cast_fp16")]; + string var_21396_equation_0 = const()[name = string("op_21396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21396_cast_fp16 = einsum(equation = var_21396_equation_0, values = (var_20870_cast_fp16, var_21272_cast_fp16))[name = string("op_21396_cast_fp16")]; + string var_21398_equation_0 = const()[name = string("op_21398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21398_cast_fp16 = einsum(equation = var_21398_equation_0, values = (var_20874_cast_fp16, var_21273_cast_fp16))[name = string("op_21398_cast_fp16")]; + string var_21400_equation_0 = const()[name = string("op_21400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21400_cast_fp16 = einsum(equation = var_21400_equation_0, values = (var_20874_cast_fp16, var_21274_cast_fp16))[name = string("op_21400_cast_fp16")]; + string var_21402_equation_0 = const()[name = string("op_21402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21402_cast_fp16 = einsum(equation = var_21402_equation_0, values = (var_20874_cast_fp16, var_21275_cast_fp16))[name = string("op_21402_cast_fp16")]; + string var_21404_equation_0 = const()[name = string("op_21404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21404_cast_fp16 = einsum(equation = var_21404_equation_0, values = (var_20874_cast_fp16, var_21276_cast_fp16))[name = string("op_21404_cast_fp16")]; + string var_21406_equation_0 = const()[name = string("op_21406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21406_cast_fp16 = einsum(equation = var_21406_equation_0, values = (var_20878_cast_fp16, var_21277_cast_fp16))[name = string("op_21406_cast_fp16")]; + string var_21408_equation_0 = const()[name = string("op_21408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21408_cast_fp16 = einsum(equation = var_21408_equation_0, values = (var_20878_cast_fp16, var_21278_cast_fp16))[name = string("op_21408_cast_fp16")]; + string var_21410_equation_0 = const()[name = string("op_21410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21410_cast_fp16 = einsum(equation = var_21410_equation_0, values = (var_20878_cast_fp16, var_21279_cast_fp16))[name = string("op_21410_cast_fp16")]; + string var_21412_equation_0 = const()[name = string("op_21412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21412_cast_fp16 = einsum(equation = var_21412_equation_0, values = (var_20878_cast_fp16, var_21280_cast_fp16))[name = string("op_21412_cast_fp16")]; + string var_21414_equation_0 = const()[name = string("op_21414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21414_cast_fp16 = einsum(equation = var_21414_equation_0, values = (var_20882_cast_fp16, var_21281_cast_fp16))[name = string("op_21414_cast_fp16")]; + string var_21416_equation_0 = const()[name = string("op_21416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21416_cast_fp16 = einsum(equation = var_21416_equation_0, values = (var_20882_cast_fp16, var_21282_cast_fp16))[name = string("op_21416_cast_fp16")]; + string var_21418_equation_0 = const()[name = string("op_21418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21418_cast_fp16 = einsum(equation = var_21418_equation_0, values = (var_20882_cast_fp16, var_21283_cast_fp16))[name = string("op_21418_cast_fp16")]; + string var_21420_equation_0 = const()[name = string("op_21420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21420_cast_fp16 = einsum(equation = var_21420_equation_0, values = (var_20882_cast_fp16, var_21284_cast_fp16))[name = string("op_21420_cast_fp16")]; + string var_21422_equation_0 = const()[name = string("op_21422_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21422_cast_fp16 = einsum(equation = var_21422_equation_0, values = (var_20886_cast_fp16, var_21285_cast_fp16))[name = string("op_21422_cast_fp16")]; + string var_21424_equation_0 = const()[name = string("op_21424_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21424_cast_fp16 = einsum(equation = var_21424_equation_0, values = (var_20886_cast_fp16, var_21286_cast_fp16))[name = string("op_21424_cast_fp16")]; + string var_21426_equation_0 = const()[name = string("op_21426_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21426_cast_fp16 = einsum(equation = var_21426_equation_0, values = (var_20886_cast_fp16, var_21287_cast_fp16))[name = string("op_21426_cast_fp16")]; + string var_21428_equation_0 = const()[name = string("op_21428_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21428_cast_fp16 = einsum(equation = var_21428_equation_0, values = (var_20886_cast_fp16, var_21288_cast_fp16))[name = string("op_21428_cast_fp16")]; + string var_21430_equation_0 = const()[name = string("op_21430_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21430_cast_fp16 = einsum(equation = var_21430_equation_0, values = (var_20890_cast_fp16, var_21289_cast_fp16))[name = string("op_21430_cast_fp16")]; + string var_21432_equation_0 = const()[name = string("op_21432_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21432_cast_fp16 = einsum(equation = var_21432_equation_0, values = (var_20890_cast_fp16, var_21290_cast_fp16))[name = string("op_21432_cast_fp16")]; + string var_21434_equation_0 = const()[name = string("op_21434_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21434_cast_fp16 = einsum(equation = var_21434_equation_0, values = (var_20890_cast_fp16, var_21291_cast_fp16))[name = string("op_21434_cast_fp16")]; + string var_21436_equation_0 = const()[name = string("op_21436_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21436_cast_fp16 = einsum(equation = var_21436_equation_0, values = (var_20890_cast_fp16, var_21292_cast_fp16))[name = string("op_21436_cast_fp16")]; + string var_21438_equation_0 = const()[name = string("op_21438_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21438_cast_fp16 = einsum(equation = var_21438_equation_0, values = (var_20894_cast_fp16, var_21293_cast_fp16))[name = string("op_21438_cast_fp16")]; + string var_21440_equation_0 = const()[name = string("op_21440_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21440_cast_fp16 = einsum(equation = var_21440_equation_0, values = (var_20894_cast_fp16, var_21294_cast_fp16))[name = string("op_21440_cast_fp16")]; + string var_21442_equation_0 = const()[name = string("op_21442_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21442_cast_fp16 = einsum(equation = var_21442_equation_0, values = (var_20894_cast_fp16, var_21295_cast_fp16))[name = string("op_21442_cast_fp16")]; + string var_21444_equation_0 = const()[name = string("op_21444_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21444_cast_fp16 = einsum(equation = var_21444_equation_0, values = (var_20894_cast_fp16, var_21296_cast_fp16))[name = string("op_21444_cast_fp16")]; + string var_21446_equation_0 = const()[name = string("op_21446_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21446_cast_fp16 = einsum(equation = var_21446_equation_0, values = (var_20898_cast_fp16, var_21297_cast_fp16))[name = string("op_21446_cast_fp16")]; + string var_21448_equation_0 = const()[name = string("op_21448_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21448_cast_fp16 = einsum(equation = var_21448_equation_0, values = (var_20898_cast_fp16, var_21298_cast_fp16))[name = string("op_21448_cast_fp16")]; + string var_21450_equation_0 = const()[name = string("op_21450_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21450_cast_fp16 = einsum(equation = var_21450_equation_0, values = (var_20898_cast_fp16, var_21299_cast_fp16))[name = string("op_21450_cast_fp16")]; + string var_21452_equation_0 = const()[name = string("op_21452_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21452_cast_fp16 = einsum(equation = var_21452_equation_0, values = (var_20898_cast_fp16, var_21300_cast_fp16))[name = string("op_21452_cast_fp16")]; + string var_21454_equation_0 = const()[name = string("op_21454_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21454_cast_fp16 = einsum(equation = var_21454_equation_0, values = (var_20902_cast_fp16, var_21301_cast_fp16))[name = string("op_21454_cast_fp16")]; + string var_21456_equation_0 = const()[name = string("op_21456_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21456_cast_fp16 = einsum(equation = var_21456_equation_0, values = (var_20902_cast_fp16, var_21302_cast_fp16))[name = string("op_21456_cast_fp16")]; + string var_21458_equation_0 = const()[name = string("op_21458_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21458_cast_fp16 = einsum(equation = var_21458_equation_0, values = (var_20902_cast_fp16, var_21303_cast_fp16))[name = string("op_21458_cast_fp16")]; + string var_21460_equation_0 = const()[name = string("op_21460_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21460_cast_fp16 = einsum(equation = var_21460_equation_0, values = (var_20902_cast_fp16, var_21304_cast_fp16))[name = string("op_21460_cast_fp16")]; + string var_21462_equation_0 = const()[name = string("op_21462_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21462_cast_fp16 = einsum(equation = var_21462_equation_0, values = (var_20906_cast_fp16, var_21305_cast_fp16))[name = string("op_21462_cast_fp16")]; + string var_21464_equation_0 = const()[name = string("op_21464_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21464_cast_fp16 = einsum(equation = var_21464_equation_0, values = (var_20906_cast_fp16, var_21306_cast_fp16))[name = string("op_21464_cast_fp16")]; + string var_21466_equation_0 = const()[name = string("op_21466_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21466_cast_fp16 = einsum(equation = var_21466_equation_0, values = (var_20906_cast_fp16, var_21307_cast_fp16))[name = string("op_21466_cast_fp16")]; + string var_21468_equation_0 = const()[name = string("op_21468_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21468_cast_fp16 = einsum(equation = var_21468_equation_0, values = (var_20906_cast_fp16, var_21308_cast_fp16))[name = string("op_21468_cast_fp16")]; + bool var_21470_interleave_0 = const()[name = string("op_21470_interleave_0"), val = bool(false)]; + tensor var_21470_cast_fp16 = concat(axis = var_20029, interleave = var_21470_interleave_0, values = (var_21310_cast_fp16, var_21312_cast_fp16, var_21314_cast_fp16, var_21316_cast_fp16))[name = string("op_21470_cast_fp16")]; + bool var_21472_interleave_0 = const()[name = string("op_21472_interleave_0"), val = bool(false)]; + tensor var_21472_cast_fp16 = concat(axis = var_20029, interleave = var_21472_interleave_0, values = (var_21318_cast_fp16, var_21320_cast_fp16, var_21322_cast_fp16, var_21324_cast_fp16))[name = string("op_21472_cast_fp16")]; + bool var_21474_interleave_0 = const()[name = string("op_21474_interleave_0"), val = bool(false)]; + tensor var_21474_cast_fp16 = concat(axis = var_20029, interleave = var_21474_interleave_0, values = (var_21326_cast_fp16, var_21328_cast_fp16, var_21330_cast_fp16, var_21332_cast_fp16))[name = string("op_21474_cast_fp16")]; + bool var_21476_interleave_0 = const()[name = string("op_21476_interleave_0"), val = bool(false)]; + tensor var_21476_cast_fp16 = concat(axis = var_20029, interleave = var_21476_interleave_0, values = (var_21334_cast_fp16, var_21336_cast_fp16, var_21338_cast_fp16, var_21340_cast_fp16))[name = string("op_21476_cast_fp16")]; + bool var_21478_interleave_0 = const()[name = string("op_21478_interleave_0"), val = bool(false)]; + tensor var_21478_cast_fp16 = concat(axis = var_20029, interleave = var_21478_interleave_0, values = (var_21342_cast_fp16, var_21344_cast_fp16, var_21346_cast_fp16, var_21348_cast_fp16))[name = string("op_21478_cast_fp16")]; + bool var_21480_interleave_0 = const()[name = string("op_21480_interleave_0"), val = bool(false)]; + tensor var_21480_cast_fp16 = concat(axis = var_20029, interleave = var_21480_interleave_0, values = (var_21350_cast_fp16, var_21352_cast_fp16, var_21354_cast_fp16, var_21356_cast_fp16))[name = string("op_21480_cast_fp16")]; + bool var_21482_interleave_0 = const()[name = string("op_21482_interleave_0"), val = bool(false)]; + tensor var_21482_cast_fp16 = concat(axis = var_20029, interleave = var_21482_interleave_0, values = (var_21358_cast_fp16, var_21360_cast_fp16, var_21362_cast_fp16, var_21364_cast_fp16))[name = string("op_21482_cast_fp16")]; + bool var_21484_interleave_0 = const()[name = string("op_21484_interleave_0"), val = bool(false)]; + tensor var_21484_cast_fp16 = concat(axis = var_20029, interleave = var_21484_interleave_0, values = (var_21366_cast_fp16, var_21368_cast_fp16, var_21370_cast_fp16, var_21372_cast_fp16))[name = string("op_21484_cast_fp16")]; + bool var_21486_interleave_0 = const()[name = string("op_21486_interleave_0"), val = bool(false)]; + tensor var_21486_cast_fp16 = concat(axis = var_20029, interleave = var_21486_interleave_0, values = (var_21374_cast_fp16, var_21376_cast_fp16, var_21378_cast_fp16, var_21380_cast_fp16))[name = string("op_21486_cast_fp16")]; + bool var_21488_interleave_0 = const()[name = string("op_21488_interleave_0"), val = bool(false)]; + tensor var_21488_cast_fp16 = concat(axis = var_20029, interleave = var_21488_interleave_0, values = (var_21382_cast_fp16, var_21384_cast_fp16, var_21386_cast_fp16, var_21388_cast_fp16))[name = string("op_21488_cast_fp16")]; + bool var_21490_interleave_0 = const()[name = string("op_21490_interleave_0"), val = bool(false)]; + tensor var_21490_cast_fp16 = concat(axis = var_20029, interleave = var_21490_interleave_0, values = (var_21390_cast_fp16, var_21392_cast_fp16, var_21394_cast_fp16, var_21396_cast_fp16))[name = string("op_21490_cast_fp16")]; + bool var_21492_interleave_0 = const()[name = string("op_21492_interleave_0"), val = bool(false)]; + tensor var_21492_cast_fp16 = concat(axis = var_20029, interleave = var_21492_interleave_0, values = (var_21398_cast_fp16, var_21400_cast_fp16, var_21402_cast_fp16, var_21404_cast_fp16))[name = string("op_21492_cast_fp16")]; + bool var_21494_interleave_0 = const()[name = string("op_21494_interleave_0"), val = bool(false)]; + tensor var_21494_cast_fp16 = concat(axis = var_20029, interleave = var_21494_interleave_0, values = (var_21406_cast_fp16, var_21408_cast_fp16, var_21410_cast_fp16, var_21412_cast_fp16))[name = string("op_21494_cast_fp16")]; + bool var_21496_interleave_0 = const()[name = string("op_21496_interleave_0"), val = bool(false)]; + tensor var_21496_cast_fp16 = concat(axis = var_20029, interleave = var_21496_interleave_0, values = (var_21414_cast_fp16, var_21416_cast_fp16, var_21418_cast_fp16, var_21420_cast_fp16))[name = string("op_21496_cast_fp16")]; + bool var_21498_interleave_0 = const()[name = string("op_21498_interleave_0"), val = bool(false)]; + tensor var_21498_cast_fp16 = concat(axis = var_20029, interleave = var_21498_interleave_0, values = (var_21422_cast_fp16, var_21424_cast_fp16, var_21426_cast_fp16, var_21428_cast_fp16))[name = string("op_21498_cast_fp16")]; + bool var_21500_interleave_0 = const()[name = string("op_21500_interleave_0"), val = bool(false)]; + tensor var_21500_cast_fp16 = concat(axis = var_20029, interleave = var_21500_interleave_0, values = (var_21430_cast_fp16, var_21432_cast_fp16, var_21434_cast_fp16, var_21436_cast_fp16))[name = string("op_21500_cast_fp16")]; + bool var_21502_interleave_0 = const()[name = string("op_21502_interleave_0"), val = bool(false)]; + tensor var_21502_cast_fp16 = concat(axis = var_20029, interleave = var_21502_interleave_0, values = (var_21438_cast_fp16, var_21440_cast_fp16, var_21442_cast_fp16, var_21444_cast_fp16))[name = string("op_21502_cast_fp16")]; + bool var_21504_interleave_0 = const()[name = string("op_21504_interleave_0"), val = bool(false)]; + tensor var_21504_cast_fp16 = concat(axis = var_20029, interleave = var_21504_interleave_0, values = (var_21446_cast_fp16, var_21448_cast_fp16, var_21450_cast_fp16, var_21452_cast_fp16))[name = string("op_21504_cast_fp16")]; + bool var_21506_interleave_0 = const()[name = string("op_21506_interleave_0"), val = bool(false)]; + tensor var_21506_cast_fp16 = concat(axis = var_20029, interleave = var_21506_interleave_0, values = (var_21454_cast_fp16, var_21456_cast_fp16, var_21458_cast_fp16, var_21460_cast_fp16))[name = string("op_21506_cast_fp16")]; + bool var_21508_interleave_0 = const()[name = string("op_21508_interleave_0"), val = bool(false)]; + tensor var_21508_cast_fp16 = concat(axis = var_20029, interleave = var_21508_interleave_0, values = (var_21462_cast_fp16, var_21464_cast_fp16, var_21466_cast_fp16, var_21468_cast_fp16))[name = string("op_21508_cast_fp16")]; + bool input_105_interleave_0 = const()[name = string("input_105_interleave_0"), val = bool(false)]; + tensor input_105_cast_fp16 = concat(axis = var_20054, interleave = input_105_interleave_0, values = (var_21470_cast_fp16, var_21472_cast_fp16, var_21474_cast_fp16, var_21476_cast_fp16, var_21478_cast_fp16, var_21480_cast_fp16, var_21482_cast_fp16, var_21484_cast_fp16, var_21486_cast_fp16, var_21488_cast_fp16, var_21490_cast_fp16, var_21492_cast_fp16, var_21494_cast_fp16, var_21496_cast_fp16, var_21498_cast_fp16, var_21500_cast_fp16, var_21502_cast_fp16, var_21504_cast_fp16, var_21506_cast_fp16, var_21508_cast_fp16))[name = string("input_105_cast_fp16")]; + string obj_55_pad_type_0 = const()[name = string("obj_55_pad_type_0"), val = string("valid")]; + tensor obj_55_strides_0 = const()[name = string("obj_55_strides_0"), val = tensor([1, 1])]; + tensor obj_55_pad_0 = const()[name = string("obj_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_55_dilations_0 = const()[name = string("obj_55_dilations_0"), val = tensor([1, 1])]; + int32 obj_55_groups_0 = const()[name = string("obj_55_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536097920)))]; + tensor layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539374784)))]; + tensor obj_55_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_55_dilations_0, groups = obj_55_groups_0, pad = obj_55_pad_0, pad_type = obj_55_pad_type_0, strides = obj_55_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("obj_55_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_55_cast_fp16)[name = string("inputs_55_cast_fp16")]; + tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; + fp16 var_21527_to_fp16 = const()[name = string("op_21527_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_21527_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; + tensor input_107_gamma_0_to_fp16 = const()[name = string("input_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539377408)))]; + tensor input_107_beta_0_to_fp16 = const()[name = string("input_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539380032)))]; + fp16 input_107_epsilon_0_to_fp16 = const()[name = string("input_107_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_107_cast_fp16 = batch_norm(beta = input_107_beta_0_to_fp16, epsilon = input_107_epsilon_0_to_fp16, gamma = input_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("input_107_cast_fp16")]; + string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("valid")]; + tensor input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor([1, 1])]; + tensor input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor([1, 1])]; + int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1)]; + tensor layers_13_fc1_weight_to_fp16 = const()[name = string("layers_13_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539382656)))]; + tensor layers_13_fc1_bias_to_fp16 = const()[name = string("layers_13_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552489920)))]; + tensor input_109_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = layers_13_fc1_weight_to_fp16, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")]; + string input_111_mode_0 = const()[name = string("input_111_mode_0"), val = string("EXACT")]; + tensor input_111_cast_fp16 = gelu(mode = input_111_mode_0, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")]; + string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")]; + tensor hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)]; + tensor layers_13_fc2_weight_to_fp16 = const()[name = string("layers_13_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552500224)))]; + tensor layers_13_fc2_bias_to_fp16 = const()[name = string("layers_13_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565607488)))]; + tensor hidden_states_31_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_13_fc2_weight_to_fp16, x = input_111_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = hidden_states_31_cast_fp16)[name = string("inputs_57_cast_fp16")]; + int32 var_21556 = const()[name = string("op_21556"), val = int32(3)]; + int32 var_21581 = const()[name = string("op_21581"), val = int32(1)]; + tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; + fp16 var_21598_to_fp16 = const()[name = string("op_21598_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_21598_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; + tensor obj_57_gamma_0_to_fp16 = const()[name = string("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565610112)))]; + tensor obj_57_beta_0_to_fp16 = const()[name = string("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565612736)))]; + fp16 obj_57_epsilon_0_to_fp16 = const()[name = string("obj_57_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_57_cast_fp16")]; + string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; + tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; + tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; + int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565615360)))]; + tensor layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568892224)))]; + tensor query_29_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("query_29_cast_fp16")]; + string key_29_pad_type_0 = const()[name = string("key_29_pad_type_0"), val = string("valid")]; + tensor key_29_strides_0 = const()[name = string("key_29_strides_0"), val = tensor([1, 1])]; + tensor key_29_pad_0 = const()[name = string("key_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_29_dilations_0 = const()[name = string("key_29_dilations_0"), val = tensor([1, 1])]; + int32 key_29_groups_0 = const()[name = string("key_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568894848)))]; + tensor key_29_cast_fp16 = conv(dilations = key_29_dilations_0, groups = key_29_groups_0, pad = key_29_pad_0, pad_type = key_29_pad_type_0, strides = key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("key_29_cast_fp16")]; + string value_29_pad_type_0 = const()[name = string("value_29_pad_type_0"), val = string("valid")]; + tensor value_29_strides_0 = const()[name = string("value_29_strides_0"), val = tensor([1, 1])]; + tensor value_29_pad_0 = const()[name = string("value_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_29_dilations_0 = const()[name = string("value_29_dilations_0"), val = tensor([1, 1])]; + int32 value_29_groups_0 = const()[name = string("value_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572171712)))]; + tensor layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575448576)))]; + tensor value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = value_29_dilations_0, groups = value_29_groups_0, pad = value_29_pad_0, pad_type = value_29_pad_type_0, strides = value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("value_29_cast_fp16")]; + tensor var_21636_begin_0 = const()[name = string("op_21636_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21636_end_0 = const()[name = string("op_21636_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21636_end_mask_0 = const()[name = string("op_21636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21636_cast_fp16 = slice_by_index(begin = var_21636_begin_0, end = var_21636_end_0, end_mask = var_21636_end_mask_0, x = query_29_cast_fp16)[name = string("op_21636_cast_fp16")]; + tensor var_21640_begin_0 = const()[name = string("op_21640_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21640_end_0 = const()[name = string("op_21640_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_21640_end_mask_0 = const()[name = string("op_21640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21640_cast_fp16 = slice_by_index(begin = var_21640_begin_0, end = var_21640_end_0, end_mask = var_21640_end_mask_0, x = query_29_cast_fp16)[name = string("op_21640_cast_fp16")]; + tensor var_21644_begin_0 = const()[name = string("op_21644_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21644_end_0 = const()[name = string("op_21644_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_21644_end_mask_0 = const()[name = string("op_21644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21644_cast_fp16 = slice_by_index(begin = var_21644_begin_0, end = var_21644_end_0, end_mask = var_21644_end_mask_0, x = query_29_cast_fp16)[name = string("op_21644_cast_fp16")]; + tensor var_21648_begin_0 = const()[name = string("op_21648_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21648_end_0 = const()[name = string("op_21648_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_21648_end_mask_0 = const()[name = string("op_21648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21648_cast_fp16 = slice_by_index(begin = var_21648_begin_0, end = var_21648_end_0, end_mask = var_21648_end_mask_0, x = query_29_cast_fp16)[name = string("op_21648_cast_fp16")]; + tensor var_21652_begin_0 = const()[name = string("op_21652_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21652_end_0 = const()[name = string("op_21652_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_21652_end_mask_0 = const()[name = string("op_21652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21652_cast_fp16 = slice_by_index(begin = var_21652_begin_0, end = var_21652_end_0, end_mask = var_21652_end_mask_0, x = query_29_cast_fp16)[name = string("op_21652_cast_fp16")]; + tensor var_21656_begin_0 = const()[name = string("op_21656_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21656_end_0 = const()[name = string("op_21656_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_21656_end_mask_0 = const()[name = string("op_21656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21656_cast_fp16 = slice_by_index(begin = var_21656_begin_0, end = var_21656_end_0, end_mask = var_21656_end_mask_0, x = query_29_cast_fp16)[name = string("op_21656_cast_fp16")]; + tensor var_21660_begin_0 = const()[name = string("op_21660_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21660_end_0 = const()[name = string("op_21660_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_21660_end_mask_0 = const()[name = string("op_21660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21660_cast_fp16 = slice_by_index(begin = var_21660_begin_0, end = var_21660_end_0, end_mask = var_21660_end_mask_0, x = query_29_cast_fp16)[name = string("op_21660_cast_fp16")]; + tensor var_21664_begin_0 = const()[name = string("op_21664_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21664_end_0 = const()[name = string("op_21664_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_21664_end_mask_0 = const()[name = string("op_21664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21664_cast_fp16 = slice_by_index(begin = var_21664_begin_0, end = var_21664_end_0, end_mask = var_21664_end_mask_0, x = query_29_cast_fp16)[name = string("op_21664_cast_fp16")]; + tensor var_21668_begin_0 = const()[name = string("op_21668_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21668_end_0 = const()[name = string("op_21668_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_21668_end_mask_0 = const()[name = string("op_21668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21668_cast_fp16 = slice_by_index(begin = var_21668_begin_0, end = var_21668_end_0, end_mask = var_21668_end_mask_0, x = query_29_cast_fp16)[name = string("op_21668_cast_fp16")]; + tensor var_21672_begin_0 = const()[name = string("op_21672_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21672_end_0 = const()[name = string("op_21672_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_21672_end_mask_0 = const()[name = string("op_21672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21672_cast_fp16 = slice_by_index(begin = var_21672_begin_0, end = var_21672_end_0, end_mask = var_21672_end_mask_0, x = query_29_cast_fp16)[name = string("op_21672_cast_fp16")]; + tensor var_21676_begin_0 = const()[name = string("op_21676_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21676_end_0 = const()[name = string("op_21676_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_21676_end_mask_0 = const()[name = string("op_21676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21676_cast_fp16 = slice_by_index(begin = var_21676_begin_0, end = var_21676_end_0, end_mask = var_21676_end_mask_0, x = query_29_cast_fp16)[name = string("op_21676_cast_fp16")]; + tensor var_21680_begin_0 = const()[name = string("op_21680_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21680_end_0 = const()[name = string("op_21680_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_21680_end_mask_0 = const()[name = string("op_21680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21680_cast_fp16 = slice_by_index(begin = var_21680_begin_0, end = var_21680_end_0, end_mask = var_21680_end_mask_0, x = query_29_cast_fp16)[name = string("op_21680_cast_fp16")]; + tensor var_21684_begin_0 = const()[name = string("op_21684_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21684_end_0 = const()[name = string("op_21684_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_21684_end_mask_0 = const()[name = string("op_21684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21684_cast_fp16 = slice_by_index(begin = var_21684_begin_0, end = var_21684_end_0, end_mask = var_21684_end_mask_0, x = query_29_cast_fp16)[name = string("op_21684_cast_fp16")]; + tensor var_21688_begin_0 = const()[name = string("op_21688_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21688_end_0 = const()[name = string("op_21688_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_21688_end_mask_0 = const()[name = string("op_21688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21688_cast_fp16 = slice_by_index(begin = var_21688_begin_0, end = var_21688_end_0, end_mask = var_21688_end_mask_0, x = query_29_cast_fp16)[name = string("op_21688_cast_fp16")]; + tensor var_21692_begin_0 = const()[name = string("op_21692_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21692_end_0 = const()[name = string("op_21692_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_21692_end_mask_0 = const()[name = string("op_21692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21692_cast_fp16 = slice_by_index(begin = var_21692_begin_0, end = var_21692_end_0, end_mask = var_21692_end_mask_0, x = query_29_cast_fp16)[name = string("op_21692_cast_fp16")]; + tensor var_21696_begin_0 = const()[name = string("op_21696_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21696_end_0 = const()[name = string("op_21696_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_21696_end_mask_0 = const()[name = string("op_21696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21696_cast_fp16 = slice_by_index(begin = var_21696_begin_0, end = var_21696_end_0, end_mask = var_21696_end_mask_0, x = query_29_cast_fp16)[name = string("op_21696_cast_fp16")]; + tensor var_21700_begin_0 = const()[name = string("op_21700_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21700_end_0 = const()[name = string("op_21700_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_21700_end_mask_0 = const()[name = string("op_21700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21700_cast_fp16 = slice_by_index(begin = var_21700_begin_0, end = var_21700_end_0, end_mask = var_21700_end_mask_0, x = query_29_cast_fp16)[name = string("op_21700_cast_fp16")]; + tensor var_21704_begin_0 = const()[name = string("op_21704_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21704_end_0 = const()[name = string("op_21704_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_21704_end_mask_0 = const()[name = string("op_21704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21704_cast_fp16 = slice_by_index(begin = var_21704_begin_0, end = var_21704_end_0, end_mask = var_21704_end_mask_0, x = query_29_cast_fp16)[name = string("op_21704_cast_fp16")]; + tensor var_21708_begin_0 = const()[name = string("op_21708_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21708_end_0 = const()[name = string("op_21708_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_21708_end_mask_0 = const()[name = string("op_21708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21708_cast_fp16 = slice_by_index(begin = var_21708_begin_0, end = var_21708_end_0, end_mask = var_21708_end_mask_0, x = query_29_cast_fp16)[name = string("op_21708_cast_fp16")]; + tensor var_21712_begin_0 = const()[name = string("op_21712_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21712_end_0 = const()[name = string("op_21712_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_21712_end_mask_0 = const()[name = string("op_21712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21712_cast_fp16 = slice_by_index(begin = var_21712_begin_0, end = var_21712_end_0, end_mask = var_21712_end_mask_0, x = query_29_cast_fp16)[name = string("op_21712_cast_fp16")]; + tensor var_21721_begin_0 = const()[name = string("op_21721_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21721_end_0 = const()[name = string("op_21721_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21721_end_mask_0 = const()[name = string("op_21721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21721_cast_fp16 = slice_by_index(begin = var_21721_begin_0, end = var_21721_end_0, end_mask = var_21721_end_mask_0, x = var_21636_cast_fp16)[name = string("op_21721_cast_fp16")]; + tensor var_21728_begin_0 = const()[name = string("op_21728_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21728_end_0 = const()[name = string("op_21728_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21728_end_mask_0 = const()[name = string("op_21728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21728_cast_fp16 = slice_by_index(begin = var_21728_begin_0, end = var_21728_end_0, end_mask = var_21728_end_mask_0, x = var_21636_cast_fp16)[name = string("op_21728_cast_fp16")]; + tensor var_21735_begin_0 = const()[name = string("op_21735_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21735_end_0 = const()[name = string("op_21735_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21735_end_mask_0 = const()[name = string("op_21735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21735_cast_fp16 = slice_by_index(begin = var_21735_begin_0, end = var_21735_end_0, end_mask = var_21735_end_mask_0, x = var_21636_cast_fp16)[name = string("op_21735_cast_fp16")]; + tensor var_21742_begin_0 = const()[name = string("op_21742_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21742_end_0 = const()[name = string("op_21742_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21742_end_mask_0 = const()[name = string("op_21742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21742_cast_fp16 = slice_by_index(begin = var_21742_begin_0, end = var_21742_end_0, end_mask = var_21742_end_mask_0, x = var_21636_cast_fp16)[name = string("op_21742_cast_fp16")]; + tensor var_21749_begin_0 = const()[name = string("op_21749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21749_end_0 = const()[name = string("op_21749_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21749_end_mask_0 = const()[name = string("op_21749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21749_cast_fp16 = slice_by_index(begin = var_21749_begin_0, end = var_21749_end_0, end_mask = var_21749_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21749_cast_fp16")]; + tensor var_21756_begin_0 = const()[name = string("op_21756_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21756_end_0 = const()[name = string("op_21756_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21756_end_mask_0 = const()[name = string("op_21756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21756_cast_fp16 = slice_by_index(begin = var_21756_begin_0, end = var_21756_end_0, end_mask = var_21756_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21756_cast_fp16")]; + tensor var_21763_begin_0 = const()[name = string("op_21763_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21763_end_0 = const()[name = string("op_21763_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21763_end_mask_0 = const()[name = string("op_21763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21763_cast_fp16 = slice_by_index(begin = var_21763_begin_0, end = var_21763_end_0, end_mask = var_21763_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21763_cast_fp16")]; + tensor var_21770_begin_0 = const()[name = string("op_21770_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21770_end_0 = const()[name = string("op_21770_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21770_end_mask_0 = const()[name = string("op_21770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21770_cast_fp16 = slice_by_index(begin = var_21770_begin_0, end = var_21770_end_0, end_mask = var_21770_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21770_cast_fp16")]; + tensor var_21777_begin_0 = const()[name = string("op_21777_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21777_end_0 = const()[name = string("op_21777_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21777_end_mask_0 = const()[name = string("op_21777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21777_cast_fp16 = slice_by_index(begin = var_21777_begin_0, end = var_21777_end_0, end_mask = var_21777_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21777_cast_fp16")]; + tensor var_21784_begin_0 = const()[name = string("op_21784_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21784_end_0 = const()[name = string("op_21784_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21784_end_mask_0 = const()[name = string("op_21784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21784_cast_fp16 = slice_by_index(begin = var_21784_begin_0, end = var_21784_end_0, end_mask = var_21784_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21784_cast_fp16")]; + tensor var_21791_begin_0 = const()[name = string("op_21791_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21791_end_0 = const()[name = string("op_21791_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21791_end_mask_0 = const()[name = string("op_21791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21791_cast_fp16 = slice_by_index(begin = var_21791_begin_0, end = var_21791_end_0, end_mask = var_21791_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21791_cast_fp16")]; + tensor var_21798_begin_0 = const()[name = string("op_21798_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21798_end_0 = const()[name = string("op_21798_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21798_end_mask_0 = const()[name = string("op_21798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21798_cast_fp16 = slice_by_index(begin = var_21798_begin_0, end = var_21798_end_0, end_mask = var_21798_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21798_cast_fp16")]; + tensor var_21805_begin_0 = const()[name = string("op_21805_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21805_end_0 = const()[name = string("op_21805_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21805_end_mask_0 = const()[name = string("op_21805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21805_cast_fp16 = slice_by_index(begin = var_21805_begin_0, end = var_21805_end_0, end_mask = var_21805_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21805_cast_fp16")]; + tensor var_21812_begin_0 = const()[name = string("op_21812_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21812_end_0 = const()[name = string("op_21812_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21812_end_mask_0 = const()[name = string("op_21812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21812_cast_fp16 = slice_by_index(begin = var_21812_begin_0, end = var_21812_end_0, end_mask = var_21812_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21812_cast_fp16")]; + tensor var_21819_begin_0 = const()[name = string("op_21819_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21819_end_0 = const()[name = string("op_21819_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21819_end_mask_0 = const()[name = string("op_21819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21819_cast_fp16 = slice_by_index(begin = var_21819_begin_0, end = var_21819_end_0, end_mask = var_21819_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21819_cast_fp16")]; + tensor var_21826_begin_0 = const()[name = string("op_21826_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21826_end_0 = const()[name = string("op_21826_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21826_end_mask_0 = const()[name = string("op_21826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21826_cast_fp16 = slice_by_index(begin = var_21826_begin_0, end = var_21826_end_0, end_mask = var_21826_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21826_cast_fp16")]; + tensor var_21833_begin_0 = const()[name = string("op_21833_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21833_end_0 = const()[name = string("op_21833_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21833_end_mask_0 = const()[name = string("op_21833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21833_cast_fp16 = slice_by_index(begin = var_21833_begin_0, end = var_21833_end_0, end_mask = var_21833_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21833_cast_fp16")]; + tensor var_21840_begin_0 = const()[name = string("op_21840_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21840_end_0 = const()[name = string("op_21840_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21840_end_mask_0 = const()[name = string("op_21840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21840_cast_fp16 = slice_by_index(begin = var_21840_begin_0, end = var_21840_end_0, end_mask = var_21840_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21840_cast_fp16")]; + tensor var_21847_begin_0 = const()[name = string("op_21847_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21847_end_0 = const()[name = string("op_21847_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21847_end_mask_0 = const()[name = string("op_21847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21847_cast_fp16 = slice_by_index(begin = var_21847_begin_0, end = var_21847_end_0, end_mask = var_21847_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21847_cast_fp16")]; + tensor var_21854_begin_0 = const()[name = string("op_21854_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21854_end_0 = const()[name = string("op_21854_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21854_end_mask_0 = const()[name = string("op_21854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21854_cast_fp16 = slice_by_index(begin = var_21854_begin_0, end = var_21854_end_0, end_mask = var_21854_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21854_cast_fp16")]; + tensor var_21861_begin_0 = const()[name = string("op_21861_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21861_end_0 = const()[name = string("op_21861_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21861_end_mask_0 = const()[name = string("op_21861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21861_cast_fp16 = slice_by_index(begin = var_21861_begin_0, end = var_21861_end_0, end_mask = var_21861_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21861_cast_fp16")]; + tensor var_21868_begin_0 = const()[name = string("op_21868_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21868_end_0 = const()[name = string("op_21868_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21868_end_mask_0 = const()[name = string("op_21868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21868_cast_fp16 = slice_by_index(begin = var_21868_begin_0, end = var_21868_end_0, end_mask = var_21868_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21868_cast_fp16")]; + tensor var_21875_begin_0 = const()[name = string("op_21875_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21875_end_0 = const()[name = string("op_21875_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21875_end_mask_0 = const()[name = string("op_21875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21875_cast_fp16 = slice_by_index(begin = var_21875_begin_0, end = var_21875_end_0, end_mask = var_21875_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21875_cast_fp16")]; + tensor var_21882_begin_0 = const()[name = string("op_21882_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21882_end_0 = const()[name = string("op_21882_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21882_end_mask_0 = const()[name = string("op_21882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21882_cast_fp16 = slice_by_index(begin = var_21882_begin_0, end = var_21882_end_0, end_mask = var_21882_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21882_cast_fp16")]; + tensor var_21889_begin_0 = const()[name = string("op_21889_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21889_end_0 = const()[name = string("op_21889_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21889_end_mask_0 = const()[name = string("op_21889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21889_cast_fp16 = slice_by_index(begin = var_21889_begin_0, end = var_21889_end_0, end_mask = var_21889_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21889_cast_fp16")]; + tensor var_21896_begin_0 = const()[name = string("op_21896_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21896_end_0 = const()[name = string("op_21896_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21896_end_mask_0 = const()[name = string("op_21896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21896_cast_fp16 = slice_by_index(begin = var_21896_begin_0, end = var_21896_end_0, end_mask = var_21896_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21896_cast_fp16")]; + tensor var_21903_begin_0 = const()[name = string("op_21903_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21903_end_0 = const()[name = string("op_21903_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21903_end_mask_0 = const()[name = string("op_21903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21903_cast_fp16 = slice_by_index(begin = var_21903_begin_0, end = var_21903_end_0, end_mask = var_21903_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21903_cast_fp16")]; + tensor var_21910_begin_0 = const()[name = string("op_21910_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21910_end_0 = const()[name = string("op_21910_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21910_end_mask_0 = const()[name = string("op_21910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21910_cast_fp16 = slice_by_index(begin = var_21910_begin_0, end = var_21910_end_0, end_mask = var_21910_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21910_cast_fp16")]; + tensor var_21917_begin_0 = const()[name = string("op_21917_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21917_end_0 = const()[name = string("op_21917_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21917_end_mask_0 = const()[name = string("op_21917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21917_cast_fp16 = slice_by_index(begin = var_21917_begin_0, end = var_21917_end_0, end_mask = var_21917_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21917_cast_fp16")]; + tensor var_21924_begin_0 = const()[name = string("op_21924_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21924_end_0 = const()[name = string("op_21924_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21924_end_mask_0 = const()[name = string("op_21924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21924_cast_fp16 = slice_by_index(begin = var_21924_begin_0, end = var_21924_end_0, end_mask = var_21924_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21924_cast_fp16")]; + tensor var_21931_begin_0 = const()[name = string("op_21931_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21931_end_0 = const()[name = string("op_21931_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21931_end_mask_0 = const()[name = string("op_21931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21931_cast_fp16 = slice_by_index(begin = var_21931_begin_0, end = var_21931_end_0, end_mask = var_21931_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21931_cast_fp16")]; + tensor var_21938_begin_0 = const()[name = string("op_21938_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21938_end_0 = const()[name = string("op_21938_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21938_end_mask_0 = const()[name = string("op_21938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21938_cast_fp16 = slice_by_index(begin = var_21938_begin_0, end = var_21938_end_0, end_mask = var_21938_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21938_cast_fp16")]; + tensor var_21945_begin_0 = const()[name = string("op_21945_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21945_end_0 = const()[name = string("op_21945_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21945_end_mask_0 = const()[name = string("op_21945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21945_cast_fp16 = slice_by_index(begin = var_21945_begin_0, end = var_21945_end_0, end_mask = var_21945_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21945_cast_fp16")]; + tensor var_21952_begin_0 = const()[name = string("op_21952_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21952_end_0 = const()[name = string("op_21952_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21952_end_mask_0 = const()[name = string("op_21952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21952_cast_fp16 = slice_by_index(begin = var_21952_begin_0, end = var_21952_end_0, end_mask = var_21952_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21952_cast_fp16")]; + tensor var_21959_begin_0 = const()[name = string("op_21959_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21959_end_0 = const()[name = string("op_21959_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21959_end_mask_0 = const()[name = string("op_21959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21959_cast_fp16 = slice_by_index(begin = var_21959_begin_0, end = var_21959_end_0, end_mask = var_21959_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21959_cast_fp16")]; + tensor var_21966_begin_0 = const()[name = string("op_21966_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21966_end_0 = const()[name = string("op_21966_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21966_end_mask_0 = const()[name = string("op_21966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21966_cast_fp16 = slice_by_index(begin = var_21966_begin_0, end = var_21966_end_0, end_mask = var_21966_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21966_cast_fp16")]; + tensor var_21973_begin_0 = const()[name = string("op_21973_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21973_end_0 = const()[name = string("op_21973_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21973_end_mask_0 = const()[name = string("op_21973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21973_cast_fp16 = slice_by_index(begin = var_21973_begin_0, end = var_21973_end_0, end_mask = var_21973_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21973_cast_fp16")]; + tensor var_21980_begin_0 = const()[name = string("op_21980_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21980_end_0 = const()[name = string("op_21980_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21980_end_mask_0 = const()[name = string("op_21980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21980_cast_fp16 = slice_by_index(begin = var_21980_begin_0, end = var_21980_end_0, end_mask = var_21980_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21980_cast_fp16")]; + tensor var_21987_begin_0 = const()[name = string("op_21987_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21987_end_0 = const()[name = string("op_21987_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21987_end_mask_0 = const()[name = string("op_21987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21987_cast_fp16 = slice_by_index(begin = var_21987_begin_0, end = var_21987_end_0, end_mask = var_21987_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21987_cast_fp16")]; + tensor var_21994_begin_0 = const()[name = string("op_21994_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21994_end_0 = const()[name = string("op_21994_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21994_end_mask_0 = const()[name = string("op_21994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21994_cast_fp16 = slice_by_index(begin = var_21994_begin_0, end = var_21994_end_0, end_mask = var_21994_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21994_cast_fp16")]; + tensor var_22001_begin_0 = const()[name = string("op_22001_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22001_end_0 = const()[name = string("op_22001_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22001_end_mask_0 = const()[name = string("op_22001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22001_cast_fp16 = slice_by_index(begin = var_22001_begin_0, end = var_22001_end_0, end_mask = var_22001_end_mask_0, x = var_21676_cast_fp16)[name = string("op_22001_cast_fp16")]; + tensor var_22008_begin_0 = const()[name = string("op_22008_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22008_end_0 = const()[name = string("op_22008_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22008_end_mask_0 = const()[name = string("op_22008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22008_cast_fp16 = slice_by_index(begin = var_22008_begin_0, end = var_22008_end_0, end_mask = var_22008_end_mask_0, x = var_21676_cast_fp16)[name = string("op_22008_cast_fp16")]; + tensor var_22015_begin_0 = const()[name = string("op_22015_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22015_end_0 = const()[name = string("op_22015_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22015_end_mask_0 = const()[name = string("op_22015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22015_cast_fp16 = slice_by_index(begin = var_22015_begin_0, end = var_22015_end_0, end_mask = var_22015_end_mask_0, x = var_21676_cast_fp16)[name = string("op_22015_cast_fp16")]; + tensor var_22022_begin_0 = const()[name = string("op_22022_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22022_end_0 = const()[name = string("op_22022_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22022_end_mask_0 = const()[name = string("op_22022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22022_cast_fp16 = slice_by_index(begin = var_22022_begin_0, end = var_22022_end_0, end_mask = var_22022_end_mask_0, x = var_21676_cast_fp16)[name = string("op_22022_cast_fp16")]; + tensor var_22029_begin_0 = const()[name = string("op_22029_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22029_end_0 = const()[name = string("op_22029_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22029_end_mask_0 = const()[name = string("op_22029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22029_cast_fp16 = slice_by_index(begin = var_22029_begin_0, end = var_22029_end_0, end_mask = var_22029_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22029_cast_fp16")]; + tensor var_22036_begin_0 = const()[name = string("op_22036_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22036_end_0 = const()[name = string("op_22036_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22036_end_mask_0 = const()[name = string("op_22036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22036_cast_fp16 = slice_by_index(begin = var_22036_begin_0, end = var_22036_end_0, end_mask = var_22036_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22036_cast_fp16")]; + tensor var_22043_begin_0 = const()[name = string("op_22043_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22043_end_0 = const()[name = string("op_22043_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22043_end_mask_0 = const()[name = string("op_22043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22043_cast_fp16 = slice_by_index(begin = var_22043_begin_0, end = var_22043_end_0, end_mask = var_22043_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22043_cast_fp16")]; + tensor var_22050_begin_0 = const()[name = string("op_22050_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22050_end_0 = const()[name = string("op_22050_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22050_end_mask_0 = const()[name = string("op_22050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22050_cast_fp16 = slice_by_index(begin = var_22050_begin_0, end = var_22050_end_0, end_mask = var_22050_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22050_cast_fp16")]; + tensor var_22057_begin_0 = const()[name = string("op_22057_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22057_end_0 = const()[name = string("op_22057_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22057_end_mask_0 = const()[name = string("op_22057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22057_cast_fp16 = slice_by_index(begin = var_22057_begin_0, end = var_22057_end_0, end_mask = var_22057_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22057_cast_fp16")]; + tensor var_22064_begin_0 = const()[name = string("op_22064_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22064_end_0 = const()[name = string("op_22064_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22064_end_mask_0 = const()[name = string("op_22064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22064_cast_fp16 = slice_by_index(begin = var_22064_begin_0, end = var_22064_end_0, end_mask = var_22064_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22064_cast_fp16")]; + tensor var_22071_begin_0 = const()[name = string("op_22071_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22071_end_0 = const()[name = string("op_22071_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22071_end_mask_0 = const()[name = string("op_22071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22071_cast_fp16 = slice_by_index(begin = var_22071_begin_0, end = var_22071_end_0, end_mask = var_22071_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22071_cast_fp16")]; + tensor var_22078_begin_0 = const()[name = string("op_22078_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22078_end_0 = const()[name = string("op_22078_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22078_end_mask_0 = const()[name = string("op_22078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22078_cast_fp16 = slice_by_index(begin = var_22078_begin_0, end = var_22078_end_0, end_mask = var_22078_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22078_cast_fp16")]; + tensor var_22085_begin_0 = const()[name = string("op_22085_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22085_end_0 = const()[name = string("op_22085_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22085_end_mask_0 = const()[name = string("op_22085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22085_cast_fp16 = slice_by_index(begin = var_22085_begin_0, end = var_22085_end_0, end_mask = var_22085_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22085_cast_fp16")]; + tensor var_22092_begin_0 = const()[name = string("op_22092_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22092_end_0 = const()[name = string("op_22092_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22092_end_mask_0 = const()[name = string("op_22092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22092_cast_fp16 = slice_by_index(begin = var_22092_begin_0, end = var_22092_end_0, end_mask = var_22092_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22092_cast_fp16")]; + tensor var_22099_begin_0 = const()[name = string("op_22099_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22099_end_0 = const()[name = string("op_22099_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22099_end_mask_0 = const()[name = string("op_22099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22099_cast_fp16 = slice_by_index(begin = var_22099_begin_0, end = var_22099_end_0, end_mask = var_22099_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22099_cast_fp16")]; + tensor var_22106_begin_0 = const()[name = string("op_22106_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22106_end_0 = const()[name = string("op_22106_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22106_end_mask_0 = const()[name = string("op_22106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22106_cast_fp16 = slice_by_index(begin = var_22106_begin_0, end = var_22106_end_0, end_mask = var_22106_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22106_cast_fp16")]; + tensor var_22113_begin_0 = const()[name = string("op_22113_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22113_end_0 = const()[name = string("op_22113_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22113_end_mask_0 = const()[name = string("op_22113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22113_cast_fp16 = slice_by_index(begin = var_22113_begin_0, end = var_22113_end_0, end_mask = var_22113_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22113_cast_fp16")]; + tensor var_22120_begin_0 = const()[name = string("op_22120_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22120_end_0 = const()[name = string("op_22120_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22120_end_mask_0 = const()[name = string("op_22120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22120_cast_fp16 = slice_by_index(begin = var_22120_begin_0, end = var_22120_end_0, end_mask = var_22120_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22120_cast_fp16")]; + tensor var_22127_begin_0 = const()[name = string("op_22127_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22127_end_0 = const()[name = string("op_22127_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22127_end_mask_0 = const()[name = string("op_22127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22127_cast_fp16 = slice_by_index(begin = var_22127_begin_0, end = var_22127_end_0, end_mask = var_22127_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22127_cast_fp16")]; + tensor var_22134_begin_0 = const()[name = string("op_22134_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22134_end_0 = const()[name = string("op_22134_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22134_end_mask_0 = const()[name = string("op_22134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22134_cast_fp16 = slice_by_index(begin = var_22134_begin_0, end = var_22134_end_0, end_mask = var_22134_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22134_cast_fp16")]; + tensor var_22141_begin_0 = const()[name = string("op_22141_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22141_end_0 = const()[name = string("op_22141_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22141_end_mask_0 = const()[name = string("op_22141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22141_cast_fp16 = slice_by_index(begin = var_22141_begin_0, end = var_22141_end_0, end_mask = var_22141_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22141_cast_fp16")]; + tensor var_22148_begin_0 = const()[name = string("op_22148_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22148_end_0 = const()[name = string("op_22148_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22148_end_mask_0 = const()[name = string("op_22148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22148_cast_fp16 = slice_by_index(begin = var_22148_begin_0, end = var_22148_end_0, end_mask = var_22148_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22148_cast_fp16")]; + tensor var_22155_begin_0 = const()[name = string("op_22155_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22155_end_0 = const()[name = string("op_22155_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22155_end_mask_0 = const()[name = string("op_22155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22155_cast_fp16 = slice_by_index(begin = var_22155_begin_0, end = var_22155_end_0, end_mask = var_22155_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22155_cast_fp16")]; + tensor var_22162_begin_0 = const()[name = string("op_22162_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22162_end_0 = const()[name = string("op_22162_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22162_end_mask_0 = const()[name = string("op_22162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22162_cast_fp16 = slice_by_index(begin = var_22162_begin_0, end = var_22162_end_0, end_mask = var_22162_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22162_cast_fp16")]; + tensor var_22169_begin_0 = const()[name = string("op_22169_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22169_end_0 = const()[name = string("op_22169_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22169_end_mask_0 = const()[name = string("op_22169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22169_cast_fp16 = slice_by_index(begin = var_22169_begin_0, end = var_22169_end_0, end_mask = var_22169_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22169_cast_fp16")]; + tensor var_22176_begin_0 = const()[name = string("op_22176_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22176_end_0 = const()[name = string("op_22176_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22176_end_mask_0 = const()[name = string("op_22176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22176_cast_fp16 = slice_by_index(begin = var_22176_begin_0, end = var_22176_end_0, end_mask = var_22176_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22176_cast_fp16")]; + tensor var_22183_begin_0 = const()[name = string("op_22183_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22183_end_0 = const()[name = string("op_22183_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22183_end_mask_0 = const()[name = string("op_22183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22183_cast_fp16 = slice_by_index(begin = var_22183_begin_0, end = var_22183_end_0, end_mask = var_22183_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22183_cast_fp16")]; + tensor var_22190_begin_0 = const()[name = string("op_22190_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22190_end_0 = const()[name = string("op_22190_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22190_end_mask_0 = const()[name = string("op_22190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22190_cast_fp16 = slice_by_index(begin = var_22190_begin_0, end = var_22190_end_0, end_mask = var_22190_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22190_cast_fp16")]; + tensor var_22197_begin_0 = const()[name = string("op_22197_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22197_end_0 = const()[name = string("op_22197_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22197_end_mask_0 = const()[name = string("op_22197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22197_cast_fp16 = slice_by_index(begin = var_22197_begin_0, end = var_22197_end_0, end_mask = var_22197_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22197_cast_fp16")]; + tensor var_22204_begin_0 = const()[name = string("op_22204_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22204_end_0 = const()[name = string("op_22204_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22204_end_mask_0 = const()[name = string("op_22204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22204_cast_fp16 = slice_by_index(begin = var_22204_begin_0, end = var_22204_end_0, end_mask = var_22204_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22204_cast_fp16")]; + tensor var_22211_begin_0 = const()[name = string("op_22211_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22211_end_0 = const()[name = string("op_22211_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22211_end_mask_0 = const()[name = string("op_22211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22211_cast_fp16 = slice_by_index(begin = var_22211_begin_0, end = var_22211_end_0, end_mask = var_22211_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22211_cast_fp16")]; + tensor var_22218_begin_0 = const()[name = string("op_22218_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22218_end_0 = const()[name = string("op_22218_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22218_end_mask_0 = const()[name = string("op_22218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22218_cast_fp16 = slice_by_index(begin = var_22218_begin_0, end = var_22218_end_0, end_mask = var_22218_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22218_cast_fp16")]; + tensor var_22225_begin_0 = const()[name = string("op_22225_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22225_end_0 = const()[name = string("op_22225_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22225_end_mask_0 = const()[name = string("op_22225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22225_cast_fp16 = slice_by_index(begin = var_22225_begin_0, end = var_22225_end_0, end_mask = var_22225_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22225_cast_fp16")]; + tensor var_22232_begin_0 = const()[name = string("op_22232_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22232_end_0 = const()[name = string("op_22232_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22232_end_mask_0 = const()[name = string("op_22232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22232_cast_fp16 = slice_by_index(begin = var_22232_begin_0, end = var_22232_end_0, end_mask = var_22232_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22232_cast_fp16")]; + tensor var_22239_begin_0 = const()[name = string("op_22239_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22239_end_0 = const()[name = string("op_22239_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22239_end_mask_0 = const()[name = string("op_22239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22239_cast_fp16 = slice_by_index(begin = var_22239_begin_0, end = var_22239_end_0, end_mask = var_22239_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22239_cast_fp16")]; + tensor var_22246_begin_0 = const()[name = string("op_22246_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22246_end_0 = const()[name = string("op_22246_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22246_end_mask_0 = const()[name = string("op_22246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22246_cast_fp16 = slice_by_index(begin = var_22246_begin_0, end = var_22246_end_0, end_mask = var_22246_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22246_cast_fp16")]; + tensor var_22253_begin_0 = const()[name = string("op_22253_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22253_end_0 = const()[name = string("op_22253_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22253_end_mask_0 = const()[name = string("op_22253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22253_cast_fp16 = slice_by_index(begin = var_22253_begin_0, end = var_22253_end_0, end_mask = var_22253_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22253_cast_fp16")]; + tensor var_22260_begin_0 = const()[name = string("op_22260_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22260_end_0 = const()[name = string("op_22260_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22260_end_mask_0 = const()[name = string("op_22260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22260_cast_fp16 = slice_by_index(begin = var_22260_begin_0, end = var_22260_end_0, end_mask = var_22260_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22260_cast_fp16")]; + tensor var_22267_begin_0 = const()[name = string("op_22267_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22267_end_0 = const()[name = string("op_22267_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22267_end_mask_0 = const()[name = string("op_22267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22267_cast_fp16 = slice_by_index(begin = var_22267_begin_0, end = var_22267_end_0, end_mask = var_22267_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22267_cast_fp16")]; + tensor var_22274_begin_0 = const()[name = string("op_22274_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22274_end_0 = const()[name = string("op_22274_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22274_end_mask_0 = const()[name = string("op_22274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22274_cast_fp16 = slice_by_index(begin = var_22274_begin_0, end = var_22274_end_0, end_mask = var_22274_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22274_cast_fp16")]; + tensor k_29_perm_0 = const()[name = string("k_29_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22279_begin_0 = const()[name = string("op_22279_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22279_end_0 = const()[name = string("op_22279_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_22279_end_mask_0 = const()[name = string("op_22279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_29_cast_fp16 = transpose(perm = k_29_perm_0, x = key_29_cast_fp16)[name = string("transpose_17")]; + tensor var_22279_cast_fp16 = slice_by_index(begin = var_22279_begin_0, end = var_22279_end_0, end_mask = var_22279_end_mask_0, x = k_29_cast_fp16)[name = string("op_22279_cast_fp16")]; + tensor var_22283_begin_0 = const()[name = string("op_22283_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22283_end_0 = const()[name = string("op_22283_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_22283_end_mask_0 = const()[name = string("op_22283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22283_cast_fp16 = slice_by_index(begin = var_22283_begin_0, end = var_22283_end_0, end_mask = var_22283_end_mask_0, x = k_29_cast_fp16)[name = string("op_22283_cast_fp16")]; + tensor var_22287_begin_0 = const()[name = string("op_22287_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22287_end_0 = const()[name = string("op_22287_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_22287_end_mask_0 = const()[name = string("op_22287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22287_cast_fp16 = slice_by_index(begin = var_22287_begin_0, end = var_22287_end_0, end_mask = var_22287_end_mask_0, x = k_29_cast_fp16)[name = string("op_22287_cast_fp16")]; + tensor var_22291_begin_0 = const()[name = string("op_22291_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22291_end_0 = const()[name = string("op_22291_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_22291_end_mask_0 = const()[name = string("op_22291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22291_cast_fp16 = slice_by_index(begin = var_22291_begin_0, end = var_22291_end_0, end_mask = var_22291_end_mask_0, x = k_29_cast_fp16)[name = string("op_22291_cast_fp16")]; + tensor var_22295_begin_0 = const()[name = string("op_22295_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22295_end_0 = const()[name = string("op_22295_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_22295_end_mask_0 = const()[name = string("op_22295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22295_cast_fp16 = slice_by_index(begin = var_22295_begin_0, end = var_22295_end_0, end_mask = var_22295_end_mask_0, x = k_29_cast_fp16)[name = string("op_22295_cast_fp16")]; + tensor var_22299_begin_0 = const()[name = string("op_22299_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22299_end_0 = const()[name = string("op_22299_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_22299_end_mask_0 = const()[name = string("op_22299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22299_cast_fp16 = slice_by_index(begin = var_22299_begin_0, end = var_22299_end_0, end_mask = var_22299_end_mask_0, x = k_29_cast_fp16)[name = string("op_22299_cast_fp16")]; + tensor var_22303_begin_0 = const()[name = string("op_22303_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22303_end_0 = const()[name = string("op_22303_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_22303_end_mask_0 = const()[name = string("op_22303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22303_cast_fp16 = slice_by_index(begin = var_22303_begin_0, end = var_22303_end_0, end_mask = var_22303_end_mask_0, x = k_29_cast_fp16)[name = string("op_22303_cast_fp16")]; + tensor var_22307_begin_0 = const()[name = string("op_22307_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22307_end_0 = const()[name = string("op_22307_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_22307_end_mask_0 = const()[name = string("op_22307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22307_cast_fp16 = slice_by_index(begin = var_22307_begin_0, end = var_22307_end_0, end_mask = var_22307_end_mask_0, x = k_29_cast_fp16)[name = string("op_22307_cast_fp16")]; + tensor var_22311_begin_0 = const()[name = string("op_22311_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22311_end_0 = const()[name = string("op_22311_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_22311_end_mask_0 = const()[name = string("op_22311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22311_cast_fp16 = slice_by_index(begin = var_22311_begin_0, end = var_22311_end_0, end_mask = var_22311_end_mask_0, x = k_29_cast_fp16)[name = string("op_22311_cast_fp16")]; + tensor var_22315_begin_0 = const()[name = string("op_22315_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22315_end_0 = const()[name = string("op_22315_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_22315_end_mask_0 = const()[name = string("op_22315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22315_cast_fp16 = slice_by_index(begin = var_22315_begin_0, end = var_22315_end_0, end_mask = var_22315_end_mask_0, x = k_29_cast_fp16)[name = string("op_22315_cast_fp16")]; + tensor var_22319_begin_0 = const()[name = string("op_22319_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22319_end_0 = const()[name = string("op_22319_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_22319_end_mask_0 = const()[name = string("op_22319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22319_cast_fp16 = slice_by_index(begin = var_22319_begin_0, end = var_22319_end_0, end_mask = var_22319_end_mask_0, x = k_29_cast_fp16)[name = string("op_22319_cast_fp16")]; + tensor var_22323_begin_0 = const()[name = string("op_22323_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22323_end_0 = const()[name = string("op_22323_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_22323_end_mask_0 = const()[name = string("op_22323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22323_cast_fp16 = slice_by_index(begin = var_22323_begin_0, end = var_22323_end_0, end_mask = var_22323_end_mask_0, x = k_29_cast_fp16)[name = string("op_22323_cast_fp16")]; + tensor var_22327_begin_0 = const()[name = string("op_22327_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22327_end_0 = const()[name = string("op_22327_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_22327_end_mask_0 = const()[name = string("op_22327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22327_cast_fp16 = slice_by_index(begin = var_22327_begin_0, end = var_22327_end_0, end_mask = var_22327_end_mask_0, x = k_29_cast_fp16)[name = string("op_22327_cast_fp16")]; + tensor var_22331_begin_0 = const()[name = string("op_22331_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22331_end_0 = const()[name = string("op_22331_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_22331_end_mask_0 = const()[name = string("op_22331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22331_cast_fp16 = slice_by_index(begin = var_22331_begin_0, end = var_22331_end_0, end_mask = var_22331_end_mask_0, x = k_29_cast_fp16)[name = string("op_22331_cast_fp16")]; + tensor var_22335_begin_0 = const()[name = string("op_22335_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22335_end_0 = const()[name = string("op_22335_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_22335_end_mask_0 = const()[name = string("op_22335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22335_cast_fp16 = slice_by_index(begin = var_22335_begin_0, end = var_22335_end_0, end_mask = var_22335_end_mask_0, x = k_29_cast_fp16)[name = string("op_22335_cast_fp16")]; + tensor var_22339_begin_0 = const()[name = string("op_22339_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22339_end_0 = const()[name = string("op_22339_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_22339_end_mask_0 = const()[name = string("op_22339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22339_cast_fp16 = slice_by_index(begin = var_22339_begin_0, end = var_22339_end_0, end_mask = var_22339_end_mask_0, x = k_29_cast_fp16)[name = string("op_22339_cast_fp16")]; + tensor var_22343_begin_0 = const()[name = string("op_22343_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22343_end_0 = const()[name = string("op_22343_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_22343_end_mask_0 = const()[name = string("op_22343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22343_cast_fp16 = slice_by_index(begin = var_22343_begin_0, end = var_22343_end_0, end_mask = var_22343_end_mask_0, x = k_29_cast_fp16)[name = string("op_22343_cast_fp16")]; + tensor var_22347_begin_0 = const()[name = string("op_22347_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22347_end_0 = const()[name = string("op_22347_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_22347_end_mask_0 = const()[name = string("op_22347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22347_cast_fp16 = slice_by_index(begin = var_22347_begin_0, end = var_22347_end_0, end_mask = var_22347_end_mask_0, x = k_29_cast_fp16)[name = string("op_22347_cast_fp16")]; + tensor var_22351_begin_0 = const()[name = string("op_22351_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22351_end_0 = const()[name = string("op_22351_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_22351_end_mask_0 = const()[name = string("op_22351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22351_cast_fp16 = slice_by_index(begin = var_22351_begin_0, end = var_22351_end_0, end_mask = var_22351_end_mask_0, x = k_29_cast_fp16)[name = string("op_22351_cast_fp16")]; + tensor var_22355_begin_0 = const()[name = string("op_22355_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22355_end_0 = const()[name = string("op_22355_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_22355_end_mask_0 = const()[name = string("op_22355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22355_cast_fp16 = slice_by_index(begin = var_22355_begin_0, end = var_22355_end_0, end_mask = var_22355_end_mask_0, x = k_29_cast_fp16)[name = string("op_22355_cast_fp16")]; + tensor var_22357_begin_0 = const()[name = string("op_22357_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22357_end_0 = const()[name = string("op_22357_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22357_end_mask_0 = const()[name = string("op_22357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22357_cast_fp16 = slice_by_index(begin = var_22357_begin_0, end = var_22357_end_0, end_mask = var_22357_end_mask_0, x = value_29_cast_fp16)[name = string("op_22357_cast_fp16")]; + tensor var_22361_begin_0 = const()[name = string("op_22361_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22361_end_0 = const()[name = string("op_22361_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_22361_end_mask_0 = const()[name = string("op_22361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22361_cast_fp16 = slice_by_index(begin = var_22361_begin_0, end = var_22361_end_0, end_mask = var_22361_end_mask_0, x = value_29_cast_fp16)[name = string("op_22361_cast_fp16")]; + tensor var_22365_begin_0 = const()[name = string("op_22365_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22365_end_0 = const()[name = string("op_22365_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_22365_end_mask_0 = const()[name = string("op_22365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22365_cast_fp16 = slice_by_index(begin = var_22365_begin_0, end = var_22365_end_0, end_mask = var_22365_end_mask_0, x = value_29_cast_fp16)[name = string("op_22365_cast_fp16")]; + tensor var_22369_begin_0 = const()[name = string("op_22369_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22369_end_0 = const()[name = string("op_22369_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_22369_end_mask_0 = const()[name = string("op_22369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22369_cast_fp16 = slice_by_index(begin = var_22369_begin_0, end = var_22369_end_0, end_mask = var_22369_end_mask_0, x = value_29_cast_fp16)[name = string("op_22369_cast_fp16")]; + tensor var_22373_begin_0 = const()[name = string("op_22373_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22373_end_0 = const()[name = string("op_22373_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_22373_end_mask_0 = const()[name = string("op_22373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22373_cast_fp16 = slice_by_index(begin = var_22373_begin_0, end = var_22373_end_0, end_mask = var_22373_end_mask_0, x = value_29_cast_fp16)[name = string("op_22373_cast_fp16")]; + tensor var_22377_begin_0 = const()[name = string("op_22377_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22377_end_0 = const()[name = string("op_22377_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_22377_end_mask_0 = const()[name = string("op_22377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22377_cast_fp16 = slice_by_index(begin = var_22377_begin_0, end = var_22377_end_0, end_mask = var_22377_end_mask_0, x = value_29_cast_fp16)[name = string("op_22377_cast_fp16")]; + tensor var_22381_begin_0 = const()[name = string("op_22381_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22381_end_0 = const()[name = string("op_22381_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_22381_end_mask_0 = const()[name = string("op_22381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22381_cast_fp16 = slice_by_index(begin = var_22381_begin_0, end = var_22381_end_0, end_mask = var_22381_end_mask_0, x = value_29_cast_fp16)[name = string("op_22381_cast_fp16")]; + tensor var_22385_begin_0 = const()[name = string("op_22385_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22385_end_0 = const()[name = string("op_22385_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_22385_end_mask_0 = const()[name = string("op_22385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22385_cast_fp16 = slice_by_index(begin = var_22385_begin_0, end = var_22385_end_0, end_mask = var_22385_end_mask_0, x = value_29_cast_fp16)[name = string("op_22385_cast_fp16")]; + tensor var_22389_begin_0 = const()[name = string("op_22389_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22389_end_0 = const()[name = string("op_22389_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_22389_end_mask_0 = const()[name = string("op_22389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22389_cast_fp16 = slice_by_index(begin = var_22389_begin_0, end = var_22389_end_0, end_mask = var_22389_end_mask_0, x = value_29_cast_fp16)[name = string("op_22389_cast_fp16")]; + tensor var_22393_begin_0 = const()[name = string("op_22393_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22393_end_0 = const()[name = string("op_22393_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_22393_end_mask_0 = const()[name = string("op_22393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22393_cast_fp16 = slice_by_index(begin = var_22393_begin_0, end = var_22393_end_0, end_mask = var_22393_end_mask_0, x = value_29_cast_fp16)[name = string("op_22393_cast_fp16")]; + tensor var_22397_begin_0 = const()[name = string("op_22397_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22397_end_0 = const()[name = string("op_22397_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_22397_end_mask_0 = const()[name = string("op_22397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22397_cast_fp16 = slice_by_index(begin = var_22397_begin_0, end = var_22397_end_0, end_mask = var_22397_end_mask_0, x = value_29_cast_fp16)[name = string("op_22397_cast_fp16")]; + tensor var_22401_begin_0 = const()[name = string("op_22401_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22401_end_0 = const()[name = string("op_22401_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_22401_end_mask_0 = const()[name = string("op_22401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22401_cast_fp16 = slice_by_index(begin = var_22401_begin_0, end = var_22401_end_0, end_mask = var_22401_end_mask_0, x = value_29_cast_fp16)[name = string("op_22401_cast_fp16")]; + tensor var_22405_begin_0 = const()[name = string("op_22405_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22405_end_0 = const()[name = string("op_22405_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_22405_end_mask_0 = const()[name = string("op_22405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22405_cast_fp16 = slice_by_index(begin = var_22405_begin_0, end = var_22405_end_0, end_mask = var_22405_end_mask_0, x = value_29_cast_fp16)[name = string("op_22405_cast_fp16")]; + tensor var_22409_begin_0 = const()[name = string("op_22409_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22409_end_0 = const()[name = string("op_22409_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_22409_end_mask_0 = const()[name = string("op_22409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22409_cast_fp16 = slice_by_index(begin = var_22409_begin_0, end = var_22409_end_0, end_mask = var_22409_end_mask_0, x = value_29_cast_fp16)[name = string("op_22409_cast_fp16")]; + tensor var_22413_begin_0 = const()[name = string("op_22413_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22413_end_0 = const()[name = string("op_22413_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_22413_end_mask_0 = const()[name = string("op_22413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22413_cast_fp16 = slice_by_index(begin = var_22413_begin_0, end = var_22413_end_0, end_mask = var_22413_end_mask_0, x = value_29_cast_fp16)[name = string("op_22413_cast_fp16")]; + tensor var_22417_begin_0 = const()[name = string("op_22417_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22417_end_0 = const()[name = string("op_22417_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_22417_end_mask_0 = const()[name = string("op_22417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22417_cast_fp16 = slice_by_index(begin = var_22417_begin_0, end = var_22417_end_0, end_mask = var_22417_end_mask_0, x = value_29_cast_fp16)[name = string("op_22417_cast_fp16")]; + tensor var_22421_begin_0 = const()[name = string("op_22421_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22421_end_0 = const()[name = string("op_22421_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_22421_end_mask_0 = const()[name = string("op_22421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22421_cast_fp16 = slice_by_index(begin = var_22421_begin_0, end = var_22421_end_0, end_mask = var_22421_end_mask_0, x = value_29_cast_fp16)[name = string("op_22421_cast_fp16")]; + tensor var_22425_begin_0 = const()[name = string("op_22425_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22425_end_0 = const()[name = string("op_22425_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_22425_end_mask_0 = const()[name = string("op_22425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22425_cast_fp16 = slice_by_index(begin = var_22425_begin_0, end = var_22425_end_0, end_mask = var_22425_end_mask_0, x = value_29_cast_fp16)[name = string("op_22425_cast_fp16")]; + tensor var_22429_begin_0 = const()[name = string("op_22429_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22429_end_0 = const()[name = string("op_22429_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_22429_end_mask_0 = const()[name = string("op_22429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22429_cast_fp16 = slice_by_index(begin = var_22429_begin_0, end = var_22429_end_0, end_mask = var_22429_end_mask_0, x = value_29_cast_fp16)[name = string("op_22429_cast_fp16")]; + tensor var_22433_begin_0 = const()[name = string("op_22433_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22433_end_0 = const()[name = string("op_22433_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_22433_end_mask_0 = const()[name = string("op_22433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22433_cast_fp16 = slice_by_index(begin = var_22433_begin_0, end = var_22433_end_0, end_mask = var_22433_end_mask_0, x = value_29_cast_fp16)[name = string("op_22433_cast_fp16")]; + string _SplitHeadsQ__mh_w_2241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2241_equation_0, values = (var_22279_cast_fp16, var_21721_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2241_cast_fp16")]; + string _SplitHeadsQ__mh_w_2243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2243_equation_0, values = (var_22279_cast_fp16, var_21728_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2243_cast_fp16")]; + string _SplitHeadsQ__mh_w_2245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2245_equation_0, values = (var_22279_cast_fp16, var_21735_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2245_cast_fp16")]; + string _SplitHeadsQ__mh_w_2247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2247_equation_0, values = (var_22279_cast_fp16, var_21742_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2247_cast_fp16")]; + string _SplitHeadsQ__mh_w_2249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2249_equation_0, values = (var_22283_cast_fp16, var_21749_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2249_cast_fp16")]; + string _SplitHeadsQ__mh_w_2251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2251_equation_0, values = (var_22283_cast_fp16, var_21756_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2251_cast_fp16")]; + string _SplitHeadsQ__mh_w_2253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2253_equation_0, values = (var_22283_cast_fp16, var_21763_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2253_cast_fp16")]; + string _SplitHeadsQ__mh_w_2255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2255_equation_0, values = (var_22283_cast_fp16, var_21770_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2255_cast_fp16")]; + string _SplitHeadsQ__mh_w_2257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2257_equation_0, values = (var_22287_cast_fp16, var_21777_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2257_cast_fp16")]; + string _SplitHeadsQ__mh_w_2259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2259_equation_0, values = (var_22287_cast_fp16, var_21784_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2259_cast_fp16")]; + string _SplitHeadsQ__mh_w_2261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2261_equation_0, values = (var_22287_cast_fp16, var_21791_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2261_cast_fp16")]; + string _SplitHeadsQ__mh_w_2263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2263_equation_0, values = (var_22287_cast_fp16, var_21798_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2263_cast_fp16")]; + string _SplitHeadsQ__mh_w_2265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2265_equation_0, values = (var_22291_cast_fp16, var_21805_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2265_cast_fp16")]; + string _SplitHeadsQ__mh_w_2267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2267_equation_0, values = (var_22291_cast_fp16, var_21812_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2267_cast_fp16")]; + string _SplitHeadsQ__mh_w_2269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2269_equation_0, values = (var_22291_cast_fp16, var_21819_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2269_cast_fp16")]; + string _SplitHeadsQ__mh_w_2271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2271_equation_0, values = (var_22291_cast_fp16, var_21826_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2271_cast_fp16")]; + string _SplitHeadsQ__mh_w_2273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2273_equation_0, values = (var_22295_cast_fp16, var_21833_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2273_cast_fp16")]; + string _SplitHeadsQ__mh_w_2275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2275_equation_0, values = (var_22295_cast_fp16, var_21840_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2275_cast_fp16")]; + string _SplitHeadsQ__mh_w_2277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2277_equation_0, values = (var_22295_cast_fp16, var_21847_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2277_cast_fp16")]; + string _SplitHeadsQ__mh_w_2279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2279_equation_0, values = (var_22295_cast_fp16, var_21854_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2279_cast_fp16")]; + string _SplitHeadsQ__mh_w_2281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2281_equation_0, values = (var_22299_cast_fp16, var_21861_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2281_cast_fp16")]; + string _SplitHeadsQ__mh_w_2283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2283_equation_0, values = (var_22299_cast_fp16, var_21868_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2283_cast_fp16")]; + string _SplitHeadsQ__mh_w_2285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2285_equation_0, values = (var_22299_cast_fp16, var_21875_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2285_cast_fp16")]; + string _SplitHeadsQ__mh_w_2287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2287_equation_0, values = (var_22299_cast_fp16, var_21882_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2287_cast_fp16")]; + string _SplitHeadsQ__mh_w_2289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2289_equation_0, values = (var_22303_cast_fp16, var_21889_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2289_cast_fp16")]; + string _SplitHeadsQ__mh_w_2291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2291_equation_0, values = (var_22303_cast_fp16, var_21896_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2291_cast_fp16")]; + string _SplitHeadsQ__mh_w_2293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2293_equation_0, values = (var_22303_cast_fp16, var_21903_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2293_cast_fp16")]; + string _SplitHeadsQ__mh_w_2295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2295_equation_0, values = (var_22303_cast_fp16, var_21910_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2295_cast_fp16")]; + string _SplitHeadsQ__mh_w_2297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2297_equation_0, values = (var_22307_cast_fp16, var_21917_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2297_cast_fp16")]; + string _SplitHeadsQ__mh_w_2299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2299_equation_0, values = (var_22307_cast_fp16, var_21924_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2299_cast_fp16")]; + string _SplitHeadsQ__mh_w_2301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2301_equation_0, values = (var_22307_cast_fp16, var_21931_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2301_cast_fp16")]; + string _SplitHeadsQ__mh_w_2303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2303_equation_0, values = (var_22307_cast_fp16, var_21938_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2303_cast_fp16")]; + string _SplitHeadsQ__mh_w_2305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2305_equation_0, values = (var_22311_cast_fp16, var_21945_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2305_cast_fp16")]; + string _SplitHeadsQ__mh_w_2307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2307_equation_0, values = (var_22311_cast_fp16, var_21952_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2307_cast_fp16")]; + string _SplitHeadsQ__mh_w_2309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2309_equation_0, values = (var_22311_cast_fp16, var_21959_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2309_cast_fp16")]; + string _SplitHeadsQ__mh_w_2311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2311_equation_0, values = (var_22311_cast_fp16, var_21966_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2311_cast_fp16")]; + string _SplitHeadsQ__mh_w_2313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2313_equation_0, values = (var_22315_cast_fp16, var_21973_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2313_cast_fp16")]; + string _SplitHeadsQ__mh_w_2315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2315_equation_0, values = (var_22315_cast_fp16, var_21980_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2315_cast_fp16")]; + string _SplitHeadsQ__mh_w_2317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2317_equation_0, values = (var_22315_cast_fp16, var_21987_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2317_cast_fp16")]; + string _SplitHeadsQ__mh_w_2319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2319_equation_0, values = (var_22315_cast_fp16, var_21994_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2319_cast_fp16")]; + string _SplitHeadsQ__mh_w_2321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2321_equation_0, values = (var_22319_cast_fp16, var_22001_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2321_cast_fp16")]; + string _SplitHeadsQ__mh_w_2323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2323_equation_0, values = (var_22319_cast_fp16, var_22008_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2323_cast_fp16")]; + string _SplitHeadsQ__mh_w_2325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2325_equation_0, values = (var_22319_cast_fp16, var_22015_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2325_cast_fp16")]; + string _SplitHeadsQ__mh_w_2327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2327_equation_0, values = (var_22319_cast_fp16, var_22022_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2327_cast_fp16")]; + string _SplitHeadsQ__mh_w_2329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2329_equation_0, values = (var_22323_cast_fp16, var_22029_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2329_cast_fp16")]; + string _SplitHeadsQ__mh_w_2331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2331_equation_0, values = (var_22323_cast_fp16, var_22036_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2331_cast_fp16")]; + string _SplitHeadsQ__mh_w_2333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2333_equation_0, values = (var_22323_cast_fp16, var_22043_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2333_cast_fp16")]; + string _SplitHeadsQ__mh_w_2335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2335_equation_0, values = (var_22323_cast_fp16, var_22050_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2335_cast_fp16")]; + string _SplitHeadsQ__mh_w_2337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2337_equation_0, values = (var_22327_cast_fp16, var_22057_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2337_cast_fp16")]; + string _SplitHeadsQ__mh_w_2339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2339_equation_0, values = (var_22327_cast_fp16, var_22064_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2339_cast_fp16")]; + string _SplitHeadsQ__mh_w_2341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2341_equation_0, values = (var_22327_cast_fp16, var_22071_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2341_cast_fp16")]; + string _SplitHeadsQ__mh_w_2343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2343_equation_0, values = (var_22327_cast_fp16, var_22078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2343_cast_fp16")]; + string _SplitHeadsQ__mh_w_2345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2345_equation_0, values = (var_22331_cast_fp16, var_22085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2345_cast_fp16")]; + string _SplitHeadsQ__mh_w_2347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2347_equation_0, values = (var_22331_cast_fp16, var_22092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2347_cast_fp16")]; + string _SplitHeadsQ__mh_w_2349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2349_equation_0, values = (var_22331_cast_fp16, var_22099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2349_cast_fp16")]; + string _SplitHeadsQ__mh_w_2351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2351_equation_0, values = (var_22331_cast_fp16, var_22106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2351_cast_fp16")]; + string _SplitHeadsQ__mh_w_2353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2353_equation_0, values = (var_22335_cast_fp16, var_22113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2353_cast_fp16")]; + string _SplitHeadsQ__mh_w_2355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2355_equation_0, values = (var_22335_cast_fp16, var_22120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2355_cast_fp16")]; + string _SplitHeadsQ__mh_w_2357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2357_equation_0, values = (var_22335_cast_fp16, var_22127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2357_cast_fp16")]; + string _SplitHeadsQ__mh_w_2359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2359_equation_0, values = (var_22335_cast_fp16, var_22134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2359_cast_fp16")]; + string _SplitHeadsQ__mh_w_2361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2361_equation_0, values = (var_22339_cast_fp16, var_22141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2361_cast_fp16")]; + string _SplitHeadsQ__mh_w_2363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2363_equation_0, values = (var_22339_cast_fp16, var_22148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2363_cast_fp16")]; + string _SplitHeadsQ__mh_w_2365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2365_equation_0, values = (var_22339_cast_fp16, var_22155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2365_cast_fp16")]; + string _SplitHeadsQ__mh_w_2367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2367_equation_0, values = (var_22339_cast_fp16, var_22162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2367_cast_fp16")]; + string _SplitHeadsQ__mh_w_2369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2369_equation_0, values = (var_22343_cast_fp16, var_22169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2369_cast_fp16")]; + string _SplitHeadsQ__mh_w_2371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2371_equation_0, values = (var_22343_cast_fp16, var_22176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2371_cast_fp16")]; + string _SplitHeadsQ__mh_w_2373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2373_equation_0, values = (var_22343_cast_fp16, var_22183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2373_cast_fp16")]; + string _SplitHeadsQ__mh_w_2375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2375_equation_0, values = (var_22343_cast_fp16, var_22190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2375_cast_fp16")]; + string _SplitHeadsQ__mh_w_2377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2377_equation_0, values = (var_22347_cast_fp16, var_22197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2377_cast_fp16")]; + string _SplitHeadsQ__mh_w_2379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2379_equation_0, values = (var_22347_cast_fp16, var_22204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2379_cast_fp16")]; + string _SplitHeadsQ__mh_w_2381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2381_equation_0, values = (var_22347_cast_fp16, var_22211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2381_cast_fp16")]; + string _SplitHeadsQ__mh_w_2383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2383_equation_0, values = (var_22347_cast_fp16, var_22218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2383_cast_fp16")]; + string _SplitHeadsQ__mh_w_2385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2385_equation_0, values = (var_22351_cast_fp16, var_22225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2385_cast_fp16")]; + string _SplitHeadsQ__mh_w_2387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2387_equation_0, values = (var_22351_cast_fp16, var_22232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2387_cast_fp16")]; + string _SplitHeadsQ__mh_w_2389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2389_equation_0, values = (var_22351_cast_fp16, var_22239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2389_cast_fp16")]; + string _SplitHeadsQ__mh_w_2391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2391_equation_0, values = (var_22351_cast_fp16, var_22246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2391_cast_fp16")]; + string _SplitHeadsQ__mh_w_2393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2393_equation_0, values = (var_22355_cast_fp16, var_22253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2393_cast_fp16")]; + string _SplitHeadsQ__mh_w_2395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2395_equation_0, values = (var_22355_cast_fp16, var_22260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2395_cast_fp16")]; + string _SplitHeadsQ__mh_w_2397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2397_equation_0, values = (var_22355_cast_fp16, var_22267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2397_cast_fp16")]; + string _SplitHeadsQ__mh_w_2399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2399_equation_0, values = (var_22355_cast_fp16, var_22274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2399_cast_fp16")]; + fp16 var_22596_to_fp16 = const()[name = string("op_22596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2241_cast_fp16, y = var_22596_to_fp16)[name = string("aw_chunk_2241_cast_fp16")]; + fp16 var_22598_to_fp16 = const()[name = string("op_22598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2243_cast_fp16, y = var_22598_to_fp16)[name = string("aw_chunk_2243_cast_fp16")]; + fp16 var_22600_to_fp16 = const()[name = string("op_22600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2245_cast_fp16, y = var_22600_to_fp16)[name = string("aw_chunk_2245_cast_fp16")]; + fp16 var_22602_to_fp16 = const()[name = string("op_22602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2247_cast_fp16, y = var_22602_to_fp16)[name = string("aw_chunk_2247_cast_fp16")]; + fp16 var_22604_to_fp16 = const()[name = string("op_22604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2249_cast_fp16, y = var_22604_to_fp16)[name = string("aw_chunk_2249_cast_fp16")]; + fp16 var_22606_to_fp16 = const()[name = string("op_22606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2251_cast_fp16, y = var_22606_to_fp16)[name = string("aw_chunk_2251_cast_fp16")]; + fp16 var_22608_to_fp16 = const()[name = string("op_22608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2253_cast_fp16, y = var_22608_to_fp16)[name = string("aw_chunk_2253_cast_fp16")]; + fp16 var_22610_to_fp16 = const()[name = string("op_22610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2255_cast_fp16, y = var_22610_to_fp16)[name = string("aw_chunk_2255_cast_fp16")]; + fp16 var_22612_to_fp16 = const()[name = string("op_22612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2257_cast_fp16, y = var_22612_to_fp16)[name = string("aw_chunk_2257_cast_fp16")]; + fp16 var_22614_to_fp16 = const()[name = string("op_22614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2259_cast_fp16, y = var_22614_to_fp16)[name = string("aw_chunk_2259_cast_fp16")]; + fp16 var_22616_to_fp16 = const()[name = string("op_22616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2261_cast_fp16, y = var_22616_to_fp16)[name = string("aw_chunk_2261_cast_fp16")]; + fp16 var_22618_to_fp16 = const()[name = string("op_22618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2263_cast_fp16, y = var_22618_to_fp16)[name = string("aw_chunk_2263_cast_fp16")]; + fp16 var_22620_to_fp16 = const()[name = string("op_22620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2265_cast_fp16, y = var_22620_to_fp16)[name = string("aw_chunk_2265_cast_fp16")]; + fp16 var_22622_to_fp16 = const()[name = string("op_22622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2267_cast_fp16, y = var_22622_to_fp16)[name = string("aw_chunk_2267_cast_fp16")]; + fp16 var_22624_to_fp16 = const()[name = string("op_22624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2269_cast_fp16, y = var_22624_to_fp16)[name = string("aw_chunk_2269_cast_fp16")]; + fp16 var_22626_to_fp16 = const()[name = string("op_22626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2271_cast_fp16, y = var_22626_to_fp16)[name = string("aw_chunk_2271_cast_fp16")]; + fp16 var_22628_to_fp16 = const()[name = string("op_22628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2273_cast_fp16, y = var_22628_to_fp16)[name = string("aw_chunk_2273_cast_fp16")]; + fp16 var_22630_to_fp16 = const()[name = string("op_22630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2275_cast_fp16, y = var_22630_to_fp16)[name = string("aw_chunk_2275_cast_fp16")]; + fp16 var_22632_to_fp16 = const()[name = string("op_22632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2277_cast_fp16, y = var_22632_to_fp16)[name = string("aw_chunk_2277_cast_fp16")]; + fp16 var_22634_to_fp16 = const()[name = string("op_22634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2279_cast_fp16, y = var_22634_to_fp16)[name = string("aw_chunk_2279_cast_fp16")]; + fp16 var_22636_to_fp16 = const()[name = string("op_22636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2281_cast_fp16, y = var_22636_to_fp16)[name = string("aw_chunk_2281_cast_fp16")]; + fp16 var_22638_to_fp16 = const()[name = string("op_22638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2283_cast_fp16, y = var_22638_to_fp16)[name = string("aw_chunk_2283_cast_fp16")]; + fp16 var_22640_to_fp16 = const()[name = string("op_22640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2285_cast_fp16, y = var_22640_to_fp16)[name = string("aw_chunk_2285_cast_fp16")]; + fp16 var_22642_to_fp16 = const()[name = string("op_22642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2287_cast_fp16, y = var_22642_to_fp16)[name = string("aw_chunk_2287_cast_fp16")]; + fp16 var_22644_to_fp16 = const()[name = string("op_22644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2289_cast_fp16, y = var_22644_to_fp16)[name = string("aw_chunk_2289_cast_fp16")]; + fp16 var_22646_to_fp16 = const()[name = string("op_22646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2291_cast_fp16, y = var_22646_to_fp16)[name = string("aw_chunk_2291_cast_fp16")]; + fp16 var_22648_to_fp16 = const()[name = string("op_22648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2293_cast_fp16, y = var_22648_to_fp16)[name = string("aw_chunk_2293_cast_fp16")]; + fp16 var_22650_to_fp16 = const()[name = string("op_22650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2295_cast_fp16, y = var_22650_to_fp16)[name = string("aw_chunk_2295_cast_fp16")]; + fp16 var_22652_to_fp16 = const()[name = string("op_22652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2297_cast_fp16, y = var_22652_to_fp16)[name = string("aw_chunk_2297_cast_fp16")]; + fp16 var_22654_to_fp16 = const()[name = string("op_22654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2299_cast_fp16, y = var_22654_to_fp16)[name = string("aw_chunk_2299_cast_fp16")]; + fp16 var_22656_to_fp16 = const()[name = string("op_22656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2301_cast_fp16, y = var_22656_to_fp16)[name = string("aw_chunk_2301_cast_fp16")]; + fp16 var_22658_to_fp16 = const()[name = string("op_22658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2303_cast_fp16, y = var_22658_to_fp16)[name = string("aw_chunk_2303_cast_fp16")]; + fp16 var_22660_to_fp16 = const()[name = string("op_22660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2305_cast_fp16, y = var_22660_to_fp16)[name = string("aw_chunk_2305_cast_fp16")]; + fp16 var_22662_to_fp16 = const()[name = string("op_22662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2307_cast_fp16, y = var_22662_to_fp16)[name = string("aw_chunk_2307_cast_fp16")]; + fp16 var_22664_to_fp16 = const()[name = string("op_22664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2309_cast_fp16, y = var_22664_to_fp16)[name = string("aw_chunk_2309_cast_fp16")]; + fp16 var_22666_to_fp16 = const()[name = string("op_22666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2311_cast_fp16, y = var_22666_to_fp16)[name = string("aw_chunk_2311_cast_fp16")]; + fp16 var_22668_to_fp16 = const()[name = string("op_22668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2313_cast_fp16, y = var_22668_to_fp16)[name = string("aw_chunk_2313_cast_fp16")]; + fp16 var_22670_to_fp16 = const()[name = string("op_22670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2315_cast_fp16, y = var_22670_to_fp16)[name = string("aw_chunk_2315_cast_fp16")]; + fp16 var_22672_to_fp16 = const()[name = string("op_22672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2317_cast_fp16, y = var_22672_to_fp16)[name = string("aw_chunk_2317_cast_fp16")]; + fp16 var_22674_to_fp16 = const()[name = string("op_22674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2319_cast_fp16, y = var_22674_to_fp16)[name = string("aw_chunk_2319_cast_fp16")]; + fp16 var_22676_to_fp16 = const()[name = string("op_22676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2321_cast_fp16, y = var_22676_to_fp16)[name = string("aw_chunk_2321_cast_fp16")]; + fp16 var_22678_to_fp16 = const()[name = string("op_22678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2323_cast_fp16, y = var_22678_to_fp16)[name = string("aw_chunk_2323_cast_fp16")]; + fp16 var_22680_to_fp16 = const()[name = string("op_22680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2325_cast_fp16, y = var_22680_to_fp16)[name = string("aw_chunk_2325_cast_fp16")]; + fp16 var_22682_to_fp16 = const()[name = string("op_22682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2327_cast_fp16, y = var_22682_to_fp16)[name = string("aw_chunk_2327_cast_fp16")]; + fp16 var_22684_to_fp16 = const()[name = string("op_22684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2329_cast_fp16, y = var_22684_to_fp16)[name = string("aw_chunk_2329_cast_fp16")]; + fp16 var_22686_to_fp16 = const()[name = string("op_22686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2331_cast_fp16, y = var_22686_to_fp16)[name = string("aw_chunk_2331_cast_fp16")]; + fp16 var_22688_to_fp16 = const()[name = string("op_22688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2333_cast_fp16, y = var_22688_to_fp16)[name = string("aw_chunk_2333_cast_fp16")]; + fp16 var_22690_to_fp16 = const()[name = string("op_22690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2335_cast_fp16, y = var_22690_to_fp16)[name = string("aw_chunk_2335_cast_fp16")]; + fp16 var_22692_to_fp16 = const()[name = string("op_22692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2337_cast_fp16, y = var_22692_to_fp16)[name = string("aw_chunk_2337_cast_fp16")]; + fp16 var_22694_to_fp16 = const()[name = string("op_22694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2339_cast_fp16, y = var_22694_to_fp16)[name = string("aw_chunk_2339_cast_fp16")]; + fp16 var_22696_to_fp16 = const()[name = string("op_22696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2341_cast_fp16, y = var_22696_to_fp16)[name = string("aw_chunk_2341_cast_fp16")]; + fp16 var_22698_to_fp16 = const()[name = string("op_22698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2343_cast_fp16, y = var_22698_to_fp16)[name = string("aw_chunk_2343_cast_fp16")]; + fp16 var_22700_to_fp16 = const()[name = string("op_22700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2345_cast_fp16, y = var_22700_to_fp16)[name = string("aw_chunk_2345_cast_fp16")]; + fp16 var_22702_to_fp16 = const()[name = string("op_22702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2347_cast_fp16, y = var_22702_to_fp16)[name = string("aw_chunk_2347_cast_fp16")]; + fp16 var_22704_to_fp16 = const()[name = string("op_22704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2349_cast_fp16, y = var_22704_to_fp16)[name = string("aw_chunk_2349_cast_fp16")]; + fp16 var_22706_to_fp16 = const()[name = string("op_22706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2351_cast_fp16, y = var_22706_to_fp16)[name = string("aw_chunk_2351_cast_fp16")]; + fp16 var_22708_to_fp16 = const()[name = string("op_22708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2353_cast_fp16, y = var_22708_to_fp16)[name = string("aw_chunk_2353_cast_fp16")]; + fp16 var_22710_to_fp16 = const()[name = string("op_22710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2355_cast_fp16, y = var_22710_to_fp16)[name = string("aw_chunk_2355_cast_fp16")]; + fp16 var_22712_to_fp16 = const()[name = string("op_22712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2357_cast_fp16, y = var_22712_to_fp16)[name = string("aw_chunk_2357_cast_fp16")]; + fp16 var_22714_to_fp16 = const()[name = string("op_22714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2359_cast_fp16, y = var_22714_to_fp16)[name = string("aw_chunk_2359_cast_fp16")]; + fp16 var_22716_to_fp16 = const()[name = string("op_22716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2361_cast_fp16, y = var_22716_to_fp16)[name = string("aw_chunk_2361_cast_fp16")]; + fp16 var_22718_to_fp16 = const()[name = string("op_22718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2363_cast_fp16, y = var_22718_to_fp16)[name = string("aw_chunk_2363_cast_fp16")]; + fp16 var_22720_to_fp16 = const()[name = string("op_22720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2365_cast_fp16, y = var_22720_to_fp16)[name = string("aw_chunk_2365_cast_fp16")]; + fp16 var_22722_to_fp16 = const()[name = string("op_22722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2367_cast_fp16, y = var_22722_to_fp16)[name = string("aw_chunk_2367_cast_fp16")]; + fp16 var_22724_to_fp16 = const()[name = string("op_22724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2369_cast_fp16, y = var_22724_to_fp16)[name = string("aw_chunk_2369_cast_fp16")]; + fp16 var_22726_to_fp16 = const()[name = string("op_22726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2371_cast_fp16, y = var_22726_to_fp16)[name = string("aw_chunk_2371_cast_fp16")]; + fp16 var_22728_to_fp16 = const()[name = string("op_22728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2373_cast_fp16, y = var_22728_to_fp16)[name = string("aw_chunk_2373_cast_fp16")]; + fp16 var_22730_to_fp16 = const()[name = string("op_22730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2375_cast_fp16, y = var_22730_to_fp16)[name = string("aw_chunk_2375_cast_fp16")]; + fp16 var_22732_to_fp16 = const()[name = string("op_22732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2377_cast_fp16, y = var_22732_to_fp16)[name = string("aw_chunk_2377_cast_fp16")]; + fp16 var_22734_to_fp16 = const()[name = string("op_22734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2379_cast_fp16, y = var_22734_to_fp16)[name = string("aw_chunk_2379_cast_fp16")]; + fp16 var_22736_to_fp16 = const()[name = string("op_22736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2381_cast_fp16, y = var_22736_to_fp16)[name = string("aw_chunk_2381_cast_fp16")]; + fp16 var_22738_to_fp16 = const()[name = string("op_22738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2383_cast_fp16, y = var_22738_to_fp16)[name = string("aw_chunk_2383_cast_fp16")]; + fp16 var_22740_to_fp16 = const()[name = string("op_22740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2385_cast_fp16, y = var_22740_to_fp16)[name = string("aw_chunk_2385_cast_fp16")]; + fp16 var_22742_to_fp16 = const()[name = string("op_22742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2387_cast_fp16, y = var_22742_to_fp16)[name = string("aw_chunk_2387_cast_fp16")]; + fp16 var_22744_to_fp16 = const()[name = string("op_22744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2389_cast_fp16, y = var_22744_to_fp16)[name = string("aw_chunk_2389_cast_fp16")]; + fp16 var_22746_to_fp16 = const()[name = string("op_22746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2391_cast_fp16, y = var_22746_to_fp16)[name = string("aw_chunk_2391_cast_fp16")]; + fp16 var_22748_to_fp16 = const()[name = string("op_22748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2393_cast_fp16, y = var_22748_to_fp16)[name = string("aw_chunk_2393_cast_fp16")]; + fp16 var_22750_to_fp16 = const()[name = string("op_22750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2395_cast_fp16, y = var_22750_to_fp16)[name = string("aw_chunk_2395_cast_fp16")]; + fp16 var_22752_to_fp16 = const()[name = string("op_22752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2397_cast_fp16, y = var_22752_to_fp16)[name = string("aw_chunk_2397_cast_fp16")]; + fp16 var_22754_to_fp16 = const()[name = string("op_22754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2399_cast_fp16, y = var_22754_to_fp16)[name = string("aw_chunk_2399_cast_fp16")]; + tensor var_22756_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2241_cast_fp16)[name = string("op_22756_cast_fp16")]; + tensor var_22757_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2243_cast_fp16)[name = string("op_22757_cast_fp16")]; + tensor var_22758_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2245_cast_fp16)[name = string("op_22758_cast_fp16")]; + tensor var_22759_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2247_cast_fp16)[name = string("op_22759_cast_fp16")]; + tensor var_22760_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2249_cast_fp16)[name = string("op_22760_cast_fp16")]; + tensor var_22761_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2251_cast_fp16)[name = string("op_22761_cast_fp16")]; + tensor var_22762_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2253_cast_fp16)[name = string("op_22762_cast_fp16")]; + tensor var_22763_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2255_cast_fp16)[name = string("op_22763_cast_fp16")]; + tensor var_22764_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2257_cast_fp16)[name = string("op_22764_cast_fp16")]; + tensor var_22765_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2259_cast_fp16)[name = string("op_22765_cast_fp16")]; + tensor var_22766_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2261_cast_fp16)[name = string("op_22766_cast_fp16")]; + tensor var_22767_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2263_cast_fp16)[name = string("op_22767_cast_fp16")]; + tensor var_22768_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2265_cast_fp16)[name = string("op_22768_cast_fp16")]; + tensor var_22769_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2267_cast_fp16)[name = string("op_22769_cast_fp16")]; + tensor var_22770_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2269_cast_fp16)[name = string("op_22770_cast_fp16")]; + tensor var_22771_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2271_cast_fp16)[name = string("op_22771_cast_fp16")]; + tensor var_22772_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2273_cast_fp16)[name = string("op_22772_cast_fp16")]; + tensor var_22773_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2275_cast_fp16)[name = string("op_22773_cast_fp16")]; + tensor var_22774_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2277_cast_fp16)[name = string("op_22774_cast_fp16")]; + tensor var_22775_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2279_cast_fp16)[name = string("op_22775_cast_fp16")]; + tensor var_22776_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2281_cast_fp16)[name = string("op_22776_cast_fp16")]; + tensor var_22777_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2283_cast_fp16)[name = string("op_22777_cast_fp16")]; + tensor var_22778_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2285_cast_fp16)[name = string("op_22778_cast_fp16")]; + tensor var_22779_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2287_cast_fp16)[name = string("op_22779_cast_fp16")]; + tensor var_22780_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2289_cast_fp16)[name = string("op_22780_cast_fp16")]; + tensor var_22781_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2291_cast_fp16)[name = string("op_22781_cast_fp16")]; + tensor var_22782_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2293_cast_fp16)[name = string("op_22782_cast_fp16")]; + tensor var_22783_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2295_cast_fp16)[name = string("op_22783_cast_fp16")]; + tensor var_22784_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2297_cast_fp16)[name = string("op_22784_cast_fp16")]; + tensor var_22785_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2299_cast_fp16)[name = string("op_22785_cast_fp16")]; + tensor var_22786_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2301_cast_fp16)[name = string("op_22786_cast_fp16")]; + tensor var_22787_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2303_cast_fp16)[name = string("op_22787_cast_fp16")]; + tensor var_22788_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2305_cast_fp16)[name = string("op_22788_cast_fp16")]; + tensor var_22789_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2307_cast_fp16)[name = string("op_22789_cast_fp16")]; + tensor var_22790_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2309_cast_fp16)[name = string("op_22790_cast_fp16")]; + tensor var_22791_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2311_cast_fp16)[name = string("op_22791_cast_fp16")]; + tensor var_22792_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2313_cast_fp16)[name = string("op_22792_cast_fp16")]; + tensor var_22793_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2315_cast_fp16)[name = string("op_22793_cast_fp16")]; + tensor var_22794_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2317_cast_fp16)[name = string("op_22794_cast_fp16")]; + tensor var_22795_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2319_cast_fp16)[name = string("op_22795_cast_fp16")]; + tensor var_22796_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2321_cast_fp16)[name = string("op_22796_cast_fp16")]; + tensor var_22797_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2323_cast_fp16)[name = string("op_22797_cast_fp16")]; + tensor var_22798_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2325_cast_fp16)[name = string("op_22798_cast_fp16")]; + tensor var_22799_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2327_cast_fp16)[name = string("op_22799_cast_fp16")]; + tensor var_22800_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2329_cast_fp16)[name = string("op_22800_cast_fp16")]; + tensor var_22801_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2331_cast_fp16)[name = string("op_22801_cast_fp16")]; + tensor var_22802_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2333_cast_fp16)[name = string("op_22802_cast_fp16")]; + tensor var_22803_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2335_cast_fp16)[name = string("op_22803_cast_fp16")]; + tensor var_22804_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2337_cast_fp16)[name = string("op_22804_cast_fp16")]; + tensor var_22805_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2339_cast_fp16)[name = string("op_22805_cast_fp16")]; + tensor var_22806_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2341_cast_fp16)[name = string("op_22806_cast_fp16")]; + tensor var_22807_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2343_cast_fp16)[name = string("op_22807_cast_fp16")]; + tensor var_22808_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2345_cast_fp16)[name = string("op_22808_cast_fp16")]; + tensor var_22809_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2347_cast_fp16)[name = string("op_22809_cast_fp16")]; + tensor var_22810_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2349_cast_fp16)[name = string("op_22810_cast_fp16")]; + tensor var_22811_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2351_cast_fp16)[name = string("op_22811_cast_fp16")]; + tensor var_22812_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2353_cast_fp16)[name = string("op_22812_cast_fp16")]; + tensor var_22813_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2355_cast_fp16)[name = string("op_22813_cast_fp16")]; + tensor var_22814_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2357_cast_fp16)[name = string("op_22814_cast_fp16")]; + tensor var_22815_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2359_cast_fp16)[name = string("op_22815_cast_fp16")]; + tensor var_22816_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2361_cast_fp16)[name = string("op_22816_cast_fp16")]; + tensor var_22817_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2363_cast_fp16)[name = string("op_22817_cast_fp16")]; + tensor var_22818_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2365_cast_fp16)[name = string("op_22818_cast_fp16")]; + tensor var_22819_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2367_cast_fp16)[name = string("op_22819_cast_fp16")]; + tensor var_22820_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2369_cast_fp16)[name = string("op_22820_cast_fp16")]; + tensor var_22821_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2371_cast_fp16)[name = string("op_22821_cast_fp16")]; + tensor var_22822_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2373_cast_fp16)[name = string("op_22822_cast_fp16")]; + tensor var_22823_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2375_cast_fp16)[name = string("op_22823_cast_fp16")]; + tensor var_22824_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2377_cast_fp16)[name = string("op_22824_cast_fp16")]; + tensor var_22825_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2379_cast_fp16)[name = string("op_22825_cast_fp16")]; + tensor var_22826_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2381_cast_fp16)[name = string("op_22826_cast_fp16")]; + tensor var_22827_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2383_cast_fp16)[name = string("op_22827_cast_fp16")]; + tensor var_22828_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2385_cast_fp16)[name = string("op_22828_cast_fp16")]; + tensor var_22829_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2387_cast_fp16)[name = string("op_22829_cast_fp16")]; + tensor var_22830_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2389_cast_fp16)[name = string("op_22830_cast_fp16")]; + tensor var_22831_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2391_cast_fp16)[name = string("op_22831_cast_fp16")]; + tensor var_22832_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2393_cast_fp16)[name = string("op_22832_cast_fp16")]; + tensor var_22833_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2395_cast_fp16)[name = string("op_22833_cast_fp16")]; + tensor var_22834_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2397_cast_fp16)[name = string("op_22834_cast_fp16")]; + tensor var_22835_cast_fp16 = softmax(axis = var_21581, x = aw_chunk_2399_cast_fp16)[name = string("op_22835_cast_fp16")]; + string var_22837_equation_0 = const()[name = string("op_22837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22837_cast_fp16 = einsum(equation = var_22837_equation_0, values = (var_22357_cast_fp16, var_22756_cast_fp16))[name = string("op_22837_cast_fp16")]; + string var_22839_equation_0 = const()[name = string("op_22839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22839_cast_fp16 = einsum(equation = var_22839_equation_0, values = (var_22357_cast_fp16, var_22757_cast_fp16))[name = string("op_22839_cast_fp16")]; + string var_22841_equation_0 = const()[name = string("op_22841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22841_cast_fp16 = einsum(equation = var_22841_equation_0, values = (var_22357_cast_fp16, var_22758_cast_fp16))[name = string("op_22841_cast_fp16")]; + string var_22843_equation_0 = const()[name = string("op_22843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22843_cast_fp16 = einsum(equation = var_22843_equation_0, values = (var_22357_cast_fp16, var_22759_cast_fp16))[name = string("op_22843_cast_fp16")]; + string var_22845_equation_0 = const()[name = string("op_22845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22845_cast_fp16 = einsum(equation = var_22845_equation_0, values = (var_22361_cast_fp16, var_22760_cast_fp16))[name = string("op_22845_cast_fp16")]; + string var_22847_equation_0 = const()[name = string("op_22847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22847_cast_fp16 = einsum(equation = var_22847_equation_0, values = (var_22361_cast_fp16, var_22761_cast_fp16))[name = string("op_22847_cast_fp16")]; + string var_22849_equation_0 = const()[name = string("op_22849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22849_cast_fp16 = einsum(equation = var_22849_equation_0, values = (var_22361_cast_fp16, var_22762_cast_fp16))[name = string("op_22849_cast_fp16")]; + string var_22851_equation_0 = const()[name = string("op_22851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22851_cast_fp16 = einsum(equation = var_22851_equation_0, values = (var_22361_cast_fp16, var_22763_cast_fp16))[name = string("op_22851_cast_fp16")]; + string var_22853_equation_0 = const()[name = string("op_22853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22853_cast_fp16 = einsum(equation = var_22853_equation_0, values = (var_22365_cast_fp16, var_22764_cast_fp16))[name = string("op_22853_cast_fp16")]; + string var_22855_equation_0 = const()[name = string("op_22855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22855_cast_fp16 = einsum(equation = var_22855_equation_0, values = (var_22365_cast_fp16, var_22765_cast_fp16))[name = string("op_22855_cast_fp16")]; + string var_22857_equation_0 = const()[name = string("op_22857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22857_cast_fp16 = einsum(equation = var_22857_equation_0, values = (var_22365_cast_fp16, var_22766_cast_fp16))[name = string("op_22857_cast_fp16")]; + string var_22859_equation_0 = const()[name = string("op_22859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22859_cast_fp16 = einsum(equation = var_22859_equation_0, values = (var_22365_cast_fp16, var_22767_cast_fp16))[name = string("op_22859_cast_fp16")]; + string var_22861_equation_0 = const()[name = string("op_22861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22861_cast_fp16 = einsum(equation = var_22861_equation_0, values = (var_22369_cast_fp16, var_22768_cast_fp16))[name = string("op_22861_cast_fp16")]; + string var_22863_equation_0 = const()[name = string("op_22863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22863_cast_fp16 = einsum(equation = var_22863_equation_0, values = (var_22369_cast_fp16, var_22769_cast_fp16))[name = string("op_22863_cast_fp16")]; + string var_22865_equation_0 = const()[name = string("op_22865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22865_cast_fp16 = einsum(equation = var_22865_equation_0, values = (var_22369_cast_fp16, var_22770_cast_fp16))[name = string("op_22865_cast_fp16")]; + string var_22867_equation_0 = const()[name = string("op_22867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22867_cast_fp16 = einsum(equation = var_22867_equation_0, values = (var_22369_cast_fp16, var_22771_cast_fp16))[name = string("op_22867_cast_fp16")]; + string var_22869_equation_0 = const()[name = string("op_22869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22869_cast_fp16 = einsum(equation = var_22869_equation_0, values = (var_22373_cast_fp16, var_22772_cast_fp16))[name = string("op_22869_cast_fp16")]; + string var_22871_equation_0 = const()[name = string("op_22871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22871_cast_fp16 = einsum(equation = var_22871_equation_0, values = (var_22373_cast_fp16, var_22773_cast_fp16))[name = string("op_22871_cast_fp16")]; + string var_22873_equation_0 = const()[name = string("op_22873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22873_cast_fp16 = einsum(equation = var_22873_equation_0, values = (var_22373_cast_fp16, var_22774_cast_fp16))[name = string("op_22873_cast_fp16")]; + string var_22875_equation_0 = const()[name = string("op_22875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22875_cast_fp16 = einsum(equation = var_22875_equation_0, values = (var_22373_cast_fp16, var_22775_cast_fp16))[name = string("op_22875_cast_fp16")]; + string var_22877_equation_0 = const()[name = string("op_22877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22877_cast_fp16 = einsum(equation = var_22877_equation_0, values = (var_22377_cast_fp16, var_22776_cast_fp16))[name = string("op_22877_cast_fp16")]; + string var_22879_equation_0 = const()[name = string("op_22879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22879_cast_fp16 = einsum(equation = var_22879_equation_0, values = (var_22377_cast_fp16, var_22777_cast_fp16))[name = string("op_22879_cast_fp16")]; + string var_22881_equation_0 = const()[name = string("op_22881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22881_cast_fp16 = einsum(equation = var_22881_equation_0, values = (var_22377_cast_fp16, var_22778_cast_fp16))[name = string("op_22881_cast_fp16")]; + string var_22883_equation_0 = const()[name = string("op_22883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22883_cast_fp16 = einsum(equation = var_22883_equation_0, values = (var_22377_cast_fp16, var_22779_cast_fp16))[name = string("op_22883_cast_fp16")]; + string var_22885_equation_0 = const()[name = string("op_22885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22885_cast_fp16 = einsum(equation = var_22885_equation_0, values = (var_22381_cast_fp16, var_22780_cast_fp16))[name = string("op_22885_cast_fp16")]; + string var_22887_equation_0 = const()[name = string("op_22887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22887_cast_fp16 = einsum(equation = var_22887_equation_0, values = (var_22381_cast_fp16, var_22781_cast_fp16))[name = string("op_22887_cast_fp16")]; + string var_22889_equation_0 = const()[name = string("op_22889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22889_cast_fp16 = einsum(equation = var_22889_equation_0, values = (var_22381_cast_fp16, var_22782_cast_fp16))[name = string("op_22889_cast_fp16")]; + string var_22891_equation_0 = const()[name = string("op_22891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22891_cast_fp16 = einsum(equation = var_22891_equation_0, values = (var_22381_cast_fp16, var_22783_cast_fp16))[name = string("op_22891_cast_fp16")]; + string var_22893_equation_0 = const()[name = string("op_22893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22893_cast_fp16 = einsum(equation = var_22893_equation_0, values = (var_22385_cast_fp16, var_22784_cast_fp16))[name = string("op_22893_cast_fp16")]; + string var_22895_equation_0 = const()[name = string("op_22895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22895_cast_fp16 = einsum(equation = var_22895_equation_0, values = (var_22385_cast_fp16, var_22785_cast_fp16))[name = string("op_22895_cast_fp16")]; + string var_22897_equation_0 = const()[name = string("op_22897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22897_cast_fp16 = einsum(equation = var_22897_equation_0, values = (var_22385_cast_fp16, var_22786_cast_fp16))[name = string("op_22897_cast_fp16")]; + string var_22899_equation_0 = const()[name = string("op_22899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22899_cast_fp16 = einsum(equation = var_22899_equation_0, values = (var_22385_cast_fp16, var_22787_cast_fp16))[name = string("op_22899_cast_fp16")]; + string var_22901_equation_0 = const()[name = string("op_22901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22901_cast_fp16 = einsum(equation = var_22901_equation_0, values = (var_22389_cast_fp16, var_22788_cast_fp16))[name = string("op_22901_cast_fp16")]; + string var_22903_equation_0 = const()[name = string("op_22903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22903_cast_fp16 = einsum(equation = var_22903_equation_0, values = (var_22389_cast_fp16, var_22789_cast_fp16))[name = string("op_22903_cast_fp16")]; + string var_22905_equation_0 = const()[name = string("op_22905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22905_cast_fp16 = einsum(equation = var_22905_equation_0, values = (var_22389_cast_fp16, var_22790_cast_fp16))[name = string("op_22905_cast_fp16")]; + string var_22907_equation_0 = const()[name = string("op_22907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22907_cast_fp16 = einsum(equation = var_22907_equation_0, values = (var_22389_cast_fp16, var_22791_cast_fp16))[name = string("op_22907_cast_fp16")]; + string var_22909_equation_0 = const()[name = string("op_22909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22909_cast_fp16 = einsum(equation = var_22909_equation_0, values = (var_22393_cast_fp16, var_22792_cast_fp16))[name = string("op_22909_cast_fp16")]; + string var_22911_equation_0 = const()[name = string("op_22911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22911_cast_fp16 = einsum(equation = var_22911_equation_0, values = (var_22393_cast_fp16, var_22793_cast_fp16))[name = string("op_22911_cast_fp16")]; + string var_22913_equation_0 = const()[name = string("op_22913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22913_cast_fp16 = einsum(equation = var_22913_equation_0, values = (var_22393_cast_fp16, var_22794_cast_fp16))[name = string("op_22913_cast_fp16")]; + string var_22915_equation_0 = const()[name = string("op_22915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22915_cast_fp16 = einsum(equation = var_22915_equation_0, values = (var_22393_cast_fp16, var_22795_cast_fp16))[name = string("op_22915_cast_fp16")]; + string var_22917_equation_0 = const()[name = string("op_22917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22917_cast_fp16 = einsum(equation = var_22917_equation_0, values = (var_22397_cast_fp16, var_22796_cast_fp16))[name = string("op_22917_cast_fp16")]; + string var_22919_equation_0 = const()[name = string("op_22919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22919_cast_fp16 = einsum(equation = var_22919_equation_0, values = (var_22397_cast_fp16, var_22797_cast_fp16))[name = string("op_22919_cast_fp16")]; + string var_22921_equation_0 = const()[name = string("op_22921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22921_cast_fp16 = einsum(equation = var_22921_equation_0, values = (var_22397_cast_fp16, var_22798_cast_fp16))[name = string("op_22921_cast_fp16")]; + string var_22923_equation_0 = const()[name = string("op_22923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22923_cast_fp16 = einsum(equation = var_22923_equation_0, values = (var_22397_cast_fp16, var_22799_cast_fp16))[name = string("op_22923_cast_fp16")]; + string var_22925_equation_0 = const()[name = string("op_22925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22925_cast_fp16 = einsum(equation = var_22925_equation_0, values = (var_22401_cast_fp16, var_22800_cast_fp16))[name = string("op_22925_cast_fp16")]; + string var_22927_equation_0 = const()[name = string("op_22927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22927_cast_fp16 = einsum(equation = var_22927_equation_0, values = (var_22401_cast_fp16, var_22801_cast_fp16))[name = string("op_22927_cast_fp16")]; + string var_22929_equation_0 = const()[name = string("op_22929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22929_cast_fp16 = einsum(equation = var_22929_equation_0, values = (var_22401_cast_fp16, var_22802_cast_fp16))[name = string("op_22929_cast_fp16")]; + string var_22931_equation_0 = const()[name = string("op_22931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22931_cast_fp16 = einsum(equation = var_22931_equation_0, values = (var_22401_cast_fp16, var_22803_cast_fp16))[name = string("op_22931_cast_fp16")]; + string var_22933_equation_0 = const()[name = string("op_22933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22933_cast_fp16 = einsum(equation = var_22933_equation_0, values = (var_22405_cast_fp16, var_22804_cast_fp16))[name = string("op_22933_cast_fp16")]; + string var_22935_equation_0 = const()[name = string("op_22935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22935_cast_fp16 = einsum(equation = var_22935_equation_0, values = (var_22405_cast_fp16, var_22805_cast_fp16))[name = string("op_22935_cast_fp16")]; + string var_22937_equation_0 = const()[name = string("op_22937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22937_cast_fp16 = einsum(equation = var_22937_equation_0, values = (var_22405_cast_fp16, var_22806_cast_fp16))[name = string("op_22937_cast_fp16")]; + string var_22939_equation_0 = const()[name = string("op_22939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22939_cast_fp16 = einsum(equation = var_22939_equation_0, values = (var_22405_cast_fp16, var_22807_cast_fp16))[name = string("op_22939_cast_fp16")]; + string var_22941_equation_0 = const()[name = string("op_22941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22941_cast_fp16 = einsum(equation = var_22941_equation_0, values = (var_22409_cast_fp16, var_22808_cast_fp16))[name = string("op_22941_cast_fp16")]; + string var_22943_equation_0 = const()[name = string("op_22943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22943_cast_fp16 = einsum(equation = var_22943_equation_0, values = (var_22409_cast_fp16, var_22809_cast_fp16))[name = string("op_22943_cast_fp16")]; + string var_22945_equation_0 = const()[name = string("op_22945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22945_cast_fp16 = einsum(equation = var_22945_equation_0, values = (var_22409_cast_fp16, var_22810_cast_fp16))[name = string("op_22945_cast_fp16")]; + string var_22947_equation_0 = const()[name = string("op_22947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22947_cast_fp16 = einsum(equation = var_22947_equation_0, values = (var_22409_cast_fp16, var_22811_cast_fp16))[name = string("op_22947_cast_fp16")]; + string var_22949_equation_0 = const()[name = string("op_22949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22949_cast_fp16 = einsum(equation = var_22949_equation_0, values = (var_22413_cast_fp16, var_22812_cast_fp16))[name = string("op_22949_cast_fp16")]; + string var_22951_equation_0 = const()[name = string("op_22951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22951_cast_fp16 = einsum(equation = var_22951_equation_0, values = (var_22413_cast_fp16, var_22813_cast_fp16))[name = string("op_22951_cast_fp16")]; + string var_22953_equation_0 = const()[name = string("op_22953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22953_cast_fp16 = einsum(equation = var_22953_equation_0, values = (var_22413_cast_fp16, var_22814_cast_fp16))[name = string("op_22953_cast_fp16")]; + string var_22955_equation_0 = const()[name = string("op_22955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22955_cast_fp16 = einsum(equation = var_22955_equation_0, values = (var_22413_cast_fp16, var_22815_cast_fp16))[name = string("op_22955_cast_fp16")]; + string var_22957_equation_0 = const()[name = string("op_22957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22957_cast_fp16 = einsum(equation = var_22957_equation_0, values = (var_22417_cast_fp16, var_22816_cast_fp16))[name = string("op_22957_cast_fp16")]; + string var_22959_equation_0 = const()[name = string("op_22959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22959_cast_fp16 = einsum(equation = var_22959_equation_0, values = (var_22417_cast_fp16, var_22817_cast_fp16))[name = string("op_22959_cast_fp16")]; + string var_22961_equation_0 = const()[name = string("op_22961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22961_cast_fp16 = einsum(equation = var_22961_equation_0, values = (var_22417_cast_fp16, var_22818_cast_fp16))[name = string("op_22961_cast_fp16")]; + string var_22963_equation_0 = const()[name = string("op_22963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22963_cast_fp16 = einsum(equation = var_22963_equation_0, values = (var_22417_cast_fp16, var_22819_cast_fp16))[name = string("op_22963_cast_fp16")]; + string var_22965_equation_0 = const()[name = string("op_22965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22965_cast_fp16 = einsum(equation = var_22965_equation_0, values = (var_22421_cast_fp16, var_22820_cast_fp16))[name = string("op_22965_cast_fp16")]; + string var_22967_equation_0 = const()[name = string("op_22967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22967_cast_fp16 = einsum(equation = var_22967_equation_0, values = (var_22421_cast_fp16, var_22821_cast_fp16))[name = string("op_22967_cast_fp16")]; + string var_22969_equation_0 = const()[name = string("op_22969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22969_cast_fp16 = einsum(equation = var_22969_equation_0, values = (var_22421_cast_fp16, var_22822_cast_fp16))[name = string("op_22969_cast_fp16")]; + string var_22971_equation_0 = const()[name = string("op_22971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22971_cast_fp16 = einsum(equation = var_22971_equation_0, values = (var_22421_cast_fp16, var_22823_cast_fp16))[name = string("op_22971_cast_fp16")]; + string var_22973_equation_0 = const()[name = string("op_22973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22973_cast_fp16 = einsum(equation = var_22973_equation_0, values = (var_22425_cast_fp16, var_22824_cast_fp16))[name = string("op_22973_cast_fp16")]; + string var_22975_equation_0 = const()[name = string("op_22975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22975_cast_fp16 = einsum(equation = var_22975_equation_0, values = (var_22425_cast_fp16, var_22825_cast_fp16))[name = string("op_22975_cast_fp16")]; + string var_22977_equation_0 = const()[name = string("op_22977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22977_cast_fp16 = einsum(equation = var_22977_equation_0, values = (var_22425_cast_fp16, var_22826_cast_fp16))[name = string("op_22977_cast_fp16")]; + string var_22979_equation_0 = const()[name = string("op_22979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22979_cast_fp16 = einsum(equation = var_22979_equation_0, values = (var_22425_cast_fp16, var_22827_cast_fp16))[name = string("op_22979_cast_fp16")]; + string var_22981_equation_0 = const()[name = string("op_22981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22981_cast_fp16 = einsum(equation = var_22981_equation_0, values = (var_22429_cast_fp16, var_22828_cast_fp16))[name = string("op_22981_cast_fp16")]; + string var_22983_equation_0 = const()[name = string("op_22983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22983_cast_fp16 = einsum(equation = var_22983_equation_0, values = (var_22429_cast_fp16, var_22829_cast_fp16))[name = string("op_22983_cast_fp16")]; + string var_22985_equation_0 = const()[name = string("op_22985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22985_cast_fp16 = einsum(equation = var_22985_equation_0, values = (var_22429_cast_fp16, var_22830_cast_fp16))[name = string("op_22985_cast_fp16")]; + string var_22987_equation_0 = const()[name = string("op_22987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22987_cast_fp16 = einsum(equation = var_22987_equation_0, values = (var_22429_cast_fp16, var_22831_cast_fp16))[name = string("op_22987_cast_fp16")]; + string var_22989_equation_0 = const()[name = string("op_22989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22989_cast_fp16 = einsum(equation = var_22989_equation_0, values = (var_22433_cast_fp16, var_22832_cast_fp16))[name = string("op_22989_cast_fp16")]; + string var_22991_equation_0 = const()[name = string("op_22991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22991_cast_fp16 = einsum(equation = var_22991_equation_0, values = (var_22433_cast_fp16, var_22833_cast_fp16))[name = string("op_22991_cast_fp16")]; + string var_22993_equation_0 = const()[name = string("op_22993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22993_cast_fp16 = einsum(equation = var_22993_equation_0, values = (var_22433_cast_fp16, var_22834_cast_fp16))[name = string("op_22993_cast_fp16")]; + string var_22995_equation_0 = const()[name = string("op_22995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22995_cast_fp16 = einsum(equation = var_22995_equation_0, values = (var_22433_cast_fp16, var_22835_cast_fp16))[name = string("op_22995_cast_fp16")]; + bool var_22997_interleave_0 = const()[name = string("op_22997_interleave_0"), val = bool(false)]; + tensor var_22997_cast_fp16 = concat(axis = var_21556, interleave = var_22997_interleave_0, values = (var_22837_cast_fp16, var_22839_cast_fp16, var_22841_cast_fp16, var_22843_cast_fp16))[name = string("op_22997_cast_fp16")]; + bool var_22999_interleave_0 = const()[name = string("op_22999_interleave_0"), val = bool(false)]; + tensor var_22999_cast_fp16 = concat(axis = var_21556, interleave = var_22999_interleave_0, values = (var_22845_cast_fp16, var_22847_cast_fp16, var_22849_cast_fp16, var_22851_cast_fp16))[name = string("op_22999_cast_fp16")]; + bool var_23001_interleave_0 = const()[name = string("op_23001_interleave_0"), val = bool(false)]; + tensor var_23001_cast_fp16 = concat(axis = var_21556, interleave = var_23001_interleave_0, values = (var_22853_cast_fp16, var_22855_cast_fp16, var_22857_cast_fp16, var_22859_cast_fp16))[name = string("op_23001_cast_fp16")]; + bool var_23003_interleave_0 = const()[name = string("op_23003_interleave_0"), val = bool(false)]; + tensor var_23003_cast_fp16 = concat(axis = var_21556, interleave = var_23003_interleave_0, values = (var_22861_cast_fp16, var_22863_cast_fp16, var_22865_cast_fp16, var_22867_cast_fp16))[name = string("op_23003_cast_fp16")]; + bool var_23005_interleave_0 = const()[name = string("op_23005_interleave_0"), val = bool(false)]; + tensor var_23005_cast_fp16 = concat(axis = var_21556, interleave = var_23005_interleave_0, values = (var_22869_cast_fp16, var_22871_cast_fp16, var_22873_cast_fp16, var_22875_cast_fp16))[name = string("op_23005_cast_fp16")]; + bool var_23007_interleave_0 = const()[name = string("op_23007_interleave_0"), val = bool(false)]; + tensor var_23007_cast_fp16 = concat(axis = var_21556, interleave = var_23007_interleave_0, values = (var_22877_cast_fp16, var_22879_cast_fp16, var_22881_cast_fp16, var_22883_cast_fp16))[name = string("op_23007_cast_fp16")]; + bool var_23009_interleave_0 = const()[name = string("op_23009_interleave_0"), val = bool(false)]; + tensor var_23009_cast_fp16 = concat(axis = var_21556, interleave = var_23009_interleave_0, values = (var_22885_cast_fp16, var_22887_cast_fp16, var_22889_cast_fp16, var_22891_cast_fp16))[name = string("op_23009_cast_fp16")]; + bool var_23011_interleave_0 = const()[name = string("op_23011_interleave_0"), val = bool(false)]; + tensor var_23011_cast_fp16 = concat(axis = var_21556, interleave = var_23011_interleave_0, values = (var_22893_cast_fp16, var_22895_cast_fp16, var_22897_cast_fp16, var_22899_cast_fp16))[name = string("op_23011_cast_fp16")]; + bool var_23013_interleave_0 = const()[name = string("op_23013_interleave_0"), val = bool(false)]; + tensor var_23013_cast_fp16 = concat(axis = var_21556, interleave = var_23013_interleave_0, values = (var_22901_cast_fp16, var_22903_cast_fp16, var_22905_cast_fp16, var_22907_cast_fp16))[name = string("op_23013_cast_fp16")]; + bool var_23015_interleave_0 = const()[name = string("op_23015_interleave_0"), val = bool(false)]; + tensor var_23015_cast_fp16 = concat(axis = var_21556, interleave = var_23015_interleave_0, values = (var_22909_cast_fp16, var_22911_cast_fp16, var_22913_cast_fp16, var_22915_cast_fp16))[name = string("op_23015_cast_fp16")]; + bool var_23017_interleave_0 = const()[name = string("op_23017_interleave_0"), val = bool(false)]; + tensor var_23017_cast_fp16 = concat(axis = var_21556, interleave = var_23017_interleave_0, values = (var_22917_cast_fp16, var_22919_cast_fp16, var_22921_cast_fp16, var_22923_cast_fp16))[name = string("op_23017_cast_fp16")]; + bool var_23019_interleave_0 = const()[name = string("op_23019_interleave_0"), val = bool(false)]; + tensor var_23019_cast_fp16 = concat(axis = var_21556, interleave = var_23019_interleave_0, values = (var_22925_cast_fp16, var_22927_cast_fp16, var_22929_cast_fp16, var_22931_cast_fp16))[name = string("op_23019_cast_fp16")]; + bool var_23021_interleave_0 = const()[name = string("op_23021_interleave_0"), val = bool(false)]; + tensor var_23021_cast_fp16 = concat(axis = var_21556, interleave = var_23021_interleave_0, values = (var_22933_cast_fp16, var_22935_cast_fp16, var_22937_cast_fp16, var_22939_cast_fp16))[name = string("op_23021_cast_fp16")]; + bool var_23023_interleave_0 = const()[name = string("op_23023_interleave_0"), val = bool(false)]; + tensor var_23023_cast_fp16 = concat(axis = var_21556, interleave = var_23023_interleave_0, values = (var_22941_cast_fp16, var_22943_cast_fp16, var_22945_cast_fp16, var_22947_cast_fp16))[name = string("op_23023_cast_fp16")]; + bool var_23025_interleave_0 = const()[name = string("op_23025_interleave_0"), val = bool(false)]; + tensor var_23025_cast_fp16 = concat(axis = var_21556, interleave = var_23025_interleave_0, values = (var_22949_cast_fp16, var_22951_cast_fp16, var_22953_cast_fp16, var_22955_cast_fp16))[name = string("op_23025_cast_fp16")]; + bool var_23027_interleave_0 = const()[name = string("op_23027_interleave_0"), val = bool(false)]; + tensor var_23027_cast_fp16 = concat(axis = var_21556, interleave = var_23027_interleave_0, values = (var_22957_cast_fp16, var_22959_cast_fp16, var_22961_cast_fp16, var_22963_cast_fp16))[name = string("op_23027_cast_fp16")]; + bool var_23029_interleave_0 = const()[name = string("op_23029_interleave_0"), val = bool(false)]; + tensor var_23029_cast_fp16 = concat(axis = var_21556, interleave = var_23029_interleave_0, values = (var_22965_cast_fp16, var_22967_cast_fp16, var_22969_cast_fp16, var_22971_cast_fp16))[name = string("op_23029_cast_fp16")]; + bool var_23031_interleave_0 = const()[name = string("op_23031_interleave_0"), val = bool(false)]; + tensor var_23031_cast_fp16 = concat(axis = var_21556, interleave = var_23031_interleave_0, values = (var_22973_cast_fp16, var_22975_cast_fp16, var_22977_cast_fp16, var_22979_cast_fp16))[name = string("op_23031_cast_fp16")]; + bool var_23033_interleave_0 = const()[name = string("op_23033_interleave_0"), val = bool(false)]; + tensor var_23033_cast_fp16 = concat(axis = var_21556, interleave = var_23033_interleave_0, values = (var_22981_cast_fp16, var_22983_cast_fp16, var_22985_cast_fp16, var_22987_cast_fp16))[name = string("op_23033_cast_fp16")]; + bool var_23035_interleave_0 = const()[name = string("op_23035_interleave_0"), val = bool(false)]; + tensor var_23035_cast_fp16 = concat(axis = var_21556, interleave = var_23035_interleave_0, values = (var_22989_cast_fp16, var_22991_cast_fp16, var_22993_cast_fp16, var_22995_cast_fp16))[name = string("op_23035_cast_fp16")]; + bool input_113_interleave_0 = const()[name = string("input_113_interleave_0"), val = bool(false)]; + tensor input_113_cast_fp16 = concat(axis = var_21581, interleave = input_113_interleave_0, values = (var_22997_cast_fp16, var_22999_cast_fp16, var_23001_cast_fp16, var_23003_cast_fp16, var_23005_cast_fp16, var_23007_cast_fp16, var_23009_cast_fp16, var_23011_cast_fp16, var_23013_cast_fp16, var_23015_cast_fp16, var_23017_cast_fp16, var_23019_cast_fp16, var_23021_cast_fp16, var_23023_cast_fp16, var_23025_cast_fp16, var_23027_cast_fp16, var_23029_cast_fp16, var_23031_cast_fp16, var_23033_cast_fp16, var_23035_cast_fp16))[name = string("input_113_cast_fp16")]; + string obj_59_pad_type_0 = const()[name = string("obj_59_pad_type_0"), val = string("valid")]; + tensor obj_59_strides_0 = const()[name = string("obj_59_strides_0"), val = tensor([1, 1])]; + tensor obj_59_pad_0 = const()[name = string("obj_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_59_dilations_0 = const()[name = string("obj_59_dilations_0"), val = tensor([1, 1])]; + int32 obj_59_groups_0 = const()[name = string("obj_59_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575451200)))]; + tensor layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578728064)))]; + tensor obj_59_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_59_dilations_0, groups = obj_59_groups_0, pad = obj_59_pad_0, pad_type = obj_59_pad_type_0, strides = obj_59_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_59_cast_fp16)[name = string("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; + fp16 var_23054_to_fp16 = const()[name = string("op_23054_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_23054_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578730688)))]; + tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578733312)))]; + fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor layers_14_fc1_weight_to_fp16 = const()[name = string("layers_14_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578735936)))]; + tensor layers_14_fc1_bias_to_fp16 = const()[name = string("layers_14_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591843200)))]; + tensor input_117_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_14_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")]; + tensor input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")]; + string hidden_states_33_pad_type_0 = const()[name = string("hidden_states_33_pad_type_0"), val = string("valid")]; + tensor hidden_states_33_strides_0 = const()[name = string("hidden_states_33_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_33_pad_0 = const()[name = string("hidden_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_33_dilations_0 = const()[name = string("hidden_states_33_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_33_groups_0 = const()[name = string("hidden_states_33_groups_0"), val = int32(1)]; + tensor layers_14_fc2_weight_to_fp16 = const()[name = string("layers_14_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591853504)))]; + tensor layers_14_fc2_bias_to_fp16 = const()[name = string("layers_14_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604960768)))]; + tensor hidden_states_33_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_14_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = string("hidden_states_33_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_33_cast_fp16)[name = string("inputs_61_cast_fp16")]; + int32 var_23083 = const()[name = string("op_23083"), val = int32(3)]; + int32 var_23108 = const()[name = string("op_23108"), val = int32(1)]; + tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; + fp16 var_23125_to_fp16 = const()[name = string("op_23125_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_23125_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604963392)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604966016)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; + tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; + tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; + int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604968640)))]; + tensor layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608245504)))]; + tensor query_31_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_31_cast_fp16")]; + string key_31_pad_type_0 = const()[name = string("key_31_pad_type_0"), val = string("valid")]; + tensor key_31_strides_0 = const()[name = string("key_31_strides_0"), val = tensor([1, 1])]; + tensor key_31_pad_0 = const()[name = string("key_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_31_dilations_0 = const()[name = string("key_31_dilations_0"), val = tensor([1, 1])]; + int32 key_31_groups_0 = const()[name = string("key_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608248128)))]; + tensor key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("key_31_cast_fp16")]; + string value_31_pad_type_0 = const()[name = string("value_31_pad_type_0"), val = string("valid")]; + tensor value_31_strides_0 = const()[name = string("value_31_strides_0"), val = tensor([1, 1])]; + tensor value_31_pad_0 = const()[name = string("value_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_31_dilations_0 = const()[name = string("value_31_dilations_0"), val = tensor([1, 1])]; + int32 value_31_groups_0 = const()[name = string("value_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(611524992)))]; + tensor layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614801856)))]; + tensor value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("value_31_cast_fp16")]; + tensor var_23163_begin_0 = const()[name = string("op_23163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23163_end_0 = const()[name = string("op_23163_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23163_end_mask_0 = const()[name = string("op_23163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23163_cast_fp16 = slice_by_index(begin = var_23163_begin_0, end = var_23163_end_0, end_mask = var_23163_end_mask_0, x = query_31_cast_fp16)[name = string("op_23163_cast_fp16")]; + tensor var_23167_begin_0 = const()[name = string("op_23167_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23167_end_0 = const()[name = string("op_23167_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_23167_end_mask_0 = const()[name = string("op_23167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23167_cast_fp16 = slice_by_index(begin = var_23167_begin_0, end = var_23167_end_0, end_mask = var_23167_end_mask_0, x = query_31_cast_fp16)[name = string("op_23167_cast_fp16")]; + tensor var_23171_begin_0 = const()[name = string("op_23171_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23171_end_0 = const()[name = string("op_23171_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_23171_end_mask_0 = const()[name = string("op_23171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23171_cast_fp16 = slice_by_index(begin = var_23171_begin_0, end = var_23171_end_0, end_mask = var_23171_end_mask_0, x = query_31_cast_fp16)[name = string("op_23171_cast_fp16")]; + tensor var_23175_begin_0 = const()[name = string("op_23175_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23175_end_0 = const()[name = string("op_23175_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_23175_end_mask_0 = const()[name = string("op_23175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23175_cast_fp16 = slice_by_index(begin = var_23175_begin_0, end = var_23175_end_0, end_mask = var_23175_end_mask_0, x = query_31_cast_fp16)[name = string("op_23175_cast_fp16")]; + tensor var_23179_begin_0 = const()[name = string("op_23179_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23179_end_0 = const()[name = string("op_23179_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_23179_end_mask_0 = const()[name = string("op_23179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23179_cast_fp16 = slice_by_index(begin = var_23179_begin_0, end = var_23179_end_0, end_mask = var_23179_end_mask_0, x = query_31_cast_fp16)[name = string("op_23179_cast_fp16")]; + tensor var_23183_begin_0 = const()[name = string("op_23183_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23183_end_0 = const()[name = string("op_23183_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_23183_end_mask_0 = const()[name = string("op_23183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23183_cast_fp16 = slice_by_index(begin = var_23183_begin_0, end = var_23183_end_0, end_mask = var_23183_end_mask_0, x = query_31_cast_fp16)[name = string("op_23183_cast_fp16")]; + tensor var_23187_begin_0 = const()[name = string("op_23187_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23187_end_0 = const()[name = string("op_23187_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_23187_end_mask_0 = const()[name = string("op_23187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23187_cast_fp16 = slice_by_index(begin = var_23187_begin_0, end = var_23187_end_0, end_mask = var_23187_end_mask_0, x = query_31_cast_fp16)[name = string("op_23187_cast_fp16")]; + tensor var_23191_begin_0 = const()[name = string("op_23191_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23191_end_0 = const()[name = string("op_23191_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_23191_end_mask_0 = const()[name = string("op_23191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23191_cast_fp16 = slice_by_index(begin = var_23191_begin_0, end = var_23191_end_0, end_mask = var_23191_end_mask_0, x = query_31_cast_fp16)[name = string("op_23191_cast_fp16")]; + tensor var_23195_begin_0 = const()[name = string("op_23195_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23195_end_0 = const()[name = string("op_23195_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_23195_end_mask_0 = const()[name = string("op_23195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23195_cast_fp16 = slice_by_index(begin = var_23195_begin_0, end = var_23195_end_0, end_mask = var_23195_end_mask_0, x = query_31_cast_fp16)[name = string("op_23195_cast_fp16")]; + tensor var_23199_begin_0 = const()[name = string("op_23199_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23199_end_0 = const()[name = string("op_23199_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_23199_end_mask_0 = const()[name = string("op_23199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23199_cast_fp16 = slice_by_index(begin = var_23199_begin_0, end = var_23199_end_0, end_mask = var_23199_end_mask_0, x = query_31_cast_fp16)[name = string("op_23199_cast_fp16")]; + tensor var_23203_begin_0 = const()[name = string("op_23203_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23203_end_0 = const()[name = string("op_23203_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_23203_end_mask_0 = const()[name = string("op_23203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23203_cast_fp16 = slice_by_index(begin = var_23203_begin_0, end = var_23203_end_0, end_mask = var_23203_end_mask_0, x = query_31_cast_fp16)[name = string("op_23203_cast_fp16")]; + tensor var_23207_begin_0 = const()[name = string("op_23207_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23207_end_0 = const()[name = string("op_23207_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_23207_end_mask_0 = const()[name = string("op_23207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23207_cast_fp16 = slice_by_index(begin = var_23207_begin_0, end = var_23207_end_0, end_mask = var_23207_end_mask_0, x = query_31_cast_fp16)[name = string("op_23207_cast_fp16")]; + tensor var_23211_begin_0 = const()[name = string("op_23211_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23211_end_0 = const()[name = string("op_23211_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_23211_end_mask_0 = const()[name = string("op_23211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23211_cast_fp16 = slice_by_index(begin = var_23211_begin_0, end = var_23211_end_0, end_mask = var_23211_end_mask_0, x = query_31_cast_fp16)[name = string("op_23211_cast_fp16")]; + tensor var_23215_begin_0 = const()[name = string("op_23215_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23215_end_0 = const()[name = string("op_23215_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_23215_end_mask_0 = const()[name = string("op_23215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23215_cast_fp16 = slice_by_index(begin = var_23215_begin_0, end = var_23215_end_0, end_mask = var_23215_end_mask_0, x = query_31_cast_fp16)[name = string("op_23215_cast_fp16")]; + tensor var_23219_begin_0 = const()[name = string("op_23219_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23219_end_0 = const()[name = string("op_23219_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_23219_end_mask_0 = const()[name = string("op_23219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23219_cast_fp16 = slice_by_index(begin = var_23219_begin_0, end = var_23219_end_0, end_mask = var_23219_end_mask_0, x = query_31_cast_fp16)[name = string("op_23219_cast_fp16")]; + tensor var_23223_begin_0 = const()[name = string("op_23223_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23223_end_0 = const()[name = string("op_23223_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_23223_end_mask_0 = const()[name = string("op_23223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23223_cast_fp16 = slice_by_index(begin = var_23223_begin_0, end = var_23223_end_0, end_mask = var_23223_end_mask_0, x = query_31_cast_fp16)[name = string("op_23223_cast_fp16")]; + tensor var_23227_begin_0 = const()[name = string("op_23227_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23227_end_0 = const()[name = string("op_23227_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_23227_end_mask_0 = const()[name = string("op_23227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23227_cast_fp16 = slice_by_index(begin = var_23227_begin_0, end = var_23227_end_0, end_mask = var_23227_end_mask_0, x = query_31_cast_fp16)[name = string("op_23227_cast_fp16")]; + tensor var_23231_begin_0 = const()[name = string("op_23231_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23231_end_0 = const()[name = string("op_23231_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_23231_end_mask_0 = const()[name = string("op_23231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23231_cast_fp16 = slice_by_index(begin = var_23231_begin_0, end = var_23231_end_0, end_mask = var_23231_end_mask_0, x = query_31_cast_fp16)[name = string("op_23231_cast_fp16")]; + tensor var_23235_begin_0 = const()[name = string("op_23235_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23235_end_0 = const()[name = string("op_23235_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_23235_end_mask_0 = const()[name = string("op_23235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23235_cast_fp16 = slice_by_index(begin = var_23235_begin_0, end = var_23235_end_0, end_mask = var_23235_end_mask_0, x = query_31_cast_fp16)[name = string("op_23235_cast_fp16")]; + tensor var_23239_begin_0 = const()[name = string("op_23239_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23239_end_0 = const()[name = string("op_23239_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_23239_end_mask_0 = const()[name = string("op_23239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23239_cast_fp16 = slice_by_index(begin = var_23239_begin_0, end = var_23239_end_0, end_mask = var_23239_end_mask_0, x = query_31_cast_fp16)[name = string("op_23239_cast_fp16")]; + tensor var_23248_begin_0 = const()[name = string("op_23248_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23248_end_0 = const()[name = string("op_23248_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23248_end_mask_0 = const()[name = string("op_23248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23248_cast_fp16 = slice_by_index(begin = var_23248_begin_0, end = var_23248_end_0, end_mask = var_23248_end_mask_0, x = var_23163_cast_fp16)[name = string("op_23248_cast_fp16")]; + tensor var_23255_begin_0 = const()[name = string("op_23255_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23255_end_0 = const()[name = string("op_23255_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23255_end_mask_0 = const()[name = string("op_23255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23255_cast_fp16 = slice_by_index(begin = var_23255_begin_0, end = var_23255_end_0, end_mask = var_23255_end_mask_0, x = var_23163_cast_fp16)[name = string("op_23255_cast_fp16")]; + tensor var_23262_begin_0 = const()[name = string("op_23262_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23262_end_0 = const()[name = string("op_23262_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23262_end_mask_0 = const()[name = string("op_23262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23262_cast_fp16 = slice_by_index(begin = var_23262_begin_0, end = var_23262_end_0, end_mask = var_23262_end_mask_0, x = var_23163_cast_fp16)[name = string("op_23262_cast_fp16")]; + tensor var_23269_begin_0 = const()[name = string("op_23269_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23269_end_0 = const()[name = string("op_23269_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23269_end_mask_0 = const()[name = string("op_23269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23269_cast_fp16 = slice_by_index(begin = var_23269_begin_0, end = var_23269_end_0, end_mask = var_23269_end_mask_0, x = var_23163_cast_fp16)[name = string("op_23269_cast_fp16")]; + tensor var_23276_begin_0 = const()[name = string("op_23276_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23276_end_0 = const()[name = string("op_23276_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23276_end_mask_0 = const()[name = string("op_23276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23276_cast_fp16 = slice_by_index(begin = var_23276_begin_0, end = var_23276_end_0, end_mask = var_23276_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23276_cast_fp16")]; + tensor var_23283_begin_0 = const()[name = string("op_23283_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23283_end_0 = const()[name = string("op_23283_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23283_end_mask_0 = const()[name = string("op_23283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23283_cast_fp16 = slice_by_index(begin = var_23283_begin_0, end = var_23283_end_0, end_mask = var_23283_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23283_cast_fp16")]; + tensor var_23290_begin_0 = const()[name = string("op_23290_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23290_end_0 = const()[name = string("op_23290_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23290_end_mask_0 = const()[name = string("op_23290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23290_cast_fp16 = slice_by_index(begin = var_23290_begin_0, end = var_23290_end_0, end_mask = var_23290_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23290_cast_fp16")]; + tensor var_23297_begin_0 = const()[name = string("op_23297_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23297_end_0 = const()[name = string("op_23297_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23297_end_mask_0 = const()[name = string("op_23297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23297_cast_fp16 = slice_by_index(begin = var_23297_begin_0, end = var_23297_end_0, end_mask = var_23297_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23297_cast_fp16")]; + tensor var_23304_begin_0 = const()[name = string("op_23304_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23304_end_0 = const()[name = string("op_23304_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23304_end_mask_0 = const()[name = string("op_23304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23304_cast_fp16 = slice_by_index(begin = var_23304_begin_0, end = var_23304_end_0, end_mask = var_23304_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23304_cast_fp16")]; + tensor var_23311_begin_0 = const()[name = string("op_23311_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23311_end_0 = const()[name = string("op_23311_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23311_end_mask_0 = const()[name = string("op_23311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23311_cast_fp16 = slice_by_index(begin = var_23311_begin_0, end = var_23311_end_0, end_mask = var_23311_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23311_cast_fp16")]; + tensor var_23318_begin_0 = const()[name = string("op_23318_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23318_end_0 = const()[name = string("op_23318_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23318_end_mask_0 = const()[name = string("op_23318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23318_cast_fp16 = slice_by_index(begin = var_23318_begin_0, end = var_23318_end_0, end_mask = var_23318_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23318_cast_fp16")]; + tensor var_23325_begin_0 = const()[name = string("op_23325_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23325_end_0 = const()[name = string("op_23325_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23325_end_mask_0 = const()[name = string("op_23325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23325_cast_fp16 = slice_by_index(begin = var_23325_begin_0, end = var_23325_end_0, end_mask = var_23325_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23325_cast_fp16")]; + tensor var_23332_begin_0 = const()[name = string("op_23332_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23332_end_0 = const()[name = string("op_23332_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23332_end_mask_0 = const()[name = string("op_23332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23332_cast_fp16 = slice_by_index(begin = var_23332_begin_0, end = var_23332_end_0, end_mask = var_23332_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23332_cast_fp16")]; + tensor var_23339_begin_0 = const()[name = string("op_23339_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23339_end_0 = const()[name = string("op_23339_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23339_end_mask_0 = const()[name = string("op_23339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23339_cast_fp16 = slice_by_index(begin = var_23339_begin_0, end = var_23339_end_0, end_mask = var_23339_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23339_cast_fp16")]; + tensor var_23346_begin_0 = const()[name = string("op_23346_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23346_end_0 = const()[name = string("op_23346_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23346_end_mask_0 = const()[name = string("op_23346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23346_cast_fp16 = slice_by_index(begin = var_23346_begin_0, end = var_23346_end_0, end_mask = var_23346_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23346_cast_fp16")]; + tensor var_23353_begin_0 = const()[name = string("op_23353_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23353_end_0 = const()[name = string("op_23353_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23353_end_mask_0 = const()[name = string("op_23353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23353_cast_fp16 = slice_by_index(begin = var_23353_begin_0, end = var_23353_end_0, end_mask = var_23353_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23353_cast_fp16")]; + tensor var_23360_begin_0 = const()[name = string("op_23360_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23360_end_0 = const()[name = string("op_23360_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23360_end_mask_0 = const()[name = string("op_23360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23360_cast_fp16 = slice_by_index(begin = var_23360_begin_0, end = var_23360_end_0, end_mask = var_23360_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23360_cast_fp16")]; + tensor var_23367_begin_0 = const()[name = string("op_23367_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23367_end_0 = const()[name = string("op_23367_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23367_end_mask_0 = const()[name = string("op_23367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23367_cast_fp16 = slice_by_index(begin = var_23367_begin_0, end = var_23367_end_0, end_mask = var_23367_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23367_cast_fp16")]; + tensor var_23374_begin_0 = const()[name = string("op_23374_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23374_end_0 = const()[name = string("op_23374_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23374_end_mask_0 = const()[name = string("op_23374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23374_cast_fp16 = slice_by_index(begin = var_23374_begin_0, end = var_23374_end_0, end_mask = var_23374_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23374_cast_fp16")]; + tensor var_23381_begin_0 = const()[name = string("op_23381_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23381_end_0 = const()[name = string("op_23381_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23381_end_mask_0 = const()[name = string("op_23381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23381_cast_fp16 = slice_by_index(begin = var_23381_begin_0, end = var_23381_end_0, end_mask = var_23381_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23381_cast_fp16")]; + tensor var_23388_begin_0 = const()[name = string("op_23388_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23388_end_0 = const()[name = string("op_23388_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23388_end_mask_0 = const()[name = string("op_23388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23388_cast_fp16 = slice_by_index(begin = var_23388_begin_0, end = var_23388_end_0, end_mask = var_23388_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23388_cast_fp16")]; + tensor var_23395_begin_0 = const()[name = string("op_23395_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23395_end_0 = const()[name = string("op_23395_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23395_end_mask_0 = const()[name = string("op_23395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23395_cast_fp16 = slice_by_index(begin = var_23395_begin_0, end = var_23395_end_0, end_mask = var_23395_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23395_cast_fp16")]; + tensor var_23402_begin_0 = const()[name = string("op_23402_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23402_end_0 = const()[name = string("op_23402_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23402_end_mask_0 = const()[name = string("op_23402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23402_cast_fp16 = slice_by_index(begin = var_23402_begin_0, end = var_23402_end_0, end_mask = var_23402_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23402_cast_fp16")]; + tensor var_23409_begin_0 = const()[name = string("op_23409_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23409_end_0 = const()[name = string("op_23409_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23409_end_mask_0 = const()[name = string("op_23409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23409_cast_fp16 = slice_by_index(begin = var_23409_begin_0, end = var_23409_end_0, end_mask = var_23409_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23409_cast_fp16")]; + tensor var_23416_begin_0 = const()[name = string("op_23416_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23416_end_0 = const()[name = string("op_23416_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23416_end_mask_0 = const()[name = string("op_23416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23416_cast_fp16 = slice_by_index(begin = var_23416_begin_0, end = var_23416_end_0, end_mask = var_23416_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23416_cast_fp16")]; + tensor var_23423_begin_0 = const()[name = string("op_23423_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23423_end_0 = const()[name = string("op_23423_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23423_end_mask_0 = const()[name = string("op_23423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23423_cast_fp16 = slice_by_index(begin = var_23423_begin_0, end = var_23423_end_0, end_mask = var_23423_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23423_cast_fp16")]; + tensor var_23430_begin_0 = const()[name = string("op_23430_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23430_end_0 = const()[name = string("op_23430_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23430_end_mask_0 = const()[name = string("op_23430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23430_cast_fp16 = slice_by_index(begin = var_23430_begin_0, end = var_23430_end_0, end_mask = var_23430_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23430_cast_fp16")]; + tensor var_23437_begin_0 = const()[name = string("op_23437_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23437_end_0 = const()[name = string("op_23437_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23437_end_mask_0 = const()[name = string("op_23437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23437_cast_fp16 = slice_by_index(begin = var_23437_begin_0, end = var_23437_end_0, end_mask = var_23437_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23437_cast_fp16")]; + tensor var_23444_begin_0 = const()[name = string("op_23444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23444_end_0 = const()[name = string("op_23444_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23444_end_mask_0 = const()[name = string("op_23444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23444_cast_fp16 = slice_by_index(begin = var_23444_begin_0, end = var_23444_end_0, end_mask = var_23444_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23444_cast_fp16")]; + tensor var_23451_begin_0 = const()[name = string("op_23451_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23451_end_0 = const()[name = string("op_23451_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23451_end_mask_0 = const()[name = string("op_23451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23451_cast_fp16 = slice_by_index(begin = var_23451_begin_0, end = var_23451_end_0, end_mask = var_23451_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23451_cast_fp16")]; + tensor var_23458_begin_0 = const()[name = string("op_23458_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23458_end_0 = const()[name = string("op_23458_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23458_end_mask_0 = const()[name = string("op_23458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23458_cast_fp16 = slice_by_index(begin = var_23458_begin_0, end = var_23458_end_0, end_mask = var_23458_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23458_cast_fp16")]; + tensor var_23465_begin_0 = const()[name = string("op_23465_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23465_end_0 = const()[name = string("op_23465_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23465_end_mask_0 = const()[name = string("op_23465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23465_cast_fp16 = slice_by_index(begin = var_23465_begin_0, end = var_23465_end_0, end_mask = var_23465_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23465_cast_fp16")]; + tensor var_23472_begin_0 = const()[name = string("op_23472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23472_end_0 = const()[name = string("op_23472_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23472_end_mask_0 = const()[name = string("op_23472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23472_cast_fp16 = slice_by_index(begin = var_23472_begin_0, end = var_23472_end_0, end_mask = var_23472_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23472_cast_fp16")]; + tensor var_23479_begin_0 = const()[name = string("op_23479_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23479_end_0 = const()[name = string("op_23479_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23479_end_mask_0 = const()[name = string("op_23479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23479_cast_fp16 = slice_by_index(begin = var_23479_begin_0, end = var_23479_end_0, end_mask = var_23479_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23479_cast_fp16")]; + tensor var_23486_begin_0 = const()[name = string("op_23486_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23486_end_0 = const()[name = string("op_23486_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23486_end_mask_0 = const()[name = string("op_23486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23486_cast_fp16 = slice_by_index(begin = var_23486_begin_0, end = var_23486_end_0, end_mask = var_23486_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23486_cast_fp16")]; + tensor var_23493_begin_0 = const()[name = string("op_23493_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23493_end_0 = const()[name = string("op_23493_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23493_end_mask_0 = const()[name = string("op_23493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23493_cast_fp16 = slice_by_index(begin = var_23493_begin_0, end = var_23493_end_0, end_mask = var_23493_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23493_cast_fp16")]; + tensor var_23500_begin_0 = const()[name = string("op_23500_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23500_end_0 = const()[name = string("op_23500_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23500_end_mask_0 = const()[name = string("op_23500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23500_cast_fp16 = slice_by_index(begin = var_23500_begin_0, end = var_23500_end_0, end_mask = var_23500_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23500_cast_fp16")]; + tensor var_23507_begin_0 = const()[name = string("op_23507_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23507_end_0 = const()[name = string("op_23507_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23507_end_mask_0 = const()[name = string("op_23507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23507_cast_fp16 = slice_by_index(begin = var_23507_begin_0, end = var_23507_end_0, end_mask = var_23507_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23507_cast_fp16")]; + tensor var_23514_begin_0 = const()[name = string("op_23514_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23514_end_0 = const()[name = string("op_23514_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23514_end_mask_0 = const()[name = string("op_23514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23514_cast_fp16 = slice_by_index(begin = var_23514_begin_0, end = var_23514_end_0, end_mask = var_23514_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23514_cast_fp16")]; + tensor var_23521_begin_0 = const()[name = string("op_23521_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23521_end_0 = const()[name = string("op_23521_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23521_end_mask_0 = const()[name = string("op_23521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23521_cast_fp16 = slice_by_index(begin = var_23521_begin_0, end = var_23521_end_0, end_mask = var_23521_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23521_cast_fp16")]; + tensor var_23528_begin_0 = const()[name = string("op_23528_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23528_end_0 = const()[name = string("op_23528_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23528_end_mask_0 = const()[name = string("op_23528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23528_cast_fp16 = slice_by_index(begin = var_23528_begin_0, end = var_23528_end_0, end_mask = var_23528_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23528_cast_fp16")]; + tensor var_23535_begin_0 = const()[name = string("op_23535_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23535_end_0 = const()[name = string("op_23535_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23535_end_mask_0 = const()[name = string("op_23535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23535_cast_fp16 = slice_by_index(begin = var_23535_begin_0, end = var_23535_end_0, end_mask = var_23535_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23535_cast_fp16")]; + tensor var_23542_begin_0 = const()[name = string("op_23542_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23542_end_0 = const()[name = string("op_23542_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23542_end_mask_0 = const()[name = string("op_23542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23542_cast_fp16 = slice_by_index(begin = var_23542_begin_0, end = var_23542_end_0, end_mask = var_23542_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23542_cast_fp16")]; + tensor var_23549_begin_0 = const()[name = string("op_23549_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23549_end_0 = const()[name = string("op_23549_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23549_end_mask_0 = const()[name = string("op_23549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23549_cast_fp16 = slice_by_index(begin = var_23549_begin_0, end = var_23549_end_0, end_mask = var_23549_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23549_cast_fp16")]; + tensor var_23556_begin_0 = const()[name = string("op_23556_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23556_end_0 = const()[name = string("op_23556_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23556_end_mask_0 = const()[name = string("op_23556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23556_cast_fp16 = slice_by_index(begin = var_23556_begin_0, end = var_23556_end_0, end_mask = var_23556_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23556_cast_fp16")]; + tensor var_23563_begin_0 = const()[name = string("op_23563_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23563_end_0 = const()[name = string("op_23563_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23563_end_mask_0 = const()[name = string("op_23563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23563_cast_fp16 = slice_by_index(begin = var_23563_begin_0, end = var_23563_end_0, end_mask = var_23563_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23563_cast_fp16")]; + tensor var_23570_begin_0 = const()[name = string("op_23570_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23570_end_0 = const()[name = string("op_23570_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23570_end_mask_0 = const()[name = string("op_23570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23570_cast_fp16 = slice_by_index(begin = var_23570_begin_0, end = var_23570_end_0, end_mask = var_23570_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23570_cast_fp16")]; + tensor var_23577_begin_0 = const()[name = string("op_23577_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23577_end_0 = const()[name = string("op_23577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23577_end_mask_0 = const()[name = string("op_23577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23577_cast_fp16 = slice_by_index(begin = var_23577_begin_0, end = var_23577_end_0, end_mask = var_23577_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23577_cast_fp16")]; + tensor var_23584_begin_0 = const()[name = string("op_23584_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23584_end_0 = const()[name = string("op_23584_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23584_end_mask_0 = const()[name = string("op_23584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23584_cast_fp16 = slice_by_index(begin = var_23584_begin_0, end = var_23584_end_0, end_mask = var_23584_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23584_cast_fp16")]; + tensor var_23591_begin_0 = const()[name = string("op_23591_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23591_end_0 = const()[name = string("op_23591_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23591_end_mask_0 = const()[name = string("op_23591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23591_cast_fp16 = slice_by_index(begin = var_23591_begin_0, end = var_23591_end_0, end_mask = var_23591_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23591_cast_fp16")]; + tensor var_23598_begin_0 = const()[name = string("op_23598_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23598_end_0 = const()[name = string("op_23598_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23598_end_mask_0 = const()[name = string("op_23598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23598_cast_fp16 = slice_by_index(begin = var_23598_begin_0, end = var_23598_end_0, end_mask = var_23598_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23598_cast_fp16")]; + tensor var_23605_begin_0 = const()[name = string("op_23605_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23605_end_0 = const()[name = string("op_23605_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23605_end_mask_0 = const()[name = string("op_23605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23605_cast_fp16 = slice_by_index(begin = var_23605_begin_0, end = var_23605_end_0, end_mask = var_23605_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23605_cast_fp16")]; + tensor var_23612_begin_0 = const()[name = string("op_23612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23612_end_0 = const()[name = string("op_23612_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23612_end_mask_0 = const()[name = string("op_23612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23612_cast_fp16 = slice_by_index(begin = var_23612_begin_0, end = var_23612_end_0, end_mask = var_23612_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23612_cast_fp16")]; + tensor var_23619_begin_0 = const()[name = string("op_23619_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23619_end_0 = const()[name = string("op_23619_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23619_end_mask_0 = const()[name = string("op_23619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23619_cast_fp16 = slice_by_index(begin = var_23619_begin_0, end = var_23619_end_0, end_mask = var_23619_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23619_cast_fp16")]; + tensor var_23626_begin_0 = const()[name = string("op_23626_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23626_end_0 = const()[name = string("op_23626_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23626_end_mask_0 = const()[name = string("op_23626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23626_cast_fp16 = slice_by_index(begin = var_23626_begin_0, end = var_23626_end_0, end_mask = var_23626_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23626_cast_fp16")]; + tensor var_23633_begin_0 = const()[name = string("op_23633_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23633_end_0 = const()[name = string("op_23633_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23633_end_mask_0 = const()[name = string("op_23633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23633_cast_fp16 = slice_by_index(begin = var_23633_begin_0, end = var_23633_end_0, end_mask = var_23633_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23633_cast_fp16")]; + tensor var_23640_begin_0 = const()[name = string("op_23640_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23640_end_0 = const()[name = string("op_23640_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23640_end_mask_0 = const()[name = string("op_23640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23640_cast_fp16 = slice_by_index(begin = var_23640_begin_0, end = var_23640_end_0, end_mask = var_23640_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23640_cast_fp16")]; + tensor var_23647_begin_0 = const()[name = string("op_23647_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23647_end_0 = const()[name = string("op_23647_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23647_end_mask_0 = const()[name = string("op_23647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23647_cast_fp16 = slice_by_index(begin = var_23647_begin_0, end = var_23647_end_0, end_mask = var_23647_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23647_cast_fp16")]; + tensor var_23654_begin_0 = const()[name = string("op_23654_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23654_end_0 = const()[name = string("op_23654_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23654_end_mask_0 = const()[name = string("op_23654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23654_cast_fp16 = slice_by_index(begin = var_23654_begin_0, end = var_23654_end_0, end_mask = var_23654_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23654_cast_fp16")]; + tensor var_23661_begin_0 = const()[name = string("op_23661_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23661_end_0 = const()[name = string("op_23661_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23661_end_mask_0 = const()[name = string("op_23661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23661_cast_fp16 = slice_by_index(begin = var_23661_begin_0, end = var_23661_end_0, end_mask = var_23661_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23661_cast_fp16")]; + tensor var_23668_begin_0 = const()[name = string("op_23668_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23668_end_0 = const()[name = string("op_23668_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23668_end_mask_0 = const()[name = string("op_23668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23668_cast_fp16 = slice_by_index(begin = var_23668_begin_0, end = var_23668_end_0, end_mask = var_23668_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23668_cast_fp16")]; + tensor var_23675_begin_0 = const()[name = string("op_23675_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23675_end_0 = const()[name = string("op_23675_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23675_end_mask_0 = const()[name = string("op_23675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23675_cast_fp16 = slice_by_index(begin = var_23675_begin_0, end = var_23675_end_0, end_mask = var_23675_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23675_cast_fp16")]; + tensor var_23682_begin_0 = const()[name = string("op_23682_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23682_end_0 = const()[name = string("op_23682_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23682_end_mask_0 = const()[name = string("op_23682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23682_cast_fp16 = slice_by_index(begin = var_23682_begin_0, end = var_23682_end_0, end_mask = var_23682_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23682_cast_fp16")]; + tensor var_23689_begin_0 = const()[name = string("op_23689_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23689_end_0 = const()[name = string("op_23689_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23689_end_mask_0 = const()[name = string("op_23689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23689_cast_fp16 = slice_by_index(begin = var_23689_begin_0, end = var_23689_end_0, end_mask = var_23689_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23689_cast_fp16")]; + tensor var_23696_begin_0 = const()[name = string("op_23696_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23696_end_0 = const()[name = string("op_23696_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23696_end_mask_0 = const()[name = string("op_23696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23696_cast_fp16 = slice_by_index(begin = var_23696_begin_0, end = var_23696_end_0, end_mask = var_23696_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23696_cast_fp16")]; + tensor var_23703_begin_0 = const()[name = string("op_23703_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23703_end_0 = const()[name = string("op_23703_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23703_end_mask_0 = const()[name = string("op_23703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23703_cast_fp16 = slice_by_index(begin = var_23703_begin_0, end = var_23703_end_0, end_mask = var_23703_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23703_cast_fp16")]; + tensor var_23710_begin_0 = const()[name = string("op_23710_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23710_end_0 = const()[name = string("op_23710_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23710_end_mask_0 = const()[name = string("op_23710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23710_cast_fp16 = slice_by_index(begin = var_23710_begin_0, end = var_23710_end_0, end_mask = var_23710_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23710_cast_fp16")]; + tensor var_23717_begin_0 = const()[name = string("op_23717_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23717_end_0 = const()[name = string("op_23717_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23717_end_mask_0 = const()[name = string("op_23717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23717_cast_fp16 = slice_by_index(begin = var_23717_begin_0, end = var_23717_end_0, end_mask = var_23717_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23717_cast_fp16")]; + tensor var_23724_begin_0 = const()[name = string("op_23724_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23724_end_0 = const()[name = string("op_23724_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23724_end_mask_0 = const()[name = string("op_23724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23724_cast_fp16 = slice_by_index(begin = var_23724_begin_0, end = var_23724_end_0, end_mask = var_23724_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23724_cast_fp16")]; + tensor var_23731_begin_0 = const()[name = string("op_23731_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23731_end_0 = const()[name = string("op_23731_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23731_end_mask_0 = const()[name = string("op_23731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23731_cast_fp16 = slice_by_index(begin = var_23731_begin_0, end = var_23731_end_0, end_mask = var_23731_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23731_cast_fp16")]; + tensor var_23738_begin_0 = const()[name = string("op_23738_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23738_end_0 = const()[name = string("op_23738_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23738_end_mask_0 = const()[name = string("op_23738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23738_cast_fp16 = slice_by_index(begin = var_23738_begin_0, end = var_23738_end_0, end_mask = var_23738_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23738_cast_fp16")]; + tensor var_23745_begin_0 = const()[name = string("op_23745_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23745_end_0 = const()[name = string("op_23745_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23745_end_mask_0 = const()[name = string("op_23745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23745_cast_fp16 = slice_by_index(begin = var_23745_begin_0, end = var_23745_end_0, end_mask = var_23745_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23745_cast_fp16")]; + tensor var_23752_begin_0 = const()[name = string("op_23752_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23752_end_0 = const()[name = string("op_23752_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23752_end_mask_0 = const()[name = string("op_23752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23752_cast_fp16 = slice_by_index(begin = var_23752_begin_0, end = var_23752_end_0, end_mask = var_23752_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23752_cast_fp16")]; + tensor var_23759_begin_0 = const()[name = string("op_23759_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23759_end_0 = const()[name = string("op_23759_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23759_end_mask_0 = const()[name = string("op_23759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23759_cast_fp16 = slice_by_index(begin = var_23759_begin_0, end = var_23759_end_0, end_mask = var_23759_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23759_cast_fp16")]; + tensor var_23766_begin_0 = const()[name = string("op_23766_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23766_end_0 = const()[name = string("op_23766_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23766_end_mask_0 = const()[name = string("op_23766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23766_cast_fp16 = slice_by_index(begin = var_23766_begin_0, end = var_23766_end_0, end_mask = var_23766_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23766_cast_fp16")]; + tensor var_23773_begin_0 = const()[name = string("op_23773_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23773_end_0 = const()[name = string("op_23773_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23773_end_mask_0 = const()[name = string("op_23773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23773_cast_fp16 = slice_by_index(begin = var_23773_begin_0, end = var_23773_end_0, end_mask = var_23773_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23773_cast_fp16")]; + tensor var_23780_begin_0 = const()[name = string("op_23780_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23780_end_0 = const()[name = string("op_23780_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23780_end_mask_0 = const()[name = string("op_23780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23780_cast_fp16 = slice_by_index(begin = var_23780_begin_0, end = var_23780_end_0, end_mask = var_23780_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23780_cast_fp16")]; + tensor var_23787_begin_0 = const()[name = string("op_23787_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23787_end_0 = const()[name = string("op_23787_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23787_end_mask_0 = const()[name = string("op_23787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23787_cast_fp16 = slice_by_index(begin = var_23787_begin_0, end = var_23787_end_0, end_mask = var_23787_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23787_cast_fp16")]; + tensor var_23794_begin_0 = const()[name = string("op_23794_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23794_end_0 = const()[name = string("op_23794_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23794_end_mask_0 = const()[name = string("op_23794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23794_cast_fp16 = slice_by_index(begin = var_23794_begin_0, end = var_23794_end_0, end_mask = var_23794_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23794_cast_fp16")]; + tensor var_23801_begin_0 = const()[name = string("op_23801_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23801_end_0 = const()[name = string("op_23801_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23801_end_mask_0 = const()[name = string("op_23801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23801_cast_fp16 = slice_by_index(begin = var_23801_begin_0, end = var_23801_end_0, end_mask = var_23801_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23801_cast_fp16")]; + tensor k_31_perm_0 = const()[name = string("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23806_begin_0 = const()[name = string("op_23806_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23806_end_0 = const()[name = string("op_23806_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_23806_end_mask_0 = const()[name = string("op_23806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_31_cast_fp16 = transpose(perm = k_31_perm_0, x = key_31_cast_fp16)[name = string("transpose_16")]; + tensor var_23806_cast_fp16 = slice_by_index(begin = var_23806_begin_0, end = var_23806_end_0, end_mask = var_23806_end_mask_0, x = k_31_cast_fp16)[name = string("op_23806_cast_fp16")]; + tensor var_23810_begin_0 = const()[name = string("op_23810_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23810_end_0 = const()[name = string("op_23810_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_23810_end_mask_0 = const()[name = string("op_23810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23810_cast_fp16 = slice_by_index(begin = var_23810_begin_0, end = var_23810_end_0, end_mask = var_23810_end_mask_0, x = k_31_cast_fp16)[name = string("op_23810_cast_fp16")]; + tensor var_23814_begin_0 = const()[name = string("op_23814_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23814_end_0 = const()[name = string("op_23814_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_23814_end_mask_0 = const()[name = string("op_23814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23814_cast_fp16 = slice_by_index(begin = var_23814_begin_0, end = var_23814_end_0, end_mask = var_23814_end_mask_0, x = k_31_cast_fp16)[name = string("op_23814_cast_fp16")]; + tensor var_23818_begin_0 = const()[name = string("op_23818_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23818_end_0 = const()[name = string("op_23818_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_23818_end_mask_0 = const()[name = string("op_23818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23818_cast_fp16 = slice_by_index(begin = var_23818_begin_0, end = var_23818_end_0, end_mask = var_23818_end_mask_0, x = k_31_cast_fp16)[name = string("op_23818_cast_fp16")]; + tensor var_23822_begin_0 = const()[name = string("op_23822_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23822_end_0 = const()[name = string("op_23822_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_23822_end_mask_0 = const()[name = string("op_23822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23822_cast_fp16 = slice_by_index(begin = var_23822_begin_0, end = var_23822_end_0, end_mask = var_23822_end_mask_0, x = k_31_cast_fp16)[name = string("op_23822_cast_fp16")]; + tensor var_23826_begin_0 = const()[name = string("op_23826_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23826_end_0 = const()[name = string("op_23826_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_23826_end_mask_0 = const()[name = string("op_23826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23826_cast_fp16 = slice_by_index(begin = var_23826_begin_0, end = var_23826_end_0, end_mask = var_23826_end_mask_0, x = k_31_cast_fp16)[name = string("op_23826_cast_fp16")]; + tensor var_23830_begin_0 = const()[name = string("op_23830_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23830_end_0 = const()[name = string("op_23830_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_23830_end_mask_0 = const()[name = string("op_23830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23830_cast_fp16 = slice_by_index(begin = var_23830_begin_0, end = var_23830_end_0, end_mask = var_23830_end_mask_0, x = k_31_cast_fp16)[name = string("op_23830_cast_fp16")]; + tensor var_23834_begin_0 = const()[name = string("op_23834_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23834_end_0 = const()[name = string("op_23834_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_23834_end_mask_0 = const()[name = string("op_23834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23834_cast_fp16 = slice_by_index(begin = var_23834_begin_0, end = var_23834_end_0, end_mask = var_23834_end_mask_0, x = k_31_cast_fp16)[name = string("op_23834_cast_fp16")]; + tensor var_23838_begin_0 = const()[name = string("op_23838_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23838_end_0 = const()[name = string("op_23838_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_23838_end_mask_0 = const()[name = string("op_23838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23838_cast_fp16 = slice_by_index(begin = var_23838_begin_0, end = var_23838_end_0, end_mask = var_23838_end_mask_0, x = k_31_cast_fp16)[name = string("op_23838_cast_fp16")]; + tensor var_23842_begin_0 = const()[name = string("op_23842_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23842_end_0 = const()[name = string("op_23842_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_23842_end_mask_0 = const()[name = string("op_23842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23842_cast_fp16 = slice_by_index(begin = var_23842_begin_0, end = var_23842_end_0, end_mask = var_23842_end_mask_0, x = k_31_cast_fp16)[name = string("op_23842_cast_fp16")]; + tensor var_23846_begin_0 = const()[name = string("op_23846_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23846_end_0 = const()[name = string("op_23846_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_23846_end_mask_0 = const()[name = string("op_23846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23846_cast_fp16 = slice_by_index(begin = var_23846_begin_0, end = var_23846_end_0, end_mask = var_23846_end_mask_0, x = k_31_cast_fp16)[name = string("op_23846_cast_fp16")]; + tensor var_23850_begin_0 = const()[name = string("op_23850_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23850_end_0 = const()[name = string("op_23850_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_23850_end_mask_0 = const()[name = string("op_23850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23850_cast_fp16 = slice_by_index(begin = var_23850_begin_0, end = var_23850_end_0, end_mask = var_23850_end_mask_0, x = k_31_cast_fp16)[name = string("op_23850_cast_fp16")]; + tensor var_23854_begin_0 = const()[name = string("op_23854_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23854_end_0 = const()[name = string("op_23854_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_23854_end_mask_0 = const()[name = string("op_23854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23854_cast_fp16 = slice_by_index(begin = var_23854_begin_0, end = var_23854_end_0, end_mask = var_23854_end_mask_0, x = k_31_cast_fp16)[name = string("op_23854_cast_fp16")]; + tensor var_23858_begin_0 = const()[name = string("op_23858_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23858_end_0 = const()[name = string("op_23858_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_23858_end_mask_0 = const()[name = string("op_23858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23858_cast_fp16 = slice_by_index(begin = var_23858_begin_0, end = var_23858_end_0, end_mask = var_23858_end_mask_0, x = k_31_cast_fp16)[name = string("op_23858_cast_fp16")]; + tensor var_23862_begin_0 = const()[name = string("op_23862_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23862_end_0 = const()[name = string("op_23862_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_23862_end_mask_0 = const()[name = string("op_23862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23862_cast_fp16 = slice_by_index(begin = var_23862_begin_0, end = var_23862_end_0, end_mask = var_23862_end_mask_0, x = k_31_cast_fp16)[name = string("op_23862_cast_fp16")]; + tensor var_23866_begin_0 = const()[name = string("op_23866_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23866_end_0 = const()[name = string("op_23866_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_23866_end_mask_0 = const()[name = string("op_23866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23866_cast_fp16 = slice_by_index(begin = var_23866_begin_0, end = var_23866_end_0, end_mask = var_23866_end_mask_0, x = k_31_cast_fp16)[name = string("op_23866_cast_fp16")]; + tensor var_23870_begin_0 = const()[name = string("op_23870_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23870_end_0 = const()[name = string("op_23870_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_23870_end_mask_0 = const()[name = string("op_23870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23870_cast_fp16 = slice_by_index(begin = var_23870_begin_0, end = var_23870_end_0, end_mask = var_23870_end_mask_0, x = k_31_cast_fp16)[name = string("op_23870_cast_fp16")]; + tensor var_23874_begin_0 = const()[name = string("op_23874_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23874_end_0 = const()[name = string("op_23874_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_23874_end_mask_0 = const()[name = string("op_23874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23874_cast_fp16 = slice_by_index(begin = var_23874_begin_0, end = var_23874_end_0, end_mask = var_23874_end_mask_0, x = k_31_cast_fp16)[name = string("op_23874_cast_fp16")]; + tensor var_23878_begin_0 = const()[name = string("op_23878_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23878_end_0 = const()[name = string("op_23878_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_23878_end_mask_0 = const()[name = string("op_23878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23878_cast_fp16 = slice_by_index(begin = var_23878_begin_0, end = var_23878_end_0, end_mask = var_23878_end_mask_0, x = k_31_cast_fp16)[name = string("op_23878_cast_fp16")]; + tensor var_23882_begin_0 = const()[name = string("op_23882_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23882_end_0 = const()[name = string("op_23882_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_23882_end_mask_0 = const()[name = string("op_23882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23882_cast_fp16 = slice_by_index(begin = var_23882_begin_0, end = var_23882_end_0, end_mask = var_23882_end_mask_0, x = k_31_cast_fp16)[name = string("op_23882_cast_fp16")]; + tensor var_23884_begin_0 = const()[name = string("op_23884_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23884_end_0 = const()[name = string("op_23884_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23884_end_mask_0 = const()[name = string("op_23884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23884_cast_fp16 = slice_by_index(begin = var_23884_begin_0, end = var_23884_end_0, end_mask = var_23884_end_mask_0, x = value_31_cast_fp16)[name = string("op_23884_cast_fp16")]; + tensor var_23888_begin_0 = const()[name = string("op_23888_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23888_end_0 = const()[name = string("op_23888_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_23888_end_mask_0 = const()[name = string("op_23888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23888_cast_fp16 = slice_by_index(begin = var_23888_begin_0, end = var_23888_end_0, end_mask = var_23888_end_mask_0, x = value_31_cast_fp16)[name = string("op_23888_cast_fp16")]; + tensor var_23892_begin_0 = const()[name = string("op_23892_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23892_end_0 = const()[name = string("op_23892_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_23892_end_mask_0 = const()[name = string("op_23892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23892_cast_fp16 = slice_by_index(begin = var_23892_begin_0, end = var_23892_end_0, end_mask = var_23892_end_mask_0, x = value_31_cast_fp16)[name = string("op_23892_cast_fp16")]; + tensor var_23896_begin_0 = const()[name = string("op_23896_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23896_end_0 = const()[name = string("op_23896_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_23896_end_mask_0 = const()[name = string("op_23896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23896_cast_fp16 = slice_by_index(begin = var_23896_begin_0, end = var_23896_end_0, end_mask = var_23896_end_mask_0, x = value_31_cast_fp16)[name = string("op_23896_cast_fp16")]; + tensor var_23900_begin_0 = const()[name = string("op_23900_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23900_end_0 = const()[name = string("op_23900_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_23900_end_mask_0 = const()[name = string("op_23900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23900_cast_fp16 = slice_by_index(begin = var_23900_begin_0, end = var_23900_end_0, end_mask = var_23900_end_mask_0, x = value_31_cast_fp16)[name = string("op_23900_cast_fp16")]; + tensor var_23904_begin_0 = const()[name = string("op_23904_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23904_end_0 = const()[name = string("op_23904_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_23904_end_mask_0 = const()[name = string("op_23904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23904_cast_fp16 = slice_by_index(begin = var_23904_begin_0, end = var_23904_end_0, end_mask = var_23904_end_mask_0, x = value_31_cast_fp16)[name = string("op_23904_cast_fp16")]; + tensor var_23908_begin_0 = const()[name = string("op_23908_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23908_end_0 = const()[name = string("op_23908_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_23908_end_mask_0 = const()[name = string("op_23908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23908_cast_fp16 = slice_by_index(begin = var_23908_begin_0, end = var_23908_end_0, end_mask = var_23908_end_mask_0, x = value_31_cast_fp16)[name = string("op_23908_cast_fp16")]; + tensor var_23912_begin_0 = const()[name = string("op_23912_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23912_end_0 = const()[name = string("op_23912_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_23912_end_mask_0 = const()[name = string("op_23912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23912_cast_fp16 = slice_by_index(begin = var_23912_begin_0, end = var_23912_end_0, end_mask = var_23912_end_mask_0, x = value_31_cast_fp16)[name = string("op_23912_cast_fp16")]; + tensor var_23916_begin_0 = const()[name = string("op_23916_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23916_end_0 = const()[name = string("op_23916_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_23916_end_mask_0 = const()[name = string("op_23916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23916_cast_fp16 = slice_by_index(begin = var_23916_begin_0, end = var_23916_end_0, end_mask = var_23916_end_mask_0, x = value_31_cast_fp16)[name = string("op_23916_cast_fp16")]; + tensor var_23920_begin_0 = const()[name = string("op_23920_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23920_end_0 = const()[name = string("op_23920_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_23920_end_mask_0 = const()[name = string("op_23920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23920_cast_fp16 = slice_by_index(begin = var_23920_begin_0, end = var_23920_end_0, end_mask = var_23920_end_mask_0, x = value_31_cast_fp16)[name = string("op_23920_cast_fp16")]; + tensor var_23924_begin_0 = const()[name = string("op_23924_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23924_end_0 = const()[name = string("op_23924_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_23924_end_mask_0 = const()[name = string("op_23924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23924_cast_fp16 = slice_by_index(begin = var_23924_begin_0, end = var_23924_end_0, end_mask = var_23924_end_mask_0, x = value_31_cast_fp16)[name = string("op_23924_cast_fp16")]; + tensor var_23928_begin_0 = const()[name = string("op_23928_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23928_end_0 = const()[name = string("op_23928_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_23928_end_mask_0 = const()[name = string("op_23928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23928_cast_fp16 = slice_by_index(begin = var_23928_begin_0, end = var_23928_end_0, end_mask = var_23928_end_mask_0, x = value_31_cast_fp16)[name = string("op_23928_cast_fp16")]; + tensor var_23932_begin_0 = const()[name = string("op_23932_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23932_end_0 = const()[name = string("op_23932_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_23932_end_mask_0 = const()[name = string("op_23932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23932_cast_fp16 = slice_by_index(begin = var_23932_begin_0, end = var_23932_end_0, end_mask = var_23932_end_mask_0, x = value_31_cast_fp16)[name = string("op_23932_cast_fp16")]; + tensor var_23936_begin_0 = const()[name = string("op_23936_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23936_end_0 = const()[name = string("op_23936_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_23936_end_mask_0 = const()[name = string("op_23936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23936_cast_fp16 = slice_by_index(begin = var_23936_begin_0, end = var_23936_end_0, end_mask = var_23936_end_mask_0, x = value_31_cast_fp16)[name = string("op_23936_cast_fp16")]; + tensor var_23940_begin_0 = const()[name = string("op_23940_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23940_end_0 = const()[name = string("op_23940_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_23940_end_mask_0 = const()[name = string("op_23940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23940_cast_fp16 = slice_by_index(begin = var_23940_begin_0, end = var_23940_end_0, end_mask = var_23940_end_mask_0, x = value_31_cast_fp16)[name = string("op_23940_cast_fp16")]; + tensor var_23944_begin_0 = const()[name = string("op_23944_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23944_end_0 = const()[name = string("op_23944_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_23944_end_mask_0 = const()[name = string("op_23944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23944_cast_fp16 = slice_by_index(begin = var_23944_begin_0, end = var_23944_end_0, end_mask = var_23944_end_mask_0, x = value_31_cast_fp16)[name = string("op_23944_cast_fp16")]; + tensor var_23948_begin_0 = const()[name = string("op_23948_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23948_end_0 = const()[name = string("op_23948_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_23948_end_mask_0 = const()[name = string("op_23948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23948_cast_fp16 = slice_by_index(begin = var_23948_begin_0, end = var_23948_end_0, end_mask = var_23948_end_mask_0, x = value_31_cast_fp16)[name = string("op_23948_cast_fp16")]; + tensor var_23952_begin_0 = const()[name = string("op_23952_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23952_end_0 = const()[name = string("op_23952_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_23952_end_mask_0 = const()[name = string("op_23952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23952_cast_fp16 = slice_by_index(begin = var_23952_begin_0, end = var_23952_end_0, end_mask = var_23952_end_mask_0, x = value_31_cast_fp16)[name = string("op_23952_cast_fp16")]; + tensor var_23956_begin_0 = const()[name = string("op_23956_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23956_end_0 = const()[name = string("op_23956_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_23956_end_mask_0 = const()[name = string("op_23956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23956_cast_fp16 = slice_by_index(begin = var_23956_begin_0, end = var_23956_end_0, end_mask = var_23956_end_mask_0, x = value_31_cast_fp16)[name = string("op_23956_cast_fp16")]; + tensor var_23960_begin_0 = const()[name = string("op_23960_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23960_end_0 = const()[name = string("op_23960_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_23960_end_mask_0 = const()[name = string("op_23960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23960_cast_fp16 = slice_by_index(begin = var_23960_begin_0, end = var_23960_end_0, end_mask = var_23960_end_mask_0, x = value_31_cast_fp16)[name = string("op_23960_cast_fp16")]; + string _SplitHeadsQ__mh_w_2401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2401_equation_0, values = (var_23806_cast_fp16, var_23248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2401_cast_fp16")]; + string _SplitHeadsQ__mh_w_2403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2403_equation_0, values = (var_23806_cast_fp16, var_23255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2403_cast_fp16")]; + string _SplitHeadsQ__mh_w_2405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2405_equation_0, values = (var_23806_cast_fp16, var_23262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2405_cast_fp16")]; + string _SplitHeadsQ__mh_w_2407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2407_equation_0, values = (var_23806_cast_fp16, var_23269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2407_cast_fp16")]; + string _SplitHeadsQ__mh_w_2409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2409_equation_0, values = (var_23810_cast_fp16, var_23276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2409_cast_fp16")]; + string _SplitHeadsQ__mh_w_2411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2411_equation_0, values = (var_23810_cast_fp16, var_23283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2411_cast_fp16")]; + string _SplitHeadsQ__mh_w_2413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2413_equation_0, values = (var_23810_cast_fp16, var_23290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2413_cast_fp16")]; + string _SplitHeadsQ__mh_w_2415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2415_equation_0, values = (var_23810_cast_fp16, var_23297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2415_cast_fp16")]; + string _SplitHeadsQ__mh_w_2417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2417_equation_0, values = (var_23814_cast_fp16, var_23304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2417_cast_fp16")]; + string _SplitHeadsQ__mh_w_2419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2419_equation_0, values = (var_23814_cast_fp16, var_23311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2419_cast_fp16")]; + string _SplitHeadsQ__mh_w_2421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2421_equation_0, values = (var_23814_cast_fp16, var_23318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2421_cast_fp16")]; + string _SplitHeadsQ__mh_w_2423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2423_equation_0, values = (var_23814_cast_fp16, var_23325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2423_cast_fp16")]; + string _SplitHeadsQ__mh_w_2425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2425_equation_0, values = (var_23818_cast_fp16, var_23332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2425_cast_fp16")]; + string _SplitHeadsQ__mh_w_2427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2427_equation_0, values = (var_23818_cast_fp16, var_23339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2427_cast_fp16")]; + string _SplitHeadsQ__mh_w_2429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2429_equation_0, values = (var_23818_cast_fp16, var_23346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2429_cast_fp16")]; + string _SplitHeadsQ__mh_w_2431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2431_equation_0, values = (var_23818_cast_fp16, var_23353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2431_cast_fp16")]; + string _SplitHeadsQ__mh_w_2433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2433_equation_0, values = (var_23822_cast_fp16, var_23360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2433_cast_fp16")]; + string _SplitHeadsQ__mh_w_2435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2435_equation_0, values = (var_23822_cast_fp16, var_23367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2435_cast_fp16")]; + string _SplitHeadsQ__mh_w_2437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2437_equation_0, values = (var_23822_cast_fp16, var_23374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2437_cast_fp16")]; + string _SplitHeadsQ__mh_w_2439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2439_equation_0, values = (var_23822_cast_fp16, var_23381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2439_cast_fp16")]; + string _SplitHeadsQ__mh_w_2441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2441_equation_0, values = (var_23826_cast_fp16, var_23388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2441_cast_fp16")]; + string _SplitHeadsQ__mh_w_2443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2443_equation_0, values = (var_23826_cast_fp16, var_23395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2443_cast_fp16")]; + string _SplitHeadsQ__mh_w_2445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2445_equation_0, values = (var_23826_cast_fp16, var_23402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2445_cast_fp16")]; + string _SplitHeadsQ__mh_w_2447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2447_equation_0, values = (var_23826_cast_fp16, var_23409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2447_cast_fp16")]; + string _SplitHeadsQ__mh_w_2449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2449_equation_0, values = (var_23830_cast_fp16, var_23416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2449_cast_fp16")]; + string _SplitHeadsQ__mh_w_2451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2451_equation_0, values = (var_23830_cast_fp16, var_23423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2451_cast_fp16")]; + string _SplitHeadsQ__mh_w_2453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2453_equation_0, values = (var_23830_cast_fp16, var_23430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2453_cast_fp16")]; + string _SplitHeadsQ__mh_w_2455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2455_equation_0, values = (var_23830_cast_fp16, var_23437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2455_cast_fp16")]; + string _SplitHeadsQ__mh_w_2457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2457_equation_0, values = (var_23834_cast_fp16, var_23444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2457_cast_fp16")]; + string _SplitHeadsQ__mh_w_2459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2459_equation_0, values = (var_23834_cast_fp16, var_23451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2459_cast_fp16")]; + string _SplitHeadsQ__mh_w_2461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2461_equation_0, values = (var_23834_cast_fp16, var_23458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2461_cast_fp16")]; + string _SplitHeadsQ__mh_w_2463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2463_equation_0, values = (var_23834_cast_fp16, var_23465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2463_cast_fp16")]; + string _SplitHeadsQ__mh_w_2465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2465_equation_0, values = (var_23838_cast_fp16, var_23472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2465_cast_fp16")]; + string _SplitHeadsQ__mh_w_2467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2467_equation_0, values = (var_23838_cast_fp16, var_23479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2467_cast_fp16")]; + string _SplitHeadsQ__mh_w_2469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2469_equation_0, values = (var_23838_cast_fp16, var_23486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2469_cast_fp16")]; + string _SplitHeadsQ__mh_w_2471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2471_equation_0, values = (var_23838_cast_fp16, var_23493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2471_cast_fp16")]; + string _SplitHeadsQ__mh_w_2473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2473_equation_0, values = (var_23842_cast_fp16, var_23500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2473_cast_fp16")]; + string _SplitHeadsQ__mh_w_2475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2475_equation_0, values = (var_23842_cast_fp16, var_23507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2475_cast_fp16")]; + string _SplitHeadsQ__mh_w_2477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2477_equation_0, values = (var_23842_cast_fp16, var_23514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2477_cast_fp16")]; + string _SplitHeadsQ__mh_w_2479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2479_equation_0, values = (var_23842_cast_fp16, var_23521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2479_cast_fp16")]; + string _SplitHeadsQ__mh_w_2481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2481_equation_0, values = (var_23846_cast_fp16, var_23528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2481_cast_fp16")]; + string _SplitHeadsQ__mh_w_2483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2483_equation_0, values = (var_23846_cast_fp16, var_23535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2483_cast_fp16")]; + string _SplitHeadsQ__mh_w_2485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2485_equation_0, values = (var_23846_cast_fp16, var_23542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2485_cast_fp16")]; + string _SplitHeadsQ__mh_w_2487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2487_equation_0, values = (var_23846_cast_fp16, var_23549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2487_cast_fp16")]; + string _SplitHeadsQ__mh_w_2489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2489_equation_0, values = (var_23850_cast_fp16, var_23556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2489_cast_fp16")]; + string _SplitHeadsQ__mh_w_2491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2491_equation_0, values = (var_23850_cast_fp16, var_23563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2491_cast_fp16")]; + string _SplitHeadsQ__mh_w_2493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2493_equation_0, values = (var_23850_cast_fp16, var_23570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2493_cast_fp16")]; + string _SplitHeadsQ__mh_w_2495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2495_equation_0, values = (var_23850_cast_fp16, var_23577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2495_cast_fp16")]; + string _SplitHeadsQ__mh_w_2497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2497_equation_0, values = (var_23854_cast_fp16, var_23584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2497_cast_fp16")]; + string _SplitHeadsQ__mh_w_2499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2499_equation_0, values = (var_23854_cast_fp16, var_23591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2499_cast_fp16")]; + string _SplitHeadsQ__mh_w_2501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2501_equation_0, values = (var_23854_cast_fp16, var_23598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2501_cast_fp16")]; + string _SplitHeadsQ__mh_w_2503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2503_equation_0, values = (var_23854_cast_fp16, var_23605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2503_cast_fp16")]; + string _SplitHeadsQ__mh_w_2505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2505_equation_0, values = (var_23858_cast_fp16, var_23612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2505_cast_fp16")]; + string _SplitHeadsQ__mh_w_2507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2507_equation_0, values = (var_23858_cast_fp16, var_23619_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2507_cast_fp16")]; + string _SplitHeadsQ__mh_w_2509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2509_equation_0, values = (var_23858_cast_fp16, var_23626_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2509_cast_fp16")]; + string _SplitHeadsQ__mh_w_2511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2511_equation_0, values = (var_23858_cast_fp16, var_23633_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2511_cast_fp16")]; + string _SplitHeadsQ__mh_w_2513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2513_equation_0, values = (var_23862_cast_fp16, var_23640_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2513_cast_fp16")]; + string _SplitHeadsQ__mh_w_2515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2515_equation_0, values = (var_23862_cast_fp16, var_23647_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2515_cast_fp16")]; + string _SplitHeadsQ__mh_w_2517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2517_equation_0, values = (var_23862_cast_fp16, var_23654_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2517_cast_fp16")]; + string _SplitHeadsQ__mh_w_2519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2519_equation_0, values = (var_23862_cast_fp16, var_23661_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2519_cast_fp16")]; + string _SplitHeadsQ__mh_w_2521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2521_equation_0, values = (var_23866_cast_fp16, var_23668_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2521_cast_fp16")]; + string _SplitHeadsQ__mh_w_2523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2523_equation_0, values = (var_23866_cast_fp16, var_23675_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2523_cast_fp16")]; + string _SplitHeadsQ__mh_w_2525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2525_equation_0, values = (var_23866_cast_fp16, var_23682_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2525_cast_fp16")]; + string _SplitHeadsQ__mh_w_2527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2527_equation_0, values = (var_23866_cast_fp16, var_23689_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2527_cast_fp16")]; + string _SplitHeadsQ__mh_w_2529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2529_equation_0, values = (var_23870_cast_fp16, var_23696_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2529_cast_fp16")]; + string _SplitHeadsQ__mh_w_2531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2531_equation_0, values = (var_23870_cast_fp16, var_23703_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2531_cast_fp16")]; + string _SplitHeadsQ__mh_w_2533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2533_equation_0, values = (var_23870_cast_fp16, var_23710_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2533_cast_fp16")]; + string _SplitHeadsQ__mh_w_2535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2535_equation_0, values = (var_23870_cast_fp16, var_23717_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2535_cast_fp16")]; + string _SplitHeadsQ__mh_w_2537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2537_equation_0, values = (var_23874_cast_fp16, var_23724_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2537_cast_fp16")]; + string _SplitHeadsQ__mh_w_2539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2539_equation_0, values = (var_23874_cast_fp16, var_23731_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2539_cast_fp16")]; + string _SplitHeadsQ__mh_w_2541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2541_equation_0, values = (var_23874_cast_fp16, var_23738_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2541_cast_fp16")]; + string _SplitHeadsQ__mh_w_2543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2543_equation_0, values = (var_23874_cast_fp16, var_23745_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2543_cast_fp16")]; + string _SplitHeadsQ__mh_w_2545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2545_equation_0, values = (var_23878_cast_fp16, var_23752_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2545_cast_fp16")]; + string _SplitHeadsQ__mh_w_2547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2547_equation_0, values = (var_23878_cast_fp16, var_23759_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2547_cast_fp16")]; + string _SplitHeadsQ__mh_w_2549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2549_equation_0, values = (var_23878_cast_fp16, var_23766_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2549_cast_fp16")]; + string _SplitHeadsQ__mh_w_2551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2551_equation_0, values = (var_23878_cast_fp16, var_23773_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2551_cast_fp16")]; + string _SplitHeadsQ__mh_w_2553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2553_equation_0, values = (var_23882_cast_fp16, var_23780_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2553_cast_fp16")]; + string _SplitHeadsQ__mh_w_2555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2555_equation_0, values = (var_23882_cast_fp16, var_23787_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2555_cast_fp16")]; + string _SplitHeadsQ__mh_w_2557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2557_equation_0, values = (var_23882_cast_fp16, var_23794_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2557_cast_fp16")]; + string _SplitHeadsQ__mh_w_2559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2559_equation_0, values = (var_23882_cast_fp16, var_23801_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2559_cast_fp16")]; + fp16 var_24123_to_fp16 = const()[name = string("op_24123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2401_cast_fp16, y = var_24123_to_fp16)[name = string("aw_chunk_2401_cast_fp16")]; + fp16 var_24125_to_fp16 = const()[name = string("op_24125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2403_cast_fp16, y = var_24125_to_fp16)[name = string("aw_chunk_2403_cast_fp16")]; + fp16 var_24127_to_fp16 = const()[name = string("op_24127_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2405_cast_fp16, y = var_24127_to_fp16)[name = string("aw_chunk_2405_cast_fp16")]; + fp16 var_24129_to_fp16 = const()[name = string("op_24129_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2407_cast_fp16, y = var_24129_to_fp16)[name = string("aw_chunk_2407_cast_fp16")]; + fp16 var_24131_to_fp16 = const()[name = string("op_24131_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2409_cast_fp16, y = var_24131_to_fp16)[name = string("aw_chunk_2409_cast_fp16")]; + fp16 var_24133_to_fp16 = const()[name = string("op_24133_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2411_cast_fp16, y = var_24133_to_fp16)[name = string("aw_chunk_2411_cast_fp16")]; + fp16 var_24135_to_fp16 = const()[name = string("op_24135_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2413_cast_fp16, y = var_24135_to_fp16)[name = string("aw_chunk_2413_cast_fp16")]; + fp16 var_24137_to_fp16 = const()[name = string("op_24137_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2415_cast_fp16, y = var_24137_to_fp16)[name = string("aw_chunk_2415_cast_fp16")]; + fp16 var_24139_to_fp16 = const()[name = string("op_24139_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2417_cast_fp16, y = var_24139_to_fp16)[name = string("aw_chunk_2417_cast_fp16")]; + fp16 var_24141_to_fp16 = const()[name = string("op_24141_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2419_cast_fp16, y = var_24141_to_fp16)[name = string("aw_chunk_2419_cast_fp16")]; + fp16 var_24143_to_fp16 = const()[name = string("op_24143_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2421_cast_fp16, y = var_24143_to_fp16)[name = string("aw_chunk_2421_cast_fp16")]; + fp16 var_24145_to_fp16 = const()[name = string("op_24145_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2423_cast_fp16, y = var_24145_to_fp16)[name = string("aw_chunk_2423_cast_fp16")]; + fp16 var_24147_to_fp16 = const()[name = string("op_24147_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2425_cast_fp16, y = var_24147_to_fp16)[name = string("aw_chunk_2425_cast_fp16")]; + fp16 var_24149_to_fp16 = const()[name = string("op_24149_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2427_cast_fp16, y = var_24149_to_fp16)[name = string("aw_chunk_2427_cast_fp16")]; + fp16 var_24151_to_fp16 = const()[name = string("op_24151_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2429_cast_fp16, y = var_24151_to_fp16)[name = string("aw_chunk_2429_cast_fp16")]; + fp16 var_24153_to_fp16 = const()[name = string("op_24153_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2431_cast_fp16, y = var_24153_to_fp16)[name = string("aw_chunk_2431_cast_fp16")]; + fp16 var_24155_to_fp16 = const()[name = string("op_24155_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2433_cast_fp16, y = var_24155_to_fp16)[name = string("aw_chunk_2433_cast_fp16")]; + fp16 var_24157_to_fp16 = const()[name = string("op_24157_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2435_cast_fp16, y = var_24157_to_fp16)[name = string("aw_chunk_2435_cast_fp16")]; + fp16 var_24159_to_fp16 = const()[name = string("op_24159_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2437_cast_fp16, y = var_24159_to_fp16)[name = string("aw_chunk_2437_cast_fp16")]; + fp16 var_24161_to_fp16 = const()[name = string("op_24161_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2439_cast_fp16, y = var_24161_to_fp16)[name = string("aw_chunk_2439_cast_fp16")]; + fp16 var_24163_to_fp16 = const()[name = string("op_24163_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2441_cast_fp16, y = var_24163_to_fp16)[name = string("aw_chunk_2441_cast_fp16")]; + fp16 var_24165_to_fp16 = const()[name = string("op_24165_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2443_cast_fp16, y = var_24165_to_fp16)[name = string("aw_chunk_2443_cast_fp16")]; + fp16 var_24167_to_fp16 = const()[name = string("op_24167_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2445_cast_fp16, y = var_24167_to_fp16)[name = string("aw_chunk_2445_cast_fp16")]; + fp16 var_24169_to_fp16 = const()[name = string("op_24169_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2447_cast_fp16, y = var_24169_to_fp16)[name = string("aw_chunk_2447_cast_fp16")]; + fp16 var_24171_to_fp16 = const()[name = string("op_24171_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2449_cast_fp16, y = var_24171_to_fp16)[name = string("aw_chunk_2449_cast_fp16")]; + fp16 var_24173_to_fp16 = const()[name = string("op_24173_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2451_cast_fp16, y = var_24173_to_fp16)[name = string("aw_chunk_2451_cast_fp16")]; + fp16 var_24175_to_fp16 = const()[name = string("op_24175_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2453_cast_fp16, y = var_24175_to_fp16)[name = string("aw_chunk_2453_cast_fp16")]; + fp16 var_24177_to_fp16 = const()[name = string("op_24177_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2455_cast_fp16, y = var_24177_to_fp16)[name = string("aw_chunk_2455_cast_fp16")]; + fp16 var_24179_to_fp16 = const()[name = string("op_24179_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2457_cast_fp16, y = var_24179_to_fp16)[name = string("aw_chunk_2457_cast_fp16")]; + fp16 var_24181_to_fp16 = const()[name = string("op_24181_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2459_cast_fp16, y = var_24181_to_fp16)[name = string("aw_chunk_2459_cast_fp16")]; + fp16 var_24183_to_fp16 = const()[name = string("op_24183_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2461_cast_fp16, y = var_24183_to_fp16)[name = string("aw_chunk_2461_cast_fp16")]; + fp16 var_24185_to_fp16 = const()[name = string("op_24185_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2463_cast_fp16, y = var_24185_to_fp16)[name = string("aw_chunk_2463_cast_fp16")]; + fp16 var_24187_to_fp16 = const()[name = string("op_24187_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2465_cast_fp16, y = var_24187_to_fp16)[name = string("aw_chunk_2465_cast_fp16")]; + fp16 var_24189_to_fp16 = const()[name = string("op_24189_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2467_cast_fp16, y = var_24189_to_fp16)[name = string("aw_chunk_2467_cast_fp16")]; + fp16 var_24191_to_fp16 = const()[name = string("op_24191_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2469_cast_fp16, y = var_24191_to_fp16)[name = string("aw_chunk_2469_cast_fp16")]; + fp16 var_24193_to_fp16 = const()[name = string("op_24193_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2471_cast_fp16, y = var_24193_to_fp16)[name = string("aw_chunk_2471_cast_fp16")]; + fp16 var_24195_to_fp16 = const()[name = string("op_24195_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2473_cast_fp16, y = var_24195_to_fp16)[name = string("aw_chunk_2473_cast_fp16")]; + fp16 var_24197_to_fp16 = const()[name = string("op_24197_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2475_cast_fp16, y = var_24197_to_fp16)[name = string("aw_chunk_2475_cast_fp16")]; + fp16 var_24199_to_fp16 = const()[name = string("op_24199_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2477_cast_fp16, y = var_24199_to_fp16)[name = string("aw_chunk_2477_cast_fp16")]; + fp16 var_24201_to_fp16 = const()[name = string("op_24201_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2479_cast_fp16, y = var_24201_to_fp16)[name = string("aw_chunk_2479_cast_fp16")]; + fp16 var_24203_to_fp16 = const()[name = string("op_24203_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2481_cast_fp16, y = var_24203_to_fp16)[name = string("aw_chunk_2481_cast_fp16")]; + fp16 var_24205_to_fp16 = const()[name = string("op_24205_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2483_cast_fp16, y = var_24205_to_fp16)[name = string("aw_chunk_2483_cast_fp16")]; + fp16 var_24207_to_fp16 = const()[name = string("op_24207_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2485_cast_fp16, y = var_24207_to_fp16)[name = string("aw_chunk_2485_cast_fp16")]; + fp16 var_24209_to_fp16 = const()[name = string("op_24209_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2487_cast_fp16, y = var_24209_to_fp16)[name = string("aw_chunk_2487_cast_fp16")]; + fp16 var_24211_to_fp16 = const()[name = string("op_24211_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2489_cast_fp16, y = var_24211_to_fp16)[name = string("aw_chunk_2489_cast_fp16")]; + fp16 var_24213_to_fp16 = const()[name = string("op_24213_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2491_cast_fp16, y = var_24213_to_fp16)[name = string("aw_chunk_2491_cast_fp16")]; + fp16 var_24215_to_fp16 = const()[name = string("op_24215_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2493_cast_fp16, y = var_24215_to_fp16)[name = string("aw_chunk_2493_cast_fp16")]; + fp16 var_24217_to_fp16 = const()[name = string("op_24217_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2495_cast_fp16, y = var_24217_to_fp16)[name = string("aw_chunk_2495_cast_fp16")]; + fp16 var_24219_to_fp16 = const()[name = string("op_24219_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2497_cast_fp16, y = var_24219_to_fp16)[name = string("aw_chunk_2497_cast_fp16")]; + fp16 var_24221_to_fp16 = const()[name = string("op_24221_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2499_cast_fp16, y = var_24221_to_fp16)[name = string("aw_chunk_2499_cast_fp16")]; + fp16 var_24223_to_fp16 = const()[name = string("op_24223_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2501_cast_fp16, y = var_24223_to_fp16)[name = string("aw_chunk_2501_cast_fp16")]; + fp16 var_24225_to_fp16 = const()[name = string("op_24225_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2503_cast_fp16, y = var_24225_to_fp16)[name = string("aw_chunk_2503_cast_fp16")]; + fp16 var_24227_to_fp16 = const()[name = string("op_24227_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2505_cast_fp16, y = var_24227_to_fp16)[name = string("aw_chunk_2505_cast_fp16")]; + fp16 var_24229_to_fp16 = const()[name = string("op_24229_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2507_cast_fp16, y = var_24229_to_fp16)[name = string("aw_chunk_2507_cast_fp16")]; + fp16 var_24231_to_fp16 = const()[name = string("op_24231_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2509_cast_fp16, y = var_24231_to_fp16)[name = string("aw_chunk_2509_cast_fp16")]; + fp16 var_24233_to_fp16 = const()[name = string("op_24233_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2511_cast_fp16, y = var_24233_to_fp16)[name = string("aw_chunk_2511_cast_fp16")]; + fp16 var_24235_to_fp16 = const()[name = string("op_24235_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2513_cast_fp16, y = var_24235_to_fp16)[name = string("aw_chunk_2513_cast_fp16")]; + fp16 var_24237_to_fp16 = const()[name = string("op_24237_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2515_cast_fp16, y = var_24237_to_fp16)[name = string("aw_chunk_2515_cast_fp16")]; + fp16 var_24239_to_fp16 = const()[name = string("op_24239_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2517_cast_fp16, y = var_24239_to_fp16)[name = string("aw_chunk_2517_cast_fp16")]; + fp16 var_24241_to_fp16 = const()[name = string("op_24241_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2519_cast_fp16, y = var_24241_to_fp16)[name = string("aw_chunk_2519_cast_fp16")]; + fp16 var_24243_to_fp16 = const()[name = string("op_24243_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2521_cast_fp16, y = var_24243_to_fp16)[name = string("aw_chunk_2521_cast_fp16")]; + fp16 var_24245_to_fp16 = const()[name = string("op_24245_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2523_cast_fp16, y = var_24245_to_fp16)[name = string("aw_chunk_2523_cast_fp16")]; + fp16 var_24247_to_fp16 = const()[name = string("op_24247_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2525_cast_fp16, y = var_24247_to_fp16)[name = string("aw_chunk_2525_cast_fp16")]; + fp16 var_24249_to_fp16 = const()[name = string("op_24249_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2527_cast_fp16, y = var_24249_to_fp16)[name = string("aw_chunk_2527_cast_fp16")]; + fp16 var_24251_to_fp16 = const()[name = string("op_24251_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2529_cast_fp16, y = var_24251_to_fp16)[name = string("aw_chunk_2529_cast_fp16")]; + fp16 var_24253_to_fp16 = const()[name = string("op_24253_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2531_cast_fp16, y = var_24253_to_fp16)[name = string("aw_chunk_2531_cast_fp16")]; + fp16 var_24255_to_fp16 = const()[name = string("op_24255_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2533_cast_fp16, y = var_24255_to_fp16)[name = string("aw_chunk_2533_cast_fp16")]; + fp16 var_24257_to_fp16 = const()[name = string("op_24257_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2535_cast_fp16, y = var_24257_to_fp16)[name = string("aw_chunk_2535_cast_fp16")]; + fp16 var_24259_to_fp16 = const()[name = string("op_24259_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2537_cast_fp16, y = var_24259_to_fp16)[name = string("aw_chunk_2537_cast_fp16")]; + fp16 var_24261_to_fp16 = const()[name = string("op_24261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2539_cast_fp16, y = var_24261_to_fp16)[name = string("aw_chunk_2539_cast_fp16")]; + fp16 var_24263_to_fp16 = const()[name = string("op_24263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2541_cast_fp16, y = var_24263_to_fp16)[name = string("aw_chunk_2541_cast_fp16")]; + fp16 var_24265_to_fp16 = const()[name = string("op_24265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2543_cast_fp16, y = var_24265_to_fp16)[name = string("aw_chunk_2543_cast_fp16")]; + fp16 var_24267_to_fp16 = const()[name = string("op_24267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2545_cast_fp16, y = var_24267_to_fp16)[name = string("aw_chunk_2545_cast_fp16")]; + fp16 var_24269_to_fp16 = const()[name = string("op_24269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2547_cast_fp16, y = var_24269_to_fp16)[name = string("aw_chunk_2547_cast_fp16")]; + fp16 var_24271_to_fp16 = const()[name = string("op_24271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2549_cast_fp16, y = var_24271_to_fp16)[name = string("aw_chunk_2549_cast_fp16")]; + fp16 var_24273_to_fp16 = const()[name = string("op_24273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2551_cast_fp16, y = var_24273_to_fp16)[name = string("aw_chunk_2551_cast_fp16")]; + fp16 var_24275_to_fp16 = const()[name = string("op_24275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2553_cast_fp16, y = var_24275_to_fp16)[name = string("aw_chunk_2553_cast_fp16")]; + fp16 var_24277_to_fp16 = const()[name = string("op_24277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2555_cast_fp16, y = var_24277_to_fp16)[name = string("aw_chunk_2555_cast_fp16")]; + fp16 var_24279_to_fp16 = const()[name = string("op_24279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2557_cast_fp16, y = var_24279_to_fp16)[name = string("aw_chunk_2557_cast_fp16")]; + fp16 var_24281_to_fp16 = const()[name = string("op_24281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2559_cast_fp16, y = var_24281_to_fp16)[name = string("aw_chunk_2559_cast_fp16")]; + tensor var_24283_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2401_cast_fp16)[name = string("op_24283_cast_fp16")]; + tensor var_24284_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2403_cast_fp16)[name = string("op_24284_cast_fp16")]; + tensor var_24285_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2405_cast_fp16)[name = string("op_24285_cast_fp16")]; + tensor var_24286_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2407_cast_fp16)[name = string("op_24286_cast_fp16")]; + tensor var_24287_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2409_cast_fp16)[name = string("op_24287_cast_fp16")]; + tensor var_24288_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2411_cast_fp16)[name = string("op_24288_cast_fp16")]; + tensor var_24289_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2413_cast_fp16)[name = string("op_24289_cast_fp16")]; + tensor var_24290_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2415_cast_fp16)[name = string("op_24290_cast_fp16")]; + tensor var_24291_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2417_cast_fp16)[name = string("op_24291_cast_fp16")]; + tensor var_24292_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2419_cast_fp16)[name = string("op_24292_cast_fp16")]; + tensor var_24293_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2421_cast_fp16)[name = string("op_24293_cast_fp16")]; + tensor var_24294_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2423_cast_fp16)[name = string("op_24294_cast_fp16")]; + tensor var_24295_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2425_cast_fp16)[name = string("op_24295_cast_fp16")]; + tensor var_24296_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2427_cast_fp16)[name = string("op_24296_cast_fp16")]; + tensor var_24297_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2429_cast_fp16)[name = string("op_24297_cast_fp16")]; + tensor var_24298_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2431_cast_fp16)[name = string("op_24298_cast_fp16")]; + tensor var_24299_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2433_cast_fp16)[name = string("op_24299_cast_fp16")]; + tensor var_24300_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2435_cast_fp16)[name = string("op_24300_cast_fp16")]; + tensor var_24301_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2437_cast_fp16)[name = string("op_24301_cast_fp16")]; + tensor var_24302_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2439_cast_fp16)[name = string("op_24302_cast_fp16")]; + tensor var_24303_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2441_cast_fp16)[name = string("op_24303_cast_fp16")]; + tensor var_24304_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2443_cast_fp16)[name = string("op_24304_cast_fp16")]; + tensor var_24305_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2445_cast_fp16)[name = string("op_24305_cast_fp16")]; + tensor var_24306_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2447_cast_fp16)[name = string("op_24306_cast_fp16")]; + tensor var_24307_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2449_cast_fp16)[name = string("op_24307_cast_fp16")]; + tensor var_24308_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2451_cast_fp16)[name = string("op_24308_cast_fp16")]; + tensor var_24309_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2453_cast_fp16)[name = string("op_24309_cast_fp16")]; + tensor var_24310_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2455_cast_fp16)[name = string("op_24310_cast_fp16")]; + tensor var_24311_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2457_cast_fp16)[name = string("op_24311_cast_fp16")]; + tensor var_24312_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2459_cast_fp16)[name = string("op_24312_cast_fp16")]; + tensor var_24313_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2461_cast_fp16)[name = string("op_24313_cast_fp16")]; + tensor var_24314_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2463_cast_fp16)[name = string("op_24314_cast_fp16")]; + tensor var_24315_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2465_cast_fp16)[name = string("op_24315_cast_fp16")]; + tensor var_24316_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2467_cast_fp16)[name = string("op_24316_cast_fp16")]; + tensor var_24317_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2469_cast_fp16)[name = string("op_24317_cast_fp16")]; + tensor var_24318_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2471_cast_fp16)[name = string("op_24318_cast_fp16")]; + tensor var_24319_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2473_cast_fp16)[name = string("op_24319_cast_fp16")]; + tensor var_24320_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2475_cast_fp16)[name = string("op_24320_cast_fp16")]; + tensor var_24321_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2477_cast_fp16)[name = string("op_24321_cast_fp16")]; + tensor var_24322_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2479_cast_fp16)[name = string("op_24322_cast_fp16")]; + tensor var_24323_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2481_cast_fp16)[name = string("op_24323_cast_fp16")]; + tensor var_24324_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2483_cast_fp16)[name = string("op_24324_cast_fp16")]; + tensor var_24325_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2485_cast_fp16)[name = string("op_24325_cast_fp16")]; + tensor var_24326_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2487_cast_fp16)[name = string("op_24326_cast_fp16")]; + tensor var_24327_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2489_cast_fp16)[name = string("op_24327_cast_fp16")]; + tensor var_24328_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2491_cast_fp16)[name = string("op_24328_cast_fp16")]; + tensor var_24329_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2493_cast_fp16)[name = string("op_24329_cast_fp16")]; + tensor var_24330_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2495_cast_fp16)[name = string("op_24330_cast_fp16")]; + tensor var_24331_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2497_cast_fp16)[name = string("op_24331_cast_fp16")]; + tensor var_24332_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2499_cast_fp16)[name = string("op_24332_cast_fp16")]; + tensor var_24333_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2501_cast_fp16)[name = string("op_24333_cast_fp16")]; + tensor var_24334_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2503_cast_fp16)[name = string("op_24334_cast_fp16")]; + tensor var_24335_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2505_cast_fp16)[name = string("op_24335_cast_fp16")]; + tensor var_24336_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2507_cast_fp16)[name = string("op_24336_cast_fp16")]; + tensor var_24337_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2509_cast_fp16)[name = string("op_24337_cast_fp16")]; + tensor var_24338_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2511_cast_fp16)[name = string("op_24338_cast_fp16")]; + tensor var_24339_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2513_cast_fp16)[name = string("op_24339_cast_fp16")]; + tensor var_24340_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2515_cast_fp16)[name = string("op_24340_cast_fp16")]; + tensor var_24341_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2517_cast_fp16)[name = string("op_24341_cast_fp16")]; + tensor var_24342_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2519_cast_fp16)[name = string("op_24342_cast_fp16")]; + tensor var_24343_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2521_cast_fp16)[name = string("op_24343_cast_fp16")]; + tensor var_24344_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2523_cast_fp16)[name = string("op_24344_cast_fp16")]; + tensor var_24345_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2525_cast_fp16)[name = string("op_24345_cast_fp16")]; + tensor var_24346_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2527_cast_fp16)[name = string("op_24346_cast_fp16")]; + tensor var_24347_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2529_cast_fp16)[name = string("op_24347_cast_fp16")]; + tensor var_24348_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2531_cast_fp16)[name = string("op_24348_cast_fp16")]; + tensor var_24349_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2533_cast_fp16)[name = string("op_24349_cast_fp16")]; + tensor var_24350_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2535_cast_fp16)[name = string("op_24350_cast_fp16")]; + tensor var_24351_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2537_cast_fp16)[name = string("op_24351_cast_fp16")]; + tensor var_24352_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2539_cast_fp16)[name = string("op_24352_cast_fp16")]; + tensor var_24353_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2541_cast_fp16)[name = string("op_24353_cast_fp16")]; + tensor var_24354_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2543_cast_fp16)[name = string("op_24354_cast_fp16")]; + tensor var_24355_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2545_cast_fp16)[name = string("op_24355_cast_fp16")]; + tensor var_24356_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2547_cast_fp16)[name = string("op_24356_cast_fp16")]; + tensor var_24357_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2549_cast_fp16)[name = string("op_24357_cast_fp16")]; + tensor var_24358_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2551_cast_fp16)[name = string("op_24358_cast_fp16")]; + tensor var_24359_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2553_cast_fp16)[name = string("op_24359_cast_fp16")]; + tensor var_24360_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2555_cast_fp16)[name = string("op_24360_cast_fp16")]; + tensor var_24361_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2557_cast_fp16)[name = string("op_24361_cast_fp16")]; + tensor var_24362_cast_fp16 = softmax(axis = var_23108, x = aw_chunk_2559_cast_fp16)[name = string("op_24362_cast_fp16")]; + string var_24364_equation_0 = const()[name = string("op_24364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24364_cast_fp16 = einsum(equation = var_24364_equation_0, values = (var_23884_cast_fp16, var_24283_cast_fp16))[name = string("op_24364_cast_fp16")]; + string var_24366_equation_0 = const()[name = string("op_24366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24366_cast_fp16 = einsum(equation = var_24366_equation_0, values = (var_23884_cast_fp16, var_24284_cast_fp16))[name = string("op_24366_cast_fp16")]; + string var_24368_equation_0 = const()[name = string("op_24368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24368_cast_fp16 = einsum(equation = var_24368_equation_0, values = (var_23884_cast_fp16, var_24285_cast_fp16))[name = string("op_24368_cast_fp16")]; + string var_24370_equation_0 = const()[name = string("op_24370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24370_cast_fp16 = einsum(equation = var_24370_equation_0, values = (var_23884_cast_fp16, var_24286_cast_fp16))[name = string("op_24370_cast_fp16")]; + string var_24372_equation_0 = const()[name = string("op_24372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24372_cast_fp16 = einsum(equation = var_24372_equation_0, values = (var_23888_cast_fp16, var_24287_cast_fp16))[name = string("op_24372_cast_fp16")]; + string var_24374_equation_0 = const()[name = string("op_24374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24374_cast_fp16 = einsum(equation = var_24374_equation_0, values = (var_23888_cast_fp16, var_24288_cast_fp16))[name = string("op_24374_cast_fp16")]; + string var_24376_equation_0 = const()[name = string("op_24376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24376_cast_fp16 = einsum(equation = var_24376_equation_0, values = (var_23888_cast_fp16, var_24289_cast_fp16))[name = string("op_24376_cast_fp16")]; + string var_24378_equation_0 = const()[name = string("op_24378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24378_cast_fp16 = einsum(equation = var_24378_equation_0, values = (var_23888_cast_fp16, var_24290_cast_fp16))[name = string("op_24378_cast_fp16")]; + string var_24380_equation_0 = const()[name = string("op_24380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24380_cast_fp16 = einsum(equation = var_24380_equation_0, values = (var_23892_cast_fp16, var_24291_cast_fp16))[name = string("op_24380_cast_fp16")]; + string var_24382_equation_0 = const()[name = string("op_24382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24382_cast_fp16 = einsum(equation = var_24382_equation_0, values = (var_23892_cast_fp16, var_24292_cast_fp16))[name = string("op_24382_cast_fp16")]; + string var_24384_equation_0 = const()[name = string("op_24384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24384_cast_fp16 = einsum(equation = var_24384_equation_0, values = (var_23892_cast_fp16, var_24293_cast_fp16))[name = string("op_24384_cast_fp16")]; + string var_24386_equation_0 = const()[name = string("op_24386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24386_cast_fp16 = einsum(equation = var_24386_equation_0, values = (var_23892_cast_fp16, var_24294_cast_fp16))[name = string("op_24386_cast_fp16")]; + string var_24388_equation_0 = const()[name = string("op_24388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24388_cast_fp16 = einsum(equation = var_24388_equation_0, values = (var_23896_cast_fp16, var_24295_cast_fp16))[name = string("op_24388_cast_fp16")]; + string var_24390_equation_0 = const()[name = string("op_24390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24390_cast_fp16 = einsum(equation = var_24390_equation_0, values = (var_23896_cast_fp16, var_24296_cast_fp16))[name = string("op_24390_cast_fp16")]; + string var_24392_equation_0 = const()[name = string("op_24392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24392_cast_fp16 = einsum(equation = var_24392_equation_0, values = (var_23896_cast_fp16, var_24297_cast_fp16))[name = string("op_24392_cast_fp16")]; + string var_24394_equation_0 = const()[name = string("op_24394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24394_cast_fp16 = einsum(equation = var_24394_equation_0, values = (var_23896_cast_fp16, var_24298_cast_fp16))[name = string("op_24394_cast_fp16")]; + string var_24396_equation_0 = const()[name = string("op_24396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24396_cast_fp16 = einsum(equation = var_24396_equation_0, values = (var_23900_cast_fp16, var_24299_cast_fp16))[name = string("op_24396_cast_fp16")]; + string var_24398_equation_0 = const()[name = string("op_24398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24398_cast_fp16 = einsum(equation = var_24398_equation_0, values = (var_23900_cast_fp16, var_24300_cast_fp16))[name = string("op_24398_cast_fp16")]; + string var_24400_equation_0 = const()[name = string("op_24400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24400_cast_fp16 = einsum(equation = var_24400_equation_0, values = (var_23900_cast_fp16, var_24301_cast_fp16))[name = string("op_24400_cast_fp16")]; + string var_24402_equation_0 = const()[name = string("op_24402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24402_cast_fp16 = einsum(equation = var_24402_equation_0, values = (var_23900_cast_fp16, var_24302_cast_fp16))[name = string("op_24402_cast_fp16")]; + string var_24404_equation_0 = const()[name = string("op_24404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24404_cast_fp16 = einsum(equation = var_24404_equation_0, values = (var_23904_cast_fp16, var_24303_cast_fp16))[name = string("op_24404_cast_fp16")]; + string var_24406_equation_0 = const()[name = string("op_24406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24406_cast_fp16 = einsum(equation = var_24406_equation_0, values = (var_23904_cast_fp16, var_24304_cast_fp16))[name = string("op_24406_cast_fp16")]; + string var_24408_equation_0 = const()[name = string("op_24408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24408_cast_fp16 = einsum(equation = var_24408_equation_0, values = (var_23904_cast_fp16, var_24305_cast_fp16))[name = string("op_24408_cast_fp16")]; + string var_24410_equation_0 = const()[name = string("op_24410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24410_cast_fp16 = einsum(equation = var_24410_equation_0, values = (var_23904_cast_fp16, var_24306_cast_fp16))[name = string("op_24410_cast_fp16")]; + string var_24412_equation_0 = const()[name = string("op_24412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24412_cast_fp16 = einsum(equation = var_24412_equation_0, values = (var_23908_cast_fp16, var_24307_cast_fp16))[name = string("op_24412_cast_fp16")]; + string var_24414_equation_0 = const()[name = string("op_24414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24414_cast_fp16 = einsum(equation = var_24414_equation_0, values = (var_23908_cast_fp16, var_24308_cast_fp16))[name = string("op_24414_cast_fp16")]; + string var_24416_equation_0 = const()[name = string("op_24416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24416_cast_fp16 = einsum(equation = var_24416_equation_0, values = (var_23908_cast_fp16, var_24309_cast_fp16))[name = string("op_24416_cast_fp16")]; + string var_24418_equation_0 = const()[name = string("op_24418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24418_cast_fp16 = einsum(equation = var_24418_equation_0, values = (var_23908_cast_fp16, var_24310_cast_fp16))[name = string("op_24418_cast_fp16")]; + string var_24420_equation_0 = const()[name = string("op_24420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24420_cast_fp16 = einsum(equation = var_24420_equation_0, values = (var_23912_cast_fp16, var_24311_cast_fp16))[name = string("op_24420_cast_fp16")]; + string var_24422_equation_0 = const()[name = string("op_24422_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24422_cast_fp16 = einsum(equation = var_24422_equation_0, values = (var_23912_cast_fp16, var_24312_cast_fp16))[name = string("op_24422_cast_fp16")]; + string var_24424_equation_0 = const()[name = string("op_24424_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24424_cast_fp16 = einsum(equation = var_24424_equation_0, values = (var_23912_cast_fp16, var_24313_cast_fp16))[name = string("op_24424_cast_fp16")]; + string var_24426_equation_0 = const()[name = string("op_24426_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24426_cast_fp16 = einsum(equation = var_24426_equation_0, values = (var_23912_cast_fp16, var_24314_cast_fp16))[name = string("op_24426_cast_fp16")]; + string var_24428_equation_0 = const()[name = string("op_24428_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24428_cast_fp16 = einsum(equation = var_24428_equation_0, values = (var_23916_cast_fp16, var_24315_cast_fp16))[name = string("op_24428_cast_fp16")]; + string var_24430_equation_0 = const()[name = string("op_24430_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24430_cast_fp16 = einsum(equation = var_24430_equation_0, values = (var_23916_cast_fp16, var_24316_cast_fp16))[name = string("op_24430_cast_fp16")]; + string var_24432_equation_0 = const()[name = string("op_24432_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24432_cast_fp16 = einsum(equation = var_24432_equation_0, values = (var_23916_cast_fp16, var_24317_cast_fp16))[name = string("op_24432_cast_fp16")]; + string var_24434_equation_0 = const()[name = string("op_24434_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24434_cast_fp16 = einsum(equation = var_24434_equation_0, values = (var_23916_cast_fp16, var_24318_cast_fp16))[name = string("op_24434_cast_fp16")]; + string var_24436_equation_0 = const()[name = string("op_24436_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24436_cast_fp16 = einsum(equation = var_24436_equation_0, values = (var_23920_cast_fp16, var_24319_cast_fp16))[name = string("op_24436_cast_fp16")]; + string var_24438_equation_0 = const()[name = string("op_24438_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24438_cast_fp16 = einsum(equation = var_24438_equation_0, values = (var_23920_cast_fp16, var_24320_cast_fp16))[name = string("op_24438_cast_fp16")]; + string var_24440_equation_0 = const()[name = string("op_24440_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24440_cast_fp16 = einsum(equation = var_24440_equation_0, values = (var_23920_cast_fp16, var_24321_cast_fp16))[name = string("op_24440_cast_fp16")]; + string var_24442_equation_0 = const()[name = string("op_24442_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24442_cast_fp16 = einsum(equation = var_24442_equation_0, values = (var_23920_cast_fp16, var_24322_cast_fp16))[name = string("op_24442_cast_fp16")]; + string var_24444_equation_0 = const()[name = string("op_24444_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24444_cast_fp16 = einsum(equation = var_24444_equation_0, values = (var_23924_cast_fp16, var_24323_cast_fp16))[name = string("op_24444_cast_fp16")]; + string var_24446_equation_0 = const()[name = string("op_24446_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24446_cast_fp16 = einsum(equation = var_24446_equation_0, values = (var_23924_cast_fp16, var_24324_cast_fp16))[name = string("op_24446_cast_fp16")]; + string var_24448_equation_0 = const()[name = string("op_24448_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24448_cast_fp16 = einsum(equation = var_24448_equation_0, values = (var_23924_cast_fp16, var_24325_cast_fp16))[name = string("op_24448_cast_fp16")]; + string var_24450_equation_0 = const()[name = string("op_24450_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24450_cast_fp16 = einsum(equation = var_24450_equation_0, values = (var_23924_cast_fp16, var_24326_cast_fp16))[name = string("op_24450_cast_fp16")]; + string var_24452_equation_0 = const()[name = string("op_24452_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24452_cast_fp16 = einsum(equation = var_24452_equation_0, values = (var_23928_cast_fp16, var_24327_cast_fp16))[name = string("op_24452_cast_fp16")]; + string var_24454_equation_0 = const()[name = string("op_24454_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24454_cast_fp16 = einsum(equation = var_24454_equation_0, values = (var_23928_cast_fp16, var_24328_cast_fp16))[name = string("op_24454_cast_fp16")]; + string var_24456_equation_0 = const()[name = string("op_24456_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24456_cast_fp16 = einsum(equation = var_24456_equation_0, values = (var_23928_cast_fp16, var_24329_cast_fp16))[name = string("op_24456_cast_fp16")]; + string var_24458_equation_0 = const()[name = string("op_24458_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24458_cast_fp16 = einsum(equation = var_24458_equation_0, values = (var_23928_cast_fp16, var_24330_cast_fp16))[name = string("op_24458_cast_fp16")]; + string var_24460_equation_0 = const()[name = string("op_24460_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24460_cast_fp16 = einsum(equation = var_24460_equation_0, values = (var_23932_cast_fp16, var_24331_cast_fp16))[name = string("op_24460_cast_fp16")]; + string var_24462_equation_0 = const()[name = string("op_24462_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24462_cast_fp16 = einsum(equation = var_24462_equation_0, values = (var_23932_cast_fp16, var_24332_cast_fp16))[name = string("op_24462_cast_fp16")]; + string var_24464_equation_0 = const()[name = string("op_24464_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24464_cast_fp16 = einsum(equation = var_24464_equation_0, values = (var_23932_cast_fp16, var_24333_cast_fp16))[name = string("op_24464_cast_fp16")]; + string var_24466_equation_0 = const()[name = string("op_24466_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24466_cast_fp16 = einsum(equation = var_24466_equation_0, values = (var_23932_cast_fp16, var_24334_cast_fp16))[name = string("op_24466_cast_fp16")]; + string var_24468_equation_0 = const()[name = string("op_24468_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24468_cast_fp16 = einsum(equation = var_24468_equation_0, values = (var_23936_cast_fp16, var_24335_cast_fp16))[name = string("op_24468_cast_fp16")]; + string var_24470_equation_0 = const()[name = string("op_24470_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24470_cast_fp16 = einsum(equation = var_24470_equation_0, values = (var_23936_cast_fp16, var_24336_cast_fp16))[name = string("op_24470_cast_fp16")]; + string var_24472_equation_0 = const()[name = string("op_24472_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24472_cast_fp16 = einsum(equation = var_24472_equation_0, values = (var_23936_cast_fp16, var_24337_cast_fp16))[name = string("op_24472_cast_fp16")]; + string var_24474_equation_0 = const()[name = string("op_24474_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24474_cast_fp16 = einsum(equation = var_24474_equation_0, values = (var_23936_cast_fp16, var_24338_cast_fp16))[name = string("op_24474_cast_fp16")]; + string var_24476_equation_0 = const()[name = string("op_24476_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24476_cast_fp16 = einsum(equation = var_24476_equation_0, values = (var_23940_cast_fp16, var_24339_cast_fp16))[name = string("op_24476_cast_fp16")]; + string var_24478_equation_0 = const()[name = string("op_24478_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24478_cast_fp16 = einsum(equation = var_24478_equation_0, values = (var_23940_cast_fp16, var_24340_cast_fp16))[name = string("op_24478_cast_fp16")]; + string var_24480_equation_0 = const()[name = string("op_24480_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24480_cast_fp16 = einsum(equation = var_24480_equation_0, values = (var_23940_cast_fp16, var_24341_cast_fp16))[name = string("op_24480_cast_fp16")]; + string var_24482_equation_0 = const()[name = string("op_24482_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24482_cast_fp16 = einsum(equation = var_24482_equation_0, values = (var_23940_cast_fp16, var_24342_cast_fp16))[name = string("op_24482_cast_fp16")]; + string var_24484_equation_0 = const()[name = string("op_24484_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24484_cast_fp16 = einsum(equation = var_24484_equation_0, values = (var_23944_cast_fp16, var_24343_cast_fp16))[name = string("op_24484_cast_fp16")]; + string var_24486_equation_0 = const()[name = string("op_24486_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24486_cast_fp16 = einsum(equation = var_24486_equation_0, values = (var_23944_cast_fp16, var_24344_cast_fp16))[name = string("op_24486_cast_fp16")]; + string var_24488_equation_0 = const()[name = string("op_24488_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24488_cast_fp16 = einsum(equation = var_24488_equation_0, values = (var_23944_cast_fp16, var_24345_cast_fp16))[name = string("op_24488_cast_fp16")]; + string var_24490_equation_0 = const()[name = string("op_24490_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24490_cast_fp16 = einsum(equation = var_24490_equation_0, values = (var_23944_cast_fp16, var_24346_cast_fp16))[name = string("op_24490_cast_fp16")]; + string var_24492_equation_0 = const()[name = string("op_24492_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24492_cast_fp16 = einsum(equation = var_24492_equation_0, values = (var_23948_cast_fp16, var_24347_cast_fp16))[name = string("op_24492_cast_fp16")]; + string var_24494_equation_0 = const()[name = string("op_24494_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24494_cast_fp16 = einsum(equation = var_24494_equation_0, values = (var_23948_cast_fp16, var_24348_cast_fp16))[name = string("op_24494_cast_fp16")]; + string var_24496_equation_0 = const()[name = string("op_24496_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24496_cast_fp16 = einsum(equation = var_24496_equation_0, values = (var_23948_cast_fp16, var_24349_cast_fp16))[name = string("op_24496_cast_fp16")]; + string var_24498_equation_0 = const()[name = string("op_24498_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24498_cast_fp16 = einsum(equation = var_24498_equation_0, values = (var_23948_cast_fp16, var_24350_cast_fp16))[name = string("op_24498_cast_fp16")]; + string var_24500_equation_0 = const()[name = string("op_24500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24500_cast_fp16 = einsum(equation = var_24500_equation_0, values = (var_23952_cast_fp16, var_24351_cast_fp16))[name = string("op_24500_cast_fp16")]; + string var_24502_equation_0 = const()[name = string("op_24502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24502_cast_fp16 = einsum(equation = var_24502_equation_0, values = (var_23952_cast_fp16, var_24352_cast_fp16))[name = string("op_24502_cast_fp16")]; + string var_24504_equation_0 = const()[name = string("op_24504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24504_cast_fp16 = einsum(equation = var_24504_equation_0, values = (var_23952_cast_fp16, var_24353_cast_fp16))[name = string("op_24504_cast_fp16")]; + string var_24506_equation_0 = const()[name = string("op_24506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24506_cast_fp16 = einsum(equation = var_24506_equation_0, values = (var_23952_cast_fp16, var_24354_cast_fp16))[name = string("op_24506_cast_fp16")]; + string var_24508_equation_0 = const()[name = string("op_24508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24508_cast_fp16 = einsum(equation = var_24508_equation_0, values = (var_23956_cast_fp16, var_24355_cast_fp16))[name = string("op_24508_cast_fp16")]; + string var_24510_equation_0 = const()[name = string("op_24510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24510_cast_fp16 = einsum(equation = var_24510_equation_0, values = (var_23956_cast_fp16, var_24356_cast_fp16))[name = string("op_24510_cast_fp16")]; + string var_24512_equation_0 = const()[name = string("op_24512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24512_cast_fp16 = einsum(equation = var_24512_equation_0, values = (var_23956_cast_fp16, var_24357_cast_fp16))[name = string("op_24512_cast_fp16")]; + string var_24514_equation_0 = const()[name = string("op_24514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24514_cast_fp16 = einsum(equation = var_24514_equation_0, values = (var_23956_cast_fp16, var_24358_cast_fp16))[name = string("op_24514_cast_fp16")]; + string var_24516_equation_0 = const()[name = string("op_24516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24516_cast_fp16 = einsum(equation = var_24516_equation_0, values = (var_23960_cast_fp16, var_24359_cast_fp16))[name = string("op_24516_cast_fp16")]; + string var_24518_equation_0 = const()[name = string("op_24518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24518_cast_fp16 = einsum(equation = var_24518_equation_0, values = (var_23960_cast_fp16, var_24360_cast_fp16))[name = string("op_24518_cast_fp16")]; + string var_24520_equation_0 = const()[name = string("op_24520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24520_cast_fp16 = einsum(equation = var_24520_equation_0, values = (var_23960_cast_fp16, var_24361_cast_fp16))[name = string("op_24520_cast_fp16")]; + string var_24522_equation_0 = const()[name = string("op_24522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24522_cast_fp16 = einsum(equation = var_24522_equation_0, values = (var_23960_cast_fp16, var_24362_cast_fp16))[name = string("op_24522_cast_fp16")]; + bool var_24524_interleave_0 = const()[name = string("op_24524_interleave_0"), val = bool(false)]; + tensor var_24524_cast_fp16 = concat(axis = var_23083, interleave = var_24524_interleave_0, values = (var_24364_cast_fp16, var_24366_cast_fp16, var_24368_cast_fp16, var_24370_cast_fp16))[name = string("op_24524_cast_fp16")]; + bool var_24526_interleave_0 = const()[name = string("op_24526_interleave_0"), val = bool(false)]; + tensor var_24526_cast_fp16 = concat(axis = var_23083, interleave = var_24526_interleave_0, values = (var_24372_cast_fp16, var_24374_cast_fp16, var_24376_cast_fp16, var_24378_cast_fp16))[name = string("op_24526_cast_fp16")]; + bool var_24528_interleave_0 = const()[name = string("op_24528_interleave_0"), val = bool(false)]; + tensor var_24528_cast_fp16 = concat(axis = var_23083, interleave = var_24528_interleave_0, values = (var_24380_cast_fp16, var_24382_cast_fp16, var_24384_cast_fp16, var_24386_cast_fp16))[name = string("op_24528_cast_fp16")]; + bool var_24530_interleave_0 = const()[name = string("op_24530_interleave_0"), val = bool(false)]; + tensor var_24530_cast_fp16 = concat(axis = var_23083, interleave = var_24530_interleave_0, values = (var_24388_cast_fp16, var_24390_cast_fp16, var_24392_cast_fp16, var_24394_cast_fp16))[name = string("op_24530_cast_fp16")]; + bool var_24532_interleave_0 = const()[name = string("op_24532_interleave_0"), val = bool(false)]; + tensor var_24532_cast_fp16 = concat(axis = var_23083, interleave = var_24532_interleave_0, values = (var_24396_cast_fp16, var_24398_cast_fp16, var_24400_cast_fp16, var_24402_cast_fp16))[name = string("op_24532_cast_fp16")]; + bool var_24534_interleave_0 = const()[name = string("op_24534_interleave_0"), val = bool(false)]; + tensor var_24534_cast_fp16 = concat(axis = var_23083, interleave = var_24534_interleave_0, values = (var_24404_cast_fp16, var_24406_cast_fp16, var_24408_cast_fp16, var_24410_cast_fp16))[name = string("op_24534_cast_fp16")]; + bool var_24536_interleave_0 = const()[name = string("op_24536_interleave_0"), val = bool(false)]; + tensor var_24536_cast_fp16 = concat(axis = var_23083, interleave = var_24536_interleave_0, values = (var_24412_cast_fp16, var_24414_cast_fp16, var_24416_cast_fp16, var_24418_cast_fp16))[name = string("op_24536_cast_fp16")]; + bool var_24538_interleave_0 = const()[name = string("op_24538_interleave_0"), val = bool(false)]; + tensor var_24538_cast_fp16 = concat(axis = var_23083, interleave = var_24538_interleave_0, values = (var_24420_cast_fp16, var_24422_cast_fp16, var_24424_cast_fp16, var_24426_cast_fp16))[name = string("op_24538_cast_fp16")]; + bool var_24540_interleave_0 = const()[name = string("op_24540_interleave_0"), val = bool(false)]; + tensor var_24540_cast_fp16 = concat(axis = var_23083, interleave = var_24540_interleave_0, values = (var_24428_cast_fp16, var_24430_cast_fp16, var_24432_cast_fp16, var_24434_cast_fp16))[name = string("op_24540_cast_fp16")]; + bool var_24542_interleave_0 = const()[name = string("op_24542_interleave_0"), val = bool(false)]; + tensor var_24542_cast_fp16 = concat(axis = var_23083, interleave = var_24542_interleave_0, values = (var_24436_cast_fp16, var_24438_cast_fp16, var_24440_cast_fp16, var_24442_cast_fp16))[name = string("op_24542_cast_fp16")]; + bool var_24544_interleave_0 = const()[name = string("op_24544_interleave_0"), val = bool(false)]; + tensor var_24544_cast_fp16 = concat(axis = var_23083, interleave = var_24544_interleave_0, values = (var_24444_cast_fp16, var_24446_cast_fp16, var_24448_cast_fp16, var_24450_cast_fp16))[name = string("op_24544_cast_fp16")]; + bool var_24546_interleave_0 = const()[name = string("op_24546_interleave_0"), val = bool(false)]; + tensor var_24546_cast_fp16 = concat(axis = var_23083, interleave = var_24546_interleave_0, values = (var_24452_cast_fp16, var_24454_cast_fp16, var_24456_cast_fp16, var_24458_cast_fp16))[name = string("op_24546_cast_fp16")]; + bool var_24548_interleave_0 = const()[name = string("op_24548_interleave_0"), val = bool(false)]; + tensor var_24548_cast_fp16 = concat(axis = var_23083, interleave = var_24548_interleave_0, values = (var_24460_cast_fp16, var_24462_cast_fp16, var_24464_cast_fp16, var_24466_cast_fp16))[name = string("op_24548_cast_fp16")]; + bool var_24550_interleave_0 = const()[name = string("op_24550_interleave_0"), val = bool(false)]; + tensor var_24550_cast_fp16 = concat(axis = var_23083, interleave = var_24550_interleave_0, values = (var_24468_cast_fp16, var_24470_cast_fp16, var_24472_cast_fp16, var_24474_cast_fp16))[name = string("op_24550_cast_fp16")]; + bool var_24552_interleave_0 = const()[name = string("op_24552_interleave_0"), val = bool(false)]; + tensor var_24552_cast_fp16 = concat(axis = var_23083, interleave = var_24552_interleave_0, values = (var_24476_cast_fp16, var_24478_cast_fp16, var_24480_cast_fp16, var_24482_cast_fp16))[name = string("op_24552_cast_fp16")]; + bool var_24554_interleave_0 = const()[name = string("op_24554_interleave_0"), val = bool(false)]; + tensor var_24554_cast_fp16 = concat(axis = var_23083, interleave = var_24554_interleave_0, values = (var_24484_cast_fp16, var_24486_cast_fp16, var_24488_cast_fp16, var_24490_cast_fp16))[name = string("op_24554_cast_fp16")]; + bool var_24556_interleave_0 = const()[name = string("op_24556_interleave_0"), val = bool(false)]; + tensor var_24556_cast_fp16 = concat(axis = var_23083, interleave = var_24556_interleave_0, values = (var_24492_cast_fp16, var_24494_cast_fp16, var_24496_cast_fp16, var_24498_cast_fp16))[name = string("op_24556_cast_fp16")]; + bool var_24558_interleave_0 = const()[name = string("op_24558_interleave_0"), val = bool(false)]; + tensor var_24558_cast_fp16 = concat(axis = var_23083, interleave = var_24558_interleave_0, values = (var_24500_cast_fp16, var_24502_cast_fp16, var_24504_cast_fp16, var_24506_cast_fp16))[name = string("op_24558_cast_fp16")]; + bool var_24560_interleave_0 = const()[name = string("op_24560_interleave_0"), val = bool(false)]; + tensor var_24560_cast_fp16 = concat(axis = var_23083, interleave = var_24560_interleave_0, values = (var_24508_cast_fp16, var_24510_cast_fp16, var_24512_cast_fp16, var_24514_cast_fp16))[name = string("op_24560_cast_fp16")]; + bool var_24562_interleave_0 = const()[name = string("op_24562_interleave_0"), val = bool(false)]; + tensor var_24562_cast_fp16 = concat(axis = var_23083, interleave = var_24562_interleave_0, values = (var_24516_cast_fp16, var_24518_cast_fp16, var_24520_cast_fp16, var_24522_cast_fp16))[name = string("op_24562_cast_fp16")]; + bool input_121_interleave_0 = const()[name = string("input_121_interleave_0"), val = bool(false)]; + tensor input_121_cast_fp16 = concat(axis = var_23108, interleave = input_121_interleave_0, values = (var_24524_cast_fp16, var_24526_cast_fp16, var_24528_cast_fp16, var_24530_cast_fp16, var_24532_cast_fp16, var_24534_cast_fp16, var_24536_cast_fp16, var_24538_cast_fp16, var_24540_cast_fp16, var_24542_cast_fp16, var_24544_cast_fp16, var_24546_cast_fp16, var_24548_cast_fp16, var_24550_cast_fp16, var_24552_cast_fp16, var_24554_cast_fp16, var_24556_cast_fp16, var_24558_cast_fp16, var_24560_cast_fp16, var_24562_cast_fp16))[name = string("input_121_cast_fp16")]; + string obj_63_pad_type_0 = const()[name = string("obj_63_pad_type_0"), val = string("valid")]; + tensor obj_63_strides_0 = const()[name = string("obj_63_strides_0"), val = tensor([1, 1])]; + tensor obj_63_pad_0 = const()[name = string("obj_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_63_dilations_0 = const()[name = string("obj_63_dilations_0"), val = tensor([1, 1])]; + int32 obj_63_groups_0 = const()[name = string("obj_63_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614804480)))]; + tensor layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618081344)))]; + tensor obj_63_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_121_cast_fp16)[name = string("obj_63_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_63_cast_fp16)[name = string("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; + fp16 var_24581_to_fp16 = const()[name = string("op_24581_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_24581_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; + tensor input_123_gamma_0_to_fp16 = const()[name = string("input_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618083968)))]; + tensor input_123_beta_0_to_fp16 = const()[name = string("input_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618086592)))]; + fp16 input_123_epsilon_0_to_fp16 = const()[name = string("input_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_123_cast_fp16 = batch_norm(beta = input_123_beta_0_to_fp16, epsilon = input_123_epsilon_0_to_fp16, gamma = input_123_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("input_123_cast_fp16")]; + string input_125_pad_type_0 = const()[name = string("input_125_pad_type_0"), val = string("valid")]; + tensor input_125_strides_0 = const()[name = string("input_125_strides_0"), val = tensor([1, 1])]; + tensor input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_125_dilations_0 = const()[name = string("input_125_dilations_0"), val = tensor([1, 1])]; + int32 input_125_groups_0 = const()[name = string("input_125_groups_0"), val = int32(1)]; + tensor layers_15_fc1_weight_to_fp16 = const()[name = string("layers_15_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618089216)))]; + tensor layers_15_fc1_bias_to_fp16 = const()[name = string("layers_15_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631196480)))]; + tensor input_125_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = layers_15_fc1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")]; + string input_127_mode_0 = const()[name = string("input_127_mode_0"), val = string("EXACT")]; + tensor input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")]; + string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")]; + tensor hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)]; + tensor layers_15_fc2_weight_to_fp16 = const()[name = string("layers_15_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631206784)))]; + tensor layers_15_fc2_bias_to_fp16 = const()[name = string("layers_15_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644314048)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_15_fc2_weight_to_fp16, x = input_127_cast_fp16)[name = string("hidden_states_35_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = hidden_states_35_cast_fp16)[name = string("inputs_65_cast_fp16")]; + int32 var_24610 = const()[name = string("op_24610"), val = int32(3)]; + int32 var_24635 = const()[name = string("op_24635"), val = int32(1)]; + tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; + fp16 var_24652_to_fp16 = const()[name = string("op_24652_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_24652_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; + tensor obj_65_gamma_0_to_fp16 = const()[name = string("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644316672)))]; + tensor obj_65_beta_0_to_fp16 = const()[name = string("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644319296)))]; + fp16 obj_65_epsilon_0_to_fp16 = const()[name = string("obj_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("obj_65_cast_fp16")]; + string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; + tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; + tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; + int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644321920)))]; + tensor layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647598784)))]; + tensor query_33_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("query_33_cast_fp16")]; + string key_33_pad_type_0 = const()[name = string("key_33_pad_type_0"), val = string("valid")]; + tensor key_33_strides_0 = const()[name = string("key_33_strides_0"), val = tensor([1, 1])]; + tensor key_33_pad_0 = const()[name = string("key_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_33_dilations_0 = const()[name = string("key_33_dilations_0"), val = tensor([1, 1])]; + int32 key_33_groups_0 = const()[name = string("key_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647601408)))]; + tensor key_33_cast_fp16 = conv(dilations = key_33_dilations_0, groups = key_33_groups_0, pad = key_33_pad_0, pad_type = key_33_pad_type_0, strides = key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("key_33_cast_fp16")]; + string value_33_pad_type_0 = const()[name = string("value_33_pad_type_0"), val = string("valid")]; + tensor value_33_strides_0 = const()[name = string("value_33_strides_0"), val = tensor([1, 1])]; + tensor value_33_pad_0 = const()[name = string("value_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_33_dilations_0 = const()[name = string("value_33_dilations_0"), val = tensor([1, 1])]; + int32 value_33_groups_0 = const()[name = string("value_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(650878272)))]; + tensor layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654155136)))]; + tensor value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = value_33_dilations_0, groups = value_33_groups_0, pad = value_33_pad_0, pad_type = value_33_pad_type_0, strides = value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("value_33_cast_fp16")]; + tensor var_24690_begin_0 = const()[name = string("op_24690_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24690_end_0 = const()[name = string("op_24690_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24690_end_mask_0 = const()[name = string("op_24690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24690_cast_fp16 = slice_by_index(begin = var_24690_begin_0, end = var_24690_end_0, end_mask = var_24690_end_mask_0, x = query_33_cast_fp16)[name = string("op_24690_cast_fp16")]; + tensor var_24694_begin_0 = const()[name = string("op_24694_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24694_end_0 = const()[name = string("op_24694_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_24694_end_mask_0 = const()[name = string("op_24694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24694_cast_fp16 = slice_by_index(begin = var_24694_begin_0, end = var_24694_end_0, end_mask = var_24694_end_mask_0, x = query_33_cast_fp16)[name = string("op_24694_cast_fp16")]; + tensor var_24698_begin_0 = const()[name = string("op_24698_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24698_end_0 = const()[name = string("op_24698_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_24698_end_mask_0 = const()[name = string("op_24698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24698_cast_fp16 = slice_by_index(begin = var_24698_begin_0, end = var_24698_end_0, end_mask = var_24698_end_mask_0, x = query_33_cast_fp16)[name = string("op_24698_cast_fp16")]; + tensor var_24702_begin_0 = const()[name = string("op_24702_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24702_end_0 = const()[name = string("op_24702_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_24702_end_mask_0 = const()[name = string("op_24702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24702_cast_fp16 = slice_by_index(begin = var_24702_begin_0, end = var_24702_end_0, end_mask = var_24702_end_mask_0, x = query_33_cast_fp16)[name = string("op_24702_cast_fp16")]; + tensor var_24706_begin_0 = const()[name = string("op_24706_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24706_end_0 = const()[name = string("op_24706_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_24706_end_mask_0 = const()[name = string("op_24706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24706_cast_fp16 = slice_by_index(begin = var_24706_begin_0, end = var_24706_end_0, end_mask = var_24706_end_mask_0, x = query_33_cast_fp16)[name = string("op_24706_cast_fp16")]; + tensor var_24710_begin_0 = const()[name = string("op_24710_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24710_end_0 = const()[name = string("op_24710_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_24710_end_mask_0 = const()[name = string("op_24710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24710_cast_fp16 = slice_by_index(begin = var_24710_begin_0, end = var_24710_end_0, end_mask = var_24710_end_mask_0, x = query_33_cast_fp16)[name = string("op_24710_cast_fp16")]; + tensor var_24714_begin_0 = const()[name = string("op_24714_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24714_end_0 = const()[name = string("op_24714_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_24714_end_mask_0 = const()[name = string("op_24714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24714_cast_fp16 = slice_by_index(begin = var_24714_begin_0, end = var_24714_end_0, end_mask = var_24714_end_mask_0, x = query_33_cast_fp16)[name = string("op_24714_cast_fp16")]; + tensor var_24718_begin_0 = const()[name = string("op_24718_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24718_end_0 = const()[name = string("op_24718_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_24718_end_mask_0 = const()[name = string("op_24718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24718_cast_fp16 = slice_by_index(begin = var_24718_begin_0, end = var_24718_end_0, end_mask = var_24718_end_mask_0, x = query_33_cast_fp16)[name = string("op_24718_cast_fp16")]; + tensor var_24722_begin_0 = const()[name = string("op_24722_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24722_end_0 = const()[name = string("op_24722_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_24722_end_mask_0 = const()[name = string("op_24722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24722_cast_fp16 = slice_by_index(begin = var_24722_begin_0, end = var_24722_end_0, end_mask = var_24722_end_mask_0, x = query_33_cast_fp16)[name = string("op_24722_cast_fp16")]; + tensor var_24726_begin_0 = const()[name = string("op_24726_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24726_end_0 = const()[name = string("op_24726_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_24726_end_mask_0 = const()[name = string("op_24726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24726_cast_fp16 = slice_by_index(begin = var_24726_begin_0, end = var_24726_end_0, end_mask = var_24726_end_mask_0, x = query_33_cast_fp16)[name = string("op_24726_cast_fp16")]; + tensor var_24730_begin_0 = const()[name = string("op_24730_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24730_end_0 = const()[name = string("op_24730_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_24730_end_mask_0 = const()[name = string("op_24730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24730_cast_fp16 = slice_by_index(begin = var_24730_begin_0, end = var_24730_end_0, end_mask = var_24730_end_mask_0, x = query_33_cast_fp16)[name = string("op_24730_cast_fp16")]; + tensor var_24734_begin_0 = const()[name = string("op_24734_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24734_end_0 = const()[name = string("op_24734_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_24734_end_mask_0 = const()[name = string("op_24734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24734_cast_fp16 = slice_by_index(begin = var_24734_begin_0, end = var_24734_end_0, end_mask = var_24734_end_mask_0, x = query_33_cast_fp16)[name = string("op_24734_cast_fp16")]; + tensor var_24738_begin_0 = const()[name = string("op_24738_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24738_end_0 = const()[name = string("op_24738_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_24738_end_mask_0 = const()[name = string("op_24738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24738_cast_fp16 = slice_by_index(begin = var_24738_begin_0, end = var_24738_end_0, end_mask = var_24738_end_mask_0, x = query_33_cast_fp16)[name = string("op_24738_cast_fp16")]; + tensor var_24742_begin_0 = const()[name = string("op_24742_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24742_end_0 = const()[name = string("op_24742_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_24742_end_mask_0 = const()[name = string("op_24742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24742_cast_fp16 = slice_by_index(begin = var_24742_begin_0, end = var_24742_end_0, end_mask = var_24742_end_mask_0, x = query_33_cast_fp16)[name = string("op_24742_cast_fp16")]; + tensor var_24746_begin_0 = const()[name = string("op_24746_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24746_end_0 = const()[name = string("op_24746_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_24746_end_mask_0 = const()[name = string("op_24746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24746_cast_fp16 = slice_by_index(begin = var_24746_begin_0, end = var_24746_end_0, end_mask = var_24746_end_mask_0, x = query_33_cast_fp16)[name = string("op_24746_cast_fp16")]; + tensor var_24750_begin_0 = const()[name = string("op_24750_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24750_end_0 = const()[name = string("op_24750_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_24750_end_mask_0 = const()[name = string("op_24750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24750_cast_fp16 = slice_by_index(begin = var_24750_begin_0, end = var_24750_end_0, end_mask = var_24750_end_mask_0, x = query_33_cast_fp16)[name = string("op_24750_cast_fp16")]; + tensor var_24754_begin_0 = const()[name = string("op_24754_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24754_end_0 = const()[name = string("op_24754_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_24754_end_mask_0 = const()[name = string("op_24754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24754_cast_fp16 = slice_by_index(begin = var_24754_begin_0, end = var_24754_end_0, end_mask = var_24754_end_mask_0, x = query_33_cast_fp16)[name = string("op_24754_cast_fp16")]; + tensor var_24758_begin_0 = const()[name = string("op_24758_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24758_end_0 = const()[name = string("op_24758_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_24758_end_mask_0 = const()[name = string("op_24758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24758_cast_fp16 = slice_by_index(begin = var_24758_begin_0, end = var_24758_end_0, end_mask = var_24758_end_mask_0, x = query_33_cast_fp16)[name = string("op_24758_cast_fp16")]; + tensor var_24762_begin_0 = const()[name = string("op_24762_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24762_end_0 = const()[name = string("op_24762_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_24762_end_mask_0 = const()[name = string("op_24762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24762_cast_fp16 = slice_by_index(begin = var_24762_begin_0, end = var_24762_end_0, end_mask = var_24762_end_mask_0, x = query_33_cast_fp16)[name = string("op_24762_cast_fp16")]; + tensor var_24766_begin_0 = const()[name = string("op_24766_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24766_end_0 = const()[name = string("op_24766_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_24766_end_mask_0 = const()[name = string("op_24766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24766_cast_fp16 = slice_by_index(begin = var_24766_begin_0, end = var_24766_end_0, end_mask = var_24766_end_mask_0, x = query_33_cast_fp16)[name = string("op_24766_cast_fp16")]; + tensor var_24775_begin_0 = const()[name = string("op_24775_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24775_end_0 = const()[name = string("op_24775_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24775_end_mask_0 = const()[name = string("op_24775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24775_cast_fp16 = slice_by_index(begin = var_24775_begin_0, end = var_24775_end_0, end_mask = var_24775_end_mask_0, x = var_24690_cast_fp16)[name = string("op_24775_cast_fp16")]; + tensor var_24782_begin_0 = const()[name = string("op_24782_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24782_end_0 = const()[name = string("op_24782_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24782_end_mask_0 = const()[name = string("op_24782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24782_cast_fp16 = slice_by_index(begin = var_24782_begin_0, end = var_24782_end_0, end_mask = var_24782_end_mask_0, x = var_24690_cast_fp16)[name = string("op_24782_cast_fp16")]; + tensor var_24789_begin_0 = const()[name = string("op_24789_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24789_end_0 = const()[name = string("op_24789_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24789_end_mask_0 = const()[name = string("op_24789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24789_cast_fp16 = slice_by_index(begin = var_24789_begin_0, end = var_24789_end_0, end_mask = var_24789_end_mask_0, x = var_24690_cast_fp16)[name = string("op_24789_cast_fp16")]; + tensor var_24796_begin_0 = const()[name = string("op_24796_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24796_end_0 = const()[name = string("op_24796_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24796_end_mask_0 = const()[name = string("op_24796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24796_cast_fp16 = slice_by_index(begin = var_24796_begin_0, end = var_24796_end_0, end_mask = var_24796_end_mask_0, x = var_24690_cast_fp16)[name = string("op_24796_cast_fp16")]; + tensor var_24803_begin_0 = const()[name = string("op_24803_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24803_end_0 = const()[name = string("op_24803_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24803_end_mask_0 = const()[name = string("op_24803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24803_cast_fp16 = slice_by_index(begin = var_24803_begin_0, end = var_24803_end_0, end_mask = var_24803_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24803_cast_fp16")]; + tensor var_24810_begin_0 = const()[name = string("op_24810_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24810_end_0 = const()[name = string("op_24810_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24810_end_mask_0 = const()[name = string("op_24810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24810_cast_fp16 = slice_by_index(begin = var_24810_begin_0, end = var_24810_end_0, end_mask = var_24810_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24810_cast_fp16")]; + tensor var_24817_begin_0 = const()[name = string("op_24817_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24817_end_0 = const()[name = string("op_24817_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24817_end_mask_0 = const()[name = string("op_24817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24817_cast_fp16 = slice_by_index(begin = var_24817_begin_0, end = var_24817_end_0, end_mask = var_24817_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24817_cast_fp16")]; + tensor var_24824_begin_0 = const()[name = string("op_24824_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24824_end_0 = const()[name = string("op_24824_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24824_end_mask_0 = const()[name = string("op_24824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24824_cast_fp16 = slice_by_index(begin = var_24824_begin_0, end = var_24824_end_0, end_mask = var_24824_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24824_cast_fp16")]; + tensor var_24831_begin_0 = const()[name = string("op_24831_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24831_end_0 = const()[name = string("op_24831_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24831_end_mask_0 = const()[name = string("op_24831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24831_cast_fp16 = slice_by_index(begin = var_24831_begin_0, end = var_24831_end_0, end_mask = var_24831_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24831_cast_fp16")]; + tensor var_24838_begin_0 = const()[name = string("op_24838_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24838_end_0 = const()[name = string("op_24838_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24838_end_mask_0 = const()[name = string("op_24838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24838_cast_fp16 = slice_by_index(begin = var_24838_begin_0, end = var_24838_end_0, end_mask = var_24838_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24838_cast_fp16")]; + tensor var_24845_begin_0 = const()[name = string("op_24845_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24845_end_0 = const()[name = string("op_24845_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24845_end_mask_0 = const()[name = string("op_24845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24845_cast_fp16 = slice_by_index(begin = var_24845_begin_0, end = var_24845_end_0, end_mask = var_24845_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24845_cast_fp16")]; + tensor var_24852_begin_0 = const()[name = string("op_24852_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24852_end_0 = const()[name = string("op_24852_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24852_end_mask_0 = const()[name = string("op_24852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24852_cast_fp16 = slice_by_index(begin = var_24852_begin_0, end = var_24852_end_0, end_mask = var_24852_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24852_cast_fp16")]; + tensor var_24859_begin_0 = const()[name = string("op_24859_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24859_end_0 = const()[name = string("op_24859_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24859_end_mask_0 = const()[name = string("op_24859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24859_cast_fp16 = slice_by_index(begin = var_24859_begin_0, end = var_24859_end_0, end_mask = var_24859_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24859_cast_fp16")]; + tensor var_24866_begin_0 = const()[name = string("op_24866_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24866_end_0 = const()[name = string("op_24866_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24866_end_mask_0 = const()[name = string("op_24866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24866_cast_fp16 = slice_by_index(begin = var_24866_begin_0, end = var_24866_end_0, end_mask = var_24866_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24866_cast_fp16")]; + tensor var_24873_begin_0 = const()[name = string("op_24873_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24873_end_0 = const()[name = string("op_24873_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24873_end_mask_0 = const()[name = string("op_24873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24873_cast_fp16 = slice_by_index(begin = var_24873_begin_0, end = var_24873_end_0, end_mask = var_24873_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24873_cast_fp16")]; + tensor var_24880_begin_0 = const()[name = string("op_24880_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24880_end_0 = const()[name = string("op_24880_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24880_end_mask_0 = const()[name = string("op_24880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24880_cast_fp16 = slice_by_index(begin = var_24880_begin_0, end = var_24880_end_0, end_mask = var_24880_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24880_cast_fp16")]; + tensor var_24887_begin_0 = const()[name = string("op_24887_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24887_end_0 = const()[name = string("op_24887_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24887_end_mask_0 = const()[name = string("op_24887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24887_cast_fp16 = slice_by_index(begin = var_24887_begin_0, end = var_24887_end_0, end_mask = var_24887_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24887_cast_fp16")]; + tensor var_24894_begin_0 = const()[name = string("op_24894_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24894_end_0 = const()[name = string("op_24894_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24894_end_mask_0 = const()[name = string("op_24894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24894_cast_fp16 = slice_by_index(begin = var_24894_begin_0, end = var_24894_end_0, end_mask = var_24894_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24894_cast_fp16")]; + tensor var_24901_begin_0 = const()[name = string("op_24901_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24901_end_0 = const()[name = string("op_24901_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24901_end_mask_0 = const()[name = string("op_24901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24901_cast_fp16 = slice_by_index(begin = var_24901_begin_0, end = var_24901_end_0, end_mask = var_24901_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24901_cast_fp16")]; + tensor var_24908_begin_0 = const()[name = string("op_24908_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24908_end_0 = const()[name = string("op_24908_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24908_end_mask_0 = const()[name = string("op_24908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24908_cast_fp16 = slice_by_index(begin = var_24908_begin_0, end = var_24908_end_0, end_mask = var_24908_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24908_cast_fp16")]; + tensor var_24915_begin_0 = const()[name = string("op_24915_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24915_end_0 = const()[name = string("op_24915_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24915_end_mask_0 = const()[name = string("op_24915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24915_cast_fp16 = slice_by_index(begin = var_24915_begin_0, end = var_24915_end_0, end_mask = var_24915_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24915_cast_fp16")]; + tensor var_24922_begin_0 = const()[name = string("op_24922_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24922_end_0 = const()[name = string("op_24922_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24922_end_mask_0 = const()[name = string("op_24922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24922_cast_fp16 = slice_by_index(begin = var_24922_begin_0, end = var_24922_end_0, end_mask = var_24922_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24922_cast_fp16")]; + tensor var_24929_begin_0 = const()[name = string("op_24929_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24929_end_0 = const()[name = string("op_24929_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24929_end_mask_0 = const()[name = string("op_24929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24929_cast_fp16 = slice_by_index(begin = var_24929_begin_0, end = var_24929_end_0, end_mask = var_24929_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24929_cast_fp16")]; + tensor var_24936_begin_0 = const()[name = string("op_24936_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24936_end_0 = const()[name = string("op_24936_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24936_end_mask_0 = const()[name = string("op_24936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24936_cast_fp16 = slice_by_index(begin = var_24936_begin_0, end = var_24936_end_0, end_mask = var_24936_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24936_cast_fp16")]; + tensor var_24943_begin_0 = const()[name = string("op_24943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24943_end_0 = const()[name = string("op_24943_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24943_end_mask_0 = const()[name = string("op_24943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24943_cast_fp16 = slice_by_index(begin = var_24943_begin_0, end = var_24943_end_0, end_mask = var_24943_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24943_cast_fp16")]; + tensor var_24950_begin_0 = const()[name = string("op_24950_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24950_end_0 = const()[name = string("op_24950_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24950_end_mask_0 = const()[name = string("op_24950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24950_cast_fp16 = slice_by_index(begin = var_24950_begin_0, end = var_24950_end_0, end_mask = var_24950_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24950_cast_fp16")]; + tensor var_24957_begin_0 = const()[name = string("op_24957_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24957_end_0 = const()[name = string("op_24957_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24957_end_mask_0 = const()[name = string("op_24957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24957_cast_fp16 = slice_by_index(begin = var_24957_begin_0, end = var_24957_end_0, end_mask = var_24957_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24957_cast_fp16")]; + tensor var_24964_begin_0 = const()[name = string("op_24964_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24964_end_0 = const()[name = string("op_24964_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24964_end_mask_0 = const()[name = string("op_24964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24964_cast_fp16 = slice_by_index(begin = var_24964_begin_0, end = var_24964_end_0, end_mask = var_24964_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24964_cast_fp16")]; + tensor var_24971_begin_0 = const()[name = string("op_24971_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24971_end_0 = const()[name = string("op_24971_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24971_end_mask_0 = const()[name = string("op_24971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24971_cast_fp16 = slice_by_index(begin = var_24971_begin_0, end = var_24971_end_0, end_mask = var_24971_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24971_cast_fp16")]; + tensor var_24978_begin_0 = const()[name = string("op_24978_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24978_end_0 = const()[name = string("op_24978_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24978_end_mask_0 = const()[name = string("op_24978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24978_cast_fp16 = slice_by_index(begin = var_24978_begin_0, end = var_24978_end_0, end_mask = var_24978_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24978_cast_fp16")]; + tensor var_24985_begin_0 = const()[name = string("op_24985_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24985_end_0 = const()[name = string("op_24985_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24985_end_mask_0 = const()[name = string("op_24985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24985_cast_fp16 = slice_by_index(begin = var_24985_begin_0, end = var_24985_end_0, end_mask = var_24985_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24985_cast_fp16")]; + tensor var_24992_begin_0 = const()[name = string("op_24992_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24992_end_0 = const()[name = string("op_24992_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24992_end_mask_0 = const()[name = string("op_24992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24992_cast_fp16 = slice_by_index(begin = var_24992_begin_0, end = var_24992_end_0, end_mask = var_24992_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24992_cast_fp16")]; + tensor var_24999_begin_0 = const()[name = string("op_24999_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24999_end_0 = const()[name = string("op_24999_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24999_end_mask_0 = const()[name = string("op_24999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24999_cast_fp16 = slice_by_index(begin = var_24999_begin_0, end = var_24999_end_0, end_mask = var_24999_end_mask_0, x = var_24722_cast_fp16)[name = string("op_24999_cast_fp16")]; + tensor var_25006_begin_0 = const()[name = string("op_25006_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25006_end_0 = const()[name = string("op_25006_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25006_end_mask_0 = const()[name = string("op_25006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25006_cast_fp16 = slice_by_index(begin = var_25006_begin_0, end = var_25006_end_0, end_mask = var_25006_end_mask_0, x = var_24722_cast_fp16)[name = string("op_25006_cast_fp16")]; + tensor var_25013_begin_0 = const()[name = string("op_25013_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25013_end_0 = const()[name = string("op_25013_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25013_end_mask_0 = const()[name = string("op_25013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25013_cast_fp16 = slice_by_index(begin = var_25013_begin_0, end = var_25013_end_0, end_mask = var_25013_end_mask_0, x = var_24722_cast_fp16)[name = string("op_25013_cast_fp16")]; + tensor var_25020_begin_0 = const()[name = string("op_25020_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25020_end_0 = const()[name = string("op_25020_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25020_end_mask_0 = const()[name = string("op_25020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25020_cast_fp16 = slice_by_index(begin = var_25020_begin_0, end = var_25020_end_0, end_mask = var_25020_end_mask_0, x = var_24722_cast_fp16)[name = string("op_25020_cast_fp16")]; + tensor var_25027_begin_0 = const()[name = string("op_25027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25027_end_0 = const()[name = string("op_25027_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25027_end_mask_0 = const()[name = string("op_25027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25027_cast_fp16 = slice_by_index(begin = var_25027_begin_0, end = var_25027_end_0, end_mask = var_25027_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25027_cast_fp16")]; + tensor var_25034_begin_0 = const()[name = string("op_25034_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25034_end_0 = const()[name = string("op_25034_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25034_end_mask_0 = const()[name = string("op_25034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25034_cast_fp16 = slice_by_index(begin = var_25034_begin_0, end = var_25034_end_0, end_mask = var_25034_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25034_cast_fp16")]; + tensor var_25041_begin_0 = const()[name = string("op_25041_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25041_end_0 = const()[name = string("op_25041_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25041_end_mask_0 = const()[name = string("op_25041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25041_cast_fp16 = slice_by_index(begin = var_25041_begin_0, end = var_25041_end_0, end_mask = var_25041_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25041_cast_fp16")]; + tensor var_25048_begin_0 = const()[name = string("op_25048_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25048_end_0 = const()[name = string("op_25048_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25048_end_mask_0 = const()[name = string("op_25048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25048_cast_fp16 = slice_by_index(begin = var_25048_begin_0, end = var_25048_end_0, end_mask = var_25048_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25048_cast_fp16")]; + tensor var_25055_begin_0 = const()[name = string("op_25055_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25055_end_0 = const()[name = string("op_25055_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25055_end_mask_0 = const()[name = string("op_25055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25055_cast_fp16 = slice_by_index(begin = var_25055_begin_0, end = var_25055_end_0, end_mask = var_25055_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25055_cast_fp16")]; + tensor var_25062_begin_0 = const()[name = string("op_25062_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25062_end_0 = const()[name = string("op_25062_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25062_end_mask_0 = const()[name = string("op_25062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25062_cast_fp16 = slice_by_index(begin = var_25062_begin_0, end = var_25062_end_0, end_mask = var_25062_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25062_cast_fp16")]; + tensor var_25069_begin_0 = const()[name = string("op_25069_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25069_end_0 = const()[name = string("op_25069_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25069_end_mask_0 = const()[name = string("op_25069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25069_cast_fp16 = slice_by_index(begin = var_25069_begin_0, end = var_25069_end_0, end_mask = var_25069_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25069_cast_fp16")]; + tensor var_25076_begin_0 = const()[name = string("op_25076_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25076_end_0 = const()[name = string("op_25076_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25076_end_mask_0 = const()[name = string("op_25076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25076_cast_fp16 = slice_by_index(begin = var_25076_begin_0, end = var_25076_end_0, end_mask = var_25076_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25076_cast_fp16")]; + tensor var_25083_begin_0 = const()[name = string("op_25083_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25083_end_0 = const()[name = string("op_25083_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25083_end_mask_0 = const()[name = string("op_25083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25083_cast_fp16 = slice_by_index(begin = var_25083_begin_0, end = var_25083_end_0, end_mask = var_25083_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25083_cast_fp16")]; + tensor var_25090_begin_0 = const()[name = string("op_25090_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25090_end_0 = const()[name = string("op_25090_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25090_end_mask_0 = const()[name = string("op_25090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25090_cast_fp16 = slice_by_index(begin = var_25090_begin_0, end = var_25090_end_0, end_mask = var_25090_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25090_cast_fp16")]; + tensor var_25097_begin_0 = const()[name = string("op_25097_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25097_end_0 = const()[name = string("op_25097_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25097_end_mask_0 = const()[name = string("op_25097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25097_cast_fp16 = slice_by_index(begin = var_25097_begin_0, end = var_25097_end_0, end_mask = var_25097_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25097_cast_fp16")]; + tensor var_25104_begin_0 = const()[name = string("op_25104_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25104_end_0 = const()[name = string("op_25104_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25104_end_mask_0 = const()[name = string("op_25104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25104_cast_fp16 = slice_by_index(begin = var_25104_begin_0, end = var_25104_end_0, end_mask = var_25104_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25104_cast_fp16")]; + tensor var_25111_begin_0 = const()[name = string("op_25111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25111_end_0 = const()[name = string("op_25111_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25111_end_mask_0 = const()[name = string("op_25111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25111_cast_fp16 = slice_by_index(begin = var_25111_begin_0, end = var_25111_end_0, end_mask = var_25111_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25111_cast_fp16")]; + tensor var_25118_begin_0 = const()[name = string("op_25118_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25118_end_0 = const()[name = string("op_25118_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25118_end_mask_0 = const()[name = string("op_25118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25118_cast_fp16 = slice_by_index(begin = var_25118_begin_0, end = var_25118_end_0, end_mask = var_25118_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25118_cast_fp16")]; + tensor var_25125_begin_0 = const()[name = string("op_25125_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25125_end_0 = const()[name = string("op_25125_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25125_end_mask_0 = const()[name = string("op_25125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25125_cast_fp16 = slice_by_index(begin = var_25125_begin_0, end = var_25125_end_0, end_mask = var_25125_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25125_cast_fp16")]; + tensor var_25132_begin_0 = const()[name = string("op_25132_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25132_end_0 = const()[name = string("op_25132_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25132_end_mask_0 = const()[name = string("op_25132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25132_cast_fp16 = slice_by_index(begin = var_25132_begin_0, end = var_25132_end_0, end_mask = var_25132_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25132_cast_fp16")]; + tensor var_25139_begin_0 = const()[name = string("op_25139_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25139_end_0 = const()[name = string("op_25139_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25139_end_mask_0 = const()[name = string("op_25139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25139_cast_fp16 = slice_by_index(begin = var_25139_begin_0, end = var_25139_end_0, end_mask = var_25139_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25139_cast_fp16")]; + tensor var_25146_begin_0 = const()[name = string("op_25146_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25146_end_0 = const()[name = string("op_25146_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25146_end_mask_0 = const()[name = string("op_25146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25146_cast_fp16 = slice_by_index(begin = var_25146_begin_0, end = var_25146_end_0, end_mask = var_25146_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25146_cast_fp16")]; + tensor var_25153_begin_0 = const()[name = string("op_25153_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25153_end_0 = const()[name = string("op_25153_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25153_end_mask_0 = const()[name = string("op_25153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25153_cast_fp16 = slice_by_index(begin = var_25153_begin_0, end = var_25153_end_0, end_mask = var_25153_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25153_cast_fp16")]; + tensor var_25160_begin_0 = const()[name = string("op_25160_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25160_end_0 = const()[name = string("op_25160_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25160_end_mask_0 = const()[name = string("op_25160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25160_cast_fp16 = slice_by_index(begin = var_25160_begin_0, end = var_25160_end_0, end_mask = var_25160_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25160_cast_fp16")]; + tensor var_25167_begin_0 = const()[name = string("op_25167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25167_end_0 = const()[name = string("op_25167_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25167_end_mask_0 = const()[name = string("op_25167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25167_cast_fp16 = slice_by_index(begin = var_25167_begin_0, end = var_25167_end_0, end_mask = var_25167_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25167_cast_fp16")]; + tensor var_25174_begin_0 = const()[name = string("op_25174_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25174_end_0 = const()[name = string("op_25174_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25174_end_mask_0 = const()[name = string("op_25174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25174_cast_fp16 = slice_by_index(begin = var_25174_begin_0, end = var_25174_end_0, end_mask = var_25174_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25174_cast_fp16")]; + tensor var_25181_begin_0 = const()[name = string("op_25181_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25181_end_0 = const()[name = string("op_25181_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25181_end_mask_0 = const()[name = string("op_25181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25181_cast_fp16 = slice_by_index(begin = var_25181_begin_0, end = var_25181_end_0, end_mask = var_25181_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25181_cast_fp16")]; + tensor var_25188_begin_0 = const()[name = string("op_25188_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25188_end_0 = const()[name = string("op_25188_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25188_end_mask_0 = const()[name = string("op_25188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25188_cast_fp16 = slice_by_index(begin = var_25188_begin_0, end = var_25188_end_0, end_mask = var_25188_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25188_cast_fp16")]; + tensor var_25195_begin_0 = const()[name = string("op_25195_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25195_end_0 = const()[name = string("op_25195_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25195_end_mask_0 = const()[name = string("op_25195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25195_cast_fp16 = slice_by_index(begin = var_25195_begin_0, end = var_25195_end_0, end_mask = var_25195_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25195_cast_fp16")]; + tensor var_25202_begin_0 = const()[name = string("op_25202_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25202_end_0 = const()[name = string("op_25202_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25202_end_mask_0 = const()[name = string("op_25202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25202_cast_fp16 = slice_by_index(begin = var_25202_begin_0, end = var_25202_end_0, end_mask = var_25202_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25202_cast_fp16")]; + tensor var_25209_begin_0 = const()[name = string("op_25209_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25209_end_0 = const()[name = string("op_25209_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25209_end_mask_0 = const()[name = string("op_25209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25209_cast_fp16 = slice_by_index(begin = var_25209_begin_0, end = var_25209_end_0, end_mask = var_25209_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25209_cast_fp16")]; + tensor var_25216_begin_0 = const()[name = string("op_25216_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25216_end_0 = const()[name = string("op_25216_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25216_end_mask_0 = const()[name = string("op_25216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25216_cast_fp16 = slice_by_index(begin = var_25216_begin_0, end = var_25216_end_0, end_mask = var_25216_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25216_cast_fp16")]; + tensor var_25223_begin_0 = const()[name = string("op_25223_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25223_end_0 = const()[name = string("op_25223_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25223_end_mask_0 = const()[name = string("op_25223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25223_cast_fp16 = slice_by_index(begin = var_25223_begin_0, end = var_25223_end_0, end_mask = var_25223_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25223_cast_fp16")]; + tensor var_25230_begin_0 = const()[name = string("op_25230_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25230_end_0 = const()[name = string("op_25230_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25230_end_mask_0 = const()[name = string("op_25230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25230_cast_fp16 = slice_by_index(begin = var_25230_begin_0, end = var_25230_end_0, end_mask = var_25230_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25230_cast_fp16")]; + tensor var_25237_begin_0 = const()[name = string("op_25237_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25237_end_0 = const()[name = string("op_25237_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25237_end_mask_0 = const()[name = string("op_25237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25237_cast_fp16 = slice_by_index(begin = var_25237_begin_0, end = var_25237_end_0, end_mask = var_25237_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25237_cast_fp16")]; + tensor var_25244_begin_0 = const()[name = string("op_25244_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25244_end_0 = const()[name = string("op_25244_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25244_end_mask_0 = const()[name = string("op_25244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25244_cast_fp16 = slice_by_index(begin = var_25244_begin_0, end = var_25244_end_0, end_mask = var_25244_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25244_cast_fp16")]; + tensor var_25251_begin_0 = const()[name = string("op_25251_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25251_end_0 = const()[name = string("op_25251_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25251_end_mask_0 = const()[name = string("op_25251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25251_cast_fp16 = slice_by_index(begin = var_25251_begin_0, end = var_25251_end_0, end_mask = var_25251_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25251_cast_fp16")]; + tensor var_25258_begin_0 = const()[name = string("op_25258_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25258_end_0 = const()[name = string("op_25258_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25258_end_mask_0 = const()[name = string("op_25258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25258_cast_fp16 = slice_by_index(begin = var_25258_begin_0, end = var_25258_end_0, end_mask = var_25258_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25258_cast_fp16")]; + tensor var_25265_begin_0 = const()[name = string("op_25265_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25265_end_0 = const()[name = string("op_25265_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25265_end_mask_0 = const()[name = string("op_25265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25265_cast_fp16 = slice_by_index(begin = var_25265_begin_0, end = var_25265_end_0, end_mask = var_25265_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25265_cast_fp16")]; + tensor var_25272_begin_0 = const()[name = string("op_25272_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25272_end_0 = const()[name = string("op_25272_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25272_end_mask_0 = const()[name = string("op_25272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25272_cast_fp16 = slice_by_index(begin = var_25272_begin_0, end = var_25272_end_0, end_mask = var_25272_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25272_cast_fp16")]; + tensor var_25279_begin_0 = const()[name = string("op_25279_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25279_end_0 = const()[name = string("op_25279_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25279_end_mask_0 = const()[name = string("op_25279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25279_cast_fp16 = slice_by_index(begin = var_25279_begin_0, end = var_25279_end_0, end_mask = var_25279_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25279_cast_fp16")]; + tensor var_25286_begin_0 = const()[name = string("op_25286_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25286_end_0 = const()[name = string("op_25286_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25286_end_mask_0 = const()[name = string("op_25286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25286_cast_fp16 = slice_by_index(begin = var_25286_begin_0, end = var_25286_end_0, end_mask = var_25286_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25286_cast_fp16")]; + tensor var_25293_begin_0 = const()[name = string("op_25293_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25293_end_0 = const()[name = string("op_25293_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25293_end_mask_0 = const()[name = string("op_25293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25293_cast_fp16 = slice_by_index(begin = var_25293_begin_0, end = var_25293_end_0, end_mask = var_25293_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25293_cast_fp16")]; + tensor var_25300_begin_0 = const()[name = string("op_25300_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25300_end_0 = const()[name = string("op_25300_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25300_end_mask_0 = const()[name = string("op_25300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25300_cast_fp16 = slice_by_index(begin = var_25300_begin_0, end = var_25300_end_0, end_mask = var_25300_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25300_cast_fp16")]; + tensor var_25307_begin_0 = const()[name = string("op_25307_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25307_end_0 = const()[name = string("op_25307_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25307_end_mask_0 = const()[name = string("op_25307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25307_cast_fp16 = slice_by_index(begin = var_25307_begin_0, end = var_25307_end_0, end_mask = var_25307_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25307_cast_fp16")]; + tensor var_25314_begin_0 = const()[name = string("op_25314_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25314_end_0 = const()[name = string("op_25314_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25314_end_mask_0 = const()[name = string("op_25314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25314_cast_fp16 = slice_by_index(begin = var_25314_begin_0, end = var_25314_end_0, end_mask = var_25314_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25314_cast_fp16")]; + tensor var_25321_begin_0 = const()[name = string("op_25321_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25321_end_0 = const()[name = string("op_25321_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25321_end_mask_0 = const()[name = string("op_25321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25321_cast_fp16 = slice_by_index(begin = var_25321_begin_0, end = var_25321_end_0, end_mask = var_25321_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25321_cast_fp16")]; + tensor var_25328_begin_0 = const()[name = string("op_25328_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25328_end_0 = const()[name = string("op_25328_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25328_end_mask_0 = const()[name = string("op_25328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25328_cast_fp16 = slice_by_index(begin = var_25328_begin_0, end = var_25328_end_0, end_mask = var_25328_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25328_cast_fp16")]; + tensor k_33_perm_0 = const()[name = string("k_33_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25333_begin_0 = const()[name = string("op_25333_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25333_end_0 = const()[name = string("op_25333_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_25333_end_mask_0 = const()[name = string("op_25333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_33_cast_fp16 = transpose(perm = k_33_perm_0, x = key_33_cast_fp16)[name = string("transpose_15")]; + tensor var_25333_cast_fp16 = slice_by_index(begin = var_25333_begin_0, end = var_25333_end_0, end_mask = var_25333_end_mask_0, x = k_33_cast_fp16)[name = string("op_25333_cast_fp16")]; + tensor var_25337_begin_0 = const()[name = string("op_25337_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25337_end_0 = const()[name = string("op_25337_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_25337_end_mask_0 = const()[name = string("op_25337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25337_cast_fp16 = slice_by_index(begin = var_25337_begin_0, end = var_25337_end_0, end_mask = var_25337_end_mask_0, x = k_33_cast_fp16)[name = string("op_25337_cast_fp16")]; + tensor var_25341_begin_0 = const()[name = string("op_25341_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25341_end_0 = const()[name = string("op_25341_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_25341_end_mask_0 = const()[name = string("op_25341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25341_cast_fp16 = slice_by_index(begin = var_25341_begin_0, end = var_25341_end_0, end_mask = var_25341_end_mask_0, x = k_33_cast_fp16)[name = string("op_25341_cast_fp16")]; + tensor var_25345_begin_0 = const()[name = string("op_25345_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25345_end_0 = const()[name = string("op_25345_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_25345_end_mask_0 = const()[name = string("op_25345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25345_cast_fp16 = slice_by_index(begin = var_25345_begin_0, end = var_25345_end_0, end_mask = var_25345_end_mask_0, x = k_33_cast_fp16)[name = string("op_25345_cast_fp16")]; + tensor var_25349_begin_0 = const()[name = string("op_25349_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25349_end_0 = const()[name = string("op_25349_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_25349_end_mask_0 = const()[name = string("op_25349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25349_cast_fp16 = slice_by_index(begin = var_25349_begin_0, end = var_25349_end_0, end_mask = var_25349_end_mask_0, x = k_33_cast_fp16)[name = string("op_25349_cast_fp16")]; + tensor var_25353_begin_0 = const()[name = string("op_25353_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25353_end_0 = const()[name = string("op_25353_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_25353_end_mask_0 = const()[name = string("op_25353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25353_cast_fp16 = slice_by_index(begin = var_25353_begin_0, end = var_25353_end_0, end_mask = var_25353_end_mask_0, x = k_33_cast_fp16)[name = string("op_25353_cast_fp16")]; + tensor var_25357_begin_0 = const()[name = string("op_25357_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25357_end_0 = const()[name = string("op_25357_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_25357_end_mask_0 = const()[name = string("op_25357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25357_cast_fp16 = slice_by_index(begin = var_25357_begin_0, end = var_25357_end_0, end_mask = var_25357_end_mask_0, x = k_33_cast_fp16)[name = string("op_25357_cast_fp16")]; + tensor var_25361_begin_0 = const()[name = string("op_25361_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25361_end_0 = const()[name = string("op_25361_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_25361_end_mask_0 = const()[name = string("op_25361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25361_cast_fp16 = slice_by_index(begin = var_25361_begin_0, end = var_25361_end_0, end_mask = var_25361_end_mask_0, x = k_33_cast_fp16)[name = string("op_25361_cast_fp16")]; + tensor var_25365_begin_0 = const()[name = string("op_25365_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25365_end_0 = const()[name = string("op_25365_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_25365_end_mask_0 = const()[name = string("op_25365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25365_cast_fp16 = slice_by_index(begin = var_25365_begin_0, end = var_25365_end_0, end_mask = var_25365_end_mask_0, x = k_33_cast_fp16)[name = string("op_25365_cast_fp16")]; + tensor var_25369_begin_0 = const()[name = string("op_25369_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25369_end_0 = const()[name = string("op_25369_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_25369_end_mask_0 = const()[name = string("op_25369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25369_cast_fp16 = slice_by_index(begin = var_25369_begin_0, end = var_25369_end_0, end_mask = var_25369_end_mask_0, x = k_33_cast_fp16)[name = string("op_25369_cast_fp16")]; + tensor var_25373_begin_0 = const()[name = string("op_25373_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25373_end_0 = const()[name = string("op_25373_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_25373_end_mask_0 = const()[name = string("op_25373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25373_cast_fp16 = slice_by_index(begin = var_25373_begin_0, end = var_25373_end_0, end_mask = var_25373_end_mask_0, x = k_33_cast_fp16)[name = string("op_25373_cast_fp16")]; + tensor var_25377_begin_0 = const()[name = string("op_25377_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25377_end_0 = const()[name = string("op_25377_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_25377_end_mask_0 = const()[name = string("op_25377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25377_cast_fp16 = slice_by_index(begin = var_25377_begin_0, end = var_25377_end_0, end_mask = var_25377_end_mask_0, x = k_33_cast_fp16)[name = string("op_25377_cast_fp16")]; + tensor var_25381_begin_0 = const()[name = string("op_25381_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25381_end_0 = const()[name = string("op_25381_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_25381_end_mask_0 = const()[name = string("op_25381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25381_cast_fp16 = slice_by_index(begin = var_25381_begin_0, end = var_25381_end_0, end_mask = var_25381_end_mask_0, x = k_33_cast_fp16)[name = string("op_25381_cast_fp16")]; + tensor var_25385_begin_0 = const()[name = string("op_25385_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25385_end_0 = const()[name = string("op_25385_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_25385_end_mask_0 = const()[name = string("op_25385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25385_cast_fp16 = slice_by_index(begin = var_25385_begin_0, end = var_25385_end_0, end_mask = var_25385_end_mask_0, x = k_33_cast_fp16)[name = string("op_25385_cast_fp16")]; + tensor var_25389_begin_0 = const()[name = string("op_25389_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25389_end_0 = const()[name = string("op_25389_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_25389_end_mask_0 = const()[name = string("op_25389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25389_cast_fp16 = slice_by_index(begin = var_25389_begin_0, end = var_25389_end_0, end_mask = var_25389_end_mask_0, x = k_33_cast_fp16)[name = string("op_25389_cast_fp16")]; + tensor var_25393_begin_0 = const()[name = string("op_25393_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25393_end_0 = const()[name = string("op_25393_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_25393_end_mask_0 = const()[name = string("op_25393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25393_cast_fp16 = slice_by_index(begin = var_25393_begin_0, end = var_25393_end_0, end_mask = var_25393_end_mask_0, x = k_33_cast_fp16)[name = string("op_25393_cast_fp16")]; + tensor var_25397_begin_0 = const()[name = string("op_25397_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25397_end_0 = const()[name = string("op_25397_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_25397_end_mask_0 = const()[name = string("op_25397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25397_cast_fp16 = slice_by_index(begin = var_25397_begin_0, end = var_25397_end_0, end_mask = var_25397_end_mask_0, x = k_33_cast_fp16)[name = string("op_25397_cast_fp16")]; + tensor var_25401_begin_0 = const()[name = string("op_25401_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25401_end_0 = const()[name = string("op_25401_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_25401_end_mask_0 = const()[name = string("op_25401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25401_cast_fp16 = slice_by_index(begin = var_25401_begin_0, end = var_25401_end_0, end_mask = var_25401_end_mask_0, x = k_33_cast_fp16)[name = string("op_25401_cast_fp16")]; + tensor var_25405_begin_0 = const()[name = string("op_25405_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25405_end_0 = const()[name = string("op_25405_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_25405_end_mask_0 = const()[name = string("op_25405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25405_cast_fp16 = slice_by_index(begin = var_25405_begin_0, end = var_25405_end_0, end_mask = var_25405_end_mask_0, x = k_33_cast_fp16)[name = string("op_25405_cast_fp16")]; + tensor var_25409_begin_0 = const()[name = string("op_25409_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25409_end_0 = const()[name = string("op_25409_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_25409_end_mask_0 = const()[name = string("op_25409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25409_cast_fp16 = slice_by_index(begin = var_25409_begin_0, end = var_25409_end_0, end_mask = var_25409_end_mask_0, x = k_33_cast_fp16)[name = string("op_25409_cast_fp16")]; + tensor var_25411_begin_0 = const()[name = string("op_25411_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25411_end_0 = const()[name = string("op_25411_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25411_end_mask_0 = const()[name = string("op_25411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25411_cast_fp16 = slice_by_index(begin = var_25411_begin_0, end = var_25411_end_0, end_mask = var_25411_end_mask_0, x = value_33_cast_fp16)[name = string("op_25411_cast_fp16")]; + tensor var_25415_begin_0 = const()[name = string("op_25415_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25415_end_0 = const()[name = string("op_25415_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_25415_end_mask_0 = const()[name = string("op_25415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25415_cast_fp16 = slice_by_index(begin = var_25415_begin_0, end = var_25415_end_0, end_mask = var_25415_end_mask_0, x = value_33_cast_fp16)[name = string("op_25415_cast_fp16")]; + tensor var_25419_begin_0 = const()[name = string("op_25419_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25419_end_0 = const()[name = string("op_25419_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_25419_end_mask_0 = const()[name = string("op_25419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25419_cast_fp16 = slice_by_index(begin = var_25419_begin_0, end = var_25419_end_0, end_mask = var_25419_end_mask_0, x = value_33_cast_fp16)[name = string("op_25419_cast_fp16")]; + tensor var_25423_begin_0 = const()[name = string("op_25423_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25423_end_0 = const()[name = string("op_25423_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_25423_end_mask_0 = const()[name = string("op_25423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25423_cast_fp16 = slice_by_index(begin = var_25423_begin_0, end = var_25423_end_0, end_mask = var_25423_end_mask_0, x = value_33_cast_fp16)[name = string("op_25423_cast_fp16")]; + tensor var_25427_begin_0 = const()[name = string("op_25427_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25427_end_0 = const()[name = string("op_25427_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_25427_end_mask_0 = const()[name = string("op_25427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25427_cast_fp16 = slice_by_index(begin = var_25427_begin_0, end = var_25427_end_0, end_mask = var_25427_end_mask_0, x = value_33_cast_fp16)[name = string("op_25427_cast_fp16")]; + tensor var_25431_begin_0 = const()[name = string("op_25431_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25431_end_0 = const()[name = string("op_25431_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_25431_end_mask_0 = const()[name = string("op_25431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25431_cast_fp16 = slice_by_index(begin = var_25431_begin_0, end = var_25431_end_0, end_mask = var_25431_end_mask_0, x = value_33_cast_fp16)[name = string("op_25431_cast_fp16")]; + tensor var_25435_begin_0 = const()[name = string("op_25435_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25435_end_0 = const()[name = string("op_25435_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_25435_end_mask_0 = const()[name = string("op_25435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25435_cast_fp16 = slice_by_index(begin = var_25435_begin_0, end = var_25435_end_0, end_mask = var_25435_end_mask_0, x = value_33_cast_fp16)[name = string("op_25435_cast_fp16")]; + tensor var_25439_begin_0 = const()[name = string("op_25439_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25439_end_0 = const()[name = string("op_25439_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_25439_end_mask_0 = const()[name = string("op_25439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25439_cast_fp16 = slice_by_index(begin = var_25439_begin_0, end = var_25439_end_0, end_mask = var_25439_end_mask_0, x = value_33_cast_fp16)[name = string("op_25439_cast_fp16")]; + tensor var_25443_begin_0 = const()[name = string("op_25443_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25443_end_0 = const()[name = string("op_25443_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_25443_end_mask_0 = const()[name = string("op_25443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25443_cast_fp16 = slice_by_index(begin = var_25443_begin_0, end = var_25443_end_0, end_mask = var_25443_end_mask_0, x = value_33_cast_fp16)[name = string("op_25443_cast_fp16")]; + tensor var_25447_begin_0 = const()[name = string("op_25447_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25447_end_0 = const()[name = string("op_25447_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_25447_end_mask_0 = const()[name = string("op_25447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25447_cast_fp16 = slice_by_index(begin = var_25447_begin_0, end = var_25447_end_0, end_mask = var_25447_end_mask_0, x = value_33_cast_fp16)[name = string("op_25447_cast_fp16")]; + tensor var_25451_begin_0 = const()[name = string("op_25451_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25451_end_0 = const()[name = string("op_25451_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_25451_end_mask_0 = const()[name = string("op_25451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25451_cast_fp16 = slice_by_index(begin = var_25451_begin_0, end = var_25451_end_0, end_mask = var_25451_end_mask_0, x = value_33_cast_fp16)[name = string("op_25451_cast_fp16")]; + tensor var_25455_begin_0 = const()[name = string("op_25455_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25455_end_0 = const()[name = string("op_25455_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_25455_end_mask_0 = const()[name = string("op_25455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25455_cast_fp16 = slice_by_index(begin = var_25455_begin_0, end = var_25455_end_0, end_mask = var_25455_end_mask_0, x = value_33_cast_fp16)[name = string("op_25455_cast_fp16")]; + tensor var_25459_begin_0 = const()[name = string("op_25459_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25459_end_0 = const()[name = string("op_25459_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_25459_end_mask_0 = const()[name = string("op_25459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25459_cast_fp16 = slice_by_index(begin = var_25459_begin_0, end = var_25459_end_0, end_mask = var_25459_end_mask_0, x = value_33_cast_fp16)[name = string("op_25459_cast_fp16")]; + tensor var_25463_begin_0 = const()[name = string("op_25463_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25463_end_0 = const()[name = string("op_25463_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_25463_end_mask_0 = const()[name = string("op_25463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25463_cast_fp16 = slice_by_index(begin = var_25463_begin_0, end = var_25463_end_0, end_mask = var_25463_end_mask_0, x = value_33_cast_fp16)[name = string("op_25463_cast_fp16")]; + tensor var_25467_begin_0 = const()[name = string("op_25467_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25467_end_0 = const()[name = string("op_25467_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_25467_end_mask_0 = const()[name = string("op_25467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25467_cast_fp16 = slice_by_index(begin = var_25467_begin_0, end = var_25467_end_0, end_mask = var_25467_end_mask_0, x = value_33_cast_fp16)[name = string("op_25467_cast_fp16")]; + tensor var_25471_begin_0 = const()[name = string("op_25471_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25471_end_0 = const()[name = string("op_25471_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_25471_end_mask_0 = const()[name = string("op_25471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25471_cast_fp16 = slice_by_index(begin = var_25471_begin_0, end = var_25471_end_0, end_mask = var_25471_end_mask_0, x = value_33_cast_fp16)[name = string("op_25471_cast_fp16")]; + tensor var_25475_begin_0 = const()[name = string("op_25475_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25475_end_0 = const()[name = string("op_25475_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_25475_end_mask_0 = const()[name = string("op_25475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25475_cast_fp16 = slice_by_index(begin = var_25475_begin_0, end = var_25475_end_0, end_mask = var_25475_end_mask_0, x = value_33_cast_fp16)[name = string("op_25475_cast_fp16")]; + tensor var_25479_begin_0 = const()[name = string("op_25479_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25479_end_0 = const()[name = string("op_25479_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_25479_end_mask_0 = const()[name = string("op_25479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25479_cast_fp16 = slice_by_index(begin = var_25479_begin_0, end = var_25479_end_0, end_mask = var_25479_end_mask_0, x = value_33_cast_fp16)[name = string("op_25479_cast_fp16")]; + tensor var_25483_begin_0 = const()[name = string("op_25483_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25483_end_0 = const()[name = string("op_25483_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_25483_end_mask_0 = const()[name = string("op_25483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25483_cast_fp16 = slice_by_index(begin = var_25483_begin_0, end = var_25483_end_0, end_mask = var_25483_end_mask_0, x = value_33_cast_fp16)[name = string("op_25483_cast_fp16")]; + tensor var_25487_begin_0 = const()[name = string("op_25487_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25487_end_0 = const()[name = string("op_25487_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_25487_end_mask_0 = const()[name = string("op_25487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25487_cast_fp16 = slice_by_index(begin = var_25487_begin_0, end = var_25487_end_0, end_mask = var_25487_end_mask_0, x = value_33_cast_fp16)[name = string("op_25487_cast_fp16")]; + string _SplitHeadsQ__mh_w_2561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2561_equation_0, values = (var_25333_cast_fp16, var_24775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2561_cast_fp16")]; + string _SplitHeadsQ__mh_w_2563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2563_equation_0, values = (var_25333_cast_fp16, var_24782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2563_cast_fp16")]; + string _SplitHeadsQ__mh_w_2565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2565_equation_0, values = (var_25333_cast_fp16, var_24789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2565_cast_fp16")]; + string _SplitHeadsQ__mh_w_2567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2567_equation_0, values = (var_25333_cast_fp16, var_24796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2567_cast_fp16")]; + string _SplitHeadsQ__mh_w_2569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2569_equation_0, values = (var_25337_cast_fp16, var_24803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2569_cast_fp16")]; + string _SplitHeadsQ__mh_w_2571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2571_equation_0, values = (var_25337_cast_fp16, var_24810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2571_cast_fp16")]; + string _SplitHeadsQ__mh_w_2573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2573_equation_0, values = (var_25337_cast_fp16, var_24817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2573_cast_fp16")]; + string _SplitHeadsQ__mh_w_2575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2575_equation_0, values = (var_25337_cast_fp16, var_24824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2575_cast_fp16")]; + string _SplitHeadsQ__mh_w_2577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2577_equation_0, values = (var_25341_cast_fp16, var_24831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2577_cast_fp16")]; + string _SplitHeadsQ__mh_w_2579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2579_equation_0, values = (var_25341_cast_fp16, var_24838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2579_cast_fp16")]; + string _SplitHeadsQ__mh_w_2581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2581_equation_0, values = (var_25341_cast_fp16, var_24845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2581_cast_fp16")]; + string _SplitHeadsQ__mh_w_2583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2583_equation_0, values = (var_25341_cast_fp16, var_24852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2583_cast_fp16")]; + string _SplitHeadsQ__mh_w_2585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2585_equation_0, values = (var_25345_cast_fp16, var_24859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2585_cast_fp16")]; + string _SplitHeadsQ__mh_w_2587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2587_equation_0, values = (var_25345_cast_fp16, var_24866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2587_cast_fp16")]; + string _SplitHeadsQ__mh_w_2589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2589_equation_0, values = (var_25345_cast_fp16, var_24873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2589_cast_fp16")]; + string _SplitHeadsQ__mh_w_2591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2591_equation_0, values = (var_25345_cast_fp16, var_24880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2591_cast_fp16")]; + string _SplitHeadsQ__mh_w_2593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2593_equation_0, values = (var_25349_cast_fp16, var_24887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2593_cast_fp16")]; + string _SplitHeadsQ__mh_w_2595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2595_equation_0, values = (var_25349_cast_fp16, var_24894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2595_cast_fp16")]; + string _SplitHeadsQ__mh_w_2597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2597_equation_0, values = (var_25349_cast_fp16, var_24901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2597_cast_fp16")]; + string _SplitHeadsQ__mh_w_2599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2599_equation_0, values = (var_25349_cast_fp16, var_24908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2599_cast_fp16")]; + string _SplitHeadsQ__mh_w_2601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2601_equation_0, values = (var_25353_cast_fp16, var_24915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2601_cast_fp16")]; + string _SplitHeadsQ__mh_w_2603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2603_equation_0, values = (var_25353_cast_fp16, var_24922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2603_cast_fp16")]; + string _SplitHeadsQ__mh_w_2605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2605_equation_0, values = (var_25353_cast_fp16, var_24929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2605_cast_fp16")]; + string _SplitHeadsQ__mh_w_2607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2607_equation_0, values = (var_25353_cast_fp16, var_24936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2607_cast_fp16")]; + string _SplitHeadsQ__mh_w_2609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2609_equation_0, values = (var_25357_cast_fp16, var_24943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2609_cast_fp16")]; + string _SplitHeadsQ__mh_w_2611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2611_equation_0, values = (var_25357_cast_fp16, var_24950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2611_cast_fp16")]; + string _SplitHeadsQ__mh_w_2613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2613_equation_0, values = (var_25357_cast_fp16, var_24957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2613_cast_fp16")]; + string _SplitHeadsQ__mh_w_2615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2615_equation_0, values = (var_25357_cast_fp16, var_24964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2615_cast_fp16")]; + string _SplitHeadsQ__mh_w_2617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2617_equation_0, values = (var_25361_cast_fp16, var_24971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2617_cast_fp16")]; + string _SplitHeadsQ__mh_w_2619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2619_equation_0, values = (var_25361_cast_fp16, var_24978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2619_cast_fp16")]; + string _SplitHeadsQ__mh_w_2621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2621_equation_0, values = (var_25361_cast_fp16, var_24985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2621_cast_fp16")]; + string _SplitHeadsQ__mh_w_2623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2623_equation_0, values = (var_25361_cast_fp16, var_24992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2623_cast_fp16")]; + string _SplitHeadsQ__mh_w_2625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2625_equation_0, values = (var_25365_cast_fp16, var_24999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2625_cast_fp16")]; + string _SplitHeadsQ__mh_w_2627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2627_equation_0, values = (var_25365_cast_fp16, var_25006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2627_cast_fp16")]; + string _SplitHeadsQ__mh_w_2629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2629_equation_0, values = (var_25365_cast_fp16, var_25013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2629_cast_fp16")]; + string _SplitHeadsQ__mh_w_2631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2631_equation_0, values = (var_25365_cast_fp16, var_25020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2631_cast_fp16")]; + string _SplitHeadsQ__mh_w_2633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2633_equation_0, values = (var_25369_cast_fp16, var_25027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2633_cast_fp16")]; + string _SplitHeadsQ__mh_w_2635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2635_equation_0, values = (var_25369_cast_fp16, var_25034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2635_cast_fp16")]; + string _SplitHeadsQ__mh_w_2637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2637_equation_0, values = (var_25369_cast_fp16, var_25041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2637_cast_fp16")]; + string _SplitHeadsQ__mh_w_2639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2639_equation_0, values = (var_25369_cast_fp16, var_25048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2639_cast_fp16")]; + string _SplitHeadsQ__mh_w_2641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2641_equation_0, values = (var_25373_cast_fp16, var_25055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2641_cast_fp16")]; + string _SplitHeadsQ__mh_w_2643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2643_equation_0, values = (var_25373_cast_fp16, var_25062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2643_cast_fp16")]; + string _SplitHeadsQ__mh_w_2645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2645_equation_0, values = (var_25373_cast_fp16, var_25069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2645_cast_fp16")]; + string _SplitHeadsQ__mh_w_2647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2647_equation_0, values = (var_25373_cast_fp16, var_25076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2647_cast_fp16")]; + string _SplitHeadsQ__mh_w_2649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2649_equation_0, values = (var_25377_cast_fp16, var_25083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2649_cast_fp16")]; + string _SplitHeadsQ__mh_w_2651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2651_equation_0, values = (var_25377_cast_fp16, var_25090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2651_cast_fp16")]; + string _SplitHeadsQ__mh_w_2653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2653_equation_0, values = (var_25377_cast_fp16, var_25097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2653_cast_fp16")]; + string _SplitHeadsQ__mh_w_2655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2655_equation_0, values = (var_25377_cast_fp16, var_25104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2655_cast_fp16")]; + string _SplitHeadsQ__mh_w_2657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2657_equation_0, values = (var_25381_cast_fp16, var_25111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2657_cast_fp16")]; + string _SplitHeadsQ__mh_w_2659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2659_equation_0, values = (var_25381_cast_fp16, var_25118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2659_cast_fp16")]; + string _SplitHeadsQ__mh_w_2661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2661_equation_0, values = (var_25381_cast_fp16, var_25125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2661_cast_fp16")]; + string _SplitHeadsQ__mh_w_2663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2663_equation_0, values = (var_25381_cast_fp16, var_25132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2663_cast_fp16")]; + string _SplitHeadsQ__mh_w_2665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2665_equation_0, values = (var_25385_cast_fp16, var_25139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2665_cast_fp16")]; + string _SplitHeadsQ__mh_w_2667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2667_equation_0, values = (var_25385_cast_fp16, var_25146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2667_cast_fp16")]; + string _SplitHeadsQ__mh_w_2669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2669_equation_0, values = (var_25385_cast_fp16, var_25153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2669_cast_fp16")]; + string _SplitHeadsQ__mh_w_2671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2671_equation_0, values = (var_25385_cast_fp16, var_25160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2671_cast_fp16")]; + string _SplitHeadsQ__mh_w_2673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2673_equation_0, values = (var_25389_cast_fp16, var_25167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2673_cast_fp16")]; + string _SplitHeadsQ__mh_w_2675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2675_equation_0, values = (var_25389_cast_fp16, var_25174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2675_cast_fp16")]; + string _SplitHeadsQ__mh_w_2677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2677_equation_0, values = (var_25389_cast_fp16, var_25181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2677_cast_fp16")]; + string _SplitHeadsQ__mh_w_2679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2679_equation_0, values = (var_25389_cast_fp16, var_25188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2679_cast_fp16")]; + string _SplitHeadsQ__mh_w_2681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2681_equation_0, values = (var_25393_cast_fp16, var_25195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2681_cast_fp16")]; + string _SplitHeadsQ__mh_w_2683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2683_equation_0, values = (var_25393_cast_fp16, var_25202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2683_cast_fp16")]; + string _SplitHeadsQ__mh_w_2685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2685_equation_0, values = (var_25393_cast_fp16, var_25209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2685_cast_fp16")]; + string _SplitHeadsQ__mh_w_2687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2687_equation_0, values = (var_25393_cast_fp16, var_25216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2687_cast_fp16")]; + string _SplitHeadsQ__mh_w_2689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2689_equation_0, values = (var_25397_cast_fp16, var_25223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2689_cast_fp16")]; + string _SplitHeadsQ__mh_w_2691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2691_equation_0, values = (var_25397_cast_fp16, var_25230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2691_cast_fp16")]; + string _SplitHeadsQ__mh_w_2693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2693_equation_0, values = (var_25397_cast_fp16, var_25237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2693_cast_fp16")]; + string _SplitHeadsQ__mh_w_2695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2695_equation_0, values = (var_25397_cast_fp16, var_25244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2695_cast_fp16")]; + string _SplitHeadsQ__mh_w_2697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2697_equation_0, values = (var_25401_cast_fp16, var_25251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2697_cast_fp16")]; + string _SplitHeadsQ__mh_w_2699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2699_equation_0, values = (var_25401_cast_fp16, var_25258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2699_cast_fp16")]; + string _SplitHeadsQ__mh_w_2701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2701_equation_0, values = (var_25401_cast_fp16, var_25265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2701_cast_fp16")]; + string _SplitHeadsQ__mh_w_2703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2703_equation_0, values = (var_25401_cast_fp16, var_25272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2703_cast_fp16")]; + string _SplitHeadsQ__mh_w_2705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2705_equation_0, values = (var_25405_cast_fp16, var_25279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2705_cast_fp16")]; + string _SplitHeadsQ__mh_w_2707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2707_equation_0, values = (var_25405_cast_fp16, var_25286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2707_cast_fp16")]; + string _SplitHeadsQ__mh_w_2709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2709_equation_0, values = (var_25405_cast_fp16, var_25293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2709_cast_fp16")]; + string _SplitHeadsQ__mh_w_2711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2711_equation_0, values = (var_25405_cast_fp16, var_25300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2711_cast_fp16")]; + string _SplitHeadsQ__mh_w_2713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2713_equation_0, values = (var_25409_cast_fp16, var_25307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2713_cast_fp16")]; + string _SplitHeadsQ__mh_w_2715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2715_equation_0, values = (var_25409_cast_fp16, var_25314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2715_cast_fp16")]; + string _SplitHeadsQ__mh_w_2717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2717_equation_0, values = (var_25409_cast_fp16, var_25321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2717_cast_fp16")]; + string _SplitHeadsQ__mh_w_2719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2719_equation_0, values = (var_25409_cast_fp16, var_25328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2719_cast_fp16")]; + fp16 var_25650_to_fp16 = const()[name = string("op_25650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2561_cast_fp16, y = var_25650_to_fp16)[name = string("aw_chunk_2561_cast_fp16")]; + fp16 var_25652_to_fp16 = const()[name = string("op_25652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2563_cast_fp16, y = var_25652_to_fp16)[name = string("aw_chunk_2563_cast_fp16")]; + fp16 var_25654_to_fp16 = const()[name = string("op_25654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2565_cast_fp16, y = var_25654_to_fp16)[name = string("aw_chunk_2565_cast_fp16")]; + fp16 var_25656_to_fp16 = const()[name = string("op_25656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2567_cast_fp16, y = var_25656_to_fp16)[name = string("aw_chunk_2567_cast_fp16")]; + fp16 var_25658_to_fp16 = const()[name = string("op_25658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2569_cast_fp16, y = var_25658_to_fp16)[name = string("aw_chunk_2569_cast_fp16")]; + fp16 var_25660_to_fp16 = const()[name = string("op_25660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2571_cast_fp16, y = var_25660_to_fp16)[name = string("aw_chunk_2571_cast_fp16")]; + fp16 var_25662_to_fp16 = const()[name = string("op_25662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2573_cast_fp16, y = var_25662_to_fp16)[name = string("aw_chunk_2573_cast_fp16")]; + fp16 var_25664_to_fp16 = const()[name = string("op_25664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2575_cast_fp16, y = var_25664_to_fp16)[name = string("aw_chunk_2575_cast_fp16")]; + fp16 var_25666_to_fp16 = const()[name = string("op_25666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2577_cast_fp16, y = var_25666_to_fp16)[name = string("aw_chunk_2577_cast_fp16")]; + fp16 var_25668_to_fp16 = const()[name = string("op_25668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2579_cast_fp16, y = var_25668_to_fp16)[name = string("aw_chunk_2579_cast_fp16")]; + fp16 var_25670_to_fp16 = const()[name = string("op_25670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2581_cast_fp16, y = var_25670_to_fp16)[name = string("aw_chunk_2581_cast_fp16")]; + fp16 var_25672_to_fp16 = const()[name = string("op_25672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2583_cast_fp16, y = var_25672_to_fp16)[name = string("aw_chunk_2583_cast_fp16")]; + fp16 var_25674_to_fp16 = const()[name = string("op_25674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2585_cast_fp16, y = var_25674_to_fp16)[name = string("aw_chunk_2585_cast_fp16")]; + fp16 var_25676_to_fp16 = const()[name = string("op_25676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2587_cast_fp16, y = var_25676_to_fp16)[name = string("aw_chunk_2587_cast_fp16")]; + fp16 var_25678_to_fp16 = const()[name = string("op_25678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2589_cast_fp16, y = var_25678_to_fp16)[name = string("aw_chunk_2589_cast_fp16")]; + fp16 var_25680_to_fp16 = const()[name = string("op_25680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2591_cast_fp16, y = var_25680_to_fp16)[name = string("aw_chunk_2591_cast_fp16")]; + fp16 var_25682_to_fp16 = const()[name = string("op_25682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2593_cast_fp16, y = var_25682_to_fp16)[name = string("aw_chunk_2593_cast_fp16")]; + fp16 var_25684_to_fp16 = const()[name = string("op_25684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2595_cast_fp16, y = var_25684_to_fp16)[name = string("aw_chunk_2595_cast_fp16")]; + fp16 var_25686_to_fp16 = const()[name = string("op_25686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2597_cast_fp16, y = var_25686_to_fp16)[name = string("aw_chunk_2597_cast_fp16")]; + fp16 var_25688_to_fp16 = const()[name = string("op_25688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2599_cast_fp16, y = var_25688_to_fp16)[name = string("aw_chunk_2599_cast_fp16")]; + fp16 var_25690_to_fp16 = const()[name = string("op_25690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2601_cast_fp16, y = var_25690_to_fp16)[name = string("aw_chunk_2601_cast_fp16")]; + fp16 var_25692_to_fp16 = const()[name = string("op_25692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2603_cast_fp16, y = var_25692_to_fp16)[name = string("aw_chunk_2603_cast_fp16")]; + fp16 var_25694_to_fp16 = const()[name = string("op_25694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2605_cast_fp16, y = var_25694_to_fp16)[name = string("aw_chunk_2605_cast_fp16")]; + fp16 var_25696_to_fp16 = const()[name = string("op_25696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2607_cast_fp16, y = var_25696_to_fp16)[name = string("aw_chunk_2607_cast_fp16")]; + fp16 var_25698_to_fp16 = const()[name = string("op_25698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2609_cast_fp16, y = var_25698_to_fp16)[name = string("aw_chunk_2609_cast_fp16")]; + fp16 var_25700_to_fp16 = const()[name = string("op_25700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2611_cast_fp16, y = var_25700_to_fp16)[name = string("aw_chunk_2611_cast_fp16")]; + fp16 var_25702_to_fp16 = const()[name = string("op_25702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2613_cast_fp16, y = var_25702_to_fp16)[name = string("aw_chunk_2613_cast_fp16")]; + fp16 var_25704_to_fp16 = const()[name = string("op_25704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2615_cast_fp16, y = var_25704_to_fp16)[name = string("aw_chunk_2615_cast_fp16")]; + fp16 var_25706_to_fp16 = const()[name = string("op_25706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2617_cast_fp16, y = var_25706_to_fp16)[name = string("aw_chunk_2617_cast_fp16")]; + fp16 var_25708_to_fp16 = const()[name = string("op_25708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2619_cast_fp16, y = var_25708_to_fp16)[name = string("aw_chunk_2619_cast_fp16")]; + fp16 var_25710_to_fp16 = const()[name = string("op_25710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2621_cast_fp16, y = var_25710_to_fp16)[name = string("aw_chunk_2621_cast_fp16")]; + fp16 var_25712_to_fp16 = const()[name = string("op_25712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2623_cast_fp16, y = var_25712_to_fp16)[name = string("aw_chunk_2623_cast_fp16")]; + fp16 var_25714_to_fp16 = const()[name = string("op_25714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2625_cast_fp16, y = var_25714_to_fp16)[name = string("aw_chunk_2625_cast_fp16")]; + fp16 var_25716_to_fp16 = const()[name = string("op_25716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2627_cast_fp16, y = var_25716_to_fp16)[name = string("aw_chunk_2627_cast_fp16")]; + fp16 var_25718_to_fp16 = const()[name = string("op_25718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2629_cast_fp16, y = var_25718_to_fp16)[name = string("aw_chunk_2629_cast_fp16")]; + fp16 var_25720_to_fp16 = const()[name = string("op_25720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2631_cast_fp16, y = var_25720_to_fp16)[name = string("aw_chunk_2631_cast_fp16")]; + fp16 var_25722_to_fp16 = const()[name = string("op_25722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2633_cast_fp16, y = var_25722_to_fp16)[name = string("aw_chunk_2633_cast_fp16")]; + fp16 var_25724_to_fp16 = const()[name = string("op_25724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2635_cast_fp16, y = var_25724_to_fp16)[name = string("aw_chunk_2635_cast_fp16")]; + fp16 var_25726_to_fp16 = const()[name = string("op_25726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2637_cast_fp16, y = var_25726_to_fp16)[name = string("aw_chunk_2637_cast_fp16")]; + fp16 var_25728_to_fp16 = const()[name = string("op_25728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2639_cast_fp16, y = var_25728_to_fp16)[name = string("aw_chunk_2639_cast_fp16")]; + fp16 var_25730_to_fp16 = const()[name = string("op_25730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2641_cast_fp16, y = var_25730_to_fp16)[name = string("aw_chunk_2641_cast_fp16")]; + fp16 var_25732_to_fp16 = const()[name = string("op_25732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2643_cast_fp16, y = var_25732_to_fp16)[name = string("aw_chunk_2643_cast_fp16")]; + fp16 var_25734_to_fp16 = const()[name = string("op_25734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2645_cast_fp16, y = var_25734_to_fp16)[name = string("aw_chunk_2645_cast_fp16")]; + fp16 var_25736_to_fp16 = const()[name = string("op_25736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2647_cast_fp16, y = var_25736_to_fp16)[name = string("aw_chunk_2647_cast_fp16")]; + fp16 var_25738_to_fp16 = const()[name = string("op_25738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2649_cast_fp16, y = var_25738_to_fp16)[name = string("aw_chunk_2649_cast_fp16")]; + fp16 var_25740_to_fp16 = const()[name = string("op_25740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2651_cast_fp16, y = var_25740_to_fp16)[name = string("aw_chunk_2651_cast_fp16")]; + fp16 var_25742_to_fp16 = const()[name = string("op_25742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2653_cast_fp16, y = var_25742_to_fp16)[name = string("aw_chunk_2653_cast_fp16")]; + fp16 var_25744_to_fp16 = const()[name = string("op_25744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2655_cast_fp16, y = var_25744_to_fp16)[name = string("aw_chunk_2655_cast_fp16")]; + fp16 var_25746_to_fp16 = const()[name = string("op_25746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2657_cast_fp16, y = var_25746_to_fp16)[name = string("aw_chunk_2657_cast_fp16")]; + fp16 var_25748_to_fp16 = const()[name = string("op_25748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2659_cast_fp16, y = var_25748_to_fp16)[name = string("aw_chunk_2659_cast_fp16")]; + fp16 var_25750_to_fp16 = const()[name = string("op_25750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2661_cast_fp16, y = var_25750_to_fp16)[name = string("aw_chunk_2661_cast_fp16")]; + fp16 var_25752_to_fp16 = const()[name = string("op_25752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2663_cast_fp16, y = var_25752_to_fp16)[name = string("aw_chunk_2663_cast_fp16")]; + fp16 var_25754_to_fp16 = const()[name = string("op_25754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2665_cast_fp16, y = var_25754_to_fp16)[name = string("aw_chunk_2665_cast_fp16")]; + fp16 var_25756_to_fp16 = const()[name = string("op_25756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2667_cast_fp16, y = var_25756_to_fp16)[name = string("aw_chunk_2667_cast_fp16")]; + fp16 var_25758_to_fp16 = const()[name = string("op_25758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2669_cast_fp16, y = var_25758_to_fp16)[name = string("aw_chunk_2669_cast_fp16")]; + fp16 var_25760_to_fp16 = const()[name = string("op_25760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2671_cast_fp16, y = var_25760_to_fp16)[name = string("aw_chunk_2671_cast_fp16")]; + fp16 var_25762_to_fp16 = const()[name = string("op_25762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2673_cast_fp16, y = var_25762_to_fp16)[name = string("aw_chunk_2673_cast_fp16")]; + fp16 var_25764_to_fp16 = const()[name = string("op_25764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2675_cast_fp16, y = var_25764_to_fp16)[name = string("aw_chunk_2675_cast_fp16")]; + fp16 var_25766_to_fp16 = const()[name = string("op_25766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2677_cast_fp16, y = var_25766_to_fp16)[name = string("aw_chunk_2677_cast_fp16")]; + fp16 var_25768_to_fp16 = const()[name = string("op_25768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2679_cast_fp16, y = var_25768_to_fp16)[name = string("aw_chunk_2679_cast_fp16")]; + fp16 var_25770_to_fp16 = const()[name = string("op_25770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2681_cast_fp16, y = var_25770_to_fp16)[name = string("aw_chunk_2681_cast_fp16")]; + fp16 var_25772_to_fp16 = const()[name = string("op_25772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2683_cast_fp16, y = var_25772_to_fp16)[name = string("aw_chunk_2683_cast_fp16")]; + fp16 var_25774_to_fp16 = const()[name = string("op_25774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2685_cast_fp16, y = var_25774_to_fp16)[name = string("aw_chunk_2685_cast_fp16")]; + fp16 var_25776_to_fp16 = const()[name = string("op_25776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2687_cast_fp16, y = var_25776_to_fp16)[name = string("aw_chunk_2687_cast_fp16")]; + fp16 var_25778_to_fp16 = const()[name = string("op_25778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2689_cast_fp16, y = var_25778_to_fp16)[name = string("aw_chunk_2689_cast_fp16")]; + fp16 var_25780_to_fp16 = const()[name = string("op_25780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2691_cast_fp16, y = var_25780_to_fp16)[name = string("aw_chunk_2691_cast_fp16")]; + fp16 var_25782_to_fp16 = const()[name = string("op_25782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2693_cast_fp16, y = var_25782_to_fp16)[name = string("aw_chunk_2693_cast_fp16")]; + fp16 var_25784_to_fp16 = const()[name = string("op_25784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2695_cast_fp16, y = var_25784_to_fp16)[name = string("aw_chunk_2695_cast_fp16")]; + fp16 var_25786_to_fp16 = const()[name = string("op_25786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2697_cast_fp16, y = var_25786_to_fp16)[name = string("aw_chunk_2697_cast_fp16")]; + fp16 var_25788_to_fp16 = const()[name = string("op_25788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2699_cast_fp16, y = var_25788_to_fp16)[name = string("aw_chunk_2699_cast_fp16")]; + fp16 var_25790_to_fp16 = const()[name = string("op_25790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2701_cast_fp16, y = var_25790_to_fp16)[name = string("aw_chunk_2701_cast_fp16")]; + fp16 var_25792_to_fp16 = const()[name = string("op_25792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2703_cast_fp16, y = var_25792_to_fp16)[name = string("aw_chunk_2703_cast_fp16")]; + fp16 var_25794_to_fp16 = const()[name = string("op_25794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2705_cast_fp16, y = var_25794_to_fp16)[name = string("aw_chunk_2705_cast_fp16")]; + fp16 var_25796_to_fp16 = const()[name = string("op_25796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2707_cast_fp16, y = var_25796_to_fp16)[name = string("aw_chunk_2707_cast_fp16")]; + fp16 var_25798_to_fp16 = const()[name = string("op_25798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2709_cast_fp16, y = var_25798_to_fp16)[name = string("aw_chunk_2709_cast_fp16")]; + fp16 var_25800_to_fp16 = const()[name = string("op_25800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2711_cast_fp16, y = var_25800_to_fp16)[name = string("aw_chunk_2711_cast_fp16")]; + fp16 var_25802_to_fp16 = const()[name = string("op_25802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2713_cast_fp16, y = var_25802_to_fp16)[name = string("aw_chunk_2713_cast_fp16")]; + fp16 var_25804_to_fp16 = const()[name = string("op_25804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2715_cast_fp16, y = var_25804_to_fp16)[name = string("aw_chunk_2715_cast_fp16")]; + fp16 var_25806_to_fp16 = const()[name = string("op_25806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2717_cast_fp16, y = var_25806_to_fp16)[name = string("aw_chunk_2717_cast_fp16")]; + fp16 var_25808_to_fp16 = const()[name = string("op_25808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2719_cast_fp16, y = var_25808_to_fp16)[name = string("aw_chunk_2719_cast_fp16")]; + tensor var_25810_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2561_cast_fp16)[name = string("op_25810_cast_fp16")]; + tensor var_25811_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2563_cast_fp16)[name = string("op_25811_cast_fp16")]; + tensor var_25812_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2565_cast_fp16)[name = string("op_25812_cast_fp16")]; + tensor var_25813_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2567_cast_fp16)[name = string("op_25813_cast_fp16")]; + tensor var_25814_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2569_cast_fp16)[name = string("op_25814_cast_fp16")]; + tensor var_25815_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2571_cast_fp16)[name = string("op_25815_cast_fp16")]; + tensor var_25816_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2573_cast_fp16)[name = string("op_25816_cast_fp16")]; + tensor var_25817_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2575_cast_fp16)[name = string("op_25817_cast_fp16")]; + tensor var_25818_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2577_cast_fp16)[name = string("op_25818_cast_fp16")]; + tensor var_25819_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2579_cast_fp16)[name = string("op_25819_cast_fp16")]; + tensor var_25820_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2581_cast_fp16)[name = string("op_25820_cast_fp16")]; + tensor var_25821_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2583_cast_fp16)[name = string("op_25821_cast_fp16")]; + tensor var_25822_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2585_cast_fp16)[name = string("op_25822_cast_fp16")]; + tensor var_25823_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2587_cast_fp16)[name = string("op_25823_cast_fp16")]; + tensor var_25824_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2589_cast_fp16)[name = string("op_25824_cast_fp16")]; + tensor var_25825_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2591_cast_fp16)[name = string("op_25825_cast_fp16")]; + tensor var_25826_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2593_cast_fp16)[name = string("op_25826_cast_fp16")]; + tensor var_25827_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2595_cast_fp16)[name = string("op_25827_cast_fp16")]; + tensor var_25828_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2597_cast_fp16)[name = string("op_25828_cast_fp16")]; + tensor var_25829_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2599_cast_fp16)[name = string("op_25829_cast_fp16")]; + tensor var_25830_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2601_cast_fp16)[name = string("op_25830_cast_fp16")]; + tensor var_25831_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2603_cast_fp16)[name = string("op_25831_cast_fp16")]; + tensor var_25832_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2605_cast_fp16)[name = string("op_25832_cast_fp16")]; + tensor var_25833_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2607_cast_fp16)[name = string("op_25833_cast_fp16")]; + tensor var_25834_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2609_cast_fp16)[name = string("op_25834_cast_fp16")]; + tensor var_25835_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2611_cast_fp16)[name = string("op_25835_cast_fp16")]; + tensor var_25836_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2613_cast_fp16)[name = string("op_25836_cast_fp16")]; + tensor var_25837_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2615_cast_fp16)[name = string("op_25837_cast_fp16")]; + tensor var_25838_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2617_cast_fp16)[name = string("op_25838_cast_fp16")]; + tensor var_25839_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2619_cast_fp16)[name = string("op_25839_cast_fp16")]; + tensor var_25840_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2621_cast_fp16)[name = string("op_25840_cast_fp16")]; + tensor var_25841_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2623_cast_fp16)[name = string("op_25841_cast_fp16")]; + tensor var_25842_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2625_cast_fp16)[name = string("op_25842_cast_fp16")]; + tensor var_25843_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2627_cast_fp16)[name = string("op_25843_cast_fp16")]; + tensor var_25844_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2629_cast_fp16)[name = string("op_25844_cast_fp16")]; + tensor var_25845_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2631_cast_fp16)[name = string("op_25845_cast_fp16")]; + tensor var_25846_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2633_cast_fp16)[name = string("op_25846_cast_fp16")]; + tensor var_25847_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2635_cast_fp16)[name = string("op_25847_cast_fp16")]; + tensor var_25848_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2637_cast_fp16)[name = string("op_25848_cast_fp16")]; + tensor var_25849_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2639_cast_fp16)[name = string("op_25849_cast_fp16")]; + tensor var_25850_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2641_cast_fp16)[name = string("op_25850_cast_fp16")]; + tensor var_25851_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2643_cast_fp16)[name = string("op_25851_cast_fp16")]; + tensor var_25852_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2645_cast_fp16)[name = string("op_25852_cast_fp16")]; + tensor var_25853_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2647_cast_fp16)[name = string("op_25853_cast_fp16")]; + tensor var_25854_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2649_cast_fp16)[name = string("op_25854_cast_fp16")]; + tensor var_25855_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2651_cast_fp16)[name = string("op_25855_cast_fp16")]; + tensor var_25856_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2653_cast_fp16)[name = string("op_25856_cast_fp16")]; + tensor var_25857_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2655_cast_fp16)[name = string("op_25857_cast_fp16")]; + tensor var_25858_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2657_cast_fp16)[name = string("op_25858_cast_fp16")]; + tensor var_25859_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2659_cast_fp16)[name = string("op_25859_cast_fp16")]; + tensor var_25860_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2661_cast_fp16)[name = string("op_25860_cast_fp16")]; + tensor var_25861_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2663_cast_fp16)[name = string("op_25861_cast_fp16")]; + tensor var_25862_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2665_cast_fp16)[name = string("op_25862_cast_fp16")]; + tensor var_25863_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2667_cast_fp16)[name = string("op_25863_cast_fp16")]; + tensor var_25864_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2669_cast_fp16)[name = string("op_25864_cast_fp16")]; + tensor var_25865_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2671_cast_fp16)[name = string("op_25865_cast_fp16")]; + tensor var_25866_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2673_cast_fp16)[name = string("op_25866_cast_fp16")]; + tensor var_25867_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2675_cast_fp16)[name = string("op_25867_cast_fp16")]; + tensor var_25868_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2677_cast_fp16)[name = string("op_25868_cast_fp16")]; + tensor var_25869_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2679_cast_fp16)[name = string("op_25869_cast_fp16")]; + tensor var_25870_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2681_cast_fp16)[name = string("op_25870_cast_fp16")]; + tensor var_25871_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2683_cast_fp16)[name = string("op_25871_cast_fp16")]; + tensor var_25872_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2685_cast_fp16)[name = string("op_25872_cast_fp16")]; + tensor var_25873_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2687_cast_fp16)[name = string("op_25873_cast_fp16")]; + tensor var_25874_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2689_cast_fp16)[name = string("op_25874_cast_fp16")]; + tensor var_25875_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2691_cast_fp16)[name = string("op_25875_cast_fp16")]; + tensor var_25876_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2693_cast_fp16)[name = string("op_25876_cast_fp16")]; + tensor var_25877_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2695_cast_fp16)[name = string("op_25877_cast_fp16")]; + tensor var_25878_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2697_cast_fp16)[name = string("op_25878_cast_fp16")]; + tensor var_25879_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2699_cast_fp16)[name = string("op_25879_cast_fp16")]; + tensor var_25880_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2701_cast_fp16)[name = string("op_25880_cast_fp16")]; + tensor var_25881_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2703_cast_fp16)[name = string("op_25881_cast_fp16")]; + tensor var_25882_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2705_cast_fp16)[name = string("op_25882_cast_fp16")]; + tensor var_25883_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2707_cast_fp16)[name = string("op_25883_cast_fp16")]; + tensor var_25884_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2709_cast_fp16)[name = string("op_25884_cast_fp16")]; + tensor var_25885_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2711_cast_fp16)[name = string("op_25885_cast_fp16")]; + tensor var_25886_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2713_cast_fp16)[name = string("op_25886_cast_fp16")]; + tensor var_25887_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2715_cast_fp16)[name = string("op_25887_cast_fp16")]; + tensor var_25888_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2717_cast_fp16)[name = string("op_25888_cast_fp16")]; + tensor var_25889_cast_fp16 = softmax(axis = var_24635, x = aw_chunk_2719_cast_fp16)[name = string("op_25889_cast_fp16")]; + string var_25891_equation_0 = const()[name = string("op_25891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25891_cast_fp16 = einsum(equation = var_25891_equation_0, values = (var_25411_cast_fp16, var_25810_cast_fp16))[name = string("op_25891_cast_fp16")]; + string var_25893_equation_0 = const()[name = string("op_25893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25893_cast_fp16 = einsum(equation = var_25893_equation_0, values = (var_25411_cast_fp16, var_25811_cast_fp16))[name = string("op_25893_cast_fp16")]; + string var_25895_equation_0 = const()[name = string("op_25895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25895_cast_fp16 = einsum(equation = var_25895_equation_0, values = (var_25411_cast_fp16, var_25812_cast_fp16))[name = string("op_25895_cast_fp16")]; + string var_25897_equation_0 = const()[name = string("op_25897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25897_cast_fp16 = einsum(equation = var_25897_equation_0, values = (var_25411_cast_fp16, var_25813_cast_fp16))[name = string("op_25897_cast_fp16")]; + string var_25899_equation_0 = const()[name = string("op_25899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25899_cast_fp16 = einsum(equation = var_25899_equation_0, values = (var_25415_cast_fp16, var_25814_cast_fp16))[name = string("op_25899_cast_fp16")]; + string var_25901_equation_0 = const()[name = string("op_25901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25901_cast_fp16 = einsum(equation = var_25901_equation_0, values = (var_25415_cast_fp16, var_25815_cast_fp16))[name = string("op_25901_cast_fp16")]; + string var_25903_equation_0 = const()[name = string("op_25903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25903_cast_fp16 = einsum(equation = var_25903_equation_0, values = (var_25415_cast_fp16, var_25816_cast_fp16))[name = string("op_25903_cast_fp16")]; + string var_25905_equation_0 = const()[name = string("op_25905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25905_cast_fp16 = einsum(equation = var_25905_equation_0, values = (var_25415_cast_fp16, var_25817_cast_fp16))[name = string("op_25905_cast_fp16")]; + string var_25907_equation_0 = const()[name = string("op_25907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25907_cast_fp16 = einsum(equation = var_25907_equation_0, values = (var_25419_cast_fp16, var_25818_cast_fp16))[name = string("op_25907_cast_fp16")]; + string var_25909_equation_0 = const()[name = string("op_25909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25909_cast_fp16 = einsum(equation = var_25909_equation_0, values = (var_25419_cast_fp16, var_25819_cast_fp16))[name = string("op_25909_cast_fp16")]; + string var_25911_equation_0 = const()[name = string("op_25911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25911_cast_fp16 = einsum(equation = var_25911_equation_0, values = (var_25419_cast_fp16, var_25820_cast_fp16))[name = string("op_25911_cast_fp16")]; + string var_25913_equation_0 = const()[name = string("op_25913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25913_cast_fp16 = einsum(equation = var_25913_equation_0, values = (var_25419_cast_fp16, var_25821_cast_fp16))[name = string("op_25913_cast_fp16")]; + string var_25915_equation_0 = const()[name = string("op_25915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25915_cast_fp16 = einsum(equation = var_25915_equation_0, values = (var_25423_cast_fp16, var_25822_cast_fp16))[name = string("op_25915_cast_fp16")]; + string var_25917_equation_0 = const()[name = string("op_25917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25917_cast_fp16 = einsum(equation = var_25917_equation_0, values = (var_25423_cast_fp16, var_25823_cast_fp16))[name = string("op_25917_cast_fp16")]; + string var_25919_equation_0 = const()[name = string("op_25919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25919_cast_fp16 = einsum(equation = var_25919_equation_0, values = (var_25423_cast_fp16, var_25824_cast_fp16))[name = string("op_25919_cast_fp16")]; + string var_25921_equation_0 = const()[name = string("op_25921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25921_cast_fp16 = einsum(equation = var_25921_equation_0, values = (var_25423_cast_fp16, var_25825_cast_fp16))[name = string("op_25921_cast_fp16")]; + string var_25923_equation_0 = const()[name = string("op_25923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25923_cast_fp16 = einsum(equation = var_25923_equation_0, values = (var_25427_cast_fp16, var_25826_cast_fp16))[name = string("op_25923_cast_fp16")]; + string var_25925_equation_0 = const()[name = string("op_25925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25925_cast_fp16 = einsum(equation = var_25925_equation_0, values = (var_25427_cast_fp16, var_25827_cast_fp16))[name = string("op_25925_cast_fp16")]; + string var_25927_equation_0 = const()[name = string("op_25927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25927_cast_fp16 = einsum(equation = var_25927_equation_0, values = (var_25427_cast_fp16, var_25828_cast_fp16))[name = string("op_25927_cast_fp16")]; + string var_25929_equation_0 = const()[name = string("op_25929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25929_cast_fp16 = einsum(equation = var_25929_equation_0, values = (var_25427_cast_fp16, var_25829_cast_fp16))[name = string("op_25929_cast_fp16")]; + string var_25931_equation_0 = const()[name = string("op_25931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25931_cast_fp16 = einsum(equation = var_25931_equation_0, values = (var_25431_cast_fp16, var_25830_cast_fp16))[name = string("op_25931_cast_fp16")]; + string var_25933_equation_0 = const()[name = string("op_25933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25933_cast_fp16 = einsum(equation = var_25933_equation_0, values = (var_25431_cast_fp16, var_25831_cast_fp16))[name = string("op_25933_cast_fp16")]; + string var_25935_equation_0 = const()[name = string("op_25935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25935_cast_fp16 = einsum(equation = var_25935_equation_0, values = (var_25431_cast_fp16, var_25832_cast_fp16))[name = string("op_25935_cast_fp16")]; + string var_25937_equation_0 = const()[name = string("op_25937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25937_cast_fp16 = einsum(equation = var_25937_equation_0, values = (var_25431_cast_fp16, var_25833_cast_fp16))[name = string("op_25937_cast_fp16")]; + string var_25939_equation_0 = const()[name = string("op_25939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25939_cast_fp16 = einsum(equation = var_25939_equation_0, values = (var_25435_cast_fp16, var_25834_cast_fp16))[name = string("op_25939_cast_fp16")]; + string var_25941_equation_0 = const()[name = string("op_25941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25941_cast_fp16 = einsum(equation = var_25941_equation_0, values = (var_25435_cast_fp16, var_25835_cast_fp16))[name = string("op_25941_cast_fp16")]; + string var_25943_equation_0 = const()[name = string("op_25943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25943_cast_fp16 = einsum(equation = var_25943_equation_0, values = (var_25435_cast_fp16, var_25836_cast_fp16))[name = string("op_25943_cast_fp16")]; + string var_25945_equation_0 = const()[name = string("op_25945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25945_cast_fp16 = einsum(equation = var_25945_equation_0, values = (var_25435_cast_fp16, var_25837_cast_fp16))[name = string("op_25945_cast_fp16")]; + string var_25947_equation_0 = const()[name = string("op_25947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25947_cast_fp16 = einsum(equation = var_25947_equation_0, values = (var_25439_cast_fp16, var_25838_cast_fp16))[name = string("op_25947_cast_fp16")]; + string var_25949_equation_0 = const()[name = string("op_25949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25949_cast_fp16 = einsum(equation = var_25949_equation_0, values = (var_25439_cast_fp16, var_25839_cast_fp16))[name = string("op_25949_cast_fp16")]; + string var_25951_equation_0 = const()[name = string("op_25951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25951_cast_fp16 = einsum(equation = var_25951_equation_0, values = (var_25439_cast_fp16, var_25840_cast_fp16))[name = string("op_25951_cast_fp16")]; + string var_25953_equation_0 = const()[name = string("op_25953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25953_cast_fp16 = einsum(equation = var_25953_equation_0, values = (var_25439_cast_fp16, var_25841_cast_fp16))[name = string("op_25953_cast_fp16")]; + string var_25955_equation_0 = const()[name = string("op_25955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25955_cast_fp16 = einsum(equation = var_25955_equation_0, values = (var_25443_cast_fp16, var_25842_cast_fp16))[name = string("op_25955_cast_fp16")]; + string var_25957_equation_0 = const()[name = string("op_25957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25957_cast_fp16 = einsum(equation = var_25957_equation_0, values = (var_25443_cast_fp16, var_25843_cast_fp16))[name = string("op_25957_cast_fp16")]; + string var_25959_equation_0 = const()[name = string("op_25959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25959_cast_fp16 = einsum(equation = var_25959_equation_0, values = (var_25443_cast_fp16, var_25844_cast_fp16))[name = string("op_25959_cast_fp16")]; + string var_25961_equation_0 = const()[name = string("op_25961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25961_cast_fp16 = einsum(equation = var_25961_equation_0, values = (var_25443_cast_fp16, var_25845_cast_fp16))[name = string("op_25961_cast_fp16")]; + string var_25963_equation_0 = const()[name = string("op_25963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25963_cast_fp16 = einsum(equation = var_25963_equation_0, values = (var_25447_cast_fp16, var_25846_cast_fp16))[name = string("op_25963_cast_fp16")]; + string var_25965_equation_0 = const()[name = string("op_25965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25965_cast_fp16 = einsum(equation = var_25965_equation_0, values = (var_25447_cast_fp16, var_25847_cast_fp16))[name = string("op_25965_cast_fp16")]; + string var_25967_equation_0 = const()[name = string("op_25967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25967_cast_fp16 = einsum(equation = var_25967_equation_0, values = (var_25447_cast_fp16, var_25848_cast_fp16))[name = string("op_25967_cast_fp16")]; + string var_25969_equation_0 = const()[name = string("op_25969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25969_cast_fp16 = einsum(equation = var_25969_equation_0, values = (var_25447_cast_fp16, var_25849_cast_fp16))[name = string("op_25969_cast_fp16")]; + string var_25971_equation_0 = const()[name = string("op_25971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25971_cast_fp16 = einsum(equation = var_25971_equation_0, values = (var_25451_cast_fp16, var_25850_cast_fp16))[name = string("op_25971_cast_fp16")]; + string var_25973_equation_0 = const()[name = string("op_25973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25973_cast_fp16 = einsum(equation = var_25973_equation_0, values = (var_25451_cast_fp16, var_25851_cast_fp16))[name = string("op_25973_cast_fp16")]; + string var_25975_equation_0 = const()[name = string("op_25975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25975_cast_fp16 = einsum(equation = var_25975_equation_0, values = (var_25451_cast_fp16, var_25852_cast_fp16))[name = string("op_25975_cast_fp16")]; + string var_25977_equation_0 = const()[name = string("op_25977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25977_cast_fp16 = einsum(equation = var_25977_equation_0, values = (var_25451_cast_fp16, var_25853_cast_fp16))[name = string("op_25977_cast_fp16")]; + string var_25979_equation_0 = const()[name = string("op_25979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25979_cast_fp16 = einsum(equation = var_25979_equation_0, values = (var_25455_cast_fp16, var_25854_cast_fp16))[name = string("op_25979_cast_fp16")]; + string var_25981_equation_0 = const()[name = string("op_25981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25981_cast_fp16 = einsum(equation = var_25981_equation_0, values = (var_25455_cast_fp16, var_25855_cast_fp16))[name = string("op_25981_cast_fp16")]; + string var_25983_equation_0 = const()[name = string("op_25983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25983_cast_fp16 = einsum(equation = var_25983_equation_0, values = (var_25455_cast_fp16, var_25856_cast_fp16))[name = string("op_25983_cast_fp16")]; + string var_25985_equation_0 = const()[name = string("op_25985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25985_cast_fp16 = einsum(equation = var_25985_equation_0, values = (var_25455_cast_fp16, var_25857_cast_fp16))[name = string("op_25985_cast_fp16")]; + string var_25987_equation_0 = const()[name = string("op_25987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25987_cast_fp16 = einsum(equation = var_25987_equation_0, values = (var_25459_cast_fp16, var_25858_cast_fp16))[name = string("op_25987_cast_fp16")]; + string var_25989_equation_0 = const()[name = string("op_25989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25989_cast_fp16 = einsum(equation = var_25989_equation_0, values = (var_25459_cast_fp16, var_25859_cast_fp16))[name = string("op_25989_cast_fp16")]; + string var_25991_equation_0 = const()[name = string("op_25991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25991_cast_fp16 = einsum(equation = var_25991_equation_0, values = (var_25459_cast_fp16, var_25860_cast_fp16))[name = string("op_25991_cast_fp16")]; + string var_25993_equation_0 = const()[name = string("op_25993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25993_cast_fp16 = einsum(equation = var_25993_equation_0, values = (var_25459_cast_fp16, var_25861_cast_fp16))[name = string("op_25993_cast_fp16")]; + string var_25995_equation_0 = const()[name = string("op_25995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25995_cast_fp16 = einsum(equation = var_25995_equation_0, values = (var_25463_cast_fp16, var_25862_cast_fp16))[name = string("op_25995_cast_fp16")]; + string var_25997_equation_0 = const()[name = string("op_25997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25997_cast_fp16 = einsum(equation = var_25997_equation_0, values = (var_25463_cast_fp16, var_25863_cast_fp16))[name = string("op_25997_cast_fp16")]; + string var_25999_equation_0 = const()[name = string("op_25999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25999_cast_fp16 = einsum(equation = var_25999_equation_0, values = (var_25463_cast_fp16, var_25864_cast_fp16))[name = string("op_25999_cast_fp16")]; + string var_26001_equation_0 = const()[name = string("op_26001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26001_cast_fp16 = einsum(equation = var_26001_equation_0, values = (var_25463_cast_fp16, var_25865_cast_fp16))[name = string("op_26001_cast_fp16")]; + string var_26003_equation_0 = const()[name = string("op_26003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26003_cast_fp16 = einsum(equation = var_26003_equation_0, values = (var_25467_cast_fp16, var_25866_cast_fp16))[name = string("op_26003_cast_fp16")]; + string var_26005_equation_0 = const()[name = string("op_26005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26005_cast_fp16 = einsum(equation = var_26005_equation_0, values = (var_25467_cast_fp16, var_25867_cast_fp16))[name = string("op_26005_cast_fp16")]; + string var_26007_equation_0 = const()[name = string("op_26007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26007_cast_fp16 = einsum(equation = var_26007_equation_0, values = (var_25467_cast_fp16, var_25868_cast_fp16))[name = string("op_26007_cast_fp16")]; + string var_26009_equation_0 = const()[name = string("op_26009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26009_cast_fp16 = einsum(equation = var_26009_equation_0, values = (var_25467_cast_fp16, var_25869_cast_fp16))[name = string("op_26009_cast_fp16")]; + string var_26011_equation_0 = const()[name = string("op_26011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26011_cast_fp16 = einsum(equation = var_26011_equation_0, values = (var_25471_cast_fp16, var_25870_cast_fp16))[name = string("op_26011_cast_fp16")]; + string var_26013_equation_0 = const()[name = string("op_26013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26013_cast_fp16 = einsum(equation = var_26013_equation_0, values = (var_25471_cast_fp16, var_25871_cast_fp16))[name = string("op_26013_cast_fp16")]; + string var_26015_equation_0 = const()[name = string("op_26015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26015_cast_fp16 = einsum(equation = var_26015_equation_0, values = (var_25471_cast_fp16, var_25872_cast_fp16))[name = string("op_26015_cast_fp16")]; + string var_26017_equation_0 = const()[name = string("op_26017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26017_cast_fp16 = einsum(equation = var_26017_equation_0, values = (var_25471_cast_fp16, var_25873_cast_fp16))[name = string("op_26017_cast_fp16")]; + string var_26019_equation_0 = const()[name = string("op_26019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26019_cast_fp16 = einsum(equation = var_26019_equation_0, values = (var_25475_cast_fp16, var_25874_cast_fp16))[name = string("op_26019_cast_fp16")]; + string var_26021_equation_0 = const()[name = string("op_26021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26021_cast_fp16 = einsum(equation = var_26021_equation_0, values = (var_25475_cast_fp16, var_25875_cast_fp16))[name = string("op_26021_cast_fp16")]; + string var_26023_equation_0 = const()[name = string("op_26023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26023_cast_fp16 = einsum(equation = var_26023_equation_0, values = (var_25475_cast_fp16, var_25876_cast_fp16))[name = string("op_26023_cast_fp16")]; + string var_26025_equation_0 = const()[name = string("op_26025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26025_cast_fp16 = einsum(equation = var_26025_equation_0, values = (var_25475_cast_fp16, var_25877_cast_fp16))[name = string("op_26025_cast_fp16")]; + string var_26027_equation_0 = const()[name = string("op_26027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26027_cast_fp16 = einsum(equation = var_26027_equation_0, values = (var_25479_cast_fp16, var_25878_cast_fp16))[name = string("op_26027_cast_fp16")]; + string var_26029_equation_0 = const()[name = string("op_26029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26029_cast_fp16 = einsum(equation = var_26029_equation_0, values = (var_25479_cast_fp16, var_25879_cast_fp16))[name = string("op_26029_cast_fp16")]; + string var_26031_equation_0 = const()[name = string("op_26031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26031_cast_fp16 = einsum(equation = var_26031_equation_0, values = (var_25479_cast_fp16, var_25880_cast_fp16))[name = string("op_26031_cast_fp16")]; + string var_26033_equation_0 = const()[name = string("op_26033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26033_cast_fp16 = einsum(equation = var_26033_equation_0, values = (var_25479_cast_fp16, var_25881_cast_fp16))[name = string("op_26033_cast_fp16")]; + string var_26035_equation_0 = const()[name = string("op_26035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26035_cast_fp16 = einsum(equation = var_26035_equation_0, values = (var_25483_cast_fp16, var_25882_cast_fp16))[name = string("op_26035_cast_fp16")]; + string var_26037_equation_0 = const()[name = string("op_26037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26037_cast_fp16 = einsum(equation = var_26037_equation_0, values = (var_25483_cast_fp16, var_25883_cast_fp16))[name = string("op_26037_cast_fp16")]; + string var_26039_equation_0 = const()[name = string("op_26039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26039_cast_fp16 = einsum(equation = var_26039_equation_0, values = (var_25483_cast_fp16, var_25884_cast_fp16))[name = string("op_26039_cast_fp16")]; + string var_26041_equation_0 = const()[name = string("op_26041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26041_cast_fp16 = einsum(equation = var_26041_equation_0, values = (var_25483_cast_fp16, var_25885_cast_fp16))[name = string("op_26041_cast_fp16")]; + string var_26043_equation_0 = const()[name = string("op_26043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26043_cast_fp16 = einsum(equation = var_26043_equation_0, values = (var_25487_cast_fp16, var_25886_cast_fp16))[name = string("op_26043_cast_fp16")]; + string var_26045_equation_0 = const()[name = string("op_26045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26045_cast_fp16 = einsum(equation = var_26045_equation_0, values = (var_25487_cast_fp16, var_25887_cast_fp16))[name = string("op_26045_cast_fp16")]; + string var_26047_equation_0 = const()[name = string("op_26047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26047_cast_fp16 = einsum(equation = var_26047_equation_0, values = (var_25487_cast_fp16, var_25888_cast_fp16))[name = string("op_26047_cast_fp16")]; + string var_26049_equation_0 = const()[name = string("op_26049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26049_cast_fp16 = einsum(equation = var_26049_equation_0, values = (var_25487_cast_fp16, var_25889_cast_fp16))[name = string("op_26049_cast_fp16")]; + bool var_26051_interleave_0 = const()[name = string("op_26051_interleave_0"), val = bool(false)]; + tensor var_26051_cast_fp16 = concat(axis = var_24610, interleave = var_26051_interleave_0, values = (var_25891_cast_fp16, var_25893_cast_fp16, var_25895_cast_fp16, var_25897_cast_fp16))[name = string("op_26051_cast_fp16")]; + bool var_26053_interleave_0 = const()[name = string("op_26053_interleave_0"), val = bool(false)]; + tensor var_26053_cast_fp16 = concat(axis = var_24610, interleave = var_26053_interleave_0, values = (var_25899_cast_fp16, var_25901_cast_fp16, var_25903_cast_fp16, var_25905_cast_fp16))[name = string("op_26053_cast_fp16")]; + bool var_26055_interleave_0 = const()[name = string("op_26055_interleave_0"), val = bool(false)]; + tensor var_26055_cast_fp16 = concat(axis = var_24610, interleave = var_26055_interleave_0, values = (var_25907_cast_fp16, var_25909_cast_fp16, var_25911_cast_fp16, var_25913_cast_fp16))[name = string("op_26055_cast_fp16")]; + bool var_26057_interleave_0 = const()[name = string("op_26057_interleave_0"), val = bool(false)]; + tensor var_26057_cast_fp16 = concat(axis = var_24610, interleave = var_26057_interleave_0, values = (var_25915_cast_fp16, var_25917_cast_fp16, var_25919_cast_fp16, var_25921_cast_fp16))[name = string("op_26057_cast_fp16")]; + bool var_26059_interleave_0 = const()[name = string("op_26059_interleave_0"), val = bool(false)]; + tensor var_26059_cast_fp16 = concat(axis = var_24610, interleave = var_26059_interleave_0, values = (var_25923_cast_fp16, var_25925_cast_fp16, var_25927_cast_fp16, var_25929_cast_fp16))[name = string("op_26059_cast_fp16")]; + bool var_26061_interleave_0 = const()[name = string("op_26061_interleave_0"), val = bool(false)]; + tensor var_26061_cast_fp16 = concat(axis = var_24610, interleave = var_26061_interleave_0, values = (var_25931_cast_fp16, var_25933_cast_fp16, var_25935_cast_fp16, var_25937_cast_fp16))[name = string("op_26061_cast_fp16")]; + bool var_26063_interleave_0 = const()[name = string("op_26063_interleave_0"), val = bool(false)]; + tensor var_26063_cast_fp16 = concat(axis = var_24610, interleave = var_26063_interleave_0, values = (var_25939_cast_fp16, var_25941_cast_fp16, var_25943_cast_fp16, var_25945_cast_fp16))[name = string("op_26063_cast_fp16")]; + bool var_26065_interleave_0 = const()[name = string("op_26065_interleave_0"), val = bool(false)]; + tensor var_26065_cast_fp16 = concat(axis = var_24610, interleave = var_26065_interleave_0, values = (var_25947_cast_fp16, var_25949_cast_fp16, var_25951_cast_fp16, var_25953_cast_fp16))[name = string("op_26065_cast_fp16")]; + bool var_26067_interleave_0 = const()[name = string("op_26067_interleave_0"), val = bool(false)]; + tensor var_26067_cast_fp16 = concat(axis = var_24610, interleave = var_26067_interleave_0, values = (var_25955_cast_fp16, var_25957_cast_fp16, var_25959_cast_fp16, var_25961_cast_fp16))[name = string("op_26067_cast_fp16")]; + bool var_26069_interleave_0 = const()[name = string("op_26069_interleave_0"), val = bool(false)]; + tensor var_26069_cast_fp16 = concat(axis = var_24610, interleave = var_26069_interleave_0, values = (var_25963_cast_fp16, var_25965_cast_fp16, var_25967_cast_fp16, var_25969_cast_fp16))[name = string("op_26069_cast_fp16")]; + bool var_26071_interleave_0 = const()[name = string("op_26071_interleave_0"), val = bool(false)]; + tensor var_26071_cast_fp16 = concat(axis = var_24610, interleave = var_26071_interleave_0, values = (var_25971_cast_fp16, var_25973_cast_fp16, var_25975_cast_fp16, var_25977_cast_fp16))[name = string("op_26071_cast_fp16")]; + bool var_26073_interleave_0 = const()[name = string("op_26073_interleave_0"), val = bool(false)]; + tensor var_26073_cast_fp16 = concat(axis = var_24610, interleave = var_26073_interleave_0, values = (var_25979_cast_fp16, var_25981_cast_fp16, var_25983_cast_fp16, var_25985_cast_fp16))[name = string("op_26073_cast_fp16")]; + bool var_26075_interleave_0 = const()[name = string("op_26075_interleave_0"), val = bool(false)]; + tensor var_26075_cast_fp16 = concat(axis = var_24610, interleave = var_26075_interleave_0, values = (var_25987_cast_fp16, var_25989_cast_fp16, var_25991_cast_fp16, var_25993_cast_fp16))[name = string("op_26075_cast_fp16")]; + bool var_26077_interleave_0 = const()[name = string("op_26077_interleave_0"), val = bool(false)]; + tensor var_26077_cast_fp16 = concat(axis = var_24610, interleave = var_26077_interleave_0, values = (var_25995_cast_fp16, var_25997_cast_fp16, var_25999_cast_fp16, var_26001_cast_fp16))[name = string("op_26077_cast_fp16")]; + bool var_26079_interleave_0 = const()[name = string("op_26079_interleave_0"), val = bool(false)]; + tensor var_26079_cast_fp16 = concat(axis = var_24610, interleave = var_26079_interleave_0, values = (var_26003_cast_fp16, var_26005_cast_fp16, var_26007_cast_fp16, var_26009_cast_fp16))[name = string("op_26079_cast_fp16")]; + bool var_26081_interleave_0 = const()[name = string("op_26081_interleave_0"), val = bool(false)]; + tensor var_26081_cast_fp16 = concat(axis = var_24610, interleave = var_26081_interleave_0, values = (var_26011_cast_fp16, var_26013_cast_fp16, var_26015_cast_fp16, var_26017_cast_fp16))[name = string("op_26081_cast_fp16")]; + bool var_26083_interleave_0 = const()[name = string("op_26083_interleave_0"), val = bool(false)]; + tensor var_26083_cast_fp16 = concat(axis = var_24610, interleave = var_26083_interleave_0, values = (var_26019_cast_fp16, var_26021_cast_fp16, var_26023_cast_fp16, var_26025_cast_fp16))[name = string("op_26083_cast_fp16")]; + bool var_26085_interleave_0 = const()[name = string("op_26085_interleave_0"), val = bool(false)]; + tensor var_26085_cast_fp16 = concat(axis = var_24610, interleave = var_26085_interleave_0, values = (var_26027_cast_fp16, var_26029_cast_fp16, var_26031_cast_fp16, var_26033_cast_fp16))[name = string("op_26085_cast_fp16")]; + bool var_26087_interleave_0 = const()[name = string("op_26087_interleave_0"), val = bool(false)]; + tensor var_26087_cast_fp16 = concat(axis = var_24610, interleave = var_26087_interleave_0, values = (var_26035_cast_fp16, var_26037_cast_fp16, var_26039_cast_fp16, var_26041_cast_fp16))[name = string("op_26087_cast_fp16")]; + bool var_26089_interleave_0 = const()[name = string("op_26089_interleave_0"), val = bool(false)]; + tensor var_26089_cast_fp16 = concat(axis = var_24610, interleave = var_26089_interleave_0, values = (var_26043_cast_fp16, var_26045_cast_fp16, var_26047_cast_fp16, var_26049_cast_fp16))[name = string("op_26089_cast_fp16")]; + bool input_129_interleave_0 = const()[name = string("input_129_interleave_0"), val = bool(false)]; + tensor input_129_cast_fp16 = concat(axis = var_24635, interleave = input_129_interleave_0, values = (var_26051_cast_fp16, var_26053_cast_fp16, var_26055_cast_fp16, var_26057_cast_fp16, var_26059_cast_fp16, var_26061_cast_fp16, var_26063_cast_fp16, var_26065_cast_fp16, var_26067_cast_fp16, var_26069_cast_fp16, var_26071_cast_fp16, var_26073_cast_fp16, var_26075_cast_fp16, var_26077_cast_fp16, var_26079_cast_fp16, var_26081_cast_fp16, var_26083_cast_fp16, var_26085_cast_fp16, var_26087_cast_fp16, var_26089_cast_fp16))[name = string("input_129_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654157760)))]; + tensor layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657434624)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_129_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_67_cast_fp16")]; + tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; + fp16 var_26108_to_fp16 = const()[name = string("op_26108_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_26108_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; + tensor input_131_gamma_0_to_fp16 = const()[name = string("input_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657437248)))]; + tensor input_131_beta_0_to_fp16 = const()[name = string("input_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657439872)))]; + fp16 input_131_epsilon_0_to_fp16 = const()[name = string("input_131_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("input_131_cast_fp16")]; + string input_133_pad_type_0 = const()[name = string("input_133_pad_type_0"), val = string("valid")]; + tensor input_133_strides_0 = const()[name = string("input_133_strides_0"), val = tensor([1, 1])]; + tensor input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_133_dilations_0 = const()[name = string("input_133_dilations_0"), val = tensor([1, 1])]; + int32 input_133_groups_0 = const()[name = string("input_133_groups_0"), val = int32(1)]; + tensor layers_16_fc1_weight_to_fp16 = const()[name = string("layers_16_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657442496)))]; + tensor layers_16_fc1_bias_to_fp16 = const()[name = string("layers_16_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670549760)))]; + tensor input_133_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = layers_16_fc1_weight_to_fp16, x = input_131_cast_fp16)[name = string("input_133_cast_fp16")]; + string input_135_mode_0 = const()[name = string("input_135_mode_0"), val = string("EXACT")]; + tensor input_135_cast_fp16 = gelu(mode = input_135_mode_0, x = input_133_cast_fp16)[name = string("input_135_cast_fp16")]; + string hidden_states_37_pad_type_0 = const()[name = string("hidden_states_37_pad_type_0"), val = string("valid")]; + tensor hidden_states_37_strides_0 = const()[name = string("hidden_states_37_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_37_pad_0 = const()[name = string("hidden_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_37_dilations_0 = const()[name = string("hidden_states_37_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_37_groups_0 = const()[name = string("hidden_states_37_groups_0"), val = int32(1)]; + tensor layers_16_fc2_weight_to_fp16 = const()[name = string("layers_16_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670560064)))]; + tensor layers_16_fc2_bias_to_fp16 = const()[name = string("layers_16_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683667328)))]; + tensor hidden_states_37_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_16_fc2_weight_to_fp16, x = input_135_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = hidden_states_37_cast_fp16)[name = string("inputs_69_cast_fp16")]; + int32 var_26137 = const()[name = string("op_26137"), val = int32(3)]; + int32 var_26162 = const()[name = string("op_26162"), val = int32(1)]; + tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; + fp16 var_26179_to_fp16 = const()[name = string("op_26179_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_26179_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683669952)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683672576)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; + tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; + tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; + int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683675200)))]; + tensor layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686952064)))]; + tensor query_35_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_35_cast_fp16")]; + string key_35_pad_type_0 = const()[name = string("key_35_pad_type_0"), val = string("valid")]; + tensor key_35_strides_0 = const()[name = string("key_35_strides_0"), val = tensor([1, 1])]; + tensor key_35_pad_0 = const()[name = string("key_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_35_dilations_0 = const()[name = string("key_35_dilations_0"), val = tensor([1, 1])]; + int32 key_35_groups_0 = const()[name = string("key_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686954688)))]; + tensor key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("key_35_cast_fp16")]; + string value_35_pad_type_0 = const()[name = string("value_35_pad_type_0"), val = string("valid")]; + tensor value_35_strides_0 = const()[name = string("value_35_strides_0"), val = tensor([1, 1])]; + tensor value_35_pad_0 = const()[name = string("value_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_35_dilations_0 = const()[name = string("value_35_dilations_0"), val = tensor([1, 1])]; + int32 value_35_groups_0 = const()[name = string("value_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(690231552)))]; + tensor layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693508416)))]; + tensor value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("value_35_cast_fp16")]; + tensor var_26217_begin_0 = const()[name = string("op_26217_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26217_end_0 = const()[name = string("op_26217_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26217_end_mask_0 = const()[name = string("op_26217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26217_cast_fp16 = slice_by_index(begin = var_26217_begin_0, end = var_26217_end_0, end_mask = var_26217_end_mask_0, x = query_35_cast_fp16)[name = string("op_26217_cast_fp16")]; + tensor var_26221_begin_0 = const()[name = string("op_26221_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26221_end_0 = const()[name = string("op_26221_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_26221_end_mask_0 = const()[name = string("op_26221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26221_cast_fp16 = slice_by_index(begin = var_26221_begin_0, end = var_26221_end_0, end_mask = var_26221_end_mask_0, x = query_35_cast_fp16)[name = string("op_26221_cast_fp16")]; + tensor var_26225_begin_0 = const()[name = string("op_26225_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26225_end_0 = const()[name = string("op_26225_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_26225_end_mask_0 = const()[name = string("op_26225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26225_cast_fp16 = slice_by_index(begin = var_26225_begin_0, end = var_26225_end_0, end_mask = var_26225_end_mask_0, x = query_35_cast_fp16)[name = string("op_26225_cast_fp16")]; + tensor var_26229_begin_0 = const()[name = string("op_26229_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26229_end_0 = const()[name = string("op_26229_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_26229_end_mask_0 = const()[name = string("op_26229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26229_cast_fp16 = slice_by_index(begin = var_26229_begin_0, end = var_26229_end_0, end_mask = var_26229_end_mask_0, x = query_35_cast_fp16)[name = string("op_26229_cast_fp16")]; + tensor var_26233_begin_0 = const()[name = string("op_26233_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26233_end_0 = const()[name = string("op_26233_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_26233_end_mask_0 = const()[name = string("op_26233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26233_cast_fp16 = slice_by_index(begin = var_26233_begin_0, end = var_26233_end_0, end_mask = var_26233_end_mask_0, x = query_35_cast_fp16)[name = string("op_26233_cast_fp16")]; + tensor var_26237_begin_0 = const()[name = string("op_26237_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26237_end_0 = const()[name = string("op_26237_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_26237_end_mask_0 = const()[name = string("op_26237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26237_cast_fp16 = slice_by_index(begin = var_26237_begin_0, end = var_26237_end_0, end_mask = var_26237_end_mask_0, x = query_35_cast_fp16)[name = string("op_26237_cast_fp16")]; + tensor var_26241_begin_0 = const()[name = string("op_26241_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26241_end_0 = const()[name = string("op_26241_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_26241_end_mask_0 = const()[name = string("op_26241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26241_cast_fp16 = slice_by_index(begin = var_26241_begin_0, end = var_26241_end_0, end_mask = var_26241_end_mask_0, x = query_35_cast_fp16)[name = string("op_26241_cast_fp16")]; + tensor var_26245_begin_0 = const()[name = string("op_26245_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26245_end_0 = const()[name = string("op_26245_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_26245_end_mask_0 = const()[name = string("op_26245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26245_cast_fp16 = slice_by_index(begin = var_26245_begin_0, end = var_26245_end_0, end_mask = var_26245_end_mask_0, x = query_35_cast_fp16)[name = string("op_26245_cast_fp16")]; + tensor var_26249_begin_0 = const()[name = string("op_26249_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26249_end_0 = const()[name = string("op_26249_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_26249_end_mask_0 = const()[name = string("op_26249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26249_cast_fp16 = slice_by_index(begin = var_26249_begin_0, end = var_26249_end_0, end_mask = var_26249_end_mask_0, x = query_35_cast_fp16)[name = string("op_26249_cast_fp16")]; + tensor var_26253_begin_0 = const()[name = string("op_26253_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26253_end_0 = const()[name = string("op_26253_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_26253_end_mask_0 = const()[name = string("op_26253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26253_cast_fp16 = slice_by_index(begin = var_26253_begin_0, end = var_26253_end_0, end_mask = var_26253_end_mask_0, x = query_35_cast_fp16)[name = string("op_26253_cast_fp16")]; + tensor var_26257_begin_0 = const()[name = string("op_26257_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26257_end_0 = const()[name = string("op_26257_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_26257_end_mask_0 = const()[name = string("op_26257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26257_cast_fp16 = slice_by_index(begin = var_26257_begin_0, end = var_26257_end_0, end_mask = var_26257_end_mask_0, x = query_35_cast_fp16)[name = string("op_26257_cast_fp16")]; + tensor var_26261_begin_0 = const()[name = string("op_26261_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26261_end_0 = const()[name = string("op_26261_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_26261_end_mask_0 = const()[name = string("op_26261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26261_cast_fp16 = slice_by_index(begin = var_26261_begin_0, end = var_26261_end_0, end_mask = var_26261_end_mask_0, x = query_35_cast_fp16)[name = string("op_26261_cast_fp16")]; + tensor var_26265_begin_0 = const()[name = string("op_26265_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26265_end_0 = const()[name = string("op_26265_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_26265_end_mask_0 = const()[name = string("op_26265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26265_cast_fp16 = slice_by_index(begin = var_26265_begin_0, end = var_26265_end_0, end_mask = var_26265_end_mask_0, x = query_35_cast_fp16)[name = string("op_26265_cast_fp16")]; + tensor var_26269_begin_0 = const()[name = string("op_26269_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26269_end_0 = const()[name = string("op_26269_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_26269_end_mask_0 = const()[name = string("op_26269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26269_cast_fp16 = slice_by_index(begin = var_26269_begin_0, end = var_26269_end_0, end_mask = var_26269_end_mask_0, x = query_35_cast_fp16)[name = string("op_26269_cast_fp16")]; + tensor var_26273_begin_0 = const()[name = string("op_26273_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26273_end_0 = const()[name = string("op_26273_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_26273_end_mask_0 = const()[name = string("op_26273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26273_cast_fp16 = slice_by_index(begin = var_26273_begin_0, end = var_26273_end_0, end_mask = var_26273_end_mask_0, x = query_35_cast_fp16)[name = string("op_26273_cast_fp16")]; + tensor var_26277_begin_0 = const()[name = string("op_26277_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26277_end_0 = const()[name = string("op_26277_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_26277_end_mask_0 = const()[name = string("op_26277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26277_cast_fp16 = slice_by_index(begin = var_26277_begin_0, end = var_26277_end_0, end_mask = var_26277_end_mask_0, x = query_35_cast_fp16)[name = string("op_26277_cast_fp16")]; + tensor var_26281_begin_0 = const()[name = string("op_26281_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26281_end_0 = const()[name = string("op_26281_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_26281_end_mask_0 = const()[name = string("op_26281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26281_cast_fp16 = slice_by_index(begin = var_26281_begin_0, end = var_26281_end_0, end_mask = var_26281_end_mask_0, x = query_35_cast_fp16)[name = string("op_26281_cast_fp16")]; + tensor var_26285_begin_0 = const()[name = string("op_26285_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26285_end_0 = const()[name = string("op_26285_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_26285_end_mask_0 = const()[name = string("op_26285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26285_cast_fp16 = slice_by_index(begin = var_26285_begin_0, end = var_26285_end_0, end_mask = var_26285_end_mask_0, x = query_35_cast_fp16)[name = string("op_26285_cast_fp16")]; + tensor var_26289_begin_0 = const()[name = string("op_26289_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26289_end_0 = const()[name = string("op_26289_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_26289_end_mask_0 = const()[name = string("op_26289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26289_cast_fp16 = slice_by_index(begin = var_26289_begin_0, end = var_26289_end_0, end_mask = var_26289_end_mask_0, x = query_35_cast_fp16)[name = string("op_26289_cast_fp16")]; + tensor var_26293_begin_0 = const()[name = string("op_26293_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26293_end_0 = const()[name = string("op_26293_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_26293_end_mask_0 = const()[name = string("op_26293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26293_cast_fp16 = slice_by_index(begin = var_26293_begin_0, end = var_26293_end_0, end_mask = var_26293_end_mask_0, x = query_35_cast_fp16)[name = string("op_26293_cast_fp16")]; + tensor var_26302_begin_0 = const()[name = string("op_26302_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26302_end_0 = const()[name = string("op_26302_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26302_end_mask_0 = const()[name = string("op_26302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26302_cast_fp16 = slice_by_index(begin = var_26302_begin_0, end = var_26302_end_0, end_mask = var_26302_end_mask_0, x = var_26217_cast_fp16)[name = string("op_26302_cast_fp16")]; + tensor var_26309_begin_0 = const()[name = string("op_26309_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26309_end_0 = const()[name = string("op_26309_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26309_end_mask_0 = const()[name = string("op_26309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26309_cast_fp16 = slice_by_index(begin = var_26309_begin_0, end = var_26309_end_0, end_mask = var_26309_end_mask_0, x = var_26217_cast_fp16)[name = string("op_26309_cast_fp16")]; + tensor var_26316_begin_0 = const()[name = string("op_26316_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26316_end_0 = const()[name = string("op_26316_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26316_end_mask_0 = const()[name = string("op_26316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26316_cast_fp16 = slice_by_index(begin = var_26316_begin_0, end = var_26316_end_0, end_mask = var_26316_end_mask_0, x = var_26217_cast_fp16)[name = string("op_26316_cast_fp16")]; + tensor var_26323_begin_0 = const()[name = string("op_26323_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26323_end_0 = const()[name = string("op_26323_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26323_end_mask_0 = const()[name = string("op_26323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26323_cast_fp16 = slice_by_index(begin = var_26323_begin_0, end = var_26323_end_0, end_mask = var_26323_end_mask_0, x = var_26217_cast_fp16)[name = string("op_26323_cast_fp16")]; + tensor var_26330_begin_0 = const()[name = string("op_26330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26330_end_0 = const()[name = string("op_26330_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26330_end_mask_0 = const()[name = string("op_26330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26330_cast_fp16 = slice_by_index(begin = var_26330_begin_0, end = var_26330_end_0, end_mask = var_26330_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26330_cast_fp16")]; + tensor var_26337_begin_0 = const()[name = string("op_26337_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26337_end_0 = const()[name = string("op_26337_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26337_end_mask_0 = const()[name = string("op_26337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26337_cast_fp16 = slice_by_index(begin = var_26337_begin_0, end = var_26337_end_0, end_mask = var_26337_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26337_cast_fp16")]; + tensor var_26344_begin_0 = const()[name = string("op_26344_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26344_end_0 = const()[name = string("op_26344_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26344_end_mask_0 = const()[name = string("op_26344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26344_cast_fp16 = slice_by_index(begin = var_26344_begin_0, end = var_26344_end_0, end_mask = var_26344_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26344_cast_fp16")]; + tensor var_26351_begin_0 = const()[name = string("op_26351_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26351_end_0 = const()[name = string("op_26351_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26351_end_mask_0 = const()[name = string("op_26351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26351_cast_fp16 = slice_by_index(begin = var_26351_begin_0, end = var_26351_end_0, end_mask = var_26351_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26351_cast_fp16")]; + tensor var_26358_begin_0 = const()[name = string("op_26358_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26358_end_0 = const()[name = string("op_26358_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26358_end_mask_0 = const()[name = string("op_26358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26358_cast_fp16 = slice_by_index(begin = var_26358_begin_0, end = var_26358_end_0, end_mask = var_26358_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26358_cast_fp16")]; + tensor var_26365_begin_0 = const()[name = string("op_26365_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26365_end_0 = const()[name = string("op_26365_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26365_end_mask_0 = const()[name = string("op_26365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26365_cast_fp16 = slice_by_index(begin = var_26365_begin_0, end = var_26365_end_0, end_mask = var_26365_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26365_cast_fp16")]; + tensor var_26372_begin_0 = const()[name = string("op_26372_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26372_end_0 = const()[name = string("op_26372_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26372_end_mask_0 = const()[name = string("op_26372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26372_cast_fp16 = slice_by_index(begin = var_26372_begin_0, end = var_26372_end_0, end_mask = var_26372_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26372_cast_fp16")]; + tensor var_26379_begin_0 = const()[name = string("op_26379_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26379_end_0 = const()[name = string("op_26379_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26379_end_mask_0 = const()[name = string("op_26379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26379_cast_fp16 = slice_by_index(begin = var_26379_begin_0, end = var_26379_end_0, end_mask = var_26379_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26379_cast_fp16")]; + tensor var_26386_begin_0 = const()[name = string("op_26386_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26386_end_0 = const()[name = string("op_26386_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26386_end_mask_0 = const()[name = string("op_26386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26386_cast_fp16 = slice_by_index(begin = var_26386_begin_0, end = var_26386_end_0, end_mask = var_26386_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26386_cast_fp16")]; + tensor var_26393_begin_0 = const()[name = string("op_26393_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26393_end_0 = const()[name = string("op_26393_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26393_end_mask_0 = const()[name = string("op_26393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26393_cast_fp16 = slice_by_index(begin = var_26393_begin_0, end = var_26393_end_0, end_mask = var_26393_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26393_cast_fp16")]; + tensor var_26400_begin_0 = const()[name = string("op_26400_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26400_end_0 = const()[name = string("op_26400_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26400_end_mask_0 = const()[name = string("op_26400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26400_cast_fp16 = slice_by_index(begin = var_26400_begin_0, end = var_26400_end_0, end_mask = var_26400_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26400_cast_fp16")]; + tensor var_26407_begin_0 = const()[name = string("op_26407_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26407_end_0 = const()[name = string("op_26407_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26407_end_mask_0 = const()[name = string("op_26407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26407_cast_fp16 = slice_by_index(begin = var_26407_begin_0, end = var_26407_end_0, end_mask = var_26407_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26407_cast_fp16")]; + tensor var_26414_begin_0 = const()[name = string("op_26414_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26414_end_0 = const()[name = string("op_26414_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26414_end_mask_0 = const()[name = string("op_26414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26414_cast_fp16 = slice_by_index(begin = var_26414_begin_0, end = var_26414_end_0, end_mask = var_26414_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26414_cast_fp16")]; + tensor var_26421_begin_0 = const()[name = string("op_26421_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26421_end_0 = const()[name = string("op_26421_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26421_end_mask_0 = const()[name = string("op_26421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26421_cast_fp16 = slice_by_index(begin = var_26421_begin_0, end = var_26421_end_0, end_mask = var_26421_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26421_cast_fp16")]; + tensor var_26428_begin_0 = const()[name = string("op_26428_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26428_end_0 = const()[name = string("op_26428_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26428_end_mask_0 = const()[name = string("op_26428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26428_cast_fp16 = slice_by_index(begin = var_26428_begin_0, end = var_26428_end_0, end_mask = var_26428_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26428_cast_fp16")]; + tensor var_26435_begin_0 = const()[name = string("op_26435_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26435_end_0 = const()[name = string("op_26435_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26435_end_mask_0 = const()[name = string("op_26435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26435_cast_fp16 = slice_by_index(begin = var_26435_begin_0, end = var_26435_end_0, end_mask = var_26435_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26435_cast_fp16")]; + tensor var_26442_begin_0 = const()[name = string("op_26442_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26442_end_0 = const()[name = string("op_26442_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26442_end_mask_0 = const()[name = string("op_26442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26442_cast_fp16 = slice_by_index(begin = var_26442_begin_0, end = var_26442_end_0, end_mask = var_26442_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26442_cast_fp16")]; + tensor var_26449_begin_0 = const()[name = string("op_26449_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26449_end_0 = const()[name = string("op_26449_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26449_end_mask_0 = const()[name = string("op_26449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26449_cast_fp16 = slice_by_index(begin = var_26449_begin_0, end = var_26449_end_0, end_mask = var_26449_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26449_cast_fp16")]; + tensor var_26456_begin_0 = const()[name = string("op_26456_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26456_end_0 = const()[name = string("op_26456_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26456_end_mask_0 = const()[name = string("op_26456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26456_cast_fp16 = slice_by_index(begin = var_26456_begin_0, end = var_26456_end_0, end_mask = var_26456_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26456_cast_fp16")]; + tensor var_26463_begin_0 = const()[name = string("op_26463_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26463_end_0 = const()[name = string("op_26463_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26463_end_mask_0 = const()[name = string("op_26463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26463_cast_fp16 = slice_by_index(begin = var_26463_begin_0, end = var_26463_end_0, end_mask = var_26463_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26463_cast_fp16")]; + tensor var_26470_begin_0 = const()[name = string("op_26470_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26470_end_0 = const()[name = string("op_26470_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26470_end_mask_0 = const()[name = string("op_26470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26470_cast_fp16 = slice_by_index(begin = var_26470_begin_0, end = var_26470_end_0, end_mask = var_26470_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26470_cast_fp16")]; + tensor var_26477_begin_0 = const()[name = string("op_26477_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26477_end_0 = const()[name = string("op_26477_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26477_end_mask_0 = const()[name = string("op_26477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26477_cast_fp16 = slice_by_index(begin = var_26477_begin_0, end = var_26477_end_0, end_mask = var_26477_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26477_cast_fp16")]; + tensor var_26484_begin_0 = const()[name = string("op_26484_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26484_end_0 = const()[name = string("op_26484_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26484_end_mask_0 = const()[name = string("op_26484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26484_cast_fp16 = slice_by_index(begin = var_26484_begin_0, end = var_26484_end_0, end_mask = var_26484_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26484_cast_fp16")]; + tensor var_26491_begin_0 = const()[name = string("op_26491_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26491_end_0 = const()[name = string("op_26491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26491_end_mask_0 = const()[name = string("op_26491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26491_cast_fp16 = slice_by_index(begin = var_26491_begin_0, end = var_26491_end_0, end_mask = var_26491_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26491_cast_fp16")]; + tensor var_26498_begin_0 = const()[name = string("op_26498_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26498_end_0 = const()[name = string("op_26498_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26498_end_mask_0 = const()[name = string("op_26498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26498_cast_fp16 = slice_by_index(begin = var_26498_begin_0, end = var_26498_end_0, end_mask = var_26498_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26498_cast_fp16")]; + tensor var_26505_begin_0 = const()[name = string("op_26505_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26505_end_0 = const()[name = string("op_26505_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26505_end_mask_0 = const()[name = string("op_26505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26505_cast_fp16 = slice_by_index(begin = var_26505_begin_0, end = var_26505_end_0, end_mask = var_26505_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26505_cast_fp16")]; + tensor var_26512_begin_0 = const()[name = string("op_26512_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26512_end_0 = const()[name = string("op_26512_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26512_end_mask_0 = const()[name = string("op_26512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26512_cast_fp16 = slice_by_index(begin = var_26512_begin_0, end = var_26512_end_0, end_mask = var_26512_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26512_cast_fp16")]; + tensor var_26519_begin_0 = const()[name = string("op_26519_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26519_end_0 = const()[name = string("op_26519_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26519_end_mask_0 = const()[name = string("op_26519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26519_cast_fp16 = slice_by_index(begin = var_26519_begin_0, end = var_26519_end_0, end_mask = var_26519_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26519_cast_fp16")]; + tensor var_26526_begin_0 = const()[name = string("op_26526_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26526_end_0 = const()[name = string("op_26526_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26526_end_mask_0 = const()[name = string("op_26526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26526_cast_fp16 = slice_by_index(begin = var_26526_begin_0, end = var_26526_end_0, end_mask = var_26526_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26526_cast_fp16")]; + tensor var_26533_begin_0 = const()[name = string("op_26533_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26533_end_0 = const()[name = string("op_26533_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26533_end_mask_0 = const()[name = string("op_26533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26533_cast_fp16 = slice_by_index(begin = var_26533_begin_0, end = var_26533_end_0, end_mask = var_26533_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26533_cast_fp16")]; + tensor var_26540_begin_0 = const()[name = string("op_26540_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26540_end_0 = const()[name = string("op_26540_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26540_end_mask_0 = const()[name = string("op_26540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26540_cast_fp16 = slice_by_index(begin = var_26540_begin_0, end = var_26540_end_0, end_mask = var_26540_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26540_cast_fp16")]; + tensor var_26547_begin_0 = const()[name = string("op_26547_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26547_end_0 = const()[name = string("op_26547_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26547_end_mask_0 = const()[name = string("op_26547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26547_cast_fp16 = slice_by_index(begin = var_26547_begin_0, end = var_26547_end_0, end_mask = var_26547_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26547_cast_fp16")]; + tensor var_26554_begin_0 = const()[name = string("op_26554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26554_end_0 = const()[name = string("op_26554_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26554_end_mask_0 = const()[name = string("op_26554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26554_cast_fp16 = slice_by_index(begin = var_26554_begin_0, end = var_26554_end_0, end_mask = var_26554_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26554_cast_fp16")]; + tensor var_26561_begin_0 = const()[name = string("op_26561_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26561_end_0 = const()[name = string("op_26561_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26561_end_mask_0 = const()[name = string("op_26561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26561_cast_fp16 = slice_by_index(begin = var_26561_begin_0, end = var_26561_end_0, end_mask = var_26561_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26561_cast_fp16")]; + tensor var_26568_begin_0 = const()[name = string("op_26568_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26568_end_0 = const()[name = string("op_26568_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26568_end_mask_0 = const()[name = string("op_26568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26568_cast_fp16 = slice_by_index(begin = var_26568_begin_0, end = var_26568_end_0, end_mask = var_26568_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26568_cast_fp16")]; + tensor var_26575_begin_0 = const()[name = string("op_26575_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26575_end_0 = const()[name = string("op_26575_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26575_end_mask_0 = const()[name = string("op_26575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26575_cast_fp16 = slice_by_index(begin = var_26575_begin_0, end = var_26575_end_0, end_mask = var_26575_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26575_cast_fp16")]; + tensor var_26582_begin_0 = const()[name = string("op_26582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26582_end_0 = const()[name = string("op_26582_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26582_end_mask_0 = const()[name = string("op_26582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26582_cast_fp16 = slice_by_index(begin = var_26582_begin_0, end = var_26582_end_0, end_mask = var_26582_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26582_cast_fp16")]; + tensor var_26589_begin_0 = const()[name = string("op_26589_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26589_end_0 = const()[name = string("op_26589_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26589_end_mask_0 = const()[name = string("op_26589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26589_cast_fp16 = slice_by_index(begin = var_26589_begin_0, end = var_26589_end_0, end_mask = var_26589_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26589_cast_fp16")]; + tensor var_26596_begin_0 = const()[name = string("op_26596_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26596_end_0 = const()[name = string("op_26596_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26596_end_mask_0 = const()[name = string("op_26596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26596_cast_fp16 = slice_by_index(begin = var_26596_begin_0, end = var_26596_end_0, end_mask = var_26596_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26596_cast_fp16")]; + tensor var_26603_begin_0 = const()[name = string("op_26603_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26603_end_0 = const()[name = string("op_26603_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26603_end_mask_0 = const()[name = string("op_26603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26603_cast_fp16 = slice_by_index(begin = var_26603_begin_0, end = var_26603_end_0, end_mask = var_26603_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26603_cast_fp16")]; + tensor var_26610_begin_0 = const()[name = string("op_26610_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26610_end_0 = const()[name = string("op_26610_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26610_end_mask_0 = const()[name = string("op_26610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26610_cast_fp16 = slice_by_index(begin = var_26610_begin_0, end = var_26610_end_0, end_mask = var_26610_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26610_cast_fp16")]; + tensor var_26617_begin_0 = const()[name = string("op_26617_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26617_end_0 = const()[name = string("op_26617_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26617_end_mask_0 = const()[name = string("op_26617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26617_cast_fp16 = slice_by_index(begin = var_26617_begin_0, end = var_26617_end_0, end_mask = var_26617_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26617_cast_fp16")]; + tensor var_26624_begin_0 = const()[name = string("op_26624_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26624_end_0 = const()[name = string("op_26624_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26624_end_mask_0 = const()[name = string("op_26624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26624_cast_fp16 = slice_by_index(begin = var_26624_begin_0, end = var_26624_end_0, end_mask = var_26624_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26624_cast_fp16")]; + tensor var_26631_begin_0 = const()[name = string("op_26631_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26631_end_0 = const()[name = string("op_26631_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26631_end_mask_0 = const()[name = string("op_26631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26631_cast_fp16 = slice_by_index(begin = var_26631_begin_0, end = var_26631_end_0, end_mask = var_26631_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26631_cast_fp16")]; + tensor var_26638_begin_0 = const()[name = string("op_26638_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26638_end_0 = const()[name = string("op_26638_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26638_end_mask_0 = const()[name = string("op_26638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26638_cast_fp16 = slice_by_index(begin = var_26638_begin_0, end = var_26638_end_0, end_mask = var_26638_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26638_cast_fp16")]; + tensor var_26645_begin_0 = const()[name = string("op_26645_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26645_end_0 = const()[name = string("op_26645_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26645_end_mask_0 = const()[name = string("op_26645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26645_cast_fp16 = slice_by_index(begin = var_26645_begin_0, end = var_26645_end_0, end_mask = var_26645_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26645_cast_fp16")]; + tensor var_26652_begin_0 = const()[name = string("op_26652_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26652_end_0 = const()[name = string("op_26652_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26652_end_mask_0 = const()[name = string("op_26652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26652_cast_fp16 = slice_by_index(begin = var_26652_begin_0, end = var_26652_end_0, end_mask = var_26652_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26652_cast_fp16")]; + tensor var_26659_begin_0 = const()[name = string("op_26659_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26659_end_0 = const()[name = string("op_26659_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26659_end_mask_0 = const()[name = string("op_26659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26659_cast_fp16 = slice_by_index(begin = var_26659_begin_0, end = var_26659_end_0, end_mask = var_26659_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26659_cast_fp16")]; + tensor var_26666_begin_0 = const()[name = string("op_26666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26666_end_0 = const()[name = string("op_26666_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26666_end_mask_0 = const()[name = string("op_26666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26666_cast_fp16 = slice_by_index(begin = var_26666_begin_0, end = var_26666_end_0, end_mask = var_26666_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26666_cast_fp16")]; + tensor var_26673_begin_0 = const()[name = string("op_26673_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26673_end_0 = const()[name = string("op_26673_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26673_end_mask_0 = const()[name = string("op_26673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26673_cast_fp16 = slice_by_index(begin = var_26673_begin_0, end = var_26673_end_0, end_mask = var_26673_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26673_cast_fp16")]; + tensor var_26680_begin_0 = const()[name = string("op_26680_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26680_end_0 = const()[name = string("op_26680_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26680_end_mask_0 = const()[name = string("op_26680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26680_cast_fp16 = slice_by_index(begin = var_26680_begin_0, end = var_26680_end_0, end_mask = var_26680_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26680_cast_fp16")]; + tensor var_26687_begin_0 = const()[name = string("op_26687_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26687_end_0 = const()[name = string("op_26687_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26687_end_mask_0 = const()[name = string("op_26687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26687_cast_fp16 = slice_by_index(begin = var_26687_begin_0, end = var_26687_end_0, end_mask = var_26687_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26687_cast_fp16")]; + tensor var_26694_begin_0 = const()[name = string("op_26694_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26694_end_0 = const()[name = string("op_26694_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26694_end_mask_0 = const()[name = string("op_26694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26694_cast_fp16 = slice_by_index(begin = var_26694_begin_0, end = var_26694_end_0, end_mask = var_26694_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26694_cast_fp16")]; + tensor var_26701_begin_0 = const()[name = string("op_26701_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26701_end_0 = const()[name = string("op_26701_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26701_end_mask_0 = const()[name = string("op_26701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26701_cast_fp16 = slice_by_index(begin = var_26701_begin_0, end = var_26701_end_0, end_mask = var_26701_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26701_cast_fp16")]; + tensor var_26708_begin_0 = const()[name = string("op_26708_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26708_end_0 = const()[name = string("op_26708_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26708_end_mask_0 = const()[name = string("op_26708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26708_cast_fp16 = slice_by_index(begin = var_26708_begin_0, end = var_26708_end_0, end_mask = var_26708_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26708_cast_fp16")]; + tensor var_26715_begin_0 = const()[name = string("op_26715_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26715_end_0 = const()[name = string("op_26715_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26715_end_mask_0 = const()[name = string("op_26715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26715_cast_fp16 = slice_by_index(begin = var_26715_begin_0, end = var_26715_end_0, end_mask = var_26715_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26715_cast_fp16")]; + tensor var_26722_begin_0 = const()[name = string("op_26722_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26722_end_0 = const()[name = string("op_26722_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26722_end_mask_0 = const()[name = string("op_26722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26722_cast_fp16 = slice_by_index(begin = var_26722_begin_0, end = var_26722_end_0, end_mask = var_26722_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26722_cast_fp16")]; + tensor var_26729_begin_0 = const()[name = string("op_26729_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26729_end_0 = const()[name = string("op_26729_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26729_end_mask_0 = const()[name = string("op_26729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26729_cast_fp16 = slice_by_index(begin = var_26729_begin_0, end = var_26729_end_0, end_mask = var_26729_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26729_cast_fp16")]; + tensor var_26736_begin_0 = const()[name = string("op_26736_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26736_end_0 = const()[name = string("op_26736_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26736_end_mask_0 = const()[name = string("op_26736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26736_cast_fp16 = slice_by_index(begin = var_26736_begin_0, end = var_26736_end_0, end_mask = var_26736_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26736_cast_fp16")]; + tensor var_26743_begin_0 = const()[name = string("op_26743_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26743_end_0 = const()[name = string("op_26743_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26743_end_mask_0 = const()[name = string("op_26743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26743_cast_fp16 = slice_by_index(begin = var_26743_begin_0, end = var_26743_end_0, end_mask = var_26743_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26743_cast_fp16")]; + tensor var_26750_begin_0 = const()[name = string("op_26750_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26750_end_0 = const()[name = string("op_26750_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26750_end_mask_0 = const()[name = string("op_26750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26750_cast_fp16 = slice_by_index(begin = var_26750_begin_0, end = var_26750_end_0, end_mask = var_26750_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26750_cast_fp16")]; + tensor var_26757_begin_0 = const()[name = string("op_26757_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26757_end_0 = const()[name = string("op_26757_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26757_end_mask_0 = const()[name = string("op_26757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26757_cast_fp16 = slice_by_index(begin = var_26757_begin_0, end = var_26757_end_0, end_mask = var_26757_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26757_cast_fp16")]; + tensor var_26764_begin_0 = const()[name = string("op_26764_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26764_end_0 = const()[name = string("op_26764_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26764_end_mask_0 = const()[name = string("op_26764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26764_cast_fp16 = slice_by_index(begin = var_26764_begin_0, end = var_26764_end_0, end_mask = var_26764_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26764_cast_fp16")]; + tensor var_26771_begin_0 = const()[name = string("op_26771_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26771_end_0 = const()[name = string("op_26771_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26771_end_mask_0 = const()[name = string("op_26771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26771_cast_fp16 = slice_by_index(begin = var_26771_begin_0, end = var_26771_end_0, end_mask = var_26771_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26771_cast_fp16")]; + tensor var_26778_begin_0 = const()[name = string("op_26778_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26778_end_0 = const()[name = string("op_26778_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26778_end_mask_0 = const()[name = string("op_26778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26778_cast_fp16 = slice_by_index(begin = var_26778_begin_0, end = var_26778_end_0, end_mask = var_26778_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26778_cast_fp16")]; + tensor var_26785_begin_0 = const()[name = string("op_26785_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26785_end_0 = const()[name = string("op_26785_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26785_end_mask_0 = const()[name = string("op_26785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26785_cast_fp16 = slice_by_index(begin = var_26785_begin_0, end = var_26785_end_0, end_mask = var_26785_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26785_cast_fp16")]; + tensor var_26792_begin_0 = const()[name = string("op_26792_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26792_end_0 = const()[name = string("op_26792_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26792_end_mask_0 = const()[name = string("op_26792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26792_cast_fp16 = slice_by_index(begin = var_26792_begin_0, end = var_26792_end_0, end_mask = var_26792_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26792_cast_fp16")]; + tensor var_26799_begin_0 = const()[name = string("op_26799_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26799_end_0 = const()[name = string("op_26799_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26799_end_mask_0 = const()[name = string("op_26799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26799_cast_fp16 = slice_by_index(begin = var_26799_begin_0, end = var_26799_end_0, end_mask = var_26799_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26799_cast_fp16")]; + tensor var_26806_begin_0 = const()[name = string("op_26806_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26806_end_0 = const()[name = string("op_26806_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26806_end_mask_0 = const()[name = string("op_26806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26806_cast_fp16 = slice_by_index(begin = var_26806_begin_0, end = var_26806_end_0, end_mask = var_26806_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26806_cast_fp16")]; + tensor var_26813_begin_0 = const()[name = string("op_26813_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26813_end_0 = const()[name = string("op_26813_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26813_end_mask_0 = const()[name = string("op_26813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26813_cast_fp16 = slice_by_index(begin = var_26813_begin_0, end = var_26813_end_0, end_mask = var_26813_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26813_cast_fp16")]; + tensor var_26820_begin_0 = const()[name = string("op_26820_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26820_end_0 = const()[name = string("op_26820_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26820_end_mask_0 = const()[name = string("op_26820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26820_cast_fp16 = slice_by_index(begin = var_26820_begin_0, end = var_26820_end_0, end_mask = var_26820_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26820_cast_fp16")]; + tensor var_26827_begin_0 = const()[name = string("op_26827_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26827_end_0 = const()[name = string("op_26827_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26827_end_mask_0 = const()[name = string("op_26827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26827_cast_fp16 = slice_by_index(begin = var_26827_begin_0, end = var_26827_end_0, end_mask = var_26827_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26827_cast_fp16")]; + tensor var_26834_begin_0 = const()[name = string("op_26834_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26834_end_0 = const()[name = string("op_26834_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26834_end_mask_0 = const()[name = string("op_26834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26834_cast_fp16 = slice_by_index(begin = var_26834_begin_0, end = var_26834_end_0, end_mask = var_26834_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26834_cast_fp16")]; + tensor var_26841_begin_0 = const()[name = string("op_26841_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26841_end_0 = const()[name = string("op_26841_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26841_end_mask_0 = const()[name = string("op_26841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26841_cast_fp16 = slice_by_index(begin = var_26841_begin_0, end = var_26841_end_0, end_mask = var_26841_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26841_cast_fp16")]; + tensor var_26848_begin_0 = const()[name = string("op_26848_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26848_end_0 = const()[name = string("op_26848_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26848_end_mask_0 = const()[name = string("op_26848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26848_cast_fp16 = slice_by_index(begin = var_26848_begin_0, end = var_26848_end_0, end_mask = var_26848_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26848_cast_fp16")]; + tensor var_26855_begin_0 = const()[name = string("op_26855_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26855_end_0 = const()[name = string("op_26855_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26855_end_mask_0 = const()[name = string("op_26855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26855_cast_fp16 = slice_by_index(begin = var_26855_begin_0, end = var_26855_end_0, end_mask = var_26855_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26855_cast_fp16")]; + tensor k_35_perm_0 = const()[name = string("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_26860_begin_0 = const()[name = string("op_26860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26860_end_0 = const()[name = string("op_26860_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_26860_end_mask_0 = const()[name = string("op_26860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_35_cast_fp16 = transpose(perm = k_35_perm_0, x = key_35_cast_fp16)[name = string("transpose_14")]; + tensor var_26860_cast_fp16 = slice_by_index(begin = var_26860_begin_0, end = var_26860_end_0, end_mask = var_26860_end_mask_0, x = k_35_cast_fp16)[name = string("op_26860_cast_fp16")]; + tensor var_26864_begin_0 = const()[name = string("op_26864_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_26864_end_0 = const()[name = string("op_26864_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_26864_end_mask_0 = const()[name = string("op_26864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26864_cast_fp16 = slice_by_index(begin = var_26864_begin_0, end = var_26864_end_0, end_mask = var_26864_end_mask_0, x = k_35_cast_fp16)[name = string("op_26864_cast_fp16")]; + tensor var_26868_begin_0 = const()[name = string("op_26868_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_26868_end_0 = const()[name = string("op_26868_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_26868_end_mask_0 = const()[name = string("op_26868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26868_cast_fp16 = slice_by_index(begin = var_26868_begin_0, end = var_26868_end_0, end_mask = var_26868_end_mask_0, x = k_35_cast_fp16)[name = string("op_26868_cast_fp16")]; + tensor var_26872_begin_0 = const()[name = string("op_26872_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_26872_end_0 = const()[name = string("op_26872_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_26872_end_mask_0 = const()[name = string("op_26872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26872_cast_fp16 = slice_by_index(begin = var_26872_begin_0, end = var_26872_end_0, end_mask = var_26872_end_mask_0, x = k_35_cast_fp16)[name = string("op_26872_cast_fp16")]; + tensor var_26876_begin_0 = const()[name = string("op_26876_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_26876_end_0 = const()[name = string("op_26876_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_26876_end_mask_0 = const()[name = string("op_26876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26876_cast_fp16 = slice_by_index(begin = var_26876_begin_0, end = var_26876_end_0, end_mask = var_26876_end_mask_0, x = k_35_cast_fp16)[name = string("op_26876_cast_fp16")]; + tensor var_26880_begin_0 = const()[name = string("op_26880_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_26880_end_0 = const()[name = string("op_26880_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_26880_end_mask_0 = const()[name = string("op_26880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26880_cast_fp16 = slice_by_index(begin = var_26880_begin_0, end = var_26880_end_0, end_mask = var_26880_end_mask_0, x = k_35_cast_fp16)[name = string("op_26880_cast_fp16")]; + tensor var_26884_begin_0 = const()[name = string("op_26884_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_26884_end_0 = const()[name = string("op_26884_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_26884_end_mask_0 = const()[name = string("op_26884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26884_cast_fp16 = slice_by_index(begin = var_26884_begin_0, end = var_26884_end_0, end_mask = var_26884_end_mask_0, x = k_35_cast_fp16)[name = string("op_26884_cast_fp16")]; + tensor var_26888_begin_0 = const()[name = string("op_26888_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_26888_end_0 = const()[name = string("op_26888_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_26888_end_mask_0 = const()[name = string("op_26888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26888_cast_fp16 = slice_by_index(begin = var_26888_begin_0, end = var_26888_end_0, end_mask = var_26888_end_mask_0, x = k_35_cast_fp16)[name = string("op_26888_cast_fp16")]; + tensor var_26892_begin_0 = const()[name = string("op_26892_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_26892_end_0 = const()[name = string("op_26892_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_26892_end_mask_0 = const()[name = string("op_26892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26892_cast_fp16 = slice_by_index(begin = var_26892_begin_0, end = var_26892_end_0, end_mask = var_26892_end_mask_0, x = k_35_cast_fp16)[name = string("op_26892_cast_fp16")]; + tensor var_26896_begin_0 = const()[name = string("op_26896_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_26896_end_0 = const()[name = string("op_26896_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_26896_end_mask_0 = const()[name = string("op_26896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26896_cast_fp16 = slice_by_index(begin = var_26896_begin_0, end = var_26896_end_0, end_mask = var_26896_end_mask_0, x = k_35_cast_fp16)[name = string("op_26896_cast_fp16")]; + tensor var_26900_begin_0 = const()[name = string("op_26900_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_26900_end_0 = const()[name = string("op_26900_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_26900_end_mask_0 = const()[name = string("op_26900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26900_cast_fp16 = slice_by_index(begin = var_26900_begin_0, end = var_26900_end_0, end_mask = var_26900_end_mask_0, x = k_35_cast_fp16)[name = string("op_26900_cast_fp16")]; + tensor var_26904_begin_0 = const()[name = string("op_26904_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_26904_end_0 = const()[name = string("op_26904_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_26904_end_mask_0 = const()[name = string("op_26904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26904_cast_fp16 = slice_by_index(begin = var_26904_begin_0, end = var_26904_end_0, end_mask = var_26904_end_mask_0, x = k_35_cast_fp16)[name = string("op_26904_cast_fp16")]; + tensor var_26908_begin_0 = const()[name = string("op_26908_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_26908_end_0 = const()[name = string("op_26908_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_26908_end_mask_0 = const()[name = string("op_26908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26908_cast_fp16 = slice_by_index(begin = var_26908_begin_0, end = var_26908_end_0, end_mask = var_26908_end_mask_0, x = k_35_cast_fp16)[name = string("op_26908_cast_fp16")]; + tensor var_26912_begin_0 = const()[name = string("op_26912_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_26912_end_0 = const()[name = string("op_26912_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_26912_end_mask_0 = const()[name = string("op_26912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26912_cast_fp16 = slice_by_index(begin = var_26912_begin_0, end = var_26912_end_0, end_mask = var_26912_end_mask_0, x = k_35_cast_fp16)[name = string("op_26912_cast_fp16")]; + tensor var_26916_begin_0 = const()[name = string("op_26916_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_26916_end_0 = const()[name = string("op_26916_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_26916_end_mask_0 = const()[name = string("op_26916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26916_cast_fp16 = slice_by_index(begin = var_26916_begin_0, end = var_26916_end_0, end_mask = var_26916_end_mask_0, x = k_35_cast_fp16)[name = string("op_26916_cast_fp16")]; + tensor var_26920_begin_0 = const()[name = string("op_26920_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_26920_end_0 = const()[name = string("op_26920_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_26920_end_mask_0 = const()[name = string("op_26920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26920_cast_fp16 = slice_by_index(begin = var_26920_begin_0, end = var_26920_end_0, end_mask = var_26920_end_mask_0, x = k_35_cast_fp16)[name = string("op_26920_cast_fp16")]; + tensor var_26924_begin_0 = const()[name = string("op_26924_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_26924_end_0 = const()[name = string("op_26924_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_26924_end_mask_0 = const()[name = string("op_26924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26924_cast_fp16 = slice_by_index(begin = var_26924_begin_0, end = var_26924_end_0, end_mask = var_26924_end_mask_0, x = k_35_cast_fp16)[name = string("op_26924_cast_fp16")]; + tensor var_26928_begin_0 = const()[name = string("op_26928_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_26928_end_0 = const()[name = string("op_26928_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_26928_end_mask_0 = const()[name = string("op_26928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26928_cast_fp16 = slice_by_index(begin = var_26928_begin_0, end = var_26928_end_0, end_mask = var_26928_end_mask_0, x = k_35_cast_fp16)[name = string("op_26928_cast_fp16")]; + tensor var_26932_begin_0 = const()[name = string("op_26932_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_26932_end_0 = const()[name = string("op_26932_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_26932_end_mask_0 = const()[name = string("op_26932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26932_cast_fp16 = slice_by_index(begin = var_26932_begin_0, end = var_26932_end_0, end_mask = var_26932_end_mask_0, x = k_35_cast_fp16)[name = string("op_26932_cast_fp16")]; + tensor var_26936_begin_0 = const()[name = string("op_26936_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_26936_end_0 = const()[name = string("op_26936_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_26936_end_mask_0 = const()[name = string("op_26936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26936_cast_fp16 = slice_by_index(begin = var_26936_begin_0, end = var_26936_end_0, end_mask = var_26936_end_mask_0, x = k_35_cast_fp16)[name = string("op_26936_cast_fp16")]; + tensor var_26938_begin_0 = const()[name = string("op_26938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26938_end_0 = const()[name = string("op_26938_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26938_end_mask_0 = const()[name = string("op_26938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26938_cast_fp16 = slice_by_index(begin = var_26938_begin_0, end = var_26938_end_0, end_mask = var_26938_end_mask_0, x = value_35_cast_fp16)[name = string("op_26938_cast_fp16")]; + tensor var_26942_begin_0 = const()[name = string("op_26942_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26942_end_0 = const()[name = string("op_26942_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_26942_end_mask_0 = const()[name = string("op_26942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26942_cast_fp16 = slice_by_index(begin = var_26942_begin_0, end = var_26942_end_0, end_mask = var_26942_end_mask_0, x = value_35_cast_fp16)[name = string("op_26942_cast_fp16")]; + tensor var_26946_begin_0 = const()[name = string("op_26946_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26946_end_0 = const()[name = string("op_26946_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_26946_end_mask_0 = const()[name = string("op_26946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26946_cast_fp16 = slice_by_index(begin = var_26946_begin_0, end = var_26946_end_0, end_mask = var_26946_end_mask_0, x = value_35_cast_fp16)[name = string("op_26946_cast_fp16")]; + tensor var_26950_begin_0 = const()[name = string("op_26950_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26950_end_0 = const()[name = string("op_26950_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_26950_end_mask_0 = const()[name = string("op_26950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26950_cast_fp16 = slice_by_index(begin = var_26950_begin_0, end = var_26950_end_0, end_mask = var_26950_end_mask_0, x = value_35_cast_fp16)[name = string("op_26950_cast_fp16")]; + tensor var_26954_begin_0 = const()[name = string("op_26954_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26954_end_0 = const()[name = string("op_26954_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_26954_end_mask_0 = const()[name = string("op_26954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26954_cast_fp16 = slice_by_index(begin = var_26954_begin_0, end = var_26954_end_0, end_mask = var_26954_end_mask_0, x = value_35_cast_fp16)[name = string("op_26954_cast_fp16")]; + tensor var_26958_begin_0 = const()[name = string("op_26958_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26958_end_0 = const()[name = string("op_26958_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_26958_end_mask_0 = const()[name = string("op_26958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26958_cast_fp16 = slice_by_index(begin = var_26958_begin_0, end = var_26958_end_0, end_mask = var_26958_end_mask_0, x = value_35_cast_fp16)[name = string("op_26958_cast_fp16")]; + tensor var_26962_begin_0 = const()[name = string("op_26962_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26962_end_0 = const()[name = string("op_26962_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_26962_end_mask_0 = const()[name = string("op_26962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26962_cast_fp16 = slice_by_index(begin = var_26962_begin_0, end = var_26962_end_0, end_mask = var_26962_end_mask_0, x = value_35_cast_fp16)[name = string("op_26962_cast_fp16")]; + tensor var_26966_begin_0 = const()[name = string("op_26966_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26966_end_0 = const()[name = string("op_26966_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_26966_end_mask_0 = const()[name = string("op_26966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26966_cast_fp16 = slice_by_index(begin = var_26966_begin_0, end = var_26966_end_0, end_mask = var_26966_end_mask_0, x = value_35_cast_fp16)[name = string("op_26966_cast_fp16")]; + tensor var_26970_begin_0 = const()[name = string("op_26970_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26970_end_0 = const()[name = string("op_26970_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_26970_end_mask_0 = const()[name = string("op_26970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26970_cast_fp16 = slice_by_index(begin = var_26970_begin_0, end = var_26970_end_0, end_mask = var_26970_end_mask_0, x = value_35_cast_fp16)[name = string("op_26970_cast_fp16")]; + tensor var_26974_begin_0 = const()[name = string("op_26974_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26974_end_0 = const()[name = string("op_26974_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_26974_end_mask_0 = const()[name = string("op_26974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26974_cast_fp16 = slice_by_index(begin = var_26974_begin_0, end = var_26974_end_0, end_mask = var_26974_end_mask_0, x = value_35_cast_fp16)[name = string("op_26974_cast_fp16")]; + tensor var_26978_begin_0 = const()[name = string("op_26978_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26978_end_0 = const()[name = string("op_26978_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_26978_end_mask_0 = const()[name = string("op_26978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26978_cast_fp16 = slice_by_index(begin = var_26978_begin_0, end = var_26978_end_0, end_mask = var_26978_end_mask_0, x = value_35_cast_fp16)[name = string("op_26978_cast_fp16")]; + tensor var_26982_begin_0 = const()[name = string("op_26982_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26982_end_0 = const()[name = string("op_26982_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_26982_end_mask_0 = const()[name = string("op_26982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26982_cast_fp16 = slice_by_index(begin = var_26982_begin_0, end = var_26982_end_0, end_mask = var_26982_end_mask_0, x = value_35_cast_fp16)[name = string("op_26982_cast_fp16")]; + tensor var_26986_begin_0 = const()[name = string("op_26986_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26986_end_0 = const()[name = string("op_26986_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_26986_end_mask_0 = const()[name = string("op_26986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26986_cast_fp16 = slice_by_index(begin = var_26986_begin_0, end = var_26986_end_0, end_mask = var_26986_end_mask_0, x = value_35_cast_fp16)[name = string("op_26986_cast_fp16")]; + tensor var_26990_begin_0 = const()[name = string("op_26990_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26990_end_0 = const()[name = string("op_26990_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_26990_end_mask_0 = const()[name = string("op_26990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26990_cast_fp16 = slice_by_index(begin = var_26990_begin_0, end = var_26990_end_0, end_mask = var_26990_end_mask_0, x = value_35_cast_fp16)[name = string("op_26990_cast_fp16")]; + tensor var_26994_begin_0 = const()[name = string("op_26994_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26994_end_0 = const()[name = string("op_26994_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_26994_end_mask_0 = const()[name = string("op_26994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26994_cast_fp16 = slice_by_index(begin = var_26994_begin_0, end = var_26994_end_0, end_mask = var_26994_end_mask_0, x = value_35_cast_fp16)[name = string("op_26994_cast_fp16")]; + tensor var_26998_begin_0 = const()[name = string("op_26998_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26998_end_0 = const()[name = string("op_26998_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_26998_end_mask_0 = const()[name = string("op_26998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26998_cast_fp16 = slice_by_index(begin = var_26998_begin_0, end = var_26998_end_0, end_mask = var_26998_end_mask_0, x = value_35_cast_fp16)[name = string("op_26998_cast_fp16")]; + tensor var_27002_begin_0 = const()[name = string("op_27002_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27002_end_0 = const()[name = string("op_27002_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_27002_end_mask_0 = const()[name = string("op_27002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27002_cast_fp16 = slice_by_index(begin = var_27002_begin_0, end = var_27002_end_0, end_mask = var_27002_end_mask_0, x = value_35_cast_fp16)[name = string("op_27002_cast_fp16")]; + tensor var_27006_begin_0 = const()[name = string("op_27006_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27006_end_0 = const()[name = string("op_27006_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_27006_end_mask_0 = const()[name = string("op_27006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27006_cast_fp16 = slice_by_index(begin = var_27006_begin_0, end = var_27006_end_0, end_mask = var_27006_end_mask_0, x = value_35_cast_fp16)[name = string("op_27006_cast_fp16")]; + tensor var_27010_begin_0 = const()[name = string("op_27010_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27010_end_0 = const()[name = string("op_27010_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_27010_end_mask_0 = const()[name = string("op_27010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27010_cast_fp16 = slice_by_index(begin = var_27010_begin_0, end = var_27010_end_0, end_mask = var_27010_end_mask_0, x = value_35_cast_fp16)[name = string("op_27010_cast_fp16")]; + tensor var_27014_begin_0 = const()[name = string("op_27014_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27014_end_0 = const()[name = string("op_27014_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_27014_end_mask_0 = const()[name = string("op_27014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27014_cast_fp16 = slice_by_index(begin = var_27014_begin_0, end = var_27014_end_0, end_mask = var_27014_end_mask_0, x = value_35_cast_fp16)[name = string("op_27014_cast_fp16")]; + string _SplitHeadsQ__mh_w_2721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2721_equation_0, values = (var_26860_cast_fp16, var_26302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2721_cast_fp16")]; + string _SplitHeadsQ__mh_w_2723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2723_equation_0, values = (var_26860_cast_fp16, var_26309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2723_cast_fp16")]; + string _SplitHeadsQ__mh_w_2725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2725_equation_0, values = (var_26860_cast_fp16, var_26316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2725_cast_fp16")]; + string _SplitHeadsQ__mh_w_2727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2727_equation_0, values = (var_26860_cast_fp16, var_26323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2727_cast_fp16")]; + string _SplitHeadsQ__mh_w_2729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2729_equation_0, values = (var_26864_cast_fp16, var_26330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2729_cast_fp16")]; + string _SplitHeadsQ__mh_w_2731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2731_equation_0, values = (var_26864_cast_fp16, var_26337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2731_cast_fp16")]; + string _SplitHeadsQ__mh_w_2733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2733_equation_0, values = (var_26864_cast_fp16, var_26344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2733_cast_fp16")]; + string _SplitHeadsQ__mh_w_2735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2735_equation_0, values = (var_26864_cast_fp16, var_26351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2735_cast_fp16")]; + string _SplitHeadsQ__mh_w_2737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2737_equation_0, values = (var_26868_cast_fp16, var_26358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2737_cast_fp16")]; + string _SplitHeadsQ__mh_w_2739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2739_equation_0, values = (var_26868_cast_fp16, var_26365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2739_cast_fp16")]; + string _SplitHeadsQ__mh_w_2741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2741_equation_0, values = (var_26868_cast_fp16, var_26372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2741_cast_fp16")]; + string _SplitHeadsQ__mh_w_2743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2743_equation_0, values = (var_26868_cast_fp16, var_26379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2743_cast_fp16")]; + string _SplitHeadsQ__mh_w_2745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2745_equation_0, values = (var_26872_cast_fp16, var_26386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2745_cast_fp16")]; + string _SplitHeadsQ__mh_w_2747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2747_equation_0, values = (var_26872_cast_fp16, var_26393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2747_cast_fp16")]; + string _SplitHeadsQ__mh_w_2749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2749_equation_0, values = (var_26872_cast_fp16, var_26400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2749_cast_fp16")]; + string _SplitHeadsQ__mh_w_2751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2751_equation_0, values = (var_26872_cast_fp16, var_26407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2751_cast_fp16")]; + string _SplitHeadsQ__mh_w_2753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2753_equation_0, values = (var_26876_cast_fp16, var_26414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2753_cast_fp16")]; + string _SplitHeadsQ__mh_w_2755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2755_equation_0, values = (var_26876_cast_fp16, var_26421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2755_cast_fp16")]; + string _SplitHeadsQ__mh_w_2757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2757_equation_0, values = (var_26876_cast_fp16, var_26428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2757_cast_fp16")]; + string _SplitHeadsQ__mh_w_2759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2759_equation_0, values = (var_26876_cast_fp16, var_26435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2759_cast_fp16")]; + string _SplitHeadsQ__mh_w_2761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2761_equation_0, values = (var_26880_cast_fp16, var_26442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2761_cast_fp16")]; + string _SplitHeadsQ__mh_w_2763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2763_equation_0, values = (var_26880_cast_fp16, var_26449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2763_cast_fp16")]; + string _SplitHeadsQ__mh_w_2765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2765_equation_0, values = (var_26880_cast_fp16, var_26456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2765_cast_fp16")]; + string _SplitHeadsQ__mh_w_2767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2767_equation_0, values = (var_26880_cast_fp16, var_26463_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2767_cast_fp16")]; + string _SplitHeadsQ__mh_w_2769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2769_equation_0, values = (var_26884_cast_fp16, var_26470_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2769_cast_fp16")]; + string _SplitHeadsQ__mh_w_2771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2771_equation_0, values = (var_26884_cast_fp16, var_26477_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2771_cast_fp16")]; + string _SplitHeadsQ__mh_w_2773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2773_equation_0, values = (var_26884_cast_fp16, var_26484_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2773_cast_fp16")]; + string _SplitHeadsQ__mh_w_2775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2775_equation_0, values = (var_26884_cast_fp16, var_26491_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2775_cast_fp16")]; + string _SplitHeadsQ__mh_w_2777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2777_equation_0, values = (var_26888_cast_fp16, var_26498_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2777_cast_fp16")]; + string _SplitHeadsQ__mh_w_2779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2779_equation_0, values = (var_26888_cast_fp16, var_26505_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2779_cast_fp16")]; + string _SplitHeadsQ__mh_w_2781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2781_equation_0, values = (var_26888_cast_fp16, var_26512_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2781_cast_fp16")]; + string _SplitHeadsQ__mh_w_2783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2783_equation_0, values = (var_26888_cast_fp16, var_26519_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2783_cast_fp16")]; + string _SplitHeadsQ__mh_w_2785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2785_equation_0, values = (var_26892_cast_fp16, var_26526_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2785_cast_fp16")]; + string _SplitHeadsQ__mh_w_2787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2787_equation_0, values = (var_26892_cast_fp16, var_26533_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2787_cast_fp16")]; + string _SplitHeadsQ__mh_w_2789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2789_equation_0, values = (var_26892_cast_fp16, var_26540_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2789_cast_fp16")]; + string _SplitHeadsQ__mh_w_2791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2791_equation_0, values = (var_26892_cast_fp16, var_26547_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2791_cast_fp16")]; + string _SplitHeadsQ__mh_w_2793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2793_equation_0, values = (var_26896_cast_fp16, var_26554_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2793_cast_fp16")]; + string _SplitHeadsQ__mh_w_2795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2795_equation_0, values = (var_26896_cast_fp16, var_26561_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2795_cast_fp16")]; + string _SplitHeadsQ__mh_w_2797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2797_equation_0, values = (var_26896_cast_fp16, var_26568_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2797_cast_fp16")]; + string _SplitHeadsQ__mh_w_2799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2799_equation_0, values = (var_26896_cast_fp16, var_26575_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2799_cast_fp16")]; + string _SplitHeadsQ__mh_w_2801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2801_equation_0, values = (var_26900_cast_fp16, var_26582_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2801_cast_fp16")]; + string _SplitHeadsQ__mh_w_2803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2803_equation_0, values = (var_26900_cast_fp16, var_26589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2803_cast_fp16")]; + string _SplitHeadsQ__mh_w_2805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2805_equation_0, values = (var_26900_cast_fp16, var_26596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2805_cast_fp16")]; + string _SplitHeadsQ__mh_w_2807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2807_equation_0, values = (var_26900_cast_fp16, var_26603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2807_cast_fp16")]; + string _SplitHeadsQ__mh_w_2809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2809_equation_0, values = (var_26904_cast_fp16, var_26610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2809_cast_fp16")]; + string _SplitHeadsQ__mh_w_2811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2811_equation_0, values = (var_26904_cast_fp16, var_26617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2811_cast_fp16")]; + string _SplitHeadsQ__mh_w_2813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2813_equation_0, values = (var_26904_cast_fp16, var_26624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2813_cast_fp16")]; + string _SplitHeadsQ__mh_w_2815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2815_equation_0, values = (var_26904_cast_fp16, var_26631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2815_cast_fp16")]; + string _SplitHeadsQ__mh_w_2817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2817_equation_0, values = (var_26908_cast_fp16, var_26638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2817_cast_fp16")]; + string _SplitHeadsQ__mh_w_2819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2819_equation_0, values = (var_26908_cast_fp16, var_26645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2819_cast_fp16")]; + string _SplitHeadsQ__mh_w_2821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2821_equation_0, values = (var_26908_cast_fp16, var_26652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2821_cast_fp16")]; + string _SplitHeadsQ__mh_w_2823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2823_equation_0, values = (var_26908_cast_fp16, var_26659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2823_cast_fp16")]; + string _SplitHeadsQ__mh_w_2825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2825_equation_0, values = (var_26912_cast_fp16, var_26666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2825_cast_fp16")]; + string _SplitHeadsQ__mh_w_2827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2827_equation_0, values = (var_26912_cast_fp16, var_26673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2827_cast_fp16")]; + string _SplitHeadsQ__mh_w_2829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2829_equation_0, values = (var_26912_cast_fp16, var_26680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2829_cast_fp16")]; + string _SplitHeadsQ__mh_w_2831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2831_equation_0, values = (var_26912_cast_fp16, var_26687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2831_cast_fp16")]; + string _SplitHeadsQ__mh_w_2833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2833_equation_0, values = (var_26916_cast_fp16, var_26694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2833_cast_fp16")]; + string _SplitHeadsQ__mh_w_2835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2835_equation_0, values = (var_26916_cast_fp16, var_26701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2835_cast_fp16")]; + string _SplitHeadsQ__mh_w_2837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2837_equation_0, values = (var_26916_cast_fp16, var_26708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2837_cast_fp16")]; + string _SplitHeadsQ__mh_w_2839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2839_equation_0, values = (var_26916_cast_fp16, var_26715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2839_cast_fp16")]; + string _SplitHeadsQ__mh_w_2841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2841_equation_0, values = (var_26920_cast_fp16, var_26722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2841_cast_fp16")]; + string _SplitHeadsQ__mh_w_2843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2843_equation_0, values = (var_26920_cast_fp16, var_26729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2843_cast_fp16")]; + string _SplitHeadsQ__mh_w_2845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2845_equation_0, values = (var_26920_cast_fp16, var_26736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2845_cast_fp16")]; + string _SplitHeadsQ__mh_w_2847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2847_equation_0, values = (var_26920_cast_fp16, var_26743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2847_cast_fp16")]; + string _SplitHeadsQ__mh_w_2849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2849_equation_0, values = (var_26924_cast_fp16, var_26750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2849_cast_fp16")]; + string _SplitHeadsQ__mh_w_2851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2851_equation_0, values = (var_26924_cast_fp16, var_26757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2851_cast_fp16")]; + string _SplitHeadsQ__mh_w_2853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2853_equation_0, values = (var_26924_cast_fp16, var_26764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2853_cast_fp16")]; + string _SplitHeadsQ__mh_w_2855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2855_equation_0, values = (var_26924_cast_fp16, var_26771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2855_cast_fp16")]; + string _SplitHeadsQ__mh_w_2857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2857_equation_0, values = (var_26928_cast_fp16, var_26778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2857_cast_fp16")]; + string _SplitHeadsQ__mh_w_2859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2859_equation_0, values = (var_26928_cast_fp16, var_26785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2859_cast_fp16")]; + string _SplitHeadsQ__mh_w_2861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2861_equation_0, values = (var_26928_cast_fp16, var_26792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2861_cast_fp16")]; + string _SplitHeadsQ__mh_w_2863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2863_equation_0, values = (var_26928_cast_fp16, var_26799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2863_cast_fp16")]; + string _SplitHeadsQ__mh_w_2865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2865_equation_0, values = (var_26932_cast_fp16, var_26806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2865_cast_fp16")]; + string _SplitHeadsQ__mh_w_2867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2867_equation_0, values = (var_26932_cast_fp16, var_26813_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2867_cast_fp16")]; + string _SplitHeadsQ__mh_w_2869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2869_equation_0, values = (var_26932_cast_fp16, var_26820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2869_cast_fp16")]; + string _SplitHeadsQ__mh_w_2871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2871_equation_0, values = (var_26932_cast_fp16, var_26827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2871_cast_fp16")]; + string _SplitHeadsQ__mh_w_2873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2873_equation_0, values = (var_26936_cast_fp16, var_26834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2873_cast_fp16")]; + string _SplitHeadsQ__mh_w_2875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2875_equation_0, values = (var_26936_cast_fp16, var_26841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2875_cast_fp16")]; + string _SplitHeadsQ__mh_w_2877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2877_equation_0, values = (var_26936_cast_fp16, var_26848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2877_cast_fp16")]; + string _SplitHeadsQ__mh_w_2879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2879_equation_0, values = (var_26936_cast_fp16, var_26855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2879_cast_fp16")]; + fp16 var_27177_to_fp16 = const()[name = string("op_27177_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2721_cast_fp16, y = var_27177_to_fp16)[name = string("aw_chunk_2721_cast_fp16")]; + fp16 var_27179_to_fp16 = const()[name = string("op_27179_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2723_cast_fp16, y = var_27179_to_fp16)[name = string("aw_chunk_2723_cast_fp16")]; + fp16 var_27181_to_fp16 = const()[name = string("op_27181_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2725_cast_fp16, y = var_27181_to_fp16)[name = string("aw_chunk_2725_cast_fp16")]; + fp16 var_27183_to_fp16 = const()[name = string("op_27183_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2727_cast_fp16, y = var_27183_to_fp16)[name = string("aw_chunk_2727_cast_fp16")]; + fp16 var_27185_to_fp16 = const()[name = string("op_27185_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2729_cast_fp16, y = var_27185_to_fp16)[name = string("aw_chunk_2729_cast_fp16")]; + fp16 var_27187_to_fp16 = const()[name = string("op_27187_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2731_cast_fp16, y = var_27187_to_fp16)[name = string("aw_chunk_2731_cast_fp16")]; + fp16 var_27189_to_fp16 = const()[name = string("op_27189_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2733_cast_fp16, y = var_27189_to_fp16)[name = string("aw_chunk_2733_cast_fp16")]; + fp16 var_27191_to_fp16 = const()[name = string("op_27191_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2735_cast_fp16, y = var_27191_to_fp16)[name = string("aw_chunk_2735_cast_fp16")]; + fp16 var_27193_to_fp16 = const()[name = string("op_27193_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2737_cast_fp16, y = var_27193_to_fp16)[name = string("aw_chunk_2737_cast_fp16")]; + fp16 var_27195_to_fp16 = const()[name = string("op_27195_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2739_cast_fp16, y = var_27195_to_fp16)[name = string("aw_chunk_2739_cast_fp16")]; + fp16 var_27197_to_fp16 = const()[name = string("op_27197_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2741_cast_fp16, y = var_27197_to_fp16)[name = string("aw_chunk_2741_cast_fp16")]; + fp16 var_27199_to_fp16 = const()[name = string("op_27199_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2743_cast_fp16, y = var_27199_to_fp16)[name = string("aw_chunk_2743_cast_fp16")]; + fp16 var_27201_to_fp16 = const()[name = string("op_27201_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2745_cast_fp16, y = var_27201_to_fp16)[name = string("aw_chunk_2745_cast_fp16")]; + fp16 var_27203_to_fp16 = const()[name = string("op_27203_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2747_cast_fp16, y = var_27203_to_fp16)[name = string("aw_chunk_2747_cast_fp16")]; + fp16 var_27205_to_fp16 = const()[name = string("op_27205_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2749_cast_fp16, y = var_27205_to_fp16)[name = string("aw_chunk_2749_cast_fp16")]; + fp16 var_27207_to_fp16 = const()[name = string("op_27207_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2751_cast_fp16, y = var_27207_to_fp16)[name = string("aw_chunk_2751_cast_fp16")]; + fp16 var_27209_to_fp16 = const()[name = string("op_27209_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2753_cast_fp16, y = var_27209_to_fp16)[name = string("aw_chunk_2753_cast_fp16")]; + fp16 var_27211_to_fp16 = const()[name = string("op_27211_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2755_cast_fp16, y = var_27211_to_fp16)[name = string("aw_chunk_2755_cast_fp16")]; + fp16 var_27213_to_fp16 = const()[name = string("op_27213_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2757_cast_fp16, y = var_27213_to_fp16)[name = string("aw_chunk_2757_cast_fp16")]; + fp16 var_27215_to_fp16 = const()[name = string("op_27215_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2759_cast_fp16, y = var_27215_to_fp16)[name = string("aw_chunk_2759_cast_fp16")]; + fp16 var_27217_to_fp16 = const()[name = string("op_27217_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2761_cast_fp16, y = var_27217_to_fp16)[name = string("aw_chunk_2761_cast_fp16")]; + fp16 var_27219_to_fp16 = const()[name = string("op_27219_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2763_cast_fp16, y = var_27219_to_fp16)[name = string("aw_chunk_2763_cast_fp16")]; + fp16 var_27221_to_fp16 = const()[name = string("op_27221_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2765_cast_fp16, y = var_27221_to_fp16)[name = string("aw_chunk_2765_cast_fp16")]; + fp16 var_27223_to_fp16 = const()[name = string("op_27223_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2767_cast_fp16, y = var_27223_to_fp16)[name = string("aw_chunk_2767_cast_fp16")]; + fp16 var_27225_to_fp16 = const()[name = string("op_27225_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2769_cast_fp16, y = var_27225_to_fp16)[name = string("aw_chunk_2769_cast_fp16")]; + fp16 var_27227_to_fp16 = const()[name = string("op_27227_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2771_cast_fp16, y = var_27227_to_fp16)[name = string("aw_chunk_2771_cast_fp16")]; + fp16 var_27229_to_fp16 = const()[name = string("op_27229_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2773_cast_fp16, y = var_27229_to_fp16)[name = string("aw_chunk_2773_cast_fp16")]; + fp16 var_27231_to_fp16 = const()[name = string("op_27231_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2775_cast_fp16, y = var_27231_to_fp16)[name = string("aw_chunk_2775_cast_fp16")]; + fp16 var_27233_to_fp16 = const()[name = string("op_27233_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2777_cast_fp16, y = var_27233_to_fp16)[name = string("aw_chunk_2777_cast_fp16")]; + fp16 var_27235_to_fp16 = const()[name = string("op_27235_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2779_cast_fp16, y = var_27235_to_fp16)[name = string("aw_chunk_2779_cast_fp16")]; + fp16 var_27237_to_fp16 = const()[name = string("op_27237_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2781_cast_fp16, y = var_27237_to_fp16)[name = string("aw_chunk_2781_cast_fp16")]; + fp16 var_27239_to_fp16 = const()[name = string("op_27239_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2783_cast_fp16, y = var_27239_to_fp16)[name = string("aw_chunk_2783_cast_fp16")]; + fp16 var_27241_to_fp16 = const()[name = string("op_27241_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2785_cast_fp16, y = var_27241_to_fp16)[name = string("aw_chunk_2785_cast_fp16")]; + fp16 var_27243_to_fp16 = const()[name = string("op_27243_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2787_cast_fp16, y = var_27243_to_fp16)[name = string("aw_chunk_2787_cast_fp16")]; + fp16 var_27245_to_fp16 = const()[name = string("op_27245_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2789_cast_fp16, y = var_27245_to_fp16)[name = string("aw_chunk_2789_cast_fp16")]; + fp16 var_27247_to_fp16 = const()[name = string("op_27247_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2791_cast_fp16, y = var_27247_to_fp16)[name = string("aw_chunk_2791_cast_fp16")]; + fp16 var_27249_to_fp16 = const()[name = string("op_27249_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2793_cast_fp16, y = var_27249_to_fp16)[name = string("aw_chunk_2793_cast_fp16")]; + fp16 var_27251_to_fp16 = const()[name = string("op_27251_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2795_cast_fp16, y = var_27251_to_fp16)[name = string("aw_chunk_2795_cast_fp16")]; + fp16 var_27253_to_fp16 = const()[name = string("op_27253_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2797_cast_fp16, y = var_27253_to_fp16)[name = string("aw_chunk_2797_cast_fp16")]; + fp16 var_27255_to_fp16 = const()[name = string("op_27255_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2799_cast_fp16, y = var_27255_to_fp16)[name = string("aw_chunk_2799_cast_fp16")]; + fp16 var_27257_to_fp16 = const()[name = string("op_27257_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2801_cast_fp16, y = var_27257_to_fp16)[name = string("aw_chunk_2801_cast_fp16")]; + fp16 var_27259_to_fp16 = const()[name = string("op_27259_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2803_cast_fp16, y = var_27259_to_fp16)[name = string("aw_chunk_2803_cast_fp16")]; + fp16 var_27261_to_fp16 = const()[name = string("op_27261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2805_cast_fp16, y = var_27261_to_fp16)[name = string("aw_chunk_2805_cast_fp16")]; + fp16 var_27263_to_fp16 = const()[name = string("op_27263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2807_cast_fp16, y = var_27263_to_fp16)[name = string("aw_chunk_2807_cast_fp16")]; + fp16 var_27265_to_fp16 = const()[name = string("op_27265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2809_cast_fp16, y = var_27265_to_fp16)[name = string("aw_chunk_2809_cast_fp16")]; + fp16 var_27267_to_fp16 = const()[name = string("op_27267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2811_cast_fp16, y = var_27267_to_fp16)[name = string("aw_chunk_2811_cast_fp16")]; + fp16 var_27269_to_fp16 = const()[name = string("op_27269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2813_cast_fp16, y = var_27269_to_fp16)[name = string("aw_chunk_2813_cast_fp16")]; + fp16 var_27271_to_fp16 = const()[name = string("op_27271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2815_cast_fp16, y = var_27271_to_fp16)[name = string("aw_chunk_2815_cast_fp16")]; + fp16 var_27273_to_fp16 = const()[name = string("op_27273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2817_cast_fp16, y = var_27273_to_fp16)[name = string("aw_chunk_2817_cast_fp16")]; + fp16 var_27275_to_fp16 = const()[name = string("op_27275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2819_cast_fp16, y = var_27275_to_fp16)[name = string("aw_chunk_2819_cast_fp16")]; + fp16 var_27277_to_fp16 = const()[name = string("op_27277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2821_cast_fp16, y = var_27277_to_fp16)[name = string("aw_chunk_2821_cast_fp16")]; + fp16 var_27279_to_fp16 = const()[name = string("op_27279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2823_cast_fp16, y = var_27279_to_fp16)[name = string("aw_chunk_2823_cast_fp16")]; + fp16 var_27281_to_fp16 = const()[name = string("op_27281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2825_cast_fp16, y = var_27281_to_fp16)[name = string("aw_chunk_2825_cast_fp16")]; + fp16 var_27283_to_fp16 = const()[name = string("op_27283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2827_cast_fp16, y = var_27283_to_fp16)[name = string("aw_chunk_2827_cast_fp16")]; + fp16 var_27285_to_fp16 = const()[name = string("op_27285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2829_cast_fp16, y = var_27285_to_fp16)[name = string("aw_chunk_2829_cast_fp16")]; + fp16 var_27287_to_fp16 = const()[name = string("op_27287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2831_cast_fp16, y = var_27287_to_fp16)[name = string("aw_chunk_2831_cast_fp16")]; + fp16 var_27289_to_fp16 = const()[name = string("op_27289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2833_cast_fp16, y = var_27289_to_fp16)[name = string("aw_chunk_2833_cast_fp16")]; + fp16 var_27291_to_fp16 = const()[name = string("op_27291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2835_cast_fp16, y = var_27291_to_fp16)[name = string("aw_chunk_2835_cast_fp16")]; + fp16 var_27293_to_fp16 = const()[name = string("op_27293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2837_cast_fp16, y = var_27293_to_fp16)[name = string("aw_chunk_2837_cast_fp16")]; + fp16 var_27295_to_fp16 = const()[name = string("op_27295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2839_cast_fp16, y = var_27295_to_fp16)[name = string("aw_chunk_2839_cast_fp16")]; + fp16 var_27297_to_fp16 = const()[name = string("op_27297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2841_cast_fp16, y = var_27297_to_fp16)[name = string("aw_chunk_2841_cast_fp16")]; + fp16 var_27299_to_fp16 = const()[name = string("op_27299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2843_cast_fp16, y = var_27299_to_fp16)[name = string("aw_chunk_2843_cast_fp16")]; + fp16 var_27301_to_fp16 = const()[name = string("op_27301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2845_cast_fp16, y = var_27301_to_fp16)[name = string("aw_chunk_2845_cast_fp16")]; + fp16 var_27303_to_fp16 = const()[name = string("op_27303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2847_cast_fp16, y = var_27303_to_fp16)[name = string("aw_chunk_2847_cast_fp16")]; + fp16 var_27305_to_fp16 = const()[name = string("op_27305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2849_cast_fp16, y = var_27305_to_fp16)[name = string("aw_chunk_2849_cast_fp16")]; + fp16 var_27307_to_fp16 = const()[name = string("op_27307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2851_cast_fp16, y = var_27307_to_fp16)[name = string("aw_chunk_2851_cast_fp16")]; + fp16 var_27309_to_fp16 = const()[name = string("op_27309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2853_cast_fp16, y = var_27309_to_fp16)[name = string("aw_chunk_2853_cast_fp16")]; + fp16 var_27311_to_fp16 = const()[name = string("op_27311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2855_cast_fp16, y = var_27311_to_fp16)[name = string("aw_chunk_2855_cast_fp16")]; + fp16 var_27313_to_fp16 = const()[name = string("op_27313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2857_cast_fp16, y = var_27313_to_fp16)[name = string("aw_chunk_2857_cast_fp16")]; + fp16 var_27315_to_fp16 = const()[name = string("op_27315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2859_cast_fp16, y = var_27315_to_fp16)[name = string("aw_chunk_2859_cast_fp16")]; + fp16 var_27317_to_fp16 = const()[name = string("op_27317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2861_cast_fp16, y = var_27317_to_fp16)[name = string("aw_chunk_2861_cast_fp16")]; + fp16 var_27319_to_fp16 = const()[name = string("op_27319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2863_cast_fp16, y = var_27319_to_fp16)[name = string("aw_chunk_2863_cast_fp16")]; + fp16 var_27321_to_fp16 = const()[name = string("op_27321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2865_cast_fp16, y = var_27321_to_fp16)[name = string("aw_chunk_2865_cast_fp16")]; + fp16 var_27323_to_fp16 = const()[name = string("op_27323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2867_cast_fp16, y = var_27323_to_fp16)[name = string("aw_chunk_2867_cast_fp16")]; + fp16 var_27325_to_fp16 = const()[name = string("op_27325_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2869_cast_fp16, y = var_27325_to_fp16)[name = string("aw_chunk_2869_cast_fp16")]; + fp16 var_27327_to_fp16 = const()[name = string("op_27327_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2871_cast_fp16, y = var_27327_to_fp16)[name = string("aw_chunk_2871_cast_fp16")]; + fp16 var_27329_to_fp16 = const()[name = string("op_27329_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2873_cast_fp16, y = var_27329_to_fp16)[name = string("aw_chunk_2873_cast_fp16")]; + fp16 var_27331_to_fp16 = const()[name = string("op_27331_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2875_cast_fp16, y = var_27331_to_fp16)[name = string("aw_chunk_2875_cast_fp16")]; + fp16 var_27333_to_fp16 = const()[name = string("op_27333_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2877_cast_fp16, y = var_27333_to_fp16)[name = string("aw_chunk_2877_cast_fp16")]; + fp16 var_27335_to_fp16 = const()[name = string("op_27335_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2879_cast_fp16, y = var_27335_to_fp16)[name = string("aw_chunk_2879_cast_fp16")]; + tensor var_27337_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2721_cast_fp16)[name = string("op_27337_cast_fp16")]; + tensor var_27338_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2723_cast_fp16)[name = string("op_27338_cast_fp16")]; + tensor var_27339_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2725_cast_fp16)[name = string("op_27339_cast_fp16")]; + tensor var_27340_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2727_cast_fp16)[name = string("op_27340_cast_fp16")]; + tensor var_27341_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2729_cast_fp16)[name = string("op_27341_cast_fp16")]; + tensor var_27342_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2731_cast_fp16)[name = string("op_27342_cast_fp16")]; + tensor var_27343_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2733_cast_fp16)[name = string("op_27343_cast_fp16")]; + tensor var_27344_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2735_cast_fp16)[name = string("op_27344_cast_fp16")]; + tensor var_27345_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2737_cast_fp16)[name = string("op_27345_cast_fp16")]; + tensor var_27346_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2739_cast_fp16)[name = string("op_27346_cast_fp16")]; + tensor var_27347_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2741_cast_fp16)[name = string("op_27347_cast_fp16")]; + tensor var_27348_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2743_cast_fp16)[name = string("op_27348_cast_fp16")]; + tensor var_27349_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2745_cast_fp16)[name = string("op_27349_cast_fp16")]; + tensor var_27350_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2747_cast_fp16)[name = string("op_27350_cast_fp16")]; + tensor var_27351_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2749_cast_fp16)[name = string("op_27351_cast_fp16")]; + tensor var_27352_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2751_cast_fp16)[name = string("op_27352_cast_fp16")]; + tensor var_27353_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2753_cast_fp16)[name = string("op_27353_cast_fp16")]; + tensor var_27354_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2755_cast_fp16)[name = string("op_27354_cast_fp16")]; + tensor var_27355_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2757_cast_fp16)[name = string("op_27355_cast_fp16")]; + tensor var_27356_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2759_cast_fp16)[name = string("op_27356_cast_fp16")]; + tensor var_27357_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2761_cast_fp16)[name = string("op_27357_cast_fp16")]; + tensor var_27358_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2763_cast_fp16)[name = string("op_27358_cast_fp16")]; + tensor var_27359_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2765_cast_fp16)[name = string("op_27359_cast_fp16")]; + tensor var_27360_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2767_cast_fp16)[name = string("op_27360_cast_fp16")]; + tensor var_27361_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2769_cast_fp16)[name = string("op_27361_cast_fp16")]; + tensor var_27362_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2771_cast_fp16)[name = string("op_27362_cast_fp16")]; + tensor var_27363_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2773_cast_fp16)[name = string("op_27363_cast_fp16")]; + tensor var_27364_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2775_cast_fp16)[name = string("op_27364_cast_fp16")]; + tensor var_27365_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2777_cast_fp16)[name = string("op_27365_cast_fp16")]; + tensor var_27366_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2779_cast_fp16)[name = string("op_27366_cast_fp16")]; + tensor var_27367_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2781_cast_fp16)[name = string("op_27367_cast_fp16")]; + tensor var_27368_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2783_cast_fp16)[name = string("op_27368_cast_fp16")]; + tensor var_27369_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2785_cast_fp16)[name = string("op_27369_cast_fp16")]; + tensor var_27370_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2787_cast_fp16)[name = string("op_27370_cast_fp16")]; + tensor var_27371_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2789_cast_fp16)[name = string("op_27371_cast_fp16")]; + tensor var_27372_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2791_cast_fp16)[name = string("op_27372_cast_fp16")]; + tensor var_27373_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2793_cast_fp16)[name = string("op_27373_cast_fp16")]; + tensor var_27374_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2795_cast_fp16)[name = string("op_27374_cast_fp16")]; + tensor var_27375_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2797_cast_fp16)[name = string("op_27375_cast_fp16")]; + tensor var_27376_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2799_cast_fp16)[name = string("op_27376_cast_fp16")]; + tensor var_27377_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2801_cast_fp16)[name = string("op_27377_cast_fp16")]; + tensor var_27378_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2803_cast_fp16)[name = string("op_27378_cast_fp16")]; + tensor var_27379_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2805_cast_fp16)[name = string("op_27379_cast_fp16")]; + tensor var_27380_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2807_cast_fp16)[name = string("op_27380_cast_fp16")]; + tensor var_27381_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2809_cast_fp16)[name = string("op_27381_cast_fp16")]; + tensor var_27382_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2811_cast_fp16)[name = string("op_27382_cast_fp16")]; + tensor var_27383_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2813_cast_fp16)[name = string("op_27383_cast_fp16")]; + tensor var_27384_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2815_cast_fp16)[name = string("op_27384_cast_fp16")]; + tensor var_27385_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2817_cast_fp16)[name = string("op_27385_cast_fp16")]; + tensor var_27386_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2819_cast_fp16)[name = string("op_27386_cast_fp16")]; + tensor var_27387_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2821_cast_fp16)[name = string("op_27387_cast_fp16")]; + tensor var_27388_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2823_cast_fp16)[name = string("op_27388_cast_fp16")]; + tensor var_27389_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2825_cast_fp16)[name = string("op_27389_cast_fp16")]; + tensor var_27390_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2827_cast_fp16)[name = string("op_27390_cast_fp16")]; + tensor var_27391_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2829_cast_fp16)[name = string("op_27391_cast_fp16")]; + tensor var_27392_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2831_cast_fp16)[name = string("op_27392_cast_fp16")]; + tensor var_27393_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2833_cast_fp16)[name = string("op_27393_cast_fp16")]; + tensor var_27394_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2835_cast_fp16)[name = string("op_27394_cast_fp16")]; + tensor var_27395_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2837_cast_fp16)[name = string("op_27395_cast_fp16")]; + tensor var_27396_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2839_cast_fp16)[name = string("op_27396_cast_fp16")]; + tensor var_27397_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2841_cast_fp16)[name = string("op_27397_cast_fp16")]; + tensor var_27398_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2843_cast_fp16)[name = string("op_27398_cast_fp16")]; + tensor var_27399_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2845_cast_fp16)[name = string("op_27399_cast_fp16")]; + tensor var_27400_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2847_cast_fp16)[name = string("op_27400_cast_fp16")]; + tensor var_27401_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2849_cast_fp16)[name = string("op_27401_cast_fp16")]; + tensor var_27402_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2851_cast_fp16)[name = string("op_27402_cast_fp16")]; + tensor var_27403_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2853_cast_fp16)[name = string("op_27403_cast_fp16")]; + tensor var_27404_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2855_cast_fp16)[name = string("op_27404_cast_fp16")]; + tensor var_27405_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2857_cast_fp16)[name = string("op_27405_cast_fp16")]; + tensor var_27406_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2859_cast_fp16)[name = string("op_27406_cast_fp16")]; + tensor var_27407_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2861_cast_fp16)[name = string("op_27407_cast_fp16")]; + tensor var_27408_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2863_cast_fp16)[name = string("op_27408_cast_fp16")]; + tensor var_27409_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2865_cast_fp16)[name = string("op_27409_cast_fp16")]; + tensor var_27410_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2867_cast_fp16)[name = string("op_27410_cast_fp16")]; + tensor var_27411_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2869_cast_fp16)[name = string("op_27411_cast_fp16")]; + tensor var_27412_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2871_cast_fp16)[name = string("op_27412_cast_fp16")]; + tensor var_27413_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2873_cast_fp16)[name = string("op_27413_cast_fp16")]; + tensor var_27414_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2875_cast_fp16)[name = string("op_27414_cast_fp16")]; + tensor var_27415_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2877_cast_fp16)[name = string("op_27415_cast_fp16")]; + tensor var_27416_cast_fp16 = softmax(axis = var_26162, x = aw_chunk_2879_cast_fp16)[name = string("op_27416_cast_fp16")]; + string var_27418_equation_0 = const()[name = string("op_27418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27418_cast_fp16 = einsum(equation = var_27418_equation_0, values = (var_26938_cast_fp16, var_27337_cast_fp16))[name = string("op_27418_cast_fp16")]; + string var_27420_equation_0 = const()[name = string("op_27420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27420_cast_fp16 = einsum(equation = var_27420_equation_0, values = (var_26938_cast_fp16, var_27338_cast_fp16))[name = string("op_27420_cast_fp16")]; + string var_27422_equation_0 = const()[name = string("op_27422_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27422_cast_fp16 = einsum(equation = var_27422_equation_0, values = (var_26938_cast_fp16, var_27339_cast_fp16))[name = string("op_27422_cast_fp16")]; + string var_27424_equation_0 = const()[name = string("op_27424_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27424_cast_fp16 = einsum(equation = var_27424_equation_0, values = (var_26938_cast_fp16, var_27340_cast_fp16))[name = string("op_27424_cast_fp16")]; + string var_27426_equation_0 = const()[name = string("op_27426_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27426_cast_fp16 = einsum(equation = var_27426_equation_0, values = (var_26942_cast_fp16, var_27341_cast_fp16))[name = string("op_27426_cast_fp16")]; + string var_27428_equation_0 = const()[name = string("op_27428_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27428_cast_fp16 = einsum(equation = var_27428_equation_0, values = (var_26942_cast_fp16, var_27342_cast_fp16))[name = string("op_27428_cast_fp16")]; + string var_27430_equation_0 = const()[name = string("op_27430_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27430_cast_fp16 = einsum(equation = var_27430_equation_0, values = (var_26942_cast_fp16, var_27343_cast_fp16))[name = string("op_27430_cast_fp16")]; + string var_27432_equation_0 = const()[name = string("op_27432_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27432_cast_fp16 = einsum(equation = var_27432_equation_0, values = (var_26942_cast_fp16, var_27344_cast_fp16))[name = string("op_27432_cast_fp16")]; + string var_27434_equation_0 = const()[name = string("op_27434_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27434_cast_fp16 = einsum(equation = var_27434_equation_0, values = (var_26946_cast_fp16, var_27345_cast_fp16))[name = string("op_27434_cast_fp16")]; + string var_27436_equation_0 = const()[name = string("op_27436_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27436_cast_fp16 = einsum(equation = var_27436_equation_0, values = (var_26946_cast_fp16, var_27346_cast_fp16))[name = string("op_27436_cast_fp16")]; + string var_27438_equation_0 = const()[name = string("op_27438_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27438_cast_fp16 = einsum(equation = var_27438_equation_0, values = (var_26946_cast_fp16, var_27347_cast_fp16))[name = string("op_27438_cast_fp16")]; + string var_27440_equation_0 = const()[name = string("op_27440_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27440_cast_fp16 = einsum(equation = var_27440_equation_0, values = (var_26946_cast_fp16, var_27348_cast_fp16))[name = string("op_27440_cast_fp16")]; + string var_27442_equation_0 = const()[name = string("op_27442_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27442_cast_fp16 = einsum(equation = var_27442_equation_0, values = (var_26950_cast_fp16, var_27349_cast_fp16))[name = string("op_27442_cast_fp16")]; + string var_27444_equation_0 = const()[name = string("op_27444_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27444_cast_fp16 = einsum(equation = var_27444_equation_0, values = (var_26950_cast_fp16, var_27350_cast_fp16))[name = string("op_27444_cast_fp16")]; + string var_27446_equation_0 = const()[name = string("op_27446_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27446_cast_fp16 = einsum(equation = var_27446_equation_0, values = (var_26950_cast_fp16, var_27351_cast_fp16))[name = string("op_27446_cast_fp16")]; + string var_27448_equation_0 = const()[name = string("op_27448_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27448_cast_fp16 = einsum(equation = var_27448_equation_0, values = (var_26950_cast_fp16, var_27352_cast_fp16))[name = string("op_27448_cast_fp16")]; + string var_27450_equation_0 = const()[name = string("op_27450_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27450_cast_fp16 = einsum(equation = var_27450_equation_0, values = (var_26954_cast_fp16, var_27353_cast_fp16))[name = string("op_27450_cast_fp16")]; + string var_27452_equation_0 = const()[name = string("op_27452_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27452_cast_fp16 = einsum(equation = var_27452_equation_0, values = (var_26954_cast_fp16, var_27354_cast_fp16))[name = string("op_27452_cast_fp16")]; + string var_27454_equation_0 = const()[name = string("op_27454_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27454_cast_fp16 = einsum(equation = var_27454_equation_0, values = (var_26954_cast_fp16, var_27355_cast_fp16))[name = string("op_27454_cast_fp16")]; + string var_27456_equation_0 = const()[name = string("op_27456_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27456_cast_fp16 = einsum(equation = var_27456_equation_0, values = (var_26954_cast_fp16, var_27356_cast_fp16))[name = string("op_27456_cast_fp16")]; + string var_27458_equation_0 = const()[name = string("op_27458_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27458_cast_fp16 = einsum(equation = var_27458_equation_0, values = (var_26958_cast_fp16, var_27357_cast_fp16))[name = string("op_27458_cast_fp16")]; + string var_27460_equation_0 = const()[name = string("op_27460_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27460_cast_fp16 = einsum(equation = var_27460_equation_0, values = (var_26958_cast_fp16, var_27358_cast_fp16))[name = string("op_27460_cast_fp16")]; + string var_27462_equation_0 = const()[name = string("op_27462_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27462_cast_fp16 = einsum(equation = var_27462_equation_0, values = (var_26958_cast_fp16, var_27359_cast_fp16))[name = string("op_27462_cast_fp16")]; + string var_27464_equation_0 = const()[name = string("op_27464_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27464_cast_fp16 = einsum(equation = var_27464_equation_0, values = (var_26958_cast_fp16, var_27360_cast_fp16))[name = string("op_27464_cast_fp16")]; + string var_27466_equation_0 = const()[name = string("op_27466_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27466_cast_fp16 = einsum(equation = var_27466_equation_0, values = (var_26962_cast_fp16, var_27361_cast_fp16))[name = string("op_27466_cast_fp16")]; + string var_27468_equation_0 = const()[name = string("op_27468_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27468_cast_fp16 = einsum(equation = var_27468_equation_0, values = (var_26962_cast_fp16, var_27362_cast_fp16))[name = string("op_27468_cast_fp16")]; + string var_27470_equation_0 = const()[name = string("op_27470_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27470_cast_fp16 = einsum(equation = var_27470_equation_0, values = (var_26962_cast_fp16, var_27363_cast_fp16))[name = string("op_27470_cast_fp16")]; + string var_27472_equation_0 = const()[name = string("op_27472_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27472_cast_fp16 = einsum(equation = var_27472_equation_0, values = (var_26962_cast_fp16, var_27364_cast_fp16))[name = string("op_27472_cast_fp16")]; + string var_27474_equation_0 = const()[name = string("op_27474_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27474_cast_fp16 = einsum(equation = var_27474_equation_0, values = (var_26966_cast_fp16, var_27365_cast_fp16))[name = string("op_27474_cast_fp16")]; + string var_27476_equation_0 = const()[name = string("op_27476_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27476_cast_fp16 = einsum(equation = var_27476_equation_0, values = (var_26966_cast_fp16, var_27366_cast_fp16))[name = string("op_27476_cast_fp16")]; + string var_27478_equation_0 = const()[name = string("op_27478_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27478_cast_fp16 = einsum(equation = var_27478_equation_0, values = (var_26966_cast_fp16, var_27367_cast_fp16))[name = string("op_27478_cast_fp16")]; + string var_27480_equation_0 = const()[name = string("op_27480_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27480_cast_fp16 = einsum(equation = var_27480_equation_0, values = (var_26966_cast_fp16, var_27368_cast_fp16))[name = string("op_27480_cast_fp16")]; + string var_27482_equation_0 = const()[name = string("op_27482_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27482_cast_fp16 = einsum(equation = var_27482_equation_0, values = (var_26970_cast_fp16, var_27369_cast_fp16))[name = string("op_27482_cast_fp16")]; + string var_27484_equation_0 = const()[name = string("op_27484_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27484_cast_fp16 = einsum(equation = var_27484_equation_0, values = (var_26970_cast_fp16, var_27370_cast_fp16))[name = string("op_27484_cast_fp16")]; + string var_27486_equation_0 = const()[name = string("op_27486_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27486_cast_fp16 = einsum(equation = var_27486_equation_0, values = (var_26970_cast_fp16, var_27371_cast_fp16))[name = string("op_27486_cast_fp16")]; + string var_27488_equation_0 = const()[name = string("op_27488_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27488_cast_fp16 = einsum(equation = var_27488_equation_0, values = (var_26970_cast_fp16, var_27372_cast_fp16))[name = string("op_27488_cast_fp16")]; + string var_27490_equation_0 = const()[name = string("op_27490_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27490_cast_fp16 = einsum(equation = var_27490_equation_0, values = (var_26974_cast_fp16, var_27373_cast_fp16))[name = string("op_27490_cast_fp16")]; + string var_27492_equation_0 = const()[name = string("op_27492_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27492_cast_fp16 = einsum(equation = var_27492_equation_0, values = (var_26974_cast_fp16, var_27374_cast_fp16))[name = string("op_27492_cast_fp16")]; + string var_27494_equation_0 = const()[name = string("op_27494_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27494_cast_fp16 = einsum(equation = var_27494_equation_0, values = (var_26974_cast_fp16, var_27375_cast_fp16))[name = string("op_27494_cast_fp16")]; + string var_27496_equation_0 = const()[name = string("op_27496_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27496_cast_fp16 = einsum(equation = var_27496_equation_0, values = (var_26974_cast_fp16, var_27376_cast_fp16))[name = string("op_27496_cast_fp16")]; + string var_27498_equation_0 = const()[name = string("op_27498_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27498_cast_fp16 = einsum(equation = var_27498_equation_0, values = (var_26978_cast_fp16, var_27377_cast_fp16))[name = string("op_27498_cast_fp16")]; + string var_27500_equation_0 = const()[name = string("op_27500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27500_cast_fp16 = einsum(equation = var_27500_equation_0, values = (var_26978_cast_fp16, var_27378_cast_fp16))[name = string("op_27500_cast_fp16")]; + string var_27502_equation_0 = const()[name = string("op_27502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27502_cast_fp16 = einsum(equation = var_27502_equation_0, values = (var_26978_cast_fp16, var_27379_cast_fp16))[name = string("op_27502_cast_fp16")]; + string var_27504_equation_0 = const()[name = string("op_27504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27504_cast_fp16 = einsum(equation = var_27504_equation_0, values = (var_26978_cast_fp16, var_27380_cast_fp16))[name = string("op_27504_cast_fp16")]; + string var_27506_equation_0 = const()[name = string("op_27506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27506_cast_fp16 = einsum(equation = var_27506_equation_0, values = (var_26982_cast_fp16, var_27381_cast_fp16))[name = string("op_27506_cast_fp16")]; + string var_27508_equation_0 = const()[name = string("op_27508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27508_cast_fp16 = einsum(equation = var_27508_equation_0, values = (var_26982_cast_fp16, var_27382_cast_fp16))[name = string("op_27508_cast_fp16")]; + string var_27510_equation_0 = const()[name = string("op_27510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27510_cast_fp16 = einsum(equation = var_27510_equation_0, values = (var_26982_cast_fp16, var_27383_cast_fp16))[name = string("op_27510_cast_fp16")]; + string var_27512_equation_0 = const()[name = string("op_27512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27512_cast_fp16 = einsum(equation = var_27512_equation_0, values = (var_26982_cast_fp16, var_27384_cast_fp16))[name = string("op_27512_cast_fp16")]; + string var_27514_equation_0 = const()[name = string("op_27514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27514_cast_fp16 = einsum(equation = var_27514_equation_0, values = (var_26986_cast_fp16, var_27385_cast_fp16))[name = string("op_27514_cast_fp16")]; + string var_27516_equation_0 = const()[name = string("op_27516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27516_cast_fp16 = einsum(equation = var_27516_equation_0, values = (var_26986_cast_fp16, var_27386_cast_fp16))[name = string("op_27516_cast_fp16")]; + string var_27518_equation_0 = const()[name = string("op_27518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27518_cast_fp16 = einsum(equation = var_27518_equation_0, values = (var_26986_cast_fp16, var_27387_cast_fp16))[name = string("op_27518_cast_fp16")]; + string var_27520_equation_0 = const()[name = string("op_27520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27520_cast_fp16 = einsum(equation = var_27520_equation_0, values = (var_26986_cast_fp16, var_27388_cast_fp16))[name = string("op_27520_cast_fp16")]; + string var_27522_equation_0 = const()[name = string("op_27522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27522_cast_fp16 = einsum(equation = var_27522_equation_0, values = (var_26990_cast_fp16, var_27389_cast_fp16))[name = string("op_27522_cast_fp16")]; + string var_27524_equation_0 = const()[name = string("op_27524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27524_cast_fp16 = einsum(equation = var_27524_equation_0, values = (var_26990_cast_fp16, var_27390_cast_fp16))[name = string("op_27524_cast_fp16")]; + string var_27526_equation_0 = const()[name = string("op_27526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27526_cast_fp16 = einsum(equation = var_27526_equation_0, values = (var_26990_cast_fp16, var_27391_cast_fp16))[name = string("op_27526_cast_fp16")]; + string var_27528_equation_0 = const()[name = string("op_27528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27528_cast_fp16 = einsum(equation = var_27528_equation_0, values = (var_26990_cast_fp16, var_27392_cast_fp16))[name = string("op_27528_cast_fp16")]; + string var_27530_equation_0 = const()[name = string("op_27530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27530_cast_fp16 = einsum(equation = var_27530_equation_0, values = (var_26994_cast_fp16, var_27393_cast_fp16))[name = string("op_27530_cast_fp16")]; + string var_27532_equation_0 = const()[name = string("op_27532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27532_cast_fp16 = einsum(equation = var_27532_equation_0, values = (var_26994_cast_fp16, var_27394_cast_fp16))[name = string("op_27532_cast_fp16")]; + string var_27534_equation_0 = const()[name = string("op_27534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27534_cast_fp16 = einsum(equation = var_27534_equation_0, values = (var_26994_cast_fp16, var_27395_cast_fp16))[name = string("op_27534_cast_fp16")]; + string var_27536_equation_0 = const()[name = string("op_27536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27536_cast_fp16 = einsum(equation = var_27536_equation_0, values = (var_26994_cast_fp16, var_27396_cast_fp16))[name = string("op_27536_cast_fp16")]; + string var_27538_equation_0 = const()[name = string("op_27538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27538_cast_fp16 = einsum(equation = var_27538_equation_0, values = (var_26998_cast_fp16, var_27397_cast_fp16))[name = string("op_27538_cast_fp16")]; + string var_27540_equation_0 = const()[name = string("op_27540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27540_cast_fp16 = einsum(equation = var_27540_equation_0, values = (var_26998_cast_fp16, var_27398_cast_fp16))[name = string("op_27540_cast_fp16")]; + string var_27542_equation_0 = const()[name = string("op_27542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27542_cast_fp16 = einsum(equation = var_27542_equation_0, values = (var_26998_cast_fp16, var_27399_cast_fp16))[name = string("op_27542_cast_fp16")]; + string var_27544_equation_0 = const()[name = string("op_27544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27544_cast_fp16 = einsum(equation = var_27544_equation_0, values = (var_26998_cast_fp16, var_27400_cast_fp16))[name = string("op_27544_cast_fp16")]; + string var_27546_equation_0 = const()[name = string("op_27546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27546_cast_fp16 = einsum(equation = var_27546_equation_0, values = (var_27002_cast_fp16, var_27401_cast_fp16))[name = string("op_27546_cast_fp16")]; + string var_27548_equation_0 = const()[name = string("op_27548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27548_cast_fp16 = einsum(equation = var_27548_equation_0, values = (var_27002_cast_fp16, var_27402_cast_fp16))[name = string("op_27548_cast_fp16")]; + string var_27550_equation_0 = const()[name = string("op_27550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27550_cast_fp16 = einsum(equation = var_27550_equation_0, values = (var_27002_cast_fp16, var_27403_cast_fp16))[name = string("op_27550_cast_fp16")]; + string var_27552_equation_0 = const()[name = string("op_27552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27552_cast_fp16 = einsum(equation = var_27552_equation_0, values = (var_27002_cast_fp16, var_27404_cast_fp16))[name = string("op_27552_cast_fp16")]; + string var_27554_equation_0 = const()[name = string("op_27554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27554_cast_fp16 = einsum(equation = var_27554_equation_0, values = (var_27006_cast_fp16, var_27405_cast_fp16))[name = string("op_27554_cast_fp16")]; + string var_27556_equation_0 = const()[name = string("op_27556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27556_cast_fp16 = einsum(equation = var_27556_equation_0, values = (var_27006_cast_fp16, var_27406_cast_fp16))[name = string("op_27556_cast_fp16")]; + string var_27558_equation_0 = const()[name = string("op_27558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27558_cast_fp16 = einsum(equation = var_27558_equation_0, values = (var_27006_cast_fp16, var_27407_cast_fp16))[name = string("op_27558_cast_fp16")]; + string var_27560_equation_0 = const()[name = string("op_27560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27560_cast_fp16 = einsum(equation = var_27560_equation_0, values = (var_27006_cast_fp16, var_27408_cast_fp16))[name = string("op_27560_cast_fp16")]; + string var_27562_equation_0 = const()[name = string("op_27562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27562_cast_fp16 = einsum(equation = var_27562_equation_0, values = (var_27010_cast_fp16, var_27409_cast_fp16))[name = string("op_27562_cast_fp16")]; + string var_27564_equation_0 = const()[name = string("op_27564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27564_cast_fp16 = einsum(equation = var_27564_equation_0, values = (var_27010_cast_fp16, var_27410_cast_fp16))[name = string("op_27564_cast_fp16")]; + string var_27566_equation_0 = const()[name = string("op_27566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27566_cast_fp16 = einsum(equation = var_27566_equation_0, values = (var_27010_cast_fp16, var_27411_cast_fp16))[name = string("op_27566_cast_fp16")]; + string var_27568_equation_0 = const()[name = string("op_27568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27568_cast_fp16 = einsum(equation = var_27568_equation_0, values = (var_27010_cast_fp16, var_27412_cast_fp16))[name = string("op_27568_cast_fp16")]; + string var_27570_equation_0 = const()[name = string("op_27570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27570_cast_fp16 = einsum(equation = var_27570_equation_0, values = (var_27014_cast_fp16, var_27413_cast_fp16))[name = string("op_27570_cast_fp16")]; + string var_27572_equation_0 = const()[name = string("op_27572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27572_cast_fp16 = einsum(equation = var_27572_equation_0, values = (var_27014_cast_fp16, var_27414_cast_fp16))[name = string("op_27572_cast_fp16")]; + string var_27574_equation_0 = const()[name = string("op_27574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27574_cast_fp16 = einsum(equation = var_27574_equation_0, values = (var_27014_cast_fp16, var_27415_cast_fp16))[name = string("op_27574_cast_fp16")]; + string var_27576_equation_0 = const()[name = string("op_27576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27576_cast_fp16 = einsum(equation = var_27576_equation_0, values = (var_27014_cast_fp16, var_27416_cast_fp16))[name = string("op_27576_cast_fp16")]; + bool var_27578_interleave_0 = const()[name = string("op_27578_interleave_0"), val = bool(false)]; + tensor var_27578_cast_fp16 = concat(axis = var_26137, interleave = var_27578_interleave_0, values = (var_27418_cast_fp16, var_27420_cast_fp16, var_27422_cast_fp16, var_27424_cast_fp16))[name = string("op_27578_cast_fp16")]; + bool var_27580_interleave_0 = const()[name = string("op_27580_interleave_0"), val = bool(false)]; + tensor var_27580_cast_fp16 = concat(axis = var_26137, interleave = var_27580_interleave_0, values = (var_27426_cast_fp16, var_27428_cast_fp16, var_27430_cast_fp16, var_27432_cast_fp16))[name = string("op_27580_cast_fp16")]; + bool var_27582_interleave_0 = const()[name = string("op_27582_interleave_0"), val = bool(false)]; + tensor var_27582_cast_fp16 = concat(axis = var_26137, interleave = var_27582_interleave_0, values = (var_27434_cast_fp16, var_27436_cast_fp16, var_27438_cast_fp16, var_27440_cast_fp16))[name = string("op_27582_cast_fp16")]; + bool var_27584_interleave_0 = const()[name = string("op_27584_interleave_0"), val = bool(false)]; + tensor var_27584_cast_fp16 = concat(axis = var_26137, interleave = var_27584_interleave_0, values = (var_27442_cast_fp16, var_27444_cast_fp16, var_27446_cast_fp16, var_27448_cast_fp16))[name = string("op_27584_cast_fp16")]; + bool var_27586_interleave_0 = const()[name = string("op_27586_interleave_0"), val = bool(false)]; + tensor var_27586_cast_fp16 = concat(axis = var_26137, interleave = var_27586_interleave_0, values = (var_27450_cast_fp16, var_27452_cast_fp16, var_27454_cast_fp16, var_27456_cast_fp16))[name = string("op_27586_cast_fp16")]; + bool var_27588_interleave_0 = const()[name = string("op_27588_interleave_0"), val = bool(false)]; + tensor var_27588_cast_fp16 = concat(axis = var_26137, interleave = var_27588_interleave_0, values = (var_27458_cast_fp16, var_27460_cast_fp16, var_27462_cast_fp16, var_27464_cast_fp16))[name = string("op_27588_cast_fp16")]; + bool var_27590_interleave_0 = const()[name = string("op_27590_interleave_0"), val = bool(false)]; + tensor var_27590_cast_fp16 = concat(axis = var_26137, interleave = var_27590_interleave_0, values = (var_27466_cast_fp16, var_27468_cast_fp16, var_27470_cast_fp16, var_27472_cast_fp16))[name = string("op_27590_cast_fp16")]; + bool var_27592_interleave_0 = const()[name = string("op_27592_interleave_0"), val = bool(false)]; + tensor var_27592_cast_fp16 = concat(axis = var_26137, interleave = var_27592_interleave_0, values = (var_27474_cast_fp16, var_27476_cast_fp16, var_27478_cast_fp16, var_27480_cast_fp16))[name = string("op_27592_cast_fp16")]; + bool var_27594_interleave_0 = const()[name = string("op_27594_interleave_0"), val = bool(false)]; + tensor var_27594_cast_fp16 = concat(axis = var_26137, interleave = var_27594_interleave_0, values = (var_27482_cast_fp16, var_27484_cast_fp16, var_27486_cast_fp16, var_27488_cast_fp16))[name = string("op_27594_cast_fp16")]; + bool var_27596_interleave_0 = const()[name = string("op_27596_interleave_0"), val = bool(false)]; + tensor var_27596_cast_fp16 = concat(axis = var_26137, interleave = var_27596_interleave_0, values = (var_27490_cast_fp16, var_27492_cast_fp16, var_27494_cast_fp16, var_27496_cast_fp16))[name = string("op_27596_cast_fp16")]; + bool var_27598_interleave_0 = const()[name = string("op_27598_interleave_0"), val = bool(false)]; + tensor var_27598_cast_fp16 = concat(axis = var_26137, interleave = var_27598_interleave_0, values = (var_27498_cast_fp16, var_27500_cast_fp16, var_27502_cast_fp16, var_27504_cast_fp16))[name = string("op_27598_cast_fp16")]; + bool var_27600_interleave_0 = const()[name = string("op_27600_interleave_0"), val = bool(false)]; + tensor var_27600_cast_fp16 = concat(axis = var_26137, interleave = var_27600_interleave_0, values = (var_27506_cast_fp16, var_27508_cast_fp16, var_27510_cast_fp16, var_27512_cast_fp16))[name = string("op_27600_cast_fp16")]; + bool var_27602_interleave_0 = const()[name = string("op_27602_interleave_0"), val = bool(false)]; + tensor var_27602_cast_fp16 = concat(axis = var_26137, interleave = var_27602_interleave_0, values = (var_27514_cast_fp16, var_27516_cast_fp16, var_27518_cast_fp16, var_27520_cast_fp16))[name = string("op_27602_cast_fp16")]; + bool var_27604_interleave_0 = const()[name = string("op_27604_interleave_0"), val = bool(false)]; + tensor var_27604_cast_fp16 = concat(axis = var_26137, interleave = var_27604_interleave_0, values = (var_27522_cast_fp16, var_27524_cast_fp16, var_27526_cast_fp16, var_27528_cast_fp16))[name = string("op_27604_cast_fp16")]; + bool var_27606_interleave_0 = const()[name = string("op_27606_interleave_0"), val = bool(false)]; + tensor var_27606_cast_fp16 = concat(axis = var_26137, interleave = var_27606_interleave_0, values = (var_27530_cast_fp16, var_27532_cast_fp16, var_27534_cast_fp16, var_27536_cast_fp16))[name = string("op_27606_cast_fp16")]; + bool var_27608_interleave_0 = const()[name = string("op_27608_interleave_0"), val = bool(false)]; + tensor var_27608_cast_fp16 = concat(axis = var_26137, interleave = var_27608_interleave_0, values = (var_27538_cast_fp16, var_27540_cast_fp16, var_27542_cast_fp16, var_27544_cast_fp16))[name = string("op_27608_cast_fp16")]; + bool var_27610_interleave_0 = const()[name = string("op_27610_interleave_0"), val = bool(false)]; + tensor var_27610_cast_fp16 = concat(axis = var_26137, interleave = var_27610_interleave_0, values = (var_27546_cast_fp16, var_27548_cast_fp16, var_27550_cast_fp16, var_27552_cast_fp16))[name = string("op_27610_cast_fp16")]; + bool var_27612_interleave_0 = const()[name = string("op_27612_interleave_0"), val = bool(false)]; + tensor var_27612_cast_fp16 = concat(axis = var_26137, interleave = var_27612_interleave_0, values = (var_27554_cast_fp16, var_27556_cast_fp16, var_27558_cast_fp16, var_27560_cast_fp16))[name = string("op_27612_cast_fp16")]; + bool var_27614_interleave_0 = const()[name = string("op_27614_interleave_0"), val = bool(false)]; + tensor var_27614_cast_fp16 = concat(axis = var_26137, interleave = var_27614_interleave_0, values = (var_27562_cast_fp16, var_27564_cast_fp16, var_27566_cast_fp16, var_27568_cast_fp16))[name = string("op_27614_cast_fp16")]; + bool var_27616_interleave_0 = const()[name = string("op_27616_interleave_0"), val = bool(false)]; + tensor var_27616_cast_fp16 = concat(axis = var_26137, interleave = var_27616_interleave_0, values = (var_27570_cast_fp16, var_27572_cast_fp16, var_27574_cast_fp16, var_27576_cast_fp16))[name = string("op_27616_cast_fp16")]; + bool input_137_interleave_0 = const()[name = string("input_137_interleave_0"), val = bool(false)]; + tensor input_137_cast_fp16 = concat(axis = var_26162, interleave = input_137_interleave_0, values = (var_27578_cast_fp16, var_27580_cast_fp16, var_27582_cast_fp16, var_27584_cast_fp16, var_27586_cast_fp16, var_27588_cast_fp16, var_27590_cast_fp16, var_27592_cast_fp16, var_27594_cast_fp16, var_27596_cast_fp16, var_27598_cast_fp16, var_27600_cast_fp16, var_27602_cast_fp16, var_27604_cast_fp16, var_27606_cast_fp16, var_27608_cast_fp16, var_27610_cast_fp16, var_27612_cast_fp16, var_27614_cast_fp16, var_27616_cast_fp16))[name = string("input_137_cast_fp16")]; + string obj_71_pad_type_0 = const()[name = string("obj_71_pad_type_0"), val = string("valid")]; + tensor obj_71_strides_0 = const()[name = string("obj_71_strides_0"), val = tensor([1, 1])]; + tensor obj_71_pad_0 = const()[name = string("obj_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_71_dilations_0 = const()[name = string("obj_71_dilations_0"), val = tensor([1, 1])]; + int32 obj_71_groups_0 = const()[name = string("obj_71_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693511040)))]; + tensor layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696787904)))]; + tensor obj_71_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_71_dilations_0, groups = obj_71_groups_0, pad = obj_71_pad_0, pad_type = obj_71_pad_type_0, strides = obj_71_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("obj_71_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_71_cast_fp16)[name = string("inputs_71_cast_fp16")]; + tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; + fp16 var_27635_to_fp16 = const()[name = string("op_27635_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_27635_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; + tensor input_139_gamma_0_to_fp16 = const()[name = string("input_139_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696790528)))]; + tensor input_139_beta_0_to_fp16 = const()[name = string("input_139_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696793152)))]; + fp16 input_139_epsilon_0_to_fp16 = const()[name = string("input_139_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_139_cast_fp16 = batch_norm(beta = input_139_beta_0_to_fp16, epsilon = input_139_epsilon_0_to_fp16, gamma = input_139_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_139_cast_fp16")]; + string input_141_pad_type_0 = const()[name = string("input_141_pad_type_0"), val = string("valid")]; + tensor input_141_strides_0 = const()[name = string("input_141_strides_0"), val = tensor([1, 1])]; + tensor input_141_pad_0 = const()[name = string("input_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_141_dilations_0 = const()[name = string("input_141_dilations_0"), val = tensor([1, 1])]; + int32 input_141_groups_0 = const()[name = string("input_141_groups_0"), val = int32(1)]; + tensor layers_17_fc1_weight_to_fp16 = const()[name = string("layers_17_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696795776)))]; + tensor layers_17_fc1_bias_to_fp16 = const()[name = string("layers_17_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709903040)))]; + tensor input_141_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_141_dilations_0, groups = input_141_groups_0, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = input_141_strides_0, weight = layers_17_fc1_weight_to_fp16, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")]; + string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("EXACT")]; + tensor input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")]; + string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")]; + tensor hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)]; + tensor layers_17_fc2_weight_to_fp16 = const()[name = string("layers_17_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709913344)))]; + tensor layers_17_fc2_bias_to_fp16 = const()[name = string("layers_17_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723020608)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_17_fc2_weight_to_fp16, x = input_143_cast_fp16)[name = string("hidden_states_39_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("inputs_73_cast_fp16")]; + int32 var_27664 = const()[name = string("op_27664"), val = int32(3)]; + int32 var_27689 = const()[name = string("op_27689"), val = int32(1)]; + tensor out_73_axes_0 = const()[name = string("out_73_axes_0"), val = tensor([1])]; + fp16 var_27706_to_fp16 = const()[name = string("op_27706_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_27706_to_fp16, x = inputs_73_cast_fp16)[name = string("out_73_cast_fp16")]; + tensor obj_73_gamma_0_to_fp16 = const()[name = string("obj_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723023232)))]; + tensor obj_73_beta_0_to_fp16 = const()[name = string("obj_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723025856)))]; + fp16 obj_73_epsilon_0_to_fp16 = const()[name = string("obj_73_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_73_cast_fp16 = batch_norm(beta = obj_73_beta_0_to_fp16, epsilon = obj_73_epsilon_0_to_fp16, gamma = obj_73_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = string("obj_73_cast_fp16")]; + string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; + tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; + tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; + int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723028480)))]; + tensor layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726305344)))]; + tensor query_37_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("query_37_cast_fp16")]; + string key_37_pad_type_0 = const()[name = string("key_37_pad_type_0"), val = string("valid")]; + tensor key_37_strides_0 = const()[name = string("key_37_strides_0"), val = tensor([1, 1])]; + tensor key_37_pad_0 = const()[name = string("key_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_37_dilations_0 = const()[name = string("key_37_dilations_0"), val = tensor([1, 1])]; + int32 key_37_groups_0 = const()[name = string("key_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726307968)))]; + tensor key_37_cast_fp16 = conv(dilations = key_37_dilations_0, groups = key_37_groups_0, pad = key_37_pad_0, pad_type = key_37_pad_type_0, strides = key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("key_37_cast_fp16")]; + string value_37_pad_type_0 = const()[name = string("value_37_pad_type_0"), val = string("valid")]; + tensor value_37_strides_0 = const()[name = string("value_37_strides_0"), val = tensor([1, 1])]; + tensor value_37_pad_0 = const()[name = string("value_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_37_dilations_0 = const()[name = string("value_37_dilations_0"), val = tensor([1, 1])]; + int32 value_37_groups_0 = const()[name = string("value_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(729584832)))]; + tensor layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732861696)))]; + tensor value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = value_37_dilations_0, groups = value_37_groups_0, pad = value_37_pad_0, pad_type = value_37_pad_type_0, strides = value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("value_37_cast_fp16")]; + tensor var_27744_begin_0 = const()[name = string("op_27744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27744_end_0 = const()[name = string("op_27744_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27744_end_mask_0 = const()[name = string("op_27744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27744_cast_fp16 = slice_by_index(begin = var_27744_begin_0, end = var_27744_end_0, end_mask = var_27744_end_mask_0, x = query_37_cast_fp16)[name = string("op_27744_cast_fp16")]; + tensor var_27748_begin_0 = const()[name = string("op_27748_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27748_end_0 = const()[name = string("op_27748_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_27748_end_mask_0 = const()[name = string("op_27748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27748_cast_fp16 = slice_by_index(begin = var_27748_begin_0, end = var_27748_end_0, end_mask = var_27748_end_mask_0, x = query_37_cast_fp16)[name = string("op_27748_cast_fp16")]; + tensor var_27752_begin_0 = const()[name = string("op_27752_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27752_end_0 = const()[name = string("op_27752_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_27752_end_mask_0 = const()[name = string("op_27752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27752_cast_fp16 = slice_by_index(begin = var_27752_begin_0, end = var_27752_end_0, end_mask = var_27752_end_mask_0, x = query_37_cast_fp16)[name = string("op_27752_cast_fp16")]; + tensor var_27756_begin_0 = const()[name = string("op_27756_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27756_end_0 = const()[name = string("op_27756_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_27756_end_mask_0 = const()[name = string("op_27756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27756_cast_fp16 = slice_by_index(begin = var_27756_begin_0, end = var_27756_end_0, end_mask = var_27756_end_mask_0, x = query_37_cast_fp16)[name = string("op_27756_cast_fp16")]; + tensor var_27760_begin_0 = const()[name = string("op_27760_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27760_end_0 = const()[name = string("op_27760_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_27760_end_mask_0 = const()[name = string("op_27760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27760_cast_fp16 = slice_by_index(begin = var_27760_begin_0, end = var_27760_end_0, end_mask = var_27760_end_mask_0, x = query_37_cast_fp16)[name = string("op_27760_cast_fp16")]; + tensor var_27764_begin_0 = const()[name = string("op_27764_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27764_end_0 = const()[name = string("op_27764_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_27764_end_mask_0 = const()[name = string("op_27764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27764_cast_fp16 = slice_by_index(begin = var_27764_begin_0, end = var_27764_end_0, end_mask = var_27764_end_mask_0, x = query_37_cast_fp16)[name = string("op_27764_cast_fp16")]; + tensor var_27768_begin_0 = const()[name = string("op_27768_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27768_end_0 = const()[name = string("op_27768_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_27768_end_mask_0 = const()[name = string("op_27768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27768_cast_fp16 = slice_by_index(begin = var_27768_begin_0, end = var_27768_end_0, end_mask = var_27768_end_mask_0, x = query_37_cast_fp16)[name = string("op_27768_cast_fp16")]; + tensor var_27772_begin_0 = const()[name = string("op_27772_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27772_end_0 = const()[name = string("op_27772_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_27772_end_mask_0 = const()[name = string("op_27772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27772_cast_fp16 = slice_by_index(begin = var_27772_begin_0, end = var_27772_end_0, end_mask = var_27772_end_mask_0, x = query_37_cast_fp16)[name = string("op_27772_cast_fp16")]; + tensor var_27776_begin_0 = const()[name = string("op_27776_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27776_end_0 = const()[name = string("op_27776_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_27776_end_mask_0 = const()[name = string("op_27776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27776_cast_fp16 = slice_by_index(begin = var_27776_begin_0, end = var_27776_end_0, end_mask = var_27776_end_mask_0, x = query_37_cast_fp16)[name = string("op_27776_cast_fp16")]; + tensor var_27780_begin_0 = const()[name = string("op_27780_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27780_end_0 = const()[name = string("op_27780_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_27780_end_mask_0 = const()[name = string("op_27780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27780_cast_fp16 = slice_by_index(begin = var_27780_begin_0, end = var_27780_end_0, end_mask = var_27780_end_mask_0, x = query_37_cast_fp16)[name = string("op_27780_cast_fp16")]; + tensor var_27784_begin_0 = const()[name = string("op_27784_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27784_end_0 = const()[name = string("op_27784_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_27784_end_mask_0 = const()[name = string("op_27784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27784_cast_fp16 = slice_by_index(begin = var_27784_begin_0, end = var_27784_end_0, end_mask = var_27784_end_mask_0, x = query_37_cast_fp16)[name = string("op_27784_cast_fp16")]; + tensor var_27788_begin_0 = const()[name = string("op_27788_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27788_end_0 = const()[name = string("op_27788_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_27788_end_mask_0 = const()[name = string("op_27788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27788_cast_fp16 = slice_by_index(begin = var_27788_begin_0, end = var_27788_end_0, end_mask = var_27788_end_mask_0, x = query_37_cast_fp16)[name = string("op_27788_cast_fp16")]; + tensor var_27792_begin_0 = const()[name = string("op_27792_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27792_end_0 = const()[name = string("op_27792_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_27792_end_mask_0 = const()[name = string("op_27792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27792_cast_fp16 = slice_by_index(begin = var_27792_begin_0, end = var_27792_end_0, end_mask = var_27792_end_mask_0, x = query_37_cast_fp16)[name = string("op_27792_cast_fp16")]; + tensor var_27796_begin_0 = const()[name = string("op_27796_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27796_end_0 = const()[name = string("op_27796_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_27796_end_mask_0 = const()[name = string("op_27796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27796_cast_fp16 = slice_by_index(begin = var_27796_begin_0, end = var_27796_end_0, end_mask = var_27796_end_mask_0, x = query_37_cast_fp16)[name = string("op_27796_cast_fp16")]; + tensor var_27800_begin_0 = const()[name = string("op_27800_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27800_end_0 = const()[name = string("op_27800_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_27800_end_mask_0 = const()[name = string("op_27800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27800_cast_fp16 = slice_by_index(begin = var_27800_begin_0, end = var_27800_end_0, end_mask = var_27800_end_mask_0, x = query_37_cast_fp16)[name = string("op_27800_cast_fp16")]; + tensor var_27804_begin_0 = const()[name = string("op_27804_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27804_end_0 = const()[name = string("op_27804_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_27804_end_mask_0 = const()[name = string("op_27804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27804_cast_fp16 = slice_by_index(begin = var_27804_begin_0, end = var_27804_end_0, end_mask = var_27804_end_mask_0, x = query_37_cast_fp16)[name = string("op_27804_cast_fp16")]; + tensor var_27808_begin_0 = const()[name = string("op_27808_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27808_end_0 = const()[name = string("op_27808_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_27808_end_mask_0 = const()[name = string("op_27808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27808_cast_fp16 = slice_by_index(begin = var_27808_begin_0, end = var_27808_end_0, end_mask = var_27808_end_mask_0, x = query_37_cast_fp16)[name = string("op_27808_cast_fp16")]; + tensor var_27812_begin_0 = const()[name = string("op_27812_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27812_end_0 = const()[name = string("op_27812_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_27812_end_mask_0 = const()[name = string("op_27812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27812_cast_fp16 = slice_by_index(begin = var_27812_begin_0, end = var_27812_end_0, end_mask = var_27812_end_mask_0, x = query_37_cast_fp16)[name = string("op_27812_cast_fp16")]; + tensor var_27816_begin_0 = const()[name = string("op_27816_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27816_end_0 = const()[name = string("op_27816_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_27816_end_mask_0 = const()[name = string("op_27816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27816_cast_fp16 = slice_by_index(begin = var_27816_begin_0, end = var_27816_end_0, end_mask = var_27816_end_mask_0, x = query_37_cast_fp16)[name = string("op_27816_cast_fp16")]; + tensor var_27820_begin_0 = const()[name = string("op_27820_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27820_end_0 = const()[name = string("op_27820_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_27820_end_mask_0 = const()[name = string("op_27820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27820_cast_fp16 = slice_by_index(begin = var_27820_begin_0, end = var_27820_end_0, end_mask = var_27820_end_mask_0, x = query_37_cast_fp16)[name = string("op_27820_cast_fp16")]; + tensor var_27829_begin_0 = const()[name = string("op_27829_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27829_end_0 = const()[name = string("op_27829_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27829_end_mask_0 = const()[name = string("op_27829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27829_cast_fp16 = slice_by_index(begin = var_27829_begin_0, end = var_27829_end_0, end_mask = var_27829_end_mask_0, x = var_27744_cast_fp16)[name = string("op_27829_cast_fp16")]; + tensor var_27836_begin_0 = const()[name = string("op_27836_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27836_end_0 = const()[name = string("op_27836_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27836_end_mask_0 = const()[name = string("op_27836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27836_cast_fp16 = slice_by_index(begin = var_27836_begin_0, end = var_27836_end_0, end_mask = var_27836_end_mask_0, x = var_27744_cast_fp16)[name = string("op_27836_cast_fp16")]; + tensor var_27843_begin_0 = const()[name = string("op_27843_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27843_end_0 = const()[name = string("op_27843_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27843_end_mask_0 = const()[name = string("op_27843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27843_cast_fp16 = slice_by_index(begin = var_27843_begin_0, end = var_27843_end_0, end_mask = var_27843_end_mask_0, x = var_27744_cast_fp16)[name = string("op_27843_cast_fp16")]; + tensor var_27850_begin_0 = const()[name = string("op_27850_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27850_end_0 = const()[name = string("op_27850_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27850_end_mask_0 = const()[name = string("op_27850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27850_cast_fp16 = slice_by_index(begin = var_27850_begin_0, end = var_27850_end_0, end_mask = var_27850_end_mask_0, x = var_27744_cast_fp16)[name = string("op_27850_cast_fp16")]; + tensor var_27857_begin_0 = const()[name = string("op_27857_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27857_end_0 = const()[name = string("op_27857_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27857_end_mask_0 = const()[name = string("op_27857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27857_cast_fp16 = slice_by_index(begin = var_27857_begin_0, end = var_27857_end_0, end_mask = var_27857_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27857_cast_fp16")]; + tensor var_27864_begin_0 = const()[name = string("op_27864_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27864_end_0 = const()[name = string("op_27864_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27864_end_mask_0 = const()[name = string("op_27864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27864_cast_fp16 = slice_by_index(begin = var_27864_begin_0, end = var_27864_end_0, end_mask = var_27864_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27864_cast_fp16")]; + tensor var_27871_begin_0 = const()[name = string("op_27871_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27871_end_0 = const()[name = string("op_27871_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27871_end_mask_0 = const()[name = string("op_27871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27871_cast_fp16 = slice_by_index(begin = var_27871_begin_0, end = var_27871_end_0, end_mask = var_27871_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27871_cast_fp16")]; + tensor var_27878_begin_0 = const()[name = string("op_27878_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27878_end_0 = const()[name = string("op_27878_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27878_end_mask_0 = const()[name = string("op_27878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27878_cast_fp16 = slice_by_index(begin = var_27878_begin_0, end = var_27878_end_0, end_mask = var_27878_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27878_cast_fp16")]; + tensor var_27885_begin_0 = const()[name = string("op_27885_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27885_end_0 = const()[name = string("op_27885_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27885_end_mask_0 = const()[name = string("op_27885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27885_cast_fp16 = slice_by_index(begin = var_27885_begin_0, end = var_27885_end_0, end_mask = var_27885_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27885_cast_fp16")]; + tensor var_27892_begin_0 = const()[name = string("op_27892_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27892_end_0 = const()[name = string("op_27892_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27892_end_mask_0 = const()[name = string("op_27892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27892_cast_fp16 = slice_by_index(begin = var_27892_begin_0, end = var_27892_end_0, end_mask = var_27892_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27892_cast_fp16")]; + tensor var_27899_begin_0 = const()[name = string("op_27899_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27899_end_0 = const()[name = string("op_27899_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27899_end_mask_0 = const()[name = string("op_27899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27899_cast_fp16 = slice_by_index(begin = var_27899_begin_0, end = var_27899_end_0, end_mask = var_27899_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27899_cast_fp16")]; + tensor var_27906_begin_0 = const()[name = string("op_27906_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27906_end_0 = const()[name = string("op_27906_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27906_end_mask_0 = const()[name = string("op_27906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27906_cast_fp16 = slice_by_index(begin = var_27906_begin_0, end = var_27906_end_0, end_mask = var_27906_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27906_cast_fp16")]; + tensor var_27913_begin_0 = const()[name = string("op_27913_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27913_end_0 = const()[name = string("op_27913_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27913_end_mask_0 = const()[name = string("op_27913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27913_cast_fp16 = slice_by_index(begin = var_27913_begin_0, end = var_27913_end_0, end_mask = var_27913_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27913_cast_fp16")]; + tensor var_27920_begin_0 = const()[name = string("op_27920_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27920_end_0 = const()[name = string("op_27920_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27920_end_mask_0 = const()[name = string("op_27920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27920_cast_fp16 = slice_by_index(begin = var_27920_begin_0, end = var_27920_end_0, end_mask = var_27920_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27920_cast_fp16")]; + tensor var_27927_begin_0 = const()[name = string("op_27927_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27927_end_0 = const()[name = string("op_27927_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27927_end_mask_0 = const()[name = string("op_27927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27927_cast_fp16 = slice_by_index(begin = var_27927_begin_0, end = var_27927_end_0, end_mask = var_27927_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27927_cast_fp16")]; + tensor var_27934_begin_0 = const()[name = string("op_27934_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27934_end_0 = const()[name = string("op_27934_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27934_end_mask_0 = const()[name = string("op_27934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27934_cast_fp16 = slice_by_index(begin = var_27934_begin_0, end = var_27934_end_0, end_mask = var_27934_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27934_cast_fp16")]; + tensor var_27941_begin_0 = const()[name = string("op_27941_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27941_end_0 = const()[name = string("op_27941_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27941_end_mask_0 = const()[name = string("op_27941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27941_cast_fp16 = slice_by_index(begin = var_27941_begin_0, end = var_27941_end_0, end_mask = var_27941_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27941_cast_fp16")]; + tensor var_27948_begin_0 = const()[name = string("op_27948_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27948_end_0 = const()[name = string("op_27948_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27948_end_mask_0 = const()[name = string("op_27948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27948_cast_fp16 = slice_by_index(begin = var_27948_begin_0, end = var_27948_end_0, end_mask = var_27948_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27948_cast_fp16")]; + tensor var_27955_begin_0 = const()[name = string("op_27955_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27955_end_0 = const()[name = string("op_27955_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27955_end_mask_0 = const()[name = string("op_27955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27955_cast_fp16 = slice_by_index(begin = var_27955_begin_0, end = var_27955_end_0, end_mask = var_27955_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27955_cast_fp16")]; + tensor var_27962_begin_0 = const()[name = string("op_27962_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27962_end_0 = const()[name = string("op_27962_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27962_end_mask_0 = const()[name = string("op_27962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27962_cast_fp16 = slice_by_index(begin = var_27962_begin_0, end = var_27962_end_0, end_mask = var_27962_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27962_cast_fp16")]; + tensor var_27969_begin_0 = const()[name = string("op_27969_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27969_end_0 = const()[name = string("op_27969_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27969_end_mask_0 = const()[name = string("op_27969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27969_cast_fp16 = slice_by_index(begin = var_27969_begin_0, end = var_27969_end_0, end_mask = var_27969_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27969_cast_fp16")]; + tensor var_27976_begin_0 = const()[name = string("op_27976_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27976_end_0 = const()[name = string("op_27976_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27976_end_mask_0 = const()[name = string("op_27976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27976_cast_fp16 = slice_by_index(begin = var_27976_begin_0, end = var_27976_end_0, end_mask = var_27976_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27976_cast_fp16")]; + tensor var_27983_begin_0 = const()[name = string("op_27983_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27983_end_0 = const()[name = string("op_27983_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27983_end_mask_0 = const()[name = string("op_27983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27983_cast_fp16 = slice_by_index(begin = var_27983_begin_0, end = var_27983_end_0, end_mask = var_27983_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27983_cast_fp16")]; + tensor var_27990_begin_0 = const()[name = string("op_27990_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27990_end_0 = const()[name = string("op_27990_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27990_end_mask_0 = const()[name = string("op_27990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27990_cast_fp16 = slice_by_index(begin = var_27990_begin_0, end = var_27990_end_0, end_mask = var_27990_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27990_cast_fp16")]; + tensor var_27997_begin_0 = const()[name = string("op_27997_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27997_end_0 = const()[name = string("op_27997_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27997_end_mask_0 = const()[name = string("op_27997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27997_cast_fp16 = slice_by_index(begin = var_27997_begin_0, end = var_27997_end_0, end_mask = var_27997_end_mask_0, x = var_27768_cast_fp16)[name = string("op_27997_cast_fp16")]; + tensor var_28004_begin_0 = const()[name = string("op_28004_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28004_end_0 = const()[name = string("op_28004_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28004_end_mask_0 = const()[name = string("op_28004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28004_cast_fp16 = slice_by_index(begin = var_28004_begin_0, end = var_28004_end_0, end_mask = var_28004_end_mask_0, x = var_27768_cast_fp16)[name = string("op_28004_cast_fp16")]; + tensor var_28011_begin_0 = const()[name = string("op_28011_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28011_end_0 = const()[name = string("op_28011_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28011_end_mask_0 = const()[name = string("op_28011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28011_cast_fp16 = slice_by_index(begin = var_28011_begin_0, end = var_28011_end_0, end_mask = var_28011_end_mask_0, x = var_27768_cast_fp16)[name = string("op_28011_cast_fp16")]; + tensor var_28018_begin_0 = const()[name = string("op_28018_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28018_end_0 = const()[name = string("op_28018_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28018_end_mask_0 = const()[name = string("op_28018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28018_cast_fp16 = slice_by_index(begin = var_28018_begin_0, end = var_28018_end_0, end_mask = var_28018_end_mask_0, x = var_27768_cast_fp16)[name = string("op_28018_cast_fp16")]; + tensor var_28025_begin_0 = const()[name = string("op_28025_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28025_end_0 = const()[name = string("op_28025_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28025_end_mask_0 = const()[name = string("op_28025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28025_cast_fp16 = slice_by_index(begin = var_28025_begin_0, end = var_28025_end_0, end_mask = var_28025_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28025_cast_fp16")]; + tensor var_28032_begin_0 = const()[name = string("op_28032_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28032_end_0 = const()[name = string("op_28032_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28032_end_mask_0 = const()[name = string("op_28032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28032_cast_fp16 = slice_by_index(begin = var_28032_begin_0, end = var_28032_end_0, end_mask = var_28032_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28032_cast_fp16")]; + tensor var_28039_begin_0 = const()[name = string("op_28039_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28039_end_0 = const()[name = string("op_28039_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28039_end_mask_0 = const()[name = string("op_28039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28039_cast_fp16 = slice_by_index(begin = var_28039_begin_0, end = var_28039_end_0, end_mask = var_28039_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28039_cast_fp16")]; + tensor var_28046_begin_0 = const()[name = string("op_28046_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28046_end_0 = const()[name = string("op_28046_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28046_end_mask_0 = const()[name = string("op_28046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28046_cast_fp16 = slice_by_index(begin = var_28046_begin_0, end = var_28046_end_0, end_mask = var_28046_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28046_cast_fp16")]; + tensor var_28053_begin_0 = const()[name = string("op_28053_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28053_end_0 = const()[name = string("op_28053_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28053_end_mask_0 = const()[name = string("op_28053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28053_cast_fp16 = slice_by_index(begin = var_28053_begin_0, end = var_28053_end_0, end_mask = var_28053_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28053_cast_fp16")]; + tensor var_28060_begin_0 = const()[name = string("op_28060_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28060_end_0 = const()[name = string("op_28060_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28060_end_mask_0 = const()[name = string("op_28060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28060_cast_fp16 = slice_by_index(begin = var_28060_begin_0, end = var_28060_end_0, end_mask = var_28060_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28060_cast_fp16")]; + tensor var_28067_begin_0 = const()[name = string("op_28067_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28067_end_0 = const()[name = string("op_28067_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28067_end_mask_0 = const()[name = string("op_28067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28067_cast_fp16 = slice_by_index(begin = var_28067_begin_0, end = var_28067_end_0, end_mask = var_28067_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28067_cast_fp16")]; + tensor var_28074_begin_0 = const()[name = string("op_28074_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28074_end_0 = const()[name = string("op_28074_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28074_end_mask_0 = const()[name = string("op_28074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28074_cast_fp16 = slice_by_index(begin = var_28074_begin_0, end = var_28074_end_0, end_mask = var_28074_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28074_cast_fp16")]; + tensor var_28081_begin_0 = const()[name = string("op_28081_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28081_end_0 = const()[name = string("op_28081_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28081_end_mask_0 = const()[name = string("op_28081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28081_cast_fp16 = slice_by_index(begin = var_28081_begin_0, end = var_28081_end_0, end_mask = var_28081_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28081_cast_fp16")]; + tensor var_28088_begin_0 = const()[name = string("op_28088_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28088_end_0 = const()[name = string("op_28088_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28088_end_mask_0 = const()[name = string("op_28088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28088_cast_fp16 = slice_by_index(begin = var_28088_begin_0, end = var_28088_end_0, end_mask = var_28088_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28088_cast_fp16")]; + tensor var_28095_begin_0 = const()[name = string("op_28095_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28095_end_0 = const()[name = string("op_28095_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28095_end_mask_0 = const()[name = string("op_28095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28095_cast_fp16 = slice_by_index(begin = var_28095_begin_0, end = var_28095_end_0, end_mask = var_28095_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28095_cast_fp16")]; + tensor var_28102_begin_0 = const()[name = string("op_28102_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28102_end_0 = const()[name = string("op_28102_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28102_end_mask_0 = const()[name = string("op_28102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28102_cast_fp16 = slice_by_index(begin = var_28102_begin_0, end = var_28102_end_0, end_mask = var_28102_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28102_cast_fp16")]; + tensor var_28109_begin_0 = const()[name = string("op_28109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28109_end_0 = const()[name = string("op_28109_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28109_end_mask_0 = const()[name = string("op_28109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28109_cast_fp16 = slice_by_index(begin = var_28109_begin_0, end = var_28109_end_0, end_mask = var_28109_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28109_cast_fp16")]; + tensor var_28116_begin_0 = const()[name = string("op_28116_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28116_end_0 = const()[name = string("op_28116_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28116_end_mask_0 = const()[name = string("op_28116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28116_cast_fp16 = slice_by_index(begin = var_28116_begin_0, end = var_28116_end_0, end_mask = var_28116_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28116_cast_fp16")]; + tensor var_28123_begin_0 = const()[name = string("op_28123_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28123_end_0 = const()[name = string("op_28123_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28123_end_mask_0 = const()[name = string("op_28123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28123_cast_fp16 = slice_by_index(begin = var_28123_begin_0, end = var_28123_end_0, end_mask = var_28123_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28123_cast_fp16")]; + tensor var_28130_begin_0 = const()[name = string("op_28130_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28130_end_0 = const()[name = string("op_28130_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28130_end_mask_0 = const()[name = string("op_28130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28130_cast_fp16 = slice_by_index(begin = var_28130_begin_0, end = var_28130_end_0, end_mask = var_28130_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28130_cast_fp16")]; + tensor var_28137_begin_0 = const()[name = string("op_28137_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28137_end_0 = const()[name = string("op_28137_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28137_end_mask_0 = const()[name = string("op_28137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28137_cast_fp16 = slice_by_index(begin = var_28137_begin_0, end = var_28137_end_0, end_mask = var_28137_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28137_cast_fp16")]; + tensor var_28144_begin_0 = const()[name = string("op_28144_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28144_end_0 = const()[name = string("op_28144_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28144_end_mask_0 = const()[name = string("op_28144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28144_cast_fp16 = slice_by_index(begin = var_28144_begin_0, end = var_28144_end_0, end_mask = var_28144_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28144_cast_fp16")]; + tensor var_28151_begin_0 = const()[name = string("op_28151_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28151_end_0 = const()[name = string("op_28151_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28151_end_mask_0 = const()[name = string("op_28151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28151_cast_fp16 = slice_by_index(begin = var_28151_begin_0, end = var_28151_end_0, end_mask = var_28151_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28151_cast_fp16")]; + tensor var_28158_begin_0 = const()[name = string("op_28158_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28158_end_0 = const()[name = string("op_28158_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28158_end_mask_0 = const()[name = string("op_28158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28158_cast_fp16 = slice_by_index(begin = var_28158_begin_0, end = var_28158_end_0, end_mask = var_28158_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28158_cast_fp16")]; + tensor var_28165_begin_0 = const()[name = string("op_28165_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28165_end_0 = const()[name = string("op_28165_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28165_end_mask_0 = const()[name = string("op_28165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28165_cast_fp16 = slice_by_index(begin = var_28165_begin_0, end = var_28165_end_0, end_mask = var_28165_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28165_cast_fp16")]; + tensor var_28172_begin_0 = const()[name = string("op_28172_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28172_end_0 = const()[name = string("op_28172_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28172_end_mask_0 = const()[name = string("op_28172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28172_cast_fp16 = slice_by_index(begin = var_28172_begin_0, end = var_28172_end_0, end_mask = var_28172_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28172_cast_fp16")]; + tensor var_28179_begin_0 = const()[name = string("op_28179_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28179_end_0 = const()[name = string("op_28179_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28179_end_mask_0 = const()[name = string("op_28179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28179_cast_fp16 = slice_by_index(begin = var_28179_begin_0, end = var_28179_end_0, end_mask = var_28179_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28179_cast_fp16")]; + tensor var_28186_begin_0 = const()[name = string("op_28186_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28186_end_0 = const()[name = string("op_28186_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28186_end_mask_0 = const()[name = string("op_28186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28186_cast_fp16 = slice_by_index(begin = var_28186_begin_0, end = var_28186_end_0, end_mask = var_28186_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28186_cast_fp16")]; + tensor var_28193_begin_0 = const()[name = string("op_28193_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28193_end_0 = const()[name = string("op_28193_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28193_end_mask_0 = const()[name = string("op_28193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28193_cast_fp16 = slice_by_index(begin = var_28193_begin_0, end = var_28193_end_0, end_mask = var_28193_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28193_cast_fp16")]; + tensor var_28200_begin_0 = const()[name = string("op_28200_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28200_end_0 = const()[name = string("op_28200_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28200_end_mask_0 = const()[name = string("op_28200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28200_cast_fp16 = slice_by_index(begin = var_28200_begin_0, end = var_28200_end_0, end_mask = var_28200_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28200_cast_fp16")]; + tensor var_28207_begin_0 = const()[name = string("op_28207_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28207_end_0 = const()[name = string("op_28207_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28207_end_mask_0 = const()[name = string("op_28207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28207_cast_fp16 = slice_by_index(begin = var_28207_begin_0, end = var_28207_end_0, end_mask = var_28207_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28207_cast_fp16")]; + tensor var_28214_begin_0 = const()[name = string("op_28214_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28214_end_0 = const()[name = string("op_28214_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28214_end_mask_0 = const()[name = string("op_28214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28214_cast_fp16 = slice_by_index(begin = var_28214_begin_0, end = var_28214_end_0, end_mask = var_28214_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28214_cast_fp16")]; + tensor var_28221_begin_0 = const()[name = string("op_28221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28221_end_0 = const()[name = string("op_28221_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28221_end_mask_0 = const()[name = string("op_28221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28221_cast_fp16 = slice_by_index(begin = var_28221_begin_0, end = var_28221_end_0, end_mask = var_28221_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28221_cast_fp16")]; + tensor var_28228_begin_0 = const()[name = string("op_28228_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28228_end_0 = const()[name = string("op_28228_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28228_end_mask_0 = const()[name = string("op_28228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28228_cast_fp16 = slice_by_index(begin = var_28228_begin_0, end = var_28228_end_0, end_mask = var_28228_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28228_cast_fp16")]; + tensor var_28235_begin_0 = const()[name = string("op_28235_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28235_end_0 = const()[name = string("op_28235_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28235_end_mask_0 = const()[name = string("op_28235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28235_cast_fp16 = slice_by_index(begin = var_28235_begin_0, end = var_28235_end_0, end_mask = var_28235_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28235_cast_fp16")]; + tensor var_28242_begin_0 = const()[name = string("op_28242_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28242_end_0 = const()[name = string("op_28242_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28242_end_mask_0 = const()[name = string("op_28242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28242_cast_fp16 = slice_by_index(begin = var_28242_begin_0, end = var_28242_end_0, end_mask = var_28242_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28242_cast_fp16")]; + tensor var_28249_begin_0 = const()[name = string("op_28249_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28249_end_0 = const()[name = string("op_28249_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28249_end_mask_0 = const()[name = string("op_28249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28249_cast_fp16 = slice_by_index(begin = var_28249_begin_0, end = var_28249_end_0, end_mask = var_28249_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28249_cast_fp16")]; + tensor var_28256_begin_0 = const()[name = string("op_28256_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28256_end_0 = const()[name = string("op_28256_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28256_end_mask_0 = const()[name = string("op_28256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28256_cast_fp16 = slice_by_index(begin = var_28256_begin_0, end = var_28256_end_0, end_mask = var_28256_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28256_cast_fp16")]; + tensor var_28263_begin_0 = const()[name = string("op_28263_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28263_end_0 = const()[name = string("op_28263_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28263_end_mask_0 = const()[name = string("op_28263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28263_cast_fp16 = slice_by_index(begin = var_28263_begin_0, end = var_28263_end_0, end_mask = var_28263_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28263_cast_fp16")]; + tensor var_28270_begin_0 = const()[name = string("op_28270_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28270_end_0 = const()[name = string("op_28270_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28270_end_mask_0 = const()[name = string("op_28270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28270_cast_fp16 = slice_by_index(begin = var_28270_begin_0, end = var_28270_end_0, end_mask = var_28270_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28270_cast_fp16")]; + tensor var_28277_begin_0 = const()[name = string("op_28277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28277_end_0 = const()[name = string("op_28277_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28277_end_mask_0 = const()[name = string("op_28277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28277_cast_fp16 = slice_by_index(begin = var_28277_begin_0, end = var_28277_end_0, end_mask = var_28277_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28277_cast_fp16")]; + tensor var_28284_begin_0 = const()[name = string("op_28284_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28284_end_0 = const()[name = string("op_28284_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28284_end_mask_0 = const()[name = string("op_28284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28284_cast_fp16 = slice_by_index(begin = var_28284_begin_0, end = var_28284_end_0, end_mask = var_28284_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28284_cast_fp16")]; + tensor var_28291_begin_0 = const()[name = string("op_28291_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28291_end_0 = const()[name = string("op_28291_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28291_end_mask_0 = const()[name = string("op_28291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28291_cast_fp16 = slice_by_index(begin = var_28291_begin_0, end = var_28291_end_0, end_mask = var_28291_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28291_cast_fp16")]; + tensor var_28298_begin_0 = const()[name = string("op_28298_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28298_end_0 = const()[name = string("op_28298_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28298_end_mask_0 = const()[name = string("op_28298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28298_cast_fp16 = slice_by_index(begin = var_28298_begin_0, end = var_28298_end_0, end_mask = var_28298_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28298_cast_fp16")]; + tensor var_28305_begin_0 = const()[name = string("op_28305_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28305_end_0 = const()[name = string("op_28305_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28305_end_mask_0 = const()[name = string("op_28305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28305_cast_fp16 = slice_by_index(begin = var_28305_begin_0, end = var_28305_end_0, end_mask = var_28305_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28305_cast_fp16")]; + tensor var_28312_begin_0 = const()[name = string("op_28312_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28312_end_0 = const()[name = string("op_28312_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28312_end_mask_0 = const()[name = string("op_28312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28312_cast_fp16 = slice_by_index(begin = var_28312_begin_0, end = var_28312_end_0, end_mask = var_28312_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28312_cast_fp16")]; + tensor var_28319_begin_0 = const()[name = string("op_28319_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28319_end_0 = const()[name = string("op_28319_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28319_end_mask_0 = const()[name = string("op_28319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28319_cast_fp16 = slice_by_index(begin = var_28319_begin_0, end = var_28319_end_0, end_mask = var_28319_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28319_cast_fp16")]; + tensor var_28326_begin_0 = const()[name = string("op_28326_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28326_end_0 = const()[name = string("op_28326_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28326_end_mask_0 = const()[name = string("op_28326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28326_cast_fp16 = slice_by_index(begin = var_28326_begin_0, end = var_28326_end_0, end_mask = var_28326_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28326_cast_fp16")]; + tensor var_28333_begin_0 = const()[name = string("op_28333_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28333_end_0 = const()[name = string("op_28333_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28333_end_mask_0 = const()[name = string("op_28333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28333_cast_fp16 = slice_by_index(begin = var_28333_begin_0, end = var_28333_end_0, end_mask = var_28333_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28333_cast_fp16")]; + tensor var_28340_begin_0 = const()[name = string("op_28340_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28340_end_0 = const()[name = string("op_28340_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28340_end_mask_0 = const()[name = string("op_28340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28340_cast_fp16 = slice_by_index(begin = var_28340_begin_0, end = var_28340_end_0, end_mask = var_28340_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28340_cast_fp16")]; + tensor var_28347_begin_0 = const()[name = string("op_28347_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28347_end_0 = const()[name = string("op_28347_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28347_end_mask_0 = const()[name = string("op_28347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28347_cast_fp16 = slice_by_index(begin = var_28347_begin_0, end = var_28347_end_0, end_mask = var_28347_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28347_cast_fp16")]; + tensor var_28354_begin_0 = const()[name = string("op_28354_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28354_end_0 = const()[name = string("op_28354_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28354_end_mask_0 = const()[name = string("op_28354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28354_cast_fp16 = slice_by_index(begin = var_28354_begin_0, end = var_28354_end_0, end_mask = var_28354_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28354_cast_fp16")]; + tensor var_28361_begin_0 = const()[name = string("op_28361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28361_end_0 = const()[name = string("op_28361_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28361_end_mask_0 = const()[name = string("op_28361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28361_cast_fp16 = slice_by_index(begin = var_28361_begin_0, end = var_28361_end_0, end_mask = var_28361_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28361_cast_fp16")]; + tensor var_28368_begin_0 = const()[name = string("op_28368_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28368_end_0 = const()[name = string("op_28368_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28368_end_mask_0 = const()[name = string("op_28368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28368_cast_fp16 = slice_by_index(begin = var_28368_begin_0, end = var_28368_end_0, end_mask = var_28368_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28368_cast_fp16")]; + tensor var_28375_begin_0 = const()[name = string("op_28375_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28375_end_0 = const()[name = string("op_28375_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28375_end_mask_0 = const()[name = string("op_28375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28375_cast_fp16 = slice_by_index(begin = var_28375_begin_0, end = var_28375_end_0, end_mask = var_28375_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28375_cast_fp16")]; + tensor var_28382_begin_0 = const()[name = string("op_28382_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28382_end_0 = const()[name = string("op_28382_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28382_end_mask_0 = const()[name = string("op_28382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28382_cast_fp16 = slice_by_index(begin = var_28382_begin_0, end = var_28382_end_0, end_mask = var_28382_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28382_cast_fp16")]; + tensor k_37_perm_0 = const()[name = string("k_37_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28387_begin_0 = const()[name = string("op_28387_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28387_end_0 = const()[name = string("op_28387_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_28387_end_mask_0 = const()[name = string("op_28387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_37_cast_fp16 = transpose(perm = k_37_perm_0, x = key_37_cast_fp16)[name = string("transpose_13")]; + tensor var_28387_cast_fp16 = slice_by_index(begin = var_28387_begin_0, end = var_28387_end_0, end_mask = var_28387_end_mask_0, x = k_37_cast_fp16)[name = string("op_28387_cast_fp16")]; + tensor var_28391_begin_0 = const()[name = string("op_28391_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28391_end_0 = const()[name = string("op_28391_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_28391_end_mask_0 = const()[name = string("op_28391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28391_cast_fp16 = slice_by_index(begin = var_28391_begin_0, end = var_28391_end_0, end_mask = var_28391_end_mask_0, x = k_37_cast_fp16)[name = string("op_28391_cast_fp16")]; + tensor var_28395_begin_0 = const()[name = string("op_28395_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28395_end_0 = const()[name = string("op_28395_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_28395_end_mask_0 = const()[name = string("op_28395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28395_cast_fp16 = slice_by_index(begin = var_28395_begin_0, end = var_28395_end_0, end_mask = var_28395_end_mask_0, x = k_37_cast_fp16)[name = string("op_28395_cast_fp16")]; + tensor var_28399_begin_0 = const()[name = string("op_28399_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28399_end_0 = const()[name = string("op_28399_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_28399_end_mask_0 = const()[name = string("op_28399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28399_cast_fp16 = slice_by_index(begin = var_28399_begin_0, end = var_28399_end_0, end_mask = var_28399_end_mask_0, x = k_37_cast_fp16)[name = string("op_28399_cast_fp16")]; + tensor var_28403_begin_0 = const()[name = string("op_28403_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28403_end_0 = const()[name = string("op_28403_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_28403_end_mask_0 = const()[name = string("op_28403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28403_cast_fp16 = slice_by_index(begin = var_28403_begin_0, end = var_28403_end_0, end_mask = var_28403_end_mask_0, x = k_37_cast_fp16)[name = string("op_28403_cast_fp16")]; + tensor var_28407_begin_0 = const()[name = string("op_28407_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28407_end_0 = const()[name = string("op_28407_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_28407_end_mask_0 = const()[name = string("op_28407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28407_cast_fp16 = slice_by_index(begin = var_28407_begin_0, end = var_28407_end_0, end_mask = var_28407_end_mask_0, x = k_37_cast_fp16)[name = string("op_28407_cast_fp16")]; + tensor var_28411_begin_0 = const()[name = string("op_28411_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28411_end_0 = const()[name = string("op_28411_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_28411_end_mask_0 = const()[name = string("op_28411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28411_cast_fp16 = slice_by_index(begin = var_28411_begin_0, end = var_28411_end_0, end_mask = var_28411_end_mask_0, x = k_37_cast_fp16)[name = string("op_28411_cast_fp16")]; + tensor var_28415_begin_0 = const()[name = string("op_28415_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28415_end_0 = const()[name = string("op_28415_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_28415_end_mask_0 = const()[name = string("op_28415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28415_cast_fp16 = slice_by_index(begin = var_28415_begin_0, end = var_28415_end_0, end_mask = var_28415_end_mask_0, x = k_37_cast_fp16)[name = string("op_28415_cast_fp16")]; + tensor var_28419_begin_0 = const()[name = string("op_28419_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28419_end_0 = const()[name = string("op_28419_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_28419_end_mask_0 = const()[name = string("op_28419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28419_cast_fp16 = slice_by_index(begin = var_28419_begin_0, end = var_28419_end_0, end_mask = var_28419_end_mask_0, x = k_37_cast_fp16)[name = string("op_28419_cast_fp16")]; + tensor var_28423_begin_0 = const()[name = string("op_28423_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28423_end_0 = const()[name = string("op_28423_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_28423_end_mask_0 = const()[name = string("op_28423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28423_cast_fp16 = slice_by_index(begin = var_28423_begin_0, end = var_28423_end_0, end_mask = var_28423_end_mask_0, x = k_37_cast_fp16)[name = string("op_28423_cast_fp16")]; + tensor var_28427_begin_0 = const()[name = string("op_28427_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28427_end_0 = const()[name = string("op_28427_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_28427_end_mask_0 = const()[name = string("op_28427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28427_cast_fp16 = slice_by_index(begin = var_28427_begin_0, end = var_28427_end_0, end_mask = var_28427_end_mask_0, x = k_37_cast_fp16)[name = string("op_28427_cast_fp16")]; + tensor var_28431_begin_0 = const()[name = string("op_28431_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28431_end_0 = const()[name = string("op_28431_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_28431_end_mask_0 = const()[name = string("op_28431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28431_cast_fp16 = slice_by_index(begin = var_28431_begin_0, end = var_28431_end_0, end_mask = var_28431_end_mask_0, x = k_37_cast_fp16)[name = string("op_28431_cast_fp16")]; + tensor var_28435_begin_0 = const()[name = string("op_28435_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28435_end_0 = const()[name = string("op_28435_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_28435_end_mask_0 = const()[name = string("op_28435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28435_cast_fp16 = slice_by_index(begin = var_28435_begin_0, end = var_28435_end_0, end_mask = var_28435_end_mask_0, x = k_37_cast_fp16)[name = string("op_28435_cast_fp16")]; + tensor var_28439_begin_0 = const()[name = string("op_28439_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28439_end_0 = const()[name = string("op_28439_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_28439_end_mask_0 = const()[name = string("op_28439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28439_cast_fp16 = slice_by_index(begin = var_28439_begin_0, end = var_28439_end_0, end_mask = var_28439_end_mask_0, x = k_37_cast_fp16)[name = string("op_28439_cast_fp16")]; + tensor var_28443_begin_0 = const()[name = string("op_28443_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28443_end_0 = const()[name = string("op_28443_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_28443_end_mask_0 = const()[name = string("op_28443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28443_cast_fp16 = slice_by_index(begin = var_28443_begin_0, end = var_28443_end_0, end_mask = var_28443_end_mask_0, x = k_37_cast_fp16)[name = string("op_28443_cast_fp16")]; + tensor var_28447_begin_0 = const()[name = string("op_28447_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28447_end_0 = const()[name = string("op_28447_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_28447_end_mask_0 = const()[name = string("op_28447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28447_cast_fp16 = slice_by_index(begin = var_28447_begin_0, end = var_28447_end_0, end_mask = var_28447_end_mask_0, x = k_37_cast_fp16)[name = string("op_28447_cast_fp16")]; + tensor var_28451_begin_0 = const()[name = string("op_28451_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28451_end_0 = const()[name = string("op_28451_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_28451_end_mask_0 = const()[name = string("op_28451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28451_cast_fp16 = slice_by_index(begin = var_28451_begin_0, end = var_28451_end_0, end_mask = var_28451_end_mask_0, x = k_37_cast_fp16)[name = string("op_28451_cast_fp16")]; + tensor var_28455_begin_0 = const()[name = string("op_28455_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28455_end_0 = const()[name = string("op_28455_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_28455_end_mask_0 = const()[name = string("op_28455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28455_cast_fp16 = slice_by_index(begin = var_28455_begin_0, end = var_28455_end_0, end_mask = var_28455_end_mask_0, x = k_37_cast_fp16)[name = string("op_28455_cast_fp16")]; + tensor var_28459_begin_0 = const()[name = string("op_28459_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28459_end_0 = const()[name = string("op_28459_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_28459_end_mask_0 = const()[name = string("op_28459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28459_cast_fp16 = slice_by_index(begin = var_28459_begin_0, end = var_28459_end_0, end_mask = var_28459_end_mask_0, x = k_37_cast_fp16)[name = string("op_28459_cast_fp16")]; + tensor var_28463_begin_0 = const()[name = string("op_28463_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28463_end_0 = const()[name = string("op_28463_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_28463_end_mask_0 = const()[name = string("op_28463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28463_cast_fp16 = slice_by_index(begin = var_28463_begin_0, end = var_28463_end_0, end_mask = var_28463_end_mask_0, x = k_37_cast_fp16)[name = string("op_28463_cast_fp16")]; + tensor var_28465_begin_0 = const()[name = string("op_28465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28465_end_0 = const()[name = string("op_28465_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28465_end_mask_0 = const()[name = string("op_28465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28465_cast_fp16 = slice_by_index(begin = var_28465_begin_0, end = var_28465_end_0, end_mask = var_28465_end_mask_0, x = value_37_cast_fp16)[name = string("op_28465_cast_fp16")]; + tensor var_28469_begin_0 = const()[name = string("op_28469_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28469_end_0 = const()[name = string("op_28469_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_28469_end_mask_0 = const()[name = string("op_28469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28469_cast_fp16 = slice_by_index(begin = var_28469_begin_0, end = var_28469_end_0, end_mask = var_28469_end_mask_0, x = value_37_cast_fp16)[name = string("op_28469_cast_fp16")]; + tensor var_28473_begin_0 = const()[name = string("op_28473_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28473_end_0 = const()[name = string("op_28473_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_28473_end_mask_0 = const()[name = string("op_28473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28473_cast_fp16 = slice_by_index(begin = var_28473_begin_0, end = var_28473_end_0, end_mask = var_28473_end_mask_0, x = value_37_cast_fp16)[name = string("op_28473_cast_fp16")]; + tensor var_28477_begin_0 = const()[name = string("op_28477_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28477_end_0 = const()[name = string("op_28477_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_28477_end_mask_0 = const()[name = string("op_28477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28477_cast_fp16 = slice_by_index(begin = var_28477_begin_0, end = var_28477_end_0, end_mask = var_28477_end_mask_0, x = value_37_cast_fp16)[name = string("op_28477_cast_fp16")]; + tensor var_28481_begin_0 = const()[name = string("op_28481_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28481_end_0 = const()[name = string("op_28481_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_28481_end_mask_0 = const()[name = string("op_28481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28481_cast_fp16 = slice_by_index(begin = var_28481_begin_0, end = var_28481_end_0, end_mask = var_28481_end_mask_0, x = value_37_cast_fp16)[name = string("op_28481_cast_fp16")]; + tensor var_28485_begin_0 = const()[name = string("op_28485_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28485_end_0 = const()[name = string("op_28485_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_28485_end_mask_0 = const()[name = string("op_28485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28485_cast_fp16 = slice_by_index(begin = var_28485_begin_0, end = var_28485_end_0, end_mask = var_28485_end_mask_0, x = value_37_cast_fp16)[name = string("op_28485_cast_fp16")]; + tensor var_28489_begin_0 = const()[name = string("op_28489_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28489_end_0 = const()[name = string("op_28489_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_28489_end_mask_0 = const()[name = string("op_28489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28489_cast_fp16 = slice_by_index(begin = var_28489_begin_0, end = var_28489_end_0, end_mask = var_28489_end_mask_0, x = value_37_cast_fp16)[name = string("op_28489_cast_fp16")]; + tensor var_28493_begin_0 = const()[name = string("op_28493_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28493_end_0 = const()[name = string("op_28493_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_28493_end_mask_0 = const()[name = string("op_28493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28493_cast_fp16 = slice_by_index(begin = var_28493_begin_0, end = var_28493_end_0, end_mask = var_28493_end_mask_0, x = value_37_cast_fp16)[name = string("op_28493_cast_fp16")]; + tensor var_28497_begin_0 = const()[name = string("op_28497_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28497_end_0 = const()[name = string("op_28497_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_28497_end_mask_0 = const()[name = string("op_28497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28497_cast_fp16 = slice_by_index(begin = var_28497_begin_0, end = var_28497_end_0, end_mask = var_28497_end_mask_0, x = value_37_cast_fp16)[name = string("op_28497_cast_fp16")]; + tensor var_28501_begin_0 = const()[name = string("op_28501_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28501_end_0 = const()[name = string("op_28501_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_28501_end_mask_0 = const()[name = string("op_28501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28501_cast_fp16 = slice_by_index(begin = var_28501_begin_0, end = var_28501_end_0, end_mask = var_28501_end_mask_0, x = value_37_cast_fp16)[name = string("op_28501_cast_fp16")]; + tensor var_28505_begin_0 = const()[name = string("op_28505_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28505_end_0 = const()[name = string("op_28505_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_28505_end_mask_0 = const()[name = string("op_28505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28505_cast_fp16 = slice_by_index(begin = var_28505_begin_0, end = var_28505_end_0, end_mask = var_28505_end_mask_0, x = value_37_cast_fp16)[name = string("op_28505_cast_fp16")]; + tensor var_28509_begin_0 = const()[name = string("op_28509_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28509_end_0 = const()[name = string("op_28509_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_28509_end_mask_0 = const()[name = string("op_28509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28509_cast_fp16 = slice_by_index(begin = var_28509_begin_0, end = var_28509_end_0, end_mask = var_28509_end_mask_0, x = value_37_cast_fp16)[name = string("op_28509_cast_fp16")]; + tensor var_28513_begin_0 = const()[name = string("op_28513_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28513_end_0 = const()[name = string("op_28513_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_28513_end_mask_0 = const()[name = string("op_28513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28513_cast_fp16 = slice_by_index(begin = var_28513_begin_0, end = var_28513_end_0, end_mask = var_28513_end_mask_0, x = value_37_cast_fp16)[name = string("op_28513_cast_fp16")]; + tensor var_28517_begin_0 = const()[name = string("op_28517_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28517_end_0 = const()[name = string("op_28517_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_28517_end_mask_0 = const()[name = string("op_28517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28517_cast_fp16 = slice_by_index(begin = var_28517_begin_0, end = var_28517_end_0, end_mask = var_28517_end_mask_0, x = value_37_cast_fp16)[name = string("op_28517_cast_fp16")]; + tensor var_28521_begin_0 = const()[name = string("op_28521_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28521_end_0 = const()[name = string("op_28521_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_28521_end_mask_0 = const()[name = string("op_28521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28521_cast_fp16 = slice_by_index(begin = var_28521_begin_0, end = var_28521_end_0, end_mask = var_28521_end_mask_0, x = value_37_cast_fp16)[name = string("op_28521_cast_fp16")]; + tensor var_28525_begin_0 = const()[name = string("op_28525_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28525_end_0 = const()[name = string("op_28525_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_28525_end_mask_0 = const()[name = string("op_28525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28525_cast_fp16 = slice_by_index(begin = var_28525_begin_0, end = var_28525_end_0, end_mask = var_28525_end_mask_0, x = value_37_cast_fp16)[name = string("op_28525_cast_fp16")]; + tensor var_28529_begin_0 = const()[name = string("op_28529_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28529_end_0 = const()[name = string("op_28529_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_28529_end_mask_0 = const()[name = string("op_28529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28529_cast_fp16 = slice_by_index(begin = var_28529_begin_0, end = var_28529_end_0, end_mask = var_28529_end_mask_0, x = value_37_cast_fp16)[name = string("op_28529_cast_fp16")]; + tensor var_28533_begin_0 = const()[name = string("op_28533_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28533_end_0 = const()[name = string("op_28533_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_28533_end_mask_0 = const()[name = string("op_28533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28533_cast_fp16 = slice_by_index(begin = var_28533_begin_0, end = var_28533_end_0, end_mask = var_28533_end_mask_0, x = value_37_cast_fp16)[name = string("op_28533_cast_fp16")]; + tensor var_28537_begin_0 = const()[name = string("op_28537_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28537_end_0 = const()[name = string("op_28537_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_28537_end_mask_0 = const()[name = string("op_28537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28537_cast_fp16 = slice_by_index(begin = var_28537_begin_0, end = var_28537_end_0, end_mask = var_28537_end_mask_0, x = value_37_cast_fp16)[name = string("op_28537_cast_fp16")]; + tensor var_28541_begin_0 = const()[name = string("op_28541_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28541_end_0 = const()[name = string("op_28541_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_28541_end_mask_0 = const()[name = string("op_28541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28541_cast_fp16 = slice_by_index(begin = var_28541_begin_0, end = var_28541_end_0, end_mask = var_28541_end_mask_0, x = value_37_cast_fp16)[name = string("op_28541_cast_fp16")]; + string _SplitHeadsQ__mh_w_2881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2881_equation_0, values = (var_28387_cast_fp16, var_27829_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2881_cast_fp16")]; + string _SplitHeadsQ__mh_w_2883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2883_equation_0, values = (var_28387_cast_fp16, var_27836_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2883_cast_fp16")]; + string _SplitHeadsQ__mh_w_2885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2885_equation_0, values = (var_28387_cast_fp16, var_27843_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2885_cast_fp16")]; + string _SplitHeadsQ__mh_w_2887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2887_equation_0, values = (var_28387_cast_fp16, var_27850_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2887_cast_fp16")]; + string _SplitHeadsQ__mh_w_2889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2889_equation_0, values = (var_28391_cast_fp16, var_27857_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2889_cast_fp16")]; + string _SplitHeadsQ__mh_w_2891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2891_equation_0, values = (var_28391_cast_fp16, var_27864_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2891_cast_fp16")]; + string _SplitHeadsQ__mh_w_2893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2893_equation_0, values = (var_28391_cast_fp16, var_27871_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2893_cast_fp16")]; + string _SplitHeadsQ__mh_w_2895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2895_equation_0, values = (var_28391_cast_fp16, var_27878_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2895_cast_fp16")]; + string _SplitHeadsQ__mh_w_2897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2897_equation_0, values = (var_28395_cast_fp16, var_27885_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2897_cast_fp16")]; + string _SplitHeadsQ__mh_w_2899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2899_equation_0, values = (var_28395_cast_fp16, var_27892_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2899_cast_fp16")]; + string _SplitHeadsQ__mh_w_2901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2901_equation_0, values = (var_28395_cast_fp16, var_27899_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2901_cast_fp16")]; + string _SplitHeadsQ__mh_w_2903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2903_equation_0, values = (var_28395_cast_fp16, var_27906_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2903_cast_fp16")]; + string _SplitHeadsQ__mh_w_2905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2905_equation_0, values = (var_28399_cast_fp16, var_27913_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2905_cast_fp16")]; + string _SplitHeadsQ__mh_w_2907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2907_equation_0, values = (var_28399_cast_fp16, var_27920_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2907_cast_fp16")]; + string _SplitHeadsQ__mh_w_2909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2909_equation_0, values = (var_28399_cast_fp16, var_27927_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2909_cast_fp16")]; + string _SplitHeadsQ__mh_w_2911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2911_equation_0, values = (var_28399_cast_fp16, var_27934_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2911_cast_fp16")]; + string _SplitHeadsQ__mh_w_2913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2913_equation_0, values = (var_28403_cast_fp16, var_27941_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2913_cast_fp16")]; + string _SplitHeadsQ__mh_w_2915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2915_equation_0, values = (var_28403_cast_fp16, var_27948_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2915_cast_fp16")]; + string _SplitHeadsQ__mh_w_2917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2917_equation_0, values = (var_28403_cast_fp16, var_27955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2917_cast_fp16")]; + string _SplitHeadsQ__mh_w_2919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2919_equation_0, values = (var_28403_cast_fp16, var_27962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2919_cast_fp16")]; + string _SplitHeadsQ__mh_w_2921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2921_equation_0, values = (var_28407_cast_fp16, var_27969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2921_cast_fp16")]; + string _SplitHeadsQ__mh_w_2923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2923_equation_0, values = (var_28407_cast_fp16, var_27976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2923_cast_fp16")]; + string _SplitHeadsQ__mh_w_2925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2925_equation_0, values = (var_28407_cast_fp16, var_27983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2925_cast_fp16")]; + string _SplitHeadsQ__mh_w_2927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2927_equation_0, values = (var_28407_cast_fp16, var_27990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2927_cast_fp16")]; + string _SplitHeadsQ__mh_w_2929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2929_equation_0, values = (var_28411_cast_fp16, var_27997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2929_cast_fp16")]; + string _SplitHeadsQ__mh_w_2931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2931_equation_0, values = (var_28411_cast_fp16, var_28004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2931_cast_fp16")]; + string _SplitHeadsQ__mh_w_2933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2933_equation_0, values = (var_28411_cast_fp16, var_28011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2933_cast_fp16")]; + string _SplitHeadsQ__mh_w_2935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2935_equation_0, values = (var_28411_cast_fp16, var_28018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2935_cast_fp16")]; + string _SplitHeadsQ__mh_w_2937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2937_equation_0, values = (var_28415_cast_fp16, var_28025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2937_cast_fp16")]; + string _SplitHeadsQ__mh_w_2939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2939_equation_0, values = (var_28415_cast_fp16, var_28032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2939_cast_fp16")]; + string _SplitHeadsQ__mh_w_2941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2941_equation_0, values = (var_28415_cast_fp16, var_28039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2941_cast_fp16")]; + string _SplitHeadsQ__mh_w_2943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2943_equation_0, values = (var_28415_cast_fp16, var_28046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2943_cast_fp16")]; + string _SplitHeadsQ__mh_w_2945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2945_equation_0, values = (var_28419_cast_fp16, var_28053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2945_cast_fp16")]; + string _SplitHeadsQ__mh_w_2947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2947_equation_0, values = (var_28419_cast_fp16, var_28060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2947_cast_fp16")]; + string _SplitHeadsQ__mh_w_2949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2949_equation_0, values = (var_28419_cast_fp16, var_28067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2949_cast_fp16")]; + string _SplitHeadsQ__mh_w_2951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2951_equation_0, values = (var_28419_cast_fp16, var_28074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2951_cast_fp16")]; + string _SplitHeadsQ__mh_w_2953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2953_equation_0, values = (var_28423_cast_fp16, var_28081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2953_cast_fp16")]; + string _SplitHeadsQ__mh_w_2955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2955_equation_0, values = (var_28423_cast_fp16, var_28088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2955_cast_fp16")]; + string _SplitHeadsQ__mh_w_2957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2957_equation_0, values = (var_28423_cast_fp16, var_28095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2957_cast_fp16")]; + string _SplitHeadsQ__mh_w_2959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2959_equation_0, values = (var_28423_cast_fp16, var_28102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2959_cast_fp16")]; + string _SplitHeadsQ__mh_w_2961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2961_equation_0, values = (var_28427_cast_fp16, var_28109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2961_cast_fp16")]; + string _SplitHeadsQ__mh_w_2963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2963_equation_0, values = (var_28427_cast_fp16, var_28116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2963_cast_fp16")]; + string _SplitHeadsQ__mh_w_2965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2965_equation_0, values = (var_28427_cast_fp16, var_28123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2965_cast_fp16")]; + string _SplitHeadsQ__mh_w_2967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2967_equation_0, values = (var_28427_cast_fp16, var_28130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2967_cast_fp16")]; + string _SplitHeadsQ__mh_w_2969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2969_equation_0, values = (var_28431_cast_fp16, var_28137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2969_cast_fp16")]; + string _SplitHeadsQ__mh_w_2971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2971_equation_0, values = (var_28431_cast_fp16, var_28144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2971_cast_fp16")]; + string _SplitHeadsQ__mh_w_2973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2973_equation_0, values = (var_28431_cast_fp16, var_28151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2973_cast_fp16")]; + string _SplitHeadsQ__mh_w_2975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2975_equation_0, values = (var_28431_cast_fp16, var_28158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2975_cast_fp16")]; + string _SplitHeadsQ__mh_w_2977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2977_equation_0, values = (var_28435_cast_fp16, var_28165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2977_cast_fp16")]; + string _SplitHeadsQ__mh_w_2979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2979_equation_0, values = (var_28435_cast_fp16, var_28172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2979_cast_fp16")]; + string _SplitHeadsQ__mh_w_2981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2981_equation_0, values = (var_28435_cast_fp16, var_28179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2981_cast_fp16")]; + string _SplitHeadsQ__mh_w_2983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2983_equation_0, values = (var_28435_cast_fp16, var_28186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2983_cast_fp16")]; + string _SplitHeadsQ__mh_w_2985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2985_equation_0, values = (var_28439_cast_fp16, var_28193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2985_cast_fp16")]; + string _SplitHeadsQ__mh_w_2987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2987_equation_0, values = (var_28439_cast_fp16, var_28200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2987_cast_fp16")]; + string _SplitHeadsQ__mh_w_2989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2989_equation_0, values = (var_28439_cast_fp16, var_28207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2989_cast_fp16")]; + string _SplitHeadsQ__mh_w_2991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2991_equation_0, values = (var_28439_cast_fp16, var_28214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2991_cast_fp16")]; + string _SplitHeadsQ__mh_w_2993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2993_equation_0, values = (var_28443_cast_fp16, var_28221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2993_cast_fp16")]; + string _SplitHeadsQ__mh_w_2995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2995_equation_0, values = (var_28443_cast_fp16, var_28228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2995_cast_fp16")]; + string _SplitHeadsQ__mh_w_2997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2997_equation_0, values = (var_28443_cast_fp16, var_28235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2997_cast_fp16")]; + string _SplitHeadsQ__mh_w_2999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2999_equation_0, values = (var_28443_cast_fp16, var_28242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2999_cast_fp16")]; + string _SplitHeadsQ__mh_w_3001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3001_equation_0, values = (var_28447_cast_fp16, var_28249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3001_cast_fp16")]; + string _SplitHeadsQ__mh_w_3003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3003_equation_0, values = (var_28447_cast_fp16, var_28256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3003_cast_fp16")]; + string _SplitHeadsQ__mh_w_3005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3005_equation_0, values = (var_28447_cast_fp16, var_28263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3005_cast_fp16")]; + string _SplitHeadsQ__mh_w_3007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3007_equation_0, values = (var_28447_cast_fp16, var_28270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3007_cast_fp16")]; + string _SplitHeadsQ__mh_w_3009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3009_equation_0, values = (var_28451_cast_fp16, var_28277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3009_cast_fp16")]; + string _SplitHeadsQ__mh_w_3011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3011_equation_0, values = (var_28451_cast_fp16, var_28284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3011_cast_fp16")]; + string _SplitHeadsQ__mh_w_3013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3013_equation_0, values = (var_28451_cast_fp16, var_28291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3013_cast_fp16")]; + string _SplitHeadsQ__mh_w_3015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3015_equation_0, values = (var_28451_cast_fp16, var_28298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3015_cast_fp16")]; + string _SplitHeadsQ__mh_w_3017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3017_equation_0, values = (var_28455_cast_fp16, var_28305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3017_cast_fp16")]; + string _SplitHeadsQ__mh_w_3019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3019_equation_0, values = (var_28455_cast_fp16, var_28312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3019_cast_fp16")]; + string _SplitHeadsQ__mh_w_3021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3021_equation_0, values = (var_28455_cast_fp16, var_28319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3021_cast_fp16")]; + string _SplitHeadsQ__mh_w_3023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3023_equation_0, values = (var_28455_cast_fp16, var_28326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3023_cast_fp16")]; + string _SplitHeadsQ__mh_w_3025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3025_equation_0, values = (var_28459_cast_fp16, var_28333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3025_cast_fp16")]; + string _SplitHeadsQ__mh_w_3027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3027_equation_0, values = (var_28459_cast_fp16, var_28340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3027_cast_fp16")]; + string _SplitHeadsQ__mh_w_3029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3029_equation_0, values = (var_28459_cast_fp16, var_28347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3029_cast_fp16")]; + string _SplitHeadsQ__mh_w_3031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3031_equation_0, values = (var_28459_cast_fp16, var_28354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3031_cast_fp16")]; + string _SplitHeadsQ__mh_w_3033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3033_equation_0, values = (var_28463_cast_fp16, var_28361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3033_cast_fp16")]; + string _SplitHeadsQ__mh_w_3035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3035_equation_0, values = (var_28463_cast_fp16, var_28368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3035_cast_fp16")]; + string _SplitHeadsQ__mh_w_3037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3037_equation_0, values = (var_28463_cast_fp16, var_28375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3037_cast_fp16")]; + string _SplitHeadsQ__mh_w_3039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3039_equation_0, values = (var_28463_cast_fp16, var_28382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3039_cast_fp16")]; + fp16 var_28704_to_fp16 = const()[name = string("op_28704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2881_cast_fp16, y = var_28704_to_fp16)[name = string("aw_chunk_2881_cast_fp16")]; + fp16 var_28706_to_fp16 = const()[name = string("op_28706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2883_cast_fp16, y = var_28706_to_fp16)[name = string("aw_chunk_2883_cast_fp16")]; + fp16 var_28708_to_fp16 = const()[name = string("op_28708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2885_cast_fp16, y = var_28708_to_fp16)[name = string("aw_chunk_2885_cast_fp16")]; + fp16 var_28710_to_fp16 = const()[name = string("op_28710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2887_cast_fp16, y = var_28710_to_fp16)[name = string("aw_chunk_2887_cast_fp16")]; + fp16 var_28712_to_fp16 = const()[name = string("op_28712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2889_cast_fp16, y = var_28712_to_fp16)[name = string("aw_chunk_2889_cast_fp16")]; + fp16 var_28714_to_fp16 = const()[name = string("op_28714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2891_cast_fp16, y = var_28714_to_fp16)[name = string("aw_chunk_2891_cast_fp16")]; + fp16 var_28716_to_fp16 = const()[name = string("op_28716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2893_cast_fp16, y = var_28716_to_fp16)[name = string("aw_chunk_2893_cast_fp16")]; + fp16 var_28718_to_fp16 = const()[name = string("op_28718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2895_cast_fp16, y = var_28718_to_fp16)[name = string("aw_chunk_2895_cast_fp16")]; + fp16 var_28720_to_fp16 = const()[name = string("op_28720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2897_cast_fp16, y = var_28720_to_fp16)[name = string("aw_chunk_2897_cast_fp16")]; + fp16 var_28722_to_fp16 = const()[name = string("op_28722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2899_cast_fp16, y = var_28722_to_fp16)[name = string("aw_chunk_2899_cast_fp16")]; + fp16 var_28724_to_fp16 = const()[name = string("op_28724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2901_cast_fp16, y = var_28724_to_fp16)[name = string("aw_chunk_2901_cast_fp16")]; + fp16 var_28726_to_fp16 = const()[name = string("op_28726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2903_cast_fp16, y = var_28726_to_fp16)[name = string("aw_chunk_2903_cast_fp16")]; + fp16 var_28728_to_fp16 = const()[name = string("op_28728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2905_cast_fp16, y = var_28728_to_fp16)[name = string("aw_chunk_2905_cast_fp16")]; + fp16 var_28730_to_fp16 = const()[name = string("op_28730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2907_cast_fp16, y = var_28730_to_fp16)[name = string("aw_chunk_2907_cast_fp16")]; + fp16 var_28732_to_fp16 = const()[name = string("op_28732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2909_cast_fp16, y = var_28732_to_fp16)[name = string("aw_chunk_2909_cast_fp16")]; + fp16 var_28734_to_fp16 = const()[name = string("op_28734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2911_cast_fp16, y = var_28734_to_fp16)[name = string("aw_chunk_2911_cast_fp16")]; + fp16 var_28736_to_fp16 = const()[name = string("op_28736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2913_cast_fp16, y = var_28736_to_fp16)[name = string("aw_chunk_2913_cast_fp16")]; + fp16 var_28738_to_fp16 = const()[name = string("op_28738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2915_cast_fp16, y = var_28738_to_fp16)[name = string("aw_chunk_2915_cast_fp16")]; + fp16 var_28740_to_fp16 = const()[name = string("op_28740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2917_cast_fp16, y = var_28740_to_fp16)[name = string("aw_chunk_2917_cast_fp16")]; + fp16 var_28742_to_fp16 = const()[name = string("op_28742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2919_cast_fp16, y = var_28742_to_fp16)[name = string("aw_chunk_2919_cast_fp16")]; + fp16 var_28744_to_fp16 = const()[name = string("op_28744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2921_cast_fp16, y = var_28744_to_fp16)[name = string("aw_chunk_2921_cast_fp16")]; + fp16 var_28746_to_fp16 = const()[name = string("op_28746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2923_cast_fp16, y = var_28746_to_fp16)[name = string("aw_chunk_2923_cast_fp16")]; + fp16 var_28748_to_fp16 = const()[name = string("op_28748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2925_cast_fp16, y = var_28748_to_fp16)[name = string("aw_chunk_2925_cast_fp16")]; + fp16 var_28750_to_fp16 = const()[name = string("op_28750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2927_cast_fp16, y = var_28750_to_fp16)[name = string("aw_chunk_2927_cast_fp16")]; + fp16 var_28752_to_fp16 = const()[name = string("op_28752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2929_cast_fp16, y = var_28752_to_fp16)[name = string("aw_chunk_2929_cast_fp16")]; + fp16 var_28754_to_fp16 = const()[name = string("op_28754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2931_cast_fp16, y = var_28754_to_fp16)[name = string("aw_chunk_2931_cast_fp16")]; + fp16 var_28756_to_fp16 = const()[name = string("op_28756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2933_cast_fp16, y = var_28756_to_fp16)[name = string("aw_chunk_2933_cast_fp16")]; + fp16 var_28758_to_fp16 = const()[name = string("op_28758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2935_cast_fp16, y = var_28758_to_fp16)[name = string("aw_chunk_2935_cast_fp16")]; + fp16 var_28760_to_fp16 = const()[name = string("op_28760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2937_cast_fp16, y = var_28760_to_fp16)[name = string("aw_chunk_2937_cast_fp16")]; + fp16 var_28762_to_fp16 = const()[name = string("op_28762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2939_cast_fp16, y = var_28762_to_fp16)[name = string("aw_chunk_2939_cast_fp16")]; + fp16 var_28764_to_fp16 = const()[name = string("op_28764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2941_cast_fp16, y = var_28764_to_fp16)[name = string("aw_chunk_2941_cast_fp16")]; + fp16 var_28766_to_fp16 = const()[name = string("op_28766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2943_cast_fp16, y = var_28766_to_fp16)[name = string("aw_chunk_2943_cast_fp16")]; + fp16 var_28768_to_fp16 = const()[name = string("op_28768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2945_cast_fp16, y = var_28768_to_fp16)[name = string("aw_chunk_2945_cast_fp16")]; + fp16 var_28770_to_fp16 = const()[name = string("op_28770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2947_cast_fp16, y = var_28770_to_fp16)[name = string("aw_chunk_2947_cast_fp16")]; + fp16 var_28772_to_fp16 = const()[name = string("op_28772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2949_cast_fp16, y = var_28772_to_fp16)[name = string("aw_chunk_2949_cast_fp16")]; + fp16 var_28774_to_fp16 = const()[name = string("op_28774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2951_cast_fp16, y = var_28774_to_fp16)[name = string("aw_chunk_2951_cast_fp16")]; + fp16 var_28776_to_fp16 = const()[name = string("op_28776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2953_cast_fp16, y = var_28776_to_fp16)[name = string("aw_chunk_2953_cast_fp16")]; + fp16 var_28778_to_fp16 = const()[name = string("op_28778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2955_cast_fp16, y = var_28778_to_fp16)[name = string("aw_chunk_2955_cast_fp16")]; + fp16 var_28780_to_fp16 = const()[name = string("op_28780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2957_cast_fp16, y = var_28780_to_fp16)[name = string("aw_chunk_2957_cast_fp16")]; + fp16 var_28782_to_fp16 = const()[name = string("op_28782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2959_cast_fp16, y = var_28782_to_fp16)[name = string("aw_chunk_2959_cast_fp16")]; + fp16 var_28784_to_fp16 = const()[name = string("op_28784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2961_cast_fp16, y = var_28784_to_fp16)[name = string("aw_chunk_2961_cast_fp16")]; + fp16 var_28786_to_fp16 = const()[name = string("op_28786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2963_cast_fp16, y = var_28786_to_fp16)[name = string("aw_chunk_2963_cast_fp16")]; + fp16 var_28788_to_fp16 = const()[name = string("op_28788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2965_cast_fp16, y = var_28788_to_fp16)[name = string("aw_chunk_2965_cast_fp16")]; + fp16 var_28790_to_fp16 = const()[name = string("op_28790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2967_cast_fp16, y = var_28790_to_fp16)[name = string("aw_chunk_2967_cast_fp16")]; + fp16 var_28792_to_fp16 = const()[name = string("op_28792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2969_cast_fp16, y = var_28792_to_fp16)[name = string("aw_chunk_2969_cast_fp16")]; + fp16 var_28794_to_fp16 = const()[name = string("op_28794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2971_cast_fp16, y = var_28794_to_fp16)[name = string("aw_chunk_2971_cast_fp16")]; + fp16 var_28796_to_fp16 = const()[name = string("op_28796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2973_cast_fp16, y = var_28796_to_fp16)[name = string("aw_chunk_2973_cast_fp16")]; + fp16 var_28798_to_fp16 = const()[name = string("op_28798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2975_cast_fp16, y = var_28798_to_fp16)[name = string("aw_chunk_2975_cast_fp16")]; + fp16 var_28800_to_fp16 = const()[name = string("op_28800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2977_cast_fp16, y = var_28800_to_fp16)[name = string("aw_chunk_2977_cast_fp16")]; + fp16 var_28802_to_fp16 = const()[name = string("op_28802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2979_cast_fp16, y = var_28802_to_fp16)[name = string("aw_chunk_2979_cast_fp16")]; + fp16 var_28804_to_fp16 = const()[name = string("op_28804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2981_cast_fp16, y = var_28804_to_fp16)[name = string("aw_chunk_2981_cast_fp16")]; + fp16 var_28806_to_fp16 = const()[name = string("op_28806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2983_cast_fp16, y = var_28806_to_fp16)[name = string("aw_chunk_2983_cast_fp16")]; + fp16 var_28808_to_fp16 = const()[name = string("op_28808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2985_cast_fp16, y = var_28808_to_fp16)[name = string("aw_chunk_2985_cast_fp16")]; + fp16 var_28810_to_fp16 = const()[name = string("op_28810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2987_cast_fp16, y = var_28810_to_fp16)[name = string("aw_chunk_2987_cast_fp16")]; + fp16 var_28812_to_fp16 = const()[name = string("op_28812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2989_cast_fp16, y = var_28812_to_fp16)[name = string("aw_chunk_2989_cast_fp16")]; + fp16 var_28814_to_fp16 = const()[name = string("op_28814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2991_cast_fp16, y = var_28814_to_fp16)[name = string("aw_chunk_2991_cast_fp16")]; + fp16 var_28816_to_fp16 = const()[name = string("op_28816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2993_cast_fp16, y = var_28816_to_fp16)[name = string("aw_chunk_2993_cast_fp16")]; + fp16 var_28818_to_fp16 = const()[name = string("op_28818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2995_cast_fp16, y = var_28818_to_fp16)[name = string("aw_chunk_2995_cast_fp16")]; + fp16 var_28820_to_fp16 = const()[name = string("op_28820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2997_cast_fp16, y = var_28820_to_fp16)[name = string("aw_chunk_2997_cast_fp16")]; + fp16 var_28822_to_fp16 = const()[name = string("op_28822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2999_cast_fp16, y = var_28822_to_fp16)[name = string("aw_chunk_2999_cast_fp16")]; + fp16 var_28824_to_fp16 = const()[name = string("op_28824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3001_cast_fp16, y = var_28824_to_fp16)[name = string("aw_chunk_3001_cast_fp16")]; + fp16 var_28826_to_fp16 = const()[name = string("op_28826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3003_cast_fp16, y = var_28826_to_fp16)[name = string("aw_chunk_3003_cast_fp16")]; + fp16 var_28828_to_fp16 = const()[name = string("op_28828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3005_cast_fp16, y = var_28828_to_fp16)[name = string("aw_chunk_3005_cast_fp16")]; + fp16 var_28830_to_fp16 = const()[name = string("op_28830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3007_cast_fp16, y = var_28830_to_fp16)[name = string("aw_chunk_3007_cast_fp16")]; + fp16 var_28832_to_fp16 = const()[name = string("op_28832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3009_cast_fp16, y = var_28832_to_fp16)[name = string("aw_chunk_3009_cast_fp16")]; + fp16 var_28834_to_fp16 = const()[name = string("op_28834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3011_cast_fp16, y = var_28834_to_fp16)[name = string("aw_chunk_3011_cast_fp16")]; + fp16 var_28836_to_fp16 = const()[name = string("op_28836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3013_cast_fp16, y = var_28836_to_fp16)[name = string("aw_chunk_3013_cast_fp16")]; + fp16 var_28838_to_fp16 = const()[name = string("op_28838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3015_cast_fp16, y = var_28838_to_fp16)[name = string("aw_chunk_3015_cast_fp16")]; + fp16 var_28840_to_fp16 = const()[name = string("op_28840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3017_cast_fp16, y = var_28840_to_fp16)[name = string("aw_chunk_3017_cast_fp16")]; + fp16 var_28842_to_fp16 = const()[name = string("op_28842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3019_cast_fp16, y = var_28842_to_fp16)[name = string("aw_chunk_3019_cast_fp16")]; + fp16 var_28844_to_fp16 = const()[name = string("op_28844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3021_cast_fp16, y = var_28844_to_fp16)[name = string("aw_chunk_3021_cast_fp16")]; + fp16 var_28846_to_fp16 = const()[name = string("op_28846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3023_cast_fp16, y = var_28846_to_fp16)[name = string("aw_chunk_3023_cast_fp16")]; + fp16 var_28848_to_fp16 = const()[name = string("op_28848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3025_cast_fp16, y = var_28848_to_fp16)[name = string("aw_chunk_3025_cast_fp16")]; + fp16 var_28850_to_fp16 = const()[name = string("op_28850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3027_cast_fp16, y = var_28850_to_fp16)[name = string("aw_chunk_3027_cast_fp16")]; + fp16 var_28852_to_fp16 = const()[name = string("op_28852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3029_cast_fp16, y = var_28852_to_fp16)[name = string("aw_chunk_3029_cast_fp16")]; + fp16 var_28854_to_fp16 = const()[name = string("op_28854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3031_cast_fp16, y = var_28854_to_fp16)[name = string("aw_chunk_3031_cast_fp16")]; + fp16 var_28856_to_fp16 = const()[name = string("op_28856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3033_cast_fp16, y = var_28856_to_fp16)[name = string("aw_chunk_3033_cast_fp16")]; + fp16 var_28858_to_fp16 = const()[name = string("op_28858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3035_cast_fp16, y = var_28858_to_fp16)[name = string("aw_chunk_3035_cast_fp16")]; + fp16 var_28860_to_fp16 = const()[name = string("op_28860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3037_cast_fp16, y = var_28860_to_fp16)[name = string("aw_chunk_3037_cast_fp16")]; + fp16 var_28862_to_fp16 = const()[name = string("op_28862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3039_cast_fp16, y = var_28862_to_fp16)[name = string("aw_chunk_3039_cast_fp16")]; + tensor var_28864_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2881_cast_fp16)[name = string("op_28864_cast_fp16")]; + tensor var_28865_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2883_cast_fp16)[name = string("op_28865_cast_fp16")]; + tensor var_28866_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2885_cast_fp16)[name = string("op_28866_cast_fp16")]; + tensor var_28867_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2887_cast_fp16)[name = string("op_28867_cast_fp16")]; + tensor var_28868_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2889_cast_fp16)[name = string("op_28868_cast_fp16")]; + tensor var_28869_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2891_cast_fp16)[name = string("op_28869_cast_fp16")]; + tensor var_28870_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2893_cast_fp16)[name = string("op_28870_cast_fp16")]; + tensor var_28871_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2895_cast_fp16)[name = string("op_28871_cast_fp16")]; + tensor var_28872_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2897_cast_fp16)[name = string("op_28872_cast_fp16")]; + tensor var_28873_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2899_cast_fp16)[name = string("op_28873_cast_fp16")]; + tensor var_28874_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2901_cast_fp16)[name = string("op_28874_cast_fp16")]; + tensor var_28875_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2903_cast_fp16)[name = string("op_28875_cast_fp16")]; + tensor var_28876_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2905_cast_fp16)[name = string("op_28876_cast_fp16")]; + tensor var_28877_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2907_cast_fp16)[name = string("op_28877_cast_fp16")]; + tensor var_28878_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2909_cast_fp16)[name = string("op_28878_cast_fp16")]; + tensor var_28879_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2911_cast_fp16)[name = string("op_28879_cast_fp16")]; + tensor var_28880_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2913_cast_fp16)[name = string("op_28880_cast_fp16")]; + tensor var_28881_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2915_cast_fp16)[name = string("op_28881_cast_fp16")]; + tensor var_28882_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2917_cast_fp16)[name = string("op_28882_cast_fp16")]; + tensor var_28883_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2919_cast_fp16)[name = string("op_28883_cast_fp16")]; + tensor var_28884_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2921_cast_fp16)[name = string("op_28884_cast_fp16")]; + tensor var_28885_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2923_cast_fp16)[name = string("op_28885_cast_fp16")]; + tensor var_28886_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2925_cast_fp16)[name = string("op_28886_cast_fp16")]; + tensor var_28887_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2927_cast_fp16)[name = string("op_28887_cast_fp16")]; + tensor var_28888_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2929_cast_fp16)[name = string("op_28888_cast_fp16")]; + tensor var_28889_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2931_cast_fp16)[name = string("op_28889_cast_fp16")]; + tensor var_28890_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2933_cast_fp16)[name = string("op_28890_cast_fp16")]; + tensor var_28891_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2935_cast_fp16)[name = string("op_28891_cast_fp16")]; + tensor var_28892_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2937_cast_fp16)[name = string("op_28892_cast_fp16")]; + tensor var_28893_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2939_cast_fp16)[name = string("op_28893_cast_fp16")]; + tensor var_28894_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2941_cast_fp16)[name = string("op_28894_cast_fp16")]; + tensor var_28895_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2943_cast_fp16)[name = string("op_28895_cast_fp16")]; + tensor var_28896_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2945_cast_fp16)[name = string("op_28896_cast_fp16")]; + tensor var_28897_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2947_cast_fp16)[name = string("op_28897_cast_fp16")]; + tensor var_28898_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2949_cast_fp16)[name = string("op_28898_cast_fp16")]; + tensor var_28899_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2951_cast_fp16)[name = string("op_28899_cast_fp16")]; + tensor var_28900_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2953_cast_fp16)[name = string("op_28900_cast_fp16")]; + tensor var_28901_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2955_cast_fp16)[name = string("op_28901_cast_fp16")]; + tensor var_28902_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2957_cast_fp16)[name = string("op_28902_cast_fp16")]; + tensor var_28903_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2959_cast_fp16)[name = string("op_28903_cast_fp16")]; + tensor var_28904_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2961_cast_fp16)[name = string("op_28904_cast_fp16")]; + tensor var_28905_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2963_cast_fp16)[name = string("op_28905_cast_fp16")]; + tensor var_28906_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2965_cast_fp16)[name = string("op_28906_cast_fp16")]; + tensor var_28907_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2967_cast_fp16)[name = string("op_28907_cast_fp16")]; + tensor var_28908_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2969_cast_fp16)[name = string("op_28908_cast_fp16")]; + tensor var_28909_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2971_cast_fp16)[name = string("op_28909_cast_fp16")]; + tensor var_28910_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2973_cast_fp16)[name = string("op_28910_cast_fp16")]; + tensor var_28911_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2975_cast_fp16)[name = string("op_28911_cast_fp16")]; + tensor var_28912_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2977_cast_fp16)[name = string("op_28912_cast_fp16")]; + tensor var_28913_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2979_cast_fp16)[name = string("op_28913_cast_fp16")]; + tensor var_28914_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2981_cast_fp16)[name = string("op_28914_cast_fp16")]; + tensor var_28915_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2983_cast_fp16)[name = string("op_28915_cast_fp16")]; + tensor var_28916_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2985_cast_fp16)[name = string("op_28916_cast_fp16")]; + tensor var_28917_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2987_cast_fp16)[name = string("op_28917_cast_fp16")]; + tensor var_28918_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2989_cast_fp16)[name = string("op_28918_cast_fp16")]; + tensor var_28919_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2991_cast_fp16)[name = string("op_28919_cast_fp16")]; + tensor var_28920_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2993_cast_fp16)[name = string("op_28920_cast_fp16")]; + tensor var_28921_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2995_cast_fp16)[name = string("op_28921_cast_fp16")]; + tensor var_28922_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2997_cast_fp16)[name = string("op_28922_cast_fp16")]; + tensor var_28923_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_2999_cast_fp16)[name = string("op_28923_cast_fp16")]; + tensor var_28924_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3001_cast_fp16)[name = string("op_28924_cast_fp16")]; + tensor var_28925_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3003_cast_fp16)[name = string("op_28925_cast_fp16")]; + tensor var_28926_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3005_cast_fp16)[name = string("op_28926_cast_fp16")]; + tensor var_28927_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3007_cast_fp16)[name = string("op_28927_cast_fp16")]; + tensor var_28928_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3009_cast_fp16)[name = string("op_28928_cast_fp16")]; + tensor var_28929_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3011_cast_fp16)[name = string("op_28929_cast_fp16")]; + tensor var_28930_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3013_cast_fp16)[name = string("op_28930_cast_fp16")]; + tensor var_28931_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3015_cast_fp16)[name = string("op_28931_cast_fp16")]; + tensor var_28932_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3017_cast_fp16)[name = string("op_28932_cast_fp16")]; + tensor var_28933_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3019_cast_fp16)[name = string("op_28933_cast_fp16")]; + tensor var_28934_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3021_cast_fp16)[name = string("op_28934_cast_fp16")]; + tensor var_28935_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3023_cast_fp16)[name = string("op_28935_cast_fp16")]; + tensor var_28936_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3025_cast_fp16)[name = string("op_28936_cast_fp16")]; + tensor var_28937_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3027_cast_fp16)[name = string("op_28937_cast_fp16")]; + tensor var_28938_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3029_cast_fp16)[name = string("op_28938_cast_fp16")]; + tensor var_28939_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3031_cast_fp16)[name = string("op_28939_cast_fp16")]; + tensor var_28940_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3033_cast_fp16)[name = string("op_28940_cast_fp16")]; + tensor var_28941_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3035_cast_fp16)[name = string("op_28941_cast_fp16")]; + tensor var_28942_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3037_cast_fp16)[name = string("op_28942_cast_fp16")]; + tensor var_28943_cast_fp16 = softmax(axis = var_27689, x = aw_chunk_3039_cast_fp16)[name = string("op_28943_cast_fp16")]; + string var_28945_equation_0 = const()[name = string("op_28945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28945_cast_fp16 = einsum(equation = var_28945_equation_0, values = (var_28465_cast_fp16, var_28864_cast_fp16))[name = string("op_28945_cast_fp16")]; + string var_28947_equation_0 = const()[name = string("op_28947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28947_cast_fp16 = einsum(equation = var_28947_equation_0, values = (var_28465_cast_fp16, var_28865_cast_fp16))[name = string("op_28947_cast_fp16")]; + string var_28949_equation_0 = const()[name = string("op_28949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28949_cast_fp16 = einsum(equation = var_28949_equation_0, values = (var_28465_cast_fp16, var_28866_cast_fp16))[name = string("op_28949_cast_fp16")]; + string var_28951_equation_0 = const()[name = string("op_28951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28951_cast_fp16 = einsum(equation = var_28951_equation_0, values = (var_28465_cast_fp16, var_28867_cast_fp16))[name = string("op_28951_cast_fp16")]; + string var_28953_equation_0 = const()[name = string("op_28953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28953_cast_fp16 = einsum(equation = var_28953_equation_0, values = (var_28469_cast_fp16, var_28868_cast_fp16))[name = string("op_28953_cast_fp16")]; + string var_28955_equation_0 = const()[name = string("op_28955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28955_cast_fp16 = einsum(equation = var_28955_equation_0, values = (var_28469_cast_fp16, var_28869_cast_fp16))[name = string("op_28955_cast_fp16")]; + string var_28957_equation_0 = const()[name = string("op_28957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28957_cast_fp16 = einsum(equation = var_28957_equation_0, values = (var_28469_cast_fp16, var_28870_cast_fp16))[name = string("op_28957_cast_fp16")]; + string var_28959_equation_0 = const()[name = string("op_28959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28959_cast_fp16 = einsum(equation = var_28959_equation_0, values = (var_28469_cast_fp16, var_28871_cast_fp16))[name = string("op_28959_cast_fp16")]; + string var_28961_equation_0 = const()[name = string("op_28961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28961_cast_fp16 = einsum(equation = var_28961_equation_0, values = (var_28473_cast_fp16, var_28872_cast_fp16))[name = string("op_28961_cast_fp16")]; + string var_28963_equation_0 = const()[name = string("op_28963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28963_cast_fp16 = einsum(equation = var_28963_equation_0, values = (var_28473_cast_fp16, var_28873_cast_fp16))[name = string("op_28963_cast_fp16")]; + string var_28965_equation_0 = const()[name = string("op_28965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28965_cast_fp16 = einsum(equation = var_28965_equation_0, values = (var_28473_cast_fp16, var_28874_cast_fp16))[name = string("op_28965_cast_fp16")]; + string var_28967_equation_0 = const()[name = string("op_28967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28967_cast_fp16 = einsum(equation = var_28967_equation_0, values = (var_28473_cast_fp16, var_28875_cast_fp16))[name = string("op_28967_cast_fp16")]; + string var_28969_equation_0 = const()[name = string("op_28969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28969_cast_fp16 = einsum(equation = var_28969_equation_0, values = (var_28477_cast_fp16, var_28876_cast_fp16))[name = string("op_28969_cast_fp16")]; + string var_28971_equation_0 = const()[name = string("op_28971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28971_cast_fp16 = einsum(equation = var_28971_equation_0, values = (var_28477_cast_fp16, var_28877_cast_fp16))[name = string("op_28971_cast_fp16")]; + string var_28973_equation_0 = const()[name = string("op_28973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28973_cast_fp16 = einsum(equation = var_28973_equation_0, values = (var_28477_cast_fp16, var_28878_cast_fp16))[name = string("op_28973_cast_fp16")]; + string var_28975_equation_0 = const()[name = string("op_28975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28975_cast_fp16 = einsum(equation = var_28975_equation_0, values = (var_28477_cast_fp16, var_28879_cast_fp16))[name = string("op_28975_cast_fp16")]; + string var_28977_equation_0 = const()[name = string("op_28977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28977_cast_fp16 = einsum(equation = var_28977_equation_0, values = (var_28481_cast_fp16, var_28880_cast_fp16))[name = string("op_28977_cast_fp16")]; + string var_28979_equation_0 = const()[name = string("op_28979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28979_cast_fp16 = einsum(equation = var_28979_equation_0, values = (var_28481_cast_fp16, var_28881_cast_fp16))[name = string("op_28979_cast_fp16")]; + string var_28981_equation_0 = const()[name = string("op_28981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28981_cast_fp16 = einsum(equation = var_28981_equation_0, values = (var_28481_cast_fp16, var_28882_cast_fp16))[name = string("op_28981_cast_fp16")]; + string var_28983_equation_0 = const()[name = string("op_28983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28983_cast_fp16 = einsum(equation = var_28983_equation_0, values = (var_28481_cast_fp16, var_28883_cast_fp16))[name = string("op_28983_cast_fp16")]; + string var_28985_equation_0 = const()[name = string("op_28985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28985_cast_fp16 = einsum(equation = var_28985_equation_0, values = (var_28485_cast_fp16, var_28884_cast_fp16))[name = string("op_28985_cast_fp16")]; + string var_28987_equation_0 = const()[name = string("op_28987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28987_cast_fp16 = einsum(equation = var_28987_equation_0, values = (var_28485_cast_fp16, var_28885_cast_fp16))[name = string("op_28987_cast_fp16")]; + string var_28989_equation_0 = const()[name = string("op_28989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28989_cast_fp16 = einsum(equation = var_28989_equation_0, values = (var_28485_cast_fp16, var_28886_cast_fp16))[name = string("op_28989_cast_fp16")]; + string var_28991_equation_0 = const()[name = string("op_28991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28991_cast_fp16 = einsum(equation = var_28991_equation_0, values = (var_28485_cast_fp16, var_28887_cast_fp16))[name = string("op_28991_cast_fp16")]; + string var_28993_equation_0 = const()[name = string("op_28993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28993_cast_fp16 = einsum(equation = var_28993_equation_0, values = (var_28489_cast_fp16, var_28888_cast_fp16))[name = string("op_28993_cast_fp16")]; + string var_28995_equation_0 = const()[name = string("op_28995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28995_cast_fp16 = einsum(equation = var_28995_equation_0, values = (var_28489_cast_fp16, var_28889_cast_fp16))[name = string("op_28995_cast_fp16")]; + string var_28997_equation_0 = const()[name = string("op_28997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28997_cast_fp16 = einsum(equation = var_28997_equation_0, values = (var_28489_cast_fp16, var_28890_cast_fp16))[name = string("op_28997_cast_fp16")]; + string var_28999_equation_0 = const()[name = string("op_28999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28999_cast_fp16 = einsum(equation = var_28999_equation_0, values = (var_28489_cast_fp16, var_28891_cast_fp16))[name = string("op_28999_cast_fp16")]; + string var_29001_equation_0 = const()[name = string("op_29001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29001_cast_fp16 = einsum(equation = var_29001_equation_0, values = (var_28493_cast_fp16, var_28892_cast_fp16))[name = string("op_29001_cast_fp16")]; + string var_29003_equation_0 = const()[name = string("op_29003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29003_cast_fp16 = einsum(equation = var_29003_equation_0, values = (var_28493_cast_fp16, var_28893_cast_fp16))[name = string("op_29003_cast_fp16")]; + string var_29005_equation_0 = const()[name = string("op_29005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29005_cast_fp16 = einsum(equation = var_29005_equation_0, values = (var_28493_cast_fp16, var_28894_cast_fp16))[name = string("op_29005_cast_fp16")]; + string var_29007_equation_0 = const()[name = string("op_29007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29007_cast_fp16 = einsum(equation = var_29007_equation_0, values = (var_28493_cast_fp16, var_28895_cast_fp16))[name = string("op_29007_cast_fp16")]; + string var_29009_equation_0 = const()[name = string("op_29009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29009_cast_fp16 = einsum(equation = var_29009_equation_0, values = (var_28497_cast_fp16, var_28896_cast_fp16))[name = string("op_29009_cast_fp16")]; + string var_29011_equation_0 = const()[name = string("op_29011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29011_cast_fp16 = einsum(equation = var_29011_equation_0, values = (var_28497_cast_fp16, var_28897_cast_fp16))[name = string("op_29011_cast_fp16")]; + string var_29013_equation_0 = const()[name = string("op_29013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29013_cast_fp16 = einsum(equation = var_29013_equation_0, values = (var_28497_cast_fp16, var_28898_cast_fp16))[name = string("op_29013_cast_fp16")]; + string var_29015_equation_0 = const()[name = string("op_29015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29015_cast_fp16 = einsum(equation = var_29015_equation_0, values = (var_28497_cast_fp16, var_28899_cast_fp16))[name = string("op_29015_cast_fp16")]; + string var_29017_equation_0 = const()[name = string("op_29017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29017_cast_fp16 = einsum(equation = var_29017_equation_0, values = (var_28501_cast_fp16, var_28900_cast_fp16))[name = string("op_29017_cast_fp16")]; + string var_29019_equation_0 = const()[name = string("op_29019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29019_cast_fp16 = einsum(equation = var_29019_equation_0, values = (var_28501_cast_fp16, var_28901_cast_fp16))[name = string("op_29019_cast_fp16")]; + string var_29021_equation_0 = const()[name = string("op_29021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29021_cast_fp16 = einsum(equation = var_29021_equation_0, values = (var_28501_cast_fp16, var_28902_cast_fp16))[name = string("op_29021_cast_fp16")]; + string var_29023_equation_0 = const()[name = string("op_29023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29023_cast_fp16 = einsum(equation = var_29023_equation_0, values = (var_28501_cast_fp16, var_28903_cast_fp16))[name = string("op_29023_cast_fp16")]; + string var_29025_equation_0 = const()[name = string("op_29025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29025_cast_fp16 = einsum(equation = var_29025_equation_0, values = (var_28505_cast_fp16, var_28904_cast_fp16))[name = string("op_29025_cast_fp16")]; + string var_29027_equation_0 = const()[name = string("op_29027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29027_cast_fp16 = einsum(equation = var_29027_equation_0, values = (var_28505_cast_fp16, var_28905_cast_fp16))[name = string("op_29027_cast_fp16")]; + string var_29029_equation_0 = const()[name = string("op_29029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29029_cast_fp16 = einsum(equation = var_29029_equation_0, values = (var_28505_cast_fp16, var_28906_cast_fp16))[name = string("op_29029_cast_fp16")]; + string var_29031_equation_0 = const()[name = string("op_29031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29031_cast_fp16 = einsum(equation = var_29031_equation_0, values = (var_28505_cast_fp16, var_28907_cast_fp16))[name = string("op_29031_cast_fp16")]; + string var_29033_equation_0 = const()[name = string("op_29033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29033_cast_fp16 = einsum(equation = var_29033_equation_0, values = (var_28509_cast_fp16, var_28908_cast_fp16))[name = string("op_29033_cast_fp16")]; + string var_29035_equation_0 = const()[name = string("op_29035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29035_cast_fp16 = einsum(equation = var_29035_equation_0, values = (var_28509_cast_fp16, var_28909_cast_fp16))[name = string("op_29035_cast_fp16")]; + string var_29037_equation_0 = const()[name = string("op_29037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29037_cast_fp16 = einsum(equation = var_29037_equation_0, values = (var_28509_cast_fp16, var_28910_cast_fp16))[name = string("op_29037_cast_fp16")]; + string var_29039_equation_0 = const()[name = string("op_29039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29039_cast_fp16 = einsum(equation = var_29039_equation_0, values = (var_28509_cast_fp16, var_28911_cast_fp16))[name = string("op_29039_cast_fp16")]; + string var_29041_equation_0 = const()[name = string("op_29041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29041_cast_fp16 = einsum(equation = var_29041_equation_0, values = (var_28513_cast_fp16, var_28912_cast_fp16))[name = string("op_29041_cast_fp16")]; + string var_29043_equation_0 = const()[name = string("op_29043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29043_cast_fp16 = einsum(equation = var_29043_equation_0, values = (var_28513_cast_fp16, var_28913_cast_fp16))[name = string("op_29043_cast_fp16")]; + string var_29045_equation_0 = const()[name = string("op_29045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29045_cast_fp16 = einsum(equation = var_29045_equation_0, values = (var_28513_cast_fp16, var_28914_cast_fp16))[name = string("op_29045_cast_fp16")]; + string var_29047_equation_0 = const()[name = string("op_29047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29047_cast_fp16 = einsum(equation = var_29047_equation_0, values = (var_28513_cast_fp16, var_28915_cast_fp16))[name = string("op_29047_cast_fp16")]; + string var_29049_equation_0 = const()[name = string("op_29049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29049_cast_fp16 = einsum(equation = var_29049_equation_0, values = (var_28517_cast_fp16, var_28916_cast_fp16))[name = string("op_29049_cast_fp16")]; + string var_29051_equation_0 = const()[name = string("op_29051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29051_cast_fp16 = einsum(equation = var_29051_equation_0, values = (var_28517_cast_fp16, var_28917_cast_fp16))[name = string("op_29051_cast_fp16")]; + string var_29053_equation_0 = const()[name = string("op_29053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29053_cast_fp16 = einsum(equation = var_29053_equation_0, values = (var_28517_cast_fp16, var_28918_cast_fp16))[name = string("op_29053_cast_fp16")]; + string var_29055_equation_0 = const()[name = string("op_29055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29055_cast_fp16 = einsum(equation = var_29055_equation_0, values = (var_28517_cast_fp16, var_28919_cast_fp16))[name = string("op_29055_cast_fp16")]; + string var_29057_equation_0 = const()[name = string("op_29057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29057_cast_fp16 = einsum(equation = var_29057_equation_0, values = (var_28521_cast_fp16, var_28920_cast_fp16))[name = string("op_29057_cast_fp16")]; + string var_29059_equation_0 = const()[name = string("op_29059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29059_cast_fp16 = einsum(equation = var_29059_equation_0, values = (var_28521_cast_fp16, var_28921_cast_fp16))[name = string("op_29059_cast_fp16")]; + string var_29061_equation_0 = const()[name = string("op_29061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29061_cast_fp16 = einsum(equation = var_29061_equation_0, values = (var_28521_cast_fp16, var_28922_cast_fp16))[name = string("op_29061_cast_fp16")]; + string var_29063_equation_0 = const()[name = string("op_29063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29063_cast_fp16 = einsum(equation = var_29063_equation_0, values = (var_28521_cast_fp16, var_28923_cast_fp16))[name = string("op_29063_cast_fp16")]; + string var_29065_equation_0 = const()[name = string("op_29065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29065_cast_fp16 = einsum(equation = var_29065_equation_0, values = (var_28525_cast_fp16, var_28924_cast_fp16))[name = string("op_29065_cast_fp16")]; + string var_29067_equation_0 = const()[name = string("op_29067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29067_cast_fp16 = einsum(equation = var_29067_equation_0, values = (var_28525_cast_fp16, var_28925_cast_fp16))[name = string("op_29067_cast_fp16")]; + string var_29069_equation_0 = const()[name = string("op_29069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29069_cast_fp16 = einsum(equation = var_29069_equation_0, values = (var_28525_cast_fp16, var_28926_cast_fp16))[name = string("op_29069_cast_fp16")]; + string var_29071_equation_0 = const()[name = string("op_29071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29071_cast_fp16 = einsum(equation = var_29071_equation_0, values = (var_28525_cast_fp16, var_28927_cast_fp16))[name = string("op_29071_cast_fp16")]; + string var_29073_equation_0 = const()[name = string("op_29073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29073_cast_fp16 = einsum(equation = var_29073_equation_0, values = (var_28529_cast_fp16, var_28928_cast_fp16))[name = string("op_29073_cast_fp16")]; + string var_29075_equation_0 = const()[name = string("op_29075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29075_cast_fp16 = einsum(equation = var_29075_equation_0, values = (var_28529_cast_fp16, var_28929_cast_fp16))[name = string("op_29075_cast_fp16")]; + string var_29077_equation_0 = const()[name = string("op_29077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29077_cast_fp16 = einsum(equation = var_29077_equation_0, values = (var_28529_cast_fp16, var_28930_cast_fp16))[name = string("op_29077_cast_fp16")]; + string var_29079_equation_0 = const()[name = string("op_29079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29079_cast_fp16 = einsum(equation = var_29079_equation_0, values = (var_28529_cast_fp16, var_28931_cast_fp16))[name = string("op_29079_cast_fp16")]; + string var_29081_equation_0 = const()[name = string("op_29081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29081_cast_fp16 = einsum(equation = var_29081_equation_0, values = (var_28533_cast_fp16, var_28932_cast_fp16))[name = string("op_29081_cast_fp16")]; + string var_29083_equation_0 = const()[name = string("op_29083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29083_cast_fp16 = einsum(equation = var_29083_equation_0, values = (var_28533_cast_fp16, var_28933_cast_fp16))[name = string("op_29083_cast_fp16")]; + string var_29085_equation_0 = const()[name = string("op_29085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29085_cast_fp16 = einsum(equation = var_29085_equation_0, values = (var_28533_cast_fp16, var_28934_cast_fp16))[name = string("op_29085_cast_fp16")]; + string var_29087_equation_0 = const()[name = string("op_29087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29087_cast_fp16 = einsum(equation = var_29087_equation_0, values = (var_28533_cast_fp16, var_28935_cast_fp16))[name = string("op_29087_cast_fp16")]; + string var_29089_equation_0 = const()[name = string("op_29089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29089_cast_fp16 = einsum(equation = var_29089_equation_0, values = (var_28537_cast_fp16, var_28936_cast_fp16))[name = string("op_29089_cast_fp16")]; + string var_29091_equation_0 = const()[name = string("op_29091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29091_cast_fp16 = einsum(equation = var_29091_equation_0, values = (var_28537_cast_fp16, var_28937_cast_fp16))[name = string("op_29091_cast_fp16")]; + string var_29093_equation_0 = const()[name = string("op_29093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29093_cast_fp16 = einsum(equation = var_29093_equation_0, values = (var_28537_cast_fp16, var_28938_cast_fp16))[name = string("op_29093_cast_fp16")]; + string var_29095_equation_0 = const()[name = string("op_29095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29095_cast_fp16 = einsum(equation = var_29095_equation_0, values = (var_28537_cast_fp16, var_28939_cast_fp16))[name = string("op_29095_cast_fp16")]; + string var_29097_equation_0 = const()[name = string("op_29097_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29097_cast_fp16 = einsum(equation = var_29097_equation_0, values = (var_28541_cast_fp16, var_28940_cast_fp16))[name = string("op_29097_cast_fp16")]; + string var_29099_equation_0 = const()[name = string("op_29099_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29099_cast_fp16 = einsum(equation = var_29099_equation_0, values = (var_28541_cast_fp16, var_28941_cast_fp16))[name = string("op_29099_cast_fp16")]; + string var_29101_equation_0 = const()[name = string("op_29101_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29101_cast_fp16 = einsum(equation = var_29101_equation_0, values = (var_28541_cast_fp16, var_28942_cast_fp16))[name = string("op_29101_cast_fp16")]; + string var_29103_equation_0 = const()[name = string("op_29103_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29103_cast_fp16 = einsum(equation = var_29103_equation_0, values = (var_28541_cast_fp16, var_28943_cast_fp16))[name = string("op_29103_cast_fp16")]; + bool var_29105_interleave_0 = const()[name = string("op_29105_interleave_0"), val = bool(false)]; + tensor var_29105_cast_fp16 = concat(axis = var_27664, interleave = var_29105_interleave_0, values = (var_28945_cast_fp16, var_28947_cast_fp16, var_28949_cast_fp16, var_28951_cast_fp16))[name = string("op_29105_cast_fp16")]; + bool var_29107_interleave_0 = const()[name = string("op_29107_interleave_0"), val = bool(false)]; + tensor var_29107_cast_fp16 = concat(axis = var_27664, interleave = var_29107_interleave_0, values = (var_28953_cast_fp16, var_28955_cast_fp16, var_28957_cast_fp16, var_28959_cast_fp16))[name = string("op_29107_cast_fp16")]; + bool var_29109_interleave_0 = const()[name = string("op_29109_interleave_0"), val = bool(false)]; + tensor var_29109_cast_fp16 = concat(axis = var_27664, interleave = var_29109_interleave_0, values = (var_28961_cast_fp16, var_28963_cast_fp16, var_28965_cast_fp16, var_28967_cast_fp16))[name = string("op_29109_cast_fp16")]; + bool var_29111_interleave_0 = const()[name = string("op_29111_interleave_0"), val = bool(false)]; + tensor var_29111_cast_fp16 = concat(axis = var_27664, interleave = var_29111_interleave_0, values = (var_28969_cast_fp16, var_28971_cast_fp16, var_28973_cast_fp16, var_28975_cast_fp16))[name = string("op_29111_cast_fp16")]; + bool var_29113_interleave_0 = const()[name = string("op_29113_interleave_0"), val = bool(false)]; + tensor var_29113_cast_fp16 = concat(axis = var_27664, interleave = var_29113_interleave_0, values = (var_28977_cast_fp16, var_28979_cast_fp16, var_28981_cast_fp16, var_28983_cast_fp16))[name = string("op_29113_cast_fp16")]; + bool var_29115_interleave_0 = const()[name = string("op_29115_interleave_0"), val = bool(false)]; + tensor var_29115_cast_fp16 = concat(axis = var_27664, interleave = var_29115_interleave_0, values = (var_28985_cast_fp16, var_28987_cast_fp16, var_28989_cast_fp16, var_28991_cast_fp16))[name = string("op_29115_cast_fp16")]; + bool var_29117_interleave_0 = const()[name = string("op_29117_interleave_0"), val = bool(false)]; + tensor var_29117_cast_fp16 = concat(axis = var_27664, interleave = var_29117_interleave_0, values = (var_28993_cast_fp16, var_28995_cast_fp16, var_28997_cast_fp16, var_28999_cast_fp16))[name = string("op_29117_cast_fp16")]; + bool var_29119_interleave_0 = const()[name = string("op_29119_interleave_0"), val = bool(false)]; + tensor var_29119_cast_fp16 = concat(axis = var_27664, interleave = var_29119_interleave_0, values = (var_29001_cast_fp16, var_29003_cast_fp16, var_29005_cast_fp16, var_29007_cast_fp16))[name = string("op_29119_cast_fp16")]; + bool var_29121_interleave_0 = const()[name = string("op_29121_interleave_0"), val = bool(false)]; + tensor var_29121_cast_fp16 = concat(axis = var_27664, interleave = var_29121_interleave_0, values = (var_29009_cast_fp16, var_29011_cast_fp16, var_29013_cast_fp16, var_29015_cast_fp16))[name = string("op_29121_cast_fp16")]; + bool var_29123_interleave_0 = const()[name = string("op_29123_interleave_0"), val = bool(false)]; + tensor var_29123_cast_fp16 = concat(axis = var_27664, interleave = var_29123_interleave_0, values = (var_29017_cast_fp16, var_29019_cast_fp16, var_29021_cast_fp16, var_29023_cast_fp16))[name = string("op_29123_cast_fp16")]; + bool var_29125_interleave_0 = const()[name = string("op_29125_interleave_0"), val = bool(false)]; + tensor var_29125_cast_fp16 = concat(axis = var_27664, interleave = var_29125_interleave_0, values = (var_29025_cast_fp16, var_29027_cast_fp16, var_29029_cast_fp16, var_29031_cast_fp16))[name = string("op_29125_cast_fp16")]; + bool var_29127_interleave_0 = const()[name = string("op_29127_interleave_0"), val = bool(false)]; + tensor var_29127_cast_fp16 = concat(axis = var_27664, interleave = var_29127_interleave_0, values = (var_29033_cast_fp16, var_29035_cast_fp16, var_29037_cast_fp16, var_29039_cast_fp16))[name = string("op_29127_cast_fp16")]; + bool var_29129_interleave_0 = const()[name = string("op_29129_interleave_0"), val = bool(false)]; + tensor var_29129_cast_fp16 = concat(axis = var_27664, interleave = var_29129_interleave_0, values = (var_29041_cast_fp16, var_29043_cast_fp16, var_29045_cast_fp16, var_29047_cast_fp16))[name = string("op_29129_cast_fp16")]; + bool var_29131_interleave_0 = const()[name = string("op_29131_interleave_0"), val = bool(false)]; + tensor var_29131_cast_fp16 = concat(axis = var_27664, interleave = var_29131_interleave_0, values = (var_29049_cast_fp16, var_29051_cast_fp16, var_29053_cast_fp16, var_29055_cast_fp16))[name = string("op_29131_cast_fp16")]; + bool var_29133_interleave_0 = const()[name = string("op_29133_interleave_0"), val = bool(false)]; + tensor var_29133_cast_fp16 = concat(axis = var_27664, interleave = var_29133_interleave_0, values = (var_29057_cast_fp16, var_29059_cast_fp16, var_29061_cast_fp16, var_29063_cast_fp16))[name = string("op_29133_cast_fp16")]; + bool var_29135_interleave_0 = const()[name = string("op_29135_interleave_0"), val = bool(false)]; + tensor var_29135_cast_fp16 = concat(axis = var_27664, interleave = var_29135_interleave_0, values = (var_29065_cast_fp16, var_29067_cast_fp16, var_29069_cast_fp16, var_29071_cast_fp16))[name = string("op_29135_cast_fp16")]; + bool var_29137_interleave_0 = const()[name = string("op_29137_interleave_0"), val = bool(false)]; + tensor var_29137_cast_fp16 = concat(axis = var_27664, interleave = var_29137_interleave_0, values = (var_29073_cast_fp16, var_29075_cast_fp16, var_29077_cast_fp16, var_29079_cast_fp16))[name = string("op_29137_cast_fp16")]; + bool var_29139_interleave_0 = const()[name = string("op_29139_interleave_0"), val = bool(false)]; + tensor var_29139_cast_fp16 = concat(axis = var_27664, interleave = var_29139_interleave_0, values = (var_29081_cast_fp16, var_29083_cast_fp16, var_29085_cast_fp16, var_29087_cast_fp16))[name = string("op_29139_cast_fp16")]; + bool var_29141_interleave_0 = const()[name = string("op_29141_interleave_0"), val = bool(false)]; + tensor var_29141_cast_fp16 = concat(axis = var_27664, interleave = var_29141_interleave_0, values = (var_29089_cast_fp16, var_29091_cast_fp16, var_29093_cast_fp16, var_29095_cast_fp16))[name = string("op_29141_cast_fp16")]; + bool var_29143_interleave_0 = const()[name = string("op_29143_interleave_0"), val = bool(false)]; + tensor var_29143_cast_fp16 = concat(axis = var_27664, interleave = var_29143_interleave_0, values = (var_29097_cast_fp16, var_29099_cast_fp16, var_29101_cast_fp16, var_29103_cast_fp16))[name = string("op_29143_cast_fp16")]; + bool input_145_interleave_0 = const()[name = string("input_145_interleave_0"), val = bool(false)]; + tensor input_145_cast_fp16 = concat(axis = var_27689, interleave = input_145_interleave_0, values = (var_29105_cast_fp16, var_29107_cast_fp16, var_29109_cast_fp16, var_29111_cast_fp16, var_29113_cast_fp16, var_29115_cast_fp16, var_29117_cast_fp16, var_29119_cast_fp16, var_29121_cast_fp16, var_29123_cast_fp16, var_29125_cast_fp16, var_29127_cast_fp16, var_29129_cast_fp16, var_29131_cast_fp16, var_29133_cast_fp16, var_29135_cast_fp16, var_29137_cast_fp16, var_29139_cast_fp16, var_29141_cast_fp16, var_29143_cast_fp16))[name = string("input_145_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732864320)))]; + tensor layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736141184)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_75_cast_fp16")]; + tensor out_75_axes_0 = const()[name = string("out_75_axes_0"), val = tensor([1])]; + fp16 var_29162_to_fp16 = const()[name = string("op_29162_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_29162_to_fp16, x = inputs_75_cast_fp16)[name = string("out_75_cast_fp16")]; + tensor input_147_gamma_0_to_fp16 = const()[name = string("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736143808)))]; + tensor input_147_beta_0_to_fp16 = const()[name = string("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736146432)))]; + fp16 input_147_epsilon_0_to_fp16 = const()[name = string("input_147_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = string("input_147_cast_fp16")]; + string input_149_pad_type_0 = const()[name = string("input_149_pad_type_0"), val = string("valid")]; + tensor input_149_strides_0 = const()[name = string("input_149_strides_0"), val = tensor([1, 1])]; + tensor input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_149_dilations_0 = const()[name = string("input_149_dilations_0"), val = tensor([1, 1])]; + int32 input_149_groups_0 = const()[name = string("input_149_groups_0"), val = int32(1)]; + tensor layers_18_fc1_weight_to_fp16 = const()[name = string("layers_18_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736149056)))]; + tensor layers_18_fc1_bias_to_fp16 = const()[name = string("layers_18_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749256320)))]; + tensor input_149_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_149_dilations_0, groups = input_149_groups_0, pad = input_149_pad_0, pad_type = input_149_pad_type_0, strides = input_149_strides_0, weight = layers_18_fc1_weight_to_fp16, x = input_147_cast_fp16)[name = string("input_149_cast_fp16")]; + string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("EXACT")]; + tensor input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")]; + string hidden_states_41_pad_type_0 = const()[name = string("hidden_states_41_pad_type_0"), val = string("valid")]; + tensor hidden_states_41_strides_0 = const()[name = string("hidden_states_41_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_41_pad_0 = const()[name = string("hidden_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_41_dilations_0 = const()[name = string("hidden_states_41_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_41_groups_0 = const()[name = string("hidden_states_41_groups_0"), val = int32(1)]; + tensor layers_18_fc2_weight_to_fp16 = const()[name = string("layers_18_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749266624)))]; + tensor layers_18_fc2_bias_to_fp16 = const()[name = string("layers_18_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762373888)))]; + tensor hidden_states_41_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_18_fc2_weight_to_fp16, x = input_151_cast_fp16)[name = string("hidden_states_41_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = hidden_states_41_cast_fp16)[name = string("inputs_77_cast_fp16")]; + int32 var_29191 = const()[name = string("op_29191"), val = int32(3)]; + int32 var_29216 = const()[name = string("op_29216"), val = int32(1)]; + tensor out_77_axes_0 = const()[name = string("out_77_axes_0"), val = tensor([1])]; + fp16 var_29233_to_fp16 = const()[name = string("op_29233_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_29233_to_fp16, x = inputs_77_cast_fp16)[name = string("out_77_cast_fp16")]; + tensor obj_77_gamma_0_to_fp16 = const()[name = string("obj_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762376512)))]; + tensor obj_77_beta_0_to_fp16 = const()[name = string("obj_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762379136)))]; + fp16 obj_77_epsilon_0_to_fp16 = const()[name = string("obj_77_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_77_cast_fp16 = batch_norm(beta = obj_77_beta_0_to_fp16, epsilon = obj_77_epsilon_0_to_fp16, gamma = obj_77_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = string("obj_77_cast_fp16")]; + string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; + tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; + tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; + int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762381760)))]; + tensor layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765658624)))]; + tensor query_39_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("query_39_cast_fp16")]; + string key_39_pad_type_0 = const()[name = string("key_39_pad_type_0"), val = string("valid")]; + tensor key_39_strides_0 = const()[name = string("key_39_strides_0"), val = tensor([1, 1])]; + tensor key_39_pad_0 = const()[name = string("key_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_39_dilations_0 = const()[name = string("key_39_dilations_0"), val = tensor([1, 1])]; + int32 key_39_groups_0 = const()[name = string("key_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765661248)))]; + tensor key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("key_39_cast_fp16")]; + string value_39_pad_type_0 = const()[name = string("value_39_pad_type_0"), val = string("valid")]; + tensor value_39_strides_0 = const()[name = string("value_39_strides_0"), val = tensor([1, 1])]; + tensor value_39_pad_0 = const()[name = string("value_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_39_dilations_0 = const()[name = string("value_39_dilations_0"), val = tensor([1, 1])]; + int32 value_39_groups_0 = const()[name = string("value_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(768938112)))]; + tensor layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772214976)))]; + tensor value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("value_39_cast_fp16")]; + tensor var_29271_begin_0 = const()[name = string("op_29271_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29271_end_0 = const()[name = string("op_29271_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29271_end_mask_0 = const()[name = string("op_29271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29271_cast_fp16 = slice_by_index(begin = var_29271_begin_0, end = var_29271_end_0, end_mask = var_29271_end_mask_0, x = query_39_cast_fp16)[name = string("op_29271_cast_fp16")]; + tensor var_29275_begin_0 = const()[name = string("op_29275_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29275_end_0 = const()[name = string("op_29275_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_29275_end_mask_0 = const()[name = string("op_29275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29275_cast_fp16 = slice_by_index(begin = var_29275_begin_0, end = var_29275_end_0, end_mask = var_29275_end_mask_0, x = query_39_cast_fp16)[name = string("op_29275_cast_fp16")]; + tensor var_29279_begin_0 = const()[name = string("op_29279_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29279_end_0 = const()[name = string("op_29279_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_29279_end_mask_0 = const()[name = string("op_29279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29279_cast_fp16 = slice_by_index(begin = var_29279_begin_0, end = var_29279_end_0, end_mask = var_29279_end_mask_0, x = query_39_cast_fp16)[name = string("op_29279_cast_fp16")]; + tensor var_29283_begin_0 = const()[name = string("op_29283_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29283_end_0 = const()[name = string("op_29283_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_29283_end_mask_0 = const()[name = string("op_29283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29283_cast_fp16 = slice_by_index(begin = var_29283_begin_0, end = var_29283_end_0, end_mask = var_29283_end_mask_0, x = query_39_cast_fp16)[name = string("op_29283_cast_fp16")]; + tensor var_29287_begin_0 = const()[name = string("op_29287_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29287_end_0 = const()[name = string("op_29287_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_29287_end_mask_0 = const()[name = string("op_29287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29287_cast_fp16 = slice_by_index(begin = var_29287_begin_0, end = var_29287_end_0, end_mask = var_29287_end_mask_0, x = query_39_cast_fp16)[name = string("op_29287_cast_fp16")]; + tensor var_29291_begin_0 = const()[name = string("op_29291_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29291_end_0 = const()[name = string("op_29291_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_29291_end_mask_0 = const()[name = string("op_29291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29291_cast_fp16 = slice_by_index(begin = var_29291_begin_0, end = var_29291_end_0, end_mask = var_29291_end_mask_0, x = query_39_cast_fp16)[name = string("op_29291_cast_fp16")]; + tensor var_29295_begin_0 = const()[name = string("op_29295_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29295_end_0 = const()[name = string("op_29295_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_29295_end_mask_0 = const()[name = string("op_29295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29295_cast_fp16 = slice_by_index(begin = var_29295_begin_0, end = var_29295_end_0, end_mask = var_29295_end_mask_0, x = query_39_cast_fp16)[name = string("op_29295_cast_fp16")]; + tensor var_29299_begin_0 = const()[name = string("op_29299_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29299_end_0 = const()[name = string("op_29299_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_29299_end_mask_0 = const()[name = string("op_29299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29299_cast_fp16 = slice_by_index(begin = var_29299_begin_0, end = var_29299_end_0, end_mask = var_29299_end_mask_0, x = query_39_cast_fp16)[name = string("op_29299_cast_fp16")]; + tensor var_29303_begin_0 = const()[name = string("op_29303_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29303_end_0 = const()[name = string("op_29303_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_29303_end_mask_0 = const()[name = string("op_29303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29303_cast_fp16 = slice_by_index(begin = var_29303_begin_0, end = var_29303_end_0, end_mask = var_29303_end_mask_0, x = query_39_cast_fp16)[name = string("op_29303_cast_fp16")]; + tensor var_29307_begin_0 = const()[name = string("op_29307_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29307_end_0 = const()[name = string("op_29307_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_29307_end_mask_0 = const()[name = string("op_29307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29307_cast_fp16 = slice_by_index(begin = var_29307_begin_0, end = var_29307_end_0, end_mask = var_29307_end_mask_0, x = query_39_cast_fp16)[name = string("op_29307_cast_fp16")]; + tensor var_29311_begin_0 = const()[name = string("op_29311_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29311_end_0 = const()[name = string("op_29311_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_29311_end_mask_0 = const()[name = string("op_29311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29311_cast_fp16 = slice_by_index(begin = var_29311_begin_0, end = var_29311_end_0, end_mask = var_29311_end_mask_0, x = query_39_cast_fp16)[name = string("op_29311_cast_fp16")]; + tensor var_29315_begin_0 = const()[name = string("op_29315_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29315_end_0 = const()[name = string("op_29315_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_29315_end_mask_0 = const()[name = string("op_29315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29315_cast_fp16 = slice_by_index(begin = var_29315_begin_0, end = var_29315_end_0, end_mask = var_29315_end_mask_0, x = query_39_cast_fp16)[name = string("op_29315_cast_fp16")]; + tensor var_29319_begin_0 = const()[name = string("op_29319_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29319_end_0 = const()[name = string("op_29319_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_29319_end_mask_0 = const()[name = string("op_29319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29319_cast_fp16 = slice_by_index(begin = var_29319_begin_0, end = var_29319_end_0, end_mask = var_29319_end_mask_0, x = query_39_cast_fp16)[name = string("op_29319_cast_fp16")]; + tensor var_29323_begin_0 = const()[name = string("op_29323_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29323_end_0 = const()[name = string("op_29323_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_29323_end_mask_0 = const()[name = string("op_29323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29323_cast_fp16 = slice_by_index(begin = var_29323_begin_0, end = var_29323_end_0, end_mask = var_29323_end_mask_0, x = query_39_cast_fp16)[name = string("op_29323_cast_fp16")]; + tensor var_29327_begin_0 = const()[name = string("op_29327_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29327_end_0 = const()[name = string("op_29327_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_29327_end_mask_0 = const()[name = string("op_29327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29327_cast_fp16 = slice_by_index(begin = var_29327_begin_0, end = var_29327_end_0, end_mask = var_29327_end_mask_0, x = query_39_cast_fp16)[name = string("op_29327_cast_fp16")]; + tensor var_29331_begin_0 = const()[name = string("op_29331_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29331_end_0 = const()[name = string("op_29331_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_29331_end_mask_0 = const()[name = string("op_29331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29331_cast_fp16 = slice_by_index(begin = var_29331_begin_0, end = var_29331_end_0, end_mask = var_29331_end_mask_0, x = query_39_cast_fp16)[name = string("op_29331_cast_fp16")]; + tensor var_29335_begin_0 = const()[name = string("op_29335_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29335_end_0 = const()[name = string("op_29335_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_29335_end_mask_0 = const()[name = string("op_29335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29335_cast_fp16 = slice_by_index(begin = var_29335_begin_0, end = var_29335_end_0, end_mask = var_29335_end_mask_0, x = query_39_cast_fp16)[name = string("op_29335_cast_fp16")]; + tensor var_29339_begin_0 = const()[name = string("op_29339_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29339_end_0 = const()[name = string("op_29339_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_29339_end_mask_0 = const()[name = string("op_29339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29339_cast_fp16 = slice_by_index(begin = var_29339_begin_0, end = var_29339_end_0, end_mask = var_29339_end_mask_0, x = query_39_cast_fp16)[name = string("op_29339_cast_fp16")]; + tensor var_29343_begin_0 = const()[name = string("op_29343_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29343_end_0 = const()[name = string("op_29343_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_29343_end_mask_0 = const()[name = string("op_29343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29343_cast_fp16 = slice_by_index(begin = var_29343_begin_0, end = var_29343_end_0, end_mask = var_29343_end_mask_0, x = query_39_cast_fp16)[name = string("op_29343_cast_fp16")]; + tensor var_29347_begin_0 = const()[name = string("op_29347_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29347_end_0 = const()[name = string("op_29347_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_29347_end_mask_0 = const()[name = string("op_29347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29347_cast_fp16 = slice_by_index(begin = var_29347_begin_0, end = var_29347_end_0, end_mask = var_29347_end_mask_0, x = query_39_cast_fp16)[name = string("op_29347_cast_fp16")]; + tensor var_29356_begin_0 = const()[name = string("op_29356_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29356_end_0 = const()[name = string("op_29356_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29356_end_mask_0 = const()[name = string("op_29356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29356_cast_fp16 = slice_by_index(begin = var_29356_begin_0, end = var_29356_end_0, end_mask = var_29356_end_mask_0, x = var_29271_cast_fp16)[name = string("op_29356_cast_fp16")]; + tensor var_29363_begin_0 = const()[name = string("op_29363_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29363_end_0 = const()[name = string("op_29363_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29363_end_mask_0 = const()[name = string("op_29363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29363_cast_fp16 = slice_by_index(begin = var_29363_begin_0, end = var_29363_end_0, end_mask = var_29363_end_mask_0, x = var_29271_cast_fp16)[name = string("op_29363_cast_fp16")]; + tensor var_29370_begin_0 = const()[name = string("op_29370_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29370_end_0 = const()[name = string("op_29370_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29370_end_mask_0 = const()[name = string("op_29370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29370_cast_fp16 = slice_by_index(begin = var_29370_begin_0, end = var_29370_end_0, end_mask = var_29370_end_mask_0, x = var_29271_cast_fp16)[name = string("op_29370_cast_fp16")]; + tensor var_29377_begin_0 = const()[name = string("op_29377_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29377_end_0 = const()[name = string("op_29377_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29377_end_mask_0 = const()[name = string("op_29377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29377_cast_fp16 = slice_by_index(begin = var_29377_begin_0, end = var_29377_end_0, end_mask = var_29377_end_mask_0, x = var_29271_cast_fp16)[name = string("op_29377_cast_fp16")]; + tensor var_29384_begin_0 = const()[name = string("op_29384_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29384_end_0 = const()[name = string("op_29384_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29384_end_mask_0 = const()[name = string("op_29384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29384_cast_fp16 = slice_by_index(begin = var_29384_begin_0, end = var_29384_end_0, end_mask = var_29384_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29384_cast_fp16")]; + tensor var_29391_begin_0 = const()[name = string("op_29391_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29391_end_0 = const()[name = string("op_29391_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29391_end_mask_0 = const()[name = string("op_29391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29391_cast_fp16 = slice_by_index(begin = var_29391_begin_0, end = var_29391_end_0, end_mask = var_29391_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29391_cast_fp16")]; + tensor var_29398_begin_0 = const()[name = string("op_29398_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29398_end_0 = const()[name = string("op_29398_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29398_end_mask_0 = const()[name = string("op_29398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29398_cast_fp16 = slice_by_index(begin = var_29398_begin_0, end = var_29398_end_0, end_mask = var_29398_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29398_cast_fp16")]; + tensor var_29405_begin_0 = const()[name = string("op_29405_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29405_end_0 = const()[name = string("op_29405_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29405_end_mask_0 = const()[name = string("op_29405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29405_cast_fp16 = slice_by_index(begin = var_29405_begin_0, end = var_29405_end_0, end_mask = var_29405_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29405_cast_fp16")]; + tensor var_29412_begin_0 = const()[name = string("op_29412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29412_end_0 = const()[name = string("op_29412_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29412_end_mask_0 = const()[name = string("op_29412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29412_cast_fp16 = slice_by_index(begin = var_29412_begin_0, end = var_29412_end_0, end_mask = var_29412_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29412_cast_fp16")]; + tensor var_29419_begin_0 = const()[name = string("op_29419_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29419_end_0 = const()[name = string("op_29419_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29419_end_mask_0 = const()[name = string("op_29419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29419_cast_fp16 = slice_by_index(begin = var_29419_begin_0, end = var_29419_end_0, end_mask = var_29419_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29419_cast_fp16")]; + tensor var_29426_begin_0 = const()[name = string("op_29426_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29426_end_0 = const()[name = string("op_29426_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29426_end_mask_0 = const()[name = string("op_29426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29426_cast_fp16 = slice_by_index(begin = var_29426_begin_0, end = var_29426_end_0, end_mask = var_29426_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29426_cast_fp16")]; + tensor var_29433_begin_0 = const()[name = string("op_29433_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29433_end_0 = const()[name = string("op_29433_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29433_end_mask_0 = const()[name = string("op_29433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29433_cast_fp16 = slice_by_index(begin = var_29433_begin_0, end = var_29433_end_0, end_mask = var_29433_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29433_cast_fp16")]; + tensor var_29440_begin_0 = const()[name = string("op_29440_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29440_end_0 = const()[name = string("op_29440_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29440_end_mask_0 = const()[name = string("op_29440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29440_cast_fp16 = slice_by_index(begin = var_29440_begin_0, end = var_29440_end_0, end_mask = var_29440_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29440_cast_fp16")]; + tensor var_29447_begin_0 = const()[name = string("op_29447_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29447_end_0 = const()[name = string("op_29447_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29447_end_mask_0 = const()[name = string("op_29447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29447_cast_fp16 = slice_by_index(begin = var_29447_begin_0, end = var_29447_end_0, end_mask = var_29447_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29447_cast_fp16")]; + tensor var_29454_begin_0 = const()[name = string("op_29454_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29454_end_0 = const()[name = string("op_29454_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29454_end_mask_0 = const()[name = string("op_29454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29454_cast_fp16 = slice_by_index(begin = var_29454_begin_0, end = var_29454_end_0, end_mask = var_29454_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29454_cast_fp16")]; + tensor var_29461_begin_0 = const()[name = string("op_29461_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29461_end_0 = const()[name = string("op_29461_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29461_end_mask_0 = const()[name = string("op_29461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29461_cast_fp16 = slice_by_index(begin = var_29461_begin_0, end = var_29461_end_0, end_mask = var_29461_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29461_cast_fp16")]; + tensor var_29468_begin_0 = const()[name = string("op_29468_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29468_end_0 = const()[name = string("op_29468_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29468_end_mask_0 = const()[name = string("op_29468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29468_cast_fp16 = slice_by_index(begin = var_29468_begin_0, end = var_29468_end_0, end_mask = var_29468_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29468_cast_fp16")]; + tensor var_29475_begin_0 = const()[name = string("op_29475_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29475_end_0 = const()[name = string("op_29475_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29475_end_mask_0 = const()[name = string("op_29475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29475_cast_fp16 = slice_by_index(begin = var_29475_begin_0, end = var_29475_end_0, end_mask = var_29475_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29475_cast_fp16")]; + tensor var_29482_begin_0 = const()[name = string("op_29482_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29482_end_0 = const()[name = string("op_29482_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29482_end_mask_0 = const()[name = string("op_29482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29482_cast_fp16 = slice_by_index(begin = var_29482_begin_0, end = var_29482_end_0, end_mask = var_29482_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29482_cast_fp16")]; + tensor var_29489_begin_0 = const()[name = string("op_29489_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29489_end_0 = const()[name = string("op_29489_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29489_end_mask_0 = const()[name = string("op_29489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29489_cast_fp16 = slice_by_index(begin = var_29489_begin_0, end = var_29489_end_0, end_mask = var_29489_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29489_cast_fp16")]; + tensor var_29496_begin_0 = const()[name = string("op_29496_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29496_end_0 = const()[name = string("op_29496_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29496_end_mask_0 = const()[name = string("op_29496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29496_cast_fp16 = slice_by_index(begin = var_29496_begin_0, end = var_29496_end_0, end_mask = var_29496_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29496_cast_fp16")]; + tensor var_29503_begin_0 = const()[name = string("op_29503_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29503_end_0 = const()[name = string("op_29503_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29503_end_mask_0 = const()[name = string("op_29503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29503_cast_fp16 = slice_by_index(begin = var_29503_begin_0, end = var_29503_end_0, end_mask = var_29503_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29503_cast_fp16")]; + tensor var_29510_begin_0 = const()[name = string("op_29510_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29510_end_0 = const()[name = string("op_29510_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29510_end_mask_0 = const()[name = string("op_29510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29510_cast_fp16 = slice_by_index(begin = var_29510_begin_0, end = var_29510_end_0, end_mask = var_29510_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29510_cast_fp16")]; + tensor var_29517_begin_0 = const()[name = string("op_29517_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29517_end_0 = const()[name = string("op_29517_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29517_end_mask_0 = const()[name = string("op_29517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29517_cast_fp16 = slice_by_index(begin = var_29517_begin_0, end = var_29517_end_0, end_mask = var_29517_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29517_cast_fp16")]; + tensor var_29524_begin_0 = const()[name = string("op_29524_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29524_end_0 = const()[name = string("op_29524_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29524_end_mask_0 = const()[name = string("op_29524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29524_cast_fp16 = slice_by_index(begin = var_29524_begin_0, end = var_29524_end_0, end_mask = var_29524_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29524_cast_fp16")]; + tensor var_29531_begin_0 = const()[name = string("op_29531_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29531_end_0 = const()[name = string("op_29531_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29531_end_mask_0 = const()[name = string("op_29531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29531_cast_fp16 = slice_by_index(begin = var_29531_begin_0, end = var_29531_end_0, end_mask = var_29531_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29531_cast_fp16")]; + tensor var_29538_begin_0 = const()[name = string("op_29538_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29538_end_0 = const()[name = string("op_29538_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29538_end_mask_0 = const()[name = string("op_29538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29538_cast_fp16 = slice_by_index(begin = var_29538_begin_0, end = var_29538_end_0, end_mask = var_29538_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29538_cast_fp16")]; + tensor var_29545_begin_0 = const()[name = string("op_29545_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29545_end_0 = const()[name = string("op_29545_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29545_end_mask_0 = const()[name = string("op_29545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29545_cast_fp16 = slice_by_index(begin = var_29545_begin_0, end = var_29545_end_0, end_mask = var_29545_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29545_cast_fp16")]; + tensor var_29552_begin_0 = const()[name = string("op_29552_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29552_end_0 = const()[name = string("op_29552_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29552_end_mask_0 = const()[name = string("op_29552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29552_cast_fp16 = slice_by_index(begin = var_29552_begin_0, end = var_29552_end_0, end_mask = var_29552_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29552_cast_fp16")]; + tensor var_29559_begin_0 = const()[name = string("op_29559_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29559_end_0 = const()[name = string("op_29559_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29559_end_mask_0 = const()[name = string("op_29559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29559_cast_fp16 = slice_by_index(begin = var_29559_begin_0, end = var_29559_end_0, end_mask = var_29559_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29559_cast_fp16")]; + tensor var_29566_begin_0 = const()[name = string("op_29566_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29566_end_0 = const()[name = string("op_29566_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29566_end_mask_0 = const()[name = string("op_29566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29566_cast_fp16 = slice_by_index(begin = var_29566_begin_0, end = var_29566_end_0, end_mask = var_29566_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29566_cast_fp16")]; + tensor var_29573_begin_0 = const()[name = string("op_29573_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29573_end_0 = const()[name = string("op_29573_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29573_end_mask_0 = const()[name = string("op_29573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29573_cast_fp16 = slice_by_index(begin = var_29573_begin_0, end = var_29573_end_0, end_mask = var_29573_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29573_cast_fp16")]; + tensor var_29580_begin_0 = const()[name = string("op_29580_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29580_end_0 = const()[name = string("op_29580_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29580_end_mask_0 = const()[name = string("op_29580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29580_cast_fp16 = slice_by_index(begin = var_29580_begin_0, end = var_29580_end_0, end_mask = var_29580_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29580_cast_fp16")]; + tensor var_29587_begin_0 = const()[name = string("op_29587_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29587_end_0 = const()[name = string("op_29587_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29587_end_mask_0 = const()[name = string("op_29587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29587_cast_fp16 = slice_by_index(begin = var_29587_begin_0, end = var_29587_end_0, end_mask = var_29587_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29587_cast_fp16")]; + tensor var_29594_begin_0 = const()[name = string("op_29594_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29594_end_0 = const()[name = string("op_29594_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29594_end_mask_0 = const()[name = string("op_29594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29594_cast_fp16 = slice_by_index(begin = var_29594_begin_0, end = var_29594_end_0, end_mask = var_29594_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29594_cast_fp16")]; + tensor var_29601_begin_0 = const()[name = string("op_29601_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29601_end_0 = const()[name = string("op_29601_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29601_end_mask_0 = const()[name = string("op_29601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29601_cast_fp16 = slice_by_index(begin = var_29601_begin_0, end = var_29601_end_0, end_mask = var_29601_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29601_cast_fp16")]; + tensor var_29608_begin_0 = const()[name = string("op_29608_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29608_end_0 = const()[name = string("op_29608_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29608_end_mask_0 = const()[name = string("op_29608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29608_cast_fp16 = slice_by_index(begin = var_29608_begin_0, end = var_29608_end_0, end_mask = var_29608_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29608_cast_fp16")]; + tensor var_29615_begin_0 = const()[name = string("op_29615_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29615_end_0 = const()[name = string("op_29615_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29615_end_mask_0 = const()[name = string("op_29615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29615_cast_fp16 = slice_by_index(begin = var_29615_begin_0, end = var_29615_end_0, end_mask = var_29615_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29615_cast_fp16")]; + tensor var_29622_begin_0 = const()[name = string("op_29622_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29622_end_0 = const()[name = string("op_29622_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29622_end_mask_0 = const()[name = string("op_29622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29622_cast_fp16 = slice_by_index(begin = var_29622_begin_0, end = var_29622_end_0, end_mask = var_29622_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29622_cast_fp16")]; + tensor var_29629_begin_0 = const()[name = string("op_29629_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29629_end_0 = const()[name = string("op_29629_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29629_end_mask_0 = const()[name = string("op_29629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29629_cast_fp16 = slice_by_index(begin = var_29629_begin_0, end = var_29629_end_0, end_mask = var_29629_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29629_cast_fp16")]; + tensor var_29636_begin_0 = const()[name = string("op_29636_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29636_end_0 = const()[name = string("op_29636_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29636_end_mask_0 = const()[name = string("op_29636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29636_cast_fp16 = slice_by_index(begin = var_29636_begin_0, end = var_29636_end_0, end_mask = var_29636_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29636_cast_fp16")]; + tensor var_29643_begin_0 = const()[name = string("op_29643_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29643_end_0 = const()[name = string("op_29643_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29643_end_mask_0 = const()[name = string("op_29643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29643_cast_fp16 = slice_by_index(begin = var_29643_begin_0, end = var_29643_end_0, end_mask = var_29643_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29643_cast_fp16")]; + tensor var_29650_begin_0 = const()[name = string("op_29650_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29650_end_0 = const()[name = string("op_29650_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29650_end_mask_0 = const()[name = string("op_29650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29650_cast_fp16 = slice_by_index(begin = var_29650_begin_0, end = var_29650_end_0, end_mask = var_29650_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29650_cast_fp16")]; + tensor var_29657_begin_0 = const()[name = string("op_29657_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29657_end_0 = const()[name = string("op_29657_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29657_end_mask_0 = const()[name = string("op_29657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29657_cast_fp16 = slice_by_index(begin = var_29657_begin_0, end = var_29657_end_0, end_mask = var_29657_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29657_cast_fp16")]; + tensor var_29664_begin_0 = const()[name = string("op_29664_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29664_end_0 = const()[name = string("op_29664_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29664_end_mask_0 = const()[name = string("op_29664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29664_cast_fp16 = slice_by_index(begin = var_29664_begin_0, end = var_29664_end_0, end_mask = var_29664_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29664_cast_fp16")]; + tensor var_29671_begin_0 = const()[name = string("op_29671_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29671_end_0 = const()[name = string("op_29671_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29671_end_mask_0 = const()[name = string("op_29671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29671_cast_fp16 = slice_by_index(begin = var_29671_begin_0, end = var_29671_end_0, end_mask = var_29671_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29671_cast_fp16")]; + tensor var_29678_begin_0 = const()[name = string("op_29678_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29678_end_0 = const()[name = string("op_29678_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29678_end_mask_0 = const()[name = string("op_29678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29678_cast_fp16 = slice_by_index(begin = var_29678_begin_0, end = var_29678_end_0, end_mask = var_29678_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29678_cast_fp16")]; + tensor var_29685_begin_0 = const()[name = string("op_29685_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29685_end_0 = const()[name = string("op_29685_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29685_end_mask_0 = const()[name = string("op_29685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29685_cast_fp16 = slice_by_index(begin = var_29685_begin_0, end = var_29685_end_0, end_mask = var_29685_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29685_cast_fp16")]; + tensor var_29692_begin_0 = const()[name = string("op_29692_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29692_end_0 = const()[name = string("op_29692_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29692_end_mask_0 = const()[name = string("op_29692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29692_cast_fp16 = slice_by_index(begin = var_29692_begin_0, end = var_29692_end_0, end_mask = var_29692_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29692_cast_fp16")]; + tensor var_29699_begin_0 = const()[name = string("op_29699_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29699_end_0 = const()[name = string("op_29699_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29699_end_mask_0 = const()[name = string("op_29699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29699_cast_fp16 = slice_by_index(begin = var_29699_begin_0, end = var_29699_end_0, end_mask = var_29699_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29699_cast_fp16")]; + tensor var_29706_begin_0 = const()[name = string("op_29706_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29706_end_0 = const()[name = string("op_29706_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29706_end_mask_0 = const()[name = string("op_29706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29706_cast_fp16 = slice_by_index(begin = var_29706_begin_0, end = var_29706_end_0, end_mask = var_29706_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29706_cast_fp16")]; + tensor var_29713_begin_0 = const()[name = string("op_29713_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29713_end_0 = const()[name = string("op_29713_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29713_end_mask_0 = const()[name = string("op_29713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29713_cast_fp16 = slice_by_index(begin = var_29713_begin_0, end = var_29713_end_0, end_mask = var_29713_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29713_cast_fp16")]; + tensor var_29720_begin_0 = const()[name = string("op_29720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29720_end_0 = const()[name = string("op_29720_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29720_end_mask_0 = const()[name = string("op_29720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29720_cast_fp16 = slice_by_index(begin = var_29720_begin_0, end = var_29720_end_0, end_mask = var_29720_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29720_cast_fp16")]; + tensor var_29727_begin_0 = const()[name = string("op_29727_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29727_end_0 = const()[name = string("op_29727_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29727_end_mask_0 = const()[name = string("op_29727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29727_cast_fp16 = slice_by_index(begin = var_29727_begin_0, end = var_29727_end_0, end_mask = var_29727_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29727_cast_fp16")]; + tensor var_29734_begin_0 = const()[name = string("op_29734_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29734_end_0 = const()[name = string("op_29734_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29734_end_mask_0 = const()[name = string("op_29734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29734_cast_fp16 = slice_by_index(begin = var_29734_begin_0, end = var_29734_end_0, end_mask = var_29734_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29734_cast_fp16")]; + tensor var_29741_begin_0 = const()[name = string("op_29741_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29741_end_0 = const()[name = string("op_29741_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29741_end_mask_0 = const()[name = string("op_29741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29741_cast_fp16 = slice_by_index(begin = var_29741_begin_0, end = var_29741_end_0, end_mask = var_29741_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29741_cast_fp16")]; + tensor var_29748_begin_0 = const()[name = string("op_29748_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29748_end_0 = const()[name = string("op_29748_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29748_end_mask_0 = const()[name = string("op_29748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29748_cast_fp16 = slice_by_index(begin = var_29748_begin_0, end = var_29748_end_0, end_mask = var_29748_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29748_cast_fp16")]; + tensor var_29755_begin_0 = const()[name = string("op_29755_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29755_end_0 = const()[name = string("op_29755_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29755_end_mask_0 = const()[name = string("op_29755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29755_cast_fp16 = slice_by_index(begin = var_29755_begin_0, end = var_29755_end_0, end_mask = var_29755_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29755_cast_fp16")]; + tensor var_29762_begin_0 = const()[name = string("op_29762_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29762_end_0 = const()[name = string("op_29762_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29762_end_mask_0 = const()[name = string("op_29762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29762_cast_fp16 = slice_by_index(begin = var_29762_begin_0, end = var_29762_end_0, end_mask = var_29762_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29762_cast_fp16")]; + tensor var_29769_begin_0 = const()[name = string("op_29769_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29769_end_0 = const()[name = string("op_29769_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29769_end_mask_0 = const()[name = string("op_29769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29769_cast_fp16 = slice_by_index(begin = var_29769_begin_0, end = var_29769_end_0, end_mask = var_29769_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29769_cast_fp16")]; + tensor var_29776_begin_0 = const()[name = string("op_29776_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29776_end_0 = const()[name = string("op_29776_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29776_end_mask_0 = const()[name = string("op_29776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29776_cast_fp16 = slice_by_index(begin = var_29776_begin_0, end = var_29776_end_0, end_mask = var_29776_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29776_cast_fp16")]; + tensor var_29783_begin_0 = const()[name = string("op_29783_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29783_end_0 = const()[name = string("op_29783_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29783_end_mask_0 = const()[name = string("op_29783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29783_cast_fp16 = slice_by_index(begin = var_29783_begin_0, end = var_29783_end_0, end_mask = var_29783_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29783_cast_fp16")]; + tensor var_29790_begin_0 = const()[name = string("op_29790_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29790_end_0 = const()[name = string("op_29790_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29790_end_mask_0 = const()[name = string("op_29790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29790_cast_fp16 = slice_by_index(begin = var_29790_begin_0, end = var_29790_end_0, end_mask = var_29790_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29790_cast_fp16")]; + tensor var_29797_begin_0 = const()[name = string("op_29797_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29797_end_0 = const()[name = string("op_29797_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29797_end_mask_0 = const()[name = string("op_29797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29797_cast_fp16 = slice_by_index(begin = var_29797_begin_0, end = var_29797_end_0, end_mask = var_29797_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29797_cast_fp16")]; + tensor var_29804_begin_0 = const()[name = string("op_29804_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29804_end_0 = const()[name = string("op_29804_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29804_end_mask_0 = const()[name = string("op_29804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29804_cast_fp16 = slice_by_index(begin = var_29804_begin_0, end = var_29804_end_0, end_mask = var_29804_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29804_cast_fp16")]; + tensor var_29811_begin_0 = const()[name = string("op_29811_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29811_end_0 = const()[name = string("op_29811_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29811_end_mask_0 = const()[name = string("op_29811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29811_cast_fp16 = slice_by_index(begin = var_29811_begin_0, end = var_29811_end_0, end_mask = var_29811_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29811_cast_fp16")]; + tensor var_29818_begin_0 = const()[name = string("op_29818_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29818_end_0 = const()[name = string("op_29818_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29818_end_mask_0 = const()[name = string("op_29818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29818_cast_fp16 = slice_by_index(begin = var_29818_begin_0, end = var_29818_end_0, end_mask = var_29818_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29818_cast_fp16")]; + tensor var_29825_begin_0 = const()[name = string("op_29825_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29825_end_0 = const()[name = string("op_29825_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29825_end_mask_0 = const()[name = string("op_29825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29825_cast_fp16 = slice_by_index(begin = var_29825_begin_0, end = var_29825_end_0, end_mask = var_29825_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29825_cast_fp16")]; + tensor var_29832_begin_0 = const()[name = string("op_29832_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29832_end_0 = const()[name = string("op_29832_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29832_end_mask_0 = const()[name = string("op_29832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29832_cast_fp16 = slice_by_index(begin = var_29832_begin_0, end = var_29832_end_0, end_mask = var_29832_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29832_cast_fp16")]; + tensor var_29839_begin_0 = const()[name = string("op_29839_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29839_end_0 = const()[name = string("op_29839_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29839_end_mask_0 = const()[name = string("op_29839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29839_cast_fp16 = slice_by_index(begin = var_29839_begin_0, end = var_29839_end_0, end_mask = var_29839_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29839_cast_fp16")]; + tensor var_29846_begin_0 = const()[name = string("op_29846_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29846_end_0 = const()[name = string("op_29846_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29846_end_mask_0 = const()[name = string("op_29846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29846_cast_fp16 = slice_by_index(begin = var_29846_begin_0, end = var_29846_end_0, end_mask = var_29846_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29846_cast_fp16")]; + tensor var_29853_begin_0 = const()[name = string("op_29853_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29853_end_0 = const()[name = string("op_29853_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29853_end_mask_0 = const()[name = string("op_29853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29853_cast_fp16 = slice_by_index(begin = var_29853_begin_0, end = var_29853_end_0, end_mask = var_29853_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29853_cast_fp16")]; + tensor var_29860_begin_0 = const()[name = string("op_29860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29860_end_0 = const()[name = string("op_29860_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29860_end_mask_0 = const()[name = string("op_29860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29860_cast_fp16 = slice_by_index(begin = var_29860_begin_0, end = var_29860_end_0, end_mask = var_29860_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29860_cast_fp16")]; + tensor var_29867_begin_0 = const()[name = string("op_29867_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29867_end_0 = const()[name = string("op_29867_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29867_end_mask_0 = const()[name = string("op_29867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29867_cast_fp16 = slice_by_index(begin = var_29867_begin_0, end = var_29867_end_0, end_mask = var_29867_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29867_cast_fp16")]; + tensor var_29874_begin_0 = const()[name = string("op_29874_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29874_end_0 = const()[name = string("op_29874_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29874_end_mask_0 = const()[name = string("op_29874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29874_cast_fp16 = slice_by_index(begin = var_29874_begin_0, end = var_29874_end_0, end_mask = var_29874_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29874_cast_fp16")]; + tensor var_29881_begin_0 = const()[name = string("op_29881_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29881_end_0 = const()[name = string("op_29881_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29881_end_mask_0 = const()[name = string("op_29881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29881_cast_fp16 = slice_by_index(begin = var_29881_begin_0, end = var_29881_end_0, end_mask = var_29881_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29881_cast_fp16")]; + tensor var_29888_begin_0 = const()[name = string("op_29888_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29888_end_0 = const()[name = string("op_29888_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29888_end_mask_0 = const()[name = string("op_29888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29888_cast_fp16 = slice_by_index(begin = var_29888_begin_0, end = var_29888_end_0, end_mask = var_29888_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29888_cast_fp16")]; + tensor var_29895_begin_0 = const()[name = string("op_29895_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29895_end_0 = const()[name = string("op_29895_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29895_end_mask_0 = const()[name = string("op_29895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29895_cast_fp16 = slice_by_index(begin = var_29895_begin_0, end = var_29895_end_0, end_mask = var_29895_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29895_cast_fp16")]; + tensor var_29902_begin_0 = const()[name = string("op_29902_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29902_end_0 = const()[name = string("op_29902_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29902_end_mask_0 = const()[name = string("op_29902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29902_cast_fp16 = slice_by_index(begin = var_29902_begin_0, end = var_29902_end_0, end_mask = var_29902_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29902_cast_fp16")]; + tensor var_29909_begin_0 = const()[name = string("op_29909_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29909_end_0 = const()[name = string("op_29909_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29909_end_mask_0 = const()[name = string("op_29909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29909_cast_fp16 = slice_by_index(begin = var_29909_begin_0, end = var_29909_end_0, end_mask = var_29909_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29909_cast_fp16")]; + tensor k_39_perm_0 = const()[name = string("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29914_begin_0 = const()[name = string("op_29914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29914_end_0 = const()[name = string("op_29914_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_29914_end_mask_0 = const()[name = string("op_29914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = key_39_cast_fp16)[name = string("transpose_12")]; + tensor var_29914_cast_fp16 = slice_by_index(begin = var_29914_begin_0, end = var_29914_end_0, end_mask = var_29914_end_mask_0, x = k_39_cast_fp16)[name = string("op_29914_cast_fp16")]; + tensor var_29918_begin_0 = const()[name = string("op_29918_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29918_end_0 = const()[name = string("op_29918_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_29918_end_mask_0 = const()[name = string("op_29918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29918_cast_fp16 = slice_by_index(begin = var_29918_begin_0, end = var_29918_end_0, end_mask = var_29918_end_mask_0, x = k_39_cast_fp16)[name = string("op_29918_cast_fp16")]; + tensor var_29922_begin_0 = const()[name = string("op_29922_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29922_end_0 = const()[name = string("op_29922_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_29922_end_mask_0 = const()[name = string("op_29922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29922_cast_fp16 = slice_by_index(begin = var_29922_begin_0, end = var_29922_end_0, end_mask = var_29922_end_mask_0, x = k_39_cast_fp16)[name = string("op_29922_cast_fp16")]; + tensor var_29926_begin_0 = const()[name = string("op_29926_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29926_end_0 = const()[name = string("op_29926_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_29926_end_mask_0 = const()[name = string("op_29926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29926_cast_fp16 = slice_by_index(begin = var_29926_begin_0, end = var_29926_end_0, end_mask = var_29926_end_mask_0, x = k_39_cast_fp16)[name = string("op_29926_cast_fp16")]; + tensor var_29930_begin_0 = const()[name = string("op_29930_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29930_end_0 = const()[name = string("op_29930_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_29930_end_mask_0 = const()[name = string("op_29930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29930_cast_fp16 = slice_by_index(begin = var_29930_begin_0, end = var_29930_end_0, end_mask = var_29930_end_mask_0, x = k_39_cast_fp16)[name = string("op_29930_cast_fp16")]; + tensor var_29934_begin_0 = const()[name = string("op_29934_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29934_end_0 = const()[name = string("op_29934_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_29934_end_mask_0 = const()[name = string("op_29934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29934_cast_fp16 = slice_by_index(begin = var_29934_begin_0, end = var_29934_end_0, end_mask = var_29934_end_mask_0, x = k_39_cast_fp16)[name = string("op_29934_cast_fp16")]; + tensor var_29938_begin_0 = const()[name = string("op_29938_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29938_end_0 = const()[name = string("op_29938_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_29938_end_mask_0 = const()[name = string("op_29938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29938_cast_fp16 = slice_by_index(begin = var_29938_begin_0, end = var_29938_end_0, end_mask = var_29938_end_mask_0, x = k_39_cast_fp16)[name = string("op_29938_cast_fp16")]; + tensor var_29942_begin_0 = const()[name = string("op_29942_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29942_end_0 = const()[name = string("op_29942_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_29942_end_mask_0 = const()[name = string("op_29942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29942_cast_fp16 = slice_by_index(begin = var_29942_begin_0, end = var_29942_end_0, end_mask = var_29942_end_mask_0, x = k_39_cast_fp16)[name = string("op_29942_cast_fp16")]; + tensor var_29946_begin_0 = const()[name = string("op_29946_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29946_end_0 = const()[name = string("op_29946_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_29946_end_mask_0 = const()[name = string("op_29946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29946_cast_fp16 = slice_by_index(begin = var_29946_begin_0, end = var_29946_end_0, end_mask = var_29946_end_mask_0, x = k_39_cast_fp16)[name = string("op_29946_cast_fp16")]; + tensor var_29950_begin_0 = const()[name = string("op_29950_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29950_end_0 = const()[name = string("op_29950_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_29950_end_mask_0 = const()[name = string("op_29950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29950_cast_fp16 = slice_by_index(begin = var_29950_begin_0, end = var_29950_end_0, end_mask = var_29950_end_mask_0, x = k_39_cast_fp16)[name = string("op_29950_cast_fp16")]; + tensor var_29954_begin_0 = const()[name = string("op_29954_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29954_end_0 = const()[name = string("op_29954_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_29954_end_mask_0 = const()[name = string("op_29954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29954_cast_fp16 = slice_by_index(begin = var_29954_begin_0, end = var_29954_end_0, end_mask = var_29954_end_mask_0, x = k_39_cast_fp16)[name = string("op_29954_cast_fp16")]; + tensor var_29958_begin_0 = const()[name = string("op_29958_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29958_end_0 = const()[name = string("op_29958_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_29958_end_mask_0 = const()[name = string("op_29958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29958_cast_fp16 = slice_by_index(begin = var_29958_begin_0, end = var_29958_end_0, end_mask = var_29958_end_mask_0, x = k_39_cast_fp16)[name = string("op_29958_cast_fp16")]; + tensor var_29962_begin_0 = const()[name = string("op_29962_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29962_end_0 = const()[name = string("op_29962_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_29962_end_mask_0 = const()[name = string("op_29962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29962_cast_fp16 = slice_by_index(begin = var_29962_begin_0, end = var_29962_end_0, end_mask = var_29962_end_mask_0, x = k_39_cast_fp16)[name = string("op_29962_cast_fp16")]; + tensor var_29966_begin_0 = const()[name = string("op_29966_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29966_end_0 = const()[name = string("op_29966_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_29966_end_mask_0 = const()[name = string("op_29966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29966_cast_fp16 = slice_by_index(begin = var_29966_begin_0, end = var_29966_end_0, end_mask = var_29966_end_mask_0, x = k_39_cast_fp16)[name = string("op_29966_cast_fp16")]; + tensor var_29970_begin_0 = const()[name = string("op_29970_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29970_end_0 = const()[name = string("op_29970_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_29970_end_mask_0 = const()[name = string("op_29970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29970_cast_fp16 = slice_by_index(begin = var_29970_begin_0, end = var_29970_end_0, end_mask = var_29970_end_mask_0, x = k_39_cast_fp16)[name = string("op_29970_cast_fp16")]; + tensor var_29974_begin_0 = const()[name = string("op_29974_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29974_end_0 = const()[name = string("op_29974_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_29974_end_mask_0 = const()[name = string("op_29974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29974_cast_fp16 = slice_by_index(begin = var_29974_begin_0, end = var_29974_end_0, end_mask = var_29974_end_mask_0, x = k_39_cast_fp16)[name = string("op_29974_cast_fp16")]; + tensor var_29978_begin_0 = const()[name = string("op_29978_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29978_end_0 = const()[name = string("op_29978_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_29978_end_mask_0 = const()[name = string("op_29978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29978_cast_fp16 = slice_by_index(begin = var_29978_begin_0, end = var_29978_end_0, end_mask = var_29978_end_mask_0, x = k_39_cast_fp16)[name = string("op_29978_cast_fp16")]; + tensor var_29982_begin_0 = const()[name = string("op_29982_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29982_end_0 = const()[name = string("op_29982_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_29982_end_mask_0 = const()[name = string("op_29982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29982_cast_fp16 = slice_by_index(begin = var_29982_begin_0, end = var_29982_end_0, end_mask = var_29982_end_mask_0, x = k_39_cast_fp16)[name = string("op_29982_cast_fp16")]; + tensor var_29986_begin_0 = const()[name = string("op_29986_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29986_end_0 = const()[name = string("op_29986_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_29986_end_mask_0 = const()[name = string("op_29986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29986_cast_fp16 = slice_by_index(begin = var_29986_begin_0, end = var_29986_end_0, end_mask = var_29986_end_mask_0, x = k_39_cast_fp16)[name = string("op_29986_cast_fp16")]; + tensor var_29990_begin_0 = const()[name = string("op_29990_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_29990_end_0 = const()[name = string("op_29990_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_29990_end_mask_0 = const()[name = string("op_29990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29990_cast_fp16 = slice_by_index(begin = var_29990_begin_0, end = var_29990_end_0, end_mask = var_29990_end_mask_0, x = k_39_cast_fp16)[name = string("op_29990_cast_fp16")]; + tensor var_29992_begin_0 = const()[name = string("op_29992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29992_end_0 = const()[name = string("op_29992_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29992_end_mask_0 = const()[name = string("op_29992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29992_cast_fp16 = slice_by_index(begin = var_29992_begin_0, end = var_29992_end_0, end_mask = var_29992_end_mask_0, x = value_39_cast_fp16)[name = string("op_29992_cast_fp16")]; + tensor var_29996_begin_0 = const()[name = string("op_29996_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29996_end_0 = const()[name = string("op_29996_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_29996_end_mask_0 = const()[name = string("op_29996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29996_cast_fp16 = slice_by_index(begin = var_29996_begin_0, end = var_29996_end_0, end_mask = var_29996_end_mask_0, x = value_39_cast_fp16)[name = string("op_29996_cast_fp16")]; + tensor var_30000_begin_0 = const()[name = string("op_30000_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30000_end_0 = const()[name = string("op_30000_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_30000_end_mask_0 = const()[name = string("op_30000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30000_cast_fp16 = slice_by_index(begin = var_30000_begin_0, end = var_30000_end_0, end_mask = var_30000_end_mask_0, x = value_39_cast_fp16)[name = string("op_30000_cast_fp16")]; + tensor var_30004_begin_0 = const()[name = string("op_30004_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30004_end_0 = const()[name = string("op_30004_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_30004_end_mask_0 = const()[name = string("op_30004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30004_cast_fp16 = slice_by_index(begin = var_30004_begin_0, end = var_30004_end_0, end_mask = var_30004_end_mask_0, x = value_39_cast_fp16)[name = string("op_30004_cast_fp16")]; + tensor var_30008_begin_0 = const()[name = string("op_30008_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30008_end_0 = const()[name = string("op_30008_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_30008_end_mask_0 = const()[name = string("op_30008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30008_cast_fp16 = slice_by_index(begin = var_30008_begin_0, end = var_30008_end_0, end_mask = var_30008_end_mask_0, x = value_39_cast_fp16)[name = string("op_30008_cast_fp16")]; + tensor var_30012_begin_0 = const()[name = string("op_30012_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30012_end_0 = const()[name = string("op_30012_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_30012_end_mask_0 = const()[name = string("op_30012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30012_cast_fp16 = slice_by_index(begin = var_30012_begin_0, end = var_30012_end_0, end_mask = var_30012_end_mask_0, x = value_39_cast_fp16)[name = string("op_30012_cast_fp16")]; + tensor var_30016_begin_0 = const()[name = string("op_30016_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30016_end_0 = const()[name = string("op_30016_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_30016_end_mask_0 = const()[name = string("op_30016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30016_cast_fp16 = slice_by_index(begin = var_30016_begin_0, end = var_30016_end_0, end_mask = var_30016_end_mask_0, x = value_39_cast_fp16)[name = string("op_30016_cast_fp16")]; + tensor var_30020_begin_0 = const()[name = string("op_30020_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30020_end_0 = const()[name = string("op_30020_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_30020_end_mask_0 = const()[name = string("op_30020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30020_cast_fp16 = slice_by_index(begin = var_30020_begin_0, end = var_30020_end_0, end_mask = var_30020_end_mask_0, x = value_39_cast_fp16)[name = string("op_30020_cast_fp16")]; + tensor var_30024_begin_0 = const()[name = string("op_30024_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30024_end_0 = const()[name = string("op_30024_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_30024_end_mask_0 = const()[name = string("op_30024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30024_cast_fp16 = slice_by_index(begin = var_30024_begin_0, end = var_30024_end_0, end_mask = var_30024_end_mask_0, x = value_39_cast_fp16)[name = string("op_30024_cast_fp16")]; + tensor var_30028_begin_0 = const()[name = string("op_30028_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30028_end_0 = const()[name = string("op_30028_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_30028_end_mask_0 = const()[name = string("op_30028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30028_cast_fp16 = slice_by_index(begin = var_30028_begin_0, end = var_30028_end_0, end_mask = var_30028_end_mask_0, x = value_39_cast_fp16)[name = string("op_30028_cast_fp16")]; + tensor var_30032_begin_0 = const()[name = string("op_30032_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30032_end_0 = const()[name = string("op_30032_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_30032_end_mask_0 = const()[name = string("op_30032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30032_cast_fp16 = slice_by_index(begin = var_30032_begin_0, end = var_30032_end_0, end_mask = var_30032_end_mask_0, x = value_39_cast_fp16)[name = string("op_30032_cast_fp16")]; + tensor var_30036_begin_0 = const()[name = string("op_30036_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30036_end_0 = const()[name = string("op_30036_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_30036_end_mask_0 = const()[name = string("op_30036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30036_cast_fp16 = slice_by_index(begin = var_30036_begin_0, end = var_30036_end_0, end_mask = var_30036_end_mask_0, x = value_39_cast_fp16)[name = string("op_30036_cast_fp16")]; + tensor var_30040_begin_0 = const()[name = string("op_30040_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30040_end_0 = const()[name = string("op_30040_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_30040_end_mask_0 = const()[name = string("op_30040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30040_cast_fp16 = slice_by_index(begin = var_30040_begin_0, end = var_30040_end_0, end_mask = var_30040_end_mask_0, x = value_39_cast_fp16)[name = string("op_30040_cast_fp16")]; + tensor var_30044_begin_0 = const()[name = string("op_30044_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30044_end_0 = const()[name = string("op_30044_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_30044_end_mask_0 = const()[name = string("op_30044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30044_cast_fp16 = slice_by_index(begin = var_30044_begin_0, end = var_30044_end_0, end_mask = var_30044_end_mask_0, x = value_39_cast_fp16)[name = string("op_30044_cast_fp16")]; + tensor var_30048_begin_0 = const()[name = string("op_30048_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30048_end_0 = const()[name = string("op_30048_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_30048_end_mask_0 = const()[name = string("op_30048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30048_cast_fp16 = slice_by_index(begin = var_30048_begin_0, end = var_30048_end_0, end_mask = var_30048_end_mask_0, x = value_39_cast_fp16)[name = string("op_30048_cast_fp16")]; + tensor var_30052_begin_0 = const()[name = string("op_30052_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30052_end_0 = const()[name = string("op_30052_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_30052_end_mask_0 = const()[name = string("op_30052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30052_cast_fp16 = slice_by_index(begin = var_30052_begin_0, end = var_30052_end_0, end_mask = var_30052_end_mask_0, x = value_39_cast_fp16)[name = string("op_30052_cast_fp16")]; + tensor var_30056_begin_0 = const()[name = string("op_30056_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30056_end_0 = const()[name = string("op_30056_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_30056_end_mask_0 = const()[name = string("op_30056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30056_cast_fp16 = slice_by_index(begin = var_30056_begin_0, end = var_30056_end_0, end_mask = var_30056_end_mask_0, x = value_39_cast_fp16)[name = string("op_30056_cast_fp16")]; + tensor var_30060_begin_0 = const()[name = string("op_30060_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30060_end_0 = const()[name = string("op_30060_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_30060_end_mask_0 = const()[name = string("op_30060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30060_cast_fp16 = slice_by_index(begin = var_30060_begin_0, end = var_30060_end_0, end_mask = var_30060_end_mask_0, x = value_39_cast_fp16)[name = string("op_30060_cast_fp16")]; + tensor var_30064_begin_0 = const()[name = string("op_30064_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30064_end_0 = const()[name = string("op_30064_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_30064_end_mask_0 = const()[name = string("op_30064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30064_cast_fp16 = slice_by_index(begin = var_30064_begin_0, end = var_30064_end_0, end_mask = var_30064_end_mask_0, x = value_39_cast_fp16)[name = string("op_30064_cast_fp16")]; + tensor var_30068_begin_0 = const()[name = string("op_30068_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30068_end_0 = const()[name = string("op_30068_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_30068_end_mask_0 = const()[name = string("op_30068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30068_cast_fp16 = slice_by_index(begin = var_30068_begin_0, end = var_30068_end_0, end_mask = var_30068_end_mask_0, x = value_39_cast_fp16)[name = string("op_30068_cast_fp16")]; + string _SplitHeadsQ__mh_w_3041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3041_equation_0, values = (var_29914_cast_fp16, var_29356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3041_cast_fp16")]; + string _SplitHeadsQ__mh_w_3043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3043_equation_0, values = (var_29914_cast_fp16, var_29363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3043_cast_fp16")]; + string _SplitHeadsQ__mh_w_3045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3045_equation_0, values = (var_29914_cast_fp16, var_29370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3045_cast_fp16")]; + string _SplitHeadsQ__mh_w_3047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3047_equation_0, values = (var_29914_cast_fp16, var_29377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3047_cast_fp16")]; + string _SplitHeadsQ__mh_w_3049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3049_equation_0, values = (var_29918_cast_fp16, var_29384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3049_cast_fp16")]; + string _SplitHeadsQ__mh_w_3051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3051_equation_0, values = (var_29918_cast_fp16, var_29391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3051_cast_fp16")]; + string _SplitHeadsQ__mh_w_3053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3053_equation_0, values = (var_29918_cast_fp16, var_29398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3053_cast_fp16")]; + string _SplitHeadsQ__mh_w_3055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3055_equation_0, values = (var_29918_cast_fp16, var_29405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3055_cast_fp16")]; + string _SplitHeadsQ__mh_w_3057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3057_equation_0, values = (var_29922_cast_fp16, var_29412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3057_cast_fp16")]; + string _SplitHeadsQ__mh_w_3059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3059_equation_0, values = (var_29922_cast_fp16, var_29419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3059_cast_fp16")]; + string _SplitHeadsQ__mh_w_3061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3061_equation_0, values = (var_29922_cast_fp16, var_29426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3061_cast_fp16")]; + string _SplitHeadsQ__mh_w_3063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3063_equation_0, values = (var_29922_cast_fp16, var_29433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3063_cast_fp16")]; + string _SplitHeadsQ__mh_w_3065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3065_equation_0, values = (var_29926_cast_fp16, var_29440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3065_cast_fp16")]; + string _SplitHeadsQ__mh_w_3067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3067_equation_0, values = (var_29926_cast_fp16, var_29447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3067_cast_fp16")]; + string _SplitHeadsQ__mh_w_3069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3069_equation_0, values = (var_29926_cast_fp16, var_29454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3069_cast_fp16")]; + string _SplitHeadsQ__mh_w_3071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3071_equation_0, values = (var_29926_cast_fp16, var_29461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3071_cast_fp16")]; + string _SplitHeadsQ__mh_w_3073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3073_equation_0, values = (var_29930_cast_fp16, var_29468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3073_cast_fp16")]; + string _SplitHeadsQ__mh_w_3075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3075_equation_0, values = (var_29930_cast_fp16, var_29475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3075_cast_fp16")]; + string _SplitHeadsQ__mh_w_3077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3077_equation_0, values = (var_29930_cast_fp16, var_29482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3077_cast_fp16")]; + string _SplitHeadsQ__mh_w_3079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3079_equation_0, values = (var_29930_cast_fp16, var_29489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3079_cast_fp16")]; + string _SplitHeadsQ__mh_w_3081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3081_equation_0, values = (var_29934_cast_fp16, var_29496_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3081_cast_fp16")]; + string _SplitHeadsQ__mh_w_3083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3083_equation_0, values = (var_29934_cast_fp16, var_29503_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3083_cast_fp16")]; + string _SplitHeadsQ__mh_w_3085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3085_equation_0, values = (var_29934_cast_fp16, var_29510_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3085_cast_fp16")]; + string _SplitHeadsQ__mh_w_3087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3087_equation_0, values = (var_29934_cast_fp16, var_29517_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3087_cast_fp16")]; + string _SplitHeadsQ__mh_w_3089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3089_equation_0, values = (var_29938_cast_fp16, var_29524_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3089_cast_fp16")]; + string _SplitHeadsQ__mh_w_3091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3091_equation_0, values = (var_29938_cast_fp16, var_29531_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3091_cast_fp16")]; + string _SplitHeadsQ__mh_w_3093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3093_equation_0, values = (var_29938_cast_fp16, var_29538_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3093_cast_fp16")]; + string _SplitHeadsQ__mh_w_3095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3095_equation_0, values = (var_29938_cast_fp16, var_29545_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3095_cast_fp16")]; + string _SplitHeadsQ__mh_w_3097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3097_equation_0, values = (var_29942_cast_fp16, var_29552_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3097_cast_fp16")]; + string _SplitHeadsQ__mh_w_3099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3099_equation_0, values = (var_29942_cast_fp16, var_29559_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3099_cast_fp16")]; + string _SplitHeadsQ__mh_w_3101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3101_equation_0, values = (var_29942_cast_fp16, var_29566_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3101_cast_fp16")]; + string _SplitHeadsQ__mh_w_3103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3103_equation_0, values = (var_29942_cast_fp16, var_29573_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3103_cast_fp16")]; + string _SplitHeadsQ__mh_w_3105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3105_equation_0, values = (var_29946_cast_fp16, var_29580_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3105_cast_fp16")]; + string _SplitHeadsQ__mh_w_3107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3107_equation_0, values = (var_29946_cast_fp16, var_29587_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3107_cast_fp16")]; + string _SplitHeadsQ__mh_w_3109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3109_equation_0, values = (var_29946_cast_fp16, var_29594_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3109_cast_fp16")]; + string _SplitHeadsQ__mh_w_3111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3111_equation_0, values = (var_29946_cast_fp16, var_29601_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3111_cast_fp16")]; + string _SplitHeadsQ__mh_w_3113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3113_equation_0, values = (var_29950_cast_fp16, var_29608_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3113_cast_fp16")]; + string _SplitHeadsQ__mh_w_3115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3115_equation_0, values = (var_29950_cast_fp16, var_29615_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3115_cast_fp16")]; + string _SplitHeadsQ__mh_w_3117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3117_equation_0, values = (var_29950_cast_fp16, var_29622_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3117_cast_fp16")]; + string _SplitHeadsQ__mh_w_3119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3119_equation_0, values = (var_29950_cast_fp16, var_29629_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3119_cast_fp16")]; + string _SplitHeadsQ__mh_w_3121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3121_equation_0, values = (var_29954_cast_fp16, var_29636_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3121_cast_fp16")]; + string _SplitHeadsQ__mh_w_3123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3123_equation_0, values = (var_29954_cast_fp16, var_29643_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3123_cast_fp16")]; + string _SplitHeadsQ__mh_w_3125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3125_equation_0, values = (var_29954_cast_fp16, var_29650_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3125_cast_fp16")]; + string _SplitHeadsQ__mh_w_3127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3127_equation_0, values = (var_29954_cast_fp16, var_29657_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3127_cast_fp16")]; + string _SplitHeadsQ__mh_w_3129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3129_equation_0, values = (var_29958_cast_fp16, var_29664_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3129_cast_fp16")]; + string _SplitHeadsQ__mh_w_3131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3131_equation_0, values = (var_29958_cast_fp16, var_29671_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3131_cast_fp16")]; + string _SplitHeadsQ__mh_w_3133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3133_equation_0, values = (var_29958_cast_fp16, var_29678_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3133_cast_fp16")]; + string _SplitHeadsQ__mh_w_3135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3135_equation_0, values = (var_29958_cast_fp16, var_29685_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3135_cast_fp16")]; + string _SplitHeadsQ__mh_w_3137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3137_equation_0, values = (var_29962_cast_fp16, var_29692_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3137_cast_fp16")]; + string _SplitHeadsQ__mh_w_3139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3139_equation_0, values = (var_29962_cast_fp16, var_29699_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3139_cast_fp16")]; + string _SplitHeadsQ__mh_w_3141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3141_equation_0, values = (var_29962_cast_fp16, var_29706_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3141_cast_fp16")]; + string _SplitHeadsQ__mh_w_3143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3143_equation_0, values = (var_29962_cast_fp16, var_29713_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3143_cast_fp16")]; + string _SplitHeadsQ__mh_w_3145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3145_equation_0, values = (var_29966_cast_fp16, var_29720_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3145_cast_fp16")]; + string _SplitHeadsQ__mh_w_3147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3147_equation_0, values = (var_29966_cast_fp16, var_29727_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3147_cast_fp16")]; + string _SplitHeadsQ__mh_w_3149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3149_equation_0, values = (var_29966_cast_fp16, var_29734_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3149_cast_fp16")]; + string _SplitHeadsQ__mh_w_3151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3151_equation_0, values = (var_29966_cast_fp16, var_29741_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3151_cast_fp16")]; + string _SplitHeadsQ__mh_w_3153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3153_equation_0, values = (var_29970_cast_fp16, var_29748_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3153_cast_fp16")]; + string _SplitHeadsQ__mh_w_3155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3155_equation_0, values = (var_29970_cast_fp16, var_29755_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3155_cast_fp16")]; + string _SplitHeadsQ__mh_w_3157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3157_equation_0, values = (var_29970_cast_fp16, var_29762_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3157_cast_fp16")]; + string _SplitHeadsQ__mh_w_3159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3159_equation_0, values = (var_29970_cast_fp16, var_29769_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3159_cast_fp16")]; + string _SplitHeadsQ__mh_w_3161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3161_equation_0, values = (var_29974_cast_fp16, var_29776_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3161_cast_fp16")]; + string _SplitHeadsQ__mh_w_3163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3163_equation_0, values = (var_29974_cast_fp16, var_29783_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3163_cast_fp16")]; + string _SplitHeadsQ__mh_w_3165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3165_equation_0, values = (var_29974_cast_fp16, var_29790_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3165_cast_fp16")]; + string _SplitHeadsQ__mh_w_3167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3167_equation_0, values = (var_29974_cast_fp16, var_29797_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3167_cast_fp16")]; + string _SplitHeadsQ__mh_w_3169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3169_equation_0, values = (var_29978_cast_fp16, var_29804_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3169_cast_fp16")]; + string _SplitHeadsQ__mh_w_3171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3171_equation_0, values = (var_29978_cast_fp16, var_29811_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3171_cast_fp16")]; + string _SplitHeadsQ__mh_w_3173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3173_equation_0, values = (var_29978_cast_fp16, var_29818_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3173_cast_fp16")]; + string _SplitHeadsQ__mh_w_3175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3175_equation_0, values = (var_29978_cast_fp16, var_29825_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3175_cast_fp16")]; + string _SplitHeadsQ__mh_w_3177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3177_equation_0, values = (var_29982_cast_fp16, var_29832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3177_cast_fp16")]; + string _SplitHeadsQ__mh_w_3179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3179_equation_0, values = (var_29982_cast_fp16, var_29839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3179_cast_fp16")]; + string _SplitHeadsQ__mh_w_3181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3181_equation_0, values = (var_29982_cast_fp16, var_29846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3181_cast_fp16")]; + string _SplitHeadsQ__mh_w_3183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3183_equation_0, values = (var_29982_cast_fp16, var_29853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3183_cast_fp16")]; + string _SplitHeadsQ__mh_w_3185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3185_equation_0, values = (var_29986_cast_fp16, var_29860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3185_cast_fp16")]; + string _SplitHeadsQ__mh_w_3187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3187_equation_0, values = (var_29986_cast_fp16, var_29867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3187_cast_fp16")]; + string _SplitHeadsQ__mh_w_3189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3189_equation_0, values = (var_29986_cast_fp16, var_29874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3189_cast_fp16")]; + string _SplitHeadsQ__mh_w_3191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3191_equation_0, values = (var_29986_cast_fp16, var_29881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3191_cast_fp16")]; + string _SplitHeadsQ__mh_w_3193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3193_equation_0, values = (var_29990_cast_fp16, var_29888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3193_cast_fp16")]; + string _SplitHeadsQ__mh_w_3195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3195_equation_0, values = (var_29990_cast_fp16, var_29895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3195_cast_fp16")]; + string _SplitHeadsQ__mh_w_3197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3197_equation_0, values = (var_29990_cast_fp16, var_29902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3197_cast_fp16")]; + string _SplitHeadsQ__mh_w_3199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3199_equation_0, values = (var_29990_cast_fp16, var_29909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3199_cast_fp16")]; + fp16 var_30231_to_fp16 = const()[name = string("op_30231_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3041_cast_fp16, y = var_30231_to_fp16)[name = string("aw_chunk_3041_cast_fp16")]; + fp16 var_30233_to_fp16 = const()[name = string("op_30233_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3043_cast_fp16, y = var_30233_to_fp16)[name = string("aw_chunk_3043_cast_fp16")]; + fp16 var_30235_to_fp16 = const()[name = string("op_30235_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3045_cast_fp16, y = var_30235_to_fp16)[name = string("aw_chunk_3045_cast_fp16")]; + fp16 var_30237_to_fp16 = const()[name = string("op_30237_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3047_cast_fp16, y = var_30237_to_fp16)[name = string("aw_chunk_3047_cast_fp16")]; + fp16 var_30239_to_fp16 = const()[name = string("op_30239_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3049_cast_fp16, y = var_30239_to_fp16)[name = string("aw_chunk_3049_cast_fp16")]; + fp16 var_30241_to_fp16 = const()[name = string("op_30241_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3051_cast_fp16, y = var_30241_to_fp16)[name = string("aw_chunk_3051_cast_fp16")]; + fp16 var_30243_to_fp16 = const()[name = string("op_30243_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3053_cast_fp16, y = var_30243_to_fp16)[name = string("aw_chunk_3053_cast_fp16")]; + fp16 var_30245_to_fp16 = const()[name = string("op_30245_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3055_cast_fp16, y = var_30245_to_fp16)[name = string("aw_chunk_3055_cast_fp16")]; + fp16 var_30247_to_fp16 = const()[name = string("op_30247_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3057_cast_fp16, y = var_30247_to_fp16)[name = string("aw_chunk_3057_cast_fp16")]; + fp16 var_30249_to_fp16 = const()[name = string("op_30249_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3059_cast_fp16, y = var_30249_to_fp16)[name = string("aw_chunk_3059_cast_fp16")]; + fp16 var_30251_to_fp16 = const()[name = string("op_30251_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3061_cast_fp16, y = var_30251_to_fp16)[name = string("aw_chunk_3061_cast_fp16")]; + fp16 var_30253_to_fp16 = const()[name = string("op_30253_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3063_cast_fp16, y = var_30253_to_fp16)[name = string("aw_chunk_3063_cast_fp16")]; + fp16 var_30255_to_fp16 = const()[name = string("op_30255_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3065_cast_fp16, y = var_30255_to_fp16)[name = string("aw_chunk_3065_cast_fp16")]; + fp16 var_30257_to_fp16 = const()[name = string("op_30257_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3067_cast_fp16, y = var_30257_to_fp16)[name = string("aw_chunk_3067_cast_fp16")]; + fp16 var_30259_to_fp16 = const()[name = string("op_30259_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3069_cast_fp16, y = var_30259_to_fp16)[name = string("aw_chunk_3069_cast_fp16")]; + fp16 var_30261_to_fp16 = const()[name = string("op_30261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3071_cast_fp16, y = var_30261_to_fp16)[name = string("aw_chunk_3071_cast_fp16")]; + fp16 var_30263_to_fp16 = const()[name = string("op_30263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3073_cast_fp16, y = var_30263_to_fp16)[name = string("aw_chunk_3073_cast_fp16")]; + fp16 var_30265_to_fp16 = const()[name = string("op_30265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3075_cast_fp16, y = var_30265_to_fp16)[name = string("aw_chunk_3075_cast_fp16")]; + fp16 var_30267_to_fp16 = const()[name = string("op_30267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3077_cast_fp16, y = var_30267_to_fp16)[name = string("aw_chunk_3077_cast_fp16")]; + fp16 var_30269_to_fp16 = const()[name = string("op_30269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3079_cast_fp16, y = var_30269_to_fp16)[name = string("aw_chunk_3079_cast_fp16")]; + fp16 var_30271_to_fp16 = const()[name = string("op_30271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3081_cast_fp16, y = var_30271_to_fp16)[name = string("aw_chunk_3081_cast_fp16")]; + fp16 var_30273_to_fp16 = const()[name = string("op_30273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3083_cast_fp16, y = var_30273_to_fp16)[name = string("aw_chunk_3083_cast_fp16")]; + fp16 var_30275_to_fp16 = const()[name = string("op_30275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3085_cast_fp16, y = var_30275_to_fp16)[name = string("aw_chunk_3085_cast_fp16")]; + fp16 var_30277_to_fp16 = const()[name = string("op_30277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3087_cast_fp16, y = var_30277_to_fp16)[name = string("aw_chunk_3087_cast_fp16")]; + fp16 var_30279_to_fp16 = const()[name = string("op_30279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3089_cast_fp16, y = var_30279_to_fp16)[name = string("aw_chunk_3089_cast_fp16")]; + fp16 var_30281_to_fp16 = const()[name = string("op_30281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3091_cast_fp16, y = var_30281_to_fp16)[name = string("aw_chunk_3091_cast_fp16")]; + fp16 var_30283_to_fp16 = const()[name = string("op_30283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3093_cast_fp16, y = var_30283_to_fp16)[name = string("aw_chunk_3093_cast_fp16")]; + fp16 var_30285_to_fp16 = const()[name = string("op_30285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3095_cast_fp16, y = var_30285_to_fp16)[name = string("aw_chunk_3095_cast_fp16")]; + fp16 var_30287_to_fp16 = const()[name = string("op_30287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3097_cast_fp16, y = var_30287_to_fp16)[name = string("aw_chunk_3097_cast_fp16")]; + fp16 var_30289_to_fp16 = const()[name = string("op_30289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3099_cast_fp16, y = var_30289_to_fp16)[name = string("aw_chunk_3099_cast_fp16")]; + fp16 var_30291_to_fp16 = const()[name = string("op_30291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3101_cast_fp16, y = var_30291_to_fp16)[name = string("aw_chunk_3101_cast_fp16")]; + fp16 var_30293_to_fp16 = const()[name = string("op_30293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3103_cast_fp16, y = var_30293_to_fp16)[name = string("aw_chunk_3103_cast_fp16")]; + fp16 var_30295_to_fp16 = const()[name = string("op_30295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3105_cast_fp16, y = var_30295_to_fp16)[name = string("aw_chunk_3105_cast_fp16")]; + fp16 var_30297_to_fp16 = const()[name = string("op_30297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3107_cast_fp16, y = var_30297_to_fp16)[name = string("aw_chunk_3107_cast_fp16")]; + fp16 var_30299_to_fp16 = const()[name = string("op_30299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3109_cast_fp16, y = var_30299_to_fp16)[name = string("aw_chunk_3109_cast_fp16")]; + fp16 var_30301_to_fp16 = const()[name = string("op_30301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3111_cast_fp16, y = var_30301_to_fp16)[name = string("aw_chunk_3111_cast_fp16")]; + fp16 var_30303_to_fp16 = const()[name = string("op_30303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3113_cast_fp16, y = var_30303_to_fp16)[name = string("aw_chunk_3113_cast_fp16")]; + fp16 var_30305_to_fp16 = const()[name = string("op_30305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3115_cast_fp16, y = var_30305_to_fp16)[name = string("aw_chunk_3115_cast_fp16")]; + fp16 var_30307_to_fp16 = const()[name = string("op_30307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3117_cast_fp16, y = var_30307_to_fp16)[name = string("aw_chunk_3117_cast_fp16")]; + fp16 var_30309_to_fp16 = const()[name = string("op_30309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3119_cast_fp16, y = var_30309_to_fp16)[name = string("aw_chunk_3119_cast_fp16")]; + fp16 var_30311_to_fp16 = const()[name = string("op_30311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3121_cast_fp16, y = var_30311_to_fp16)[name = string("aw_chunk_3121_cast_fp16")]; + fp16 var_30313_to_fp16 = const()[name = string("op_30313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3123_cast_fp16, y = var_30313_to_fp16)[name = string("aw_chunk_3123_cast_fp16")]; + fp16 var_30315_to_fp16 = const()[name = string("op_30315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3125_cast_fp16, y = var_30315_to_fp16)[name = string("aw_chunk_3125_cast_fp16")]; + fp16 var_30317_to_fp16 = const()[name = string("op_30317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3127_cast_fp16, y = var_30317_to_fp16)[name = string("aw_chunk_3127_cast_fp16")]; + fp16 var_30319_to_fp16 = const()[name = string("op_30319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3129_cast_fp16, y = var_30319_to_fp16)[name = string("aw_chunk_3129_cast_fp16")]; + fp16 var_30321_to_fp16 = const()[name = string("op_30321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3131_cast_fp16, y = var_30321_to_fp16)[name = string("aw_chunk_3131_cast_fp16")]; + fp16 var_30323_to_fp16 = const()[name = string("op_30323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3133_cast_fp16, y = var_30323_to_fp16)[name = string("aw_chunk_3133_cast_fp16")]; + fp16 var_30325_to_fp16 = const()[name = string("op_30325_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3135_cast_fp16, y = var_30325_to_fp16)[name = string("aw_chunk_3135_cast_fp16")]; + fp16 var_30327_to_fp16 = const()[name = string("op_30327_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3137_cast_fp16, y = var_30327_to_fp16)[name = string("aw_chunk_3137_cast_fp16")]; + fp16 var_30329_to_fp16 = const()[name = string("op_30329_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3139_cast_fp16, y = var_30329_to_fp16)[name = string("aw_chunk_3139_cast_fp16")]; + fp16 var_30331_to_fp16 = const()[name = string("op_30331_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3141_cast_fp16, y = var_30331_to_fp16)[name = string("aw_chunk_3141_cast_fp16")]; + fp16 var_30333_to_fp16 = const()[name = string("op_30333_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3143_cast_fp16, y = var_30333_to_fp16)[name = string("aw_chunk_3143_cast_fp16")]; + fp16 var_30335_to_fp16 = const()[name = string("op_30335_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3145_cast_fp16, y = var_30335_to_fp16)[name = string("aw_chunk_3145_cast_fp16")]; + fp16 var_30337_to_fp16 = const()[name = string("op_30337_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3147_cast_fp16, y = var_30337_to_fp16)[name = string("aw_chunk_3147_cast_fp16")]; + fp16 var_30339_to_fp16 = const()[name = string("op_30339_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3149_cast_fp16, y = var_30339_to_fp16)[name = string("aw_chunk_3149_cast_fp16")]; + fp16 var_30341_to_fp16 = const()[name = string("op_30341_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3151_cast_fp16, y = var_30341_to_fp16)[name = string("aw_chunk_3151_cast_fp16")]; + fp16 var_30343_to_fp16 = const()[name = string("op_30343_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3153_cast_fp16, y = var_30343_to_fp16)[name = string("aw_chunk_3153_cast_fp16")]; + fp16 var_30345_to_fp16 = const()[name = string("op_30345_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3155_cast_fp16, y = var_30345_to_fp16)[name = string("aw_chunk_3155_cast_fp16")]; + fp16 var_30347_to_fp16 = const()[name = string("op_30347_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3157_cast_fp16, y = var_30347_to_fp16)[name = string("aw_chunk_3157_cast_fp16")]; + fp16 var_30349_to_fp16 = const()[name = string("op_30349_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3159_cast_fp16, y = var_30349_to_fp16)[name = string("aw_chunk_3159_cast_fp16")]; + fp16 var_30351_to_fp16 = const()[name = string("op_30351_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3161_cast_fp16, y = var_30351_to_fp16)[name = string("aw_chunk_3161_cast_fp16")]; + fp16 var_30353_to_fp16 = const()[name = string("op_30353_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3163_cast_fp16, y = var_30353_to_fp16)[name = string("aw_chunk_3163_cast_fp16")]; + fp16 var_30355_to_fp16 = const()[name = string("op_30355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3165_cast_fp16, y = var_30355_to_fp16)[name = string("aw_chunk_3165_cast_fp16")]; + fp16 var_30357_to_fp16 = const()[name = string("op_30357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3167_cast_fp16, y = var_30357_to_fp16)[name = string("aw_chunk_3167_cast_fp16")]; + fp16 var_30359_to_fp16 = const()[name = string("op_30359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3169_cast_fp16, y = var_30359_to_fp16)[name = string("aw_chunk_3169_cast_fp16")]; + fp16 var_30361_to_fp16 = const()[name = string("op_30361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3171_cast_fp16, y = var_30361_to_fp16)[name = string("aw_chunk_3171_cast_fp16")]; + fp16 var_30363_to_fp16 = const()[name = string("op_30363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3173_cast_fp16, y = var_30363_to_fp16)[name = string("aw_chunk_3173_cast_fp16")]; + fp16 var_30365_to_fp16 = const()[name = string("op_30365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3175_cast_fp16, y = var_30365_to_fp16)[name = string("aw_chunk_3175_cast_fp16")]; + fp16 var_30367_to_fp16 = const()[name = string("op_30367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3177_cast_fp16, y = var_30367_to_fp16)[name = string("aw_chunk_3177_cast_fp16")]; + fp16 var_30369_to_fp16 = const()[name = string("op_30369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3179_cast_fp16, y = var_30369_to_fp16)[name = string("aw_chunk_3179_cast_fp16")]; + fp16 var_30371_to_fp16 = const()[name = string("op_30371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3181_cast_fp16, y = var_30371_to_fp16)[name = string("aw_chunk_3181_cast_fp16")]; + fp16 var_30373_to_fp16 = const()[name = string("op_30373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3183_cast_fp16, y = var_30373_to_fp16)[name = string("aw_chunk_3183_cast_fp16")]; + fp16 var_30375_to_fp16 = const()[name = string("op_30375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3185_cast_fp16, y = var_30375_to_fp16)[name = string("aw_chunk_3185_cast_fp16")]; + fp16 var_30377_to_fp16 = const()[name = string("op_30377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3187_cast_fp16, y = var_30377_to_fp16)[name = string("aw_chunk_3187_cast_fp16")]; + fp16 var_30379_to_fp16 = const()[name = string("op_30379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3189_cast_fp16, y = var_30379_to_fp16)[name = string("aw_chunk_3189_cast_fp16")]; + fp16 var_30381_to_fp16 = const()[name = string("op_30381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3191_cast_fp16, y = var_30381_to_fp16)[name = string("aw_chunk_3191_cast_fp16")]; + fp16 var_30383_to_fp16 = const()[name = string("op_30383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3193_cast_fp16, y = var_30383_to_fp16)[name = string("aw_chunk_3193_cast_fp16")]; + fp16 var_30385_to_fp16 = const()[name = string("op_30385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3195_cast_fp16, y = var_30385_to_fp16)[name = string("aw_chunk_3195_cast_fp16")]; + fp16 var_30387_to_fp16 = const()[name = string("op_30387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3197_cast_fp16, y = var_30387_to_fp16)[name = string("aw_chunk_3197_cast_fp16")]; + fp16 var_30389_to_fp16 = const()[name = string("op_30389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3199_cast_fp16, y = var_30389_to_fp16)[name = string("aw_chunk_3199_cast_fp16")]; + tensor var_30391_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3041_cast_fp16)[name = string("op_30391_cast_fp16")]; + tensor var_30392_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3043_cast_fp16)[name = string("op_30392_cast_fp16")]; + tensor var_30393_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3045_cast_fp16)[name = string("op_30393_cast_fp16")]; + tensor var_30394_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3047_cast_fp16)[name = string("op_30394_cast_fp16")]; + tensor var_30395_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3049_cast_fp16)[name = string("op_30395_cast_fp16")]; + tensor var_30396_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3051_cast_fp16)[name = string("op_30396_cast_fp16")]; + tensor var_30397_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3053_cast_fp16)[name = string("op_30397_cast_fp16")]; + tensor var_30398_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3055_cast_fp16)[name = string("op_30398_cast_fp16")]; + tensor var_30399_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3057_cast_fp16)[name = string("op_30399_cast_fp16")]; + tensor var_30400_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3059_cast_fp16)[name = string("op_30400_cast_fp16")]; + tensor var_30401_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3061_cast_fp16)[name = string("op_30401_cast_fp16")]; + tensor var_30402_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3063_cast_fp16)[name = string("op_30402_cast_fp16")]; + tensor var_30403_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3065_cast_fp16)[name = string("op_30403_cast_fp16")]; + tensor var_30404_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3067_cast_fp16)[name = string("op_30404_cast_fp16")]; + tensor var_30405_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3069_cast_fp16)[name = string("op_30405_cast_fp16")]; + tensor var_30406_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3071_cast_fp16)[name = string("op_30406_cast_fp16")]; + tensor var_30407_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3073_cast_fp16)[name = string("op_30407_cast_fp16")]; + tensor var_30408_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3075_cast_fp16)[name = string("op_30408_cast_fp16")]; + tensor var_30409_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3077_cast_fp16)[name = string("op_30409_cast_fp16")]; + tensor var_30410_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3079_cast_fp16)[name = string("op_30410_cast_fp16")]; + tensor var_30411_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3081_cast_fp16)[name = string("op_30411_cast_fp16")]; + tensor var_30412_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3083_cast_fp16)[name = string("op_30412_cast_fp16")]; + tensor var_30413_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3085_cast_fp16)[name = string("op_30413_cast_fp16")]; + tensor var_30414_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3087_cast_fp16)[name = string("op_30414_cast_fp16")]; + tensor var_30415_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3089_cast_fp16)[name = string("op_30415_cast_fp16")]; + tensor var_30416_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3091_cast_fp16)[name = string("op_30416_cast_fp16")]; + tensor var_30417_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3093_cast_fp16)[name = string("op_30417_cast_fp16")]; + tensor var_30418_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3095_cast_fp16)[name = string("op_30418_cast_fp16")]; + tensor var_30419_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3097_cast_fp16)[name = string("op_30419_cast_fp16")]; + tensor var_30420_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3099_cast_fp16)[name = string("op_30420_cast_fp16")]; + tensor var_30421_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3101_cast_fp16)[name = string("op_30421_cast_fp16")]; + tensor var_30422_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3103_cast_fp16)[name = string("op_30422_cast_fp16")]; + tensor var_30423_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3105_cast_fp16)[name = string("op_30423_cast_fp16")]; + tensor var_30424_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3107_cast_fp16)[name = string("op_30424_cast_fp16")]; + tensor var_30425_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3109_cast_fp16)[name = string("op_30425_cast_fp16")]; + tensor var_30426_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3111_cast_fp16)[name = string("op_30426_cast_fp16")]; + tensor var_30427_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3113_cast_fp16)[name = string("op_30427_cast_fp16")]; + tensor var_30428_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3115_cast_fp16)[name = string("op_30428_cast_fp16")]; + tensor var_30429_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3117_cast_fp16)[name = string("op_30429_cast_fp16")]; + tensor var_30430_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3119_cast_fp16)[name = string("op_30430_cast_fp16")]; + tensor var_30431_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3121_cast_fp16)[name = string("op_30431_cast_fp16")]; + tensor var_30432_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3123_cast_fp16)[name = string("op_30432_cast_fp16")]; + tensor var_30433_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3125_cast_fp16)[name = string("op_30433_cast_fp16")]; + tensor var_30434_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3127_cast_fp16)[name = string("op_30434_cast_fp16")]; + tensor var_30435_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3129_cast_fp16)[name = string("op_30435_cast_fp16")]; + tensor var_30436_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3131_cast_fp16)[name = string("op_30436_cast_fp16")]; + tensor var_30437_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3133_cast_fp16)[name = string("op_30437_cast_fp16")]; + tensor var_30438_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3135_cast_fp16)[name = string("op_30438_cast_fp16")]; + tensor var_30439_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3137_cast_fp16)[name = string("op_30439_cast_fp16")]; + tensor var_30440_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3139_cast_fp16)[name = string("op_30440_cast_fp16")]; + tensor var_30441_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3141_cast_fp16)[name = string("op_30441_cast_fp16")]; + tensor var_30442_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3143_cast_fp16)[name = string("op_30442_cast_fp16")]; + tensor var_30443_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3145_cast_fp16)[name = string("op_30443_cast_fp16")]; + tensor var_30444_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3147_cast_fp16)[name = string("op_30444_cast_fp16")]; + tensor var_30445_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3149_cast_fp16)[name = string("op_30445_cast_fp16")]; + tensor var_30446_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3151_cast_fp16)[name = string("op_30446_cast_fp16")]; + tensor var_30447_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3153_cast_fp16)[name = string("op_30447_cast_fp16")]; + tensor var_30448_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3155_cast_fp16)[name = string("op_30448_cast_fp16")]; + tensor var_30449_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3157_cast_fp16)[name = string("op_30449_cast_fp16")]; + tensor var_30450_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3159_cast_fp16)[name = string("op_30450_cast_fp16")]; + tensor var_30451_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3161_cast_fp16)[name = string("op_30451_cast_fp16")]; + tensor var_30452_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3163_cast_fp16)[name = string("op_30452_cast_fp16")]; + tensor var_30453_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3165_cast_fp16)[name = string("op_30453_cast_fp16")]; + tensor var_30454_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3167_cast_fp16)[name = string("op_30454_cast_fp16")]; + tensor var_30455_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3169_cast_fp16)[name = string("op_30455_cast_fp16")]; + tensor var_30456_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3171_cast_fp16)[name = string("op_30456_cast_fp16")]; + tensor var_30457_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3173_cast_fp16)[name = string("op_30457_cast_fp16")]; + tensor var_30458_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3175_cast_fp16)[name = string("op_30458_cast_fp16")]; + tensor var_30459_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3177_cast_fp16)[name = string("op_30459_cast_fp16")]; + tensor var_30460_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3179_cast_fp16)[name = string("op_30460_cast_fp16")]; + tensor var_30461_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3181_cast_fp16)[name = string("op_30461_cast_fp16")]; + tensor var_30462_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3183_cast_fp16)[name = string("op_30462_cast_fp16")]; + tensor var_30463_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3185_cast_fp16)[name = string("op_30463_cast_fp16")]; + tensor var_30464_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3187_cast_fp16)[name = string("op_30464_cast_fp16")]; + tensor var_30465_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3189_cast_fp16)[name = string("op_30465_cast_fp16")]; + tensor var_30466_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3191_cast_fp16)[name = string("op_30466_cast_fp16")]; + tensor var_30467_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3193_cast_fp16)[name = string("op_30467_cast_fp16")]; + tensor var_30468_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3195_cast_fp16)[name = string("op_30468_cast_fp16")]; + tensor var_30469_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3197_cast_fp16)[name = string("op_30469_cast_fp16")]; + tensor var_30470_cast_fp16 = softmax(axis = var_29216, x = aw_chunk_3199_cast_fp16)[name = string("op_30470_cast_fp16")]; + string var_30472_equation_0 = const()[name = string("op_30472_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30472_cast_fp16 = einsum(equation = var_30472_equation_0, values = (var_29992_cast_fp16, var_30391_cast_fp16))[name = string("op_30472_cast_fp16")]; + string var_30474_equation_0 = const()[name = string("op_30474_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30474_cast_fp16 = einsum(equation = var_30474_equation_0, values = (var_29992_cast_fp16, var_30392_cast_fp16))[name = string("op_30474_cast_fp16")]; + string var_30476_equation_0 = const()[name = string("op_30476_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30476_cast_fp16 = einsum(equation = var_30476_equation_0, values = (var_29992_cast_fp16, var_30393_cast_fp16))[name = string("op_30476_cast_fp16")]; + string var_30478_equation_0 = const()[name = string("op_30478_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30478_cast_fp16 = einsum(equation = var_30478_equation_0, values = (var_29992_cast_fp16, var_30394_cast_fp16))[name = string("op_30478_cast_fp16")]; + string var_30480_equation_0 = const()[name = string("op_30480_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30480_cast_fp16 = einsum(equation = var_30480_equation_0, values = (var_29996_cast_fp16, var_30395_cast_fp16))[name = string("op_30480_cast_fp16")]; + string var_30482_equation_0 = const()[name = string("op_30482_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30482_cast_fp16 = einsum(equation = var_30482_equation_0, values = (var_29996_cast_fp16, var_30396_cast_fp16))[name = string("op_30482_cast_fp16")]; + string var_30484_equation_0 = const()[name = string("op_30484_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30484_cast_fp16 = einsum(equation = var_30484_equation_0, values = (var_29996_cast_fp16, var_30397_cast_fp16))[name = string("op_30484_cast_fp16")]; + string var_30486_equation_0 = const()[name = string("op_30486_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30486_cast_fp16 = einsum(equation = var_30486_equation_0, values = (var_29996_cast_fp16, var_30398_cast_fp16))[name = string("op_30486_cast_fp16")]; + string var_30488_equation_0 = const()[name = string("op_30488_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30488_cast_fp16 = einsum(equation = var_30488_equation_0, values = (var_30000_cast_fp16, var_30399_cast_fp16))[name = string("op_30488_cast_fp16")]; + string var_30490_equation_0 = const()[name = string("op_30490_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30490_cast_fp16 = einsum(equation = var_30490_equation_0, values = (var_30000_cast_fp16, var_30400_cast_fp16))[name = string("op_30490_cast_fp16")]; + string var_30492_equation_0 = const()[name = string("op_30492_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30492_cast_fp16 = einsum(equation = var_30492_equation_0, values = (var_30000_cast_fp16, var_30401_cast_fp16))[name = string("op_30492_cast_fp16")]; + string var_30494_equation_0 = const()[name = string("op_30494_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30494_cast_fp16 = einsum(equation = var_30494_equation_0, values = (var_30000_cast_fp16, var_30402_cast_fp16))[name = string("op_30494_cast_fp16")]; + string var_30496_equation_0 = const()[name = string("op_30496_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30496_cast_fp16 = einsum(equation = var_30496_equation_0, values = (var_30004_cast_fp16, var_30403_cast_fp16))[name = string("op_30496_cast_fp16")]; + string var_30498_equation_0 = const()[name = string("op_30498_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30498_cast_fp16 = einsum(equation = var_30498_equation_0, values = (var_30004_cast_fp16, var_30404_cast_fp16))[name = string("op_30498_cast_fp16")]; + string var_30500_equation_0 = const()[name = string("op_30500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30500_cast_fp16 = einsum(equation = var_30500_equation_0, values = (var_30004_cast_fp16, var_30405_cast_fp16))[name = string("op_30500_cast_fp16")]; + string var_30502_equation_0 = const()[name = string("op_30502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30502_cast_fp16 = einsum(equation = var_30502_equation_0, values = (var_30004_cast_fp16, var_30406_cast_fp16))[name = string("op_30502_cast_fp16")]; + string var_30504_equation_0 = const()[name = string("op_30504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30504_cast_fp16 = einsum(equation = var_30504_equation_0, values = (var_30008_cast_fp16, var_30407_cast_fp16))[name = string("op_30504_cast_fp16")]; + string var_30506_equation_0 = const()[name = string("op_30506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30506_cast_fp16 = einsum(equation = var_30506_equation_0, values = (var_30008_cast_fp16, var_30408_cast_fp16))[name = string("op_30506_cast_fp16")]; + string var_30508_equation_0 = const()[name = string("op_30508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30508_cast_fp16 = einsum(equation = var_30508_equation_0, values = (var_30008_cast_fp16, var_30409_cast_fp16))[name = string("op_30508_cast_fp16")]; + string var_30510_equation_0 = const()[name = string("op_30510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30510_cast_fp16 = einsum(equation = var_30510_equation_0, values = (var_30008_cast_fp16, var_30410_cast_fp16))[name = string("op_30510_cast_fp16")]; + string var_30512_equation_0 = const()[name = string("op_30512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30512_cast_fp16 = einsum(equation = var_30512_equation_0, values = (var_30012_cast_fp16, var_30411_cast_fp16))[name = string("op_30512_cast_fp16")]; + string var_30514_equation_0 = const()[name = string("op_30514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30514_cast_fp16 = einsum(equation = var_30514_equation_0, values = (var_30012_cast_fp16, var_30412_cast_fp16))[name = string("op_30514_cast_fp16")]; + string var_30516_equation_0 = const()[name = string("op_30516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30516_cast_fp16 = einsum(equation = var_30516_equation_0, values = (var_30012_cast_fp16, var_30413_cast_fp16))[name = string("op_30516_cast_fp16")]; + string var_30518_equation_0 = const()[name = string("op_30518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30518_cast_fp16 = einsum(equation = var_30518_equation_0, values = (var_30012_cast_fp16, var_30414_cast_fp16))[name = string("op_30518_cast_fp16")]; + string var_30520_equation_0 = const()[name = string("op_30520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30520_cast_fp16 = einsum(equation = var_30520_equation_0, values = (var_30016_cast_fp16, var_30415_cast_fp16))[name = string("op_30520_cast_fp16")]; + string var_30522_equation_0 = const()[name = string("op_30522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30522_cast_fp16 = einsum(equation = var_30522_equation_0, values = (var_30016_cast_fp16, var_30416_cast_fp16))[name = string("op_30522_cast_fp16")]; + string var_30524_equation_0 = const()[name = string("op_30524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30524_cast_fp16 = einsum(equation = var_30524_equation_0, values = (var_30016_cast_fp16, var_30417_cast_fp16))[name = string("op_30524_cast_fp16")]; + string var_30526_equation_0 = const()[name = string("op_30526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30526_cast_fp16 = einsum(equation = var_30526_equation_0, values = (var_30016_cast_fp16, var_30418_cast_fp16))[name = string("op_30526_cast_fp16")]; + string var_30528_equation_0 = const()[name = string("op_30528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30528_cast_fp16 = einsum(equation = var_30528_equation_0, values = (var_30020_cast_fp16, var_30419_cast_fp16))[name = string("op_30528_cast_fp16")]; + string var_30530_equation_0 = const()[name = string("op_30530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30530_cast_fp16 = einsum(equation = var_30530_equation_0, values = (var_30020_cast_fp16, var_30420_cast_fp16))[name = string("op_30530_cast_fp16")]; + string var_30532_equation_0 = const()[name = string("op_30532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30532_cast_fp16 = einsum(equation = var_30532_equation_0, values = (var_30020_cast_fp16, var_30421_cast_fp16))[name = string("op_30532_cast_fp16")]; + string var_30534_equation_0 = const()[name = string("op_30534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30534_cast_fp16 = einsum(equation = var_30534_equation_0, values = (var_30020_cast_fp16, var_30422_cast_fp16))[name = string("op_30534_cast_fp16")]; + string var_30536_equation_0 = const()[name = string("op_30536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30536_cast_fp16 = einsum(equation = var_30536_equation_0, values = (var_30024_cast_fp16, var_30423_cast_fp16))[name = string("op_30536_cast_fp16")]; + string var_30538_equation_0 = const()[name = string("op_30538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30538_cast_fp16 = einsum(equation = var_30538_equation_0, values = (var_30024_cast_fp16, var_30424_cast_fp16))[name = string("op_30538_cast_fp16")]; + string var_30540_equation_0 = const()[name = string("op_30540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30540_cast_fp16 = einsum(equation = var_30540_equation_0, values = (var_30024_cast_fp16, var_30425_cast_fp16))[name = string("op_30540_cast_fp16")]; + string var_30542_equation_0 = const()[name = string("op_30542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30542_cast_fp16 = einsum(equation = var_30542_equation_0, values = (var_30024_cast_fp16, var_30426_cast_fp16))[name = string("op_30542_cast_fp16")]; + string var_30544_equation_0 = const()[name = string("op_30544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30544_cast_fp16 = einsum(equation = var_30544_equation_0, values = (var_30028_cast_fp16, var_30427_cast_fp16))[name = string("op_30544_cast_fp16")]; + string var_30546_equation_0 = const()[name = string("op_30546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30546_cast_fp16 = einsum(equation = var_30546_equation_0, values = (var_30028_cast_fp16, var_30428_cast_fp16))[name = string("op_30546_cast_fp16")]; + string var_30548_equation_0 = const()[name = string("op_30548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30548_cast_fp16 = einsum(equation = var_30548_equation_0, values = (var_30028_cast_fp16, var_30429_cast_fp16))[name = string("op_30548_cast_fp16")]; + string var_30550_equation_0 = const()[name = string("op_30550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30550_cast_fp16 = einsum(equation = var_30550_equation_0, values = (var_30028_cast_fp16, var_30430_cast_fp16))[name = string("op_30550_cast_fp16")]; + string var_30552_equation_0 = const()[name = string("op_30552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30552_cast_fp16 = einsum(equation = var_30552_equation_0, values = (var_30032_cast_fp16, var_30431_cast_fp16))[name = string("op_30552_cast_fp16")]; + string var_30554_equation_0 = const()[name = string("op_30554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30554_cast_fp16 = einsum(equation = var_30554_equation_0, values = (var_30032_cast_fp16, var_30432_cast_fp16))[name = string("op_30554_cast_fp16")]; + string var_30556_equation_0 = const()[name = string("op_30556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30556_cast_fp16 = einsum(equation = var_30556_equation_0, values = (var_30032_cast_fp16, var_30433_cast_fp16))[name = string("op_30556_cast_fp16")]; + string var_30558_equation_0 = const()[name = string("op_30558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30558_cast_fp16 = einsum(equation = var_30558_equation_0, values = (var_30032_cast_fp16, var_30434_cast_fp16))[name = string("op_30558_cast_fp16")]; + string var_30560_equation_0 = const()[name = string("op_30560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30560_cast_fp16 = einsum(equation = var_30560_equation_0, values = (var_30036_cast_fp16, var_30435_cast_fp16))[name = string("op_30560_cast_fp16")]; + string var_30562_equation_0 = const()[name = string("op_30562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30562_cast_fp16 = einsum(equation = var_30562_equation_0, values = (var_30036_cast_fp16, var_30436_cast_fp16))[name = string("op_30562_cast_fp16")]; + string var_30564_equation_0 = const()[name = string("op_30564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30564_cast_fp16 = einsum(equation = var_30564_equation_0, values = (var_30036_cast_fp16, var_30437_cast_fp16))[name = string("op_30564_cast_fp16")]; + string var_30566_equation_0 = const()[name = string("op_30566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30566_cast_fp16 = einsum(equation = var_30566_equation_0, values = (var_30036_cast_fp16, var_30438_cast_fp16))[name = string("op_30566_cast_fp16")]; + string var_30568_equation_0 = const()[name = string("op_30568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30568_cast_fp16 = einsum(equation = var_30568_equation_0, values = (var_30040_cast_fp16, var_30439_cast_fp16))[name = string("op_30568_cast_fp16")]; + string var_30570_equation_0 = const()[name = string("op_30570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30570_cast_fp16 = einsum(equation = var_30570_equation_0, values = (var_30040_cast_fp16, var_30440_cast_fp16))[name = string("op_30570_cast_fp16")]; + string var_30572_equation_0 = const()[name = string("op_30572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30572_cast_fp16 = einsum(equation = var_30572_equation_0, values = (var_30040_cast_fp16, var_30441_cast_fp16))[name = string("op_30572_cast_fp16")]; + string var_30574_equation_0 = const()[name = string("op_30574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30574_cast_fp16 = einsum(equation = var_30574_equation_0, values = (var_30040_cast_fp16, var_30442_cast_fp16))[name = string("op_30574_cast_fp16")]; + string var_30576_equation_0 = const()[name = string("op_30576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30576_cast_fp16 = einsum(equation = var_30576_equation_0, values = (var_30044_cast_fp16, var_30443_cast_fp16))[name = string("op_30576_cast_fp16")]; + string var_30578_equation_0 = const()[name = string("op_30578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30578_cast_fp16 = einsum(equation = var_30578_equation_0, values = (var_30044_cast_fp16, var_30444_cast_fp16))[name = string("op_30578_cast_fp16")]; + string var_30580_equation_0 = const()[name = string("op_30580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30580_cast_fp16 = einsum(equation = var_30580_equation_0, values = (var_30044_cast_fp16, var_30445_cast_fp16))[name = string("op_30580_cast_fp16")]; + string var_30582_equation_0 = const()[name = string("op_30582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30582_cast_fp16 = einsum(equation = var_30582_equation_0, values = (var_30044_cast_fp16, var_30446_cast_fp16))[name = string("op_30582_cast_fp16")]; + string var_30584_equation_0 = const()[name = string("op_30584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30584_cast_fp16 = einsum(equation = var_30584_equation_0, values = (var_30048_cast_fp16, var_30447_cast_fp16))[name = string("op_30584_cast_fp16")]; + string var_30586_equation_0 = const()[name = string("op_30586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30586_cast_fp16 = einsum(equation = var_30586_equation_0, values = (var_30048_cast_fp16, var_30448_cast_fp16))[name = string("op_30586_cast_fp16")]; + string var_30588_equation_0 = const()[name = string("op_30588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30588_cast_fp16 = einsum(equation = var_30588_equation_0, values = (var_30048_cast_fp16, var_30449_cast_fp16))[name = string("op_30588_cast_fp16")]; + string var_30590_equation_0 = const()[name = string("op_30590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30590_cast_fp16 = einsum(equation = var_30590_equation_0, values = (var_30048_cast_fp16, var_30450_cast_fp16))[name = string("op_30590_cast_fp16")]; + string var_30592_equation_0 = const()[name = string("op_30592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30592_cast_fp16 = einsum(equation = var_30592_equation_0, values = (var_30052_cast_fp16, var_30451_cast_fp16))[name = string("op_30592_cast_fp16")]; + string var_30594_equation_0 = const()[name = string("op_30594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30594_cast_fp16 = einsum(equation = var_30594_equation_0, values = (var_30052_cast_fp16, var_30452_cast_fp16))[name = string("op_30594_cast_fp16")]; + string var_30596_equation_0 = const()[name = string("op_30596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30596_cast_fp16 = einsum(equation = var_30596_equation_0, values = (var_30052_cast_fp16, var_30453_cast_fp16))[name = string("op_30596_cast_fp16")]; + string var_30598_equation_0 = const()[name = string("op_30598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30598_cast_fp16 = einsum(equation = var_30598_equation_0, values = (var_30052_cast_fp16, var_30454_cast_fp16))[name = string("op_30598_cast_fp16")]; + string var_30600_equation_0 = const()[name = string("op_30600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30600_cast_fp16 = einsum(equation = var_30600_equation_0, values = (var_30056_cast_fp16, var_30455_cast_fp16))[name = string("op_30600_cast_fp16")]; + string var_30602_equation_0 = const()[name = string("op_30602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30602_cast_fp16 = einsum(equation = var_30602_equation_0, values = (var_30056_cast_fp16, var_30456_cast_fp16))[name = string("op_30602_cast_fp16")]; + string var_30604_equation_0 = const()[name = string("op_30604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30604_cast_fp16 = einsum(equation = var_30604_equation_0, values = (var_30056_cast_fp16, var_30457_cast_fp16))[name = string("op_30604_cast_fp16")]; + string var_30606_equation_0 = const()[name = string("op_30606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30606_cast_fp16 = einsum(equation = var_30606_equation_0, values = (var_30056_cast_fp16, var_30458_cast_fp16))[name = string("op_30606_cast_fp16")]; + string var_30608_equation_0 = const()[name = string("op_30608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30608_cast_fp16 = einsum(equation = var_30608_equation_0, values = (var_30060_cast_fp16, var_30459_cast_fp16))[name = string("op_30608_cast_fp16")]; + string var_30610_equation_0 = const()[name = string("op_30610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30610_cast_fp16 = einsum(equation = var_30610_equation_0, values = (var_30060_cast_fp16, var_30460_cast_fp16))[name = string("op_30610_cast_fp16")]; + string var_30612_equation_0 = const()[name = string("op_30612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30612_cast_fp16 = einsum(equation = var_30612_equation_0, values = (var_30060_cast_fp16, var_30461_cast_fp16))[name = string("op_30612_cast_fp16")]; + string var_30614_equation_0 = const()[name = string("op_30614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30614_cast_fp16 = einsum(equation = var_30614_equation_0, values = (var_30060_cast_fp16, var_30462_cast_fp16))[name = string("op_30614_cast_fp16")]; + string var_30616_equation_0 = const()[name = string("op_30616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30616_cast_fp16 = einsum(equation = var_30616_equation_0, values = (var_30064_cast_fp16, var_30463_cast_fp16))[name = string("op_30616_cast_fp16")]; + string var_30618_equation_0 = const()[name = string("op_30618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30618_cast_fp16 = einsum(equation = var_30618_equation_0, values = (var_30064_cast_fp16, var_30464_cast_fp16))[name = string("op_30618_cast_fp16")]; + string var_30620_equation_0 = const()[name = string("op_30620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30620_cast_fp16 = einsum(equation = var_30620_equation_0, values = (var_30064_cast_fp16, var_30465_cast_fp16))[name = string("op_30620_cast_fp16")]; + string var_30622_equation_0 = const()[name = string("op_30622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30622_cast_fp16 = einsum(equation = var_30622_equation_0, values = (var_30064_cast_fp16, var_30466_cast_fp16))[name = string("op_30622_cast_fp16")]; + string var_30624_equation_0 = const()[name = string("op_30624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30624_cast_fp16 = einsum(equation = var_30624_equation_0, values = (var_30068_cast_fp16, var_30467_cast_fp16))[name = string("op_30624_cast_fp16")]; + string var_30626_equation_0 = const()[name = string("op_30626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30626_cast_fp16 = einsum(equation = var_30626_equation_0, values = (var_30068_cast_fp16, var_30468_cast_fp16))[name = string("op_30626_cast_fp16")]; + string var_30628_equation_0 = const()[name = string("op_30628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30628_cast_fp16 = einsum(equation = var_30628_equation_0, values = (var_30068_cast_fp16, var_30469_cast_fp16))[name = string("op_30628_cast_fp16")]; + string var_30630_equation_0 = const()[name = string("op_30630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30630_cast_fp16 = einsum(equation = var_30630_equation_0, values = (var_30068_cast_fp16, var_30470_cast_fp16))[name = string("op_30630_cast_fp16")]; + bool var_30632_interleave_0 = const()[name = string("op_30632_interleave_0"), val = bool(false)]; + tensor var_30632_cast_fp16 = concat(axis = var_29191, interleave = var_30632_interleave_0, values = (var_30472_cast_fp16, var_30474_cast_fp16, var_30476_cast_fp16, var_30478_cast_fp16))[name = string("op_30632_cast_fp16")]; + bool var_30634_interleave_0 = const()[name = string("op_30634_interleave_0"), val = bool(false)]; + tensor var_30634_cast_fp16 = concat(axis = var_29191, interleave = var_30634_interleave_0, values = (var_30480_cast_fp16, var_30482_cast_fp16, var_30484_cast_fp16, var_30486_cast_fp16))[name = string("op_30634_cast_fp16")]; + bool var_30636_interleave_0 = const()[name = string("op_30636_interleave_0"), val = bool(false)]; + tensor var_30636_cast_fp16 = concat(axis = var_29191, interleave = var_30636_interleave_0, values = (var_30488_cast_fp16, var_30490_cast_fp16, var_30492_cast_fp16, var_30494_cast_fp16))[name = string("op_30636_cast_fp16")]; + bool var_30638_interleave_0 = const()[name = string("op_30638_interleave_0"), val = bool(false)]; + tensor var_30638_cast_fp16 = concat(axis = var_29191, interleave = var_30638_interleave_0, values = (var_30496_cast_fp16, var_30498_cast_fp16, var_30500_cast_fp16, var_30502_cast_fp16))[name = string("op_30638_cast_fp16")]; + bool var_30640_interleave_0 = const()[name = string("op_30640_interleave_0"), val = bool(false)]; + tensor var_30640_cast_fp16 = concat(axis = var_29191, interleave = var_30640_interleave_0, values = (var_30504_cast_fp16, var_30506_cast_fp16, var_30508_cast_fp16, var_30510_cast_fp16))[name = string("op_30640_cast_fp16")]; + bool var_30642_interleave_0 = const()[name = string("op_30642_interleave_0"), val = bool(false)]; + tensor var_30642_cast_fp16 = concat(axis = var_29191, interleave = var_30642_interleave_0, values = (var_30512_cast_fp16, var_30514_cast_fp16, var_30516_cast_fp16, var_30518_cast_fp16))[name = string("op_30642_cast_fp16")]; + bool var_30644_interleave_0 = const()[name = string("op_30644_interleave_0"), val = bool(false)]; + tensor var_30644_cast_fp16 = concat(axis = var_29191, interleave = var_30644_interleave_0, values = (var_30520_cast_fp16, var_30522_cast_fp16, var_30524_cast_fp16, var_30526_cast_fp16))[name = string("op_30644_cast_fp16")]; + bool var_30646_interleave_0 = const()[name = string("op_30646_interleave_0"), val = bool(false)]; + tensor var_30646_cast_fp16 = concat(axis = var_29191, interleave = var_30646_interleave_0, values = (var_30528_cast_fp16, var_30530_cast_fp16, var_30532_cast_fp16, var_30534_cast_fp16))[name = string("op_30646_cast_fp16")]; + bool var_30648_interleave_0 = const()[name = string("op_30648_interleave_0"), val = bool(false)]; + tensor var_30648_cast_fp16 = concat(axis = var_29191, interleave = var_30648_interleave_0, values = (var_30536_cast_fp16, var_30538_cast_fp16, var_30540_cast_fp16, var_30542_cast_fp16))[name = string("op_30648_cast_fp16")]; + bool var_30650_interleave_0 = const()[name = string("op_30650_interleave_0"), val = bool(false)]; + tensor var_30650_cast_fp16 = concat(axis = var_29191, interleave = var_30650_interleave_0, values = (var_30544_cast_fp16, var_30546_cast_fp16, var_30548_cast_fp16, var_30550_cast_fp16))[name = string("op_30650_cast_fp16")]; + bool var_30652_interleave_0 = const()[name = string("op_30652_interleave_0"), val = bool(false)]; + tensor var_30652_cast_fp16 = concat(axis = var_29191, interleave = var_30652_interleave_0, values = (var_30552_cast_fp16, var_30554_cast_fp16, var_30556_cast_fp16, var_30558_cast_fp16))[name = string("op_30652_cast_fp16")]; + bool var_30654_interleave_0 = const()[name = string("op_30654_interleave_0"), val = bool(false)]; + tensor var_30654_cast_fp16 = concat(axis = var_29191, interleave = var_30654_interleave_0, values = (var_30560_cast_fp16, var_30562_cast_fp16, var_30564_cast_fp16, var_30566_cast_fp16))[name = string("op_30654_cast_fp16")]; + bool var_30656_interleave_0 = const()[name = string("op_30656_interleave_0"), val = bool(false)]; + tensor var_30656_cast_fp16 = concat(axis = var_29191, interleave = var_30656_interleave_0, values = (var_30568_cast_fp16, var_30570_cast_fp16, var_30572_cast_fp16, var_30574_cast_fp16))[name = string("op_30656_cast_fp16")]; + bool var_30658_interleave_0 = const()[name = string("op_30658_interleave_0"), val = bool(false)]; + tensor var_30658_cast_fp16 = concat(axis = var_29191, interleave = var_30658_interleave_0, values = (var_30576_cast_fp16, var_30578_cast_fp16, var_30580_cast_fp16, var_30582_cast_fp16))[name = string("op_30658_cast_fp16")]; + bool var_30660_interleave_0 = const()[name = string("op_30660_interleave_0"), val = bool(false)]; + tensor var_30660_cast_fp16 = concat(axis = var_29191, interleave = var_30660_interleave_0, values = (var_30584_cast_fp16, var_30586_cast_fp16, var_30588_cast_fp16, var_30590_cast_fp16))[name = string("op_30660_cast_fp16")]; + bool var_30662_interleave_0 = const()[name = string("op_30662_interleave_0"), val = bool(false)]; + tensor var_30662_cast_fp16 = concat(axis = var_29191, interleave = var_30662_interleave_0, values = (var_30592_cast_fp16, var_30594_cast_fp16, var_30596_cast_fp16, var_30598_cast_fp16))[name = string("op_30662_cast_fp16")]; + bool var_30664_interleave_0 = const()[name = string("op_30664_interleave_0"), val = bool(false)]; + tensor var_30664_cast_fp16 = concat(axis = var_29191, interleave = var_30664_interleave_0, values = (var_30600_cast_fp16, var_30602_cast_fp16, var_30604_cast_fp16, var_30606_cast_fp16))[name = string("op_30664_cast_fp16")]; + bool var_30666_interleave_0 = const()[name = string("op_30666_interleave_0"), val = bool(false)]; + tensor var_30666_cast_fp16 = concat(axis = var_29191, interleave = var_30666_interleave_0, values = (var_30608_cast_fp16, var_30610_cast_fp16, var_30612_cast_fp16, var_30614_cast_fp16))[name = string("op_30666_cast_fp16")]; + bool var_30668_interleave_0 = const()[name = string("op_30668_interleave_0"), val = bool(false)]; + tensor var_30668_cast_fp16 = concat(axis = var_29191, interleave = var_30668_interleave_0, values = (var_30616_cast_fp16, var_30618_cast_fp16, var_30620_cast_fp16, var_30622_cast_fp16))[name = string("op_30668_cast_fp16")]; + bool var_30670_interleave_0 = const()[name = string("op_30670_interleave_0"), val = bool(false)]; + tensor var_30670_cast_fp16 = concat(axis = var_29191, interleave = var_30670_interleave_0, values = (var_30624_cast_fp16, var_30626_cast_fp16, var_30628_cast_fp16, var_30630_cast_fp16))[name = string("op_30670_cast_fp16")]; + bool input_153_interleave_0 = const()[name = string("input_153_interleave_0"), val = bool(false)]; + tensor input_153_cast_fp16 = concat(axis = var_29216, interleave = input_153_interleave_0, values = (var_30632_cast_fp16, var_30634_cast_fp16, var_30636_cast_fp16, var_30638_cast_fp16, var_30640_cast_fp16, var_30642_cast_fp16, var_30644_cast_fp16, var_30646_cast_fp16, var_30648_cast_fp16, var_30650_cast_fp16, var_30652_cast_fp16, var_30654_cast_fp16, var_30656_cast_fp16, var_30658_cast_fp16, var_30660_cast_fp16, var_30662_cast_fp16, var_30664_cast_fp16, var_30666_cast_fp16, var_30668_cast_fp16, var_30670_cast_fp16))[name = string("input_153_cast_fp16")]; + string obj_79_pad_type_0 = const()[name = string("obj_79_pad_type_0"), val = string("valid")]; + tensor obj_79_strides_0 = const()[name = string("obj_79_strides_0"), val = tensor([1, 1])]; + tensor obj_79_pad_0 = const()[name = string("obj_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_79_dilations_0 = const()[name = string("obj_79_dilations_0"), val = tensor([1, 1])]; + int32 obj_79_groups_0 = const()[name = string("obj_79_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772217600)))]; + tensor layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775494464)))]; + tensor obj_79_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_79_dilations_0, groups = obj_79_groups_0, pad = obj_79_pad_0, pad_type = obj_79_pad_type_0, strides = obj_79_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = string("obj_79_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = obj_79_cast_fp16)[name = string("inputs_79_cast_fp16")]; + tensor out_79_axes_0 = const()[name = string("out_79_axes_0"), val = tensor([1])]; + fp16 var_30689_to_fp16 = const()[name = string("op_30689_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_30689_to_fp16, x = inputs_79_cast_fp16)[name = string("out_79_cast_fp16")]; + tensor input_155_gamma_0_to_fp16 = const()[name = string("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775497088)))]; + tensor input_155_beta_0_to_fp16 = const()[name = string("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775499712)))]; + fp16 input_155_epsilon_0_to_fp16 = const()[name = string("input_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = string("input_155_cast_fp16")]; + string input_157_pad_type_0 = const()[name = string("input_157_pad_type_0"), val = string("valid")]; + tensor input_157_strides_0 = const()[name = string("input_157_strides_0"), val = tensor([1, 1])]; + tensor input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_157_dilations_0 = const()[name = string("input_157_dilations_0"), val = tensor([1, 1])]; + int32 input_157_groups_0 = const()[name = string("input_157_groups_0"), val = int32(1)]; + tensor layers_19_fc1_weight_to_fp16 = const()[name = string("layers_19_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775502336)))]; + tensor layers_19_fc1_bias_to_fp16 = const()[name = string("layers_19_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788609600)))]; + tensor input_157_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_19_fc1_weight_to_fp16, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")]; + string input_159_mode_0 = const()[name = string("input_159_mode_0"), val = string("EXACT")]; + tensor input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")]; + string hidden_states_43_pad_type_0 = const()[name = string("hidden_states_43_pad_type_0"), val = string("valid")]; + tensor hidden_states_43_strides_0 = const()[name = string("hidden_states_43_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_43_pad_0 = const()[name = string("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_43_dilations_0 = const()[name = string("hidden_states_43_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_43_groups_0 = const()[name = string("hidden_states_43_groups_0"), val = int32(1)]; + tensor layers_19_fc2_weight_to_fp16 = const()[name = string("layers_19_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788619904)))]; + tensor layers_19_fc2_bias_to_fp16 = const()[name = string("layers_19_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801727168)))]; + tensor hidden_states_43_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_19_fc2_weight_to_fp16, x = input_159_cast_fp16)[name = string("hidden_states_43_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = hidden_states_43_cast_fp16)[name = string("inputs_81_cast_fp16")]; + int32 var_30718 = const()[name = string("op_30718"), val = int32(3)]; + int32 var_30743 = const()[name = string("op_30743"), val = int32(1)]; + tensor out_81_axes_0 = const()[name = string("out_81_axes_0"), val = tensor([1])]; + fp16 var_30760_to_fp16 = const()[name = string("op_30760_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_30760_to_fp16, x = inputs_81_cast_fp16)[name = string("out_81_cast_fp16")]; + tensor obj_81_gamma_0_to_fp16 = const()[name = string("obj_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801729792)))]; + tensor obj_81_beta_0_to_fp16 = const()[name = string("obj_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801732416)))]; + fp16 obj_81_epsilon_0_to_fp16 = const()[name = string("obj_81_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_81_cast_fp16 = batch_norm(beta = obj_81_beta_0_to_fp16, epsilon = obj_81_epsilon_0_to_fp16, gamma = obj_81_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = string("obj_81_cast_fp16")]; + string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; + tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; + tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; + int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801735040)))]; + tensor layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805011904)))]; + tensor query_41_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("query_41_cast_fp16")]; + string key_41_pad_type_0 = const()[name = string("key_41_pad_type_0"), val = string("valid")]; + tensor key_41_strides_0 = const()[name = string("key_41_strides_0"), val = tensor([1, 1])]; + tensor key_41_pad_0 = const()[name = string("key_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_41_dilations_0 = const()[name = string("key_41_dilations_0"), val = tensor([1, 1])]; + int32 key_41_groups_0 = const()[name = string("key_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805014528)))]; + tensor key_41_cast_fp16 = conv(dilations = key_41_dilations_0, groups = key_41_groups_0, pad = key_41_pad_0, pad_type = key_41_pad_type_0, strides = key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("key_41_cast_fp16")]; + string value_41_pad_type_0 = const()[name = string("value_41_pad_type_0"), val = string("valid")]; + tensor value_41_strides_0 = const()[name = string("value_41_strides_0"), val = tensor([1, 1])]; + tensor value_41_pad_0 = const()[name = string("value_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_41_dilations_0 = const()[name = string("value_41_dilations_0"), val = tensor([1, 1])]; + int32 value_41_groups_0 = const()[name = string("value_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(808291392)))]; + tensor layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811568256)))]; + tensor value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = value_41_dilations_0, groups = value_41_groups_0, pad = value_41_pad_0, pad_type = value_41_pad_type_0, strides = value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("value_41_cast_fp16")]; + tensor var_30798_begin_0 = const()[name = string("op_30798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30798_end_0 = const()[name = string("op_30798_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30798_end_mask_0 = const()[name = string("op_30798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30798_cast_fp16 = slice_by_index(begin = var_30798_begin_0, end = var_30798_end_0, end_mask = var_30798_end_mask_0, x = query_41_cast_fp16)[name = string("op_30798_cast_fp16")]; + tensor var_30802_begin_0 = const()[name = string("op_30802_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30802_end_0 = const()[name = string("op_30802_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_30802_end_mask_0 = const()[name = string("op_30802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30802_cast_fp16 = slice_by_index(begin = var_30802_begin_0, end = var_30802_end_0, end_mask = var_30802_end_mask_0, x = query_41_cast_fp16)[name = string("op_30802_cast_fp16")]; + tensor var_30806_begin_0 = const()[name = string("op_30806_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30806_end_0 = const()[name = string("op_30806_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_30806_end_mask_0 = const()[name = string("op_30806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30806_cast_fp16 = slice_by_index(begin = var_30806_begin_0, end = var_30806_end_0, end_mask = var_30806_end_mask_0, x = query_41_cast_fp16)[name = string("op_30806_cast_fp16")]; + tensor var_30810_begin_0 = const()[name = string("op_30810_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30810_end_0 = const()[name = string("op_30810_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_30810_end_mask_0 = const()[name = string("op_30810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30810_cast_fp16 = slice_by_index(begin = var_30810_begin_0, end = var_30810_end_0, end_mask = var_30810_end_mask_0, x = query_41_cast_fp16)[name = string("op_30810_cast_fp16")]; + tensor var_30814_begin_0 = const()[name = string("op_30814_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30814_end_0 = const()[name = string("op_30814_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_30814_end_mask_0 = const()[name = string("op_30814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30814_cast_fp16 = slice_by_index(begin = var_30814_begin_0, end = var_30814_end_0, end_mask = var_30814_end_mask_0, x = query_41_cast_fp16)[name = string("op_30814_cast_fp16")]; + tensor var_30818_begin_0 = const()[name = string("op_30818_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30818_end_0 = const()[name = string("op_30818_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_30818_end_mask_0 = const()[name = string("op_30818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30818_cast_fp16 = slice_by_index(begin = var_30818_begin_0, end = var_30818_end_0, end_mask = var_30818_end_mask_0, x = query_41_cast_fp16)[name = string("op_30818_cast_fp16")]; + tensor var_30822_begin_0 = const()[name = string("op_30822_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30822_end_0 = const()[name = string("op_30822_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_30822_end_mask_0 = const()[name = string("op_30822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30822_cast_fp16 = slice_by_index(begin = var_30822_begin_0, end = var_30822_end_0, end_mask = var_30822_end_mask_0, x = query_41_cast_fp16)[name = string("op_30822_cast_fp16")]; + tensor var_30826_begin_0 = const()[name = string("op_30826_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30826_end_0 = const()[name = string("op_30826_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_30826_end_mask_0 = const()[name = string("op_30826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30826_cast_fp16 = slice_by_index(begin = var_30826_begin_0, end = var_30826_end_0, end_mask = var_30826_end_mask_0, x = query_41_cast_fp16)[name = string("op_30826_cast_fp16")]; + tensor var_30830_begin_0 = const()[name = string("op_30830_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30830_end_0 = const()[name = string("op_30830_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_30830_end_mask_0 = const()[name = string("op_30830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30830_cast_fp16 = slice_by_index(begin = var_30830_begin_0, end = var_30830_end_0, end_mask = var_30830_end_mask_0, x = query_41_cast_fp16)[name = string("op_30830_cast_fp16")]; + tensor var_30834_begin_0 = const()[name = string("op_30834_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30834_end_0 = const()[name = string("op_30834_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_30834_end_mask_0 = const()[name = string("op_30834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30834_cast_fp16 = slice_by_index(begin = var_30834_begin_0, end = var_30834_end_0, end_mask = var_30834_end_mask_0, x = query_41_cast_fp16)[name = string("op_30834_cast_fp16")]; + tensor var_30838_begin_0 = const()[name = string("op_30838_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30838_end_0 = const()[name = string("op_30838_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_30838_end_mask_0 = const()[name = string("op_30838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30838_cast_fp16 = slice_by_index(begin = var_30838_begin_0, end = var_30838_end_0, end_mask = var_30838_end_mask_0, x = query_41_cast_fp16)[name = string("op_30838_cast_fp16")]; + tensor var_30842_begin_0 = const()[name = string("op_30842_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30842_end_0 = const()[name = string("op_30842_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_30842_end_mask_0 = const()[name = string("op_30842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30842_cast_fp16 = slice_by_index(begin = var_30842_begin_0, end = var_30842_end_0, end_mask = var_30842_end_mask_0, x = query_41_cast_fp16)[name = string("op_30842_cast_fp16")]; + tensor var_30846_begin_0 = const()[name = string("op_30846_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30846_end_0 = const()[name = string("op_30846_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_30846_end_mask_0 = const()[name = string("op_30846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30846_cast_fp16 = slice_by_index(begin = var_30846_begin_0, end = var_30846_end_0, end_mask = var_30846_end_mask_0, x = query_41_cast_fp16)[name = string("op_30846_cast_fp16")]; + tensor var_30850_begin_0 = const()[name = string("op_30850_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30850_end_0 = const()[name = string("op_30850_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_30850_end_mask_0 = const()[name = string("op_30850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30850_cast_fp16 = slice_by_index(begin = var_30850_begin_0, end = var_30850_end_0, end_mask = var_30850_end_mask_0, x = query_41_cast_fp16)[name = string("op_30850_cast_fp16")]; + tensor var_30854_begin_0 = const()[name = string("op_30854_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30854_end_0 = const()[name = string("op_30854_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_30854_end_mask_0 = const()[name = string("op_30854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30854_cast_fp16 = slice_by_index(begin = var_30854_begin_0, end = var_30854_end_0, end_mask = var_30854_end_mask_0, x = query_41_cast_fp16)[name = string("op_30854_cast_fp16")]; + tensor var_30858_begin_0 = const()[name = string("op_30858_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30858_end_0 = const()[name = string("op_30858_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_30858_end_mask_0 = const()[name = string("op_30858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30858_cast_fp16 = slice_by_index(begin = var_30858_begin_0, end = var_30858_end_0, end_mask = var_30858_end_mask_0, x = query_41_cast_fp16)[name = string("op_30858_cast_fp16")]; + tensor var_30862_begin_0 = const()[name = string("op_30862_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30862_end_0 = const()[name = string("op_30862_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_30862_end_mask_0 = const()[name = string("op_30862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30862_cast_fp16 = slice_by_index(begin = var_30862_begin_0, end = var_30862_end_0, end_mask = var_30862_end_mask_0, x = query_41_cast_fp16)[name = string("op_30862_cast_fp16")]; + tensor var_30866_begin_0 = const()[name = string("op_30866_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30866_end_0 = const()[name = string("op_30866_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_30866_end_mask_0 = const()[name = string("op_30866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30866_cast_fp16 = slice_by_index(begin = var_30866_begin_0, end = var_30866_end_0, end_mask = var_30866_end_mask_0, x = query_41_cast_fp16)[name = string("op_30866_cast_fp16")]; + tensor var_30870_begin_0 = const()[name = string("op_30870_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30870_end_0 = const()[name = string("op_30870_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_30870_end_mask_0 = const()[name = string("op_30870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30870_cast_fp16 = slice_by_index(begin = var_30870_begin_0, end = var_30870_end_0, end_mask = var_30870_end_mask_0, x = query_41_cast_fp16)[name = string("op_30870_cast_fp16")]; + tensor var_30874_begin_0 = const()[name = string("op_30874_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30874_end_0 = const()[name = string("op_30874_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_30874_end_mask_0 = const()[name = string("op_30874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30874_cast_fp16 = slice_by_index(begin = var_30874_begin_0, end = var_30874_end_0, end_mask = var_30874_end_mask_0, x = query_41_cast_fp16)[name = string("op_30874_cast_fp16")]; + tensor var_30883_begin_0 = const()[name = string("op_30883_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30883_end_0 = const()[name = string("op_30883_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30883_end_mask_0 = const()[name = string("op_30883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30883_cast_fp16 = slice_by_index(begin = var_30883_begin_0, end = var_30883_end_0, end_mask = var_30883_end_mask_0, x = var_30798_cast_fp16)[name = string("op_30883_cast_fp16")]; + tensor var_30890_begin_0 = const()[name = string("op_30890_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30890_end_0 = const()[name = string("op_30890_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30890_end_mask_0 = const()[name = string("op_30890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30890_cast_fp16 = slice_by_index(begin = var_30890_begin_0, end = var_30890_end_0, end_mask = var_30890_end_mask_0, x = var_30798_cast_fp16)[name = string("op_30890_cast_fp16")]; + tensor var_30897_begin_0 = const()[name = string("op_30897_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30897_end_0 = const()[name = string("op_30897_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30897_end_mask_0 = const()[name = string("op_30897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30897_cast_fp16 = slice_by_index(begin = var_30897_begin_0, end = var_30897_end_0, end_mask = var_30897_end_mask_0, x = var_30798_cast_fp16)[name = string("op_30897_cast_fp16")]; + tensor var_30904_begin_0 = const()[name = string("op_30904_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30904_end_0 = const()[name = string("op_30904_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30904_end_mask_0 = const()[name = string("op_30904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30904_cast_fp16 = slice_by_index(begin = var_30904_begin_0, end = var_30904_end_0, end_mask = var_30904_end_mask_0, x = var_30798_cast_fp16)[name = string("op_30904_cast_fp16")]; + tensor var_30911_begin_0 = const()[name = string("op_30911_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30911_end_0 = const()[name = string("op_30911_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30911_end_mask_0 = const()[name = string("op_30911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30911_cast_fp16 = slice_by_index(begin = var_30911_begin_0, end = var_30911_end_0, end_mask = var_30911_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30911_cast_fp16")]; + tensor var_30918_begin_0 = const()[name = string("op_30918_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30918_end_0 = const()[name = string("op_30918_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30918_end_mask_0 = const()[name = string("op_30918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30918_cast_fp16 = slice_by_index(begin = var_30918_begin_0, end = var_30918_end_0, end_mask = var_30918_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30918_cast_fp16")]; + tensor var_30925_begin_0 = const()[name = string("op_30925_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30925_end_0 = const()[name = string("op_30925_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30925_end_mask_0 = const()[name = string("op_30925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30925_cast_fp16 = slice_by_index(begin = var_30925_begin_0, end = var_30925_end_0, end_mask = var_30925_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30925_cast_fp16")]; + tensor var_30932_begin_0 = const()[name = string("op_30932_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30932_end_0 = const()[name = string("op_30932_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30932_end_mask_0 = const()[name = string("op_30932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30932_cast_fp16 = slice_by_index(begin = var_30932_begin_0, end = var_30932_end_0, end_mask = var_30932_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30932_cast_fp16")]; + tensor var_30939_begin_0 = const()[name = string("op_30939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30939_end_0 = const()[name = string("op_30939_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30939_end_mask_0 = const()[name = string("op_30939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30939_cast_fp16 = slice_by_index(begin = var_30939_begin_0, end = var_30939_end_0, end_mask = var_30939_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30939_cast_fp16")]; + tensor var_30946_begin_0 = const()[name = string("op_30946_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30946_end_0 = const()[name = string("op_30946_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30946_end_mask_0 = const()[name = string("op_30946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30946_cast_fp16 = slice_by_index(begin = var_30946_begin_0, end = var_30946_end_0, end_mask = var_30946_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30946_cast_fp16")]; + tensor var_30953_begin_0 = const()[name = string("op_30953_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30953_end_0 = const()[name = string("op_30953_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30953_end_mask_0 = const()[name = string("op_30953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30953_cast_fp16 = slice_by_index(begin = var_30953_begin_0, end = var_30953_end_0, end_mask = var_30953_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30953_cast_fp16")]; + tensor var_30960_begin_0 = const()[name = string("op_30960_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30960_end_0 = const()[name = string("op_30960_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30960_end_mask_0 = const()[name = string("op_30960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30960_cast_fp16 = slice_by_index(begin = var_30960_begin_0, end = var_30960_end_0, end_mask = var_30960_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30960_cast_fp16")]; + tensor var_30967_begin_0 = const()[name = string("op_30967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30967_end_0 = const()[name = string("op_30967_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30967_end_mask_0 = const()[name = string("op_30967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30967_cast_fp16 = slice_by_index(begin = var_30967_begin_0, end = var_30967_end_0, end_mask = var_30967_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30967_cast_fp16")]; + tensor var_30974_begin_0 = const()[name = string("op_30974_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30974_end_0 = const()[name = string("op_30974_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30974_end_mask_0 = const()[name = string("op_30974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30974_cast_fp16 = slice_by_index(begin = var_30974_begin_0, end = var_30974_end_0, end_mask = var_30974_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30974_cast_fp16")]; + tensor var_30981_begin_0 = const()[name = string("op_30981_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30981_end_0 = const()[name = string("op_30981_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30981_end_mask_0 = const()[name = string("op_30981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30981_cast_fp16 = slice_by_index(begin = var_30981_begin_0, end = var_30981_end_0, end_mask = var_30981_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30981_cast_fp16")]; + tensor var_30988_begin_0 = const()[name = string("op_30988_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30988_end_0 = const()[name = string("op_30988_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30988_end_mask_0 = const()[name = string("op_30988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30988_cast_fp16 = slice_by_index(begin = var_30988_begin_0, end = var_30988_end_0, end_mask = var_30988_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30988_cast_fp16")]; + tensor var_30995_begin_0 = const()[name = string("op_30995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30995_end_0 = const()[name = string("op_30995_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30995_end_mask_0 = const()[name = string("op_30995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30995_cast_fp16 = slice_by_index(begin = var_30995_begin_0, end = var_30995_end_0, end_mask = var_30995_end_mask_0, x = var_30814_cast_fp16)[name = string("op_30995_cast_fp16")]; + tensor var_31002_begin_0 = const()[name = string("op_31002_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31002_end_0 = const()[name = string("op_31002_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31002_end_mask_0 = const()[name = string("op_31002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31002_cast_fp16 = slice_by_index(begin = var_31002_begin_0, end = var_31002_end_0, end_mask = var_31002_end_mask_0, x = var_30814_cast_fp16)[name = string("op_31002_cast_fp16")]; + tensor var_31009_begin_0 = const()[name = string("op_31009_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31009_end_0 = const()[name = string("op_31009_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31009_end_mask_0 = const()[name = string("op_31009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31009_cast_fp16 = slice_by_index(begin = var_31009_begin_0, end = var_31009_end_0, end_mask = var_31009_end_mask_0, x = var_30814_cast_fp16)[name = string("op_31009_cast_fp16")]; + tensor var_31016_begin_0 = const()[name = string("op_31016_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31016_end_0 = const()[name = string("op_31016_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31016_end_mask_0 = const()[name = string("op_31016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31016_cast_fp16 = slice_by_index(begin = var_31016_begin_0, end = var_31016_end_0, end_mask = var_31016_end_mask_0, x = var_30814_cast_fp16)[name = string("op_31016_cast_fp16")]; + tensor var_31023_begin_0 = const()[name = string("op_31023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31023_end_0 = const()[name = string("op_31023_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31023_end_mask_0 = const()[name = string("op_31023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31023_cast_fp16 = slice_by_index(begin = var_31023_begin_0, end = var_31023_end_0, end_mask = var_31023_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31023_cast_fp16")]; + tensor var_31030_begin_0 = const()[name = string("op_31030_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31030_end_0 = const()[name = string("op_31030_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31030_end_mask_0 = const()[name = string("op_31030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31030_cast_fp16 = slice_by_index(begin = var_31030_begin_0, end = var_31030_end_0, end_mask = var_31030_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31030_cast_fp16")]; + tensor var_31037_begin_0 = const()[name = string("op_31037_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31037_end_0 = const()[name = string("op_31037_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31037_end_mask_0 = const()[name = string("op_31037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31037_cast_fp16 = slice_by_index(begin = var_31037_begin_0, end = var_31037_end_0, end_mask = var_31037_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31037_cast_fp16")]; + tensor var_31044_begin_0 = const()[name = string("op_31044_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31044_end_0 = const()[name = string("op_31044_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31044_end_mask_0 = const()[name = string("op_31044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31044_cast_fp16 = slice_by_index(begin = var_31044_begin_0, end = var_31044_end_0, end_mask = var_31044_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31044_cast_fp16")]; + tensor var_31051_begin_0 = const()[name = string("op_31051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31051_end_0 = const()[name = string("op_31051_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31051_end_mask_0 = const()[name = string("op_31051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31051_cast_fp16 = slice_by_index(begin = var_31051_begin_0, end = var_31051_end_0, end_mask = var_31051_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31051_cast_fp16")]; + tensor var_31058_begin_0 = const()[name = string("op_31058_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31058_end_0 = const()[name = string("op_31058_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31058_end_mask_0 = const()[name = string("op_31058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31058_cast_fp16 = slice_by_index(begin = var_31058_begin_0, end = var_31058_end_0, end_mask = var_31058_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31058_cast_fp16")]; + tensor var_31065_begin_0 = const()[name = string("op_31065_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31065_end_0 = const()[name = string("op_31065_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31065_end_mask_0 = const()[name = string("op_31065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31065_cast_fp16 = slice_by_index(begin = var_31065_begin_0, end = var_31065_end_0, end_mask = var_31065_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31065_cast_fp16")]; + tensor var_31072_begin_0 = const()[name = string("op_31072_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31072_end_0 = const()[name = string("op_31072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31072_end_mask_0 = const()[name = string("op_31072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31072_cast_fp16 = slice_by_index(begin = var_31072_begin_0, end = var_31072_end_0, end_mask = var_31072_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31072_cast_fp16")]; + tensor var_31079_begin_0 = const()[name = string("op_31079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31079_end_0 = const()[name = string("op_31079_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31079_end_mask_0 = const()[name = string("op_31079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31079_cast_fp16 = slice_by_index(begin = var_31079_begin_0, end = var_31079_end_0, end_mask = var_31079_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31079_cast_fp16")]; + tensor var_31086_begin_0 = const()[name = string("op_31086_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31086_end_0 = const()[name = string("op_31086_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31086_end_mask_0 = const()[name = string("op_31086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31086_cast_fp16 = slice_by_index(begin = var_31086_begin_0, end = var_31086_end_0, end_mask = var_31086_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31086_cast_fp16")]; + tensor var_31093_begin_0 = const()[name = string("op_31093_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31093_end_0 = const()[name = string("op_31093_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31093_end_mask_0 = const()[name = string("op_31093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31093_cast_fp16 = slice_by_index(begin = var_31093_begin_0, end = var_31093_end_0, end_mask = var_31093_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31093_cast_fp16")]; + tensor var_31100_begin_0 = const()[name = string("op_31100_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31100_end_0 = const()[name = string("op_31100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31100_end_mask_0 = const()[name = string("op_31100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31100_cast_fp16 = slice_by_index(begin = var_31100_begin_0, end = var_31100_end_0, end_mask = var_31100_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31100_cast_fp16")]; + tensor var_31107_begin_0 = const()[name = string("op_31107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31107_end_0 = const()[name = string("op_31107_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31107_end_mask_0 = const()[name = string("op_31107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31107_cast_fp16 = slice_by_index(begin = var_31107_begin_0, end = var_31107_end_0, end_mask = var_31107_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31107_cast_fp16")]; + tensor var_31114_begin_0 = const()[name = string("op_31114_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31114_end_0 = const()[name = string("op_31114_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31114_end_mask_0 = const()[name = string("op_31114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31114_cast_fp16 = slice_by_index(begin = var_31114_begin_0, end = var_31114_end_0, end_mask = var_31114_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31114_cast_fp16")]; + tensor var_31121_begin_0 = const()[name = string("op_31121_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31121_end_0 = const()[name = string("op_31121_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31121_end_mask_0 = const()[name = string("op_31121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31121_cast_fp16 = slice_by_index(begin = var_31121_begin_0, end = var_31121_end_0, end_mask = var_31121_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31121_cast_fp16")]; + tensor var_31128_begin_0 = const()[name = string("op_31128_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31128_end_0 = const()[name = string("op_31128_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31128_end_mask_0 = const()[name = string("op_31128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31128_cast_fp16 = slice_by_index(begin = var_31128_begin_0, end = var_31128_end_0, end_mask = var_31128_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31128_cast_fp16")]; + tensor var_31135_begin_0 = const()[name = string("op_31135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31135_end_0 = const()[name = string("op_31135_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31135_end_mask_0 = const()[name = string("op_31135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31135_cast_fp16 = slice_by_index(begin = var_31135_begin_0, end = var_31135_end_0, end_mask = var_31135_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31135_cast_fp16")]; + tensor var_31142_begin_0 = const()[name = string("op_31142_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31142_end_0 = const()[name = string("op_31142_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31142_end_mask_0 = const()[name = string("op_31142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31142_cast_fp16 = slice_by_index(begin = var_31142_begin_0, end = var_31142_end_0, end_mask = var_31142_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31142_cast_fp16")]; + tensor var_31149_begin_0 = const()[name = string("op_31149_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31149_end_0 = const()[name = string("op_31149_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31149_end_mask_0 = const()[name = string("op_31149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31149_cast_fp16 = slice_by_index(begin = var_31149_begin_0, end = var_31149_end_0, end_mask = var_31149_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31149_cast_fp16")]; + tensor var_31156_begin_0 = const()[name = string("op_31156_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31156_end_0 = const()[name = string("op_31156_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31156_end_mask_0 = const()[name = string("op_31156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31156_cast_fp16 = slice_by_index(begin = var_31156_begin_0, end = var_31156_end_0, end_mask = var_31156_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31156_cast_fp16")]; + tensor var_31163_begin_0 = const()[name = string("op_31163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31163_end_0 = const()[name = string("op_31163_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31163_end_mask_0 = const()[name = string("op_31163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31163_cast_fp16 = slice_by_index(begin = var_31163_begin_0, end = var_31163_end_0, end_mask = var_31163_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31163_cast_fp16")]; + tensor var_31170_begin_0 = const()[name = string("op_31170_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31170_end_0 = const()[name = string("op_31170_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31170_end_mask_0 = const()[name = string("op_31170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31170_cast_fp16 = slice_by_index(begin = var_31170_begin_0, end = var_31170_end_0, end_mask = var_31170_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31170_cast_fp16")]; + tensor var_31177_begin_0 = const()[name = string("op_31177_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31177_end_0 = const()[name = string("op_31177_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31177_end_mask_0 = const()[name = string("op_31177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31177_cast_fp16 = slice_by_index(begin = var_31177_begin_0, end = var_31177_end_0, end_mask = var_31177_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31177_cast_fp16")]; + tensor var_31184_begin_0 = const()[name = string("op_31184_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31184_end_0 = const()[name = string("op_31184_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31184_end_mask_0 = const()[name = string("op_31184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31184_cast_fp16 = slice_by_index(begin = var_31184_begin_0, end = var_31184_end_0, end_mask = var_31184_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31184_cast_fp16")]; + tensor var_31191_begin_0 = const()[name = string("op_31191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31191_end_0 = const()[name = string("op_31191_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31191_end_mask_0 = const()[name = string("op_31191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31191_cast_fp16 = slice_by_index(begin = var_31191_begin_0, end = var_31191_end_0, end_mask = var_31191_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31191_cast_fp16")]; + tensor var_31198_begin_0 = const()[name = string("op_31198_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31198_end_0 = const()[name = string("op_31198_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31198_end_mask_0 = const()[name = string("op_31198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31198_cast_fp16 = slice_by_index(begin = var_31198_begin_0, end = var_31198_end_0, end_mask = var_31198_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31198_cast_fp16")]; + tensor var_31205_begin_0 = const()[name = string("op_31205_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31205_end_0 = const()[name = string("op_31205_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31205_end_mask_0 = const()[name = string("op_31205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31205_cast_fp16 = slice_by_index(begin = var_31205_begin_0, end = var_31205_end_0, end_mask = var_31205_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31205_cast_fp16")]; + tensor var_31212_begin_0 = const()[name = string("op_31212_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31212_end_0 = const()[name = string("op_31212_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31212_end_mask_0 = const()[name = string("op_31212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31212_cast_fp16 = slice_by_index(begin = var_31212_begin_0, end = var_31212_end_0, end_mask = var_31212_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31212_cast_fp16")]; + tensor var_31219_begin_0 = const()[name = string("op_31219_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31219_end_0 = const()[name = string("op_31219_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31219_end_mask_0 = const()[name = string("op_31219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31219_cast_fp16 = slice_by_index(begin = var_31219_begin_0, end = var_31219_end_0, end_mask = var_31219_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31219_cast_fp16")]; + tensor var_31226_begin_0 = const()[name = string("op_31226_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31226_end_0 = const()[name = string("op_31226_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31226_end_mask_0 = const()[name = string("op_31226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31226_cast_fp16 = slice_by_index(begin = var_31226_begin_0, end = var_31226_end_0, end_mask = var_31226_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31226_cast_fp16")]; + tensor var_31233_begin_0 = const()[name = string("op_31233_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31233_end_0 = const()[name = string("op_31233_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31233_end_mask_0 = const()[name = string("op_31233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31233_cast_fp16 = slice_by_index(begin = var_31233_begin_0, end = var_31233_end_0, end_mask = var_31233_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31233_cast_fp16")]; + tensor var_31240_begin_0 = const()[name = string("op_31240_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31240_end_0 = const()[name = string("op_31240_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31240_end_mask_0 = const()[name = string("op_31240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31240_cast_fp16 = slice_by_index(begin = var_31240_begin_0, end = var_31240_end_0, end_mask = var_31240_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31240_cast_fp16")]; + tensor var_31247_begin_0 = const()[name = string("op_31247_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31247_end_0 = const()[name = string("op_31247_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31247_end_mask_0 = const()[name = string("op_31247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31247_cast_fp16 = slice_by_index(begin = var_31247_begin_0, end = var_31247_end_0, end_mask = var_31247_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31247_cast_fp16")]; + tensor var_31254_begin_0 = const()[name = string("op_31254_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31254_end_0 = const()[name = string("op_31254_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31254_end_mask_0 = const()[name = string("op_31254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31254_cast_fp16 = slice_by_index(begin = var_31254_begin_0, end = var_31254_end_0, end_mask = var_31254_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31254_cast_fp16")]; + tensor var_31261_begin_0 = const()[name = string("op_31261_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31261_end_0 = const()[name = string("op_31261_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31261_end_mask_0 = const()[name = string("op_31261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31261_cast_fp16 = slice_by_index(begin = var_31261_begin_0, end = var_31261_end_0, end_mask = var_31261_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31261_cast_fp16")]; + tensor var_31268_begin_0 = const()[name = string("op_31268_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31268_end_0 = const()[name = string("op_31268_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31268_end_mask_0 = const()[name = string("op_31268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31268_cast_fp16 = slice_by_index(begin = var_31268_begin_0, end = var_31268_end_0, end_mask = var_31268_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31268_cast_fp16")]; + tensor var_31275_begin_0 = const()[name = string("op_31275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31275_end_0 = const()[name = string("op_31275_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31275_end_mask_0 = const()[name = string("op_31275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31275_cast_fp16 = slice_by_index(begin = var_31275_begin_0, end = var_31275_end_0, end_mask = var_31275_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31275_cast_fp16")]; + tensor var_31282_begin_0 = const()[name = string("op_31282_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31282_end_0 = const()[name = string("op_31282_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31282_end_mask_0 = const()[name = string("op_31282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31282_cast_fp16 = slice_by_index(begin = var_31282_begin_0, end = var_31282_end_0, end_mask = var_31282_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31282_cast_fp16")]; + tensor var_31289_begin_0 = const()[name = string("op_31289_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31289_end_0 = const()[name = string("op_31289_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31289_end_mask_0 = const()[name = string("op_31289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31289_cast_fp16 = slice_by_index(begin = var_31289_begin_0, end = var_31289_end_0, end_mask = var_31289_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31289_cast_fp16")]; + tensor var_31296_begin_0 = const()[name = string("op_31296_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31296_end_0 = const()[name = string("op_31296_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31296_end_mask_0 = const()[name = string("op_31296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31296_cast_fp16 = slice_by_index(begin = var_31296_begin_0, end = var_31296_end_0, end_mask = var_31296_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31296_cast_fp16")]; + tensor var_31303_begin_0 = const()[name = string("op_31303_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31303_end_0 = const()[name = string("op_31303_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31303_end_mask_0 = const()[name = string("op_31303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31303_cast_fp16 = slice_by_index(begin = var_31303_begin_0, end = var_31303_end_0, end_mask = var_31303_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31303_cast_fp16")]; + tensor var_31310_begin_0 = const()[name = string("op_31310_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31310_end_0 = const()[name = string("op_31310_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31310_end_mask_0 = const()[name = string("op_31310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31310_cast_fp16 = slice_by_index(begin = var_31310_begin_0, end = var_31310_end_0, end_mask = var_31310_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31310_cast_fp16")]; + tensor var_31317_begin_0 = const()[name = string("op_31317_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31317_end_0 = const()[name = string("op_31317_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31317_end_mask_0 = const()[name = string("op_31317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31317_cast_fp16 = slice_by_index(begin = var_31317_begin_0, end = var_31317_end_0, end_mask = var_31317_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31317_cast_fp16")]; + tensor var_31324_begin_0 = const()[name = string("op_31324_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31324_end_0 = const()[name = string("op_31324_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31324_end_mask_0 = const()[name = string("op_31324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31324_cast_fp16 = slice_by_index(begin = var_31324_begin_0, end = var_31324_end_0, end_mask = var_31324_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31324_cast_fp16")]; + tensor var_31331_begin_0 = const()[name = string("op_31331_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31331_end_0 = const()[name = string("op_31331_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31331_end_mask_0 = const()[name = string("op_31331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31331_cast_fp16 = slice_by_index(begin = var_31331_begin_0, end = var_31331_end_0, end_mask = var_31331_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31331_cast_fp16")]; + tensor var_31338_begin_0 = const()[name = string("op_31338_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31338_end_0 = const()[name = string("op_31338_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31338_end_mask_0 = const()[name = string("op_31338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31338_cast_fp16 = slice_by_index(begin = var_31338_begin_0, end = var_31338_end_0, end_mask = var_31338_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31338_cast_fp16")]; + tensor var_31345_begin_0 = const()[name = string("op_31345_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31345_end_0 = const()[name = string("op_31345_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31345_end_mask_0 = const()[name = string("op_31345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31345_cast_fp16 = slice_by_index(begin = var_31345_begin_0, end = var_31345_end_0, end_mask = var_31345_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31345_cast_fp16")]; + tensor var_31352_begin_0 = const()[name = string("op_31352_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31352_end_0 = const()[name = string("op_31352_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31352_end_mask_0 = const()[name = string("op_31352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31352_cast_fp16 = slice_by_index(begin = var_31352_begin_0, end = var_31352_end_0, end_mask = var_31352_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31352_cast_fp16")]; + tensor var_31359_begin_0 = const()[name = string("op_31359_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31359_end_0 = const()[name = string("op_31359_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31359_end_mask_0 = const()[name = string("op_31359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31359_cast_fp16 = slice_by_index(begin = var_31359_begin_0, end = var_31359_end_0, end_mask = var_31359_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31359_cast_fp16")]; + tensor var_31366_begin_0 = const()[name = string("op_31366_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31366_end_0 = const()[name = string("op_31366_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31366_end_mask_0 = const()[name = string("op_31366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31366_cast_fp16 = slice_by_index(begin = var_31366_begin_0, end = var_31366_end_0, end_mask = var_31366_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31366_cast_fp16")]; + tensor var_31373_begin_0 = const()[name = string("op_31373_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31373_end_0 = const()[name = string("op_31373_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31373_end_mask_0 = const()[name = string("op_31373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31373_cast_fp16 = slice_by_index(begin = var_31373_begin_0, end = var_31373_end_0, end_mask = var_31373_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31373_cast_fp16")]; + tensor var_31380_begin_0 = const()[name = string("op_31380_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31380_end_0 = const()[name = string("op_31380_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31380_end_mask_0 = const()[name = string("op_31380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31380_cast_fp16 = slice_by_index(begin = var_31380_begin_0, end = var_31380_end_0, end_mask = var_31380_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31380_cast_fp16")]; + tensor var_31387_begin_0 = const()[name = string("op_31387_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31387_end_0 = const()[name = string("op_31387_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31387_end_mask_0 = const()[name = string("op_31387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31387_cast_fp16 = slice_by_index(begin = var_31387_begin_0, end = var_31387_end_0, end_mask = var_31387_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31387_cast_fp16")]; + tensor var_31394_begin_0 = const()[name = string("op_31394_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31394_end_0 = const()[name = string("op_31394_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31394_end_mask_0 = const()[name = string("op_31394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31394_cast_fp16 = slice_by_index(begin = var_31394_begin_0, end = var_31394_end_0, end_mask = var_31394_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31394_cast_fp16")]; + tensor var_31401_begin_0 = const()[name = string("op_31401_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31401_end_0 = const()[name = string("op_31401_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31401_end_mask_0 = const()[name = string("op_31401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31401_cast_fp16 = slice_by_index(begin = var_31401_begin_0, end = var_31401_end_0, end_mask = var_31401_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31401_cast_fp16")]; + tensor var_31408_begin_0 = const()[name = string("op_31408_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31408_end_0 = const()[name = string("op_31408_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31408_end_mask_0 = const()[name = string("op_31408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31408_cast_fp16 = slice_by_index(begin = var_31408_begin_0, end = var_31408_end_0, end_mask = var_31408_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31408_cast_fp16")]; + tensor var_31415_begin_0 = const()[name = string("op_31415_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31415_end_0 = const()[name = string("op_31415_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31415_end_mask_0 = const()[name = string("op_31415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31415_cast_fp16 = slice_by_index(begin = var_31415_begin_0, end = var_31415_end_0, end_mask = var_31415_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31415_cast_fp16")]; + tensor var_31422_begin_0 = const()[name = string("op_31422_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31422_end_0 = const()[name = string("op_31422_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31422_end_mask_0 = const()[name = string("op_31422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31422_cast_fp16 = slice_by_index(begin = var_31422_begin_0, end = var_31422_end_0, end_mask = var_31422_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31422_cast_fp16")]; + tensor var_31429_begin_0 = const()[name = string("op_31429_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31429_end_0 = const()[name = string("op_31429_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31429_end_mask_0 = const()[name = string("op_31429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31429_cast_fp16 = slice_by_index(begin = var_31429_begin_0, end = var_31429_end_0, end_mask = var_31429_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31429_cast_fp16")]; + tensor var_31436_begin_0 = const()[name = string("op_31436_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31436_end_0 = const()[name = string("op_31436_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31436_end_mask_0 = const()[name = string("op_31436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31436_cast_fp16 = slice_by_index(begin = var_31436_begin_0, end = var_31436_end_0, end_mask = var_31436_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31436_cast_fp16")]; + tensor k_41_perm_0 = const()[name = string("k_41_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_31441_begin_0 = const()[name = string("op_31441_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31441_end_0 = const()[name = string("op_31441_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_31441_end_mask_0 = const()[name = string("op_31441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_41_cast_fp16 = transpose(perm = k_41_perm_0, x = key_41_cast_fp16)[name = string("transpose_11")]; + tensor var_31441_cast_fp16 = slice_by_index(begin = var_31441_begin_0, end = var_31441_end_0, end_mask = var_31441_end_mask_0, x = k_41_cast_fp16)[name = string("op_31441_cast_fp16")]; + tensor var_31445_begin_0 = const()[name = string("op_31445_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_31445_end_0 = const()[name = string("op_31445_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_31445_end_mask_0 = const()[name = string("op_31445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31445_cast_fp16 = slice_by_index(begin = var_31445_begin_0, end = var_31445_end_0, end_mask = var_31445_end_mask_0, x = k_41_cast_fp16)[name = string("op_31445_cast_fp16")]; + tensor var_31449_begin_0 = const()[name = string("op_31449_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_31449_end_0 = const()[name = string("op_31449_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_31449_end_mask_0 = const()[name = string("op_31449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31449_cast_fp16 = slice_by_index(begin = var_31449_begin_0, end = var_31449_end_0, end_mask = var_31449_end_mask_0, x = k_41_cast_fp16)[name = string("op_31449_cast_fp16")]; + tensor var_31453_begin_0 = const()[name = string("op_31453_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_31453_end_0 = const()[name = string("op_31453_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_31453_end_mask_0 = const()[name = string("op_31453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31453_cast_fp16 = slice_by_index(begin = var_31453_begin_0, end = var_31453_end_0, end_mask = var_31453_end_mask_0, x = k_41_cast_fp16)[name = string("op_31453_cast_fp16")]; + tensor var_31457_begin_0 = const()[name = string("op_31457_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_31457_end_0 = const()[name = string("op_31457_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_31457_end_mask_0 = const()[name = string("op_31457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31457_cast_fp16 = slice_by_index(begin = var_31457_begin_0, end = var_31457_end_0, end_mask = var_31457_end_mask_0, x = k_41_cast_fp16)[name = string("op_31457_cast_fp16")]; + tensor var_31461_begin_0 = const()[name = string("op_31461_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_31461_end_0 = const()[name = string("op_31461_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_31461_end_mask_0 = const()[name = string("op_31461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31461_cast_fp16 = slice_by_index(begin = var_31461_begin_0, end = var_31461_end_0, end_mask = var_31461_end_mask_0, x = k_41_cast_fp16)[name = string("op_31461_cast_fp16")]; + tensor var_31465_begin_0 = const()[name = string("op_31465_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_31465_end_0 = const()[name = string("op_31465_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_31465_end_mask_0 = const()[name = string("op_31465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31465_cast_fp16 = slice_by_index(begin = var_31465_begin_0, end = var_31465_end_0, end_mask = var_31465_end_mask_0, x = k_41_cast_fp16)[name = string("op_31465_cast_fp16")]; + tensor var_31469_begin_0 = const()[name = string("op_31469_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_31469_end_0 = const()[name = string("op_31469_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_31469_end_mask_0 = const()[name = string("op_31469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31469_cast_fp16 = slice_by_index(begin = var_31469_begin_0, end = var_31469_end_0, end_mask = var_31469_end_mask_0, x = k_41_cast_fp16)[name = string("op_31469_cast_fp16")]; + tensor var_31473_begin_0 = const()[name = string("op_31473_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_31473_end_0 = const()[name = string("op_31473_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_31473_end_mask_0 = const()[name = string("op_31473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31473_cast_fp16 = slice_by_index(begin = var_31473_begin_0, end = var_31473_end_0, end_mask = var_31473_end_mask_0, x = k_41_cast_fp16)[name = string("op_31473_cast_fp16")]; + tensor var_31477_begin_0 = const()[name = string("op_31477_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_31477_end_0 = const()[name = string("op_31477_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_31477_end_mask_0 = const()[name = string("op_31477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31477_cast_fp16 = slice_by_index(begin = var_31477_begin_0, end = var_31477_end_0, end_mask = var_31477_end_mask_0, x = k_41_cast_fp16)[name = string("op_31477_cast_fp16")]; + tensor var_31481_begin_0 = const()[name = string("op_31481_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_31481_end_0 = const()[name = string("op_31481_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_31481_end_mask_0 = const()[name = string("op_31481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31481_cast_fp16 = slice_by_index(begin = var_31481_begin_0, end = var_31481_end_0, end_mask = var_31481_end_mask_0, x = k_41_cast_fp16)[name = string("op_31481_cast_fp16")]; + tensor var_31485_begin_0 = const()[name = string("op_31485_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_31485_end_0 = const()[name = string("op_31485_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_31485_end_mask_0 = const()[name = string("op_31485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31485_cast_fp16 = slice_by_index(begin = var_31485_begin_0, end = var_31485_end_0, end_mask = var_31485_end_mask_0, x = k_41_cast_fp16)[name = string("op_31485_cast_fp16")]; + tensor var_31489_begin_0 = const()[name = string("op_31489_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_31489_end_0 = const()[name = string("op_31489_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_31489_end_mask_0 = const()[name = string("op_31489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31489_cast_fp16 = slice_by_index(begin = var_31489_begin_0, end = var_31489_end_0, end_mask = var_31489_end_mask_0, x = k_41_cast_fp16)[name = string("op_31489_cast_fp16")]; + tensor var_31493_begin_0 = const()[name = string("op_31493_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_31493_end_0 = const()[name = string("op_31493_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_31493_end_mask_0 = const()[name = string("op_31493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31493_cast_fp16 = slice_by_index(begin = var_31493_begin_0, end = var_31493_end_0, end_mask = var_31493_end_mask_0, x = k_41_cast_fp16)[name = string("op_31493_cast_fp16")]; + tensor var_31497_begin_0 = const()[name = string("op_31497_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_31497_end_0 = const()[name = string("op_31497_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_31497_end_mask_0 = const()[name = string("op_31497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31497_cast_fp16 = slice_by_index(begin = var_31497_begin_0, end = var_31497_end_0, end_mask = var_31497_end_mask_0, x = k_41_cast_fp16)[name = string("op_31497_cast_fp16")]; + tensor var_31501_begin_0 = const()[name = string("op_31501_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_31501_end_0 = const()[name = string("op_31501_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_31501_end_mask_0 = const()[name = string("op_31501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31501_cast_fp16 = slice_by_index(begin = var_31501_begin_0, end = var_31501_end_0, end_mask = var_31501_end_mask_0, x = k_41_cast_fp16)[name = string("op_31501_cast_fp16")]; + tensor var_31505_begin_0 = const()[name = string("op_31505_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_31505_end_0 = const()[name = string("op_31505_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_31505_end_mask_0 = const()[name = string("op_31505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31505_cast_fp16 = slice_by_index(begin = var_31505_begin_0, end = var_31505_end_0, end_mask = var_31505_end_mask_0, x = k_41_cast_fp16)[name = string("op_31505_cast_fp16")]; + tensor var_31509_begin_0 = const()[name = string("op_31509_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_31509_end_0 = const()[name = string("op_31509_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_31509_end_mask_0 = const()[name = string("op_31509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31509_cast_fp16 = slice_by_index(begin = var_31509_begin_0, end = var_31509_end_0, end_mask = var_31509_end_mask_0, x = k_41_cast_fp16)[name = string("op_31509_cast_fp16")]; + tensor var_31513_begin_0 = const()[name = string("op_31513_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_31513_end_0 = const()[name = string("op_31513_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_31513_end_mask_0 = const()[name = string("op_31513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31513_cast_fp16 = slice_by_index(begin = var_31513_begin_0, end = var_31513_end_0, end_mask = var_31513_end_mask_0, x = k_41_cast_fp16)[name = string("op_31513_cast_fp16")]; + tensor var_31517_begin_0 = const()[name = string("op_31517_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_31517_end_0 = const()[name = string("op_31517_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_31517_end_mask_0 = const()[name = string("op_31517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31517_cast_fp16 = slice_by_index(begin = var_31517_begin_0, end = var_31517_end_0, end_mask = var_31517_end_mask_0, x = k_41_cast_fp16)[name = string("op_31517_cast_fp16")]; + tensor var_31519_begin_0 = const()[name = string("op_31519_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31519_end_0 = const()[name = string("op_31519_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31519_end_mask_0 = const()[name = string("op_31519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31519_cast_fp16 = slice_by_index(begin = var_31519_begin_0, end = var_31519_end_0, end_mask = var_31519_end_mask_0, x = value_41_cast_fp16)[name = string("op_31519_cast_fp16")]; + tensor var_31523_begin_0 = const()[name = string("op_31523_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31523_end_0 = const()[name = string("op_31523_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_31523_end_mask_0 = const()[name = string("op_31523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31523_cast_fp16 = slice_by_index(begin = var_31523_begin_0, end = var_31523_end_0, end_mask = var_31523_end_mask_0, x = value_41_cast_fp16)[name = string("op_31523_cast_fp16")]; + tensor var_31527_begin_0 = const()[name = string("op_31527_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31527_end_0 = const()[name = string("op_31527_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_31527_end_mask_0 = const()[name = string("op_31527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31527_cast_fp16 = slice_by_index(begin = var_31527_begin_0, end = var_31527_end_0, end_mask = var_31527_end_mask_0, x = value_41_cast_fp16)[name = string("op_31527_cast_fp16")]; + tensor var_31531_begin_0 = const()[name = string("op_31531_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31531_end_0 = const()[name = string("op_31531_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_31531_end_mask_0 = const()[name = string("op_31531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31531_cast_fp16 = slice_by_index(begin = var_31531_begin_0, end = var_31531_end_0, end_mask = var_31531_end_mask_0, x = value_41_cast_fp16)[name = string("op_31531_cast_fp16")]; + tensor var_31535_begin_0 = const()[name = string("op_31535_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31535_end_0 = const()[name = string("op_31535_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_31535_end_mask_0 = const()[name = string("op_31535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31535_cast_fp16 = slice_by_index(begin = var_31535_begin_0, end = var_31535_end_0, end_mask = var_31535_end_mask_0, x = value_41_cast_fp16)[name = string("op_31535_cast_fp16")]; + tensor var_31539_begin_0 = const()[name = string("op_31539_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31539_end_0 = const()[name = string("op_31539_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_31539_end_mask_0 = const()[name = string("op_31539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31539_cast_fp16 = slice_by_index(begin = var_31539_begin_0, end = var_31539_end_0, end_mask = var_31539_end_mask_0, x = value_41_cast_fp16)[name = string("op_31539_cast_fp16")]; + tensor var_31543_begin_0 = const()[name = string("op_31543_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31543_end_0 = const()[name = string("op_31543_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_31543_end_mask_0 = const()[name = string("op_31543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31543_cast_fp16 = slice_by_index(begin = var_31543_begin_0, end = var_31543_end_0, end_mask = var_31543_end_mask_0, x = value_41_cast_fp16)[name = string("op_31543_cast_fp16")]; + tensor var_31547_begin_0 = const()[name = string("op_31547_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31547_end_0 = const()[name = string("op_31547_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_31547_end_mask_0 = const()[name = string("op_31547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31547_cast_fp16 = slice_by_index(begin = var_31547_begin_0, end = var_31547_end_0, end_mask = var_31547_end_mask_0, x = value_41_cast_fp16)[name = string("op_31547_cast_fp16")]; + tensor var_31551_begin_0 = const()[name = string("op_31551_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31551_end_0 = const()[name = string("op_31551_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_31551_end_mask_0 = const()[name = string("op_31551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31551_cast_fp16 = slice_by_index(begin = var_31551_begin_0, end = var_31551_end_0, end_mask = var_31551_end_mask_0, x = value_41_cast_fp16)[name = string("op_31551_cast_fp16")]; + tensor var_31555_begin_0 = const()[name = string("op_31555_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31555_end_0 = const()[name = string("op_31555_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_31555_end_mask_0 = const()[name = string("op_31555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31555_cast_fp16 = slice_by_index(begin = var_31555_begin_0, end = var_31555_end_0, end_mask = var_31555_end_mask_0, x = value_41_cast_fp16)[name = string("op_31555_cast_fp16")]; + tensor var_31559_begin_0 = const()[name = string("op_31559_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31559_end_0 = const()[name = string("op_31559_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_31559_end_mask_0 = const()[name = string("op_31559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31559_cast_fp16 = slice_by_index(begin = var_31559_begin_0, end = var_31559_end_0, end_mask = var_31559_end_mask_0, x = value_41_cast_fp16)[name = string("op_31559_cast_fp16")]; + tensor var_31563_begin_0 = const()[name = string("op_31563_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31563_end_0 = const()[name = string("op_31563_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_31563_end_mask_0 = const()[name = string("op_31563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31563_cast_fp16 = slice_by_index(begin = var_31563_begin_0, end = var_31563_end_0, end_mask = var_31563_end_mask_0, x = value_41_cast_fp16)[name = string("op_31563_cast_fp16")]; + tensor var_31567_begin_0 = const()[name = string("op_31567_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31567_end_0 = const()[name = string("op_31567_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_31567_end_mask_0 = const()[name = string("op_31567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31567_cast_fp16 = slice_by_index(begin = var_31567_begin_0, end = var_31567_end_0, end_mask = var_31567_end_mask_0, x = value_41_cast_fp16)[name = string("op_31567_cast_fp16")]; + tensor var_31571_begin_0 = const()[name = string("op_31571_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31571_end_0 = const()[name = string("op_31571_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_31571_end_mask_0 = const()[name = string("op_31571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31571_cast_fp16 = slice_by_index(begin = var_31571_begin_0, end = var_31571_end_0, end_mask = var_31571_end_mask_0, x = value_41_cast_fp16)[name = string("op_31571_cast_fp16")]; + tensor var_31575_begin_0 = const()[name = string("op_31575_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31575_end_0 = const()[name = string("op_31575_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_31575_end_mask_0 = const()[name = string("op_31575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31575_cast_fp16 = slice_by_index(begin = var_31575_begin_0, end = var_31575_end_0, end_mask = var_31575_end_mask_0, x = value_41_cast_fp16)[name = string("op_31575_cast_fp16")]; + tensor var_31579_begin_0 = const()[name = string("op_31579_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31579_end_0 = const()[name = string("op_31579_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_31579_end_mask_0 = const()[name = string("op_31579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31579_cast_fp16 = slice_by_index(begin = var_31579_begin_0, end = var_31579_end_0, end_mask = var_31579_end_mask_0, x = value_41_cast_fp16)[name = string("op_31579_cast_fp16")]; + tensor var_31583_begin_0 = const()[name = string("op_31583_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31583_end_0 = const()[name = string("op_31583_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_31583_end_mask_0 = const()[name = string("op_31583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31583_cast_fp16 = slice_by_index(begin = var_31583_begin_0, end = var_31583_end_0, end_mask = var_31583_end_mask_0, x = value_41_cast_fp16)[name = string("op_31583_cast_fp16")]; + tensor var_31587_begin_0 = const()[name = string("op_31587_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31587_end_0 = const()[name = string("op_31587_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_31587_end_mask_0 = const()[name = string("op_31587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31587_cast_fp16 = slice_by_index(begin = var_31587_begin_0, end = var_31587_end_0, end_mask = var_31587_end_mask_0, x = value_41_cast_fp16)[name = string("op_31587_cast_fp16")]; + tensor var_31591_begin_0 = const()[name = string("op_31591_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31591_end_0 = const()[name = string("op_31591_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_31591_end_mask_0 = const()[name = string("op_31591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31591_cast_fp16 = slice_by_index(begin = var_31591_begin_0, end = var_31591_end_0, end_mask = var_31591_end_mask_0, x = value_41_cast_fp16)[name = string("op_31591_cast_fp16")]; + tensor var_31595_begin_0 = const()[name = string("op_31595_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31595_end_0 = const()[name = string("op_31595_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_31595_end_mask_0 = const()[name = string("op_31595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31595_cast_fp16 = slice_by_index(begin = var_31595_begin_0, end = var_31595_end_0, end_mask = var_31595_end_mask_0, x = value_41_cast_fp16)[name = string("op_31595_cast_fp16")]; + string _SplitHeadsQ__mh_w_3201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3201_equation_0, values = (var_31441_cast_fp16, var_30883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3201_cast_fp16")]; + string _SplitHeadsQ__mh_w_3203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3203_equation_0, values = (var_31441_cast_fp16, var_30890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3203_cast_fp16")]; + string _SplitHeadsQ__mh_w_3205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3205_equation_0, values = (var_31441_cast_fp16, var_30897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3205_cast_fp16")]; + string _SplitHeadsQ__mh_w_3207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3207_equation_0, values = (var_31441_cast_fp16, var_30904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3207_cast_fp16")]; + string _SplitHeadsQ__mh_w_3209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3209_equation_0, values = (var_31445_cast_fp16, var_30911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3209_cast_fp16")]; + string _SplitHeadsQ__mh_w_3211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3211_equation_0, values = (var_31445_cast_fp16, var_30918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3211_cast_fp16")]; + string _SplitHeadsQ__mh_w_3213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3213_equation_0, values = (var_31445_cast_fp16, var_30925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3213_cast_fp16")]; + string _SplitHeadsQ__mh_w_3215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3215_equation_0, values = (var_31445_cast_fp16, var_30932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3215_cast_fp16")]; + string _SplitHeadsQ__mh_w_3217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3217_equation_0, values = (var_31449_cast_fp16, var_30939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3217_cast_fp16")]; + string _SplitHeadsQ__mh_w_3219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3219_equation_0, values = (var_31449_cast_fp16, var_30946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3219_cast_fp16")]; + string _SplitHeadsQ__mh_w_3221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3221_equation_0, values = (var_31449_cast_fp16, var_30953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3221_cast_fp16")]; + string _SplitHeadsQ__mh_w_3223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3223_equation_0, values = (var_31449_cast_fp16, var_30960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3223_cast_fp16")]; + string _SplitHeadsQ__mh_w_3225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3225_equation_0, values = (var_31453_cast_fp16, var_30967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3225_cast_fp16")]; + string _SplitHeadsQ__mh_w_3227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3227_equation_0, values = (var_31453_cast_fp16, var_30974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3227_cast_fp16")]; + string _SplitHeadsQ__mh_w_3229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3229_equation_0, values = (var_31453_cast_fp16, var_30981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3229_cast_fp16")]; + string _SplitHeadsQ__mh_w_3231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3231_equation_0, values = (var_31453_cast_fp16, var_30988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3231_cast_fp16")]; + string _SplitHeadsQ__mh_w_3233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3233_equation_0, values = (var_31457_cast_fp16, var_30995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3233_cast_fp16")]; + string _SplitHeadsQ__mh_w_3235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3235_equation_0, values = (var_31457_cast_fp16, var_31002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3235_cast_fp16")]; + string _SplitHeadsQ__mh_w_3237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3237_equation_0, values = (var_31457_cast_fp16, var_31009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3237_cast_fp16")]; + string _SplitHeadsQ__mh_w_3239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3239_equation_0, values = (var_31457_cast_fp16, var_31016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3239_cast_fp16")]; + string _SplitHeadsQ__mh_w_3241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3241_equation_0, values = (var_31461_cast_fp16, var_31023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3241_cast_fp16")]; + string _SplitHeadsQ__mh_w_3243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3243_equation_0, values = (var_31461_cast_fp16, var_31030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3243_cast_fp16")]; + string _SplitHeadsQ__mh_w_3245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3245_equation_0, values = (var_31461_cast_fp16, var_31037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3245_cast_fp16")]; + string _SplitHeadsQ__mh_w_3247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3247_equation_0, values = (var_31461_cast_fp16, var_31044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3247_cast_fp16")]; + string _SplitHeadsQ__mh_w_3249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3249_equation_0, values = (var_31465_cast_fp16, var_31051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3249_cast_fp16")]; + string _SplitHeadsQ__mh_w_3251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3251_equation_0, values = (var_31465_cast_fp16, var_31058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3251_cast_fp16")]; + string _SplitHeadsQ__mh_w_3253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3253_equation_0, values = (var_31465_cast_fp16, var_31065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3253_cast_fp16")]; + string _SplitHeadsQ__mh_w_3255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3255_equation_0, values = (var_31465_cast_fp16, var_31072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3255_cast_fp16")]; + string _SplitHeadsQ__mh_w_3257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3257_equation_0, values = (var_31469_cast_fp16, var_31079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3257_cast_fp16")]; + string _SplitHeadsQ__mh_w_3259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3259_equation_0, values = (var_31469_cast_fp16, var_31086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3259_cast_fp16")]; + string _SplitHeadsQ__mh_w_3261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3261_equation_0, values = (var_31469_cast_fp16, var_31093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3261_cast_fp16")]; + string _SplitHeadsQ__mh_w_3263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3263_equation_0, values = (var_31469_cast_fp16, var_31100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3263_cast_fp16")]; + string _SplitHeadsQ__mh_w_3265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3265_equation_0, values = (var_31473_cast_fp16, var_31107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3265_cast_fp16")]; + string _SplitHeadsQ__mh_w_3267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3267_equation_0, values = (var_31473_cast_fp16, var_31114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3267_cast_fp16")]; + string _SplitHeadsQ__mh_w_3269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3269_equation_0, values = (var_31473_cast_fp16, var_31121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3269_cast_fp16")]; + string _SplitHeadsQ__mh_w_3271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3271_equation_0, values = (var_31473_cast_fp16, var_31128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3271_cast_fp16")]; + string _SplitHeadsQ__mh_w_3273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3273_equation_0, values = (var_31477_cast_fp16, var_31135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3273_cast_fp16")]; + string _SplitHeadsQ__mh_w_3275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3275_equation_0, values = (var_31477_cast_fp16, var_31142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3275_cast_fp16")]; + string _SplitHeadsQ__mh_w_3277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3277_equation_0, values = (var_31477_cast_fp16, var_31149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3277_cast_fp16")]; + string _SplitHeadsQ__mh_w_3279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3279_equation_0, values = (var_31477_cast_fp16, var_31156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3279_cast_fp16")]; + string _SplitHeadsQ__mh_w_3281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3281_equation_0, values = (var_31481_cast_fp16, var_31163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3281_cast_fp16")]; + string _SplitHeadsQ__mh_w_3283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3283_equation_0, values = (var_31481_cast_fp16, var_31170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3283_cast_fp16")]; + string _SplitHeadsQ__mh_w_3285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3285_equation_0, values = (var_31481_cast_fp16, var_31177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3285_cast_fp16")]; + string _SplitHeadsQ__mh_w_3287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3287_equation_0, values = (var_31481_cast_fp16, var_31184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3287_cast_fp16")]; + string _SplitHeadsQ__mh_w_3289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3289_equation_0, values = (var_31485_cast_fp16, var_31191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3289_cast_fp16")]; + string _SplitHeadsQ__mh_w_3291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3291_equation_0, values = (var_31485_cast_fp16, var_31198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3291_cast_fp16")]; + string _SplitHeadsQ__mh_w_3293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3293_equation_0, values = (var_31485_cast_fp16, var_31205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3293_cast_fp16")]; + string _SplitHeadsQ__mh_w_3295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3295_equation_0, values = (var_31485_cast_fp16, var_31212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3295_cast_fp16")]; + string _SplitHeadsQ__mh_w_3297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3297_equation_0, values = (var_31489_cast_fp16, var_31219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3297_cast_fp16")]; + string _SplitHeadsQ__mh_w_3299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3299_equation_0, values = (var_31489_cast_fp16, var_31226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3299_cast_fp16")]; + string _SplitHeadsQ__mh_w_3301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3301_equation_0, values = (var_31489_cast_fp16, var_31233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3301_cast_fp16")]; + string _SplitHeadsQ__mh_w_3303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3303_equation_0, values = (var_31489_cast_fp16, var_31240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3303_cast_fp16")]; + string _SplitHeadsQ__mh_w_3305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3305_equation_0, values = (var_31493_cast_fp16, var_31247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3305_cast_fp16")]; + string _SplitHeadsQ__mh_w_3307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3307_equation_0, values = (var_31493_cast_fp16, var_31254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3307_cast_fp16")]; + string _SplitHeadsQ__mh_w_3309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3309_equation_0, values = (var_31493_cast_fp16, var_31261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3309_cast_fp16")]; + string _SplitHeadsQ__mh_w_3311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3311_equation_0, values = (var_31493_cast_fp16, var_31268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3311_cast_fp16")]; + string _SplitHeadsQ__mh_w_3313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3313_equation_0, values = (var_31497_cast_fp16, var_31275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3313_cast_fp16")]; + string _SplitHeadsQ__mh_w_3315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3315_equation_0, values = (var_31497_cast_fp16, var_31282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3315_cast_fp16")]; + string _SplitHeadsQ__mh_w_3317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3317_equation_0, values = (var_31497_cast_fp16, var_31289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3317_cast_fp16")]; + string _SplitHeadsQ__mh_w_3319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3319_equation_0, values = (var_31497_cast_fp16, var_31296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3319_cast_fp16")]; + string _SplitHeadsQ__mh_w_3321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3321_equation_0, values = (var_31501_cast_fp16, var_31303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3321_cast_fp16")]; + string _SplitHeadsQ__mh_w_3323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3323_equation_0, values = (var_31501_cast_fp16, var_31310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3323_cast_fp16")]; + string _SplitHeadsQ__mh_w_3325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3325_equation_0, values = (var_31501_cast_fp16, var_31317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3325_cast_fp16")]; + string _SplitHeadsQ__mh_w_3327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3327_equation_0, values = (var_31501_cast_fp16, var_31324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3327_cast_fp16")]; + string _SplitHeadsQ__mh_w_3329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3329_equation_0, values = (var_31505_cast_fp16, var_31331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3329_cast_fp16")]; + string _SplitHeadsQ__mh_w_3331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3331_equation_0, values = (var_31505_cast_fp16, var_31338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3331_cast_fp16")]; + string _SplitHeadsQ__mh_w_3333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3333_equation_0, values = (var_31505_cast_fp16, var_31345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3333_cast_fp16")]; + string _SplitHeadsQ__mh_w_3335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3335_equation_0, values = (var_31505_cast_fp16, var_31352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3335_cast_fp16")]; + string _SplitHeadsQ__mh_w_3337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3337_equation_0, values = (var_31509_cast_fp16, var_31359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3337_cast_fp16")]; + string _SplitHeadsQ__mh_w_3339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3339_equation_0, values = (var_31509_cast_fp16, var_31366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3339_cast_fp16")]; + string _SplitHeadsQ__mh_w_3341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3341_equation_0, values = (var_31509_cast_fp16, var_31373_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3341_cast_fp16")]; + string _SplitHeadsQ__mh_w_3343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3343_equation_0, values = (var_31509_cast_fp16, var_31380_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3343_cast_fp16")]; + string _SplitHeadsQ__mh_w_3345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3345_equation_0, values = (var_31513_cast_fp16, var_31387_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3345_cast_fp16")]; + string _SplitHeadsQ__mh_w_3347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3347_equation_0, values = (var_31513_cast_fp16, var_31394_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3347_cast_fp16")]; + string _SplitHeadsQ__mh_w_3349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3349_equation_0, values = (var_31513_cast_fp16, var_31401_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3349_cast_fp16")]; + string _SplitHeadsQ__mh_w_3351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3351_equation_0, values = (var_31513_cast_fp16, var_31408_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3351_cast_fp16")]; + string _SplitHeadsQ__mh_w_3353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3353_equation_0, values = (var_31517_cast_fp16, var_31415_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3353_cast_fp16")]; + string _SplitHeadsQ__mh_w_3355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3355_equation_0, values = (var_31517_cast_fp16, var_31422_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3355_cast_fp16")]; + string _SplitHeadsQ__mh_w_3357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3357_equation_0, values = (var_31517_cast_fp16, var_31429_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3357_cast_fp16")]; + string _SplitHeadsQ__mh_w_3359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3359_equation_0, values = (var_31517_cast_fp16, var_31436_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3359_cast_fp16")]; + fp16 var_31758_to_fp16 = const()[name = string("op_31758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3201_cast_fp16, y = var_31758_to_fp16)[name = string("aw_chunk_3201_cast_fp16")]; + fp16 var_31760_to_fp16 = const()[name = string("op_31760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3203_cast_fp16, y = var_31760_to_fp16)[name = string("aw_chunk_3203_cast_fp16")]; + fp16 var_31762_to_fp16 = const()[name = string("op_31762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3205_cast_fp16, y = var_31762_to_fp16)[name = string("aw_chunk_3205_cast_fp16")]; + fp16 var_31764_to_fp16 = const()[name = string("op_31764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3207_cast_fp16, y = var_31764_to_fp16)[name = string("aw_chunk_3207_cast_fp16")]; + fp16 var_31766_to_fp16 = const()[name = string("op_31766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3209_cast_fp16, y = var_31766_to_fp16)[name = string("aw_chunk_3209_cast_fp16")]; + fp16 var_31768_to_fp16 = const()[name = string("op_31768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3211_cast_fp16, y = var_31768_to_fp16)[name = string("aw_chunk_3211_cast_fp16")]; + fp16 var_31770_to_fp16 = const()[name = string("op_31770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3213_cast_fp16, y = var_31770_to_fp16)[name = string("aw_chunk_3213_cast_fp16")]; + fp16 var_31772_to_fp16 = const()[name = string("op_31772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3215_cast_fp16, y = var_31772_to_fp16)[name = string("aw_chunk_3215_cast_fp16")]; + fp16 var_31774_to_fp16 = const()[name = string("op_31774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3217_cast_fp16, y = var_31774_to_fp16)[name = string("aw_chunk_3217_cast_fp16")]; + fp16 var_31776_to_fp16 = const()[name = string("op_31776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3219_cast_fp16, y = var_31776_to_fp16)[name = string("aw_chunk_3219_cast_fp16")]; + fp16 var_31778_to_fp16 = const()[name = string("op_31778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3221_cast_fp16, y = var_31778_to_fp16)[name = string("aw_chunk_3221_cast_fp16")]; + fp16 var_31780_to_fp16 = const()[name = string("op_31780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3223_cast_fp16, y = var_31780_to_fp16)[name = string("aw_chunk_3223_cast_fp16")]; + fp16 var_31782_to_fp16 = const()[name = string("op_31782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3225_cast_fp16, y = var_31782_to_fp16)[name = string("aw_chunk_3225_cast_fp16")]; + fp16 var_31784_to_fp16 = const()[name = string("op_31784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3227_cast_fp16, y = var_31784_to_fp16)[name = string("aw_chunk_3227_cast_fp16")]; + fp16 var_31786_to_fp16 = const()[name = string("op_31786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3229_cast_fp16, y = var_31786_to_fp16)[name = string("aw_chunk_3229_cast_fp16")]; + fp16 var_31788_to_fp16 = const()[name = string("op_31788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3231_cast_fp16, y = var_31788_to_fp16)[name = string("aw_chunk_3231_cast_fp16")]; + fp16 var_31790_to_fp16 = const()[name = string("op_31790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3233_cast_fp16, y = var_31790_to_fp16)[name = string("aw_chunk_3233_cast_fp16")]; + fp16 var_31792_to_fp16 = const()[name = string("op_31792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3235_cast_fp16, y = var_31792_to_fp16)[name = string("aw_chunk_3235_cast_fp16")]; + fp16 var_31794_to_fp16 = const()[name = string("op_31794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3237_cast_fp16, y = var_31794_to_fp16)[name = string("aw_chunk_3237_cast_fp16")]; + fp16 var_31796_to_fp16 = const()[name = string("op_31796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3239_cast_fp16, y = var_31796_to_fp16)[name = string("aw_chunk_3239_cast_fp16")]; + fp16 var_31798_to_fp16 = const()[name = string("op_31798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3241_cast_fp16, y = var_31798_to_fp16)[name = string("aw_chunk_3241_cast_fp16")]; + fp16 var_31800_to_fp16 = const()[name = string("op_31800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3243_cast_fp16, y = var_31800_to_fp16)[name = string("aw_chunk_3243_cast_fp16")]; + fp16 var_31802_to_fp16 = const()[name = string("op_31802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3245_cast_fp16, y = var_31802_to_fp16)[name = string("aw_chunk_3245_cast_fp16")]; + fp16 var_31804_to_fp16 = const()[name = string("op_31804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3247_cast_fp16, y = var_31804_to_fp16)[name = string("aw_chunk_3247_cast_fp16")]; + fp16 var_31806_to_fp16 = const()[name = string("op_31806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3249_cast_fp16, y = var_31806_to_fp16)[name = string("aw_chunk_3249_cast_fp16")]; + fp16 var_31808_to_fp16 = const()[name = string("op_31808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3251_cast_fp16, y = var_31808_to_fp16)[name = string("aw_chunk_3251_cast_fp16")]; + fp16 var_31810_to_fp16 = const()[name = string("op_31810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3253_cast_fp16, y = var_31810_to_fp16)[name = string("aw_chunk_3253_cast_fp16")]; + fp16 var_31812_to_fp16 = const()[name = string("op_31812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3255_cast_fp16, y = var_31812_to_fp16)[name = string("aw_chunk_3255_cast_fp16")]; + fp16 var_31814_to_fp16 = const()[name = string("op_31814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3257_cast_fp16, y = var_31814_to_fp16)[name = string("aw_chunk_3257_cast_fp16")]; + fp16 var_31816_to_fp16 = const()[name = string("op_31816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3259_cast_fp16, y = var_31816_to_fp16)[name = string("aw_chunk_3259_cast_fp16")]; + fp16 var_31818_to_fp16 = const()[name = string("op_31818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3261_cast_fp16, y = var_31818_to_fp16)[name = string("aw_chunk_3261_cast_fp16")]; + fp16 var_31820_to_fp16 = const()[name = string("op_31820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3263_cast_fp16, y = var_31820_to_fp16)[name = string("aw_chunk_3263_cast_fp16")]; + fp16 var_31822_to_fp16 = const()[name = string("op_31822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3265_cast_fp16, y = var_31822_to_fp16)[name = string("aw_chunk_3265_cast_fp16")]; + fp16 var_31824_to_fp16 = const()[name = string("op_31824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3267_cast_fp16, y = var_31824_to_fp16)[name = string("aw_chunk_3267_cast_fp16")]; + fp16 var_31826_to_fp16 = const()[name = string("op_31826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3269_cast_fp16, y = var_31826_to_fp16)[name = string("aw_chunk_3269_cast_fp16")]; + fp16 var_31828_to_fp16 = const()[name = string("op_31828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3271_cast_fp16, y = var_31828_to_fp16)[name = string("aw_chunk_3271_cast_fp16")]; + fp16 var_31830_to_fp16 = const()[name = string("op_31830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3273_cast_fp16, y = var_31830_to_fp16)[name = string("aw_chunk_3273_cast_fp16")]; + fp16 var_31832_to_fp16 = const()[name = string("op_31832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3275_cast_fp16, y = var_31832_to_fp16)[name = string("aw_chunk_3275_cast_fp16")]; + fp16 var_31834_to_fp16 = const()[name = string("op_31834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3277_cast_fp16, y = var_31834_to_fp16)[name = string("aw_chunk_3277_cast_fp16")]; + fp16 var_31836_to_fp16 = const()[name = string("op_31836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3279_cast_fp16, y = var_31836_to_fp16)[name = string("aw_chunk_3279_cast_fp16")]; + fp16 var_31838_to_fp16 = const()[name = string("op_31838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3281_cast_fp16, y = var_31838_to_fp16)[name = string("aw_chunk_3281_cast_fp16")]; + fp16 var_31840_to_fp16 = const()[name = string("op_31840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3283_cast_fp16, y = var_31840_to_fp16)[name = string("aw_chunk_3283_cast_fp16")]; + fp16 var_31842_to_fp16 = const()[name = string("op_31842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3285_cast_fp16, y = var_31842_to_fp16)[name = string("aw_chunk_3285_cast_fp16")]; + fp16 var_31844_to_fp16 = const()[name = string("op_31844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3287_cast_fp16, y = var_31844_to_fp16)[name = string("aw_chunk_3287_cast_fp16")]; + fp16 var_31846_to_fp16 = const()[name = string("op_31846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3289_cast_fp16, y = var_31846_to_fp16)[name = string("aw_chunk_3289_cast_fp16")]; + fp16 var_31848_to_fp16 = const()[name = string("op_31848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3291_cast_fp16, y = var_31848_to_fp16)[name = string("aw_chunk_3291_cast_fp16")]; + fp16 var_31850_to_fp16 = const()[name = string("op_31850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3293_cast_fp16, y = var_31850_to_fp16)[name = string("aw_chunk_3293_cast_fp16")]; + fp16 var_31852_to_fp16 = const()[name = string("op_31852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3295_cast_fp16, y = var_31852_to_fp16)[name = string("aw_chunk_3295_cast_fp16")]; + fp16 var_31854_to_fp16 = const()[name = string("op_31854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3297_cast_fp16, y = var_31854_to_fp16)[name = string("aw_chunk_3297_cast_fp16")]; + fp16 var_31856_to_fp16 = const()[name = string("op_31856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3299_cast_fp16, y = var_31856_to_fp16)[name = string("aw_chunk_3299_cast_fp16")]; + fp16 var_31858_to_fp16 = const()[name = string("op_31858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3301_cast_fp16, y = var_31858_to_fp16)[name = string("aw_chunk_3301_cast_fp16")]; + fp16 var_31860_to_fp16 = const()[name = string("op_31860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3303_cast_fp16, y = var_31860_to_fp16)[name = string("aw_chunk_3303_cast_fp16")]; + fp16 var_31862_to_fp16 = const()[name = string("op_31862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3305_cast_fp16, y = var_31862_to_fp16)[name = string("aw_chunk_3305_cast_fp16")]; + fp16 var_31864_to_fp16 = const()[name = string("op_31864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3307_cast_fp16, y = var_31864_to_fp16)[name = string("aw_chunk_3307_cast_fp16")]; + fp16 var_31866_to_fp16 = const()[name = string("op_31866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3309_cast_fp16, y = var_31866_to_fp16)[name = string("aw_chunk_3309_cast_fp16")]; + fp16 var_31868_to_fp16 = const()[name = string("op_31868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3311_cast_fp16, y = var_31868_to_fp16)[name = string("aw_chunk_3311_cast_fp16")]; + fp16 var_31870_to_fp16 = const()[name = string("op_31870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3313_cast_fp16, y = var_31870_to_fp16)[name = string("aw_chunk_3313_cast_fp16")]; + fp16 var_31872_to_fp16 = const()[name = string("op_31872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3315_cast_fp16, y = var_31872_to_fp16)[name = string("aw_chunk_3315_cast_fp16")]; + fp16 var_31874_to_fp16 = const()[name = string("op_31874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3317_cast_fp16, y = var_31874_to_fp16)[name = string("aw_chunk_3317_cast_fp16")]; + fp16 var_31876_to_fp16 = const()[name = string("op_31876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3319_cast_fp16, y = var_31876_to_fp16)[name = string("aw_chunk_3319_cast_fp16")]; + fp16 var_31878_to_fp16 = const()[name = string("op_31878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3321_cast_fp16, y = var_31878_to_fp16)[name = string("aw_chunk_3321_cast_fp16")]; + fp16 var_31880_to_fp16 = const()[name = string("op_31880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3323_cast_fp16, y = var_31880_to_fp16)[name = string("aw_chunk_3323_cast_fp16")]; + fp16 var_31882_to_fp16 = const()[name = string("op_31882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3325_cast_fp16, y = var_31882_to_fp16)[name = string("aw_chunk_3325_cast_fp16")]; + fp16 var_31884_to_fp16 = const()[name = string("op_31884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3327_cast_fp16, y = var_31884_to_fp16)[name = string("aw_chunk_3327_cast_fp16")]; + fp16 var_31886_to_fp16 = const()[name = string("op_31886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3329_cast_fp16, y = var_31886_to_fp16)[name = string("aw_chunk_3329_cast_fp16")]; + fp16 var_31888_to_fp16 = const()[name = string("op_31888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3331_cast_fp16, y = var_31888_to_fp16)[name = string("aw_chunk_3331_cast_fp16")]; + fp16 var_31890_to_fp16 = const()[name = string("op_31890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3333_cast_fp16, y = var_31890_to_fp16)[name = string("aw_chunk_3333_cast_fp16")]; + fp16 var_31892_to_fp16 = const()[name = string("op_31892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3335_cast_fp16, y = var_31892_to_fp16)[name = string("aw_chunk_3335_cast_fp16")]; + fp16 var_31894_to_fp16 = const()[name = string("op_31894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3337_cast_fp16, y = var_31894_to_fp16)[name = string("aw_chunk_3337_cast_fp16")]; + fp16 var_31896_to_fp16 = const()[name = string("op_31896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3339_cast_fp16, y = var_31896_to_fp16)[name = string("aw_chunk_3339_cast_fp16")]; + fp16 var_31898_to_fp16 = const()[name = string("op_31898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3341_cast_fp16, y = var_31898_to_fp16)[name = string("aw_chunk_3341_cast_fp16")]; + fp16 var_31900_to_fp16 = const()[name = string("op_31900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3343_cast_fp16, y = var_31900_to_fp16)[name = string("aw_chunk_3343_cast_fp16")]; + fp16 var_31902_to_fp16 = const()[name = string("op_31902_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3345_cast_fp16, y = var_31902_to_fp16)[name = string("aw_chunk_3345_cast_fp16")]; + fp16 var_31904_to_fp16 = const()[name = string("op_31904_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3347_cast_fp16, y = var_31904_to_fp16)[name = string("aw_chunk_3347_cast_fp16")]; + fp16 var_31906_to_fp16 = const()[name = string("op_31906_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3349_cast_fp16, y = var_31906_to_fp16)[name = string("aw_chunk_3349_cast_fp16")]; + fp16 var_31908_to_fp16 = const()[name = string("op_31908_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3351_cast_fp16, y = var_31908_to_fp16)[name = string("aw_chunk_3351_cast_fp16")]; + fp16 var_31910_to_fp16 = const()[name = string("op_31910_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3353_cast_fp16, y = var_31910_to_fp16)[name = string("aw_chunk_3353_cast_fp16")]; + fp16 var_31912_to_fp16 = const()[name = string("op_31912_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3355_cast_fp16, y = var_31912_to_fp16)[name = string("aw_chunk_3355_cast_fp16")]; + fp16 var_31914_to_fp16 = const()[name = string("op_31914_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3357_cast_fp16, y = var_31914_to_fp16)[name = string("aw_chunk_3357_cast_fp16")]; + fp16 var_31916_to_fp16 = const()[name = string("op_31916_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3359_cast_fp16, y = var_31916_to_fp16)[name = string("aw_chunk_3359_cast_fp16")]; + tensor var_31918_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3201_cast_fp16)[name = string("op_31918_cast_fp16")]; + tensor var_31919_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3203_cast_fp16)[name = string("op_31919_cast_fp16")]; + tensor var_31920_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3205_cast_fp16)[name = string("op_31920_cast_fp16")]; + tensor var_31921_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3207_cast_fp16)[name = string("op_31921_cast_fp16")]; + tensor var_31922_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3209_cast_fp16)[name = string("op_31922_cast_fp16")]; + tensor var_31923_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3211_cast_fp16)[name = string("op_31923_cast_fp16")]; + tensor var_31924_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3213_cast_fp16)[name = string("op_31924_cast_fp16")]; + tensor var_31925_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3215_cast_fp16)[name = string("op_31925_cast_fp16")]; + tensor var_31926_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3217_cast_fp16)[name = string("op_31926_cast_fp16")]; + tensor var_31927_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3219_cast_fp16)[name = string("op_31927_cast_fp16")]; + tensor var_31928_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3221_cast_fp16)[name = string("op_31928_cast_fp16")]; + tensor var_31929_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3223_cast_fp16)[name = string("op_31929_cast_fp16")]; + tensor var_31930_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3225_cast_fp16)[name = string("op_31930_cast_fp16")]; + tensor var_31931_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3227_cast_fp16)[name = string("op_31931_cast_fp16")]; + tensor var_31932_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3229_cast_fp16)[name = string("op_31932_cast_fp16")]; + tensor var_31933_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3231_cast_fp16)[name = string("op_31933_cast_fp16")]; + tensor var_31934_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3233_cast_fp16)[name = string("op_31934_cast_fp16")]; + tensor var_31935_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3235_cast_fp16)[name = string("op_31935_cast_fp16")]; + tensor var_31936_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3237_cast_fp16)[name = string("op_31936_cast_fp16")]; + tensor var_31937_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3239_cast_fp16)[name = string("op_31937_cast_fp16")]; + tensor var_31938_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3241_cast_fp16)[name = string("op_31938_cast_fp16")]; + tensor var_31939_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3243_cast_fp16)[name = string("op_31939_cast_fp16")]; + tensor var_31940_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3245_cast_fp16)[name = string("op_31940_cast_fp16")]; + tensor var_31941_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3247_cast_fp16)[name = string("op_31941_cast_fp16")]; + tensor var_31942_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3249_cast_fp16)[name = string("op_31942_cast_fp16")]; + tensor var_31943_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3251_cast_fp16)[name = string("op_31943_cast_fp16")]; + tensor var_31944_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3253_cast_fp16)[name = string("op_31944_cast_fp16")]; + tensor var_31945_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3255_cast_fp16)[name = string("op_31945_cast_fp16")]; + tensor var_31946_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3257_cast_fp16)[name = string("op_31946_cast_fp16")]; + tensor var_31947_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3259_cast_fp16)[name = string("op_31947_cast_fp16")]; + tensor var_31948_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3261_cast_fp16)[name = string("op_31948_cast_fp16")]; + tensor var_31949_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3263_cast_fp16)[name = string("op_31949_cast_fp16")]; + tensor var_31950_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3265_cast_fp16)[name = string("op_31950_cast_fp16")]; + tensor var_31951_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3267_cast_fp16)[name = string("op_31951_cast_fp16")]; + tensor var_31952_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3269_cast_fp16)[name = string("op_31952_cast_fp16")]; + tensor var_31953_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3271_cast_fp16)[name = string("op_31953_cast_fp16")]; + tensor var_31954_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3273_cast_fp16)[name = string("op_31954_cast_fp16")]; + tensor var_31955_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3275_cast_fp16)[name = string("op_31955_cast_fp16")]; + tensor var_31956_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3277_cast_fp16)[name = string("op_31956_cast_fp16")]; + tensor var_31957_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3279_cast_fp16)[name = string("op_31957_cast_fp16")]; + tensor var_31958_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3281_cast_fp16)[name = string("op_31958_cast_fp16")]; + tensor var_31959_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3283_cast_fp16)[name = string("op_31959_cast_fp16")]; + tensor var_31960_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3285_cast_fp16)[name = string("op_31960_cast_fp16")]; + tensor var_31961_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3287_cast_fp16)[name = string("op_31961_cast_fp16")]; + tensor var_31962_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3289_cast_fp16)[name = string("op_31962_cast_fp16")]; + tensor var_31963_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3291_cast_fp16)[name = string("op_31963_cast_fp16")]; + tensor var_31964_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3293_cast_fp16)[name = string("op_31964_cast_fp16")]; + tensor var_31965_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3295_cast_fp16)[name = string("op_31965_cast_fp16")]; + tensor var_31966_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3297_cast_fp16)[name = string("op_31966_cast_fp16")]; + tensor var_31967_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3299_cast_fp16)[name = string("op_31967_cast_fp16")]; + tensor var_31968_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3301_cast_fp16)[name = string("op_31968_cast_fp16")]; + tensor var_31969_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3303_cast_fp16)[name = string("op_31969_cast_fp16")]; + tensor var_31970_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3305_cast_fp16)[name = string("op_31970_cast_fp16")]; + tensor var_31971_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3307_cast_fp16)[name = string("op_31971_cast_fp16")]; + tensor var_31972_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3309_cast_fp16)[name = string("op_31972_cast_fp16")]; + tensor var_31973_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3311_cast_fp16)[name = string("op_31973_cast_fp16")]; + tensor var_31974_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3313_cast_fp16)[name = string("op_31974_cast_fp16")]; + tensor var_31975_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3315_cast_fp16)[name = string("op_31975_cast_fp16")]; + tensor var_31976_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3317_cast_fp16)[name = string("op_31976_cast_fp16")]; + tensor var_31977_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3319_cast_fp16)[name = string("op_31977_cast_fp16")]; + tensor var_31978_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3321_cast_fp16)[name = string("op_31978_cast_fp16")]; + tensor var_31979_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3323_cast_fp16)[name = string("op_31979_cast_fp16")]; + tensor var_31980_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3325_cast_fp16)[name = string("op_31980_cast_fp16")]; + tensor var_31981_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3327_cast_fp16)[name = string("op_31981_cast_fp16")]; + tensor var_31982_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3329_cast_fp16)[name = string("op_31982_cast_fp16")]; + tensor var_31983_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3331_cast_fp16)[name = string("op_31983_cast_fp16")]; + tensor var_31984_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3333_cast_fp16)[name = string("op_31984_cast_fp16")]; + tensor var_31985_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3335_cast_fp16)[name = string("op_31985_cast_fp16")]; + tensor var_31986_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3337_cast_fp16)[name = string("op_31986_cast_fp16")]; + tensor var_31987_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3339_cast_fp16)[name = string("op_31987_cast_fp16")]; + tensor var_31988_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3341_cast_fp16)[name = string("op_31988_cast_fp16")]; + tensor var_31989_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3343_cast_fp16)[name = string("op_31989_cast_fp16")]; + tensor var_31990_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3345_cast_fp16)[name = string("op_31990_cast_fp16")]; + tensor var_31991_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3347_cast_fp16)[name = string("op_31991_cast_fp16")]; + tensor var_31992_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3349_cast_fp16)[name = string("op_31992_cast_fp16")]; + tensor var_31993_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3351_cast_fp16)[name = string("op_31993_cast_fp16")]; + tensor var_31994_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3353_cast_fp16)[name = string("op_31994_cast_fp16")]; + tensor var_31995_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3355_cast_fp16)[name = string("op_31995_cast_fp16")]; + tensor var_31996_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3357_cast_fp16)[name = string("op_31996_cast_fp16")]; + tensor var_31997_cast_fp16 = softmax(axis = var_30743, x = aw_chunk_3359_cast_fp16)[name = string("op_31997_cast_fp16")]; + string var_31999_equation_0 = const()[name = string("op_31999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_31999_cast_fp16 = einsum(equation = var_31999_equation_0, values = (var_31519_cast_fp16, var_31918_cast_fp16))[name = string("op_31999_cast_fp16")]; + string var_32001_equation_0 = const()[name = string("op_32001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32001_cast_fp16 = einsum(equation = var_32001_equation_0, values = (var_31519_cast_fp16, var_31919_cast_fp16))[name = string("op_32001_cast_fp16")]; + string var_32003_equation_0 = const()[name = string("op_32003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32003_cast_fp16 = einsum(equation = var_32003_equation_0, values = (var_31519_cast_fp16, var_31920_cast_fp16))[name = string("op_32003_cast_fp16")]; + string var_32005_equation_0 = const()[name = string("op_32005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32005_cast_fp16 = einsum(equation = var_32005_equation_0, values = (var_31519_cast_fp16, var_31921_cast_fp16))[name = string("op_32005_cast_fp16")]; + string var_32007_equation_0 = const()[name = string("op_32007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32007_cast_fp16 = einsum(equation = var_32007_equation_0, values = (var_31523_cast_fp16, var_31922_cast_fp16))[name = string("op_32007_cast_fp16")]; + string var_32009_equation_0 = const()[name = string("op_32009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32009_cast_fp16 = einsum(equation = var_32009_equation_0, values = (var_31523_cast_fp16, var_31923_cast_fp16))[name = string("op_32009_cast_fp16")]; + string var_32011_equation_0 = const()[name = string("op_32011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32011_cast_fp16 = einsum(equation = var_32011_equation_0, values = (var_31523_cast_fp16, var_31924_cast_fp16))[name = string("op_32011_cast_fp16")]; + string var_32013_equation_0 = const()[name = string("op_32013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32013_cast_fp16 = einsum(equation = var_32013_equation_0, values = (var_31523_cast_fp16, var_31925_cast_fp16))[name = string("op_32013_cast_fp16")]; + string var_32015_equation_0 = const()[name = string("op_32015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32015_cast_fp16 = einsum(equation = var_32015_equation_0, values = (var_31527_cast_fp16, var_31926_cast_fp16))[name = string("op_32015_cast_fp16")]; + string var_32017_equation_0 = const()[name = string("op_32017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32017_cast_fp16 = einsum(equation = var_32017_equation_0, values = (var_31527_cast_fp16, var_31927_cast_fp16))[name = string("op_32017_cast_fp16")]; + string var_32019_equation_0 = const()[name = string("op_32019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32019_cast_fp16 = einsum(equation = var_32019_equation_0, values = (var_31527_cast_fp16, var_31928_cast_fp16))[name = string("op_32019_cast_fp16")]; + string var_32021_equation_0 = const()[name = string("op_32021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32021_cast_fp16 = einsum(equation = var_32021_equation_0, values = (var_31527_cast_fp16, var_31929_cast_fp16))[name = string("op_32021_cast_fp16")]; + string var_32023_equation_0 = const()[name = string("op_32023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32023_cast_fp16 = einsum(equation = var_32023_equation_0, values = (var_31531_cast_fp16, var_31930_cast_fp16))[name = string("op_32023_cast_fp16")]; + string var_32025_equation_0 = const()[name = string("op_32025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32025_cast_fp16 = einsum(equation = var_32025_equation_0, values = (var_31531_cast_fp16, var_31931_cast_fp16))[name = string("op_32025_cast_fp16")]; + string var_32027_equation_0 = const()[name = string("op_32027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32027_cast_fp16 = einsum(equation = var_32027_equation_0, values = (var_31531_cast_fp16, var_31932_cast_fp16))[name = string("op_32027_cast_fp16")]; + string var_32029_equation_0 = const()[name = string("op_32029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32029_cast_fp16 = einsum(equation = var_32029_equation_0, values = (var_31531_cast_fp16, var_31933_cast_fp16))[name = string("op_32029_cast_fp16")]; + string var_32031_equation_0 = const()[name = string("op_32031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32031_cast_fp16 = einsum(equation = var_32031_equation_0, values = (var_31535_cast_fp16, var_31934_cast_fp16))[name = string("op_32031_cast_fp16")]; + string var_32033_equation_0 = const()[name = string("op_32033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32033_cast_fp16 = einsum(equation = var_32033_equation_0, values = (var_31535_cast_fp16, var_31935_cast_fp16))[name = string("op_32033_cast_fp16")]; + string var_32035_equation_0 = const()[name = string("op_32035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32035_cast_fp16 = einsum(equation = var_32035_equation_0, values = (var_31535_cast_fp16, var_31936_cast_fp16))[name = string("op_32035_cast_fp16")]; + string var_32037_equation_0 = const()[name = string("op_32037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32037_cast_fp16 = einsum(equation = var_32037_equation_0, values = (var_31535_cast_fp16, var_31937_cast_fp16))[name = string("op_32037_cast_fp16")]; + string var_32039_equation_0 = const()[name = string("op_32039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32039_cast_fp16 = einsum(equation = var_32039_equation_0, values = (var_31539_cast_fp16, var_31938_cast_fp16))[name = string("op_32039_cast_fp16")]; + string var_32041_equation_0 = const()[name = string("op_32041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32041_cast_fp16 = einsum(equation = var_32041_equation_0, values = (var_31539_cast_fp16, var_31939_cast_fp16))[name = string("op_32041_cast_fp16")]; + string var_32043_equation_0 = const()[name = string("op_32043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32043_cast_fp16 = einsum(equation = var_32043_equation_0, values = (var_31539_cast_fp16, var_31940_cast_fp16))[name = string("op_32043_cast_fp16")]; + string var_32045_equation_0 = const()[name = string("op_32045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32045_cast_fp16 = einsum(equation = var_32045_equation_0, values = (var_31539_cast_fp16, var_31941_cast_fp16))[name = string("op_32045_cast_fp16")]; + string var_32047_equation_0 = const()[name = string("op_32047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32047_cast_fp16 = einsum(equation = var_32047_equation_0, values = (var_31543_cast_fp16, var_31942_cast_fp16))[name = string("op_32047_cast_fp16")]; + string var_32049_equation_0 = const()[name = string("op_32049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32049_cast_fp16 = einsum(equation = var_32049_equation_0, values = (var_31543_cast_fp16, var_31943_cast_fp16))[name = string("op_32049_cast_fp16")]; + string var_32051_equation_0 = const()[name = string("op_32051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32051_cast_fp16 = einsum(equation = var_32051_equation_0, values = (var_31543_cast_fp16, var_31944_cast_fp16))[name = string("op_32051_cast_fp16")]; + string var_32053_equation_0 = const()[name = string("op_32053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32053_cast_fp16 = einsum(equation = var_32053_equation_0, values = (var_31543_cast_fp16, var_31945_cast_fp16))[name = string("op_32053_cast_fp16")]; + string var_32055_equation_0 = const()[name = string("op_32055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32055_cast_fp16 = einsum(equation = var_32055_equation_0, values = (var_31547_cast_fp16, var_31946_cast_fp16))[name = string("op_32055_cast_fp16")]; + string var_32057_equation_0 = const()[name = string("op_32057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32057_cast_fp16 = einsum(equation = var_32057_equation_0, values = (var_31547_cast_fp16, var_31947_cast_fp16))[name = string("op_32057_cast_fp16")]; + string var_32059_equation_0 = const()[name = string("op_32059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32059_cast_fp16 = einsum(equation = var_32059_equation_0, values = (var_31547_cast_fp16, var_31948_cast_fp16))[name = string("op_32059_cast_fp16")]; + string var_32061_equation_0 = const()[name = string("op_32061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32061_cast_fp16 = einsum(equation = var_32061_equation_0, values = (var_31547_cast_fp16, var_31949_cast_fp16))[name = string("op_32061_cast_fp16")]; + string var_32063_equation_0 = const()[name = string("op_32063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32063_cast_fp16 = einsum(equation = var_32063_equation_0, values = (var_31551_cast_fp16, var_31950_cast_fp16))[name = string("op_32063_cast_fp16")]; + string var_32065_equation_0 = const()[name = string("op_32065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32065_cast_fp16 = einsum(equation = var_32065_equation_0, values = (var_31551_cast_fp16, var_31951_cast_fp16))[name = string("op_32065_cast_fp16")]; + string var_32067_equation_0 = const()[name = string("op_32067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32067_cast_fp16 = einsum(equation = var_32067_equation_0, values = (var_31551_cast_fp16, var_31952_cast_fp16))[name = string("op_32067_cast_fp16")]; + string var_32069_equation_0 = const()[name = string("op_32069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32069_cast_fp16 = einsum(equation = var_32069_equation_0, values = (var_31551_cast_fp16, var_31953_cast_fp16))[name = string("op_32069_cast_fp16")]; + string var_32071_equation_0 = const()[name = string("op_32071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32071_cast_fp16 = einsum(equation = var_32071_equation_0, values = (var_31555_cast_fp16, var_31954_cast_fp16))[name = string("op_32071_cast_fp16")]; + string var_32073_equation_0 = const()[name = string("op_32073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32073_cast_fp16 = einsum(equation = var_32073_equation_0, values = (var_31555_cast_fp16, var_31955_cast_fp16))[name = string("op_32073_cast_fp16")]; + string var_32075_equation_0 = const()[name = string("op_32075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32075_cast_fp16 = einsum(equation = var_32075_equation_0, values = (var_31555_cast_fp16, var_31956_cast_fp16))[name = string("op_32075_cast_fp16")]; + string var_32077_equation_0 = const()[name = string("op_32077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32077_cast_fp16 = einsum(equation = var_32077_equation_0, values = (var_31555_cast_fp16, var_31957_cast_fp16))[name = string("op_32077_cast_fp16")]; + string var_32079_equation_0 = const()[name = string("op_32079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32079_cast_fp16 = einsum(equation = var_32079_equation_0, values = (var_31559_cast_fp16, var_31958_cast_fp16))[name = string("op_32079_cast_fp16")]; + string var_32081_equation_0 = const()[name = string("op_32081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32081_cast_fp16 = einsum(equation = var_32081_equation_0, values = (var_31559_cast_fp16, var_31959_cast_fp16))[name = string("op_32081_cast_fp16")]; + string var_32083_equation_0 = const()[name = string("op_32083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32083_cast_fp16 = einsum(equation = var_32083_equation_0, values = (var_31559_cast_fp16, var_31960_cast_fp16))[name = string("op_32083_cast_fp16")]; + string var_32085_equation_0 = const()[name = string("op_32085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32085_cast_fp16 = einsum(equation = var_32085_equation_0, values = (var_31559_cast_fp16, var_31961_cast_fp16))[name = string("op_32085_cast_fp16")]; + string var_32087_equation_0 = const()[name = string("op_32087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32087_cast_fp16 = einsum(equation = var_32087_equation_0, values = (var_31563_cast_fp16, var_31962_cast_fp16))[name = string("op_32087_cast_fp16")]; + string var_32089_equation_0 = const()[name = string("op_32089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32089_cast_fp16 = einsum(equation = var_32089_equation_0, values = (var_31563_cast_fp16, var_31963_cast_fp16))[name = string("op_32089_cast_fp16")]; + string var_32091_equation_0 = const()[name = string("op_32091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32091_cast_fp16 = einsum(equation = var_32091_equation_0, values = (var_31563_cast_fp16, var_31964_cast_fp16))[name = string("op_32091_cast_fp16")]; + string var_32093_equation_0 = const()[name = string("op_32093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32093_cast_fp16 = einsum(equation = var_32093_equation_0, values = (var_31563_cast_fp16, var_31965_cast_fp16))[name = string("op_32093_cast_fp16")]; + string var_32095_equation_0 = const()[name = string("op_32095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32095_cast_fp16 = einsum(equation = var_32095_equation_0, values = (var_31567_cast_fp16, var_31966_cast_fp16))[name = string("op_32095_cast_fp16")]; + string var_32097_equation_0 = const()[name = string("op_32097_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32097_cast_fp16 = einsum(equation = var_32097_equation_0, values = (var_31567_cast_fp16, var_31967_cast_fp16))[name = string("op_32097_cast_fp16")]; + string var_32099_equation_0 = const()[name = string("op_32099_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32099_cast_fp16 = einsum(equation = var_32099_equation_0, values = (var_31567_cast_fp16, var_31968_cast_fp16))[name = string("op_32099_cast_fp16")]; + string var_32101_equation_0 = const()[name = string("op_32101_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32101_cast_fp16 = einsum(equation = var_32101_equation_0, values = (var_31567_cast_fp16, var_31969_cast_fp16))[name = string("op_32101_cast_fp16")]; + string var_32103_equation_0 = const()[name = string("op_32103_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32103_cast_fp16 = einsum(equation = var_32103_equation_0, values = (var_31571_cast_fp16, var_31970_cast_fp16))[name = string("op_32103_cast_fp16")]; + string var_32105_equation_0 = const()[name = string("op_32105_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32105_cast_fp16 = einsum(equation = var_32105_equation_0, values = (var_31571_cast_fp16, var_31971_cast_fp16))[name = string("op_32105_cast_fp16")]; + string var_32107_equation_0 = const()[name = string("op_32107_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32107_cast_fp16 = einsum(equation = var_32107_equation_0, values = (var_31571_cast_fp16, var_31972_cast_fp16))[name = string("op_32107_cast_fp16")]; + string var_32109_equation_0 = const()[name = string("op_32109_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32109_cast_fp16 = einsum(equation = var_32109_equation_0, values = (var_31571_cast_fp16, var_31973_cast_fp16))[name = string("op_32109_cast_fp16")]; + string var_32111_equation_0 = const()[name = string("op_32111_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32111_cast_fp16 = einsum(equation = var_32111_equation_0, values = (var_31575_cast_fp16, var_31974_cast_fp16))[name = string("op_32111_cast_fp16")]; + string var_32113_equation_0 = const()[name = string("op_32113_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32113_cast_fp16 = einsum(equation = var_32113_equation_0, values = (var_31575_cast_fp16, var_31975_cast_fp16))[name = string("op_32113_cast_fp16")]; + string var_32115_equation_0 = const()[name = string("op_32115_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32115_cast_fp16 = einsum(equation = var_32115_equation_0, values = (var_31575_cast_fp16, var_31976_cast_fp16))[name = string("op_32115_cast_fp16")]; + string var_32117_equation_0 = const()[name = string("op_32117_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32117_cast_fp16 = einsum(equation = var_32117_equation_0, values = (var_31575_cast_fp16, var_31977_cast_fp16))[name = string("op_32117_cast_fp16")]; + string var_32119_equation_0 = const()[name = string("op_32119_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32119_cast_fp16 = einsum(equation = var_32119_equation_0, values = (var_31579_cast_fp16, var_31978_cast_fp16))[name = string("op_32119_cast_fp16")]; + string var_32121_equation_0 = const()[name = string("op_32121_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32121_cast_fp16 = einsum(equation = var_32121_equation_0, values = (var_31579_cast_fp16, var_31979_cast_fp16))[name = string("op_32121_cast_fp16")]; + string var_32123_equation_0 = const()[name = string("op_32123_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32123_cast_fp16 = einsum(equation = var_32123_equation_0, values = (var_31579_cast_fp16, var_31980_cast_fp16))[name = string("op_32123_cast_fp16")]; + string var_32125_equation_0 = const()[name = string("op_32125_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32125_cast_fp16 = einsum(equation = var_32125_equation_0, values = (var_31579_cast_fp16, var_31981_cast_fp16))[name = string("op_32125_cast_fp16")]; + string var_32127_equation_0 = const()[name = string("op_32127_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32127_cast_fp16 = einsum(equation = var_32127_equation_0, values = (var_31583_cast_fp16, var_31982_cast_fp16))[name = string("op_32127_cast_fp16")]; + string var_32129_equation_0 = const()[name = string("op_32129_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32129_cast_fp16 = einsum(equation = var_32129_equation_0, values = (var_31583_cast_fp16, var_31983_cast_fp16))[name = string("op_32129_cast_fp16")]; + string var_32131_equation_0 = const()[name = string("op_32131_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32131_cast_fp16 = einsum(equation = var_32131_equation_0, values = (var_31583_cast_fp16, var_31984_cast_fp16))[name = string("op_32131_cast_fp16")]; + string var_32133_equation_0 = const()[name = string("op_32133_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32133_cast_fp16 = einsum(equation = var_32133_equation_0, values = (var_31583_cast_fp16, var_31985_cast_fp16))[name = string("op_32133_cast_fp16")]; + string var_32135_equation_0 = const()[name = string("op_32135_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32135_cast_fp16 = einsum(equation = var_32135_equation_0, values = (var_31587_cast_fp16, var_31986_cast_fp16))[name = string("op_32135_cast_fp16")]; + string var_32137_equation_0 = const()[name = string("op_32137_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32137_cast_fp16 = einsum(equation = var_32137_equation_0, values = (var_31587_cast_fp16, var_31987_cast_fp16))[name = string("op_32137_cast_fp16")]; + string var_32139_equation_0 = const()[name = string("op_32139_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32139_cast_fp16 = einsum(equation = var_32139_equation_0, values = (var_31587_cast_fp16, var_31988_cast_fp16))[name = string("op_32139_cast_fp16")]; + string var_32141_equation_0 = const()[name = string("op_32141_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32141_cast_fp16 = einsum(equation = var_32141_equation_0, values = (var_31587_cast_fp16, var_31989_cast_fp16))[name = string("op_32141_cast_fp16")]; + string var_32143_equation_0 = const()[name = string("op_32143_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32143_cast_fp16 = einsum(equation = var_32143_equation_0, values = (var_31591_cast_fp16, var_31990_cast_fp16))[name = string("op_32143_cast_fp16")]; + string var_32145_equation_0 = const()[name = string("op_32145_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32145_cast_fp16 = einsum(equation = var_32145_equation_0, values = (var_31591_cast_fp16, var_31991_cast_fp16))[name = string("op_32145_cast_fp16")]; + string var_32147_equation_0 = const()[name = string("op_32147_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32147_cast_fp16 = einsum(equation = var_32147_equation_0, values = (var_31591_cast_fp16, var_31992_cast_fp16))[name = string("op_32147_cast_fp16")]; + string var_32149_equation_0 = const()[name = string("op_32149_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32149_cast_fp16 = einsum(equation = var_32149_equation_0, values = (var_31591_cast_fp16, var_31993_cast_fp16))[name = string("op_32149_cast_fp16")]; + string var_32151_equation_0 = const()[name = string("op_32151_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32151_cast_fp16 = einsum(equation = var_32151_equation_0, values = (var_31595_cast_fp16, var_31994_cast_fp16))[name = string("op_32151_cast_fp16")]; + string var_32153_equation_0 = const()[name = string("op_32153_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32153_cast_fp16 = einsum(equation = var_32153_equation_0, values = (var_31595_cast_fp16, var_31995_cast_fp16))[name = string("op_32153_cast_fp16")]; + string var_32155_equation_0 = const()[name = string("op_32155_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32155_cast_fp16 = einsum(equation = var_32155_equation_0, values = (var_31595_cast_fp16, var_31996_cast_fp16))[name = string("op_32155_cast_fp16")]; + string var_32157_equation_0 = const()[name = string("op_32157_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32157_cast_fp16 = einsum(equation = var_32157_equation_0, values = (var_31595_cast_fp16, var_31997_cast_fp16))[name = string("op_32157_cast_fp16")]; + bool var_32159_interleave_0 = const()[name = string("op_32159_interleave_0"), val = bool(false)]; + tensor var_32159_cast_fp16 = concat(axis = var_30718, interleave = var_32159_interleave_0, values = (var_31999_cast_fp16, var_32001_cast_fp16, var_32003_cast_fp16, var_32005_cast_fp16))[name = string("op_32159_cast_fp16")]; + bool var_32161_interleave_0 = const()[name = string("op_32161_interleave_0"), val = bool(false)]; + tensor var_32161_cast_fp16 = concat(axis = var_30718, interleave = var_32161_interleave_0, values = (var_32007_cast_fp16, var_32009_cast_fp16, var_32011_cast_fp16, var_32013_cast_fp16))[name = string("op_32161_cast_fp16")]; + bool var_32163_interleave_0 = const()[name = string("op_32163_interleave_0"), val = bool(false)]; + tensor var_32163_cast_fp16 = concat(axis = var_30718, interleave = var_32163_interleave_0, values = (var_32015_cast_fp16, var_32017_cast_fp16, var_32019_cast_fp16, var_32021_cast_fp16))[name = string("op_32163_cast_fp16")]; + bool var_32165_interleave_0 = const()[name = string("op_32165_interleave_0"), val = bool(false)]; + tensor var_32165_cast_fp16 = concat(axis = var_30718, interleave = var_32165_interleave_0, values = (var_32023_cast_fp16, var_32025_cast_fp16, var_32027_cast_fp16, var_32029_cast_fp16))[name = string("op_32165_cast_fp16")]; + bool var_32167_interleave_0 = const()[name = string("op_32167_interleave_0"), val = bool(false)]; + tensor var_32167_cast_fp16 = concat(axis = var_30718, interleave = var_32167_interleave_0, values = (var_32031_cast_fp16, var_32033_cast_fp16, var_32035_cast_fp16, var_32037_cast_fp16))[name = string("op_32167_cast_fp16")]; + bool var_32169_interleave_0 = const()[name = string("op_32169_interleave_0"), val = bool(false)]; + tensor var_32169_cast_fp16 = concat(axis = var_30718, interleave = var_32169_interleave_0, values = (var_32039_cast_fp16, var_32041_cast_fp16, var_32043_cast_fp16, var_32045_cast_fp16))[name = string("op_32169_cast_fp16")]; + bool var_32171_interleave_0 = const()[name = string("op_32171_interleave_0"), val = bool(false)]; + tensor var_32171_cast_fp16 = concat(axis = var_30718, interleave = var_32171_interleave_0, values = (var_32047_cast_fp16, var_32049_cast_fp16, var_32051_cast_fp16, var_32053_cast_fp16))[name = string("op_32171_cast_fp16")]; + bool var_32173_interleave_0 = const()[name = string("op_32173_interleave_0"), val = bool(false)]; + tensor var_32173_cast_fp16 = concat(axis = var_30718, interleave = var_32173_interleave_0, values = (var_32055_cast_fp16, var_32057_cast_fp16, var_32059_cast_fp16, var_32061_cast_fp16))[name = string("op_32173_cast_fp16")]; + bool var_32175_interleave_0 = const()[name = string("op_32175_interleave_0"), val = bool(false)]; + tensor var_32175_cast_fp16 = concat(axis = var_30718, interleave = var_32175_interleave_0, values = (var_32063_cast_fp16, var_32065_cast_fp16, var_32067_cast_fp16, var_32069_cast_fp16))[name = string("op_32175_cast_fp16")]; + bool var_32177_interleave_0 = const()[name = string("op_32177_interleave_0"), val = bool(false)]; + tensor var_32177_cast_fp16 = concat(axis = var_30718, interleave = var_32177_interleave_0, values = (var_32071_cast_fp16, var_32073_cast_fp16, var_32075_cast_fp16, var_32077_cast_fp16))[name = string("op_32177_cast_fp16")]; + bool var_32179_interleave_0 = const()[name = string("op_32179_interleave_0"), val = bool(false)]; + tensor var_32179_cast_fp16 = concat(axis = var_30718, interleave = var_32179_interleave_0, values = (var_32079_cast_fp16, var_32081_cast_fp16, var_32083_cast_fp16, var_32085_cast_fp16))[name = string("op_32179_cast_fp16")]; + bool var_32181_interleave_0 = const()[name = string("op_32181_interleave_0"), val = bool(false)]; + tensor var_32181_cast_fp16 = concat(axis = var_30718, interleave = var_32181_interleave_0, values = (var_32087_cast_fp16, var_32089_cast_fp16, var_32091_cast_fp16, var_32093_cast_fp16))[name = string("op_32181_cast_fp16")]; + bool var_32183_interleave_0 = const()[name = string("op_32183_interleave_0"), val = bool(false)]; + tensor var_32183_cast_fp16 = concat(axis = var_30718, interleave = var_32183_interleave_0, values = (var_32095_cast_fp16, var_32097_cast_fp16, var_32099_cast_fp16, var_32101_cast_fp16))[name = string("op_32183_cast_fp16")]; + bool var_32185_interleave_0 = const()[name = string("op_32185_interleave_0"), val = bool(false)]; + tensor var_32185_cast_fp16 = concat(axis = var_30718, interleave = var_32185_interleave_0, values = (var_32103_cast_fp16, var_32105_cast_fp16, var_32107_cast_fp16, var_32109_cast_fp16))[name = string("op_32185_cast_fp16")]; + bool var_32187_interleave_0 = const()[name = string("op_32187_interleave_0"), val = bool(false)]; + tensor var_32187_cast_fp16 = concat(axis = var_30718, interleave = var_32187_interleave_0, values = (var_32111_cast_fp16, var_32113_cast_fp16, var_32115_cast_fp16, var_32117_cast_fp16))[name = string("op_32187_cast_fp16")]; + bool var_32189_interleave_0 = const()[name = string("op_32189_interleave_0"), val = bool(false)]; + tensor var_32189_cast_fp16 = concat(axis = var_30718, interleave = var_32189_interleave_0, values = (var_32119_cast_fp16, var_32121_cast_fp16, var_32123_cast_fp16, var_32125_cast_fp16))[name = string("op_32189_cast_fp16")]; + bool var_32191_interleave_0 = const()[name = string("op_32191_interleave_0"), val = bool(false)]; + tensor var_32191_cast_fp16 = concat(axis = var_30718, interleave = var_32191_interleave_0, values = (var_32127_cast_fp16, var_32129_cast_fp16, var_32131_cast_fp16, var_32133_cast_fp16))[name = string("op_32191_cast_fp16")]; + bool var_32193_interleave_0 = const()[name = string("op_32193_interleave_0"), val = bool(false)]; + tensor var_32193_cast_fp16 = concat(axis = var_30718, interleave = var_32193_interleave_0, values = (var_32135_cast_fp16, var_32137_cast_fp16, var_32139_cast_fp16, var_32141_cast_fp16))[name = string("op_32193_cast_fp16")]; + bool var_32195_interleave_0 = const()[name = string("op_32195_interleave_0"), val = bool(false)]; + tensor var_32195_cast_fp16 = concat(axis = var_30718, interleave = var_32195_interleave_0, values = (var_32143_cast_fp16, var_32145_cast_fp16, var_32147_cast_fp16, var_32149_cast_fp16))[name = string("op_32195_cast_fp16")]; + bool var_32197_interleave_0 = const()[name = string("op_32197_interleave_0"), val = bool(false)]; + tensor var_32197_cast_fp16 = concat(axis = var_30718, interleave = var_32197_interleave_0, values = (var_32151_cast_fp16, var_32153_cast_fp16, var_32155_cast_fp16, var_32157_cast_fp16))[name = string("op_32197_cast_fp16")]; + bool input_161_interleave_0 = const()[name = string("input_161_interleave_0"), val = bool(false)]; + tensor input_161_cast_fp16 = concat(axis = var_30743, interleave = input_161_interleave_0, values = (var_32159_cast_fp16, var_32161_cast_fp16, var_32163_cast_fp16, var_32165_cast_fp16, var_32167_cast_fp16, var_32169_cast_fp16, var_32171_cast_fp16, var_32173_cast_fp16, var_32175_cast_fp16, var_32177_cast_fp16, var_32179_cast_fp16, var_32181_cast_fp16, var_32183_cast_fp16, var_32185_cast_fp16, var_32187_cast_fp16, var_32189_cast_fp16, var_32191_cast_fp16, var_32193_cast_fp16, var_32195_cast_fp16, var_32197_cast_fp16))[name = string("input_161_cast_fp16")]; + string obj_83_pad_type_0 = const()[name = string("obj_83_pad_type_0"), val = string("valid")]; + tensor obj_83_strides_0 = const()[name = string("obj_83_strides_0"), val = tensor([1, 1])]; + tensor obj_83_pad_0 = const()[name = string("obj_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_83_dilations_0 = const()[name = string("obj_83_dilations_0"), val = tensor([1, 1])]; + int32 obj_83_groups_0 = const()[name = string("obj_83_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811570880)))]; + tensor layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814847744)))]; + tensor obj_83_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_83_dilations_0, groups = obj_83_groups_0, pad = obj_83_pad_0, pad_type = obj_83_pad_type_0, strides = obj_83_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_161_cast_fp16)[name = string("obj_83_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_83_cast_fp16)[name = string("inputs_83_cast_fp16")]; + tensor out_83_axes_0 = const()[name = string("out_83_axes_0"), val = tensor([1])]; + fp16 var_32216_to_fp16 = const()[name = string("op_32216_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_32216_to_fp16, x = inputs_83_cast_fp16)[name = string("out_83_cast_fp16")]; + tensor input_163_gamma_0_to_fp16 = const()[name = string("input_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814850368)))]; + tensor input_163_beta_0_to_fp16 = const()[name = string("input_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814852992)))]; + fp16 input_163_epsilon_0_to_fp16 = const()[name = string("input_163_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_163_cast_fp16 = batch_norm(beta = input_163_beta_0_to_fp16, epsilon = input_163_epsilon_0_to_fp16, gamma = input_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = string("input_163_cast_fp16")]; + string input_165_pad_type_0 = const()[name = string("input_165_pad_type_0"), val = string("valid")]; + tensor input_165_strides_0 = const()[name = string("input_165_strides_0"), val = tensor([1, 1])]; + tensor input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_165_dilations_0 = const()[name = string("input_165_dilations_0"), val = tensor([1, 1])]; + int32 input_165_groups_0 = const()[name = string("input_165_groups_0"), val = int32(1)]; + tensor layers_20_fc1_weight_to_fp16 = const()[name = string("layers_20_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814855616)))]; + tensor layers_20_fc1_bias_to_fp16 = const()[name = string("layers_20_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827962880)))]; + tensor input_165_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_165_dilations_0, groups = input_165_groups_0, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = input_165_strides_0, weight = layers_20_fc1_weight_to_fp16, x = input_163_cast_fp16)[name = string("input_165_cast_fp16")]; + string input_167_mode_0 = const()[name = string("input_167_mode_0"), val = string("EXACT")]; + tensor input_167_cast_fp16 = gelu(mode = input_167_mode_0, x = input_165_cast_fp16)[name = string("input_167_cast_fp16")]; + string hidden_states_45_pad_type_0 = const()[name = string("hidden_states_45_pad_type_0"), val = string("valid")]; + tensor hidden_states_45_strides_0 = const()[name = string("hidden_states_45_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_45_pad_0 = const()[name = string("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_45_dilations_0 = const()[name = string("hidden_states_45_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_45_groups_0 = const()[name = string("hidden_states_45_groups_0"), val = int32(1)]; + tensor layers_20_fc2_weight_to_fp16 = const()[name = string("layers_20_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827973184)))]; + tensor layers_20_fc2_bias_to_fp16 = const()[name = string("layers_20_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841080448)))]; + tensor hidden_states_45_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_20_fc2_weight_to_fp16, x = input_167_cast_fp16)[name = string("hidden_states_45_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_45_cast_fp16)[name = string("inputs_85_cast_fp16")]; + int32 var_32245 = const()[name = string("op_32245"), val = int32(3)]; + int32 var_32270 = const()[name = string("op_32270"), val = int32(1)]; + tensor out_85_axes_0 = const()[name = string("out_85_axes_0"), val = tensor([1])]; + fp16 var_32287_to_fp16 = const()[name = string("op_32287_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_32287_to_fp16, x = inputs_85_cast_fp16)[name = string("out_85_cast_fp16")]; + tensor obj_85_gamma_0_to_fp16 = const()[name = string("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841083072)))]; + tensor obj_85_beta_0_to_fp16 = const()[name = string("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841085696)))]; + fp16 obj_85_epsilon_0_to_fp16 = const()[name = string("obj_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = string("obj_85_cast_fp16")]; + string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; + tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; + tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; + int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841088320)))]; + tensor layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844365184)))]; + tensor query_43_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("query_43_cast_fp16")]; + string key_43_pad_type_0 = const()[name = string("key_43_pad_type_0"), val = string("valid")]; + tensor key_43_strides_0 = const()[name = string("key_43_strides_0"), val = tensor([1, 1])]; + tensor key_43_pad_0 = const()[name = string("key_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_43_dilations_0 = const()[name = string("key_43_dilations_0"), val = tensor([1, 1])]; + int32 key_43_groups_0 = const()[name = string("key_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844367808)))]; + tensor key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("key_43_cast_fp16")]; + string value_43_pad_type_0 = const()[name = string("value_43_pad_type_0"), val = string("valid")]; + tensor value_43_strides_0 = const()[name = string("value_43_strides_0"), val = tensor([1, 1])]; + tensor value_43_pad_0 = const()[name = string("value_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_43_dilations_0 = const()[name = string("value_43_dilations_0"), val = tensor([1, 1])]; + int32 value_43_groups_0 = const()[name = string("value_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(847644672)))]; + tensor layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850921536)))]; + tensor value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("value_43_cast_fp16")]; + tensor var_32325_begin_0 = const()[name = string("op_32325_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32325_end_0 = const()[name = string("op_32325_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32325_end_mask_0 = const()[name = string("op_32325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32325_cast_fp16 = slice_by_index(begin = var_32325_begin_0, end = var_32325_end_0, end_mask = var_32325_end_mask_0, x = query_43_cast_fp16)[name = string("op_32325_cast_fp16")]; + tensor var_32329_begin_0 = const()[name = string("op_32329_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32329_end_0 = const()[name = string("op_32329_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_32329_end_mask_0 = const()[name = string("op_32329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32329_cast_fp16 = slice_by_index(begin = var_32329_begin_0, end = var_32329_end_0, end_mask = var_32329_end_mask_0, x = query_43_cast_fp16)[name = string("op_32329_cast_fp16")]; + tensor var_32333_begin_0 = const()[name = string("op_32333_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32333_end_0 = const()[name = string("op_32333_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_32333_end_mask_0 = const()[name = string("op_32333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32333_cast_fp16 = slice_by_index(begin = var_32333_begin_0, end = var_32333_end_0, end_mask = var_32333_end_mask_0, x = query_43_cast_fp16)[name = string("op_32333_cast_fp16")]; + tensor var_32337_begin_0 = const()[name = string("op_32337_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32337_end_0 = const()[name = string("op_32337_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_32337_end_mask_0 = const()[name = string("op_32337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32337_cast_fp16 = slice_by_index(begin = var_32337_begin_0, end = var_32337_end_0, end_mask = var_32337_end_mask_0, x = query_43_cast_fp16)[name = string("op_32337_cast_fp16")]; + tensor var_32341_begin_0 = const()[name = string("op_32341_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32341_end_0 = const()[name = string("op_32341_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_32341_end_mask_0 = const()[name = string("op_32341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32341_cast_fp16 = slice_by_index(begin = var_32341_begin_0, end = var_32341_end_0, end_mask = var_32341_end_mask_0, x = query_43_cast_fp16)[name = string("op_32341_cast_fp16")]; + tensor var_32345_begin_0 = const()[name = string("op_32345_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32345_end_0 = const()[name = string("op_32345_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_32345_end_mask_0 = const()[name = string("op_32345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32345_cast_fp16 = slice_by_index(begin = var_32345_begin_0, end = var_32345_end_0, end_mask = var_32345_end_mask_0, x = query_43_cast_fp16)[name = string("op_32345_cast_fp16")]; + tensor var_32349_begin_0 = const()[name = string("op_32349_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32349_end_0 = const()[name = string("op_32349_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_32349_end_mask_0 = const()[name = string("op_32349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32349_cast_fp16 = slice_by_index(begin = var_32349_begin_0, end = var_32349_end_0, end_mask = var_32349_end_mask_0, x = query_43_cast_fp16)[name = string("op_32349_cast_fp16")]; + tensor var_32353_begin_0 = const()[name = string("op_32353_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32353_end_0 = const()[name = string("op_32353_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_32353_end_mask_0 = const()[name = string("op_32353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32353_cast_fp16 = slice_by_index(begin = var_32353_begin_0, end = var_32353_end_0, end_mask = var_32353_end_mask_0, x = query_43_cast_fp16)[name = string("op_32353_cast_fp16")]; + tensor var_32357_begin_0 = const()[name = string("op_32357_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32357_end_0 = const()[name = string("op_32357_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_32357_end_mask_0 = const()[name = string("op_32357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32357_cast_fp16 = slice_by_index(begin = var_32357_begin_0, end = var_32357_end_0, end_mask = var_32357_end_mask_0, x = query_43_cast_fp16)[name = string("op_32357_cast_fp16")]; + tensor var_32361_begin_0 = const()[name = string("op_32361_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32361_end_0 = const()[name = string("op_32361_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_32361_end_mask_0 = const()[name = string("op_32361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32361_cast_fp16 = slice_by_index(begin = var_32361_begin_0, end = var_32361_end_0, end_mask = var_32361_end_mask_0, x = query_43_cast_fp16)[name = string("op_32361_cast_fp16")]; + tensor var_32365_begin_0 = const()[name = string("op_32365_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32365_end_0 = const()[name = string("op_32365_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_32365_end_mask_0 = const()[name = string("op_32365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32365_cast_fp16 = slice_by_index(begin = var_32365_begin_0, end = var_32365_end_0, end_mask = var_32365_end_mask_0, x = query_43_cast_fp16)[name = string("op_32365_cast_fp16")]; + tensor var_32369_begin_0 = const()[name = string("op_32369_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32369_end_0 = const()[name = string("op_32369_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_32369_end_mask_0 = const()[name = string("op_32369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32369_cast_fp16 = slice_by_index(begin = var_32369_begin_0, end = var_32369_end_0, end_mask = var_32369_end_mask_0, x = query_43_cast_fp16)[name = string("op_32369_cast_fp16")]; + tensor var_32373_begin_0 = const()[name = string("op_32373_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32373_end_0 = const()[name = string("op_32373_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_32373_end_mask_0 = const()[name = string("op_32373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32373_cast_fp16 = slice_by_index(begin = var_32373_begin_0, end = var_32373_end_0, end_mask = var_32373_end_mask_0, x = query_43_cast_fp16)[name = string("op_32373_cast_fp16")]; + tensor var_32377_begin_0 = const()[name = string("op_32377_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32377_end_0 = const()[name = string("op_32377_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_32377_end_mask_0 = const()[name = string("op_32377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32377_cast_fp16 = slice_by_index(begin = var_32377_begin_0, end = var_32377_end_0, end_mask = var_32377_end_mask_0, x = query_43_cast_fp16)[name = string("op_32377_cast_fp16")]; + tensor var_32381_begin_0 = const()[name = string("op_32381_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32381_end_0 = const()[name = string("op_32381_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_32381_end_mask_0 = const()[name = string("op_32381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32381_cast_fp16 = slice_by_index(begin = var_32381_begin_0, end = var_32381_end_0, end_mask = var_32381_end_mask_0, x = query_43_cast_fp16)[name = string("op_32381_cast_fp16")]; + tensor var_32385_begin_0 = const()[name = string("op_32385_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32385_end_0 = const()[name = string("op_32385_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_32385_end_mask_0 = const()[name = string("op_32385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32385_cast_fp16 = slice_by_index(begin = var_32385_begin_0, end = var_32385_end_0, end_mask = var_32385_end_mask_0, x = query_43_cast_fp16)[name = string("op_32385_cast_fp16")]; + tensor var_32389_begin_0 = const()[name = string("op_32389_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32389_end_0 = const()[name = string("op_32389_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_32389_end_mask_0 = const()[name = string("op_32389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32389_cast_fp16 = slice_by_index(begin = var_32389_begin_0, end = var_32389_end_0, end_mask = var_32389_end_mask_0, x = query_43_cast_fp16)[name = string("op_32389_cast_fp16")]; + tensor var_32393_begin_0 = const()[name = string("op_32393_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32393_end_0 = const()[name = string("op_32393_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_32393_end_mask_0 = const()[name = string("op_32393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32393_cast_fp16 = slice_by_index(begin = var_32393_begin_0, end = var_32393_end_0, end_mask = var_32393_end_mask_0, x = query_43_cast_fp16)[name = string("op_32393_cast_fp16")]; + tensor var_32397_begin_0 = const()[name = string("op_32397_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32397_end_0 = const()[name = string("op_32397_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_32397_end_mask_0 = const()[name = string("op_32397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32397_cast_fp16 = slice_by_index(begin = var_32397_begin_0, end = var_32397_end_0, end_mask = var_32397_end_mask_0, x = query_43_cast_fp16)[name = string("op_32397_cast_fp16")]; + tensor var_32401_begin_0 = const()[name = string("op_32401_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32401_end_0 = const()[name = string("op_32401_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_32401_end_mask_0 = const()[name = string("op_32401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32401_cast_fp16 = slice_by_index(begin = var_32401_begin_0, end = var_32401_end_0, end_mask = var_32401_end_mask_0, x = query_43_cast_fp16)[name = string("op_32401_cast_fp16")]; + tensor var_32410_begin_0 = const()[name = string("op_32410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32410_end_0 = const()[name = string("op_32410_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32410_end_mask_0 = const()[name = string("op_32410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32410_cast_fp16 = slice_by_index(begin = var_32410_begin_0, end = var_32410_end_0, end_mask = var_32410_end_mask_0, x = var_32325_cast_fp16)[name = string("op_32410_cast_fp16")]; + tensor var_32417_begin_0 = const()[name = string("op_32417_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32417_end_0 = const()[name = string("op_32417_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32417_end_mask_0 = const()[name = string("op_32417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32417_cast_fp16 = slice_by_index(begin = var_32417_begin_0, end = var_32417_end_0, end_mask = var_32417_end_mask_0, x = var_32325_cast_fp16)[name = string("op_32417_cast_fp16")]; + tensor var_32424_begin_0 = const()[name = string("op_32424_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32424_end_0 = const()[name = string("op_32424_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32424_end_mask_0 = const()[name = string("op_32424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32424_cast_fp16 = slice_by_index(begin = var_32424_begin_0, end = var_32424_end_0, end_mask = var_32424_end_mask_0, x = var_32325_cast_fp16)[name = string("op_32424_cast_fp16")]; + tensor var_32431_begin_0 = const()[name = string("op_32431_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32431_end_0 = const()[name = string("op_32431_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32431_end_mask_0 = const()[name = string("op_32431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32431_cast_fp16 = slice_by_index(begin = var_32431_begin_0, end = var_32431_end_0, end_mask = var_32431_end_mask_0, x = var_32325_cast_fp16)[name = string("op_32431_cast_fp16")]; + tensor var_32438_begin_0 = const()[name = string("op_32438_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32438_end_0 = const()[name = string("op_32438_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32438_end_mask_0 = const()[name = string("op_32438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32438_cast_fp16 = slice_by_index(begin = var_32438_begin_0, end = var_32438_end_0, end_mask = var_32438_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32438_cast_fp16")]; + tensor var_32445_begin_0 = const()[name = string("op_32445_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32445_end_0 = const()[name = string("op_32445_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32445_end_mask_0 = const()[name = string("op_32445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32445_cast_fp16 = slice_by_index(begin = var_32445_begin_0, end = var_32445_end_0, end_mask = var_32445_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32445_cast_fp16")]; + tensor var_32452_begin_0 = const()[name = string("op_32452_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32452_end_0 = const()[name = string("op_32452_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32452_end_mask_0 = const()[name = string("op_32452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32452_cast_fp16 = slice_by_index(begin = var_32452_begin_0, end = var_32452_end_0, end_mask = var_32452_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32452_cast_fp16")]; + tensor var_32459_begin_0 = const()[name = string("op_32459_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32459_end_0 = const()[name = string("op_32459_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32459_end_mask_0 = const()[name = string("op_32459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32459_cast_fp16 = slice_by_index(begin = var_32459_begin_0, end = var_32459_end_0, end_mask = var_32459_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32459_cast_fp16")]; + tensor var_32466_begin_0 = const()[name = string("op_32466_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32466_end_0 = const()[name = string("op_32466_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32466_end_mask_0 = const()[name = string("op_32466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32466_cast_fp16 = slice_by_index(begin = var_32466_begin_0, end = var_32466_end_0, end_mask = var_32466_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32466_cast_fp16")]; + tensor var_32473_begin_0 = const()[name = string("op_32473_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32473_end_0 = const()[name = string("op_32473_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32473_end_mask_0 = const()[name = string("op_32473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32473_cast_fp16 = slice_by_index(begin = var_32473_begin_0, end = var_32473_end_0, end_mask = var_32473_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32473_cast_fp16")]; + tensor var_32480_begin_0 = const()[name = string("op_32480_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32480_end_0 = const()[name = string("op_32480_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32480_end_mask_0 = const()[name = string("op_32480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32480_cast_fp16 = slice_by_index(begin = var_32480_begin_0, end = var_32480_end_0, end_mask = var_32480_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32480_cast_fp16")]; + tensor var_32487_begin_0 = const()[name = string("op_32487_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32487_end_0 = const()[name = string("op_32487_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32487_end_mask_0 = const()[name = string("op_32487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32487_cast_fp16 = slice_by_index(begin = var_32487_begin_0, end = var_32487_end_0, end_mask = var_32487_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32487_cast_fp16")]; + tensor var_32494_begin_0 = const()[name = string("op_32494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32494_end_0 = const()[name = string("op_32494_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32494_end_mask_0 = const()[name = string("op_32494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32494_cast_fp16 = slice_by_index(begin = var_32494_begin_0, end = var_32494_end_0, end_mask = var_32494_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32494_cast_fp16")]; + tensor var_32501_begin_0 = const()[name = string("op_32501_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32501_end_0 = const()[name = string("op_32501_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32501_end_mask_0 = const()[name = string("op_32501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32501_cast_fp16 = slice_by_index(begin = var_32501_begin_0, end = var_32501_end_0, end_mask = var_32501_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32501_cast_fp16")]; + tensor var_32508_begin_0 = const()[name = string("op_32508_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32508_end_0 = const()[name = string("op_32508_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32508_end_mask_0 = const()[name = string("op_32508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32508_cast_fp16 = slice_by_index(begin = var_32508_begin_0, end = var_32508_end_0, end_mask = var_32508_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32508_cast_fp16")]; + tensor var_32515_begin_0 = const()[name = string("op_32515_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32515_end_0 = const()[name = string("op_32515_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32515_end_mask_0 = const()[name = string("op_32515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32515_cast_fp16 = slice_by_index(begin = var_32515_begin_0, end = var_32515_end_0, end_mask = var_32515_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32515_cast_fp16")]; + tensor var_32522_begin_0 = const()[name = string("op_32522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32522_end_0 = const()[name = string("op_32522_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32522_end_mask_0 = const()[name = string("op_32522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32522_cast_fp16 = slice_by_index(begin = var_32522_begin_0, end = var_32522_end_0, end_mask = var_32522_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32522_cast_fp16")]; + tensor var_32529_begin_0 = const()[name = string("op_32529_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32529_end_0 = const()[name = string("op_32529_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32529_end_mask_0 = const()[name = string("op_32529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32529_cast_fp16 = slice_by_index(begin = var_32529_begin_0, end = var_32529_end_0, end_mask = var_32529_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32529_cast_fp16")]; + tensor var_32536_begin_0 = const()[name = string("op_32536_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32536_end_0 = const()[name = string("op_32536_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32536_end_mask_0 = const()[name = string("op_32536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32536_cast_fp16 = slice_by_index(begin = var_32536_begin_0, end = var_32536_end_0, end_mask = var_32536_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32536_cast_fp16")]; + tensor var_32543_begin_0 = const()[name = string("op_32543_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32543_end_0 = const()[name = string("op_32543_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32543_end_mask_0 = const()[name = string("op_32543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32543_cast_fp16 = slice_by_index(begin = var_32543_begin_0, end = var_32543_end_0, end_mask = var_32543_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32543_cast_fp16")]; + tensor var_32550_begin_0 = const()[name = string("op_32550_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32550_end_0 = const()[name = string("op_32550_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32550_end_mask_0 = const()[name = string("op_32550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32550_cast_fp16 = slice_by_index(begin = var_32550_begin_0, end = var_32550_end_0, end_mask = var_32550_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32550_cast_fp16")]; + tensor var_32557_begin_0 = const()[name = string("op_32557_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32557_end_0 = const()[name = string("op_32557_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32557_end_mask_0 = const()[name = string("op_32557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32557_cast_fp16 = slice_by_index(begin = var_32557_begin_0, end = var_32557_end_0, end_mask = var_32557_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32557_cast_fp16")]; + tensor var_32564_begin_0 = const()[name = string("op_32564_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32564_end_0 = const()[name = string("op_32564_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32564_end_mask_0 = const()[name = string("op_32564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32564_cast_fp16 = slice_by_index(begin = var_32564_begin_0, end = var_32564_end_0, end_mask = var_32564_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32564_cast_fp16")]; + tensor var_32571_begin_0 = const()[name = string("op_32571_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32571_end_0 = const()[name = string("op_32571_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32571_end_mask_0 = const()[name = string("op_32571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32571_cast_fp16 = slice_by_index(begin = var_32571_begin_0, end = var_32571_end_0, end_mask = var_32571_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32571_cast_fp16")]; + tensor var_32578_begin_0 = const()[name = string("op_32578_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32578_end_0 = const()[name = string("op_32578_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32578_end_mask_0 = const()[name = string("op_32578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32578_cast_fp16 = slice_by_index(begin = var_32578_begin_0, end = var_32578_end_0, end_mask = var_32578_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32578_cast_fp16")]; + tensor var_32585_begin_0 = const()[name = string("op_32585_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32585_end_0 = const()[name = string("op_32585_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32585_end_mask_0 = const()[name = string("op_32585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32585_cast_fp16 = slice_by_index(begin = var_32585_begin_0, end = var_32585_end_0, end_mask = var_32585_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32585_cast_fp16")]; + tensor var_32592_begin_0 = const()[name = string("op_32592_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32592_end_0 = const()[name = string("op_32592_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32592_end_mask_0 = const()[name = string("op_32592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32592_cast_fp16 = slice_by_index(begin = var_32592_begin_0, end = var_32592_end_0, end_mask = var_32592_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32592_cast_fp16")]; + tensor var_32599_begin_0 = const()[name = string("op_32599_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32599_end_0 = const()[name = string("op_32599_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32599_end_mask_0 = const()[name = string("op_32599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32599_cast_fp16 = slice_by_index(begin = var_32599_begin_0, end = var_32599_end_0, end_mask = var_32599_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32599_cast_fp16")]; + tensor var_32606_begin_0 = const()[name = string("op_32606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32606_end_0 = const()[name = string("op_32606_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32606_end_mask_0 = const()[name = string("op_32606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32606_cast_fp16 = slice_by_index(begin = var_32606_begin_0, end = var_32606_end_0, end_mask = var_32606_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32606_cast_fp16")]; + tensor var_32613_begin_0 = const()[name = string("op_32613_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32613_end_0 = const()[name = string("op_32613_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32613_end_mask_0 = const()[name = string("op_32613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32613_cast_fp16 = slice_by_index(begin = var_32613_begin_0, end = var_32613_end_0, end_mask = var_32613_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32613_cast_fp16")]; + tensor var_32620_begin_0 = const()[name = string("op_32620_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32620_end_0 = const()[name = string("op_32620_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32620_end_mask_0 = const()[name = string("op_32620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32620_cast_fp16 = slice_by_index(begin = var_32620_begin_0, end = var_32620_end_0, end_mask = var_32620_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32620_cast_fp16")]; + tensor var_32627_begin_0 = const()[name = string("op_32627_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32627_end_0 = const()[name = string("op_32627_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32627_end_mask_0 = const()[name = string("op_32627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32627_cast_fp16 = slice_by_index(begin = var_32627_begin_0, end = var_32627_end_0, end_mask = var_32627_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32627_cast_fp16")]; + tensor var_32634_begin_0 = const()[name = string("op_32634_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32634_end_0 = const()[name = string("op_32634_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32634_end_mask_0 = const()[name = string("op_32634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32634_cast_fp16 = slice_by_index(begin = var_32634_begin_0, end = var_32634_end_0, end_mask = var_32634_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32634_cast_fp16")]; + tensor var_32641_begin_0 = const()[name = string("op_32641_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32641_end_0 = const()[name = string("op_32641_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32641_end_mask_0 = const()[name = string("op_32641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32641_cast_fp16 = slice_by_index(begin = var_32641_begin_0, end = var_32641_end_0, end_mask = var_32641_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32641_cast_fp16")]; + tensor var_32648_begin_0 = const()[name = string("op_32648_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32648_end_0 = const()[name = string("op_32648_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32648_end_mask_0 = const()[name = string("op_32648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32648_cast_fp16 = slice_by_index(begin = var_32648_begin_0, end = var_32648_end_0, end_mask = var_32648_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32648_cast_fp16")]; + tensor var_32655_begin_0 = const()[name = string("op_32655_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32655_end_0 = const()[name = string("op_32655_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32655_end_mask_0 = const()[name = string("op_32655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32655_cast_fp16 = slice_by_index(begin = var_32655_begin_0, end = var_32655_end_0, end_mask = var_32655_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32655_cast_fp16")]; + tensor var_32662_begin_0 = const()[name = string("op_32662_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32662_end_0 = const()[name = string("op_32662_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32662_end_mask_0 = const()[name = string("op_32662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32662_cast_fp16 = slice_by_index(begin = var_32662_begin_0, end = var_32662_end_0, end_mask = var_32662_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32662_cast_fp16")]; + tensor var_32669_begin_0 = const()[name = string("op_32669_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32669_end_0 = const()[name = string("op_32669_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32669_end_mask_0 = const()[name = string("op_32669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32669_cast_fp16 = slice_by_index(begin = var_32669_begin_0, end = var_32669_end_0, end_mask = var_32669_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32669_cast_fp16")]; + tensor var_32676_begin_0 = const()[name = string("op_32676_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32676_end_0 = const()[name = string("op_32676_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32676_end_mask_0 = const()[name = string("op_32676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32676_cast_fp16 = slice_by_index(begin = var_32676_begin_0, end = var_32676_end_0, end_mask = var_32676_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32676_cast_fp16")]; + tensor var_32683_begin_0 = const()[name = string("op_32683_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32683_end_0 = const()[name = string("op_32683_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32683_end_mask_0 = const()[name = string("op_32683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32683_cast_fp16 = slice_by_index(begin = var_32683_begin_0, end = var_32683_end_0, end_mask = var_32683_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32683_cast_fp16")]; + tensor var_32690_begin_0 = const()[name = string("op_32690_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32690_end_0 = const()[name = string("op_32690_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32690_end_mask_0 = const()[name = string("op_32690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32690_cast_fp16 = slice_by_index(begin = var_32690_begin_0, end = var_32690_end_0, end_mask = var_32690_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32690_cast_fp16")]; + tensor var_32697_begin_0 = const()[name = string("op_32697_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32697_end_0 = const()[name = string("op_32697_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32697_end_mask_0 = const()[name = string("op_32697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32697_cast_fp16 = slice_by_index(begin = var_32697_begin_0, end = var_32697_end_0, end_mask = var_32697_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32697_cast_fp16")]; + tensor var_32704_begin_0 = const()[name = string("op_32704_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32704_end_0 = const()[name = string("op_32704_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32704_end_mask_0 = const()[name = string("op_32704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32704_cast_fp16 = slice_by_index(begin = var_32704_begin_0, end = var_32704_end_0, end_mask = var_32704_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32704_cast_fp16")]; + tensor var_32711_begin_0 = const()[name = string("op_32711_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32711_end_0 = const()[name = string("op_32711_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32711_end_mask_0 = const()[name = string("op_32711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32711_cast_fp16 = slice_by_index(begin = var_32711_begin_0, end = var_32711_end_0, end_mask = var_32711_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32711_cast_fp16")]; + tensor var_32718_begin_0 = const()[name = string("op_32718_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32718_end_0 = const()[name = string("op_32718_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32718_end_mask_0 = const()[name = string("op_32718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32718_cast_fp16 = slice_by_index(begin = var_32718_begin_0, end = var_32718_end_0, end_mask = var_32718_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32718_cast_fp16")]; + tensor var_32725_begin_0 = const()[name = string("op_32725_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32725_end_0 = const()[name = string("op_32725_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32725_end_mask_0 = const()[name = string("op_32725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32725_cast_fp16 = slice_by_index(begin = var_32725_begin_0, end = var_32725_end_0, end_mask = var_32725_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32725_cast_fp16")]; + tensor var_32732_begin_0 = const()[name = string("op_32732_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32732_end_0 = const()[name = string("op_32732_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32732_end_mask_0 = const()[name = string("op_32732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32732_cast_fp16 = slice_by_index(begin = var_32732_begin_0, end = var_32732_end_0, end_mask = var_32732_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32732_cast_fp16")]; + tensor var_32739_begin_0 = const()[name = string("op_32739_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32739_end_0 = const()[name = string("op_32739_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32739_end_mask_0 = const()[name = string("op_32739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32739_cast_fp16 = slice_by_index(begin = var_32739_begin_0, end = var_32739_end_0, end_mask = var_32739_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32739_cast_fp16")]; + tensor var_32746_begin_0 = const()[name = string("op_32746_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32746_end_0 = const()[name = string("op_32746_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32746_end_mask_0 = const()[name = string("op_32746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32746_cast_fp16 = slice_by_index(begin = var_32746_begin_0, end = var_32746_end_0, end_mask = var_32746_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32746_cast_fp16")]; + tensor var_32753_begin_0 = const()[name = string("op_32753_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32753_end_0 = const()[name = string("op_32753_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32753_end_mask_0 = const()[name = string("op_32753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32753_cast_fp16 = slice_by_index(begin = var_32753_begin_0, end = var_32753_end_0, end_mask = var_32753_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32753_cast_fp16")]; + tensor var_32760_begin_0 = const()[name = string("op_32760_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32760_end_0 = const()[name = string("op_32760_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32760_end_mask_0 = const()[name = string("op_32760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32760_cast_fp16 = slice_by_index(begin = var_32760_begin_0, end = var_32760_end_0, end_mask = var_32760_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32760_cast_fp16")]; + tensor var_32767_begin_0 = const()[name = string("op_32767_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32767_end_0 = const()[name = string("op_32767_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32767_end_mask_0 = const()[name = string("op_32767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32767_cast_fp16 = slice_by_index(begin = var_32767_begin_0, end = var_32767_end_0, end_mask = var_32767_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32767_cast_fp16")]; + tensor var_32774_begin_0 = const()[name = string("op_32774_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32774_end_0 = const()[name = string("op_32774_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32774_end_mask_0 = const()[name = string("op_32774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32774_cast_fp16 = slice_by_index(begin = var_32774_begin_0, end = var_32774_end_0, end_mask = var_32774_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32774_cast_fp16")]; + tensor var_32781_begin_0 = const()[name = string("op_32781_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32781_end_0 = const()[name = string("op_32781_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32781_end_mask_0 = const()[name = string("op_32781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32781_cast_fp16 = slice_by_index(begin = var_32781_begin_0, end = var_32781_end_0, end_mask = var_32781_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32781_cast_fp16")]; + tensor var_32788_begin_0 = const()[name = string("op_32788_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32788_end_0 = const()[name = string("op_32788_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32788_end_mask_0 = const()[name = string("op_32788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32788_cast_fp16 = slice_by_index(begin = var_32788_begin_0, end = var_32788_end_0, end_mask = var_32788_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32788_cast_fp16")]; + tensor var_32795_begin_0 = const()[name = string("op_32795_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32795_end_0 = const()[name = string("op_32795_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32795_end_mask_0 = const()[name = string("op_32795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32795_cast_fp16 = slice_by_index(begin = var_32795_begin_0, end = var_32795_end_0, end_mask = var_32795_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32795_cast_fp16")]; + tensor var_32802_begin_0 = const()[name = string("op_32802_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32802_end_0 = const()[name = string("op_32802_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32802_end_mask_0 = const()[name = string("op_32802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32802_cast_fp16 = slice_by_index(begin = var_32802_begin_0, end = var_32802_end_0, end_mask = var_32802_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32802_cast_fp16")]; + tensor var_32809_begin_0 = const()[name = string("op_32809_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32809_end_0 = const()[name = string("op_32809_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32809_end_mask_0 = const()[name = string("op_32809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32809_cast_fp16 = slice_by_index(begin = var_32809_begin_0, end = var_32809_end_0, end_mask = var_32809_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32809_cast_fp16")]; + tensor var_32816_begin_0 = const()[name = string("op_32816_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32816_end_0 = const()[name = string("op_32816_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32816_end_mask_0 = const()[name = string("op_32816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32816_cast_fp16 = slice_by_index(begin = var_32816_begin_0, end = var_32816_end_0, end_mask = var_32816_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32816_cast_fp16")]; + tensor var_32823_begin_0 = const()[name = string("op_32823_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32823_end_0 = const()[name = string("op_32823_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32823_end_mask_0 = const()[name = string("op_32823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32823_cast_fp16 = slice_by_index(begin = var_32823_begin_0, end = var_32823_end_0, end_mask = var_32823_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32823_cast_fp16")]; + tensor var_32830_begin_0 = const()[name = string("op_32830_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32830_end_0 = const()[name = string("op_32830_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32830_end_mask_0 = const()[name = string("op_32830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32830_cast_fp16 = slice_by_index(begin = var_32830_begin_0, end = var_32830_end_0, end_mask = var_32830_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32830_cast_fp16")]; + tensor var_32837_begin_0 = const()[name = string("op_32837_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32837_end_0 = const()[name = string("op_32837_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32837_end_mask_0 = const()[name = string("op_32837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32837_cast_fp16 = slice_by_index(begin = var_32837_begin_0, end = var_32837_end_0, end_mask = var_32837_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32837_cast_fp16")]; + tensor var_32844_begin_0 = const()[name = string("op_32844_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32844_end_0 = const()[name = string("op_32844_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32844_end_mask_0 = const()[name = string("op_32844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32844_cast_fp16 = slice_by_index(begin = var_32844_begin_0, end = var_32844_end_0, end_mask = var_32844_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32844_cast_fp16")]; + tensor var_32851_begin_0 = const()[name = string("op_32851_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32851_end_0 = const()[name = string("op_32851_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32851_end_mask_0 = const()[name = string("op_32851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32851_cast_fp16 = slice_by_index(begin = var_32851_begin_0, end = var_32851_end_0, end_mask = var_32851_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32851_cast_fp16")]; + tensor var_32858_begin_0 = const()[name = string("op_32858_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32858_end_0 = const()[name = string("op_32858_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32858_end_mask_0 = const()[name = string("op_32858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32858_cast_fp16 = slice_by_index(begin = var_32858_begin_0, end = var_32858_end_0, end_mask = var_32858_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32858_cast_fp16")]; + tensor var_32865_begin_0 = const()[name = string("op_32865_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32865_end_0 = const()[name = string("op_32865_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32865_end_mask_0 = const()[name = string("op_32865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32865_cast_fp16 = slice_by_index(begin = var_32865_begin_0, end = var_32865_end_0, end_mask = var_32865_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32865_cast_fp16")]; + tensor var_32872_begin_0 = const()[name = string("op_32872_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32872_end_0 = const()[name = string("op_32872_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32872_end_mask_0 = const()[name = string("op_32872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32872_cast_fp16 = slice_by_index(begin = var_32872_begin_0, end = var_32872_end_0, end_mask = var_32872_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32872_cast_fp16")]; + tensor var_32879_begin_0 = const()[name = string("op_32879_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32879_end_0 = const()[name = string("op_32879_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32879_end_mask_0 = const()[name = string("op_32879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32879_cast_fp16 = slice_by_index(begin = var_32879_begin_0, end = var_32879_end_0, end_mask = var_32879_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32879_cast_fp16")]; + tensor var_32886_begin_0 = const()[name = string("op_32886_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32886_end_0 = const()[name = string("op_32886_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32886_end_mask_0 = const()[name = string("op_32886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32886_cast_fp16 = slice_by_index(begin = var_32886_begin_0, end = var_32886_end_0, end_mask = var_32886_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32886_cast_fp16")]; + tensor var_32893_begin_0 = const()[name = string("op_32893_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32893_end_0 = const()[name = string("op_32893_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32893_end_mask_0 = const()[name = string("op_32893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32893_cast_fp16 = slice_by_index(begin = var_32893_begin_0, end = var_32893_end_0, end_mask = var_32893_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32893_cast_fp16")]; + tensor var_32900_begin_0 = const()[name = string("op_32900_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32900_end_0 = const()[name = string("op_32900_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32900_end_mask_0 = const()[name = string("op_32900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32900_cast_fp16 = slice_by_index(begin = var_32900_begin_0, end = var_32900_end_0, end_mask = var_32900_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32900_cast_fp16")]; + tensor var_32907_begin_0 = const()[name = string("op_32907_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32907_end_0 = const()[name = string("op_32907_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32907_end_mask_0 = const()[name = string("op_32907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32907_cast_fp16 = slice_by_index(begin = var_32907_begin_0, end = var_32907_end_0, end_mask = var_32907_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32907_cast_fp16")]; + tensor var_32914_begin_0 = const()[name = string("op_32914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32914_end_0 = const()[name = string("op_32914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32914_end_mask_0 = const()[name = string("op_32914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32914_cast_fp16 = slice_by_index(begin = var_32914_begin_0, end = var_32914_end_0, end_mask = var_32914_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32914_cast_fp16")]; + tensor var_32921_begin_0 = const()[name = string("op_32921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32921_end_0 = const()[name = string("op_32921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32921_end_mask_0 = const()[name = string("op_32921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32921_cast_fp16 = slice_by_index(begin = var_32921_begin_0, end = var_32921_end_0, end_mask = var_32921_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32921_cast_fp16")]; + tensor var_32928_begin_0 = const()[name = string("op_32928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32928_end_0 = const()[name = string("op_32928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32928_end_mask_0 = const()[name = string("op_32928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32928_cast_fp16 = slice_by_index(begin = var_32928_begin_0, end = var_32928_end_0, end_mask = var_32928_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32928_cast_fp16")]; + tensor var_32935_begin_0 = const()[name = string("op_32935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32935_end_0 = const()[name = string("op_32935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32935_end_mask_0 = const()[name = string("op_32935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32935_cast_fp16 = slice_by_index(begin = var_32935_begin_0, end = var_32935_end_0, end_mask = var_32935_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32935_cast_fp16")]; + tensor var_32942_begin_0 = const()[name = string("op_32942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32942_end_0 = const()[name = string("op_32942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32942_end_mask_0 = const()[name = string("op_32942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32942_cast_fp16 = slice_by_index(begin = var_32942_begin_0, end = var_32942_end_0, end_mask = var_32942_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32942_cast_fp16")]; + tensor var_32949_begin_0 = const()[name = string("op_32949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32949_end_0 = const()[name = string("op_32949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32949_end_mask_0 = const()[name = string("op_32949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32949_cast_fp16 = slice_by_index(begin = var_32949_begin_0, end = var_32949_end_0, end_mask = var_32949_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32949_cast_fp16")]; + tensor var_32956_begin_0 = const()[name = string("op_32956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32956_end_0 = const()[name = string("op_32956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32956_end_mask_0 = const()[name = string("op_32956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32956_cast_fp16 = slice_by_index(begin = var_32956_begin_0, end = var_32956_end_0, end_mask = var_32956_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32956_cast_fp16")]; + tensor var_32963_begin_0 = const()[name = string("op_32963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32963_end_0 = const()[name = string("op_32963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32963_end_mask_0 = const()[name = string("op_32963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32963_cast_fp16 = slice_by_index(begin = var_32963_begin_0, end = var_32963_end_0, end_mask = var_32963_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32963_cast_fp16")]; + tensor k_43_perm_0 = const()[name = string("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_32968_begin_0 = const()[name = string("op_32968_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32968_end_0 = const()[name = string("op_32968_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_32968_end_mask_0 = const()[name = string("op_32968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_43_cast_fp16 = transpose(perm = k_43_perm_0, x = key_43_cast_fp16)[name = string("transpose_10")]; + tensor var_32968_cast_fp16 = slice_by_index(begin = var_32968_begin_0, end = var_32968_end_0, end_mask = var_32968_end_mask_0, x = k_43_cast_fp16)[name = string("op_32968_cast_fp16")]; + tensor var_32972_begin_0 = const()[name = string("op_32972_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_32972_end_0 = const()[name = string("op_32972_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_32972_end_mask_0 = const()[name = string("op_32972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32972_cast_fp16 = slice_by_index(begin = var_32972_begin_0, end = var_32972_end_0, end_mask = var_32972_end_mask_0, x = k_43_cast_fp16)[name = string("op_32972_cast_fp16")]; + tensor var_32976_begin_0 = const()[name = string("op_32976_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_32976_end_0 = const()[name = string("op_32976_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_32976_end_mask_0 = const()[name = string("op_32976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32976_cast_fp16 = slice_by_index(begin = var_32976_begin_0, end = var_32976_end_0, end_mask = var_32976_end_mask_0, x = k_43_cast_fp16)[name = string("op_32976_cast_fp16")]; + tensor var_32980_begin_0 = const()[name = string("op_32980_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_32980_end_0 = const()[name = string("op_32980_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_32980_end_mask_0 = const()[name = string("op_32980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32980_cast_fp16 = slice_by_index(begin = var_32980_begin_0, end = var_32980_end_0, end_mask = var_32980_end_mask_0, x = k_43_cast_fp16)[name = string("op_32980_cast_fp16")]; + tensor var_32984_begin_0 = const()[name = string("op_32984_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_32984_end_0 = const()[name = string("op_32984_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_32984_end_mask_0 = const()[name = string("op_32984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32984_cast_fp16 = slice_by_index(begin = var_32984_begin_0, end = var_32984_end_0, end_mask = var_32984_end_mask_0, x = k_43_cast_fp16)[name = string("op_32984_cast_fp16")]; + tensor var_32988_begin_0 = const()[name = string("op_32988_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_32988_end_0 = const()[name = string("op_32988_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_32988_end_mask_0 = const()[name = string("op_32988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32988_cast_fp16 = slice_by_index(begin = var_32988_begin_0, end = var_32988_end_0, end_mask = var_32988_end_mask_0, x = k_43_cast_fp16)[name = string("op_32988_cast_fp16")]; + tensor var_32992_begin_0 = const()[name = string("op_32992_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_32992_end_0 = const()[name = string("op_32992_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_32992_end_mask_0 = const()[name = string("op_32992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32992_cast_fp16 = slice_by_index(begin = var_32992_begin_0, end = var_32992_end_0, end_mask = var_32992_end_mask_0, x = k_43_cast_fp16)[name = string("op_32992_cast_fp16")]; + tensor var_32996_begin_0 = const()[name = string("op_32996_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_32996_end_0 = const()[name = string("op_32996_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_32996_end_mask_0 = const()[name = string("op_32996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32996_cast_fp16 = slice_by_index(begin = var_32996_begin_0, end = var_32996_end_0, end_mask = var_32996_end_mask_0, x = k_43_cast_fp16)[name = string("op_32996_cast_fp16")]; + tensor var_33000_begin_0 = const()[name = string("op_33000_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_33000_end_0 = const()[name = string("op_33000_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_33000_end_mask_0 = const()[name = string("op_33000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33000_cast_fp16 = slice_by_index(begin = var_33000_begin_0, end = var_33000_end_0, end_mask = var_33000_end_mask_0, x = k_43_cast_fp16)[name = string("op_33000_cast_fp16")]; + tensor var_33004_begin_0 = const()[name = string("op_33004_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_33004_end_0 = const()[name = string("op_33004_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_33004_end_mask_0 = const()[name = string("op_33004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33004_cast_fp16 = slice_by_index(begin = var_33004_begin_0, end = var_33004_end_0, end_mask = var_33004_end_mask_0, x = k_43_cast_fp16)[name = string("op_33004_cast_fp16")]; + tensor var_33008_begin_0 = const()[name = string("op_33008_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_33008_end_0 = const()[name = string("op_33008_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_33008_end_mask_0 = const()[name = string("op_33008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33008_cast_fp16 = slice_by_index(begin = var_33008_begin_0, end = var_33008_end_0, end_mask = var_33008_end_mask_0, x = k_43_cast_fp16)[name = string("op_33008_cast_fp16")]; + tensor var_33012_begin_0 = const()[name = string("op_33012_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_33012_end_0 = const()[name = string("op_33012_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_33012_end_mask_0 = const()[name = string("op_33012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33012_cast_fp16 = slice_by_index(begin = var_33012_begin_0, end = var_33012_end_0, end_mask = var_33012_end_mask_0, x = k_43_cast_fp16)[name = string("op_33012_cast_fp16")]; + tensor var_33016_begin_0 = const()[name = string("op_33016_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_33016_end_0 = const()[name = string("op_33016_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_33016_end_mask_0 = const()[name = string("op_33016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33016_cast_fp16 = slice_by_index(begin = var_33016_begin_0, end = var_33016_end_0, end_mask = var_33016_end_mask_0, x = k_43_cast_fp16)[name = string("op_33016_cast_fp16")]; + tensor var_33020_begin_0 = const()[name = string("op_33020_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_33020_end_0 = const()[name = string("op_33020_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_33020_end_mask_0 = const()[name = string("op_33020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33020_cast_fp16 = slice_by_index(begin = var_33020_begin_0, end = var_33020_end_0, end_mask = var_33020_end_mask_0, x = k_43_cast_fp16)[name = string("op_33020_cast_fp16")]; + tensor var_33024_begin_0 = const()[name = string("op_33024_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_33024_end_0 = const()[name = string("op_33024_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_33024_end_mask_0 = const()[name = string("op_33024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33024_cast_fp16 = slice_by_index(begin = var_33024_begin_0, end = var_33024_end_0, end_mask = var_33024_end_mask_0, x = k_43_cast_fp16)[name = string("op_33024_cast_fp16")]; + tensor var_33028_begin_0 = const()[name = string("op_33028_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_33028_end_0 = const()[name = string("op_33028_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_33028_end_mask_0 = const()[name = string("op_33028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33028_cast_fp16 = slice_by_index(begin = var_33028_begin_0, end = var_33028_end_0, end_mask = var_33028_end_mask_0, x = k_43_cast_fp16)[name = string("op_33028_cast_fp16")]; + tensor var_33032_begin_0 = const()[name = string("op_33032_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_33032_end_0 = const()[name = string("op_33032_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_33032_end_mask_0 = const()[name = string("op_33032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33032_cast_fp16 = slice_by_index(begin = var_33032_begin_0, end = var_33032_end_0, end_mask = var_33032_end_mask_0, x = k_43_cast_fp16)[name = string("op_33032_cast_fp16")]; + tensor var_33036_begin_0 = const()[name = string("op_33036_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_33036_end_0 = const()[name = string("op_33036_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_33036_end_mask_0 = const()[name = string("op_33036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33036_cast_fp16 = slice_by_index(begin = var_33036_begin_0, end = var_33036_end_0, end_mask = var_33036_end_mask_0, x = k_43_cast_fp16)[name = string("op_33036_cast_fp16")]; + tensor var_33040_begin_0 = const()[name = string("op_33040_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_33040_end_0 = const()[name = string("op_33040_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_33040_end_mask_0 = const()[name = string("op_33040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33040_cast_fp16 = slice_by_index(begin = var_33040_begin_0, end = var_33040_end_0, end_mask = var_33040_end_mask_0, x = k_43_cast_fp16)[name = string("op_33040_cast_fp16")]; + tensor var_33044_begin_0 = const()[name = string("op_33044_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_33044_end_0 = const()[name = string("op_33044_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_33044_end_mask_0 = const()[name = string("op_33044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33044_cast_fp16 = slice_by_index(begin = var_33044_begin_0, end = var_33044_end_0, end_mask = var_33044_end_mask_0, x = k_43_cast_fp16)[name = string("op_33044_cast_fp16")]; + tensor var_33046_begin_0 = const()[name = string("op_33046_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33046_end_0 = const()[name = string("op_33046_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33046_end_mask_0 = const()[name = string("op_33046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33046_cast_fp16 = slice_by_index(begin = var_33046_begin_0, end = var_33046_end_0, end_mask = var_33046_end_mask_0, x = value_43_cast_fp16)[name = string("op_33046_cast_fp16")]; + tensor var_33050_begin_0 = const()[name = string("op_33050_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33050_end_0 = const()[name = string("op_33050_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_33050_end_mask_0 = const()[name = string("op_33050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33050_cast_fp16 = slice_by_index(begin = var_33050_begin_0, end = var_33050_end_0, end_mask = var_33050_end_mask_0, x = value_43_cast_fp16)[name = string("op_33050_cast_fp16")]; + tensor var_33054_begin_0 = const()[name = string("op_33054_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33054_end_0 = const()[name = string("op_33054_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_33054_end_mask_0 = const()[name = string("op_33054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33054_cast_fp16 = slice_by_index(begin = var_33054_begin_0, end = var_33054_end_0, end_mask = var_33054_end_mask_0, x = value_43_cast_fp16)[name = string("op_33054_cast_fp16")]; + tensor var_33058_begin_0 = const()[name = string("op_33058_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33058_end_0 = const()[name = string("op_33058_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_33058_end_mask_0 = const()[name = string("op_33058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33058_cast_fp16 = slice_by_index(begin = var_33058_begin_0, end = var_33058_end_0, end_mask = var_33058_end_mask_0, x = value_43_cast_fp16)[name = string("op_33058_cast_fp16")]; + tensor var_33062_begin_0 = const()[name = string("op_33062_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33062_end_0 = const()[name = string("op_33062_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_33062_end_mask_0 = const()[name = string("op_33062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33062_cast_fp16 = slice_by_index(begin = var_33062_begin_0, end = var_33062_end_0, end_mask = var_33062_end_mask_0, x = value_43_cast_fp16)[name = string("op_33062_cast_fp16")]; + tensor var_33066_begin_0 = const()[name = string("op_33066_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33066_end_0 = const()[name = string("op_33066_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_33066_end_mask_0 = const()[name = string("op_33066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33066_cast_fp16 = slice_by_index(begin = var_33066_begin_0, end = var_33066_end_0, end_mask = var_33066_end_mask_0, x = value_43_cast_fp16)[name = string("op_33066_cast_fp16")]; + tensor var_33070_begin_0 = const()[name = string("op_33070_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33070_end_0 = const()[name = string("op_33070_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_33070_end_mask_0 = const()[name = string("op_33070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33070_cast_fp16 = slice_by_index(begin = var_33070_begin_0, end = var_33070_end_0, end_mask = var_33070_end_mask_0, x = value_43_cast_fp16)[name = string("op_33070_cast_fp16")]; + tensor var_33074_begin_0 = const()[name = string("op_33074_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33074_end_0 = const()[name = string("op_33074_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_33074_end_mask_0 = const()[name = string("op_33074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33074_cast_fp16 = slice_by_index(begin = var_33074_begin_0, end = var_33074_end_0, end_mask = var_33074_end_mask_0, x = value_43_cast_fp16)[name = string("op_33074_cast_fp16")]; + tensor var_33078_begin_0 = const()[name = string("op_33078_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33078_end_0 = const()[name = string("op_33078_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_33078_end_mask_0 = const()[name = string("op_33078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33078_cast_fp16 = slice_by_index(begin = var_33078_begin_0, end = var_33078_end_0, end_mask = var_33078_end_mask_0, x = value_43_cast_fp16)[name = string("op_33078_cast_fp16")]; + tensor var_33082_begin_0 = const()[name = string("op_33082_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33082_end_0 = const()[name = string("op_33082_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_33082_end_mask_0 = const()[name = string("op_33082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33082_cast_fp16 = slice_by_index(begin = var_33082_begin_0, end = var_33082_end_0, end_mask = var_33082_end_mask_0, x = value_43_cast_fp16)[name = string("op_33082_cast_fp16")]; + tensor var_33086_begin_0 = const()[name = string("op_33086_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33086_end_0 = const()[name = string("op_33086_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_33086_end_mask_0 = const()[name = string("op_33086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33086_cast_fp16 = slice_by_index(begin = var_33086_begin_0, end = var_33086_end_0, end_mask = var_33086_end_mask_0, x = value_43_cast_fp16)[name = string("op_33086_cast_fp16")]; + tensor var_33090_begin_0 = const()[name = string("op_33090_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33090_end_0 = const()[name = string("op_33090_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_33090_end_mask_0 = const()[name = string("op_33090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33090_cast_fp16 = slice_by_index(begin = var_33090_begin_0, end = var_33090_end_0, end_mask = var_33090_end_mask_0, x = value_43_cast_fp16)[name = string("op_33090_cast_fp16")]; + tensor var_33094_begin_0 = const()[name = string("op_33094_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33094_end_0 = const()[name = string("op_33094_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_33094_end_mask_0 = const()[name = string("op_33094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33094_cast_fp16 = slice_by_index(begin = var_33094_begin_0, end = var_33094_end_0, end_mask = var_33094_end_mask_0, x = value_43_cast_fp16)[name = string("op_33094_cast_fp16")]; + tensor var_33098_begin_0 = const()[name = string("op_33098_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33098_end_0 = const()[name = string("op_33098_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_33098_end_mask_0 = const()[name = string("op_33098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33098_cast_fp16 = slice_by_index(begin = var_33098_begin_0, end = var_33098_end_0, end_mask = var_33098_end_mask_0, x = value_43_cast_fp16)[name = string("op_33098_cast_fp16")]; + tensor var_33102_begin_0 = const()[name = string("op_33102_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33102_end_0 = const()[name = string("op_33102_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_33102_end_mask_0 = const()[name = string("op_33102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33102_cast_fp16 = slice_by_index(begin = var_33102_begin_0, end = var_33102_end_0, end_mask = var_33102_end_mask_0, x = value_43_cast_fp16)[name = string("op_33102_cast_fp16")]; + tensor var_33106_begin_0 = const()[name = string("op_33106_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33106_end_0 = const()[name = string("op_33106_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_33106_end_mask_0 = const()[name = string("op_33106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33106_cast_fp16 = slice_by_index(begin = var_33106_begin_0, end = var_33106_end_0, end_mask = var_33106_end_mask_0, x = value_43_cast_fp16)[name = string("op_33106_cast_fp16")]; + tensor var_33110_begin_0 = const()[name = string("op_33110_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33110_end_0 = const()[name = string("op_33110_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_33110_end_mask_0 = const()[name = string("op_33110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33110_cast_fp16 = slice_by_index(begin = var_33110_begin_0, end = var_33110_end_0, end_mask = var_33110_end_mask_0, x = value_43_cast_fp16)[name = string("op_33110_cast_fp16")]; + tensor var_33114_begin_0 = const()[name = string("op_33114_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33114_end_0 = const()[name = string("op_33114_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_33114_end_mask_0 = const()[name = string("op_33114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33114_cast_fp16 = slice_by_index(begin = var_33114_begin_0, end = var_33114_end_0, end_mask = var_33114_end_mask_0, x = value_43_cast_fp16)[name = string("op_33114_cast_fp16")]; + tensor var_33118_begin_0 = const()[name = string("op_33118_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33118_end_0 = const()[name = string("op_33118_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_33118_end_mask_0 = const()[name = string("op_33118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33118_cast_fp16 = slice_by_index(begin = var_33118_begin_0, end = var_33118_end_0, end_mask = var_33118_end_mask_0, x = value_43_cast_fp16)[name = string("op_33118_cast_fp16")]; + tensor var_33122_begin_0 = const()[name = string("op_33122_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33122_end_0 = const()[name = string("op_33122_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_33122_end_mask_0 = const()[name = string("op_33122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33122_cast_fp16 = slice_by_index(begin = var_33122_begin_0, end = var_33122_end_0, end_mask = var_33122_end_mask_0, x = value_43_cast_fp16)[name = string("op_33122_cast_fp16")]; + string _SplitHeadsQ__mh_w_3361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3361_equation_0, values = (var_32968_cast_fp16, var_32410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3361_cast_fp16")]; + string _SplitHeadsQ__mh_w_3363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3363_equation_0, values = (var_32968_cast_fp16, var_32417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3363_cast_fp16")]; + string _SplitHeadsQ__mh_w_3365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3365_equation_0, values = (var_32968_cast_fp16, var_32424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3365_cast_fp16")]; + string _SplitHeadsQ__mh_w_3367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3367_equation_0, values = (var_32968_cast_fp16, var_32431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3367_cast_fp16")]; + string _SplitHeadsQ__mh_w_3369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3369_equation_0, values = (var_32972_cast_fp16, var_32438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3369_cast_fp16")]; + string _SplitHeadsQ__mh_w_3371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3371_equation_0, values = (var_32972_cast_fp16, var_32445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3371_cast_fp16")]; + string _SplitHeadsQ__mh_w_3373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3373_equation_0, values = (var_32972_cast_fp16, var_32452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3373_cast_fp16")]; + string _SplitHeadsQ__mh_w_3375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3375_equation_0, values = (var_32972_cast_fp16, var_32459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3375_cast_fp16")]; + string _SplitHeadsQ__mh_w_3377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3377_equation_0, values = (var_32976_cast_fp16, var_32466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3377_cast_fp16")]; + string _SplitHeadsQ__mh_w_3379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3379_equation_0, values = (var_32976_cast_fp16, var_32473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3379_cast_fp16")]; + string _SplitHeadsQ__mh_w_3381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3381_equation_0, values = (var_32976_cast_fp16, var_32480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3381_cast_fp16")]; + string _SplitHeadsQ__mh_w_3383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3383_equation_0, values = (var_32976_cast_fp16, var_32487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3383_cast_fp16")]; + string _SplitHeadsQ__mh_w_3385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3385_equation_0, values = (var_32980_cast_fp16, var_32494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3385_cast_fp16")]; + string _SplitHeadsQ__mh_w_3387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3387_equation_0, values = (var_32980_cast_fp16, var_32501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3387_cast_fp16")]; + string _SplitHeadsQ__mh_w_3389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3389_equation_0, values = (var_32980_cast_fp16, var_32508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3389_cast_fp16")]; + string _SplitHeadsQ__mh_w_3391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3391_equation_0, values = (var_32980_cast_fp16, var_32515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3391_cast_fp16")]; + string _SplitHeadsQ__mh_w_3393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3393_equation_0, values = (var_32984_cast_fp16, var_32522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3393_cast_fp16")]; + string _SplitHeadsQ__mh_w_3395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3395_equation_0, values = (var_32984_cast_fp16, var_32529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3395_cast_fp16")]; + string _SplitHeadsQ__mh_w_3397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3397_equation_0, values = (var_32984_cast_fp16, var_32536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3397_cast_fp16")]; + string _SplitHeadsQ__mh_w_3399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3399_equation_0, values = (var_32984_cast_fp16, var_32543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3399_cast_fp16")]; + string _SplitHeadsQ__mh_w_3401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3401_equation_0, values = (var_32988_cast_fp16, var_32550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3401_cast_fp16")]; + string _SplitHeadsQ__mh_w_3403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3403_equation_0, values = (var_32988_cast_fp16, var_32557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3403_cast_fp16")]; + string _SplitHeadsQ__mh_w_3405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3405_equation_0, values = (var_32988_cast_fp16, var_32564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3405_cast_fp16")]; + string _SplitHeadsQ__mh_w_3407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3407_equation_0, values = (var_32988_cast_fp16, var_32571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3407_cast_fp16")]; + string _SplitHeadsQ__mh_w_3409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3409_equation_0, values = (var_32992_cast_fp16, var_32578_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3409_cast_fp16")]; + string _SplitHeadsQ__mh_w_3411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3411_equation_0, values = (var_32992_cast_fp16, var_32585_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3411_cast_fp16")]; + string _SplitHeadsQ__mh_w_3413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3413_equation_0, values = (var_32992_cast_fp16, var_32592_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3413_cast_fp16")]; + string _SplitHeadsQ__mh_w_3415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3415_equation_0, values = (var_32992_cast_fp16, var_32599_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3415_cast_fp16")]; + string _SplitHeadsQ__mh_w_3417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3417_equation_0, values = (var_32996_cast_fp16, var_32606_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3417_cast_fp16")]; + string _SplitHeadsQ__mh_w_3419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3419_equation_0, values = (var_32996_cast_fp16, var_32613_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3419_cast_fp16")]; + string _SplitHeadsQ__mh_w_3421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3421_equation_0, values = (var_32996_cast_fp16, var_32620_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3421_cast_fp16")]; + string _SplitHeadsQ__mh_w_3423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3423_equation_0, values = (var_32996_cast_fp16, var_32627_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3423_cast_fp16")]; + string _SplitHeadsQ__mh_w_3425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3425_equation_0, values = (var_33000_cast_fp16, var_32634_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3425_cast_fp16")]; + string _SplitHeadsQ__mh_w_3427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3427_equation_0, values = (var_33000_cast_fp16, var_32641_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3427_cast_fp16")]; + string _SplitHeadsQ__mh_w_3429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3429_equation_0, values = (var_33000_cast_fp16, var_32648_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3429_cast_fp16")]; + string _SplitHeadsQ__mh_w_3431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3431_equation_0, values = (var_33000_cast_fp16, var_32655_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3431_cast_fp16")]; + string _SplitHeadsQ__mh_w_3433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3433_equation_0, values = (var_33004_cast_fp16, var_32662_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3433_cast_fp16")]; + string _SplitHeadsQ__mh_w_3435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3435_equation_0, values = (var_33004_cast_fp16, var_32669_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3435_cast_fp16")]; + string _SplitHeadsQ__mh_w_3437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3437_equation_0, values = (var_33004_cast_fp16, var_32676_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3437_cast_fp16")]; + string _SplitHeadsQ__mh_w_3439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3439_equation_0, values = (var_33004_cast_fp16, var_32683_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3439_cast_fp16")]; + string _SplitHeadsQ__mh_w_3441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3441_equation_0, values = (var_33008_cast_fp16, var_32690_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3441_cast_fp16")]; + string _SplitHeadsQ__mh_w_3443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3443_equation_0, values = (var_33008_cast_fp16, var_32697_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3443_cast_fp16")]; + string _SplitHeadsQ__mh_w_3445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3445_equation_0, values = (var_33008_cast_fp16, var_32704_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3445_cast_fp16")]; + string _SplitHeadsQ__mh_w_3447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3447_equation_0, values = (var_33008_cast_fp16, var_32711_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3447_cast_fp16")]; + string _SplitHeadsQ__mh_w_3449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3449_equation_0, values = (var_33012_cast_fp16, var_32718_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3449_cast_fp16")]; + string _SplitHeadsQ__mh_w_3451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3451_equation_0, values = (var_33012_cast_fp16, var_32725_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3451_cast_fp16")]; + string _SplitHeadsQ__mh_w_3453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3453_equation_0, values = (var_33012_cast_fp16, var_32732_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3453_cast_fp16")]; + string _SplitHeadsQ__mh_w_3455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3455_equation_0, values = (var_33012_cast_fp16, var_32739_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3455_cast_fp16")]; + string _SplitHeadsQ__mh_w_3457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3457_equation_0, values = (var_33016_cast_fp16, var_32746_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3457_cast_fp16")]; + string _SplitHeadsQ__mh_w_3459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3459_equation_0, values = (var_33016_cast_fp16, var_32753_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3459_cast_fp16")]; + string _SplitHeadsQ__mh_w_3461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3461_equation_0, values = (var_33016_cast_fp16, var_32760_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3461_cast_fp16")]; + string _SplitHeadsQ__mh_w_3463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3463_equation_0, values = (var_33016_cast_fp16, var_32767_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3463_cast_fp16")]; + string _SplitHeadsQ__mh_w_3465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3465_equation_0, values = (var_33020_cast_fp16, var_32774_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3465_cast_fp16")]; + string _SplitHeadsQ__mh_w_3467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3467_equation_0, values = (var_33020_cast_fp16, var_32781_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3467_cast_fp16")]; + string _SplitHeadsQ__mh_w_3469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3469_equation_0, values = (var_33020_cast_fp16, var_32788_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3469_cast_fp16")]; + string _SplitHeadsQ__mh_w_3471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3471_equation_0, values = (var_33020_cast_fp16, var_32795_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3471_cast_fp16")]; + string _SplitHeadsQ__mh_w_3473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3473_equation_0, values = (var_33024_cast_fp16, var_32802_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3473_cast_fp16")]; + string _SplitHeadsQ__mh_w_3475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3475_equation_0, values = (var_33024_cast_fp16, var_32809_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3475_cast_fp16")]; + string _SplitHeadsQ__mh_w_3477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3477_equation_0, values = (var_33024_cast_fp16, var_32816_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3477_cast_fp16")]; + string _SplitHeadsQ__mh_w_3479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3479_equation_0, values = (var_33024_cast_fp16, var_32823_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3479_cast_fp16")]; + string _SplitHeadsQ__mh_w_3481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3481_equation_0, values = (var_33028_cast_fp16, var_32830_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3481_cast_fp16")]; + string _SplitHeadsQ__mh_w_3483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3483_equation_0, values = (var_33028_cast_fp16, var_32837_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3483_cast_fp16")]; + string _SplitHeadsQ__mh_w_3485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3485_equation_0, values = (var_33028_cast_fp16, var_32844_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3485_cast_fp16")]; + string _SplitHeadsQ__mh_w_3487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3487_equation_0, values = (var_33028_cast_fp16, var_32851_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3487_cast_fp16")]; + string _SplitHeadsQ__mh_w_3489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3489_equation_0, values = (var_33032_cast_fp16, var_32858_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3489_cast_fp16")]; + string _SplitHeadsQ__mh_w_3491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3491_equation_0, values = (var_33032_cast_fp16, var_32865_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3491_cast_fp16")]; + string _SplitHeadsQ__mh_w_3493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3493_equation_0, values = (var_33032_cast_fp16, var_32872_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3493_cast_fp16")]; + string _SplitHeadsQ__mh_w_3495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3495_equation_0, values = (var_33032_cast_fp16, var_32879_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3495_cast_fp16")]; + string _SplitHeadsQ__mh_w_3497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3497_equation_0, values = (var_33036_cast_fp16, var_32886_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3497_cast_fp16")]; + string _SplitHeadsQ__mh_w_3499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3499_equation_0, values = (var_33036_cast_fp16, var_32893_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3499_cast_fp16")]; + string _SplitHeadsQ__mh_w_3501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3501_equation_0, values = (var_33036_cast_fp16, var_32900_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3501_cast_fp16")]; + string _SplitHeadsQ__mh_w_3503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3503_equation_0, values = (var_33036_cast_fp16, var_32907_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3503_cast_fp16")]; + string _SplitHeadsQ__mh_w_3505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3505_equation_0, values = (var_33040_cast_fp16, var_32914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3505_cast_fp16")]; + string _SplitHeadsQ__mh_w_3507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3507_equation_0, values = (var_33040_cast_fp16, var_32921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3507_cast_fp16")]; + string _SplitHeadsQ__mh_w_3509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3509_equation_0, values = (var_33040_cast_fp16, var_32928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3509_cast_fp16")]; + string _SplitHeadsQ__mh_w_3511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3511_equation_0, values = (var_33040_cast_fp16, var_32935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3511_cast_fp16")]; + string _SplitHeadsQ__mh_w_3513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3513_equation_0, values = (var_33044_cast_fp16, var_32942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3513_cast_fp16")]; + string _SplitHeadsQ__mh_w_3515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3515_equation_0, values = (var_33044_cast_fp16, var_32949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3515_cast_fp16")]; + string _SplitHeadsQ__mh_w_3517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3517_equation_0, values = (var_33044_cast_fp16, var_32956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3517_cast_fp16")]; + string _SplitHeadsQ__mh_w_3519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3519_equation_0, values = (var_33044_cast_fp16, var_32963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3519_cast_fp16")]; + fp16 var_33285_to_fp16 = const()[name = string("op_33285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3361_cast_fp16, y = var_33285_to_fp16)[name = string("aw_chunk_3361_cast_fp16")]; + fp16 var_33287_to_fp16 = const()[name = string("op_33287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3363_cast_fp16, y = var_33287_to_fp16)[name = string("aw_chunk_3363_cast_fp16")]; + fp16 var_33289_to_fp16 = const()[name = string("op_33289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3365_cast_fp16, y = var_33289_to_fp16)[name = string("aw_chunk_3365_cast_fp16")]; + fp16 var_33291_to_fp16 = const()[name = string("op_33291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3367_cast_fp16, y = var_33291_to_fp16)[name = string("aw_chunk_3367_cast_fp16")]; + fp16 var_33293_to_fp16 = const()[name = string("op_33293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3369_cast_fp16, y = var_33293_to_fp16)[name = string("aw_chunk_3369_cast_fp16")]; + fp16 var_33295_to_fp16 = const()[name = string("op_33295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3371_cast_fp16, y = var_33295_to_fp16)[name = string("aw_chunk_3371_cast_fp16")]; + fp16 var_33297_to_fp16 = const()[name = string("op_33297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3373_cast_fp16, y = var_33297_to_fp16)[name = string("aw_chunk_3373_cast_fp16")]; + fp16 var_33299_to_fp16 = const()[name = string("op_33299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3375_cast_fp16, y = var_33299_to_fp16)[name = string("aw_chunk_3375_cast_fp16")]; + fp16 var_33301_to_fp16 = const()[name = string("op_33301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3377_cast_fp16, y = var_33301_to_fp16)[name = string("aw_chunk_3377_cast_fp16")]; + fp16 var_33303_to_fp16 = const()[name = string("op_33303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3379_cast_fp16, y = var_33303_to_fp16)[name = string("aw_chunk_3379_cast_fp16")]; + fp16 var_33305_to_fp16 = const()[name = string("op_33305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3381_cast_fp16, y = var_33305_to_fp16)[name = string("aw_chunk_3381_cast_fp16")]; + fp16 var_33307_to_fp16 = const()[name = string("op_33307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3383_cast_fp16, y = var_33307_to_fp16)[name = string("aw_chunk_3383_cast_fp16")]; + fp16 var_33309_to_fp16 = const()[name = string("op_33309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3385_cast_fp16, y = var_33309_to_fp16)[name = string("aw_chunk_3385_cast_fp16")]; + fp16 var_33311_to_fp16 = const()[name = string("op_33311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3387_cast_fp16, y = var_33311_to_fp16)[name = string("aw_chunk_3387_cast_fp16")]; + fp16 var_33313_to_fp16 = const()[name = string("op_33313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3389_cast_fp16, y = var_33313_to_fp16)[name = string("aw_chunk_3389_cast_fp16")]; + fp16 var_33315_to_fp16 = const()[name = string("op_33315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3391_cast_fp16, y = var_33315_to_fp16)[name = string("aw_chunk_3391_cast_fp16")]; + fp16 var_33317_to_fp16 = const()[name = string("op_33317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3393_cast_fp16, y = var_33317_to_fp16)[name = string("aw_chunk_3393_cast_fp16")]; + fp16 var_33319_to_fp16 = const()[name = string("op_33319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3395_cast_fp16, y = var_33319_to_fp16)[name = string("aw_chunk_3395_cast_fp16")]; + fp16 var_33321_to_fp16 = const()[name = string("op_33321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3397_cast_fp16, y = var_33321_to_fp16)[name = string("aw_chunk_3397_cast_fp16")]; + fp16 var_33323_to_fp16 = const()[name = string("op_33323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3399_cast_fp16, y = var_33323_to_fp16)[name = string("aw_chunk_3399_cast_fp16")]; + fp16 var_33325_to_fp16 = const()[name = string("op_33325_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3401_cast_fp16, y = var_33325_to_fp16)[name = string("aw_chunk_3401_cast_fp16")]; + fp16 var_33327_to_fp16 = const()[name = string("op_33327_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3403_cast_fp16, y = var_33327_to_fp16)[name = string("aw_chunk_3403_cast_fp16")]; + fp16 var_33329_to_fp16 = const()[name = string("op_33329_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3405_cast_fp16, y = var_33329_to_fp16)[name = string("aw_chunk_3405_cast_fp16")]; + fp16 var_33331_to_fp16 = const()[name = string("op_33331_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3407_cast_fp16, y = var_33331_to_fp16)[name = string("aw_chunk_3407_cast_fp16")]; + fp16 var_33333_to_fp16 = const()[name = string("op_33333_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3409_cast_fp16, y = var_33333_to_fp16)[name = string("aw_chunk_3409_cast_fp16")]; + fp16 var_33335_to_fp16 = const()[name = string("op_33335_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3411_cast_fp16, y = var_33335_to_fp16)[name = string("aw_chunk_3411_cast_fp16")]; + fp16 var_33337_to_fp16 = const()[name = string("op_33337_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3413_cast_fp16, y = var_33337_to_fp16)[name = string("aw_chunk_3413_cast_fp16")]; + fp16 var_33339_to_fp16 = const()[name = string("op_33339_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3415_cast_fp16, y = var_33339_to_fp16)[name = string("aw_chunk_3415_cast_fp16")]; + fp16 var_33341_to_fp16 = const()[name = string("op_33341_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3417_cast_fp16, y = var_33341_to_fp16)[name = string("aw_chunk_3417_cast_fp16")]; + fp16 var_33343_to_fp16 = const()[name = string("op_33343_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3419_cast_fp16, y = var_33343_to_fp16)[name = string("aw_chunk_3419_cast_fp16")]; + fp16 var_33345_to_fp16 = const()[name = string("op_33345_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3421_cast_fp16, y = var_33345_to_fp16)[name = string("aw_chunk_3421_cast_fp16")]; + fp16 var_33347_to_fp16 = const()[name = string("op_33347_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3423_cast_fp16, y = var_33347_to_fp16)[name = string("aw_chunk_3423_cast_fp16")]; + fp16 var_33349_to_fp16 = const()[name = string("op_33349_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3425_cast_fp16, y = var_33349_to_fp16)[name = string("aw_chunk_3425_cast_fp16")]; + fp16 var_33351_to_fp16 = const()[name = string("op_33351_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3427_cast_fp16, y = var_33351_to_fp16)[name = string("aw_chunk_3427_cast_fp16")]; + fp16 var_33353_to_fp16 = const()[name = string("op_33353_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3429_cast_fp16, y = var_33353_to_fp16)[name = string("aw_chunk_3429_cast_fp16")]; + fp16 var_33355_to_fp16 = const()[name = string("op_33355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3431_cast_fp16, y = var_33355_to_fp16)[name = string("aw_chunk_3431_cast_fp16")]; + fp16 var_33357_to_fp16 = const()[name = string("op_33357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3433_cast_fp16, y = var_33357_to_fp16)[name = string("aw_chunk_3433_cast_fp16")]; + fp16 var_33359_to_fp16 = const()[name = string("op_33359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3435_cast_fp16, y = var_33359_to_fp16)[name = string("aw_chunk_3435_cast_fp16")]; + fp16 var_33361_to_fp16 = const()[name = string("op_33361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3437_cast_fp16, y = var_33361_to_fp16)[name = string("aw_chunk_3437_cast_fp16")]; + fp16 var_33363_to_fp16 = const()[name = string("op_33363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3439_cast_fp16, y = var_33363_to_fp16)[name = string("aw_chunk_3439_cast_fp16")]; + fp16 var_33365_to_fp16 = const()[name = string("op_33365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3441_cast_fp16, y = var_33365_to_fp16)[name = string("aw_chunk_3441_cast_fp16")]; + fp16 var_33367_to_fp16 = const()[name = string("op_33367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3443_cast_fp16, y = var_33367_to_fp16)[name = string("aw_chunk_3443_cast_fp16")]; + fp16 var_33369_to_fp16 = const()[name = string("op_33369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3445_cast_fp16, y = var_33369_to_fp16)[name = string("aw_chunk_3445_cast_fp16")]; + fp16 var_33371_to_fp16 = const()[name = string("op_33371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3447_cast_fp16, y = var_33371_to_fp16)[name = string("aw_chunk_3447_cast_fp16")]; + fp16 var_33373_to_fp16 = const()[name = string("op_33373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3449_cast_fp16, y = var_33373_to_fp16)[name = string("aw_chunk_3449_cast_fp16")]; + fp16 var_33375_to_fp16 = const()[name = string("op_33375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3451_cast_fp16, y = var_33375_to_fp16)[name = string("aw_chunk_3451_cast_fp16")]; + fp16 var_33377_to_fp16 = const()[name = string("op_33377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3453_cast_fp16, y = var_33377_to_fp16)[name = string("aw_chunk_3453_cast_fp16")]; + fp16 var_33379_to_fp16 = const()[name = string("op_33379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3455_cast_fp16, y = var_33379_to_fp16)[name = string("aw_chunk_3455_cast_fp16")]; + fp16 var_33381_to_fp16 = const()[name = string("op_33381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3457_cast_fp16, y = var_33381_to_fp16)[name = string("aw_chunk_3457_cast_fp16")]; + fp16 var_33383_to_fp16 = const()[name = string("op_33383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3459_cast_fp16, y = var_33383_to_fp16)[name = string("aw_chunk_3459_cast_fp16")]; + fp16 var_33385_to_fp16 = const()[name = string("op_33385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3461_cast_fp16, y = var_33385_to_fp16)[name = string("aw_chunk_3461_cast_fp16")]; + fp16 var_33387_to_fp16 = const()[name = string("op_33387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3463_cast_fp16, y = var_33387_to_fp16)[name = string("aw_chunk_3463_cast_fp16")]; + fp16 var_33389_to_fp16 = const()[name = string("op_33389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3465_cast_fp16, y = var_33389_to_fp16)[name = string("aw_chunk_3465_cast_fp16")]; + fp16 var_33391_to_fp16 = const()[name = string("op_33391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3467_cast_fp16, y = var_33391_to_fp16)[name = string("aw_chunk_3467_cast_fp16")]; + fp16 var_33393_to_fp16 = const()[name = string("op_33393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3469_cast_fp16, y = var_33393_to_fp16)[name = string("aw_chunk_3469_cast_fp16")]; + fp16 var_33395_to_fp16 = const()[name = string("op_33395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3471_cast_fp16, y = var_33395_to_fp16)[name = string("aw_chunk_3471_cast_fp16")]; + fp16 var_33397_to_fp16 = const()[name = string("op_33397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3473_cast_fp16, y = var_33397_to_fp16)[name = string("aw_chunk_3473_cast_fp16")]; + fp16 var_33399_to_fp16 = const()[name = string("op_33399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3475_cast_fp16, y = var_33399_to_fp16)[name = string("aw_chunk_3475_cast_fp16")]; + fp16 var_33401_to_fp16 = const()[name = string("op_33401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3477_cast_fp16, y = var_33401_to_fp16)[name = string("aw_chunk_3477_cast_fp16")]; + fp16 var_33403_to_fp16 = const()[name = string("op_33403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3479_cast_fp16, y = var_33403_to_fp16)[name = string("aw_chunk_3479_cast_fp16")]; + fp16 var_33405_to_fp16 = const()[name = string("op_33405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3481_cast_fp16, y = var_33405_to_fp16)[name = string("aw_chunk_3481_cast_fp16")]; + fp16 var_33407_to_fp16 = const()[name = string("op_33407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3483_cast_fp16, y = var_33407_to_fp16)[name = string("aw_chunk_3483_cast_fp16")]; + fp16 var_33409_to_fp16 = const()[name = string("op_33409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3485_cast_fp16, y = var_33409_to_fp16)[name = string("aw_chunk_3485_cast_fp16")]; + fp16 var_33411_to_fp16 = const()[name = string("op_33411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3487_cast_fp16, y = var_33411_to_fp16)[name = string("aw_chunk_3487_cast_fp16")]; + fp16 var_33413_to_fp16 = const()[name = string("op_33413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3489_cast_fp16, y = var_33413_to_fp16)[name = string("aw_chunk_3489_cast_fp16")]; + fp16 var_33415_to_fp16 = const()[name = string("op_33415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3491_cast_fp16, y = var_33415_to_fp16)[name = string("aw_chunk_3491_cast_fp16")]; + fp16 var_33417_to_fp16 = const()[name = string("op_33417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3493_cast_fp16, y = var_33417_to_fp16)[name = string("aw_chunk_3493_cast_fp16")]; + fp16 var_33419_to_fp16 = const()[name = string("op_33419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3495_cast_fp16, y = var_33419_to_fp16)[name = string("aw_chunk_3495_cast_fp16")]; + fp16 var_33421_to_fp16 = const()[name = string("op_33421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3497_cast_fp16, y = var_33421_to_fp16)[name = string("aw_chunk_3497_cast_fp16")]; + fp16 var_33423_to_fp16 = const()[name = string("op_33423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3499_cast_fp16, y = var_33423_to_fp16)[name = string("aw_chunk_3499_cast_fp16")]; + fp16 var_33425_to_fp16 = const()[name = string("op_33425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3501_cast_fp16, y = var_33425_to_fp16)[name = string("aw_chunk_3501_cast_fp16")]; + fp16 var_33427_to_fp16 = const()[name = string("op_33427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3503_cast_fp16, y = var_33427_to_fp16)[name = string("aw_chunk_3503_cast_fp16")]; + fp16 var_33429_to_fp16 = const()[name = string("op_33429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3505_cast_fp16, y = var_33429_to_fp16)[name = string("aw_chunk_3505_cast_fp16")]; + fp16 var_33431_to_fp16 = const()[name = string("op_33431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3507_cast_fp16, y = var_33431_to_fp16)[name = string("aw_chunk_3507_cast_fp16")]; + fp16 var_33433_to_fp16 = const()[name = string("op_33433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3509_cast_fp16, y = var_33433_to_fp16)[name = string("aw_chunk_3509_cast_fp16")]; + fp16 var_33435_to_fp16 = const()[name = string("op_33435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3511_cast_fp16, y = var_33435_to_fp16)[name = string("aw_chunk_3511_cast_fp16")]; + fp16 var_33437_to_fp16 = const()[name = string("op_33437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3513_cast_fp16, y = var_33437_to_fp16)[name = string("aw_chunk_3513_cast_fp16")]; + fp16 var_33439_to_fp16 = const()[name = string("op_33439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3515_cast_fp16, y = var_33439_to_fp16)[name = string("aw_chunk_3515_cast_fp16")]; + fp16 var_33441_to_fp16 = const()[name = string("op_33441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3517_cast_fp16, y = var_33441_to_fp16)[name = string("aw_chunk_3517_cast_fp16")]; + fp16 var_33443_to_fp16 = const()[name = string("op_33443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3519_cast_fp16, y = var_33443_to_fp16)[name = string("aw_chunk_3519_cast_fp16")]; + tensor var_33445_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3361_cast_fp16)[name = string("op_33445_cast_fp16")]; + tensor var_33446_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3363_cast_fp16)[name = string("op_33446_cast_fp16")]; + tensor var_33447_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3365_cast_fp16)[name = string("op_33447_cast_fp16")]; + tensor var_33448_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3367_cast_fp16)[name = string("op_33448_cast_fp16")]; + tensor var_33449_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3369_cast_fp16)[name = string("op_33449_cast_fp16")]; + tensor var_33450_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3371_cast_fp16)[name = string("op_33450_cast_fp16")]; + tensor var_33451_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3373_cast_fp16)[name = string("op_33451_cast_fp16")]; + tensor var_33452_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3375_cast_fp16)[name = string("op_33452_cast_fp16")]; + tensor var_33453_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3377_cast_fp16)[name = string("op_33453_cast_fp16")]; + tensor var_33454_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3379_cast_fp16)[name = string("op_33454_cast_fp16")]; + tensor var_33455_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3381_cast_fp16)[name = string("op_33455_cast_fp16")]; + tensor var_33456_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3383_cast_fp16)[name = string("op_33456_cast_fp16")]; + tensor var_33457_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3385_cast_fp16)[name = string("op_33457_cast_fp16")]; + tensor var_33458_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3387_cast_fp16)[name = string("op_33458_cast_fp16")]; + tensor var_33459_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3389_cast_fp16)[name = string("op_33459_cast_fp16")]; + tensor var_33460_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3391_cast_fp16)[name = string("op_33460_cast_fp16")]; + tensor var_33461_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3393_cast_fp16)[name = string("op_33461_cast_fp16")]; + tensor var_33462_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3395_cast_fp16)[name = string("op_33462_cast_fp16")]; + tensor var_33463_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3397_cast_fp16)[name = string("op_33463_cast_fp16")]; + tensor var_33464_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3399_cast_fp16)[name = string("op_33464_cast_fp16")]; + tensor var_33465_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3401_cast_fp16)[name = string("op_33465_cast_fp16")]; + tensor var_33466_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3403_cast_fp16)[name = string("op_33466_cast_fp16")]; + tensor var_33467_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3405_cast_fp16)[name = string("op_33467_cast_fp16")]; + tensor var_33468_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3407_cast_fp16)[name = string("op_33468_cast_fp16")]; + tensor var_33469_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3409_cast_fp16)[name = string("op_33469_cast_fp16")]; + tensor var_33470_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3411_cast_fp16)[name = string("op_33470_cast_fp16")]; + tensor var_33471_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3413_cast_fp16)[name = string("op_33471_cast_fp16")]; + tensor var_33472_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3415_cast_fp16)[name = string("op_33472_cast_fp16")]; + tensor var_33473_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3417_cast_fp16)[name = string("op_33473_cast_fp16")]; + tensor var_33474_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3419_cast_fp16)[name = string("op_33474_cast_fp16")]; + tensor var_33475_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3421_cast_fp16)[name = string("op_33475_cast_fp16")]; + tensor var_33476_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3423_cast_fp16)[name = string("op_33476_cast_fp16")]; + tensor var_33477_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3425_cast_fp16)[name = string("op_33477_cast_fp16")]; + tensor var_33478_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3427_cast_fp16)[name = string("op_33478_cast_fp16")]; + tensor var_33479_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3429_cast_fp16)[name = string("op_33479_cast_fp16")]; + tensor var_33480_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3431_cast_fp16)[name = string("op_33480_cast_fp16")]; + tensor var_33481_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3433_cast_fp16)[name = string("op_33481_cast_fp16")]; + tensor var_33482_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3435_cast_fp16)[name = string("op_33482_cast_fp16")]; + tensor var_33483_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3437_cast_fp16)[name = string("op_33483_cast_fp16")]; + tensor var_33484_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3439_cast_fp16)[name = string("op_33484_cast_fp16")]; + tensor var_33485_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3441_cast_fp16)[name = string("op_33485_cast_fp16")]; + tensor var_33486_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3443_cast_fp16)[name = string("op_33486_cast_fp16")]; + tensor var_33487_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3445_cast_fp16)[name = string("op_33487_cast_fp16")]; + tensor var_33488_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3447_cast_fp16)[name = string("op_33488_cast_fp16")]; + tensor var_33489_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3449_cast_fp16)[name = string("op_33489_cast_fp16")]; + tensor var_33490_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3451_cast_fp16)[name = string("op_33490_cast_fp16")]; + tensor var_33491_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3453_cast_fp16)[name = string("op_33491_cast_fp16")]; + tensor var_33492_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3455_cast_fp16)[name = string("op_33492_cast_fp16")]; + tensor var_33493_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3457_cast_fp16)[name = string("op_33493_cast_fp16")]; + tensor var_33494_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3459_cast_fp16)[name = string("op_33494_cast_fp16")]; + tensor var_33495_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3461_cast_fp16)[name = string("op_33495_cast_fp16")]; + tensor var_33496_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3463_cast_fp16)[name = string("op_33496_cast_fp16")]; + tensor var_33497_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3465_cast_fp16)[name = string("op_33497_cast_fp16")]; + tensor var_33498_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3467_cast_fp16)[name = string("op_33498_cast_fp16")]; + tensor var_33499_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3469_cast_fp16)[name = string("op_33499_cast_fp16")]; + tensor var_33500_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3471_cast_fp16)[name = string("op_33500_cast_fp16")]; + tensor var_33501_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3473_cast_fp16)[name = string("op_33501_cast_fp16")]; + tensor var_33502_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3475_cast_fp16)[name = string("op_33502_cast_fp16")]; + tensor var_33503_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3477_cast_fp16)[name = string("op_33503_cast_fp16")]; + tensor var_33504_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3479_cast_fp16)[name = string("op_33504_cast_fp16")]; + tensor var_33505_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3481_cast_fp16)[name = string("op_33505_cast_fp16")]; + tensor var_33506_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3483_cast_fp16)[name = string("op_33506_cast_fp16")]; + tensor var_33507_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3485_cast_fp16)[name = string("op_33507_cast_fp16")]; + tensor var_33508_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3487_cast_fp16)[name = string("op_33508_cast_fp16")]; + tensor var_33509_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3489_cast_fp16)[name = string("op_33509_cast_fp16")]; + tensor var_33510_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3491_cast_fp16)[name = string("op_33510_cast_fp16")]; + tensor var_33511_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3493_cast_fp16)[name = string("op_33511_cast_fp16")]; + tensor var_33512_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3495_cast_fp16)[name = string("op_33512_cast_fp16")]; + tensor var_33513_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3497_cast_fp16)[name = string("op_33513_cast_fp16")]; + tensor var_33514_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3499_cast_fp16)[name = string("op_33514_cast_fp16")]; + tensor var_33515_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3501_cast_fp16)[name = string("op_33515_cast_fp16")]; + tensor var_33516_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3503_cast_fp16)[name = string("op_33516_cast_fp16")]; + tensor var_33517_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3505_cast_fp16)[name = string("op_33517_cast_fp16")]; + tensor var_33518_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3507_cast_fp16)[name = string("op_33518_cast_fp16")]; + tensor var_33519_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3509_cast_fp16)[name = string("op_33519_cast_fp16")]; + tensor var_33520_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3511_cast_fp16)[name = string("op_33520_cast_fp16")]; + tensor var_33521_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3513_cast_fp16)[name = string("op_33521_cast_fp16")]; + tensor var_33522_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3515_cast_fp16)[name = string("op_33522_cast_fp16")]; + tensor var_33523_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3517_cast_fp16)[name = string("op_33523_cast_fp16")]; + tensor var_33524_cast_fp16 = softmax(axis = var_32270, x = aw_chunk_3519_cast_fp16)[name = string("op_33524_cast_fp16")]; + string var_33526_equation_0 = const()[name = string("op_33526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33526_cast_fp16 = einsum(equation = var_33526_equation_0, values = (var_33046_cast_fp16, var_33445_cast_fp16))[name = string("op_33526_cast_fp16")]; + string var_33528_equation_0 = const()[name = string("op_33528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33528_cast_fp16 = einsum(equation = var_33528_equation_0, values = (var_33046_cast_fp16, var_33446_cast_fp16))[name = string("op_33528_cast_fp16")]; + string var_33530_equation_0 = const()[name = string("op_33530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33530_cast_fp16 = einsum(equation = var_33530_equation_0, values = (var_33046_cast_fp16, var_33447_cast_fp16))[name = string("op_33530_cast_fp16")]; + string var_33532_equation_0 = const()[name = string("op_33532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33532_cast_fp16 = einsum(equation = var_33532_equation_0, values = (var_33046_cast_fp16, var_33448_cast_fp16))[name = string("op_33532_cast_fp16")]; + string var_33534_equation_0 = const()[name = string("op_33534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33534_cast_fp16 = einsum(equation = var_33534_equation_0, values = (var_33050_cast_fp16, var_33449_cast_fp16))[name = string("op_33534_cast_fp16")]; + string var_33536_equation_0 = const()[name = string("op_33536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33536_cast_fp16 = einsum(equation = var_33536_equation_0, values = (var_33050_cast_fp16, var_33450_cast_fp16))[name = string("op_33536_cast_fp16")]; + string var_33538_equation_0 = const()[name = string("op_33538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33538_cast_fp16 = einsum(equation = var_33538_equation_0, values = (var_33050_cast_fp16, var_33451_cast_fp16))[name = string("op_33538_cast_fp16")]; + string var_33540_equation_0 = const()[name = string("op_33540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33540_cast_fp16 = einsum(equation = var_33540_equation_0, values = (var_33050_cast_fp16, var_33452_cast_fp16))[name = string("op_33540_cast_fp16")]; + string var_33542_equation_0 = const()[name = string("op_33542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33542_cast_fp16 = einsum(equation = var_33542_equation_0, values = (var_33054_cast_fp16, var_33453_cast_fp16))[name = string("op_33542_cast_fp16")]; + string var_33544_equation_0 = const()[name = string("op_33544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33544_cast_fp16 = einsum(equation = var_33544_equation_0, values = (var_33054_cast_fp16, var_33454_cast_fp16))[name = string("op_33544_cast_fp16")]; + string var_33546_equation_0 = const()[name = string("op_33546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33546_cast_fp16 = einsum(equation = var_33546_equation_0, values = (var_33054_cast_fp16, var_33455_cast_fp16))[name = string("op_33546_cast_fp16")]; + string var_33548_equation_0 = const()[name = string("op_33548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33548_cast_fp16 = einsum(equation = var_33548_equation_0, values = (var_33054_cast_fp16, var_33456_cast_fp16))[name = string("op_33548_cast_fp16")]; + string var_33550_equation_0 = const()[name = string("op_33550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33550_cast_fp16 = einsum(equation = var_33550_equation_0, values = (var_33058_cast_fp16, var_33457_cast_fp16))[name = string("op_33550_cast_fp16")]; + string var_33552_equation_0 = const()[name = string("op_33552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33552_cast_fp16 = einsum(equation = var_33552_equation_0, values = (var_33058_cast_fp16, var_33458_cast_fp16))[name = string("op_33552_cast_fp16")]; + string var_33554_equation_0 = const()[name = string("op_33554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33554_cast_fp16 = einsum(equation = var_33554_equation_0, values = (var_33058_cast_fp16, var_33459_cast_fp16))[name = string("op_33554_cast_fp16")]; + string var_33556_equation_0 = const()[name = string("op_33556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33556_cast_fp16 = einsum(equation = var_33556_equation_0, values = (var_33058_cast_fp16, var_33460_cast_fp16))[name = string("op_33556_cast_fp16")]; + string var_33558_equation_0 = const()[name = string("op_33558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33558_cast_fp16 = einsum(equation = var_33558_equation_0, values = (var_33062_cast_fp16, var_33461_cast_fp16))[name = string("op_33558_cast_fp16")]; + string var_33560_equation_0 = const()[name = string("op_33560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33560_cast_fp16 = einsum(equation = var_33560_equation_0, values = (var_33062_cast_fp16, var_33462_cast_fp16))[name = string("op_33560_cast_fp16")]; + string var_33562_equation_0 = const()[name = string("op_33562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33562_cast_fp16 = einsum(equation = var_33562_equation_0, values = (var_33062_cast_fp16, var_33463_cast_fp16))[name = string("op_33562_cast_fp16")]; + string var_33564_equation_0 = const()[name = string("op_33564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33564_cast_fp16 = einsum(equation = var_33564_equation_0, values = (var_33062_cast_fp16, var_33464_cast_fp16))[name = string("op_33564_cast_fp16")]; + string var_33566_equation_0 = const()[name = string("op_33566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33566_cast_fp16 = einsum(equation = var_33566_equation_0, values = (var_33066_cast_fp16, var_33465_cast_fp16))[name = string("op_33566_cast_fp16")]; + string var_33568_equation_0 = const()[name = string("op_33568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33568_cast_fp16 = einsum(equation = var_33568_equation_0, values = (var_33066_cast_fp16, var_33466_cast_fp16))[name = string("op_33568_cast_fp16")]; + string var_33570_equation_0 = const()[name = string("op_33570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33570_cast_fp16 = einsum(equation = var_33570_equation_0, values = (var_33066_cast_fp16, var_33467_cast_fp16))[name = string("op_33570_cast_fp16")]; + string var_33572_equation_0 = const()[name = string("op_33572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33572_cast_fp16 = einsum(equation = var_33572_equation_0, values = (var_33066_cast_fp16, var_33468_cast_fp16))[name = string("op_33572_cast_fp16")]; + string var_33574_equation_0 = const()[name = string("op_33574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33574_cast_fp16 = einsum(equation = var_33574_equation_0, values = (var_33070_cast_fp16, var_33469_cast_fp16))[name = string("op_33574_cast_fp16")]; + string var_33576_equation_0 = const()[name = string("op_33576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33576_cast_fp16 = einsum(equation = var_33576_equation_0, values = (var_33070_cast_fp16, var_33470_cast_fp16))[name = string("op_33576_cast_fp16")]; + string var_33578_equation_0 = const()[name = string("op_33578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33578_cast_fp16 = einsum(equation = var_33578_equation_0, values = (var_33070_cast_fp16, var_33471_cast_fp16))[name = string("op_33578_cast_fp16")]; + string var_33580_equation_0 = const()[name = string("op_33580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33580_cast_fp16 = einsum(equation = var_33580_equation_0, values = (var_33070_cast_fp16, var_33472_cast_fp16))[name = string("op_33580_cast_fp16")]; + string var_33582_equation_0 = const()[name = string("op_33582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33582_cast_fp16 = einsum(equation = var_33582_equation_0, values = (var_33074_cast_fp16, var_33473_cast_fp16))[name = string("op_33582_cast_fp16")]; + string var_33584_equation_0 = const()[name = string("op_33584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33584_cast_fp16 = einsum(equation = var_33584_equation_0, values = (var_33074_cast_fp16, var_33474_cast_fp16))[name = string("op_33584_cast_fp16")]; + string var_33586_equation_0 = const()[name = string("op_33586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33586_cast_fp16 = einsum(equation = var_33586_equation_0, values = (var_33074_cast_fp16, var_33475_cast_fp16))[name = string("op_33586_cast_fp16")]; + string var_33588_equation_0 = const()[name = string("op_33588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33588_cast_fp16 = einsum(equation = var_33588_equation_0, values = (var_33074_cast_fp16, var_33476_cast_fp16))[name = string("op_33588_cast_fp16")]; + string var_33590_equation_0 = const()[name = string("op_33590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33590_cast_fp16 = einsum(equation = var_33590_equation_0, values = (var_33078_cast_fp16, var_33477_cast_fp16))[name = string("op_33590_cast_fp16")]; + string var_33592_equation_0 = const()[name = string("op_33592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33592_cast_fp16 = einsum(equation = var_33592_equation_0, values = (var_33078_cast_fp16, var_33478_cast_fp16))[name = string("op_33592_cast_fp16")]; + string var_33594_equation_0 = const()[name = string("op_33594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33594_cast_fp16 = einsum(equation = var_33594_equation_0, values = (var_33078_cast_fp16, var_33479_cast_fp16))[name = string("op_33594_cast_fp16")]; + string var_33596_equation_0 = const()[name = string("op_33596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33596_cast_fp16 = einsum(equation = var_33596_equation_0, values = (var_33078_cast_fp16, var_33480_cast_fp16))[name = string("op_33596_cast_fp16")]; + string var_33598_equation_0 = const()[name = string("op_33598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33598_cast_fp16 = einsum(equation = var_33598_equation_0, values = (var_33082_cast_fp16, var_33481_cast_fp16))[name = string("op_33598_cast_fp16")]; + string var_33600_equation_0 = const()[name = string("op_33600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33600_cast_fp16 = einsum(equation = var_33600_equation_0, values = (var_33082_cast_fp16, var_33482_cast_fp16))[name = string("op_33600_cast_fp16")]; + string var_33602_equation_0 = const()[name = string("op_33602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33602_cast_fp16 = einsum(equation = var_33602_equation_0, values = (var_33082_cast_fp16, var_33483_cast_fp16))[name = string("op_33602_cast_fp16")]; + string var_33604_equation_0 = const()[name = string("op_33604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33604_cast_fp16 = einsum(equation = var_33604_equation_0, values = (var_33082_cast_fp16, var_33484_cast_fp16))[name = string("op_33604_cast_fp16")]; + string var_33606_equation_0 = const()[name = string("op_33606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33606_cast_fp16 = einsum(equation = var_33606_equation_0, values = (var_33086_cast_fp16, var_33485_cast_fp16))[name = string("op_33606_cast_fp16")]; + string var_33608_equation_0 = const()[name = string("op_33608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33608_cast_fp16 = einsum(equation = var_33608_equation_0, values = (var_33086_cast_fp16, var_33486_cast_fp16))[name = string("op_33608_cast_fp16")]; + string var_33610_equation_0 = const()[name = string("op_33610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33610_cast_fp16 = einsum(equation = var_33610_equation_0, values = (var_33086_cast_fp16, var_33487_cast_fp16))[name = string("op_33610_cast_fp16")]; + string var_33612_equation_0 = const()[name = string("op_33612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33612_cast_fp16 = einsum(equation = var_33612_equation_0, values = (var_33086_cast_fp16, var_33488_cast_fp16))[name = string("op_33612_cast_fp16")]; + string var_33614_equation_0 = const()[name = string("op_33614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33614_cast_fp16 = einsum(equation = var_33614_equation_0, values = (var_33090_cast_fp16, var_33489_cast_fp16))[name = string("op_33614_cast_fp16")]; + string var_33616_equation_0 = const()[name = string("op_33616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33616_cast_fp16 = einsum(equation = var_33616_equation_0, values = (var_33090_cast_fp16, var_33490_cast_fp16))[name = string("op_33616_cast_fp16")]; + string var_33618_equation_0 = const()[name = string("op_33618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33618_cast_fp16 = einsum(equation = var_33618_equation_0, values = (var_33090_cast_fp16, var_33491_cast_fp16))[name = string("op_33618_cast_fp16")]; + string var_33620_equation_0 = const()[name = string("op_33620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33620_cast_fp16 = einsum(equation = var_33620_equation_0, values = (var_33090_cast_fp16, var_33492_cast_fp16))[name = string("op_33620_cast_fp16")]; + string var_33622_equation_0 = const()[name = string("op_33622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33622_cast_fp16 = einsum(equation = var_33622_equation_0, values = (var_33094_cast_fp16, var_33493_cast_fp16))[name = string("op_33622_cast_fp16")]; + string var_33624_equation_0 = const()[name = string("op_33624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33624_cast_fp16 = einsum(equation = var_33624_equation_0, values = (var_33094_cast_fp16, var_33494_cast_fp16))[name = string("op_33624_cast_fp16")]; + string var_33626_equation_0 = const()[name = string("op_33626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33626_cast_fp16 = einsum(equation = var_33626_equation_0, values = (var_33094_cast_fp16, var_33495_cast_fp16))[name = string("op_33626_cast_fp16")]; + string var_33628_equation_0 = const()[name = string("op_33628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33628_cast_fp16 = einsum(equation = var_33628_equation_0, values = (var_33094_cast_fp16, var_33496_cast_fp16))[name = string("op_33628_cast_fp16")]; + string var_33630_equation_0 = const()[name = string("op_33630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33630_cast_fp16 = einsum(equation = var_33630_equation_0, values = (var_33098_cast_fp16, var_33497_cast_fp16))[name = string("op_33630_cast_fp16")]; + string var_33632_equation_0 = const()[name = string("op_33632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33632_cast_fp16 = einsum(equation = var_33632_equation_0, values = (var_33098_cast_fp16, var_33498_cast_fp16))[name = string("op_33632_cast_fp16")]; + string var_33634_equation_0 = const()[name = string("op_33634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33634_cast_fp16 = einsum(equation = var_33634_equation_0, values = (var_33098_cast_fp16, var_33499_cast_fp16))[name = string("op_33634_cast_fp16")]; + string var_33636_equation_0 = const()[name = string("op_33636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33636_cast_fp16 = einsum(equation = var_33636_equation_0, values = (var_33098_cast_fp16, var_33500_cast_fp16))[name = string("op_33636_cast_fp16")]; + string var_33638_equation_0 = const()[name = string("op_33638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33638_cast_fp16 = einsum(equation = var_33638_equation_0, values = (var_33102_cast_fp16, var_33501_cast_fp16))[name = string("op_33638_cast_fp16")]; + string var_33640_equation_0 = const()[name = string("op_33640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33640_cast_fp16 = einsum(equation = var_33640_equation_0, values = (var_33102_cast_fp16, var_33502_cast_fp16))[name = string("op_33640_cast_fp16")]; + string var_33642_equation_0 = const()[name = string("op_33642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33642_cast_fp16 = einsum(equation = var_33642_equation_0, values = (var_33102_cast_fp16, var_33503_cast_fp16))[name = string("op_33642_cast_fp16")]; + string var_33644_equation_0 = const()[name = string("op_33644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33644_cast_fp16 = einsum(equation = var_33644_equation_0, values = (var_33102_cast_fp16, var_33504_cast_fp16))[name = string("op_33644_cast_fp16")]; + string var_33646_equation_0 = const()[name = string("op_33646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33646_cast_fp16 = einsum(equation = var_33646_equation_0, values = (var_33106_cast_fp16, var_33505_cast_fp16))[name = string("op_33646_cast_fp16")]; + string var_33648_equation_0 = const()[name = string("op_33648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33648_cast_fp16 = einsum(equation = var_33648_equation_0, values = (var_33106_cast_fp16, var_33506_cast_fp16))[name = string("op_33648_cast_fp16")]; + string var_33650_equation_0 = const()[name = string("op_33650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33650_cast_fp16 = einsum(equation = var_33650_equation_0, values = (var_33106_cast_fp16, var_33507_cast_fp16))[name = string("op_33650_cast_fp16")]; + string var_33652_equation_0 = const()[name = string("op_33652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33652_cast_fp16 = einsum(equation = var_33652_equation_0, values = (var_33106_cast_fp16, var_33508_cast_fp16))[name = string("op_33652_cast_fp16")]; + string var_33654_equation_0 = const()[name = string("op_33654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33654_cast_fp16 = einsum(equation = var_33654_equation_0, values = (var_33110_cast_fp16, var_33509_cast_fp16))[name = string("op_33654_cast_fp16")]; + string var_33656_equation_0 = const()[name = string("op_33656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33656_cast_fp16 = einsum(equation = var_33656_equation_0, values = (var_33110_cast_fp16, var_33510_cast_fp16))[name = string("op_33656_cast_fp16")]; + string var_33658_equation_0 = const()[name = string("op_33658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33658_cast_fp16 = einsum(equation = var_33658_equation_0, values = (var_33110_cast_fp16, var_33511_cast_fp16))[name = string("op_33658_cast_fp16")]; + string var_33660_equation_0 = const()[name = string("op_33660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33660_cast_fp16 = einsum(equation = var_33660_equation_0, values = (var_33110_cast_fp16, var_33512_cast_fp16))[name = string("op_33660_cast_fp16")]; + string var_33662_equation_0 = const()[name = string("op_33662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33662_cast_fp16 = einsum(equation = var_33662_equation_0, values = (var_33114_cast_fp16, var_33513_cast_fp16))[name = string("op_33662_cast_fp16")]; + string var_33664_equation_0 = const()[name = string("op_33664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33664_cast_fp16 = einsum(equation = var_33664_equation_0, values = (var_33114_cast_fp16, var_33514_cast_fp16))[name = string("op_33664_cast_fp16")]; + string var_33666_equation_0 = const()[name = string("op_33666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33666_cast_fp16 = einsum(equation = var_33666_equation_0, values = (var_33114_cast_fp16, var_33515_cast_fp16))[name = string("op_33666_cast_fp16")]; + string var_33668_equation_0 = const()[name = string("op_33668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33668_cast_fp16 = einsum(equation = var_33668_equation_0, values = (var_33114_cast_fp16, var_33516_cast_fp16))[name = string("op_33668_cast_fp16")]; + string var_33670_equation_0 = const()[name = string("op_33670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33670_cast_fp16 = einsum(equation = var_33670_equation_0, values = (var_33118_cast_fp16, var_33517_cast_fp16))[name = string("op_33670_cast_fp16")]; + string var_33672_equation_0 = const()[name = string("op_33672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33672_cast_fp16 = einsum(equation = var_33672_equation_0, values = (var_33118_cast_fp16, var_33518_cast_fp16))[name = string("op_33672_cast_fp16")]; + string var_33674_equation_0 = const()[name = string("op_33674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33674_cast_fp16 = einsum(equation = var_33674_equation_0, values = (var_33118_cast_fp16, var_33519_cast_fp16))[name = string("op_33674_cast_fp16")]; + string var_33676_equation_0 = const()[name = string("op_33676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33676_cast_fp16 = einsum(equation = var_33676_equation_0, values = (var_33118_cast_fp16, var_33520_cast_fp16))[name = string("op_33676_cast_fp16")]; + string var_33678_equation_0 = const()[name = string("op_33678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33678_cast_fp16 = einsum(equation = var_33678_equation_0, values = (var_33122_cast_fp16, var_33521_cast_fp16))[name = string("op_33678_cast_fp16")]; + string var_33680_equation_0 = const()[name = string("op_33680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33680_cast_fp16 = einsum(equation = var_33680_equation_0, values = (var_33122_cast_fp16, var_33522_cast_fp16))[name = string("op_33680_cast_fp16")]; + string var_33682_equation_0 = const()[name = string("op_33682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33682_cast_fp16 = einsum(equation = var_33682_equation_0, values = (var_33122_cast_fp16, var_33523_cast_fp16))[name = string("op_33682_cast_fp16")]; + string var_33684_equation_0 = const()[name = string("op_33684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33684_cast_fp16 = einsum(equation = var_33684_equation_0, values = (var_33122_cast_fp16, var_33524_cast_fp16))[name = string("op_33684_cast_fp16")]; + bool var_33686_interleave_0 = const()[name = string("op_33686_interleave_0"), val = bool(false)]; + tensor var_33686_cast_fp16 = concat(axis = var_32245, interleave = var_33686_interleave_0, values = (var_33526_cast_fp16, var_33528_cast_fp16, var_33530_cast_fp16, var_33532_cast_fp16))[name = string("op_33686_cast_fp16")]; + bool var_33688_interleave_0 = const()[name = string("op_33688_interleave_0"), val = bool(false)]; + tensor var_33688_cast_fp16 = concat(axis = var_32245, interleave = var_33688_interleave_0, values = (var_33534_cast_fp16, var_33536_cast_fp16, var_33538_cast_fp16, var_33540_cast_fp16))[name = string("op_33688_cast_fp16")]; + bool var_33690_interleave_0 = const()[name = string("op_33690_interleave_0"), val = bool(false)]; + tensor var_33690_cast_fp16 = concat(axis = var_32245, interleave = var_33690_interleave_0, values = (var_33542_cast_fp16, var_33544_cast_fp16, var_33546_cast_fp16, var_33548_cast_fp16))[name = string("op_33690_cast_fp16")]; + bool var_33692_interleave_0 = const()[name = string("op_33692_interleave_0"), val = bool(false)]; + tensor var_33692_cast_fp16 = concat(axis = var_32245, interleave = var_33692_interleave_0, values = (var_33550_cast_fp16, var_33552_cast_fp16, var_33554_cast_fp16, var_33556_cast_fp16))[name = string("op_33692_cast_fp16")]; + bool var_33694_interleave_0 = const()[name = string("op_33694_interleave_0"), val = bool(false)]; + tensor var_33694_cast_fp16 = concat(axis = var_32245, interleave = var_33694_interleave_0, values = (var_33558_cast_fp16, var_33560_cast_fp16, var_33562_cast_fp16, var_33564_cast_fp16))[name = string("op_33694_cast_fp16")]; + bool var_33696_interleave_0 = const()[name = string("op_33696_interleave_0"), val = bool(false)]; + tensor var_33696_cast_fp16 = concat(axis = var_32245, interleave = var_33696_interleave_0, values = (var_33566_cast_fp16, var_33568_cast_fp16, var_33570_cast_fp16, var_33572_cast_fp16))[name = string("op_33696_cast_fp16")]; + bool var_33698_interleave_0 = const()[name = string("op_33698_interleave_0"), val = bool(false)]; + tensor var_33698_cast_fp16 = concat(axis = var_32245, interleave = var_33698_interleave_0, values = (var_33574_cast_fp16, var_33576_cast_fp16, var_33578_cast_fp16, var_33580_cast_fp16))[name = string("op_33698_cast_fp16")]; + bool var_33700_interleave_0 = const()[name = string("op_33700_interleave_0"), val = bool(false)]; + tensor var_33700_cast_fp16 = concat(axis = var_32245, interleave = var_33700_interleave_0, values = (var_33582_cast_fp16, var_33584_cast_fp16, var_33586_cast_fp16, var_33588_cast_fp16))[name = string("op_33700_cast_fp16")]; + bool var_33702_interleave_0 = const()[name = string("op_33702_interleave_0"), val = bool(false)]; + tensor var_33702_cast_fp16 = concat(axis = var_32245, interleave = var_33702_interleave_0, values = (var_33590_cast_fp16, var_33592_cast_fp16, var_33594_cast_fp16, var_33596_cast_fp16))[name = string("op_33702_cast_fp16")]; + bool var_33704_interleave_0 = const()[name = string("op_33704_interleave_0"), val = bool(false)]; + tensor var_33704_cast_fp16 = concat(axis = var_32245, interleave = var_33704_interleave_0, values = (var_33598_cast_fp16, var_33600_cast_fp16, var_33602_cast_fp16, var_33604_cast_fp16))[name = string("op_33704_cast_fp16")]; + bool var_33706_interleave_0 = const()[name = string("op_33706_interleave_0"), val = bool(false)]; + tensor var_33706_cast_fp16 = concat(axis = var_32245, interleave = var_33706_interleave_0, values = (var_33606_cast_fp16, var_33608_cast_fp16, var_33610_cast_fp16, var_33612_cast_fp16))[name = string("op_33706_cast_fp16")]; + bool var_33708_interleave_0 = const()[name = string("op_33708_interleave_0"), val = bool(false)]; + tensor var_33708_cast_fp16 = concat(axis = var_32245, interleave = var_33708_interleave_0, values = (var_33614_cast_fp16, var_33616_cast_fp16, var_33618_cast_fp16, var_33620_cast_fp16))[name = string("op_33708_cast_fp16")]; + bool var_33710_interleave_0 = const()[name = string("op_33710_interleave_0"), val = bool(false)]; + tensor var_33710_cast_fp16 = concat(axis = var_32245, interleave = var_33710_interleave_0, values = (var_33622_cast_fp16, var_33624_cast_fp16, var_33626_cast_fp16, var_33628_cast_fp16))[name = string("op_33710_cast_fp16")]; + bool var_33712_interleave_0 = const()[name = string("op_33712_interleave_0"), val = bool(false)]; + tensor var_33712_cast_fp16 = concat(axis = var_32245, interleave = var_33712_interleave_0, values = (var_33630_cast_fp16, var_33632_cast_fp16, var_33634_cast_fp16, var_33636_cast_fp16))[name = string("op_33712_cast_fp16")]; + bool var_33714_interleave_0 = const()[name = string("op_33714_interleave_0"), val = bool(false)]; + tensor var_33714_cast_fp16 = concat(axis = var_32245, interleave = var_33714_interleave_0, values = (var_33638_cast_fp16, var_33640_cast_fp16, var_33642_cast_fp16, var_33644_cast_fp16))[name = string("op_33714_cast_fp16")]; + bool var_33716_interleave_0 = const()[name = string("op_33716_interleave_0"), val = bool(false)]; + tensor var_33716_cast_fp16 = concat(axis = var_32245, interleave = var_33716_interleave_0, values = (var_33646_cast_fp16, var_33648_cast_fp16, var_33650_cast_fp16, var_33652_cast_fp16))[name = string("op_33716_cast_fp16")]; + bool var_33718_interleave_0 = const()[name = string("op_33718_interleave_0"), val = bool(false)]; + tensor var_33718_cast_fp16 = concat(axis = var_32245, interleave = var_33718_interleave_0, values = (var_33654_cast_fp16, var_33656_cast_fp16, var_33658_cast_fp16, var_33660_cast_fp16))[name = string("op_33718_cast_fp16")]; + bool var_33720_interleave_0 = const()[name = string("op_33720_interleave_0"), val = bool(false)]; + tensor var_33720_cast_fp16 = concat(axis = var_32245, interleave = var_33720_interleave_0, values = (var_33662_cast_fp16, var_33664_cast_fp16, var_33666_cast_fp16, var_33668_cast_fp16))[name = string("op_33720_cast_fp16")]; + bool var_33722_interleave_0 = const()[name = string("op_33722_interleave_0"), val = bool(false)]; + tensor var_33722_cast_fp16 = concat(axis = var_32245, interleave = var_33722_interleave_0, values = (var_33670_cast_fp16, var_33672_cast_fp16, var_33674_cast_fp16, var_33676_cast_fp16))[name = string("op_33722_cast_fp16")]; + bool var_33724_interleave_0 = const()[name = string("op_33724_interleave_0"), val = bool(false)]; + tensor var_33724_cast_fp16 = concat(axis = var_32245, interleave = var_33724_interleave_0, values = (var_33678_cast_fp16, var_33680_cast_fp16, var_33682_cast_fp16, var_33684_cast_fp16))[name = string("op_33724_cast_fp16")]; + bool input_169_interleave_0 = const()[name = string("input_169_interleave_0"), val = bool(false)]; + tensor input_169_cast_fp16 = concat(axis = var_32270, interleave = input_169_interleave_0, values = (var_33686_cast_fp16, var_33688_cast_fp16, var_33690_cast_fp16, var_33692_cast_fp16, var_33694_cast_fp16, var_33696_cast_fp16, var_33698_cast_fp16, var_33700_cast_fp16, var_33702_cast_fp16, var_33704_cast_fp16, var_33706_cast_fp16, var_33708_cast_fp16, var_33710_cast_fp16, var_33712_cast_fp16, var_33714_cast_fp16, var_33716_cast_fp16, var_33718_cast_fp16, var_33720_cast_fp16, var_33722_cast_fp16, var_33724_cast_fp16))[name = string("input_169_cast_fp16")]; + string obj_87_pad_type_0 = const()[name = string("obj_87_pad_type_0"), val = string("valid")]; + tensor obj_87_strides_0 = const()[name = string("obj_87_strides_0"), val = tensor([1, 1])]; + tensor obj_87_pad_0 = const()[name = string("obj_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_87_dilations_0 = const()[name = string("obj_87_dilations_0"), val = tensor([1, 1])]; + int32 obj_87_groups_0 = const()[name = string("obj_87_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850924160)))]; + tensor layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854201024)))]; + tensor obj_87_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_87_dilations_0, groups = obj_87_groups_0, pad = obj_87_pad_0, pad_type = obj_87_pad_type_0, strides = obj_87_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_169_cast_fp16)[name = string("obj_87_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_87_cast_fp16)[name = string("inputs_87_cast_fp16")]; + tensor out_87_axes_0 = const()[name = string("out_87_axes_0"), val = tensor([1])]; + fp16 var_33743_to_fp16 = const()[name = string("op_33743_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_33743_to_fp16, x = inputs_87_cast_fp16)[name = string("out_87_cast_fp16")]; + tensor input_171_gamma_0_to_fp16 = const()[name = string("input_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854203648)))]; + tensor input_171_beta_0_to_fp16 = const()[name = string("input_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854206272)))]; + fp16 input_171_epsilon_0_to_fp16 = const()[name = string("input_171_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_171_cast_fp16 = batch_norm(beta = input_171_beta_0_to_fp16, epsilon = input_171_epsilon_0_to_fp16, gamma = input_171_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = string("input_171_cast_fp16")]; + string input_173_pad_type_0 = const()[name = string("input_173_pad_type_0"), val = string("valid")]; + tensor input_173_strides_0 = const()[name = string("input_173_strides_0"), val = tensor([1, 1])]; + tensor input_173_pad_0 = const()[name = string("input_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_173_dilations_0 = const()[name = string("input_173_dilations_0"), val = tensor([1, 1])]; + int32 input_173_groups_0 = const()[name = string("input_173_groups_0"), val = int32(1)]; + tensor layers_21_fc1_weight_to_fp16 = const()[name = string("layers_21_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854208896)))]; + tensor layers_21_fc1_bias_to_fp16 = const()[name = string("layers_21_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867316160)))]; + tensor input_173_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_173_dilations_0, groups = input_173_groups_0, pad = input_173_pad_0, pad_type = input_173_pad_type_0, strides = input_173_strides_0, weight = layers_21_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = string("input_173_cast_fp16")]; + string input_175_mode_0 = const()[name = string("input_175_mode_0"), val = string("EXACT")]; + tensor input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")]; + string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")]; + tensor hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)]; + tensor layers_21_fc2_weight_to_fp16 = const()[name = string("layers_21_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867326464)))]; + tensor layers_21_fc2_bias_to_fp16 = const()[name = string("layers_21_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880433728)))]; + tensor hidden_states_47_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_21_fc2_weight_to_fp16, x = input_175_cast_fp16)[name = string("hidden_states_47_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = hidden_states_47_cast_fp16)[name = string("inputs_89_cast_fp16")]; + int32 var_33772 = const()[name = string("op_33772"), val = int32(3)]; + int32 var_33797 = const()[name = string("op_33797"), val = int32(1)]; + tensor out_89_axes_0 = const()[name = string("out_89_axes_0"), val = tensor([1])]; + fp16 var_33814_to_fp16 = const()[name = string("op_33814_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_33814_to_fp16, x = inputs_89_cast_fp16)[name = string("out_89_cast_fp16")]; + tensor obj_89_gamma_0_to_fp16 = const()[name = string("obj_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880436352)))]; + tensor obj_89_beta_0_to_fp16 = const()[name = string("obj_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880438976)))]; + fp16 obj_89_epsilon_0_to_fp16 = const()[name = string("obj_89_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_89_cast_fp16 = batch_norm(beta = obj_89_beta_0_to_fp16, epsilon = obj_89_epsilon_0_to_fp16, gamma = obj_89_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = string("obj_89_cast_fp16")]; + string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; + tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; + tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; + int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880441600)))]; + tensor layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883718464)))]; + tensor query_45_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("query_45_cast_fp16")]; + string key_45_pad_type_0 = const()[name = string("key_45_pad_type_0"), val = string("valid")]; + tensor key_45_strides_0 = const()[name = string("key_45_strides_0"), val = tensor([1, 1])]; + tensor key_45_pad_0 = const()[name = string("key_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_45_dilations_0 = const()[name = string("key_45_dilations_0"), val = tensor([1, 1])]; + int32 key_45_groups_0 = const()[name = string("key_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883721088)))]; + tensor key_45_cast_fp16 = conv(dilations = key_45_dilations_0, groups = key_45_groups_0, pad = key_45_pad_0, pad_type = key_45_pad_type_0, strides = key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("key_45_cast_fp16")]; + string value_45_pad_type_0 = const()[name = string("value_45_pad_type_0"), val = string("valid")]; + tensor value_45_strides_0 = const()[name = string("value_45_strides_0"), val = tensor([1, 1])]; + tensor value_45_pad_0 = const()[name = string("value_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_45_dilations_0 = const()[name = string("value_45_dilations_0"), val = tensor([1, 1])]; + int32 value_45_groups_0 = const()[name = string("value_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(886997952)))]; + tensor layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890274816)))]; + tensor value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = value_45_dilations_0, groups = value_45_groups_0, pad = value_45_pad_0, pad_type = value_45_pad_type_0, strides = value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("value_45_cast_fp16")]; + tensor var_33852_begin_0 = const()[name = string("op_33852_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33852_end_0 = const()[name = string("op_33852_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33852_end_mask_0 = const()[name = string("op_33852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33852_cast_fp16 = slice_by_index(begin = var_33852_begin_0, end = var_33852_end_0, end_mask = var_33852_end_mask_0, x = query_45_cast_fp16)[name = string("op_33852_cast_fp16")]; + tensor var_33856_begin_0 = const()[name = string("op_33856_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33856_end_0 = const()[name = string("op_33856_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_33856_end_mask_0 = const()[name = string("op_33856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33856_cast_fp16 = slice_by_index(begin = var_33856_begin_0, end = var_33856_end_0, end_mask = var_33856_end_mask_0, x = query_45_cast_fp16)[name = string("op_33856_cast_fp16")]; + tensor var_33860_begin_0 = const()[name = string("op_33860_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33860_end_0 = const()[name = string("op_33860_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_33860_end_mask_0 = const()[name = string("op_33860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33860_cast_fp16 = slice_by_index(begin = var_33860_begin_0, end = var_33860_end_0, end_mask = var_33860_end_mask_0, x = query_45_cast_fp16)[name = string("op_33860_cast_fp16")]; + tensor var_33864_begin_0 = const()[name = string("op_33864_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33864_end_0 = const()[name = string("op_33864_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_33864_end_mask_0 = const()[name = string("op_33864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33864_cast_fp16 = slice_by_index(begin = var_33864_begin_0, end = var_33864_end_0, end_mask = var_33864_end_mask_0, x = query_45_cast_fp16)[name = string("op_33864_cast_fp16")]; + tensor var_33868_begin_0 = const()[name = string("op_33868_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33868_end_0 = const()[name = string("op_33868_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_33868_end_mask_0 = const()[name = string("op_33868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33868_cast_fp16 = slice_by_index(begin = var_33868_begin_0, end = var_33868_end_0, end_mask = var_33868_end_mask_0, x = query_45_cast_fp16)[name = string("op_33868_cast_fp16")]; + tensor var_33872_begin_0 = const()[name = string("op_33872_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33872_end_0 = const()[name = string("op_33872_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_33872_end_mask_0 = const()[name = string("op_33872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33872_cast_fp16 = slice_by_index(begin = var_33872_begin_0, end = var_33872_end_0, end_mask = var_33872_end_mask_0, x = query_45_cast_fp16)[name = string("op_33872_cast_fp16")]; + tensor var_33876_begin_0 = const()[name = string("op_33876_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33876_end_0 = const()[name = string("op_33876_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_33876_end_mask_0 = const()[name = string("op_33876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33876_cast_fp16 = slice_by_index(begin = var_33876_begin_0, end = var_33876_end_0, end_mask = var_33876_end_mask_0, x = query_45_cast_fp16)[name = string("op_33876_cast_fp16")]; + tensor var_33880_begin_0 = const()[name = string("op_33880_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33880_end_0 = const()[name = string("op_33880_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_33880_end_mask_0 = const()[name = string("op_33880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33880_cast_fp16 = slice_by_index(begin = var_33880_begin_0, end = var_33880_end_0, end_mask = var_33880_end_mask_0, x = query_45_cast_fp16)[name = string("op_33880_cast_fp16")]; + tensor var_33884_begin_0 = const()[name = string("op_33884_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33884_end_0 = const()[name = string("op_33884_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_33884_end_mask_0 = const()[name = string("op_33884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33884_cast_fp16 = slice_by_index(begin = var_33884_begin_0, end = var_33884_end_0, end_mask = var_33884_end_mask_0, x = query_45_cast_fp16)[name = string("op_33884_cast_fp16")]; + tensor var_33888_begin_0 = const()[name = string("op_33888_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33888_end_0 = const()[name = string("op_33888_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_33888_end_mask_0 = const()[name = string("op_33888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33888_cast_fp16 = slice_by_index(begin = var_33888_begin_0, end = var_33888_end_0, end_mask = var_33888_end_mask_0, x = query_45_cast_fp16)[name = string("op_33888_cast_fp16")]; + tensor var_33892_begin_0 = const()[name = string("op_33892_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33892_end_0 = const()[name = string("op_33892_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_33892_end_mask_0 = const()[name = string("op_33892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33892_cast_fp16 = slice_by_index(begin = var_33892_begin_0, end = var_33892_end_0, end_mask = var_33892_end_mask_0, x = query_45_cast_fp16)[name = string("op_33892_cast_fp16")]; + tensor var_33896_begin_0 = const()[name = string("op_33896_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33896_end_0 = const()[name = string("op_33896_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_33896_end_mask_0 = const()[name = string("op_33896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33896_cast_fp16 = slice_by_index(begin = var_33896_begin_0, end = var_33896_end_0, end_mask = var_33896_end_mask_0, x = query_45_cast_fp16)[name = string("op_33896_cast_fp16")]; + tensor var_33900_begin_0 = const()[name = string("op_33900_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33900_end_0 = const()[name = string("op_33900_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_33900_end_mask_0 = const()[name = string("op_33900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33900_cast_fp16 = slice_by_index(begin = var_33900_begin_0, end = var_33900_end_0, end_mask = var_33900_end_mask_0, x = query_45_cast_fp16)[name = string("op_33900_cast_fp16")]; + tensor var_33904_begin_0 = const()[name = string("op_33904_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33904_end_0 = const()[name = string("op_33904_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_33904_end_mask_0 = const()[name = string("op_33904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33904_cast_fp16 = slice_by_index(begin = var_33904_begin_0, end = var_33904_end_0, end_mask = var_33904_end_mask_0, x = query_45_cast_fp16)[name = string("op_33904_cast_fp16")]; + tensor var_33908_begin_0 = const()[name = string("op_33908_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33908_end_0 = const()[name = string("op_33908_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_33908_end_mask_0 = const()[name = string("op_33908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33908_cast_fp16 = slice_by_index(begin = var_33908_begin_0, end = var_33908_end_0, end_mask = var_33908_end_mask_0, x = query_45_cast_fp16)[name = string("op_33908_cast_fp16")]; + tensor var_33912_begin_0 = const()[name = string("op_33912_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33912_end_0 = const()[name = string("op_33912_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_33912_end_mask_0 = const()[name = string("op_33912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33912_cast_fp16 = slice_by_index(begin = var_33912_begin_0, end = var_33912_end_0, end_mask = var_33912_end_mask_0, x = query_45_cast_fp16)[name = string("op_33912_cast_fp16")]; + tensor var_33916_begin_0 = const()[name = string("op_33916_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33916_end_0 = const()[name = string("op_33916_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_33916_end_mask_0 = const()[name = string("op_33916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33916_cast_fp16 = slice_by_index(begin = var_33916_begin_0, end = var_33916_end_0, end_mask = var_33916_end_mask_0, x = query_45_cast_fp16)[name = string("op_33916_cast_fp16")]; + tensor var_33920_begin_0 = const()[name = string("op_33920_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33920_end_0 = const()[name = string("op_33920_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_33920_end_mask_0 = const()[name = string("op_33920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33920_cast_fp16 = slice_by_index(begin = var_33920_begin_0, end = var_33920_end_0, end_mask = var_33920_end_mask_0, x = query_45_cast_fp16)[name = string("op_33920_cast_fp16")]; + tensor var_33924_begin_0 = const()[name = string("op_33924_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33924_end_0 = const()[name = string("op_33924_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_33924_end_mask_0 = const()[name = string("op_33924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33924_cast_fp16 = slice_by_index(begin = var_33924_begin_0, end = var_33924_end_0, end_mask = var_33924_end_mask_0, x = query_45_cast_fp16)[name = string("op_33924_cast_fp16")]; + tensor var_33928_begin_0 = const()[name = string("op_33928_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33928_end_0 = const()[name = string("op_33928_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_33928_end_mask_0 = const()[name = string("op_33928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33928_cast_fp16 = slice_by_index(begin = var_33928_begin_0, end = var_33928_end_0, end_mask = var_33928_end_mask_0, x = query_45_cast_fp16)[name = string("op_33928_cast_fp16")]; + tensor var_33937_begin_0 = const()[name = string("op_33937_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33937_end_0 = const()[name = string("op_33937_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33937_end_mask_0 = const()[name = string("op_33937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33937_cast_fp16 = slice_by_index(begin = var_33937_begin_0, end = var_33937_end_0, end_mask = var_33937_end_mask_0, x = var_33852_cast_fp16)[name = string("op_33937_cast_fp16")]; + tensor var_33944_begin_0 = const()[name = string("op_33944_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33944_end_0 = const()[name = string("op_33944_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33944_end_mask_0 = const()[name = string("op_33944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33944_cast_fp16 = slice_by_index(begin = var_33944_begin_0, end = var_33944_end_0, end_mask = var_33944_end_mask_0, x = var_33852_cast_fp16)[name = string("op_33944_cast_fp16")]; + tensor var_33951_begin_0 = const()[name = string("op_33951_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33951_end_0 = const()[name = string("op_33951_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33951_end_mask_0 = const()[name = string("op_33951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33951_cast_fp16 = slice_by_index(begin = var_33951_begin_0, end = var_33951_end_0, end_mask = var_33951_end_mask_0, x = var_33852_cast_fp16)[name = string("op_33951_cast_fp16")]; + tensor var_33958_begin_0 = const()[name = string("op_33958_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33958_end_0 = const()[name = string("op_33958_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33958_end_mask_0 = const()[name = string("op_33958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33958_cast_fp16 = slice_by_index(begin = var_33958_begin_0, end = var_33958_end_0, end_mask = var_33958_end_mask_0, x = var_33852_cast_fp16)[name = string("op_33958_cast_fp16")]; + tensor var_33965_begin_0 = const()[name = string("op_33965_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33965_end_0 = const()[name = string("op_33965_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33965_end_mask_0 = const()[name = string("op_33965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33965_cast_fp16 = slice_by_index(begin = var_33965_begin_0, end = var_33965_end_0, end_mask = var_33965_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33965_cast_fp16")]; + tensor var_33972_begin_0 = const()[name = string("op_33972_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33972_end_0 = const()[name = string("op_33972_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33972_end_mask_0 = const()[name = string("op_33972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33972_cast_fp16 = slice_by_index(begin = var_33972_begin_0, end = var_33972_end_0, end_mask = var_33972_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33972_cast_fp16")]; + tensor var_33979_begin_0 = const()[name = string("op_33979_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33979_end_0 = const()[name = string("op_33979_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33979_end_mask_0 = const()[name = string("op_33979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33979_cast_fp16 = slice_by_index(begin = var_33979_begin_0, end = var_33979_end_0, end_mask = var_33979_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33979_cast_fp16")]; + tensor var_33986_begin_0 = const()[name = string("op_33986_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33986_end_0 = const()[name = string("op_33986_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33986_end_mask_0 = const()[name = string("op_33986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33986_cast_fp16 = slice_by_index(begin = var_33986_begin_0, end = var_33986_end_0, end_mask = var_33986_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33986_cast_fp16")]; + tensor var_33993_begin_0 = const()[name = string("op_33993_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33993_end_0 = const()[name = string("op_33993_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33993_end_mask_0 = const()[name = string("op_33993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33993_cast_fp16 = slice_by_index(begin = var_33993_begin_0, end = var_33993_end_0, end_mask = var_33993_end_mask_0, x = var_33860_cast_fp16)[name = string("op_33993_cast_fp16")]; + tensor var_34000_begin_0 = const()[name = string("op_34000_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34000_end_0 = const()[name = string("op_34000_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34000_end_mask_0 = const()[name = string("op_34000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34000_cast_fp16 = slice_by_index(begin = var_34000_begin_0, end = var_34000_end_0, end_mask = var_34000_end_mask_0, x = var_33860_cast_fp16)[name = string("op_34000_cast_fp16")]; + tensor var_34007_begin_0 = const()[name = string("op_34007_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34007_end_0 = const()[name = string("op_34007_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34007_end_mask_0 = const()[name = string("op_34007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34007_cast_fp16 = slice_by_index(begin = var_34007_begin_0, end = var_34007_end_0, end_mask = var_34007_end_mask_0, x = var_33860_cast_fp16)[name = string("op_34007_cast_fp16")]; + tensor var_34014_begin_0 = const()[name = string("op_34014_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34014_end_0 = const()[name = string("op_34014_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34014_end_mask_0 = const()[name = string("op_34014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34014_cast_fp16 = slice_by_index(begin = var_34014_begin_0, end = var_34014_end_0, end_mask = var_34014_end_mask_0, x = var_33860_cast_fp16)[name = string("op_34014_cast_fp16")]; + tensor var_34021_begin_0 = const()[name = string("op_34021_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34021_end_0 = const()[name = string("op_34021_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34021_end_mask_0 = const()[name = string("op_34021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34021_cast_fp16 = slice_by_index(begin = var_34021_begin_0, end = var_34021_end_0, end_mask = var_34021_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34021_cast_fp16")]; + tensor var_34028_begin_0 = const()[name = string("op_34028_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34028_end_0 = const()[name = string("op_34028_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34028_end_mask_0 = const()[name = string("op_34028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34028_cast_fp16 = slice_by_index(begin = var_34028_begin_0, end = var_34028_end_0, end_mask = var_34028_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34028_cast_fp16")]; + tensor var_34035_begin_0 = const()[name = string("op_34035_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34035_end_0 = const()[name = string("op_34035_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34035_end_mask_0 = const()[name = string("op_34035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34035_cast_fp16 = slice_by_index(begin = var_34035_begin_0, end = var_34035_end_0, end_mask = var_34035_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34035_cast_fp16")]; + tensor var_34042_begin_0 = const()[name = string("op_34042_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34042_end_0 = const()[name = string("op_34042_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34042_end_mask_0 = const()[name = string("op_34042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34042_cast_fp16 = slice_by_index(begin = var_34042_begin_0, end = var_34042_end_0, end_mask = var_34042_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34042_cast_fp16")]; + tensor var_34049_begin_0 = const()[name = string("op_34049_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34049_end_0 = const()[name = string("op_34049_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34049_end_mask_0 = const()[name = string("op_34049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34049_cast_fp16 = slice_by_index(begin = var_34049_begin_0, end = var_34049_end_0, end_mask = var_34049_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34049_cast_fp16")]; + tensor var_34056_begin_0 = const()[name = string("op_34056_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34056_end_0 = const()[name = string("op_34056_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34056_end_mask_0 = const()[name = string("op_34056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34056_cast_fp16 = slice_by_index(begin = var_34056_begin_0, end = var_34056_end_0, end_mask = var_34056_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34056_cast_fp16")]; + tensor var_34063_begin_0 = const()[name = string("op_34063_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34063_end_0 = const()[name = string("op_34063_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34063_end_mask_0 = const()[name = string("op_34063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34063_cast_fp16 = slice_by_index(begin = var_34063_begin_0, end = var_34063_end_0, end_mask = var_34063_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34063_cast_fp16")]; + tensor var_34070_begin_0 = const()[name = string("op_34070_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34070_end_0 = const()[name = string("op_34070_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34070_end_mask_0 = const()[name = string("op_34070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34070_cast_fp16 = slice_by_index(begin = var_34070_begin_0, end = var_34070_end_0, end_mask = var_34070_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34070_cast_fp16")]; + tensor var_34077_begin_0 = const()[name = string("op_34077_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34077_end_0 = const()[name = string("op_34077_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34077_end_mask_0 = const()[name = string("op_34077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34077_cast_fp16 = slice_by_index(begin = var_34077_begin_0, end = var_34077_end_0, end_mask = var_34077_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34077_cast_fp16")]; + tensor var_34084_begin_0 = const()[name = string("op_34084_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34084_end_0 = const()[name = string("op_34084_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34084_end_mask_0 = const()[name = string("op_34084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34084_cast_fp16 = slice_by_index(begin = var_34084_begin_0, end = var_34084_end_0, end_mask = var_34084_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34084_cast_fp16")]; + tensor var_34091_begin_0 = const()[name = string("op_34091_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34091_end_0 = const()[name = string("op_34091_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34091_end_mask_0 = const()[name = string("op_34091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34091_cast_fp16 = slice_by_index(begin = var_34091_begin_0, end = var_34091_end_0, end_mask = var_34091_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34091_cast_fp16")]; + tensor var_34098_begin_0 = const()[name = string("op_34098_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34098_end_0 = const()[name = string("op_34098_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34098_end_mask_0 = const()[name = string("op_34098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34098_cast_fp16 = slice_by_index(begin = var_34098_begin_0, end = var_34098_end_0, end_mask = var_34098_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34098_cast_fp16")]; + tensor var_34105_begin_0 = const()[name = string("op_34105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34105_end_0 = const()[name = string("op_34105_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34105_end_mask_0 = const()[name = string("op_34105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34105_cast_fp16 = slice_by_index(begin = var_34105_begin_0, end = var_34105_end_0, end_mask = var_34105_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34105_cast_fp16")]; + tensor var_34112_begin_0 = const()[name = string("op_34112_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34112_end_0 = const()[name = string("op_34112_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34112_end_mask_0 = const()[name = string("op_34112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34112_cast_fp16 = slice_by_index(begin = var_34112_begin_0, end = var_34112_end_0, end_mask = var_34112_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34112_cast_fp16")]; + tensor var_34119_begin_0 = const()[name = string("op_34119_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34119_end_0 = const()[name = string("op_34119_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34119_end_mask_0 = const()[name = string("op_34119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34119_cast_fp16 = slice_by_index(begin = var_34119_begin_0, end = var_34119_end_0, end_mask = var_34119_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34119_cast_fp16")]; + tensor var_34126_begin_0 = const()[name = string("op_34126_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34126_end_0 = const()[name = string("op_34126_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34126_end_mask_0 = const()[name = string("op_34126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34126_cast_fp16 = slice_by_index(begin = var_34126_begin_0, end = var_34126_end_0, end_mask = var_34126_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34126_cast_fp16")]; + tensor var_34133_begin_0 = const()[name = string("op_34133_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34133_end_0 = const()[name = string("op_34133_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34133_end_mask_0 = const()[name = string("op_34133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34133_cast_fp16 = slice_by_index(begin = var_34133_begin_0, end = var_34133_end_0, end_mask = var_34133_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34133_cast_fp16")]; + tensor var_34140_begin_0 = const()[name = string("op_34140_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34140_end_0 = const()[name = string("op_34140_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34140_end_mask_0 = const()[name = string("op_34140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34140_cast_fp16 = slice_by_index(begin = var_34140_begin_0, end = var_34140_end_0, end_mask = var_34140_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34140_cast_fp16")]; + tensor var_34147_begin_0 = const()[name = string("op_34147_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34147_end_0 = const()[name = string("op_34147_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34147_end_mask_0 = const()[name = string("op_34147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34147_cast_fp16 = slice_by_index(begin = var_34147_begin_0, end = var_34147_end_0, end_mask = var_34147_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34147_cast_fp16")]; + tensor var_34154_begin_0 = const()[name = string("op_34154_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34154_end_0 = const()[name = string("op_34154_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34154_end_mask_0 = const()[name = string("op_34154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34154_cast_fp16 = slice_by_index(begin = var_34154_begin_0, end = var_34154_end_0, end_mask = var_34154_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34154_cast_fp16")]; + tensor var_34161_begin_0 = const()[name = string("op_34161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34161_end_0 = const()[name = string("op_34161_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34161_end_mask_0 = const()[name = string("op_34161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34161_cast_fp16 = slice_by_index(begin = var_34161_begin_0, end = var_34161_end_0, end_mask = var_34161_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34161_cast_fp16")]; + tensor var_34168_begin_0 = const()[name = string("op_34168_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34168_end_0 = const()[name = string("op_34168_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34168_end_mask_0 = const()[name = string("op_34168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34168_cast_fp16 = slice_by_index(begin = var_34168_begin_0, end = var_34168_end_0, end_mask = var_34168_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34168_cast_fp16")]; + tensor var_34175_begin_0 = const()[name = string("op_34175_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34175_end_0 = const()[name = string("op_34175_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34175_end_mask_0 = const()[name = string("op_34175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34175_cast_fp16 = slice_by_index(begin = var_34175_begin_0, end = var_34175_end_0, end_mask = var_34175_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34175_cast_fp16")]; + tensor var_34182_begin_0 = const()[name = string("op_34182_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34182_end_0 = const()[name = string("op_34182_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34182_end_mask_0 = const()[name = string("op_34182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34182_cast_fp16 = slice_by_index(begin = var_34182_begin_0, end = var_34182_end_0, end_mask = var_34182_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34182_cast_fp16")]; + tensor var_34189_begin_0 = const()[name = string("op_34189_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34189_end_0 = const()[name = string("op_34189_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34189_end_mask_0 = const()[name = string("op_34189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34189_cast_fp16 = slice_by_index(begin = var_34189_begin_0, end = var_34189_end_0, end_mask = var_34189_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34189_cast_fp16")]; + tensor var_34196_begin_0 = const()[name = string("op_34196_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34196_end_0 = const()[name = string("op_34196_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34196_end_mask_0 = const()[name = string("op_34196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34196_cast_fp16 = slice_by_index(begin = var_34196_begin_0, end = var_34196_end_0, end_mask = var_34196_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34196_cast_fp16")]; + tensor var_34203_begin_0 = const()[name = string("op_34203_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34203_end_0 = const()[name = string("op_34203_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34203_end_mask_0 = const()[name = string("op_34203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34203_cast_fp16 = slice_by_index(begin = var_34203_begin_0, end = var_34203_end_0, end_mask = var_34203_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34203_cast_fp16")]; + tensor var_34210_begin_0 = const()[name = string("op_34210_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34210_end_0 = const()[name = string("op_34210_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34210_end_mask_0 = const()[name = string("op_34210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34210_cast_fp16 = slice_by_index(begin = var_34210_begin_0, end = var_34210_end_0, end_mask = var_34210_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34210_cast_fp16")]; + tensor var_34217_begin_0 = const()[name = string("op_34217_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34217_end_0 = const()[name = string("op_34217_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34217_end_mask_0 = const()[name = string("op_34217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34217_cast_fp16 = slice_by_index(begin = var_34217_begin_0, end = var_34217_end_0, end_mask = var_34217_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34217_cast_fp16")]; + tensor var_34224_begin_0 = const()[name = string("op_34224_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34224_end_0 = const()[name = string("op_34224_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34224_end_mask_0 = const()[name = string("op_34224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34224_cast_fp16 = slice_by_index(begin = var_34224_begin_0, end = var_34224_end_0, end_mask = var_34224_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34224_cast_fp16")]; + tensor var_34231_begin_0 = const()[name = string("op_34231_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34231_end_0 = const()[name = string("op_34231_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34231_end_mask_0 = const()[name = string("op_34231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34231_cast_fp16 = slice_by_index(begin = var_34231_begin_0, end = var_34231_end_0, end_mask = var_34231_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34231_cast_fp16")]; + tensor var_34238_begin_0 = const()[name = string("op_34238_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34238_end_0 = const()[name = string("op_34238_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34238_end_mask_0 = const()[name = string("op_34238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34238_cast_fp16 = slice_by_index(begin = var_34238_begin_0, end = var_34238_end_0, end_mask = var_34238_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34238_cast_fp16")]; + tensor var_34245_begin_0 = const()[name = string("op_34245_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34245_end_0 = const()[name = string("op_34245_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34245_end_mask_0 = const()[name = string("op_34245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34245_cast_fp16 = slice_by_index(begin = var_34245_begin_0, end = var_34245_end_0, end_mask = var_34245_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34245_cast_fp16")]; + tensor var_34252_begin_0 = const()[name = string("op_34252_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34252_end_0 = const()[name = string("op_34252_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34252_end_mask_0 = const()[name = string("op_34252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34252_cast_fp16 = slice_by_index(begin = var_34252_begin_0, end = var_34252_end_0, end_mask = var_34252_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34252_cast_fp16")]; + tensor var_34259_begin_0 = const()[name = string("op_34259_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34259_end_0 = const()[name = string("op_34259_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34259_end_mask_0 = const()[name = string("op_34259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34259_cast_fp16 = slice_by_index(begin = var_34259_begin_0, end = var_34259_end_0, end_mask = var_34259_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34259_cast_fp16")]; + tensor var_34266_begin_0 = const()[name = string("op_34266_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34266_end_0 = const()[name = string("op_34266_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34266_end_mask_0 = const()[name = string("op_34266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34266_cast_fp16 = slice_by_index(begin = var_34266_begin_0, end = var_34266_end_0, end_mask = var_34266_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34266_cast_fp16")]; + tensor var_34273_begin_0 = const()[name = string("op_34273_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34273_end_0 = const()[name = string("op_34273_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34273_end_mask_0 = const()[name = string("op_34273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34273_cast_fp16 = slice_by_index(begin = var_34273_begin_0, end = var_34273_end_0, end_mask = var_34273_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34273_cast_fp16")]; + tensor var_34280_begin_0 = const()[name = string("op_34280_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34280_end_0 = const()[name = string("op_34280_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34280_end_mask_0 = const()[name = string("op_34280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34280_cast_fp16 = slice_by_index(begin = var_34280_begin_0, end = var_34280_end_0, end_mask = var_34280_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34280_cast_fp16")]; + tensor var_34287_begin_0 = const()[name = string("op_34287_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34287_end_0 = const()[name = string("op_34287_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34287_end_mask_0 = const()[name = string("op_34287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34287_cast_fp16 = slice_by_index(begin = var_34287_begin_0, end = var_34287_end_0, end_mask = var_34287_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34287_cast_fp16")]; + tensor var_34294_begin_0 = const()[name = string("op_34294_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34294_end_0 = const()[name = string("op_34294_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34294_end_mask_0 = const()[name = string("op_34294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34294_cast_fp16 = slice_by_index(begin = var_34294_begin_0, end = var_34294_end_0, end_mask = var_34294_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34294_cast_fp16")]; + tensor var_34301_begin_0 = const()[name = string("op_34301_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34301_end_0 = const()[name = string("op_34301_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34301_end_mask_0 = const()[name = string("op_34301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34301_cast_fp16 = slice_by_index(begin = var_34301_begin_0, end = var_34301_end_0, end_mask = var_34301_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34301_cast_fp16")]; + tensor var_34308_begin_0 = const()[name = string("op_34308_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34308_end_0 = const()[name = string("op_34308_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34308_end_mask_0 = const()[name = string("op_34308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34308_cast_fp16 = slice_by_index(begin = var_34308_begin_0, end = var_34308_end_0, end_mask = var_34308_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34308_cast_fp16")]; + tensor var_34315_begin_0 = const()[name = string("op_34315_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34315_end_0 = const()[name = string("op_34315_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34315_end_mask_0 = const()[name = string("op_34315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34315_cast_fp16 = slice_by_index(begin = var_34315_begin_0, end = var_34315_end_0, end_mask = var_34315_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34315_cast_fp16")]; + tensor var_34322_begin_0 = const()[name = string("op_34322_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34322_end_0 = const()[name = string("op_34322_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34322_end_mask_0 = const()[name = string("op_34322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34322_cast_fp16 = slice_by_index(begin = var_34322_begin_0, end = var_34322_end_0, end_mask = var_34322_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34322_cast_fp16")]; + tensor var_34329_begin_0 = const()[name = string("op_34329_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34329_end_0 = const()[name = string("op_34329_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34329_end_mask_0 = const()[name = string("op_34329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34329_cast_fp16 = slice_by_index(begin = var_34329_begin_0, end = var_34329_end_0, end_mask = var_34329_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34329_cast_fp16")]; + tensor var_34336_begin_0 = const()[name = string("op_34336_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34336_end_0 = const()[name = string("op_34336_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34336_end_mask_0 = const()[name = string("op_34336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34336_cast_fp16 = slice_by_index(begin = var_34336_begin_0, end = var_34336_end_0, end_mask = var_34336_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34336_cast_fp16")]; + tensor var_34343_begin_0 = const()[name = string("op_34343_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34343_end_0 = const()[name = string("op_34343_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34343_end_mask_0 = const()[name = string("op_34343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34343_cast_fp16 = slice_by_index(begin = var_34343_begin_0, end = var_34343_end_0, end_mask = var_34343_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34343_cast_fp16")]; + tensor var_34350_begin_0 = const()[name = string("op_34350_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34350_end_0 = const()[name = string("op_34350_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34350_end_mask_0 = const()[name = string("op_34350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34350_cast_fp16 = slice_by_index(begin = var_34350_begin_0, end = var_34350_end_0, end_mask = var_34350_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34350_cast_fp16")]; + tensor var_34357_begin_0 = const()[name = string("op_34357_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34357_end_0 = const()[name = string("op_34357_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34357_end_mask_0 = const()[name = string("op_34357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34357_cast_fp16 = slice_by_index(begin = var_34357_begin_0, end = var_34357_end_0, end_mask = var_34357_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34357_cast_fp16")]; + tensor var_34364_begin_0 = const()[name = string("op_34364_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34364_end_0 = const()[name = string("op_34364_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34364_end_mask_0 = const()[name = string("op_34364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34364_cast_fp16 = slice_by_index(begin = var_34364_begin_0, end = var_34364_end_0, end_mask = var_34364_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34364_cast_fp16")]; + tensor var_34371_begin_0 = const()[name = string("op_34371_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34371_end_0 = const()[name = string("op_34371_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34371_end_mask_0 = const()[name = string("op_34371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34371_cast_fp16 = slice_by_index(begin = var_34371_begin_0, end = var_34371_end_0, end_mask = var_34371_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34371_cast_fp16")]; + tensor var_34378_begin_0 = const()[name = string("op_34378_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34378_end_0 = const()[name = string("op_34378_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34378_end_mask_0 = const()[name = string("op_34378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34378_cast_fp16 = slice_by_index(begin = var_34378_begin_0, end = var_34378_end_0, end_mask = var_34378_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34378_cast_fp16")]; + tensor var_34385_begin_0 = const()[name = string("op_34385_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34385_end_0 = const()[name = string("op_34385_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34385_end_mask_0 = const()[name = string("op_34385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34385_cast_fp16 = slice_by_index(begin = var_34385_begin_0, end = var_34385_end_0, end_mask = var_34385_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34385_cast_fp16")]; + tensor var_34392_begin_0 = const()[name = string("op_34392_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34392_end_0 = const()[name = string("op_34392_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34392_end_mask_0 = const()[name = string("op_34392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34392_cast_fp16 = slice_by_index(begin = var_34392_begin_0, end = var_34392_end_0, end_mask = var_34392_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34392_cast_fp16")]; + tensor var_34399_begin_0 = const()[name = string("op_34399_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34399_end_0 = const()[name = string("op_34399_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34399_end_mask_0 = const()[name = string("op_34399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34399_cast_fp16 = slice_by_index(begin = var_34399_begin_0, end = var_34399_end_0, end_mask = var_34399_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34399_cast_fp16")]; + tensor var_34406_begin_0 = const()[name = string("op_34406_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34406_end_0 = const()[name = string("op_34406_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34406_end_mask_0 = const()[name = string("op_34406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34406_cast_fp16 = slice_by_index(begin = var_34406_begin_0, end = var_34406_end_0, end_mask = var_34406_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34406_cast_fp16")]; + tensor var_34413_begin_0 = const()[name = string("op_34413_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34413_end_0 = const()[name = string("op_34413_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34413_end_mask_0 = const()[name = string("op_34413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34413_cast_fp16 = slice_by_index(begin = var_34413_begin_0, end = var_34413_end_0, end_mask = var_34413_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34413_cast_fp16")]; + tensor var_34420_begin_0 = const()[name = string("op_34420_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34420_end_0 = const()[name = string("op_34420_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34420_end_mask_0 = const()[name = string("op_34420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34420_cast_fp16 = slice_by_index(begin = var_34420_begin_0, end = var_34420_end_0, end_mask = var_34420_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34420_cast_fp16")]; + tensor var_34427_begin_0 = const()[name = string("op_34427_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34427_end_0 = const()[name = string("op_34427_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34427_end_mask_0 = const()[name = string("op_34427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34427_cast_fp16 = slice_by_index(begin = var_34427_begin_0, end = var_34427_end_0, end_mask = var_34427_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34427_cast_fp16")]; + tensor var_34434_begin_0 = const()[name = string("op_34434_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34434_end_0 = const()[name = string("op_34434_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34434_end_mask_0 = const()[name = string("op_34434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34434_cast_fp16 = slice_by_index(begin = var_34434_begin_0, end = var_34434_end_0, end_mask = var_34434_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34434_cast_fp16")]; + tensor var_34441_begin_0 = const()[name = string("op_34441_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34441_end_0 = const()[name = string("op_34441_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34441_end_mask_0 = const()[name = string("op_34441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34441_cast_fp16 = slice_by_index(begin = var_34441_begin_0, end = var_34441_end_0, end_mask = var_34441_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34441_cast_fp16")]; + tensor var_34448_begin_0 = const()[name = string("op_34448_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34448_end_0 = const()[name = string("op_34448_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34448_end_mask_0 = const()[name = string("op_34448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34448_cast_fp16 = slice_by_index(begin = var_34448_begin_0, end = var_34448_end_0, end_mask = var_34448_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34448_cast_fp16")]; + tensor var_34455_begin_0 = const()[name = string("op_34455_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34455_end_0 = const()[name = string("op_34455_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34455_end_mask_0 = const()[name = string("op_34455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34455_cast_fp16 = slice_by_index(begin = var_34455_begin_0, end = var_34455_end_0, end_mask = var_34455_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34455_cast_fp16")]; + tensor var_34462_begin_0 = const()[name = string("op_34462_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34462_end_0 = const()[name = string("op_34462_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34462_end_mask_0 = const()[name = string("op_34462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34462_cast_fp16 = slice_by_index(begin = var_34462_begin_0, end = var_34462_end_0, end_mask = var_34462_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34462_cast_fp16")]; + tensor var_34469_begin_0 = const()[name = string("op_34469_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34469_end_0 = const()[name = string("op_34469_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34469_end_mask_0 = const()[name = string("op_34469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34469_cast_fp16 = slice_by_index(begin = var_34469_begin_0, end = var_34469_end_0, end_mask = var_34469_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34469_cast_fp16")]; + tensor var_34476_begin_0 = const()[name = string("op_34476_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34476_end_0 = const()[name = string("op_34476_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34476_end_mask_0 = const()[name = string("op_34476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34476_cast_fp16 = slice_by_index(begin = var_34476_begin_0, end = var_34476_end_0, end_mask = var_34476_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34476_cast_fp16")]; + tensor var_34483_begin_0 = const()[name = string("op_34483_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34483_end_0 = const()[name = string("op_34483_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34483_end_mask_0 = const()[name = string("op_34483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34483_cast_fp16 = slice_by_index(begin = var_34483_begin_0, end = var_34483_end_0, end_mask = var_34483_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34483_cast_fp16")]; + tensor var_34490_begin_0 = const()[name = string("op_34490_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34490_end_0 = const()[name = string("op_34490_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34490_end_mask_0 = const()[name = string("op_34490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34490_cast_fp16 = slice_by_index(begin = var_34490_begin_0, end = var_34490_end_0, end_mask = var_34490_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34490_cast_fp16")]; + tensor k_45_perm_0 = const()[name = string("k_45_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_34495_begin_0 = const()[name = string("op_34495_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34495_end_0 = const()[name = string("op_34495_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_34495_end_mask_0 = const()[name = string("op_34495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_45_cast_fp16 = transpose(perm = k_45_perm_0, x = key_45_cast_fp16)[name = string("transpose_9")]; + tensor var_34495_cast_fp16 = slice_by_index(begin = var_34495_begin_0, end = var_34495_end_0, end_mask = var_34495_end_mask_0, x = k_45_cast_fp16)[name = string("op_34495_cast_fp16")]; + tensor var_34499_begin_0 = const()[name = string("op_34499_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_34499_end_0 = const()[name = string("op_34499_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_34499_end_mask_0 = const()[name = string("op_34499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34499_cast_fp16 = slice_by_index(begin = var_34499_begin_0, end = var_34499_end_0, end_mask = var_34499_end_mask_0, x = k_45_cast_fp16)[name = string("op_34499_cast_fp16")]; + tensor var_34503_begin_0 = const()[name = string("op_34503_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_34503_end_0 = const()[name = string("op_34503_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_34503_end_mask_0 = const()[name = string("op_34503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34503_cast_fp16 = slice_by_index(begin = var_34503_begin_0, end = var_34503_end_0, end_mask = var_34503_end_mask_0, x = k_45_cast_fp16)[name = string("op_34503_cast_fp16")]; + tensor var_34507_begin_0 = const()[name = string("op_34507_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_34507_end_0 = const()[name = string("op_34507_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_34507_end_mask_0 = const()[name = string("op_34507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34507_cast_fp16 = slice_by_index(begin = var_34507_begin_0, end = var_34507_end_0, end_mask = var_34507_end_mask_0, x = k_45_cast_fp16)[name = string("op_34507_cast_fp16")]; + tensor var_34511_begin_0 = const()[name = string("op_34511_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_34511_end_0 = const()[name = string("op_34511_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_34511_end_mask_0 = const()[name = string("op_34511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34511_cast_fp16 = slice_by_index(begin = var_34511_begin_0, end = var_34511_end_0, end_mask = var_34511_end_mask_0, x = k_45_cast_fp16)[name = string("op_34511_cast_fp16")]; + tensor var_34515_begin_0 = const()[name = string("op_34515_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_34515_end_0 = const()[name = string("op_34515_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_34515_end_mask_0 = const()[name = string("op_34515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34515_cast_fp16 = slice_by_index(begin = var_34515_begin_0, end = var_34515_end_0, end_mask = var_34515_end_mask_0, x = k_45_cast_fp16)[name = string("op_34515_cast_fp16")]; + tensor var_34519_begin_0 = const()[name = string("op_34519_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_34519_end_0 = const()[name = string("op_34519_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_34519_end_mask_0 = const()[name = string("op_34519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34519_cast_fp16 = slice_by_index(begin = var_34519_begin_0, end = var_34519_end_0, end_mask = var_34519_end_mask_0, x = k_45_cast_fp16)[name = string("op_34519_cast_fp16")]; + tensor var_34523_begin_0 = const()[name = string("op_34523_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_34523_end_0 = const()[name = string("op_34523_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_34523_end_mask_0 = const()[name = string("op_34523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34523_cast_fp16 = slice_by_index(begin = var_34523_begin_0, end = var_34523_end_0, end_mask = var_34523_end_mask_0, x = k_45_cast_fp16)[name = string("op_34523_cast_fp16")]; + tensor var_34527_begin_0 = const()[name = string("op_34527_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_34527_end_0 = const()[name = string("op_34527_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_34527_end_mask_0 = const()[name = string("op_34527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34527_cast_fp16 = slice_by_index(begin = var_34527_begin_0, end = var_34527_end_0, end_mask = var_34527_end_mask_0, x = k_45_cast_fp16)[name = string("op_34527_cast_fp16")]; + tensor var_34531_begin_0 = const()[name = string("op_34531_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_34531_end_0 = const()[name = string("op_34531_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_34531_end_mask_0 = const()[name = string("op_34531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34531_cast_fp16 = slice_by_index(begin = var_34531_begin_0, end = var_34531_end_0, end_mask = var_34531_end_mask_0, x = k_45_cast_fp16)[name = string("op_34531_cast_fp16")]; + tensor var_34535_begin_0 = const()[name = string("op_34535_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_34535_end_0 = const()[name = string("op_34535_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_34535_end_mask_0 = const()[name = string("op_34535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34535_cast_fp16 = slice_by_index(begin = var_34535_begin_0, end = var_34535_end_0, end_mask = var_34535_end_mask_0, x = k_45_cast_fp16)[name = string("op_34535_cast_fp16")]; + tensor var_34539_begin_0 = const()[name = string("op_34539_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_34539_end_0 = const()[name = string("op_34539_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_34539_end_mask_0 = const()[name = string("op_34539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34539_cast_fp16 = slice_by_index(begin = var_34539_begin_0, end = var_34539_end_0, end_mask = var_34539_end_mask_0, x = k_45_cast_fp16)[name = string("op_34539_cast_fp16")]; + tensor var_34543_begin_0 = const()[name = string("op_34543_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_34543_end_0 = const()[name = string("op_34543_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_34543_end_mask_0 = const()[name = string("op_34543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34543_cast_fp16 = slice_by_index(begin = var_34543_begin_0, end = var_34543_end_0, end_mask = var_34543_end_mask_0, x = k_45_cast_fp16)[name = string("op_34543_cast_fp16")]; + tensor var_34547_begin_0 = const()[name = string("op_34547_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_34547_end_0 = const()[name = string("op_34547_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_34547_end_mask_0 = const()[name = string("op_34547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34547_cast_fp16 = slice_by_index(begin = var_34547_begin_0, end = var_34547_end_0, end_mask = var_34547_end_mask_0, x = k_45_cast_fp16)[name = string("op_34547_cast_fp16")]; + tensor var_34551_begin_0 = const()[name = string("op_34551_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_34551_end_0 = const()[name = string("op_34551_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_34551_end_mask_0 = const()[name = string("op_34551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34551_cast_fp16 = slice_by_index(begin = var_34551_begin_0, end = var_34551_end_0, end_mask = var_34551_end_mask_0, x = k_45_cast_fp16)[name = string("op_34551_cast_fp16")]; + tensor var_34555_begin_0 = const()[name = string("op_34555_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_34555_end_0 = const()[name = string("op_34555_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_34555_end_mask_0 = const()[name = string("op_34555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34555_cast_fp16 = slice_by_index(begin = var_34555_begin_0, end = var_34555_end_0, end_mask = var_34555_end_mask_0, x = k_45_cast_fp16)[name = string("op_34555_cast_fp16")]; + tensor var_34559_begin_0 = const()[name = string("op_34559_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_34559_end_0 = const()[name = string("op_34559_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_34559_end_mask_0 = const()[name = string("op_34559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34559_cast_fp16 = slice_by_index(begin = var_34559_begin_0, end = var_34559_end_0, end_mask = var_34559_end_mask_0, x = k_45_cast_fp16)[name = string("op_34559_cast_fp16")]; + tensor var_34563_begin_0 = const()[name = string("op_34563_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_34563_end_0 = const()[name = string("op_34563_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_34563_end_mask_0 = const()[name = string("op_34563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34563_cast_fp16 = slice_by_index(begin = var_34563_begin_0, end = var_34563_end_0, end_mask = var_34563_end_mask_0, x = k_45_cast_fp16)[name = string("op_34563_cast_fp16")]; + tensor var_34567_begin_0 = const()[name = string("op_34567_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_34567_end_0 = const()[name = string("op_34567_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_34567_end_mask_0 = const()[name = string("op_34567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34567_cast_fp16 = slice_by_index(begin = var_34567_begin_0, end = var_34567_end_0, end_mask = var_34567_end_mask_0, x = k_45_cast_fp16)[name = string("op_34567_cast_fp16")]; + tensor var_34571_begin_0 = const()[name = string("op_34571_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_34571_end_0 = const()[name = string("op_34571_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_34571_end_mask_0 = const()[name = string("op_34571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34571_cast_fp16 = slice_by_index(begin = var_34571_begin_0, end = var_34571_end_0, end_mask = var_34571_end_mask_0, x = k_45_cast_fp16)[name = string("op_34571_cast_fp16")]; + tensor var_34573_begin_0 = const()[name = string("op_34573_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34573_end_0 = const()[name = string("op_34573_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34573_end_mask_0 = const()[name = string("op_34573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34573_cast_fp16 = slice_by_index(begin = var_34573_begin_0, end = var_34573_end_0, end_mask = var_34573_end_mask_0, x = value_45_cast_fp16)[name = string("op_34573_cast_fp16")]; + tensor var_34577_begin_0 = const()[name = string("op_34577_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34577_end_0 = const()[name = string("op_34577_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_34577_end_mask_0 = const()[name = string("op_34577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34577_cast_fp16 = slice_by_index(begin = var_34577_begin_0, end = var_34577_end_0, end_mask = var_34577_end_mask_0, x = value_45_cast_fp16)[name = string("op_34577_cast_fp16")]; + tensor var_34581_begin_0 = const()[name = string("op_34581_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34581_end_0 = const()[name = string("op_34581_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_34581_end_mask_0 = const()[name = string("op_34581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34581_cast_fp16 = slice_by_index(begin = var_34581_begin_0, end = var_34581_end_0, end_mask = var_34581_end_mask_0, x = value_45_cast_fp16)[name = string("op_34581_cast_fp16")]; + tensor var_34585_begin_0 = const()[name = string("op_34585_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34585_end_0 = const()[name = string("op_34585_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_34585_end_mask_0 = const()[name = string("op_34585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34585_cast_fp16 = slice_by_index(begin = var_34585_begin_0, end = var_34585_end_0, end_mask = var_34585_end_mask_0, x = value_45_cast_fp16)[name = string("op_34585_cast_fp16")]; + tensor var_34589_begin_0 = const()[name = string("op_34589_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34589_end_0 = const()[name = string("op_34589_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_34589_end_mask_0 = const()[name = string("op_34589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34589_cast_fp16 = slice_by_index(begin = var_34589_begin_0, end = var_34589_end_0, end_mask = var_34589_end_mask_0, x = value_45_cast_fp16)[name = string("op_34589_cast_fp16")]; + tensor var_34593_begin_0 = const()[name = string("op_34593_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34593_end_0 = const()[name = string("op_34593_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_34593_end_mask_0 = const()[name = string("op_34593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34593_cast_fp16 = slice_by_index(begin = var_34593_begin_0, end = var_34593_end_0, end_mask = var_34593_end_mask_0, x = value_45_cast_fp16)[name = string("op_34593_cast_fp16")]; + tensor var_34597_begin_0 = const()[name = string("op_34597_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34597_end_0 = const()[name = string("op_34597_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_34597_end_mask_0 = const()[name = string("op_34597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34597_cast_fp16 = slice_by_index(begin = var_34597_begin_0, end = var_34597_end_0, end_mask = var_34597_end_mask_0, x = value_45_cast_fp16)[name = string("op_34597_cast_fp16")]; + tensor var_34601_begin_0 = const()[name = string("op_34601_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34601_end_0 = const()[name = string("op_34601_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_34601_end_mask_0 = const()[name = string("op_34601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34601_cast_fp16 = slice_by_index(begin = var_34601_begin_0, end = var_34601_end_0, end_mask = var_34601_end_mask_0, x = value_45_cast_fp16)[name = string("op_34601_cast_fp16")]; + tensor var_34605_begin_0 = const()[name = string("op_34605_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34605_end_0 = const()[name = string("op_34605_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_34605_end_mask_0 = const()[name = string("op_34605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34605_cast_fp16 = slice_by_index(begin = var_34605_begin_0, end = var_34605_end_0, end_mask = var_34605_end_mask_0, x = value_45_cast_fp16)[name = string("op_34605_cast_fp16")]; + tensor var_34609_begin_0 = const()[name = string("op_34609_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34609_end_0 = const()[name = string("op_34609_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_34609_end_mask_0 = const()[name = string("op_34609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34609_cast_fp16 = slice_by_index(begin = var_34609_begin_0, end = var_34609_end_0, end_mask = var_34609_end_mask_0, x = value_45_cast_fp16)[name = string("op_34609_cast_fp16")]; + tensor var_34613_begin_0 = const()[name = string("op_34613_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34613_end_0 = const()[name = string("op_34613_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_34613_end_mask_0 = const()[name = string("op_34613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34613_cast_fp16 = slice_by_index(begin = var_34613_begin_0, end = var_34613_end_0, end_mask = var_34613_end_mask_0, x = value_45_cast_fp16)[name = string("op_34613_cast_fp16")]; + tensor var_34617_begin_0 = const()[name = string("op_34617_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34617_end_0 = const()[name = string("op_34617_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_34617_end_mask_0 = const()[name = string("op_34617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34617_cast_fp16 = slice_by_index(begin = var_34617_begin_0, end = var_34617_end_0, end_mask = var_34617_end_mask_0, x = value_45_cast_fp16)[name = string("op_34617_cast_fp16")]; + tensor var_34621_begin_0 = const()[name = string("op_34621_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34621_end_0 = const()[name = string("op_34621_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_34621_end_mask_0 = const()[name = string("op_34621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34621_cast_fp16 = slice_by_index(begin = var_34621_begin_0, end = var_34621_end_0, end_mask = var_34621_end_mask_0, x = value_45_cast_fp16)[name = string("op_34621_cast_fp16")]; + tensor var_34625_begin_0 = const()[name = string("op_34625_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34625_end_0 = const()[name = string("op_34625_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_34625_end_mask_0 = const()[name = string("op_34625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34625_cast_fp16 = slice_by_index(begin = var_34625_begin_0, end = var_34625_end_0, end_mask = var_34625_end_mask_0, x = value_45_cast_fp16)[name = string("op_34625_cast_fp16")]; + tensor var_34629_begin_0 = const()[name = string("op_34629_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34629_end_0 = const()[name = string("op_34629_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_34629_end_mask_0 = const()[name = string("op_34629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34629_cast_fp16 = slice_by_index(begin = var_34629_begin_0, end = var_34629_end_0, end_mask = var_34629_end_mask_0, x = value_45_cast_fp16)[name = string("op_34629_cast_fp16")]; + tensor var_34633_begin_0 = const()[name = string("op_34633_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34633_end_0 = const()[name = string("op_34633_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_34633_end_mask_0 = const()[name = string("op_34633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34633_cast_fp16 = slice_by_index(begin = var_34633_begin_0, end = var_34633_end_0, end_mask = var_34633_end_mask_0, x = value_45_cast_fp16)[name = string("op_34633_cast_fp16")]; + tensor var_34637_begin_0 = const()[name = string("op_34637_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34637_end_0 = const()[name = string("op_34637_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_34637_end_mask_0 = const()[name = string("op_34637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34637_cast_fp16 = slice_by_index(begin = var_34637_begin_0, end = var_34637_end_0, end_mask = var_34637_end_mask_0, x = value_45_cast_fp16)[name = string("op_34637_cast_fp16")]; + tensor var_34641_begin_0 = const()[name = string("op_34641_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34641_end_0 = const()[name = string("op_34641_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_34641_end_mask_0 = const()[name = string("op_34641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34641_cast_fp16 = slice_by_index(begin = var_34641_begin_0, end = var_34641_end_0, end_mask = var_34641_end_mask_0, x = value_45_cast_fp16)[name = string("op_34641_cast_fp16")]; + tensor var_34645_begin_0 = const()[name = string("op_34645_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34645_end_0 = const()[name = string("op_34645_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_34645_end_mask_0 = const()[name = string("op_34645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34645_cast_fp16 = slice_by_index(begin = var_34645_begin_0, end = var_34645_end_0, end_mask = var_34645_end_mask_0, x = value_45_cast_fp16)[name = string("op_34645_cast_fp16")]; + tensor var_34649_begin_0 = const()[name = string("op_34649_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34649_end_0 = const()[name = string("op_34649_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_34649_end_mask_0 = const()[name = string("op_34649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34649_cast_fp16 = slice_by_index(begin = var_34649_begin_0, end = var_34649_end_0, end_mask = var_34649_end_mask_0, x = value_45_cast_fp16)[name = string("op_34649_cast_fp16")]; + string _SplitHeadsQ__mh_w_3521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3521_equation_0, values = (var_34495_cast_fp16, var_33937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3521_cast_fp16")]; + string _SplitHeadsQ__mh_w_3523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3523_equation_0, values = (var_34495_cast_fp16, var_33944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3523_cast_fp16")]; + string _SplitHeadsQ__mh_w_3525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3525_equation_0, values = (var_34495_cast_fp16, var_33951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3525_cast_fp16")]; + string _SplitHeadsQ__mh_w_3527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3527_equation_0, values = (var_34495_cast_fp16, var_33958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3527_cast_fp16")]; + string _SplitHeadsQ__mh_w_3529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3529_equation_0, values = (var_34499_cast_fp16, var_33965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3529_cast_fp16")]; + string _SplitHeadsQ__mh_w_3531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3531_equation_0, values = (var_34499_cast_fp16, var_33972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3531_cast_fp16")]; + string _SplitHeadsQ__mh_w_3533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3533_equation_0, values = (var_34499_cast_fp16, var_33979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3533_cast_fp16")]; + string _SplitHeadsQ__mh_w_3535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3535_equation_0, values = (var_34499_cast_fp16, var_33986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3535_cast_fp16")]; + string _SplitHeadsQ__mh_w_3537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3537_equation_0, values = (var_34503_cast_fp16, var_33993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3537_cast_fp16")]; + string _SplitHeadsQ__mh_w_3539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3539_equation_0, values = (var_34503_cast_fp16, var_34000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3539_cast_fp16")]; + string _SplitHeadsQ__mh_w_3541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3541_equation_0, values = (var_34503_cast_fp16, var_34007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3541_cast_fp16")]; + string _SplitHeadsQ__mh_w_3543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3543_equation_0, values = (var_34503_cast_fp16, var_34014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3543_cast_fp16")]; + string _SplitHeadsQ__mh_w_3545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3545_equation_0, values = (var_34507_cast_fp16, var_34021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3545_cast_fp16")]; + string _SplitHeadsQ__mh_w_3547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3547_equation_0, values = (var_34507_cast_fp16, var_34028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3547_cast_fp16")]; + string _SplitHeadsQ__mh_w_3549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3549_equation_0, values = (var_34507_cast_fp16, var_34035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3549_cast_fp16")]; + string _SplitHeadsQ__mh_w_3551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3551_equation_0, values = (var_34507_cast_fp16, var_34042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3551_cast_fp16")]; + string _SplitHeadsQ__mh_w_3553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3553_equation_0, values = (var_34511_cast_fp16, var_34049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3553_cast_fp16")]; + string _SplitHeadsQ__mh_w_3555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3555_equation_0, values = (var_34511_cast_fp16, var_34056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3555_cast_fp16")]; + string _SplitHeadsQ__mh_w_3557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3557_equation_0, values = (var_34511_cast_fp16, var_34063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3557_cast_fp16")]; + string _SplitHeadsQ__mh_w_3559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3559_equation_0, values = (var_34511_cast_fp16, var_34070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3559_cast_fp16")]; + string _SplitHeadsQ__mh_w_3561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3561_equation_0, values = (var_34515_cast_fp16, var_34077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3561_cast_fp16")]; + string _SplitHeadsQ__mh_w_3563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3563_equation_0, values = (var_34515_cast_fp16, var_34084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3563_cast_fp16")]; + string _SplitHeadsQ__mh_w_3565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3565_equation_0, values = (var_34515_cast_fp16, var_34091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3565_cast_fp16")]; + string _SplitHeadsQ__mh_w_3567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3567_equation_0, values = (var_34515_cast_fp16, var_34098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3567_cast_fp16")]; + string _SplitHeadsQ__mh_w_3569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3569_equation_0, values = (var_34519_cast_fp16, var_34105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3569_cast_fp16")]; + string _SplitHeadsQ__mh_w_3571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3571_equation_0, values = (var_34519_cast_fp16, var_34112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3571_cast_fp16")]; + string _SplitHeadsQ__mh_w_3573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3573_equation_0, values = (var_34519_cast_fp16, var_34119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3573_cast_fp16")]; + string _SplitHeadsQ__mh_w_3575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3575_equation_0, values = (var_34519_cast_fp16, var_34126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3575_cast_fp16")]; + string _SplitHeadsQ__mh_w_3577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3577_equation_0, values = (var_34523_cast_fp16, var_34133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3577_cast_fp16")]; + string _SplitHeadsQ__mh_w_3579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3579_equation_0, values = (var_34523_cast_fp16, var_34140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3579_cast_fp16")]; + string _SplitHeadsQ__mh_w_3581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3581_equation_0, values = (var_34523_cast_fp16, var_34147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3581_cast_fp16")]; + string _SplitHeadsQ__mh_w_3583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3583_equation_0, values = (var_34523_cast_fp16, var_34154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3583_cast_fp16")]; + string _SplitHeadsQ__mh_w_3585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3585_equation_0, values = (var_34527_cast_fp16, var_34161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3585_cast_fp16")]; + string _SplitHeadsQ__mh_w_3587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3587_equation_0, values = (var_34527_cast_fp16, var_34168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3587_cast_fp16")]; + string _SplitHeadsQ__mh_w_3589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3589_equation_0, values = (var_34527_cast_fp16, var_34175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3589_cast_fp16")]; + string _SplitHeadsQ__mh_w_3591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3591_equation_0, values = (var_34527_cast_fp16, var_34182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3591_cast_fp16")]; + string _SplitHeadsQ__mh_w_3593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3593_equation_0, values = (var_34531_cast_fp16, var_34189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3593_cast_fp16")]; + string _SplitHeadsQ__mh_w_3595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3595_equation_0, values = (var_34531_cast_fp16, var_34196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3595_cast_fp16")]; + string _SplitHeadsQ__mh_w_3597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3597_equation_0, values = (var_34531_cast_fp16, var_34203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3597_cast_fp16")]; + string _SplitHeadsQ__mh_w_3599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3599_equation_0, values = (var_34531_cast_fp16, var_34210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3599_cast_fp16")]; + string _SplitHeadsQ__mh_w_3601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3601_equation_0, values = (var_34535_cast_fp16, var_34217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3601_cast_fp16")]; + string _SplitHeadsQ__mh_w_3603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3603_equation_0, values = (var_34535_cast_fp16, var_34224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3603_cast_fp16")]; + string _SplitHeadsQ__mh_w_3605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3605_equation_0, values = (var_34535_cast_fp16, var_34231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3605_cast_fp16")]; + string _SplitHeadsQ__mh_w_3607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3607_equation_0, values = (var_34535_cast_fp16, var_34238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3607_cast_fp16")]; + string _SplitHeadsQ__mh_w_3609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3609_equation_0, values = (var_34539_cast_fp16, var_34245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3609_cast_fp16")]; + string _SplitHeadsQ__mh_w_3611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3611_equation_0, values = (var_34539_cast_fp16, var_34252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3611_cast_fp16")]; + string _SplitHeadsQ__mh_w_3613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3613_equation_0, values = (var_34539_cast_fp16, var_34259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3613_cast_fp16")]; + string _SplitHeadsQ__mh_w_3615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3615_equation_0, values = (var_34539_cast_fp16, var_34266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3615_cast_fp16")]; + string _SplitHeadsQ__mh_w_3617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3617_equation_0, values = (var_34543_cast_fp16, var_34273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3617_cast_fp16")]; + string _SplitHeadsQ__mh_w_3619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3619_equation_0, values = (var_34543_cast_fp16, var_34280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3619_cast_fp16")]; + string _SplitHeadsQ__mh_w_3621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3621_equation_0, values = (var_34543_cast_fp16, var_34287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3621_cast_fp16")]; + string _SplitHeadsQ__mh_w_3623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3623_equation_0, values = (var_34543_cast_fp16, var_34294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3623_cast_fp16")]; + string _SplitHeadsQ__mh_w_3625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3625_equation_0, values = (var_34547_cast_fp16, var_34301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3625_cast_fp16")]; + string _SplitHeadsQ__mh_w_3627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3627_equation_0, values = (var_34547_cast_fp16, var_34308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3627_cast_fp16")]; + string _SplitHeadsQ__mh_w_3629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3629_equation_0, values = (var_34547_cast_fp16, var_34315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3629_cast_fp16")]; + string _SplitHeadsQ__mh_w_3631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3631_equation_0, values = (var_34547_cast_fp16, var_34322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3631_cast_fp16")]; + string _SplitHeadsQ__mh_w_3633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3633_equation_0, values = (var_34551_cast_fp16, var_34329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3633_cast_fp16")]; + string _SplitHeadsQ__mh_w_3635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3635_equation_0, values = (var_34551_cast_fp16, var_34336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3635_cast_fp16")]; + string _SplitHeadsQ__mh_w_3637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3637_equation_0, values = (var_34551_cast_fp16, var_34343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3637_cast_fp16")]; + string _SplitHeadsQ__mh_w_3639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3639_equation_0, values = (var_34551_cast_fp16, var_34350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3639_cast_fp16")]; + string _SplitHeadsQ__mh_w_3641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3641_equation_0, values = (var_34555_cast_fp16, var_34357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3641_cast_fp16")]; + string _SplitHeadsQ__mh_w_3643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3643_equation_0, values = (var_34555_cast_fp16, var_34364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3643_cast_fp16")]; + string _SplitHeadsQ__mh_w_3645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3645_equation_0, values = (var_34555_cast_fp16, var_34371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3645_cast_fp16")]; + string _SplitHeadsQ__mh_w_3647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3647_equation_0, values = (var_34555_cast_fp16, var_34378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3647_cast_fp16")]; + string _SplitHeadsQ__mh_w_3649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3649_equation_0, values = (var_34559_cast_fp16, var_34385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3649_cast_fp16")]; + string _SplitHeadsQ__mh_w_3651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3651_equation_0, values = (var_34559_cast_fp16, var_34392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3651_cast_fp16")]; + string _SplitHeadsQ__mh_w_3653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3653_equation_0, values = (var_34559_cast_fp16, var_34399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3653_cast_fp16")]; + string _SplitHeadsQ__mh_w_3655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3655_equation_0, values = (var_34559_cast_fp16, var_34406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3655_cast_fp16")]; + string _SplitHeadsQ__mh_w_3657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3657_equation_0, values = (var_34563_cast_fp16, var_34413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3657_cast_fp16")]; + string _SplitHeadsQ__mh_w_3659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3659_equation_0, values = (var_34563_cast_fp16, var_34420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3659_cast_fp16")]; + string _SplitHeadsQ__mh_w_3661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3661_equation_0, values = (var_34563_cast_fp16, var_34427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3661_cast_fp16")]; + string _SplitHeadsQ__mh_w_3663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3663_equation_0, values = (var_34563_cast_fp16, var_34434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3663_cast_fp16")]; + string _SplitHeadsQ__mh_w_3665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3665_equation_0, values = (var_34567_cast_fp16, var_34441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3665_cast_fp16")]; + string _SplitHeadsQ__mh_w_3667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3667_equation_0, values = (var_34567_cast_fp16, var_34448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3667_cast_fp16")]; + string _SplitHeadsQ__mh_w_3669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3669_equation_0, values = (var_34567_cast_fp16, var_34455_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3669_cast_fp16")]; + string _SplitHeadsQ__mh_w_3671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3671_equation_0, values = (var_34567_cast_fp16, var_34462_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3671_cast_fp16")]; + string _SplitHeadsQ__mh_w_3673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3673_equation_0, values = (var_34571_cast_fp16, var_34469_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3673_cast_fp16")]; + string _SplitHeadsQ__mh_w_3675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3675_equation_0, values = (var_34571_cast_fp16, var_34476_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3675_cast_fp16")]; + string _SplitHeadsQ__mh_w_3677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3677_equation_0, values = (var_34571_cast_fp16, var_34483_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3677_cast_fp16")]; + string _SplitHeadsQ__mh_w_3679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3679_equation_0, values = (var_34571_cast_fp16, var_34490_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3679_cast_fp16")]; + fp16 var_34812_to_fp16 = const()[name = string("op_34812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3521_cast_fp16, y = var_34812_to_fp16)[name = string("aw_chunk_3521_cast_fp16")]; + fp16 var_34814_to_fp16 = const()[name = string("op_34814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3523_cast_fp16, y = var_34814_to_fp16)[name = string("aw_chunk_3523_cast_fp16")]; + fp16 var_34816_to_fp16 = const()[name = string("op_34816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3525_cast_fp16, y = var_34816_to_fp16)[name = string("aw_chunk_3525_cast_fp16")]; + fp16 var_34818_to_fp16 = const()[name = string("op_34818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3527_cast_fp16, y = var_34818_to_fp16)[name = string("aw_chunk_3527_cast_fp16")]; + fp16 var_34820_to_fp16 = const()[name = string("op_34820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3529_cast_fp16, y = var_34820_to_fp16)[name = string("aw_chunk_3529_cast_fp16")]; + fp16 var_34822_to_fp16 = const()[name = string("op_34822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3531_cast_fp16, y = var_34822_to_fp16)[name = string("aw_chunk_3531_cast_fp16")]; + fp16 var_34824_to_fp16 = const()[name = string("op_34824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3533_cast_fp16, y = var_34824_to_fp16)[name = string("aw_chunk_3533_cast_fp16")]; + fp16 var_34826_to_fp16 = const()[name = string("op_34826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3535_cast_fp16, y = var_34826_to_fp16)[name = string("aw_chunk_3535_cast_fp16")]; + fp16 var_34828_to_fp16 = const()[name = string("op_34828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3537_cast_fp16, y = var_34828_to_fp16)[name = string("aw_chunk_3537_cast_fp16")]; + fp16 var_34830_to_fp16 = const()[name = string("op_34830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3539_cast_fp16, y = var_34830_to_fp16)[name = string("aw_chunk_3539_cast_fp16")]; + fp16 var_34832_to_fp16 = const()[name = string("op_34832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3541_cast_fp16, y = var_34832_to_fp16)[name = string("aw_chunk_3541_cast_fp16")]; + fp16 var_34834_to_fp16 = const()[name = string("op_34834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3543_cast_fp16, y = var_34834_to_fp16)[name = string("aw_chunk_3543_cast_fp16")]; + fp16 var_34836_to_fp16 = const()[name = string("op_34836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3545_cast_fp16, y = var_34836_to_fp16)[name = string("aw_chunk_3545_cast_fp16")]; + fp16 var_34838_to_fp16 = const()[name = string("op_34838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3547_cast_fp16, y = var_34838_to_fp16)[name = string("aw_chunk_3547_cast_fp16")]; + fp16 var_34840_to_fp16 = const()[name = string("op_34840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3549_cast_fp16, y = var_34840_to_fp16)[name = string("aw_chunk_3549_cast_fp16")]; + fp16 var_34842_to_fp16 = const()[name = string("op_34842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3551_cast_fp16, y = var_34842_to_fp16)[name = string("aw_chunk_3551_cast_fp16")]; + fp16 var_34844_to_fp16 = const()[name = string("op_34844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3553_cast_fp16, y = var_34844_to_fp16)[name = string("aw_chunk_3553_cast_fp16")]; + fp16 var_34846_to_fp16 = const()[name = string("op_34846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3555_cast_fp16, y = var_34846_to_fp16)[name = string("aw_chunk_3555_cast_fp16")]; + fp16 var_34848_to_fp16 = const()[name = string("op_34848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3557_cast_fp16, y = var_34848_to_fp16)[name = string("aw_chunk_3557_cast_fp16")]; + fp16 var_34850_to_fp16 = const()[name = string("op_34850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3559_cast_fp16, y = var_34850_to_fp16)[name = string("aw_chunk_3559_cast_fp16")]; + fp16 var_34852_to_fp16 = const()[name = string("op_34852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3561_cast_fp16, y = var_34852_to_fp16)[name = string("aw_chunk_3561_cast_fp16")]; + fp16 var_34854_to_fp16 = const()[name = string("op_34854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3563_cast_fp16, y = var_34854_to_fp16)[name = string("aw_chunk_3563_cast_fp16")]; + fp16 var_34856_to_fp16 = const()[name = string("op_34856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3565_cast_fp16, y = var_34856_to_fp16)[name = string("aw_chunk_3565_cast_fp16")]; + fp16 var_34858_to_fp16 = const()[name = string("op_34858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3567_cast_fp16, y = var_34858_to_fp16)[name = string("aw_chunk_3567_cast_fp16")]; + fp16 var_34860_to_fp16 = const()[name = string("op_34860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3569_cast_fp16, y = var_34860_to_fp16)[name = string("aw_chunk_3569_cast_fp16")]; + fp16 var_34862_to_fp16 = const()[name = string("op_34862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3571_cast_fp16, y = var_34862_to_fp16)[name = string("aw_chunk_3571_cast_fp16")]; + fp16 var_34864_to_fp16 = const()[name = string("op_34864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3573_cast_fp16, y = var_34864_to_fp16)[name = string("aw_chunk_3573_cast_fp16")]; + fp16 var_34866_to_fp16 = const()[name = string("op_34866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3575_cast_fp16, y = var_34866_to_fp16)[name = string("aw_chunk_3575_cast_fp16")]; + fp16 var_34868_to_fp16 = const()[name = string("op_34868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3577_cast_fp16, y = var_34868_to_fp16)[name = string("aw_chunk_3577_cast_fp16")]; + fp16 var_34870_to_fp16 = const()[name = string("op_34870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3579_cast_fp16, y = var_34870_to_fp16)[name = string("aw_chunk_3579_cast_fp16")]; + fp16 var_34872_to_fp16 = const()[name = string("op_34872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3581_cast_fp16, y = var_34872_to_fp16)[name = string("aw_chunk_3581_cast_fp16")]; + fp16 var_34874_to_fp16 = const()[name = string("op_34874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3583_cast_fp16, y = var_34874_to_fp16)[name = string("aw_chunk_3583_cast_fp16")]; + fp16 var_34876_to_fp16 = const()[name = string("op_34876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3585_cast_fp16, y = var_34876_to_fp16)[name = string("aw_chunk_3585_cast_fp16")]; + fp16 var_34878_to_fp16 = const()[name = string("op_34878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3587_cast_fp16, y = var_34878_to_fp16)[name = string("aw_chunk_3587_cast_fp16")]; + fp16 var_34880_to_fp16 = const()[name = string("op_34880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3589_cast_fp16, y = var_34880_to_fp16)[name = string("aw_chunk_3589_cast_fp16")]; + fp16 var_34882_to_fp16 = const()[name = string("op_34882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3591_cast_fp16, y = var_34882_to_fp16)[name = string("aw_chunk_3591_cast_fp16")]; + fp16 var_34884_to_fp16 = const()[name = string("op_34884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3593_cast_fp16, y = var_34884_to_fp16)[name = string("aw_chunk_3593_cast_fp16")]; + fp16 var_34886_to_fp16 = const()[name = string("op_34886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3595_cast_fp16, y = var_34886_to_fp16)[name = string("aw_chunk_3595_cast_fp16")]; + fp16 var_34888_to_fp16 = const()[name = string("op_34888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3597_cast_fp16, y = var_34888_to_fp16)[name = string("aw_chunk_3597_cast_fp16")]; + fp16 var_34890_to_fp16 = const()[name = string("op_34890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3599_cast_fp16, y = var_34890_to_fp16)[name = string("aw_chunk_3599_cast_fp16")]; + fp16 var_34892_to_fp16 = const()[name = string("op_34892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3601_cast_fp16, y = var_34892_to_fp16)[name = string("aw_chunk_3601_cast_fp16")]; + fp16 var_34894_to_fp16 = const()[name = string("op_34894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3603_cast_fp16, y = var_34894_to_fp16)[name = string("aw_chunk_3603_cast_fp16")]; + fp16 var_34896_to_fp16 = const()[name = string("op_34896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3605_cast_fp16, y = var_34896_to_fp16)[name = string("aw_chunk_3605_cast_fp16")]; + fp16 var_34898_to_fp16 = const()[name = string("op_34898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3607_cast_fp16, y = var_34898_to_fp16)[name = string("aw_chunk_3607_cast_fp16")]; + fp16 var_34900_to_fp16 = const()[name = string("op_34900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3609_cast_fp16, y = var_34900_to_fp16)[name = string("aw_chunk_3609_cast_fp16")]; + fp16 var_34902_to_fp16 = const()[name = string("op_34902_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3611_cast_fp16, y = var_34902_to_fp16)[name = string("aw_chunk_3611_cast_fp16")]; + fp16 var_34904_to_fp16 = const()[name = string("op_34904_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3613_cast_fp16, y = var_34904_to_fp16)[name = string("aw_chunk_3613_cast_fp16")]; + fp16 var_34906_to_fp16 = const()[name = string("op_34906_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3615_cast_fp16, y = var_34906_to_fp16)[name = string("aw_chunk_3615_cast_fp16")]; + fp16 var_34908_to_fp16 = const()[name = string("op_34908_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3617_cast_fp16, y = var_34908_to_fp16)[name = string("aw_chunk_3617_cast_fp16")]; + fp16 var_34910_to_fp16 = const()[name = string("op_34910_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3619_cast_fp16, y = var_34910_to_fp16)[name = string("aw_chunk_3619_cast_fp16")]; + fp16 var_34912_to_fp16 = const()[name = string("op_34912_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3621_cast_fp16, y = var_34912_to_fp16)[name = string("aw_chunk_3621_cast_fp16")]; + fp16 var_34914_to_fp16 = const()[name = string("op_34914_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3623_cast_fp16, y = var_34914_to_fp16)[name = string("aw_chunk_3623_cast_fp16")]; + fp16 var_34916_to_fp16 = const()[name = string("op_34916_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3625_cast_fp16, y = var_34916_to_fp16)[name = string("aw_chunk_3625_cast_fp16")]; + fp16 var_34918_to_fp16 = const()[name = string("op_34918_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3627_cast_fp16, y = var_34918_to_fp16)[name = string("aw_chunk_3627_cast_fp16")]; + fp16 var_34920_to_fp16 = const()[name = string("op_34920_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3629_cast_fp16, y = var_34920_to_fp16)[name = string("aw_chunk_3629_cast_fp16")]; + fp16 var_34922_to_fp16 = const()[name = string("op_34922_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3631_cast_fp16, y = var_34922_to_fp16)[name = string("aw_chunk_3631_cast_fp16")]; + fp16 var_34924_to_fp16 = const()[name = string("op_34924_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3633_cast_fp16, y = var_34924_to_fp16)[name = string("aw_chunk_3633_cast_fp16")]; + fp16 var_34926_to_fp16 = const()[name = string("op_34926_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3635_cast_fp16, y = var_34926_to_fp16)[name = string("aw_chunk_3635_cast_fp16")]; + fp16 var_34928_to_fp16 = const()[name = string("op_34928_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3637_cast_fp16, y = var_34928_to_fp16)[name = string("aw_chunk_3637_cast_fp16")]; + fp16 var_34930_to_fp16 = const()[name = string("op_34930_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3639_cast_fp16, y = var_34930_to_fp16)[name = string("aw_chunk_3639_cast_fp16")]; + fp16 var_34932_to_fp16 = const()[name = string("op_34932_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3641_cast_fp16, y = var_34932_to_fp16)[name = string("aw_chunk_3641_cast_fp16")]; + fp16 var_34934_to_fp16 = const()[name = string("op_34934_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3643_cast_fp16, y = var_34934_to_fp16)[name = string("aw_chunk_3643_cast_fp16")]; + fp16 var_34936_to_fp16 = const()[name = string("op_34936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3645_cast_fp16, y = var_34936_to_fp16)[name = string("aw_chunk_3645_cast_fp16")]; + fp16 var_34938_to_fp16 = const()[name = string("op_34938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3647_cast_fp16, y = var_34938_to_fp16)[name = string("aw_chunk_3647_cast_fp16")]; + fp16 var_34940_to_fp16 = const()[name = string("op_34940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3649_cast_fp16, y = var_34940_to_fp16)[name = string("aw_chunk_3649_cast_fp16")]; + fp16 var_34942_to_fp16 = const()[name = string("op_34942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3651_cast_fp16, y = var_34942_to_fp16)[name = string("aw_chunk_3651_cast_fp16")]; + fp16 var_34944_to_fp16 = const()[name = string("op_34944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3653_cast_fp16, y = var_34944_to_fp16)[name = string("aw_chunk_3653_cast_fp16")]; + fp16 var_34946_to_fp16 = const()[name = string("op_34946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3655_cast_fp16, y = var_34946_to_fp16)[name = string("aw_chunk_3655_cast_fp16")]; + fp16 var_34948_to_fp16 = const()[name = string("op_34948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3657_cast_fp16, y = var_34948_to_fp16)[name = string("aw_chunk_3657_cast_fp16")]; + fp16 var_34950_to_fp16 = const()[name = string("op_34950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3659_cast_fp16, y = var_34950_to_fp16)[name = string("aw_chunk_3659_cast_fp16")]; + fp16 var_34952_to_fp16 = const()[name = string("op_34952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3661_cast_fp16, y = var_34952_to_fp16)[name = string("aw_chunk_3661_cast_fp16")]; + fp16 var_34954_to_fp16 = const()[name = string("op_34954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3663_cast_fp16, y = var_34954_to_fp16)[name = string("aw_chunk_3663_cast_fp16")]; + fp16 var_34956_to_fp16 = const()[name = string("op_34956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3665_cast_fp16, y = var_34956_to_fp16)[name = string("aw_chunk_3665_cast_fp16")]; + fp16 var_34958_to_fp16 = const()[name = string("op_34958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3667_cast_fp16, y = var_34958_to_fp16)[name = string("aw_chunk_3667_cast_fp16")]; + fp16 var_34960_to_fp16 = const()[name = string("op_34960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3669_cast_fp16, y = var_34960_to_fp16)[name = string("aw_chunk_3669_cast_fp16")]; + fp16 var_34962_to_fp16 = const()[name = string("op_34962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3671_cast_fp16, y = var_34962_to_fp16)[name = string("aw_chunk_3671_cast_fp16")]; + fp16 var_34964_to_fp16 = const()[name = string("op_34964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3673_cast_fp16, y = var_34964_to_fp16)[name = string("aw_chunk_3673_cast_fp16")]; + fp16 var_34966_to_fp16 = const()[name = string("op_34966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3675_cast_fp16, y = var_34966_to_fp16)[name = string("aw_chunk_3675_cast_fp16")]; + fp16 var_34968_to_fp16 = const()[name = string("op_34968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3677_cast_fp16, y = var_34968_to_fp16)[name = string("aw_chunk_3677_cast_fp16")]; + fp16 var_34970_to_fp16 = const()[name = string("op_34970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3679_cast_fp16, y = var_34970_to_fp16)[name = string("aw_chunk_3679_cast_fp16")]; + tensor var_34972_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3521_cast_fp16)[name = string("op_34972_cast_fp16")]; + tensor var_34973_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3523_cast_fp16)[name = string("op_34973_cast_fp16")]; + tensor var_34974_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3525_cast_fp16)[name = string("op_34974_cast_fp16")]; + tensor var_34975_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3527_cast_fp16)[name = string("op_34975_cast_fp16")]; + tensor var_34976_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3529_cast_fp16)[name = string("op_34976_cast_fp16")]; + tensor var_34977_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3531_cast_fp16)[name = string("op_34977_cast_fp16")]; + tensor var_34978_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3533_cast_fp16)[name = string("op_34978_cast_fp16")]; + tensor var_34979_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3535_cast_fp16)[name = string("op_34979_cast_fp16")]; + tensor var_34980_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3537_cast_fp16)[name = string("op_34980_cast_fp16")]; + tensor var_34981_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3539_cast_fp16)[name = string("op_34981_cast_fp16")]; + tensor var_34982_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3541_cast_fp16)[name = string("op_34982_cast_fp16")]; + tensor var_34983_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3543_cast_fp16)[name = string("op_34983_cast_fp16")]; + tensor var_34984_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3545_cast_fp16)[name = string("op_34984_cast_fp16")]; + tensor var_34985_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3547_cast_fp16)[name = string("op_34985_cast_fp16")]; + tensor var_34986_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3549_cast_fp16)[name = string("op_34986_cast_fp16")]; + tensor var_34987_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3551_cast_fp16)[name = string("op_34987_cast_fp16")]; + tensor var_34988_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3553_cast_fp16)[name = string("op_34988_cast_fp16")]; + tensor var_34989_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3555_cast_fp16)[name = string("op_34989_cast_fp16")]; + tensor var_34990_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3557_cast_fp16)[name = string("op_34990_cast_fp16")]; + tensor var_34991_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3559_cast_fp16)[name = string("op_34991_cast_fp16")]; + tensor var_34992_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3561_cast_fp16)[name = string("op_34992_cast_fp16")]; + tensor var_34993_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3563_cast_fp16)[name = string("op_34993_cast_fp16")]; + tensor var_34994_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3565_cast_fp16)[name = string("op_34994_cast_fp16")]; + tensor var_34995_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3567_cast_fp16)[name = string("op_34995_cast_fp16")]; + tensor var_34996_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3569_cast_fp16)[name = string("op_34996_cast_fp16")]; + tensor var_34997_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3571_cast_fp16)[name = string("op_34997_cast_fp16")]; + tensor var_34998_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3573_cast_fp16)[name = string("op_34998_cast_fp16")]; + tensor var_34999_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3575_cast_fp16)[name = string("op_34999_cast_fp16")]; + tensor var_35000_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3577_cast_fp16)[name = string("op_35000_cast_fp16")]; + tensor var_35001_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3579_cast_fp16)[name = string("op_35001_cast_fp16")]; + tensor var_35002_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3581_cast_fp16)[name = string("op_35002_cast_fp16")]; + tensor var_35003_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3583_cast_fp16)[name = string("op_35003_cast_fp16")]; + tensor var_35004_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3585_cast_fp16)[name = string("op_35004_cast_fp16")]; + tensor var_35005_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3587_cast_fp16)[name = string("op_35005_cast_fp16")]; + tensor var_35006_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3589_cast_fp16)[name = string("op_35006_cast_fp16")]; + tensor var_35007_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3591_cast_fp16)[name = string("op_35007_cast_fp16")]; + tensor var_35008_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3593_cast_fp16)[name = string("op_35008_cast_fp16")]; + tensor var_35009_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3595_cast_fp16)[name = string("op_35009_cast_fp16")]; + tensor var_35010_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3597_cast_fp16)[name = string("op_35010_cast_fp16")]; + tensor var_35011_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3599_cast_fp16)[name = string("op_35011_cast_fp16")]; + tensor var_35012_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3601_cast_fp16)[name = string("op_35012_cast_fp16")]; + tensor var_35013_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3603_cast_fp16)[name = string("op_35013_cast_fp16")]; + tensor var_35014_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3605_cast_fp16)[name = string("op_35014_cast_fp16")]; + tensor var_35015_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3607_cast_fp16)[name = string("op_35015_cast_fp16")]; + tensor var_35016_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3609_cast_fp16)[name = string("op_35016_cast_fp16")]; + tensor var_35017_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3611_cast_fp16)[name = string("op_35017_cast_fp16")]; + tensor var_35018_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3613_cast_fp16)[name = string("op_35018_cast_fp16")]; + tensor var_35019_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3615_cast_fp16)[name = string("op_35019_cast_fp16")]; + tensor var_35020_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3617_cast_fp16)[name = string("op_35020_cast_fp16")]; + tensor var_35021_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3619_cast_fp16)[name = string("op_35021_cast_fp16")]; + tensor var_35022_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3621_cast_fp16)[name = string("op_35022_cast_fp16")]; + tensor var_35023_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3623_cast_fp16)[name = string("op_35023_cast_fp16")]; + tensor var_35024_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3625_cast_fp16)[name = string("op_35024_cast_fp16")]; + tensor var_35025_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3627_cast_fp16)[name = string("op_35025_cast_fp16")]; + tensor var_35026_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3629_cast_fp16)[name = string("op_35026_cast_fp16")]; + tensor var_35027_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3631_cast_fp16)[name = string("op_35027_cast_fp16")]; + tensor var_35028_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3633_cast_fp16)[name = string("op_35028_cast_fp16")]; + tensor var_35029_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3635_cast_fp16)[name = string("op_35029_cast_fp16")]; + tensor var_35030_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3637_cast_fp16)[name = string("op_35030_cast_fp16")]; + tensor var_35031_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3639_cast_fp16)[name = string("op_35031_cast_fp16")]; + tensor var_35032_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3641_cast_fp16)[name = string("op_35032_cast_fp16")]; + tensor var_35033_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3643_cast_fp16)[name = string("op_35033_cast_fp16")]; + tensor var_35034_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3645_cast_fp16)[name = string("op_35034_cast_fp16")]; + tensor var_35035_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3647_cast_fp16)[name = string("op_35035_cast_fp16")]; + tensor var_35036_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3649_cast_fp16)[name = string("op_35036_cast_fp16")]; + tensor var_35037_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3651_cast_fp16)[name = string("op_35037_cast_fp16")]; + tensor var_35038_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3653_cast_fp16)[name = string("op_35038_cast_fp16")]; + tensor var_35039_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3655_cast_fp16)[name = string("op_35039_cast_fp16")]; + tensor var_35040_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3657_cast_fp16)[name = string("op_35040_cast_fp16")]; + tensor var_35041_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3659_cast_fp16)[name = string("op_35041_cast_fp16")]; + tensor var_35042_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3661_cast_fp16)[name = string("op_35042_cast_fp16")]; + tensor var_35043_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3663_cast_fp16)[name = string("op_35043_cast_fp16")]; + tensor var_35044_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3665_cast_fp16)[name = string("op_35044_cast_fp16")]; + tensor var_35045_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3667_cast_fp16)[name = string("op_35045_cast_fp16")]; + tensor var_35046_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3669_cast_fp16)[name = string("op_35046_cast_fp16")]; + tensor var_35047_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3671_cast_fp16)[name = string("op_35047_cast_fp16")]; + tensor var_35048_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3673_cast_fp16)[name = string("op_35048_cast_fp16")]; + tensor var_35049_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3675_cast_fp16)[name = string("op_35049_cast_fp16")]; + tensor var_35050_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3677_cast_fp16)[name = string("op_35050_cast_fp16")]; + tensor var_35051_cast_fp16 = softmax(axis = var_33797, x = aw_chunk_3679_cast_fp16)[name = string("op_35051_cast_fp16")]; + string var_35053_equation_0 = const()[name = string("op_35053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35053_cast_fp16 = einsum(equation = var_35053_equation_0, values = (var_34573_cast_fp16, var_34972_cast_fp16))[name = string("op_35053_cast_fp16")]; + string var_35055_equation_0 = const()[name = string("op_35055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35055_cast_fp16 = einsum(equation = var_35055_equation_0, values = (var_34573_cast_fp16, var_34973_cast_fp16))[name = string("op_35055_cast_fp16")]; + string var_35057_equation_0 = const()[name = string("op_35057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35057_cast_fp16 = einsum(equation = var_35057_equation_0, values = (var_34573_cast_fp16, var_34974_cast_fp16))[name = string("op_35057_cast_fp16")]; + string var_35059_equation_0 = const()[name = string("op_35059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35059_cast_fp16 = einsum(equation = var_35059_equation_0, values = (var_34573_cast_fp16, var_34975_cast_fp16))[name = string("op_35059_cast_fp16")]; + string var_35061_equation_0 = const()[name = string("op_35061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35061_cast_fp16 = einsum(equation = var_35061_equation_0, values = (var_34577_cast_fp16, var_34976_cast_fp16))[name = string("op_35061_cast_fp16")]; + string var_35063_equation_0 = const()[name = string("op_35063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35063_cast_fp16 = einsum(equation = var_35063_equation_0, values = (var_34577_cast_fp16, var_34977_cast_fp16))[name = string("op_35063_cast_fp16")]; + string var_35065_equation_0 = const()[name = string("op_35065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35065_cast_fp16 = einsum(equation = var_35065_equation_0, values = (var_34577_cast_fp16, var_34978_cast_fp16))[name = string("op_35065_cast_fp16")]; + string var_35067_equation_0 = const()[name = string("op_35067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35067_cast_fp16 = einsum(equation = var_35067_equation_0, values = (var_34577_cast_fp16, var_34979_cast_fp16))[name = string("op_35067_cast_fp16")]; + string var_35069_equation_0 = const()[name = string("op_35069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35069_cast_fp16 = einsum(equation = var_35069_equation_0, values = (var_34581_cast_fp16, var_34980_cast_fp16))[name = string("op_35069_cast_fp16")]; + string var_35071_equation_0 = const()[name = string("op_35071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35071_cast_fp16 = einsum(equation = var_35071_equation_0, values = (var_34581_cast_fp16, var_34981_cast_fp16))[name = string("op_35071_cast_fp16")]; + string var_35073_equation_0 = const()[name = string("op_35073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35073_cast_fp16 = einsum(equation = var_35073_equation_0, values = (var_34581_cast_fp16, var_34982_cast_fp16))[name = string("op_35073_cast_fp16")]; + string var_35075_equation_0 = const()[name = string("op_35075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35075_cast_fp16 = einsum(equation = var_35075_equation_0, values = (var_34581_cast_fp16, var_34983_cast_fp16))[name = string("op_35075_cast_fp16")]; + string var_35077_equation_0 = const()[name = string("op_35077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35077_cast_fp16 = einsum(equation = var_35077_equation_0, values = (var_34585_cast_fp16, var_34984_cast_fp16))[name = string("op_35077_cast_fp16")]; + string var_35079_equation_0 = const()[name = string("op_35079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35079_cast_fp16 = einsum(equation = var_35079_equation_0, values = (var_34585_cast_fp16, var_34985_cast_fp16))[name = string("op_35079_cast_fp16")]; + string var_35081_equation_0 = const()[name = string("op_35081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35081_cast_fp16 = einsum(equation = var_35081_equation_0, values = (var_34585_cast_fp16, var_34986_cast_fp16))[name = string("op_35081_cast_fp16")]; + string var_35083_equation_0 = const()[name = string("op_35083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35083_cast_fp16 = einsum(equation = var_35083_equation_0, values = (var_34585_cast_fp16, var_34987_cast_fp16))[name = string("op_35083_cast_fp16")]; + string var_35085_equation_0 = const()[name = string("op_35085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35085_cast_fp16 = einsum(equation = var_35085_equation_0, values = (var_34589_cast_fp16, var_34988_cast_fp16))[name = string("op_35085_cast_fp16")]; + string var_35087_equation_0 = const()[name = string("op_35087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35087_cast_fp16 = einsum(equation = var_35087_equation_0, values = (var_34589_cast_fp16, var_34989_cast_fp16))[name = string("op_35087_cast_fp16")]; + string var_35089_equation_0 = const()[name = string("op_35089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35089_cast_fp16 = einsum(equation = var_35089_equation_0, values = (var_34589_cast_fp16, var_34990_cast_fp16))[name = string("op_35089_cast_fp16")]; + string var_35091_equation_0 = const()[name = string("op_35091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35091_cast_fp16 = einsum(equation = var_35091_equation_0, values = (var_34589_cast_fp16, var_34991_cast_fp16))[name = string("op_35091_cast_fp16")]; + string var_35093_equation_0 = const()[name = string("op_35093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35093_cast_fp16 = einsum(equation = var_35093_equation_0, values = (var_34593_cast_fp16, var_34992_cast_fp16))[name = string("op_35093_cast_fp16")]; + string var_35095_equation_0 = const()[name = string("op_35095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35095_cast_fp16 = einsum(equation = var_35095_equation_0, values = (var_34593_cast_fp16, var_34993_cast_fp16))[name = string("op_35095_cast_fp16")]; + string var_35097_equation_0 = const()[name = string("op_35097_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35097_cast_fp16 = einsum(equation = var_35097_equation_0, values = (var_34593_cast_fp16, var_34994_cast_fp16))[name = string("op_35097_cast_fp16")]; + string var_35099_equation_0 = const()[name = string("op_35099_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35099_cast_fp16 = einsum(equation = var_35099_equation_0, values = (var_34593_cast_fp16, var_34995_cast_fp16))[name = string("op_35099_cast_fp16")]; + string var_35101_equation_0 = const()[name = string("op_35101_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35101_cast_fp16 = einsum(equation = var_35101_equation_0, values = (var_34597_cast_fp16, var_34996_cast_fp16))[name = string("op_35101_cast_fp16")]; + string var_35103_equation_0 = const()[name = string("op_35103_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35103_cast_fp16 = einsum(equation = var_35103_equation_0, values = (var_34597_cast_fp16, var_34997_cast_fp16))[name = string("op_35103_cast_fp16")]; + string var_35105_equation_0 = const()[name = string("op_35105_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35105_cast_fp16 = einsum(equation = var_35105_equation_0, values = (var_34597_cast_fp16, var_34998_cast_fp16))[name = string("op_35105_cast_fp16")]; + string var_35107_equation_0 = const()[name = string("op_35107_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35107_cast_fp16 = einsum(equation = var_35107_equation_0, values = (var_34597_cast_fp16, var_34999_cast_fp16))[name = string("op_35107_cast_fp16")]; + string var_35109_equation_0 = const()[name = string("op_35109_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35109_cast_fp16 = einsum(equation = var_35109_equation_0, values = (var_34601_cast_fp16, var_35000_cast_fp16))[name = string("op_35109_cast_fp16")]; + string var_35111_equation_0 = const()[name = string("op_35111_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35111_cast_fp16 = einsum(equation = var_35111_equation_0, values = (var_34601_cast_fp16, var_35001_cast_fp16))[name = string("op_35111_cast_fp16")]; + string var_35113_equation_0 = const()[name = string("op_35113_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35113_cast_fp16 = einsum(equation = var_35113_equation_0, values = (var_34601_cast_fp16, var_35002_cast_fp16))[name = string("op_35113_cast_fp16")]; + string var_35115_equation_0 = const()[name = string("op_35115_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35115_cast_fp16 = einsum(equation = var_35115_equation_0, values = (var_34601_cast_fp16, var_35003_cast_fp16))[name = string("op_35115_cast_fp16")]; + string var_35117_equation_0 = const()[name = string("op_35117_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35117_cast_fp16 = einsum(equation = var_35117_equation_0, values = (var_34605_cast_fp16, var_35004_cast_fp16))[name = string("op_35117_cast_fp16")]; + string var_35119_equation_0 = const()[name = string("op_35119_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35119_cast_fp16 = einsum(equation = var_35119_equation_0, values = (var_34605_cast_fp16, var_35005_cast_fp16))[name = string("op_35119_cast_fp16")]; + string var_35121_equation_0 = const()[name = string("op_35121_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35121_cast_fp16 = einsum(equation = var_35121_equation_0, values = (var_34605_cast_fp16, var_35006_cast_fp16))[name = string("op_35121_cast_fp16")]; + string var_35123_equation_0 = const()[name = string("op_35123_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35123_cast_fp16 = einsum(equation = var_35123_equation_0, values = (var_34605_cast_fp16, var_35007_cast_fp16))[name = string("op_35123_cast_fp16")]; + string var_35125_equation_0 = const()[name = string("op_35125_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35125_cast_fp16 = einsum(equation = var_35125_equation_0, values = (var_34609_cast_fp16, var_35008_cast_fp16))[name = string("op_35125_cast_fp16")]; + string var_35127_equation_0 = const()[name = string("op_35127_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35127_cast_fp16 = einsum(equation = var_35127_equation_0, values = (var_34609_cast_fp16, var_35009_cast_fp16))[name = string("op_35127_cast_fp16")]; + string var_35129_equation_0 = const()[name = string("op_35129_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35129_cast_fp16 = einsum(equation = var_35129_equation_0, values = (var_34609_cast_fp16, var_35010_cast_fp16))[name = string("op_35129_cast_fp16")]; + string var_35131_equation_0 = const()[name = string("op_35131_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35131_cast_fp16 = einsum(equation = var_35131_equation_0, values = (var_34609_cast_fp16, var_35011_cast_fp16))[name = string("op_35131_cast_fp16")]; + string var_35133_equation_0 = const()[name = string("op_35133_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35133_cast_fp16 = einsum(equation = var_35133_equation_0, values = (var_34613_cast_fp16, var_35012_cast_fp16))[name = string("op_35133_cast_fp16")]; + string var_35135_equation_0 = const()[name = string("op_35135_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35135_cast_fp16 = einsum(equation = var_35135_equation_0, values = (var_34613_cast_fp16, var_35013_cast_fp16))[name = string("op_35135_cast_fp16")]; + string var_35137_equation_0 = const()[name = string("op_35137_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35137_cast_fp16 = einsum(equation = var_35137_equation_0, values = (var_34613_cast_fp16, var_35014_cast_fp16))[name = string("op_35137_cast_fp16")]; + string var_35139_equation_0 = const()[name = string("op_35139_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35139_cast_fp16 = einsum(equation = var_35139_equation_0, values = (var_34613_cast_fp16, var_35015_cast_fp16))[name = string("op_35139_cast_fp16")]; + string var_35141_equation_0 = const()[name = string("op_35141_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35141_cast_fp16 = einsum(equation = var_35141_equation_0, values = (var_34617_cast_fp16, var_35016_cast_fp16))[name = string("op_35141_cast_fp16")]; + string var_35143_equation_0 = const()[name = string("op_35143_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35143_cast_fp16 = einsum(equation = var_35143_equation_0, values = (var_34617_cast_fp16, var_35017_cast_fp16))[name = string("op_35143_cast_fp16")]; + string var_35145_equation_0 = const()[name = string("op_35145_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35145_cast_fp16 = einsum(equation = var_35145_equation_0, values = (var_34617_cast_fp16, var_35018_cast_fp16))[name = string("op_35145_cast_fp16")]; + string var_35147_equation_0 = const()[name = string("op_35147_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35147_cast_fp16 = einsum(equation = var_35147_equation_0, values = (var_34617_cast_fp16, var_35019_cast_fp16))[name = string("op_35147_cast_fp16")]; + string var_35149_equation_0 = const()[name = string("op_35149_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35149_cast_fp16 = einsum(equation = var_35149_equation_0, values = (var_34621_cast_fp16, var_35020_cast_fp16))[name = string("op_35149_cast_fp16")]; + string var_35151_equation_0 = const()[name = string("op_35151_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35151_cast_fp16 = einsum(equation = var_35151_equation_0, values = (var_34621_cast_fp16, var_35021_cast_fp16))[name = string("op_35151_cast_fp16")]; + string var_35153_equation_0 = const()[name = string("op_35153_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35153_cast_fp16 = einsum(equation = var_35153_equation_0, values = (var_34621_cast_fp16, var_35022_cast_fp16))[name = string("op_35153_cast_fp16")]; + string var_35155_equation_0 = const()[name = string("op_35155_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35155_cast_fp16 = einsum(equation = var_35155_equation_0, values = (var_34621_cast_fp16, var_35023_cast_fp16))[name = string("op_35155_cast_fp16")]; + string var_35157_equation_0 = const()[name = string("op_35157_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35157_cast_fp16 = einsum(equation = var_35157_equation_0, values = (var_34625_cast_fp16, var_35024_cast_fp16))[name = string("op_35157_cast_fp16")]; + string var_35159_equation_0 = const()[name = string("op_35159_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35159_cast_fp16 = einsum(equation = var_35159_equation_0, values = (var_34625_cast_fp16, var_35025_cast_fp16))[name = string("op_35159_cast_fp16")]; + string var_35161_equation_0 = const()[name = string("op_35161_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35161_cast_fp16 = einsum(equation = var_35161_equation_0, values = (var_34625_cast_fp16, var_35026_cast_fp16))[name = string("op_35161_cast_fp16")]; + string var_35163_equation_0 = const()[name = string("op_35163_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35163_cast_fp16 = einsum(equation = var_35163_equation_0, values = (var_34625_cast_fp16, var_35027_cast_fp16))[name = string("op_35163_cast_fp16")]; + string var_35165_equation_0 = const()[name = string("op_35165_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35165_cast_fp16 = einsum(equation = var_35165_equation_0, values = (var_34629_cast_fp16, var_35028_cast_fp16))[name = string("op_35165_cast_fp16")]; + string var_35167_equation_0 = const()[name = string("op_35167_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35167_cast_fp16 = einsum(equation = var_35167_equation_0, values = (var_34629_cast_fp16, var_35029_cast_fp16))[name = string("op_35167_cast_fp16")]; + string var_35169_equation_0 = const()[name = string("op_35169_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35169_cast_fp16 = einsum(equation = var_35169_equation_0, values = (var_34629_cast_fp16, var_35030_cast_fp16))[name = string("op_35169_cast_fp16")]; + string var_35171_equation_0 = const()[name = string("op_35171_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35171_cast_fp16 = einsum(equation = var_35171_equation_0, values = (var_34629_cast_fp16, var_35031_cast_fp16))[name = string("op_35171_cast_fp16")]; + string var_35173_equation_0 = const()[name = string("op_35173_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35173_cast_fp16 = einsum(equation = var_35173_equation_0, values = (var_34633_cast_fp16, var_35032_cast_fp16))[name = string("op_35173_cast_fp16")]; + string var_35175_equation_0 = const()[name = string("op_35175_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35175_cast_fp16 = einsum(equation = var_35175_equation_0, values = (var_34633_cast_fp16, var_35033_cast_fp16))[name = string("op_35175_cast_fp16")]; + string var_35177_equation_0 = const()[name = string("op_35177_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35177_cast_fp16 = einsum(equation = var_35177_equation_0, values = (var_34633_cast_fp16, var_35034_cast_fp16))[name = string("op_35177_cast_fp16")]; + string var_35179_equation_0 = const()[name = string("op_35179_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35179_cast_fp16 = einsum(equation = var_35179_equation_0, values = (var_34633_cast_fp16, var_35035_cast_fp16))[name = string("op_35179_cast_fp16")]; + string var_35181_equation_0 = const()[name = string("op_35181_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35181_cast_fp16 = einsum(equation = var_35181_equation_0, values = (var_34637_cast_fp16, var_35036_cast_fp16))[name = string("op_35181_cast_fp16")]; + string var_35183_equation_0 = const()[name = string("op_35183_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35183_cast_fp16 = einsum(equation = var_35183_equation_0, values = (var_34637_cast_fp16, var_35037_cast_fp16))[name = string("op_35183_cast_fp16")]; + string var_35185_equation_0 = const()[name = string("op_35185_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35185_cast_fp16 = einsum(equation = var_35185_equation_0, values = (var_34637_cast_fp16, var_35038_cast_fp16))[name = string("op_35185_cast_fp16")]; + string var_35187_equation_0 = const()[name = string("op_35187_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35187_cast_fp16 = einsum(equation = var_35187_equation_0, values = (var_34637_cast_fp16, var_35039_cast_fp16))[name = string("op_35187_cast_fp16")]; + string var_35189_equation_0 = const()[name = string("op_35189_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35189_cast_fp16 = einsum(equation = var_35189_equation_0, values = (var_34641_cast_fp16, var_35040_cast_fp16))[name = string("op_35189_cast_fp16")]; + string var_35191_equation_0 = const()[name = string("op_35191_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35191_cast_fp16 = einsum(equation = var_35191_equation_0, values = (var_34641_cast_fp16, var_35041_cast_fp16))[name = string("op_35191_cast_fp16")]; + string var_35193_equation_0 = const()[name = string("op_35193_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35193_cast_fp16 = einsum(equation = var_35193_equation_0, values = (var_34641_cast_fp16, var_35042_cast_fp16))[name = string("op_35193_cast_fp16")]; + string var_35195_equation_0 = const()[name = string("op_35195_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35195_cast_fp16 = einsum(equation = var_35195_equation_0, values = (var_34641_cast_fp16, var_35043_cast_fp16))[name = string("op_35195_cast_fp16")]; + string var_35197_equation_0 = const()[name = string("op_35197_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35197_cast_fp16 = einsum(equation = var_35197_equation_0, values = (var_34645_cast_fp16, var_35044_cast_fp16))[name = string("op_35197_cast_fp16")]; + string var_35199_equation_0 = const()[name = string("op_35199_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35199_cast_fp16 = einsum(equation = var_35199_equation_0, values = (var_34645_cast_fp16, var_35045_cast_fp16))[name = string("op_35199_cast_fp16")]; + string var_35201_equation_0 = const()[name = string("op_35201_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35201_cast_fp16 = einsum(equation = var_35201_equation_0, values = (var_34645_cast_fp16, var_35046_cast_fp16))[name = string("op_35201_cast_fp16")]; + string var_35203_equation_0 = const()[name = string("op_35203_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35203_cast_fp16 = einsum(equation = var_35203_equation_0, values = (var_34645_cast_fp16, var_35047_cast_fp16))[name = string("op_35203_cast_fp16")]; + string var_35205_equation_0 = const()[name = string("op_35205_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35205_cast_fp16 = einsum(equation = var_35205_equation_0, values = (var_34649_cast_fp16, var_35048_cast_fp16))[name = string("op_35205_cast_fp16")]; + string var_35207_equation_0 = const()[name = string("op_35207_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35207_cast_fp16 = einsum(equation = var_35207_equation_0, values = (var_34649_cast_fp16, var_35049_cast_fp16))[name = string("op_35207_cast_fp16")]; + string var_35209_equation_0 = const()[name = string("op_35209_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35209_cast_fp16 = einsum(equation = var_35209_equation_0, values = (var_34649_cast_fp16, var_35050_cast_fp16))[name = string("op_35209_cast_fp16")]; + string var_35211_equation_0 = const()[name = string("op_35211_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35211_cast_fp16 = einsum(equation = var_35211_equation_0, values = (var_34649_cast_fp16, var_35051_cast_fp16))[name = string("op_35211_cast_fp16")]; + bool var_35213_interleave_0 = const()[name = string("op_35213_interleave_0"), val = bool(false)]; + tensor var_35213_cast_fp16 = concat(axis = var_33772, interleave = var_35213_interleave_0, values = (var_35053_cast_fp16, var_35055_cast_fp16, var_35057_cast_fp16, var_35059_cast_fp16))[name = string("op_35213_cast_fp16")]; + bool var_35215_interleave_0 = const()[name = string("op_35215_interleave_0"), val = bool(false)]; + tensor var_35215_cast_fp16 = concat(axis = var_33772, interleave = var_35215_interleave_0, values = (var_35061_cast_fp16, var_35063_cast_fp16, var_35065_cast_fp16, var_35067_cast_fp16))[name = string("op_35215_cast_fp16")]; + bool var_35217_interleave_0 = const()[name = string("op_35217_interleave_0"), val = bool(false)]; + tensor var_35217_cast_fp16 = concat(axis = var_33772, interleave = var_35217_interleave_0, values = (var_35069_cast_fp16, var_35071_cast_fp16, var_35073_cast_fp16, var_35075_cast_fp16))[name = string("op_35217_cast_fp16")]; + bool var_35219_interleave_0 = const()[name = string("op_35219_interleave_0"), val = bool(false)]; + tensor var_35219_cast_fp16 = concat(axis = var_33772, interleave = var_35219_interleave_0, values = (var_35077_cast_fp16, var_35079_cast_fp16, var_35081_cast_fp16, var_35083_cast_fp16))[name = string("op_35219_cast_fp16")]; + bool var_35221_interleave_0 = const()[name = string("op_35221_interleave_0"), val = bool(false)]; + tensor var_35221_cast_fp16 = concat(axis = var_33772, interleave = var_35221_interleave_0, values = (var_35085_cast_fp16, var_35087_cast_fp16, var_35089_cast_fp16, var_35091_cast_fp16))[name = string("op_35221_cast_fp16")]; + bool var_35223_interleave_0 = const()[name = string("op_35223_interleave_0"), val = bool(false)]; + tensor var_35223_cast_fp16 = concat(axis = var_33772, interleave = var_35223_interleave_0, values = (var_35093_cast_fp16, var_35095_cast_fp16, var_35097_cast_fp16, var_35099_cast_fp16))[name = string("op_35223_cast_fp16")]; + bool var_35225_interleave_0 = const()[name = string("op_35225_interleave_0"), val = bool(false)]; + tensor var_35225_cast_fp16 = concat(axis = var_33772, interleave = var_35225_interleave_0, values = (var_35101_cast_fp16, var_35103_cast_fp16, var_35105_cast_fp16, var_35107_cast_fp16))[name = string("op_35225_cast_fp16")]; + bool var_35227_interleave_0 = const()[name = string("op_35227_interleave_0"), val = bool(false)]; + tensor var_35227_cast_fp16 = concat(axis = var_33772, interleave = var_35227_interleave_0, values = (var_35109_cast_fp16, var_35111_cast_fp16, var_35113_cast_fp16, var_35115_cast_fp16))[name = string("op_35227_cast_fp16")]; + bool var_35229_interleave_0 = const()[name = string("op_35229_interleave_0"), val = bool(false)]; + tensor var_35229_cast_fp16 = concat(axis = var_33772, interleave = var_35229_interleave_0, values = (var_35117_cast_fp16, var_35119_cast_fp16, var_35121_cast_fp16, var_35123_cast_fp16))[name = string("op_35229_cast_fp16")]; + bool var_35231_interleave_0 = const()[name = string("op_35231_interleave_0"), val = bool(false)]; + tensor var_35231_cast_fp16 = concat(axis = var_33772, interleave = var_35231_interleave_0, values = (var_35125_cast_fp16, var_35127_cast_fp16, var_35129_cast_fp16, var_35131_cast_fp16))[name = string("op_35231_cast_fp16")]; + bool var_35233_interleave_0 = const()[name = string("op_35233_interleave_0"), val = bool(false)]; + tensor var_35233_cast_fp16 = concat(axis = var_33772, interleave = var_35233_interleave_0, values = (var_35133_cast_fp16, var_35135_cast_fp16, var_35137_cast_fp16, var_35139_cast_fp16))[name = string("op_35233_cast_fp16")]; + bool var_35235_interleave_0 = const()[name = string("op_35235_interleave_0"), val = bool(false)]; + tensor var_35235_cast_fp16 = concat(axis = var_33772, interleave = var_35235_interleave_0, values = (var_35141_cast_fp16, var_35143_cast_fp16, var_35145_cast_fp16, var_35147_cast_fp16))[name = string("op_35235_cast_fp16")]; + bool var_35237_interleave_0 = const()[name = string("op_35237_interleave_0"), val = bool(false)]; + tensor var_35237_cast_fp16 = concat(axis = var_33772, interleave = var_35237_interleave_0, values = (var_35149_cast_fp16, var_35151_cast_fp16, var_35153_cast_fp16, var_35155_cast_fp16))[name = string("op_35237_cast_fp16")]; + bool var_35239_interleave_0 = const()[name = string("op_35239_interleave_0"), val = bool(false)]; + tensor var_35239_cast_fp16 = concat(axis = var_33772, interleave = var_35239_interleave_0, values = (var_35157_cast_fp16, var_35159_cast_fp16, var_35161_cast_fp16, var_35163_cast_fp16))[name = string("op_35239_cast_fp16")]; + bool var_35241_interleave_0 = const()[name = string("op_35241_interleave_0"), val = bool(false)]; + tensor var_35241_cast_fp16 = concat(axis = var_33772, interleave = var_35241_interleave_0, values = (var_35165_cast_fp16, var_35167_cast_fp16, var_35169_cast_fp16, var_35171_cast_fp16))[name = string("op_35241_cast_fp16")]; + bool var_35243_interleave_0 = const()[name = string("op_35243_interleave_0"), val = bool(false)]; + tensor var_35243_cast_fp16 = concat(axis = var_33772, interleave = var_35243_interleave_0, values = (var_35173_cast_fp16, var_35175_cast_fp16, var_35177_cast_fp16, var_35179_cast_fp16))[name = string("op_35243_cast_fp16")]; + bool var_35245_interleave_0 = const()[name = string("op_35245_interleave_0"), val = bool(false)]; + tensor var_35245_cast_fp16 = concat(axis = var_33772, interleave = var_35245_interleave_0, values = (var_35181_cast_fp16, var_35183_cast_fp16, var_35185_cast_fp16, var_35187_cast_fp16))[name = string("op_35245_cast_fp16")]; + bool var_35247_interleave_0 = const()[name = string("op_35247_interleave_0"), val = bool(false)]; + tensor var_35247_cast_fp16 = concat(axis = var_33772, interleave = var_35247_interleave_0, values = (var_35189_cast_fp16, var_35191_cast_fp16, var_35193_cast_fp16, var_35195_cast_fp16))[name = string("op_35247_cast_fp16")]; + bool var_35249_interleave_0 = const()[name = string("op_35249_interleave_0"), val = bool(false)]; + tensor var_35249_cast_fp16 = concat(axis = var_33772, interleave = var_35249_interleave_0, values = (var_35197_cast_fp16, var_35199_cast_fp16, var_35201_cast_fp16, var_35203_cast_fp16))[name = string("op_35249_cast_fp16")]; + bool var_35251_interleave_0 = const()[name = string("op_35251_interleave_0"), val = bool(false)]; + tensor var_35251_cast_fp16 = concat(axis = var_33772, interleave = var_35251_interleave_0, values = (var_35205_cast_fp16, var_35207_cast_fp16, var_35209_cast_fp16, var_35211_cast_fp16))[name = string("op_35251_cast_fp16")]; + bool input_177_interleave_0 = const()[name = string("input_177_interleave_0"), val = bool(false)]; + tensor input_177_cast_fp16 = concat(axis = var_33797, interleave = input_177_interleave_0, values = (var_35213_cast_fp16, var_35215_cast_fp16, var_35217_cast_fp16, var_35219_cast_fp16, var_35221_cast_fp16, var_35223_cast_fp16, var_35225_cast_fp16, var_35227_cast_fp16, var_35229_cast_fp16, var_35231_cast_fp16, var_35233_cast_fp16, var_35235_cast_fp16, var_35237_cast_fp16, var_35239_cast_fp16, var_35241_cast_fp16, var_35243_cast_fp16, var_35245_cast_fp16, var_35247_cast_fp16, var_35249_cast_fp16, var_35251_cast_fp16))[name = string("input_177_cast_fp16")]; + string obj_91_pad_type_0 = const()[name = string("obj_91_pad_type_0"), val = string("valid")]; + tensor obj_91_strides_0 = const()[name = string("obj_91_strides_0"), val = tensor([1, 1])]; + tensor obj_91_pad_0 = const()[name = string("obj_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_91_dilations_0 = const()[name = string("obj_91_dilations_0"), val = tensor([1, 1])]; + int32 obj_91_groups_0 = const()[name = string("obj_91_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890277440)))]; + tensor layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893554304)))]; + tensor obj_91_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_177_cast_fp16)[name = string("obj_91_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = obj_91_cast_fp16)[name = string("inputs_91_cast_fp16")]; + tensor out_91_axes_0 = const()[name = string("out_91_axes_0"), val = tensor([1])]; + fp16 var_35270_to_fp16 = const()[name = string("op_35270_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_35270_to_fp16, x = inputs_91_cast_fp16)[name = string("out_91_cast_fp16")]; + tensor input_179_gamma_0_to_fp16 = const()[name = string("input_179_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893556928)))]; + tensor input_179_beta_0_to_fp16 = const()[name = string("input_179_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893559552)))]; + fp16 input_179_epsilon_0_to_fp16 = const()[name = string("input_179_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_179_cast_fp16 = batch_norm(beta = input_179_beta_0_to_fp16, epsilon = input_179_epsilon_0_to_fp16, gamma = input_179_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = string("input_179_cast_fp16")]; + string input_181_pad_type_0 = const()[name = string("input_181_pad_type_0"), val = string("valid")]; + tensor input_181_strides_0 = const()[name = string("input_181_strides_0"), val = tensor([1, 1])]; + tensor input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_181_dilations_0 = const()[name = string("input_181_dilations_0"), val = tensor([1, 1])]; + int32 input_181_groups_0 = const()[name = string("input_181_groups_0"), val = int32(1)]; + tensor layers_22_fc1_weight_to_fp16 = const()[name = string("layers_22_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893562176)))]; + tensor layers_22_fc1_bias_to_fp16 = const()[name = string("layers_22_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906669440)))]; + tensor input_181_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_181_dilations_0, groups = input_181_groups_0, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = input_181_strides_0, weight = layers_22_fc1_weight_to_fp16, x = input_179_cast_fp16)[name = string("input_181_cast_fp16")]; + string input_183_mode_0 = const()[name = string("input_183_mode_0"), val = string("EXACT")]; + tensor input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = input_181_cast_fp16)[name = string("input_183_cast_fp16")]; + string hidden_states_49_pad_type_0 = const()[name = string("hidden_states_49_pad_type_0"), val = string("valid")]; + tensor hidden_states_49_strides_0 = const()[name = string("hidden_states_49_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_49_pad_0 = const()[name = string("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_49_dilations_0 = const()[name = string("hidden_states_49_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_49_groups_0 = const()[name = string("hidden_states_49_groups_0"), val = int32(1)]; + tensor layers_22_fc2_weight_to_fp16 = const()[name = string("layers_22_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906679744)))]; + tensor layers_22_fc2_bias_to_fp16 = const()[name = string("layers_22_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919787008)))]; + tensor hidden_states_49_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_22_fc2_weight_to_fp16, x = input_183_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("inputs_93_cast_fp16")]; + int32 var_35299 = const()[name = string("op_35299"), val = int32(3)]; + int32 var_35324 = const()[name = string("op_35324"), val = int32(1)]; + tensor out_93_axes_0 = const()[name = string("out_93_axes_0"), val = tensor([1])]; + fp16 var_35341_to_fp16 = const()[name = string("op_35341_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_35341_to_fp16, x = inputs_93_cast_fp16)[name = string("out_93_cast_fp16")]; + tensor obj_93_gamma_0_to_fp16 = const()[name = string("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919789632)))]; + tensor obj_93_beta_0_to_fp16 = const()[name = string("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919792256)))]; + fp16 obj_93_epsilon_0_to_fp16 = const()[name = string("obj_93_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = string("obj_93_cast_fp16")]; + string query_47_pad_type_0 = const()[name = string("query_47_pad_type_0"), val = string("valid")]; + tensor query_47_strides_0 = const()[name = string("query_47_strides_0"), val = tensor([1, 1])]; + tensor query_47_pad_0 = const()[name = string("query_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_47_dilations_0 = const()[name = string("query_47_dilations_0"), val = tensor([1, 1])]; + int32 query_47_groups_0 = const()[name = string("query_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919794880)))]; + tensor layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923071744)))]; + tensor query_47_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("query_47_cast_fp16")]; + string key_47_pad_type_0 = const()[name = string("key_47_pad_type_0"), val = string("valid")]; + tensor key_47_strides_0 = const()[name = string("key_47_strides_0"), val = tensor([1, 1])]; + tensor key_47_pad_0 = const()[name = string("key_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_47_dilations_0 = const()[name = string("key_47_dilations_0"), val = tensor([1, 1])]; + int32 key_47_groups_0 = const()[name = string("key_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923074368)))]; + tensor key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("key_47_cast_fp16")]; + string value_47_pad_type_0 = const()[name = string("value_47_pad_type_0"), val = string("valid")]; + tensor value_47_strides_0 = const()[name = string("value_47_strides_0"), val = tensor([1, 1])]; + tensor value_47_pad_0 = const()[name = string("value_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_47_dilations_0 = const()[name = string("value_47_dilations_0"), val = tensor([1, 1])]; + int32 value_47_groups_0 = const()[name = string("value_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(926351232)))]; + tensor layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929628096)))]; + tensor value_47_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("value_47_cast_fp16")]; + tensor var_35379_begin_0 = const()[name = string("op_35379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35379_end_0 = const()[name = string("op_35379_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35379_end_mask_0 = const()[name = string("op_35379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35379_cast_fp16 = slice_by_index(begin = var_35379_begin_0, end = var_35379_end_0, end_mask = var_35379_end_mask_0, x = query_47_cast_fp16)[name = string("op_35379_cast_fp16")]; + tensor var_35383_begin_0 = const()[name = string("op_35383_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35383_end_0 = const()[name = string("op_35383_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_35383_end_mask_0 = const()[name = string("op_35383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35383_cast_fp16 = slice_by_index(begin = var_35383_begin_0, end = var_35383_end_0, end_mask = var_35383_end_mask_0, x = query_47_cast_fp16)[name = string("op_35383_cast_fp16")]; + tensor var_35387_begin_0 = const()[name = string("op_35387_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35387_end_0 = const()[name = string("op_35387_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_35387_end_mask_0 = const()[name = string("op_35387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35387_cast_fp16 = slice_by_index(begin = var_35387_begin_0, end = var_35387_end_0, end_mask = var_35387_end_mask_0, x = query_47_cast_fp16)[name = string("op_35387_cast_fp16")]; + tensor var_35391_begin_0 = const()[name = string("op_35391_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35391_end_0 = const()[name = string("op_35391_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_35391_end_mask_0 = const()[name = string("op_35391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35391_cast_fp16 = slice_by_index(begin = var_35391_begin_0, end = var_35391_end_0, end_mask = var_35391_end_mask_0, x = query_47_cast_fp16)[name = string("op_35391_cast_fp16")]; + tensor var_35395_begin_0 = const()[name = string("op_35395_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35395_end_0 = const()[name = string("op_35395_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_35395_end_mask_0 = const()[name = string("op_35395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35395_cast_fp16 = slice_by_index(begin = var_35395_begin_0, end = var_35395_end_0, end_mask = var_35395_end_mask_0, x = query_47_cast_fp16)[name = string("op_35395_cast_fp16")]; + tensor var_35399_begin_0 = const()[name = string("op_35399_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35399_end_0 = const()[name = string("op_35399_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_35399_end_mask_0 = const()[name = string("op_35399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35399_cast_fp16 = slice_by_index(begin = var_35399_begin_0, end = var_35399_end_0, end_mask = var_35399_end_mask_0, x = query_47_cast_fp16)[name = string("op_35399_cast_fp16")]; + tensor var_35403_begin_0 = const()[name = string("op_35403_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35403_end_0 = const()[name = string("op_35403_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_35403_end_mask_0 = const()[name = string("op_35403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35403_cast_fp16 = slice_by_index(begin = var_35403_begin_0, end = var_35403_end_0, end_mask = var_35403_end_mask_0, x = query_47_cast_fp16)[name = string("op_35403_cast_fp16")]; + tensor var_35407_begin_0 = const()[name = string("op_35407_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35407_end_0 = const()[name = string("op_35407_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_35407_end_mask_0 = const()[name = string("op_35407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35407_cast_fp16 = slice_by_index(begin = var_35407_begin_0, end = var_35407_end_0, end_mask = var_35407_end_mask_0, x = query_47_cast_fp16)[name = string("op_35407_cast_fp16")]; + tensor var_35411_begin_0 = const()[name = string("op_35411_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35411_end_0 = const()[name = string("op_35411_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_35411_end_mask_0 = const()[name = string("op_35411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35411_cast_fp16 = slice_by_index(begin = var_35411_begin_0, end = var_35411_end_0, end_mask = var_35411_end_mask_0, x = query_47_cast_fp16)[name = string("op_35411_cast_fp16")]; + tensor var_35415_begin_0 = const()[name = string("op_35415_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35415_end_0 = const()[name = string("op_35415_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_35415_end_mask_0 = const()[name = string("op_35415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35415_cast_fp16 = slice_by_index(begin = var_35415_begin_0, end = var_35415_end_0, end_mask = var_35415_end_mask_0, x = query_47_cast_fp16)[name = string("op_35415_cast_fp16")]; + tensor var_35419_begin_0 = const()[name = string("op_35419_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35419_end_0 = const()[name = string("op_35419_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_35419_end_mask_0 = const()[name = string("op_35419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35419_cast_fp16 = slice_by_index(begin = var_35419_begin_0, end = var_35419_end_0, end_mask = var_35419_end_mask_0, x = query_47_cast_fp16)[name = string("op_35419_cast_fp16")]; + tensor var_35423_begin_0 = const()[name = string("op_35423_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35423_end_0 = const()[name = string("op_35423_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_35423_end_mask_0 = const()[name = string("op_35423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35423_cast_fp16 = slice_by_index(begin = var_35423_begin_0, end = var_35423_end_0, end_mask = var_35423_end_mask_0, x = query_47_cast_fp16)[name = string("op_35423_cast_fp16")]; + tensor var_35427_begin_0 = const()[name = string("op_35427_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35427_end_0 = const()[name = string("op_35427_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_35427_end_mask_0 = const()[name = string("op_35427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35427_cast_fp16 = slice_by_index(begin = var_35427_begin_0, end = var_35427_end_0, end_mask = var_35427_end_mask_0, x = query_47_cast_fp16)[name = string("op_35427_cast_fp16")]; + tensor var_35431_begin_0 = const()[name = string("op_35431_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35431_end_0 = const()[name = string("op_35431_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_35431_end_mask_0 = const()[name = string("op_35431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35431_cast_fp16 = slice_by_index(begin = var_35431_begin_0, end = var_35431_end_0, end_mask = var_35431_end_mask_0, x = query_47_cast_fp16)[name = string("op_35431_cast_fp16")]; + tensor var_35435_begin_0 = const()[name = string("op_35435_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35435_end_0 = const()[name = string("op_35435_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_35435_end_mask_0 = const()[name = string("op_35435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35435_cast_fp16 = slice_by_index(begin = var_35435_begin_0, end = var_35435_end_0, end_mask = var_35435_end_mask_0, x = query_47_cast_fp16)[name = string("op_35435_cast_fp16")]; + tensor var_35439_begin_0 = const()[name = string("op_35439_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35439_end_0 = const()[name = string("op_35439_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_35439_end_mask_0 = const()[name = string("op_35439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35439_cast_fp16 = slice_by_index(begin = var_35439_begin_0, end = var_35439_end_0, end_mask = var_35439_end_mask_0, x = query_47_cast_fp16)[name = string("op_35439_cast_fp16")]; + tensor var_35443_begin_0 = const()[name = string("op_35443_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35443_end_0 = const()[name = string("op_35443_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_35443_end_mask_0 = const()[name = string("op_35443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35443_cast_fp16 = slice_by_index(begin = var_35443_begin_0, end = var_35443_end_0, end_mask = var_35443_end_mask_0, x = query_47_cast_fp16)[name = string("op_35443_cast_fp16")]; + tensor var_35447_begin_0 = const()[name = string("op_35447_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35447_end_0 = const()[name = string("op_35447_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_35447_end_mask_0 = const()[name = string("op_35447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35447_cast_fp16 = slice_by_index(begin = var_35447_begin_0, end = var_35447_end_0, end_mask = var_35447_end_mask_0, x = query_47_cast_fp16)[name = string("op_35447_cast_fp16")]; + tensor var_35451_begin_0 = const()[name = string("op_35451_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35451_end_0 = const()[name = string("op_35451_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_35451_end_mask_0 = const()[name = string("op_35451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35451_cast_fp16 = slice_by_index(begin = var_35451_begin_0, end = var_35451_end_0, end_mask = var_35451_end_mask_0, x = query_47_cast_fp16)[name = string("op_35451_cast_fp16")]; + tensor var_35455_begin_0 = const()[name = string("op_35455_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35455_end_0 = const()[name = string("op_35455_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_35455_end_mask_0 = const()[name = string("op_35455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35455_cast_fp16 = slice_by_index(begin = var_35455_begin_0, end = var_35455_end_0, end_mask = var_35455_end_mask_0, x = query_47_cast_fp16)[name = string("op_35455_cast_fp16")]; + tensor var_35464_begin_0 = const()[name = string("op_35464_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35464_end_0 = const()[name = string("op_35464_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35464_end_mask_0 = const()[name = string("op_35464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35464_cast_fp16 = slice_by_index(begin = var_35464_begin_0, end = var_35464_end_0, end_mask = var_35464_end_mask_0, x = var_35379_cast_fp16)[name = string("op_35464_cast_fp16")]; + tensor var_35471_begin_0 = const()[name = string("op_35471_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35471_end_0 = const()[name = string("op_35471_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35471_end_mask_0 = const()[name = string("op_35471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35471_cast_fp16 = slice_by_index(begin = var_35471_begin_0, end = var_35471_end_0, end_mask = var_35471_end_mask_0, x = var_35379_cast_fp16)[name = string("op_35471_cast_fp16")]; + tensor var_35478_begin_0 = const()[name = string("op_35478_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35478_end_0 = const()[name = string("op_35478_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35478_end_mask_0 = const()[name = string("op_35478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35478_cast_fp16 = slice_by_index(begin = var_35478_begin_0, end = var_35478_end_0, end_mask = var_35478_end_mask_0, x = var_35379_cast_fp16)[name = string("op_35478_cast_fp16")]; + tensor var_35485_begin_0 = const()[name = string("op_35485_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35485_end_0 = const()[name = string("op_35485_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35485_end_mask_0 = const()[name = string("op_35485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35485_cast_fp16 = slice_by_index(begin = var_35485_begin_0, end = var_35485_end_0, end_mask = var_35485_end_mask_0, x = var_35379_cast_fp16)[name = string("op_35485_cast_fp16")]; + tensor var_35492_begin_0 = const()[name = string("op_35492_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35492_end_0 = const()[name = string("op_35492_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35492_end_mask_0 = const()[name = string("op_35492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35492_cast_fp16 = slice_by_index(begin = var_35492_begin_0, end = var_35492_end_0, end_mask = var_35492_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35492_cast_fp16")]; + tensor var_35499_begin_0 = const()[name = string("op_35499_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35499_end_0 = const()[name = string("op_35499_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35499_end_mask_0 = const()[name = string("op_35499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35499_cast_fp16 = slice_by_index(begin = var_35499_begin_0, end = var_35499_end_0, end_mask = var_35499_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35499_cast_fp16")]; + tensor var_35506_begin_0 = const()[name = string("op_35506_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35506_end_0 = const()[name = string("op_35506_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35506_end_mask_0 = const()[name = string("op_35506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35506_cast_fp16 = slice_by_index(begin = var_35506_begin_0, end = var_35506_end_0, end_mask = var_35506_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35506_cast_fp16")]; + tensor var_35513_begin_0 = const()[name = string("op_35513_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35513_end_0 = const()[name = string("op_35513_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35513_end_mask_0 = const()[name = string("op_35513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35513_cast_fp16 = slice_by_index(begin = var_35513_begin_0, end = var_35513_end_0, end_mask = var_35513_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35513_cast_fp16")]; + tensor var_35520_begin_0 = const()[name = string("op_35520_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35520_end_0 = const()[name = string("op_35520_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35520_end_mask_0 = const()[name = string("op_35520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35520_cast_fp16 = slice_by_index(begin = var_35520_begin_0, end = var_35520_end_0, end_mask = var_35520_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35520_cast_fp16")]; + tensor var_35527_begin_0 = const()[name = string("op_35527_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35527_end_0 = const()[name = string("op_35527_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35527_end_mask_0 = const()[name = string("op_35527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35527_cast_fp16 = slice_by_index(begin = var_35527_begin_0, end = var_35527_end_0, end_mask = var_35527_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35527_cast_fp16")]; + tensor var_35534_begin_0 = const()[name = string("op_35534_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35534_end_0 = const()[name = string("op_35534_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35534_end_mask_0 = const()[name = string("op_35534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35534_cast_fp16 = slice_by_index(begin = var_35534_begin_0, end = var_35534_end_0, end_mask = var_35534_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35534_cast_fp16")]; + tensor var_35541_begin_0 = const()[name = string("op_35541_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35541_end_0 = const()[name = string("op_35541_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35541_end_mask_0 = const()[name = string("op_35541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35541_cast_fp16 = slice_by_index(begin = var_35541_begin_0, end = var_35541_end_0, end_mask = var_35541_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35541_cast_fp16")]; + tensor var_35548_begin_0 = const()[name = string("op_35548_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35548_end_0 = const()[name = string("op_35548_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35548_end_mask_0 = const()[name = string("op_35548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35548_cast_fp16 = slice_by_index(begin = var_35548_begin_0, end = var_35548_end_0, end_mask = var_35548_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35548_cast_fp16")]; + tensor var_35555_begin_0 = const()[name = string("op_35555_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35555_end_0 = const()[name = string("op_35555_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35555_end_mask_0 = const()[name = string("op_35555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35555_cast_fp16 = slice_by_index(begin = var_35555_begin_0, end = var_35555_end_0, end_mask = var_35555_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35555_cast_fp16")]; + tensor var_35562_begin_0 = const()[name = string("op_35562_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35562_end_0 = const()[name = string("op_35562_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35562_end_mask_0 = const()[name = string("op_35562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35562_cast_fp16 = slice_by_index(begin = var_35562_begin_0, end = var_35562_end_0, end_mask = var_35562_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35562_cast_fp16")]; + tensor var_35569_begin_0 = const()[name = string("op_35569_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35569_end_0 = const()[name = string("op_35569_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35569_end_mask_0 = const()[name = string("op_35569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35569_cast_fp16 = slice_by_index(begin = var_35569_begin_0, end = var_35569_end_0, end_mask = var_35569_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35569_cast_fp16")]; + tensor var_35576_begin_0 = const()[name = string("op_35576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35576_end_0 = const()[name = string("op_35576_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35576_end_mask_0 = const()[name = string("op_35576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35576_cast_fp16 = slice_by_index(begin = var_35576_begin_0, end = var_35576_end_0, end_mask = var_35576_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35576_cast_fp16")]; + tensor var_35583_begin_0 = const()[name = string("op_35583_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35583_end_0 = const()[name = string("op_35583_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35583_end_mask_0 = const()[name = string("op_35583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35583_cast_fp16 = slice_by_index(begin = var_35583_begin_0, end = var_35583_end_0, end_mask = var_35583_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35583_cast_fp16")]; + tensor var_35590_begin_0 = const()[name = string("op_35590_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35590_end_0 = const()[name = string("op_35590_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35590_end_mask_0 = const()[name = string("op_35590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35590_cast_fp16 = slice_by_index(begin = var_35590_begin_0, end = var_35590_end_0, end_mask = var_35590_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35590_cast_fp16")]; + tensor var_35597_begin_0 = const()[name = string("op_35597_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35597_end_0 = const()[name = string("op_35597_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35597_end_mask_0 = const()[name = string("op_35597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35597_cast_fp16 = slice_by_index(begin = var_35597_begin_0, end = var_35597_end_0, end_mask = var_35597_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35597_cast_fp16")]; + tensor var_35604_begin_0 = const()[name = string("op_35604_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35604_end_0 = const()[name = string("op_35604_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35604_end_mask_0 = const()[name = string("op_35604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35604_cast_fp16 = slice_by_index(begin = var_35604_begin_0, end = var_35604_end_0, end_mask = var_35604_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35604_cast_fp16")]; + tensor var_35611_begin_0 = const()[name = string("op_35611_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35611_end_0 = const()[name = string("op_35611_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35611_end_mask_0 = const()[name = string("op_35611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35611_cast_fp16 = slice_by_index(begin = var_35611_begin_0, end = var_35611_end_0, end_mask = var_35611_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35611_cast_fp16")]; + tensor var_35618_begin_0 = const()[name = string("op_35618_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35618_end_0 = const()[name = string("op_35618_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35618_end_mask_0 = const()[name = string("op_35618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35618_cast_fp16 = slice_by_index(begin = var_35618_begin_0, end = var_35618_end_0, end_mask = var_35618_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35618_cast_fp16")]; + tensor var_35625_begin_0 = const()[name = string("op_35625_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35625_end_0 = const()[name = string("op_35625_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35625_end_mask_0 = const()[name = string("op_35625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35625_cast_fp16 = slice_by_index(begin = var_35625_begin_0, end = var_35625_end_0, end_mask = var_35625_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35625_cast_fp16")]; + tensor var_35632_begin_0 = const()[name = string("op_35632_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35632_end_0 = const()[name = string("op_35632_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35632_end_mask_0 = const()[name = string("op_35632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35632_cast_fp16 = slice_by_index(begin = var_35632_begin_0, end = var_35632_end_0, end_mask = var_35632_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35632_cast_fp16")]; + tensor var_35639_begin_0 = const()[name = string("op_35639_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35639_end_0 = const()[name = string("op_35639_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35639_end_mask_0 = const()[name = string("op_35639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35639_cast_fp16 = slice_by_index(begin = var_35639_begin_0, end = var_35639_end_0, end_mask = var_35639_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35639_cast_fp16")]; + tensor var_35646_begin_0 = const()[name = string("op_35646_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35646_end_0 = const()[name = string("op_35646_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35646_end_mask_0 = const()[name = string("op_35646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35646_cast_fp16 = slice_by_index(begin = var_35646_begin_0, end = var_35646_end_0, end_mask = var_35646_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35646_cast_fp16")]; + tensor var_35653_begin_0 = const()[name = string("op_35653_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35653_end_0 = const()[name = string("op_35653_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35653_end_mask_0 = const()[name = string("op_35653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35653_cast_fp16 = slice_by_index(begin = var_35653_begin_0, end = var_35653_end_0, end_mask = var_35653_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35653_cast_fp16")]; + tensor var_35660_begin_0 = const()[name = string("op_35660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35660_end_0 = const()[name = string("op_35660_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35660_end_mask_0 = const()[name = string("op_35660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35660_cast_fp16 = slice_by_index(begin = var_35660_begin_0, end = var_35660_end_0, end_mask = var_35660_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35660_cast_fp16")]; + tensor var_35667_begin_0 = const()[name = string("op_35667_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35667_end_0 = const()[name = string("op_35667_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35667_end_mask_0 = const()[name = string("op_35667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35667_cast_fp16 = slice_by_index(begin = var_35667_begin_0, end = var_35667_end_0, end_mask = var_35667_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35667_cast_fp16")]; + tensor var_35674_begin_0 = const()[name = string("op_35674_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35674_end_0 = const()[name = string("op_35674_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35674_end_mask_0 = const()[name = string("op_35674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35674_cast_fp16 = slice_by_index(begin = var_35674_begin_0, end = var_35674_end_0, end_mask = var_35674_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35674_cast_fp16")]; + tensor var_35681_begin_0 = const()[name = string("op_35681_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35681_end_0 = const()[name = string("op_35681_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35681_end_mask_0 = const()[name = string("op_35681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35681_cast_fp16 = slice_by_index(begin = var_35681_begin_0, end = var_35681_end_0, end_mask = var_35681_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35681_cast_fp16")]; + tensor var_35688_begin_0 = const()[name = string("op_35688_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35688_end_0 = const()[name = string("op_35688_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35688_end_mask_0 = const()[name = string("op_35688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35688_cast_fp16 = slice_by_index(begin = var_35688_begin_0, end = var_35688_end_0, end_mask = var_35688_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35688_cast_fp16")]; + tensor var_35695_begin_0 = const()[name = string("op_35695_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35695_end_0 = const()[name = string("op_35695_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35695_end_mask_0 = const()[name = string("op_35695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35695_cast_fp16 = slice_by_index(begin = var_35695_begin_0, end = var_35695_end_0, end_mask = var_35695_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35695_cast_fp16")]; + tensor var_35702_begin_0 = const()[name = string("op_35702_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35702_end_0 = const()[name = string("op_35702_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35702_end_mask_0 = const()[name = string("op_35702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35702_cast_fp16 = slice_by_index(begin = var_35702_begin_0, end = var_35702_end_0, end_mask = var_35702_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35702_cast_fp16")]; + tensor var_35709_begin_0 = const()[name = string("op_35709_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35709_end_0 = const()[name = string("op_35709_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35709_end_mask_0 = const()[name = string("op_35709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35709_cast_fp16 = slice_by_index(begin = var_35709_begin_0, end = var_35709_end_0, end_mask = var_35709_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35709_cast_fp16")]; + tensor var_35716_begin_0 = const()[name = string("op_35716_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35716_end_0 = const()[name = string("op_35716_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35716_end_mask_0 = const()[name = string("op_35716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35716_cast_fp16 = slice_by_index(begin = var_35716_begin_0, end = var_35716_end_0, end_mask = var_35716_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35716_cast_fp16")]; + tensor var_35723_begin_0 = const()[name = string("op_35723_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35723_end_0 = const()[name = string("op_35723_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35723_end_mask_0 = const()[name = string("op_35723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35723_cast_fp16 = slice_by_index(begin = var_35723_begin_0, end = var_35723_end_0, end_mask = var_35723_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35723_cast_fp16")]; + tensor var_35730_begin_0 = const()[name = string("op_35730_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35730_end_0 = const()[name = string("op_35730_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35730_end_mask_0 = const()[name = string("op_35730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35730_cast_fp16 = slice_by_index(begin = var_35730_begin_0, end = var_35730_end_0, end_mask = var_35730_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35730_cast_fp16")]; + tensor var_35737_begin_0 = const()[name = string("op_35737_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35737_end_0 = const()[name = string("op_35737_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35737_end_mask_0 = const()[name = string("op_35737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35737_cast_fp16 = slice_by_index(begin = var_35737_begin_0, end = var_35737_end_0, end_mask = var_35737_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35737_cast_fp16")]; + tensor var_35744_begin_0 = const()[name = string("op_35744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35744_end_0 = const()[name = string("op_35744_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35744_end_mask_0 = const()[name = string("op_35744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35744_cast_fp16 = slice_by_index(begin = var_35744_begin_0, end = var_35744_end_0, end_mask = var_35744_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35744_cast_fp16")]; + tensor var_35751_begin_0 = const()[name = string("op_35751_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35751_end_0 = const()[name = string("op_35751_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35751_end_mask_0 = const()[name = string("op_35751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35751_cast_fp16 = slice_by_index(begin = var_35751_begin_0, end = var_35751_end_0, end_mask = var_35751_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35751_cast_fp16")]; + tensor var_35758_begin_0 = const()[name = string("op_35758_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35758_end_0 = const()[name = string("op_35758_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35758_end_mask_0 = const()[name = string("op_35758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35758_cast_fp16 = slice_by_index(begin = var_35758_begin_0, end = var_35758_end_0, end_mask = var_35758_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35758_cast_fp16")]; + tensor var_35765_begin_0 = const()[name = string("op_35765_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35765_end_0 = const()[name = string("op_35765_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35765_end_mask_0 = const()[name = string("op_35765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35765_cast_fp16 = slice_by_index(begin = var_35765_begin_0, end = var_35765_end_0, end_mask = var_35765_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35765_cast_fp16")]; + tensor var_35772_begin_0 = const()[name = string("op_35772_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35772_end_0 = const()[name = string("op_35772_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35772_end_mask_0 = const()[name = string("op_35772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35772_cast_fp16 = slice_by_index(begin = var_35772_begin_0, end = var_35772_end_0, end_mask = var_35772_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35772_cast_fp16")]; + tensor var_35779_begin_0 = const()[name = string("op_35779_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35779_end_0 = const()[name = string("op_35779_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35779_end_mask_0 = const()[name = string("op_35779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35779_cast_fp16 = slice_by_index(begin = var_35779_begin_0, end = var_35779_end_0, end_mask = var_35779_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35779_cast_fp16")]; + tensor var_35786_begin_0 = const()[name = string("op_35786_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35786_end_0 = const()[name = string("op_35786_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35786_end_mask_0 = const()[name = string("op_35786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35786_cast_fp16 = slice_by_index(begin = var_35786_begin_0, end = var_35786_end_0, end_mask = var_35786_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35786_cast_fp16")]; + tensor var_35793_begin_0 = const()[name = string("op_35793_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35793_end_0 = const()[name = string("op_35793_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35793_end_mask_0 = const()[name = string("op_35793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35793_cast_fp16 = slice_by_index(begin = var_35793_begin_0, end = var_35793_end_0, end_mask = var_35793_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35793_cast_fp16")]; + tensor var_35800_begin_0 = const()[name = string("op_35800_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35800_end_0 = const()[name = string("op_35800_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35800_end_mask_0 = const()[name = string("op_35800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35800_cast_fp16 = slice_by_index(begin = var_35800_begin_0, end = var_35800_end_0, end_mask = var_35800_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35800_cast_fp16")]; + tensor var_35807_begin_0 = const()[name = string("op_35807_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35807_end_0 = const()[name = string("op_35807_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35807_end_mask_0 = const()[name = string("op_35807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35807_cast_fp16 = slice_by_index(begin = var_35807_begin_0, end = var_35807_end_0, end_mask = var_35807_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35807_cast_fp16")]; + tensor var_35814_begin_0 = const()[name = string("op_35814_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35814_end_0 = const()[name = string("op_35814_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35814_end_mask_0 = const()[name = string("op_35814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35814_cast_fp16 = slice_by_index(begin = var_35814_begin_0, end = var_35814_end_0, end_mask = var_35814_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35814_cast_fp16")]; + tensor var_35821_begin_0 = const()[name = string("op_35821_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35821_end_0 = const()[name = string("op_35821_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35821_end_mask_0 = const()[name = string("op_35821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35821_cast_fp16 = slice_by_index(begin = var_35821_begin_0, end = var_35821_end_0, end_mask = var_35821_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35821_cast_fp16")]; + tensor var_35828_begin_0 = const()[name = string("op_35828_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35828_end_0 = const()[name = string("op_35828_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35828_end_mask_0 = const()[name = string("op_35828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35828_cast_fp16 = slice_by_index(begin = var_35828_begin_0, end = var_35828_end_0, end_mask = var_35828_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35828_cast_fp16")]; + tensor var_35835_begin_0 = const()[name = string("op_35835_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35835_end_0 = const()[name = string("op_35835_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35835_end_mask_0 = const()[name = string("op_35835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35835_cast_fp16 = slice_by_index(begin = var_35835_begin_0, end = var_35835_end_0, end_mask = var_35835_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35835_cast_fp16")]; + tensor var_35842_begin_0 = const()[name = string("op_35842_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35842_end_0 = const()[name = string("op_35842_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35842_end_mask_0 = const()[name = string("op_35842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35842_cast_fp16 = slice_by_index(begin = var_35842_begin_0, end = var_35842_end_0, end_mask = var_35842_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35842_cast_fp16")]; + tensor var_35849_begin_0 = const()[name = string("op_35849_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35849_end_0 = const()[name = string("op_35849_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35849_end_mask_0 = const()[name = string("op_35849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35849_cast_fp16 = slice_by_index(begin = var_35849_begin_0, end = var_35849_end_0, end_mask = var_35849_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35849_cast_fp16")]; + tensor var_35856_begin_0 = const()[name = string("op_35856_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35856_end_0 = const()[name = string("op_35856_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35856_end_mask_0 = const()[name = string("op_35856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35856_cast_fp16 = slice_by_index(begin = var_35856_begin_0, end = var_35856_end_0, end_mask = var_35856_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35856_cast_fp16")]; + tensor var_35863_begin_0 = const()[name = string("op_35863_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35863_end_0 = const()[name = string("op_35863_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35863_end_mask_0 = const()[name = string("op_35863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35863_cast_fp16 = slice_by_index(begin = var_35863_begin_0, end = var_35863_end_0, end_mask = var_35863_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35863_cast_fp16")]; + tensor var_35870_begin_0 = const()[name = string("op_35870_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35870_end_0 = const()[name = string("op_35870_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35870_end_mask_0 = const()[name = string("op_35870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35870_cast_fp16 = slice_by_index(begin = var_35870_begin_0, end = var_35870_end_0, end_mask = var_35870_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35870_cast_fp16")]; + tensor var_35877_begin_0 = const()[name = string("op_35877_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35877_end_0 = const()[name = string("op_35877_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35877_end_mask_0 = const()[name = string("op_35877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35877_cast_fp16 = slice_by_index(begin = var_35877_begin_0, end = var_35877_end_0, end_mask = var_35877_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35877_cast_fp16")]; + tensor var_35884_begin_0 = const()[name = string("op_35884_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35884_end_0 = const()[name = string("op_35884_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35884_end_mask_0 = const()[name = string("op_35884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35884_cast_fp16 = slice_by_index(begin = var_35884_begin_0, end = var_35884_end_0, end_mask = var_35884_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35884_cast_fp16")]; + tensor var_35891_begin_0 = const()[name = string("op_35891_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35891_end_0 = const()[name = string("op_35891_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35891_end_mask_0 = const()[name = string("op_35891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35891_cast_fp16 = slice_by_index(begin = var_35891_begin_0, end = var_35891_end_0, end_mask = var_35891_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35891_cast_fp16")]; + tensor var_35898_begin_0 = const()[name = string("op_35898_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35898_end_0 = const()[name = string("op_35898_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35898_end_mask_0 = const()[name = string("op_35898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35898_cast_fp16 = slice_by_index(begin = var_35898_begin_0, end = var_35898_end_0, end_mask = var_35898_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35898_cast_fp16")]; + tensor var_35905_begin_0 = const()[name = string("op_35905_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35905_end_0 = const()[name = string("op_35905_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35905_end_mask_0 = const()[name = string("op_35905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35905_cast_fp16 = slice_by_index(begin = var_35905_begin_0, end = var_35905_end_0, end_mask = var_35905_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35905_cast_fp16")]; + tensor var_35912_begin_0 = const()[name = string("op_35912_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35912_end_0 = const()[name = string("op_35912_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35912_end_mask_0 = const()[name = string("op_35912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35912_cast_fp16 = slice_by_index(begin = var_35912_begin_0, end = var_35912_end_0, end_mask = var_35912_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35912_cast_fp16")]; + tensor var_35919_begin_0 = const()[name = string("op_35919_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35919_end_0 = const()[name = string("op_35919_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35919_end_mask_0 = const()[name = string("op_35919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35919_cast_fp16 = slice_by_index(begin = var_35919_begin_0, end = var_35919_end_0, end_mask = var_35919_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35919_cast_fp16")]; + tensor var_35926_begin_0 = const()[name = string("op_35926_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35926_end_0 = const()[name = string("op_35926_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35926_end_mask_0 = const()[name = string("op_35926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35926_cast_fp16 = slice_by_index(begin = var_35926_begin_0, end = var_35926_end_0, end_mask = var_35926_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35926_cast_fp16")]; + tensor var_35933_begin_0 = const()[name = string("op_35933_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35933_end_0 = const()[name = string("op_35933_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35933_end_mask_0 = const()[name = string("op_35933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35933_cast_fp16 = slice_by_index(begin = var_35933_begin_0, end = var_35933_end_0, end_mask = var_35933_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35933_cast_fp16")]; + tensor var_35940_begin_0 = const()[name = string("op_35940_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35940_end_0 = const()[name = string("op_35940_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35940_end_mask_0 = const()[name = string("op_35940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35940_cast_fp16 = slice_by_index(begin = var_35940_begin_0, end = var_35940_end_0, end_mask = var_35940_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35940_cast_fp16")]; + tensor var_35947_begin_0 = const()[name = string("op_35947_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35947_end_0 = const()[name = string("op_35947_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35947_end_mask_0 = const()[name = string("op_35947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35947_cast_fp16 = slice_by_index(begin = var_35947_begin_0, end = var_35947_end_0, end_mask = var_35947_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35947_cast_fp16")]; + tensor var_35954_begin_0 = const()[name = string("op_35954_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35954_end_0 = const()[name = string("op_35954_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35954_end_mask_0 = const()[name = string("op_35954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35954_cast_fp16 = slice_by_index(begin = var_35954_begin_0, end = var_35954_end_0, end_mask = var_35954_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35954_cast_fp16")]; + tensor var_35961_begin_0 = const()[name = string("op_35961_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35961_end_0 = const()[name = string("op_35961_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35961_end_mask_0 = const()[name = string("op_35961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35961_cast_fp16 = slice_by_index(begin = var_35961_begin_0, end = var_35961_end_0, end_mask = var_35961_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35961_cast_fp16")]; + tensor var_35968_begin_0 = const()[name = string("op_35968_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35968_end_0 = const()[name = string("op_35968_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35968_end_mask_0 = const()[name = string("op_35968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35968_cast_fp16 = slice_by_index(begin = var_35968_begin_0, end = var_35968_end_0, end_mask = var_35968_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35968_cast_fp16")]; + tensor var_35975_begin_0 = const()[name = string("op_35975_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35975_end_0 = const()[name = string("op_35975_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35975_end_mask_0 = const()[name = string("op_35975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35975_cast_fp16 = slice_by_index(begin = var_35975_begin_0, end = var_35975_end_0, end_mask = var_35975_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35975_cast_fp16")]; + tensor var_35982_begin_0 = const()[name = string("op_35982_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35982_end_0 = const()[name = string("op_35982_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35982_end_mask_0 = const()[name = string("op_35982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35982_cast_fp16 = slice_by_index(begin = var_35982_begin_0, end = var_35982_end_0, end_mask = var_35982_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35982_cast_fp16")]; + tensor var_35989_begin_0 = const()[name = string("op_35989_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35989_end_0 = const()[name = string("op_35989_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35989_end_mask_0 = const()[name = string("op_35989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35989_cast_fp16 = slice_by_index(begin = var_35989_begin_0, end = var_35989_end_0, end_mask = var_35989_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35989_cast_fp16")]; + tensor var_35996_begin_0 = const()[name = string("op_35996_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35996_end_0 = const()[name = string("op_35996_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35996_end_mask_0 = const()[name = string("op_35996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35996_cast_fp16 = slice_by_index(begin = var_35996_begin_0, end = var_35996_end_0, end_mask = var_35996_end_mask_0, x = var_35455_cast_fp16)[name = string("op_35996_cast_fp16")]; + tensor var_36003_begin_0 = const()[name = string("op_36003_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36003_end_0 = const()[name = string("op_36003_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36003_end_mask_0 = const()[name = string("op_36003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36003_cast_fp16 = slice_by_index(begin = var_36003_begin_0, end = var_36003_end_0, end_mask = var_36003_end_mask_0, x = var_35455_cast_fp16)[name = string("op_36003_cast_fp16")]; + tensor var_36010_begin_0 = const()[name = string("op_36010_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36010_end_0 = const()[name = string("op_36010_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36010_end_mask_0 = const()[name = string("op_36010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36010_cast_fp16 = slice_by_index(begin = var_36010_begin_0, end = var_36010_end_0, end_mask = var_36010_end_mask_0, x = var_35455_cast_fp16)[name = string("op_36010_cast_fp16")]; + tensor var_36017_begin_0 = const()[name = string("op_36017_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36017_end_0 = const()[name = string("op_36017_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36017_end_mask_0 = const()[name = string("op_36017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36017_cast_fp16 = slice_by_index(begin = var_36017_begin_0, end = var_36017_end_0, end_mask = var_36017_end_mask_0, x = var_35455_cast_fp16)[name = string("op_36017_cast_fp16")]; + tensor k_47_perm_0 = const()[name = string("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_36022_begin_0 = const()[name = string("op_36022_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36022_end_0 = const()[name = string("op_36022_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_36022_end_mask_0 = const()[name = string("op_36022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_47_cast_fp16 = transpose(perm = k_47_perm_0, x = key_47_cast_fp16)[name = string("transpose_8")]; + tensor var_36022_cast_fp16 = slice_by_index(begin = var_36022_begin_0, end = var_36022_end_0, end_mask = var_36022_end_mask_0, x = k_47_cast_fp16)[name = string("op_36022_cast_fp16")]; + tensor var_36026_begin_0 = const()[name = string("op_36026_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_36026_end_0 = const()[name = string("op_36026_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_36026_end_mask_0 = const()[name = string("op_36026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36026_cast_fp16 = slice_by_index(begin = var_36026_begin_0, end = var_36026_end_0, end_mask = var_36026_end_mask_0, x = k_47_cast_fp16)[name = string("op_36026_cast_fp16")]; + tensor var_36030_begin_0 = const()[name = string("op_36030_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_36030_end_0 = const()[name = string("op_36030_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_36030_end_mask_0 = const()[name = string("op_36030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36030_cast_fp16 = slice_by_index(begin = var_36030_begin_0, end = var_36030_end_0, end_mask = var_36030_end_mask_0, x = k_47_cast_fp16)[name = string("op_36030_cast_fp16")]; + tensor var_36034_begin_0 = const()[name = string("op_36034_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_36034_end_0 = const()[name = string("op_36034_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_36034_end_mask_0 = const()[name = string("op_36034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36034_cast_fp16 = slice_by_index(begin = var_36034_begin_0, end = var_36034_end_0, end_mask = var_36034_end_mask_0, x = k_47_cast_fp16)[name = string("op_36034_cast_fp16")]; + tensor var_36038_begin_0 = const()[name = string("op_36038_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_36038_end_0 = const()[name = string("op_36038_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_36038_end_mask_0 = const()[name = string("op_36038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36038_cast_fp16 = slice_by_index(begin = var_36038_begin_0, end = var_36038_end_0, end_mask = var_36038_end_mask_0, x = k_47_cast_fp16)[name = string("op_36038_cast_fp16")]; + tensor var_36042_begin_0 = const()[name = string("op_36042_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_36042_end_0 = const()[name = string("op_36042_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_36042_end_mask_0 = const()[name = string("op_36042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36042_cast_fp16 = slice_by_index(begin = var_36042_begin_0, end = var_36042_end_0, end_mask = var_36042_end_mask_0, x = k_47_cast_fp16)[name = string("op_36042_cast_fp16")]; + tensor var_36046_begin_0 = const()[name = string("op_36046_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_36046_end_0 = const()[name = string("op_36046_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_36046_end_mask_0 = const()[name = string("op_36046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36046_cast_fp16 = slice_by_index(begin = var_36046_begin_0, end = var_36046_end_0, end_mask = var_36046_end_mask_0, x = k_47_cast_fp16)[name = string("op_36046_cast_fp16")]; + tensor var_36050_begin_0 = const()[name = string("op_36050_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_36050_end_0 = const()[name = string("op_36050_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_36050_end_mask_0 = const()[name = string("op_36050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36050_cast_fp16 = slice_by_index(begin = var_36050_begin_0, end = var_36050_end_0, end_mask = var_36050_end_mask_0, x = k_47_cast_fp16)[name = string("op_36050_cast_fp16")]; + tensor var_36054_begin_0 = const()[name = string("op_36054_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_36054_end_0 = const()[name = string("op_36054_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_36054_end_mask_0 = const()[name = string("op_36054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36054_cast_fp16 = slice_by_index(begin = var_36054_begin_0, end = var_36054_end_0, end_mask = var_36054_end_mask_0, x = k_47_cast_fp16)[name = string("op_36054_cast_fp16")]; + tensor var_36058_begin_0 = const()[name = string("op_36058_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_36058_end_0 = const()[name = string("op_36058_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_36058_end_mask_0 = const()[name = string("op_36058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36058_cast_fp16 = slice_by_index(begin = var_36058_begin_0, end = var_36058_end_0, end_mask = var_36058_end_mask_0, x = k_47_cast_fp16)[name = string("op_36058_cast_fp16")]; + tensor var_36062_begin_0 = const()[name = string("op_36062_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_36062_end_0 = const()[name = string("op_36062_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_36062_end_mask_0 = const()[name = string("op_36062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36062_cast_fp16 = slice_by_index(begin = var_36062_begin_0, end = var_36062_end_0, end_mask = var_36062_end_mask_0, x = k_47_cast_fp16)[name = string("op_36062_cast_fp16")]; + tensor var_36066_begin_0 = const()[name = string("op_36066_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_36066_end_0 = const()[name = string("op_36066_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_36066_end_mask_0 = const()[name = string("op_36066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36066_cast_fp16 = slice_by_index(begin = var_36066_begin_0, end = var_36066_end_0, end_mask = var_36066_end_mask_0, x = k_47_cast_fp16)[name = string("op_36066_cast_fp16")]; + tensor var_36070_begin_0 = const()[name = string("op_36070_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_36070_end_0 = const()[name = string("op_36070_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_36070_end_mask_0 = const()[name = string("op_36070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36070_cast_fp16 = slice_by_index(begin = var_36070_begin_0, end = var_36070_end_0, end_mask = var_36070_end_mask_0, x = k_47_cast_fp16)[name = string("op_36070_cast_fp16")]; + tensor var_36074_begin_0 = const()[name = string("op_36074_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_36074_end_0 = const()[name = string("op_36074_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_36074_end_mask_0 = const()[name = string("op_36074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36074_cast_fp16 = slice_by_index(begin = var_36074_begin_0, end = var_36074_end_0, end_mask = var_36074_end_mask_0, x = k_47_cast_fp16)[name = string("op_36074_cast_fp16")]; + tensor var_36078_begin_0 = const()[name = string("op_36078_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_36078_end_0 = const()[name = string("op_36078_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_36078_end_mask_0 = const()[name = string("op_36078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36078_cast_fp16 = slice_by_index(begin = var_36078_begin_0, end = var_36078_end_0, end_mask = var_36078_end_mask_0, x = k_47_cast_fp16)[name = string("op_36078_cast_fp16")]; + tensor var_36082_begin_0 = const()[name = string("op_36082_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_36082_end_0 = const()[name = string("op_36082_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_36082_end_mask_0 = const()[name = string("op_36082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36082_cast_fp16 = slice_by_index(begin = var_36082_begin_0, end = var_36082_end_0, end_mask = var_36082_end_mask_0, x = k_47_cast_fp16)[name = string("op_36082_cast_fp16")]; + tensor var_36086_begin_0 = const()[name = string("op_36086_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_36086_end_0 = const()[name = string("op_36086_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_36086_end_mask_0 = const()[name = string("op_36086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36086_cast_fp16 = slice_by_index(begin = var_36086_begin_0, end = var_36086_end_0, end_mask = var_36086_end_mask_0, x = k_47_cast_fp16)[name = string("op_36086_cast_fp16")]; + tensor var_36090_begin_0 = const()[name = string("op_36090_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_36090_end_0 = const()[name = string("op_36090_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_36090_end_mask_0 = const()[name = string("op_36090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36090_cast_fp16 = slice_by_index(begin = var_36090_begin_0, end = var_36090_end_0, end_mask = var_36090_end_mask_0, x = k_47_cast_fp16)[name = string("op_36090_cast_fp16")]; + tensor var_36094_begin_0 = const()[name = string("op_36094_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_36094_end_0 = const()[name = string("op_36094_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_36094_end_mask_0 = const()[name = string("op_36094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36094_cast_fp16 = slice_by_index(begin = var_36094_begin_0, end = var_36094_end_0, end_mask = var_36094_end_mask_0, x = k_47_cast_fp16)[name = string("op_36094_cast_fp16")]; + tensor var_36098_begin_0 = const()[name = string("op_36098_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_36098_end_0 = const()[name = string("op_36098_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_36098_end_mask_0 = const()[name = string("op_36098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36098_cast_fp16 = slice_by_index(begin = var_36098_begin_0, end = var_36098_end_0, end_mask = var_36098_end_mask_0, x = k_47_cast_fp16)[name = string("op_36098_cast_fp16")]; + tensor var_36100_begin_0 = const()[name = string("op_36100_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36100_end_0 = const()[name = string("op_36100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36100_end_mask_0 = const()[name = string("op_36100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36100_cast_fp16 = slice_by_index(begin = var_36100_begin_0, end = var_36100_end_0, end_mask = var_36100_end_mask_0, x = value_47_cast_fp16)[name = string("op_36100_cast_fp16")]; + tensor var_36104_begin_0 = const()[name = string("op_36104_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36104_end_0 = const()[name = string("op_36104_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_36104_end_mask_0 = const()[name = string("op_36104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36104_cast_fp16 = slice_by_index(begin = var_36104_begin_0, end = var_36104_end_0, end_mask = var_36104_end_mask_0, x = value_47_cast_fp16)[name = string("op_36104_cast_fp16")]; + tensor var_36108_begin_0 = const()[name = string("op_36108_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36108_end_0 = const()[name = string("op_36108_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_36108_end_mask_0 = const()[name = string("op_36108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36108_cast_fp16 = slice_by_index(begin = var_36108_begin_0, end = var_36108_end_0, end_mask = var_36108_end_mask_0, x = value_47_cast_fp16)[name = string("op_36108_cast_fp16")]; + tensor var_36112_begin_0 = const()[name = string("op_36112_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36112_end_0 = const()[name = string("op_36112_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_36112_end_mask_0 = const()[name = string("op_36112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36112_cast_fp16 = slice_by_index(begin = var_36112_begin_0, end = var_36112_end_0, end_mask = var_36112_end_mask_0, x = value_47_cast_fp16)[name = string("op_36112_cast_fp16")]; + tensor var_36116_begin_0 = const()[name = string("op_36116_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36116_end_0 = const()[name = string("op_36116_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_36116_end_mask_0 = const()[name = string("op_36116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36116_cast_fp16 = slice_by_index(begin = var_36116_begin_0, end = var_36116_end_0, end_mask = var_36116_end_mask_0, x = value_47_cast_fp16)[name = string("op_36116_cast_fp16")]; + tensor var_36120_begin_0 = const()[name = string("op_36120_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36120_end_0 = const()[name = string("op_36120_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_36120_end_mask_0 = const()[name = string("op_36120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36120_cast_fp16 = slice_by_index(begin = var_36120_begin_0, end = var_36120_end_0, end_mask = var_36120_end_mask_0, x = value_47_cast_fp16)[name = string("op_36120_cast_fp16")]; + tensor var_36124_begin_0 = const()[name = string("op_36124_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36124_end_0 = const()[name = string("op_36124_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_36124_end_mask_0 = const()[name = string("op_36124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36124_cast_fp16 = slice_by_index(begin = var_36124_begin_0, end = var_36124_end_0, end_mask = var_36124_end_mask_0, x = value_47_cast_fp16)[name = string("op_36124_cast_fp16")]; + tensor var_36128_begin_0 = const()[name = string("op_36128_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36128_end_0 = const()[name = string("op_36128_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_36128_end_mask_0 = const()[name = string("op_36128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36128_cast_fp16 = slice_by_index(begin = var_36128_begin_0, end = var_36128_end_0, end_mask = var_36128_end_mask_0, x = value_47_cast_fp16)[name = string("op_36128_cast_fp16")]; + tensor var_36132_begin_0 = const()[name = string("op_36132_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36132_end_0 = const()[name = string("op_36132_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_36132_end_mask_0 = const()[name = string("op_36132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36132_cast_fp16 = slice_by_index(begin = var_36132_begin_0, end = var_36132_end_0, end_mask = var_36132_end_mask_0, x = value_47_cast_fp16)[name = string("op_36132_cast_fp16")]; + tensor var_36136_begin_0 = const()[name = string("op_36136_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36136_end_0 = const()[name = string("op_36136_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_36136_end_mask_0 = const()[name = string("op_36136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36136_cast_fp16 = slice_by_index(begin = var_36136_begin_0, end = var_36136_end_0, end_mask = var_36136_end_mask_0, x = value_47_cast_fp16)[name = string("op_36136_cast_fp16")]; + tensor var_36140_begin_0 = const()[name = string("op_36140_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36140_end_0 = const()[name = string("op_36140_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_36140_end_mask_0 = const()[name = string("op_36140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36140_cast_fp16 = slice_by_index(begin = var_36140_begin_0, end = var_36140_end_0, end_mask = var_36140_end_mask_0, x = value_47_cast_fp16)[name = string("op_36140_cast_fp16")]; + tensor var_36144_begin_0 = const()[name = string("op_36144_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36144_end_0 = const()[name = string("op_36144_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_36144_end_mask_0 = const()[name = string("op_36144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36144_cast_fp16 = slice_by_index(begin = var_36144_begin_0, end = var_36144_end_0, end_mask = var_36144_end_mask_0, x = value_47_cast_fp16)[name = string("op_36144_cast_fp16")]; + tensor var_36148_begin_0 = const()[name = string("op_36148_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36148_end_0 = const()[name = string("op_36148_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_36148_end_mask_0 = const()[name = string("op_36148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36148_cast_fp16 = slice_by_index(begin = var_36148_begin_0, end = var_36148_end_0, end_mask = var_36148_end_mask_0, x = value_47_cast_fp16)[name = string("op_36148_cast_fp16")]; + tensor var_36152_begin_0 = const()[name = string("op_36152_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36152_end_0 = const()[name = string("op_36152_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_36152_end_mask_0 = const()[name = string("op_36152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36152_cast_fp16 = slice_by_index(begin = var_36152_begin_0, end = var_36152_end_0, end_mask = var_36152_end_mask_0, x = value_47_cast_fp16)[name = string("op_36152_cast_fp16")]; + tensor var_36156_begin_0 = const()[name = string("op_36156_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36156_end_0 = const()[name = string("op_36156_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_36156_end_mask_0 = const()[name = string("op_36156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36156_cast_fp16 = slice_by_index(begin = var_36156_begin_0, end = var_36156_end_0, end_mask = var_36156_end_mask_0, x = value_47_cast_fp16)[name = string("op_36156_cast_fp16")]; + tensor var_36160_begin_0 = const()[name = string("op_36160_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36160_end_0 = const()[name = string("op_36160_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_36160_end_mask_0 = const()[name = string("op_36160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36160_cast_fp16 = slice_by_index(begin = var_36160_begin_0, end = var_36160_end_0, end_mask = var_36160_end_mask_0, x = value_47_cast_fp16)[name = string("op_36160_cast_fp16")]; + tensor var_36164_begin_0 = const()[name = string("op_36164_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36164_end_0 = const()[name = string("op_36164_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_36164_end_mask_0 = const()[name = string("op_36164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36164_cast_fp16 = slice_by_index(begin = var_36164_begin_0, end = var_36164_end_0, end_mask = var_36164_end_mask_0, x = value_47_cast_fp16)[name = string("op_36164_cast_fp16")]; + tensor var_36168_begin_0 = const()[name = string("op_36168_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36168_end_0 = const()[name = string("op_36168_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_36168_end_mask_0 = const()[name = string("op_36168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36168_cast_fp16 = slice_by_index(begin = var_36168_begin_0, end = var_36168_end_0, end_mask = var_36168_end_mask_0, x = value_47_cast_fp16)[name = string("op_36168_cast_fp16")]; + tensor var_36172_begin_0 = const()[name = string("op_36172_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36172_end_0 = const()[name = string("op_36172_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_36172_end_mask_0 = const()[name = string("op_36172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36172_cast_fp16 = slice_by_index(begin = var_36172_begin_0, end = var_36172_end_0, end_mask = var_36172_end_mask_0, x = value_47_cast_fp16)[name = string("op_36172_cast_fp16")]; + tensor var_36176_begin_0 = const()[name = string("op_36176_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36176_end_0 = const()[name = string("op_36176_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_36176_end_mask_0 = const()[name = string("op_36176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36176_cast_fp16 = slice_by_index(begin = var_36176_begin_0, end = var_36176_end_0, end_mask = var_36176_end_mask_0, x = value_47_cast_fp16)[name = string("op_36176_cast_fp16")]; + string _SplitHeadsQ__mh_w_3681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3681_equation_0, values = (var_36022_cast_fp16, var_35464_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3681_cast_fp16")]; + string _SplitHeadsQ__mh_w_3683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3683_equation_0, values = (var_36022_cast_fp16, var_35471_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3683_cast_fp16")]; + string _SplitHeadsQ__mh_w_3685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3685_equation_0, values = (var_36022_cast_fp16, var_35478_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3685_cast_fp16")]; + string _SplitHeadsQ__mh_w_3687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3687_equation_0, values = (var_36022_cast_fp16, var_35485_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3687_cast_fp16")]; + string _SplitHeadsQ__mh_w_3689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3689_equation_0, values = (var_36026_cast_fp16, var_35492_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3689_cast_fp16")]; + string _SplitHeadsQ__mh_w_3691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3691_equation_0, values = (var_36026_cast_fp16, var_35499_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3691_cast_fp16")]; + string _SplitHeadsQ__mh_w_3693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3693_equation_0, values = (var_36026_cast_fp16, var_35506_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3693_cast_fp16")]; + string _SplitHeadsQ__mh_w_3695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3695_equation_0, values = (var_36026_cast_fp16, var_35513_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3695_cast_fp16")]; + string _SplitHeadsQ__mh_w_3697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3697_equation_0, values = (var_36030_cast_fp16, var_35520_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3697_cast_fp16")]; + string _SplitHeadsQ__mh_w_3699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3699_equation_0, values = (var_36030_cast_fp16, var_35527_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3699_cast_fp16")]; + string _SplitHeadsQ__mh_w_3701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3701_equation_0, values = (var_36030_cast_fp16, var_35534_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3701_cast_fp16")]; + string _SplitHeadsQ__mh_w_3703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3703_equation_0, values = (var_36030_cast_fp16, var_35541_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3703_cast_fp16")]; + string _SplitHeadsQ__mh_w_3705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3705_equation_0, values = (var_36034_cast_fp16, var_35548_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3705_cast_fp16")]; + string _SplitHeadsQ__mh_w_3707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3707_equation_0, values = (var_36034_cast_fp16, var_35555_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3707_cast_fp16")]; + string _SplitHeadsQ__mh_w_3709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3709_equation_0, values = (var_36034_cast_fp16, var_35562_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3709_cast_fp16")]; + string _SplitHeadsQ__mh_w_3711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3711_equation_0, values = (var_36034_cast_fp16, var_35569_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3711_cast_fp16")]; + string _SplitHeadsQ__mh_w_3713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3713_equation_0, values = (var_36038_cast_fp16, var_35576_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3713_cast_fp16")]; + string _SplitHeadsQ__mh_w_3715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3715_equation_0, values = (var_36038_cast_fp16, var_35583_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3715_cast_fp16")]; + string _SplitHeadsQ__mh_w_3717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3717_equation_0, values = (var_36038_cast_fp16, var_35590_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3717_cast_fp16")]; + string _SplitHeadsQ__mh_w_3719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3719_equation_0, values = (var_36038_cast_fp16, var_35597_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3719_cast_fp16")]; + string _SplitHeadsQ__mh_w_3721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3721_equation_0, values = (var_36042_cast_fp16, var_35604_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3721_cast_fp16")]; + string _SplitHeadsQ__mh_w_3723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3723_equation_0, values = (var_36042_cast_fp16, var_35611_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3723_cast_fp16")]; + string _SplitHeadsQ__mh_w_3725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3725_equation_0, values = (var_36042_cast_fp16, var_35618_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3725_cast_fp16")]; + string _SplitHeadsQ__mh_w_3727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3727_equation_0, values = (var_36042_cast_fp16, var_35625_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3727_cast_fp16")]; + string _SplitHeadsQ__mh_w_3729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3729_equation_0, values = (var_36046_cast_fp16, var_35632_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3729_cast_fp16")]; + string _SplitHeadsQ__mh_w_3731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3731_equation_0, values = (var_36046_cast_fp16, var_35639_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3731_cast_fp16")]; + string _SplitHeadsQ__mh_w_3733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3733_equation_0, values = (var_36046_cast_fp16, var_35646_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3733_cast_fp16")]; + string _SplitHeadsQ__mh_w_3735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3735_equation_0, values = (var_36046_cast_fp16, var_35653_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3735_cast_fp16")]; + string _SplitHeadsQ__mh_w_3737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3737_equation_0, values = (var_36050_cast_fp16, var_35660_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3737_cast_fp16")]; + string _SplitHeadsQ__mh_w_3739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3739_equation_0, values = (var_36050_cast_fp16, var_35667_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3739_cast_fp16")]; + string _SplitHeadsQ__mh_w_3741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3741_equation_0, values = (var_36050_cast_fp16, var_35674_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3741_cast_fp16")]; + string _SplitHeadsQ__mh_w_3743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3743_equation_0, values = (var_36050_cast_fp16, var_35681_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3743_cast_fp16")]; + string _SplitHeadsQ__mh_w_3745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3745_equation_0, values = (var_36054_cast_fp16, var_35688_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3745_cast_fp16")]; + string _SplitHeadsQ__mh_w_3747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3747_equation_0, values = (var_36054_cast_fp16, var_35695_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3747_cast_fp16")]; + string _SplitHeadsQ__mh_w_3749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3749_equation_0, values = (var_36054_cast_fp16, var_35702_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3749_cast_fp16")]; + string _SplitHeadsQ__mh_w_3751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3751_equation_0, values = (var_36054_cast_fp16, var_35709_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3751_cast_fp16")]; + string _SplitHeadsQ__mh_w_3753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3753_equation_0, values = (var_36058_cast_fp16, var_35716_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3753_cast_fp16")]; + string _SplitHeadsQ__mh_w_3755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3755_equation_0, values = (var_36058_cast_fp16, var_35723_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3755_cast_fp16")]; + string _SplitHeadsQ__mh_w_3757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3757_equation_0, values = (var_36058_cast_fp16, var_35730_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3757_cast_fp16")]; + string _SplitHeadsQ__mh_w_3759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3759_equation_0, values = (var_36058_cast_fp16, var_35737_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3759_cast_fp16")]; + string _SplitHeadsQ__mh_w_3761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3761_equation_0, values = (var_36062_cast_fp16, var_35744_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3761_cast_fp16")]; + string _SplitHeadsQ__mh_w_3763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3763_equation_0, values = (var_36062_cast_fp16, var_35751_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3763_cast_fp16")]; + string _SplitHeadsQ__mh_w_3765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3765_equation_0, values = (var_36062_cast_fp16, var_35758_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3765_cast_fp16")]; + string _SplitHeadsQ__mh_w_3767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3767_equation_0, values = (var_36062_cast_fp16, var_35765_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3767_cast_fp16")]; + string _SplitHeadsQ__mh_w_3769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3769_equation_0, values = (var_36066_cast_fp16, var_35772_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3769_cast_fp16")]; + string _SplitHeadsQ__mh_w_3771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3771_equation_0, values = (var_36066_cast_fp16, var_35779_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3771_cast_fp16")]; + string _SplitHeadsQ__mh_w_3773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3773_equation_0, values = (var_36066_cast_fp16, var_35786_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3773_cast_fp16")]; + string _SplitHeadsQ__mh_w_3775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3775_equation_0, values = (var_36066_cast_fp16, var_35793_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3775_cast_fp16")]; + string _SplitHeadsQ__mh_w_3777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3777_equation_0, values = (var_36070_cast_fp16, var_35800_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3777_cast_fp16")]; + string _SplitHeadsQ__mh_w_3779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3779_equation_0, values = (var_36070_cast_fp16, var_35807_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3779_cast_fp16")]; + string _SplitHeadsQ__mh_w_3781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3781_equation_0, values = (var_36070_cast_fp16, var_35814_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3781_cast_fp16")]; + string _SplitHeadsQ__mh_w_3783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3783_equation_0, values = (var_36070_cast_fp16, var_35821_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3783_cast_fp16")]; + string _SplitHeadsQ__mh_w_3785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3785_equation_0, values = (var_36074_cast_fp16, var_35828_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3785_cast_fp16")]; + string _SplitHeadsQ__mh_w_3787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3787_equation_0, values = (var_36074_cast_fp16, var_35835_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3787_cast_fp16")]; + string _SplitHeadsQ__mh_w_3789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3789_equation_0, values = (var_36074_cast_fp16, var_35842_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3789_cast_fp16")]; + string _SplitHeadsQ__mh_w_3791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3791_equation_0, values = (var_36074_cast_fp16, var_35849_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3791_cast_fp16")]; + string _SplitHeadsQ__mh_w_3793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3793_equation_0, values = (var_36078_cast_fp16, var_35856_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3793_cast_fp16")]; + string _SplitHeadsQ__mh_w_3795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3795_equation_0, values = (var_36078_cast_fp16, var_35863_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3795_cast_fp16")]; + string _SplitHeadsQ__mh_w_3797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3797_equation_0, values = (var_36078_cast_fp16, var_35870_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3797_cast_fp16")]; + string _SplitHeadsQ__mh_w_3799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3799_equation_0, values = (var_36078_cast_fp16, var_35877_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3799_cast_fp16")]; + string _SplitHeadsQ__mh_w_3801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3801_equation_0, values = (var_36082_cast_fp16, var_35884_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3801_cast_fp16")]; + string _SplitHeadsQ__mh_w_3803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3803_equation_0, values = (var_36082_cast_fp16, var_35891_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3803_cast_fp16")]; + string _SplitHeadsQ__mh_w_3805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3805_equation_0, values = (var_36082_cast_fp16, var_35898_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3805_cast_fp16")]; + string _SplitHeadsQ__mh_w_3807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3807_equation_0, values = (var_36082_cast_fp16, var_35905_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3807_cast_fp16")]; + string _SplitHeadsQ__mh_w_3809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3809_equation_0, values = (var_36086_cast_fp16, var_35912_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3809_cast_fp16")]; + string _SplitHeadsQ__mh_w_3811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3811_equation_0, values = (var_36086_cast_fp16, var_35919_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3811_cast_fp16")]; + string _SplitHeadsQ__mh_w_3813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3813_equation_0, values = (var_36086_cast_fp16, var_35926_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3813_cast_fp16")]; + string _SplitHeadsQ__mh_w_3815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3815_equation_0, values = (var_36086_cast_fp16, var_35933_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3815_cast_fp16")]; + string _SplitHeadsQ__mh_w_3817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3817_equation_0, values = (var_36090_cast_fp16, var_35940_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3817_cast_fp16")]; + string _SplitHeadsQ__mh_w_3819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3819_equation_0, values = (var_36090_cast_fp16, var_35947_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3819_cast_fp16")]; + string _SplitHeadsQ__mh_w_3821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3821_equation_0, values = (var_36090_cast_fp16, var_35954_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3821_cast_fp16")]; + string _SplitHeadsQ__mh_w_3823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3823_equation_0, values = (var_36090_cast_fp16, var_35961_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3823_cast_fp16")]; + string _SplitHeadsQ__mh_w_3825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3825_equation_0, values = (var_36094_cast_fp16, var_35968_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3825_cast_fp16")]; + string _SplitHeadsQ__mh_w_3827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3827_equation_0, values = (var_36094_cast_fp16, var_35975_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3827_cast_fp16")]; + string _SplitHeadsQ__mh_w_3829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3829_equation_0, values = (var_36094_cast_fp16, var_35982_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3829_cast_fp16")]; + string _SplitHeadsQ__mh_w_3831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3831_equation_0, values = (var_36094_cast_fp16, var_35989_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3831_cast_fp16")]; + string _SplitHeadsQ__mh_w_3833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3833_equation_0, values = (var_36098_cast_fp16, var_35996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3833_cast_fp16")]; + string _SplitHeadsQ__mh_w_3835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3835_equation_0, values = (var_36098_cast_fp16, var_36003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3835_cast_fp16")]; + string _SplitHeadsQ__mh_w_3837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3837_equation_0, values = (var_36098_cast_fp16, var_36010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3837_cast_fp16")]; + string _SplitHeadsQ__mh_w_3839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3839_equation_0, values = (var_36098_cast_fp16, var_36017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3839_cast_fp16")]; + fp16 var_36339_to_fp16 = const()[name = string("op_36339_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3681_cast_fp16, y = var_36339_to_fp16)[name = string("aw_chunk_3681_cast_fp16")]; + fp16 var_36341_to_fp16 = const()[name = string("op_36341_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3683_cast_fp16, y = var_36341_to_fp16)[name = string("aw_chunk_3683_cast_fp16")]; + fp16 var_36343_to_fp16 = const()[name = string("op_36343_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3685_cast_fp16, y = var_36343_to_fp16)[name = string("aw_chunk_3685_cast_fp16")]; + fp16 var_36345_to_fp16 = const()[name = string("op_36345_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3687_cast_fp16, y = var_36345_to_fp16)[name = string("aw_chunk_3687_cast_fp16")]; + fp16 var_36347_to_fp16 = const()[name = string("op_36347_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3689_cast_fp16, y = var_36347_to_fp16)[name = string("aw_chunk_3689_cast_fp16")]; + fp16 var_36349_to_fp16 = const()[name = string("op_36349_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3691_cast_fp16, y = var_36349_to_fp16)[name = string("aw_chunk_3691_cast_fp16")]; + fp16 var_36351_to_fp16 = const()[name = string("op_36351_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3693_cast_fp16, y = var_36351_to_fp16)[name = string("aw_chunk_3693_cast_fp16")]; + fp16 var_36353_to_fp16 = const()[name = string("op_36353_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3695_cast_fp16, y = var_36353_to_fp16)[name = string("aw_chunk_3695_cast_fp16")]; + fp16 var_36355_to_fp16 = const()[name = string("op_36355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3697_cast_fp16, y = var_36355_to_fp16)[name = string("aw_chunk_3697_cast_fp16")]; + fp16 var_36357_to_fp16 = const()[name = string("op_36357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3699_cast_fp16, y = var_36357_to_fp16)[name = string("aw_chunk_3699_cast_fp16")]; + fp16 var_36359_to_fp16 = const()[name = string("op_36359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3701_cast_fp16, y = var_36359_to_fp16)[name = string("aw_chunk_3701_cast_fp16")]; + fp16 var_36361_to_fp16 = const()[name = string("op_36361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3703_cast_fp16, y = var_36361_to_fp16)[name = string("aw_chunk_3703_cast_fp16")]; + fp16 var_36363_to_fp16 = const()[name = string("op_36363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3705_cast_fp16, y = var_36363_to_fp16)[name = string("aw_chunk_3705_cast_fp16")]; + fp16 var_36365_to_fp16 = const()[name = string("op_36365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3707_cast_fp16, y = var_36365_to_fp16)[name = string("aw_chunk_3707_cast_fp16")]; + fp16 var_36367_to_fp16 = const()[name = string("op_36367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3709_cast_fp16, y = var_36367_to_fp16)[name = string("aw_chunk_3709_cast_fp16")]; + fp16 var_36369_to_fp16 = const()[name = string("op_36369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3711_cast_fp16, y = var_36369_to_fp16)[name = string("aw_chunk_3711_cast_fp16")]; + fp16 var_36371_to_fp16 = const()[name = string("op_36371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3713_cast_fp16, y = var_36371_to_fp16)[name = string("aw_chunk_3713_cast_fp16")]; + fp16 var_36373_to_fp16 = const()[name = string("op_36373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3715_cast_fp16, y = var_36373_to_fp16)[name = string("aw_chunk_3715_cast_fp16")]; + fp16 var_36375_to_fp16 = const()[name = string("op_36375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3717_cast_fp16, y = var_36375_to_fp16)[name = string("aw_chunk_3717_cast_fp16")]; + fp16 var_36377_to_fp16 = const()[name = string("op_36377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3719_cast_fp16, y = var_36377_to_fp16)[name = string("aw_chunk_3719_cast_fp16")]; + fp16 var_36379_to_fp16 = const()[name = string("op_36379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3721_cast_fp16, y = var_36379_to_fp16)[name = string("aw_chunk_3721_cast_fp16")]; + fp16 var_36381_to_fp16 = const()[name = string("op_36381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3723_cast_fp16, y = var_36381_to_fp16)[name = string("aw_chunk_3723_cast_fp16")]; + fp16 var_36383_to_fp16 = const()[name = string("op_36383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3725_cast_fp16, y = var_36383_to_fp16)[name = string("aw_chunk_3725_cast_fp16")]; + fp16 var_36385_to_fp16 = const()[name = string("op_36385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3727_cast_fp16, y = var_36385_to_fp16)[name = string("aw_chunk_3727_cast_fp16")]; + fp16 var_36387_to_fp16 = const()[name = string("op_36387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3729_cast_fp16, y = var_36387_to_fp16)[name = string("aw_chunk_3729_cast_fp16")]; + fp16 var_36389_to_fp16 = const()[name = string("op_36389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3731_cast_fp16, y = var_36389_to_fp16)[name = string("aw_chunk_3731_cast_fp16")]; + fp16 var_36391_to_fp16 = const()[name = string("op_36391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3733_cast_fp16, y = var_36391_to_fp16)[name = string("aw_chunk_3733_cast_fp16")]; + fp16 var_36393_to_fp16 = const()[name = string("op_36393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3735_cast_fp16, y = var_36393_to_fp16)[name = string("aw_chunk_3735_cast_fp16")]; + fp16 var_36395_to_fp16 = const()[name = string("op_36395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3737_cast_fp16, y = var_36395_to_fp16)[name = string("aw_chunk_3737_cast_fp16")]; + fp16 var_36397_to_fp16 = const()[name = string("op_36397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3739_cast_fp16, y = var_36397_to_fp16)[name = string("aw_chunk_3739_cast_fp16")]; + fp16 var_36399_to_fp16 = const()[name = string("op_36399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3741_cast_fp16, y = var_36399_to_fp16)[name = string("aw_chunk_3741_cast_fp16")]; + fp16 var_36401_to_fp16 = const()[name = string("op_36401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3743_cast_fp16, y = var_36401_to_fp16)[name = string("aw_chunk_3743_cast_fp16")]; + fp16 var_36403_to_fp16 = const()[name = string("op_36403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3745_cast_fp16, y = var_36403_to_fp16)[name = string("aw_chunk_3745_cast_fp16")]; + fp16 var_36405_to_fp16 = const()[name = string("op_36405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3747_cast_fp16, y = var_36405_to_fp16)[name = string("aw_chunk_3747_cast_fp16")]; + fp16 var_36407_to_fp16 = const()[name = string("op_36407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3749_cast_fp16, y = var_36407_to_fp16)[name = string("aw_chunk_3749_cast_fp16")]; + fp16 var_36409_to_fp16 = const()[name = string("op_36409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3751_cast_fp16, y = var_36409_to_fp16)[name = string("aw_chunk_3751_cast_fp16")]; + fp16 var_36411_to_fp16 = const()[name = string("op_36411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3753_cast_fp16, y = var_36411_to_fp16)[name = string("aw_chunk_3753_cast_fp16")]; + fp16 var_36413_to_fp16 = const()[name = string("op_36413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3755_cast_fp16, y = var_36413_to_fp16)[name = string("aw_chunk_3755_cast_fp16")]; + fp16 var_36415_to_fp16 = const()[name = string("op_36415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3757_cast_fp16, y = var_36415_to_fp16)[name = string("aw_chunk_3757_cast_fp16")]; + fp16 var_36417_to_fp16 = const()[name = string("op_36417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3759_cast_fp16, y = var_36417_to_fp16)[name = string("aw_chunk_3759_cast_fp16")]; + fp16 var_36419_to_fp16 = const()[name = string("op_36419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3761_cast_fp16, y = var_36419_to_fp16)[name = string("aw_chunk_3761_cast_fp16")]; + fp16 var_36421_to_fp16 = const()[name = string("op_36421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3763_cast_fp16, y = var_36421_to_fp16)[name = string("aw_chunk_3763_cast_fp16")]; + fp16 var_36423_to_fp16 = const()[name = string("op_36423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3765_cast_fp16, y = var_36423_to_fp16)[name = string("aw_chunk_3765_cast_fp16")]; + fp16 var_36425_to_fp16 = const()[name = string("op_36425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3767_cast_fp16, y = var_36425_to_fp16)[name = string("aw_chunk_3767_cast_fp16")]; + fp16 var_36427_to_fp16 = const()[name = string("op_36427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3769_cast_fp16, y = var_36427_to_fp16)[name = string("aw_chunk_3769_cast_fp16")]; + fp16 var_36429_to_fp16 = const()[name = string("op_36429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3771_cast_fp16, y = var_36429_to_fp16)[name = string("aw_chunk_3771_cast_fp16")]; + fp16 var_36431_to_fp16 = const()[name = string("op_36431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3773_cast_fp16, y = var_36431_to_fp16)[name = string("aw_chunk_3773_cast_fp16")]; + fp16 var_36433_to_fp16 = const()[name = string("op_36433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3775_cast_fp16, y = var_36433_to_fp16)[name = string("aw_chunk_3775_cast_fp16")]; + fp16 var_36435_to_fp16 = const()[name = string("op_36435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3777_cast_fp16, y = var_36435_to_fp16)[name = string("aw_chunk_3777_cast_fp16")]; + fp16 var_36437_to_fp16 = const()[name = string("op_36437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3779_cast_fp16, y = var_36437_to_fp16)[name = string("aw_chunk_3779_cast_fp16")]; + fp16 var_36439_to_fp16 = const()[name = string("op_36439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3781_cast_fp16, y = var_36439_to_fp16)[name = string("aw_chunk_3781_cast_fp16")]; + fp16 var_36441_to_fp16 = const()[name = string("op_36441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3783_cast_fp16, y = var_36441_to_fp16)[name = string("aw_chunk_3783_cast_fp16")]; + fp16 var_36443_to_fp16 = const()[name = string("op_36443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3785_cast_fp16, y = var_36443_to_fp16)[name = string("aw_chunk_3785_cast_fp16")]; + fp16 var_36445_to_fp16 = const()[name = string("op_36445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3787_cast_fp16, y = var_36445_to_fp16)[name = string("aw_chunk_3787_cast_fp16")]; + fp16 var_36447_to_fp16 = const()[name = string("op_36447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3789_cast_fp16, y = var_36447_to_fp16)[name = string("aw_chunk_3789_cast_fp16")]; + fp16 var_36449_to_fp16 = const()[name = string("op_36449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3791_cast_fp16, y = var_36449_to_fp16)[name = string("aw_chunk_3791_cast_fp16")]; + fp16 var_36451_to_fp16 = const()[name = string("op_36451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3793_cast_fp16, y = var_36451_to_fp16)[name = string("aw_chunk_3793_cast_fp16")]; + fp16 var_36453_to_fp16 = const()[name = string("op_36453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3795_cast_fp16, y = var_36453_to_fp16)[name = string("aw_chunk_3795_cast_fp16")]; + fp16 var_36455_to_fp16 = const()[name = string("op_36455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3797_cast_fp16, y = var_36455_to_fp16)[name = string("aw_chunk_3797_cast_fp16")]; + fp16 var_36457_to_fp16 = const()[name = string("op_36457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3799_cast_fp16, y = var_36457_to_fp16)[name = string("aw_chunk_3799_cast_fp16")]; + fp16 var_36459_to_fp16 = const()[name = string("op_36459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3801_cast_fp16, y = var_36459_to_fp16)[name = string("aw_chunk_3801_cast_fp16")]; + fp16 var_36461_to_fp16 = const()[name = string("op_36461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3803_cast_fp16, y = var_36461_to_fp16)[name = string("aw_chunk_3803_cast_fp16")]; + fp16 var_36463_to_fp16 = const()[name = string("op_36463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3805_cast_fp16, y = var_36463_to_fp16)[name = string("aw_chunk_3805_cast_fp16")]; + fp16 var_36465_to_fp16 = const()[name = string("op_36465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3807_cast_fp16, y = var_36465_to_fp16)[name = string("aw_chunk_3807_cast_fp16")]; + fp16 var_36467_to_fp16 = const()[name = string("op_36467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3809_cast_fp16, y = var_36467_to_fp16)[name = string("aw_chunk_3809_cast_fp16")]; + fp16 var_36469_to_fp16 = const()[name = string("op_36469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3811_cast_fp16, y = var_36469_to_fp16)[name = string("aw_chunk_3811_cast_fp16")]; + fp16 var_36471_to_fp16 = const()[name = string("op_36471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3813_cast_fp16, y = var_36471_to_fp16)[name = string("aw_chunk_3813_cast_fp16")]; + fp16 var_36473_to_fp16 = const()[name = string("op_36473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3815_cast_fp16, y = var_36473_to_fp16)[name = string("aw_chunk_3815_cast_fp16")]; + fp16 var_36475_to_fp16 = const()[name = string("op_36475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3817_cast_fp16, y = var_36475_to_fp16)[name = string("aw_chunk_3817_cast_fp16")]; + fp16 var_36477_to_fp16 = const()[name = string("op_36477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3819_cast_fp16, y = var_36477_to_fp16)[name = string("aw_chunk_3819_cast_fp16")]; + fp16 var_36479_to_fp16 = const()[name = string("op_36479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3821_cast_fp16, y = var_36479_to_fp16)[name = string("aw_chunk_3821_cast_fp16")]; + fp16 var_36481_to_fp16 = const()[name = string("op_36481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3823_cast_fp16, y = var_36481_to_fp16)[name = string("aw_chunk_3823_cast_fp16")]; + fp16 var_36483_to_fp16 = const()[name = string("op_36483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3825_cast_fp16, y = var_36483_to_fp16)[name = string("aw_chunk_3825_cast_fp16")]; + fp16 var_36485_to_fp16 = const()[name = string("op_36485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3827_cast_fp16, y = var_36485_to_fp16)[name = string("aw_chunk_3827_cast_fp16")]; + fp16 var_36487_to_fp16 = const()[name = string("op_36487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3829_cast_fp16, y = var_36487_to_fp16)[name = string("aw_chunk_3829_cast_fp16")]; + fp16 var_36489_to_fp16 = const()[name = string("op_36489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3831_cast_fp16, y = var_36489_to_fp16)[name = string("aw_chunk_3831_cast_fp16")]; + fp16 var_36491_to_fp16 = const()[name = string("op_36491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3833_cast_fp16, y = var_36491_to_fp16)[name = string("aw_chunk_3833_cast_fp16")]; + fp16 var_36493_to_fp16 = const()[name = string("op_36493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3835_cast_fp16, y = var_36493_to_fp16)[name = string("aw_chunk_3835_cast_fp16")]; + fp16 var_36495_to_fp16 = const()[name = string("op_36495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3837_cast_fp16, y = var_36495_to_fp16)[name = string("aw_chunk_3837_cast_fp16")]; + fp16 var_36497_to_fp16 = const()[name = string("op_36497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3839_cast_fp16, y = var_36497_to_fp16)[name = string("aw_chunk_3839_cast_fp16")]; + tensor var_36499_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3681_cast_fp16)[name = string("op_36499_cast_fp16")]; + tensor var_36500_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3683_cast_fp16)[name = string("op_36500_cast_fp16")]; + tensor var_36501_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3685_cast_fp16)[name = string("op_36501_cast_fp16")]; + tensor var_36502_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3687_cast_fp16)[name = string("op_36502_cast_fp16")]; + tensor var_36503_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3689_cast_fp16)[name = string("op_36503_cast_fp16")]; + tensor var_36504_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3691_cast_fp16)[name = string("op_36504_cast_fp16")]; + tensor var_36505_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3693_cast_fp16)[name = string("op_36505_cast_fp16")]; + tensor var_36506_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3695_cast_fp16)[name = string("op_36506_cast_fp16")]; + tensor var_36507_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3697_cast_fp16)[name = string("op_36507_cast_fp16")]; + tensor var_36508_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3699_cast_fp16)[name = string("op_36508_cast_fp16")]; + tensor var_36509_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3701_cast_fp16)[name = string("op_36509_cast_fp16")]; + tensor var_36510_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3703_cast_fp16)[name = string("op_36510_cast_fp16")]; + tensor var_36511_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3705_cast_fp16)[name = string("op_36511_cast_fp16")]; + tensor var_36512_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3707_cast_fp16)[name = string("op_36512_cast_fp16")]; + tensor var_36513_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3709_cast_fp16)[name = string("op_36513_cast_fp16")]; + tensor var_36514_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3711_cast_fp16)[name = string("op_36514_cast_fp16")]; + tensor var_36515_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3713_cast_fp16)[name = string("op_36515_cast_fp16")]; + tensor var_36516_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3715_cast_fp16)[name = string("op_36516_cast_fp16")]; + tensor var_36517_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3717_cast_fp16)[name = string("op_36517_cast_fp16")]; + tensor var_36518_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3719_cast_fp16)[name = string("op_36518_cast_fp16")]; + tensor var_36519_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3721_cast_fp16)[name = string("op_36519_cast_fp16")]; + tensor var_36520_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3723_cast_fp16)[name = string("op_36520_cast_fp16")]; + tensor var_36521_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3725_cast_fp16)[name = string("op_36521_cast_fp16")]; + tensor var_36522_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3727_cast_fp16)[name = string("op_36522_cast_fp16")]; + tensor var_36523_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3729_cast_fp16)[name = string("op_36523_cast_fp16")]; + tensor var_36524_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3731_cast_fp16)[name = string("op_36524_cast_fp16")]; + tensor var_36525_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3733_cast_fp16)[name = string("op_36525_cast_fp16")]; + tensor var_36526_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3735_cast_fp16)[name = string("op_36526_cast_fp16")]; + tensor var_36527_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3737_cast_fp16)[name = string("op_36527_cast_fp16")]; + tensor var_36528_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3739_cast_fp16)[name = string("op_36528_cast_fp16")]; + tensor var_36529_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3741_cast_fp16)[name = string("op_36529_cast_fp16")]; + tensor var_36530_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3743_cast_fp16)[name = string("op_36530_cast_fp16")]; + tensor var_36531_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3745_cast_fp16)[name = string("op_36531_cast_fp16")]; + tensor var_36532_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3747_cast_fp16)[name = string("op_36532_cast_fp16")]; + tensor var_36533_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3749_cast_fp16)[name = string("op_36533_cast_fp16")]; + tensor var_36534_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3751_cast_fp16)[name = string("op_36534_cast_fp16")]; + tensor var_36535_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3753_cast_fp16)[name = string("op_36535_cast_fp16")]; + tensor var_36536_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3755_cast_fp16)[name = string("op_36536_cast_fp16")]; + tensor var_36537_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3757_cast_fp16)[name = string("op_36537_cast_fp16")]; + tensor var_36538_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3759_cast_fp16)[name = string("op_36538_cast_fp16")]; + tensor var_36539_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3761_cast_fp16)[name = string("op_36539_cast_fp16")]; + tensor var_36540_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3763_cast_fp16)[name = string("op_36540_cast_fp16")]; + tensor var_36541_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3765_cast_fp16)[name = string("op_36541_cast_fp16")]; + tensor var_36542_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3767_cast_fp16)[name = string("op_36542_cast_fp16")]; + tensor var_36543_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3769_cast_fp16)[name = string("op_36543_cast_fp16")]; + tensor var_36544_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3771_cast_fp16)[name = string("op_36544_cast_fp16")]; + tensor var_36545_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3773_cast_fp16)[name = string("op_36545_cast_fp16")]; + tensor var_36546_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3775_cast_fp16)[name = string("op_36546_cast_fp16")]; + tensor var_36547_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3777_cast_fp16)[name = string("op_36547_cast_fp16")]; + tensor var_36548_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3779_cast_fp16)[name = string("op_36548_cast_fp16")]; + tensor var_36549_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3781_cast_fp16)[name = string("op_36549_cast_fp16")]; + tensor var_36550_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3783_cast_fp16)[name = string("op_36550_cast_fp16")]; + tensor var_36551_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3785_cast_fp16)[name = string("op_36551_cast_fp16")]; + tensor var_36552_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3787_cast_fp16)[name = string("op_36552_cast_fp16")]; + tensor var_36553_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3789_cast_fp16)[name = string("op_36553_cast_fp16")]; + tensor var_36554_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3791_cast_fp16)[name = string("op_36554_cast_fp16")]; + tensor var_36555_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3793_cast_fp16)[name = string("op_36555_cast_fp16")]; + tensor var_36556_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3795_cast_fp16)[name = string("op_36556_cast_fp16")]; + tensor var_36557_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3797_cast_fp16)[name = string("op_36557_cast_fp16")]; + tensor var_36558_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3799_cast_fp16)[name = string("op_36558_cast_fp16")]; + tensor var_36559_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3801_cast_fp16)[name = string("op_36559_cast_fp16")]; + tensor var_36560_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3803_cast_fp16)[name = string("op_36560_cast_fp16")]; + tensor var_36561_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3805_cast_fp16)[name = string("op_36561_cast_fp16")]; + tensor var_36562_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3807_cast_fp16)[name = string("op_36562_cast_fp16")]; + tensor var_36563_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3809_cast_fp16)[name = string("op_36563_cast_fp16")]; + tensor var_36564_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3811_cast_fp16)[name = string("op_36564_cast_fp16")]; + tensor var_36565_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3813_cast_fp16)[name = string("op_36565_cast_fp16")]; + tensor var_36566_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3815_cast_fp16)[name = string("op_36566_cast_fp16")]; + tensor var_36567_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3817_cast_fp16)[name = string("op_36567_cast_fp16")]; + tensor var_36568_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3819_cast_fp16)[name = string("op_36568_cast_fp16")]; + tensor var_36569_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3821_cast_fp16)[name = string("op_36569_cast_fp16")]; + tensor var_36570_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3823_cast_fp16)[name = string("op_36570_cast_fp16")]; + tensor var_36571_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3825_cast_fp16)[name = string("op_36571_cast_fp16")]; + tensor var_36572_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3827_cast_fp16)[name = string("op_36572_cast_fp16")]; + tensor var_36573_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3829_cast_fp16)[name = string("op_36573_cast_fp16")]; + tensor var_36574_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3831_cast_fp16)[name = string("op_36574_cast_fp16")]; + tensor var_36575_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3833_cast_fp16)[name = string("op_36575_cast_fp16")]; + tensor var_36576_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3835_cast_fp16)[name = string("op_36576_cast_fp16")]; + tensor var_36577_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3837_cast_fp16)[name = string("op_36577_cast_fp16")]; + tensor var_36578_cast_fp16 = softmax(axis = var_35324, x = aw_chunk_3839_cast_fp16)[name = string("op_36578_cast_fp16")]; + string var_36580_equation_0 = const()[name = string("op_36580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36580_cast_fp16 = einsum(equation = var_36580_equation_0, values = (var_36100_cast_fp16, var_36499_cast_fp16))[name = string("op_36580_cast_fp16")]; + string var_36582_equation_0 = const()[name = string("op_36582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36582_cast_fp16 = einsum(equation = var_36582_equation_0, values = (var_36100_cast_fp16, var_36500_cast_fp16))[name = string("op_36582_cast_fp16")]; + string var_36584_equation_0 = const()[name = string("op_36584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36584_cast_fp16 = einsum(equation = var_36584_equation_0, values = (var_36100_cast_fp16, var_36501_cast_fp16))[name = string("op_36584_cast_fp16")]; + string var_36586_equation_0 = const()[name = string("op_36586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36586_cast_fp16 = einsum(equation = var_36586_equation_0, values = (var_36100_cast_fp16, var_36502_cast_fp16))[name = string("op_36586_cast_fp16")]; + string var_36588_equation_0 = const()[name = string("op_36588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36588_cast_fp16 = einsum(equation = var_36588_equation_0, values = (var_36104_cast_fp16, var_36503_cast_fp16))[name = string("op_36588_cast_fp16")]; + string var_36590_equation_0 = const()[name = string("op_36590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36590_cast_fp16 = einsum(equation = var_36590_equation_0, values = (var_36104_cast_fp16, var_36504_cast_fp16))[name = string("op_36590_cast_fp16")]; + string var_36592_equation_0 = const()[name = string("op_36592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36592_cast_fp16 = einsum(equation = var_36592_equation_0, values = (var_36104_cast_fp16, var_36505_cast_fp16))[name = string("op_36592_cast_fp16")]; + string var_36594_equation_0 = const()[name = string("op_36594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36594_cast_fp16 = einsum(equation = var_36594_equation_0, values = (var_36104_cast_fp16, var_36506_cast_fp16))[name = string("op_36594_cast_fp16")]; + string var_36596_equation_0 = const()[name = string("op_36596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36596_cast_fp16 = einsum(equation = var_36596_equation_0, values = (var_36108_cast_fp16, var_36507_cast_fp16))[name = string("op_36596_cast_fp16")]; + string var_36598_equation_0 = const()[name = string("op_36598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36598_cast_fp16 = einsum(equation = var_36598_equation_0, values = (var_36108_cast_fp16, var_36508_cast_fp16))[name = string("op_36598_cast_fp16")]; + string var_36600_equation_0 = const()[name = string("op_36600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36600_cast_fp16 = einsum(equation = var_36600_equation_0, values = (var_36108_cast_fp16, var_36509_cast_fp16))[name = string("op_36600_cast_fp16")]; + string var_36602_equation_0 = const()[name = string("op_36602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36602_cast_fp16 = einsum(equation = var_36602_equation_0, values = (var_36108_cast_fp16, var_36510_cast_fp16))[name = string("op_36602_cast_fp16")]; + string var_36604_equation_0 = const()[name = string("op_36604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36604_cast_fp16 = einsum(equation = var_36604_equation_0, values = (var_36112_cast_fp16, var_36511_cast_fp16))[name = string("op_36604_cast_fp16")]; + string var_36606_equation_0 = const()[name = string("op_36606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36606_cast_fp16 = einsum(equation = var_36606_equation_0, values = (var_36112_cast_fp16, var_36512_cast_fp16))[name = string("op_36606_cast_fp16")]; + string var_36608_equation_0 = const()[name = string("op_36608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36608_cast_fp16 = einsum(equation = var_36608_equation_0, values = (var_36112_cast_fp16, var_36513_cast_fp16))[name = string("op_36608_cast_fp16")]; + string var_36610_equation_0 = const()[name = string("op_36610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36610_cast_fp16 = einsum(equation = var_36610_equation_0, values = (var_36112_cast_fp16, var_36514_cast_fp16))[name = string("op_36610_cast_fp16")]; + string var_36612_equation_0 = const()[name = string("op_36612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36612_cast_fp16 = einsum(equation = var_36612_equation_0, values = (var_36116_cast_fp16, var_36515_cast_fp16))[name = string("op_36612_cast_fp16")]; + string var_36614_equation_0 = const()[name = string("op_36614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36614_cast_fp16 = einsum(equation = var_36614_equation_0, values = (var_36116_cast_fp16, var_36516_cast_fp16))[name = string("op_36614_cast_fp16")]; + string var_36616_equation_0 = const()[name = string("op_36616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36616_cast_fp16 = einsum(equation = var_36616_equation_0, values = (var_36116_cast_fp16, var_36517_cast_fp16))[name = string("op_36616_cast_fp16")]; + string var_36618_equation_0 = const()[name = string("op_36618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36618_cast_fp16 = einsum(equation = var_36618_equation_0, values = (var_36116_cast_fp16, var_36518_cast_fp16))[name = string("op_36618_cast_fp16")]; + string var_36620_equation_0 = const()[name = string("op_36620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36620_cast_fp16 = einsum(equation = var_36620_equation_0, values = (var_36120_cast_fp16, var_36519_cast_fp16))[name = string("op_36620_cast_fp16")]; + string var_36622_equation_0 = const()[name = string("op_36622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36622_cast_fp16 = einsum(equation = var_36622_equation_0, values = (var_36120_cast_fp16, var_36520_cast_fp16))[name = string("op_36622_cast_fp16")]; + string var_36624_equation_0 = const()[name = string("op_36624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36624_cast_fp16 = einsum(equation = var_36624_equation_0, values = (var_36120_cast_fp16, var_36521_cast_fp16))[name = string("op_36624_cast_fp16")]; + string var_36626_equation_0 = const()[name = string("op_36626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36626_cast_fp16 = einsum(equation = var_36626_equation_0, values = (var_36120_cast_fp16, var_36522_cast_fp16))[name = string("op_36626_cast_fp16")]; + string var_36628_equation_0 = const()[name = string("op_36628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36628_cast_fp16 = einsum(equation = var_36628_equation_0, values = (var_36124_cast_fp16, var_36523_cast_fp16))[name = string("op_36628_cast_fp16")]; + string var_36630_equation_0 = const()[name = string("op_36630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36630_cast_fp16 = einsum(equation = var_36630_equation_0, values = (var_36124_cast_fp16, var_36524_cast_fp16))[name = string("op_36630_cast_fp16")]; + string var_36632_equation_0 = const()[name = string("op_36632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36632_cast_fp16 = einsum(equation = var_36632_equation_0, values = (var_36124_cast_fp16, var_36525_cast_fp16))[name = string("op_36632_cast_fp16")]; + string var_36634_equation_0 = const()[name = string("op_36634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36634_cast_fp16 = einsum(equation = var_36634_equation_0, values = (var_36124_cast_fp16, var_36526_cast_fp16))[name = string("op_36634_cast_fp16")]; + string var_36636_equation_0 = const()[name = string("op_36636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36636_cast_fp16 = einsum(equation = var_36636_equation_0, values = (var_36128_cast_fp16, var_36527_cast_fp16))[name = string("op_36636_cast_fp16")]; + string var_36638_equation_0 = const()[name = string("op_36638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36638_cast_fp16 = einsum(equation = var_36638_equation_0, values = (var_36128_cast_fp16, var_36528_cast_fp16))[name = string("op_36638_cast_fp16")]; + string var_36640_equation_0 = const()[name = string("op_36640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36640_cast_fp16 = einsum(equation = var_36640_equation_0, values = (var_36128_cast_fp16, var_36529_cast_fp16))[name = string("op_36640_cast_fp16")]; + string var_36642_equation_0 = const()[name = string("op_36642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36642_cast_fp16 = einsum(equation = var_36642_equation_0, values = (var_36128_cast_fp16, var_36530_cast_fp16))[name = string("op_36642_cast_fp16")]; + string var_36644_equation_0 = const()[name = string("op_36644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36644_cast_fp16 = einsum(equation = var_36644_equation_0, values = (var_36132_cast_fp16, var_36531_cast_fp16))[name = string("op_36644_cast_fp16")]; + string var_36646_equation_0 = const()[name = string("op_36646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36646_cast_fp16 = einsum(equation = var_36646_equation_0, values = (var_36132_cast_fp16, var_36532_cast_fp16))[name = string("op_36646_cast_fp16")]; + string var_36648_equation_0 = const()[name = string("op_36648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36648_cast_fp16 = einsum(equation = var_36648_equation_0, values = (var_36132_cast_fp16, var_36533_cast_fp16))[name = string("op_36648_cast_fp16")]; + string var_36650_equation_0 = const()[name = string("op_36650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36650_cast_fp16 = einsum(equation = var_36650_equation_0, values = (var_36132_cast_fp16, var_36534_cast_fp16))[name = string("op_36650_cast_fp16")]; + string var_36652_equation_0 = const()[name = string("op_36652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36652_cast_fp16 = einsum(equation = var_36652_equation_0, values = (var_36136_cast_fp16, var_36535_cast_fp16))[name = string("op_36652_cast_fp16")]; + string var_36654_equation_0 = const()[name = string("op_36654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36654_cast_fp16 = einsum(equation = var_36654_equation_0, values = (var_36136_cast_fp16, var_36536_cast_fp16))[name = string("op_36654_cast_fp16")]; + string var_36656_equation_0 = const()[name = string("op_36656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36656_cast_fp16 = einsum(equation = var_36656_equation_0, values = (var_36136_cast_fp16, var_36537_cast_fp16))[name = string("op_36656_cast_fp16")]; + string var_36658_equation_0 = const()[name = string("op_36658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36658_cast_fp16 = einsum(equation = var_36658_equation_0, values = (var_36136_cast_fp16, var_36538_cast_fp16))[name = string("op_36658_cast_fp16")]; + string var_36660_equation_0 = const()[name = string("op_36660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36660_cast_fp16 = einsum(equation = var_36660_equation_0, values = (var_36140_cast_fp16, var_36539_cast_fp16))[name = string("op_36660_cast_fp16")]; + string var_36662_equation_0 = const()[name = string("op_36662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36662_cast_fp16 = einsum(equation = var_36662_equation_0, values = (var_36140_cast_fp16, var_36540_cast_fp16))[name = string("op_36662_cast_fp16")]; + string var_36664_equation_0 = const()[name = string("op_36664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36664_cast_fp16 = einsum(equation = var_36664_equation_0, values = (var_36140_cast_fp16, var_36541_cast_fp16))[name = string("op_36664_cast_fp16")]; + string var_36666_equation_0 = const()[name = string("op_36666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36666_cast_fp16 = einsum(equation = var_36666_equation_0, values = (var_36140_cast_fp16, var_36542_cast_fp16))[name = string("op_36666_cast_fp16")]; + string var_36668_equation_0 = const()[name = string("op_36668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36668_cast_fp16 = einsum(equation = var_36668_equation_0, values = (var_36144_cast_fp16, var_36543_cast_fp16))[name = string("op_36668_cast_fp16")]; + string var_36670_equation_0 = const()[name = string("op_36670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36670_cast_fp16 = einsum(equation = var_36670_equation_0, values = (var_36144_cast_fp16, var_36544_cast_fp16))[name = string("op_36670_cast_fp16")]; + string var_36672_equation_0 = const()[name = string("op_36672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36672_cast_fp16 = einsum(equation = var_36672_equation_0, values = (var_36144_cast_fp16, var_36545_cast_fp16))[name = string("op_36672_cast_fp16")]; + string var_36674_equation_0 = const()[name = string("op_36674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36674_cast_fp16 = einsum(equation = var_36674_equation_0, values = (var_36144_cast_fp16, var_36546_cast_fp16))[name = string("op_36674_cast_fp16")]; + string var_36676_equation_0 = const()[name = string("op_36676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36676_cast_fp16 = einsum(equation = var_36676_equation_0, values = (var_36148_cast_fp16, var_36547_cast_fp16))[name = string("op_36676_cast_fp16")]; + string var_36678_equation_0 = const()[name = string("op_36678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36678_cast_fp16 = einsum(equation = var_36678_equation_0, values = (var_36148_cast_fp16, var_36548_cast_fp16))[name = string("op_36678_cast_fp16")]; + string var_36680_equation_0 = const()[name = string("op_36680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36680_cast_fp16 = einsum(equation = var_36680_equation_0, values = (var_36148_cast_fp16, var_36549_cast_fp16))[name = string("op_36680_cast_fp16")]; + string var_36682_equation_0 = const()[name = string("op_36682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36682_cast_fp16 = einsum(equation = var_36682_equation_0, values = (var_36148_cast_fp16, var_36550_cast_fp16))[name = string("op_36682_cast_fp16")]; + string var_36684_equation_0 = const()[name = string("op_36684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36684_cast_fp16 = einsum(equation = var_36684_equation_0, values = (var_36152_cast_fp16, var_36551_cast_fp16))[name = string("op_36684_cast_fp16")]; + string var_36686_equation_0 = const()[name = string("op_36686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36686_cast_fp16 = einsum(equation = var_36686_equation_0, values = (var_36152_cast_fp16, var_36552_cast_fp16))[name = string("op_36686_cast_fp16")]; + string var_36688_equation_0 = const()[name = string("op_36688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36688_cast_fp16 = einsum(equation = var_36688_equation_0, values = (var_36152_cast_fp16, var_36553_cast_fp16))[name = string("op_36688_cast_fp16")]; + string var_36690_equation_0 = const()[name = string("op_36690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36690_cast_fp16 = einsum(equation = var_36690_equation_0, values = (var_36152_cast_fp16, var_36554_cast_fp16))[name = string("op_36690_cast_fp16")]; + string var_36692_equation_0 = const()[name = string("op_36692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36692_cast_fp16 = einsum(equation = var_36692_equation_0, values = (var_36156_cast_fp16, var_36555_cast_fp16))[name = string("op_36692_cast_fp16")]; + string var_36694_equation_0 = const()[name = string("op_36694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36694_cast_fp16 = einsum(equation = var_36694_equation_0, values = (var_36156_cast_fp16, var_36556_cast_fp16))[name = string("op_36694_cast_fp16")]; + string var_36696_equation_0 = const()[name = string("op_36696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36696_cast_fp16 = einsum(equation = var_36696_equation_0, values = (var_36156_cast_fp16, var_36557_cast_fp16))[name = string("op_36696_cast_fp16")]; + string var_36698_equation_0 = const()[name = string("op_36698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36698_cast_fp16 = einsum(equation = var_36698_equation_0, values = (var_36156_cast_fp16, var_36558_cast_fp16))[name = string("op_36698_cast_fp16")]; + string var_36700_equation_0 = const()[name = string("op_36700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36700_cast_fp16 = einsum(equation = var_36700_equation_0, values = (var_36160_cast_fp16, var_36559_cast_fp16))[name = string("op_36700_cast_fp16")]; + string var_36702_equation_0 = const()[name = string("op_36702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36702_cast_fp16 = einsum(equation = var_36702_equation_0, values = (var_36160_cast_fp16, var_36560_cast_fp16))[name = string("op_36702_cast_fp16")]; + string var_36704_equation_0 = const()[name = string("op_36704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36704_cast_fp16 = einsum(equation = var_36704_equation_0, values = (var_36160_cast_fp16, var_36561_cast_fp16))[name = string("op_36704_cast_fp16")]; + string var_36706_equation_0 = const()[name = string("op_36706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36706_cast_fp16 = einsum(equation = var_36706_equation_0, values = (var_36160_cast_fp16, var_36562_cast_fp16))[name = string("op_36706_cast_fp16")]; + string var_36708_equation_0 = const()[name = string("op_36708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36708_cast_fp16 = einsum(equation = var_36708_equation_0, values = (var_36164_cast_fp16, var_36563_cast_fp16))[name = string("op_36708_cast_fp16")]; + string var_36710_equation_0 = const()[name = string("op_36710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36710_cast_fp16 = einsum(equation = var_36710_equation_0, values = (var_36164_cast_fp16, var_36564_cast_fp16))[name = string("op_36710_cast_fp16")]; + string var_36712_equation_0 = const()[name = string("op_36712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36712_cast_fp16 = einsum(equation = var_36712_equation_0, values = (var_36164_cast_fp16, var_36565_cast_fp16))[name = string("op_36712_cast_fp16")]; + string var_36714_equation_0 = const()[name = string("op_36714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36714_cast_fp16 = einsum(equation = var_36714_equation_0, values = (var_36164_cast_fp16, var_36566_cast_fp16))[name = string("op_36714_cast_fp16")]; + string var_36716_equation_0 = const()[name = string("op_36716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36716_cast_fp16 = einsum(equation = var_36716_equation_0, values = (var_36168_cast_fp16, var_36567_cast_fp16))[name = string("op_36716_cast_fp16")]; + string var_36718_equation_0 = const()[name = string("op_36718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36718_cast_fp16 = einsum(equation = var_36718_equation_0, values = (var_36168_cast_fp16, var_36568_cast_fp16))[name = string("op_36718_cast_fp16")]; + string var_36720_equation_0 = const()[name = string("op_36720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36720_cast_fp16 = einsum(equation = var_36720_equation_0, values = (var_36168_cast_fp16, var_36569_cast_fp16))[name = string("op_36720_cast_fp16")]; + string var_36722_equation_0 = const()[name = string("op_36722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36722_cast_fp16 = einsum(equation = var_36722_equation_0, values = (var_36168_cast_fp16, var_36570_cast_fp16))[name = string("op_36722_cast_fp16")]; + string var_36724_equation_0 = const()[name = string("op_36724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36724_cast_fp16 = einsum(equation = var_36724_equation_0, values = (var_36172_cast_fp16, var_36571_cast_fp16))[name = string("op_36724_cast_fp16")]; + string var_36726_equation_0 = const()[name = string("op_36726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36726_cast_fp16 = einsum(equation = var_36726_equation_0, values = (var_36172_cast_fp16, var_36572_cast_fp16))[name = string("op_36726_cast_fp16")]; + string var_36728_equation_0 = const()[name = string("op_36728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36728_cast_fp16 = einsum(equation = var_36728_equation_0, values = (var_36172_cast_fp16, var_36573_cast_fp16))[name = string("op_36728_cast_fp16")]; + string var_36730_equation_0 = const()[name = string("op_36730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36730_cast_fp16 = einsum(equation = var_36730_equation_0, values = (var_36172_cast_fp16, var_36574_cast_fp16))[name = string("op_36730_cast_fp16")]; + string var_36732_equation_0 = const()[name = string("op_36732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36732_cast_fp16 = einsum(equation = var_36732_equation_0, values = (var_36176_cast_fp16, var_36575_cast_fp16))[name = string("op_36732_cast_fp16")]; + string var_36734_equation_0 = const()[name = string("op_36734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36734_cast_fp16 = einsum(equation = var_36734_equation_0, values = (var_36176_cast_fp16, var_36576_cast_fp16))[name = string("op_36734_cast_fp16")]; + string var_36736_equation_0 = const()[name = string("op_36736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36736_cast_fp16 = einsum(equation = var_36736_equation_0, values = (var_36176_cast_fp16, var_36577_cast_fp16))[name = string("op_36736_cast_fp16")]; + string var_36738_equation_0 = const()[name = string("op_36738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36738_cast_fp16 = einsum(equation = var_36738_equation_0, values = (var_36176_cast_fp16, var_36578_cast_fp16))[name = string("op_36738_cast_fp16")]; + bool var_36740_interleave_0 = const()[name = string("op_36740_interleave_0"), val = bool(false)]; + tensor var_36740_cast_fp16 = concat(axis = var_35299, interleave = var_36740_interleave_0, values = (var_36580_cast_fp16, var_36582_cast_fp16, var_36584_cast_fp16, var_36586_cast_fp16))[name = string("op_36740_cast_fp16")]; + bool var_36742_interleave_0 = const()[name = string("op_36742_interleave_0"), val = bool(false)]; + tensor var_36742_cast_fp16 = concat(axis = var_35299, interleave = var_36742_interleave_0, values = (var_36588_cast_fp16, var_36590_cast_fp16, var_36592_cast_fp16, var_36594_cast_fp16))[name = string("op_36742_cast_fp16")]; + bool var_36744_interleave_0 = const()[name = string("op_36744_interleave_0"), val = bool(false)]; + tensor var_36744_cast_fp16 = concat(axis = var_35299, interleave = var_36744_interleave_0, values = (var_36596_cast_fp16, var_36598_cast_fp16, var_36600_cast_fp16, var_36602_cast_fp16))[name = string("op_36744_cast_fp16")]; + bool var_36746_interleave_0 = const()[name = string("op_36746_interleave_0"), val = bool(false)]; + tensor var_36746_cast_fp16 = concat(axis = var_35299, interleave = var_36746_interleave_0, values = (var_36604_cast_fp16, var_36606_cast_fp16, var_36608_cast_fp16, var_36610_cast_fp16))[name = string("op_36746_cast_fp16")]; + bool var_36748_interleave_0 = const()[name = string("op_36748_interleave_0"), val = bool(false)]; + tensor var_36748_cast_fp16 = concat(axis = var_35299, interleave = var_36748_interleave_0, values = (var_36612_cast_fp16, var_36614_cast_fp16, var_36616_cast_fp16, var_36618_cast_fp16))[name = string("op_36748_cast_fp16")]; + bool var_36750_interleave_0 = const()[name = string("op_36750_interleave_0"), val = bool(false)]; + tensor var_36750_cast_fp16 = concat(axis = var_35299, interleave = var_36750_interleave_0, values = (var_36620_cast_fp16, var_36622_cast_fp16, var_36624_cast_fp16, var_36626_cast_fp16))[name = string("op_36750_cast_fp16")]; + bool var_36752_interleave_0 = const()[name = string("op_36752_interleave_0"), val = bool(false)]; + tensor var_36752_cast_fp16 = concat(axis = var_35299, interleave = var_36752_interleave_0, values = (var_36628_cast_fp16, var_36630_cast_fp16, var_36632_cast_fp16, var_36634_cast_fp16))[name = string("op_36752_cast_fp16")]; + bool var_36754_interleave_0 = const()[name = string("op_36754_interleave_0"), val = bool(false)]; + tensor var_36754_cast_fp16 = concat(axis = var_35299, interleave = var_36754_interleave_0, values = (var_36636_cast_fp16, var_36638_cast_fp16, var_36640_cast_fp16, var_36642_cast_fp16))[name = string("op_36754_cast_fp16")]; + bool var_36756_interleave_0 = const()[name = string("op_36756_interleave_0"), val = bool(false)]; + tensor var_36756_cast_fp16 = concat(axis = var_35299, interleave = var_36756_interleave_0, values = (var_36644_cast_fp16, var_36646_cast_fp16, var_36648_cast_fp16, var_36650_cast_fp16))[name = string("op_36756_cast_fp16")]; + bool var_36758_interleave_0 = const()[name = string("op_36758_interleave_0"), val = bool(false)]; + tensor var_36758_cast_fp16 = concat(axis = var_35299, interleave = var_36758_interleave_0, values = (var_36652_cast_fp16, var_36654_cast_fp16, var_36656_cast_fp16, var_36658_cast_fp16))[name = string("op_36758_cast_fp16")]; + bool var_36760_interleave_0 = const()[name = string("op_36760_interleave_0"), val = bool(false)]; + tensor var_36760_cast_fp16 = concat(axis = var_35299, interleave = var_36760_interleave_0, values = (var_36660_cast_fp16, var_36662_cast_fp16, var_36664_cast_fp16, var_36666_cast_fp16))[name = string("op_36760_cast_fp16")]; + bool var_36762_interleave_0 = const()[name = string("op_36762_interleave_0"), val = bool(false)]; + tensor var_36762_cast_fp16 = concat(axis = var_35299, interleave = var_36762_interleave_0, values = (var_36668_cast_fp16, var_36670_cast_fp16, var_36672_cast_fp16, var_36674_cast_fp16))[name = string("op_36762_cast_fp16")]; + bool var_36764_interleave_0 = const()[name = string("op_36764_interleave_0"), val = bool(false)]; + tensor var_36764_cast_fp16 = concat(axis = var_35299, interleave = var_36764_interleave_0, values = (var_36676_cast_fp16, var_36678_cast_fp16, var_36680_cast_fp16, var_36682_cast_fp16))[name = string("op_36764_cast_fp16")]; + bool var_36766_interleave_0 = const()[name = string("op_36766_interleave_0"), val = bool(false)]; + tensor var_36766_cast_fp16 = concat(axis = var_35299, interleave = var_36766_interleave_0, values = (var_36684_cast_fp16, var_36686_cast_fp16, var_36688_cast_fp16, var_36690_cast_fp16))[name = string("op_36766_cast_fp16")]; + bool var_36768_interleave_0 = const()[name = string("op_36768_interleave_0"), val = bool(false)]; + tensor var_36768_cast_fp16 = concat(axis = var_35299, interleave = var_36768_interleave_0, values = (var_36692_cast_fp16, var_36694_cast_fp16, var_36696_cast_fp16, var_36698_cast_fp16))[name = string("op_36768_cast_fp16")]; + bool var_36770_interleave_0 = const()[name = string("op_36770_interleave_0"), val = bool(false)]; + tensor var_36770_cast_fp16 = concat(axis = var_35299, interleave = var_36770_interleave_0, values = (var_36700_cast_fp16, var_36702_cast_fp16, var_36704_cast_fp16, var_36706_cast_fp16))[name = string("op_36770_cast_fp16")]; + bool var_36772_interleave_0 = const()[name = string("op_36772_interleave_0"), val = bool(false)]; + tensor var_36772_cast_fp16 = concat(axis = var_35299, interleave = var_36772_interleave_0, values = (var_36708_cast_fp16, var_36710_cast_fp16, var_36712_cast_fp16, var_36714_cast_fp16))[name = string("op_36772_cast_fp16")]; + bool var_36774_interleave_0 = const()[name = string("op_36774_interleave_0"), val = bool(false)]; + tensor var_36774_cast_fp16 = concat(axis = var_35299, interleave = var_36774_interleave_0, values = (var_36716_cast_fp16, var_36718_cast_fp16, var_36720_cast_fp16, var_36722_cast_fp16))[name = string("op_36774_cast_fp16")]; + bool var_36776_interleave_0 = const()[name = string("op_36776_interleave_0"), val = bool(false)]; + tensor var_36776_cast_fp16 = concat(axis = var_35299, interleave = var_36776_interleave_0, values = (var_36724_cast_fp16, var_36726_cast_fp16, var_36728_cast_fp16, var_36730_cast_fp16))[name = string("op_36776_cast_fp16")]; + bool var_36778_interleave_0 = const()[name = string("op_36778_interleave_0"), val = bool(false)]; + tensor var_36778_cast_fp16 = concat(axis = var_35299, interleave = var_36778_interleave_0, values = (var_36732_cast_fp16, var_36734_cast_fp16, var_36736_cast_fp16, var_36738_cast_fp16))[name = string("op_36778_cast_fp16")]; + bool input_185_interleave_0 = const()[name = string("input_185_interleave_0"), val = bool(false)]; + tensor input_185_cast_fp16 = concat(axis = var_35324, interleave = input_185_interleave_0, values = (var_36740_cast_fp16, var_36742_cast_fp16, var_36744_cast_fp16, var_36746_cast_fp16, var_36748_cast_fp16, var_36750_cast_fp16, var_36752_cast_fp16, var_36754_cast_fp16, var_36756_cast_fp16, var_36758_cast_fp16, var_36760_cast_fp16, var_36762_cast_fp16, var_36764_cast_fp16, var_36766_cast_fp16, var_36768_cast_fp16, var_36770_cast_fp16, var_36772_cast_fp16, var_36774_cast_fp16, var_36776_cast_fp16, var_36778_cast_fp16))[name = string("input_185_cast_fp16")]; + string obj_95_pad_type_0 = const()[name = string("obj_95_pad_type_0"), val = string("valid")]; + tensor obj_95_strides_0 = const()[name = string("obj_95_strides_0"), val = tensor([1, 1])]; + tensor obj_95_pad_0 = const()[name = string("obj_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_95_dilations_0 = const()[name = string("obj_95_dilations_0"), val = tensor([1, 1])]; + int32 obj_95_groups_0 = const()[name = string("obj_95_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929630720)))]; + tensor layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932907584)))]; + tensor obj_95_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_95_cast_fp16)[name = string("inputs_95_cast_fp16")]; + tensor out_95_axes_0 = const()[name = string("out_95_axes_0"), val = tensor([1])]; + fp16 var_36797_to_fp16 = const()[name = string("op_36797_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_36797_to_fp16, x = inputs_95_cast_fp16)[name = string("out_95_cast_fp16")]; + tensor input_187_gamma_0_to_fp16 = const()[name = string("input_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932910208)))]; + tensor input_187_beta_0_to_fp16 = const()[name = string("input_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932912832)))]; + fp16 input_187_epsilon_0_to_fp16 = const()[name = string("input_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_187_cast_fp16 = batch_norm(beta = input_187_beta_0_to_fp16, epsilon = input_187_epsilon_0_to_fp16, gamma = input_187_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = string("input_187_cast_fp16")]; + string input_189_pad_type_0 = const()[name = string("input_189_pad_type_0"), val = string("valid")]; + tensor input_189_strides_0 = const()[name = string("input_189_strides_0"), val = tensor([1, 1])]; + tensor input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_189_dilations_0 = const()[name = string("input_189_dilations_0"), val = tensor([1, 1])]; + int32 input_189_groups_0 = const()[name = string("input_189_groups_0"), val = int32(1)]; + tensor layers_23_fc1_weight_to_fp16 = const()[name = string("layers_23_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932915456)))]; + tensor layers_23_fc1_bias_to_fp16 = const()[name = string("layers_23_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946022720)))]; + tensor input_189_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = layers_23_fc1_weight_to_fp16, x = input_187_cast_fp16)[name = string("input_189_cast_fp16")]; + string input_191_mode_0 = const()[name = string("input_191_mode_0"), val = string("EXACT")]; + tensor input_191_cast_fp16 = gelu(mode = input_191_mode_0, x = input_189_cast_fp16)[name = string("input_191_cast_fp16")]; + string hidden_states_51_pad_type_0 = const()[name = string("hidden_states_51_pad_type_0"), val = string("valid")]; + tensor hidden_states_51_strides_0 = const()[name = string("hidden_states_51_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_51_pad_0 = const()[name = string("hidden_states_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_51_dilations_0 = const()[name = string("hidden_states_51_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_51_groups_0 = const()[name = string("hidden_states_51_groups_0"), val = int32(1)]; + tensor layers_23_fc2_weight_to_fp16 = const()[name = string("layers_23_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946033024)))]; + tensor layers_23_fc2_bias_to_fp16 = const()[name = string("layers_23_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959140288)))]; + tensor hidden_states_51_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_51_dilations_0, groups = hidden_states_51_groups_0, pad = hidden_states_51_pad_0, pad_type = hidden_states_51_pad_type_0, strides = hidden_states_51_strides_0, weight = layers_23_fc2_weight_to_fp16, x = input_191_cast_fp16)[name = string("hidden_states_51_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_51_cast_fp16)[name = string("inputs_97_cast_fp16")]; + int32 var_36826 = const()[name = string("op_36826"), val = int32(3)]; + int32 var_36851 = const()[name = string("op_36851"), val = int32(1)]; + tensor out_97_axes_0 = const()[name = string("out_97_axes_0"), val = tensor([1])]; + fp16 var_36868_to_fp16 = const()[name = string("op_36868_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_36868_to_fp16, x = inputs_97_cast_fp16)[name = string("out_97_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959142912)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959145536)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_49_pad_type_0 = const()[name = string("query_49_pad_type_0"), val = string("valid")]; + tensor query_49_strides_0 = const()[name = string("query_49_strides_0"), val = tensor([1, 1])]; + tensor query_49_pad_0 = const()[name = string("query_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_49_dilations_0 = const()[name = string("query_49_dilations_0"), val = tensor([1, 1])]; + int32 query_49_groups_0 = const()[name = string("query_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959148160)))]; + tensor layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962425024)))]; + tensor query_49_cast_fp16 = conv(bias = layers_24_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_24_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_49_cast_fp16")]; + string key_49_pad_type_0 = const()[name = string("key_49_pad_type_0"), val = string("valid")]; + tensor key_49_strides_0 = const()[name = string("key_49_strides_0"), val = tensor([1, 1])]; + tensor key_49_pad_0 = const()[name = string("key_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_49_dilations_0 = const()[name = string("key_49_dilations_0"), val = tensor([1, 1])]; + int32 key_49_groups_0 = const()[name = string("key_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962427648)))]; + tensor key_49_cast_fp16 = conv(dilations = key_49_dilations_0, groups = key_49_groups_0, pad = key_49_pad_0, pad_type = key_49_pad_type_0, strides = key_49_strides_0, weight = layers_24_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("key_49_cast_fp16")]; + string value_49_pad_type_0 = const()[name = string("value_49_pad_type_0"), val = string("valid")]; + tensor value_49_strides_0 = const()[name = string("value_49_strides_0"), val = tensor([1, 1])]; + tensor value_49_pad_0 = const()[name = string("value_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_49_dilations_0 = const()[name = string("value_49_dilations_0"), val = tensor([1, 1])]; + int32 value_49_groups_0 = const()[name = string("value_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(965704512)))]; + tensor layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968981376)))]; + tensor value_49_cast_fp16 = conv(bias = layers_24_self_attn_v_proj_bias_to_fp16, dilations = value_49_dilations_0, groups = value_49_groups_0, pad = value_49_pad_0, pad_type = value_49_pad_type_0, strides = value_49_strides_0, weight = layers_24_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("value_49_cast_fp16")]; + tensor var_36906_begin_0 = const()[name = string("op_36906_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36906_end_0 = const()[name = string("op_36906_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36906_end_mask_0 = const()[name = string("op_36906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36906_cast_fp16 = slice_by_index(begin = var_36906_begin_0, end = var_36906_end_0, end_mask = var_36906_end_mask_0, x = query_49_cast_fp16)[name = string("op_36906_cast_fp16")]; + tensor var_36910_begin_0 = const()[name = string("op_36910_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36910_end_0 = const()[name = string("op_36910_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_36910_end_mask_0 = const()[name = string("op_36910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36910_cast_fp16 = slice_by_index(begin = var_36910_begin_0, end = var_36910_end_0, end_mask = var_36910_end_mask_0, x = query_49_cast_fp16)[name = string("op_36910_cast_fp16")]; + tensor var_36914_begin_0 = const()[name = string("op_36914_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36914_end_0 = const()[name = string("op_36914_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_36914_end_mask_0 = const()[name = string("op_36914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36914_cast_fp16 = slice_by_index(begin = var_36914_begin_0, end = var_36914_end_0, end_mask = var_36914_end_mask_0, x = query_49_cast_fp16)[name = string("op_36914_cast_fp16")]; + tensor var_36918_begin_0 = const()[name = string("op_36918_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36918_end_0 = const()[name = string("op_36918_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_36918_end_mask_0 = const()[name = string("op_36918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36918_cast_fp16 = slice_by_index(begin = var_36918_begin_0, end = var_36918_end_0, end_mask = var_36918_end_mask_0, x = query_49_cast_fp16)[name = string("op_36918_cast_fp16")]; + tensor var_36922_begin_0 = const()[name = string("op_36922_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36922_end_0 = const()[name = string("op_36922_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_36922_end_mask_0 = const()[name = string("op_36922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36922_cast_fp16 = slice_by_index(begin = var_36922_begin_0, end = var_36922_end_0, end_mask = var_36922_end_mask_0, x = query_49_cast_fp16)[name = string("op_36922_cast_fp16")]; + tensor var_36926_begin_0 = const()[name = string("op_36926_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36926_end_0 = const()[name = string("op_36926_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_36926_end_mask_0 = const()[name = string("op_36926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36926_cast_fp16 = slice_by_index(begin = var_36926_begin_0, end = var_36926_end_0, end_mask = var_36926_end_mask_0, x = query_49_cast_fp16)[name = string("op_36926_cast_fp16")]; + tensor var_36930_begin_0 = const()[name = string("op_36930_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36930_end_0 = const()[name = string("op_36930_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_36930_end_mask_0 = const()[name = string("op_36930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36930_cast_fp16 = slice_by_index(begin = var_36930_begin_0, end = var_36930_end_0, end_mask = var_36930_end_mask_0, x = query_49_cast_fp16)[name = string("op_36930_cast_fp16")]; + tensor var_36934_begin_0 = const()[name = string("op_36934_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36934_end_0 = const()[name = string("op_36934_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_36934_end_mask_0 = const()[name = string("op_36934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36934_cast_fp16 = slice_by_index(begin = var_36934_begin_0, end = var_36934_end_0, end_mask = var_36934_end_mask_0, x = query_49_cast_fp16)[name = string("op_36934_cast_fp16")]; + tensor var_36938_begin_0 = const()[name = string("op_36938_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36938_end_0 = const()[name = string("op_36938_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_36938_end_mask_0 = const()[name = string("op_36938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36938_cast_fp16 = slice_by_index(begin = var_36938_begin_0, end = var_36938_end_0, end_mask = var_36938_end_mask_0, x = query_49_cast_fp16)[name = string("op_36938_cast_fp16")]; + tensor var_36942_begin_0 = const()[name = string("op_36942_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36942_end_0 = const()[name = string("op_36942_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_36942_end_mask_0 = const()[name = string("op_36942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36942_cast_fp16 = slice_by_index(begin = var_36942_begin_0, end = var_36942_end_0, end_mask = var_36942_end_mask_0, x = query_49_cast_fp16)[name = string("op_36942_cast_fp16")]; + tensor var_36946_begin_0 = const()[name = string("op_36946_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36946_end_0 = const()[name = string("op_36946_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_36946_end_mask_0 = const()[name = string("op_36946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36946_cast_fp16 = slice_by_index(begin = var_36946_begin_0, end = var_36946_end_0, end_mask = var_36946_end_mask_0, x = query_49_cast_fp16)[name = string("op_36946_cast_fp16")]; + tensor var_36950_begin_0 = const()[name = string("op_36950_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36950_end_0 = const()[name = string("op_36950_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_36950_end_mask_0 = const()[name = string("op_36950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36950_cast_fp16 = slice_by_index(begin = var_36950_begin_0, end = var_36950_end_0, end_mask = var_36950_end_mask_0, x = query_49_cast_fp16)[name = string("op_36950_cast_fp16")]; + tensor var_36954_begin_0 = const()[name = string("op_36954_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36954_end_0 = const()[name = string("op_36954_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_36954_end_mask_0 = const()[name = string("op_36954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36954_cast_fp16 = slice_by_index(begin = var_36954_begin_0, end = var_36954_end_0, end_mask = var_36954_end_mask_0, x = query_49_cast_fp16)[name = string("op_36954_cast_fp16")]; + tensor var_36958_begin_0 = const()[name = string("op_36958_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36958_end_0 = const()[name = string("op_36958_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_36958_end_mask_0 = const()[name = string("op_36958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36958_cast_fp16 = slice_by_index(begin = var_36958_begin_0, end = var_36958_end_0, end_mask = var_36958_end_mask_0, x = query_49_cast_fp16)[name = string("op_36958_cast_fp16")]; + tensor var_36962_begin_0 = const()[name = string("op_36962_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36962_end_0 = const()[name = string("op_36962_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_36962_end_mask_0 = const()[name = string("op_36962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36962_cast_fp16 = slice_by_index(begin = var_36962_begin_0, end = var_36962_end_0, end_mask = var_36962_end_mask_0, x = query_49_cast_fp16)[name = string("op_36962_cast_fp16")]; + tensor var_36966_begin_0 = const()[name = string("op_36966_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36966_end_0 = const()[name = string("op_36966_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_36966_end_mask_0 = const()[name = string("op_36966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36966_cast_fp16 = slice_by_index(begin = var_36966_begin_0, end = var_36966_end_0, end_mask = var_36966_end_mask_0, x = query_49_cast_fp16)[name = string("op_36966_cast_fp16")]; + tensor var_36970_begin_0 = const()[name = string("op_36970_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36970_end_0 = const()[name = string("op_36970_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_36970_end_mask_0 = const()[name = string("op_36970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36970_cast_fp16 = slice_by_index(begin = var_36970_begin_0, end = var_36970_end_0, end_mask = var_36970_end_mask_0, x = query_49_cast_fp16)[name = string("op_36970_cast_fp16")]; + tensor var_36974_begin_0 = const()[name = string("op_36974_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36974_end_0 = const()[name = string("op_36974_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_36974_end_mask_0 = const()[name = string("op_36974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36974_cast_fp16 = slice_by_index(begin = var_36974_begin_0, end = var_36974_end_0, end_mask = var_36974_end_mask_0, x = query_49_cast_fp16)[name = string("op_36974_cast_fp16")]; + tensor var_36978_begin_0 = const()[name = string("op_36978_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36978_end_0 = const()[name = string("op_36978_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_36978_end_mask_0 = const()[name = string("op_36978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36978_cast_fp16 = slice_by_index(begin = var_36978_begin_0, end = var_36978_end_0, end_mask = var_36978_end_mask_0, x = query_49_cast_fp16)[name = string("op_36978_cast_fp16")]; + tensor var_36982_begin_0 = const()[name = string("op_36982_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36982_end_0 = const()[name = string("op_36982_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_36982_end_mask_0 = const()[name = string("op_36982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36982_cast_fp16 = slice_by_index(begin = var_36982_begin_0, end = var_36982_end_0, end_mask = var_36982_end_mask_0, x = query_49_cast_fp16)[name = string("op_36982_cast_fp16")]; + tensor var_36991_begin_0 = const()[name = string("op_36991_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36991_end_0 = const()[name = string("op_36991_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36991_end_mask_0 = const()[name = string("op_36991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36991_cast_fp16 = slice_by_index(begin = var_36991_begin_0, end = var_36991_end_0, end_mask = var_36991_end_mask_0, x = var_36906_cast_fp16)[name = string("op_36991_cast_fp16")]; + tensor var_36998_begin_0 = const()[name = string("op_36998_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36998_end_0 = const()[name = string("op_36998_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36998_end_mask_0 = const()[name = string("op_36998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36998_cast_fp16 = slice_by_index(begin = var_36998_begin_0, end = var_36998_end_0, end_mask = var_36998_end_mask_0, x = var_36906_cast_fp16)[name = string("op_36998_cast_fp16")]; + tensor var_37005_begin_0 = const()[name = string("op_37005_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37005_end_0 = const()[name = string("op_37005_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37005_end_mask_0 = const()[name = string("op_37005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37005_cast_fp16 = slice_by_index(begin = var_37005_begin_0, end = var_37005_end_0, end_mask = var_37005_end_mask_0, x = var_36906_cast_fp16)[name = string("op_37005_cast_fp16")]; + tensor var_37012_begin_0 = const()[name = string("op_37012_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37012_end_0 = const()[name = string("op_37012_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37012_end_mask_0 = const()[name = string("op_37012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37012_cast_fp16 = slice_by_index(begin = var_37012_begin_0, end = var_37012_end_0, end_mask = var_37012_end_mask_0, x = var_36906_cast_fp16)[name = string("op_37012_cast_fp16")]; + tensor var_37019_begin_0 = const()[name = string("op_37019_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37019_end_0 = const()[name = string("op_37019_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37019_end_mask_0 = const()[name = string("op_37019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37019_cast_fp16 = slice_by_index(begin = var_37019_begin_0, end = var_37019_end_0, end_mask = var_37019_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37019_cast_fp16")]; + tensor var_37026_begin_0 = const()[name = string("op_37026_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37026_end_0 = const()[name = string("op_37026_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37026_end_mask_0 = const()[name = string("op_37026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37026_cast_fp16 = slice_by_index(begin = var_37026_begin_0, end = var_37026_end_0, end_mask = var_37026_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37026_cast_fp16")]; + tensor var_37033_begin_0 = const()[name = string("op_37033_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37033_end_0 = const()[name = string("op_37033_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37033_end_mask_0 = const()[name = string("op_37033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37033_cast_fp16 = slice_by_index(begin = var_37033_begin_0, end = var_37033_end_0, end_mask = var_37033_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37033_cast_fp16")]; + tensor var_37040_begin_0 = const()[name = string("op_37040_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37040_end_0 = const()[name = string("op_37040_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37040_end_mask_0 = const()[name = string("op_37040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37040_cast_fp16 = slice_by_index(begin = var_37040_begin_0, end = var_37040_end_0, end_mask = var_37040_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37040_cast_fp16")]; + tensor var_37047_begin_0 = const()[name = string("op_37047_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37047_end_0 = const()[name = string("op_37047_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37047_end_mask_0 = const()[name = string("op_37047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37047_cast_fp16 = slice_by_index(begin = var_37047_begin_0, end = var_37047_end_0, end_mask = var_37047_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37047_cast_fp16")]; + tensor var_37054_begin_0 = const()[name = string("op_37054_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37054_end_0 = const()[name = string("op_37054_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37054_end_mask_0 = const()[name = string("op_37054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37054_cast_fp16 = slice_by_index(begin = var_37054_begin_0, end = var_37054_end_0, end_mask = var_37054_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37054_cast_fp16")]; + tensor var_37061_begin_0 = const()[name = string("op_37061_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37061_end_0 = const()[name = string("op_37061_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37061_end_mask_0 = const()[name = string("op_37061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37061_cast_fp16 = slice_by_index(begin = var_37061_begin_0, end = var_37061_end_0, end_mask = var_37061_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37061_cast_fp16")]; + tensor var_37068_begin_0 = const()[name = string("op_37068_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37068_end_0 = const()[name = string("op_37068_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37068_end_mask_0 = const()[name = string("op_37068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37068_cast_fp16 = slice_by_index(begin = var_37068_begin_0, end = var_37068_end_0, end_mask = var_37068_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37068_cast_fp16")]; + tensor var_37075_begin_0 = const()[name = string("op_37075_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37075_end_0 = const()[name = string("op_37075_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37075_end_mask_0 = const()[name = string("op_37075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37075_cast_fp16 = slice_by_index(begin = var_37075_begin_0, end = var_37075_end_0, end_mask = var_37075_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37075_cast_fp16")]; + tensor var_37082_begin_0 = const()[name = string("op_37082_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37082_end_0 = const()[name = string("op_37082_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37082_end_mask_0 = const()[name = string("op_37082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37082_cast_fp16 = slice_by_index(begin = var_37082_begin_0, end = var_37082_end_0, end_mask = var_37082_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37082_cast_fp16")]; + tensor var_37089_begin_0 = const()[name = string("op_37089_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37089_end_0 = const()[name = string("op_37089_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37089_end_mask_0 = const()[name = string("op_37089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37089_cast_fp16 = slice_by_index(begin = var_37089_begin_0, end = var_37089_end_0, end_mask = var_37089_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37089_cast_fp16")]; + tensor var_37096_begin_0 = const()[name = string("op_37096_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37096_end_0 = const()[name = string("op_37096_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37096_end_mask_0 = const()[name = string("op_37096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37096_cast_fp16 = slice_by_index(begin = var_37096_begin_0, end = var_37096_end_0, end_mask = var_37096_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37096_cast_fp16")]; + tensor var_37103_begin_0 = const()[name = string("op_37103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37103_end_0 = const()[name = string("op_37103_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37103_end_mask_0 = const()[name = string("op_37103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37103_cast_fp16 = slice_by_index(begin = var_37103_begin_0, end = var_37103_end_0, end_mask = var_37103_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37103_cast_fp16")]; + tensor var_37110_begin_0 = const()[name = string("op_37110_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37110_end_0 = const()[name = string("op_37110_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37110_end_mask_0 = const()[name = string("op_37110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37110_cast_fp16 = slice_by_index(begin = var_37110_begin_0, end = var_37110_end_0, end_mask = var_37110_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37110_cast_fp16")]; + tensor var_37117_begin_0 = const()[name = string("op_37117_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37117_end_0 = const()[name = string("op_37117_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37117_end_mask_0 = const()[name = string("op_37117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37117_cast_fp16 = slice_by_index(begin = var_37117_begin_0, end = var_37117_end_0, end_mask = var_37117_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37117_cast_fp16")]; + tensor var_37124_begin_0 = const()[name = string("op_37124_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37124_end_0 = const()[name = string("op_37124_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37124_end_mask_0 = const()[name = string("op_37124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37124_cast_fp16 = slice_by_index(begin = var_37124_begin_0, end = var_37124_end_0, end_mask = var_37124_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37124_cast_fp16")]; + tensor var_37131_begin_0 = const()[name = string("op_37131_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37131_end_0 = const()[name = string("op_37131_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37131_end_mask_0 = const()[name = string("op_37131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37131_cast_fp16 = slice_by_index(begin = var_37131_begin_0, end = var_37131_end_0, end_mask = var_37131_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37131_cast_fp16")]; + tensor var_37138_begin_0 = const()[name = string("op_37138_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37138_end_0 = const()[name = string("op_37138_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37138_end_mask_0 = const()[name = string("op_37138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37138_cast_fp16 = slice_by_index(begin = var_37138_begin_0, end = var_37138_end_0, end_mask = var_37138_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37138_cast_fp16")]; + tensor var_37145_begin_0 = const()[name = string("op_37145_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37145_end_0 = const()[name = string("op_37145_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37145_end_mask_0 = const()[name = string("op_37145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37145_cast_fp16 = slice_by_index(begin = var_37145_begin_0, end = var_37145_end_0, end_mask = var_37145_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37145_cast_fp16")]; + tensor var_37152_begin_0 = const()[name = string("op_37152_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37152_end_0 = const()[name = string("op_37152_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37152_end_mask_0 = const()[name = string("op_37152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37152_cast_fp16 = slice_by_index(begin = var_37152_begin_0, end = var_37152_end_0, end_mask = var_37152_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37152_cast_fp16")]; + tensor var_37159_begin_0 = const()[name = string("op_37159_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37159_end_0 = const()[name = string("op_37159_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37159_end_mask_0 = const()[name = string("op_37159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37159_cast_fp16 = slice_by_index(begin = var_37159_begin_0, end = var_37159_end_0, end_mask = var_37159_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37159_cast_fp16")]; + tensor var_37166_begin_0 = const()[name = string("op_37166_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37166_end_0 = const()[name = string("op_37166_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37166_end_mask_0 = const()[name = string("op_37166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37166_cast_fp16 = slice_by_index(begin = var_37166_begin_0, end = var_37166_end_0, end_mask = var_37166_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37166_cast_fp16")]; + tensor var_37173_begin_0 = const()[name = string("op_37173_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37173_end_0 = const()[name = string("op_37173_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37173_end_mask_0 = const()[name = string("op_37173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37173_cast_fp16 = slice_by_index(begin = var_37173_begin_0, end = var_37173_end_0, end_mask = var_37173_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37173_cast_fp16")]; + tensor var_37180_begin_0 = const()[name = string("op_37180_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37180_end_0 = const()[name = string("op_37180_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37180_end_mask_0 = const()[name = string("op_37180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37180_cast_fp16 = slice_by_index(begin = var_37180_begin_0, end = var_37180_end_0, end_mask = var_37180_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37180_cast_fp16")]; + tensor var_37187_begin_0 = const()[name = string("op_37187_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37187_end_0 = const()[name = string("op_37187_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37187_end_mask_0 = const()[name = string("op_37187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37187_cast_fp16 = slice_by_index(begin = var_37187_begin_0, end = var_37187_end_0, end_mask = var_37187_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37187_cast_fp16")]; + tensor var_37194_begin_0 = const()[name = string("op_37194_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37194_end_0 = const()[name = string("op_37194_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37194_end_mask_0 = const()[name = string("op_37194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37194_cast_fp16 = slice_by_index(begin = var_37194_begin_0, end = var_37194_end_0, end_mask = var_37194_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37194_cast_fp16")]; + tensor var_37201_begin_0 = const()[name = string("op_37201_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37201_end_0 = const()[name = string("op_37201_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37201_end_mask_0 = const()[name = string("op_37201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37201_cast_fp16 = slice_by_index(begin = var_37201_begin_0, end = var_37201_end_0, end_mask = var_37201_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37201_cast_fp16")]; + tensor var_37208_begin_0 = const()[name = string("op_37208_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37208_end_0 = const()[name = string("op_37208_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37208_end_mask_0 = const()[name = string("op_37208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37208_cast_fp16 = slice_by_index(begin = var_37208_begin_0, end = var_37208_end_0, end_mask = var_37208_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37208_cast_fp16")]; + tensor var_37215_begin_0 = const()[name = string("op_37215_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37215_end_0 = const()[name = string("op_37215_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37215_end_mask_0 = const()[name = string("op_37215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37215_cast_fp16 = slice_by_index(begin = var_37215_begin_0, end = var_37215_end_0, end_mask = var_37215_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37215_cast_fp16")]; + tensor var_37222_begin_0 = const()[name = string("op_37222_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37222_end_0 = const()[name = string("op_37222_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37222_end_mask_0 = const()[name = string("op_37222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37222_cast_fp16 = slice_by_index(begin = var_37222_begin_0, end = var_37222_end_0, end_mask = var_37222_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37222_cast_fp16")]; + tensor var_37229_begin_0 = const()[name = string("op_37229_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37229_end_0 = const()[name = string("op_37229_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37229_end_mask_0 = const()[name = string("op_37229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37229_cast_fp16 = slice_by_index(begin = var_37229_begin_0, end = var_37229_end_0, end_mask = var_37229_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37229_cast_fp16")]; + tensor var_37236_begin_0 = const()[name = string("op_37236_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37236_end_0 = const()[name = string("op_37236_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37236_end_mask_0 = const()[name = string("op_37236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37236_cast_fp16 = slice_by_index(begin = var_37236_begin_0, end = var_37236_end_0, end_mask = var_37236_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37236_cast_fp16")]; + tensor var_37243_begin_0 = const()[name = string("op_37243_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37243_end_0 = const()[name = string("op_37243_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37243_end_mask_0 = const()[name = string("op_37243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37243_cast_fp16 = slice_by_index(begin = var_37243_begin_0, end = var_37243_end_0, end_mask = var_37243_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37243_cast_fp16")]; + tensor var_37250_begin_0 = const()[name = string("op_37250_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37250_end_0 = const()[name = string("op_37250_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37250_end_mask_0 = const()[name = string("op_37250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37250_cast_fp16 = slice_by_index(begin = var_37250_begin_0, end = var_37250_end_0, end_mask = var_37250_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37250_cast_fp16")]; + tensor var_37257_begin_0 = const()[name = string("op_37257_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37257_end_0 = const()[name = string("op_37257_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37257_end_mask_0 = const()[name = string("op_37257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37257_cast_fp16 = slice_by_index(begin = var_37257_begin_0, end = var_37257_end_0, end_mask = var_37257_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37257_cast_fp16")]; + tensor var_37264_begin_0 = const()[name = string("op_37264_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37264_end_0 = const()[name = string("op_37264_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37264_end_mask_0 = const()[name = string("op_37264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37264_cast_fp16 = slice_by_index(begin = var_37264_begin_0, end = var_37264_end_0, end_mask = var_37264_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37264_cast_fp16")]; + tensor var_37271_begin_0 = const()[name = string("op_37271_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37271_end_0 = const()[name = string("op_37271_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37271_end_mask_0 = const()[name = string("op_37271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37271_cast_fp16 = slice_by_index(begin = var_37271_begin_0, end = var_37271_end_0, end_mask = var_37271_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37271_cast_fp16")]; + tensor var_37278_begin_0 = const()[name = string("op_37278_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37278_end_0 = const()[name = string("op_37278_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37278_end_mask_0 = const()[name = string("op_37278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37278_cast_fp16 = slice_by_index(begin = var_37278_begin_0, end = var_37278_end_0, end_mask = var_37278_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37278_cast_fp16")]; + tensor var_37285_begin_0 = const()[name = string("op_37285_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37285_end_0 = const()[name = string("op_37285_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37285_end_mask_0 = const()[name = string("op_37285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37285_cast_fp16 = slice_by_index(begin = var_37285_begin_0, end = var_37285_end_0, end_mask = var_37285_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37285_cast_fp16")]; + tensor var_37292_begin_0 = const()[name = string("op_37292_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37292_end_0 = const()[name = string("op_37292_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37292_end_mask_0 = const()[name = string("op_37292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37292_cast_fp16 = slice_by_index(begin = var_37292_begin_0, end = var_37292_end_0, end_mask = var_37292_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37292_cast_fp16")]; + tensor var_37299_begin_0 = const()[name = string("op_37299_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37299_end_0 = const()[name = string("op_37299_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37299_end_mask_0 = const()[name = string("op_37299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37299_cast_fp16 = slice_by_index(begin = var_37299_begin_0, end = var_37299_end_0, end_mask = var_37299_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37299_cast_fp16")]; + tensor var_37306_begin_0 = const()[name = string("op_37306_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37306_end_0 = const()[name = string("op_37306_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37306_end_mask_0 = const()[name = string("op_37306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37306_cast_fp16 = slice_by_index(begin = var_37306_begin_0, end = var_37306_end_0, end_mask = var_37306_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37306_cast_fp16")]; + tensor var_37313_begin_0 = const()[name = string("op_37313_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37313_end_0 = const()[name = string("op_37313_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37313_end_mask_0 = const()[name = string("op_37313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37313_cast_fp16 = slice_by_index(begin = var_37313_begin_0, end = var_37313_end_0, end_mask = var_37313_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37313_cast_fp16")]; + tensor var_37320_begin_0 = const()[name = string("op_37320_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37320_end_0 = const()[name = string("op_37320_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37320_end_mask_0 = const()[name = string("op_37320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37320_cast_fp16 = slice_by_index(begin = var_37320_begin_0, end = var_37320_end_0, end_mask = var_37320_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37320_cast_fp16")]; + tensor var_37327_begin_0 = const()[name = string("op_37327_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37327_end_0 = const()[name = string("op_37327_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37327_end_mask_0 = const()[name = string("op_37327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37327_cast_fp16 = slice_by_index(begin = var_37327_begin_0, end = var_37327_end_0, end_mask = var_37327_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37327_cast_fp16")]; + tensor var_37334_begin_0 = const()[name = string("op_37334_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37334_end_0 = const()[name = string("op_37334_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37334_end_mask_0 = const()[name = string("op_37334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37334_cast_fp16 = slice_by_index(begin = var_37334_begin_0, end = var_37334_end_0, end_mask = var_37334_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37334_cast_fp16")]; + tensor var_37341_begin_0 = const()[name = string("op_37341_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37341_end_0 = const()[name = string("op_37341_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37341_end_mask_0 = const()[name = string("op_37341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37341_cast_fp16 = slice_by_index(begin = var_37341_begin_0, end = var_37341_end_0, end_mask = var_37341_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37341_cast_fp16")]; + tensor var_37348_begin_0 = const()[name = string("op_37348_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37348_end_0 = const()[name = string("op_37348_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37348_end_mask_0 = const()[name = string("op_37348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37348_cast_fp16 = slice_by_index(begin = var_37348_begin_0, end = var_37348_end_0, end_mask = var_37348_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37348_cast_fp16")]; + tensor var_37355_begin_0 = const()[name = string("op_37355_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37355_end_0 = const()[name = string("op_37355_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37355_end_mask_0 = const()[name = string("op_37355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37355_cast_fp16 = slice_by_index(begin = var_37355_begin_0, end = var_37355_end_0, end_mask = var_37355_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37355_cast_fp16")]; + tensor var_37362_begin_0 = const()[name = string("op_37362_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37362_end_0 = const()[name = string("op_37362_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37362_end_mask_0 = const()[name = string("op_37362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37362_cast_fp16 = slice_by_index(begin = var_37362_begin_0, end = var_37362_end_0, end_mask = var_37362_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37362_cast_fp16")]; + tensor var_37369_begin_0 = const()[name = string("op_37369_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37369_end_0 = const()[name = string("op_37369_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37369_end_mask_0 = const()[name = string("op_37369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37369_cast_fp16 = slice_by_index(begin = var_37369_begin_0, end = var_37369_end_0, end_mask = var_37369_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37369_cast_fp16")]; + tensor var_37376_begin_0 = const()[name = string("op_37376_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37376_end_0 = const()[name = string("op_37376_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37376_end_mask_0 = const()[name = string("op_37376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37376_cast_fp16 = slice_by_index(begin = var_37376_begin_0, end = var_37376_end_0, end_mask = var_37376_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37376_cast_fp16")]; + tensor var_37383_begin_0 = const()[name = string("op_37383_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37383_end_0 = const()[name = string("op_37383_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37383_end_mask_0 = const()[name = string("op_37383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37383_cast_fp16 = slice_by_index(begin = var_37383_begin_0, end = var_37383_end_0, end_mask = var_37383_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37383_cast_fp16")]; + tensor var_37390_begin_0 = const()[name = string("op_37390_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37390_end_0 = const()[name = string("op_37390_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37390_end_mask_0 = const()[name = string("op_37390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37390_cast_fp16 = slice_by_index(begin = var_37390_begin_0, end = var_37390_end_0, end_mask = var_37390_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37390_cast_fp16")]; + tensor var_37397_begin_0 = const()[name = string("op_37397_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37397_end_0 = const()[name = string("op_37397_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37397_end_mask_0 = const()[name = string("op_37397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37397_cast_fp16 = slice_by_index(begin = var_37397_begin_0, end = var_37397_end_0, end_mask = var_37397_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37397_cast_fp16")]; + tensor var_37404_begin_0 = const()[name = string("op_37404_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37404_end_0 = const()[name = string("op_37404_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37404_end_mask_0 = const()[name = string("op_37404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37404_cast_fp16 = slice_by_index(begin = var_37404_begin_0, end = var_37404_end_0, end_mask = var_37404_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37404_cast_fp16")]; + tensor var_37411_begin_0 = const()[name = string("op_37411_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37411_end_0 = const()[name = string("op_37411_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37411_end_mask_0 = const()[name = string("op_37411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37411_cast_fp16 = slice_by_index(begin = var_37411_begin_0, end = var_37411_end_0, end_mask = var_37411_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37411_cast_fp16")]; + tensor var_37418_begin_0 = const()[name = string("op_37418_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37418_end_0 = const()[name = string("op_37418_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37418_end_mask_0 = const()[name = string("op_37418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37418_cast_fp16 = slice_by_index(begin = var_37418_begin_0, end = var_37418_end_0, end_mask = var_37418_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37418_cast_fp16")]; + tensor var_37425_begin_0 = const()[name = string("op_37425_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37425_end_0 = const()[name = string("op_37425_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37425_end_mask_0 = const()[name = string("op_37425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37425_cast_fp16 = slice_by_index(begin = var_37425_begin_0, end = var_37425_end_0, end_mask = var_37425_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37425_cast_fp16")]; + tensor var_37432_begin_0 = const()[name = string("op_37432_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37432_end_0 = const()[name = string("op_37432_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37432_end_mask_0 = const()[name = string("op_37432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37432_cast_fp16 = slice_by_index(begin = var_37432_begin_0, end = var_37432_end_0, end_mask = var_37432_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37432_cast_fp16")]; + tensor var_37439_begin_0 = const()[name = string("op_37439_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37439_end_0 = const()[name = string("op_37439_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37439_end_mask_0 = const()[name = string("op_37439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37439_cast_fp16 = slice_by_index(begin = var_37439_begin_0, end = var_37439_end_0, end_mask = var_37439_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37439_cast_fp16")]; + tensor var_37446_begin_0 = const()[name = string("op_37446_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37446_end_0 = const()[name = string("op_37446_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37446_end_mask_0 = const()[name = string("op_37446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37446_cast_fp16 = slice_by_index(begin = var_37446_begin_0, end = var_37446_end_0, end_mask = var_37446_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37446_cast_fp16")]; + tensor var_37453_begin_0 = const()[name = string("op_37453_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37453_end_0 = const()[name = string("op_37453_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37453_end_mask_0 = const()[name = string("op_37453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37453_cast_fp16 = slice_by_index(begin = var_37453_begin_0, end = var_37453_end_0, end_mask = var_37453_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37453_cast_fp16")]; + tensor var_37460_begin_0 = const()[name = string("op_37460_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37460_end_0 = const()[name = string("op_37460_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37460_end_mask_0 = const()[name = string("op_37460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37460_cast_fp16 = slice_by_index(begin = var_37460_begin_0, end = var_37460_end_0, end_mask = var_37460_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37460_cast_fp16")]; + tensor var_37467_begin_0 = const()[name = string("op_37467_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37467_end_0 = const()[name = string("op_37467_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37467_end_mask_0 = const()[name = string("op_37467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37467_cast_fp16 = slice_by_index(begin = var_37467_begin_0, end = var_37467_end_0, end_mask = var_37467_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37467_cast_fp16")]; + tensor var_37474_begin_0 = const()[name = string("op_37474_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37474_end_0 = const()[name = string("op_37474_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37474_end_mask_0 = const()[name = string("op_37474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37474_cast_fp16 = slice_by_index(begin = var_37474_begin_0, end = var_37474_end_0, end_mask = var_37474_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37474_cast_fp16")]; + tensor var_37481_begin_0 = const()[name = string("op_37481_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37481_end_0 = const()[name = string("op_37481_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37481_end_mask_0 = const()[name = string("op_37481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37481_cast_fp16 = slice_by_index(begin = var_37481_begin_0, end = var_37481_end_0, end_mask = var_37481_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37481_cast_fp16")]; + tensor var_37488_begin_0 = const()[name = string("op_37488_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37488_end_0 = const()[name = string("op_37488_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37488_end_mask_0 = const()[name = string("op_37488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37488_cast_fp16 = slice_by_index(begin = var_37488_begin_0, end = var_37488_end_0, end_mask = var_37488_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37488_cast_fp16")]; + tensor var_37495_begin_0 = const()[name = string("op_37495_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37495_end_0 = const()[name = string("op_37495_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37495_end_mask_0 = const()[name = string("op_37495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37495_cast_fp16 = slice_by_index(begin = var_37495_begin_0, end = var_37495_end_0, end_mask = var_37495_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37495_cast_fp16")]; + tensor var_37502_begin_0 = const()[name = string("op_37502_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37502_end_0 = const()[name = string("op_37502_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37502_end_mask_0 = const()[name = string("op_37502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37502_cast_fp16 = slice_by_index(begin = var_37502_begin_0, end = var_37502_end_0, end_mask = var_37502_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37502_cast_fp16")]; + tensor var_37509_begin_0 = const()[name = string("op_37509_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37509_end_0 = const()[name = string("op_37509_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37509_end_mask_0 = const()[name = string("op_37509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37509_cast_fp16 = slice_by_index(begin = var_37509_begin_0, end = var_37509_end_0, end_mask = var_37509_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37509_cast_fp16")]; + tensor var_37516_begin_0 = const()[name = string("op_37516_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37516_end_0 = const()[name = string("op_37516_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37516_end_mask_0 = const()[name = string("op_37516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37516_cast_fp16 = slice_by_index(begin = var_37516_begin_0, end = var_37516_end_0, end_mask = var_37516_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37516_cast_fp16")]; + tensor var_37523_begin_0 = const()[name = string("op_37523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37523_end_0 = const()[name = string("op_37523_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37523_end_mask_0 = const()[name = string("op_37523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37523_cast_fp16 = slice_by_index(begin = var_37523_begin_0, end = var_37523_end_0, end_mask = var_37523_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37523_cast_fp16")]; + tensor var_37530_begin_0 = const()[name = string("op_37530_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37530_end_0 = const()[name = string("op_37530_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37530_end_mask_0 = const()[name = string("op_37530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37530_cast_fp16 = slice_by_index(begin = var_37530_begin_0, end = var_37530_end_0, end_mask = var_37530_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37530_cast_fp16")]; + tensor var_37537_begin_0 = const()[name = string("op_37537_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37537_end_0 = const()[name = string("op_37537_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37537_end_mask_0 = const()[name = string("op_37537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37537_cast_fp16 = slice_by_index(begin = var_37537_begin_0, end = var_37537_end_0, end_mask = var_37537_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37537_cast_fp16")]; + tensor var_37544_begin_0 = const()[name = string("op_37544_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37544_end_0 = const()[name = string("op_37544_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37544_end_mask_0 = const()[name = string("op_37544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37544_cast_fp16 = slice_by_index(begin = var_37544_begin_0, end = var_37544_end_0, end_mask = var_37544_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37544_cast_fp16")]; + tensor k_49_perm_0 = const()[name = string("k_49_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_37549_begin_0 = const()[name = string("op_37549_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37549_end_0 = const()[name = string("op_37549_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_37549_end_mask_0 = const()[name = string("op_37549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_49_cast_fp16 = transpose(perm = k_49_perm_0, x = key_49_cast_fp16)[name = string("transpose_7")]; + tensor var_37549_cast_fp16 = slice_by_index(begin = var_37549_begin_0, end = var_37549_end_0, end_mask = var_37549_end_mask_0, x = k_49_cast_fp16)[name = string("op_37549_cast_fp16")]; + tensor var_37553_begin_0 = const()[name = string("op_37553_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_37553_end_0 = const()[name = string("op_37553_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_37553_end_mask_0 = const()[name = string("op_37553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37553_cast_fp16 = slice_by_index(begin = var_37553_begin_0, end = var_37553_end_0, end_mask = var_37553_end_mask_0, x = k_49_cast_fp16)[name = string("op_37553_cast_fp16")]; + tensor var_37557_begin_0 = const()[name = string("op_37557_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_37557_end_0 = const()[name = string("op_37557_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_37557_end_mask_0 = const()[name = string("op_37557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37557_cast_fp16 = slice_by_index(begin = var_37557_begin_0, end = var_37557_end_0, end_mask = var_37557_end_mask_0, x = k_49_cast_fp16)[name = string("op_37557_cast_fp16")]; + tensor var_37561_begin_0 = const()[name = string("op_37561_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_37561_end_0 = const()[name = string("op_37561_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_37561_end_mask_0 = const()[name = string("op_37561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37561_cast_fp16 = slice_by_index(begin = var_37561_begin_0, end = var_37561_end_0, end_mask = var_37561_end_mask_0, x = k_49_cast_fp16)[name = string("op_37561_cast_fp16")]; + tensor var_37565_begin_0 = const()[name = string("op_37565_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_37565_end_0 = const()[name = string("op_37565_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_37565_end_mask_0 = const()[name = string("op_37565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37565_cast_fp16 = slice_by_index(begin = var_37565_begin_0, end = var_37565_end_0, end_mask = var_37565_end_mask_0, x = k_49_cast_fp16)[name = string("op_37565_cast_fp16")]; + tensor var_37569_begin_0 = const()[name = string("op_37569_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_37569_end_0 = const()[name = string("op_37569_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_37569_end_mask_0 = const()[name = string("op_37569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37569_cast_fp16 = slice_by_index(begin = var_37569_begin_0, end = var_37569_end_0, end_mask = var_37569_end_mask_0, x = k_49_cast_fp16)[name = string("op_37569_cast_fp16")]; + tensor var_37573_begin_0 = const()[name = string("op_37573_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_37573_end_0 = const()[name = string("op_37573_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_37573_end_mask_0 = const()[name = string("op_37573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37573_cast_fp16 = slice_by_index(begin = var_37573_begin_0, end = var_37573_end_0, end_mask = var_37573_end_mask_0, x = k_49_cast_fp16)[name = string("op_37573_cast_fp16")]; + tensor var_37577_begin_0 = const()[name = string("op_37577_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_37577_end_0 = const()[name = string("op_37577_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_37577_end_mask_0 = const()[name = string("op_37577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37577_cast_fp16 = slice_by_index(begin = var_37577_begin_0, end = var_37577_end_0, end_mask = var_37577_end_mask_0, x = k_49_cast_fp16)[name = string("op_37577_cast_fp16")]; + tensor var_37581_begin_0 = const()[name = string("op_37581_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_37581_end_0 = const()[name = string("op_37581_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_37581_end_mask_0 = const()[name = string("op_37581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37581_cast_fp16 = slice_by_index(begin = var_37581_begin_0, end = var_37581_end_0, end_mask = var_37581_end_mask_0, x = k_49_cast_fp16)[name = string("op_37581_cast_fp16")]; + tensor var_37585_begin_0 = const()[name = string("op_37585_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_37585_end_0 = const()[name = string("op_37585_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_37585_end_mask_0 = const()[name = string("op_37585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37585_cast_fp16 = slice_by_index(begin = var_37585_begin_0, end = var_37585_end_0, end_mask = var_37585_end_mask_0, x = k_49_cast_fp16)[name = string("op_37585_cast_fp16")]; + tensor var_37589_begin_0 = const()[name = string("op_37589_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_37589_end_0 = const()[name = string("op_37589_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_37589_end_mask_0 = const()[name = string("op_37589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37589_cast_fp16 = slice_by_index(begin = var_37589_begin_0, end = var_37589_end_0, end_mask = var_37589_end_mask_0, x = k_49_cast_fp16)[name = string("op_37589_cast_fp16")]; + tensor var_37593_begin_0 = const()[name = string("op_37593_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_37593_end_0 = const()[name = string("op_37593_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_37593_end_mask_0 = const()[name = string("op_37593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37593_cast_fp16 = slice_by_index(begin = var_37593_begin_0, end = var_37593_end_0, end_mask = var_37593_end_mask_0, x = k_49_cast_fp16)[name = string("op_37593_cast_fp16")]; + tensor var_37597_begin_0 = const()[name = string("op_37597_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_37597_end_0 = const()[name = string("op_37597_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_37597_end_mask_0 = const()[name = string("op_37597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37597_cast_fp16 = slice_by_index(begin = var_37597_begin_0, end = var_37597_end_0, end_mask = var_37597_end_mask_0, x = k_49_cast_fp16)[name = string("op_37597_cast_fp16")]; + tensor var_37601_begin_0 = const()[name = string("op_37601_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_37601_end_0 = const()[name = string("op_37601_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_37601_end_mask_0 = const()[name = string("op_37601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37601_cast_fp16 = slice_by_index(begin = var_37601_begin_0, end = var_37601_end_0, end_mask = var_37601_end_mask_0, x = k_49_cast_fp16)[name = string("op_37601_cast_fp16")]; + tensor var_37605_begin_0 = const()[name = string("op_37605_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_37605_end_0 = const()[name = string("op_37605_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_37605_end_mask_0 = const()[name = string("op_37605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37605_cast_fp16 = slice_by_index(begin = var_37605_begin_0, end = var_37605_end_0, end_mask = var_37605_end_mask_0, x = k_49_cast_fp16)[name = string("op_37605_cast_fp16")]; + tensor var_37609_begin_0 = const()[name = string("op_37609_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_37609_end_0 = const()[name = string("op_37609_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_37609_end_mask_0 = const()[name = string("op_37609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37609_cast_fp16 = slice_by_index(begin = var_37609_begin_0, end = var_37609_end_0, end_mask = var_37609_end_mask_0, x = k_49_cast_fp16)[name = string("op_37609_cast_fp16")]; + tensor var_37613_begin_0 = const()[name = string("op_37613_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_37613_end_0 = const()[name = string("op_37613_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_37613_end_mask_0 = const()[name = string("op_37613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37613_cast_fp16 = slice_by_index(begin = var_37613_begin_0, end = var_37613_end_0, end_mask = var_37613_end_mask_0, x = k_49_cast_fp16)[name = string("op_37613_cast_fp16")]; + tensor var_37617_begin_0 = const()[name = string("op_37617_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_37617_end_0 = const()[name = string("op_37617_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_37617_end_mask_0 = const()[name = string("op_37617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37617_cast_fp16 = slice_by_index(begin = var_37617_begin_0, end = var_37617_end_0, end_mask = var_37617_end_mask_0, x = k_49_cast_fp16)[name = string("op_37617_cast_fp16")]; + tensor var_37621_begin_0 = const()[name = string("op_37621_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_37621_end_0 = const()[name = string("op_37621_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_37621_end_mask_0 = const()[name = string("op_37621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37621_cast_fp16 = slice_by_index(begin = var_37621_begin_0, end = var_37621_end_0, end_mask = var_37621_end_mask_0, x = k_49_cast_fp16)[name = string("op_37621_cast_fp16")]; + tensor var_37625_begin_0 = const()[name = string("op_37625_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_37625_end_0 = const()[name = string("op_37625_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_37625_end_mask_0 = const()[name = string("op_37625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37625_cast_fp16 = slice_by_index(begin = var_37625_begin_0, end = var_37625_end_0, end_mask = var_37625_end_mask_0, x = k_49_cast_fp16)[name = string("op_37625_cast_fp16")]; + tensor var_37627_begin_0 = const()[name = string("op_37627_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37627_end_0 = const()[name = string("op_37627_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37627_end_mask_0 = const()[name = string("op_37627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37627_cast_fp16 = slice_by_index(begin = var_37627_begin_0, end = var_37627_end_0, end_mask = var_37627_end_mask_0, x = value_49_cast_fp16)[name = string("op_37627_cast_fp16")]; + tensor var_37631_begin_0 = const()[name = string("op_37631_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37631_end_0 = const()[name = string("op_37631_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_37631_end_mask_0 = const()[name = string("op_37631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37631_cast_fp16 = slice_by_index(begin = var_37631_begin_0, end = var_37631_end_0, end_mask = var_37631_end_mask_0, x = value_49_cast_fp16)[name = string("op_37631_cast_fp16")]; + tensor var_37635_begin_0 = const()[name = string("op_37635_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37635_end_0 = const()[name = string("op_37635_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_37635_end_mask_0 = const()[name = string("op_37635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37635_cast_fp16 = slice_by_index(begin = var_37635_begin_0, end = var_37635_end_0, end_mask = var_37635_end_mask_0, x = value_49_cast_fp16)[name = string("op_37635_cast_fp16")]; + tensor var_37639_begin_0 = const()[name = string("op_37639_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37639_end_0 = const()[name = string("op_37639_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_37639_end_mask_0 = const()[name = string("op_37639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37639_cast_fp16 = slice_by_index(begin = var_37639_begin_0, end = var_37639_end_0, end_mask = var_37639_end_mask_0, x = value_49_cast_fp16)[name = string("op_37639_cast_fp16")]; + tensor var_37643_begin_0 = const()[name = string("op_37643_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37643_end_0 = const()[name = string("op_37643_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_37643_end_mask_0 = const()[name = string("op_37643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37643_cast_fp16 = slice_by_index(begin = var_37643_begin_0, end = var_37643_end_0, end_mask = var_37643_end_mask_0, x = value_49_cast_fp16)[name = string("op_37643_cast_fp16")]; + tensor var_37647_begin_0 = const()[name = string("op_37647_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37647_end_0 = const()[name = string("op_37647_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_37647_end_mask_0 = const()[name = string("op_37647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37647_cast_fp16 = slice_by_index(begin = var_37647_begin_0, end = var_37647_end_0, end_mask = var_37647_end_mask_0, x = value_49_cast_fp16)[name = string("op_37647_cast_fp16")]; + tensor var_37651_begin_0 = const()[name = string("op_37651_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37651_end_0 = const()[name = string("op_37651_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_37651_end_mask_0 = const()[name = string("op_37651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37651_cast_fp16 = slice_by_index(begin = var_37651_begin_0, end = var_37651_end_0, end_mask = var_37651_end_mask_0, x = value_49_cast_fp16)[name = string("op_37651_cast_fp16")]; + tensor var_37655_begin_0 = const()[name = string("op_37655_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37655_end_0 = const()[name = string("op_37655_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_37655_end_mask_0 = const()[name = string("op_37655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37655_cast_fp16 = slice_by_index(begin = var_37655_begin_0, end = var_37655_end_0, end_mask = var_37655_end_mask_0, x = value_49_cast_fp16)[name = string("op_37655_cast_fp16")]; + tensor var_37659_begin_0 = const()[name = string("op_37659_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37659_end_0 = const()[name = string("op_37659_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_37659_end_mask_0 = const()[name = string("op_37659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37659_cast_fp16 = slice_by_index(begin = var_37659_begin_0, end = var_37659_end_0, end_mask = var_37659_end_mask_0, x = value_49_cast_fp16)[name = string("op_37659_cast_fp16")]; + tensor var_37663_begin_0 = const()[name = string("op_37663_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37663_end_0 = const()[name = string("op_37663_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_37663_end_mask_0 = const()[name = string("op_37663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37663_cast_fp16 = slice_by_index(begin = var_37663_begin_0, end = var_37663_end_0, end_mask = var_37663_end_mask_0, x = value_49_cast_fp16)[name = string("op_37663_cast_fp16")]; + tensor var_37667_begin_0 = const()[name = string("op_37667_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37667_end_0 = const()[name = string("op_37667_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_37667_end_mask_0 = const()[name = string("op_37667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37667_cast_fp16 = slice_by_index(begin = var_37667_begin_0, end = var_37667_end_0, end_mask = var_37667_end_mask_0, x = value_49_cast_fp16)[name = string("op_37667_cast_fp16")]; + tensor var_37671_begin_0 = const()[name = string("op_37671_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37671_end_0 = const()[name = string("op_37671_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_37671_end_mask_0 = const()[name = string("op_37671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37671_cast_fp16 = slice_by_index(begin = var_37671_begin_0, end = var_37671_end_0, end_mask = var_37671_end_mask_0, x = value_49_cast_fp16)[name = string("op_37671_cast_fp16")]; + tensor var_37675_begin_0 = const()[name = string("op_37675_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37675_end_0 = const()[name = string("op_37675_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_37675_end_mask_0 = const()[name = string("op_37675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37675_cast_fp16 = slice_by_index(begin = var_37675_begin_0, end = var_37675_end_0, end_mask = var_37675_end_mask_0, x = value_49_cast_fp16)[name = string("op_37675_cast_fp16")]; + tensor var_37679_begin_0 = const()[name = string("op_37679_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37679_end_0 = const()[name = string("op_37679_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_37679_end_mask_0 = const()[name = string("op_37679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37679_cast_fp16 = slice_by_index(begin = var_37679_begin_0, end = var_37679_end_0, end_mask = var_37679_end_mask_0, x = value_49_cast_fp16)[name = string("op_37679_cast_fp16")]; + tensor var_37683_begin_0 = const()[name = string("op_37683_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37683_end_0 = const()[name = string("op_37683_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_37683_end_mask_0 = const()[name = string("op_37683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37683_cast_fp16 = slice_by_index(begin = var_37683_begin_0, end = var_37683_end_0, end_mask = var_37683_end_mask_0, x = value_49_cast_fp16)[name = string("op_37683_cast_fp16")]; + tensor var_37687_begin_0 = const()[name = string("op_37687_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37687_end_0 = const()[name = string("op_37687_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_37687_end_mask_0 = const()[name = string("op_37687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37687_cast_fp16 = slice_by_index(begin = var_37687_begin_0, end = var_37687_end_0, end_mask = var_37687_end_mask_0, x = value_49_cast_fp16)[name = string("op_37687_cast_fp16")]; + tensor var_37691_begin_0 = const()[name = string("op_37691_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37691_end_0 = const()[name = string("op_37691_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_37691_end_mask_0 = const()[name = string("op_37691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37691_cast_fp16 = slice_by_index(begin = var_37691_begin_0, end = var_37691_end_0, end_mask = var_37691_end_mask_0, x = value_49_cast_fp16)[name = string("op_37691_cast_fp16")]; + tensor var_37695_begin_0 = const()[name = string("op_37695_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37695_end_0 = const()[name = string("op_37695_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_37695_end_mask_0 = const()[name = string("op_37695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37695_cast_fp16 = slice_by_index(begin = var_37695_begin_0, end = var_37695_end_0, end_mask = var_37695_end_mask_0, x = value_49_cast_fp16)[name = string("op_37695_cast_fp16")]; + tensor var_37699_begin_0 = const()[name = string("op_37699_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37699_end_0 = const()[name = string("op_37699_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_37699_end_mask_0 = const()[name = string("op_37699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37699_cast_fp16 = slice_by_index(begin = var_37699_begin_0, end = var_37699_end_0, end_mask = var_37699_end_mask_0, x = value_49_cast_fp16)[name = string("op_37699_cast_fp16")]; + tensor var_37703_begin_0 = const()[name = string("op_37703_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37703_end_0 = const()[name = string("op_37703_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_37703_end_mask_0 = const()[name = string("op_37703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37703_cast_fp16 = slice_by_index(begin = var_37703_begin_0, end = var_37703_end_0, end_mask = var_37703_end_mask_0, x = value_49_cast_fp16)[name = string("op_37703_cast_fp16")]; + string _SplitHeadsQ__mh_w_3841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3841_equation_0, values = (var_37549_cast_fp16, var_36991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3841_cast_fp16")]; + string _SplitHeadsQ__mh_w_3843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3843_equation_0, values = (var_37549_cast_fp16, var_36998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3843_cast_fp16")]; + string _SplitHeadsQ__mh_w_3845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3845_equation_0, values = (var_37549_cast_fp16, var_37005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3845_cast_fp16")]; + string _SplitHeadsQ__mh_w_3847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3847_equation_0, values = (var_37549_cast_fp16, var_37012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3847_cast_fp16")]; + string _SplitHeadsQ__mh_w_3849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3849_equation_0, values = (var_37553_cast_fp16, var_37019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3849_cast_fp16")]; + string _SplitHeadsQ__mh_w_3851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3851_equation_0, values = (var_37553_cast_fp16, var_37026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3851_cast_fp16")]; + string _SplitHeadsQ__mh_w_3853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3853_equation_0, values = (var_37553_cast_fp16, var_37033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3853_cast_fp16")]; + string _SplitHeadsQ__mh_w_3855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3855_equation_0, values = (var_37553_cast_fp16, var_37040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3855_cast_fp16")]; + string _SplitHeadsQ__mh_w_3857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3857_equation_0, values = (var_37557_cast_fp16, var_37047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3857_cast_fp16")]; + string _SplitHeadsQ__mh_w_3859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3859_equation_0, values = (var_37557_cast_fp16, var_37054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3859_cast_fp16")]; + string _SplitHeadsQ__mh_w_3861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3861_equation_0, values = (var_37557_cast_fp16, var_37061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3861_cast_fp16")]; + string _SplitHeadsQ__mh_w_3863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3863_equation_0, values = (var_37557_cast_fp16, var_37068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3863_cast_fp16")]; + string _SplitHeadsQ__mh_w_3865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3865_equation_0, values = (var_37561_cast_fp16, var_37075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3865_cast_fp16")]; + string _SplitHeadsQ__mh_w_3867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3867_equation_0, values = (var_37561_cast_fp16, var_37082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3867_cast_fp16")]; + string _SplitHeadsQ__mh_w_3869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3869_equation_0, values = (var_37561_cast_fp16, var_37089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3869_cast_fp16")]; + string _SplitHeadsQ__mh_w_3871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3871_equation_0, values = (var_37561_cast_fp16, var_37096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3871_cast_fp16")]; + string _SplitHeadsQ__mh_w_3873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3873_equation_0, values = (var_37565_cast_fp16, var_37103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3873_cast_fp16")]; + string _SplitHeadsQ__mh_w_3875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3875_equation_0, values = (var_37565_cast_fp16, var_37110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3875_cast_fp16")]; + string _SplitHeadsQ__mh_w_3877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3877_equation_0, values = (var_37565_cast_fp16, var_37117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3877_cast_fp16")]; + string _SplitHeadsQ__mh_w_3879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3879_equation_0, values = (var_37565_cast_fp16, var_37124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3879_cast_fp16")]; + string _SplitHeadsQ__mh_w_3881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3881_equation_0, values = (var_37569_cast_fp16, var_37131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3881_cast_fp16")]; + string _SplitHeadsQ__mh_w_3883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3883_equation_0, values = (var_37569_cast_fp16, var_37138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3883_cast_fp16")]; + string _SplitHeadsQ__mh_w_3885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3885_equation_0, values = (var_37569_cast_fp16, var_37145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3885_cast_fp16")]; + string _SplitHeadsQ__mh_w_3887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3887_equation_0, values = (var_37569_cast_fp16, var_37152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3887_cast_fp16")]; + string _SplitHeadsQ__mh_w_3889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3889_equation_0, values = (var_37573_cast_fp16, var_37159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3889_cast_fp16")]; + string _SplitHeadsQ__mh_w_3891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3891_equation_0, values = (var_37573_cast_fp16, var_37166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3891_cast_fp16")]; + string _SplitHeadsQ__mh_w_3893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3893_equation_0, values = (var_37573_cast_fp16, var_37173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3893_cast_fp16")]; + string _SplitHeadsQ__mh_w_3895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3895_equation_0, values = (var_37573_cast_fp16, var_37180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3895_cast_fp16")]; + string _SplitHeadsQ__mh_w_3897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3897_equation_0, values = (var_37577_cast_fp16, var_37187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3897_cast_fp16")]; + string _SplitHeadsQ__mh_w_3899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3899_equation_0, values = (var_37577_cast_fp16, var_37194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3899_cast_fp16")]; + string _SplitHeadsQ__mh_w_3901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3901_equation_0, values = (var_37577_cast_fp16, var_37201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3901_cast_fp16")]; + string _SplitHeadsQ__mh_w_3903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3903_equation_0, values = (var_37577_cast_fp16, var_37208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3903_cast_fp16")]; + string _SplitHeadsQ__mh_w_3905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3905_equation_0, values = (var_37581_cast_fp16, var_37215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3905_cast_fp16")]; + string _SplitHeadsQ__mh_w_3907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3907_equation_0, values = (var_37581_cast_fp16, var_37222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3907_cast_fp16")]; + string _SplitHeadsQ__mh_w_3909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3909_equation_0, values = (var_37581_cast_fp16, var_37229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3909_cast_fp16")]; + string _SplitHeadsQ__mh_w_3911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3911_equation_0, values = (var_37581_cast_fp16, var_37236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3911_cast_fp16")]; + string _SplitHeadsQ__mh_w_3913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3913_equation_0, values = (var_37585_cast_fp16, var_37243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3913_cast_fp16")]; + string _SplitHeadsQ__mh_w_3915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3915_equation_0, values = (var_37585_cast_fp16, var_37250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3915_cast_fp16")]; + string _SplitHeadsQ__mh_w_3917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3917_equation_0, values = (var_37585_cast_fp16, var_37257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3917_cast_fp16")]; + string _SplitHeadsQ__mh_w_3919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3919_equation_0, values = (var_37585_cast_fp16, var_37264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3919_cast_fp16")]; + string _SplitHeadsQ__mh_w_3921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3921_equation_0, values = (var_37589_cast_fp16, var_37271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3921_cast_fp16")]; + string _SplitHeadsQ__mh_w_3923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3923_equation_0, values = (var_37589_cast_fp16, var_37278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3923_cast_fp16")]; + string _SplitHeadsQ__mh_w_3925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3925_equation_0, values = (var_37589_cast_fp16, var_37285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3925_cast_fp16")]; + string _SplitHeadsQ__mh_w_3927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3927_equation_0, values = (var_37589_cast_fp16, var_37292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3927_cast_fp16")]; + string _SplitHeadsQ__mh_w_3929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3929_equation_0, values = (var_37593_cast_fp16, var_37299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3929_cast_fp16")]; + string _SplitHeadsQ__mh_w_3931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3931_equation_0, values = (var_37593_cast_fp16, var_37306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3931_cast_fp16")]; + string _SplitHeadsQ__mh_w_3933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3933_equation_0, values = (var_37593_cast_fp16, var_37313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3933_cast_fp16")]; + string _SplitHeadsQ__mh_w_3935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3935_equation_0, values = (var_37593_cast_fp16, var_37320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3935_cast_fp16")]; + string _SplitHeadsQ__mh_w_3937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3937_equation_0, values = (var_37597_cast_fp16, var_37327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3937_cast_fp16")]; + string _SplitHeadsQ__mh_w_3939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3939_equation_0, values = (var_37597_cast_fp16, var_37334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3939_cast_fp16")]; + string _SplitHeadsQ__mh_w_3941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3941_equation_0, values = (var_37597_cast_fp16, var_37341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3941_cast_fp16")]; + string _SplitHeadsQ__mh_w_3943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3943_equation_0, values = (var_37597_cast_fp16, var_37348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3943_cast_fp16")]; + string _SplitHeadsQ__mh_w_3945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3945_equation_0, values = (var_37601_cast_fp16, var_37355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3945_cast_fp16")]; + string _SplitHeadsQ__mh_w_3947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3947_equation_0, values = (var_37601_cast_fp16, var_37362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3947_cast_fp16")]; + string _SplitHeadsQ__mh_w_3949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3949_equation_0, values = (var_37601_cast_fp16, var_37369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3949_cast_fp16")]; + string _SplitHeadsQ__mh_w_3951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3951_equation_0, values = (var_37601_cast_fp16, var_37376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3951_cast_fp16")]; + string _SplitHeadsQ__mh_w_3953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3953_equation_0, values = (var_37605_cast_fp16, var_37383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3953_cast_fp16")]; + string _SplitHeadsQ__mh_w_3955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3955_equation_0, values = (var_37605_cast_fp16, var_37390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3955_cast_fp16")]; + string _SplitHeadsQ__mh_w_3957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3957_equation_0, values = (var_37605_cast_fp16, var_37397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3957_cast_fp16")]; + string _SplitHeadsQ__mh_w_3959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3959_equation_0, values = (var_37605_cast_fp16, var_37404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3959_cast_fp16")]; + string _SplitHeadsQ__mh_w_3961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3961_equation_0, values = (var_37609_cast_fp16, var_37411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3961_cast_fp16")]; + string _SplitHeadsQ__mh_w_3963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3963_equation_0, values = (var_37609_cast_fp16, var_37418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3963_cast_fp16")]; + string _SplitHeadsQ__mh_w_3965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3965_equation_0, values = (var_37609_cast_fp16, var_37425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3965_cast_fp16")]; + string _SplitHeadsQ__mh_w_3967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3967_equation_0, values = (var_37609_cast_fp16, var_37432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3967_cast_fp16")]; + string _SplitHeadsQ__mh_w_3969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3969_equation_0, values = (var_37613_cast_fp16, var_37439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3969_cast_fp16")]; + string _SplitHeadsQ__mh_w_3971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3971_equation_0, values = (var_37613_cast_fp16, var_37446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3971_cast_fp16")]; + string _SplitHeadsQ__mh_w_3973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3973_equation_0, values = (var_37613_cast_fp16, var_37453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3973_cast_fp16")]; + string _SplitHeadsQ__mh_w_3975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3975_equation_0, values = (var_37613_cast_fp16, var_37460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3975_cast_fp16")]; + string _SplitHeadsQ__mh_w_3977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3977_equation_0, values = (var_37617_cast_fp16, var_37467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3977_cast_fp16")]; + string _SplitHeadsQ__mh_w_3979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3979_equation_0, values = (var_37617_cast_fp16, var_37474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3979_cast_fp16")]; + string _SplitHeadsQ__mh_w_3981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3981_equation_0, values = (var_37617_cast_fp16, var_37481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3981_cast_fp16")]; + string _SplitHeadsQ__mh_w_3983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3983_equation_0, values = (var_37617_cast_fp16, var_37488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3983_cast_fp16")]; + string _SplitHeadsQ__mh_w_3985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3985_equation_0, values = (var_37621_cast_fp16, var_37495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3985_cast_fp16")]; + string _SplitHeadsQ__mh_w_3987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3987_equation_0, values = (var_37621_cast_fp16, var_37502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3987_cast_fp16")]; + string _SplitHeadsQ__mh_w_3989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3989_equation_0, values = (var_37621_cast_fp16, var_37509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3989_cast_fp16")]; + string _SplitHeadsQ__mh_w_3991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3991_equation_0, values = (var_37621_cast_fp16, var_37516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3991_cast_fp16")]; + string _SplitHeadsQ__mh_w_3993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3993_equation_0, values = (var_37625_cast_fp16, var_37523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3993_cast_fp16")]; + string _SplitHeadsQ__mh_w_3995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3995_equation_0, values = (var_37625_cast_fp16, var_37530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3995_cast_fp16")]; + string _SplitHeadsQ__mh_w_3997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3997_equation_0, values = (var_37625_cast_fp16, var_37537_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3997_cast_fp16")]; + string _SplitHeadsQ__mh_w_3999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3999_equation_0, values = (var_37625_cast_fp16, var_37544_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3999_cast_fp16")]; + fp16 var_37866_to_fp16 = const()[name = string("op_37866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3841_cast_fp16, y = var_37866_to_fp16)[name = string("aw_chunk_3841_cast_fp16")]; + fp16 var_37868_to_fp16 = const()[name = string("op_37868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3843_cast_fp16, y = var_37868_to_fp16)[name = string("aw_chunk_3843_cast_fp16")]; + fp16 var_37870_to_fp16 = const()[name = string("op_37870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3845_cast_fp16, y = var_37870_to_fp16)[name = string("aw_chunk_3845_cast_fp16")]; + fp16 var_37872_to_fp16 = const()[name = string("op_37872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3847_cast_fp16, y = var_37872_to_fp16)[name = string("aw_chunk_3847_cast_fp16")]; + fp16 var_37874_to_fp16 = const()[name = string("op_37874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3849_cast_fp16, y = var_37874_to_fp16)[name = string("aw_chunk_3849_cast_fp16")]; + fp16 var_37876_to_fp16 = const()[name = string("op_37876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3851_cast_fp16, y = var_37876_to_fp16)[name = string("aw_chunk_3851_cast_fp16")]; + fp16 var_37878_to_fp16 = const()[name = string("op_37878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3853_cast_fp16, y = var_37878_to_fp16)[name = string("aw_chunk_3853_cast_fp16")]; + fp16 var_37880_to_fp16 = const()[name = string("op_37880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3855_cast_fp16, y = var_37880_to_fp16)[name = string("aw_chunk_3855_cast_fp16")]; + fp16 var_37882_to_fp16 = const()[name = string("op_37882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3857_cast_fp16, y = var_37882_to_fp16)[name = string("aw_chunk_3857_cast_fp16")]; + fp16 var_37884_to_fp16 = const()[name = string("op_37884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3859_cast_fp16, y = var_37884_to_fp16)[name = string("aw_chunk_3859_cast_fp16")]; + fp16 var_37886_to_fp16 = const()[name = string("op_37886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3861_cast_fp16, y = var_37886_to_fp16)[name = string("aw_chunk_3861_cast_fp16")]; + fp16 var_37888_to_fp16 = const()[name = string("op_37888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3863_cast_fp16, y = var_37888_to_fp16)[name = string("aw_chunk_3863_cast_fp16")]; + fp16 var_37890_to_fp16 = const()[name = string("op_37890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3865_cast_fp16, y = var_37890_to_fp16)[name = string("aw_chunk_3865_cast_fp16")]; + fp16 var_37892_to_fp16 = const()[name = string("op_37892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3867_cast_fp16, y = var_37892_to_fp16)[name = string("aw_chunk_3867_cast_fp16")]; + fp16 var_37894_to_fp16 = const()[name = string("op_37894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3869_cast_fp16, y = var_37894_to_fp16)[name = string("aw_chunk_3869_cast_fp16")]; + fp16 var_37896_to_fp16 = const()[name = string("op_37896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3871_cast_fp16, y = var_37896_to_fp16)[name = string("aw_chunk_3871_cast_fp16")]; + fp16 var_37898_to_fp16 = const()[name = string("op_37898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3873_cast_fp16, y = var_37898_to_fp16)[name = string("aw_chunk_3873_cast_fp16")]; + fp16 var_37900_to_fp16 = const()[name = string("op_37900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3875_cast_fp16, y = var_37900_to_fp16)[name = string("aw_chunk_3875_cast_fp16")]; + fp16 var_37902_to_fp16 = const()[name = string("op_37902_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3877_cast_fp16, y = var_37902_to_fp16)[name = string("aw_chunk_3877_cast_fp16")]; + fp16 var_37904_to_fp16 = const()[name = string("op_37904_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3879_cast_fp16, y = var_37904_to_fp16)[name = string("aw_chunk_3879_cast_fp16")]; + fp16 var_37906_to_fp16 = const()[name = string("op_37906_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3881_cast_fp16, y = var_37906_to_fp16)[name = string("aw_chunk_3881_cast_fp16")]; + fp16 var_37908_to_fp16 = const()[name = string("op_37908_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3883_cast_fp16, y = var_37908_to_fp16)[name = string("aw_chunk_3883_cast_fp16")]; + fp16 var_37910_to_fp16 = const()[name = string("op_37910_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3885_cast_fp16, y = var_37910_to_fp16)[name = string("aw_chunk_3885_cast_fp16")]; + fp16 var_37912_to_fp16 = const()[name = string("op_37912_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3887_cast_fp16, y = var_37912_to_fp16)[name = string("aw_chunk_3887_cast_fp16")]; + fp16 var_37914_to_fp16 = const()[name = string("op_37914_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3889_cast_fp16, y = var_37914_to_fp16)[name = string("aw_chunk_3889_cast_fp16")]; + fp16 var_37916_to_fp16 = const()[name = string("op_37916_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3891_cast_fp16, y = var_37916_to_fp16)[name = string("aw_chunk_3891_cast_fp16")]; + fp16 var_37918_to_fp16 = const()[name = string("op_37918_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3893_cast_fp16, y = var_37918_to_fp16)[name = string("aw_chunk_3893_cast_fp16")]; + fp16 var_37920_to_fp16 = const()[name = string("op_37920_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3895_cast_fp16, y = var_37920_to_fp16)[name = string("aw_chunk_3895_cast_fp16")]; + fp16 var_37922_to_fp16 = const()[name = string("op_37922_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3897_cast_fp16, y = var_37922_to_fp16)[name = string("aw_chunk_3897_cast_fp16")]; + fp16 var_37924_to_fp16 = const()[name = string("op_37924_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3899_cast_fp16, y = var_37924_to_fp16)[name = string("aw_chunk_3899_cast_fp16")]; + fp16 var_37926_to_fp16 = const()[name = string("op_37926_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3901_cast_fp16, y = var_37926_to_fp16)[name = string("aw_chunk_3901_cast_fp16")]; + fp16 var_37928_to_fp16 = const()[name = string("op_37928_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3903_cast_fp16, y = var_37928_to_fp16)[name = string("aw_chunk_3903_cast_fp16")]; + fp16 var_37930_to_fp16 = const()[name = string("op_37930_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3905_cast_fp16, y = var_37930_to_fp16)[name = string("aw_chunk_3905_cast_fp16")]; + fp16 var_37932_to_fp16 = const()[name = string("op_37932_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3907_cast_fp16, y = var_37932_to_fp16)[name = string("aw_chunk_3907_cast_fp16")]; + fp16 var_37934_to_fp16 = const()[name = string("op_37934_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3909_cast_fp16, y = var_37934_to_fp16)[name = string("aw_chunk_3909_cast_fp16")]; + fp16 var_37936_to_fp16 = const()[name = string("op_37936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3911_cast_fp16, y = var_37936_to_fp16)[name = string("aw_chunk_3911_cast_fp16")]; + fp16 var_37938_to_fp16 = const()[name = string("op_37938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3913_cast_fp16, y = var_37938_to_fp16)[name = string("aw_chunk_3913_cast_fp16")]; + fp16 var_37940_to_fp16 = const()[name = string("op_37940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3915_cast_fp16, y = var_37940_to_fp16)[name = string("aw_chunk_3915_cast_fp16")]; + fp16 var_37942_to_fp16 = const()[name = string("op_37942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3917_cast_fp16, y = var_37942_to_fp16)[name = string("aw_chunk_3917_cast_fp16")]; + fp16 var_37944_to_fp16 = const()[name = string("op_37944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3919_cast_fp16, y = var_37944_to_fp16)[name = string("aw_chunk_3919_cast_fp16")]; + fp16 var_37946_to_fp16 = const()[name = string("op_37946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3921_cast_fp16, y = var_37946_to_fp16)[name = string("aw_chunk_3921_cast_fp16")]; + fp16 var_37948_to_fp16 = const()[name = string("op_37948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3923_cast_fp16, y = var_37948_to_fp16)[name = string("aw_chunk_3923_cast_fp16")]; + fp16 var_37950_to_fp16 = const()[name = string("op_37950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3925_cast_fp16, y = var_37950_to_fp16)[name = string("aw_chunk_3925_cast_fp16")]; + fp16 var_37952_to_fp16 = const()[name = string("op_37952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3927_cast_fp16, y = var_37952_to_fp16)[name = string("aw_chunk_3927_cast_fp16")]; + fp16 var_37954_to_fp16 = const()[name = string("op_37954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3929_cast_fp16, y = var_37954_to_fp16)[name = string("aw_chunk_3929_cast_fp16")]; + fp16 var_37956_to_fp16 = const()[name = string("op_37956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3931_cast_fp16, y = var_37956_to_fp16)[name = string("aw_chunk_3931_cast_fp16")]; + fp16 var_37958_to_fp16 = const()[name = string("op_37958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3933_cast_fp16, y = var_37958_to_fp16)[name = string("aw_chunk_3933_cast_fp16")]; + fp16 var_37960_to_fp16 = const()[name = string("op_37960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3935_cast_fp16, y = var_37960_to_fp16)[name = string("aw_chunk_3935_cast_fp16")]; + fp16 var_37962_to_fp16 = const()[name = string("op_37962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3937_cast_fp16, y = var_37962_to_fp16)[name = string("aw_chunk_3937_cast_fp16")]; + fp16 var_37964_to_fp16 = const()[name = string("op_37964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3939_cast_fp16, y = var_37964_to_fp16)[name = string("aw_chunk_3939_cast_fp16")]; + fp16 var_37966_to_fp16 = const()[name = string("op_37966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3941_cast_fp16, y = var_37966_to_fp16)[name = string("aw_chunk_3941_cast_fp16")]; + fp16 var_37968_to_fp16 = const()[name = string("op_37968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3943_cast_fp16, y = var_37968_to_fp16)[name = string("aw_chunk_3943_cast_fp16")]; + fp16 var_37970_to_fp16 = const()[name = string("op_37970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3945_cast_fp16, y = var_37970_to_fp16)[name = string("aw_chunk_3945_cast_fp16")]; + fp16 var_37972_to_fp16 = const()[name = string("op_37972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3947_cast_fp16, y = var_37972_to_fp16)[name = string("aw_chunk_3947_cast_fp16")]; + fp16 var_37974_to_fp16 = const()[name = string("op_37974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3949_cast_fp16, y = var_37974_to_fp16)[name = string("aw_chunk_3949_cast_fp16")]; + fp16 var_37976_to_fp16 = const()[name = string("op_37976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3951_cast_fp16, y = var_37976_to_fp16)[name = string("aw_chunk_3951_cast_fp16")]; + fp16 var_37978_to_fp16 = const()[name = string("op_37978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3953_cast_fp16, y = var_37978_to_fp16)[name = string("aw_chunk_3953_cast_fp16")]; + fp16 var_37980_to_fp16 = const()[name = string("op_37980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3955_cast_fp16, y = var_37980_to_fp16)[name = string("aw_chunk_3955_cast_fp16")]; + fp16 var_37982_to_fp16 = const()[name = string("op_37982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3957_cast_fp16, y = var_37982_to_fp16)[name = string("aw_chunk_3957_cast_fp16")]; + fp16 var_37984_to_fp16 = const()[name = string("op_37984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3959_cast_fp16, y = var_37984_to_fp16)[name = string("aw_chunk_3959_cast_fp16")]; + fp16 var_37986_to_fp16 = const()[name = string("op_37986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3961_cast_fp16, y = var_37986_to_fp16)[name = string("aw_chunk_3961_cast_fp16")]; + fp16 var_37988_to_fp16 = const()[name = string("op_37988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3963_cast_fp16, y = var_37988_to_fp16)[name = string("aw_chunk_3963_cast_fp16")]; + fp16 var_37990_to_fp16 = const()[name = string("op_37990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3965_cast_fp16, y = var_37990_to_fp16)[name = string("aw_chunk_3965_cast_fp16")]; + fp16 var_37992_to_fp16 = const()[name = string("op_37992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3967_cast_fp16, y = var_37992_to_fp16)[name = string("aw_chunk_3967_cast_fp16")]; + fp16 var_37994_to_fp16 = const()[name = string("op_37994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3969_cast_fp16, y = var_37994_to_fp16)[name = string("aw_chunk_3969_cast_fp16")]; + fp16 var_37996_to_fp16 = const()[name = string("op_37996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3971_cast_fp16, y = var_37996_to_fp16)[name = string("aw_chunk_3971_cast_fp16")]; + fp16 var_37998_to_fp16 = const()[name = string("op_37998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3973_cast_fp16, y = var_37998_to_fp16)[name = string("aw_chunk_3973_cast_fp16")]; + fp16 var_38000_to_fp16 = const()[name = string("op_38000_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3975_cast_fp16, y = var_38000_to_fp16)[name = string("aw_chunk_3975_cast_fp16")]; + fp16 var_38002_to_fp16 = const()[name = string("op_38002_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3977_cast_fp16, y = var_38002_to_fp16)[name = string("aw_chunk_3977_cast_fp16")]; + fp16 var_38004_to_fp16 = const()[name = string("op_38004_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3979_cast_fp16, y = var_38004_to_fp16)[name = string("aw_chunk_3979_cast_fp16")]; + fp16 var_38006_to_fp16 = const()[name = string("op_38006_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3981_cast_fp16, y = var_38006_to_fp16)[name = string("aw_chunk_3981_cast_fp16")]; + fp16 var_38008_to_fp16 = const()[name = string("op_38008_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3983_cast_fp16, y = var_38008_to_fp16)[name = string("aw_chunk_3983_cast_fp16")]; + fp16 var_38010_to_fp16 = const()[name = string("op_38010_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3985_cast_fp16, y = var_38010_to_fp16)[name = string("aw_chunk_3985_cast_fp16")]; + fp16 var_38012_to_fp16 = const()[name = string("op_38012_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3987_cast_fp16, y = var_38012_to_fp16)[name = string("aw_chunk_3987_cast_fp16")]; + fp16 var_38014_to_fp16 = const()[name = string("op_38014_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3989_cast_fp16, y = var_38014_to_fp16)[name = string("aw_chunk_3989_cast_fp16")]; + fp16 var_38016_to_fp16 = const()[name = string("op_38016_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3991_cast_fp16, y = var_38016_to_fp16)[name = string("aw_chunk_3991_cast_fp16")]; + fp16 var_38018_to_fp16 = const()[name = string("op_38018_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3993_cast_fp16, y = var_38018_to_fp16)[name = string("aw_chunk_3993_cast_fp16")]; + fp16 var_38020_to_fp16 = const()[name = string("op_38020_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3995_cast_fp16, y = var_38020_to_fp16)[name = string("aw_chunk_3995_cast_fp16")]; + fp16 var_38022_to_fp16 = const()[name = string("op_38022_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3997_cast_fp16, y = var_38022_to_fp16)[name = string("aw_chunk_3997_cast_fp16")]; + fp16 var_38024_to_fp16 = const()[name = string("op_38024_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3999_cast_fp16, y = var_38024_to_fp16)[name = string("aw_chunk_3999_cast_fp16")]; + tensor var_38026_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3841_cast_fp16)[name = string("op_38026_cast_fp16")]; + tensor var_38027_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3843_cast_fp16)[name = string("op_38027_cast_fp16")]; + tensor var_38028_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3845_cast_fp16)[name = string("op_38028_cast_fp16")]; + tensor var_38029_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3847_cast_fp16)[name = string("op_38029_cast_fp16")]; + tensor var_38030_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3849_cast_fp16)[name = string("op_38030_cast_fp16")]; + tensor var_38031_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3851_cast_fp16)[name = string("op_38031_cast_fp16")]; + tensor var_38032_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3853_cast_fp16)[name = string("op_38032_cast_fp16")]; + tensor var_38033_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3855_cast_fp16)[name = string("op_38033_cast_fp16")]; + tensor var_38034_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3857_cast_fp16)[name = string("op_38034_cast_fp16")]; + tensor var_38035_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3859_cast_fp16)[name = string("op_38035_cast_fp16")]; + tensor var_38036_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3861_cast_fp16)[name = string("op_38036_cast_fp16")]; + tensor var_38037_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3863_cast_fp16)[name = string("op_38037_cast_fp16")]; + tensor var_38038_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3865_cast_fp16)[name = string("op_38038_cast_fp16")]; + tensor var_38039_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3867_cast_fp16)[name = string("op_38039_cast_fp16")]; + tensor var_38040_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3869_cast_fp16)[name = string("op_38040_cast_fp16")]; + tensor var_38041_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3871_cast_fp16)[name = string("op_38041_cast_fp16")]; + tensor var_38042_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3873_cast_fp16)[name = string("op_38042_cast_fp16")]; + tensor var_38043_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3875_cast_fp16)[name = string("op_38043_cast_fp16")]; + tensor var_38044_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3877_cast_fp16)[name = string("op_38044_cast_fp16")]; + tensor var_38045_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3879_cast_fp16)[name = string("op_38045_cast_fp16")]; + tensor var_38046_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3881_cast_fp16)[name = string("op_38046_cast_fp16")]; + tensor var_38047_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3883_cast_fp16)[name = string("op_38047_cast_fp16")]; + tensor var_38048_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3885_cast_fp16)[name = string("op_38048_cast_fp16")]; + tensor var_38049_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3887_cast_fp16)[name = string("op_38049_cast_fp16")]; + tensor var_38050_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3889_cast_fp16)[name = string("op_38050_cast_fp16")]; + tensor var_38051_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3891_cast_fp16)[name = string("op_38051_cast_fp16")]; + tensor var_38052_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3893_cast_fp16)[name = string("op_38052_cast_fp16")]; + tensor var_38053_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3895_cast_fp16)[name = string("op_38053_cast_fp16")]; + tensor var_38054_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3897_cast_fp16)[name = string("op_38054_cast_fp16")]; + tensor var_38055_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3899_cast_fp16)[name = string("op_38055_cast_fp16")]; + tensor var_38056_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3901_cast_fp16)[name = string("op_38056_cast_fp16")]; + tensor var_38057_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3903_cast_fp16)[name = string("op_38057_cast_fp16")]; + tensor var_38058_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3905_cast_fp16)[name = string("op_38058_cast_fp16")]; + tensor var_38059_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3907_cast_fp16)[name = string("op_38059_cast_fp16")]; + tensor var_38060_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3909_cast_fp16)[name = string("op_38060_cast_fp16")]; + tensor var_38061_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3911_cast_fp16)[name = string("op_38061_cast_fp16")]; + tensor var_38062_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3913_cast_fp16)[name = string("op_38062_cast_fp16")]; + tensor var_38063_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3915_cast_fp16)[name = string("op_38063_cast_fp16")]; + tensor var_38064_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3917_cast_fp16)[name = string("op_38064_cast_fp16")]; + tensor var_38065_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3919_cast_fp16)[name = string("op_38065_cast_fp16")]; + tensor var_38066_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3921_cast_fp16)[name = string("op_38066_cast_fp16")]; + tensor var_38067_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3923_cast_fp16)[name = string("op_38067_cast_fp16")]; + tensor var_38068_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3925_cast_fp16)[name = string("op_38068_cast_fp16")]; + tensor var_38069_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3927_cast_fp16)[name = string("op_38069_cast_fp16")]; + tensor var_38070_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3929_cast_fp16)[name = string("op_38070_cast_fp16")]; + tensor var_38071_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3931_cast_fp16)[name = string("op_38071_cast_fp16")]; + tensor var_38072_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3933_cast_fp16)[name = string("op_38072_cast_fp16")]; + tensor var_38073_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3935_cast_fp16)[name = string("op_38073_cast_fp16")]; + tensor var_38074_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3937_cast_fp16)[name = string("op_38074_cast_fp16")]; + tensor var_38075_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3939_cast_fp16)[name = string("op_38075_cast_fp16")]; + tensor var_38076_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3941_cast_fp16)[name = string("op_38076_cast_fp16")]; + tensor var_38077_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3943_cast_fp16)[name = string("op_38077_cast_fp16")]; + tensor var_38078_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3945_cast_fp16)[name = string("op_38078_cast_fp16")]; + tensor var_38079_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3947_cast_fp16)[name = string("op_38079_cast_fp16")]; + tensor var_38080_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3949_cast_fp16)[name = string("op_38080_cast_fp16")]; + tensor var_38081_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3951_cast_fp16)[name = string("op_38081_cast_fp16")]; + tensor var_38082_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3953_cast_fp16)[name = string("op_38082_cast_fp16")]; + tensor var_38083_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3955_cast_fp16)[name = string("op_38083_cast_fp16")]; + tensor var_38084_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3957_cast_fp16)[name = string("op_38084_cast_fp16")]; + tensor var_38085_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3959_cast_fp16)[name = string("op_38085_cast_fp16")]; + tensor var_38086_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3961_cast_fp16)[name = string("op_38086_cast_fp16")]; + tensor var_38087_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3963_cast_fp16)[name = string("op_38087_cast_fp16")]; + tensor var_38088_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3965_cast_fp16)[name = string("op_38088_cast_fp16")]; + tensor var_38089_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3967_cast_fp16)[name = string("op_38089_cast_fp16")]; + tensor var_38090_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3969_cast_fp16)[name = string("op_38090_cast_fp16")]; + tensor var_38091_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3971_cast_fp16)[name = string("op_38091_cast_fp16")]; + tensor var_38092_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3973_cast_fp16)[name = string("op_38092_cast_fp16")]; + tensor var_38093_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3975_cast_fp16)[name = string("op_38093_cast_fp16")]; + tensor var_38094_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3977_cast_fp16)[name = string("op_38094_cast_fp16")]; + tensor var_38095_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3979_cast_fp16)[name = string("op_38095_cast_fp16")]; + tensor var_38096_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3981_cast_fp16)[name = string("op_38096_cast_fp16")]; + tensor var_38097_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3983_cast_fp16)[name = string("op_38097_cast_fp16")]; + tensor var_38098_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3985_cast_fp16)[name = string("op_38098_cast_fp16")]; + tensor var_38099_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3987_cast_fp16)[name = string("op_38099_cast_fp16")]; + tensor var_38100_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3989_cast_fp16)[name = string("op_38100_cast_fp16")]; + tensor var_38101_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3991_cast_fp16)[name = string("op_38101_cast_fp16")]; + tensor var_38102_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3993_cast_fp16)[name = string("op_38102_cast_fp16")]; + tensor var_38103_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3995_cast_fp16)[name = string("op_38103_cast_fp16")]; + tensor var_38104_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3997_cast_fp16)[name = string("op_38104_cast_fp16")]; + tensor var_38105_cast_fp16 = softmax(axis = var_36851, x = aw_chunk_3999_cast_fp16)[name = string("op_38105_cast_fp16")]; + string var_38107_equation_0 = const()[name = string("op_38107_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38107_cast_fp16 = einsum(equation = var_38107_equation_0, values = (var_37627_cast_fp16, var_38026_cast_fp16))[name = string("op_38107_cast_fp16")]; + string var_38109_equation_0 = const()[name = string("op_38109_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38109_cast_fp16 = einsum(equation = var_38109_equation_0, values = (var_37627_cast_fp16, var_38027_cast_fp16))[name = string("op_38109_cast_fp16")]; + string var_38111_equation_0 = const()[name = string("op_38111_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38111_cast_fp16 = einsum(equation = var_38111_equation_0, values = (var_37627_cast_fp16, var_38028_cast_fp16))[name = string("op_38111_cast_fp16")]; + string var_38113_equation_0 = const()[name = string("op_38113_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38113_cast_fp16 = einsum(equation = var_38113_equation_0, values = (var_37627_cast_fp16, var_38029_cast_fp16))[name = string("op_38113_cast_fp16")]; + string var_38115_equation_0 = const()[name = string("op_38115_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38115_cast_fp16 = einsum(equation = var_38115_equation_0, values = (var_37631_cast_fp16, var_38030_cast_fp16))[name = string("op_38115_cast_fp16")]; + string var_38117_equation_0 = const()[name = string("op_38117_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38117_cast_fp16 = einsum(equation = var_38117_equation_0, values = (var_37631_cast_fp16, var_38031_cast_fp16))[name = string("op_38117_cast_fp16")]; + string var_38119_equation_0 = const()[name = string("op_38119_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38119_cast_fp16 = einsum(equation = var_38119_equation_0, values = (var_37631_cast_fp16, var_38032_cast_fp16))[name = string("op_38119_cast_fp16")]; + string var_38121_equation_0 = const()[name = string("op_38121_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38121_cast_fp16 = einsum(equation = var_38121_equation_0, values = (var_37631_cast_fp16, var_38033_cast_fp16))[name = string("op_38121_cast_fp16")]; + string var_38123_equation_0 = const()[name = string("op_38123_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38123_cast_fp16 = einsum(equation = var_38123_equation_0, values = (var_37635_cast_fp16, var_38034_cast_fp16))[name = string("op_38123_cast_fp16")]; + string var_38125_equation_0 = const()[name = string("op_38125_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38125_cast_fp16 = einsum(equation = var_38125_equation_0, values = (var_37635_cast_fp16, var_38035_cast_fp16))[name = string("op_38125_cast_fp16")]; + string var_38127_equation_0 = const()[name = string("op_38127_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38127_cast_fp16 = einsum(equation = var_38127_equation_0, values = (var_37635_cast_fp16, var_38036_cast_fp16))[name = string("op_38127_cast_fp16")]; + string var_38129_equation_0 = const()[name = string("op_38129_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38129_cast_fp16 = einsum(equation = var_38129_equation_0, values = (var_37635_cast_fp16, var_38037_cast_fp16))[name = string("op_38129_cast_fp16")]; + string var_38131_equation_0 = const()[name = string("op_38131_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38131_cast_fp16 = einsum(equation = var_38131_equation_0, values = (var_37639_cast_fp16, var_38038_cast_fp16))[name = string("op_38131_cast_fp16")]; + string var_38133_equation_0 = const()[name = string("op_38133_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38133_cast_fp16 = einsum(equation = var_38133_equation_0, values = (var_37639_cast_fp16, var_38039_cast_fp16))[name = string("op_38133_cast_fp16")]; + string var_38135_equation_0 = const()[name = string("op_38135_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38135_cast_fp16 = einsum(equation = var_38135_equation_0, values = (var_37639_cast_fp16, var_38040_cast_fp16))[name = string("op_38135_cast_fp16")]; + string var_38137_equation_0 = const()[name = string("op_38137_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38137_cast_fp16 = einsum(equation = var_38137_equation_0, values = (var_37639_cast_fp16, var_38041_cast_fp16))[name = string("op_38137_cast_fp16")]; + string var_38139_equation_0 = const()[name = string("op_38139_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38139_cast_fp16 = einsum(equation = var_38139_equation_0, values = (var_37643_cast_fp16, var_38042_cast_fp16))[name = string("op_38139_cast_fp16")]; + string var_38141_equation_0 = const()[name = string("op_38141_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38141_cast_fp16 = einsum(equation = var_38141_equation_0, values = (var_37643_cast_fp16, var_38043_cast_fp16))[name = string("op_38141_cast_fp16")]; + string var_38143_equation_0 = const()[name = string("op_38143_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38143_cast_fp16 = einsum(equation = var_38143_equation_0, values = (var_37643_cast_fp16, var_38044_cast_fp16))[name = string("op_38143_cast_fp16")]; + string var_38145_equation_0 = const()[name = string("op_38145_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38145_cast_fp16 = einsum(equation = var_38145_equation_0, values = (var_37643_cast_fp16, var_38045_cast_fp16))[name = string("op_38145_cast_fp16")]; + string var_38147_equation_0 = const()[name = string("op_38147_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38147_cast_fp16 = einsum(equation = var_38147_equation_0, values = (var_37647_cast_fp16, var_38046_cast_fp16))[name = string("op_38147_cast_fp16")]; + string var_38149_equation_0 = const()[name = string("op_38149_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38149_cast_fp16 = einsum(equation = var_38149_equation_0, values = (var_37647_cast_fp16, var_38047_cast_fp16))[name = string("op_38149_cast_fp16")]; + string var_38151_equation_0 = const()[name = string("op_38151_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38151_cast_fp16 = einsum(equation = var_38151_equation_0, values = (var_37647_cast_fp16, var_38048_cast_fp16))[name = string("op_38151_cast_fp16")]; + string var_38153_equation_0 = const()[name = string("op_38153_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38153_cast_fp16 = einsum(equation = var_38153_equation_0, values = (var_37647_cast_fp16, var_38049_cast_fp16))[name = string("op_38153_cast_fp16")]; + string var_38155_equation_0 = const()[name = string("op_38155_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38155_cast_fp16 = einsum(equation = var_38155_equation_0, values = (var_37651_cast_fp16, var_38050_cast_fp16))[name = string("op_38155_cast_fp16")]; + string var_38157_equation_0 = const()[name = string("op_38157_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38157_cast_fp16 = einsum(equation = var_38157_equation_0, values = (var_37651_cast_fp16, var_38051_cast_fp16))[name = string("op_38157_cast_fp16")]; + string var_38159_equation_0 = const()[name = string("op_38159_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38159_cast_fp16 = einsum(equation = var_38159_equation_0, values = (var_37651_cast_fp16, var_38052_cast_fp16))[name = string("op_38159_cast_fp16")]; + string var_38161_equation_0 = const()[name = string("op_38161_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38161_cast_fp16 = einsum(equation = var_38161_equation_0, values = (var_37651_cast_fp16, var_38053_cast_fp16))[name = string("op_38161_cast_fp16")]; + string var_38163_equation_0 = const()[name = string("op_38163_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38163_cast_fp16 = einsum(equation = var_38163_equation_0, values = (var_37655_cast_fp16, var_38054_cast_fp16))[name = string("op_38163_cast_fp16")]; + string var_38165_equation_0 = const()[name = string("op_38165_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38165_cast_fp16 = einsum(equation = var_38165_equation_0, values = (var_37655_cast_fp16, var_38055_cast_fp16))[name = string("op_38165_cast_fp16")]; + string var_38167_equation_0 = const()[name = string("op_38167_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38167_cast_fp16 = einsum(equation = var_38167_equation_0, values = (var_37655_cast_fp16, var_38056_cast_fp16))[name = string("op_38167_cast_fp16")]; + string var_38169_equation_0 = const()[name = string("op_38169_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38169_cast_fp16 = einsum(equation = var_38169_equation_0, values = (var_37655_cast_fp16, var_38057_cast_fp16))[name = string("op_38169_cast_fp16")]; + string var_38171_equation_0 = const()[name = string("op_38171_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38171_cast_fp16 = einsum(equation = var_38171_equation_0, values = (var_37659_cast_fp16, var_38058_cast_fp16))[name = string("op_38171_cast_fp16")]; + string var_38173_equation_0 = const()[name = string("op_38173_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38173_cast_fp16 = einsum(equation = var_38173_equation_0, values = (var_37659_cast_fp16, var_38059_cast_fp16))[name = string("op_38173_cast_fp16")]; + string var_38175_equation_0 = const()[name = string("op_38175_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38175_cast_fp16 = einsum(equation = var_38175_equation_0, values = (var_37659_cast_fp16, var_38060_cast_fp16))[name = string("op_38175_cast_fp16")]; + string var_38177_equation_0 = const()[name = string("op_38177_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38177_cast_fp16 = einsum(equation = var_38177_equation_0, values = (var_37659_cast_fp16, var_38061_cast_fp16))[name = string("op_38177_cast_fp16")]; + string var_38179_equation_0 = const()[name = string("op_38179_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38179_cast_fp16 = einsum(equation = var_38179_equation_0, values = (var_37663_cast_fp16, var_38062_cast_fp16))[name = string("op_38179_cast_fp16")]; + string var_38181_equation_0 = const()[name = string("op_38181_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38181_cast_fp16 = einsum(equation = var_38181_equation_0, values = (var_37663_cast_fp16, var_38063_cast_fp16))[name = string("op_38181_cast_fp16")]; + string var_38183_equation_0 = const()[name = string("op_38183_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38183_cast_fp16 = einsum(equation = var_38183_equation_0, values = (var_37663_cast_fp16, var_38064_cast_fp16))[name = string("op_38183_cast_fp16")]; + string var_38185_equation_0 = const()[name = string("op_38185_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38185_cast_fp16 = einsum(equation = var_38185_equation_0, values = (var_37663_cast_fp16, var_38065_cast_fp16))[name = string("op_38185_cast_fp16")]; + string var_38187_equation_0 = const()[name = string("op_38187_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38187_cast_fp16 = einsum(equation = var_38187_equation_0, values = (var_37667_cast_fp16, var_38066_cast_fp16))[name = string("op_38187_cast_fp16")]; + string var_38189_equation_0 = const()[name = string("op_38189_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38189_cast_fp16 = einsum(equation = var_38189_equation_0, values = (var_37667_cast_fp16, var_38067_cast_fp16))[name = string("op_38189_cast_fp16")]; + string var_38191_equation_0 = const()[name = string("op_38191_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38191_cast_fp16 = einsum(equation = var_38191_equation_0, values = (var_37667_cast_fp16, var_38068_cast_fp16))[name = string("op_38191_cast_fp16")]; + string var_38193_equation_0 = const()[name = string("op_38193_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38193_cast_fp16 = einsum(equation = var_38193_equation_0, values = (var_37667_cast_fp16, var_38069_cast_fp16))[name = string("op_38193_cast_fp16")]; + string var_38195_equation_0 = const()[name = string("op_38195_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38195_cast_fp16 = einsum(equation = var_38195_equation_0, values = (var_37671_cast_fp16, var_38070_cast_fp16))[name = string("op_38195_cast_fp16")]; + string var_38197_equation_0 = const()[name = string("op_38197_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38197_cast_fp16 = einsum(equation = var_38197_equation_0, values = (var_37671_cast_fp16, var_38071_cast_fp16))[name = string("op_38197_cast_fp16")]; + string var_38199_equation_0 = const()[name = string("op_38199_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38199_cast_fp16 = einsum(equation = var_38199_equation_0, values = (var_37671_cast_fp16, var_38072_cast_fp16))[name = string("op_38199_cast_fp16")]; + string var_38201_equation_0 = const()[name = string("op_38201_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38201_cast_fp16 = einsum(equation = var_38201_equation_0, values = (var_37671_cast_fp16, var_38073_cast_fp16))[name = string("op_38201_cast_fp16")]; + string var_38203_equation_0 = const()[name = string("op_38203_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38203_cast_fp16 = einsum(equation = var_38203_equation_0, values = (var_37675_cast_fp16, var_38074_cast_fp16))[name = string("op_38203_cast_fp16")]; + string var_38205_equation_0 = const()[name = string("op_38205_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38205_cast_fp16 = einsum(equation = var_38205_equation_0, values = (var_37675_cast_fp16, var_38075_cast_fp16))[name = string("op_38205_cast_fp16")]; + string var_38207_equation_0 = const()[name = string("op_38207_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38207_cast_fp16 = einsum(equation = var_38207_equation_0, values = (var_37675_cast_fp16, var_38076_cast_fp16))[name = string("op_38207_cast_fp16")]; + string var_38209_equation_0 = const()[name = string("op_38209_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38209_cast_fp16 = einsum(equation = var_38209_equation_0, values = (var_37675_cast_fp16, var_38077_cast_fp16))[name = string("op_38209_cast_fp16")]; + string var_38211_equation_0 = const()[name = string("op_38211_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38211_cast_fp16 = einsum(equation = var_38211_equation_0, values = (var_37679_cast_fp16, var_38078_cast_fp16))[name = string("op_38211_cast_fp16")]; + string var_38213_equation_0 = const()[name = string("op_38213_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38213_cast_fp16 = einsum(equation = var_38213_equation_0, values = (var_37679_cast_fp16, var_38079_cast_fp16))[name = string("op_38213_cast_fp16")]; + string var_38215_equation_0 = const()[name = string("op_38215_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38215_cast_fp16 = einsum(equation = var_38215_equation_0, values = (var_37679_cast_fp16, var_38080_cast_fp16))[name = string("op_38215_cast_fp16")]; + string var_38217_equation_0 = const()[name = string("op_38217_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38217_cast_fp16 = einsum(equation = var_38217_equation_0, values = (var_37679_cast_fp16, var_38081_cast_fp16))[name = string("op_38217_cast_fp16")]; + string var_38219_equation_0 = const()[name = string("op_38219_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38219_cast_fp16 = einsum(equation = var_38219_equation_0, values = (var_37683_cast_fp16, var_38082_cast_fp16))[name = string("op_38219_cast_fp16")]; + string var_38221_equation_0 = const()[name = string("op_38221_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38221_cast_fp16 = einsum(equation = var_38221_equation_0, values = (var_37683_cast_fp16, var_38083_cast_fp16))[name = string("op_38221_cast_fp16")]; + string var_38223_equation_0 = const()[name = string("op_38223_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38223_cast_fp16 = einsum(equation = var_38223_equation_0, values = (var_37683_cast_fp16, var_38084_cast_fp16))[name = string("op_38223_cast_fp16")]; + string var_38225_equation_0 = const()[name = string("op_38225_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38225_cast_fp16 = einsum(equation = var_38225_equation_0, values = (var_37683_cast_fp16, var_38085_cast_fp16))[name = string("op_38225_cast_fp16")]; + string var_38227_equation_0 = const()[name = string("op_38227_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38227_cast_fp16 = einsum(equation = var_38227_equation_0, values = (var_37687_cast_fp16, var_38086_cast_fp16))[name = string("op_38227_cast_fp16")]; + string var_38229_equation_0 = const()[name = string("op_38229_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38229_cast_fp16 = einsum(equation = var_38229_equation_0, values = (var_37687_cast_fp16, var_38087_cast_fp16))[name = string("op_38229_cast_fp16")]; + string var_38231_equation_0 = const()[name = string("op_38231_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38231_cast_fp16 = einsum(equation = var_38231_equation_0, values = (var_37687_cast_fp16, var_38088_cast_fp16))[name = string("op_38231_cast_fp16")]; + string var_38233_equation_0 = const()[name = string("op_38233_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38233_cast_fp16 = einsum(equation = var_38233_equation_0, values = (var_37687_cast_fp16, var_38089_cast_fp16))[name = string("op_38233_cast_fp16")]; + string var_38235_equation_0 = const()[name = string("op_38235_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38235_cast_fp16 = einsum(equation = var_38235_equation_0, values = (var_37691_cast_fp16, var_38090_cast_fp16))[name = string("op_38235_cast_fp16")]; + string var_38237_equation_0 = const()[name = string("op_38237_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38237_cast_fp16 = einsum(equation = var_38237_equation_0, values = (var_37691_cast_fp16, var_38091_cast_fp16))[name = string("op_38237_cast_fp16")]; + string var_38239_equation_0 = const()[name = string("op_38239_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38239_cast_fp16 = einsum(equation = var_38239_equation_0, values = (var_37691_cast_fp16, var_38092_cast_fp16))[name = string("op_38239_cast_fp16")]; + string var_38241_equation_0 = const()[name = string("op_38241_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38241_cast_fp16 = einsum(equation = var_38241_equation_0, values = (var_37691_cast_fp16, var_38093_cast_fp16))[name = string("op_38241_cast_fp16")]; + string var_38243_equation_0 = const()[name = string("op_38243_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38243_cast_fp16 = einsum(equation = var_38243_equation_0, values = (var_37695_cast_fp16, var_38094_cast_fp16))[name = string("op_38243_cast_fp16")]; + string var_38245_equation_0 = const()[name = string("op_38245_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38245_cast_fp16 = einsum(equation = var_38245_equation_0, values = (var_37695_cast_fp16, var_38095_cast_fp16))[name = string("op_38245_cast_fp16")]; + string var_38247_equation_0 = const()[name = string("op_38247_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38247_cast_fp16 = einsum(equation = var_38247_equation_0, values = (var_37695_cast_fp16, var_38096_cast_fp16))[name = string("op_38247_cast_fp16")]; + string var_38249_equation_0 = const()[name = string("op_38249_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38249_cast_fp16 = einsum(equation = var_38249_equation_0, values = (var_37695_cast_fp16, var_38097_cast_fp16))[name = string("op_38249_cast_fp16")]; + string var_38251_equation_0 = const()[name = string("op_38251_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38251_cast_fp16 = einsum(equation = var_38251_equation_0, values = (var_37699_cast_fp16, var_38098_cast_fp16))[name = string("op_38251_cast_fp16")]; + string var_38253_equation_0 = const()[name = string("op_38253_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38253_cast_fp16 = einsum(equation = var_38253_equation_0, values = (var_37699_cast_fp16, var_38099_cast_fp16))[name = string("op_38253_cast_fp16")]; + string var_38255_equation_0 = const()[name = string("op_38255_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38255_cast_fp16 = einsum(equation = var_38255_equation_0, values = (var_37699_cast_fp16, var_38100_cast_fp16))[name = string("op_38255_cast_fp16")]; + string var_38257_equation_0 = const()[name = string("op_38257_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38257_cast_fp16 = einsum(equation = var_38257_equation_0, values = (var_37699_cast_fp16, var_38101_cast_fp16))[name = string("op_38257_cast_fp16")]; + string var_38259_equation_0 = const()[name = string("op_38259_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38259_cast_fp16 = einsum(equation = var_38259_equation_0, values = (var_37703_cast_fp16, var_38102_cast_fp16))[name = string("op_38259_cast_fp16")]; + string var_38261_equation_0 = const()[name = string("op_38261_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38261_cast_fp16 = einsum(equation = var_38261_equation_0, values = (var_37703_cast_fp16, var_38103_cast_fp16))[name = string("op_38261_cast_fp16")]; + string var_38263_equation_0 = const()[name = string("op_38263_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38263_cast_fp16 = einsum(equation = var_38263_equation_0, values = (var_37703_cast_fp16, var_38104_cast_fp16))[name = string("op_38263_cast_fp16")]; + string var_38265_equation_0 = const()[name = string("op_38265_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38265_cast_fp16 = einsum(equation = var_38265_equation_0, values = (var_37703_cast_fp16, var_38105_cast_fp16))[name = string("op_38265_cast_fp16")]; + bool var_38267_interleave_0 = const()[name = string("op_38267_interleave_0"), val = bool(false)]; + tensor var_38267_cast_fp16 = concat(axis = var_36826, interleave = var_38267_interleave_0, values = (var_38107_cast_fp16, var_38109_cast_fp16, var_38111_cast_fp16, var_38113_cast_fp16))[name = string("op_38267_cast_fp16")]; + bool var_38269_interleave_0 = const()[name = string("op_38269_interleave_0"), val = bool(false)]; + tensor var_38269_cast_fp16 = concat(axis = var_36826, interleave = var_38269_interleave_0, values = (var_38115_cast_fp16, var_38117_cast_fp16, var_38119_cast_fp16, var_38121_cast_fp16))[name = string("op_38269_cast_fp16")]; + bool var_38271_interleave_0 = const()[name = string("op_38271_interleave_0"), val = bool(false)]; + tensor var_38271_cast_fp16 = concat(axis = var_36826, interleave = var_38271_interleave_0, values = (var_38123_cast_fp16, var_38125_cast_fp16, var_38127_cast_fp16, var_38129_cast_fp16))[name = string("op_38271_cast_fp16")]; + bool var_38273_interleave_0 = const()[name = string("op_38273_interleave_0"), val = bool(false)]; + tensor var_38273_cast_fp16 = concat(axis = var_36826, interleave = var_38273_interleave_0, values = (var_38131_cast_fp16, var_38133_cast_fp16, var_38135_cast_fp16, var_38137_cast_fp16))[name = string("op_38273_cast_fp16")]; + bool var_38275_interleave_0 = const()[name = string("op_38275_interleave_0"), val = bool(false)]; + tensor var_38275_cast_fp16 = concat(axis = var_36826, interleave = var_38275_interleave_0, values = (var_38139_cast_fp16, var_38141_cast_fp16, var_38143_cast_fp16, var_38145_cast_fp16))[name = string("op_38275_cast_fp16")]; + bool var_38277_interleave_0 = const()[name = string("op_38277_interleave_0"), val = bool(false)]; + tensor var_38277_cast_fp16 = concat(axis = var_36826, interleave = var_38277_interleave_0, values = (var_38147_cast_fp16, var_38149_cast_fp16, var_38151_cast_fp16, var_38153_cast_fp16))[name = string("op_38277_cast_fp16")]; + bool var_38279_interleave_0 = const()[name = string("op_38279_interleave_0"), val = bool(false)]; + tensor var_38279_cast_fp16 = concat(axis = var_36826, interleave = var_38279_interleave_0, values = (var_38155_cast_fp16, var_38157_cast_fp16, var_38159_cast_fp16, var_38161_cast_fp16))[name = string("op_38279_cast_fp16")]; + bool var_38281_interleave_0 = const()[name = string("op_38281_interleave_0"), val = bool(false)]; + tensor var_38281_cast_fp16 = concat(axis = var_36826, interleave = var_38281_interleave_0, values = (var_38163_cast_fp16, var_38165_cast_fp16, var_38167_cast_fp16, var_38169_cast_fp16))[name = string("op_38281_cast_fp16")]; + bool var_38283_interleave_0 = const()[name = string("op_38283_interleave_0"), val = bool(false)]; + tensor var_38283_cast_fp16 = concat(axis = var_36826, interleave = var_38283_interleave_0, values = (var_38171_cast_fp16, var_38173_cast_fp16, var_38175_cast_fp16, var_38177_cast_fp16))[name = string("op_38283_cast_fp16")]; + bool var_38285_interleave_0 = const()[name = string("op_38285_interleave_0"), val = bool(false)]; + tensor var_38285_cast_fp16 = concat(axis = var_36826, interleave = var_38285_interleave_0, values = (var_38179_cast_fp16, var_38181_cast_fp16, var_38183_cast_fp16, var_38185_cast_fp16))[name = string("op_38285_cast_fp16")]; + bool var_38287_interleave_0 = const()[name = string("op_38287_interleave_0"), val = bool(false)]; + tensor var_38287_cast_fp16 = concat(axis = var_36826, interleave = var_38287_interleave_0, values = (var_38187_cast_fp16, var_38189_cast_fp16, var_38191_cast_fp16, var_38193_cast_fp16))[name = string("op_38287_cast_fp16")]; + bool var_38289_interleave_0 = const()[name = string("op_38289_interleave_0"), val = bool(false)]; + tensor var_38289_cast_fp16 = concat(axis = var_36826, interleave = var_38289_interleave_0, values = (var_38195_cast_fp16, var_38197_cast_fp16, var_38199_cast_fp16, var_38201_cast_fp16))[name = string("op_38289_cast_fp16")]; + bool var_38291_interleave_0 = const()[name = string("op_38291_interleave_0"), val = bool(false)]; + tensor var_38291_cast_fp16 = concat(axis = var_36826, interleave = var_38291_interleave_0, values = (var_38203_cast_fp16, var_38205_cast_fp16, var_38207_cast_fp16, var_38209_cast_fp16))[name = string("op_38291_cast_fp16")]; + bool var_38293_interleave_0 = const()[name = string("op_38293_interleave_0"), val = bool(false)]; + tensor var_38293_cast_fp16 = concat(axis = var_36826, interleave = var_38293_interleave_0, values = (var_38211_cast_fp16, var_38213_cast_fp16, var_38215_cast_fp16, var_38217_cast_fp16))[name = string("op_38293_cast_fp16")]; + bool var_38295_interleave_0 = const()[name = string("op_38295_interleave_0"), val = bool(false)]; + tensor var_38295_cast_fp16 = concat(axis = var_36826, interleave = var_38295_interleave_0, values = (var_38219_cast_fp16, var_38221_cast_fp16, var_38223_cast_fp16, var_38225_cast_fp16))[name = string("op_38295_cast_fp16")]; + bool var_38297_interleave_0 = const()[name = string("op_38297_interleave_0"), val = bool(false)]; + tensor var_38297_cast_fp16 = concat(axis = var_36826, interleave = var_38297_interleave_0, values = (var_38227_cast_fp16, var_38229_cast_fp16, var_38231_cast_fp16, var_38233_cast_fp16))[name = string("op_38297_cast_fp16")]; + bool var_38299_interleave_0 = const()[name = string("op_38299_interleave_0"), val = bool(false)]; + tensor var_38299_cast_fp16 = concat(axis = var_36826, interleave = var_38299_interleave_0, values = (var_38235_cast_fp16, var_38237_cast_fp16, var_38239_cast_fp16, var_38241_cast_fp16))[name = string("op_38299_cast_fp16")]; + bool var_38301_interleave_0 = const()[name = string("op_38301_interleave_0"), val = bool(false)]; + tensor var_38301_cast_fp16 = concat(axis = var_36826, interleave = var_38301_interleave_0, values = (var_38243_cast_fp16, var_38245_cast_fp16, var_38247_cast_fp16, var_38249_cast_fp16))[name = string("op_38301_cast_fp16")]; + bool var_38303_interleave_0 = const()[name = string("op_38303_interleave_0"), val = bool(false)]; + tensor var_38303_cast_fp16 = concat(axis = var_36826, interleave = var_38303_interleave_0, values = (var_38251_cast_fp16, var_38253_cast_fp16, var_38255_cast_fp16, var_38257_cast_fp16))[name = string("op_38303_cast_fp16")]; + bool var_38305_interleave_0 = const()[name = string("op_38305_interleave_0"), val = bool(false)]; + tensor var_38305_cast_fp16 = concat(axis = var_36826, interleave = var_38305_interleave_0, values = (var_38259_cast_fp16, var_38261_cast_fp16, var_38263_cast_fp16, var_38265_cast_fp16))[name = string("op_38305_cast_fp16")]; + bool input_193_interleave_0 = const()[name = string("input_193_interleave_0"), val = bool(false)]; + tensor input_193_cast_fp16 = concat(axis = var_36851, interleave = input_193_interleave_0, values = (var_38267_cast_fp16, var_38269_cast_fp16, var_38271_cast_fp16, var_38273_cast_fp16, var_38275_cast_fp16, var_38277_cast_fp16, var_38279_cast_fp16, var_38281_cast_fp16, var_38283_cast_fp16, var_38285_cast_fp16, var_38287_cast_fp16, var_38289_cast_fp16, var_38291_cast_fp16, var_38293_cast_fp16, var_38295_cast_fp16, var_38297_cast_fp16, var_38299_cast_fp16, var_38301_cast_fp16, var_38303_cast_fp16, var_38305_cast_fp16))[name = string("input_193_cast_fp16")]; + string obj_99_pad_type_0 = const()[name = string("obj_99_pad_type_0"), val = string("valid")]; + tensor obj_99_strides_0 = const()[name = string("obj_99_strides_0"), val = tensor([1, 1])]; + tensor obj_99_pad_0 = const()[name = string("obj_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_99_dilations_0 = const()[name = string("obj_99_dilations_0"), val = tensor([1, 1])]; + int32 obj_99_groups_0 = const()[name = string("obj_99_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968984000)))]; + tensor layers_24_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972260864)))]; + tensor obj_99_cast_fp16 = conv(bias = layers_24_self_attn_o_proj_bias_to_fp16, dilations = obj_99_dilations_0, groups = obj_99_groups_0, pad = obj_99_pad_0, pad_type = obj_99_pad_type_0, strides = obj_99_strides_0, weight = layers_24_self_attn_o_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("obj_99_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_99_cast_fp16)[name = string("inputs_99_cast_fp16")]; + tensor out_99_axes_0 = const()[name = string("out_99_axes_0"), val = tensor([1])]; + fp16 var_38324_to_fp16 = const()[name = string("op_38324_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_38324_to_fp16, x = inputs_99_cast_fp16)[name = string("out_99_cast_fp16")]; + tensor input_195_gamma_0_to_fp16 = const()[name = string("input_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972263488)))]; + tensor input_195_beta_0_to_fp16 = const()[name = string("input_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972266112)))]; + fp16 input_195_epsilon_0_to_fp16 = const()[name = string("input_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = string("input_195_cast_fp16")]; + string input_197_pad_type_0 = const()[name = string("input_197_pad_type_0"), val = string("valid")]; + tensor input_197_strides_0 = const()[name = string("input_197_strides_0"), val = tensor([1, 1])]; + tensor input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_197_dilations_0 = const()[name = string("input_197_dilations_0"), val = tensor([1, 1])]; + int32 input_197_groups_0 = const()[name = string("input_197_groups_0"), val = int32(1)]; + tensor layers_24_fc1_weight_to_fp16 = const()[name = string("layers_24_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972268736)))]; + tensor layers_24_fc1_bias_to_fp16 = const()[name = string("layers_24_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985376000)))]; + tensor input_197_cast_fp16 = conv(bias = layers_24_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_24_fc1_weight_to_fp16, x = input_195_cast_fp16)[name = string("input_197_cast_fp16")]; + string input_199_mode_0 = const()[name = string("input_199_mode_0"), val = string("EXACT")]; + tensor input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")]; + string hidden_states_53_pad_type_0 = const()[name = string("hidden_states_53_pad_type_0"), val = string("valid")]; + tensor hidden_states_53_strides_0 = const()[name = string("hidden_states_53_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_53_pad_0 = const()[name = string("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_53_dilations_0 = const()[name = string("hidden_states_53_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_53_groups_0 = const()[name = string("hidden_states_53_groups_0"), val = int32(1)]; + tensor layers_24_fc2_weight_to_fp16 = const()[name = string("layers_24_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985386304)))]; + tensor layers_24_fc2_bias_to_fp16 = const()[name = string("layers_24_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998493568)))]; + tensor hidden_states_53_cast_fp16 = conv(bias = layers_24_fc2_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = layers_24_fc2_weight_to_fp16, x = input_199_cast_fp16)[name = string("hidden_states_53_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = hidden_states_53_cast_fp16)[name = string("inputs_101_cast_fp16")]; + int32 var_38353 = const()[name = string("op_38353"), val = int32(3)]; + int32 var_38378 = const()[name = string("op_38378"), val = int32(1)]; + tensor out_101_axes_0 = const()[name = string("out_101_axes_0"), val = tensor([1])]; + fp16 var_38395_to_fp16 = const()[name = string("op_38395_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_38395_to_fp16, x = inputs_101_cast_fp16)[name = string("out_101_cast_fp16")]; + tensor obj_101_gamma_0_to_fp16 = const()[name = string("obj_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998496192)))]; + tensor obj_101_beta_0_to_fp16 = const()[name = string("obj_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998498816)))]; + fp16 obj_101_epsilon_0_to_fp16 = const()[name = string("obj_101_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_101_cast_fp16 = batch_norm(beta = obj_101_beta_0_to_fp16, epsilon = obj_101_epsilon_0_to_fp16, gamma = obj_101_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = string("obj_101_cast_fp16")]; + string query_51_pad_type_0 = const()[name = string("query_51_pad_type_0"), val = string("valid")]; + tensor query_51_strides_0 = const()[name = string("query_51_strides_0"), val = tensor([1, 1])]; + tensor query_51_pad_0 = const()[name = string("query_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_51_dilations_0 = const()[name = string("query_51_dilations_0"), val = tensor([1, 1])]; + int32 query_51_groups_0 = const()[name = string("query_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998501440)))]; + tensor layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001778304)))]; + tensor query_51_cast_fp16 = conv(bias = layers_25_self_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_25_self_attn_q_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("query_51_cast_fp16")]; + string key_51_pad_type_0 = const()[name = string("key_51_pad_type_0"), val = string("valid")]; + tensor key_51_strides_0 = const()[name = string("key_51_strides_0"), val = tensor([1, 1])]; + tensor key_51_pad_0 = const()[name = string("key_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_51_dilations_0 = const()[name = string("key_51_dilations_0"), val = tensor([1, 1])]; + int32 key_51_groups_0 = const()[name = string("key_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001780928)))]; + tensor key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_25_self_attn_k_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("key_51_cast_fp16")]; + string value_51_pad_type_0 = const()[name = string("value_51_pad_type_0"), val = string("valid")]; + tensor value_51_strides_0 = const()[name = string("value_51_strides_0"), val = tensor([1, 1])]; + tensor value_51_pad_0 = const()[name = string("value_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_51_dilations_0 = const()[name = string("value_51_dilations_0"), val = tensor([1, 1])]; + int32 value_51_groups_0 = const()[name = string("value_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1005057792)))]; + tensor layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008334656)))]; + tensor value_51_cast_fp16 = conv(bias = layers_25_self_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_25_self_attn_v_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("value_51_cast_fp16")]; + tensor var_38433_begin_0 = const()[name = string("op_38433_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38433_end_0 = const()[name = string("op_38433_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38433_end_mask_0 = const()[name = string("op_38433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38433_cast_fp16 = slice_by_index(begin = var_38433_begin_0, end = var_38433_end_0, end_mask = var_38433_end_mask_0, x = query_51_cast_fp16)[name = string("op_38433_cast_fp16")]; + tensor var_38437_begin_0 = const()[name = string("op_38437_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38437_end_0 = const()[name = string("op_38437_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_38437_end_mask_0 = const()[name = string("op_38437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38437_cast_fp16 = slice_by_index(begin = var_38437_begin_0, end = var_38437_end_0, end_mask = var_38437_end_mask_0, x = query_51_cast_fp16)[name = string("op_38437_cast_fp16")]; + tensor var_38441_begin_0 = const()[name = string("op_38441_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38441_end_0 = const()[name = string("op_38441_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_38441_end_mask_0 = const()[name = string("op_38441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38441_cast_fp16 = slice_by_index(begin = var_38441_begin_0, end = var_38441_end_0, end_mask = var_38441_end_mask_0, x = query_51_cast_fp16)[name = string("op_38441_cast_fp16")]; + tensor var_38445_begin_0 = const()[name = string("op_38445_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38445_end_0 = const()[name = string("op_38445_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_38445_end_mask_0 = const()[name = string("op_38445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38445_cast_fp16 = slice_by_index(begin = var_38445_begin_0, end = var_38445_end_0, end_mask = var_38445_end_mask_0, x = query_51_cast_fp16)[name = string("op_38445_cast_fp16")]; + tensor var_38449_begin_0 = const()[name = string("op_38449_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38449_end_0 = const()[name = string("op_38449_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_38449_end_mask_0 = const()[name = string("op_38449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38449_cast_fp16 = slice_by_index(begin = var_38449_begin_0, end = var_38449_end_0, end_mask = var_38449_end_mask_0, x = query_51_cast_fp16)[name = string("op_38449_cast_fp16")]; + tensor var_38453_begin_0 = const()[name = string("op_38453_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38453_end_0 = const()[name = string("op_38453_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_38453_end_mask_0 = const()[name = string("op_38453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38453_cast_fp16 = slice_by_index(begin = var_38453_begin_0, end = var_38453_end_0, end_mask = var_38453_end_mask_0, x = query_51_cast_fp16)[name = string("op_38453_cast_fp16")]; + tensor var_38457_begin_0 = const()[name = string("op_38457_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38457_end_0 = const()[name = string("op_38457_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_38457_end_mask_0 = const()[name = string("op_38457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38457_cast_fp16 = slice_by_index(begin = var_38457_begin_0, end = var_38457_end_0, end_mask = var_38457_end_mask_0, x = query_51_cast_fp16)[name = string("op_38457_cast_fp16")]; + tensor var_38461_begin_0 = const()[name = string("op_38461_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38461_end_0 = const()[name = string("op_38461_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_38461_end_mask_0 = const()[name = string("op_38461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38461_cast_fp16 = slice_by_index(begin = var_38461_begin_0, end = var_38461_end_0, end_mask = var_38461_end_mask_0, x = query_51_cast_fp16)[name = string("op_38461_cast_fp16")]; + tensor var_38465_begin_0 = const()[name = string("op_38465_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38465_end_0 = const()[name = string("op_38465_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_38465_end_mask_0 = const()[name = string("op_38465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38465_cast_fp16 = slice_by_index(begin = var_38465_begin_0, end = var_38465_end_0, end_mask = var_38465_end_mask_0, x = query_51_cast_fp16)[name = string("op_38465_cast_fp16")]; + tensor var_38469_begin_0 = const()[name = string("op_38469_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38469_end_0 = const()[name = string("op_38469_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_38469_end_mask_0 = const()[name = string("op_38469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38469_cast_fp16 = slice_by_index(begin = var_38469_begin_0, end = var_38469_end_0, end_mask = var_38469_end_mask_0, x = query_51_cast_fp16)[name = string("op_38469_cast_fp16")]; + tensor var_38473_begin_0 = const()[name = string("op_38473_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38473_end_0 = const()[name = string("op_38473_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_38473_end_mask_0 = const()[name = string("op_38473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38473_cast_fp16 = slice_by_index(begin = var_38473_begin_0, end = var_38473_end_0, end_mask = var_38473_end_mask_0, x = query_51_cast_fp16)[name = string("op_38473_cast_fp16")]; + tensor var_38477_begin_0 = const()[name = string("op_38477_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38477_end_0 = const()[name = string("op_38477_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_38477_end_mask_0 = const()[name = string("op_38477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38477_cast_fp16 = slice_by_index(begin = var_38477_begin_0, end = var_38477_end_0, end_mask = var_38477_end_mask_0, x = query_51_cast_fp16)[name = string("op_38477_cast_fp16")]; + tensor var_38481_begin_0 = const()[name = string("op_38481_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38481_end_0 = const()[name = string("op_38481_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_38481_end_mask_0 = const()[name = string("op_38481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38481_cast_fp16 = slice_by_index(begin = var_38481_begin_0, end = var_38481_end_0, end_mask = var_38481_end_mask_0, x = query_51_cast_fp16)[name = string("op_38481_cast_fp16")]; + tensor var_38485_begin_0 = const()[name = string("op_38485_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38485_end_0 = const()[name = string("op_38485_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_38485_end_mask_0 = const()[name = string("op_38485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38485_cast_fp16 = slice_by_index(begin = var_38485_begin_0, end = var_38485_end_0, end_mask = var_38485_end_mask_0, x = query_51_cast_fp16)[name = string("op_38485_cast_fp16")]; + tensor var_38489_begin_0 = const()[name = string("op_38489_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38489_end_0 = const()[name = string("op_38489_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_38489_end_mask_0 = const()[name = string("op_38489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38489_cast_fp16 = slice_by_index(begin = var_38489_begin_0, end = var_38489_end_0, end_mask = var_38489_end_mask_0, x = query_51_cast_fp16)[name = string("op_38489_cast_fp16")]; + tensor var_38493_begin_0 = const()[name = string("op_38493_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38493_end_0 = const()[name = string("op_38493_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_38493_end_mask_0 = const()[name = string("op_38493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38493_cast_fp16 = slice_by_index(begin = var_38493_begin_0, end = var_38493_end_0, end_mask = var_38493_end_mask_0, x = query_51_cast_fp16)[name = string("op_38493_cast_fp16")]; + tensor var_38497_begin_0 = const()[name = string("op_38497_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38497_end_0 = const()[name = string("op_38497_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_38497_end_mask_0 = const()[name = string("op_38497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38497_cast_fp16 = slice_by_index(begin = var_38497_begin_0, end = var_38497_end_0, end_mask = var_38497_end_mask_0, x = query_51_cast_fp16)[name = string("op_38497_cast_fp16")]; + tensor var_38501_begin_0 = const()[name = string("op_38501_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38501_end_0 = const()[name = string("op_38501_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_38501_end_mask_0 = const()[name = string("op_38501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38501_cast_fp16 = slice_by_index(begin = var_38501_begin_0, end = var_38501_end_0, end_mask = var_38501_end_mask_0, x = query_51_cast_fp16)[name = string("op_38501_cast_fp16")]; + tensor var_38505_begin_0 = const()[name = string("op_38505_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38505_end_0 = const()[name = string("op_38505_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_38505_end_mask_0 = const()[name = string("op_38505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38505_cast_fp16 = slice_by_index(begin = var_38505_begin_0, end = var_38505_end_0, end_mask = var_38505_end_mask_0, x = query_51_cast_fp16)[name = string("op_38505_cast_fp16")]; + tensor var_38509_begin_0 = const()[name = string("op_38509_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38509_end_0 = const()[name = string("op_38509_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_38509_end_mask_0 = const()[name = string("op_38509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38509_cast_fp16 = slice_by_index(begin = var_38509_begin_0, end = var_38509_end_0, end_mask = var_38509_end_mask_0, x = query_51_cast_fp16)[name = string("op_38509_cast_fp16")]; + tensor var_38518_begin_0 = const()[name = string("op_38518_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38518_end_0 = const()[name = string("op_38518_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38518_end_mask_0 = const()[name = string("op_38518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38518_cast_fp16 = slice_by_index(begin = var_38518_begin_0, end = var_38518_end_0, end_mask = var_38518_end_mask_0, x = var_38433_cast_fp16)[name = string("op_38518_cast_fp16")]; + tensor var_38525_begin_0 = const()[name = string("op_38525_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38525_end_0 = const()[name = string("op_38525_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38525_end_mask_0 = const()[name = string("op_38525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38525_cast_fp16 = slice_by_index(begin = var_38525_begin_0, end = var_38525_end_0, end_mask = var_38525_end_mask_0, x = var_38433_cast_fp16)[name = string("op_38525_cast_fp16")]; + tensor var_38532_begin_0 = const()[name = string("op_38532_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38532_end_0 = const()[name = string("op_38532_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38532_end_mask_0 = const()[name = string("op_38532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38532_cast_fp16 = slice_by_index(begin = var_38532_begin_0, end = var_38532_end_0, end_mask = var_38532_end_mask_0, x = var_38433_cast_fp16)[name = string("op_38532_cast_fp16")]; + tensor var_38539_begin_0 = const()[name = string("op_38539_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38539_end_0 = const()[name = string("op_38539_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38539_end_mask_0 = const()[name = string("op_38539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38539_cast_fp16 = slice_by_index(begin = var_38539_begin_0, end = var_38539_end_0, end_mask = var_38539_end_mask_0, x = var_38433_cast_fp16)[name = string("op_38539_cast_fp16")]; + tensor var_38546_begin_0 = const()[name = string("op_38546_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38546_end_0 = const()[name = string("op_38546_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38546_end_mask_0 = const()[name = string("op_38546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38546_cast_fp16 = slice_by_index(begin = var_38546_begin_0, end = var_38546_end_0, end_mask = var_38546_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38546_cast_fp16")]; + tensor var_38553_begin_0 = const()[name = string("op_38553_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38553_end_0 = const()[name = string("op_38553_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38553_end_mask_0 = const()[name = string("op_38553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38553_cast_fp16 = slice_by_index(begin = var_38553_begin_0, end = var_38553_end_0, end_mask = var_38553_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38553_cast_fp16")]; + tensor var_38560_begin_0 = const()[name = string("op_38560_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38560_end_0 = const()[name = string("op_38560_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38560_end_mask_0 = const()[name = string("op_38560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38560_cast_fp16 = slice_by_index(begin = var_38560_begin_0, end = var_38560_end_0, end_mask = var_38560_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38560_cast_fp16")]; + tensor var_38567_begin_0 = const()[name = string("op_38567_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38567_end_0 = const()[name = string("op_38567_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38567_end_mask_0 = const()[name = string("op_38567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38567_cast_fp16 = slice_by_index(begin = var_38567_begin_0, end = var_38567_end_0, end_mask = var_38567_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38567_cast_fp16")]; + tensor var_38574_begin_0 = const()[name = string("op_38574_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38574_end_0 = const()[name = string("op_38574_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38574_end_mask_0 = const()[name = string("op_38574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38574_cast_fp16 = slice_by_index(begin = var_38574_begin_0, end = var_38574_end_0, end_mask = var_38574_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38574_cast_fp16")]; + tensor var_38581_begin_0 = const()[name = string("op_38581_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38581_end_0 = const()[name = string("op_38581_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38581_end_mask_0 = const()[name = string("op_38581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38581_cast_fp16 = slice_by_index(begin = var_38581_begin_0, end = var_38581_end_0, end_mask = var_38581_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38581_cast_fp16")]; + tensor var_38588_begin_0 = const()[name = string("op_38588_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38588_end_0 = const()[name = string("op_38588_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38588_end_mask_0 = const()[name = string("op_38588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38588_cast_fp16 = slice_by_index(begin = var_38588_begin_0, end = var_38588_end_0, end_mask = var_38588_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38588_cast_fp16")]; + tensor var_38595_begin_0 = const()[name = string("op_38595_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38595_end_0 = const()[name = string("op_38595_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38595_end_mask_0 = const()[name = string("op_38595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38595_cast_fp16 = slice_by_index(begin = var_38595_begin_0, end = var_38595_end_0, end_mask = var_38595_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38595_cast_fp16")]; + tensor var_38602_begin_0 = const()[name = string("op_38602_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38602_end_0 = const()[name = string("op_38602_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38602_end_mask_0 = const()[name = string("op_38602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38602_cast_fp16 = slice_by_index(begin = var_38602_begin_0, end = var_38602_end_0, end_mask = var_38602_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38602_cast_fp16")]; + tensor var_38609_begin_0 = const()[name = string("op_38609_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38609_end_0 = const()[name = string("op_38609_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38609_end_mask_0 = const()[name = string("op_38609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38609_cast_fp16 = slice_by_index(begin = var_38609_begin_0, end = var_38609_end_0, end_mask = var_38609_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38609_cast_fp16")]; + tensor var_38616_begin_0 = const()[name = string("op_38616_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38616_end_0 = const()[name = string("op_38616_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38616_end_mask_0 = const()[name = string("op_38616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38616_cast_fp16 = slice_by_index(begin = var_38616_begin_0, end = var_38616_end_0, end_mask = var_38616_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38616_cast_fp16")]; + tensor var_38623_begin_0 = const()[name = string("op_38623_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38623_end_0 = const()[name = string("op_38623_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38623_end_mask_0 = const()[name = string("op_38623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38623_cast_fp16 = slice_by_index(begin = var_38623_begin_0, end = var_38623_end_0, end_mask = var_38623_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38623_cast_fp16")]; + tensor var_38630_begin_0 = const()[name = string("op_38630_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38630_end_0 = const()[name = string("op_38630_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38630_end_mask_0 = const()[name = string("op_38630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38630_cast_fp16 = slice_by_index(begin = var_38630_begin_0, end = var_38630_end_0, end_mask = var_38630_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38630_cast_fp16")]; + tensor var_38637_begin_0 = const()[name = string("op_38637_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38637_end_0 = const()[name = string("op_38637_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38637_end_mask_0 = const()[name = string("op_38637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38637_cast_fp16 = slice_by_index(begin = var_38637_begin_0, end = var_38637_end_0, end_mask = var_38637_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38637_cast_fp16")]; + tensor var_38644_begin_0 = const()[name = string("op_38644_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38644_end_0 = const()[name = string("op_38644_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38644_end_mask_0 = const()[name = string("op_38644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38644_cast_fp16 = slice_by_index(begin = var_38644_begin_0, end = var_38644_end_0, end_mask = var_38644_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38644_cast_fp16")]; + tensor var_38651_begin_0 = const()[name = string("op_38651_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38651_end_0 = const()[name = string("op_38651_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38651_end_mask_0 = const()[name = string("op_38651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38651_cast_fp16 = slice_by_index(begin = var_38651_begin_0, end = var_38651_end_0, end_mask = var_38651_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38651_cast_fp16")]; + tensor var_38658_begin_0 = const()[name = string("op_38658_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38658_end_0 = const()[name = string("op_38658_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38658_end_mask_0 = const()[name = string("op_38658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38658_cast_fp16 = slice_by_index(begin = var_38658_begin_0, end = var_38658_end_0, end_mask = var_38658_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38658_cast_fp16")]; + tensor var_38665_begin_0 = const()[name = string("op_38665_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38665_end_0 = const()[name = string("op_38665_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38665_end_mask_0 = const()[name = string("op_38665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38665_cast_fp16 = slice_by_index(begin = var_38665_begin_0, end = var_38665_end_0, end_mask = var_38665_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38665_cast_fp16")]; + tensor var_38672_begin_0 = const()[name = string("op_38672_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38672_end_0 = const()[name = string("op_38672_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38672_end_mask_0 = const()[name = string("op_38672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38672_cast_fp16 = slice_by_index(begin = var_38672_begin_0, end = var_38672_end_0, end_mask = var_38672_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38672_cast_fp16")]; + tensor var_38679_begin_0 = const()[name = string("op_38679_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38679_end_0 = const()[name = string("op_38679_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38679_end_mask_0 = const()[name = string("op_38679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38679_cast_fp16 = slice_by_index(begin = var_38679_begin_0, end = var_38679_end_0, end_mask = var_38679_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38679_cast_fp16")]; + tensor var_38686_begin_0 = const()[name = string("op_38686_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38686_end_0 = const()[name = string("op_38686_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38686_end_mask_0 = const()[name = string("op_38686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38686_cast_fp16 = slice_by_index(begin = var_38686_begin_0, end = var_38686_end_0, end_mask = var_38686_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38686_cast_fp16")]; + tensor var_38693_begin_0 = const()[name = string("op_38693_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38693_end_0 = const()[name = string("op_38693_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38693_end_mask_0 = const()[name = string("op_38693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38693_cast_fp16 = slice_by_index(begin = var_38693_begin_0, end = var_38693_end_0, end_mask = var_38693_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38693_cast_fp16")]; + tensor var_38700_begin_0 = const()[name = string("op_38700_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38700_end_0 = const()[name = string("op_38700_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38700_end_mask_0 = const()[name = string("op_38700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38700_cast_fp16 = slice_by_index(begin = var_38700_begin_0, end = var_38700_end_0, end_mask = var_38700_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38700_cast_fp16")]; + tensor var_38707_begin_0 = const()[name = string("op_38707_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38707_end_0 = const()[name = string("op_38707_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38707_end_mask_0 = const()[name = string("op_38707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38707_cast_fp16 = slice_by_index(begin = var_38707_begin_0, end = var_38707_end_0, end_mask = var_38707_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38707_cast_fp16")]; + tensor var_38714_begin_0 = const()[name = string("op_38714_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38714_end_0 = const()[name = string("op_38714_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38714_end_mask_0 = const()[name = string("op_38714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38714_cast_fp16 = slice_by_index(begin = var_38714_begin_0, end = var_38714_end_0, end_mask = var_38714_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38714_cast_fp16")]; + tensor var_38721_begin_0 = const()[name = string("op_38721_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38721_end_0 = const()[name = string("op_38721_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38721_end_mask_0 = const()[name = string("op_38721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38721_cast_fp16 = slice_by_index(begin = var_38721_begin_0, end = var_38721_end_0, end_mask = var_38721_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38721_cast_fp16")]; + tensor var_38728_begin_0 = const()[name = string("op_38728_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38728_end_0 = const()[name = string("op_38728_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38728_end_mask_0 = const()[name = string("op_38728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38728_cast_fp16 = slice_by_index(begin = var_38728_begin_0, end = var_38728_end_0, end_mask = var_38728_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38728_cast_fp16")]; + tensor var_38735_begin_0 = const()[name = string("op_38735_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38735_end_0 = const()[name = string("op_38735_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38735_end_mask_0 = const()[name = string("op_38735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38735_cast_fp16 = slice_by_index(begin = var_38735_begin_0, end = var_38735_end_0, end_mask = var_38735_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38735_cast_fp16")]; + tensor var_38742_begin_0 = const()[name = string("op_38742_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38742_end_0 = const()[name = string("op_38742_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38742_end_mask_0 = const()[name = string("op_38742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38742_cast_fp16 = slice_by_index(begin = var_38742_begin_0, end = var_38742_end_0, end_mask = var_38742_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38742_cast_fp16")]; + tensor var_38749_begin_0 = const()[name = string("op_38749_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38749_end_0 = const()[name = string("op_38749_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38749_end_mask_0 = const()[name = string("op_38749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38749_cast_fp16 = slice_by_index(begin = var_38749_begin_0, end = var_38749_end_0, end_mask = var_38749_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38749_cast_fp16")]; + tensor var_38756_begin_0 = const()[name = string("op_38756_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38756_end_0 = const()[name = string("op_38756_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38756_end_mask_0 = const()[name = string("op_38756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38756_cast_fp16 = slice_by_index(begin = var_38756_begin_0, end = var_38756_end_0, end_mask = var_38756_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38756_cast_fp16")]; + tensor var_38763_begin_0 = const()[name = string("op_38763_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38763_end_0 = const()[name = string("op_38763_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38763_end_mask_0 = const()[name = string("op_38763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38763_cast_fp16 = slice_by_index(begin = var_38763_begin_0, end = var_38763_end_0, end_mask = var_38763_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38763_cast_fp16")]; + tensor var_38770_begin_0 = const()[name = string("op_38770_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38770_end_0 = const()[name = string("op_38770_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38770_end_mask_0 = const()[name = string("op_38770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38770_cast_fp16 = slice_by_index(begin = var_38770_begin_0, end = var_38770_end_0, end_mask = var_38770_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38770_cast_fp16")]; + tensor var_38777_begin_0 = const()[name = string("op_38777_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38777_end_0 = const()[name = string("op_38777_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38777_end_mask_0 = const()[name = string("op_38777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38777_cast_fp16 = slice_by_index(begin = var_38777_begin_0, end = var_38777_end_0, end_mask = var_38777_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38777_cast_fp16")]; + tensor var_38784_begin_0 = const()[name = string("op_38784_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38784_end_0 = const()[name = string("op_38784_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38784_end_mask_0 = const()[name = string("op_38784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38784_cast_fp16 = slice_by_index(begin = var_38784_begin_0, end = var_38784_end_0, end_mask = var_38784_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38784_cast_fp16")]; + tensor var_38791_begin_0 = const()[name = string("op_38791_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38791_end_0 = const()[name = string("op_38791_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38791_end_mask_0 = const()[name = string("op_38791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38791_cast_fp16 = slice_by_index(begin = var_38791_begin_0, end = var_38791_end_0, end_mask = var_38791_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38791_cast_fp16")]; + tensor var_38798_begin_0 = const()[name = string("op_38798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38798_end_0 = const()[name = string("op_38798_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38798_end_mask_0 = const()[name = string("op_38798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38798_cast_fp16 = slice_by_index(begin = var_38798_begin_0, end = var_38798_end_0, end_mask = var_38798_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38798_cast_fp16")]; + tensor var_38805_begin_0 = const()[name = string("op_38805_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38805_end_0 = const()[name = string("op_38805_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38805_end_mask_0 = const()[name = string("op_38805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38805_cast_fp16 = slice_by_index(begin = var_38805_begin_0, end = var_38805_end_0, end_mask = var_38805_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38805_cast_fp16")]; + tensor var_38812_begin_0 = const()[name = string("op_38812_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38812_end_0 = const()[name = string("op_38812_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38812_end_mask_0 = const()[name = string("op_38812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38812_cast_fp16 = slice_by_index(begin = var_38812_begin_0, end = var_38812_end_0, end_mask = var_38812_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38812_cast_fp16")]; + tensor var_38819_begin_0 = const()[name = string("op_38819_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38819_end_0 = const()[name = string("op_38819_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38819_end_mask_0 = const()[name = string("op_38819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38819_cast_fp16 = slice_by_index(begin = var_38819_begin_0, end = var_38819_end_0, end_mask = var_38819_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38819_cast_fp16")]; + tensor var_38826_begin_0 = const()[name = string("op_38826_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38826_end_0 = const()[name = string("op_38826_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38826_end_mask_0 = const()[name = string("op_38826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38826_cast_fp16 = slice_by_index(begin = var_38826_begin_0, end = var_38826_end_0, end_mask = var_38826_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38826_cast_fp16")]; + tensor var_38833_begin_0 = const()[name = string("op_38833_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38833_end_0 = const()[name = string("op_38833_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38833_end_mask_0 = const()[name = string("op_38833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38833_cast_fp16 = slice_by_index(begin = var_38833_begin_0, end = var_38833_end_0, end_mask = var_38833_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38833_cast_fp16")]; + tensor var_38840_begin_0 = const()[name = string("op_38840_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38840_end_0 = const()[name = string("op_38840_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38840_end_mask_0 = const()[name = string("op_38840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38840_cast_fp16 = slice_by_index(begin = var_38840_begin_0, end = var_38840_end_0, end_mask = var_38840_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38840_cast_fp16")]; + tensor var_38847_begin_0 = const()[name = string("op_38847_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38847_end_0 = const()[name = string("op_38847_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38847_end_mask_0 = const()[name = string("op_38847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38847_cast_fp16 = slice_by_index(begin = var_38847_begin_0, end = var_38847_end_0, end_mask = var_38847_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38847_cast_fp16")]; + tensor var_38854_begin_0 = const()[name = string("op_38854_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38854_end_0 = const()[name = string("op_38854_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38854_end_mask_0 = const()[name = string("op_38854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38854_cast_fp16 = slice_by_index(begin = var_38854_begin_0, end = var_38854_end_0, end_mask = var_38854_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38854_cast_fp16")]; + tensor var_38861_begin_0 = const()[name = string("op_38861_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38861_end_0 = const()[name = string("op_38861_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38861_end_mask_0 = const()[name = string("op_38861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38861_cast_fp16 = slice_by_index(begin = var_38861_begin_0, end = var_38861_end_0, end_mask = var_38861_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38861_cast_fp16")]; + tensor var_38868_begin_0 = const()[name = string("op_38868_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38868_end_0 = const()[name = string("op_38868_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38868_end_mask_0 = const()[name = string("op_38868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38868_cast_fp16 = slice_by_index(begin = var_38868_begin_0, end = var_38868_end_0, end_mask = var_38868_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38868_cast_fp16")]; + tensor var_38875_begin_0 = const()[name = string("op_38875_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38875_end_0 = const()[name = string("op_38875_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38875_end_mask_0 = const()[name = string("op_38875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38875_cast_fp16 = slice_by_index(begin = var_38875_begin_0, end = var_38875_end_0, end_mask = var_38875_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38875_cast_fp16")]; + tensor var_38882_begin_0 = const()[name = string("op_38882_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38882_end_0 = const()[name = string("op_38882_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38882_end_mask_0 = const()[name = string("op_38882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38882_cast_fp16 = slice_by_index(begin = var_38882_begin_0, end = var_38882_end_0, end_mask = var_38882_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38882_cast_fp16")]; + tensor var_38889_begin_0 = const()[name = string("op_38889_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38889_end_0 = const()[name = string("op_38889_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38889_end_mask_0 = const()[name = string("op_38889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38889_cast_fp16 = slice_by_index(begin = var_38889_begin_0, end = var_38889_end_0, end_mask = var_38889_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38889_cast_fp16")]; + tensor var_38896_begin_0 = const()[name = string("op_38896_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38896_end_0 = const()[name = string("op_38896_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38896_end_mask_0 = const()[name = string("op_38896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38896_cast_fp16 = slice_by_index(begin = var_38896_begin_0, end = var_38896_end_0, end_mask = var_38896_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38896_cast_fp16")]; + tensor var_38903_begin_0 = const()[name = string("op_38903_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38903_end_0 = const()[name = string("op_38903_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38903_end_mask_0 = const()[name = string("op_38903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38903_cast_fp16 = slice_by_index(begin = var_38903_begin_0, end = var_38903_end_0, end_mask = var_38903_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38903_cast_fp16")]; + tensor var_38910_begin_0 = const()[name = string("op_38910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38910_end_0 = const()[name = string("op_38910_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38910_end_mask_0 = const()[name = string("op_38910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38910_cast_fp16 = slice_by_index(begin = var_38910_begin_0, end = var_38910_end_0, end_mask = var_38910_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38910_cast_fp16")]; + tensor var_38917_begin_0 = const()[name = string("op_38917_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38917_end_0 = const()[name = string("op_38917_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38917_end_mask_0 = const()[name = string("op_38917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38917_cast_fp16 = slice_by_index(begin = var_38917_begin_0, end = var_38917_end_0, end_mask = var_38917_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38917_cast_fp16")]; + tensor var_38924_begin_0 = const()[name = string("op_38924_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38924_end_0 = const()[name = string("op_38924_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38924_end_mask_0 = const()[name = string("op_38924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38924_cast_fp16 = slice_by_index(begin = var_38924_begin_0, end = var_38924_end_0, end_mask = var_38924_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38924_cast_fp16")]; + tensor var_38931_begin_0 = const()[name = string("op_38931_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38931_end_0 = const()[name = string("op_38931_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38931_end_mask_0 = const()[name = string("op_38931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38931_cast_fp16 = slice_by_index(begin = var_38931_begin_0, end = var_38931_end_0, end_mask = var_38931_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38931_cast_fp16")]; + tensor var_38938_begin_0 = const()[name = string("op_38938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38938_end_0 = const()[name = string("op_38938_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38938_end_mask_0 = const()[name = string("op_38938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38938_cast_fp16 = slice_by_index(begin = var_38938_begin_0, end = var_38938_end_0, end_mask = var_38938_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38938_cast_fp16")]; + tensor var_38945_begin_0 = const()[name = string("op_38945_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38945_end_0 = const()[name = string("op_38945_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38945_end_mask_0 = const()[name = string("op_38945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38945_cast_fp16 = slice_by_index(begin = var_38945_begin_0, end = var_38945_end_0, end_mask = var_38945_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38945_cast_fp16")]; + tensor var_38952_begin_0 = const()[name = string("op_38952_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38952_end_0 = const()[name = string("op_38952_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38952_end_mask_0 = const()[name = string("op_38952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38952_cast_fp16 = slice_by_index(begin = var_38952_begin_0, end = var_38952_end_0, end_mask = var_38952_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38952_cast_fp16")]; + tensor var_38959_begin_0 = const()[name = string("op_38959_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38959_end_0 = const()[name = string("op_38959_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38959_end_mask_0 = const()[name = string("op_38959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38959_cast_fp16 = slice_by_index(begin = var_38959_begin_0, end = var_38959_end_0, end_mask = var_38959_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38959_cast_fp16")]; + tensor var_38966_begin_0 = const()[name = string("op_38966_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38966_end_0 = const()[name = string("op_38966_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38966_end_mask_0 = const()[name = string("op_38966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38966_cast_fp16 = slice_by_index(begin = var_38966_begin_0, end = var_38966_end_0, end_mask = var_38966_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38966_cast_fp16")]; + tensor var_38973_begin_0 = const()[name = string("op_38973_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38973_end_0 = const()[name = string("op_38973_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38973_end_mask_0 = const()[name = string("op_38973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38973_cast_fp16 = slice_by_index(begin = var_38973_begin_0, end = var_38973_end_0, end_mask = var_38973_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38973_cast_fp16")]; + tensor var_38980_begin_0 = const()[name = string("op_38980_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38980_end_0 = const()[name = string("op_38980_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38980_end_mask_0 = const()[name = string("op_38980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38980_cast_fp16 = slice_by_index(begin = var_38980_begin_0, end = var_38980_end_0, end_mask = var_38980_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38980_cast_fp16")]; + tensor var_38987_begin_0 = const()[name = string("op_38987_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38987_end_0 = const()[name = string("op_38987_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38987_end_mask_0 = const()[name = string("op_38987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38987_cast_fp16 = slice_by_index(begin = var_38987_begin_0, end = var_38987_end_0, end_mask = var_38987_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38987_cast_fp16")]; + tensor var_38994_begin_0 = const()[name = string("op_38994_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38994_end_0 = const()[name = string("op_38994_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38994_end_mask_0 = const()[name = string("op_38994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38994_cast_fp16 = slice_by_index(begin = var_38994_begin_0, end = var_38994_end_0, end_mask = var_38994_end_mask_0, x = var_38501_cast_fp16)[name = string("op_38994_cast_fp16")]; + tensor var_39001_begin_0 = const()[name = string("op_39001_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39001_end_0 = const()[name = string("op_39001_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39001_end_mask_0 = const()[name = string("op_39001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39001_cast_fp16 = slice_by_index(begin = var_39001_begin_0, end = var_39001_end_0, end_mask = var_39001_end_mask_0, x = var_38501_cast_fp16)[name = string("op_39001_cast_fp16")]; + tensor var_39008_begin_0 = const()[name = string("op_39008_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39008_end_0 = const()[name = string("op_39008_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39008_end_mask_0 = const()[name = string("op_39008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39008_cast_fp16 = slice_by_index(begin = var_39008_begin_0, end = var_39008_end_0, end_mask = var_39008_end_mask_0, x = var_38501_cast_fp16)[name = string("op_39008_cast_fp16")]; + tensor var_39015_begin_0 = const()[name = string("op_39015_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39015_end_0 = const()[name = string("op_39015_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39015_end_mask_0 = const()[name = string("op_39015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39015_cast_fp16 = slice_by_index(begin = var_39015_begin_0, end = var_39015_end_0, end_mask = var_39015_end_mask_0, x = var_38501_cast_fp16)[name = string("op_39015_cast_fp16")]; + tensor var_39022_begin_0 = const()[name = string("op_39022_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39022_end_0 = const()[name = string("op_39022_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39022_end_mask_0 = const()[name = string("op_39022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39022_cast_fp16 = slice_by_index(begin = var_39022_begin_0, end = var_39022_end_0, end_mask = var_39022_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39022_cast_fp16")]; + tensor var_39029_begin_0 = const()[name = string("op_39029_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39029_end_0 = const()[name = string("op_39029_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39029_end_mask_0 = const()[name = string("op_39029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39029_cast_fp16 = slice_by_index(begin = var_39029_begin_0, end = var_39029_end_0, end_mask = var_39029_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39029_cast_fp16")]; + tensor var_39036_begin_0 = const()[name = string("op_39036_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39036_end_0 = const()[name = string("op_39036_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39036_end_mask_0 = const()[name = string("op_39036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39036_cast_fp16 = slice_by_index(begin = var_39036_begin_0, end = var_39036_end_0, end_mask = var_39036_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39036_cast_fp16")]; + tensor var_39043_begin_0 = const()[name = string("op_39043_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39043_end_0 = const()[name = string("op_39043_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39043_end_mask_0 = const()[name = string("op_39043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39043_cast_fp16 = slice_by_index(begin = var_39043_begin_0, end = var_39043_end_0, end_mask = var_39043_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39043_cast_fp16")]; + tensor var_39050_begin_0 = const()[name = string("op_39050_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39050_end_0 = const()[name = string("op_39050_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39050_end_mask_0 = const()[name = string("op_39050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39050_cast_fp16 = slice_by_index(begin = var_39050_begin_0, end = var_39050_end_0, end_mask = var_39050_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39050_cast_fp16")]; + tensor var_39057_begin_0 = const()[name = string("op_39057_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39057_end_0 = const()[name = string("op_39057_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39057_end_mask_0 = const()[name = string("op_39057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39057_cast_fp16 = slice_by_index(begin = var_39057_begin_0, end = var_39057_end_0, end_mask = var_39057_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39057_cast_fp16")]; + tensor var_39064_begin_0 = const()[name = string("op_39064_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39064_end_0 = const()[name = string("op_39064_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39064_end_mask_0 = const()[name = string("op_39064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39064_cast_fp16 = slice_by_index(begin = var_39064_begin_0, end = var_39064_end_0, end_mask = var_39064_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39064_cast_fp16")]; + tensor var_39071_begin_0 = const()[name = string("op_39071_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39071_end_0 = const()[name = string("op_39071_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39071_end_mask_0 = const()[name = string("op_39071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39071_cast_fp16 = slice_by_index(begin = var_39071_begin_0, end = var_39071_end_0, end_mask = var_39071_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39071_cast_fp16")]; + tensor k_51_perm_0 = const()[name = string("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_39076_begin_0 = const()[name = string("op_39076_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39076_end_0 = const()[name = string("op_39076_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_39076_end_mask_0 = const()[name = string("op_39076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = key_51_cast_fp16)[name = string("transpose_6")]; + tensor var_39076_cast_fp16 = slice_by_index(begin = var_39076_begin_0, end = var_39076_end_0, end_mask = var_39076_end_mask_0, x = k_51_cast_fp16)[name = string("op_39076_cast_fp16")]; + tensor var_39080_begin_0 = const()[name = string("op_39080_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_39080_end_0 = const()[name = string("op_39080_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_39080_end_mask_0 = const()[name = string("op_39080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39080_cast_fp16 = slice_by_index(begin = var_39080_begin_0, end = var_39080_end_0, end_mask = var_39080_end_mask_0, x = k_51_cast_fp16)[name = string("op_39080_cast_fp16")]; + tensor var_39084_begin_0 = const()[name = string("op_39084_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_39084_end_0 = const()[name = string("op_39084_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_39084_end_mask_0 = const()[name = string("op_39084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39084_cast_fp16 = slice_by_index(begin = var_39084_begin_0, end = var_39084_end_0, end_mask = var_39084_end_mask_0, x = k_51_cast_fp16)[name = string("op_39084_cast_fp16")]; + tensor var_39088_begin_0 = const()[name = string("op_39088_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_39088_end_0 = const()[name = string("op_39088_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_39088_end_mask_0 = const()[name = string("op_39088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39088_cast_fp16 = slice_by_index(begin = var_39088_begin_0, end = var_39088_end_0, end_mask = var_39088_end_mask_0, x = k_51_cast_fp16)[name = string("op_39088_cast_fp16")]; + tensor var_39092_begin_0 = const()[name = string("op_39092_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_39092_end_0 = const()[name = string("op_39092_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_39092_end_mask_0 = const()[name = string("op_39092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39092_cast_fp16 = slice_by_index(begin = var_39092_begin_0, end = var_39092_end_0, end_mask = var_39092_end_mask_0, x = k_51_cast_fp16)[name = string("op_39092_cast_fp16")]; + tensor var_39096_begin_0 = const()[name = string("op_39096_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_39096_end_0 = const()[name = string("op_39096_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_39096_end_mask_0 = const()[name = string("op_39096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39096_cast_fp16 = slice_by_index(begin = var_39096_begin_0, end = var_39096_end_0, end_mask = var_39096_end_mask_0, x = k_51_cast_fp16)[name = string("op_39096_cast_fp16")]; + tensor var_39100_begin_0 = const()[name = string("op_39100_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_39100_end_0 = const()[name = string("op_39100_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_39100_end_mask_0 = const()[name = string("op_39100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39100_cast_fp16 = slice_by_index(begin = var_39100_begin_0, end = var_39100_end_0, end_mask = var_39100_end_mask_0, x = k_51_cast_fp16)[name = string("op_39100_cast_fp16")]; + tensor var_39104_begin_0 = const()[name = string("op_39104_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_39104_end_0 = const()[name = string("op_39104_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_39104_end_mask_0 = const()[name = string("op_39104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39104_cast_fp16 = slice_by_index(begin = var_39104_begin_0, end = var_39104_end_0, end_mask = var_39104_end_mask_0, x = k_51_cast_fp16)[name = string("op_39104_cast_fp16")]; + tensor var_39108_begin_0 = const()[name = string("op_39108_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_39108_end_0 = const()[name = string("op_39108_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_39108_end_mask_0 = const()[name = string("op_39108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39108_cast_fp16 = slice_by_index(begin = var_39108_begin_0, end = var_39108_end_0, end_mask = var_39108_end_mask_0, x = k_51_cast_fp16)[name = string("op_39108_cast_fp16")]; + tensor var_39112_begin_0 = const()[name = string("op_39112_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_39112_end_0 = const()[name = string("op_39112_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_39112_end_mask_0 = const()[name = string("op_39112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39112_cast_fp16 = slice_by_index(begin = var_39112_begin_0, end = var_39112_end_0, end_mask = var_39112_end_mask_0, x = k_51_cast_fp16)[name = string("op_39112_cast_fp16")]; + tensor var_39116_begin_0 = const()[name = string("op_39116_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_39116_end_0 = const()[name = string("op_39116_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_39116_end_mask_0 = const()[name = string("op_39116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39116_cast_fp16 = slice_by_index(begin = var_39116_begin_0, end = var_39116_end_0, end_mask = var_39116_end_mask_0, x = k_51_cast_fp16)[name = string("op_39116_cast_fp16")]; + tensor var_39120_begin_0 = const()[name = string("op_39120_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_39120_end_0 = const()[name = string("op_39120_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_39120_end_mask_0 = const()[name = string("op_39120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39120_cast_fp16 = slice_by_index(begin = var_39120_begin_0, end = var_39120_end_0, end_mask = var_39120_end_mask_0, x = k_51_cast_fp16)[name = string("op_39120_cast_fp16")]; + tensor var_39124_begin_0 = const()[name = string("op_39124_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_39124_end_0 = const()[name = string("op_39124_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_39124_end_mask_0 = const()[name = string("op_39124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39124_cast_fp16 = slice_by_index(begin = var_39124_begin_0, end = var_39124_end_0, end_mask = var_39124_end_mask_0, x = k_51_cast_fp16)[name = string("op_39124_cast_fp16")]; + tensor var_39128_begin_0 = const()[name = string("op_39128_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_39128_end_0 = const()[name = string("op_39128_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_39128_end_mask_0 = const()[name = string("op_39128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39128_cast_fp16 = slice_by_index(begin = var_39128_begin_0, end = var_39128_end_0, end_mask = var_39128_end_mask_0, x = k_51_cast_fp16)[name = string("op_39128_cast_fp16")]; + tensor var_39132_begin_0 = const()[name = string("op_39132_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_39132_end_0 = const()[name = string("op_39132_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_39132_end_mask_0 = const()[name = string("op_39132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39132_cast_fp16 = slice_by_index(begin = var_39132_begin_0, end = var_39132_end_0, end_mask = var_39132_end_mask_0, x = k_51_cast_fp16)[name = string("op_39132_cast_fp16")]; + tensor var_39136_begin_0 = const()[name = string("op_39136_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_39136_end_0 = const()[name = string("op_39136_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_39136_end_mask_0 = const()[name = string("op_39136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39136_cast_fp16 = slice_by_index(begin = var_39136_begin_0, end = var_39136_end_0, end_mask = var_39136_end_mask_0, x = k_51_cast_fp16)[name = string("op_39136_cast_fp16")]; + tensor var_39140_begin_0 = const()[name = string("op_39140_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_39140_end_0 = const()[name = string("op_39140_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_39140_end_mask_0 = const()[name = string("op_39140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39140_cast_fp16 = slice_by_index(begin = var_39140_begin_0, end = var_39140_end_0, end_mask = var_39140_end_mask_0, x = k_51_cast_fp16)[name = string("op_39140_cast_fp16")]; + tensor var_39144_begin_0 = const()[name = string("op_39144_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_39144_end_0 = const()[name = string("op_39144_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_39144_end_mask_0 = const()[name = string("op_39144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39144_cast_fp16 = slice_by_index(begin = var_39144_begin_0, end = var_39144_end_0, end_mask = var_39144_end_mask_0, x = k_51_cast_fp16)[name = string("op_39144_cast_fp16")]; + tensor var_39148_begin_0 = const()[name = string("op_39148_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_39148_end_0 = const()[name = string("op_39148_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_39148_end_mask_0 = const()[name = string("op_39148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39148_cast_fp16 = slice_by_index(begin = var_39148_begin_0, end = var_39148_end_0, end_mask = var_39148_end_mask_0, x = k_51_cast_fp16)[name = string("op_39148_cast_fp16")]; + tensor var_39152_begin_0 = const()[name = string("op_39152_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_39152_end_0 = const()[name = string("op_39152_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_39152_end_mask_0 = const()[name = string("op_39152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39152_cast_fp16 = slice_by_index(begin = var_39152_begin_0, end = var_39152_end_0, end_mask = var_39152_end_mask_0, x = k_51_cast_fp16)[name = string("op_39152_cast_fp16")]; + tensor var_39154_begin_0 = const()[name = string("op_39154_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39154_end_0 = const()[name = string("op_39154_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39154_end_mask_0 = const()[name = string("op_39154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39154_cast_fp16 = slice_by_index(begin = var_39154_begin_0, end = var_39154_end_0, end_mask = var_39154_end_mask_0, x = value_51_cast_fp16)[name = string("op_39154_cast_fp16")]; + tensor var_39158_begin_0 = const()[name = string("op_39158_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39158_end_0 = const()[name = string("op_39158_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_39158_end_mask_0 = const()[name = string("op_39158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39158_cast_fp16 = slice_by_index(begin = var_39158_begin_0, end = var_39158_end_0, end_mask = var_39158_end_mask_0, x = value_51_cast_fp16)[name = string("op_39158_cast_fp16")]; + tensor var_39162_begin_0 = const()[name = string("op_39162_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39162_end_0 = const()[name = string("op_39162_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_39162_end_mask_0 = const()[name = string("op_39162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39162_cast_fp16 = slice_by_index(begin = var_39162_begin_0, end = var_39162_end_0, end_mask = var_39162_end_mask_0, x = value_51_cast_fp16)[name = string("op_39162_cast_fp16")]; + tensor var_39166_begin_0 = const()[name = string("op_39166_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39166_end_0 = const()[name = string("op_39166_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_39166_end_mask_0 = const()[name = string("op_39166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39166_cast_fp16 = slice_by_index(begin = var_39166_begin_0, end = var_39166_end_0, end_mask = var_39166_end_mask_0, x = value_51_cast_fp16)[name = string("op_39166_cast_fp16")]; + tensor var_39170_begin_0 = const()[name = string("op_39170_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39170_end_0 = const()[name = string("op_39170_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_39170_end_mask_0 = const()[name = string("op_39170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39170_cast_fp16 = slice_by_index(begin = var_39170_begin_0, end = var_39170_end_0, end_mask = var_39170_end_mask_0, x = value_51_cast_fp16)[name = string("op_39170_cast_fp16")]; + tensor var_39174_begin_0 = const()[name = string("op_39174_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39174_end_0 = const()[name = string("op_39174_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_39174_end_mask_0 = const()[name = string("op_39174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39174_cast_fp16 = slice_by_index(begin = var_39174_begin_0, end = var_39174_end_0, end_mask = var_39174_end_mask_0, x = value_51_cast_fp16)[name = string("op_39174_cast_fp16")]; + tensor var_39178_begin_0 = const()[name = string("op_39178_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39178_end_0 = const()[name = string("op_39178_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_39178_end_mask_0 = const()[name = string("op_39178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39178_cast_fp16 = slice_by_index(begin = var_39178_begin_0, end = var_39178_end_0, end_mask = var_39178_end_mask_0, x = value_51_cast_fp16)[name = string("op_39178_cast_fp16")]; + tensor var_39182_begin_0 = const()[name = string("op_39182_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39182_end_0 = const()[name = string("op_39182_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_39182_end_mask_0 = const()[name = string("op_39182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39182_cast_fp16 = slice_by_index(begin = var_39182_begin_0, end = var_39182_end_0, end_mask = var_39182_end_mask_0, x = value_51_cast_fp16)[name = string("op_39182_cast_fp16")]; + tensor var_39186_begin_0 = const()[name = string("op_39186_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39186_end_0 = const()[name = string("op_39186_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_39186_end_mask_0 = const()[name = string("op_39186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39186_cast_fp16 = slice_by_index(begin = var_39186_begin_0, end = var_39186_end_0, end_mask = var_39186_end_mask_0, x = value_51_cast_fp16)[name = string("op_39186_cast_fp16")]; + tensor var_39190_begin_0 = const()[name = string("op_39190_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39190_end_0 = const()[name = string("op_39190_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_39190_end_mask_0 = const()[name = string("op_39190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39190_cast_fp16 = slice_by_index(begin = var_39190_begin_0, end = var_39190_end_0, end_mask = var_39190_end_mask_0, x = value_51_cast_fp16)[name = string("op_39190_cast_fp16")]; + tensor var_39194_begin_0 = const()[name = string("op_39194_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39194_end_0 = const()[name = string("op_39194_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_39194_end_mask_0 = const()[name = string("op_39194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39194_cast_fp16 = slice_by_index(begin = var_39194_begin_0, end = var_39194_end_0, end_mask = var_39194_end_mask_0, x = value_51_cast_fp16)[name = string("op_39194_cast_fp16")]; + tensor var_39198_begin_0 = const()[name = string("op_39198_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39198_end_0 = const()[name = string("op_39198_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_39198_end_mask_0 = const()[name = string("op_39198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39198_cast_fp16 = slice_by_index(begin = var_39198_begin_0, end = var_39198_end_0, end_mask = var_39198_end_mask_0, x = value_51_cast_fp16)[name = string("op_39198_cast_fp16")]; + tensor var_39202_begin_0 = const()[name = string("op_39202_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39202_end_0 = const()[name = string("op_39202_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_39202_end_mask_0 = const()[name = string("op_39202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39202_cast_fp16 = slice_by_index(begin = var_39202_begin_0, end = var_39202_end_0, end_mask = var_39202_end_mask_0, x = value_51_cast_fp16)[name = string("op_39202_cast_fp16")]; + tensor var_39206_begin_0 = const()[name = string("op_39206_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39206_end_0 = const()[name = string("op_39206_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_39206_end_mask_0 = const()[name = string("op_39206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39206_cast_fp16 = slice_by_index(begin = var_39206_begin_0, end = var_39206_end_0, end_mask = var_39206_end_mask_0, x = value_51_cast_fp16)[name = string("op_39206_cast_fp16")]; + tensor var_39210_begin_0 = const()[name = string("op_39210_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39210_end_0 = const()[name = string("op_39210_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_39210_end_mask_0 = const()[name = string("op_39210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39210_cast_fp16 = slice_by_index(begin = var_39210_begin_0, end = var_39210_end_0, end_mask = var_39210_end_mask_0, x = value_51_cast_fp16)[name = string("op_39210_cast_fp16")]; + tensor var_39214_begin_0 = const()[name = string("op_39214_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39214_end_0 = const()[name = string("op_39214_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_39214_end_mask_0 = const()[name = string("op_39214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39214_cast_fp16 = slice_by_index(begin = var_39214_begin_0, end = var_39214_end_0, end_mask = var_39214_end_mask_0, x = value_51_cast_fp16)[name = string("op_39214_cast_fp16")]; + tensor var_39218_begin_0 = const()[name = string("op_39218_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39218_end_0 = const()[name = string("op_39218_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_39218_end_mask_0 = const()[name = string("op_39218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39218_cast_fp16 = slice_by_index(begin = var_39218_begin_0, end = var_39218_end_0, end_mask = var_39218_end_mask_0, x = value_51_cast_fp16)[name = string("op_39218_cast_fp16")]; + tensor var_39222_begin_0 = const()[name = string("op_39222_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39222_end_0 = const()[name = string("op_39222_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_39222_end_mask_0 = const()[name = string("op_39222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39222_cast_fp16 = slice_by_index(begin = var_39222_begin_0, end = var_39222_end_0, end_mask = var_39222_end_mask_0, x = value_51_cast_fp16)[name = string("op_39222_cast_fp16")]; + tensor var_39226_begin_0 = const()[name = string("op_39226_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39226_end_0 = const()[name = string("op_39226_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_39226_end_mask_0 = const()[name = string("op_39226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39226_cast_fp16 = slice_by_index(begin = var_39226_begin_0, end = var_39226_end_0, end_mask = var_39226_end_mask_0, x = value_51_cast_fp16)[name = string("op_39226_cast_fp16")]; + tensor var_39230_begin_0 = const()[name = string("op_39230_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39230_end_0 = const()[name = string("op_39230_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_39230_end_mask_0 = const()[name = string("op_39230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39230_cast_fp16 = slice_by_index(begin = var_39230_begin_0, end = var_39230_end_0, end_mask = var_39230_end_mask_0, x = value_51_cast_fp16)[name = string("op_39230_cast_fp16")]; + string _SplitHeadsQ__mh_w_4001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4001_equation_0, values = (var_39076_cast_fp16, var_38518_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4001_cast_fp16")]; + string _SplitHeadsQ__mh_w_4003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4003_equation_0, values = (var_39076_cast_fp16, var_38525_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4003_cast_fp16")]; + string _SplitHeadsQ__mh_w_4005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4005_equation_0, values = (var_39076_cast_fp16, var_38532_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4005_cast_fp16")]; + string _SplitHeadsQ__mh_w_4007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4007_equation_0, values = (var_39076_cast_fp16, var_38539_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4007_cast_fp16")]; + string _SplitHeadsQ__mh_w_4009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4009_equation_0, values = (var_39080_cast_fp16, var_38546_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4009_cast_fp16")]; + string _SplitHeadsQ__mh_w_4011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4011_equation_0, values = (var_39080_cast_fp16, var_38553_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4011_cast_fp16")]; + string _SplitHeadsQ__mh_w_4013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4013_equation_0, values = (var_39080_cast_fp16, var_38560_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4013_cast_fp16")]; + string _SplitHeadsQ__mh_w_4015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4015_equation_0, values = (var_39080_cast_fp16, var_38567_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4015_cast_fp16")]; + string _SplitHeadsQ__mh_w_4017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4017_equation_0, values = (var_39084_cast_fp16, var_38574_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4017_cast_fp16")]; + string _SplitHeadsQ__mh_w_4019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4019_equation_0, values = (var_39084_cast_fp16, var_38581_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4019_cast_fp16")]; + string _SplitHeadsQ__mh_w_4021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4021_equation_0, values = (var_39084_cast_fp16, var_38588_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4021_cast_fp16")]; + string _SplitHeadsQ__mh_w_4023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4023_equation_0, values = (var_39084_cast_fp16, var_38595_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4023_cast_fp16")]; + string _SplitHeadsQ__mh_w_4025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4025_equation_0, values = (var_39088_cast_fp16, var_38602_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4025_cast_fp16")]; + string _SplitHeadsQ__mh_w_4027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4027_equation_0, values = (var_39088_cast_fp16, var_38609_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4027_cast_fp16")]; + string _SplitHeadsQ__mh_w_4029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4029_equation_0, values = (var_39088_cast_fp16, var_38616_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4029_cast_fp16")]; + string _SplitHeadsQ__mh_w_4031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4031_equation_0, values = (var_39088_cast_fp16, var_38623_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4031_cast_fp16")]; + string _SplitHeadsQ__mh_w_4033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4033_equation_0, values = (var_39092_cast_fp16, var_38630_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4033_cast_fp16")]; + string _SplitHeadsQ__mh_w_4035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4035_equation_0, values = (var_39092_cast_fp16, var_38637_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4035_cast_fp16")]; + string _SplitHeadsQ__mh_w_4037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4037_equation_0, values = (var_39092_cast_fp16, var_38644_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4037_cast_fp16")]; + string _SplitHeadsQ__mh_w_4039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4039_equation_0, values = (var_39092_cast_fp16, var_38651_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4039_cast_fp16")]; + string _SplitHeadsQ__mh_w_4041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4041_equation_0, values = (var_39096_cast_fp16, var_38658_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4041_cast_fp16")]; + string _SplitHeadsQ__mh_w_4043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4043_equation_0, values = (var_39096_cast_fp16, var_38665_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4043_cast_fp16")]; + string _SplitHeadsQ__mh_w_4045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4045_equation_0, values = (var_39096_cast_fp16, var_38672_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4045_cast_fp16")]; + string _SplitHeadsQ__mh_w_4047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4047_equation_0, values = (var_39096_cast_fp16, var_38679_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4047_cast_fp16")]; + string _SplitHeadsQ__mh_w_4049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4049_equation_0, values = (var_39100_cast_fp16, var_38686_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4049_cast_fp16")]; + string _SplitHeadsQ__mh_w_4051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4051_equation_0, values = (var_39100_cast_fp16, var_38693_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4051_cast_fp16")]; + string _SplitHeadsQ__mh_w_4053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4053_equation_0, values = (var_39100_cast_fp16, var_38700_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4053_cast_fp16")]; + string _SplitHeadsQ__mh_w_4055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4055_equation_0, values = (var_39100_cast_fp16, var_38707_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4055_cast_fp16")]; + string _SplitHeadsQ__mh_w_4057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4057_equation_0, values = (var_39104_cast_fp16, var_38714_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4057_cast_fp16")]; + string _SplitHeadsQ__mh_w_4059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4059_equation_0, values = (var_39104_cast_fp16, var_38721_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4059_cast_fp16")]; + string _SplitHeadsQ__mh_w_4061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4061_equation_0, values = (var_39104_cast_fp16, var_38728_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4061_cast_fp16")]; + string _SplitHeadsQ__mh_w_4063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4063_equation_0, values = (var_39104_cast_fp16, var_38735_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4063_cast_fp16")]; + string _SplitHeadsQ__mh_w_4065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4065_equation_0, values = (var_39108_cast_fp16, var_38742_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4065_cast_fp16")]; + string _SplitHeadsQ__mh_w_4067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4067_equation_0, values = (var_39108_cast_fp16, var_38749_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4067_cast_fp16")]; + string _SplitHeadsQ__mh_w_4069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4069_equation_0, values = (var_39108_cast_fp16, var_38756_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4069_cast_fp16")]; + string _SplitHeadsQ__mh_w_4071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4071_equation_0, values = (var_39108_cast_fp16, var_38763_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4071_cast_fp16")]; + string _SplitHeadsQ__mh_w_4073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4073_equation_0, values = (var_39112_cast_fp16, var_38770_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4073_cast_fp16")]; + string _SplitHeadsQ__mh_w_4075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4075_equation_0, values = (var_39112_cast_fp16, var_38777_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4075_cast_fp16")]; + string _SplitHeadsQ__mh_w_4077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4077_equation_0, values = (var_39112_cast_fp16, var_38784_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4077_cast_fp16")]; + string _SplitHeadsQ__mh_w_4079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4079_equation_0, values = (var_39112_cast_fp16, var_38791_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4079_cast_fp16")]; + string _SplitHeadsQ__mh_w_4081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4081_equation_0, values = (var_39116_cast_fp16, var_38798_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4081_cast_fp16")]; + string _SplitHeadsQ__mh_w_4083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4083_equation_0, values = (var_39116_cast_fp16, var_38805_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4083_cast_fp16")]; + string _SplitHeadsQ__mh_w_4085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4085_equation_0, values = (var_39116_cast_fp16, var_38812_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4085_cast_fp16")]; + string _SplitHeadsQ__mh_w_4087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4087_equation_0, values = (var_39116_cast_fp16, var_38819_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4087_cast_fp16")]; + string _SplitHeadsQ__mh_w_4089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4089_equation_0, values = (var_39120_cast_fp16, var_38826_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4089_cast_fp16")]; + string _SplitHeadsQ__mh_w_4091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4091_equation_0, values = (var_39120_cast_fp16, var_38833_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4091_cast_fp16")]; + string _SplitHeadsQ__mh_w_4093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4093_equation_0, values = (var_39120_cast_fp16, var_38840_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4093_cast_fp16")]; + string _SplitHeadsQ__mh_w_4095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4095_equation_0, values = (var_39120_cast_fp16, var_38847_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4095_cast_fp16")]; + string _SplitHeadsQ__mh_w_4097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4097_equation_0, values = (var_39124_cast_fp16, var_38854_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4097_cast_fp16")]; + string _SplitHeadsQ__mh_w_4099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4099_equation_0, values = (var_39124_cast_fp16, var_38861_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4099_cast_fp16")]; + string _SplitHeadsQ__mh_w_4101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4101_equation_0, values = (var_39124_cast_fp16, var_38868_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4101_cast_fp16")]; + string _SplitHeadsQ__mh_w_4103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4103_equation_0, values = (var_39124_cast_fp16, var_38875_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4103_cast_fp16")]; + string _SplitHeadsQ__mh_w_4105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4105_equation_0, values = (var_39128_cast_fp16, var_38882_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4105_cast_fp16")]; + string _SplitHeadsQ__mh_w_4107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4107_equation_0, values = (var_39128_cast_fp16, var_38889_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4107_cast_fp16")]; + string _SplitHeadsQ__mh_w_4109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4109_equation_0, values = (var_39128_cast_fp16, var_38896_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4109_cast_fp16")]; + string _SplitHeadsQ__mh_w_4111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4111_equation_0, values = (var_39128_cast_fp16, var_38903_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4111_cast_fp16")]; + string _SplitHeadsQ__mh_w_4113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4113_equation_0, values = (var_39132_cast_fp16, var_38910_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4113_cast_fp16")]; + string _SplitHeadsQ__mh_w_4115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4115_equation_0, values = (var_39132_cast_fp16, var_38917_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4115_cast_fp16")]; + string _SplitHeadsQ__mh_w_4117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4117_equation_0, values = (var_39132_cast_fp16, var_38924_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4117_cast_fp16")]; + string _SplitHeadsQ__mh_w_4119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4119_equation_0, values = (var_39132_cast_fp16, var_38931_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4119_cast_fp16")]; + string _SplitHeadsQ__mh_w_4121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4121_equation_0, values = (var_39136_cast_fp16, var_38938_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4121_cast_fp16")]; + string _SplitHeadsQ__mh_w_4123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4123_equation_0, values = (var_39136_cast_fp16, var_38945_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4123_cast_fp16")]; + string _SplitHeadsQ__mh_w_4125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4125_equation_0, values = (var_39136_cast_fp16, var_38952_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4125_cast_fp16")]; + string _SplitHeadsQ__mh_w_4127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4127_equation_0, values = (var_39136_cast_fp16, var_38959_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4127_cast_fp16")]; + string _SplitHeadsQ__mh_w_4129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4129_equation_0, values = (var_39140_cast_fp16, var_38966_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4129_cast_fp16")]; + string _SplitHeadsQ__mh_w_4131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4131_equation_0, values = (var_39140_cast_fp16, var_38973_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4131_cast_fp16")]; + string _SplitHeadsQ__mh_w_4133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4133_equation_0, values = (var_39140_cast_fp16, var_38980_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4133_cast_fp16")]; + string _SplitHeadsQ__mh_w_4135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4135_equation_0, values = (var_39140_cast_fp16, var_38987_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4135_cast_fp16")]; + string _SplitHeadsQ__mh_w_4137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4137_equation_0, values = (var_39144_cast_fp16, var_38994_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4137_cast_fp16")]; + string _SplitHeadsQ__mh_w_4139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4139_equation_0, values = (var_39144_cast_fp16, var_39001_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4139_cast_fp16")]; + string _SplitHeadsQ__mh_w_4141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4141_equation_0, values = (var_39144_cast_fp16, var_39008_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4141_cast_fp16")]; + string _SplitHeadsQ__mh_w_4143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4143_equation_0, values = (var_39144_cast_fp16, var_39015_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4143_cast_fp16")]; + string _SplitHeadsQ__mh_w_4145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4145_equation_0, values = (var_39148_cast_fp16, var_39022_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4145_cast_fp16")]; + string _SplitHeadsQ__mh_w_4147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4147_equation_0, values = (var_39148_cast_fp16, var_39029_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4147_cast_fp16")]; + string _SplitHeadsQ__mh_w_4149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4149_equation_0, values = (var_39148_cast_fp16, var_39036_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4149_cast_fp16")]; + string _SplitHeadsQ__mh_w_4151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4151_equation_0, values = (var_39148_cast_fp16, var_39043_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4151_cast_fp16")]; + string _SplitHeadsQ__mh_w_4153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4153_equation_0, values = (var_39152_cast_fp16, var_39050_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4153_cast_fp16")]; + string _SplitHeadsQ__mh_w_4155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4155_equation_0, values = (var_39152_cast_fp16, var_39057_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4155_cast_fp16")]; + string _SplitHeadsQ__mh_w_4157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4157_equation_0, values = (var_39152_cast_fp16, var_39064_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4157_cast_fp16")]; + string _SplitHeadsQ__mh_w_4159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4159_equation_0, values = (var_39152_cast_fp16, var_39071_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4159_cast_fp16")]; + fp16 var_39393_to_fp16 = const()[name = string("op_39393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4001_cast_fp16, y = var_39393_to_fp16)[name = string("aw_chunk_4001_cast_fp16")]; + fp16 var_39395_to_fp16 = const()[name = string("op_39395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4003_cast_fp16, y = var_39395_to_fp16)[name = string("aw_chunk_4003_cast_fp16")]; + fp16 var_39397_to_fp16 = const()[name = string("op_39397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4005_cast_fp16, y = var_39397_to_fp16)[name = string("aw_chunk_4005_cast_fp16")]; + fp16 var_39399_to_fp16 = const()[name = string("op_39399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4007_cast_fp16, y = var_39399_to_fp16)[name = string("aw_chunk_4007_cast_fp16")]; + fp16 var_39401_to_fp16 = const()[name = string("op_39401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4009_cast_fp16, y = var_39401_to_fp16)[name = string("aw_chunk_4009_cast_fp16")]; + fp16 var_39403_to_fp16 = const()[name = string("op_39403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4011_cast_fp16, y = var_39403_to_fp16)[name = string("aw_chunk_4011_cast_fp16")]; + fp16 var_39405_to_fp16 = const()[name = string("op_39405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4013_cast_fp16, y = var_39405_to_fp16)[name = string("aw_chunk_4013_cast_fp16")]; + fp16 var_39407_to_fp16 = const()[name = string("op_39407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4015_cast_fp16, y = var_39407_to_fp16)[name = string("aw_chunk_4015_cast_fp16")]; + fp16 var_39409_to_fp16 = const()[name = string("op_39409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4017_cast_fp16, y = var_39409_to_fp16)[name = string("aw_chunk_4017_cast_fp16")]; + fp16 var_39411_to_fp16 = const()[name = string("op_39411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4019_cast_fp16, y = var_39411_to_fp16)[name = string("aw_chunk_4019_cast_fp16")]; + fp16 var_39413_to_fp16 = const()[name = string("op_39413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4021_cast_fp16, y = var_39413_to_fp16)[name = string("aw_chunk_4021_cast_fp16")]; + fp16 var_39415_to_fp16 = const()[name = string("op_39415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4023_cast_fp16, y = var_39415_to_fp16)[name = string("aw_chunk_4023_cast_fp16")]; + fp16 var_39417_to_fp16 = const()[name = string("op_39417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4025_cast_fp16, y = var_39417_to_fp16)[name = string("aw_chunk_4025_cast_fp16")]; + fp16 var_39419_to_fp16 = const()[name = string("op_39419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4027_cast_fp16, y = var_39419_to_fp16)[name = string("aw_chunk_4027_cast_fp16")]; + fp16 var_39421_to_fp16 = const()[name = string("op_39421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4029_cast_fp16, y = var_39421_to_fp16)[name = string("aw_chunk_4029_cast_fp16")]; + fp16 var_39423_to_fp16 = const()[name = string("op_39423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4031_cast_fp16, y = var_39423_to_fp16)[name = string("aw_chunk_4031_cast_fp16")]; + fp16 var_39425_to_fp16 = const()[name = string("op_39425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4033_cast_fp16, y = var_39425_to_fp16)[name = string("aw_chunk_4033_cast_fp16")]; + fp16 var_39427_to_fp16 = const()[name = string("op_39427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4035_cast_fp16, y = var_39427_to_fp16)[name = string("aw_chunk_4035_cast_fp16")]; + fp16 var_39429_to_fp16 = const()[name = string("op_39429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4037_cast_fp16, y = var_39429_to_fp16)[name = string("aw_chunk_4037_cast_fp16")]; + fp16 var_39431_to_fp16 = const()[name = string("op_39431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4039_cast_fp16, y = var_39431_to_fp16)[name = string("aw_chunk_4039_cast_fp16")]; + fp16 var_39433_to_fp16 = const()[name = string("op_39433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4041_cast_fp16, y = var_39433_to_fp16)[name = string("aw_chunk_4041_cast_fp16")]; + fp16 var_39435_to_fp16 = const()[name = string("op_39435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4043_cast_fp16, y = var_39435_to_fp16)[name = string("aw_chunk_4043_cast_fp16")]; + fp16 var_39437_to_fp16 = const()[name = string("op_39437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4045_cast_fp16, y = var_39437_to_fp16)[name = string("aw_chunk_4045_cast_fp16")]; + fp16 var_39439_to_fp16 = const()[name = string("op_39439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4047_cast_fp16, y = var_39439_to_fp16)[name = string("aw_chunk_4047_cast_fp16")]; + fp16 var_39441_to_fp16 = const()[name = string("op_39441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4049_cast_fp16, y = var_39441_to_fp16)[name = string("aw_chunk_4049_cast_fp16")]; + fp16 var_39443_to_fp16 = const()[name = string("op_39443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4051_cast_fp16, y = var_39443_to_fp16)[name = string("aw_chunk_4051_cast_fp16")]; + fp16 var_39445_to_fp16 = const()[name = string("op_39445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4053_cast_fp16, y = var_39445_to_fp16)[name = string("aw_chunk_4053_cast_fp16")]; + fp16 var_39447_to_fp16 = const()[name = string("op_39447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4055_cast_fp16, y = var_39447_to_fp16)[name = string("aw_chunk_4055_cast_fp16")]; + fp16 var_39449_to_fp16 = const()[name = string("op_39449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4057_cast_fp16, y = var_39449_to_fp16)[name = string("aw_chunk_4057_cast_fp16")]; + fp16 var_39451_to_fp16 = const()[name = string("op_39451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4059_cast_fp16, y = var_39451_to_fp16)[name = string("aw_chunk_4059_cast_fp16")]; + fp16 var_39453_to_fp16 = const()[name = string("op_39453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4061_cast_fp16, y = var_39453_to_fp16)[name = string("aw_chunk_4061_cast_fp16")]; + fp16 var_39455_to_fp16 = const()[name = string("op_39455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4063_cast_fp16, y = var_39455_to_fp16)[name = string("aw_chunk_4063_cast_fp16")]; + fp16 var_39457_to_fp16 = const()[name = string("op_39457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4065_cast_fp16, y = var_39457_to_fp16)[name = string("aw_chunk_4065_cast_fp16")]; + fp16 var_39459_to_fp16 = const()[name = string("op_39459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4067_cast_fp16, y = var_39459_to_fp16)[name = string("aw_chunk_4067_cast_fp16")]; + fp16 var_39461_to_fp16 = const()[name = string("op_39461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4069_cast_fp16, y = var_39461_to_fp16)[name = string("aw_chunk_4069_cast_fp16")]; + fp16 var_39463_to_fp16 = const()[name = string("op_39463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4071_cast_fp16, y = var_39463_to_fp16)[name = string("aw_chunk_4071_cast_fp16")]; + fp16 var_39465_to_fp16 = const()[name = string("op_39465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4073_cast_fp16, y = var_39465_to_fp16)[name = string("aw_chunk_4073_cast_fp16")]; + fp16 var_39467_to_fp16 = const()[name = string("op_39467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4075_cast_fp16, y = var_39467_to_fp16)[name = string("aw_chunk_4075_cast_fp16")]; + fp16 var_39469_to_fp16 = const()[name = string("op_39469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4077_cast_fp16, y = var_39469_to_fp16)[name = string("aw_chunk_4077_cast_fp16")]; + fp16 var_39471_to_fp16 = const()[name = string("op_39471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4079_cast_fp16, y = var_39471_to_fp16)[name = string("aw_chunk_4079_cast_fp16")]; + fp16 var_39473_to_fp16 = const()[name = string("op_39473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4081_cast_fp16, y = var_39473_to_fp16)[name = string("aw_chunk_4081_cast_fp16")]; + fp16 var_39475_to_fp16 = const()[name = string("op_39475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4083_cast_fp16, y = var_39475_to_fp16)[name = string("aw_chunk_4083_cast_fp16")]; + fp16 var_39477_to_fp16 = const()[name = string("op_39477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4085_cast_fp16, y = var_39477_to_fp16)[name = string("aw_chunk_4085_cast_fp16")]; + fp16 var_39479_to_fp16 = const()[name = string("op_39479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4087_cast_fp16, y = var_39479_to_fp16)[name = string("aw_chunk_4087_cast_fp16")]; + fp16 var_39481_to_fp16 = const()[name = string("op_39481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4089_cast_fp16, y = var_39481_to_fp16)[name = string("aw_chunk_4089_cast_fp16")]; + fp16 var_39483_to_fp16 = const()[name = string("op_39483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4091_cast_fp16, y = var_39483_to_fp16)[name = string("aw_chunk_4091_cast_fp16")]; + fp16 var_39485_to_fp16 = const()[name = string("op_39485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4093_cast_fp16, y = var_39485_to_fp16)[name = string("aw_chunk_4093_cast_fp16")]; + fp16 var_39487_to_fp16 = const()[name = string("op_39487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4095_cast_fp16, y = var_39487_to_fp16)[name = string("aw_chunk_4095_cast_fp16")]; + fp16 var_39489_to_fp16 = const()[name = string("op_39489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4097_cast_fp16, y = var_39489_to_fp16)[name = string("aw_chunk_4097_cast_fp16")]; + fp16 var_39491_to_fp16 = const()[name = string("op_39491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4099_cast_fp16, y = var_39491_to_fp16)[name = string("aw_chunk_4099_cast_fp16")]; + fp16 var_39493_to_fp16 = const()[name = string("op_39493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4101_cast_fp16, y = var_39493_to_fp16)[name = string("aw_chunk_4101_cast_fp16")]; + fp16 var_39495_to_fp16 = const()[name = string("op_39495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4103_cast_fp16, y = var_39495_to_fp16)[name = string("aw_chunk_4103_cast_fp16")]; + fp16 var_39497_to_fp16 = const()[name = string("op_39497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4105_cast_fp16, y = var_39497_to_fp16)[name = string("aw_chunk_4105_cast_fp16")]; + fp16 var_39499_to_fp16 = const()[name = string("op_39499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4107_cast_fp16, y = var_39499_to_fp16)[name = string("aw_chunk_4107_cast_fp16")]; + fp16 var_39501_to_fp16 = const()[name = string("op_39501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4109_cast_fp16, y = var_39501_to_fp16)[name = string("aw_chunk_4109_cast_fp16")]; + fp16 var_39503_to_fp16 = const()[name = string("op_39503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4111_cast_fp16, y = var_39503_to_fp16)[name = string("aw_chunk_4111_cast_fp16")]; + fp16 var_39505_to_fp16 = const()[name = string("op_39505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4113_cast_fp16, y = var_39505_to_fp16)[name = string("aw_chunk_4113_cast_fp16")]; + fp16 var_39507_to_fp16 = const()[name = string("op_39507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4115_cast_fp16, y = var_39507_to_fp16)[name = string("aw_chunk_4115_cast_fp16")]; + fp16 var_39509_to_fp16 = const()[name = string("op_39509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4117_cast_fp16, y = var_39509_to_fp16)[name = string("aw_chunk_4117_cast_fp16")]; + fp16 var_39511_to_fp16 = const()[name = string("op_39511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4119_cast_fp16, y = var_39511_to_fp16)[name = string("aw_chunk_4119_cast_fp16")]; + fp16 var_39513_to_fp16 = const()[name = string("op_39513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4121_cast_fp16, y = var_39513_to_fp16)[name = string("aw_chunk_4121_cast_fp16")]; + fp16 var_39515_to_fp16 = const()[name = string("op_39515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4123_cast_fp16, y = var_39515_to_fp16)[name = string("aw_chunk_4123_cast_fp16")]; + fp16 var_39517_to_fp16 = const()[name = string("op_39517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4125_cast_fp16, y = var_39517_to_fp16)[name = string("aw_chunk_4125_cast_fp16")]; + fp16 var_39519_to_fp16 = const()[name = string("op_39519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4127_cast_fp16, y = var_39519_to_fp16)[name = string("aw_chunk_4127_cast_fp16")]; + fp16 var_39521_to_fp16 = const()[name = string("op_39521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4129_cast_fp16, y = var_39521_to_fp16)[name = string("aw_chunk_4129_cast_fp16")]; + fp16 var_39523_to_fp16 = const()[name = string("op_39523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4131_cast_fp16, y = var_39523_to_fp16)[name = string("aw_chunk_4131_cast_fp16")]; + fp16 var_39525_to_fp16 = const()[name = string("op_39525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4133_cast_fp16, y = var_39525_to_fp16)[name = string("aw_chunk_4133_cast_fp16")]; + fp16 var_39527_to_fp16 = const()[name = string("op_39527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4135_cast_fp16, y = var_39527_to_fp16)[name = string("aw_chunk_4135_cast_fp16")]; + fp16 var_39529_to_fp16 = const()[name = string("op_39529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4137_cast_fp16, y = var_39529_to_fp16)[name = string("aw_chunk_4137_cast_fp16")]; + fp16 var_39531_to_fp16 = const()[name = string("op_39531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4139_cast_fp16, y = var_39531_to_fp16)[name = string("aw_chunk_4139_cast_fp16")]; + fp16 var_39533_to_fp16 = const()[name = string("op_39533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4141_cast_fp16, y = var_39533_to_fp16)[name = string("aw_chunk_4141_cast_fp16")]; + fp16 var_39535_to_fp16 = const()[name = string("op_39535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4143_cast_fp16, y = var_39535_to_fp16)[name = string("aw_chunk_4143_cast_fp16")]; + fp16 var_39537_to_fp16 = const()[name = string("op_39537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4145_cast_fp16, y = var_39537_to_fp16)[name = string("aw_chunk_4145_cast_fp16")]; + fp16 var_39539_to_fp16 = const()[name = string("op_39539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4147_cast_fp16, y = var_39539_to_fp16)[name = string("aw_chunk_4147_cast_fp16")]; + fp16 var_39541_to_fp16 = const()[name = string("op_39541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4149_cast_fp16, y = var_39541_to_fp16)[name = string("aw_chunk_4149_cast_fp16")]; + fp16 var_39543_to_fp16 = const()[name = string("op_39543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4151_cast_fp16, y = var_39543_to_fp16)[name = string("aw_chunk_4151_cast_fp16")]; + fp16 var_39545_to_fp16 = const()[name = string("op_39545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4153_cast_fp16, y = var_39545_to_fp16)[name = string("aw_chunk_4153_cast_fp16")]; + fp16 var_39547_to_fp16 = const()[name = string("op_39547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4155_cast_fp16, y = var_39547_to_fp16)[name = string("aw_chunk_4155_cast_fp16")]; + fp16 var_39549_to_fp16 = const()[name = string("op_39549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4157_cast_fp16, y = var_39549_to_fp16)[name = string("aw_chunk_4157_cast_fp16")]; + fp16 var_39551_to_fp16 = const()[name = string("op_39551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4159_cast_fp16, y = var_39551_to_fp16)[name = string("aw_chunk_4159_cast_fp16")]; + tensor var_39553_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4001_cast_fp16)[name = string("op_39553_cast_fp16")]; + tensor var_39554_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4003_cast_fp16)[name = string("op_39554_cast_fp16")]; + tensor var_39555_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4005_cast_fp16)[name = string("op_39555_cast_fp16")]; + tensor var_39556_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4007_cast_fp16)[name = string("op_39556_cast_fp16")]; + tensor var_39557_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4009_cast_fp16)[name = string("op_39557_cast_fp16")]; + tensor var_39558_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4011_cast_fp16)[name = string("op_39558_cast_fp16")]; + tensor var_39559_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4013_cast_fp16)[name = string("op_39559_cast_fp16")]; + tensor var_39560_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4015_cast_fp16)[name = string("op_39560_cast_fp16")]; + tensor var_39561_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4017_cast_fp16)[name = string("op_39561_cast_fp16")]; + tensor var_39562_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4019_cast_fp16)[name = string("op_39562_cast_fp16")]; + tensor var_39563_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4021_cast_fp16)[name = string("op_39563_cast_fp16")]; + tensor var_39564_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4023_cast_fp16)[name = string("op_39564_cast_fp16")]; + tensor var_39565_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4025_cast_fp16)[name = string("op_39565_cast_fp16")]; + tensor var_39566_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4027_cast_fp16)[name = string("op_39566_cast_fp16")]; + tensor var_39567_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4029_cast_fp16)[name = string("op_39567_cast_fp16")]; + tensor var_39568_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4031_cast_fp16)[name = string("op_39568_cast_fp16")]; + tensor var_39569_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4033_cast_fp16)[name = string("op_39569_cast_fp16")]; + tensor var_39570_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4035_cast_fp16)[name = string("op_39570_cast_fp16")]; + tensor var_39571_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4037_cast_fp16)[name = string("op_39571_cast_fp16")]; + tensor var_39572_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4039_cast_fp16)[name = string("op_39572_cast_fp16")]; + tensor var_39573_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4041_cast_fp16)[name = string("op_39573_cast_fp16")]; + tensor var_39574_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4043_cast_fp16)[name = string("op_39574_cast_fp16")]; + tensor var_39575_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4045_cast_fp16)[name = string("op_39575_cast_fp16")]; + tensor var_39576_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4047_cast_fp16)[name = string("op_39576_cast_fp16")]; + tensor var_39577_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4049_cast_fp16)[name = string("op_39577_cast_fp16")]; + tensor var_39578_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4051_cast_fp16)[name = string("op_39578_cast_fp16")]; + tensor var_39579_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4053_cast_fp16)[name = string("op_39579_cast_fp16")]; + tensor var_39580_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4055_cast_fp16)[name = string("op_39580_cast_fp16")]; + tensor var_39581_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4057_cast_fp16)[name = string("op_39581_cast_fp16")]; + tensor var_39582_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4059_cast_fp16)[name = string("op_39582_cast_fp16")]; + tensor var_39583_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4061_cast_fp16)[name = string("op_39583_cast_fp16")]; + tensor var_39584_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4063_cast_fp16)[name = string("op_39584_cast_fp16")]; + tensor var_39585_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4065_cast_fp16)[name = string("op_39585_cast_fp16")]; + tensor var_39586_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4067_cast_fp16)[name = string("op_39586_cast_fp16")]; + tensor var_39587_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4069_cast_fp16)[name = string("op_39587_cast_fp16")]; + tensor var_39588_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4071_cast_fp16)[name = string("op_39588_cast_fp16")]; + tensor var_39589_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4073_cast_fp16)[name = string("op_39589_cast_fp16")]; + tensor var_39590_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4075_cast_fp16)[name = string("op_39590_cast_fp16")]; + tensor var_39591_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4077_cast_fp16)[name = string("op_39591_cast_fp16")]; + tensor var_39592_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4079_cast_fp16)[name = string("op_39592_cast_fp16")]; + tensor var_39593_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4081_cast_fp16)[name = string("op_39593_cast_fp16")]; + tensor var_39594_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4083_cast_fp16)[name = string("op_39594_cast_fp16")]; + tensor var_39595_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4085_cast_fp16)[name = string("op_39595_cast_fp16")]; + tensor var_39596_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4087_cast_fp16)[name = string("op_39596_cast_fp16")]; + tensor var_39597_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4089_cast_fp16)[name = string("op_39597_cast_fp16")]; + tensor var_39598_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4091_cast_fp16)[name = string("op_39598_cast_fp16")]; + tensor var_39599_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4093_cast_fp16)[name = string("op_39599_cast_fp16")]; + tensor var_39600_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4095_cast_fp16)[name = string("op_39600_cast_fp16")]; + tensor var_39601_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4097_cast_fp16)[name = string("op_39601_cast_fp16")]; + tensor var_39602_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4099_cast_fp16)[name = string("op_39602_cast_fp16")]; + tensor var_39603_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4101_cast_fp16)[name = string("op_39603_cast_fp16")]; + tensor var_39604_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4103_cast_fp16)[name = string("op_39604_cast_fp16")]; + tensor var_39605_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4105_cast_fp16)[name = string("op_39605_cast_fp16")]; + tensor var_39606_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4107_cast_fp16)[name = string("op_39606_cast_fp16")]; + tensor var_39607_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4109_cast_fp16)[name = string("op_39607_cast_fp16")]; + tensor var_39608_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4111_cast_fp16)[name = string("op_39608_cast_fp16")]; + tensor var_39609_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4113_cast_fp16)[name = string("op_39609_cast_fp16")]; + tensor var_39610_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4115_cast_fp16)[name = string("op_39610_cast_fp16")]; + tensor var_39611_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4117_cast_fp16)[name = string("op_39611_cast_fp16")]; + tensor var_39612_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4119_cast_fp16)[name = string("op_39612_cast_fp16")]; + tensor var_39613_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4121_cast_fp16)[name = string("op_39613_cast_fp16")]; + tensor var_39614_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4123_cast_fp16)[name = string("op_39614_cast_fp16")]; + tensor var_39615_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4125_cast_fp16)[name = string("op_39615_cast_fp16")]; + tensor var_39616_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4127_cast_fp16)[name = string("op_39616_cast_fp16")]; + tensor var_39617_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4129_cast_fp16)[name = string("op_39617_cast_fp16")]; + tensor var_39618_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4131_cast_fp16)[name = string("op_39618_cast_fp16")]; + tensor var_39619_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4133_cast_fp16)[name = string("op_39619_cast_fp16")]; + tensor var_39620_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4135_cast_fp16)[name = string("op_39620_cast_fp16")]; + tensor var_39621_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4137_cast_fp16)[name = string("op_39621_cast_fp16")]; + tensor var_39622_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4139_cast_fp16)[name = string("op_39622_cast_fp16")]; + tensor var_39623_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4141_cast_fp16)[name = string("op_39623_cast_fp16")]; + tensor var_39624_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4143_cast_fp16)[name = string("op_39624_cast_fp16")]; + tensor var_39625_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4145_cast_fp16)[name = string("op_39625_cast_fp16")]; + tensor var_39626_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4147_cast_fp16)[name = string("op_39626_cast_fp16")]; + tensor var_39627_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4149_cast_fp16)[name = string("op_39627_cast_fp16")]; + tensor var_39628_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4151_cast_fp16)[name = string("op_39628_cast_fp16")]; + tensor var_39629_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4153_cast_fp16)[name = string("op_39629_cast_fp16")]; + tensor var_39630_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4155_cast_fp16)[name = string("op_39630_cast_fp16")]; + tensor var_39631_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4157_cast_fp16)[name = string("op_39631_cast_fp16")]; + tensor var_39632_cast_fp16 = softmax(axis = var_38378, x = aw_chunk_4159_cast_fp16)[name = string("op_39632_cast_fp16")]; + string var_39634_equation_0 = const()[name = string("op_39634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39634_cast_fp16 = einsum(equation = var_39634_equation_0, values = (var_39154_cast_fp16, var_39553_cast_fp16))[name = string("op_39634_cast_fp16")]; + string var_39636_equation_0 = const()[name = string("op_39636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39636_cast_fp16 = einsum(equation = var_39636_equation_0, values = (var_39154_cast_fp16, var_39554_cast_fp16))[name = string("op_39636_cast_fp16")]; + string var_39638_equation_0 = const()[name = string("op_39638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39638_cast_fp16 = einsum(equation = var_39638_equation_0, values = (var_39154_cast_fp16, var_39555_cast_fp16))[name = string("op_39638_cast_fp16")]; + string var_39640_equation_0 = const()[name = string("op_39640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39640_cast_fp16 = einsum(equation = var_39640_equation_0, values = (var_39154_cast_fp16, var_39556_cast_fp16))[name = string("op_39640_cast_fp16")]; + string var_39642_equation_0 = const()[name = string("op_39642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39642_cast_fp16 = einsum(equation = var_39642_equation_0, values = (var_39158_cast_fp16, var_39557_cast_fp16))[name = string("op_39642_cast_fp16")]; + string var_39644_equation_0 = const()[name = string("op_39644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39644_cast_fp16 = einsum(equation = var_39644_equation_0, values = (var_39158_cast_fp16, var_39558_cast_fp16))[name = string("op_39644_cast_fp16")]; + string var_39646_equation_0 = const()[name = string("op_39646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39646_cast_fp16 = einsum(equation = var_39646_equation_0, values = (var_39158_cast_fp16, var_39559_cast_fp16))[name = string("op_39646_cast_fp16")]; + string var_39648_equation_0 = const()[name = string("op_39648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39648_cast_fp16 = einsum(equation = var_39648_equation_0, values = (var_39158_cast_fp16, var_39560_cast_fp16))[name = string("op_39648_cast_fp16")]; + string var_39650_equation_0 = const()[name = string("op_39650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39650_cast_fp16 = einsum(equation = var_39650_equation_0, values = (var_39162_cast_fp16, var_39561_cast_fp16))[name = string("op_39650_cast_fp16")]; + string var_39652_equation_0 = const()[name = string("op_39652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39652_cast_fp16 = einsum(equation = var_39652_equation_0, values = (var_39162_cast_fp16, var_39562_cast_fp16))[name = string("op_39652_cast_fp16")]; + string var_39654_equation_0 = const()[name = string("op_39654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39654_cast_fp16 = einsum(equation = var_39654_equation_0, values = (var_39162_cast_fp16, var_39563_cast_fp16))[name = string("op_39654_cast_fp16")]; + string var_39656_equation_0 = const()[name = string("op_39656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39656_cast_fp16 = einsum(equation = var_39656_equation_0, values = (var_39162_cast_fp16, var_39564_cast_fp16))[name = string("op_39656_cast_fp16")]; + string var_39658_equation_0 = const()[name = string("op_39658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39658_cast_fp16 = einsum(equation = var_39658_equation_0, values = (var_39166_cast_fp16, var_39565_cast_fp16))[name = string("op_39658_cast_fp16")]; + string var_39660_equation_0 = const()[name = string("op_39660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39660_cast_fp16 = einsum(equation = var_39660_equation_0, values = (var_39166_cast_fp16, var_39566_cast_fp16))[name = string("op_39660_cast_fp16")]; + string var_39662_equation_0 = const()[name = string("op_39662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39662_cast_fp16 = einsum(equation = var_39662_equation_0, values = (var_39166_cast_fp16, var_39567_cast_fp16))[name = string("op_39662_cast_fp16")]; + string var_39664_equation_0 = const()[name = string("op_39664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39664_cast_fp16 = einsum(equation = var_39664_equation_0, values = (var_39166_cast_fp16, var_39568_cast_fp16))[name = string("op_39664_cast_fp16")]; + string var_39666_equation_0 = const()[name = string("op_39666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39666_cast_fp16 = einsum(equation = var_39666_equation_0, values = (var_39170_cast_fp16, var_39569_cast_fp16))[name = string("op_39666_cast_fp16")]; + string var_39668_equation_0 = const()[name = string("op_39668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39668_cast_fp16 = einsum(equation = var_39668_equation_0, values = (var_39170_cast_fp16, var_39570_cast_fp16))[name = string("op_39668_cast_fp16")]; + string var_39670_equation_0 = const()[name = string("op_39670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39670_cast_fp16 = einsum(equation = var_39670_equation_0, values = (var_39170_cast_fp16, var_39571_cast_fp16))[name = string("op_39670_cast_fp16")]; + string var_39672_equation_0 = const()[name = string("op_39672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39672_cast_fp16 = einsum(equation = var_39672_equation_0, values = (var_39170_cast_fp16, var_39572_cast_fp16))[name = string("op_39672_cast_fp16")]; + string var_39674_equation_0 = const()[name = string("op_39674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39674_cast_fp16 = einsum(equation = var_39674_equation_0, values = (var_39174_cast_fp16, var_39573_cast_fp16))[name = string("op_39674_cast_fp16")]; + string var_39676_equation_0 = const()[name = string("op_39676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39676_cast_fp16 = einsum(equation = var_39676_equation_0, values = (var_39174_cast_fp16, var_39574_cast_fp16))[name = string("op_39676_cast_fp16")]; + string var_39678_equation_0 = const()[name = string("op_39678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39678_cast_fp16 = einsum(equation = var_39678_equation_0, values = (var_39174_cast_fp16, var_39575_cast_fp16))[name = string("op_39678_cast_fp16")]; + string var_39680_equation_0 = const()[name = string("op_39680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39680_cast_fp16 = einsum(equation = var_39680_equation_0, values = (var_39174_cast_fp16, var_39576_cast_fp16))[name = string("op_39680_cast_fp16")]; + string var_39682_equation_0 = const()[name = string("op_39682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39682_cast_fp16 = einsum(equation = var_39682_equation_0, values = (var_39178_cast_fp16, var_39577_cast_fp16))[name = string("op_39682_cast_fp16")]; + string var_39684_equation_0 = const()[name = string("op_39684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39684_cast_fp16 = einsum(equation = var_39684_equation_0, values = (var_39178_cast_fp16, var_39578_cast_fp16))[name = string("op_39684_cast_fp16")]; + string var_39686_equation_0 = const()[name = string("op_39686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39686_cast_fp16 = einsum(equation = var_39686_equation_0, values = (var_39178_cast_fp16, var_39579_cast_fp16))[name = string("op_39686_cast_fp16")]; + string var_39688_equation_0 = const()[name = string("op_39688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39688_cast_fp16 = einsum(equation = var_39688_equation_0, values = (var_39178_cast_fp16, var_39580_cast_fp16))[name = string("op_39688_cast_fp16")]; + string var_39690_equation_0 = const()[name = string("op_39690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39690_cast_fp16 = einsum(equation = var_39690_equation_0, values = (var_39182_cast_fp16, var_39581_cast_fp16))[name = string("op_39690_cast_fp16")]; + string var_39692_equation_0 = const()[name = string("op_39692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39692_cast_fp16 = einsum(equation = var_39692_equation_0, values = (var_39182_cast_fp16, var_39582_cast_fp16))[name = string("op_39692_cast_fp16")]; + string var_39694_equation_0 = const()[name = string("op_39694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39694_cast_fp16 = einsum(equation = var_39694_equation_0, values = (var_39182_cast_fp16, var_39583_cast_fp16))[name = string("op_39694_cast_fp16")]; + string var_39696_equation_0 = const()[name = string("op_39696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39696_cast_fp16 = einsum(equation = var_39696_equation_0, values = (var_39182_cast_fp16, var_39584_cast_fp16))[name = string("op_39696_cast_fp16")]; + string var_39698_equation_0 = const()[name = string("op_39698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39698_cast_fp16 = einsum(equation = var_39698_equation_0, values = (var_39186_cast_fp16, var_39585_cast_fp16))[name = string("op_39698_cast_fp16")]; + string var_39700_equation_0 = const()[name = string("op_39700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39700_cast_fp16 = einsum(equation = var_39700_equation_0, values = (var_39186_cast_fp16, var_39586_cast_fp16))[name = string("op_39700_cast_fp16")]; + string var_39702_equation_0 = const()[name = string("op_39702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39702_cast_fp16 = einsum(equation = var_39702_equation_0, values = (var_39186_cast_fp16, var_39587_cast_fp16))[name = string("op_39702_cast_fp16")]; + string var_39704_equation_0 = const()[name = string("op_39704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39704_cast_fp16 = einsum(equation = var_39704_equation_0, values = (var_39186_cast_fp16, var_39588_cast_fp16))[name = string("op_39704_cast_fp16")]; + string var_39706_equation_0 = const()[name = string("op_39706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39706_cast_fp16 = einsum(equation = var_39706_equation_0, values = (var_39190_cast_fp16, var_39589_cast_fp16))[name = string("op_39706_cast_fp16")]; + string var_39708_equation_0 = const()[name = string("op_39708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39708_cast_fp16 = einsum(equation = var_39708_equation_0, values = (var_39190_cast_fp16, var_39590_cast_fp16))[name = string("op_39708_cast_fp16")]; + string var_39710_equation_0 = const()[name = string("op_39710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39710_cast_fp16 = einsum(equation = var_39710_equation_0, values = (var_39190_cast_fp16, var_39591_cast_fp16))[name = string("op_39710_cast_fp16")]; + string var_39712_equation_0 = const()[name = string("op_39712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39712_cast_fp16 = einsum(equation = var_39712_equation_0, values = (var_39190_cast_fp16, var_39592_cast_fp16))[name = string("op_39712_cast_fp16")]; + string var_39714_equation_0 = const()[name = string("op_39714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39714_cast_fp16 = einsum(equation = var_39714_equation_0, values = (var_39194_cast_fp16, var_39593_cast_fp16))[name = string("op_39714_cast_fp16")]; + string var_39716_equation_0 = const()[name = string("op_39716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39716_cast_fp16 = einsum(equation = var_39716_equation_0, values = (var_39194_cast_fp16, var_39594_cast_fp16))[name = string("op_39716_cast_fp16")]; + string var_39718_equation_0 = const()[name = string("op_39718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39718_cast_fp16 = einsum(equation = var_39718_equation_0, values = (var_39194_cast_fp16, var_39595_cast_fp16))[name = string("op_39718_cast_fp16")]; + string var_39720_equation_0 = const()[name = string("op_39720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39720_cast_fp16 = einsum(equation = var_39720_equation_0, values = (var_39194_cast_fp16, var_39596_cast_fp16))[name = string("op_39720_cast_fp16")]; + string var_39722_equation_0 = const()[name = string("op_39722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39722_cast_fp16 = einsum(equation = var_39722_equation_0, values = (var_39198_cast_fp16, var_39597_cast_fp16))[name = string("op_39722_cast_fp16")]; + string var_39724_equation_0 = const()[name = string("op_39724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39724_cast_fp16 = einsum(equation = var_39724_equation_0, values = (var_39198_cast_fp16, var_39598_cast_fp16))[name = string("op_39724_cast_fp16")]; + string var_39726_equation_0 = const()[name = string("op_39726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39726_cast_fp16 = einsum(equation = var_39726_equation_0, values = (var_39198_cast_fp16, var_39599_cast_fp16))[name = string("op_39726_cast_fp16")]; + string var_39728_equation_0 = const()[name = string("op_39728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39728_cast_fp16 = einsum(equation = var_39728_equation_0, values = (var_39198_cast_fp16, var_39600_cast_fp16))[name = string("op_39728_cast_fp16")]; + string var_39730_equation_0 = const()[name = string("op_39730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39730_cast_fp16 = einsum(equation = var_39730_equation_0, values = (var_39202_cast_fp16, var_39601_cast_fp16))[name = string("op_39730_cast_fp16")]; + string var_39732_equation_0 = const()[name = string("op_39732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39732_cast_fp16 = einsum(equation = var_39732_equation_0, values = (var_39202_cast_fp16, var_39602_cast_fp16))[name = string("op_39732_cast_fp16")]; + string var_39734_equation_0 = const()[name = string("op_39734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39734_cast_fp16 = einsum(equation = var_39734_equation_0, values = (var_39202_cast_fp16, var_39603_cast_fp16))[name = string("op_39734_cast_fp16")]; + string var_39736_equation_0 = const()[name = string("op_39736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39736_cast_fp16 = einsum(equation = var_39736_equation_0, values = (var_39202_cast_fp16, var_39604_cast_fp16))[name = string("op_39736_cast_fp16")]; + string var_39738_equation_0 = const()[name = string("op_39738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39738_cast_fp16 = einsum(equation = var_39738_equation_0, values = (var_39206_cast_fp16, var_39605_cast_fp16))[name = string("op_39738_cast_fp16")]; + string var_39740_equation_0 = const()[name = string("op_39740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39740_cast_fp16 = einsum(equation = var_39740_equation_0, values = (var_39206_cast_fp16, var_39606_cast_fp16))[name = string("op_39740_cast_fp16")]; + string var_39742_equation_0 = const()[name = string("op_39742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39742_cast_fp16 = einsum(equation = var_39742_equation_0, values = (var_39206_cast_fp16, var_39607_cast_fp16))[name = string("op_39742_cast_fp16")]; + string var_39744_equation_0 = const()[name = string("op_39744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39744_cast_fp16 = einsum(equation = var_39744_equation_0, values = (var_39206_cast_fp16, var_39608_cast_fp16))[name = string("op_39744_cast_fp16")]; + string var_39746_equation_0 = const()[name = string("op_39746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39746_cast_fp16 = einsum(equation = var_39746_equation_0, values = (var_39210_cast_fp16, var_39609_cast_fp16))[name = string("op_39746_cast_fp16")]; + string var_39748_equation_0 = const()[name = string("op_39748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39748_cast_fp16 = einsum(equation = var_39748_equation_0, values = (var_39210_cast_fp16, var_39610_cast_fp16))[name = string("op_39748_cast_fp16")]; + string var_39750_equation_0 = const()[name = string("op_39750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39750_cast_fp16 = einsum(equation = var_39750_equation_0, values = (var_39210_cast_fp16, var_39611_cast_fp16))[name = string("op_39750_cast_fp16")]; + string var_39752_equation_0 = const()[name = string("op_39752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39752_cast_fp16 = einsum(equation = var_39752_equation_0, values = (var_39210_cast_fp16, var_39612_cast_fp16))[name = string("op_39752_cast_fp16")]; + string var_39754_equation_0 = const()[name = string("op_39754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39754_cast_fp16 = einsum(equation = var_39754_equation_0, values = (var_39214_cast_fp16, var_39613_cast_fp16))[name = string("op_39754_cast_fp16")]; + string var_39756_equation_0 = const()[name = string("op_39756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39756_cast_fp16 = einsum(equation = var_39756_equation_0, values = (var_39214_cast_fp16, var_39614_cast_fp16))[name = string("op_39756_cast_fp16")]; + string var_39758_equation_0 = const()[name = string("op_39758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39758_cast_fp16 = einsum(equation = var_39758_equation_0, values = (var_39214_cast_fp16, var_39615_cast_fp16))[name = string("op_39758_cast_fp16")]; + string var_39760_equation_0 = const()[name = string("op_39760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39760_cast_fp16 = einsum(equation = var_39760_equation_0, values = (var_39214_cast_fp16, var_39616_cast_fp16))[name = string("op_39760_cast_fp16")]; + string var_39762_equation_0 = const()[name = string("op_39762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39762_cast_fp16 = einsum(equation = var_39762_equation_0, values = (var_39218_cast_fp16, var_39617_cast_fp16))[name = string("op_39762_cast_fp16")]; + string var_39764_equation_0 = const()[name = string("op_39764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39764_cast_fp16 = einsum(equation = var_39764_equation_0, values = (var_39218_cast_fp16, var_39618_cast_fp16))[name = string("op_39764_cast_fp16")]; + string var_39766_equation_0 = const()[name = string("op_39766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39766_cast_fp16 = einsum(equation = var_39766_equation_0, values = (var_39218_cast_fp16, var_39619_cast_fp16))[name = string("op_39766_cast_fp16")]; + string var_39768_equation_0 = const()[name = string("op_39768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39768_cast_fp16 = einsum(equation = var_39768_equation_0, values = (var_39218_cast_fp16, var_39620_cast_fp16))[name = string("op_39768_cast_fp16")]; + string var_39770_equation_0 = const()[name = string("op_39770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39770_cast_fp16 = einsum(equation = var_39770_equation_0, values = (var_39222_cast_fp16, var_39621_cast_fp16))[name = string("op_39770_cast_fp16")]; + string var_39772_equation_0 = const()[name = string("op_39772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39772_cast_fp16 = einsum(equation = var_39772_equation_0, values = (var_39222_cast_fp16, var_39622_cast_fp16))[name = string("op_39772_cast_fp16")]; + string var_39774_equation_0 = const()[name = string("op_39774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39774_cast_fp16 = einsum(equation = var_39774_equation_0, values = (var_39222_cast_fp16, var_39623_cast_fp16))[name = string("op_39774_cast_fp16")]; + string var_39776_equation_0 = const()[name = string("op_39776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39776_cast_fp16 = einsum(equation = var_39776_equation_0, values = (var_39222_cast_fp16, var_39624_cast_fp16))[name = string("op_39776_cast_fp16")]; + string var_39778_equation_0 = const()[name = string("op_39778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39778_cast_fp16 = einsum(equation = var_39778_equation_0, values = (var_39226_cast_fp16, var_39625_cast_fp16))[name = string("op_39778_cast_fp16")]; + string var_39780_equation_0 = const()[name = string("op_39780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39780_cast_fp16 = einsum(equation = var_39780_equation_0, values = (var_39226_cast_fp16, var_39626_cast_fp16))[name = string("op_39780_cast_fp16")]; + string var_39782_equation_0 = const()[name = string("op_39782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39782_cast_fp16 = einsum(equation = var_39782_equation_0, values = (var_39226_cast_fp16, var_39627_cast_fp16))[name = string("op_39782_cast_fp16")]; + string var_39784_equation_0 = const()[name = string("op_39784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39784_cast_fp16 = einsum(equation = var_39784_equation_0, values = (var_39226_cast_fp16, var_39628_cast_fp16))[name = string("op_39784_cast_fp16")]; + string var_39786_equation_0 = const()[name = string("op_39786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39786_cast_fp16 = einsum(equation = var_39786_equation_0, values = (var_39230_cast_fp16, var_39629_cast_fp16))[name = string("op_39786_cast_fp16")]; + string var_39788_equation_0 = const()[name = string("op_39788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39788_cast_fp16 = einsum(equation = var_39788_equation_0, values = (var_39230_cast_fp16, var_39630_cast_fp16))[name = string("op_39788_cast_fp16")]; + string var_39790_equation_0 = const()[name = string("op_39790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39790_cast_fp16 = einsum(equation = var_39790_equation_0, values = (var_39230_cast_fp16, var_39631_cast_fp16))[name = string("op_39790_cast_fp16")]; + string var_39792_equation_0 = const()[name = string("op_39792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39792_cast_fp16 = einsum(equation = var_39792_equation_0, values = (var_39230_cast_fp16, var_39632_cast_fp16))[name = string("op_39792_cast_fp16")]; + bool var_39794_interleave_0 = const()[name = string("op_39794_interleave_0"), val = bool(false)]; + tensor var_39794_cast_fp16 = concat(axis = var_38353, interleave = var_39794_interleave_0, values = (var_39634_cast_fp16, var_39636_cast_fp16, var_39638_cast_fp16, var_39640_cast_fp16))[name = string("op_39794_cast_fp16")]; + bool var_39796_interleave_0 = const()[name = string("op_39796_interleave_0"), val = bool(false)]; + tensor var_39796_cast_fp16 = concat(axis = var_38353, interleave = var_39796_interleave_0, values = (var_39642_cast_fp16, var_39644_cast_fp16, var_39646_cast_fp16, var_39648_cast_fp16))[name = string("op_39796_cast_fp16")]; + bool var_39798_interleave_0 = const()[name = string("op_39798_interleave_0"), val = bool(false)]; + tensor var_39798_cast_fp16 = concat(axis = var_38353, interleave = var_39798_interleave_0, values = (var_39650_cast_fp16, var_39652_cast_fp16, var_39654_cast_fp16, var_39656_cast_fp16))[name = string("op_39798_cast_fp16")]; + bool var_39800_interleave_0 = const()[name = string("op_39800_interleave_0"), val = bool(false)]; + tensor var_39800_cast_fp16 = concat(axis = var_38353, interleave = var_39800_interleave_0, values = (var_39658_cast_fp16, var_39660_cast_fp16, var_39662_cast_fp16, var_39664_cast_fp16))[name = string("op_39800_cast_fp16")]; + bool var_39802_interleave_0 = const()[name = string("op_39802_interleave_0"), val = bool(false)]; + tensor var_39802_cast_fp16 = concat(axis = var_38353, interleave = var_39802_interleave_0, values = (var_39666_cast_fp16, var_39668_cast_fp16, var_39670_cast_fp16, var_39672_cast_fp16))[name = string("op_39802_cast_fp16")]; + bool var_39804_interleave_0 = const()[name = string("op_39804_interleave_0"), val = bool(false)]; + tensor var_39804_cast_fp16 = concat(axis = var_38353, interleave = var_39804_interleave_0, values = (var_39674_cast_fp16, var_39676_cast_fp16, var_39678_cast_fp16, var_39680_cast_fp16))[name = string("op_39804_cast_fp16")]; + bool var_39806_interleave_0 = const()[name = string("op_39806_interleave_0"), val = bool(false)]; + tensor var_39806_cast_fp16 = concat(axis = var_38353, interleave = var_39806_interleave_0, values = (var_39682_cast_fp16, var_39684_cast_fp16, var_39686_cast_fp16, var_39688_cast_fp16))[name = string("op_39806_cast_fp16")]; + bool var_39808_interleave_0 = const()[name = string("op_39808_interleave_0"), val = bool(false)]; + tensor var_39808_cast_fp16 = concat(axis = var_38353, interleave = var_39808_interleave_0, values = (var_39690_cast_fp16, var_39692_cast_fp16, var_39694_cast_fp16, var_39696_cast_fp16))[name = string("op_39808_cast_fp16")]; + bool var_39810_interleave_0 = const()[name = string("op_39810_interleave_0"), val = bool(false)]; + tensor var_39810_cast_fp16 = concat(axis = var_38353, interleave = var_39810_interleave_0, values = (var_39698_cast_fp16, var_39700_cast_fp16, var_39702_cast_fp16, var_39704_cast_fp16))[name = string("op_39810_cast_fp16")]; + bool var_39812_interleave_0 = const()[name = string("op_39812_interleave_0"), val = bool(false)]; + tensor var_39812_cast_fp16 = concat(axis = var_38353, interleave = var_39812_interleave_0, values = (var_39706_cast_fp16, var_39708_cast_fp16, var_39710_cast_fp16, var_39712_cast_fp16))[name = string("op_39812_cast_fp16")]; + bool var_39814_interleave_0 = const()[name = string("op_39814_interleave_0"), val = bool(false)]; + tensor var_39814_cast_fp16 = concat(axis = var_38353, interleave = var_39814_interleave_0, values = (var_39714_cast_fp16, var_39716_cast_fp16, var_39718_cast_fp16, var_39720_cast_fp16))[name = string("op_39814_cast_fp16")]; + bool var_39816_interleave_0 = const()[name = string("op_39816_interleave_0"), val = bool(false)]; + tensor var_39816_cast_fp16 = concat(axis = var_38353, interleave = var_39816_interleave_0, values = (var_39722_cast_fp16, var_39724_cast_fp16, var_39726_cast_fp16, var_39728_cast_fp16))[name = string("op_39816_cast_fp16")]; + bool var_39818_interleave_0 = const()[name = string("op_39818_interleave_0"), val = bool(false)]; + tensor var_39818_cast_fp16 = concat(axis = var_38353, interleave = var_39818_interleave_0, values = (var_39730_cast_fp16, var_39732_cast_fp16, var_39734_cast_fp16, var_39736_cast_fp16))[name = string("op_39818_cast_fp16")]; + bool var_39820_interleave_0 = const()[name = string("op_39820_interleave_0"), val = bool(false)]; + tensor var_39820_cast_fp16 = concat(axis = var_38353, interleave = var_39820_interleave_0, values = (var_39738_cast_fp16, var_39740_cast_fp16, var_39742_cast_fp16, var_39744_cast_fp16))[name = string("op_39820_cast_fp16")]; + bool var_39822_interleave_0 = const()[name = string("op_39822_interleave_0"), val = bool(false)]; + tensor var_39822_cast_fp16 = concat(axis = var_38353, interleave = var_39822_interleave_0, values = (var_39746_cast_fp16, var_39748_cast_fp16, var_39750_cast_fp16, var_39752_cast_fp16))[name = string("op_39822_cast_fp16")]; + bool var_39824_interleave_0 = const()[name = string("op_39824_interleave_0"), val = bool(false)]; + tensor var_39824_cast_fp16 = concat(axis = var_38353, interleave = var_39824_interleave_0, values = (var_39754_cast_fp16, var_39756_cast_fp16, var_39758_cast_fp16, var_39760_cast_fp16))[name = string("op_39824_cast_fp16")]; + bool var_39826_interleave_0 = const()[name = string("op_39826_interleave_0"), val = bool(false)]; + tensor var_39826_cast_fp16 = concat(axis = var_38353, interleave = var_39826_interleave_0, values = (var_39762_cast_fp16, var_39764_cast_fp16, var_39766_cast_fp16, var_39768_cast_fp16))[name = string("op_39826_cast_fp16")]; + bool var_39828_interleave_0 = const()[name = string("op_39828_interleave_0"), val = bool(false)]; + tensor var_39828_cast_fp16 = concat(axis = var_38353, interleave = var_39828_interleave_0, values = (var_39770_cast_fp16, var_39772_cast_fp16, var_39774_cast_fp16, var_39776_cast_fp16))[name = string("op_39828_cast_fp16")]; + bool var_39830_interleave_0 = const()[name = string("op_39830_interleave_0"), val = bool(false)]; + tensor var_39830_cast_fp16 = concat(axis = var_38353, interleave = var_39830_interleave_0, values = (var_39778_cast_fp16, var_39780_cast_fp16, var_39782_cast_fp16, var_39784_cast_fp16))[name = string("op_39830_cast_fp16")]; + bool var_39832_interleave_0 = const()[name = string("op_39832_interleave_0"), val = bool(false)]; + tensor var_39832_cast_fp16 = concat(axis = var_38353, interleave = var_39832_interleave_0, values = (var_39786_cast_fp16, var_39788_cast_fp16, var_39790_cast_fp16, var_39792_cast_fp16))[name = string("op_39832_cast_fp16")]; + bool input_201_interleave_0 = const()[name = string("input_201_interleave_0"), val = bool(false)]; + tensor input_201_cast_fp16 = concat(axis = var_38378, interleave = input_201_interleave_0, values = (var_39794_cast_fp16, var_39796_cast_fp16, var_39798_cast_fp16, var_39800_cast_fp16, var_39802_cast_fp16, var_39804_cast_fp16, var_39806_cast_fp16, var_39808_cast_fp16, var_39810_cast_fp16, var_39812_cast_fp16, var_39814_cast_fp16, var_39816_cast_fp16, var_39818_cast_fp16, var_39820_cast_fp16, var_39822_cast_fp16, var_39824_cast_fp16, var_39826_cast_fp16, var_39828_cast_fp16, var_39830_cast_fp16, var_39832_cast_fp16))[name = string("input_201_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008337280)))]; + tensor layers_25_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011614144)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_25_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_25_self_attn_o_proj_weight_to_fp16, x = input_201_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_103_cast_fp16")]; + tensor out_103_axes_0 = const()[name = string("out_103_axes_0"), val = tensor([1])]; + fp16 var_39851_to_fp16 = const()[name = string("op_39851_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_39851_to_fp16, x = inputs_103_cast_fp16)[name = string("out_103_cast_fp16")]; + tensor input_203_gamma_0_to_fp16 = const()[name = string("input_203_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011616768)))]; + tensor input_203_beta_0_to_fp16 = const()[name = string("input_203_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011619392)))]; + fp16 input_203_epsilon_0_to_fp16 = const()[name = string("input_203_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_203_cast_fp16 = batch_norm(beta = input_203_beta_0_to_fp16, epsilon = input_203_epsilon_0_to_fp16, gamma = input_203_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = string("input_203_cast_fp16")]; + string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")]; + tensor input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor([1, 1])]; + tensor input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor([1, 1])]; + int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1)]; + tensor layers_25_fc1_weight_to_fp16 = const()[name = string("layers_25_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011622016)))]; + tensor layers_25_fc1_bias_to_fp16 = const()[name = string("layers_25_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024729280)))]; + tensor input_205_cast_fp16 = conv(bias = layers_25_fc1_bias_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = layers_25_fc1_weight_to_fp16, x = input_203_cast_fp16)[name = string("input_205_cast_fp16")]; + string input_207_mode_0 = const()[name = string("input_207_mode_0"), val = string("EXACT")]; + tensor input_207_cast_fp16 = gelu(mode = input_207_mode_0, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")]; + string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")]; + tensor hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)]; + tensor layers_25_fc2_weight_to_fp16 = const()[name = string("layers_25_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024739584)))]; + tensor layers_25_fc2_bias_to_fp16 = const()[name = string("layers_25_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037846848)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = layers_25_fc2_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = layers_25_fc2_weight_to_fp16, x = input_207_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = hidden_states_55_cast_fp16)[name = string("inputs_105_cast_fp16")]; + int32 var_39880 = const()[name = string("op_39880"), val = int32(3)]; + int32 var_39905 = const()[name = string("op_39905"), val = int32(1)]; + tensor out_105_axes_0 = const()[name = string("out_105_axes_0"), val = tensor([1])]; + fp16 var_39922_to_fp16 = const()[name = string("op_39922_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_39922_to_fp16, x = inputs_105_cast_fp16)[name = string("out_105_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037849472)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037852096)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_53_pad_type_0 = const()[name = string("query_53_pad_type_0"), val = string("valid")]; + tensor query_53_strides_0 = const()[name = string("query_53_strides_0"), val = tensor([1, 1])]; + tensor query_53_pad_0 = const()[name = string("query_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_53_dilations_0 = const()[name = string("query_53_dilations_0"), val = tensor([1, 1])]; + int32 query_53_groups_0 = const()[name = string("query_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037854720)))]; + tensor layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041131584)))]; + tensor query_53_cast_fp16 = conv(bias = layers_26_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_26_self_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_53_cast_fp16")]; + string key_53_pad_type_0 = const()[name = string("key_53_pad_type_0"), val = string("valid")]; + tensor key_53_strides_0 = const()[name = string("key_53_strides_0"), val = tensor([1, 1])]; + tensor key_53_pad_0 = const()[name = string("key_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_53_dilations_0 = const()[name = string("key_53_dilations_0"), val = tensor([1, 1])]; + int32 key_53_groups_0 = const()[name = string("key_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041134208)))]; + tensor key_53_cast_fp16 = conv(dilations = key_53_dilations_0, groups = key_53_groups_0, pad = key_53_pad_0, pad_type = key_53_pad_type_0, strides = key_53_strides_0, weight = layers_26_self_attn_k_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("key_53_cast_fp16")]; + string value_53_pad_type_0 = const()[name = string("value_53_pad_type_0"), val = string("valid")]; + tensor value_53_strides_0 = const()[name = string("value_53_strides_0"), val = tensor([1, 1])]; + tensor value_53_pad_0 = const()[name = string("value_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_53_dilations_0 = const()[name = string("value_53_dilations_0"), val = tensor([1, 1])]; + int32 value_53_groups_0 = const()[name = string("value_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1044411072)))]; + tensor layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047687936)))]; + tensor value_53_cast_fp16 = conv(bias = layers_26_self_attn_v_proj_bias_to_fp16, dilations = value_53_dilations_0, groups = value_53_groups_0, pad = value_53_pad_0, pad_type = value_53_pad_type_0, strides = value_53_strides_0, weight = layers_26_self_attn_v_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("value_53_cast_fp16")]; + tensor var_39960_begin_0 = const()[name = string("op_39960_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39960_end_0 = const()[name = string("op_39960_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39960_end_mask_0 = const()[name = string("op_39960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39960_cast_fp16 = slice_by_index(begin = var_39960_begin_0, end = var_39960_end_0, end_mask = var_39960_end_mask_0, x = query_53_cast_fp16)[name = string("op_39960_cast_fp16")]; + tensor var_39964_begin_0 = const()[name = string("op_39964_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39964_end_0 = const()[name = string("op_39964_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_39964_end_mask_0 = const()[name = string("op_39964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39964_cast_fp16 = slice_by_index(begin = var_39964_begin_0, end = var_39964_end_0, end_mask = var_39964_end_mask_0, x = query_53_cast_fp16)[name = string("op_39964_cast_fp16")]; + tensor var_39968_begin_0 = const()[name = string("op_39968_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39968_end_0 = const()[name = string("op_39968_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_39968_end_mask_0 = const()[name = string("op_39968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39968_cast_fp16 = slice_by_index(begin = var_39968_begin_0, end = var_39968_end_0, end_mask = var_39968_end_mask_0, x = query_53_cast_fp16)[name = string("op_39968_cast_fp16")]; + tensor var_39972_begin_0 = const()[name = string("op_39972_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39972_end_0 = const()[name = string("op_39972_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_39972_end_mask_0 = const()[name = string("op_39972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39972_cast_fp16 = slice_by_index(begin = var_39972_begin_0, end = var_39972_end_0, end_mask = var_39972_end_mask_0, x = query_53_cast_fp16)[name = string("op_39972_cast_fp16")]; + tensor var_39976_begin_0 = const()[name = string("op_39976_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39976_end_0 = const()[name = string("op_39976_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_39976_end_mask_0 = const()[name = string("op_39976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39976_cast_fp16 = slice_by_index(begin = var_39976_begin_0, end = var_39976_end_0, end_mask = var_39976_end_mask_0, x = query_53_cast_fp16)[name = string("op_39976_cast_fp16")]; + tensor var_39980_begin_0 = const()[name = string("op_39980_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39980_end_0 = const()[name = string("op_39980_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_39980_end_mask_0 = const()[name = string("op_39980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39980_cast_fp16 = slice_by_index(begin = var_39980_begin_0, end = var_39980_end_0, end_mask = var_39980_end_mask_0, x = query_53_cast_fp16)[name = string("op_39980_cast_fp16")]; + tensor var_39984_begin_0 = const()[name = string("op_39984_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39984_end_0 = const()[name = string("op_39984_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_39984_end_mask_0 = const()[name = string("op_39984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39984_cast_fp16 = slice_by_index(begin = var_39984_begin_0, end = var_39984_end_0, end_mask = var_39984_end_mask_0, x = query_53_cast_fp16)[name = string("op_39984_cast_fp16")]; + tensor var_39988_begin_0 = const()[name = string("op_39988_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39988_end_0 = const()[name = string("op_39988_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_39988_end_mask_0 = const()[name = string("op_39988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39988_cast_fp16 = slice_by_index(begin = var_39988_begin_0, end = var_39988_end_0, end_mask = var_39988_end_mask_0, x = query_53_cast_fp16)[name = string("op_39988_cast_fp16")]; + tensor var_39992_begin_0 = const()[name = string("op_39992_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39992_end_0 = const()[name = string("op_39992_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_39992_end_mask_0 = const()[name = string("op_39992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39992_cast_fp16 = slice_by_index(begin = var_39992_begin_0, end = var_39992_end_0, end_mask = var_39992_end_mask_0, x = query_53_cast_fp16)[name = string("op_39992_cast_fp16")]; + tensor var_39996_begin_0 = const()[name = string("op_39996_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39996_end_0 = const()[name = string("op_39996_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_39996_end_mask_0 = const()[name = string("op_39996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39996_cast_fp16 = slice_by_index(begin = var_39996_begin_0, end = var_39996_end_0, end_mask = var_39996_end_mask_0, x = query_53_cast_fp16)[name = string("op_39996_cast_fp16")]; + tensor var_40000_begin_0 = const()[name = string("op_40000_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40000_end_0 = const()[name = string("op_40000_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_40000_end_mask_0 = const()[name = string("op_40000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40000_cast_fp16 = slice_by_index(begin = var_40000_begin_0, end = var_40000_end_0, end_mask = var_40000_end_mask_0, x = query_53_cast_fp16)[name = string("op_40000_cast_fp16")]; + tensor var_40004_begin_0 = const()[name = string("op_40004_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40004_end_0 = const()[name = string("op_40004_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_40004_end_mask_0 = const()[name = string("op_40004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40004_cast_fp16 = slice_by_index(begin = var_40004_begin_0, end = var_40004_end_0, end_mask = var_40004_end_mask_0, x = query_53_cast_fp16)[name = string("op_40004_cast_fp16")]; + tensor var_40008_begin_0 = const()[name = string("op_40008_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40008_end_0 = const()[name = string("op_40008_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_40008_end_mask_0 = const()[name = string("op_40008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40008_cast_fp16 = slice_by_index(begin = var_40008_begin_0, end = var_40008_end_0, end_mask = var_40008_end_mask_0, x = query_53_cast_fp16)[name = string("op_40008_cast_fp16")]; + tensor var_40012_begin_0 = const()[name = string("op_40012_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40012_end_0 = const()[name = string("op_40012_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_40012_end_mask_0 = const()[name = string("op_40012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40012_cast_fp16 = slice_by_index(begin = var_40012_begin_0, end = var_40012_end_0, end_mask = var_40012_end_mask_0, x = query_53_cast_fp16)[name = string("op_40012_cast_fp16")]; + tensor var_40016_begin_0 = const()[name = string("op_40016_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40016_end_0 = const()[name = string("op_40016_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_40016_end_mask_0 = const()[name = string("op_40016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40016_cast_fp16 = slice_by_index(begin = var_40016_begin_0, end = var_40016_end_0, end_mask = var_40016_end_mask_0, x = query_53_cast_fp16)[name = string("op_40016_cast_fp16")]; + tensor var_40020_begin_0 = const()[name = string("op_40020_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40020_end_0 = const()[name = string("op_40020_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_40020_end_mask_0 = const()[name = string("op_40020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40020_cast_fp16 = slice_by_index(begin = var_40020_begin_0, end = var_40020_end_0, end_mask = var_40020_end_mask_0, x = query_53_cast_fp16)[name = string("op_40020_cast_fp16")]; + tensor var_40024_begin_0 = const()[name = string("op_40024_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40024_end_0 = const()[name = string("op_40024_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_40024_end_mask_0 = const()[name = string("op_40024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40024_cast_fp16 = slice_by_index(begin = var_40024_begin_0, end = var_40024_end_0, end_mask = var_40024_end_mask_0, x = query_53_cast_fp16)[name = string("op_40024_cast_fp16")]; + tensor var_40028_begin_0 = const()[name = string("op_40028_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40028_end_0 = const()[name = string("op_40028_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_40028_end_mask_0 = const()[name = string("op_40028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40028_cast_fp16 = slice_by_index(begin = var_40028_begin_0, end = var_40028_end_0, end_mask = var_40028_end_mask_0, x = query_53_cast_fp16)[name = string("op_40028_cast_fp16")]; + tensor var_40032_begin_0 = const()[name = string("op_40032_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40032_end_0 = const()[name = string("op_40032_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_40032_end_mask_0 = const()[name = string("op_40032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40032_cast_fp16 = slice_by_index(begin = var_40032_begin_0, end = var_40032_end_0, end_mask = var_40032_end_mask_0, x = query_53_cast_fp16)[name = string("op_40032_cast_fp16")]; + tensor var_40036_begin_0 = const()[name = string("op_40036_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40036_end_0 = const()[name = string("op_40036_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_40036_end_mask_0 = const()[name = string("op_40036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40036_cast_fp16 = slice_by_index(begin = var_40036_begin_0, end = var_40036_end_0, end_mask = var_40036_end_mask_0, x = query_53_cast_fp16)[name = string("op_40036_cast_fp16")]; + tensor var_40045_begin_0 = const()[name = string("op_40045_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40045_end_0 = const()[name = string("op_40045_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40045_end_mask_0 = const()[name = string("op_40045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40045_cast_fp16 = slice_by_index(begin = var_40045_begin_0, end = var_40045_end_0, end_mask = var_40045_end_mask_0, x = var_39960_cast_fp16)[name = string("op_40045_cast_fp16")]; + tensor var_40052_begin_0 = const()[name = string("op_40052_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40052_end_0 = const()[name = string("op_40052_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40052_end_mask_0 = const()[name = string("op_40052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40052_cast_fp16 = slice_by_index(begin = var_40052_begin_0, end = var_40052_end_0, end_mask = var_40052_end_mask_0, x = var_39960_cast_fp16)[name = string("op_40052_cast_fp16")]; + tensor var_40059_begin_0 = const()[name = string("op_40059_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40059_end_0 = const()[name = string("op_40059_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40059_end_mask_0 = const()[name = string("op_40059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40059_cast_fp16 = slice_by_index(begin = var_40059_begin_0, end = var_40059_end_0, end_mask = var_40059_end_mask_0, x = var_39960_cast_fp16)[name = string("op_40059_cast_fp16")]; + tensor var_40066_begin_0 = const()[name = string("op_40066_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40066_end_0 = const()[name = string("op_40066_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40066_end_mask_0 = const()[name = string("op_40066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40066_cast_fp16 = slice_by_index(begin = var_40066_begin_0, end = var_40066_end_0, end_mask = var_40066_end_mask_0, x = var_39960_cast_fp16)[name = string("op_40066_cast_fp16")]; + tensor var_40073_begin_0 = const()[name = string("op_40073_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40073_end_0 = const()[name = string("op_40073_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40073_end_mask_0 = const()[name = string("op_40073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40073_cast_fp16 = slice_by_index(begin = var_40073_begin_0, end = var_40073_end_0, end_mask = var_40073_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40073_cast_fp16")]; + tensor var_40080_begin_0 = const()[name = string("op_40080_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40080_end_0 = const()[name = string("op_40080_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40080_end_mask_0 = const()[name = string("op_40080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40080_cast_fp16 = slice_by_index(begin = var_40080_begin_0, end = var_40080_end_0, end_mask = var_40080_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40080_cast_fp16")]; + tensor var_40087_begin_0 = const()[name = string("op_40087_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40087_end_0 = const()[name = string("op_40087_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40087_end_mask_0 = const()[name = string("op_40087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40087_cast_fp16 = slice_by_index(begin = var_40087_begin_0, end = var_40087_end_0, end_mask = var_40087_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40087_cast_fp16")]; + tensor var_40094_begin_0 = const()[name = string("op_40094_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40094_end_0 = const()[name = string("op_40094_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40094_end_mask_0 = const()[name = string("op_40094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40094_cast_fp16 = slice_by_index(begin = var_40094_begin_0, end = var_40094_end_0, end_mask = var_40094_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40094_cast_fp16")]; + tensor var_40101_begin_0 = const()[name = string("op_40101_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40101_end_0 = const()[name = string("op_40101_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40101_end_mask_0 = const()[name = string("op_40101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40101_cast_fp16 = slice_by_index(begin = var_40101_begin_0, end = var_40101_end_0, end_mask = var_40101_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40101_cast_fp16")]; + tensor var_40108_begin_0 = const()[name = string("op_40108_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40108_end_0 = const()[name = string("op_40108_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40108_end_mask_0 = const()[name = string("op_40108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40108_cast_fp16 = slice_by_index(begin = var_40108_begin_0, end = var_40108_end_0, end_mask = var_40108_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40108_cast_fp16")]; + tensor var_40115_begin_0 = const()[name = string("op_40115_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40115_end_0 = const()[name = string("op_40115_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40115_end_mask_0 = const()[name = string("op_40115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40115_cast_fp16 = slice_by_index(begin = var_40115_begin_0, end = var_40115_end_0, end_mask = var_40115_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40115_cast_fp16")]; + tensor var_40122_begin_0 = const()[name = string("op_40122_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40122_end_0 = const()[name = string("op_40122_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40122_end_mask_0 = const()[name = string("op_40122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40122_cast_fp16 = slice_by_index(begin = var_40122_begin_0, end = var_40122_end_0, end_mask = var_40122_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40122_cast_fp16")]; + tensor var_40129_begin_0 = const()[name = string("op_40129_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40129_end_0 = const()[name = string("op_40129_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40129_end_mask_0 = const()[name = string("op_40129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40129_cast_fp16 = slice_by_index(begin = var_40129_begin_0, end = var_40129_end_0, end_mask = var_40129_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40129_cast_fp16")]; + tensor var_40136_begin_0 = const()[name = string("op_40136_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40136_end_0 = const()[name = string("op_40136_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40136_end_mask_0 = const()[name = string("op_40136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40136_cast_fp16 = slice_by_index(begin = var_40136_begin_0, end = var_40136_end_0, end_mask = var_40136_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40136_cast_fp16")]; + tensor var_40143_begin_0 = const()[name = string("op_40143_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40143_end_0 = const()[name = string("op_40143_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40143_end_mask_0 = const()[name = string("op_40143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40143_cast_fp16 = slice_by_index(begin = var_40143_begin_0, end = var_40143_end_0, end_mask = var_40143_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40143_cast_fp16")]; + tensor var_40150_begin_0 = const()[name = string("op_40150_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40150_end_0 = const()[name = string("op_40150_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40150_end_mask_0 = const()[name = string("op_40150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40150_cast_fp16 = slice_by_index(begin = var_40150_begin_0, end = var_40150_end_0, end_mask = var_40150_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40150_cast_fp16")]; + tensor var_40157_begin_0 = const()[name = string("op_40157_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40157_end_0 = const()[name = string("op_40157_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40157_end_mask_0 = const()[name = string("op_40157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40157_cast_fp16 = slice_by_index(begin = var_40157_begin_0, end = var_40157_end_0, end_mask = var_40157_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40157_cast_fp16")]; + tensor var_40164_begin_0 = const()[name = string("op_40164_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40164_end_0 = const()[name = string("op_40164_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40164_end_mask_0 = const()[name = string("op_40164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40164_cast_fp16 = slice_by_index(begin = var_40164_begin_0, end = var_40164_end_0, end_mask = var_40164_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40164_cast_fp16")]; + tensor var_40171_begin_0 = const()[name = string("op_40171_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40171_end_0 = const()[name = string("op_40171_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40171_end_mask_0 = const()[name = string("op_40171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40171_cast_fp16 = slice_by_index(begin = var_40171_begin_0, end = var_40171_end_0, end_mask = var_40171_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40171_cast_fp16")]; + tensor var_40178_begin_0 = const()[name = string("op_40178_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40178_end_0 = const()[name = string("op_40178_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40178_end_mask_0 = const()[name = string("op_40178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40178_cast_fp16 = slice_by_index(begin = var_40178_begin_0, end = var_40178_end_0, end_mask = var_40178_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40178_cast_fp16")]; + tensor var_40185_begin_0 = const()[name = string("op_40185_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40185_end_0 = const()[name = string("op_40185_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40185_end_mask_0 = const()[name = string("op_40185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40185_cast_fp16 = slice_by_index(begin = var_40185_begin_0, end = var_40185_end_0, end_mask = var_40185_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40185_cast_fp16")]; + tensor var_40192_begin_0 = const()[name = string("op_40192_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40192_end_0 = const()[name = string("op_40192_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40192_end_mask_0 = const()[name = string("op_40192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40192_cast_fp16 = slice_by_index(begin = var_40192_begin_0, end = var_40192_end_0, end_mask = var_40192_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40192_cast_fp16")]; + tensor var_40199_begin_0 = const()[name = string("op_40199_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40199_end_0 = const()[name = string("op_40199_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40199_end_mask_0 = const()[name = string("op_40199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40199_cast_fp16 = slice_by_index(begin = var_40199_begin_0, end = var_40199_end_0, end_mask = var_40199_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40199_cast_fp16")]; + tensor var_40206_begin_0 = const()[name = string("op_40206_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40206_end_0 = const()[name = string("op_40206_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40206_end_mask_0 = const()[name = string("op_40206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40206_cast_fp16 = slice_by_index(begin = var_40206_begin_0, end = var_40206_end_0, end_mask = var_40206_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40206_cast_fp16")]; + tensor var_40213_begin_0 = const()[name = string("op_40213_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40213_end_0 = const()[name = string("op_40213_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40213_end_mask_0 = const()[name = string("op_40213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40213_cast_fp16 = slice_by_index(begin = var_40213_begin_0, end = var_40213_end_0, end_mask = var_40213_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40213_cast_fp16")]; + tensor var_40220_begin_0 = const()[name = string("op_40220_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40220_end_0 = const()[name = string("op_40220_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40220_end_mask_0 = const()[name = string("op_40220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40220_cast_fp16 = slice_by_index(begin = var_40220_begin_0, end = var_40220_end_0, end_mask = var_40220_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40220_cast_fp16")]; + tensor var_40227_begin_0 = const()[name = string("op_40227_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40227_end_0 = const()[name = string("op_40227_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40227_end_mask_0 = const()[name = string("op_40227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40227_cast_fp16 = slice_by_index(begin = var_40227_begin_0, end = var_40227_end_0, end_mask = var_40227_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40227_cast_fp16")]; + tensor var_40234_begin_0 = const()[name = string("op_40234_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40234_end_0 = const()[name = string("op_40234_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40234_end_mask_0 = const()[name = string("op_40234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40234_cast_fp16 = slice_by_index(begin = var_40234_begin_0, end = var_40234_end_0, end_mask = var_40234_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40234_cast_fp16")]; + tensor var_40241_begin_0 = const()[name = string("op_40241_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40241_end_0 = const()[name = string("op_40241_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40241_end_mask_0 = const()[name = string("op_40241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40241_cast_fp16 = slice_by_index(begin = var_40241_begin_0, end = var_40241_end_0, end_mask = var_40241_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40241_cast_fp16")]; + tensor var_40248_begin_0 = const()[name = string("op_40248_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40248_end_0 = const()[name = string("op_40248_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40248_end_mask_0 = const()[name = string("op_40248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40248_cast_fp16 = slice_by_index(begin = var_40248_begin_0, end = var_40248_end_0, end_mask = var_40248_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40248_cast_fp16")]; + tensor var_40255_begin_0 = const()[name = string("op_40255_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40255_end_0 = const()[name = string("op_40255_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40255_end_mask_0 = const()[name = string("op_40255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40255_cast_fp16 = slice_by_index(begin = var_40255_begin_0, end = var_40255_end_0, end_mask = var_40255_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40255_cast_fp16")]; + tensor var_40262_begin_0 = const()[name = string("op_40262_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40262_end_0 = const()[name = string("op_40262_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40262_end_mask_0 = const()[name = string("op_40262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40262_cast_fp16 = slice_by_index(begin = var_40262_begin_0, end = var_40262_end_0, end_mask = var_40262_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40262_cast_fp16")]; + tensor var_40269_begin_0 = const()[name = string("op_40269_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40269_end_0 = const()[name = string("op_40269_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40269_end_mask_0 = const()[name = string("op_40269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40269_cast_fp16 = slice_by_index(begin = var_40269_begin_0, end = var_40269_end_0, end_mask = var_40269_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40269_cast_fp16")]; + tensor var_40276_begin_0 = const()[name = string("op_40276_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40276_end_0 = const()[name = string("op_40276_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40276_end_mask_0 = const()[name = string("op_40276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40276_cast_fp16 = slice_by_index(begin = var_40276_begin_0, end = var_40276_end_0, end_mask = var_40276_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40276_cast_fp16")]; + tensor var_40283_begin_0 = const()[name = string("op_40283_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40283_end_0 = const()[name = string("op_40283_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40283_end_mask_0 = const()[name = string("op_40283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40283_cast_fp16 = slice_by_index(begin = var_40283_begin_0, end = var_40283_end_0, end_mask = var_40283_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40283_cast_fp16")]; + tensor var_40290_begin_0 = const()[name = string("op_40290_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40290_end_0 = const()[name = string("op_40290_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40290_end_mask_0 = const()[name = string("op_40290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40290_cast_fp16 = slice_by_index(begin = var_40290_begin_0, end = var_40290_end_0, end_mask = var_40290_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40290_cast_fp16")]; + tensor var_40297_begin_0 = const()[name = string("op_40297_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40297_end_0 = const()[name = string("op_40297_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40297_end_mask_0 = const()[name = string("op_40297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40297_cast_fp16 = slice_by_index(begin = var_40297_begin_0, end = var_40297_end_0, end_mask = var_40297_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40297_cast_fp16")]; + tensor var_40304_begin_0 = const()[name = string("op_40304_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40304_end_0 = const()[name = string("op_40304_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40304_end_mask_0 = const()[name = string("op_40304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40304_cast_fp16 = slice_by_index(begin = var_40304_begin_0, end = var_40304_end_0, end_mask = var_40304_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40304_cast_fp16")]; + tensor var_40311_begin_0 = const()[name = string("op_40311_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40311_end_0 = const()[name = string("op_40311_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40311_end_mask_0 = const()[name = string("op_40311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40311_cast_fp16 = slice_by_index(begin = var_40311_begin_0, end = var_40311_end_0, end_mask = var_40311_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40311_cast_fp16")]; + tensor var_40318_begin_0 = const()[name = string("op_40318_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40318_end_0 = const()[name = string("op_40318_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40318_end_mask_0 = const()[name = string("op_40318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40318_cast_fp16 = slice_by_index(begin = var_40318_begin_0, end = var_40318_end_0, end_mask = var_40318_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40318_cast_fp16")]; + tensor var_40325_begin_0 = const()[name = string("op_40325_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40325_end_0 = const()[name = string("op_40325_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40325_end_mask_0 = const()[name = string("op_40325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40325_cast_fp16 = slice_by_index(begin = var_40325_begin_0, end = var_40325_end_0, end_mask = var_40325_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40325_cast_fp16")]; + tensor var_40332_begin_0 = const()[name = string("op_40332_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40332_end_0 = const()[name = string("op_40332_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40332_end_mask_0 = const()[name = string("op_40332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40332_cast_fp16 = slice_by_index(begin = var_40332_begin_0, end = var_40332_end_0, end_mask = var_40332_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40332_cast_fp16")]; + tensor var_40339_begin_0 = const()[name = string("op_40339_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40339_end_0 = const()[name = string("op_40339_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40339_end_mask_0 = const()[name = string("op_40339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40339_cast_fp16 = slice_by_index(begin = var_40339_begin_0, end = var_40339_end_0, end_mask = var_40339_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40339_cast_fp16")]; + tensor var_40346_begin_0 = const()[name = string("op_40346_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40346_end_0 = const()[name = string("op_40346_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40346_end_mask_0 = const()[name = string("op_40346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40346_cast_fp16 = slice_by_index(begin = var_40346_begin_0, end = var_40346_end_0, end_mask = var_40346_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40346_cast_fp16")]; + tensor var_40353_begin_0 = const()[name = string("op_40353_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40353_end_0 = const()[name = string("op_40353_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40353_end_mask_0 = const()[name = string("op_40353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40353_cast_fp16 = slice_by_index(begin = var_40353_begin_0, end = var_40353_end_0, end_mask = var_40353_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40353_cast_fp16")]; + tensor var_40360_begin_0 = const()[name = string("op_40360_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40360_end_0 = const()[name = string("op_40360_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40360_end_mask_0 = const()[name = string("op_40360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40360_cast_fp16 = slice_by_index(begin = var_40360_begin_0, end = var_40360_end_0, end_mask = var_40360_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40360_cast_fp16")]; + tensor var_40367_begin_0 = const()[name = string("op_40367_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40367_end_0 = const()[name = string("op_40367_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40367_end_mask_0 = const()[name = string("op_40367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40367_cast_fp16 = slice_by_index(begin = var_40367_begin_0, end = var_40367_end_0, end_mask = var_40367_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40367_cast_fp16")]; + tensor var_40374_begin_0 = const()[name = string("op_40374_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40374_end_0 = const()[name = string("op_40374_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40374_end_mask_0 = const()[name = string("op_40374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40374_cast_fp16 = slice_by_index(begin = var_40374_begin_0, end = var_40374_end_0, end_mask = var_40374_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40374_cast_fp16")]; + tensor var_40381_begin_0 = const()[name = string("op_40381_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40381_end_0 = const()[name = string("op_40381_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40381_end_mask_0 = const()[name = string("op_40381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40381_cast_fp16 = slice_by_index(begin = var_40381_begin_0, end = var_40381_end_0, end_mask = var_40381_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40381_cast_fp16")]; + tensor var_40388_begin_0 = const()[name = string("op_40388_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40388_end_0 = const()[name = string("op_40388_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40388_end_mask_0 = const()[name = string("op_40388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40388_cast_fp16 = slice_by_index(begin = var_40388_begin_0, end = var_40388_end_0, end_mask = var_40388_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40388_cast_fp16")]; + tensor var_40395_begin_0 = const()[name = string("op_40395_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40395_end_0 = const()[name = string("op_40395_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40395_end_mask_0 = const()[name = string("op_40395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40395_cast_fp16 = slice_by_index(begin = var_40395_begin_0, end = var_40395_end_0, end_mask = var_40395_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40395_cast_fp16")]; + tensor var_40402_begin_0 = const()[name = string("op_40402_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40402_end_0 = const()[name = string("op_40402_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40402_end_mask_0 = const()[name = string("op_40402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40402_cast_fp16 = slice_by_index(begin = var_40402_begin_0, end = var_40402_end_0, end_mask = var_40402_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40402_cast_fp16")]; + tensor var_40409_begin_0 = const()[name = string("op_40409_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40409_end_0 = const()[name = string("op_40409_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40409_end_mask_0 = const()[name = string("op_40409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40409_cast_fp16 = slice_by_index(begin = var_40409_begin_0, end = var_40409_end_0, end_mask = var_40409_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40409_cast_fp16")]; + tensor var_40416_begin_0 = const()[name = string("op_40416_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40416_end_0 = const()[name = string("op_40416_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40416_end_mask_0 = const()[name = string("op_40416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40416_cast_fp16 = slice_by_index(begin = var_40416_begin_0, end = var_40416_end_0, end_mask = var_40416_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40416_cast_fp16")]; + tensor var_40423_begin_0 = const()[name = string("op_40423_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40423_end_0 = const()[name = string("op_40423_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40423_end_mask_0 = const()[name = string("op_40423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40423_cast_fp16 = slice_by_index(begin = var_40423_begin_0, end = var_40423_end_0, end_mask = var_40423_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40423_cast_fp16")]; + tensor var_40430_begin_0 = const()[name = string("op_40430_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40430_end_0 = const()[name = string("op_40430_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40430_end_mask_0 = const()[name = string("op_40430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40430_cast_fp16 = slice_by_index(begin = var_40430_begin_0, end = var_40430_end_0, end_mask = var_40430_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40430_cast_fp16")]; + tensor var_40437_begin_0 = const()[name = string("op_40437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40437_end_0 = const()[name = string("op_40437_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40437_end_mask_0 = const()[name = string("op_40437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40437_cast_fp16 = slice_by_index(begin = var_40437_begin_0, end = var_40437_end_0, end_mask = var_40437_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40437_cast_fp16")]; + tensor var_40444_begin_0 = const()[name = string("op_40444_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40444_end_0 = const()[name = string("op_40444_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40444_end_mask_0 = const()[name = string("op_40444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40444_cast_fp16 = slice_by_index(begin = var_40444_begin_0, end = var_40444_end_0, end_mask = var_40444_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40444_cast_fp16")]; + tensor var_40451_begin_0 = const()[name = string("op_40451_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40451_end_0 = const()[name = string("op_40451_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40451_end_mask_0 = const()[name = string("op_40451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40451_cast_fp16 = slice_by_index(begin = var_40451_begin_0, end = var_40451_end_0, end_mask = var_40451_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40451_cast_fp16")]; + tensor var_40458_begin_0 = const()[name = string("op_40458_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40458_end_0 = const()[name = string("op_40458_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40458_end_mask_0 = const()[name = string("op_40458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40458_cast_fp16 = slice_by_index(begin = var_40458_begin_0, end = var_40458_end_0, end_mask = var_40458_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40458_cast_fp16")]; + tensor var_40465_begin_0 = const()[name = string("op_40465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40465_end_0 = const()[name = string("op_40465_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40465_end_mask_0 = const()[name = string("op_40465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40465_cast_fp16 = slice_by_index(begin = var_40465_begin_0, end = var_40465_end_0, end_mask = var_40465_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40465_cast_fp16")]; + tensor var_40472_begin_0 = const()[name = string("op_40472_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40472_end_0 = const()[name = string("op_40472_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40472_end_mask_0 = const()[name = string("op_40472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40472_cast_fp16 = slice_by_index(begin = var_40472_begin_0, end = var_40472_end_0, end_mask = var_40472_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40472_cast_fp16")]; + tensor var_40479_begin_0 = const()[name = string("op_40479_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40479_end_0 = const()[name = string("op_40479_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40479_end_mask_0 = const()[name = string("op_40479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40479_cast_fp16 = slice_by_index(begin = var_40479_begin_0, end = var_40479_end_0, end_mask = var_40479_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40479_cast_fp16")]; + tensor var_40486_begin_0 = const()[name = string("op_40486_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40486_end_0 = const()[name = string("op_40486_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40486_end_mask_0 = const()[name = string("op_40486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40486_cast_fp16 = slice_by_index(begin = var_40486_begin_0, end = var_40486_end_0, end_mask = var_40486_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40486_cast_fp16")]; + tensor var_40493_begin_0 = const()[name = string("op_40493_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40493_end_0 = const()[name = string("op_40493_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40493_end_mask_0 = const()[name = string("op_40493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40493_cast_fp16 = slice_by_index(begin = var_40493_begin_0, end = var_40493_end_0, end_mask = var_40493_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40493_cast_fp16")]; + tensor var_40500_begin_0 = const()[name = string("op_40500_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40500_end_0 = const()[name = string("op_40500_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40500_end_mask_0 = const()[name = string("op_40500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40500_cast_fp16 = slice_by_index(begin = var_40500_begin_0, end = var_40500_end_0, end_mask = var_40500_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40500_cast_fp16")]; + tensor var_40507_begin_0 = const()[name = string("op_40507_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40507_end_0 = const()[name = string("op_40507_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40507_end_mask_0 = const()[name = string("op_40507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40507_cast_fp16 = slice_by_index(begin = var_40507_begin_0, end = var_40507_end_0, end_mask = var_40507_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40507_cast_fp16")]; + tensor var_40514_begin_0 = const()[name = string("op_40514_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40514_end_0 = const()[name = string("op_40514_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40514_end_mask_0 = const()[name = string("op_40514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40514_cast_fp16 = slice_by_index(begin = var_40514_begin_0, end = var_40514_end_0, end_mask = var_40514_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40514_cast_fp16")]; + tensor var_40521_begin_0 = const()[name = string("op_40521_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40521_end_0 = const()[name = string("op_40521_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40521_end_mask_0 = const()[name = string("op_40521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40521_cast_fp16 = slice_by_index(begin = var_40521_begin_0, end = var_40521_end_0, end_mask = var_40521_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40521_cast_fp16")]; + tensor var_40528_begin_0 = const()[name = string("op_40528_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40528_end_0 = const()[name = string("op_40528_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40528_end_mask_0 = const()[name = string("op_40528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40528_cast_fp16 = slice_by_index(begin = var_40528_begin_0, end = var_40528_end_0, end_mask = var_40528_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40528_cast_fp16")]; + tensor var_40535_begin_0 = const()[name = string("op_40535_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40535_end_0 = const()[name = string("op_40535_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40535_end_mask_0 = const()[name = string("op_40535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40535_cast_fp16 = slice_by_index(begin = var_40535_begin_0, end = var_40535_end_0, end_mask = var_40535_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40535_cast_fp16")]; + tensor var_40542_begin_0 = const()[name = string("op_40542_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40542_end_0 = const()[name = string("op_40542_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40542_end_mask_0 = const()[name = string("op_40542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40542_cast_fp16 = slice_by_index(begin = var_40542_begin_0, end = var_40542_end_0, end_mask = var_40542_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40542_cast_fp16")]; + tensor var_40549_begin_0 = const()[name = string("op_40549_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40549_end_0 = const()[name = string("op_40549_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40549_end_mask_0 = const()[name = string("op_40549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40549_cast_fp16 = slice_by_index(begin = var_40549_begin_0, end = var_40549_end_0, end_mask = var_40549_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40549_cast_fp16")]; + tensor var_40556_begin_0 = const()[name = string("op_40556_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40556_end_0 = const()[name = string("op_40556_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40556_end_mask_0 = const()[name = string("op_40556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40556_cast_fp16 = slice_by_index(begin = var_40556_begin_0, end = var_40556_end_0, end_mask = var_40556_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40556_cast_fp16")]; + tensor var_40563_begin_0 = const()[name = string("op_40563_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40563_end_0 = const()[name = string("op_40563_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40563_end_mask_0 = const()[name = string("op_40563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40563_cast_fp16 = slice_by_index(begin = var_40563_begin_0, end = var_40563_end_0, end_mask = var_40563_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40563_cast_fp16")]; + tensor var_40570_begin_0 = const()[name = string("op_40570_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40570_end_0 = const()[name = string("op_40570_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40570_end_mask_0 = const()[name = string("op_40570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40570_cast_fp16 = slice_by_index(begin = var_40570_begin_0, end = var_40570_end_0, end_mask = var_40570_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40570_cast_fp16")]; + tensor var_40577_begin_0 = const()[name = string("op_40577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40577_end_0 = const()[name = string("op_40577_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40577_end_mask_0 = const()[name = string("op_40577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40577_cast_fp16 = slice_by_index(begin = var_40577_begin_0, end = var_40577_end_0, end_mask = var_40577_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40577_cast_fp16")]; + tensor var_40584_begin_0 = const()[name = string("op_40584_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40584_end_0 = const()[name = string("op_40584_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40584_end_mask_0 = const()[name = string("op_40584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40584_cast_fp16 = slice_by_index(begin = var_40584_begin_0, end = var_40584_end_0, end_mask = var_40584_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40584_cast_fp16")]; + tensor var_40591_begin_0 = const()[name = string("op_40591_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40591_end_0 = const()[name = string("op_40591_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40591_end_mask_0 = const()[name = string("op_40591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40591_cast_fp16 = slice_by_index(begin = var_40591_begin_0, end = var_40591_end_0, end_mask = var_40591_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40591_cast_fp16")]; + tensor var_40598_begin_0 = const()[name = string("op_40598_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40598_end_0 = const()[name = string("op_40598_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40598_end_mask_0 = const()[name = string("op_40598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40598_cast_fp16 = slice_by_index(begin = var_40598_begin_0, end = var_40598_end_0, end_mask = var_40598_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40598_cast_fp16")]; + tensor k_53_perm_0 = const()[name = string("k_53_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_40603_begin_0 = const()[name = string("op_40603_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40603_end_0 = const()[name = string("op_40603_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_40603_end_mask_0 = const()[name = string("op_40603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_53_cast_fp16 = transpose(perm = k_53_perm_0, x = key_53_cast_fp16)[name = string("transpose_5")]; + tensor var_40603_cast_fp16 = slice_by_index(begin = var_40603_begin_0, end = var_40603_end_0, end_mask = var_40603_end_mask_0, x = k_53_cast_fp16)[name = string("op_40603_cast_fp16")]; + tensor var_40607_begin_0 = const()[name = string("op_40607_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_40607_end_0 = const()[name = string("op_40607_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_40607_end_mask_0 = const()[name = string("op_40607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40607_cast_fp16 = slice_by_index(begin = var_40607_begin_0, end = var_40607_end_0, end_mask = var_40607_end_mask_0, x = k_53_cast_fp16)[name = string("op_40607_cast_fp16")]; + tensor var_40611_begin_0 = const()[name = string("op_40611_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_40611_end_0 = const()[name = string("op_40611_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_40611_end_mask_0 = const()[name = string("op_40611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40611_cast_fp16 = slice_by_index(begin = var_40611_begin_0, end = var_40611_end_0, end_mask = var_40611_end_mask_0, x = k_53_cast_fp16)[name = string("op_40611_cast_fp16")]; + tensor var_40615_begin_0 = const()[name = string("op_40615_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_40615_end_0 = const()[name = string("op_40615_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_40615_end_mask_0 = const()[name = string("op_40615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40615_cast_fp16 = slice_by_index(begin = var_40615_begin_0, end = var_40615_end_0, end_mask = var_40615_end_mask_0, x = k_53_cast_fp16)[name = string("op_40615_cast_fp16")]; + tensor var_40619_begin_0 = const()[name = string("op_40619_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_40619_end_0 = const()[name = string("op_40619_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_40619_end_mask_0 = const()[name = string("op_40619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40619_cast_fp16 = slice_by_index(begin = var_40619_begin_0, end = var_40619_end_0, end_mask = var_40619_end_mask_0, x = k_53_cast_fp16)[name = string("op_40619_cast_fp16")]; + tensor var_40623_begin_0 = const()[name = string("op_40623_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_40623_end_0 = const()[name = string("op_40623_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_40623_end_mask_0 = const()[name = string("op_40623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40623_cast_fp16 = slice_by_index(begin = var_40623_begin_0, end = var_40623_end_0, end_mask = var_40623_end_mask_0, x = k_53_cast_fp16)[name = string("op_40623_cast_fp16")]; + tensor var_40627_begin_0 = const()[name = string("op_40627_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_40627_end_0 = const()[name = string("op_40627_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_40627_end_mask_0 = const()[name = string("op_40627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40627_cast_fp16 = slice_by_index(begin = var_40627_begin_0, end = var_40627_end_0, end_mask = var_40627_end_mask_0, x = k_53_cast_fp16)[name = string("op_40627_cast_fp16")]; + tensor var_40631_begin_0 = const()[name = string("op_40631_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_40631_end_0 = const()[name = string("op_40631_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_40631_end_mask_0 = const()[name = string("op_40631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40631_cast_fp16 = slice_by_index(begin = var_40631_begin_0, end = var_40631_end_0, end_mask = var_40631_end_mask_0, x = k_53_cast_fp16)[name = string("op_40631_cast_fp16")]; + tensor var_40635_begin_0 = const()[name = string("op_40635_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_40635_end_0 = const()[name = string("op_40635_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_40635_end_mask_0 = const()[name = string("op_40635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40635_cast_fp16 = slice_by_index(begin = var_40635_begin_0, end = var_40635_end_0, end_mask = var_40635_end_mask_0, x = k_53_cast_fp16)[name = string("op_40635_cast_fp16")]; + tensor var_40639_begin_0 = const()[name = string("op_40639_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_40639_end_0 = const()[name = string("op_40639_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_40639_end_mask_0 = const()[name = string("op_40639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40639_cast_fp16 = slice_by_index(begin = var_40639_begin_0, end = var_40639_end_0, end_mask = var_40639_end_mask_0, x = k_53_cast_fp16)[name = string("op_40639_cast_fp16")]; + tensor var_40643_begin_0 = const()[name = string("op_40643_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_40643_end_0 = const()[name = string("op_40643_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_40643_end_mask_0 = const()[name = string("op_40643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40643_cast_fp16 = slice_by_index(begin = var_40643_begin_0, end = var_40643_end_0, end_mask = var_40643_end_mask_0, x = k_53_cast_fp16)[name = string("op_40643_cast_fp16")]; + tensor var_40647_begin_0 = const()[name = string("op_40647_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_40647_end_0 = const()[name = string("op_40647_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_40647_end_mask_0 = const()[name = string("op_40647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40647_cast_fp16 = slice_by_index(begin = var_40647_begin_0, end = var_40647_end_0, end_mask = var_40647_end_mask_0, x = k_53_cast_fp16)[name = string("op_40647_cast_fp16")]; + tensor var_40651_begin_0 = const()[name = string("op_40651_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_40651_end_0 = const()[name = string("op_40651_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_40651_end_mask_0 = const()[name = string("op_40651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40651_cast_fp16 = slice_by_index(begin = var_40651_begin_0, end = var_40651_end_0, end_mask = var_40651_end_mask_0, x = k_53_cast_fp16)[name = string("op_40651_cast_fp16")]; + tensor var_40655_begin_0 = const()[name = string("op_40655_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_40655_end_0 = const()[name = string("op_40655_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_40655_end_mask_0 = const()[name = string("op_40655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40655_cast_fp16 = slice_by_index(begin = var_40655_begin_0, end = var_40655_end_0, end_mask = var_40655_end_mask_0, x = k_53_cast_fp16)[name = string("op_40655_cast_fp16")]; + tensor var_40659_begin_0 = const()[name = string("op_40659_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_40659_end_0 = const()[name = string("op_40659_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_40659_end_mask_0 = const()[name = string("op_40659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40659_cast_fp16 = slice_by_index(begin = var_40659_begin_0, end = var_40659_end_0, end_mask = var_40659_end_mask_0, x = k_53_cast_fp16)[name = string("op_40659_cast_fp16")]; + tensor var_40663_begin_0 = const()[name = string("op_40663_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_40663_end_0 = const()[name = string("op_40663_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_40663_end_mask_0 = const()[name = string("op_40663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40663_cast_fp16 = slice_by_index(begin = var_40663_begin_0, end = var_40663_end_0, end_mask = var_40663_end_mask_0, x = k_53_cast_fp16)[name = string("op_40663_cast_fp16")]; + tensor var_40667_begin_0 = const()[name = string("op_40667_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_40667_end_0 = const()[name = string("op_40667_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_40667_end_mask_0 = const()[name = string("op_40667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40667_cast_fp16 = slice_by_index(begin = var_40667_begin_0, end = var_40667_end_0, end_mask = var_40667_end_mask_0, x = k_53_cast_fp16)[name = string("op_40667_cast_fp16")]; + tensor var_40671_begin_0 = const()[name = string("op_40671_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_40671_end_0 = const()[name = string("op_40671_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_40671_end_mask_0 = const()[name = string("op_40671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40671_cast_fp16 = slice_by_index(begin = var_40671_begin_0, end = var_40671_end_0, end_mask = var_40671_end_mask_0, x = k_53_cast_fp16)[name = string("op_40671_cast_fp16")]; + tensor var_40675_begin_0 = const()[name = string("op_40675_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_40675_end_0 = const()[name = string("op_40675_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_40675_end_mask_0 = const()[name = string("op_40675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40675_cast_fp16 = slice_by_index(begin = var_40675_begin_0, end = var_40675_end_0, end_mask = var_40675_end_mask_0, x = k_53_cast_fp16)[name = string("op_40675_cast_fp16")]; + tensor var_40679_begin_0 = const()[name = string("op_40679_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_40679_end_0 = const()[name = string("op_40679_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_40679_end_mask_0 = const()[name = string("op_40679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40679_cast_fp16 = slice_by_index(begin = var_40679_begin_0, end = var_40679_end_0, end_mask = var_40679_end_mask_0, x = k_53_cast_fp16)[name = string("op_40679_cast_fp16")]; + tensor var_40681_begin_0 = const()[name = string("op_40681_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40681_end_0 = const()[name = string("op_40681_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40681_end_mask_0 = const()[name = string("op_40681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40681_cast_fp16 = slice_by_index(begin = var_40681_begin_0, end = var_40681_end_0, end_mask = var_40681_end_mask_0, x = value_53_cast_fp16)[name = string("op_40681_cast_fp16")]; + tensor var_40685_begin_0 = const()[name = string("op_40685_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40685_end_0 = const()[name = string("op_40685_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_40685_end_mask_0 = const()[name = string("op_40685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40685_cast_fp16 = slice_by_index(begin = var_40685_begin_0, end = var_40685_end_0, end_mask = var_40685_end_mask_0, x = value_53_cast_fp16)[name = string("op_40685_cast_fp16")]; + tensor var_40689_begin_0 = const()[name = string("op_40689_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40689_end_0 = const()[name = string("op_40689_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_40689_end_mask_0 = const()[name = string("op_40689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40689_cast_fp16 = slice_by_index(begin = var_40689_begin_0, end = var_40689_end_0, end_mask = var_40689_end_mask_0, x = value_53_cast_fp16)[name = string("op_40689_cast_fp16")]; + tensor var_40693_begin_0 = const()[name = string("op_40693_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40693_end_0 = const()[name = string("op_40693_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_40693_end_mask_0 = const()[name = string("op_40693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40693_cast_fp16 = slice_by_index(begin = var_40693_begin_0, end = var_40693_end_0, end_mask = var_40693_end_mask_0, x = value_53_cast_fp16)[name = string("op_40693_cast_fp16")]; + tensor var_40697_begin_0 = const()[name = string("op_40697_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40697_end_0 = const()[name = string("op_40697_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_40697_end_mask_0 = const()[name = string("op_40697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40697_cast_fp16 = slice_by_index(begin = var_40697_begin_0, end = var_40697_end_0, end_mask = var_40697_end_mask_0, x = value_53_cast_fp16)[name = string("op_40697_cast_fp16")]; + tensor var_40701_begin_0 = const()[name = string("op_40701_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40701_end_0 = const()[name = string("op_40701_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_40701_end_mask_0 = const()[name = string("op_40701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40701_cast_fp16 = slice_by_index(begin = var_40701_begin_0, end = var_40701_end_0, end_mask = var_40701_end_mask_0, x = value_53_cast_fp16)[name = string("op_40701_cast_fp16")]; + tensor var_40705_begin_0 = const()[name = string("op_40705_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40705_end_0 = const()[name = string("op_40705_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_40705_end_mask_0 = const()[name = string("op_40705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40705_cast_fp16 = slice_by_index(begin = var_40705_begin_0, end = var_40705_end_0, end_mask = var_40705_end_mask_0, x = value_53_cast_fp16)[name = string("op_40705_cast_fp16")]; + tensor var_40709_begin_0 = const()[name = string("op_40709_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40709_end_0 = const()[name = string("op_40709_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_40709_end_mask_0 = const()[name = string("op_40709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40709_cast_fp16 = slice_by_index(begin = var_40709_begin_0, end = var_40709_end_0, end_mask = var_40709_end_mask_0, x = value_53_cast_fp16)[name = string("op_40709_cast_fp16")]; + tensor var_40713_begin_0 = const()[name = string("op_40713_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40713_end_0 = const()[name = string("op_40713_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_40713_end_mask_0 = const()[name = string("op_40713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40713_cast_fp16 = slice_by_index(begin = var_40713_begin_0, end = var_40713_end_0, end_mask = var_40713_end_mask_0, x = value_53_cast_fp16)[name = string("op_40713_cast_fp16")]; + tensor var_40717_begin_0 = const()[name = string("op_40717_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40717_end_0 = const()[name = string("op_40717_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_40717_end_mask_0 = const()[name = string("op_40717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40717_cast_fp16 = slice_by_index(begin = var_40717_begin_0, end = var_40717_end_0, end_mask = var_40717_end_mask_0, x = value_53_cast_fp16)[name = string("op_40717_cast_fp16")]; + tensor var_40721_begin_0 = const()[name = string("op_40721_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40721_end_0 = const()[name = string("op_40721_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_40721_end_mask_0 = const()[name = string("op_40721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40721_cast_fp16 = slice_by_index(begin = var_40721_begin_0, end = var_40721_end_0, end_mask = var_40721_end_mask_0, x = value_53_cast_fp16)[name = string("op_40721_cast_fp16")]; + tensor var_40725_begin_0 = const()[name = string("op_40725_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40725_end_0 = const()[name = string("op_40725_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_40725_end_mask_0 = const()[name = string("op_40725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40725_cast_fp16 = slice_by_index(begin = var_40725_begin_0, end = var_40725_end_0, end_mask = var_40725_end_mask_0, x = value_53_cast_fp16)[name = string("op_40725_cast_fp16")]; + tensor var_40729_begin_0 = const()[name = string("op_40729_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40729_end_0 = const()[name = string("op_40729_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_40729_end_mask_0 = const()[name = string("op_40729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40729_cast_fp16 = slice_by_index(begin = var_40729_begin_0, end = var_40729_end_0, end_mask = var_40729_end_mask_0, x = value_53_cast_fp16)[name = string("op_40729_cast_fp16")]; + tensor var_40733_begin_0 = const()[name = string("op_40733_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40733_end_0 = const()[name = string("op_40733_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_40733_end_mask_0 = const()[name = string("op_40733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40733_cast_fp16 = slice_by_index(begin = var_40733_begin_0, end = var_40733_end_0, end_mask = var_40733_end_mask_0, x = value_53_cast_fp16)[name = string("op_40733_cast_fp16")]; + tensor var_40737_begin_0 = const()[name = string("op_40737_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40737_end_0 = const()[name = string("op_40737_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_40737_end_mask_0 = const()[name = string("op_40737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40737_cast_fp16 = slice_by_index(begin = var_40737_begin_0, end = var_40737_end_0, end_mask = var_40737_end_mask_0, x = value_53_cast_fp16)[name = string("op_40737_cast_fp16")]; + tensor var_40741_begin_0 = const()[name = string("op_40741_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40741_end_0 = const()[name = string("op_40741_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_40741_end_mask_0 = const()[name = string("op_40741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40741_cast_fp16 = slice_by_index(begin = var_40741_begin_0, end = var_40741_end_0, end_mask = var_40741_end_mask_0, x = value_53_cast_fp16)[name = string("op_40741_cast_fp16")]; + tensor var_40745_begin_0 = const()[name = string("op_40745_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40745_end_0 = const()[name = string("op_40745_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_40745_end_mask_0 = const()[name = string("op_40745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40745_cast_fp16 = slice_by_index(begin = var_40745_begin_0, end = var_40745_end_0, end_mask = var_40745_end_mask_0, x = value_53_cast_fp16)[name = string("op_40745_cast_fp16")]; + tensor var_40749_begin_0 = const()[name = string("op_40749_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40749_end_0 = const()[name = string("op_40749_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_40749_end_mask_0 = const()[name = string("op_40749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40749_cast_fp16 = slice_by_index(begin = var_40749_begin_0, end = var_40749_end_0, end_mask = var_40749_end_mask_0, x = value_53_cast_fp16)[name = string("op_40749_cast_fp16")]; + tensor var_40753_begin_0 = const()[name = string("op_40753_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40753_end_0 = const()[name = string("op_40753_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_40753_end_mask_0 = const()[name = string("op_40753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40753_cast_fp16 = slice_by_index(begin = var_40753_begin_0, end = var_40753_end_0, end_mask = var_40753_end_mask_0, x = value_53_cast_fp16)[name = string("op_40753_cast_fp16")]; + tensor var_40757_begin_0 = const()[name = string("op_40757_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40757_end_0 = const()[name = string("op_40757_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_40757_end_mask_0 = const()[name = string("op_40757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40757_cast_fp16 = slice_by_index(begin = var_40757_begin_0, end = var_40757_end_0, end_mask = var_40757_end_mask_0, x = value_53_cast_fp16)[name = string("op_40757_cast_fp16")]; + string _SplitHeadsQ__mh_w_4161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4161_equation_0, values = (var_40603_cast_fp16, var_40045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4161_cast_fp16")]; + string _SplitHeadsQ__mh_w_4163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4163_equation_0, values = (var_40603_cast_fp16, var_40052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4163_cast_fp16")]; + string _SplitHeadsQ__mh_w_4165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4165_equation_0, values = (var_40603_cast_fp16, var_40059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4165_cast_fp16")]; + string _SplitHeadsQ__mh_w_4167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4167_equation_0, values = (var_40603_cast_fp16, var_40066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4167_cast_fp16")]; + string _SplitHeadsQ__mh_w_4169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4169_equation_0, values = (var_40607_cast_fp16, var_40073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4169_cast_fp16")]; + string _SplitHeadsQ__mh_w_4171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4171_equation_0, values = (var_40607_cast_fp16, var_40080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4171_cast_fp16")]; + string _SplitHeadsQ__mh_w_4173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4173_equation_0, values = (var_40607_cast_fp16, var_40087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4173_cast_fp16")]; + string _SplitHeadsQ__mh_w_4175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4175_equation_0, values = (var_40607_cast_fp16, var_40094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4175_cast_fp16")]; + string _SplitHeadsQ__mh_w_4177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4177_equation_0, values = (var_40611_cast_fp16, var_40101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4177_cast_fp16")]; + string _SplitHeadsQ__mh_w_4179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4179_equation_0, values = (var_40611_cast_fp16, var_40108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4179_cast_fp16")]; + string _SplitHeadsQ__mh_w_4181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4181_equation_0, values = (var_40611_cast_fp16, var_40115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4181_cast_fp16")]; + string _SplitHeadsQ__mh_w_4183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4183_equation_0, values = (var_40611_cast_fp16, var_40122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4183_cast_fp16")]; + string _SplitHeadsQ__mh_w_4185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4185_equation_0, values = (var_40615_cast_fp16, var_40129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4185_cast_fp16")]; + string _SplitHeadsQ__mh_w_4187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4187_equation_0, values = (var_40615_cast_fp16, var_40136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4187_cast_fp16")]; + string _SplitHeadsQ__mh_w_4189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4189_equation_0, values = (var_40615_cast_fp16, var_40143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4189_cast_fp16")]; + string _SplitHeadsQ__mh_w_4191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4191_equation_0, values = (var_40615_cast_fp16, var_40150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4191_cast_fp16")]; + string _SplitHeadsQ__mh_w_4193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4193_equation_0, values = (var_40619_cast_fp16, var_40157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4193_cast_fp16")]; + string _SplitHeadsQ__mh_w_4195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4195_equation_0, values = (var_40619_cast_fp16, var_40164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4195_cast_fp16")]; + string _SplitHeadsQ__mh_w_4197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4197_equation_0, values = (var_40619_cast_fp16, var_40171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4197_cast_fp16")]; + string _SplitHeadsQ__mh_w_4199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4199_equation_0, values = (var_40619_cast_fp16, var_40178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4199_cast_fp16")]; + string _SplitHeadsQ__mh_w_4201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4201_equation_0, values = (var_40623_cast_fp16, var_40185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4201_cast_fp16")]; + string _SplitHeadsQ__mh_w_4203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4203_equation_0, values = (var_40623_cast_fp16, var_40192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4203_cast_fp16")]; + string _SplitHeadsQ__mh_w_4205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4205_equation_0, values = (var_40623_cast_fp16, var_40199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4205_cast_fp16")]; + string _SplitHeadsQ__mh_w_4207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4207_equation_0, values = (var_40623_cast_fp16, var_40206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4207_cast_fp16")]; + string _SplitHeadsQ__mh_w_4209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4209_equation_0, values = (var_40627_cast_fp16, var_40213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4209_cast_fp16")]; + string _SplitHeadsQ__mh_w_4211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4211_equation_0, values = (var_40627_cast_fp16, var_40220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4211_cast_fp16")]; + string _SplitHeadsQ__mh_w_4213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4213_equation_0, values = (var_40627_cast_fp16, var_40227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4213_cast_fp16")]; + string _SplitHeadsQ__mh_w_4215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4215_equation_0, values = (var_40627_cast_fp16, var_40234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4215_cast_fp16")]; + string _SplitHeadsQ__mh_w_4217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4217_equation_0, values = (var_40631_cast_fp16, var_40241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4217_cast_fp16")]; + string _SplitHeadsQ__mh_w_4219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4219_equation_0, values = (var_40631_cast_fp16, var_40248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4219_cast_fp16")]; + string _SplitHeadsQ__mh_w_4221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4221_equation_0, values = (var_40631_cast_fp16, var_40255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4221_cast_fp16")]; + string _SplitHeadsQ__mh_w_4223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4223_equation_0, values = (var_40631_cast_fp16, var_40262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4223_cast_fp16")]; + string _SplitHeadsQ__mh_w_4225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4225_equation_0, values = (var_40635_cast_fp16, var_40269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4225_cast_fp16")]; + string _SplitHeadsQ__mh_w_4227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4227_equation_0, values = (var_40635_cast_fp16, var_40276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4227_cast_fp16")]; + string _SplitHeadsQ__mh_w_4229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4229_equation_0, values = (var_40635_cast_fp16, var_40283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4229_cast_fp16")]; + string _SplitHeadsQ__mh_w_4231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4231_equation_0, values = (var_40635_cast_fp16, var_40290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4231_cast_fp16")]; + string _SplitHeadsQ__mh_w_4233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4233_equation_0, values = (var_40639_cast_fp16, var_40297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4233_cast_fp16")]; + string _SplitHeadsQ__mh_w_4235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4235_equation_0, values = (var_40639_cast_fp16, var_40304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4235_cast_fp16")]; + string _SplitHeadsQ__mh_w_4237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4237_equation_0, values = (var_40639_cast_fp16, var_40311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4237_cast_fp16")]; + string _SplitHeadsQ__mh_w_4239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4239_equation_0, values = (var_40639_cast_fp16, var_40318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4239_cast_fp16")]; + string _SplitHeadsQ__mh_w_4241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4241_equation_0, values = (var_40643_cast_fp16, var_40325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4241_cast_fp16")]; + string _SplitHeadsQ__mh_w_4243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4243_equation_0, values = (var_40643_cast_fp16, var_40332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4243_cast_fp16")]; + string _SplitHeadsQ__mh_w_4245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4245_equation_0, values = (var_40643_cast_fp16, var_40339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4245_cast_fp16")]; + string _SplitHeadsQ__mh_w_4247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4247_equation_0, values = (var_40643_cast_fp16, var_40346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4247_cast_fp16")]; + string _SplitHeadsQ__mh_w_4249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4249_equation_0, values = (var_40647_cast_fp16, var_40353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4249_cast_fp16")]; + string _SplitHeadsQ__mh_w_4251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4251_equation_0, values = (var_40647_cast_fp16, var_40360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4251_cast_fp16")]; + string _SplitHeadsQ__mh_w_4253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4253_equation_0, values = (var_40647_cast_fp16, var_40367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4253_cast_fp16")]; + string _SplitHeadsQ__mh_w_4255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4255_equation_0, values = (var_40647_cast_fp16, var_40374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4255_cast_fp16")]; + string _SplitHeadsQ__mh_w_4257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4257_equation_0, values = (var_40651_cast_fp16, var_40381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4257_cast_fp16")]; + string _SplitHeadsQ__mh_w_4259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4259_equation_0, values = (var_40651_cast_fp16, var_40388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4259_cast_fp16")]; + string _SplitHeadsQ__mh_w_4261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4261_equation_0, values = (var_40651_cast_fp16, var_40395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4261_cast_fp16")]; + string _SplitHeadsQ__mh_w_4263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4263_equation_0, values = (var_40651_cast_fp16, var_40402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4263_cast_fp16")]; + string _SplitHeadsQ__mh_w_4265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4265_equation_0, values = (var_40655_cast_fp16, var_40409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4265_cast_fp16")]; + string _SplitHeadsQ__mh_w_4267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4267_equation_0, values = (var_40655_cast_fp16, var_40416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4267_cast_fp16")]; + string _SplitHeadsQ__mh_w_4269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4269_equation_0, values = (var_40655_cast_fp16, var_40423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4269_cast_fp16")]; + string _SplitHeadsQ__mh_w_4271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4271_equation_0, values = (var_40655_cast_fp16, var_40430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4271_cast_fp16")]; + string _SplitHeadsQ__mh_w_4273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4273_equation_0, values = (var_40659_cast_fp16, var_40437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4273_cast_fp16")]; + string _SplitHeadsQ__mh_w_4275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4275_equation_0, values = (var_40659_cast_fp16, var_40444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4275_cast_fp16")]; + string _SplitHeadsQ__mh_w_4277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4277_equation_0, values = (var_40659_cast_fp16, var_40451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4277_cast_fp16")]; + string _SplitHeadsQ__mh_w_4279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4279_equation_0, values = (var_40659_cast_fp16, var_40458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4279_cast_fp16")]; + string _SplitHeadsQ__mh_w_4281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4281_equation_0, values = (var_40663_cast_fp16, var_40465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4281_cast_fp16")]; + string _SplitHeadsQ__mh_w_4283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4283_equation_0, values = (var_40663_cast_fp16, var_40472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4283_cast_fp16")]; + string _SplitHeadsQ__mh_w_4285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4285_equation_0, values = (var_40663_cast_fp16, var_40479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4285_cast_fp16")]; + string _SplitHeadsQ__mh_w_4287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4287_equation_0, values = (var_40663_cast_fp16, var_40486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4287_cast_fp16")]; + string _SplitHeadsQ__mh_w_4289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4289_equation_0, values = (var_40667_cast_fp16, var_40493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4289_cast_fp16")]; + string _SplitHeadsQ__mh_w_4291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4291_equation_0, values = (var_40667_cast_fp16, var_40500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4291_cast_fp16")]; + string _SplitHeadsQ__mh_w_4293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4293_equation_0, values = (var_40667_cast_fp16, var_40507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4293_cast_fp16")]; + string _SplitHeadsQ__mh_w_4295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4295_equation_0, values = (var_40667_cast_fp16, var_40514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4295_cast_fp16")]; + string _SplitHeadsQ__mh_w_4297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4297_equation_0, values = (var_40671_cast_fp16, var_40521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4297_cast_fp16")]; + string _SplitHeadsQ__mh_w_4299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4299_equation_0, values = (var_40671_cast_fp16, var_40528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4299_cast_fp16")]; + string _SplitHeadsQ__mh_w_4301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4301_equation_0, values = (var_40671_cast_fp16, var_40535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4301_cast_fp16")]; + string _SplitHeadsQ__mh_w_4303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4303_equation_0, values = (var_40671_cast_fp16, var_40542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4303_cast_fp16")]; + string _SplitHeadsQ__mh_w_4305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4305_equation_0, values = (var_40675_cast_fp16, var_40549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4305_cast_fp16")]; + string _SplitHeadsQ__mh_w_4307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4307_equation_0, values = (var_40675_cast_fp16, var_40556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4307_cast_fp16")]; + string _SplitHeadsQ__mh_w_4309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4309_equation_0, values = (var_40675_cast_fp16, var_40563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4309_cast_fp16")]; + string _SplitHeadsQ__mh_w_4311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4311_equation_0, values = (var_40675_cast_fp16, var_40570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4311_cast_fp16")]; + string _SplitHeadsQ__mh_w_4313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4313_equation_0, values = (var_40679_cast_fp16, var_40577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4313_cast_fp16")]; + string _SplitHeadsQ__mh_w_4315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4315_equation_0, values = (var_40679_cast_fp16, var_40584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4315_cast_fp16")]; + string _SplitHeadsQ__mh_w_4317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4317_equation_0, values = (var_40679_cast_fp16, var_40591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4317_cast_fp16")]; + string _SplitHeadsQ__mh_w_4319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4319_equation_0, values = (var_40679_cast_fp16, var_40598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4319_cast_fp16")]; + fp16 var_40920_to_fp16 = const()[name = string("op_40920_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4161_cast_fp16, y = var_40920_to_fp16)[name = string("aw_chunk_4161_cast_fp16")]; + fp16 var_40922_to_fp16 = const()[name = string("op_40922_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4163_cast_fp16, y = var_40922_to_fp16)[name = string("aw_chunk_4163_cast_fp16")]; + fp16 var_40924_to_fp16 = const()[name = string("op_40924_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4165_cast_fp16, y = var_40924_to_fp16)[name = string("aw_chunk_4165_cast_fp16")]; + fp16 var_40926_to_fp16 = const()[name = string("op_40926_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4167_cast_fp16, y = var_40926_to_fp16)[name = string("aw_chunk_4167_cast_fp16")]; + fp16 var_40928_to_fp16 = const()[name = string("op_40928_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4169_cast_fp16, y = var_40928_to_fp16)[name = string("aw_chunk_4169_cast_fp16")]; + fp16 var_40930_to_fp16 = const()[name = string("op_40930_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4171_cast_fp16, y = var_40930_to_fp16)[name = string("aw_chunk_4171_cast_fp16")]; + fp16 var_40932_to_fp16 = const()[name = string("op_40932_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4173_cast_fp16, y = var_40932_to_fp16)[name = string("aw_chunk_4173_cast_fp16")]; + fp16 var_40934_to_fp16 = const()[name = string("op_40934_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4175_cast_fp16, y = var_40934_to_fp16)[name = string("aw_chunk_4175_cast_fp16")]; + fp16 var_40936_to_fp16 = const()[name = string("op_40936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4177_cast_fp16, y = var_40936_to_fp16)[name = string("aw_chunk_4177_cast_fp16")]; + fp16 var_40938_to_fp16 = const()[name = string("op_40938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4179_cast_fp16, y = var_40938_to_fp16)[name = string("aw_chunk_4179_cast_fp16")]; + fp16 var_40940_to_fp16 = const()[name = string("op_40940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4181_cast_fp16, y = var_40940_to_fp16)[name = string("aw_chunk_4181_cast_fp16")]; + fp16 var_40942_to_fp16 = const()[name = string("op_40942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4183_cast_fp16, y = var_40942_to_fp16)[name = string("aw_chunk_4183_cast_fp16")]; + fp16 var_40944_to_fp16 = const()[name = string("op_40944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4185_cast_fp16, y = var_40944_to_fp16)[name = string("aw_chunk_4185_cast_fp16")]; + fp16 var_40946_to_fp16 = const()[name = string("op_40946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4187_cast_fp16, y = var_40946_to_fp16)[name = string("aw_chunk_4187_cast_fp16")]; + fp16 var_40948_to_fp16 = const()[name = string("op_40948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4189_cast_fp16, y = var_40948_to_fp16)[name = string("aw_chunk_4189_cast_fp16")]; + fp16 var_40950_to_fp16 = const()[name = string("op_40950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4191_cast_fp16, y = var_40950_to_fp16)[name = string("aw_chunk_4191_cast_fp16")]; + fp16 var_40952_to_fp16 = const()[name = string("op_40952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4193_cast_fp16, y = var_40952_to_fp16)[name = string("aw_chunk_4193_cast_fp16")]; + fp16 var_40954_to_fp16 = const()[name = string("op_40954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4195_cast_fp16, y = var_40954_to_fp16)[name = string("aw_chunk_4195_cast_fp16")]; + fp16 var_40956_to_fp16 = const()[name = string("op_40956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4197_cast_fp16, y = var_40956_to_fp16)[name = string("aw_chunk_4197_cast_fp16")]; + fp16 var_40958_to_fp16 = const()[name = string("op_40958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4199_cast_fp16, y = var_40958_to_fp16)[name = string("aw_chunk_4199_cast_fp16")]; + fp16 var_40960_to_fp16 = const()[name = string("op_40960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4201_cast_fp16, y = var_40960_to_fp16)[name = string("aw_chunk_4201_cast_fp16")]; + fp16 var_40962_to_fp16 = const()[name = string("op_40962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4203_cast_fp16, y = var_40962_to_fp16)[name = string("aw_chunk_4203_cast_fp16")]; + fp16 var_40964_to_fp16 = const()[name = string("op_40964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4205_cast_fp16, y = var_40964_to_fp16)[name = string("aw_chunk_4205_cast_fp16")]; + fp16 var_40966_to_fp16 = const()[name = string("op_40966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4207_cast_fp16, y = var_40966_to_fp16)[name = string("aw_chunk_4207_cast_fp16")]; + fp16 var_40968_to_fp16 = const()[name = string("op_40968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4209_cast_fp16, y = var_40968_to_fp16)[name = string("aw_chunk_4209_cast_fp16")]; + fp16 var_40970_to_fp16 = const()[name = string("op_40970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4211_cast_fp16, y = var_40970_to_fp16)[name = string("aw_chunk_4211_cast_fp16")]; + fp16 var_40972_to_fp16 = const()[name = string("op_40972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4213_cast_fp16, y = var_40972_to_fp16)[name = string("aw_chunk_4213_cast_fp16")]; + fp16 var_40974_to_fp16 = const()[name = string("op_40974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4215_cast_fp16, y = var_40974_to_fp16)[name = string("aw_chunk_4215_cast_fp16")]; + fp16 var_40976_to_fp16 = const()[name = string("op_40976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4217_cast_fp16, y = var_40976_to_fp16)[name = string("aw_chunk_4217_cast_fp16")]; + fp16 var_40978_to_fp16 = const()[name = string("op_40978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4219_cast_fp16, y = var_40978_to_fp16)[name = string("aw_chunk_4219_cast_fp16")]; + fp16 var_40980_to_fp16 = const()[name = string("op_40980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4221_cast_fp16, y = var_40980_to_fp16)[name = string("aw_chunk_4221_cast_fp16")]; + fp16 var_40982_to_fp16 = const()[name = string("op_40982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4223_cast_fp16, y = var_40982_to_fp16)[name = string("aw_chunk_4223_cast_fp16")]; + fp16 var_40984_to_fp16 = const()[name = string("op_40984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4225_cast_fp16, y = var_40984_to_fp16)[name = string("aw_chunk_4225_cast_fp16")]; + fp16 var_40986_to_fp16 = const()[name = string("op_40986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4227_cast_fp16, y = var_40986_to_fp16)[name = string("aw_chunk_4227_cast_fp16")]; + fp16 var_40988_to_fp16 = const()[name = string("op_40988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4229_cast_fp16, y = var_40988_to_fp16)[name = string("aw_chunk_4229_cast_fp16")]; + fp16 var_40990_to_fp16 = const()[name = string("op_40990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4231_cast_fp16, y = var_40990_to_fp16)[name = string("aw_chunk_4231_cast_fp16")]; + fp16 var_40992_to_fp16 = const()[name = string("op_40992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4233_cast_fp16, y = var_40992_to_fp16)[name = string("aw_chunk_4233_cast_fp16")]; + fp16 var_40994_to_fp16 = const()[name = string("op_40994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4235_cast_fp16, y = var_40994_to_fp16)[name = string("aw_chunk_4235_cast_fp16")]; + fp16 var_40996_to_fp16 = const()[name = string("op_40996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4237_cast_fp16, y = var_40996_to_fp16)[name = string("aw_chunk_4237_cast_fp16")]; + fp16 var_40998_to_fp16 = const()[name = string("op_40998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4239_cast_fp16, y = var_40998_to_fp16)[name = string("aw_chunk_4239_cast_fp16")]; + fp16 var_41000_to_fp16 = const()[name = string("op_41000_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4241_cast_fp16, y = var_41000_to_fp16)[name = string("aw_chunk_4241_cast_fp16")]; + fp16 var_41002_to_fp16 = const()[name = string("op_41002_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4243_cast_fp16, y = var_41002_to_fp16)[name = string("aw_chunk_4243_cast_fp16")]; + fp16 var_41004_to_fp16 = const()[name = string("op_41004_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4245_cast_fp16, y = var_41004_to_fp16)[name = string("aw_chunk_4245_cast_fp16")]; + fp16 var_41006_to_fp16 = const()[name = string("op_41006_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4247_cast_fp16, y = var_41006_to_fp16)[name = string("aw_chunk_4247_cast_fp16")]; + fp16 var_41008_to_fp16 = const()[name = string("op_41008_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4249_cast_fp16, y = var_41008_to_fp16)[name = string("aw_chunk_4249_cast_fp16")]; + fp16 var_41010_to_fp16 = const()[name = string("op_41010_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4251_cast_fp16, y = var_41010_to_fp16)[name = string("aw_chunk_4251_cast_fp16")]; + fp16 var_41012_to_fp16 = const()[name = string("op_41012_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4253_cast_fp16, y = var_41012_to_fp16)[name = string("aw_chunk_4253_cast_fp16")]; + fp16 var_41014_to_fp16 = const()[name = string("op_41014_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4255_cast_fp16, y = var_41014_to_fp16)[name = string("aw_chunk_4255_cast_fp16")]; + fp16 var_41016_to_fp16 = const()[name = string("op_41016_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4257_cast_fp16, y = var_41016_to_fp16)[name = string("aw_chunk_4257_cast_fp16")]; + fp16 var_41018_to_fp16 = const()[name = string("op_41018_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4259_cast_fp16, y = var_41018_to_fp16)[name = string("aw_chunk_4259_cast_fp16")]; + fp16 var_41020_to_fp16 = const()[name = string("op_41020_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4261_cast_fp16, y = var_41020_to_fp16)[name = string("aw_chunk_4261_cast_fp16")]; + fp16 var_41022_to_fp16 = const()[name = string("op_41022_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4263_cast_fp16, y = var_41022_to_fp16)[name = string("aw_chunk_4263_cast_fp16")]; + fp16 var_41024_to_fp16 = const()[name = string("op_41024_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4265_cast_fp16, y = var_41024_to_fp16)[name = string("aw_chunk_4265_cast_fp16")]; + fp16 var_41026_to_fp16 = const()[name = string("op_41026_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4267_cast_fp16, y = var_41026_to_fp16)[name = string("aw_chunk_4267_cast_fp16")]; + fp16 var_41028_to_fp16 = const()[name = string("op_41028_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4269_cast_fp16, y = var_41028_to_fp16)[name = string("aw_chunk_4269_cast_fp16")]; + fp16 var_41030_to_fp16 = const()[name = string("op_41030_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4271_cast_fp16, y = var_41030_to_fp16)[name = string("aw_chunk_4271_cast_fp16")]; + fp16 var_41032_to_fp16 = const()[name = string("op_41032_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4273_cast_fp16, y = var_41032_to_fp16)[name = string("aw_chunk_4273_cast_fp16")]; + fp16 var_41034_to_fp16 = const()[name = string("op_41034_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4275_cast_fp16, y = var_41034_to_fp16)[name = string("aw_chunk_4275_cast_fp16")]; + fp16 var_41036_to_fp16 = const()[name = string("op_41036_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4277_cast_fp16, y = var_41036_to_fp16)[name = string("aw_chunk_4277_cast_fp16")]; + fp16 var_41038_to_fp16 = const()[name = string("op_41038_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4279_cast_fp16, y = var_41038_to_fp16)[name = string("aw_chunk_4279_cast_fp16")]; + fp16 var_41040_to_fp16 = const()[name = string("op_41040_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4281_cast_fp16, y = var_41040_to_fp16)[name = string("aw_chunk_4281_cast_fp16")]; + fp16 var_41042_to_fp16 = const()[name = string("op_41042_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4283_cast_fp16, y = var_41042_to_fp16)[name = string("aw_chunk_4283_cast_fp16")]; + fp16 var_41044_to_fp16 = const()[name = string("op_41044_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4285_cast_fp16, y = var_41044_to_fp16)[name = string("aw_chunk_4285_cast_fp16")]; + fp16 var_41046_to_fp16 = const()[name = string("op_41046_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4287_cast_fp16, y = var_41046_to_fp16)[name = string("aw_chunk_4287_cast_fp16")]; + fp16 var_41048_to_fp16 = const()[name = string("op_41048_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4289_cast_fp16, y = var_41048_to_fp16)[name = string("aw_chunk_4289_cast_fp16")]; + fp16 var_41050_to_fp16 = const()[name = string("op_41050_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4291_cast_fp16, y = var_41050_to_fp16)[name = string("aw_chunk_4291_cast_fp16")]; + fp16 var_41052_to_fp16 = const()[name = string("op_41052_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4293_cast_fp16, y = var_41052_to_fp16)[name = string("aw_chunk_4293_cast_fp16")]; + fp16 var_41054_to_fp16 = const()[name = string("op_41054_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4295_cast_fp16, y = var_41054_to_fp16)[name = string("aw_chunk_4295_cast_fp16")]; + fp16 var_41056_to_fp16 = const()[name = string("op_41056_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4297_cast_fp16, y = var_41056_to_fp16)[name = string("aw_chunk_4297_cast_fp16")]; + fp16 var_41058_to_fp16 = const()[name = string("op_41058_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4299_cast_fp16, y = var_41058_to_fp16)[name = string("aw_chunk_4299_cast_fp16")]; + fp16 var_41060_to_fp16 = const()[name = string("op_41060_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4301_cast_fp16, y = var_41060_to_fp16)[name = string("aw_chunk_4301_cast_fp16")]; + fp16 var_41062_to_fp16 = const()[name = string("op_41062_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4303_cast_fp16, y = var_41062_to_fp16)[name = string("aw_chunk_4303_cast_fp16")]; + fp16 var_41064_to_fp16 = const()[name = string("op_41064_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4305_cast_fp16, y = var_41064_to_fp16)[name = string("aw_chunk_4305_cast_fp16")]; + fp16 var_41066_to_fp16 = const()[name = string("op_41066_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4307_cast_fp16, y = var_41066_to_fp16)[name = string("aw_chunk_4307_cast_fp16")]; + fp16 var_41068_to_fp16 = const()[name = string("op_41068_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4309_cast_fp16, y = var_41068_to_fp16)[name = string("aw_chunk_4309_cast_fp16")]; + fp16 var_41070_to_fp16 = const()[name = string("op_41070_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4311_cast_fp16, y = var_41070_to_fp16)[name = string("aw_chunk_4311_cast_fp16")]; + fp16 var_41072_to_fp16 = const()[name = string("op_41072_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4313_cast_fp16, y = var_41072_to_fp16)[name = string("aw_chunk_4313_cast_fp16")]; + fp16 var_41074_to_fp16 = const()[name = string("op_41074_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4315_cast_fp16, y = var_41074_to_fp16)[name = string("aw_chunk_4315_cast_fp16")]; + fp16 var_41076_to_fp16 = const()[name = string("op_41076_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4317_cast_fp16, y = var_41076_to_fp16)[name = string("aw_chunk_4317_cast_fp16")]; + fp16 var_41078_to_fp16 = const()[name = string("op_41078_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4319_cast_fp16, y = var_41078_to_fp16)[name = string("aw_chunk_4319_cast_fp16")]; + tensor var_41080_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4161_cast_fp16)[name = string("op_41080_cast_fp16")]; + tensor var_41081_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4163_cast_fp16)[name = string("op_41081_cast_fp16")]; + tensor var_41082_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4165_cast_fp16)[name = string("op_41082_cast_fp16")]; + tensor var_41083_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4167_cast_fp16)[name = string("op_41083_cast_fp16")]; + tensor var_41084_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4169_cast_fp16)[name = string("op_41084_cast_fp16")]; + tensor var_41085_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4171_cast_fp16)[name = string("op_41085_cast_fp16")]; + tensor var_41086_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4173_cast_fp16)[name = string("op_41086_cast_fp16")]; + tensor var_41087_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4175_cast_fp16)[name = string("op_41087_cast_fp16")]; + tensor var_41088_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4177_cast_fp16)[name = string("op_41088_cast_fp16")]; + tensor var_41089_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4179_cast_fp16)[name = string("op_41089_cast_fp16")]; + tensor var_41090_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4181_cast_fp16)[name = string("op_41090_cast_fp16")]; + tensor var_41091_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4183_cast_fp16)[name = string("op_41091_cast_fp16")]; + tensor var_41092_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4185_cast_fp16)[name = string("op_41092_cast_fp16")]; + tensor var_41093_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4187_cast_fp16)[name = string("op_41093_cast_fp16")]; + tensor var_41094_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4189_cast_fp16)[name = string("op_41094_cast_fp16")]; + tensor var_41095_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4191_cast_fp16)[name = string("op_41095_cast_fp16")]; + tensor var_41096_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4193_cast_fp16)[name = string("op_41096_cast_fp16")]; + tensor var_41097_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4195_cast_fp16)[name = string("op_41097_cast_fp16")]; + tensor var_41098_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4197_cast_fp16)[name = string("op_41098_cast_fp16")]; + tensor var_41099_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4199_cast_fp16)[name = string("op_41099_cast_fp16")]; + tensor var_41100_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4201_cast_fp16)[name = string("op_41100_cast_fp16")]; + tensor var_41101_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4203_cast_fp16)[name = string("op_41101_cast_fp16")]; + tensor var_41102_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4205_cast_fp16)[name = string("op_41102_cast_fp16")]; + tensor var_41103_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4207_cast_fp16)[name = string("op_41103_cast_fp16")]; + tensor var_41104_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4209_cast_fp16)[name = string("op_41104_cast_fp16")]; + tensor var_41105_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4211_cast_fp16)[name = string("op_41105_cast_fp16")]; + tensor var_41106_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4213_cast_fp16)[name = string("op_41106_cast_fp16")]; + tensor var_41107_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4215_cast_fp16)[name = string("op_41107_cast_fp16")]; + tensor var_41108_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4217_cast_fp16)[name = string("op_41108_cast_fp16")]; + tensor var_41109_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4219_cast_fp16)[name = string("op_41109_cast_fp16")]; + tensor var_41110_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4221_cast_fp16)[name = string("op_41110_cast_fp16")]; + tensor var_41111_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4223_cast_fp16)[name = string("op_41111_cast_fp16")]; + tensor var_41112_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4225_cast_fp16)[name = string("op_41112_cast_fp16")]; + tensor var_41113_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4227_cast_fp16)[name = string("op_41113_cast_fp16")]; + tensor var_41114_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4229_cast_fp16)[name = string("op_41114_cast_fp16")]; + tensor var_41115_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4231_cast_fp16)[name = string("op_41115_cast_fp16")]; + tensor var_41116_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4233_cast_fp16)[name = string("op_41116_cast_fp16")]; + tensor var_41117_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4235_cast_fp16)[name = string("op_41117_cast_fp16")]; + tensor var_41118_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4237_cast_fp16)[name = string("op_41118_cast_fp16")]; + tensor var_41119_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4239_cast_fp16)[name = string("op_41119_cast_fp16")]; + tensor var_41120_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4241_cast_fp16)[name = string("op_41120_cast_fp16")]; + tensor var_41121_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4243_cast_fp16)[name = string("op_41121_cast_fp16")]; + tensor var_41122_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4245_cast_fp16)[name = string("op_41122_cast_fp16")]; + tensor var_41123_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4247_cast_fp16)[name = string("op_41123_cast_fp16")]; + tensor var_41124_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4249_cast_fp16)[name = string("op_41124_cast_fp16")]; + tensor var_41125_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4251_cast_fp16)[name = string("op_41125_cast_fp16")]; + tensor var_41126_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4253_cast_fp16)[name = string("op_41126_cast_fp16")]; + tensor var_41127_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4255_cast_fp16)[name = string("op_41127_cast_fp16")]; + tensor var_41128_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4257_cast_fp16)[name = string("op_41128_cast_fp16")]; + tensor var_41129_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4259_cast_fp16)[name = string("op_41129_cast_fp16")]; + tensor var_41130_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4261_cast_fp16)[name = string("op_41130_cast_fp16")]; + tensor var_41131_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4263_cast_fp16)[name = string("op_41131_cast_fp16")]; + tensor var_41132_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4265_cast_fp16)[name = string("op_41132_cast_fp16")]; + tensor var_41133_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4267_cast_fp16)[name = string("op_41133_cast_fp16")]; + tensor var_41134_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4269_cast_fp16)[name = string("op_41134_cast_fp16")]; + tensor var_41135_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4271_cast_fp16)[name = string("op_41135_cast_fp16")]; + tensor var_41136_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4273_cast_fp16)[name = string("op_41136_cast_fp16")]; + tensor var_41137_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4275_cast_fp16)[name = string("op_41137_cast_fp16")]; + tensor var_41138_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4277_cast_fp16)[name = string("op_41138_cast_fp16")]; + tensor var_41139_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4279_cast_fp16)[name = string("op_41139_cast_fp16")]; + tensor var_41140_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4281_cast_fp16)[name = string("op_41140_cast_fp16")]; + tensor var_41141_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4283_cast_fp16)[name = string("op_41141_cast_fp16")]; + tensor var_41142_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4285_cast_fp16)[name = string("op_41142_cast_fp16")]; + tensor var_41143_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4287_cast_fp16)[name = string("op_41143_cast_fp16")]; + tensor var_41144_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4289_cast_fp16)[name = string("op_41144_cast_fp16")]; + tensor var_41145_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4291_cast_fp16)[name = string("op_41145_cast_fp16")]; + tensor var_41146_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4293_cast_fp16)[name = string("op_41146_cast_fp16")]; + tensor var_41147_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4295_cast_fp16)[name = string("op_41147_cast_fp16")]; + tensor var_41148_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4297_cast_fp16)[name = string("op_41148_cast_fp16")]; + tensor var_41149_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4299_cast_fp16)[name = string("op_41149_cast_fp16")]; + tensor var_41150_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4301_cast_fp16)[name = string("op_41150_cast_fp16")]; + tensor var_41151_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4303_cast_fp16)[name = string("op_41151_cast_fp16")]; + tensor var_41152_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4305_cast_fp16)[name = string("op_41152_cast_fp16")]; + tensor var_41153_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4307_cast_fp16)[name = string("op_41153_cast_fp16")]; + tensor var_41154_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4309_cast_fp16)[name = string("op_41154_cast_fp16")]; + tensor var_41155_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4311_cast_fp16)[name = string("op_41155_cast_fp16")]; + tensor var_41156_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4313_cast_fp16)[name = string("op_41156_cast_fp16")]; + tensor var_41157_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4315_cast_fp16)[name = string("op_41157_cast_fp16")]; + tensor var_41158_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4317_cast_fp16)[name = string("op_41158_cast_fp16")]; + tensor var_41159_cast_fp16 = softmax(axis = var_39905, x = aw_chunk_4319_cast_fp16)[name = string("op_41159_cast_fp16")]; + string var_41161_equation_0 = const()[name = string("op_41161_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41161_cast_fp16 = einsum(equation = var_41161_equation_0, values = (var_40681_cast_fp16, var_41080_cast_fp16))[name = string("op_41161_cast_fp16")]; + string var_41163_equation_0 = const()[name = string("op_41163_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41163_cast_fp16 = einsum(equation = var_41163_equation_0, values = (var_40681_cast_fp16, var_41081_cast_fp16))[name = string("op_41163_cast_fp16")]; + string var_41165_equation_0 = const()[name = string("op_41165_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41165_cast_fp16 = einsum(equation = var_41165_equation_0, values = (var_40681_cast_fp16, var_41082_cast_fp16))[name = string("op_41165_cast_fp16")]; + string var_41167_equation_0 = const()[name = string("op_41167_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41167_cast_fp16 = einsum(equation = var_41167_equation_0, values = (var_40681_cast_fp16, var_41083_cast_fp16))[name = string("op_41167_cast_fp16")]; + string var_41169_equation_0 = const()[name = string("op_41169_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41169_cast_fp16 = einsum(equation = var_41169_equation_0, values = (var_40685_cast_fp16, var_41084_cast_fp16))[name = string("op_41169_cast_fp16")]; + string var_41171_equation_0 = const()[name = string("op_41171_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41171_cast_fp16 = einsum(equation = var_41171_equation_0, values = (var_40685_cast_fp16, var_41085_cast_fp16))[name = string("op_41171_cast_fp16")]; + string var_41173_equation_0 = const()[name = string("op_41173_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41173_cast_fp16 = einsum(equation = var_41173_equation_0, values = (var_40685_cast_fp16, var_41086_cast_fp16))[name = string("op_41173_cast_fp16")]; + string var_41175_equation_0 = const()[name = string("op_41175_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41175_cast_fp16 = einsum(equation = var_41175_equation_0, values = (var_40685_cast_fp16, var_41087_cast_fp16))[name = string("op_41175_cast_fp16")]; + string var_41177_equation_0 = const()[name = string("op_41177_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41177_cast_fp16 = einsum(equation = var_41177_equation_0, values = (var_40689_cast_fp16, var_41088_cast_fp16))[name = string("op_41177_cast_fp16")]; + string var_41179_equation_0 = const()[name = string("op_41179_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41179_cast_fp16 = einsum(equation = var_41179_equation_0, values = (var_40689_cast_fp16, var_41089_cast_fp16))[name = string("op_41179_cast_fp16")]; + string var_41181_equation_0 = const()[name = string("op_41181_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41181_cast_fp16 = einsum(equation = var_41181_equation_0, values = (var_40689_cast_fp16, var_41090_cast_fp16))[name = string("op_41181_cast_fp16")]; + string var_41183_equation_0 = const()[name = string("op_41183_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41183_cast_fp16 = einsum(equation = var_41183_equation_0, values = (var_40689_cast_fp16, var_41091_cast_fp16))[name = string("op_41183_cast_fp16")]; + string var_41185_equation_0 = const()[name = string("op_41185_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41185_cast_fp16 = einsum(equation = var_41185_equation_0, values = (var_40693_cast_fp16, var_41092_cast_fp16))[name = string("op_41185_cast_fp16")]; + string var_41187_equation_0 = const()[name = string("op_41187_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41187_cast_fp16 = einsum(equation = var_41187_equation_0, values = (var_40693_cast_fp16, var_41093_cast_fp16))[name = string("op_41187_cast_fp16")]; + string var_41189_equation_0 = const()[name = string("op_41189_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41189_cast_fp16 = einsum(equation = var_41189_equation_0, values = (var_40693_cast_fp16, var_41094_cast_fp16))[name = string("op_41189_cast_fp16")]; + string var_41191_equation_0 = const()[name = string("op_41191_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41191_cast_fp16 = einsum(equation = var_41191_equation_0, values = (var_40693_cast_fp16, var_41095_cast_fp16))[name = string("op_41191_cast_fp16")]; + string var_41193_equation_0 = const()[name = string("op_41193_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41193_cast_fp16 = einsum(equation = var_41193_equation_0, values = (var_40697_cast_fp16, var_41096_cast_fp16))[name = string("op_41193_cast_fp16")]; + string var_41195_equation_0 = const()[name = string("op_41195_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41195_cast_fp16 = einsum(equation = var_41195_equation_0, values = (var_40697_cast_fp16, var_41097_cast_fp16))[name = string("op_41195_cast_fp16")]; + string var_41197_equation_0 = const()[name = string("op_41197_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41197_cast_fp16 = einsum(equation = var_41197_equation_0, values = (var_40697_cast_fp16, var_41098_cast_fp16))[name = string("op_41197_cast_fp16")]; + string var_41199_equation_0 = const()[name = string("op_41199_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41199_cast_fp16 = einsum(equation = var_41199_equation_0, values = (var_40697_cast_fp16, var_41099_cast_fp16))[name = string("op_41199_cast_fp16")]; + string var_41201_equation_0 = const()[name = string("op_41201_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41201_cast_fp16 = einsum(equation = var_41201_equation_0, values = (var_40701_cast_fp16, var_41100_cast_fp16))[name = string("op_41201_cast_fp16")]; + string var_41203_equation_0 = const()[name = string("op_41203_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41203_cast_fp16 = einsum(equation = var_41203_equation_0, values = (var_40701_cast_fp16, var_41101_cast_fp16))[name = string("op_41203_cast_fp16")]; + string var_41205_equation_0 = const()[name = string("op_41205_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41205_cast_fp16 = einsum(equation = var_41205_equation_0, values = (var_40701_cast_fp16, var_41102_cast_fp16))[name = string("op_41205_cast_fp16")]; + string var_41207_equation_0 = const()[name = string("op_41207_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41207_cast_fp16 = einsum(equation = var_41207_equation_0, values = (var_40701_cast_fp16, var_41103_cast_fp16))[name = string("op_41207_cast_fp16")]; + string var_41209_equation_0 = const()[name = string("op_41209_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41209_cast_fp16 = einsum(equation = var_41209_equation_0, values = (var_40705_cast_fp16, var_41104_cast_fp16))[name = string("op_41209_cast_fp16")]; + string var_41211_equation_0 = const()[name = string("op_41211_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41211_cast_fp16 = einsum(equation = var_41211_equation_0, values = (var_40705_cast_fp16, var_41105_cast_fp16))[name = string("op_41211_cast_fp16")]; + string var_41213_equation_0 = const()[name = string("op_41213_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41213_cast_fp16 = einsum(equation = var_41213_equation_0, values = (var_40705_cast_fp16, var_41106_cast_fp16))[name = string("op_41213_cast_fp16")]; + string var_41215_equation_0 = const()[name = string("op_41215_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41215_cast_fp16 = einsum(equation = var_41215_equation_0, values = (var_40705_cast_fp16, var_41107_cast_fp16))[name = string("op_41215_cast_fp16")]; + string var_41217_equation_0 = const()[name = string("op_41217_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41217_cast_fp16 = einsum(equation = var_41217_equation_0, values = (var_40709_cast_fp16, var_41108_cast_fp16))[name = string("op_41217_cast_fp16")]; + string var_41219_equation_0 = const()[name = string("op_41219_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41219_cast_fp16 = einsum(equation = var_41219_equation_0, values = (var_40709_cast_fp16, var_41109_cast_fp16))[name = string("op_41219_cast_fp16")]; + string var_41221_equation_0 = const()[name = string("op_41221_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41221_cast_fp16 = einsum(equation = var_41221_equation_0, values = (var_40709_cast_fp16, var_41110_cast_fp16))[name = string("op_41221_cast_fp16")]; + string var_41223_equation_0 = const()[name = string("op_41223_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41223_cast_fp16 = einsum(equation = var_41223_equation_0, values = (var_40709_cast_fp16, var_41111_cast_fp16))[name = string("op_41223_cast_fp16")]; + string var_41225_equation_0 = const()[name = string("op_41225_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41225_cast_fp16 = einsum(equation = var_41225_equation_0, values = (var_40713_cast_fp16, var_41112_cast_fp16))[name = string("op_41225_cast_fp16")]; + string var_41227_equation_0 = const()[name = string("op_41227_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41227_cast_fp16 = einsum(equation = var_41227_equation_0, values = (var_40713_cast_fp16, var_41113_cast_fp16))[name = string("op_41227_cast_fp16")]; + string var_41229_equation_0 = const()[name = string("op_41229_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41229_cast_fp16 = einsum(equation = var_41229_equation_0, values = (var_40713_cast_fp16, var_41114_cast_fp16))[name = string("op_41229_cast_fp16")]; + string var_41231_equation_0 = const()[name = string("op_41231_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41231_cast_fp16 = einsum(equation = var_41231_equation_0, values = (var_40713_cast_fp16, var_41115_cast_fp16))[name = string("op_41231_cast_fp16")]; + string var_41233_equation_0 = const()[name = string("op_41233_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41233_cast_fp16 = einsum(equation = var_41233_equation_0, values = (var_40717_cast_fp16, var_41116_cast_fp16))[name = string("op_41233_cast_fp16")]; + string var_41235_equation_0 = const()[name = string("op_41235_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41235_cast_fp16 = einsum(equation = var_41235_equation_0, values = (var_40717_cast_fp16, var_41117_cast_fp16))[name = string("op_41235_cast_fp16")]; + string var_41237_equation_0 = const()[name = string("op_41237_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41237_cast_fp16 = einsum(equation = var_41237_equation_0, values = (var_40717_cast_fp16, var_41118_cast_fp16))[name = string("op_41237_cast_fp16")]; + string var_41239_equation_0 = const()[name = string("op_41239_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41239_cast_fp16 = einsum(equation = var_41239_equation_0, values = (var_40717_cast_fp16, var_41119_cast_fp16))[name = string("op_41239_cast_fp16")]; + string var_41241_equation_0 = const()[name = string("op_41241_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41241_cast_fp16 = einsum(equation = var_41241_equation_0, values = (var_40721_cast_fp16, var_41120_cast_fp16))[name = string("op_41241_cast_fp16")]; + string var_41243_equation_0 = const()[name = string("op_41243_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41243_cast_fp16 = einsum(equation = var_41243_equation_0, values = (var_40721_cast_fp16, var_41121_cast_fp16))[name = string("op_41243_cast_fp16")]; + string var_41245_equation_0 = const()[name = string("op_41245_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41245_cast_fp16 = einsum(equation = var_41245_equation_0, values = (var_40721_cast_fp16, var_41122_cast_fp16))[name = string("op_41245_cast_fp16")]; + string var_41247_equation_0 = const()[name = string("op_41247_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41247_cast_fp16 = einsum(equation = var_41247_equation_0, values = (var_40721_cast_fp16, var_41123_cast_fp16))[name = string("op_41247_cast_fp16")]; + string var_41249_equation_0 = const()[name = string("op_41249_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41249_cast_fp16 = einsum(equation = var_41249_equation_0, values = (var_40725_cast_fp16, var_41124_cast_fp16))[name = string("op_41249_cast_fp16")]; + string var_41251_equation_0 = const()[name = string("op_41251_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41251_cast_fp16 = einsum(equation = var_41251_equation_0, values = (var_40725_cast_fp16, var_41125_cast_fp16))[name = string("op_41251_cast_fp16")]; + string var_41253_equation_0 = const()[name = string("op_41253_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41253_cast_fp16 = einsum(equation = var_41253_equation_0, values = (var_40725_cast_fp16, var_41126_cast_fp16))[name = string("op_41253_cast_fp16")]; + string var_41255_equation_0 = const()[name = string("op_41255_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41255_cast_fp16 = einsum(equation = var_41255_equation_0, values = (var_40725_cast_fp16, var_41127_cast_fp16))[name = string("op_41255_cast_fp16")]; + string var_41257_equation_0 = const()[name = string("op_41257_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41257_cast_fp16 = einsum(equation = var_41257_equation_0, values = (var_40729_cast_fp16, var_41128_cast_fp16))[name = string("op_41257_cast_fp16")]; + string var_41259_equation_0 = const()[name = string("op_41259_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41259_cast_fp16 = einsum(equation = var_41259_equation_0, values = (var_40729_cast_fp16, var_41129_cast_fp16))[name = string("op_41259_cast_fp16")]; + string var_41261_equation_0 = const()[name = string("op_41261_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41261_cast_fp16 = einsum(equation = var_41261_equation_0, values = (var_40729_cast_fp16, var_41130_cast_fp16))[name = string("op_41261_cast_fp16")]; + string var_41263_equation_0 = const()[name = string("op_41263_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41263_cast_fp16 = einsum(equation = var_41263_equation_0, values = (var_40729_cast_fp16, var_41131_cast_fp16))[name = string("op_41263_cast_fp16")]; + string var_41265_equation_0 = const()[name = string("op_41265_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41265_cast_fp16 = einsum(equation = var_41265_equation_0, values = (var_40733_cast_fp16, var_41132_cast_fp16))[name = string("op_41265_cast_fp16")]; + string var_41267_equation_0 = const()[name = string("op_41267_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41267_cast_fp16 = einsum(equation = var_41267_equation_0, values = (var_40733_cast_fp16, var_41133_cast_fp16))[name = string("op_41267_cast_fp16")]; + string var_41269_equation_0 = const()[name = string("op_41269_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41269_cast_fp16 = einsum(equation = var_41269_equation_0, values = (var_40733_cast_fp16, var_41134_cast_fp16))[name = string("op_41269_cast_fp16")]; + string var_41271_equation_0 = const()[name = string("op_41271_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41271_cast_fp16 = einsum(equation = var_41271_equation_0, values = (var_40733_cast_fp16, var_41135_cast_fp16))[name = string("op_41271_cast_fp16")]; + string var_41273_equation_0 = const()[name = string("op_41273_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41273_cast_fp16 = einsum(equation = var_41273_equation_0, values = (var_40737_cast_fp16, var_41136_cast_fp16))[name = string("op_41273_cast_fp16")]; + string var_41275_equation_0 = const()[name = string("op_41275_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41275_cast_fp16 = einsum(equation = var_41275_equation_0, values = (var_40737_cast_fp16, var_41137_cast_fp16))[name = string("op_41275_cast_fp16")]; + string var_41277_equation_0 = const()[name = string("op_41277_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41277_cast_fp16 = einsum(equation = var_41277_equation_0, values = (var_40737_cast_fp16, var_41138_cast_fp16))[name = string("op_41277_cast_fp16")]; + string var_41279_equation_0 = const()[name = string("op_41279_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41279_cast_fp16 = einsum(equation = var_41279_equation_0, values = (var_40737_cast_fp16, var_41139_cast_fp16))[name = string("op_41279_cast_fp16")]; + string var_41281_equation_0 = const()[name = string("op_41281_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41281_cast_fp16 = einsum(equation = var_41281_equation_0, values = (var_40741_cast_fp16, var_41140_cast_fp16))[name = string("op_41281_cast_fp16")]; + string var_41283_equation_0 = const()[name = string("op_41283_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41283_cast_fp16 = einsum(equation = var_41283_equation_0, values = (var_40741_cast_fp16, var_41141_cast_fp16))[name = string("op_41283_cast_fp16")]; + string var_41285_equation_0 = const()[name = string("op_41285_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41285_cast_fp16 = einsum(equation = var_41285_equation_0, values = (var_40741_cast_fp16, var_41142_cast_fp16))[name = string("op_41285_cast_fp16")]; + string var_41287_equation_0 = const()[name = string("op_41287_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41287_cast_fp16 = einsum(equation = var_41287_equation_0, values = (var_40741_cast_fp16, var_41143_cast_fp16))[name = string("op_41287_cast_fp16")]; + string var_41289_equation_0 = const()[name = string("op_41289_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41289_cast_fp16 = einsum(equation = var_41289_equation_0, values = (var_40745_cast_fp16, var_41144_cast_fp16))[name = string("op_41289_cast_fp16")]; + string var_41291_equation_0 = const()[name = string("op_41291_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41291_cast_fp16 = einsum(equation = var_41291_equation_0, values = (var_40745_cast_fp16, var_41145_cast_fp16))[name = string("op_41291_cast_fp16")]; + string var_41293_equation_0 = const()[name = string("op_41293_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41293_cast_fp16 = einsum(equation = var_41293_equation_0, values = (var_40745_cast_fp16, var_41146_cast_fp16))[name = string("op_41293_cast_fp16")]; + string var_41295_equation_0 = const()[name = string("op_41295_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41295_cast_fp16 = einsum(equation = var_41295_equation_0, values = (var_40745_cast_fp16, var_41147_cast_fp16))[name = string("op_41295_cast_fp16")]; + string var_41297_equation_0 = const()[name = string("op_41297_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41297_cast_fp16 = einsum(equation = var_41297_equation_0, values = (var_40749_cast_fp16, var_41148_cast_fp16))[name = string("op_41297_cast_fp16")]; + string var_41299_equation_0 = const()[name = string("op_41299_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41299_cast_fp16 = einsum(equation = var_41299_equation_0, values = (var_40749_cast_fp16, var_41149_cast_fp16))[name = string("op_41299_cast_fp16")]; + string var_41301_equation_0 = const()[name = string("op_41301_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41301_cast_fp16 = einsum(equation = var_41301_equation_0, values = (var_40749_cast_fp16, var_41150_cast_fp16))[name = string("op_41301_cast_fp16")]; + string var_41303_equation_0 = const()[name = string("op_41303_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41303_cast_fp16 = einsum(equation = var_41303_equation_0, values = (var_40749_cast_fp16, var_41151_cast_fp16))[name = string("op_41303_cast_fp16")]; + string var_41305_equation_0 = const()[name = string("op_41305_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41305_cast_fp16 = einsum(equation = var_41305_equation_0, values = (var_40753_cast_fp16, var_41152_cast_fp16))[name = string("op_41305_cast_fp16")]; + string var_41307_equation_0 = const()[name = string("op_41307_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41307_cast_fp16 = einsum(equation = var_41307_equation_0, values = (var_40753_cast_fp16, var_41153_cast_fp16))[name = string("op_41307_cast_fp16")]; + string var_41309_equation_0 = const()[name = string("op_41309_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41309_cast_fp16 = einsum(equation = var_41309_equation_0, values = (var_40753_cast_fp16, var_41154_cast_fp16))[name = string("op_41309_cast_fp16")]; + string var_41311_equation_0 = const()[name = string("op_41311_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41311_cast_fp16 = einsum(equation = var_41311_equation_0, values = (var_40753_cast_fp16, var_41155_cast_fp16))[name = string("op_41311_cast_fp16")]; + string var_41313_equation_0 = const()[name = string("op_41313_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41313_cast_fp16 = einsum(equation = var_41313_equation_0, values = (var_40757_cast_fp16, var_41156_cast_fp16))[name = string("op_41313_cast_fp16")]; + string var_41315_equation_0 = const()[name = string("op_41315_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41315_cast_fp16 = einsum(equation = var_41315_equation_0, values = (var_40757_cast_fp16, var_41157_cast_fp16))[name = string("op_41315_cast_fp16")]; + string var_41317_equation_0 = const()[name = string("op_41317_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41317_cast_fp16 = einsum(equation = var_41317_equation_0, values = (var_40757_cast_fp16, var_41158_cast_fp16))[name = string("op_41317_cast_fp16")]; + string var_41319_equation_0 = const()[name = string("op_41319_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41319_cast_fp16 = einsum(equation = var_41319_equation_0, values = (var_40757_cast_fp16, var_41159_cast_fp16))[name = string("op_41319_cast_fp16")]; + bool var_41321_interleave_0 = const()[name = string("op_41321_interleave_0"), val = bool(false)]; + tensor var_41321_cast_fp16 = concat(axis = var_39880, interleave = var_41321_interleave_0, values = (var_41161_cast_fp16, var_41163_cast_fp16, var_41165_cast_fp16, var_41167_cast_fp16))[name = string("op_41321_cast_fp16")]; + bool var_41323_interleave_0 = const()[name = string("op_41323_interleave_0"), val = bool(false)]; + tensor var_41323_cast_fp16 = concat(axis = var_39880, interleave = var_41323_interleave_0, values = (var_41169_cast_fp16, var_41171_cast_fp16, var_41173_cast_fp16, var_41175_cast_fp16))[name = string("op_41323_cast_fp16")]; + bool var_41325_interleave_0 = const()[name = string("op_41325_interleave_0"), val = bool(false)]; + tensor var_41325_cast_fp16 = concat(axis = var_39880, interleave = var_41325_interleave_0, values = (var_41177_cast_fp16, var_41179_cast_fp16, var_41181_cast_fp16, var_41183_cast_fp16))[name = string("op_41325_cast_fp16")]; + bool var_41327_interleave_0 = const()[name = string("op_41327_interleave_0"), val = bool(false)]; + tensor var_41327_cast_fp16 = concat(axis = var_39880, interleave = var_41327_interleave_0, values = (var_41185_cast_fp16, var_41187_cast_fp16, var_41189_cast_fp16, var_41191_cast_fp16))[name = string("op_41327_cast_fp16")]; + bool var_41329_interleave_0 = const()[name = string("op_41329_interleave_0"), val = bool(false)]; + tensor var_41329_cast_fp16 = concat(axis = var_39880, interleave = var_41329_interleave_0, values = (var_41193_cast_fp16, var_41195_cast_fp16, var_41197_cast_fp16, var_41199_cast_fp16))[name = string("op_41329_cast_fp16")]; + bool var_41331_interleave_0 = const()[name = string("op_41331_interleave_0"), val = bool(false)]; + tensor var_41331_cast_fp16 = concat(axis = var_39880, interleave = var_41331_interleave_0, values = (var_41201_cast_fp16, var_41203_cast_fp16, var_41205_cast_fp16, var_41207_cast_fp16))[name = string("op_41331_cast_fp16")]; + bool var_41333_interleave_0 = const()[name = string("op_41333_interleave_0"), val = bool(false)]; + tensor var_41333_cast_fp16 = concat(axis = var_39880, interleave = var_41333_interleave_0, values = (var_41209_cast_fp16, var_41211_cast_fp16, var_41213_cast_fp16, var_41215_cast_fp16))[name = string("op_41333_cast_fp16")]; + bool var_41335_interleave_0 = const()[name = string("op_41335_interleave_0"), val = bool(false)]; + tensor var_41335_cast_fp16 = concat(axis = var_39880, interleave = var_41335_interleave_0, values = (var_41217_cast_fp16, var_41219_cast_fp16, var_41221_cast_fp16, var_41223_cast_fp16))[name = string("op_41335_cast_fp16")]; + bool var_41337_interleave_0 = const()[name = string("op_41337_interleave_0"), val = bool(false)]; + tensor var_41337_cast_fp16 = concat(axis = var_39880, interleave = var_41337_interleave_0, values = (var_41225_cast_fp16, var_41227_cast_fp16, var_41229_cast_fp16, var_41231_cast_fp16))[name = string("op_41337_cast_fp16")]; + bool var_41339_interleave_0 = const()[name = string("op_41339_interleave_0"), val = bool(false)]; + tensor var_41339_cast_fp16 = concat(axis = var_39880, interleave = var_41339_interleave_0, values = (var_41233_cast_fp16, var_41235_cast_fp16, var_41237_cast_fp16, var_41239_cast_fp16))[name = string("op_41339_cast_fp16")]; + bool var_41341_interleave_0 = const()[name = string("op_41341_interleave_0"), val = bool(false)]; + tensor var_41341_cast_fp16 = concat(axis = var_39880, interleave = var_41341_interleave_0, values = (var_41241_cast_fp16, var_41243_cast_fp16, var_41245_cast_fp16, var_41247_cast_fp16))[name = string("op_41341_cast_fp16")]; + bool var_41343_interleave_0 = const()[name = string("op_41343_interleave_0"), val = bool(false)]; + tensor var_41343_cast_fp16 = concat(axis = var_39880, interleave = var_41343_interleave_0, values = (var_41249_cast_fp16, var_41251_cast_fp16, var_41253_cast_fp16, var_41255_cast_fp16))[name = string("op_41343_cast_fp16")]; + bool var_41345_interleave_0 = const()[name = string("op_41345_interleave_0"), val = bool(false)]; + tensor var_41345_cast_fp16 = concat(axis = var_39880, interleave = var_41345_interleave_0, values = (var_41257_cast_fp16, var_41259_cast_fp16, var_41261_cast_fp16, var_41263_cast_fp16))[name = string("op_41345_cast_fp16")]; + bool var_41347_interleave_0 = const()[name = string("op_41347_interleave_0"), val = bool(false)]; + tensor var_41347_cast_fp16 = concat(axis = var_39880, interleave = var_41347_interleave_0, values = (var_41265_cast_fp16, var_41267_cast_fp16, var_41269_cast_fp16, var_41271_cast_fp16))[name = string("op_41347_cast_fp16")]; + bool var_41349_interleave_0 = const()[name = string("op_41349_interleave_0"), val = bool(false)]; + tensor var_41349_cast_fp16 = concat(axis = var_39880, interleave = var_41349_interleave_0, values = (var_41273_cast_fp16, var_41275_cast_fp16, var_41277_cast_fp16, var_41279_cast_fp16))[name = string("op_41349_cast_fp16")]; + bool var_41351_interleave_0 = const()[name = string("op_41351_interleave_0"), val = bool(false)]; + tensor var_41351_cast_fp16 = concat(axis = var_39880, interleave = var_41351_interleave_0, values = (var_41281_cast_fp16, var_41283_cast_fp16, var_41285_cast_fp16, var_41287_cast_fp16))[name = string("op_41351_cast_fp16")]; + bool var_41353_interleave_0 = const()[name = string("op_41353_interleave_0"), val = bool(false)]; + tensor var_41353_cast_fp16 = concat(axis = var_39880, interleave = var_41353_interleave_0, values = (var_41289_cast_fp16, var_41291_cast_fp16, var_41293_cast_fp16, var_41295_cast_fp16))[name = string("op_41353_cast_fp16")]; + bool var_41355_interleave_0 = const()[name = string("op_41355_interleave_0"), val = bool(false)]; + tensor var_41355_cast_fp16 = concat(axis = var_39880, interleave = var_41355_interleave_0, values = (var_41297_cast_fp16, var_41299_cast_fp16, var_41301_cast_fp16, var_41303_cast_fp16))[name = string("op_41355_cast_fp16")]; + bool var_41357_interleave_0 = const()[name = string("op_41357_interleave_0"), val = bool(false)]; + tensor var_41357_cast_fp16 = concat(axis = var_39880, interleave = var_41357_interleave_0, values = (var_41305_cast_fp16, var_41307_cast_fp16, var_41309_cast_fp16, var_41311_cast_fp16))[name = string("op_41357_cast_fp16")]; + bool var_41359_interleave_0 = const()[name = string("op_41359_interleave_0"), val = bool(false)]; + tensor var_41359_cast_fp16 = concat(axis = var_39880, interleave = var_41359_interleave_0, values = (var_41313_cast_fp16, var_41315_cast_fp16, var_41317_cast_fp16, var_41319_cast_fp16))[name = string("op_41359_cast_fp16")]; + bool input_209_interleave_0 = const()[name = string("input_209_interleave_0"), val = bool(false)]; + tensor input_209_cast_fp16 = concat(axis = var_39905, interleave = input_209_interleave_0, values = (var_41321_cast_fp16, var_41323_cast_fp16, var_41325_cast_fp16, var_41327_cast_fp16, var_41329_cast_fp16, var_41331_cast_fp16, var_41333_cast_fp16, var_41335_cast_fp16, var_41337_cast_fp16, var_41339_cast_fp16, var_41341_cast_fp16, var_41343_cast_fp16, var_41345_cast_fp16, var_41347_cast_fp16, var_41349_cast_fp16, var_41351_cast_fp16, var_41353_cast_fp16, var_41355_cast_fp16, var_41357_cast_fp16, var_41359_cast_fp16))[name = string("input_209_cast_fp16")]; + string obj_107_pad_type_0 = const()[name = string("obj_107_pad_type_0"), val = string("valid")]; + tensor obj_107_strides_0 = const()[name = string("obj_107_strides_0"), val = tensor([1, 1])]; + tensor obj_107_pad_0 = const()[name = string("obj_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_107_dilations_0 = const()[name = string("obj_107_dilations_0"), val = tensor([1, 1])]; + int32 obj_107_groups_0 = const()[name = string("obj_107_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047690560)))]; + tensor layers_26_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050967424)))]; + tensor obj_107_cast_fp16 = conv(bias = layers_26_self_attn_o_proj_bias_to_fp16, dilations = obj_107_dilations_0, groups = obj_107_groups_0, pad = obj_107_pad_0, pad_type = obj_107_pad_type_0, strides = obj_107_strides_0, weight = layers_26_self_attn_o_proj_weight_to_fp16, x = input_209_cast_fp16)[name = string("obj_107_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_107_cast_fp16)[name = string("inputs_107_cast_fp16")]; + tensor out_107_axes_0 = const()[name = string("out_107_axes_0"), val = tensor([1])]; + fp16 var_41378_to_fp16 = const()[name = string("op_41378_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_41378_to_fp16, x = inputs_107_cast_fp16)[name = string("out_107_cast_fp16")]; + tensor input_211_gamma_0_to_fp16 = const()[name = string("input_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050970048)))]; + tensor input_211_beta_0_to_fp16 = const()[name = string("input_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050972672)))]; + fp16 input_211_epsilon_0_to_fp16 = const()[name = string("input_211_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_211_cast_fp16 = batch_norm(beta = input_211_beta_0_to_fp16, epsilon = input_211_epsilon_0_to_fp16, gamma = input_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = string("input_211_cast_fp16")]; + string input_213_pad_type_0 = const()[name = string("input_213_pad_type_0"), val = string("valid")]; + tensor input_213_strides_0 = const()[name = string("input_213_strides_0"), val = tensor([1, 1])]; + tensor input_213_pad_0 = const()[name = string("input_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_213_dilations_0 = const()[name = string("input_213_dilations_0"), val = tensor([1, 1])]; + int32 input_213_groups_0 = const()[name = string("input_213_groups_0"), val = int32(1)]; + tensor layers_26_fc1_weight_to_fp16 = const()[name = string("layers_26_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050975296)))]; + tensor layers_26_fc1_bias_to_fp16 = const()[name = string("layers_26_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064082560)))]; + tensor input_213_cast_fp16 = conv(bias = layers_26_fc1_bias_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = layers_26_fc1_weight_to_fp16, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")]; + string input_215_mode_0 = const()[name = string("input_215_mode_0"), val = string("EXACT")]; + tensor input_215_cast_fp16 = gelu(mode = input_215_mode_0, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")]; + string hidden_states_57_pad_type_0 = const()[name = string("hidden_states_57_pad_type_0"), val = string("valid")]; + tensor hidden_states_57_strides_0 = const()[name = string("hidden_states_57_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_57_pad_0 = const()[name = string("hidden_states_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_57_dilations_0 = const()[name = string("hidden_states_57_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_57_groups_0 = const()[name = string("hidden_states_57_groups_0"), val = int32(1)]; + tensor layers_26_fc2_weight_to_fp16 = const()[name = string("layers_26_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064092864)))]; + tensor layers_26_fc2_bias_to_fp16 = const()[name = string("layers_26_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077200128)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = layers_26_fc2_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = layers_26_fc2_weight_to_fp16, x = input_215_cast_fp16)[name = string("hidden_states_57_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_57_cast_fp16)[name = string("inputs_109_cast_fp16")]; + int32 var_41407 = const()[name = string("op_41407"), val = int32(3)]; + int32 var_41432 = const()[name = string("op_41432"), val = int32(1)]; + tensor out_109_axes_0 = const()[name = string("out_109_axes_0"), val = tensor([1])]; + fp16 var_41449_to_fp16 = const()[name = string("op_41449_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_41449_to_fp16, x = inputs_109_cast_fp16)[name = string("out_109_cast_fp16")]; + tensor obj_109_gamma_0_to_fp16 = const()[name = string("obj_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077202752)))]; + tensor obj_109_beta_0_to_fp16 = const()[name = string("obj_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077205376)))]; + fp16 obj_109_epsilon_0_to_fp16 = const()[name = string("obj_109_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_109_cast_fp16 = batch_norm(beta = obj_109_beta_0_to_fp16, epsilon = obj_109_epsilon_0_to_fp16, gamma = obj_109_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = string("obj_109_cast_fp16")]; + string query_55_pad_type_0 = const()[name = string("query_55_pad_type_0"), val = string("valid")]; + tensor query_55_strides_0 = const()[name = string("query_55_strides_0"), val = tensor([1, 1])]; + tensor query_55_pad_0 = const()[name = string("query_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_55_dilations_0 = const()[name = string("query_55_dilations_0"), val = tensor([1, 1])]; + int32 query_55_groups_0 = const()[name = string("query_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077208000)))]; + tensor layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080484864)))]; + tensor query_55_cast_fp16 = conv(bias = layers_27_self_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_27_self_attn_q_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("query_55_cast_fp16")]; + string key_55_pad_type_0 = const()[name = string("key_55_pad_type_0"), val = string("valid")]; + tensor key_55_strides_0 = const()[name = string("key_55_strides_0"), val = tensor([1, 1])]; + tensor key_55_pad_0 = const()[name = string("key_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_55_dilations_0 = const()[name = string("key_55_dilations_0"), val = tensor([1, 1])]; + int32 key_55_groups_0 = const()[name = string("key_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080487488)))]; + tensor key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_27_self_attn_k_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("key_55_cast_fp16")]; + string value_55_pad_type_0 = const()[name = string("value_55_pad_type_0"), val = string("valid")]; + tensor value_55_strides_0 = const()[name = string("value_55_strides_0"), val = tensor([1, 1])]; + tensor value_55_pad_0 = const()[name = string("value_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_55_dilations_0 = const()[name = string("value_55_dilations_0"), val = tensor([1, 1])]; + int32 value_55_groups_0 = const()[name = string("value_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1083764352)))]; + tensor layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087041216)))]; + tensor value_55_cast_fp16 = conv(bias = layers_27_self_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_27_self_attn_v_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("value_55_cast_fp16")]; + tensor var_41487_begin_0 = const()[name = string("op_41487_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41487_end_0 = const()[name = string("op_41487_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41487_end_mask_0 = const()[name = string("op_41487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41487_cast_fp16 = slice_by_index(begin = var_41487_begin_0, end = var_41487_end_0, end_mask = var_41487_end_mask_0, x = query_55_cast_fp16)[name = string("op_41487_cast_fp16")]; + tensor var_41491_begin_0 = const()[name = string("op_41491_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41491_end_0 = const()[name = string("op_41491_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_41491_end_mask_0 = const()[name = string("op_41491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41491_cast_fp16 = slice_by_index(begin = var_41491_begin_0, end = var_41491_end_0, end_mask = var_41491_end_mask_0, x = query_55_cast_fp16)[name = string("op_41491_cast_fp16")]; + tensor var_41495_begin_0 = const()[name = string("op_41495_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41495_end_0 = const()[name = string("op_41495_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_41495_end_mask_0 = const()[name = string("op_41495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41495_cast_fp16 = slice_by_index(begin = var_41495_begin_0, end = var_41495_end_0, end_mask = var_41495_end_mask_0, x = query_55_cast_fp16)[name = string("op_41495_cast_fp16")]; + tensor var_41499_begin_0 = const()[name = string("op_41499_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41499_end_0 = const()[name = string("op_41499_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_41499_end_mask_0 = const()[name = string("op_41499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41499_cast_fp16 = slice_by_index(begin = var_41499_begin_0, end = var_41499_end_0, end_mask = var_41499_end_mask_0, x = query_55_cast_fp16)[name = string("op_41499_cast_fp16")]; + tensor var_41503_begin_0 = const()[name = string("op_41503_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41503_end_0 = const()[name = string("op_41503_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_41503_end_mask_0 = const()[name = string("op_41503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41503_cast_fp16 = slice_by_index(begin = var_41503_begin_0, end = var_41503_end_0, end_mask = var_41503_end_mask_0, x = query_55_cast_fp16)[name = string("op_41503_cast_fp16")]; + tensor var_41507_begin_0 = const()[name = string("op_41507_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41507_end_0 = const()[name = string("op_41507_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_41507_end_mask_0 = const()[name = string("op_41507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41507_cast_fp16 = slice_by_index(begin = var_41507_begin_0, end = var_41507_end_0, end_mask = var_41507_end_mask_0, x = query_55_cast_fp16)[name = string("op_41507_cast_fp16")]; + tensor var_41511_begin_0 = const()[name = string("op_41511_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41511_end_0 = const()[name = string("op_41511_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_41511_end_mask_0 = const()[name = string("op_41511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41511_cast_fp16 = slice_by_index(begin = var_41511_begin_0, end = var_41511_end_0, end_mask = var_41511_end_mask_0, x = query_55_cast_fp16)[name = string("op_41511_cast_fp16")]; + tensor var_41515_begin_0 = const()[name = string("op_41515_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41515_end_0 = const()[name = string("op_41515_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_41515_end_mask_0 = const()[name = string("op_41515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41515_cast_fp16 = slice_by_index(begin = var_41515_begin_0, end = var_41515_end_0, end_mask = var_41515_end_mask_0, x = query_55_cast_fp16)[name = string("op_41515_cast_fp16")]; + tensor var_41519_begin_0 = const()[name = string("op_41519_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41519_end_0 = const()[name = string("op_41519_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_41519_end_mask_0 = const()[name = string("op_41519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41519_cast_fp16 = slice_by_index(begin = var_41519_begin_0, end = var_41519_end_0, end_mask = var_41519_end_mask_0, x = query_55_cast_fp16)[name = string("op_41519_cast_fp16")]; + tensor var_41523_begin_0 = const()[name = string("op_41523_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41523_end_0 = const()[name = string("op_41523_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_41523_end_mask_0 = const()[name = string("op_41523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41523_cast_fp16 = slice_by_index(begin = var_41523_begin_0, end = var_41523_end_0, end_mask = var_41523_end_mask_0, x = query_55_cast_fp16)[name = string("op_41523_cast_fp16")]; + tensor var_41527_begin_0 = const()[name = string("op_41527_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41527_end_0 = const()[name = string("op_41527_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_41527_end_mask_0 = const()[name = string("op_41527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41527_cast_fp16 = slice_by_index(begin = var_41527_begin_0, end = var_41527_end_0, end_mask = var_41527_end_mask_0, x = query_55_cast_fp16)[name = string("op_41527_cast_fp16")]; + tensor var_41531_begin_0 = const()[name = string("op_41531_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41531_end_0 = const()[name = string("op_41531_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_41531_end_mask_0 = const()[name = string("op_41531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41531_cast_fp16 = slice_by_index(begin = var_41531_begin_0, end = var_41531_end_0, end_mask = var_41531_end_mask_0, x = query_55_cast_fp16)[name = string("op_41531_cast_fp16")]; + tensor var_41535_begin_0 = const()[name = string("op_41535_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41535_end_0 = const()[name = string("op_41535_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_41535_end_mask_0 = const()[name = string("op_41535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41535_cast_fp16 = slice_by_index(begin = var_41535_begin_0, end = var_41535_end_0, end_mask = var_41535_end_mask_0, x = query_55_cast_fp16)[name = string("op_41535_cast_fp16")]; + tensor var_41539_begin_0 = const()[name = string("op_41539_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41539_end_0 = const()[name = string("op_41539_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_41539_end_mask_0 = const()[name = string("op_41539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41539_cast_fp16 = slice_by_index(begin = var_41539_begin_0, end = var_41539_end_0, end_mask = var_41539_end_mask_0, x = query_55_cast_fp16)[name = string("op_41539_cast_fp16")]; + tensor var_41543_begin_0 = const()[name = string("op_41543_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41543_end_0 = const()[name = string("op_41543_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_41543_end_mask_0 = const()[name = string("op_41543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41543_cast_fp16 = slice_by_index(begin = var_41543_begin_0, end = var_41543_end_0, end_mask = var_41543_end_mask_0, x = query_55_cast_fp16)[name = string("op_41543_cast_fp16")]; + tensor var_41547_begin_0 = const()[name = string("op_41547_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41547_end_0 = const()[name = string("op_41547_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_41547_end_mask_0 = const()[name = string("op_41547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41547_cast_fp16 = slice_by_index(begin = var_41547_begin_0, end = var_41547_end_0, end_mask = var_41547_end_mask_0, x = query_55_cast_fp16)[name = string("op_41547_cast_fp16")]; + tensor var_41551_begin_0 = const()[name = string("op_41551_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41551_end_0 = const()[name = string("op_41551_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_41551_end_mask_0 = const()[name = string("op_41551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41551_cast_fp16 = slice_by_index(begin = var_41551_begin_0, end = var_41551_end_0, end_mask = var_41551_end_mask_0, x = query_55_cast_fp16)[name = string("op_41551_cast_fp16")]; + tensor var_41555_begin_0 = const()[name = string("op_41555_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41555_end_0 = const()[name = string("op_41555_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_41555_end_mask_0 = const()[name = string("op_41555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41555_cast_fp16 = slice_by_index(begin = var_41555_begin_0, end = var_41555_end_0, end_mask = var_41555_end_mask_0, x = query_55_cast_fp16)[name = string("op_41555_cast_fp16")]; + tensor var_41559_begin_0 = const()[name = string("op_41559_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41559_end_0 = const()[name = string("op_41559_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_41559_end_mask_0 = const()[name = string("op_41559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41559_cast_fp16 = slice_by_index(begin = var_41559_begin_0, end = var_41559_end_0, end_mask = var_41559_end_mask_0, x = query_55_cast_fp16)[name = string("op_41559_cast_fp16")]; + tensor var_41563_begin_0 = const()[name = string("op_41563_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41563_end_0 = const()[name = string("op_41563_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_41563_end_mask_0 = const()[name = string("op_41563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41563_cast_fp16 = slice_by_index(begin = var_41563_begin_0, end = var_41563_end_0, end_mask = var_41563_end_mask_0, x = query_55_cast_fp16)[name = string("op_41563_cast_fp16")]; + tensor var_41572_begin_0 = const()[name = string("op_41572_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41572_end_0 = const()[name = string("op_41572_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41572_end_mask_0 = const()[name = string("op_41572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41572_cast_fp16 = slice_by_index(begin = var_41572_begin_0, end = var_41572_end_0, end_mask = var_41572_end_mask_0, x = var_41487_cast_fp16)[name = string("op_41572_cast_fp16")]; + tensor var_41579_begin_0 = const()[name = string("op_41579_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41579_end_0 = const()[name = string("op_41579_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41579_end_mask_0 = const()[name = string("op_41579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41579_cast_fp16 = slice_by_index(begin = var_41579_begin_0, end = var_41579_end_0, end_mask = var_41579_end_mask_0, x = var_41487_cast_fp16)[name = string("op_41579_cast_fp16")]; + tensor var_41586_begin_0 = const()[name = string("op_41586_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41586_end_0 = const()[name = string("op_41586_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41586_end_mask_0 = const()[name = string("op_41586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41586_cast_fp16 = slice_by_index(begin = var_41586_begin_0, end = var_41586_end_0, end_mask = var_41586_end_mask_0, x = var_41487_cast_fp16)[name = string("op_41586_cast_fp16")]; + tensor var_41593_begin_0 = const()[name = string("op_41593_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41593_end_0 = const()[name = string("op_41593_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41593_end_mask_0 = const()[name = string("op_41593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41593_cast_fp16 = slice_by_index(begin = var_41593_begin_0, end = var_41593_end_0, end_mask = var_41593_end_mask_0, x = var_41487_cast_fp16)[name = string("op_41593_cast_fp16")]; + tensor var_41600_begin_0 = const()[name = string("op_41600_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41600_end_0 = const()[name = string("op_41600_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41600_end_mask_0 = const()[name = string("op_41600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41600_cast_fp16 = slice_by_index(begin = var_41600_begin_0, end = var_41600_end_0, end_mask = var_41600_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41600_cast_fp16")]; + tensor var_41607_begin_0 = const()[name = string("op_41607_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41607_end_0 = const()[name = string("op_41607_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41607_end_mask_0 = const()[name = string("op_41607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41607_cast_fp16 = slice_by_index(begin = var_41607_begin_0, end = var_41607_end_0, end_mask = var_41607_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41607_cast_fp16")]; + tensor var_41614_begin_0 = const()[name = string("op_41614_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41614_end_0 = const()[name = string("op_41614_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41614_end_mask_0 = const()[name = string("op_41614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41614_cast_fp16 = slice_by_index(begin = var_41614_begin_0, end = var_41614_end_0, end_mask = var_41614_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41614_cast_fp16")]; + tensor var_41621_begin_0 = const()[name = string("op_41621_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41621_end_0 = const()[name = string("op_41621_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41621_end_mask_0 = const()[name = string("op_41621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41621_cast_fp16 = slice_by_index(begin = var_41621_begin_0, end = var_41621_end_0, end_mask = var_41621_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41621_cast_fp16")]; + tensor var_41628_begin_0 = const()[name = string("op_41628_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41628_end_0 = const()[name = string("op_41628_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41628_end_mask_0 = const()[name = string("op_41628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41628_cast_fp16 = slice_by_index(begin = var_41628_begin_0, end = var_41628_end_0, end_mask = var_41628_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41628_cast_fp16")]; + tensor var_41635_begin_0 = const()[name = string("op_41635_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41635_end_0 = const()[name = string("op_41635_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41635_end_mask_0 = const()[name = string("op_41635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41635_cast_fp16 = slice_by_index(begin = var_41635_begin_0, end = var_41635_end_0, end_mask = var_41635_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41635_cast_fp16")]; + tensor var_41642_begin_0 = const()[name = string("op_41642_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41642_end_0 = const()[name = string("op_41642_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41642_end_mask_0 = const()[name = string("op_41642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41642_cast_fp16 = slice_by_index(begin = var_41642_begin_0, end = var_41642_end_0, end_mask = var_41642_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41642_cast_fp16")]; + tensor var_41649_begin_0 = const()[name = string("op_41649_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41649_end_0 = const()[name = string("op_41649_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41649_end_mask_0 = const()[name = string("op_41649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41649_cast_fp16 = slice_by_index(begin = var_41649_begin_0, end = var_41649_end_0, end_mask = var_41649_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41649_cast_fp16")]; + tensor var_41656_begin_0 = const()[name = string("op_41656_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41656_end_0 = const()[name = string("op_41656_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41656_end_mask_0 = const()[name = string("op_41656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41656_cast_fp16 = slice_by_index(begin = var_41656_begin_0, end = var_41656_end_0, end_mask = var_41656_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41656_cast_fp16")]; + tensor var_41663_begin_0 = const()[name = string("op_41663_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41663_end_0 = const()[name = string("op_41663_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41663_end_mask_0 = const()[name = string("op_41663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41663_cast_fp16 = slice_by_index(begin = var_41663_begin_0, end = var_41663_end_0, end_mask = var_41663_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41663_cast_fp16")]; + tensor var_41670_begin_0 = const()[name = string("op_41670_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41670_end_0 = const()[name = string("op_41670_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41670_end_mask_0 = const()[name = string("op_41670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41670_cast_fp16 = slice_by_index(begin = var_41670_begin_0, end = var_41670_end_0, end_mask = var_41670_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41670_cast_fp16")]; + tensor var_41677_begin_0 = const()[name = string("op_41677_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41677_end_0 = const()[name = string("op_41677_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41677_end_mask_0 = const()[name = string("op_41677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41677_cast_fp16 = slice_by_index(begin = var_41677_begin_0, end = var_41677_end_0, end_mask = var_41677_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41677_cast_fp16")]; + tensor var_41684_begin_0 = const()[name = string("op_41684_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41684_end_0 = const()[name = string("op_41684_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41684_end_mask_0 = const()[name = string("op_41684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41684_cast_fp16 = slice_by_index(begin = var_41684_begin_0, end = var_41684_end_0, end_mask = var_41684_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41684_cast_fp16")]; + tensor var_41691_begin_0 = const()[name = string("op_41691_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41691_end_0 = const()[name = string("op_41691_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41691_end_mask_0 = const()[name = string("op_41691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41691_cast_fp16 = slice_by_index(begin = var_41691_begin_0, end = var_41691_end_0, end_mask = var_41691_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41691_cast_fp16")]; + tensor var_41698_begin_0 = const()[name = string("op_41698_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41698_end_0 = const()[name = string("op_41698_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41698_end_mask_0 = const()[name = string("op_41698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41698_cast_fp16 = slice_by_index(begin = var_41698_begin_0, end = var_41698_end_0, end_mask = var_41698_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41698_cast_fp16")]; + tensor var_41705_begin_0 = const()[name = string("op_41705_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41705_end_0 = const()[name = string("op_41705_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41705_end_mask_0 = const()[name = string("op_41705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41705_cast_fp16 = slice_by_index(begin = var_41705_begin_0, end = var_41705_end_0, end_mask = var_41705_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41705_cast_fp16")]; + tensor var_41712_begin_0 = const()[name = string("op_41712_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41712_end_0 = const()[name = string("op_41712_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41712_end_mask_0 = const()[name = string("op_41712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41712_cast_fp16 = slice_by_index(begin = var_41712_begin_0, end = var_41712_end_0, end_mask = var_41712_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41712_cast_fp16")]; + tensor var_41719_begin_0 = const()[name = string("op_41719_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41719_end_0 = const()[name = string("op_41719_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41719_end_mask_0 = const()[name = string("op_41719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41719_cast_fp16 = slice_by_index(begin = var_41719_begin_0, end = var_41719_end_0, end_mask = var_41719_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41719_cast_fp16")]; + tensor var_41726_begin_0 = const()[name = string("op_41726_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41726_end_0 = const()[name = string("op_41726_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41726_end_mask_0 = const()[name = string("op_41726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41726_cast_fp16 = slice_by_index(begin = var_41726_begin_0, end = var_41726_end_0, end_mask = var_41726_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41726_cast_fp16")]; + tensor var_41733_begin_0 = const()[name = string("op_41733_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41733_end_0 = const()[name = string("op_41733_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41733_end_mask_0 = const()[name = string("op_41733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41733_cast_fp16 = slice_by_index(begin = var_41733_begin_0, end = var_41733_end_0, end_mask = var_41733_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41733_cast_fp16")]; + tensor var_41740_begin_0 = const()[name = string("op_41740_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41740_end_0 = const()[name = string("op_41740_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41740_end_mask_0 = const()[name = string("op_41740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41740_cast_fp16 = slice_by_index(begin = var_41740_begin_0, end = var_41740_end_0, end_mask = var_41740_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41740_cast_fp16")]; + tensor var_41747_begin_0 = const()[name = string("op_41747_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41747_end_0 = const()[name = string("op_41747_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41747_end_mask_0 = const()[name = string("op_41747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41747_cast_fp16 = slice_by_index(begin = var_41747_begin_0, end = var_41747_end_0, end_mask = var_41747_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41747_cast_fp16")]; + tensor var_41754_begin_0 = const()[name = string("op_41754_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41754_end_0 = const()[name = string("op_41754_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41754_end_mask_0 = const()[name = string("op_41754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41754_cast_fp16 = slice_by_index(begin = var_41754_begin_0, end = var_41754_end_0, end_mask = var_41754_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41754_cast_fp16")]; + tensor var_41761_begin_0 = const()[name = string("op_41761_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41761_end_0 = const()[name = string("op_41761_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41761_end_mask_0 = const()[name = string("op_41761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41761_cast_fp16 = slice_by_index(begin = var_41761_begin_0, end = var_41761_end_0, end_mask = var_41761_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41761_cast_fp16")]; + tensor var_41768_begin_0 = const()[name = string("op_41768_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41768_end_0 = const()[name = string("op_41768_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41768_end_mask_0 = const()[name = string("op_41768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41768_cast_fp16 = slice_by_index(begin = var_41768_begin_0, end = var_41768_end_0, end_mask = var_41768_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41768_cast_fp16")]; + tensor var_41775_begin_0 = const()[name = string("op_41775_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41775_end_0 = const()[name = string("op_41775_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41775_end_mask_0 = const()[name = string("op_41775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41775_cast_fp16 = slice_by_index(begin = var_41775_begin_0, end = var_41775_end_0, end_mask = var_41775_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41775_cast_fp16")]; + tensor var_41782_begin_0 = const()[name = string("op_41782_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41782_end_0 = const()[name = string("op_41782_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41782_end_mask_0 = const()[name = string("op_41782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41782_cast_fp16 = slice_by_index(begin = var_41782_begin_0, end = var_41782_end_0, end_mask = var_41782_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41782_cast_fp16")]; + tensor var_41789_begin_0 = const()[name = string("op_41789_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41789_end_0 = const()[name = string("op_41789_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41789_end_mask_0 = const()[name = string("op_41789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41789_cast_fp16 = slice_by_index(begin = var_41789_begin_0, end = var_41789_end_0, end_mask = var_41789_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41789_cast_fp16")]; + tensor var_41796_begin_0 = const()[name = string("op_41796_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41796_end_0 = const()[name = string("op_41796_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41796_end_mask_0 = const()[name = string("op_41796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41796_cast_fp16 = slice_by_index(begin = var_41796_begin_0, end = var_41796_end_0, end_mask = var_41796_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41796_cast_fp16")]; + tensor var_41803_begin_0 = const()[name = string("op_41803_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41803_end_0 = const()[name = string("op_41803_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41803_end_mask_0 = const()[name = string("op_41803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41803_cast_fp16 = slice_by_index(begin = var_41803_begin_0, end = var_41803_end_0, end_mask = var_41803_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41803_cast_fp16")]; + tensor var_41810_begin_0 = const()[name = string("op_41810_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41810_end_0 = const()[name = string("op_41810_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41810_end_mask_0 = const()[name = string("op_41810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41810_cast_fp16 = slice_by_index(begin = var_41810_begin_0, end = var_41810_end_0, end_mask = var_41810_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41810_cast_fp16")]; + tensor var_41817_begin_0 = const()[name = string("op_41817_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41817_end_0 = const()[name = string("op_41817_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41817_end_mask_0 = const()[name = string("op_41817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41817_cast_fp16 = slice_by_index(begin = var_41817_begin_0, end = var_41817_end_0, end_mask = var_41817_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41817_cast_fp16")]; + tensor var_41824_begin_0 = const()[name = string("op_41824_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41824_end_0 = const()[name = string("op_41824_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41824_end_mask_0 = const()[name = string("op_41824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41824_cast_fp16 = slice_by_index(begin = var_41824_begin_0, end = var_41824_end_0, end_mask = var_41824_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41824_cast_fp16")]; + tensor var_41831_begin_0 = const()[name = string("op_41831_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41831_end_0 = const()[name = string("op_41831_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41831_end_mask_0 = const()[name = string("op_41831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41831_cast_fp16 = slice_by_index(begin = var_41831_begin_0, end = var_41831_end_0, end_mask = var_41831_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41831_cast_fp16")]; + tensor var_41838_begin_0 = const()[name = string("op_41838_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41838_end_0 = const()[name = string("op_41838_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41838_end_mask_0 = const()[name = string("op_41838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41838_cast_fp16 = slice_by_index(begin = var_41838_begin_0, end = var_41838_end_0, end_mask = var_41838_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41838_cast_fp16")]; + tensor var_41845_begin_0 = const()[name = string("op_41845_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41845_end_0 = const()[name = string("op_41845_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41845_end_mask_0 = const()[name = string("op_41845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41845_cast_fp16 = slice_by_index(begin = var_41845_begin_0, end = var_41845_end_0, end_mask = var_41845_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41845_cast_fp16")]; + tensor var_41852_begin_0 = const()[name = string("op_41852_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41852_end_0 = const()[name = string("op_41852_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41852_end_mask_0 = const()[name = string("op_41852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41852_cast_fp16 = slice_by_index(begin = var_41852_begin_0, end = var_41852_end_0, end_mask = var_41852_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41852_cast_fp16")]; + tensor var_41859_begin_0 = const()[name = string("op_41859_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41859_end_0 = const()[name = string("op_41859_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41859_end_mask_0 = const()[name = string("op_41859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41859_cast_fp16 = slice_by_index(begin = var_41859_begin_0, end = var_41859_end_0, end_mask = var_41859_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41859_cast_fp16")]; + tensor var_41866_begin_0 = const()[name = string("op_41866_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41866_end_0 = const()[name = string("op_41866_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41866_end_mask_0 = const()[name = string("op_41866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41866_cast_fp16 = slice_by_index(begin = var_41866_begin_0, end = var_41866_end_0, end_mask = var_41866_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41866_cast_fp16")]; + tensor var_41873_begin_0 = const()[name = string("op_41873_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41873_end_0 = const()[name = string("op_41873_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41873_end_mask_0 = const()[name = string("op_41873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41873_cast_fp16 = slice_by_index(begin = var_41873_begin_0, end = var_41873_end_0, end_mask = var_41873_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41873_cast_fp16")]; + tensor var_41880_begin_0 = const()[name = string("op_41880_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41880_end_0 = const()[name = string("op_41880_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41880_end_mask_0 = const()[name = string("op_41880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41880_cast_fp16 = slice_by_index(begin = var_41880_begin_0, end = var_41880_end_0, end_mask = var_41880_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41880_cast_fp16")]; + tensor var_41887_begin_0 = const()[name = string("op_41887_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41887_end_0 = const()[name = string("op_41887_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41887_end_mask_0 = const()[name = string("op_41887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41887_cast_fp16 = slice_by_index(begin = var_41887_begin_0, end = var_41887_end_0, end_mask = var_41887_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41887_cast_fp16")]; + tensor var_41894_begin_0 = const()[name = string("op_41894_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41894_end_0 = const()[name = string("op_41894_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41894_end_mask_0 = const()[name = string("op_41894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41894_cast_fp16 = slice_by_index(begin = var_41894_begin_0, end = var_41894_end_0, end_mask = var_41894_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41894_cast_fp16")]; + tensor var_41901_begin_0 = const()[name = string("op_41901_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41901_end_0 = const()[name = string("op_41901_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41901_end_mask_0 = const()[name = string("op_41901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41901_cast_fp16 = slice_by_index(begin = var_41901_begin_0, end = var_41901_end_0, end_mask = var_41901_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41901_cast_fp16")]; + tensor var_41908_begin_0 = const()[name = string("op_41908_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41908_end_0 = const()[name = string("op_41908_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41908_end_mask_0 = const()[name = string("op_41908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41908_cast_fp16 = slice_by_index(begin = var_41908_begin_0, end = var_41908_end_0, end_mask = var_41908_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41908_cast_fp16")]; + tensor var_41915_begin_0 = const()[name = string("op_41915_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41915_end_0 = const()[name = string("op_41915_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41915_end_mask_0 = const()[name = string("op_41915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41915_cast_fp16 = slice_by_index(begin = var_41915_begin_0, end = var_41915_end_0, end_mask = var_41915_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41915_cast_fp16")]; + tensor var_41922_begin_0 = const()[name = string("op_41922_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41922_end_0 = const()[name = string("op_41922_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41922_end_mask_0 = const()[name = string("op_41922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41922_cast_fp16 = slice_by_index(begin = var_41922_begin_0, end = var_41922_end_0, end_mask = var_41922_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41922_cast_fp16")]; + tensor var_41929_begin_0 = const()[name = string("op_41929_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41929_end_0 = const()[name = string("op_41929_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41929_end_mask_0 = const()[name = string("op_41929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41929_cast_fp16 = slice_by_index(begin = var_41929_begin_0, end = var_41929_end_0, end_mask = var_41929_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41929_cast_fp16")]; + tensor var_41936_begin_0 = const()[name = string("op_41936_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41936_end_0 = const()[name = string("op_41936_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41936_end_mask_0 = const()[name = string("op_41936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41936_cast_fp16 = slice_by_index(begin = var_41936_begin_0, end = var_41936_end_0, end_mask = var_41936_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41936_cast_fp16")]; + tensor var_41943_begin_0 = const()[name = string("op_41943_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41943_end_0 = const()[name = string("op_41943_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41943_end_mask_0 = const()[name = string("op_41943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41943_cast_fp16 = slice_by_index(begin = var_41943_begin_0, end = var_41943_end_0, end_mask = var_41943_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41943_cast_fp16")]; + tensor var_41950_begin_0 = const()[name = string("op_41950_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41950_end_0 = const()[name = string("op_41950_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41950_end_mask_0 = const()[name = string("op_41950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41950_cast_fp16 = slice_by_index(begin = var_41950_begin_0, end = var_41950_end_0, end_mask = var_41950_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41950_cast_fp16")]; + tensor var_41957_begin_0 = const()[name = string("op_41957_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41957_end_0 = const()[name = string("op_41957_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41957_end_mask_0 = const()[name = string("op_41957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41957_cast_fp16 = slice_by_index(begin = var_41957_begin_0, end = var_41957_end_0, end_mask = var_41957_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41957_cast_fp16")]; + tensor var_41964_begin_0 = const()[name = string("op_41964_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41964_end_0 = const()[name = string("op_41964_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41964_end_mask_0 = const()[name = string("op_41964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41964_cast_fp16 = slice_by_index(begin = var_41964_begin_0, end = var_41964_end_0, end_mask = var_41964_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41964_cast_fp16")]; + tensor var_41971_begin_0 = const()[name = string("op_41971_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41971_end_0 = const()[name = string("op_41971_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41971_end_mask_0 = const()[name = string("op_41971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41971_cast_fp16 = slice_by_index(begin = var_41971_begin_0, end = var_41971_end_0, end_mask = var_41971_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41971_cast_fp16")]; + tensor var_41978_begin_0 = const()[name = string("op_41978_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41978_end_0 = const()[name = string("op_41978_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41978_end_mask_0 = const()[name = string("op_41978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41978_cast_fp16 = slice_by_index(begin = var_41978_begin_0, end = var_41978_end_0, end_mask = var_41978_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41978_cast_fp16")]; + tensor var_41985_begin_0 = const()[name = string("op_41985_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41985_end_0 = const()[name = string("op_41985_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41985_end_mask_0 = const()[name = string("op_41985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41985_cast_fp16 = slice_by_index(begin = var_41985_begin_0, end = var_41985_end_0, end_mask = var_41985_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41985_cast_fp16")]; + tensor var_41992_begin_0 = const()[name = string("op_41992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41992_end_0 = const()[name = string("op_41992_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41992_end_mask_0 = const()[name = string("op_41992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41992_cast_fp16 = slice_by_index(begin = var_41992_begin_0, end = var_41992_end_0, end_mask = var_41992_end_mask_0, x = var_41547_cast_fp16)[name = string("op_41992_cast_fp16")]; + tensor var_41999_begin_0 = const()[name = string("op_41999_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41999_end_0 = const()[name = string("op_41999_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41999_end_mask_0 = const()[name = string("op_41999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41999_cast_fp16 = slice_by_index(begin = var_41999_begin_0, end = var_41999_end_0, end_mask = var_41999_end_mask_0, x = var_41547_cast_fp16)[name = string("op_41999_cast_fp16")]; + tensor var_42006_begin_0 = const()[name = string("op_42006_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42006_end_0 = const()[name = string("op_42006_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42006_end_mask_0 = const()[name = string("op_42006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42006_cast_fp16 = slice_by_index(begin = var_42006_begin_0, end = var_42006_end_0, end_mask = var_42006_end_mask_0, x = var_41547_cast_fp16)[name = string("op_42006_cast_fp16")]; + tensor var_42013_begin_0 = const()[name = string("op_42013_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42013_end_0 = const()[name = string("op_42013_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42013_end_mask_0 = const()[name = string("op_42013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42013_cast_fp16 = slice_by_index(begin = var_42013_begin_0, end = var_42013_end_0, end_mask = var_42013_end_mask_0, x = var_41547_cast_fp16)[name = string("op_42013_cast_fp16")]; + tensor var_42020_begin_0 = const()[name = string("op_42020_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42020_end_0 = const()[name = string("op_42020_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42020_end_mask_0 = const()[name = string("op_42020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42020_cast_fp16 = slice_by_index(begin = var_42020_begin_0, end = var_42020_end_0, end_mask = var_42020_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42020_cast_fp16")]; + tensor var_42027_begin_0 = const()[name = string("op_42027_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42027_end_0 = const()[name = string("op_42027_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42027_end_mask_0 = const()[name = string("op_42027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42027_cast_fp16 = slice_by_index(begin = var_42027_begin_0, end = var_42027_end_0, end_mask = var_42027_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42027_cast_fp16")]; + tensor var_42034_begin_0 = const()[name = string("op_42034_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42034_end_0 = const()[name = string("op_42034_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42034_end_mask_0 = const()[name = string("op_42034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42034_cast_fp16 = slice_by_index(begin = var_42034_begin_0, end = var_42034_end_0, end_mask = var_42034_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42034_cast_fp16")]; + tensor var_42041_begin_0 = const()[name = string("op_42041_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42041_end_0 = const()[name = string("op_42041_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42041_end_mask_0 = const()[name = string("op_42041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42041_cast_fp16 = slice_by_index(begin = var_42041_begin_0, end = var_42041_end_0, end_mask = var_42041_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42041_cast_fp16")]; + tensor var_42048_begin_0 = const()[name = string("op_42048_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42048_end_0 = const()[name = string("op_42048_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42048_end_mask_0 = const()[name = string("op_42048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42048_cast_fp16 = slice_by_index(begin = var_42048_begin_0, end = var_42048_end_0, end_mask = var_42048_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42048_cast_fp16")]; + tensor var_42055_begin_0 = const()[name = string("op_42055_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42055_end_0 = const()[name = string("op_42055_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42055_end_mask_0 = const()[name = string("op_42055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42055_cast_fp16 = slice_by_index(begin = var_42055_begin_0, end = var_42055_end_0, end_mask = var_42055_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42055_cast_fp16")]; + tensor var_42062_begin_0 = const()[name = string("op_42062_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42062_end_0 = const()[name = string("op_42062_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42062_end_mask_0 = const()[name = string("op_42062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42062_cast_fp16 = slice_by_index(begin = var_42062_begin_0, end = var_42062_end_0, end_mask = var_42062_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42062_cast_fp16")]; + tensor var_42069_begin_0 = const()[name = string("op_42069_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42069_end_0 = const()[name = string("op_42069_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42069_end_mask_0 = const()[name = string("op_42069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42069_cast_fp16 = slice_by_index(begin = var_42069_begin_0, end = var_42069_end_0, end_mask = var_42069_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42069_cast_fp16")]; + tensor var_42076_begin_0 = const()[name = string("op_42076_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42076_end_0 = const()[name = string("op_42076_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42076_end_mask_0 = const()[name = string("op_42076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42076_cast_fp16 = slice_by_index(begin = var_42076_begin_0, end = var_42076_end_0, end_mask = var_42076_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42076_cast_fp16")]; + tensor var_42083_begin_0 = const()[name = string("op_42083_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42083_end_0 = const()[name = string("op_42083_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42083_end_mask_0 = const()[name = string("op_42083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42083_cast_fp16 = slice_by_index(begin = var_42083_begin_0, end = var_42083_end_0, end_mask = var_42083_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42083_cast_fp16")]; + tensor var_42090_begin_0 = const()[name = string("op_42090_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42090_end_0 = const()[name = string("op_42090_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42090_end_mask_0 = const()[name = string("op_42090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42090_cast_fp16 = slice_by_index(begin = var_42090_begin_0, end = var_42090_end_0, end_mask = var_42090_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42090_cast_fp16")]; + tensor var_42097_begin_0 = const()[name = string("op_42097_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42097_end_0 = const()[name = string("op_42097_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42097_end_mask_0 = const()[name = string("op_42097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42097_cast_fp16 = slice_by_index(begin = var_42097_begin_0, end = var_42097_end_0, end_mask = var_42097_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42097_cast_fp16")]; + tensor var_42104_begin_0 = const()[name = string("op_42104_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42104_end_0 = const()[name = string("op_42104_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42104_end_mask_0 = const()[name = string("op_42104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42104_cast_fp16 = slice_by_index(begin = var_42104_begin_0, end = var_42104_end_0, end_mask = var_42104_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42104_cast_fp16")]; + tensor var_42111_begin_0 = const()[name = string("op_42111_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42111_end_0 = const()[name = string("op_42111_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42111_end_mask_0 = const()[name = string("op_42111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42111_cast_fp16 = slice_by_index(begin = var_42111_begin_0, end = var_42111_end_0, end_mask = var_42111_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42111_cast_fp16")]; + tensor var_42118_begin_0 = const()[name = string("op_42118_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42118_end_0 = const()[name = string("op_42118_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42118_end_mask_0 = const()[name = string("op_42118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42118_cast_fp16 = slice_by_index(begin = var_42118_begin_0, end = var_42118_end_0, end_mask = var_42118_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42118_cast_fp16")]; + tensor var_42125_begin_0 = const()[name = string("op_42125_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42125_end_0 = const()[name = string("op_42125_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42125_end_mask_0 = const()[name = string("op_42125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42125_cast_fp16 = slice_by_index(begin = var_42125_begin_0, end = var_42125_end_0, end_mask = var_42125_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42125_cast_fp16")]; + tensor k_55_perm_0 = const()[name = string("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_42130_begin_0 = const()[name = string("op_42130_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42130_end_0 = const()[name = string("op_42130_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_42130_end_mask_0 = const()[name = string("op_42130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_55_cast_fp16 = transpose(perm = k_55_perm_0, x = key_55_cast_fp16)[name = string("transpose_4")]; + tensor var_42130_cast_fp16 = slice_by_index(begin = var_42130_begin_0, end = var_42130_end_0, end_mask = var_42130_end_mask_0, x = k_55_cast_fp16)[name = string("op_42130_cast_fp16")]; + tensor var_42134_begin_0 = const()[name = string("op_42134_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_42134_end_0 = const()[name = string("op_42134_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_42134_end_mask_0 = const()[name = string("op_42134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42134_cast_fp16 = slice_by_index(begin = var_42134_begin_0, end = var_42134_end_0, end_mask = var_42134_end_mask_0, x = k_55_cast_fp16)[name = string("op_42134_cast_fp16")]; + tensor var_42138_begin_0 = const()[name = string("op_42138_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_42138_end_0 = const()[name = string("op_42138_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_42138_end_mask_0 = const()[name = string("op_42138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42138_cast_fp16 = slice_by_index(begin = var_42138_begin_0, end = var_42138_end_0, end_mask = var_42138_end_mask_0, x = k_55_cast_fp16)[name = string("op_42138_cast_fp16")]; + tensor var_42142_begin_0 = const()[name = string("op_42142_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_42142_end_0 = const()[name = string("op_42142_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_42142_end_mask_0 = const()[name = string("op_42142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42142_cast_fp16 = slice_by_index(begin = var_42142_begin_0, end = var_42142_end_0, end_mask = var_42142_end_mask_0, x = k_55_cast_fp16)[name = string("op_42142_cast_fp16")]; + tensor var_42146_begin_0 = const()[name = string("op_42146_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_42146_end_0 = const()[name = string("op_42146_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_42146_end_mask_0 = const()[name = string("op_42146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42146_cast_fp16 = slice_by_index(begin = var_42146_begin_0, end = var_42146_end_0, end_mask = var_42146_end_mask_0, x = k_55_cast_fp16)[name = string("op_42146_cast_fp16")]; + tensor var_42150_begin_0 = const()[name = string("op_42150_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_42150_end_0 = const()[name = string("op_42150_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_42150_end_mask_0 = const()[name = string("op_42150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42150_cast_fp16 = slice_by_index(begin = var_42150_begin_0, end = var_42150_end_0, end_mask = var_42150_end_mask_0, x = k_55_cast_fp16)[name = string("op_42150_cast_fp16")]; + tensor var_42154_begin_0 = const()[name = string("op_42154_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_42154_end_0 = const()[name = string("op_42154_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_42154_end_mask_0 = const()[name = string("op_42154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42154_cast_fp16 = slice_by_index(begin = var_42154_begin_0, end = var_42154_end_0, end_mask = var_42154_end_mask_0, x = k_55_cast_fp16)[name = string("op_42154_cast_fp16")]; + tensor var_42158_begin_0 = const()[name = string("op_42158_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_42158_end_0 = const()[name = string("op_42158_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_42158_end_mask_0 = const()[name = string("op_42158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42158_cast_fp16 = slice_by_index(begin = var_42158_begin_0, end = var_42158_end_0, end_mask = var_42158_end_mask_0, x = k_55_cast_fp16)[name = string("op_42158_cast_fp16")]; + tensor var_42162_begin_0 = const()[name = string("op_42162_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_42162_end_0 = const()[name = string("op_42162_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_42162_end_mask_0 = const()[name = string("op_42162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42162_cast_fp16 = slice_by_index(begin = var_42162_begin_0, end = var_42162_end_0, end_mask = var_42162_end_mask_0, x = k_55_cast_fp16)[name = string("op_42162_cast_fp16")]; + tensor var_42166_begin_0 = const()[name = string("op_42166_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_42166_end_0 = const()[name = string("op_42166_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_42166_end_mask_0 = const()[name = string("op_42166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42166_cast_fp16 = slice_by_index(begin = var_42166_begin_0, end = var_42166_end_0, end_mask = var_42166_end_mask_0, x = k_55_cast_fp16)[name = string("op_42166_cast_fp16")]; + tensor var_42170_begin_0 = const()[name = string("op_42170_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_42170_end_0 = const()[name = string("op_42170_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_42170_end_mask_0 = const()[name = string("op_42170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42170_cast_fp16 = slice_by_index(begin = var_42170_begin_0, end = var_42170_end_0, end_mask = var_42170_end_mask_0, x = k_55_cast_fp16)[name = string("op_42170_cast_fp16")]; + tensor var_42174_begin_0 = const()[name = string("op_42174_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_42174_end_0 = const()[name = string("op_42174_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_42174_end_mask_0 = const()[name = string("op_42174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42174_cast_fp16 = slice_by_index(begin = var_42174_begin_0, end = var_42174_end_0, end_mask = var_42174_end_mask_0, x = k_55_cast_fp16)[name = string("op_42174_cast_fp16")]; + tensor var_42178_begin_0 = const()[name = string("op_42178_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_42178_end_0 = const()[name = string("op_42178_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_42178_end_mask_0 = const()[name = string("op_42178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42178_cast_fp16 = slice_by_index(begin = var_42178_begin_0, end = var_42178_end_0, end_mask = var_42178_end_mask_0, x = k_55_cast_fp16)[name = string("op_42178_cast_fp16")]; + tensor var_42182_begin_0 = const()[name = string("op_42182_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_42182_end_0 = const()[name = string("op_42182_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_42182_end_mask_0 = const()[name = string("op_42182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42182_cast_fp16 = slice_by_index(begin = var_42182_begin_0, end = var_42182_end_0, end_mask = var_42182_end_mask_0, x = k_55_cast_fp16)[name = string("op_42182_cast_fp16")]; + tensor var_42186_begin_0 = const()[name = string("op_42186_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_42186_end_0 = const()[name = string("op_42186_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_42186_end_mask_0 = const()[name = string("op_42186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42186_cast_fp16 = slice_by_index(begin = var_42186_begin_0, end = var_42186_end_0, end_mask = var_42186_end_mask_0, x = k_55_cast_fp16)[name = string("op_42186_cast_fp16")]; + tensor var_42190_begin_0 = const()[name = string("op_42190_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_42190_end_0 = const()[name = string("op_42190_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_42190_end_mask_0 = const()[name = string("op_42190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42190_cast_fp16 = slice_by_index(begin = var_42190_begin_0, end = var_42190_end_0, end_mask = var_42190_end_mask_0, x = k_55_cast_fp16)[name = string("op_42190_cast_fp16")]; + tensor var_42194_begin_0 = const()[name = string("op_42194_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_42194_end_0 = const()[name = string("op_42194_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_42194_end_mask_0 = const()[name = string("op_42194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42194_cast_fp16 = slice_by_index(begin = var_42194_begin_0, end = var_42194_end_0, end_mask = var_42194_end_mask_0, x = k_55_cast_fp16)[name = string("op_42194_cast_fp16")]; + tensor var_42198_begin_0 = const()[name = string("op_42198_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_42198_end_0 = const()[name = string("op_42198_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_42198_end_mask_0 = const()[name = string("op_42198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42198_cast_fp16 = slice_by_index(begin = var_42198_begin_0, end = var_42198_end_0, end_mask = var_42198_end_mask_0, x = k_55_cast_fp16)[name = string("op_42198_cast_fp16")]; + tensor var_42202_begin_0 = const()[name = string("op_42202_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_42202_end_0 = const()[name = string("op_42202_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_42202_end_mask_0 = const()[name = string("op_42202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42202_cast_fp16 = slice_by_index(begin = var_42202_begin_0, end = var_42202_end_0, end_mask = var_42202_end_mask_0, x = k_55_cast_fp16)[name = string("op_42202_cast_fp16")]; + tensor var_42206_begin_0 = const()[name = string("op_42206_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_42206_end_0 = const()[name = string("op_42206_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_42206_end_mask_0 = const()[name = string("op_42206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42206_cast_fp16 = slice_by_index(begin = var_42206_begin_0, end = var_42206_end_0, end_mask = var_42206_end_mask_0, x = k_55_cast_fp16)[name = string("op_42206_cast_fp16")]; + tensor var_42208_begin_0 = const()[name = string("op_42208_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42208_end_0 = const()[name = string("op_42208_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42208_end_mask_0 = const()[name = string("op_42208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42208_cast_fp16 = slice_by_index(begin = var_42208_begin_0, end = var_42208_end_0, end_mask = var_42208_end_mask_0, x = value_55_cast_fp16)[name = string("op_42208_cast_fp16")]; + tensor var_42212_begin_0 = const()[name = string("op_42212_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42212_end_0 = const()[name = string("op_42212_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_42212_end_mask_0 = const()[name = string("op_42212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42212_cast_fp16 = slice_by_index(begin = var_42212_begin_0, end = var_42212_end_0, end_mask = var_42212_end_mask_0, x = value_55_cast_fp16)[name = string("op_42212_cast_fp16")]; + tensor var_42216_begin_0 = const()[name = string("op_42216_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42216_end_0 = const()[name = string("op_42216_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_42216_end_mask_0 = const()[name = string("op_42216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42216_cast_fp16 = slice_by_index(begin = var_42216_begin_0, end = var_42216_end_0, end_mask = var_42216_end_mask_0, x = value_55_cast_fp16)[name = string("op_42216_cast_fp16")]; + tensor var_42220_begin_0 = const()[name = string("op_42220_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42220_end_0 = const()[name = string("op_42220_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_42220_end_mask_0 = const()[name = string("op_42220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42220_cast_fp16 = slice_by_index(begin = var_42220_begin_0, end = var_42220_end_0, end_mask = var_42220_end_mask_0, x = value_55_cast_fp16)[name = string("op_42220_cast_fp16")]; + tensor var_42224_begin_0 = const()[name = string("op_42224_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42224_end_0 = const()[name = string("op_42224_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_42224_end_mask_0 = const()[name = string("op_42224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42224_cast_fp16 = slice_by_index(begin = var_42224_begin_0, end = var_42224_end_0, end_mask = var_42224_end_mask_0, x = value_55_cast_fp16)[name = string("op_42224_cast_fp16")]; + tensor var_42228_begin_0 = const()[name = string("op_42228_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42228_end_0 = const()[name = string("op_42228_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_42228_end_mask_0 = const()[name = string("op_42228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42228_cast_fp16 = slice_by_index(begin = var_42228_begin_0, end = var_42228_end_0, end_mask = var_42228_end_mask_0, x = value_55_cast_fp16)[name = string("op_42228_cast_fp16")]; + tensor var_42232_begin_0 = const()[name = string("op_42232_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42232_end_0 = const()[name = string("op_42232_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_42232_end_mask_0 = const()[name = string("op_42232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42232_cast_fp16 = slice_by_index(begin = var_42232_begin_0, end = var_42232_end_0, end_mask = var_42232_end_mask_0, x = value_55_cast_fp16)[name = string("op_42232_cast_fp16")]; + tensor var_42236_begin_0 = const()[name = string("op_42236_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42236_end_0 = const()[name = string("op_42236_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_42236_end_mask_0 = const()[name = string("op_42236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42236_cast_fp16 = slice_by_index(begin = var_42236_begin_0, end = var_42236_end_0, end_mask = var_42236_end_mask_0, x = value_55_cast_fp16)[name = string("op_42236_cast_fp16")]; + tensor var_42240_begin_0 = const()[name = string("op_42240_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42240_end_0 = const()[name = string("op_42240_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_42240_end_mask_0 = const()[name = string("op_42240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42240_cast_fp16 = slice_by_index(begin = var_42240_begin_0, end = var_42240_end_0, end_mask = var_42240_end_mask_0, x = value_55_cast_fp16)[name = string("op_42240_cast_fp16")]; + tensor var_42244_begin_0 = const()[name = string("op_42244_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42244_end_0 = const()[name = string("op_42244_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_42244_end_mask_0 = const()[name = string("op_42244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42244_cast_fp16 = slice_by_index(begin = var_42244_begin_0, end = var_42244_end_0, end_mask = var_42244_end_mask_0, x = value_55_cast_fp16)[name = string("op_42244_cast_fp16")]; + tensor var_42248_begin_0 = const()[name = string("op_42248_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42248_end_0 = const()[name = string("op_42248_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_42248_end_mask_0 = const()[name = string("op_42248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42248_cast_fp16 = slice_by_index(begin = var_42248_begin_0, end = var_42248_end_0, end_mask = var_42248_end_mask_0, x = value_55_cast_fp16)[name = string("op_42248_cast_fp16")]; + tensor var_42252_begin_0 = const()[name = string("op_42252_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42252_end_0 = const()[name = string("op_42252_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_42252_end_mask_0 = const()[name = string("op_42252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42252_cast_fp16 = slice_by_index(begin = var_42252_begin_0, end = var_42252_end_0, end_mask = var_42252_end_mask_0, x = value_55_cast_fp16)[name = string("op_42252_cast_fp16")]; + tensor var_42256_begin_0 = const()[name = string("op_42256_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42256_end_0 = const()[name = string("op_42256_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_42256_end_mask_0 = const()[name = string("op_42256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42256_cast_fp16 = slice_by_index(begin = var_42256_begin_0, end = var_42256_end_0, end_mask = var_42256_end_mask_0, x = value_55_cast_fp16)[name = string("op_42256_cast_fp16")]; + tensor var_42260_begin_0 = const()[name = string("op_42260_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42260_end_0 = const()[name = string("op_42260_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_42260_end_mask_0 = const()[name = string("op_42260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42260_cast_fp16 = slice_by_index(begin = var_42260_begin_0, end = var_42260_end_0, end_mask = var_42260_end_mask_0, x = value_55_cast_fp16)[name = string("op_42260_cast_fp16")]; + tensor var_42264_begin_0 = const()[name = string("op_42264_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42264_end_0 = const()[name = string("op_42264_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_42264_end_mask_0 = const()[name = string("op_42264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42264_cast_fp16 = slice_by_index(begin = var_42264_begin_0, end = var_42264_end_0, end_mask = var_42264_end_mask_0, x = value_55_cast_fp16)[name = string("op_42264_cast_fp16")]; + tensor var_42268_begin_0 = const()[name = string("op_42268_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42268_end_0 = const()[name = string("op_42268_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_42268_end_mask_0 = const()[name = string("op_42268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42268_cast_fp16 = slice_by_index(begin = var_42268_begin_0, end = var_42268_end_0, end_mask = var_42268_end_mask_0, x = value_55_cast_fp16)[name = string("op_42268_cast_fp16")]; + tensor var_42272_begin_0 = const()[name = string("op_42272_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42272_end_0 = const()[name = string("op_42272_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_42272_end_mask_0 = const()[name = string("op_42272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42272_cast_fp16 = slice_by_index(begin = var_42272_begin_0, end = var_42272_end_0, end_mask = var_42272_end_mask_0, x = value_55_cast_fp16)[name = string("op_42272_cast_fp16")]; + tensor var_42276_begin_0 = const()[name = string("op_42276_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42276_end_0 = const()[name = string("op_42276_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_42276_end_mask_0 = const()[name = string("op_42276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42276_cast_fp16 = slice_by_index(begin = var_42276_begin_0, end = var_42276_end_0, end_mask = var_42276_end_mask_0, x = value_55_cast_fp16)[name = string("op_42276_cast_fp16")]; + tensor var_42280_begin_0 = const()[name = string("op_42280_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42280_end_0 = const()[name = string("op_42280_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_42280_end_mask_0 = const()[name = string("op_42280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42280_cast_fp16 = slice_by_index(begin = var_42280_begin_0, end = var_42280_end_0, end_mask = var_42280_end_mask_0, x = value_55_cast_fp16)[name = string("op_42280_cast_fp16")]; + tensor var_42284_begin_0 = const()[name = string("op_42284_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42284_end_0 = const()[name = string("op_42284_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_42284_end_mask_0 = const()[name = string("op_42284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42284_cast_fp16 = slice_by_index(begin = var_42284_begin_0, end = var_42284_end_0, end_mask = var_42284_end_mask_0, x = value_55_cast_fp16)[name = string("op_42284_cast_fp16")]; + string _SplitHeadsQ__mh_w_4321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4321_equation_0, values = (var_42130_cast_fp16, var_41572_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4321_cast_fp16")]; + string _SplitHeadsQ__mh_w_4323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4323_equation_0, values = (var_42130_cast_fp16, var_41579_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4323_cast_fp16")]; + string _SplitHeadsQ__mh_w_4325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4325_equation_0, values = (var_42130_cast_fp16, var_41586_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4325_cast_fp16")]; + string _SplitHeadsQ__mh_w_4327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4327_equation_0, values = (var_42130_cast_fp16, var_41593_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4327_cast_fp16")]; + string _SplitHeadsQ__mh_w_4329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4329_equation_0, values = (var_42134_cast_fp16, var_41600_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4329_cast_fp16")]; + string _SplitHeadsQ__mh_w_4331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4331_equation_0, values = (var_42134_cast_fp16, var_41607_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4331_cast_fp16")]; + string _SplitHeadsQ__mh_w_4333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4333_equation_0, values = (var_42134_cast_fp16, var_41614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4333_cast_fp16")]; + string _SplitHeadsQ__mh_w_4335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4335_equation_0, values = (var_42134_cast_fp16, var_41621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4335_cast_fp16")]; + string _SplitHeadsQ__mh_w_4337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4337_equation_0, values = (var_42138_cast_fp16, var_41628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4337_cast_fp16")]; + string _SplitHeadsQ__mh_w_4339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4339_equation_0, values = (var_42138_cast_fp16, var_41635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4339_cast_fp16")]; + string _SplitHeadsQ__mh_w_4341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4341_equation_0, values = (var_42138_cast_fp16, var_41642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4341_cast_fp16")]; + string _SplitHeadsQ__mh_w_4343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4343_equation_0, values = (var_42138_cast_fp16, var_41649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4343_cast_fp16")]; + string _SplitHeadsQ__mh_w_4345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4345_equation_0, values = (var_42142_cast_fp16, var_41656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4345_cast_fp16")]; + string _SplitHeadsQ__mh_w_4347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4347_equation_0, values = (var_42142_cast_fp16, var_41663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4347_cast_fp16")]; + string _SplitHeadsQ__mh_w_4349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4349_equation_0, values = (var_42142_cast_fp16, var_41670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4349_cast_fp16")]; + string _SplitHeadsQ__mh_w_4351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4351_equation_0, values = (var_42142_cast_fp16, var_41677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4351_cast_fp16")]; + string _SplitHeadsQ__mh_w_4353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4353_equation_0, values = (var_42146_cast_fp16, var_41684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4353_cast_fp16")]; + string _SplitHeadsQ__mh_w_4355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4355_equation_0, values = (var_42146_cast_fp16, var_41691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4355_cast_fp16")]; + string _SplitHeadsQ__mh_w_4357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4357_equation_0, values = (var_42146_cast_fp16, var_41698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4357_cast_fp16")]; + string _SplitHeadsQ__mh_w_4359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4359_equation_0, values = (var_42146_cast_fp16, var_41705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4359_cast_fp16")]; + string _SplitHeadsQ__mh_w_4361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4361_equation_0, values = (var_42150_cast_fp16, var_41712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4361_cast_fp16")]; + string _SplitHeadsQ__mh_w_4363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4363_equation_0, values = (var_42150_cast_fp16, var_41719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4363_cast_fp16")]; + string _SplitHeadsQ__mh_w_4365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4365_equation_0, values = (var_42150_cast_fp16, var_41726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4365_cast_fp16")]; + string _SplitHeadsQ__mh_w_4367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4367_equation_0, values = (var_42150_cast_fp16, var_41733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4367_cast_fp16")]; + string _SplitHeadsQ__mh_w_4369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4369_equation_0, values = (var_42154_cast_fp16, var_41740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4369_cast_fp16")]; + string _SplitHeadsQ__mh_w_4371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4371_equation_0, values = (var_42154_cast_fp16, var_41747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4371_cast_fp16")]; + string _SplitHeadsQ__mh_w_4373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4373_equation_0, values = (var_42154_cast_fp16, var_41754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4373_cast_fp16")]; + string _SplitHeadsQ__mh_w_4375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4375_equation_0, values = (var_42154_cast_fp16, var_41761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4375_cast_fp16")]; + string _SplitHeadsQ__mh_w_4377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4377_equation_0, values = (var_42158_cast_fp16, var_41768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4377_cast_fp16")]; + string _SplitHeadsQ__mh_w_4379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4379_equation_0, values = (var_42158_cast_fp16, var_41775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4379_cast_fp16")]; + string _SplitHeadsQ__mh_w_4381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4381_equation_0, values = (var_42158_cast_fp16, var_41782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4381_cast_fp16")]; + string _SplitHeadsQ__mh_w_4383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4383_equation_0, values = (var_42158_cast_fp16, var_41789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4383_cast_fp16")]; + string _SplitHeadsQ__mh_w_4385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4385_equation_0, values = (var_42162_cast_fp16, var_41796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4385_cast_fp16")]; + string _SplitHeadsQ__mh_w_4387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4387_equation_0, values = (var_42162_cast_fp16, var_41803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4387_cast_fp16")]; + string _SplitHeadsQ__mh_w_4389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4389_equation_0, values = (var_42162_cast_fp16, var_41810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4389_cast_fp16")]; + string _SplitHeadsQ__mh_w_4391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4391_equation_0, values = (var_42162_cast_fp16, var_41817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4391_cast_fp16")]; + string _SplitHeadsQ__mh_w_4393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4393_equation_0, values = (var_42166_cast_fp16, var_41824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4393_cast_fp16")]; + string _SplitHeadsQ__mh_w_4395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4395_equation_0, values = (var_42166_cast_fp16, var_41831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4395_cast_fp16")]; + string _SplitHeadsQ__mh_w_4397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4397_equation_0, values = (var_42166_cast_fp16, var_41838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4397_cast_fp16")]; + string _SplitHeadsQ__mh_w_4399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4399_equation_0, values = (var_42166_cast_fp16, var_41845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4399_cast_fp16")]; + string _SplitHeadsQ__mh_w_4401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4401_equation_0, values = (var_42170_cast_fp16, var_41852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4401_cast_fp16")]; + string _SplitHeadsQ__mh_w_4403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4403_equation_0, values = (var_42170_cast_fp16, var_41859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4403_cast_fp16")]; + string _SplitHeadsQ__mh_w_4405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4405_equation_0, values = (var_42170_cast_fp16, var_41866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4405_cast_fp16")]; + string _SplitHeadsQ__mh_w_4407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4407_equation_0, values = (var_42170_cast_fp16, var_41873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4407_cast_fp16")]; + string _SplitHeadsQ__mh_w_4409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4409_equation_0, values = (var_42174_cast_fp16, var_41880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4409_cast_fp16")]; + string _SplitHeadsQ__mh_w_4411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4411_equation_0, values = (var_42174_cast_fp16, var_41887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4411_cast_fp16")]; + string _SplitHeadsQ__mh_w_4413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4413_equation_0, values = (var_42174_cast_fp16, var_41894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4413_cast_fp16")]; + string _SplitHeadsQ__mh_w_4415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4415_equation_0, values = (var_42174_cast_fp16, var_41901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4415_cast_fp16")]; + string _SplitHeadsQ__mh_w_4417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4417_equation_0, values = (var_42178_cast_fp16, var_41908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4417_cast_fp16")]; + string _SplitHeadsQ__mh_w_4419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4419_equation_0, values = (var_42178_cast_fp16, var_41915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4419_cast_fp16")]; + string _SplitHeadsQ__mh_w_4421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4421_equation_0, values = (var_42178_cast_fp16, var_41922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4421_cast_fp16")]; + string _SplitHeadsQ__mh_w_4423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4423_equation_0, values = (var_42178_cast_fp16, var_41929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4423_cast_fp16")]; + string _SplitHeadsQ__mh_w_4425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4425_equation_0, values = (var_42182_cast_fp16, var_41936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4425_cast_fp16")]; + string _SplitHeadsQ__mh_w_4427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4427_equation_0, values = (var_42182_cast_fp16, var_41943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4427_cast_fp16")]; + string _SplitHeadsQ__mh_w_4429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4429_equation_0, values = (var_42182_cast_fp16, var_41950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4429_cast_fp16")]; + string _SplitHeadsQ__mh_w_4431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4431_equation_0, values = (var_42182_cast_fp16, var_41957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4431_cast_fp16")]; + string _SplitHeadsQ__mh_w_4433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4433_equation_0, values = (var_42186_cast_fp16, var_41964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4433_cast_fp16")]; + string _SplitHeadsQ__mh_w_4435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4435_equation_0, values = (var_42186_cast_fp16, var_41971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4435_cast_fp16")]; + string _SplitHeadsQ__mh_w_4437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4437_equation_0, values = (var_42186_cast_fp16, var_41978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4437_cast_fp16")]; + string _SplitHeadsQ__mh_w_4439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4439_equation_0, values = (var_42186_cast_fp16, var_41985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4439_cast_fp16")]; + string _SplitHeadsQ__mh_w_4441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4441_equation_0, values = (var_42190_cast_fp16, var_41992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4441_cast_fp16")]; + string _SplitHeadsQ__mh_w_4443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4443_equation_0, values = (var_42190_cast_fp16, var_41999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4443_cast_fp16")]; + string _SplitHeadsQ__mh_w_4445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4445_equation_0, values = (var_42190_cast_fp16, var_42006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4445_cast_fp16")]; + string _SplitHeadsQ__mh_w_4447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4447_equation_0, values = (var_42190_cast_fp16, var_42013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4447_cast_fp16")]; + string _SplitHeadsQ__mh_w_4449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4449_equation_0, values = (var_42194_cast_fp16, var_42020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4449_cast_fp16")]; + string _SplitHeadsQ__mh_w_4451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4451_equation_0, values = (var_42194_cast_fp16, var_42027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4451_cast_fp16")]; + string _SplitHeadsQ__mh_w_4453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4453_equation_0, values = (var_42194_cast_fp16, var_42034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4453_cast_fp16")]; + string _SplitHeadsQ__mh_w_4455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4455_equation_0, values = (var_42194_cast_fp16, var_42041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4455_cast_fp16")]; + string _SplitHeadsQ__mh_w_4457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4457_equation_0, values = (var_42198_cast_fp16, var_42048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4457_cast_fp16")]; + string _SplitHeadsQ__mh_w_4459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4459_equation_0, values = (var_42198_cast_fp16, var_42055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4459_cast_fp16")]; + string _SplitHeadsQ__mh_w_4461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4461_equation_0, values = (var_42198_cast_fp16, var_42062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4461_cast_fp16")]; + string _SplitHeadsQ__mh_w_4463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4463_equation_0, values = (var_42198_cast_fp16, var_42069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4463_cast_fp16")]; + string _SplitHeadsQ__mh_w_4465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4465_equation_0, values = (var_42202_cast_fp16, var_42076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4465_cast_fp16")]; + string _SplitHeadsQ__mh_w_4467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4467_equation_0, values = (var_42202_cast_fp16, var_42083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4467_cast_fp16")]; + string _SplitHeadsQ__mh_w_4469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4469_equation_0, values = (var_42202_cast_fp16, var_42090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4469_cast_fp16")]; + string _SplitHeadsQ__mh_w_4471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4471_equation_0, values = (var_42202_cast_fp16, var_42097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4471_cast_fp16")]; + string _SplitHeadsQ__mh_w_4473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4473_equation_0, values = (var_42206_cast_fp16, var_42104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4473_cast_fp16")]; + string _SplitHeadsQ__mh_w_4475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4475_equation_0, values = (var_42206_cast_fp16, var_42111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4475_cast_fp16")]; + string _SplitHeadsQ__mh_w_4477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4477_equation_0, values = (var_42206_cast_fp16, var_42118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4477_cast_fp16")]; + string _SplitHeadsQ__mh_w_4479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4479_equation_0, values = (var_42206_cast_fp16, var_42125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4479_cast_fp16")]; + fp16 var_42447_to_fp16 = const()[name = string("op_42447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4321_cast_fp16, y = var_42447_to_fp16)[name = string("aw_chunk_4321_cast_fp16")]; + fp16 var_42449_to_fp16 = const()[name = string("op_42449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4323_cast_fp16, y = var_42449_to_fp16)[name = string("aw_chunk_4323_cast_fp16")]; + fp16 var_42451_to_fp16 = const()[name = string("op_42451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4325_cast_fp16, y = var_42451_to_fp16)[name = string("aw_chunk_4325_cast_fp16")]; + fp16 var_42453_to_fp16 = const()[name = string("op_42453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4327_cast_fp16, y = var_42453_to_fp16)[name = string("aw_chunk_4327_cast_fp16")]; + fp16 var_42455_to_fp16 = const()[name = string("op_42455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4329_cast_fp16, y = var_42455_to_fp16)[name = string("aw_chunk_4329_cast_fp16")]; + fp16 var_42457_to_fp16 = const()[name = string("op_42457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4331_cast_fp16, y = var_42457_to_fp16)[name = string("aw_chunk_4331_cast_fp16")]; + fp16 var_42459_to_fp16 = const()[name = string("op_42459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4333_cast_fp16, y = var_42459_to_fp16)[name = string("aw_chunk_4333_cast_fp16")]; + fp16 var_42461_to_fp16 = const()[name = string("op_42461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4335_cast_fp16, y = var_42461_to_fp16)[name = string("aw_chunk_4335_cast_fp16")]; + fp16 var_42463_to_fp16 = const()[name = string("op_42463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4337_cast_fp16, y = var_42463_to_fp16)[name = string("aw_chunk_4337_cast_fp16")]; + fp16 var_42465_to_fp16 = const()[name = string("op_42465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4339_cast_fp16, y = var_42465_to_fp16)[name = string("aw_chunk_4339_cast_fp16")]; + fp16 var_42467_to_fp16 = const()[name = string("op_42467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4341_cast_fp16, y = var_42467_to_fp16)[name = string("aw_chunk_4341_cast_fp16")]; + fp16 var_42469_to_fp16 = const()[name = string("op_42469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4343_cast_fp16, y = var_42469_to_fp16)[name = string("aw_chunk_4343_cast_fp16")]; + fp16 var_42471_to_fp16 = const()[name = string("op_42471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4345_cast_fp16, y = var_42471_to_fp16)[name = string("aw_chunk_4345_cast_fp16")]; + fp16 var_42473_to_fp16 = const()[name = string("op_42473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4347_cast_fp16, y = var_42473_to_fp16)[name = string("aw_chunk_4347_cast_fp16")]; + fp16 var_42475_to_fp16 = const()[name = string("op_42475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4349_cast_fp16, y = var_42475_to_fp16)[name = string("aw_chunk_4349_cast_fp16")]; + fp16 var_42477_to_fp16 = const()[name = string("op_42477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4351_cast_fp16, y = var_42477_to_fp16)[name = string("aw_chunk_4351_cast_fp16")]; + fp16 var_42479_to_fp16 = const()[name = string("op_42479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4353_cast_fp16, y = var_42479_to_fp16)[name = string("aw_chunk_4353_cast_fp16")]; + fp16 var_42481_to_fp16 = const()[name = string("op_42481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4355_cast_fp16, y = var_42481_to_fp16)[name = string("aw_chunk_4355_cast_fp16")]; + fp16 var_42483_to_fp16 = const()[name = string("op_42483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4357_cast_fp16, y = var_42483_to_fp16)[name = string("aw_chunk_4357_cast_fp16")]; + fp16 var_42485_to_fp16 = const()[name = string("op_42485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4359_cast_fp16, y = var_42485_to_fp16)[name = string("aw_chunk_4359_cast_fp16")]; + fp16 var_42487_to_fp16 = const()[name = string("op_42487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4361_cast_fp16, y = var_42487_to_fp16)[name = string("aw_chunk_4361_cast_fp16")]; + fp16 var_42489_to_fp16 = const()[name = string("op_42489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4363_cast_fp16, y = var_42489_to_fp16)[name = string("aw_chunk_4363_cast_fp16")]; + fp16 var_42491_to_fp16 = const()[name = string("op_42491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4365_cast_fp16, y = var_42491_to_fp16)[name = string("aw_chunk_4365_cast_fp16")]; + fp16 var_42493_to_fp16 = const()[name = string("op_42493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4367_cast_fp16, y = var_42493_to_fp16)[name = string("aw_chunk_4367_cast_fp16")]; + fp16 var_42495_to_fp16 = const()[name = string("op_42495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4369_cast_fp16, y = var_42495_to_fp16)[name = string("aw_chunk_4369_cast_fp16")]; + fp16 var_42497_to_fp16 = const()[name = string("op_42497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4371_cast_fp16, y = var_42497_to_fp16)[name = string("aw_chunk_4371_cast_fp16")]; + fp16 var_42499_to_fp16 = const()[name = string("op_42499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4373_cast_fp16, y = var_42499_to_fp16)[name = string("aw_chunk_4373_cast_fp16")]; + fp16 var_42501_to_fp16 = const()[name = string("op_42501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4375_cast_fp16, y = var_42501_to_fp16)[name = string("aw_chunk_4375_cast_fp16")]; + fp16 var_42503_to_fp16 = const()[name = string("op_42503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4377_cast_fp16, y = var_42503_to_fp16)[name = string("aw_chunk_4377_cast_fp16")]; + fp16 var_42505_to_fp16 = const()[name = string("op_42505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4379_cast_fp16, y = var_42505_to_fp16)[name = string("aw_chunk_4379_cast_fp16")]; + fp16 var_42507_to_fp16 = const()[name = string("op_42507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4381_cast_fp16, y = var_42507_to_fp16)[name = string("aw_chunk_4381_cast_fp16")]; + fp16 var_42509_to_fp16 = const()[name = string("op_42509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4383_cast_fp16, y = var_42509_to_fp16)[name = string("aw_chunk_4383_cast_fp16")]; + fp16 var_42511_to_fp16 = const()[name = string("op_42511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4385_cast_fp16, y = var_42511_to_fp16)[name = string("aw_chunk_4385_cast_fp16")]; + fp16 var_42513_to_fp16 = const()[name = string("op_42513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4387_cast_fp16, y = var_42513_to_fp16)[name = string("aw_chunk_4387_cast_fp16")]; + fp16 var_42515_to_fp16 = const()[name = string("op_42515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4389_cast_fp16, y = var_42515_to_fp16)[name = string("aw_chunk_4389_cast_fp16")]; + fp16 var_42517_to_fp16 = const()[name = string("op_42517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4391_cast_fp16, y = var_42517_to_fp16)[name = string("aw_chunk_4391_cast_fp16")]; + fp16 var_42519_to_fp16 = const()[name = string("op_42519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4393_cast_fp16, y = var_42519_to_fp16)[name = string("aw_chunk_4393_cast_fp16")]; + fp16 var_42521_to_fp16 = const()[name = string("op_42521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4395_cast_fp16, y = var_42521_to_fp16)[name = string("aw_chunk_4395_cast_fp16")]; + fp16 var_42523_to_fp16 = const()[name = string("op_42523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4397_cast_fp16, y = var_42523_to_fp16)[name = string("aw_chunk_4397_cast_fp16")]; + fp16 var_42525_to_fp16 = const()[name = string("op_42525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4399_cast_fp16, y = var_42525_to_fp16)[name = string("aw_chunk_4399_cast_fp16")]; + fp16 var_42527_to_fp16 = const()[name = string("op_42527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4401_cast_fp16, y = var_42527_to_fp16)[name = string("aw_chunk_4401_cast_fp16")]; + fp16 var_42529_to_fp16 = const()[name = string("op_42529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4403_cast_fp16, y = var_42529_to_fp16)[name = string("aw_chunk_4403_cast_fp16")]; + fp16 var_42531_to_fp16 = const()[name = string("op_42531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4405_cast_fp16, y = var_42531_to_fp16)[name = string("aw_chunk_4405_cast_fp16")]; + fp16 var_42533_to_fp16 = const()[name = string("op_42533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4407_cast_fp16, y = var_42533_to_fp16)[name = string("aw_chunk_4407_cast_fp16")]; + fp16 var_42535_to_fp16 = const()[name = string("op_42535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4409_cast_fp16, y = var_42535_to_fp16)[name = string("aw_chunk_4409_cast_fp16")]; + fp16 var_42537_to_fp16 = const()[name = string("op_42537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4411_cast_fp16, y = var_42537_to_fp16)[name = string("aw_chunk_4411_cast_fp16")]; + fp16 var_42539_to_fp16 = const()[name = string("op_42539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4413_cast_fp16, y = var_42539_to_fp16)[name = string("aw_chunk_4413_cast_fp16")]; + fp16 var_42541_to_fp16 = const()[name = string("op_42541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4415_cast_fp16, y = var_42541_to_fp16)[name = string("aw_chunk_4415_cast_fp16")]; + fp16 var_42543_to_fp16 = const()[name = string("op_42543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4417_cast_fp16, y = var_42543_to_fp16)[name = string("aw_chunk_4417_cast_fp16")]; + fp16 var_42545_to_fp16 = const()[name = string("op_42545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4419_cast_fp16, y = var_42545_to_fp16)[name = string("aw_chunk_4419_cast_fp16")]; + fp16 var_42547_to_fp16 = const()[name = string("op_42547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4421_cast_fp16, y = var_42547_to_fp16)[name = string("aw_chunk_4421_cast_fp16")]; + fp16 var_42549_to_fp16 = const()[name = string("op_42549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4423_cast_fp16, y = var_42549_to_fp16)[name = string("aw_chunk_4423_cast_fp16")]; + fp16 var_42551_to_fp16 = const()[name = string("op_42551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4425_cast_fp16, y = var_42551_to_fp16)[name = string("aw_chunk_4425_cast_fp16")]; + fp16 var_42553_to_fp16 = const()[name = string("op_42553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4427_cast_fp16, y = var_42553_to_fp16)[name = string("aw_chunk_4427_cast_fp16")]; + fp16 var_42555_to_fp16 = const()[name = string("op_42555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4429_cast_fp16, y = var_42555_to_fp16)[name = string("aw_chunk_4429_cast_fp16")]; + fp16 var_42557_to_fp16 = const()[name = string("op_42557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4431_cast_fp16, y = var_42557_to_fp16)[name = string("aw_chunk_4431_cast_fp16")]; + fp16 var_42559_to_fp16 = const()[name = string("op_42559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4433_cast_fp16, y = var_42559_to_fp16)[name = string("aw_chunk_4433_cast_fp16")]; + fp16 var_42561_to_fp16 = const()[name = string("op_42561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4435_cast_fp16, y = var_42561_to_fp16)[name = string("aw_chunk_4435_cast_fp16")]; + fp16 var_42563_to_fp16 = const()[name = string("op_42563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4437_cast_fp16, y = var_42563_to_fp16)[name = string("aw_chunk_4437_cast_fp16")]; + fp16 var_42565_to_fp16 = const()[name = string("op_42565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4439_cast_fp16, y = var_42565_to_fp16)[name = string("aw_chunk_4439_cast_fp16")]; + fp16 var_42567_to_fp16 = const()[name = string("op_42567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4441_cast_fp16, y = var_42567_to_fp16)[name = string("aw_chunk_4441_cast_fp16")]; + fp16 var_42569_to_fp16 = const()[name = string("op_42569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4443_cast_fp16, y = var_42569_to_fp16)[name = string("aw_chunk_4443_cast_fp16")]; + fp16 var_42571_to_fp16 = const()[name = string("op_42571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4445_cast_fp16, y = var_42571_to_fp16)[name = string("aw_chunk_4445_cast_fp16")]; + fp16 var_42573_to_fp16 = const()[name = string("op_42573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4447_cast_fp16, y = var_42573_to_fp16)[name = string("aw_chunk_4447_cast_fp16")]; + fp16 var_42575_to_fp16 = const()[name = string("op_42575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4449_cast_fp16, y = var_42575_to_fp16)[name = string("aw_chunk_4449_cast_fp16")]; + fp16 var_42577_to_fp16 = const()[name = string("op_42577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4451_cast_fp16, y = var_42577_to_fp16)[name = string("aw_chunk_4451_cast_fp16")]; + fp16 var_42579_to_fp16 = const()[name = string("op_42579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4453_cast_fp16, y = var_42579_to_fp16)[name = string("aw_chunk_4453_cast_fp16")]; + fp16 var_42581_to_fp16 = const()[name = string("op_42581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4455_cast_fp16, y = var_42581_to_fp16)[name = string("aw_chunk_4455_cast_fp16")]; + fp16 var_42583_to_fp16 = const()[name = string("op_42583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4457_cast_fp16, y = var_42583_to_fp16)[name = string("aw_chunk_4457_cast_fp16")]; + fp16 var_42585_to_fp16 = const()[name = string("op_42585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4459_cast_fp16, y = var_42585_to_fp16)[name = string("aw_chunk_4459_cast_fp16")]; + fp16 var_42587_to_fp16 = const()[name = string("op_42587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4461_cast_fp16, y = var_42587_to_fp16)[name = string("aw_chunk_4461_cast_fp16")]; + fp16 var_42589_to_fp16 = const()[name = string("op_42589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4463_cast_fp16, y = var_42589_to_fp16)[name = string("aw_chunk_4463_cast_fp16")]; + fp16 var_42591_to_fp16 = const()[name = string("op_42591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4465_cast_fp16, y = var_42591_to_fp16)[name = string("aw_chunk_4465_cast_fp16")]; + fp16 var_42593_to_fp16 = const()[name = string("op_42593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4467_cast_fp16, y = var_42593_to_fp16)[name = string("aw_chunk_4467_cast_fp16")]; + fp16 var_42595_to_fp16 = const()[name = string("op_42595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4469_cast_fp16, y = var_42595_to_fp16)[name = string("aw_chunk_4469_cast_fp16")]; + fp16 var_42597_to_fp16 = const()[name = string("op_42597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4471_cast_fp16, y = var_42597_to_fp16)[name = string("aw_chunk_4471_cast_fp16")]; + fp16 var_42599_to_fp16 = const()[name = string("op_42599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4473_cast_fp16, y = var_42599_to_fp16)[name = string("aw_chunk_4473_cast_fp16")]; + fp16 var_42601_to_fp16 = const()[name = string("op_42601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4475_cast_fp16, y = var_42601_to_fp16)[name = string("aw_chunk_4475_cast_fp16")]; + fp16 var_42603_to_fp16 = const()[name = string("op_42603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4477_cast_fp16, y = var_42603_to_fp16)[name = string("aw_chunk_4477_cast_fp16")]; + fp16 var_42605_to_fp16 = const()[name = string("op_42605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4479_cast_fp16, y = var_42605_to_fp16)[name = string("aw_chunk_4479_cast_fp16")]; + tensor var_42607_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4321_cast_fp16)[name = string("op_42607_cast_fp16")]; + tensor var_42608_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4323_cast_fp16)[name = string("op_42608_cast_fp16")]; + tensor var_42609_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4325_cast_fp16)[name = string("op_42609_cast_fp16")]; + tensor var_42610_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4327_cast_fp16)[name = string("op_42610_cast_fp16")]; + tensor var_42611_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4329_cast_fp16)[name = string("op_42611_cast_fp16")]; + tensor var_42612_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4331_cast_fp16)[name = string("op_42612_cast_fp16")]; + tensor var_42613_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4333_cast_fp16)[name = string("op_42613_cast_fp16")]; + tensor var_42614_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4335_cast_fp16)[name = string("op_42614_cast_fp16")]; + tensor var_42615_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4337_cast_fp16)[name = string("op_42615_cast_fp16")]; + tensor var_42616_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4339_cast_fp16)[name = string("op_42616_cast_fp16")]; + tensor var_42617_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4341_cast_fp16)[name = string("op_42617_cast_fp16")]; + tensor var_42618_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4343_cast_fp16)[name = string("op_42618_cast_fp16")]; + tensor var_42619_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4345_cast_fp16)[name = string("op_42619_cast_fp16")]; + tensor var_42620_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4347_cast_fp16)[name = string("op_42620_cast_fp16")]; + tensor var_42621_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4349_cast_fp16)[name = string("op_42621_cast_fp16")]; + tensor var_42622_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4351_cast_fp16)[name = string("op_42622_cast_fp16")]; + tensor var_42623_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4353_cast_fp16)[name = string("op_42623_cast_fp16")]; + tensor var_42624_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4355_cast_fp16)[name = string("op_42624_cast_fp16")]; + tensor var_42625_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4357_cast_fp16)[name = string("op_42625_cast_fp16")]; + tensor var_42626_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4359_cast_fp16)[name = string("op_42626_cast_fp16")]; + tensor var_42627_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4361_cast_fp16)[name = string("op_42627_cast_fp16")]; + tensor var_42628_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4363_cast_fp16)[name = string("op_42628_cast_fp16")]; + tensor var_42629_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4365_cast_fp16)[name = string("op_42629_cast_fp16")]; + tensor var_42630_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4367_cast_fp16)[name = string("op_42630_cast_fp16")]; + tensor var_42631_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4369_cast_fp16)[name = string("op_42631_cast_fp16")]; + tensor var_42632_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4371_cast_fp16)[name = string("op_42632_cast_fp16")]; + tensor var_42633_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4373_cast_fp16)[name = string("op_42633_cast_fp16")]; + tensor var_42634_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4375_cast_fp16)[name = string("op_42634_cast_fp16")]; + tensor var_42635_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4377_cast_fp16)[name = string("op_42635_cast_fp16")]; + tensor var_42636_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4379_cast_fp16)[name = string("op_42636_cast_fp16")]; + tensor var_42637_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4381_cast_fp16)[name = string("op_42637_cast_fp16")]; + tensor var_42638_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4383_cast_fp16)[name = string("op_42638_cast_fp16")]; + tensor var_42639_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4385_cast_fp16)[name = string("op_42639_cast_fp16")]; + tensor var_42640_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4387_cast_fp16)[name = string("op_42640_cast_fp16")]; + tensor var_42641_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4389_cast_fp16)[name = string("op_42641_cast_fp16")]; + tensor var_42642_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4391_cast_fp16)[name = string("op_42642_cast_fp16")]; + tensor var_42643_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4393_cast_fp16)[name = string("op_42643_cast_fp16")]; + tensor var_42644_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4395_cast_fp16)[name = string("op_42644_cast_fp16")]; + tensor var_42645_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4397_cast_fp16)[name = string("op_42645_cast_fp16")]; + tensor var_42646_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4399_cast_fp16)[name = string("op_42646_cast_fp16")]; + tensor var_42647_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4401_cast_fp16)[name = string("op_42647_cast_fp16")]; + tensor var_42648_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4403_cast_fp16)[name = string("op_42648_cast_fp16")]; + tensor var_42649_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4405_cast_fp16)[name = string("op_42649_cast_fp16")]; + tensor var_42650_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4407_cast_fp16)[name = string("op_42650_cast_fp16")]; + tensor var_42651_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4409_cast_fp16)[name = string("op_42651_cast_fp16")]; + tensor var_42652_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4411_cast_fp16)[name = string("op_42652_cast_fp16")]; + tensor var_42653_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4413_cast_fp16)[name = string("op_42653_cast_fp16")]; + tensor var_42654_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4415_cast_fp16)[name = string("op_42654_cast_fp16")]; + tensor var_42655_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4417_cast_fp16)[name = string("op_42655_cast_fp16")]; + tensor var_42656_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4419_cast_fp16)[name = string("op_42656_cast_fp16")]; + tensor var_42657_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4421_cast_fp16)[name = string("op_42657_cast_fp16")]; + tensor var_42658_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4423_cast_fp16)[name = string("op_42658_cast_fp16")]; + tensor var_42659_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4425_cast_fp16)[name = string("op_42659_cast_fp16")]; + tensor var_42660_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4427_cast_fp16)[name = string("op_42660_cast_fp16")]; + tensor var_42661_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4429_cast_fp16)[name = string("op_42661_cast_fp16")]; + tensor var_42662_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4431_cast_fp16)[name = string("op_42662_cast_fp16")]; + tensor var_42663_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4433_cast_fp16)[name = string("op_42663_cast_fp16")]; + tensor var_42664_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4435_cast_fp16)[name = string("op_42664_cast_fp16")]; + tensor var_42665_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4437_cast_fp16)[name = string("op_42665_cast_fp16")]; + tensor var_42666_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4439_cast_fp16)[name = string("op_42666_cast_fp16")]; + tensor var_42667_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4441_cast_fp16)[name = string("op_42667_cast_fp16")]; + tensor var_42668_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4443_cast_fp16)[name = string("op_42668_cast_fp16")]; + tensor var_42669_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4445_cast_fp16)[name = string("op_42669_cast_fp16")]; + tensor var_42670_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4447_cast_fp16)[name = string("op_42670_cast_fp16")]; + tensor var_42671_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4449_cast_fp16)[name = string("op_42671_cast_fp16")]; + tensor var_42672_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4451_cast_fp16)[name = string("op_42672_cast_fp16")]; + tensor var_42673_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4453_cast_fp16)[name = string("op_42673_cast_fp16")]; + tensor var_42674_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4455_cast_fp16)[name = string("op_42674_cast_fp16")]; + tensor var_42675_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4457_cast_fp16)[name = string("op_42675_cast_fp16")]; + tensor var_42676_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4459_cast_fp16)[name = string("op_42676_cast_fp16")]; + tensor var_42677_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4461_cast_fp16)[name = string("op_42677_cast_fp16")]; + tensor var_42678_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4463_cast_fp16)[name = string("op_42678_cast_fp16")]; + tensor var_42679_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4465_cast_fp16)[name = string("op_42679_cast_fp16")]; + tensor var_42680_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4467_cast_fp16)[name = string("op_42680_cast_fp16")]; + tensor var_42681_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4469_cast_fp16)[name = string("op_42681_cast_fp16")]; + tensor var_42682_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4471_cast_fp16)[name = string("op_42682_cast_fp16")]; + tensor var_42683_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4473_cast_fp16)[name = string("op_42683_cast_fp16")]; + tensor var_42684_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4475_cast_fp16)[name = string("op_42684_cast_fp16")]; + tensor var_42685_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4477_cast_fp16)[name = string("op_42685_cast_fp16")]; + tensor var_42686_cast_fp16 = softmax(axis = var_41432, x = aw_chunk_4479_cast_fp16)[name = string("op_42686_cast_fp16")]; + string var_42688_equation_0 = const()[name = string("op_42688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42688_cast_fp16 = einsum(equation = var_42688_equation_0, values = (var_42208_cast_fp16, var_42607_cast_fp16))[name = string("op_42688_cast_fp16")]; + string var_42690_equation_0 = const()[name = string("op_42690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42690_cast_fp16 = einsum(equation = var_42690_equation_0, values = (var_42208_cast_fp16, var_42608_cast_fp16))[name = string("op_42690_cast_fp16")]; + string var_42692_equation_0 = const()[name = string("op_42692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42692_cast_fp16 = einsum(equation = var_42692_equation_0, values = (var_42208_cast_fp16, var_42609_cast_fp16))[name = string("op_42692_cast_fp16")]; + string var_42694_equation_0 = const()[name = string("op_42694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42694_cast_fp16 = einsum(equation = var_42694_equation_0, values = (var_42208_cast_fp16, var_42610_cast_fp16))[name = string("op_42694_cast_fp16")]; + string var_42696_equation_0 = const()[name = string("op_42696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42696_cast_fp16 = einsum(equation = var_42696_equation_0, values = (var_42212_cast_fp16, var_42611_cast_fp16))[name = string("op_42696_cast_fp16")]; + string var_42698_equation_0 = const()[name = string("op_42698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42698_cast_fp16 = einsum(equation = var_42698_equation_0, values = (var_42212_cast_fp16, var_42612_cast_fp16))[name = string("op_42698_cast_fp16")]; + string var_42700_equation_0 = const()[name = string("op_42700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42700_cast_fp16 = einsum(equation = var_42700_equation_0, values = (var_42212_cast_fp16, var_42613_cast_fp16))[name = string("op_42700_cast_fp16")]; + string var_42702_equation_0 = const()[name = string("op_42702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42702_cast_fp16 = einsum(equation = var_42702_equation_0, values = (var_42212_cast_fp16, var_42614_cast_fp16))[name = string("op_42702_cast_fp16")]; + string var_42704_equation_0 = const()[name = string("op_42704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42704_cast_fp16 = einsum(equation = var_42704_equation_0, values = (var_42216_cast_fp16, var_42615_cast_fp16))[name = string("op_42704_cast_fp16")]; + string var_42706_equation_0 = const()[name = string("op_42706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42706_cast_fp16 = einsum(equation = var_42706_equation_0, values = (var_42216_cast_fp16, var_42616_cast_fp16))[name = string("op_42706_cast_fp16")]; + string var_42708_equation_0 = const()[name = string("op_42708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42708_cast_fp16 = einsum(equation = var_42708_equation_0, values = (var_42216_cast_fp16, var_42617_cast_fp16))[name = string("op_42708_cast_fp16")]; + string var_42710_equation_0 = const()[name = string("op_42710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42710_cast_fp16 = einsum(equation = var_42710_equation_0, values = (var_42216_cast_fp16, var_42618_cast_fp16))[name = string("op_42710_cast_fp16")]; + string var_42712_equation_0 = const()[name = string("op_42712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42712_cast_fp16 = einsum(equation = var_42712_equation_0, values = (var_42220_cast_fp16, var_42619_cast_fp16))[name = string("op_42712_cast_fp16")]; + string var_42714_equation_0 = const()[name = string("op_42714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42714_cast_fp16 = einsum(equation = var_42714_equation_0, values = (var_42220_cast_fp16, var_42620_cast_fp16))[name = string("op_42714_cast_fp16")]; + string var_42716_equation_0 = const()[name = string("op_42716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42716_cast_fp16 = einsum(equation = var_42716_equation_0, values = (var_42220_cast_fp16, var_42621_cast_fp16))[name = string("op_42716_cast_fp16")]; + string var_42718_equation_0 = const()[name = string("op_42718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42718_cast_fp16 = einsum(equation = var_42718_equation_0, values = (var_42220_cast_fp16, var_42622_cast_fp16))[name = string("op_42718_cast_fp16")]; + string var_42720_equation_0 = const()[name = string("op_42720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42720_cast_fp16 = einsum(equation = var_42720_equation_0, values = (var_42224_cast_fp16, var_42623_cast_fp16))[name = string("op_42720_cast_fp16")]; + string var_42722_equation_0 = const()[name = string("op_42722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42722_cast_fp16 = einsum(equation = var_42722_equation_0, values = (var_42224_cast_fp16, var_42624_cast_fp16))[name = string("op_42722_cast_fp16")]; + string var_42724_equation_0 = const()[name = string("op_42724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42724_cast_fp16 = einsum(equation = var_42724_equation_0, values = (var_42224_cast_fp16, var_42625_cast_fp16))[name = string("op_42724_cast_fp16")]; + string var_42726_equation_0 = const()[name = string("op_42726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42726_cast_fp16 = einsum(equation = var_42726_equation_0, values = (var_42224_cast_fp16, var_42626_cast_fp16))[name = string("op_42726_cast_fp16")]; + string var_42728_equation_0 = const()[name = string("op_42728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42728_cast_fp16 = einsum(equation = var_42728_equation_0, values = (var_42228_cast_fp16, var_42627_cast_fp16))[name = string("op_42728_cast_fp16")]; + string var_42730_equation_0 = const()[name = string("op_42730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42730_cast_fp16 = einsum(equation = var_42730_equation_0, values = (var_42228_cast_fp16, var_42628_cast_fp16))[name = string("op_42730_cast_fp16")]; + string var_42732_equation_0 = const()[name = string("op_42732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42732_cast_fp16 = einsum(equation = var_42732_equation_0, values = (var_42228_cast_fp16, var_42629_cast_fp16))[name = string("op_42732_cast_fp16")]; + string var_42734_equation_0 = const()[name = string("op_42734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42734_cast_fp16 = einsum(equation = var_42734_equation_0, values = (var_42228_cast_fp16, var_42630_cast_fp16))[name = string("op_42734_cast_fp16")]; + string var_42736_equation_0 = const()[name = string("op_42736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42736_cast_fp16 = einsum(equation = var_42736_equation_0, values = (var_42232_cast_fp16, var_42631_cast_fp16))[name = string("op_42736_cast_fp16")]; + string var_42738_equation_0 = const()[name = string("op_42738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42738_cast_fp16 = einsum(equation = var_42738_equation_0, values = (var_42232_cast_fp16, var_42632_cast_fp16))[name = string("op_42738_cast_fp16")]; + string var_42740_equation_0 = const()[name = string("op_42740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42740_cast_fp16 = einsum(equation = var_42740_equation_0, values = (var_42232_cast_fp16, var_42633_cast_fp16))[name = string("op_42740_cast_fp16")]; + string var_42742_equation_0 = const()[name = string("op_42742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42742_cast_fp16 = einsum(equation = var_42742_equation_0, values = (var_42232_cast_fp16, var_42634_cast_fp16))[name = string("op_42742_cast_fp16")]; + string var_42744_equation_0 = const()[name = string("op_42744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42744_cast_fp16 = einsum(equation = var_42744_equation_0, values = (var_42236_cast_fp16, var_42635_cast_fp16))[name = string("op_42744_cast_fp16")]; + string var_42746_equation_0 = const()[name = string("op_42746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42746_cast_fp16 = einsum(equation = var_42746_equation_0, values = (var_42236_cast_fp16, var_42636_cast_fp16))[name = string("op_42746_cast_fp16")]; + string var_42748_equation_0 = const()[name = string("op_42748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42748_cast_fp16 = einsum(equation = var_42748_equation_0, values = (var_42236_cast_fp16, var_42637_cast_fp16))[name = string("op_42748_cast_fp16")]; + string var_42750_equation_0 = const()[name = string("op_42750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42750_cast_fp16 = einsum(equation = var_42750_equation_0, values = (var_42236_cast_fp16, var_42638_cast_fp16))[name = string("op_42750_cast_fp16")]; + string var_42752_equation_0 = const()[name = string("op_42752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42752_cast_fp16 = einsum(equation = var_42752_equation_0, values = (var_42240_cast_fp16, var_42639_cast_fp16))[name = string("op_42752_cast_fp16")]; + string var_42754_equation_0 = const()[name = string("op_42754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42754_cast_fp16 = einsum(equation = var_42754_equation_0, values = (var_42240_cast_fp16, var_42640_cast_fp16))[name = string("op_42754_cast_fp16")]; + string var_42756_equation_0 = const()[name = string("op_42756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42756_cast_fp16 = einsum(equation = var_42756_equation_0, values = (var_42240_cast_fp16, var_42641_cast_fp16))[name = string("op_42756_cast_fp16")]; + string var_42758_equation_0 = const()[name = string("op_42758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42758_cast_fp16 = einsum(equation = var_42758_equation_0, values = (var_42240_cast_fp16, var_42642_cast_fp16))[name = string("op_42758_cast_fp16")]; + string var_42760_equation_0 = const()[name = string("op_42760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42760_cast_fp16 = einsum(equation = var_42760_equation_0, values = (var_42244_cast_fp16, var_42643_cast_fp16))[name = string("op_42760_cast_fp16")]; + string var_42762_equation_0 = const()[name = string("op_42762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42762_cast_fp16 = einsum(equation = var_42762_equation_0, values = (var_42244_cast_fp16, var_42644_cast_fp16))[name = string("op_42762_cast_fp16")]; + string var_42764_equation_0 = const()[name = string("op_42764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42764_cast_fp16 = einsum(equation = var_42764_equation_0, values = (var_42244_cast_fp16, var_42645_cast_fp16))[name = string("op_42764_cast_fp16")]; + string var_42766_equation_0 = const()[name = string("op_42766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42766_cast_fp16 = einsum(equation = var_42766_equation_0, values = (var_42244_cast_fp16, var_42646_cast_fp16))[name = string("op_42766_cast_fp16")]; + string var_42768_equation_0 = const()[name = string("op_42768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42768_cast_fp16 = einsum(equation = var_42768_equation_0, values = (var_42248_cast_fp16, var_42647_cast_fp16))[name = string("op_42768_cast_fp16")]; + string var_42770_equation_0 = const()[name = string("op_42770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42770_cast_fp16 = einsum(equation = var_42770_equation_0, values = (var_42248_cast_fp16, var_42648_cast_fp16))[name = string("op_42770_cast_fp16")]; + string var_42772_equation_0 = const()[name = string("op_42772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42772_cast_fp16 = einsum(equation = var_42772_equation_0, values = (var_42248_cast_fp16, var_42649_cast_fp16))[name = string("op_42772_cast_fp16")]; + string var_42774_equation_0 = const()[name = string("op_42774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42774_cast_fp16 = einsum(equation = var_42774_equation_0, values = (var_42248_cast_fp16, var_42650_cast_fp16))[name = string("op_42774_cast_fp16")]; + string var_42776_equation_0 = const()[name = string("op_42776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42776_cast_fp16 = einsum(equation = var_42776_equation_0, values = (var_42252_cast_fp16, var_42651_cast_fp16))[name = string("op_42776_cast_fp16")]; + string var_42778_equation_0 = const()[name = string("op_42778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42778_cast_fp16 = einsum(equation = var_42778_equation_0, values = (var_42252_cast_fp16, var_42652_cast_fp16))[name = string("op_42778_cast_fp16")]; + string var_42780_equation_0 = const()[name = string("op_42780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42780_cast_fp16 = einsum(equation = var_42780_equation_0, values = (var_42252_cast_fp16, var_42653_cast_fp16))[name = string("op_42780_cast_fp16")]; + string var_42782_equation_0 = const()[name = string("op_42782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42782_cast_fp16 = einsum(equation = var_42782_equation_0, values = (var_42252_cast_fp16, var_42654_cast_fp16))[name = string("op_42782_cast_fp16")]; + string var_42784_equation_0 = const()[name = string("op_42784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42784_cast_fp16 = einsum(equation = var_42784_equation_0, values = (var_42256_cast_fp16, var_42655_cast_fp16))[name = string("op_42784_cast_fp16")]; + string var_42786_equation_0 = const()[name = string("op_42786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42786_cast_fp16 = einsum(equation = var_42786_equation_0, values = (var_42256_cast_fp16, var_42656_cast_fp16))[name = string("op_42786_cast_fp16")]; + string var_42788_equation_0 = const()[name = string("op_42788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42788_cast_fp16 = einsum(equation = var_42788_equation_0, values = (var_42256_cast_fp16, var_42657_cast_fp16))[name = string("op_42788_cast_fp16")]; + string var_42790_equation_0 = const()[name = string("op_42790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42790_cast_fp16 = einsum(equation = var_42790_equation_0, values = (var_42256_cast_fp16, var_42658_cast_fp16))[name = string("op_42790_cast_fp16")]; + string var_42792_equation_0 = const()[name = string("op_42792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42792_cast_fp16 = einsum(equation = var_42792_equation_0, values = (var_42260_cast_fp16, var_42659_cast_fp16))[name = string("op_42792_cast_fp16")]; + string var_42794_equation_0 = const()[name = string("op_42794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42794_cast_fp16 = einsum(equation = var_42794_equation_0, values = (var_42260_cast_fp16, var_42660_cast_fp16))[name = string("op_42794_cast_fp16")]; + string var_42796_equation_0 = const()[name = string("op_42796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42796_cast_fp16 = einsum(equation = var_42796_equation_0, values = (var_42260_cast_fp16, var_42661_cast_fp16))[name = string("op_42796_cast_fp16")]; + string var_42798_equation_0 = const()[name = string("op_42798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42798_cast_fp16 = einsum(equation = var_42798_equation_0, values = (var_42260_cast_fp16, var_42662_cast_fp16))[name = string("op_42798_cast_fp16")]; + string var_42800_equation_0 = const()[name = string("op_42800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42800_cast_fp16 = einsum(equation = var_42800_equation_0, values = (var_42264_cast_fp16, var_42663_cast_fp16))[name = string("op_42800_cast_fp16")]; + string var_42802_equation_0 = const()[name = string("op_42802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42802_cast_fp16 = einsum(equation = var_42802_equation_0, values = (var_42264_cast_fp16, var_42664_cast_fp16))[name = string("op_42802_cast_fp16")]; + string var_42804_equation_0 = const()[name = string("op_42804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42804_cast_fp16 = einsum(equation = var_42804_equation_0, values = (var_42264_cast_fp16, var_42665_cast_fp16))[name = string("op_42804_cast_fp16")]; + string var_42806_equation_0 = const()[name = string("op_42806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42806_cast_fp16 = einsum(equation = var_42806_equation_0, values = (var_42264_cast_fp16, var_42666_cast_fp16))[name = string("op_42806_cast_fp16")]; + string var_42808_equation_0 = const()[name = string("op_42808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42808_cast_fp16 = einsum(equation = var_42808_equation_0, values = (var_42268_cast_fp16, var_42667_cast_fp16))[name = string("op_42808_cast_fp16")]; + string var_42810_equation_0 = const()[name = string("op_42810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42810_cast_fp16 = einsum(equation = var_42810_equation_0, values = (var_42268_cast_fp16, var_42668_cast_fp16))[name = string("op_42810_cast_fp16")]; + string var_42812_equation_0 = const()[name = string("op_42812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42812_cast_fp16 = einsum(equation = var_42812_equation_0, values = (var_42268_cast_fp16, var_42669_cast_fp16))[name = string("op_42812_cast_fp16")]; + string var_42814_equation_0 = const()[name = string("op_42814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42814_cast_fp16 = einsum(equation = var_42814_equation_0, values = (var_42268_cast_fp16, var_42670_cast_fp16))[name = string("op_42814_cast_fp16")]; + string var_42816_equation_0 = const()[name = string("op_42816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42816_cast_fp16 = einsum(equation = var_42816_equation_0, values = (var_42272_cast_fp16, var_42671_cast_fp16))[name = string("op_42816_cast_fp16")]; + string var_42818_equation_0 = const()[name = string("op_42818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42818_cast_fp16 = einsum(equation = var_42818_equation_0, values = (var_42272_cast_fp16, var_42672_cast_fp16))[name = string("op_42818_cast_fp16")]; + string var_42820_equation_0 = const()[name = string("op_42820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42820_cast_fp16 = einsum(equation = var_42820_equation_0, values = (var_42272_cast_fp16, var_42673_cast_fp16))[name = string("op_42820_cast_fp16")]; + string var_42822_equation_0 = const()[name = string("op_42822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42822_cast_fp16 = einsum(equation = var_42822_equation_0, values = (var_42272_cast_fp16, var_42674_cast_fp16))[name = string("op_42822_cast_fp16")]; + string var_42824_equation_0 = const()[name = string("op_42824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42824_cast_fp16 = einsum(equation = var_42824_equation_0, values = (var_42276_cast_fp16, var_42675_cast_fp16))[name = string("op_42824_cast_fp16")]; + string var_42826_equation_0 = const()[name = string("op_42826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42826_cast_fp16 = einsum(equation = var_42826_equation_0, values = (var_42276_cast_fp16, var_42676_cast_fp16))[name = string("op_42826_cast_fp16")]; + string var_42828_equation_0 = const()[name = string("op_42828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42828_cast_fp16 = einsum(equation = var_42828_equation_0, values = (var_42276_cast_fp16, var_42677_cast_fp16))[name = string("op_42828_cast_fp16")]; + string var_42830_equation_0 = const()[name = string("op_42830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42830_cast_fp16 = einsum(equation = var_42830_equation_0, values = (var_42276_cast_fp16, var_42678_cast_fp16))[name = string("op_42830_cast_fp16")]; + string var_42832_equation_0 = const()[name = string("op_42832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42832_cast_fp16 = einsum(equation = var_42832_equation_0, values = (var_42280_cast_fp16, var_42679_cast_fp16))[name = string("op_42832_cast_fp16")]; + string var_42834_equation_0 = const()[name = string("op_42834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42834_cast_fp16 = einsum(equation = var_42834_equation_0, values = (var_42280_cast_fp16, var_42680_cast_fp16))[name = string("op_42834_cast_fp16")]; + string var_42836_equation_0 = const()[name = string("op_42836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42836_cast_fp16 = einsum(equation = var_42836_equation_0, values = (var_42280_cast_fp16, var_42681_cast_fp16))[name = string("op_42836_cast_fp16")]; + string var_42838_equation_0 = const()[name = string("op_42838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42838_cast_fp16 = einsum(equation = var_42838_equation_0, values = (var_42280_cast_fp16, var_42682_cast_fp16))[name = string("op_42838_cast_fp16")]; + string var_42840_equation_0 = const()[name = string("op_42840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42840_cast_fp16 = einsum(equation = var_42840_equation_0, values = (var_42284_cast_fp16, var_42683_cast_fp16))[name = string("op_42840_cast_fp16")]; + string var_42842_equation_0 = const()[name = string("op_42842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42842_cast_fp16 = einsum(equation = var_42842_equation_0, values = (var_42284_cast_fp16, var_42684_cast_fp16))[name = string("op_42842_cast_fp16")]; + string var_42844_equation_0 = const()[name = string("op_42844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42844_cast_fp16 = einsum(equation = var_42844_equation_0, values = (var_42284_cast_fp16, var_42685_cast_fp16))[name = string("op_42844_cast_fp16")]; + string var_42846_equation_0 = const()[name = string("op_42846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42846_cast_fp16 = einsum(equation = var_42846_equation_0, values = (var_42284_cast_fp16, var_42686_cast_fp16))[name = string("op_42846_cast_fp16")]; + bool var_42848_interleave_0 = const()[name = string("op_42848_interleave_0"), val = bool(false)]; + tensor var_42848_cast_fp16 = concat(axis = var_41407, interleave = var_42848_interleave_0, values = (var_42688_cast_fp16, var_42690_cast_fp16, var_42692_cast_fp16, var_42694_cast_fp16))[name = string("op_42848_cast_fp16")]; + bool var_42850_interleave_0 = const()[name = string("op_42850_interleave_0"), val = bool(false)]; + tensor var_42850_cast_fp16 = concat(axis = var_41407, interleave = var_42850_interleave_0, values = (var_42696_cast_fp16, var_42698_cast_fp16, var_42700_cast_fp16, var_42702_cast_fp16))[name = string("op_42850_cast_fp16")]; + bool var_42852_interleave_0 = const()[name = string("op_42852_interleave_0"), val = bool(false)]; + tensor var_42852_cast_fp16 = concat(axis = var_41407, interleave = var_42852_interleave_0, values = (var_42704_cast_fp16, var_42706_cast_fp16, var_42708_cast_fp16, var_42710_cast_fp16))[name = string("op_42852_cast_fp16")]; + bool var_42854_interleave_0 = const()[name = string("op_42854_interleave_0"), val = bool(false)]; + tensor var_42854_cast_fp16 = concat(axis = var_41407, interleave = var_42854_interleave_0, values = (var_42712_cast_fp16, var_42714_cast_fp16, var_42716_cast_fp16, var_42718_cast_fp16))[name = string("op_42854_cast_fp16")]; + bool var_42856_interleave_0 = const()[name = string("op_42856_interleave_0"), val = bool(false)]; + tensor var_42856_cast_fp16 = concat(axis = var_41407, interleave = var_42856_interleave_0, values = (var_42720_cast_fp16, var_42722_cast_fp16, var_42724_cast_fp16, var_42726_cast_fp16))[name = string("op_42856_cast_fp16")]; + bool var_42858_interleave_0 = const()[name = string("op_42858_interleave_0"), val = bool(false)]; + tensor var_42858_cast_fp16 = concat(axis = var_41407, interleave = var_42858_interleave_0, values = (var_42728_cast_fp16, var_42730_cast_fp16, var_42732_cast_fp16, var_42734_cast_fp16))[name = string("op_42858_cast_fp16")]; + bool var_42860_interleave_0 = const()[name = string("op_42860_interleave_0"), val = bool(false)]; + tensor var_42860_cast_fp16 = concat(axis = var_41407, interleave = var_42860_interleave_0, values = (var_42736_cast_fp16, var_42738_cast_fp16, var_42740_cast_fp16, var_42742_cast_fp16))[name = string("op_42860_cast_fp16")]; + bool var_42862_interleave_0 = const()[name = string("op_42862_interleave_0"), val = bool(false)]; + tensor var_42862_cast_fp16 = concat(axis = var_41407, interleave = var_42862_interleave_0, values = (var_42744_cast_fp16, var_42746_cast_fp16, var_42748_cast_fp16, var_42750_cast_fp16))[name = string("op_42862_cast_fp16")]; + bool var_42864_interleave_0 = const()[name = string("op_42864_interleave_0"), val = bool(false)]; + tensor var_42864_cast_fp16 = concat(axis = var_41407, interleave = var_42864_interleave_0, values = (var_42752_cast_fp16, var_42754_cast_fp16, var_42756_cast_fp16, var_42758_cast_fp16))[name = string("op_42864_cast_fp16")]; + bool var_42866_interleave_0 = const()[name = string("op_42866_interleave_0"), val = bool(false)]; + tensor var_42866_cast_fp16 = concat(axis = var_41407, interleave = var_42866_interleave_0, values = (var_42760_cast_fp16, var_42762_cast_fp16, var_42764_cast_fp16, var_42766_cast_fp16))[name = string("op_42866_cast_fp16")]; + bool var_42868_interleave_0 = const()[name = string("op_42868_interleave_0"), val = bool(false)]; + tensor var_42868_cast_fp16 = concat(axis = var_41407, interleave = var_42868_interleave_0, values = (var_42768_cast_fp16, var_42770_cast_fp16, var_42772_cast_fp16, var_42774_cast_fp16))[name = string("op_42868_cast_fp16")]; + bool var_42870_interleave_0 = const()[name = string("op_42870_interleave_0"), val = bool(false)]; + tensor var_42870_cast_fp16 = concat(axis = var_41407, interleave = var_42870_interleave_0, values = (var_42776_cast_fp16, var_42778_cast_fp16, var_42780_cast_fp16, var_42782_cast_fp16))[name = string("op_42870_cast_fp16")]; + bool var_42872_interleave_0 = const()[name = string("op_42872_interleave_0"), val = bool(false)]; + tensor var_42872_cast_fp16 = concat(axis = var_41407, interleave = var_42872_interleave_0, values = (var_42784_cast_fp16, var_42786_cast_fp16, var_42788_cast_fp16, var_42790_cast_fp16))[name = string("op_42872_cast_fp16")]; + bool var_42874_interleave_0 = const()[name = string("op_42874_interleave_0"), val = bool(false)]; + tensor var_42874_cast_fp16 = concat(axis = var_41407, interleave = var_42874_interleave_0, values = (var_42792_cast_fp16, var_42794_cast_fp16, var_42796_cast_fp16, var_42798_cast_fp16))[name = string("op_42874_cast_fp16")]; + bool var_42876_interleave_0 = const()[name = string("op_42876_interleave_0"), val = bool(false)]; + tensor var_42876_cast_fp16 = concat(axis = var_41407, interleave = var_42876_interleave_0, values = (var_42800_cast_fp16, var_42802_cast_fp16, var_42804_cast_fp16, var_42806_cast_fp16))[name = string("op_42876_cast_fp16")]; + bool var_42878_interleave_0 = const()[name = string("op_42878_interleave_0"), val = bool(false)]; + tensor var_42878_cast_fp16 = concat(axis = var_41407, interleave = var_42878_interleave_0, values = (var_42808_cast_fp16, var_42810_cast_fp16, var_42812_cast_fp16, var_42814_cast_fp16))[name = string("op_42878_cast_fp16")]; + bool var_42880_interleave_0 = const()[name = string("op_42880_interleave_0"), val = bool(false)]; + tensor var_42880_cast_fp16 = concat(axis = var_41407, interleave = var_42880_interleave_0, values = (var_42816_cast_fp16, var_42818_cast_fp16, var_42820_cast_fp16, var_42822_cast_fp16))[name = string("op_42880_cast_fp16")]; + bool var_42882_interleave_0 = const()[name = string("op_42882_interleave_0"), val = bool(false)]; + tensor var_42882_cast_fp16 = concat(axis = var_41407, interleave = var_42882_interleave_0, values = (var_42824_cast_fp16, var_42826_cast_fp16, var_42828_cast_fp16, var_42830_cast_fp16))[name = string("op_42882_cast_fp16")]; + bool var_42884_interleave_0 = const()[name = string("op_42884_interleave_0"), val = bool(false)]; + tensor var_42884_cast_fp16 = concat(axis = var_41407, interleave = var_42884_interleave_0, values = (var_42832_cast_fp16, var_42834_cast_fp16, var_42836_cast_fp16, var_42838_cast_fp16))[name = string("op_42884_cast_fp16")]; + bool var_42886_interleave_0 = const()[name = string("op_42886_interleave_0"), val = bool(false)]; + tensor var_42886_cast_fp16 = concat(axis = var_41407, interleave = var_42886_interleave_0, values = (var_42840_cast_fp16, var_42842_cast_fp16, var_42844_cast_fp16, var_42846_cast_fp16))[name = string("op_42886_cast_fp16")]; + bool input_217_interleave_0 = const()[name = string("input_217_interleave_0"), val = bool(false)]; + tensor input_217_cast_fp16 = concat(axis = var_41432, interleave = input_217_interleave_0, values = (var_42848_cast_fp16, var_42850_cast_fp16, var_42852_cast_fp16, var_42854_cast_fp16, var_42856_cast_fp16, var_42858_cast_fp16, var_42860_cast_fp16, var_42862_cast_fp16, var_42864_cast_fp16, var_42866_cast_fp16, var_42868_cast_fp16, var_42870_cast_fp16, var_42872_cast_fp16, var_42874_cast_fp16, var_42876_cast_fp16, var_42878_cast_fp16, var_42880_cast_fp16, var_42882_cast_fp16, var_42884_cast_fp16, var_42886_cast_fp16))[name = string("input_217_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087043840)))]; + tensor layers_27_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090320704)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_27_self_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_27_self_attn_o_proj_weight_to_fp16, x = input_217_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_111_cast_fp16")]; + tensor out_111_axes_0 = const()[name = string("out_111_axes_0"), val = tensor([1])]; + fp16 var_42905_to_fp16 = const()[name = string("op_42905_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_42905_to_fp16, x = inputs_111_cast_fp16)[name = string("out_111_cast_fp16")]; + tensor input_219_gamma_0_to_fp16 = const()[name = string("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090323328)))]; + tensor input_219_beta_0_to_fp16 = const()[name = string("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090325952)))]; + fp16 input_219_epsilon_0_to_fp16 = const()[name = string("input_219_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = string("input_219_cast_fp16")]; + string input_221_pad_type_0 = const()[name = string("input_221_pad_type_0"), val = string("valid")]; + tensor input_221_strides_0 = const()[name = string("input_221_strides_0"), val = tensor([1, 1])]; + tensor input_221_pad_0 = const()[name = string("input_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_221_dilations_0 = const()[name = string("input_221_dilations_0"), val = tensor([1, 1])]; + int32 input_221_groups_0 = const()[name = string("input_221_groups_0"), val = int32(1)]; + tensor layers_27_fc1_weight_to_fp16 = const()[name = string("layers_27_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090328576)))]; + tensor layers_27_fc1_bias_to_fp16 = const()[name = string("layers_27_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103435840)))]; + tensor input_221_cast_fp16 = conv(bias = layers_27_fc1_bias_to_fp16, dilations = input_221_dilations_0, groups = input_221_groups_0, pad = input_221_pad_0, pad_type = input_221_pad_type_0, strides = input_221_strides_0, weight = layers_27_fc1_weight_to_fp16, x = input_219_cast_fp16)[name = string("input_221_cast_fp16")]; + string input_223_mode_0 = const()[name = string("input_223_mode_0"), val = string("EXACT")]; + tensor input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")]; + string hidden_states_59_pad_type_0 = const()[name = string("hidden_states_59_pad_type_0"), val = string("valid")]; + tensor hidden_states_59_strides_0 = const()[name = string("hidden_states_59_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_59_pad_0 = const()[name = string("hidden_states_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_59_dilations_0 = const()[name = string("hidden_states_59_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_59_groups_0 = const()[name = string("hidden_states_59_groups_0"), val = int32(1)]; + tensor layers_27_fc2_weight_to_fp16 = const()[name = string("layers_27_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103446144)))]; + tensor layers_27_fc2_bias_to_fp16 = const()[name = string("layers_27_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116553408)))]; + tensor hidden_states_59_cast_fp16 = conv(bias = layers_27_fc2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = layers_27_fc2_weight_to_fp16, x = input_223_cast_fp16)[name = string("hidden_states_59_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = hidden_states_59_cast_fp16)[name = string("inputs_113_cast_fp16")]; + int32 var_42934 = const()[name = string("op_42934"), val = int32(3)]; + int32 var_42959 = const()[name = string("op_42959"), val = int32(1)]; + tensor out_113_axes_0 = const()[name = string("out_113_axes_0"), val = tensor([1])]; + fp16 var_42976_to_fp16 = const()[name = string("op_42976_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_42976_to_fp16, x = inputs_113_cast_fp16)[name = string("out_113_cast_fp16")]; + tensor obj_113_gamma_0_to_fp16 = const()[name = string("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116556032)))]; + tensor obj_113_beta_0_to_fp16 = const()[name = string("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116558656)))]; + fp16 obj_113_epsilon_0_to_fp16 = const()[name = string("obj_113_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = string("obj_113_cast_fp16")]; + string query_57_pad_type_0 = const()[name = string("query_57_pad_type_0"), val = string("valid")]; + tensor query_57_strides_0 = const()[name = string("query_57_strides_0"), val = tensor([1, 1])]; + tensor query_57_pad_0 = const()[name = string("query_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_57_dilations_0 = const()[name = string("query_57_dilations_0"), val = tensor([1, 1])]; + int32 query_57_groups_0 = const()[name = string("query_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116561280)))]; + tensor layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119838144)))]; + tensor query_57_cast_fp16 = conv(bias = layers_28_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_28_self_attn_q_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("query_57_cast_fp16")]; + string key_57_pad_type_0 = const()[name = string("key_57_pad_type_0"), val = string("valid")]; + tensor key_57_strides_0 = const()[name = string("key_57_strides_0"), val = tensor([1, 1])]; + tensor key_57_pad_0 = const()[name = string("key_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_57_dilations_0 = const()[name = string("key_57_dilations_0"), val = tensor([1, 1])]; + int32 key_57_groups_0 = const()[name = string("key_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119840768)))]; + tensor key_57_cast_fp16 = conv(dilations = key_57_dilations_0, groups = key_57_groups_0, pad = key_57_pad_0, pad_type = key_57_pad_type_0, strides = key_57_strides_0, weight = layers_28_self_attn_k_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("key_57_cast_fp16")]; + string value_57_pad_type_0 = const()[name = string("value_57_pad_type_0"), val = string("valid")]; + tensor value_57_strides_0 = const()[name = string("value_57_strides_0"), val = tensor([1, 1])]; + tensor value_57_pad_0 = const()[name = string("value_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_57_dilations_0 = const()[name = string("value_57_dilations_0"), val = tensor([1, 1])]; + int32 value_57_groups_0 = const()[name = string("value_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1123117632)))]; + tensor layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126394496)))]; + tensor value_57_cast_fp16 = conv(bias = layers_28_self_attn_v_proj_bias_to_fp16, dilations = value_57_dilations_0, groups = value_57_groups_0, pad = value_57_pad_0, pad_type = value_57_pad_type_0, strides = value_57_strides_0, weight = layers_28_self_attn_v_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("value_57_cast_fp16")]; + tensor var_43014_begin_0 = const()[name = string("op_43014_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43014_end_0 = const()[name = string("op_43014_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43014_end_mask_0 = const()[name = string("op_43014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43014_cast_fp16 = slice_by_index(begin = var_43014_begin_0, end = var_43014_end_0, end_mask = var_43014_end_mask_0, x = query_57_cast_fp16)[name = string("op_43014_cast_fp16")]; + tensor var_43018_begin_0 = const()[name = string("op_43018_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43018_end_0 = const()[name = string("op_43018_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_43018_end_mask_0 = const()[name = string("op_43018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43018_cast_fp16 = slice_by_index(begin = var_43018_begin_0, end = var_43018_end_0, end_mask = var_43018_end_mask_0, x = query_57_cast_fp16)[name = string("op_43018_cast_fp16")]; + tensor var_43022_begin_0 = const()[name = string("op_43022_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43022_end_0 = const()[name = string("op_43022_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_43022_end_mask_0 = const()[name = string("op_43022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43022_cast_fp16 = slice_by_index(begin = var_43022_begin_0, end = var_43022_end_0, end_mask = var_43022_end_mask_0, x = query_57_cast_fp16)[name = string("op_43022_cast_fp16")]; + tensor var_43026_begin_0 = const()[name = string("op_43026_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43026_end_0 = const()[name = string("op_43026_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_43026_end_mask_0 = const()[name = string("op_43026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43026_cast_fp16 = slice_by_index(begin = var_43026_begin_0, end = var_43026_end_0, end_mask = var_43026_end_mask_0, x = query_57_cast_fp16)[name = string("op_43026_cast_fp16")]; + tensor var_43030_begin_0 = const()[name = string("op_43030_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43030_end_0 = const()[name = string("op_43030_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_43030_end_mask_0 = const()[name = string("op_43030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43030_cast_fp16 = slice_by_index(begin = var_43030_begin_0, end = var_43030_end_0, end_mask = var_43030_end_mask_0, x = query_57_cast_fp16)[name = string("op_43030_cast_fp16")]; + tensor var_43034_begin_0 = const()[name = string("op_43034_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43034_end_0 = const()[name = string("op_43034_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_43034_end_mask_0 = const()[name = string("op_43034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43034_cast_fp16 = slice_by_index(begin = var_43034_begin_0, end = var_43034_end_0, end_mask = var_43034_end_mask_0, x = query_57_cast_fp16)[name = string("op_43034_cast_fp16")]; + tensor var_43038_begin_0 = const()[name = string("op_43038_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43038_end_0 = const()[name = string("op_43038_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_43038_end_mask_0 = const()[name = string("op_43038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43038_cast_fp16 = slice_by_index(begin = var_43038_begin_0, end = var_43038_end_0, end_mask = var_43038_end_mask_0, x = query_57_cast_fp16)[name = string("op_43038_cast_fp16")]; + tensor var_43042_begin_0 = const()[name = string("op_43042_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43042_end_0 = const()[name = string("op_43042_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_43042_end_mask_0 = const()[name = string("op_43042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43042_cast_fp16 = slice_by_index(begin = var_43042_begin_0, end = var_43042_end_0, end_mask = var_43042_end_mask_0, x = query_57_cast_fp16)[name = string("op_43042_cast_fp16")]; + tensor var_43046_begin_0 = const()[name = string("op_43046_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43046_end_0 = const()[name = string("op_43046_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_43046_end_mask_0 = const()[name = string("op_43046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43046_cast_fp16 = slice_by_index(begin = var_43046_begin_0, end = var_43046_end_0, end_mask = var_43046_end_mask_0, x = query_57_cast_fp16)[name = string("op_43046_cast_fp16")]; + tensor var_43050_begin_0 = const()[name = string("op_43050_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43050_end_0 = const()[name = string("op_43050_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_43050_end_mask_0 = const()[name = string("op_43050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43050_cast_fp16 = slice_by_index(begin = var_43050_begin_0, end = var_43050_end_0, end_mask = var_43050_end_mask_0, x = query_57_cast_fp16)[name = string("op_43050_cast_fp16")]; + tensor var_43054_begin_0 = const()[name = string("op_43054_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43054_end_0 = const()[name = string("op_43054_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_43054_end_mask_0 = const()[name = string("op_43054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43054_cast_fp16 = slice_by_index(begin = var_43054_begin_0, end = var_43054_end_0, end_mask = var_43054_end_mask_0, x = query_57_cast_fp16)[name = string("op_43054_cast_fp16")]; + tensor var_43058_begin_0 = const()[name = string("op_43058_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43058_end_0 = const()[name = string("op_43058_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_43058_end_mask_0 = const()[name = string("op_43058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43058_cast_fp16 = slice_by_index(begin = var_43058_begin_0, end = var_43058_end_0, end_mask = var_43058_end_mask_0, x = query_57_cast_fp16)[name = string("op_43058_cast_fp16")]; + tensor var_43062_begin_0 = const()[name = string("op_43062_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43062_end_0 = const()[name = string("op_43062_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_43062_end_mask_0 = const()[name = string("op_43062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43062_cast_fp16 = slice_by_index(begin = var_43062_begin_0, end = var_43062_end_0, end_mask = var_43062_end_mask_0, x = query_57_cast_fp16)[name = string("op_43062_cast_fp16")]; + tensor var_43066_begin_0 = const()[name = string("op_43066_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43066_end_0 = const()[name = string("op_43066_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_43066_end_mask_0 = const()[name = string("op_43066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43066_cast_fp16 = slice_by_index(begin = var_43066_begin_0, end = var_43066_end_0, end_mask = var_43066_end_mask_0, x = query_57_cast_fp16)[name = string("op_43066_cast_fp16")]; + tensor var_43070_begin_0 = const()[name = string("op_43070_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43070_end_0 = const()[name = string("op_43070_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_43070_end_mask_0 = const()[name = string("op_43070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43070_cast_fp16 = slice_by_index(begin = var_43070_begin_0, end = var_43070_end_0, end_mask = var_43070_end_mask_0, x = query_57_cast_fp16)[name = string("op_43070_cast_fp16")]; + tensor var_43074_begin_0 = const()[name = string("op_43074_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43074_end_0 = const()[name = string("op_43074_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_43074_end_mask_0 = const()[name = string("op_43074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43074_cast_fp16 = slice_by_index(begin = var_43074_begin_0, end = var_43074_end_0, end_mask = var_43074_end_mask_0, x = query_57_cast_fp16)[name = string("op_43074_cast_fp16")]; + tensor var_43078_begin_0 = const()[name = string("op_43078_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43078_end_0 = const()[name = string("op_43078_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_43078_end_mask_0 = const()[name = string("op_43078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43078_cast_fp16 = slice_by_index(begin = var_43078_begin_0, end = var_43078_end_0, end_mask = var_43078_end_mask_0, x = query_57_cast_fp16)[name = string("op_43078_cast_fp16")]; + tensor var_43082_begin_0 = const()[name = string("op_43082_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43082_end_0 = const()[name = string("op_43082_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_43082_end_mask_0 = const()[name = string("op_43082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43082_cast_fp16 = slice_by_index(begin = var_43082_begin_0, end = var_43082_end_0, end_mask = var_43082_end_mask_0, x = query_57_cast_fp16)[name = string("op_43082_cast_fp16")]; + tensor var_43086_begin_0 = const()[name = string("op_43086_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43086_end_0 = const()[name = string("op_43086_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_43086_end_mask_0 = const()[name = string("op_43086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43086_cast_fp16 = slice_by_index(begin = var_43086_begin_0, end = var_43086_end_0, end_mask = var_43086_end_mask_0, x = query_57_cast_fp16)[name = string("op_43086_cast_fp16")]; + tensor var_43090_begin_0 = const()[name = string("op_43090_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43090_end_0 = const()[name = string("op_43090_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_43090_end_mask_0 = const()[name = string("op_43090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43090_cast_fp16 = slice_by_index(begin = var_43090_begin_0, end = var_43090_end_0, end_mask = var_43090_end_mask_0, x = query_57_cast_fp16)[name = string("op_43090_cast_fp16")]; + tensor var_43099_begin_0 = const()[name = string("op_43099_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43099_end_0 = const()[name = string("op_43099_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43099_end_mask_0 = const()[name = string("op_43099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43099_cast_fp16 = slice_by_index(begin = var_43099_begin_0, end = var_43099_end_0, end_mask = var_43099_end_mask_0, x = var_43014_cast_fp16)[name = string("op_43099_cast_fp16")]; + tensor var_43106_begin_0 = const()[name = string("op_43106_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43106_end_0 = const()[name = string("op_43106_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43106_end_mask_0 = const()[name = string("op_43106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43106_cast_fp16 = slice_by_index(begin = var_43106_begin_0, end = var_43106_end_0, end_mask = var_43106_end_mask_0, x = var_43014_cast_fp16)[name = string("op_43106_cast_fp16")]; + tensor var_43113_begin_0 = const()[name = string("op_43113_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43113_end_0 = const()[name = string("op_43113_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43113_end_mask_0 = const()[name = string("op_43113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43113_cast_fp16 = slice_by_index(begin = var_43113_begin_0, end = var_43113_end_0, end_mask = var_43113_end_mask_0, x = var_43014_cast_fp16)[name = string("op_43113_cast_fp16")]; + tensor var_43120_begin_0 = const()[name = string("op_43120_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43120_end_0 = const()[name = string("op_43120_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43120_end_mask_0 = const()[name = string("op_43120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43120_cast_fp16 = slice_by_index(begin = var_43120_begin_0, end = var_43120_end_0, end_mask = var_43120_end_mask_0, x = var_43014_cast_fp16)[name = string("op_43120_cast_fp16")]; + tensor var_43127_begin_0 = const()[name = string("op_43127_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43127_end_0 = const()[name = string("op_43127_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43127_end_mask_0 = const()[name = string("op_43127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43127_cast_fp16 = slice_by_index(begin = var_43127_begin_0, end = var_43127_end_0, end_mask = var_43127_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43127_cast_fp16")]; + tensor var_43134_begin_0 = const()[name = string("op_43134_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43134_end_0 = const()[name = string("op_43134_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43134_end_mask_0 = const()[name = string("op_43134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43134_cast_fp16 = slice_by_index(begin = var_43134_begin_0, end = var_43134_end_0, end_mask = var_43134_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43134_cast_fp16")]; + tensor var_43141_begin_0 = const()[name = string("op_43141_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43141_end_0 = const()[name = string("op_43141_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43141_end_mask_0 = const()[name = string("op_43141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43141_cast_fp16 = slice_by_index(begin = var_43141_begin_0, end = var_43141_end_0, end_mask = var_43141_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43141_cast_fp16")]; + tensor var_43148_begin_0 = const()[name = string("op_43148_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43148_end_0 = const()[name = string("op_43148_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43148_end_mask_0 = const()[name = string("op_43148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43148_cast_fp16 = slice_by_index(begin = var_43148_begin_0, end = var_43148_end_0, end_mask = var_43148_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43148_cast_fp16")]; + tensor var_43155_begin_0 = const()[name = string("op_43155_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43155_end_0 = const()[name = string("op_43155_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43155_end_mask_0 = const()[name = string("op_43155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43155_cast_fp16 = slice_by_index(begin = var_43155_begin_0, end = var_43155_end_0, end_mask = var_43155_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43155_cast_fp16")]; + tensor var_43162_begin_0 = const()[name = string("op_43162_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43162_end_0 = const()[name = string("op_43162_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43162_end_mask_0 = const()[name = string("op_43162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43162_cast_fp16 = slice_by_index(begin = var_43162_begin_0, end = var_43162_end_0, end_mask = var_43162_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43162_cast_fp16")]; + tensor var_43169_begin_0 = const()[name = string("op_43169_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43169_end_0 = const()[name = string("op_43169_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43169_end_mask_0 = const()[name = string("op_43169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43169_cast_fp16 = slice_by_index(begin = var_43169_begin_0, end = var_43169_end_0, end_mask = var_43169_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43169_cast_fp16")]; + tensor var_43176_begin_0 = const()[name = string("op_43176_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43176_end_0 = const()[name = string("op_43176_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43176_end_mask_0 = const()[name = string("op_43176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43176_cast_fp16 = slice_by_index(begin = var_43176_begin_0, end = var_43176_end_0, end_mask = var_43176_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43176_cast_fp16")]; + tensor var_43183_begin_0 = const()[name = string("op_43183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43183_end_0 = const()[name = string("op_43183_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43183_end_mask_0 = const()[name = string("op_43183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43183_cast_fp16 = slice_by_index(begin = var_43183_begin_0, end = var_43183_end_0, end_mask = var_43183_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43183_cast_fp16")]; + tensor var_43190_begin_0 = const()[name = string("op_43190_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43190_end_0 = const()[name = string("op_43190_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43190_end_mask_0 = const()[name = string("op_43190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43190_cast_fp16 = slice_by_index(begin = var_43190_begin_0, end = var_43190_end_0, end_mask = var_43190_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43190_cast_fp16")]; + tensor var_43197_begin_0 = const()[name = string("op_43197_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43197_end_0 = const()[name = string("op_43197_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43197_end_mask_0 = const()[name = string("op_43197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43197_cast_fp16 = slice_by_index(begin = var_43197_begin_0, end = var_43197_end_0, end_mask = var_43197_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43197_cast_fp16")]; + tensor var_43204_begin_0 = const()[name = string("op_43204_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43204_end_0 = const()[name = string("op_43204_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43204_end_mask_0 = const()[name = string("op_43204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43204_cast_fp16 = slice_by_index(begin = var_43204_begin_0, end = var_43204_end_0, end_mask = var_43204_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43204_cast_fp16")]; + tensor var_43211_begin_0 = const()[name = string("op_43211_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43211_end_0 = const()[name = string("op_43211_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43211_end_mask_0 = const()[name = string("op_43211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43211_cast_fp16 = slice_by_index(begin = var_43211_begin_0, end = var_43211_end_0, end_mask = var_43211_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43211_cast_fp16")]; + tensor var_43218_begin_0 = const()[name = string("op_43218_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43218_end_0 = const()[name = string("op_43218_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43218_end_mask_0 = const()[name = string("op_43218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43218_cast_fp16 = slice_by_index(begin = var_43218_begin_0, end = var_43218_end_0, end_mask = var_43218_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43218_cast_fp16")]; + tensor var_43225_begin_0 = const()[name = string("op_43225_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43225_end_0 = const()[name = string("op_43225_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43225_end_mask_0 = const()[name = string("op_43225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43225_cast_fp16 = slice_by_index(begin = var_43225_begin_0, end = var_43225_end_0, end_mask = var_43225_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43225_cast_fp16")]; + tensor var_43232_begin_0 = const()[name = string("op_43232_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43232_end_0 = const()[name = string("op_43232_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43232_end_mask_0 = const()[name = string("op_43232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43232_cast_fp16 = slice_by_index(begin = var_43232_begin_0, end = var_43232_end_0, end_mask = var_43232_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43232_cast_fp16")]; + tensor var_43239_begin_0 = const()[name = string("op_43239_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43239_end_0 = const()[name = string("op_43239_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43239_end_mask_0 = const()[name = string("op_43239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43239_cast_fp16 = slice_by_index(begin = var_43239_begin_0, end = var_43239_end_0, end_mask = var_43239_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43239_cast_fp16")]; + tensor var_43246_begin_0 = const()[name = string("op_43246_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43246_end_0 = const()[name = string("op_43246_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43246_end_mask_0 = const()[name = string("op_43246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43246_cast_fp16 = slice_by_index(begin = var_43246_begin_0, end = var_43246_end_0, end_mask = var_43246_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43246_cast_fp16")]; + tensor var_43253_begin_0 = const()[name = string("op_43253_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43253_end_0 = const()[name = string("op_43253_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43253_end_mask_0 = const()[name = string("op_43253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43253_cast_fp16 = slice_by_index(begin = var_43253_begin_0, end = var_43253_end_0, end_mask = var_43253_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43253_cast_fp16")]; + tensor var_43260_begin_0 = const()[name = string("op_43260_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43260_end_0 = const()[name = string("op_43260_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43260_end_mask_0 = const()[name = string("op_43260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43260_cast_fp16 = slice_by_index(begin = var_43260_begin_0, end = var_43260_end_0, end_mask = var_43260_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43260_cast_fp16")]; + tensor var_43267_begin_0 = const()[name = string("op_43267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43267_end_0 = const()[name = string("op_43267_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43267_end_mask_0 = const()[name = string("op_43267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43267_cast_fp16 = slice_by_index(begin = var_43267_begin_0, end = var_43267_end_0, end_mask = var_43267_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43267_cast_fp16")]; + tensor var_43274_begin_0 = const()[name = string("op_43274_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43274_end_0 = const()[name = string("op_43274_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43274_end_mask_0 = const()[name = string("op_43274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43274_cast_fp16 = slice_by_index(begin = var_43274_begin_0, end = var_43274_end_0, end_mask = var_43274_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43274_cast_fp16")]; + tensor var_43281_begin_0 = const()[name = string("op_43281_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43281_end_0 = const()[name = string("op_43281_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43281_end_mask_0 = const()[name = string("op_43281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43281_cast_fp16 = slice_by_index(begin = var_43281_begin_0, end = var_43281_end_0, end_mask = var_43281_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43281_cast_fp16")]; + tensor var_43288_begin_0 = const()[name = string("op_43288_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43288_end_0 = const()[name = string("op_43288_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43288_end_mask_0 = const()[name = string("op_43288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43288_cast_fp16 = slice_by_index(begin = var_43288_begin_0, end = var_43288_end_0, end_mask = var_43288_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43288_cast_fp16")]; + tensor var_43295_begin_0 = const()[name = string("op_43295_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43295_end_0 = const()[name = string("op_43295_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43295_end_mask_0 = const()[name = string("op_43295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43295_cast_fp16 = slice_by_index(begin = var_43295_begin_0, end = var_43295_end_0, end_mask = var_43295_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43295_cast_fp16")]; + tensor var_43302_begin_0 = const()[name = string("op_43302_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43302_end_0 = const()[name = string("op_43302_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43302_end_mask_0 = const()[name = string("op_43302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43302_cast_fp16 = slice_by_index(begin = var_43302_begin_0, end = var_43302_end_0, end_mask = var_43302_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43302_cast_fp16")]; + tensor var_43309_begin_0 = const()[name = string("op_43309_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43309_end_0 = const()[name = string("op_43309_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43309_end_mask_0 = const()[name = string("op_43309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43309_cast_fp16 = slice_by_index(begin = var_43309_begin_0, end = var_43309_end_0, end_mask = var_43309_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43309_cast_fp16")]; + tensor var_43316_begin_0 = const()[name = string("op_43316_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43316_end_0 = const()[name = string("op_43316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43316_end_mask_0 = const()[name = string("op_43316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43316_cast_fp16 = slice_by_index(begin = var_43316_begin_0, end = var_43316_end_0, end_mask = var_43316_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43316_cast_fp16")]; + tensor var_43323_begin_0 = const()[name = string("op_43323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43323_end_0 = const()[name = string("op_43323_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43323_end_mask_0 = const()[name = string("op_43323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43323_cast_fp16 = slice_by_index(begin = var_43323_begin_0, end = var_43323_end_0, end_mask = var_43323_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43323_cast_fp16")]; + tensor var_43330_begin_0 = const()[name = string("op_43330_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43330_end_0 = const()[name = string("op_43330_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43330_end_mask_0 = const()[name = string("op_43330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43330_cast_fp16 = slice_by_index(begin = var_43330_begin_0, end = var_43330_end_0, end_mask = var_43330_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43330_cast_fp16")]; + tensor var_43337_begin_0 = const()[name = string("op_43337_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43337_end_0 = const()[name = string("op_43337_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43337_end_mask_0 = const()[name = string("op_43337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43337_cast_fp16 = slice_by_index(begin = var_43337_begin_0, end = var_43337_end_0, end_mask = var_43337_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43337_cast_fp16")]; + tensor var_43344_begin_0 = const()[name = string("op_43344_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43344_end_0 = const()[name = string("op_43344_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43344_end_mask_0 = const()[name = string("op_43344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43344_cast_fp16 = slice_by_index(begin = var_43344_begin_0, end = var_43344_end_0, end_mask = var_43344_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43344_cast_fp16")]; + tensor var_43351_begin_0 = const()[name = string("op_43351_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43351_end_0 = const()[name = string("op_43351_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43351_end_mask_0 = const()[name = string("op_43351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43351_cast_fp16 = slice_by_index(begin = var_43351_begin_0, end = var_43351_end_0, end_mask = var_43351_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43351_cast_fp16")]; + tensor var_43358_begin_0 = const()[name = string("op_43358_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43358_end_0 = const()[name = string("op_43358_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43358_end_mask_0 = const()[name = string("op_43358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43358_cast_fp16 = slice_by_index(begin = var_43358_begin_0, end = var_43358_end_0, end_mask = var_43358_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43358_cast_fp16")]; + tensor var_43365_begin_0 = const()[name = string("op_43365_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43365_end_0 = const()[name = string("op_43365_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43365_end_mask_0 = const()[name = string("op_43365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43365_cast_fp16 = slice_by_index(begin = var_43365_begin_0, end = var_43365_end_0, end_mask = var_43365_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43365_cast_fp16")]; + tensor var_43372_begin_0 = const()[name = string("op_43372_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43372_end_0 = const()[name = string("op_43372_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43372_end_mask_0 = const()[name = string("op_43372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43372_cast_fp16 = slice_by_index(begin = var_43372_begin_0, end = var_43372_end_0, end_mask = var_43372_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43372_cast_fp16")]; + tensor var_43379_begin_0 = const()[name = string("op_43379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43379_end_0 = const()[name = string("op_43379_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43379_end_mask_0 = const()[name = string("op_43379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43379_cast_fp16 = slice_by_index(begin = var_43379_begin_0, end = var_43379_end_0, end_mask = var_43379_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43379_cast_fp16")]; + tensor var_43386_begin_0 = const()[name = string("op_43386_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43386_end_0 = const()[name = string("op_43386_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43386_end_mask_0 = const()[name = string("op_43386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43386_cast_fp16 = slice_by_index(begin = var_43386_begin_0, end = var_43386_end_0, end_mask = var_43386_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43386_cast_fp16")]; + tensor var_43393_begin_0 = const()[name = string("op_43393_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43393_end_0 = const()[name = string("op_43393_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43393_end_mask_0 = const()[name = string("op_43393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43393_cast_fp16 = slice_by_index(begin = var_43393_begin_0, end = var_43393_end_0, end_mask = var_43393_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43393_cast_fp16")]; + tensor var_43400_begin_0 = const()[name = string("op_43400_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43400_end_0 = const()[name = string("op_43400_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43400_end_mask_0 = const()[name = string("op_43400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43400_cast_fp16 = slice_by_index(begin = var_43400_begin_0, end = var_43400_end_0, end_mask = var_43400_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43400_cast_fp16")]; + tensor var_43407_begin_0 = const()[name = string("op_43407_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43407_end_0 = const()[name = string("op_43407_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43407_end_mask_0 = const()[name = string("op_43407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43407_cast_fp16 = slice_by_index(begin = var_43407_begin_0, end = var_43407_end_0, end_mask = var_43407_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43407_cast_fp16")]; + tensor var_43414_begin_0 = const()[name = string("op_43414_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43414_end_0 = const()[name = string("op_43414_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43414_end_mask_0 = const()[name = string("op_43414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43414_cast_fp16 = slice_by_index(begin = var_43414_begin_0, end = var_43414_end_0, end_mask = var_43414_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43414_cast_fp16")]; + tensor var_43421_begin_0 = const()[name = string("op_43421_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43421_end_0 = const()[name = string("op_43421_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43421_end_mask_0 = const()[name = string("op_43421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43421_cast_fp16 = slice_by_index(begin = var_43421_begin_0, end = var_43421_end_0, end_mask = var_43421_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43421_cast_fp16")]; + tensor var_43428_begin_0 = const()[name = string("op_43428_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43428_end_0 = const()[name = string("op_43428_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43428_end_mask_0 = const()[name = string("op_43428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43428_cast_fp16 = slice_by_index(begin = var_43428_begin_0, end = var_43428_end_0, end_mask = var_43428_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43428_cast_fp16")]; + tensor var_43435_begin_0 = const()[name = string("op_43435_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43435_end_0 = const()[name = string("op_43435_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43435_end_mask_0 = const()[name = string("op_43435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43435_cast_fp16 = slice_by_index(begin = var_43435_begin_0, end = var_43435_end_0, end_mask = var_43435_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43435_cast_fp16")]; + tensor var_43442_begin_0 = const()[name = string("op_43442_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43442_end_0 = const()[name = string("op_43442_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43442_end_mask_0 = const()[name = string("op_43442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43442_cast_fp16 = slice_by_index(begin = var_43442_begin_0, end = var_43442_end_0, end_mask = var_43442_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43442_cast_fp16")]; + tensor var_43449_begin_0 = const()[name = string("op_43449_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43449_end_0 = const()[name = string("op_43449_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43449_end_mask_0 = const()[name = string("op_43449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43449_cast_fp16 = slice_by_index(begin = var_43449_begin_0, end = var_43449_end_0, end_mask = var_43449_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43449_cast_fp16")]; + tensor var_43456_begin_0 = const()[name = string("op_43456_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43456_end_0 = const()[name = string("op_43456_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43456_end_mask_0 = const()[name = string("op_43456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43456_cast_fp16 = slice_by_index(begin = var_43456_begin_0, end = var_43456_end_0, end_mask = var_43456_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43456_cast_fp16")]; + tensor var_43463_begin_0 = const()[name = string("op_43463_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43463_end_0 = const()[name = string("op_43463_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43463_end_mask_0 = const()[name = string("op_43463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43463_cast_fp16 = slice_by_index(begin = var_43463_begin_0, end = var_43463_end_0, end_mask = var_43463_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43463_cast_fp16")]; + tensor var_43470_begin_0 = const()[name = string("op_43470_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43470_end_0 = const()[name = string("op_43470_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43470_end_mask_0 = const()[name = string("op_43470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43470_cast_fp16 = slice_by_index(begin = var_43470_begin_0, end = var_43470_end_0, end_mask = var_43470_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43470_cast_fp16")]; + tensor var_43477_begin_0 = const()[name = string("op_43477_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43477_end_0 = const()[name = string("op_43477_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43477_end_mask_0 = const()[name = string("op_43477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43477_cast_fp16 = slice_by_index(begin = var_43477_begin_0, end = var_43477_end_0, end_mask = var_43477_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43477_cast_fp16")]; + tensor var_43484_begin_0 = const()[name = string("op_43484_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43484_end_0 = const()[name = string("op_43484_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43484_end_mask_0 = const()[name = string("op_43484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43484_cast_fp16 = slice_by_index(begin = var_43484_begin_0, end = var_43484_end_0, end_mask = var_43484_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43484_cast_fp16")]; + tensor var_43491_begin_0 = const()[name = string("op_43491_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43491_end_0 = const()[name = string("op_43491_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43491_end_mask_0 = const()[name = string("op_43491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43491_cast_fp16 = slice_by_index(begin = var_43491_begin_0, end = var_43491_end_0, end_mask = var_43491_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43491_cast_fp16")]; + tensor var_43498_begin_0 = const()[name = string("op_43498_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43498_end_0 = const()[name = string("op_43498_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43498_end_mask_0 = const()[name = string("op_43498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43498_cast_fp16 = slice_by_index(begin = var_43498_begin_0, end = var_43498_end_0, end_mask = var_43498_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43498_cast_fp16")]; + tensor var_43505_begin_0 = const()[name = string("op_43505_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43505_end_0 = const()[name = string("op_43505_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43505_end_mask_0 = const()[name = string("op_43505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43505_cast_fp16 = slice_by_index(begin = var_43505_begin_0, end = var_43505_end_0, end_mask = var_43505_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43505_cast_fp16")]; + tensor var_43512_begin_0 = const()[name = string("op_43512_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43512_end_0 = const()[name = string("op_43512_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43512_end_mask_0 = const()[name = string("op_43512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43512_cast_fp16 = slice_by_index(begin = var_43512_begin_0, end = var_43512_end_0, end_mask = var_43512_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43512_cast_fp16")]; + tensor var_43519_begin_0 = const()[name = string("op_43519_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43519_end_0 = const()[name = string("op_43519_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43519_end_mask_0 = const()[name = string("op_43519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43519_cast_fp16 = slice_by_index(begin = var_43519_begin_0, end = var_43519_end_0, end_mask = var_43519_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43519_cast_fp16")]; + tensor var_43526_begin_0 = const()[name = string("op_43526_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43526_end_0 = const()[name = string("op_43526_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43526_end_mask_0 = const()[name = string("op_43526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43526_cast_fp16 = slice_by_index(begin = var_43526_begin_0, end = var_43526_end_0, end_mask = var_43526_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43526_cast_fp16")]; + tensor var_43533_begin_0 = const()[name = string("op_43533_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43533_end_0 = const()[name = string("op_43533_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43533_end_mask_0 = const()[name = string("op_43533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43533_cast_fp16 = slice_by_index(begin = var_43533_begin_0, end = var_43533_end_0, end_mask = var_43533_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43533_cast_fp16")]; + tensor var_43540_begin_0 = const()[name = string("op_43540_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43540_end_0 = const()[name = string("op_43540_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43540_end_mask_0 = const()[name = string("op_43540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43540_cast_fp16 = slice_by_index(begin = var_43540_begin_0, end = var_43540_end_0, end_mask = var_43540_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43540_cast_fp16")]; + tensor var_43547_begin_0 = const()[name = string("op_43547_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43547_end_0 = const()[name = string("op_43547_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43547_end_mask_0 = const()[name = string("op_43547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43547_cast_fp16 = slice_by_index(begin = var_43547_begin_0, end = var_43547_end_0, end_mask = var_43547_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43547_cast_fp16")]; + tensor var_43554_begin_0 = const()[name = string("op_43554_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43554_end_0 = const()[name = string("op_43554_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43554_end_mask_0 = const()[name = string("op_43554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43554_cast_fp16 = slice_by_index(begin = var_43554_begin_0, end = var_43554_end_0, end_mask = var_43554_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43554_cast_fp16")]; + tensor var_43561_begin_0 = const()[name = string("op_43561_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43561_end_0 = const()[name = string("op_43561_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43561_end_mask_0 = const()[name = string("op_43561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43561_cast_fp16 = slice_by_index(begin = var_43561_begin_0, end = var_43561_end_0, end_mask = var_43561_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43561_cast_fp16")]; + tensor var_43568_begin_0 = const()[name = string("op_43568_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43568_end_0 = const()[name = string("op_43568_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43568_end_mask_0 = const()[name = string("op_43568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43568_cast_fp16 = slice_by_index(begin = var_43568_begin_0, end = var_43568_end_0, end_mask = var_43568_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43568_cast_fp16")]; + tensor var_43575_begin_0 = const()[name = string("op_43575_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43575_end_0 = const()[name = string("op_43575_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43575_end_mask_0 = const()[name = string("op_43575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43575_cast_fp16 = slice_by_index(begin = var_43575_begin_0, end = var_43575_end_0, end_mask = var_43575_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43575_cast_fp16")]; + tensor var_43582_begin_0 = const()[name = string("op_43582_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43582_end_0 = const()[name = string("op_43582_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43582_end_mask_0 = const()[name = string("op_43582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43582_cast_fp16 = slice_by_index(begin = var_43582_begin_0, end = var_43582_end_0, end_mask = var_43582_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43582_cast_fp16")]; + tensor var_43589_begin_0 = const()[name = string("op_43589_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43589_end_0 = const()[name = string("op_43589_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43589_end_mask_0 = const()[name = string("op_43589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43589_cast_fp16 = slice_by_index(begin = var_43589_begin_0, end = var_43589_end_0, end_mask = var_43589_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43589_cast_fp16")]; + tensor var_43596_begin_0 = const()[name = string("op_43596_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43596_end_0 = const()[name = string("op_43596_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43596_end_mask_0 = const()[name = string("op_43596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43596_cast_fp16 = slice_by_index(begin = var_43596_begin_0, end = var_43596_end_0, end_mask = var_43596_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43596_cast_fp16")]; + tensor var_43603_begin_0 = const()[name = string("op_43603_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43603_end_0 = const()[name = string("op_43603_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43603_end_mask_0 = const()[name = string("op_43603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43603_cast_fp16 = slice_by_index(begin = var_43603_begin_0, end = var_43603_end_0, end_mask = var_43603_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43603_cast_fp16")]; + tensor var_43610_begin_0 = const()[name = string("op_43610_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43610_end_0 = const()[name = string("op_43610_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43610_end_mask_0 = const()[name = string("op_43610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43610_cast_fp16 = slice_by_index(begin = var_43610_begin_0, end = var_43610_end_0, end_mask = var_43610_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43610_cast_fp16")]; + tensor var_43617_begin_0 = const()[name = string("op_43617_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43617_end_0 = const()[name = string("op_43617_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43617_end_mask_0 = const()[name = string("op_43617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43617_cast_fp16 = slice_by_index(begin = var_43617_begin_0, end = var_43617_end_0, end_mask = var_43617_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43617_cast_fp16")]; + tensor var_43624_begin_0 = const()[name = string("op_43624_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43624_end_0 = const()[name = string("op_43624_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43624_end_mask_0 = const()[name = string("op_43624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43624_cast_fp16 = slice_by_index(begin = var_43624_begin_0, end = var_43624_end_0, end_mask = var_43624_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43624_cast_fp16")]; + tensor var_43631_begin_0 = const()[name = string("op_43631_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43631_end_0 = const()[name = string("op_43631_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43631_end_mask_0 = const()[name = string("op_43631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43631_cast_fp16 = slice_by_index(begin = var_43631_begin_0, end = var_43631_end_0, end_mask = var_43631_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43631_cast_fp16")]; + tensor var_43638_begin_0 = const()[name = string("op_43638_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43638_end_0 = const()[name = string("op_43638_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43638_end_mask_0 = const()[name = string("op_43638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43638_cast_fp16 = slice_by_index(begin = var_43638_begin_0, end = var_43638_end_0, end_mask = var_43638_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43638_cast_fp16")]; + tensor var_43645_begin_0 = const()[name = string("op_43645_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43645_end_0 = const()[name = string("op_43645_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43645_end_mask_0 = const()[name = string("op_43645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43645_cast_fp16 = slice_by_index(begin = var_43645_begin_0, end = var_43645_end_0, end_mask = var_43645_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43645_cast_fp16")]; + tensor var_43652_begin_0 = const()[name = string("op_43652_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43652_end_0 = const()[name = string("op_43652_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43652_end_mask_0 = const()[name = string("op_43652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43652_cast_fp16 = slice_by_index(begin = var_43652_begin_0, end = var_43652_end_0, end_mask = var_43652_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43652_cast_fp16")]; + tensor k_57_perm_0 = const()[name = string("k_57_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_43657_begin_0 = const()[name = string("op_43657_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43657_end_0 = const()[name = string("op_43657_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_43657_end_mask_0 = const()[name = string("op_43657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_57_cast_fp16 = transpose(perm = k_57_perm_0, x = key_57_cast_fp16)[name = string("transpose_3")]; + tensor var_43657_cast_fp16 = slice_by_index(begin = var_43657_begin_0, end = var_43657_end_0, end_mask = var_43657_end_mask_0, x = k_57_cast_fp16)[name = string("op_43657_cast_fp16")]; + tensor var_43661_begin_0 = const()[name = string("op_43661_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_43661_end_0 = const()[name = string("op_43661_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_43661_end_mask_0 = const()[name = string("op_43661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43661_cast_fp16 = slice_by_index(begin = var_43661_begin_0, end = var_43661_end_0, end_mask = var_43661_end_mask_0, x = k_57_cast_fp16)[name = string("op_43661_cast_fp16")]; + tensor var_43665_begin_0 = const()[name = string("op_43665_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_43665_end_0 = const()[name = string("op_43665_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_43665_end_mask_0 = const()[name = string("op_43665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43665_cast_fp16 = slice_by_index(begin = var_43665_begin_0, end = var_43665_end_0, end_mask = var_43665_end_mask_0, x = k_57_cast_fp16)[name = string("op_43665_cast_fp16")]; + tensor var_43669_begin_0 = const()[name = string("op_43669_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_43669_end_0 = const()[name = string("op_43669_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_43669_end_mask_0 = const()[name = string("op_43669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43669_cast_fp16 = slice_by_index(begin = var_43669_begin_0, end = var_43669_end_0, end_mask = var_43669_end_mask_0, x = k_57_cast_fp16)[name = string("op_43669_cast_fp16")]; + tensor var_43673_begin_0 = const()[name = string("op_43673_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_43673_end_0 = const()[name = string("op_43673_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_43673_end_mask_0 = const()[name = string("op_43673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43673_cast_fp16 = slice_by_index(begin = var_43673_begin_0, end = var_43673_end_0, end_mask = var_43673_end_mask_0, x = k_57_cast_fp16)[name = string("op_43673_cast_fp16")]; + tensor var_43677_begin_0 = const()[name = string("op_43677_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_43677_end_0 = const()[name = string("op_43677_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_43677_end_mask_0 = const()[name = string("op_43677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43677_cast_fp16 = slice_by_index(begin = var_43677_begin_0, end = var_43677_end_0, end_mask = var_43677_end_mask_0, x = k_57_cast_fp16)[name = string("op_43677_cast_fp16")]; + tensor var_43681_begin_0 = const()[name = string("op_43681_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_43681_end_0 = const()[name = string("op_43681_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_43681_end_mask_0 = const()[name = string("op_43681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43681_cast_fp16 = slice_by_index(begin = var_43681_begin_0, end = var_43681_end_0, end_mask = var_43681_end_mask_0, x = k_57_cast_fp16)[name = string("op_43681_cast_fp16")]; + tensor var_43685_begin_0 = const()[name = string("op_43685_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_43685_end_0 = const()[name = string("op_43685_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_43685_end_mask_0 = const()[name = string("op_43685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43685_cast_fp16 = slice_by_index(begin = var_43685_begin_0, end = var_43685_end_0, end_mask = var_43685_end_mask_0, x = k_57_cast_fp16)[name = string("op_43685_cast_fp16")]; + tensor var_43689_begin_0 = const()[name = string("op_43689_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_43689_end_0 = const()[name = string("op_43689_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_43689_end_mask_0 = const()[name = string("op_43689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43689_cast_fp16 = slice_by_index(begin = var_43689_begin_0, end = var_43689_end_0, end_mask = var_43689_end_mask_0, x = k_57_cast_fp16)[name = string("op_43689_cast_fp16")]; + tensor var_43693_begin_0 = const()[name = string("op_43693_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_43693_end_0 = const()[name = string("op_43693_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_43693_end_mask_0 = const()[name = string("op_43693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43693_cast_fp16 = slice_by_index(begin = var_43693_begin_0, end = var_43693_end_0, end_mask = var_43693_end_mask_0, x = k_57_cast_fp16)[name = string("op_43693_cast_fp16")]; + tensor var_43697_begin_0 = const()[name = string("op_43697_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_43697_end_0 = const()[name = string("op_43697_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_43697_end_mask_0 = const()[name = string("op_43697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43697_cast_fp16 = slice_by_index(begin = var_43697_begin_0, end = var_43697_end_0, end_mask = var_43697_end_mask_0, x = k_57_cast_fp16)[name = string("op_43697_cast_fp16")]; + tensor var_43701_begin_0 = const()[name = string("op_43701_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_43701_end_0 = const()[name = string("op_43701_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_43701_end_mask_0 = const()[name = string("op_43701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43701_cast_fp16 = slice_by_index(begin = var_43701_begin_0, end = var_43701_end_0, end_mask = var_43701_end_mask_0, x = k_57_cast_fp16)[name = string("op_43701_cast_fp16")]; + tensor var_43705_begin_0 = const()[name = string("op_43705_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_43705_end_0 = const()[name = string("op_43705_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_43705_end_mask_0 = const()[name = string("op_43705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43705_cast_fp16 = slice_by_index(begin = var_43705_begin_0, end = var_43705_end_0, end_mask = var_43705_end_mask_0, x = k_57_cast_fp16)[name = string("op_43705_cast_fp16")]; + tensor var_43709_begin_0 = const()[name = string("op_43709_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_43709_end_0 = const()[name = string("op_43709_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_43709_end_mask_0 = const()[name = string("op_43709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43709_cast_fp16 = slice_by_index(begin = var_43709_begin_0, end = var_43709_end_0, end_mask = var_43709_end_mask_0, x = k_57_cast_fp16)[name = string("op_43709_cast_fp16")]; + tensor var_43713_begin_0 = const()[name = string("op_43713_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_43713_end_0 = const()[name = string("op_43713_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_43713_end_mask_0 = const()[name = string("op_43713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43713_cast_fp16 = slice_by_index(begin = var_43713_begin_0, end = var_43713_end_0, end_mask = var_43713_end_mask_0, x = k_57_cast_fp16)[name = string("op_43713_cast_fp16")]; + tensor var_43717_begin_0 = const()[name = string("op_43717_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_43717_end_0 = const()[name = string("op_43717_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_43717_end_mask_0 = const()[name = string("op_43717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43717_cast_fp16 = slice_by_index(begin = var_43717_begin_0, end = var_43717_end_0, end_mask = var_43717_end_mask_0, x = k_57_cast_fp16)[name = string("op_43717_cast_fp16")]; + tensor var_43721_begin_0 = const()[name = string("op_43721_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_43721_end_0 = const()[name = string("op_43721_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_43721_end_mask_0 = const()[name = string("op_43721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43721_cast_fp16 = slice_by_index(begin = var_43721_begin_0, end = var_43721_end_0, end_mask = var_43721_end_mask_0, x = k_57_cast_fp16)[name = string("op_43721_cast_fp16")]; + tensor var_43725_begin_0 = const()[name = string("op_43725_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_43725_end_0 = const()[name = string("op_43725_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_43725_end_mask_0 = const()[name = string("op_43725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43725_cast_fp16 = slice_by_index(begin = var_43725_begin_0, end = var_43725_end_0, end_mask = var_43725_end_mask_0, x = k_57_cast_fp16)[name = string("op_43725_cast_fp16")]; + tensor var_43729_begin_0 = const()[name = string("op_43729_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_43729_end_0 = const()[name = string("op_43729_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_43729_end_mask_0 = const()[name = string("op_43729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43729_cast_fp16 = slice_by_index(begin = var_43729_begin_0, end = var_43729_end_0, end_mask = var_43729_end_mask_0, x = k_57_cast_fp16)[name = string("op_43729_cast_fp16")]; + tensor var_43733_begin_0 = const()[name = string("op_43733_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_43733_end_0 = const()[name = string("op_43733_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_43733_end_mask_0 = const()[name = string("op_43733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43733_cast_fp16 = slice_by_index(begin = var_43733_begin_0, end = var_43733_end_0, end_mask = var_43733_end_mask_0, x = k_57_cast_fp16)[name = string("op_43733_cast_fp16")]; + tensor var_43735_begin_0 = const()[name = string("op_43735_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43735_end_0 = const()[name = string("op_43735_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43735_end_mask_0 = const()[name = string("op_43735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43735_cast_fp16 = slice_by_index(begin = var_43735_begin_0, end = var_43735_end_0, end_mask = var_43735_end_mask_0, x = value_57_cast_fp16)[name = string("op_43735_cast_fp16")]; + tensor var_43739_begin_0 = const()[name = string("op_43739_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43739_end_0 = const()[name = string("op_43739_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_43739_end_mask_0 = const()[name = string("op_43739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43739_cast_fp16 = slice_by_index(begin = var_43739_begin_0, end = var_43739_end_0, end_mask = var_43739_end_mask_0, x = value_57_cast_fp16)[name = string("op_43739_cast_fp16")]; + tensor var_43743_begin_0 = const()[name = string("op_43743_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43743_end_0 = const()[name = string("op_43743_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_43743_end_mask_0 = const()[name = string("op_43743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43743_cast_fp16 = slice_by_index(begin = var_43743_begin_0, end = var_43743_end_0, end_mask = var_43743_end_mask_0, x = value_57_cast_fp16)[name = string("op_43743_cast_fp16")]; + tensor var_43747_begin_0 = const()[name = string("op_43747_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43747_end_0 = const()[name = string("op_43747_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_43747_end_mask_0 = const()[name = string("op_43747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43747_cast_fp16 = slice_by_index(begin = var_43747_begin_0, end = var_43747_end_0, end_mask = var_43747_end_mask_0, x = value_57_cast_fp16)[name = string("op_43747_cast_fp16")]; + tensor var_43751_begin_0 = const()[name = string("op_43751_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43751_end_0 = const()[name = string("op_43751_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_43751_end_mask_0 = const()[name = string("op_43751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43751_cast_fp16 = slice_by_index(begin = var_43751_begin_0, end = var_43751_end_0, end_mask = var_43751_end_mask_0, x = value_57_cast_fp16)[name = string("op_43751_cast_fp16")]; + tensor var_43755_begin_0 = const()[name = string("op_43755_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43755_end_0 = const()[name = string("op_43755_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_43755_end_mask_0 = const()[name = string("op_43755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43755_cast_fp16 = slice_by_index(begin = var_43755_begin_0, end = var_43755_end_0, end_mask = var_43755_end_mask_0, x = value_57_cast_fp16)[name = string("op_43755_cast_fp16")]; + tensor var_43759_begin_0 = const()[name = string("op_43759_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43759_end_0 = const()[name = string("op_43759_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_43759_end_mask_0 = const()[name = string("op_43759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43759_cast_fp16 = slice_by_index(begin = var_43759_begin_0, end = var_43759_end_0, end_mask = var_43759_end_mask_0, x = value_57_cast_fp16)[name = string("op_43759_cast_fp16")]; + tensor var_43763_begin_0 = const()[name = string("op_43763_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43763_end_0 = const()[name = string("op_43763_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_43763_end_mask_0 = const()[name = string("op_43763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43763_cast_fp16 = slice_by_index(begin = var_43763_begin_0, end = var_43763_end_0, end_mask = var_43763_end_mask_0, x = value_57_cast_fp16)[name = string("op_43763_cast_fp16")]; + tensor var_43767_begin_0 = const()[name = string("op_43767_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43767_end_0 = const()[name = string("op_43767_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_43767_end_mask_0 = const()[name = string("op_43767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43767_cast_fp16 = slice_by_index(begin = var_43767_begin_0, end = var_43767_end_0, end_mask = var_43767_end_mask_0, x = value_57_cast_fp16)[name = string("op_43767_cast_fp16")]; + tensor var_43771_begin_0 = const()[name = string("op_43771_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43771_end_0 = const()[name = string("op_43771_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_43771_end_mask_0 = const()[name = string("op_43771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43771_cast_fp16 = slice_by_index(begin = var_43771_begin_0, end = var_43771_end_0, end_mask = var_43771_end_mask_0, x = value_57_cast_fp16)[name = string("op_43771_cast_fp16")]; + tensor var_43775_begin_0 = const()[name = string("op_43775_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43775_end_0 = const()[name = string("op_43775_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_43775_end_mask_0 = const()[name = string("op_43775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43775_cast_fp16 = slice_by_index(begin = var_43775_begin_0, end = var_43775_end_0, end_mask = var_43775_end_mask_0, x = value_57_cast_fp16)[name = string("op_43775_cast_fp16")]; + tensor var_43779_begin_0 = const()[name = string("op_43779_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43779_end_0 = const()[name = string("op_43779_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_43779_end_mask_0 = const()[name = string("op_43779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43779_cast_fp16 = slice_by_index(begin = var_43779_begin_0, end = var_43779_end_0, end_mask = var_43779_end_mask_0, x = value_57_cast_fp16)[name = string("op_43779_cast_fp16")]; + tensor var_43783_begin_0 = const()[name = string("op_43783_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43783_end_0 = const()[name = string("op_43783_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_43783_end_mask_0 = const()[name = string("op_43783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43783_cast_fp16 = slice_by_index(begin = var_43783_begin_0, end = var_43783_end_0, end_mask = var_43783_end_mask_0, x = value_57_cast_fp16)[name = string("op_43783_cast_fp16")]; + tensor var_43787_begin_0 = const()[name = string("op_43787_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43787_end_0 = const()[name = string("op_43787_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_43787_end_mask_0 = const()[name = string("op_43787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43787_cast_fp16 = slice_by_index(begin = var_43787_begin_0, end = var_43787_end_0, end_mask = var_43787_end_mask_0, x = value_57_cast_fp16)[name = string("op_43787_cast_fp16")]; + tensor var_43791_begin_0 = const()[name = string("op_43791_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43791_end_0 = const()[name = string("op_43791_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_43791_end_mask_0 = const()[name = string("op_43791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43791_cast_fp16 = slice_by_index(begin = var_43791_begin_0, end = var_43791_end_0, end_mask = var_43791_end_mask_0, x = value_57_cast_fp16)[name = string("op_43791_cast_fp16")]; + tensor var_43795_begin_0 = const()[name = string("op_43795_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43795_end_0 = const()[name = string("op_43795_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_43795_end_mask_0 = const()[name = string("op_43795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43795_cast_fp16 = slice_by_index(begin = var_43795_begin_0, end = var_43795_end_0, end_mask = var_43795_end_mask_0, x = value_57_cast_fp16)[name = string("op_43795_cast_fp16")]; + tensor var_43799_begin_0 = const()[name = string("op_43799_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43799_end_0 = const()[name = string("op_43799_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_43799_end_mask_0 = const()[name = string("op_43799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43799_cast_fp16 = slice_by_index(begin = var_43799_begin_0, end = var_43799_end_0, end_mask = var_43799_end_mask_0, x = value_57_cast_fp16)[name = string("op_43799_cast_fp16")]; + tensor var_43803_begin_0 = const()[name = string("op_43803_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43803_end_0 = const()[name = string("op_43803_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_43803_end_mask_0 = const()[name = string("op_43803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43803_cast_fp16 = slice_by_index(begin = var_43803_begin_0, end = var_43803_end_0, end_mask = var_43803_end_mask_0, x = value_57_cast_fp16)[name = string("op_43803_cast_fp16")]; + tensor var_43807_begin_0 = const()[name = string("op_43807_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43807_end_0 = const()[name = string("op_43807_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_43807_end_mask_0 = const()[name = string("op_43807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43807_cast_fp16 = slice_by_index(begin = var_43807_begin_0, end = var_43807_end_0, end_mask = var_43807_end_mask_0, x = value_57_cast_fp16)[name = string("op_43807_cast_fp16")]; + tensor var_43811_begin_0 = const()[name = string("op_43811_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43811_end_0 = const()[name = string("op_43811_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_43811_end_mask_0 = const()[name = string("op_43811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43811_cast_fp16 = slice_by_index(begin = var_43811_begin_0, end = var_43811_end_0, end_mask = var_43811_end_mask_0, x = value_57_cast_fp16)[name = string("op_43811_cast_fp16")]; + string _SplitHeadsQ__mh_w_4481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4481_equation_0, values = (var_43657_cast_fp16, var_43099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4481_cast_fp16")]; + string _SplitHeadsQ__mh_w_4483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4483_equation_0, values = (var_43657_cast_fp16, var_43106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4483_cast_fp16")]; + string _SplitHeadsQ__mh_w_4485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4485_equation_0, values = (var_43657_cast_fp16, var_43113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4485_cast_fp16")]; + string _SplitHeadsQ__mh_w_4487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4487_equation_0, values = (var_43657_cast_fp16, var_43120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4487_cast_fp16")]; + string _SplitHeadsQ__mh_w_4489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4489_equation_0, values = (var_43661_cast_fp16, var_43127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4489_cast_fp16")]; + string _SplitHeadsQ__mh_w_4491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4491_equation_0, values = (var_43661_cast_fp16, var_43134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4491_cast_fp16")]; + string _SplitHeadsQ__mh_w_4493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4493_equation_0, values = (var_43661_cast_fp16, var_43141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4493_cast_fp16")]; + string _SplitHeadsQ__mh_w_4495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4495_equation_0, values = (var_43661_cast_fp16, var_43148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4495_cast_fp16")]; + string _SplitHeadsQ__mh_w_4497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4497_equation_0, values = (var_43665_cast_fp16, var_43155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4497_cast_fp16")]; + string _SplitHeadsQ__mh_w_4499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4499_equation_0, values = (var_43665_cast_fp16, var_43162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4499_cast_fp16")]; + string _SplitHeadsQ__mh_w_4501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4501_equation_0, values = (var_43665_cast_fp16, var_43169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4501_cast_fp16")]; + string _SplitHeadsQ__mh_w_4503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4503_equation_0, values = (var_43665_cast_fp16, var_43176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4503_cast_fp16")]; + string _SplitHeadsQ__mh_w_4505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4505_equation_0, values = (var_43669_cast_fp16, var_43183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4505_cast_fp16")]; + string _SplitHeadsQ__mh_w_4507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4507_equation_0, values = (var_43669_cast_fp16, var_43190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4507_cast_fp16")]; + string _SplitHeadsQ__mh_w_4509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4509_equation_0, values = (var_43669_cast_fp16, var_43197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4509_cast_fp16")]; + string _SplitHeadsQ__mh_w_4511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4511_equation_0, values = (var_43669_cast_fp16, var_43204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4511_cast_fp16")]; + string _SplitHeadsQ__mh_w_4513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4513_equation_0, values = (var_43673_cast_fp16, var_43211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4513_cast_fp16")]; + string _SplitHeadsQ__mh_w_4515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4515_equation_0, values = (var_43673_cast_fp16, var_43218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4515_cast_fp16")]; + string _SplitHeadsQ__mh_w_4517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4517_equation_0, values = (var_43673_cast_fp16, var_43225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4517_cast_fp16")]; + string _SplitHeadsQ__mh_w_4519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4519_equation_0, values = (var_43673_cast_fp16, var_43232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4519_cast_fp16")]; + string _SplitHeadsQ__mh_w_4521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4521_equation_0, values = (var_43677_cast_fp16, var_43239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4521_cast_fp16")]; + string _SplitHeadsQ__mh_w_4523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4523_equation_0, values = (var_43677_cast_fp16, var_43246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4523_cast_fp16")]; + string _SplitHeadsQ__mh_w_4525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4525_equation_0, values = (var_43677_cast_fp16, var_43253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4525_cast_fp16")]; + string _SplitHeadsQ__mh_w_4527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4527_equation_0, values = (var_43677_cast_fp16, var_43260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4527_cast_fp16")]; + string _SplitHeadsQ__mh_w_4529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4529_equation_0, values = (var_43681_cast_fp16, var_43267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4529_cast_fp16")]; + string _SplitHeadsQ__mh_w_4531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4531_equation_0, values = (var_43681_cast_fp16, var_43274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4531_cast_fp16")]; + string _SplitHeadsQ__mh_w_4533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4533_equation_0, values = (var_43681_cast_fp16, var_43281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4533_cast_fp16")]; + string _SplitHeadsQ__mh_w_4535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4535_equation_0, values = (var_43681_cast_fp16, var_43288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4535_cast_fp16")]; + string _SplitHeadsQ__mh_w_4537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4537_equation_0, values = (var_43685_cast_fp16, var_43295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4537_cast_fp16")]; + string _SplitHeadsQ__mh_w_4539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4539_equation_0, values = (var_43685_cast_fp16, var_43302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4539_cast_fp16")]; + string _SplitHeadsQ__mh_w_4541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4541_equation_0, values = (var_43685_cast_fp16, var_43309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4541_cast_fp16")]; + string _SplitHeadsQ__mh_w_4543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4543_equation_0, values = (var_43685_cast_fp16, var_43316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4543_cast_fp16")]; + string _SplitHeadsQ__mh_w_4545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4545_equation_0, values = (var_43689_cast_fp16, var_43323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4545_cast_fp16")]; + string _SplitHeadsQ__mh_w_4547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4547_equation_0, values = (var_43689_cast_fp16, var_43330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4547_cast_fp16")]; + string _SplitHeadsQ__mh_w_4549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4549_equation_0, values = (var_43689_cast_fp16, var_43337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4549_cast_fp16")]; + string _SplitHeadsQ__mh_w_4551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4551_equation_0, values = (var_43689_cast_fp16, var_43344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4551_cast_fp16")]; + string _SplitHeadsQ__mh_w_4553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4553_equation_0, values = (var_43693_cast_fp16, var_43351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4553_cast_fp16")]; + string _SplitHeadsQ__mh_w_4555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4555_equation_0, values = (var_43693_cast_fp16, var_43358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4555_cast_fp16")]; + string _SplitHeadsQ__mh_w_4557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4557_equation_0, values = (var_43693_cast_fp16, var_43365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4557_cast_fp16")]; + string _SplitHeadsQ__mh_w_4559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4559_equation_0, values = (var_43693_cast_fp16, var_43372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4559_cast_fp16")]; + string _SplitHeadsQ__mh_w_4561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4561_equation_0, values = (var_43697_cast_fp16, var_43379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4561_cast_fp16")]; + string _SplitHeadsQ__mh_w_4563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4563_equation_0, values = (var_43697_cast_fp16, var_43386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4563_cast_fp16")]; + string _SplitHeadsQ__mh_w_4565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4565_equation_0, values = (var_43697_cast_fp16, var_43393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4565_cast_fp16")]; + string _SplitHeadsQ__mh_w_4567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4567_equation_0, values = (var_43697_cast_fp16, var_43400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4567_cast_fp16")]; + string _SplitHeadsQ__mh_w_4569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4569_equation_0, values = (var_43701_cast_fp16, var_43407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4569_cast_fp16")]; + string _SplitHeadsQ__mh_w_4571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4571_equation_0, values = (var_43701_cast_fp16, var_43414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4571_cast_fp16")]; + string _SplitHeadsQ__mh_w_4573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4573_equation_0, values = (var_43701_cast_fp16, var_43421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4573_cast_fp16")]; + string _SplitHeadsQ__mh_w_4575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4575_equation_0, values = (var_43701_cast_fp16, var_43428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4575_cast_fp16")]; + string _SplitHeadsQ__mh_w_4577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4577_equation_0, values = (var_43705_cast_fp16, var_43435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4577_cast_fp16")]; + string _SplitHeadsQ__mh_w_4579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4579_equation_0, values = (var_43705_cast_fp16, var_43442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4579_cast_fp16")]; + string _SplitHeadsQ__mh_w_4581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4581_equation_0, values = (var_43705_cast_fp16, var_43449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4581_cast_fp16")]; + string _SplitHeadsQ__mh_w_4583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4583_equation_0, values = (var_43705_cast_fp16, var_43456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4583_cast_fp16")]; + string _SplitHeadsQ__mh_w_4585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4585_equation_0, values = (var_43709_cast_fp16, var_43463_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4585_cast_fp16")]; + string _SplitHeadsQ__mh_w_4587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4587_equation_0, values = (var_43709_cast_fp16, var_43470_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4587_cast_fp16")]; + string _SplitHeadsQ__mh_w_4589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4589_equation_0, values = (var_43709_cast_fp16, var_43477_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4589_cast_fp16")]; + string _SplitHeadsQ__mh_w_4591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4591_equation_0, values = (var_43709_cast_fp16, var_43484_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4591_cast_fp16")]; + string _SplitHeadsQ__mh_w_4593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4593_equation_0, values = (var_43713_cast_fp16, var_43491_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4593_cast_fp16")]; + string _SplitHeadsQ__mh_w_4595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4595_equation_0, values = (var_43713_cast_fp16, var_43498_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4595_cast_fp16")]; + string _SplitHeadsQ__mh_w_4597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4597_equation_0, values = (var_43713_cast_fp16, var_43505_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4597_cast_fp16")]; + string _SplitHeadsQ__mh_w_4599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4599_equation_0, values = (var_43713_cast_fp16, var_43512_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4599_cast_fp16")]; + string _SplitHeadsQ__mh_w_4601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4601_equation_0, values = (var_43717_cast_fp16, var_43519_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4601_cast_fp16")]; + string _SplitHeadsQ__mh_w_4603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4603_equation_0, values = (var_43717_cast_fp16, var_43526_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4603_cast_fp16")]; + string _SplitHeadsQ__mh_w_4605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4605_equation_0, values = (var_43717_cast_fp16, var_43533_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4605_cast_fp16")]; + string _SplitHeadsQ__mh_w_4607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4607_equation_0, values = (var_43717_cast_fp16, var_43540_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4607_cast_fp16")]; + string _SplitHeadsQ__mh_w_4609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4609_equation_0, values = (var_43721_cast_fp16, var_43547_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4609_cast_fp16")]; + string _SplitHeadsQ__mh_w_4611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4611_equation_0, values = (var_43721_cast_fp16, var_43554_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4611_cast_fp16")]; + string _SplitHeadsQ__mh_w_4613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4613_equation_0, values = (var_43721_cast_fp16, var_43561_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4613_cast_fp16")]; + string _SplitHeadsQ__mh_w_4615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4615_equation_0, values = (var_43721_cast_fp16, var_43568_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4615_cast_fp16")]; + string _SplitHeadsQ__mh_w_4617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4617_equation_0, values = (var_43725_cast_fp16, var_43575_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4617_cast_fp16")]; + string _SplitHeadsQ__mh_w_4619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4619_equation_0, values = (var_43725_cast_fp16, var_43582_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4619_cast_fp16")]; + string _SplitHeadsQ__mh_w_4621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4621_equation_0, values = (var_43725_cast_fp16, var_43589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4621_cast_fp16")]; + string _SplitHeadsQ__mh_w_4623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4623_equation_0, values = (var_43725_cast_fp16, var_43596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4623_cast_fp16")]; + string _SplitHeadsQ__mh_w_4625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4625_equation_0, values = (var_43729_cast_fp16, var_43603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4625_cast_fp16")]; + string _SplitHeadsQ__mh_w_4627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4627_equation_0, values = (var_43729_cast_fp16, var_43610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4627_cast_fp16")]; + string _SplitHeadsQ__mh_w_4629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4629_equation_0, values = (var_43729_cast_fp16, var_43617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4629_cast_fp16")]; + string _SplitHeadsQ__mh_w_4631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4631_equation_0, values = (var_43729_cast_fp16, var_43624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4631_cast_fp16")]; + string _SplitHeadsQ__mh_w_4633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4633_equation_0, values = (var_43733_cast_fp16, var_43631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4633_cast_fp16")]; + string _SplitHeadsQ__mh_w_4635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4635_equation_0, values = (var_43733_cast_fp16, var_43638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4635_cast_fp16")]; + string _SplitHeadsQ__mh_w_4637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4637_equation_0, values = (var_43733_cast_fp16, var_43645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4637_cast_fp16")]; + string _SplitHeadsQ__mh_w_4639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4639_equation_0, values = (var_43733_cast_fp16, var_43652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4639_cast_fp16")]; + fp16 var_43974_to_fp16 = const()[name = string("op_43974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4481_cast_fp16, y = var_43974_to_fp16)[name = string("aw_chunk_4481_cast_fp16")]; + fp16 var_43976_to_fp16 = const()[name = string("op_43976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4483_cast_fp16, y = var_43976_to_fp16)[name = string("aw_chunk_4483_cast_fp16")]; + fp16 var_43978_to_fp16 = const()[name = string("op_43978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4485_cast_fp16, y = var_43978_to_fp16)[name = string("aw_chunk_4485_cast_fp16")]; + fp16 var_43980_to_fp16 = const()[name = string("op_43980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4487_cast_fp16, y = var_43980_to_fp16)[name = string("aw_chunk_4487_cast_fp16")]; + fp16 var_43982_to_fp16 = const()[name = string("op_43982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4489_cast_fp16, y = var_43982_to_fp16)[name = string("aw_chunk_4489_cast_fp16")]; + fp16 var_43984_to_fp16 = const()[name = string("op_43984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4491_cast_fp16, y = var_43984_to_fp16)[name = string("aw_chunk_4491_cast_fp16")]; + fp16 var_43986_to_fp16 = const()[name = string("op_43986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4493_cast_fp16, y = var_43986_to_fp16)[name = string("aw_chunk_4493_cast_fp16")]; + fp16 var_43988_to_fp16 = const()[name = string("op_43988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4495_cast_fp16, y = var_43988_to_fp16)[name = string("aw_chunk_4495_cast_fp16")]; + fp16 var_43990_to_fp16 = const()[name = string("op_43990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4497_cast_fp16, y = var_43990_to_fp16)[name = string("aw_chunk_4497_cast_fp16")]; + fp16 var_43992_to_fp16 = const()[name = string("op_43992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4499_cast_fp16, y = var_43992_to_fp16)[name = string("aw_chunk_4499_cast_fp16")]; + fp16 var_43994_to_fp16 = const()[name = string("op_43994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4501_cast_fp16, y = var_43994_to_fp16)[name = string("aw_chunk_4501_cast_fp16")]; + fp16 var_43996_to_fp16 = const()[name = string("op_43996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4503_cast_fp16, y = var_43996_to_fp16)[name = string("aw_chunk_4503_cast_fp16")]; + fp16 var_43998_to_fp16 = const()[name = string("op_43998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4505_cast_fp16, y = var_43998_to_fp16)[name = string("aw_chunk_4505_cast_fp16")]; + fp16 var_44000_to_fp16 = const()[name = string("op_44000_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4507_cast_fp16, y = var_44000_to_fp16)[name = string("aw_chunk_4507_cast_fp16")]; + fp16 var_44002_to_fp16 = const()[name = string("op_44002_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4509_cast_fp16, y = var_44002_to_fp16)[name = string("aw_chunk_4509_cast_fp16")]; + fp16 var_44004_to_fp16 = const()[name = string("op_44004_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4511_cast_fp16, y = var_44004_to_fp16)[name = string("aw_chunk_4511_cast_fp16")]; + fp16 var_44006_to_fp16 = const()[name = string("op_44006_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4513_cast_fp16, y = var_44006_to_fp16)[name = string("aw_chunk_4513_cast_fp16")]; + fp16 var_44008_to_fp16 = const()[name = string("op_44008_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4515_cast_fp16, y = var_44008_to_fp16)[name = string("aw_chunk_4515_cast_fp16")]; + fp16 var_44010_to_fp16 = const()[name = string("op_44010_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4517_cast_fp16, y = var_44010_to_fp16)[name = string("aw_chunk_4517_cast_fp16")]; + fp16 var_44012_to_fp16 = const()[name = string("op_44012_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4519_cast_fp16, y = var_44012_to_fp16)[name = string("aw_chunk_4519_cast_fp16")]; + fp16 var_44014_to_fp16 = const()[name = string("op_44014_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4521_cast_fp16, y = var_44014_to_fp16)[name = string("aw_chunk_4521_cast_fp16")]; + fp16 var_44016_to_fp16 = const()[name = string("op_44016_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4523_cast_fp16, y = var_44016_to_fp16)[name = string("aw_chunk_4523_cast_fp16")]; + fp16 var_44018_to_fp16 = const()[name = string("op_44018_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4525_cast_fp16, y = var_44018_to_fp16)[name = string("aw_chunk_4525_cast_fp16")]; + fp16 var_44020_to_fp16 = const()[name = string("op_44020_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4527_cast_fp16, y = var_44020_to_fp16)[name = string("aw_chunk_4527_cast_fp16")]; + fp16 var_44022_to_fp16 = const()[name = string("op_44022_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4529_cast_fp16, y = var_44022_to_fp16)[name = string("aw_chunk_4529_cast_fp16")]; + fp16 var_44024_to_fp16 = const()[name = string("op_44024_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4531_cast_fp16, y = var_44024_to_fp16)[name = string("aw_chunk_4531_cast_fp16")]; + fp16 var_44026_to_fp16 = const()[name = string("op_44026_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4533_cast_fp16, y = var_44026_to_fp16)[name = string("aw_chunk_4533_cast_fp16")]; + fp16 var_44028_to_fp16 = const()[name = string("op_44028_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4535_cast_fp16, y = var_44028_to_fp16)[name = string("aw_chunk_4535_cast_fp16")]; + fp16 var_44030_to_fp16 = const()[name = string("op_44030_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4537_cast_fp16, y = var_44030_to_fp16)[name = string("aw_chunk_4537_cast_fp16")]; + fp16 var_44032_to_fp16 = const()[name = string("op_44032_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4539_cast_fp16, y = var_44032_to_fp16)[name = string("aw_chunk_4539_cast_fp16")]; + fp16 var_44034_to_fp16 = const()[name = string("op_44034_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4541_cast_fp16, y = var_44034_to_fp16)[name = string("aw_chunk_4541_cast_fp16")]; + fp16 var_44036_to_fp16 = const()[name = string("op_44036_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4543_cast_fp16, y = var_44036_to_fp16)[name = string("aw_chunk_4543_cast_fp16")]; + fp16 var_44038_to_fp16 = const()[name = string("op_44038_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4545_cast_fp16, y = var_44038_to_fp16)[name = string("aw_chunk_4545_cast_fp16")]; + fp16 var_44040_to_fp16 = const()[name = string("op_44040_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4547_cast_fp16, y = var_44040_to_fp16)[name = string("aw_chunk_4547_cast_fp16")]; + fp16 var_44042_to_fp16 = const()[name = string("op_44042_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4549_cast_fp16, y = var_44042_to_fp16)[name = string("aw_chunk_4549_cast_fp16")]; + fp16 var_44044_to_fp16 = const()[name = string("op_44044_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4551_cast_fp16, y = var_44044_to_fp16)[name = string("aw_chunk_4551_cast_fp16")]; + fp16 var_44046_to_fp16 = const()[name = string("op_44046_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4553_cast_fp16, y = var_44046_to_fp16)[name = string("aw_chunk_4553_cast_fp16")]; + fp16 var_44048_to_fp16 = const()[name = string("op_44048_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4555_cast_fp16, y = var_44048_to_fp16)[name = string("aw_chunk_4555_cast_fp16")]; + fp16 var_44050_to_fp16 = const()[name = string("op_44050_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4557_cast_fp16, y = var_44050_to_fp16)[name = string("aw_chunk_4557_cast_fp16")]; + fp16 var_44052_to_fp16 = const()[name = string("op_44052_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4559_cast_fp16, y = var_44052_to_fp16)[name = string("aw_chunk_4559_cast_fp16")]; + fp16 var_44054_to_fp16 = const()[name = string("op_44054_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4561_cast_fp16, y = var_44054_to_fp16)[name = string("aw_chunk_4561_cast_fp16")]; + fp16 var_44056_to_fp16 = const()[name = string("op_44056_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4563_cast_fp16, y = var_44056_to_fp16)[name = string("aw_chunk_4563_cast_fp16")]; + fp16 var_44058_to_fp16 = const()[name = string("op_44058_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4565_cast_fp16, y = var_44058_to_fp16)[name = string("aw_chunk_4565_cast_fp16")]; + fp16 var_44060_to_fp16 = const()[name = string("op_44060_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4567_cast_fp16, y = var_44060_to_fp16)[name = string("aw_chunk_4567_cast_fp16")]; + fp16 var_44062_to_fp16 = const()[name = string("op_44062_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4569_cast_fp16, y = var_44062_to_fp16)[name = string("aw_chunk_4569_cast_fp16")]; + fp16 var_44064_to_fp16 = const()[name = string("op_44064_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4571_cast_fp16, y = var_44064_to_fp16)[name = string("aw_chunk_4571_cast_fp16")]; + fp16 var_44066_to_fp16 = const()[name = string("op_44066_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4573_cast_fp16, y = var_44066_to_fp16)[name = string("aw_chunk_4573_cast_fp16")]; + fp16 var_44068_to_fp16 = const()[name = string("op_44068_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4575_cast_fp16, y = var_44068_to_fp16)[name = string("aw_chunk_4575_cast_fp16")]; + fp16 var_44070_to_fp16 = const()[name = string("op_44070_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4577_cast_fp16, y = var_44070_to_fp16)[name = string("aw_chunk_4577_cast_fp16")]; + fp16 var_44072_to_fp16 = const()[name = string("op_44072_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4579_cast_fp16, y = var_44072_to_fp16)[name = string("aw_chunk_4579_cast_fp16")]; + fp16 var_44074_to_fp16 = const()[name = string("op_44074_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4581_cast_fp16, y = var_44074_to_fp16)[name = string("aw_chunk_4581_cast_fp16")]; + fp16 var_44076_to_fp16 = const()[name = string("op_44076_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4583_cast_fp16, y = var_44076_to_fp16)[name = string("aw_chunk_4583_cast_fp16")]; + fp16 var_44078_to_fp16 = const()[name = string("op_44078_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4585_cast_fp16, y = var_44078_to_fp16)[name = string("aw_chunk_4585_cast_fp16")]; + fp16 var_44080_to_fp16 = const()[name = string("op_44080_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4587_cast_fp16, y = var_44080_to_fp16)[name = string("aw_chunk_4587_cast_fp16")]; + fp16 var_44082_to_fp16 = const()[name = string("op_44082_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4589_cast_fp16, y = var_44082_to_fp16)[name = string("aw_chunk_4589_cast_fp16")]; + fp16 var_44084_to_fp16 = const()[name = string("op_44084_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4591_cast_fp16, y = var_44084_to_fp16)[name = string("aw_chunk_4591_cast_fp16")]; + fp16 var_44086_to_fp16 = const()[name = string("op_44086_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4593_cast_fp16, y = var_44086_to_fp16)[name = string("aw_chunk_4593_cast_fp16")]; + fp16 var_44088_to_fp16 = const()[name = string("op_44088_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4595_cast_fp16, y = var_44088_to_fp16)[name = string("aw_chunk_4595_cast_fp16")]; + fp16 var_44090_to_fp16 = const()[name = string("op_44090_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4597_cast_fp16, y = var_44090_to_fp16)[name = string("aw_chunk_4597_cast_fp16")]; + fp16 var_44092_to_fp16 = const()[name = string("op_44092_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4599_cast_fp16, y = var_44092_to_fp16)[name = string("aw_chunk_4599_cast_fp16")]; + fp16 var_44094_to_fp16 = const()[name = string("op_44094_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4601_cast_fp16, y = var_44094_to_fp16)[name = string("aw_chunk_4601_cast_fp16")]; + fp16 var_44096_to_fp16 = const()[name = string("op_44096_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4603_cast_fp16, y = var_44096_to_fp16)[name = string("aw_chunk_4603_cast_fp16")]; + fp16 var_44098_to_fp16 = const()[name = string("op_44098_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4605_cast_fp16, y = var_44098_to_fp16)[name = string("aw_chunk_4605_cast_fp16")]; + fp16 var_44100_to_fp16 = const()[name = string("op_44100_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4607_cast_fp16, y = var_44100_to_fp16)[name = string("aw_chunk_4607_cast_fp16")]; + fp16 var_44102_to_fp16 = const()[name = string("op_44102_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4609_cast_fp16, y = var_44102_to_fp16)[name = string("aw_chunk_4609_cast_fp16")]; + fp16 var_44104_to_fp16 = const()[name = string("op_44104_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4611_cast_fp16, y = var_44104_to_fp16)[name = string("aw_chunk_4611_cast_fp16")]; + fp16 var_44106_to_fp16 = const()[name = string("op_44106_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4613_cast_fp16, y = var_44106_to_fp16)[name = string("aw_chunk_4613_cast_fp16")]; + fp16 var_44108_to_fp16 = const()[name = string("op_44108_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4615_cast_fp16, y = var_44108_to_fp16)[name = string("aw_chunk_4615_cast_fp16")]; + fp16 var_44110_to_fp16 = const()[name = string("op_44110_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4617_cast_fp16, y = var_44110_to_fp16)[name = string("aw_chunk_4617_cast_fp16")]; + fp16 var_44112_to_fp16 = const()[name = string("op_44112_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4619_cast_fp16, y = var_44112_to_fp16)[name = string("aw_chunk_4619_cast_fp16")]; + fp16 var_44114_to_fp16 = const()[name = string("op_44114_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4621_cast_fp16, y = var_44114_to_fp16)[name = string("aw_chunk_4621_cast_fp16")]; + fp16 var_44116_to_fp16 = const()[name = string("op_44116_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4623_cast_fp16, y = var_44116_to_fp16)[name = string("aw_chunk_4623_cast_fp16")]; + fp16 var_44118_to_fp16 = const()[name = string("op_44118_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4625_cast_fp16, y = var_44118_to_fp16)[name = string("aw_chunk_4625_cast_fp16")]; + fp16 var_44120_to_fp16 = const()[name = string("op_44120_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4627_cast_fp16, y = var_44120_to_fp16)[name = string("aw_chunk_4627_cast_fp16")]; + fp16 var_44122_to_fp16 = const()[name = string("op_44122_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4629_cast_fp16, y = var_44122_to_fp16)[name = string("aw_chunk_4629_cast_fp16")]; + fp16 var_44124_to_fp16 = const()[name = string("op_44124_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4631_cast_fp16, y = var_44124_to_fp16)[name = string("aw_chunk_4631_cast_fp16")]; + fp16 var_44126_to_fp16 = const()[name = string("op_44126_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4633_cast_fp16, y = var_44126_to_fp16)[name = string("aw_chunk_4633_cast_fp16")]; + fp16 var_44128_to_fp16 = const()[name = string("op_44128_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4635_cast_fp16, y = var_44128_to_fp16)[name = string("aw_chunk_4635_cast_fp16")]; + fp16 var_44130_to_fp16 = const()[name = string("op_44130_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4637_cast_fp16, y = var_44130_to_fp16)[name = string("aw_chunk_4637_cast_fp16")]; + fp16 var_44132_to_fp16 = const()[name = string("op_44132_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4639_cast_fp16, y = var_44132_to_fp16)[name = string("aw_chunk_4639_cast_fp16")]; + tensor var_44134_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4481_cast_fp16)[name = string("op_44134_cast_fp16")]; + tensor var_44135_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4483_cast_fp16)[name = string("op_44135_cast_fp16")]; + tensor var_44136_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4485_cast_fp16)[name = string("op_44136_cast_fp16")]; + tensor var_44137_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4487_cast_fp16)[name = string("op_44137_cast_fp16")]; + tensor var_44138_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4489_cast_fp16)[name = string("op_44138_cast_fp16")]; + tensor var_44139_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4491_cast_fp16)[name = string("op_44139_cast_fp16")]; + tensor var_44140_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4493_cast_fp16)[name = string("op_44140_cast_fp16")]; + tensor var_44141_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4495_cast_fp16)[name = string("op_44141_cast_fp16")]; + tensor var_44142_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4497_cast_fp16)[name = string("op_44142_cast_fp16")]; + tensor var_44143_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4499_cast_fp16)[name = string("op_44143_cast_fp16")]; + tensor var_44144_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4501_cast_fp16)[name = string("op_44144_cast_fp16")]; + tensor var_44145_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4503_cast_fp16)[name = string("op_44145_cast_fp16")]; + tensor var_44146_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4505_cast_fp16)[name = string("op_44146_cast_fp16")]; + tensor var_44147_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4507_cast_fp16)[name = string("op_44147_cast_fp16")]; + tensor var_44148_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4509_cast_fp16)[name = string("op_44148_cast_fp16")]; + tensor var_44149_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4511_cast_fp16)[name = string("op_44149_cast_fp16")]; + tensor var_44150_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4513_cast_fp16)[name = string("op_44150_cast_fp16")]; + tensor var_44151_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4515_cast_fp16)[name = string("op_44151_cast_fp16")]; + tensor var_44152_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4517_cast_fp16)[name = string("op_44152_cast_fp16")]; + tensor var_44153_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4519_cast_fp16)[name = string("op_44153_cast_fp16")]; + tensor var_44154_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4521_cast_fp16)[name = string("op_44154_cast_fp16")]; + tensor var_44155_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4523_cast_fp16)[name = string("op_44155_cast_fp16")]; + tensor var_44156_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4525_cast_fp16)[name = string("op_44156_cast_fp16")]; + tensor var_44157_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4527_cast_fp16)[name = string("op_44157_cast_fp16")]; + tensor var_44158_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4529_cast_fp16)[name = string("op_44158_cast_fp16")]; + tensor var_44159_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4531_cast_fp16)[name = string("op_44159_cast_fp16")]; + tensor var_44160_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4533_cast_fp16)[name = string("op_44160_cast_fp16")]; + tensor var_44161_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4535_cast_fp16)[name = string("op_44161_cast_fp16")]; + tensor var_44162_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4537_cast_fp16)[name = string("op_44162_cast_fp16")]; + tensor var_44163_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4539_cast_fp16)[name = string("op_44163_cast_fp16")]; + tensor var_44164_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4541_cast_fp16)[name = string("op_44164_cast_fp16")]; + tensor var_44165_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4543_cast_fp16)[name = string("op_44165_cast_fp16")]; + tensor var_44166_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4545_cast_fp16)[name = string("op_44166_cast_fp16")]; + tensor var_44167_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4547_cast_fp16)[name = string("op_44167_cast_fp16")]; + tensor var_44168_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4549_cast_fp16)[name = string("op_44168_cast_fp16")]; + tensor var_44169_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4551_cast_fp16)[name = string("op_44169_cast_fp16")]; + tensor var_44170_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4553_cast_fp16)[name = string("op_44170_cast_fp16")]; + tensor var_44171_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4555_cast_fp16)[name = string("op_44171_cast_fp16")]; + tensor var_44172_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4557_cast_fp16)[name = string("op_44172_cast_fp16")]; + tensor var_44173_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4559_cast_fp16)[name = string("op_44173_cast_fp16")]; + tensor var_44174_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4561_cast_fp16)[name = string("op_44174_cast_fp16")]; + tensor var_44175_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4563_cast_fp16)[name = string("op_44175_cast_fp16")]; + tensor var_44176_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4565_cast_fp16)[name = string("op_44176_cast_fp16")]; + tensor var_44177_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4567_cast_fp16)[name = string("op_44177_cast_fp16")]; + tensor var_44178_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4569_cast_fp16)[name = string("op_44178_cast_fp16")]; + tensor var_44179_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4571_cast_fp16)[name = string("op_44179_cast_fp16")]; + tensor var_44180_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4573_cast_fp16)[name = string("op_44180_cast_fp16")]; + tensor var_44181_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4575_cast_fp16)[name = string("op_44181_cast_fp16")]; + tensor var_44182_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4577_cast_fp16)[name = string("op_44182_cast_fp16")]; + tensor var_44183_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4579_cast_fp16)[name = string("op_44183_cast_fp16")]; + tensor var_44184_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4581_cast_fp16)[name = string("op_44184_cast_fp16")]; + tensor var_44185_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4583_cast_fp16)[name = string("op_44185_cast_fp16")]; + tensor var_44186_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4585_cast_fp16)[name = string("op_44186_cast_fp16")]; + tensor var_44187_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4587_cast_fp16)[name = string("op_44187_cast_fp16")]; + tensor var_44188_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4589_cast_fp16)[name = string("op_44188_cast_fp16")]; + tensor var_44189_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4591_cast_fp16)[name = string("op_44189_cast_fp16")]; + tensor var_44190_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4593_cast_fp16)[name = string("op_44190_cast_fp16")]; + tensor var_44191_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4595_cast_fp16)[name = string("op_44191_cast_fp16")]; + tensor var_44192_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4597_cast_fp16)[name = string("op_44192_cast_fp16")]; + tensor var_44193_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4599_cast_fp16)[name = string("op_44193_cast_fp16")]; + tensor var_44194_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4601_cast_fp16)[name = string("op_44194_cast_fp16")]; + tensor var_44195_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4603_cast_fp16)[name = string("op_44195_cast_fp16")]; + tensor var_44196_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4605_cast_fp16)[name = string("op_44196_cast_fp16")]; + tensor var_44197_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4607_cast_fp16)[name = string("op_44197_cast_fp16")]; + tensor var_44198_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4609_cast_fp16)[name = string("op_44198_cast_fp16")]; + tensor var_44199_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4611_cast_fp16)[name = string("op_44199_cast_fp16")]; + tensor var_44200_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4613_cast_fp16)[name = string("op_44200_cast_fp16")]; + tensor var_44201_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4615_cast_fp16)[name = string("op_44201_cast_fp16")]; + tensor var_44202_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4617_cast_fp16)[name = string("op_44202_cast_fp16")]; + tensor var_44203_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4619_cast_fp16)[name = string("op_44203_cast_fp16")]; + tensor var_44204_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4621_cast_fp16)[name = string("op_44204_cast_fp16")]; + tensor var_44205_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4623_cast_fp16)[name = string("op_44205_cast_fp16")]; + tensor var_44206_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4625_cast_fp16)[name = string("op_44206_cast_fp16")]; + tensor var_44207_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4627_cast_fp16)[name = string("op_44207_cast_fp16")]; + tensor var_44208_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4629_cast_fp16)[name = string("op_44208_cast_fp16")]; + tensor var_44209_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4631_cast_fp16)[name = string("op_44209_cast_fp16")]; + tensor var_44210_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4633_cast_fp16)[name = string("op_44210_cast_fp16")]; + tensor var_44211_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4635_cast_fp16)[name = string("op_44211_cast_fp16")]; + tensor var_44212_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4637_cast_fp16)[name = string("op_44212_cast_fp16")]; + tensor var_44213_cast_fp16 = softmax(axis = var_42959, x = aw_chunk_4639_cast_fp16)[name = string("op_44213_cast_fp16")]; + string var_44215_equation_0 = const()[name = string("op_44215_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44215_cast_fp16 = einsum(equation = var_44215_equation_0, values = (var_43735_cast_fp16, var_44134_cast_fp16))[name = string("op_44215_cast_fp16")]; + string var_44217_equation_0 = const()[name = string("op_44217_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44217_cast_fp16 = einsum(equation = var_44217_equation_0, values = (var_43735_cast_fp16, var_44135_cast_fp16))[name = string("op_44217_cast_fp16")]; + string var_44219_equation_0 = const()[name = string("op_44219_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44219_cast_fp16 = einsum(equation = var_44219_equation_0, values = (var_43735_cast_fp16, var_44136_cast_fp16))[name = string("op_44219_cast_fp16")]; + string var_44221_equation_0 = const()[name = string("op_44221_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44221_cast_fp16 = einsum(equation = var_44221_equation_0, values = (var_43735_cast_fp16, var_44137_cast_fp16))[name = string("op_44221_cast_fp16")]; + string var_44223_equation_0 = const()[name = string("op_44223_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44223_cast_fp16 = einsum(equation = var_44223_equation_0, values = (var_43739_cast_fp16, var_44138_cast_fp16))[name = string("op_44223_cast_fp16")]; + string var_44225_equation_0 = const()[name = string("op_44225_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44225_cast_fp16 = einsum(equation = var_44225_equation_0, values = (var_43739_cast_fp16, var_44139_cast_fp16))[name = string("op_44225_cast_fp16")]; + string var_44227_equation_0 = const()[name = string("op_44227_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44227_cast_fp16 = einsum(equation = var_44227_equation_0, values = (var_43739_cast_fp16, var_44140_cast_fp16))[name = string("op_44227_cast_fp16")]; + string var_44229_equation_0 = const()[name = string("op_44229_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44229_cast_fp16 = einsum(equation = var_44229_equation_0, values = (var_43739_cast_fp16, var_44141_cast_fp16))[name = string("op_44229_cast_fp16")]; + string var_44231_equation_0 = const()[name = string("op_44231_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44231_cast_fp16 = einsum(equation = var_44231_equation_0, values = (var_43743_cast_fp16, var_44142_cast_fp16))[name = string("op_44231_cast_fp16")]; + string var_44233_equation_0 = const()[name = string("op_44233_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44233_cast_fp16 = einsum(equation = var_44233_equation_0, values = (var_43743_cast_fp16, var_44143_cast_fp16))[name = string("op_44233_cast_fp16")]; + string var_44235_equation_0 = const()[name = string("op_44235_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44235_cast_fp16 = einsum(equation = var_44235_equation_0, values = (var_43743_cast_fp16, var_44144_cast_fp16))[name = string("op_44235_cast_fp16")]; + string var_44237_equation_0 = const()[name = string("op_44237_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44237_cast_fp16 = einsum(equation = var_44237_equation_0, values = (var_43743_cast_fp16, var_44145_cast_fp16))[name = string("op_44237_cast_fp16")]; + string var_44239_equation_0 = const()[name = string("op_44239_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44239_cast_fp16 = einsum(equation = var_44239_equation_0, values = (var_43747_cast_fp16, var_44146_cast_fp16))[name = string("op_44239_cast_fp16")]; + string var_44241_equation_0 = const()[name = string("op_44241_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44241_cast_fp16 = einsum(equation = var_44241_equation_0, values = (var_43747_cast_fp16, var_44147_cast_fp16))[name = string("op_44241_cast_fp16")]; + string var_44243_equation_0 = const()[name = string("op_44243_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44243_cast_fp16 = einsum(equation = var_44243_equation_0, values = (var_43747_cast_fp16, var_44148_cast_fp16))[name = string("op_44243_cast_fp16")]; + string var_44245_equation_0 = const()[name = string("op_44245_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44245_cast_fp16 = einsum(equation = var_44245_equation_0, values = (var_43747_cast_fp16, var_44149_cast_fp16))[name = string("op_44245_cast_fp16")]; + string var_44247_equation_0 = const()[name = string("op_44247_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44247_cast_fp16 = einsum(equation = var_44247_equation_0, values = (var_43751_cast_fp16, var_44150_cast_fp16))[name = string("op_44247_cast_fp16")]; + string var_44249_equation_0 = const()[name = string("op_44249_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44249_cast_fp16 = einsum(equation = var_44249_equation_0, values = (var_43751_cast_fp16, var_44151_cast_fp16))[name = string("op_44249_cast_fp16")]; + string var_44251_equation_0 = const()[name = string("op_44251_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44251_cast_fp16 = einsum(equation = var_44251_equation_0, values = (var_43751_cast_fp16, var_44152_cast_fp16))[name = string("op_44251_cast_fp16")]; + string var_44253_equation_0 = const()[name = string("op_44253_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44253_cast_fp16 = einsum(equation = var_44253_equation_0, values = (var_43751_cast_fp16, var_44153_cast_fp16))[name = string("op_44253_cast_fp16")]; + string var_44255_equation_0 = const()[name = string("op_44255_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44255_cast_fp16 = einsum(equation = var_44255_equation_0, values = (var_43755_cast_fp16, var_44154_cast_fp16))[name = string("op_44255_cast_fp16")]; + string var_44257_equation_0 = const()[name = string("op_44257_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44257_cast_fp16 = einsum(equation = var_44257_equation_0, values = (var_43755_cast_fp16, var_44155_cast_fp16))[name = string("op_44257_cast_fp16")]; + string var_44259_equation_0 = const()[name = string("op_44259_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44259_cast_fp16 = einsum(equation = var_44259_equation_0, values = (var_43755_cast_fp16, var_44156_cast_fp16))[name = string("op_44259_cast_fp16")]; + string var_44261_equation_0 = const()[name = string("op_44261_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44261_cast_fp16 = einsum(equation = var_44261_equation_0, values = (var_43755_cast_fp16, var_44157_cast_fp16))[name = string("op_44261_cast_fp16")]; + string var_44263_equation_0 = const()[name = string("op_44263_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44263_cast_fp16 = einsum(equation = var_44263_equation_0, values = (var_43759_cast_fp16, var_44158_cast_fp16))[name = string("op_44263_cast_fp16")]; + string var_44265_equation_0 = const()[name = string("op_44265_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44265_cast_fp16 = einsum(equation = var_44265_equation_0, values = (var_43759_cast_fp16, var_44159_cast_fp16))[name = string("op_44265_cast_fp16")]; + string var_44267_equation_0 = const()[name = string("op_44267_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44267_cast_fp16 = einsum(equation = var_44267_equation_0, values = (var_43759_cast_fp16, var_44160_cast_fp16))[name = string("op_44267_cast_fp16")]; + string var_44269_equation_0 = const()[name = string("op_44269_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44269_cast_fp16 = einsum(equation = var_44269_equation_0, values = (var_43759_cast_fp16, var_44161_cast_fp16))[name = string("op_44269_cast_fp16")]; + string var_44271_equation_0 = const()[name = string("op_44271_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44271_cast_fp16 = einsum(equation = var_44271_equation_0, values = (var_43763_cast_fp16, var_44162_cast_fp16))[name = string("op_44271_cast_fp16")]; + string var_44273_equation_0 = const()[name = string("op_44273_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44273_cast_fp16 = einsum(equation = var_44273_equation_0, values = (var_43763_cast_fp16, var_44163_cast_fp16))[name = string("op_44273_cast_fp16")]; + string var_44275_equation_0 = const()[name = string("op_44275_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44275_cast_fp16 = einsum(equation = var_44275_equation_0, values = (var_43763_cast_fp16, var_44164_cast_fp16))[name = string("op_44275_cast_fp16")]; + string var_44277_equation_0 = const()[name = string("op_44277_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44277_cast_fp16 = einsum(equation = var_44277_equation_0, values = (var_43763_cast_fp16, var_44165_cast_fp16))[name = string("op_44277_cast_fp16")]; + string var_44279_equation_0 = const()[name = string("op_44279_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44279_cast_fp16 = einsum(equation = var_44279_equation_0, values = (var_43767_cast_fp16, var_44166_cast_fp16))[name = string("op_44279_cast_fp16")]; + string var_44281_equation_0 = const()[name = string("op_44281_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44281_cast_fp16 = einsum(equation = var_44281_equation_0, values = (var_43767_cast_fp16, var_44167_cast_fp16))[name = string("op_44281_cast_fp16")]; + string var_44283_equation_0 = const()[name = string("op_44283_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44283_cast_fp16 = einsum(equation = var_44283_equation_0, values = (var_43767_cast_fp16, var_44168_cast_fp16))[name = string("op_44283_cast_fp16")]; + string var_44285_equation_0 = const()[name = string("op_44285_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44285_cast_fp16 = einsum(equation = var_44285_equation_0, values = (var_43767_cast_fp16, var_44169_cast_fp16))[name = string("op_44285_cast_fp16")]; + string var_44287_equation_0 = const()[name = string("op_44287_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44287_cast_fp16 = einsum(equation = var_44287_equation_0, values = (var_43771_cast_fp16, var_44170_cast_fp16))[name = string("op_44287_cast_fp16")]; + string var_44289_equation_0 = const()[name = string("op_44289_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44289_cast_fp16 = einsum(equation = var_44289_equation_0, values = (var_43771_cast_fp16, var_44171_cast_fp16))[name = string("op_44289_cast_fp16")]; + string var_44291_equation_0 = const()[name = string("op_44291_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44291_cast_fp16 = einsum(equation = var_44291_equation_0, values = (var_43771_cast_fp16, var_44172_cast_fp16))[name = string("op_44291_cast_fp16")]; + string var_44293_equation_0 = const()[name = string("op_44293_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44293_cast_fp16 = einsum(equation = var_44293_equation_0, values = (var_43771_cast_fp16, var_44173_cast_fp16))[name = string("op_44293_cast_fp16")]; + string var_44295_equation_0 = const()[name = string("op_44295_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44295_cast_fp16 = einsum(equation = var_44295_equation_0, values = (var_43775_cast_fp16, var_44174_cast_fp16))[name = string("op_44295_cast_fp16")]; + string var_44297_equation_0 = const()[name = string("op_44297_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44297_cast_fp16 = einsum(equation = var_44297_equation_0, values = (var_43775_cast_fp16, var_44175_cast_fp16))[name = string("op_44297_cast_fp16")]; + string var_44299_equation_0 = const()[name = string("op_44299_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44299_cast_fp16 = einsum(equation = var_44299_equation_0, values = (var_43775_cast_fp16, var_44176_cast_fp16))[name = string("op_44299_cast_fp16")]; + string var_44301_equation_0 = const()[name = string("op_44301_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44301_cast_fp16 = einsum(equation = var_44301_equation_0, values = (var_43775_cast_fp16, var_44177_cast_fp16))[name = string("op_44301_cast_fp16")]; + string var_44303_equation_0 = const()[name = string("op_44303_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44303_cast_fp16 = einsum(equation = var_44303_equation_0, values = (var_43779_cast_fp16, var_44178_cast_fp16))[name = string("op_44303_cast_fp16")]; + string var_44305_equation_0 = const()[name = string("op_44305_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44305_cast_fp16 = einsum(equation = var_44305_equation_0, values = (var_43779_cast_fp16, var_44179_cast_fp16))[name = string("op_44305_cast_fp16")]; + string var_44307_equation_0 = const()[name = string("op_44307_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44307_cast_fp16 = einsum(equation = var_44307_equation_0, values = (var_43779_cast_fp16, var_44180_cast_fp16))[name = string("op_44307_cast_fp16")]; + string var_44309_equation_0 = const()[name = string("op_44309_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44309_cast_fp16 = einsum(equation = var_44309_equation_0, values = (var_43779_cast_fp16, var_44181_cast_fp16))[name = string("op_44309_cast_fp16")]; + string var_44311_equation_0 = const()[name = string("op_44311_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44311_cast_fp16 = einsum(equation = var_44311_equation_0, values = (var_43783_cast_fp16, var_44182_cast_fp16))[name = string("op_44311_cast_fp16")]; + string var_44313_equation_0 = const()[name = string("op_44313_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44313_cast_fp16 = einsum(equation = var_44313_equation_0, values = (var_43783_cast_fp16, var_44183_cast_fp16))[name = string("op_44313_cast_fp16")]; + string var_44315_equation_0 = const()[name = string("op_44315_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44315_cast_fp16 = einsum(equation = var_44315_equation_0, values = (var_43783_cast_fp16, var_44184_cast_fp16))[name = string("op_44315_cast_fp16")]; + string var_44317_equation_0 = const()[name = string("op_44317_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44317_cast_fp16 = einsum(equation = var_44317_equation_0, values = (var_43783_cast_fp16, var_44185_cast_fp16))[name = string("op_44317_cast_fp16")]; + string var_44319_equation_0 = const()[name = string("op_44319_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44319_cast_fp16 = einsum(equation = var_44319_equation_0, values = (var_43787_cast_fp16, var_44186_cast_fp16))[name = string("op_44319_cast_fp16")]; + string var_44321_equation_0 = const()[name = string("op_44321_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44321_cast_fp16 = einsum(equation = var_44321_equation_0, values = (var_43787_cast_fp16, var_44187_cast_fp16))[name = string("op_44321_cast_fp16")]; + string var_44323_equation_0 = const()[name = string("op_44323_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44323_cast_fp16 = einsum(equation = var_44323_equation_0, values = (var_43787_cast_fp16, var_44188_cast_fp16))[name = string("op_44323_cast_fp16")]; + string var_44325_equation_0 = const()[name = string("op_44325_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44325_cast_fp16 = einsum(equation = var_44325_equation_0, values = (var_43787_cast_fp16, var_44189_cast_fp16))[name = string("op_44325_cast_fp16")]; + string var_44327_equation_0 = const()[name = string("op_44327_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44327_cast_fp16 = einsum(equation = var_44327_equation_0, values = (var_43791_cast_fp16, var_44190_cast_fp16))[name = string("op_44327_cast_fp16")]; + string var_44329_equation_0 = const()[name = string("op_44329_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44329_cast_fp16 = einsum(equation = var_44329_equation_0, values = (var_43791_cast_fp16, var_44191_cast_fp16))[name = string("op_44329_cast_fp16")]; + string var_44331_equation_0 = const()[name = string("op_44331_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44331_cast_fp16 = einsum(equation = var_44331_equation_0, values = (var_43791_cast_fp16, var_44192_cast_fp16))[name = string("op_44331_cast_fp16")]; + string var_44333_equation_0 = const()[name = string("op_44333_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44333_cast_fp16 = einsum(equation = var_44333_equation_0, values = (var_43791_cast_fp16, var_44193_cast_fp16))[name = string("op_44333_cast_fp16")]; + string var_44335_equation_0 = const()[name = string("op_44335_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44335_cast_fp16 = einsum(equation = var_44335_equation_0, values = (var_43795_cast_fp16, var_44194_cast_fp16))[name = string("op_44335_cast_fp16")]; + string var_44337_equation_0 = const()[name = string("op_44337_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44337_cast_fp16 = einsum(equation = var_44337_equation_0, values = (var_43795_cast_fp16, var_44195_cast_fp16))[name = string("op_44337_cast_fp16")]; + string var_44339_equation_0 = const()[name = string("op_44339_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44339_cast_fp16 = einsum(equation = var_44339_equation_0, values = (var_43795_cast_fp16, var_44196_cast_fp16))[name = string("op_44339_cast_fp16")]; + string var_44341_equation_0 = const()[name = string("op_44341_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44341_cast_fp16 = einsum(equation = var_44341_equation_0, values = (var_43795_cast_fp16, var_44197_cast_fp16))[name = string("op_44341_cast_fp16")]; + string var_44343_equation_0 = const()[name = string("op_44343_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44343_cast_fp16 = einsum(equation = var_44343_equation_0, values = (var_43799_cast_fp16, var_44198_cast_fp16))[name = string("op_44343_cast_fp16")]; + string var_44345_equation_0 = const()[name = string("op_44345_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44345_cast_fp16 = einsum(equation = var_44345_equation_0, values = (var_43799_cast_fp16, var_44199_cast_fp16))[name = string("op_44345_cast_fp16")]; + string var_44347_equation_0 = const()[name = string("op_44347_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44347_cast_fp16 = einsum(equation = var_44347_equation_0, values = (var_43799_cast_fp16, var_44200_cast_fp16))[name = string("op_44347_cast_fp16")]; + string var_44349_equation_0 = const()[name = string("op_44349_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44349_cast_fp16 = einsum(equation = var_44349_equation_0, values = (var_43799_cast_fp16, var_44201_cast_fp16))[name = string("op_44349_cast_fp16")]; + string var_44351_equation_0 = const()[name = string("op_44351_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44351_cast_fp16 = einsum(equation = var_44351_equation_0, values = (var_43803_cast_fp16, var_44202_cast_fp16))[name = string("op_44351_cast_fp16")]; + string var_44353_equation_0 = const()[name = string("op_44353_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44353_cast_fp16 = einsum(equation = var_44353_equation_0, values = (var_43803_cast_fp16, var_44203_cast_fp16))[name = string("op_44353_cast_fp16")]; + string var_44355_equation_0 = const()[name = string("op_44355_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44355_cast_fp16 = einsum(equation = var_44355_equation_0, values = (var_43803_cast_fp16, var_44204_cast_fp16))[name = string("op_44355_cast_fp16")]; + string var_44357_equation_0 = const()[name = string("op_44357_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44357_cast_fp16 = einsum(equation = var_44357_equation_0, values = (var_43803_cast_fp16, var_44205_cast_fp16))[name = string("op_44357_cast_fp16")]; + string var_44359_equation_0 = const()[name = string("op_44359_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44359_cast_fp16 = einsum(equation = var_44359_equation_0, values = (var_43807_cast_fp16, var_44206_cast_fp16))[name = string("op_44359_cast_fp16")]; + string var_44361_equation_0 = const()[name = string("op_44361_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44361_cast_fp16 = einsum(equation = var_44361_equation_0, values = (var_43807_cast_fp16, var_44207_cast_fp16))[name = string("op_44361_cast_fp16")]; + string var_44363_equation_0 = const()[name = string("op_44363_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44363_cast_fp16 = einsum(equation = var_44363_equation_0, values = (var_43807_cast_fp16, var_44208_cast_fp16))[name = string("op_44363_cast_fp16")]; + string var_44365_equation_0 = const()[name = string("op_44365_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44365_cast_fp16 = einsum(equation = var_44365_equation_0, values = (var_43807_cast_fp16, var_44209_cast_fp16))[name = string("op_44365_cast_fp16")]; + string var_44367_equation_0 = const()[name = string("op_44367_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44367_cast_fp16 = einsum(equation = var_44367_equation_0, values = (var_43811_cast_fp16, var_44210_cast_fp16))[name = string("op_44367_cast_fp16")]; + string var_44369_equation_0 = const()[name = string("op_44369_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44369_cast_fp16 = einsum(equation = var_44369_equation_0, values = (var_43811_cast_fp16, var_44211_cast_fp16))[name = string("op_44369_cast_fp16")]; + string var_44371_equation_0 = const()[name = string("op_44371_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44371_cast_fp16 = einsum(equation = var_44371_equation_0, values = (var_43811_cast_fp16, var_44212_cast_fp16))[name = string("op_44371_cast_fp16")]; + string var_44373_equation_0 = const()[name = string("op_44373_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44373_cast_fp16 = einsum(equation = var_44373_equation_0, values = (var_43811_cast_fp16, var_44213_cast_fp16))[name = string("op_44373_cast_fp16")]; + bool var_44375_interleave_0 = const()[name = string("op_44375_interleave_0"), val = bool(false)]; + tensor var_44375_cast_fp16 = concat(axis = var_42934, interleave = var_44375_interleave_0, values = (var_44215_cast_fp16, var_44217_cast_fp16, var_44219_cast_fp16, var_44221_cast_fp16))[name = string("op_44375_cast_fp16")]; + bool var_44377_interleave_0 = const()[name = string("op_44377_interleave_0"), val = bool(false)]; + tensor var_44377_cast_fp16 = concat(axis = var_42934, interleave = var_44377_interleave_0, values = (var_44223_cast_fp16, var_44225_cast_fp16, var_44227_cast_fp16, var_44229_cast_fp16))[name = string("op_44377_cast_fp16")]; + bool var_44379_interleave_0 = const()[name = string("op_44379_interleave_0"), val = bool(false)]; + tensor var_44379_cast_fp16 = concat(axis = var_42934, interleave = var_44379_interleave_0, values = (var_44231_cast_fp16, var_44233_cast_fp16, var_44235_cast_fp16, var_44237_cast_fp16))[name = string("op_44379_cast_fp16")]; + bool var_44381_interleave_0 = const()[name = string("op_44381_interleave_0"), val = bool(false)]; + tensor var_44381_cast_fp16 = concat(axis = var_42934, interleave = var_44381_interleave_0, values = (var_44239_cast_fp16, var_44241_cast_fp16, var_44243_cast_fp16, var_44245_cast_fp16))[name = string("op_44381_cast_fp16")]; + bool var_44383_interleave_0 = const()[name = string("op_44383_interleave_0"), val = bool(false)]; + tensor var_44383_cast_fp16 = concat(axis = var_42934, interleave = var_44383_interleave_0, values = (var_44247_cast_fp16, var_44249_cast_fp16, var_44251_cast_fp16, var_44253_cast_fp16))[name = string("op_44383_cast_fp16")]; + bool var_44385_interleave_0 = const()[name = string("op_44385_interleave_0"), val = bool(false)]; + tensor var_44385_cast_fp16 = concat(axis = var_42934, interleave = var_44385_interleave_0, values = (var_44255_cast_fp16, var_44257_cast_fp16, var_44259_cast_fp16, var_44261_cast_fp16))[name = string("op_44385_cast_fp16")]; + bool var_44387_interleave_0 = const()[name = string("op_44387_interleave_0"), val = bool(false)]; + tensor var_44387_cast_fp16 = concat(axis = var_42934, interleave = var_44387_interleave_0, values = (var_44263_cast_fp16, var_44265_cast_fp16, var_44267_cast_fp16, var_44269_cast_fp16))[name = string("op_44387_cast_fp16")]; + bool var_44389_interleave_0 = const()[name = string("op_44389_interleave_0"), val = bool(false)]; + tensor var_44389_cast_fp16 = concat(axis = var_42934, interleave = var_44389_interleave_0, values = (var_44271_cast_fp16, var_44273_cast_fp16, var_44275_cast_fp16, var_44277_cast_fp16))[name = string("op_44389_cast_fp16")]; + bool var_44391_interleave_0 = const()[name = string("op_44391_interleave_0"), val = bool(false)]; + tensor var_44391_cast_fp16 = concat(axis = var_42934, interleave = var_44391_interleave_0, values = (var_44279_cast_fp16, var_44281_cast_fp16, var_44283_cast_fp16, var_44285_cast_fp16))[name = string("op_44391_cast_fp16")]; + bool var_44393_interleave_0 = const()[name = string("op_44393_interleave_0"), val = bool(false)]; + tensor var_44393_cast_fp16 = concat(axis = var_42934, interleave = var_44393_interleave_0, values = (var_44287_cast_fp16, var_44289_cast_fp16, var_44291_cast_fp16, var_44293_cast_fp16))[name = string("op_44393_cast_fp16")]; + bool var_44395_interleave_0 = const()[name = string("op_44395_interleave_0"), val = bool(false)]; + tensor var_44395_cast_fp16 = concat(axis = var_42934, interleave = var_44395_interleave_0, values = (var_44295_cast_fp16, var_44297_cast_fp16, var_44299_cast_fp16, var_44301_cast_fp16))[name = string("op_44395_cast_fp16")]; + bool var_44397_interleave_0 = const()[name = string("op_44397_interleave_0"), val = bool(false)]; + tensor var_44397_cast_fp16 = concat(axis = var_42934, interleave = var_44397_interleave_0, values = (var_44303_cast_fp16, var_44305_cast_fp16, var_44307_cast_fp16, var_44309_cast_fp16))[name = string("op_44397_cast_fp16")]; + bool var_44399_interleave_0 = const()[name = string("op_44399_interleave_0"), val = bool(false)]; + tensor var_44399_cast_fp16 = concat(axis = var_42934, interleave = var_44399_interleave_0, values = (var_44311_cast_fp16, var_44313_cast_fp16, var_44315_cast_fp16, var_44317_cast_fp16))[name = string("op_44399_cast_fp16")]; + bool var_44401_interleave_0 = const()[name = string("op_44401_interleave_0"), val = bool(false)]; + tensor var_44401_cast_fp16 = concat(axis = var_42934, interleave = var_44401_interleave_0, values = (var_44319_cast_fp16, var_44321_cast_fp16, var_44323_cast_fp16, var_44325_cast_fp16))[name = string("op_44401_cast_fp16")]; + bool var_44403_interleave_0 = const()[name = string("op_44403_interleave_0"), val = bool(false)]; + tensor var_44403_cast_fp16 = concat(axis = var_42934, interleave = var_44403_interleave_0, values = (var_44327_cast_fp16, var_44329_cast_fp16, var_44331_cast_fp16, var_44333_cast_fp16))[name = string("op_44403_cast_fp16")]; + bool var_44405_interleave_0 = const()[name = string("op_44405_interleave_0"), val = bool(false)]; + tensor var_44405_cast_fp16 = concat(axis = var_42934, interleave = var_44405_interleave_0, values = (var_44335_cast_fp16, var_44337_cast_fp16, var_44339_cast_fp16, var_44341_cast_fp16))[name = string("op_44405_cast_fp16")]; + bool var_44407_interleave_0 = const()[name = string("op_44407_interleave_0"), val = bool(false)]; + tensor var_44407_cast_fp16 = concat(axis = var_42934, interleave = var_44407_interleave_0, values = (var_44343_cast_fp16, var_44345_cast_fp16, var_44347_cast_fp16, var_44349_cast_fp16))[name = string("op_44407_cast_fp16")]; + bool var_44409_interleave_0 = const()[name = string("op_44409_interleave_0"), val = bool(false)]; + tensor var_44409_cast_fp16 = concat(axis = var_42934, interleave = var_44409_interleave_0, values = (var_44351_cast_fp16, var_44353_cast_fp16, var_44355_cast_fp16, var_44357_cast_fp16))[name = string("op_44409_cast_fp16")]; + bool var_44411_interleave_0 = const()[name = string("op_44411_interleave_0"), val = bool(false)]; + tensor var_44411_cast_fp16 = concat(axis = var_42934, interleave = var_44411_interleave_0, values = (var_44359_cast_fp16, var_44361_cast_fp16, var_44363_cast_fp16, var_44365_cast_fp16))[name = string("op_44411_cast_fp16")]; + bool var_44413_interleave_0 = const()[name = string("op_44413_interleave_0"), val = bool(false)]; + tensor var_44413_cast_fp16 = concat(axis = var_42934, interleave = var_44413_interleave_0, values = (var_44367_cast_fp16, var_44369_cast_fp16, var_44371_cast_fp16, var_44373_cast_fp16))[name = string("op_44413_cast_fp16")]; + bool input_225_interleave_0 = const()[name = string("input_225_interleave_0"), val = bool(false)]; + tensor input_225_cast_fp16 = concat(axis = var_42959, interleave = input_225_interleave_0, values = (var_44375_cast_fp16, var_44377_cast_fp16, var_44379_cast_fp16, var_44381_cast_fp16, var_44383_cast_fp16, var_44385_cast_fp16, var_44387_cast_fp16, var_44389_cast_fp16, var_44391_cast_fp16, var_44393_cast_fp16, var_44395_cast_fp16, var_44397_cast_fp16, var_44399_cast_fp16, var_44401_cast_fp16, var_44403_cast_fp16, var_44405_cast_fp16, var_44407_cast_fp16, var_44409_cast_fp16, var_44411_cast_fp16, var_44413_cast_fp16))[name = string("input_225_cast_fp16")]; + string obj_115_pad_type_0 = const()[name = string("obj_115_pad_type_0"), val = string("valid")]; + tensor obj_115_strides_0 = const()[name = string("obj_115_strides_0"), val = tensor([1, 1])]; + tensor obj_115_pad_0 = const()[name = string("obj_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_115_dilations_0 = const()[name = string("obj_115_dilations_0"), val = tensor([1, 1])]; + int32 obj_115_groups_0 = const()[name = string("obj_115_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126397120)))]; + tensor layers_28_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129673984)))]; + tensor obj_115_cast_fp16 = conv(bias = layers_28_self_attn_o_proj_bias_to_fp16, dilations = obj_115_dilations_0, groups = obj_115_groups_0, pad = obj_115_pad_0, pad_type = obj_115_pad_type_0, strides = obj_115_strides_0, weight = layers_28_self_attn_o_proj_weight_to_fp16, x = input_225_cast_fp16)[name = string("obj_115_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_115_cast_fp16)[name = string("inputs_115_cast_fp16")]; + tensor out_115_axes_0 = const()[name = string("out_115_axes_0"), val = tensor([1])]; + fp16 var_44432_to_fp16 = const()[name = string("op_44432_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_44432_to_fp16, x = inputs_115_cast_fp16)[name = string("out_115_cast_fp16")]; + tensor input_227_gamma_0_to_fp16 = const()[name = string("input_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129676608)))]; + tensor input_227_beta_0_to_fp16 = const()[name = string("input_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129679232)))]; + fp16 input_227_epsilon_0_to_fp16 = const()[name = string("input_227_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_227_cast_fp16 = batch_norm(beta = input_227_beta_0_to_fp16, epsilon = input_227_epsilon_0_to_fp16, gamma = input_227_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = string("input_227_cast_fp16")]; + string input_229_pad_type_0 = const()[name = string("input_229_pad_type_0"), val = string("valid")]; + tensor input_229_strides_0 = const()[name = string("input_229_strides_0"), val = tensor([1, 1])]; + tensor input_229_pad_0 = const()[name = string("input_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_229_dilations_0 = const()[name = string("input_229_dilations_0"), val = tensor([1, 1])]; + int32 input_229_groups_0 = const()[name = string("input_229_groups_0"), val = int32(1)]; + tensor layers_28_fc1_weight_to_fp16 = const()[name = string("layers_28_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129681856)))]; + tensor layers_28_fc1_bias_to_fp16 = const()[name = string("layers_28_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142789120)))]; + tensor input_229_cast_fp16 = conv(bias = layers_28_fc1_bias_to_fp16, dilations = input_229_dilations_0, groups = input_229_groups_0, pad = input_229_pad_0, pad_type = input_229_pad_type_0, strides = input_229_strides_0, weight = layers_28_fc1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")]; + string input_231_mode_0 = const()[name = string("input_231_mode_0"), val = string("EXACT")]; + tensor input_231_cast_fp16 = gelu(mode = input_231_mode_0, x = input_229_cast_fp16)[name = string("input_231_cast_fp16")]; + string hidden_states_61_pad_type_0 = const()[name = string("hidden_states_61_pad_type_0"), val = string("valid")]; + tensor hidden_states_61_strides_0 = const()[name = string("hidden_states_61_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_61_pad_0 = const()[name = string("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_61_dilations_0 = const()[name = string("hidden_states_61_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_61_groups_0 = const()[name = string("hidden_states_61_groups_0"), val = int32(1)]; + tensor layers_28_fc2_weight_to_fp16 = const()[name = string("layers_28_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142799424)))]; + tensor layers_28_fc2_bias_to_fp16 = const()[name = string("layers_28_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155906688)))]; + tensor hidden_states_61_cast_fp16 = conv(bias = layers_28_fc2_bias_to_fp16, dilations = hidden_states_61_dilations_0, groups = hidden_states_61_groups_0, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = hidden_states_61_strides_0, weight = layers_28_fc2_weight_to_fp16, x = input_231_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = hidden_states_61_cast_fp16)[name = string("inputs_117_cast_fp16")]; + int32 var_44461 = const()[name = string("op_44461"), val = int32(3)]; + int32 var_44486 = const()[name = string("op_44486"), val = int32(1)]; + tensor out_117_axes_0 = const()[name = string("out_117_axes_0"), val = tensor([1])]; + fp16 var_44503_to_fp16 = const()[name = string("op_44503_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_44503_to_fp16, x = inputs_117_cast_fp16)[name = string("out_117_cast_fp16")]; + tensor obj_117_gamma_0_to_fp16 = const()[name = string("obj_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155909312)))]; + tensor obj_117_beta_0_to_fp16 = const()[name = string("obj_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155911936)))]; + fp16 obj_117_epsilon_0_to_fp16 = const()[name = string("obj_117_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_117_cast_fp16 = batch_norm(beta = obj_117_beta_0_to_fp16, epsilon = obj_117_epsilon_0_to_fp16, gamma = obj_117_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = string("obj_117_cast_fp16")]; + string query_59_pad_type_0 = const()[name = string("query_59_pad_type_0"), val = string("valid")]; + tensor query_59_strides_0 = const()[name = string("query_59_strides_0"), val = tensor([1, 1])]; + tensor query_59_pad_0 = const()[name = string("query_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_59_dilations_0 = const()[name = string("query_59_dilations_0"), val = tensor([1, 1])]; + int32 query_59_groups_0 = const()[name = string("query_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155914560)))]; + tensor layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159191424)))]; + tensor query_59_cast_fp16 = conv(bias = layers_29_self_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_29_self_attn_q_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("query_59_cast_fp16")]; + string key_59_pad_type_0 = const()[name = string("key_59_pad_type_0"), val = string("valid")]; + tensor key_59_strides_0 = const()[name = string("key_59_strides_0"), val = tensor([1, 1])]; + tensor key_59_pad_0 = const()[name = string("key_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_59_dilations_0 = const()[name = string("key_59_dilations_0"), val = tensor([1, 1])]; + int32 key_59_groups_0 = const()[name = string("key_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159194048)))]; + tensor key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_29_self_attn_k_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("key_59_cast_fp16")]; + string value_59_pad_type_0 = const()[name = string("value_59_pad_type_0"), val = string("valid")]; + tensor value_59_strides_0 = const()[name = string("value_59_strides_0"), val = tensor([1, 1])]; + tensor value_59_pad_0 = const()[name = string("value_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_59_dilations_0 = const()[name = string("value_59_dilations_0"), val = tensor([1, 1])]; + int32 value_59_groups_0 = const()[name = string("value_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1162470912)))]; + tensor layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165747776)))]; + tensor value_59_cast_fp16 = conv(bias = layers_29_self_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_29_self_attn_v_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("value_59_cast_fp16")]; + tensor var_44541_begin_0 = const()[name = string("op_44541_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44541_end_0 = const()[name = string("op_44541_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44541_end_mask_0 = const()[name = string("op_44541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44541_cast_fp16 = slice_by_index(begin = var_44541_begin_0, end = var_44541_end_0, end_mask = var_44541_end_mask_0, x = query_59_cast_fp16)[name = string("op_44541_cast_fp16")]; + tensor var_44545_begin_0 = const()[name = string("op_44545_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44545_end_0 = const()[name = string("op_44545_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_44545_end_mask_0 = const()[name = string("op_44545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44545_cast_fp16 = slice_by_index(begin = var_44545_begin_0, end = var_44545_end_0, end_mask = var_44545_end_mask_0, x = query_59_cast_fp16)[name = string("op_44545_cast_fp16")]; + tensor var_44549_begin_0 = const()[name = string("op_44549_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44549_end_0 = const()[name = string("op_44549_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_44549_end_mask_0 = const()[name = string("op_44549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44549_cast_fp16 = slice_by_index(begin = var_44549_begin_0, end = var_44549_end_0, end_mask = var_44549_end_mask_0, x = query_59_cast_fp16)[name = string("op_44549_cast_fp16")]; + tensor var_44553_begin_0 = const()[name = string("op_44553_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44553_end_0 = const()[name = string("op_44553_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_44553_end_mask_0 = const()[name = string("op_44553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44553_cast_fp16 = slice_by_index(begin = var_44553_begin_0, end = var_44553_end_0, end_mask = var_44553_end_mask_0, x = query_59_cast_fp16)[name = string("op_44553_cast_fp16")]; + tensor var_44557_begin_0 = const()[name = string("op_44557_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44557_end_0 = const()[name = string("op_44557_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_44557_end_mask_0 = const()[name = string("op_44557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44557_cast_fp16 = slice_by_index(begin = var_44557_begin_0, end = var_44557_end_0, end_mask = var_44557_end_mask_0, x = query_59_cast_fp16)[name = string("op_44557_cast_fp16")]; + tensor var_44561_begin_0 = const()[name = string("op_44561_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44561_end_0 = const()[name = string("op_44561_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_44561_end_mask_0 = const()[name = string("op_44561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44561_cast_fp16 = slice_by_index(begin = var_44561_begin_0, end = var_44561_end_0, end_mask = var_44561_end_mask_0, x = query_59_cast_fp16)[name = string("op_44561_cast_fp16")]; + tensor var_44565_begin_0 = const()[name = string("op_44565_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44565_end_0 = const()[name = string("op_44565_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_44565_end_mask_0 = const()[name = string("op_44565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44565_cast_fp16 = slice_by_index(begin = var_44565_begin_0, end = var_44565_end_0, end_mask = var_44565_end_mask_0, x = query_59_cast_fp16)[name = string("op_44565_cast_fp16")]; + tensor var_44569_begin_0 = const()[name = string("op_44569_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44569_end_0 = const()[name = string("op_44569_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_44569_end_mask_0 = const()[name = string("op_44569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44569_cast_fp16 = slice_by_index(begin = var_44569_begin_0, end = var_44569_end_0, end_mask = var_44569_end_mask_0, x = query_59_cast_fp16)[name = string("op_44569_cast_fp16")]; + tensor var_44573_begin_0 = const()[name = string("op_44573_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44573_end_0 = const()[name = string("op_44573_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_44573_end_mask_0 = const()[name = string("op_44573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44573_cast_fp16 = slice_by_index(begin = var_44573_begin_0, end = var_44573_end_0, end_mask = var_44573_end_mask_0, x = query_59_cast_fp16)[name = string("op_44573_cast_fp16")]; + tensor var_44577_begin_0 = const()[name = string("op_44577_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44577_end_0 = const()[name = string("op_44577_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_44577_end_mask_0 = const()[name = string("op_44577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44577_cast_fp16 = slice_by_index(begin = var_44577_begin_0, end = var_44577_end_0, end_mask = var_44577_end_mask_0, x = query_59_cast_fp16)[name = string("op_44577_cast_fp16")]; + tensor var_44581_begin_0 = const()[name = string("op_44581_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44581_end_0 = const()[name = string("op_44581_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_44581_end_mask_0 = const()[name = string("op_44581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44581_cast_fp16 = slice_by_index(begin = var_44581_begin_0, end = var_44581_end_0, end_mask = var_44581_end_mask_0, x = query_59_cast_fp16)[name = string("op_44581_cast_fp16")]; + tensor var_44585_begin_0 = const()[name = string("op_44585_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44585_end_0 = const()[name = string("op_44585_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_44585_end_mask_0 = const()[name = string("op_44585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44585_cast_fp16 = slice_by_index(begin = var_44585_begin_0, end = var_44585_end_0, end_mask = var_44585_end_mask_0, x = query_59_cast_fp16)[name = string("op_44585_cast_fp16")]; + tensor var_44589_begin_0 = const()[name = string("op_44589_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44589_end_0 = const()[name = string("op_44589_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_44589_end_mask_0 = const()[name = string("op_44589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44589_cast_fp16 = slice_by_index(begin = var_44589_begin_0, end = var_44589_end_0, end_mask = var_44589_end_mask_0, x = query_59_cast_fp16)[name = string("op_44589_cast_fp16")]; + tensor var_44593_begin_0 = const()[name = string("op_44593_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44593_end_0 = const()[name = string("op_44593_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_44593_end_mask_0 = const()[name = string("op_44593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44593_cast_fp16 = slice_by_index(begin = var_44593_begin_0, end = var_44593_end_0, end_mask = var_44593_end_mask_0, x = query_59_cast_fp16)[name = string("op_44593_cast_fp16")]; + tensor var_44597_begin_0 = const()[name = string("op_44597_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44597_end_0 = const()[name = string("op_44597_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_44597_end_mask_0 = const()[name = string("op_44597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44597_cast_fp16 = slice_by_index(begin = var_44597_begin_0, end = var_44597_end_0, end_mask = var_44597_end_mask_0, x = query_59_cast_fp16)[name = string("op_44597_cast_fp16")]; + tensor var_44601_begin_0 = const()[name = string("op_44601_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44601_end_0 = const()[name = string("op_44601_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_44601_end_mask_0 = const()[name = string("op_44601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44601_cast_fp16 = slice_by_index(begin = var_44601_begin_0, end = var_44601_end_0, end_mask = var_44601_end_mask_0, x = query_59_cast_fp16)[name = string("op_44601_cast_fp16")]; + tensor var_44605_begin_0 = const()[name = string("op_44605_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44605_end_0 = const()[name = string("op_44605_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_44605_end_mask_0 = const()[name = string("op_44605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44605_cast_fp16 = slice_by_index(begin = var_44605_begin_0, end = var_44605_end_0, end_mask = var_44605_end_mask_0, x = query_59_cast_fp16)[name = string("op_44605_cast_fp16")]; + tensor var_44609_begin_0 = const()[name = string("op_44609_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44609_end_0 = const()[name = string("op_44609_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_44609_end_mask_0 = const()[name = string("op_44609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44609_cast_fp16 = slice_by_index(begin = var_44609_begin_0, end = var_44609_end_0, end_mask = var_44609_end_mask_0, x = query_59_cast_fp16)[name = string("op_44609_cast_fp16")]; + tensor var_44613_begin_0 = const()[name = string("op_44613_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44613_end_0 = const()[name = string("op_44613_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_44613_end_mask_0 = const()[name = string("op_44613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44613_cast_fp16 = slice_by_index(begin = var_44613_begin_0, end = var_44613_end_0, end_mask = var_44613_end_mask_0, x = query_59_cast_fp16)[name = string("op_44613_cast_fp16")]; + tensor var_44617_begin_0 = const()[name = string("op_44617_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44617_end_0 = const()[name = string("op_44617_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_44617_end_mask_0 = const()[name = string("op_44617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44617_cast_fp16 = slice_by_index(begin = var_44617_begin_0, end = var_44617_end_0, end_mask = var_44617_end_mask_0, x = query_59_cast_fp16)[name = string("op_44617_cast_fp16")]; + tensor var_44626_begin_0 = const()[name = string("op_44626_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44626_end_0 = const()[name = string("op_44626_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44626_end_mask_0 = const()[name = string("op_44626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44626_cast_fp16 = slice_by_index(begin = var_44626_begin_0, end = var_44626_end_0, end_mask = var_44626_end_mask_0, x = var_44541_cast_fp16)[name = string("op_44626_cast_fp16")]; + tensor var_44633_begin_0 = const()[name = string("op_44633_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44633_end_0 = const()[name = string("op_44633_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44633_end_mask_0 = const()[name = string("op_44633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44633_cast_fp16 = slice_by_index(begin = var_44633_begin_0, end = var_44633_end_0, end_mask = var_44633_end_mask_0, x = var_44541_cast_fp16)[name = string("op_44633_cast_fp16")]; + tensor var_44640_begin_0 = const()[name = string("op_44640_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44640_end_0 = const()[name = string("op_44640_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44640_end_mask_0 = const()[name = string("op_44640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44640_cast_fp16 = slice_by_index(begin = var_44640_begin_0, end = var_44640_end_0, end_mask = var_44640_end_mask_0, x = var_44541_cast_fp16)[name = string("op_44640_cast_fp16")]; + tensor var_44647_begin_0 = const()[name = string("op_44647_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44647_end_0 = const()[name = string("op_44647_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44647_end_mask_0 = const()[name = string("op_44647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44647_cast_fp16 = slice_by_index(begin = var_44647_begin_0, end = var_44647_end_0, end_mask = var_44647_end_mask_0, x = var_44541_cast_fp16)[name = string("op_44647_cast_fp16")]; + tensor var_44654_begin_0 = const()[name = string("op_44654_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44654_end_0 = const()[name = string("op_44654_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44654_end_mask_0 = const()[name = string("op_44654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44654_cast_fp16 = slice_by_index(begin = var_44654_begin_0, end = var_44654_end_0, end_mask = var_44654_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44654_cast_fp16")]; + tensor var_44661_begin_0 = const()[name = string("op_44661_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44661_end_0 = const()[name = string("op_44661_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44661_end_mask_0 = const()[name = string("op_44661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44661_cast_fp16 = slice_by_index(begin = var_44661_begin_0, end = var_44661_end_0, end_mask = var_44661_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44661_cast_fp16")]; + tensor var_44668_begin_0 = const()[name = string("op_44668_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44668_end_0 = const()[name = string("op_44668_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44668_end_mask_0 = const()[name = string("op_44668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44668_cast_fp16 = slice_by_index(begin = var_44668_begin_0, end = var_44668_end_0, end_mask = var_44668_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44668_cast_fp16")]; + tensor var_44675_begin_0 = const()[name = string("op_44675_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44675_end_0 = const()[name = string("op_44675_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44675_end_mask_0 = const()[name = string("op_44675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44675_cast_fp16 = slice_by_index(begin = var_44675_begin_0, end = var_44675_end_0, end_mask = var_44675_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44675_cast_fp16")]; + tensor var_44682_begin_0 = const()[name = string("op_44682_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44682_end_0 = const()[name = string("op_44682_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44682_end_mask_0 = const()[name = string("op_44682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44682_cast_fp16 = slice_by_index(begin = var_44682_begin_0, end = var_44682_end_0, end_mask = var_44682_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44682_cast_fp16")]; + tensor var_44689_begin_0 = const()[name = string("op_44689_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44689_end_0 = const()[name = string("op_44689_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44689_end_mask_0 = const()[name = string("op_44689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44689_cast_fp16 = slice_by_index(begin = var_44689_begin_0, end = var_44689_end_0, end_mask = var_44689_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44689_cast_fp16")]; + tensor var_44696_begin_0 = const()[name = string("op_44696_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44696_end_0 = const()[name = string("op_44696_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44696_end_mask_0 = const()[name = string("op_44696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44696_cast_fp16 = slice_by_index(begin = var_44696_begin_0, end = var_44696_end_0, end_mask = var_44696_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44696_cast_fp16")]; + tensor var_44703_begin_0 = const()[name = string("op_44703_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44703_end_0 = const()[name = string("op_44703_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44703_end_mask_0 = const()[name = string("op_44703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44703_cast_fp16 = slice_by_index(begin = var_44703_begin_0, end = var_44703_end_0, end_mask = var_44703_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44703_cast_fp16")]; + tensor var_44710_begin_0 = const()[name = string("op_44710_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44710_end_0 = const()[name = string("op_44710_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44710_end_mask_0 = const()[name = string("op_44710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44710_cast_fp16 = slice_by_index(begin = var_44710_begin_0, end = var_44710_end_0, end_mask = var_44710_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44710_cast_fp16")]; + tensor var_44717_begin_0 = const()[name = string("op_44717_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44717_end_0 = const()[name = string("op_44717_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44717_end_mask_0 = const()[name = string("op_44717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44717_cast_fp16 = slice_by_index(begin = var_44717_begin_0, end = var_44717_end_0, end_mask = var_44717_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44717_cast_fp16")]; + tensor var_44724_begin_0 = const()[name = string("op_44724_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44724_end_0 = const()[name = string("op_44724_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44724_end_mask_0 = const()[name = string("op_44724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44724_cast_fp16 = slice_by_index(begin = var_44724_begin_0, end = var_44724_end_0, end_mask = var_44724_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44724_cast_fp16")]; + tensor var_44731_begin_0 = const()[name = string("op_44731_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44731_end_0 = const()[name = string("op_44731_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44731_end_mask_0 = const()[name = string("op_44731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44731_cast_fp16 = slice_by_index(begin = var_44731_begin_0, end = var_44731_end_0, end_mask = var_44731_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44731_cast_fp16")]; + tensor var_44738_begin_0 = const()[name = string("op_44738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44738_end_0 = const()[name = string("op_44738_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44738_end_mask_0 = const()[name = string("op_44738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44738_cast_fp16 = slice_by_index(begin = var_44738_begin_0, end = var_44738_end_0, end_mask = var_44738_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44738_cast_fp16")]; + tensor var_44745_begin_0 = const()[name = string("op_44745_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44745_end_0 = const()[name = string("op_44745_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44745_end_mask_0 = const()[name = string("op_44745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44745_cast_fp16 = slice_by_index(begin = var_44745_begin_0, end = var_44745_end_0, end_mask = var_44745_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44745_cast_fp16")]; + tensor var_44752_begin_0 = const()[name = string("op_44752_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44752_end_0 = const()[name = string("op_44752_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44752_end_mask_0 = const()[name = string("op_44752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44752_cast_fp16 = slice_by_index(begin = var_44752_begin_0, end = var_44752_end_0, end_mask = var_44752_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44752_cast_fp16")]; + tensor var_44759_begin_0 = const()[name = string("op_44759_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44759_end_0 = const()[name = string("op_44759_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44759_end_mask_0 = const()[name = string("op_44759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44759_cast_fp16 = slice_by_index(begin = var_44759_begin_0, end = var_44759_end_0, end_mask = var_44759_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44759_cast_fp16")]; + tensor var_44766_begin_0 = const()[name = string("op_44766_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44766_end_0 = const()[name = string("op_44766_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44766_end_mask_0 = const()[name = string("op_44766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44766_cast_fp16 = slice_by_index(begin = var_44766_begin_0, end = var_44766_end_0, end_mask = var_44766_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44766_cast_fp16")]; + tensor var_44773_begin_0 = const()[name = string("op_44773_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44773_end_0 = const()[name = string("op_44773_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44773_end_mask_0 = const()[name = string("op_44773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44773_cast_fp16 = slice_by_index(begin = var_44773_begin_0, end = var_44773_end_0, end_mask = var_44773_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44773_cast_fp16")]; + tensor var_44780_begin_0 = const()[name = string("op_44780_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44780_end_0 = const()[name = string("op_44780_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44780_end_mask_0 = const()[name = string("op_44780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44780_cast_fp16 = slice_by_index(begin = var_44780_begin_0, end = var_44780_end_0, end_mask = var_44780_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44780_cast_fp16")]; + tensor var_44787_begin_0 = const()[name = string("op_44787_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44787_end_0 = const()[name = string("op_44787_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44787_end_mask_0 = const()[name = string("op_44787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44787_cast_fp16 = slice_by_index(begin = var_44787_begin_0, end = var_44787_end_0, end_mask = var_44787_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44787_cast_fp16")]; + tensor var_44794_begin_0 = const()[name = string("op_44794_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44794_end_0 = const()[name = string("op_44794_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44794_end_mask_0 = const()[name = string("op_44794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44794_cast_fp16 = slice_by_index(begin = var_44794_begin_0, end = var_44794_end_0, end_mask = var_44794_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44794_cast_fp16")]; + tensor var_44801_begin_0 = const()[name = string("op_44801_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44801_end_0 = const()[name = string("op_44801_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44801_end_mask_0 = const()[name = string("op_44801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44801_cast_fp16 = slice_by_index(begin = var_44801_begin_0, end = var_44801_end_0, end_mask = var_44801_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44801_cast_fp16")]; + tensor var_44808_begin_0 = const()[name = string("op_44808_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44808_end_0 = const()[name = string("op_44808_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44808_end_mask_0 = const()[name = string("op_44808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44808_cast_fp16 = slice_by_index(begin = var_44808_begin_0, end = var_44808_end_0, end_mask = var_44808_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44808_cast_fp16")]; + tensor var_44815_begin_0 = const()[name = string("op_44815_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44815_end_0 = const()[name = string("op_44815_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44815_end_mask_0 = const()[name = string("op_44815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44815_cast_fp16 = slice_by_index(begin = var_44815_begin_0, end = var_44815_end_0, end_mask = var_44815_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44815_cast_fp16")]; + tensor var_44822_begin_0 = const()[name = string("op_44822_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44822_end_0 = const()[name = string("op_44822_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44822_end_mask_0 = const()[name = string("op_44822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44822_cast_fp16 = slice_by_index(begin = var_44822_begin_0, end = var_44822_end_0, end_mask = var_44822_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44822_cast_fp16")]; + tensor var_44829_begin_0 = const()[name = string("op_44829_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44829_end_0 = const()[name = string("op_44829_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44829_end_mask_0 = const()[name = string("op_44829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44829_cast_fp16 = slice_by_index(begin = var_44829_begin_0, end = var_44829_end_0, end_mask = var_44829_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44829_cast_fp16")]; + tensor var_44836_begin_0 = const()[name = string("op_44836_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44836_end_0 = const()[name = string("op_44836_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44836_end_mask_0 = const()[name = string("op_44836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44836_cast_fp16 = slice_by_index(begin = var_44836_begin_0, end = var_44836_end_0, end_mask = var_44836_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44836_cast_fp16")]; + tensor var_44843_begin_0 = const()[name = string("op_44843_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44843_end_0 = const()[name = string("op_44843_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44843_end_mask_0 = const()[name = string("op_44843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44843_cast_fp16 = slice_by_index(begin = var_44843_begin_0, end = var_44843_end_0, end_mask = var_44843_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44843_cast_fp16")]; + tensor var_44850_begin_0 = const()[name = string("op_44850_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44850_end_0 = const()[name = string("op_44850_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44850_end_mask_0 = const()[name = string("op_44850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44850_cast_fp16 = slice_by_index(begin = var_44850_begin_0, end = var_44850_end_0, end_mask = var_44850_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44850_cast_fp16")]; + tensor var_44857_begin_0 = const()[name = string("op_44857_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44857_end_0 = const()[name = string("op_44857_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44857_end_mask_0 = const()[name = string("op_44857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44857_cast_fp16 = slice_by_index(begin = var_44857_begin_0, end = var_44857_end_0, end_mask = var_44857_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44857_cast_fp16")]; + tensor var_44864_begin_0 = const()[name = string("op_44864_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44864_end_0 = const()[name = string("op_44864_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44864_end_mask_0 = const()[name = string("op_44864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44864_cast_fp16 = slice_by_index(begin = var_44864_begin_0, end = var_44864_end_0, end_mask = var_44864_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44864_cast_fp16")]; + tensor var_44871_begin_0 = const()[name = string("op_44871_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44871_end_0 = const()[name = string("op_44871_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44871_end_mask_0 = const()[name = string("op_44871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44871_cast_fp16 = slice_by_index(begin = var_44871_begin_0, end = var_44871_end_0, end_mask = var_44871_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44871_cast_fp16")]; + tensor var_44878_begin_0 = const()[name = string("op_44878_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44878_end_0 = const()[name = string("op_44878_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44878_end_mask_0 = const()[name = string("op_44878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44878_cast_fp16 = slice_by_index(begin = var_44878_begin_0, end = var_44878_end_0, end_mask = var_44878_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44878_cast_fp16")]; + tensor var_44885_begin_0 = const()[name = string("op_44885_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44885_end_0 = const()[name = string("op_44885_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44885_end_mask_0 = const()[name = string("op_44885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44885_cast_fp16 = slice_by_index(begin = var_44885_begin_0, end = var_44885_end_0, end_mask = var_44885_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44885_cast_fp16")]; + tensor var_44892_begin_0 = const()[name = string("op_44892_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44892_end_0 = const()[name = string("op_44892_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44892_end_mask_0 = const()[name = string("op_44892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44892_cast_fp16 = slice_by_index(begin = var_44892_begin_0, end = var_44892_end_0, end_mask = var_44892_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44892_cast_fp16")]; + tensor var_44899_begin_0 = const()[name = string("op_44899_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44899_end_0 = const()[name = string("op_44899_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44899_end_mask_0 = const()[name = string("op_44899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44899_cast_fp16 = slice_by_index(begin = var_44899_begin_0, end = var_44899_end_0, end_mask = var_44899_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44899_cast_fp16")]; + tensor var_44906_begin_0 = const()[name = string("op_44906_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44906_end_0 = const()[name = string("op_44906_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44906_end_mask_0 = const()[name = string("op_44906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44906_cast_fp16 = slice_by_index(begin = var_44906_begin_0, end = var_44906_end_0, end_mask = var_44906_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44906_cast_fp16")]; + tensor var_44913_begin_0 = const()[name = string("op_44913_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44913_end_0 = const()[name = string("op_44913_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44913_end_mask_0 = const()[name = string("op_44913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44913_cast_fp16 = slice_by_index(begin = var_44913_begin_0, end = var_44913_end_0, end_mask = var_44913_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44913_cast_fp16")]; + tensor var_44920_begin_0 = const()[name = string("op_44920_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44920_end_0 = const()[name = string("op_44920_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44920_end_mask_0 = const()[name = string("op_44920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44920_cast_fp16 = slice_by_index(begin = var_44920_begin_0, end = var_44920_end_0, end_mask = var_44920_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44920_cast_fp16")]; + tensor var_44927_begin_0 = const()[name = string("op_44927_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44927_end_0 = const()[name = string("op_44927_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44927_end_mask_0 = const()[name = string("op_44927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44927_cast_fp16 = slice_by_index(begin = var_44927_begin_0, end = var_44927_end_0, end_mask = var_44927_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44927_cast_fp16")]; + tensor var_44934_begin_0 = const()[name = string("op_44934_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44934_end_0 = const()[name = string("op_44934_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44934_end_mask_0 = const()[name = string("op_44934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44934_cast_fp16 = slice_by_index(begin = var_44934_begin_0, end = var_44934_end_0, end_mask = var_44934_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44934_cast_fp16")]; + tensor var_44941_begin_0 = const()[name = string("op_44941_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44941_end_0 = const()[name = string("op_44941_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44941_end_mask_0 = const()[name = string("op_44941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44941_cast_fp16 = slice_by_index(begin = var_44941_begin_0, end = var_44941_end_0, end_mask = var_44941_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44941_cast_fp16")]; + tensor var_44948_begin_0 = const()[name = string("op_44948_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44948_end_0 = const()[name = string("op_44948_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44948_end_mask_0 = const()[name = string("op_44948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44948_cast_fp16 = slice_by_index(begin = var_44948_begin_0, end = var_44948_end_0, end_mask = var_44948_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44948_cast_fp16")]; + tensor var_44955_begin_0 = const()[name = string("op_44955_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44955_end_0 = const()[name = string("op_44955_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44955_end_mask_0 = const()[name = string("op_44955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44955_cast_fp16 = slice_by_index(begin = var_44955_begin_0, end = var_44955_end_0, end_mask = var_44955_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44955_cast_fp16")]; + tensor var_44962_begin_0 = const()[name = string("op_44962_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44962_end_0 = const()[name = string("op_44962_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44962_end_mask_0 = const()[name = string("op_44962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44962_cast_fp16 = slice_by_index(begin = var_44962_begin_0, end = var_44962_end_0, end_mask = var_44962_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44962_cast_fp16")]; + tensor var_44969_begin_0 = const()[name = string("op_44969_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44969_end_0 = const()[name = string("op_44969_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44969_end_mask_0 = const()[name = string("op_44969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44969_cast_fp16 = slice_by_index(begin = var_44969_begin_0, end = var_44969_end_0, end_mask = var_44969_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44969_cast_fp16")]; + tensor var_44976_begin_0 = const()[name = string("op_44976_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44976_end_0 = const()[name = string("op_44976_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44976_end_mask_0 = const()[name = string("op_44976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44976_cast_fp16 = slice_by_index(begin = var_44976_begin_0, end = var_44976_end_0, end_mask = var_44976_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44976_cast_fp16")]; + tensor var_44983_begin_0 = const()[name = string("op_44983_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44983_end_0 = const()[name = string("op_44983_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44983_end_mask_0 = const()[name = string("op_44983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44983_cast_fp16 = slice_by_index(begin = var_44983_begin_0, end = var_44983_end_0, end_mask = var_44983_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44983_cast_fp16")]; + tensor var_44990_begin_0 = const()[name = string("op_44990_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44990_end_0 = const()[name = string("op_44990_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44990_end_mask_0 = const()[name = string("op_44990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44990_cast_fp16 = slice_by_index(begin = var_44990_begin_0, end = var_44990_end_0, end_mask = var_44990_end_mask_0, x = var_44593_cast_fp16)[name = string("op_44990_cast_fp16")]; + tensor var_44997_begin_0 = const()[name = string("op_44997_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44997_end_0 = const()[name = string("op_44997_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44997_end_mask_0 = const()[name = string("op_44997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44997_cast_fp16 = slice_by_index(begin = var_44997_begin_0, end = var_44997_end_0, end_mask = var_44997_end_mask_0, x = var_44593_cast_fp16)[name = string("op_44997_cast_fp16")]; + tensor var_45004_begin_0 = const()[name = string("op_45004_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45004_end_0 = const()[name = string("op_45004_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45004_end_mask_0 = const()[name = string("op_45004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45004_cast_fp16 = slice_by_index(begin = var_45004_begin_0, end = var_45004_end_0, end_mask = var_45004_end_mask_0, x = var_44593_cast_fp16)[name = string("op_45004_cast_fp16")]; + tensor var_45011_begin_0 = const()[name = string("op_45011_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45011_end_0 = const()[name = string("op_45011_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45011_end_mask_0 = const()[name = string("op_45011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45011_cast_fp16 = slice_by_index(begin = var_45011_begin_0, end = var_45011_end_0, end_mask = var_45011_end_mask_0, x = var_44593_cast_fp16)[name = string("op_45011_cast_fp16")]; + tensor var_45018_begin_0 = const()[name = string("op_45018_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45018_end_0 = const()[name = string("op_45018_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45018_end_mask_0 = const()[name = string("op_45018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45018_cast_fp16 = slice_by_index(begin = var_45018_begin_0, end = var_45018_end_0, end_mask = var_45018_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45018_cast_fp16")]; + tensor var_45025_begin_0 = const()[name = string("op_45025_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45025_end_0 = const()[name = string("op_45025_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45025_end_mask_0 = const()[name = string("op_45025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45025_cast_fp16 = slice_by_index(begin = var_45025_begin_0, end = var_45025_end_0, end_mask = var_45025_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45025_cast_fp16")]; + tensor var_45032_begin_0 = const()[name = string("op_45032_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45032_end_0 = const()[name = string("op_45032_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45032_end_mask_0 = const()[name = string("op_45032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45032_cast_fp16 = slice_by_index(begin = var_45032_begin_0, end = var_45032_end_0, end_mask = var_45032_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45032_cast_fp16")]; + tensor var_45039_begin_0 = const()[name = string("op_45039_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45039_end_0 = const()[name = string("op_45039_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45039_end_mask_0 = const()[name = string("op_45039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45039_cast_fp16 = slice_by_index(begin = var_45039_begin_0, end = var_45039_end_0, end_mask = var_45039_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45039_cast_fp16")]; + tensor var_45046_begin_0 = const()[name = string("op_45046_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45046_end_0 = const()[name = string("op_45046_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45046_end_mask_0 = const()[name = string("op_45046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45046_cast_fp16 = slice_by_index(begin = var_45046_begin_0, end = var_45046_end_0, end_mask = var_45046_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45046_cast_fp16")]; + tensor var_45053_begin_0 = const()[name = string("op_45053_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45053_end_0 = const()[name = string("op_45053_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45053_end_mask_0 = const()[name = string("op_45053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45053_cast_fp16 = slice_by_index(begin = var_45053_begin_0, end = var_45053_end_0, end_mask = var_45053_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45053_cast_fp16")]; + tensor var_45060_begin_0 = const()[name = string("op_45060_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45060_end_0 = const()[name = string("op_45060_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45060_end_mask_0 = const()[name = string("op_45060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45060_cast_fp16 = slice_by_index(begin = var_45060_begin_0, end = var_45060_end_0, end_mask = var_45060_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45060_cast_fp16")]; + tensor var_45067_begin_0 = const()[name = string("op_45067_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45067_end_0 = const()[name = string("op_45067_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45067_end_mask_0 = const()[name = string("op_45067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45067_cast_fp16 = slice_by_index(begin = var_45067_begin_0, end = var_45067_end_0, end_mask = var_45067_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45067_cast_fp16")]; + tensor var_45074_begin_0 = const()[name = string("op_45074_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45074_end_0 = const()[name = string("op_45074_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45074_end_mask_0 = const()[name = string("op_45074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45074_cast_fp16 = slice_by_index(begin = var_45074_begin_0, end = var_45074_end_0, end_mask = var_45074_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45074_cast_fp16")]; + tensor var_45081_begin_0 = const()[name = string("op_45081_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45081_end_0 = const()[name = string("op_45081_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45081_end_mask_0 = const()[name = string("op_45081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45081_cast_fp16 = slice_by_index(begin = var_45081_begin_0, end = var_45081_end_0, end_mask = var_45081_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45081_cast_fp16")]; + tensor var_45088_begin_0 = const()[name = string("op_45088_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45088_end_0 = const()[name = string("op_45088_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45088_end_mask_0 = const()[name = string("op_45088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45088_cast_fp16 = slice_by_index(begin = var_45088_begin_0, end = var_45088_end_0, end_mask = var_45088_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45088_cast_fp16")]; + tensor var_45095_begin_0 = const()[name = string("op_45095_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45095_end_0 = const()[name = string("op_45095_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45095_end_mask_0 = const()[name = string("op_45095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45095_cast_fp16 = slice_by_index(begin = var_45095_begin_0, end = var_45095_end_0, end_mask = var_45095_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45095_cast_fp16")]; + tensor var_45102_begin_0 = const()[name = string("op_45102_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45102_end_0 = const()[name = string("op_45102_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45102_end_mask_0 = const()[name = string("op_45102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45102_cast_fp16 = slice_by_index(begin = var_45102_begin_0, end = var_45102_end_0, end_mask = var_45102_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45102_cast_fp16")]; + tensor var_45109_begin_0 = const()[name = string("op_45109_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45109_end_0 = const()[name = string("op_45109_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45109_end_mask_0 = const()[name = string("op_45109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45109_cast_fp16 = slice_by_index(begin = var_45109_begin_0, end = var_45109_end_0, end_mask = var_45109_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45109_cast_fp16")]; + tensor var_45116_begin_0 = const()[name = string("op_45116_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45116_end_0 = const()[name = string("op_45116_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45116_end_mask_0 = const()[name = string("op_45116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45116_cast_fp16 = slice_by_index(begin = var_45116_begin_0, end = var_45116_end_0, end_mask = var_45116_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45116_cast_fp16")]; + tensor var_45123_begin_0 = const()[name = string("op_45123_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45123_end_0 = const()[name = string("op_45123_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45123_end_mask_0 = const()[name = string("op_45123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45123_cast_fp16 = slice_by_index(begin = var_45123_begin_0, end = var_45123_end_0, end_mask = var_45123_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45123_cast_fp16")]; + tensor var_45130_begin_0 = const()[name = string("op_45130_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45130_end_0 = const()[name = string("op_45130_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45130_end_mask_0 = const()[name = string("op_45130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45130_cast_fp16 = slice_by_index(begin = var_45130_begin_0, end = var_45130_end_0, end_mask = var_45130_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45130_cast_fp16")]; + tensor var_45137_begin_0 = const()[name = string("op_45137_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45137_end_0 = const()[name = string("op_45137_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45137_end_mask_0 = const()[name = string("op_45137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45137_cast_fp16 = slice_by_index(begin = var_45137_begin_0, end = var_45137_end_0, end_mask = var_45137_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45137_cast_fp16")]; + tensor var_45144_begin_0 = const()[name = string("op_45144_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45144_end_0 = const()[name = string("op_45144_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45144_end_mask_0 = const()[name = string("op_45144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45144_cast_fp16 = slice_by_index(begin = var_45144_begin_0, end = var_45144_end_0, end_mask = var_45144_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45144_cast_fp16")]; + tensor var_45151_begin_0 = const()[name = string("op_45151_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45151_end_0 = const()[name = string("op_45151_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45151_end_mask_0 = const()[name = string("op_45151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45151_cast_fp16 = slice_by_index(begin = var_45151_begin_0, end = var_45151_end_0, end_mask = var_45151_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45151_cast_fp16")]; + tensor var_45158_begin_0 = const()[name = string("op_45158_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45158_end_0 = const()[name = string("op_45158_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45158_end_mask_0 = const()[name = string("op_45158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45158_cast_fp16 = slice_by_index(begin = var_45158_begin_0, end = var_45158_end_0, end_mask = var_45158_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45158_cast_fp16")]; + tensor var_45165_begin_0 = const()[name = string("op_45165_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45165_end_0 = const()[name = string("op_45165_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45165_end_mask_0 = const()[name = string("op_45165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45165_cast_fp16 = slice_by_index(begin = var_45165_begin_0, end = var_45165_end_0, end_mask = var_45165_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45165_cast_fp16")]; + tensor var_45172_begin_0 = const()[name = string("op_45172_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45172_end_0 = const()[name = string("op_45172_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45172_end_mask_0 = const()[name = string("op_45172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45172_cast_fp16 = slice_by_index(begin = var_45172_begin_0, end = var_45172_end_0, end_mask = var_45172_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45172_cast_fp16")]; + tensor var_45179_begin_0 = const()[name = string("op_45179_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45179_end_0 = const()[name = string("op_45179_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45179_end_mask_0 = const()[name = string("op_45179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45179_cast_fp16 = slice_by_index(begin = var_45179_begin_0, end = var_45179_end_0, end_mask = var_45179_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45179_cast_fp16")]; + tensor k_59_perm_0 = const()[name = string("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_45184_begin_0 = const()[name = string("op_45184_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45184_end_0 = const()[name = string("op_45184_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_45184_end_mask_0 = const()[name = string("op_45184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_59_cast_fp16 = transpose(perm = k_59_perm_0, x = key_59_cast_fp16)[name = string("transpose_2")]; + tensor var_45184_cast_fp16 = slice_by_index(begin = var_45184_begin_0, end = var_45184_end_0, end_mask = var_45184_end_mask_0, x = k_59_cast_fp16)[name = string("op_45184_cast_fp16")]; + tensor var_45188_begin_0 = const()[name = string("op_45188_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_45188_end_0 = const()[name = string("op_45188_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_45188_end_mask_0 = const()[name = string("op_45188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45188_cast_fp16 = slice_by_index(begin = var_45188_begin_0, end = var_45188_end_0, end_mask = var_45188_end_mask_0, x = k_59_cast_fp16)[name = string("op_45188_cast_fp16")]; + tensor var_45192_begin_0 = const()[name = string("op_45192_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_45192_end_0 = const()[name = string("op_45192_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_45192_end_mask_0 = const()[name = string("op_45192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45192_cast_fp16 = slice_by_index(begin = var_45192_begin_0, end = var_45192_end_0, end_mask = var_45192_end_mask_0, x = k_59_cast_fp16)[name = string("op_45192_cast_fp16")]; + tensor var_45196_begin_0 = const()[name = string("op_45196_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_45196_end_0 = const()[name = string("op_45196_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_45196_end_mask_0 = const()[name = string("op_45196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45196_cast_fp16 = slice_by_index(begin = var_45196_begin_0, end = var_45196_end_0, end_mask = var_45196_end_mask_0, x = k_59_cast_fp16)[name = string("op_45196_cast_fp16")]; + tensor var_45200_begin_0 = const()[name = string("op_45200_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_45200_end_0 = const()[name = string("op_45200_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_45200_end_mask_0 = const()[name = string("op_45200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45200_cast_fp16 = slice_by_index(begin = var_45200_begin_0, end = var_45200_end_0, end_mask = var_45200_end_mask_0, x = k_59_cast_fp16)[name = string("op_45200_cast_fp16")]; + tensor var_45204_begin_0 = const()[name = string("op_45204_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_45204_end_0 = const()[name = string("op_45204_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_45204_end_mask_0 = const()[name = string("op_45204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45204_cast_fp16 = slice_by_index(begin = var_45204_begin_0, end = var_45204_end_0, end_mask = var_45204_end_mask_0, x = k_59_cast_fp16)[name = string("op_45204_cast_fp16")]; + tensor var_45208_begin_0 = const()[name = string("op_45208_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_45208_end_0 = const()[name = string("op_45208_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_45208_end_mask_0 = const()[name = string("op_45208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45208_cast_fp16 = slice_by_index(begin = var_45208_begin_0, end = var_45208_end_0, end_mask = var_45208_end_mask_0, x = k_59_cast_fp16)[name = string("op_45208_cast_fp16")]; + tensor var_45212_begin_0 = const()[name = string("op_45212_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_45212_end_0 = const()[name = string("op_45212_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_45212_end_mask_0 = const()[name = string("op_45212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45212_cast_fp16 = slice_by_index(begin = var_45212_begin_0, end = var_45212_end_0, end_mask = var_45212_end_mask_0, x = k_59_cast_fp16)[name = string("op_45212_cast_fp16")]; + tensor var_45216_begin_0 = const()[name = string("op_45216_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_45216_end_0 = const()[name = string("op_45216_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_45216_end_mask_0 = const()[name = string("op_45216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45216_cast_fp16 = slice_by_index(begin = var_45216_begin_0, end = var_45216_end_0, end_mask = var_45216_end_mask_0, x = k_59_cast_fp16)[name = string("op_45216_cast_fp16")]; + tensor var_45220_begin_0 = const()[name = string("op_45220_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_45220_end_0 = const()[name = string("op_45220_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_45220_end_mask_0 = const()[name = string("op_45220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45220_cast_fp16 = slice_by_index(begin = var_45220_begin_0, end = var_45220_end_0, end_mask = var_45220_end_mask_0, x = k_59_cast_fp16)[name = string("op_45220_cast_fp16")]; + tensor var_45224_begin_0 = const()[name = string("op_45224_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_45224_end_0 = const()[name = string("op_45224_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_45224_end_mask_0 = const()[name = string("op_45224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45224_cast_fp16 = slice_by_index(begin = var_45224_begin_0, end = var_45224_end_0, end_mask = var_45224_end_mask_0, x = k_59_cast_fp16)[name = string("op_45224_cast_fp16")]; + tensor var_45228_begin_0 = const()[name = string("op_45228_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_45228_end_0 = const()[name = string("op_45228_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_45228_end_mask_0 = const()[name = string("op_45228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45228_cast_fp16 = slice_by_index(begin = var_45228_begin_0, end = var_45228_end_0, end_mask = var_45228_end_mask_0, x = k_59_cast_fp16)[name = string("op_45228_cast_fp16")]; + tensor var_45232_begin_0 = const()[name = string("op_45232_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_45232_end_0 = const()[name = string("op_45232_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_45232_end_mask_0 = const()[name = string("op_45232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45232_cast_fp16 = slice_by_index(begin = var_45232_begin_0, end = var_45232_end_0, end_mask = var_45232_end_mask_0, x = k_59_cast_fp16)[name = string("op_45232_cast_fp16")]; + tensor var_45236_begin_0 = const()[name = string("op_45236_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_45236_end_0 = const()[name = string("op_45236_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_45236_end_mask_0 = const()[name = string("op_45236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45236_cast_fp16 = slice_by_index(begin = var_45236_begin_0, end = var_45236_end_0, end_mask = var_45236_end_mask_0, x = k_59_cast_fp16)[name = string("op_45236_cast_fp16")]; + tensor var_45240_begin_0 = const()[name = string("op_45240_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_45240_end_0 = const()[name = string("op_45240_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_45240_end_mask_0 = const()[name = string("op_45240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45240_cast_fp16 = slice_by_index(begin = var_45240_begin_0, end = var_45240_end_0, end_mask = var_45240_end_mask_0, x = k_59_cast_fp16)[name = string("op_45240_cast_fp16")]; + tensor var_45244_begin_0 = const()[name = string("op_45244_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_45244_end_0 = const()[name = string("op_45244_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_45244_end_mask_0 = const()[name = string("op_45244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45244_cast_fp16 = slice_by_index(begin = var_45244_begin_0, end = var_45244_end_0, end_mask = var_45244_end_mask_0, x = k_59_cast_fp16)[name = string("op_45244_cast_fp16")]; + tensor var_45248_begin_0 = const()[name = string("op_45248_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_45248_end_0 = const()[name = string("op_45248_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_45248_end_mask_0 = const()[name = string("op_45248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45248_cast_fp16 = slice_by_index(begin = var_45248_begin_0, end = var_45248_end_0, end_mask = var_45248_end_mask_0, x = k_59_cast_fp16)[name = string("op_45248_cast_fp16")]; + tensor var_45252_begin_0 = const()[name = string("op_45252_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_45252_end_0 = const()[name = string("op_45252_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_45252_end_mask_0 = const()[name = string("op_45252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45252_cast_fp16 = slice_by_index(begin = var_45252_begin_0, end = var_45252_end_0, end_mask = var_45252_end_mask_0, x = k_59_cast_fp16)[name = string("op_45252_cast_fp16")]; + tensor var_45256_begin_0 = const()[name = string("op_45256_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_45256_end_0 = const()[name = string("op_45256_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_45256_end_mask_0 = const()[name = string("op_45256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45256_cast_fp16 = slice_by_index(begin = var_45256_begin_0, end = var_45256_end_0, end_mask = var_45256_end_mask_0, x = k_59_cast_fp16)[name = string("op_45256_cast_fp16")]; + tensor var_45260_begin_0 = const()[name = string("op_45260_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_45260_end_0 = const()[name = string("op_45260_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_45260_end_mask_0 = const()[name = string("op_45260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45260_cast_fp16 = slice_by_index(begin = var_45260_begin_0, end = var_45260_end_0, end_mask = var_45260_end_mask_0, x = k_59_cast_fp16)[name = string("op_45260_cast_fp16")]; + tensor var_45262_begin_0 = const()[name = string("op_45262_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45262_end_0 = const()[name = string("op_45262_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45262_end_mask_0 = const()[name = string("op_45262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45262_cast_fp16 = slice_by_index(begin = var_45262_begin_0, end = var_45262_end_0, end_mask = var_45262_end_mask_0, x = value_59_cast_fp16)[name = string("op_45262_cast_fp16")]; + tensor var_45266_begin_0 = const()[name = string("op_45266_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45266_end_0 = const()[name = string("op_45266_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_45266_end_mask_0 = const()[name = string("op_45266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45266_cast_fp16 = slice_by_index(begin = var_45266_begin_0, end = var_45266_end_0, end_mask = var_45266_end_mask_0, x = value_59_cast_fp16)[name = string("op_45266_cast_fp16")]; + tensor var_45270_begin_0 = const()[name = string("op_45270_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45270_end_0 = const()[name = string("op_45270_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_45270_end_mask_0 = const()[name = string("op_45270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45270_cast_fp16 = slice_by_index(begin = var_45270_begin_0, end = var_45270_end_0, end_mask = var_45270_end_mask_0, x = value_59_cast_fp16)[name = string("op_45270_cast_fp16")]; + tensor var_45274_begin_0 = const()[name = string("op_45274_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45274_end_0 = const()[name = string("op_45274_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_45274_end_mask_0 = const()[name = string("op_45274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45274_cast_fp16 = slice_by_index(begin = var_45274_begin_0, end = var_45274_end_0, end_mask = var_45274_end_mask_0, x = value_59_cast_fp16)[name = string("op_45274_cast_fp16")]; + tensor var_45278_begin_0 = const()[name = string("op_45278_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45278_end_0 = const()[name = string("op_45278_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_45278_end_mask_0 = const()[name = string("op_45278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45278_cast_fp16 = slice_by_index(begin = var_45278_begin_0, end = var_45278_end_0, end_mask = var_45278_end_mask_0, x = value_59_cast_fp16)[name = string("op_45278_cast_fp16")]; + tensor var_45282_begin_0 = const()[name = string("op_45282_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45282_end_0 = const()[name = string("op_45282_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_45282_end_mask_0 = const()[name = string("op_45282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45282_cast_fp16 = slice_by_index(begin = var_45282_begin_0, end = var_45282_end_0, end_mask = var_45282_end_mask_0, x = value_59_cast_fp16)[name = string("op_45282_cast_fp16")]; + tensor var_45286_begin_0 = const()[name = string("op_45286_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45286_end_0 = const()[name = string("op_45286_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_45286_end_mask_0 = const()[name = string("op_45286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45286_cast_fp16 = slice_by_index(begin = var_45286_begin_0, end = var_45286_end_0, end_mask = var_45286_end_mask_0, x = value_59_cast_fp16)[name = string("op_45286_cast_fp16")]; + tensor var_45290_begin_0 = const()[name = string("op_45290_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45290_end_0 = const()[name = string("op_45290_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_45290_end_mask_0 = const()[name = string("op_45290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45290_cast_fp16 = slice_by_index(begin = var_45290_begin_0, end = var_45290_end_0, end_mask = var_45290_end_mask_0, x = value_59_cast_fp16)[name = string("op_45290_cast_fp16")]; + tensor var_45294_begin_0 = const()[name = string("op_45294_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45294_end_0 = const()[name = string("op_45294_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_45294_end_mask_0 = const()[name = string("op_45294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45294_cast_fp16 = slice_by_index(begin = var_45294_begin_0, end = var_45294_end_0, end_mask = var_45294_end_mask_0, x = value_59_cast_fp16)[name = string("op_45294_cast_fp16")]; + tensor var_45298_begin_0 = const()[name = string("op_45298_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45298_end_0 = const()[name = string("op_45298_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_45298_end_mask_0 = const()[name = string("op_45298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45298_cast_fp16 = slice_by_index(begin = var_45298_begin_0, end = var_45298_end_0, end_mask = var_45298_end_mask_0, x = value_59_cast_fp16)[name = string("op_45298_cast_fp16")]; + tensor var_45302_begin_0 = const()[name = string("op_45302_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45302_end_0 = const()[name = string("op_45302_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_45302_end_mask_0 = const()[name = string("op_45302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45302_cast_fp16 = slice_by_index(begin = var_45302_begin_0, end = var_45302_end_0, end_mask = var_45302_end_mask_0, x = value_59_cast_fp16)[name = string("op_45302_cast_fp16")]; + tensor var_45306_begin_0 = const()[name = string("op_45306_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45306_end_0 = const()[name = string("op_45306_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_45306_end_mask_0 = const()[name = string("op_45306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45306_cast_fp16 = slice_by_index(begin = var_45306_begin_0, end = var_45306_end_0, end_mask = var_45306_end_mask_0, x = value_59_cast_fp16)[name = string("op_45306_cast_fp16")]; + tensor var_45310_begin_0 = const()[name = string("op_45310_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45310_end_0 = const()[name = string("op_45310_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_45310_end_mask_0 = const()[name = string("op_45310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45310_cast_fp16 = slice_by_index(begin = var_45310_begin_0, end = var_45310_end_0, end_mask = var_45310_end_mask_0, x = value_59_cast_fp16)[name = string("op_45310_cast_fp16")]; + tensor var_45314_begin_0 = const()[name = string("op_45314_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45314_end_0 = const()[name = string("op_45314_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_45314_end_mask_0 = const()[name = string("op_45314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45314_cast_fp16 = slice_by_index(begin = var_45314_begin_0, end = var_45314_end_0, end_mask = var_45314_end_mask_0, x = value_59_cast_fp16)[name = string("op_45314_cast_fp16")]; + tensor var_45318_begin_0 = const()[name = string("op_45318_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45318_end_0 = const()[name = string("op_45318_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_45318_end_mask_0 = const()[name = string("op_45318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45318_cast_fp16 = slice_by_index(begin = var_45318_begin_0, end = var_45318_end_0, end_mask = var_45318_end_mask_0, x = value_59_cast_fp16)[name = string("op_45318_cast_fp16")]; + tensor var_45322_begin_0 = const()[name = string("op_45322_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45322_end_0 = const()[name = string("op_45322_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_45322_end_mask_0 = const()[name = string("op_45322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45322_cast_fp16 = slice_by_index(begin = var_45322_begin_0, end = var_45322_end_0, end_mask = var_45322_end_mask_0, x = value_59_cast_fp16)[name = string("op_45322_cast_fp16")]; + tensor var_45326_begin_0 = const()[name = string("op_45326_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45326_end_0 = const()[name = string("op_45326_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_45326_end_mask_0 = const()[name = string("op_45326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45326_cast_fp16 = slice_by_index(begin = var_45326_begin_0, end = var_45326_end_0, end_mask = var_45326_end_mask_0, x = value_59_cast_fp16)[name = string("op_45326_cast_fp16")]; + tensor var_45330_begin_0 = const()[name = string("op_45330_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45330_end_0 = const()[name = string("op_45330_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_45330_end_mask_0 = const()[name = string("op_45330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45330_cast_fp16 = slice_by_index(begin = var_45330_begin_0, end = var_45330_end_0, end_mask = var_45330_end_mask_0, x = value_59_cast_fp16)[name = string("op_45330_cast_fp16")]; + tensor var_45334_begin_0 = const()[name = string("op_45334_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45334_end_0 = const()[name = string("op_45334_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_45334_end_mask_0 = const()[name = string("op_45334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45334_cast_fp16 = slice_by_index(begin = var_45334_begin_0, end = var_45334_end_0, end_mask = var_45334_end_mask_0, x = value_59_cast_fp16)[name = string("op_45334_cast_fp16")]; + tensor var_45338_begin_0 = const()[name = string("op_45338_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45338_end_0 = const()[name = string("op_45338_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_45338_end_mask_0 = const()[name = string("op_45338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45338_cast_fp16 = slice_by_index(begin = var_45338_begin_0, end = var_45338_end_0, end_mask = var_45338_end_mask_0, x = value_59_cast_fp16)[name = string("op_45338_cast_fp16")]; + string _SplitHeadsQ__mh_w_4641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4641_equation_0, values = (var_45184_cast_fp16, var_44626_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4641_cast_fp16")]; + string _SplitHeadsQ__mh_w_4643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4643_equation_0, values = (var_45184_cast_fp16, var_44633_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4643_cast_fp16")]; + string _SplitHeadsQ__mh_w_4645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4645_equation_0, values = (var_45184_cast_fp16, var_44640_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4645_cast_fp16")]; + string _SplitHeadsQ__mh_w_4647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4647_equation_0, values = (var_45184_cast_fp16, var_44647_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4647_cast_fp16")]; + string _SplitHeadsQ__mh_w_4649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4649_equation_0, values = (var_45188_cast_fp16, var_44654_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4649_cast_fp16")]; + string _SplitHeadsQ__mh_w_4651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4651_equation_0, values = (var_45188_cast_fp16, var_44661_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4651_cast_fp16")]; + string _SplitHeadsQ__mh_w_4653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4653_equation_0, values = (var_45188_cast_fp16, var_44668_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4653_cast_fp16")]; + string _SplitHeadsQ__mh_w_4655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4655_equation_0, values = (var_45188_cast_fp16, var_44675_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4655_cast_fp16")]; + string _SplitHeadsQ__mh_w_4657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4657_equation_0, values = (var_45192_cast_fp16, var_44682_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4657_cast_fp16")]; + string _SplitHeadsQ__mh_w_4659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4659_equation_0, values = (var_45192_cast_fp16, var_44689_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4659_cast_fp16")]; + string _SplitHeadsQ__mh_w_4661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4661_equation_0, values = (var_45192_cast_fp16, var_44696_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4661_cast_fp16")]; + string _SplitHeadsQ__mh_w_4663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4663_equation_0, values = (var_45192_cast_fp16, var_44703_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4663_cast_fp16")]; + string _SplitHeadsQ__mh_w_4665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4665_equation_0, values = (var_45196_cast_fp16, var_44710_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4665_cast_fp16")]; + string _SplitHeadsQ__mh_w_4667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4667_equation_0, values = (var_45196_cast_fp16, var_44717_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4667_cast_fp16")]; + string _SplitHeadsQ__mh_w_4669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4669_equation_0, values = (var_45196_cast_fp16, var_44724_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4669_cast_fp16")]; + string _SplitHeadsQ__mh_w_4671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4671_equation_0, values = (var_45196_cast_fp16, var_44731_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4671_cast_fp16")]; + string _SplitHeadsQ__mh_w_4673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4673_equation_0, values = (var_45200_cast_fp16, var_44738_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4673_cast_fp16")]; + string _SplitHeadsQ__mh_w_4675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4675_equation_0, values = (var_45200_cast_fp16, var_44745_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4675_cast_fp16")]; + string _SplitHeadsQ__mh_w_4677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4677_equation_0, values = (var_45200_cast_fp16, var_44752_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4677_cast_fp16")]; + string _SplitHeadsQ__mh_w_4679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4679_equation_0, values = (var_45200_cast_fp16, var_44759_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4679_cast_fp16")]; + string _SplitHeadsQ__mh_w_4681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4681_equation_0, values = (var_45204_cast_fp16, var_44766_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4681_cast_fp16")]; + string _SplitHeadsQ__mh_w_4683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4683_equation_0, values = (var_45204_cast_fp16, var_44773_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4683_cast_fp16")]; + string _SplitHeadsQ__mh_w_4685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4685_equation_0, values = (var_45204_cast_fp16, var_44780_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4685_cast_fp16")]; + string _SplitHeadsQ__mh_w_4687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4687_equation_0, values = (var_45204_cast_fp16, var_44787_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4687_cast_fp16")]; + string _SplitHeadsQ__mh_w_4689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4689_equation_0, values = (var_45208_cast_fp16, var_44794_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4689_cast_fp16")]; + string _SplitHeadsQ__mh_w_4691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4691_equation_0, values = (var_45208_cast_fp16, var_44801_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4691_cast_fp16")]; + string _SplitHeadsQ__mh_w_4693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4693_equation_0, values = (var_45208_cast_fp16, var_44808_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4693_cast_fp16")]; + string _SplitHeadsQ__mh_w_4695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4695_equation_0, values = (var_45208_cast_fp16, var_44815_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4695_cast_fp16")]; + string _SplitHeadsQ__mh_w_4697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4697_equation_0, values = (var_45212_cast_fp16, var_44822_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4697_cast_fp16")]; + string _SplitHeadsQ__mh_w_4699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4699_equation_0, values = (var_45212_cast_fp16, var_44829_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4699_cast_fp16")]; + string _SplitHeadsQ__mh_w_4701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4701_equation_0, values = (var_45212_cast_fp16, var_44836_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4701_cast_fp16")]; + string _SplitHeadsQ__mh_w_4703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4703_equation_0, values = (var_45212_cast_fp16, var_44843_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4703_cast_fp16")]; + string _SplitHeadsQ__mh_w_4705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4705_equation_0, values = (var_45216_cast_fp16, var_44850_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4705_cast_fp16")]; + string _SplitHeadsQ__mh_w_4707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4707_equation_0, values = (var_45216_cast_fp16, var_44857_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4707_cast_fp16")]; + string _SplitHeadsQ__mh_w_4709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4709_equation_0, values = (var_45216_cast_fp16, var_44864_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4709_cast_fp16")]; + string _SplitHeadsQ__mh_w_4711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4711_equation_0, values = (var_45216_cast_fp16, var_44871_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4711_cast_fp16")]; + string _SplitHeadsQ__mh_w_4713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4713_equation_0, values = (var_45220_cast_fp16, var_44878_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4713_cast_fp16")]; + string _SplitHeadsQ__mh_w_4715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4715_equation_0, values = (var_45220_cast_fp16, var_44885_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4715_cast_fp16")]; + string _SplitHeadsQ__mh_w_4717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4717_equation_0, values = (var_45220_cast_fp16, var_44892_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4717_cast_fp16")]; + string _SplitHeadsQ__mh_w_4719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4719_equation_0, values = (var_45220_cast_fp16, var_44899_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4719_cast_fp16")]; + string _SplitHeadsQ__mh_w_4721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4721_equation_0, values = (var_45224_cast_fp16, var_44906_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4721_cast_fp16")]; + string _SplitHeadsQ__mh_w_4723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4723_equation_0, values = (var_45224_cast_fp16, var_44913_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4723_cast_fp16")]; + string _SplitHeadsQ__mh_w_4725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4725_equation_0, values = (var_45224_cast_fp16, var_44920_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4725_cast_fp16")]; + string _SplitHeadsQ__mh_w_4727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4727_equation_0, values = (var_45224_cast_fp16, var_44927_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4727_cast_fp16")]; + string _SplitHeadsQ__mh_w_4729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4729_equation_0, values = (var_45228_cast_fp16, var_44934_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4729_cast_fp16")]; + string _SplitHeadsQ__mh_w_4731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4731_equation_0, values = (var_45228_cast_fp16, var_44941_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4731_cast_fp16")]; + string _SplitHeadsQ__mh_w_4733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4733_equation_0, values = (var_45228_cast_fp16, var_44948_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4733_cast_fp16")]; + string _SplitHeadsQ__mh_w_4735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4735_equation_0, values = (var_45228_cast_fp16, var_44955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4735_cast_fp16")]; + string _SplitHeadsQ__mh_w_4737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4737_equation_0, values = (var_45232_cast_fp16, var_44962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4737_cast_fp16")]; + string _SplitHeadsQ__mh_w_4739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4739_equation_0, values = (var_45232_cast_fp16, var_44969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4739_cast_fp16")]; + string _SplitHeadsQ__mh_w_4741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4741_equation_0, values = (var_45232_cast_fp16, var_44976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4741_cast_fp16")]; + string _SplitHeadsQ__mh_w_4743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4743_equation_0, values = (var_45232_cast_fp16, var_44983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4743_cast_fp16")]; + string _SplitHeadsQ__mh_w_4745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4745_equation_0, values = (var_45236_cast_fp16, var_44990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4745_cast_fp16")]; + string _SplitHeadsQ__mh_w_4747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4747_equation_0, values = (var_45236_cast_fp16, var_44997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4747_cast_fp16")]; + string _SplitHeadsQ__mh_w_4749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4749_equation_0, values = (var_45236_cast_fp16, var_45004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4749_cast_fp16")]; + string _SplitHeadsQ__mh_w_4751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4751_equation_0, values = (var_45236_cast_fp16, var_45011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4751_cast_fp16")]; + string _SplitHeadsQ__mh_w_4753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4753_equation_0, values = (var_45240_cast_fp16, var_45018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4753_cast_fp16")]; + string _SplitHeadsQ__mh_w_4755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4755_equation_0, values = (var_45240_cast_fp16, var_45025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4755_cast_fp16")]; + string _SplitHeadsQ__mh_w_4757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4757_equation_0, values = (var_45240_cast_fp16, var_45032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4757_cast_fp16")]; + string _SplitHeadsQ__mh_w_4759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4759_equation_0, values = (var_45240_cast_fp16, var_45039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4759_cast_fp16")]; + string _SplitHeadsQ__mh_w_4761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4761_equation_0, values = (var_45244_cast_fp16, var_45046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4761_cast_fp16")]; + string _SplitHeadsQ__mh_w_4763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4763_equation_0, values = (var_45244_cast_fp16, var_45053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4763_cast_fp16")]; + string _SplitHeadsQ__mh_w_4765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4765_equation_0, values = (var_45244_cast_fp16, var_45060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4765_cast_fp16")]; + string _SplitHeadsQ__mh_w_4767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4767_equation_0, values = (var_45244_cast_fp16, var_45067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4767_cast_fp16")]; + string _SplitHeadsQ__mh_w_4769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4769_equation_0, values = (var_45248_cast_fp16, var_45074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4769_cast_fp16")]; + string _SplitHeadsQ__mh_w_4771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4771_equation_0, values = (var_45248_cast_fp16, var_45081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4771_cast_fp16")]; + string _SplitHeadsQ__mh_w_4773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4773_equation_0, values = (var_45248_cast_fp16, var_45088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4773_cast_fp16")]; + string _SplitHeadsQ__mh_w_4775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4775_equation_0, values = (var_45248_cast_fp16, var_45095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4775_cast_fp16")]; + string _SplitHeadsQ__mh_w_4777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4777_equation_0, values = (var_45252_cast_fp16, var_45102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4777_cast_fp16")]; + string _SplitHeadsQ__mh_w_4779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4779_equation_0, values = (var_45252_cast_fp16, var_45109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4779_cast_fp16")]; + string _SplitHeadsQ__mh_w_4781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4781_equation_0, values = (var_45252_cast_fp16, var_45116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4781_cast_fp16")]; + string _SplitHeadsQ__mh_w_4783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4783_equation_0, values = (var_45252_cast_fp16, var_45123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4783_cast_fp16")]; + string _SplitHeadsQ__mh_w_4785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4785_equation_0, values = (var_45256_cast_fp16, var_45130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4785_cast_fp16")]; + string _SplitHeadsQ__mh_w_4787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4787_equation_0, values = (var_45256_cast_fp16, var_45137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4787_cast_fp16")]; + string _SplitHeadsQ__mh_w_4789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4789_equation_0, values = (var_45256_cast_fp16, var_45144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4789_cast_fp16")]; + string _SplitHeadsQ__mh_w_4791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4791_equation_0, values = (var_45256_cast_fp16, var_45151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4791_cast_fp16")]; + string _SplitHeadsQ__mh_w_4793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4793_equation_0, values = (var_45260_cast_fp16, var_45158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4793_cast_fp16")]; + string _SplitHeadsQ__mh_w_4795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4795_equation_0, values = (var_45260_cast_fp16, var_45165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4795_cast_fp16")]; + string _SplitHeadsQ__mh_w_4797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4797_equation_0, values = (var_45260_cast_fp16, var_45172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4797_cast_fp16")]; + string _SplitHeadsQ__mh_w_4799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4799_equation_0, values = (var_45260_cast_fp16, var_45179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4799_cast_fp16")]; + fp16 var_45501_to_fp16 = const()[name = string("op_45501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4641_cast_fp16, y = var_45501_to_fp16)[name = string("aw_chunk_4641_cast_fp16")]; + fp16 var_45503_to_fp16 = const()[name = string("op_45503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4643_cast_fp16, y = var_45503_to_fp16)[name = string("aw_chunk_4643_cast_fp16")]; + fp16 var_45505_to_fp16 = const()[name = string("op_45505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4645_cast_fp16, y = var_45505_to_fp16)[name = string("aw_chunk_4645_cast_fp16")]; + fp16 var_45507_to_fp16 = const()[name = string("op_45507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4647_cast_fp16, y = var_45507_to_fp16)[name = string("aw_chunk_4647_cast_fp16")]; + fp16 var_45509_to_fp16 = const()[name = string("op_45509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4649_cast_fp16, y = var_45509_to_fp16)[name = string("aw_chunk_4649_cast_fp16")]; + fp16 var_45511_to_fp16 = const()[name = string("op_45511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4651_cast_fp16, y = var_45511_to_fp16)[name = string("aw_chunk_4651_cast_fp16")]; + fp16 var_45513_to_fp16 = const()[name = string("op_45513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4653_cast_fp16, y = var_45513_to_fp16)[name = string("aw_chunk_4653_cast_fp16")]; + fp16 var_45515_to_fp16 = const()[name = string("op_45515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4655_cast_fp16, y = var_45515_to_fp16)[name = string("aw_chunk_4655_cast_fp16")]; + fp16 var_45517_to_fp16 = const()[name = string("op_45517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4657_cast_fp16, y = var_45517_to_fp16)[name = string("aw_chunk_4657_cast_fp16")]; + fp16 var_45519_to_fp16 = const()[name = string("op_45519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4659_cast_fp16, y = var_45519_to_fp16)[name = string("aw_chunk_4659_cast_fp16")]; + fp16 var_45521_to_fp16 = const()[name = string("op_45521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4661_cast_fp16, y = var_45521_to_fp16)[name = string("aw_chunk_4661_cast_fp16")]; + fp16 var_45523_to_fp16 = const()[name = string("op_45523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4663_cast_fp16, y = var_45523_to_fp16)[name = string("aw_chunk_4663_cast_fp16")]; + fp16 var_45525_to_fp16 = const()[name = string("op_45525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4665_cast_fp16, y = var_45525_to_fp16)[name = string("aw_chunk_4665_cast_fp16")]; + fp16 var_45527_to_fp16 = const()[name = string("op_45527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4667_cast_fp16, y = var_45527_to_fp16)[name = string("aw_chunk_4667_cast_fp16")]; + fp16 var_45529_to_fp16 = const()[name = string("op_45529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4669_cast_fp16, y = var_45529_to_fp16)[name = string("aw_chunk_4669_cast_fp16")]; + fp16 var_45531_to_fp16 = const()[name = string("op_45531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4671_cast_fp16, y = var_45531_to_fp16)[name = string("aw_chunk_4671_cast_fp16")]; + fp16 var_45533_to_fp16 = const()[name = string("op_45533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4673_cast_fp16, y = var_45533_to_fp16)[name = string("aw_chunk_4673_cast_fp16")]; + fp16 var_45535_to_fp16 = const()[name = string("op_45535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4675_cast_fp16, y = var_45535_to_fp16)[name = string("aw_chunk_4675_cast_fp16")]; + fp16 var_45537_to_fp16 = const()[name = string("op_45537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4677_cast_fp16, y = var_45537_to_fp16)[name = string("aw_chunk_4677_cast_fp16")]; + fp16 var_45539_to_fp16 = const()[name = string("op_45539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4679_cast_fp16, y = var_45539_to_fp16)[name = string("aw_chunk_4679_cast_fp16")]; + fp16 var_45541_to_fp16 = const()[name = string("op_45541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4681_cast_fp16, y = var_45541_to_fp16)[name = string("aw_chunk_4681_cast_fp16")]; + fp16 var_45543_to_fp16 = const()[name = string("op_45543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4683_cast_fp16, y = var_45543_to_fp16)[name = string("aw_chunk_4683_cast_fp16")]; + fp16 var_45545_to_fp16 = const()[name = string("op_45545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4685_cast_fp16, y = var_45545_to_fp16)[name = string("aw_chunk_4685_cast_fp16")]; + fp16 var_45547_to_fp16 = const()[name = string("op_45547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4687_cast_fp16, y = var_45547_to_fp16)[name = string("aw_chunk_4687_cast_fp16")]; + fp16 var_45549_to_fp16 = const()[name = string("op_45549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4689_cast_fp16, y = var_45549_to_fp16)[name = string("aw_chunk_4689_cast_fp16")]; + fp16 var_45551_to_fp16 = const()[name = string("op_45551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4691_cast_fp16, y = var_45551_to_fp16)[name = string("aw_chunk_4691_cast_fp16")]; + fp16 var_45553_to_fp16 = const()[name = string("op_45553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4693_cast_fp16, y = var_45553_to_fp16)[name = string("aw_chunk_4693_cast_fp16")]; + fp16 var_45555_to_fp16 = const()[name = string("op_45555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4695_cast_fp16, y = var_45555_to_fp16)[name = string("aw_chunk_4695_cast_fp16")]; + fp16 var_45557_to_fp16 = const()[name = string("op_45557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4697_cast_fp16, y = var_45557_to_fp16)[name = string("aw_chunk_4697_cast_fp16")]; + fp16 var_45559_to_fp16 = const()[name = string("op_45559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4699_cast_fp16, y = var_45559_to_fp16)[name = string("aw_chunk_4699_cast_fp16")]; + fp16 var_45561_to_fp16 = const()[name = string("op_45561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4701_cast_fp16, y = var_45561_to_fp16)[name = string("aw_chunk_4701_cast_fp16")]; + fp16 var_45563_to_fp16 = const()[name = string("op_45563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4703_cast_fp16, y = var_45563_to_fp16)[name = string("aw_chunk_4703_cast_fp16")]; + fp16 var_45565_to_fp16 = const()[name = string("op_45565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4705_cast_fp16, y = var_45565_to_fp16)[name = string("aw_chunk_4705_cast_fp16")]; + fp16 var_45567_to_fp16 = const()[name = string("op_45567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4707_cast_fp16, y = var_45567_to_fp16)[name = string("aw_chunk_4707_cast_fp16")]; + fp16 var_45569_to_fp16 = const()[name = string("op_45569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4709_cast_fp16, y = var_45569_to_fp16)[name = string("aw_chunk_4709_cast_fp16")]; + fp16 var_45571_to_fp16 = const()[name = string("op_45571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4711_cast_fp16, y = var_45571_to_fp16)[name = string("aw_chunk_4711_cast_fp16")]; + fp16 var_45573_to_fp16 = const()[name = string("op_45573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4713_cast_fp16, y = var_45573_to_fp16)[name = string("aw_chunk_4713_cast_fp16")]; + fp16 var_45575_to_fp16 = const()[name = string("op_45575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4715_cast_fp16, y = var_45575_to_fp16)[name = string("aw_chunk_4715_cast_fp16")]; + fp16 var_45577_to_fp16 = const()[name = string("op_45577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4717_cast_fp16, y = var_45577_to_fp16)[name = string("aw_chunk_4717_cast_fp16")]; + fp16 var_45579_to_fp16 = const()[name = string("op_45579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4719_cast_fp16, y = var_45579_to_fp16)[name = string("aw_chunk_4719_cast_fp16")]; + fp16 var_45581_to_fp16 = const()[name = string("op_45581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4721_cast_fp16, y = var_45581_to_fp16)[name = string("aw_chunk_4721_cast_fp16")]; + fp16 var_45583_to_fp16 = const()[name = string("op_45583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4723_cast_fp16, y = var_45583_to_fp16)[name = string("aw_chunk_4723_cast_fp16")]; + fp16 var_45585_to_fp16 = const()[name = string("op_45585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4725_cast_fp16, y = var_45585_to_fp16)[name = string("aw_chunk_4725_cast_fp16")]; + fp16 var_45587_to_fp16 = const()[name = string("op_45587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4727_cast_fp16, y = var_45587_to_fp16)[name = string("aw_chunk_4727_cast_fp16")]; + fp16 var_45589_to_fp16 = const()[name = string("op_45589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4729_cast_fp16, y = var_45589_to_fp16)[name = string("aw_chunk_4729_cast_fp16")]; + fp16 var_45591_to_fp16 = const()[name = string("op_45591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4731_cast_fp16, y = var_45591_to_fp16)[name = string("aw_chunk_4731_cast_fp16")]; + fp16 var_45593_to_fp16 = const()[name = string("op_45593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4733_cast_fp16, y = var_45593_to_fp16)[name = string("aw_chunk_4733_cast_fp16")]; + fp16 var_45595_to_fp16 = const()[name = string("op_45595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4735_cast_fp16, y = var_45595_to_fp16)[name = string("aw_chunk_4735_cast_fp16")]; + fp16 var_45597_to_fp16 = const()[name = string("op_45597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4737_cast_fp16, y = var_45597_to_fp16)[name = string("aw_chunk_4737_cast_fp16")]; + fp16 var_45599_to_fp16 = const()[name = string("op_45599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4739_cast_fp16, y = var_45599_to_fp16)[name = string("aw_chunk_4739_cast_fp16")]; + fp16 var_45601_to_fp16 = const()[name = string("op_45601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4741_cast_fp16, y = var_45601_to_fp16)[name = string("aw_chunk_4741_cast_fp16")]; + fp16 var_45603_to_fp16 = const()[name = string("op_45603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4743_cast_fp16, y = var_45603_to_fp16)[name = string("aw_chunk_4743_cast_fp16")]; + fp16 var_45605_to_fp16 = const()[name = string("op_45605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4745_cast_fp16, y = var_45605_to_fp16)[name = string("aw_chunk_4745_cast_fp16")]; + fp16 var_45607_to_fp16 = const()[name = string("op_45607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4747_cast_fp16, y = var_45607_to_fp16)[name = string("aw_chunk_4747_cast_fp16")]; + fp16 var_45609_to_fp16 = const()[name = string("op_45609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4749_cast_fp16, y = var_45609_to_fp16)[name = string("aw_chunk_4749_cast_fp16")]; + fp16 var_45611_to_fp16 = const()[name = string("op_45611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4751_cast_fp16, y = var_45611_to_fp16)[name = string("aw_chunk_4751_cast_fp16")]; + fp16 var_45613_to_fp16 = const()[name = string("op_45613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4753_cast_fp16, y = var_45613_to_fp16)[name = string("aw_chunk_4753_cast_fp16")]; + fp16 var_45615_to_fp16 = const()[name = string("op_45615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4755_cast_fp16, y = var_45615_to_fp16)[name = string("aw_chunk_4755_cast_fp16")]; + fp16 var_45617_to_fp16 = const()[name = string("op_45617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4757_cast_fp16, y = var_45617_to_fp16)[name = string("aw_chunk_4757_cast_fp16")]; + fp16 var_45619_to_fp16 = const()[name = string("op_45619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4759_cast_fp16, y = var_45619_to_fp16)[name = string("aw_chunk_4759_cast_fp16")]; + fp16 var_45621_to_fp16 = const()[name = string("op_45621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4761_cast_fp16, y = var_45621_to_fp16)[name = string("aw_chunk_4761_cast_fp16")]; + fp16 var_45623_to_fp16 = const()[name = string("op_45623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4763_cast_fp16, y = var_45623_to_fp16)[name = string("aw_chunk_4763_cast_fp16")]; + fp16 var_45625_to_fp16 = const()[name = string("op_45625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4765_cast_fp16, y = var_45625_to_fp16)[name = string("aw_chunk_4765_cast_fp16")]; + fp16 var_45627_to_fp16 = const()[name = string("op_45627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4767_cast_fp16, y = var_45627_to_fp16)[name = string("aw_chunk_4767_cast_fp16")]; + fp16 var_45629_to_fp16 = const()[name = string("op_45629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4769_cast_fp16, y = var_45629_to_fp16)[name = string("aw_chunk_4769_cast_fp16")]; + fp16 var_45631_to_fp16 = const()[name = string("op_45631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4771_cast_fp16, y = var_45631_to_fp16)[name = string("aw_chunk_4771_cast_fp16")]; + fp16 var_45633_to_fp16 = const()[name = string("op_45633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4773_cast_fp16, y = var_45633_to_fp16)[name = string("aw_chunk_4773_cast_fp16")]; + fp16 var_45635_to_fp16 = const()[name = string("op_45635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4775_cast_fp16, y = var_45635_to_fp16)[name = string("aw_chunk_4775_cast_fp16")]; + fp16 var_45637_to_fp16 = const()[name = string("op_45637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4777_cast_fp16, y = var_45637_to_fp16)[name = string("aw_chunk_4777_cast_fp16")]; + fp16 var_45639_to_fp16 = const()[name = string("op_45639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4779_cast_fp16, y = var_45639_to_fp16)[name = string("aw_chunk_4779_cast_fp16")]; + fp16 var_45641_to_fp16 = const()[name = string("op_45641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4781_cast_fp16, y = var_45641_to_fp16)[name = string("aw_chunk_4781_cast_fp16")]; + fp16 var_45643_to_fp16 = const()[name = string("op_45643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4783_cast_fp16, y = var_45643_to_fp16)[name = string("aw_chunk_4783_cast_fp16")]; + fp16 var_45645_to_fp16 = const()[name = string("op_45645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4785_cast_fp16, y = var_45645_to_fp16)[name = string("aw_chunk_4785_cast_fp16")]; + fp16 var_45647_to_fp16 = const()[name = string("op_45647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4787_cast_fp16, y = var_45647_to_fp16)[name = string("aw_chunk_4787_cast_fp16")]; + fp16 var_45649_to_fp16 = const()[name = string("op_45649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4789_cast_fp16, y = var_45649_to_fp16)[name = string("aw_chunk_4789_cast_fp16")]; + fp16 var_45651_to_fp16 = const()[name = string("op_45651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4791_cast_fp16, y = var_45651_to_fp16)[name = string("aw_chunk_4791_cast_fp16")]; + fp16 var_45653_to_fp16 = const()[name = string("op_45653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4793_cast_fp16, y = var_45653_to_fp16)[name = string("aw_chunk_4793_cast_fp16")]; + fp16 var_45655_to_fp16 = const()[name = string("op_45655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4795_cast_fp16, y = var_45655_to_fp16)[name = string("aw_chunk_4795_cast_fp16")]; + fp16 var_45657_to_fp16 = const()[name = string("op_45657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4797_cast_fp16, y = var_45657_to_fp16)[name = string("aw_chunk_4797_cast_fp16")]; + fp16 var_45659_to_fp16 = const()[name = string("op_45659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4799_cast_fp16, y = var_45659_to_fp16)[name = string("aw_chunk_4799_cast_fp16")]; + tensor var_45661_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4641_cast_fp16)[name = string("op_45661_cast_fp16")]; + tensor var_45662_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4643_cast_fp16)[name = string("op_45662_cast_fp16")]; + tensor var_45663_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4645_cast_fp16)[name = string("op_45663_cast_fp16")]; + tensor var_45664_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4647_cast_fp16)[name = string("op_45664_cast_fp16")]; + tensor var_45665_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4649_cast_fp16)[name = string("op_45665_cast_fp16")]; + tensor var_45666_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4651_cast_fp16)[name = string("op_45666_cast_fp16")]; + tensor var_45667_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4653_cast_fp16)[name = string("op_45667_cast_fp16")]; + tensor var_45668_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4655_cast_fp16)[name = string("op_45668_cast_fp16")]; + tensor var_45669_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4657_cast_fp16)[name = string("op_45669_cast_fp16")]; + tensor var_45670_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4659_cast_fp16)[name = string("op_45670_cast_fp16")]; + tensor var_45671_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4661_cast_fp16)[name = string("op_45671_cast_fp16")]; + tensor var_45672_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4663_cast_fp16)[name = string("op_45672_cast_fp16")]; + tensor var_45673_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4665_cast_fp16)[name = string("op_45673_cast_fp16")]; + tensor var_45674_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4667_cast_fp16)[name = string("op_45674_cast_fp16")]; + tensor var_45675_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4669_cast_fp16)[name = string("op_45675_cast_fp16")]; + tensor var_45676_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4671_cast_fp16)[name = string("op_45676_cast_fp16")]; + tensor var_45677_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4673_cast_fp16)[name = string("op_45677_cast_fp16")]; + tensor var_45678_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4675_cast_fp16)[name = string("op_45678_cast_fp16")]; + tensor var_45679_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4677_cast_fp16)[name = string("op_45679_cast_fp16")]; + tensor var_45680_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4679_cast_fp16)[name = string("op_45680_cast_fp16")]; + tensor var_45681_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4681_cast_fp16)[name = string("op_45681_cast_fp16")]; + tensor var_45682_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4683_cast_fp16)[name = string("op_45682_cast_fp16")]; + tensor var_45683_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4685_cast_fp16)[name = string("op_45683_cast_fp16")]; + tensor var_45684_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4687_cast_fp16)[name = string("op_45684_cast_fp16")]; + tensor var_45685_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4689_cast_fp16)[name = string("op_45685_cast_fp16")]; + tensor var_45686_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4691_cast_fp16)[name = string("op_45686_cast_fp16")]; + tensor var_45687_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4693_cast_fp16)[name = string("op_45687_cast_fp16")]; + tensor var_45688_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4695_cast_fp16)[name = string("op_45688_cast_fp16")]; + tensor var_45689_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4697_cast_fp16)[name = string("op_45689_cast_fp16")]; + tensor var_45690_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4699_cast_fp16)[name = string("op_45690_cast_fp16")]; + tensor var_45691_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4701_cast_fp16)[name = string("op_45691_cast_fp16")]; + tensor var_45692_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4703_cast_fp16)[name = string("op_45692_cast_fp16")]; + tensor var_45693_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4705_cast_fp16)[name = string("op_45693_cast_fp16")]; + tensor var_45694_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4707_cast_fp16)[name = string("op_45694_cast_fp16")]; + tensor var_45695_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4709_cast_fp16)[name = string("op_45695_cast_fp16")]; + tensor var_45696_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4711_cast_fp16)[name = string("op_45696_cast_fp16")]; + tensor var_45697_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4713_cast_fp16)[name = string("op_45697_cast_fp16")]; + tensor var_45698_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4715_cast_fp16)[name = string("op_45698_cast_fp16")]; + tensor var_45699_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4717_cast_fp16)[name = string("op_45699_cast_fp16")]; + tensor var_45700_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4719_cast_fp16)[name = string("op_45700_cast_fp16")]; + tensor var_45701_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4721_cast_fp16)[name = string("op_45701_cast_fp16")]; + tensor var_45702_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4723_cast_fp16)[name = string("op_45702_cast_fp16")]; + tensor var_45703_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4725_cast_fp16)[name = string("op_45703_cast_fp16")]; + tensor var_45704_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4727_cast_fp16)[name = string("op_45704_cast_fp16")]; + tensor var_45705_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4729_cast_fp16)[name = string("op_45705_cast_fp16")]; + tensor var_45706_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4731_cast_fp16)[name = string("op_45706_cast_fp16")]; + tensor var_45707_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4733_cast_fp16)[name = string("op_45707_cast_fp16")]; + tensor var_45708_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4735_cast_fp16)[name = string("op_45708_cast_fp16")]; + tensor var_45709_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4737_cast_fp16)[name = string("op_45709_cast_fp16")]; + tensor var_45710_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4739_cast_fp16)[name = string("op_45710_cast_fp16")]; + tensor var_45711_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4741_cast_fp16)[name = string("op_45711_cast_fp16")]; + tensor var_45712_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4743_cast_fp16)[name = string("op_45712_cast_fp16")]; + tensor var_45713_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4745_cast_fp16)[name = string("op_45713_cast_fp16")]; + tensor var_45714_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4747_cast_fp16)[name = string("op_45714_cast_fp16")]; + tensor var_45715_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4749_cast_fp16)[name = string("op_45715_cast_fp16")]; + tensor var_45716_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4751_cast_fp16)[name = string("op_45716_cast_fp16")]; + tensor var_45717_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4753_cast_fp16)[name = string("op_45717_cast_fp16")]; + tensor var_45718_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4755_cast_fp16)[name = string("op_45718_cast_fp16")]; + tensor var_45719_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4757_cast_fp16)[name = string("op_45719_cast_fp16")]; + tensor var_45720_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4759_cast_fp16)[name = string("op_45720_cast_fp16")]; + tensor var_45721_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4761_cast_fp16)[name = string("op_45721_cast_fp16")]; + tensor var_45722_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4763_cast_fp16)[name = string("op_45722_cast_fp16")]; + tensor var_45723_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4765_cast_fp16)[name = string("op_45723_cast_fp16")]; + tensor var_45724_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4767_cast_fp16)[name = string("op_45724_cast_fp16")]; + tensor var_45725_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4769_cast_fp16)[name = string("op_45725_cast_fp16")]; + tensor var_45726_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4771_cast_fp16)[name = string("op_45726_cast_fp16")]; + tensor var_45727_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4773_cast_fp16)[name = string("op_45727_cast_fp16")]; + tensor var_45728_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4775_cast_fp16)[name = string("op_45728_cast_fp16")]; + tensor var_45729_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4777_cast_fp16)[name = string("op_45729_cast_fp16")]; + tensor var_45730_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4779_cast_fp16)[name = string("op_45730_cast_fp16")]; + tensor var_45731_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4781_cast_fp16)[name = string("op_45731_cast_fp16")]; + tensor var_45732_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4783_cast_fp16)[name = string("op_45732_cast_fp16")]; + tensor var_45733_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4785_cast_fp16)[name = string("op_45733_cast_fp16")]; + tensor var_45734_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4787_cast_fp16)[name = string("op_45734_cast_fp16")]; + tensor var_45735_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4789_cast_fp16)[name = string("op_45735_cast_fp16")]; + tensor var_45736_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4791_cast_fp16)[name = string("op_45736_cast_fp16")]; + tensor var_45737_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4793_cast_fp16)[name = string("op_45737_cast_fp16")]; + tensor var_45738_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4795_cast_fp16)[name = string("op_45738_cast_fp16")]; + tensor var_45739_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4797_cast_fp16)[name = string("op_45739_cast_fp16")]; + tensor var_45740_cast_fp16 = softmax(axis = var_44486, x = aw_chunk_4799_cast_fp16)[name = string("op_45740_cast_fp16")]; + string var_45742_equation_0 = const()[name = string("op_45742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45742_cast_fp16 = einsum(equation = var_45742_equation_0, values = (var_45262_cast_fp16, var_45661_cast_fp16))[name = string("op_45742_cast_fp16")]; + string var_45744_equation_0 = const()[name = string("op_45744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45744_cast_fp16 = einsum(equation = var_45744_equation_0, values = (var_45262_cast_fp16, var_45662_cast_fp16))[name = string("op_45744_cast_fp16")]; + string var_45746_equation_0 = const()[name = string("op_45746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45746_cast_fp16 = einsum(equation = var_45746_equation_0, values = (var_45262_cast_fp16, var_45663_cast_fp16))[name = string("op_45746_cast_fp16")]; + string var_45748_equation_0 = const()[name = string("op_45748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45748_cast_fp16 = einsum(equation = var_45748_equation_0, values = (var_45262_cast_fp16, var_45664_cast_fp16))[name = string("op_45748_cast_fp16")]; + string var_45750_equation_0 = const()[name = string("op_45750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45750_cast_fp16 = einsum(equation = var_45750_equation_0, values = (var_45266_cast_fp16, var_45665_cast_fp16))[name = string("op_45750_cast_fp16")]; + string var_45752_equation_0 = const()[name = string("op_45752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45752_cast_fp16 = einsum(equation = var_45752_equation_0, values = (var_45266_cast_fp16, var_45666_cast_fp16))[name = string("op_45752_cast_fp16")]; + string var_45754_equation_0 = const()[name = string("op_45754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45754_cast_fp16 = einsum(equation = var_45754_equation_0, values = (var_45266_cast_fp16, var_45667_cast_fp16))[name = string("op_45754_cast_fp16")]; + string var_45756_equation_0 = const()[name = string("op_45756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45756_cast_fp16 = einsum(equation = var_45756_equation_0, values = (var_45266_cast_fp16, var_45668_cast_fp16))[name = string("op_45756_cast_fp16")]; + string var_45758_equation_0 = const()[name = string("op_45758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45758_cast_fp16 = einsum(equation = var_45758_equation_0, values = (var_45270_cast_fp16, var_45669_cast_fp16))[name = string("op_45758_cast_fp16")]; + string var_45760_equation_0 = const()[name = string("op_45760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45760_cast_fp16 = einsum(equation = var_45760_equation_0, values = (var_45270_cast_fp16, var_45670_cast_fp16))[name = string("op_45760_cast_fp16")]; + string var_45762_equation_0 = const()[name = string("op_45762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45762_cast_fp16 = einsum(equation = var_45762_equation_0, values = (var_45270_cast_fp16, var_45671_cast_fp16))[name = string("op_45762_cast_fp16")]; + string var_45764_equation_0 = const()[name = string("op_45764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45764_cast_fp16 = einsum(equation = var_45764_equation_0, values = (var_45270_cast_fp16, var_45672_cast_fp16))[name = string("op_45764_cast_fp16")]; + string var_45766_equation_0 = const()[name = string("op_45766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45766_cast_fp16 = einsum(equation = var_45766_equation_0, values = (var_45274_cast_fp16, var_45673_cast_fp16))[name = string("op_45766_cast_fp16")]; + string var_45768_equation_0 = const()[name = string("op_45768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45768_cast_fp16 = einsum(equation = var_45768_equation_0, values = (var_45274_cast_fp16, var_45674_cast_fp16))[name = string("op_45768_cast_fp16")]; + string var_45770_equation_0 = const()[name = string("op_45770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45770_cast_fp16 = einsum(equation = var_45770_equation_0, values = (var_45274_cast_fp16, var_45675_cast_fp16))[name = string("op_45770_cast_fp16")]; + string var_45772_equation_0 = const()[name = string("op_45772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45772_cast_fp16 = einsum(equation = var_45772_equation_0, values = (var_45274_cast_fp16, var_45676_cast_fp16))[name = string("op_45772_cast_fp16")]; + string var_45774_equation_0 = const()[name = string("op_45774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45774_cast_fp16 = einsum(equation = var_45774_equation_0, values = (var_45278_cast_fp16, var_45677_cast_fp16))[name = string("op_45774_cast_fp16")]; + string var_45776_equation_0 = const()[name = string("op_45776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45776_cast_fp16 = einsum(equation = var_45776_equation_0, values = (var_45278_cast_fp16, var_45678_cast_fp16))[name = string("op_45776_cast_fp16")]; + string var_45778_equation_0 = const()[name = string("op_45778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45778_cast_fp16 = einsum(equation = var_45778_equation_0, values = (var_45278_cast_fp16, var_45679_cast_fp16))[name = string("op_45778_cast_fp16")]; + string var_45780_equation_0 = const()[name = string("op_45780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45780_cast_fp16 = einsum(equation = var_45780_equation_0, values = (var_45278_cast_fp16, var_45680_cast_fp16))[name = string("op_45780_cast_fp16")]; + string var_45782_equation_0 = const()[name = string("op_45782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45782_cast_fp16 = einsum(equation = var_45782_equation_0, values = (var_45282_cast_fp16, var_45681_cast_fp16))[name = string("op_45782_cast_fp16")]; + string var_45784_equation_0 = const()[name = string("op_45784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45784_cast_fp16 = einsum(equation = var_45784_equation_0, values = (var_45282_cast_fp16, var_45682_cast_fp16))[name = string("op_45784_cast_fp16")]; + string var_45786_equation_0 = const()[name = string("op_45786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45786_cast_fp16 = einsum(equation = var_45786_equation_0, values = (var_45282_cast_fp16, var_45683_cast_fp16))[name = string("op_45786_cast_fp16")]; + string var_45788_equation_0 = const()[name = string("op_45788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45788_cast_fp16 = einsum(equation = var_45788_equation_0, values = (var_45282_cast_fp16, var_45684_cast_fp16))[name = string("op_45788_cast_fp16")]; + string var_45790_equation_0 = const()[name = string("op_45790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45790_cast_fp16 = einsum(equation = var_45790_equation_0, values = (var_45286_cast_fp16, var_45685_cast_fp16))[name = string("op_45790_cast_fp16")]; + string var_45792_equation_0 = const()[name = string("op_45792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45792_cast_fp16 = einsum(equation = var_45792_equation_0, values = (var_45286_cast_fp16, var_45686_cast_fp16))[name = string("op_45792_cast_fp16")]; + string var_45794_equation_0 = const()[name = string("op_45794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45794_cast_fp16 = einsum(equation = var_45794_equation_0, values = (var_45286_cast_fp16, var_45687_cast_fp16))[name = string("op_45794_cast_fp16")]; + string var_45796_equation_0 = const()[name = string("op_45796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45796_cast_fp16 = einsum(equation = var_45796_equation_0, values = (var_45286_cast_fp16, var_45688_cast_fp16))[name = string("op_45796_cast_fp16")]; + string var_45798_equation_0 = const()[name = string("op_45798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45798_cast_fp16 = einsum(equation = var_45798_equation_0, values = (var_45290_cast_fp16, var_45689_cast_fp16))[name = string("op_45798_cast_fp16")]; + string var_45800_equation_0 = const()[name = string("op_45800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45800_cast_fp16 = einsum(equation = var_45800_equation_0, values = (var_45290_cast_fp16, var_45690_cast_fp16))[name = string("op_45800_cast_fp16")]; + string var_45802_equation_0 = const()[name = string("op_45802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45802_cast_fp16 = einsum(equation = var_45802_equation_0, values = (var_45290_cast_fp16, var_45691_cast_fp16))[name = string("op_45802_cast_fp16")]; + string var_45804_equation_0 = const()[name = string("op_45804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45804_cast_fp16 = einsum(equation = var_45804_equation_0, values = (var_45290_cast_fp16, var_45692_cast_fp16))[name = string("op_45804_cast_fp16")]; + string var_45806_equation_0 = const()[name = string("op_45806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45806_cast_fp16 = einsum(equation = var_45806_equation_0, values = (var_45294_cast_fp16, var_45693_cast_fp16))[name = string("op_45806_cast_fp16")]; + string var_45808_equation_0 = const()[name = string("op_45808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45808_cast_fp16 = einsum(equation = var_45808_equation_0, values = (var_45294_cast_fp16, var_45694_cast_fp16))[name = string("op_45808_cast_fp16")]; + string var_45810_equation_0 = const()[name = string("op_45810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45810_cast_fp16 = einsum(equation = var_45810_equation_0, values = (var_45294_cast_fp16, var_45695_cast_fp16))[name = string("op_45810_cast_fp16")]; + string var_45812_equation_0 = const()[name = string("op_45812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45812_cast_fp16 = einsum(equation = var_45812_equation_0, values = (var_45294_cast_fp16, var_45696_cast_fp16))[name = string("op_45812_cast_fp16")]; + string var_45814_equation_0 = const()[name = string("op_45814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45814_cast_fp16 = einsum(equation = var_45814_equation_0, values = (var_45298_cast_fp16, var_45697_cast_fp16))[name = string("op_45814_cast_fp16")]; + string var_45816_equation_0 = const()[name = string("op_45816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45816_cast_fp16 = einsum(equation = var_45816_equation_0, values = (var_45298_cast_fp16, var_45698_cast_fp16))[name = string("op_45816_cast_fp16")]; + string var_45818_equation_0 = const()[name = string("op_45818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45818_cast_fp16 = einsum(equation = var_45818_equation_0, values = (var_45298_cast_fp16, var_45699_cast_fp16))[name = string("op_45818_cast_fp16")]; + string var_45820_equation_0 = const()[name = string("op_45820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45820_cast_fp16 = einsum(equation = var_45820_equation_0, values = (var_45298_cast_fp16, var_45700_cast_fp16))[name = string("op_45820_cast_fp16")]; + string var_45822_equation_0 = const()[name = string("op_45822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45822_cast_fp16 = einsum(equation = var_45822_equation_0, values = (var_45302_cast_fp16, var_45701_cast_fp16))[name = string("op_45822_cast_fp16")]; + string var_45824_equation_0 = const()[name = string("op_45824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45824_cast_fp16 = einsum(equation = var_45824_equation_0, values = (var_45302_cast_fp16, var_45702_cast_fp16))[name = string("op_45824_cast_fp16")]; + string var_45826_equation_0 = const()[name = string("op_45826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45826_cast_fp16 = einsum(equation = var_45826_equation_0, values = (var_45302_cast_fp16, var_45703_cast_fp16))[name = string("op_45826_cast_fp16")]; + string var_45828_equation_0 = const()[name = string("op_45828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45828_cast_fp16 = einsum(equation = var_45828_equation_0, values = (var_45302_cast_fp16, var_45704_cast_fp16))[name = string("op_45828_cast_fp16")]; + string var_45830_equation_0 = const()[name = string("op_45830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45830_cast_fp16 = einsum(equation = var_45830_equation_0, values = (var_45306_cast_fp16, var_45705_cast_fp16))[name = string("op_45830_cast_fp16")]; + string var_45832_equation_0 = const()[name = string("op_45832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45832_cast_fp16 = einsum(equation = var_45832_equation_0, values = (var_45306_cast_fp16, var_45706_cast_fp16))[name = string("op_45832_cast_fp16")]; + string var_45834_equation_0 = const()[name = string("op_45834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45834_cast_fp16 = einsum(equation = var_45834_equation_0, values = (var_45306_cast_fp16, var_45707_cast_fp16))[name = string("op_45834_cast_fp16")]; + string var_45836_equation_0 = const()[name = string("op_45836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45836_cast_fp16 = einsum(equation = var_45836_equation_0, values = (var_45306_cast_fp16, var_45708_cast_fp16))[name = string("op_45836_cast_fp16")]; + string var_45838_equation_0 = const()[name = string("op_45838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45838_cast_fp16 = einsum(equation = var_45838_equation_0, values = (var_45310_cast_fp16, var_45709_cast_fp16))[name = string("op_45838_cast_fp16")]; + string var_45840_equation_0 = const()[name = string("op_45840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45840_cast_fp16 = einsum(equation = var_45840_equation_0, values = (var_45310_cast_fp16, var_45710_cast_fp16))[name = string("op_45840_cast_fp16")]; + string var_45842_equation_0 = const()[name = string("op_45842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45842_cast_fp16 = einsum(equation = var_45842_equation_0, values = (var_45310_cast_fp16, var_45711_cast_fp16))[name = string("op_45842_cast_fp16")]; + string var_45844_equation_0 = const()[name = string("op_45844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45844_cast_fp16 = einsum(equation = var_45844_equation_0, values = (var_45310_cast_fp16, var_45712_cast_fp16))[name = string("op_45844_cast_fp16")]; + string var_45846_equation_0 = const()[name = string("op_45846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45846_cast_fp16 = einsum(equation = var_45846_equation_0, values = (var_45314_cast_fp16, var_45713_cast_fp16))[name = string("op_45846_cast_fp16")]; + string var_45848_equation_0 = const()[name = string("op_45848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45848_cast_fp16 = einsum(equation = var_45848_equation_0, values = (var_45314_cast_fp16, var_45714_cast_fp16))[name = string("op_45848_cast_fp16")]; + string var_45850_equation_0 = const()[name = string("op_45850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45850_cast_fp16 = einsum(equation = var_45850_equation_0, values = (var_45314_cast_fp16, var_45715_cast_fp16))[name = string("op_45850_cast_fp16")]; + string var_45852_equation_0 = const()[name = string("op_45852_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45852_cast_fp16 = einsum(equation = var_45852_equation_0, values = (var_45314_cast_fp16, var_45716_cast_fp16))[name = string("op_45852_cast_fp16")]; + string var_45854_equation_0 = const()[name = string("op_45854_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45854_cast_fp16 = einsum(equation = var_45854_equation_0, values = (var_45318_cast_fp16, var_45717_cast_fp16))[name = string("op_45854_cast_fp16")]; + string var_45856_equation_0 = const()[name = string("op_45856_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45856_cast_fp16 = einsum(equation = var_45856_equation_0, values = (var_45318_cast_fp16, var_45718_cast_fp16))[name = string("op_45856_cast_fp16")]; + string var_45858_equation_0 = const()[name = string("op_45858_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45858_cast_fp16 = einsum(equation = var_45858_equation_0, values = (var_45318_cast_fp16, var_45719_cast_fp16))[name = string("op_45858_cast_fp16")]; + string var_45860_equation_0 = const()[name = string("op_45860_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45860_cast_fp16 = einsum(equation = var_45860_equation_0, values = (var_45318_cast_fp16, var_45720_cast_fp16))[name = string("op_45860_cast_fp16")]; + string var_45862_equation_0 = const()[name = string("op_45862_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45862_cast_fp16 = einsum(equation = var_45862_equation_0, values = (var_45322_cast_fp16, var_45721_cast_fp16))[name = string("op_45862_cast_fp16")]; + string var_45864_equation_0 = const()[name = string("op_45864_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45864_cast_fp16 = einsum(equation = var_45864_equation_0, values = (var_45322_cast_fp16, var_45722_cast_fp16))[name = string("op_45864_cast_fp16")]; + string var_45866_equation_0 = const()[name = string("op_45866_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45866_cast_fp16 = einsum(equation = var_45866_equation_0, values = (var_45322_cast_fp16, var_45723_cast_fp16))[name = string("op_45866_cast_fp16")]; + string var_45868_equation_0 = const()[name = string("op_45868_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45868_cast_fp16 = einsum(equation = var_45868_equation_0, values = (var_45322_cast_fp16, var_45724_cast_fp16))[name = string("op_45868_cast_fp16")]; + string var_45870_equation_0 = const()[name = string("op_45870_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45870_cast_fp16 = einsum(equation = var_45870_equation_0, values = (var_45326_cast_fp16, var_45725_cast_fp16))[name = string("op_45870_cast_fp16")]; + string var_45872_equation_0 = const()[name = string("op_45872_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45872_cast_fp16 = einsum(equation = var_45872_equation_0, values = (var_45326_cast_fp16, var_45726_cast_fp16))[name = string("op_45872_cast_fp16")]; + string var_45874_equation_0 = const()[name = string("op_45874_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45874_cast_fp16 = einsum(equation = var_45874_equation_0, values = (var_45326_cast_fp16, var_45727_cast_fp16))[name = string("op_45874_cast_fp16")]; + string var_45876_equation_0 = const()[name = string("op_45876_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45876_cast_fp16 = einsum(equation = var_45876_equation_0, values = (var_45326_cast_fp16, var_45728_cast_fp16))[name = string("op_45876_cast_fp16")]; + string var_45878_equation_0 = const()[name = string("op_45878_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45878_cast_fp16 = einsum(equation = var_45878_equation_0, values = (var_45330_cast_fp16, var_45729_cast_fp16))[name = string("op_45878_cast_fp16")]; + string var_45880_equation_0 = const()[name = string("op_45880_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45880_cast_fp16 = einsum(equation = var_45880_equation_0, values = (var_45330_cast_fp16, var_45730_cast_fp16))[name = string("op_45880_cast_fp16")]; + string var_45882_equation_0 = const()[name = string("op_45882_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45882_cast_fp16 = einsum(equation = var_45882_equation_0, values = (var_45330_cast_fp16, var_45731_cast_fp16))[name = string("op_45882_cast_fp16")]; + string var_45884_equation_0 = const()[name = string("op_45884_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45884_cast_fp16 = einsum(equation = var_45884_equation_0, values = (var_45330_cast_fp16, var_45732_cast_fp16))[name = string("op_45884_cast_fp16")]; + string var_45886_equation_0 = const()[name = string("op_45886_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45886_cast_fp16 = einsum(equation = var_45886_equation_0, values = (var_45334_cast_fp16, var_45733_cast_fp16))[name = string("op_45886_cast_fp16")]; + string var_45888_equation_0 = const()[name = string("op_45888_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45888_cast_fp16 = einsum(equation = var_45888_equation_0, values = (var_45334_cast_fp16, var_45734_cast_fp16))[name = string("op_45888_cast_fp16")]; + string var_45890_equation_0 = const()[name = string("op_45890_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45890_cast_fp16 = einsum(equation = var_45890_equation_0, values = (var_45334_cast_fp16, var_45735_cast_fp16))[name = string("op_45890_cast_fp16")]; + string var_45892_equation_0 = const()[name = string("op_45892_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45892_cast_fp16 = einsum(equation = var_45892_equation_0, values = (var_45334_cast_fp16, var_45736_cast_fp16))[name = string("op_45892_cast_fp16")]; + string var_45894_equation_0 = const()[name = string("op_45894_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45894_cast_fp16 = einsum(equation = var_45894_equation_0, values = (var_45338_cast_fp16, var_45737_cast_fp16))[name = string("op_45894_cast_fp16")]; + string var_45896_equation_0 = const()[name = string("op_45896_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45896_cast_fp16 = einsum(equation = var_45896_equation_0, values = (var_45338_cast_fp16, var_45738_cast_fp16))[name = string("op_45896_cast_fp16")]; + string var_45898_equation_0 = const()[name = string("op_45898_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45898_cast_fp16 = einsum(equation = var_45898_equation_0, values = (var_45338_cast_fp16, var_45739_cast_fp16))[name = string("op_45898_cast_fp16")]; + string var_45900_equation_0 = const()[name = string("op_45900_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45900_cast_fp16 = einsum(equation = var_45900_equation_0, values = (var_45338_cast_fp16, var_45740_cast_fp16))[name = string("op_45900_cast_fp16")]; + bool var_45902_interleave_0 = const()[name = string("op_45902_interleave_0"), val = bool(false)]; + tensor var_45902_cast_fp16 = concat(axis = var_44461, interleave = var_45902_interleave_0, values = (var_45742_cast_fp16, var_45744_cast_fp16, var_45746_cast_fp16, var_45748_cast_fp16))[name = string("op_45902_cast_fp16")]; + bool var_45904_interleave_0 = const()[name = string("op_45904_interleave_0"), val = bool(false)]; + tensor var_45904_cast_fp16 = concat(axis = var_44461, interleave = var_45904_interleave_0, values = (var_45750_cast_fp16, var_45752_cast_fp16, var_45754_cast_fp16, var_45756_cast_fp16))[name = string("op_45904_cast_fp16")]; + bool var_45906_interleave_0 = const()[name = string("op_45906_interleave_0"), val = bool(false)]; + tensor var_45906_cast_fp16 = concat(axis = var_44461, interleave = var_45906_interleave_0, values = (var_45758_cast_fp16, var_45760_cast_fp16, var_45762_cast_fp16, var_45764_cast_fp16))[name = string("op_45906_cast_fp16")]; + bool var_45908_interleave_0 = const()[name = string("op_45908_interleave_0"), val = bool(false)]; + tensor var_45908_cast_fp16 = concat(axis = var_44461, interleave = var_45908_interleave_0, values = (var_45766_cast_fp16, var_45768_cast_fp16, var_45770_cast_fp16, var_45772_cast_fp16))[name = string("op_45908_cast_fp16")]; + bool var_45910_interleave_0 = const()[name = string("op_45910_interleave_0"), val = bool(false)]; + tensor var_45910_cast_fp16 = concat(axis = var_44461, interleave = var_45910_interleave_0, values = (var_45774_cast_fp16, var_45776_cast_fp16, var_45778_cast_fp16, var_45780_cast_fp16))[name = string("op_45910_cast_fp16")]; + bool var_45912_interleave_0 = const()[name = string("op_45912_interleave_0"), val = bool(false)]; + tensor var_45912_cast_fp16 = concat(axis = var_44461, interleave = var_45912_interleave_0, values = (var_45782_cast_fp16, var_45784_cast_fp16, var_45786_cast_fp16, var_45788_cast_fp16))[name = string("op_45912_cast_fp16")]; + bool var_45914_interleave_0 = const()[name = string("op_45914_interleave_0"), val = bool(false)]; + tensor var_45914_cast_fp16 = concat(axis = var_44461, interleave = var_45914_interleave_0, values = (var_45790_cast_fp16, var_45792_cast_fp16, var_45794_cast_fp16, var_45796_cast_fp16))[name = string("op_45914_cast_fp16")]; + bool var_45916_interleave_0 = const()[name = string("op_45916_interleave_0"), val = bool(false)]; + tensor var_45916_cast_fp16 = concat(axis = var_44461, interleave = var_45916_interleave_0, values = (var_45798_cast_fp16, var_45800_cast_fp16, var_45802_cast_fp16, var_45804_cast_fp16))[name = string("op_45916_cast_fp16")]; + bool var_45918_interleave_0 = const()[name = string("op_45918_interleave_0"), val = bool(false)]; + tensor var_45918_cast_fp16 = concat(axis = var_44461, interleave = var_45918_interleave_0, values = (var_45806_cast_fp16, var_45808_cast_fp16, var_45810_cast_fp16, var_45812_cast_fp16))[name = string("op_45918_cast_fp16")]; + bool var_45920_interleave_0 = const()[name = string("op_45920_interleave_0"), val = bool(false)]; + tensor var_45920_cast_fp16 = concat(axis = var_44461, interleave = var_45920_interleave_0, values = (var_45814_cast_fp16, var_45816_cast_fp16, var_45818_cast_fp16, var_45820_cast_fp16))[name = string("op_45920_cast_fp16")]; + bool var_45922_interleave_0 = const()[name = string("op_45922_interleave_0"), val = bool(false)]; + tensor var_45922_cast_fp16 = concat(axis = var_44461, interleave = var_45922_interleave_0, values = (var_45822_cast_fp16, var_45824_cast_fp16, var_45826_cast_fp16, var_45828_cast_fp16))[name = string("op_45922_cast_fp16")]; + bool var_45924_interleave_0 = const()[name = string("op_45924_interleave_0"), val = bool(false)]; + tensor var_45924_cast_fp16 = concat(axis = var_44461, interleave = var_45924_interleave_0, values = (var_45830_cast_fp16, var_45832_cast_fp16, var_45834_cast_fp16, var_45836_cast_fp16))[name = string("op_45924_cast_fp16")]; + bool var_45926_interleave_0 = const()[name = string("op_45926_interleave_0"), val = bool(false)]; + tensor var_45926_cast_fp16 = concat(axis = var_44461, interleave = var_45926_interleave_0, values = (var_45838_cast_fp16, var_45840_cast_fp16, var_45842_cast_fp16, var_45844_cast_fp16))[name = string("op_45926_cast_fp16")]; + bool var_45928_interleave_0 = const()[name = string("op_45928_interleave_0"), val = bool(false)]; + tensor var_45928_cast_fp16 = concat(axis = var_44461, interleave = var_45928_interleave_0, values = (var_45846_cast_fp16, var_45848_cast_fp16, var_45850_cast_fp16, var_45852_cast_fp16))[name = string("op_45928_cast_fp16")]; + bool var_45930_interleave_0 = const()[name = string("op_45930_interleave_0"), val = bool(false)]; + tensor var_45930_cast_fp16 = concat(axis = var_44461, interleave = var_45930_interleave_0, values = (var_45854_cast_fp16, var_45856_cast_fp16, var_45858_cast_fp16, var_45860_cast_fp16))[name = string("op_45930_cast_fp16")]; + bool var_45932_interleave_0 = const()[name = string("op_45932_interleave_0"), val = bool(false)]; + tensor var_45932_cast_fp16 = concat(axis = var_44461, interleave = var_45932_interleave_0, values = (var_45862_cast_fp16, var_45864_cast_fp16, var_45866_cast_fp16, var_45868_cast_fp16))[name = string("op_45932_cast_fp16")]; + bool var_45934_interleave_0 = const()[name = string("op_45934_interleave_0"), val = bool(false)]; + tensor var_45934_cast_fp16 = concat(axis = var_44461, interleave = var_45934_interleave_0, values = (var_45870_cast_fp16, var_45872_cast_fp16, var_45874_cast_fp16, var_45876_cast_fp16))[name = string("op_45934_cast_fp16")]; + bool var_45936_interleave_0 = const()[name = string("op_45936_interleave_0"), val = bool(false)]; + tensor var_45936_cast_fp16 = concat(axis = var_44461, interleave = var_45936_interleave_0, values = (var_45878_cast_fp16, var_45880_cast_fp16, var_45882_cast_fp16, var_45884_cast_fp16))[name = string("op_45936_cast_fp16")]; + bool var_45938_interleave_0 = const()[name = string("op_45938_interleave_0"), val = bool(false)]; + tensor var_45938_cast_fp16 = concat(axis = var_44461, interleave = var_45938_interleave_0, values = (var_45886_cast_fp16, var_45888_cast_fp16, var_45890_cast_fp16, var_45892_cast_fp16))[name = string("op_45938_cast_fp16")]; + bool var_45940_interleave_0 = const()[name = string("op_45940_interleave_0"), val = bool(false)]; + tensor var_45940_cast_fp16 = concat(axis = var_44461, interleave = var_45940_interleave_0, values = (var_45894_cast_fp16, var_45896_cast_fp16, var_45898_cast_fp16, var_45900_cast_fp16))[name = string("op_45940_cast_fp16")]; + bool input_233_interleave_0 = const()[name = string("input_233_interleave_0"), val = bool(false)]; + tensor input_233_cast_fp16 = concat(axis = var_44486, interleave = input_233_interleave_0, values = (var_45902_cast_fp16, var_45904_cast_fp16, var_45906_cast_fp16, var_45908_cast_fp16, var_45910_cast_fp16, var_45912_cast_fp16, var_45914_cast_fp16, var_45916_cast_fp16, var_45918_cast_fp16, var_45920_cast_fp16, var_45922_cast_fp16, var_45924_cast_fp16, var_45926_cast_fp16, var_45928_cast_fp16, var_45930_cast_fp16, var_45932_cast_fp16, var_45934_cast_fp16, var_45936_cast_fp16, var_45938_cast_fp16, var_45940_cast_fp16))[name = string("input_233_cast_fp16")]; + string obj_119_pad_type_0 = const()[name = string("obj_119_pad_type_0"), val = string("valid")]; + tensor obj_119_strides_0 = const()[name = string("obj_119_strides_0"), val = tensor([1, 1])]; + tensor obj_119_pad_0 = const()[name = string("obj_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_119_dilations_0 = const()[name = string("obj_119_dilations_0"), val = tensor([1, 1])]; + int32 obj_119_groups_0 = const()[name = string("obj_119_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165750400)))]; + tensor layers_29_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169027264)))]; + tensor obj_119_cast_fp16 = conv(bias = layers_29_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_29_self_attn_o_proj_weight_to_fp16, x = input_233_cast_fp16)[name = string("obj_119_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_119_cast_fp16)[name = string("inputs_119_cast_fp16")]; + tensor out_119_axes_0 = const()[name = string("out_119_axes_0"), val = tensor([1])]; + fp16 var_45959_to_fp16 = const()[name = string("op_45959_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_45959_to_fp16, x = inputs_119_cast_fp16)[name = string("out_119_cast_fp16")]; + tensor input_235_gamma_0_to_fp16 = const()[name = string("input_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169029888)))]; + tensor input_235_beta_0_to_fp16 = const()[name = string("input_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169032512)))]; + fp16 input_235_epsilon_0_to_fp16 = const()[name = string("input_235_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = string("input_235_cast_fp16")]; + string input_237_pad_type_0 = const()[name = string("input_237_pad_type_0"), val = string("valid")]; + tensor input_237_strides_0 = const()[name = string("input_237_strides_0"), val = tensor([1, 1])]; + tensor input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_237_dilations_0 = const()[name = string("input_237_dilations_0"), val = tensor([1, 1])]; + int32 input_237_groups_0 = const()[name = string("input_237_groups_0"), val = int32(1)]; + tensor layers_29_fc1_weight_to_fp16 = const()[name = string("layers_29_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169035136)))]; + tensor layers_29_fc1_bias_to_fp16 = const()[name = string("layers_29_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182142400)))]; + tensor input_237_cast_fp16 = conv(bias = layers_29_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_29_fc1_weight_to_fp16, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")]; + string input_239_mode_0 = const()[name = string("input_239_mode_0"), val = string("EXACT")]; + tensor input_239_cast_fp16 = gelu(mode = input_239_mode_0, x = input_237_cast_fp16)[name = string("input_239_cast_fp16")]; + string hidden_states_63_pad_type_0 = const()[name = string("hidden_states_63_pad_type_0"), val = string("valid")]; + tensor hidden_states_63_strides_0 = const()[name = string("hidden_states_63_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_63_pad_0 = const()[name = string("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_63_dilations_0 = const()[name = string("hidden_states_63_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_63_groups_0 = const()[name = string("hidden_states_63_groups_0"), val = int32(1)]; + tensor layers_29_fc2_weight_to_fp16 = const()[name = string("layers_29_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182152704)))]; + tensor layers_29_fc2_bias_to_fp16 = const()[name = string("layers_29_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195259968)))]; + tensor hidden_states_63_cast_fp16 = conv(bias = layers_29_fc2_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = layers_29_fc2_weight_to_fp16, x = input_239_cast_fp16)[name = string("hidden_states_63_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_63_cast_fp16)[name = string("inputs_121_cast_fp16")]; + int32 var_45988 = const()[name = string("op_45988"), val = int32(3)]; + int32 var_46013 = const()[name = string("op_46013"), val = int32(1)]; + tensor out_121_axes_0 = const()[name = string("out_121_axes_0"), val = tensor([1])]; + fp16 var_46030_to_fp16 = const()[name = string("op_46030_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_46030_to_fp16, x = inputs_121_cast_fp16)[name = string("out_121_cast_fp16")]; + tensor obj_121_gamma_0_to_fp16 = const()[name = string("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195262592)))]; + tensor obj_121_beta_0_to_fp16 = const()[name = string("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195265216)))]; + fp16 obj_121_epsilon_0_to_fp16 = const()[name = string("obj_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = string("obj_121_cast_fp16")]; + string query_61_pad_type_0 = const()[name = string("query_61_pad_type_0"), val = string("valid")]; + tensor query_61_strides_0 = const()[name = string("query_61_strides_0"), val = tensor([1, 1])]; + tensor query_61_pad_0 = const()[name = string("query_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_61_dilations_0 = const()[name = string("query_61_dilations_0"), val = tensor([1, 1])]; + int32 query_61_groups_0 = const()[name = string("query_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195267840)))]; + tensor layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198544704)))]; + tensor query_61_cast_fp16 = conv(bias = layers_30_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_30_self_attn_q_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("query_61_cast_fp16")]; + string key_61_pad_type_0 = const()[name = string("key_61_pad_type_0"), val = string("valid")]; + tensor key_61_strides_0 = const()[name = string("key_61_strides_0"), val = tensor([1, 1])]; + tensor key_61_pad_0 = const()[name = string("key_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_61_dilations_0 = const()[name = string("key_61_dilations_0"), val = tensor([1, 1])]; + int32 key_61_groups_0 = const()[name = string("key_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198547328)))]; + tensor key_61_cast_fp16 = conv(dilations = key_61_dilations_0, groups = key_61_groups_0, pad = key_61_pad_0, pad_type = key_61_pad_type_0, strides = key_61_strides_0, weight = layers_30_self_attn_k_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("key_61_cast_fp16")]; + string value_61_pad_type_0 = const()[name = string("value_61_pad_type_0"), val = string("valid")]; + tensor value_61_strides_0 = const()[name = string("value_61_strides_0"), val = tensor([1, 1])]; + tensor value_61_pad_0 = const()[name = string("value_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_61_dilations_0 = const()[name = string("value_61_dilations_0"), val = tensor([1, 1])]; + int32 value_61_groups_0 = const()[name = string("value_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1201824192)))]; + tensor layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205101056)))]; + tensor value_61_cast_fp16 = conv(bias = layers_30_self_attn_v_proj_bias_to_fp16, dilations = value_61_dilations_0, groups = value_61_groups_0, pad = value_61_pad_0, pad_type = value_61_pad_type_0, strides = value_61_strides_0, weight = layers_30_self_attn_v_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("value_61_cast_fp16")]; + tensor var_46068_begin_0 = const()[name = string("op_46068_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46068_end_0 = const()[name = string("op_46068_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46068_end_mask_0 = const()[name = string("op_46068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46068_cast_fp16 = slice_by_index(begin = var_46068_begin_0, end = var_46068_end_0, end_mask = var_46068_end_mask_0, x = query_61_cast_fp16)[name = string("op_46068_cast_fp16")]; + tensor var_46072_begin_0 = const()[name = string("op_46072_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46072_end_0 = const()[name = string("op_46072_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_46072_end_mask_0 = const()[name = string("op_46072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46072_cast_fp16 = slice_by_index(begin = var_46072_begin_0, end = var_46072_end_0, end_mask = var_46072_end_mask_0, x = query_61_cast_fp16)[name = string("op_46072_cast_fp16")]; + tensor var_46076_begin_0 = const()[name = string("op_46076_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46076_end_0 = const()[name = string("op_46076_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_46076_end_mask_0 = const()[name = string("op_46076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46076_cast_fp16 = slice_by_index(begin = var_46076_begin_0, end = var_46076_end_0, end_mask = var_46076_end_mask_0, x = query_61_cast_fp16)[name = string("op_46076_cast_fp16")]; + tensor var_46080_begin_0 = const()[name = string("op_46080_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46080_end_0 = const()[name = string("op_46080_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_46080_end_mask_0 = const()[name = string("op_46080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46080_cast_fp16 = slice_by_index(begin = var_46080_begin_0, end = var_46080_end_0, end_mask = var_46080_end_mask_0, x = query_61_cast_fp16)[name = string("op_46080_cast_fp16")]; + tensor var_46084_begin_0 = const()[name = string("op_46084_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46084_end_0 = const()[name = string("op_46084_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_46084_end_mask_0 = const()[name = string("op_46084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46084_cast_fp16 = slice_by_index(begin = var_46084_begin_0, end = var_46084_end_0, end_mask = var_46084_end_mask_0, x = query_61_cast_fp16)[name = string("op_46084_cast_fp16")]; + tensor var_46088_begin_0 = const()[name = string("op_46088_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46088_end_0 = const()[name = string("op_46088_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_46088_end_mask_0 = const()[name = string("op_46088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46088_cast_fp16 = slice_by_index(begin = var_46088_begin_0, end = var_46088_end_0, end_mask = var_46088_end_mask_0, x = query_61_cast_fp16)[name = string("op_46088_cast_fp16")]; + tensor var_46092_begin_0 = const()[name = string("op_46092_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46092_end_0 = const()[name = string("op_46092_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_46092_end_mask_0 = const()[name = string("op_46092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46092_cast_fp16 = slice_by_index(begin = var_46092_begin_0, end = var_46092_end_0, end_mask = var_46092_end_mask_0, x = query_61_cast_fp16)[name = string("op_46092_cast_fp16")]; + tensor var_46096_begin_0 = const()[name = string("op_46096_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46096_end_0 = const()[name = string("op_46096_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_46096_end_mask_0 = const()[name = string("op_46096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46096_cast_fp16 = slice_by_index(begin = var_46096_begin_0, end = var_46096_end_0, end_mask = var_46096_end_mask_0, x = query_61_cast_fp16)[name = string("op_46096_cast_fp16")]; + tensor var_46100_begin_0 = const()[name = string("op_46100_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46100_end_0 = const()[name = string("op_46100_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_46100_end_mask_0 = const()[name = string("op_46100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46100_cast_fp16 = slice_by_index(begin = var_46100_begin_0, end = var_46100_end_0, end_mask = var_46100_end_mask_0, x = query_61_cast_fp16)[name = string("op_46100_cast_fp16")]; + tensor var_46104_begin_0 = const()[name = string("op_46104_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46104_end_0 = const()[name = string("op_46104_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_46104_end_mask_0 = const()[name = string("op_46104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46104_cast_fp16 = slice_by_index(begin = var_46104_begin_0, end = var_46104_end_0, end_mask = var_46104_end_mask_0, x = query_61_cast_fp16)[name = string("op_46104_cast_fp16")]; + tensor var_46108_begin_0 = const()[name = string("op_46108_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46108_end_0 = const()[name = string("op_46108_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_46108_end_mask_0 = const()[name = string("op_46108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46108_cast_fp16 = slice_by_index(begin = var_46108_begin_0, end = var_46108_end_0, end_mask = var_46108_end_mask_0, x = query_61_cast_fp16)[name = string("op_46108_cast_fp16")]; + tensor var_46112_begin_0 = const()[name = string("op_46112_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46112_end_0 = const()[name = string("op_46112_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_46112_end_mask_0 = const()[name = string("op_46112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46112_cast_fp16 = slice_by_index(begin = var_46112_begin_0, end = var_46112_end_0, end_mask = var_46112_end_mask_0, x = query_61_cast_fp16)[name = string("op_46112_cast_fp16")]; + tensor var_46116_begin_0 = const()[name = string("op_46116_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46116_end_0 = const()[name = string("op_46116_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_46116_end_mask_0 = const()[name = string("op_46116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46116_cast_fp16 = slice_by_index(begin = var_46116_begin_0, end = var_46116_end_0, end_mask = var_46116_end_mask_0, x = query_61_cast_fp16)[name = string("op_46116_cast_fp16")]; + tensor var_46120_begin_0 = const()[name = string("op_46120_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46120_end_0 = const()[name = string("op_46120_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_46120_end_mask_0 = const()[name = string("op_46120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46120_cast_fp16 = slice_by_index(begin = var_46120_begin_0, end = var_46120_end_0, end_mask = var_46120_end_mask_0, x = query_61_cast_fp16)[name = string("op_46120_cast_fp16")]; + tensor var_46124_begin_0 = const()[name = string("op_46124_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46124_end_0 = const()[name = string("op_46124_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_46124_end_mask_0 = const()[name = string("op_46124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46124_cast_fp16 = slice_by_index(begin = var_46124_begin_0, end = var_46124_end_0, end_mask = var_46124_end_mask_0, x = query_61_cast_fp16)[name = string("op_46124_cast_fp16")]; + tensor var_46128_begin_0 = const()[name = string("op_46128_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46128_end_0 = const()[name = string("op_46128_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_46128_end_mask_0 = const()[name = string("op_46128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46128_cast_fp16 = slice_by_index(begin = var_46128_begin_0, end = var_46128_end_0, end_mask = var_46128_end_mask_0, x = query_61_cast_fp16)[name = string("op_46128_cast_fp16")]; + tensor var_46132_begin_0 = const()[name = string("op_46132_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46132_end_0 = const()[name = string("op_46132_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_46132_end_mask_0 = const()[name = string("op_46132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46132_cast_fp16 = slice_by_index(begin = var_46132_begin_0, end = var_46132_end_0, end_mask = var_46132_end_mask_0, x = query_61_cast_fp16)[name = string("op_46132_cast_fp16")]; + tensor var_46136_begin_0 = const()[name = string("op_46136_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46136_end_0 = const()[name = string("op_46136_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_46136_end_mask_0 = const()[name = string("op_46136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46136_cast_fp16 = slice_by_index(begin = var_46136_begin_0, end = var_46136_end_0, end_mask = var_46136_end_mask_0, x = query_61_cast_fp16)[name = string("op_46136_cast_fp16")]; + tensor var_46140_begin_0 = const()[name = string("op_46140_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46140_end_0 = const()[name = string("op_46140_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_46140_end_mask_0 = const()[name = string("op_46140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46140_cast_fp16 = slice_by_index(begin = var_46140_begin_0, end = var_46140_end_0, end_mask = var_46140_end_mask_0, x = query_61_cast_fp16)[name = string("op_46140_cast_fp16")]; + tensor var_46144_begin_0 = const()[name = string("op_46144_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46144_end_0 = const()[name = string("op_46144_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_46144_end_mask_0 = const()[name = string("op_46144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46144_cast_fp16 = slice_by_index(begin = var_46144_begin_0, end = var_46144_end_0, end_mask = var_46144_end_mask_0, x = query_61_cast_fp16)[name = string("op_46144_cast_fp16")]; + tensor var_46153_begin_0 = const()[name = string("op_46153_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46153_end_0 = const()[name = string("op_46153_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46153_end_mask_0 = const()[name = string("op_46153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46153_cast_fp16 = slice_by_index(begin = var_46153_begin_0, end = var_46153_end_0, end_mask = var_46153_end_mask_0, x = var_46068_cast_fp16)[name = string("op_46153_cast_fp16")]; + tensor var_46160_begin_0 = const()[name = string("op_46160_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46160_end_0 = const()[name = string("op_46160_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46160_end_mask_0 = const()[name = string("op_46160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46160_cast_fp16 = slice_by_index(begin = var_46160_begin_0, end = var_46160_end_0, end_mask = var_46160_end_mask_0, x = var_46068_cast_fp16)[name = string("op_46160_cast_fp16")]; + tensor var_46167_begin_0 = const()[name = string("op_46167_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46167_end_0 = const()[name = string("op_46167_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46167_end_mask_0 = const()[name = string("op_46167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46167_cast_fp16 = slice_by_index(begin = var_46167_begin_0, end = var_46167_end_0, end_mask = var_46167_end_mask_0, x = var_46068_cast_fp16)[name = string("op_46167_cast_fp16")]; + tensor var_46174_begin_0 = const()[name = string("op_46174_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46174_end_0 = const()[name = string("op_46174_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46174_end_mask_0 = const()[name = string("op_46174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46174_cast_fp16 = slice_by_index(begin = var_46174_begin_0, end = var_46174_end_0, end_mask = var_46174_end_mask_0, x = var_46068_cast_fp16)[name = string("op_46174_cast_fp16")]; + tensor var_46181_begin_0 = const()[name = string("op_46181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46181_end_0 = const()[name = string("op_46181_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46181_end_mask_0 = const()[name = string("op_46181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46181_cast_fp16 = slice_by_index(begin = var_46181_begin_0, end = var_46181_end_0, end_mask = var_46181_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46181_cast_fp16")]; + tensor var_46188_begin_0 = const()[name = string("op_46188_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46188_end_0 = const()[name = string("op_46188_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46188_end_mask_0 = const()[name = string("op_46188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46188_cast_fp16 = slice_by_index(begin = var_46188_begin_0, end = var_46188_end_0, end_mask = var_46188_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46188_cast_fp16")]; + tensor var_46195_begin_0 = const()[name = string("op_46195_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46195_end_0 = const()[name = string("op_46195_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46195_end_mask_0 = const()[name = string("op_46195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46195_cast_fp16 = slice_by_index(begin = var_46195_begin_0, end = var_46195_end_0, end_mask = var_46195_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46195_cast_fp16")]; + tensor var_46202_begin_0 = const()[name = string("op_46202_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46202_end_0 = const()[name = string("op_46202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46202_end_mask_0 = const()[name = string("op_46202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46202_cast_fp16 = slice_by_index(begin = var_46202_begin_0, end = var_46202_end_0, end_mask = var_46202_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46202_cast_fp16")]; + tensor var_46209_begin_0 = const()[name = string("op_46209_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46209_end_0 = const()[name = string("op_46209_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46209_end_mask_0 = const()[name = string("op_46209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46209_cast_fp16 = slice_by_index(begin = var_46209_begin_0, end = var_46209_end_0, end_mask = var_46209_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46209_cast_fp16")]; + tensor var_46216_begin_0 = const()[name = string("op_46216_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46216_end_0 = const()[name = string("op_46216_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46216_end_mask_0 = const()[name = string("op_46216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46216_cast_fp16 = slice_by_index(begin = var_46216_begin_0, end = var_46216_end_0, end_mask = var_46216_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46216_cast_fp16")]; + tensor var_46223_begin_0 = const()[name = string("op_46223_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46223_end_0 = const()[name = string("op_46223_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46223_end_mask_0 = const()[name = string("op_46223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46223_cast_fp16 = slice_by_index(begin = var_46223_begin_0, end = var_46223_end_0, end_mask = var_46223_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46223_cast_fp16")]; + tensor var_46230_begin_0 = const()[name = string("op_46230_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46230_end_0 = const()[name = string("op_46230_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46230_end_mask_0 = const()[name = string("op_46230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46230_cast_fp16 = slice_by_index(begin = var_46230_begin_0, end = var_46230_end_0, end_mask = var_46230_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46230_cast_fp16")]; + tensor var_46237_begin_0 = const()[name = string("op_46237_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46237_end_0 = const()[name = string("op_46237_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46237_end_mask_0 = const()[name = string("op_46237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46237_cast_fp16 = slice_by_index(begin = var_46237_begin_0, end = var_46237_end_0, end_mask = var_46237_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46237_cast_fp16")]; + tensor var_46244_begin_0 = const()[name = string("op_46244_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46244_end_0 = const()[name = string("op_46244_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46244_end_mask_0 = const()[name = string("op_46244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46244_cast_fp16 = slice_by_index(begin = var_46244_begin_0, end = var_46244_end_0, end_mask = var_46244_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46244_cast_fp16")]; + tensor var_46251_begin_0 = const()[name = string("op_46251_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46251_end_0 = const()[name = string("op_46251_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46251_end_mask_0 = const()[name = string("op_46251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46251_cast_fp16 = slice_by_index(begin = var_46251_begin_0, end = var_46251_end_0, end_mask = var_46251_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46251_cast_fp16")]; + tensor var_46258_begin_0 = const()[name = string("op_46258_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46258_end_0 = const()[name = string("op_46258_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46258_end_mask_0 = const()[name = string("op_46258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46258_cast_fp16 = slice_by_index(begin = var_46258_begin_0, end = var_46258_end_0, end_mask = var_46258_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46258_cast_fp16")]; + tensor var_46265_begin_0 = const()[name = string("op_46265_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46265_end_0 = const()[name = string("op_46265_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46265_end_mask_0 = const()[name = string("op_46265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46265_cast_fp16 = slice_by_index(begin = var_46265_begin_0, end = var_46265_end_0, end_mask = var_46265_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46265_cast_fp16")]; + tensor var_46272_begin_0 = const()[name = string("op_46272_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46272_end_0 = const()[name = string("op_46272_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46272_end_mask_0 = const()[name = string("op_46272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46272_cast_fp16 = slice_by_index(begin = var_46272_begin_0, end = var_46272_end_0, end_mask = var_46272_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46272_cast_fp16")]; + tensor var_46279_begin_0 = const()[name = string("op_46279_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46279_end_0 = const()[name = string("op_46279_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46279_end_mask_0 = const()[name = string("op_46279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46279_cast_fp16 = slice_by_index(begin = var_46279_begin_0, end = var_46279_end_0, end_mask = var_46279_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46279_cast_fp16")]; + tensor var_46286_begin_0 = const()[name = string("op_46286_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46286_end_0 = const()[name = string("op_46286_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46286_end_mask_0 = const()[name = string("op_46286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46286_cast_fp16 = slice_by_index(begin = var_46286_begin_0, end = var_46286_end_0, end_mask = var_46286_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46286_cast_fp16")]; + tensor var_46293_begin_0 = const()[name = string("op_46293_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46293_end_0 = const()[name = string("op_46293_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46293_end_mask_0 = const()[name = string("op_46293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46293_cast_fp16 = slice_by_index(begin = var_46293_begin_0, end = var_46293_end_0, end_mask = var_46293_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46293_cast_fp16")]; + tensor var_46300_begin_0 = const()[name = string("op_46300_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46300_end_0 = const()[name = string("op_46300_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46300_end_mask_0 = const()[name = string("op_46300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46300_cast_fp16 = slice_by_index(begin = var_46300_begin_0, end = var_46300_end_0, end_mask = var_46300_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46300_cast_fp16")]; + tensor var_46307_begin_0 = const()[name = string("op_46307_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46307_end_0 = const()[name = string("op_46307_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46307_end_mask_0 = const()[name = string("op_46307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46307_cast_fp16 = slice_by_index(begin = var_46307_begin_0, end = var_46307_end_0, end_mask = var_46307_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46307_cast_fp16")]; + tensor var_46314_begin_0 = const()[name = string("op_46314_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46314_end_0 = const()[name = string("op_46314_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46314_end_mask_0 = const()[name = string("op_46314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46314_cast_fp16 = slice_by_index(begin = var_46314_begin_0, end = var_46314_end_0, end_mask = var_46314_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46314_cast_fp16")]; + tensor var_46321_begin_0 = const()[name = string("op_46321_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46321_end_0 = const()[name = string("op_46321_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46321_end_mask_0 = const()[name = string("op_46321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46321_cast_fp16 = slice_by_index(begin = var_46321_begin_0, end = var_46321_end_0, end_mask = var_46321_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46321_cast_fp16")]; + tensor var_46328_begin_0 = const()[name = string("op_46328_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46328_end_0 = const()[name = string("op_46328_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46328_end_mask_0 = const()[name = string("op_46328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46328_cast_fp16 = slice_by_index(begin = var_46328_begin_0, end = var_46328_end_0, end_mask = var_46328_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46328_cast_fp16")]; + tensor var_46335_begin_0 = const()[name = string("op_46335_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46335_end_0 = const()[name = string("op_46335_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46335_end_mask_0 = const()[name = string("op_46335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46335_cast_fp16 = slice_by_index(begin = var_46335_begin_0, end = var_46335_end_0, end_mask = var_46335_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46335_cast_fp16")]; + tensor var_46342_begin_0 = const()[name = string("op_46342_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46342_end_0 = const()[name = string("op_46342_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46342_end_mask_0 = const()[name = string("op_46342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46342_cast_fp16 = slice_by_index(begin = var_46342_begin_0, end = var_46342_end_0, end_mask = var_46342_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46342_cast_fp16")]; + tensor var_46349_begin_0 = const()[name = string("op_46349_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46349_end_0 = const()[name = string("op_46349_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46349_end_mask_0 = const()[name = string("op_46349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46349_cast_fp16 = slice_by_index(begin = var_46349_begin_0, end = var_46349_end_0, end_mask = var_46349_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46349_cast_fp16")]; + tensor var_46356_begin_0 = const()[name = string("op_46356_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46356_end_0 = const()[name = string("op_46356_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46356_end_mask_0 = const()[name = string("op_46356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46356_cast_fp16 = slice_by_index(begin = var_46356_begin_0, end = var_46356_end_0, end_mask = var_46356_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46356_cast_fp16")]; + tensor var_46363_begin_0 = const()[name = string("op_46363_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46363_end_0 = const()[name = string("op_46363_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46363_end_mask_0 = const()[name = string("op_46363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46363_cast_fp16 = slice_by_index(begin = var_46363_begin_0, end = var_46363_end_0, end_mask = var_46363_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46363_cast_fp16")]; + tensor var_46370_begin_0 = const()[name = string("op_46370_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46370_end_0 = const()[name = string("op_46370_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46370_end_mask_0 = const()[name = string("op_46370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46370_cast_fp16 = slice_by_index(begin = var_46370_begin_0, end = var_46370_end_0, end_mask = var_46370_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46370_cast_fp16")]; + tensor var_46377_begin_0 = const()[name = string("op_46377_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46377_end_0 = const()[name = string("op_46377_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46377_end_mask_0 = const()[name = string("op_46377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46377_cast_fp16 = slice_by_index(begin = var_46377_begin_0, end = var_46377_end_0, end_mask = var_46377_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46377_cast_fp16")]; + tensor var_46384_begin_0 = const()[name = string("op_46384_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46384_end_0 = const()[name = string("op_46384_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46384_end_mask_0 = const()[name = string("op_46384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46384_cast_fp16 = slice_by_index(begin = var_46384_begin_0, end = var_46384_end_0, end_mask = var_46384_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46384_cast_fp16")]; + tensor var_46391_begin_0 = const()[name = string("op_46391_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46391_end_0 = const()[name = string("op_46391_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46391_end_mask_0 = const()[name = string("op_46391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46391_cast_fp16 = slice_by_index(begin = var_46391_begin_0, end = var_46391_end_0, end_mask = var_46391_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46391_cast_fp16")]; + tensor var_46398_begin_0 = const()[name = string("op_46398_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46398_end_0 = const()[name = string("op_46398_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46398_end_mask_0 = const()[name = string("op_46398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46398_cast_fp16 = slice_by_index(begin = var_46398_begin_0, end = var_46398_end_0, end_mask = var_46398_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46398_cast_fp16")]; + tensor var_46405_begin_0 = const()[name = string("op_46405_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46405_end_0 = const()[name = string("op_46405_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46405_end_mask_0 = const()[name = string("op_46405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46405_cast_fp16 = slice_by_index(begin = var_46405_begin_0, end = var_46405_end_0, end_mask = var_46405_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46405_cast_fp16")]; + tensor var_46412_begin_0 = const()[name = string("op_46412_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46412_end_0 = const()[name = string("op_46412_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46412_end_mask_0 = const()[name = string("op_46412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46412_cast_fp16 = slice_by_index(begin = var_46412_begin_0, end = var_46412_end_0, end_mask = var_46412_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46412_cast_fp16")]; + tensor var_46419_begin_0 = const()[name = string("op_46419_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46419_end_0 = const()[name = string("op_46419_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46419_end_mask_0 = const()[name = string("op_46419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46419_cast_fp16 = slice_by_index(begin = var_46419_begin_0, end = var_46419_end_0, end_mask = var_46419_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46419_cast_fp16")]; + tensor var_46426_begin_0 = const()[name = string("op_46426_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46426_end_0 = const()[name = string("op_46426_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46426_end_mask_0 = const()[name = string("op_46426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46426_cast_fp16 = slice_by_index(begin = var_46426_begin_0, end = var_46426_end_0, end_mask = var_46426_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46426_cast_fp16")]; + tensor var_46433_begin_0 = const()[name = string("op_46433_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46433_end_0 = const()[name = string("op_46433_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46433_end_mask_0 = const()[name = string("op_46433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46433_cast_fp16 = slice_by_index(begin = var_46433_begin_0, end = var_46433_end_0, end_mask = var_46433_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46433_cast_fp16")]; + tensor var_46440_begin_0 = const()[name = string("op_46440_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46440_end_0 = const()[name = string("op_46440_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46440_end_mask_0 = const()[name = string("op_46440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46440_cast_fp16 = slice_by_index(begin = var_46440_begin_0, end = var_46440_end_0, end_mask = var_46440_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46440_cast_fp16")]; + tensor var_46447_begin_0 = const()[name = string("op_46447_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46447_end_0 = const()[name = string("op_46447_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46447_end_mask_0 = const()[name = string("op_46447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46447_cast_fp16 = slice_by_index(begin = var_46447_begin_0, end = var_46447_end_0, end_mask = var_46447_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46447_cast_fp16")]; + tensor var_46454_begin_0 = const()[name = string("op_46454_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46454_end_0 = const()[name = string("op_46454_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46454_end_mask_0 = const()[name = string("op_46454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46454_cast_fp16 = slice_by_index(begin = var_46454_begin_0, end = var_46454_end_0, end_mask = var_46454_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46454_cast_fp16")]; + tensor var_46461_begin_0 = const()[name = string("op_46461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46461_end_0 = const()[name = string("op_46461_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46461_end_mask_0 = const()[name = string("op_46461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46461_cast_fp16 = slice_by_index(begin = var_46461_begin_0, end = var_46461_end_0, end_mask = var_46461_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46461_cast_fp16")]; + tensor var_46468_begin_0 = const()[name = string("op_46468_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46468_end_0 = const()[name = string("op_46468_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46468_end_mask_0 = const()[name = string("op_46468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46468_cast_fp16 = slice_by_index(begin = var_46468_begin_0, end = var_46468_end_0, end_mask = var_46468_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46468_cast_fp16")]; + tensor var_46475_begin_0 = const()[name = string("op_46475_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46475_end_0 = const()[name = string("op_46475_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46475_end_mask_0 = const()[name = string("op_46475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46475_cast_fp16 = slice_by_index(begin = var_46475_begin_0, end = var_46475_end_0, end_mask = var_46475_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46475_cast_fp16")]; + tensor var_46482_begin_0 = const()[name = string("op_46482_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46482_end_0 = const()[name = string("op_46482_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46482_end_mask_0 = const()[name = string("op_46482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46482_cast_fp16 = slice_by_index(begin = var_46482_begin_0, end = var_46482_end_0, end_mask = var_46482_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46482_cast_fp16")]; + tensor var_46489_begin_0 = const()[name = string("op_46489_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46489_end_0 = const()[name = string("op_46489_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46489_end_mask_0 = const()[name = string("op_46489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46489_cast_fp16 = slice_by_index(begin = var_46489_begin_0, end = var_46489_end_0, end_mask = var_46489_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46489_cast_fp16")]; + tensor var_46496_begin_0 = const()[name = string("op_46496_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46496_end_0 = const()[name = string("op_46496_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46496_end_mask_0 = const()[name = string("op_46496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46496_cast_fp16 = slice_by_index(begin = var_46496_begin_0, end = var_46496_end_0, end_mask = var_46496_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46496_cast_fp16")]; + tensor var_46503_begin_0 = const()[name = string("op_46503_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46503_end_0 = const()[name = string("op_46503_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46503_end_mask_0 = const()[name = string("op_46503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46503_cast_fp16 = slice_by_index(begin = var_46503_begin_0, end = var_46503_end_0, end_mask = var_46503_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46503_cast_fp16")]; + tensor var_46510_begin_0 = const()[name = string("op_46510_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46510_end_0 = const()[name = string("op_46510_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46510_end_mask_0 = const()[name = string("op_46510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46510_cast_fp16 = slice_by_index(begin = var_46510_begin_0, end = var_46510_end_0, end_mask = var_46510_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46510_cast_fp16")]; + tensor var_46517_begin_0 = const()[name = string("op_46517_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46517_end_0 = const()[name = string("op_46517_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46517_end_mask_0 = const()[name = string("op_46517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46517_cast_fp16 = slice_by_index(begin = var_46517_begin_0, end = var_46517_end_0, end_mask = var_46517_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46517_cast_fp16")]; + tensor var_46524_begin_0 = const()[name = string("op_46524_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46524_end_0 = const()[name = string("op_46524_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46524_end_mask_0 = const()[name = string("op_46524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46524_cast_fp16 = slice_by_index(begin = var_46524_begin_0, end = var_46524_end_0, end_mask = var_46524_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46524_cast_fp16")]; + tensor var_46531_begin_0 = const()[name = string("op_46531_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46531_end_0 = const()[name = string("op_46531_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46531_end_mask_0 = const()[name = string("op_46531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46531_cast_fp16 = slice_by_index(begin = var_46531_begin_0, end = var_46531_end_0, end_mask = var_46531_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46531_cast_fp16")]; + tensor var_46538_begin_0 = const()[name = string("op_46538_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46538_end_0 = const()[name = string("op_46538_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46538_end_mask_0 = const()[name = string("op_46538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46538_cast_fp16 = slice_by_index(begin = var_46538_begin_0, end = var_46538_end_0, end_mask = var_46538_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46538_cast_fp16")]; + tensor var_46545_begin_0 = const()[name = string("op_46545_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46545_end_0 = const()[name = string("op_46545_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46545_end_mask_0 = const()[name = string("op_46545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46545_cast_fp16 = slice_by_index(begin = var_46545_begin_0, end = var_46545_end_0, end_mask = var_46545_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46545_cast_fp16")]; + tensor var_46552_begin_0 = const()[name = string("op_46552_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46552_end_0 = const()[name = string("op_46552_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46552_end_mask_0 = const()[name = string("op_46552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46552_cast_fp16 = slice_by_index(begin = var_46552_begin_0, end = var_46552_end_0, end_mask = var_46552_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46552_cast_fp16")]; + tensor var_46559_begin_0 = const()[name = string("op_46559_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46559_end_0 = const()[name = string("op_46559_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46559_end_mask_0 = const()[name = string("op_46559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46559_cast_fp16 = slice_by_index(begin = var_46559_begin_0, end = var_46559_end_0, end_mask = var_46559_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46559_cast_fp16")]; + tensor var_46566_begin_0 = const()[name = string("op_46566_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46566_end_0 = const()[name = string("op_46566_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46566_end_mask_0 = const()[name = string("op_46566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46566_cast_fp16 = slice_by_index(begin = var_46566_begin_0, end = var_46566_end_0, end_mask = var_46566_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46566_cast_fp16")]; + tensor var_46573_begin_0 = const()[name = string("op_46573_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46573_end_0 = const()[name = string("op_46573_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46573_end_mask_0 = const()[name = string("op_46573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46573_cast_fp16 = slice_by_index(begin = var_46573_begin_0, end = var_46573_end_0, end_mask = var_46573_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46573_cast_fp16")]; + tensor var_46580_begin_0 = const()[name = string("op_46580_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46580_end_0 = const()[name = string("op_46580_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46580_end_mask_0 = const()[name = string("op_46580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46580_cast_fp16 = slice_by_index(begin = var_46580_begin_0, end = var_46580_end_0, end_mask = var_46580_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46580_cast_fp16")]; + tensor var_46587_begin_0 = const()[name = string("op_46587_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46587_end_0 = const()[name = string("op_46587_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46587_end_mask_0 = const()[name = string("op_46587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46587_cast_fp16 = slice_by_index(begin = var_46587_begin_0, end = var_46587_end_0, end_mask = var_46587_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46587_cast_fp16")]; + tensor var_46594_begin_0 = const()[name = string("op_46594_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46594_end_0 = const()[name = string("op_46594_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46594_end_mask_0 = const()[name = string("op_46594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46594_cast_fp16 = slice_by_index(begin = var_46594_begin_0, end = var_46594_end_0, end_mask = var_46594_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46594_cast_fp16")]; + tensor var_46601_begin_0 = const()[name = string("op_46601_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46601_end_0 = const()[name = string("op_46601_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46601_end_mask_0 = const()[name = string("op_46601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46601_cast_fp16 = slice_by_index(begin = var_46601_begin_0, end = var_46601_end_0, end_mask = var_46601_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46601_cast_fp16")]; + tensor var_46608_begin_0 = const()[name = string("op_46608_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46608_end_0 = const()[name = string("op_46608_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46608_end_mask_0 = const()[name = string("op_46608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46608_cast_fp16 = slice_by_index(begin = var_46608_begin_0, end = var_46608_end_0, end_mask = var_46608_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46608_cast_fp16")]; + tensor var_46615_begin_0 = const()[name = string("op_46615_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46615_end_0 = const()[name = string("op_46615_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46615_end_mask_0 = const()[name = string("op_46615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46615_cast_fp16 = slice_by_index(begin = var_46615_begin_0, end = var_46615_end_0, end_mask = var_46615_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46615_cast_fp16")]; + tensor var_46622_begin_0 = const()[name = string("op_46622_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46622_end_0 = const()[name = string("op_46622_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46622_end_mask_0 = const()[name = string("op_46622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46622_cast_fp16 = slice_by_index(begin = var_46622_begin_0, end = var_46622_end_0, end_mask = var_46622_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46622_cast_fp16")]; + tensor var_46629_begin_0 = const()[name = string("op_46629_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46629_end_0 = const()[name = string("op_46629_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46629_end_mask_0 = const()[name = string("op_46629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46629_cast_fp16 = slice_by_index(begin = var_46629_begin_0, end = var_46629_end_0, end_mask = var_46629_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46629_cast_fp16")]; + tensor var_46636_begin_0 = const()[name = string("op_46636_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46636_end_0 = const()[name = string("op_46636_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46636_end_mask_0 = const()[name = string("op_46636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46636_cast_fp16 = slice_by_index(begin = var_46636_begin_0, end = var_46636_end_0, end_mask = var_46636_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46636_cast_fp16")]; + tensor var_46643_begin_0 = const()[name = string("op_46643_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46643_end_0 = const()[name = string("op_46643_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46643_end_mask_0 = const()[name = string("op_46643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46643_cast_fp16 = slice_by_index(begin = var_46643_begin_0, end = var_46643_end_0, end_mask = var_46643_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46643_cast_fp16")]; + tensor var_46650_begin_0 = const()[name = string("op_46650_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46650_end_0 = const()[name = string("op_46650_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46650_end_mask_0 = const()[name = string("op_46650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46650_cast_fp16 = slice_by_index(begin = var_46650_begin_0, end = var_46650_end_0, end_mask = var_46650_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46650_cast_fp16")]; + tensor var_46657_begin_0 = const()[name = string("op_46657_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46657_end_0 = const()[name = string("op_46657_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46657_end_mask_0 = const()[name = string("op_46657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46657_cast_fp16 = slice_by_index(begin = var_46657_begin_0, end = var_46657_end_0, end_mask = var_46657_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46657_cast_fp16")]; + tensor var_46664_begin_0 = const()[name = string("op_46664_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46664_end_0 = const()[name = string("op_46664_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46664_end_mask_0 = const()[name = string("op_46664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46664_cast_fp16 = slice_by_index(begin = var_46664_begin_0, end = var_46664_end_0, end_mask = var_46664_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46664_cast_fp16")]; + tensor var_46671_begin_0 = const()[name = string("op_46671_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46671_end_0 = const()[name = string("op_46671_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46671_end_mask_0 = const()[name = string("op_46671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46671_cast_fp16 = slice_by_index(begin = var_46671_begin_0, end = var_46671_end_0, end_mask = var_46671_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46671_cast_fp16")]; + tensor var_46678_begin_0 = const()[name = string("op_46678_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46678_end_0 = const()[name = string("op_46678_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46678_end_mask_0 = const()[name = string("op_46678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46678_cast_fp16 = slice_by_index(begin = var_46678_begin_0, end = var_46678_end_0, end_mask = var_46678_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46678_cast_fp16")]; + tensor var_46685_begin_0 = const()[name = string("op_46685_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46685_end_0 = const()[name = string("op_46685_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46685_end_mask_0 = const()[name = string("op_46685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46685_cast_fp16 = slice_by_index(begin = var_46685_begin_0, end = var_46685_end_0, end_mask = var_46685_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46685_cast_fp16")]; + tensor var_46692_begin_0 = const()[name = string("op_46692_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46692_end_0 = const()[name = string("op_46692_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46692_end_mask_0 = const()[name = string("op_46692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46692_cast_fp16 = slice_by_index(begin = var_46692_begin_0, end = var_46692_end_0, end_mask = var_46692_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46692_cast_fp16")]; + tensor var_46699_begin_0 = const()[name = string("op_46699_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46699_end_0 = const()[name = string("op_46699_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46699_end_mask_0 = const()[name = string("op_46699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46699_cast_fp16 = slice_by_index(begin = var_46699_begin_0, end = var_46699_end_0, end_mask = var_46699_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46699_cast_fp16")]; + tensor var_46706_begin_0 = const()[name = string("op_46706_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46706_end_0 = const()[name = string("op_46706_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46706_end_mask_0 = const()[name = string("op_46706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46706_cast_fp16 = slice_by_index(begin = var_46706_begin_0, end = var_46706_end_0, end_mask = var_46706_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46706_cast_fp16")]; + tensor k_61_perm_0 = const()[name = string("k_61_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_46711_begin_0 = const()[name = string("op_46711_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46711_end_0 = const()[name = string("op_46711_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_46711_end_mask_0 = const()[name = string("op_46711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_61_cast_fp16 = transpose(perm = k_61_perm_0, x = key_61_cast_fp16)[name = string("transpose_1")]; + tensor var_46711_cast_fp16 = slice_by_index(begin = var_46711_begin_0, end = var_46711_end_0, end_mask = var_46711_end_mask_0, x = k_61_cast_fp16)[name = string("op_46711_cast_fp16")]; + tensor var_46715_begin_0 = const()[name = string("op_46715_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_46715_end_0 = const()[name = string("op_46715_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_46715_end_mask_0 = const()[name = string("op_46715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46715_cast_fp16 = slice_by_index(begin = var_46715_begin_0, end = var_46715_end_0, end_mask = var_46715_end_mask_0, x = k_61_cast_fp16)[name = string("op_46715_cast_fp16")]; + tensor var_46719_begin_0 = const()[name = string("op_46719_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_46719_end_0 = const()[name = string("op_46719_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_46719_end_mask_0 = const()[name = string("op_46719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46719_cast_fp16 = slice_by_index(begin = var_46719_begin_0, end = var_46719_end_0, end_mask = var_46719_end_mask_0, x = k_61_cast_fp16)[name = string("op_46719_cast_fp16")]; + tensor var_46723_begin_0 = const()[name = string("op_46723_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_46723_end_0 = const()[name = string("op_46723_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_46723_end_mask_0 = const()[name = string("op_46723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46723_cast_fp16 = slice_by_index(begin = var_46723_begin_0, end = var_46723_end_0, end_mask = var_46723_end_mask_0, x = k_61_cast_fp16)[name = string("op_46723_cast_fp16")]; + tensor var_46727_begin_0 = const()[name = string("op_46727_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_46727_end_0 = const()[name = string("op_46727_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_46727_end_mask_0 = const()[name = string("op_46727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46727_cast_fp16 = slice_by_index(begin = var_46727_begin_0, end = var_46727_end_0, end_mask = var_46727_end_mask_0, x = k_61_cast_fp16)[name = string("op_46727_cast_fp16")]; + tensor var_46731_begin_0 = const()[name = string("op_46731_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_46731_end_0 = const()[name = string("op_46731_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_46731_end_mask_0 = const()[name = string("op_46731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46731_cast_fp16 = slice_by_index(begin = var_46731_begin_0, end = var_46731_end_0, end_mask = var_46731_end_mask_0, x = k_61_cast_fp16)[name = string("op_46731_cast_fp16")]; + tensor var_46735_begin_0 = const()[name = string("op_46735_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_46735_end_0 = const()[name = string("op_46735_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_46735_end_mask_0 = const()[name = string("op_46735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46735_cast_fp16 = slice_by_index(begin = var_46735_begin_0, end = var_46735_end_0, end_mask = var_46735_end_mask_0, x = k_61_cast_fp16)[name = string("op_46735_cast_fp16")]; + tensor var_46739_begin_0 = const()[name = string("op_46739_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_46739_end_0 = const()[name = string("op_46739_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_46739_end_mask_0 = const()[name = string("op_46739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46739_cast_fp16 = slice_by_index(begin = var_46739_begin_0, end = var_46739_end_0, end_mask = var_46739_end_mask_0, x = k_61_cast_fp16)[name = string("op_46739_cast_fp16")]; + tensor var_46743_begin_0 = const()[name = string("op_46743_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_46743_end_0 = const()[name = string("op_46743_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_46743_end_mask_0 = const()[name = string("op_46743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46743_cast_fp16 = slice_by_index(begin = var_46743_begin_0, end = var_46743_end_0, end_mask = var_46743_end_mask_0, x = k_61_cast_fp16)[name = string("op_46743_cast_fp16")]; + tensor var_46747_begin_0 = const()[name = string("op_46747_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_46747_end_0 = const()[name = string("op_46747_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_46747_end_mask_0 = const()[name = string("op_46747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46747_cast_fp16 = slice_by_index(begin = var_46747_begin_0, end = var_46747_end_0, end_mask = var_46747_end_mask_0, x = k_61_cast_fp16)[name = string("op_46747_cast_fp16")]; + tensor var_46751_begin_0 = const()[name = string("op_46751_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_46751_end_0 = const()[name = string("op_46751_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_46751_end_mask_0 = const()[name = string("op_46751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46751_cast_fp16 = slice_by_index(begin = var_46751_begin_0, end = var_46751_end_0, end_mask = var_46751_end_mask_0, x = k_61_cast_fp16)[name = string("op_46751_cast_fp16")]; + tensor var_46755_begin_0 = const()[name = string("op_46755_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_46755_end_0 = const()[name = string("op_46755_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_46755_end_mask_0 = const()[name = string("op_46755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46755_cast_fp16 = slice_by_index(begin = var_46755_begin_0, end = var_46755_end_0, end_mask = var_46755_end_mask_0, x = k_61_cast_fp16)[name = string("op_46755_cast_fp16")]; + tensor var_46759_begin_0 = const()[name = string("op_46759_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_46759_end_0 = const()[name = string("op_46759_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_46759_end_mask_0 = const()[name = string("op_46759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46759_cast_fp16 = slice_by_index(begin = var_46759_begin_0, end = var_46759_end_0, end_mask = var_46759_end_mask_0, x = k_61_cast_fp16)[name = string("op_46759_cast_fp16")]; + tensor var_46763_begin_0 = const()[name = string("op_46763_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_46763_end_0 = const()[name = string("op_46763_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_46763_end_mask_0 = const()[name = string("op_46763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46763_cast_fp16 = slice_by_index(begin = var_46763_begin_0, end = var_46763_end_0, end_mask = var_46763_end_mask_0, x = k_61_cast_fp16)[name = string("op_46763_cast_fp16")]; + tensor var_46767_begin_0 = const()[name = string("op_46767_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_46767_end_0 = const()[name = string("op_46767_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_46767_end_mask_0 = const()[name = string("op_46767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46767_cast_fp16 = slice_by_index(begin = var_46767_begin_0, end = var_46767_end_0, end_mask = var_46767_end_mask_0, x = k_61_cast_fp16)[name = string("op_46767_cast_fp16")]; + tensor var_46771_begin_0 = const()[name = string("op_46771_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_46771_end_0 = const()[name = string("op_46771_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_46771_end_mask_0 = const()[name = string("op_46771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46771_cast_fp16 = slice_by_index(begin = var_46771_begin_0, end = var_46771_end_0, end_mask = var_46771_end_mask_0, x = k_61_cast_fp16)[name = string("op_46771_cast_fp16")]; + tensor var_46775_begin_0 = const()[name = string("op_46775_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_46775_end_0 = const()[name = string("op_46775_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_46775_end_mask_0 = const()[name = string("op_46775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46775_cast_fp16 = slice_by_index(begin = var_46775_begin_0, end = var_46775_end_0, end_mask = var_46775_end_mask_0, x = k_61_cast_fp16)[name = string("op_46775_cast_fp16")]; + tensor var_46779_begin_0 = const()[name = string("op_46779_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_46779_end_0 = const()[name = string("op_46779_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_46779_end_mask_0 = const()[name = string("op_46779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46779_cast_fp16 = slice_by_index(begin = var_46779_begin_0, end = var_46779_end_0, end_mask = var_46779_end_mask_0, x = k_61_cast_fp16)[name = string("op_46779_cast_fp16")]; + tensor var_46783_begin_0 = const()[name = string("op_46783_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_46783_end_0 = const()[name = string("op_46783_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_46783_end_mask_0 = const()[name = string("op_46783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46783_cast_fp16 = slice_by_index(begin = var_46783_begin_0, end = var_46783_end_0, end_mask = var_46783_end_mask_0, x = k_61_cast_fp16)[name = string("op_46783_cast_fp16")]; + tensor var_46787_begin_0 = const()[name = string("op_46787_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_46787_end_0 = const()[name = string("op_46787_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_46787_end_mask_0 = const()[name = string("op_46787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46787_cast_fp16 = slice_by_index(begin = var_46787_begin_0, end = var_46787_end_0, end_mask = var_46787_end_mask_0, x = k_61_cast_fp16)[name = string("op_46787_cast_fp16")]; + tensor var_46789_begin_0 = const()[name = string("op_46789_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46789_end_0 = const()[name = string("op_46789_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46789_end_mask_0 = const()[name = string("op_46789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46789_cast_fp16 = slice_by_index(begin = var_46789_begin_0, end = var_46789_end_0, end_mask = var_46789_end_mask_0, x = value_61_cast_fp16)[name = string("op_46789_cast_fp16")]; + tensor var_46793_begin_0 = const()[name = string("op_46793_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46793_end_0 = const()[name = string("op_46793_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_46793_end_mask_0 = const()[name = string("op_46793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46793_cast_fp16 = slice_by_index(begin = var_46793_begin_0, end = var_46793_end_0, end_mask = var_46793_end_mask_0, x = value_61_cast_fp16)[name = string("op_46793_cast_fp16")]; + tensor var_46797_begin_0 = const()[name = string("op_46797_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46797_end_0 = const()[name = string("op_46797_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_46797_end_mask_0 = const()[name = string("op_46797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46797_cast_fp16 = slice_by_index(begin = var_46797_begin_0, end = var_46797_end_0, end_mask = var_46797_end_mask_0, x = value_61_cast_fp16)[name = string("op_46797_cast_fp16")]; + tensor var_46801_begin_0 = const()[name = string("op_46801_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46801_end_0 = const()[name = string("op_46801_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_46801_end_mask_0 = const()[name = string("op_46801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46801_cast_fp16 = slice_by_index(begin = var_46801_begin_0, end = var_46801_end_0, end_mask = var_46801_end_mask_0, x = value_61_cast_fp16)[name = string("op_46801_cast_fp16")]; + tensor var_46805_begin_0 = const()[name = string("op_46805_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46805_end_0 = const()[name = string("op_46805_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_46805_end_mask_0 = const()[name = string("op_46805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46805_cast_fp16 = slice_by_index(begin = var_46805_begin_0, end = var_46805_end_0, end_mask = var_46805_end_mask_0, x = value_61_cast_fp16)[name = string("op_46805_cast_fp16")]; + tensor var_46809_begin_0 = const()[name = string("op_46809_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46809_end_0 = const()[name = string("op_46809_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_46809_end_mask_0 = const()[name = string("op_46809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46809_cast_fp16 = slice_by_index(begin = var_46809_begin_0, end = var_46809_end_0, end_mask = var_46809_end_mask_0, x = value_61_cast_fp16)[name = string("op_46809_cast_fp16")]; + tensor var_46813_begin_0 = const()[name = string("op_46813_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46813_end_0 = const()[name = string("op_46813_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_46813_end_mask_0 = const()[name = string("op_46813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46813_cast_fp16 = slice_by_index(begin = var_46813_begin_0, end = var_46813_end_0, end_mask = var_46813_end_mask_0, x = value_61_cast_fp16)[name = string("op_46813_cast_fp16")]; + tensor var_46817_begin_0 = const()[name = string("op_46817_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46817_end_0 = const()[name = string("op_46817_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_46817_end_mask_0 = const()[name = string("op_46817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46817_cast_fp16 = slice_by_index(begin = var_46817_begin_0, end = var_46817_end_0, end_mask = var_46817_end_mask_0, x = value_61_cast_fp16)[name = string("op_46817_cast_fp16")]; + tensor var_46821_begin_0 = const()[name = string("op_46821_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46821_end_0 = const()[name = string("op_46821_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_46821_end_mask_0 = const()[name = string("op_46821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46821_cast_fp16 = slice_by_index(begin = var_46821_begin_0, end = var_46821_end_0, end_mask = var_46821_end_mask_0, x = value_61_cast_fp16)[name = string("op_46821_cast_fp16")]; + tensor var_46825_begin_0 = const()[name = string("op_46825_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46825_end_0 = const()[name = string("op_46825_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_46825_end_mask_0 = const()[name = string("op_46825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46825_cast_fp16 = slice_by_index(begin = var_46825_begin_0, end = var_46825_end_0, end_mask = var_46825_end_mask_0, x = value_61_cast_fp16)[name = string("op_46825_cast_fp16")]; + tensor var_46829_begin_0 = const()[name = string("op_46829_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46829_end_0 = const()[name = string("op_46829_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_46829_end_mask_0 = const()[name = string("op_46829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46829_cast_fp16 = slice_by_index(begin = var_46829_begin_0, end = var_46829_end_0, end_mask = var_46829_end_mask_0, x = value_61_cast_fp16)[name = string("op_46829_cast_fp16")]; + tensor var_46833_begin_0 = const()[name = string("op_46833_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46833_end_0 = const()[name = string("op_46833_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_46833_end_mask_0 = const()[name = string("op_46833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46833_cast_fp16 = slice_by_index(begin = var_46833_begin_0, end = var_46833_end_0, end_mask = var_46833_end_mask_0, x = value_61_cast_fp16)[name = string("op_46833_cast_fp16")]; + tensor var_46837_begin_0 = const()[name = string("op_46837_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46837_end_0 = const()[name = string("op_46837_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_46837_end_mask_0 = const()[name = string("op_46837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46837_cast_fp16 = slice_by_index(begin = var_46837_begin_0, end = var_46837_end_0, end_mask = var_46837_end_mask_0, x = value_61_cast_fp16)[name = string("op_46837_cast_fp16")]; + tensor var_46841_begin_0 = const()[name = string("op_46841_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46841_end_0 = const()[name = string("op_46841_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_46841_end_mask_0 = const()[name = string("op_46841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46841_cast_fp16 = slice_by_index(begin = var_46841_begin_0, end = var_46841_end_0, end_mask = var_46841_end_mask_0, x = value_61_cast_fp16)[name = string("op_46841_cast_fp16")]; + tensor var_46845_begin_0 = const()[name = string("op_46845_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46845_end_0 = const()[name = string("op_46845_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_46845_end_mask_0 = const()[name = string("op_46845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46845_cast_fp16 = slice_by_index(begin = var_46845_begin_0, end = var_46845_end_0, end_mask = var_46845_end_mask_0, x = value_61_cast_fp16)[name = string("op_46845_cast_fp16")]; + tensor var_46849_begin_0 = const()[name = string("op_46849_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46849_end_0 = const()[name = string("op_46849_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_46849_end_mask_0 = const()[name = string("op_46849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46849_cast_fp16 = slice_by_index(begin = var_46849_begin_0, end = var_46849_end_0, end_mask = var_46849_end_mask_0, x = value_61_cast_fp16)[name = string("op_46849_cast_fp16")]; + tensor var_46853_begin_0 = const()[name = string("op_46853_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46853_end_0 = const()[name = string("op_46853_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_46853_end_mask_0 = const()[name = string("op_46853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46853_cast_fp16 = slice_by_index(begin = var_46853_begin_0, end = var_46853_end_0, end_mask = var_46853_end_mask_0, x = value_61_cast_fp16)[name = string("op_46853_cast_fp16")]; + tensor var_46857_begin_0 = const()[name = string("op_46857_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46857_end_0 = const()[name = string("op_46857_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_46857_end_mask_0 = const()[name = string("op_46857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46857_cast_fp16 = slice_by_index(begin = var_46857_begin_0, end = var_46857_end_0, end_mask = var_46857_end_mask_0, x = value_61_cast_fp16)[name = string("op_46857_cast_fp16")]; + tensor var_46861_begin_0 = const()[name = string("op_46861_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46861_end_0 = const()[name = string("op_46861_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_46861_end_mask_0 = const()[name = string("op_46861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46861_cast_fp16 = slice_by_index(begin = var_46861_begin_0, end = var_46861_end_0, end_mask = var_46861_end_mask_0, x = value_61_cast_fp16)[name = string("op_46861_cast_fp16")]; + tensor var_46865_begin_0 = const()[name = string("op_46865_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46865_end_0 = const()[name = string("op_46865_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_46865_end_mask_0 = const()[name = string("op_46865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46865_cast_fp16 = slice_by_index(begin = var_46865_begin_0, end = var_46865_end_0, end_mask = var_46865_end_mask_0, x = value_61_cast_fp16)[name = string("op_46865_cast_fp16")]; + string _SplitHeadsQ__mh_w_4801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4801_equation_0, values = (var_46711_cast_fp16, var_46153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4801_cast_fp16")]; + string _SplitHeadsQ__mh_w_4803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4803_equation_0, values = (var_46711_cast_fp16, var_46160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4803_cast_fp16")]; + string _SplitHeadsQ__mh_w_4805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4805_equation_0, values = (var_46711_cast_fp16, var_46167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4805_cast_fp16")]; + string _SplitHeadsQ__mh_w_4807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4807_equation_0, values = (var_46711_cast_fp16, var_46174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4807_cast_fp16")]; + string _SplitHeadsQ__mh_w_4809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4809_equation_0, values = (var_46715_cast_fp16, var_46181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4809_cast_fp16")]; + string _SplitHeadsQ__mh_w_4811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4811_equation_0, values = (var_46715_cast_fp16, var_46188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4811_cast_fp16")]; + string _SplitHeadsQ__mh_w_4813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4813_equation_0, values = (var_46715_cast_fp16, var_46195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4813_cast_fp16")]; + string _SplitHeadsQ__mh_w_4815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4815_equation_0, values = (var_46715_cast_fp16, var_46202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4815_cast_fp16")]; + string _SplitHeadsQ__mh_w_4817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4817_equation_0, values = (var_46719_cast_fp16, var_46209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4817_cast_fp16")]; + string _SplitHeadsQ__mh_w_4819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4819_equation_0, values = (var_46719_cast_fp16, var_46216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4819_cast_fp16")]; + string _SplitHeadsQ__mh_w_4821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4821_equation_0, values = (var_46719_cast_fp16, var_46223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4821_cast_fp16")]; + string _SplitHeadsQ__mh_w_4823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4823_equation_0, values = (var_46719_cast_fp16, var_46230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4823_cast_fp16")]; + string _SplitHeadsQ__mh_w_4825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4825_equation_0, values = (var_46723_cast_fp16, var_46237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4825_cast_fp16")]; + string _SplitHeadsQ__mh_w_4827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4827_equation_0, values = (var_46723_cast_fp16, var_46244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4827_cast_fp16")]; + string _SplitHeadsQ__mh_w_4829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4829_equation_0, values = (var_46723_cast_fp16, var_46251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4829_cast_fp16")]; + string _SplitHeadsQ__mh_w_4831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4831_equation_0, values = (var_46723_cast_fp16, var_46258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4831_cast_fp16")]; + string _SplitHeadsQ__mh_w_4833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4833_equation_0, values = (var_46727_cast_fp16, var_46265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4833_cast_fp16")]; + string _SplitHeadsQ__mh_w_4835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4835_equation_0, values = (var_46727_cast_fp16, var_46272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4835_cast_fp16")]; + string _SplitHeadsQ__mh_w_4837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4837_equation_0, values = (var_46727_cast_fp16, var_46279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4837_cast_fp16")]; + string _SplitHeadsQ__mh_w_4839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4839_equation_0, values = (var_46727_cast_fp16, var_46286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4839_cast_fp16")]; + string _SplitHeadsQ__mh_w_4841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4841_equation_0, values = (var_46731_cast_fp16, var_46293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4841_cast_fp16")]; + string _SplitHeadsQ__mh_w_4843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4843_equation_0, values = (var_46731_cast_fp16, var_46300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4843_cast_fp16")]; + string _SplitHeadsQ__mh_w_4845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4845_equation_0, values = (var_46731_cast_fp16, var_46307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4845_cast_fp16")]; + string _SplitHeadsQ__mh_w_4847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4847_equation_0, values = (var_46731_cast_fp16, var_46314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4847_cast_fp16")]; + string _SplitHeadsQ__mh_w_4849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4849_equation_0, values = (var_46735_cast_fp16, var_46321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4849_cast_fp16")]; + string _SplitHeadsQ__mh_w_4851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4851_equation_0, values = (var_46735_cast_fp16, var_46328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4851_cast_fp16")]; + string _SplitHeadsQ__mh_w_4853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4853_equation_0, values = (var_46735_cast_fp16, var_46335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4853_cast_fp16")]; + string _SplitHeadsQ__mh_w_4855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4855_equation_0, values = (var_46735_cast_fp16, var_46342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4855_cast_fp16")]; + string _SplitHeadsQ__mh_w_4857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4857_equation_0, values = (var_46739_cast_fp16, var_46349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4857_cast_fp16")]; + string _SplitHeadsQ__mh_w_4859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4859_equation_0, values = (var_46739_cast_fp16, var_46356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4859_cast_fp16")]; + string _SplitHeadsQ__mh_w_4861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4861_equation_0, values = (var_46739_cast_fp16, var_46363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4861_cast_fp16")]; + string _SplitHeadsQ__mh_w_4863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4863_equation_0, values = (var_46739_cast_fp16, var_46370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4863_cast_fp16")]; + string _SplitHeadsQ__mh_w_4865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4865_equation_0, values = (var_46743_cast_fp16, var_46377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4865_cast_fp16")]; + string _SplitHeadsQ__mh_w_4867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4867_equation_0, values = (var_46743_cast_fp16, var_46384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4867_cast_fp16")]; + string _SplitHeadsQ__mh_w_4869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4869_equation_0, values = (var_46743_cast_fp16, var_46391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4869_cast_fp16")]; + string _SplitHeadsQ__mh_w_4871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4871_equation_0, values = (var_46743_cast_fp16, var_46398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4871_cast_fp16")]; + string _SplitHeadsQ__mh_w_4873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4873_equation_0, values = (var_46747_cast_fp16, var_46405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4873_cast_fp16")]; + string _SplitHeadsQ__mh_w_4875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4875_equation_0, values = (var_46747_cast_fp16, var_46412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4875_cast_fp16")]; + string _SplitHeadsQ__mh_w_4877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4877_equation_0, values = (var_46747_cast_fp16, var_46419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4877_cast_fp16")]; + string _SplitHeadsQ__mh_w_4879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4879_equation_0, values = (var_46747_cast_fp16, var_46426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4879_cast_fp16")]; + string _SplitHeadsQ__mh_w_4881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4881_equation_0, values = (var_46751_cast_fp16, var_46433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4881_cast_fp16")]; + string _SplitHeadsQ__mh_w_4883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4883_equation_0, values = (var_46751_cast_fp16, var_46440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4883_cast_fp16")]; + string _SplitHeadsQ__mh_w_4885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4885_equation_0, values = (var_46751_cast_fp16, var_46447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4885_cast_fp16")]; + string _SplitHeadsQ__mh_w_4887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4887_equation_0, values = (var_46751_cast_fp16, var_46454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4887_cast_fp16")]; + string _SplitHeadsQ__mh_w_4889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4889_equation_0, values = (var_46755_cast_fp16, var_46461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4889_cast_fp16")]; + string _SplitHeadsQ__mh_w_4891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4891_equation_0, values = (var_46755_cast_fp16, var_46468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4891_cast_fp16")]; + string _SplitHeadsQ__mh_w_4893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4893_equation_0, values = (var_46755_cast_fp16, var_46475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4893_cast_fp16")]; + string _SplitHeadsQ__mh_w_4895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4895_equation_0, values = (var_46755_cast_fp16, var_46482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4895_cast_fp16")]; + string _SplitHeadsQ__mh_w_4897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4897_equation_0, values = (var_46759_cast_fp16, var_46489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4897_cast_fp16")]; + string _SplitHeadsQ__mh_w_4899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4899_equation_0, values = (var_46759_cast_fp16, var_46496_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4899_cast_fp16")]; + string _SplitHeadsQ__mh_w_4901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4901_equation_0, values = (var_46759_cast_fp16, var_46503_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4901_cast_fp16")]; + string _SplitHeadsQ__mh_w_4903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4903_equation_0, values = (var_46759_cast_fp16, var_46510_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4903_cast_fp16")]; + string _SplitHeadsQ__mh_w_4905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4905_equation_0, values = (var_46763_cast_fp16, var_46517_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4905_cast_fp16")]; + string _SplitHeadsQ__mh_w_4907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4907_equation_0, values = (var_46763_cast_fp16, var_46524_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4907_cast_fp16")]; + string _SplitHeadsQ__mh_w_4909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4909_equation_0, values = (var_46763_cast_fp16, var_46531_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4909_cast_fp16")]; + string _SplitHeadsQ__mh_w_4911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4911_equation_0, values = (var_46763_cast_fp16, var_46538_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4911_cast_fp16")]; + string _SplitHeadsQ__mh_w_4913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4913_equation_0, values = (var_46767_cast_fp16, var_46545_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4913_cast_fp16")]; + string _SplitHeadsQ__mh_w_4915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4915_equation_0, values = (var_46767_cast_fp16, var_46552_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4915_cast_fp16")]; + string _SplitHeadsQ__mh_w_4917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4917_equation_0, values = (var_46767_cast_fp16, var_46559_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4917_cast_fp16")]; + string _SplitHeadsQ__mh_w_4919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4919_equation_0, values = (var_46767_cast_fp16, var_46566_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4919_cast_fp16")]; + string _SplitHeadsQ__mh_w_4921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4921_equation_0, values = (var_46771_cast_fp16, var_46573_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4921_cast_fp16")]; + string _SplitHeadsQ__mh_w_4923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4923_equation_0, values = (var_46771_cast_fp16, var_46580_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4923_cast_fp16")]; + string _SplitHeadsQ__mh_w_4925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4925_equation_0, values = (var_46771_cast_fp16, var_46587_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4925_cast_fp16")]; + string _SplitHeadsQ__mh_w_4927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4927_equation_0, values = (var_46771_cast_fp16, var_46594_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4927_cast_fp16")]; + string _SplitHeadsQ__mh_w_4929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4929_equation_0, values = (var_46775_cast_fp16, var_46601_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4929_cast_fp16")]; + string _SplitHeadsQ__mh_w_4931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4931_equation_0, values = (var_46775_cast_fp16, var_46608_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4931_cast_fp16")]; + string _SplitHeadsQ__mh_w_4933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4933_equation_0, values = (var_46775_cast_fp16, var_46615_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4933_cast_fp16")]; + string _SplitHeadsQ__mh_w_4935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4935_equation_0, values = (var_46775_cast_fp16, var_46622_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4935_cast_fp16")]; + string _SplitHeadsQ__mh_w_4937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4937_equation_0, values = (var_46779_cast_fp16, var_46629_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4937_cast_fp16")]; + string _SplitHeadsQ__mh_w_4939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4939_equation_0, values = (var_46779_cast_fp16, var_46636_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4939_cast_fp16")]; + string _SplitHeadsQ__mh_w_4941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4941_equation_0, values = (var_46779_cast_fp16, var_46643_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4941_cast_fp16")]; + string _SplitHeadsQ__mh_w_4943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4943_equation_0, values = (var_46779_cast_fp16, var_46650_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4943_cast_fp16")]; + string _SplitHeadsQ__mh_w_4945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4945_equation_0, values = (var_46783_cast_fp16, var_46657_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4945_cast_fp16")]; + string _SplitHeadsQ__mh_w_4947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4947_equation_0, values = (var_46783_cast_fp16, var_46664_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4947_cast_fp16")]; + string _SplitHeadsQ__mh_w_4949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4949_equation_0, values = (var_46783_cast_fp16, var_46671_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4949_cast_fp16")]; + string _SplitHeadsQ__mh_w_4951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4951_equation_0, values = (var_46783_cast_fp16, var_46678_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4951_cast_fp16")]; + string _SplitHeadsQ__mh_w_4953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4953_equation_0, values = (var_46787_cast_fp16, var_46685_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4953_cast_fp16")]; + string _SplitHeadsQ__mh_w_4955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4955_equation_0, values = (var_46787_cast_fp16, var_46692_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4955_cast_fp16")]; + string _SplitHeadsQ__mh_w_4957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4957_equation_0, values = (var_46787_cast_fp16, var_46699_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4957_cast_fp16")]; + string _SplitHeadsQ__mh_w_4959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4959_equation_0, values = (var_46787_cast_fp16, var_46706_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4959_cast_fp16")]; + fp16 var_47028_to_fp16 = const()[name = string("op_47028_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4801_cast_fp16, y = var_47028_to_fp16)[name = string("aw_chunk_4801_cast_fp16")]; + fp16 var_47030_to_fp16 = const()[name = string("op_47030_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4803_cast_fp16, y = var_47030_to_fp16)[name = string("aw_chunk_4803_cast_fp16")]; + fp16 var_47032_to_fp16 = const()[name = string("op_47032_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4805_cast_fp16, y = var_47032_to_fp16)[name = string("aw_chunk_4805_cast_fp16")]; + fp16 var_47034_to_fp16 = const()[name = string("op_47034_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4807_cast_fp16, y = var_47034_to_fp16)[name = string("aw_chunk_4807_cast_fp16")]; + fp16 var_47036_to_fp16 = const()[name = string("op_47036_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4809_cast_fp16, y = var_47036_to_fp16)[name = string("aw_chunk_4809_cast_fp16")]; + fp16 var_47038_to_fp16 = const()[name = string("op_47038_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4811_cast_fp16, y = var_47038_to_fp16)[name = string("aw_chunk_4811_cast_fp16")]; + fp16 var_47040_to_fp16 = const()[name = string("op_47040_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4813_cast_fp16, y = var_47040_to_fp16)[name = string("aw_chunk_4813_cast_fp16")]; + fp16 var_47042_to_fp16 = const()[name = string("op_47042_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4815_cast_fp16, y = var_47042_to_fp16)[name = string("aw_chunk_4815_cast_fp16")]; + fp16 var_47044_to_fp16 = const()[name = string("op_47044_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4817_cast_fp16, y = var_47044_to_fp16)[name = string("aw_chunk_4817_cast_fp16")]; + fp16 var_47046_to_fp16 = const()[name = string("op_47046_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4819_cast_fp16, y = var_47046_to_fp16)[name = string("aw_chunk_4819_cast_fp16")]; + fp16 var_47048_to_fp16 = const()[name = string("op_47048_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4821_cast_fp16, y = var_47048_to_fp16)[name = string("aw_chunk_4821_cast_fp16")]; + fp16 var_47050_to_fp16 = const()[name = string("op_47050_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4823_cast_fp16, y = var_47050_to_fp16)[name = string("aw_chunk_4823_cast_fp16")]; + fp16 var_47052_to_fp16 = const()[name = string("op_47052_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4825_cast_fp16, y = var_47052_to_fp16)[name = string("aw_chunk_4825_cast_fp16")]; + fp16 var_47054_to_fp16 = const()[name = string("op_47054_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4827_cast_fp16, y = var_47054_to_fp16)[name = string("aw_chunk_4827_cast_fp16")]; + fp16 var_47056_to_fp16 = const()[name = string("op_47056_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4829_cast_fp16, y = var_47056_to_fp16)[name = string("aw_chunk_4829_cast_fp16")]; + fp16 var_47058_to_fp16 = const()[name = string("op_47058_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4831_cast_fp16, y = var_47058_to_fp16)[name = string("aw_chunk_4831_cast_fp16")]; + fp16 var_47060_to_fp16 = const()[name = string("op_47060_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4833_cast_fp16, y = var_47060_to_fp16)[name = string("aw_chunk_4833_cast_fp16")]; + fp16 var_47062_to_fp16 = const()[name = string("op_47062_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4835_cast_fp16, y = var_47062_to_fp16)[name = string("aw_chunk_4835_cast_fp16")]; + fp16 var_47064_to_fp16 = const()[name = string("op_47064_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4837_cast_fp16, y = var_47064_to_fp16)[name = string("aw_chunk_4837_cast_fp16")]; + fp16 var_47066_to_fp16 = const()[name = string("op_47066_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4839_cast_fp16, y = var_47066_to_fp16)[name = string("aw_chunk_4839_cast_fp16")]; + fp16 var_47068_to_fp16 = const()[name = string("op_47068_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4841_cast_fp16, y = var_47068_to_fp16)[name = string("aw_chunk_4841_cast_fp16")]; + fp16 var_47070_to_fp16 = const()[name = string("op_47070_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4843_cast_fp16, y = var_47070_to_fp16)[name = string("aw_chunk_4843_cast_fp16")]; + fp16 var_47072_to_fp16 = const()[name = string("op_47072_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4845_cast_fp16, y = var_47072_to_fp16)[name = string("aw_chunk_4845_cast_fp16")]; + fp16 var_47074_to_fp16 = const()[name = string("op_47074_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4847_cast_fp16, y = var_47074_to_fp16)[name = string("aw_chunk_4847_cast_fp16")]; + fp16 var_47076_to_fp16 = const()[name = string("op_47076_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4849_cast_fp16, y = var_47076_to_fp16)[name = string("aw_chunk_4849_cast_fp16")]; + fp16 var_47078_to_fp16 = const()[name = string("op_47078_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4851_cast_fp16, y = var_47078_to_fp16)[name = string("aw_chunk_4851_cast_fp16")]; + fp16 var_47080_to_fp16 = const()[name = string("op_47080_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4853_cast_fp16, y = var_47080_to_fp16)[name = string("aw_chunk_4853_cast_fp16")]; + fp16 var_47082_to_fp16 = const()[name = string("op_47082_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4855_cast_fp16, y = var_47082_to_fp16)[name = string("aw_chunk_4855_cast_fp16")]; + fp16 var_47084_to_fp16 = const()[name = string("op_47084_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4857_cast_fp16, y = var_47084_to_fp16)[name = string("aw_chunk_4857_cast_fp16")]; + fp16 var_47086_to_fp16 = const()[name = string("op_47086_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4859_cast_fp16, y = var_47086_to_fp16)[name = string("aw_chunk_4859_cast_fp16")]; + fp16 var_47088_to_fp16 = const()[name = string("op_47088_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4861_cast_fp16, y = var_47088_to_fp16)[name = string("aw_chunk_4861_cast_fp16")]; + fp16 var_47090_to_fp16 = const()[name = string("op_47090_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4863_cast_fp16, y = var_47090_to_fp16)[name = string("aw_chunk_4863_cast_fp16")]; + fp16 var_47092_to_fp16 = const()[name = string("op_47092_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4865_cast_fp16, y = var_47092_to_fp16)[name = string("aw_chunk_4865_cast_fp16")]; + fp16 var_47094_to_fp16 = const()[name = string("op_47094_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4867_cast_fp16, y = var_47094_to_fp16)[name = string("aw_chunk_4867_cast_fp16")]; + fp16 var_47096_to_fp16 = const()[name = string("op_47096_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4869_cast_fp16, y = var_47096_to_fp16)[name = string("aw_chunk_4869_cast_fp16")]; + fp16 var_47098_to_fp16 = const()[name = string("op_47098_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4871_cast_fp16, y = var_47098_to_fp16)[name = string("aw_chunk_4871_cast_fp16")]; + fp16 var_47100_to_fp16 = const()[name = string("op_47100_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4873_cast_fp16, y = var_47100_to_fp16)[name = string("aw_chunk_4873_cast_fp16")]; + fp16 var_47102_to_fp16 = const()[name = string("op_47102_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4875_cast_fp16, y = var_47102_to_fp16)[name = string("aw_chunk_4875_cast_fp16")]; + fp16 var_47104_to_fp16 = const()[name = string("op_47104_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4877_cast_fp16, y = var_47104_to_fp16)[name = string("aw_chunk_4877_cast_fp16")]; + fp16 var_47106_to_fp16 = const()[name = string("op_47106_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4879_cast_fp16, y = var_47106_to_fp16)[name = string("aw_chunk_4879_cast_fp16")]; + fp16 var_47108_to_fp16 = const()[name = string("op_47108_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4881_cast_fp16, y = var_47108_to_fp16)[name = string("aw_chunk_4881_cast_fp16")]; + fp16 var_47110_to_fp16 = const()[name = string("op_47110_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4883_cast_fp16, y = var_47110_to_fp16)[name = string("aw_chunk_4883_cast_fp16")]; + fp16 var_47112_to_fp16 = const()[name = string("op_47112_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4885_cast_fp16, y = var_47112_to_fp16)[name = string("aw_chunk_4885_cast_fp16")]; + fp16 var_47114_to_fp16 = const()[name = string("op_47114_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4887_cast_fp16, y = var_47114_to_fp16)[name = string("aw_chunk_4887_cast_fp16")]; + fp16 var_47116_to_fp16 = const()[name = string("op_47116_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4889_cast_fp16, y = var_47116_to_fp16)[name = string("aw_chunk_4889_cast_fp16")]; + fp16 var_47118_to_fp16 = const()[name = string("op_47118_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4891_cast_fp16, y = var_47118_to_fp16)[name = string("aw_chunk_4891_cast_fp16")]; + fp16 var_47120_to_fp16 = const()[name = string("op_47120_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4893_cast_fp16, y = var_47120_to_fp16)[name = string("aw_chunk_4893_cast_fp16")]; + fp16 var_47122_to_fp16 = const()[name = string("op_47122_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4895_cast_fp16, y = var_47122_to_fp16)[name = string("aw_chunk_4895_cast_fp16")]; + fp16 var_47124_to_fp16 = const()[name = string("op_47124_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4897_cast_fp16, y = var_47124_to_fp16)[name = string("aw_chunk_4897_cast_fp16")]; + fp16 var_47126_to_fp16 = const()[name = string("op_47126_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4899_cast_fp16, y = var_47126_to_fp16)[name = string("aw_chunk_4899_cast_fp16")]; + fp16 var_47128_to_fp16 = const()[name = string("op_47128_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4901_cast_fp16, y = var_47128_to_fp16)[name = string("aw_chunk_4901_cast_fp16")]; + fp16 var_47130_to_fp16 = const()[name = string("op_47130_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4903_cast_fp16, y = var_47130_to_fp16)[name = string("aw_chunk_4903_cast_fp16")]; + fp16 var_47132_to_fp16 = const()[name = string("op_47132_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4905_cast_fp16, y = var_47132_to_fp16)[name = string("aw_chunk_4905_cast_fp16")]; + fp16 var_47134_to_fp16 = const()[name = string("op_47134_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4907_cast_fp16, y = var_47134_to_fp16)[name = string("aw_chunk_4907_cast_fp16")]; + fp16 var_47136_to_fp16 = const()[name = string("op_47136_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4909_cast_fp16, y = var_47136_to_fp16)[name = string("aw_chunk_4909_cast_fp16")]; + fp16 var_47138_to_fp16 = const()[name = string("op_47138_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4911_cast_fp16, y = var_47138_to_fp16)[name = string("aw_chunk_4911_cast_fp16")]; + fp16 var_47140_to_fp16 = const()[name = string("op_47140_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4913_cast_fp16, y = var_47140_to_fp16)[name = string("aw_chunk_4913_cast_fp16")]; + fp16 var_47142_to_fp16 = const()[name = string("op_47142_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4915_cast_fp16, y = var_47142_to_fp16)[name = string("aw_chunk_4915_cast_fp16")]; + fp16 var_47144_to_fp16 = const()[name = string("op_47144_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4917_cast_fp16, y = var_47144_to_fp16)[name = string("aw_chunk_4917_cast_fp16")]; + fp16 var_47146_to_fp16 = const()[name = string("op_47146_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4919_cast_fp16, y = var_47146_to_fp16)[name = string("aw_chunk_4919_cast_fp16")]; + fp16 var_47148_to_fp16 = const()[name = string("op_47148_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4921_cast_fp16, y = var_47148_to_fp16)[name = string("aw_chunk_4921_cast_fp16")]; + fp16 var_47150_to_fp16 = const()[name = string("op_47150_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4923_cast_fp16, y = var_47150_to_fp16)[name = string("aw_chunk_4923_cast_fp16")]; + fp16 var_47152_to_fp16 = const()[name = string("op_47152_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4925_cast_fp16, y = var_47152_to_fp16)[name = string("aw_chunk_4925_cast_fp16")]; + fp16 var_47154_to_fp16 = const()[name = string("op_47154_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4927_cast_fp16, y = var_47154_to_fp16)[name = string("aw_chunk_4927_cast_fp16")]; + fp16 var_47156_to_fp16 = const()[name = string("op_47156_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4929_cast_fp16, y = var_47156_to_fp16)[name = string("aw_chunk_4929_cast_fp16")]; + fp16 var_47158_to_fp16 = const()[name = string("op_47158_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4931_cast_fp16, y = var_47158_to_fp16)[name = string("aw_chunk_4931_cast_fp16")]; + fp16 var_47160_to_fp16 = const()[name = string("op_47160_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4933_cast_fp16, y = var_47160_to_fp16)[name = string("aw_chunk_4933_cast_fp16")]; + fp16 var_47162_to_fp16 = const()[name = string("op_47162_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4935_cast_fp16, y = var_47162_to_fp16)[name = string("aw_chunk_4935_cast_fp16")]; + fp16 var_47164_to_fp16 = const()[name = string("op_47164_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4937_cast_fp16, y = var_47164_to_fp16)[name = string("aw_chunk_4937_cast_fp16")]; + fp16 var_47166_to_fp16 = const()[name = string("op_47166_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4939_cast_fp16, y = var_47166_to_fp16)[name = string("aw_chunk_4939_cast_fp16")]; + fp16 var_47168_to_fp16 = const()[name = string("op_47168_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4941_cast_fp16, y = var_47168_to_fp16)[name = string("aw_chunk_4941_cast_fp16")]; + fp16 var_47170_to_fp16 = const()[name = string("op_47170_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4943_cast_fp16, y = var_47170_to_fp16)[name = string("aw_chunk_4943_cast_fp16")]; + fp16 var_47172_to_fp16 = const()[name = string("op_47172_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4945_cast_fp16, y = var_47172_to_fp16)[name = string("aw_chunk_4945_cast_fp16")]; + fp16 var_47174_to_fp16 = const()[name = string("op_47174_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4947_cast_fp16, y = var_47174_to_fp16)[name = string("aw_chunk_4947_cast_fp16")]; + fp16 var_47176_to_fp16 = const()[name = string("op_47176_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4949_cast_fp16, y = var_47176_to_fp16)[name = string("aw_chunk_4949_cast_fp16")]; + fp16 var_47178_to_fp16 = const()[name = string("op_47178_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4951_cast_fp16, y = var_47178_to_fp16)[name = string("aw_chunk_4951_cast_fp16")]; + fp16 var_47180_to_fp16 = const()[name = string("op_47180_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4953_cast_fp16, y = var_47180_to_fp16)[name = string("aw_chunk_4953_cast_fp16")]; + fp16 var_47182_to_fp16 = const()[name = string("op_47182_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4955_cast_fp16, y = var_47182_to_fp16)[name = string("aw_chunk_4955_cast_fp16")]; + fp16 var_47184_to_fp16 = const()[name = string("op_47184_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4957_cast_fp16, y = var_47184_to_fp16)[name = string("aw_chunk_4957_cast_fp16")]; + fp16 var_47186_to_fp16 = const()[name = string("op_47186_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4959_cast_fp16, y = var_47186_to_fp16)[name = string("aw_chunk_4959_cast_fp16")]; + tensor var_47188_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4801_cast_fp16)[name = string("op_47188_cast_fp16")]; + tensor var_47189_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4803_cast_fp16)[name = string("op_47189_cast_fp16")]; + tensor var_47190_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4805_cast_fp16)[name = string("op_47190_cast_fp16")]; + tensor var_47191_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4807_cast_fp16)[name = string("op_47191_cast_fp16")]; + tensor var_47192_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4809_cast_fp16)[name = string("op_47192_cast_fp16")]; + tensor var_47193_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4811_cast_fp16)[name = string("op_47193_cast_fp16")]; + tensor var_47194_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4813_cast_fp16)[name = string("op_47194_cast_fp16")]; + tensor var_47195_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4815_cast_fp16)[name = string("op_47195_cast_fp16")]; + tensor var_47196_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4817_cast_fp16)[name = string("op_47196_cast_fp16")]; + tensor var_47197_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4819_cast_fp16)[name = string("op_47197_cast_fp16")]; + tensor var_47198_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4821_cast_fp16)[name = string("op_47198_cast_fp16")]; + tensor var_47199_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4823_cast_fp16)[name = string("op_47199_cast_fp16")]; + tensor var_47200_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4825_cast_fp16)[name = string("op_47200_cast_fp16")]; + tensor var_47201_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4827_cast_fp16)[name = string("op_47201_cast_fp16")]; + tensor var_47202_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4829_cast_fp16)[name = string("op_47202_cast_fp16")]; + tensor var_47203_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4831_cast_fp16)[name = string("op_47203_cast_fp16")]; + tensor var_47204_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4833_cast_fp16)[name = string("op_47204_cast_fp16")]; + tensor var_47205_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4835_cast_fp16)[name = string("op_47205_cast_fp16")]; + tensor var_47206_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4837_cast_fp16)[name = string("op_47206_cast_fp16")]; + tensor var_47207_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4839_cast_fp16)[name = string("op_47207_cast_fp16")]; + tensor var_47208_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4841_cast_fp16)[name = string("op_47208_cast_fp16")]; + tensor var_47209_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4843_cast_fp16)[name = string("op_47209_cast_fp16")]; + tensor var_47210_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4845_cast_fp16)[name = string("op_47210_cast_fp16")]; + tensor var_47211_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4847_cast_fp16)[name = string("op_47211_cast_fp16")]; + tensor var_47212_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4849_cast_fp16)[name = string("op_47212_cast_fp16")]; + tensor var_47213_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4851_cast_fp16)[name = string("op_47213_cast_fp16")]; + tensor var_47214_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4853_cast_fp16)[name = string("op_47214_cast_fp16")]; + tensor var_47215_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4855_cast_fp16)[name = string("op_47215_cast_fp16")]; + tensor var_47216_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4857_cast_fp16)[name = string("op_47216_cast_fp16")]; + tensor var_47217_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4859_cast_fp16)[name = string("op_47217_cast_fp16")]; + tensor var_47218_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4861_cast_fp16)[name = string("op_47218_cast_fp16")]; + tensor var_47219_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4863_cast_fp16)[name = string("op_47219_cast_fp16")]; + tensor var_47220_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4865_cast_fp16)[name = string("op_47220_cast_fp16")]; + tensor var_47221_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4867_cast_fp16)[name = string("op_47221_cast_fp16")]; + tensor var_47222_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4869_cast_fp16)[name = string("op_47222_cast_fp16")]; + tensor var_47223_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4871_cast_fp16)[name = string("op_47223_cast_fp16")]; + tensor var_47224_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4873_cast_fp16)[name = string("op_47224_cast_fp16")]; + tensor var_47225_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4875_cast_fp16)[name = string("op_47225_cast_fp16")]; + tensor var_47226_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4877_cast_fp16)[name = string("op_47226_cast_fp16")]; + tensor var_47227_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4879_cast_fp16)[name = string("op_47227_cast_fp16")]; + tensor var_47228_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4881_cast_fp16)[name = string("op_47228_cast_fp16")]; + tensor var_47229_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4883_cast_fp16)[name = string("op_47229_cast_fp16")]; + tensor var_47230_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4885_cast_fp16)[name = string("op_47230_cast_fp16")]; + tensor var_47231_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4887_cast_fp16)[name = string("op_47231_cast_fp16")]; + tensor var_47232_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4889_cast_fp16)[name = string("op_47232_cast_fp16")]; + tensor var_47233_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4891_cast_fp16)[name = string("op_47233_cast_fp16")]; + tensor var_47234_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4893_cast_fp16)[name = string("op_47234_cast_fp16")]; + tensor var_47235_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4895_cast_fp16)[name = string("op_47235_cast_fp16")]; + tensor var_47236_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4897_cast_fp16)[name = string("op_47236_cast_fp16")]; + tensor var_47237_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4899_cast_fp16)[name = string("op_47237_cast_fp16")]; + tensor var_47238_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4901_cast_fp16)[name = string("op_47238_cast_fp16")]; + tensor var_47239_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4903_cast_fp16)[name = string("op_47239_cast_fp16")]; + tensor var_47240_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4905_cast_fp16)[name = string("op_47240_cast_fp16")]; + tensor var_47241_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4907_cast_fp16)[name = string("op_47241_cast_fp16")]; + tensor var_47242_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4909_cast_fp16)[name = string("op_47242_cast_fp16")]; + tensor var_47243_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4911_cast_fp16)[name = string("op_47243_cast_fp16")]; + tensor var_47244_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4913_cast_fp16)[name = string("op_47244_cast_fp16")]; + tensor var_47245_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4915_cast_fp16)[name = string("op_47245_cast_fp16")]; + tensor var_47246_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4917_cast_fp16)[name = string("op_47246_cast_fp16")]; + tensor var_47247_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4919_cast_fp16)[name = string("op_47247_cast_fp16")]; + tensor var_47248_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4921_cast_fp16)[name = string("op_47248_cast_fp16")]; + tensor var_47249_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4923_cast_fp16)[name = string("op_47249_cast_fp16")]; + tensor var_47250_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4925_cast_fp16)[name = string("op_47250_cast_fp16")]; + tensor var_47251_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4927_cast_fp16)[name = string("op_47251_cast_fp16")]; + tensor var_47252_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4929_cast_fp16)[name = string("op_47252_cast_fp16")]; + tensor var_47253_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4931_cast_fp16)[name = string("op_47253_cast_fp16")]; + tensor var_47254_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4933_cast_fp16)[name = string("op_47254_cast_fp16")]; + tensor var_47255_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4935_cast_fp16)[name = string("op_47255_cast_fp16")]; + tensor var_47256_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4937_cast_fp16)[name = string("op_47256_cast_fp16")]; + tensor var_47257_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4939_cast_fp16)[name = string("op_47257_cast_fp16")]; + tensor var_47258_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4941_cast_fp16)[name = string("op_47258_cast_fp16")]; + tensor var_47259_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4943_cast_fp16)[name = string("op_47259_cast_fp16")]; + tensor var_47260_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4945_cast_fp16)[name = string("op_47260_cast_fp16")]; + tensor var_47261_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4947_cast_fp16)[name = string("op_47261_cast_fp16")]; + tensor var_47262_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4949_cast_fp16)[name = string("op_47262_cast_fp16")]; + tensor var_47263_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4951_cast_fp16)[name = string("op_47263_cast_fp16")]; + tensor var_47264_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4953_cast_fp16)[name = string("op_47264_cast_fp16")]; + tensor var_47265_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4955_cast_fp16)[name = string("op_47265_cast_fp16")]; + tensor var_47266_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4957_cast_fp16)[name = string("op_47266_cast_fp16")]; + tensor var_47267_cast_fp16 = softmax(axis = var_46013, x = aw_chunk_4959_cast_fp16)[name = string("op_47267_cast_fp16")]; + string var_47269_equation_0 = const()[name = string("op_47269_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47269_cast_fp16 = einsum(equation = var_47269_equation_0, values = (var_46789_cast_fp16, var_47188_cast_fp16))[name = string("op_47269_cast_fp16")]; + string var_47271_equation_0 = const()[name = string("op_47271_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47271_cast_fp16 = einsum(equation = var_47271_equation_0, values = (var_46789_cast_fp16, var_47189_cast_fp16))[name = string("op_47271_cast_fp16")]; + string var_47273_equation_0 = const()[name = string("op_47273_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47273_cast_fp16 = einsum(equation = var_47273_equation_0, values = (var_46789_cast_fp16, var_47190_cast_fp16))[name = string("op_47273_cast_fp16")]; + string var_47275_equation_0 = const()[name = string("op_47275_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47275_cast_fp16 = einsum(equation = var_47275_equation_0, values = (var_46789_cast_fp16, var_47191_cast_fp16))[name = string("op_47275_cast_fp16")]; + string var_47277_equation_0 = const()[name = string("op_47277_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47277_cast_fp16 = einsum(equation = var_47277_equation_0, values = (var_46793_cast_fp16, var_47192_cast_fp16))[name = string("op_47277_cast_fp16")]; + string var_47279_equation_0 = const()[name = string("op_47279_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47279_cast_fp16 = einsum(equation = var_47279_equation_0, values = (var_46793_cast_fp16, var_47193_cast_fp16))[name = string("op_47279_cast_fp16")]; + string var_47281_equation_0 = const()[name = string("op_47281_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47281_cast_fp16 = einsum(equation = var_47281_equation_0, values = (var_46793_cast_fp16, var_47194_cast_fp16))[name = string("op_47281_cast_fp16")]; + string var_47283_equation_0 = const()[name = string("op_47283_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47283_cast_fp16 = einsum(equation = var_47283_equation_0, values = (var_46793_cast_fp16, var_47195_cast_fp16))[name = string("op_47283_cast_fp16")]; + string var_47285_equation_0 = const()[name = string("op_47285_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47285_cast_fp16 = einsum(equation = var_47285_equation_0, values = (var_46797_cast_fp16, var_47196_cast_fp16))[name = string("op_47285_cast_fp16")]; + string var_47287_equation_0 = const()[name = string("op_47287_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47287_cast_fp16 = einsum(equation = var_47287_equation_0, values = (var_46797_cast_fp16, var_47197_cast_fp16))[name = string("op_47287_cast_fp16")]; + string var_47289_equation_0 = const()[name = string("op_47289_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47289_cast_fp16 = einsum(equation = var_47289_equation_0, values = (var_46797_cast_fp16, var_47198_cast_fp16))[name = string("op_47289_cast_fp16")]; + string var_47291_equation_0 = const()[name = string("op_47291_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47291_cast_fp16 = einsum(equation = var_47291_equation_0, values = (var_46797_cast_fp16, var_47199_cast_fp16))[name = string("op_47291_cast_fp16")]; + string var_47293_equation_0 = const()[name = string("op_47293_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47293_cast_fp16 = einsum(equation = var_47293_equation_0, values = (var_46801_cast_fp16, var_47200_cast_fp16))[name = string("op_47293_cast_fp16")]; + string var_47295_equation_0 = const()[name = string("op_47295_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47295_cast_fp16 = einsum(equation = var_47295_equation_0, values = (var_46801_cast_fp16, var_47201_cast_fp16))[name = string("op_47295_cast_fp16")]; + string var_47297_equation_0 = const()[name = string("op_47297_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47297_cast_fp16 = einsum(equation = var_47297_equation_0, values = (var_46801_cast_fp16, var_47202_cast_fp16))[name = string("op_47297_cast_fp16")]; + string var_47299_equation_0 = const()[name = string("op_47299_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47299_cast_fp16 = einsum(equation = var_47299_equation_0, values = (var_46801_cast_fp16, var_47203_cast_fp16))[name = string("op_47299_cast_fp16")]; + string var_47301_equation_0 = const()[name = string("op_47301_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47301_cast_fp16 = einsum(equation = var_47301_equation_0, values = (var_46805_cast_fp16, var_47204_cast_fp16))[name = string("op_47301_cast_fp16")]; + string var_47303_equation_0 = const()[name = string("op_47303_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47303_cast_fp16 = einsum(equation = var_47303_equation_0, values = (var_46805_cast_fp16, var_47205_cast_fp16))[name = string("op_47303_cast_fp16")]; + string var_47305_equation_0 = const()[name = string("op_47305_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47305_cast_fp16 = einsum(equation = var_47305_equation_0, values = (var_46805_cast_fp16, var_47206_cast_fp16))[name = string("op_47305_cast_fp16")]; + string var_47307_equation_0 = const()[name = string("op_47307_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47307_cast_fp16 = einsum(equation = var_47307_equation_0, values = (var_46805_cast_fp16, var_47207_cast_fp16))[name = string("op_47307_cast_fp16")]; + string var_47309_equation_0 = const()[name = string("op_47309_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47309_cast_fp16 = einsum(equation = var_47309_equation_0, values = (var_46809_cast_fp16, var_47208_cast_fp16))[name = string("op_47309_cast_fp16")]; + string var_47311_equation_0 = const()[name = string("op_47311_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47311_cast_fp16 = einsum(equation = var_47311_equation_0, values = (var_46809_cast_fp16, var_47209_cast_fp16))[name = string("op_47311_cast_fp16")]; + string var_47313_equation_0 = const()[name = string("op_47313_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47313_cast_fp16 = einsum(equation = var_47313_equation_0, values = (var_46809_cast_fp16, var_47210_cast_fp16))[name = string("op_47313_cast_fp16")]; + string var_47315_equation_0 = const()[name = string("op_47315_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47315_cast_fp16 = einsum(equation = var_47315_equation_0, values = (var_46809_cast_fp16, var_47211_cast_fp16))[name = string("op_47315_cast_fp16")]; + string var_47317_equation_0 = const()[name = string("op_47317_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47317_cast_fp16 = einsum(equation = var_47317_equation_0, values = (var_46813_cast_fp16, var_47212_cast_fp16))[name = string("op_47317_cast_fp16")]; + string var_47319_equation_0 = const()[name = string("op_47319_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47319_cast_fp16 = einsum(equation = var_47319_equation_0, values = (var_46813_cast_fp16, var_47213_cast_fp16))[name = string("op_47319_cast_fp16")]; + string var_47321_equation_0 = const()[name = string("op_47321_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47321_cast_fp16 = einsum(equation = var_47321_equation_0, values = (var_46813_cast_fp16, var_47214_cast_fp16))[name = string("op_47321_cast_fp16")]; + string var_47323_equation_0 = const()[name = string("op_47323_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47323_cast_fp16 = einsum(equation = var_47323_equation_0, values = (var_46813_cast_fp16, var_47215_cast_fp16))[name = string("op_47323_cast_fp16")]; + string var_47325_equation_0 = const()[name = string("op_47325_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47325_cast_fp16 = einsum(equation = var_47325_equation_0, values = (var_46817_cast_fp16, var_47216_cast_fp16))[name = string("op_47325_cast_fp16")]; + string var_47327_equation_0 = const()[name = string("op_47327_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47327_cast_fp16 = einsum(equation = var_47327_equation_0, values = (var_46817_cast_fp16, var_47217_cast_fp16))[name = string("op_47327_cast_fp16")]; + string var_47329_equation_0 = const()[name = string("op_47329_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47329_cast_fp16 = einsum(equation = var_47329_equation_0, values = (var_46817_cast_fp16, var_47218_cast_fp16))[name = string("op_47329_cast_fp16")]; + string var_47331_equation_0 = const()[name = string("op_47331_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47331_cast_fp16 = einsum(equation = var_47331_equation_0, values = (var_46817_cast_fp16, var_47219_cast_fp16))[name = string("op_47331_cast_fp16")]; + string var_47333_equation_0 = const()[name = string("op_47333_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47333_cast_fp16 = einsum(equation = var_47333_equation_0, values = (var_46821_cast_fp16, var_47220_cast_fp16))[name = string("op_47333_cast_fp16")]; + string var_47335_equation_0 = const()[name = string("op_47335_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47335_cast_fp16 = einsum(equation = var_47335_equation_0, values = (var_46821_cast_fp16, var_47221_cast_fp16))[name = string("op_47335_cast_fp16")]; + string var_47337_equation_0 = const()[name = string("op_47337_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47337_cast_fp16 = einsum(equation = var_47337_equation_0, values = (var_46821_cast_fp16, var_47222_cast_fp16))[name = string("op_47337_cast_fp16")]; + string var_47339_equation_0 = const()[name = string("op_47339_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47339_cast_fp16 = einsum(equation = var_47339_equation_0, values = (var_46821_cast_fp16, var_47223_cast_fp16))[name = string("op_47339_cast_fp16")]; + string var_47341_equation_0 = const()[name = string("op_47341_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47341_cast_fp16 = einsum(equation = var_47341_equation_0, values = (var_46825_cast_fp16, var_47224_cast_fp16))[name = string("op_47341_cast_fp16")]; + string var_47343_equation_0 = const()[name = string("op_47343_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47343_cast_fp16 = einsum(equation = var_47343_equation_0, values = (var_46825_cast_fp16, var_47225_cast_fp16))[name = string("op_47343_cast_fp16")]; + string var_47345_equation_0 = const()[name = string("op_47345_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47345_cast_fp16 = einsum(equation = var_47345_equation_0, values = (var_46825_cast_fp16, var_47226_cast_fp16))[name = string("op_47345_cast_fp16")]; + string var_47347_equation_0 = const()[name = string("op_47347_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47347_cast_fp16 = einsum(equation = var_47347_equation_0, values = (var_46825_cast_fp16, var_47227_cast_fp16))[name = string("op_47347_cast_fp16")]; + string var_47349_equation_0 = const()[name = string("op_47349_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47349_cast_fp16 = einsum(equation = var_47349_equation_0, values = (var_46829_cast_fp16, var_47228_cast_fp16))[name = string("op_47349_cast_fp16")]; + string var_47351_equation_0 = const()[name = string("op_47351_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47351_cast_fp16 = einsum(equation = var_47351_equation_0, values = (var_46829_cast_fp16, var_47229_cast_fp16))[name = string("op_47351_cast_fp16")]; + string var_47353_equation_0 = const()[name = string("op_47353_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47353_cast_fp16 = einsum(equation = var_47353_equation_0, values = (var_46829_cast_fp16, var_47230_cast_fp16))[name = string("op_47353_cast_fp16")]; + string var_47355_equation_0 = const()[name = string("op_47355_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47355_cast_fp16 = einsum(equation = var_47355_equation_0, values = (var_46829_cast_fp16, var_47231_cast_fp16))[name = string("op_47355_cast_fp16")]; + string var_47357_equation_0 = const()[name = string("op_47357_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47357_cast_fp16 = einsum(equation = var_47357_equation_0, values = (var_46833_cast_fp16, var_47232_cast_fp16))[name = string("op_47357_cast_fp16")]; + string var_47359_equation_0 = const()[name = string("op_47359_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47359_cast_fp16 = einsum(equation = var_47359_equation_0, values = (var_46833_cast_fp16, var_47233_cast_fp16))[name = string("op_47359_cast_fp16")]; + string var_47361_equation_0 = const()[name = string("op_47361_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47361_cast_fp16 = einsum(equation = var_47361_equation_0, values = (var_46833_cast_fp16, var_47234_cast_fp16))[name = string("op_47361_cast_fp16")]; + string var_47363_equation_0 = const()[name = string("op_47363_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47363_cast_fp16 = einsum(equation = var_47363_equation_0, values = (var_46833_cast_fp16, var_47235_cast_fp16))[name = string("op_47363_cast_fp16")]; + string var_47365_equation_0 = const()[name = string("op_47365_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47365_cast_fp16 = einsum(equation = var_47365_equation_0, values = (var_46837_cast_fp16, var_47236_cast_fp16))[name = string("op_47365_cast_fp16")]; + string var_47367_equation_0 = const()[name = string("op_47367_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47367_cast_fp16 = einsum(equation = var_47367_equation_0, values = (var_46837_cast_fp16, var_47237_cast_fp16))[name = string("op_47367_cast_fp16")]; + string var_47369_equation_0 = const()[name = string("op_47369_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47369_cast_fp16 = einsum(equation = var_47369_equation_0, values = (var_46837_cast_fp16, var_47238_cast_fp16))[name = string("op_47369_cast_fp16")]; + string var_47371_equation_0 = const()[name = string("op_47371_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47371_cast_fp16 = einsum(equation = var_47371_equation_0, values = (var_46837_cast_fp16, var_47239_cast_fp16))[name = string("op_47371_cast_fp16")]; + string var_47373_equation_0 = const()[name = string("op_47373_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47373_cast_fp16 = einsum(equation = var_47373_equation_0, values = (var_46841_cast_fp16, var_47240_cast_fp16))[name = string("op_47373_cast_fp16")]; + string var_47375_equation_0 = const()[name = string("op_47375_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47375_cast_fp16 = einsum(equation = var_47375_equation_0, values = (var_46841_cast_fp16, var_47241_cast_fp16))[name = string("op_47375_cast_fp16")]; + string var_47377_equation_0 = const()[name = string("op_47377_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47377_cast_fp16 = einsum(equation = var_47377_equation_0, values = (var_46841_cast_fp16, var_47242_cast_fp16))[name = string("op_47377_cast_fp16")]; + string var_47379_equation_0 = const()[name = string("op_47379_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47379_cast_fp16 = einsum(equation = var_47379_equation_0, values = (var_46841_cast_fp16, var_47243_cast_fp16))[name = string("op_47379_cast_fp16")]; + string var_47381_equation_0 = const()[name = string("op_47381_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47381_cast_fp16 = einsum(equation = var_47381_equation_0, values = (var_46845_cast_fp16, var_47244_cast_fp16))[name = string("op_47381_cast_fp16")]; + string var_47383_equation_0 = const()[name = string("op_47383_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47383_cast_fp16 = einsum(equation = var_47383_equation_0, values = (var_46845_cast_fp16, var_47245_cast_fp16))[name = string("op_47383_cast_fp16")]; + string var_47385_equation_0 = const()[name = string("op_47385_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47385_cast_fp16 = einsum(equation = var_47385_equation_0, values = (var_46845_cast_fp16, var_47246_cast_fp16))[name = string("op_47385_cast_fp16")]; + string var_47387_equation_0 = const()[name = string("op_47387_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47387_cast_fp16 = einsum(equation = var_47387_equation_0, values = (var_46845_cast_fp16, var_47247_cast_fp16))[name = string("op_47387_cast_fp16")]; + string var_47389_equation_0 = const()[name = string("op_47389_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47389_cast_fp16 = einsum(equation = var_47389_equation_0, values = (var_46849_cast_fp16, var_47248_cast_fp16))[name = string("op_47389_cast_fp16")]; + string var_47391_equation_0 = const()[name = string("op_47391_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47391_cast_fp16 = einsum(equation = var_47391_equation_0, values = (var_46849_cast_fp16, var_47249_cast_fp16))[name = string("op_47391_cast_fp16")]; + string var_47393_equation_0 = const()[name = string("op_47393_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47393_cast_fp16 = einsum(equation = var_47393_equation_0, values = (var_46849_cast_fp16, var_47250_cast_fp16))[name = string("op_47393_cast_fp16")]; + string var_47395_equation_0 = const()[name = string("op_47395_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47395_cast_fp16 = einsum(equation = var_47395_equation_0, values = (var_46849_cast_fp16, var_47251_cast_fp16))[name = string("op_47395_cast_fp16")]; + string var_47397_equation_0 = const()[name = string("op_47397_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47397_cast_fp16 = einsum(equation = var_47397_equation_0, values = (var_46853_cast_fp16, var_47252_cast_fp16))[name = string("op_47397_cast_fp16")]; + string var_47399_equation_0 = const()[name = string("op_47399_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47399_cast_fp16 = einsum(equation = var_47399_equation_0, values = (var_46853_cast_fp16, var_47253_cast_fp16))[name = string("op_47399_cast_fp16")]; + string var_47401_equation_0 = const()[name = string("op_47401_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47401_cast_fp16 = einsum(equation = var_47401_equation_0, values = (var_46853_cast_fp16, var_47254_cast_fp16))[name = string("op_47401_cast_fp16")]; + string var_47403_equation_0 = const()[name = string("op_47403_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47403_cast_fp16 = einsum(equation = var_47403_equation_0, values = (var_46853_cast_fp16, var_47255_cast_fp16))[name = string("op_47403_cast_fp16")]; + string var_47405_equation_0 = const()[name = string("op_47405_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47405_cast_fp16 = einsum(equation = var_47405_equation_0, values = (var_46857_cast_fp16, var_47256_cast_fp16))[name = string("op_47405_cast_fp16")]; + string var_47407_equation_0 = const()[name = string("op_47407_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47407_cast_fp16 = einsum(equation = var_47407_equation_0, values = (var_46857_cast_fp16, var_47257_cast_fp16))[name = string("op_47407_cast_fp16")]; + string var_47409_equation_0 = const()[name = string("op_47409_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47409_cast_fp16 = einsum(equation = var_47409_equation_0, values = (var_46857_cast_fp16, var_47258_cast_fp16))[name = string("op_47409_cast_fp16")]; + string var_47411_equation_0 = const()[name = string("op_47411_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47411_cast_fp16 = einsum(equation = var_47411_equation_0, values = (var_46857_cast_fp16, var_47259_cast_fp16))[name = string("op_47411_cast_fp16")]; + string var_47413_equation_0 = const()[name = string("op_47413_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47413_cast_fp16 = einsum(equation = var_47413_equation_0, values = (var_46861_cast_fp16, var_47260_cast_fp16))[name = string("op_47413_cast_fp16")]; + string var_47415_equation_0 = const()[name = string("op_47415_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47415_cast_fp16 = einsum(equation = var_47415_equation_0, values = (var_46861_cast_fp16, var_47261_cast_fp16))[name = string("op_47415_cast_fp16")]; + string var_47417_equation_0 = const()[name = string("op_47417_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47417_cast_fp16 = einsum(equation = var_47417_equation_0, values = (var_46861_cast_fp16, var_47262_cast_fp16))[name = string("op_47417_cast_fp16")]; + string var_47419_equation_0 = const()[name = string("op_47419_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47419_cast_fp16 = einsum(equation = var_47419_equation_0, values = (var_46861_cast_fp16, var_47263_cast_fp16))[name = string("op_47419_cast_fp16")]; + string var_47421_equation_0 = const()[name = string("op_47421_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47421_cast_fp16 = einsum(equation = var_47421_equation_0, values = (var_46865_cast_fp16, var_47264_cast_fp16))[name = string("op_47421_cast_fp16")]; + string var_47423_equation_0 = const()[name = string("op_47423_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47423_cast_fp16 = einsum(equation = var_47423_equation_0, values = (var_46865_cast_fp16, var_47265_cast_fp16))[name = string("op_47423_cast_fp16")]; + string var_47425_equation_0 = const()[name = string("op_47425_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47425_cast_fp16 = einsum(equation = var_47425_equation_0, values = (var_46865_cast_fp16, var_47266_cast_fp16))[name = string("op_47425_cast_fp16")]; + string var_47427_equation_0 = const()[name = string("op_47427_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47427_cast_fp16 = einsum(equation = var_47427_equation_0, values = (var_46865_cast_fp16, var_47267_cast_fp16))[name = string("op_47427_cast_fp16")]; + bool var_47429_interleave_0 = const()[name = string("op_47429_interleave_0"), val = bool(false)]; + tensor var_47429_cast_fp16 = concat(axis = var_45988, interleave = var_47429_interleave_0, values = (var_47269_cast_fp16, var_47271_cast_fp16, var_47273_cast_fp16, var_47275_cast_fp16))[name = string("op_47429_cast_fp16")]; + bool var_47431_interleave_0 = const()[name = string("op_47431_interleave_0"), val = bool(false)]; + tensor var_47431_cast_fp16 = concat(axis = var_45988, interleave = var_47431_interleave_0, values = (var_47277_cast_fp16, var_47279_cast_fp16, var_47281_cast_fp16, var_47283_cast_fp16))[name = string("op_47431_cast_fp16")]; + bool var_47433_interleave_0 = const()[name = string("op_47433_interleave_0"), val = bool(false)]; + tensor var_47433_cast_fp16 = concat(axis = var_45988, interleave = var_47433_interleave_0, values = (var_47285_cast_fp16, var_47287_cast_fp16, var_47289_cast_fp16, var_47291_cast_fp16))[name = string("op_47433_cast_fp16")]; + bool var_47435_interleave_0 = const()[name = string("op_47435_interleave_0"), val = bool(false)]; + tensor var_47435_cast_fp16 = concat(axis = var_45988, interleave = var_47435_interleave_0, values = (var_47293_cast_fp16, var_47295_cast_fp16, var_47297_cast_fp16, var_47299_cast_fp16))[name = string("op_47435_cast_fp16")]; + bool var_47437_interleave_0 = const()[name = string("op_47437_interleave_0"), val = bool(false)]; + tensor var_47437_cast_fp16 = concat(axis = var_45988, interleave = var_47437_interleave_0, values = (var_47301_cast_fp16, var_47303_cast_fp16, var_47305_cast_fp16, var_47307_cast_fp16))[name = string("op_47437_cast_fp16")]; + bool var_47439_interleave_0 = const()[name = string("op_47439_interleave_0"), val = bool(false)]; + tensor var_47439_cast_fp16 = concat(axis = var_45988, interleave = var_47439_interleave_0, values = (var_47309_cast_fp16, var_47311_cast_fp16, var_47313_cast_fp16, var_47315_cast_fp16))[name = string("op_47439_cast_fp16")]; + bool var_47441_interleave_0 = const()[name = string("op_47441_interleave_0"), val = bool(false)]; + tensor var_47441_cast_fp16 = concat(axis = var_45988, interleave = var_47441_interleave_0, values = (var_47317_cast_fp16, var_47319_cast_fp16, var_47321_cast_fp16, var_47323_cast_fp16))[name = string("op_47441_cast_fp16")]; + bool var_47443_interleave_0 = const()[name = string("op_47443_interleave_0"), val = bool(false)]; + tensor var_47443_cast_fp16 = concat(axis = var_45988, interleave = var_47443_interleave_0, values = (var_47325_cast_fp16, var_47327_cast_fp16, var_47329_cast_fp16, var_47331_cast_fp16))[name = string("op_47443_cast_fp16")]; + bool var_47445_interleave_0 = const()[name = string("op_47445_interleave_0"), val = bool(false)]; + tensor var_47445_cast_fp16 = concat(axis = var_45988, interleave = var_47445_interleave_0, values = (var_47333_cast_fp16, var_47335_cast_fp16, var_47337_cast_fp16, var_47339_cast_fp16))[name = string("op_47445_cast_fp16")]; + bool var_47447_interleave_0 = const()[name = string("op_47447_interleave_0"), val = bool(false)]; + tensor var_47447_cast_fp16 = concat(axis = var_45988, interleave = var_47447_interleave_0, values = (var_47341_cast_fp16, var_47343_cast_fp16, var_47345_cast_fp16, var_47347_cast_fp16))[name = string("op_47447_cast_fp16")]; + bool var_47449_interleave_0 = const()[name = string("op_47449_interleave_0"), val = bool(false)]; + tensor var_47449_cast_fp16 = concat(axis = var_45988, interleave = var_47449_interleave_0, values = (var_47349_cast_fp16, var_47351_cast_fp16, var_47353_cast_fp16, var_47355_cast_fp16))[name = string("op_47449_cast_fp16")]; + bool var_47451_interleave_0 = const()[name = string("op_47451_interleave_0"), val = bool(false)]; + tensor var_47451_cast_fp16 = concat(axis = var_45988, interleave = var_47451_interleave_0, values = (var_47357_cast_fp16, var_47359_cast_fp16, var_47361_cast_fp16, var_47363_cast_fp16))[name = string("op_47451_cast_fp16")]; + bool var_47453_interleave_0 = const()[name = string("op_47453_interleave_0"), val = bool(false)]; + tensor var_47453_cast_fp16 = concat(axis = var_45988, interleave = var_47453_interleave_0, values = (var_47365_cast_fp16, var_47367_cast_fp16, var_47369_cast_fp16, var_47371_cast_fp16))[name = string("op_47453_cast_fp16")]; + bool var_47455_interleave_0 = const()[name = string("op_47455_interleave_0"), val = bool(false)]; + tensor var_47455_cast_fp16 = concat(axis = var_45988, interleave = var_47455_interleave_0, values = (var_47373_cast_fp16, var_47375_cast_fp16, var_47377_cast_fp16, var_47379_cast_fp16))[name = string("op_47455_cast_fp16")]; + bool var_47457_interleave_0 = const()[name = string("op_47457_interleave_0"), val = bool(false)]; + tensor var_47457_cast_fp16 = concat(axis = var_45988, interleave = var_47457_interleave_0, values = (var_47381_cast_fp16, var_47383_cast_fp16, var_47385_cast_fp16, var_47387_cast_fp16))[name = string("op_47457_cast_fp16")]; + bool var_47459_interleave_0 = const()[name = string("op_47459_interleave_0"), val = bool(false)]; + tensor var_47459_cast_fp16 = concat(axis = var_45988, interleave = var_47459_interleave_0, values = (var_47389_cast_fp16, var_47391_cast_fp16, var_47393_cast_fp16, var_47395_cast_fp16))[name = string("op_47459_cast_fp16")]; + bool var_47461_interleave_0 = const()[name = string("op_47461_interleave_0"), val = bool(false)]; + tensor var_47461_cast_fp16 = concat(axis = var_45988, interleave = var_47461_interleave_0, values = (var_47397_cast_fp16, var_47399_cast_fp16, var_47401_cast_fp16, var_47403_cast_fp16))[name = string("op_47461_cast_fp16")]; + bool var_47463_interleave_0 = const()[name = string("op_47463_interleave_0"), val = bool(false)]; + tensor var_47463_cast_fp16 = concat(axis = var_45988, interleave = var_47463_interleave_0, values = (var_47405_cast_fp16, var_47407_cast_fp16, var_47409_cast_fp16, var_47411_cast_fp16))[name = string("op_47463_cast_fp16")]; + bool var_47465_interleave_0 = const()[name = string("op_47465_interleave_0"), val = bool(false)]; + tensor var_47465_cast_fp16 = concat(axis = var_45988, interleave = var_47465_interleave_0, values = (var_47413_cast_fp16, var_47415_cast_fp16, var_47417_cast_fp16, var_47419_cast_fp16))[name = string("op_47465_cast_fp16")]; + bool var_47467_interleave_0 = const()[name = string("op_47467_interleave_0"), val = bool(false)]; + tensor var_47467_cast_fp16 = concat(axis = var_45988, interleave = var_47467_interleave_0, values = (var_47421_cast_fp16, var_47423_cast_fp16, var_47425_cast_fp16, var_47427_cast_fp16))[name = string("op_47467_cast_fp16")]; + bool input_241_interleave_0 = const()[name = string("input_241_interleave_0"), val = bool(false)]; + tensor input_241_cast_fp16 = concat(axis = var_46013, interleave = input_241_interleave_0, values = (var_47429_cast_fp16, var_47431_cast_fp16, var_47433_cast_fp16, var_47435_cast_fp16, var_47437_cast_fp16, var_47439_cast_fp16, var_47441_cast_fp16, var_47443_cast_fp16, var_47445_cast_fp16, var_47447_cast_fp16, var_47449_cast_fp16, var_47451_cast_fp16, var_47453_cast_fp16, var_47455_cast_fp16, var_47457_cast_fp16, var_47459_cast_fp16, var_47461_cast_fp16, var_47463_cast_fp16, var_47465_cast_fp16, var_47467_cast_fp16))[name = string("input_241_cast_fp16")]; + string obj_123_pad_type_0 = const()[name = string("obj_123_pad_type_0"), val = string("valid")]; + tensor obj_123_strides_0 = const()[name = string("obj_123_strides_0"), val = tensor([1, 1])]; + tensor obj_123_pad_0 = const()[name = string("obj_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_123_dilations_0 = const()[name = string("obj_123_dilations_0"), val = tensor([1, 1])]; + int32 obj_123_groups_0 = const()[name = string("obj_123_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205103680)))]; + tensor layers_30_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208380544)))]; + tensor obj_123_cast_fp16 = conv(bias = layers_30_self_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_30_self_attn_o_proj_weight_to_fp16, x = input_241_cast_fp16)[name = string("obj_123_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_123_cast_fp16)[name = string("inputs_123_cast_fp16")]; + tensor out_123_axes_0 = const()[name = string("out_123_axes_0"), val = tensor([1])]; + fp16 var_47486_to_fp16 = const()[name = string("op_47486_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_47486_to_fp16, x = inputs_123_cast_fp16)[name = string("out_123_cast_fp16")]; + tensor input_243_gamma_0_to_fp16 = const()[name = string("input_243_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208383168)))]; + tensor input_243_beta_0_to_fp16 = const()[name = string("input_243_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208385792)))]; + fp16 input_243_epsilon_0_to_fp16 = const()[name = string("input_243_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_243_cast_fp16 = batch_norm(beta = input_243_beta_0_to_fp16, epsilon = input_243_epsilon_0_to_fp16, gamma = input_243_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = string("input_243_cast_fp16")]; + string input_245_pad_type_0 = const()[name = string("input_245_pad_type_0"), val = string("valid")]; + tensor input_245_strides_0 = const()[name = string("input_245_strides_0"), val = tensor([1, 1])]; + tensor input_245_pad_0 = const()[name = string("input_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_245_dilations_0 = const()[name = string("input_245_dilations_0"), val = tensor([1, 1])]; + int32 input_245_groups_0 = const()[name = string("input_245_groups_0"), val = int32(1)]; + tensor layers_30_fc1_weight_to_fp16 = const()[name = string("layers_30_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208388416)))]; + tensor layers_30_fc1_bias_to_fp16 = const()[name = string("layers_30_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221495680)))]; + tensor input_245_cast_fp16 = conv(bias = layers_30_fc1_bias_to_fp16, dilations = input_245_dilations_0, groups = input_245_groups_0, pad = input_245_pad_0, pad_type = input_245_pad_type_0, strides = input_245_strides_0, weight = layers_30_fc1_weight_to_fp16, x = input_243_cast_fp16)[name = string("input_245_cast_fp16")]; + string input_247_mode_0 = const()[name = string("input_247_mode_0"), val = string("EXACT")]; + tensor input_247_cast_fp16 = gelu(mode = input_247_mode_0, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")]; + string hidden_states_65_pad_type_0 = const()[name = string("hidden_states_65_pad_type_0"), val = string("valid")]; + tensor hidden_states_65_strides_0 = const()[name = string("hidden_states_65_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_65_pad_0 = const()[name = string("hidden_states_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_65_dilations_0 = const()[name = string("hidden_states_65_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_65_groups_0 = const()[name = string("hidden_states_65_groups_0"), val = int32(1)]; + tensor layers_30_fc2_weight_to_fp16 = const()[name = string("layers_30_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221505984)))]; + tensor layers_30_fc2_bias_to_fp16 = const()[name = string("layers_30_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234613248)))]; + tensor hidden_states_65_cast_fp16 = conv(bias = layers_30_fc2_bias_to_fp16, dilations = hidden_states_65_dilations_0, groups = hidden_states_65_groups_0, pad = hidden_states_65_pad_0, pad_type = hidden_states_65_pad_type_0, strides = hidden_states_65_strides_0, weight = layers_30_fc2_weight_to_fp16, x = input_247_cast_fp16)[name = string("hidden_states_65_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = hidden_states_65_cast_fp16)[name = string("inputs_125_cast_fp16")]; + int32 var_47515 = const()[name = string("op_47515"), val = int32(3)]; + int32 var_47540 = const()[name = string("op_47540"), val = int32(1)]; + tensor out_125_axes_0 = const()[name = string("out_125_axes_0"), val = tensor([1])]; + fp16 var_47557_to_fp16 = const()[name = string("op_47557_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_47557_to_fp16, x = inputs_125_cast_fp16)[name = string("out_125_cast_fp16")]; + tensor obj_125_gamma_0_to_fp16 = const()[name = string("obj_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234615872)))]; + tensor obj_125_beta_0_to_fp16 = const()[name = string("obj_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234618496)))]; + fp16 obj_125_epsilon_0_to_fp16 = const()[name = string("obj_125_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_125_cast_fp16 = batch_norm(beta = obj_125_beta_0_to_fp16, epsilon = obj_125_epsilon_0_to_fp16, gamma = obj_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = string("obj_125_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234621120)))]; + tensor layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237897984)))]; + tensor query_cast_fp16 = conv(bias = layers_31_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_31_self_attn_q_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237900608)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_31_self_attn_k_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1241177472)))]; + tensor layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244454336)))]; + tensor value_cast_fp16 = conv(bias = layers_31_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_31_self_attn_v_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_47595_begin_0 = const()[name = string("op_47595_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47595_end_0 = const()[name = string("op_47595_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47595_end_mask_0 = const()[name = string("op_47595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47595_cast_fp16 = slice_by_index(begin = var_47595_begin_0, end = var_47595_end_0, end_mask = var_47595_end_mask_0, x = query_cast_fp16)[name = string("op_47595_cast_fp16")]; + tensor var_47599_begin_0 = const()[name = string("op_47599_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47599_end_0 = const()[name = string("op_47599_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_47599_end_mask_0 = const()[name = string("op_47599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47599_cast_fp16 = slice_by_index(begin = var_47599_begin_0, end = var_47599_end_0, end_mask = var_47599_end_mask_0, x = query_cast_fp16)[name = string("op_47599_cast_fp16")]; + tensor var_47603_begin_0 = const()[name = string("op_47603_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47603_end_0 = const()[name = string("op_47603_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_47603_end_mask_0 = const()[name = string("op_47603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47603_cast_fp16 = slice_by_index(begin = var_47603_begin_0, end = var_47603_end_0, end_mask = var_47603_end_mask_0, x = query_cast_fp16)[name = string("op_47603_cast_fp16")]; + tensor var_47607_begin_0 = const()[name = string("op_47607_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47607_end_0 = const()[name = string("op_47607_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_47607_end_mask_0 = const()[name = string("op_47607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47607_cast_fp16 = slice_by_index(begin = var_47607_begin_0, end = var_47607_end_0, end_mask = var_47607_end_mask_0, x = query_cast_fp16)[name = string("op_47607_cast_fp16")]; + tensor var_47611_begin_0 = const()[name = string("op_47611_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47611_end_0 = const()[name = string("op_47611_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_47611_end_mask_0 = const()[name = string("op_47611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47611_cast_fp16 = slice_by_index(begin = var_47611_begin_0, end = var_47611_end_0, end_mask = var_47611_end_mask_0, x = query_cast_fp16)[name = string("op_47611_cast_fp16")]; + tensor var_47615_begin_0 = const()[name = string("op_47615_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47615_end_0 = const()[name = string("op_47615_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_47615_end_mask_0 = const()[name = string("op_47615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47615_cast_fp16 = slice_by_index(begin = var_47615_begin_0, end = var_47615_end_0, end_mask = var_47615_end_mask_0, x = query_cast_fp16)[name = string("op_47615_cast_fp16")]; + tensor var_47619_begin_0 = const()[name = string("op_47619_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47619_end_0 = const()[name = string("op_47619_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_47619_end_mask_0 = const()[name = string("op_47619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47619_cast_fp16 = slice_by_index(begin = var_47619_begin_0, end = var_47619_end_0, end_mask = var_47619_end_mask_0, x = query_cast_fp16)[name = string("op_47619_cast_fp16")]; + tensor var_47623_begin_0 = const()[name = string("op_47623_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47623_end_0 = const()[name = string("op_47623_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_47623_end_mask_0 = const()[name = string("op_47623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47623_cast_fp16 = slice_by_index(begin = var_47623_begin_0, end = var_47623_end_0, end_mask = var_47623_end_mask_0, x = query_cast_fp16)[name = string("op_47623_cast_fp16")]; + tensor var_47627_begin_0 = const()[name = string("op_47627_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47627_end_0 = const()[name = string("op_47627_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_47627_end_mask_0 = const()[name = string("op_47627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47627_cast_fp16 = slice_by_index(begin = var_47627_begin_0, end = var_47627_end_0, end_mask = var_47627_end_mask_0, x = query_cast_fp16)[name = string("op_47627_cast_fp16")]; + tensor var_47631_begin_0 = const()[name = string("op_47631_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47631_end_0 = const()[name = string("op_47631_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_47631_end_mask_0 = const()[name = string("op_47631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47631_cast_fp16 = slice_by_index(begin = var_47631_begin_0, end = var_47631_end_0, end_mask = var_47631_end_mask_0, x = query_cast_fp16)[name = string("op_47631_cast_fp16")]; + tensor var_47635_begin_0 = const()[name = string("op_47635_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47635_end_0 = const()[name = string("op_47635_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_47635_end_mask_0 = const()[name = string("op_47635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47635_cast_fp16 = slice_by_index(begin = var_47635_begin_0, end = var_47635_end_0, end_mask = var_47635_end_mask_0, x = query_cast_fp16)[name = string("op_47635_cast_fp16")]; + tensor var_47639_begin_0 = const()[name = string("op_47639_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47639_end_0 = const()[name = string("op_47639_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_47639_end_mask_0 = const()[name = string("op_47639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47639_cast_fp16 = slice_by_index(begin = var_47639_begin_0, end = var_47639_end_0, end_mask = var_47639_end_mask_0, x = query_cast_fp16)[name = string("op_47639_cast_fp16")]; + tensor var_47643_begin_0 = const()[name = string("op_47643_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47643_end_0 = const()[name = string("op_47643_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_47643_end_mask_0 = const()[name = string("op_47643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47643_cast_fp16 = slice_by_index(begin = var_47643_begin_0, end = var_47643_end_0, end_mask = var_47643_end_mask_0, x = query_cast_fp16)[name = string("op_47643_cast_fp16")]; + tensor var_47647_begin_0 = const()[name = string("op_47647_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47647_end_0 = const()[name = string("op_47647_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_47647_end_mask_0 = const()[name = string("op_47647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47647_cast_fp16 = slice_by_index(begin = var_47647_begin_0, end = var_47647_end_0, end_mask = var_47647_end_mask_0, x = query_cast_fp16)[name = string("op_47647_cast_fp16")]; + tensor var_47651_begin_0 = const()[name = string("op_47651_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47651_end_0 = const()[name = string("op_47651_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_47651_end_mask_0 = const()[name = string("op_47651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47651_cast_fp16 = slice_by_index(begin = var_47651_begin_0, end = var_47651_end_0, end_mask = var_47651_end_mask_0, x = query_cast_fp16)[name = string("op_47651_cast_fp16")]; + tensor var_47655_begin_0 = const()[name = string("op_47655_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47655_end_0 = const()[name = string("op_47655_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_47655_end_mask_0 = const()[name = string("op_47655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47655_cast_fp16 = slice_by_index(begin = var_47655_begin_0, end = var_47655_end_0, end_mask = var_47655_end_mask_0, x = query_cast_fp16)[name = string("op_47655_cast_fp16")]; + tensor var_47659_begin_0 = const()[name = string("op_47659_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47659_end_0 = const()[name = string("op_47659_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_47659_end_mask_0 = const()[name = string("op_47659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47659_cast_fp16 = slice_by_index(begin = var_47659_begin_0, end = var_47659_end_0, end_mask = var_47659_end_mask_0, x = query_cast_fp16)[name = string("op_47659_cast_fp16")]; + tensor var_47663_begin_0 = const()[name = string("op_47663_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47663_end_0 = const()[name = string("op_47663_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_47663_end_mask_0 = const()[name = string("op_47663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47663_cast_fp16 = slice_by_index(begin = var_47663_begin_0, end = var_47663_end_0, end_mask = var_47663_end_mask_0, x = query_cast_fp16)[name = string("op_47663_cast_fp16")]; + tensor var_47667_begin_0 = const()[name = string("op_47667_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47667_end_0 = const()[name = string("op_47667_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_47667_end_mask_0 = const()[name = string("op_47667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47667_cast_fp16 = slice_by_index(begin = var_47667_begin_0, end = var_47667_end_0, end_mask = var_47667_end_mask_0, x = query_cast_fp16)[name = string("op_47667_cast_fp16")]; + tensor var_47671_begin_0 = const()[name = string("op_47671_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47671_end_0 = const()[name = string("op_47671_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_47671_end_mask_0 = const()[name = string("op_47671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47671_cast_fp16 = slice_by_index(begin = var_47671_begin_0, end = var_47671_end_0, end_mask = var_47671_end_mask_0, x = query_cast_fp16)[name = string("op_47671_cast_fp16")]; + tensor var_47680_begin_0 = const()[name = string("op_47680_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47680_end_0 = const()[name = string("op_47680_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47680_end_mask_0 = const()[name = string("op_47680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47680_cast_fp16 = slice_by_index(begin = var_47680_begin_0, end = var_47680_end_0, end_mask = var_47680_end_mask_0, x = var_47595_cast_fp16)[name = string("op_47680_cast_fp16")]; + tensor var_47687_begin_0 = const()[name = string("op_47687_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47687_end_0 = const()[name = string("op_47687_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47687_end_mask_0 = const()[name = string("op_47687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47687_cast_fp16 = slice_by_index(begin = var_47687_begin_0, end = var_47687_end_0, end_mask = var_47687_end_mask_0, x = var_47595_cast_fp16)[name = string("op_47687_cast_fp16")]; + tensor var_47694_begin_0 = const()[name = string("op_47694_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47694_end_0 = const()[name = string("op_47694_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47694_end_mask_0 = const()[name = string("op_47694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47694_cast_fp16 = slice_by_index(begin = var_47694_begin_0, end = var_47694_end_0, end_mask = var_47694_end_mask_0, x = var_47595_cast_fp16)[name = string("op_47694_cast_fp16")]; + tensor var_47701_begin_0 = const()[name = string("op_47701_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47701_end_0 = const()[name = string("op_47701_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47701_end_mask_0 = const()[name = string("op_47701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47701_cast_fp16 = slice_by_index(begin = var_47701_begin_0, end = var_47701_end_0, end_mask = var_47701_end_mask_0, x = var_47595_cast_fp16)[name = string("op_47701_cast_fp16")]; + tensor var_47708_begin_0 = const()[name = string("op_47708_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47708_end_0 = const()[name = string("op_47708_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47708_end_mask_0 = const()[name = string("op_47708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47708_cast_fp16 = slice_by_index(begin = var_47708_begin_0, end = var_47708_end_0, end_mask = var_47708_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47708_cast_fp16")]; + tensor var_47715_begin_0 = const()[name = string("op_47715_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47715_end_0 = const()[name = string("op_47715_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47715_end_mask_0 = const()[name = string("op_47715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47715_cast_fp16 = slice_by_index(begin = var_47715_begin_0, end = var_47715_end_0, end_mask = var_47715_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47715_cast_fp16")]; + tensor var_47722_begin_0 = const()[name = string("op_47722_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47722_end_0 = const()[name = string("op_47722_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47722_end_mask_0 = const()[name = string("op_47722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47722_cast_fp16 = slice_by_index(begin = var_47722_begin_0, end = var_47722_end_0, end_mask = var_47722_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47722_cast_fp16")]; + tensor var_47729_begin_0 = const()[name = string("op_47729_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47729_end_0 = const()[name = string("op_47729_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47729_end_mask_0 = const()[name = string("op_47729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47729_cast_fp16 = slice_by_index(begin = var_47729_begin_0, end = var_47729_end_0, end_mask = var_47729_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47729_cast_fp16")]; + tensor var_47736_begin_0 = const()[name = string("op_47736_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47736_end_0 = const()[name = string("op_47736_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47736_end_mask_0 = const()[name = string("op_47736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47736_cast_fp16 = slice_by_index(begin = var_47736_begin_0, end = var_47736_end_0, end_mask = var_47736_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47736_cast_fp16")]; + tensor var_47743_begin_0 = const()[name = string("op_47743_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47743_end_0 = const()[name = string("op_47743_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47743_end_mask_0 = const()[name = string("op_47743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47743_cast_fp16 = slice_by_index(begin = var_47743_begin_0, end = var_47743_end_0, end_mask = var_47743_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47743_cast_fp16")]; + tensor var_47750_begin_0 = const()[name = string("op_47750_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47750_end_0 = const()[name = string("op_47750_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47750_end_mask_0 = const()[name = string("op_47750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47750_cast_fp16 = slice_by_index(begin = var_47750_begin_0, end = var_47750_end_0, end_mask = var_47750_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47750_cast_fp16")]; + tensor var_47757_begin_0 = const()[name = string("op_47757_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47757_end_0 = const()[name = string("op_47757_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47757_end_mask_0 = const()[name = string("op_47757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47757_cast_fp16 = slice_by_index(begin = var_47757_begin_0, end = var_47757_end_0, end_mask = var_47757_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47757_cast_fp16")]; + tensor var_47764_begin_0 = const()[name = string("op_47764_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47764_end_0 = const()[name = string("op_47764_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47764_end_mask_0 = const()[name = string("op_47764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47764_cast_fp16 = slice_by_index(begin = var_47764_begin_0, end = var_47764_end_0, end_mask = var_47764_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47764_cast_fp16")]; + tensor var_47771_begin_0 = const()[name = string("op_47771_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47771_end_0 = const()[name = string("op_47771_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47771_end_mask_0 = const()[name = string("op_47771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47771_cast_fp16 = slice_by_index(begin = var_47771_begin_0, end = var_47771_end_0, end_mask = var_47771_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47771_cast_fp16")]; + tensor var_47778_begin_0 = const()[name = string("op_47778_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47778_end_0 = const()[name = string("op_47778_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47778_end_mask_0 = const()[name = string("op_47778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47778_cast_fp16 = slice_by_index(begin = var_47778_begin_0, end = var_47778_end_0, end_mask = var_47778_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47778_cast_fp16")]; + tensor var_47785_begin_0 = const()[name = string("op_47785_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47785_end_0 = const()[name = string("op_47785_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47785_end_mask_0 = const()[name = string("op_47785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47785_cast_fp16 = slice_by_index(begin = var_47785_begin_0, end = var_47785_end_0, end_mask = var_47785_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47785_cast_fp16")]; + tensor var_47792_begin_0 = const()[name = string("op_47792_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47792_end_0 = const()[name = string("op_47792_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47792_end_mask_0 = const()[name = string("op_47792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47792_cast_fp16 = slice_by_index(begin = var_47792_begin_0, end = var_47792_end_0, end_mask = var_47792_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47792_cast_fp16")]; + tensor var_47799_begin_0 = const()[name = string("op_47799_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47799_end_0 = const()[name = string("op_47799_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47799_end_mask_0 = const()[name = string("op_47799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47799_cast_fp16 = slice_by_index(begin = var_47799_begin_0, end = var_47799_end_0, end_mask = var_47799_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47799_cast_fp16")]; + tensor var_47806_begin_0 = const()[name = string("op_47806_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47806_end_0 = const()[name = string("op_47806_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47806_end_mask_0 = const()[name = string("op_47806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47806_cast_fp16 = slice_by_index(begin = var_47806_begin_0, end = var_47806_end_0, end_mask = var_47806_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47806_cast_fp16")]; + tensor var_47813_begin_0 = const()[name = string("op_47813_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47813_end_0 = const()[name = string("op_47813_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47813_end_mask_0 = const()[name = string("op_47813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47813_cast_fp16 = slice_by_index(begin = var_47813_begin_0, end = var_47813_end_0, end_mask = var_47813_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47813_cast_fp16")]; + tensor var_47820_begin_0 = const()[name = string("op_47820_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47820_end_0 = const()[name = string("op_47820_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47820_end_mask_0 = const()[name = string("op_47820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47820_cast_fp16 = slice_by_index(begin = var_47820_begin_0, end = var_47820_end_0, end_mask = var_47820_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47820_cast_fp16")]; + tensor var_47827_begin_0 = const()[name = string("op_47827_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47827_end_0 = const()[name = string("op_47827_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47827_end_mask_0 = const()[name = string("op_47827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47827_cast_fp16 = slice_by_index(begin = var_47827_begin_0, end = var_47827_end_0, end_mask = var_47827_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47827_cast_fp16")]; + tensor var_47834_begin_0 = const()[name = string("op_47834_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47834_end_0 = const()[name = string("op_47834_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47834_end_mask_0 = const()[name = string("op_47834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47834_cast_fp16 = slice_by_index(begin = var_47834_begin_0, end = var_47834_end_0, end_mask = var_47834_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47834_cast_fp16")]; + tensor var_47841_begin_0 = const()[name = string("op_47841_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47841_end_0 = const()[name = string("op_47841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47841_end_mask_0 = const()[name = string("op_47841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47841_cast_fp16 = slice_by_index(begin = var_47841_begin_0, end = var_47841_end_0, end_mask = var_47841_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47841_cast_fp16")]; + tensor var_47848_begin_0 = const()[name = string("op_47848_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47848_end_0 = const()[name = string("op_47848_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47848_end_mask_0 = const()[name = string("op_47848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47848_cast_fp16 = slice_by_index(begin = var_47848_begin_0, end = var_47848_end_0, end_mask = var_47848_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47848_cast_fp16")]; + tensor var_47855_begin_0 = const()[name = string("op_47855_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47855_end_0 = const()[name = string("op_47855_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47855_end_mask_0 = const()[name = string("op_47855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47855_cast_fp16 = slice_by_index(begin = var_47855_begin_0, end = var_47855_end_0, end_mask = var_47855_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47855_cast_fp16")]; + tensor var_47862_begin_0 = const()[name = string("op_47862_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47862_end_0 = const()[name = string("op_47862_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47862_end_mask_0 = const()[name = string("op_47862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47862_cast_fp16 = slice_by_index(begin = var_47862_begin_0, end = var_47862_end_0, end_mask = var_47862_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47862_cast_fp16")]; + tensor var_47869_begin_0 = const()[name = string("op_47869_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47869_end_0 = const()[name = string("op_47869_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47869_end_mask_0 = const()[name = string("op_47869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47869_cast_fp16 = slice_by_index(begin = var_47869_begin_0, end = var_47869_end_0, end_mask = var_47869_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47869_cast_fp16")]; + tensor var_47876_begin_0 = const()[name = string("op_47876_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47876_end_0 = const()[name = string("op_47876_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47876_end_mask_0 = const()[name = string("op_47876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47876_cast_fp16 = slice_by_index(begin = var_47876_begin_0, end = var_47876_end_0, end_mask = var_47876_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47876_cast_fp16")]; + tensor var_47883_begin_0 = const()[name = string("op_47883_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47883_end_0 = const()[name = string("op_47883_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47883_end_mask_0 = const()[name = string("op_47883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47883_cast_fp16 = slice_by_index(begin = var_47883_begin_0, end = var_47883_end_0, end_mask = var_47883_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47883_cast_fp16")]; + tensor var_47890_begin_0 = const()[name = string("op_47890_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47890_end_0 = const()[name = string("op_47890_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47890_end_mask_0 = const()[name = string("op_47890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47890_cast_fp16 = slice_by_index(begin = var_47890_begin_0, end = var_47890_end_0, end_mask = var_47890_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47890_cast_fp16")]; + tensor var_47897_begin_0 = const()[name = string("op_47897_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47897_end_0 = const()[name = string("op_47897_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47897_end_mask_0 = const()[name = string("op_47897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47897_cast_fp16 = slice_by_index(begin = var_47897_begin_0, end = var_47897_end_0, end_mask = var_47897_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47897_cast_fp16")]; + tensor var_47904_begin_0 = const()[name = string("op_47904_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47904_end_0 = const()[name = string("op_47904_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47904_end_mask_0 = const()[name = string("op_47904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47904_cast_fp16 = slice_by_index(begin = var_47904_begin_0, end = var_47904_end_0, end_mask = var_47904_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47904_cast_fp16")]; + tensor var_47911_begin_0 = const()[name = string("op_47911_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47911_end_0 = const()[name = string("op_47911_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47911_end_mask_0 = const()[name = string("op_47911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47911_cast_fp16 = slice_by_index(begin = var_47911_begin_0, end = var_47911_end_0, end_mask = var_47911_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47911_cast_fp16")]; + tensor var_47918_begin_0 = const()[name = string("op_47918_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47918_end_0 = const()[name = string("op_47918_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47918_end_mask_0 = const()[name = string("op_47918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47918_cast_fp16 = slice_by_index(begin = var_47918_begin_0, end = var_47918_end_0, end_mask = var_47918_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47918_cast_fp16")]; + tensor var_47925_begin_0 = const()[name = string("op_47925_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47925_end_0 = const()[name = string("op_47925_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47925_end_mask_0 = const()[name = string("op_47925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47925_cast_fp16 = slice_by_index(begin = var_47925_begin_0, end = var_47925_end_0, end_mask = var_47925_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47925_cast_fp16")]; + tensor var_47932_begin_0 = const()[name = string("op_47932_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47932_end_0 = const()[name = string("op_47932_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47932_end_mask_0 = const()[name = string("op_47932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47932_cast_fp16 = slice_by_index(begin = var_47932_begin_0, end = var_47932_end_0, end_mask = var_47932_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47932_cast_fp16")]; + tensor var_47939_begin_0 = const()[name = string("op_47939_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47939_end_0 = const()[name = string("op_47939_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47939_end_mask_0 = const()[name = string("op_47939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47939_cast_fp16 = slice_by_index(begin = var_47939_begin_0, end = var_47939_end_0, end_mask = var_47939_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47939_cast_fp16")]; + tensor var_47946_begin_0 = const()[name = string("op_47946_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47946_end_0 = const()[name = string("op_47946_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47946_end_mask_0 = const()[name = string("op_47946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47946_cast_fp16 = slice_by_index(begin = var_47946_begin_0, end = var_47946_end_0, end_mask = var_47946_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47946_cast_fp16")]; + tensor var_47953_begin_0 = const()[name = string("op_47953_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47953_end_0 = const()[name = string("op_47953_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47953_end_mask_0 = const()[name = string("op_47953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47953_cast_fp16 = slice_by_index(begin = var_47953_begin_0, end = var_47953_end_0, end_mask = var_47953_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47953_cast_fp16")]; + tensor var_47960_begin_0 = const()[name = string("op_47960_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47960_end_0 = const()[name = string("op_47960_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47960_end_mask_0 = const()[name = string("op_47960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47960_cast_fp16 = slice_by_index(begin = var_47960_begin_0, end = var_47960_end_0, end_mask = var_47960_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47960_cast_fp16")]; + tensor var_47967_begin_0 = const()[name = string("op_47967_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47967_end_0 = const()[name = string("op_47967_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47967_end_mask_0 = const()[name = string("op_47967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47967_cast_fp16 = slice_by_index(begin = var_47967_begin_0, end = var_47967_end_0, end_mask = var_47967_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47967_cast_fp16")]; + tensor var_47974_begin_0 = const()[name = string("op_47974_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47974_end_0 = const()[name = string("op_47974_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47974_end_mask_0 = const()[name = string("op_47974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47974_cast_fp16 = slice_by_index(begin = var_47974_begin_0, end = var_47974_end_0, end_mask = var_47974_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47974_cast_fp16")]; + tensor var_47981_begin_0 = const()[name = string("op_47981_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47981_end_0 = const()[name = string("op_47981_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47981_end_mask_0 = const()[name = string("op_47981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47981_cast_fp16 = slice_by_index(begin = var_47981_begin_0, end = var_47981_end_0, end_mask = var_47981_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47981_cast_fp16")]; + tensor var_47988_begin_0 = const()[name = string("op_47988_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47988_end_0 = const()[name = string("op_47988_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47988_end_mask_0 = const()[name = string("op_47988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47988_cast_fp16 = slice_by_index(begin = var_47988_begin_0, end = var_47988_end_0, end_mask = var_47988_end_mask_0, x = var_47639_cast_fp16)[name = string("op_47988_cast_fp16")]; + tensor var_47995_begin_0 = const()[name = string("op_47995_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47995_end_0 = const()[name = string("op_47995_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47995_end_mask_0 = const()[name = string("op_47995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47995_cast_fp16 = slice_by_index(begin = var_47995_begin_0, end = var_47995_end_0, end_mask = var_47995_end_mask_0, x = var_47639_cast_fp16)[name = string("op_47995_cast_fp16")]; + tensor var_48002_begin_0 = const()[name = string("op_48002_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48002_end_0 = const()[name = string("op_48002_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48002_end_mask_0 = const()[name = string("op_48002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48002_cast_fp16 = slice_by_index(begin = var_48002_begin_0, end = var_48002_end_0, end_mask = var_48002_end_mask_0, x = var_47639_cast_fp16)[name = string("op_48002_cast_fp16")]; + tensor var_48009_begin_0 = const()[name = string("op_48009_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48009_end_0 = const()[name = string("op_48009_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48009_end_mask_0 = const()[name = string("op_48009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48009_cast_fp16 = slice_by_index(begin = var_48009_begin_0, end = var_48009_end_0, end_mask = var_48009_end_mask_0, x = var_47639_cast_fp16)[name = string("op_48009_cast_fp16")]; + tensor var_48016_begin_0 = const()[name = string("op_48016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48016_end_0 = const()[name = string("op_48016_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48016_end_mask_0 = const()[name = string("op_48016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48016_cast_fp16 = slice_by_index(begin = var_48016_begin_0, end = var_48016_end_0, end_mask = var_48016_end_mask_0, x = var_47643_cast_fp16)[name = string("op_48016_cast_fp16")]; + tensor var_48023_begin_0 = const()[name = string("op_48023_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48023_end_0 = const()[name = string("op_48023_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48023_end_mask_0 = const()[name = string("op_48023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48023_cast_fp16 = slice_by_index(begin = var_48023_begin_0, end = var_48023_end_0, end_mask = var_48023_end_mask_0, x = var_47643_cast_fp16)[name = string("op_48023_cast_fp16")]; + tensor var_48030_begin_0 = const()[name = string("op_48030_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48030_end_0 = const()[name = string("op_48030_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48030_end_mask_0 = const()[name = string("op_48030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48030_cast_fp16 = slice_by_index(begin = var_48030_begin_0, end = var_48030_end_0, end_mask = var_48030_end_mask_0, x = var_47643_cast_fp16)[name = string("op_48030_cast_fp16")]; + tensor var_48037_begin_0 = const()[name = string("op_48037_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48037_end_0 = const()[name = string("op_48037_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48037_end_mask_0 = const()[name = string("op_48037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48037_cast_fp16 = slice_by_index(begin = var_48037_begin_0, end = var_48037_end_0, end_mask = var_48037_end_mask_0, x = var_47643_cast_fp16)[name = string("op_48037_cast_fp16")]; + tensor var_48044_begin_0 = const()[name = string("op_48044_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48044_end_0 = const()[name = string("op_48044_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48044_end_mask_0 = const()[name = string("op_48044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48044_cast_fp16 = slice_by_index(begin = var_48044_begin_0, end = var_48044_end_0, end_mask = var_48044_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48044_cast_fp16")]; + tensor var_48051_begin_0 = const()[name = string("op_48051_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48051_end_0 = const()[name = string("op_48051_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48051_end_mask_0 = const()[name = string("op_48051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48051_cast_fp16 = slice_by_index(begin = var_48051_begin_0, end = var_48051_end_0, end_mask = var_48051_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48051_cast_fp16")]; + tensor var_48058_begin_0 = const()[name = string("op_48058_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48058_end_0 = const()[name = string("op_48058_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48058_end_mask_0 = const()[name = string("op_48058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48058_cast_fp16 = slice_by_index(begin = var_48058_begin_0, end = var_48058_end_0, end_mask = var_48058_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48058_cast_fp16")]; + tensor var_48065_begin_0 = const()[name = string("op_48065_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48065_end_0 = const()[name = string("op_48065_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48065_end_mask_0 = const()[name = string("op_48065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48065_cast_fp16 = slice_by_index(begin = var_48065_begin_0, end = var_48065_end_0, end_mask = var_48065_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48065_cast_fp16")]; + tensor var_48072_begin_0 = const()[name = string("op_48072_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48072_end_0 = const()[name = string("op_48072_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48072_end_mask_0 = const()[name = string("op_48072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48072_cast_fp16 = slice_by_index(begin = var_48072_begin_0, end = var_48072_end_0, end_mask = var_48072_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48072_cast_fp16")]; + tensor var_48079_begin_0 = const()[name = string("op_48079_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48079_end_0 = const()[name = string("op_48079_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48079_end_mask_0 = const()[name = string("op_48079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48079_cast_fp16 = slice_by_index(begin = var_48079_begin_0, end = var_48079_end_0, end_mask = var_48079_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48079_cast_fp16")]; + tensor var_48086_begin_0 = const()[name = string("op_48086_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48086_end_0 = const()[name = string("op_48086_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48086_end_mask_0 = const()[name = string("op_48086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48086_cast_fp16 = slice_by_index(begin = var_48086_begin_0, end = var_48086_end_0, end_mask = var_48086_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48086_cast_fp16")]; + tensor var_48093_begin_0 = const()[name = string("op_48093_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48093_end_0 = const()[name = string("op_48093_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48093_end_mask_0 = const()[name = string("op_48093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48093_cast_fp16 = slice_by_index(begin = var_48093_begin_0, end = var_48093_end_0, end_mask = var_48093_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48093_cast_fp16")]; + tensor var_48100_begin_0 = const()[name = string("op_48100_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48100_end_0 = const()[name = string("op_48100_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48100_end_mask_0 = const()[name = string("op_48100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48100_cast_fp16 = slice_by_index(begin = var_48100_begin_0, end = var_48100_end_0, end_mask = var_48100_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48100_cast_fp16")]; + tensor var_48107_begin_0 = const()[name = string("op_48107_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48107_end_0 = const()[name = string("op_48107_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48107_end_mask_0 = const()[name = string("op_48107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48107_cast_fp16 = slice_by_index(begin = var_48107_begin_0, end = var_48107_end_0, end_mask = var_48107_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48107_cast_fp16")]; + tensor var_48114_begin_0 = const()[name = string("op_48114_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48114_end_0 = const()[name = string("op_48114_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48114_end_mask_0 = const()[name = string("op_48114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48114_cast_fp16 = slice_by_index(begin = var_48114_begin_0, end = var_48114_end_0, end_mask = var_48114_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48114_cast_fp16")]; + tensor var_48121_begin_0 = const()[name = string("op_48121_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48121_end_0 = const()[name = string("op_48121_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48121_end_mask_0 = const()[name = string("op_48121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48121_cast_fp16 = slice_by_index(begin = var_48121_begin_0, end = var_48121_end_0, end_mask = var_48121_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48121_cast_fp16")]; + tensor var_48128_begin_0 = const()[name = string("op_48128_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48128_end_0 = const()[name = string("op_48128_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48128_end_mask_0 = const()[name = string("op_48128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48128_cast_fp16 = slice_by_index(begin = var_48128_begin_0, end = var_48128_end_0, end_mask = var_48128_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48128_cast_fp16")]; + tensor var_48135_begin_0 = const()[name = string("op_48135_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48135_end_0 = const()[name = string("op_48135_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48135_end_mask_0 = const()[name = string("op_48135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48135_cast_fp16 = slice_by_index(begin = var_48135_begin_0, end = var_48135_end_0, end_mask = var_48135_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48135_cast_fp16")]; + tensor var_48142_begin_0 = const()[name = string("op_48142_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48142_end_0 = const()[name = string("op_48142_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48142_end_mask_0 = const()[name = string("op_48142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48142_cast_fp16 = slice_by_index(begin = var_48142_begin_0, end = var_48142_end_0, end_mask = var_48142_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48142_cast_fp16")]; + tensor var_48149_begin_0 = const()[name = string("op_48149_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48149_end_0 = const()[name = string("op_48149_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48149_end_mask_0 = const()[name = string("op_48149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48149_cast_fp16 = slice_by_index(begin = var_48149_begin_0, end = var_48149_end_0, end_mask = var_48149_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48149_cast_fp16")]; + tensor var_48156_begin_0 = const()[name = string("op_48156_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48156_end_0 = const()[name = string("op_48156_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48156_end_mask_0 = const()[name = string("op_48156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48156_cast_fp16 = slice_by_index(begin = var_48156_begin_0, end = var_48156_end_0, end_mask = var_48156_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48156_cast_fp16")]; + tensor var_48163_begin_0 = const()[name = string("op_48163_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48163_end_0 = const()[name = string("op_48163_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48163_end_mask_0 = const()[name = string("op_48163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48163_cast_fp16 = slice_by_index(begin = var_48163_begin_0, end = var_48163_end_0, end_mask = var_48163_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48163_cast_fp16")]; + tensor var_48170_begin_0 = const()[name = string("op_48170_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48170_end_0 = const()[name = string("op_48170_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48170_end_mask_0 = const()[name = string("op_48170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48170_cast_fp16 = slice_by_index(begin = var_48170_begin_0, end = var_48170_end_0, end_mask = var_48170_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48170_cast_fp16")]; + tensor var_48177_begin_0 = const()[name = string("op_48177_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48177_end_0 = const()[name = string("op_48177_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48177_end_mask_0 = const()[name = string("op_48177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48177_cast_fp16 = slice_by_index(begin = var_48177_begin_0, end = var_48177_end_0, end_mask = var_48177_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48177_cast_fp16")]; + tensor var_48184_begin_0 = const()[name = string("op_48184_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48184_end_0 = const()[name = string("op_48184_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48184_end_mask_0 = const()[name = string("op_48184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48184_cast_fp16 = slice_by_index(begin = var_48184_begin_0, end = var_48184_end_0, end_mask = var_48184_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48184_cast_fp16")]; + tensor var_48191_begin_0 = const()[name = string("op_48191_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48191_end_0 = const()[name = string("op_48191_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48191_end_mask_0 = const()[name = string("op_48191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48191_cast_fp16 = slice_by_index(begin = var_48191_begin_0, end = var_48191_end_0, end_mask = var_48191_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48191_cast_fp16")]; + tensor var_48198_begin_0 = const()[name = string("op_48198_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48198_end_0 = const()[name = string("op_48198_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48198_end_mask_0 = const()[name = string("op_48198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48198_cast_fp16 = slice_by_index(begin = var_48198_begin_0, end = var_48198_end_0, end_mask = var_48198_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48198_cast_fp16")]; + tensor var_48205_begin_0 = const()[name = string("op_48205_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48205_end_0 = const()[name = string("op_48205_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48205_end_mask_0 = const()[name = string("op_48205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48205_cast_fp16 = slice_by_index(begin = var_48205_begin_0, end = var_48205_end_0, end_mask = var_48205_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48205_cast_fp16")]; + tensor var_48212_begin_0 = const()[name = string("op_48212_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48212_end_0 = const()[name = string("op_48212_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48212_end_mask_0 = const()[name = string("op_48212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48212_cast_fp16 = slice_by_index(begin = var_48212_begin_0, end = var_48212_end_0, end_mask = var_48212_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48212_cast_fp16")]; + tensor var_48219_begin_0 = const()[name = string("op_48219_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48219_end_0 = const()[name = string("op_48219_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48219_end_mask_0 = const()[name = string("op_48219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48219_cast_fp16 = slice_by_index(begin = var_48219_begin_0, end = var_48219_end_0, end_mask = var_48219_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48219_cast_fp16")]; + tensor var_48226_begin_0 = const()[name = string("op_48226_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48226_end_0 = const()[name = string("op_48226_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48226_end_mask_0 = const()[name = string("op_48226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48226_cast_fp16 = slice_by_index(begin = var_48226_begin_0, end = var_48226_end_0, end_mask = var_48226_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48226_cast_fp16")]; + tensor var_48233_begin_0 = const()[name = string("op_48233_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48233_end_0 = const()[name = string("op_48233_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48233_end_mask_0 = const()[name = string("op_48233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48233_cast_fp16 = slice_by_index(begin = var_48233_begin_0, end = var_48233_end_0, end_mask = var_48233_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48233_cast_fp16")]; + tensor k_63_perm_0 = const()[name = string("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_48238_begin_0 = const()[name = string("op_48238_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48238_end_0 = const()[name = string("op_48238_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_48238_end_mask_0 = const()[name = string("op_48238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_48238_cast_fp16 = slice_by_index(begin = var_48238_begin_0, end = var_48238_end_0, end_mask = var_48238_end_mask_0, x = k_63_cast_fp16)[name = string("op_48238_cast_fp16")]; + tensor var_48242_begin_0 = const()[name = string("op_48242_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_48242_end_0 = const()[name = string("op_48242_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_48242_end_mask_0 = const()[name = string("op_48242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48242_cast_fp16 = slice_by_index(begin = var_48242_begin_0, end = var_48242_end_0, end_mask = var_48242_end_mask_0, x = k_63_cast_fp16)[name = string("op_48242_cast_fp16")]; + tensor var_48246_begin_0 = const()[name = string("op_48246_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_48246_end_0 = const()[name = string("op_48246_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_48246_end_mask_0 = const()[name = string("op_48246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48246_cast_fp16 = slice_by_index(begin = var_48246_begin_0, end = var_48246_end_0, end_mask = var_48246_end_mask_0, x = k_63_cast_fp16)[name = string("op_48246_cast_fp16")]; + tensor var_48250_begin_0 = const()[name = string("op_48250_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_48250_end_0 = const()[name = string("op_48250_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_48250_end_mask_0 = const()[name = string("op_48250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48250_cast_fp16 = slice_by_index(begin = var_48250_begin_0, end = var_48250_end_0, end_mask = var_48250_end_mask_0, x = k_63_cast_fp16)[name = string("op_48250_cast_fp16")]; + tensor var_48254_begin_0 = const()[name = string("op_48254_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_48254_end_0 = const()[name = string("op_48254_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_48254_end_mask_0 = const()[name = string("op_48254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48254_cast_fp16 = slice_by_index(begin = var_48254_begin_0, end = var_48254_end_0, end_mask = var_48254_end_mask_0, x = k_63_cast_fp16)[name = string("op_48254_cast_fp16")]; + tensor var_48258_begin_0 = const()[name = string("op_48258_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_48258_end_0 = const()[name = string("op_48258_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_48258_end_mask_0 = const()[name = string("op_48258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48258_cast_fp16 = slice_by_index(begin = var_48258_begin_0, end = var_48258_end_0, end_mask = var_48258_end_mask_0, x = k_63_cast_fp16)[name = string("op_48258_cast_fp16")]; + tensor var_48262_begin_0 = const()[name = string("op_48262_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_48262_end_0 = const()[name = string("op_48262_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_48262_end_mask_0 = const()[name = string("op_48262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48262_cast_fp16 = slice_by_index(begin = var_48262_begin_0, end = var_48262_end_0, end_mask = var_48262_end_mask_0, x = k_63_cast_fp16)[name = string("op_48262_cast_fp16")]; + tensor var_48266_begin_0 = const()[name = string("op_48266_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_48266_end_0 = const()[name = string("op_48266_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_48266_end_mask_0 = const()[name = string("op_48266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48266_cast_fp16 = slice_by_index(begin = var_48266_begin_0, end = var_48266_end_0, end_mask = var_48266_end_mask_0, x = k_63_cast_fp16)[name = string("op_48266_cast_fp16")]; + tensor var_48270_begin_0 = const()[name = string("op_48270_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_48270_end_0 = const()[name = string("op_48270_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_48270_end_mask_0 = const()[name = string("op_48270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48270_cast_fp16 = slice_by_index(begin = var_48270_begin_0, end = var_48270_end_0, end_mask = var_48270_end_mask_0, x = k_63_cast_fp16)[name = string("op_48270_cast_fp16")]; + tensor var_48274_begin_0 = const()[name = string("op_48274_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_48274_end_0 = const()[name = string("op_48274_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_48274_end_mask_0 = const()[name = string("op_48274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48274_cast_fp16 = slice_by_index(begin = var_48274_begin_0, end = var_48274_end_0, end_mask = var_48274_end_mask_0, x = k_63_cast_fp16)[name = string("op_48274_cast_fp16")]; + tensor var_48278_begin_0 = const()[name = string("op_48278_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_48278_end_0 = const()[name = string("op_48278_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_48278_end_mask_0 = const()[name = string("op_48278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48278_cast_fp16 = slice_by_index(begin = var_48278_begin_0, end = var_48278_end_0, end_mask = var_48278_end_mask_0, x = k_63_cast_fp16)[name = string("op_48278_cast_fp16")]; + tensor var_48282_begin_0 = const()[name = string("op_48282_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_48282_end_0 = const()[name = string("op_48282_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_48282_end_mask_0 = const()[name = string("op_48282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48282_cast_fp16 = slice_by_index(begin = var_48282_begin_0, end = var_48282_end_0, end_mask = var_48282_end_mask_0, x = k_63_cast_fp16)[name = string("op_48282_cast_fp16")]; + tensor var_48286_begin_0 = const()[name = string("op_48286_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_48286_end_0 = const()[name = string("op_48286_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_48286_end_mask_0 = const()[name = string("op_48286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48286_cast_fp16 = slice_by_index(begin = var_48286_begin_0, end = var_48286_end_0, end_mask = var_48286_end_mask_0, x = k_63_cast_fp16)[name = string("op_48286_cast_fp16")]; + tensor var_48290_begin_0 = const()[name = string("op_48290_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_48290_end_0 = const()[name = string("op_48290_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_48290_end_mask_0 = const()[name = string("op_48290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48290_cast_fp16 = slice_by_index(begin = var_48290_begin_0, end = var_48290_end_0, end_mask = var_48290_end_mask_0, x = k_63_cast_fp16)[name = string("op_48290_cast_fp16")]; + tensor var_48294_begin_0 = const()[name = string("op_48294_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_48294_end_0 = const()[name = string("op_48294_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_48294_end_mask_0 = const()[name = string("op_48294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48294_cast_fp16 = slice_by_index(begin = var_48294_begin_0, end = var_48294_end_0, end_mask = var_48294_end_mask_0, x = k_63_cast_fp16)[name = string("op_48294_cast_fp16")]; + tensor var_48298_begin_0 = const()[name = string("op_48298_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_48298_end_0 = const()[name = string("op_48298_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_48298_end_mask_0 = const()[name = string("op_48298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48298_cast_fp16 = slice_by_index(begin = var_48298_begin_0, end = var_48298_end_0, end_mask = var_48298_end_mask_0, x = k_63_cast_fp16)[name = string("op_48298_cast_fp16")]; + tensor var_48302_begin_0 = const()[name = string("op_48302_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_48302_end_0 = const()[name = string("op_48302_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_48302_end_mask_0 = const()[name = string("op_48302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48302_cast_fp16 = slice_by_index(begin = var_48302_begin_0, end = var_48302_end_0, end_mask = var_48302_end_mask_0, x = k_63_cast_fp16)[name = string("op_48302_cast_fp16")]; + tensor var_48306_begin_0 = const()[name = string("op_48306_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_48306_end_0 = const()[name = string("op_48306_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_48306_end_mask_0 = const()[name = string("op_48306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48306_cast_fp16 = slice_by_index(begin = var_48306_begin_0, end = var_48306_end_0, end_mask = var_48306_end_mask_0, x = k_63_cast_fp16)[name = string("op_48306_cast_fp16")]; + tensor var_48310_begin_0 = const()[name = string("op_48310_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_48310_end_0 = const()[name = string("op_48310_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_48310_end_mask_0 = const()[name = string("op_48310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48310_cast_fp16 = slice_by_index(begin = var_48310_begin_0, end = var_48310_end_0, end_mask = var_48310_end_mask_0, x = k_63_cast_fp16)[name = string("op_48310_cast_fp16")]; + tensor var_48314_begin_0 = const()[name = string("op_48314_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_48314_end_0 = const()[name = string("op_48314_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_48314_end_mask_0 = const()[name = string("op_48314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48314_cast_fp16 = slice_by_index(begin = var_48314_begin_0, end = var_48314_end_0, end_mask = var_48314_end_mask_0, x = k_63_cast_fp16)[name = string("op_48314_cast_fp16")]; + tensor var_48316_begin_0 = const()[name = string("op_48316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48316_end_0 = const()[name = string("op_48316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48316_end_mask_0 = const()[name = string("op_48316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48316_cast_fp16 = slice_by_index(begin = var_48316_begin_0, end = var_48316_end_0, end_mask = var_48316_end_mask_0, x = value_cast_fp16)[name = string("op_48316_cast_fp16")]; + tensor var_48320_begin_0 = const()[name = string("op_48320_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48320_end_0 = const()[name = string("op_48320_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_48320_end_mask_0 = const()[name = string("op_48320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48320_cast_fp16 = slice_by_index(begin = var_48320_begin_0, end = var_48320_end_0, end_mask = var_48320_end_mask_0, x = value_cast_fp16)[name = string("op_48320_cast_fp16")]; + tensor var_48324_begin_0 = const()[name = string("op_48324_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48324_end_0 = const()[name = string("op_48324_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_48324_end_mask_0 = const()[name = string("op_48324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48324_cast_fp16 = slice_by_index(begin = var_48324_begin_0, end = var_48324_end_0, end_mask = var_48324_end_mask_0, x = value_cast_fp16)[name = string("op_48324_cast_fp16")]; + tensor var_48328_begin_0 = const()[name = string("op_48328_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48328_end_0 = const()[name = string("op_48328_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_48328_end_mask_0 = const()[name = string("op_48328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48328_cast_fp16 = slice_by_index(begin = var_48328_begin_0, end = var_48328_end_0, end_mask = var_48328_end_mask_0, x = value_cast_fp16)[name = string("op_48328_cast_fp16")]; + tensor var_48332_begin_0 = const()[name = string("op_48332_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48332_end_0 = const()[name = string("op_48332_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_48332_end_mask_0 = const()[name = string("op_48332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48332_cast_fp16 = slice_by_index(begin = var_48332_begin_0, end = var_48332_end_0, end_mask = var_48332_end_mask_0, x = value_cast_fp16)[name = string("op_48332_cast_fp16")]; + tensor var_48336_begin_0 = const()[name = string("op_48336_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48336_end_0 = const()[name = string("op_48336_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_48336_end_mask_0 = const()[name = string("op_48336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48336_cast_fp16 = slice_by_index(begin = var_48336_begin_0, end = var_48336_end_0, end_mask = var_48336_end_mask_0, x = value_cast_fp16)[name = string("op_48336_cast_fp16")]; + tensor var_48340_begin_0 = const()[name = string("op_48340_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48340_end_0 = const()[name = string("op_48340_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_48340_end_mask_0 = const()[name = string("op_48340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48340_cast_fp16 = slice_by_index(begin = var_48340_begin_0, end = var_48340_end_0, end_mask = var_48340_end_mask_0, x = value_cast_fp16)[name = string("op_48340_cast_fp16")]; + tensor var_48344_begin_0 = const()[name = string("op_48344_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48344_end_0 = const()[name = string("op_48344_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_48344_end_mask_0 = const()[name = string("op_48344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48344_cast_fp16 = slice_by_index(begin = var_48344_begin_0, end = var_48344_end_0, end_mask = var_48344_end_mask_0, x = value_cast_fp16)[name = string("op_48344_cast_fp16")]; + tensor var_48348_begin_0 = const()[name = string("op_48348_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48348_end_0 = const()[name = string("op_48348_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_48348_end_mask_0 = const()[name = string("op_48348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48348_cast_fp16 = slice_by_index(begin = var_48348_begin_0, end = var_48348_end_0, end_mask = var_48348_end_mask_0, x = value_cast_fp16)[name = string("op_48348_cast_fp16")]; + tensor var_48352_begin_0 = const()[name = string("op_48352_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48352_end_0 = const()[name = string("op_48352_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_48352_end_mask_0 = const()[name = string("op_48352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48352_cast_fp16 = slice_by_index(begin = var_48352_begin_0, end = var_48352_end_0, end_mask = var_48352_end_mask_0, x = value_cast_fp16)[name = string("op_48352_cast_fp16")]; + tensor var_48356_begin_0 = const()[name = string("op_48356_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48356_end_0 = const()[name = string("op_48356_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_48356_end_mask_0 = const()[name = string("op_48356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48356_cast_fp16 = slice_by_index(begin = var_48356_begin_0, end = var_48356_end_0, end_mask = var_48356_end_mask_0, x = value_cast_fp16)[name = string("op_48356_cast_fp16")]; + tensor var_48360_begin_0 = const()[name = string("op_48360_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48360_end_0 = const()[name = string("op_48360_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_48360_end_mask_0 = const()[name = string("op_48360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48360_cast_fp16 = slice_by_index(begin = var_48360_begin_0, end = var_48360_end_0, end_mask = var_48360_end_mask_0, x = value_cast_fp16)[name = string("op_48360_cast_fp16")]; + tensor var_48364_begin_0 = const()[name = string("op_48364_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48364_end_0 = const()[name = string("op_48364_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_48364_end_mask_0 = const()[name = string("op_48364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48364_cast_fp16 = slice_by_index(begin = var_48364_begin_0, end = var_48364_end_0, end_mask = var_48364_end_mask_0, x = value_cast_fp16)[name = string("op_48364_cast_fp16")]; + tensor var_48368_begin_0 = const()[name = string("op_48368_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48368_end_0 = const()[name = string("op_48368_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_48368_end_mask_0 = const()[name = string("op_48368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48368_cast_fp16 = slice_by_index(begin = var_48368_begin_0, end = var_48368_end_0, end_mask = var_48368_end_mask_0, x = value_cast_fp16)[name = string("op_48368_cast_fp16")]; + tensor var_48372_begin_0 = const()[name = string("op_48372_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48372_end_0 = const()[name = string("op_48372_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_48372_end_mask_0 = const()[name = string("op_48372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48372_cast_fp16 = slice_by_index(begin = var_48372_begin_0, end = var_48372_end_0, end_mask = var_48372_end_mask_0, x = value_cast_fp16)[name = string("op_48372_cast_fp16")]; + tensor var_48376_begin_0 = const()[name = string("op_48376_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48376_end_0 = const()[name = string("op_48376_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_48376_end_mask_0 = const()[name = string("op_48376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48376_cast_fp16 = slice_by_index(begin = var_48376_begin_0, end = var_48376_end_0, end_mask = var_48376_end_mask_0, x = value_cast_fp16)[name = string("op_48376_cast_fp16")]; + tensor var_48380_begin_0 = const()[name = string("op_48380_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48380_end_0 = const()[name = string("op_48380_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_48380_end_mask_0 = const()[name = string("op_48380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48380_cast_fp16 = slice_by_index(begin = var_48380_begin_0, end = var_48380_end_0, end_mask = var_48380_end_mask_0, x = value_cast_fp16)[name = string("op_48380_cast_fp16")]; + tensor var_48384_begin_0 = const()[name = string("op_48384_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48384_end_0 = const()[name = string("op_48384_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_48384_end_mask_0 = const()[name = string("op_48384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48384_cast_fp16 = slice_by_index(begin = var_48384_begin_0, end = var_48384_end_0, end_mask = var_48384_end_mask_0, x = value_cast_fp16)[name = string("op_48384_cast_fp16")]; + tensor var_48388_begin_0 = const()[name = string("op_48388_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48388_end_0 = const()[name = string("op_48388_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_48388_end_mask_0 = const()[name = string("op_48388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48388_cast_fp16 = slice_by_index(begin = var_48388_begin_0, end = var_48388_end_0, end_mask = var_48388_end_mask_0, x = value_cast_fp16)[name = string("op_48388_cast_fp16")]; + tensor var_48392_begin_0 = const()[name = string("op_48392_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_48392_end_0 = const()[name = string("op_48392_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_48392_end_mask_0 = const()[name = string("op_48392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48392_cast_fp16 = slice_by_index(begin = var_48392_begin_0, end = var_48392_end_0, end_mask = var_48392_end_mask_0, x = value_cast_fp16)[name = string("op_48392_cast_fp16")]; + string _SplitHeadsQ__mh_w_4961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4961_equation_0, values = (var_48238_cast_fp16, var_47680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4961_cast_fp16")]; + string _SplitHeadsQ__mh_w_4963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4963_equation_0, values = (var_48238_cast_fp16, var_47687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4963_cast_fp16")]; + string _SplitHeadsQ__mh_w_4965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4965_equation_0, values = (var_48238_cast_fp16, var_47694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4965_cast_fp16")]; + string _SplitHeadsQ__mh_w_4967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4967_equation_0, values = (var_48238_cast_fp16, var_47701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4967_cast_fp16")]; + string _SplitHeadsQ__mh_w_4969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4969_equation_0, values = (var_48242_cast_fp16, var_47708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4969_cast_fp16")]; + string _SplitHeadsQ__mh_w_4971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4971_equation_0, values = (var_48242_cast_fp16, var_47715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4971_cast_fp16")]; + string _SplitHeadsQ__mh_w_4973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4973_equation_0, values = (var_48242_cast_fp16, var_47722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4973_cast_fp16")]; + string _SplitHeadsQ__mh_w_4975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4975_equation_0, values = (var_48242_cast_fp16, var_47729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4975_cast_fp16")]; + string _SplitHeadsQ__mh_w_4977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4977_equation_0, values = (var_48246_cast_fp16, var_47736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4977_cast_fp16")]; + string _SplitHeadsQ__mh_w_4979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4979_equation_0, values = (var_48246_cast_fp16, var_47743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4979_cast_fp16")]; + string _SplitHeadsQ__mh_w_4981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4981_equation_0, values = (var_48246_cast_fp16, var_47750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4981_cast_fp16")]; + string _SplitHeadsQ__mh_w_4983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4983_equation_0, values = (var_48246_cast_fp16, var_47757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4983_cast_fp16")]; + string _SplitHeadsQ__mh_w_4985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4985_equation_0, values = (var_48250_cast_fp16, var_47764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4985_cast_fp16")]; + string _SplitHeadsQ__mh_w_4987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4987_equation_0, values = (var_48250_cast_fp16, var_47771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4987_cast_fp16")]; + string _SplitHeadsQ__mh_w_4989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4989_equation_0, values = (var_48250_cast_fp16, var_47778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4989_cast_fp16")]; + string _SplitHeadsQ__mh_w_4991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4991_equation_0, values = (var_48250_cast_fp16, var_47785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4991_cast_fp16")]; + string _SplitHeadsQ__mh_w_4993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4993_equation_0, values = (var_48254_cast_fp16, var_47792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4993_cast_fp16")]; + string _SplitHeadsQ__mh_w_4995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4995_equation_0, values = (var_48254_cast_fp16, var_47799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4995_cast_fp16")]; + string _SplitHeadsQ__mh_w_4997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4997_equation_0, values = (var_48254_cast_fp16, var_47806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4997_cast_fp16")]; + string _SplitHeadsQ__mh_w_4999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4999_equation_0, values = (var_48254_cast_fp16, var_47813_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4999_cast_fp16")]; + string _SplitHeadsQ__mh_w_5001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5001_equation_0, values = (var_48258_cast_fp16, var_47820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5001_cast_fp16")]; + string _SplitHeadsQ__mh_w_5003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5003_equation_0, values = (var_48258_cast_fp16, var_47827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5003_cast_fp16")]; + string _SplitHeadsQ__mh_w_5005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5005_equation_0, values = (var_48258_cast_fp16, var_47834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5005_cast_fp16")]; + string _SplitHeadsQ__mh_w_5007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5007_equation_0, values = (var_48258_cast_fp16, var_47841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5007_cast_fp16")]; + string _SplitHeadsQ__mh_w_5009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5009_equation_0, values = (var_48262_cast_fp16, var_47848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5009_cast_fp16")]; + string _SplitHeadsQ__mh_w_5011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5011_equation_0, values = (var_48262_cast_fp16, var_47855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5011_cast_fp16")]; + string _SplitHeadsQ__mh_w_5013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5013_equation_0, values = (var_48262_cast_fp16, var_47862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5013_cast_fp16")]; + string _SplitHeadsQ__mh_w_5015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5015_equation_0, values = (var_48262_cast_fp16, var_47869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5015_cast_fp16")]; + string _SplitHeadsQ__mh_w_5017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5017_equation_0, values = (var_48266_cast_fp16, var_47876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5017_cast_fp16")]; + string _SplitHeadsQ__mh_w_5019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5019_equation_0, values = (var_48266_cast_fp16, var_47883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5019_cast_fp16")]; + string _SplitHeadsQ__mh_w_5021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5021_equation_0, values = (var_48266_cast_fp16, var_47890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5021_cast_fp16")]; + string _SplitHeadsQ__mh_w_5023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5023_equation_0, values = (var_48266_cast_fp16, var_47897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5023_cast_fp16")]; + string _SplitHeadsQ__mh_w_5025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5025_equation_0, values = (var_48270_cast_fp16, var_47904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5025_cast_fp16")]; + string _SplitHeadsQ__mh_w_5027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5027_equation_0, values = (var_48270_cast_fp16, var_47911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5027_cast_fp16")]; + string _SplitHeadsQ__mh_w_5029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5029_equation_0, values = (var_48270_cast_fp16, var_47918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5029_cast_fp16")]; + string _SplitHeadsQ__mh_w_5031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5031_equation_0, values = (var_48270_cast_fp16, var_47925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5031_cast_fp16")]; + string _SplitHeadsQ__mh_w_5033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5033_equation_0, values = (var_48274_cast_fp16, var_47932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5033_cast_fp16")]; + string _SplitHeadsQ__mh_w_5035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5035_equation_0, values = (var_48274_cast_fp16, var_47939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5035_cast_fp16")]; + string _SplitHeadsQ__mh_w_5037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5037_equation_0, values = (var_48274_cast_fp16, var_47946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5037_cast_fp16")]; + string _SplitHeadsQ__mh_w_5039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5039_equation_0, values = (var_48274_cast_fp16, var_47953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5039_cast_fp16")]; + string _SplitHeadsQ__mh_w_5041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5041_equation_0, values = (var_48278_cast_fp16, var_47960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5041_cast_fp16")]; + string _SplitHeadsQ__mh_w_5043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5043_equation_0, values = (var_48278_cast_fp16, var_47967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5043_cast_fp16")]; + string _SplitHeadsQ__mh_w_5045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5045_equation_0, values = (var_48278_cast_fp16, var_47974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5045_cast_fp16")]; + string _SplitHeadsQ__mh_w_5047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5047_equation_0, values = (var_48278_cast_fp16, var_47981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5047_cast_fp16")]; + string _SplitHeadsQ__mh_w_5049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5049_equation_0, values = (var_48282_cast_fp16, var_47988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5049_cast_fp16")]; + string _SplitHeadsQ__mh_w_5051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5051_equation_0, values = (var_48282_cast_fp16, var_47995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5051_cast_fp16")]; + string _SplitHeadsQ__mh_w_5053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5053_equation_0, values = (var_48282_cast_fp16, var_48002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5053_cast_fp16")]; + string _SplitHeadsQ__mh_w_5055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5055_equation_0, values = (var_48282_cast_fp16, var_48009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5055_cast_fp16")]; + string _SplitHeadsQ__mh_w_5057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5057_equation_0, values = (var_48286_cast_fp16, var_48016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5057_cast_fp16")]; + string _SplitHeadsQ__mh_w_5059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5059_equation_0, values = (var_48286_cast_fp16, var_48023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5059_cast_fp16")]; + string _SplitHeadsQ__mh_w_5061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5061_equation_0, values = (var_48286_cast_fp16, var_48030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5061_cast_fp16")]; + string _SplitHeadsQ__mh_w_5063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5063_equation_0, values = (var_48286_cast_fp16, var_48037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5063_cast_fp16")]; + string _SplitHeadsQ__mh_w_5065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5065_equation_0, values = (var_48290_cast_fp16, var_48044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5065_cast_fp16")]; + string _SplitHeadsQ__mh_w_5067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5067_equation_0, values = (var_48290_cast_fp16, var_48051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5067_cast_fp16")]; + string _SplitHeadsQ__mh_w_5069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5069_equation_0, values = (var_48290_cast_fp16, var_48058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5069_cast_fp16")]; + string _SplitHeadsQ__mh_w_5071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5071_equation_0, values = (var_48290_cast_fp16, var_48065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5071_cast_fp16")]; + string _SplitHeadsQ__mh_w_5073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5073_equation_0, values = (var_48294_cast_fp16, var_48072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5073_cast_fp16")]; + string _SplitHeadsQ__mh_w_5075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5075_equation_0, values = (var_48294_cast_fp16, var_48079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5075_cast_fp16")]; + string _SplitHeadsQ__mh_w_5077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5077_equation_0, values = (var_48294_cast_fp16, var_48086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5077_cast_fp16")]; + string _SplitHeadsQ__mh_w_5079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5079_equation_0, values = (var_48294_cast_fp16, var_48093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5079_cast_fp16")]; + string _SplitHeadsQ__mh_w_5081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5081_equation_0, values = (var_48298_cast_fp16, var_48100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5081_cast_fp16")]; + string _SplitHeadsQ__mh_w_5083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5083_equation_0, values = (var_48298_cast_fp16, var_48107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5083_cast_fp16")]; + string _SplitHeadsQ__mh_w_5085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5085_equation_0, values = (var_48298_cast_fp16, var_48114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5085_cast_fp16")]; + string _SplitHeadsQ__mh_w_5087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5087_equation_0, values = (var_48298_cast_fp16, var_48121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5087_cast_fp16")]; + string _SplitHeadsQ__mh_w_5089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5089_equation_0, values = (var_48302_cast_fp16, var_48128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5089_cast_fp16")]; + string _SplitHeadsQ__mh_w_5091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5091_equation_0, values = (var_48302_cast_fp16, var_48135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5091_cast_fp16")]; + string _SplitHeadsQ__mh_w_5093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5093_equation_0, values = (var_48302_cast_fp16, var_48142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5093_cast_fp16")]; + string _SplitHeadsQ__mh_w_5095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5095_equation_0, values = (var_48302_cast_fp16, var_48149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5095_cast_fp16")]; + string _SplitHeadsQ__mh_w_5097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5097_equation_0, values = (var_48306_cast_fp16, var_48156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5097_cast_fp16")]; + string _SplitHeadsQ__mh_w_5099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5099_equation_0, values = (var_48306_cast_fp16, var_48163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5099_cast_fp16")]; + string _SplitHeadsQ__mh_w_5101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5101_equation_0, values = (var_48306_cast_fp16, var_48170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5101_cast_fp16")]; + string _SplitHeadsQ__mh_w_5103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5103_equation_0, values = (var_48306_cast_fp16, var_48177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5103_cast_fp16")]; + string _SplitHeadsQ__mh_w_5105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5105_equation_0, values = (var_48310_cast_fp16, var_48184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5105_cast_fp16")]; + string _SplitHeadsQ__mh_w_5107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5107_equation_0, values = (var_48310_cast_fp16, var_48191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5107_cast_fp16")]; + string _SplitHeadsQ__mh_w_5109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5109_equation_0, values = (var_48310_cast_fp16, var_48198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5109_cast_fp16")]; + string _SplitHeadsQ__mh_w_5111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5111_equation_0, values = (var_48310_cast_fp16, var_48205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5111_cast_fp16")]; + string _SplitHeadsQ__mh_w_5113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5113_equation_0, values = (var_48314_cast_fp16, var_48212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5113_cast_fp16")]; + string _SplitHeadsQ__mh_w_5115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5115_equation_0, values = (var_48314_cast_fp16, var_48219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5115_cast_fp16")]; + string _SplitHeadsQ__mh_w_5117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5117_equation_0, values = (var_48314_cast_fp16, var_48226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5117_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_48314_cast_fp16, var_48233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_48555_to_fp16 = const()[name = string("op_48555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4961_cast_fp16, y = var_48555_to_fp16)[name = string("aw_chunk_4961_cast_fp16")]; + fp16 var_48557_to_fp16 = const()[name = string("op_48557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4963_cast_fp16, y = var_48557_to_fp16)[name = string("aw_chunk_4963_cast_fp16")]; + fp16 var_48559_to_fp16 = const()[name = string("op_48559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4965_cast_fp16, y = var_48559_to_fp16)[name = string("aw_chunk_4965_cast_fp16")]; + fp16 var_48561_to_fp16 = const()[name = string("op_48561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4967_cast_fp16, y = var_48561_to_fp16)[name = string("aw_chunk_4967_cast_fp16")]; + fp16 var_48563_to_fp16 = const()[name = string("op_48563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4969_cast_fp16, y = var_48563_to_fp16)[name = string("aw_chunk_4969_cast_fp16")]; + fp16 var_48565_to_fp16 = const()[name = string("op_48565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4971_cast_fp16, y = var_48565_to_fp16)[name = string("aw_chunk_4971_cast_fp16")]; + fp16 var_48567_to_fp16 = const()[name = string("op_48567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4973_cast_fp16, y = var_48567_to_fp16)[name = string("aw_chunk_4973_cast_fp16")]; + fp16 var_48569_to_fp16 = const()[name = string("op_48569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4975_cast_fp16, y = var_48569_to_fp16)[name = string("aw_chunk_4975_cast_fp16")]; + fp16 var_48571_to_fp16 = const()[name = string("op_48571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4977_cast_fp16, y = var_48571_to_fp16)[name = string("aw_chunk_4977_cast_fp16")]; + fp16 var_48573_to_fp16 = const()[name = string("op_48573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4979_cast_fp16, y = var_48573_to_fp16)[name = string("aw_chunk_4979_cast_fp16")]; + fp16 var_48575_to_fp16 = const()[name = string("op_48575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4981_cast_fp16, y = var_48575_to_fp16)[name = string("aw_chunk_4981_cast_fp16")]; + fp16 var_48577_to_fp16 = const()[name = string("op_48577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4983_cast_fp16, y = var_48577_to_fp16)[name = string("aw_chunk_4983_cast_fp16")]; + fp16 var_48579_to_fp16 = const()[name = string("op_48579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4985_cast_fp16, y = var_48579_to_fp16)[name = string("aw_chunk_4985_cast_fp16")]; + fp16 var_48581_to_fp16 = const()[name = string("op_48581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4987_cast_fp16, y = var_48581_to_fp16)[name = string("aw_chunk_4987_cast_fp16")]; + fp16 var_48583_to_fp16 = const()[name = string("op_48583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4989_cast_fp16, y = var_48583_to_fp16)[name = string("aw_chunk_4989_cast_fp16")]; + fp16 var_48585_to_fp16 = const()[name = string("op_48585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4991_cast_fp16, y = var_48585_to_fp16)[name = string("aw_chunk_4991_cast_fp16")]; + fp16 var_48587_to_fp16 = const()[name = string("op_48587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4993_cast_fp16, y = var_48587_to_fp16)[name = string("aw_chunk_4993_cast_fp16")]; + fp16 var_48589_to_fp16 = const()[name = string("op_48589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4995_cast_fp16, y = var_48589_to_fp16)[name = string("aw_chunk_4995_cast_fp16")]; + fp16 var_48591_to_fp16 = const()[name = string("op_48591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4997_cast_fp16, y = var_48591_to_fp16)[name = string("aw_chunk_4997_cast_fp16")]; + fp16 var_48593_to_fp16 = const()[name = string("op_48593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4999_cast_fp16, y = var_48593_to_fp16)[name = string("aw_chunk_4999_cast_fp16")]; + fp16 var_48595_to_fp16 = const()[name = string("op_48595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5001_cast_fp16, y = var_48595_to_fp16)[name = string("aw_chunk_5001_cast_fp16")]; + fp16 var_48597_to_fp16 = const()[name = string("op_48597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5003_cast_fp16, y = var_48597_to_fp16)[name = string("aw_chunk_5003_cast_fp16")]; + fp16 var_48599_to_fp16 = const()[name = string("op_48599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5005_cast_fp16, y = var_48599_to_fp16)[name = string("aw_chunk_5005_cast_fp16")]; + fp16 var_48601_to_fp16 = const()[name = string("op_48601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5007_cast_fp16, y = var_48601_to_fp16)[name = string("aw_chunk_5007_cast_fp16")]; + fp16 var_48603_to_fp16 = const()[name = string("op_48603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5009_cast_fp16, y = var_48603_to_fp16)[name = string("aw_chunk_5009_cast_fp16")]; + fp16 var_48605_to_fp16 = const()[name = string("op_48605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5011_cast_fp16, y = var_48605_to_fp16)[name = string("aw_chunk_5011_cast_fp16")]; + fp16 var_48607_to_fp16 = const()[name = string("op_48607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5013_cast_fp16, y = var_48607_to_fp16)[name = string("aw_chunk_5013_cast_fp16")]; + fp16 var_48609_to_fp16 = const()[name = string("op_48609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5015_cast_fp16, y = var_48609_to_fp16)[name = string("aw_chunk_5015_cast_fp16")]; + fp16 var_48611_to_fp16 = const()[name = string("op_48611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5017_cast_fp16, y = var_48611_to_fp16)[name = string("aw_chunk_5017_cast_fp16")]; + fp16 var_48613_to_fp16 = const()[name = string("op_48613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5019_cast_fp16, y = var_48613_to_fp16)[name = string("aw_chunk_5019_cast_fp16")]; + fp16 var_48615_to_fp16 = const()[name = string("op_48615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5021_cast_fp16, y = var_48615_to_fp16)[name = string("aw_chunk_5021_cast_fp16")]; + fp16 var_48617_to_fp16 = const()[name = string("op_48617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5023_cast_fp16, y = var_48617_to_fp16)[name = string("aw_chunk_5023_cast_fp16")]; + fp16 var_48619_to_fp16 = const()[name = string("op_48619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5025_cast_fp16, y = var_48619_to_fp16)[name = string("aw_chunk_5025_cast_fp16")]; + fp16 var_48621_to_fp16 = const()[name = string("op_48621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5027_cast_fp16, y = var_48621_to_fp16)[name = string("aw_chunk_5027_cast_fp16")]; + fp16 var_48623_to_fp16 = const()[name = string("op_48623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5029_cast_fp16, y = var_48623_to_fp16)[name = string("aw_chunk_5029_cast_fp16")]; + fp16 var_48625_to_fp16 = const()[name = string("op_48625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5031_cast_fp16, y = var_48625_to_fp16)[name = string("aw_chunk_5031_cast_fp16")]; + fp16 var_48627_to_fp16 = const()[name = string("op_48627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5033_cast_fp16, y = var_48627_to_fp16)[name = string("aw_chunk_5033_cast_fp16")]; + fp16 var_48629_to_fp16 = const()[name = string("op_48629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5035_cast_fp16, y = var_48629_to_fp16)[name = string("aw_chunk_5035_cast_fp16")]; + fp16 var_48631_to_fp16 = const()[name = string("op_48631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5037_cast_fp16, y = var_48631_to_fp16)[name = string("aw_chunk_5037_cast_fp16")]; + fp16 var_48633_to_fp16 = const()[name = string("op_48633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5039_cast_fp16, y = var_48633_to_fp16)[name = string("aw_chunk_5039_cast_fp16")]; + fp16 var_48635_to_fp16 = const()[name = string("op_48635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5041_cast_fp16, y = var_48635_to_fp16)[name = string("aw_chunk_5041_cast_fp16")]; + fp16 var_48637_to_fp16 = const()[name = string("op_48637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5043_cast_fp16, y = var_48637_to_fp16)[name = string("aw_chunk_5043_cast_fp16")]; + fp16 var_48639_to_fp16 = const()[name = string("op_48639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5045_cast_fp16, y = var_48639_to_fp16)[name = string("aw_chunk_5045_cast_fp16")]; + fp16 var_48641_to_fp16 = const()[name = string("op_48641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5047_cast_fp16, y = var_48641_to_fp16)[name = string("aw_chunk_5047_cast_fp16")]; + fp16 var_48643_to_fp16 = const()[name = string("op_48643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5049_cast_fp16, y = var_48643_to_fp16)[name = string("aw_chunk_5049_cast_fp16")]; + fp16 var_48645_to_fp16 = const()[name = string("op_48645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5051_cast_fp16, y = var_48645_to_fp16)[name = string("aw_chunk_5051_cast_fp16")]; + fp16 var_48647_to_fp16 = const()[name = string("op_48647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5053_cast_fp16, y = var_48647_to_fp16)[name = string("aw_chunk_5053_cast_fp16")]; + fp16 var_48649_to_fp16 = const()[name = string("op_48649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5055_cast_fp16, y = var_48649_to_fp16)[name = string("aw_chunk_5055_cast_fp16")]; + fp16 var_48651_to_fp16 = const()[name = string("op_48651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5057_cast_fp16, y = var_48651_to_fp16)[name = string("aw_chunk_5057_cast_fp16")]; + fp16 var_48653_to_fp16 = const()[name = string("op_48653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5059_cast_fp16, y = var_48653_to_fp16)[name = string("aw_chunk_5059_cast_fp16")]; + fp16 var_48655_to_fp16 = const()[name = string("op_48655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5061_cast_fp16, y = var_48655_to_fp16)[name = string("aw_chunk_5061_cast_fp16")]; + fp16 var_48657_to_fp16 = const()[name = string("op_48657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5063_cast_fp16, y = var_48657_to_fp16)[name = string("aw_chunk_5063_cast_fp16")]; + fp16 var_48659_to_fp16 = const()[name = string("op_48659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5065_cast_fp16, y = var_48659_to_fp16)[name = string("aw_chunk_5065_cast_fp16")]; + fp16 var_48661_to_fp16 = const()[name = string("op_48661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5067_cast_fp16, y = var_48661_to_fp16)[name = string("aw_chunk_5067_cast_fp16")]; + fp16 var_48663_to_fp16 = const()[name = string("op_48663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5069_cast_fp16, y = var_48663_to_fp16)[name = string("aw_chunk_5069_cast_fp16")]; + fp16 var_48665_to_fp16 = const()[name = string("op_48665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5071_cast_fp16, y = var_48665_to_fp16)[name = string("aw_chunk_5071_cast_fp16")]; + fp16 var_48667_to_fp16 = const()[name = string("op_48667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5073_cast_fp16, y = var_48667_to_fp16)[name = string("aw_chunk_5073_cast_fp16")]; + fp16 var_48669_to_fp16 = const()[name = string("op_48669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5075_cast_fp16, y = var_48669_to_fp16)[name = string("aw_chunk_5075_cast_fp16")]; + fp16 var_48671_to_fp16 = const()[name = string("op_48671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5077_cast_fp16, y = var_48671_to_fp16)[name = string("aw_chunk_5077_cast_fp16")]; + fp16 var_48673_to_fp16 = const()[name = string("op_48673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5079_cast_fp16, y = var_48673_to_fp16)[name = string("aw_chunk_5079_cast_fp16")]; + fp16 var_48675_to_fp16 = const()[name = string("op_48675_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5081_cast_fp16, y = var_48675_to_fp16)[name = string("aw_chunk_5081_cast_fp16")]; + fp16 var_48677_to_fp16 = const()[name = string("op_48677_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5083_cast_fp16, y = var_48677_to_fp16)[name = string("aw_chunk_5083_cast_fp16")]; + fp16 var_48679_to_fp16 = const()[name = string("op_48679_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5085_cast_fp16, y = var_48679_to_fp16)[name = string("aw_chunk_5085_cast_fp16")]; + fp16 var_48681_to_fp16 = const()[name = string("op_48681_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5087_cast_fp16, y = var_48681_to_fp16)[name = string("aw_chunk_5087_cast_fp16")]; + fp16 var_48683_to_fp16 = const()[name = string("op_48683_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5089_cast_fp16, y = var_48683_to_fp16)[name = string("aw_chunk_5089_cast_fp16")]; + fp16 var_48685_to_fp16 = const()[name = string("op_48685_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5091_cast_fp16, y = var_48685_to_fp16)[name = string("aw_chunk_5091_cast_fp16")]; + fp16 var_48687_to_fp16 = const()[name = string("op_48687_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5093_cast_fp16, y = var_48687_to_fp16)[name = string("aw_chunk_5093_cast_fp16")]; + fp16 var_48689_to_fp16 = const()[name = string("op_48689_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5095_cast_fp16, y = var_48689_to_fp16)[name = string("aw_chunk_5095_cast_fp16")]; + fp16 var_48691_to_fp16 = const()[name = string("op_48691_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5097_cast_fp16, y = var_48691_to_fp16)[name = string("aw_chunk_5097_cast_fp16")]; + fp16 var_48693_to_fp16 = const()[name = string("op_48693_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5099_cast_fp16, y = var_48693_to_fp16)[name = string("aw_chunk_5099_cast_fp16")]; + fp16 var_48695_to_fp16 = const()[name = string("op_48695_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5101_cast_fp16, y = var_48695_to_fp16)[name = string("aw_chunk_5101_cast_fp16")]; + fp16 var_48697_to_fp16 = const()[name = string("op_48697_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5103_cast_fp16, y = var_48697_to_fp16)[name = string("aw_chunk_5103_cast_fp16")]; + fp16 var_48699_to_fp16 = const()[name = string("op_48699_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5105_cast_fp16, y = var_48699_to_fp16)[name = string("aw_chunk_5105_cast_fp16")]; + fp16 var_48701_to_fp16 = const()[name = string("op_48701_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5107_cast_fp16, y = var_48701_to_fp16)[name = string("aw_chunk_5107_cast_fp16")]; + fp16 var_48703_to_fp16 = const()[name = string("op_48703_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5109_cast_fp16, y = var_48703_to_fp16)[name = string("aw_chunk_5109_cast_fp16")]; + fp16 var_48705_to_fp16 = const()[name = string("op_48705_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5111_cast_fp16, y = var_48705_to_fp16)[name = string("aw_chunk_5111_cast_fp16")]; + fp16 var_48707_to_fp16 = const()[name = string("op_48707_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5113_cast_fp16, y = var_48707_to_fp16)[name = string("aw_chunk_5113_cast_fp16")]; + fp16 var_48709_to_fp16 = const()[name = string("op_48709_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5115_cast_fp16, y = var_48709_to_fp16)[name = string("aw_chunk_5115_cast_fp16")]; + fp16 var_48711_to_fp16 = const()[name = string("op_48711_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5117_cast_fp16, y = var_48711_to_fp16)[name = string("aw_chunk_5117_cast_fp16")]; + fp16 var_48713_to_fp16 = const()[name = string("op_48713_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_48713_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_48715_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4961_cast_fp16)[name = string("op_48715_cast_fp16")]; + tensor var_48716_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4963_cast_fp16)[name = string("op_48716_cast_fp16")]; + tensor var_48717_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4965_cast_fp16)[name = string("op_48717_cast_fp16")]; + tensor var_48718_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4967_cast_fp16)[name = string("op_48718_cast_fp16")]; + tensor var_48719_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4969_cast_fp16)[name = string("op_48719_cast_fp16")]; + tensor var_48720_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4971_cast_fp16)[name = string("op_48720_cast_fp16")]; + tensor var_48721_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4973_cast_fp16)[name = string("op_48721_cast_fp16")]; + tensor var_48722_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4975_cast_fp16)[name = string("op_48722_cast_fp16")]; + tensor var_48723_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4977_cast_fp16)[name = string("op_48723_cast_fp16")]; + tensor var_48724_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4979_cast_fp16)[name = string("op_48724_cast_fp16")]; + tensor var_48725_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4981_cast_fp16)[name = string("op_48725_cast_fp16")]; + tensor var_48726_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4983_cast_fp16)[name = string("op_48726_cast_fp16")]; + tensor var_48727_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4985_cast_fp16)[name = string("op_48727_cast_fp16")]; + tensor var_48728_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4987_cast_fp16)[name = string("op_48728_cast_fp16")]; + tensor var_48729_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4989_cast_fp16)[name = string("op_48729_cast_fp16")]; + tensor var_48730_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4991_cast_fp16)[name = string("op_48730_cast_fp16")]; + tensor var_48731_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4993_cast_fp16)[name = string("op_48731_cast_fp16")]; + tensor var_48732_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4995_cast_fp16)[name = string("op_48732_cast_fp16")]; + tensor var_48733_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4997_cast_fp16)[name = string("op_48733_cast_fp16")]; + tensor var_48734_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_4999_cast_fp16)[name = string("op_48734_cast_fp16")]; + tensor var_48735_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5001_cast_fp16)[name = string("op_48735_cast_fp16")]; + tensor var_48736_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5003_cast_fp16)[name = string("op_48736_cast_fp16")]; + tensor var_48737_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5005_cast_fp16)[name = string("op_48737_cast_fp16")]; + tensor var_48738_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5007_cast_fp16)[name = string("op_48738_cast_fp16")]; + tensor var_48739_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5009_cast_fp16)[name = string("op_48739_cast_fp16")]; + tensor var_48740_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5011_cast_fp16)[name = string("op_48740_cast_fp16")]; + tensor var_48741_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5013_cast_fp16)[name = string("op_48741_cast_fp16")]; + tensor var_48742_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5015_cast_fp16)[name = string("op_48742_cast_fp16")]; + tensor var_48743_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5017_cast_fp16)[name = string("op_48743_cast_fp16")]; + tensor var_48744_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5019_cast_fp16)[name = string("op_48744_cast_fp16")]; + tensor var_48745_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5021_cast_fp16)[name = string("op_48745_cast_fp16")]; + tensor var_48746_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5023_cast_fp16)[name = string("op_48746_cast_fp16")]; + tensor var_48747_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5025_cast_fp16)[name = string("op_48747_cast_fp16")]; + tensor var_48748_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5027_cast_fp16)[name = string("op_48748_cast_fp16")]; + tensor var_48749_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5029_cast_fp16)[name = string("op_48749_cast_fp16")]; + tensor var_48750_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5031_cast_fp16)[name = string("op_48750_cast_fp16")]; + tensor var_48751_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5033_cast_fp16)[name = string("op_48751_cast_fp16")]; + tensor var_48752_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5035_cast_fp16)[name = string("op_48752_cast_fp16")]; + tensor var_48753_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5037_cast_fp16)[name = string("op_48753_cast_fp16")]; + tensor var_48754_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5039_cast_fp16)[name = string("op_48754_cast_fp16")]; + tensor var_48755_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5041_cast_fp16)[name = string("op_48755_cast_fp16")]; + tensor var_48756_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5043_cast_fp16)[name = string("op_48756_cast_fp16")]; + tensor var_48757_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5045_cast_fp16)[name = string("op_48757_cast_fp16")]; + tensor var_48758_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5047_cast_fp16)[name = string("op_48758_cast_fp16")]; + tensor var_48759_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5049_cast_fp16)[name = string("op_48759_cast_fp16")]; + tensor var_48760_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5051_cast_fp16)[name = string("op_48760_cast_fp16")]; + tensor var_48761_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5053_cast_fp16)[name = string("op_48761_cast_fp16")]; + tensor var_48762_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5055_cast_fp16)[name = string("op_48762_cast_fp16")]; + tensor var_48763_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5057_cast_fp16)[name = string("op_48763_cast_fp16")]; + tensor var_48764_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5059_cast_fp16)[name = string("op_48764_cast_fp16")]; + tensor var_48765_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5061_cast_fp16)[name = string("op_48765_cast_fp16")]; + tensor var_48766_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5063_cast_fp16)[name = string("op_48766_cast_fp16")]; + tensor var_48767_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5065_cast_fp16)[name = string("op_48767_cast_fp16")]; + tensor var_48768_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5067_cast_fp16)[name = string("op_48768_cast_fp16")]; + tensor var_48769_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5069_cast_fp16)[name = string("op_48769_cast_fp16")]; + tensor var_48770_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5071_cast_fp16)[name = string("op_48770_cast_fp16")]; + tensor var_48771_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5073_cast_fp16)[name = string("op_48771_cast_fp16")]; + tensor var_48772_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5075_cast_fp16)[name = string("op_48772_cast_fp16")]; + tensor var_48773_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5077_cast_fp16)[name = string("op_48773_cast_fp16")]; + tensor var_48774_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5079_cast_fp16)[name = string("op_48774_cast_fp16")]; + tensor var_48775_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5081_cast_fp16)[name = string("op_48775_cast_fp16")]; + tensor var_48776_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5083_cast_fp16)[name = string("op_48776_cast_fp16")]; + tensor var_48777_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5085_cast_fp16)[name = string("op_48777_cast_fp16")]; + tensor var_48778_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5087_cast_fp16)[name = string("op_48778_cast_fp16")]; + tensor var_48779_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5089_cast_fp16)[name = string("op_48779_cast_fp16")]; + tensor var_48780_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5091_cast_fp16)[name = string("op_48780_cast_fp16")]; + tensor var_48781_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5093_cast_fp16)[name = string("op_48781_cast_fp16")]; + tensor var_48782_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5095_cast_fp16)[name = string("op_48782_cast_fp16")]; + tensor var_48783_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5097_cast_fp16)[name = string("op_48783_cast_fp16")]; + tensor var_48784_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5099_cast_fp16)[name = string("op_48784_cast_fp16")]; + tensor var_48785_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5101_cast_fp16)[name = string("op_48785_cast_fp16")]; + tensor var_48786_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5103_cast_fp16)[name = string("op_48786_cast_fp16")]; + tensor var_48787_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5105_cast_fp16)[name = string("op_48787_cast_fp16")]; + tensor var_48788_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5107_cast_fp16)[name = string("op_48788_cast_fp16")]; + tensor var_48789_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5109_cast_fp16)[name = string("op_48789_cast_fp16")]; + tensor var_48790_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5111_cast_fp16)[name = string("op_48790_cast_fp16")]; + tensor var_48791_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5113_cast_fp16)[name = string("op_48791_cast_fp16")]; + tensor var_48792_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5115_cast_fp16)[name = string("op_48792_cast_fp16")]; + tensor var_48793_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_5117_cast_fp16)[name = string("op_48793_cast_fp16")]; + tensor var_48794_cast_fp16 = softmax(axis = var_47540, x = aw_chunk_cast_fp16)[name = string("op_48794_cast_fp16")]; + string var_48796_equation_0 = const()[name = string("op_48796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48796_cast_fp16 = einsum(equation = var_48796_equation_0, values = (var_48316_cast_fp16, var_48715_cast_fp16))[name = string("op_48796_cast_fp16")]; + string var_48798_equation_0 = const()[name = string("op_48798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48798_cast_fp16 = einsum(equation = var_48798_equation_0, values = (var_48316_cast_fp16, var_48716_cast_fp16))[name = string("op_48798_cast_fp16")]; + string var_48800_equation_0 = const()[name = string("op_48800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48800_cast_fp16 = einsum(equation = var_48800_equation_0, values = (var_48316_cast_fp16, var_48717_cast_fp16))[name = string("op_48800_cast_fp16")]; + string var_48802_equation_0 = const()[name = string("op_48802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48802_cast_fp16 = einsum(equation = var_48802_equation_0, values = (var_48316_cast_fp16, var_48718_cast_fp16))[name = string("op_48802_cast_fp16")]; + string var_48804_equation_0 = const()[name = string("op_48804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48804_cast_fp16 = einsum(equation = var_48804_equation_0, values = (var_48320_cast_fp16, var_48719_cast_fp16))[name = string("op_48804_cast_fp16")]; + string var_48806_equation_0 = const()[name = string("op_48806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48806_cast_fp16 = einsum(equation = var_48806_equation_0, values = (var_48320_cast_fp16, var_48720_cast_fp16))[name = string("op_48806_cast_fp16")]; + string var_48808_equation_0 = const()[name = string("op_48808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48808_cast_fp16 = einsum(equation = var_48808_equation_0, values = (var_48320_cast_fp16, var_48721_cast_fp16))[name = string("op_48808_cast_fp16")]; + string var_48810_equation_0 = const()[name = string("op_48810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48810_cast_fp16 = einsum(equation = var_48810_equation_0, values = (var_48320_cast_fp16, var_48722_cast_fp16))[name = string("op_48810_cast_fp16")]; + string var_48812_equation_0 = const()[name = string("op_48812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48812_cast_fp16 = einsum(equation = var_48812_equation_0, values = (var_48324_cast_fp16, var_48723_cast_fp16))[name = string("op_48812_cast_fp16")]; + string var_48814_equation_0 = const()[name = string("op_48814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48814_cast_fp16 = einsum(equation = var_48814_equation_0, values = (var_48324_cast_fp16, var_48724_cast_fp16))[name = string("op_48814_cast_fp16")]; + string var_48816_equation_0 = const()[name = string("op_48816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48816_cast_fp16 = einsum(equation = var_48816_equation_0, values = (var_48324_cast_fp16, var_48725_cast_fp16))[name = string("op_48816_cast_fp16")]; + string var_48818_equation_0 = const()[name = string("op_48818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48818_cast_fp16 = einsum(equation = var_48818_equation_0, values = (var_48324_cast_fp16, var_48726_cast_fp16))[name = string("op_48818_cast_fp16")]; + string var_48820_equation_0 = const()[name = string("op_48820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48820_cast_fp16 = einsum(equation = var_48820_equation_0, values = (var_48328_cast_fp16, var_48727_cast_fp16))[name = string("op_48820_cast_fp16")]; + string var_48822_equation_0 = const()[name = string("op_48822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48822_cast_fp16 = einsum(equation = var_48822_equation_0, values = (var_48328_cast_fp16, var_48728_cast_fp16))[name = string("op_48822_cast_fp16")]; + string var_48824_equation_0 = const()[name = string("op_48824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48824_cast_fp16 = einsum(equation = var_48824_equation_0, values = (var_48328_cast_fp16, var_48729_cast_fp16))[name = string("op_48824_cast_fp16")]; + string var_48826_equation_0 = const()[name = string("op_48826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48826_cast_fp16 = einsum(equation = var_48826_equation_0, values = (var_48328_cast_fp16, var_48730_cast_fp16))[name = string("op_48826_cast_fp16")]; + string var_48828_equation_0 = const()[name = string("op_48828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48828_cast_fp16 = einsum(equation = var_48828_equation_0, values = (var_48332_cast_fp16, var_48731_cast_fp16))[name = string("op_48828_cast_fp16")]; + string var_48830_equation_0 = const()[name = string("op_48830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48830_cast_fp16 = einsum(equation = var_48830_equation_0, values = (var_48332_cast_fp16, var_48732_cast_fp16))[name = string("op_48830_cast_fp16")]; + string var_48832_equation_0 = const()[name = string("op_48832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48832_cast_fp16 = einsum(equation = var_48832_equation_0, values = (var_48332_cast_fp16, var_48733_cast_fp16))[name = string("op_48832_cast_fp16")]; + string var_48834_equation_0 = const()[name = string("op_48834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48834_cast_fp16 = einsum(equation = var_48834_equation_0, values = (var_48332_cast_fp16, var_48734_cast_fp16))[name = string("op_48834_cast_fp16")]; + string var_48836_equation_0 = const()[name = string("op_48836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48836_cast_fp16 = einsum(equation = var_48836_equation_0, values = (var_48336_cast_fp16, var_48735_cast_fp16))[name = string("op_48836_cast_fp16")]; + string var_48838_equation_0 = const()[name = string("op_48838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48838_cast_fp16 = einsum(equation = var_48838_equation_0, values = (var_48336_cast_fp16, var_48736_cast_fp16))[name = string("op_48838_cast_fp16")]; + string var_48840_equation_0 = const()[name = string("op_48840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48840_cast_fp16 = einsum(equation = var_48840_equation_0, values = (var_48336_cast_fp16, var_48737_cast_fp16))[name = string("op_48840_cast_fp16")]; + string var_48842_equation_0 = const()[name = string("op_48842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48842_cast_fp16 = einsum(equation = var_48842_equation_0, values = (var_48336_cast_fp16, var_48738_cast_fp16))[name = string("op_48842_cast_fp16")]; + string var_48844_equation_0 = const()[name = string("op_48844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48844_cast_fp16 = einsum(equation = var_48844_equation_0, values = (var_48340_cast_fp16, var_48739_cast_fp16))[name = string("op_48844_cast_fp16")]; + string var_48846_equation_0 = const()[name = string("op_48846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48846_cast_fp16 = einsum(equation = var_48846_equation_0, values = (var_48340_cast_fp16, var_48740_cast_fp16))[name = string("op_48846_cast_fp16")]; + string var_48848_equation_0 = const()[name = string("op_48848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48848_cast_fp16 = einsum(equation = var_48848_equation_0, values = (var_48340_cast_fp16, var_48741_cast_fp16))[name = string("op_48848_cast_fp16")]; + string var_48850_equation_0 = const()[name = string("op_48850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48850_cast_fp16 = einsum(equation = var_48850_equation_0, values = (var_48340_cast_fp16, var_48742_cast_fp16))[name = string("op_48850_cast_fp16")]; + string var_48852_equation_0 = const()[name = string("op_48852_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48852_cast_fp16 = einsum(equation = var_48852_equation_0, values = (var_48344_cast_fp16, var_48743_cast_fp16))[name = string("op_48852_cast_fp16")]; + string var_48854_equation_0 = const()[name = string("op_48854_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48854_cast_fp16 = einsum(equation = var_48854_equation_0, values = (var_48344_cast_fp16, var_48744_cast_fp16))[name = string("op_48854_cast_fp16")]; + string var_48856_equation_0 = const()[name = string("op_48856_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48856_cast_fp16 = einsum(equation = var_48856_equation_0, values = (var_48344_cast_fp16, var_48745_cast_fp16))[name = string("op_48856_cast_fp16")]; + string var_48858_equation_0 = const()[name = string("op_48858_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48858_cast_fp16 = einsum(equation = var_48858_equation_0, values = (var_48344_cast_fp16, var_48746_cast_fp16))[name = string("op_48858_cast_fp16")]; + string var_48860_equation_0 = const()[name = string("op_48860_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48860_cast_fp16 = einsum(equation = var_48860_equation_0, values = (var_48348_cast_fp16, var_48747_cast_fp16))[name = string("op_48860_cast_fp16")]; + string var_48862_equation_0 = const()[name = string("op_48862_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48862_cast_fp16 = einsum(equation = var_48862_equation_0, values = (var_48348_cast_fp16, var_48748_cast_fp16))[name = string("op_48862_cast_fp16")]; + string var_48864_equation_0 = const()[name = string("op_48864_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48864_cast_fp16 = einsum(equation = var_48864_equation_0, values = (var_48348_cast_fp16, var_48749_cast_fp16))[name = string("op_48864_cast_fp16")]; + string var_48866_equation_0 = const()[name = string("op_48866_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48866_cast_fp16 = einsum(equation = var_48866_equation_0, values = (var_48348_cast_fp16, var_48750_cast_fp16))[name = string("op_48866_cast_fp16")]; + string var_48868_equation_0 = const()[name = string("op_48868_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48868_cast_fp16 = einsum(equation = var_48868_equation_0, values = (var_48352_cast_fp16, var_48751_cast_fp16))[name = string("op_48868_cast_fp16")]; + string var_48870_equation_0 = const()[name = string("op_48870_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48870_cast_fp16 = einsum(equation = var_48870_equation_0, values = (var_48352_cast_fp16, var_48752_cast_fp16))[name = string("op_48870_cast_fp16")]; + string var_48872_equation_0 = const()[name = string("op_48872_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48872_cast_fp16 = einsum(equation = var_48872_equation_0, values = (var_48352_cast_fp16, var_48753_cast_fp16))[name = string("op_48872_cast_fp16")]; + string var_48874_equation_0 = const()[name = string("op_48874_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48874_cast_fp16 = einsum(equation = var_48874_equation_0, values = (var_48352_cast_fp16, var_48754_cast_fp16))[name = string("op_48874_cast_fp16")]; + string var_48876_equation_0 = const()[name = string("op_48876_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48876_cast_fp16 = einsum(equation = var_48876_equation_0, values = (var_48356_cast_fp16, var_48755_cast_fp16))[name = string("op_48876_cast_fp16")]; + string var_48878_equation_0 = const()[name = string("op_48878_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48878_cast_fp16 = einsum(equation = var_48878_equation_0, values = (var_48356_cast_fp16, var_48756_cast_fp16))[name = string("op_48878_cast_fp16")]; + string var_48880_equation_0 = const()[name = string("op_48880_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48880_cast_fp16 = einsum(equation = var_48880_equation_0, values = (var_48356_cast_fp16, var_48757_cast_fp16))[name = string("op_48880_cast_fp16")]; + string var_48882_equation_0 = const()[name = string("op_48882_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48882_cast_fp16 = einsum(equation = var_48882_equation_0, values = (var_48356_cast_fp16, var_48758_cast_fp16))[name = string("op_48882_cast_fp16")]; + string var_48884_equation_0 = const()[name = string("op_48884_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48884_cast_fp16 = einsum(equation = var_48884_equation_0, values = (var_48360_cast_fp16, var_48759_cast_fp16))[name = string("op_48884_cast_fp16")]; + string var_48886_equation_0 = const()[name = string("op_48886_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48886_cast_fp16 = einsum(equation = var_48886_equation_0, values = (var_48360_cast_fp16, var_48760_cast_fp16))[name = string("op_48886_cast_fp16")]; + string var_48888_equation_0 = const()[name = string("op_48888_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48888_cast_fp16 = einsum(equation = var_48888_equation_0, values = (var_48360_cast_fp16, var_48761_cast_fp16))[name = string("op_48888_cast_fp16")]; + string var_48890_equation_0 = const()[name = string("op_48890_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48890_cast_fp16 = einsum(equation = var_48890_equation_0, values = (var_48360_cast_fp16, var_48762_cast_fp16))[name = string("op_48890_cast_fp16")]; + string var_48892_equation_0 = const()[name = string("op_48892_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48892_cast_fp16 = einsum(equation = var_48892_equation_0, values = (var_48364_cast_fp16, var_48763_cast_fp16))[name = string("op_48892_cast_fp16")]; + string var_48894_equation_0 = const()[name = string("op_48894_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48894_cast_fp16 = einsum(equation = var_48894_equation_0, values = (var_48364_cast_fp16, var_48764_cast_fp16))[name = string("op_48894_cast_fp16")]; + string var_48896_equation_0 = const()[name = string("op_48896_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48896_cast_fp16 = einsum(equation = var_48896_equation_0, values = (var_48364_cast_fp16, var_48765_cast_fp16))[name = string("op_48896_cast_fp16")]; + string var_48898_equation_0 = const()[name = string("op_48898_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48898_cast_fp16 = einsum(equation = var_48898_equation_0, values = (var_48364_cast_fp16, var_48766_cast_fp16))[name = string("op_48898_cast_fp16")]; + string var_48900_equation_0 = const()[name = string("op_48900_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48900_cast_fp16 = einsum(equation = var_48900_equation_0, values = (var_48368_cast_fp16, var_48767_cast_fp16))[name = string("op_48900_cast_fp16")]; + string var_48902_equation_0 = const()[name = string("op_48902_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48902_cast_fp16 = einsum(equation = var_48902_equation_0, values = (var_48368_cast_fp16, var_48768_cast_fp16))[name = string("op_48902_cast_fp16")]; + string var_48904_equation_0 = const()[name = string("op_48904_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48904_cast_fp16 = einsum(equation = var_48904_equation_0, values = (var_48368_cast_fp16, var_48769_cast_fp16))[name = string("op_48904_cast_fp16")]; + string var_48906_equation_0 = const()[name = string("op_48906_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48906_cast_fp16 = einsum(equation = var_48906_equation_0, values = (var_48368_cast_fp16, var_48770_cast_fp16))[name = string("op_48906_cast_fp16")]; + string var_48908_equation_0 = const()[name = string("op_48908_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48908_cast_fp16 = einsum(equation = var_48908_equation_0, values = (var_48372_cast_fp16, var_48771_cast_fp16))[name = string("op_48908_cast_fp16")]; + string var_48910_equation_0 = const()[name = string("op_48910_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48910_cast_fp16 = einsum(equation = var_48910_equation_0, values = (var_48372_cast_fp16, var_48772_cast_fp16))[name = string("op_48910_cast_fp16")]; + string var_48912_equation_0 = const()[name = string("op_48912_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48912_cast_fp16 = einsum(equation = var_48912_equation_0, values = (var_48372_cast_fp16, var_48773_cast_fp16))[name = string("op_48912_cast_fp16")]; + string var_48914_equation_0 = const()[name = string("op_48914_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48914_cast_fp16 = einsum(equation = var_48914_equation_0, values = (var_48372_cast_fp16, var_48774_cast_fp16))[name = string("op_48914_cast_fp16")]; + string var_48916_equation_0 = const()[name = string("op_48916_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48916_cast_fp16 = einsum(equation = var_48916_equation_0, values = (var_48376_cast_fp16, var_48775_cast_fp16))[name = string("op_48916_cast_fp16")]; + string var_48918_equation_0 = const()[name = string("op_48918_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48918_cast_fp16 = einsum(equation = var_48918_equation_0, values = (var_48376_cast_fp16, var_48776_cast_fp16))[name = string("op_48918_cast_fp16")]; + string var_48920_equation_0 = const()[name = string("op_48920_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48920_cast_fp16 = einsum(equation = var_48920_equation_0, values = (var_48376_cast_fp16, var_48777_cast_fp16))[name = string("op_48920_cast_fp16")]; + string var_48922_equation_0 = const()[name = string("op_48922_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48922_cast_fp16 = einsum(equation = var_48922_equation_0, values = (var_48376_cast_fp16, var_48778_cast_fp16))[name = string("op_48922_cast_fp16")]; + string var_48924_equation_0 = const()[name = string("op_48924_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48924_cast_fp16 = einsum(equation = var_48924_equation_0, values = (var_48380_cast_fp16, var_48779_cast_fp16))[name = string("op_48924_cast_fp16")]; + string var_48926_equation_0 = const()[name = string("op_48926_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48926_cast_fp16 = einsum(equation = var_48926_equation_0, values = (var_48380_cast_fp16, var_48780_cast_fp16))[name = string("op_48926_cast_fp16")]; + string var_48928_equation_0 = const()[name = string("op_48928_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48928_cast_fp16 = einsum(equation = var_48928_equation_0, values = (var_48380_cast_fp16, var_48781_cast_fp16))[name = string("op_48928_cast_fp16")]; + string var_48930_equation_0 = const()[name = string("op_48930_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48930_cast_fp16 = einsum(equation = var_48930_equation_0, values = (var_48380_cast_fp16, var_48782_cast_fp16))[name = string("op_48930_cast_fp16")]; + string var_48932_equation_0 = const()[name = string("op_48932_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48932_cast_fp16 = einsum(equation = var_48932_equation_0, values = (var_48384_cast_fp16, var_48783_cast_fp16))[name = string("op_48932_cast_fp16")]; + string var_48934_equation_0 = const()[name = string("op_48934_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48934_cast_fp16 = einsum(equation = var_48934_equation_0, values = (var_48384_cast_fp16, var_48784_cast_fp16))[name = string("op_48934_cast_fp16")]; + string var_48936_equation_0 = const()[name = string("op_48936_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48936_cast_fp16 = einsum(equation = var_48936_equation_0, values = (var_48384_cast_fp16, var_48785_cast_fp16))[name = string("op_48936_cast_fp16")]; + string var_48938_equation_0 = const()[name = string("op_48938_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48938_cast_fp16 = einsum(equation = var_48938_equation_0, values = (var_48384_cast_fp16, var_48786_cast_fp16))[name = string("op_48938_cast_fp16")]; + string var_48940_equation_0 = const()[name = string("op_48940_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48940_cast_fp16 = einsum(equation = var_48940_equation_0, values = (var_48388_cast_fp16, var_48787_cast_fp16))[name = string("op_48940_cast_fp16")]; + string var_48942_equation_0 = const()[name = string("op_48942_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48942_cast_fp16 = einsum(equation = var_48942_equation_0, values = (var_48388_cast_fp16, var_48788_cast_fp16))[name = string("op_48942_cast_fp16")]; + string var_48944_equation_0 = const()[name = string("op_48944_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48944_cast_fp16 = einsum(equation = var_48944_equation_0, values = (var_48388_cast_fp16, var_48789_cast_fp16))[name = string("op_48944_cast_fp16")]; + string var_48946_equation_0 = const()[name = string("op_48946_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48946_cast_fp16 = einsum(equation = var_48946_equation_0, values = (var_48388_cast_fp16, var_48790_cast_fp16))[name = string("op_48946_cast_fp16")]; + string var_48948_equation_0 = const()[name = string("op_48948_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48948_cast_fp16 = einsum(equation = var_48948_equation_0, values = (var_48392_cast_fp16, var_48791_cast_fp16))[name = string("op_48948_cast_fp16")]; + string var_48950_equation_0 = const()[name = string("op_48950_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48950_cast_fp16 = einsum(equation = var_48950_equation_0, values = (var_48392_cast_fp16, var_48792_cast_fp16))[name = string("op_48950_cast_fp16")]; + string var_48952_equation_0 = const()[name = string("op_48952_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48952_cast_fp16 = einsum(equation = var_48952_equation_0, values = (var_48392_cast_fp16, var_48793_cast_fp16))[name = string("op_48952_cast_fp16")]; + string var_48954_equation_0 = const()[name = string("op_48954_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48954_cast_fp16 = einsum(equation = var_48954_equation_0, values = (var_48392_cast_fp16, var_48794_cast_fp16))[name = string("op_48954_cast_fp16")]; + bool var_48956_interleave_0 = const()[name = string("op_48956_interleave_0"), val = bool(false)]; + tensor var_48956_cast_fp16 = concat(axis = var_47515, interleave = var_48956_interleave_0, values = (var_48796_cast_fp16, var_48798_cast_fp16, var_48800_cast_fp16, var_48802_cast_fp16))[name = string("op_48956_cast_fp16")]; + bool var_48958_interleave_0 = const()[name = string("op_48958_interleave_0"), val = bool(false)]; + tensor var_48958_cast_fp16 = concat(axis = var_47515, interleave = var_48958_interleave_0, values = (var_48804_cast_fp16, var_48806_cast_fp16, var_48808_cast_fp16, var_48810_cast_fp16))[name = string("op_48958_cast_fp16")]; + bool var_48960_interleave_0 = const()[name = string("op_48960_interleave_0"), val = bool(false)]; + tensor var_48960_cast_fp16 = concat(axis = var_47515, interleave = var_48960_interleave_0, values = (var_48812_cast_fp16, var_48814_cast_fp16, var_48816_cast_fp16, var_48818_cast_fp16))[name = string("op_48960_cast_fp16")]; + bool var_48962_interleave_0 = const()[name = string("op_48962_interleave_0"), val = bool(false)]; + tensor var_48962_cast_fp16 = concat(axis = var_47515, interleave = var_48962_interleave_0, values = (var_48820_cast_fp16, var_48822_cast_fp16, var_48824_cast_fp16, var_48826_cast_fp16))[name = string("op_48962_cast_fp16")]; + bool var_48964_interleave_0 = const()[name = string("op_48964_interleave_0"), val = bool(false)]; + tensor var_48964_cast_fp16 = concat(axis = var_47515, interleave = var_48964_interleave_0, values = (var_48828_cast_fp16, var_48830_cast_fp16, var_48832_cast_fp16, var_48834_cast_fp16))[name = string("op_48964_cast_fp16")]; + bool var_48966_interleave_0 = const()[name = string("op_48966_interleave_0"), val = bool(false)]; + tensor var_48966_cast_fp16 = concat(axis = var_47515, interleave = var_48966_interleave_0, values = (var_48836_cast_fp16, var_48838_cast_fp16, var_48840_cast_fp16, var_48842_cast_fp16))[name = string("op_48966_cast_fp16")]; + bool var_48968_interleave_0 = const()[name = string("op_48968_interleave_0"), val = bool(false)]; + tensor var_48968_cast_fp16 = concat(axis = var_47515, interleave = var_48968_interleave_0, values = (var_48844_cast_fp16, var_48846_cast_fp16, var_48848_cast_fp16, var_48850_cast_fp16))[name = string("op_48968_cast_fp16")]; + bool var_48970_interleave_0 = const()[name = string("op_48970_interleave_0"), val = bool(false)]; + tensor var_48970_cast_fp16 = concat(axis = var_47515, interleave = var_48970_interleave_0, values = (var_48852_cast_fp16, var_48854_cast_fp16, var_48856_cast_fp16, var_48858_cast_fp16))[name = string("op_48970_cast_fp16")]; + bool var_48972_interleave_0 = const()[name = string("op_48972_interleave_0"), val = bool(false)]; + tensor var_48972_cast_fp16 = concat(axis = var_47515, interleave = var_48972_interleave_0, values = (var_48860_cast_fp16, var_48862_cast_fp16, var_48864_cast_fp16, var_48866_cast_fp16))[name = string("op_48972_cast_fp16")]; + bool var_48974_interleave_0 = const()[name = string("op_48974_interleave_0"), val = bool(false)]; + tensor var_48974_cast_fp16 = concat(axis = var_47515, interleave = var_48974_interleave_0, values = (var_48868_cast_fp16, var_48870_cast_fp16, var_48872_cast_fp16, var_48874_cast_fp16))[name = string("op_48974_cast_fp16")]; + bool var_48976_interleave_0 = const()[name = string("op_48976_interleave_0"), val = bool(false)]; + tensor var_48976_cast_fp16 = concat(axis = var_47515, interleave = var_48976_interleave_0, values = (var_48876_cast_fp16, var_48878_cast_fp16, var_48880_cast_fp16, var_48882_cast_fp16))[name = string("op_48976_cast_fp16")]; + bool var_48978_interleave_0 = const()[name = string("op_48978_interleave_0"), val = bool(false)]; + tensor var_48978_cast_fp16 = concat(axis = var_47515, interleave = var_48978_interleave_0, values = (var_48884_cast_fp16, var_48886_cast_fp16, var_48888_cast_fp16, var_48890_cast_fp16))[name = string("op_48978_cast_fp16")]; + bool var_48980_interleave_0 = const()[name = string("op_48980_interleave_0"), val = bool(false)]; + tensor var_48980_cast_fp16 = concat(axis = var_47515, interleave = var_48980_interleave_0, values = (var_48892_cast_fp16, var_48894_cast_fp16, var_48896_cast_fp16, var_48898_cast_fp16))[name = string("op_48980_cast_fp16")]; + bool var_48982_interleave_0 = const()[name = string("op_48982_interleave_0"), val = bool(false)]; + tensor var_48982_cast_fp16 = concat(axis = var_47515, interleave = var_48982_interleave_0, values = (var_48900_cast_fp16, var_48902_cast_fp16, var_48904_cast_fp16, var_48906_cast_fp16))[name = string("op_48982_cast_fp16")]; + bool var_48984_interleave_0 = const()[name = string("op_48984_interleave_0"), val = bool(false)]; + tensor var_48984_cast_fp16 = concat(axis = var_47515, interleave = var_48984_interleave_0, values = (var_48908_cast_fp16, var_48910_cast_fp16, var_48912_cast_fp16, var_48914_cast_fp16))[name = string("op_48984_cast_fp16")]; + bool var_48986_interleave_0 = const()[name = string("op_48986_interleave_0"), val = bool(false)]; + tensor var_48986_cast_fp16 = concat(axis = var_47515, interleave = var_48986_interleave_0, values = (var_48916_cast_fp16, var_48918_cast_fp16, var_48920_cast_fp16, var_48922_cast_fp16))[name = string("op_48986_cast_fp16")]; + bool var_48988_interleave_0 = const()[name = string("op_48988_interleave_0"), val = bool(false)]; + tensor var_48988_cast_fp16 = concat(axis = var_47515, interleave = var_48988_interleave_0, values = (var_48924_cast_fp16, var_48926_cast_fp16, var_48928_cast_fp16, var_48930_cast_fp16))[name = string("op_48988_cast_fp16")]; + bool var_48990_interleave_0 = const()[name = string("op_48990_interleave_0"), val = bool(false)]; + tensor var_48990_cast_fp16 = concat(axis = var_47515, interleave = var_48990_interleave_0, values = (var_48932_cast_fp16, var_48934_cast_fp16, var_48936_cast_fp16, var_48938_cast_fp16))[name = string("op_48990_cast_fp16")]; + bool var_48992_interleave_0 = const()[name = string("op_48992_interleave_0"), val = bool(false)]; + tensor var_48992_cast_fp16 = concat(axis = var_47515, interleave = var_48992_interleave_0, values = (var_48940_cast_fp16, var_48942_cast_fp16, var_48944_cast_fp16, var_48946_cast_fp16))[name = string("op_48992_cast_fp16")]; + bool var_48994_interleave_0 = const()[name = string("op_48994_interleave_0"), val = bool(false)]; + tensor var_48994_cast_fp16 = concat(axis = var_47515, interleave = var_48994_interleave_0, values = (var_48948_cast_fp16, var_48950_cast_fp16, var_48952_cast_fp16, var_48954_cast_fp16))[name = string("op_48994_cast_fp16")]; + bool input_249_interleave_0 = const()[name = string("input_249_interleave_0"), val = bool(false)]; + tensor input_249_cast_fp16 = concat(axis = var_47540, interleave = input_249_interleave_0, values = (var_48956_cast_fp16, var_48958_cast_fp16, var_48960_cast_fp16, var_48962_cast_fp16, var_48964_cast_fp16, var_48966_cast_fp16, var_48968_cast_fp16, var_48970_cast_fp16, var_48972_cast_fp16, var_48974_cast_fp16, var_48976_cast_fp16, var_48978_cast_fp16, var_48980_cast_fp16, var_48982_cast_fp16, var_48984_cast_fp16, var_48986_cast_fp16, var_48988_cast_fp16, var_48990_cast_fp16, var_48992_cast_fp16, var_48994_cast_fp16))[name = string("input_249_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244456960)))]; + tensor layers_31_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247733824)))]; + tensor obj_cast_fp16 = conv(bias = layers_31_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_31_self_attn_o_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = obj_cast_fp16)[name = string("inputs_127_cast_fp16")]; + tensor out_127_axes_0 = const()[name = string("out_127_axes_0"), val = tensor([1])]; + fp16 var_49013_to_fp16 = const()[name = string("op_49013_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_49013_to_fp16, x = inputs_127_cast_fp16)[name = string("out_127_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = string("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247736448)))]; + tensor input_251_beta_0_to_fp16 = const()[name = string("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247739072)))]; + fp16 input_251_epsilon_0_to_fp16 = const()[name = string("input_251_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = string("input_251_cast_fp16")]; + string input_253_pad_type_0 = const()[name = string("input_253_pad_type_0"), val = string("valid")]; + tensor input_253_strides_0 = const()[name = string("input_253_strides_0"), val = tensor([1, 1])]; + tensor input_253_pad_0 = const()[name = string("input_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_253_dilations_0 = const()[name = string("input_253_dilations_0"), val = tensor([1, 1])]; + int32 input_253_groups_0 = const()[name = string("input_253_groups_0"), val = int32(1)]; + tensor layers_31_fc1_weight_to_fp16 = const()[name = string("layers_31_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247741696)))]; + tensor layers_31_fc1_bias_to_fp16 = const()[name = string("layers_31_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260848960)))]; + tensor input_253_cast_fp16 = conv(bias = layers_31_fc1_bias_to_fp16, dilations = input_253_dilations_0, groups = input_253_groups_0, pad = input_253_pad_0, pad_type = input_253_pad_type_0, strides = input_253_strides_0, weight = layers_31_fc1_weight_to_fp16, x = input_251_cast_fp16)[name = string("input_253_cast_fp16")]; + string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("EXACT")]; + tensor input_255_cast_fp16 = gelu(mode = input_255_mode_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_31_fc2_weight_to_fp16 = const()[name = string("layers_31_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260859264)))]; + tensor layers_31_fc2_bias_to_fp16 = const()[name = string("layers_31_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273966528)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_31_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_31_fc2_weight_to_fp16, x = input_255_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_127_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_49051_to_fp16 = const()[name = string("op_49051_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_49051_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273969152)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273971776)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_49075_pad_type_0 = const()[name = string("op_49075_pad_type_0"), val = string("valid")]; + tensor var_49075_strides_0 = const()[name = string("op_49075_strides_0"), val = tensor([1, 1])]; + tensor var_49075_pad_0 = const()[name = string("op_49075_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49075_dilations_0 = const()[name = string("op_49075_dilations_0"), val = tensor([1, 1])]; + int32 var_49075_groups_0 = const()[name = string("op_49075_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273974400)))]; + tensor var_49075_cast_fp16 = conv(dilations = var_49075_dilations_0, groups = var_49075_groups_0, pad = var_49075_pad_0, pad_type = var_49075_pad_type_0, strides = var_49075_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49075_cast_fp16")]; + string var_49082_pad_type_0 = const()[name = string("op_49082_pad_type_0"), val = string("valid")]; + tensor var_49082_strides_0 = const()[name = string("op_49082_strides_0"), val = tensor([1, 1])]; + tensor var_49082_pad_0 = const()[name = string("op_49082_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49082_dilations_0 = const()[name = string("op_49082_dilations_0"), val = tensor([1, 1])]; + int32 var_49082_groups_0 = const()[name = string("op_49082_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1277251264)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280528128)))]; + tensor var_49082_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_49082_dilations_0, groups = var_49082_groups_0, pad = var_49082_pad_0, pad_type = var_49082_pad_type_0, strides = var_49082_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49082_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280530752)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1283807616)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1287084480)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_49112 = const()[name = string("op_49112"), val = int32(0)]; + bool input_259_interleave_0 = const()[name = string("input_259_interleave_0"), val = bool(false)]; + tensor input_259_cast_fp16 = concat(axis = var_49112, interleave = input_259_interleave_0, values = (var_49075_cast_fp16, k_cast_fp16))[name = string("input_259_cast_fp16")]; + int32 var_49115 = const()[name = string("op_49115"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_49115, interleave = input_interleave_0, values = (var_49082_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_49122_pad_0 = const()[name = string("op_49122_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_49122_mode_0 = const()[name = string("op_49122_mode_0"), val = string("constant")]; + fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_33_to_fp16, mode = var_49122_mode_0, pad = var_49122_pad_0, x = input_259_cast_fp16)[name = string("op_49122_cast_fp16")]; + tensor var_49128_pad_0 = const()[name = string("op_49128_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_49128_mode_0 = const()[name = string("op_49128_mode_0"), val = string("constant")]; + fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_34_to_fp16, mode = var_49128_mode_0, pad = var_49128_pad_0, x = input_cast_fp16)[name = string("op_49128_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/weights/weight.bin b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..d381c5cf6c7d4bbd79f9efc62d1372cd5cf6e67a --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b43a5d9e21e95067e0af8cf4b8fcbd16cc8e6f99993084f5e67cdf81bde16e79 +size 1287087104 diff --git a/distil-whisper_distil-large-v3_turbo/LICENSE_NOTICE.txt b/distil-whisper_distil-large-v3_turbo/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..3ba3246801c85f92f79ac029f59b94e7fb646f85 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0980462db89a546e1e90888ea38e0a5ddf1f1fec84608802cdbb12f8a5cc7215 +size 243 diff --git a/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/coremldata.bin b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..780171e73cd57a772ec0457470f0c8b86f4c73cd --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6475c6649047ce609e3fe84b2525843c03342820662404540baf28146c174014 +size 329 diff --git a/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/metadata.json b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..65be90aad1d0e5f73a1f50b19705ccad3c0da822 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/model.mil b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..6cf57d7dbf15af35e56636caf15aff60353296f0 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/weights/weight.bin b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..2ae170c9000db89326cc2600450001654bb10f7f --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:009d9fb8f6b589accfa08cebf1c712ef07c3405229ce3cfb3a57ee033c9d8a49 +size 373376 diff --git a/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/analytics/coremldata.bin b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..99e84c4a1fe76264a47fffd68e9bb0b795e5458f --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77cb1b565a336e7fc01586698e50aa32d9a2a8f1ca5c439172564f4af0515f5d +size 243 diff --git a/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/coremldata.bin b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..bd5b35d879f406171df9765f277e32012e734dac --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a5e6f62b5ae897c8f846e22cacbe7d4f7d6bdbeb5f46366e2387f1082676b62 +size 754 diff --git a/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/metadata.json b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..d916f4d101c50c85e7cd58cb4f3baca787cacd86 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51866)", + "shortDescription" : "", + "shape" : "[1, 1, 51866]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 2560 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 2560, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 2560 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 2560, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 4, + "Ios18.mul" : 8, + "Ios18.matmul" : 8, + "Ios18.batchNorm" : 7, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 15, + "Ios18.layerNorm" : 7, + "Ios18.reshape" : 16, + "Ios18.linear" : 1, + "Ios18.conv" : 16, + "Ios18.gelu" : 2, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 44, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 2 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[2, 1280, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/model.mil b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..5be4fa08fa43ad78d542009812ac7702c5a5d9fc --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,529 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_22_axis_0 = const()[name = string("op_22_axis_0"), val = int32(0)]; + int32 var_22_batch_dims_0 = const()[name = string("op_22_batch_dims_0"), val = int32(0)]; + bool var_22_validate_indices_0 = const()[name = string("op_22_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_22_cast_fp16 = gather(axis = var_22_axis_0, batch_dims = var_22_batch_dims_0, indices = input_ids, validate_indices = var_22_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_22_cast_fp16")]; + int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; + int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; + bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132777088)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_43")]; + tensor var_26_cast_fp16_cast_uint16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_26_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_26_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_22_cast_fp16, y = var_26_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_40_axes_0 = const()[name = string("op_40_axes_0"), val = tensor([2])]; + tensor var_40_cast_fp16 = expand_dims(axes = var_40_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_40_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_40_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1])]; + int32 var_45_axis_0 = const()[name = string("op_45_axis_0"), val = int32(0)]; + tensor var_45_cast_fp16_0, tensor var_45_cast_fp16_1 = split(axis = var_45_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_45_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1])]; + int32 var_50_axis_0 = const()[name = string("op_50_axis_0"), val = int32(0)]; + tensor var_50_cast_fp16_0, tensor var_50_cast_fp16_1 = split(axis = var_50_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_50_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_68 = const()[name = string("op_68"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_93_to_fp16 = const()[name = string("op_93_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_93_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133924032)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133926656)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133929280)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133931904)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133934528)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137211392)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137214016)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140490880)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143767744)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_128_axes_0 = const()[name = string("op_128_axes_0"), val = tensor([1])]; + tensor var_128_cast_fp16 = expand_dims(axes = var_128_axes_0, x = kv_cache_update_mask)[name = string("op_128_cast_fp16")]; + tensor var_129_axes_0 = const()[name = string("op_129_axes_0"), val = tensor([2])]; + tensor var_129_cast_fp16 = expand_dims(axes = var_129_axes_0, x = var_128_cast_fp16)[name = string("op_129_cast_fp16")]; + tensor var_131_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_129_cast_fp16)[name = string("op_131_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_45_cast_fp16_0, y = var_131_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_133_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_129_cast_fp16)[name = string("op_133_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_50_cast_fp16_0, y = var_133_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_136 = const()[name = string("op_136"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_136, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_138_to_fp16 = const()[name = string("op_138_to_fp16"), val = fp16(0x1p-3)]; + tensor var_139_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_138_to_fp16)[name = string("op_139_cast_fp16")]; + tensor var_140 = const()[name = string("op_140"), val = tensor([1, 20, 64, -1])]; + tensor var_141_cast_fp16 = reshape(shape = var_140, x = key_1_cast_fp16)[name = string("op_141_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_139_cast_fp16, y = var_141_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_145_axes_0 = const()[name = string("op_145_axes_0"), val = tensor([1])]; + tensor var_145_cast_fp16 = expand_dims(axes = var_145_axes_0, x = decoder_key_padding_mask)[name = string("op_145_cast_fp16")]; + tensor var_146_axes_0 = const()[name = string("op_146_axes_0"), val = tensor([2])]; + tensor var_146_cast_fp16 = expand_dims(axes = var_146_axes_0, x = var_145_cast_fp16)[name = string("op_146_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_146_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_149_cast_fp16 = softmax(axis = var_68, x = mh_w_3_cast_fp16)[name = string("op_149_cast_fp16")]; + tensor var_150 = const()[name = string("op_150"), val = tensor([1, 20, 64, -1])]; + tensor var_151_cast_fp16 = reshape(shape = var_150, x = value_1_cast_fp16)[name = string("op_151_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_151_cast_fp16, y = var_149_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_154 = const()[name = string("op_154"), val = tensor([1, 1280, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_154, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143770368)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147047232)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_176_to_fp16 = const()[name = string("op_176_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_176_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147049856)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147052480)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147055104)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331968)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_196 = const()[name = string("op_196"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_196, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_198_to_fp16 = const()[name = string("op_198_to_fp16"), val = fp16(0x1p-3)]; + tensor var_199_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_198_to_fp16)[name = string("op_199_cast_fp16")]; + tensor var_200 = const()[name = string("op_200"), val = tensor([1, 20, 64, -1])]; + tensor var_201_cast_fp16 = reshape(shape = var_200, x = obj_17_cast_fp16)[name = string("op_201_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_199_cast_fp16, y = var_201_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_205_axes_0 = const()[name = string("op_205_axes_0"), val = tensor([1])]; + tensor var_205_cast_fp16 = expand_dims(axes = var_205_axes_0, x = read_state_4)[name = string("op_205_cast_fp16")]; + tensor var_206_axes_0 = const()[name = string("op_206_axes_0"), val = tensor([2])]; + tensor var_206_cast_fp16 = expand_dims(axes = var_206_axes_0, x = var_205_cast_fp16)[name = string("op_206_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_206_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_68, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_210 = const()[name = string("op_210"), val = tensor([1, 20, 64, -1])]; + tensor var_211_cast_fp16 = reshape(shape = var_210, x = obj_19_cast_fp16)[name = string("op_211_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_211_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_214 = const()[name = string("op_214"), val = tensor([1, 1280, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_214, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334592)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_232_to_fp16 = const()[name = string("op_232_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_232_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614080)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616704)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153619328)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166726592)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166736896)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179844160)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_277 = const()[name = string("op_277"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_302_to_fp16 = const()[name = string("op_302_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_302_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179846784)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179849408)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179852032)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128896)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183131520)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186408384)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189685248)))]; + tensor current_value_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_340_cast_fp16 = mul(x = current_key_cast_fp16, y = var_129_cast_fp16)[name = string("op_340_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_45_cast_fp16_1, y = var_340_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_342_cast_fp16 = mul(x = current_value_cast_fp16, y = var_129_cast_fp16)[name = string("op_342_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_50_cast_fp16_1, y = var_342_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_345 = const()[name = string("op_345"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_345, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_347_to_fp16 = const()[name = string("op_347_to_fp16"), val = fp16(0x1p-3)]; + tensor var_348_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_347_to_fp16)[name = string("op_348_cast_fp16")]; + tensor var_349 = const()[name = string("op_349"), val = tensor([1, 20, 64, -1])]; + tensor var_350_cast_fp16 = reshape(shape = var_349, x = key_cast_fp16)[name = string("op_350_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_348_cast_fp16, y = var_350_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_146_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_358_cast_fp16 = softmax(axis = var_277, x = mh_w_11_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor var_359 = const()[name = string("op_359"), val = tensor([1, 20, 64, -1])]; + tensor var_360_cast_fp16 = reshape(shape = var_359, x = value_cast_fp16)[name = string("op_360_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_360_cast_fp16, y = var_358_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_363 = const()[name = string("op_363"), val = tensor([1, 1280, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_363, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189687872)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192964736)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_385_to_fp16 = const()[name = string("op_385_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_385_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192967360)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192969984)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192972608)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196249472)))]; + tensor query_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_405 = const()[name = string("op_405"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_405, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_407_to_fp16 = const()[name = string("op_407_to_fp16"), val = fp16(0x1p-3)]; + tensor var_408_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_407_to_fp16)[name = string("op_408_cast_fp16")]; + tensor var_409 = const()[name = string("op_409"), val = tensor([1, 20, 64, -1])]; + tensor var_410_cast_fp16 = reshape(shape = var_409, x = obj_35_cast_fp16)[name = string("op_410_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_408_cast_fp16, y = var_410_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_206_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_277, x = mh_w_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_419 = const()[name = string("op_419"), val = tensor([1, 20, 64, -1])]; + tensor var_420_cast_fp16 = reshape(shape = var_419, x = obj_37_cast_fp16)[name = string("op_420_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_420_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_423 = const()[name = string("op_423"), val = tensor([1, 1280, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_423, x = attn_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196252096)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199528960)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_444_to_fp16 = const()[name = string("op_444_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_444_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199531584)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199534208)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199536832)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212644096)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_17_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212654400)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225761664)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_487_to_fp16 = const()[name = string("op_487_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_487_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225764288)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225766912)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_498_axes_0 = const()[name = string("op_498_axes_0"), val = tensor([2])]; + tensor var_498_cast_fp16 = squeeze(axes = var_498_axes_0, x = hidden_states_cast_fp16)[name = string("op_498_cast_fp16")]; + tensor var_501_perm_0 = const()[name = string("op_501_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225769536)))]; + tensor var_501_cast_fp16 = transpose(perm = var_501_perm_0, x = var_498_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_501_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_505 = const()[name = string("op_505"), val = int32(1)]; + bool obj_45_interleave_0 = const()[name = string("obj_45_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_505, interleave = obj_45_interleave_0, values = (current_key_1_cast_fp16, current_key_cast_fp16))[name = string("obj_45_cast_fp16")]; + int32 var_508 = const()[name = string("op_508"), val = int32(1)]; + bool obj_47_interleave_0 = const()[name = string("obj_47_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_508, interleave = obj_47_interleave_0, values = (current_value_1_cast_fp16, current_value_cast_fp16))[name = string("obj_47_cast_fp16")]; + tensor var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = obj_41_cast_fp16)[name = string("op_519_cast_fp16")]; + tensor var_522_begin_0 = const()[name = string("op_522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_522_end_0 = const()[name = string("op_522_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_522_end_mask_0 = const()[name = string("op_522_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_522_squeeze_mask_0 = const()[name = string("op_522_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_522_cast_fp16 = slice_by_index(begin = var_522_begin_0, end = var_522_end_0, end_mask = var_522_end_mask_0, squeeze_mask = var_522_squeeze_mask_0, x = var_519_cast_fp16)[name = string("op_522_cast_fp16")]; + tensor var_537_begin_0 = const()[name = string("op_537_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_537_end_0 = const()[name = string("op_537_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_537_end_mask_0 = const()[name = string("op_537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_537_cast_fp16 = slice_by_index(begin = var_537_begin_0, end = var_537_end_0, end_mask = var_537_end_mask_0, x = obj_41_cast_fp16)[name = string("op_537_cast_fp16")]; + tensor var_540_begin_0 = const()[name = string("op_540_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_540_end_0 = const()[name = string("op_540_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_540_end_mask_0 = const()[name = string("op_540_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_540_squeeze_mask_0 = const()[name = string("op_540_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_540_cast_fp16 = slice_by_index(begin = var_540_begin_0, end = var_540_end_0, end_mask = var_540_end_mask_0, squeeze_mask = var_540_squeeze_mask_0, x = var_537_cast_fp16)[name = string("op_540_cast_fp16")]; + tensor var_555_begin_0 = const()[name = string("op_555_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_555_end_0 = const()[name = string("op_555_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_555_end_mask_0 = const()[name = string("op_555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_555_cast_fp16 = slice_by_index(begin = var_555_begin_0, end = var_555_end_0, end_mask = var_555_end_mask_0, x = obj_41_cast_fp16)[name = string("op_555_cast_fp16")]; + tensor var_558_begin_0 = const()[name = string("op_558_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_558_end_0 = const()[name = string("op_558_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_558_end_mask_0 = const()[name = string("op_558_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_558_squeeze_mask_0 = const()[name = string("op_558_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_558_cast_fp16 = slice_by_index(begin = var_558_begin_0, end = var_558_end_0, end_mask = var_558_end_mask_0, squeeze_mask = var_558_squeeze_mask_0, x = var_555_cast_fp16)[name = string("op_558_cast_fp16")]; + tensor var_573_begin_0 = const()[name = string("op_573_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_573_end_0 = const()[name = string("op_573_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_573_end_mask_0 = const()[name = string("op_573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_573_cast_fp16 = slice_by_index(begin = var_573_begin_0, end = var_573_end_0, end_mask = var_573_end_mask_0, x = obj_41_cast_fp16)[name = string("op_573_cast_fp16")]; + tensor var_576_begin_0 = const()[name = string("op_576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_576_end_0 = const()[name = string("op_576_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_576_end_mask_0 = const()[name = string("op_576_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_576_squeeze_mask_0 = const()[name = string("op_576_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_576_cast_fp16 = slice_by_index(begin = var_576_begin_0, end = var_576_end_0, end_mask = var_576_end_mask_0, squeeze_mask = var_576_squeeze_mask_0, x = var_573_cast_fp16)[name = string("op_576_cast_fp16")]; + tensor var_591_begin_0 = const()[name = string("op_591_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_591_end_0 = const()[name = string("op_591_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_591_end_mask_0 = const()[name = string("op_591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_591_cast_fp16 = slice_by_index(begin = var_591_begin_0, end = var_591_end_0, end_mask = var_591_end_mask_0, x = obj_41_cast_fp16)[name = string("op_591_cast_fp16")]; + tensor var_594_begin_0 = const()[name = string("op_594_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_594_end_0 = const()[name = string("op_594_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_594_end_mask_0 = const()[name = string("op_594_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_594_squeeze_mask_0 = const()[name = string("op_594_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_594_cast_fp16 = slice_by_index(begin = var_594_begin_0, end = var_594_end_0, end_mask = var_594_end_mask_0, squeeze_mask = var_594_squeeze_mask_0, x = var_591_cast_fp16)[name = string("op_594_cast_fp16")]; + tensor var_609_begin_0 = const()[name = string("op_609_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_609_end_0 = const()[name = string("op_609_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_609_end_mask_0 = const()[name = string("op_609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_609_cast_fp16 = slice_by_index(begin = var_609_begin_0, end = var_609_end_0, end_mask = var_609_end_mask_0, x = obj_41_cast_fp16)[name = string("op_609_cast_fp16")]; + tensor var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_612_squeeze_mask_0 = const()[name = string("op_612_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, squeeze_mask = var_612_squeeze_mask_0, x = var_609_cast_fp16)[name = string("op_612_cast_fp16")]; + tensor var_627_begin_0 = const()[name = string("op_627_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_627_end_0 = const()[name = string("op_627_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_627_end_mask_0 = const()[name = string("op_627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_627_cast_fp16 = slice_by_index(begin = var_627_begin_0, end = var_627_end_0, end_mask = var_627_end_mask_0, x = obj_41_cast_fp16)[name = string("op_627_cast_fp16")]; + tensor var_630_begin_0 = const()[name = string("op_630_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_630_end_0 = const()[name = string("op_630_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_630_end_mask_0 = const()[name = string("op_630_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_630_squeeze_mask_0 = const()[name = string("op_630_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_630_cast_fp16 = slice_by_index(begin = var_630_begin_0, end = var_630_end_0, end_mask = var_630_end_mask_0, squeeze_mask = var_630_squeeze_mask_0, x = var_627_cast_fp16)[name = string("op_630_cast_fp16")]; + tensor var_645_begin_0 = const()[name = string("op_645_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_645_end_0 = const()[name = string("op_645_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_645_end_mask_0 = const()[name = string("op_645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = obj_41_cast_fp16)[name = string("op_645_cast_fp16")]; + tensor var_648_begin_0 = const()[name = string("op_648_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_648_end_0 = const()[name = string("op_648_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_648_end_mask_0 = const()[name = string("op_648_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_648_squeeze_mask_0 = const()[name = string("op_648_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_648_cast_fp16 = slice_by_index(begin = var_648_begin_0, end = var_648_end_0, end_mask = var_648_end_mask_0, squeeze_mask = var_648_squeeze_mask_0, x = var_645_cast_fp16)[name = string("op_648_cast_fp16")]; + tensor var_663_begin_0 = const()[name = string("op_663_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_663_end_0 = const()[name = string("op_663_end_0"), val = tensor([1, 9, 1, 1536])]; + tensor var_663_end_mask_0 = const()[name = string("op_663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = obj_41_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_666_begin_0 = const()[name = string("op_666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_666_end_0 = const()[name = string("op_666_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_666_end_mask_0 = const()[name = string("op_666_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_666_squeeze_mask_0 = const()[name = string("op_666_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_666_cast_fp16 = slice_by_index(begin = var_666_begin_0, end = var_666_end_0, end_mask = var_666_end_mask_0, squeeze_mask = var_666_squeeze_mask_0, x = var_663_cast_fp16)[name = string("op_666_cast_fp16")]; + tensor var_681_begin_0 = const()[name = string("op_681_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_681_end_0 = const()[name = string("op_681_end_0"), val = tensor([1, 10, 1, 1536])]; + tensor var_681_end_mask_0 = const()[name = string("op_681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_681_cast_fp16 = slice_by_index(begin = var_681_begin_0, end = var_681_end_0, end_mask = var_681_end_mask_0, x = obj_41_cast_fp16)[name = string("op_681_cast_fp16")]; + tensor var_684_begin_0 = const()[name = string("op_684_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_684_end_0 = const()[name = string("op_684_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_684_end_mask_0 = const()[name = string("op_684_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_684_squeeze_mask_0 = const()[name = string("op_684_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_684_cast_fp16 = slice_by_index(begin = var_684_begin_0, end = var_684_end_0, end_mask = var_684_end_mask_0, squeeze_mask = var_684_squeeze_mask_0, x = var_681_cast_fp16)[name = string("op_684_cast_fp16")]; + tensor var_699_begin_0 = const()[name = string("op_699_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_699_end_0 = const()[name = string("op_699_end_0"), val = tensor([1, 11, 1, 1536])]; + tensor var_699_end_mask_0 = const()[name = string("op_699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_699_cast_fp16 = slice_by_index(begin = var_699_begin_0, end = var_699_end_0, end_mask = var_699_end_mask_0, x = obj_41_cast_fp16)[name = string("op_699_cast_fp16")]; + tensor var_702_begin_0 = const()[name = string("op_702_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_702_end_0 = const()[name = string("op_702_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_702_end_mask_0 = const()[name = string("op_702_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_702_squeeze_mask_0 = const()[name = string("op_702_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_702_cast_fp16 = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, squeeze_mask = var_702_squeeze_mask_0, x = var_699_cast_fp16)[name = string("op_702_cast_fp16")]; + tensor var_717_begin_0 = const()[name = string("op_717_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_717_end_0 = const()[name = string("op_717_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_717_end_mask_0 = const()[name = string("op_717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_717_cast_fp16 = slice_by_index(begin = var_717_begin_0, end = var_717_end_0, end_mask = var_717_end_mask_0, x = obj_41_cast_fp16)[name = string("op_717_cast_fp16")]; + tensor var_720_begin_0 = const()[name = string("op_720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_720_end_0 = const()[name = string("op_720_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_720_end_mask_0 = const()[name = string("op_720_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_720_squeeze_mask_0 = const()[name = string("op_720_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_720_cast_fp16 = slice_by_index(begin = var_720_begin_0, end = var_720_end_0, end_mask = var_720_end_mask_0, squeeze_mask = var_720_squeeze_mask_0, x = var_717_cast_fp16)[name = string("op_720_cast_fp16")]; + tensor var_735_begin_0 = const()[name = string("op_735_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_735_end_0 = const()[name = string("op_735_end_0"), val = tensor([1, 13, 1, 1536])]; + tensor var_735_end_mask_0 = const()[name = string("op_735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_735_cast_fp16 = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = obj_41_cast_fp16)[name = string("op_735_cast_fp16")]; + tensor var_738_begin_0 = const()[name = string("op_738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_738_end_0 = const()[name = string("op_738_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_738_end_mask_0 = const()[name = string("op_738_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_738_squeeze_mask_0 = const()[name = string("op_738_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_738_cast_fp16 = slice_by_index(begin = var_738_begin_0, end = var_738_end_0, end_mask = var_738_end_mask_0, squeeze_mask = var_738_squeeze_mask_0, x = var_735_cast_fp16)[name = string("op_738_cast_fp16")]; + tensor var_753_begin_0 = const()[name = string("op_753_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_753_end_0 = const()[name = string("op_753_end_0"), val = tensor([1, 14, 1, 1536])]; + tensor var_753_end_mask_0 = const()[name = string("op_753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = obj_41_cast_fp16)[name = string("op_753_cast_fp16")]; + tensor var_756_begin_0 = const()[name = string("op_756_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_756_end_0 = const()[name = string("op_756_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_756_end_mask_0 = const()[name = string("op_756_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_756_squeeze_mask_0 = const()[name = string("op_756_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_756_cast_fp16 = slice_by_index(begin = var_756_begin_0, end = var_756_end_0, end_mask = var_756_end_mask_0, squeeze_mask = var_756_squeeze_mask_0, x = var_753_cast_fp16)[name = string("op_756_cast_fp16")]; + tensor var_771_begin_0 = const()[name = string("op_771_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_771_end_0 = const()[name = string("op_771_end_0"), val = tensor([1, 15, 1, 1536])]; + tensor var_771_end_mask_0 = const()[name = string("op_771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_771_cast_fp16 = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = obj_41_cast_fp16)[name = string("op_771_cast_fp16")]; + tensor var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_774_squeeze_mask_0 = const()[name = string("op_774_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, squeeze_mask = var_774_squeeze_mask_0, x = var_771_cast_fp16)[name = string("op_774_cast_fp16")]; + tensor var_789_begin_0 = const()[name = string("op_789_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_789_end_0 = const()[name = string("op_789_end_0"), val = tensor([1, 16, 1, 1536])]; + tensor var_789_end_mask_0 = const()[name = string("op_789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_789_cast_fp16 = slice_by_index(begin = var_789_begin_0, end = var_789_end_0, end_mask = var_789_end_mask_0, x = obj_41_cast_fp16)[name = string("op_789_cast_fp16")]; + tensor var_792_begin_0 = const()[name = string("op_792_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_792_end_0 = const()[name = string("op_792_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_792_end_mask_0 = const()[name = string("op_792_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_792_squeeze_mask_0 = const()[name = string("op_792_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_792_cast_fp16 = slice_by_index(begin = var_792_begin_0, end = var_792_end_0, end_mask = var_792_end_mask_0, squeeze_mask = var_792_squeeze_mask_0, x = var_789_cast_fp16)[name = string("op_792_cast_fp16")]; + tensor var_807_begin_0 = const()[name = string("op_807_begin_0"), val = tensor([0, 16, 0, 0])]; + tensor var_807_end_0 = const()[name = string("op_807_end_0"), val = tensor([1, 17, 1, 1536])]; + tensor var_807_end_mask_0 = const()[name = string("op_807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_807_cast_fp16 = slice_by_index(begin = var_807_begin_0, end = var_807_end_0, end_mask = var_807_end_mask_0, x = obj_41_cast_fp16)[name = string("op_807_cast_fp16")]; + tensor var_810_begin_0 = const()[name = string("op_810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_810_end_0 = const()[name = string("op_810_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_810_end_mask_0 = const()[name = string("op_810_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_810_squeeze_mask_0 = const()[name = string("op_810_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_810_cast_fp16 = slice_by_index(begin = var_810_begin_0, end = var_810_end_0, end_mask = var_810_end_mask_0, squeeze_mask = var_810_squeeze_mask_0, x = var_807_cast_fp16)[name = string("op_810_cast_fp16")]; + tensor var_825_begin_0 = const()[name = string("op_825_begin_0"), val = tensor([0, 17, 0, 0])]; + tensor var_825_end_0 = const()[name = string("op_825_end_0"), val = tensor([1, 18, 1, 1536])]; + tensor var_825_end_mask_0 = const()[name = string("op_825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_825_cast_fp16 = slice_by_index(begin = var_825_begin_0, end = var_825_end_0, end_mask = var_825_end_mask_0, x = obj_41_cast_fp16)[name = string("op_825_cast_fp16")]; + tensor var_828_begin_0 = const()[name = string("op_828_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_828_end_0 = const()[name = string("op_828_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_828_end_mask_0 = const()[name = string("op_828_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_828_squeeze_mask_0 = const()[name = string("op_828_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_828_cast_fp16 = slice_by_index(begin = var_828_begin_0, end = var_828_end_0, end_mask = var_828_end_mask_0, squeeze_mask = var_828_squeeze_mask_0, x = var_825_cast_fp16)[name = string("op_828_cast_fp16")]; + tensor var_843_begin_0 = const()[name = string("op_843_begin_0"), val = tensor([0, 18, 0, 0])]; + tensor var_843_end_0 = const()[name = string("op_843_end_0"), val = tensor([1, 19, 1, 1536])]; + tensor var_843_end_mask_0 = const()[name = string("op_843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_843_cast_fp16 = slice_by_index(begin = var_843_begin_0, end = var_843_end_0, end_mask = var_843_end_mask_0, x = obj_41_cast_fp16)[name = string("op_843_cast_fp16")]; + tensor var_846_begin_0 = const()[name = string("op_846_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_846_end_0 = const()[name = string("op_846_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_846_end_mask_0 = const()[name = string("op_846_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_846_squeeze_mask_0 = const()[name = string("op_846_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_846_cast_fp16 = slice_by_index(begin = var_846_begin_0, end = var_846_end_0, end_mask = var_846_end_mask_0, squeeze_mask = var_846_squeeze_mask_0, x = var_843_cast_fp16)[name = string("op_846_cast_fp16")]; + tensor var_861_begin_0 = const()[name = string("op_861_begin_0"), val = tensor([0, 19, 0, 0])]; + tensor var_861_end_0 = const()[name = string("op_861_end_0"), val = tensor([1, 20, 1, 1536])]; + tensor var_861_end_mask_0 = const()[name = string("op_861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_861_cast_fp16 = slice_by_index(begin = var_861_begin_0, end = var_861_end_0, end_mask = var_861_end_mask_0, x = obj_41_cast_fp16)[name = string("op_861_cast_fp16")]; + tensor var_864_begin_0 = const()[name = string("op_864_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_864_end_0 = const()[name = string("op_864_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_864_end_mask_0 = const()[name = string("op_864_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_864_squeeze_mask_0 = const()[name = string("op_864_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_864_cast_fp16 = slice_by_index(begin = var_864_begin_0, end = var_864_end_0, end_mask = var_864_end_mask_0, squeeze_mask = var_864_squeeze_mask_0, x = var_861_cast_fp16)[name = string("op_864_cast_fp16")]; + int32 var_871 = const()[name = string("op_871"), val = int32(1)]; + bool var_872_interleave_0 = const()[name = string("op_872_interleave_0"), val = bool(false)]; + tensor var_872_cast_fp16 = concat(axis = var_871, interleave = var_872_interleave_0, values = (var_522_cast_fp16, var_540_cast_fp16, var_558_cast_fp16, var_576_cast_fp16, var_594_cast_fp16, var_612_cast_fp16, var_630_cast_fp16, var_648_cast_fp16, var_666_cast_fp16, var_684_cast_fp16, var_702_cast_fp16, var_720_cast_fp16, var_738_cast_fp16, var_756_cast_fp16, var_774_cast_fp16, var_792_cast_fp16, var_810_cast_fp16, var_828_cast_fp16, var_846_cast_fp16, var_864_cast_fp16))[name = string("op_872_cast_fp16")]; + bool var_875 = const()[name = string("op_875"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_875, x = var_872_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/weights/weight.bin b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..9929465fd5f57561fddd5514a2227d4c6c31a160 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c1afaacaec2fac64e8867d758742347e10c849fdbf81c8761344b5c56a55b5d +size 225873332 diff --git a/distil-whisper_distil-large-v3_turbo/config.json b/distil-whisper_distil-large-v3_turbo/config.json new file mode 100644 index 0000000000000000000000000000000000000000..35fb9504486c4c3df76f5f84aad0bdffd3de00f6 --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/config.json @@ -0,0 +1 @@ +{"_name_or_path": "./distil-large-v3", "activation_dropout": 0.0, "activation_function": "gelu", "apply_spec_augment": false, "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "classifier_proj_size": 256, "d_model": 1280, "decoder_attention_heads": 20, "decoder_ffn_dim": 5120, "decoder_layerdrop": 0.0, "decoder_layers": 2, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 20, "encoder_ffn_dim": 5120, "encoder_layerdrop": 0.0, "encoder_layers": 32, "eos_token_id": 50257, "init_std": 0.02, "is_encoder_decoder": true, "mask_feature_length": 10, "mask_feature_min_masks": 0, "mask_feature_prob": 0.0, "mask_time_length": 10, "mask_time_min_masks": 2, "mask_time_prob": 0.05, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "median_filter_width": 7, "model_type": "whisper", "num_hidden_layers": 32, "num_mel_bins": 128, "pad_token_id": 50256, "scale_embedding": false, "torch_dtype": "float16", "transformers_version": "4.38.0.dev0", "use_cache": true, "use_weighted_layer_sum": false, "vocab_size": 51866} \ No newline at end of file diff --git a/distil-whisper_distil-large-v3_turbo/generation_config.json b/distil-whisper_distil-large-v3_turbo/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..8d1525665be3225ab0cbd947863a3f75fe5a5b2b --- /dev/null +++ b/distil-whisper_distil-large-v3_turbo/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[7, 0], [10, 17], [12, 18], [13, 12], [16, 1], [17, 14], [19, 11], [21, 4], [24, 1], [25, 6]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50360]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|yue|>": 50358, "<|zh|>": 50260}, "language": "<|en|>", "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50364, "pad_token_id": 50257, "prev_sot_token_id": 50362, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50359, 50360, 50361, 50362, 50363], "task": "transcribe", "task_to_id": {"transcribe": 50360, "translate": 50359}, "transformers_version": "4.38.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-base.en/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-base.en/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..6781142f32ee6e81c445b39773e9f3623f438be6 --- /dev/null +++ b/openai_whisper-base.en/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:384d19c754b6ca6a7ad6dd457406dd9c9de44e43034cbfaf3f343e0278e43ac9 +size 243 diff --git a/openai_whisper-base.en/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-base.en/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..47ddd91836e24b3659e2ac0d879cc15bfc49a74b --- /dev/null +++ b/openai_whisper-base.en/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a536e74da525305d998542cdad99de17f18771834664969738d6fa2ab99fd115 +size 433 diff --git a/openai_whisper-base.en/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-base.en/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..3db663b3e0d7c210978a6267eba2821c4e38906d --- /dev/null +++ b/openai_whisper-base.en/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 512, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 13, + "Ios18.conv" : 50, + "Ios18.gelu" : 8, + "Ios18.concat" : 56, + "Ios16.einsum" : 384, + "Ios18.add" : 13, + "Ios18.softmax" : 192, + "Ios18.sliceByIndex" : 336, + "Ios18.layerNorm" : 13, + "Ios18.transpose" : 6, + "Ios18.mul" : 192 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-base.en/AudioEncoder.mlmodelc/model.mil b/openai_whisper-base.en/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cec64545eb7b80d0ff0100fd8a83a36d68918483 --- /dev/null +++ b/openai_whisper-base.en/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,3351 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_66_pad_type_0 = const()[name = string("op_66_pad_type_0"), val = string("custom")]; + tensor var_66_pad_0 = const()[name = string("op_66_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_66_strides_0 = const()[name = string("op_66_strides_0"), val = tensor([1, 1])]; + tensor var_66_dilations_0 = const()[name = string("op_66_dilations_0"), val = tensor([1, 1])]; + int32 var_66_groups_0 = const()[name = string("op_66_groups_0"), val = int32(1)]; + tensor var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_47_to_fp16 = const()[name = string("op_47_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245888)))]; + tensor var_66_cast_fp16 = conv(bias = var_47_to_fp16, dilations = var_66_dilations_0, groups = var_66_groups_0, pad = var_66_pad_0, pad_type = var_66_pad_type_0, strides = var_66_strides_0, weight = var_41_to_fp16, x = melspectrogram_features)[name = string("op_66_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_66_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_106_pad_type_0 = const()[name = string("op_106_pad_type_0"), val = string("custom")]; + tensor var_106_pad_0 = const()[name = string("op_106_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_106_strides_0 = const()[name = string("op_106_strides_0"), val = tensor([2, 2])]; + tensor var_106_dilations_0 = const()[name = string("op_106_dilations_0"), val = tensor([1, 1])]; + int32 var_106_groups_0 = const()[name = string("op_106_groups_0"), val = int32(1)]; + tensor var_81_to_fp16 = const()[name = string("op_81_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246976)))]; + tensor var_87_to_fp16 = const()[name = string("op_87_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1819904)))]; + tensor var_106_cast_fp16 = conv(bias = var_87_to_fp16, dilations = var_106_dilations_0, groups = var_106_groups_0, pad = var_106_pad_0, pad_type = var_106_pad_type_0, strides = var_106_strides_0, weight = var_81_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_106_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_106_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_124_to_fp16 = const()[name = string("op_124_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1820992)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_124_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_134 = const()[name = string("op_134"), val = int32(3)]; + int32 var_147 = const()[name = string("op_147"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_164_to_fp16 = const()[name = string("op_164_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_164_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3357056)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3358144)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3359232)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3360320)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3361408)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3885760)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3886848)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4411200)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4935552)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_202_begin_0 = const()[name = string("op_202_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_202_end_0 = const()[name = string("op_202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_202_end_mask_0 = const()[name = string("op_202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_202_cast_fp16 = slice_by_index(begin = var_202_begin_0, end = var_202_end_0, end_mask = var_202_end_mask_0, x = query_1_cast_fp16)[name = string("op_202_cast_fp16")]; + tensor var_206_begin_0 = const()[name = string("op_206_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_206_end_0 = const()[name = string("op_206_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_206_end_mask_0 = const()[name = string("op_206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_206_cast_fp16 = slice_by_index(begin = var_206_begin_0, end = var_206_end_0, end_mask = var_206_end_mask_0, x = query_1_cast_fp16)[name = string("op_206_cast_fp16")]; + tensor var_210_begin_0 = const()[name = string("op_210_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_210_end_0 = const()[name = string("op_210_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_210_end_mask_0 = const()[name = string("op_210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_210_cast_fp16 = slice_by_index(begin = var_210_begin_0, end = var_210_end_0, end_mask = var_210_end_mask_0, x = query_1_cast_fp16)[name = string("op_210_cast_fp16")]; + tensor var_214_begin_0 = const()[name = string("op_214_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_214_end_0 = const()[name = string("op_214_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_214_end_mask_0 = const()[name = string("op_214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_214_cast_fp16 = slice_by_index(begin = var_214_begin_0, end = var_214_end_0, end_mask = var_214_end_mask_0, x = query_1_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor var_218_begin_0 = const()[name = string("op_218_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_218_end_0 = const()[name = string("op_218_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_218_end_mask_0 = const()[name = string("op_218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_218_cast_fp16 = slice_by_index(begin = var_218_begin_0, end = var_218_end_0, end_mask = var_218_end_mask_0, x = query_1_cast_fp16)[name = string("op_218_cast_fp16")]; + tensor var_222_begin_0 = const()[name = string("op_222_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_222_end_0 = const()[name = string("op_222_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_222_end_mask_0 = const()[name = string("op_222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_222_cast_fp16 = slice_by_index(begin = var_222_begin_0, end = var_222_end_0, end_mask = var_222_end_mask_0, x = query_1_cast_fp16)[name = string("op_222_cast_fp16")]; + tensor var_226_begin_0 = const()[name = string("op_226_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_226_end_0 = const()[name = string("op_226_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_226_end_mask_0 = const()[name = string("op_226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_226_cast_fp16 = slice_by_index(begin = var_226_begin_0, end = var_226_end_0, end_mask = var_226_end_mask_0, x = query_1_cast_fp16)[name = string("op_226_cast_fp16")]; + tensor var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = string("op_230_cast_fp16")]; + tensor var_239_begin_0 = const()[name = string("op_239_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_239_end_0 = const()[name = string("op_239_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_239_end_mask_0 = const()[name = string("op_239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_239_cast_fp16 = slice_by_index(begin = var_239_begin_0, end = var_239_end_0, end_mask = var_239_end_mask_0, x = var_202_cast_fp16)[name = string("op_239_cast_fp16")]; + tensor var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = var_202_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor var_253_begin_0 = const()[name = string("op_253_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_253_end_0 = const()[name = string("op_253_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_253_end_mask_0 = const()[name = string("op_253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_253_cast_fp16 = slice_by_index(begin = var_253_begin_0, end = var_253_end_0, end_mask = var_253_end_mask_0, x = var_202_cast_fp16)[name = string("op_253_cast_fp16")]; + tensor var_260_begin_0 = const()[name = string("op_260_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_260_end_0 = const()[name = string("op_260_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_260_end_mask_0 = const()[name = string("op_260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_260_cast_fp16 = slice_by_index(begin = var_260_begin_0, end = var_260_end_0, end_mask = var_260_end_mask_0, x = var_202_cast_fp16)[name = string("op_260_cast_fp16")]; + tensor var_267_begin_0 = const()[name = string("op_267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_267_end_0 = const()[name = string("op_267_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_267_end_mask_0 = const()[name = string("op_267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_267_cast_fp16 = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = var_206_cast_fp16)[name = string("op_267_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_206_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_281_begin_0 = const()[name = string("op_281_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_281_end_0 = const()[name = string("op_281_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_281_end_mask_0 = const()[name = string("op_281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_281_cast_fp16 = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_206_cast_fp16)[name = string("op_281_cast_fp16")]; + tensor var_288_begin_0 = const()[name = string("op_288_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_288_end_0 = const()[name = string("op_288_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_288_end_mask_0 = const()[name = string("op_288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_288_cast_fp16 = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_206_cast_fp16)[name = string("op_288_cast_fp16")]; + tensor var_295_begin_0 = const()[name = string("op_295_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_295_end_0 = const()[name = string("op_295_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_295_end_mask_0 = const()[name = string("op_295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_295_cast_fp16 = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_210_cast_fp16)[name = string("op_295_cast_fp16")]; + tensor var_302_begin_0 = const()[name = string("op_302_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_302_end_0 = const()[name = string("op_302_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_302_end_mask_0 = const()[name = string("op_302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = var_210_cast_fp16)[name = string("op_302_cast_fp16")]; + tensor var_309_begin_0 = const()[name = string("op_309_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_309_end_0 = const()[name = string("op_309_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_309_end_mask_0 = const()[name = string("op_309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_309_cast_fp16 = slice_by_index(begin = var_309_begin_0, end = var_309_end_0, end_mask = var_309_end_mask_0, x = var_210_cast_fp16)[name = string("op_309_cast_fp16")]; + tensor var_316_begin_0 = const()[name = string("op_316_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_316_end_0 = const()[name = string("op_316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_316_end_mask_0 = const()[name = string("op_316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_316_cast_fp16 = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = var_210_cast_fp16)[name = string("op_316_cast_fp16")]; + tensor var_323_begin_0 = const()[name = string("op_323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_323_end_0 = const()[name = string("op_323_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_323_end_mask_0 = const()[name = string("op_323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_323_cast_fp16 = slice_by_index(begin = var_323_begin_0, end = var_323_end_0, end_mask = var_323_end_mask_0, x = var_214_cast_fp16)[name = string("op_323_cast_fp16")]; + tensor var_330_begin_0 = const()[name = string("op_330_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_330_end_0 = const()[name = string("op_330_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_330_end_mask_0 = const()[name = string("op_330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_330_cast_fp16 = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = var_214_cast_fp16)[name = string("op_330_cast_fp16")]; + tensor var_337_begin_0 = const()[name = string("op_337_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_337_end_0 = const()[name = string("op_337_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_337_end_mask_0 = const()[name = string("op_337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_337_cast_fp16 = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = var_214_cast_fp16)[name = string("op_337_cast_fp16")]; + tensor var_344_begin_0 = const()[name = string("op_344_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_344_end_0 = const()[name = string("op_344_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_344_end_mask_0 = const()[name = string("op_344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_344_cast_fp16 = slice_by_index(begin = var_344_begin_0, end = var_344_end_0, end_mask = var_344_end_mask_0, x = var_214_cast_fp16)[name = string("op_344_cast_fp16")]; + tensor var_351_begin_0 = const()[name = string("op_351_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_351_end_0 = const()[name = string("op_351_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_351_end_mask_0 = const()[name = string("op_351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_351_cast_fp16 = slice_by_index(begin = var_351_begin_0, end = var_351_end_0, end_mask = var_351_end_mask_0, x = var_218_cast_fp16)[name = string("op_351_cast_fp16")]; + tensor var_358_begin_0 = const()[name = string("op_358_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_358_end_0 = const()[name = string("op_358_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_358_end_mask_0 = const()[name = string("op_358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_358_cast_fp16 = slice_by_index(begin = var_358_begin_0, end = var_358_end_0, end_mask = var_358_end_mask_0, x = var_218_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor var_365_begin_0 = const()[name = string("op_365_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_365_end_0 = const()[name = string("op_365_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_365_end_mask_0 = const()[name = string("op_365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_365_cast_fp16 = slice_by_index(begin = var_365_begin_0, end = var_365_end_0, end_mask = var_365_end_mask_0, x = var_218_cast_fp16)[name = string("op_365_cast_fp16")]; + tensor var_372_begin_0 = const()[name = string("op_372_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_372_end_0 = const()[name = string("op_372_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_372_end_mask_0 = const()[name = string("op_372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_372_cast_fp16 = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, x = var_218_cast_fp16)[name = string("op_372_cast_fp16")]; + tensor var_379_begin_0 = const()[name = string("op_379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_379_end_0 = const()[name = string("op_379_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_379_end_mask_0 = const()[name = string("op_379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_379_cast_fp16 = slice_by_index(begin = var_379_begin_0, end = var_379_end_0, end_mask = var_379_end_mask_0, x = var_222_cast_fp16)[name = string("op_379_cast_fp16")]; + tensor var_386_begin_0 = const()[name = string("op_386_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_386_end_0 = const()[name = string("op_386_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_386_end_mask_0 = const()[name = string("op_386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_386_cast_fp16 = slice_by_index(begin = var_386_begin_0, end = var_386_end_0, end_mask = var_386_end_mask_0, x = var_222_cast_fp16)[name = string("op_386_cast_fp16")]; + tensor var_393_begin_0 = const()[name = string("op_393_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_393_end_0 = const()[name = string("op_393_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_393_end_mask_0 = const()[name = string("op_393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_393_cast_fp16 = slice_by_index(begin = var_393_begin_0, end = var_393_end_0, end_mask = var_393_end_mask_0, x = var_222_cast_fp16)[name = string("op_393_cast_fp16")]; + tensor var_400_begin_0 = const()[name = string("op_400_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_400_end_0 = const()[name = string("op_400_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_400_end_mask_0 = const()[name = string("op_400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_400_cast_fp16 = slice_by_index(begin = var_400_begin_0, end = var_400_end_0, end_mask = var_400_end_mask_0, x = var_222_cast_fp16)[name = string("op_400_cast_fp16")]; + tensor var_407_begin_0 = const()[name = string("op_407_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_407_end_0 = const()[name = string("op_407_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_407_end_mask_0 = const()[name = string("op_407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = var_226_cast_fp16)[name = string("op_407_cast_fp16")]; + tensor var_414_begin_0 = const()[name = string("op_414_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_414_end_0 = const()[name = string("op_414_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_414_end_mask_0 = const()[name = string("op_414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_414_cast_fp16 = slice_by_index(begin = var_414_begin_0, end = var_414_end_0, end_mask = var_414_end_mask_0, x = var_226_cast_fp16)[name = string("op_414_cast_fp16")]; + tensor var_421_begin_0 = const()[name = string("op_421_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_421_end_0 = const()[name = string("op_421_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_421_end_mask_0 = const()[name = string("op_421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = var_226_cast_fp16)[name = string("op_421_cast_fp16")]; + tensor var_428_begin_0 = const()[name = string("op_428_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_428_end_0 = const()[name = string("op_428_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_428_end_mask_0 = const()[name = string("op_428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_428_cast_fp16 = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = var_226_cast_fp16)[name = string("op_428_cast_fp16")]; + tensor var_435_begin_0 = const()[name = string("op_435_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_435_end_0 = const()[name = string("op_435_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_435_end_mask_0 = const()[name = string("op_435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_435_cast_fp16 = slice_by_index(begin = var_435_begin_0, end = var_435_end_0, end_mask = var_435_end_mask_0, x = var_230_cast_fp16)[name = string("op_435_cast_fp16")]; + tensor var_442_begin_0 = const()[name = string("op_442_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_442_end_0 = const()[name = string("op_442_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_442_end_mask_0 = const()[name = string("op_442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = var_230_cast_fp16)[name = string("op_442_cast_fp16")]; + tensor var_449_begin_0 = const()[name = string("op_449_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_449_end_0 = const()[name = string("op_449_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_449_end_mask_0 = const()[name = string("op_449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_449_cast_fp16 = slice_by_index(begin = var_449_begin_0, end = var_449_end_0, end_mask = var_449_end_mask_0, x = var_230_cast_fp16)[name = string("op_449_cast_fp16")]; + tensor var_456_begin_0 = const()[name = string("op_456_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_456_end_0 = const()[name = string("op_456_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_456_end_mask_0 = const()[name = string("op_456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_456_cast_fp16 = slice_by_index(begin = var_456_begin_0, end = var_456_end_0, end_mask = var_456_end_mask_0, x = var_230_cast_fp16)[name = string("op_456_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_461_begin_0 = const()[name = string("op_461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_461_end_0 = const()[name = string("op_461_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_461_end_mask_0 = const()[name = string("op_461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_5")]; + tensor var_461_cast_fp16 = slice_by_index(begin = var_461_begin_0, end = var_461_end_0, end_mask = var_461_end_mask_0, x = k_1_cast_fp16)[name = string("op_461_cast_fp16")]; + tensor var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = k_1_cast_fp16)[name = string("op_465_cast_fp16")]; + tensor var_469_begin_0 = const()[name = string("op_469_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_469_end_0 = const()[name = string("op_469_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_469_end_mask_0 = const()[name = string("op_469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = k_1_cast_fp16)[name = string("op_469_cast_fp16")]; + tensor var_473_begin_0 = const()[name = string("op_473_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_473_end_0 = const()[name = string("op_473_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_473_end_mask_0 = const()[name = string("op_473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = k_1_cast_fp16)[name = string("op_473_cast_fp16")]; + tensor var_477_begin_0 = const()[name = string("op_477_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_477_end_0 = const()[name = string("op_477_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_477_end_mask_0 = const()[name = string("op_477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = k_1_cast_fp16)[name = string("op_477_cast_fp16")]; + tensor var_481_begin_0 = const()[name = string("op_481_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_481_end_0 = const()[name = string("op_481_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_481_end_mask_0 = const()[name = string("op_481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_481_cast_fp16 = slice_by_index(begin = var_481_begin_0, end = var_481_end_0, end_mask = var_481_end_mask_0, x = k_1_cast_fp16)[name = string("op_481_cast_fp16")]; + tensor var_485_begin_0 = const()[name = string("op_485_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_485_end_0 = const()[name = string("op_485_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_485_end_mask_0 = const()[name = string("op_485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_485_cast_fp16 = slice_by_index(begin = var_485_begin_0, end = var_485_end_0, end_mask = var_485_end_mask_0, x = k_1_cast_fp16)[name = string("op_485_cast_fp16")]; + tensor var_489_begin_0 = const()[name = string("op_489_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_489_end_0 = const()[name = string("op_489_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_489_end_mask_0 = const()[name = string("op_489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_489_cast_fp16 = slice_by_index(begin = var_489_begin_0, end = var_489_end_0, end_mask = var_489_end_mask_0, x = k_1_cast_fp16)[name = string("op_489_cast_fp16")]; + tensor var_491_begin_0 = const()[name = string("op_491_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_491_end_0 = const()[name = string("op_491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_491_end_mask_0 = const()[name = string("op_491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_491_cast_fp16 = slice_by_index(begin = var_491_begin_0, end = var_491_end_0, end_mask = var_491_end_mask_0, x = value_1_cast_fp16)[name = string("op_491_cast_fp16")]; + tensor var_495_begin_0 = const()[name = string("op_495_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_495_end_0 = const()[name = string("op_495_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_495_end_mask_0 = const()[name = string("op_495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_495_cast_fp16 = slice_by_index(begin = var_495_begin_0, end = var_495_end_0, end_mask = var_495_end_mask_0, x = value_1_cast_fp16)[name = string("op_495_cast_fp16")]; + tensor var_499_begin_0 = const()[name = string("op_499_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_499_end_0 = const()[name = string("op_499_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_499_end_mask_0 = const()[name = string("op_499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_499_cast_fp16 = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = value_1_cast_fp16)[name = string("op_499_cast_fp16")]; + tensor var_503_begin_0 = const()[name = string("op_503_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_503_end_0 = const()[name = string("op_503_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_503_end_mask_0 = const()[name = string("op_503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_503_cast_fp16 = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = value_1_cast_fp16)[name = string("op_503_cast_fp16")]; + tensor var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = value_1_cast_fp16)[name = string("op_507_cast_fp16")]; + tensor var_511_begin_0 = const()[name = string("op_511_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_511_end_0 = const()[name = string("op_511_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_511_end_mask_0 = const()[name = string("op_511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = value_1_cast_fp16)[name = string("op_511_cast_fp16")]; + tensor var_515_begin_0 = const()[name = string("op_515_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_515_end_0 = const()[name = string("op_515_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_515_end_mask_0 = const()[name = string("op_515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = value_1_cast_fp16)[name = string("op_515_cast_fp16")]; + tensor var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = value_1_cast_fp16)[name = string("op_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_461_cast_fp16, var_239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_461_cast_fp16, var_246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_461_cast_fp16, var_253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_461_cast_fp16, var_260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_465_cast_fp16, var_267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_465_cast_fp16, var_274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_465_cast_fp16, var_281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_465_cast_fp16, var_288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_469_cast_fp16, var_295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_469_cast_fp16, var_302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_469_cast_fp16, var_309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_469_cast_fp16, var_316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_473_cast_fp16, var_323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_473_cast_fp16, var_330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_473_cast_fp16, var_337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_473_cast_fp16, var_344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_477_cast_fp16, var_351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_477_cast_fp16, var_358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_477_cast_fp16, var_365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_477_cast_fp16, var_372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_481_cast_fp16, var_379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_481_cast_fp16, var_386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_481_cast_fp16, var_393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_481_cast_fp16, var_400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_485_cast_fp16, var_407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_485_cast_fp16, var_414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_485_cast_fp16, var_421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_485_cast_fp16, var_428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_489_cast_fp16, var_435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_489_cast_fp16, var_442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_489_cast_fp16, var_449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_489_cast_fp16, var_456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + fp16 var_586_to_fp16 = const()[name = string("op_586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_586_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_588_to_fp16 = const()[name = string("op_588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_588_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_590_to_fp16 = const()[name = string("op_590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_590_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_592_to_fp16 = const()[name = string("op_592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_592_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_594_to_fp16 = const()[name = string("op_594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_594_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_596_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_598_to_fp16 = const()[name = string("op_598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_598_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_600_to_fp16 = const()[name = string("op_600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_600_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_602_to_fp16 = const()[name = string("op_602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_602_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_604_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_606_to_fp16 = const()[name = string("op_606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_606_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_608_to_fp16 = const()[name = string("op_608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_608_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_610_to_fp16 = const()[name = string("op_610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_610_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_612_to_fp16 = const()[name = string("op_612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_612_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_614_to_fp16 = const()[name = string("op_614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_614_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_616_to_fp16 = const()[name = string("op_616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_616_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_618_to_fp16 = const()[name = string("op_618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_618_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_620_to_fp16 = const()[name = string("op_620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_620_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_622_to_fp16 = const()[name = string("op_622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_622_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_624_to_fp16 = const()[name = string("op_624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_624_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_626_to_fp16 = const()[name = string("op_626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_626_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_628_to_fp16 = const()[name = string("op_628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_628_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_630_to_fp16 = const()[name = string("op_630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_630_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_632_to_fp16 = const()[name = string("op_632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_632_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_634_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_636_to_fp16 = const()[name = string("op_636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_636_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_638_to_fp16 = const()[name = string("op_638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_638_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_640_to_fp16 = const()[name = string("op_640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_640_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_642_to_fp16 = const()[name = string("op_642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_642_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_644_to_fp16 = const()[name = string("op_644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_644_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_646_to_fp16 = const()[name = string("op_646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_646_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_648_to_fp16 = const()[name = string("op_648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_648_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + tensor var_650_cast_fp16 = softmax(axis = var_147, x = aw_chunk_1_cast_fp16)[name = string("op_650_cast_fp16")]; + tensor var_651_cast_fp16 = softmax(axis = var_147, x = aw_chunk_3_cast_fp16)[name = string("op_651_cast_fp16")]; + tensor var_652_cast_fp16 = softmax(axis = var_147, x = aw_chunk_5_cast_fp16)[name = string("op_652_cast_fp16")]; + tensor var_653_cast_fp16 = softmax(axis = var_147, x = aw_chunk_7_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor var_654_cast_fp16 = softmax(axis = var_147, x = aw_chunk_9_cast_fp16)[name = string("op_654_cast_fp16")]; + tensor var_655_cast_fp16 = softmax(axis = var_147, x = aw_chunk_11_cast_fp16)[name = string("op_655_cast_fp16")]; + tensor var_656_cast_fp16 = softmax(axis = var_147, x = aw_chunk_13_cast_fp16)[name = string("op_656_cast_fp16")]; + tensor var_657_cast_fp16 = softmax(axis = var_147, x = aw_chunk_15_cast_fp16)[name = string("op_657_cast_fp16")]; + tensor var_658_cast_fp16 = softmax(axis = var_147, x = aw_chunk_17_cast_fp16)[name = string("op_658_cast_fp16")]; + tensor var_659_cast_fp16 = softmax(axis = var_147, x = aw_chunk_19_cast_fp16)[name = string("op_659_cast_fp16")]; + tensor var_660_cast_fp16 = softmax(axis = var_147, x = aw_chunk_21_cast_fp16)[name = string("op_660_cast_fp16")]; + tensor var_661_cast_fp16 = softmax(axis = var_147, x = aw_chunk_23_cast_fp16)[name = string("op_661_cast_fp16")]; + tensor var_662_cast_fp16 = softmax(axis = var_147, x = aw_chunk_25_cast_fp16)[name = string("op_662_cast_fp16")]; + tensor var_663_cast_fp16 = softmax(axis = var_147, x = aw_chunk_27_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_664_cast_fp16 = softmax(axis = var_147, x = aw_chunk_29_cast_fp16)[name = string("op_664_cast_fp16")]; + tensor var_665_cast_fp16 = softmax(axis = var_147, x = aw_chunk_31_cast_fp16)[name = string("op_665_cast_fp16")]; + tensor var_666_cast_fp16 = softmax(axis = var_147, x = aw_chunk_33_cast_fp16)[name = string("op_666_cast_fp16")]; + tensor var_667_cast_fp16 = softmax(axis = var_147, x = aw_chunk_35_cast_fp16)[name = string("op_667_cast_fp16")]; + tensor var_668_cast_fp16 = softmax(axis = var_147, x = aw_chunk_37_cast_fp16)[name = string("op_668_cast_fp16")]; + tensor var_669_cast_fp16 = softmax(axis = var_147, x = aw_chunk_39_cast_fp16)[name = string("op_669_cast_fp16")]; + tensor var_670_cast_fp16 = softmax(axis = var_147, x = aw_chunk_41_cast_fp16)[name = string("op_670_cast_fp16")]; + tensor var_671_cast_fp16 = softmax(axis = var_147, x = aw_chunk_43_cast_fp16)[name = string("op_671_cast_fp16")]; + tensor var_672_cast_fp16 = softmax(axis = var_147, x = aw_chunk_45_cast_fp16)[name = string("op_672_cast_fp16")]; + tensor var_673_cast_fp16 = softmax(axis = var_147, x = aw_chunk_47_cast_fp16)[name = string("op_673_cast_fp16")]; + tensor var_674_cast_fp16 = softmax(axis = var_147, x = aw_chunk_49_cast_fp16)[name = string("op_674_cast_fp16")]; + tensor var_675_cast_fp16 = softmax(axis = var_147, x = aw_chunk_51_cast_fp16)[name = string("op_675_cast_fp16")]; + tensor var_676_cast_fp16 = softmax(axis = var_147, x = aw_chunk_53_cast_fp16)[name = string("op_676_cast_fp16")]; + tensor var_677_cast_fp16 = softmax(axis = var_147, x = aw_chunk_55_cast_fp16)[name = string("op_677_cast_fp16")]; + tensor var_678_cast_fp16 = softmax(axis = var_147, x = aw_chunk_57_cast_fp16)[name = string("op_678_cast_fp16")]; + tensor var_679_cast_fp16 = softmax(axis = var_147, x = aw_chunk_59_cast_fp16)[name = string("op_679_cast_fp16")]; + tensor var_680_cast_fp16 = softmax(axis = var_147, x = aw_chunk_61_cast_fp16)[name = string("op_680_cast_fp16")]; + tensor var_681_cast_fp16 = softmax(axis = var_147, x = aw_chunk_63_cast_fp16)[name = string("op_681_cast_fp16")]; + string var_683_equation_0 = const()[name = string("op_683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_683_cast_fp16 = einsum(equation = var_683_equation_0, values = (var_491_cast_fp16, var_650_cast_fp16))[name = string("op_683_cast_fp16")]; + string var_685_equation_0 = const()[name = string("op_685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_685_cast_fp16 = einsum(equation = var_685_equation_0, values = (var_491_cast_fp16, var_651_cast_fp16))[name = string("op_685_cast_fp16")]; + string var_687_equation_0 = const()[name = string("op_687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_687_cast_fp16 = einsum(equation = var_687_equation_0, values = (var_491_cast_fp16, var_652_cast_fp16))[name = string("op_687_cast_fp16")]; + string var_689_equation_0 = const()[name = string("op_689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_689_cast_fp16 = einsum(equation = var_689_equation_0, values = (var_491_cast_fp16, var_653_cast_fp16))[name = string("op_689_cast_fp16")]; + string var_691_equation_0 = const()[name = string("op_691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_691_cast_fp16 = einsum(equation = var_691_equation_0, values = (var_495_cast_fp16, var_654_cast_fp16))[name = string("op_691_cast_fp16")]; + string var_693_equation_0 = const()[name = string("op_693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_693_cast_fp16 = einsum(equation = var_693_equation_0, values = (var_495_cast_fp16, var_655_cast_fp16))[name = string("op_693_cast_fp16")]; + string var_695_equation_0 = const()[name = string("op_695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_695_cast_fp16 = einsum(equation = var_695_equation_0, values = (var_495_cast_fp16, var_656_cast_fp16))[name = string("op_695_cast_fp16")]; + string var_697_equation_0 = const()[name = string("op_697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_697_cast_fp16 = einsum(equation = var_697_equation_0, values = (var_495_cast_fp16, var_657_cast_fp16))[name = string("op_697_cast_fp16")]; + string var_699_equation_0 = const()[name = string("op_699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_699_cast_fp16 = einsum(equation = var_699_equation_0, values = (var_499_cast_fp16, var_658_cast_fp16))[name = string("op_699_cast_fp16")]; + string var_701_equation_0 = const()[name = string("op_701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_701_cast_fp16 = einsum(equation = var_701_equation_0, values = (var_499_cast_fp16, var_659_cast_fp16))[name = string("op_701_cast_fp16")]; + string var_703_equation_0 = const()[name = string("op_703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_703_cast_fp16 = einsum(equation = var_703_equation_0, values = (var_499_cast_fp16, var_660_cast_fp16))[name = string("op_703_cast_fp16")]; + string var_705_equation_0 = const()[name = string("op_705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_705_cast_fp16 = einsum(equation = var_705_equation_0, values = (var_499_cast_fp16, var_661_cast_fp16))[name = string("op_705_cast_fp16")]; + string var_707_equation_0 = const()[name = string("op_707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_707_cast_fp16 = einsum(equation = var_707_equation_0, values = (var_503_cast_fp16, var_662_cast_fp16))[name = string("op_707_cast_fp16")]; + string var_709_equation_0 = const()[name = string("op_709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_709_cast_fp16 = einsum(equation = var_709_equation_0, values = (var_503_cast_fp16, var_663_cast_fp16))[name = string("op_709_cast_fp16")]; + string var_711_equation_0 = const()[name = string("op_711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_711_cast_fp16 = einsum(equation = var_711_equation_0, values = (var_503_cast_fp16, var_664_cast_fp16))[name = string("op_711_cast_fp16")]; + string var_713_equation_0 = const()[name = string("op_713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_713_cast_fp16 = einsum(equation = var_713_equation_0, values = (var_503_cast_fp16, var_665_cast_fp16))[name = string("op_713_cast_fp16")]; + string var_715_equation_0 = const()[name = string("op_715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_715_cast_fp16 = einsum(equation = var_715_equation_0, values = (var_507_cast_fp16, var_666_cast_fp16))[name = string("op_715_cast_fp16")]; + string var_717_equation_0 = const()[name = string("op_717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_717_cast_fp16 = einsum(equation = var_717_equation_0, values = (var_507_cast_fp16, var_667_cast_fp16))[name = string("op_717_cast_fp16")]; + string var_719_equation_0 = const()[name = string("op_719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_719_cast_fp16 = einsum(equation = var_719_equation_0, values = (var_507_cast_fp16, var_668_cast_fp16))[name = string("op_719_cast_fp16")]; + string var_721_equation_0 = const()[name = string("op_721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_721_cast_fp16 = einsum(equation = var_721_equation_0, values = (var_507_cast_fp16, var_669_cast_fp16))[name = string("op_721_cast_fp16")]; + string var_723_equation_0 = const()[name = string("op_723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_723_cast_fp16 = einsum(equation = var_723_equation_0, values = (var_511_cast_fp16, var_670_cast_fp16))[name = string("op_723_cast_fp16")]; + string var_725_equation_0 = const()[name = string("op_725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_725_cast_fp16 = einsum(equation = var_725_equation_0, values = (var_511_cast_fp16, var_671_cast_fp16))[name = string("op_725_cast_fp16")]; + string var_727_equation_0 = const()[name = string("op_727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_727_cast_fp16 = einsum(equation = var_727_equation_0, values = (var_511_cast_fp16, var_672_cast_fp16))[name = string("op_727_cast_fp16")]; + string var_729_equation_0 = const()[name = string("op_729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_729_cast_fp16 = einsum(equation = var_729_equation_0, values = (var_511_cast_fp16, var_673_cast_fp16))[name = string("op_729_cast_fp16")]; + string var_731_equation_0 = const()[name = string("op_731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_731_cast_fp16 = einsum(equation = var_731_equation_0, values = (var_515_cast_fp16, var_674_cast_fp16))[name = string("op_731_cast_fp16")]; + string var_733_equation_0 = const()[name = string("op_733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_733_cast_fp16 = einsum(equation = var_733_equation_0, values = (var_515_cast_fp16, var_675_cast_fp16))[name = string("op_733_cast_fp16")]; + string var_735_equation_0 = const()[name = string("op_735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_735_cast_fp16 = einsum(equation = var_735_equation_0, values = (var_515_cast_fp16, var_676_cast_fp16))[name = string("op_735_cast_fp16")]; + string var_737_equation_0 = const()[name = string("op_737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_737_cast_fp16 = einsum(equation = var_737_equation_0, values = (var_515_cast_fp16, var_677_cast_fp16))[name = string("op_737_cast_fp16")]; + string var_739_equation_0 = const()[name = string("op_739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_739_cast_fp16 = einsum(equation = var_739_equation_0, values = (var_519_cast_fp16, var_678_cast_fp16))[name = string("op_739_cast_fp16")]; + string var_741_equation_0 = const()[name = string("op_741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_741_cast_fp16 = einsum(equation = var_741_equation_0, values = (var_519_cast_fp16, var_679_cast_fp16))[name = string("op_741_cast_fp16")]; + string var_743_equation_0 = const()[name = string("op_743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_743_cast_fp16 = einsum(equation = var_743_equation_0, values = (var_519_cast_fp16, var_680_cast_fp16))[name = string("op_743_cast_fp16")]; + string var_745_equation_0 = const()[name = string("op_745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_745_cast_fp16 = einsum(equation = var_745_equation_0, values = (var_519_cast_fp16, var_681_cast_fp16))[name = string("op_745_cast_fp16")]; + bool var_747_interleave_0 = const()[name = string("op_747_interleave_0"), val = bool(false)]; + tensor var_747_cast_fp16 = concat(axis = var_134, interleave = var_747_interleave_0, values = (var_683_cast_fp16, var_685_cast_fp16, var_687_cast_fp16, var_689_cast_fp16))[name = string("op_747_cast_fp16")]; + bool var_749_interleave_0 = const()[name = string("op_749_interleave_0"), val = bool(false)]; + tensor var_749_cast_fp16 = concat(axis = var_134, interleave = var_749_interleave_0, values = (var_691_cast_fp16, var_693_cast_fp16, var_695_cast_fp16, var_697_cast_fp16))[name = string("op_749_cast_fp16")]; + bool var_751_interleave_0 = const()[name = string("op_751_interleave_0"), val = bool(false)]; + tensor var_751_cast_fp16 = concat(axis = var_134, interleave = var_751_interleave_0, values = (var_699_cast_fp16, var_701_cast_fp16, var_703_cast_fp16, var_705_cast_fp16))[name = string("op_751_cast_fp16")]; + bool var_753_interleave_0 = const()[name = string("op_753_interleave_0"), val = bool(false)]; + tensor var_753_cast_fp16 = concat(axis = var_134, interleave = var_753_interleave_0, values = (var_707_cast_fp16, var_709_cast_fp16, var_711_cast_fp16, var_713_cast_fp16))[name = string("op_753_cast_fp16")]; + bool var_755_interleave_0 = const()[name = string("op_755_interleave_0"), val = bool(false)]; + tensor var_755_cast_fp16 = concat(axis = var_134, interleave = var_755_interleave_0, values = (var_715_cast_fp16, var_717_cast_fp16, var_719_cast_fp16, var_721_cast_fp16))[name = string("op_755_cast_fp16")]; + bool var_757_interleave_0 = const()[name = string("op_757_interleave_0"), val = bool(false)]; + tensor var_757_cast_fp16 = concat(axis = var_134, interleave = var_757_interleave_0, values = (var_723_cast_fp16, var_725_cast_fp16, var_727_cast_fp16, var_729_cast_fp16))[name = string("op_757_cast_fp16")]; + bool var_759_interleave_0 = const()[name = string("op_759_interleave_0"), val = bool(false)]; + tensor var_759_cast_fp16 = concat(axis = var_134, interleave = var_759_interleave_0, values = (var_731_cast_fp16, var_733_cast_fp16, var_735_cast_fp16, var_737_cast_fp16))[name = string("op_759_cast_fp16")]; + bool var_761_interleave_0 = const()[name = string("op_761_interleave_0"), val = bool(false)]; + tensor var_761_cast_fp16 = concat(axis = var_134, interleave = var_761_interleave_0, values = (var_739_cast_fp16, var_741_cast_fp16, var_743_cast_fp16, var_745_cast_fp16))[name = string("op_761_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_147, interleave = input_1_interleave_0, values = (var_747_cast_fp16, var_749_cast_fp16, var_751_cast_fp16, var_753_cast_fp16, var_755_cast_fp16, var_757_cast_fp16, var_759_cast_fp16, var_761_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936640)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5460992)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_780_to_fp16 = const()[name = string("op_780_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_780_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5462080)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5463168)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5464256)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7561472)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7565632)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9662848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_809 = const()[name = string("op_809"), val = int32(3)]; + int32 var_822 = const()[name = string("op_822"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_839_to_fp16 = const()[name = string("op_839_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_839_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9663936)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9665024)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9666112)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10190464)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10191552)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10715904)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11240256)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_877_begin_0 = const()[name = string("op_877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_877_end_0 = const()[name = string("op_877_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_877_end_mask_0 = const()[name = string("op_877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_877_cast_fp16 = slice_by_index(begin = var_877_begin_0, end = var_877_end_0, end_mask = var_877_end_mask_0, x = query_3_cast_fp16)[name = string("op_877_cast_fp16")]; + tensor var_881_begin_0 = const()[name = string("op_881_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_881_end_0 = const()[name = string("op_881_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_881_end_mask_0 = const()[name = string("op_881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_881_cast_fp16 = slice_by_index(begin = var_881_begin_0, end = var_881_end_0, end_mask = var_881_end_mask_0, x = query_3_cast_fp16)[name = string("op_881_cast_fp16")]; + tensor var_885_begin_0 = const()[name = string("op_885_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_885_end_0 = const()[name = string("op_885_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_885_end_mask_0 = const()[name = string("op_885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_885_cast_fp16 = slice_by_index(begin = var_885_begin_0, end = var_885_end_0, end_mask = var_885_end_mask_0, x = query_3_cast_fp16)[name = string("op_885_cast_fp16")]; + tensor var_889_begin_0 = const()[name = string("op_889_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_889_end_0 = const()[name = string("op_889_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_889_end_mask_0 = const()[name = string("op_889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_889_cast_fp16 = slice_by_index(begin = var_889_begin_0, end = var_889_end_0, end_mask = var_889_end_mask_0, x = query_3_cast_fp16)[name = string("op_889_cast_fp16")]; + tensor var_893_begin_0 = const()[name = string("op_893_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_893_end_0 = const()[name = string("op_893_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_893_end_mask_0 = const()[name = string("op_893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_893_cast_fp16 = slice_by_index(begin = var_893_begin_0, end = var_893_end_0, end_mask = var_893_end_mask_0, x = query_3_cast_fp16)[name = string("op_893_cast_fp16")]; + tensor var_897_begin_0 = const()[name = string("op_897_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_897_end_0 = const()[name = string("op_897_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_897_end_mask_0 = const()[name = string("op_897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_897_cast_fp16 = slice_by_index(begin = var_897_begin_0, end = var_897_end_0, end_mask = var_897_end_mask_0, x = query_3_cast_fp16)[name = string("op_897_cast_fp16")]; + tensor var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = query_3_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_905_begin_0 = const()[name = string("op_905_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_905_end_0 = const()[name = string("op_905_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_905_end_mask_0 = const()[name = string("op_905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_905_cast_fp16 = slice_by_index(begin = var_905_begin_0, end = var_905_end_0, end_mask = var_905_end_mask_0, x = query_3_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_914_begin_0 = const()[name = string("op_914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_914_end_0 = const()[name = string("op_914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_914_end_mask_0 = const()[name = string("op_914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_914_cast_fp16 = slice_by_index(begin = var_914_begin_0, end = var_914_end_0, end_mask = var_914_end_mask_0, x = var_877_cast_fp16)[name = string("op_914_cast_fp16")]; + tensor var_921_begin_0 = const()[name = string("op_921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_921_end_0 = const()[name = string("op_921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_921_end_mask_0 = const()[name = string("op_921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_921_cast_fp16 = slice_by_index(begin = var_921_begin_0, end = var_921_end_0, end_mask = var_921_end_mask_0, x = var_877_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_928_begin_0 = const()[name = string("op_928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_928_end_0 = const()[name = string("op_928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_928_end_mask_0 = const()[name = string("op_928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_928_cast_fp16 = slice_by_index(begin = var_928_begin_0, end = var_928_end_0, end_mask = var_928_end_mask_0, x = var_877_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_935_begin_0 = const()[name = string("op_935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_935_end_0 = const()[name = string("op_935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_935_end_mask_0 = const()[name = string("op_935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_935_cast_fp16 = slice_by_index(begin = var_935_begin_0, end = var_935_end_0, end_mask = var_935_end_mask_0, x = var_877_cast_fp16)[name = string("op_935_cast_fp16")]; + tensor var_942_begin_0 = const()[name = string("op_942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_942_end_0 = const()[name = string("op_942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_942_end_mask_0 = const()[name = string("op_942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = var_881_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = var_881_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_956_begin_0 = const()[name = string("op_956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_956_end_0 = const()[name = string("op_956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_956_end_mask_0 = const()[name = string("op_956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_956_cast_fp16 = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, x = var_881_cast_fp16)[name = string("op_956_cast_fp16")]; + tensor var_963_begin_0 = const()[name = string("op_963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_963_end_0 = const()[name = string("op_963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_963_end_mask_0 = const()[name = string("op_963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_963_cast_fp16 = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = var_881_cast_fp16)[name = string("op_963_cast_fp16")]; + tensor var_970_begin_0 = const()[name = string("op_970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_end_0 = const()[name = string("op_970_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_970_end_mask_0 = const()[name = string("op_970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, x = var_885_cast_fp16)[name = string("op_970_cast_fp16")]; + tensor var_977_begin_0 = const()[name = string("op_977_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_977_end_0 = const()[name = string("op_977_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_977_end_mask_0 = const()[name = string("op_977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = var_885_cast_fp16)[name = string("op_977_cast_fp16")]; + tensor var_984_begin_0 = const()[name = string("op_984_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_984_end_0 = const()[name = string("op_984_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_984_end_mask_0 = const()[name = string("op_984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_984_cast_fp16 = slice_by_index(begin = var_984_begin_0, end = var_984_end_0, end_mask = var_984_end_mask_0, x = var_885_cast_fp16)[name = string("op_984_cast_fp16")]; + tensor var_991_begin_0 = const()[name = string("op_991_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_991_end_0 = const()[name = string("op_991_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_991_end_mask_0 = const()[name = string("op_991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = var_885_cast_fp16)[name = string("op_991_cast_fp16")]; + tensor var_998_begin_0 = const()[name = string("op_998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_998_end_0 = const()[name = string("op_998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_998_end_mask_0 = const()[name = string("op_998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_998_cast_fp16 = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = var_889_cast_fp16)[name = string("op_998_cast_fp16")]; + tensor var_1005_begin_0 = const()[name = string("op_1005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1005_end_0 = const()[name = string("op_1005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1005_end_mask_0 = const()[name = string("op_1005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1005_cast_fp16 = slice_by_index(begin = var_1005_begin_0, end = var_1005_end_0, end_mask = var_1005_end_mask_0, x = var_889_cast_fp16)[name = string("op_1005_cast_fp16")]; + tensor var_1012_begin_0 = const()[name = string("op_1012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1012_end_0 = const()[name = string("op_1012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1012_end_mask_0 = const()[name = string("op_1012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = var_889_cast_fp16)[name = string("op_1012_cast_fp16")]; + tensor var_1019_begin_0 = const()[name = string("op_1019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1019_end_0 = const()[name = string("op_1019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1019_end_mask_0 = const()[name = string("op_1019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = var_1019_end_0, end_mask = var_1019_end_mask_0, x = var_889_cast_fp16)[name = string("op_1019_cast_fp16")]; + tensor var_1026_begin_0 = const()[name = string("op_1026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1026_end_0 = const()[name = string("op_1026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1026_end_mask_0 = const()[name = string("op_1026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1026_cast_fp16 = slice_by_index(begin = var_1026_begin_0, end = var_1026_end_0, end_mask = var_1026_end_mask_0, x = var_893_cast_fp16)[name = string("op_1026_cast_fp16")]; + tensor var_1033_begin_0 = const()[name = string("op_1033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1033_end_0 = const()[name = string("op_1033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1033_end_mask_0 = const()[name = string("op_1033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1033_cast_fp16 = slice_by_index(begin = var_1033_begin_0, end = var_1033_end_0, end_mask = var_1033_end_mask_0, x = var_893_cast_fp16)[name = string("op_1033_cast_fp16")]; + tensor var_1040_begin_0 = const()[name = string("op_1040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1040_end_0 = const()[name = string("op_1040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1040_end_mask_0 = const()[name = string("op_1040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1040_cast_fp16 = slice_by_index(begin = var_1040_begin_0, end = var_1040_end_0, end_mask = var_1040_end_mask_0, x = var_893_cast_fp16)[name = string("op_1040_cast_fp16")]; + tensor var_1047_begin_0 = const()[name = string("op_1047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1047_end_0 = const()[name = string("op_1047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1047_end_mask_0 = const()[name = string("op_1047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = var_893_cast_fp16)[name = string("op_1047_cast_fp16")]; + tensor var_1054_begin_0 = const()[name = string("op_1054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1054_end_0 = const()[name = string("op_1054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1054_end_mask_0 = const()[name = string("op_1054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1054_cast_fp16 = slice_by_index(begin = var_1054_begin_0, end = var_1054_end_0, end_mask = var_1054_end_mask_0, x = var_897_cast_fp16)[name = string("op_1054_cast_fp16")]; + tensor var_1061_begin_0 = const()[name = string("op_1061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1061_end_0 = const()[name = string("op_1061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1061_end_mask_0 = const()[name = string("op_1061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1061_cast_fp16 = slice_by_index(begin = var_1061_begin_0, end = var_1061_end_0, end_mask = var_1061_end_mask_0, x = var_897_cast_fp16)[name = string("op_1061_cast_fp16")]; + tensor var_1068_begin_0 = const()[name = string("op_1068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1068_end_0 = const()[name = string("op_1068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1068_end_mask_0 = const()[name = string("op_1068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1068_cast_fp16 = slice_by_index(begin = var_1068_begin_0, end = var_1068_end_0, end_mask = var_1068_end_mask_0, x = var_897_cast_fp16)[name = string("op_1068_cast_fp16")]; + tensor var_1075_begin_0 = const()[name = string("op_1075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1075_end_0 = const()[name = string("op_1075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1075_end_mask_0 = const()[name = string("op_1075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1075_cast_fp16 = slice_by_index(begin = var_1075_begin_0, end = var_1075_end_0, end_mask = var_1075_end_mask_0, x = var_897_cast_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1082_begin_0 = const()[name = string("op_1082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1082_end_0 = const()[name = string("op_1082_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1082_end_mask_0 = const()[name = string("op_1082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1082_cast_fp16 = slice_by_index(begin = var_1082_begin_0, end = var_1082_end_0, end_mask = var_1082_end_mask_0, x = var_901_cast_fp16)[name = string("op_1082_cast_fp16")]; + tensor var_1089_begin_0 = const()[name = string("op_1089_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1089_end_0 = const()[name = string("op_1089_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1089_end_mask_0 = const()[name = string("op_1089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1089_cast_fp16 = slice_by_index(begin = var_1089_begin_0, end = var_1089_end_0, end_mask = var_1089_end_mask_0, x = var_901_cast_fp16)[name = string("op_1089_cast_fp16")]; + tensor var_1096_begin_0 = const()[name = string("op_1096_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1096_end_0 = const()[name = string("op_1096_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1096_end_mask_0 = const()[name = string("op_1096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1096_cast_fp16 = slice_by_index(begin = var_1096_begin_0, end = var_1096_end_0, end_mask = var_1096_end_mask_0, x = var_901_cast_fp16)[name = string("op_1096_cast_fp16")]; + tensor var_1103_begin_0 = const()[name = string("op_1103_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1103_end_0 = const()[name = string("op_1103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1103_end_mask_0 = const()[name = string("op_1103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1103_cast_fp16 = slice_by_index(begin = var_1103_begin_0, end = var_1103_end_0, end_mask = var_1103_end_mask_0, x = var_901_cast_fp16)[name = string("op_1103_cast_fp16")]; + tensor var_1110_begin_0 = const()[name = string("op_1110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1110_end_0 = const()[name = string("op_1110_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1110_end_mask_0 = const()[name = string("op_1110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1110_cast_fp16 = slice_by_index(begin = var_1110_begin_0, end = var_1110_end_0, end_mask = var_1110_end_mask_0, x = var_905_cast_fp16)[name = string("op_1110_cast_fp16")]; + tensor var_1117_begin_0 = const()[name = string("op_1117_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1117_end_0 = const()[name = string("op_1117_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1117_end_mask_0 = const()[name = string("op_1117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1117_cast_fp16 = slice_by_index(begin = var_1117_begin_0, end = var_1117_end_0, end_mask = var_1117_end_mask_0, x = var_905_cast_fp16)[name = string("op_1117_cast_fp16")]; + tensor var_1124_begin_0 = const()[name = string("op_1124_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1124_end_0 = const()[name = string("op_1124_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1124_end_mask_0 = const()[name = string("op_1124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1124_cast_fp16 = slice_by_index(begin = var_1124_begin_0, end = var_1124_end_0, end_mask = var_1124_end_mask_0, x = var_905_cast_fp16)[name = string("op_1124_cast_fp16")]; + tensor var_1131_begin_0 = const()[name = string("op_1131_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1131_end_0 = const()[name = string("op_1131_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1131_end_mask_0 = const()[name = string("op_1131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1131_cast_fp16 = slice_by_index(begin = var_1131_begin_0, end = var_1131_end_0, end_mask = var_1131_end_mask_0, x = var_905_cast_fp16)[name = string("op_1131_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1136_begin_0 = const()[name = string("op_1136_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1136_end_0 = const()[name = string("op_1136_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1136_end_mask_0 = const()[name = string("op_1136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_4")]; + tensor var_1136_cast_fp16 = slice_by_index(begin = var_1136_begin_0, end = var_1136_end_0, end_mask = var_1136_end_mask_0, x = k_3_cast_fp16)[name = string("op_1136_cast_fp16")]; + tensor var_1140_begin_0 = const()[name = string("op_1140_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1140_end_0 = const()[name = string("op_1140_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1140_end_mask_0 = const()[name = string("op_1140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1140_cast_fp16 = slice_by_index(begin = var_1140_begin_0, end = var_1140_end_0, end_mask = var_1140_end_mask_0, x = k_3_cast_fp16)[name = string("op_1140_cast_fp16")]; + tensor var_1144_begin_0 = const()[name = string("op_1144_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1144_end_0 = const()[name = string("op_1144_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1144_end_mask_0 = const()[name = string("op_1144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1144_cast_fp16 = slice_by_index(begin = var_1144_begin_0, end = var_1144_end_0, end_mask = var_1144_end_mask_0, x = k_3_cast_fp16)[name = string("op_1144_cast_fp16")]; + tensor var_1148_begin_0 = const()[name = string("op_1148_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1148_end_0 = const()[name = string("op_1148_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1148_end_mask_0 = const()[name = string("op_1148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1148_cast_fp16 = slice_by_index(begin = var_1148_begin_0, end = var_1148_end_0, end_mask = var_1148_end_mask_0, x = k_3_cast_fp16)[name = string("op_1148_cast_fp16")]; + tensor var_1152_begin_0 = const()[name = string("op_1152_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1152_end_0 = const()[name = string("op_1152_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1152_end_mask_0 = const()[name = string("op_1152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1152_cast_fp16 = slice_by_index(begin = var_1152_begin_0, end = var_1152_end_0, end_mask = var_1152_end_mask_0, x = k_3_cast_fp16)[name = string("op_1152_cast_fp16")]; + tensor var_1156_begin_0 = const()[name = string("op_1156_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1156_end_0 = const()[name = string("op_1156_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1156_end_mask_0 = const()[name = string("op_1156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1156_cast_fp16 = slice_by_index(begin = var_1156_begin_0, end = var_1156_end_0, end_mask = var_1156_end_mask_0, x = k_3_cast_fp16)[name = string("op_1156_cast_fp16")]; + tensor var_1160_begin_0 = const()[name = string("op_1160_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1160_end_0 = const()[name = string("op_1160_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1160_end_mask_0 = const()[name = string("op_1160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1160_cast_fp16 = slice_by_index(begin = var_1160_begin_0, end = var_1160_end_0, end_mask = var_1160_end_mask_0, x = k_3_cast_fp16)[name = string("op_1160_cast_fp16")]; + tensor var_1164_begin_0 = const()[name = string("op_1164_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1164_end_0 = const()[name = string("op_1164_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1164_end_mask_0 = const()[name = string("op_1164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1164_cast_fp16 = slice_by_index(begin = var_1164_begin_0, end = var_1164_end_0, end_mask = var_1164_end_mask_0, x = k_3_cast_fp16)[name = string("op_1164_cast_fp16")]; + tensor var_1166_begin_0 = const()[name = string("op_1166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1166_end_0 = const()[name = string("op_1166_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1166_end_mask_0 = const()[name = string("op_1166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1166_cast_fp16 = slice_by_index(begin = var_1166_begin_0, end = var_1166_end_0, end_mask = var_1166_end_mask_0, x = value_3_cast_fp16)[name = string("op_1166_cast_fp16")]; + tensor var_1170_begin_0 = const()[name = string("op_1170_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1170_end_0 = const()[name = string("op_1170_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1170_end_mask_0 = const()[name = string("op_1170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1170_cast_fp16 = slice_by_index(begin = var_1170_begin_0, end = var_1170_end_0, end_mask = var_1170_end_mask_0, x = value_3_cast_fp16)[name = string("op_1170_cast_fp16")]; + tensor var_1174_begin_0 = const()[name = string("op_1174_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1174_end_0 = const()[name = string("op_1174_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1174_end_mask_0 = const()[name = string("op_1174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1174_cast_fp16 = slice_by_index(begin = var_1174_begin_0, end = var_1174_end_0, end_mask = var_1174_end_mask_0, x = value_3_cast_fp16)[name = string("op_1174_cast_fp16")]; + tensor var_1178_begin_0 = const()[name = string("op_1178_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1178_end_0 = const()[name = string("op_1178_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1178_end_mask_0 = const()[name = string("op_1178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1178_cast_fp16 = slice_by_index(begin = var_1178_begin_0, end = var_1178_end_0, end_mask = var_1178_end_mask_0, x = value_3_cast_fp16)[name = string("op_1178_cast_fp16")]; + tensor var_1182_begin_0 = const()[name = string("op_1182_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1182_end_0 = const()[name = string("op_1182_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1182_end_mask_0 = const()[name = string("op_1182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1182_cast_fp16 = slice_by_index(begin = var_1182_begin_0, end = var_1182_end_0, end_mask = var_1182_end_mask_0, x = value_3_cast_fp16)[name = string("op_1182_cast_fp16")]; + tensor var_1186_begin_0 = const()[name = string("op_1186_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1186_end_0 = const()[name = string("op_1186_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1186_end_mask_0 = const()[name = string("op_1186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1186_cast_fp16 = slice_by_index(begin = var_1186_begin_0, end = var_1186_end_0, end_mask = var_1186_end_mask_0, x = value_3_cast_fp16)[name = string("op_1186_cast_fp16")]; + tensor var_1190_begin_0 = const()[name = string("op_1190_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1190_end_0 = const()[name = string("op_1190_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1190_end_mask_0 = const()[name = string("op_1190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1190_cast_fp16 = slice_by_index(begin = var_1190_begin_0, end = var_1190_end_0, end_mask = var_1190_end_mask_0, x = value_3_cast_fp16)[name = string("op_1190_cast_fp16")]; + tensor var_1194_begin_0 = const()[name = string("op_1194_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1194_end_0 = const()[name = string("op_1194_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1194_end_mask_0 = const()[name = string("op_1194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, x = value_3_cast_fp16)[name = string("op_1194_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_1136_cast_fp16, var_914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_1136_cast_fp16, var_921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_1136_cast_fp16, var_928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_1136_cast_fp16, var_935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_1140_cast_fp16, var_942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_1140_cast_fp16, var_949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_1140_cast_fp16, var_956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_1140_cast_fp16, var_963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_1144_cast_fp16, var_970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_1144_cast_fp16, var_977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_1144_cast_fp16, var_984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_1144_cast_fp16, var_991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_1148_cast_fp16, var_998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_1148_cast_fp16, var_1005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_1148_cast_fp16, var_1012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_1148_cast_fp16, var_1019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1152_cast_fp16, var_1026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1152_cast_fp16, var_1033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1152_cast_fp16, var_1040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1152_cast_fp16, var_1047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1156_cast_fp16, var_1054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1156_cast_fp16, var_1061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1156_cast_fp16, var_1068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1156_cast_fp16, var_1075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1160_cast_fp16, var_1082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1160_cast_fp16, var_1089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1160_cast_fp16, var_1096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1160_cast_fp16, var_1103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1164_cast_fp16, var_1110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1164_cast_fp16, var_1117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1164_cast_fp16, var_1124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1164_cast_fp16, var_1131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + fp16 var_1261_to_fp16 = const()[name = string("op_1261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1261_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1263_to_fp16 = const()[name = string("op_1263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1263_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1265_to_fp16 = const()[name = string("op_1265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1265_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1267_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1269_to_fp16 = const()[name = string("op_1269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1269_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1271_to_fp16 = const()[name = string("op_1271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1271_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1273_to_fp16 = const()[name = string("op_1273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1273_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1275_to_fp16 = const()[name = string("op_1275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1275_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1277_to_fp16 = const()[name = string("op_1277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1277_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1279_to_fp16 = const()[name = string("op_1279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1279_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1281_to_fp16 = const()[name = string("op_1281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1281_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1283_to_fp16 = const()[name = string("op_1283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1283_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1285_to_fp16 = const()[name = string("op_1285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1285_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1287_to_fp16 = const()[name = string("op_1287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1287_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1289_to_fp16 = const()[name = string("op_1289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1289_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1291_to_fp16 = const()[name = string("op_1291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1291_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + fp16 var_1293_to_fp16 = const()[name = string("op_1293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1293_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1295_to_fp16 = const()[name = string("op_1295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1295_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1297_to_fp16 = const()[name = string("op_1297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1297_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1299_to_fp16 = const()[name = string("op_1299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1299_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1301_to_fp16 = const()[name = string("op_1301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1301_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1303_to_fp16 = const()[name = string("op_1303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1303_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1305_to_fp16 = const()[name = string("op_1305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1305_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1307_to_fp16 = const()[name = string("op_1307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1307_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1309_to_fp16 = const()[name = string("op_1309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1309_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1311_to_fp16 = const()[name = string("op_1311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1311_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1313_to_fp16 = const()[name = string("op_1313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1313_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1315_to_fp16 = const()[name = string("op_1315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1315_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1317_to_fp16 = const()[name = string("op_1317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1317_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1319_to_fp16 = const()[name = string("op_1319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1319_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1321_to_fp16 = const()[name = string("op_1321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1321_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1323_to_fp16 = const()[name = string("op_1323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1323_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + tensor var_1325_cast_fp16 = softmax(axis = var_822, x = aw_chunk_65_cast_fp16)[name = string("op_1325_cast_fp16")]; + tensor var_1326_cast_fp16 = softmax(axis = var_822, x = aw_chunk_67_cast_fp16)[name = string("op_1326_cast_fp16")]; + tensor var_1327_cast_fp16 = softmax(axis = var_822, x = aw_chunk_69_cast_fp16)[name = string("op_1327_cast_fp16")]; + tensor var_1328_cast_fp16 = softmax(axis = var_822, x = aw_chunk_71_cast_fp16)[name = string("op_1328_cast_fp16")]; + tensor var_1329_cast_fp16 = softmax(axis = var_822, x = aw_chunk_73_cast_fp16)[name = string("op_1329_cast_fp16")]; + tensor var_1330_cast_fp16 = softmax(axis = var_822, x = aw_chunk_75_cast_fp16)[name = string("op_1330_cast_fp16")]; + tensor var_1331_cast_fp16 = softmax(axis = var_822, x = aw_chunk_77_cast_fp16)[name = string("op_1331_cast_fp16")]; + tensor var_1332_cast_fp16 = softmax(axis = var_822, x = aw_chunk_79_cast_fp16)[name = string("op_1332_cast_fp16")]; + tensor var_1333_cast_fp16 = softmax(axis = var_822, x = aw_chunk_81_cast_fp16)[name = string("op_1333_cast_fp16")]; + tensor var_1334_cast_fp16 = softmax(axis = var_822, x = aw_chunk_83_cast_fp16)[name = string("op_1334_cast_fp16")]; + tensor var_1335_cast_fp16 = softmax(axis = var_822, x = aw_chunk_85_cast_fp16)[name = string("op_1335_cast_fp16")]; + tensor var_1336_cast_fp16 = softmax(axis = var_822, x = aw_chunk_87_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor var_1337_cast_fp16 = softmax(axis = var_822, x = aw_chunk_89_cast_fp16)[name = string("op_1337_cast_fp16")]; + tensor var_1338_cast_fp16 = softmax(axis = var_822, x = aw_chunk_91_cast_fp16)[name = string("op_1338_cast_fp16")]; + tensor var_1339_cast_fp16 = softmax(axis = var_822, x = aw_chunk_93_cast_fp16)[name = string("op_1339_cast_fp16")]; + tensor var_1340_cast_fp16 = softmax(axis = var_822, x = aw_chunk_95_cast_fp16)[name = string("op_1340_cast_fp16")]; + tensor var_1341_cast_fp16 = softmax(axis = var_822, x = aw_chunk_97_cast_fp16)[name = string("op_1341_cast_fp16")]; + tensor var_1342_cast_fp16 = softmax(axis = var_822, x = aw_chunk_99_cast_fp16)[name = string("op_1342_cast_fp16")]; + tensor var_1343_cast_fp16 = softmax(axis = var_822, x = aw_chunk_101_cast_fp16)[name = string("op_1343_cast_fp16")]; + tensor var_1344_cast_fp16 = softmax(axis = var_822, x = aw_chunk_103_cast_fp16)[name = string("op_1344_cast_fp16")]; + tensor var_1345_cast_fp16 = softmax(axis = var_822, x = aw_chunk_105_cast_fp16)[name = string("op_1345_cast_fp16")]; + tensor var_1346_cast_fp16 = softmax(axis = var_822, x = aw_chunk_107_cast_fp16)[name = string("op_1346_cast_fp16")]; + tensor var_1347_cast_fp16 = softmax(axis = var_822, x = aw_chunk_109_cast_fp16)[name = string("op_1347_cast_fp16")]; + tensor var_1348_cast_fp16 = softmax(axis = var_822, x = aw_chunk_111_cast_fp16)[name = string("op_1348_cast_fp16")]; + tensor var_1349_cast_fp16 = softmax(axis = var_822, x = aw_chunk_113_cast_fp16)[name = string("op_1349_cast_fp16")]; + tensor var_1350_cast_fp16 = softmax(axis = var_822, x = aw_chunk_115_cast_fp16)[name = string("op_1350_cast_fp16")]; + tensor var_1351_cast_fp16 = softmax(axis = var_822, x = aw_chunk_117_cast_fp16)[name = string("op_1351_cast_fp16")]; + tensor var_1352_cast_fp16 = softmax(axis = var_822, x = aw_chunk_119_cast_fp16)[name = string("op_1352_cast_fp16")]; + tensor var_1353_cast_fp16 = softmax(axis = var_822, x = aw_chunk_121_cast_fp16)[name = string("op_1353_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_822, x = aw_chunk_123_cast_fp16)[name = string("op_1354_cast_fp16")]; + tensor var_1355_cast_fp16 = softmax(axis = var_822, x = aw_chunk_125_cast_fp16)[name = string("op_1355_cast_fp16")]; + tensor var_1356_cast_fp16 = softmax(axis = var_822, x = aw_chunk_127_cast_fp16)[name = string("op_1356_cast_fp16")]; + string var_1358_equation_0 = const()[name = string("op_1358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1358_cast_fp16 = einsum(equation = var_1358_equation_0, values = (var_1166_cast_fp16, var_1325_cast_fp16))[name = string("op_1358_cast_fp16")]; + string var_1360_equation_0 = const()[name = string("op_1360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1360_cast_fp16 = einsum(equation = var_1360_equation_0, values = (var_1166_cast_fp16, var_1326_cast_fp16))[name = string("op_1360_cast_fp16")]; + string var_1362_equation_0 = const()[name = string("op_1362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1362_cast_fp16 = einsum(equation = var_1362_equation_0, values = (var_1166_cast_fp16, var_1327_cast_fp16))[name = string("op_1362_cast_fp16")]; + string var_1364_equation_0 = const()[name = string("op_1364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1364_cast_fp16 = einsum(equation = var_1364_equation_0, values = (var_1166_cast_fp16, var_1328_cast_fp16))[name = string("op_1364_cast_fp16")]; + string var_1366_equation_0 = const()[name = string("op_1366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1366_cast_fp16 = einsum(equation = var_1366_equation_0, values = (var_1170_cast_fp16, var_1329_cast_fp16))[name = string("op_1366_cast_fp16")]; + string var_1368_equation_0 = const()[name = string("op_1368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1368_cast_fp16 = einsum(equation = var_1368_equation_0, values = (var_1170_cast_fp16, var_1330_cast_fp16))[name = string("op_1368_cast_fp16")]; + string var_1370_equation_0 = const()[name = string("op_1370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1370_cast_fp16 = einsum(equation = var_1370_equation_0, values = (var_1170_cast_fp16, var_1331_cast_fp16))[name = string("op_1370_cast_fp16")]; + string var_1372_equation_0 = const()[name = string("op_1372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1372_cast_fp16 = einsum(equation = var_1372_equation_0, values = (var_1170_cast_fp16, var_1332_cast_fp16))[name = string("op_1372_cast_fp16")]; + string var_1374_equation_0 = const()[name = string("op_1374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1374_cast_fp16 = einsum(equation = var_1374_equation_0, values = (var_1174_cast_fp16, var_1333_cast_fp16))[name = string("op_1374_cast_fp16")]; + string var_1376_equation_0 = const()[name = string("op_1376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1376_cast_fp16 = einsum(equation = var_1376_equation_0, values = (var_1174_cast_fp16, var_1334_cast_fp16))[name = string("op_1376_cast_fp16")]; + string var_1378_equation_0 = const()[name = string("op_1378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1378_cast_fp16 = einsum(equation = var_1378_equation_0, values = (var_1174_cast_fp16, var_1335_cast_fp16))[name = string("op_1378_cast_fp16")]; + string var_1380_equation_0 = const()[name = string("op_1380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1380_cast_fp16 = einsum(equation = var_1380_equation_0, values = (var_1174_cast_fp16, var_1336_cast_fp16))[name = string("op_1380_cast_fp16")]; + string var_1382_equation_0 = const()[name = string("op_1382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1382_cast_fp16 = einsum(equation = var_1382_equation_0, values = (var_1178_cast_fp16, var_1337_cast_fp16))[name = string("op_1382_cast_fp16")]; + string var_1384_equation_0 = const()[name = string("op_1384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1384_cast_fp16 = einsum(equation = var_1384_equation_0, values = (var_1178_cast_fp16, var_1338_cast_fp16))[name = string("op_1384_cast_fp16")]; + string var_1386_equation_0 = const()[name = string("op_1386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1386_cast_fp16 = einsum(equation = var_1386_equation_0, values = (var_1178_cast_fp16, var_1339_cast_fp16))[name = string("op_1386_cast_fp16")]; + string var_1388_equation_0 = const()[name = string("op_1388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1388_cast_fp16 = einsum(equation = var_1388_equation_0, values = (var_1178_cast_fp16, var_1340_cast_fp16))[name = string("op_1388_cast_fp16")]; + string var_1390_equation_0 = const()[name = string("op_1390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1390_cast_fp16 = einsum(equation = var_1390_equation_0, values = (var_1182_cast_fp16, var_1341_cast_fp16))[name = string("op_1390_cast_fp16")]; + string var_1392_equation_0 = const()[name = string("op_1392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1392_cast_fp16 = einsum(equation = var_1392_equation_0, values = (var_1182_cast_fp16, var_1342_cast_fp16))[name = string("op_1392_cast_fp16")]; + string var_1394_equation_0 = const()[name = string("op_1394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1394_cast_fp16 = einsum(equation = var_1394_equation_0, values = (var_1182_cast_fp16, var_1343_cast_fp16))[name = string("op_1394_cast_fp16")]; + string var_1396_equation_0 = const()[name = string("op_1396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1396_cast_fp16 = einsum(equation = var_1396_equation_0, values = (var_1182_cast_fp16, var_1344_cast_fp16))[name = string("op_1396_cast_fp16")]; + string var_1398_equation_0 = const()[name = string("op_1398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1398_cast_fp16 = einsum(equation = var_1398_equation_0, values = (var_1186_cast_fp16, var_1345_cast_fp16))[name = string("op_1398_cast_fp16")]; + string var_1400_equation_0 = const()[name = string("op_1400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1400_cast_fp16 = einsum(equation = var_1400_equation_0, values = (var_1186_cast_fp16, var_1346_cast_fp16))[name = string("op_1400_cast_fp16")]; + string var_1402_equation_0 = const()[name = string("op_1402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1402_cast_fp16 = einsum(equation = var_1402_equation_0, values = (var_1186_cast_fp16, var_1347_cast_fp16))[name = string("op_1402_cast_fp16")]; + string var_1404_equation_0 = const()[name = string("op_1404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1404_cast_fp16 = einsum(equation = var_1404_equation_0, values = (var_1186_cast_fp16, var_1348_cast_fp16))[name = string("op_1404_cast_fp16")]; + string var_1406_equation_0 = const()[name = string("op_1406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1406_cast_fp16 = einsum(equation = var_1406_equation_0, values = (var_1190_cast_fp16, var_1349_cast_fp16))[name = string("op_1406_cast_fp16")]; + string var_1408_equation_0 = const()[name = string("op_1408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1408_cast_fp16 = einsum(equation = var_1408_equation_0, values = (var_1190_cast_fp16, var_1350_cast_fp16))[name = string("op_1408_cast_fp16")]; + string var_1410_equation_0 = const()[name = string("op_1410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1410_cast_fp16 = einsum(equation = var_1410_equation_0, values = (var_1190_cast_fp16, var_1351_cast_fp16))[name = string("op_1410_cast_fp16")]; + string var_1412_equation_0 = const()[name = string("op_1412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1412_cast_fp16 = einsum(equation = var_1412_equation_0, values = (var_1190_cast_fp16, var_1352_cast_fp16))[name = string("op_1412_cast_fp16")]; + string var_1414_equation_0 = const()[name = string("op_1414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1414_cast_fp16 = einsum(equation = var_1414_equation_0, values = (var_1194_cast_fp16, var_1353_cast_fp16))[name = string("op_1414_cast_fp16")]; + string var_1416_equation_0 = const()[name = string("op_1416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1416_cast_fp16 = einsum(equation = var_1416_equation_0, values = (var_1194_cast_fp16, var_1354_cast_fp16))[name = string("op_1416_cast_fp16")]; + string var_1418_equation_0 = const()[name = string("op_1418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1418_cast_fp16 = einsum(equation = var_1418_equation_0, values = (var_1194_cast_fp16, var_1355_cast_fp16))[name = string("op_1418_cast_fp16")]; + string var_1420_equation_0 = const()[name = string("op_1420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1420_cast_fp16 = einsum(equation = var_1420_equation_0, values = (var_1194_cast_fp16, var_1356_cast_fp16))[name = string("op_1420_cast_fp16")]; + bool var_1422_interleave_0 = const()[name = string("op_1422_interleave_0"), val = bool(false)]; + tensor var_1422_cast_fp16 = concat(axis = var_809, interleave = var_1422_interleave_0, values = (var_1358_cast_fp16, var_1360_cast_fp16, var_1362_cast_fp16, var_1364_cast_fp16))[name = string("op_1422_cast_fp16")]; + bool var_1424_interleave_0 = const()[name = string("op_1424_interleave_0"), val = bool(false)]; + tensor var_1424_cast_fp16 = concat(axis = var_809, interleave = var_1424_interleave_0, values = (var_1366_cast_fp16, var_1368_cast_fp16, var_1370_cast_fp16, var_1372_cast_fp16))[name = string("op_1424_cast_fp16")]; + bool var_1426_interleave_0 = const()[name = string("op_1426_interleave_0"), val = bool(false)]; + tensor var_1426_cast_fp16 = concat(axis = var_809, interleave = var_1426_interleave_0, values = (var_1374_cast_fp16, var_1376_cast_fp16, var_1378_cast_fp16, var_1380_cast_fp16))[name = string("op_1426_cast_fp16")]; + bool var_1428_interleave_0 = const()[name = string("op_1428_interleave_0"), val = bool(false)]; + tensor var_1428_cast_fp16 = concat(axis = var_809, interleave = var_1428_interleave_0, values = (var_1382_cast_fp16, var_1384_cast_fp16, var_1386_cast_fp16, var_1388_cast_fp16))[name = string("op_1428_cast_fp16")]; + bool var_1430_interleave_0 = const()[name = string("op_1430_interleave_0"), val = bool(false)]; + tensor var_1430_cast_fp16 = concat(axis = var_809, interleave = var_1430_interleave_0, values = (var_1390_cast_fp16, var_1392_cast_fp16, var_1394_cast_fp16, var_1396_cast_fp16))[name = string("op_1430_cast_fp16")]; + bool var_1432_interleave_0 = const()[name = string("op_1432_interleave_0"), val = bool(false)]; + tensor var_1432_cast_fp16 = concat(axis = var_809, interleave = var_1432_interleave_0, values = (var_1398_cast_fp16, var_1400_cast_fp16, var_1402_cast_fp16, var_1404_cast_fp16))[name = string("op_1432_cast_fp16")]; + bool var_1434_interleave_0 = const()[name = string("op_1434_interleave_0"), val = bool(false)]; + tensor var_1434_cast_fp16 = concat(axis = var_809, interleave = var_1434_interleave_0, values = (var_1406_cast_fp16, var_1408_cast_fp16, var_1410_cast_fp16, var_1412_cast_fp16))[name = string("op_1434_cast_fp16")]; + bool var_1436_interleave_0 = const()[name = string("op_1436_interleave_0"), val = bool(false)]; + tensor var_1436_cast_fp16 = concat(axis = var_809, interleave = var_1436_interleave_0, values = (var_1414_cast_fp16, var_1416_cast_fp16, var_1418_cast_fp16, var_1420_cast_fp16))[name = string("op_1436_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_822, interleave = input_9_interleave_0, values = (var_1422_cast_fp16, var_1424_cast_fp16, var_1426_cast_fp16, var_1428_cast_fp16, var_1430_cast_fp16, var_1432_cast_fp16, var_1434_cast_fp16, var_1436_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11241344)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11765696)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_1455_to_fp16 = const()[name = string("op_1455_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_1455_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11766784)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11767872)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11768960)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13866176)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13870336)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15967552)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_1484 = const()[name = string("op_1484"), val = int32(3)]; + int32 var_1497 = const()[name = string("op_1497"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_1514_to_fp16 = const()[name = string("op_1514_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_1514_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15968640)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15969728)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15970816)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16495168)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16496256)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17020608)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17544960)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_1552_begin_0 = const()[name = string("op_1552_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1552_end_0 = const()[name = string("op_1552_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1552_end_mask_0 = const()[name = string("op_1552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1552_cast_fp16 = slice_by_index(begin = var_1552_begin_0, end = var_1552_end_0, end_mask = var_1552_end_mask_0, x = query_5_cast_fp16)[name = string("op_1552_cast_fp16")]; + tensor var_1556_begin_0 = const()[name = string("op_1556_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1556_end_0 = const()[name = string("op_1556_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1556_end_mask_0 = const()[name = string("op_1556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1556_cast_fp16 = slice_by_index(begin = var_1556_begin_0, end = var_1556_end_0, end_mask = var_1556_end_mask_0, x = query_5_cast_fp16)[name = string("op_1556_cast_fp16")]; + tensor var_1560_begin_0 = const()[name = string("op_1560_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1560_end_0 = const()[name = string("op_1560_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1560_end_mask_0 = const()[name = string("op_1560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1560_cast_fp16 = slice_by_index(begin = var_1560_begin_0, end = var_1560_end_0, end_mask = var_1560_end_mask_0, x = query_5_cast_fp16)[name = string("op_1560_cast_fp16")]; + tensor var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = query_5_cast_fp16)[name = string("op_1564_cast_fp16")]; + tensor var_1568_begin_0 = const()[name = string("op_1568_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1568_end_0 = const()[name = string("op_1568_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1568_end_mask_0 = const()[name = string("op_1568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1568_cast_fp16 = slice_by_index(begin = var_1568_begin_0, end = var_1568_end_0, end_mask = var_1568_end_mask_0, x = query_5_cast_fp16)[name = string("op_1568_cast_fp16")]; + tensor var_1572_begin_0 = const()[name = string("op_1572_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1572_end_0 = const()[name = string("op_1572_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1572_end_mask_0 = const()[name = string("op_1572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1572_cast_fp16 = slice_by_index(begin = var_1572_begin_0, end = var_1572_end_0, end_mask = var_1572_end_mask_0, x = query_5_cast_fp16)[name = string("op_1572_cast_fp16")]; + tensor var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = query_5_cast_fp16)[name = string("op_1576_cast_fp16")]; + tensor var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = query_5_cast_fp16)[name = string("op_1580_cast_fp16")]; + tensor var_1589_begin_0 = const()[name = string("op_1589_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1589_end_mask_0 = const()[name = string("op_1589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1589_cast_fp16")]; + tensor var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1603_begin_0 = const()[name = string("op_1603_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1603_end_0 = const()[name = string("op_1603_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1603_end_mask_0 = const()[name = string("op_1603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1603_cast_fp16 = slice_by_index(begin = var_1603_begin_0, end = var_1603_end_0, end_mask = var_1603_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1603_cast_fp16")]; + tensor var_1610_begin_0 = const()[name = string("op_1610_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1610_end_0 = const()[name = string("op_1610_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1610_end_mask_0 = const()[name = string("op_1610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1610_cast_fp16 = slice_by_index(begin = var_1610_begin_0, end = var_1610_end_0, end_mask = var_1610_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1610_cast_fp16")]; + tensor var_1617_begin_0 = const()[name = string("op_1617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1617_end_0 = const()[name = string("op_1617_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1617_end_mask_0 = const()[name = string("op_1617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1617_cast_fp16 = slice_by_index(begin = var_1617_begin_0, end = var_1617_end_0, end_mask = var_1617_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1617_cast_fp16")]; + tensor var_1624_begin_0 = const()[name = string("op_1624_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1624_end_0 = const()[name = string("op_1624_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1624_end_mask_0 = const()[name = string("op_1624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1624_cast_fp16 = slice_by_index(begin = var_1624_begin_0, end = var_1624_end_0, end_mask = var_1624_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1624_cast_fp16")]; + tensor var_1631_begin_0 = const()[name = string("op_1631_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1631_end_0 = const()[name = string("op_1631_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1631_end_mask_0 = const()[name = string("op_1631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1631_cast_fp16 = slice_by_index(begin = var_1631_begin_0, end = var_1631_end_0, end_mask = var_1631_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1631_cast_fp16")]; + tensor var_1638_begin_0 = const()[name = string("op_1638_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1638_end_0 = const()[name = string("op_1638_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1638_end_mask_0 = const()[name = string("op_1638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor var_1645_begin_0 = const()[name = string("op_1645_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1645_end_0 = const()[name = string("op_1645_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1645_end_mask_0 = const()[name = string("op_1645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1645_cast_fp16 = slice_by_index(begin = var_1645_begin_0, end = var_1645_end_0, end_mask = var_1645_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1645_cast_fp16")]; + tensor var_1652_begin_0 = const()[name = string("op_1652_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1652_end_0 = const()[name = string("op_1652_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1652_end_mask_0 = const()[name = string("op_1652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1652_cast_fp16 = slice_by_index(begin = var_1652_begin_0, end = var_1652_end_0, end_mask = var_1652_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1652_cast_fp16")]; + tensor var_1659_begin_0 = const()[name = string("op_1659_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1659_end_0 = const()[name = string("op_1659_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1659_end_mask_0 = const()[name = string("op_1659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1659_cast_fp16 = slice_by_index(begin = var_1659_begin_0, end = var_1659_end_0, end_mask = var_1659_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1659_cast_fp16")]; + tensor var_1666_begin_0 = const()[name = string("op_1666_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1666_end_0 = const()[name = string("op_1666_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1666_end_mask_0 = const()[name = string("op_1666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1666_cast_fp16 = slice_by_index(begin = var_1666_begin_0, end = var_1666_end_0, end_mask = var_1666_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1666_cast_fp16")]; + tensor var_1673_begin_0 = const()[name = string("op_1673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1673_end_0 = const()[name = string("op_1673_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1673_end_mask_0 = const()[name = string("op_1673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1673_cast_fp16 = slice_by_index(begin = var_1673_begin_0, end = var_1673_end_0, end_mask = var_1673_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1673_cast_fp16")]; + tensor var_1680_begin_0 = const()[name = string("op_1680_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1680_end_0 = const()[name = string("op_1680_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1680_end_mask_0 = const()[name = string("op_1680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1680_cast_fp16 = slice_by_index(begin = var_1680_begin_0, end = var_1680_end_0, end_mask = var_1680_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1680_cast_fp16")]; + tensor var_1687_begin_0 = const()[name = string("op_1687_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1687_end_0 = const()[name = string("op_1687_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1687_end_mask_0 = const()[name = string("op_1687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1687_cast_fp16 = slice_by_index(begin = var_1687_begin_0, end = var_1687_end_0, end_mask = var_1687_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1687_cast_fp16")]; + tensor var_1694_begin_0 = const()[name = string("op_1694_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1694_end_0 = const()[name = string("op_1694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1694_end_mask_0 = const()[name = string("op_1694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1694_cast_fp16 = slice_by_index(begin = var_1694_begin_0, end = var_1694_end_0, end_mask = var_1694_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1694_cast_fp16")]; + tensor var_1701_begin_0 = const()[name = string("op_1701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1701_end_0 = const()[name = string("op_1701_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1701_end_mask_0 = const()[name = string("op_1701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1701_cast_fp16 = slice_by_index(begin = var_1701_begin_0, end = var_1701_end_0, end_mask = var_1701_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1701_cast_fp16")]; + tensor var_1708_begin_0 = const()[name = string("op_1708_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1708_end_0 = const()[name = string("op_1708_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1708_end_mask_0 = const()[name = string("op_1708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1708_cast_fp16 = slice_by_index(begin = var_1708_begin_0, end = var_1708_end_0, end_mask = var_1708_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1708_cast_fp16")]; + tensor var_1715_begin_0 = const()[name = string("op_1715_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1715_end_0 = const()[name = string("op_1715_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1715_end_mask_0 = const()[name = string("op_1715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1715_cast_fp16 = slice_by_index(begin = var_1715_begin_0, end = var_1715_end_0, end_mask = var_1715_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1715_cast_fp16")]; + tensor var_1722_begin_0 = const()[name = string("op_1722_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1722_end_0 = const()[name = string("op_1722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1722_end_mask_0 = const()[name = string("op_1722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1722_cast_fp16 = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1722_cast_fp16")]; + tensor var_1729_begin_0 = const()[name = string("op_1729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1729_end_0 = const()[name = string("op_1729_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1729_end_mask_0 = const()[name = string("op_1729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1729_cast_fp16 = slice_by_index(begin = var_1729_begin_0, end = var_1729_end_0, end_mask = var_1729_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1729_cast_fp16")]; + tensor var_1736_begin_0 = const()[name = string("op_1736_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1736_end_0 = const()[name = string("op_1736_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1736_end_mask_0 = const()[name = string("op_1736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1736_cast_fp16 = slice_by_index(begin = var_1736_begin_0, end = var_1736_end_0, end_mask = var_1736_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1736_cast_fp16")]; + tensor var_1743_begin_0 = const()[name = string("op_1743_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1743_end_0 = const()[name = string("op_1743_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1743_end_mask_0 = const()[name = string("op_1743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1743_cast_fp16 = slice_by_index(begin = var_1743_begin_0, end = var_1743_end_0, end_mask = var_1743_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1743_cast_fp16")]; + tensor var_1750_begin_0 = const()[name = string("op_1750_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1750_end_0 = const()[name = string("op_1750_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1750_end_mask_0 = const()[name = string("op_1750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1750_cast_fp16 = slice_by_index(begin = var_1750_begin_0, end = var_1750_end_0, end_mask = var_1750_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1750_cast_fp16")]; + tensor var_1757_begin_0 = const()[name = string("op_1757_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1757_end_0 = const()[name = string("op_1757_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1757_end_mask_0 = const()[name = string("op_1757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1757_cast_fp16 = slice_by_index(begin = var_1757_begin_0, end = var_1757_end_0, end_mask = var_1757_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1757_cast_fp16")]; + tensor var_1764_begin_0 = const()[name = string("op_1764_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1764_end_0 = const()[name = string("op_1764_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1764_end_mask_0 = const()[name = string("op_1764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1764_cast_fp16 = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1764_cast_fp16")]; + tensor var_1771_begin_0 = const()[name = string("op_1771_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1771_end_0 = const()[name = string("op_1771_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1771_end_mask_0 = const()[name = string("op_1771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1771_cast_fp16 = slice_by_index(begin = var_1771_begin_0, end = var_1771_end_0, end_mask = var_1771_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1771_cast_fp16")]; + tensor var_1778_begin_0 = const()[name = string("op_1778_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1778_end_0 = const()[name = string("op_1778_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1778_end_mask_0 = const()[name = string("op_1778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1778_cast_fp16 = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1778_cast_fp16")]; + tensor var_1785_begin_0 = const()[name = string("op_1785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1785_end_0 = const()[name = string("op_1785_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1785_end_mask_0 = const()[name = string("op_1785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1785_cast_fp16 = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1785_cast_fp16")]; + tensor var_1792_begin_0 = const()[name = string("op_1792_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1792_end_0 = const()[name = string("op_1792_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1792_end_mask_0 = const()[name = string("op_1792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1792_cast_fp16 = slice_by_index(begin = var_1792_begin_0, end = var_1792_end_0, end_mask = var_1792_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1792_cast_fp16")]; + tensor var_1799_begin_0 = const()[name = string("op_1799_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1799_end_0 = const()[name = string("op_1799_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1799_end_mask_0 = const()[name = string("op_1799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1799_cast_fp16")]; + tensor var_1806_begin_0 = const()[name = string("op_1806_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1806_end_0 = const()[name = string("op_1806_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1806_end_mask_0 = const()[name = string("op_1806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1806_cast_fp16 = slice_by_index(begin = var_1806_begin_0, end = var_1806_end_0, end_mask = var_1806_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1806_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1811_begin_0 = const()[name = string("op_1811_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1811_end_0 = const()[name = string("op_1811_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1811_end_mask_0 = const()[name = string("op_1811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_3")]; + tensor var_1811_cast_fp16 = slice_by_index(begin = var_1811_begin_0, end = var_1811_end_0, end_mask = var_1811_end_mask_0, x = k_5_cast_fp16)[name = string("op_1811_cast_fp16")]; + tensor var_1815_begin_0 = const()[name = string("op_1815_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1815_end_0 = const()[name = string("op_1815_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1815_end_mask_0 = const()[name = string("op_1815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1815_cast_fp16 = slice_by_index(begin = var_1815_begin_0, end = var_1815_end_0, end_mask = var_1815_end_mask_0, x = k_5_cast_fp16)[name = string("op_1815_cast_fp16")]; + tensor var_1819_begin_0 = const()[name = string("op_1819_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1819_end_0 = const()[name = string("op_1819_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1819_end_mask_0 = const()[name = string("op_1819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1819_cast_fp16 = slice_by_index(begin = var_1819_begin_0, end = var_1819_end_0, end_mask = var_1819_end_mask_0, x = k_5_cast_fp16)[name = string("op_1819_cast_fp16")]; + tensor var_1823_begin_0 = const()[name = string("op_1823_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1823_end_0 = const()[name = string("op_1823_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1823_end_mask_0 = const()[name = string("op_1823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1823_cast_fp16 = slice_by_index(begin = var_1823_begin_0, end = var_1823_end_0, end_mask = var_1823_end_mask_0, x = k_5_cast_fp16)[name = string("op_1823_cast_fp16")]; + tensor var_1827_begin_0 = const()[name = string("op_1827_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1827_end_0 = const()[name = string("op_1827_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1827_end_mask_0 = const()[name = string("op_1827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1827_cast_fp16 = slice_by_index(begin = var_1827_begin_0, end = var_1827_end_0, end_mask = var_1827_end_mask_0, x = k_5_cast_fp16)[name = string("op_1827_cast_fp16")]; + tensor var_1831_begin_0 = const()[name = string("op_1831_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1831_end_0 = const()[name = string("op_1831_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1831_end_mask_0 = const()[name = string("op_1831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1831_cast_fp16 = slice_by_index(begin = var_1831_begin_0, end = var_1831_end_0, end_mask = var_1831_end_mask_0, x = k_5_cast_fp16)[name = string("op_1831_cast_fp16")]; + tensor var_1835_begin_0 = const()[name = string("op_1835_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1835_end_0 = const()[name = string("op_1835_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1835_end_mask_0 = const()[name = string("op_1835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1835_cast_fp16 = slice_by_index(begin = var_1835_begin_0, end = var_1835_end_0, end_mask = var_1835_end_mask_0, x = k_5_cast_fp16)[name = string("op_1835_cast_fp16")]; + tensor var_1839_begin_0 = const()[name = string("op_1839_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1839_end_0 = const()[name = string("op_1839_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1839_end_mask_0 = const()[name = string("op_1839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1839_cast_fp16 = slice_by_index(begin = var_1839_begin_0, end = var_1839_end_0, end_mask = var_1839_end_mask_0, x = k_5_cast_fp16)[name = string("op_1839_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = value_5_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1845_begin_0 = const()[name = string("op_1845_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1845_end_0 = const()[name = string("op_1845_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1845_end_mask_0 = const()[name = string("op_1845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1845_cast_fp16 = slice_by_index(begin = var_1845_begin_0, end = var_1845_end_0, end_mask = var_1845_end_mask_0, x = value_5_cast_fp16)[name = string("op_1845_cast_fp16")]; + tensor var_1849_begin_0 = const()[name = string("op_1849_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1849_end_0 = const()[name = string("op_1849_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1849_end_mask_0 = const()[name = string("op_1849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1849_cast_fp16 = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, x = value_5_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor var_1853_begin_0 = const()[name = string("op_1853_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1853_end_0 = const()[name = string("op_1853_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1853_end_mask_0 = const()[name = string("op_1853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1853_cast_fp16 = slice_by_index(begin = var_1853_begin_0, end = var_1853_end_0, end_mask = var_1853_end_mask_0, x = value_5_cast_fp16)[name = string("op_1853_cast_fp16")]; + tensor var_1857_begin_0 = const()[name = string("op_1857_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1857_end_0 = const()[name = string("op_1857_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1857_end_mask_0 = const()[name = string("op_1857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1857_cast_fp16 = slice_by_index(begin = var_1857_begin_0, end = var_1857_end_0, end_mask = var_1857_end_mask_0, x = value_5_cast_fp16)[name = string("op_1857_cast_fp16")]; + tensor var_1861_begin_0 = const()[name = string("op_1861_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1861_end_0 = const()[name = string("op_1861_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1861_end_mask_0 = const()[name = string("op_1861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = value_5_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1865_begin_0 = const()[name = string("op_1865_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1865_end_0 = const()[name = string("op_1865_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1865_end_mask_0 = const()[name = string("op_1865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1865_cast_fp16 = slice_by_index(begin = var_1865_begin_0, end = var_1865_end_0, end_mask = var_1865_end_mask_0, x = value_5_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1869_begin_0 = const()[name = string("op_1869_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1869_end_0 = const()[name = string("op_1869_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1869_end_mask_0 = const()[name = string("op_1869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1869_cast_fp16 = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = value_5_cast_fp16)[name = string("op_1869_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1811_cast_fp16, var_1589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1811_cast_fp16, var_1596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1811_cast_fp16, var_1603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1811_cast_fp16, var_1610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1815_cast_fp16, var_1617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1815_cast_fp16, var_1624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1815_cast_fp16, var_1631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1815_cast_fp16, var_1638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1819_cast_fp16, var_1645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1819_cast_fp16, var_1652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1819_cast_fp16, var_1659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1819_cast_fp16, var_1666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1823_cast_fp16, var_1673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1823_cast_fp16, var_1680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1823_cast_fp16, var_1687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1823_cast_fp16, var_1694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1827_cast_fp16, var_1701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1827_cast_fp16, var_1708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1827_cast_fp16, var_1715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1827_cast_fp16, var_1722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1831_cast_fp16, var_1729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1831_cast_fp16, var_1736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1831_cast_fp16, var_1743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1831_cast_fp16, var_1750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_1835_cast_fp16, var_1757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_1835_cast_fp16, var_1764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_1835_cast_fp16, var_1771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_1835_cast_fp16, var_1778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_1839_cast_fp16, var_1785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_1839_cast_fp16, var_1792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_1839_cast_fp16, var_1799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_1839_cast_fp16, var_1806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + fp16 var_1936_to_fp16 = const()[name = string("op_1936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1936_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1938_to_fp16 = const()[name = string("op_1938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1938_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1940_to_fp16 = const()[name = string("op_1940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1940_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1942_to_fp16 = const()[name = string("op_1942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1942_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1944_to_fp16 = const()[name = string("op_1944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1944_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1946_to_fp16 = const()[name = string("op_1946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1946_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1948_to_fp16 = const()[name = string("op_1948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1948_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1950_to_fp16 = const()[name = string("op_1950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1950_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1952_to_fp16 = const()[name = string("op_1952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1952_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1954_to_fp16 = const()[name = string("op_1954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1954_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1956_to_fp16 = const()[name = string("op_1956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1956_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1958_to_fp16 = const()[name = string("op_1958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1958_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1960_to_fp16 = const()[name = string("op_1960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1960_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1962_to_fp16 = const()[name = string("op_1962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1962_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1964_to_fp16 = const()[name = string("op_1964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1964_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1966_to_fp16 = const()[name = string("op_1966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1966_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_1968_to_fp16 = const()[name = string("op_1968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_1968_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_1970_to_fp16 = const()[name = string("op_1970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_1970_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_1972_to_fp16 = const()[name = string("op_1972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_1972_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_1974_to_fp16 = const()[name = string("op_1974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_1974_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_1976_to_fp16 = const()[name = string("op_1976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_1976_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_1978_to_fp16 = const()[name = string("op_1978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_1978_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_1980_to_fp16 = const()[name = string("op_1980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_1980_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_1982_to_fp16 = const()[name = string("op_1982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_1982_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_1984_to_fp16 = const()[name = string("op_1984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_1984_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_1986_to_fp16 = const()[name = string("op_1986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_1986_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_1988_to_fp16 = const()[name = string("op_1988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_1988_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_1990_to_fp16 = const()[name = string("op_1990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_1990_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_1992_to_fp16 = const()[name = string("op_1992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_1992_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_1994_to_fp16 = const()[name = string("op_1994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_1994_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_1996_to_fp16 = const()[name = string("op_1996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_1996_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_1998_to_fp16 = const()[name = string("op_1998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_1998_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + tensor var_2000_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_129_cast_fp16)[name = string("op_2000_cast_fp16")]; + tensor var_2001_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_131_cast_fp16)[name = string("op_2001_cast_fp16")]; + tensor var_2002_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_133_cast_fp16)[name = string("op_2002_cast_fp16")]; + tensor var_2003_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_135_cast_fp16)[name = string("op_2003_cast_fp16")]; + tensor var_2004_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_137_cast_fp16)[name = string("op_2004_cast_fp16")]; + tensor var_2005_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_139_cast_fp16)[name = string("op_2005_cast_fp16")]; + tensor var_2006_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_141_cast_fp16)[name = string("op_2006_cast_fp16")]; + tensor var_2007_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_143_cast_fp16)[name = string("op_2007_cast_fp16")]; + tensor var_2008_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_145_cast_fp16)[name = string("op_2008_cast_fp16")]; + tensor var_2009_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_147_cast_fp16)[name = string("op_2009_cast_fp16")]; + tensor var_2010_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_149_cast_fp16)[name = string("op_2010_cast_fp16")]; + tensor var_2011_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_151_cast_fp16)[name = string("op_2011_cast_fp16")]; + tensor var_2012_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_153_cast_fp16)[name = string("op_2012_cast_fp16")]; + tensor var_2013_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_155_cast_fp16)[name = string("op_2013_cast_fp16")]; + tensor var_2014_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_157_cast_fp16)[name = string("op_2014_cast_fp16")]; + tensor var_2015_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_159_cast_fp16)[name = string("op_2015_cast_fp16")]; + tensor var_2016_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_161_cast_fp16)[name = string("op_2016_cast_fp16")]; + tensor var_2017_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_163_cast_fp16)[name = string("op_2017_cast_fp16")]; + tensor var_2018_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_165_cast_fp16)[name = string("op_2018_cast_fp16")]; + tensor var_2019_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_167_cast_fp16)[name = string("op_2019_cast_fp16")]; + tensor var_2020_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_169_cast_fp16)[name = string("op_2020_cast_fp16")]; + tensor var_2021_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_171_cast_fp16)[name = string("op_2021_cast_fp16")]; + tensor var_2022_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_173_cast_fp16)[name = string("op_2022_cast_fp16")]; + tensor var_2023_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_175_cast_fp16)[name = string("op_2023_cast_fp16")]; + tensor var_2024_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_177_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2025_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_179_cast_fp16)[name = string("op_2025_cast_fp16")]; + tensor var_2026_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_181_cast_fp16)[name = string("op_2026_cast_fp16")]; + tensor var_2027_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_183_cast_fp16)[name = string("op_2027_cast_fp16")]; + tensor var_2028_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_185_cast_fp16)[name = string("op_2028_cast_fp16")]; + tensor var_2029_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_187_cast_fp16)[name = string("op_2029_cast_fp16")]; + tensor var_2030_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_189_cast_fp16)[name = string("op_2030_cast_fp16")]; + tensor var_2031_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_191_cast_fp16)[name = string("op_2031_cast_fp16")]; + string var_2033_equation_0 = const()[name = string("op_2033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2033_cast_fp16 = einsum(equation = var_2033_equation_0, values = (var_1841_cast_fp16, var_2000_cast_fp16))[name = string("op_2033_cast_fp16")]; + string var_2035_equation_0 = const()[name = string("op_2035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2035_cast_fp16 = einsum(equation = var_2035_equation_0, values = (var_1841_cast_fp16, var_2001_cast_fp16))[name = string("op_2035_cast_fp16")]; + string var_2037_equation_0 = const()[name = string("op_2037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2037_cast_fp16 = einsum(equation = var_2037_equation_0, values = (var_1841_cast_fp16, var_2002_cast_fp16))[name = string("op_2037_cast_fp16")]; + string var_2039_equation_0 = const()[name = string("op_2039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2039_cast_fp16 = einsum(equation = var_2039_equation_0, values = (var_1841_cast_fp16, var_2003_cast_fp16))[name = string("op_2039_cast_fp16")]; + string var_2041_equation_0 = const()[name = string("op_2041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2041_cast_fp16 = einsum(equation = var_2041_equation_0, values = (var_1845_cast_fp16, var_2004_cast_fp16))[name = string("op_2041_cast_fp16")]; + string var_2043_equation_0 = const()[name = string("op_2043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2043_cast_fp16 = einsum(equation = var_2043_equation_0, values = (var_1845_cast_fp16, var_2005_cast_fp16))[name = string("op_2043_cast_fp16")]; + string var_2045_equation_0 = const()[name = string("op_2045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2045_cast_fp16 = einsum(equation = var_2045_equation_0, values = (var_1845_cast_fp16, var_2006_cast_fp16))[name = string("op_2045_cast_fp16")]; + string var_2047_equation_0 = const()[name = string("op_2047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2047_cast_fp16 = einsum(equation = var_2047_equation_0, values = (var_1845_cast_fp16, var_2007_cast_fp16))[name = string("op_2047_cast_fp16")]; + string var_2049_equation_0 = const()[name = string("op_2049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2049_cast_fp16 = einsum(equation = var_2049_equation_0, values = (var_1849_cast_fp16, var_2008_cast_fp16))[name = string("op_2049_cast_fp16")]; + string var_2051_equation_0 = const()[name = string("op_2051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2051_cast_fp16 = einsum(equation = var_2051_equation_0, values = (var_1849_cast_fp16, var_2009_cast_fp16))[name = string("op_2051_cast_fp16")]; + string var_2053_equation_0 = const()[name = string("op_2053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2053_cast_fp16 = einsum(equation = var_2053_equation_0, values = (var_1849_cast_fp16, var_2010_cast_fp16))[name = string("op_2053_cast_fp16")]; + string var_2055_equation_0 = const()[name = string("op_2055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2055_cast_fp16 = einsum(equation = var_2055_equation_0, values = (var_1849_cast_fp16, var_2011_cast_fp16))[name = string("op_2055_cast_fp16")]; + string var_2057_equation_0 = const()[name = string("op_2057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2057_cast_fp16 = einsum(equation = var_2057_equation_0, values = (var_1853_cast_fp16, var_2012_cast_fp16))[name = string("op_2057_cast_fp16")]; + string var_2059_equation_0 = const()[name = string("op_2059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2059_cast_fp16 = einsum(equation = var_2059_equation_0, values = (var_1853_cast_fp16, var_2013_cast_fp16))[name = string("op_2059_cast_fp16")]; + string var_2061_equation_0 = const()[name = string("op_2061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2061_cast_fp16 = einsum(equation = var_2061_equation_0, values = (var_1853_cast_fp16, var_2014_cast_fp16))[name = string("op_2061_cast_fp16")]; + string var_2063_equation_0 = const()[name = string("op_2063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2063_cast_fp16 = einsum(equation = var_2063_equation_0, values = (var_1853_cast_fp16, var_2015_cast_fp16))[name = string("op_2063_cast_fp16")]; + string var_2065_equation_0 = const()[name = string("op_2065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2065_cast_fp16 = einsum(equation = var_2065_equation_0, values = (var_1857_cast_fp16, var_2016_cast_fp16))[name = string("op_2065_cast_fp16")]; + string var_2067_equation_0 = const()[name = string("op_2067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2067_cast_fp16 = einsum(equation = var_2067_equation_0, values = (var_1857_cast_fp16, var_2017_cast_fp16))[name = string("op_2067_cast_fp16")]; + string var_2069_equation_0 = const()[name = string("op_2069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2069_cast_fp16 = einsum(equation = var_2069_equation_0, values = (var_1857_cast_fp16, var_2018_cast_fp16))[name = string("op_2069_cast_fp16")]; + string var_2071_equation_0 = const()[name = string("op_2071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2071_cast_fp16 = einsum(equation = var_2071_equation_0, values = (var_1857_cast_fp16, var_2019_cast_fp16))[name = string("op_2071_cast_fp16")]; + string var_2073_equation_0 = const()[name = string("op_2073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2073_cast_fp16 = einsum(equation = var_2073_equation_0, values = (var_1861_cast_fp16, var_2020_cast_fp16))[name = string("op_2073_cast_fp16")]; + string var_2075_equation_0 = const()[name = string("op_2075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2075_cast_fp16 = einsum(equation = var_2075_equation_0, values = (var_1861_cast_fp16, var_2021_cast_fp16))[name = string("op_2075_cast_fp16")]; + string var_2077_equation_0 = const()[name = string("op_2077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2077_cast_fp16 = einsum(equation = var_2077_equation_0, values = (var_1861_cast_fp16, var_2022_cast_fp16))[name = string("op_2077_cast_fp16")]; + string var_2079_equation_0 = const()[name = string("op_2079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2079_cast_fp16 = einsum(equation = var_2079_equation_0, values = (var_1861_cast_fp16, var_2023_cast_fp16))[name = string("op_2079_cast_fp16")]; + string var_2081_equation_0 = const()[name = string("op_2081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2081_cast_fp16 = einsum(equation = var_2081_equation_0, values = (var_1865_cast_fp16, var_2024_cast_fp16))[name = string("op_2081_cast_fp16")]; + string var_2083_equation_0 = const()[name = string("op_2083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2083_cast_fp16 = einsum(equation = var_2083_equation_0, values = (var_1865_cast_fp16, var_2025_cast_fp16))[name = string("op_2083_cast_fp16")]; + string var_2085_equation_0 = const()[name = string("op_2085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2085_cast_fp16 = einsum(equation = var_2085_equation_0, values = (var_1865_cast_fp16, var_2026_cast_fp16))[name = string("op_2085_cast_fp16")]; + string var_2087_equation_0 = const()[name = string("op_2087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2087_cast_fp16 = einsum(equation = var_2087_equation_0, values = (var_1865_cast_fp16, var_2027_cast_fp16))[name = string("op_2087_cast_fp16")]; + string var_2089_equation_0 = const()[name = string("op_2089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2089_cast_fp16 = einsum(equation = var_2089_equation_0, values = (var_1869_cast_fp16, var_2028_cast_fp16))[name = string("op_2089_cast_fp16")]; + string var_2091_equation_0 = const()[name = string("op_2091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2091_cast_fp16 = einsum(equation = var_2091_equation_0, values = (var_1869_cast_fp16, var_2029_cast_fp16))[name = string("op_2091_cast_fp16")]; + string var_2093_equation_0 = const()[name = string("op_2093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2093_cast_fp16 = einsum(equation = var_2093_equation_0, values = (var_1869_cast_fp16, var_2030_cast_fp16))[name = string("op_2093_cast_fp16")]; + string var_2095_equation_0 = const()[name = string("op_2095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2095_cast_fp16 = einsum(equation = var_2095_equation_0, values = (var_1869_cast_fp16, var_2031_cast_fp16))[name = string("op_2095_cast_fp16")]; + bool var_2097_interleave_0 = const()[name = string("op_2097_interleave_0"), val = bool(false)]; + tensor var_2097_cast_fp16 = concat(axis = var_1484, interleave = var_2097_interleave_0, values = (var_2033_cast_fp16, var_2035_cast_fp16, var_2037_cast_fp16, var_2039_cast_fp16))[name = string("op_2097_cast_fp16")]; + bool var_2099_interleave_0 = const()[name = string("op_2099_interleave_0"), val = bool(false)]; + tensor var_2099_cast_fp16 = concat(axis = var_1484, interleave = var_2099_interleave_0, values = (var_2041_cast_fp16, var_2043_cast_fp16, var_2045_cast_fp16, var_2047_cast_fp16))[name = string("op_2099_cast_fp16")]; + bool var_2101_interleave_0 = const()[name = string("op_2101_interleave_0"), val = bool(false)]; + tensor var_2101_cast_fp16 = concat(axis = var_1484, interleave = var_2101_interleave_0, values = (var_2049_cast_fp16, var_2051_cast_fp16, var_2053_cast_fp16, var_2055_cast_fp16))[name = string("op_2101_cast_fp16")]; + bool var_2103_interleave_0 = const()[name = string("op_2103_interleave_0"), val = bool(false)]; + tensor var_2103_cast_fp16 = concat(axis = var_1484, interleave = var_2103_interleave_0, values = (var_2057_cast_fp16, var_2059_cast_fp16, var_2061_cast_fp16, var_2063_cast_fp16))[name = string("op_2103_cast_fp16")]; + bool var_2105_interleave_0 = const()[name = string("op_2105_interleave_0"), val = bool(false)]; + tensor var_2105_cast_fp16 = concat(axis = var_1484, interleave = var_2105_interleave_0, values = (var_2065_cast_fp16, var_2067_cast_fp16, var_2069_cast_fp16, var_2071_cast_fp16))[name = string("op_2105_cast_fp16")]; + bool var_2107_interleave_0 = const()[name = string("op_2107_interleave_0"), val = bool(false)]; + tensor var_2107_cast_fp16 = concat(axis = var_1484, interleave = var_2107_interleave_0, values = (var_2073_cast_fp16, var_2075_cast_fp16, var_2077_cast_fp16, var_2079_cast_fp16))[name = string("op_2107_cast_fp16")]; + bool var_2109_interleave_0 = const()[name = string("op_2109_interleave_0"), val = bool(false)]; + tensor var_2109_cast_fp16 = concat(axis = var_1484, interleave = var_2109_interleave_0, values = (var_2081_cast_fp16, var_2083_cast_fp16, var_2085_cast_fp16, var_2087_cast_fp16))[name = string("op_2109_cast_fp16")]; + bool var_2111_interleave_0 = const()[name = string("op_2111_interleave_0"), val = bool(false)]; + tensor var_2111_cast_fp16 = concat(axis = var_1484, interleave = var_2111_interleave_0, values = (var_2089_cast_fp16, var_2091_cast_fp16, var_2093_cast_fp16, var_2095_cast_fp16))[name = string("op_2111_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_1497, interleave = input_17_interleave_0, values = (var_2097_cast_fp16, var_2099_cast_fp16, var_2101_cast_fp16, var_2103_cast_fp16, var_2105_cast_fp16, var_2107_cast_fp16, var_2109_cast_fp16, var_2111_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17546048)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18070400)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_2130_to_fp16 = const()[name = string("op_2130_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_2130_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18071488)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18072576)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18073664)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20170880)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20175040)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22272256)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_2159 = const()[name = string("op_2159"), val = int32(3)]; + int32 var_2172 = const()[name = string("op_2172"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_2189_to_fp16 = const()[name = string("op_2189_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_2189_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22273344)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22274432)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22275520)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22799872)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22800960)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23325312)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23849664)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_2227_begin_0 = const()[name = string("op_2227_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2227_end_0 = const()[name = string("op_2227_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2227_end_mask_0 = const()[name = string("op_2227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2227_cast_fp16 = slice_by_index(begin = var_2227_begin_0, end = var_2227_end_0, end_mask = var_2227_end_mask_0, x = query_7_cast_fp16)[name = string("op_2227_cast_fp16")]; + tensor var_2231_begin_0 = const()[name = string("op_2231_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2231_end_0 = const()[name = string("op_2231_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2231_end_mask_0 = const()[name = string("op_2231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, x = query_7_cast_fp16)[name = string("op_2231_cast_fp16")]; + tensor var_2235_begin_0 = const()[name = string("op_2235_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2235_end_0 = const()[name = string("op_2235_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2235_end_mask_0 = const()[name = string("op_2235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2235_cast_fp16 = slice_by_index(begin = var_2235_begin_0, end = var_2235_end_0, end_mask = var_2235_end_mask_0, x = query_7_cast_fp16)[name = string("op_2235_cast_fp16")]; + tensor var_2239_begin_0 = const()[name = string("op_2239_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2239_end_0 = const()[name = string("op_2239_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2239_end_mask_0 = const()[name = string("op_2239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2239_cast_fp16 = slice_by_index(begin = var_2239_begin_0, end = var_2239_end_0, end_mask = var_2239_end_mask_0, x = query_7_cast_fp16)[name = string("op_2239_cast_fp16")]; + tensor var_2243_begin_0 = const()[name = string("op_2243_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2243_end_0 = const()[name = string("op_2243_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2243_end_mask_0 = const()[name = string("op_2243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = query_7_cast_fp16)[name = string("op_2243_cast_fp16")]; + tensor var_2247_begin_0 = const()[name = string("op_2247_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2247_end_0 = const()[name = string("op_2247_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2247_end_mask_0 = const()[name = string("op_2247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2247_cast_fp16 = slice_by_index(begin = var_2247_begin_0, end = var_2247_end_0, end_mask = var_2247_end_mask_0, x = query_7_cast_fp16)[name = string("op_2247_cast_fp16")]; + tensor var_2251_begin_0 = const()[name = string("op_2251_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2251_end_0 = const()[name = string("op_2251_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2251_end_mask_0 = const()[name = string("op_2251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2251_cast_fp16 = slice_by_index(begin = var_2251_begin_0, end = var_2251_end_0, end_mask = var_2251_end_mask_0, x = query_7_cast_fp16)[name = string("op_2251_cast_fp16")]; + tensor var_2255_begin_0 = const()[name = string("op_2255_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2255_end_0 = const()[name = string("op_2255_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2255_end_mask_0 = const()[name = string("op_2255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2255_cast_fp16 = slice_by_index(begin = var_2255_begin_0, end = var_2255_end_0, end_mask = var_2255_end_mask_0, x = query_7_cast_fp16)[name = string("op_2255_cast_fp16")]; + tensor var_2264_begin_0 = const()[name = string("op_2264_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2264_end_0 = const()[name = string("op_2264_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2264_end_mask_0 = const()[name = string("op_2264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2264_cast_fp16 = slice_by_index(begin = var_2264_begin_0, end = var_2264_end_0, end_mask = var_2264_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2264_cast_fp16")]; + tensor var_2271_begin_0 = const()[name = string("op_2271_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2271_end_0 = const()[name = string("op_2271_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2271_end_mask_0 = const()[name = string("op_2271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2271_cast_fp16 = slice_by_index(begin = var_2271_begin_0, end = var_2271_end_0, end_mask = var_2271_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2271_cast_fp16")]; + tensor var_2278_begin_0 = const()[name = string("op_2278_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2278_end_0 = const()[name = string("op_2278_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2278_end_mask_0 = const()[name = string("op_2278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2278_cast_fp16 = slice_by_index(begin = var_2278_begin_0, end = var_2278_end_0, end_mask = var_2278_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2278_cast_fp16")]; + tensor var_2285_begin_0 = const()[name = string("op_2285_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2285_end_0 = const()[name = string("op_2285_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2285_end_mask_0 = const()[name = string("op_2285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2285_cast_fp16 = slice_by_index(begin = var_2285_begin_0, end = var_2285_end_0, end_mask = var_2285_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2285_cast_fp16")]; + tensor var_2292_begin_0 = const()[name = string("op_2292_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2292_end_0 = const()[name = string("op_2292_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2292_end_mask_0 = const()[name = string("op_2292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2292_cast_fp16 = slice_by_index(begin = var_2292_begin_0, end = var_2292_end_0, end_mask = var_2292_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2292_cast_fp16")]; + tensor var_2299_begin_0 = const()[name = string("op_2299_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2299_end_0 = const()[name = string("op_2299_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2299_end_mask_0 = const()[name = string("op_2299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2299_cast_fp16 = slice_by_index(begin = var_2299_begin_0, end = var_2299_end_0, end_mask = var_2299_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2299_cast_fp16")]; + tensor var_2306_begin_0 = const()[name = string("op_2306_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2306_end_0 = const()[name = string("op_2306_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2306_end_mask_0 = const()[name = string("op_2306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2306_cast_fp16 = slice_by_index(begin = var_2306_begin_0, end = var_2306_end_0, end_mask = var_2306_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2306_cast_fp16")]; + tensor var_2313_begin_0 = const()[name = string("op_2313_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2313_end_0 = const()[name = string("op_2313_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2313_end_mask_0 = const()[name = string("op_2313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2313_cast_fp16 = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2313_cast_fp16")]; + tensor var_2320_begin_0 = const()[name = string("op_2320_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2320_end_0 = const()[name = string("op_2320_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2320_end_mask_0 = const()[name = string("op_2320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2320_cast_fp16 = slice_by_index(begin = var_2320_begin_0, end = var_2320_end_0, end_mask = var_2320_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2320_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = string("op_2327_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2327_end_0 = const()[name = string("op_2327_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2327_end_mask_0 = const()[name = string("op_2327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2327_cast_fp16")]; + tensor var_2334_begin_0 = const()[name = string("op_2334_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2334_end_0 = const()[name = string("op_2334_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2334_end_mask_0 = const()[name = string("op_2334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2334_cast_fp16 = slice_by_index(begin = var_2334_begin_0, end = var_2334_end_0, end_mask = var_2334_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2334_cast_fp16")]; + tensor var_2341_begin_0 = const()[name = string("op_2341_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2341_end_0 = const()[name = string("op_2341_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2341_end_mask_0 = const()[name = string("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2341_cast_fp16 = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2341_cast_fp16")]; + tensor var_2348_begin_0 = const()[name = string("op_2348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2348_end_0 = const()[name = string("op_2348_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2348_end_mask_0 = const()[name = string("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2348_cast_fp16 = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2348_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = string("op_2355_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2355_end_0 = const()[name = string("op_2355_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2355_end_mask_0 = const()[name = string("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2355_cast_fp16")]; + tensor var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2362_cast_fp16")]; + tensor var_2369_begin_0 = const()[name = string("op_2369_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2369_end_0 = const()[name = string("op_2369_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2369_end_mask_0 = const()[name = string("op_2369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2369_cast_fp16 = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2369_cast_fp16")]; + tensor var_2376_begin_0 = const()[name = string("op_2376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2376_end_0 = const()[name = string("op_2376_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2376_end_mask_0 = const()[name = string("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2376_cast_fp16 = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2376_cast_fp16")]; + tensor var_2383_begin_0 = const()[name = string("op_2383_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2383_end_0 = const()[name = string("op_2383_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2383_end_mask_0 = const()[name = string("op_2383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2383_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = string("op_2390_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2390_end_0 = const()[name = string("op_2390_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2390_end_mask_0 = const()[name = string("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2390_cast_fp16")]; + tensor var_2397_begin_0 = const()[name = string("op_2397_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2397_end_0 = const()[name = string("op_2397_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2397_end_mask_0 = const()[name = string("op_2397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2397_cast_fp16 = slice_by_index(begin = var_2397_begin_0, end = var_2397_end_0, end_mask = var_2397_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2397_cast_fp16")]; + tensor var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2404_cast_fp16")]; + tensor var_2411_begin_0 = const()[name = string("op_2411_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2411_end_0 = const()[name = string("op_2411_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2411_end_mask_0 = const()[name = string("op_2411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2411_cast_fp16 = slice_by_index(begin = var_2411_begin_0, end = var_2411_end_0, end_mask = var_2411_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2411_cast_fp16")]; + tensor var_2418_begin_0 = const()[name = string("op_2418_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2418_end_0 = const()[name = string("op_2418_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2418_end_mask_0 = const()[name = string("op_2418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2418_cast_fp16")]; + tensor var_2425_begin_0 = const()[name = string("op_2425_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2425_end_0 = const()[name = string("op_2425_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2425_end_mask_0 = const()[name = string("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2425_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2439_begin_0 = const()[name = string("op_2439_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2439_end_0 = const()[name = string("op_2439_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2439_end_mask_0 = const()[name = string("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2439_cast_fp16")]; + tensor var_2446_begin_0 = const()[name = string("op_2446_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2446_end_0 = const()[name = string("op_2446_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2446_end_mask_0 = const()[name = string("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2446_cast_fp16")]; + tensor var_2453_begin_0 = const()[name = string("op_2453_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2453_end_0 = const()[name = string("op_2453_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2453_end_mask_0 = const()[name = string("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2453_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2467_begin_0 = const()[name = string("op_2467_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2467_end_0 = const()[name = string("op_2467_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2467_end_mask_0 = const()[name = string("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2467_cast_fp16")]; + tensor var_2474_begin_0 = const()[name = string("op_2474_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2474_end_0 = const()[name = string("op_2474_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2474_end_mask_0 = const()[name = string("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2474_cast_fp16")]; + tensor var_2481_begin_0 = const()[name = string("op_2481_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2481_end_0 = const()[name = string("op_2481_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2481_end_mask_0 = const()[name = string("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2481_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2486_begin_0 = const()[name = string("op_2486_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2486_end_0 = const()[name = string("op_2486_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2486_end_mask_0 = const()[name = string("op_2486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_2")]; + tensor var_2486_cast_fp16 = slice_by_index(begin = var_2486_begin_0, end = var_2486_end_0, end_mask = var_2486_end_mask_0, x = k_7_cast_fp16)[name = string("op_2486_cast_fp16")]; + tensor var_2490_begin_0 = const()[name = string("op_2490_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2490_end_0 = const()[name = string("op_2490_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2490_end_mask_0 = const()[name = string("op_2490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2490_cast_fp16 = slice_by_index(begin = var_2490_begin_0, end = var_2490_end_0, end_mask = var_2490_end_mask_0, x = k_7_cast_fp16)[name = string("op_2490_cast_fp16")]; + tensor var_2494_begin_0 = const()[name = string("op_2494_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2494_end_0 = const()[name = string("op_2494_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2494_end_mask_0 = const()[name = string("op_2494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2494_cast_fp16 = slice_by_index(begin = var_2494_begin_0, end = var_2494_end_0, end_mask = var_2494_end_mask_0, x = k_7_cast_fp16)[name = string("op_2494_cast_fp16")]; + tensor var_2498_begin_0 = const()[name = string("op_2498_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2498_end_0 = const()[name = string("op_2498_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2498_end_mask_0 = const()[name = string("op_2498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2498_cast_fp16 = slice_by_index(begin = var_2498_begin_0, end = var_2498_end_0, end_mask = var_2498_end_mask_0, x = k_7_cast_fp16)[name = string("op_2498_cast_fp16")]; + tensor var_2502_begin_0 = const()[name = string("op_2502_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2502_end_0 = const()[name = string("op_2502_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2502_end_mask_0 = const()[name = string("op_2502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = k_7_cast_fp16)[name = string("op_2502_cast_fp16")]; + tensor var_2506_begin_0 = const()[name = string("op_2506_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2506_end_0 = const()[name = string("op_2506_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2506_end_mask_0 = const()[name = string("op_2506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2506_cast_fp16 = slice_by_index(begin = var_2506_begin_0, end = var_2506_end_0, end_mask = var_2506_end_mask_0, x = k_7_cast_fp16)[name = string("op_2506_cast_fp16")]; + tensor var_2510_begin_0 = const()[name = string("op_2510_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2510_end_0 = const()[name = string("op_2510_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2510_end_mask_0 = const()[name = string("op_2510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2510_cast_fp16 = slice_by_index(begin = var_2510_begin_0, end = var_2510_end_0, end_mask = var_2510_end_mask_0, x = k_7_cast_fp16)[name = string("op_2510_cast_fp16")]; + tensor var_2514_begin_0 = const()[name = string("op_2514_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2514_end_0 = const()[name = string("op_2514_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2514_end_mask_0 = const()[name = string("op_2514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2514_cast_fp16 = slice_by_index(begin = var_2514_begin_0, end = var_2514_end_0, end_mask = var_2514_end_mask_0, x = k_7_cast_fp16)[name = string("op_2514_cast_fp16")]; + tensor var_2516_begin_0 = const()[name = string("op_2516_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2516_end_0 = const()[name = string("op_2516_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2516_end_mask_0 = const()[name = string("op_2516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2516_cast_fp16 = slice_by_index(begin = var_2516_begin_0, end = var_2516_end_0, end_mask = var_2516_end_mask_0, x = value_7_cast_fp16)[name = string("op_2516_cast_fp16")]; + tensor var_2520_begin_0 = const()[name = string("op_2520_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2520_end_0 = const()[name = string("op_2520_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2520_end_mask_0 = const()[name = string("op_2520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2520_cast_fp16 = slice_by_index(begin = var_2520_begin_0, end = var_2520_end_0, end_mask = var_2520_end_mask_0, x = value_7_cast_fp16)[name = string("op_2520_cast_fp16")]; + tensor var_2524_begin_0 = const()[name = string("op_2524_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2524_end_0 = const()[name = string("op_2524_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2524_end_mask_0 = const()[name = string("op_2524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2524_cast_fp16 = slice_by_index(begin = var_2524_begin_0, end = var_2524_end_0, end_mask = var_2524_end_mask_0, x = value_7_cast_fp16)[name = string("op_2524_cast_fp16")]; + tensor var_2528_begin_0 = const()[name = string("op_2528_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2528_end_0 = const()[name = string("op_2528_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2528_end_mask_0 = const()[name = string("op_2528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2528_cast_fp16 = slice_by_index(begin = var_2528_begin_0, end = var_2528_end_0, end_mask = var_2528_end_mask_0, x = value_7_cast_fp16)[name = string("op_2528_cast_fp16")]; + tensor var_2532_begin_0 = const()[name = string("op_2532_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2532_end_0 = const()[name = string("op_2532_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2532_end_mask_0 = const()[name = string("op_2532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2532_cast_fp16 = slice_by_index(begin = var_2532_begin_0, end = var_2532_end_0, end_mask = var_2532_end_mask_0, x = value_7_cast_fp16)[name = string("op_2532_cast_fp16")]; + tensor var_2536_begin_0 = const()[name = string("op_2536_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2536_end_0 = const()[name = string("op_2536_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2536_end_mask_0 = const()[name = string("op_2536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2536_cast_fp16 = slice_by_index(begin = var_2536_begin_0, end = var_2536_end_0, end_mask = var_2536_end_mask_0, x = value_7_cast_fp16)[name = string("op_2536_cast_fp16")]; + tensor var_2540_begin_0 = const()[name = string("op_2540_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2540_end_0 = const()[name = string("op_2540_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2540_end_mask_0 = const()[name = string("op_2540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2540_cast_fp16 = slice_by_index(begin = var_2540_begin_0, end = var_2540_end_0, end_mask = var_2540_end_mask_0, x = value_7_cast_fp16)[name = string("op_2540_cast_fp16")]; + tensor var_2544_begin_0 = const()[name = string("op_2544_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2544_end_0 = const()[name = string("op_2544_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2544_end_mask_0 = const()[name = string("op_2544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2544_cast_fp16 = slice_by_index(begin = var_2544_begin_0, end = var_2544_end_0, end_mask = var_2544_end_mask_0, x = value_7_cast_fp16)[name = string("op_2544_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2486_cast_fp16, var_2264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2486_cast_fp16, var_2271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2486_cast_fp16, var_2278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2486_cast_fp16, var_2285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2490_cast_fp16, var_2292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2490_cast_fp16, var_2299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2490_cast_fp16, var_2306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2490_cast_fp16, var_2313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2494_cast_fp16, var_2320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2494_cast_fp16, var_2327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2494_cast_fp16, var_2334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2494_cast_fp16, var_2341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2498_cast_fp16, var_2348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2498_cast_fp16, var_2355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2498_cast_fp16, var_2362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2498_cast_fp16, var_2369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2502_cast_fp16, var_2376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2502_cast_fp16, var_2383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2502_cast_fp16, var_2390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2502_cast_fp16, var_2397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2506_cast_fp16, var_2404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2506_cast_fp16, var_2411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2506_cast_fp16, var_2418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2506_cast_fp16, var_2425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2510_cast_fp16, var_2432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2510_cast_fp16, var_2439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2510_cast_fp16, var_2446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2510_cast_fp16, var_2453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2514_cast_fp16, var_2460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2514_cast_fp16, var_2467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2514_cast_fp16, var_2474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2514_cast_fp16, var_2481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + fp16 var_2611_to_fp16 = const()[name = string("op_2611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2611_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2613_to_fp16 = const()[name = string("op_2613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2613_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2615_to_fp16 = const()[name = string("op_2615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2615_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2617_to_fp16 = const()[name = string("op_2617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2617_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2619_to_fp16 = const()[name = string("op_2619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2619_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2621_to_fp16 = const()[name = string("op_2621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2621_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2623_to_fp16 = const()[name = string("op_2623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2623_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2625_to_fp16 = const()[name = string("op_2625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2625_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2627_to_fp16 = const()[name = string("op_2627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2627_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2629_to_fp16 = const()[name = string("op_2629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2629_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2631_to_fp16 = const()[name = string("op_2631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2631_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2633_to_fp16 = const()[name = string("op_2633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2633_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2635_to_fp16 = const()[name = string("op_2635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2635_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2637_to_fp16 = const()[name = string("op_2637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2637_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2639_to_fp16 = const()[name = string("op_2639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2639_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2641_to_fp16 = const()[name = string("op_2641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2641_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2643_to_fp16 = const()[name = string("op_2643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2643_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2645_to_fp16 = const()[name = string("op_2645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2645_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2647_to_fp16 = const()[name = string("op_2647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2647_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2649_to_fp16 = const()[name = string("op_2649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2649_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2651_to_fp16 = const()[name = string("op_2651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2651_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2653_to_fp16 = const()[name = string("op_2653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2653_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2655_to_fp16 = const()[name = string("op_2655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2655_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2657_to_fp16 = const()[name = string("op_2657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2657_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2659_to_fp16 = const()[name = string("op_2659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2659_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2661_to_fp16 = const()[name = string("op_2661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2661_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2663_to_fp16 = const()[name = string("op_2663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2663_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2665_to_fp16 = const()[name = string("op_2665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2665_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2667_to_fp16 = const()[name = string("op_2667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2667_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2669_to_fp16 = const()[name = string("op_2669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2669_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2671_to_fp16 = const()[name = string("op_2671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2671_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2673_to_fp16 = const()[name = string("op_2673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2673_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + tensor var_2675_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_193_cast_fp16)[name = string("op_2675_cast_fp16")]; + tensor var_2676_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_195_cast_fp16)[name = string("op_2676_cast_fp16")]; + tensor var_2677_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_197_cast_fp16)[name = string("op_2677_cast_fp16")]; + tensor var_2678_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_199_cast_fp16)[name = string("op_2678_cast_fp16")]; + tensor var_2679_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_201_cast_fp16)[name = string("op_2679_cast_fp16")]; + tensor var_2680_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_203_cast_fp16)[name = string("op_2680_cast_fp16")]; + tensor var_2681_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_205_cast_fp16)[name = string("op_2681_cast_fp16")]; + tensor var_2682_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_207_cast_fp16)[name = string("op_2682_cast_fp16")]; + tensor var_2683_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_209_cast_fp16)[name = string("op_2683_cast_fp16")]; + tensor var_2684_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_211_cast_fp16)[name = string("op_2684_cast_fp16")]; + tensor var_2685_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_213_cast_fp16)[name = string("op_2685_cast_fp16")]; + tensor var_2686_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_215_cast_fp16)[name = string("op_2686_cast_fp16")]; + tensor var_2687_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_217_cast_fp16)[name = string("op_2687_cast_fp16")]; + tensor var_2688_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_219_cast_fp16)[name = string("op_2688_cast_fp16")]; + tensor var_2689_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_221_cast_fp16)[name = string("op_2689_cast_fp16")]; + tensor var_2690_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_223_cast_fp16)[name = string("op_2690_cast_fp16")]; + tensor var_2691_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_225_cast_fp16)[name = string("op_2691_cast_fp16")]; + tensor var_2692_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_227_cast_fp16)[name = string("op_2692_cast_fp16")]; + tensor var_2693_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_229_cast_fp16)[name = string("op_2693_cast_fp16")]; + tensor var_2694_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_231_cast_fp16)[name = string("op_2694_cast_fp16")]; + tensor var_2695_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_233_cast_fp16)[name = string("op_2695_cast_fp16")]; + tensor var_2696_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_235_cast_fp16)[name = string("op_2696_cast_fp16")]; + tensor var_2697_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_237_cast_fp16)[name = string("op_2697_cast_fp16")]; + tensor var_2698_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_239_cast_fp16)[name = string("op_2698_cast_fp16")]; + tensor var_2699_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_241_cast_fp16)[name = string("op_2699_cast_fp16")]; + tensor var_2700_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_243_cast_fp16)[name = string("op_2700_cast_fp16")]; + tensor var_2701_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_245_cast_fp16)[name = string("op_2701_cast_fp16")]; + tensor var_2702_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_247_cast_fp16)[name = string("op_2702_cast_fp16")]; + tensor var_2703_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_249_cast_fp16)[name = string("op_2703_cast_fp16")]; + tensor var_2704_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_251_cast_fp16)[name = string("op_2704_cast_fp16")]; + tensor var_2705_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_253_cast_fp16)[name = string("op_2705_cast_fp16")]; + tensor var_2706_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_255_cast_fp16)[name = string("op_2706_cast_fp16")]; + string var_2708_equation_0 = const()[name = string("op_2708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2708_cast_fp16 = einsum(equation = var_2708_equation_0, values = (var_2516_cast_fp16, var_2675_cast_fp16))[name = string("op_2708_cast_fp16")]; + string var_2710_equation_0 = const()[name = string("op_2710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2710_cast_fp16 = einsum(equation = var_2710_equation_0, values = (var_2516_cast_fp16, var_2676_cast_fp16))[name = string("op_2710_cast_fp16")]; + string var_2712_equation_0 = const()[name = string("op_2712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2712_cast_fp16 = einsum(equation = var_2712_equation_0, values = (var_2516_cast_fp16, var_2677_cast_fp16))[name = string("op_2712_cast_fp16")]; + string var_2714_equation_0 = const()[name = string("op_2714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2714_cast_fp16 = einsum(equation = var_2714_equation_0, values = (var_2516_cast_fp16, var_2678_cast_fp16))[name = string("op_2714_cast_fp16")]; + string var_2716_equation_0 = const()[name = string("op_2716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2716_cast_fp16 = einsum(equation = var_2716_equation_0, values = (var_2520_cast_fp16, var_2679_cast_fp16))[name = string("op_2716_cast_fp16")]; + string var_2718_equation_0 = const()[name = string("op_2718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2718_cast_fp16 = einsum(equation = var_2718_equation_0, values = (var_2520_cast_fp16, var_2680_cast_fp16))[name = string("op_2718_cast_fp16")]; + string var_2720_equation_0 = const()[name = string("op_2720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2720_cast_fp16 = einsum(equation = var_2720_equation_0, values = (var_2520_cast_fp16, var_2681_cast_fp16))[name = string("op_2720_cast_fp16")]; + string var_2722_equation_0 = const()[name = string("op_2722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2722_cast_fp16 = einsum(equation = var_2722_equation_0, values = (var_2520_cast_fp16, var_2682_cast_fp16))[name = string("op_2722_cast_fp16")]; + string var_2724_equation_0 = const()[name = string("op_2724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2724_cast_fp16 = einsum(equation = var_2724_equation_0, values = (var_2524_cast_fp16, var_2683_cast_fp16))[name = string("op_2724_cast_fp16")]; + string var_2726_equation_0 = const()[name = string("op_2726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2726_cast_fp16 = einsum(equation = var_2726_equation_0, values = (var_2524_cast_fp16, var_2684_cast_fp16))[name = string("op_2726_cast_fp16")]; + string var_2728_equation_0 = const()[name = string("op_2728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2728_cast_fp16 = einsum(equation = var_2728_equation_0, values = (var_2524_cast_fp16, var_2685_cast_fp16))[name = string("op_2728_cast_fp16")]; + string var_2730_equation_0 = const()[name = string("op_2730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2730_cast_fp16 = einsum(equation = var_2730_equation_0, values = (var_2524_cast_fp16, var_2686_cast_fp16))[name = string("op_2730_cast_fp16")]; + string var_2732_equation_0 = const()[name = string("op_2732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2732_cast_fp16 = einsum(equation = var_2732_equation_0, values = (var_2528_cast_fp16, var_2687_cast_fp16))[name = string("op_2732_cast_fp16")]; + string var_2734_equation_0 = const()[name = string("op_2734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2734_cast_fp16 = einsum(equation = var_2734_equation_0, values = (var_2528_cast_fp16, var_2688_cast_fp16))[name = string("op_2734_cast_fp16")]; + string var_2736_equation_0 = const()[name = string("op_2736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2736_cast_fp16 = einsum(equation = var_2736_equation_0, values = (var_2528_cast_fp16, var_2689_cast_fp16))[name = string("op_2736_cast_fp16")]; + string var_2738_equation_0 = const()[name = string("op_2738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2738_cast_fp16 = einsum(equation = var_2738_equation_0, values = (var_2528_cast_fp16, var_2690_cast_fp16))[name = string("op_2738_cast_fp16")]; + string var_2740_equation_0 = const()[name = string("op_2740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2740_cast_fp16 = einsum(equation = var_2740_equation_0, values = (var_2532_cast_fp16, var_2691_cast_fp16))[name = string("op_2740_cast_fp16")]; + string var_2742_equation_0 = const()[name = string("op_2742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2742_cast_fp16 = einsum(equation = var_2742_equation_0, values = (var_2532_cast_fp16, var_2692_cast_fp16))[name = string("op_2742_cast_fp16")]; + string var_2744_equation_0 = const()[name = string("op_2744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2744_cast_fp16 = einsum(equation = var_2744_equation_0, values = (var_2532_cast_fp16, var_2693_cast_fp16))[name = string("op_2744_cast_fp16")]; + string var_2746_equation_0 = const()[name = string("op_2746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2746_cast_fp16 = einsum(equation = var_2746_equation_0, values = (var_2532_cast_fp16, var_2694_cast_fp16))[name = string("op_2746_cast_fp16")]; + string var_2748_equation_0 = const()[name = string("op_2748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2748_cast_fp16 = einsum(equation = var_2748_equation_0, values = (var_2536_cast_fp16, var_2695_cast_fp16))[name = string("op_2748_cast_fp16")]; + string var_2750_equation_0 = const()[name = string("op_2750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2750_cast_fp16 = einsum(equation = var_2750_equation_0, values = (var_2536_cast_fp16, var_2696_cast_fp16))[name = string("op_2750_cast_fp16")]; + string var_2752_equation_0 = const()[name = string("op_2752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2752_cast_fp16 = einsum(equation = var_2752_equation_0, values = (var_2536_cast_fp16, var_2697_cast_fp16))[name = string("op_2752_cast_fp16")]; + string var_2754_equation_0 = const()[name = string("op_2754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2754_cast_fp16 = einsum(equation = var_2754_equation_0, values = (var_2536_cast_fp16, var_2698_cast_fp16))[name = string("op_2754_cast_fp16")]; + string var_2756_equation_0 = const()[name = string("op_2756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2756_cast_fp16 = einsum(equation = var_2756_equation_0, values = (var_2540_cast_fp16, var_2699_cast_fp16))[name = string("op_2756_cast_fp16")]; + string var_2758_equation_0 = const()[name = string("op_2758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2758_cast_fp16 = einsum(equation = var_2758_equation_0, values = (var_2540_cast_fp16, var_2700_cast_fp16))[name = string("op_2758_cast_fp16")]; + string var_2760_equation_0 = const()[name = string("op_2760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2760_cast_fp16 = einsum(equation = var_2760_equation_0, values = (var_2540_cast_fp16, var_2701_cast_fp16))[name = string("op_2760_cast_fp16")]; + string var_2762_equation_0 = const()[name = string("op_2762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2762_cast_fp16 = einsum(equation = var_2762_equation_0, values = (var_2540_cast_fp16, var_2702_cast_fp16))[name = string("op_2762_cast_fp16")]; + string var_2764_equation_0 = const()[name = string("op_2764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2764_cast_fp16 = einsum(equation = var_2764_equation_0, values = (var_2544_cast_fp16, var_2703_cast_fp16))[name = string("op_2764_cast_fp16")]; + string var_2766_equation_0 = const()[name = string("op_2766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2766_cast_fp16 = einsum(equation = var_2766_equation_0, values = (var_2544_cast_fp16, var_2704_cast_fp16))[name = string("op_2766_cast_fp16")]; + string var_2768_equation_0 = const()[name = string("op_2768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2768_cast_fp16 = einsum(equation = var_2768_equation_0, values = (var_2544_cast_fp16, var_2705_cast_fp16))[name = string("op_2768_cast_fp16")]; + string var_2770_equation_0 = const()[name = string("op_2770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2770_cast_fp16 = einsum(equation = var_2770_equation_0, values = (var_2544_cast_fp16, var_2706_cast_fp16))[name = string("op_2770_cast_fp16")]; + bool var_2772_interleave_0 = const()[name = string("op_2772_interleave_0"), val = bool(false)]; + tensor var_2772_cast_fp16 = concat(axis = var_2159, interleave = var_2772_interleave_0, values = (var_2708_cast_fp16, var_2710_cast_fp16, var_2712_cast_fp16, var_2714_cast_fp16))[name = string("op_2772_cast_fp16")]; + bool var_2774_interleave_0 = const()[name = string("op_2774_interleave_0"), val = bool(false)]; + tensor var_2774_cast_fp16 = concat(axis = var_2159, interleave = var_2774_interleave_0, values = (var_2716_cast_fp16, var_2718_cast_fp16, var_2720_cast_fp16, var_2722_cast_fp16))[name = string("op_2774_cast_fp16")]; + bool var_2776_interleave_0 = const()[name = string("op_2776_interleave_0"), val = bool(false)]; + tensor var_2776_cast_fp16 = concat(axis = var_2159, interleave = var_2776_interleave_0, values = (var_2724_cast_fp16, var_2726_cast_fp16, var_2728_cast_fp16, var_2730_cast_fp16))[name = string("op_2776_cast_fp16")]; + bool var_2778_interleave_0 = const()[name = string("op_2778_interleave_0"), val = bool(false)]; + tensor var_2778_cast_fp16 = concat(axis = var_2159, interleave = var_2778_interleave_0, values = (var_2732_cast_fp16, var_2734_cast_fp16, var_2736_cast_fp16, var_2738_cast_fp16))[name = string("op_2778_cast_fp16")]; + bool var_2780_interleave_0 = const()[name = string("op_2780_interleave_0"), val = bool(false)]; + tensor var_2780_cast_fp16 = concat(axis = var_2159, interleave = var_2780_interleave_0, values = (var_2740_cast_fp16, var_2742_cast_fp16, var_2744_cast_fp16, var_2746_cast_fp16))[name = string("op_2780_cast_fp16")]; + bool var_2782_interleave_0 = const()[name = string("op_2782_interleave_0"), val = bool(false)]; + tensor var_2782_cast_fp16 = concat(axis = var_2159, interleave = var_2782_interleave_0, values = (var_2748_cast_fp16, var_2750_cast_fp16, var_2752_cast_fp16, var_2754_cast_fp16))[name = string("op_2782_cast_fp16")]; + bool var_2784_interleave_0 = const()[name = string("op_2784_interleave_0"), val = bool(false)]; + tensor var_2784_cast_fp16 = concat(axis = var_2159, interleave = var_2784_interleave_0, values = (var_2756_cast_fp16, var_2758_cast_fp16, var_2760_cast_fp16, var_2762_cast_fp16))[name = string("op_2784_cast_fp16")]; + bool var_2786_interleave_0 = const()[name = string("op_2786_interleave_0"), val = bool(false)]; + tensor var_2786_cast_fp16 = concat(axis = var_2159, interleave = var_2786_interleave_0, values = (var_2764_cast_fp16, var_2766_cast_fp16, var_2768_cast_fp16, var_2770_cast_fp16))[name = string("op_2786_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_2172, interleave = input_25_interleave_0, values = (var_2772_cast_fp16, var_2774_cast_fp16, var_2776_cast_fp16, var_2778_cast_fp16, var_2780_cast_fp16, var_2782_cast_fp16, var_2784_cast_fp16, var_2786_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23850752)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24375104)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_2805_to_fp16 = const()[name = string("op_2805_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_2805_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24376192)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24377280)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24378368)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26475584)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26479744)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28576960)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_2834 = const()[name = string("op_2834"), val = int32(3)]; + int32 var_2847 = const()[name = string("op_2847"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_2864_to_fp16 = const()[name = string("op_2864_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_2864_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28578048)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28579136)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28580224)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29104576)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29105664)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29630016)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30154368)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_2902_begin_0 = const()[name = string("op_2902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2902_end_0 = const()[name = string("op_2902_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2902_end_mask_0 = const()[name = string("op_2902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2902_cast_fp16 = slice_by_index(begin = var_2902_begin_0, end = var_2902_end_0, end_mask = var_2902_end_mask_0, x = query_9_cast_fp16)[name = string("op_2902_cast_fp16")]; + tensor var_2906_begin_0 = const()[name = string("op_2906_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2906_end_0 = const()[name = string("op_2906_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2906_end_mask_0 = const()[name = string("op_2906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2906_cast_fp16 = slice_by_index(begin = var_2906_begin_0, end = var_2906_end_0, end_mask = var_2906_end_mask_0, x = query_9_cast_fp16)[name = string("op_2906_cast_fp16")]; + tensor var_2910_begin_0 = const()[name = string("op_2910_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2910_end_0 = const()[name = string("op_2910_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2910_end_mask_0 = const()[name = string("op_2910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2910_cast_fp16 = slice_by_index(begin = var_2910_begin_0, end = var_2910_end_0, end_mask = var_2910_end_mask_0, x = query_9_cast_fp16)[name = string("op_2910_cast_fp16")]; + tensor var_2914_begin_0 = const()[name = string("op_2914_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2914_end_0 = const()[name = string("op_2914_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2914_end_mask_0 = const()[name = string("op_2914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2914_cast_fp16 = slice_by_index(begin = var_2914_begin_0, end = var_2914_end_0, end_mask = var_2914_end_mask_0, x = query_9_cast_fp16)[name = string("op_2914_cast_fp16")]; + tensor var_2918_begin_0 = const()[name = string("op_2918_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2918_end_0 = const()[name = string("op_2918_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2918_end_mask_0 = const()[name = string("op_2918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2918_cast_fp16 = slice_by_index(begin = var_2918_begin_0, end = var_2918_end_0, end_mask = var_2918_end_mask_0, x = query_9_cast_fp16)[name = string("op_2918_cast_fp16")]; + tensor var_2922_begin_0 = const()[name = string("op_2922_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2922_end_0 = const()[name = string("op_2922_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2922_end_mask_0 = const()[name = string("op_2922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2922_cast_fp16 = slice_by_index(begin = var_2922_begin_0, end = var_2922_end_0, end_mask = var_2922_end_mask_0, x = query_9_cast_fp16)[name = string("op_2922_cast_fp16")]; + tensor var_2926_begin_0 = const()[name = string("op_2926_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2926_end_0 = const()[name = string("op_2926_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2926_end_mask_0 = const()[name = string("op_2926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2926_cast_fp16 = slice_by_index(begin = var_2926_begin_0, end = var_2926_end_0, end_mask = var_2926_end_mask_0, x = query_9_cast_fp16)[name = string("op_2926_cast_fp16")]; + tensor var_2930_begin_0 = const()[name = string("op_2930_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2930_end_0 = const()[name = string("op_2930_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2930_end_mask_0 = const()[name = string("op_2930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2930_cast_fp16 = slice_by_index(begin = var_2930_begin_0, end = var_2930_end_0, end_mask = var_2930_end_mask_0, x = query_9_cast_fp16)[name = string("op_2930_cast_fp16")]; + tensor var_2939_begin_0 = const()[name = string("op_2939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2939_end_0 = const()[name = string("op_2939_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2939_end_mask_0 = const()[name = string("op_2939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2939_cast_fp16 = slice_by_index(begin = var_2939_begin_0, end = var_2939_end_0, end_mask = var_2939_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2939_cast_fp16")]; + tensor var_2946_begin_0 = const()[name = string("op_2946_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2946_end_0 = const()[name = string("op_2946_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2946_end_mask_0 = const()[name = string("op_2946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2946_cast_fp16 = slice_by_index(begin = var_2946_begin_0, end = var_2946_end_0, end_mask = var_2946_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2946_cast_fp16")]; + tensor var_2953_begin_0 = const()[name = string("op_2953_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2953_end_0 = const()[name = string("op_2953_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2953_end_mask_0 = const()[name = string("op_2953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2953_cast_fp16 = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2960_begin_0 = const()[name = string("op_2960_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2960_end_0 = const()[name = string("op_2960_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2960_end_mask_0 = const()[name = string("op_2960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2960_cast_fp16 = slice_by_index(begin = var_2960_begin_0, end = var_2960_end_0, end_mask = var_2960_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2960_cast_fp16")]; + tensor var_2967_begin_0 = const()[name = string("op_2967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2967_end_0 = const()[name = string("op_2967_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2967_end_mask_0 = const()[name = string("op_2967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2967_cast_fp16 = slice_by_index(begin = var_2967_begin_0, end = var_2967_end_0, end_mask = var_2967_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2967_cast_fp16")]; + tensor var_2974_begin_0 = const()[name = string("op_2974_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2974_end_0 = const()[name = string("op_2974_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2974_end_mask_0 = const()[name = string("op_2974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2974_cast_fp16 = slice_by_index(begin = var_2974_begin_0, end = var_2974_end_0, end_mask = var_2974_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2974_cast_fp16")]; + tensor var_2981_begin_0 = const()[name = string("op_2981_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2981_end_0 = const()[name = string("op_2981_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2981_end_mask_0 = const()[name = string("op_2981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2981_cast_fp16 = slice_by_index(begin = var_2981_begin_0, end = var_2981_end_0, end_mask = var_2981_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2981_cast_fp16")]; + tensor var_2988_begin_0 = const()[name = string("op_2988_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2988_end_0 = const()[name = string("op_2988_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2988_end_mask_0 = const()[name = string("op_2988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2988_cast_fp16 = slice_by_index(begin = var_2988_begin_0, end = var_2988_end_0, end_mask = var_2988_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2988_cast_fp16")]; + tensor var_2995_begin_0 = const()[name = string("op_2995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2995_end_0 = const()[name = string("op_2995_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2995_end_mask_0 = const()[name = string("op_2995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2995_cast_fp16 = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, x = var_2910_cast_fp16)[name = string("op_2995_cast_fp16")]; + tensor var_3002_begin_0 = const()[name = string("op_3002_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3002_end_0 = const()[name = string("op_3002_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3002_end_mask_0 = const()[name = string("op_3002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3002_cast_fp16 = slice_by_index(begin = var_3002_begin_0, end = var_3002_end_0, end_mask = var_3002_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3002_cast_fp16")]; + tensor var_3009_begin_0 = const()[name = string("op_3009_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3009_end_0 = const()[name = string("op_3009_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3009_end_mask_0 = const()[name = string("op_3009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3009_cast_fp16 = slice_by_index(begin = var_3009_begin_0, end = var_3009_end_0, end_mask = var_3009_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3009_cast_fp16")]; + tensor var_3016_begin_0 = const()[name = string("op_3016_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3016_end_0 = const()[name = string("op_3016_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3016_end_mask_0 = const()[name = string("op_3016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3016_cast_fp16 = slice_by_index(begin = var_3016_begin_0, end = var_3016_end_0, end_mask = var_3016_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3016_cast_fp16")]; + tensor var_3023_begin_0 = const()[name = string("op_3023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3023_end_0 = const()[name = string("op_3023_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3023_end_mask_0 = const()[name = string("op_3023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3023_cast_fp16 = slice_by_index(begin = var_3023_begin_0, end = var_3023_end_0, end_mask = var_3023_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3023_cast_fp16")]; + tensor var_3030_begin_0 = const()[name = string("op_3030_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3030_end_0 = const()[name = string("op_3030_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3030_end_mask_0 = const()[name = string("op_3030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3030_cast_fp16 = slice_by_index(begin = var_3030_begin_0, end = var_3030_end_0, end_mask = var_3030_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3030_cast_fp16")]; + tensor var_3037_begin_0 = const()[name = string("op_3037_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3037_end_0 = const()[name = string("op_3037_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3037_end_mask_0 = const()[name = string("op_3037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3037_cast_fp16 = slice_by_index(begin = var_3037_begin_0, end = var_3037_end_0, end_mask = var_3037_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3037_cast_fp16")]; + tensor var_3044_begin_0 = const()[name = string("op_3044_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3044_end_0 = const()[name = string("op_3044_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3044_end_mask_0 = const()[name = string("op_3044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3044_cast_fp16 = slice_by_index(begin = var_3044_begin_0, end = var_3044_end_0, end_mask = var_3044_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3044_cast_fp16")]; + tensor var_3051_begin_0 = const()[name = string("op_3051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3051_end_0 = const()[name = string("op_3051_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3051_end_mask_0 = const()[name = string("op_3051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3051_cast_fp16 = slice_by_index(begin = var_3051_begin_0, end = var_3051_end_0, end_mask = var_3051_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3051_cast_fp16")]; + tensor var_3058_begin_0 = const()[name = string("op_3058_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3058_end_0 = const()[name = string("op_3058_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3058_end_mask_0 = const()[name = string("op_3058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3058_cast_fp16 = slice_by_index(begin = var_3058_begin_0, end = var_3058_end_0, end_mask = var_3058_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3058_cast_fp16")]; + tensor var_3065_begin_0 = const()[name = string("op_3065_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3065_end_0 = const()[name = string("op_3065_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3065_end_mask_0 = const()[name = string("op_3065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3065_cast_fp16 = slice_by_index(begin = var_3065_begin_0, end = var_3065_end_0, end_mask = var_3065_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3065_cast_fp16")]; + tensor var_3072_begin_0 = const()[name = string("op_3072_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3072_end_0 = const()[name = string("op_3072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3072_end_mask_0 = const()[name = string("op_3072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3072_cast_fp16 = slice_by_index(begin = var_3072_begin_0, end = var_3072_end_0, end_mask = var_3072_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3072_cast_fp16")]; + tensor var_3079_begin_0 = const()[name = string("op_3079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3079_end_0 = const()[name = string("op_3079_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3079_end_mask_0 = const()[name = string("op_3079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3079_cast_fp16 = slice_by_index(begin = var_3079_begin_0, end = var_3079_end_0, end_mask = var_3079_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3079_cast_fp16")]; + tensor var_3086_begin_0 = const()[name = string("op_3086_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3086_end_0 = const()[name = string("op_3086_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3086_end_mask_0 = const()[name = string("op_3086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3086_cast_fp16 = slice_by_index(begin = var_3086_begin_0, end = var_3086_end_0, end_mask = var_3086_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3086_cast_fp16")]; + tensor var_3093_begin_0 = const()[name = string("op_3093_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3093_end_0 = const()[name = string("op_3093_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3093_end_mask_0 = const()[name = string("op_3093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3093_cast_fp16 = slice_by_index(begin = var_3093_begin_0, end = var_3093_end_0, end_mask = var_3093_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3093_cast_fp16")]; + tensor var_3100_begin_0 = const()[name = string("op_3100_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3100_end_0 = const()[name = string("op_3100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3100_end_mask_0 = const()[name = string("op_3100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3100_cast_fp16 = slice_by_index(begin = var_3100_begin_0, end = var_3100_end_0, end_mask = var_3100_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3100_cast_fp16")]; + tensor var_3107_begin_0 = const()[name = string("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3107_end_0 = const()[name = string("op_3107_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3107_end_mask_0 = const()[name = string("op_3107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3107_cast_fp16")]; + tensor var_3114_begin_0 = const()[name = string("op_3114_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3114_end_0 = const()[name = string("op_3114_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3114_end_mask_0 = const()[name = string("op_3114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3114_cast_fp16 = slice_by_index(begin = var_3114_begin_0, end = var_3114_end_0, end_mask = var_3114_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3114_cast_fp16")]; + tensor var_3121_begin_0 = const()[name = string("op_3121_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3121_end_0 = const()[name = string("op_3121_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3121_end_mask_0 = const()[name = string("op_3121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3121_cast_fp16 = slice_by_index(begin = var_3121_begin_0, end = var_3121_end_0, end_mask = var_3121_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3121_cast_fp16")]; + tensor var_3128_begin_0 = const()[name = string("op_3128_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3128_end_0 = const()[name = string("op_3128_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3128_end_mask_0 = const()[name = string("op_3128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3128_cast_fp16 = slice_by_index(begin = var_3128_begin_0, end = var_3128_end_0, end_mask = var_3128_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3128_cast_fp16")]; + tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3135_cast_fp16")]; + tensor var_3142_begin_0 = const()[name = string("op_3142_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3142_end_0 = const()[name = string("op_3142_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3142_end_mask_0 = const()[name = string("op_3142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3142_cast_fp16 = slice_by_index(begin = var_3142_begin_0, end = var_3142_end_0, end_mask = var_3142_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3142_cast_fp16")]; + tensor var_3149_begin_0 = const()[name = string("op_3149_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3149_end_0 = const()[name = string("op_3149_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3149_end_mask_0 = const()[name = string("op_3149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3149_cast_fp16 = slice_by_index(begin = var_3149_begin_0, end = var_3149_end_0, end_mask = var_3149_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3149_cast_fp16")]; + tensor var_3156_begin_0 = const()[name = string("op_3156_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3156_end_0 = const()[name = string("op_3156_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3156_end_mask_0 = const()[name = string("op_3156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3156_cast_fp16 = slice_by_index(begin = var_3156_begin_0, end = var_3156_end_0, end_mask = var_3156_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3156_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3161_begin_0 = const()[name = string("op_3161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3161_end_0 = const()[name = string("op_3161_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3161_end_mask_0 = const()[name = string("op_3161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_1")]; + tensor var_3161_cast_fp16 = slice_by_index(begin = var_3161_begin_0, end = var_3161_end_0, end_mask = var_3161_end_mask_0, x = k_9_cast_fp16)[name = string("op_3161_cast_fp16")]; + tensor var_3165_begin_0 = const()[name = string("op_3165_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3165_end_0 = const()[name = string("op_3165_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3165_end_mask_0 = const()[name = string("op_3165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3165_cast_fp16 = slice_by_index(begin = var_3165_begin_0, end = var_3165_end_0, end_mask = var_3165_end_mask_0, x = k_9_cast_fp16)[name = string("op_3165_cast_fp16")]; + tensor var_3169_begin_0 = const()[name = string("op_3169_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3169_end_0 = const()[name = string("op_3169_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3169_end_mask_0 = const()[name = string("op_3169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3169_cast_fp16 = slice_by_index(begin = var_3169_begin_0, end = var_3169_end_0, end_mask = var_3169_end_mask_0, x = k_9_cast_fp16)[name = string("op_3169_cast_fp16")]; + tensor var_3173_begin_0 = const()[name = string("op_3173_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3173_end_0 = const()[name = string("op_3173_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3173_end_mask_0 = const()[name = string("op_3173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3173_cast_fp16 = slice_by_index(begin = var_3173_begin_0, end = var_3173_end_0, end_mask = var_3173_end_mask_0, x = k_9_cast_fp16)[name = string("op_3173_cast_fp16")]; + tensor var_3177_begin_0 = const()[name = string("op_3177_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3177_end_0 = const()[name = string("op_3177_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3177_end_mask_0 = const()[name = string("op_3177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3177_cast_fp16 = slice_by_index(begin = var_3177_begin_0, end = var_3177_end_0, end_mask = var_3177_end_mask_0, x = k_9_cast_fp16)[name = string("op_3177_cast_fp16")]; + tensor var_3181_begin_0 = const()[name = string("op_3181_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3181_end_0 = const()[name = string("op_3181_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3181_end_mask_0 = const()[name = string("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3181_cast_fp16 = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = k_9_cast_fp16)[name = string("op_3181_cast_fp16")]; + tensor var_3185_begin_0 = const()[name = string("op_3185_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3185_end_0 = const()[name = string("op_3185_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3185_end_mask_0 = const()[name = string("op_3185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3185_cast_fp16 = slice_by_index(begin = var_3185_begin_0, end = var_3185_end_0, end_mask = var_3185_end_mask_0, x = k_9_cast_fp16)[name = string("op_3185_cast_fp16")]; + tensor var_3189_begin_0 = const()[name = string("op_3189_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3189_end_0 = const()[name = string("op_3189_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3189_end_mask_0 = const()[name = string("op_3189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3189_cast_fp16 = slice_by_index(begin = var_3189_begin_0, end = var_3189_end_0, end_mask = var_3189_end_mask_0, x = k_9_cast_fp16)[name = string("op_3189_cast_fp16")]; + tensor var_3191_begin_0 = const()[name = string("op_3191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3191_end_0 = const()[name = string("op_3191_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3191_end_mask_0 = const()[name = string("op_3191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3191_cast_fp16 = slice_by_index(begin = var_3191_begin_0, end = var_3191_end_0, end_mask = var_3191_end_mask_0, x = value_9_cast_fp16)[name = string("op_3191_cast_fp16")]; + tensor var_3195_begin_0 = const()[name = string("op_3195_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3195_end_0 = const()[name = string("op_3195_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3195_end_mask_0 = const()[name = string("op_3195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3195_cast_fp16 = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = value_9_cast_fp16)[name = string("op_3195_cast_fp16")]; + tensor var_3199_begin_0 = const()[name = string("op_3199_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3199_end_0 = const()[name = string("op_3199_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3199_end_mask_0 = const()[name = string("op_3199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3199_cast_fp16 = slice_by_index(begin = var_3199_begin_0, end = var_3199_end_0, end_mask = var_3199_end_mask_0, x = value_9_cast_fp16)[name = string("op_3199_cast_fp16")]; + tensor var_3203_begin_0 = const()[name = string("op_3203_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3203_end_0 = const()[name = string("op_3203_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3203_end_mask_0 = const()[name = string("op_3203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3203_cast_fp16 = slice_by_index(begin = var_3203_begin_0, end = var_3203_end_0, end_mask = var_3203_end_mask_0, x = value_9_cast_fp16)[name = string("op_3203_cast_fp16")]; + tensor var_3207_begin_0 = const()[name = string("op_3207_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3207_end_0 = const()[name = string("op_3207_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3207_end_mask_0 = const()[name = string("op_3207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3207_cast_fp16 = slice_by_index(begin = var_3207_begin_0, end = var_3207_end_0, end_mask = var_3207_end_mask_0, x = value_9_cast_fp16)[name = string("op_3207_cast_fp16")]; + tensor var_3211_begin_0 = const()[name = string("op_3211_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3211_end_0 = const()[name = string("op_3211_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3211_end_mask_0 = const()[name = string("op_3211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3211_cast_fp16 = slice_by_index(begin = var_3211_begin_0, end = var_3211_end_0, end_mask = var_3211_end_mask_0, x = value_9_cast_fp16)[name = string("op_3211_cast_fp16")]; + tensor var_3215_begin_0 = const()[name = string("op_3215_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3215_end_0 = const()[name = string("op_3215_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3215_end_mask_0 = const()[name = string("op_3215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3215_cast_fp16 = slice_by_index(begin = var_3215_begin_0, end = var_3215_end_0, end_mask = var_3215_end_mask_0, x = value_9_cast_fp16)[name = string("op_3215_cast_fp16")]; + tensor var_3219_begin_0 = const()[name = string("op_3219_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3219_end_mask_0 = const()[name = string("op_3219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = value_9_cast_fp16)[name = string("op_3219_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_3161_cast_fp16, var_2939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_3161_cast_fp16, var_2946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_3161_cast_fp16, var_2953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_3161_cast_fp16, var_2960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_3165_cast_fp16, var_2967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_3165_cast_fp16, var_2974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_3165_cast_fp16, var_2981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_3165_cast_fp16, var_2988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_3169_cast_fp16, var_2995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_3169_cast_fp16, var_3002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_3169_cast_fp16, var_3009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_3169_cast_fp16, var_3016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_3173_cast_fp16, var_3023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_3173_cast_fp16, var_3030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_3173_cast_fp16, var_3037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_3173_cast_fp16, var_3044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_3177_cast_fp16, var_3051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_3177_cast_fp16, var_3058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_3177_cast_fp16, var_3065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_3177_cast_fp16, var_3072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_3181_cast_fp16, var_3079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_3181_cast_fp16, var_3086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_3181_cast_fp16, var_3093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_3181_cast_fp16, var_3100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_3185_cast_fp16, var_3107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_3185_cast_fp16, var_3114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_3185_cast_fp16, var_3121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_3185_cast_fp16, var_3128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_3189_cast_fp16, var_3135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_3189_cast_fp16, var_3142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_3189_cast_fp16, var_3149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_3189_cast_fp16, var_3156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + fp16 var_3286_to_fp16 = const()[name = string("op_3286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_3286_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_3288_to_fp16 = const()[name = string("op_3288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_3288_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_3290_to_fp16 = const()[name = string("op_3290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_3290_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_3292_to_fp16 = const()[name = string("op_3292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_3292_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_3294_to_fp16 = const()[name = string("op_3294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_3294_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_3296_to_fp16 = const()[name = string("op_3296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_3296_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_3298_to_fp16 = const()[name = string("op_3298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_3298_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_3300_to_fp16 = const()[name = string("op_3300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_3300_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_3302_to_fp16 = const()[name = string("op_3302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_3302_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_3304_to_fp16 = const()[name = string("op_3304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_3304_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_3306_to_fp16 = const()[name = string("op_3306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_3306_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_3308_to_fp16 = const()[name = string("op_3308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_3308_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_3310_to_fp16 = const()[name = string("op_3310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_3310_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_3312_to_fp16 = const()[name = string("op_3312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_3312_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_3314_to_fp16 = const()[name = string("op_3314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_3314_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_3316_to_fp16 = const()[name = string("op_3316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_3316_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + fp16 var_3318_to_fp16 = const()[name = string("op_3318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_3318_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_3320_to_fp16 = const()[name = string("op_3320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_3320_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_3322_to_fp16 = const()[name = string("op_3322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_3322_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_3324_to_fp16 = const()[name = string("op_3324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_3324_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_3326_to_fp16 = const()[name = string("op_3326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_3326_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_3328_to_fp16 = const()[name = string("op_3328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_3328_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_3330_to_fp16 = const()[name = string("op_3330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_3330_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_3332_to_fp16 = const()[name = string("op_3332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_3332_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_3334_to_fp16 = const()[name = string("op_3334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_3334_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_3336_to_fp16 = const()[name = string("op_3336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_3336_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_3338_to_fp16 = const()[name = string("op_3338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_3338_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_3340_to_fp16 = const()[name = string("op_3340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_3340_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_3342_to_fp16 = const()[name = string("op_3342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_3342_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_3344_to_fp16 = const()[name = string("op_3344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_3344_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_3346_to_fp16 = const()[name = string("op_3346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_3346_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_3348_to_fp16 = const()[name = string("op_3348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_3348_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + tensor var_3350_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_257_cast_fp16)[name = string("op_3350_cast_fp16")]; + tensor var_3351_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_259_cast_fp16)[name = string("op_3351_cast_fp16")]; + tensor var_3352_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_261_cast_fp16)[name = string("op_3352_cast_fp16")]; + tensor var_3353_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_263_cast_fp16)[name = string("op_3353_cast_fp16")]; + tensor var_3354_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_265_cast_fp16)[name = string("op_3354_cast_fp16")]; + tensor var_3355_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_267_cast_fp16)[name = string("op_3355_cast_fp16")]; + tensor var_3356_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_269_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3357_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_271_cast_fp16)[name = string("op_3357_cast_fp16")]; + tensor var_3358_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_273_cast_fp16)[name = string("op_3358_cast_fp16")]; + tensor var_3359_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_275_cast_fp16)[name = string("op_3359_cast_fp16")]; + tensor var_3360_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_277_cast_fp16)[name = string("op_3360_cast_fp16")]; + tensor var_3361_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_279_cast_fp16)[name = string("op_3361_cast_fp16")]; + tensor var_3362_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_281_cast_fp16)[name = string("op_3362_cast_fp16")]; + tensor var_3363_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_283_cast_fp16)[name = string("op_3363_cast_fp16")]; + tensor var_3364_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_285_cast_fp16)[name = string("op_3364_cast_fp16")]; + tensor var_3365_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_287_cast_fp16)[name = string("op_3365_cast_fp16")]; + tensor var_3366_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_289_cast_fp16)[name = string("op_3366_cast_fp16")]; + tensor var_3367_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_291_cast_fp16)[name = string("op_3367_cast_fp16")]; + tensor var_3368_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_293_cast_fp16)[name = string("op_3368_cast_fp16")]; + tensor var_3369_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_295_cast_fp16)[name = string("op_3369_cast_fp16")]; + tensor var_3370_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_297_cast_fp16)[name = string("op_3370_cast_fp16")]; + tensor var_3371_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_299_cast_fp16)[name = string("op_3371_cast_fp16")]; + tensor var_3372_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_301_cast_fp16)[name = string("op_3372_cast_fp16")]; + tensor var_3373_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_303_cast_fp16)[name = string("op_3373_cast_fp16")]; + tensor var_3374_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_305_cast_fp16)[name = string("op_3374_cast_fp16")]; + tensor var_3375_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_307_cast_fp16)[name = string("op_3375_cast_fp16")]; + tensor var_3376_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_309_cast_fp16)[name = string("op_3376_cast_fp16")]; + tensor var_3377_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_311_cast_fp16)[name = string("op_3377_cast_fp16")]; + tensor var_3378_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_313_cast_fp16)[name = string("op_3378_cast_fp16")]; + tensor var_3379_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_315_cast_fp16)[name = string("op_3379_cast_fp16")]; + tensor var_3380_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_317_cast_fp16)[name = string("op_3380_cast_fp16")]; + tensor var_3381_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_319_cast_fp16)[name = string("op_3381_cast_fp16")]; + string var_3383_equation_0 = const()[name = string("op_3383_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3383_cast_fp16 = einsum(equation = var_3383_equation_0, values = (var_3191_cast_fp16, var_3350_cast_fp16))[name = string("op_3383_cast_fp16")]; + string var_3385_equation_0 = const()[name = string("op_3385_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3385_cast_fp16 = einsum(equation = var_3385_equation_0, values = (var_3191_cast_fp16, var_3351_cast_fp16))[name = string("op_3385_cast_fp16")]; + string var_3387_equation_0 = const()[name = string("op_3387_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3387_cast_fp16 = einsum(equation = var_3387_equation_0, values = (var_3191_cast_fp16, var_3352_cast_fp16))[name = string("op_3387_cast_fp16")]; + string var_3389_equation_0 = const()[name = string("op_3389_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3389_cast_fp16 = einsum(equation = var_3389_equation_0, values = (var_3191_cast_fp16, var_3353_cast_fp16))[name = string("op_3389_cast_fp16")]; + string var_3391_equation_0 = const()[name = string("op_3391_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3391_cast_fp16 = einsum(equation = var_3391_equation_0, values = (var_3195_cast_fp16, var_3354_cast_fp16))[name = string("op_3391_cast_fp16")]; + string var_3393_equation_0 = const()[name = string("op_3393_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3393_cast_fp16 = einsum(equation = var_3393_equation_0, values = (var_3195_cast_fp16, var_3355_cast_fp16))[name = string("op_3393_cast_fp16")]; + string var_3395_equation_0 = const()[name = string("op_3395_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3395_cast_fp16 = einsum(equation = var_3395_equation_0, values = (var_3195_cast_fp16, var_3356_cast_fp16))[name = string("op_3395_cast_fp16")]; + string var_3397_equation_0 = const()[name = string("op_3397_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3397_cast_fp16 = einsum(equation = var_3397_equation_0, values = (var_3195_cast_fp16, var_3357_cast_fp16))[name = string("op_3397_cast_fp16")]; + string var_3399_equation_0 = const()[name = string("op_3399_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3399_cast_fp16 = einsum(equation = var_3399_equation_0, values = (var_3199_cast_fp16, var_3358_cast_fp16))[name = string("op_3399_cast_fp16")]; + string var_3401_equation_0 = const()[name = string("op_3401_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3401_cast_fp16 = einsum(equation = var_3401_equation_0, values = (var_3199_cast_fp16, var_3359_cast_fp16))[name = string("op_3401_cast_fp16")]; + string var_3403_equation_0 = const()[name = string("op_3403_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3403_cast_fp16 = einsum(equation = var_3403_equation_0, values = (var_3199_cast_fp16, var_3360_cast_fp16))[name = string("op_3403_cast_fp16")]; + string var_3405_equation_0 = const()[name = string("op_3405_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3405_cast_fp16 = einsum(equation = var_3405_equation_0, values = (var_3199_cast_fp16, var_3361_cast_fp16))[name = string("op_3405_cast_fp16")]; + string var_3407_equation_0 = const()[name = string("op_3407_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3407_cast_fp16 = einsum(equation = var_3407_equation_0, values = (var_3203_cast_fp16, var_3362_cast_fp16))[name = string("op_3407_cast_fp16")]; + string var_3409_equation_0 = const()[name = string("op_3409_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3409_cast_fp16 = einsum(equation = var_3409_equation_0, values = (var_3203_cast_fp16, var_3363_cast_fp16))[name = string("op_3409_cast_fp16")]; + string var_3411_equation_0 = const()[name = string("op_3411_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3411_cast_fp16 = einsum(equation = var_3411_equation_0, values = (var_3203_cast_fp16, var_3364_cast_fp16))[name = string("op_3411_cast_fp16")]; + string var_3413_equation_0 = const()[name = string("op_3413_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3413_cast_fp16 = einsum(equation = var_3413_equation_0, values = (var_3203_cast_fp16, var_3365_cast_fp16))[name = string("op_3413_cast_fp16")]; + string var_3415_equation_0 = const()[name = string("op_3415_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3415_cast_fp16 = einsum(equation = var_3415_equation_0, values = (var_3207_cast_fp16, var_3366_cast_fp16))[name = string("op_3415_cast_fp16")]; + string var_3417_equation_0 = const()[name = string("op_3417_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3417_cast_fp16 = einsum(equation = var_3417_equation_0, values = (var_3207_cast_fp16, var_3367_cast_fp16))[name = string("op_3417_cast_fp16")]; + string var_3419_equation_0 = const()[name = string("op_3419_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3419_cast_fp16 = einsum(equation = var_3419_equation_0, values = (var_3207_cast_fp16, var_3368_cast_fp16))[name = string("op_3419_cast_fp16")]; + string var_3421_equation_0 = const()[name = string("op_3421_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3421_cast_fp16 = einsum(equation = var_3421_equation_0, values = (var_3207_cast_fp16, var_3369_cast_fp16))[name = string("op_3421_cast_fp16")]; + string var_3423_equation_0 = const()[name = string("op_3423_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3423_cast_fp16 = einsum(equation = var_3423_equation_0, values = (var_3211_cast_fp16, var_3370_cast_fp16))[name = string("op_3423_cast_fp16")]; + string var_3425_equation_0 = const()[name = string("op_3425_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3425_cast_fp16 = einsum(equation = var_3425_equation_0, values = (var_3211_cast_fp16, var_3371_cast_fp16))[name = string("op_3425_cast_fp16")]; + string var_3427_equation_0 = const()[name = string("op_3427_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3427_cast_fp16 = einsum(equation = var_3427_equation_0, values = (var_3211_cast_fp16, var_3372_cast_fp16))[name = string("op_3427_cast_fp16")]; + string var_3429_equation_0 = const()[name = string("op_3429_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3429_cast_fp16 = einsum(equation = var_3429_equation_0, values = (var_3211_cast_fp16, var_3373_cast_fp16))[name = string("op_3429_cast_fp16")]; + string var_3431_equation_0 = const()[name = string("op_3431_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3431_cast_fp16 = einsum(equation = var_3431_equation_0, values = (var_3215_cast_fp16, var_3374_cast_fp16))[name = string("op_3431_cast_fp16")]; + string var_3433_equation_0 = const()[name = string("op_3433_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3433_cast_fp16 = einsum(equation = var_3433_equation_0, values = (var_3215_cast_fp16, var_3375_cast_fp16))[name = string("op_3433_cast_fp16")]; + string var_3435_equation_0 = const()[name = string("op_3435_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3435_cast_fp16 = einsum(equation = var_3435_equation_0, values = (var_3215_cast_fp16, var_3376_cast_fp16))[name = string("op_3435_cast_fp16")]; + string var_3437_equation_0 = const()[name = string("op_3437_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3437_cast_fp16 = einsum(equation = var_3437_equation_0, values = (var_3215_cast_fp16, var_3377_cast_fp16))[name = string("op_3437_cast_fp16")]; + string var_3439_equation_0 = const()[name = string("op_3439_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3439_cast_fp16 = einsum(equation = var_3439_equation_0, values = (var_3219_cast_fp16, var_3378_cast_fp16))[name = string("op_3439_cast_fp16")]; + string var_3441_equation_0 = const()[name = string("op_3441_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3441_cast_fp16 = einsum(equation = var_3441_equation_0, values = (var_3219_cast_fp16, var_3379_cast_fp16))[name = string("op_3441_cast_fp16")]; + string var_3443_equation_0 = const()[name = string("op_3443_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3443_cast_fp16 = einsum(equation = var_3443_equation_0, values = (var_3219_cast_fp16, var_3380_cast_fp16))[name = string("op_3443_cast_fp16")]; + string var_3445_equation_0 = const()[name = string("op_3445_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3445_cast_fp16 = einsum(equation = var_3445_equation_0, values = (var_3219_cast_fp16, var_3381_cast_fp16))[name = string("op_3445_cast_fp16")]; + bool var_3447_interleave_0 = const()[name = string("op_3447_interleave_0"), val = bool(false)]; + tensor var_3447_cast_fp16 = concat(axis = var_2834, interleave = var_3447_interleave_0, values = (var_3383_cast_fp16, var_3385_cast_fp16, var_3387_cast_fp16, var_3389_cast_fp16))[name = string("op_3447_cast_fp16")]; + bool var_3449_interleave_0 = const()[name = string("op_3449_interleave_0"), val = bool(false)]; + tensor var_3449_cast_fp16 = concat(axis = var_2834, interleave = var_3449_interleave_0, values = (var_3391_cast_fp16, var_3393_cast_fp16, var_3395_cast_fp16, var_3397_cast_fp16))[name = string("op_3449_cast_fp16")]; + bool var_3451_interleave_0 = const()[name = string("op_3451_interleave_0"), val = bool(false)]; + tensor var_3451_cast_fp16 = concat(axis = var_2834, interleave = var_3451_interleave_0, values = (var_3399_cast_fp16, var_3401_cast_fp16, var_3403_cast_fp16, var_3405_cast_fp16))[name = string("op_3451_cast_fp16")]; + bool var_3453_interleave_0 = const()[name = string("op_3453_interleave_0"), val = bool(false)]; + tensor var_3453_cast_fp16 = concat(axis = var_2834, interleave = var_3453_interleave_0, values = (var_3407_cast_fp16, var_3409_cast_fp16, var_3411_cast_fp16, var_3413_cast_fp16))[name = string("op_3453_cast_fp16")]; + bool var_3455_interleave_0 = const()[name = string("op_3455_interleave_0"), val = bool(false)]; + tensor var_3455_cast_fp16 = concat(axis = var_2834, interleave = var_3455_interleave_0, values = (var_3415_cast_fp16, var_3417_cast_fp16, var_3419_cast_fp16, var_3421_cast_fp16))[name = string("op_3455_cast_fp16")]; + bool var_3457_interleave_0 = const()[name = string("op_3457_interleave_0"), val = bool(false)]; + tensor var_3457_cast_fp16 = concat(axis = var_2834, interleave = var_3457_interleave_0, values = (var_3423_cast_fp16, var_3425_cast_fp16, var_3427_cast_fp16, var_3429_cast_fp16))[name = string("op_3457_cast_fp16")]; + bool var_3459_interleave_0 = const()[name = string("op_3459_interleave_0"), val = bool(false)]; + tensor var_3459_cast_fp16 = concat(axis = var_2834, interleave = var_3459_interleave_0, values = (var_3431_cast_fp16, var_3433_cast_fp16, var_3435_cast_fp16, var_3437_cast_fp16))[name = string("op_3459_cast_fp16")]; + bool var_3461_interleave_0 = const()[name = string("op_3461_interleave_0"), val = bool(false)]; + tensor var_3461_cast_fp16 = concat(axis = var_2834, interleave = var_3461_interleave_0, values = (var_3439_cast_fp16, var_3441_cast_fp16, var_3443_cast_fp16, var_3445_cast_fp16))[name = string("op_3461_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_2847, interleave = input_33_interleave_0, values = (var_3447_cast_fp16, var_3449_cast_fp16, var_3451_cast_fp16, var_3453_cast_fp16, var_3455_cast_fp16, var_3457_cast_fp16, var_3459_cast_fp16, var_3461_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30155456)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30679808)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_3480_to_fp16 = const()[name = string("op_3480_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_3480_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30680896)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30681984)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30683072)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32780288)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32784448)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34881664)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_3509 = const()[name = string("op_3509"), val = int32(3)]; + int32 var_3522 = const()[name = string("op_3522"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_3539_to_fp16 = const()[name = string("op_3539_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_3539_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34882752)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34883840)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34884928)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35409280)))]; + tensor query_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35410368)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35934720)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36459072)))]; + tensor value_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_3577_begin_0 = const()[name = string("op_3577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3577_end_0 = const()[name = string("op_3577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3577_end_mask_0 = const()[name = string("op_3577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3577_cast_fp16 = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = query_cast_fp16)[name = string("op_3577_cast_fp16")]; + tensor var_3581_begin_0 = const()[name = string("op_3581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3581_end_0 = const()[name = string("op_3581_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3581_end_mask_0 = const()[name = string("op_3581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3581_cast_fp16 = slice_by_index(begin = var_3581_begin_0, end = var_3581_end_0, end_mask = var_3581_end_mask_0, x = query_cast_fp16)[name = string("op_3581_cast_fp16")]; + tensor var_3585_begin_0 = const()[name = string("op_3585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3585_end_0 = const()[name = string("op_3585_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3585_end_mask_0 = const()[name = string("op_3585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3585_cast_fp16 = slice_by_index(begin = var_3585_begin_0, end = var_3585_end_0, end_mask = var_3585_end_mask_0, x = query_cast_fp16)[name = string("op_3585_cast_fp16")]; + tensor var_3589_begin_0 = const()[name = string("op_3589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3589_end_0 = const()[name = string("op_3589_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3589_end_mask_0 = const()[name = string("op_3589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3589_cast_fp16 = slice_by_index(begin = var_3589_begin_0, end = var_3589_end_0, end_mask = var_3589_end_mask_0, x = query_cast_fp16)[name = string("op_3589_cast_fp16")]; + tensor var_3593_begin_0 = const()[name = string("op_3593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3593_end_0 = const()[name = string("op_3593_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3593_end_mask_0 = const()[name = string("op_3593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3593_cast_fp16 = slice_by_index(begin = var_3593_begin_0, end = var_3593_end_0, end_mask = var_3593_end_mask_0, x = query_cast_fp16)[name = string("op_3593_cast_fp16")]; + tensor var_3597_begin_0 = const()[name = string("op_3597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3597_end_0 = const()[name = string("op_3597_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3597_end_mask_0 = const()[name = string("op_3597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3597_cast_fp16 = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, x = query_cast_fp16)[name = string("op_3597_cast_fp16")]; + tensor var_3601_begin_0 = const()[name = string("op_3601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3601_end_0 = const()[name = string("op_3601_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3601_end_mask_0 = const()[name = string("op_3601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3601_cast_fp16 = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, x = query_cast_fp16)[name = string("op_3601_cast_fp16")]; + tensor var_3605_begin_0 = const()[name = string("op_3605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3605_end_0 = const()[name = string("op_3605_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3605_end_mask_0 = const()[name = string("op_3605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3605_cast_fp16 = slice_by_index(begin = var_3605_begin_0, end = var_3605_end_0, end_mask = var_3605_end_mask_0, x = query_cast_fp16)[name = string("op_3605_cast_fp16")]; + tensor var_3614_begin_0 = const()[name = string("op_3614_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3614_end_0 = const()[name = string("op_3614_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3614_end_mask_0 = const()[name = string("op_3614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3614_cast_fp16 = slice_by_index(begin = var_3614_begin_0, end = var_3614_end_0, end_mask = var_3614_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3614_cast_fp16")]; + tensor var_3621_begin_0 = const()[name = string("op_3621_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3621_end_0 = const()[name = string("op_3621_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3621_end_mask_0 = const()[name = string("op_3621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3621_cast_fp16 = slice_by_index(begin = var_3621_begin_0, end = var_3621_end_0, end_mask = var_3621_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3621_cast_fp16")]; + tensor var_3628_begin_0 = const()[name = string("op_3628_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3628_end_0 = const()[name = string("op_3628_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3628_end_mask_0 = const()[name = string("op_3628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3628_cast_fp16 = slice_by_index(begin = var_3628_begin_0, end = var_3628_end_0, end_mask = var_3628_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3628_cast_fp16")]; + tensor var_3635_begin_0 = const()[name = string("op_3635_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3635_end_0 = const()[name = string("op_3635_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3635_end_mask_0 = const()[name = string("op_3635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3635_cast_fp16 = slice_by_index(begin = var_3635_begin_0, end = var_3635_end_0, end_mask = var_3635_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3635_cast_fp16")]; + tensor var_3642_begin_0 = const()[name = string("op_3642_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3642_end_0 = const()[name = string("op_3642_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3642_end_mask_0 = const()[name = string("op_3642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3642_cast_fp16 = slice_by_index(begin = var_3642_begin_0, end = var_3642_end_0, end_mask = var_3642_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3642_cast_fp16")]; + tensor var_3649_begin_0 = const()[name = string("op_3649_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3649_end_0 = const()[name = string("op_3649_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3649_end_mask_0 = const()[name = string("op_3649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3649_cast_fp16 = slice_by_index(begin = var_3649_begin_0, end = var_3649_end_0, end_mask = var_3649_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3649_cast_fp16")]; + tensor var_3656_begin_0 = const()[name = string("op_3656_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3656_end_0 = const()[name = string("op_3656_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3656_end_mask_0 = const()[name = string("op_3656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3656_cast_fp16 = slice_by_index(begin = var_3656_begin_0, end = var_3656_end_0, end_mask = var_3656_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3656_cast_fp16")]; + tensor var_3663_begin_0 = const()[name = string("op_3663_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3663_end_0 = const()[name = string("op_3663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3663_end_mask_0 = const()[name = string("op_3663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3663_cast_fp16 = slice_by_index(begin = var_3663_begin_0, end = var_3663_end_0, end_mask = var_3663_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3663_cast_fp16")]; + tensor var_3670_begin_0 = const()[name = string("op_3670_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3670_end_0 = const()[name = string("op_3670_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3670_end_mask_0 = const()[name = string("op_3670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3670_cast_fp16 = slice_by_index(begin = var_3670_begin_0, end = var_3670_end_0, end_mask = var_3670_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3670_cast_fp16")]; + tensor var_3677_begin_0 = const()[name = string("op_3677_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3677_end_0 = const()[name = string("op_3677_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3677_end_mask_0 = const()[name = string("op_3677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3677_cast_fp16 = slice_by_index(begin = var_3677_begin_0, end = var_3677_end_0, end_mask = var_3677_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3677_cast_fp16")]; + tensor var_3684_begin_0 = const()[name = string("op_3684_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3684_end_0 = const()[name = string("op_3684_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3684_end_mask_0 = const()[name = string("op_3684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3684_cast_fp16 = slice_by_index(begin = var_3684_begin_0, end = var_3684_end_0, end_mask = var_3684_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3684_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = string("op_3691_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3691_end_0 = const()[name = string("op_3691_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3691_end_mask_0 = const()[name = string("op_3691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3691_cast_fp16")]; + tensor var_3698_begin_0 = const()[name = string("op_3698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3698_end_0 = const()[name = string("op_3698_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3698_end_mask_0 = const()[name = string("op_3698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3698_cast_fp16 = slice_by_index(begin = var_3698_begin_0, end = var_3698_end_0, end_mask = var_3698_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3698_cast_fp16")]; + tensor var_3705_begin_0 = const()[name = string("op_3705_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3705_end_0 = const()[name = string("op_3705_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3705_end_mask_0 = const()[name = string("op_3705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3705_cast_fp16 = slice_by_index(begin = var_3705_begin_0, end = var_3705_end_0, end_mask = var_3705_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3705_cast_fp16")]; + tensor var_3712_begin_0 = const()[name = string("op_3712_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3712_end_0 = const()[name = string("op_3712_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3712_end_mask_0 = const()[name = string("op_3712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3712_cast_fp16 = slice_by_index(begin = var_3712_begin_0, end = var_3712_end_0, end_mask = var_3712_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3712_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = string("op_3719_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3719_end_0 = const()[name = string("op_3719_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3719_end_mask_0 = const()[name = string("op_3719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3719_cast_fp16")]; + tensor var_3726_begin_0 = const()[name = string("op_3726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3726_end_0 = const()[name = string("op_3726_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3726_end_mask_0 = const()[name = string("op_3726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3726_cast_fp16 = slice_by_index(begin = var_3726_begin_0, end = var_3726_end_0, end_mask = var_3726_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3726_cast_fp16")]; + tensor var_3733_begin_0 = const()[name = string("op_3733_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3733_end_0 = const()[name = string("op_3733_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3733_end_mask_0 = const()[name = string("op_3733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3733_cast_fp16 = slice_by_index(begin = var_3733_begin_0, end = var_3733_end_0, end_mask = var_3733_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3733_cast_fp16")]; + tensor var_3740_begin_0 = const()[name = string("op_3740_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3740_end_0 = const()[name = string("op_3740_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3740_end_mask_0 = const()[name = string("op_3740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3740_cast_fp16 = slice_by_index(begin = var_3740_begin_0, end = var_3740_end_0, end_mask = var_3740_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3740_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = string("op_3747_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3747_end_0 = const()[name = string("op_3747_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3747_end_mask_0 = const()[name = string("op_3747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3747_cast_fp16")]; + tensor var_3754_begin_0 = const()[name = string("op_3754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3754_end_0 = const()[name = string("op_3754_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3754_end_mask_0 = const()[name = string("op_3754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3754_cast_fp16 = slice_by_index(begin = var_3754_begin_0, end = var_3754_end_0, end_mask = var_3754_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3754_cast_fp16")]; + tensor var_3761_begin_0 = const()[name = string("op_3761_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3761_end_0 = const()[name = string("op_3761_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3761_end_mask_0 = const()[name = string("op_3761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3761_cast_fp16 = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3761_cast_fp16")]; + tensor var_3768_begin_0 = const()[name = string("op_3768_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3768_end_0 = const()[name = string("op_3768_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3768_end_mask_0 = const()[name = string("op_3768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3768_cast_fp16 = slice_by_index(begin = var_3768_begin_0, end = var_3768_end_0, end_mask = var_3768_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3768_cast_fp16")]; + tensor var_3775_begin_0 = const()[name = string("op_3775_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3775_end_0 = const()[name = string("op_3775_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3775_end_mask_0 = const()[name = string("op_3775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3775_cast_fp16 = slice_by_index(begin = var_3775_begin_0, end = var_3775_end_0, end_mask = var_3775_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3775_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = string("op_3782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3782_end_0 = const()[name = string("op_3782_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3782_end_mask_0 = const()[name = string("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3782_cast_fp16")]; + tensor var_3789_begin_0 = const()[name = string("op_3789_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3789_end_0 = const()[name = string("op_3789_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3789_end_mask_0 = const()[name = string("op_3789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3789_cast_fp16 = slice_by_index(begin = var_3789_begin_0, end = var_3789_end_0, end_mask = var_3789_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3789_cast_fp16")]; + tensor var_3796_begin_0 = const()[name = string("op_3796_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3796_end_0 = const()[name = string("op_3796_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3796_end_mask_0 = const()[name = string("op_3796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3796_cast_fp16 = slice_by_index(begin = var_3796_begin_0, end = var_3796_end_0, end_mask = var_3796_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3796_cast_fp16")]; + tensor var_3803_begin_0 = const()[name = string("op_3803_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3803_end_0 = const()[name = string("op_3803_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3803_end_mask_0 = const()[name = string("op_3803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3803_cast_fp16 = slice_by_index(begin = var_3803_begin_0, end = var_3803_end_0, end_mask = var_3803_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3803_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = string("op_3810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3810_end_0 = const()[name = string("op_3810_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3810_end_mask_0 = const()[name = string("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3810_cast_fp16")]; + tensor var_3817_begin_0 = const()[name = string("op_3817_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3817_end_0 = const()[name = string("op_3817_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3817_end_mask_0 = const()[name = string("op_3817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3817_cast_fp16 = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3817_cast_fp16")]; + tensor var_3824_begin_0 = const()[name = string("op_3824_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3824_end_0 = const()[name = string("op_3824_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3824_end_mask_0 = const()[name = string("op_3824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3824_cast_fp16 = slice_by_index(begin = var_3824_begin_0, end = var_3824_end_0, end_mask = var_3824_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3824_cast_fp16")]; + tensor var_3831_begin_0 = const()[name = string("op_3831_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3831_end_0 = const()[name = string("op_3831_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3831_end_mask_0 = const()[name = string("op_3831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3831_cast_fp16 = slice_by_index(begin = var_3831_begin_0, end = var_3831_end_0, end_mask = var_3831_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3831_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3836_begin_0 = const()[name = string("op_3836_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3836_end_0 = const()[name = string("op_3836_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3836_end_mask_0 = const()[name = string("op_3836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_3836_cast_fp16 = slice_by_index(begin = var_3836_begin_0, end = var_3836_end_0, end_mask = var_3836_end_mask_0, x = k_11_cast_fp16)[name = string("op_3836_cast_fp16")]; + tensor var_3840_begin_0 = const()[name = string("op_3840_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3840_end_0 = const()[name = string("op_3840_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3840_end_mask_0 = const()[name = string("op_3840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3840_cast_fp16 = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, x = k_11_cast_fp16)[name = string("op_3840_cast_fp16")]; + tensor var_3844_begin_0 = const()[name = string("op_3844_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3844_end_0 = const()[name = string("op_3844_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3844_end_mask_0 = const()[name = string("op_3844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3844_cast_fp16 = slice_by_index(begin = var_3844_begin_0, end = var_3844_end_0, end_mask = var_3844_end_mask_0, x = k_11_cast_fp16)[name = string("op_3844_cast_fp16")]; + tensor var_3848_begin_0 = const()[name = string("op_3848_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3848_end_0 = const()[name = string("op_3848_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3848_end_mask_0 = const()[name = string("op_3848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3848_cast_fp16 = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = k_11_cast_fp16)[name = string("op_3848_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = string("op_3852_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3852_end_0 = const()[name = string("op_3852_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3852_end_mask_0 = const()[name = string("op_3852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = k_11_cast_fp16)[name = string("op_3852_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = string("op_3856_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3856_end_0 = const()[name = string("op_3856_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3856_end_mask_0 = const()[name = string("op_3856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = k_11_cast_fp16)[name = string("op_3856_cast_fp16")]; + tensor var_3860_begin_0 = const()[name = string("op_3860_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3860_end_0 = const()[name = string("op_3860_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3860_end_mask_0 = const()[name = string("op_3860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3860_cast_fp16 = slice_by_index(begin = var_3860_begin_0, end = var_3860_end_0, end_mask = var_3860_end_mask_0, x = k_11_cast_fp16)[name = string("op_3860_cast_fp16")]; + tensor var_3864_begin_0 = const()[name = string("op_3864_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3864_end_0 = const()[name = string("op_3864_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3864_end_mask_0 = const()[name = string("op_3864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3864_cast_fp16 = slice_by_index(begin = var_3864_begin_0, end = var_3864_end_0, end_mask = var_3864_end_mask_0, x = k_11_cast_fp16)[name = string("op_3864_cast_fp16")]; + tensor var_3866_begin_0 = const()[name = string("op_3866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3866_end_0 = const()[name = string("op_3866_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3866_end_mask_0 = const()[name = string("op_3866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3866_cast_fp16 = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = value_cast_fp16)[name = string("op_3866_cast_fp16")]; + tensor var_3870_begin_0 = const()[name = string("op_3870_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3870_end_0 = const()[name = string("op_3870_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3870_end_mask_0 = const()[name = string("op_3870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3870_cast_fp16 = slice_by_index(begin = var_3870_begin_0, end = var_3870_end_0, end_mask = var_3870_end_mask_0, x = value_cast_fp16)[name = string("op_3870_cast_fp16")]; + tensor var_3874_begin_0 = const()[name = string("op_3874_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3874_end_0 = const()[name = string("op_3874_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3874_end_mask_0 = const()[name = string("op_3874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3874_cast_fp16 = slice_by_index(begin = var_3874_begin_0, end = var_3874_end_0, end_mask = var_3874_end_mask_0, x = value_cast_fp16)[name = string("op_3874_cast_fp16")]; + tensor var_3878_begin_0 = const()[name = string("op_3878_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3878_end_0 = const()[name = string("op_3878_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3878_end_mask_0 = const()[name = string("op_3878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3878_cast_fp16 = slice_by_index(begin = var_3878_begin_0, end = var_3878_end_0, end_mask = var_3878_end_mask_0, x = value_cast_fp16)[name = string("op_3878_cast_fp16")]; + tensor var_3882_begin_0 = const()[name = string("op_3882_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3882_end_0 = const()[name = string("op_3882_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3882_end_mask_0 = const()[name = string("op_3882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3882_cast_fp16 = slice_by_index(begin = var_3882_begin_0, end = var_3882_end_0, end_mask = var_3882_end_mask_0, x = value_cast_fp16)[name = string("op_3882_cast_fp16")]; + tensor var_3886_begin_0 = const()[name = string("op_3886_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3886_end_0 = const()[name = string("op_3886_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3886_end_mask_0 = const()[name = string("op_3886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3886_cast_fp16 = slice_by_index(begin = var_3886_begin_0, end = var_3886_end_0, end_mask = var_3886_end_mask_0, x = value_cast_fp16)[name = string("op_3886_cast_fp16")]; + tensor var_3890_begin_0 = const()[name = string("op_3890_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3890_end_0 = const()[name = string("op_3890_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3890_end_mask_0 = const()[name = string("op_3890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3890_cast_fp16 = slice_by_index(begin = var_3890_begin_0, end = var_3890_end_0, end_mask = var_3890_end_mask_0, x = value_cast_fp16)[name = string("op_3890_cast_fp16")]; + tensor var_3894_begin_0 = const()[name = string("op_3894_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3894_end_0 = const()[name = string("op_3894_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3894_end_mask_0 = const()[name = string("op_3894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3894_cast_fp16 = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = value_cast_fp16)[name = string("op_3894_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3836_cast_fp16, var_3614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3836_cast_fp16, var_3621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3836_cast_fp16, var_3628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3836_cast_fp16, var_3635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3840_cast_fp16, var_3642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3840_cast_fp16, var_3649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3840_cast_fp16, var_3656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3840_cast_fp16, var_3663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3844_cast_fp16, var_3670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3844_cast_fp16, var_3677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3844_cast_fp16, var_3684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3844_cast_fp16, var_3691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3848_cast_fp16, var_3698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3848_cast_fp16, var_3705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3848_cast_fp16, var_3712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3848_cast_fp16, var_3719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3852_cast_fp16, var_3726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3852_cast_fp16, var_3733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3852_cast_fp16, var_3740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3852_cast_fp16, var_3747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3856_cast_fp16, var_3754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3856_cast_fp16, var_3761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3856_cast_fp16, var_3768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3856_cast_fp16, var_3775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3860_cast_fp16, var_3782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3860_cast_fp16, var_3789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3860_cast_fp16, var_3796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3860_cast_fp16, var_3803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3864_cast_fp16, var_3810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3864_cast_fp16, var_3817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3864_cast_fp16, var_3824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_3864_cast_fp16, var_3831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_3961_to_fp16 = const()[name = string("op_3961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_3961_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_3963_to_fp16 = const()[name = string("op_3963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_3963_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_3965_to_fp16 = const()[name = string("op_3965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_3965_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_3967_to_fp16 = const()[name = string("op_3967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_3967_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_3969_to_fp16 = const()[name = string("op_3969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_3969_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_3971_to_fp16 = const()[name = string("op_3971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_3971_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_3973_to_fp16 = const()[name = string("op_3973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_3973_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_3975_to_fp16 = const()[name = string("op_3975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_3975_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_3977_to_fp16 = const()[name = string("op_3977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_3977_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_3979_to_fp16 = const()[name = string("op_3979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_3979_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_3981_to_fp16 = const()[name = string("op_3981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_3981_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_3983_to_fp16 = const()[name = string("op_3983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_3983_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_3985_to_fp16 = const()[name = string("op_3985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_3985_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_3987_to_fp16 = const()[name = string("op_3987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_3987_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_3989_to_fp16 = const()[name = string("op_3989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_3989_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_3991_to_fp16 = const()[name = string("op_3991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_3991_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_3993_to_fp16 = const()[name = string("op_3993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_3993_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_3995_to_fp16 = const()[name = string("op_3995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_3995_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_3997_to_fp16 = const()[name = string("op_3997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_3997_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_3999_to_fp16 = const()[name = string("op_3999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_3999_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_4001_to_fp16 = const()[name = string("op_4001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_4001_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_4003_to_fp16 = const()[name = string("op_4003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_4003_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_4005_to_fp16 = const()[name = string("op_4005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_4005_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_4007_to_fp16 = const()[name = string("op_4007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_4007_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_4009_to_fp16 = const()[name = string("op_4009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_4009_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_4011_to_fp16 = const()[name = string("op_4011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_4011_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_4013_to_fp16 = const()[name = string("op_4013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_4013_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_4015_to_fp16 = const()[name = string("op_4015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_4015_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_4017_to_fp16 = const()[name = string("op_4017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_4017_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_4019_to_fp16 = const()[name = string("op_4019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_4019_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_4021_to_fp16 = const()[name = string("op_4021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_4021_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_4023_to_fp16 = const()[name = string("op_4023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_4023_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_4025_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_321_cast_fp16)[name = string("op_4025_cast_fp16")]; + tensor var_4026_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_323_cast_fp16)[name = string("op_4026_cast_fp16")]; + tensor var_4027_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_325_cast_fp16)[name = string("op_4027_cast_fp16")]; + tensor var_4028_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_327_cast_fp16)[name = string("op_4028_cast_fp16")]; + tensor var_4029_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_329_cast_fp16)[name = string("op_4029_cast_fp16")]; + tensor var_4030_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_331_cast_fp16)[name = string("op_4030_cast_fp16")]; + tensor var_4031_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_333_cast_fp16)[name = string("op_4031_cast_fp16")]; + tensor var_4032_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_335_cast_fp16)[name = string("op_4032_cast_fp16")]; + tensor var_4033_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_337_cast_fp16)[name = string("op_4033_cast_fp16")]; + tensor var_4034_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_339_cast_fp16)[name = string("op_4034_cast_fp16")]; + tensor var_4035_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_341_cast_fp16)[name = string("op_4035_cast_fp16")]; + tensor var_4036_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_343_cast_fp16)[name = string("op_4036_cast_fp16")]; + tensor var_4037_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_345_cast_fp16)[name = string("op_4037_cast_fp16")]; + tensor var_4038_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_347_cast_fp16)[name = string("op_4038_cast_fp16")]; + tensor var_4039_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_349_cast_fp16)[name = string("op_4039_cast_fp16")]; + tensor var_4040_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_351_cast_fp16)[name = string("op_4040_cast_fp16")]; + tensor var_4041_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_353_cast_fp16)[name = string("op_4041_cast_fp16")]; + tensor var_4042_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_355_cast_fp16)[name = string("op_4042_cast_fp16")]; + tensor var_4043_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_357_cast_fp16)[name = string("op_4043_cast_fp16")]; + tensor var_4044_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_359_cast_fp16)[name = string("op_4044_cast_fp16")]; + tensor var_4045_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_361_cast_fp16)[name = string("op_4045_cast_fp16")]; + tensor var_4046_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_363_cast_fp16)[name = string("op_4046_cast_fp16")]; + tensor var_4047_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_365_cast_fp16)[name = string("op_4047_cast_fp16")]; + tensor var_4048_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_367_cast_fp16)[name = string("op_4048_cast_fp16")]; + tensor var_4049_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_369_cast_fp16)[name = string("op_4049_cast_fp16")]; + tensor var_4050_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_371_cast_fp16)[name = string("op_4050_cast_fp16")]; + tensor var_4051_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_373_cast_fp16)[name = string("op_4051_cast_fp16")]; + tensor var_4052_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_375_cast_fp16)[name = string("op_4052_cast_fp16")]; + tensor var_4053_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_377_cast_fp16)[name = string("op_4053_cast_fp16")]; + tensor var_4054_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_379_cast_fp16)[name = string("op_4054_cast_fp16")]; + tensor var_4055_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_381_cast_fp16)[name = string("op_4055_cast_fp16")]; + tensor var_4056_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_cast_fp16)[name = string("op_4056_cast_fp16")]; + string var_4058_equation_0 = const()[name = string("op_4058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4058_cast_fp16 = einsum(equation = var_4058_equation_0, values = (var_3866_cast_fp16, var_4025_cast_fp16))[name = string("op_4058_cast_fp16")]; + string var_4060_equation_0 = const()[name = string("op_4060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4060_cast_fp16 = einsum(equation = var_4060_equation_0, values = (var_3866_cast_fp16, var_4026_cast_fp16))[name = string("op_4060_cast_fp16")]; + string var_4062_equation_0 = const()[name = string("op_4062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4062_cast_fp16 = einsum(equation = var_4062_equation_0, values = (var_3866_cast_fp16, var_4027_cast_fp16))[name = string("op_4062_cast_fp16")]; + string var_4064_equation_0 = const()[name = string("op_4064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4064_cast_fp16 = einsum(equation = var_4064_equation_0, values = (var_3866_cast_fp16, var_4028_cast_fp16))[name = string("op_4064_cast_fp16")]; + string var_4066_equation_0 = const()[name = string("op_4066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4066_cast_fp16 = einsum(equation = var_4066_equation_0, values = (var_3870_cast_fp16, var_4029_cast_fp16))[name = string("op_4066_cast_fp16")]; + string var_4068_equation_0 = const()[name = string("op_4068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4068_cast_fp16 = einsum(equation = var_4068_equation_0, values = (var_3870_cast_fp16, var_4030_cast_fp16))[name = string("op_4068_cast_fp16")]; + string var_4070_equation_0 = const()[name = string("op_4070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4070_cast_fp16 = einsum(equation = var_4070_equation_0, values = (var_3870_cast_fp16, var_4031_cast_fp16))[name = string("op_4070_cast_fp16")]; + string var_4072_equation_0 = const()[name = string("op_4072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4072_cast_fp16 = einsum(equation = var_4072_equation_0, values = (var_3870_cast_fp16, var_4032_cast_fp16))[name = string("op_4072_cast_fp16")]; + string var_4074_equation_0 = const()[name = string("op_4074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4074_cast_fp16 = einsum(equation = var_4074_equation_0, values = (var_3874_cast_fp16, var_4033_cast_fp16))[name = string("op_4074_cast_fp16")]; + string var_4076_equation_0 = const()[name = string("op_4076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4076_cast_fp16 = einsum(equation = var_4076_equation_0, values = (var_3874_cast_fp16, var_4034_cast_fp16))[name = string("op_4076_cast_fp16")]; + string var_4078_equation_0 = const()[name = string("op_4078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4078_cast_fp16 = einsum(equation = var_4078_equation_0, values = (var_3874_cast_fp16, var_4035_cast_fp16))[name = string("op_4078_cast_fp16")]; + string var_4080_equation_0 = const()[name = string("op_4080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4080_cast_fp16 = einsum(equation = var_4080_equation_0, values = (var_3874_cast_fp16, var_4036_cast_fp16))[name = string("op_4080_cast_fp16")]; + string var_4082_equation_0 = const()[name = string("op_4082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4082_cast_fp16 = einsum(equation = var_4082_equation_0, values = (var_3878_cast_fp16, var_4037_cast_fp16))[name = string("op_4082_cast_fp16")]; + string var_4084_equation_0 = const()[name = string("op_4084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4084_cast_fp16 = einsum(equation = var_4084_equation_0, values = (var_3878_cast_fp16, var_4038_cast_fp16))[name = string("op_4084_cast_fp16")]; + string var_4086_equation_0 = const()[name = string("op_4086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4086_cast_fp16 = einsum(equation = var_4086_equation_0, values = (var_3878_cast_fp16, var_4039_cast_fp16))[name = string("op_4086_cast_fp16")]; + string var_4088_equation_0 = const()[name = string("op_4088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4088_cast_fp16 = einsum(equation = var_4088_equation_0, values = (var_3878_cast_fp16, var_4040_cast_fp16))[name = string("op_4088_cast_fp16")]; + string var_4090_equation_0 = const()[name = string("op_4090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4090_cast_fp16 = einsum(equation = var_4090_equation_0, values = (var_3882_cast_fp16, var_4041_cast_fp16))[name = string("op_4090_cast_fp16")]; + string var_4092_equation_0 = const()[name = string("op_4092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4092_cast_fp16 = einsum(equation = var_4092_equation_0, values = (var_3882_cast_fp16, var_4042_cast_fp16))[name = string("op_4092_cast_fp16")]; + string var_4094_equation_0 = const()[name = string("op_4094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4094_cast_fp16 = einsum(equation = var_4094_equation_0, values = (var_3882_cast_fp16, var_4043_cast_fp16))[name = string("op_4094_cast_fp16")]; + string var_4096_equation_0 = const()[name = string("op_4096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4096_cast_fp16 = einsum(equation = var_4096_equation_0, values = (var_3882_cast_fp16, var_4044_cast_fp16))[name = string("op_4096_cast_fp16")]; + string var_4098_equation_0 = const()[name = string("op_4098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4098_cast_fp16 = einsum(equation = var_4098_equation_0, values = (var_3886_cast_fp16, var_4045_cast_fp16))[name = string("op_4098_cast_fp16")]; + string var_4100_equation_0 = const()[name = string("op_4100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4100_cast_fp16 = einsum(equation = var_4100_equation_0, values = (var_3886_cast_fp16, var_4046_cast_fp16))[name = string("op_4100_cast_fp16")]; + string var_4102_equation_0 = const()[name = string("op_4102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4102_cast_fp16 = einsum(equation = var_4102_equation_0, values = (var_3886_cast_fp16, var_4047_cast_fp16))[name = string("op_4102_cast_fp16")]; + string var_4104_equation_0 = const()[name = string("op_4104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4104_cast_fp16 = einsum(equation = var_4104_equation_0, values = (var_3886_cast_fp16, var_4048_cast_fp16))[name = string("op_4104_cast_fp16")]; + string var_4106_equation_0 = const()[name = string("op_4106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4106_cast_fp16 = einsum(equation = var_4106_equation_0, values = (var_3890_cast_fp16, var_4049_cast_fp16))[name = string("op_4106_cast_fp16")]; + string var_4108_equation_0 = const()[name = string("op_4108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4108_cast_fp16 = einsum(equation = var_4108_equation_0, values = (var_3890_cast_fp16, var_4050_cast_fp16))[name = string("op_4108_cast_fp16")]; + string var_4110_equation_0 = const()[name = string("op_4110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4110_cast_fp16 = einsum(equation = var_4110_equation_0, values = (var_3890_cast_fp16, var_4051_cast_fp16))[name = string("op_4110_cast_fp16")]; + string var_4112_equation_0 = const()[name = string("op_4112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4112_cast_fp16 = einsum(equation = var_4112_equation_0, values = (var_3890_cast_fp16, var_4052_cast_fp16))[name = string("op_4112_cast_fp16")]; + string var_4114_equation_0 = const()[name = string("op_4114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4114_cast_fp16 = einsum(equation = var_4114_equation_0, values = (var_3894_cast_fp16, var_4053_cast_fp16))[name = string("op_4114_cast_fp16")]; + string var_4116_equation_0 = const()[name = string("op_4116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4116_cast_fp16 = einsum(equation = var_4116_equation_0, values = (var_3894_cast_fp16, var_4054_cast_fp16))[name = string("op_4116_cast_fp16")]; + string var_4118_equation_0 = const()[name = string("op_4118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4118_cast_fp16 = einsum(equation = var_4118_equation_0, values = (var_3894_cast_fp16, var_4055_cast_fp16))[name = string("op_4118_cast_fp16")]; + string var_4120_equation_0 = const()[name = string("op_4120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4120_cast_fp16 = einsum(equation = var_4120_equation_0, values = (var_3894_cast_fp16, var_4056_cast_fp16))[name = string("op_4120_cast_fp16")]; + bool var_4122_interleave_0 = const()[name = string("op_4122_interleave_0"), val = bool(false)]; + tensor var_4122_cast_fp16 = concat(axis = var_3509, interleave = var_4122_interleave_0, values = (var_4058_cast_fp16, var_4060_cast_fp16, var_4062_cast_fp16, var_4064_cast_fp16))[name = string("op_4122_cast_fp16")]; + bool var_4124_interleave_0 = const()[name = string("op_4124_interleave_0"), val = bool(false)]; + tensor var_4124_cast_fp16 = concat(axis = var_3509, interleave = var_4124_interleave_0, values = (var_4066_cast_fp16, var_4068_cast_fp16, var_4070_cast_fp16, var_4072_cast_fp16))[name = string("op_4124_cast_fp16")]; + bool var_4126_interleave_0 = const()[name = string("op_4126_interleave_0"), val = bool(false)]; + tensor var_4126_cast_fp16 = concat(axis = var_3509, interleave = var_4126_interleave_0, values = (var_4074_cast_fp16, var_4076_cast_fp16, var_4078_cast_fp16, var_4080_cast_fp16))[name = string("op_4126_cast_fp16")]; + bool var_4128_interleave_0 = const()[name = string("op_4128_interleave_0"), val = bool(false)]; + tensor var_4128_cast_fp16 = concat(axis = var_3509, interleave = var_4128_interleave_0, values = (var_4082_cast_fp16, var_4084_cast_fp16, var_4086_cast_fp16, var_4088_cast_fp16))[name = string("op_4128_cast_fp16")]; + bool var_4130_interleave_0 = const()[name = string("op_4130_interleave_0"), val = bool(false)]; + tensor var_4130_cast_fp16 = concat(axis = var_3509, interleave = var_4130_interleave_0, values = (var_4090_cast_fp16, var_4092_cast_fp16, var_4094_cast_fp16, var_4096_cast_fp16))[name = string("op_4130_cast_fp16")]; + bool var_4132_interleave_0 = const()[name = string("op_4132_interleave_0"), val = bool(false)]; + tensor var_4132_cast_fp16 = concat(axis = var_3509, interleave = var_4132_interleave_0, values = (var_4098_cast_fp16, var_4100_cast_fp16, var_4102_cast_fp16, var_4104_cast_fp16))[name = string("op_4132_cast_fp16")]; + bool var_4134_interleave_0 = const()[name = string("op_4134_interleave_0"), val = bool(false)]; + tensor var_4134_cast_fp16 = concat(axis = var_3509, interleave = var_4134_interleave_0, values = (var_4106_cast_fp16, var_4108_cast_fp16, var_4110_cast_fp16, var_4112_cast_fp16))[name = string("op_4134_cast_fp16")]; + bool var_4136_interleave_0 = const()[name = string("op_4136_interleave_0"), val = bool(false)]; + tensor var_4136_cast_fp16 = concat(axis = var_3509, interleave = var_4136_interleave_0, values = (var_4114_cast_fp16, var_4116_cast_fp16, var_4118_cast_fp16, var_4120_cast_fp16))[name = string("op_4136_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_3522, interleave = input_41_interleave_0, values = (var_4122_cast_fp16, var_4124_cast_fp16, var_4126_cast_fp16, var_4128_cast_fp16, var_4130_cast_fp16, var_4132_cast_fp16, var_4134_cast_fp16, var_4136_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36460160)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36984512)))]; + tensor obj_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_4155_to_fp16 = const()[name = string("op_4155_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_4155_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36985600)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36986688)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36987776)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39084992)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39089152)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41186368)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_4193_to_fp16 = const()[name = string("op_4193_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_4193_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41187456)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41188544)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_4217_pad_type_0 = const()[name = string("op_4217_pad_type_0"), val = string("valid")]; + tensor var_4217_strides_0 = const()[name = string("op_4217_strides_0"), val = tensor([1, 1])]; + tensor var_4217_pad_0 = const()[name = string("op_4217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4217_dilations_0 = const()[name = string("op_4217_dilations_0"), val = tensor([1, 1])]; + int32 var_4217_groups_0 = const()[name = string("op_4217_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41189632)))]; + tensor var_4217_cast_fp16 = conv(dilations = var_4217_dilations_0, groups = var_4217_groups_0, pad = var_4217_pad_0, pad_type = var_4217_pad_type_0, strides = var_4217_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4217_cast_fp16")]; + string var_4224_pad_type_0 = const()[name = string("op_4224_pad_type_0"), val = string("valid")]; + tensor var_4224_strides_0 = const()[name = string("op_4224_strides_0"), val = tensor([1, 1])]; + tensor var_4224_pad_0 = const()[name = string("op_4224_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4224_dilations_0 = const()[name = string("op_4224_dilations_0"), val = tensor([1, 1])]; + int32 var_4224_groups_0 = const()[name = string("op_4224_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41713984)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42238336)))]; + tensor var_4224_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_4224_dilations_0, groups = var_4224_groups_0, pad = var_4224_pad_0, pad_type = var_4224_pad_type_0, strides = var_4224_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4224_cast_fp16")]; + string var_4242_pad_type_0 = const()[name = string("op_4242_pad_type_0"), val = string("valid")]; + tensor var_4242_strides_0 = const()[name = string("op_4242_strides_0"), val = tensor([1, 1])]; + tensor var_4242_pad_0 = const()[name = string("op_4242_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4242_dilations_0 = const()[name = string("op_4242_dilations_0"), val = tensor([1, 1])]; + int32 var_4242_groups_0 = const()[name = string("op_4242_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42239424)))]; + tensor var_4242_cast_fp16 = conv(dilations = var_4242_dilations_0, groups = var_4242_groups_0, pad = var_4242_pad_0, pad_type = var_4242_pad_type_0, strides = var_4242_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4242_cast_fp16")]; + string var_4249_pad_type_0 = const()[name = string("op_4249_pad_type_0"), val = string("valid")]; + tensor var_4249_strides_0 = const()[name = string("op_4249_strides_0"), val = tensor([1, 1])]; + tensor var_4249_pad_0 = const()[name = string("op_4249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4249_dilations_0 = const()[name = string("op_4249_dilations_0"), val = tensor([1, 1])]; + int32 var_4249_groups_0 = const()[name = string("op_4249_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42763776)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43288128)))]; + tensor var_4249_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_4249_dilations_0, groups = var_4249_groups_0, pad = var_4249_pad_0, pad_type = var_4249_pad_type_0, strides = var_4249_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4249_cast_fp16")]; + string var_4267_pad_type_0 = const()[name = string("op_4267_pad_type_0"), val = string("valid")]; + tensor var_4267_strides_0 = const()[name = string("op_4267_strides_0"), val = tensor([1, 1])]; + tensor var_4267_pad_0 = const()[name = string("op_4267_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4267_dilations_0 = const()[name = string("op_4267_dilations_0"), val = tensor([1, 1])]; + int32 var_4267_groups_0 = const()[name = string("op_4267_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43289216)))]; + tensor var_4267_cast_fp16 = conv(dilations = var_4267_dilations_0, groups = var_4267_groups_0, pad = var_4267_pad_0, pad_type = var_4267_pad_type_0, strides = var_4267_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4267_cast_fp16")]; + string var_4274_pad_type_0 = const()[name = string("op_4274_pad_type_0"), val = string("valid")]; + tensor var_4274_strides_0 = const()[name = string("op_4274_strides_0"), val = tensor([1, 1])]; + tensor var_4274_pad_0 = const()[name = string("op_4274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4274_dilations_0 = const()[name = string("op_4274_dilations_0"), val = tensor([1, 1])]; + int32 var_4274_groups_0 = const()[name = string("op_4274_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43813568)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44337920)))]; + tensor var_4274_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_4274_dilations_0, groups = var_4274_groups_0, pad = var_4274_pad_0, pad_type = var_4274_pad_type_0, strides = var_4274_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4274_cast_fp16")]; + string var_4292_pad_type_0 = const()[name = string("op_4292_pad_type_0"), val = string("valid")]; + tensor var_4292_strides_0 = const()[name = string("op_4292_strides_0"), val = tensor([1, 1])]; + tensor var_4292_pad_0 = const()[name = string("op_4292_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4292_dilations_0 = const()[name = string("op_4292_dilations_0"), val = tensor([1, 1])]; + int32 var_4292_groups_0 = const()[name = string("op_4292_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44339008)))]; + tensor var_4292_cast_fp16 = conv(dilations = var_4292_dilations_0, groups = var_4292_groups_0, pad = var_4292_pad_0, pad_type = var_4292_pad_type_0, strides = var_4292_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4292_cast_fp16")]; + string var_4299_pad_type_0 = const()[name = string("op_4299_pad_type_0"), val = string("valid")]; + tensor var_4299_strides_0 = const()[name = string("op_4299_strides_0"), val = tensor([1, 1])]; + tensor var_4299_pad_0 = const()[name = string("op_4299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4299_dilations_0 = const()[name = string("op_4299_dilations_0"), val = tensor([1, 1])]; + int32 var_4299_groups_0 = const()[name = string("op_4299_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44863360)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45387712)))]; + tensor var_4299_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = var_4299_dilations_0, groups = var_4299_groups_0, pad = var_4299_pad_0, pad_type = var_4299_pad_type_0, strides = var_4299_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4299_cast_fp16")]; + string var_4317_pad_type_0 = const()[name = string("op_4317_pad_type_0"), val = string("valid")]; + tensor var_4317_strides_0 = const()[name = string("op_4317_strides_0"), val = tensor([1, 1])]; + tensor var_4317_pad_0 = const()[name = string("op_4317_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4317_dilations_0 = const()[name = string("op_4317_dilations_0"), val = tensor([1, 1])]; + int32 var_4317_groups_0 = const()[name = string("op_4317_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45388800)))]; + tensor var_4317_cast_fp16 = conv(dilations = var_4317_dilations_0, groups = var_4317_groups_0, pad = var_4317_pad_0, pad_type = var_4317_pad_type_0, strides = var_4317_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4317_cast_fp16")]; + string var_4324_pad_type_0 = const()[name = string("op_4324_pad_type_0"), val = string("valid")]; + tensor var_4324_strides_0 = const()[name = string("op_4324_strides_0"), val = tensor([1, 1])]; + tensor var_4324_pad_0 = const()[name = string("op_4324_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4324_dilations_0 = const()[name = string("op_4324_dilations_0"), val = tensor([1, 1])]; + int32 var_4324_groups_0 = const()[name = string("op_4324_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45913152)))]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46437504)))]; + tensor var_4324_cast_fp16 = conv(bias = decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16, dilations = var_4324_dilations_0, groups = var_4324_groups_0, pad = var_4324_pad_0, pad_type = var_4324_pad_type_0, strides = var_4324_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4324_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46438592)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46962944)))]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47487296)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_4354 = const()[name = string("op_4354"), val = int32(0)]; + bool input_51_interleave_0 = const()[name = string("input_51_interleave_0"), val = bool(false)]; + tensor input_51_cast_fp16 = concat(axis = var_4354, interleave = input_51_interleave_0, values = (var_4217_cast_fp16, var_4242_cast_fp16, var_4267_cast_fp16, var_4292_cast_fp16, var_4317_cast_fp16, k_cast_fp16))[name = string("input_51_cast_fp16")]; + int32 var_4357 = const()[name = string("op_4357"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_4357, interleave = input_interleave_0, values = (var_4224_cast_fp16, var_4249_cast_fp16, var_4274_cast_fp16, var_4299_cast_fp16, var_4324_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_4364_pad_0 = const()[name = string("op_4364_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4364_mode_0 = const()[name = string("op_4364_mode_0"), val = string("constant")]; + fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_7_to_fp16, mode = var_4364_mode_0, pad = var_4364_pad_0, x = input_51_cast_fp16)[name = string("op_4364_cast_fp16")]; + tensor var_4370_pad_0 = const()[name = string("op_4370_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4370_mode_0 = const()[name = string("op_4370_mode_0"), val = string("constant")]; + fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_8_to_fp16, mode = var_4370_mode_0, pad = var_4370_pad_0, x = input_cast_fp16)[name = string("op_4370_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-base.en/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-base.en/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..4880b9e3041b7a5ae2f963e16b0adab5840d96a7 --- /dev/null +++ b/openai_whisper-base.en/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf7338475244e4dfd0df94300cf8449d6df39c592d5f9488360b59265a9cc80d +size 47488384 diff --git a/openai_whisper-base.en/LICENSE_NOTICE.txt b/openai_whisper-base.en/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-base.en/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-base.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-base.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a11fbb2cd75b96eb2120a672afefa298c2ef857b --- /dev/null +++ b/openai_whisper-base.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc05e563ee0c556e3f578e04be5fb67b4e7520124403f2561f39102f0f2b33d +size 243 diff --git a/openai_whisper-base.en/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-base.en/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a3544b6644c1af93ca6bdabb67a1c51e80eaa552 --- /dev/null +++ b/openai_whisper-base.en/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ef11ea703011eab03287ec661f999e19c2c78cf67d531b5e6afa02e18f913d +size 328 diff --git a/openai_whisper-base.en/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-base.en/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1a60dd494a857817b67d87cd920baa6824e74b61 --- /dev/null +++ b/openai_whisper-base.en/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-base.en/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-base.en/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cf4cd446f68b88655d00a7df7063aa46937a9bdd --- /dev/null +++ b/openai_whisper-base.en/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-base.en/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-base.en/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f7d28ffac464e9e7086a526930f0059187de8d01 --- /dev/null +++ b/openai_whisper-base.en/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83 +size 354080 diff --git a/openai_whisper-base.en/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-base.en/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..860a9b6dc2432d8e35a976f365d09577a3a57753 --- /dev/null +++ b/openai_whisper-base.en/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:feed440922c687f323c64b1c2ffabdde7b2905e434add975ee0df75654b1a34c +size 243 diff --git a/openai_whisper-base.en/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-base.en/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..de9067d352926cbd932c5fac115ea8d97a206deb --- /dev/null +++ b/openai_whisper-base.en/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b532f135b1b8cec3dac9fdffb77a7304f3f03dda37ac478beeae0f4c6fbd669 +size 754 diff --git a/openai_whisper-base.en/TextDecoder.mlmodelc/metadata.json b/openai_whisper-base.en/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..2fa9916dbedfb86c78f06e54137b53dd848755b0 --- /dev/null +++ b/openai_whisper-base.en/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51864)", + "shortDescription" : "", + "shape" : "[1, 1, 51864]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 3072 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 3072, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 3072 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 3072, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 12, + "Ios18.mul" : 24, + "Ios18.matmul" : 24, + "Ios18.batchNorm" : 19, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 43, + "Ios18.layerNorm" : 19, + "Ios18.reshape" : 48, + "Ios18.linear" : 1, + "Ios18.conv" : 48, + "Ios18.gelu" : 6, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 22, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 448)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 448)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-base.en/TextDecoder.mlmodelc/model.mil b/openai_whisper-base.en/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..1d187982910a661ec68e374bad0b6c49202f316d --- /dev/null +++ b/openai_whisper-base.en/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,946 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; + int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; + bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_30_cast_fp16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = input_ids, validate_indices = var_30_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_30_cast_fp16")]; + int32 var_34_axis_0 = const()[name = string("op_34_axis_0"), val = int32(0)]; + int32 var_34_batch_dims_0 = const()[name = string("op_34_batch_dims_0"), val = int32(0)]; + bool var_34_validate_indices_0 = const()[name = string("op_34_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53108864)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_99")]; + tensor var_34_cast_fp16_cast_uint16 = gather(axis = var_34_axis_0, batch_dims = var_34_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_34_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_34_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_30_cast_fp16, y = var_34_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_48_axes_0 = const()[name = string("op_48_axes_0"), val = tensor([2])]; + tensor var_48_cast_fp16 = expand_dims(axes = var_48_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_48_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_48_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1, 1, 1])]; + int32 var_53_axis_0 = const()[name = string("op_53_axis_0"), val = int32(0)]; + tensor var_53_cast_fp16_0, tensor var_53_cast_fp16_1, tensor var_53_cast_fp16_2, tensor var_53_cast_fp16_3, tensor var_53_cast_fp16_4, tensor var_53_cast_fp16_5 = split(axis = var_53_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_53_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1, 1, 1])]; + int32 var_62_axis_0 = const()[name = string("op_62_axis_0"), val = int32(0)]; + tensor var_62_cast_fp16_0, tensor var_62_cast_fp16_1, tensor var_62_cast_fp16_2, tensor var_62_cast_fp16_3, tensor var_62_cast_fp16_4, tensor var_62_cast_fp16_5 = split(axis = var_62_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_62_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 512, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 512, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_84 = const()[name = string("op_84"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_109_to_fp16 = const()[name = string("op_109_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_109_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53567680)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53568768)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53569856)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53570944)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53572032)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54096384)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54097472)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54621824)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55146176)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_144_axes_0 = const()[name = string("op_144_axes_0"), val = tensor([1])]; + tensor var_144_cast_fp16 = expand_dims(axes = var_144_axes_0, x = kv_cache_update_mask)[name = string("op_144_cast_fp16")]; + tensor var_145_axes_0 = const()[name = string("op_145_axes_0"), val = tensor([2])]; + tensor var_145_cast_fp16 = expand_dims(axes = var_145_axes_0, x = var_144_cast_fp16)[name = string("op_145_cast_fp16")]; + tensor var_147_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_145_cast_fp16)[name = string("op_147_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_53_cast_fp16_0, y = var_147_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_149_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_145_cast_fp16)[name = string("op_149_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_62_cast_fp16_0, y = var_149_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_152 = const()[name = string("op_152"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_152, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_154_to_fp16 = const()[name = string("op_154_to_fp16"), val = fp16(0x1p-3)]; + tensor var_155_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_154_to_fp16)[name = string("op_155_cast_fp16")]; + tensor var_156 = const()[name = string("op_156"), val = tensor([1, 8, 64, -1])]; + tensor var_157_cast_fp16 = reshape(shape = var_156, x = key_1_cast_fp16)[name = string("op_157_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_155_cast_fp16, y = var_157_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_161_axes_0 = const()[name = string("op_161_axes_0"), val = tensor([1])]; + tensor var_161_cast_fp16 = expand_dims(axes = var_161_axes_0, x = decoder_key_padding_mask)[name = string("op_161_cast_fp16")]; + tensor var_162_axes_0 = const()[name = string("op_162_axes_0"), val = tensor([2])]; + tensor var_162_cast_fp16 = expand_dims(axes = var_162_axes_0, x = var_161_cast_fp16)[name = string("op_162_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_165_cast_fp16 = softmax(axis = var_84, x = mh_w_3_cast_fp16)[name = string("op_165_cast_fp16")]; + tensor var_166 = const()[name = string("op_166"), val = tensor([1, 8, 64, -1])]; + tensor var_167_cast_fp16 = reshape(shape = var_166, x = value_1_cast_fp16)[name = string("op_167_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_167_cast_fp16, y = var_165_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_170 = const()[name = string("op_170"), val = tensor([1, 512, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_170, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55147264)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55671616)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_192_to_fp16 = const()[name = string("op_192_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_192_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55672704)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55673792)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55674880)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56199232)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_212 = const()[name = string("op_212"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_212, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_214_to_fp16 = const()[name = string("op_214_to_fp16"), val = fp16(0x1p-3)]; + tensor var_215_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_214_to_fp16)[name = string("op_215_cast_fp16")]; + tensor var_216 = const()[name = string("op_216"), val = tensor([1, 8, 64, -1])]; + tensor var_217_cast_fp16 = reshape(shape = var_216, x = obj_17_cast_fp16)[name = string("op_217_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_215_cast_fp16, y = var_217_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_221_axes_0 = const()[name = string("op_221_axes_0"), val = tensor([1])]; + tensor var_221_cast_fp16 = expand_dims(axes = var_221_axes_0, x = read_state_4)[name = string("op_221_cast_fp16")]; + tensor var_222_axes_0 = const()[name = string("op_222_axes_0"), val = tensor([2])]; + tensor var_222_cast_fp16 = expand_dims(axes = var_222_axes_0, x = var_221_cast_fp16)[name = string("op_222_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_84, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_226 = const()[name = string("op_226"), val = tensor([1, 8, 64, -1])]; + tensor var_227_cast_fp16 = reshape(shape = var_226, x = obj_19_cast_fp16)[name = string("op_227_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_227_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_230 = const()[name = string("op_230"), val = tensor([1, 512, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_230, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56200320)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56724672)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_248_to_fp16 = const()[name = string("op_248_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_248_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56725760)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56726848)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56727936)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58825152)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58829312)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60926528)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 512, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 512, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_293 = const()[name = string("op_293"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_318_to_fp16 = const()[name = string("op_318_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_318_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60927616)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60928704)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60929792)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61454144)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61455232)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61979584)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62503936)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_356_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_145_cast_fp16)[name = string("op_356_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_53_cast_fp16_1, y = var_356_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_358_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_145_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_62_cast_fp16_1, y = var_358_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_361 = const()[name = string("op_361"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_361, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_363_to_fp16 = const()[name = string("op_363_to_fp16"), val = fp16(0x1p-3)]; + tensor var_364_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_363_to_fp16)[name = string("op_364_cast_fp16")]; + tensor var_365 = const()[name = string("op_365"), val = tensor([1, 8, 64, -1])]; + tensor var_366_cast_fp16 = reshape(shape = var_365, x = key_3_cast_fp16)[name = string("op_366_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_364_cast_fp16, y = var_366_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_374_cast_fp16 = softmax(axis = var_293, x = mh_w_11_cast_fp16)[name = string("op_374_cast_fp16")]; + tensor var_375 = const()[name = string("op_375"), val = tensor([1, 8, 64, -1])]; + tensor var_376_cast_fp16 = reshape(shape = var_375, x = value_3_cast_fp16)[name = string("op_376_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_376_cast_fp16, y = var_374_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_379 = const()[name = string("op_379"), val = tensor([1, 512, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_379, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62505024)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63029376)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_401_to_fp16 = const()[name = string("op_401_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_401_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63030464)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63031552)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63032640)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63556992)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_421 = const()[name = string("op_421"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_421, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_423_to_fp16 = const()[name = string("op_423_to_fp16"), val = fp16(0x1p-3)]; + tensor var_424_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_423_to_fp16)[name = string("op_424_cast_fp16")]; + tensor var_425 = const()[name = string("op_425"), val = tensor([1, 8, 64, -1])]; + tensor var_426_cast_fp16 = reshape(shape = var_425, x = obj_35_cast_fp16)[name = string("op_426_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_424_cast_fp16, y = var_426_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_293, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_435 = const()[name = string("op_435"), val = tensor([1, 8, 64, -1])]; + tensor var_436_cast_fp16 = reshape(shape = var_435, x = obj_37_cast_fp16)[name = string("op_436_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_436_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_439 = const()[name = string("op_439"), val = tensor([1, 512, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_439, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63558080)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64082432)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_457_to_fp16 = const()[name = string("op_457_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_457_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64083520)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64084608)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64085696)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66182912)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66187072)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68284288)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 512, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 512, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_502 = const()[name = string("op_502"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_527_to_fp16 = const()[name = string("op_527_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_527_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68285376)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68286464)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68287552)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68811904)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68812992)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69337344)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69861696)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_565_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_145_cast_fp16)[name = string("op_565_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_53_cast_fp16_2, y = var_565_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_567_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_145_cast_fp16)[name = string("op_567_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_62_cast_fp16_2, y = var_567_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_570 = const()[name = string("op_570"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_570, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_572_to_fp16 = const()[name = string("op_572_to_fp16"), val = fp16(0x1p-3)]; + tensor var_573_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_572_to_fp16)[name = string("op_573_cast_fp16")]; + tensor var_574 = const()[name = string("op_574"), val = tensor([1, 8, 64, -1])]; + tensor var_575_cast_fp16 = reshape(shape = var_574, x = key_5_cast_fp16)[name = string("op_575_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_573_cast_fp16, y = var_575_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_583_cast_fp16 = softmax(axis = var_502, x = mh_w_19_cast_fp16)[name = string("op_583_cast_fp16")]; + tensor var_584 = const()[name = string("op_584"), val = tensor([1, 8, 64, -1])]; + tensor var_585_cast_fp16 = reshape(shape = var_584, x = value_5_cast_fp16)[name = string("op_585_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_585_cast_fp16, y = var_583_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_588 = const()[name = string("op_588"), val = tensor([1, 512, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_588, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69862784)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70387136)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_610_to_fp16 = const()[name = string("op_610_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_610_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70388224)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70389312)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70390400)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70914752)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_630 = const()[name = string("op_630"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_630, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_632_to_fp16 = const()[name = string("op_632_to_fp16"), val = fp16(0x1p-3)]; + tensor var_633_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_632_to_fp16)[name = string("op_633_cast_fp16")]; + tensor var_634 = const()[name = string("op_634"), val = tensor([1, 8, 64, -1])]; + tensor var_635_cast_fp16 = reshape(shape = var_634, x = obj_53_cast_fp16)[name = string("op_635_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_633_cast_fp16, y = var_635_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_502, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_644 = const()[name = string("op_644"), val = tensor([1, 8, 64, -1])]; + tensor var_645_cast_fp16 = reshape(shape = var_644, x = obj_55_cast_fp16)[name = string("op_645_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_645_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_648 = const()[name = string("op_648"), val = tensor([1, 512, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_648, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70915840)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71440192)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_666_to_fp16 = const()[name = string("op_666_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_666_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71441280)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71442368)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71443456)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73540672)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73544832)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75642048)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 512, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 512, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_711 = const()[name = string("op_711"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_736_to_fp16 = const()[name = string("op_736_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_736_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75643136)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75644224)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75645312)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76169664)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_7_pad_type_0 = const()[name = string("current_key_7_pad_type_0"), val = string("valid")]; + tensor current_key_7_strides_0 = const()[name = string("current_key_7_strides_0"), val = tensor([1, 1])]; + tensor current_key_7_pad_0 = const()[name = string("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_7_dilations_0 = const()[name = string("current_key_7_dilations_0"), val = tensor([1, 1])]; + int32 current_key_7_groups_0 = const()[name = string("current_key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76170752)))]; + tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_7_cast_fp16")]; + string current_value_7_pad_type_0 = const()[name = string("current_value_7_pad_type_0"), val = string("valid")]; + tensor current_value_7_strides_0 = const()[name = string("current_value_7_strides_0"), val = tensor([1, 1])]; + tensor current_value_7_pad_0 = const()[name = string("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_7_dilations_0 = const()[name = string("current_value_7_dilations_0"), val = tensor([1, 1])]; + int32 current_value_7_groups_0 = const()[name = string("current_value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76695104)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77219456)))]; + tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_7_cast_fp16")]; + tensor var_774_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_145_cast_fp16)[name = string("op_774_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = var_53_cast_fp16_3, y = var_774_cast_fp16)[name = string("key_7_cast_fp16")]; + tensor var_776_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_145_cast_fp16)[name = string("op_776_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = var_62_cast_fp16_3, y = var_776_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_779 = const()[name = string("op_779"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_779, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_781_to_fp16 = const()[name = string("op_781_to_fp16"), val = fp16(0x1p-3)]; + tensor var_782_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_781_to_fp16)[name = string("op_782_cast_fp16")]; + tensor var_783 = const()[name = string("op_783"), val = tensor([1, 8, 64, -1])]; + tensor var_784_cast_fp16 = reshape(shape = var_783, x = key_7_cast_fp16)[name = string("op_784_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_782_cast_fp16, y = var_784_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_792_cast_fp16 = softmax(axis = var_711, x = mh_w_27_cast_fp16)[name = string("op_792_cast_fp16")]; + tensor var_793 = const()[name = string("op_793"), val = tensor([1, 8, 64, -1])]; + tensor var_794_cast_fp16 = reshape(shape = var_793, x = value_7_cast_fp16)[name = string("op_794_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_794_cast_fp16, y = var_792_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_797 = const()[name = string("op_797"), val = tensor([1, 512, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_797, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77220544)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77744896)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_819_to_fp16 = const()[name = string("op_819_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_819_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77745984)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77747072)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77748160)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78272512)))]; + tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_15_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_839, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; + fp16 var_841_to_fp16 = const()[name = string("op_841_to_fp16"), val = fp16(0x1p-3)]; + tensor var_842_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_841_to_fp16)[name = string("op_842_cast_fp16")]; + tensor var_843 = const()[name = string("op_843"), val = tensor([1, 8, 64, -1])]; + tensor var_844_cast_fp16 = reshape(shape = var_843, x = obj_71_cast_fp16)[name = string("op_844_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_842_cast_fp16, y = var_844_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_31_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_711, x = mh_w_31_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_853 = const()[name = string("op_853"), val = tensor([1, 8, 64, -1])]; + tensor var_854_cast_fp16 = reshape(shape = var_853, x = obj_73_cast_fp16)[name = string("op_854_cast_fp16")]; + bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; + bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_854_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_15_cast_fp16")]; + tensor var_857 = const()[name = string("op_857"), val = tensor([1, 512, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_857, x = attn_15_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78273600)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78797952)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_878_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78799040)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78800128)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78801216)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80898432)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80902592)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82999808)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_25_cast_fp16")]; + tensor obj_89_begin_0 = const()[name = string("obj_89_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_89_end_0 = const()[name = string("obj_89_end_0"), val = tensor([5, 512, 1, 1536])]; + tensor obj_89_end_mask_0 = const()[name = string("obj_89_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_89_cast_fp16 = slice_by_index(begin = obj_89_begin_0, end = obj_89_end_0, end_mask = obj_89_end_mask_0, x = read_state_2)[name = string("obj_89_cast_fp16")]; + tensor obj_91_begin_0 = const()[name = string("obj_91_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_91_end_0 = const()[name = string("obj_91_end_0"), val = tensor([5, 512, 1, 1536])]; + tensor obj_91_end_mask_0 = const()[name = string("obj_91_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_91_cast_fp16 = slice_by_index(begin = obj_91_begin_0, end = obj_91_end_0, end_mask = obj_91_end_mask_0, x = read_state_3)[name = string("obj_91_cast_fp16")]; + int32 var_924 = const()[name = string("op_924"), val = int32(3)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_949_to_fp16 = const()[name = string("op_949_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_949_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83000896)))]; + tensor obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83001984)))]; + fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_79_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83003072)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83527424)))]; + tensor query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("query_17_cast_fp16")]; + string current_key_9_pad_type_0 = const()[name = string("current_key_9_pad_type_0"), val = string("valid")]; + tensor current_key_9_strides_0 = const()[name = string("current_key_9_strides_0"), val = tensor([1, 1])]; + tensor current_key_9_pad_0 = const()[name = string("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_9_dilations_0 = const()[name = string("current_key_9_dilations_0"), val = tensor([1, 1])]; + int32 current_key_9_groups_0 = const()[name = string("current_key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83528512)))]; + tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_key_9_cast_fp16")]; + string current_value_9_pad_type_0 = const()[name = string("current_value_9_pad_type_0"), val = string("valid")]; + tensor current_value_9_strides_0 = const()[name = string("current_value_9_strides_0"), val = tensor([1, 1])]; + tensor current_value_9_pad_0 = const()[name = string("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_9_dilations_0 = const()[name = string("current_value_9_dilations_0"), val = tensor([1, 1])]; + int32 current_value_9_groups_0 = const()[name = string("current_value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84052864)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84577216)))]; + tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_value_9_cast_fp16")]; + tensor var_987_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_145_cast_fp16)[name = string("op_987_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = var_53_cast_fp16_4, y = var_987_cast_fp16)[name = string("key_9_cast_fp16")]; + tensor var_989_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_145_cast_fp16)[name = string("op_989_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = var_62_cast_fp16_4, y = var_989_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_992 = const()[name = string("op_992"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_992, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; + fp16 var_994_to_fp16 = const()[name = string("op_994_to_fp16"), val = fp16(0x1p-3)]; + tensor var_995_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_994_to_fp16)[name = string("op_995_cast_fp16")]; + tensor var_996 = const()[name = string("op_996"), val = tensor([1, 8, 64, -1])]; + tensor var_997_cast_fp16 = reshape(shape = var_996, x = key_9_cast_fp16)[name = string("op_997_cast_fp16")]; + bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; + bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_995_cast_fp16, y = var_997_cast_fp16)[name = string("mh_w_33_cast_fp16")]; + tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_35_cast_fp16")]; + tensor var_1005_cast_fp16 = softmax(axis = var_924, x = mh_w_35_cast_fp16)[name = string("op_1005_cast_fp16")]; + tensor var_1006 = const()[name = string("op_1006"), val = tensor([1, 8, 64, -1])]; + tensor var_1007_cast_fp16 = reshape(shape = var_1006, x = value_9_cast_fp16)[name = string("op_1007_cast_fp16")]; + bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; + bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1007_cast_fp16, y = var_1005_cast_fp16)[name = string("attn_17_cast_fp16")]; + tensor var_1010 = const()[name = string("op_1010"), val = tensor([1, 512, 1, -1])]; + tensor input_41_cast_fp16 = reshape(shape = var_1010, x = attn_17_cast_fp16)[name = string("input_41_cast_fp16")]; + string obj_85_pad_type_0 = const()[name = string("obj_85_pad_type_0"), val = string("valid")]; + tensor obj_85_strides_0 = const()[name = string("obj_85_strides_0"), val = tensor([1, 1])]; + tensor obj_85_pad_0 = const()[name = string("obj_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_85_dilations_0 = const()[name = string("obj_85_dilations_0"), val = tensor([1, 1])]; + int32 obj_85_groups_0 = const()[name = string("obj_85_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84578304)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85102656)))]; + tensor obj_85_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_85_dilations_0, groups = obj_85_groups_0, pad = obj_85_pad_0, pad_type = obj_85_pad_type_0, strides = obj_85_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_85_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_85_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_1032_to_fp16 = const()[name = string("op_1032_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1032_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor obj_87_gamma_0_to_fp16 = const()[name = string("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85103744)))]; + tensor obj_87_beta_0_to_fp16 = const()[name = string("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85104832)))]; + fp16 obj_87_epsilon_0_to_fp16 = const()[name = string("obj_87_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("obj_87_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85105920)))]; + tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85630272)))]; + tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16, x = obj_87_cast_fp16)[name = string("query_19_cast_fp16")]; + tensor var_1052 = const()[name = string("op_1052"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_1052, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; + fp16 var_1054_to_fp16 = const()[name = string("op_1054_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1055_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1054_to_fp16)[name = string("op_1055_cast_fp16")]; + tensor var_1056 = const()[name = string("op_1056"), val = tensor([1, 8, 64, -1])]; + tensor var_1057_cast_fp16 = reshape(shape = var_1056, x = obj_89_cast_fp16)[name = string("op_1057_cast_fp16")]; + bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; + bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1055_cast_fp16, y = var_1057_cast_fp16)[name = string("mh_w_37_cast_fp16")]; + tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_39_cast_fp16")]; + tensor obj_95_cast_fp16 = softmax(axis = var_924, x = mh_w_39_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor var_1066 = const()[name = string("op_1066"), val = tensor([1, 8, 64, -1])]; + tensor var_1067_cast_fp16 = reshape(shape = var_1066, x = obj_91_cast_fp16)[name = string("op_1067_cast_fp16")]; + bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; + bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1067_cast_fp16, y = obj_95_cast_fp16)[name = string("attn_19_cast_fp16")]; + tensor var_1070 = const()[name = string("op_1070"), val = tensor([1, 512, 1, -1])]; + tensor input_43_cast_fp16 = reshape(shape = var_1070, x = attn_19_cast_fp16)[name = string("input_43_cast_fp16")]; + string obj_93_pad_type_0 = const()[name = string("obj_93_pad_type_0"), val = string("valid")]; + tensor obj_93_strides_0 = const()[name = string("obj_93_strides_0"), val = tensor([1, 1])]; + tensor obj_93_pad_0 = const()[name = string("obj_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_93_dilations_0 = const()[name = string("obj_93_dilations_0"), val = tensor([1, 1])]; + int32 obj_93_groups_0 = const()[name = string("obj_93_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85631360)))]; + tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86155712)))]; + tensor obj_93_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_93_dilations_0, groups = obj_93_groups_0, pad = obj_93_pad_0, pad_type = obj_93_pad_type_0, strides = obj_93_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("obj_93_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_93_cast_fp16)[name = string("inputs_29_cast_fp16")]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_1091_to_fp16 = const()[name = string("op_1091_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1091_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86156800)))]; + tensor input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86157888)))]; + fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")]; + tensor input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor([1, 1])]; + tensor input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor([1, 1])]; + int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86158976)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88256192)))]; + tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")]; + tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = string("input_49_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88260352)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90357568)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor obj_107_begin_0 = const()[name = string("obj_107_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_107_end_0 = const()[name = string("obj_107_end_0"), val = tensor([6, 512, 1, 1536])]; + tensor obj_107_end_mask_0 = const()[name = string("obj_107_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_107_cast_fp16 = slice_by_index(begin = obj_107_begin_0, end = obj_107_end_0, end_mask = obj_107_end_mask_0, x = read_state_2)[name = string("obj_107_cast_fp16")]; + tensor obj_109_begin_0 = const()[name = string("obj_109_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_109_end_0 = const()[name = string("obj_109_end_0"), val = tensor([6, 512, 1, 1536])]; + tensor obj_109_end_mask_0 = const()[name = string("obj_109_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_109_cast_fp16 = slice_by_index(begin = obj_109_begin_0, end = obj_109_end_0, end_mask = obj_109_end_mask_0, x = read_state_3)[name = string("obj_109_cast_fp16")]; + int32 var_1137 = const()[name = string("op_1137"), val = int32(3)]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_1162_to_fp16 = const()[name = string("op_1162_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1162_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90358656)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90359744)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90360832)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90885184)))]; + tensor query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_21_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90886272)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91410624)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91934976)))]; + tensor current_value_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_1200_cast_fp16 = mul(x = current_key_cast_fp16, y = var_145_cast_fp16)[name = string("op_1200_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_53_cast_fp16_5, y = var_1200_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_1202_cast_fp16 = mul(x = current_value_cast_fp16, y = var_145_cast_fp16)[name = string("op_1202_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_62_cast_fp16_5, y = var_1202_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_1205 = const()[name = string("op_1205"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_1205, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; + fp16 var_1207_to_fp16 = const()[name = string("op_1207_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1208_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1207_to_fp16)[name = string("op_1208_cast_fp16")]; + tensor var_1209 = const()[name = string("op_1209"), val = tensor([1, 8, 64, -1])]; + tensor var_1210_cast_fp16 = reshape(shape = var_1209, x = key_cast_fp16)[name = string("op_1210_cast_fp16")]; + bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; + bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1208_cast_fp16, y = var_1210_cast_fp16)[name = string("mh_w_41_cast_fp16")]; + tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_43_cast_fp16")]; + tensor var_1218_cast_fp16 = softmax(axis = var_1137, x = mh_w_43_cast_fp16)[name = string("op_1218_cast_fp16")]; + tensor var_1219 = const()[name = string("op_1219"), val = tensor([1, 8, 64, -1])]; + tensor var_1220_cast_fp16 = reshape(shape = var_1219, x = value_cast_fp16)[name = string("op_1220_cast_fp16")]; + bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; + bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1220_cast_fp16, y = var_1218_cast_fp16)[name = string("attn_21_cast_fp16")]; + tensor var_1223 = const()[name = string("op_1223"), val = tensor([1, 512, 1, -1])]; + tensor input_51_cast_fp16 = reshape(shape = var_1223, x = attn_21_cast_fp16)[name = string("input_51_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91936064)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92460416)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_33_cast_fp16")]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_1245_to_fp16 = const()[name = string("op_1245_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1245_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92461504)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92462592)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92463680)))]; + tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92988032)))]; + tensor query_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_1265 = const()[name = string("op_1265"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_1265, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1268_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_1267_to_fp16)[name = string("op_1268_cast_fp16")]; + tensor var_1269 = const()[name = string("op_1269"), val = tensor([1, 8, 64, -1])]; + tensor var_1270_cast_fp16 = reshape(shape = var_1269, x = obj_107_cast_fp16)[name = string("op_1270_cast_fp16")]; + bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; + bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_1268_cast_fp16, y = var_1270_cast_fp16)[name = string("mh_w_45_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_45_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_113_cast_fp16 = softmax(axis = var_1137, x = mh_w_cast_fp16)[name = string("obj_113_cast_fp16")]; + tensor var_1279 = const()[name = string("op_1279"), val = tensor([1, 8, 64, -1])]; + tensor var_1280_cast_fp16 = reshape(shape = var_1279, x = obj_109_cast_fp16)[name = string("op_1280_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_1280_cast_fp16, y = obj_113_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_1283 = const()[name = string("op_1283"), val = tensor([1, 512, 1, -1])]; + tensor input_53_cast_fp16 = reshape(shape = var_1283, x = attn_cast_fp16)[name = string("input_53_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92989120)))]; + tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93513472)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_1304_to_fp16 = const()[name = string("op_1304_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1304_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_55_gamma_0_to_fp16 = const()[name = string("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93514560)))]; + tensor input_55_beta_0_to_fp16 = const()[name = string("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93515648)))]; + fp16 input_55_epsilon_0_to_fp16 = const()[name = string("input_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_55_cast_fp16")]; + string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")]; + tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; + int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93516736)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95613952)))]; + tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_57_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95618112)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97715328)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_1347_to_fp16 = const()[name = string("op_1347_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_1347_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97716416)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97717504)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_1358_axes_0 = const()[name = string("op_1358_axes_0"), val = tensor([2])]; + tensor var_1358_cast_fp16 = squeeze(axes = var_1358_axes_0, x = hidden_states_cast_fp16)[name = string("op_1358_cast_fp16")]; + tensor var_1361_perm_0 = const()[name = string("op_1361_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97718592)))]; + tensor var_1361_cast_fp16 = transpose(perm = var_1361_perm_0, x = var_1358_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_1361_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_1365 = const()[name = string("op_1365"), val = int32(1)]; + bool obj_117_interleave_0 = const()[name = string("obj_117_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_1365, interleave = obj_117_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_cast_fp16))[name = string("obj_117_cast_fp16")]; + int32 var_1368 = const()[name = string("op_1368"), val = int32(1)]; + bool obj_119_interleave_0 = const()[name = string("obj_119_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_1368, interleave = obj_119_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_cast_fp16))[name = string("obj_119_cast_fp16")]; + tensor var_1379_begin_0 = const()[name = string("op_1379_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1379_end_0 = const()[name = string("op_1379_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_1379_end_mask_0 = const()[name = string("op_1379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1379_cast_fp16 = slice_by_index(begin = var_1379_begin_0, end = var_1379_end_0, end_mask = var_1379_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1379_cast_fp16")]; + tensor var_1382_begin_0 = const()[name = string("op_1382_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1382_end_0 = const()[name = string("op_1382_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1382_end_mask_0 = const()[name = string("op_1382_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1382_squeeze_mask_0 = const()[name = string("op_1382_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1382_cast_fp16 = slice_by_index(begin = var_1382_begin_0, end = var_1382_end_0, end_mask = var_1382_end_mask_0, squeeze_mask = var_1382_squeeze_mask_0, x = var_1379_cast_fp16)[name = string("op_1382_cast_fp16")]; + tensor var_1397_begin_0 = const()[name = string("op_1397_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1397_end_0 = const()[name = string("op_1397_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_1397_end_mask_0 = const()[name = string("op_1397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1397_cast_fp16 = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, x = obj_95_cast_fp16)[name = string("op_1397_cast_fp16")]; + tensor var_1400_begin_0 = const()[name = string("op_1400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1400_end_0 = const()[name = string("op_1400_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1400_end_mask_0 = const()[name = string("op_1400_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1400_squeeze_mask_0 = const()[name = string("op_1400_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1400_cast_fp16 = slice_by_index(begin = var_1400_begin_0, end = var_1400_end_0, end_mask = var_1400_end_mask_0, squeeze_mask = var_1400_squeeze_mask_0, x = var_1397_cast_fp16)[name = string("op_1400_cast_fp16")]; + tensor var_1415_begin_0 = const()[name = string("op_1415_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1415_end_0 = const()[name = string("op_1415_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_1415_end_mask_0 = const()[name = string("op_1415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1415_cast_fp16 = slice_by_index(begin = var_1415_begin_0, end = var_1415_end_0, end_mask = var_1415_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1415_cast_fp16")]; + tensor var_1418_begin_0 = const()[name = string("op_1418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1418_end_0 = const()[name = string("op_1418_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1418_end_mask_0 = const()[name = string("op_1418_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1418_squeeze_mask_0 = const()[name = string("op_1418_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1418_cast_fp16 = slice_by_index(begin = var_1418_begin_0, end = var_1418_end_0, end_mask = var_1418_end_mask_0, squeeze_mask = var_1418_squeeze_mask_0, x = var_1415_cast_fp16)[name = string("op_1418_cast_fp16")]; + tensor var_1433_begin_0 = const()[name = string("op_1433_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1433_end_0 = const()[name = string("op_1433_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_1433_end_mask_0 = const()[name = string("op_1433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1433_cast_fp16 = slice_by_index(begin = var_1433_begin_0, end = var_1433_end_0, end_mask = var_1433_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1433_cast_fp16")]; + tensor var_1436_begin_0 = const()[name = string("op_1436_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1436_end_0 = const()[name = string("op_1436_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1436_end_mask_0 = const()[name = string("op_1436_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1436_squeeze_mask_0 = const()[name = string("op_1436_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1436_cast_fp16 = slice_by_index(begin = var_1436_begin_0, end = var_1436_end_0, end_mask = var_1436_end_mask_0, squeeze_mask = var_1436_squeeze_mask_0, x = var_1433_cast_fp16)[name = string("op_1436_cast_fp16")]; + tensor var_1451_begin_0 = const()[name = string("op_1451_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1451_end_0 = const()[name = string("op_1451_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_1451_end_mask_0 = const()[name = string("op_1451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1451_cast_fp16 = slice_by_index(begin = var_1451_begin_0, end = var_1451_end_0, end_mask = var_1451_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1451_cast_fp16")]; + tensor var_1454_begin_0 = const()[name = string("op_1454_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1454_end_0 = const()[name = string("op_1454_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1454_end_mask_0 = const()[name = string("op_1454_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1454_squeeze_mask_0 = const()[name = string("op_1454_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1454_cast_fp16 = slice_by_index(begin = var_1454_begin_0, end = var_1454_end_0, end_mask = var_1454_end_mask_0, squeeze_mask = var_1454_squeeze_mask_0, x = var_1451_cast_fp16)[name = string("op_1454_cast_fp16")]; + int32 var_1461 = const()[name = string("op_1461"), val = int32(1)]; + bool var_1462_interleave_0 = const()[name = string("op_1462_interleave_0"), val = bool(false)]; + tensor var_1462_cast_fp16 = concat(axis = var_1461, interleave = var_1462_interleave_0, values = (var_1382_cast_fp16, var_1400_cast_fp16, var_1418_cast_fp16, var_1436_cast_fp16, var_1454_cast_fp16))[name = string("op_1462_cast_fp16")]; + bool var_1465 = const()[name = string("op_1465"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1465, x = var_1462_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-base.en/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-base.en/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..42d53571ab1bbcd98e890088c1ba7f74eedcf7f1 --- /dev/null +++ b/openai_whisper-base.en/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca6b88a4147c3879aae03f15e21806ed19841e10f7849f25180f361a5dffac5d +size 97822384 diff --git a/openai_whisper-base.en/config.json b/openai_whisper-base.en/config.json new file mode 100644 index 0000000000000000000000000000000000000000..7af931c5fd7052d3211621654d93a2a3e33dd307 --- /dev/null +++ b/openai_whisper-base.en/config.json @@ -0,0 +1 @@ +{"_name_or_path": "openai/whisper-base.en", "activation_dropout": 0.0, "activation_function": "gelu", "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50256], "bos_token_id": 50257, "d_model": 512, "decoder_attention_heads": 8, "decoder_ffn_dim": 2048, "decoder_layerdrop": 0.0, "decoder_layers": 6, "decoder_start_token_id": 50257, "dropout": 0.0, "encoder_attention_heads": 8, "encoder_ffn_dim": 2048, "encoder_layerdrop": 0.0, "encoder_layers": 6, "eos_token_id": 50256, "forced_decoder_ids": [[1, 50362]], "init_std": 0.02, "is_encoder_decoder": true, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "model_type": "whisper", "num_hidden_layers": 6, "num_mel_bins": 80, "pad_token_id": 50256, "scale_embedding": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 357, 366, 438, 532, 685, 705, 796, 930, 1058, 1220, 1267, 1279, 1303, 1343, 1377, 1391, 1635, 1782, 1875, 2162, 2361, 2488, 3467, 4008, 4211, 4600, 4808, 5299, 5855, 6329, 7203, 9609, 9959, 10563, 10786, 11420, 11709, 11907, 13163, 13697, 13700, 14808, 15306, 16410, 16791, 17992, 19203, 19510, 20724, 22305, 22935, 27007, 30109, 30420, 33409, 34949, 40283, 40493, 40549, 47282, 49146, 50257, 50357, 50358, 50359, 50360, 50361], "torch_dtype": "float32", "transformers_version": "4.27.0.dev0", "use_cache": true, "vocab_size": 51864} \ No newline at end of file diff --git a/openai_whisper-base.en/generation_config.json b/openai_whisper-base.en/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..cc2edd4b48ab2b8389169e32f930abd8dd8b209d --- /dev/null +++ b/openai_whisper-base.en/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[3, 3], [4, 7], [5, 1], [5, 5], [5, 7]], "begin_suppress_tokens": [220, 50256], "bos_token_id": 50257, "decoder_start_token_id": 50257, "eos_token_id": 50256, "forced_decoder_ids": [[1, 50362]], "is_multilingual": false, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50362, "pad_token_id": 50256, "prev_sot_token_id": 50360, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 357, 366, 438, 532, 685, 705, 796, 930, 1058, 1220, 1267, 1279, 1303, 1343, 1377, 1391, 1635, 1782, 1875, 2162, 2361, 2488, 3467, 4008, 4211, 4600, 4808, 5299, 5855, 6329, 7203, 9609, 9959, 10563, 10786, 11420, 11709, 11907, 13163, 13697, 13700, 14808, 15306, 16410, 16791, 17992, 19203, 19510, 20724, 22305, 22935, 27007, 30109, 30420, 33409, 34949, 40283, 40493, 40549, 47282, 49146, 50257, 50357, 50358, 50359, 50360, 50361], "transformers_version": "4.31.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-base/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-base/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..6781142f32ee6e81c445b39773e9f3623f438be6 --- /dev/null +++ b/openai_whisper-base/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:384d19c754b6ca6a7ad6dd457406dd9c9de44e43034cbfaf3f343e0278e43ac9 +size 243 diff --git a/openai_whisper-base/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-base/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..47ddd91836e24b3659e2ac0d879cc15bfc49a74b --- /dev/null +++ b/openai_whisper-base/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a536e74da525305d998542cdad99de17f18771834664969738d6fa2ab99fd115 +size 433 diff --git a/openai_whisper-base/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-base/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..3db663b3e0d7c210978a6267eba2821c4e38906d --- /dev/null +++ b/openai_whisper-base/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 512 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 512, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 13, + "Ios18.conv" : 50, + "Ios18.gelu" : 8, + "Ios18.concat" : 56, + "Ios16.einsum" : 384, + "Ios18.add" : 13, + "Ios18.softmax" : 192, + "Ios18.sliceByIndex" : 336, + "Ios18.layerNorm" : 13, + "Ios18.transpose" : 6, + "Ios18.mul" : 192 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-base/AudioEncoder.mlmodelc/model.mil b/openai_whisper-base/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cec64545eb7b80d0ff0100fd8a83a36d68918483 --- /dev/null +++ b/openai_whisper-base/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,3351 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_66_pad_type_0 = const()[name = string("op_66_pad_type_0"), val = string("custom")]; + tensor var_66_pad_0 = const()[name = string("op_66_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_66_strides_0 = const()[name = string("op_66_strides_0"), val = tensor([1, 1])]; + tensor var_66_dilations_0 = const()[name = string("op_66_dilations_0"), val = tensor([1, 1])]; + int32 var_66_groups_0 = const()[name = string("op_66_groups_0"), val = int32(1)]; + tensor var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_47_to_fp16 = const()[name = string("op_47_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245888)))]; + tensor var_66_cast_fp16 = conv(bias = var_47_to_fp16, dilations = var_66_dilations_0, groups = var_66_groups_0, pad = var_66_pad_0, pad_type = var_66_pad_type_0, strides = var_66_strides_0, weight = var_41_to_fp16, x = melspectrogram_features)[name = string("op_66_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_66_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_106_pad_type_0 = const()[name = string("op_106_pad_type_0"), val = string("custom")]; + tensor var_106_pad_0 = const()[name = string("op_106_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_106_strides_0 = const()[name = string("op_106_strides_0"), val = tensor([2, 2])]; + tensor var_106_dilations_0 = const()[name = string("op_106_dilations_0"), val = tensor([1, 1])]; + int32 var_106_groups_0 = const()[name = string("op_106_groups_0"), val = int32(1)]; + tensor var_81_to_fp16 = const()[name = string("op_81_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246976)))]; + tensor var_87_to_fp16 = const()[name = string("op_87_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1819904)))]; + tensor var_106_cast_fp16 = conv(bias = var_87_to_fp16, dilations = var_106_dilations_0, groups = var_106_groups_0, pad = var_106_pad_0, pad_type = var_106_pad_type_0, strides = var_106_strides_0, weight = var_81_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_106_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_106_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_124_to_fp16 = const()[name = string("op_124_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1820992)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_124_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_134 = const()[name = string("op_134"), val = int32(3)]; + int32 var_147 = const()[name = string("op_147"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_164_to_fp16 = const()[name = string("op_164_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_164_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3357056)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3358144)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3359232)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3360320)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3361408)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3885760)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3886848)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4411200)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4935552)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_202_begin_0 = const()[name = string("op_202_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_202_end_0 = const()[name = string("op_202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_202_end_mask_0 = const()[name = string("op_202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_202_cast_fp16 = slice_by_index(begin = var_202_begin_0, end = var_202_end_0, end_mask = var_202_end_mask_0, x = query_1_cast_fp16)[name = string("op_202_cast_fp16")]; + tensor var_206_begin_0 = const()[name = string("op_206_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_206_end_0 = const()[name = string("op_206_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_206_end_mask_0 = const()[name = string("op_206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_206_cast_fp16 = slice_by_index(begin = var_206_begin_0, end = var_206_end_0, end_mask = var_206_end_mask_0, x = query_1_cast_fp16)[name = string("op_206_cast_fp16")]; + tensor var_210_begin_0 = const()[name = string("op_210_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_210_end_0 = const()[name = string("op_210_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_210_end_mask_0 = const()[name = string("op_210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_210_cast_fp16 = slice_by_index(begin = var_210_begin_0, end = var_210_end_0, end_mask = var_210_end_mask_0, x = query_1_cast_fp16)[name = string("op_210_cast_fp16")]; + tensor var_214_begin_0 = const()[name = string("op_214_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_214_end_0 = const()[name = string("op_214_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_214_end_mask_0 = const()[name = string("op_214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_214_cast_fp16 = slice_by_index(begin = var_214_begin_0, end = var_214_end_0, end_mask = var_214_end_mask_0, x = query_1_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor var_218_begin_0 = const()[name = string("op_218_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_218_end_0 = const()[name = string("op_218_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_218_end_mask_0 = const()[name = string("op_218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_218_cast_fp16 = slice_by_index(begin = var_218_begin_0, end = var_218_end_0, end_mask = var_218_end_mask_0, x = query_1_cast_fp16)[name = string("op_218_cast_fp16")]; + tensor var_222_begin_0 = const()[name = string("op_222_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_222_end_0 = const()[name = string("op_222_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_222_end_mask_0 = const()[name = string("op_222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_222_cast_fp16 = slice_by_index(begin = var_222_begin_0, end = var_222_end_0, end_mask = var_222_end_mask_0, x = query_1_cast_fp16)[name = string("op_222_cast_fp16")]; + tensor var_226_begin_0 = const()[name = string("op_226_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_226_end_0 = const()[name = string("op_226_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_226_end_mask_0 = const()[name = string("op_226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_226_cast_fp16 = slice_by_index(begin = var_226_begin_0, end = var_226_end_0, end_mask = var_226_end_mask_0, x = query_1_cast_fp16)[name = string("op_226_cast_fp16")]; + tensor var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = string("op_230_cast_fp16")]; + tensor var_239_begin_0 = const()[name = string("op_239_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_239_end_0 = const()[name = string("op_239_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_239_end_mask_0 = const()[name = string("op_239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_239_cast_fp16 = slice_by_index(begin = var_239_begin_0, end = var_239_end_0, end_mask = var_239_end_mask_0, x = var_202_cast_fp16)[name = string("op_239_cast_fp16")]; + tensor var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = var_202_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor var_253_begin_0 = const()[name = string("op_253_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_253_end_0 = const()[name = string("op_253_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_253_end_mask_0 = const()[name = string("op_253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_253_cast_fp16 = slice_by_index(begin = var_253_begin_0, end = var_253_end_0, end_mask = var_253_end_mask_0, x = var_202_cast_fp16)[name = string("op_253_cast_fp16")]; + tensor var_260_begin_0 = const()[name = string("op_260_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_260_end_0 = const()[name = string("op_260_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_260_end_mask_0 = const()[name = string("op_260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_260_cast_fp16 = slice_by_index(begin = var_260_begin_0, end = var_260_end_0, end_mask = var_260_end_mask_0, x = var_202_cast_fp16)[name = string("op_260_cast_fp16")]; + tensor var_267_begin_0 = const()[name = string("op_267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_267_end_0 = const()[name = string("op_267_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_267_end_mask_0 = const()[name = string("op_267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_267_cast_fp16 = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = var_206_cast_fp16)[name = string("op_267_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_206_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_281_begin_0 = const()[name = string("op_281_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_281_end_0 = const()[name = string("op_281_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_281_end_mask_0 = const()[name = string("op_281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_281_cast_fp16 = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_206_cast_fp16)[name = string("op_281_cast_fp16")]; + tensor var_288_begin_0 = const()[name = string("op_288_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_288_end_0 = const()[name = string("op_288_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_288_end_mask_0 = const()[name = string("op_288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_288_cast_fp16 = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_206_cast_fp16)[name = string("op_288_cast_fp16")]; + tensor var_295_begin_0 = const()[name = string("op_295_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_295_end_0 = const()[name = string("op_295_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_295_end_mask_0 = const()[name = string("op_295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_295_cast_fp16 = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_210_cast_fp16)[name = string("op_295_cast_fp16")]; + tensor var_302_begin_0 = const()[name = string("op_302_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_302_end_0 = const()[name = string("op_302_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_302_end_mask_0 = const()[name = string("op_302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = var_210_cast_fp16)[name = string("op_302_cast_fp16")]; + tensor var_309_begin_0 = const()[name = string("op_309_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_309_end_0 = const()[name = string("op_309_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_309_end_mask_0 = const()[name = string("op_309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_309_cast_fp16 = slice_by_index(begin = var_309_begin_0, end = var_309_end_0, end_mask = var_309_end_mask_0, x = var_210_cast_fp16)[name = string("op_309_cast_fp16")]; + tensor var_316_begin_0 = const()[name = string("op_316_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_316_end_0 = const()[name = string("op_316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_316_end_mask_0 = const()[name = string("op_316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_316_cast_fp16 = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = var_210_cast_fp16)[name = string("op_316_cast_fp16")]; + tensor var_323_begin_0 = const()[name = string("op_323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_323_end_0 = const()[name = string("op_323_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_323_end_mask_0 = const()[name = string("op_323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_323_cast_fp16 = slice_by_index(begin = var_323_begin_0, end = var_323_end_0, end_mask = var_323_end_mask_0, x = var_214_cast_fp16)[name = string("op_323_cast_fp16")]; + tensor var_330_begin_0 = const()[name = string("op_330_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_330_end_0 = const()[name = string("op_330_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_330_end_mask_0 = const()[name = string("op_330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_330_cast_fp16 = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = var_214_cast_fp16)[name = string("op_330_cast_fp16")]; + tensor var_337_begin_0 = const()[name = string("op_337_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_337_end_0 = const()[name = string("op_337_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_337_end_mask_0 = const()[name = string("op_337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_337_cast_fp16 = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = var_214_cast_fp16)[name = string("op_337_cast_fp16")]; + tensor var_344_begin_0 = const()[name = string("op_344_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_344_end_0 = const()[name = string("op_344_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_344_end_mask_0 = const()[name = string("op_344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_344_cast_fp16 = slice_by_index(begin = var_344_begin_0, end = var_344_end_0, end_mask = var_344_end_mask_0, x = var_214_cast_fp16)[name = string("op_344_cast_fp16")]; + tensor var_351_begin_0 = const()[name = string("op_351_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_351_end_0 = const()[name = string("op_351_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_351_end_mask_0 = const()[name = string("op_351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_351_cast_fp16 = slice_by_index(begin = var_351_begin_0, end = var_351_end_0, end_mask = var_351_end_mask_0, x = var_218_cast_fp16)[name = string("op_351_cast_fp16")]; + tensor var_358_begin_0 = const()[name = string("op_358_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_358_end_0 = const()[name = string("op_358_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_358_end_mask_0 = const()[name = string("op_358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_358_cast_fp16 = slice_by_index(begin = var_358_begin_0, end = var_358_end_0, end_mask = var_358_end_mask_0, x = var_218_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor var_365_begin_0 = const()[name = string("op_365_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_365_end_0 = const()[name = string("op_365_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_365_end_mask_0 = const()[name = string("op_365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_365_cast_fp16 = slice_by_index(begin = var_365_begin_0, end = var_365_end_0, end_mask = var_365_end_mask_0, x = var_218_cast_fp16)[name = string("op_365_cast_fp16")]; + tensor var_372_begin_0 = const()[name = string("op_372_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_372_end_0 = const()[name = string("op_372_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_372_end_mask_0 = const()[name = string("op_372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_372_cast_fp16 = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, x = var_218_cast_fp16)[name = string("op_372_cast_fp16")]; + tensor var_379_begin_0 = const()[name = string("op_379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_379_end_0 = const()[name = string("op_379_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_379_end_mask_0 = const()[name = string("op_379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_379_cast_fp16 = slice_by_index(begin = var_379_begin_0, end = var_379_end_0, end_mask = var_379_end_mask_0, x = var_222_cast_fp16)[name = string("op_379_cast_fp16")]; + tensor var_386_begin_0 = const()[name = string("op_386_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_386_end_0 = const()[name = string("op_386_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_386_end_mask_0 = const()[name = string("op_386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_386_cast_fp16 = slice_by_index(begin = var_386_begin_0, end = var_386_end_0, end_mask = var_386_end_mask_0, x = var_222_cast_fp16)[name = string("op_386_cast_fp16")]; + tensor var_393_begin_0 = const()[name = string("op_393_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_393_end_0 = const()[name = string("op_393_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_393_end_mask_0 = const()[name = string("op_393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_393_cast_fp16 = slice_by_index(begin = var_393_begin_0, end = var_393_end_0, end_mask = var_393_end_mask_0, x = var_222_cast_fp16)[name = string("op_393_cast_fp16")]; + tensor var_400_begin_0 = const()[name = string("op_400_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_400_end_0 = const()[name = string("op_400_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_400_end_mask_0 = const()[name = string("op_400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_400_cast_fp16 = slice_by_index(begin = var_400_begin_0, end = var_400_end_0, end_mask = var_400_end_mask_0, x = var_222_cast_fp16)[name = string("op_400_cast_fp16")]; + tensor var_407_begin_0 = const()[name = string("op_407_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_407_end_0 = const()[name = string("op_407_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_407_end_mask_0 = const()[name = string("op_407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = var_226_cast_fp16)[name = string("op_407_cast_fp16")]; + tensor var_414_begin_0 = const()[name = string("op_414_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_414_end_0 = const()[name = string("op_414_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_414_end_mask_0 = const()[name = string("op_414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_414_cast_fp16 = slice_by_index(begin = var_414_begin_0, end = var_414_end_0, end_mask = var_414_end_mask_0, x = var_226_cast_fp16)[name = string("op_414_cast_fp16")]; + tensor var_421_begin_0 = const()[name = string("op_421_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_421_end_0 = const()[name = string("op_421_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_421_end_mask_0 = const()[name = string("op_421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = var_226_cast_fp16)[name = string("op_421_cast_fp16")]; + tensor var_428_begin_0 = const()[name = string("op_428_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_428_end_0 = const()[name = string("op_428_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_428_end_mask_0 = const()[name = string("op_428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_428_cast_fp16 = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = var_226_cast_fp16)[name = string("op_428_cast_fp16")]; + tensor var_435_begin_0 = const()[name = string("op_435_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_435_end_0 = const()[name = string("op_435_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_435_end_mask_0 = const()[name = string("op_435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_435_cast_fp16 = slice_by_index(begin = var_435_begin_0, end = var_435_end_0, end_mask = var_435_end_mask_0, x = var_230_cast_fp16)[name = string("op_435_cast_fp16")]; + tensor var_442_begin_0 = const()[name = string("op_442_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_442_end_0 = const()[name = string("op_442_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_442_end_mask_0 = const()[name = string("op_442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = var_230_cast_fp16)[name = string("op_442_cast_fp16")]; + tensor var_449_begin_0 = const()[name = string("op_449_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_449_end_0 = const()[name = string("op_449_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_449_end_mask_0 = const()[name = string("op_449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_449_cast_fp16 = slice_by_index(begin = var_449_begin_0, end = var_449_end_0, end_mask = var_449_end_mask_0, x = var_230_cast_fp16)[name = string("op_449_cast_fp16")]; + tensor var_456_begin_0 = const()[name = string("op_456_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_456_end_0 = const()[name = string("op_456_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_456_end_mask_0 = const()[name = string("op_456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_456_cast_fp16 = slice_by_index(begin = var_456_begin_0, end = var_456_end_0, end_mask = var_456_end_mask_0, x = var_230_cast_fp16)[name = string("op_456_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_461_begin_0 = const()[name = string("op_461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_461_end_0 = const()[name = string("op_461_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_461_end_mask_0 = const()[name = string("op_461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_5")]; + tensor var_461_cast_fp16 = slice_by_index(begin = var_461_begin_0, end = var_461_end_0, end_mask = var_461_end_mask_0, x = k_1_cast_fp16)[name = string("op_461_cast_fp16")]; + tensor var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = k_1_cast_fp16)[name = string("op_465_cast_fp16")]; + tensor var_469_begin_0 = const()[name = string("op_469_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_469_end_0 = const()[name = string("op_469_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_469_end_mask_0 = const()[name = string("op_469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = k_1_cast_fp16)[name = string("op_469_cast_fp16")]; + tensor var_473_begin_0 = const()[name = string("op_473_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_473_end_0 = const()[name = string("op_473_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_473_end_mask_0 = const()[name = string("op_473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = k_1_cast_fp16)[name = string("op_473_cast_fp16")]; + tensor var_477_begin_0 = const()[name = string("op_477_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_477_end_0 = const()[name = string("op_477_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_477_end_mask_0 = const()[name = string("op_477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = k_1_cast_fp16)[name = string("op_477_cast_fp16")]; + tensor var_481_begin_0 = const()[name = string("op_481_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_481_end_0 = const()[name = string("op_481_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_481_end_mask_0 = const()[name = string("op_481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_481_cast_fp16 = slice_by_index(begin = var_481_begin_0, end = var_481_end_0, end_mask = var_481_end_mask_0, x = k_1_cast_fp16)[name = string("op_481_cast_fp16")]; + tensor var_485_begin_0 = const()[name = string("op_485_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_485_end_0 = const()[name = string("op_485_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_485_end_mask_0 = const()[name = string("op_485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_485_cast_fp16 = slice_by_index(begin = var_485_begin_0, end = var_485_end_0, end_mask = var_485_end_mask_0, x = k_1_cast_fp16)[name = string("op_485_cast_fp16")]; + tensor var_489_begin_0 = const()[name = string("op_489_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_489_end_0 = const()[name = string("op_489_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_489_end_mask_0 = const()[name = string("op_489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_489_cast_fp16 = slice_by_index(begin = var_489_begin_0, end = var_489_end_0, end_mask = var_489_end_mask_0, x = k_1_cast_fp16)[name = string("op_489_cast_fp16")]; + tensor var_491_begin_0 = const()[name = string("op_491_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_491_end_0 = const()[name = string("op_491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_491_end_mask_0 = const()[name = string("op_491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_491_cast_fp16 = slice_by_index(begin = var_491_begin_0, end = var_491_end_0, end_mask = var_491_end_mask_0, x = value_1_cast_fp16)[name = string("op_491_cast_fp16")]; + tensor var_495_begin_0 = const()[name = string("op_495_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_495_end_0 = const()[name = string("op_495_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_495_end_mask_0 = const()[name = string("op_495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_495_cast_fp16 = slice_by_index(begin = var_495_begin_0, end = var_495_end_0, end_mask = var_495_end_mask_0, x = value_1_cast_fp16)[name = string("op_495_cast_fp16")]; + tensor var_499_begin_0 = const()[name = string("op_499_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_499_end_0 = const()[name = string("op_499_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_499_end_mask_0 = const()[name = string("op_499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_499_cast_fp16 = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = value_1_cast_fp16)[name = string("op_499_cast_fp16")]; + tensor var_503_begin_0 = const()[name = string("op_503_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_503_end_0 = const()[name = string("op_503_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_503_end_mask_0 = const()[name = string("op_503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_503_cast_fp16 = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = value_1_cast_fp16)[name = string("op_503_cast_fp16")]; + tensor var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = value_1_cast_fp16)[name = string("op_507_cast_fp16")]; + tensor var_511_begin_0 = const()[name = string("op_511_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_511_end_0 = const()[name = string("op_511_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_511_end_mask_0 = const()[name = string("op_511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = value_1_cast_fp16)[name = string("op_511_cast_fp16")]; + tensor var_515_begin_0 = const()[name = string("op_515_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_515_end_0 = const()[name = string("op_515_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_515_end_mask_0 = const()[name = string("op_515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = value_1_cast_fp16)[name = string("op_515_cast_fp16")]; + tensor var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = value_1_cast_fp16)[name = string("op_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_461_cast_fp16, var_239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_461_cast_fp16, var_246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_461_cast_fp16, var_253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_461_cast_fp16, var_260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_465_cast_fp16, var_267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_465_cast_fp16, var_274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_465_cast_fp16, var_281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_465_cast_fp16, var_288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_469_cast_fp16, var_295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_469_cast_fp16, var_302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_469_cast_fp16, var_309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_469_cast_fp16, var_316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_473_cast_fp16, var_323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_473_cast_fp16, var_330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_473_cast_fp16, var_337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_473_cast_fp16, var_344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_477_cast_fp16, var_351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_477_cast_fp16, var_358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_477_cast_fp16, var_365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_477_cast_fp16, var_372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_481_cast_fp16, var_379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_481_cast_fp16, var_386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_481_cast_fp16, var_393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_481_cast_fp16, var_400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_485_cast_fp16, var_407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_485_cast_fp16, var_414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_485_cast_fp16, var_421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_485_cast_fp16, var_428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_489_cast_fp16, var_435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_489_cast_fp16, var_442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_489_cast_fp16, var_449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_489_cast_fp16, var_456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + fp16 var_586_to_fp16 = const()[name = string("op_586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_586_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_588_to_fp16 = const()[name = string("op_588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_588_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_590_to_fp16 = const()[name = string("op_590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_590_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_592_to_fp16 = const()[name = string("op_592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_592_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_594_to_fp16 = const()[name = string("op_594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_594_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_596_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_598_to_fp16 = const()[name = string("op_598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_598_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_600_to_fp16 = const()[name = string("op_600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_600_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_602_to_fp16 = const()[name = string("op_602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_602_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_604_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_606_to_fp16 = const()[name = string("op_606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_606_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_608_to_fp16 = const()[name = string("op_608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_608_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_610_to_fp16 = const()[name = string("op_610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_610_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_612_to_fp16 = const()[name = string("op_612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_612_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_614_to_fp16 = const()[name = string("op_614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_614_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_616_to_fp16 = const()[name = string("op_616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_616_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_618_to_fp16 = const()[name = string("op_618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_618_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_620_to_fp16 = const()[name = string("op_620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_620_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_622_to_fp16 = const()[name = string("op_622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_622_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_624_to_fp16 = const()[name = string("op_624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_624_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_626_to_fp16 = const()[name = string("op_626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_626_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_628_to_fp16 = const()[name = string("op_628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_628_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_630_to_fp16 = const()[name = string("op_630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_630_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_632_to_fp16 = const()[name = string("op_632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_632_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_634_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_636_to_fp16 = const()[name = string("op_636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_636_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_638_to_fp16 = const()[name = string("op_638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_638_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_640_to_fp16 = const()[name = string("op_640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_640_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_642_to_fp16 = const()[name = string("op_642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_642_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_644_to_fp16 = const()[name = string("op_644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_644_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_646_to_fp16 = const()[name = string("op_646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_646_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_648_to_fp16 = const()[name = string("op_648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_648_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + tensor var_650_cast_fp16 = softmax(axis = var_147, x = aw_chunk_1_cast_fp16)[name = string("op_650_cast_fp16")]; + tensor var_651_cast_fp16 = softmax(axis = var_147, x = aw_chunk_3_cast_fp16)[name = string("op_651_cast_fp16")]; + tensor var_652_cast_fp16 = softmax(axis = var_147, x = aw_chunk_5_cast_fp16)[name = string("op_652_cast_fp16")]; + tensor var_653_cast_fp16 = softmax(axis = var_147, x = aw_chunk_7_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor var_654_cast_fp16 = softmax(axis = var_147, x = aw_chunk_9_cast_fp16)[name = string("op_654_cast_fp16")]; + tensor var_655_cast_fp16 = softmax(axis = var_147, x = aw_chunk_11_cast_fp16)[name = string("op_655_cast_fp16")]; + tensor var_656_cast_fp16 = softmax(axis = var_147, x = aw_chunk_13_cast_fp16)[name = string("op_656_cast_fp16")]; + tensor var_657_cast_fp16 = softmax(axis = var_147, x = aw_chunk_15_cast_fp16)[name = string("op_657_cast_fp16")]; + tensor var_658_cast_fp16 = softmax(axis = var_147, x = aw_chunk_17_cast_fp16)[name = string("op_658_cast_fp16")]; + tensor var_659_cast_fp16 = softmax(axis = var_147, x = aw_chunk_19_cast_fp16)[name = string("op_659_cast_fp16")]; + tensor var_660_cast_fp16 = softmax(axis = var_147, x = aw_chunk_21_cast_fp16)[name = string("op_660_cast_fp16")]; + tensor var_661_cast_fp16 = softmax(axis = var_147, x = aw_chunk_23_cast_fp16)[name = string("op_661_cast_fp16")]; + tensor var_662_cast_fp16 = softmax(axis = var_147, x = aw_chunk_25_cast_fp16)[name = string("op_662_cast_fp16")]; + tensor var_663_cast_fp16 = softmax(axis = var_147, x = aw_chunk_27_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_664_cast_fp16 = softmax(axis = var_147, x = aw_chunk_29_cast_fp16)[name = string("op_664_cast_fp16")]; + tensor var_665_cast_fp16 = softmax(axis = var_147, x = aw_chunk_31_cast_fp16)[name = string("op_665_cast_fp16")]; + tensor var_666_cast_fp16 = softmax(axis = var_147, x = aw_chunk_33_cast_fp16)[name = string("op_666_cast_fp16")]; + tensor var_667_cast_fp16 = softmax(axis = var_147, x = aw_chunk_35_cast_fp16)[name = string("op_667_cast_fp16")]; + tensor var_668_cast_fp16 = softmax(axis = var_147, x = aw_chunk_37_cast_fp16)[name = string("op_668_cast_fp16")]; + tensor var_669_cast_fp16 = softmax(axis = var_147, x = aw_chunk_39_cast_fp16)[name = string("op_669_cast_fp16")]; + tensor var_670_cast_fp16 = softmax(axis = var_147, x = aw_chunk_41_cast_fp16)[name = string("op_670_cast_fp16")]; + tensor var_671_cast_fp16 = softmax(axis = var_147, x = aw_chunk_43_cast_fp16)[name = string("op_671_cast_fp16")]; + tensor var_672_cast_fp16 = softmax(axis = var_147, x = aw_chunk_45_cast_fp16)[name = string("op_672_cast_fp16")]; + tensor var_673_cast_fp16 = softmax(axis = var_147, x = aw_chunk_47_cast_fp16)[name = string("op_673_cast_fp16")]; + tensor var_674_cast_fp16 = softmax(axis = var_147, x = aw_chunk_49_cast_fp16)[name = string("op_674_cast_fp16")]; + tensor var_675_cast_fp16 = softmax(axis = var_147, x = aw_chunk_51_cast_fp16)[name = string("op_675_cast_fp16")]; + tensor var_676_cast_fp16 = softmax(axis = var_147, x = aw_chunk_53_cast_fp16)[name = string("op_676_cast_fp16")]; + tensor var_677_cast_fp16 = softmax(axis = var_147, x = aw_chunk_55_cast_fp16)[name = string("op_677_cast_fp16")]; + tensor var_678_cast_fp16 = softmax(axis = var_147, x = aw_chunk_57_cast_fp16)[name = string("op_678_cast_fp16")]; + tensor var_679_cast_fp16 = softmax(axis = var_147, x = aw_chunk_59_cast_fp16)[name = string("op_679_cast_fp16")]; + tensor var_680_cast_fp16 = softmax(axis = var_147, x = aw_chunk_61_cast_fp16)[name = string("op_680_cast_fp16")]; + tensor var_681_cast_fp16 = softmax(axis = var_147, x = aw_chunk_63_cast_fp16)[name = string("op_681_cast_fp16")]; + string var_683_equation_0 = const()[name = string("op_683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_683_cast_fp16 = einsum(equation = var_683_equation_0, values = (var_491_cast_fp16, var_650_cast_fp16))[name = string("op_683_cast_fp16")]; + string var_685_equation_0 = const()[name = string("op_685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_685_cast_fp16 = einsum(equation = var_685_equation_0, values = (var_491_cast_fp16, var_651_cast_fp16))[name = string("op_685_cast_fp16")]; + string var_687_equation_0 = const()[name = string("op_687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_687_cast_fp16 = einsum(equation = var_687_equation_0, values = (var_491_cast_fp16, var_652_cast_fp16))[name = string("op_687_cast_fp16")]; + string var_689_equation_0 = const()[name = string("op_689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_689_cast_fp16 = einsum(equation = var_689_equation_0, values = (var_491_cast_fp16, var_653_cast_fp16))[name = string("op_689_cast_fp16")]; + string var_691_equation_0 = const()[name = string("op_691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_691_cast_fp16 = einsum(equation = var_691_equation_0, values = (var_495_cast_fp16, var_654_cast_fp16))[name = string("op_691_cast_fp16")]; + string var_693_equation_0 = const()[name = string("op_693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_693_cast_fp16 = einsum(equation = var_693_equation_0, values = (var_495_cast_fp16, var_655_cast_fp16))[name = string("op_693_cast_fp16")]; + string var_695_equation_0 = const()[name = string("op_695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_695_cast_fp16 = einsum(equation = var_695_equation_0, values = (var_495_cast_fp16, var_656_cast_fp16))[name = string("op_695_cast_fp16")]; + string var_697_equation_0 = const()[name = string("op_697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_697_cast_fp16 = einsum(equation = var_697_equation_0, values = (var_495_cast_fp16, var_657_cast_fp16))[name = string("op_697_cast_fp16")]; + string var_699_equation_0 = const()[name = string("op_699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_699_cast_fp16 = einsum(equation = var_699_equation_0, values = (var_499_cast_fp16, var_658_cast_fp16))[name = string("op_699_cast_fp16")]; + string var_701_equation_0 = const()[name = string("op_701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_701_cast_fp16 = einsum(equation = var_701_equation_0, values = (var_499_cast_fp16, var_659_cast_fp16))[name = string("op_701_cast_fp16")]; + string var_703_equation_0 = const()[name = string("op_703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_703_cast_fp16 = einsum(equation = var_703_equation_0, values = (var_499_cast_fp16, var_660_cast_fp16))[name = string("op_703_cast_fp16")]; + string var_705_equation_0 = const()[name = string("op_705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_705_cast_fp16 = einsum(equation = var_705_equation_0, values = (var_499_cast_fp16, var_661_cast_fp16))[name = string("op_705_cast_fp16")]; + string var_707_equation_0 = const()[name = string("op_707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_707_cast_fp16 = einsum(equation = var_707_equation_0, values = (var_503_cast_fp16, var_662_cast_fp16))[name = string("op_707_cast_fp16")]; + string var_709_equation_0 = const()[name = string("op_709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_709_cast_fp16 = einsum(equation = var_709_equation_0, values = (var_503_cast_fp16, var_663_cast_fp16))[name = string("op_709_cast_fp16")]; + string var_711_equation_0 = const()[name = string("op_711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_711_cast_fp16 = einsum(equation = var_711_equation_0, values = (var_503_cast_fp16, var_664_cast_fp16))[name = string("op_711_cast_fp16")]; + string var_713_equation_0 = const()[name = string("op_713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_713_cast_fp16 = einsum(equation = var_713_equation_0, values = (var_503_cast_fp16, var_665_cast_fp16))[name = string("op_713_cast_fp16")]; + string var_715_equation_0 = const()[name = string("op_715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_715_cast_fp16 = einsum(equation = var_715_equation_0, values = (var_507_cast_fp16, var_666_cast_fp16))[name = string("op_715_cast_fp16")]; + string var_717_equation_0 = const()[name = string("op_717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_717_cast_fp16 = einsum(equation = var_717_equation_0, values = (var_507_cast_fp16, var_667_cast_fp16))[name = string("op_717_cast_fp16")]; + string var_719_equation_0 = const()[name = string("op_719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_719_cast_fp16 = einsum(equation = var_719_equation_0, values = (var_507_cast_fp16, var_668_cast_fp16))[name = string("op_719_cast_fp16")]; + string var_721_equation_0 = const()[name = string("op_721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_721_cast_fp16 = einsum(equation = var_721_equation_0, values = (var_507_cast_fp16, var_669_cast_fp16))[name = string("op_721_cast_fp16")]; + string var_723_equation_0 = const()[name = string("op_723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_723_cast_fp16 = einsum(equation = var_723_equation_0, values = (var_511_cast_fp16, var_670_cast_fp16))[name = string("op_723_cast_fp16")]; + string var_725_equation_0 = const()[name = string("op_725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_725_cast_fp16 = einsum(equation = var_725_equation_0, values = (var_511_cast_fp16, var_671_cast_fp16))[name = string("op_725_cast_fp16")]; + string var_727_equation_0 = const()[name = string("op_727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_727_cast_fp16 = einsum(equation = var_727_equation_0, values = (var_511_cast_fp16, var_672_cast_fp16))[name = string("op_727_cast_fp16")]; + string var_729_equation_0 = const()[name = string("op_729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_729_cast_fp16 = einsum(equation = var_729_equation_0, values = (var_511_cast_fp16, var_673_cast_fp16))[name = string("op_729_cast_fp16")]; + string var_731_equation_0 = const()[name = string("op_731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_731_cast_fp16 = einsum(equation = var_731_equation_0, values = (var_515_cast_fp16, var_674_cast_fp16))[name = string("op_731_cast_fp16")]; + string var_733_equation_0 = const()[name = string("op_733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_733_cast_fp16 = einsum(equation = var_733_equation_0, values = (var_515_cast_fp16, var_675_cast_fp16))[name = string("op_733_cast_fp16")]; + string var_735_equation_0 = const()[name = string("op_735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_735_cast_fp16 = einsum(equation = var_735_equation_0, values = (var_515_cast_fp16, var_676_cast_fp16))[name = string("op_735_cast_fp16")]; + string var_737_equation_0 = const()[name = string("op_737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_737_cast_fp16 = einsum(equation = var_737_equation_0, values = (var_515_cast_fp16, var_677_cast_fp16))[name = string("op_737_cast_fp16")]; + string var_739_equation_0 = const()[name = string("op_739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_739_cast_fp16 = einsum(equation = var_739_equation_0, values = (var_519_cast_fp16, var_678_cast_fp16))[name = string("op_739_cast_fp16")]; + string var_741_equation_0 = const()[name = string("op_741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_741_cast_fp16 = einsum(equation = var_741_equation_0, values = (var_519_cast_fp16, var_679_cast_fp16))[name = string("op_741_cast_fp16")]; + string var_743_equation_0 = const()[name = string("op_743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_743_cast_fp16 = einsum(equation = var_743_equation_0, values = (var_519_cast_fp16, var_680_cast_fp16))[name = string("op_743_cast_fp16")]; + string var_745_equation_0 = const()[name = string("op_745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_745_cast_fp16 = einsum(equation = var_745_equation_0, values = (var_519_cast_fp16, var_681_cast_fp16))[name = string("op_745_cast_fp16")]; + bool var_747_interleave_0 = const()[name = string("op_747_interleave_0"), val = bool(false)]; + tensor var_747_cast_fp16 = concat(axis = var_134, interleave = var_747_interleave_0, values = (var_683_cast_fp16, var_685_cast_fp16, var_687_cast_fp16, var_689_cast_fp16))[name = string("op_747_cast_fp16")]; + bool var_749_interleave_0 = const()[name = string("op_749_interleave_0"), val = bool(false)]; + tensor var_749_cast_fp16 = concat(axis = var_134, interleave = var_749_interleave_0, values = (var_691_cast_fp16, var_693_cast_fp16, var_695_cast_fp16, var_697_cast_fp16))[name = string("op_749_cast_fp16")]; + bool var_751_interleave_0 = const()[name = string("op_751_interleave_0"), val = bool(false)]; + tensor var_751_cast_fp16 = concat(axis = var_134, interleave = var_751_interleave_0, values = (var_699_cast_fp16, var_701_cast_fp16, var_703_cast_fp16, var_705_cast_fp16))[name = string("op_751_cast_fp16")]; + bool var_753_interleave_0 = const()[name = string("op_753_interleave_0"), val = bool(false)]; + tensor var_753_cast_fp16 = concat(axis = var_134, interleave = var_753_interleave_0, values = (var_707_cast_fp16, var_709_cast_fp16, var_711_cast_fp16, var_713_cast_fp16))[name = string("op_753_cast_fp16")]; + bool var_755_interleave_0 = const()[name = string("op_755_interleave_0"), val = bool(false)]; + tensor var_755_cast_fp16 = concat(axis = var_134, interleave = var_755_interleave_0, values = (var_715_cast_fp16, var_717_cast_fp16, var_719_cast_fp16, var_721_cast_fp16))[name = string("op_755_cast_fp16")]; + bool var_757_interleave_0 = const()[name = string("op_757_interleave_0"), val = bool(false)]; + tensor var_757_cast_fp16 = concat(axis = var_134, interleave = var_757_interleave_0, values = (var_723_cast_fp16, var_725_cast_fp16, var_727_cast_fp16, var_729_cast_fp16))[name = string("op_757_cast_fp16")]; + bool var_759_interleave_0 = const()[name = string("op_759_interleave_0"), val = bool(false)]; + tensor var_759_cast_fp16 = concat(axis = var_134, interleave = var_759_interleave_0, values = (var_731_cast_fp16, var_733_cast_fp16, var_735_cast_fp16, var_737_cast_fp16))[name = string("op_759_cast_fp16")]; + bool var_761_interleave_0 = const()[name = string("op_761_interleave_0"), val = bool(false)]; + tensor var_761_cast_fp16 = concat(axis = var_134, interleave = var_761_interleave_0, values = (var_739_cast_fp16, var_741_cast_fp16, var_743_cast_fp16, var_745_cast_fp16))[name = string("op_761_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_147, interleave = input_1_interleave_0, values = (var_747_cast_fp16, var_749_cast_fp16, var_751_cast_fp16, var_753_cast_fp16, var_755_cast_fp16, var_757_cast_fp16, var_759_cast_fp16, var_761_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936640)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5460992)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_780_to_fp16 = const()[name = string("op_780_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_780_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5462080)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5463168)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5464256)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7561472)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7565632)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9662848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_809 = const()[name = string("op_809"), val = int32(3)]; + int32 var_822 = const()[name = string("op_822"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_839_to_fp16 = const()[name = string("op_839_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_839_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9663936)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9665024)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9666112)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10190464)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10191552)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10715904)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11240256)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_877_begin_0 = const()[name = string("op_877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_877_end_0 = const()[name = string("op_877_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_877_end_mask_0 = const()[name = string("op_877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_877_cast_fp16 = slice_by_index(begin = var_877_begin_0, end = var_877_end_0, end_mask = var_877_end_mask_0, x = query_3_cast_fp16)[name = string("op_877_cast_fp16")]; + tensor var_881_begin_0 = const()[name = string("op_881_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_881_end_0 = const()[name = string("op_881_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_881_end_mask_0 = const()[name = string("op_881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_881_cast_fp16 = slice_by_index(begin = var_881_begin_0, end = var_881_end_0, end_mask = var_881_end_mask_0, x = query_3_cast_fp16)[name = string("op_881_cast_fp16")]; + tensor var_885_begin_0 = const()[name = string("op_885_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_885_end_0 = const()[name = string("op_885_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_885_end_mask_0 = const()[name = string("op_885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_885_cast_fp16 = slice_by_index(begin = var_885_begin_0, end = var_885_end_0, end_mask = var_885_end_mask_0, x = query_3_cast_fp16)[name = string("op_885_cast_fp16")]; + tensor var_889_begin_0 = const()[name = string("op_889_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_889_end_0 = const()[name = string("op_889_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_889_end_mask_0 = const()[name = string("op_889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_889_cast_fp16 = slice_by_index(begin = var_889_begin_0, end = var_889_end_0, end_mask = var_889_end_mask_0, x = query_3_cast_fp16)[name = string("op_889_cast_fp16")]; + tensor var_893_begin_0 = const()[name = string("op_893_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_893_end_0 = const()[name = string("op_893_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_893_end_mask_0 = const()[name = string("op_893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_893_cast_fp16 = slice_by_index(begin = var_893_begin_0, end = var_893_end_0, end_mask = var_893_end_mask_0, x = query_3_cast_fp16)[name = string("op_893_cast_fp16")]; + tensor var_897_begin_0 = const()[name = string("op_897_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_897_end_0 = const()[name = string("op_897_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_897_end_mask_0 = const()[name = string("op_897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_897_cast_fp16 = slice_by_index(begin = var_897_begin_0, end = var_897_end_0, end_mask = var_897_end_mask_0, x = query_3_cast_fp16)[name = string("op_897_cast_fp16")]; + tensor var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = query_3_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_905_begin_0 = const()[name = string("op_905_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_905_end_0 = const()[name = string("op_905_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_905_end_mask_0 = const()[name = string("op_905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_905_cast_fp16 = slice_by_index(begin = var_905_begin_0, end = var_905_end_0, end_mask = var_905_end_mask_0, x = query_3_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_914_begin_0 = const()[name = string("op_914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_914_end_0 = const()[name = string("op_914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_914_end_mask_0 = const()[name = string("op_914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_914_cast_fp16 = slice_by_index(begin = var_914_begin_0, end = var_914_end_0, end_mask = var_914_end_mask_0, x = var_877_cast_fp16)[name = string("op_914_cast_fp16")]; + tensor var_921_begin_0 = const()[name = string("op_921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_921_end_0 = const()[name = string("op_921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_921_end_mask_0 = const()[name = string("op_921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_921_cast_fp16 = slice_by_index(begin = var_921_begin_0, end = var_921_end_0, end_mask = var_921_end_mask_0, x = var_877_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_928_begin_0 = const()[name = string("op_928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_928_end_0 = const()[name = string("op_928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_928_end_mask_0 = const()[name = string("op_928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_928_cast_fp16 = slice_by_index(begin = var_928_begin_0, end = var_928_end_0, end_mask = var_928_end_mask_0, x = var_877_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_935_begin_0 = const()[name = string("op_935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_935_end_0 = const()[name = string("op_935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_935_end_mask_0 = const()[name = string("op_935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_935_cast_fp16 = slice_by_index(begin = var_935_begin_0, end = var_935_end_0, end_mask = var_935_end_mask_0, x = var_877_cast_fp16)[name = string("op_935_cast_fp16")]; + tensor var_942_begin_0 = const()[name = string("op_942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_942_end_0 = const()[name = string("op_942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_942_end_mask_0 = const()[name = string("op_942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = var_881_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = var_881_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_956_begin_0 = const()[name = string("op_956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_956_end_0 = const()[name = string("op_956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_956_end_mask_0 = const()[name = string("op_956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_956_cast_fp16 = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, x = var_881_cast_fp16)[name = string("op_956_cast_fp16")]; + tensor var_963_begin_0 = const()[name = string("op_963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_963_end_0 = const()[name = string("op_963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_963_end_mask_0 = const()[name = string("op_963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_963_cast_fp16 = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = var_881_cast_fp16)[name = string("op_963_cast_fp16")]; + tensor var_970_begin_0 = const()[name = string("op_970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_end_0 = const()[name = string("op_970_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_970_end_mask_0 = const()[name = string("op_970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, x = var_885_cast_fp16)[name = string("op_970_cast_fp16")]; + tensor var_977_begin_0 = const()[name = string("op_977_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_977_end_0 = const()[name = string("op_977_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_977_end_mask_0 = const()[name = string("op_977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = var_885_cast_fp16)[name = string("op_977_cast_fp16")]; + tensor var_984_begin_0 = const()[name = string("op_984_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_984_end_0 = const()[name = string("op_984_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_984_end_mask_0 = const()[name = string("op_984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_984_cast_fp16 = slice_by_index(begin = var_984_begin_0, end = var_984_end_0, end_mask = var_984_end_mask_0, x = var_885_cast_fp16)[name = string("op_984_cast_fp16")]; + tensor var_991_begin_0 = const()[name = string("op_991_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_991_end_0 = const()[name = string("op_991_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_991_end_mask_0 = const()[name = string("op_991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = var_885_cast_fp16)[name = string("op_991_cast_fp16")]; + tensor var_998_begin_0 = const()[name = string("op_998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_998_end_0 = const()[name = string("op_998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_998_end_mask_0 = const()[name = string("op_998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_998_cast_fp16 = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = var_889_cast_fp16)[name = string("op_998_cast_fp16")]; + tensor var_1005_begin_0 = const()[name = string("op_1005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1005_end_0 = const()[name = string("op_1005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1005_end_mask_0 = const()[name = string("op_1005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1005_cast_fp16 = slice_by_index(begin = var_1005_begin_0, end = var_1005_end_0, end_mask = var_1005_end_mask_0, x = var_889_cast_fp16)[name = string("op_1005_cast_fp16")]; + tensor var_1012_begin_0 = const()[name = string("op_1012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1012_end_0 = const()[name = string("op_1012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1012_end_mask_0 = const()[name = string("op_1012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = var_889_cast_fp16)[name = string("op_1012_cast_fp16")]; + tensor var_1019_begin_0 = const()[name = string("op_1019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1019_end_0 = const()[name = string("op_1019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1019_end_mask_0 = const()[name = string("op_1019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = var_1019_end_0, end_mask = var_1019_end_mask_0, x = var_889_cast_fp16)[name = string("op_1019_cast_fp16")]; + tensor var_1026_begin_0 = const()[name = string("op_1026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1026_end_0 = const()[name = string("op_1026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1026_end_mask_0 = const()[name = string("op_1026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1026_cast_fp16 = slice_by_index(begin = var_1026_begin_0, end = var_1026_end_0, end_mask = var_1026_end_mask_0, x = var_893_cast_fp16)[name = string("op_1026_cast_fp16")]; + tensor var_1033_begin_0 = const()[name = string("op_1033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1033_end_0 = const()[name = string("op_1033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1033_end_mask_0 = const()[name = string("op_1033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1033_cast_fp16 = slice_by_index(begin = var_1033_begin_0, end = var_1033_end_0, end_mask = var_1033_end_mask_0, x = var_893_cast_fp16)[name = string("op_1033_cast_fp16")]; + tensor var_1040_begin_0 = const()[name = string("op_1040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1040_end_0 = const()[name = string("op_1040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1040_end_mask_0 = const()[name = string("op_1040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1040_cast_fp16 = slice_by_index(begin = var_1040_begin_0, end = var_1040_end_0, end_mask = var_1040_end_mask_0, x = var_893_cast_fp16)[name = string("op_1040_cast_fp16")]; + tensor var_1047_begin_0 = const()[name = string("op_1047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1047_end_0 = const()[name = string("op_1047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1047_end_mask_0 = const()[name = string("op_1047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = var_893_cast_fp16)[name = string("op_1047_cast_fp16")]; + tensor var_1054_begin_0 = const()[name = string("op_1054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1054_end_0 = const()[name = string("op_1054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1054_end_mask_0 = const()[name = string("op_1054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1054_cast_fp16 = slice_by_index(begin = var_1054_begin_0, end = var_1054_end_0, end_mask = var_1054_end_mask_0, x = var_897_cast_fp16)[name = string("op_1054_cast_fp16")]; + tensor var_1061_begin_0 = const()[name = string("op_1061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1061_end_0 = const()[name = string("op_1061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1061_end_mask_0 = const()[name = string("op_1061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1061_cast_fp16 = slice_by_index(begin = var_1061_begin_0, end = var_1061_end_0, end_mask = var_1061_end_mask_0, x = var_897_cast_fp16)[name = string("op_1061_cast_fp16")]; + tensor var_1068_begin_0 = const()[name = string("op_1068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1068_end_0 = const()[name = string("op_1068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1068_end_mask_0 = const()[name = string("op_1068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1068_cast_fp16 = slice_by_index(begin = var_1068_begin_0, end = var_1068_end_0, end_mask = var_1068_end_mask_0, x = var_897_cast_fp16)[name = string("op_1068_cast_fp16")]; + tensor var_1075_begin_0 = const()[name = string("op_1075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1075_end_0 = const()[name = string("op_1075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1075_end_mask_0 = const()[name = string("op_1075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1075_cast_fp16 = slice_by_index(begin = var_1075_begin_0, end = var_1075_end_0, end_mask = var_1075_end_mask_0, x = var_897_cast_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1082_begin_0 = const()[name = string("op_1082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1082_end_0 = const()[name = string("op_1082_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1082_end_mask_0 = const()[name = string("op_1082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1082_cast_fp16 = slice_by_index(begin = var_1082_begin_0, end = var_1082_end_0, end_mask = var_1082_end_mask_0, x = var_901_cast_fp16)[name = string("op_1082_cast_fp16")]; + tensor var_1089_begin_0 = const()[name = string("op_1089_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1089_end_0 = const()[name = string("op_1089_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1089_end_mask_0 = const()[name = string("op_1089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1089_cast_fp16 = slice_by_index(begin = var_1089_begin_0, end = var_1089_end_0, end_mask = var_1089_end_mask_0, x = var_901_cast_fp16)[name = string("op_1089_cast_fp16")]; + tensor var_1096_begin_0 = const()[name = string("op_1096_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1096_end_0 = const()[name = string("op_1096_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1096_end_mask_0 = const()[name = string("op_1096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1096_cast_fp16 = slice_by_index(begin = var_1096_begin_0, end = var_1096_end_0, end_mask = var_1096_end_mask_0, x = var_901_cast_fp16)[name = string("op_1096_cast_fp16")]; + tensor var_1103_begin_0 = const()[name = string("op_1103_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1103_end_0 = const()[name = string("op_1103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1103_end_mask_0 = const()[name = string("op_1103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1103_cast_fp16 = slice_by_index(begin = var_1103_begin_0, end = var_1103_end_0, end_mask = var_1103_end_mask_0, x = var_901_cast_fp16)[name = string("op_1103_cast_fp16")]; + tensor var_1110_begin_0 = const()[name = string("op_1110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1110_end_0 = const()[name = string("op_1110_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1110_end_mask_0 = const()[name = string("op_1110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1110_cast_fp16 = slice_by_index(begin = var_1110_begin_0, end = var_1110_end_0, end_mask = var_1110_end_mask_0, x = var_905_cast_fp16)[name = string("op_1110_cast_fp16")]; + tensor var_1117_begin_0 = const()[name = string("op_1117_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1117_end_0 = const()[name = string("op_1117_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1117_end_mask_0 = const()[name = string("op_1117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1117_cast_fp16 = slice_by_index(begin = var_1117_begin_0, end = var_1117_end_0, end_mask = var_1117_end_mask_0, x = var_905_cast_fp16)[name = string("op_1117_cast_fp16")]; + tensor var_1124_begin_0 = const()[name = string("op_1124_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1124_end_0 = const()[name = string("op_1124_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1124_end_mask_0 = const()[name = string("op_1124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1124_cast_fp16 = slice_by_index(begin = var_1124_begin_0, end = var_1124_end_0, end_mask = var_1124_end_mask_0, x = var_905_cast_fp16)[name = string("op_1124_cast_fp16")]; + tensor var_1131_begin_0 = const()[name = string("op_1131_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1131_end_0 = const()[name = string("op_1131_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1131_end_mask_0 = const()[name = string("op_1131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1131_cast_fp16 = slice_by_index(begin = var_1131_begin_0, end = var_1131_end_0, end_mask = var_1131_end_mask_0, x = var_905_cast_fp16)[name = string("op_1131_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1136_begin_0 = const()[name = string("op_1136_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1136_end_0 = const()[name = string("op_1136_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1136_end_mask_0 = const()[name = string("op_1136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_4")]; + tensor var_1136_cast_fp16 = slice_by_index(begin = var_1136_begin_0, end = var_1136_end_0, end_mask = var_1136_end_mask_0, x = k_3_cast_fp16)[name = string("op_1136_cast_fp16")]; + tensor var_1140_begin_0 = const()[name = string("op_1140_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1140_end_0 = const()[name = string("op_1140_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1140_end_mask_0 = const()[name = string("op_1140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1140_cast_fp16 = slice_by_index(begin = var_1140_begin_0, end = var_1140_end_0, end_mask = var_1140_end_mask_0, x = k_3_cast_fp16)[name = string("op_1140_cast_fp16")]; + tensor var_1144_begin_0 = const()[name = string("op_1144_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1144_end_0 = const()[name = string("op_1144_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1144_end_mask_0 = const()[name = string("op_1144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1144_cast_fp16 = slice_by_index(begin = var_1144_begin_0, end = var_1144_end_0, end_mask = var_1144_end_mask_0, x = k_3_cast_fp16)[name = string("op_1144_cast_fp16")]; + tensor var_1148_begin_0 = const()[name = string("op_1148_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1148_end_0 = const()[name = string("op_1148_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1148_end_mask_0 = const()[name = string("op_1148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1148_cast_fp16 = slice_by_index(begin = var_1148_begin_0, end = var_1148_end_0, end_mask = var_1148_end_mask_0, x = k_3_cast_fp16)[name = string("op_1148_cast_fp16")]; + tensor var_1152_begin_0 = const()[name = string("op_1152_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1152_end_0 = const()[name = string("op_1152_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1152_end_mask_0 = const()[name = string("op_1152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1152_cast_fp16 = slice_by_index(begin = var_1152_begin_0, end = var_1152_end_0, end_mask = var_1152_end_mask_0, x = k_3_cast_fp16)[name = string("op_1152_cast_fp16")]; + tensor var_1156_begin_0 = const()[name = string("op_1156_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1156_end_0 = const()[name = string("op_1156_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1156_end_mask_0 = const()[name = string("op_1156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1156_cast_fp16 = slice_by_index(begin = var_1156_begin_0, end = var_1156_end_0, end_mask = var_1156_end_mask_0, x = k_3_cast_fp16)[name = string("op_1156_cast_fp16")]; + tensor var_1160_begin_0 = const()[name = string("op_1160_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1160_end_0 = const()[name = string("op_1160_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1160_end_mask_0 = const()[name = string("op_1160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1160_cast_fp16 = slice_by_index(begin = var_1160_begin_0, end = var_1160_end_0, end_mask = var_1160_end_mask_0, x = k_3_cast_fp16)[name = string("op_1160_cast_fp16")]; + tensor var_1164_begin_0 = const()[name = string("op_1164_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1164_end_0 = const()[name = string("op_1164_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1164_end_mask_0 = const()[name = string("op_1164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1164_cast_fp16 = slice_by_index(begin = var_1164_begin_0, end = var_1164_end_0, end_mask = var_1164_end_mask_0, x = k_3_cast_fp16)[name = string("op_1164_cast_fp16")]; + tensor var_1166_begin_0 = const()[name = string("op_1166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1166_end_0 = const()[name = string("op_1166_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1166_end_mask_0 = const()[name = string("op_1166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1166_cast_fp16 = slice_by_index(begin = var_1166_begin_0, end = var_1166_end_0, end_mask = var_1166_end_mask_0, x = value_3_cast_fp16)[name = string("op_1166_cast_fp16")]; + tensor var_1170_begin_0 = const()[name = string("op_1170_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1170_end_0 = const()[name = string("op_1170_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1170_end_mask_0 = const()[name = string("op_1170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1170_cast_fp16 = slice_by_index(begin = var_1170_begin_0, end = var_1170_end_0, end_mask = var_1170_end_mask_0, x = value_3_cast_fp16)[name = string("op_1170_cast_fp16")]; + tensor var_1174_begin_0 = const()[name = string("op_1174_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1174_end_0 = const()[name = string("op_1174_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1174_end_mask_0 = const()[name = string("op_1174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1174_cast_fp16 = slice_by_index(begin = var_1174_begin_0, end = var_1174_end_0, end_mask = var_1174_end_mask_0, x = value_3_cast_fp16)[name = string("op_1174_cast_fp16")]; + tensor var_1178_begin_0 = const()[name = string("op_1178_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1178_end_0 = const()[name = string("op_1178_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1178_end_mask_0 = const()[name = string("op_1178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1178_cast_fp16 = slice_by_index(begin = var_1178_begin_0, end = var_1178_end_0, end_mask = var_1178_end_mask_0, x = value_3_cast_fp16)[name = string("op_1178_cast_fp16")]; + tensor var_1182_begin_0 = const()[name = string("op_1182_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1182_end_0 = const()[name = string("op_1182_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1182_end_mask_0 = const()[name = string("op_1182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1182_cast_fp16 = slice_by_index(begin = var_1182_begin_0, end = var_1182_end_0, end_mask = var_1182_end_mask_0, x = value_3_cast_fp16)[name = string("op_1182_cast_fp16")]; + tensor var_1186_begin_0 = const()[name = string("op_1186_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1186_end_0 = const()[name = string("op_1186_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1186_end_mask_0 = const()[name = string("op_1186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1186_cast_fp16 = slice_by_index(begin = var_1186_begin_0, end = var_1186_end_0, end_mask = var_1186_end_mask_0, x = value_3_cast_fp16)[name = string("op_1186_cast_fp16")]; + tensor var_1190_begin_0 = const()[name = string("op_1190_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1190_end_0 = const()[name = string("op_1190_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1190_end_mask_0 = const()[name = string("op_1190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1190_cast_fp16 = slice_by_index(begin = var_1190_begin_0, end = var_1190_end_0, end_mask = var_1190_end_mask_0, x = value_3_cast_fp16)[name = string("op_1190_cast_fp16")]; + tensor var_1194_begin_0 = const()[name = string("op_1194_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1194_end_0 = const()[name = string("op_1194_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1194_end_mask_0 = const()[name = string("op_1194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, x = value_3_cast_fp16)[name = string("op_1194_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_1136_cast_fp16, var_914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_1136_cast_fp16, var_921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_1136_cast_fp16, var_928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_1136_cast_fp16, var_935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_1140_cast_fp16, var_942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_1140_cast_fp16, var_949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_1140_cast_fp16, var_956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_1140_cast_fp16, var_963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_1144_cast_fp16, var_970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_1144_cast_fp16, var_977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_1144_cast_fp16, var_984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_1144_cast_fp16, var_991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_1148_cast_fp16, var_998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_1148_cast_fp16, var_1005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_1148_cast_fp16, var_1012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_1148_cast_fp16, var_1019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1152_cast_fp16, var_1026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1152_cast_fp16, var_1033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1152_cast_fp16, var_1040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1152_cast_fp16, var_1047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1156_cast_fp16, var_1054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1156_cast_fp16, var_1061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1156_cast_fp16, var_1068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1156_cast_fp16, var_1075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1160_cast_fp16, var_1082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1160_cast_fp16, var_1089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1160_cast_fp16, var_1096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1160_cast_fp16, var_1103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1164_cast_fp16, var_1110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1164_cast_fp16, var_1117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1164_cast_fp16, var_1124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1164_cast_fp16, var_1131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + fp16 var_1261_to_fp16 = const()[name = string("op_1261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1261_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1263_to_fp16 = const()[name = string("op_1263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1263_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1265_to_fp16 = const()[name = string("op_1265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1265_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1267_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1269_to_fp16 = const()[name = string("op_1269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1269_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1271_to_fp16 = const()[name = string("op_1271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1271_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1273_to_fp16 = const()[name = string("op_1273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1273_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1275_to_fp16 = const()[name = string("op_1275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1275_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1277_to_fp16 = const()[name = string("op_1277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1277_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1279_to_fp16 = const()[name = string("op_1279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1279_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1281_to_fp16 = const()[name = string("op_1281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1281_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1283_to_fp16 = const()[name = string("op_1283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1283_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1285_to_fp16 = const()[name = string("op_1285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1285_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1287_to_fp16 = const()[name = string("op_1287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1287_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1289_to_fp16 = const()[name = string("op_1289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1289_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1291_to_fp16 = const()[name = string("op_1291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1291_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + fp16 var_1293_to_fp16 = const()[name = string("op_1293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1293_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1295_to_fp16 = const()[name = string("op_1295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1295_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1297_to_fp16 = const()[name = string("op_1297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1297_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1299_to_fp16 = const()[name = string("op_1299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1299_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1301_to_fp16 = const()[name = string("op_1301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1301_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1303_to_fp16 = const()[name = string("op_1303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1303_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1305_to_fp16 = const()[name = string("op_1305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1305_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1307_to_fp16 = const()[name = string("op_1307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1307_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1309_to_fp16 = const()[name = string("op_1309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1309_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1311_to_fp16 = const()[name = string("op_1311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1311_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1313_to_fp16 = const()[name = string("op_1313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1313_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1315_to_fp16 = const()[name = string("op_1315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1315_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1317_to_fp16 = const()[name = string("op_1317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1317_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1319_to_fp16 = const()[name = string("op_1319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1319_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1321_to_fp16 = const()[name = string("op_1321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1321_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1323_to_fp16 = const()[name = string("op_1323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1323_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + tensor var_1325_cast_fp16 = softmax(axis = var_822, x = aw_chunk_65_cast_fp16)[name = string("op_1325_cast_fp16")]; + tensor var_1326_cast_fp16 = softmax(axis = var_822, x = aw_chunk_67_cast_fp16)[name = string("op_1326_cast_fp16")]; + tensor var_1327_cast_fp16 = softmax(axis = var_822, x = aw_chunk_69_cast_fp16)[name = string("op_1327_cast_fp16")]; + tensor var_1328_cast_fp16 = softmax(axis = var_822, x = aw_chunk_71_cast_fp16)[name = string("op_1328_cast_fp16")]; + tensor var_1329_cast_fp16 = softmax(axis = var_822, x = aw_chunk_73_cast_fp16)[name = string("op_1329_cast_fp16")]; + tensor var_1330_cast_fp16 = softmax(axis = var_822, x = aw_chunk_75_cast_fp16)[name = string("op_1330_cast_fp16")]; + tensor var_1331_cast_fp16 = softmax(axis = var_822, x = aw_chunk_77_cast_fp16)[name = string("op_1331_cast_fp16")]; + tensor var_1332_cast_fp16 = softmax(axis = var_822, x = aw_chunk_79_cast_fp16)[name = string("op_1332_cast_fp16")]; + tensor var_1333_cast_fp16 = softmax(axis = var_822, x = aw_chunk_81_cast_fp16)[name = string("op_1333_cast_fp16")]; + tensor var_1334_cast_fp16 = softmax(axis = var_822, x = aw_chunk_83_cast_fp16)[name = string("op_1334_cast_fp16")]; + tensor var_1335_cast_fp16 = softmax(axis = var_822, x = aw_chunk_85_cast_fp16)[name = string("op_1335_cast_fp16")]; + tensor var_1336_cast_fp16 = softmax(axis = var_822, x = aw_chunk_87_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor var_1337_cast_fp16 = softmax(axis = var_822, x = aw_chunk_89_cast_fp16)[name = string("op_1337_cast_fp16")]; + tensor var_1338_cast_fp16 = softmax(axis = var_822, x = aw_chunk_91_cast_fp16)[name = string("op_1338_cast_fp16")]; + tensor var_1339_cast_fp16 = softmax(axis = var_822, x = aw_chunk_93_cast_fp16)[name = string("op_1339_cast_fp16")]; + tensor var_1340_cast_fp16 = softmax(axis = var_822, x = aw_chunk_95_cast_fp16)[name = string("op_1340_cast_fp16")]; + tensor var_1341_cast_fp16 = softmax(axis = var_822, x = aw_chunk_97_cast_fp16)[name = string("op_1341_cast_fp16")]; + tensor var_1342_cast_fp16 = softmax(axis = var_822, x = aw_chunk_99_cast_fp16)[name = string("op_1342_cast_fp16")]; + tensor var_1343_cast_fp16 = softmax(axis = var_822, x = aw_chunk_101_cast_fp16)[name = string("op_1343_cast_fp16")]; + tensor var_1344_cast_fp16 = softmax(axis = var_822, x = aw_chunk_103_cast_fp16)[name = string("op_1344_cast_fp16")]; + tensor var_1345_cast_fp16 = softmax(axis = var_822, x = aw_chunk_105_cast_fp16)[name = string("op_1345_cast_fp16")]; + tensor var_1346_cast_fp16 = softmax(axis = var_822, x = aw_chunk_107_cast_fp16)[name = string("op_1346_cast_fp16")]; + tensor var_1347_cast_fp16 = softmax(axis = var_822, x = aw_chunk_109_cast_fp16)[name = string("op_1347_cast_fp16")]; + tensor var_1348_cast_fp16 = softmax(axis = var_822, x = aw_chunk_111_cast_fp16)[name = string("op_1348_cast_fp16")]; + tensor var_1349_cast_fp16 = softmax(axis = var_822, x = aw_chunk_113_cast_fp16)[name = string("op_1349_cast_fp16")]; + tensor var_1350_cast_fp16 = softmax(axis = var_822, x = aw_chunk_115_cast_fp16)[name = string("op_1350_cast_fp16")]; + tensor var_1351_cast_fp16 = softmax(axis = var_822, x = aw_chunk_117_cast_fp16)[name = string("op_1351_cast_fp16")]; + tensor var_1352_cast_fp16 = softmax(axis = var_822, x = aw_chunk_119_cast_fp16)[name = string("op_1352_cast_fp16")]; + tensor var_1353_cast_fp16 = softmax(axis = var_822, x = aw_chunk_121_cast_fp16)[name = string("op_1353_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_822, x = aw_chunk_123_cast_fp16)[name = string("op_1354_cast_fp16")]; + tensor var_1355_cast_fp16 = softmax(axis = var_822, x = aw_chunk_125_cast_fp16)[name = string("op_1355_cast_fp16")]; + tensor var_1356_cast_fp16 = softmax(axis = var_822, x = aw_chunk_127_cast_fp16)[name = string("op_1356_cast_fp16")]; + string var_1358_equation_0 = const()[name = string("op_1358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1358_cast_fp16 = einsum(equation = var_1358_equation_0, values = (var_1166_cast_fp16, var_1325_cast_fp16))[name = string("op_1358_cast_fp16")]; + string var_1360_equation_0 = const()[name = string("op_1360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1360_cast_fp16 = einsum(equation = var_1360_equation_0, values = (var_1166_cast_fp16, var_1326_cast_fp16))[name = string("op_1360_cast_fp16")]; + string var_1362_equation_0 = const()[name = string("op_1362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1362_cast_fp16 = einsum(equation = var_1362_equation_0, values = (var_1166_cast_fp16, var_1327_cast_fp16))[name = string("op_1362_cast_fp16")]; + string var_1364_equation_0 = const()[name = string("op_1364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1364_cast_fp16 = einsum(equation = var_1364_equation_0, values = (var_1166_cast_fp16, var_1328_cast_fp16))[name = string("op_1364_cast_fp16")]; + string var_1366_equation_0 = const()[name = string("op_1366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1366_cast_fp16 = einsum(equation = var_1366_equation_0, values = (var_1170_cast_fp16, var_1329_cast_fp16))[name = string("op_1366_cast_fp16")]; + string var_1368_equation_0 = const()[name = string("op_1368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1368_cast_fp16 = einsum(equation = var_1368_equation_0, values = (var_1170_cast_fp16, var_1330_cast_fp16))[name = string("op_1368_cast_fp16")]; + string var_1370_equation_0 = const()[name = string("op_1370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1370_cast_fp16 = einsum(equation = var_1370_equation_0, values = (var_1170_cast_fp16, var_1331_cast_fp16))[name = string("op_1370_cast_fp16")]; + string var_1372_equation_0 = const()[name = string("op_1372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1372_cast_fp16 = einsum(equation = var_1372_equation_0, values = (var_1170_cast_fp16, var_1332_cast_fp16))[name = string("op_1372_cast_fp16")]; + string var_1374_equation_0 = const()[name = string("op_1374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1374_cast_fp16 = einsum(equation = var_1374_equation_0, values = (var_1174_cast_fp16, var_1333_cast_fp16))[name = string("op_1374_cast_fp16")]; + string var_1376_equation_0 = const()[name = string("op_1376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1376_cast_fp16 = einsum(equation = var_1376_equation_0, values = (var_1174_cast_fp16, var_1334_cast_fp16))[name = string("op_1376_cast_fp16")]; + string var_1378_equation_0 = const()[name = string("op_1378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1378_cast_fp16 = einsum(equation = var_1378_equation_0, values = (var_1174_cast_fp16, var_1335_cast_fp16))[name = string("op_1378_cast_fp16")]; + string var_1380_equation_0 = const()[name = string("op_1380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1380_cast_fp16 = einsum(equation = var_1380_equation_0, values = (var_1174_cast_fp16, var_1336_cast_fp16))[name = string("op_1380_cast_fp16")]; + string var_1382_equation_0 = const()[name = string("op_1382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1382_cast_fp16 = einsum(equation = var_1382_equation_0, values = (var_1178_cast_fp16, var_1337_cast_fp16))[name = string("op_1382_cast_fp16")]; + string var_1384_equation_0 = const()[name = string("op_1384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1384_cast_fp16 = einsum(equation = var_1384_equation_0, values = (var_1178_cast_fp16, var_1338_cast_fp16))[name = string("op_1384_cast_fp16")]; + string var_1386_equation_0 = const()[name = string("op_1386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1386_cast_fp16 = einsum(equation = var_1386_equation_0, values = (var_1178_cast_fp16, var_1339_cast_fp16))[name = string("op_1386_cast_fp16")]; + string var_1388_equation_0 = const()[name = string("op_1388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1388_cast_fp16 = einsum(equation = var_1388_equation_0, values = (var_1178_cast_fp16, var_1340_cast_fp16))[name = string("op_1388_cast_fp16")]; + string var_1390_equation_0 = const()[name = string("op_1390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1390_cast_fp16 = einsum(equation = var_1390_equation_0, values = (var_1182_cast_fp16, var_1341_cast_fp16))[name = string("op_1390_cast_fp16")]; + string var_1392_equation_0 = const()[name = string("op_1392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1392_cast_fp16 = einsum(equation = var_1392_equation_0, values = (var_1182_cast_fp16, var_1342_cast_fp16))[name = string("op_1392_cast_fp16")]; + string var_1394_equation_0 = const()[name = string("op_1394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1394_cast_fp16 = einsum(equation = var_1394_equation_0, values = (var_1182_cast_fp16, var_1343_cast_fp16))[name = string("op_1394_cast_fp16")]; + string var_1396_equation_0 = const()[name = string("op_1396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1396_cast_fp16 = einsum(equation = var_1396_equation_0, values = (var_1182_cast_fp16, var_1344_cast_fp16))[name = string("op_1396_cast_fp16")]; + string var_1398_equation_0 = const()[name = string("op_1398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1398_cast_fp16 = einsum(equation = var_1398_equation_0, values = (var_1186_cast_fp16, var_1345_cast_fp16))[name = string("op_1398_cast_fp16")]; + string var_1400_equation_0 = const()[name = string("op_1400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1400_cast_fp16 = einsum(equation = var_1400_equation_0, values = (var_1186_cast_fp16, var_1346_cast_fp16))[name = string("op_1400_cast_fp16")]; + string var_1402_equation_0 = const()[name = string("op_1402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1402_cast_fp16 = einsum(equation = var_1402_equation_0, values = (var_1186_cast_fp16, var_1347_cast_fp16))[name = string("op_1402_cast_fp16")]; + string var_1404_equation_0 = const()[name = string("op_1404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1404_cast_fp16 = einsum(equation = var_1404_equation_0, values = (var_1186_cast_fp16, var_1348_cast_fp16))[name = string("op_1404_cast_fp16")]; + string var_1406_equation_0 = const()[name = string("op_1406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1406_cast_fp16 = einsum(equation = var_1406_equation_0, values = (var_1190_cast_fp16, var_1349_cast_fp16))[name = string("op_1406_cast_fp16")]; + string var_1408_equation_0 = const()[name = string("op_1408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1408_cast_fp16 = einsum(equation = var_1408_equation_0, values = (var_1190_cast_fp16, var_1350_cast_fp16))[name = string("op_1408_cast_fp16")]; + string var_1410_equation_0 = const()[name = string("op_1410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1410_cast_fp16 = einsum(equation = var_1410_equation_0, values = (var_1190_cast_fp16, var_1351_cast_fp16))[name = string("op_1410_cast_fp16")]; + string var_1412_equation_0 = const()[name = string("op_1412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1412_cast_fp16 = einsum(equation = var_1412_equation_0, values = (var_1190_cast_fp16, var_1352_cast_fp16))[name = string("op_1412_cast_fp16")]; + string var_1414_equation_0 = const()[name = string("op_1414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1414_cast_fp16 = einsum(equation = var_1414_equation_0, values = (var_1194_cast_fp16, var_1353_cast_fp16))[name = string("op_1414_cast_fp16")]; + string var_1416_equation_0 = const()[name = string("op_1416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1416_cast_fp16 = einsum(equation = var_1416_equation_0, values = (var_1194_cast_fp16, var_1354_cast_fp16))[name = string("op_1416_cast_fp16")]; + string var_1418_equation_0 = const()[name = string("op_1418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1418_cast_fp16 = einsum(equation = var_1418_equation_0, values = (var_1194_cast_fp16, var_1355_cast_fp16))[name = string("op_1418_cast_fp16")]; + string var_1420_equation_0 = const()[name = string("op_1420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1420_cast_fp16 = einsum(equation = var_1420_equation_0, values = (var_1194_cast_fp16, var_1356_cast_fp16))[name = string("op_1420_cast_fp16")]; + bool var_1422_interleave_0 = const()[name = string("op_1422_interleave_0"), val = bool(false)]; + tensor var_1422_cast_fp16 = concat(axis = var_809, interleave = var_1422_interleave_0, values = (var_1358_cast_fp16, var_1360_cast_fp16, var_1362_cast_fp16, var_1364_cast_fp16))[name = string("op_1422_cast_fp16")]; + bool var_1424_interleave_0 = const()[name = string("op_1424_interleave_0"), val = bool(false)]; + tensor var_1424_cast_fp16 = concat(axis = var_809, interleave = var_1424_interleave_0, values = (var_1366_cast_fp16, var_1368_cast_fp16, var_1370_cast_fp16, var_1372_cast_fp16))[name = string("op_1424_cast_fp16")]; + bool var_1426_interleave_0 = const()[name = string("op_1426_interleave_0"), val = bool(false)]; + tensor var_1426_cast_fp16 = concat(axis = var_809, interleave = var_1426_interleave_0, values = (var_1374_cast_fp16, var_1376_cast_fp16, var_1378_cast_fp16, var_1380_cast_fp16))[name = string("op_1426_cast_fp16")]; + bool var_1428_interleave_0 = const()[name = string("op_1428_interleave_0"), val = bool(false)]; + tensor var_1428_cast_fp16 = concat(axis = var_809, interleave = var_1428_interleave_0, values = (var_1382_cast_fp16, var_1384_cast_fp16, var_1386_cast_fp16, var_1388_cast_fp16))[name = string("op_1428_cast_fp16")]; + bool var_1430_interleave_0 = const()[name = string("op_1430_interleave_0"), val = bool(false)]; + tensor var_1430_cast_fp16 = concat(axis = var_809, interleave = var_1430_interleave_0, values = (var_1390_cast_fp16, var_1392_cast_fp16, var_1394_cast_fp16, var_1396_cast_fp16))[name = string("op_1430_cast_fp16")]; + bool var_1432_interleave_0 = const()[name = string("op_1432_interleave_0"), val = bool(false)]; + tensor var_1432_cast_fp16 = concat(axis = var_809, interleave = var_1432_interleave_0, values = (var_1398_cast_fp16, var_1400_cast_fp16, var_1402_cast_fp16, var_1404_cast_fp16))[name = string("op_1432_cast_fp16")]; + bool var_1434_interleave_0 = const()[name = string("op_1434_interleave_0"), val = bool(false)]; + tensor var_1434_cast_fp16 = concat(axis = var_809, interleave = var_1434_interleave_0, values = (var_1406_cast_fp16, var_1408_cast_fp16, var_1410_cast_fp16, var_1412_cast_fp16))[name = string("op_1434_cast_fp16")]; + bool var_1436_interleave_0 = const()[name = string("op_1436_interleave_0"), val = bool(false)]; + tensor var_1436_cast_fp16 = concat(axis = var_809, interleave = var_1436_interleave_0, values = (var_1414_cast_fp16, var_1416_cast_fp16, var_1418_cast_fp16, var_1420_cast_fp16))[name = string("op_1436_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_822, interleave = input_9_interleave_0, values = (var_1422_cast_fp16, var_1424_cast_fp16, var_1426_cast_fp16, var_1428_cast_fp16, var_1430_cast_fp16, var_1432_cast_fp16, var_1434_cast_fp16, var_1436_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11241344)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11765696)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_1455_to_fp16 = const()[name = string("op_1455_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_1455_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11766784)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11767872)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11768960)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13866176)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13870336)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15967552)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_1484 = const()[name = string("op_1484"), val = int32(3)]; + int32 var_1497 = const()[name = string("op_1497"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_1514_to_fp16 = const()[name = string("op_1514_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_1514_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15968640)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15969728)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15970816)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16495168)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16496256)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17020608)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17544960)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_1552_begin_0 = const()[name = string("op_1552_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1552_end_0 = const()[name = string("op_1552_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1552_end_mask_0 = const()[name = string("op_1552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1552_cast_fp16 = slice_by_index(begin = var_1552_begin_0, end = var_1552_end_0, end_mask = var_1552_end_mask_0, x = query_5_cast_fp16)[name = string("op_1552_cast_fp16")]; + tensor var_1556_begin_0 = const()[name = string("op_1556_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1556_end_0 = const()[name = string("op_1556_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1556_end_mask_0 = const()[name = string("op_1556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1556_cast_fp16 = slice_by_index(begin = var_1556_begin_0, end = var_1556_end_0, end_mask = var_1556_end_mask_0, x = query_5_cast_fp16)[name = string("op_1556_cast_fp16")]; + tensor var_1560_begin_0 = const()[name = string("op_1560_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1560_end_0 = const()[name = string("op_1560_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1560_end_mask_0 = const()[name = string("op_1560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1560_cast_fp16 = slice_by_index(begin = var_1560_begin_0, end = var_1560_end_0, end_mask = var_1560_end_mask_0, x = query_5_cast_fp16)[name = string("op_1560_cast_fp16")]; + tensor var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = query_5_cast_fp16)[name = string("op_1564_cast_fp16")]; + tensor var_1568_begin_0 = const()[name = string("op_1568_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1568_end_0 = const()[name = string("op_1568_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1568_end_mask_0 = const()[name = string("op_1568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1568_cast_fp16 = slice_by_index(begin = var_1568_begin_0, end = var_1568_end_0, end_mask = var_1568_end_mask_0, x = query_5_cast_fp16)[name = string("op_1568_cast_fp16")]; + tensor var_1572_begin_0 = const()[name = string("op_1572_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1572_end_0 = const()[name = string("op_1572_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1572_end_mask_0 = const()[name = string("op_1572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1572_cast_fp16 = slice_by_index(begin = var_1572_begin_0, end = var_1572_end_0, end_mask = var_1572_end_mask_0, x = query_5_cast_fp16)[name = string("op_1572_cast_fp16")]; + tensor var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = query_5_cast_fp16)[name = string("op_1576_cast_fp16")]; + tensor var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = query_5_cast_fp16)[name = string("op_1580_cast_fp16")]; + tensor var_1589_begin_0 = const()[name = string("op_1589_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1589_end_mask_0 = const()[name = string("op_1589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1589_cast_fp16")]; + tensor var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1603_begin_0 = const()[name = string("op_1603_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1603_end_0 = const()[name = string("op_1603_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1603_end_mask_0 = const()[name = string("op_1603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1603_cast_fp16 = slice_by_index(begin = var_1603_begin_0, end = var_1603_end_0, end_mask = var_1603_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1603_cast_fp16")]; + tensor var_1610_begin_0 = const()[name = string("op_1610_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1610_end_0 = const()[name = string("op_1610_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1610_end_mask_0 = const()[name = string("op_1610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1610_cast_fp16 = slice_by_index(begin = var_1610_begin_0, end = var_1610_end_0, end_mask = var_1610_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1610_cast_fp16")]; + tensor var_1617_begin_0 = const()[name = string("op_1617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1617_end_0 = const()[name = string("op_1617_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1617_end_mask_0 = const()[name = string("op_1617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1617_cast_fp16 = slice_by_index(begin = var_1617_begin_0, end = var_1617_end_0, end_mask = var_1617_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1617_cast_fp16")]; + tensor var_1624_begin_0 = const()[name = string("op_1624_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1624_end_0 = const()[name = string("op_1624_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1624_end_mask_0 = const()[name = string("op_1624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1624_cast_fp16 = slice_by_index(begin = var_1624_begin_0, end = var_1624_end_0, end_mask = var_1624_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1624_cast_fp16")]; + tensor var_1631_begin_0 = const()[name = string("op_1631_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1631_end_0 = const()[name = string("op_1631_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1631_end_mask_0 = const()[name = string("op_1631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1631_cast_fp16 = slice_by_index(begin = var_1631_begin_0, end = var_1631_end_0, end_mask = var_1631_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1631_cast_fp16")]; + tensor var_1638_begin_0 = const()[name = string("op_1638_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1638_end_0 = const()[name = string("op_1638_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1638_end_mask_0 = const()[name = string("op_1638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor var_1645_begin_0 = const()[name = string("op_1645_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1645_end_0 = const()[name = string("op_1645_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1645_end_mask_0 = const()[name = string("op_1645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1645_cast_fp16 = slice_by_index(begin = var_1645_begin_0, end = var_1645_end_0, end_mask = var_1645_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1645_cast_fp16")]; + tensor var_1652_begin_0 = const()[name = string("op_1652_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1652_end_0 = const()[name = string("op_1652_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1652_end_mask_0 = const()[name = string("op_1652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1652_cast_fp16 = slice_by_index(begin = var_1652_begin_0, end = var_1652_end_0, end_mask = var_1652_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1652_cast_fp16")]; + tensor var_1659_begin_0 = const()[name = string("op_1659_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1659_end_0 = const()[name = string("op_1659_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1659_end_mask_0 = const()[name = string("op_1659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1659_cast_fp16 = slice_by_index(begin = var_1659_begin_0, end = var_1659_end_0, end_mask = var_1659_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1659_cast_fp16")]; + tensor var_1666_begin_0 = const()[name = string("op_1666_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1666_end_0 = const()[name = string("op_1666_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1666_end_mask_0 = const()[name = string("op_1666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1666_cast_fp16 = slice_by_index(begin = var_1666_begin_0, end = var_1666_end_0, end_mask = var_1666_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1666_cast_fp16")]; + tensor var_1673_begin_0 = const()[name = string("op_1673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1673_end_0 = const()[name = string("op_1673_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1673_end_mask_0 = const()[name = string("op_1673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1673_cast_fp16 = slice_by_index(begin = var_1673_begin_0, end = var_1673_end_0, end_mask = var_1673_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1673_cast_fp16")]; + tensor var_1680_begin_0 = const()[name = string("op_1680_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1680_end_0 = const()[name = string("op_1680_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1680_end_mask_0 = const()[name = string("op_1680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1680_cast_fp16 = slice_by_index(begin = var_1680_begin_0, end = var_1680_end_0, end_mask = var_1680_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1680_cast_fp16")]; + tensor var_1687_begin_0 = const()[name = string("op_1687_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1687_end_0 = const()[name = string("op_1687_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1687_end_mask_0 = const()[name = string("op_1687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1687_cast_fp16 = slice_by_index(begin = var_1687_begin_0, end = var_1687_end_0, end_mask = var_1687_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1687_cast_fp16")]; + tensor var_1694_begin_0 = const()[name = string("op_1694_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1694_end_0 = const()[name = string("op_1694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1694_end_mask_0 = const()[name = string("op_1694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1694_cast_fp16 = slice_by_index(begin = var_1694_begin_0, end = var_1694_end_0, end_mask = var_1694_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1694_cast_fp16")]; + tensor var_1701_begin_0 = const()[name = string("op_1701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1701_end_0 = const()[name = string("op_1701_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1701_end_mask_0 = const()[name = string("op_1701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1701_cast_fp16 = slice_by_index(begin = var_1701_begin_0, end = var_1701_end_0, end_mask = var_1701_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1701_cast_fp16")]; + tensor var_1708_begin_0 = const()[name = string("op_1708_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1708_end_0 = const()[name = string("op_1708_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1708_end_mask_0 = const()[name = string("op_1708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1708_cast_fp16 = slice_by_index(begin = var_1708_begin_0, end = var_1708_end_0, end_mask = var_1708_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1708_cast_fp16")]; + tensor var_1715_begin_0 = const()[name = string("op_1715_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1715_end_0 = const()[name = string("op_1715_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1715_end_mask_0 = const()[name = string("op_1715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1715_cast_fp16 = slice_by_index(begin = var_1715_begin_0, end = var_1715_end_0, end_mask = var_1715_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1715_cast_fp16")]; + tensor var_1722_begin_0 = const()[name = string("op_1722_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1722_end_0 = const()[name = string("op_1722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1722_end_mask_0 = const()[name = string("op_1722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1722_cast_fp16 = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1722_cast_fp16")]; + tensor var_1729_begin_0 = const()[name = string("op_1729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1729_end_0 = const()[name = string("op_1729_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1729_end_mask_0 = const()[name = string("op_1729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1729_cast_fp16 = slice_by_index(begin = var_1729_begin_0, end = var_1729_end_0, end_mask = var_1729_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1729_cast_fp16")]; + tensor var_1736_begin_0 = const()[name = string("op_1736_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1736_end_0 = const()[name = string("op_1736_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1736_end_mask_0 = const()[name = string("op_1736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1736_cast_fp16 = slice_by_index(begin = var_1736_begin_0, end = var_1736_end_0, end_mask = var_1736_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1736_cast_fp16")]; + tensor var_1743_begin_0 = const()[name = string("op_1743_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1743_end_0 = const()[name = string("op_1743_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1743_end_mask_0 = const()[name = string("op_1743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1743_cast_fp16 = slice_by_index(begin = var_1743_begin_0, end = var_1743_end_0, end_mask = var_1743_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1743_cast_fp16")]; + tensor var_1750_begin_0 = const()[name = string("op_1750_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1750_end_0 = const()[name = string("op_1750_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1750_end_mask_0 = const()[name = string("op_1750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1750_cast_fp16 = slice_by_index(begin = var_1750_begin_0, end = var_1750_end_0, end_mask = var_1750_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1750_cast_fp16")]; + tensor var_1757_begin_0 = const()[name = string("op_1757_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1757_end_0 = const()[name = string("op_1757_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1757_end_mask_0 = const()[name = string("op_1757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1757_cast_fp16 = slice_by_index(begin = var_1757_begin_0, end = var_1757_end_0, end_mask = var_1757_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1757_cast_fp16")]; + tensor var_1764_begin_0 = const()[name = string("op_1764_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1764_end_0 = const()[name = string("op_1764_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1764_end_mask_0 = const()[name = string("op_1764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1764_cast_fp16 = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1764_cast_fp16")]; + tensor var_1771_begin_0 = const()[name = string("op_1771_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1771_end_0 = const()[name = string("op_1771_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1771_end_mask_0 = const()[name = string("op_1771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1771_cast_fp16 = slice_by_index(begin = var_1771_begin_0, end = var_1771_end_0, end_mask = var_1771_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1771_cast_fp16")]; + tensor var_1778_begin_0 = const()[name = string("op_1778_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1778_end_0 = const()[name = string("op_1778_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1778_end_mask_0 = const()[name = string("op_1778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1778_cast_fp16 = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1778_cast_fp16")]; + tensor var_1785_begin_0 = const()[name = string("op_1785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1785_end_0 = const()[name = string("op_1785_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1785_end_mask_0 = const()[name = string("op_1785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1785_cast_fp16 = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1785_cast_fp16")]; + tensor var_1792_begin_0 = const()[name = string("op_1792_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1792_end_0 = const()[name = string("op_1792_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1792_end_mask_0 = const()[name = string("op_1792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1792_cast_fp16 = slice_by_index(begin = var_1792_begin_0, end = var_1792_end_0, end_mask = var_1792_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1792_cast_fp16")]; + tensor var_1799_begin_0 = const()[name = string("op_1799_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1799_end_0 = const()[name = string("op_1799_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1799_end_mask_0 = const()[name = string("op_1799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1799_cast_fp16")]; + tensor var_1806_begin_0 = const()[name = string("op_1806_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1806_end_0 = const()[name = string("op_1806_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1806_end_mask_0 = const()[name = string("op_1806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1806_cast_fp16 = slice_by_index(begin = var_1806_begin_0, end = var_1806_end_0, end_mask = var_1806_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1806_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1811_begin_0 = const()[name = string("op_1811_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1811_end_0 = const()[name = string("op_1811_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1811_end_mask_0 = const()[name = string("op_1811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_3")]; + tensor var_1811_cast_fp16 = slice_by_index(begin = var_1811_begin_0, end = var_1811_end_0, end_mask = var_1811_end_mask_0, x = k_5_cast_fp16)[name = string("op_1811_cast_fp16")]; + tensor var_1815_begin_0 = const()[name = string("op_1815_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1815_end_0 = const()[name = string("op_1815_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1815_end_mask_0 = const()[name = string("op_1815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1815_cast_fp16 = slice_by_index(begin = var_1815_begin_0, end = var_1815_end_0, end_mask = var_1815_end_mask_0, x = k_5_cast_fp16)[name = string("op_1815_cast_fp16")]; + tensor var_1819_begin_0 = const()[name = string("op_1819_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1819_end_0 = const()[name = string("op_1819_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1819_end_mask_0 = const()[name = string("op_1819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1819_cast_fp16 = slice_by_index(begin = var_1819_begin_0, end = var_1819_end_0, end_mask = var_1819_end_mask_0, x = k_5_cast_fp16)[name = string("op_1819_cast_fp16")]; + tensor var_1823_begin_0 = const()[name = string("op_1823_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1823_end_0 = const()[name = string("op_1823_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1823_end_mask_0 = const()[name = string("op_1823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1823_cast_fp16 = slice_by_index(begin = var_1823_begin_0, end = var_1823_end_0, end_mask = var_1823_end_mask_0, x = k_5_cast_fp16)[name = string("op_1823_cast_fp16")]; + tensor var_1827_begin_0 = const()[name = string("op_1827_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1827_end_0 = const()[name = string("op_1827_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1827_end_mask_0 = const()[name = string("op_1827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1827_cast_fp16 = slice_by_index(begin = var_1827_begin_0, end = var_1827_end_0, end_mask = var_1827_end_mask_0, x = k_5_cast_fp16)[name = string("op_1827_cast_fp16")]; + tensor var_1831_begin_0 = const()[name = string("op_1831_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1831_end_0 = const()[name = string("op_1831_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1831_end_mask_0 = const()[name = string("op_1831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1831_cast_fp16 = slice_by_index(begin = var_1831_begin_0, end = var_1831_end_0, end_mask = var_1831_end_mask_0, x = k_5_cast_fp16)[name = string("op_1831_cast_fp16")]; + tensor var_1835_begin_0 = const()[name = string("op_1835_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1835_end_0 = const()[name = string("op_1835_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1835_end_mask_0 = const()[name = string("op_1835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1835_cast_fp16 = slice_by_index(begin = var_1835_begin_0, end = var_1835_end_0, end_mask = var_1835_end_mask_0, x = k_5_cast_fp16)[name = string("op_1835_cast_fp16")]; + tensor var_1839_begin_0 = const()[name = string("op_1839_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1839_end_0 = const()[name = string("op_1839_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1839_end_mask_0 = const()[name = string("op_1839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1839_cast_fp16 = slice_by_index(begin = var_1839_begin_0, end = var_1839_end_0, end_mask = var_1839_end_mask_0, x = k_5_cast_fp16)[name = string("op_1839_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = value_5_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1845_begin_0 = const()[name = string("op_1845_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1845_end_0 = const()[name = string("op_1845_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1845_end_mask_0 = const()[name = string("op_1845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1845_cast_fp16 = slice_by_index(begin = var_1845_begin_0, end = var_1845_end_0, end_mask = var_1845_end_mask_0, x = value_5_cast_fp16)[name = string("op_1845_cast_fp16")]; + tensor var_1849_begin_0 = const()[name = string("op_1849_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1849_end_0 = const()[name = string("op_1849_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1849_end_mask_0 = const()[name = string("op_1849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1849_cast_fp16 = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, x = value_5_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor var_1853_begin_0 = const()[name = string("op_1853_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1853_end_0 = const()[name = string("op_1853_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1853_end_mask_0 = const()[name = string("op_1853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1853_cast_fp16 = slice_by_index(begin = var_1853_begin_0, end = var_1853_end_0, end_mask = var_1853_end_mask_0, x = value_5_cast_fp16)[name = string("op_1853_cast_fp16")]; + tensor var_1857_begin_0 = const()[name = string("op_1857_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1857_end_0 = const()[name = string("op_1857_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1857_end_mask_0 = const()[name = string("op_1857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1857_cast_fp16 = slice_by_index(begin = var_1857_begin_0, end = var_1857_end_0, end_mask = var_1857_end_mask_0, x = value_5_cast_fp16)[name = string("op_1857_cast_fp16")]; + tensor var_1861_begin_0 = const()[name = string("op_1861_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1861_end_0 = const()[name = string("op_1861_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1861_end_mask_0 = const()[name = string("op_1861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = value_5_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1865_begin_0 = const()[name = string("op_1865_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1865_end_0 = const()[name = string("op_1865_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1865_end_mask_0 = const()[name = string("op_1865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1865_cast_fp16 = slice_by_index(begin = var_1865_begin_0, end = var_1865_end_0, end_mask = var_1865_end_mask_0, x = value_5_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1869_begin_0 = const()[name = string("op_1869_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1869_end_0 = const()[name = string("op_1869_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1869_end_mask_0 = const()[name = string("op_1869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1869_cast_fp16 = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = value_5_cast_fp16)[name = string("op_1869_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1811_cast_fp16, var_1589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1811_cast_fp16, var_1596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1811_cast_fp16, var_1603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1811_cast_fp16, var_1610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1815_cast_fp16, var_1617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1815_cast_fp16, var_1624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1815_cast_fp16, var_1631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1815_cast_fp16, var_1638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1819_cast_fp16, var_1645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1819_cast_fp16, var_1652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1819_cast_fp16, var_1659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1819_cast_fp16, var_1666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1823_cast_fp16, var_1673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1823_cast_fp16, var_1680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1823_cast_fp16, var_1687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1823_cast_fp16, var_1694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1827_cast_fp16, var_1701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1827_cast_fp16, var_1708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1827_cast_fp16, var_1715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1827_cast_fp16, var_1722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1831_cast_fp16, var_1729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1831_cast_fp16, var_1736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1831_cast_fp16, var_1743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1831_cast_fp16, var_1750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_1835_cast_fp16, var_1757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_1835_cast_fp16, var_1764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_1835_cast_fp16, var_1771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_1835_cast_fp16, var_1778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_1839_cast_fp16, var_1785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_1839_cast_fp16, var_1792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_1839_cast_fp16, var_1799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_1839_cast_fp16, var_1806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + fp16 var_1936_to_fp16 = const()[name = string("op_1936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1936_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1938_to_fp16 = const()[name = string("op_1938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1938_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1940_to_fp16 = const()[name = string("op_1940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1940_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1942_to_fp16 = const()[name = string("op_1942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1942_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1944_to_fp16 = const()[name = string("op_1944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1944_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1946_to_fp16 = const()[name = string("op_1946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1946_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1948_to_fp16 = const()[name = string("op_1948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1948_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1950_to_fp16 = const()[name = string("op_1950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1950_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1952_to_fp16 = const()[name = string("op_1952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1952_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1954_to_fp16 = const()[name = string("op_1954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1954_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1956_to_fp16 = const()[name = string("op_1956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1956_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1958_to_fp16 = const()[name = string("op_1958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1958_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1960_to_fp16 = const()[name = string("op_1960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1960_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1962_to_fp16 = const()[name = string("op_1962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1962_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1964_to_fp16 = const()[name = string("op_1964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1964_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1966_to_fp16 = const()[name = string("op_1966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1966_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_1968_to_fp16 = const()[name = string("op_1968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_1968_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_1970_to_fp16 = const()[name = string("op_1970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_1970_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_1972_to_fp16 = const()[name = string("op_1972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_1972_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_1974_to_fp16 = const()[name = string("op_1974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_1974_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_1976_to_fp16 = const()[name = string("op_1976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_1976_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_1978_to_fp16 = const()[name = string("op_1978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_1978_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_1980_to_fp16 = const()[name = string("op_1980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_1980_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_1982_to_fp16 = const()[name = string("op_1982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_1982_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_1984_to_fp16 = const()[name = string("op_1984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_1984_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_1986_to_fp16 = const()[name = string("op_1986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_1986_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_1988_to_fp16 = const()[name = string("op_1988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_1988_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_1990_to_fp16 = const()[name = string("op_1990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_1990_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_1992_to_fp16 = const()[name = string("op_1992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_1992_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_1994_to_fp16 = const()[name = string("op_1994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_1994_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_1996_to_fp16 = const()[name = string("op_1996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_1996_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_1998_to_fp16 = const()[name = string("op_1998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_1998_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + tensor var_2000_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_129_cast_fp16)[name = string("op_2000_cast_fp16")]; + tensor var_2001_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_131_cast_fp16)[name = string("op_2001_cast_fp16")]; + tensor var_2002_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_133_cast_fp16)[name = string("op_2002_cast_fp16")]; + tensor var_2003_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_135_cast_fp16)[name = string("op_2003_cast_fp16")]; + tensor var_2004_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_137_cast_fp16)[name = string("op_2004_cast_fp16")]; + tensor var_2005_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_139_cast_fp16)[name = string("op_2005_cast_fp16")]; + tensor var_2006_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_141_cast_fp16)[name = string("op_2006_cast_fp16")]; + tensor var_2007_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_143_cast_fp16)[name = string("op_2007_cast_fp16")]; + tensor var_2008_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_145_cast_fp16)[name = string("op_2008_cast_fp16")]; + tensor var_2009_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_147_cast_fp16)[name = string("op_2009_cast_fp16")]; + tensor var_2010_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_149_cast_fp16)[name = string("op_2010_cast_fp16")]; + tensor var_2011_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_151_cast_fp16)[name = string("op_2011_cast_fp16")]; + tensor var_2012_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_153_cast_fp16)[name = string("op_2012_cast_fp16")]; + tensor var_2013_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_155_cast_fp16)[name = string("op_2013_cast_fp16")]; + tensor var_2014_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_157_cast_fp16)[name = string("op_2014_cast_fp16")]; + tensor var_2015_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_159_cast_fp16)[name = string("op_2015_cast_fp16")]; + tensor var_2016_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_161_cast_fp16)[name = string("op_2016_cast_fp16")]; + tensor var_2017_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_163_cast_fp16)[name = string("op_2017_cast_fp16")]; + tensor var_2018_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_165_cast_fp16)[name = string("op_2018_cast_fp16")]; + tensor var_2019_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_167_cast_fp16)[name = string("op_2019_cast_fp16")]; + tensor var_2020_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_169_cast_fp16)[name = string("op_2020_cast_fp16")]; + tensor var_2021_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_171_cast_fp16)[name = string("op_2021_cast_fp16")]; + tensor var_2022_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_173_cast_fp16)[name = string("op_2022_cast_fp16")]; + tensor var_2023_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_175_cast_fp16)[name = string("op_2023_cast_fp16")]; + tensor var_2024_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_177_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2025_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_179_cast_fp16)[name = string("op_2025_cast_fp16")]; + tensor var_2026_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_181_cast_fp16)[name = string("op_2026_cast_fp16")]; + tensor var_2027_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_183_cast_fp16)[name = string("op_2027_cast_fp16")]; + tensor var_2028_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_185_cast_fp16)[name = string("op_2028_cast_fp16")]; + tensor var_2029_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_187_cast_fp16)[name = string("op_2029_cast_fp16")]; + tensor var_2030_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_189_cast_fp16)[name = string("op_2030_cast_fp16")]; + tensor var_2031_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_191_cast_fp16)[name = string("op_2031_cast_fp16")]; + string var_2033_equation_0 = const()[name = string("op_2033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2033_cast_fp16 = einsum(equation = var_2033_equation_0, values = (var_1841_cast_fp16, var_2000_cast_fp16))[name = string("op_2033_cast_fp16")]; + string var_2035_equation_0 = const()[name = string("op_2035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2035_cast_fp16 = einsum(equation = var_2035_equation_0, values = (var_1841_cast_fp16, var_2001_cast_fp16))[name = string("op_2035_cast_fp16")]; + string var_2037_equation_0 = const()[name = string("op_2037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2037_cast_fp16 = einsum(equation = var_2037_equation_0, values = (var_1841_cast_fp16, var_2002_cast_fp16))[name = string("op_2037_cast_fp16")]; + string var_2039_equation_0 = const()[name = string("op_2039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2039_cast_fp16 = einsum(equation = var_2039_equation_0, values = (var_1841_cast_fp16, var_2003_cast_fp16))[name = string("op_2039_cast_fp16")]; + string var_2041_equation_0 = const()[name = string("op_2041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2041_cast_fp16 = einsum(equation = var_2041_equation_0, values = (var_1845_cast_fp16, var_2004_cast_fp16))[name = string("op_2041_cast_fp16")]; + string var_2043_equation_0 = const()[name = string("op_2043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2043_cast_fp16 = einsum(equation = var_2043_equation_0, values = (var_1845_cast_fp16, var_2005_cast_fp16))[name = string("op_2043_cast_fp16")]; + string var_2045_equation_0 = const()[name = string("op_2045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2045_cast_fp16 = einsum(equation = var_2045_equation_0, values = (var_1845_cast_fp16, var_2006_cast_fp16))[name = string("op_2045_cast_fp16")]; + string var_2047_equation_0 = const()[name = string("op_2047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2047_cast_fp16 = einsum(equation = var_2047_equation_0, values = (var_1845_cast_fp16, var_2007_cast_fp16))[name = string("op_2047_cast_fp16")]; + string var_2049_equation_0 = const()[name = string("op_2049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2049_cast_fp16 = einsum(equation = var_2049_equation_0, values = (var_1849_cast_fp16, var_2008_cast_fp16))[name = string("op_2049_cast_fp16")]; + string var_2051_equation_0 = const()[name = string("op_2051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2051_cast_fp16 = einsum(equation = var_2051_equation_0, values = (var_1849_cast_fp16, var_2009_cast_fp16))[name = string("op_2051_cast_fp16")]; + string var_2053_equation_0 = const()[name = string("op_2053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2053_cast_fp16 = einsum(equation = var_2053_equation_0, values = (var_1849_cast_fp16, var_2010_cast_fp16))[name = string("op_2053_cast_fp16")]; + string var_2055_equation_0 = const()[name = string("op_2055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2055_cast_fp16 = einsum(equation = var_2055_equation_0, values = (var_1849_cast_fp16, var_2011_cast_fp16))[name = string("op_2055_cast_fp16")]; + string var_2057_equation_0 = const()[name = string("op_2057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2057_cast_fp16 = einsum(equation = var_2057_equation_0, values = (var_1853_cast_fp16, var_2012_cast_fp16))[name = string("op_2057_cast_fp16")]; + string var_2059_equation_0 = const()[name = string("op_2059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2059_cast_fp16 = einsum(equation = var_2059_equation_0, values = (var_1853_cast_fp16, var_2013_cast_fp16))[name = string("op_2059_cast_fp16")]; + string var_2061_equation_0 = const()[name = string("op_2061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2061_cast_fp16 = einsum(equation = var_2061_equation_0, values = (var_1853_cast_fp16, var_2014_cast_fp16))[name = string("op_2061_cast_fp16")]; + string var_2063_equation_0 = const()[name = string("op_2063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2063_cast_fp16 = einsum(equation = var_2063_equation_0, values = (var_1853_cast_fp16, var_2015_cast_fp16))[name = string("op_2063_cast_fp16")]; + string var_2065_equation_0 = const()[name = string("op_2065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2065_cast_fp16 = einsum(equation = var_2065_equation_0, values = (var_1857_cast_fp16, var_2016_cast_fp16))[name = string("op_2065_cast_fp16")]; + string var_2067_equation_0 = const()[name = string("op_2067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2067_cast_fp16 = einsum(equation = var_2067_equation_0, values = (var_1857_cast_fp16, var_2017_cast_fp16))[name = string("op_2067_cast_fp16")]; + string var_2069_equation_0 = const()[name = string("op_2069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2069_cast_fp16 = einsum(equation = var_2069_equation_0, values = (var_1857_cast_fp16, var_2018_cast_fp16))[name = string("op_2069_cast_fp16")]; + string var_2071_equation_0 = const()[name = string("op_2071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2071_cast_fp16 = einsum(equation = var_2071_equation_0, values = (var_1857_cast_fp16, var_2019_cast_fp16))[name = string("op_2071_cast_fp16")]; + string var_2073_equation_0 = const()[name = string("op_2073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2073_cast_fp16 = einsum(equation = var_2073_equation_0, values = (var_1861_cast_fp16, var_2020_cast_fp16))[name = string("op_2073_cast_fp16")]; + string var_2075_equation_0 = const()[name = string("op_2075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2075_cast_fp16 = einsum(equation = var_2075_equation_0, values = (var_1861_cast_fp16, var_2021_cast_fp16))[name = string("op_2075_cast_fp16")]; + string var_2077_equation_0 = const()[name = string("op_2077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2077_cast_fp16 = einsum(equation = var_2077_equation_0, values = (var_1861_cast_fp16, var_2022_cast_fp16))[name = string("op_2077_cast_fp16")]; + string var_2079_equation_0 = const()[name = string("op_2079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2079_cast_fp16 = einsum(equation = var_2079_equation_0, values = (var_1861_cast_fp16, var_2023_cast_fp16))[name = string("op_2079_cast_fp16")]; + string var_2081_equation_0 = const()[name = string("op_2081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2081_cast_fp16 = einsum(equation = var_2081_equation_0, values = (var_1865_cast_fp16, var_2024_cast_fp16))[name = string("op_2081_cast_fp16")]; + string var_2083_equation_0 = const()[name = string("op_2083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2083_cast_fp16 = einsum(equation = var_2083_equation_0, values = (var_1865_cast_fp16, var_2025_cast_fp16))[name = string("op_2083_cast_fp16")]; + string var_2085_equation_0 = const()[name = string("op_2085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2085_cast_fp16 = einsum(equation = var_2085_equation_0, values = (var_1865_cast_fp16, var_2026_cast_fp16))[name = string("op_2085_cast_fp16")]; + string var_2087_equation_0 = const()[name = string("op_2087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2087_cast_fp16 = einsum(equation = var_2087_equation_0, values = (var_1865_cast_fp16, var_2027_cast_fp16))[name = string("op_2087_cast_fp16")]; + string var_2089_equation_0 = const()[name = string("op_2089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2089_cast_fp16 = einsum(equation = var_2089_equation_0, values = (var_1869_cast_fp16, var_2028_cast_fp16))[name = string("op_2089_cast_fp16")]; + string var_2091_equation_0 = const()[name = string("op_2091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2091_cast_fp16 = einsum(equation = var_2091_equation_0, values = (var_1869_cast_fp16, var_2029_cast_fp16))[name = string("op_2091_cast_fp16")]; + string var_2093_equation_0 = const()[name = string("op_2093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2093_cast_fp16 = einsum(equation = var_2093_equation_0, values = (var_1869_cast_fp16, var_2030_cast_fp16))[name = string("op_2093_cast_fp16")]; + string var_2095_equation_0 = const()[name = string("op_2095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2095_cast_fp16 = einsum(equation = var_2095_equation_0, values = (var_1869_cast_fp16, var_2031_cast_fp16))[name = string("op_2095_cast_fp16")]; + bool var_2097_interleave_0 = const()[name = string("op_2097_interleave_0"), val = bool(false)]; + tensor var_2097_cast_fp16 = concat(axis = var_1484, interleave = var_2097_interleave_0, values = (var_2033_cast_fp16, var_2035_cast_fp16, var_2037_cast_fp16, var_2039_cast_fp16))[name = string("op_2097_cast_fp16")]; + bool var_2099_interleave_0 = const()[name = string("op_2099_interleave_0"), val = bool(false)]; + tensor var_2099_cast_fp16 = concat(axis = var_1484, interleave = var_2099_interleave_0, values = (var_2041_cast_fp16, var_2043_cast_fp16, var_2045_cast_fp16, var_2047_cast_fp16))[name = string("op_2099_cast_fp16")]; + bool var_2101_interleave_0 = const()[name = string("op_2101_interleave_0"), val = bool(false)]; + tensor var_2101_cast_fp16 = concat(axis = var_1484, interleave = var_2101_interleave_0, values = (var_2049_cast_fp16, var_2051_cast_fp16, var_2053_cast_fp16, var_2055_cast_fp16))[name = string("op_2101_cast_fp16")]; + bool var_2103_interleave_0 = const()[name = string("op_2103_interleave_0"), val = bool(false)]; + tensor var_2103_cast_fp16 = concat(axis = var_1484, interleave = var_2103_interleave_0, values = (var_2057_cast_fp16, var_2059_cast_fp16, var_2061_cast_fp16, var_2063_cast_fp16))[name = string("op_2103_cast_fp16")]; + bool var_2105_interleave_0 = const()[name = string("op_2105_interleave_0"), val = bool(false)]; + tensor var_2105_cast_fp16 = concat(axis = var_1484, interleave = var_2105_interleave_0, values = (var_2065_cast_fp16, var_2067_cast_fp16, var_2069_cast_fp16, var_2071_cast_fp16))[name = string("op_2105_cast_fp16")]; + bool var_2107_interleave_0 = const()[name = string("op_2107_interleave_0"), val = bool(false)]; + tensor var_2107_cast_fp16 = concat(axis = var_1484, interleave = var_2107_interleave_0, values = (var_2073_cast_fp16, var_2075_cast_fp16, var_2077_cast_fp16, var_2079_cast_fp16))[name = string("op_2107_cast_fp16")]; + bool var_2109_interleave_0 = const()[name = string("op_2109_interleave_0"), val = bool(false)]; + tensor var_2109_cast_fp16 = concat(axis = var_1484, interleave = var_2109_interleave_0, values = (var_2081_cast_fp16, var_2083_cast_fp16, var_2085_cast_fp16, var_2087_cast_fp16))[name = string("op_2109_cast_fp16")]; + bool var_2111_interleave_0 = const()[name = string("op_2111_interleave_0"), val = bool(false)]; + tensor var_2111_cast_fp16 = concat(axis = var_1484, interleave = var_2111_interleave_0, values = (var_2089_cast_fp16, var_2091_cast_fp16, var_2093_cast_fp16, var_2095_cast_fp16))[name = string("op_2111_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_1497, interleave = input_17_interleave_0, values = (var_2097_cast_fp16, var_2099_cast_fp16, var_2101_cast_fp16, var_2103_cast_fp16, var_2105_cast_fp16, var_2107_cast_fp16, var_2109_cast_fp16, var_2111_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17546048)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18070400)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_2130_to_fp16 = const()[name = string("op_2130_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_2130_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18071488)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18072576)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18073664)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20170880)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20175040)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22272256)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_2159 = const()[name = string("op_2159"), val = int32(3)]; + int32 var_2172 = const()[name = string("op_2172"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_2189_to_fp16 = const()[name = string("op_2189_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_2189_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22273344)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22274432)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22275520)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22799872)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22800960)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23325312)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23849664)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_2227_begin_0 = const()[name = string("op_2227_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2227_end_0 = const()[name = string("op_2227_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2227_end_mask_0 = const()[name = string("op_2227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2227_cast_fp16 = slice_by_index(begin = var_2227_begin_0, end = var_2227_end_0, end_mask = var_2227_end_mask_0, x = query_7_cast_fp16)[name = string("op_2227_cast_fp16")]; + tensor var_2231_begin_0 = const()[name = string("op_2231_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2231_end_0 = const()[name = string("op_2231_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2231_end_mask_0 = const()[name = string("op_2231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, x = query_7_cast_fp16)[name = string("op_2231_cast_fp16")]; + tensor var_2235_begin_0 = const()[name = string("op_2235_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2235_end_0 = const()[name = string("op_2235_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2235_end_mask_0 = const()[name = string("op_2235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2235_cast_fp16 = slice_by_index(begin = var_2235_begin_0, end = var_2235_end_0, end_mask = var_2235_end_mask_0, x = query_7_cast_fp16)[name = string("op_2235_cast_fp16")]; + tensor var_2239_begin_0 = const()[name = string("op_2239_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2239_end_0 = const()[name = string("op_2239_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2239_end_mask_0 = const()[name = string("op_2239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2239_cast_fp16 = slice_by_index(begin = var_2239_begin_0, end = var_2239_end_0, end_mask = var_2239_end_mask_0, x = query_7_cast_fp16)[name = string("op_2239_cast_fp16")]; + tensor var_2243_begin_0 = const()[name = string("op_2243_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2243_end_0 = const()[name = string("op_2243_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2243_end_mask_0 = const()[name = string("op_2243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = query_7_cast_fp16)[name = string("op_2243_cast_fp16")]; + tensor var_2247_begin_0 = const()[name = string("op_2247_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2247_end_0 = const()[name = string("op_2247_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2247_end_mask_0 = const()[name = string("op_2247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2247_cast_fp16 = slice_by_index(begin = var_2247_begin_0, end = var_2247_end_0, end_mask = var_2247_end_mask_0, x = query_7_cast_fp16)[name = string("op_2247_cast_fp16")]; + tensor var_2251_begin_0 = const()[name = string("op_2251_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2251_end_0 = const()[name = string("op_2251_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2251_end_mask_0 = const()[name = string("op_2251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2251_cast_fp16 = slice_by_index(begin = var_2251_begin_0, end = var_2251_end_0, end_mask = var_2251_end_mask_0, x = query_7_cast_fp16)[name = string("op_2251_cast_fp16")]; + tensor var_2255_begin_0 = const()[name = string("op_2255_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2255_end_0 = const()[name = string("op_2255_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2255_end_mask_0 = const()[name = string("op_2255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2255_cast_fp16 = slice_by_index(begin = var_2255_begin_0, end = var_2255_end_0, end_mask = var_2255_end_mask_0, x = query_7_cast_fp16)[name = string("op_2255_cast_fp16")]; + tensor var_2264_begin_0 = const()[name = string("op_2264_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2264_end_0 = const()[name = string("op_2264_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2264_end_mask_0 = const()[name = string("op_2264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2264_cast_fp16 = slice_by_index(begin = var_2264_begin_0, end = var_2264_end_0, end_mask = var_2264_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2264_cast_fp16")]; + tensor var_2271_begin_0 = const()[name = string("op_2271_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2271_end_0 = const()[name = string("op_2271_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2271_end_mask_0 = const()[name = string("op_2271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2271_cast_fp16 = slice_by_index(begin = var_2271_begin_0, end = var_2271_end_0, end_mask = var_2271_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2271_cast_fp16")]; + tensor var_2278_begin_0 = const()[name = string("op_2278_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2278_end_0 = const()[name = string("op_2278_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2278_end_mask_0 = const()[name = string("op_2278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2278_cast_fp16 = slice_by_index(begin = var_2278_begin_0, end = var_2278_end_0, end_mask = var_2278_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2278_cast_fp16")]; + tensor var_2285_begin_0 = const()[name = string("op_2285_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2285_end_0 = const()[name = string("op_2285_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2285_end_mask_0 = const()[name = string("op_2285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2285_cast_fp16 = slice_by_index(begin = var_2285_begin_0, end = var_2285_end_0, end_mask = var_2285_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2285_cast_fp16")]; + tensor var_2292_begin_0 = const()[name = string("op_2292_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2292_end_0 = const()[name = string("op_2292_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2292_end_mask_0 = const()[name = string("op_2292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2292_cast_fp16 = slice_by_index(begin = var_2292_begin_0, end = var_2292_end_0, end_mask = var_2292_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2292_cast_fp16")]; + tensor var_2299_begin_0 = const()[name = string("op_2299_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2299_end_0 = const()[name = string("op_2299_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2299_end_mask_0 = const()[name = string("op_2299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2299_cast_fp16 = slice_by_index(begin = var_2299_begin_0, end = var_2299_end_0, end_mask = var_2299_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2299_cast_fp16")]; + tensor var_2306_begin_0 = const()[name = string("op_2306_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2306_end_0 = const()[name = string("op_2306_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2306_end_mask_0 = const()[name = string("op_2306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2306_cast_fp16 = slice_by_index(begin = var_2306_begin_0, end = var_2306_end_0, end_mask = var_2306_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2306_cast_fp16")]; + tensor var_2313_begin_0 = const()[name = string("op_2313_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2313_end_0 = const()[name = string("op_2313_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2313_end_mask_0 = const()[name = string("op_2313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2313_cast_fp16 = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2313_cast_fp16")]; + tensor var_2320_begin_0 = const()[name = string("op_2320_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2320_end_0 = const()[name = string("op_2320_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2320_end_mask_0 = const()[name = string("op_2320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2320_cast_fp16 = slice_by_index(begin = var_2320_begin_0, end = var_2320_end_0, end_mask = var_2320_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2320_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = string("op_2327_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2327_end_0 = const()[name = string("op_2327_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2327_end_mask_0 = const()[name = string("op_2327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2327_cast_fp16")]; + tensor var_2334_begin_0 = const()[name = string("op_2334_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2334_end_0 = const()[name = string("op_2334_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2334_end_mask_0 = const()[name = string("op_2334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2334_cast_fp16 = slice_by_index(begin = var_2334_begin_0, end = var_2334_end_0, end_mask = var_2334_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2334_cast_fp16")]; + tensor var_2341_begin_0 = const()[name = string("op_2341_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2341_end_0 = const()[name = string("op_2341_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2341_end_mask_0 = const()[name = string("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2341_cast_fp16 = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2341_cast_fp16")]; + tensor var_2348_begin_0 = const()[name = string("op_2348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2348_end_0 = const()[name = string("op_2348_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2348_end_mask_0 = const()[name = string("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2348_cast_fp16 = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2348_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = string("op_2355_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2355_end_0 = const()[name = string("op_2355_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2355_end_mask_0 = const()[name = string("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2355_cast_fp16")]; + tensor var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2362_cast_fp16")]; + tensor var_2369_begin_0 = const()[name = string("op_2369_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2369_end_0 = const()[name = string("op_2369_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2369_end_mask_0 = const()[name = string("op_2369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2369_cast_fp16 = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2369_cast_fp16")]; + tensor var_2376_begin_0 = const()[name = string("op_2376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2376_end_0 = const()[name = string("op_2376_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2376_end_mask_0 = const()[name = string("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2376_cast_fp16 = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2376_cast_fp16")]; + tensor var_2383_begin_0 = const()[name = string("op_2383_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2383_end_0 = const()[name = string("op_2383_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2383_end_mask_0 = const()[name = string("op_2383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2383_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = string("op_2390_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2390_end_0 = const()[name = string("op_2390_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2390_end_mask_0 = const()[name = string("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2390_cast_fp16")]; + tensor var_2397_begin_0 = const()[name = string("op_2397_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2397_end_0 = const()[name = string("op_2397_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2397_end_mask_0 = const()[name = string("op_2397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2397_cast_fp16 = slice_by_index(begin = var_2397_begin_0, end = var_2397_end_0, end_mask = var_2397_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2397_cast_fp16")]; + tensor var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2404_cast_fp16")]; + tensor var_2411_begin_0 = const()[name = string("op_2411_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2411_end_0 = const()[name = string("op_2411_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2411_end_mask_0 = const()[name = string("op_2411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2411_cast_fp16 = slice_by_index(begin = var_2411_begin_0, end = var_2411_end_0, end_mask = var_2411_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2411_cast_fp16")]; + tensor var_2418_begin_0 = const()[name = string("op_2418_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2418_end_0 = const()[name = string("op_2418_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2418_end_mask_0 = const()[name = string("op_2418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2418_cast_fp16")]; + tensor var_2425_begin_0 = const()[name = string("op_2425_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2425_end_0 = const()[name = string("op_2425_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2425_end_mask_0 = const()[name = string("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2425_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2439_begin_0 = const()[name = string("op_2439_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2439_end_0 = const()[name = string("op_2439_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2439_end_mask_0 = const()[name = string("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2439_cast_fp16")]; + tensor var_2446_begin_0 = const()[name = string("op_2446_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2446_end_0 = const()[name = string("op_2446_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2446_end_mask_0 = const()[name = string("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2446_cast_fp16")]; + tensor var_2453_begin_0 = const()[name = string("op_2453_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2453_end_0 = const()[name = string("op_2453_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2453_end_mask_0 = const()[name = string("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2453_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2467_begin_0 = const()[name = string("op_2467_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2467_end_0 = const()[name = string("op_2467_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2467_end_mask_0 = const()[name = string("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2467_cast_fp16")]; + tensor var_2474_begin_0 = const()[name = string("op_2474_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2474_end_0 = const()[name = string("op_2474_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2474_end_mask_0 = const()[name = string("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2474_cast_fp16")]; + tensor var_2481_begin_0 = const()[name = string("op_2481_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2481_end_0 = const()[name = string("op_2481_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2481_end_mask_0 = const()[name = string("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2481_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2486_begin_0 = const()[name = string("op_2486_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2486_end_0 = const()[name = string("op_2486_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2486_end_mask_0 = const()[name = string("op_2486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_2")]; + tensor var_2486_cast_fp16 = slice_by_index(begin = var_2486_begin_0, end = var_2486_end_0, end_mask = var_2486_end_mask_0, x = k_7_cast_fp16)[name = string("op_2486_cast_fp16")]; + tensor var_2490_begin_0 = const()[name = string("op_2490_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2490_end_0 = const()[name = string("op_2490_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2490_end_mask_0 = const()[name = string("op_2490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2490_cast_fp16 = slice_by_index(begin = var_2490_begin_0, end = var_2490_end_0, end_mask = var_2490_end_mask_0, x = k_7_cast_fp16)[name = string("op_2490_cast_fp16")]; + tensor var_2494_begin_0 = const()[name = string("op_2494_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2494_end_0 = const()[name = string("op_2494_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2494_end_mask_0 = const()[name = string("op_2494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2494_cast_fp16 = slice_by_index(begin = var_2494_begin_0, end = var_2494_end_0, end_mask = var_2494_end_mask_0, x = k_7_cast_fp16)[name = string("op_2494_cast_fp16")]; + tensor var_2498_begin_0 = const()[name = string("op_2498_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2498_end_0 = const()[name = string("op_2498_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2498_end_mask_0 = const()[name = string("op_2498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2498_cast_fp16 = slice_by_index(begin = var_2498_begin_0, end = var_2498_end_0, end_mask = var_2498_end_mask_0, x = k_7_cast_fp16)[name = string("op_2498_cast_fp16")]; + tensor var_2502_begin_0 = const()[name = string("op_2502_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2502_end_0 = const()[name = string("op_2502_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2502_end_mask_0 = const()[name = string("op_2502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = k_7_cast_fp16)[name = string("op_2502_cast_fp16")]; + tensor var_2506_begin_0 = const()[name = string("op_2506_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2506_end_0 = const()[name = string("op_2506_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2506_end_mask_0 = const()[name = string("op_2506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2506_cast_fp16 = slice_by_index(begin = var_2506_begin_0, end = var_2506_end_0, end_mask = var_2506_end_mask_0, x = k_7_cast_fp16)[name = string("op_2506_cast_fp16")]; + tensor var_2510_begin_0 = const()[name = string("op_2510_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2510_end_0 = const()[name = string("op_2510_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2510_end_mask_0 = const()[name = string("op_2510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2510_cast_fp16 = slice_by_index(begin = var_2510_begin_0, end = var_2510_end_0, end_mask = var_2510_end_mask_0, x = k_7_cast_fp16)[name = string("op_2510_cast_fp16")]; + tensor var_2514_begin_0 = const()[name = string("op_2514_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2514_end_0 = const()[name = string("op_2514_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2514_end_mask_0 = const()[name = string("op_2514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2514_cast_fp16 = slice_by_index(begin = var_2514_begin_0, end = var_2514_end_0, end_mask = var_2514_end_mask_0, x = k_7_cast_fp16)[name = string("op_2514_cast_fp16")]; + tensor var_2516_begin_0 = const()[name = string("op_2516_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2516_end_0 = const()[name = string("op_2516_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2516_end_mask_0 = const()[name = string("op_2516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2516_cast_fp16 = slice_by_index(begin = var_2516_begin_0, end = var_2516_end_0, end_mask = var_2516_end_mask_0, x = value_7_cast_fp16)[name = string("op_2516_cast_fp16")]; + tensor var_2520_begin_0 = const()[name = string("op_2520_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2520_end_0 = const()[name = string("op_2520_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2520_end_mask_0 = const()[name = string("op_2520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2520_cast_fp16 = slice_by_index(begin = var_2520_begin_0, end = var_2520_end_0, end_mask = var_2520_end_mask_0, x = value_7_cast_fp16)[name = string("op_2520_cast_fp16")]; + tensor var_2524_begin_0 = const()[name = string("op_2524_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2524_end_0 = const()[name = string("op_2524_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2524_end_mask_0 = const()[name = string("op_2524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2524_cast_fp16 = slice_by_index(begin = var_2524_begin_0, end = var_2524_end_0, end_mask = var_2524_end_mask_0, x = value_7_cast_fp16)[name = string("op_2524_cast_fp16")]; + tensor var_2528_begin_0 = const()[name = string("op_2528_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2528_end_0 = const()[name = string("op_2528_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2528_end_mask_0 = const()[name = string("op_2528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2528_cast_fp16 = slice_by_index(begin = var_2528_begin_0, end = var_2528_end_0, end_mask = var_2528_end_mask_0, x = value_7_cast_fp16)[name = string("op_2528_cast_fp16")]; + tensor var_2532_begin_0 = const()[name = string("op_2532_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2532_end_0 = const()[name = string("op_2532_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2532_end_mask_0 = const()[name = string("op_2532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2532_cast_fp16 = slice_by_index(begin = var_2532_begin_0, end = var_2532_end_0, end_mask = var_2532_end_mask_0, x = value_7_cast_fp16)[name = string("op_2532_cast_fp16")]; + tensor var_2536_begin_0 = const()[name = string("op_2536_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2536_end_0 = const()[name = string("op_2536_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2536_end_mask_0 = const()[name = string("op_2536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2536_cast_fp16 = slice_by_index(begin = var_2536_begin_0, end = var_2536_end_0, end_mask = var_2536_end_mask_0, x = value_7_cast_fp16)[name = string("op_2536_cast_fp16")]; + tensor var_2540_begin_0 = const()[name = string("op_2540_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2540_end_0 = const()[name = string("op_2540_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2540_end_mask_0 = const()[name = string("op_2540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2540_cast_fp16 = slice_by_index(begin = var_2540_begin_0, end = var_2540_end_0, end_mask = var_2540_end_mask_0, x = value_7_cast_fp16)[name = string("op_2540_cast_fp16")]; + tensor var_2544_begin_0 = const()[name = string("op_2544_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2544_end_0 = const()[name = string("op_2544_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2544_end_mask_0 = const()[name = string("op_2544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2544_cast_fp16 = slice_by_index(begin = var_2544_begin_0, end = var_2544_end_0, end_mask = var_2544_end_mask_0, x = value_7_cast_fp16)[name = string("op_2544_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2486_cast_fp16, var_2264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2486_cast_fp16, var_2271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2486_cast_fp16, var_2278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2486_cast_fp16, var_2285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2490_cast_fp16, var_2292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2490_cast_fp16, var_2299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2490_cast_fp16, var_2306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2490_cast_fp16, var_2313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2494_cast_fp16, var_2320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2494_cast_fp16, var_2327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2494_cast_fp16, var_2334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2494_cast_fp16, var_2341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2498_cast_fp16, var_2348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2498_cast_fp16, var_2355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2498_cast_fp16, var_2362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2498_cast_fp16, var_2369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2502_cast_fp16, var_2376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2502_cast_fp16, var_2383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2502_cast_fp16, var_2390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2502_cast_fp16, var_2397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2506_cast_fp16, var_2404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2506_cast_fp16, var_2411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2506_cast_fp16, var_2418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2506_cast_fp16, var_2425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2510_cast_fp16, var_2432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2510_cast_fp16, var_2439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2510_cast_fp16, var_2446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2510_cast_fp16, var_2453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2514_cast_fp16, var_2460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2514_cast_fp16, var_2467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2514_cast_fp16, var_2474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2514_cast_fp16, var_2481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + fp16 var_2611_to_fp16 = const()[name = string("op_2611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2611_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2613_to_fp16 = const()[name = string("op_2613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2613_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2615_to_fp16 = const()[name = string("op_2615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2615_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2617_to_fp16 = const()[name = string("op_2617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2617_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2619_to_fp16 = const()[name = string("op_2619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2619_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2621_to_fp16 = const()[name = string("op_2621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2621_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2623_to_fp16 = const()[name = string("op_2623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2623_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2625_to_fp16 = const()[name = string("op_2625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2625_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2627_to_fp16 = const()[name = string("op_2627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2627_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2629_to_fp16 = const()[name = string("op_2629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2629_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2631_to_fp16 = const()[name = string("op_2631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2631_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2633_to_fp16 = const()[name = string("op_2633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2633_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2635_to_fp16 = const()[name = string("op_2635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2635_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2637_to_fp16 = const()[name = string("op_2637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2637_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2639_to_fp16 = const()[name = string("op_2639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2639_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2641_to_fp16 = const()[name = string("op_2641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2641_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2643_to_fp16 = const()[name = string("op_2643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2643_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2645_to_fp16 = const()[name = string("op_2645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2645_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2647_to_fp16 = const()[name = string("op_2647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2647_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2649_to_fp16 = const()[name = string("op_2649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2649_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2651_to_fp16 = const()[name = string("op_2651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2651_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2653_to_fp16 = const()[name = string("op_2653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2653_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2655_to_fp16 = const()[name = string("op_2655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2655_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2657_to_fp16 = const()[name = string("op_2657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2657_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2659_to_fp16 = const()[name = string("op_2659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2659_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2661_to_fp16 = const()[name = string("op_2661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2661_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2663_to_fp16 = const()[name = string("op_2663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2663_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2665_to_fp16 = const()[name = string("op_2665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2665_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2667_to_fp16 = const()[name = string("op_2667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2667_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2669_to_fp16 = const()[name = string("op_2669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2669_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2671_to_fp16 = const()[name = string("op_2671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2671_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2673_to_fp16 = const()[name = string("op_2673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2673_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + tensor var_2675_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_193_cast_fp16)[name = string("op_2675_cast_fp16")]; + tensor var_2676_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_195_cast_fp16)[name = string("op_2676_cast_fp16")]; + tensor var_2677_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_197_cast_fp16)[name = string("op_2677_cast_fp16")]; + tensor var_2678_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_199_cast_fp16)[name = string("op_2678_cast_fp16")]; + tensor var_2679_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_201_cast_fp16)[name = string("op_2679_cast_fp16")]; + tensor var_2680_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_203_cast_fp16)[name = string("op_2680_cast_fp16")]; + tensor var_2681_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_205_cast_fp16)[name = string("op_2681_cast_fp16")]; + tensor var_2682_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_207_cast_fp16)[name = string("op_2682_cast_fp16")]; + tensor var_2683_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_209_cast_fp16)[name = string("op_2683_cast_fp16")]; + tensor var_2684_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_211_cast_fp16)[name = string("op_2684_cast_fp16")]; + tensor var_2685_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_213_cast_fp16)[name = string("op_2685_cast_fp16")]; + tensor var_2686_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_215_cast_fp16)[name = string("op_2686_cast_fp16")]; + tensor var_2687_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_217_cast_fp16)[name = string("op_2687_cast_fp16")]; + tensor var_2688_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_219_cast_fp16)[name = string("op_2688_cast_fp16")]; + tensor var_2689_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_221_cast_fp16)[name = string("op_2689_cast_fp16")]; + tensor var_2690_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_223_cast_fp16)[name = string("op_2690_cast_fp16")]; + tensor var_2691_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_225_cast_fp16)[name = string("op_2691_cast_fp16")]; + tensor var_2692_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_227_cast_fp16)[name = string("op_2692_cast_fp16")]; + tensor var_2693_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_229_cast_fp16)[name = string("op_2693_cast_fp16")]; + tensor var_2694_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_231_cast_fp16)[name = string("op_2694_cast_fp16")]; + tensor var_2695_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_233_cast_fp16)[name = string("op_2695_cast_fp16")]; + tensor var_2696_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_235_cast_fp16)[name = string("op_2696_cast_fp16")]; + tensor var_2697_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_237_cast_fp16)[name = string("op_2697_cast_fp16")]; + tensor var_2698_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_239_cast_fp16)[name = string("op_2698_cast_fp16")]; + tensor var_2699_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_241_cast_fp16)[name = string("op_2699_cast_fp16")]; + tensor var_2700_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_243_cast_fp16)[name = string("op_2700_cast_fp16")]; + tensor var_2701_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_245_cast_fp16)[name = string("op_2701_cast_fp16")]; + tensor var_2702_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_247_cast_fp16)[name = string("op_2702_cast_fp16")]; + tensor var_2703_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_249_cast_fp16)[name = string("op_2703_cast_fp16")]; + tensor var_2704_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_251_cast_fp16)[name = string("op_2704_cast_fp16")]; + tensor var_2705_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_253_cast_fp16)[name = string("op_2705_cast_fp16")]; + tensor var_2706_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_255_cast_fp16)[name = string("op_2706_cast_fp16")]; + string var_2708_equation_0 = const()[name = string("op_2708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2708_cast_fp16 = einsum(equation = var_2708_equation_0, values = (var_2516_cast_fp16, var_2675_cast_fp16))[name = string("op_2708_cast_fp16")]; + string var_2710_equation_0 = const()[name = string("op_2710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2710_cast_fp16 = einsum(equation = var_2710_equation_0, values = (var_2516_cast_fp16, var_2676_cast_fp16))[name = string("op_2710_cast_fp16")]; + string var_2712_equation_0 = const()[name = string("op_2712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2712_cast_fp16 = einsum(equation = var_2712_equation_0, values = (var_2516_cast_fp16, var_2677_cast_fp16))[name = string("op_2712_cast_fp16")]; + string var_2714_equation_0 = const()[name = string("op_2714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2714_cast_fp16 = einsum(equation = var_2714_equation_0, values = (var_2516_cast_fp16, var_2678_cast_fp16))[name = string("op_2714_cast_fp16")]; + string var_2716_equation_0 = const()[name = string("op_2716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2716_cast_fp16 = einsum(equation = var_2716_equation_0, values = (var_2520_cast_fp16, var_2679_cast_fp16))[name = string("op_2716_cast_fp16")]; + string var_2718_equation_0 = const()[name = string("op_2718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2718_cast_fp16 = einsum(equation = var_2718_equation_0, values = (var_2520_cast_fp16, var_2680_cast_fp16))[name = string("op_2718_cast_fp16")]; + string var_2720_equation_0 = const()[name = string("op_2720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2720_cast_fp16 = einsum(equation = var_2720_equation_0, values = (var_2520_cast_fp16, var_2681_cast_fp16))[name = string("op_2720_cast_fp16")]; + string var_2722_equation_0 = const()[name = string("op_2722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2722_cast_fp16 = einsum(equation = var_2722_equation_0, values = (var_2520_cast_fp16, var_2682_cast_fp16))[name = string("op_2722_cast_fp16")]; + string var_2724_equation_0 = const()[name = string("op_2724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2724_cast_fp16 = einsum(equation = var_2724_equation_0, values = (var_2524_cast_fp16, var_2683_cast_fp16))[name = string("op_2724_cast_fp16")]; + string var_2726_equation_0 = const()[name = string("op_2726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2726_cast_fp16 = einsum(equation = var_2726_equation_0, values = (var_2524_cast_fp16, var_2684_cast_fp16))[name = string("op_2726_cast_fp16")]; + string var_2728_equation_0 = const()[name = string("op_2728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2728_cast_fp16 = einsum(equation = var_2728_equation_0, values = (var_2524_cast_fp16, var_2685_cast_fp16))[name = string("op_2728_cast_fp16")]; + string var_2730_equation_0 = const()[name = string("op_2730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2730_cast_fp16 = einsum(equation = var_2730_equation_0, values = (var_2524_cast_fp16, var_2686_cast_fp16))[name = string("op_2730_cast_fp16")]; + string var_2732_equation_0 = const()[name = string("op_2732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2732_cast_fp16 = einsum(equation = var_2732_equation_0, values = (var_2528_cast_fp16, var_2687_cast_fp16))[name = string("op_2732_cast_fp16")]; + string var_2734_equation_0 = const()[name = string("op_2734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2734_cast_fp16 = einsum(equation = var_2734_equation_0, values = (var_2528_cast_fp16, var_2688_cast_fp16))[name = string("op_2734_cast_fp16")]; + string var_2736_equation_0 = const()[name = string("op_2736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2736_cast_fp16 = einsum(equation = var_2736_equation_0, values = (var_2528_cast_fp16, var_2689_cast_fp16))[name = string("op_2736_cast_fp16")]; + string var_2738_equation_0 = const()[name = string("op_2738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2738_cast_fp16 = einsum(equation = var_2738_equation_0, values = (var_2528_cast_fp16, var_2690_cast_fp16))[name = string("op_2738_cast_fp16")]; + string var_2740_equation_0 = const()[name = string("op_2740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2740_cast_fp16 = einsum(equation = var_2740_equation_0, values = (var_2532_cast_fp16, var_2691_cast_fp16))[name = string("op_2740_cast_fp16")]; + string var_2742_equation_0 = const()[name = string("op_2742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2742_cast_fp16 = einsum(equation = var_2742_equation_0, values = (var_2532_cast_fp16, var_2692_cast_fp16))[name = string("op_2742_cast_fp16")]; + string var_2744_equation_0 = const()[name = string("op_2744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2744_cast_fp16 = einsum(equation = var_2744_equation_0, values = (var_2532_cast_fp16, var_2693_cast_fp16))[name = string("op_2744_cast_fp16")]; + string var_2746_equation_0 = const()[name = string("op_2746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2746_cast_fp16 = einsum(equation = var_2746_equation_0, values = (var_2532_cast_fp16, var_2694_cast_fp16))[name = string("op_2746_cast_fp16")]; + string var_2748_equation_0 = const()[name = string("op_2748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2748_cast_fp16 = einsum(equation = var_2748_equation_0, values = (var_2536_cast_fp16, var_2695_cast_fp16))[name = string("op_2748_cast_fp16")]; + string var_2750_equation_0 = const()[name = string("op_2750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2750_cast_fp16 = einsum(equation = var_2750_equation_0, values = (var_2536_cast_fp16, var_2696_cast_fp16))[name = string("op_2750_cast_fp16")]; + string var_2752_equation_0 = const()[name = string("op_2752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2752_cast_fp16 = einsum(equation = var_2752_equation_0, values = (var_2536_cast_fp16, var_2697_cast_fp16))[name = string("op_2752_cast_fp16")]; + string var_2754_equation_0 = const()[name = string("op_2754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2754_cast_fp16 = einsum(equation = var_2754_equation_0, values = (var_2536_cast_fp16, var_2698_cast_fp16))[name = string("op_2754_cast_fp16")]; + string var_2756_equation_0 = const()[name = string("op_2756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2756_cast_fp16 = einsum(equation = var_2756_equation_0, values = (var_2540_cast_fp16, var_2699_cast_fp16))[name = string("op_2756_cast_fp16")]; + string var_2758_equation_0 = const()[name = string("op_2758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2758_cast_fp16 = einsum(equation = var_2758_equation_0, values = (var_2540_cast_fp16, var_2700_cast_fp16))[name = string("op_2758_cast_fp16")]; + string var_2760_equation_0 = const()[name = string("op_2760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2760_cast_fp16 = einsum(equation = var_2760_equation_0, values = (var_2540_cast_fp16, var_2701_cast_fp16))[name = string("op_2760_cast_fp16")]; + string var_2762_equation_0 = const()[name = string("op_2762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2762_cast_fp16 = einsum(equation = var_2762_equation_0, values = (var_2540_cast_fp16, var_2702_cast_fp16))[name = string("op_2762_cast_fp16")]; + string var_2764_equation_0 = const()[name = string("op_2764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2764_cast_fp16 = einsum(equation = var_2764_equation_0, values = (var_2544_cast_fp16, var_2703_cast_fp16))[name = string("op_2764_cast_fp16")]; + string var_2766_equation_0 = const()[name = string("op_2766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2766_cast_fp16 = einsum(equation = var_2766_equation_0, values = (var_2544_cast_fp16, var_2704_cast_fp16))[name = string("op_2766_cast_fp16")]; + string var_2768_equation_0 = const()[name = string("op_2768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2768_cast_fp16 = einsum(equation = var_2768_equation_0, values = (var_2544_cast_fp16, var_2705_cast_fp16))[name = string("op_2768_cast_fp16")]; + string var_2770_equation_0 = const()[name = string("op_2770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2770_cast_fp16 = einsum(equation = var_2770_equation_0, values = (var_2544_cast_fp16, var_2706_cast_fp16))[name = string("op_2770_cast_fp16")]; + bool var_2772_interleave_0 = const()[name = string("op_2772_interleave_0"), val = bool(false)]; + tensor var_2772_cast_fp16 = concat(axis = var_2159, interleave = var_2772_interleave_0, values = (var_2708_cast_fp16, var_2710_cast_fp16, var_2712_cast_fp16, var_2714_cast_fp16))[name = string("op_2772_cast_fp16")]; + bool var_2774_interleave_0 = const()[name = string("op_2774_interleave_0"), val = bool(false)]; + tensor var_2774_cast_fp16 = concat(axis = var_2159, interleave = var_2774_interleave_0, values = (var_2716_cast_fp16, var_2718_cast_fp16, var_2720_cast_fp16, var_2722_cast_fp16))[name = string("op_2774_cast_fp16")]; + bool var_2776_interleave_0 = const()[name = string("op_2776_interleave_0"), val = bool(false)]; + tensor var_2776_cast_fp16 = concat(axis = var_2159, interleave = var_2776_interleave_0, values = (var_2724_cast_fp16, var_2726_cast_fp16, var_2728_cast_fp16, var_2730_cast_fp16))[name = string("op_2776_cast_fp16")]; + bool var_2778_interleave_0 = const()[name = string("op_2778_interleave_0"), val = bool(false)]; + tensor var_2778_cast_fp16 = concat(axis = var_2159, interleave = var_2778_interleave_0, values = (var_2732_cast_fp16, var_2734_cast_fp16, var_2736_cast_fp16, var_2738_cast_fp16))[name = string("op_2778_cast_fp16")]; + bool var_2780_interleave_0 = const()[name = string("op_2780_interleave_0"), val = bool(false)]; + tensor var_2780_cast_fp16 = concat(axis = var_2159, interleave = var_2780_interleave_0, values = (var_2740_cast_fp16, var_2742_cast_fp16, var_2744_cast_fp16, var_2746_cast_fp16))[name = string("op_2780_cast_fp16")]; + bool var_2782_interleave_0 = const()[name = string("op_2782_interleave_0"), val = bool(false)]; + tensor var_2782_cast_fp16 = concat(axis = var_2159, interleave = var_2782_interleave_0, values = (var_2748_cast_fp16, var_2750_cast_fp16, var_2752_cast_fp16, var_2754_cast_fp16))[name = string("op_2782_cast_fp16")]; + bool var_2784_interleave_0 = const()[name = string("op_2784_interleave_0"), val = bool(false)]; + tensor var_2784_cast_fp16 = concat(axis = var_2159, interleave = var_2784_interleave_0, values = (var_2756_cast_fp16, var_2758_cast_fp16, var_2760_cast_fp16, var_2762_cast_fp16))[name = string("op_2784_cast_fp16")]; + bool var_2786_interleave_0 = const()[name = string("op_2786_interleave_0"), val = bool(false)]; + tensor var_2786_cast_fp16 = concat(axis = var_2159, interleave = var_2786_interleave_0, values = (var_2764_cast_fp16, var_2766_cast_fp16, var_2768_cast_fp16, var_2770_cast_fp16))[name = string("op_2786_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_2172, interleave = input_25_interleave_0, values = (var_2772_cast_fp16, var_2774_cast_fp16, var_2776_cast_fp16, var_2778_cast_fp16, var_2780_cast_fp16, var_2782_cast_fp16, var_2784_cast_fp16, var_2786_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23850752)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24375104)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_2805_to_fp16 = const()[name = string("op_2805_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_2805_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24376192)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24377280)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24378368)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26475584)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26479744)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28576960)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_2834 = const()[name = string("op_2834"), val = int32(3)]; + int32 var_2847 = const()[name = string("op_2847"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_2864_to_fp16 = const()[name = string("op_2864_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_2864_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28578048)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28579136)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28580224)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29104576)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29105664)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29630016)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30154368)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_2902_begin_0 = const()[name = string("op_2902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2902_end_0 = const()[name = string("op_2902_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2902_end_mask_0 = const()[name = string("op_2902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2902_cast_fp16 = slice_by_index(begin = var_2902_begin_0, end = var_2902_end_0, end_mask = var_2902_end_mask_0, x = query_9_cast_fp16)[name = string("op_2902_cast_fp16")]; + tensor var_2906_begin_0 = const()[name = string("op_2906_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2906_end_0 = const()[name = string("op_2906_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2906_end_mask_0 = const()[name = string("op_2906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2906_cast_fp16 = slice_by_index(begin = var_2906_begin_0, end = var_2906_end_0, end_mask = var_2906_end_mask_0, x = query_9_cast_fp16)[name = string("op_2906_cast_fp16")]; + tensor var_2910_begin_0 = const()[name = string("op_2910_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2910_end_0 = const()[name = string("op_2910_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2910_end_mask_0 = const()[name = string("op_2910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2910_cast_fp16 = slice_by_index(begin = var_2910_begin_0, end = var_2910_end_0, end_mask = var_2910_end_mask_0, x = query_9_cast_fp16)[name = string("op_2910_cast_fp16")]; + tensor var_2914_begin_0 = const()[name = string("op_2914_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2914_end_0 = const()[name = string("op_2914_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2914_end_mask_0 = const()[name = string("op_2914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2914_cast_fp16 = slice_by_index(begin = var_2914_begin_0, end = var_2914_end_0, end_mask = var_2914_end_mask_0, x = query_9_cast_fp16)[name = string("op_2914_cast_fp16")]; + tensor var_2918_begin_0 = const()[name = string("op_2918_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2918_end_0 = const()[name = string("op_2918_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2918_end_mask_0 = const()[name = string("op_2918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2918_cast_fp16 = slice_by_index(begin = var_2918_begin_0, end = var_2918_end_0, end_mask = var_2918_end_mask_0, x = query_9_cast_fp16)[name = string("op_2918_cast_fp16")]; + tensor var_2922_begin_0 = const()[name = string("op_2922_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2922_end_0 = const()[name = string("op_2922_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2922_end_mask_0 = const()[name = string("op_2922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2922_cast_fp16 = slice_by_index(begin = var_2922_begin_0, end = var_2922_end_0, end_mask = var_2922_end_mask_0, x = query_9_cast_fp16)[name = string("op_2922_cast_fp16")]; + tensor var_2926_begin_0 = const()[name = string("op_2926_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2926_end_0 = const()[name = string("op_2926_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2926_end_mask_0 = const()[name = string("op_2926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2926_cast_fp16 = slice_by_index(begin = var_2926_begin_0, end = var_2926_end_0, end_mask = var_2926_end_mask_0, x = query_9_cast_fp16)[name = string("op_2926_cast_fp16")]; + tensor var_2930_begin_0 = const()[name = string("op_2930_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2930_end_0 = const()[name = string("op_2930_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2930_end_mask_0 = const()[name = string("op_2930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2930_cast_fp16 = slice_by_index(begin = var_2930_begin_0, end = var_2930_end_0, end_mask = var_2930_end_mask_0, x = query_9_cast_fp16)[name = string("op_2930_cast_fp16")]; + tensor var_2939_begin_0 = const()[name = string("op_2939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2939_end_0 = const()[name = string("op_2939_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2939_end_mask_0 = const()[name = string("op_2939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2939_cast_fp16 = slice_by_index(begin = var_2939_begin_0, end = var_2939_end_0, end_mask = var_2939_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2939_cast_fp16")]; + tensor var_2946_begin_0 = const()[name = string("op_2946_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2946_end_0 = const()[name = string("op_2946_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2946_end_mask_0 = const()[name = string("op_2946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2946_cast_fp16 = slice_by_index(begin = var_2946_begin_0, end = var_2946_end_0, end_mask = var_2946_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2946_cast_fp16")]; + tensor var_2953_begin_0 = const()[name = string("op_2953_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2953_end_0 = const()[name = string("op_2953_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2953_end_mask_0 = const()[name = string("op_2953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2953_cast_fp16 = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2960_begin_0 = const()[name = string("op_2960_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2960_end_0 = const()[name = string("op_2960_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2960_end_mask_0 = const()[name = string("op_2960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2960_cast_fp16 = slice_by_index(begin = var_2960_begin_0, end = var_2960_end_0, end_mask = var_2960_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2960_cast_fp16")]; + tensor var_2967_begin_0 = const()[name = string("op_2967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2967_end_0 = const()[name = string("op_2967_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2967_end_mask_0 = const()[name = string("op_2967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2967_cast_fp16 = slice_by_index(begin = var_2967_begin_0, end = var_2967_end_0, end_mask = var_2967_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2967_cast_fp16")]; + tensor var_2974_begin_0 = const()[name = string("op_2974_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2974_end_0 = const()[name = string("op_2974_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2974_end_mask_0 = const()[name = string("op_2974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2974_cast_fp16 = slice_by_index(begin = var_2974_begin_0, end = var_2974_end_0, end_mask = var_2974_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2974_cast_fp16")]; + tensor var_2981_begin_0 = const()[name = string("op_2981_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2981_end_0 = const()[name = string("op_2981_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2981_end_mask_0 = const()[name = string("op_2981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2981_cast_fp16 = slice_by_index(begin = var_2981_begin_0, end = var_2981_end_0, end_mask = var_2981_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2981_cast_fp16")]; + tensor var_2988_begin_0 = const()[name = string("op_2988_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2988_end_0 = const()[name = string("op_2988_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2988_end_mask_0 = const()[name = string("op_2988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2988_cast_fp16 = slice_by_index(begin = var_2988_begin_0, end = var_2988_end_0, end_mask = var_2988_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2988_cast_fp16")]; + tensor var_2995_begin_0 = const()[name = string("op_2995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2995_end_0 = const()[name = string("op_2995_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2995_end_mask_0 = const()[name = string("op_2995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2995_cast_fp16 = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, x = var_2910_cast_fp16)[name = string("op_2995_cast_fp16")]; + tensor var_3002_begin_0 = const()[name = string("op_3002_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3002_end_0 = const()[name = string("op_3002_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3002_end_mask_0 = const()[name = string("op_3002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3002_cast_fp16 = slice_by_index(begin = var_3002_begin_0, end = var_3002_end_0, end_mask = var_3002_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3002_cast_fp16")]; + tensor var_3009_begin_0 = const()[name = string("op_3009_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3009_end_0 = const()[name = string("op_3009_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3009_end_mask_0 = const()[name = string("op_3009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3009_cast_fp16 = slice_by_index(begin = var_3009_begin_0, end = var_3009_end_0, end_mask = var_3009_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3009_cast_fp16")]; + tensor var_3016_begin_0 = const()[name = string("op_3016_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3016_end_0 = const()[name = string("op_3016_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3016_end_mask_0 = const()[name = string("op_3016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3016_cast_fp16 = slice_by_index(begin = var_3016_begin_0, end = var_3016_end_0, end_mask = var_3016_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3016_cast_fp16")]; + tensor var_3023_begin_0 = const()[name = string("op_3023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3023_end_0 = const()[name = string("op_3023_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3023_end_mask_0 = const()[name = string("op_3023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3023_cast_fp16 = slice_by_index(begin = var_3023_begin_0, end = var_3023_end_0, end_mask = var_3023_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3023_cast_fp16")]; + tensor var_3030_begin_0 = const()[name = string("op_3030_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3030_end_0 = const()[name = string("op_3030_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3030_end_mask_0 = const()[name = string("op_3030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3030_cast_fp16 = slice_by_index(begin = var_3030_begin_0, end = var_3030_end_0, end_mask = var_3030_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3030_cast_fp16")]; + tensor var_3037_begin_0 = const()[name = string("op_3037_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3037_end_0 = const()[name = string("op_3037_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3037_end_mask_0 = const()[name = string("op_3037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3037_cast_fp16 = slice_by_index(begin = var_3037_begin_0, end = var_3037_end_0, end_mask = var_3037_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3037_cast_fp16")]; + tensor var_3044_begin_0 = const()[name = string("op_3044_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3044_end_0 = const()[name = string("op_3044_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3044_end_mask_0 = const()[name = string("op_3044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3044_cast_fp16 = slice_by_index(begin = var_3044_begin_0, end = var_3044_end_0, end_mask = var_3044_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3044_cast_fp16")]; + tensor var_3051_begin_0 = const()[name = string("op_3051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3051_end_0 = const()[name = string("op_3051_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3051_end_mask_0 = const()[name = string("op_3051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3051_cast_fp16 = slice_by_index(begin = var_3051_begin_0, end = var_3051_end_0, end_mask = var_3051_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3051_cast_fp16")]; + tensor var_3058_begin_0 = const()[name = string("op_3058_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3058_end_0 = const()[name = string("op_3058_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3058_end_mask_0 = const()[name = string("op_3058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3058_cast_fp16 = slice_by_index(begin = var_3058_begin_0, end = var_3058_end_0, end_mask = var_3058_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3058_cast_fp16")]; + tensor var_3065_begin_0 = const()[name = string("op_3065_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3065_end_0 = const()[name = string("op_3065_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3065_end_mask_0 = const()[name = string("op_3065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3065_cast_fp16 = slice_by_index(begin = var_3065_begin_0, end = var_3065_end_0, end_mask = var_3065_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3065_cast_fp16")]; + tensor var_3072_begin_0 = const()[name = string("op_3072_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3072_end_0 = const()[name = string("op_3072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3072_end_mask_0 = const()[name = string("op_3072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3072_cast_fp16 = slice_by_index(begin = var_3072_begin_0, end = var_3072_end_0, end_mask = var_3072_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3072_cast_fp16")]; + tensor var_3079_begin_0 = const()[name = string("op_3079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3079_end_0 = const()[name = string("op_3079_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3079_end_mask_0 = const()[name = string("op_3079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3079_cast_fp16 = slice_by_index(begin = var_3079_begin_0, end = var_3079_end_0, end_mask = var_3079_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3079_cast_fp16")]; + tensor var_3086_begin_0 = const()[name = string("op_3086_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3086_end_0 = const()[name = string("op_3086_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3086_end_mask_0 = const()[name = string("op_3086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3086_cast_fp16 = slice_by_index(begin = var_3086_begin_0, end = var_3086_end_0, end_mask = var_3086_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3086_cast_fp16")]; + tensor var_3093_begin_0 = const()[name = string("op_3093_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3093_end_0 = const()[name = string("op_3093_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3093_end_mask_0 = const()[name = string("op_3093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3093_cast_fp16 = slice_by_index(begin = var_3093_begin_0, end = var_3093_end_0, end_mask = var_3093_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3093_cast_fp16")]; + tensor var_3100_begin_0 = const()[name = string("op_3100_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3100_end_0 = const()[name = string("op_3100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3100_end_mask_0 = const()[name = string("op_3100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3100_cast_fp16 = slice_by_index(begin = var_3100_begin_0, end = var_3100_end_0, end_mask = var_3100_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3100_cast_fp16")]; + tensor var_3107_begin_0 = const()[name = string("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3107_end_0 = const()[name = string("op_3107_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3107_end_mask_0 = const()[name = string("op_3107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3107_cast_fp16")]; + tensor var_3114_begin_0 = const()[name = string("op_3114_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3114_end_0 = const()[name = string("op_3114_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3114_end_mask_0 = const()[name = string("op_3114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3114_cast_fp16 = slice_by_index(begin = var_3114_begin_0, end = var_3114_end_0, end_mask = var_3114_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3114_cast_fp16")]; + tensor var_3121_begin_0 = const()[name = string("op_3121_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3121_end_0 = const()[name = string("op_3121_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3121_end_mask_0 = const()[name = string("op_3121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3121_cast_fp16 = slice_by_index(begin = var_3121_begin_0, end = var_3121_end_0, end_mask = var_3121_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3121_cast_fp16")]; + tensor var_3128_begin_0 = const()[name = string("op_3128_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3128_end_0 = const()[name = string("op_3128_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3128_end_mask_0 = const()[name = string("op_3128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3128_cast_fp16 = slice_by_index(begin = var_3128_begin_0, end = var_3128_end_0, end_mask = var_3128_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3128_cast_fp16")]; + tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3135_cast_fp16")]; + tensor var_3142_begin_0 = const()[name = string("op_3142_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3142_end_0 = const()[name = string("op_3142_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3142_end_mask_0 = const()[name = string("op_3142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3142_cast_fp16 = slice_by_index(begin = var_3142_begin_0, end = var_3142_end_0, end_mask = var_3142_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3142_cast_fp16")]; + tensor var_3149_begin_0 = const()[name = string("op_3149_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3149_end_0 = const()[name = string("op_3149_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3149_end_mask_0 = const()[name = string("op_3149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3149_cast_fp16 = slice_by_index(begin = var_3149_begin_0, end = var_3149_end_0, end_mask = var_3149_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3149_cast_fp16")]; + tensor var_3156_begin_0 = const()[name = string("op_3156_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3156_end_0 = const()[name = string("op_3156_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3156_end_mask_0 = const()[name = string("op_3156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3156_cast_fp16 = slice_by_index(begin = var_3156_begin_0, end = var_3156_end_0, end_mask = var_3156_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3156_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3161_begin_0 = const()[name = string("op_3161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3161_end_0 = const()[name = string("op_3161_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3161_end_mask_0 = const()[name = string("op_3161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_1")]; + tensor var_3161_cast_fp16 = slice_by_index(begin = var_3161_begin_0, end = var_3161_end_0, end_mask = var_3161_end_mask_0, x = k_9_cast_fp16)[name = string("op_3161_cast_fp16")]; + tensor var_3165_begin_0 = const()[name = string("op_3165_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3165_end_0 = const()[name = string("op_3165_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3165_end_mask_0 = const()[name = string("op_3165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3165_cast_fp16 = slice_by_index(begin = var_3165_begin_0, end = var_3165_end_0, end_mask = var_3165_end_mask_0, x = k_9_cast_fp16)[name = string("op_3165_cast_fp16")]; + tensor var_3169_begin_0 = const()[name = string("op_3169_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3169_end_0 = const()[name = string("op_3169_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3169_end_mask_0 = const()[name = string("op_3169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3169_cast_fp16 = slice_by_index(begin = var_3169_begin_0, end = var_3169_end_0, end_mask = var_3169_end_mask_0, x = k_9_cast_fp16)[name = string("op_3169_cast_fp16")]; + tensor var_3173_begin_0 = const()[name = string("op_3173_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3173_end_0 = const()[name = string("op_3173_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3173_end_mask_0 = const()[name = string("op_3173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3173_cast_fp16 = slice_by_index(begin = var_3173_begin_0, end = var_3173_end_0, end_mask = var_3173_end_mask_0, x = k_9_cast_fp16)[name = string("op_3173_cast_fp16")]; + tensor var_3177_begin_0 = const()[name = string("op_3177_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3177_end_0 = const()[name = string("op_3177_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3177_end_mask_0 = const()[name = string("op_3177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3177_cast_fp16 = slice_by_index(begin = var_3177_begin_0, end = var_3177_end_0, end_mask = var_3177_end_mask_0, x = k_9_cast_fp16)[name = string("op_3177_cast_fp16")]; + tensor var_3181_begin_0 = const()[name = string("op_3181_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3181_end_0 = const()[name = string("op_3181_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3181_end_mask_0 = const()[name = string("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3181_cast_fp16 = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = k_9_cast_fp16)[name = string("op_3181_cast_fp16")]; + tensor var_3185_begin_0 = const()[name = string("op_3185_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3185_end_0 = const()[name = string("op_3185_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3185_end_mask_0 = const()[name = string("op_3185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3185_cast_fp16 = slice_by_index(begin = var_3185_begin_0, end = var_3185_end_0, end_mask = var_3185_end_mask_0, x = k_9_cast_fp16)[name = string("op_3185_cast_fp16")]; + tensor var_3189_begin_0 = const()[name = string("op_3189_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3189_end_0 = const()[name = string("op_3189_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3189_end_mask_0 = const()[name = string("op_3189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3189_cast_fp16 = slice_by_index(begin = var_3189_begin_0, end = var_3189_end_0, end_mask = var_3189_end_mask_0, x = k_9_cast_fp16)[name = string("op_3189_cast_fp16")]; + tensor var_3191_begin_0 = const()[name = string("op_3191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3191_end_0 = const()[name = string("op_3191_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3191_end_mask_0 = const()[name = string("op_3191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3191_cast_fp16 = slice_by_index(begin = var_3191_begin_0, end = var_3191_end_0, end_mask = var_3191_end_mask_0, x = value_9_cast_fp16)[name = string("op_3191_cast_fp16")]; + tensor var_3195_begin_0 = const()[name = string("op_3195_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3195_end_0 = const()[name = string("op_3195_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3195_end_mask_0 = const()[name = string("op_3195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3195_cast_fp16 = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = value_9_cast_fp16)[name = string("op_3195_cast_fp16")]; + tensor var_3199_begin_0 = const()[name = string("op_3199_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3199_end_0 = const()[name = string("op_3199_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3199_end_mask_0 = const()[name = string("op_3199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3199_cast_fp16 = slice_by_index(begin = var_3199_begin_0, end = var_3199_end_0, end_mask = var_3199_end_mask_0, x = value_9_cast_fp16)[name = string("op_3199_cast_fp16")]; + tensor var_3203_begin_0 = const()[name = string("op_3203_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3203_end_0 = const()[name = string("op_3203_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3203_end_mask_0 = const()[name = string("op_3203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3203_cast_fp16 = slice_by_index(begin = var_3203_begin_0, end = var_3203_end_0, end_mask = var_3203_end_mask_0, x = value_9_cast_fp16)[name = string("op_3203_cast_fp16")]; + tensor var_3207_begin_0 = const()[name = string("op_3207_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3207_end_0 = const()[name = string("op_3207_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3207_end_mask_0 = const()[name = string("op_3207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3207_cast_fp16 = slice_by_index(begin = var_3207_begin_0, end = var_3207_end_0, end_mask = var_3207_end_mask_0, x = value_9_cast_fp16)[name = string("op_3207_cast_fp16")]; + tensor var_3211_begin_0 = const()[name = string("op_3211_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3211_end_0 = const()[name = string("op_3211_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3211_end_mask_0 = const()[name = string("op_3211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3211_cast_fp16 = slice_by_index(begin = var_3211_begin_0, end = var_3211_end_0, end_mask = var_3211_end_mask_0, x = value_9_cast_fp16)[name = string("op_3211_cast_fp16")]; + tensor var_3215_begin_0 = const()[name = string("op_3215_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3215_end_0 = const()[name = string("op_3215_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3215_end_mask_0 = const()[name = string("op_3215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3215_cast_fp16 = slice_by_index(begin = var_3215_begin_0, end = var_3215_end_0, end_mask = var_3215_end_mask_0, x = value_9_cast_fp16)[name = string("op_3215_cast_fp16")]; + tensor var_3219_begin_0 = const()[name = string("op_3219_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3219_end_mask_0 = const()[name = string("op_3219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = value_9_cast_fp16)[name = string("op_3219_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_3161_cast_fp16, var_2939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_3161_cast_fp16, var_2946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_3161_cast_fp16, var_2953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_3161_cast_fp16, var_2960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_3165_cast_fp16, var_2967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_3165_cast_fp16, var_2974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_3165_cast_fp16, var_2981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_3165_cast_fp16, var_2988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_3169_cast_fp16, var_2995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_3169_cast_fp16, var_3002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_3169_cast_fp16, var_3009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_3169_cast_fp16, var_3016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_3173_cast_fp16, var_3023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_3173_cast_fp16, var_3030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_3173_cast_fp16, var_3037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_3173_cast_fp16, var_3044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_3177_cast_fp16, var_3051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_3177_cast_fp16, var_3058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_3177_cast_fp16, var_3065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_3177_cast_fp16, var_3072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_3181_cast_fp16, var_3079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_3181_cast_fp16, var_3086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_3181_cast_fp16, var_3093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_3181_cast_fp16, var_3100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_3185_cast_fp16, var_3107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_3185_cast_fp16, var_3114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_3185_cast_fp16, var_3121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_3185_cast_fp16, var_3128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_3189_cast_fp16, var_3135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_3189_cast_fp16, var_3142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_3189_cast_fp16, var_3149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_3189_cast_fp16, var_3156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + fp16 var_3286_to_fp16 = const()[name = string("op_3286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_3286_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_3288_to_fp16 = const()[name = string("op_3288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_3288_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_3290_to_fp16 = const()[name = string("op_3290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_3290_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_3292_to_fp16 = const()[name = string("op_3292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_3292_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_3294_to_fp16 = const()[name = string("op_3294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_3294_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_3296_to_fp16 = const()[name = string("op_3296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_3296_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_3298_to_fp16 = const()[name = string("op_3298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_3298_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_3300_to_fp16 = const()[name = string("op_3300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_3300_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_3302_to_fp16 = const()[name = string("op_3302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_3302_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_3304_to_fp16 = const()[name = string("op_3304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_3304_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_3306_to_fp16 = const()[name = string("op_3306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_3306_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_3308_to_fp16 = const()[name = string("op_3308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_3308_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_3310_to_fp16 = const()[name = string("op_3310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_3310_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_3312_to_fp16 = const()[name = string("op_3312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_3312_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_3314_to_fp16 = const()[name = string("op_3314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_3314_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_3316_to_fp16 = const()[name = string("op_3316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_3316_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + fp16 var_3318_to_fp16 = const()[name = string("op_3318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_3318_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_3320_to_fp16 = const()[name = string("op_3320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_3320_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_3322_to_fp16 = const()[name = string("op_3322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_3322_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_3324_to_fp16 = const()[name = string("op_3324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_3324_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_3326_to_fp16 = const()[name = string("op_3326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_3326_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_3328_to_fp16 = const()[name = string("op_3328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_3328_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_3330_to_fp16 = const()[name = string("op_3330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_3330_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_3332_to_fp16 = const()[name = string("op_3332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_3332_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_3334_to_fp16 = const()[name = string("op_3334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_3334_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_3336_to_fp16 = const()[name = string("op_3336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_3336_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_3338_to_fp16 = const()[name = string("op_3338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_3338_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_3340_to_fp16 = const()[name = string("op_3340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_3340_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_3342_to_fp16 = const()[name = string("op_3342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_3342_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_3344_to_fp16 = const()[name = string("op_3344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_3344_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_3346_to_fp16 = const()[name = string("op_3346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_3346_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_3348_to_fp16 = const()[name = string("op_3348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_3348_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + tensor var_3350_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_257_cast_fp16)[name = string("op_3350_cast_fp16")]; + tensor var_3351_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_259_cast_fp16)[name = string("op_3351_cast_fp16")]; + tensor var_3352_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_261_cast_fp16)[name = string("op_3352_cast_fp16")]; + tensor var_3353_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_263_cast_fp16)[name = string("op_3353_cast_fp16")]; + tensor var_3354_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_265_cast_fp16)[name = string("op_3354_cast_fp16")]; + tensor var_3355_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_267_cast_fp16)[name = string("op_3355_cast_fp16")]; + tensor var_3356_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_269_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3357_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_271_cast_fp16)[name = string("op_3357_cast_fp16")]; + tensor var_3358_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_273_cast_fp16)[name = string("op_3358_cast_fp16")]; + tensor var_3359_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_275_cast_fp16)[name = string("op_3359_cast_fp16")]; + tensor var_3360_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_277_cast_fp16)[name = string("op_3360_cast_fp16")]; + tensor var_3361_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_279_cast_fp16)[name = string("op_3361_cast_fp16")]; + tensor var_3362_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_281_cast_fp16)[name = string("op_3362_cast_fp16")]; + tensor var_3363_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_283_cast_fp16)[name = string("op_3363_cast_fp16")]; + tensor var_3364_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_285_cast_fp16)[name = string("op_3364_cast_fp16")]; + tensor var_3365_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_287_cast_fp16)[name = string("op_3365_cast_fp16")]; + tensor var_3366_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_289_cast_fp16)[name = string("op_3366_cast_fp16")]; + tensor var_3367_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_291_cast_fp16)[name = string("op_3367_cast_fp16")]; + tensor var_3368_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_293_cast_fp16)[name = string("op_3368_cast_fp16")]; + tensor var_3369_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_295_cast_fp16)[name = string("op_3369_cast_fp16")]; + tensor var_3370_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_297_cast_fp16)[name = string("op_3370_cast_fp16")]; + tensor var_3371_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_299_cast_fp16)[name = string("op_3371_cast_fp16")]; + tensor var_3372_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_301_cast_fp16)[name = string("op_3372_cast_fp16")]; + tensor var_3373_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_303_cast_fp16)[name = string("op_3373_cast_fp16")]; + tensor var_3374_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_305_cast_fp16)[name = string("op_3374_cast_fp16")]; + tensor var_3375_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_307_cast_fp16)[name = string("op_3375_cast_fp16")]; + tensor var_3376_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_309_cast_fp16)[name = string("op_3376_cast_fp16")]; + tensor var_3377_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_311_cast_fp16)[name = string("op_3377_cast_fp16")]; + tensor var_3378_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_313_cast_fp16)[name = string("op_3378_cast_fp16")]; + tensor var_3379_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_315_cast_fp16)[name = string("op_3379_cast_fp16")]; + tensor var_3380_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_317_cast_fp16)[name = string("op_3380_cast_fp16")]; + tensor var_3381_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_319_cast_fp16)[name = string("op_3381_cast_fp16")]; + string var_3383_equation_0 = const()[name = string("op_3383_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3383_cast_fp16 = einsum(equation = var_3383_equation_0, values = (var_3191_cast_fp16, var_3350_cast_fp16))[name = string("op_3383_cast_fp16")]; + string var_3385_equation_0 = const()[name = string("op_3385_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3385_cast_fp16 = einsum(equation = var_3385_equation_0, values = (var_3191_cast_fp16, var_3351_cast_fp16))[name = string("op_3385_cast_fp16")]; + string var_3387_equation_0 = const()[name = string("op_3387_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3387_cast_fp16 = einsum(equation = var_3387_equation_0, values = (var_3191_cast_fp16, var_3352_cast_fp16))[name = string("op_3387_cast_fp16")]; + string var_3389_equation_0 = const()[name = string("op_3389_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3389_cast_fp16 = einsum(equation = var_3389_equation_0, values = (var_3191_cast_fp16, var_3353_cast_fp16))[name = string("op_3389_cast_fp16")]; + string var_3391_equation_0 = const()[name = string("op_3391_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3391_cast_fp16 = einsum(equation = var_3391_equation_0, values = (var_3195_cast_fp16, var_3354_cast_fp16))[name = string("op_3391_cast_fp16")]; + string var_3393_equation_0 = const()[name = string("op_3393_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3393_cast_fp16 = einsum(equation = var_3393_equation_0, values = (var_3195_cast_fp16, var_3355_cast_fp16))[name = string("op_3393_cast_fp16")]; + string var_3395_equation_0 = const()[name = string("op_3395_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3395_cast_fp16 = einsum(equation = var_3395_equation_0, values = (var_3195_cast_fp16, var_3356_cast_fp16))[name = string("op_3395_cast_fp16")]; + string var_3397_equation_0 = const()[name = string("op_3397_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3397_cast_fp16 = einsum(equation = var_3397_equation_0, values = (var_3195_cast_fp16, var_3357_cast_fp16))[name = string("op_3397_cast_fp16")]; + string var_3399_equation_0 = const()[name = string("op_3399_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3399_cast_fp16 = einsum(equation = var_3399_equation_0, values = (var_3199_cast_fp16, var_3358_cast_fp16))[name = string("op_3399_cast_fp16")]; + string var_3401_equation_0 = const()[name = string("op_3401_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3401_cast_fp16 = einsum(equation = var_3401_equation_0, values = (var_3199_cast_fp16, var_3359_cast_fp16))[name = string("op_3401_cast_fp16")]; + string var_3403_equation_0 = const()[name = string("op_3403_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3403_cast_fp16 = einsum(equation = var_3403_equation_0, values = (var_3199_cast_fp16, var_3360_cast_fp16))[name = string("op_3403_cast_fp16")]; + string var_3405_equation_0 = const()[name = string("op_3405_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3405_cast_fp16 = einsum(equation = var_3405_equation_0, values = (var_3199_cast_fp16, var_3361_cast_fp16))[name = string("op_3405_cast_fp16")]; + string var_3407_equation_0 = const()[name = string("op_3407_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3407_cast_fp16 = einsum(equation = var_3407_equation_0, values = (var_3203_cast_fp16, var_3362_cast_fp16))[name = string("op_3407_cast_fp16")]; + string var_3409_equation_0 = const()[name = string("op_3409_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3409_cast_fp16 = einsum(equation = var_3409_equation_0, values = (var_3203_cast_fp16, var_3363_cast_fp16))[name = string("op_3409_cast_fp16")]; + string var_3411_equation_0 = const()[name = string("op_3411_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3411_cast_fp16 = einsum(equation = var_3411_equation_0, values = (var_3203_cast_fp16, var_3364_cast_fp16))[name = string("op_3411_cast_fp16")]; + string var_3413_equation_0 = const()[name = string("op_3413_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3413_cast_fp16 = einsum(equation = var_3413_equation_0, values = (var_3203_cast_fp16, var_3365_cast_fp16))[name = string("op_3413_cast_fp16")]; + string var_3415_equation_0 = const()[name = string("op_3415_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3415_cast_fp16 = einsum(equation = var_3415_equation_0, values = (var_3207_cast_fp16, var_3366_cast_fp16))[name = string("op_3415_cast_fp16")]; + string var_3417_equation_0 = const()[name = string("op_3417_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3417_cast_fp16 = einsum(equation = var_3417_equation_0, values = (var_3207_cast_fp16, var_3367_cast_fp16))[name = string("op_3417_cast_fp16")]; + string var_3419_equation_0 = const()[name = string("op_3419_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3419_cast_fp16 = einsum(equation = var_3419_equation_0, values = (var_3207_cast_fp16, var_3368_cast_fp16))[name = string("op_3419_cast_fp16")]; + string var_3421_equation_0 = const()[name = string("op_3421_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3421_cast_fp16 = einsum(equation = var_3421_equation_0, values = (var_3207_cast_fp16, var_3369_cast_fp16))[name = string("op_3421_cast_fp16")]; + string var_3423_equation_0 = const()[name = string("op_3423_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3423_cast_fp16 = einsum(equation = var_3423_equation_0, values = (var_3211_cast_fp16, var_3370_cast_fp16))[name = string("op_3423_cast_fp16")]; + string var_3425_equation_0 = const()[name = string("op_3425_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3425_cast_fp16 = einsum(equation = var_3425_equation_0, values = (var_3211_cast_fp16, var_3371_cast_fp16))[name = string("op_3425_cast_fp16")]; + string var_3427_equation_0 = const()[name = string("op_3427_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3427_cast_fp16 = einsum(equation = var_3427_equation_0, values = (var_3211_cast_fp16, var_3372_cast_fp16))[name = string("op_3427_cast_fp16")]; + string var_3429_equation_0 = const()[name = string("op_3429_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3429_cast_fp16 = einsum(equation = var_3429_equation_0, values = (var_3211_cast_fp16, var_3373_cast_fp16))[name = string("op_3429_cast_fp16")]; + string var_3431_equation_0 = const()[name = string("op_3431_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3431_cast_fp16 = einsum(equation = var_3431_equation_0, values = (var_3215_cast_fp16, var_3374_cast_fp16))[name = string("op_3431_cast_fp16")]; + string var_3433_equation_0 = const()[name = string("op_3433_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3433_cast_fp16 = einsum(equation = var_3433_equation_0, values = (var_3215_cast_fp16, var_3375_cast_fp16))[name = string("op_3433_cast_fp16")]; + string var_3435_equation_0 = const()[name = string("op_3435_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3435_cast_fp16 = einsum(equation = var_3435_equation_0, values = (var_3215_cast_fp16, var_3376_cast_fp16))[name = string("op_3435_cast_fp16")]; + string var_3437_equation_0 = const()[name = string("op_3437_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3437_cast_fp16 = einsum(equation = var_3437_equation_0, values = (var_3215_cast_fp16, var_3377_cast_fp16))[name = string("op_3437_cast_fp16")]; + string var_3439_equation_0 = const()[name = string("op_3439_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3439_cast_fp16 = einsum(equation = var_3439_equation_0, values = (var_3219_cast_fp16, var_3378_cast_fp16))[name = string("op_3439_cast_fp16")]; + string var_3441_equation_0 = const()[name = string("op_3441_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3441_cast_fp16 = einsum(equation = var_3441_equation_0, values = (var_3219_cast_fp16, var_3379_cast_fp16))[name = string("op_3441_cast_fp16")]; + string var_3443_equation_0 = const()[name = string("op_3443_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3443_cast_fp16 = einsum(equation = var_3443_equation_0, values = (var_3219_cast_fp16, var_3380_cast_fp16))[name = string("op_3443_cast_fp16")]; + string var_3445_equation_0 = const()[name = string("op_3445_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3445_cast_fp16 = einsum(equation = var_3445_equation_0, values = (var_3219_cast_fp16, var_3381_cast_fp16))[name = string("op_3445_cast_fp16")]; + bool var_3447_interleave_0 = const()[name = string("op_3447_interleave_0"), val = bool(false)]; + tensor var_3447_cast_fp16 = concat(axis = var_2834, interleave = var_3447_interleave_0, values = (var_3383_cast_fp16, var_3385_cast_fp16, var_3387_cast_fp16, var_3389_cast_fp16))[name = string("op_3447_cast_fp16")]; + bool var_3449_interleave_0 = const()[name = string("op_3449_interleave_0"), val = bool(false)]; + tensor var_3449_cast_fp16 = concat(axis = var_2834, interleave = var_3449_interleave_0, values = (var_3391_cast_fp16, var_3393_cast_fp16, var_3395_cast_fp16, var_3397_cast_fp16))[name = string("op_3449_cast_fp16")]; + bool var_3451_interleave_0 = const()[name = string("op_3451_interleave_0"), val = bool(false)]; + tensor var_3451_cast_fp16 = concat(axis = var_2834, interleave = var_3451_interleave_0, values = (var_3399_cast_fp16, var_3401_cast_fp16, var_3403_cast_fp16, var_3405_cast_fp16))[name = string("op_3451_cast_fp16")]; + bool var_3453_interleave_0 = const()[name = string("op_3453_interleave_0"), val = bool(false)]; + tensor var_3453_cast_fp16 = concat(axis = var_2834, interleave = var_3453_interleave_0, values = (var_3407_cast_fp16, var_3409_cast_fp16, var_3411_cast_fp16, var_3413_cast_fp16))[name = string("op_3453_cast_fp16")]; + bool var_3455_interleave_0 = const()[name = string("op_3455_interleave_0"), val = bool(false)]; + tensor var_3455_cast_fp16 = concat(axis = var_2834, interleave = var_3455_interleave_0, values = (var_3415_cast_fp16, var_3417_cast_fp16, var_3419_cast_fp16, var_3421_cast_fp16))[name = string("op_3455_cast_fp16")]; + bool var_3457_interleave_0 = const()[name = string("op_3457_interleave_0"), val = bool(false)]; + tensor var_3457_cast_fp16 = concat(axis = var_2834, interleave = var_3457_interleave_0, values = (var_3423_cast_fp16, var_3425_cast_fp16, var_3427_cast_fp16, var_3429_cast_fp16))[name = string("op_3457_cast_fp16")]; + bool var_3459_interleave_0 = const()[name = string("op_3459_interleave_0"), val = bool(false)]; + tensor var_3459_cast_fp16 = concat(axis = var_2834, interleave = var_3459_interleave_0, values = (var_3431_cast_fp16, var_3433_cast_fp16, var_3435_cast_fp16, var_3437_cast_fp16))[name = string("op_3459_cast_fp16")]; + bool var_3461_interleave_0 = const()[name = string("op_3461_interleave_0"), val = bool(false)]; + tensor var_3461_cast_fp16 = concat(axis = var_2834, interleave = var_3461_interleave_0, values = (var_3439_cast_fp16, var_3441_cast_fp16, var_3443_cast_fp16, var_3445_cast_fp16))[name = string("op_3461_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_2847, interleave = input_33_interleave_0, values = (var_3447_cast_fp16, var_3449_cast_fp16, var_3451_cast_fp16, var_3453_cast_fp16, var_3455_cast_fp16, var_3457_cast_fp16, var_3459_cast_fp16, var_3461_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30155456)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30679808)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_3480_to_fp16 = const()[name = string("op_3480_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_3480_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30680896)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30681984)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30683072)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32780288)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32784448)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34881664)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_3509 = const()[name = string("op_3509"), val = int32(3)]; + int32 var_3522 = const()[name = string("op_3522"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_3539_to_fp16 = const()[name = string("op_3539_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_3539_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34882752)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34883840)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34884928)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35409280)))]; + tensor query_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35410368)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35934720)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36459072)))]; + tensor value_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_3577_begin_0 = const()[name = string("op_3577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3577_end_0 = const()[name = string("op_3577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3577_end_mask_0 = const()[name = string("op_3577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3577_cast_fp16 = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = query_cast_fp16)[name = string("op_3577_cast_fp16")]; + tensor var_3581_begin_0 = const()[name = string("op_3581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3581_end_0 = const()[name = string("op_3581_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3581_end_mask_0 = const()[name = string("op_3581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3581_cast_fp16 = slice_by_index(begin = var_3581_begin_0, end = var_3581_end_0, end_mask = var_3581_end_mask_0, x = query_cast_fp16)[name = string("op_3581_cast_fp16")]; + tensor var_3585_begin_0 = const()[name = string("op_3585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3585_end_0 = const()[name = string("op_3585_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3585_end_mask_0 = const()[name = string("op_3585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3585_cast_fp16 = slice_by_index(begin = var_3585_begin_0, end = var_3585_end_0, end_mask = var_3585_end_mask_0, x = query_cast_fp16)[name = string("op_3585_cast_fp16")]; + tensor var_3589_begin_0 = const()[name = string("op_3589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3589_end_0 = const()[name = string("op_3589_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3589_end_mask_0 = const()[name = string("op_3589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3589_cast_fp16 = slice_by_index(begin = var_3589_begin_0, end = var_3589_end_0, end_mask = var_3589_end_mask_0, x = query_cast_fp16)[name = string("op_3589_cast_fp16")]; + tensor var_3593_begin_0 = const()[name = string("op_3593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3593_end_0 = const()[name = string("op_3593_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3593_end_mask_0 = const()[name = string("op_3593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3593_cast_fp16 = slice_by_index(begin = var_3593_begin_0, end = var_3593_end_0, end_mask = var_3593_end_mask_0, x = query_cast_fp16)[name = string("op_3593_cast_fp16")]; + tensor var_3597_begin_0 = const()[name = string("op_3597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3597_end_0 = const()[name = string("op_3597_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3597_end_mask_0 = const()[name = string("op_3597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3597_cast_fp16 = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, x = query_cast_fp16)[name = string("op_3597_cast_fp16")]; + tensor var_3601_begin_0 = const()[name = string("op_3601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3601_end_0 = const()[name = string("op_3601_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3601_end_mask_0 = const()[name = string("op_3601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3601_cast_fp16 = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, x = query_cast_fp16)[name = string("op_3601_cast_fp16")]; + tensor var_3605_begin_0 = const()[name = string("op_3605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3605_end_0 = const()[name = string("op_3605_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3605_end_mask_0 = const()[name = string("op_3605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3605_cast_fp16 = slice_by_index(begin = var_3605_begin_0, end = var_3605_end_0, end_mask = var_3605_end_mask_0, x = query_cast_fp16)[name = string("op_3605_cast_fp16")]; + tensor var_3614_begin_0 = const()[name = string("op_3614_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3614_end_0 = const()[name = string("op_3614_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3614_end_mask_0 = const()[name = string("op_3614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3614_cast_fp16 = slice_by_index(begin = var_3614_begin_0, end = var_3614_end_0, end_mask = var_3614_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3614_cast_fp16")]; + tensor var_3621_begin_0 = const()[name = string("op_3621_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3621_end_0 = const()[name = string("op_3621_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3621_end_mask_0 = const()[name = string("op_3621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3621_cast_fp16 = slice_by_index(begin = var_3621_begin_0, end = var_3621_end_0, end_mask = var_3621_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3621_cast_fp16")]; + tensor var_3628_begin_0 = const()[name = string("op_3628_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3628_end_0 = const()[name = string("op_3628_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3628_end_mask_0 = const()[name = string("op_3628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3628_cast_fp16 = slice_by_index(begin = var_3628_begin_0, end = var_3628_end_0, end_mask = var_3628_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3628_cast_fp16")]; + tensor var_3635_begin_0 = const()[name = string("op_3635_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3635_end_0 = const()[name = string("op_3635_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3635_end_mask_0 = const()[name = string("op_3635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3635_cast_fp16 = slice_by_index(begin = var_3635_begin_0, end = var_3635_end_0, end_mask = var_3635_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3635_cast_fp16")]; + tensor var_3642_begin_0 = const()[name = string("op_3642_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3642_end_0 = const()[name = string("op_3642_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3642_end_mask_0 = const()[name = string("op_3642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3642_cast_fp16 = slice_by_index(begin = var_3642_begin_0, end = var_3642_end_0, end_mask = var_3642_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3642_cast_fp16")]; + tensor var_3649_begin_0 = const()[name = string("op_3649_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3649_end_0 = const()[name = string("op_3649_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3649_end_mask_0 = const()[name = string("op_3649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3649_cast_fp16 = slice_by_index(begin = var_3649_begin_0, end = var_3649_end_0, end_mask = var_3649_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3649_cast_fp16")]; + tensor var_3656_begin_0 = const()[name = string("op_3656_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3656_end_0 = const()[name = string("op_3656_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3656_end_mask_0 = const()[name = string("op_3656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3656_cast_fp16 = slice_by_index(begin = var_3656_begin_0, end = var_3656_end_0, end_mask = var_3656_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3656_cast_fp16")]; + tensor var_3663_begin_0 = const()[name = string("op_3663_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3663_end_0 = const()[name = string("op_3663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3663_end_mask_0 = const()[name = string("op_3663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3663_cast_fp16 = slice_by_index(begin = var_3663_begin_0, end = var_3663_end_0, end_mask = var_3663_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3663_cast_fp16")]; + tensor var_3670_begin_0 = const()[name = string("op_3670_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3670_end_0 = const()[name = string("op_3670_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3670_end_mask_0 = const()[name = string("op_3670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3670_cast_fp16 = slice_by_index(begin = var_3670_begin_0, end = var_3670_end_0, end_mask = var_3670_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3670_cast_fp16")]; + tensor var_3677_begin_0 = const()[name = string("op_3677_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3677_end_0 = const()[name = string("op_3677_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3677_end_mask_0 = const()[name = string("op_3677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3677_cast_fp16 = slice_by_index(begin = var_3677_begin_0, end = var_3677_end_0, end_mask = var_3677_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3677_cast_fp16")]; + tensor var_3684_begin_0 = const()[name = string("op_3684_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3684_end_0 = const()[name = string("op_3684_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3684_end_mask_0 = const()[name = string("op_3684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3684_cast_fp16 = slice_by_index(begin = var_3684_begin_0, end = var_3684_end_0, end_mask = var_3684_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3684_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = string("op_3691_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3691_end_0 = const()[name = string("op_3691_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3691_end_mask_0 = const()[name = string("op_3691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3691_cast_fp16")]; + tensor var_3698_begin_0 = const()[name = string("op_3698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3698_end_0 = const()[name = string("op_3698_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3698_end_mask_0 = const()[name = string("op_3698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3698_cast_fp16 = slice_by_index(begin = var_3698_begin_0, end = var_3698_end_0, end_mask = var_3698_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3698_cast_fp16")]; + tensor var_3705_begin_0 = const()[name = string("op_3705_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3705_end_0 = const()[name = string("op_3705_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3705_end_mask_0 = const()[name = string("op_3705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3705_cast_fp16 = slice_by_index(begin = var_3705_begin_0, end = var_3705_end_0, end_mask = var_3705_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3705_cast_fp16")]; + tensor var_3712_begin_0 = const()[name = string("op_3712_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3712_end_0 = const()[name = string("op_3712_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3712_end_mask_0 = const()[name = string("op_3712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3712_cast_fp16 = slice_by_index(begin = var_3712_begin_0, end = var_3712_end_0, end_mask = var_3712_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3712_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = string("op_3719_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3719_end_0 = const()[name = string("op_3719_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3719_end_mask_0 = const()[name = string("op_3719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3719_cast_fp16")]; + tensor var_3726_begin_0 = const()[name = string("op_3726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3726_end_0 = const()[name = string("op_3726_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3726_end_mask_0 = const()[name = string("op_3726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3726_cast_fp16 = slice_by_index(begin = var_3726_begin_0, end = var_3726_end_0, end_mask = var_3726_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3726_cast_fp16")]; + tensor var_3733_begin_0 = const()[name = string("op_3733_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3733_end_0 = const()[name = string("op_3733_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3733_end_mask_0 = const()[name = string("op_3733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3733_cast_fp16 = slice_by_index(begin = var_3733_begin_0, end = var_3733_end_0, end_mask = var_3733_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3733_cast_fp16")]; + tensor var_3740_begin_0 = const()[name = string("op_3740_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3740_end_0 = const()[name = string("op_3740_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3740_end_mask_0 = const()[name = string("op_3740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3740_cast_fp16 = slice_by_index(begin = var_3740_begin_0, end = var_3740_end_0, end_mask = var_3740_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3740_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = string("op_3747_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3747_end_0 = const()[name = string("op_3747_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3747_end_mask_0 = const()[name = string("op_3747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3747_cast_fp16")]; + tensor var_3754_begin_0 = const()[name = string("op_3754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3754_end_0 = const()[name = string("op_3754_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3754_end_mask_0 = const()[name = string("op_3754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3754_cast_fp16 = slice_by_index(begin = var_3754_begin_0, end = var_3754_end_0, end_mask = var_3754_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3754_cast_fp16")]; + tensor var_3761_begin_0 = const()[name = string("op_3761_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3761_end_0 = const()[name = string("op_3761_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3761_end_mask_0 = const()[name = string("op_3761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3761_cast_fp16 = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3761_cast_fp16")]; + tensor var_3768_begin_0 = const()[name = string("op_3768_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3768_end_0 = const()[name = string("op_3768_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3768_end_mask_0 = const()[name = string("op_3768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3768_cast_fp16 = slice_by_index(begin = var_3768_begin_0, end = var_3768_end_0, end_mask = var_3768_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3768_cast_fp16")]; + tensor var_3775_begin_0 = const()[name = string("op_3775_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3775_end_0 = const()[name = string("op_3775_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3775_end_mask_0 = const()[name = string("op_3775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3775_cast_fp16 = slice_by_index(begin = var_3775_begin_0, end = var_3775_end_0, end_mask = var_3775_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3775_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = string("op_3782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3782_end_0 = const()[name = string("op_3782_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3782_end_mask_0 = const()[name = string("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3782_cast_fp16")]; + tensor var_3789_begin_0 = const()[name = string("op_3789_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3789_end_0 = const()[name = string("op_3789_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3789_end_mask_0 = const()[name = string("op_3789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3789_cast_fp16 = slice_by_index(begin = var_3789_begin_0, end = var_3789_end_0, end_mask = var_3789_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3789_cast_fp16")]; + tensor var_3796_begin_0 = const()[name = string("op_3796_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3796_end_0 = const()[name = string("op_3796_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3796_end_mask_0 = const()[name = string("op_3796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3796_cast_fp16 = slice_by_index(begin = var_3796_begin_0, end = var_3796_end_0, end_mask = var_3796_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3796_cast_fp16")]; + tensor var_3803_begin_0 = const()[name = string("op_3803_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3803_end_0 = const()[name = string("op_3803_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3803_end_mask_0 = const()[name = string("op_3803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3803_cast_fp16 = slice_by_index(begin = var_3803_begin_0, end = var_3803_end_0, end_mask = var_3803_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3803_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = string("op_3810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3810_end_0 = const()[name = string("op_3810_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3810_end_mask_0 = const()[name = string("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3810_cast_fp16")]; + tensor var_3817_begin_0 = const()[name = string("op_3817_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3817_end_0 = const()[name = string("op_3817_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3817_end_mask_0 = const()[name = string("op_3817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3817_cast_fp16 = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3817_cast_fp16")]; + tensor var_3824_begin_0 = const()[name = string("op_3824_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3824_end_0 = const()[name = string("op_3824_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3824_end_mask_0 = const()[name = string("op_3824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3824_cast_fp16 = slice_by_index(begin = var_3824_begin_0, end = var_3824_end_0, end_mask = var_3824_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3824_cast_fp16")]; + tensor var_3831_begin_0 = const()[name = string("op_3831_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3831_end_0 = const()[name = string("op_3831_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3831_end_mask_0 = const()[name = string("op_3831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3831_cast_fp16 = slice_by_index(begin = var_3831_begin_0, end = var_3831_end_0, end_mask = var_3831_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3831_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3836_begin_0 = const()[name = string("op_3836_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3836_end_0 = const()[name = string("op_3836_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3836_end_mask_0 = const()[name = string("op_3836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_3836_cast_fp16 = slice_by_index(begin = var_3836_begin_0, end = var_3836_end_0, end_mask = var_3836_end_mask_0, x = k_11_cast_fp16)[name = string("op_3836_cast_fp16")]; + tensor var_3840_begin_0 = const()[name = string("op_3840_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3840_end_0 = const()[name = string("op_3840_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3840_end_mask_0 = const()[name = string("op_3840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3840_cast_fp16 = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, x = k_11_cast_fp16)[name = string("op_3840_cast_fp16")]; + tensor var_3844_begin_0 = const()[name = string("op_3844_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3844_end_0 = const()[name = string("op_3844_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3844_end_mask_0 = const()[name = string("op_3844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3844_cast_fp16 = slice_by_index(begin = var_3844_begin_0, end = var_3844_end_0, end_mask = var_3844_end_mask_0, x = k_11_cast_fp16)[name = string("op_3844_cast_fp16")]; + tensor var_3848_begin_0 = const()[name = string("op_3848_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3848_end_0 = const()[name = string("op_3848_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3848_end_mask_0 = const()[name = string("op_3848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3848_cast_fp16 = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = k_11_cast_fp16)[name = string("op_3848_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = string("op_3852_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3852_end_0 = const()[name = string("op_3852_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3852_end_mask_0 = const()[name = string("op_3852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = k_11_cast_fp16)[name = string("op_3852_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = string("op_3856_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3856_end_0 = const()[name = string("op_3856_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3856_end_mask_0 = const()[name = string("op_3856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = k_11_cast_fp16)[name = string("op_3856_cast_fp16")]; + tensor var_3860_begin_0 = const()[name = string("op_3860_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3860_end_0 = const()[name = string("op_3860_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3860_end_mask_0 = const()[name = string("op_3860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3860_cast_fp16 = slice_by_index(begin = var_3860_begin_0, end = var_3860_end_0, end_mask = var_3860_end_mask_0, x = k_11_cast_fp16)[name = string("op_3860_cast_fp16")]; + tensor var_3864_begin_0 = const()[name = string("op_3864_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3864_end_0 = const()[name = string("op_3864_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3864_end_mask_0 = const()[name = string("op_3864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3864_cast_fp16 = slice_by_index(begin = var_3864_begin_0, end = var_3864_end_0, end_mask = var_3864_end_mask_0, x = k_11_cast_fp16)[name = string("op_3864_cast_fp16")]; + tensor var_3866_begin_0 = const()[name = string("op_3866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3866_end_0 = const()[name = string("op_3866_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3866_end_mask_0 = const()[name = string("op_3866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3866_cast_fp16 = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = value_cast_fp16)[name = string("op_3866_cast_fp16")]; + tensor var_3870_begin_0 = const()[name = string("op_3870_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3870_end_0 = const()[name = string("op_3870_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3870_end_mask_0 = const()[name = string("op_3870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3870_cast_fp16 = slice_by_index(begin = var_3870_begin_0, end = var_3870_end_0, end_mask = var_3870_end_mask_0, x = value_cast_fp16)[name = string("op_3870_cast_fp16")]; + tensor var_3874_begin_0 = const()[name = string("op_3874_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3874_end_0 = const()[name = string("op_3874_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3874_end_mask_0 = const()[name = string("op_3874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3874_cast_fp16 = slice_by_index(begin = var_3874_begin_0, end = var_3874_end_0, end_mask = var_3874_end_mask_0, x = value_cast_fp16)[name = string("op_3874_cast_fp16")]; + tensor var_3878_begin_0 = const()[name = string("op_3878_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3878_end_0 = const()[name = string("op_3878_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3878_end_mask_0 = const()[name = string("op_3878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3878_cast_fp16 = slice_by_index(begin = var_3878_begin_0, end = var_3878_end_0, end_mask = var_3878_end_mask_0, x = value_cast_fp16)[name = string("op_3878_cast_fp16")]; + tensor var_3882_begin_0 = const()[name = string("op_3882_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3882_end_0 = const()[name = string("op_3882_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3882_end_mask_0 = const()[name = string("op_3882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3882_cast_fp16 = slice_by_index(begin = var_3882_begin_0, end = var_3882_end_0, end_mask = var_3882_end_mask_0, x = value_cast_fp16)[name = string("op_3882_cast_fp16")]; + tensor var_3886_begin_0 = const()[name = string("op_3886_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3886_end_0 = const()[name = string("op_3886_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3886_end_mask_0 = const()[name = string("op_3886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3886_cast_fp16 = slice_by_index(begin = var_3886_begin_0, end = var_3886_end_0, end_mask = var_3886_end_mask_0, x = value_cast_fp16)[name = string("op_3886_cast_fp16")]; + tensor var_3890_begin_0 = const()[name = string("op_3890_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3890_end_0 = const()[name = string("op_3890_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3890_end_mask_0 = const()[name = string("op_3890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3890_cast_fp16 = slice_by_index(begin = var_3890_begin_0, end = var_3890_end_0, end_mask = var_3890_end_mask_0, x = value_cast_fp16)[name = string("op_3890_cast_fp16")]; + tensor var_3894_begin_0 = const()[name = string("op_3894_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3894_end_0 = const()[name = string("op_3894_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3894_end_mask_0 = const()[name = string("op_3894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3894_cast_fp16 = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = value_cast_fp16)[name = string("op_3894_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3836_cast_fp16, var_3614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3836_cast_fp16, var_3621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3836_cast_fp16, var_3628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3836_cast_fp16, var_3635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3840_cast_fp16, var_3642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3840_cast_fp16, var_3649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3840_cast_fp16, var_3656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3840_cast_fp16, var_3663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3844_cast_fp16, var_3670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3844_cast_fp16, var_3677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3844_cast_fp16, var_3684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3844_cast_fp16, var_3691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3848_cast_fp16, var_3698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3848_cast_fp16, var_3705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3848_cast_fp16, var_3712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3848_cast_fp16, var_3719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3852_cast_fp16, var_3726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3852_cast_fp16, var_3733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3852_cast_fp16, var_3740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3852_cast_fp16, var_3747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3856_cast_fp16, var_3754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3856_cast_fp16, var_3761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3856_cast_fp16, var_3768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3856_cast_fp16, var_3775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3860_cast_fp16, var_3782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3860_cast_fp16, var_3789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3860_cast_fp16, var_3796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3860_cast_fp16, var_3803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3864_cast_fp16, var_3810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3864_cast_fp16, var_3817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3864_cast_fp16, var_3824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_3864_cast_fp16, var_3831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_3961_to_fp16 = const()[name = string("op_3961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_3961_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_3963_to_fp16 = const()[name = string("op_3963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_3963_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_3965_to_fp16 = const()[name = string("op_3965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_3965_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_3967_to_fp16 = const()[name = string("op_3967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_3967_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_3969_to_fp16 = const()[name = string("op_3969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_3969_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_3971_to_fp16 = const()[name = string("op_3971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_3971_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_3973_to_fp16 = const()[name = string("op_3973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_3973_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_3975_to_fp16 = const()[name = string("op_3975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_3975_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_3977_to_fp16 = const()[name = string("op_3977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_3977_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_3979_to_fp16 = const()[name = string("op_3979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_3979_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_3981_to_fp16 = const()[name = string("op_3981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_3981_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_3983_to_fp16 = const()[name = string("op_3983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_3983_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_3985_to_fp16 = const()[name = string("op_3985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_3985_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_3987_to_fp16 = const()[name = string("op_3987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_3987_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_3989_to_fp16 = const()[name = string("op_3989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_3989_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_3991_to_fp16 = const()[name = string("op_3991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_3991_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_3993_to_fp16 = const()[name = string("op_3993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_3993_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_3995_to_fp16 = const()[name = string("op_3995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_3995_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_3997_to_fp16 = const()[name = string("op_3997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_3997_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_3999_to_fp16 = const()[name = string("op_3999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_3999_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_4001_to_fp16 = const()[name = string("op_4001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_4001_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_4003_to_fp16 = const()[name = string("op_4003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_4003_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_4005_to_fp16 = const()[name = string("op_4005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_4005_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_4007_to_fp16 = const()[name = string("op_4007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_4007_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_4009_to_fp16 = const()[name = string("op_4009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_4009_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_4011_to_fp16 = const()[name = string("op_4011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_4011_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_4013_to_fp16 = const()[name = string("op_4013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_4013_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_4015_to_fp16 = const()[name = string("op_4015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_4015_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_4017_to_fp16 = const()[name = string("op_4017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_4017_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_4019_to_fp16 = const()[name = string("op_4019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_4019_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_4021_to_fp16 = const()[name = string("op_4021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_4021_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_4023_to_fp16 = const()[name = string("op_4023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_4023_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_4025_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_321_cast_fp16)[name = string("op_4025_cast_fp16")]; + tensor var_4026_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_323_cast_fp16)[name = string("op_4026_cast_fp16")]; + tensor var_4027_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_325_cast_fp16)[name = string("op_4027_cast_fp16")]; + tensor var_4028_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_327_cast_fp16)[name = string("op_4028_cast_fp16")]; + tensor var_4029_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_329_cast_fp16)[name = string("op_4029_cast_fp16")]; + tensor var_4030_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_331_cast_fp16)[name = string("op_4030_cast_fp16")]; + tensor var_4031_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_333_cast_fp16)[name = string("op_4031_cast_fp16")]; + tensor var_4032_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_335_cast_fp16)[name = string("op_4032_cast_fp16")]; + tensor var_4033_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_337_cast_fp16)[name = string("op_4033_cast_fp16")]; + tensor var_4034_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_339_cast_fp16)[name = string("op_4034_cast_fp16")]; + tensor var_4035_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_341_cast_fp16)[name = string("op_4035_cast_fp16")]; + tensor var_4036_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_343_cast_fp16)[name = string("op_4036_cast_fp16")]; + tensor var_4037_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_345_cast_fp16)[name = string("op_4037_cast_fp16")]; + tensor var_4038_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_347_cast_fp16)[name = string("op_4038_cast_fp16")]; + tensor var_4039_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_349_cast_fp16)[name = string("op_4039_cast_fp16")]; + tensor var_4040_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_351_cast_fp16)[name = string("op_4040_cast_fp16")]; + tensor var_4041_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_353_cast_fp16)[name = string("op_4041_cast_fp16")]; + tensor var_4042_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_355_cast_fp16)[name = string("op_4042_cast_fp16")]; + tensor var_4043_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_357_cast_fp16)[name = string("op_4043_cast_fp16")]; + tensor var_4044_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_359_cast_fp16)[name = string("op_4044_cast_fp16")]; + tensor var_4045_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_361_cast_fp16)[name = string("op_4045_cast_fp16")]; + tensor var_4046_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_363_cast_fp16)[name = string("op_4046_cast_fp16")]; + tensor var_4047_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_365_cast_fp16)[name = string("op_4047_cast_fp16")]; + tensor var_4048_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_367_cast_fp16)[name = string("op_4048_cast_fp16")]; + tensor var_4049_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_369_cast_fp16)[name = string("op_4049_cast_fp16")]; + tensor var_4050_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_371_cast_fp16)[name = string("op_4050_cast_fp16")]; + tensor var_4051_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_373_cast_fp16)[name = string("op_4051_cast_fp16")]; + tensor var_4052_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_375_cast_fp16)[name = string("op_4052_cast_fp16")]; + tensor var_4053_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_377_cast_fp16)[name = string("op_4053_cast_fp16")]; + tensor var_4054_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_379_cast_fp16)[name = string("op_4054_cast_fp16")]; + tensor var_4055_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_381_cast_fp16)[name = string("op_4055_cast_fp16")]; + tensor var_4056_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_cast_fp16)[name = string("op_4056_cast_fp16")]; + string var_4058_equation_0 = const()[name = string("op_4058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4058_cast_fp16 = einsum(equation = var_4058_equation_0, values = (var_3866_cast_fp16, var_4025_cast_fp16))[name = string("op_4058_cast_fp16")]; + string var_4060_equation_0 = const()[name = string("op_4060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4060_cast_fp16 = einsum(equation = var_4060_equation_0, values = (var_3866_cast_fp16, var_4026_cast_fp16))[name = string("op_4060_cast_fp16")]; + string var_4062_equation_0 = const()[name = string("op_4062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4062_cast_fp16 = einsum(equation = var_4062_equation_0, values = (var_3866_cast_fp16, var_4027_cast_fp16))[name = string("op_4062_cast_fp16")]; + string var_4064_equation_0 = const()[name = string("op_4064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4064_cast_fp16 = einsum(equation = var_4064_equation_0, values = (var_3866_cast_fp16, var_4028_cast_fp16))[name = string("op_4064_cast_fp16")]; + string var_4066_equation_0 = const()[name = string("op_4066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4066_cast_fp16 = einsum(equation = var_4066_equation_0, values = (var_3870_cast_fp16, var_4029_cast_fp16))[name = string("op_4066_cast_fp16")]; + string var_4068_equation_0 = const()[name = string("op_4068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4068_cast_fp16 = einsum(equation = var_4068_equation_0, values = (var_3870_cast_fp16, var_4030_cast_fp16))[name = string("op_4068_cast_fp16")]; + string var_4070_equation_0 = const()[name = string("op_4070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4070_cast_fp16 = einsum(equation = var_4070_equation_0, values = (var_3870_cast_fp16, var_4031_cast_fp16))[name = string("op_4070_cast_fp16")]; + string var_4072_equation_0 = const()[name = string("op_4072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4072_cast_fp16 = einsum(equation = var_4072_equation_0, values = (var_3870_cast_fp16, var_4032_cast_fp16))[name = string("op_4072_cast_fp16")]; + string var_4074_equation_0 = const()[name = string("op_4074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4074_cast_fp16 = einsum(equation = var_4074_equation_0, values = (var_3874_cast_fp16, var_4033_cast_fp16))[name = string("op_4074_cast_fp16")]; + string var_4076_equation_0 = const()[name = string("op_4076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4076_cast_fp16 = einsum(equation = var_4076_equation_0, values = (var_3874_cast_fp16, var_4034_cast_fp16))[name = string("op_4076_cast_fp16")]; + string var_4078_equation_0 = const()[name = string("op_4078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4078_cast_fp16 = einsum(equation = var_4078_equation_0, values = (var_3874_cast_fp16, var_4035_cast_fp16))[name = string("op_4078_cast_fp16")]; + string var_4080_equation_0 = const()[name = string("op_4080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4080_cast_fp16 = einsum(equation = var_4080_equation_0, values = (var_3874_cast_fp16, var_4036_cast_fp16))[name = string("op_4080_cast_fp16")]; + string var_4082_equation_0 = const()[name = string("op_4082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4082_cast_fp16 = einsum(equation = var_4082_equation_0, values = (var_3878_cast_fp16, var_4037_cast_fp16))[name = string("op_4082_cast_fp16")]; + string var_4084_equation_0 = const()[name = string("op_4084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4084_cast_fp16 = einsum(equation = var_4084_equation_0, values = (var_3878_cast_fp16, var_4038_cast_fp16))[name = string("op_4084_cast_fp16")]; + string var_4086_equation_0 = const()[name = string("op_4086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4086_cast_fp16 = einsum(equation = var_4086_equation_0, values = (var_3878_cast_fp16, var_4039_cast_fp16))[name = string("op_4086_cast_fp16")]; + string var_4088_equation_0 = const()[name = string("op_4088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4088_cast_fp16 = einsum(equation = var_4088_equation_0, values = (var_3878_cast_fp16, var_4040_cast_fp16))[name = string("op_4088_cast_fp16")]; + string var_4090_equation_0 = const()[name = string("op_4090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4090_cast_fp16 = einsum(equation = var_4090_equation_0, values = (var_3882_cast_fp16, var_4041_cast_fp16))[name = string("op_4090_cast_fp16")]; + string var_4092_equation_0 = const()[name = string("op_4092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4092_cast_fp16 = einsum(equation = var_4092_equation_0, values = (var_3882_cast_fp16, var_4042_cast_fp16))[name = string("op_4092_cast_fp16")]; + string var_4094_equation_0 = const()[name = string("op_4094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4094_cast_fp16 = einsum(equation = var_4094_equation_0, values = (var_3882_cast_fp16, var_4043_cast_fp16))[name = string("op_4094_cast_fp16")]; + string var_4096_equation_0 = const()[name = string("op_4096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4096_cast_fp16 = einsum(equation = var_4096_equation_0, values = (var_3882_cast_fp16, var_4044_cast_fp16))[name = string("op_4096_cast_fp16")]; + string var_4098_equation_0 = const()[name = string("op_4098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4098_cast_fp16 = einsum(equation = var_4098_equation_0, values = (var_3886_cast_fp16, var_4045_cast_fp16))[name = string("op_4098_cast_fp16")]; + string var_4100_equation_0 = const()[name = string("op_4100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4100_cast_fp16 = einsum(equation = var_4100_equation_0, values = (var_3886_cast_fp16, var_4046_cast_fp16))[name = string("op_4100_cast_fp16")]; + string var_4102_equation_0 = const()[name = string("op_4102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4102_cast_fp16 = einsum(equation = var_4102_equation_0, values = (var_3886_cast_fp16, var_4047_cast_fp16))[name = string("op_4102_cast_fp16")]; + string var_4104_equation_0 = const()[name = string("op_4104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4104_cast_fp16 = einsum(equation = var_4104_equation_0, values = (var_3886_cast_fp16, var_4048_cast_fp16))[name = string("op_4104_cast_fp16")]; + string var_4106_equation_0 = const()[name = string("op_4106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4106_cast_fp16 = einsum(equation = var_4106_equation_0, values = (var_3890_cast_fp16, var_4049_cast_fp16))[name = string("op_4106_cast_fp16")]; + string var_4108_equation_0 = const()[name = string("op_4108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4108_cast_fp16 = einsum(equation = var_4108_equation_0, values = (var_3890_cast_fp16, var_4050_cast_fp16))[name = string("op_4108_cast_fp16")]; + string var_4110_equation_0 = const()[name = string("op_4110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4110_cast_fp16 = einsum(equation = var_4110_equation_0, values = (var_3890_cast_fp16, var_4051_cast_fp16))[name = string("op_4110_cast_fp16")]; + string var_4112_equation_0 = const()[name = string("op_4112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4112_cast_fp16 = einsum(equation = var_4112_equation_0, values = (var_3890_cast_fp16, var_4052_cast_fp16))[name = string("op_4112_cast_fp16")]; + string var_4114_equation_0 = const()[name = string("op_4114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4114_cast_fp16 = einsum(equation = var_4114_equation_0, values = (var_3894_cast_fp16, var_4053_cast_fp16))[name = string("op_4114_cast_fp16")]; + string var_4116_equation_0 = const()[name = string("op_4116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4116_cast_fp16 = einsum(equation = var_4116_equation_0, values = (var_3894_cast_fp16, var_4054_cast_fp16))[name = string("op_4116_cast_fp16")]; + string var_4118_equation_0 = const()[name = string("op_4118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4118_cast_fp16 = einsum(equation = var_4118_equation_0, values = (var_3894_cast_fp16, var_4055_cast_fp16))[name = string("op_4118_cast_fp16")]; + string var_4120_equation_0 = const()[name = string("op_4120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4120_cast_fp16 = einsum(equation = var_4120_equation_0, values = (var_3894_cast_fp16, var_4056_cast_fp16))[name = string("op_4120_cast_fp16")]; + bool var_4122_interleave_0 = const()[name = string("op_4122_interleave_0"), val = bool(false)]; + tensor var_4122_cast_fp16 = concat(axis = var_3509, interleave = var_4122_interleave_0, values = (var_4058_cast_fp16, var_4060_cast_fp16, var_4062_cast_fp16, var_4064_cast_fp16))[name = string("op_4122_cast_fp16")]; + bool var_4124_interleave_0 = const()[name = string("op_4124_interleave_0"), val = bool(false)]; + tensor var_4124_cast_fp16 = concat(axis = var_3509, interleave = var_4124_interleave_0, values = (var_4066_cast_fp16, var_4068_cast_fp16, var_4070_cast_fp16, var_4072_cast_fp16))[name = string("op_4124_cast_fp16")]; + bool var_4126_interleave_0 = const()[name = string("op_4126_interleave_0"), val = bool(false)]; + tensor var_4126_cast_fp16 = concat(axis = var_3509, interleave = var_4126_interleave_0, values = (var_4074_cast_fp16, var_4076_cast_fp16, var_4078_cast_fp16, var_4080_cast_fp16))[name = string("op_4126_cast_fp16")]; + bool var_4128_interleave_0 = const()[name = string("op_4128_interleave_0"), val = bool(false)]; + tensor var_4128_cast_fp16 = concat(axis = var_3509, interleave = var_4128_interleave_0, values = (var_4082_cast_fp16, var_4084_cast_fp16, var_4086_cast_fp16, var_4088_cast_fp16))[name = string("op_4128_cast_fp16")]; + bool var_4130_interleave_0 = const()[name = string("op_4130_interleave_0"), val = bool(false)]; + tensor var_4130_cast_fp16 = concat(axis = var_3509, interleave = var_4130_interleave_0, values = (var_4090_cast_fp16, var_4092_cast_fp16, var_4094_cast_fp16, var_4096_cast_fp16))[name = string("op_4130_cast_fp16")]; + bool var_4132_interleave_0 = const()[name = string("op_4132_interleave_0"), val = bool(false)]; + tensor var_4132_cast_fp16 = concat(axis = var_3509, interleave = var_4132_interleave_0, values = (var_4098_cast_fp16, var_4100_cast_fp16, var_4102_cast_fp16, var_4104_cast_fp16))[name = string("op_4132_cast_fp16")]; + bool var_4134_interleave_0 = const()[name = string("op_4134_interleave_0"), val = bool(false)]; + tensor var_4134_cast_fp16 = concat(axis = var_3509, interleave = var_4134_interleave_0, values = (var_4106_cast_fp16, var_4108_cast_fp16, var_4110_cast_fp16, var_4112_cast_fp16))[name = string("op_4134_cast_fp16")]; + bool var_4136_interleave_0 = const()[name = string("op_4136_interleave_0"), val = bool(false)]; + tensor var_4136_cast_fp16 = concat(axis = var_3509, interleave = var_4136_interleave_0, values = (var_4114_cast_fp16, var_4116_cast_fp16, var_4118_cast_fp16, var_4120_cast_fp16))[name = string("op_4136_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_3522, interleave = input_41_interleave_0, values = (var_4122_cast_fp16, var_4124_cast_fp16, var_4126_cast_fp16, var_4128_cast_fp16, var_4130_cast_fp16, var_4132_cast_fp16, var_4134_cast_fp16, var_4136_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36460160)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36984512)))]; + tensor obj_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_4155_to_fp16 = const()[name = string("op_4155_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_4155_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36985600)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36986688)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36987776)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39084992)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39089152)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41186368)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_4193_to_fp16 = const()[name = string("op_4193_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_4193_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41187456)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41188544)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_4217_pad_type_0 = const()[name = string("op_4217_pad_type_0"), val = string("valid")]; + tensor var_4217_strides_0 = const()[name = string("op_4217_strides_0"), val = tensor([1, 1])]; + tensor var_4217_pad_0 = const()[name = string("op_4217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4217_dilations_0 = const()[name = string("op_4217_dilations_0"), val = tensor([1, 1])]; + int32 var_4217_groups_0 = const()[name = string("op_4217_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41189632)))]; + tensor var_4217_cast_fp16 = conv(dilations = var_4217_dilations_0, groups = var_4217_groups_0, pad = var_4217_pad_0, pad_type = var_4217_pad_type_0, strides = var_4217_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4217_cast_fp16")]; + string var_4224_pad_type_0 = const()[name = string("op_4224_pad_type_0"), val = string("valid")]; + tensor var_4224_strides_0 = const()[name = string("op_4224_strides_0"), val = tensor([1, 1])]; + tensor var_4224_pad_0 = const()[name = string("op_4224_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4224_dilations_0 = const()[name = string("op_4224_dilations_0"), val = tensor([1, 1])]; + int32 var_4224_groups_0 = const()[name = string("op_4224_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41713984)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42238336)))]; + tensor var_4224_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_4224_dilations_0, groups = var_4224_groups_0, pad = var_4224_pad_0, pad_type = var_4224_pad_type_0, strides = var_4224_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4224_cast_fp16")]; + string var_4242_pad_type_0 = const()[name = string("op_4242_pad_type_0"), val = string("valid")]; + tensor var_4242_strides_0 = const()[name = string("op_4242_strides_0"), val = tensor([1, 1])]; + tensor var_4242_pad_0 = const()[name = string("op_4242_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4242_dilations_0 = const()[name = string("op_4242_dilations_0"), val = tensor([1, 1])]; + int32 var_4242_groups_0 = const()[name = string("op_4242_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42239424)))]; + tensor var_4242_cast_fp16 = conv(dilations = var_4242_dilations_0, groups = var_4242_groups_0, pad = var_4242_pad_0, pad_type = var_4242_pad_type_0, strides = var_4242_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4242_cast_fp16")]; + string var_4249_pad_type_0 = const()[name = string("op_4249_pad_type_0"), val = string("valid")]; + tensor var_4249_strides_0 = const()[name = string("op_4249_strides_0"), val = tensor([1, 1])]; + tensor var_4249_pad_0 = const()[name = string("op_4249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4249_dilations_0 = const()[name = string("op_4249_dilations_0"), val = tensor([1, 1])]; + int32 var_4249_groups_0 = const()[name = string("op_4249_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42763776)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43288128)))]; + tensor var_4249_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_4249_dilations_0, groups = var_4249_groups_0, pad = var_4249_pad_0, pad_type = var_4249_pad_type_0, strides = var_4249_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4249_cast_fp16")]; + string var_4267_pad_type_0 = const()[name = string("op_4267_pad_type_0"), val = string("valid")]; + tensor var_4267_strides_0 = const()[name = string("op_4267_strides_0"), val = tensor([1, 1])]; + tensor var_4267_pad_0 = const()[name = string("op_4267_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4267_dilations_0 = const()[name = string("op_4267_dilations_0"), val = tensor([1, 1])]; + int32 var_4267_groups_0 = const()[name = string("op_4267_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43289216)))]; + tensor var_4267_cast_fp16 = conv(dilations = var_4267_dilations_0, groups = var_4267_groups_0, pad = var_4267_pad_0, pad_type = var_4267_pad_type_0, strides = var_4267_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4267_cast_fp16")]; + string var_4274_pad_type_0 = const()[name = string("op_4274_pad_type_0"), val = string("valid")]; + tensor var_4274_strides_0 = const()[name = string("op_4274_strides_0"), val = tensor([1, 1])]; + tensor var_4274_pad_0 = const()[name = string("op_4274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4274_dilations_0 = const()[name = string("op_4274_dilations_0"), val = tensor([1, 1])]; + int32 var_4274_groups_0 = const()[name = string("op_4274_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43813568)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44337920)))]; + tensor var_4274_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_4274_dilations_0, groups = var_4274_groups_0, pad = var_4274_pad_0, pad_type = var_4274_pad_type_0, strides = var_4274_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4274_cast_fp16")]; + string var_4292_pad_type_0 = const()[name = string("op_4292_pad_type_0"), val = string("valid")]; + tensor var_4292_strides_0 = const()[name = string("op_4292_strides_0"), val = tensor([1, 1])]; + tensor var_4292_pad_0 = const()[name = string("op_4292_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4292_dilations_0 = const()[name = string("op_4292_dilations_0"), val = tensor([1, 1])]; + int32 var_4292_groups_0 = const()[name = string("op_4292_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44339008)))]; + tensor var_4292_cast_fp16 = conv(dilations = var_4292_dilations_0, groups = var_4292_groups_0, pad = var_4292_pad_0, pad_type = var_4292_pad_type_0, strides = var_4292_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4292_cast_fp16")]; + string var_4299_pad_type_0 = const()[name = string("op_4299_pad_type_0"), val = string("valid")]; + tensor var_4299_strides_0 = const()[name = string("op_4299_strides_0"), val = tensor([1, 1])]; + tensor var_4299_pad_0 = const()[name = string("op_4299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4299_dilations_0 = const()[name = string("op_4299_dilations_0"), val = tensor([1, 1])]; + int32 var_4299_groups_0 = const()[name = string("op_4299_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44863360)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45387712)))]; + tensor var_4299_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = var_4299_dilations_0, groups = var_4299_groups_0, pad = var_4299_pad_0, pad_type = var_4299_pad_type_0, strides = var_4299_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4299_cast_fp16")]; + string var_4317_pad_type_0 = const()[name = string("op_4317_pad_type_0"), val = string("valid")]; + tensor var_4317_strides_0 = const()[name = string("op_4317_strides_0"), val = tensor([1, 1])]; + tensor var_4317_pad_0 = const()[name = string("op_4317_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4317_dilations_0 = const()[name = string("op_4317_dilations_0"), val = tensor([1, 1])]; + int32 var_4317_groups_0 = const()[name = string("op_4317_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45388800)))]; + tensor var_4317_cast_fp16 = conv(dilations = var_4317_dilations_0, groups = var_4317_groups_0, pad = var_4317_pad_0, pad_type = var_4317_pad_type_0, strides = var_4317_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4317_cast_fp16")]; + string var_4324_pad_type_0 = const()[name = string("op_4324_pad_type_0"), val = string("valid")]; + tensor var_4324_strides_0 = const()[name = string("op_4324_strides_0"), val = tensor([1, 1])]; + tensor var_4324_pad_0 = const()[name = string("op_4324_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4324_dilations_0 = const()[name = string("op_4324_dilations_0"), val = tensor([1, 1])]; + int32 var_4324_groups_0 = const()[name = string("op_4324_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45913152)))]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46437504)))]; + tensor var_4324_cast_fp16 = conv(bias = decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16, dilations = var_4324_dilations_0, groups = var_4324_groups_0, pad = var_4324_pad_0, pad_type = var_4324_pad_type_0, strides = var_4324_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4324_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46438592)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46962944)))]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47487296)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_4354 = const()[name = string("op_4354"), val = int32(0)]; + bool input_51_interleave_0 = const()[name = string("input_51_interleave_0"), val = bool(false)]; + tensor input_51_cast_fp16 = concat(axis = var_4354, interleave = input_51_interleave_0, values = (var_4217_cast_fp16, var_4242_cast_fp16, var_4267_cast_fp16, var_4292_cast_fp16, var_4317_cast_fp16, k_cast_fp16))[name = string("input_51_cast_fp16")]; + int32 var_4357 = const()[name = string("op_4357"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_4357, interleave = input_interleave_0, values = (var_4224_cast_fp16, var_4249_cast_fp16, var_4274_cast_fp16, var_4299_cast_fp16, var_4324_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_4364_pad_0 = const()[name = string("op_4364_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4364_mode_0 = const()[name = string("op_4364_mode_0"), val = string("constant")]; + fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_7_to_fp16, mode = var_4364_mode_0, pad = var_4364_pad_0, x = input_51_cast_fp16)[name = string("op_4364_cast_fp16")]; + tensor var_4370_pad_0 = const()[name = string("op_4370_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4370_mode_0 = const()[name = string("op_4370_mode_0"), val = string("constant")]; + fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_8_to_fp16, mode = var_4370_mode_0, pad = var_4370_pad_0, x = input_cast_fp16)[name = string("op_4370_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-base/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-base/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..a4a151dbf644bd4496bbd2814a99308d1d5a8fe2 --- /dev/null +++ b/openai_whisper-base/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8af1204cd8809eab4254e1bd92d2df6204ecfed529557e5d959de09196661d7e +size 47488384 diff --git a/openai_whisper-base/LICENSE_NOTICE.txt b/openai_whisper-base/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-base/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-base/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-base/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a11fbb2cd75b96eb2120a672afefa298c2ef857b --- /dev/null +++ b/openai_whisper-base/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc05e563ee0c556e3f578e04be5fb67b4e7520124403f2561f39102f0f2b33d +size 243 diff --git a/openai_whisper-base/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-base/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a3544b6644c1af93ca6bdabb67a1c51e80eaa552 --- /dev/null +++ b/openai_whisper-base/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ef11ea703011eab03287ec661f999e19c2c78cf67d531b5e6afa02e18f913d +size 328 diff --git a/openai_whisper-base/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-base/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..14c5b740c647e540d629abcaf72ee1cfddce2c7c --- /dev/null +++ b/openai_whisper-base/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.version" : "8.0", + "com.github.apple.coremltools.source" : "torch==2.5.1" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-base/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-base/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cf4cd446f68b88655d00a7df7063aa46937a9bdd --- /dev/null +++ b/openai_whisper-base/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-base/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-base/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f7d28ffac464e9e7086a526930f0059187de8d01 --- /dev/null +++ b/openai_whisper-base/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83 +size 354080 diff --git a/openai_whisper-base/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-base/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..bb0c8291d04f62c72ac417ea75132e9ba81ead64 --- /dev/null +++ b/openai_whisper-base/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1142d9c54d1ed6e913db8c2a3c0fb5bd7eff60848de7c69d2c62e861f0a8217 +size 243 diff --git a/openai_whisper-base/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-base/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a5ec536bb16cfb96f3663270f59dd1341a78145c --- /dev/null +++ b/openai_whisper-base/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6187c26f9b51fb464b068dfc833cb82d691054bffbfdfe534340a4e31e9ab919 +size 754 diff --git a/openai_whisper-base/TextDecoder.mlmodelc/metadata.json b/openai_whisper-base/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1deeccf581f79b7c26cf4ad8440146eddd38db7c --- /dev/null +++ b/openai_whisper-base/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51865)", + "shortDescription" : "", + "shape" : "[1, 1, 51865]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 3072 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 3072, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 3072 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 3072, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 12, + "Ios18.mul" : 24, + "Ios18.matmul" : 24, + "Ios18.batchNorm" : 19, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 43, + "Ios18.layerNorm" : 19, + "Ios18.reshape" : 48, + "Ios18.linear" : 1, + "Ios18.conv" : 48, + "Ios18.gelu" : 6, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 28, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 448)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 6 × 512 × 1 × 448)", + "shortDescription" : "", + "shape" : "[6, 512, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-base/TextDecoder.mlmodelc/model.mil b/openai_whisper-base/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..32488499c80679eeac89041cbc278add73db1544 --- /dev/null +++ b/openai_whisper-base/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,973 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; + int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; + bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_30_cast_fp16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = input_ids, validate_indices = var_30_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_30_cast_fp16")]; + int32 var_34_axis_0 = const()[name = string("op_34_axis_0"), val = int32(0)]; + int32 var_34_batch_dims_0 = const()[name = string("op_34_batch_dims_0"), val = int32(0)]; + bool var_34_validate_indices_0 = const()[name = string("op_34_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53109888)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_99")]; + tensor var_34_cast_fp16_cast_uint16 = gather(axis = var_34_axis_0, batch_dims = var_34_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_34_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_34_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_30_cast_fp16, y = var_34_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_48_axes_0 = const()[name = string("op_48_axes_0"), val = tensor([2])]; + tensor var_48_cast_fp16 = expand_dims(axes = var_48_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_48_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_48_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1, 1, 1])]; + int32 var_53_axis_0 = const()[name = string("op_53_axis_0"), val = int32(0)]; + tensor var_53_cast_fp16_0, tensor var_53_cast_fp16_1, tensor var_53_cast_fp16_2, tensor var_53_cast_fp16_3, tensor var_53_cast_fp16_4, tensor var_53_cast_fp16_5 = split(axis = var_53_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_53_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1, 1, 1])]; + int32 var_62_axis_0 = const()[name = string("op_62_axis_0"), val = int32(0)]; + tensor var_62_cast_fp16_0, tensor var_62_cast_fp16_1, tensor var_62_cast_fp16_2, tensor var_62_cast_fp16_3, tensor var_62_cast_fp16_4, tensor var_62_cast_fp16_5 = split(axis = var_62_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_62_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 512, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 512, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_84 = const()[name = string("op_84"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_109_to_fp16 = const()[name = string("op_109_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_109_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53568704)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53569792)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53570880)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53571968)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53573056)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54097408)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54098496)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54622848)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55147200)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_144_axes_0 = const()[name = string("op_144_axes_0"), val = tensor([1])]; + tensor var_144_cast_fp16 = expand_dims(axes = var_144_axes_0, x = kv_cache_update_mask)[name = string("op_144_cast_fp16")]; + tensor var_145_axes_0 = const()[name = string("op_145_axes_0"), val = tensor([2])]; + tensor var_145_cast_fp16 = expand_dims(axes = var_145_axes_0, x = var_144_cast_fp16)[name = string("op_145_cast_fp16")]; + tensor var_147_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_145_cast_fp16)[name = string("op_147_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_53_cast_fp16_0, y = var_147_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_149_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_145_cast_fp16)[name = string("op_149_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_62_cast_fp16_0, y = var_149_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_152 = const()[name = string("op_152"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_152, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_154_to_fp16 = const()[name = string("op_154_to_fp16"), val = fp16(0x1p-3)]; + tensor var_155_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_154_to_fp16)[name = string("op_155_cast_fp16")]; + tensor var_156 = const()[name = string("op_156"), val = tensor([1, 8, 64, -1])]; + tensor var_157_cast_fp16 = reshape(shape = var_156, x = key_1_cast_fp16)[name = string("op_157_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_155_cast_fp16, y = var_157_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_161_axes_0 = const()[name = string("op_161_axes_0"), val = tensor([1])]; + tensor var_161_cast_fp16 = expand_dims(axes = var_161_axes_0, x = decoder_key_padding_mask)[name = string("op_161_cast_fp16")]; + tensor var_162_axes_0 = const()[name = string("op_162_axes_0"), val = tensor([2])]; + tensor var_162_cast_fp16 = expand_dims(axes = var_162_axes_0, x = var_161_cast_fp16)[name = string("op_162_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_165_cast_fp16 = softmax(axis = var_84, x = mh_w_3_cast_fp16)[name = string("op_165_cast_fp16")]; + tensor var_166 = const()[name = string("op_166"), val = tensor([1, 8, 64, -1])]; + tensor var_167_cast_fp16 = reshape(shape = var_166, x = value_1_cast_fp16)[name = string("op_167_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_167_cast_fp16, y = var_165_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_170 = const()[name = string("op_170"), val = tensor([1, 512, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_170, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55148288)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55672640)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_192_to_fp16 = const()[name = string("op_192_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_192_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55673728)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55674816)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55675904)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56200256)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_212 = const()[name = string("op_212"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_212, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_214_to_fp16 = const()[name = string("op_214_to_fp16"), val = fp16(0x1p-3)]; + tensor var_215_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_214_to_fp16)[name = string("op_215_cast_fp16")]; + tensor var_216 = const()[name = string("op_216"), val = tensor([1, 8, 64, -1])]; + tensor var_217_cast_fp16 = reshape(shape = var_216, x = obj_17_cast_fp16)[name = string("op_217_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_215_cast_fp16, y = var_217_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_221_axes_0 = const()[name = string("op_221_axes_0"), val = tensor([1])]; + tensor var_221_cast_fp16 = expand_dims(axes = var_221_axes_0, x = read_state_4)[name = string("op_221_cast_fp16")]; + tensor var_222_axes_0 = const()[name = string("op_222_axes_0"), val = tensor([2])]; + tensor var_222_cast_fp16 = expand_dims(axes = var_222_axes_0, x = var_221_cast_fp16)[name = string("op_222_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_84, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_226 = const()[name = string("op_226"), val = tensor([1, 8, 64, -1])]; + tensor var_227_cast_fp16 = reshape(shape = var_226, x = obj_19_cast_fp16)[name = string("op_227_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_227_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_230 = const()[name = string("op_230"), val = tensor([1, 512, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_230, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56201344)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56725696)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_248_to_fp16 = const()[name = string("op_248_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_248_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56726784)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56727872)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56728960)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58826176)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58830336)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60927552)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 512, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 512, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_293 = const()[name = string("op_293"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_318_to_fp16 = const()[name = string("op_318_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_318_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60928640)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60929728)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60930816)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61455168)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61456256)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(61980608)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62504960)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_356_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_145_cast_fp16)[name = string("op_356_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_53_cast_fp16_1, y = var_356_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_358_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_145_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_62_cast_fp16_1, y = var_358_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_361 = const()[name = string("op_361"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_361, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_363_to_fp16 = const()[name = string("op_363_to_fp16"), val = fp16(0x1p-3)]; + tensor var_364_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_363_to_fp16)[name = string("op_364_cast_fp16")]; + tensor var_365 = const()[name = string("op_365"), val = tensor([1, 8, 64, -1])]; + tensor var_366_cast_fp16 = reshape(shape = var_365, x = key_3_cast_fp16)[name = string("op_366_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_364_cast_fp16, y = var_366_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_374_cast_fp16 = softmax(axis = var_293, x = mh_w_11_cast_fp16)[name = string("op_374_cast_fp16")]; + tensor var_375 = const()[name = string("op_375"), val = tensor([1, 8, 64, -1])]; + tensor var_376_cast_fp16 = reshape(shape = var_375, x = value_3_cast_fp16)[name = string("op_376_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_376_cast_fp16, y = var_374_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_379 = const()[name = string("op_379"), val = tensor([1, 512, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_379, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62506048)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63030400)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_401_to_fp16 = const()[name = string("op_401_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_401_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63031488)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63032576)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63033664)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63558016)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_421 = const()[name = string("op_421"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_421, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_423_to_fp16 = const()[name = string("op_423_to_fp16"), val = fp16(0x1p-3)]; + tensor var_424_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_423_to_fp16)[name = string("op_424_cast_fp16")]; + tensor var_425 = const()[name = string("op_425"), val = tensor([1, 8, 64, -1])]; + tensor var_426_cast_fp16 = reshape(shape = var_425, x = obj_35_cast_fp16)[name = string("op_426_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_424_cast_fp16, y = var_426_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_293, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_435 = const()[name = string("op_435"), val = tensor([1, 8, 64, -1])]; + tensor var_436_cast_fp16 = reshape(shape = var_435, x = obj_37_cast_fp16)[name = string("op_436_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_436_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_439 = const()[name = string("op_439"), val = tensor([1, 512, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_439, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63559104)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64083456)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_457_to_fp16 = const()[name = string("op_457_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_457_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64084544)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64085632)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64086720)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66183936)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66188096)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68285312)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 512, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 512, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_502 = const()[name = string("op_502"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_527_to_fp16 = const()[name = string("op_527_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_527_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68286400)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68287488)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68288576)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68812928)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68814016)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69338368)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69862720)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_565_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_145_cast_fp16)[name = string("op_565_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_53_cast_fp16_2, y = var_565_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_567_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_145_cast_fp16)[name = string("op_567_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_62_cast_fp16_2, y = var_567_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_570 = const()[name = string("op_570"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_570, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_572_to_fp16 = const()[name = string("op_572_to_fp16"), val = fp16(0x1p-3)]; + tensor var_573_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_572_to_fp16)[name = string("op_573_cast_fp16")]; + tensor var_574 = const()[name = string("op_574"), val = tensor([1, 8, 64, -1])]; + tensor var_575_cast_fp16 = reshape(shape = var_574, x = key_5_cast_fp16)[name = string("op_575_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_573_cast_fp16, y = var_575_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_583_cast_fp16 = softmax(axis = var_502, x = mh_w_19_cast_fp16)[name = string("op_583_cast_fp16")]; + tensor var_584 = const()[name = string("op_584"), val = tensor([1, 8, 64, -1])]; + tensor var_585_cast_fp16 = reshape(shape = var_584, x = value_5_cast_fp16)[name = string("op_585_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_585_cast_fp16, y = var_583_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_588 = const()[name = string("op_588"), val = tensor([1, 512, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_588, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69863808)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70388160)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_610_to_fp16 = const()[name = string("op_610_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_610_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70389248)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70390336)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70391424)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70915776)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_630 = const()[name = string("op_630"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_630, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_632_to_fp16 = const()[name = string("op_632_to_fp16"), val = fp16(0x1p-3)]; + tensor var_633_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_632_to_fp16)[name = string("op_633_cast_fp16")]; + tensor var_634 = const()[name = string("op_634"), val = tensor([1, 8, 64, -1])]; + tensor var_635_cast_fp16 = reshape(shape = var_634, x = obj_53_cast_fp16)[name = string("op_635_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_633_cast_fp16, y = var_635_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_502, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_644 = const()[name = string("op_644"), val = tensor([1, 8, 64, -1])]; + tensor var_645_cast_fp16 = reshape(shape = var_644, x = obj_55_cast_fp16)[name = string("op_645_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_645_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_648 = const()[name = string("op_648"), val = tensor([1, 512, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_648, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70916864)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71441216)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_666_to_fp16 = const()[name = string("op_666_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_666_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71442304)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71443392)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71444480)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73541696)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73545856)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75643072)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 512, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 512, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_711 = const()[name = string("op_711"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_736_to_fp16 = const()[name = string("op_736_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_736_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75644160)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75645248)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75646336)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76170688)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_7_pad_type_0 = const()[name = string("current_key_7_pad_type_0"), val = string("valid")]; + tensor current_key_7_strides_0 = const()[name = string("current_key_7_strides_0"), val = tensor([1, 1])]; + tensor current_key_7_pad_0 = const()[name = string("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_7_dilations_0 = const()[name = string("current_key_7_dilations_0"), val = tensor([1, 1])]; + int32 current_key_7_groups_0 = const()[name = string("current_key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76171776)))]; + tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_7_cast_fp16")]; + string current_value_7_pad_type_0 = const()[name = string("current_value_7_pad_type_0"), val = string("valid")]; + tensor current_value_7_strides_0 = const()[name = string("current_value_7_strides_0"), val = tensor([1, 1])]; + tensor current_value_7_pad_0 = const()[name = string("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_7_dilations_0 = const()[name = string("current_value_7_dilations_0"), val = tensor([1, 1])]; + int32 current_value_7_groups_0 = const()[name = string("current_value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76696128)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77220480)))]; + tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_7_cast_fp16")]; + tensor var_774_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_145_cast_fp16)[name = string("op_774_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = var_53_cast_fp16_3, y = var_774_cast_fp16)[name = string("key_7_cast_fp16")]; + tensor var_776_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_145_cast_fp16)[name = string("op_776_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = var_62_cast_fp16_3, y = var_776_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_779 = const()[name = string("op_779"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_779, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_781_to_fp16 = const()[name = string("op_781_to_fp16"), val = fp16(0x1p-3)]; + tensor var_782_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_781_to_fp16)[name = string("op_782_cast_fp16")]; + tensor var_783 = const()[name = string("op_783"), val = tensor([1, 8, 64, -1])]; + tensor var_784_cast_fp16 = reshape(shape = var_783, x = key_7_cast_fp16)[name = string("op_784_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_782_cast_fp16, y = var_784_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_792_cast_fp16 = softmax(axis = var_711, x = mh_w_27_cast_fp16)[name = string("op_792_cast_fp16")]; + tensor var_793 = const()[name = string("op_793"), val = tensor([1, 8, 64, -1])]; + tensor var_794_cast_fp16 = reshape(shape = var_793, x = value_7_cast_fp16)[name = string("op_794_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_794_cast_fp16, y = var_792_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_797 = const()[name = string("op_797"), val = tensor([1, 512, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_797, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77221568)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77745920)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_819_to_fp16 = const()[name = string("op_819_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_819_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77747008)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77748096)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77749184)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78273536)))]; + tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_15_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_839, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; + fp16 var_841_to_fp16 = const()[name = string("op_841_to_fp16"), val = fp16(0x1p-3)]; + tensor var_842_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_841_to_fp16)[name = string("op_842_cast_fp16")]; + tensor var_843 = const()[name = string("op_843"), val = tensor([1, 8, 64, -1])]; + tensor var_844_cast_fp16 = reshape(shape = var_843, x = obj_71_cast_fp16)[name = string("op_844_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_842_cast_fp16, y = var_844_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_31_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_711, x = mh_w_31_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_853 = const()[name = string("op_853"), val = tensor([1, 8, 64, -1])]; + tensor var_854_cast_fp16 = reshape(shape = var_853, x = obj_73_cast_fp16)[name = string("op_854_cast_fp16")]; + bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; + bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_854_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_15_cast_fp16")]; + tensor var_857 = const()[name = string("op_857"), val = tensor([1, 512, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_857, x = attn_15_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78274624)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78798976)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_878_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78800064)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78801152)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78802240)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80899456)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80903616)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83000832)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_25_cast_fp16")]; + tensor obj_89_begin_0 = const()[name = string("obj_89_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_89_end_0 = const()[name = string("obj_89_end_0"), val = tensor([5, 512, 1, 1536])]; + tensor obj_89_end_mask_0 = const()[name = string("obj_89_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_89_cast_fp16 = slice_by_index(begin = obj_89_begin_0, end = obj_89_end_0, end_mask = obj_89_end_mask_0, x = read_state_2)[name = string("obj_89_cast_fp16")]; + tensor obj_91_begin_0 = const()[name = string("obj_91_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_91_end_0 = const()[name = string("obj_91_end_0"), val = tensor([5, 512, 1, 1536])]; + tensor obj_91_end_mask_0 = const()[name = string("obj_91_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_91_cast_fp16 = slice_by_index(begin = obj_91_begin_0, end = obj_91_end_0, end_mask = obj_91_end_mask_0, x = read_state_3)[name = string("obj_91_cast_fp16")]; + int32 var_924 = const()[name = string("op_924"), val = int32(3)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_949_to_fp16 = const()[name = string("op_949_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_949_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83001920)))]; + tensor obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83003008)))]; + fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_79_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83004096)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83528448)))]; + tensor query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("query_17_cast_fp16")]; + string current_key_9_pad_type_0 = const()[name = string("current_key_9_pad_type_0"), val = string("valid")]; + tensor current_key_9_strides_0 = const()[name = string("current_key_9_strides_0"), val = tensor([1, 1])]; + tensor current_key_9_pad_0 = const()[name = string("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_9_dilations_0 = const()[name = string("current_key_9_dilations_0"), val = tensor([1, 1])]; + int32 current_key_9_groups_0 = const()[name = string("current_key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83529536)))]; + tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_key_9_cast_fp16")]; + string current_value_9_pad_type_0 = const()[name = string("current_value_9_pad_type_0"), val = string("valid")]; + tensor current_value_9_strides_0 = const()[name = string("current_value_9_strides_0"), val = tensor([1, 1])]; + tensor current_value_9_pad_0 = const()[name = string("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_9_dilations_0 = const()[name = string("current_value_9_dilations_0"), val = tensor([1, 1])]; + int32 current_value_9_groups_0 = const()[name = string("current_value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84053888)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84578240)))]; + tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_value_9_cast_fp16")]; + tensor var_987_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_145_cast_fp16)[name = string("op_987_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = var_53_cast_fp16_4, y = var_987_cast_fp16)[name = string("key_9_cast_fp16")]; + tensor var_989_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_145_cast_fp16)[name = string("op_989_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = var_62_cast_fp16_4, y = var_989_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_992 = const()[name = string("op_992"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_992, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; + fp16 var_994_to_fp16 = const()[name = string("op_994_to_fp16"), val = fp16(0x1p-3)]; + tensor var_995_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_994_to_fp16)[name = string("op_995_cast_fp16")]; + tensor var_996 = const()[name = string("op_996"), val = tensor([1, 8, 64, -1])]; + tensor var_997_cast_fp16 = reshape(shape = var_996, x = key_9_cast_fp16)[name = string("op_997_cast_fp16")]; + bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; + bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_995_cast_fp16, y = var_997_cast_fp16)[name = string("mh_w_33_cast_fp16")]; + tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_35_cast_fp16")]; + tensor var_1005_cast_fp16 = softmax(axis = var_924, x = mh_w_35_cast_fp16)[name = string("op_1005_cast_fp16")]; + tensor var_1006 = const()[name = string("op_1006"), val = tensor([1, 8, 64, -1])]; + tensor var_1007_cast_fp16 = reshape(shape = var_1006, x = value_9_cast_fp16)[name = string("op_1007_cast_fp16")]; + bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; + bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1007_cast_fp16, y = var_1005_cast_fp16)[name = string("attn_17_cast_fp16")]; + tensor var_1010 = const()[name = string("op_1010"), val = tensor([1, 512, 1, -1])]; + tensor input_41_cast_fp16 = reshape(shape = var_1010, x = attn_17_cast_fp16)[name = string("input_41_cast_fp16")]; + string obj_85_pad_type_0 = const()[name = string("obj_85_pad_type_0"), val = string("valid")]; + tensor obj_85_strides_0 = const()[name = string("obj_85_strides_0"), val = tensor([1, 1])]; + tensor obj_85_pad_0 = const()[name = string("obj_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_85_dilations_0 = const()[name = string("obj_85_dilations_0"), val = tensor([1, 1])]; + int32 obj_85_groups_0 = const()[name = string("obj_85_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84579328)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85103680)))]; + tensor obj_85_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_85_dilations_0, groups = obj_85_groups_0, pad = obj_85_pad_0, pad_type = obj_85_pad_type_0, strides = obj_85_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_85_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_85_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_1032_to_fp16 = const()[name = string("op_1032_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1032_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor obj_87_gamma_0_to_fp16 = const()[name = string("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85104768)))]; + tensor obj_87_beta_0_to_fp16 = const()[name = string("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85105856)))]; + fp16 obj_87_epsilon_0_to_fp16 = const()[name = string("obj_87_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("obj_87_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85106944)))]; + tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85631296)))]; + tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16, x = obj_87_cast_fp16)[name = string("query_19_cast_fp16")]; + tensor var_1052 = const()[name = string("op_1052"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_1052, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; + fp16 var_1054_to_fp16 = const()[name = string("op_1054_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1055_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1054_to_fp16)[name = string("op_1055_cast_fp16")]; + tensor var_1056 = const()[name = string("op_1056"), val = tensor([1, 8, 64, -1])]; + tensor var_1057_cast_fp16 = reshape(shape = var_1056, x = obj_89_cast_fp16)[name = string("op_1057_cast_fp16")]; + bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; + bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1055_cast_fp16, y = var_1057_cast_fp16)[name = string("mh_w_37_cast_fp16")]; + tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_39_cast_fp16")]; + tensor obj_95_cast_fp16 = softmax(axis = var_924, x = mh_w_39_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor var_1066 = const()[name = string("op_1066"), val = tensor([1, 8, 64, -1])]; + tensor var_1067_cast_fp16 = reshape(shape = var_1066, x = obj_91_cast_fp16)[name = string("op_1067_cast_fp16")]; + bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; + bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1067_cast_fp16, y = obj_95_cast_fp16)[name = string("attn_19_cast_fp16")]; + tensor var_1070 = const()[name = string("op_1070"), val = tensor([1, 512, 1, -1])]; + tensor input_43_cast_fp16 = reshape(shape = var_1070, x = attn_19_cast_fp16)[name = string("input_43_cast_fp16")]; + string obj_93_pad_type_0 = const()[name = string("obj_93_pad_type_0"), val = string("valid")]; + tensor obj_93_strides_0 = const()[name = string("obj_93_strides_0"), val = tensor([1, 1])]; + tensor obj_93_pad_0 = const()[name = string("obj_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_93_dilations_0 = const()[name = string("obj_93_dilations_0"), val = tensor([1, 1])]; + int32 obj_93_groups_0 = const()[name = string("obj_93_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85632384)))]; + tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86156736)))]; + tensor obj_93_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_93_dilations_0, groups = obj_93_groups_0, pad = obj_93_pad_0, pad_type = obj_93_pad_type_0, strides = obj_93_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("obj_93_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_93_cast_fp16)[name = string("inputs_29_cast_fp16")]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_1091_to_fp16 = const()[name = string("op_1091_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1091_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86157824)))]; + tensor input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86158912)))]; + fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")]; + tensor input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor([1, 1])]; + tensor input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor([1, 1])]; + int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86160000)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88257216)))]; + tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")]; + tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = string("input_49_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88261376)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90358592)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor obj_107_begin_0 = const()[name = string("obj_107_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_107_end_0 = const()[name = string("obj_107_end_0"), val = tensor([6, 512, 1, 1536])]; + tensor obj_107_end_mask_0 = const()[name = string("obj_107_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_107_cast_fp16 = slice_by_index(begin = obj_107_begin_0, end = obj_107_end_0, end_mask = obj_107_end_mask_0, x = read_state_2)[name = string("obj_107_cast_fp16")]; + tensor obj_109_begin_0 = const()[name = string("obj_109_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_109_end_0 = const()[name = string("obj_109_end_0"), val = tensor([6, 512, 1, 1536])]; + tensor obj_109_end_mask_0 = const()[name = string("obj_109_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_109_cast_fp16 = slice_by_index(begin = obj_109_begin_0, end = obj_109_end_0, end_mask = obj_109_end_mask_0, x = read_state_3)[name = string("obj_109_cast_fp16")]; + int32 var_1137 = const()[name = string("op_1137"), val = int32(3)]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_1162_to_fp16 = const()[name = string("op_1162_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1162_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90359680)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90360768)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90361856)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90886208)))]; + tensor query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_21_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90887296)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91411648)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91936000)))]; + tensor current_value_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_1200_cast_fp16 = mul(x = current_key_cast_fp16, y = var_145_cast_fp16)[name = string("op_1200_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_53_cast_fp16_5, y = var_1200_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_1202_cast_fp16 = mul(x = current_value_cast_fp16, y = var_145_cast_fp16)[name = string("op_1202_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_62_cast_fp16_5, y = var_1202_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_1205 = const()[name = string("op_1205"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_1205, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; + fp16 var_1207_to_fp16 = const()[name = string("op_1207_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1208_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1207_to_fp16)[name = string("op_1208_cast_fp16")]; + tensor var_1209 = const()[name = string("op_1209"), val = tensor([1, 8, 64, -1])]; + tensor var_1210_cast_fp16 = reshape(shape = var_1209, x = key_cast_fp16)[name = string("op_1210_cast_fp16")]; + bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; + bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1208_cast_fp16, y = var_1210_cast_fp16)[name = string("mh_w_41_cast_fp16")]; + tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = var_162_cast_fp16)[name = string("mh_w_43_cast_fp16")]; + tensor var_1218_cast_fp16 = softmax(axis = var_1137, x = mh_w_43_cast_fp16)[name = string("op_1218_cast_fp16")]; + tensor var_1219 = const()[name = string("op_1219"), val = tensor([1, 8, 64, -1])]; + tensor var_1220_cast_fp16 = reshape(shape = var_1219, x = value_cast_fp16)[name = string("op_1220_cast_fp16")]; + bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; + bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1220_cast_fp16, y = var_1218_cast_fp16)[name = string("attn_21_cast_fp16")]; + tensor var_1223 = const()[name = string("op_1223"), val = tensor([1, 512, 1, -1])]; + tensor input_51_cast_fp16 = reshape(shape = var_1223, x = attn_21_cast_fp16)[name = string("input_51_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91937088)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92461440)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_33_cast_fp16")]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_1245_to_fp16 = const()[name = string("op_1245_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1245_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92462528)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92463616)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92464704)))]; + tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92989056)))]; + tensor query_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_1265 = const()[name = string("op_1265"), val = tensor([1, 8, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_1265, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1268_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_1267_to_fp16)[name = string("op_1268_cast_fp16")]; + tensor var_1269 = const()[name = string("op_1269"), val = tensor([1, 8, 64, -1])]; + tensor var_1270_cast_fp16 = reshape(shape = var_1269, x = obj_107_cast_fp16)[name = string("op_1270_cast_fp16")]; + bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; + bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_1268_cast_fp16, y = var_1270_cast_fp16)[name = string("mh_w_45_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_45_cast_fp16, y = var_222_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_113_cast_fp16 = softmax(axis = var_1137, x = mh_w_cast_fp16)[name = string("obj_113_cast_fp16")]; + tensor var_1279 = const()[name = string("op_1279"), val = tensor([1, 8, 64, -1])]; + tensor var_1280_cast_fp16 = reshape(shape = var_1279, x = obj_109_cast_fp16)[name = string("op_1280_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_1280_cast_fp16, y = obj_113_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_1283 = const()[name = string("op_1283"), val = tensor([1, 512, 1, -1])]; + tensor input_53_cast_fp16 = reshape(shape = var_1283, x = attn_cast_fp16)[name = string("input_53_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92990144)))]; + tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93514496)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_1304_to_fp16 = const()[name = string("op_1304_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1304_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_55_gamma_0_to_fp16 = const()[name = string("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93515584)))]; + tensor input_55_beta_0_to_fp16 = const()[name = string("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93516672)))]; + fp16 input_55_epsilon_0_to_fp16 = const()[name = string("input_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_55_cast_fp16")]; + string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")]; + tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; + int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93517760)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95614976)))]; + tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_57_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95619136)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97716352)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_1347_to_fp16 = const()[name = string("op_1347_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_1347_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97717440)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97718528)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_1358_axes_0 = const()[name = string("op_1358_axes_0"), val = tensor([2])]; + tensor var_1358_cast_fp16 = squeeze(axes = var_1358_axes_0, x = hidden_states_cast_fp16)[name = string("op_1358_cast_fp16")]; + tensor var_1361_perm_0 = const()[name = string("op_1361_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97719616)))]; + tensor var_1361_cast_fp16 = transpose(perm = var_1361_perm_0, x = var_1358_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_1361_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_1365 = const()[name = string("op_1365"), val = int32(1)]; + bool obj_117_interleave_0 = const()[name = string("obj_117_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_1365, interleave = obj_117_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_cast_fp16))[name = string("obj_117_cast_fp16")]; + int32 var_1368 = const()[name = string("op_1368"), val = int32(1)]; + bool obj_119_interleave_0 = const()[name = string("obj_119_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_1368, interleave = obj_119_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_cast_fp16))[name = string("obj_119_cast_fp16")]; + tensor var_1379_begin_0 = const()[name = string("op_1379_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1379_end_0 = const()[name = string("op_1379_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_1379_end_mask_0 = const()[name = string("op_1379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1379_cast_fp16 = slice_by_index(begin = var_1379_begin_0, end = var_1379_end_0, end_mask = var_1379_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1379_cast_fp16")]; + tensor var_1382_begin_0 = const()[name = string("op_1382_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1382_end_0 = const()[name = string("op_1382_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1382_end_mask_0 = const()[name = string("op_1382_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1382_squeeze_mask_0 = const()[name = string("op_1382_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1382_cast_fp16 = slice_by_index(begin = var_1382_begin_0, end = var_1382_end_0, end_mask = var_1382_end_mask_0, squeeze_mask = var_1382_squeeze_mask_0, x = var_1379_cast_fp16)[name = string("op_1382_cast_fp16")]; + tensor var_1397_begin_0 = const()[name = string("op_1397_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1397_end_0 = const()[name = string("op_1397_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_1397_end_mask_0 = const()[name = string("op_1397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1397_cast_fp16 = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, x = obj_95_cast_fp16)[name = string("op_1397_cast_fp16")]; + tensor var_1400_begin_0 = const()[name = string("op_1400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1400_end_0 = const()[name = string("op_1400_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1400_end_mask_0 = const()[name = string("op_1400_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1400_squeeze_mask_0 = const()[name = string("op_1400_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1400_cast_fp16 = slice_by_index(begin = var_1400_begin_0, end = var_1400_end_0, end_mask = var_1400_end_mask_0, squeeze_mask = var_1400_squeeze_mask_0, x = var_1397_cast_fp16)[name = string("op_1400_cast_fp16")]; + tensor var_1415_begin_0 = const()[name = string("op_1415_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1415_end_0 = const()[name = string("op_1415_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_1415_end_mask_0 = const()[name = string("op_1415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1415_cast_fp16 = slice_by_index(begin = var_1415_begin_0, end = var_1415_end_0, end_mask = var_1415_end_mask_0, x = obj_95_cast_fp16)[name = string("op_1415_cast_fp16")]; + tensor var_1418_begin_0 = const()[name = string("op_1418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1418_end_0 = const()[name = string("op_1418_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1418_end_mask_0 = const()[name = string("op_1418_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1418_squeeze_mask_0 = const()[name = string("op_1418_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1418_cast_fp16 = slice_by_index(begin = var_1418_begin_0, end = var_1418_end_0, end_mask = var_1418_end_mask_0, squeeze_mask = var_1418_squeeze_mask_0, x = var_1415_cast_fp16)[name = string("op_1418_cast_fp16")]; + tensor var_1433_begin_0 = const()[name = string("op_1433_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1433_end_0 = const()[name = string("op_1433_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_1433_end_mask_0 = const()[name = string("op_1433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1433_cast_fp16 = slice_by_index(begin = var_1433_begin_0, end = var_1433_end_0, end_mask = var_1433_end_mask_0, x = obj_95_cast_fp16)[name = string("op_1433_cast_fp16")]; + tensor var_1436_begin_0 = const()[name = string("op_1436_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1436_end_0 = const()[name = string("op_1436_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1436_end_mask_0 = const()[name = string("op_1436_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1436_squeeze_mask_0 = const()[name = string("op_1436_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1436_cast_fp16 = slice_by_index(begin = var_1436_begin_0, end = var_1436_end_0, end_mask = var_1436_end_mask_0, squeeze_mask = var_1436_squeeze_mask_0, x = var_1433_cast_fp16)[name = string("op_1436_cast_fp16")]; + tensor var_1451_begin_0 = const()[name = string("op_1451_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1451_end_0 = const()[name = string("op_1451_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_1451_end_mask_0 = const()[name = string("op_1451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1451_cast_fp16 = slice_by_index(begin = var_1451_begin_0, end = var_1451_end_0, end_mask = var_1451_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1451_cast_fp16")]; + tensor var_1454_begin_0 = const()[name = string("op_1454_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1454_end_0 = const()[name = string("op_1454_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1454_end_mask_0 = const()[name = string("op_1454_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1454_squeeze_mask_0 = const()[name = string("op_1454_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1454_cast_fp16 = slice_by_index(begin = var_1454_begin_0, end = var_1454_end_0, end_mask = var_1454_end_mask_0, squeeze_mask = var_1454_squeeze_mask_0, x = var_1451_cast_fp16)[name = string("op_1454_cast_fp16")]; + tensor var_1469_begin_0 = const()[name = string("op_1469_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1469_end_0 = const()[name = string("op_1469_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_1469_end_mask_0 = const()[name = string("op_1469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1469_cast_fp16 = slice_by_index(begin = var_1469_begin_0, end = var_1469_end_0, end_mask = var_1469_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1469_cast_fp16")]; + tensor var_1472_begin_0 = const()[name = string("op_1472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1472_end_0 = const()[name = string("op_1472_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1472_end_mask_0 = const()[name = string("op_1472_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1472_squeeze_mask_0 = const()[name = string("op_1472_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1472_cast_fp16 = slice_by_index(begin = var_1472_begin_0, end = var_1472_end_0, end_mask = var_1472_end_mask_0, squeeze_mask = var_1472_squeeze_mask_0, x = var_1469_cast_fp16)[name = string("op_1472_cast_fp16")]; + tensor var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1487_cast_fp16")]; + tensor var_1490_begin_0 = const()[name = string("op_1490_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1490_end_0 = const()[name = string("op_1490_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1490_end_mask_0 = const()[name = string("op_1490_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1490_squeeze_mask_0 = const()[name = string("op_1490_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1490_cast_fp16 = slice_by_index(begin = var_1490_begin_0, end = var_1490_end_0, end_mask = var_1490_end_mask_0, squeeze_mask = var_1490_squeeze_mask_0, x = var_1487_cast_fp16)[name = string("op_1490_cast_fp16")]; + tensor var_1505_begin_0 = const()[name = string("op_1505_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1505_end_0 = const()[name = string("op_1505_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_1505_end_mask_0 = const()[name = string("op_1505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1505_cast_fp16 = slice_by_index(begin = var_1505_begin_0, end = var_1505_end_0, end_mask = var_1505_end_mask_0, x = obj_113_cast_fp16)[name = string("op_1505_cast_fp16")]; + tensor var_1508_begin_0 = const()[name = string("op_1508_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1508_end_0 = const()[name = string("op_1508_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1508_end_mask_0 = const()[name = string("op_1508_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1508_squeeze_mask_0 = const()[name = string("op_1508_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1508_cast_fp16 = slice_by_index(begin = var_1508_begin_0, end = var_1508_end_0, end_mask = var_1508_end_mask_0, squeeze_mask = var_1508_squeeze_mask_0, x = var_1505_cast_fp16)[name = string("op_1508_cast_fp16")]; + int32 var_1515 = const()[name = string("op_1515"), val = int32(1)]; + bool var_1516_interleave_0 = const()[name = string("op_1516_interleave_0"), val = bool(false)]; + tensor var_1516_cast_fp16 = concat(axis = var_1515, interleave = var_1516_interleave_0, values = (var_1382_cast_fp16, var_1400_cast_fp16, var_1418_cast_fp16, var_1436_cast_fp16, var_1454_cast_fp16, var_1472_cast_fp16, var_1490_cast_fp16, var_1508_cast_fp16))[name = string("op_1516_cast_fp16")]; + bool var_1519 = const()[name = string("op_1519"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1519, x = var_1516_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-base/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-base/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..92adad72fccbab4214f6bc256ddd9a157bfb0674 --- /dev/null +++ b/openai_whisper-base/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a8419ea11f839c055bdbbab4229b2d2fdf0f46761ad314e387a2060fd4a4dd5 +size 97823410 diff --git a/openai_whisper-base/config.json b/openai_whisper-base/config.json new file mode 100644 index 0000000000000000000000000000000000000000..61e1ae6f9129c8c6fda294e069baf047f7366e1b --- /dev/null +++ b/openai_whisper-base/config.json @@ -0,0 +1 @@ +{"_name_or_path": "openai/whisper-base", "activation_dropout": 0.0, "activation_function": "gelu", "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "d_model": 512, "decoder_attention_heads": 8, "decoder_ffn_dim": 2048, "decoder_layerdrop": 0.0, "decoder_layers": 6, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 8, "encoder_ffn_dim": 2048, "encoder_layerdrop": 0.0, "encoder_layers": 6, "eos_token_id": 50257, "forced_decoder_ids": [[1, 50259], [2, 50359], [3, 50363]], "init_std": 0.02, "is_encoder_decoder": true, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "model_type": "whisper", "num_hidden_layers": 6, "num_mel_bins": 80, "pad_token_id": 50257, "scale_embedding": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "torch_dtype": "float32", "transformers_version": "4.27.0.dev0", "use_cache": true, "vocab_size": 51865} \ No newline at end of file diff --git a/openai_whisper-base/generation_config.json b/openai_whisper-base/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..ce587327c4acabdbbb31865d28a3d79696608181 --- /dev/null +++ b/openai_whisper-base/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[3, 1], [4, 2], [4, 3], [4, 7], [5, 1], [5, 2], [5, 4], [5, 6]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50359]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50363, "pad_token_id": 50257, "prev_sot_token_id": 50361, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "task_to_id": {"transcribe": 50359, "translate": 50358}, "transformers_version": "4.31.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..d43516664e7fbe4cd031c8e46ae50db90cf498dc --- /dev/null +++ b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf5638f45b54e7967f4295530c5cbd0a4faa6362418c54394bdfb366c63c5889 +size 243 diff --git a/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..744160e29b93fbb6f726f655af6eec4eea765cb2 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d7522fa9a913828818c7bd24b63a02f7c9eed38197dd609b95a238dcdca40b8 +size 434 diff --git a/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..4ec9517ea8d152cf90dee830cd3805a1da4d64d0 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,90 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1280 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 1280, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.reshape" : 128, + "Ios18.batchNorm" : 65, + "Ios18.softmax" : 32, + "Pad" : 2, + "Ios18.concat" : 2, + "Ios18.gelu" : 34, + "Ios18.layerNorm" : 65, + "Ios18.matmul" : 64, + "Ios18.conv" : 202, + "Ios18.mul" : 32, + "Ios18.add" : 65 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/model.mil b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..df575772f293808374de1386112ade3bdf3c6d80 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,2767 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_114_pad_type_0 = const()[name = string("op_114_pad_type_0"), val = string("custom")]; + tensor var_114_pad_0 = const()[name = string("op_114_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_114_strides_0 = const()[name = string("op_114_strides_0"), val = tensor([1, 1])]; + tensor var_114_dilations_0 = const()[name = string("op_114_dilations_0"), val = tensor([1, 1])]; + int32 var_114_groups_0 = const()[name = string("op_114_groups_0"), val = int32(1)]; + tensor var_89_to_fp16 = const()[name = string("op_89_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_95_to_fp16 = const()[name = string("op_95_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(983168)))]; + tensor var_114_cast_fp16 = conv(bias = var_95_to_fp16, dilations = var_114_dilations_0, groups = var_114_groups_0, pad = var_114_pad_0, pad_type = var_114_pad_type_0, strides = var_114_strides_0, weight = var_89_to_fp16, x = melspectrogram_features)[name = string("op_114_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_114_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_154_pad_type_0 = const()[name = string("op_154_pad_type_0"), val = string("custom")]; + tensor var_154_pad_0 = const()[name = string("op_154_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_154_strides_0 = const()[name = string("op_154_strides_0"), val = tensor([2, 2])]; + tensor var_154_dilations_0 = const()[name = string("op_154_dilations_0"), val = tensor([1, 1])]; + int32 var_154_groups_0 = const()[name = string("op_154_groups_0"), val = int32(1)]; + tensor var_129_to_fp16 = const()[name = string("op_129_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985792)))]; + tensor var_135_to_fp16 = const()[name = string("op_135_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10816256)))]; + tensor var_154_cast_fp16 = conv(bias = var_135_to_fp16, dilations = var_154_dilations_0, groups = var_154_groups_0, pad = var_154_pad_0, pad_type = var_154_pad_type_0, strides = var_154_strides_0, weight = var_129_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_154_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_154_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_172_to_fp16 = const()[name = string("op_172_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_172_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_186 = const()[name = string("op_186"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_205_to_fp16 = const()[name = string("op_205_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_205_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14658944)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14661568)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14664192)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14666816)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14669440)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17946304)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17948928)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21225792)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24502656)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_240 = const()[name = string("op_240"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_240, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_242_to_fp16 = const()[name = string("op_242_to_fp16"), val = fp16(0x1p-3)]; + tensor var_243_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_242_to_fp16)[name = string("op_243_cast_fp16")]; + tensor var_244 = const()[name = string("op_244"), val = tensor([1, 20, 64, -1])]; + tensor var_245_cast_fp16 = reshape(shape = var_244, x = key_1_cast_fp16)[name = string("op_245_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_243_cast_fp16, y = var_245_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_248_cast_fp16 = softmax(axis = var_186, x = mh_w_1_cast_fp16)[name = string("op_248_cast_fp16")]; + tensor var_249 = const()[name = string("op_249"), val = tensor([1, 20, 64, -1])]; + tensor var_250_cast_fp16 = reshape(shape = var_249, x = value_1_cast_fp16)[name = string("op_250_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_250_cast_fp16, y = var_248_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_253 = const()[name = string("op_253"), val = tensor([1, 1280, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_253, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24505280)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27782144)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_271_to_fp16 = const()[name = string("op_271_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_271_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27784768)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27787392)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27790016)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40897280)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40907584)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54014848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_304 = const()[name = string("op_304"), val = int32(3)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_323_to_fp16 = const()[name = string("op_323_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_323_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54017472)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54020096)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54022720)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57299584)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57302208)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60579072)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63855936)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_358 = const()[name = string("op_358"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_358, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_360_to_fp16 = const()[name = string("op_360_to_fp16"), val = fp16(0x1p-3)]; + tensor var_361_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_360_to_fp16)[name = string("op_361_cast_fp16")]; + tensor var_362 = const()[name = string("op_362"), val = tensor([1, 20, 64, -1])]; + tensor var_363_cast_fp16 = reshape(shape = var_362, x = key_3_cast_fp16)[name = string("op_363_cast_fp16")]; + bool mh_w_3_transpose_x_0 = const()[name = string("mh_w_3_transpose_x_0"), val = bool(true)]; + bool mh_w_3_transpose_y_0 = const()[name = string("mh_w_3_transpose_y_0"), val = bool(false)]; + tensor mh_w_3_cast_fp16 = matmul(transpose_x = mh_w_3_transpose_x_0, transpose_y = mh_w_3_transpose_y_0, x = var_361_cast_fp16, y = var_363_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_366_cast_fp16 = softmax(axis = var_304, x = mh_w_3_cast_fp16)[name = string("op_366_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([1, 20, 64, -1])]; + tensor var_368_cast_fp16 = reshape(shape = var_367, x = value_3_cast_fp16)[name = string("op_368_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_368_cast_fp16, y = var_366_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_371 = const()[name = string("op_371"), val = tensor([1, 1280, 1, -1])]; + tensor input_9_cast_fp16 = reshape(shape = var_371, x = attn_3_cast_fp16)[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63858560)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67135424)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_389_to_fp16 = const()[name = string("op_389_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_389_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67138048)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67140672)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67143296)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80250560)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80260864)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93368128)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_422 = const()[name = string("op_422"), val = int32(3)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_441_to_fp16 = const()[name = string("op_441_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_441_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93370752)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93373376)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93376000)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96652864)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96655488)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99932352)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103209216)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_476 = const()[name = string("op_476"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_476, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_478_to_fp16 = const()[name = string("op_478_to_fp16"), val = fp16(0x1p-3)]; + tensor var_479_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_478_to_fp16)[name = string("op_479_cast_fp16")]; + tensor var_480 = const()[name = string("op_480"), val = tensor([1, 20, 64, -1])]; + tensor var_481_cast_fp16 = reshape(shape = var_480, x = key_5_cast_fp16)[name = string("op_481_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_479_cast_fp16, y = var_481_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor var_484_cast_fp16 = softmax(axis = var_422, x = mh_w_5_cast_fp16)[name = string("op_484_cast_fp16")]; + tensor var_485 = const()[name = string("op_485"), val = tensor([1, 20, 64, -1])]; + tensor var_486_cast_fp16 = reshape(shape = var_485, x = value_5_cast_fp16)[name = string("op_486_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_486_cast_fp16, y = var_484_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_489 = const()[name = string("op_489"), val = tensor([1, 1280, 1, -1])]; + tensor input_17_cast_fp16 = reshape(shape = var_489, x = attn_5_cast_fp16)[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103211840)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106488704)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_507_to_fp16 = const()[name = string("op_507_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_507_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106491328)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106493952)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106496576)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119603840)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119614144)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132721408)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_540 = const()[name = string("op_540"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_559_to_fp16 = const()[name = string("op_559_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_559_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132724032)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132726656)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132729280)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136006144)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136008768)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139285632)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142562496)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_594 = const()[name = string("op_594"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_594, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; + tensor var_597_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_596_to_fp16)[name = string("op_597_cast_fp16")]; + tensor var_598 = const()[name = string("op_598"), val = tensor([1, 20, 64, -1])]; + tensor var_599_cast_fp16 = reshape(shape = var_598, x = key_7_cast_fp16)[name = string("op_599_cast_fp16")]; + bool mh_w_7_transpose_x_0 = const()[name = string("mh_w_7_transpose_x_0"), val = bool(true)]; + bool mh_w_7_transpose_y_0 = const()[name = string("mh_w_7_transpose_y_0"), val = bool(false)]; + tensor mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_597_cast_fp16, y = var_599_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor var_602_cast_fp16 = softmax(axis = var_540, x = mh_w_7_cast_fp16)[name = string("op_602_cast_fp16")]; + tensor var_603 = const()[name = string("op_603"), val = tensor([1, 20, 64, -1])]; + tensor var_604_cast_fp16 = reshape(shape = var_603, x = value_7_cast_fp16)[name = string("op_604_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_604_cast_fp16, y = var_602_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_607 = const()[name = string("op_607"), val = tensor([1, 1280, 1, -1])]; + tensor input_25_cast_fp16 = reshape(shape = var_607, x = attn_7_cast_fp16)[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142565120)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145841984)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_625_to_fp16 = const()[name = string("op_625_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_625_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145844608)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145847232)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145849856)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158957120)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158967424)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172074688)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_658 = const()[name = string("op_658"), val = int32(3)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_677_to_fp16 = const()[name = string("op_677_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_677_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172077312)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172079936)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172082560)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175359424)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175362048)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178638912)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181915776)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_712 = const()[name = string("op_712"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_712, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_714_to_fp16 = const()[name = string("op_714_to_fp16"), val = fp16(0x1p-3)]; + tensor var_715_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_714_to_fp16)[name = string("op_715_cast_fp16")]; + tensor var_716 = const()[name = string("op_716"), val = tensor([1, 20, 64, -1])]; + tensor var_717_cast_fp16 = reshape(shape = var_716, x = key_9_cast_fp16)[name = string("op_717_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_715_cast_fp16, y = var_717_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor var_720_cast_fp16 = softmax(axis = var_658, x = mh_w_9_cast_fp16)[name = string("op_720_cast_fp16")]; + tensor var_721 = const()[name = string("op_721"), val = tensor([1, 20, 64, -1])]; + tensor var_722_cast_fp16 = reshape(shape = var_721, x = value_9_cast_fp16)[name = string("op_722_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_722_cast_fp16, y = var_720_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_725 = const()[name = string("op_725"), val = tensor([1, 1280, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_725, x = attn_9_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181918400)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185195264)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_743_to_fp16 = const()[name = string("op_743_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_743_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185197888)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185200512)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185203136)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198310400)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198320704)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211427968)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_776 = const()[name = string("op_776"), val = int32(3)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_795_to_fp16 = const()[name = string("op_795_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_795_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211430592)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211433216)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211435840)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214712704)))]; + tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; + string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; + tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; + tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; + int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214715328)))]; + tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; + string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; + tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; + tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; + int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217992192)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221269056)))]; + tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_830 = const()[name = string("op_830"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_830, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_832_to_fp16 = const()[name = string("op_832_to_fp16"), val = fp16(0x1p-3)]; + tensor var_833_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_832_to_fp16)[name = string("op_833_cast_fp16")]; + tensor var_834 = const()[name = string("op_834"), val = tensor([1, 20, 64, -1])]; + tensor var_835_cast_fp16 = reshape(shape = var_834, x = key_11_cast_fp16)[name = string("op_835_cast_fp16")]; + bool mh_w_11_transpose_x_0 = const()[name = string("mh_w_11_transpose_x_0"), val = bool(true)]; + bool mh_w_11_transpose_y_0 = const()[name = string("mh_w_11_transpose_y_0"), val = bool(false)]; + tensor mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_833_cast_fp16, y = var_835_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_838_cast_fp16 = softmax(axis = var_776, x = mh_w_11_cast_fp16)[name = string("op_838_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 20, 64, -1])]; + tensor var_840_cast_fp16 = reshape(shape = var_839, x = value_11_cast_fp16)[name = string("op_840_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_840_cast_fp16, y = var_838_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_843 = const()[name = string("op_843"), val = tensor([1, 1280, 1, -1])]; + tensor input_41_cast_fp16 = reshape(shape = var_843, x = attn_11_cast_fp16)[name = string("input_41_cast_fp16")]; + string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; + tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; + tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; + int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221271680)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224548544)))]; + tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_861_to_fp16 = const()[name = string("op_861_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_861_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224551168)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224553792)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224556416)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237663680)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237673984)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250781248)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; + int32 var_894 = const()[name = string("op_894"), val = int32(3)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_913_to_fp16 = const()[name = string("op_913_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_913_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250783872)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250786496)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250789120)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254065984)))]; + tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; + string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; + tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; + tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; + int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254068608)))]; + tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; + string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; + tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; + tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; + int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257345472)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260622336)))]; + tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_948 = const()[name = string("op_948"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_948, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_950_to_fp16 = const()[name = string("op_950_to_fp16"), val = fp16(0x1p-3)]; + tensor var_951_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_950_to_fp16)[name = string("op_951_cast_fp16")]; + tensor var_952 = const()[name = string("op_952"), val = tensor([1, 20, 64, -1])]; + tensor var_953_cast_fp16 = reshape(shape = var_952, x = key_13_cast_fp16)[name = string("op_953_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_951_cast_fp16, y = var_953_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor var_956_cast_fp16 = softmax(axis = var_894, x = mh_w_13_cast_fp16)[name = string("op_956_cast_fp16")]; + tensor var_957 = const()[name = string("op_957"), val = tensor([1, 20, 64, -1])]; + tensor var_958_cast_fp16 = reshape(shape = var_957, x = value_13_cast_fp16)[name = string("op_958_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_958_cast_fp16, y = var_956_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_961 = const()[name = string("op_961"), val = tensor([1, 1280, 1, -1])]; + tensor input_49_cast_fp16 = reshape(shape = var_961, x = attn_13_cast_fp16)[name = string("input_49_cast_fp16")]; + string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; + tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; + tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; + int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260624960)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263901824)))]; + tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_979_to_fp16 = const()[name = string("op_979_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_979_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263904448)))]; + tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263907072)))]; + fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; + string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; + tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; + tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; + int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263909696)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277016960)))]; + tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; + string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; + tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277027264)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290134528)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; + int32 var_1012 = const()[name = string("op_1012"), val = int32(3)]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_1031_to_fp16 = const()[name = string("op_1031_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1031_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290137152)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290139776)))]; + fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290142400)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293419264)))]; + tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; + string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; + tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; + tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; + int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293421888)))]; + tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; + string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; + tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; + tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; + int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296698752)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299975616)))]; + tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_1066 = const()[name = string("op_1066"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_1066, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; + fp16 var_1068_to_fp16 = const()[name = string("op_1068_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1069_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_1068_to_fp16)[name = string("op_1069_cast_fp16")]; + tensor var_1070 = const()[name = string("op_1070"), val = tensor([1, 20, 64, -1])]; + tensor var_1071_cast_fp16 = reshape(shape = var_1070, x = key_15_cast_fp16)[name = string("op_1071_cast_fp16")]; + bool mh_w_15_transpose_x_0 = const()[name = string("mh_w_15_transpose_x_0"), val = bool(true)]; + bool mh_w_15_transpose_y_0 = const()[name = string("mh_w_15_transpose_y_0"), val = bool(false)]; + tensor mh_w_15_cast_fp16 = matmul(transpose_x = mh_w_15_transpose_x_0, transpose_y = mh_w_15_transpose_y_0, x = var_1069_cast_fp16, y = var_1071_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor var_1074_cast_fp16 = softmax(axis = var_1012, x = mh_w_15_cast_fp16)[name = string("op_1074_cast_fp16")]; + tensor var_1075 = const()[name = string("op_1075"), val = tensor([1, 20, 64, -1])]; + tensor var_1076_cast_fp16 = reshape(shape = var_1075, x = value_15_cast_fp16)[name = string("op_1076_cast_fp16")]; + bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; + bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_1076_cast_fp16, y = var_1074_cast_fp16)[name = string("attn_15_cast_fp16")]; + tensor var_1079 = const()[name = string("op_1079"), val = tensor([1, 1280, 1, -1])]; + tensor input_57_cast_fp16 = reshape(shape = var_1079, x = attn_15_cast_fp16)[name = string("input_57_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299978240)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303255104)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_1097_to_fp16 = const()[name = string("op_1097_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1097_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303257728)))]; + tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303260352)))]; + fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303262976)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316370240)))]; + tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; + tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316380544)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329487808)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; + int32 var_1130 = const()[name = string("op_1130"), val = int32(3)]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_1149_to_fp16 = const()[name = string("op_1149_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1149_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329490432)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329493056)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329495680)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332772544)))]; + tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; + string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; + tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; + tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; + int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332775168)))]; + tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; + string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; + tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; + tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; + int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336052032)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339328896)))]; + tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_1184 = const()[name = string("op_1184"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_1184, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; + fp16 var_1186_to_fp16 = const()[name = string("op_1186_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1187_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1186_to_fp16)[name = string("op_1187_cast_fp16")]; + tensor var_1188 = const()[name = string("op_1188"), val = tensor([1, 20, 64, -1])]; + tensor var_1189_cast_fp16 = reshape(shape = var_1188, x = key_17_cast_fp16)[name = string("op_1189_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1187_cast_fp16, y = var_1189_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor var_1192_cast_fp16 = softmax(axis = var_1130, x = mh_w_17_cast_fp16)[name = string("op_1192_cast_fp16")]; + tensor var_1193 = const()[name = string("op_1193"), val = tensor([1, 20, 64, -1])]; + tensor var_1194_cast_fp16 = reshape(shape = var_1193, x = value_17_cast_fp16)[name = string("op_1194_cast_fp16")]; + bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; + bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1194_cast_fp16, y = var_1192_cast_fp16)[name = string("attn_17_cast_fp16")]; + tensor var_1197 = const()[name = string("op_1197"), val = tensor([1, 1280, 1, -1])]; + tensor input_65_cast_fp16 = reshape(shape = var_1197, x = attn_17_cast_fp16)[name = string("input_65_cast_fp16")]; + string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; + tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; + tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; + int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339331520)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342608384)))]; + tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_1215_to_fp16 = const()[name = string("op_1215_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1215_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342611008)))]; + tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342613632)))]; + fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; + tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; + tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; + int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342616256)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355723520)))]; + tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; + tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355733824)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368841088)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; + int32 var_1248 = const()[name = string("op_1248"), val = int32(3)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1267_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368843712)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368846336)))]; + fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368848960)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372125824)))]; + tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; + string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; + tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; + tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; + int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372128448)))]; + tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; + string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; + tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; + tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; + int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375405312)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378682176)))]; + tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_1302 = const()[name = string("op_1302"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_1302, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; + fp16 var_1304_to_fp16 = const()[name = string("op_1304_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1305_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1304_to_fp16)[name = string("op_1305_cast_fp16")]; + tensor var_1306 = const()[name = string("op_1306"), val = tensor([1, 20, 64, -1])]; + tensor var_1307_cast_fp16 = reshape(shape = var_1306, x = key_19_cast_fp16)[name = string("op_1307_cast_fp16")]; + bool mh_w_19_transpose_x_0 = const()[name = string("mh_w_19_transpose_x_0"), val = bool(true)]; + bool mh_w_19_transpose_y_0 = const()[name = string("mh_w_19_transpose_y_0"), val = bool(false)]; + tensor mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_1305_cast_fp16, y = var_1307_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_1310_cast_fp16 = softmax(axis = var_1248, x = mh_w_19_cast_fp16)[name = string("op_1310_cast_fp16")]; + tensor var_1311 = const()[name = string("op_1311"), val = tensor([1, 20, 64, -1])]; + tensor var_1312_cast_fp16 = reshape(shape = var_1311, x = value_19_cast_fp16)[name = string("op_1312_cast_fp16")]; + bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; + bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1312_cast_fp16, y = var_1310_cast_fp16)[name = string("attn_19_cast_fp16")]; + tensor var_1315 = const()[name = string("op_1315"), val = tensor([1, 1280, 1, -1])]; + tensor input_73_cast_fp16 = reshape(shape = var_1315, x = attn_19_cast_fp16)[name = string("input_73_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378684800)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381961664)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_1333_to_fp16 = const()[name = string("op_1333_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1333_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381964288)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381966912)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381969536)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395076800)))]; + tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395087104)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408194368)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; + int32 var_1366 = const()[name = string("op_1366"), val = int32(3)]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_1385_to_fp16 = const()[name = string("op_1385_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1385_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408196992)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408199616)))]; + fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408202240)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411479104)))]; + tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; + string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; + tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; + tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; + int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411481728)))]; + tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; + string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; + tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; + tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; + int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414758592)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418035456)))]; + tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_1420 = const()[name = string("op_1420"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_1420, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; + fp16 var_1422_to_fp16 = const()[name = string("op_1422_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1423_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1422_to_fp16)[name = string("op_1423_cast_fp16")]; + tensor var_1424 = const()[name = string("op_1424"), val = tensor([1, 20, 64, -1])]; + tensor var_1425_cast_fp16 = reshape(shape = var_1424, x = key_21_cast_fp16)[name = string("op_1425_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_1423_cast_fp16, y = var_1425_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor var_1428_cast_fp16 = softmax(axis = var_1366, x = mh_w_21_cast_fp16)[name = string("op_1428_cast_fp16")]; + tensor var_1429 = const()[name = string("op_1429"), val = tensor([1, 20, 64, -1])]; + tensor var_1430_cast_fp16 = reshape(shape = var_1429, x = value_21_cast_fp16)[name = string("op_1430_cast_fp16")]; + bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; + bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1430_cast_fp16, y = var_1428_cast_fp16)[name = string("attn_21_cast_fp16")]; + tensor var_1433 = const()[name = string("op_1433"), val = tensor([1, 1280, 1, -1])]; + tensor input_81_cast_fp16 = reshape(shape = var_1433, x = attn_21_cast_fp16)[name = string("input_81_cast_fp16")]; + string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; + tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; + tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; + int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418038080)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421314944)))]; + tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_1451_to_fp16 = const()[name = string("op_1451_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1451_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421317568)))]; + tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421320192)))]; + fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; + string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; + tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; + tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; + int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421322816)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434430080)))]; + tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; + tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434440384)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447547648)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; + int32 var_1484 = const()[name = string("op_1484"), val = int32(3)]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_1503_to_fp16 = const()[name = string("op_1503_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1503_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447550272)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447552896)))]; + fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; + string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; + tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; + tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; + int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447555520)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450832384)))]; + tensor query_23_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_23_cast_fp16")]; + string key_23_pad_type_0 = const()[name = string("key_23_pad_type_0"), val = string("valid")]; + tensor key_23_strides_0 = const()[name = string("key_23_strides_0"), val = tensor([1, 1])]; + tensor key_23_pad_0 = const()[name = string("key_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_23_dilations_0 = const()[name = string("key_23_dilations_0"), val = tensor([1, 1])]; + int32 key_23_groups_0 = const()[name = string("key_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450835008)))]; + tensor key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_23_cast_fp16")]; + string value_23_pad_type_0 = const()[name = string("value_23_pad_type_0"), val = string("valid")]; + tensor value_23_strides_0 = const()[name = string("value_23_strides_0"), val = tensor([1, 1])]; + tensor value_23_pad_0 = const()[name = string("value_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_23_dilations_0 = const()[name = string("value_23_dilations_0"), val = tensor([1, 1])]; + int32 value_23_groups_0 = const()[name = string("value_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454111872)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457388736)))]; + tensor value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_23_cast_fp16")]; + tensor var_1538 = const()[name = string("op_1538"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_23_cast_fp16 = reshape(shape = var_1538, x = query_23_cast_fp16)[name = string("mh_q_23_cast_fp16")]; + fp16 var_1540_to_fp16 = const()[name = string("op_1540_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1541_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1540_to_fp16)[name = string("op_1541_cast_fp16")]; + tensor var_1542 = const()[name = string("op_1542"), val = tensor([1, 20, 64, -1])]; + tensor var_1543_cast_fp16 = reshape(shape = var_1542, x = key_23_cast_fp16)[name = string("op_1543_cast_fp16")]; + bool mh_w_23_transpose_x_0 = const()[name = string("mh_w_23_transpose_x_0"), val = bool(true)]; + bool mh_w_23_transpose_y_0 = const()[name = string("mh_w_23_transpose_y_0"), val = bool(false)]; + tensor mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_1541_cast_fp16, y = var_1543_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor var_1546_cast_fp16 = softmax(axis = var_1484, x = mh_w_23_cast_fp16)[name = string("op_1546_cast_fp16")]; + tensor var_1547 = const()[name = string("op_1547"), val = tensor([1, 20, 64, -1])]; + tensor var_1548_cast_fp16 = reshape(shape = var_1547, x = value_23_cast_fp16)[name = string("op_1548_cast_fp16")]; + bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)]; + bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)]; + tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1548_cast_fp16, y = var_1546_cast_fp16)[name = string("attn_23_cast_fp16")]; + tensor var_1551 = const()[name = string("op_1551"), val = tensor([1, 1280, 1, -1])]; + tensor input_89_cast_fp16 = reshape(shape = var_1551, x = attn_23_cast_fp16)[name = string("input_89_cast_fp16")]; + string obj_47_pad_type_0 = const()[name = string("obj_47_pad_type_0"), val = string("valid")]; + tensor obj_47_strides_0 = const()[name = string("obj_47_strides_0"), val = tensor([1, 1])]; + tensor obj_47_pad_0 = const()[name = string("obj_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_47_dilations_0 = const()[name = string("obj_47_dilations_0"), val = tensor([1, 1])]; + int32 obj_47_groups_0 = const()[name = string("obj_47_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457391360)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460668224)))]; + tensor obj_47_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_47_dilations_0, groups = obj_47_groups_0, pad = obj_47_pad_0, pad_type = obj_47_pad_type_0, strides = obj_47_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_47_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_47_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_1569_to_fp16 = const()[name = string("op_1569_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1569_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460670848)))]; + tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460673472)))]; + fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460676096)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473783360)))]; + tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; + tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + string hidden_states_27_pad_type_0 = const()[name = string("hidden_states_27_pad_type_0"), val = string("valid")]; + tensor hidden_states_27_strides_0 = const()[name = string("hidden_states_27_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_27_pad_0 = const()[name = string("hidden_states_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_27_dilations_0 = const()[name = string("hidden_states_27_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_27_groups_0 = const()[name = string("hidden_states_27_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473793664)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486900928)))]; + tensor hidden_states_27_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_27_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_27_cast_fp16)[name = string("inputs_49_cast_fp16")]; + int32 var_1602 = const()[name = string("op_1602"), val = int32(3)]; + tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; + fp16 var_1621_to_fp16 = const()[name = string("op_1621_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1621_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; + tensor obj_49_gamma_0_to_fp16 = const()[name = string("obj_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486903552)))]; + tensor obj_49_beta_0_to_fp16 = const()[name = string("obj_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486906176)))]; + fp16 obj_49_epsilon_0_to_fp16 = const()[name = string("obj_49_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_49_cast_fp16 = batch_norm(beta = obj_49_beta_0_to_fp16, epsilon = obj_49_epsilon_0_to_fp16, gamma = obj_49_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_49_cast_fp16")]; + string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; + tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; + tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; + int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486908800)))]; + tensor layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490185664)))]; + tensor query_25_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("query_25_cast_fp16")]; + string key_25_pad_type_0 = const()[name = string("key_25_pad_type_0"), val = string("valid")]; + tensor key_25_strides_0 = const()[name = string("key_25_strides_0"), val = tensor([1, 1])]; + tensor key_25_pad_0 = const()[name = string("key_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_25_dilations_0 = const()[name = string("key_25_dilations_0"), val = tensor([1, 1])]; + int32 key_25_groups_0 = const()[name = string("key_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490188288)))]; + tensor key_25_cast_fp16 = conv(dilations = key_25_dilations_0, groups = key_25_groups_0, pad = key_25_pad_0, pad_type = key_25_pad_type_0, strides = key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("key_25_cast_fp16")]; + string value_25_pad_type_0 = const()[name = string("value_25_pad_type_0"), val = string("valid")]; + tensor value_25_strides_0 = const()[name = string("value_25_strides_0"), val = tensor([1, 1])]; + tensor value_25_pad_0 = const()[name = string("value_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_25_dilations_0 = const()[name = string("value_25_dilations_0"), val = tensor([1, 1])]; + int32 value_25_groups_0 = const()[name = string("value_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(493465152)))]; + tensor layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496742016)))]; + tensor value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = value_25_dilations_0, groups = value_25_groups_0, pad = value_25_pad_0, pad_type = value_25_pad_type_0, strides = value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("value_25_cast_fp16")]; + tensor var_1656 = const()[name = string("op_1656"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_25_cast_fp16 = reshape(shape = var_1656, x = query_25_cast_fp16)[name = string("mh_q_25_cast_fp16")]; + fp16 var_1658_to_fp16 = const()[name = string("op_1658_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1659_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1658_to_fp16)[name = string("op_1659_cast_fp16")]; + tensor var_1660 = const()[name = string("op_1660"), val = tensor([1, 20, 64, -1])]; + tensor var_1661_cast_fp16 = reshape(shape = var_1660, x = key_25_cast_fp16)[name = string("op_1661_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1659_cast_fp16, y = var_1661_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor var_1664_cast_fp16 = softmax(axis = var_1602, x = mh_w_25_cast_fp16)[name = string("op_1664_cast_fp16")]; + tensor var_1665 = const()[name = string("op_1665"), val = tensor([1, 20, 64, -1])]; + tensor var_1666_cast_fp16 = reshape(shape = var_1665, x = value_25_cast_fp16)[name = string("op_1666_cast_fp16")]; + bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)]; + bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)]; + tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1666_cast_fp16, y = var_1664_cast_fp16)[name = string("attn_25_cast_fp16")]; + tensor var_1669 = const()[name = string("op_1669"), val = tensor([1, 1280, 1, -1])]; + tensor input_97_cast_fp16 = reshape(shape = var_1669, x = attn_25_cast_fp16)[name = string("input_97_cast_fp16")]; + string obj_51_pad_type_0 = const()[name = string("obj_51_pad_type_0"), val = string("valid")]; + tensor obj_51_strides_0 = const()[name = string("obj_51_strides_0"), val = tensor([1, 1])]; + tensor obj_51_pad_0 = const()[name = string("obj_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_51_dilations_0 = const()[name = string("obj_51_dilations_0"), val = tensor([1, 1])]; + int32 obj_51_groups_0 = const()[name = string("obj_51_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496744640)))]; + tensor layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500021504)))]; + tensor obj_51_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_51_dilations_0, groups = obj_51_groups_0, pad = obj_51_pad_0, pad_type = obj_51_pad_type_0, strides = obj_51_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("obj_51_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_51_cast_fp16)[name = string("inputs_51_cast_fp16")]; + tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; + fp16 var_1687_to_fp16 = const()[name = string("op_1687_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1687_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; + tensor input_99_gamma_0_to_fp16 = const()[name = string("input_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500024128)))]; + tensor input_99_beta_0_to_fp16 = const()[name = string("input_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500026752)))]; + fp16 input_99_epsilon_0_to_fp16 = const()[name = string("input_99_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_99_cast_fp16 = batch_norm(beta = input_99_beta_0_to_fp16, epsilon = input_99_epsilon_0_to_fp16, gamma = input_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("input_99_cast_fp16")]; + string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")]; + tensor input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor([1, 1])]; + tensor input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor([1, 1])]; + int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1)]; + tensor layers_12_fc1_weight_to_fp16 = const()[name = string("layers_12_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500029376)))]; + tensor layers_12_fc1_bias_to_fp16 = const()[name = string("layers_12_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513136640)))]; + tensor input_101_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = layers_12_fc1_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")]; + string input_103_mode_0 = const()[name = string("input_103_mode_0"), val = string("EXACT")]; + tensor input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")]; + string hidden_states_29_pad_type_0 = const()[name = string("hidden_states_29_pad_type_0"), val = string("valid")]; + tensor hidden_states_29_strides_0 = const()[name = string("hidden_states_29_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_29_pad_0 = const()[name = string("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_29_dilations_0 = const()[name = string("hidden_states_29_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_29_groups_0 = const()[name = string("hidden_states_29_groups_0"), val = int32(1)]; + tensor layers_12_fc2_weight_to_fp16 = const()[name = string("layers_12_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513146944)))]; + tensor layers_12_fc2_bias_to_fp16 = const()[name = string("layers_12_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526254208)))]; + tensor hidden_states_29_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_12_fc2_weight_to_fp16, x = input_103_cast_fp16)[name = string("hidden_states_29_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = hidden_states_29_cast_fp16)[name = string("inputs_53_cast_fp16")]; + int32 var_1720 = const()[name = string("op_1720"), val = int32(3)]; + tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; + fp16 var_1739_to_fp16 = const()[name = string("op_1739_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_1739_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; + tensor obj_53_gamma_0_to_fp16 = const()[name = string("obj_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526256832)))]; + tensor obj_53_beta_0_to_fp16 = const()[name = string("obj_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526259456)))]; + fp16 obj_53_epsilon_0_to_fp16 = const()[name = string("obj_53_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_53_cast_fp16 = batch_norm(beta = obj_53_beta_0_to_fp16, epsilon = obj_53_epsilon_0_to_fp16, gamma = obj_53_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("obj_53_cast_fp16")]; + string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; + tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; + tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; + int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526262080)))]; + tensor layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529538944)))]; + tensor query_27_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("query_27_cast_fp16")]; + string key_27_pad_type_0 = const()[name = string("key_27_pad_type_0"), val = string("valid")]; + tensor key_27_strides_0 = const()[name = string("key_27_strides_0"), val = tensor([1, 1])]; + tensor key_27_pad_0 = const()[name = string("key_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_27_dilations_0 = const()[name = string("key_27_dilations_0"), val = tensor([1, 1])]; + int32 key_27_groups_0 = const()[name = string("key_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529541568)))]; + tensor key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("key_27_cast_fp16")]; + string value_27_pad_type_0 = const()[name = string("value_27_pad_type_0"), val = string("valid")]; + tensor value_27_strides_0 = const()[name = string("value_27_strides_0"), val = tensor([1, 1])]; + tensor value_27_pad_0 = const()[name = string("value_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_27_dilations_0 = const()[name = string("value_27_dilations_0"), val = tensor([1, 1])]; + int32 value_27_groups_0 = const()[name = string("value_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(532818432)))]; + tensor layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536095296)))]; + tensor value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("value_27_cast_fp16")]; + tensor var_1774 = const()[name = string("op_1774"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_27_cast_fp16 = reshape(shape = var_1774, x = query_27_cast_fp16)[name = string("mh_q_27_cast_fp16")]; + fp16 var_1776_to_fp16 = const()[name = string("op_1776_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1777_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1776_to_fp16)[name = string("op_1777_cast_fp16")]; + tensor var_1778 = const()[name = string("op_1778"), val = tensor([1, 20, 64, -1])]; + tensor var_1779_cast_fp16 = reshape(shape = var_1778, x = key_27_cast_fp16)[name = string("op_1779_cast_fp16")]; + bool mh_w_27_transpose_x_0 = const()[name = string("mh_w_27_transpose_x_0"), val = bool(true)]; + bool mh_w_27_transpose_y_0 = const()[name = string("mh_w_27_transpose_y_0"), val = bool(false)]; + tensor mh_w_27_cast_fp16 = matmul(transpose_x = mh_w_27_transpose_x_0, transpose_y = mh_w_27_transpose_y_0, x = var_1777_cast_fp16, y = var_1779_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_1782_cast_fp16 = softmax(axis = var_1720, x = mh_w_27_cast_fp16)[name = string("op_1782_cast_fp16")]; + tensor var_1783 = const()[name = string("op_1783"), val = tensor([1, 20, 64, -1])]; + tensor var_1784_cast_fp16 = reshape(shape = var_1783, x = value_27_cast_fp16)[name = string("op_1784_cast_fp16")]; + bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)]; + bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)]; + tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1784_cast_fp16, y = var_1782_cast_fp16)[name = string("attn_27_cast_fp16")]; + tensor var_1787 = const()[name = string("op_1787"), val = tensor([1, 1280, 1, -1])]; + tensor input_105_cast_fp16 = reshape(shape = var_1787, x = attn_27_cast_fp16)[name = string("input_105_cast_fp16")]; + string obj_55_pad_type_0 = const()[name = string("obj_55_pad_type_0"), val = string("valid")]; + tensor obj_55_strides_0 = const()[name = string("obj_55_strides_0"), val = tensor([1, 1])]; + tensor obj_55_pad_0 = const()[name = string("obj_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_55_dilations_0 = const()[name = string("obj_55_dilations_0"), val = tensor([1, 1])]; + int32 obj_55_groups_0 = const()[name = string("obj_55_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536097920)))]; + tensor layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539374784)))]; + tensor obj_55_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_55_dilations_0, groups = obj_55_groups_0, pad = obj_55_pad_0, pad_type = obj_55_pad_type_0, strides = obj_55_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("obj_55_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_55_cast_fp16)[name = string("inputs_55_cast_fp16")]; + tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; + fp16 var_1805_to_fp16 = const()[name = string("op_1805_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_1805_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; + tensor input_107_gamma_0_to_fp16 = const()[name = string("input_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539377408)))]; + tensor input_107_beta_0_to_fp16 = const()[name = string("input_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539380032)))]; + fp16 input_107_epsilon_0_to_fp16 = const()[name = string("input_107_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_107_cast_fp16 = batch_norm(beta = input_107_beta_0_to_fp16, epsilon = input_107_epsilon_0_to_fp16, gamma = input_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("input_107_cast_fp16")]; + string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("valid")]; + tensor input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor([1, 1])]; + tensor input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor([1, 1])]; + int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1)]; + tensor layers_13_fc1_weight_to_fp16 = const()[name = string("layers_13_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539382656)))]; + tensor layers_13_fc1_bias_to_fp16 = const()[name = string("layers_13_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552489920)))]; + tensor input_109_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = layers_13_fc1_weight_to_fp16, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")]; + string input_111_mode_0 = const()[name = string("input_111_mode_0"), val = string("EXACT")]; + tensor input_111_cast_fp16 = gelu(mode = input_111_mode_0, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")]; + string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")]; + tensor hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)]; + tensor layers_13_fc2_weight_to_fp16 = const()[name = string("layers_13_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552500224)))]; + tensor layers_13_fc2_bias_to_fp16 = const()[name = string("layers_13_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565607488)))]; + tensor hidden_states_31_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_13_fc2_weight_to_fp16, x = input_111_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = hidden_states_31_cast_fp16)[name = string("inputs_57_cast_fp16")]; + int32 var_1838 = const()[name = string("op_1838"), val = int32(3)]; + tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; + fp16 var_1857_to_fp16 = const()[name = string("op_1857_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_1857_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; + tensor obj_57_gamma_0_to_fp16 = const()[name = string("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565610112)))]; + tensor obj_57_beta_0_to_fp16 = const()[name = string("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565612736)))]; + fp16 obj_57_epsilon_0_to_fp16 = const()[name = string("obj_57_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_57_cast_fp16")]; + string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; + tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; + tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; + int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565615360)))]; + tensor layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568892224)))]; + tensor query_29_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("query_29_cast_fp16")]; + string key_29_pad_type_0 = const()[name = string("key_29_pad_type_0"), val = string("valid")]; + tensor key_29_strides_0 = const()[name = string("key_29_strides_0"), val = tensor([1, 1])]; + tensor key_29_pad_0 = const()[name = string("key_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_29_dilations_0 = const()[name = string("key_29_dilations_0"), val = tensor([1, 1])]; + int32 key_29_groups_0 = const()[name = string("key_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568894848)))]; + tensor key_29_cast_fp16 = conv(dilations = key_29_dilations_0, groups = key_29_groups_0, pad = key_29_pad_0, pad_type = key_29_pad_type_0, strides = key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("key_29_cast_fp16")]; + string value_29_pad_type_0 = const()[name = string("value_29_pad_type_0"), val = string("valid")]; + tensor value_29_strides_0 = const()[name = string("value_29_strides_0"), val = tensor([1, 1])]; + tensor value_29_pad_0 = const()[name = string("value_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_29_dilations_0 = const()[name = string("value_29_dilations_0"), val = tensor([1, 1])]; + int32 value_29_groups_0 = const()[name = string("value_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572171712)))]; + tensor layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575448576)))]; + tensor value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = value_29_dilations_0, groups = value_29_groups_0, pad = value_29_pad_0, pad_type = value_29_pad_type_0, strides = value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("value_29_cast_fp16")]; + tensor var_1892 = const()[name = string("op_1892"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_29_cast_fp16 = reshape(shape = var_1892, x = query_29_cast_fp16)[name = string("mh_q_29_cast_fp16")]; + fp16 var_1894_to_fp16 = const()[name = string("op_1894_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1895_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1894_to_fp16)[name = string("op_1895_cast_fp16")]; + tensor var_1896 = const()[name = string("op_1896"), val = tensor([1, 20, 64, -1])]; + tensor var_1897_cast_fp16 = reshape(shape = var_1896, x = key_29_cast_fp16)[name = string("op_1897_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1895_cast_fp16, y = var_1897_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor var_1900_cast_fp16 = softmax(axis = var_1838, x = mh_w_29_cast_fp16)[name = string("op_1900_cast_fp16")]; + tensor var_1901 = const()[name = string("op_1901"), val = tensor([1, 20, 64, -1])]; + tensor var_1902_cast_fp16 = reshape(shape = var_1901, x = value_29_cast_fp16)[name = string("op_1902_cast_fp16")]; + bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)]; + bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)]; + tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1902_cast_fp16, y = var_1900_cast_fp16)[name = string("attn_29_cast_fp16")]; + tensor var_1905 = const()[name = string("op_1905"), val = tensor([1, 1280, 1, -1])]; + tensor input_113_cast_fp16 = reshape(shape = var_1905, x = attn_29_cast_fp16)[name = string("input_113_cast_fp16")]; + string obj_59_pad_type_0 = const()[name = string("obj_59_pad_type_0"), val = string("valid")]; + tensor obj_59_strides_0 = const()[name = string("obj_59_strides_0"), val = tensor([1, 1])]; + tensor obj_59_pad_0 = const()[name = string("obj_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_59_dilations_0 = const()[name = string("obj_59_dilations_0"), val = tensor([1, 1])]; + int32 obj_59_groups_0 = const()[name = string("obj_59_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575451200)))]; + tensor layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578728064)))]; + tensor obj_59_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_59_dilations_0, groups = obj_59_groups_0, pad = obj_59_pad_0, pad_type = obj_59_pad_type_0, strides = obj_59_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_59_cast_fp16)[name = string("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; + fp16 var_1923_to_fp16 = const()[name = string("op_1923_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_1923_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578730688)))]; + tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578733312)))]; + fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor layers_14_fc1_weight_to_fp16 = const()[name = string("layers_14_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578735936)))]; + tensor layers_14_fc1_bias_to_fp16 = const()[name = string("layers_14_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591843200)))]; + tensor input_117_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_14_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")]; + tensor input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")]; + string hidden_states_33_pad_type_0 = const()[name = string("hidden_states_33_pad_type_0"), val = string("valid")]; + tensor hidden_states_33_strides_0 = const()[name = string("hidden_states_33_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_33_pad_0 = const()[name = string("hidden_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_33_dilations_0 = const()[name = string("hidden_states_33_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_33_groups_0 = const()[name = string("hidden_states_33_groups_0"), val = int32(1)]; + tensor layers_14_fc2_weight_to_fp16 = const()[name = string("layers_14_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591853504)))]; + tensor layers_14_fc2_bias_to_fp16 = const()[name = string("layers_14_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604960768)))]; + tensor hidden_states_33_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_14_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = string("hidden_states_33_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_33_cast_fp16)[name = string("inputs_61_cast_fp16")]; + int32 var_1956 = const()[name = string("op_1956"), val = int32(3)]; + tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; + fp16 var_1975_to_fp16 = const()[name = string("op_1975_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_1975_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604963392)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604966016)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; + tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; + tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; + int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604968640)))]; + tensor layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608245504)))]; + tensor query_31_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_31_cast_fp16")]; + string key_31_pad_type_0 = const()[name = string("key_31_pad_type_0"), val = string("valid")]; + tensor key_31_strides_0 = const()[name = string("key_31_strides_0"), val = tensor([1, 1])]; + tensor key_31_pad_0 = const()[name = string("key_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_31_dilations_0 = const()[name = string("key_31_dilations_0"), val = tensor([1, 1])]; + int32 key_31_groups_0 = const()[name = string("key_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608248128)))]; + tensor key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("key_31_cast_fp16")]; + string value_31_pad_type_0 = const()[name = string("value_31_pad_type_0"), val = string("valid")]; + tensor value_31_strides_0 = const()[name = string("value_31_strides_0"), val = tensor([1, 1])]; + tensor value_31_pad_0 = const()[name = string("value_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_31_dilations_0 = const()[name = string("value_31_dilations_0"), val = tensor([1, 1])]; + int32 value_31_groups_0 = const()[name = string("value_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(611524992)))]; + tensor layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614801856)))]; + tensor value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("value_31_cast_fp16")]; + tensor var_2010 = const()[name = string("op_2010"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_31_cast_fp16 = reshape(shape = var_2010, x = query_31_cast_fp16)[name = string("mh_q_31_cast_fp16")]; + fp16 var_2012_to_fp16 = const()[name = string("op_2012_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2013_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_2012_to_fp16)[name = string("op_2013_cast_fp16")]; + tensor var_2014 = const()[name = string("op_2014"), val = tensor([1, 20, 64, -1])]; + tensor var_2015_cast_fp16 = reshape(shape = var_2014, x = key_31_cast_fp16)[name = string("op_2015_cast_fp16")]; + bool mh_w_31_transpose_x_0 = const()[name = string("mh_w_31_transpose_x_0"), val = bool(true)]; + bool mh_w_31_transpose_y_0 = const()[name = string("mh_w_31_transpose_y_0"), val = bool(false)]; + tensor mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_2013_cast_fp16, y = var_2015_cast_fp16)[name = string("mh_w_31_cast_fp16")]; + tensor var_2018_cast_fp16 = softmax(axis = var_1956, x = mh_w_31_cast_fp16)[name = string("op_2018_cast_fp16")]; + tensor var_2019 = const()[name = string("op_2019"), val = tensor([1, 20, 64, -1])]; + tensor var_2020_cast_fp16 = reshape(shape = var_2019, x = value_31_cast_fp16)[name = string("op_2020_cast_fp16")]; + bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)]; + bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)]; + tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_2020_cast_fp16, y = var_2018_cast_fp16)[name = string("attn_31_cast_fp16")]; + tensor var_2023 = const()[name = string("op_2023"), val = tensor([1, 1280, 1, -1])]; + tensor input_121_cast_fp16 = reshape(shape = var_2023, x = attn_31_cast_fp16)[name = string("input_121_cast_fp16")]; + string obj_63_pad_type_0 = const()[name = string("obj_63_pad_type_0"), val = string("valid")]; + tensor obj_63_strides_0 = const()[name = string("obj_63_strides_0"), val = tensor([1, 1])]; + tensor obj_63_pad_0 = const()[name = string("obj_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_63_dilations_0 = const()[name = string("obj_63_dilations_0"), val = tensor([1, 1])]; + int32 obj_63_groups_0 = const()[name = string("obj_63_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614804480)))]; + tensor layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618081344)))]; + tensor obj_63_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_121_cast_fp16)[name = string("obj_63_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_63_cast_fp16)[name = string("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; + fp16 var_2041_to_fp16 = const()[name = string("op_2041_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2041_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; + tensor input_123_gamma_0_to_fp16 = const()[name = string("input_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618083968)))]; + tensor input_123_beta_0_to_fp16 = const()[name = string("input_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618086592)))]; + fp16 input_123_epsilon_0_to_fp16 = const()[name = string("input_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_123_cast_fp16 = batch_norm(beta = input_123_beta_0_to_fp16, epsilon = input_123_epsilon_0_to_fp16, gamma = input_123_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("input_123_cast_fp16")]; + string input_125_pad_type_0 = const()[name = string("input_125_pad_type_0"), val = string("valid")]; + tensor input_125_strides_0 = const()[name = string("input_125_strides_0"), val = tensor([1, 1])]; + tensor input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_125_dilations_0 = const()[name = string("input_125_dilations_0"), val = tensor([1, 1])]; + int32 input_125_groups_0 = const()[name = string("input_125_groups_0"), val = int32(1)]; + tensor layers_15_fc1_weight_to_fp16 = const()[name = string("layers_15_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618089216)))]; + tensor layers_15_fc1_bias_to_fp16 = const()[name = string("layers_15_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631196480)))]; + tensor input_125_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = layers_15_fc1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")]; + string input_127_mode_0 = const()[name = string("input_127_mode_0"), val = string("EXACT")]; + tensor input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")]; + string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")]; + tensor hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)]; + tensor layers_15_fc2_weight_to_fp16 = const()[name = string("layers_15_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631206784)))]; + tensor layers_15_fc2_bias_to_fp16 = const()[name = string("layers_15_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644314048)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_15_fc2_weight_to_fp16, x = input_127_cast_fp16)[name = string("hidden_states_35_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = hidden_states_35_cast_fp16)[name = string("inputs_65_cast_fp16")]; + int32 var_2074 = const()[name = string("op_2074"), val = int32(3)]; + tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; + fp16 var_2093_to_fp16 = const()[name = string("op_2093_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2093_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; + tensor obj_65_gamma_0_to_fp16 = const()[name = string("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644316672)))]; + tensor obj_65_beta_0_to_fp16 = const()[name = string("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644319296)))]; + fp16 obj_65_epsilon_0_to_fp16 = const()[name = string("obj_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("obj_65_cast_fp16")]; + string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; + tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; + tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; + int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644321920)))]; + tensor layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647598784)))]; + tensor query_33_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("query_33_cast_fp16")]; + string key_33_pad_type_0 = const()[name = string("key_33_pad_type_0"), val = string("valid")]; + tensor key_33_strides_0 = const()[name = string("key_33_strides_0"), val = tensor([1, 1])]; + tensor key_33_pad_0 = const()[name = string("key_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_33_dilations_0 = const()[name = string("key_33_dilations_0"), val = tensor([1, 1])]; + int32 key_33_groups_0 = const()[name = string("key_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647601408)))]; + tensor key_33_cast_fp16 = conv(dilations = key_33_dilations_0, groups = key_33_groups_0, pad = key_33_pad_0, pad_type = key_33_pad_type_0, strides = key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("key_33_cast_fp16")]; + string value_33_pad_type_0 = const()[name = string("value_33_pad_type_0"), val = string("valid")]; + tensor value_33_strides_0 = const()[name = string("value_33_strides_0"), val = tensor([1, 1])]; + tensor value_33_pad_0 = const()[name = string("value_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_33_dilations_0 = const()[name = string("value_33_dilations_0"), val = tensor([1, 1])]; + int32 value_33_groups_0 = const()[name = string("value_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(650878272)))]; + tensor layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654155136)))]; + tensor value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = value_33_dilations_0, groups = value_33_groups_0, pad = value_33_pad_0, pad_type = value_33_pad_type_0, strides = value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("value_33_cast_fp16")]; + tensor var_2128 = const()[name = string("op_2128"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_33_cast_fp16 = reshape(shape = var_2128, x = query_33_cast_fp16)[name = string("mh_q_33_cast_fp16")]; + fp16 var_2130_to_fp16 = const()[name = string("op_2130_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2131_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_2130_to_fp16)[name = string("op_2131_cast_fp16")]; + tensor var_2132 = const()[name = string("op_2132"), val = tensor([1, 20, 64, -1])]; + tensor var_2133_cast_fp16 = reshape(shape = var_2132, x = key_33_cast_fp16)[name = string("op_2133_cast_fp16")]; + bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; + bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_2131_cast_fp16, y = var_2133_cast_fp16)[name = string("mh_w_33_cast_fp16")]; + tensor var_2136_cast_fp16 = softmax(axis = var_2074, x = mh_w_33_cast_fp16)[name = string("op_2136_cast_fp16")]; + tensor var_2137 = const()[name = string("op_2137"), val = tensor([1, 20, 64, -1])]; + tensor var_2138_cast_fp16 = reshape(shape = var_2137, x = value_33_cast_fp16)[name = string("op_2138_cast_fp16")]; + bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)]; + bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)]; + tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_2138_cast_fp16, y = var_2136_cast_fp16)[name = string("attn_33_cast_fp16")]; + tensor var_2141 = const()[name = string("op_2141"), val = tensor([1, 1280, 1, -1])]; + tensor input_129_cast_fp16 = reshape(shape = var_2141, x = attn_33_cast_fp16)[name = string("input_129_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654157760)))]; + tensor layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657434624)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_129_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_67_cast_fp16")]; + tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; + fp16 var_2159_to_fp16 = const()[name = string("op_2159_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2159_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; + tensor input_131_gamma_0_to_fp16 = const()[name = string("input_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657437248)))]; + tensor input_131_beta_0_to_fp16 = const()[name = string("input_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657439872)))]; + fp16 input_131_epsilon_0_to_fp16 = const()[name = string("input_131_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("input_131_cast_fp16")]; + string input_133_pad_type_0 = const()[name = string("input_133_pad_type_0"), val = string("valid")]; + tensor input_133_strides_0 = const()[name = string("input_133_strides_0"), val = tensor([1, 1])]; + tensor input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_133_dilations_0 = const()[name = string("input_133_dilations_0"), val = tensor([1, 1])]; + int32 input_133_groups_0 = const()[name = string("input_133_groups_0"), val = int32(1)]; + tensor layers_16_fc1_weight_to_fp16 = const()[name = string("layers_16_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657442496)))]; + tensor layers_16_fc1_bias_to_fp16 = const()[name = string("layers_16_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670549760)))]; + tensor input_133_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = layers_16_fc1_weight_to_fp16, x = input_131_cast_fp16)[name = string("input_133_cast_fp16")]; + string input_135_mode_0 = const()[name = string("input_135_mode_0"), val = string("EXACT")]; + tensor input_135_cast_fp16 = gelu(mode = input_135_mode_0, x = input_133_cast_fp16)[name = string("input_135_cast_fp16")]; + string hidden_states_37_pad_type_0 = const()[name = string("hidden_states_37_pad_type_0"), val = string("valid")]; + tensor hidden_states_37_strides_0 = const()[name = string("hidden_states_37_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_37_pad_0 = const()[name = string("hidden_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_37_dilations_0 = const()[name = string("hidden_states_37_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_37_groups_0 = const()[name = string("hidden_states_37_groups_0"), val = int32(1)]; + tensor layers_16_fc2_weight_to_fp16 = const()[name = string("layers_16_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670560064)))]; + tensor layers_16_fc2_bias_to_fp16 = const()[name = string("layers_16_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683667328)))]; + tensor hidden_states_37_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_16_fc2_weight_to_fp16, x = input_135_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = hidden_states_37_cast_fp16)[name = string("inputs_69_cast_fp16")]; + int32 var_2192 = const()[name = string("op_2192"), val = int32(3)]; + tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; + fp16 var_2211_to_fp16 = const()[name = string("op_2211_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2211_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683669952)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683672576)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; + tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; + tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; + int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683675200)))]; + tensor layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686952064)))]; + tensor query_35_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_35_cast_fp16")]; + string key_35_pad_type_0 = const()[name = string("key_35_pad_type_0"), val = string("valid")]; + tensor key_35_strides_0 = const()[name = string("key_35_strides_0"), val = tensor([1, 1])]; + tensor key_35_pad_0 = const()[name = string("key_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_35_dilations_0 = const()[name = string("key_35_dilations_0"), val = tensor([1, 1])]; + int32 key_35_groups_0 = const()[name = string("key_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686954688)))]; + tensor key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("key_35_cast_fp16")]; + string value_35_pad_type_0 = const()[name = string("value_35_pad_type_0"), val = string("valid")]; + tensor value_35_strides_0 = const()[name = string("value_35_strides_0"), val = tensor([1, 1])]; + tensor value_35_pad_0 = const()[name = string("value_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_35_dilations_0 = const()[name = string("value_35_dilations_0"), val = tensor([1, 1])]; + int32 value_35_groups_0 = const()[name = string("value_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(690231552)))]; + tensor layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693508416)))]; + tensor value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("value_35_cast_fp16")]; + tensor var_2246 = const()[name = string("op_2246"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_35_cast_fp16 = reshape(shape = var_2246, x = query_35_cast_fp16)[name = string("mh_q_35_cast_fp16")]; + fp16 var_2248_to_fp16 = const()[name = string("op_2248_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2249_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2248_to_fp16)[name = string("op_2249_cast_fp16")]; + tensor var_2250 = const()[name = string("op_2250"), val = tensor([1, 20, 64, -1])]; + tensor var_2251_cast_fp16 = reshape(shape = var_2250, x = key_35_cast_fp16)[name = string("op_2251_cast_fp16")]; + bool mh_w_35_transpose_x_0 = const()[name = string("mh_w_35_transpose_x_0"), val = bool(true)]; + bool mh_w_35_transpose_y_0 = const()[name = string("mh_w_35_transpose_y_0"), val = bool(false)]; + tensor mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_2249_cast_fp16, y = var_2251_cast_fp16)[name = string("mh_w_35_cast_fp16")]; + tensor var_2254_cast_fp16 = softmax(axis = var_2192, x = mh_w_35_cast_fp16)[name = string("op_2254_cast_fp16")]; + tensor var_2255 = const()[name = string("op_2255"), val = tensor([1, 20, 64, -1])]; + tensor var_2256_cast_fp16 = reshape(shape = var_2255, x = value_35_cast_fp16)[name = string("op_2256_cast_fp16")]; + bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)]; + bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)]; + tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2256_cast_fp16, y = var_2254_cast_fp16)[name = string("attn_35_cast_fp16")]; + tensor var_2259 = const()[name = string("op_2259"), val = tensor([1, 1280, 1, -1])]; + tensor input_137_cast_fp16 = reshape(shape = var_2259, x = attn_35_cast_fp16)[name = string("input_137_cast_fp16")]; + string obj_71_pad_type_0 = const()[name = string("obj_71_pad_type_0"), val = string("valid")]; + tensor obj_71_strides_0 = const()[name = string("obj_71_strides_0"), val = tensor([1, 1])]; + tensor obj_71_pad_0 = const()[name = string("obj_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_71_dilations_0 = const()[name = string("obj_71_dilations_0"), val = tensor([1, 1])]; + int32 obj_71_groups_0 = const()[name = string("obj_71_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693511040)))]; + tensor layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696787904)))]; + tensor obj_71_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_71_dilations_0, groups = obj_71_groups_0, pad = obj_71_pad_0, pad_type = obj_71_pad_type_0, strides = obj_71_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("obj_71_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_71_cast_fp16)[name = string("inputs_71_cast_fp16")]; + tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; + fp16 var_2277_to_fp16 = const()[name = string("op_2277_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2277_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; + tensor input_139_gamma_0_to_fp16 = const()[name = string("input_139_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696790528)))]; + tensor input_139_beta_0_to_fp16 = const()[name = string("input_139_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696793152)))]; + fp16 input_139_epsilon_0_to_fp16 = const()[name = string("input_139_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_139_cast_fp16 = batch_norm(beta = input_139_beta_0_to_fp16, epsilon = input_139_epsilon_0_to_fp16, gamma = input_139_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_139_cast_fp16")]; + string input_141_pad_type_0 = const()[name = string("input_141_pad_type_0"), val = string("valid")]; + tensor input_141_strides_0 = const()[name = string("input_141_strides_0"), val = tensor([1, 1])]; + tensor input_141_pad_0 = const()[name = string("input_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_141_dilations_0 = const()[name = string("input_141_dilations_0"), val = tensor([1, 1])]; + int32 input_141_groups_0 = const()[name = string("input_141_groups_0"), val = int32(1)]; + tensor layers_17_fc1_weight_to_fp16 = const()[name = string("layers_17_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696795776)))]; + tensor layers_17_fc1_bias_to_fp16 = const()[name = string("layers_17_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709903040)))]; + tensor input_141_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_141_dilations_0, groups = input_141_groups_0, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = input_141_strides_0, weight = layers_17_fc1_weight_to_fp16, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")]; + string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("EXACT")]; + tensor input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")]; + string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")]; + tensor hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)]; + tensor layers_17_fc2_weight_to_fp16 = const()[name = string("layers_17_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709913344)))]; + tensor layers_17_fc2_bias_to_fp16 = const()[name = string("layers_17_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723020608)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_17_fc2_weight_to_fp16, x = input_143_cast_fp16)[name = string("hidden_states_39_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("inputs_73_cast_fp16")]; + int32 var_2310 = const()[name = string("op_2310"), val = int32(3)]; + tensor out_73_axes_0 = const()[name = string("out_73_axes_0"), val = tensor([1])]; + fp16 var_2329_to_fp16 = const()[name = string("op_2329_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2329_to_fp16, x = inputs_73_cast_fp16)[name = string("out_73_cast_fp16")]; + tensor obj_73_gamma_0_to_fp16 = const()[name = string("obj_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723023232)))]; + tensor obj_73_beta_0_to_fp16 = const()[name = string("obj_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723025856)))]; + fp16 obj_73_epsilon_0_to_fp16 = const()[name = string("obj_73_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_73_cast_fp16 = batch_norm(beta = obj_73_beta_0_to_fp16, epsilon = obj_73_epsilon_0_to_fp16, gamma = obj_73_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = string("obj_73_cast_fp16")]; + string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; + tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; + tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; + int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723028480)))]; + tensor layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726305344)))]; + tensor query_37_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("query_37_cast_fp16")]; + string key_37_pad_type_0 = const()[name = string("key_37_pad_type_0"), val = string("valid")]; + tensor key_37_strides_0 = const()[name = string("key_37_strides_0"), val = tensor([1, 1])]; + tensor key_37_pad_0 = const()[name = string("key_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_37_dilations_0 = const()[name = string("key_37_dilations_0"), val = tensor([1, 1])]; + int32 key_37_groups_0 = const()[name = string("key_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726307968)))]; + tensor key_37_cast_fp16 = conv(dilations = key_37_dilations_0, groups = key_37_groups_0, pad = key_37_pad_0, pad_type = key_37_pad_type_0, strides = key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("key_37_cast_fp16")]; + string value_37_pad_type_0 = const()[name = string("value_37_pad_type_0"), val = string("valid")]; + tensor value_37_strides_0 = const()[name = string("value_37_strides_0"), val = tensor([1, 1])]; + tensor value_37_pad_0 = const()[name = string("value_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_37_dilations_0 = const()[name = string("value_37_dilations_0"), val = tensor([1, 1])]; + int32 value_37_groups_0 = const()[name = string("value_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(729584832)))]; + tensor layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732861696)))]; + tensor value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = value_37_dilations_0, groups = value_37_groups_0, pad = value_37_pad_0, pad_type = value_37_pad_type_0, strides = value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("value_37_cast_fp16")]; + tensor var_2364 = const()[name = string("op_2364"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_37_cast_fp16 = reshape(shape = var_2364, x = query_37_cast_fp16)[name = string("mh_q_37_cast_fp16")]; + fp16 var_2366_to_fp16 = const()[name = string("op_2366_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2367_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2366_to_fp16)[name = string("op_2367_cast_fp16")]; + tensor var_2368 = const()[name = string("op_2368"), val = tensor([1, 20, 64, -1])]; + tensor var_2369_cast_fp16 = reshape(shape = var_2368, x = key_37_cast_fp16)[name = string("op_2369_cast_fp16")]; + bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; + bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_2367_cast_fp16, y = var_2369_cast_fp16)[name = string("mh_w_37_cast_fp16")]; + tensor var_2372_cast_fp16 = softmax(axis = var_2310, x = mh_w_37_cast_fp16)[name = string("op_2372_cast_fp16")]; + tensor var_2373 = const()[name = string("op_2373"), val = tensor([1, 20, 64, -1])]; + tensor var_2374_cast_fp16 = reshape(shape = var_2373, x = value_37_cast_fp16)[name = string("op_2374_cast_fp16")]; + bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)]; + bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)]; + tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2374_cast_fp16, y = var_2372_cast_fp16)[name = string("attn_37_cast_fp16")]; + tensor var_2377 = const()[name = string("op_2377"), val = tensor([1, 1280, 1, -1])]; + tensor input_145_cast_fp16 = reshape(shape = var_2377, x = attn_37_cast_fp16)[name = string("input_145_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732864320)))]; + tensor layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736141184)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_75_cast_fp16")]; + tensor out_75_axes_0 = const()[name = string("out_75_axes_0"), val = tensor([1])]; + fp16 var_2395_to_fp16 = const()[name = string("op_2395_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_2395_to_fp16, x = inputs_75_cast_fp16)[name = string("out_75_cast_fp16")]; + tensor input_147_gamma_0_to_fp16 = const()[name = string("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736143808)))]; + tensor input_147_beta_0_to_fp16 = const()[name = string("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736146432)))]; + fp16 input_147_epsilon_0_to_fp16 = const()[name = string("input_147_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = string("input_147_cast_fp16")]; + string input_149_pad_type_0 = const()[name = string("input_149_pad_type_0"), val = string("valid")]; + tensor input_149_strides_0 = const()[name = string("input_149_strides_0"), val = tensor([1, 1])]; + tensor input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_149_dilations_0 = const()[name = string("input_149_dilations_0"), val = tensor([1, 1])]; + int32 input_149_groups_0 = const()[name = string("input_149_groups_0"), val = int32(1)]; + tensor layers_18_fc1_weight_to_fp16 = const()[name = string("layers_18_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736149056)))]; + tensor layers_18_fc1_bias_to_fp16 = const()[name = string("layers_18_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749256320)))]; + tensor input_149_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_149_dilations_0, groups = input_149_groups_0, pad = input_149_pad_0, pad_type = input_149_pad_type_0, strides = input_149_strides_0, weight = layers_18_fc1_weight_to_fp16, x = input_147_cast_fp16)[name = string("input_149_cast_fp16")]; + string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("EXACT")]; + tensor input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")]; + string hidden_states_41_pad_type_0 = const()[name = string("hidden_states_41_pad_type_0"), val = string("valid")]; + tensor hidden_states_41_strides_0 = const()[name = string("hidden_states_41_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_41_pad_0 = const()[name = string("hidden_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_41_dilations_0 = const()[name = string("hidden_states_41_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_41_groups_0 = const()[name = string("hidden_states_41_groups_0"), val = int32(1)]; + tensor layers_18_fc2_weight_to_fp16 = const()[name = string("layers_18_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749266624)))]; + tensor layers_18_fc2_bias_to_fp16 = const()[name = string("layers_18_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762373888)))]; + tensor hidden_states_41_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_18_fc2_weight_to_fp16, x = input_151_cast_fp16)[name = string("hidden_states_41_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = hidden_states_41_cast_fp16)[name = string("inputs_77_cast_fp16")]; + int32 var_2428 = const()[name = string("op_2428"), val = int32(3)]; + tensor out_77_axes_0 = const()[name = string("out_77_axes_0"), val = tensor([1])]; + fp16 var_2447_to_fp16 = const()[name = string("op_2447_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_2447_to_fp16, x = inputs_77_cast_fp16)[name = string("out_77_cast_fp16")]; + tensor obj_77_gamma_0_to_fp16 = const()[name = string("obj_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762376512)))]; + tensor obj_77_beta_0_to_fp16 = const()[name = string("obj_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762379136)))]; + fp16 obj_77_epsilon_0_to_fp16 = const()[name = string("obj_77_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_77_cast_fp16 = batch_norm(beta = obj_77_beta_0_to_fp16, epsilon = obj_77_epsilon_0_to_fp16, gamma = obj_77_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = string("obj_77_cast_fp16")]; + string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; + tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; + tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; + int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762381760)))]; + tensor layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765658624)))]; + tensor query_39_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("query_39_cast_fp16")]; + string key_39_pad_type_0 = const()[name = string("key_39_pad_type_0"), val = string("valid")]; + tensor key_39_strides_0 = const()[name = string("key_39_strides_0"), val = tensor([1, 1])]; + tensor key_39_pad_0 = const()[name = string("key_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_39_dilations_0 = const()[name = string("key_39_dilations_0"), val = tensor([1, 1])]; + int32 key_39_groups_0 = const()[name = string("key_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765661248)))]; + tensor key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("key_39_cast_fp16")]; + string value_39_pad_type_0 = const()[name = string("value_39_pad_type_0"), val = string("valid")]; + tensor value_39_strides_0 = const()[name = string("value_39_strides_0"), val = tensor([1, 1])]; + tensor value_39_pad_0 = const()[name = string("value_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_39_dilations_0 = const()[name = string("value_39_dilations_0"), val = tensor([1, 1])]; + int32 value_39_groups_0 = const()[name = string("value_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(768938112)))]; + tensor layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772214976)))]; + tensor value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("value_39_cast_fp16")]; + tensor var_2482 = const()[name = string("op_2482"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_39_cast_fp16 = reshape(shape = var_2482, x = query_39_cast_fp16)[name = string("mh_q_39_cast_fp16")]; + fp16 var_2484_to_fp16 = const()[name = string("op_2484_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2485_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2484_to_fp16)[name = string("op_2485_cast_fp16")]; + tensor var_2486 = const()[name = string("op_2486"), val = tensor([1, 20, 64, -1])]; + tensor var_2487_cast_fp16 = reshape(shape = var_2486, x = key_39_cast_fp16)[name = string("op_2487_cast_fp16")]; + bool mh_w_39_transpose_x_0 = const()[name = string("mh_w_39_transpose_x_0"), val = bool(true)]; + bool mh_w_39_transpose_y_0 = const()[name = string("mh_w_39_transpose_y_0"), val = bool(false)]; + tensor mh_w_39_cast_fp16 = matmul(transpose_x = mh_w_39_transpose_x_0, transpose_y = mh_w_39_transpose_y_0, x = var_2485_cast_fp16, y = var_2487_cast_fp16)[name = string("mh_w_39_cast_fp16")]; + tensor var_2490_cast_fp16 = softmax(axis = var_2428, x = mh_w_39_cast_fp16)[name = string("op_2490_cast_fp16")]; + tensor var_2491 = const()[name = string("op_2491"), val = tensor([1, 20, 64, -1])]; + tensor var_2492_cast_fp16 = reshape(shape = var_2491, x = value_39_cast_fp16)[name = string("op_2492_cast_fp16")]; + bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)]; + bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)]; + tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2492_cast_fp16, y = var_2490_cast_fp16)[name = string("attn_39_cast_fp16")]; + tensor var_2495 = const()[name = string("op_2495"), val = tensor([1, 1280, 1, -1])]; + tensor input_153_cast_fp16 = reshape(shape = var_2495, x = attn_39_cast_fp16)[name = string("input_153_cast_fp16")]; + string obj_79_pad_type_0 = const()[name = string("obj_79_pad_type_0"), val = string("valid")]; + tensor obj_79_strides_0 = const()[name = string("obj_79_strides_0"), val = tensor([1, 1])]; + tensor obj_79_pad_0 = const()[name = string("obj_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_79_dilations_0 = const()[name = string("obj_79_dilations_0"), val = tensor([1, 1])]; + int32 obj_79_groups_0 = const()[name = string("obj_79_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772217600)))]; + tensor layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775494464)))]; + tensor obj_79_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_79_dilations_0, groups = obj_79_groups_0, pad = obj_79_pad_0, pad_type = obj_79_pad_type_0, strides = obj_79_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = string("obj_79_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = obj_79_cast_fp16)[name = string("inputs_79_cast_fp16")]; + tensor out_79_axes_0 = const()[name = string("out_79_axes_0"), val = tensor([1])]; + fp16 var_2513_to_fp16 = const()[name = string("op_2513_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_2513_to_fp16, x = inputs_79_cast_fp16)[name = string("out_79_cast_fp16")]; + tensor input_155_gamma_0_to_fp16 = const()[name = string("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775497088)))]; + tensor input_155_beta_0_to_fp16 = const()[name = string("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775499712)))]; + fp16 input_155_epsilon_0_to_fp16 = const()[name = string("input_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = string("input_155_cast_fp16")]; + string input_157_pad_type_0 = const()[name = string("input_157_pad_type_0"), val = string("valid")]; + tensor input_157_strides_0 = const()[name = string("input_157_strides_0"), val = tensor([1, 1])]; + tensor input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_157_dilations_0 = const()[name = string("input_157_dilations_0"), val = tensor([1, 1])]; + int32 input_157_groups_0 = const()[name = string("input_157_groups_0"), val = int32(1)]; + tensor layers_19_fc1_weight_to_fp16 = const()[name = string("layers_19_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775502336)))]; + tensor layers_19_fc1_bias_to_fp16 = const()[name = string("layers_19_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788609600)))]; + tensor input_157_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_19_fc1_weight_to_fp16, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")]; + string input_159_mode_0 = const()[name = string("input_159_mode_0"), val = string("EXACT")]; + tensor input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")]; + string hidden_states_43_pad_type_0 = const()[name = string("hidden_states_43_pad_type_0"), val = string("valid")]; + tensor hidden_states_43_strides_0 = const()[name = string("hidden_states_43_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_43_pad_0 = const()[name = string("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_43_dilations_0 = const()[name = string("hidden_states_43_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_43_groups_0 = const()[name = string("hidden_states_43_groups_0"), val = int32(1)]; + tensor layers_19_fc2_weight_to_fp16 = const()[name = string("layers_19_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788619904)))]; + tensor layers_19_fc2_bias_to_fp16 = const()[name = string("layers_19_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801727168)))]; + tensor hidden_states_43_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_19_fc2_weight_to_fp16, x = input_159_cast_fp16)[name = string("hidden_states_43_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = hidden_states_43_cast_fp16)[name = string("inputs_81_cast_fp16")]; + int32 var_2546 = const()[name = string("op_2546"), val = int32(3)]; + tensor out_81_axes_0 = const()[name = string("out_81_axes_0"), val = tensor([1])]; + fp16 var_2565_to_fp16 = const()[name = string("op_2565_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_2565_to_fp16, x = inputs_81_cast_fp16)[name = string("out_81_cast_fp16")]; + tensor obj_81_gamma_0_to_fp16 = const()[name = string("obj_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801729792)))]; + tensor obj_81_beta_0_to_fp16 = const()[name = string("obj_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801732416)))]; + fp16 obj_81_epsilon_0_to_fp16 = const()[name = string("obj_81_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_81_cast_fp16 = batch_norm(beta = obj_81_beta_0_to_fp16, epsilon = obj_81_epsilon_0_to_fp16, gamma = obj_81_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = string("obj_81_cast_fp16")]; + string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; + tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; + tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; + int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801735040)))]; + tensor layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805011904)))]; + tensor query_41_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("query_41_cast_fp16")]; + string key_41_pad_type_0 = const()[name = string("key_41_pad_type_0"), val = string("valid")]; + tensor key_41_strides_0 = const()[name = string("key_41_strides_0"), val = tensor([1, 1])]; + tensor key_41_pad_0 = const()[name = string("key_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_41_dilations_0 = const()[name = string("key_41_dilations_0"), val = tensor([1, 1])]; + int32 key_41_groups_0 = const()[name = string("key_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805014528)))]; + tensor key_41_cast_fp16 = conv(dilations = key_41_dilations_0, groups = key_41_groups_0, pad = key_41_pad_0, pad_type = key_41_pad_type_0, strides = key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("key_41_cast_fp16")]; + string value_41_pad_type_0 = const()[name = string("value_41_pad_type_0"), val = string("valid")]; + tensor value_41_strides_0 = const()[name = string("value_41_strides_0"), val = tensor([1, 1])]; + tensor value_41_pad_0 = const()[name = string("value_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_41_dilations_0 = const()[name = string("value_41_dilations_0"), val = tensor([1, 1])]; + int32 value_41_groups_0 = const()[name = string("value_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(808291392)))]; + tensor layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811568256)))]; + tensor value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = value_41_dilations_0, groups = value_41_groups_0, pad = value_41_pad_0, pad_type = value_41_pad_type_0, strides = value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("value_41_cast_fp16")]; + tensor var_2600 = const()[name = string("op_2600"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_41_cast_fp16 = reshape(shape = var_2600, x = query_41_cast_fp16)[name = string("mh_q_41_cast_fp16")]; + fp16 var_2602_to_fp16 = const()[name = string("op_2602_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2603_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2602_to_fp16)[name = string("op_2603_cast_fp16")]; + tensor var_2604 = const()[name = string("op_2604"), val = tensor([1, 20, 64, -1])]; + tensor var_2605_cast_fp16 = reshape(shape = var_2604, x = key_41_cast_fp16)[name = string("op_2605_cast_fp16")]; + bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; + bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_2603_cast_fp16, y = var_2605_cast_fp16)[name = string("mh_w_41_cast_fp16")]; + tensor var_2608_cast_fp16 = softmax(axis = var_2546, x = mh_w_41_cast_fp16)[name = string("op_2608_cast_fp16")]; + tensor var_2609 = const()[name = string("op_2609"), val = tensor([1, 20, 64, -1])]; + tensor var_2610_cast_fp16 = reshape(shape = var_2609, x = value_41_cast_fp16)[name = string("op_2610_cast_fp16")]; + bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)]; + bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)]; + tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2610_cast_fp16, y = var_2608_cast_fp16)[name = string("attn_41_cast_fp16")]; + tensor var_2613 = const()[name = string("op_2613"), val = tensor([1, 1280, 1, -1])]; + tensor input_161_cast_fp16 = reshape(shape = var_2613, x = attn_41_cast_fp16)[name = string("input_161_cast_fp16")]; + string obj_83_pad_type_0 = const()[name = string("obj_83_pad_type_0"), val = string("valid")]; + tensor obj_83_strides_0 = const()[name = string("obj_83_strides_0"), val = tensor([1, 1])]; + tensor obj_83_pad_0 = const()[name = string("obj_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_83_dilations_0 = const()[name = string("obj_83_dilations_0"), val = tensor([1, 1])]; + int32 obj_83_groups_0 = const()[name = string("obj_83_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811570880)))]; + tensor layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814847744)))]; + tensor obj_83_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_83_dilations_0, groups = obj_83_groups_0, pad = obj_83_pad_0, pad_type = obj_83_pad_type_0, strides = obj_83_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_161_cast_fp16)[name = string("obj_83_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_83_cast_fp16)[name = string("inputs_83_cast_fp16")]; + tensor out_83_axes_0 = const()[name = string("out_83_axes_0"), val = tensor([1])]; + fp16 var_2631_to_fp16 = const()[name = string("op_2631_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_2631_to_fp16, x = inputs_83_cast_fp16)[name = string("out_83_cast_fp16")]; + tensor input_163_gamma_0_to_fp16 = const()[name = string("input_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814850368)))]; + tensor input_163_beta_0_to_fp16 = const()[name = string("input_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814852992)))]; + fp16 input_163_epsilon_0_to_fp16 = const()[name = string("input_163_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_163_cast_fp16 = batch_norm(beta = input_163_beta_0_to_fp16, epsilon = input_163_epsilon_0_to_fp16, gamma = input_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = string("input_163_cast_fp16")]; + string input_165_pad_type_0 = const()[name = string("input_165_pad_type_0"), val = string("valid")]; + tensor input_165_strides_0 = const()[name = string("input_165_strides_0"), val = tensor([1, 1])]; + tensor input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_165_dilations_0 = const()[name = string("input_165_dilations_0"), val = tensor([1, 1])]; + int32 input_165_groups_0 = const()[name = string("input_165_groups_0"), val = int32(1)]; + tensor layers_20_fc1_weight_to_fp16 = const()[name = string("layers_20_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814855616)))]; + tensor layers_20_fc1_bias_to_fp16 = const()[name = string("layers_20_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827962880)))]; + tensor input_165_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_165_dilations_0, groups = input_165_groups_0, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = input_165_strides_0, weight = layers_20_fc1_weight_to_fp16, x = input_163_cast_fp16)[name = string("input_165_cast_fp16")]; + string input_167_mode_0 = const()[name = string("input_167_mode_0"), val = string("EXACT")]; + tensor input_167_cast_fp16 = gelu(mode = input_167_mode_0, x = input_165_cast_fp16)[name = string("input_167_cast_fp16")]; + string hidden_states_45_pad_type_0 = const()[name = string("hidden_states_45_pad_type_0"), val = string("valid")]; + tensor hidden_states_45_strides_0 = const()[name = string("hidden_states_45_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_45_pad_0 = const()[name = string("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_45_dilations_0 = const()[name = string("hidden_states_45_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_45_groups_0 = const()[name = string("hidden_states_45_groups_0"), val = int32(1)]; + tensor layers_20_fc2_weight_to_fp16 = const()[name = string("layers_20_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827973184)))]; + tensor layers_20_fc2_bias_to_fp16 = const()[name = string("layers_20_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841080448)))]; + tensor hidden_states_45_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_20_fc2_weight_to_fp16, x = input_167_cast_fp16)[name = string("hidden_states_45_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_45_cast_fp16)[name = string("inputs_85_cast_fp16")]; + int32 var_2664 = const()[name = string("op_2664"), val = int32(3)]; + tensor out_85_axes_0 = const()[name = string("out_85_axes_0"), val = tensor([1])]; + fp16 var_2683_to_fp16 = const()[name = string("op_2683_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_2683_to_fp16, x = inputs_85_cast_fp16)[name = string("out_85_cast_fp16")]; + tensor obj_85_gamma_0_to_fp16 = const()[name = string("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841083072)))]; + tensor obj_85_beta_0_to_fp16 = const()[name = string("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841085696)))]; + fp16 obj_85_epsilon_0_to_fp16 = const()[name = string("obj_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = string("obj_85_cast_fp16")]; + string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; + tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; + tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; + int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841088320)))]; + tensor layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844365184)))]; + tensor query_43_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("query_43_cast_fp16")]; + string key_43_pad_type_0 = const()[name = string("key_43_pad_type_0"), val = string("valid")]; + tensor key_43_strides_0 = const()[name = string("key_43_strides_0"), val = tensor([1, 1])]; + tensor key_43_pad_0 = const()[name = string("key_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_43_dilations_0 = const()[name = string("key_43_dilations_0"), val = tensor([1, 1])]; + int32 key_43_groups_0 = const()[name = string("key_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844367808)))]; + tensor key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("key_43_cast_fp16")]; + string value_43_pad_type_0 = const()[name = string("value_43_pad_type_0"), val = string("valid")]; + tensor value_43_strides_0 = const()[name = string("value_43_strides_0"), val = tensor([1, 1])]; + tensor value_43_pad_0 = const()[name = string("value_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_43_dilations_0 = const()[name = string("value_43_dilations_0"), val = tensor([1, 1])]; + int32 value_43_groups_0 = const()[name = string("value_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(847644672)))]; + tensor layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850921536)))]; + tensor value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("value_43_cast_fp16")]; + tensor var_2718 = const()[name = string("op_2718"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_43_cast_fp16 = reshape(shape = var_2718, x = query_43_cast_fp16)[name = string("mh_q_43_cast_fp16")]; + fp16 var_2720_to_fp16 = const()[name = string("op_2720_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2721_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2720_to_fp16)[name = string("op_2721_cast_fp16")]; + tensor var_2722 = const()[name = string("op_2722"), val = tensor([1, 20, 64, -1])]; + tensor var_2723_cast_fp16 = reshape(shape = var_2722, x = key_43_cast_fp16)[name = string("op_2723_cast_fp16")]; + bool mh_w_43_transpose_x_0 = const()[name = string("mh_w_43_transpose_x_0"), val = bool(true)]; + bool mh_w_43_transpose_y_0 = const()[name = string("mh_w_43_transpose_y_0"), val = bool(false)]; + tensor mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_2721_cast_fp16, y = var_2723_cast_fp16)[name = string("mh_w_43_cast_fp16")]; + tensor var_2726_cast_fp16 = softmax(axis = var_2664, x = mh_w_43_cast_fp16)[name = string("op_2726_cast_fp16")]; + tensor var_2727 = const()[name = string("op_2727"), val = tensor([1, 20, 64, -1])]; + tensor var_2728_cast_fp16 = reshape(shape = var_2727, x = value_43_cast_fp16)[name = string("op_2728_cast_fp16")]; + bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)]; + bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)]; + tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2728_cast_fp16, y = var_2726_cast_fp16)[name = string("attn_43_cast_fp16")]; + tensor var_2731 = const()[name = string("op_2731"), val = tensor([1, 1280, 1, -1])]; + tensor input_169_cast_fp16 = reshape(shape = var_2731, x = attn_43_cast_fp16)[name = string("input_169_cast_fp16")]; + string obj_87_pad_type_0 = const()[name = string("obj_87_pad_type_0"), val = string("valid")]; + tensor obj_87_strides_0 = const()[name = string("obj_87_strides_0"), val = tensor([1, 1])]; + tensor obj_87_pad_0 = const()[name = string("obj_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_87_dilations_0 = const()[name = string("obj_87_dilations_0"), val = tensor([1, 1])]; + int32 obj_87_groups_0 = const()[name = string("obj_87_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850924160)))]; + tensor layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854201024)))]; + tensor obj_87_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_87_dilations_0, groups = obj_87_groups_0, pad = obj_87_pad_0, pad_type = obj_87_pad_type_0, strides = obj_87_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_169_cast_fp16)[name = string("obj_87_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_87_cast_fp16)[name = string("inputs_87_cast_fp16")]; + tensor out_87_axes_0 = const()[name = string("out_87_axes_0"), val = tensor([1])]; + fp16 var_2749_to_fp16 = const()[name = string("op_2749_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_2749_to_fp16, x = inputs_87_cast_fp16)[name = string("out_87_cast_fp16")]; + tensor input_171_gamma_0_to_fp16 = const()[name = string("input_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854203648)))]; + tensor input_171_beta_0_to_fp16 = const()[name = string("input_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854206272)))]; + fp16 input_171_epsilon_0_to_fp16 = const()[name = string("input_171_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_171_cast_fp16 = batch_norm(beta = input_171_beta_0_to_fp16, epsilon = input_171_epsilon_0_to_fp16, gamma = input_171_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = string("input_171_cast_fp16")]; + string input_173_pad_type_0 = const()[name = string("input_173_pad_type_0"), val = string("valid")]; + tensor input_173_strides_0 = const()[name = string("input_173_strides_0"), val = tensor([1, 1])]; + tensor input_173_pad_0 = const()[name = string("input_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_173_dilations_0 = const()[name = string("input_173_dilations_0"), val = tensor([1, 1])]; + int32 input_173_groups_0 = const()[name = string("input_173_groups_0"), val = int32(1)]; + tensor layers_21_fc1_weight_to_fp16 = const()[name = string("layers_21_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854208896)))]; + tensor layers_21_fc1_bias_to_fp16 = const()[name = string("layers_21_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867316160)))]; + tensor input_173_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_173_dilations_0, groups = input_173_groups_0, pad = input_173_pad_0, pad_type = input_173_pad_type_0, strides = input_173_strides_0, weight = layers_21_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = string("input_173_cast_fp16")]; + string input_175_mode_0 = const()[name = string("input_175_mode_0"), val = string("EXACT")]; + tensor input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")]; + string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")]; + tensor hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)]; + tensor layers_21_fc2_weight_to_fp16 = const()[name = string("layers_21_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867326464)))]; + tensor layers_21_fc2_bias_to_fp16 = const()[name = string("layers_21_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880433728)))]; + tensor hidden_states_47_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_21_fc2_weight_to_fp16, x = input_175_cast_fp16)[name = string("hidden_states_47_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = hidden_states_47_cast_fp16)[name = string("inputs_89_cast_fp16")]; + int32 var_2782 = const()[name = string("op_2782"), val = int32(3)]; + tensor out_89_axes_0 = const()[name = string("out_89_axes_0"), val = tensor([1])]; + fp16 var_2801_to_fp16 = const()[name = string("op_2801_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_2801_to_fp16, x = inputs_89_cast_fp16)[name = string("out_89_cast_fp16")]; + tensor obj_89_gamma_0_to_fp16 = const()[name = string("obj_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880436352)))]; + tensor obj_89_beta_0_to_fp16 = const()[name = string("obj_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880438976)))]; + fp16 obj_89_epsilon_0_to_fp16 = const()[name = string("obj_89_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_89_cast_fp16 = batch_norm(beta = obj_89_beta_0_to_fp16, epsilon = obj_89_epsilon_0_to_fp16, gamma = obj_89_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = string("obj_89_cast_fp16")]; + string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; + tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; + tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; + int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880441600)))]; + tensor layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883718464)))]; + tensor query_45_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("query_45_cast_fp16")]; + string key_45_pad_type_0 = const()[name = string("key_45_pad_type_0"), val = string("valid")]; + tensor key_45_strides_0 = const()[name = string("key_45_strides_0"), val = tensor([1, 1])]; + tensor key_45_pad_0 = const()[name = string("key_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_45_dilations_0 = const()[name = string("key_45_dilations_0"), val = tensor([1, 1])]; + int32 key_45_groups_0 = const()[name = string("key_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883721088)))]; + tensor key_45_cast_fp16 = conv(dilations = key_45_dilations_0, groups = key_45_groups_0, pad = key_45_pad_0, pad_type = key_45_pad_type_0, strides = key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("key_45_cast_fp16")]; + string value_45_pad_type_0 = const()[name = string("value_45_pad_type_0"), val = string("valid")]; + tensor value_45_strides_0 = const()[name = string("value_45_strides_0"), val = tensor([1, 1])]; + tensor value_45_pad_0 = const()[name = string("value_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_45_dilations_0 = const()[name = string("value_45_dilations_0"), val = tensor([1, 1])]; + int32 value_45_groups_0 = const()[name = string("value_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(886997952)))]; + tensor layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890274816)))]; + tensor value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = value_45_dilations_0, groups = value_45_groups_0, pad = value_45_pad_0, pad_type = value_45_pad_type_0, strides = value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("value_45_cast_fp16")]; + tensor var_2836 = const()[name = string("op_2836"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_45_cast_fp16 = reshape(shape = var_2836, x = query_45_cast_fp16)[name = string("mh_q_45_cast_fp16")]; + fp16 var_2838_to_fp16 = const()[name = string("op_2838_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2839_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2838_to_fp16)[name = string("op_2839_cast_fp16")]; + tensor var_2840 = const()[name = string("op_2840"), val = tensor([1, 20, 64, -1])]; + tensor var_2841_cast_fp16 = reshape(shape = var_2840, x = key_45_cast_fp16)[name = string("op_2841_cast_fp16")]; + bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; + bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_2839_cast_fp16, y = var_2841_cast_fp16)[name = string("mh_w_45_cast_fp16")]; + tensor var_2844_cast_fp16 = softmax(axis = var_2782, x = mh_w_45_cast_fp16)[name = string("op_2844_cast_fp16")]; + tensor var_2845 = const()[name = string("op_2845"), val = tensor([1, 20, 64, -1])]; + tensor var_2846_cast_fp16 = reshape(shape = var_2845, x = value_45_cast_fp16)[name = string("op_2846_cast_fp16")]; + bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)]; + bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)]; + tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2846_cast_fp16, y = var_2844_cast_fp16)[name = string("attn_45_cast_fp16")]; + tensor var_2849 = const()[name = string("op_2849"), val = tensor([1, 1280, 1, -1])]; + tensor input_177_cast_fp16 = reshape(shape = var_2849, x = attn_45_cast_fp16)[name = string("input_177_cast_fp16")]; + string obj_91_pad_type_0 = const()[name = string("obj_91_pad_type_0"), val = string("valid")]; + tensor obj_91_strides_0 = const()[name = string("obj_91_strides_0"), val = tensor([1, 1])]; + tensor obj_91_pad_0 = const()[name = string("obj_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_91_dilations_0 = const()[name = string("obj_91_dilations_0"), val = tensor([1, 1])]; + int32 obj_91_groups_0 = const()[name = string("obj_91_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890277440)))]; + tensor layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893554304)))]; + tensor obj_91_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_177_cast_fp16)[name = string("obj_91_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = obj_91_cast_fp16)[name = string("inputs_91_cast_fp16")]; + tensor out_91_axes_0 = const()[name = string("out_91_axes_0"), val = tensor([1])]; + fp16 var_2867_to_fp16 = const()[name = string("op_2867_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_2867_to_fp16, x = inputs_91_cast_fp16)[name = string("out_91_cast_fp16")]; + tensor input_179_gamma_0_to_fp16 = const()[name = string("input_179_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893556928)))]; + tensor input_179_beta_0_to_fp16 = const()[name = string("input_179_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893559552)))]; + fp16 input_179_epsilon_0_to_fp16 = const()[name = string("input_179_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_179_cast_fp16 = batch_norm(beta = input_179_beta_0_to_fp16, epsilon = input_179_epsilon_0_to_fp16, gamma = input_179_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = string("input_179_cast_fp16")]; + string input_181_pad_type_0 = const()[name = string("input_181_pad_type_0"), val = string("valid")]; + tensor input_181_strides_0 = const()[name = string("input_181_strides_0"), val = tensor([1, 1])]; + tensor input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_181_dilations_0 = const()[name = string("input_181_dilations_0"), val = tensor([1, 1])]; + int32 input_181_groups_0 = const()[name = string("input_181_groups_0"), val = int32(1)]; + tensor layers_22_fc1_weight_to_fp16 = const()[name = string("layers_22_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893562176)))]; + tensor layers_22_fc1_bias_to_fp16 = const()[name = string("layers_22_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906669440)))]; + tensor input_181_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_181_dilations_0, groups = input_181_groups_0, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = input_181_strides_0, weight = layers_22_fc1_weight_to_fp16, x = input_179_cast_fp16)[name = string("input_181_cast_fp16")]; + string input_183_mode_0 = const()[name = string("input_183_mode_0"), val = string("EXACT")]; + tensor input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = input_181_cast_fp16)[name = string("input_183_cast_fp16")]; + string hidden_states_49_pad_type_0 = const()[name = string("hidden_states_49_pad_type_0"), val = string("valid")]; + tensor hidden_states_49_strides_0 = const()[name = string("hidden_states_49_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_49_pad_0 = const()[name = string("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_49_dilations_0 = const()[name = string("hidden_states_49_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_49_groups_0 = const()[name = string("hidden_states_49_groups_0"), val = int32(1)]; + tensor layers_22_fc2_weight_to_fp16 = const()[name = string("layers_22_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906679744)))]; + tensor layers_22_fc2_bias_to_fp16 = const()[name = string("layers_22_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919787008)))]; + tensor hidden_states_49_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_22_fc2_weight_to_fp16, x = input_183_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("inputs_93_cast_fp16")]; + int32 var_2900 = const()[name = string("op_2900"), val = int32(3)]; + tensor out_93_axes_0 = const()[name = string("out_93_axes_0"), val = tensor([1])]; + fp16 var_2919_to_fp16 = const()[name = string("op_2919_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_2919_to_fp16, x = inputs_93_cast_fp16)[name = string("out_93_cast_fp16")]; + tensor obj_93_gamma_0_to_fp16 = const()[name = string("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919789632)))]; + tensor obj_93_beta_0_to_fp16 = const()[name = string("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919792256)))]; + fp16 obj_93_epsilon_0_to_fp16 = const()[name = string("obj_93_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = string("obj_93_cast_fp16")]; + string query_47_pad_type_0 = const()[name = string("query_47_pad_type_0"), val = string("valid")]; + tensor query_47_strides_0 = const()[name = string("query_47_strides_0"), val = tensor([1, 1])]; + tensor query_47_pad_0 = const()[name = string("query_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_47_dilations_0 = const()[name = string("query_47_dilations_0"), val = tensor([1, 1])]; + int32 query_47_groups_0 = const()[name = string("query_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919794880)))]; + tensor layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923071744)))]; + tensor query_47_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("query_47_cast_fp16")]; + string key_47_pad_type_0 = const()[name = string("key_47_pad_type_0"), val = string("valid")]; + tensor key_47_strides_0 = const()[name = string("key_47_strides_0"), val = tensor([1, 1])]; + tensor key_47_pad_0 = const()[name = string("key_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_47_dilations_0 = const()[name = string("key_47_dilations_0"), val = tensor([1, 1])]; + int32 key_47_groups_0 = const()[name = string("key_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923074368)))]; + tensor key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("key_47_cast_fp16")]; + string value_47_pad_type_0 = const()[name = string("value_47_pad_type_0"), val = string("valid")]; + tensor value_47_strides_0 = const()[name = string("value_47_strides_0"), val = tensor([1, 1])]; + tensor value_47_pad_0 = const()[name = string("value_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_47_dilations_0 = const()[name = string("value_47_dilations_0"), val = tensor([1, 1])]; + int32 value_47_groups_0 = const()[name = string("value_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(926351232)))]; + tensor layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929628096)))]; + tensor value_47_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("value_47_cast_fp16")]; + tensor var_2954 = const()[name = string("op_2954"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_47_cast_fp16 = reshape(shape = var_2954, x = query_47_cast_fp16)[name = string("mh_q_47_cast_fp16")]; + fp16 var_2956_to_fp16 = const()[name = string("op_2956_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2957_cast_fp16 = mul(x = mh_q_47_cast_fp16, y = var_2956_to_fp16)[name = string("op_2957_cast_fp16")]; + tensor var_2958 = const()[name = string("op_2958"), val = tensor([1, 20, 64, -1])]; + tensor var_2959_cast_fp16 = reshape(shape = var_2958, x = key_47_cast_fp16)[name = string("op_2959_cast_fp16")]; + bool mh_w_47_transpose_x_0 = const()[name = string("mh_w_47_transpose_x_0"), val = bool(true)]; + bool mh_w_47_transpose_y_0 = const()[name = string("mh_w_47_transpose_y_0"), val = bool(false)]; + tensor mh_w_47_cast_fp16 = matmul(transpose_x = mh_w_47_transpose_x_0, transpose_y = mh_w_47_transpose_y_0, x = var_2957_cast_fp16, y = var_2959_cast_fp16)[name = string("mh_w_47_cast_fp16")]; + tensor var_2962_cast_fp16 = softmax(axis = var_2900, x = mh_w_47_cast_fp16)[name = string("op_2962_cast_fp16")]; + tensor var_2963 = const()[name = string("op_2963"), val = tensor([1, 20, 64, -1])]; + tensor var_2964_cast_fp16 = reshape(shape = var_2963, x = value_47_cast_fp16)[name = string("op_2964_cast_fp16")]; + bool attn_47_transpose_x_0 = const()[name = string("attn_47_transpose_x_0"), val = bool(false)]; + bool attn_47_transpose_y_0 = const()[name = string("attn_47_transpose_y_0"), val = bool(true)]; + tensor attn_47_cast_fp16 = matmul(transpose_x = attn_47_transpose_x_0, transpose_y = attn_47_transpose_y_0, x = var_2964_cast_fp16, y = var_2962_cast_fp16)[name = string("attn_47_cast_fp16")]; + tensor var_2967 = const()[name = string("op_2967"), val = tensor([1, 1280, 1, -1])]; + tensor input_185_cast_fp16 = reshape(shape = var_2967, x = attn_47_cast_fp16)[name = string("input_185_cast_fp16")]; + string obj_95_pad_type_0 = const()[name = string("obj_95_pad_type_0"), val = string("valid")]; + tensor obj_95_strides_0 = const()[name = string("obj_95_strides_0"), val = tensor([1, 1])]; + tensor obj_95_pad_0 = const()[name = string("obj_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_95_dilations_0 = const()[name = string("obj_95_dilations_0"), val = tensor([1, 1])]; + int32 obj_95_groups_0 = const()[name = string("obj_95_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929630720)))]; + tensor layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932907584)))]; + tensor obj_95_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_95_cast_fp16)[name = string("inputs_95_cast_fp16")]; + tensor out_95_axes_0 = const()[name = string("out_95_axes_0"), val = tensor([1])]; + fp16 var_2985_to_fp16 = const()[name = string("op_2985_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_2985_to_fp16, x = inputs_95_cast_fp16)[name = string("out_95_cast_fp16")]; + tensor input_187_gamma_0_to_fp16 = const()[name = string("input_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932910208)))]; + tensor input_187_beta_0_to_fp16 = const()[name = string("input_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932912832)))]; + fp16 input_187_epsilon_0_to_fp16 = const()[name = string("input_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_187_cast_fp16 = batch_norm(beta = input_187_beta_0_to_fp16, epsilon = input_187_epsilon_0_to_fp16, gamma = input_187_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = string("input_187_cast_fp16")]; + string input_189_pad_type_0 = const()[name = string("input_189_pad_type_0"), val = string("valid")]; + tensor input_189_strides_0 = const()[name = string("input_189_strides_0"), val = tensor([1, 1])]; + tensor input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_189_dilations_0 = const()[name = string("input_189_dilations_0"), val = tensor([1, 1])]; + int32 input_189_groups_0 = const()[name = string("input_189_groups_0"), val = int32(1)]; + tensor layers_23_fc1_weight_to_fp16 = const()[name = string("layers_23_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932915456)))]; + tensor layers_23_fc1_bias_to_fp16 = const()[name = string("layers_23_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946022720)))]; + tensor input_189_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = layers_23_fc1_weight_to_fp16, x = input_187_cast_fp16)[name = string("input_189_cast_fp16")]; + string input_191_mode_0 = const()[name = string("input_191_mode_0"), val = string("EXACT")]; + tensor input_191_cast_fp16 = gelu(mode = input_191_mode_0, x = input_189_cast_fp16)[name = string("input_191_cast_fp16")]; + string hidden_states_51_pad_type_0 = const()[name = string("hidden_states_51_pad_type_0"), val = string("valid")]; + tensor hidden_states_51_strides_0 = const()[name = string("hidden_states_51_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_51_pad_0 = const()[name = string("hidden_states_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_51_dilations_0 = const()[name = string("hidden_states_51_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_51_groups_0 = const()[name = string("hidden_states_51_groups_0"), val = int32(1)]; + tensor layers_23_fc2_weight_to_fp16 = const()[name = string("layers_23_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946033024)))]; + tensor layers_23_fc2_bias_to_fp16 = const()[name = string("layers_23_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959140288)))]; + tensor hidden_states_51_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_51_dilations_0, groups = hidden_states_51_groups_0, pad = hidden_states_51_pad_0, pad_type = hidden_states_51_pad_type_0, strides = hidden_states_51_strides_0, weight = layers_23_fc2_weight_to_fp16, x = input_191_cast_fp16)[name = string("hidden_states_51_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_51_cast_fp16)[name = string("inputs_97_cast_fp16")]; + int32 var_3018 = const()[name = string("op_3018"), val = int32(3)]; + tensor out_97_axes_0 = const()[name = string("out_97_axes_0"), val = tensor([1])]; + fp16 var_3037_to_fp16 = const()[name = string("op_3037_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3037_to_fp16, x = inputs_97_cast_fp16)[name = string("out_97_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959142912)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959145536)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_49_pad_type_0 = const()[name = string("query_49_pad_type_0"), val = string("valid")]; + tensor query_49_strides_0 = const()[name = string("query_49_strides_0"), val = tensor([1, 1])]; + tensor query_49_pad_0 = const()[name = string("query_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_49_dilations_0 = const()[name = string("query_49_dilations_0"), val = tensor([1, 1])]; + int32 query_49_groups_0 = const()[name = string("query_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959148160)))]; + tensor layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962425024)))]; + tensor query_49_cast_fp16 = conv(bias = layers_24_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_24_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_49_cast_fp16")]; + string key_49_pad_type_0 = const()[name = string("key_49_pad_type_0"), val = string("valid")]; + tensor key_49_strides_0 = const()[name = string("key_49_strides_0"), val = tensor([1, 1])]; + tensor key_49_pad_0 = const()[name = string("key_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_49_dilations_0 = const()[name = string("key_49_dilations_0"), val = tensor([1, 1])]; + int32 key_49_groups_0 = const()[name = string("key_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962427648)))]; + tensor key_49_cast_fp16 = conv(dilations = key_49_dilations_0, groups = key_49_groups_0, pad = key_49_pad_0, pad_type = key_49_pad_type_0, strides = key_49_strides_0, weight = layers_24_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("key_49_cast_fp16")]; + string value_49_pad_type_0 = const()[name = string("value_49_pad_type_0"), val = string("valid")]; + tensor value_49_strides_0 = const()[name = string("value_49_strides_0"), val = tensor([1, 1])]; + tensor value_49_pad_0 = const()[name = string("value_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_49_dilations_0 = const()[name = string("value_49_dilations_0"), val = tensor([1, 1])]; + int32 value_49_groups_0 = const()[name = string("value_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(965704512)))]; + tensor layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968981376)))]; + tensor value_49_cast_fp16 = conv(bias = layers_24_self_attn_v_proj_bias_to_fp16, dilations = value_49_dilations_0, groups = value_49_groups_0, pad = value_49_pad_0, pad_type = value_49_pad_type_0, strides = value_49_strides_0, weight = layers_24_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("value_49_cast_fp16")]; + tensor var_3072 = const()[name = string("op_3072"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_49_cast_fp16 = reshape(shape = var_3072, x = query_49_cast_fp16)[name = string("mh_q_49_cast_fp16")]; + fp16 var_3074_to_fp16 = const()[name = string("op_3074_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3075_cast_fp16 = mul(x = mh_q_49_cast_fp16, y = var_3074_to_fp16)[name = string("op_3075_cast_fp16")]; + tensor var_3076 = const()[name = string("op_3076"), val = tensor([1, 20, 64, -1])]; + tensor var_3077_cast_fp16 = reshape(shape = var_3076, x = key_49_cast_fp16)[name = string("op_3077_cast_fp16")]; + bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)]; + bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)]; + tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_3075_cast_fp16, y = var_3077_cast_fp16)[name = string("mh_w_49_cast_fp16")]; + tensor var_3080_cast_fp16 = softmax(axis = var_3018, x = mh_w_49_cast_fp16)[name = string("op_3080_cast_fp16")]; + tensor var_3081 = const()[name = string("op_3081"), val = tensor([1, 20, 64, -1])]; + tensor var_3082_cast_fp16 = reshape(shape = var_3081, x = value_49_cast_fp16)[name = string("op_3082_cast_fp16")]; + bool attn_49_transpose_x_0 = const()[name = string("attn_49_transpose_x_0"), val = bool(false)]; + bool attn_49_transpose_y_0 = const()[name = string("attn_49_transpose_y_0"), val = bool(true)]; + tensor attn_49_cast_fp16 = matmul(transpose_x = attn_49_transpose_x_0, transpose_y = attn_49_transpose_y_0, x = var_3082_cast_fp16, y = var_3080_cast_fp16)[name = string("attn_49_cast_fp16")]; + tensor var_3085 = const()[name = string("op_3085"), val = tensor([1, 1280, 1, -1])]; + tensor input_193_cast_fp16 = reshape(shape = var_3085, x = attn_49_cast_fp16)[name = string("input_193_cast_fp16")]; + string obj_99_pad_type_0 = const()[name = string("obj_99_pad_type_0"), val = string("valid")]; + tensor obj_99_strides_0 = const()[name = string("obj_99_strides_0"), val = tensor([1, 1])]; + tensor obj_99_pad_0 = const()[name = string("obj_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_99_dilations_0 = const()[name = string("obj_99_dilations_0"), val = tensor([1, 1])]; + int32 obj_99_groups_0 = const()[name = string("obj_99_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968984000)))]; + tensor layers_24_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972260864)))]; + tensor obj_99_cast_fp16 = conv(bias = layers_24_self_attn_o_proj_bias_to_fp16, dilations = obj_99_dilations_0, groups = obj_99_groups_0, pad = obj_99_pad_0, pad_type = obj_99_pad_type_0, strides = obj_99_strides_0, weight = layers_24_self_attn_o_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("obj_99_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_99_cast_fp16)[name = string("inputs_99_cast_fp16")]; + tensor out_99_axes_0 = const()[name = string("out_99_axes_0"), val = tensor([1])]; + fp16 var_3103_to_fp16 = const()[name = string("op_3103_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3103_to_fp16, x = inputs_99_cast_fp16)[name = string("out_99_cast_fp16")]; + tensor input_195_gamma_0_to_fp16 = const()[name = string("input_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972263488)))]; + tensor input_195_beta_0_to_fp16 = const()[name = string("input_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972266112)))]; + fp16 input_195_epsilon_0_to_fp16 = const()[name = string("input_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = string("input_195_cast_fp16")]; + string input_197_pad_type_0 = const()[name = string("input_197_pad_type_0"), val = string("valid")]; + tensor input_197_strides_0 = const()[name = string("input_197_strides_0"), val = tensor([1, 1])]; + tensor input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_197_dilations_0 = const()[name = string("input_197_dilations_0"), val = tensor([1, 1])]; + int32 input_197_groups_0 = const()[name = string("input_197_groups_0"), val = int32(1)]; + tensor layers_24_fc1_weight_to_fp16 = const()[name = string("layers_24_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972268736)))]; + tensor layers_24_fc1_bias_to_fp16 = const()[name = string("layers_24_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985376000)))]; + tensor input_197_cast_fp16 = conv(bias = layers_24_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_24_fc1_weight_to_fp16, x = input_195_cast_fp16)[name = string("input_197_cast_fp16")]; + string input_199_mode_0 = const()[name = string("input_199_mode_0"), val = string("EXACT")]; + tensor input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")]; + string hidden_states_53_pad_type_0 = const()[name = string("hidden_states_53_pad_type_0"), val = string("valid")]; + tensor hidden_states_53_strides_0 = const()[name = string("hidden_states_53_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_53_pad_0 = const()[name = string("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_53_dilations_0 = const()[name = string("hidden_states_53_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_53_groups_0 = const()[name = string("hidden_states_53_groups_0"), val = int32(1)]; + tensor layers_24_fc2_weight_to_fp16 = const()[name = string("layers_24_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985386304)))]; + tensor layers_24_fc2_bias_to_fp16 = const()[name = string("layers_24_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998493568)))]; + tensor hidden_states_53_cast_fp16 = conv(bias = layers_24_fc2_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = layers_24_fc2_weight_to_fp16, x = input_199_cast_fp16)[name = string("hidden_states_53_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = hidden_states_53_cast_fp16)[name = string("inputs_101_cast_fp16")]; + int32 var_3136 = const()[name = string("op_3136"), val = int32(3)]; + tensor out_101_axes_0 = const()[name = string("out_101_axes_0"), val = tensor([1])]; + fp16 var_3155_to_fp16 = const()[name = string("op_3155_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3155_to_fp16, x = inputs_101_cast_fp16)[name = string("out_101_cast_fp16")]; + tensor obj_101_gamma_0_to_fp16 = const()[name = string("obj_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998496192)))]; + tensor obj_101_beta_0_to_fp16 = const()[name = string("obj_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998498816)))]; + fp16 obj_101_epsilon_0_to_fp16 = const()[name = string("obj_101_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_101_cast_fp16 = batch_norm(beta = obj_101_beta_0_to_fp16, epsilon = obj_101_epsilon_0_to_fp16, gamma = obj_101_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = string("obj_101_cast_fp16")]; + string query_51_pad_type_0 = const()[name = string("query_51_pad_type_0"), val = string("valid")]; + tensor query_51_strides_0 = const()[name = string("query_51_strides_0"), val = tensor([1, 1])]; + tensor query_51_pad_0 = const()[name = string("query_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_51_dilations_0 = const()[name = string("query_51_dilations_0"), val = tensor([1, 1])]; + int32 query_51_groups_0 = const()[name = string("query_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998501440)))]; + tensor layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001778304)))]; + tensor query_51_cast_fp16 = conv(bias = layers_25_self_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_25_self_attn_q_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("query_51_cast_fp16")]; + string key_51_pad_type_0 = const()[name = string("key_51_pad_type_0"), val = string("valid")]; + tensor key_51_strides_0 = const()[name = string("key_51_strides_0"), val = tensor([1, 1])]; + tensor key_51_pad_0 = const()[name = string("key_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_51_dilations_0 = const()[name = string("key_51_dilations_0"), val = tensor([1, 1])]; + int32 key_51_groups_0 = const()[name = string("key_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001780928)))]; + tensor key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_25_self_attn_k_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("key_51_cast_fp16")]; + string value_51_pad_type_0 = const()[name = string("value_51_pad_type_0"), val = string("valid")]; + tensor value_51_strides_0 = const()[name = string("value_51_strides_0"), val = tensor([1, 1])]; + tensor value_51_pad_0 = const()[name = string("value_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_51_dilations_0 = const()[name = string("value_51_dilations_0"), val = tensor([1, 1])]; + int32 value_51_groups_0 = const()[name = string("value_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1005057792)))]; + tensor layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008334656)))]; + tensor value_51_cast_fp16 = conv(bias = layers_25_self_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_25_self_attn_v_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("value_51_cast_fp16")]; + tensor var_3190 = const()[name = string("op_3190"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_51_cast_fp16 = reshape(shape = var_3190, x = query_51_cast_fp16)[name = string("mh_q_51_cast_fp16")]; + fp16 var_3192_to_fp16 = const()[name = string("op_3192_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3193_cast_fp16 = mul(x = mh_q_51_cast_fp16, y = var_3192_to_fp16)[name = string("op_3193_cast_fp16")]; + tensor var_3194 = const()[name = string("op_3194"), val = tensor([1, 20, 64, -1])]; + tensor var_3195_cast_fp16 = reshape(shape = var_3194, x = key_51_cast_fp16)[name = string("op_3195_cast_fp16")]; + bool mh_w_51_transpose_x_0 = const()[name = string("mh_w_51_transpose_x_0"), val = bool(true)]; + bool mh_w_51_transpose_y_0 = const()[name = string("mh_w_51_transpose_y_0"), val = bool(false)]; + tensor mh_w_51_cast_fp16 = matmul(transpose_x = mh_w_51_transpose_x_0, transpose_y = mh_w_51_transpose_y_0, x = var_3193_cast_fp16, y = var_3195_cast_fp16)[name = string("mh_w_51_cast_fp16")]; + tensor var_3198_cast_fp16 = softmax(axis = var_3136, x = mh_w_51_cast_fp16)[name = string("op_3198_cast_fp16")]; + tensor var_3199 = const()[name = string("op_3199"), val = tensor([1, 20, 64, -1])]; + tensor var_3200_cast_fp16 = reshape(shape = var_3199, x = value_51_cast_fp16)[name = string("op_3200_cast_fp16")]; + bool attn_51_transpose_x_0 = const()[name = string("attn_51_transpose_x_0"), val = bool(false)]; + bool attn_51_transpose_y_0 = const()[name = string("attn_51_transpose_y_0"), val = bool(true)]; + tensor attn_51_cast_fp16 = matmul(transpose_x = attn_51_transpose_x_0, transpose_y = attn_51_transpose_y_0, x = var_3200_cast_fp16, y = var_3198_cast_fp16)[name = string("attn_51_cast_fp16")]; + tensor var_3203 = const()[name = string("op_3203"), val = tensor([1, 1280, 1, -1])]; + tensor input_201_cast_fp16 = reshape(shape = var_3203, x = attn_51_cast_fp16)[name = string("input_201_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008337280)))]; + tensor layers_25_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011614144)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_25_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_25_self_attn_o_proj_weight_to_fp16, x = input_201_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_103_cast_fp16")]; + tensor out_103_axes_0 = const()[name = string("out_103_axes_0"), val = tensor([1])]; + fp16 var_3221_to_fp16 = const()[name = string("op_3221_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_3221_to_fp16, x = inputs_103_cast_fp16)[name = string("out_103_cast_fp16")]; + tensor input_203_gamma_0_to_fp16 = const()[name = string("input_203_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011616768)))]; + tensor input_203_beta_0_to_fp16 = const()[name = string("input_203_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011619392)))]; + fp16 input_203_epsilon_0_to_fp16 = const()[name = string("input_203_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_203_cast_fp16 = batch_norm(beta = input_203_beta_0_to_fp16, epsilon = input_203_epsilon_0_to_fp16, gamma = input_203_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = string("input_203_cast_fp16")]; + string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")]; + tensor input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor([1, 1])]; + tensor input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor([1, 1])]; + int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1)]; + tensor layers_25_fc1_weight_to_fp16 = const()[name = string("layers_25_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011622016)))]; + tensor layers_25_fc1_bias_to_fp16 = const()[name = string("layers_25_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024729280)))]; + tensor input_205_cast_fp16 = conv(bias = layers_25_fc1_bias_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = layers_25_fc1_weight_to_fp16, x = input_203_cast_fp16)[name = string("input_205_cast_fp16")]; + string input_207_mode_0 = const()[name = string("input_207_mode_0"), val = string("EXACT")]; + tensor input_207_cast_fp16 = gelu(mode = input_207_mode_0, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")]; + string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")]; + tensor hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)]; + tensor layers_25_fc2_weight_to_fp16 = const()[name = string("layers_25_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024739584)))]; + tensor layers_25_fc2_bias_to_fp16 = const()[name = string("layers_25_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037846848)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = layers_25_fc2_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = layers_25_fc2_weight_to_fp16, x = input_207_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = hidden_states_55_cast_fp16)[name = string("inputs_105_cast_fp16")]; + int32 var_3254 = const()[name = string("op_3254"), val = int32(3)]; + tensor out_105_axes_0 = const()[name = string("out_105_axes_0"), val = tensor([1])]; + fp16 var_3273_to_fp16 = const()[name = string("op_3273_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_3273_to_fp16, x = inputs_105_cast_fp16)[name = string("out_105_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037849472)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037852096)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_53_pad_type_0 = const()[name = string("query_53_pad_type_0"), val = string("valid")]; + tensor query_53_strides_0 = const()[name = string("query_53_strides_0"), val = tensor([1, 1])]; + tensor query_53_pad_0 = const()[name = string("query_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_53_dilations_0 = const()[name = string("query_53_dilations_0"), val = tensor([1, 1])]; + int32 query_53_groups_0 = const()[name = string("query_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037854720)))]; + tensor layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041131584)))]; + tensor query_53_cast_fp16 = conv(bias = layers_26_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_26_self_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_53_cast_fp16")]; + string key_53_pad_type_0 = const()[name = string("key_53_pad_type_0"), val = string("valid")]; + tensor key_53_strides_0 = const()[name = string("key_53_strides_0"), val = tensor([1, 1])]; + tensor key_53_pad_0 = const()[name = string("key_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_53_dilations_0 = const()[name = string("key_53_dilations_0"), val = tensor([1, 1])]; + int32 key_53_groups_0 = const()[name = string("key_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041134208)))]; + tensor key_53_cast_fp16 = conv(dilations = key_53_dilations_0, groups = key_53_groups_0, pad = key_53_pad_0, pad_type = key_53_pad_type_0, strides = key_53_strides_0, weight = layers_26_self_attn_k_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("key_53_cast_fp16")]; + string value_53_pad_type_0 = const()[name = string("value_53_pad_type_0"), val = string("valid")]; + tensor value_53_strides_0 = const()[name = string("value_53_strides_0"), val = tensor([1, 1])]; + tensor value_53_pad_0 = const()[name = string("value_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_53_dilations_0 = const()[name = string("value_53_dilations_0"), val = tensor([1, 1])]; + int32 value_53_groups_0 = const()[name = string("value_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1044411072)))]; + tensor layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047687936)))]; + tensor value_53_cast_fp16 = conv(bias = layers_26_self_attn_v_proj_bias_to_fp16, dilations = value_53_dilations_0, groups = value_53_groups_0, pad = value_53_pad_0, pad_type = value_53_pad_type_0, strides = value_53_strides_0, weight = layers_26_self_attn_v_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("value_53_cast_fp16")]; + tensor var_3308 = const()[name = string("op_3308"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_53_cast_fp16 = reshape(shape = var_3308, x = query_53_cast_fp16)[name = string("mh_q_53_cast_fp16")]; + fp16 var_3310_to_fp16 = const()[name = string("op_3310_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3311_cast_fp16 = mul(x = mh_q_53_cast_fp16, y = var_3310_to_fp16)[name = string("op_3311_cast_fp16")]; + tensor var_3312 = const()[name = string("op_3312"), val = tensor([1, 20, 64, -1])]; + tensor var_3313_cast_fp16 = reshape(shape = var_3312, x = key_53_cast_fp16)[name = string("op_3313_cast_fp16")]; + bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)]; + bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)]; + tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_3311_cast_fp16, y = var_3313_cast_fp16)[name = string("mh_w_53_cast_fp16")]; + tensor var_3316_cast_fp16 = softmax(axis = var_3254, x = mh_w_53_cast_fp16)[name = string("op_3316_cast_fp16")]; + tensor var_3317 = const()[name = string("op_3317"), val = tensor([1, 20, 64, -1])]; + tensor var_3318_cast_fp16 = reshape(shape = var_3317, x = value_53_cast_fp16)[name = string("op_3318_cast_fp16")]; + bool attn_53_transpose_x_0 = const()[name = string("attn_53_transpose_x_0"), val = bool(false)]; + bool attn_53_transpose_y_0 = const()[name = string("attn_53_transpose_y_0"), val = bool(true)]; + tensor attn_53_cast_fp16 = matmul(transpose_x = attn_53_transpose_x_0, transpose_y = attn_53_transpose_y_0, x = var_3318_cast_fp16, y = var_3316_cast_fp16)[name = string("attn_53_cast_fp16")]; + tensor var_3321 = const()[name = string("op_3321"), val = tensor([1, 1280, 1, -1])]; + tensor input_209_cast_fp16 = reshape(shape = var_3321, x = attn_53_cast_fp16)[name = string("input_209_cast_fp16")]; + string obj_107_pad_type_0 = const()[name = string("obj_107_pad_type_0"), val = string("valid")]; + tensor obj_107_strides_0 = const()[name = string("obj_107_strides_0"), val = tensor([1, 1])]; + tensor obj_107_pad_0 = const()[name = string("obj_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_107_dilations_0 = const()[name = string("obj_107_dilations_0"), val = tensor([1, 1])]; + int32 obj_107_groups_0 = const()[name = string("obj_107_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047690560)))]; + tensor layers_26_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050967424)))]; + tensor obj_107_cast_fp16 = conv(bias = layers_26_self_attn_o_proj_bias_to_fp16, dilations = obj_107_dilations_0, groups = obj_107_groups_0, pad = obj_107_pad_0, pad_type = obj_107_pad_type_0, strides = obj_107_strides_0, weight = layers_26_self_attn_o_proj_weight_to_fp16, x = input_209_cast_fp16)[name = string("obj_107_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_107_cast_fp16)[name = string("inputs_107_cast_fp16")]; + tensor out_107_axes_0 = const()[name = string("out_107_axes_0"), val = tensor([1])]; + fp16 var_3339_to_fp16 = const()[name = string("op_3339_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_3339_to_fp16, x = inputs_107_cast_fp16)[name = string("out_107_cast_fp16")]; + tensor input_211_gamma_0_to_fp16 = const()[name = string("input_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050970048)))]; + tensor input_211_beta_0_to_fp16 = const()[name = string("input_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050972672)))]; + fp16 input_211_epsilon_0_to_fp16 = const()[name = string("input_211_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_211_cast_fp16 = batch_norm(beta = input_211_beta_0_to_fp16, epsilon = input_211_epsilon_0_to_fp16, gamma = input_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = string("input_211_cast_fp16")]; + string input_213_pad_type_0 = const()[name = string("input_213_pad_type_0"), val = string("valid")]; + tensor input_213_strides_0 = const()[name = string("input_213_strides_0"), val = tensor([1, 1])]; + tensor input_213_pad_0 = const()[name = string("input_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_213_dilations_0 = const()[name = string("input_213_dilations_0"), val = tensor([1, 1])]; + int32 input_213_groups_0 = const()[name = string("input_213_groups_0"), val = int32(1)]; + tensor layers_26_fc1_weight_to_fp16 = const()[name = string("layers_26_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050975296)))]; + tensor layers_26_fc1_bias_to_fp16 = const()[name = string("layers_26_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064082560)))]; + tensor input_213_cast_fp16 = conv(bias = layers_26_fc1_bias_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = layers_26_fc1_weight_to_fp16, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")]; + string input_215_mode_0 = const()[name = string("input_215_mode_0"), val = string("EXACT")]; + tensor input_215_cast_fp16 = gelu(mode = input_215_mode_0, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")]; + string hidden_states_57_pad_type_0 = const()[name = string("hidden_states_57_pad_type_0"), val = string("valid")]; + tensor hidden_states_57_strides_0 = const()[name = string("hidden_states_57_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_57_pad_0 = const()[name = string("hidden_states_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_57_dilations_0 = const()[name = string("hidden_states_57_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_57_groups_0 = const()[name = string("hidden_states_57_groups_0"), val = int32(1)]; + tensor layers_26_fc2_weight_to_fp16 = const()[name = string("layers_26_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064092864)))]; + tensor layers_26_fc2_bias_to_fp16 = const()[name = string("layers_26_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077200128)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = layers_26_fc2_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = layers_26_fc2_weight_to_fp16, x = input_215_cast_fp16)[name = string("hidden_states_57_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_57_cast_fp16)[name = string("inputs_109_cast_fp16")]; + int32 var_3372 = const()[name = string("op_3372"), val = int32(3)]; + tensor out_109_axes_0 = const()[name = string("out_109_axes_0"), val = tensor([1])]; + fp16 var_3391_to_fp16 = const()[name = string("op_3391_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_3391_to_fp16, x = inputs_109_cast_fp16)[name = string("out_109_cast_fp16")]; + tensor obj_109_gamma_0_to_fp16 = const()[name = string("obj_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077202752)))]; + tensor obj_109_beta_0_to_fp16 = const()[name = string("obj_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077205376)))]; + fp16 obj_109_epsilon_0_to_fp16 = const()[name = string("obj_109_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_109_cast_fp16 = batch_norm(beta = obj_109_beta_0_to_fp16, epsilon = obj_109_epsilon_0_to_fp16, gamma = obj_109_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = string("obj_109_cast_fp16")]; + string query_55_pad_type_0 = const()[name = string("query_55_pad_type_0"), val = string("valid")]; + tensor query_55_strides_0 = const()[name = string("query_55_strides_0"), val = tensor([1, 1])]; + tensor query_55_pad_0 = const()[name = string("query_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_55_dilations_0 = const()[name = string("query_55_dilations_0"), val = tensor([1, 1])]; + int32 query_55_groups_0 = const()[name = string("query_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077208000)))]; + tensor layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080484864)))]; + tensor query_55_cast_fp16 = conv(bias = layers_27_self_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_27_self_attn_q_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("query_55_cast_fp16")]; + string key_55_pad_type_0 = const()[name = string("key_55_pad_type_0"), val = string("valid")]; + tensor key_55_strides_0 = const()[name = string("key_55_strides_0"), val = tensor([1, 1])]; + tensor key_55_pad_0 = const()[name = string("key_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_55_dilations_0 = const()[name = string("key_55_dilations_0"), val = tensor([1, 1])]; + int32 key_55_groups_0 = const()[name = string("key_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080487488)))]; + tensor key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_27_self_attn_k_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("key_55_cast_fp16")]; + string value_55_pad_type_0 = const()[name = string("value_55_pad_type_0"), val = string("valid")]; + tensor value_55_strides_0 = const()[name = string("value_55_strides_0"), val = tensor([1, 1])]; + tensor value_55_pad_0 = const()[name = string("value_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_55_dilations_0 = const()[name = string("value_55_dilations_0"), val = tensor([1, 1])]; + int32 value_55_groups_0 = const()[name = string("value_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1083764352)))]; + tensor layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087041216)))]; + tensor value_55_cast_fp16 = conv(bias = layers_27_self_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_27_self_attn_v_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("value_55_cast_fp16")]; + tensor var_3426 = const()[name = string("op_3426"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_55_cast_fp16 = reshape(shape = var_3426, x = query_55_cast_fp16)[name = string("mh_q_55_cast_fp16")]; + fp16 var_3428_to_fp16 = const()[name = string("op_3428_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3429_cast_fp16 = mul(x = mh_q_55_cast_fp16, y = var_3428_to_fp16)[name = string("op_3429_cast_fp16")]; + tensor var_3430 = const()[name = string("op_3430"), val = tensor([1, 20, 64, -1])]; + tensor var_3431_cast_fp16 = reshape(shape = var_3430, x = key_55_cast_fp16)[name = string("op_3431_cast_fp16")]; + bool mh_w_55_transpose_x_0 = const()[name = string("mh_w_55_transpose_x_0"), val = bool(true)]; + bool mh_w_55_transpose_y_0 = const()[name = string("mh_w_55_transpose_y_0"), val = bool(false)]; + tensor mh_w_55_cast_fp16 = matmul(transpose_x = mh_w_55_transpose_x_0, transpose_y = mh_w_55_transpose_y_0, x = var_3429_cast_fp16, y = var_3431_cast_fp16)[name = string("mh_w_55_cast_fp16")]; + tensor var_3434_cast_fp16 = softmax(axis = var_3372, x = mh_w_55_cast_fp16)[name = string("op_3434_cast_fp16")]; + tensor var_3435 = const()[name = string("op_3435"), val = tensor([1, 20, 64, -1])]; + tensor var_3436_cast_fp16 = reshape(shape = var_3435, x = value_55_cast_fp16)[name = string("op_3436_cast_fp16")]; + bool attn_55_transpose_x_0 = const()[name = string("attn_55_transpose_x_0"), val = bool(false)]; + bool attn_55_transpose_y_0 = const()[name = string("attn_55_transpose_y_0"), val = bool(true)]; + tensor attn_55_cast_fp16 = matmul(transpose_x = attn_55_transpose_x_0, transpose_y = attn_55_transpose_y_0, x = var_3436_cast_fp16, y = var_3434_cast_fp16)[name = string("attn_55_cast_fp16")]; + tensor var_3439 = const()[name = string("op_3439"), val = tensor([1, 1280, 1, -1])]; + tensor input_217_cast_fp16 = reshape(shape = var_3439, x = attn_55_cast_fp16)[name = string("input_217_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087043840)))]; + tensor layers_27_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090320704)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_27_self_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_27_self_attn_o_proj_weight_to_fp16, x = input_217_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_111_cast_fp16")]; + tensor out_111_axes_0 = const()[name = string("out_111_axes_0"), val = tensor([1])]; + fp16 var_3457_to_fp16 = const()[name = string("op_3457_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_3457_to_fp16, x = inputs_111_cast_fp16)[name = string("out_111_cast_fp16")]; + tensor input_219_gamma_0_to_fp16 = const()[name = string("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090323328)))]; + tensor input_219_beta_0_to_fp16 = const()[name = string("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090325952)))]; + fp16 input_219_epsilon_0_to_fp16 = const()[name = string("input_219_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = string("input_219_cast_fp16")]; + string input_221_pad_type_0 = const()[name = string("input_221_pad_type_0"), val = string("valid")]; + tensor input_221_strides_0 = const()[name = string("input_221_strides_0"), val = tensor([1, 1])]; + tensor input_221_pad_0 = const()[name = string("input_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_221_dilations_0 = const()[name = string("input_221_dilations_0"), val = tensor([1, 1])]; + int32 input_221_groups_0 = const()[name = string("input_221_groups_0"), val = int32(1)]; + tensor layers_27_fc1_weight_to_fp16 = const()[name = string("layers_27_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090328576)))]; + tensor layers_27_fc1_bias_to_fp16 = const()[name = string("layers_27_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103435840)))]; + tensor input_221_cast_fp16 = conv(bias = layers_27_fc1_bias_to_fp16, dilations = input_221_dilations_0, groups = input_221_groups_0, pad = input_221_pad_0, pad_type = input_221_pad_type_0, strides = input_221_strides_0, weight = layers_27_fc1_weight_to_fp16, x = input_219_cast_fp16)[name = string("input_221_cast_fp16")]; + string input_223_mode_0 = const()[name = string("input_223_mode_0"), val = string("EXACT")]; + tensor input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")]; + string hidden_states_59_pad_type_0 = const()[name = string("hidden_states_59_pad_type_0"), val = string("valid")]; + tensor hidden_states_59_strides_0 = const()[name = string("hidden_states_59_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_59_pad_0 = const()[name = string("hidden_states_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_59_dilations_0 = const()[name = string("hidden_states_59_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_59_groups_0 = const()[name = string("hidden_states_59_groups_0"), val = int32(1)]; + tensor layers_27_fc2_weight_to_fp16 = const()[name = string("layers_27_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103446144)))]; + tensor layers_27_fc2_bias_to_fp16 = const()[name = string("layers_27_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116553408)))]; + tensor hidden_states_59_cast_fp16 = conv(bias = layers_27_fc2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = layers_27_fc2_weight_to_fp16, x = input_223_cast_fp16)[name = string("hidden_states_59_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = hidden_states_59_cast_fp16)[name = string("inputs_113_cast_fp16")]; + int32 var_3490 = const()[name = string("op_3490"), val = int32(3)]; + tensor out_113_axes_0 = const()[name = string("out_113_axes_0"), val = tensor([1])]; + fp16 var_3509_to_fp16 = const()[name = string("op_3509_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_3509_to_fp16, x = inputs_113_cast_fp16)[name = string("out_113_cast_fp16")]; + tensor obj_113_gamma_0_to_fp16 = const()[name = string("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116556032)))]; + tensor obj_113_beta_0_to_fp16 = const()[name = string("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116558656)))]; + fp16 obj_113_epsilon_0_to_fp16 = const()[name = string("obj_113_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = string("obj_113_cast_fp16")]; + string query_57_pad_type_0 = const()[name = string("query_57_pad_type_0"), val = string("valid")]; + tensor query_57_strides_0 = const()[name = string("query_57_strides_0"), val = tensor([1, 1])]; + tensor query_57_pad_0 = const()[name = string("query_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_57_dilations_0 = const()[name = string("query_57_dilations_0"), val = tensor([1, 1])]; + int32 query_57_groups_0 = const()[name = string("query_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116561280)))]; + tensor layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119838144)))]; + tensor query_57_cast_fp16 = conv(bias = layers_28_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_28_self_attn_q_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("query_57_cast_fp16")]; + string key_57_pad_type_0 = const()[name = string("key_57_pad_type_0"), val = string("valid")]; + tensor key_57_strides_0 = const()[name = string("key_57_strides_0"), val = tensor([1, 1])]; + tensor key_57_pad_0 = const()[name = string("key_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_57_dilations_0 = const()[name = string("key_57_dilations_0"), val = tensor([1, 1])]; + int32 key_57_groups_0 = const()[name = string("key_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119840768)))]; + tensor key_57_cast_fp16 = conv(dilations = key_57_dilations_0, groups = key_57_groups_0, pad = key_57_pad_0, pad_type = key_57_pad_type_0, strides = key_57_strides_0, weight = layers_28_self_attn_k_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("key_57_cast_fp16")]; + string value_57_pad_type_0 = const()[name = string("value_57_pad_type_0"), val = string("valid")]; + tensor value_57_strides_0 = const()[name = string("value_57_strides_0"), val = tensor([1, 1])]; + tensor value_57_pad_0 = const()[name = string("value_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_57_dilations_0 = const()[name = string("value_57_dilations_0"), val = tensor([1, 1])]; + int32 value_57_groups_0 = const()[name = string("value_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1123117632)))]; + tensor layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126394496)))]; + tensor value_57_cast_fp16 = conv(bias = layers_28_self_attn_v_proj_bias_to_fp16, dilations = value_57_dilations_0, groups = value_57_groups_0, pad = value_57_pad_0, pad_type = value_57_pad_type_0, strides = value_57_strides_0, weight = layers_28_self_attn_v_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("value_57_cast_fp16")]; + tensor var_3544 = const()[name = string("op_3544"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_57_cast_fp16 = reshape(shape = var_3544, x = query_57_cast_fp16)[name = string("mh_q_57_cast_fp16")]; + fp16 var_3546_to_fp16 = const()[name = string("op_3546_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3547_cast_fp16 = mul(x = mh_q_57_cast_fp16, y = var_3546_to_fp16)[name = string("op_3547_cast_fp16")]; + tensor var_3548 = const()[name = string("op_3548"), val = tensor([1, 20, 64, -1])]; + tensor var_3549_cast_fp16 = reshape(shape = var_3548, x = key_57_cast_fp16)[name = string("op_3549_cast_fp16")]; + bool mh_w_57_transpose_x_0 = const()[name = string("mh_w_57_transpose_x_0"), val = bool(true)]; + bool mh_w_57_transpose_y_0 = const()[name = string("mh_w_57_transpose_y_0"), val = bool(false)]; + tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_3547_cast_fp16, y = var_3549_cast_fp16)[name = string("mh_w_57_cast_fp16")]; + tensor var_3552_cast_fp16 = softmax(axis = var_3490, x = mh_w_57_cast_fp16)[name = string("op_3552_cast_fp16")]; + tensor var_3553 = const()[name = string("op_3553"), val = tensor([1, 20, 64, -1])]; + tensor var_3554_cast_fp16 = reshape(shape = var_3553, x = value_57_cast_fp16)[name = string("op_3554_cast_fp16")]; + bool attn_57_transpose_x_0 = const()[name = string("attn_57_transpose_x_0"), val = bool(false)]; + bool attn_57_transpose_y_0 = const()[name = string("attn_57_transpose_y_0"), val = bool(true)]; + tensor attn_57_cast_fp16 = matmul(transpose_x = attn_57_transpose_x_0, transpose_y = attn_57_transpose_y_0, x = var_3554_cast_fp16, y = var_3552_cast_fp16)[name = string("attn_57_cast_fp16")]; + tensor var_3557 = const()[name = string("op_3557"), val = tensor([1, 1280, 1, -1])]; + tensor input_225_cast_fp16 = reshape(shape = var_3557, x = attn_57_cast_fp16)[name = string("input_225_cast_fp16")]; + string obj_115_pad_type_0 = const()[name = string("obj_115_pad_type_0"), val = string("valid")]; + tensor obj_115_strides_0 = const()[name = string("obj_115_strides_0"), val = tensor([1, 1])]; + tensor obj_115_pad_0 = const()[name = string("obj_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_115_dilations_0 = const()[name = string("obj_115_dilations_0"), val = tensor([1, 1])]; + int32 obj_115_groups_0 = const()[name = string("obj_115_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126397120)))]; + tensor layers_28_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129673984)))]; + tensor obj_115_cast_fp16 = conv(bias = layers_28_self_attn_o_proj_bias_to_fp16, dilations = obj_115_dilations_0, groups = obj_115_groups_0, pad = obj_115_pad_0, pad_type = obj_115_pad_type_0, strides = obj_115_strides_0, weight = layers_28_self_attn_o_proj_weight_to_fp16, x = input_225_cast_fp16)[name = string("obj_115_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_115_cast_fp16)[name = string("inputs_115_cast_fp16")]; + tensor out_115_axes_0 = const()[name = string("out_115_axes_0"), val = tensor([1])]; + fp16 var_3575_to_fp16 = const()[name = string("op_3575_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_3575_to_fp16, x = inputs_115_cast_fp16)[name = string("out_115_cast_fp16")]; + tensor input_227_gamma_0_to_fp16 = const()[name = string("input_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129676608)))]; + tensor input_227_beta_0_to_fp16 = const()[name = string("input_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129679232)))]; + fp16 input_227_epsilon_0_to_fp16 = const()[name = string("input_227_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_227_cast_fp16 = batch_norm(beta = input_227_beta_0_to_fp16, epsilon = input_227_epsilon_0_to_fp16, gamma = input_227_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = string("input_227_cast_fp16")]; + string input_229_pad_type_0 = const()[name = string("input_229_pad_type_0"), val = string("valid")]; + tensor input_229_strides_0 = const()[name = string("input_229_strides_0"), val = tensor([1, 1])]; + tensor input_229_pad_0 = const()[name = string("input_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_229_dilations_0 = const()[name = string("input_229_dilations_0"), val = tensor([1, 1])]; + int32 input_229_groups_0 = const()[name = string("input_229_groups_0"), val = int32(1)]; + tensor layers_28_fc1_weight_to_fp16 = const()[name = string("layers_28_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129681856)))]; + tensor layers_28_fc1_bias_to_fp16 = const()[name = string("layers_28_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142789120)))]; + tensor input_229_cast_fp16 = conv(bias = layers_28_fc1_bias_to_fp16, dilations = input_229_dilations_0, groups = input_229_groups_0, pad = input_229_pad_0, pad_type = input_229_pad_type_0, strides = input_229_strides_0, weight = layers_28_fc1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")]; + string input_231_mode_0 = const()[name = string("input_231_mode_0"), val = string("EXACT")]; + tensor input_231_cast_fp16 = gelu(mode = input_231_mode_0, x = input_229_cast_fp16)[name = string("input_231_cast_fp16")]; + string hidden_states_61_pad_type_0 = const()[name = string("hidden_states_61_pad_type_0"), val = string("valid")]; + tensor hidden_states_61_strides_0 = const()[name = string("hidden_states_61_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_61_pad_0 = const()[name = string("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_61_dilations_0 = const()[name = string("hidden_states_61_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_61_groups_0 = const()[name = string("hidden_states_61_groups_0"), val = int32(1)]; + tensor layers_28_fc2_weight_to_fp16 = const()[name = string("layers_28_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142799424)))]; + tensor layers_28_fc2_bias_to_fp16 = const()[name = string("layers_28_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155906688)))]; + tensor hidden_states_61_cast_fp16 = conv(bias = layers_28_fc2_bias_to_fp16, dilations = hidden_states_61_dilations_0, groups = hidden_states_61_groups_0, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = hidden_states_61_strides_0, weight = layers_28_fc2_weight_to_fp16, x = input_231_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = hidden_states_61_cast_fp16)[name = string("inputs_117_cast_fp16")]; + int32 var_3608 = const()[name = string("op_3608"), val = int32(3)]; + tensor out_117_axes_0 = const()[name = string("out_117_axes_0"), val = tensor([1])]; + fp16 var_3627_to_fp16 = const()[name = string("op_3627_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_3627_to_fp16, x = inputs_117_cast_fp16)[name = string("out_117_cast_fp16")]; + tensor obj_117_gamma_0_to_fp16 = const()[name = string("obj_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155909312)))]; + tensor obj_117_beta_0_to_fp16 = const()[name = string("obj_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155911936)))]; + fp16 obj_117_epsilon_0_to_fp16 = const()[name = string("obj_117_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_117_cast_fp16 = batch_norm(beta = obj_117_beta_0_to_fp16, epsilon = obj_117_epsilon_0_to_fp16, gamma = obj_117_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = string("obj_117_cast_fp16")]; + string query_59_pad_type_0 = const()[name = string("query_59_pad_type_0"), val = string("valid")]; + tensor query_59_strides_0 = const()[name = string("query_59_strides_0"), val = tensor([1, 1])]; + tensor query_59_pad_0 = const()[name = string("query_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_59_dilations_0 = const()[name = string("query_59_dilations_0"), val = tensor([1, 1])]; + int32 query_59_groups_0 = const()[name = string("query_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155914560)))]; + tensor layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159191424)))]; + tensor query_59_cast_fp16 = conv(bias = layers_29_self_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_29_self_attn_q_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("query_59_cast_fp16")]; + string key_59_pad_type_0 = const()[name = string("key_59_pad_type_0"), val = string("valid")]; + tensor key_59_strides_0 = const()[name = string("key_59_strides_0"), val = tensor([1, 1])]; + tensor key_59_pad_0 = const()[name = string("key_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_59_dilations_0 = const()[name = string("key_59_dilations_0"), val = tensor([1, 1])]; + int32 key_59_groups_0 = const()[name = string("key_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159194048)))]; + tensor key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_29_self_attn_k_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("key_59_cast_fp16")]; + string value_59_pad_type_0 = const()[name = string("value_59_pad_type_0"), val = string("valid")]; + tensor value_59_strides_0 = const()[name = string("value_59_strides_0"), val = tensor([1, 1])]; + tensor value_59_pad_0 = const()[name = string("value_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_59_dilations_0 = const()[name = string("value_59_dilations_0"), val = tensor([1, 1])]; + int32 value_59_groups_0 = const()[name = string("value_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1162470912)))]; + tensor layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165747776)))]; + tensor value_59_cast_fp16 = conv(bias = layers_29_self_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_29_self_attn_v_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("value_59_cast_fp16")]; + tensor var_3662 = const()[name = string("op_3662"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_59_cast_fp16 = reshape(shape = var_3662, x = query_59_cast_fp16)[name = string("mh_q_59_cast_fp16")]; + fp16 var_3664_to_fp16 = const()[name = string("op_3664_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3665_cast_fp16 = mul(x = mh_q_59_cast_fp16, y = var_3664_to_fp16)[name = string("op_3665_cast_fp16")]; + tensor var_3666 = const()[name = string("op_3666"), val = tensor([1, 20, 64, -1])]; + tensor var_3667_cast_fp16 = reshape(shape = var_3666, x = key_59_cast_fp16)[name = string("op_3667_cast_fp16")]; + bool mh_w_59_transpose_x_0 = const()[name = string("mh_w_59_transpose_x_0"), val = bool(true)]; + bool mh_w_59_transpose_y_0 = const()[name = string("mh_w_59_transpose_y_0"), val = bool(false)]; + tensor mh_w_59_cast_fp16 = matmul(transpose_x = mh_w_59_transpose_x_0, transpose_y = mh_w_59_transpose_y_0, x = var_3665_cast_fp16, y = var_3667_cast_fp16)[name = string("mh_w_59_cast_fp16")]; + tensor var_3670_cast_fp16 = softmax(axis = var_3608, x = mh_w_59_cast_fp16)[name = string("op_3670_cast_fp16")]; + tensor var_3671 = const()[name = string("op_3671"), val = tensor([1, 20, 64, -1])]; + tensor var_3672_cast_fp16 = reshape(shape = var_3671, x = value_59_cast_fp16)[name = string("op_3672_cast_fp16")]; + bool attn_59_transpose_x_0 = const()[name = string("attn_59_transpose_x_0"), val = bool(false)]; + bool attn_59_transpose_y_0 = const()[name = string("attn_59_transpose_y_0"), val = bool(true)]; + tensor attn_59_cast_fp16 = matmul(transpose_x = attn_59_transpose_x_0, transpose_y = attn_59_transpose_y_0, x = var_3672_cast_fp16, y = var_3670_cast_fp16)[name = string("attn_59_cast_fp16")]; + tensor var_3675 = const()[name = string("op_3675"), val = tensor([1, 1280, 1, -1])]; + tensor input_233_cast_fp16 = reshape(shape = var_3675, x = attn_59_cast_fp16)[name = string("input_233_cast_fp16")]; + string obj_119_pad_type_0 = const()[name = string("obj_119_pad_type_0"), val = string("valid")]; + tensor obj_119_strides_0 = const()[name = string("obj_119_strides_0"), val = tensor([1, 1])]; + tensor obj_119_pad_0 = const()[name = string("obj_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_119_dilations_0 = const()[name = string("obj_119_dilations_0"), val = tensor([1, 1])]; + int32 obj_119_groups_0 = const()[name = string("obj_119_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165750400)))]; + tensor layers_29_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169027264)))]; + tensor obj_119_cast_fp16 = conv(bias = layers_29_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_29_self_attn_o_proj_weight_to_fp16, x = input_233_cast_fp16)[name = string("obj_119_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_119_cast_fp16)[name = string("inputs_119_cast_fp16")]; + tensor out_119_axes_0 = const()[name = string("out_119_axes_0"), val = tensor([1])]; + fp16 var_3693_to_fp16 = const()[name = string("op_3693_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_3693_to_fp16, x = inputs_119_cast_fp16)[name = string("out_119_cast_fp16")]; + tensor input_235_gamma_0_to_fp16 = const()[name = string("input_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169029888)))]; + tensor input_235_beta_0_to_fp16 = const()[name = string("input_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169032512)))]; + fp16 input_235_epsilon_0_to_fp16 = const()[name = string("input_235_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = string("input_235_cast_fp16")]; + string input_237_pad_type_0 = const()[name = string("input_237_pad_type_0"), val = string("valid")]; + tensor input_237_strides_0 = const()[name = string("input_237_strides_0"), val = tensor([1, 1])]; + tensor input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_237_dilations_0 = const()[name = string("input_237_dilations_0"), val = tensor([1, 1])]; + int32 input_237_groups_0 = const()[name = string("input_237_groups_0"), val = int32(1)]; + tensor layers_29_fc1_weight_to_fp16 = const()[name = string("layers_29_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169035136)))]; + tensor layers_29_fc1_bias_to_fp16 = const()[name = string("layers_29_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182142400)))]; + tensor input_237_cast_fp16 = conv(bias = layers_29_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_29_fc1_weight_to_fp16, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")]; + string input_239_mode_0 = const()[name = string("input_239_mode_0"), val = string("EXACT")]; + tensor input_239_cast_fp16 = gelu(mode = input_239_mode_0, x = input_237_cast_fp16)[name = string("input_239_cast_fp16")]; + string hidden_states_63_pad_type_0 = const()[name = string("hidden_states_63_pad_type_0"), val = string("valid")]; + tensor hidden_states_63_strides_0 = const()[name = string("hidden_states_63_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_63_pad_0 = const()[name = string("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_63_dilations_0 = const()[name = string("hidden_states_63_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_63_groups_0 = const()[name = string("hidden_states_63_groups_0"), val = int32(1)]; + tensor layers_29_fc2_weight_to_fp16 = const()[name = string("layers_29_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182152704)))]; + tensor layers_29_fc2_bias_to_fp16 = const()[name = string("layers_29_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195259968)))]; + tensor hidden_states_63_cast_fp16 = conv(bias = layers_29_fc2_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = layers_29_fc2_weight_to_fp16, x = input_239_cast_fp16)[name = string("hidden_states_63_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_63_cast_fp16)[name = string("inputs_121_cast_fp16")]; + int32 var_3726 = const()[name = string("op_3726"), val = int32(3)]; + tensor out_121_axes_0 = const()[name = string("out_121_axes_0"), val = tensor([1])]; + fp16 var_3745_to_fp16 = const()[name = string("op_3745_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_3745_to_fp16, x = inputs_121_cast_fp16)[name = string("out_121_cast_fp16")]; + tensor obj_121_gamma_0_to_fp16 = const()[name = string("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195262592)))]; + tensor obj_121_beta_0_to_fp16 = const()[name = string("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195265216)))]; + fp16 obj_121_epsilon_0_to_fp16 = const()[name = string("obj_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = string("obj_121_cast_fp16")]; + string query_61_pad_type_0 = const()[name = string("query_61_pad_type_0"), val = string("valid")]; + tensor query_61_strides_0 = const()[name = string("query_61_strides_0"), val = tensor([1, 1])]; + tensor query_61_pad_0 = const()[name = string("query_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_61_dilations_0 = const()[name = string("query_61_dilations_0"), val = tensor([1, 1])]; + int32 query_61_groups_0 = const()[name = string("query_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195267840)))]; + tensor layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198544704)))]; + tensor query_61_cast_fp16 = conv(bias = layers_30_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_30_self_attn_q_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("query_61_cast_fp16")]; + string key_61_pad_type_0 = const()[name = string("key_61_pad_type_0"), val = string("valid")]; + tensor key_61_strides_0 = const()[name = string("key_61_strides_0"), val = tensor([1, 1])]; + tensor key_61_pad_0 = const()[name = string("key_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_61_dilations_0 = const()[name = string("key_61_dilations_0"), val = tensor([1, 1])]; + int32 key_61_groups_0 = const()[name = string("key_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198547328)))]; + tensor key_61_cast_fp16 = conv(dilations = key_61_dilations_0, groups = key_61_groups_0, pad = key_61_pad_0, pad_type = key_61_pad_type_0, strides = key_61_strides_0, weight = layers_30_self_attn_k_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("key_61_cast_fp16")]; + string value_61_pad_type_0 = const()[name = string("value_61_pad_type_0"), val = string("valid")]; + tensor value_61_strides_0 = const()[name = string("value_61_strides_0"), val = tensor([1, 1])]; + tensor value_61_pad_0 = const()[name = string("value_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_61_dilations_0 = const()[name = string("value_61_dilations_0"), val = tensor([1, 1])]; + int32 value_61_groups_0 = const()[name = string("value_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1201824192)))]; + tensor layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205101056)))]; + tensor value_61_cast_fp16 = conv(bias = layers_30_self_attn_v_proj_bias_to_fp16, dilations = value_61_dilations_0, groups = value_61_groups_0, pad = value_61_pad_0, pad_type = value_61_pad_type_0, strides = value_61_strides_0, weight = layers_30_self_attn_v_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("value_61_cast_fp16")]; + tensor var_3780 = const()[name = string("op_3780"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_61_cast_fp16 = reshape(shape = var_3780, x = query_61_cast_fp16)[name = string("mh_q_61_cast_fp16")]; + fp16 var_3782_to_fp16 = const()[name = string("op_3782_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3783_cast_fp16 = mul(x = mh_q_61_cast_fp16, y = var_3782_to_fp16)[name = string("op_3783_cast_fp16")]; + tensor var_3784 = const()[name = string("op_3784"), val = tensor([1, 20, 64, -1])]; + tensor var_3785_cast_fp16 = reshape(shape = var_3784, x = key_61_cast_fp16)[name = string("op_3785_cast_fp16")]; + bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)]; + bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)]; + tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_3783_cast_fp16, y = var_3785_cast_fp16)[name = string("mh_w_61_cast_fp16")]; + tensor var_3788_cast_fp16 = softmax(axis = var_3726, x = mh_w_61_cast_fp16)[name = string("op_3788_cast_fp16")]; + tensor var_3789 = const()[name = string("op_3789"), val = tensor([1, 20, 64, -1])]; + tensor var_3790_cast_fp16 = reshape(shape = var_3789, x = value_61_cast_fp16)[name = string("op_3790_cast_fp16")]; + bool attn_61_transpose_x_0 = const()[name = string("attn_61_transpose_x_0"), val = bool(false)]; + bool attn_61_transpose_y_0 = const()[name = string("attn_61_transpose_y_0"), val = bool(true)]; + tensor attn_61_cast_fp16 = matmul(transpose_x = attn_61_transpose_x_0, transpose_y = attn_61_transpose_y_0, x = var_3790_cast_fp16, y = var_3788_cast_fp16)[name = string("attn_61_cast_fp16")]; + tensor var_3793 = const()[name = string("op_3793"), val = tensor([1, 1280, 1, -1])]; + tensor input_241_cast_fp16 = reshape(shape = var_3793, x = attn_61_cast_fp16)[name = string("input_241_cast_fp16")]; + string obj_123_pad_type_0 = const()[name = string("obj_123_pad_type_0"), val = string("valid")]; + tensor obj_123_strides_0 = const()[name = string("obj_123_strides_0"), val = tensor([1, 1])]; + tensor obj_123_pad_0 = const()[name = string("obj_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_123_dilations_0 = const()[name = string("obj_123_dilations_0"), val = tensor([1, 1])]; + int32 obj_123_groups_0 = const()[name = string("obj_123_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205103680)))]; + tensor layers_30_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208380544)))]; + tensor obj_123_cast_fp16 = conv(bias = layers_30_self_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_30_self_attn_o_proj_weight_to_fp16, x = input_241_cast_fp16)[name = string("obj_123_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_123_cast_fp16)[name = string("inputs_123_cast_fp16")]; + tensor out_123_axes_0 = const()[name = string("out_123_axes_0"), val = tensor([1])]; + fp16 var_3811_to_fp16 = const()[name = string("op_3811_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_3811_to_fp16, x = inputs_123_cast_fp16)[name = string("out_123_cast_fp16")]; + tensor input_243_gamma_0_to_fp16 = const()[name = string("input_243_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208383168)))]; + tensor input_243_beta_0_to_fp16 = const()[name = string("input_243_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208385792)))]; + fp16 input_243_epsilon_0_to_fp16 = const()[name = string("input_243_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_243_cast_fp16 = batch_norm(beta = input_243_beta_0_to_fp16, epsilon = input_243_epsilon_0_to_fp16, gamma = input_243_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = string("input_243_cast_fp16")]; + string input_245_pad_type_0 = const()[name = string("input_245_pad_type_0"), val = string("valid")]; + tensor input_245_strides_0 = const()[name = string("input_245_strides_0"), val = tensor([1, 1])]; + tensor input_245_pad_0 = const()[name = string("input_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_245_dilations_0 = const()[name = string("input_245_dilations_0"), val = tensor([1, 1])]; + int32 input_245_groups_0 = const()[name = string("input_245_groups_0"), val = int32(1)]; + tensor layers_30_fc1_weight_to_fp16 = const()[name = string("layers_30_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208388416)))]; + tensor layers_30_fc1_bias_to_fp16 = const()[name = string("layers_30_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221495680)))]; + tensor input_245_cast_fp16 = conv(bias = layers_30_fc1_bias_to_fp16, dilations = input_245_dilations_0, groups = input_245_groups_0, pad = input_245_pad_0, pad_type = input_245_pad_type_0, strides = input_245_strides_0, weight = layers_30_fc1_weight_to_fp16, x = input_243_cast_fp16)[name = string("input_245_cast_fp16")]; + string input_247_mode_0 = const()[name = string("input_247_mode_0"), val = string("EXACT")]; + tensor input_247_cast_fp16 = gelu(mode = input_247_mode_0, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")]; + string hidden_states_65_pad_type_0 = const()[name = string("hidden_states_65_pad_type_0"), val = string("valid")]; + tensor hidden_states_65_strides_0 = const()[name = string("hidden_states_65_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_65_pad_0 = const()[name = string("hidden_states_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_65_dilations_0 = const()[name = string("hidden_states_65_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_65_groups_0 = const()[name = string("hidden_states_65_groups_0"), val = int32(1)]; + tensor layers_30_fc2_weight_to_fp16 = const()[name = string("layers_30_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221505984)))]; + tensor layers_30_fc2_bias_to_fp16 = const()[name = string("layers_30_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234613248)))]; + tensor hidden_states_65_cast_fp16 = conv(bias = layers_30_fc2_bias_to_fp16, dilations = hidden_states_65_dilations_0, groups = hidden_states_65_groups_0, pad = hidden_states_65_pad_0, pad_type = hidden_states_65_pad_type_0, strides = hidden_states_65_strides_0, weight = layers_30_fc2_weight_to_fp16, x = input_247_cast_fp16)[name = string("hidden_states_65_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = hidden_states_65_cast_fp16)[name = string("inputs_125_cast_fp16")]; + int32 var_3844 = const()[name = string("op_3844"), val = int32(3)]; + tensor out_125_axes_0 = const()[name = string("out_125_axes_0"), val = tensor([1])]; + fp16 var_3863_to_fp16 = const()[name = string("op_3863_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_3863_to_fp16, x = inputs_125_cast_fp16)[name = string("out_125_cast_fp16")]; + tensor obj_125_gamma_0_to_fp16 = const()[name = string("obj_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234615872)))]; + tensor obj_125_beta_0_to_fp16 = const()[name = string("obj_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234618496)))]; + fp16 obj_125_epsilon_0_to_fp16 = const()[name = string("obj_125_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_125_cast_fp16 = batch_norm(beta = obj_125_beta_0_to_fp16, epsilon = obj_125_epsilon_0_to_fp16, gamma = obj_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = string("obj_125_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234621120)))]; + tensor layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237897984)))]; + tensor query_cast_fp16 = conv(bias = layers_31_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_31_self_attn_q_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237900608)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_31_self_attn_k_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1241177472)))]; + tensor layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244454336)))]; + tensor value_cast_fp16 = conv(bias = layers_31_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_31_self_attn_v_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_3898 = const()[name = string("op_3898"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_3898, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_3900_to_fp16 = const()[name = string("op_3900_to_fp16"), val = fp16(0x1p-3)]; + tensor var_3901_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_3900_to_fp16)[name = string("op_3901_cast_fp16")]; + tensor var_3902 = const()[name = string("op_3902"), val = tensor([1, 20, 64, -1])]; + tensor var_3903_cast_fp16 = reshape(shape = var_3902, x = key_cast_fp16)[name = string("op_3903_cast_fp16")]; + bool mh_w_transpose_x_0 = const()[name = string("mh_w_transpose_x_0"), val = bool(true)]; + bool mh_w_transpose_y_0 = const()[name = string("mh_w_transpose_y_0"), val = bool(false)]; + tensor mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_3901_cast_fp16, y = var_3903_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor var_3906_cast_fp16 = softmax(axis = var_3844, x = mh_w_cast_fp16)[name = string("op_3906_cast_fp16")]; + tensor var_3907 = const()[name = string("op_3907"), val = tensor([1, 20, 64, -1])]; + tensor var_3908_cast_fp16 = reshape(shape = var_3907, x = value_cast_fp16)[name = string("op_3908_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_3908_cast_fp16, y = var_3906_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_3911 = const()[name = string("op_3911"), val = tensor([1, 1280, 1, -1])]; + tensor input_249_cast_fp16 = reshape(shape = var_3911, x = attn_cast_fp16)[name = string("input_249_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244456960)))]; + tensor layers_31_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247733824)))]; + tensor obj_cast_fp16 = conv(bias = layers_31_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_31_self_attn_o_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = obj_cast_fp16)[name = string("inputs_127_cast_fp16")]; + tensor out_127_axes_0 = const()[name = string("out_127_axes_0"), val = tensor([1])]; + fp16 var_3929_to_fp16 = const()[name = string("op_3929_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_3929_to_fp16, x = inputs_127_cast_fp16)[name = string("out_127_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = string("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247736448)))]; + tensor input_251_beta_0_to_fp16 = const()[name = string("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247739072)))]; + fp16 input_251_epsilon_0_to_fp16 = const()[name = string("input_251_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = string("input_251_cast_fp16")]; + string input_253_pad_type_0 = const()[name = string("input_253_pad_type_0"), val = string("valid")]; + tensor input_253_strides_0 = const()[name = string("input_253_strides_0"), val = tensor([1, 1])]; + tensor input_253_pad_0 = const()[name = string("input_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_253_dilations_0 = const()[name = string("input_253_dilations_0"), val = tensor([1, 1])]; + int32 input_253_groups_0 = const()[name = string("input_253_groups_0"), val = int32(1)]; + tensor layers_31_fc1_weight_to_fp16 = const()[name = string("layers_31_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247741696)))]; + tensor layers_31_fc1_bias_to_fp16 = const()[name = string("layers_31_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260848960)))]; + tensor input_253_cast_fp16 = conv(bias = layers_31_fc1_bias_to_fp16, dilations = input_253_dilations_0, groups = input_253_groups_0, pad = input_253_pad_0, pad_type = input_253_pad_type_0, strides = input_253_strides_0, weight = layers_31_fc1_weight_to_fp16, x = input_251_cast_fp16)[name = string("input_253_cast_fp16")]; + string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("EXACT")]; + tensor input_255_cast_fp16 = gelu(mode = input_255_mode_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_31_fc2_weight_to_fp16 = const()[name = string("layers_31_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260859264)))]; + tensor layers_31_fc2_bias_to_fp16 = const()[name = string("layers_31_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273966528)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_31_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_31_fc2_weight_to_fp16, x = input_255_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_127_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_3967_to_fp16 = const()[name = string("op_3967_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_3967_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273969152)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273971776)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_3991_pad_type_0 = const()[name = string("op_3991_pad_type_0"), val = string("valid")]; + tensor var_3991_strides_0 = const()[name = string("op_3991_strides_0"), val = tensor([1, 1])]; + tensor var_3991_pad_0 = const()[name = string("op_3991_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3991_dilations_0 = const()[name = string("op_3991_dilations_0"), val = tensor([1, 1])]; + int32 var_3991_groups_0 = const()[name = string("op_3991_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273974400)))]; + tensor var_3991_cast_fp16 = conv(dilations = var_3991_dilations_0, groups = var_3991_groups_0, pad = var_3991_pad_0, pad_type = var_3991_pad_type_0, strides = var_3991_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_3991_cast_fp16")]; + string var_3998_pad_type_0 = const()[name = string("op_3998_pad_type_0"), val = string("valid")]; + tensor var_3998_strides_0 = const()[name = string("op_3998_strides_0"), val = tensor([1, 1])]; + tensor var_3998_pad_0 = const()[name = string("op_3998_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3998_dilations_0 = const()[name = string("op_3998_dilations_0"), val = tensor([1, 1])]; + int32 var_3998_groups_0 = const()[name = string("op_3998_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1277251264)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280528128)))]; + tensor var_3998_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_3998_dilations_0, groups = var_3998_groups_0, pad = var_3998_pad_0, pad_type = var_3998_pad_type_0, strides = var_3998_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_3998_cast_fp16")]; + string var_4016_pad_type_0 = const()[name = string("op_4016_pad_type_0"), val = string("valid")]; + tensor var_4016_strides_0 = const()[name = string("op_4016_strides_0"), val = tensor([1, 1])]; + tensor var_4016_pad_0 = const()[name = string("op_4016_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4016_dilations_0 = const()[name = string("op_4016_dilations_0"), val = tensor([1, 1])]; + int32 var_4016_groups_0 = const()[name = string("op_4016_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280530752)))]; + tensor var_4016_cast_fp16 = conv(dilations = var_4016_dilations_0, groups = var_4016_groups_0, pad = var_4016_pad_0, pad_type = var_4016_pad_type_0, strides = var_4016_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4016_cast_fp16")]; + string var_4023_pad_type_0 = const()[name = string("op_4023_pad_type_0"), val = string("valid")]; + tensor var_4023_strides_0 = const()[name = string("op_4023_strides_0"), val = tensor([1, 1])]; + tensor var_4023_pad_0 = const()[name = string("op_4023_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4023_dilations_0 = const()[name = string("op_4023_dilations_0"), val = tensor([1, 1])]; + int32 var_4023_groups_0 = const()[name = string("op_4023_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1283807616)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1287084480)))]; + tensor var_4023_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_4023_dilations_0, groups = var_4023_groups_0, pad = var_4023_pad_0, pad_type = var_4023_pad_type_0, strides = var_4023_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4023_cast_fp16")]; + string var_4041_pad_type_0 = const()[name = string("op_4041_pad_type_0"), val = string("valid")]; + tensor var_4041_strides_0 = const()[name = string("op_4041_strides_0"), val = tensor([1, 1])]; + tensor var_4041_pad_0 = const()[name = string("op_4041_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4041_dilations_0 = const()[name = string("op_4041_dilations_0"), val = tensor([1, 1])]; + int32 var_4041_groups_0 = const()[name = string("op_4041_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1287087104)))]; + tensor var_4041_cast_fp16 = conv(dilations = var_4041_dilations_0, groups = var_4041_groups_0, pad = var_4041_pad_0, pad_type = var_4041_pad_type_0, strides = var_4041_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4041_cast_fp16")]; + string var_4048_pad_type_0 = const()[name = string("op_4048_pad_type_0"), val = string("valid")]; + tensor var_4048_strides_0 = const()[name = string("op_4048_strides_0"), val = tensor([1, 1])]; + tensor var_4048_pad_0 = const()[name = string("op_4048_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4048_dilations_0 = const()[name = string("op_4048_dilations_0"), val = tensor([1, 1])]; + int32 var_4048_groups_0 = const()[name = string("op_4048_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1290363968)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1293640832)))]; + tensor var_4048_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_4048_dilations_0, groups = var_4048_groups_0, pad = var_4048_pad_0, pad_type = var_4048_pad_type_0, strides = var_4048_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4048_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1293643456)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1296920320)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1300197184)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_4078 = const()[name = string("op_4078"), val = int32(0)]; + bool input_259_interleave_0 = const()[name = string("input_259_interleave_0"), val = bool(false)]; + tensor input_259_cast_fp16 = concat(axis = var_4078, interleave = input_259_interleave_0, values = (var_3991_cast_fp16, var_4016_cast_fp16, var_4041_cast_fp16, k_cast_fp16))[name = string("input_259_cast_fp16")]; + int32 var_4081 = const()[name = string("op_4081"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_4081, interleave = input_interleave_0, values = (var_3998_cast_fp16, var_4023_cast_fp16, var_4048_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_4088_pad_0 = const()[name = string("op_4088_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4088_mode_0 = const()[name = string("op_4088_mode_0"), val = string("constant")]; + fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_33_to_fp16, mode = var_4088_mode_0, pad = var_4088_pad_0, x = input_259_cast_fp16)[name = string("op_4088_cast_fp16")]; + tensor var_4094_pad_0 = const()[name = string("op_4094_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4094_mode_0 = const()[name = string("op_4094_mode_0"), val = string("constant")]; + fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_34_to_fp16, mode = var_4094_mode_0, pad = var_4094_pad_0, x = input_cast_fp16)[name = string("op_4094_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..b7782cab9f13e7f9bfb526d1e60e6b169f9cf6df --- /dev/null +++ b/openai_whisper-large-v3-v20240930/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:151f7b8578b144f50f29cd3542a1247bf6f2ba707d6ac0a480b630d82fe3d236 +size 1300199808 diff --git a/openai_whisper-large-v3-v20240930/LICENSE_NOTICE.txt b/openai_whisper-large-v3-v20240930/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-large-v3-v20240930/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..3ba3246801c85f92f79ac029f59b94e7fb646f85 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0980462db89a546e1e90888ea38e0a5ddf1f1fec84608802cdbb12f8a5cc7215 +size 243 diff --git a/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..780171e73cd57a772ec0457470f0c8b86f4c73cd --- /dev/null +++ b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6475c6649047ce609e3fe84b2525843c03342820662404540baf28146c174014 +size 329 diff --git a/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..65be90aad1d0e5f73a1f50b19705ccad3c0da822 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..6cf57d7dbf15af35e56636caf15aff60353296f0 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..2ae170c9000db89326cc2600450001654bb10f7f --- /dev/null +++ b/openai_whisper-large-v3-v20240930/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:009d9fb8f6b589accfa08cebf1c712ef07c3405229ce3cfb3a57ee033c9d8a49 +size 373376 diff --git a/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..e663d681b043c4befaf670427a5aa56c5e6f35f7 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:554d32ced57d792d052df3888d8de65b1bd12c8ad56f876979de0beff2a07abc +size 243 diff --git a/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..89928605833c60a71a2f6f55f19c9a96d133f403 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4901c5e4249e42e8f37325412eb6fcf9ca9c5e22660271613675afed77cff8f +size 754 diff --git a/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/metadata.json b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..cd8c99cc3198103882587e3d40bf5da8335b3664 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51866)", + "shortDescription" : "", + "shape" : "[1, 1, 51866]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 5120 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 5120, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 5120 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 5120, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 8, + "Ios18.mul" : 16, + "Ios18.matmul" : 16, + "Ios18.batchNorm" : 13, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 29, + "Ios18.layerNorm" : 13, + "Ios18.reshape" : 32, + "Ios18.linear" : 1, + "Ios18.conv" : 32, + "Ios18.gelu" : 4, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 20, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/model.mil b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..9f060563f9c7e6cab241feb43aab826c7414c868 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,679 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; + int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; + bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_26_cast_fp16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = input_ids, validate_indices = var_26_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_26_cast_fp16")]; + int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; + int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; + bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132777088)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_71")]; + tensor var_30_cast_fp16_cast_uint16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_30_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_30_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_26_cast_fp16, y = var_30_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_44_axes_0 = const()[name = string("op_44_axes_0"), val = tensor([2])]; + tensor var_44_cast_fp16 = expand_dims(axes = var_44_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_44_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_44_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1])]; + int32 var_49_axis_0 = const()[name = string("op_49_axis_0"), val = int32(0)]; + tensor var_49_cast_fp16_0, tensor var_49_cast_fp16_1, tensor var_49_cast_fp16_2, tensor var_49_cast_fp16_3 = split(axis = var_49_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_49_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1])]; + int32 var_56_axis_0 = const()[name = string("op_56_axis_0"), val = int32(0)]; + tensor var_56_cast_fp16_0, tensor var_56_cast_fp16_1, tensor var_56_cast_fp16_2, tensor var_56_cast_fp16_3 = split(axis = var_56_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_56_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_76 = const()[name = string("op_76"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_101_to_fp16 = const()[name = string("op_101_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_101_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133924032)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133926656)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133929280)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133931904)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133934528)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137211392)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137214016)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140490880)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143767744)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor([1])]; + tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = kv_cache_update_mask)[name = string("op_136_cast_fp16")]; + tensor var_137_axes_0 = const()[name = string("op_137_axes_0"), val = tensor([2])]; + tensor var_137_cast_fp16 = expand_dims(axes = var_137_axes_0, x = var_136_cast_fp16)[name = string("op_137_cast_fp16")]; + tensor var_139_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_139_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_49_cast_fp16_0, y = var_139_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_141_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_141_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_56_cast_fp16_0, y = var_141_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_144 = const()[name = string("op_144"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_144, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_146_to_fp16 = const()[name = string("op_146_to_fp16"), val = fp16(0x1p-3)]; + tensor var_147_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_146_to_fp16)[name = string("op_147_cast_fp16")]; + tensor var_148 = const()[name = string("op_148"), val = tensor([1, 20, 64, -1])]; + tensor var_149_cast_fp16 = reshape(shape = var_148, x = key_1_cast_fp16)[name = string("op_149_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_147_cast_fp16, y = var_149_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_153_axes_0 = const()[name = string("op_153_axes_0"), val = tensor([1])]; + tensor var_153_cast_fp16 = expand_dims(axes = var_153_axes_0, x = decoder_key_padding_mask)[name = string("op_153_cast_fp16")]; + tensor var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor([2])]; + tensor var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = var_153_cast_fp16)[name = string("op_154_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_157_cast_fp16 = softmax(axis = var_76, x = mh_w_3_cast_fp16)[name = string("op_157_cast_fp16")]; + tensor var_158 = const()[name = string("op_158"), val = tensor([1, 20, 64, -1])]; + tensor var_159_cast_fp16 = reshape(shape = var_158, x = value_1_cast_fp16)[name = string("op_159_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_159_cast_fp16, y = var_157_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_162 = const()[name = string("op_162"), val = tensor([1, 1280, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_162, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143770368)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147047232)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_184_to_fp16 = const()[name = string("op_184_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_184_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147049856)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147052480)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147055104)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331968)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_204 = const()[name = string("op_204"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_204, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_206_to_fp16 = const()[name = string("op_206_to_fp16"), val = fp16(0x1p-3)]; + tensor var_207_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_206_to_fp16)[name = string("op_207_cast_fp16")]; + tensor var_208 = const()[name = string("op_208"), val = tensor([1, 20, 64, -1])]; + tensor var_209_cast_fp16 = reshape(shape = var_208, x = obj_17_cast_fp16)[name = string("op_209_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_207_cast_fp16, y = var_209_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_213_axes_0 = const()[name = string("op_213_axes_0"), val = tensor([1])]; + tensor var_213_cast_fp16 = expand_dims(axes = var_213_axes_0, x = read_state_4)[name = string("op_213_cast_fp16")]; + tensor var_214_axes_0 = const()[name = string("op_214_axes_0"), val = tensor([2])]; + tensor var_214_cast_fp16 = expand_dims(axes = var_214_axes_0, x = var_213_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_76, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_218 = const()[name = string("op_218"), val = tensor([1, 20, 64, -1])]; + tensor var_219_cast_fp16 = reshape(shape = var_218, x = obj_19_cast_fp16)[name = string("op_219_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_219_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_222 = const()[name = string("op_222"), val = tensor([1, 1280, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_222, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334592)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_240_to_fp16 = const()[name = string("op_240_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_240_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614080)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616704)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153619328)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166726592)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166736896)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179844160)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_285 = const()[name = string("op_285"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_310_to_fp16 = const()[name = string("op_310_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_310_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179846784)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179849408)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179852032)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128896)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183131520)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186408384)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189685248)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_348_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_348_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_49_cast_fp16_1, y = var_348_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_350_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_350_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_56_cast_fp16_1, y = var_350_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_353 = const()[name = string("op_353"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_353, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_355_to_fp16 = const()[name = string("op_355_to_fp16"), val = fp16(0x1p-3)]; + tensor var_356_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_355_to_fp16)[name = string("op_356_cast_fp16")]; + tensor var_357 = const()[name = string("op_357"), val = tensor([1, 20, 64, -1])]; + tensor var_358_cast_fp16 = reshape(shape = var_357, x = key_3_cast_fp16)[name = string("op_358_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_356_cast_fp16, y = var_358_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_366_cast_fp16 = softmax(axis = var_285, x = mh_w_11_cast_fp16)[name = string("op_366_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([1, 20, 64, -1])]; + tensor var_368_cast_fp16 = reshape(shape = var_367, x = value_3_cast_fp16)[name = string("op_368_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_368_cast_fp16, y = var_366_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_371 = const()[name = string("op_371"), val = tensor([1, 1280, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_371, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189687872)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192964736)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_393_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192967360)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192969984)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192972608)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196249472)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_413 = const()[name = string("op_413"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_413, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_415_to_fp16 = const()[name = string("op_415_to_fp16"), val = fp16(0x1p-3)]; + tensor var_416_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_415_to_fp16)[name = string("op_416_cast_fp16")]; + tensor var_417 = const()[name = string("op_417"), val = tensor([1, 20, 64, -1])]; + tensor var_418_cast_fp16 = reshape(shape = var_417, x = obj_35_cast_fp16)[name = string("op_418_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_416_cast_fp16, y = var_418_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_285, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_427 = const()[name = string("op_427"), val = tensor([1, 20, 64, -1])]; + tensor var_428_cast_fp16 = reshape(shape = var_427, x = obj_37_cast_fp16)[name = string("op_428_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_428_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_431 = const()[name = string("op_431"), val = tensor([1, 1280, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_431, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196252096)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199528960)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_449_to_fp16 = const()[name = string("op_449_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_449_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199531584)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199534208)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199536832)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212644096)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212654400)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225761664)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 1280, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 1280, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_494 = const()[name = string("op_494"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_519_to_fp16 = const()[name = string("op_519_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_519_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225764288)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225766912)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225769536)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229046400)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229049024)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232325888)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235602752)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_557_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_557_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_49_cast_fp16_2, y = var_557_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_559_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_559_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_56_cast_fp16_2, y = var_559_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_562 = const()[name = string("op_562"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_562, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_564_to_fp16 = const()[name = string("op_564_to_fp16"), val = fp16(0x1p-3)]; + tensor var_565_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_564_to_fp16)[name = string("op_565_cast_fp16")]; + tensor var_566 = const()[name = string("op_566"), val = tensor([1, 20, 64, -1])]; + tensor var_567_cast_fp16 = reshape(shape = var_566, x = key_5_cast_fp16)[name = string("op_567_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_565_cast_fp16, y = var_567_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_575_cast_fp16 = softmax(axis = var_494, x = mh_w_19_cast_fp16)[name = string("op_575_cast_fp16")]; + tensor var_576 = const()[name = string("op_576"), val = tensor([1, 20, 64, -1])]; + tensor var_577_cast_fp16 = reshape(shape = var_576, x = value_5_cast_fp16)[name = string("op_577_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_577_cast_fp16, y = var_575_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_580 = const()[name = string("op_580"), val = tensor([1, 1280, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_580, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235605376)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238882240)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_602_to_fp16 = const()[name = string("op_602_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_602_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238884864)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238887488)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238890112)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242166976)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_622 = const()[name = string("op_622"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_622, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_624_to_fp16 = const()[name = string("op_624_to_fp16"), val = fp16(0x1p-3)]; + tensor var_625_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_624_to_fp16)[name = string("op_625_cast_fp16")]; + tensor var_626 = const()[name = string("op_626"), val = tensor([1, 20, 64, -1])]; + tensor var_627_cast_fp16 = reshape(shape = var_626, x = obj_53_cast_fp16)[name = string("op_627_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_625_cast_fp16, y = var_627_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_494, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_636 = const()[name = string("op_636"), val = tensor([1, 20, 64, -1])]; + tensor var_637_cast_fp16 = reshape(shape = var_636, x = obj_55_cast_fp16)[name = string("op_637_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_637_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_640 = const()[name = string("op_640"), val = tensor([1, 1280, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_640, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242169600)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245446464)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_661_to_fp16 = const()[name = string("op_661_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_661_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245449088)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245451712)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245454336)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258561600)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258571904)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271679168)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 1280, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 1280, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_707 = const()[name = string("op_707"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_732_to_fp16 = const()[name = string("op_732_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_732_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271681792)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271684416)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271687040)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274963904)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274966528)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278243392)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281520256)))]; + tensor current_value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_770_cast_fp16 = mul(x = current_key_cast_fp16, y = var_137_cast_fp16)[name = string("op_770_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_49_cast_fp16_3, y = var_770_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_772_cast_fp16 = mul(x = current_value_cast_fp16, y = var_137_cast_fp16)[name = string("op_772_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_56_cast_fp16_3, y = var_772_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_775 = const()[name = string("op_775"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_775, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_777_to_fp16 = const()[name = string("op_777_to_fp16"), val = fp16(0x1p-3)]; + tensor var_778_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_777_to_fp16)[name = string("op_778_cast_fp16")]; + tensor var_779 = const()[name = string("op_779"), val = tensor([1, 20, 64, -1])]; + tensor var_780_cast_fp16 = reshape(shape = var_779, x = key_cast_fp16)[name = string("op_780_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_778_cast_fp16, y = var_780_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_788_cast_fp16 = softmax(axis = var_707, x = mh_w_27_cast_fp16)[name = string("op_788_cast_fp16")]; + tensor var_789 = const()[name = string("op_789"), val = tensor([1, 20, 64, -1])]; + tensor var_790_cast_fp16 = reshape(shape = var_789, x = value_cast_fp16)[name = string("op_790_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_790_cast_fp16, y = var_788_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_793 = const()[name = string("op_793"), val = tensor([1, 1280, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_793, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281522880)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284799744)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_815_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284802368)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284804992)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284807616)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288084480)))]; + tensor query_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_835 = const()[name = string("op_835"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_835, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_837_to_fp16 = const()[name = string("op_837_to_fp16"), val = fp16(0x1p-3)]; + tensor var_838_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_837_to_fp16)[name = string("op_838_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 20, 64, -1])]; + tensor var_840_cast_fp16 = reshape(shape = var_839, x = obj_71_cast_fp16)[name = string("op_840_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_838_cast_fp16, y = var_840_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_707, x = mh_w_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_849 = const()[name = string("op_849"), val = tensor([1, 20, 64, -1])]; + tensor var_850_cast_fp16 = reshape(shape = var_849, x = obj_73_cast_fp16)[name = string("op_850_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_850_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_853 = const()[name = string("op_853"), val = tensor([1, 1280, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_853, x = attn_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288087104)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291363968)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_874_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291366592)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291369216)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291371840)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304479104)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_37_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304489408)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317596672)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_917_to_fp16 = const()[name = string("op_917_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_917_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317599296)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317601920)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_928_axes_0 = const()[name = string("op_928_axes_0"), val = tensor([2])]; + tensor var_928_cast_fp16 = squeeze(axes = var_928_axes_0, x = hidden_states_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_931_perm_0 = const()[name = string("op_931_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317604544)))]; + tensor var_931_cast_fp16 = transpose(perm = var_931_perm_0, x = var_928_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_931_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_935 = const()[name = string("op_935"), val = int32(1)]; + bool obj_81_interleave_0 = const()[name = string("obj_81_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_935, interleave = obj_81_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_cast_fp16))[name = string("obj_81_cast_fp16")]; + int32 var_938 = const()[name = string("op_938"), val = int32(1)]; + bool obj_83_interleave_0 = const()[name = string("obj_83_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_938, interleave = obj_83_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_cast_fp16))[name = string("obj_83_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = obj_59_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_952_begin_0 = const()[name = string("op_952_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_952_end_0 = const()[name = string("op_952_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_952_end_mask_0 = const()[name = string("op_952_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_952_squeeze_mask_0 = const()[name = string("op_952_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_952_cast_fp16 = slice_by_index(begin = var_952_begin_0, end = var_952_end_0, end_mask = var_952_end_mask_0, squeeze_mask = var_952_squeeze_mask_0, x = var_949_cast_fp16)[name = string("op_952_cast_fp16")]; + tensor var_967_begin_0 = const()[name = string("op_967_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_967_end_0 = const()[name = string("op_967_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_967_end_mask_0 = const()[name = string("op_967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_967_cast_fp16 = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = obj_59_cast_fp16)[name = string("op_967_cast_fp16")]; + tensor var_970_begin_0 = const()[name = string("op_970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_end_0 = const()[name = string("op_970_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_970_end_mask_0 = const()[name = string("op_970_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_970_squeeze_mask_0 = const()[name = string("op_970_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, squeeze_mask = var_970_squeeze_mask_0, x = var_967_cast_fp16)[name = string("op_970_cast_fp16")]; + tensor var_985_begin_0 = const()[name = string("op_985_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_985_end_0 = const()[name = string("op_985_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_985_end_mask_0 = const()[name = string("op_985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = obj_77_cast_fp16)[name = string("op_985_cast_fp16")]; + tensor var_988_begin_0 = const()[name = string("op_988_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_988_end_0 = const()[name = string("op_988_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_988_end_mask_0 = const()[name = string("op_988_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_988_squeeze_mask_0 = const()[name = string("op_988_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_988_cast_fp16 = slice_by_index(begin = var_988_begin_0, end = var_988_end_0, end_mask = var_988_end_mask_0, squeeze_mask = var_988_squeeze_mask_0, x = var_985_cast_fp16)[name = string("op_988_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = string("op_1003_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1003_end_0 = const()[name = string("op_1003_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_1003_end_mask_0 = const()[name = string("op_1003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1003_cast_fp16")]; + tensor var_1006_begin_0 = const()[name = string("op_1006_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1006_end_0 = const()[name = string("op_1006_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1006_end_mask_0 = const()[name = string("op_1006_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1006_squeeze_mask_0 = const()[name = string("op_1006_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1006_cast_fp16 = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, squeeze_mask = var_1006_squeeze_mask_0, x = var_1003_cast_fp16)[name = string("op_1006_cast_fp16")]; + tensor var_1021_begin_0 = const()[name = string("op_1021_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1021_end_0 = const()[name = string("op_1021_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_1021_end_mask_0 = const()[name = string("op_1021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1021_cast_fp16 = slice_by_index(begin = var_1021_begin_0, end = var_1021_end_0, end_mask = var_1021_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1021_cast_fp16")]; + tensor var_1024_begin_0 = const()[name = string("op_1024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1024_end_0 = const()[name = string("op_1024_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1024_end_mask_0 = const()[name = string("op_1024_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1024_squeeze_mask_0 = const()[name = string("op_1024_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1024_cast_fp16 = slice_by_index(begin = var_1024_begin_0, end = var_1024_end_0, end_mask = var_1024_end_mask_0, squeeze_mask = var_1024_squeeze_mask_0, x = var_1021_cast_fp16)[name = string("op_1024_cast_fp16")]; + tensor var_1039_begin_0 = const()[name = string("op_1039_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1039_end_0 = const()[name = string("op_1039_end_0"), val = tensor([1, 15, 1, 1536])]; + tensor var_1039_end_mask_0 = const()[name = string("op_1039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1039_cast_fp16")]; + tensor var_1042_begin_0 = const()[name = string("op_1042_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1042_end_0 = const()[name = string("op_1042_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1042_end_mask_0 = const()[name = string("op_1042_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1042_squeeze_mask_0 = const()[name = string("op_1042_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1042_cast_fp16 = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, squeeze_mask = var_1042_squeeze_mask_0, x = var_1039_cast_fp16)[name = string("op_1042_cast_fp16")]; + int32 var_1049 = const()[name = string("op_1049"), val = int32(1)]; + bool var_1050_interleave_0 = const()[name = string("op_1050_interleave_0"), val = bool(false)]; + tensor var_1050_cast_fp16 = concat(axis = var_1049, interleave = var_1050_interleave_0, values = (var_952_cast_fp16, var_970_cast_fp16, var_988_cast_fp16, var_1006_cast_fp16, var_1024_cast_fp16, var_1042_cast_fp16))[name = string("op_1050_cast_fp16")]; + bool var_1053 = const()[name = string("op_1053"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1053, x = var_1050_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..ce17ffd9db34a87639997c3cd17fa5a72513d3eb --- /dev/null +++ b/openai_whisper-large-v3-v20240930/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9be9fd71781e6e440994496f915a5b21f02745788dfb0e6e84978c8f8c89bfc8 +size 317708340 diff --git a/openai_whisper-large-v3-v20240930/config.json b/openai_whisper-large-v3-v20240930/config.json new file mode 100644 index 0000000000000000000000000000000000000000..f6c3db01a3124f90b3bfbf3d58e47ba5de1f5ccd --- /dev/null +++ b/openai_whisper-large-v3-v20240930/config.json @@ -0,0 +1 @@ +{"_name_or_path": "/raid/yoach/tmp_whisper_turbo", "activation_dropout": 0.0, "activation_function": "gelu", "apply_spec_augment": false, "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50256], "bos_token_id": 50257, "classifier_proj_size": 256, "d_model": 1280, "decoder_attention_heads": 20, "decoder_ffn_dim": 5120, "decoder_layerdrop": 0.0, "decoder_layers": 4, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 20, "encoder_ffn_dim": 5120, "encoder_layerdrop": 0.0, "encoder_layers": 32, "eos_token_id": 50257, "init_std": 0.02, "is_encoder_decoder": true, "mask_feature_length": 10, "mask_feature_min_masks": 0, "mask_feature_prob": 0.0, "mask_time_length": 10, "mask_time_min_masks": 2, "mask_time_prob": 0.05, "max_source_positions": 1500, "max_target_positions": 448, "median_filter_width": 7, "model_type": "whisper", "num_hidden_layers": 32, "num_mel_bins": 128, "pad_token_id": 50257, "scale_embedding": false, "torch_dtype": "float16", "transformers_version": "4.46.0.dev0", "use_cache": true, "use_weighted_layer_sum": false, "vocab_size": 51866} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930/generation_config.json b/openai_whisper-large-v3-v20240930/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..a023f6f947b0fc4ab7a2986a1c23b014836ec720 --- /dev/null +++ b/openai_whisper-large-v3-v20240930/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[2, 4], [2, 11], [3, 3], [3, 6], [3, 11], [3, 14]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50360]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|yue|>": 50358, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50364, "pad_token_id": 50257, "prev_sot_token_id": 50362, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50359, 50360, 50361, 50362, 50363], "task_to_id": {"transcribe": 50360, "translate": 50359}, "transformers_version": "4.46.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..7dff7e174bc89d642476cff4cac285454c56b87e --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08b7b4ce51d098601b3195f9a38ffadcd4e619b05135e276b1a3beeda66f0949 +size 243 diff --git a/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..744160e29b93fbb6f726f655af6eec4eea765cb2 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d7522fa9a913828818c7bd24b63a02f7c9eed38197dd609b95a238dcdca40b8 +size 434 diff --git a/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..6eed1b32b711175cf9b4089811cd15d232507ffd --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1280 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 1280, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 65, + "Ios18.conv" : 202, + "Ios18.gelu" : 34, + "Ios18.concat" : 674, + "Ios16.einsum" : 5120, + "Ios18.add" : 65, + "Ios18.softmax" : 2560, + "Ios18.sliceByIndex" : 4480, + "Ios18.layerNorm" : 65, + "Ios18.transpose" : 32, + "Ios18.mul" : 2560 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/model.mil b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..da2e6c970f29e5b78cb28c9cd1f8649a6b98518a --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,39503 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_114_pad_type_0 = const()[name = string("op_114_pad_type_0"), val = string("custom")]; + tensor var_114_pad_0 = const()[name = string("op_114_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_114_strides_0 = const()[name = string("op_114_strides_0"), val = tensor([1, 1])]; + tensor var_114_dilations_0 = const()[name = string("op_114_dilations_0"), val = tensor([1, 1])]; + int32 var_114_groups_0 = const()[name = string("op_114_groups_0"), val = int32(1)]; + tensor var_89_to_fp16 = const()[name = string("op_89_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_95_to_fp16 = const()[name = string("op_95_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(983168)))]; + tensor var_114_cast_fp16 = conv(bias = var_95_to_fp16, dilations = var_114_dilations_0, groups = var_114_groups_0, pad = var_114_pad_0, pad_type = var_114_pad_type_0, strides = var_114_strides_0, weight = var_89_to_fp16, x = melspectrogram_features)[name = string("op_114_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_114_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_154_pad_type_0 = const()[name = string("op_154_pad_type_0"), val = string("custom")]; + tensor var_154_pad_0 = const()[name = string("op_154_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_154_strides_0 = const()[name = string("op_154_strides_0"), val = tensor([2, 2])]; + tensor var_154_dilations_0 = const()[name = string("op_154_dilations_0"), val = tensor([1, 1])]; + int32 var_154_groups_0 = const()[name = string("op_154_groups_0"), val = int32(1)]; + tensor var_129_to_fp16 = const()[name = string("op_129_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985792)))]; + tensor var_135_to_fp16 = const()[name = string("op_135_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10816256)))]; + tensor var_154_cast_fp16 = conv(bias = var_135_to_fp16, dilations = var_154_dilations_0, groups = var_154_groups_0, pad = var_154_pad_0, pad_type = var_154_pad_type_0, strides = var_154_strides_0, weight = var_129_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_154_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_154_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_172_to_fp16 = const()[name = string("op_172_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10818880)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_172_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_182 = const()[name = string("op_182"), val = int32(3)]; + int32 var_207 = const()[name = string("op_207"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_224_to_fp16 = const()[name = string("op_224_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_224_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14658944)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14661568)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14664192)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14666816)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14669440)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17946304)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17948928)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21225792)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24502656)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_262_begin_0 = const()[name = string("op_262_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_262_end_0 = const()[name = string("op_262_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_262_end_mask_0 = const()[name = string("op_262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_262_cast_fp16 = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = query_1_cast_fp16)[name = string("op_262_cast_fp16")]; + tensor var_266_begin_0 = const()[name = string("op_266_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_266_end_0 = const()[name = string("op_266_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_266_end_mask_0 = const()[name = string("op_266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = query_1_cast_fp16)[name = string("op_266_cast_fp16")]; + tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = query_1_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = query_1_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_278_begin_0 = const()[name = string("op_278_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_278_end_0 = const()[name = string("op_278_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_278_end_mask_0 = const()[name = string("op_278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_278_cast_fp16 = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = query_1_cast_fp16)[name = string("op_278_cast_fp16")]; + tensor var_282_begin_0 = const()[name = string("op_282_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_282_end_0 = const()[name = string("op_282_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_282_end_mask_0 = const()[name = string("op_282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_282_cast_fp16 = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, x = query_1_cast_fp16)[name = string("op_282_cast_fp16")]; + tensor var_286_begin_0 = const()[name = string("op_286_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_286_end_0 = const()[name = string("op_286_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_286_end_mask_0 = const()[name = string("op_286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_286_cast_fp16 = slice_by_index(begin = var_286_begin_0, end = var_286_end_0, end_mask = var_286_end_mask_0, x = query_1_cast_fp16)[name = string("op_286_cast_fp16")]; + tensor var_290_begin_0 = const()[name = string("op_290_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_290_end_0 = const()[name = string("op_290_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_290_end_mask_0 = const()[name = string("op_290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = query_1_cast_fp16)[name = string("op_290_cast_fp16")]; + tensor var_294_begin_0 = const()[name = string("op_294_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_294_end_0 = const()[name = string("op_294_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_294_end_mask_0 = const()[name = string("op_294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_294_cast_fp16 = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = query_1_cast_fp16)[name = string("op_294_cast_fp16")]; + tensor var_298_begin_0 = const()[name = string("op_298_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_298_end_0 = const()[name = string("op_298_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_298_end_mask_0 = const()[name = string("op_298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = query_1_cast_fp16)[name = string("op_298_cast_fp16")]; + tensor var_302_begin_0 = const()[name = string("op_302_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_302_end_0 = const()[name = string("op_302_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_302_end_mask_0 = const()[name = string("op_302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = query_1_cast_fp16)[name = string("op_302_cast_fp16")]; + tensor var_306_begin_0 = const()[name = string("op_306_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_306_end_0 = const()[name = string("op_306_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_306_end_mask_0 = const()[name = string("op_306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_306_cast_fp16 = slice_by_index(begin = var_306_begin_0, end = var_306_end_0, end_mask = var_306_end_mask_0, x = query_1_cast_fp16)[name = string("op_306_cast_fp16")]; + tensor var_310_begin_0 = const()[name = string("op_310_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_310_end_0 = const()[name = string("op_310_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_310_end_mask_0 = const()[name = string("op_310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_310_cast_fp16 = slice_by_index(begin = var_310_begin_0, end = var_310_end_0, end_mask = var_310_end_mask_0, x = query_1_cast_fp16)[name = string("op_310_cast_fp16")]; + tensor var_314_begin_0 = const()[name = string("op_314_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_314_end_0 = const()[name = string("op_314_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_314_end_mask_0 = const()[name = string("op_314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_314_cast_fp16 = slice_by_index(begin = var_314_begin_0, end = var_314_end_0, end_mask = var_314_end_mask_0, x = query_1_cast_fp16)[name = string("op_314_cast_fp16")]; + tensor var_318_begin_0 = const()[name = string("op_318_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_318_end_0 = const()[name = string("op_318_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_318_end_mask_0 = const()[name = string("op_318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = query_1_cast_fp16)[name = string("op_318_cast_fp16")]; + tensor var_322_begin_0 = const()[name = string("op_322_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_322_end_0 = const()[name = string("op_322_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_322_end_mask_0 = const()[name = string("op_322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_322_cast_fp16 = slice_by_index(begin = var_322_begin_0, end = var_322_end_0, end_mask = var_322_end_mask_0, x = query_1_cast_fp16)[name = string("op_322_cast_fp16")]; + tensor var_326_begin_0 = const()[name = string("op_326_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_326_end_0 = const()[name = string("op_326_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_326_end_mask_0 = const()[name = string("op_326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = query_1_cast_fp16)[name = string("op_326_cast_fp16")]; + tensor var_330_begin_0 = const()[name = string("op_330_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_330_end_0 = const()[name = string("op_330_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_330_end_mask_0 = const()[name = string("op_330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_330_cast_fp16 = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = query_1_cast_fp16)[name = string("op_330_cast_fp16")]; + tensor var_334_begin_0 = const()[name = string("op_334_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_334_end_0 = const()[name = string("op_334_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_334_end_mask_0 = const()[name = string("op_334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_334_cast_fp16 = slice_by_index(begin = var_334_begin_0, end = var_334_end_0, end_mask = var_334_end_mask_0, x = query_1_cast_fp16)[name = string("op_334_cast_fp16")]; + tensor var_338_begin_0 = const()[name = string("op_338_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_338_end_0 = const()[name = string("op_338_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_338_end_mask_0 = const()[name = string("op_338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_338_cast_fp16 = slice_by_index(begin = var_338_begin_0, end = var_338_end_0, end_mask = var_338_end_mask_0, x = query_1_cast_fp16)[name = string("op_338_cast_fp16")]; + tensor var_347_begin_0 = const()[name = string("op_347_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_347_end_0 = const()[name = string("op_347_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_347_end_mask_0 = const()[name = string("op_347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_347_cast_fp16 = slice_by_index(begin = var_347_begin_0, end = var_347_end_0, end_mask = var_347_end_mask_0, x = var_262_cast_fp16)[name = string("op_347_cast_fp16")]; + tensor var_354_begin_0 = const()[name = string("op_354_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_354_end_0 = const()[name = string("op_354_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_354_end_mask_0 = const()[name = string("op_354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_354_cast_fp16 = slice_by_index(begin = var_354_begin_0, end = var_354_end_0, end_mask = var_354_end_mask_0, x = var_262_cast_fp16)[name = string("op_354_cast_fp16")]; + tensor var_361_begin_0 = const()[name = string("op_361_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_361_end_0 = const()[name = string("op_361_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_361_end_mask_0 = const()[name = string("op_361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_361_cast_fp16 = slice_by_index(begin = var_361_begin_0, end = var_361_end_0, end_mask = var_361_end_mask_0, x = var_262_cast_fp16)[name = string("op_361_cast_fp16")]; + tensor var_368_begin_0 = const()[name = string("op_368_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_368_end_0 = const()[name = string("op_368_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_368_end_mask_0 = const()[name = string("op_368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_368_cast_fp16 = slice_by_index(begin = var_368_begin_0, end = var_368_end_0, end_mask = var_368_end_mask_0, x = var_262_cast_fp16)[name = string("op_368_cast_fp16")]; + tensor var_375_begin_0 = const()[name = string("op_375_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_375_end_0 = const()[name = string("op_375_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_375_end_mask_0 = const()[name = string("op_375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_375_cast_fp16 = slice_by_index(begin = var_375_begin_0, end = var_375_end_0, end_mask = var_375_end_mask_0, x = var_266_cast_fp16)[name = string("op_375_cast_fp16")]; + tensor var_382_begin_0 = const()[name = string("op_382_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_382_end_0 = const()[name = string("op_382_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_382_end_mask_0 = const()[name = string("op_382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_382_cast_fp16 = slice_by_index(begin = var_382_begin_0, end = var_382_end_0, end_mask = var_382_end_mask_0, x = var_266_cast_fp16)[name = string("op_382_cast_fp16")]; + tensor var_389_begin_0 = const()[name = string("op_389_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_389_end_0 = const()[name = string("op_389_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_389_end_mask_0 = const()[name = string("op_389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_389_cast_fp16 = slice_by_index(begin = var_389_begin_0, end = var_389_end_0, end_mask = var_389_end_mask_0, x = var_266_cast_fp16)[name = string("op_389_cast_fp16")]; + tensor var_396_begin_0 = const()[name = string("op_396_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_396_end_0 = const()[name = string("op_396_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_396_end_mask_0 = const()[name = string("op_396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_396_cast_fp16 = slice_by_index(begin = var_396_begin_0, end = var_396_end_0, end_mask = var_396_end_mask_0, x = var_266_cast_fp16)[name = string("op_396_cast_fp16")]; + tensor var_403_begin_0 = const()[name = string("op_403_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_403_end_0 = const()[name = string("op_403_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_403_end_mask_0 = const()[name = string("op_403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_403_cast_fp16 = slice_by_index(begin = var_403_begin_0, end = var_403_end_0, end_mask = var_403_end_mask_0, x = var_270_cast_fp16)[name = string("op_403_cast_fp16")]; + tensor var_410_begin_0 = const()[name = string("op_410_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_410_end_0 = const()[name = string("op_410_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_410_end_mask_0 = const()[name = string("op_410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_410_cast_fp16 = slice_by_index(begin = var_410_begin_0, end = var_410_end_0, end_mask = var_410_end_mask_0, x = var_270_cast_fp16)[name = string("op_410_cast_fp16")]; + tensor var_417_begin_0 = const()[name = string("op_417_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_417_end_0 = const()[name = string("op_417_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_417_end_mask_0 = const()[name = string("op_417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_417_cast_fp16 = slice_by_index(begin = var_417_begin_0, end = var_417_end_0, end_mask = var_417_end_mask_0, x = var_270_cast_fp16)[name = string("op_417_cast_fp16")]; + tensor var_424_begin_0 = const()[name = string("op_424_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_424_end_0 = const()[name = string("op_424_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_424_end_mask_0 = const()[name = string("op_424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_424_cast_fp16 = slice_by_index(begin = var_424_begin_0, end = var_424_end_0, end_mask = var_424_end_mask_0, x = var_270_cast_fp16)[name = string("op_424_cast_fp16")]; + tensor var_431_begin_0 = const()[name = string("op_431_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_431_end_0 = const()[name = string("op_431_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_431_end_mask_0 = const()[name = string("op_431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_431_cast_fp16 = slice_by_index(begin = var_431_begin_0, end = var_431_end_0, end_mask = var_431_end_mask_0, x = var_274_cast_fp16)[name = string("op_431_cast_fp16")]; + tensor var_438_begin_0 = const()[name = string("op_438_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_438_end_0 = const()[name = string("op_438_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_438_end_mask_0 = const()[name = string("op_438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_438_cast_fp16 = slice_by_index(begin = var_438_begin_0, end = var_438_end_0, end_mask = var_438_end_mask_0, x = var_274_cast_fp16)[name = string("op_438_cast_fp16")]; + tensor var_445_begin_0 = const()[name = string("op_445_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_445_end_0 = const()[name = string("op_445_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_445_end_mask_0 = const()[name = string("op_445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_445_cast_fp16 = slice_by_index(begin = var_445_begin_0, end = var_445_end_0, end_mask = var_445_end_mask_0, x = var_274_cast_fp16)[name = string("op_445_cast_fp16")]; + tensor var_452_begin_0 = const()[name = string("op_452_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_452_end_0 = const()[name = string("op_452_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_452_end_mask_0 = const()[name = string("op_452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_452_cast_fp16 = slice_by_index(begin = var_452_begin_0, end = var_452_end_0, end_mask = var_452_end_mask_0, x = var_274_cast_fp16)[name = string("op_452_cast_fp16")]; + tensor var_459_begin_0 = const()[name = string("op_459_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_459_end_0 = const()[name = string("op_459_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_459_end_mask_0 = const()[name = string("op_459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_459_cast_fp16 = slice_by_index(begin = var_459_begin_0, end = var_459_end_0, end_mask = var_459_end_mask_0, x = var_278_cast_fp16)[name = string("op_459_cast_fp16")]; + tensor var_466_begin_0 = const()[name = string("op_466_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_466_end_0 = const()[name = string("op_466_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_466_end_mask_0 = const()[name = string("op_466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_466_cast_fp16 = slice_by_index(begin = var_466_begin_0, end = var_466_end_0, end_mask = var_466_end_mask_0, x = var_278_cast_fp16)[name = string("op_466_cast_fp16")]; + tensor var_473_begin_0 = const()[name = string("op_473_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_473_end_0 = const()[name = string("op_473_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_473_end_mask_0 = const()[name = string("op_473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = var_278_cast_fp16)[name = string("op_473_cast_fp16")]; + tensor var_480_begin_0 = const()[name = string("op_480_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_480_end_0 = const()[name = string("op_480_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_480_end_mask_0 = const()[name = string("op_480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_480_cast_fp16 = slice_by_index(begin = var_480_begin_0, end = var_480_end_0, end_mask = var_480_end_mask_0, x = var_278_cast_fp16)[name = string("op_480_cast_fp16")]; + tensor var_487_begin_0 = const()[name = string("op_487_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_487_end_0 = const()[name = string("op_487_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_487_end_mask_0 = const()[name = string("op_487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_487_cast_fp16 = slice_by_index(begin = var_487_begin_0, end = var_487_end_0, end_mask = var_487_end_mask_0, x = var_282_cast_fp16)[name = string("op_487_cast_fp16")]; + tensor var_494_begin_0 = const()[name = string("op_494_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_494_end_0 = const()[name = string("op_494_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_494_end_mask_0 = const()[name = string("op_494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_494_cast_fp16 = slice_by_index(begin = var_494_begin_0, end = var_494_end_0, end_mask = var_494_end_mask_0, x = var_282_cast_fp16)[name = string("op_494_cast_fp16")]; + tensor var_501_begin_0 = const()[name = string("op_501_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_501_end_0 = const()[name = string("op_501_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_501_end_mask_0 = const()[name = string("op_501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_501_cast_fp16 = slice_by_index(begin = var_501_begin_0, end = var_501_end_0, end_mask = var_501_end_mask_0, x = var_282_cast_fp16)[name = string("op_501_cast_fp16")]; + tensor var_508_begin_0 = const()[name = string("op_508_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_508_end_0 = const()[name = string("op_508_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_508_end_mask_0 = const()[name = string("op_508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_508_cast_fp16 = slice_by_index(begin = var_508_begin_0, end = var_508_end_0, end_mask = var_508_end_mask_0, x = var_282_cast_fp16)[name = string("op_508_cast_fp16")]; + tensor var_515_begin_0 = const()[name = string("op_515_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_515_end_0 = const()[name = string("op_515_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_515_end_mask_0 = const()[name = string("op_515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = var_286_cast_fp16)[name = string("op_515_cast_fp16")]; + tensor var_522_begin_0 = const()[name = string("op_522_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_522_end_0 = const()[name = string("op_522_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_522_end_mask_0 = const()[name = string("op_522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_522_cast_fp16 = slice_by_index(begin = var_522_begin_0, end = var_522_end_0, end_mask = var_522_end_mask_0, x = var_286_cast_fp16)[name = string("op_522_cast_fp16")]; + tensor var_529_begin_0 = const()[name = string("op_529_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_529_end_0 = const()[name = string("op_529_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_529_end_mask_0 = const()[name = string("op_529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_529_cast_fp16 = slice_by_index(begin = var_529_begin_0, end = var_529_end_0, end_mask = var_529_end_mask_0, x = var_286_cast_fp16)[name = string("op_529_cast_fp16")]; + tensor var_536_begin_0 = const()[name = string("op_536_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_536_end_0 = const()[name = string("op_536_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_536_end_mask_0 = const()[name = string("op_536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_536_cast_fp16 = slice_by_index(begin = var_536_begin_0, end = var_536_end_0, end_mask = var_536_end_mask_0, x = var_286_cast_fp16)[name = string("op_536_cast_fp16")]; + tensor var_543_begin_0 = const()[name = string("op_543_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_543_end_0 = const()[name = string("op_543_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_543_end_mask_0 = const()[name = string("op_543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_543_cast_fp16 = slice_by_index(begin = var_543_begin_0, end = var_543_end_0, end_mask = var_543_end_mask_0, x = var_290_cast_fp16)[name = string("op_543_cast_fp16")]; + tensor var_550_begin_0 = const()[name = string("op_550_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_550_end_0 = const()[name = string("op_550_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_550_end_mask_0 = const()[name = string("op_550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_550_cast_fp16 = slice_by_index(begin = var_550_begin_0, end = var_550_end_0, end_mask = var_550_end_mask_0, x = var_290_cast_fp16)[name = string("op_550_cast_fp16")]; + tensor var_557_begin_0 = const()[name = string("op_557_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_557_end_0 = const()[name = string("op_557_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_557_end_mask_0 = const()[name = string("op_557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_557_cast_fp16 = slice_by_index(begin = var_557_begin_0, end = var_557_end_0, end_mask = var_557_end_mask_0, x = var_290_cast_fp16)[name = string("op_557_cast_fp16")]; + tensor var_564_begin_0 = const()[name = string("op_564_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_564_end_0 = const()[name = string("op_564_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_564_end_mask_0 = const()[name = string("op_564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_564_cast_fp16 = slice_by_index(begin = var_564_begin_0, end = var_564_end_0, end_mask = var_564_end_mask_0, x = var_290_cast_fp16)[name = string("op_564_cast_fp16")]; + tensor var_571_begin_0 = const()[name = string("op_571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_571_end_0 = const()[name = string("op_571_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_571_end_mask_0 = const()[name = string("op_571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_571_cast_fp16 = slice_by_index(begin = var_571_begin_0, end = var_571_end_0, end_mask = var_571_end_mask_0, x = var_294_cast_fp16)[name = string("op_571_cast_fp16")]; + tensor var_578_begin_0 = const()[name = string("op_578_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_578_end_0 = const()[name = string("op_578_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_578_end_mask_0 = const()[name = string("op_578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_578_cast_fp16 = slice_by_index(begin = var_578_begin_0, end = var_578_end_0, end_mask = var_578_end_mask_0, x = var_294_cast_fp16)[name = string("op_578_cast_fp16")]; + tensor var_585_begin_0 = const()[name = string("op_585_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_585_end_0 = const()[name = string("op_585_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_585_end_mask_0 = const()[name = string("op_585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_585_cast_fp16 = slice_by_index(begin = var_585_begin_0, end = var_585_end_0, end_mask = var_585_end_mask_0, x = var_294_cast_fp16)[name = string("op_585_cast_fp16")]; + tensor var_592_begin_0 = const()[name = string("op_592_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_592_end_0 = const()[name = string("op_592_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_592_end_mask_0 = const()[name = string("op_592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_592_cast_fp16 = slice_by_index(begin = var_592_begin_0, end = var_592_end_0, end_mask = var_592_end_mask_0, x = var_294_cast_fp16)[name = string("op_592_cast_fp16")]; + tensor var_599_begin_0 = const()[name = string("op_599_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_599_end_0 = const()[name = string("op_599_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_599_end_mask_0 = const()[name = string("op_599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_599_cast_fp16 = slice_by_index(begin = var_599_begin_0, end = var_599_end_0, end_mask = var_599_end_mask_0, x = var_298_cast_fp16)[name = string("op_599_cast_fp16")]; + tensor var_606_begin_0 = const()[name = string("op_606_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_606_end_0 = const()[name = string("op_606_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_606_end_mask_0 = const()[name = string("op_606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_606_cast_fp16 = slice_by_index(begin = var_606_begin_0, end = var_606_end_0, end_mask = var_606_end_mask_0, x = var_298_cast_fp16)[name = string("op_606_cast_fp16")]; + tensor var_613_begin_0 = const()[name = string("op_613_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_613_end_0 = const()[name = string("op_613_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_613_end_mask_0 = const()[name = string("op_613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_613_cast_fp16 = slice_by_index(begin = var_613_begin_0, end = var_613_end_0, end_mask = var_613_end_mask_0, x = var_298_cast_fp16)[name = string("op_613_cast_fp16")]; + tensor var_620_begin_0 = const()[name = string("op_620_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_620_end_0 = const()[name = string("op_620_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_620_end_mask_0 = const()[name = string("op_620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_620_cast_fp16 = slice_by_index(begin = var_620_begin_0, end = var_620_end_0, end_mask = var_620_end_mask_0, x = var_298_cast_fp16)[name = string("op_620_cast_fp16")]; + tensor var_627_begin_0 = const()[name = string("op_627_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_627_end_0 = const()[name = string("op_627_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_627_end_mask_0 = const()[name = string("op_627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_627_cast_fp16 = slice_by_index(begin = var_627_begin_0, end = var_627_end_0, end_mask = var_627_end_mask_0, x = var_302_cast_fp16)[name = string("op_627_cast_fp16")]; + tensor var_634_begin_0 = const()[name = string("op_634_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_634_end_0 = const()[name = string("op_634_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_634_end_mask_0 = const()[name = string("op_634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_634_cast_fp16 = slice_by_index(begin = var_634_begin_0, end = var_634_end_0, end_mask = var_634_end_mask_0, x = var_302_cast_fp16)[name = string("op_634_cast_fp16")]; + tensor var_641_begin_0 = const()[name = string("op_641_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_641_end_0 = const()[name = string("op_641_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_641_end_mask_0 = const()[name = string("op_641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_641_cast_fp16 = slice_by_index(begin = var_641_begin_0, end = var_641_end_0, end_mask = var_641_end_mask_0, x = var_302_cast_fp16)[name = string("op_641_cast_fp16")]; + tensor var_648_begin_0 = const()[name = string("op_648_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_648_end_0 = const()[name = string("op_648_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_648_end_mask_0 = const()[name = string("op_648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_648_cast_fp16 = slice_by_index(begin = var_648_begin_0, end = var_648_end_0, end_mask = var_648_end_mask_0, x = var_302_cast_fp16)[name = string("op_648_cast_fp16")]; + tensor var_655_begin_0 = const()[name = string("op_655_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_655_end_0 = const()[name = string("op_655_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_655_end_mask_0 = const()[name = string("op_655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_655_cast_fp16 = slice_by_index(begin = var_655_begin_0, end = var_655_end_0, end_mask = var_655_end_mask_0, x = var_306_cast_fp16)[name = string("op_655_cast_fp16")]; + tensor var_662_begin_0 = const()[name = string("op_662_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_662_end_0 = const()[name = string("op_662_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_662_end_mask_0 = const()[name = string("op_662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_662_cast_fp16 = slice_by_index(begin = var_662_begin_0, end = var_662_end_0, end_mask = var_662_end_mask_0, x = var_306_cast_fp16)[name = string("op_662_cast_fp16")]; + tensor var_669_begin_0 = const()[name = string("op_669_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_669_end_0 = const()[name = string("op_669_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_669_end_mask_0 = const()[name = string("op_669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_669_cast_fp16 = slice_by_index(begin = var_669_begin_0, end = var_669_end_0, end_mask = var_669_end_mask_0, x = var_306_cast_fp16)[name = string("op_669_cast_fp16")]; + tensor var_676_begin_0 = const()[name = string("op_676_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_676_end_0 = const()[name = string("op_676_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_676_end_mask_0 = const()[name = string("op_676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_676_cast_fp16 = slice_by_index(begin = var_676_begin_0, end = var_676_end_0, end_mask = var_676_end_mask_0, x = var_306_cast_fp16)[name = string("op_676_cast_fp16")]; + tensor var_683_begin_0 = const()[name = string("op_683_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_683_end_0 = const()[name = string("op_683_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_683_end_mask_0 = const()[name = string("op_683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_683_cast_fp16 = slice_by_index(begin = var_683_begin_0, end = var_683_end_0, end_mask = var_683_end_mask_0, x = var_310_cast_fp16)[name = string("op_683_cast_fp16")]; + tensor var_690_begin_0 = const()[name = string("op_690_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_690_end_0 = const()[name = string("op_690_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_690_end_mask_0 = const()[name = string("op_690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_690_cast_fp16 = slice_by_index(begin = var_690_begin_0, end = var_690_end_0, end_mask = var_690_end_mask_0, x = var_310_cast_fp16)[name = string("op_690_cast_fp16")]; + tensor var_697_begin_0 = const()[name = string("op_697_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_697_end_0 = const()[name = string("op_697_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_697_end_mask_0 = const()[name = string("op_697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_697_cast_fp16 = slice_by_index(begin = var_697_begin_0, end = var_697_end_0, end_mask = var_697_end_mask_0, x = var_310_cast_fp16)[name = string("op_697_cast_fp16")]; + tensor var_704_begin_0 = const()[name = string("op_704_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_704_end_0 = const()[name = string("op_704_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_704_end_mask_0 = const()[name = string("op_704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_704_cast_fp16 = slice_by_index(begin = var_704_begin_0, end = var_704_end_0, end_mask = var_704_end_mask_0, x = var_310_cast_fp16)[name = string("op_704_cast_fp16")]; + tensor var_711_begin_0 = const()[name = string("op_711_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_711_end_0 = const()[name = string("op_711_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_711_end_mask_0 = const()[name = string("op_711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_711_cast_fp16 = slice_by_index(begin = var_711_begin_0, end = var_711_end_0, end_mask = var_711_end_mask_0, x = var_314_cast_fp16)[name = string("op_711_cast_fp16")]; + tensor var_718_begin_0 = const()[name = string("op_718_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_718_end_0 = const()[name = string("op_718_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_718_end_mask_0 = const()[name = string("op_718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_718_cast_fp16 = slice_by_index(begin = var_718_begin_0, end = var_718_end_0, end_mask = var_718_end_mask_0, x = var_314_cast_fp16)[name = string("op_718_cast_fp16")]; + tensor var_725_begin_0 = const()[name = string("op_725_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_725_end_0 = const()[name = string("op_725_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_725_end_mask_0 = const()[name = string("op_725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_725_cast_fp16 = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = var_314_cast_fp16)[name = string("op_725_cast_fp16")]; + tensor var_732_begin_0 = const()[name = string("op_732_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_732_end_0 = const()[name = string("op_732_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_732_end_mask_0 = const()[name = string("op_732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_732_cast_fp16 = slice_by_index(begin = var_732_begin_0, end = var_732_end_0, end_mask = var_732_end_mask_0, x = var_314_cast_fp16)[name = string("op_732_cast_fp16")]; + tensor var_739_begin_0 = const()[name = string("op_739_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_739_end_0 = const()[name = string("op_739_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_739_end_mask_0 = const()[name = string("op_739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_739_cast_fp16 = slice_by_index(begin = var_739_begin_0, end = var_739_end_0, end_mask = var_739_end_mask_0, x = var_318_cast_fp16)[name = string("op_739_cast_fp16")]; + tensor var_746_begin_0 = const()[name = string("op_746_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_746_end_0 = const()[name = string("op_746_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_746_end_mask_0 = const()[name = string("op_746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_746_cast_fp16 = slice_by_index(begin = var_746_begin_0, end = var_746_end_0, end_mask = var_746_end_mask_0, x = var_318_cast_fp16)[name = string("op_746_cast_fp16")]; + tensor var_753_begin_0 = const()[name = string("op_753_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_753_end_0 = const()[name = string("op_753_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_753_end_mask_0 = const()[name = string("op_753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = var_318_cast_fp16)[name = string("op_753_cast_fp16")]; + tensor var_760_begin_0 = const()[name = string("op_760_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_760_end_0 = const()[name = string("op_760_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_760_end_mask_0 = const()[name = string("op_760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_760_cast_fp16 = slice_by_index(begin = var_760_begin_0, end = var_760_end_0, end_mask = var_760_end_mask_0, x = var_318_cast_fp16)[name = string("op_760_cast_fp16")]; + tensor var_767_begin_0 = const()[name = string("op_767_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_767_end_0 = const()[name = string("op_767_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_767_end_mask_0 = const()[name = string("op_767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_767_cast_fp16 = slice_by_index(begin = var_767_begin_0, end = var_767_end_0, end_mask = var_767_end_mask_0, x = var_322_cast_fp16)[name = string("op_767_cast_fp16")]; + tensor var_774_begin_0 = const()[name = string("op_774_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_774_end_0 = const()[name = string("op_774_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_774_end_mask_0 = const()[name = string("op_774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_774_cast_fp16 = slice_by_index(begin = var_774_begin_0, end = var_774_end_0, end_mask = var_774_end_mask_0, x = var_322_cast_fp16)[name = string("op_774_cast_fp16")]; + tensor var_781_begin_0 = const()[name = string("op_781_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_781_end_0 = const()[name = string("op_781_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_781_end_mask_0 = const()[name = string("op_781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_781_cast_fp16 = slice_by_index(begin = var_781_begin_0, end = var_781_end_0, end_mask = var_781_end_mask_0, x = var_322_cast_fp16)[name = string("op_781_cast_fp16")]; + tensor var_788_begin_0 = const()[name = string("op_788_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_788_end_0 = const()[name = string("op_788_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_788_end_mask_0 = const()[name = string("op_788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_788_cast_fp16 = slice_by_index(begin = var_788_begin_0, end = var_788_end_0, end_mask = var_788_end_mask_0, x = var_322_cast_fp16)[name = string("op_788_cast_fp16")]; + tensor var_795_begin_0 = const()[name = string("op_795_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_795_end_0 = const()[name = string("op_795_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_795_end_mask_0 = const()[name = string("op_795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_795_cast_fp16 = slice_by_index(begin = var_795_begin_0, end = var_795_end_0, end_mask = var_795_end_mask_0, x = var_326_cast_fp16)[name = string("op_795_cast_fp16")]; + tensor var_802_begin_0 = const()[name = string("op_802_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_802_end_0 = const()[name = string("op_802_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_802_end_mask_0 = const()[name = string("op_802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_802_cast_fp16 = slice_by_index(begin = var_802_begin_0, end = var_802_end_0, end_mask = var_802_end_mask_0, x = var_326_cast_fp16)[name = string("op_802_cast_fp16")]; + tensor var_809_begin_0 = const()[name = string("op_809_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_809_end_0 = const()[name = string("op_809_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_809_end_mask_0 = const()[name = string("op_809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_809_cast_fp16 = slice_by_index(begin = var_809_begin_0, end = var_809_end_0, end_mask = var_809_end_mask_0, x = var_326_cast_fp16)[name = string("op_809_cast_fp16")]; + tensor var_816_begin_0 = const()[name = string("op_816_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_816_end_0 = const()[name = string("op_816_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_816_end_mask_0 = const()[name = string("op_816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_816_cast_fp16 = slice_by_index(begin = var_816_begin_0, end = var_816_end_0, end_mask = var_816_end_mask_0, x = var_326_cast_fp16)[name = string("op_816_cast_fp16")]; + tensor var_823_begin_0 = const()[name = string("op_823_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_823_end_0 = const()[name = string("op_823_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_823_end_mask_0 = const()[name = string("op_823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_823_cast_fp16 = slice_by_index(begin = var_823_begin_0, end = var_823_end_0, end_mask = var_823_end_mask_0, x = var_330_cast_fp16)[name = string("op_823_cast_fp16")]; + tensor var_830_begin_0 = const()[name = string("op_830_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_830_end_0 = const()[name = string("op_830_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_830_end_mask_0 = const()[name = string("op_830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_830_cast_fp16 = slice_by_index(begin = var_830_begin_0, end = var_830_end_0, end_mask = var_830_end_mask_0, x = var_330_cast_fp16)[name = string("op_830_cast_fp16")]; + tensor var_837_begin_0 = const()[name = string("op_837_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_837_end_0 = const()[name = string("op_837_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_837_end_mask_0 = const()[name = string("op_837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_837_cast_fp16 = slice_by_index(begin = var_837_begin_0, end = var_837_end_0, end_mask = var_837_end_mask_0, x = var_330_cast_fp16)[name = string("op_837_cast_fp16")]; + tensor var_844_begin_0 = const()[name = string("op_844_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_844_end_0 = const()[name = string("op_844_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_844_end_mask_0 = const()[name = string("op_844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_844_cast_fp16 = slice_by_index(begin = var_844_begin_0, end = var_844_end_0, end_mask = var_844_end_mask_0, x = var_330_cast_fp16)[name = string("op_844_cast_fp16")]; + tensor var_851_begin_0 = const()[name = string("op_851_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_851_end_0 = const()[name = string("op_851_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_851_end_mask_0 = const()[name = string("op_851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_851_cast_fp16 = slice_by_index(begin = var_851_begin_0, end = var_851_end_0, end_mask = var_851_end_mask_0, x = var_334_cast_fp16)[name = string("op_851_cast_fp16")]; + tensor var_858_begin_0 = const()[name = string("op_858_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_858_end_0 = const()[name = string("op_858_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_858_end_mask_0 = const()[name = string("op_858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_858_cast_fp16 = slice_by_index(begin = var_858_begin_0, end = var_858_end_0, end_mask = var_858_end_mask_0, x = var_334_cast_fp16)[name = string("op_858_cast_fp16")]; + tensor var_865_begin_0 = const()[name = string("op_865_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_865_end_0 = const()[name = string("op_865_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_865_end_mask_0 = const()[name = string("op_865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_865_cast_fp16 = slice_by_index(begin = var_865_begin_0, end = var_865_end_0, end_mask = var_865_end_mask_0, x = var_334_cast_fp16)[name = string("op_865_cast_fp16")]; + tensor var_872_begin_0 = const()[name = string("op_872_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_872_end_0 = const()[name = string("op_872_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_872_end_mask_0 = const()[name = string("op_872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_872_cast_fp16 = slice_by_index(begin = var_872_begin_0, end = var_872_end_0, end_mask = var_872_end_mask_0, x = var_334_cast_fp16)[name = string("op_872_cast_fp16")]; + tensor var_879_begin_0 = const()[name = string("op_879_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_879_end_0 = const()[name = string("op_879_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_879_end_mask_0 = const()[name = string("op_879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_879_cast_fp16 = slice_by_index(begin = var_879_begin_0, end = var_879_end_0, end_mask = var_879_end_mask_0, x = var_338_cast_fp16)[name = string("op_879_cast_fp16")]; + tensor var_886_begin_0 = const()[name = string("op_886_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_886_end_0 = const()[name = string("op_886_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_886_end_mask_0 = const()[name = string("op_886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_886_cast_fp16 = slice_by_index(begin = var_886_begin_0, end = var_886_end_0, end_mask = var_886_end_mask_0, x = var_338_cast_fp16)[name = string("op_886_cast_fp16")]; + tensor var_893_begin_0 = const()[name = string("op_893_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_893_end_0 = const()[name = string("op_893_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_893_end_mask_0 = const()[name = string("op_893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_893_cast_fp16 = slice_by_index(begin = var_893_begin_0, end = var_893_end_0, end_mask = var_893_end_mask_0, x = var_338_cast_fp16)[name = string("op_893_cast_fp16")]; + tensor var_900_begin_0 = const()[name = string("op_900_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_900_end_0 = const()[name = string("op_900_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_900_end_mask_0 = const()[name = string("op_900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_900_cast_fp16 = slice_by_index(begin = var_900_begin_0, end = var_900_end_0, end_mask = var_900_end_mask_0, x = var_338_cast_fp16)[name = string("op_900_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_905_begin_0 = const()[name = string("op_905_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_905_end_0 = const()[name = string("op_905_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_905_end_mask_0 = const()[name = string("op_905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_31")]; + tensor var_905_cast_fp16 = slice_by_index(begin = var_905_begin_0, end = var_905_end_0, end_mask = var_905_end_mask_0, x = k_1_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_909_begin_0 = const()[name = string("op_909_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_909_end_0 = const()[name = string("op_909_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_909_end_mask_0 = const()[name = string("op_909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_909_cast_fp16 = slice_by_index(begin = var_909_begin_0, end = var_909_end_0, end_mask = var_909_end_mask_0, x = k_1_cast_fp16)[name = string("op_909_cast_fp16")]; + tensor var_913_begin_0 = const()[name = string("op_913_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_913_end_0 = const()[name = string("op_913_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_913_end_mask_0 = const()[name = string("op_913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_913_cast_fp16 = slice_by_index(begin = var_913_begin_0, end = var_913_end_0, end_mask = var_913_end_mask_0, x = k_1_cast_fp16)[name = string("op_913_cast_fp16")]; + tensor var_917_begin_0 = const()[name = string("op_917_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_917_end_0 = const()[name = string("op_917_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_917_end_mask_0 = const()[name = string("op_917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_917_cast_fp16 = slice_by_index(begin = var_917_begin_0, end = var_917_end_0, end_mask = var_917_end_mask_0, x = k_1_cast_fp16)[name = string("op_917_cast_fp16")]; + tensor var_921_begin_0 = const()[name = string("op_921_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_921_end_0 = const()[name = string("op_921_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_921_end_mask_0 = const()[name = string("op_921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_921_cast_fp16 = slice_by_index(begin = var_921_begin_0, end = var_921_end_0, end_mask = var_921_end_mask_0, x = k_1_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_925_begin_0 = const()[name = string("op_925_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_925_end_0 = const()[name = string("op_925_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_925_end_mask_0 = const()[name = string("op_925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_925_cast_fp16 = slice_by_index(begin = var_925_begin_0, end = var_925_end_0, end_mask = var_925_end_mask_0, x = k_1_cast_fp16)[name = string("op_925_cast_fp16")]; + tensor var_929_begin_0 = const()[name = string("op_929_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_929_end_0 = const()[name = string("op_929_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_929_end_mask_0 = const()[name = string("op_929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_929_cast_fp16 = slice_by_index(begin = var_929_begin_0, end = var_929_end_0, end_mask = var_929_end_mask_0, x = k_1_cast_fp16)[name = string("op_929_cast_fp16")]; + tensor var_933_begin_0 = const()[name = string("op_933_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_933_end_0 = const()[name = string("op_933_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_933_end_mask_0 = const()[name = string("op_933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_933_cast_fp16 = slice_by_index(begin = var_933_begin_0, end = var_933_end_0, end_mask = var_933_end_mask_0, x = k_1_cast_fp16)[name = string("op_933_cast_fp16")]; + tensor var_937_begin_0 = const()[name = string("op_937_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_937_end_0 = const()[name = string("op_937_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_937_end_mask_0 = const()[name = string("op_937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_937_cast_fp16 = slice_by_index(begin = var_937_begin_0, end = var_937_end_0, end_mask = var_937_end_mask_0, x = k_1_cast_fp16)[name = string("op_937_cast_fp16")]; + tensor var_941_begin_0 = const()[name = string("op_941_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_941_end_0 = const()[name = string("op_941_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_941_end_mask_0 = const()[name = string("op_941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_941_cast_fp16 = slice_by_index(begin = var_941_begin_0, end = var_941_end_0, end_mask = var_941_end_mask_0, x = k_1_cast_fp16)[name = string("op_941_cast_fp16")]; + tensor var_945_begin_0 = const()[name = string("op_945_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_945_end_0 = const()[name = string("op_945_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_945_end_mask_0 = const()[name = string("op_945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_945_cast_fp16 = slice_by_index(begin = var_945_begin_0, end = var_945_end_0, end_mask = var_945_end_mask_0, x = k_1_cast_fp16)[name = string("op_945_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = k_1_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_953_begin_0 = const()[name = string("op_953_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_953_end_0 = const()[name = string("op_953_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_953_end_mask_0 = const()[name = string("op_953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = k_1_cast_fp16)[name = string("op_953_cast_fp16")]; + tensor var_957_begin_0 = const()[name = string("op_957_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_957_end_0 = const()[name = string("op_957_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_957_end_mask_0 = const()[name = string("op_957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_957_cast_fp16 = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = k_1_cast_fp16)[name = string("op_957_cast_fp16")]; + tensor var_961_begin_0 = const()[name = string("op_961_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_961_end_0 = const()[name = string("op_961_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_961_end_mask_0 = const()[name = string("op_961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_961_cast_fp16 = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = k_1_cast_fp16)[name = string("op_961_cast_fp16")]; + tensor var_965_begin_0 = const()[name = string("op_965_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_965_end_0 = const()[name = string("op_965_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_965_end_mask_0 = const()[name = string("op_965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_965_cast_fp16 = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = k_1_cast_fp16)[name = string("op_965_cast_fp16")]; + tensor var_969_begin_0 = const()[name = string("op_969_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_969_end_0 = const()[name = string("op_969_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_969_end_mask_0 = const()[name = string("op_969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_969_cast_fp16 = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = k_1_cast_fp16)[name = string("op_969_cast_fp16")]; + tensor var_973_begin_0 = const()[name = string("op_973_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_973_end_0 = const()[name = string("op_973_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_973_end_mask_0 = const()[name = string("op_973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = k_1_cast_fp16)[name = string("op_973_cast_fp16")]; + tensor var_977_begin_0 = const()[name = string("op_977_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_977_end_0 = const()[name = string("op_977_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_977_end_mask_0 = const()[name = string("op_977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = k_1_cast_fp16)[name = string("op_977_cast_fp16")]; + tensor var_981_begin_0 = const()[name = string("op_981_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_981_end_0 = const()[name = string("op_981_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_981_end_mask_0 = const()[name = string("op_981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_981_cast_fp16 = slice_by_index(begin = var_981_begin_0, end = var_981_end_0, end_mask = var_981_end_mask_0, x = k_1_cast_fp16)[name = string("op_981_cast_fp16")]; + tensor var_983_begin_0 = const()[name = string("op_983_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_983_end_0 = const()[name = string("op_983_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_983_end_mask_0 = const()[name = string("op_983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_983_cast_fp16 = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = value_1_cast_fp16)[name = string("op_983_cast_fp16")]; + tensor var_987_begin_0 = const()[name = string("op_987_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_987_end_0 = const()[name = string("op_987_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_987_end_mask_0 = const()[name = string("op_987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_987_cast_fp16 = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = value_1_cast_fp16)[name = string("op_987_cast_fp16")]; + tensor var_991_begin_0 = const()[name = string("op_991_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_991_end_0 = const()[name = string("op_991_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_991_end_mask_0 = const()[name = string("op_991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = value_1_cast_fp16)[name = string("op_991_cast_fp16")]; + tensor var_995_begin_0 = const()[name = string("op_995_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_995_end_0 = const()[name = string("op_995_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_995_end_mask_0 = const()[name = string("op_995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_995_cast_fp16 = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = value_1_cast_fp16)[name = string("op_995_cast_fp16")]; + tensor var_999_begin_0 = const()[name = string("op_999_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_999_end_0 = const()[name = string("op_999_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_999_end_mask_0 = const()[name = string("op_999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_999_cast_fp16 = slice_by_index(begin = var_999_begin_0, end = var_999_end_0, end_mask = var_999_end_mask_0, x = value_1_cast_fp16)[name = string("op_999_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = string("op_1003_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1003_end_0 = const()[name = string("op_1003_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1003_end_mask_0 = const()[name = string("op_1003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = value_1_cast_fp16)[name = string("op_1003_cast_fp16")]; + tensor var_1007_begin_0 = const()[name = string("op_1007_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1007_end_0 = const()[name = string("op_1007_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1007_end_mask_0 = const()[name = string("op_1007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1007_cast_fp16 = slice_by_index(begin = var_1007_begin_0, end = var_1007_end_0, end_mask = var_1007_end_mask_0, x = value_1_cast_fp16)[name = string("op_1007_cast_fp16")]; + tensor var_1011_begin_0 = const()[name = string("op_1011_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1011_end_0 = const()[name = string("op_1011_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1011_end_mask_0 = const()[name = string("op_1011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1011_cast_fp16 = slice_by_index(begin = var_1011_begin_0, end = var_1011_end_0, end_mask = var_1011_end_mask_0, x = value_1_cast_fp16)[name = string("op_1011_cast_fp16")]; + tensor var_1015_begin_0 = const()[name = string("op_1015_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1015_end_0 = const()[name = string("op_1015_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1015_end_mask_0 = const()[name = string("op_1015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1015_cast_fp16 = slice_by_index(begin = var_1015_begin_0, end = var_1015_end_0, end_mask = var_1015_end_mask_0, x = value_1_cast_fp16)[name = string("op_1015_cast_fp16")]; + tensor var_1019_begin_0 = const()[name = string("op_1019_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1019_end_0 = const()[name = string("op_1019_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1019_end_mask_0 = const()[name = string("op_1019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = var_1019_end_0, end_mask = var_1019_end_mask_0, x = value_1_cast_fp16)[name = string("op_1019_cast_fp16")]; + tensor var_1023_begin_0 = const()[name = string("op_1023_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1023_end_0 = const()[name = string("op_1023_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1023_end_mask_0 = const()[name = string("op_1023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1023_cast_fp16 = slice_by_index(begin = var_1023_begin_0, end = var_1023_end_0, end_mask = var_1023_end_mask_0, x = value_1_cast_fp16)[name = string("op_1023_cast_fp16")]; + tensor var_1027_begin_0 = const()[name = string("op_1027_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1027_end_0 = const()[name = string("op_1027_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1027_end_mask_0 = const()[name = string("op_1027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1027_cast_fp16 = slice_by_index(begin = var_1027_begin_0, end = var_1027_end_0, end_mask = var_1027_end_mask_0, x = value_1_cast_fp16)[name = string("op_1027_cast_fp16")]; + tensor var_1031_begin_0 = const()[name = string("op_1031_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_1031_end_0 = const()[name = string("op_1031_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_1031_end_mask_0 = const()[name = string("op_1031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1031_cast_fp16 = slice_by_index(begin = var_1031_begin_0, end = var_1031_end_0, end_mask = var_1031_end_mask_0, x = value_1_cast_fp16)[name = string("op_1031_cast_fp16")]; + tensor var_1035_begin_0 = const()[name = string("op_1035_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_1035_end_0 = const()[name = string("op_1035_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_1035_end_mask_0 = const()[name = string("op_1035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1035_cast_fp16 = slice_by_index(begin = var_1035_begin_0, end = var_1035_end_0, end_mask = var_1035_end_mask_0, x = value_1_cast_fp16)[name = string("op_1035_cast_fp16")]; + tensor var_1039_begin_0 = const()[name = string("op_1039_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_1039_end_0 = const()[name = string("op_1039_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_1039_end_mask_0 = const()[name = string("op_1039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = value_1_cast_fp16)[name = string("op_1039_cast_fp16")]; + tensor var_1043_begin_0 = const()[name = string("op_1043_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_1043_end_0 = const()[name = string("op_1043_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_1043_end_mask_0 = const()[name = string("op_1043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1043_cast_fp16 = slice_by_index(begin = var_1043_begin_0, end = var_1043_end_0, end_mask = var_1043_end_mask_0, x = value_1_cast_fp16)[name = string("op_1043_cast_fp16")]; + tensor var_1047_begin_0 = const()[name = string("op_1047_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_1047_end_0 = const()[name = string("op_1047_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_1047_end_mask_0 = const()[name = string("op_1047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = value_1_cast_fp16)[name = string("op_1047_cast_fp16")]; + tensor var_1051_begin_0 = const()[name = string("op_1051_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_1051_end_0 = const()[name = string("op_1051_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_1051_end_mask_0 = const()[name = string("op_1051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1051_cast_fp16 = slice_by_index(begin = var_1051_begin_0, end = var_1051_end_0, end_mask = var_1051_end_mask_0, x = value_1_cast_fp16)[name = string("op_1051_cast_fp16")]; + tensor var_1055_begin_0 = const()[name = string("op_1055_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_1055_end_0 = const()[name = string("op_1055_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_1055_end_mask_0 = const()[name = string("op_1055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1055_cast_fp16 = slice_by_index(begin = var_1055_begin_0, end = var_1055_end_0, end_mask = var_1055_end_mask_0, x = value_1_cast_fp16)[name = string("op_1055_cast_fp16")]; + tensor var_1059_begin_0 = const()[name = string("op_1059_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_1059_end_0 = const()[name = string("op_1059_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_1059_end_mask_0 = const()[name = string("op_1059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1059_cast_fp16 = slice_by_index(begin = var_1059_begin_0, end = var_1059_end_0, end_mask = var_1059_end_mask_0, x = value_1_cast_fp16)[name = string("op_1059_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_905_cast_fp16, var_347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_905_cast_fp16, var_354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_905_cast_fp16, var_361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_905_cast_fp16, var_368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_909_cast_fp16, var_375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_909_cast_fp16, var_382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_909_cast_fp16, var_389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_909_cast_fp16, var_396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_913_cast_fp16, var_403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_913_cast_fp16, var_410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_913_cast_fp16, var_417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_913_cast_fp16, var_424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_917_cast_fp16, var_431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_917_cast_fp16, var_438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_917_cast_fp16, var_445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_917_cast_fp16, var_452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_921_cast_fp16, var_459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_921_cast_fp16, var_466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_921_cast_fp16, var_473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_921_cast_fp16, var_480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_925_cast_fp16, var_487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_925_cast_fp16, var_494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_925_cast_fp16, var_501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_925_cast_fp16, var_508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_929_cast_fp16, var_515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_929_cast_fp16, var_522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_929_cast_fp16, var_529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_929_cast_fp16, var_536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_933_cast_fp16, var_543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_933_cast_fp16, var_550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_933_cast_fp16, var_557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_933_cast_fp16, var_564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_937_cast_fp16, var_571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_937_cast_fp16, var_578_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_937_cast_fp16, var_585_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_937_cast_fp16, var_592_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_941_cast_fp16, var_599_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_941_cast_fp16, var_606_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_941_cast_fp16, var_613_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_941_cast_fp16, var_620_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_945_cast_fp16, var_627_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_945_cast_fp16, var_634_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_945_cast_fp16, var_641_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_945_cast_fp16, var_648_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_949_cast_fp16, var_655_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_949_cast_fp16, var_662_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_949_cast_fp16, var_669_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_949_cast_fp16, var_676_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_953_cast_fp16, var_683_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_953_cast_fp16, var_690_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_953_cast_fp16, var_697_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_953_cast_fp16, var_704_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_957_cast_fp16, var_711_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_957_cast_fp16, var_718_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_957_cast_fp16, var_725_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_957_cast_fp16, var_732_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_961_cast_fp16, var_739_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_961_cast_fp16, var_746_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_961_cast_fp16, var_753_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_961_cast_fp16, var_760_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_965_cast_fp16, var_767_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_965_cast_fp16, var_774_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_965_cast_fp16, var_781_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_965_cast_fp16, var_788_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_969_cast_fp16, var_795_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_969_cast_fp16, var_802_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_969_cast_fp16, var_809_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_969_cast_fp16, var_816_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_973_cast_fp16, var_823_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_973_cast_fp16, var_830_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_973_cast_fp16, var_837_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_973_cast_fp16, var_844_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_977_cast_fp16, var_851_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_977_cast_fp16, var_858_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_977_cast_fp16, var_865_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_977_cast_fp16, var_872_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_981_cast_fp16, var_879_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_981_cast_fp16, var_886_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_981_cast_fp16, var_893_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_981_cast_fp16, var_900_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + fp16 var_1222_to_fp16 = const()[name = string("op_1222_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_1222_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_1224_to_fp16 = const()[name = string("op_1224_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_1224_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_1226_to_fp16 = const()[name = string("op_1226_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_1226_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_1228_to_fp16 = const()[name = string("op_1228_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_1228_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_1230_to_fp16 = const()[name = string("op_1230_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_1230_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_1232_to_fp16 = const()[name = string("op_1232_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_1232_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_1234_to_fp16 = const()[name = string("op_1234_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_1234_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_1236_to_fp16 = const()[name = string("op_1236_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_1236_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_1238_to_fp16 = const()[name = string("op_1238_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_1238_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_1240_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_1242_to_fp16 = const()[name = string("op_1242_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_1242_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_1244_to_fp16 = const()[name = string("op_1244_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_1244_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_1246_to_fp16 = const()[name = string("op_1246_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_1246_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_1248_to_fp16 = const()[name = string("op_1248_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_1248_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_1250_to_fp16 = const()[name = string("op_1250_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_1250_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_1252_to_fp16 = const()[name = string("op_1252_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_1252_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_1254_to_fp16 = const()[name = string("op_1254_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_1254_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_1256_to_fp16 = const()[name = string("op_1256_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_1256_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_1258_to_fp16 = const()[name = string("op_1258_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_1258_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_1260_to_fp16 = const()[name = string("op_1260_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_1260_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_1262_to_fp16 = const()[name = string("op_1262_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_1262_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_1264_to_fp16 = const()[name = string("op_1264_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_1264_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_1266_to_fp16 = const()[name = string("op_1266_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_1266_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_1268_to_fp16 = const()[name = string("op_1268_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_1268_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_1270_to_fp16 = const()[name = string("op_1270_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_1270_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_1272_to_fp16 = const()[name = string("op_1272_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_1272_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_1274_to_fp16 = const()[name = string("op_1274_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_1274_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_1276_to_fp16 = const()[name = string("op_1276_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_1276_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_1278_to_fp16 = const()[name = string("op_1278_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_1278_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_1280_to_fp16 = const()[name = string("op_1280_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_1280_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_1282_to_fp16 = const()[name = string("op_1282_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_1282_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_1284_to_fp16 = const()[name = string("op_1284_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_1284_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + fp16 var_1286_to_fp16 = const()[name = string("op_1286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1286_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1288_to_fp16 = const()[name = string("op_1288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1288_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1290_to_fp16 = const()[name = string("op_1290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1290_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1292_to_fp16 = const()[name = string("op_1292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1292_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1294_to_fp16 = const()[name = string("op_1294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1294_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1296_to_fp16 = const()[name = string("op_1296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1296_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1298_to_fp16 = const()[name = string("op_1298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1298_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1300_to_fp16 = const()[name = string("op_1300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1300_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1302_to_fp16 = const()[name = string("op_1302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1302_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1304_to_fp16 = const()[name = string("op_1304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1304_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1306_to_fp16 = const()[name = string("op_1306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1306_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1308_to_fp16 = const()[name = string("op_1308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1308_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1310_to_fp16 = const()[name = string("op_1310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1310_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1312_to_fp16 = const()[name = string("op_1312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1312_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1314_to_fp16 = const()[name = string("op_1314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1314_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1316_to_fp16 = const()[name = string("op_1316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1316_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + fp16 var_1318_to_fp16 = const()[name = string("op_1318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1318_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1320_to_fp16 = const()[name = string("op_1320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1320_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1322_to_fp16 = const()[name = string("op_1322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1322_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1324_to_fp16 = const()[name = string("op_1324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1324_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1326_to_fp16 = const()[name = string("op_1326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1326_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1328_to_fp16 = const()[name = string("op_1328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1328_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1330_to_fp16 = const()[name = string("op_1330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1330_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1332_to_fp16 = const()[name = string("op_1332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1332_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1334_to_fp16 = const()[name = string("op_1334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1334_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1336_to_fp16 = const()[name = string("op_1336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1336_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1338_to_fp16 = const()[name = string("op_1338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1338_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1340_to_fp16 = const()[name = string("op_1340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1340_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1342_to_fp16 = const()[name = string("op_1342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1342_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1344_to_fp16 = const()[name = string("op_1344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1344_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1346_to_fp16 = const()[name = string("op_1346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1346_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1348_to_fp16 = const()[name = string("op_1348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1348_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + fp16 var_1350_to_fp16 = const()[name = string("op_1350_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1350_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1352_to_fp16 = const()[name = string("op_1352_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1352_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1354_to_fp16 = const()[name = string("op_1354_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1354_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1356_to_fp16 = const()[name = string("op_1356_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1356_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1358_to_fp16 = const()[name = string("op_1358_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1358_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1360_to_fp16 = const()[name = string("op_1360_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1360_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1362_to_fp16 = const()[name = string("op_1362_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1362_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1364_to_fp16 = const()[name = string("op_1364_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1364_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1366_to_fp16 = const()[name = string("op_1366_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1366_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1368_to_fp16 = const()[name = string("op_1368_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1368_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1370_to_fp16 = const()[name = string("op_1370_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1370_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1372_to_fp16 = const()[name = string("op_1372_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1372_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1374_to_fp16 = const()[name = string("op_1374_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1374_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1376_to_fp16 = const()[name = string("op_1376_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1376_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1378_to_fp16 = const()[name = string("op_1378_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1378_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1380_to_fp16 = const()[name = string("op_1380_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1380_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + tensor var_1382_cast_fp16 = softmax(axis = var_207, x = aw_chunk_1_cast_fp16)[name = string("op_1382_cast_fp16")]; + tensor var_1383_cast_fp16 = softmax(axis = var_207, x = aw_chunk_3_cast_fp16)[name = string("op_1383_cast_fp16")]; + tensor var_1384_cast_fp16 = softmax(axis = var_207, x = aw_chunk_5_cast_fp16)[name = string("op_1384_cast_fp16")]; + tensor var_1385_cast_fp16 = softmax(axis = var_207, x = aw_chunk_7_cast_fp16)[name = string("op_1385_cast_fp16")]; + tensor var_1386_cast_fp16 = softmax(axis = var_207, x = aw_chunk_9_cast_fp16)[name = string("op_1386_cast_fp16")]; + tensor var_1387_cast_fp16 = softmax(axis = var_207, x = aw_chunk_11_cast_fp16)[name = string("op_1387_cast_fp16")]; + tensor var_1388_cast_fp16 = softmax(axis = var_207, x = aw_chunk_13_cast_fp16)[name = string("op_1388_cast_fp16")]; + tensor var_1389_cast_fp16 = softmax(axis = var_207, x = aw_chunk_15_cast_fp16)[name = string("op_1389_cast_fp16")]; + tensor var_1390_cast_fp16 = softmax(axis = var_207, x = aw_chunk_17_cast_fp16)[name = string("op_1390_cast_fp16")]; + tensor var_1391_cast_fp16 = softmax(axis = var_207, x = aw_chunk_19_cast_fp16)[name = string("op_1391_cast_fp16")]; + tensor var_1392_cast_fp16 = softmax(axis = var_207, x = aw_chunk_21_cast_fp16)[name = string("op_1392_cast_fp16")]; + tensor var_1393_cast_fp16 = softmax(axis = var_207, x = aw_chunk_23_cast_fp16)[name = string("op_1393_cast_fp16")]; + tensor var_1394_cast_fp16 = softmax(axis = var_207, x = aw_chunk_25_cast_fp16)[name = string("op_1394_cast_fp16")]; + tensor var_1395_cast_fp16 = softmax(axis = var_207, x = aw_chunk_27_cast_fp16)[name = string("op_1395_cast_fp16")]; + tensor var_1396_cast_fp16 = softmax(axis = var_207, x = aw_chunk_29_cast_fp16)[name = string("op_1396_cast_fp16")]; + tensor var_1397_cast_fp16 = softmax(axis = var_207, x = aw_chunk_31_cast_fp16)[name = string("op_1397_cast_fp16")]; + tensor var_1398_cast_fp16 = softmax(axis = var_207, x = aw_chunk_33_cast_fp16)[name = string("op_1398_cast_fp16")]; + tensor var_1399_cast_fp16 = softmax(axis = var_207, x = aw_chunk_35_cast_fp16)[name = string("op_1399_cast_fp16")]; + tensor var_1400_cast_fp16 = softmax(axis = var_207, x = aw_chunk_37_cast_fp16)[name = string("op_1400_cast_fp16")]; + tensor var_1401_cast_fp16 = softmax(axis = var_207, x = aw_chunk_39_cast_fp16)[name = string("op_1401_cast_fp16")]; + tensor var_1402_cast_fp16 = softmax(axis = var_207, x = aw_chunk_41_cast_fp16)[name = string("op_1402_cast_fp16")]; + tensor var_1403_cast_fp16 = softmax(axis = var_207, x = aw_chunk_43_cast_fp16)[name = string("op_1403_cast_fp16")]; + tensor var_1404_cast_fp16 = softmax(axis = var_207, x = aw_chunk_45_cast_fp16)[name = string("op_1404_cast_fp16")]; + tensor var_1405_cast_fp16 = softmax(axis = var_207, x = aw_chunk_47_cast_fp16)[name = string("op_1405_cast_fp16")]; + tensor var_1406_cast_fp16 = softmax(axis = var_207, x = aw_chunk_49_cast_fp16)[name = string("op_1406_cast_fp16")]; + tensor var_1407_cast_fp16 = softmax(axis = var_207, x = aw_chunk_51_cast_fp16)[name = string("op_1407_cast_fp16")]; + tensor var_1408_cast_fp16 = softmax(axis = var_207, x = aw_chunk_53_cast_fp16)[name = string("op_1408_cast_fp16")]; + tensor var_1409_cast_fp16 = softmax(axis = var_207, x = aw_chunk_55_cast_fp16)[name = string("op_1409_cast_fp16")]; + tensor var_1410_cast_fp16 = softmax(axis = var_207, x = aw_chunk_57_cast_fp16)[name = string("op_1410_cast_fp16")]; + tensor var_1411_cast_fp16 = softmax(axis = var_207, x = aw_chunk_59_cast_fp16)[name = string("op_1411_cast_fp16")]; + tensor var_1412_cast_fp16 = softmax(axis = var_207, x = aw_chunk_61_cast_fp16)[name = string("op_1412_cast_fp16")]; + tensor var_1413_cast_fp16 = softmax(axis = var_207, x = aw_chunk_63_cast_fp16)[name = string("op_1413_cast_fp16")]; + tensor var_1414_cast_fp16 = softmax(axis = var_207, x = aw_chunk_65_cast_fp16)[name = string("op_1414_cast_fp16")]; + tensor var_1415_cast_fp16 = softmax(axis = var_207, x = aw_chunk_67_cast_fp16)[name = string("op_1415_cast_fp16")]; + tensor var_1416_cast_fp16 = softmax(axis = var_207, x = aw_chunk_69_cast_fp16)[name = string("op_1416_cast_fp16")]; + tensor var_1417_cast_fp16 = softmax(axis = var_207, x = aw_chunk_71_cast_fp16)[name = string("op_1417_cast_fp16")]; + tensor var_1418_cast_fp16 = softmax(axis = var_207, x = aw_chunk_73_cast_fp16)[name = string("op_1418_cast_fp16")]; + tensor var_1419_cast_fp16 = softmax(axis = var_207, x = aw_chunk_75_cast_fp16)[name = string("op_1419_cast_fp16")]; + tensor var_1420_cast_fp16 = softmax(axis = var_207, x = aw_chunk_77_cast_fp16)[name = string("op_1420_cast_fp16")]; + tensor var_1421_cast_fp16 = softmax(axis = var_207, x = aw_chunk_79_cast_fp16)[name = string("op_1421_cast_fp16")]; + tensor var_1422_cast_fp16 = softmax(axis = var_207, x = aw_chunk_81_cast_fp16)[name = string("op_1422_cast_fp16")]; + tensor var_1423_cast_fp16 = softmax(axis = var_207, x = aw_chunk_83_cast_fp16)[name = string("op_1423_cast_fp16")]; + tensor var_1424_cast_fp16 = softmax(axis = var_207, x = aw_chunk_85_cast_fp16)[name = string("op_1424_cast_fp16")]; + tensor var_1425_cast_fp16 = softmax(axis = var_207, x = aw_chunk_87_cast_fp16)[name = string("op_1425_cast_fp16")]; + tensor var_1426_cast_fp16 = softmax(axis = var_207, x = aw_chunk_89_cast_fp16)[name = string("op_1426_cast_fp16")]; + tensor var_1427_cast_fp16 = softmax(axis = var_207, x = aw_chunk_91_cast_fp16)[name = string("op_1427_cast_fp16")]; + tensor var_1428_cast_fp16 = softmax(axis = var_207, x = aw_chunk_93_cast_fp16)[name = string("op_1428_cast_fp16")]; + tensor var_1429_cast_fp16 = softmax(axis = var_207, x = aw_chunk_95_cast_fp16)[name = string("op_1429_cast_fp16")]; + tensor var_1430_cast_fp16 = softmax(axis = var_207, x = aw_chunk_97_cast_fp16)[name = string("op_1430_cast_fp16")]; + tensor var_1431_cast_fp16 = softmax(axis = var_207, x = aw_chunk_99_cast_fp16)[name = string("op_1431_cast_fp16")]; + tensor var_1432_cast_fp16 = softmax(axis = var_207, x = aw_chunk_101_cast_fp16)[name = string("op_1432_cast_fp16")]; + tensor var_1433_cast_fp16 = softmax(axis = var_207, x = aw_chunk_103_cast_fp16)[name = string("op_1433_cast_fp16")]; + tensor var_1434_cast_fp16 = softmax(axis = var_207, x = aw_chunk_105_cast_fp16)[name = string("op_1434_cast_fp16")]; + tensor var_1435_cast_fp16 = softmax(axis = var_207, x = aw_chunk_107_cast_fp16)[name = string("op_1435_cast_fp16")]; + tensor var_1436_cast_fp16 = softmax(axis = var_207, x = aw_chunk_109_cast_fp16)[name = string("op_1436_cast_fp16")]; + tensor var_1437_cast_fp16 = softmax(axis = var_207, x = aw_chunk_111_cast_fp16)[name = string("op_1437_cast_fp16")]; + tensor var_1438_cast_fp16 = softmax(axis = var_207, x = aw_chunk_113_cast_fp16)[name = string("op_1438_cast_fp16")]; + tensor var_1439_cast_fp16 = softmax(axis = var_207, x = aw_chunk_115_cast_fp16)[name = string("op_1439_cast_fp16")]; + tensor var_1440_cast_fp16 = softmax(axis = var_207, x = aw_chunk_117_cast_fp16)[name = string("op_1440_cast_fp16")]; + tensor var_1441_cast_fp16 = softmax(axis = var_207, x = aw_chunk_119_cast_fp16)[name = string("op_1441_cast_fp16")]; + tensor var_1442_cast_fp16 = softmax(axis = var_207, x = aw_chunk_121_cast_fp16)[name = string("op_1442_cast_fp16")]; + tensor var_1443_cast_fp16 = softmax(axis = var_207, x = aw_chunk_123_cast_fp16)[name = string("op_1443_cast_fp16")]; + tensor var_1444_cast_fp16 = softmax(axis = var_207, x = aw_chunk_125_cast_fp16)[name = string("op_1444_cast_fp16")]; + tensor var_1445_cast_fp16 = softmax(axis = var_207, x = aw_chunk_127_cast_fp16)[name = string("op_1445_cast_fp16")]; + tensor var_1446_cast_fp16 = softmax(axis = var_207, x = aw_chunk_129_cast_fp16)[name = string("op_1446_cast_fp16")]; + tensor var_1447_cast_fp16 = softmax(axis = var_207, x = aw_chunk_131_cast_fp16)[name = string("op_1447_cast_fp16")]; + tensor var_1448_cast_fp16 = softmax(axis = var_207, x = aw_chunk_133_cast_fp16)[name = string("op_1448_cast_fp16")]; + tensor var_1449_cast_fp16 = softmax(axis = var_207, x = aw_chunk_135_cast_fp16)[name = string("op_1449_cast_fp16")]; + tensor var_1450_cast_fp16 = softmax(axis = var_207, x = aw_chunk_137_cast_fp16)[name = string("op_1450_cast_fp16")]; + tensor var_1451_cast_fp16 = softmax(axis = var_207, x = aw_chunk_139_cast_fp16)[name = string("op_1451_cast_fp16")]; + tensor var_1452_cast_fp16 = softmax(axis = var_207, x = aw_chunk_141_cast_fp16)[name = string("op_1452_cast_fp16")]; + tensor var_1453_cast_fp16 = softmax(axis = var_207, x = aw_chunk_143_cast_fp16)[name = string("op_1453_cast_fp16")]; + tensor var_1454_cast_fp16 = softmax(axis = var_207, x = aw_chunk_145_cast_fp16)[name = string("op_1454_cast_fp16")]; + tensor var_1455_cast_fp16 = softmax(axis = var_207, x = aw_chunk_147_cast_fp16)[name = string("op_1455_cast_fp16")]; + tensor var_1456_cast_fp16 = softmax(axis = var_207, x = aw_chunk_149_cast_fp16)[name = string("op_1456_cast_fp16")]; + tensor var_1457_cast_fp16 = softmax(axis = var_207, x = aw_chunk_151_cast_fp16)[name = string("op_1457_cast_fp16")]; + tensor var_1458_cast_fp16 = softmax(axis = var_207, x = aw_chunk_153_cast_fp16)[name = string("op_1458_cast_fp16")]; + tensor var_1459_cast_fp16 = softmax(axis = var_207, x = aw_chunk_155_cast_fp16)[name = string("op_1459_cast_fp16")]; + tensor var_1460_cast_fp16 = softmax(axis = var_207, x = aw_chunk_157_cast_fp16)[name = string("op_1460_cast_fp16")]; + tensor var_1461_cast_fp16 = softmax(axis = var_207, x = aw_chunk_159_cast_fp16)[name = string("op_1461_cast_fp16")]; + string var_1463_equation_0 = const()[name = string("op_1463_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1463_cast_fp16 = einsum(equation = var_1463_equation_0, values = (var_983_cast_fp16, var_1382_cast_fp16))[name = string("op_1463_cast_fp16")]; + string var_1465_equation_0 = const()[name = string("op_1465_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1465_cast_fp16 = einsum(equation = var_1465_equation_0, values = (var_983_cast_fp16, var_1383_cast_fp16))[name = string("op_1465_cast_fp16")]; + string var_1467_equation_0 = const()[name = string("op_1467_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1467_cast_fp16 = einsum(equation = var_1467_equation_0, values = (var_983_cast_fp16, var_1384_cast_fp16))[name = string("op_1467_cast_fp16")]; + string var_1469_equation_0 = const()[name = string("op_1469_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1469_cast_fp16 = einsum(equation = var_1469_equation_0, values = (var_983_cast_fp16, var_1385_cast_fp16))[name = string("op_1469_cast_fp16")]; + string var_1471_equation_0 = const()[name = string("op_1471_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1471_cast_fp16 = einsum(equation = var_1471_equation_0, values = (var_987_cast_fp16, var_1386_cast_fp16))[name = string("op_1471_cast_fp16")]; + string var_1473_equation_0 = const()[name = string("op_1473_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1473_cast_fp16 = einsum(equation = var_1473_equation_0, values = (var_987_cast_fp16, var_1387_cast_fp16))[name = string("op_1473_cast_fp16")]; + string var_1475_equation_0 = const()[name = string("op_1475_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1475_cast_fp16 = einsum(equation = var_1475_equation_0, values = (var_987_cast_fp16, var_1388_cast_fp16))[name = string("op_1475_cast_fp16")]; + string var_1477_equation_0 = const()[name = string("op_1477_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1477_cast_fp16 = einsum(equation = var_1477_equation_0, values = (var_987_cast_fp16, var_1389_cast_fp16))[name = string("op_1477_cast_fp16")]; + string var_1479_equation_0 = const()[name = string("op_1479_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1479_cast_fp16 = einsum(equation = var_1479_equation_0, values = (var_991_cast_fp16, var_1390_cast_fp16))[name = string("op_1479_cast_fp16")]; + string var_1481_equation_0 = const()[name = string("op_1481_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1481_cast_fp16 = einsum(equation = var_1481_equation_0, values = (var_991_cast_fp16, var_1391_cast_fp16))[name = string("op_1481_cast_fp16")]; + string var_1483_equation_0 = const()[name = string("op_1483_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1483_cast_fp16 = einsum(equation = var_1483_equation_0, values = (var_991_cast_fp16, var_1392_cast_fp16))[name = string("op_1483_cast_fp16")]; + string var_1485_equation_0 = const()[name = string("op_1485_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1485_cast_fp16 = einsum(equation = var_1485_equation_0, values = (var_991_cast_fp16, var_1393_cast_fp16))[name = string("op_1485_cast_fp16")]; + string var_1487_equation_0 = const()[name = string("op_1487_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1487_cast_fp16 = einsum(equation = var_1487_equation_0, values = (var_995_cast_fp16, var_1394_cast_fp16))[name = string("op_1487_cast_fp16")]; + string var_1489_equation_0 = const()[name = string("op_1489_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1489_cast_fp16 = einsum(equation = var_1489_equation_0, values = (var_995_cast_fp16, var_1395_cast_fp16))[name = string("op_1489_cast_fp16")]; + string var_1491_equation_0 = const()[name = string("op_1491_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1491_cast_fp16 = einsum(equation = var_1491_equation_0, values = (var_995_cast_fp16, var_1396_cast_fp16))[name = string("op_1491_cast_fp16")]; + string var_1493_equation_0 = const()[name = string("op_1493_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1493_cast_fp16 = einsum(equation = var_1493_equation_0, values = (var_995_cast_fp16, var_1397_cast_fp16))[name = string("op_1493_cast_fp16")]; + string var_1495_equation_0 = const()[name = string("op_1495_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1495_cast_fp16 = einsum(equation = var_1495_equation_0, values = (var_999_cast_fp16, var_1398_cast_fp16))[name = string("op_1495_cast_fp16")]; + string var_1497_equation_0 = const()[name = string("op_1497_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1497_cast_fp16 = einsum(equation = var_1497_equation_0, values = (var_999_cast_fp16, var_1399_cast_fp16))[name = string("op_1497_cast_fp16")]; + string var_1499_equation_0 = const()[name = string("op_1499_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1499_cast_fp16 = einsum(equation = var_1499_equation_0, values = (var_999_cast_fp16, var_1400_cast_fp16))[name = string("op_1499_cast_fp16")]; + string var_1501_equation_0 = const()[name = string("op_1501_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1501_cast_fp16 = einsum(equation = var_1501_equation_0, values = (var_999_cast_fp16, var_1401_cast_fp16))[name = string("op_1501_cast_fp16")]; + string var_1503_equation_0 = const()[name = string("op_1503_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1503_cast_fp16 = einsum(equation = var_1503_equation_0, values = (var_1003_cast_fp16, var_1402_cast_fp16))[name = string("op_1503_cast_fp16")]; + string var_1505_equation_0 = const()[name = string("op_1505_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1505_cast_fp16 = einsum(equation = var_1505_equation_0, values = (var_1003_cast_fp16, var_1403_cast_fp16))[name = string("op_1505_cast_fp16")]; + string var_1507_equation_0 = const()[name = string("op_1507_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1507_cast_fp16 = einsum(equation = var_1507_equation_0, values = (var_1003_cast_fp16, var_1404_cast_fp16))[name = string("op_1507_cast_fp16")]; + string var_1509_equation_0 = const()[name = string("op_1509_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1509_cast_fp16 = einsum(equation = var_1509_equation_0, values = (var_1003_cast_fp16, var_1405_cast_fp16))[name = string("op_1509_cast_fp16")]; + string var_1511_equation_0 = const()[name = string("op_1511_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1511_cast_fp16 = einsum(equation = var_1511_equation_0, values = (var_1007_cast_fp16, var_1406_cast_fp16))[name = string("op_1511_cast_fp16")]; + string var_1513_equation_0 = const()[name = string("op_1513_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1513_cast_fp16 = einsum(equation = var_1513_equation_0, values = (var_1007_cast_fp16, var_1407_cast_fp16))[name = string("op_1513_cast_fp16")]; + string var_1515_equation_0 = const()[name = string("op_1515_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1515_cast_fp16 = einsum(equation = var_1515_equation_0, values = (var_1007_cast_fp16, var_1408_cast_fp16))[name = string("op_1515_cast_fp16")]; + string var_1517_equation_0 = const()[name = string("op_1517_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1517_cast_fp16 = einsum(equation = var_1517_equation_0, values = (var_1007_cast_fp16, var_1409_cast_fp16))[name = string("op_1517_cast_fp16")]; + string var_1519_equation_0 = const()[name = string("op_1519_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1519_cast_fp16 = einsum(equation = var_1519_equation_0, values = (var_1011_cast_fp16, var_1410_cast_fp16))[name = string("op_1519_cast_fp16")]; + string var_1521_equation_0 = const()[name = string("op_1521_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1521_cast_fp16 = einsum(equation = var_1521_equation_0, values = (var_1011_cast_fp16, var_1411_cast_fp16))[name = string("op_1521_cast_fp16")]; + string var_1523_equation_0 = const()[name = string("op_1523_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1523_cast_fp16 = einsum(equation = var_1523_equation_0, values = (var_1011_cast_fp16, var_1412_cast_fp16))[name = string("op_1523_cast_fp16")]; + string var_1525_equation_0 = const()[name = string("op_1525_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1525_cast_fp16 = einsum(equation = var_1525_equation_0, values = (var_1011_cast_fp16, var_1413_cast_fp16))[name = string("op_1525_cast_fp16")]; + string var_1527_equation_0 = const()[name = string("op_1527_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1527_cast_fp16 = einsum(equation = var_1527_equation_0, values = (var_1015_cast_fp16, var_1414_cast_fp16))[name = string("op_1527_cast_fp16")]; + string var_1529_equation_0 = const()[name = string("op_1529_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1529_cast_fp16 = einsum(equation = var_1529_equation_0, values = (var_1015_cast_fp16, var_1415_cast_fp16))[name = string("op_1529_cast_fp16")]; + string var_1531_equation_0 = const()[name = string("op_1531_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1531_cast_fp16 = einsum(equation = var_1531_equation_0, values = (var_1015_cast_fp16, var_1416_cast_fp16))[name = string("op_1531_cast_fp16")]; + string var_1533_equation_0 = const()[name = string("op_1533_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1533_cast_fp16 = einsum(equation = var_1533_equation_0, values = (var_1015_cast_fp16, var_1417_cast_fp16))[name = string("op_1533_cast_fp16")]; + string var_1535_equation_0 = const()[name = string("op_1535_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1535_cast_fp16 = einsum(equation = var_1535_equation_0, values = (var_1019_cast_fp16, var_1418_cast_fp16))[name = string("op_1535_cast_fp16")]; + string var_1537_equation_0 = const()[name = string("op_1537_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1537_cast_fp16 = einsum(equation = var_1537_equation_0, values = (var_1019_cast_fp16, var_1419_cast_fp16))[name = string("op_1537_cast_fp16")]; + string var_1539_equation_0 = const()[name = string("op_1539_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1539_cast_fp16 = einsum(equation = var_1539_equation_0, values = (var_1019_cast_fp16, var_1420_cast_fp16))[name = string("op_1539_cast_fp16")]; + string var_1541_equation_0 = const()[name = string("op_1541_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1541_cast_fp16 = einsum(equation = var_1541_equation_0, values = (var_1019_cast_fp16, var_1421_cast_fp16))[name = string("op_1541_cast_fp16")]; + string var_1543_equation_0 = const()[name = string("op_1543_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1543_cast_fp16 = einsum(equation = var_1543_equation_0, values = (var_1023_cast_fp16, var_1422_cast_fp16))[name = string("op_1543_cast_fp16")]; + string var_1545_equation_0 = const()[name = string("op_1545_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1545_cast_fp16 = einsum(equation = var_1545_equation_0, values = (var_1023_cast_fp16, var_1423_cast_fp16))[name = string("op_1545_cast_fp16")]; + string var_1547_equation_0 = const()[name = string("op_1547_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1547_cast_fp16 = einsum(equation = var_1547_equation_0, values = (var_1023_cast_fp16, var_1424_cast_fp16))[name = string("op_1547_cast_fp16")]; + string var_1549_equation_0 = const()[name = string("op_1549_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1549_cast_fp16 = einsum(equation = var_1549_equation_0, values = (var_1023_cast_fp16, var_1425_cast_fp16))[name = string("op_1549_cast_fp16")]; + string var_1551_equation_0 = const()[name = string("op_1551_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1551_cast_fp16 = einsum(equation = var_1551_equation_0, values = (var_1027_cast_fp16, var_1426_cast_fp16))[name = string("op_1551_cast_fp16")]; + string var_1553_equation_0 = const()[name = string("op_1553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1553_cast_fp16 = einsum(equation = var_1553_equation_0, values = (var_1027_cast_fp16, var_1427_cast_fp16))[name = string("op_1553_cast_fp16")]; + string var_1555_equation_0 = const()[name = string("op_1555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1555_cast_fp16 = einsum(equation = var_1555_equation_0, values = (var_1027_cast_fp16, var_1428_cast_fp16))[name = string("op_1555_cast_fp16")]; + string var_1557_equation_0 = const()[name = string("op_1557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1557_cast_fp16 = einsum(equation = var_1557_equation_0, values = (var_1027_cast_fp16, var_1429_cast_fp16))[name = string("op_1557_cast_fp16")]; + string var_1559_equation_0 = const()[name = string("op_1559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1559_cast_fp16 = einsum(equation = var_1559_equation_0, values = (var_1031_cast_fp16, var_1430_cast_fp16))[name = string("op_1559_cast_fp16")]; + string var_1561_equation_0 = const()[name = string("op_1561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1561_cast_fp16 = einsum(equation = var_1561_equation_0, values = (var_1031_cast_fp16, var_1431_cast_fp16))[name = string("op_1561_cast_fp16")]; + string var_1563_equation_0 = const()[name = string("op_1563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1563_cast_fp16 = einsum(equation = var_1563_equation_0, values = (var_1031_cast_fp16, var_1432_cast_fp16))[name = string("op_1563_cast_fp16")]; + string var_1565_equation_0 = const()[name = string("op_1565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1565_cast_fp16 = einsum(equation = var_1565_equation_0, values = (var_1031_cast_fp16, var_1433_cast_fp16))[name = string("op_1565_cast_fp16")]; + string var_1567_equation_0 = const()[name = string("op_1567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1567_cast_fp16 = einsum(equation = var_1567_equation_0, values = (var_1035_cast_fp16, var_1434_cast_fp16))[name = string("op_1567_cast_fp16")]; + string var_1569_equation_0 = const()[name = string("op_1569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1569_cast_fp16 = einsum(equation = var_1569_equation_0, values = (var_1035_cast_fp16, var_1435_cast_fp16))[name = string("op_1569_cast_fp16")]; + string var_1571_equation_0 = const()[name = string("op_1571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1571_cast_fp16 = einsum(equation = var_1571_equation_0, values = (var_1035_cast_fp16, var_1436_cast_fp16))[name = string("op_1571_cast_fp16")]; + string var_1573_equation_0 = const()[name = string("op_1573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1573_cast_fp16 = einsum(equation = var_1573_equation_0, values = (var_1035_cast_fp16, var_1437_cast_fp16))[name = string("op_1573_cast_fp16")]; + string var_1575_equation_0 = const()[name = string("op_1575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1575_cast_fp16 = einsum(equation = var_1575_equation_0, values = (var_1039_cast_fp16, var_1438_cast_fp16))[name = string("op_1575_cast_fp16")]; + string var_1577_equation_0 = const()[name = string("op_1577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1577_cast_fp16 = einsum(equation = var_1577_equation_0, values = (var_1039_cast_fp16, var_1439_cast_fp16))[name = string("op_1577_cast_fp16")]; + string var_1579_equation_0 = const()[name = string("op_1579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1579_cast_fp16 = einsum(equation = var_1579_equation_0, values = (var_1039_cast_fp16, var_1440_cast_fp16))[name = string("op_1579_cast_fp16")]; + string var_1581_equation_0 = const()[name = string("op_1581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1581_cast_fp16 = einsum(equation = var_1581_equation_0, values = (var_1039_cast_fp16, var_1441_cast_fp16))[name = string("op_1581_cast_fp16")]; + string var_1583_equation_0 = const()[name = string("op_1583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1583_cast_fp16 = einsum(equation = var_1583_equation_0, values = (var_1043_cast_fp16, var_1442_cast_fp16))[name = string("op_1583_cast_fp16")]; + string var_1585_equation_0 = const()[name = string("op_1585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1585_cast_fp16 = einsum(equation = var_1585_equation_0, values = (var_1043_cast_fp16, var_1443_cast_fp16))[name = string("op_1585_cast_fp16")]; + string var_1587_equation_0 = const()[name = string("op_1587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1587_cast_fp16 = einsum(equation = var_1587_equation_0, values = (var_1043_cast_fp16, var_1444_cast_fp16))[name = string("op_1587_cast_fp16")]; + string var_1589_equation_0 = const()[name = string("op_1589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1589_cast_fp16 = einsum(equation = var_1589_equation_0, values = (var_1043_cast_fp16, var_1445_cast_fp16))[name = string("op_1589_cast_fp16")]; + string var_1591_equation_0 = const()[name = string("op_1591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1591_cast_fp16 = einsum(equation = var_1591_equation_0, values = (var_1047_cast_fp16, var_1446_cast_fp16))[name = string("op_1591_cast_fp16")]; + string var_1593_equation_0 = const()[name = string("op_1593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1593_cast_fp16 = einsum(equation = var_1593_equation_0, values = (var_1047_cast_fp16, var_1447_cast_fp16))[name = string("op_1593_cast_fp16")]; + string var_1595_equation_0 = const()[name = string("op_1595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1595_cast_fp16 = einsum(equation = var_1595_equation_0, values = (var_1047_cast_fp16, var_1448_cast_fp16))[name = string("op_1595_cast_fp16")]; + string var_1597_equation_0 = const()[name = string("op_1597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1597_cast_fp16 = einsum(equation = var_1597_equation_0, values = (var_1047_cast_fp16, var_1449_cast_fp16))[name = string("op_1597_cast_fp16")]; + string var_1599_equation_0 = const()[name = string("op_1599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1599_cast_fp16 = einsum(equation = var_1599_equation_0, values = (var_1051_cast_fp16, var_1450_cast_fp16))[name = string("op_1599_cast_fp16")]; + string var_1601_equation_0 = const()[name = string("op_1601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1601_cast_fp16 = einsum(equation = var_1601_equation_0, values = (var_1051_cast_fp16, var_1451_cast_fp16))[name = string("op_1601_cast_fp16")]; + string var_1603_equation_0 = const()[name = string("op_1603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1603_cast_fp16 = einsum(equation = var_1603_equation_0, values = (var_1051_cast_fp16, var_1452_cast_fp16))[name = string("op_1603_cast_fp16")]; + string var_1605_equation_0 = const()[name = string("op_1605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1605_cast_fp16 = einsum(equation = var_1605_equation_0, values = (var_1051_cast_fp16, var_1453_cast_fp16))[name = string("op_1605_cast_fp16")]; + string var_1607_equation_0 = const()[name = string("op_1607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1607_cast_fp16 = einsum(equation = var_1607_equation_0, values = (var_1055_cast_fp16, var_1454_cast_fp16))[name = string("op_1607_cast_fp16")]; + string var_1609_equation_0 = const()[name = string("op_1609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1609_cast_fp16 = einsum(equation = var_1609_equation_0, values = (var_1055_cast_fp16, var_1455_cast_fp16))[name = string("op_1609_cast_fp16")]; + string var_1611_equation_0 = const()[name = string("op_1611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1611_cast_fp16 = einsum(equation = var_1611_equation_0, values = (var_1055_cast_fp16, var_1456_cast_fp16))[name = string("op_1611_cast_fp16")]; + string var_1613_equation_0 = const()[name = string("op_1613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1613_cast_fp16 = einsum(equation = var_1613_equation_0, values = (var_1055_cast_fp16, var_1457_cast_fp16))[name = string("op_1613_cast_fp16")]; + string var_1615_equation_0 = const()[name = string("op_1615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1615_cast_fp16 = einsum(equation = var_1615_equation_0, values = (var_1059_cast_fp16, var_1458_cast_fp16))[name = string("op_1615_cast_fp16")]; + string var_1617_equation_0 = const()[name = string("op_1617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1617_cast_fp16 = einsum(equation = var_1617_equation_0, values = (var_1059_cast_fp16, var_1459_cast_fp16))[name = string("op_1617_cast_fp16")]; + string var_1619_equation_0 = const()[name = string("op_1619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1619_cast_fp16 = einsum(equation = var_1619_equation_0, values = (var_1059_cast_fp16, var_1460_cast_fp16))[name = string("op_1619_cast_fp16")]; + string var_1621_equation_0 = const()[name = string("op_1621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1621_cast_fp16 = einsum(equation = var_1621_equation_0, values = (var_1059_cast_fp16, var_1461_cast_fp16))[name = string("op_1621_cast_fp16")]; + bool var_1623_interleave_0 = const()[name = string("op_1623_interleave_0"), val = bool(false)]; + tensor var_1623_cast_fp16 = concat(axis = var_182, interleave = var_1623_interleave_0, values = (var_1463_cast_fp16, var_1465_cast_fp16, var_1467_cast_fp16, var_1469_cast_fp16))[name = string("op_1623_cast_fp16")]; + bool var_1625_interleave_0 = const()[name = string("op_1625_interleave_0"), val = bool(false)]; + tensor var_1625_cast_fp16 = concat(axis = var_182, interleave = var_1625_interleave_0, values = (var_1471_cast_fp16, var_1473_cast_fp16, var_1475_cast_fp16, var_1477_cast_fp16))[name = string("op_1625_cast_fp16")]; + bool var_1627_interleave_0 = const()[name = string("op_1627_interleave_0"), val = bool(false)]; + tensor var_1627_cast_fp16 = concat(axis = var_182, interleave = var_1627_interleave_0, values = (var_1479_cast_fp16, var_1481_cast_fp16, var_1483_cast_fp16, var_1485_cast_fp16))[name = string("op_1627_cast_fp16")]; + bool var_1629_interleave_0 = const()[name = string("op_1629_interleave_0"), val = bool(false)]; + tensor var_1629_cast_fp16 = concat(axis = var_182, interleave = var_1629_interleave_0, values = (var_1487_cast_fp16, var_1489_cast_fp16, var_1491_cast_fp16, var_1493_cast_fp16))[name = string("op_1629_cast_fp16")]; + bool var_1631_interleave_0 = const()[name = string("op_1631_interleave_0"), val = bool(false)]; + tensor var_1631_cast_fp16 = concat(axis = var_182, interleave = var_1631_interleave_0, values = (var_1495_cast_fp16, var_1497_cast_fp16, var_1499_cast_fp16, var_1501_cast_fp16))[name = string("op_1631_cast_fp16")]; + bool var_1633_interleave_0 = const()[name = string("op_1633_interleave_0"), val = bool(false)]; + tensor var_1633_cast_fp16 = concat(axis = var_182, interleave = var_1633_interleave_0, values = (var_1503_cast_fp16, var_1505_cast_fp16, var_1507_cast_fp16, var_1509_cast_fp16))[name = string("op_1633_cast_fp16")]; + bool var_1635_interleave_0 = const()[name = string("op_1635_interleave_0"), val = bool(false)]; + tensor var_1635_cast_fp16 = concat(axis = var_182, interleave = var_1635_interleave_0, values = (var_1511_cast_fp16, var_1513_cast_fp16, var_1515_cast_fp16, var_1517_cast_fp16))[name = string("op_1635_cast_fp16")]; + bool var_1637_interleave_0 = const()[name = string("op_1637_interleave_0"), val = bool(false)]; + tensor var_1637_cast_fp16 = concat(axis = var_182, interleave = var_1637_interleave_0, values = (var_1519_cast_fp16, var_1521_cast_fp16, var_1523_cast_fp16, var_1525_cast_fp16))[name = string("op_1637_cast_fp16")]; + bool var_1639_interleave_0 = const()[name = string("op_1639_interleave_0"), val = bool(false)]; + tensor var_1639_cast_fp16 = concat(axis = var_182, interleave = var_1639_interleave_0, values = (var_1527_cast_fp16, var_1529_cast_fp16, var_1531_cast_fp16, var_1533_cast_fp16))[name = string("op_1639_cast_fp16")]; + bool var_1641_interleave_0 = const()[name = string("op_1641_interleave_0"), val = bool(false)]; + tensor var_1641_cast_fp16 = concat(axis = var_182, interleave = var_1641_interleave_0, values = (var_1535_cast_fp16, var_1537_cast_fp16, var_1539_cast_fp16, var_1541_cast_fp16))[name = string("op_1641_cast_fp16")]; + bool var_1643_interleave_0 = const()[name = string("op_1643_interleave_0"), val = bool(false)]; + tensor var_1643_cast_fp16 = concat(axis = var_182, interleave = var_1643_interleave_0, values = (var_1543_cast_fp16, var_1545_cast_fp16, var_1547_cast_fp16, var_1549_cast_fp16))[name = string("op_1643_cast_fp16")]; + bool var_1645_interleave_0 = const()[name = string("op_1645_interleave_0"), val = bool(false)]; + tensor var_1645_cast_fp16 = concat(axis = var_182, interleave = var_1645_interleave_0, values = (var_1551_cast_fp16, var_1553_cast_fp16, var_1555_cast_fp16, var_1557_cast_fp16))[name = string("op_1645_cast_fp16")]; + bool var_1647_interleave_0 = const()[name = string("op_1647_interleave_0"), val = bool(false)]; + tensor var_1647_cast_fp16 = concat(axis = var_182, interleave = var_1647_interleave_0, values = (var_1559_cast_fp16, var_1561_cast_fp16, var_1563_cast_fp16, var_1565_cast_fp16))[name = string("op_1647_cast_fp16")]; + bool var_1649_interleave_0 = const()[name = string("op_1649_interleave_0"), val = bool(false)]; + tensor var_1649_cast_fp16 = concat(axis = var_182, interleave = var_1649_interleave_0, values = (var_1567_cast_fp16, var_1569_cast_fp16, var_1571_cast_fp16, var_1573_cast_fp16))[name = string("op_1649_cast_fp16")]; + bool var_1651_interleave_0 = const()[name = string("op_1651_interleave_0"), val = bool(false)]; + tensor var_1651_cast_fp16 = concat(axis = var_182, interleave = var_1651_interleave_0, values = (var_1575_cast_fp16, var_1577_cast_fp16, var_1579_cast_fp16, var_1581_cast_fp16))[name = string("op_1651_cast_fp16")]; + bool var_1653_interleave_0 = const()[name = string("op_1653_interleave_0"), val = bool(false)]; + tensor var_1653_cast_fp16 = concat(axis = var_182, interleave = var_1653_interleave_0, values = (var_1583_cast_fp16, var_1585_cast_fp16, var_1587_cast_fp16, var_1589_cast_fp16))[name = string("op_1653_cast_fp16")]; + bool var_1655_interleave_0 = const()[name = string("op_1655_interleave_0"), val = bool(false)]; + tensor var_1655_cast_fp16 = concat(axis = var_182, interleave = var_1655_interleave_0, values = (var_1591_cast_fp16, var_1593_cast_fp16, var_1595_cast_fp16, var_1597_cast_fp16))[name = string("op_1655_cast_fp16")]; + bool var_1657_interleave_0 = const()[name = string("op_1657_interleave_0"), val = bool(false)]; + tensor var_1657_cast_fp16 = concat(axis = var_182, interleave = var_1657_interleave_0, values = (var_1599_cast_fp16, var_1601_cast_fp16, var_1603_cast_fp16, var_1605_cast_fp16))[name = string("op_1657_cast_fp16")]; + bool var_1659_interleave_0 = const()[name = string("op_1659_interleave_0"), val = bool(false)]; + tensor var_1659_cast_fp16 = concat(axis = var_182, interleave = var_1659_interleave_0, values = (var_1607_cast_fp16, var_1609_cast_fp16, var_1611_cast_fp16, var_1613_cast_fp16))[name = string("op_1659_cast_fp16")]; + bool var_1661_interleave_0 = const()[name = string("op_1661_interleave_0"), val = bool(false)]; + tensor var_1661_cast_fp16 = concat(axis = var_182, interleave = var_1661_interleave_0, values = (var_1615_cast_fp16, var_1617_cast_fp16, var_1619_cast_fp16, var_1621_cast_fp16))[name = string("op_1661_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_207, interleave = input_1_interleave_0, values = (var_1623_cast_fp16, var_1625_cast_fp16, var_1627_cast_fp16, var_1629_cast_fp16, var_1631_cast_fp16, var_1633_cast_fp16, var_1635_cast_fp16, var_1637_cast_fp16, var_1639_cast_fp16, var_1641_cast_fp16, var_1643_cast_fp16, var_1645_cast_fp16, var_1647_cast_fp16, var_1649_cast_fp16, var_1651_cast_fp16, var_1653_cast_fp16, var_1655_cast_fp16, var_1657_cast_fp16, var_1659_cast_fp16, var_1661_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24505280)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27782144)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_1680_to_fp16 = const()[name = string("op_1680_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_1680_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27784768)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27787392)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27790016)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40897280)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40907584)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54014848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_1709 = const()[name = string("op_1709"), val = int32(3)]; + int32 var_1734 = const()[name = string("op_1734"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_1751_to_fp16 = const()[name = string("op_1751_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_1751_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54017472)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54020096)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54022720)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57299584)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57302208)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60579072)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63855936)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_1789_begin_0 = const()[name = string("op_1789_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1789_end_0 = const()[name = string("op_1789_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1789_end_mask_0 = const()[name = string("op_1789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1789_cast_fp16 = slice_by_index(begin = var_1789_begin_0, end = var_1789_end_0, end_mask = var_1789_end_mask_0, x = query_3_cast_fp16)[name = string("op_1789_cast_fp16")]; + tensor var_1793_begin_0 = const()[name = string("op_1793_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1793_end_0 = const()[name = string("op_1793_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1793_end_mask_0 = const()[name = string("op_1793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1793_cast_fp16 = slice_by_index(begin = var_1793_begin_0, end = var_1793_end_0, end_mask = var_1793_end_mask_0, x = query_3_cast_fp16)[name = string("op_1793_cast_fp16")]; + tensor var_1797_begin_0 = const()[name = string("op_1797_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1797_end_0 = const()[name = string("op_1797_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1797_end_mask_0 = const()[name = string("op_1797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1797_cast_fp16 = slice_by_index(begin = var_1797_begin_0, end = var_1797_end_0, end_mask = var_1797_end_mask_0, x = query_3_cast_fp16)[name = string("op_1797_cast_fp16")]; + tensor var_1801_begin_0 = const()[name = string("op_1801_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1801_end_0 = const()[name = string("op_1801_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1801_end_mask_0 = const()[name = string("op_1801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1801_cast_fp16 = slice_by_index(begin = var_1801_begin_0, end = var_1801_end_0, end_mask = var_1801_end_mask_0, x = query_3_cast_fp16)[name = string("op_1801_cast_fp16")]; + tensor var_1805_begin_0 = const()[name = string("op_1805_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1805_end_0 = const()[name = string("op_1805_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1805_end_mask_0 = const()[name = string("op_1805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1805_cast_fp16 = slice_by_index(begin = var_1805_begin_0, end = var_1805_end_0, end_mask = var_1805_end_mask_0, x = query_3_cast_fp16)[name = string("op_1805_cast_fp16")]; + tensor var_1809_begin_0 = const()[name = string("op_1809_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1809_end_0 = const()[name = string("op_1809_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1809_end_mask_0 = const()[name = string("op_1809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1809_cast_fp16 = slice_by_index(begin = var_1809_begin_0, end = var_1809_end_0, end_mask = var_1809_end_mask_0, x = query_3_cast_fp16)[name = string("op_1809_cast_fp16")]; + tensor var_1813_begin_0 = const()[name = string("op_1813_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1813_end_0 = const()[name = string("op_1813_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1813_end_mask_0 = const()[name = string("op_1813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1813_cast_fp16 = slice_by_index(begin = var_1813_begin_0, end = var_1813_end_0, end_mask = var_1813_end_mask_0, x = query_3_cast_fp16)[name = string("op_1813_cast_fp16")]; + tensor var_1817_begin_0 = const()[name = string("op_1817_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1817_end_0 = const()[name = string("op_1817_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1817_end_mask_0 = const()[name = string("op_1817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1817_cast_fp16 = slice_by_index(begin = var_1817_begin_0, end = var_1817_end_0, end_mask = var_1817_end_mask_0, x = query_3_cast_fp16)[name = string("op_1817_cast_fp16")]; + tensor var_1821_begin_0 = const()[name = string("op_1821_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1821_end_0 = const()[name = string("op_1821_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1821_end_mask_0 = const()[name = string("op_1821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1821_cast_fp16 = slice_by_index(begin = var_1821_begin_0, end = var_1821_end_0, end_mask = var_1821_end_mask_0, x = query_3_cast_fp16)[name = string("op_1821_cast_fp16")]; + tensor var_1825_begin_0 = const()[name = string("op_1825_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1825_end_0 = const()[name = string("op_1825_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1825_end_mask_0 = const()[name = string("op_1825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1825_cast_fp16 = slice_by_index(begin = var_1825_begin_0, end = var_1825_end_0, end_mask = var_1825_end_mask_0, x = query_3_cast_fp16)[name = string("op_1825_cast_fp16")]; + tensor var_1829_begin_0 = const()[name = string("op_1829_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1829_end_0 = const()[name = string("op_1829_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1829_end_mask_0 = const()[name = string("op_1829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1829_cast_fp16 = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = query_3_cast_fp16)[name = string("op_1829_cast_fp16")]; + tensor var_1833_begin_0 = const()[name = string("op_1833_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1833_end_0 = const()[name = string("op_1833_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1833_end_mask_0 = const()[name = string("op_1833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1833_cast_fp16 = slice_by_index(begin = var_1833_begin_0, end = var_1833_end_0, end_mask = var_1833_end_mask_0, x = query_3_cast_fp16)[name = string("op_1833_cast_fp16")]; + tensor var_1837_begin_0 = const()[name = string("op_1837_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_1837_end_0 = const()[name = string("op_1837_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_1837_end_mask_0 = const()[name = string("op_1837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1837_cast_fp16 = slice_by_index(begin = var_1837_begin_0, end = var_1837_end_0, end_mask = var_1837_end_mask_0, x = query_3_cast_fp16)[name = string("op_1837_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = query_3_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1845_begin_0 = const()[name = string("op_1845_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_1845_end_0 = const()[name = string("op_1845_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_1845_end_mask_0 = const()[name = string("op_1845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1845_cast_fp16 = slice_by_index(begin = var_1845_begin_0, end = var_1845_end_0, end_mask = var_1845_end_mask_0, x = query_3_cast_fp16)[name = string("op_1845_cast_fp16")]; + tensor var_1849_begin_0 = const()[name = string("op_1849_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_1849_end_0 = const()[name = string("op_1849_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_1849_end_mask_0 = const()[name = string("op_1849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1849_cast_fp16 = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, x = query_3_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor var_1853_begin_0 = const()[name = string("op_1853_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_1853_end_0 = const()[name = string("op_1853_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_1853_end_mask_0 = const()[name = string("op_1853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1853_cast_fp16 = slice_by_index(begin = var_1853_begin_0, end = var_1853_end_0, end_mask = var_1853_end_mask_0, x = query_3_cast_fp16)[name = string("op_1853_cast_fp16")]; + tensor var_1857_begin_0 = const()[name = string("op_1857_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_1857_end_0 = const()[name = string("op_1857_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_1857_end_mask_0 = const()[name = string("op_1857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1857_cast_fp16 = slice_by_index(begin = var_1857_begin_0, end = var_1857_end_0, end_mask = var_1857_end_mask_0, x = query_3_cast_fp16)[name = string("op_1857_cast_fp16")]; + tensor var_1861_begin_0 = const()[name = string("op_1861_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_1861_end_0 = const()[name = string("op_1861_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_1861_end_mask_0 = const()[name = string("op_1861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = query_3_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1865_begin_0 = const()[name = string("op_1865_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_1865_end_0 = const()[name = string("op_1865_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_1865_end_mask_0 = const()[name = string("op_1865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1865_cast_fp16 = slice_by_index(begin = var_1865_begin_0, end = var_1865_end_0, end_mask = var_1865_end_mask_0, x = query_3_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1874_begin_0 = const()[name = string("op_1874_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1874_end_0 = const()[name = string("op_1874_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1874_end_mask_0 = const()[name = string("op_1874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1874_cast_fp16 = slice_by_index(begin = var_1874_begin_0, end = var_1874_end_0, end_mask = var_1874_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1874_cast_fp16")]; + tensor var_1881_begin_0 = const()[name = string("op_1881_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1881_end_0 = const()[name = string("op_1881_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1881_end_mask_0 = const()[name = string("op_1881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1881_cast_fp16 = slice_by_index(begin = var_1881_begin_0, end = var_1881_end_0, end_mask = var_1881_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1881_cast_fp16")]; + tensor var_1888_begin_0 = const()[name = string("op_1888_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1888_end_0 = const()[name = string("op_1888_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1888_end_mask_0 = const()[name = string("op_1888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1888_cast_fp16 = slice_by_index(begin = var_1888_begin_0, end = var_1888_end_0, end_mask = var_1888_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1888_cast_fp16")]; + tensor var_1895_begin_0 = const()[name = string("op_1895_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1895_end_0 = const()[name = string("op_1895_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1895_end_mask_0 = const()[name = string("op_1895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1895_cast_fp16 = slice_by_index(begin = var_1895_begin_0, end = var_1895_end_0, end_mask = var_1895_end_mask_0, x = var_1789_cast_fp16)[name = string("op_1895_cast_fp16")]; + tensor var_1902_begin_0 = const()[name = string("op_1902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1902_end_0 = const()[name = string("op_1902_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1902_end_mask_0 = const()[name = string("op_1902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1902_cast_fp16 = slice_by_index(begin = var_1902_begin_0, end = var_1902_end_0, end_mask = var_1902_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1902_cast_fp16")]; + tensor var_1909_begin_0 = const()[name = string("op_1909_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1909_end_0 = const()[name = string("op_1909_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1909_end_mask_0 = const()[name = string("op_1909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1909_cast_fp16 = slice_by_index(begin = var_1909_begin_0, end = var_1909_end_0, end_mask = var_1909_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1909_cast_fp16")]; + tensor var_1916_begin_0 = const()[name = string("op_1916_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1916_end_0 = const()[name = string("op_1916_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1916_end_mask_0 = const()[name = string("op_1916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1916_cast_fp16 = slice_by_index(begin = var_1916_begin_0, end = var_1916_end_0, end_mask = var_1916_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1916_cast_fp16")]; + tensor var_1923_begin_0 = const()[name = string("op_1923_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1923_end_0 = const()[name = string("op_1923_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1923_end_mask_0 = const()[name = string("op_1923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1923_cast_fp16 = slice_by_index(begin = var_1923_begin_0, end = var_1923_end_0, end_mask = var_1923_end_mask_0, x = var_1793_cast_fp16)[name = string("op_1923_cast_fp16")]; + tensor var_1930_begin_0 = const()[name = string("op_1930_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1930_end_0 = const()[name = string("op_1930_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1930_end_mask_0 = const()[name = string("op_1930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1930_cast_fp16 = slice_by_index(begin = var_1930_begin_0, end = var_1930_end_0, end_mask = var_1930_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1930_cast_fp16")]; + tensor var_1937_begin_0 = const()[name = string("op_1937_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1937_end_0 = const()[name = string("op_1937_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1937_end_mask_0 = const()[name = string("op_1937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1937_cast_fp16 = slice_by_index(begin = var_1937_begin_0, end = var_1937_end_0, end_mask = var_1937_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1937_cast_fp16")]; + tensor var_1944_begin_0 = const()[name = string("op_1944_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1944_end_0 = const()[name = string("op_1944_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1944_end_mask_0 = const()[name = string("op_1944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1944_cast_fp16 = slice_by_index(begin = var_1944_begin_0, end = var_1944_end_0, end_mask = var_1944_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1944_cast_fp16")]; + tensor var_1951_begin_0 = const()[name = string("op_1951_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1951_end_0 = const()[name = string("op_1951_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1951_end_mask_0 = const()[name = string("op_1951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1951_cast_fp16 = slice_by_index(begin = var_1951_begin_0, end = var_1951_end_0, end_mask = var_1951_end_mask_0, x = var_1797_cast_fp16)[name = string("op_1951_cast_fp16")]; + tensor var_1958_begin_0 = const()[name = string("op_1958_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1958_end_0 = const()[name = string("op_1958_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1958_end_mask_0 = const()[name = string("op_1958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1958_cast_fp16 = slice_by_index(begin = var_1958_begin_0, end = var_1958_end_0, end_mask = var_1958_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1958_cast_fp16")]; + tensor var_1965_begin_0 = const()[name = string("op_1965_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1965_end_0 = const()[name = string("op_1965_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1965_end_mask_0 = const()[name = string("op_1965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1965_cast_fp16 = slice_by_index(begin = var_1965_begin_0, end = var_1965_end_0, end_mask = var_1965_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1965_cast_fp16")]; + tensor var_1972_begin_0 = const()[name = string("op_1972_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1972_end_0 = const()[name = string("op_1972_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1972_end_mask_0 = const()[name = string("op_1972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1972_cast_fp16 = slice_by_index(begin = var_1972_begin_0, end = var_1972_end_0, end_mask = var_1972_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1972_cast_fp16")]; + tensor var_1979_begin_0 = const()[name = string("op_1979_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1979_end_0 = const()[name = string("op_1979_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1979_end_mask_0 = const()[name = string("op_1979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1979_cast_fp16 = slice_by_index(begin = var_1979_begin_0, end = var_1979_end_0, end_mask = var_1979_end_mask_0, x = var_1801_cast_fp16)[name = string("op_1979_cast_fp16")]; + tensor var_1986_begin_0 = const()[name = string("op_1986_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1986_end_0 = const()[name = string("op_1986_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1986_end_mask_0 = const()[name = string("op_1986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1986_cast_fp16 = slice_by_index(begin = var_1986_begin_0, end = var_1986_end_0, end_mask = var_1986_end_mask_0, x = var_1805_cast_fp16)[name = string("op_1986_cast_fp16")]; + tensor var_1993_begin_0 = const()[name = string("op_1993_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1993_end_0 = const()[name = string("op_1993_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1993_end_mask_0 = const()[name = string("op_1993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1993_cast_fp16 = slice_by_index(begin = var_1993_begin_0, end = var_1993_end_0, end_mask = var_1993_end_mask_0, x = var_1805_cast_fp16)[name = string("op_1993_cast_fp16")]; + tensor var_2000_begin_0 = const()[name = string("op_2000_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2000_end_0 = const()[name = string("op_2000_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2000_end_mask_0 = const()[name = string("op_2000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2000_cast_fp16 = slice_by_index(begin = var_2000_begin_0, end = var_2000_end_0, end_mask = var_2000_end_mask_0, x = var_1805_cast_fp16)[name = string("op_2000_cast_fp16")]; + tensor var_2007_begin_0 = const()[name = string("op_2007_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2007_end_0 = const()[name = string("op_2007_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2007_end_mask_0 = const()[name = string("op_2007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2007_cast_fp16 = slice_by_index(begin = var_2007_begin_0, end = var_2007_end_0, end_mask = var_2007_end_mask_0, x = var_1805_cast_fp16)[name = string("op_2007_cast_fp16")]; + tensor var_2014_begin_0 = const()[name = string("op_2014_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2014_end_0 = const()[name = string("op_2014_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2014_end_mask_0 = const()[name = string("op_2014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2014_cast_fp16 = slice_by_index(begin = var_2014_begin_0, end = var_2014_end_0, end_mask = var_2014_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2014_cast_fp16")]; + tensor var_2021_begin_0 = const()[name = string("op_2021_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2021_end_0 = const()[name = string("op_2021_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2021_end_mask_0 = const()[name = string("op_2021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2021_cast_fp16 = slice_by_index(begin = var_2021_begin_0, end = var_2021_end_0, end_mask = var_2021_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2021_cast_fp16")]; + tensor var_2028_begin_0 = const()[name = string("op_2028_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2028_end_0 = const()[name = string("op_2028_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2028_end_mask_0 = const()[name = string("op_2028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2028_cast_fp16 = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2028_cast_fp16")]; + tensor var_2035_begin_0 = const()[name = string("op_2035_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2035_end_0 = const()[name = string("op_2035_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2035_end_mask_0 = const()[name = string("op_2035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2035_cast_fp16 = slice_by_index(begin = var_2035_begin_0, end = var_2035_end_0, end_mask = var_2035_end_mask_0, x = var_1809_cast_fp16)[name = string("op_2035_cast_fp16")]; + tensor var_2042_begin_0 = const()[name = string("op_2042_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2042_end_0 = const()[name = string("op_2042_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2042_end_mask_0 = const()[name = string("op_2042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2042_cast_fp16 = slice_by_index(begin = var_2042_begin_0, end = var_2042_end_0, end_mask = var_2042_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2042_cast_fp16")]; + tensor var_2049_begin_0 = const()[name = string("op_2049_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2049_end_0 = const()[name = string("op_2049_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2049_end_mask_0 = const()[name = string("op_2049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2049_cast_fp16 = slice_by_index(begin = var_2049_begin_0, end = var_2049_end_0, end_mask = var_2049_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2049_cast_fp16")]; + tensor var_2056_begin_0 = const()[name = string("op_2056_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2056_end_0 = const()[name = string("op_2056_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2056_end_mask_0 = const()[name = string("op_2056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2056_cast_fp16 = slice_by_index(begin = var_2056_begin_0, end = var_2056_end_0, end_mask = var_2056_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2056_cast_fp16")]; + tensor var_2063_begin_0 = const()[name = string("op_2063_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2063_end_0 = const()[name = string("op_2063_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2063_end_mask_0 = const()[name = string("op_2063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2063_cast_fp16 = slice_by_index(begin = var_2063_begin_0, end = var_2063_end_0, end_mask = var_2063_end_mask_0, x = var_1813_cast_fp16)[name = string("op_2063_cast_fp16")]; + tensor var_2070_begin_0 = const()[name = string("op_2070_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2070_end_0 = const()[name = string("op_2070_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2070_end_mask_0 = const()[name = string("op_2070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2070_cast_fp16 = slice_by_index(begin = var_2070_begin_0, end = var_2070_end_0, end_mask = var_2070_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2070_cast_fp16")]; + tensor var_2077_begin_0 = const()[name = string("op_2077_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2077_end_0 = const()[name = string("op_2077_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2077_end_mask_0 = const()[name = string("op_2077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2077_cast_fp16 = slice_by_index(begin = var_2077_begin_0, end = var_2077_end_0, end_mask = var_2077_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2077_cast_fp16")]; + tensor var_2084_begin_0 = const()[name = string("op_2084_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2084_end_0 = const()[name = string("op_2084_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2084_end_mask_0 = const()[name = string("op_2084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2084_cast_fp16 = slice_by_index(begin = var_2084_begin_0, end = var_2084_end_0, end_mask = var_2084_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2084_cast_fp16")]; + tensor var_2091_begin_0 = const()[name = string("op_2091_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2091_end_0 = const()[name = string("op_2091_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2091_end_mask_0 = const()[name = string("op_2091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2091_cast_fp16 = slice_by_index(begin = var_2091_begin_0, end = var_2091_end_0, end_mask = var_2091_end_mask_0, x = var_1817_cast_fp16)[name = string("op_2091_cast_fp16")]; + tensor var_2098_begin_0 = const()[name = string("op_2098_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2098_end_0 = const()[name = string("op_2098_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2098_end_mask_0 = const()[name = string("op_2098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2098_cast_fp16 = slice_by_index(begin = var_2098_begin_0, end = var_2098_end_0, end_mask = var_2098_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2098_cast_fp16")]; + tensor var_2105_begin_0 = const()[name = string("op_2105_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2105_end_0 = const()[name = string("op_2105_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2105_end_mask_0 = const()[name = string("op_2105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2105_cast_fp16 = slice_by_index(begin = var_2105_begin_0, end = var_2105_end_0, end_mask = var_2105_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2105_cast_fp16")]; + tensor var_2112_begin_0 = const()[name = string("op_2112_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2112_end_0 = const()[name = string("op_2112_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2112_end_mask_0 = const()[name = string("op_2112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2112_cast_fp16 = slice_by_index(begin = var_2112_begin_0, end = var_2112_end_0, end_mask = var_2112_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2112_cast_fp16")]; + tensor var_2119_begin_0 = const()[name = string("op_2119_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2119_end_0 = const()[name = string("op_2119_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2119_end_mask_0 = const()[name = string("op_2119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2119_cast_fp16 = slice_by_index(begin = var_2119_begin_0, end = var_2119_end_0, end_mask = var_2119_end_mask_0, x = var_1821_cast_fp16)[name = string("op_2119_cast_fp16")]; + tensor var_2126_begin_0 = const()[name = string("op_2126_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2126_end_0 = const()[name = string("op_2126_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2126_end_mask_0 = const()[name = string("op_2126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2126_cast_fp16 = slice_by_index(begin = var_2126_begin_0, end = var_2126_end_0, end_mask = var_2126_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2126_cast_fp16")]; + tensor var_2133_begin_0 = const()[name = string("op_2133_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2133_end_0 = const()[name = string("op_2133_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2133_end_mask_0 = const()[name = string("op_2133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2133_cast_fp16 = slice_by_index(begin = var_2133_begin_0, end = var_2133_end_0, end_mask = var_2133_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2133_cast_fp16")]; + tensor var_2140_begin_0 = const()[name = string("op_2140_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2140_end_0 = const()[name = string("op_2140_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2140_end_mask_0 = const()[name = string("op_2140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2140_cast_fp16 = slice_by_index(begin = var_2140_begin_0, end = var_2140_end_0, end_mask = var_2140_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2140_cast_fp16")]; + tensor var_2147_begin_0 = const()[name = string("op_2147_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2147_end_0 = const()[name = string("op_2147_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2147_end_mask_0 = const()[name = string("op_2147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2147_cast_fp16 = slice_by_index(begin = var_2147_begin_0, end = var_2147_end_0, end_mask = var_2147_end_mask_0, x = var_1825_cast_fp16)[name = string("op_2147_cast_fp16")]; + tensor var_2154_begin_0 = const()[name = string("op_2154_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2154_end_0 = const()[name = string("op_2154_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2154_end_mask_0 = const()[name = string("op_2154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2154_cast_fp16 = slice_by_index(begin = var_2154_begin_0, end = var_2154_end_0, end_mask = var_2154_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2154_cast_fp16")]; + tensor var_2161_begin_0 = const()[name = string("op_2161_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2161_end_0 = const()[name = string("op_2161_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2161_end_mask_0 = const()[name = string("op_2161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2161_cast_fp16 = slice_by_index(begin = var_2161_begin_0, end = var_2161_end_0, end_mask = var_2161_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2161_cast_fp16")]; + tensor var_2168_begin_0 = const()[name = string("op_2168_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2168_end_0 = const()[name = string("op_2168_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2168_end_mask_0 = const()[name = string("op_2168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2168_cast_fp16 = slice_by_index(begin = var_2168_begin_0, end = var_2168_end_0, end_mask = var_2168_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2168_cast_fp16")]; + tensor var_2175_begin_0 = const()[name = string("op_2175_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2175_end_0 = const()[name = string("op_2175_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2175_end_mask_0 = const()[name = string("op_2175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2175_cast_fp16 = slice_by_index(begin = var_2175_begin_0, end = var_2175_end_0, end_mask = var_2175_end_mask_0, x = var_1829_cast_fp16)[name = string("op_2175_cast_fp16")]; + tensor var_2182_begin_0 = const()[name = string("op_2182_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2182_end_0 = const()[name = string("op_2182_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2182_end_mask_0 = const()[name = string("op_2182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2182_cast_fp16 = slice_by_index(begin = var_2182_begin_0, end = var_2182_end_0, end_mask = var_2182_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2182_cast_fp16")]; + tensor var_2189_begin_0 = const()[name = string("op_2189_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2189_end_0 = const()[name = string("op_2189_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2189_end_mask_0 = const()[name = string("op_2189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2189_cast_fp16 = slice_by_index(begin = var_2189_begin_0, end = var_2189_end_0, end_mask = var_2189_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2189_cast_fp16")]; + tensor var_2196_begin_0 = const()[name = string("op_2196_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2196_end_0 = const()[name = string("op_2196_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2196_end_mask_0 = const()[name = string("op_2196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2196_cast_fp16 = slice_by_index(begin = var_2196_begin_0, end = var_2196_end_0, end_mask = var_2196_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2196_cast_fp16")]; + tensor var_2203_begin_0 = const()[name = string("op_2203_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2203_end_0 = const()[name = string("op_2203_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2203_end_mask_0 = const()[name = string("op_2203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2203_cast_fp16 = slice_by_index(begin = var_2203_begin_0, end = var_2203_end_0, end_mask = var_2203_end_mask_0, x = var_1833_cast_fp16)[name = string("op_2203_cast_fp16")]; + tensor var_2210_begin_0 = const()[name = string("op_2210_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2210_end_0 = const()[name = string("op_2210_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2210_end_mask_0 = const()[name = string("op_2210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2210_cast_fp16 = slice_by_index(begin = var_2210_begin_0, end = var_2210_end_0, end_mask = var_2210_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2210_cast_fp16")]; + tensor var_2217_begin_0 = const()[name = string("op_2217_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2217_end_0 = const()[name = string("op_2217_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2217_end_mask_0 = const()[name = string("op_2217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2217_cast_fp16 = slice_by_index(begin = var_2217_begin_0, end = var_2217_end_0, end_mask = var_2217_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2217_cast_fp16")]; + tensor var_2224_begin_0 = const()[name = string("op_2224_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2224_end_0 = const()[name = string("op_2224_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2224_end_mask_0 = const()[name = string("op_2224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2224_cast_fp16 = slice_by_index(begin = var_2224_begin_0, end = var_2224_end_0, end_mask = var_2224_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2224_cast_fp16")]; + tensor var_2231_begin_0 = const()[name = string("op_2231_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2231_end_0 = const()[name = string("op_2231_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2231_end_mask_0 = const()[name = string("op_2231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, x = var_1837_cast_fp16)[name = string("op_2231_cast_fp16")]; + tensor var_2238_begin_0 = const()[name = string("op_2238_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2238_end_0 = const()[name = string("op_2238_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2238_end_mask_0 = const()[name = string("op_2238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2238_cast_fp16 = slice_by_index(begin = var_2238_begin_0, end = var_2238_end_0, end_mask = var_2238_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2238_cast_fp16")]; + tensor var_2245_begin_0 = const()[name = string("op_2245_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2245_end_0 = const()[name = string("op_2245_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2245_end_mask_0 = const()[name = string("op_2245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2245_cast_fp16 = slice_by_index(begin = var_2245_begin_0, end = var_2245_end_0, end_mask = var_2245_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2245_cast_fp16")]; + tensor var_2252_begin_0 = const()[name = string("op_2252_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2252_end_0 = const()[name = string("op_2252_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2252_end_mask_0 = const()[name = string("op_2252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2252_cast_fp16 = slice_by_index(begin = var_2252_begin_0, end = var_2252_end_0, end_mask = var_2252_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2252_cast_fp16")]; + tensor var_2259_begin_0 = const()[name = string("op_2259_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2259_end_0 = const()[name = string("op_2259_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2259_end_mask_0 = const()[name = string("op_2259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2259_cast_fp16 = slice_by_index(begin = var_2259_begin_0, end = var_2259_end_0, end_mask = var_2259_end_mask_0, x = var_1841_cast_fp16)[name = string("op_2259_cast_fp16")]; + tensor var_2266_begin_0 = const()[name = string("op_2266_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2266_end_0 = const()[name = string("op_2266_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2266_end_mask_0 = const()[name = string("op_2266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2266_cast_fp16 = slice_by_index(begin = var_2266_begin_0, end = var_2266_end_0, end_mask = var_2266_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2266_cast_fp16")]; + tensor var_2273_begin_0 = const()[name = string("op_2273_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2273_end_0 = const()[name = string("op_2273_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2273_end_mask_0 = const()[name = string("op_2273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2273_cast_fp16 = slice_by_index(begin = var_2273_begin_0, end = var_2273_end_0, end_mask = var_2273_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2273_cast_fp16")]; + tensor var_2280_begin_0 = const()[name = string("op_2280_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2280_end_0 = const()[name = string("op_2280_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2280_end_mask_0 = const()[name = string("op_2280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2280_cast_fp16 = slice_by_index(begin = var_2280_begin_0, end = var_2280_end_0, end_mask = var_2280_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2280_cast_fp16")]; + tensor var_2287_begin_0 = const()[name = string("op_2287_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2287_end_0 = const()[name = string("op_2287_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2287_end_mask_0 = const()[name = string("op_2287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2287_cast_fp16 = slice_by_index(begin = var_2287_begin_0, end = var_2287_end_0, end_mask = var_2287_end_mask_0, x = var_1845_cast_fp16)[name = string("op_2287_cast_fp16")]; + tensor var_2294_begin_0 = const()[name = string("op_2294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2294_end_0 = const()[name = string("op_2294_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2294_end_mask_0 = const()[name = string("op_2294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2294_cast_fp16 = slice_by_index(begin = var_2294_begin_0, end = var_2294_end_0, end_mask = var_2294_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2294_cast_fp16")]; + tensor var_2301_begin_0 = const()[name = string("op_2301_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2301_end_0 = const()[name = string("op_2301_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2301_end_mask_0 = const()[name = string("op_2301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2301_cast_fp16 = slice_by_index(begin = var_2301_begin_0, end = var_2301_end_0, end_mask = var_2301_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2301_cast_fp16")]; + tensor var_2308_begin_0 = const()[name = string("op_2308_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2308_end_0 = const()[name = string("op_2308_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2308_end_mask_0 = const()[name = string("op_2308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2308_cast_fp16 = slice_by_index(begin = var_2308_begin_0, end = var_2308_end_0, end_mask = var_2308_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2308_cast_fp16")]; + tensor var_2315_begin_0 = const()[name = string("op_2315_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2315_end_0 = const()[name = string("op_2315_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2315_end_mask_0 = const()[name = string("op_2315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2315_cast_fp16 = slice_by_index(begin = var_2315_begin_0, end = var_2315_end_0, end_mask = var_2315_end_mask_0, x = var_1849_cast_fp16)[name = string("op_2315_cast_fp16")]; + tensor var_2322_begin_0 = const()[name = string("op_2322_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2322_end_0 = const()[name = string("op_2322_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2322_end_mask_0 = const()[name = string("op_2322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2322_cast_fp16 = slice_by_index(begin = var_2322_begin_0, end = var_2322_end_0, end_mask = var_2322_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2322_cast_fp16")]; + tensor var_2329_begin_0 = const()[name = string("op_2329_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2329_end_0 = const()[name = string("op_2329_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2329_end_mask_0 = const()[name = string("op_2329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2329_cast_fp16 = slice_by_index(begin = var_2329_begin_0, end = var_2329_end_0, end_mask = var_2329_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2329_cast_fp16")]; + tensor var_2336_begin_0 = const()[name = string("op_2336_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2336_end_0 = const()[name = string("op_2336_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2336_end_mask_0 = const()[name = string("op_2336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2336_cast_fp16 = slice_by_index(begin = var_2336_begin_0, end = var_2336_end_0, end_mask = var_2336_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2336_cast_fp16")]; + tensor var_2343_begin_0 = const()[name = string("op_2343_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2343_end_0 = const()[name = string("op_2343_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2343_end_mask_0 = const()[name = string("op_2343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2343_cast_fp16 = slice_by_index(begin = var_2343_begin_0, end = var_2343_end_0, end_mask = var_2343_end_mask_0, x = var_1853_cast_fp16)[name = string("op_2343_cast_fp16")]; + tensor var_2350_begin_0 = const()[name = string("op_2350_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2350_end_0 = const()[name = string("op_2350_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2350_end_mask_0 = const()[name = string("op_2350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2350_cast_fp16 = slice_by_index(begin = var_2350_begin_0, end = var_2350_end_0, end_mask = var_2350_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2350_cast_fp16")]; + tensor var_2357_begin_0 = const()[name = string("op_2357_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2357_end_0 = const()[name = string("op_2357_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2357_end_mask_0 = const()[name = string("op_2357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2357_cast_fp16 = slice_by_index(begin = var_2357_begin_0, end = var_2357_end_0, end_mask = var_2357_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2357_cast_fp16")]; + tensor var_2364_begin_0 = const()[name = string("op_2364_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2364_end_0 = const()[name = string("op_2364_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2364_end_mask_0 = const()[name = string("op_2364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2364_cast_fp16 = slice_by_index(begin = var_2364_begin_0, end = var_2364_end_0, end_mask = var_2364_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2364_cast_fp16")]; + tensor var_2371_begin_0 = const()[name = string("op_2371_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2371_end_0 = const()[name = string("op_2371_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2371_end_mask_0 = const()[name = string("op_2371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2371_cast_fp16 = slice_by_index(begin = var_2371_begin_0, end = var_2371_end_0, end_mask = var_2371_end_mask_0, x = var_1857_cast_fp16)[name = string("op_2371_cast_fp16")]; + tensor var_2378_begin_0 = const()[name = string("op_2378_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2378_end_0 = const()[name = string("op_2378_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2378_end_mask_0 = const()[name = string("op_2378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2378_cast_fp16 = slice_by_index(begin = var_2378_begin_0, end = var_2378_end_0, end_mask = var_2378_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2378_cast_fp16")]; + tensor var_2385_begin_0 = const()[name = string("op_2385_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2385_end_0 = const()[name = string("op_2385_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2385_end_mask_0 = const()[name = string("op_2385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2385_cast_fp16 = slice_by_index(begin = var_2385_begin_0, end = var_2385_end_0, end_mask = var_2385_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2385_cast_fp16")]; + tensor var_2392_begin_0 = const()[name = string("op_2392_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2392_end_0 = const()[name = string("op_2392_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2392_end_mask_0 = const()[name = string("op_2392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2392_cast_fp16 = slice_by_index(begin = var_2392_begin_0, end = var_2392_end_0, end_mask = var_2392_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2392_cast_fp16")]; + tensor var_2399_begin_0 = const()[name = string("op_2399_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2399_end_0 = const()[name = string("op_2399_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2399_end_mask_0 = const()[name = string("op_2399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2399_cast_fp16 = slice_by_index(begin = var_2399_begin_0, end = var_2399_end_0, end_mask = var_2399_end_mask_0, x = var_1861_cast_fp16)[name = string("op_2399_cast_fp16")]; + tensor var_2406_begin_0 = const()[name = string("op_2406_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2406_end_0 = const()[name = string("op_2406_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2406_end_mask_0 = const()[name = string("op_2406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2406_cast_fp16 = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = var_1865_cast_fp16)[name = string("op_2406_cast_fp16")]; + tensor var_2413_begin_0 = const()[name = string("op_2413_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2413_end_0 = const()[name = string("op_2413_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2413_end_mask_0 = const()[name = string("op_2413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2413_cast_fp16 = slice_by_index(begin = var_2413_begin_0, end = var_2413_end_0, end_mask = var_2413_end_mask_0, x = var_1865_cast_fp16)[name = string("op_2413_cast_fp16")]; + tensor var_2420_begin_0 = const()[name = string("op_2420_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2420_end_0 = const()[name = string("op_2420_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2420_end_mask_0 = const()[name = string("op_2420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2420_cast_fp16 = slice_by_index(begin = var_2420_begin_0, end = var_2420_end_0, end_mask = var_2420_end_mask_0, x = var_1865_cast_fp16)[name = string("op_2420_cast_fp16")]; + tensor var_2427_begin_0 = const()[name = string("op_2427_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2427_end_0 = const()[name = string("op_2427_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2427_end_mask_0 = const()[name = string("op_2427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2427_cast_fp16 = slice_by_index(begin = var_2427_begin_0, end = var_2427_end_0, end_mask = var_2427_end_mask_0, x = var_1865_cast_fp16)[name = string("op_2427_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_30")]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = k_3_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2436_begin_0 = const()[name = string("op_2436_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2436_end_0 = const()[name = string("op_2436_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2436_end_mask_0 = const()[name = string("op_2436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2436_cast_fp16 = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = k_3_cast_fp16)[name = string("op_2436_cast_fp16")]; + tensor var_2440_begin_0 = const()[name = string("op_2440_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2440_end_0 = const()[name = string("op_2440_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2440_end_mask_0 = const()[name = string("op_2440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2440_cast_fp16 = slice_by_index(begin = var_2440_begin_0, end = var_2440_end_0, end_mask = var_2440_end_mask_0, x = k_3_cast_fp16)[name = string("op_2440_cast_fp16")]; + tensor var_2444_begin_0 = const()[name = string("op_2444_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2444_end_0 = const()[name = string("op_2444_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2444_end_mask_0 = const()[name = string("op_2444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2444_cast_fp16 = slice_by_index(begin = var_2444_begin_0, end = var_2444_end_0, end_mask = var_2444_end_mask_0, x = k_3_cast_fp16)[name = string("op_2444_cast_fp16")]; + tensor var_2448_begin_0 = const()[name = string("op_2448_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2448_end_0 = const()[name = string("op_2448_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2448_end_mask_0 = const()[name = string("op_2448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2448_cast_fp16 = slice_by_index(begin = var_2448_begin_0, end = var_2448_end_0, end_mask = var_2448_end_mask_0, x = k_3_cast_fp16)[name = string("op_2448_cast_fp16")]; + tensor var_2452_begin_0 = const()[name = string("op_2452_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2452_end_0 = const()[name = string("op_2452_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2452_end_mask_0 = const()[name = string("op_2452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2452_cast_fp16 = slice_by_index(begin = var_2452_begin_0, end = var_2452_end_0, end_mask = var_2452_end_mask_0, x = k_3_cast_fp16)[name = string("op_2452_cast_fp16")]; + tensor var_2456_begin_0 = const()[name = string("op_2456_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2456_end_0 = const()[name = string("op_2456_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2456_end_mask_0 = const()[name = string("op_2456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2456_cast_fp16 = slice_by_index(begin = var_2456_begin_0, end = var_2456_end_0, end_mask = var_2456_end_mask_0, x = k_3_cast_fp16)[name = string("op_2456_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = k_3_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2464_begin_0 = const()[name = string("op_2464_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2464_end_0 = const()[name = string("op_2464_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_2464_end_mask_0 = const()[name = string("op_2464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2464_cast_fp16 = slice_by_index(begin = var_2464_begin_0, end = var_2464_end_0, end_mask = var_2464_end_mask_0, x = k_3_cast_fp16)[name = string("op_2464_cast_fp16")]; + tensor var_2468_begin_0 = const()[name = string("op_2468_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2468_end_0 = const()[name = string("op_2468_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_2468_end_mask_0 = const()[name = string("op_2468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2468_cast_fp16 = slice_by_index(begin = var_2468_begin_0, end = var_2468_end_0, end_mask = var_2468_end_mask_0, x = k_3_cast_fp16)[name = string("op_2468_cast_fp16")]; + tensor var_2472_begin_0 = const()[name = string("op_2472_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2472_end_0 = const()[name = string("op_2472_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_2472_end_mask_0 = const()[name = string("op_2472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2472_cast_fp16 = slice_by_index(begin = var_2472_begin_0, end = var_2472_end_0, end_mask = var_2472_end_mask_0, x = k_3_cast_fp16)[name = string("op_2472_cast_fp16")]; + tensor var_2476_begin_0 = const()[name = string("op_2476_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2476_end_0 = const()[name = string("op_2476_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_2476_end_mask_0 = const()[name = string("op_2476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2476_cast_fp16 = slice_by_index(begin = var_2476_begin_0, end = var_2476_end_0, end_mask = var_2476_end_mask_0, x = k_3_cast_fp16)[name = string("op_2476_cast_fp16")]; + tensor var_2480_begin_0 = const()[name = string("op_2480_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_2480_end_0 = const()[name = string("op_2480_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_2480_end_mask_0 = const()[name = string("op_2480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2480_cast_fp16 = slice_by_index(begin = var_2480_begin_0, end = var_2480_end_0, end_mask = var_2480_end_mask_0, x = k_3_cast_fp16)[name = string("op_2480_cast_fp16")]; + tensor var_2484_begin_0 = const()[name = string("op_2484_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_2484_end_0 = const()[name = string("op_2484_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_2484_end_mask_0 = const()[name = string("op_2484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2484_cast_fp16 = slice_by_index(begin = var_2484_begin_0, end = var_2484_end_0, end_mask = var_2484_end_mask_0, x = k_3_cast_fp16)[name = string("op_2484_cast_fp16")]; + tensor var_2488_begin_0 = const()[name = string("op_2488_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_2488_end_0 = const()[name = string("op_2488_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_2488_end_mask_0 = const()[name = string("op_2488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2488_cast_fp16 = slice_by_index(begin = var_2488_begin_0, end = var_2488_end_0, end_mask = var_2488_end_mask_0, x = k_3_cast_fp16)[name = string("op_2488_cast_fp16")]; + tensor var_2492_begin_0 = const()[name = string("op_2492_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_2492_end_0 = const()[name = string("op_2492_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_2492_end_mask_0 = const()[name = string("op_2492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2492_cast_fp16 = slice_by_index(begin = var_2492_begin_0, end = var_2492_end_0, end_mask = var_2492_end_mask_0, x = k_3_cast_fp16)[name = string("op_2492_cast_fp16")]; + tensor var_2496_begin_0 = const()[name = string("op_2496_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2496_end_0 = const()[name = string("op_2496_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_2496_end_mask_0 = const()[name = string("op_2496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2496_cast_fp16 = slice_by_index(begin = var_2496_begin_0, end = var_2496_end_0, end_mask = var_2496_end_mask_0, x = k_3_cast_fp16)[name = string("op_2496_cast_fp16")]; + tensor var_2500_begin_0 = const()[name = string("op_2500_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_2500_end_0 = const()[name = string("op_2500_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_2500_end_mask_0 = const()[name = string("op_2500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2500_cast_fp16 = slice_by_index(begin = var_2500_begin_0, end = var_2500_end_0, end_mask = var_2500_end_mask_0, x = k_3_cast_fp16)[name = string("op_2500_cast_fp16")]; + tensor var_2504_begin_0 = const()[name = string("op_2504_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_2504_end_0 = const()[name = string("op_2504_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_2504_end_mask_0 = const()[name = string("op_2504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2504_cast_fp16 = slice_by_index(begin = var_2504_begin_0, end = var_2504_end_0, end_mask = var_2504_end_mask_0, x = k_3_cast_fp16)[name = string("op_2504_cast_fp16")]; + tensor var_2508_begin_0 = const()[name = string("op_2508_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_2508_end_0 = const()[name = string("op_2508_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_2508_end_mask_0 = const()[name = string("op_2508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2508_cast_fp16 = slice_by_index(begin = var_2508_begin_0, end = var_2508_end_0, end_mask = var_2508_end_mask_0, x = k_3_cast_fp16)[name = string("op_2508_cast_fp16")]; + tensor var_2510_begin_0 = const()[name = string("op_2510_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2510_end_0 = const()[name = string("op_2510_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2510_end_mask_0 = const()[name = string("op_2510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2510_cast_fp16 = slice_by_index(begin = var_2510_begin_0, end = var_2510_end_0, end_mask = var_2510_end_mask_0, x = value_3_cast_fp16)[name = string("op_2510_cast_fp16")]; + tensor var_2514_begin_0 = const()[name = string("op_2514_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2514_end_0 = const()[name = string("op_2514_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2514_end_mask_0 = const()[name = string("op_2514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2514_cast_fp16 = slice_by_index(begin = var_2514_begin_0, end = var_2514_end_0, end_mask = var_2514_end_mask_0, x = value_3_cast_fp16)[name = string("op_2514_cast_fp16")]; + tensor var_2518_begin_0 = const()[name = string("op_2518_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2518_end_0 = const()[name = string("op_2518_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2518_end_mask_0 = const()[name = string("op_2518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2518_cast_fp16 = slice_by_index(begin = var_2518_begin_0, end = var_2518_end_0, end_mask = var_2518_end_mask_0, x = value_3_cast_fp16)[name = string("op_2518_cast_fp16")]; + tensor var_2522_begin_0 = const()[name = string("op_2522_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2522_end_0 = const()[name = string("op_2522_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2522_end_mask_0 = const()[name = string("op_2522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2522_cast_fp16 = slice_by_index(begin = var_2522_begin_0, end = var_2522_end_0, end_mask = var_2522_end_mask_0, x = value_3_cast_fp16)[name = string("op_2522_cast_fp16")]; + tensor var_2526_begin_0 = const()[name = string("op_2526_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2526_end_0 = const()[name = string("op_2526_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2526_end_mask_0 = const()[name = string("op_2526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2526_cast_fp16 = slice_by_index(begin = var_2526_begin_0, end = var_2526_end_0, end_mask = var_2526_end_mask_0, x = value_3_cast_fp16)[name = string("op_2526_cast_fp16")]; + tensor var_2530_begin_0 = const()[name = string("op_2530_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2530_end_0 = const()[name = string("op_2530_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2530_end_mask_0 = const()[name = string("op_2530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2530_cast_fp16 = slice_by_index(begin = var_2530_begin_0, end = var_2530_end_0, end_mask = var_2530_end_mask_0, x = value_3_cast_fp16)[name = string("op_2530_cast_fp16")]; + tensor var_2534_begin_0 = const()[name = string("op_2534_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2534_end_0 = const()[name = string("op_2534_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2534_end_mask_0 = const()[name = string("op_2534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2534_cast_fp16 = slice_by_index(begin = var_2534_begin_0, end = var_2534_end_0, end_mask = var_2534_end_mask_0, x = value_3_cast_fp16)[name = string("op_2534_cast_fp16")]; + tensor var_2538_begin_0 = const()[name = string("op_2538_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2538_end_0 = const()[name = string("op_2538_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2538_end_mask_0 = const()[name = string("op_2538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2538_cast_fp16 = slice_by_index(begin = var_2538_begin_0, end = var_2538_end_0, end_mask = var_2538_end_mask_0, x = value_3_cast_fp16)[name = string("op_2538_cast_fp16")]; + tensor var_2542_begin_0 = const()[name = string("op_2542_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2542_end_0 = const()[name = string("op_2542_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2542_end_mask_0 = const()[name = string("op_2542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2542_cast_fp16 = slice_by_index(begin = var_2542_begin_0, end = var_2542_end_0, end_mask = var_2542_end_mask_0, x = value_3_cast_fp16)[name = string("op_2542_cast_fp16")]; + tensor var_2546_begin_0 = const()[name = string("op_2546_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2546_end_0 = const()[name = string("op_2546_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2546_end_mask_0 = const()[name = string("op_2546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2546_cast_fp16 = slice_by_index(begin = var_2546_begin_0, end = var_2546_end_0, end_mask = var_2546_end_mask_0, x = value_3_cast_fp16)[name = string("op_2546_cast_fp16")]; + tensor var_2550_begin_0 = const()[name = string("op_2550_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2550_end_0 = const()[name = string("op_2550_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2550_end_mask_0 = const()[name = string("op_2550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2550_cast_fp16 = slice_by_index(begin = var_2550_begin_0, end = var_2550_end_0, end_mask = var_2550_end_mask_0, x = value_3_cast_fp16)[name = string("op_2550_cast_fp16")]; + tensor var_2554_begin_0 = const()[name = string("op_2554_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2554_end_0 = const()[name = string("op_2554_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2554_end_mask_0 = const()[name = string("op_2554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2554_cast_fp16 = slice_by_index(begin = var_2554_begin_0, end = var_2554_end_0, end_mask = var_2554_end_mask_0, x = value_3_cast_fp16)[name = string("op_2554_cast_fp16")]; + tensor var_2558_begin_0 = const()[name = string("op_2558_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2558_end_0 = const()[name = string("op_2558_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_2558_end_mask_0 = const()[name = string("op_2558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2558_cast_fp16 = slice_by_index(begin = var_2558_begin_0, end = var_2558_end_0, end_mask = var_2558_end_mask_0, x = value_3_cast_fp16)[name = string("op_2558_cast_fp16")]; + tensor var_2562_begin_0 = const()[name = string("op_2562_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2562_end_0 = const()[name = string("op_2562_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_2562_end_mask_0 = const()[name = string("op_2562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2562_cast_fp16 = slice_by_index(begin = var_2562_begin_0, end = var_2562_end_0, end_mask = var_2562_end_mask_0, x = value_3_cast_fp16)[name = string("op_2562_cast_fp16")]; + tensor var_2566_begin_0 = const()[name = string("op_2566_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2566_end_0 = const()[name = string("op_2566_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_2566_end_mask_0 = const()[name = string("op_2566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2566_cast_fp16 = slice_by_index(begin = var_2566_begin_0, end = var_2566_end_0, end_mask = var_2566_end_mask_0, x = value_3_cast_fp16)[name = string("op_2566_cast_fp16")]; + tensor var_2570_begin_0 = const()[name = string("op_2570_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2570_end_0 = const()[name = string("op_2570_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_2570_end_mask_0 = const()[name = string("op_2570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2570_cast_fp16 = slice_by_index(begin = var_2570_begin_0, end = var_2570_end_0, end_mask = var_2570_end_mask_0, x = value_3_cast_fp16)[name = string("op_2570_cast_fp16")]; + tensor var_2574_begin_0 = const()[name = string("op_2574_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2574_end_0 = const()[name = string("op_2574_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_2574_end_mask_0 = const()[name = string("op_2574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2574_cast_fp16 = slice_by_index(begin = var_2574_begin_0, end = var_2574_end_0, end_mask = var_2574_end_mask_0, x = value_3_cast_fp16)[name = string("op_2574_cast_fp16")]; + tensor var_2578_begin_0 = const()[name = string("op_2578_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2578_end_0 = const()[name = string("op_2578_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_2578_end_mask_0 = const()[name = string("op_2578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2578_cast_fp16 = slice_by_index(begin = var_2578_begin_0, end = var_2578_end_0, end_mask = var_2578_end_mask_0, x = value_3_cast_fp16)[name = string("op_2578_cast_fp16")]; + tensor var_2582_begin_0 = const()[name = string("op_2582_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_2582_end_0 = const()[name = string("op_2582_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_2582_end_mask_0 = const()[name = string("op_2582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2582_cast_fp16 = slice_by_index(begin = var_2582_begin_0, end = var_2582_end_0, end_mask = var_2582_end_mask_0, x = value_3_cast_fp16)[name = string("op_2582_cast_fp16")]; + tensor var_2586_begin_0 = const()[name = string("op_2586_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_2586_end_0 = const()[name = string("op_2586_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_2586_end_mask_0 = const()[name = string("op_2586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2586_cast_fp16 = slice_by_index(begin = var_2586_begin_0, end = var_2586_end_0, end_mask = var_2586_end_mask_0, x = value_3_cast_fp16)[name = string("op_2586_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_2432_cast_fp16, var_1874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_2432_cast_fp16, var_1881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_2432_cast_fp16, var_1888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_2432_cast_fp16, var_1895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_2436_cast_fp16, var_1902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_2436_cast_fp16, var_1909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_2436_cast_fp16, var_1916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_2436_cast_fp16, var_1923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_2440_cast_fp16, var_1930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_2440_cast_fp16, var_1937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_2440_cast_fp16, var_1944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_2440_cast_fp16, var_1951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_2444_cast_fp16, var_1958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_2444_cast_fp16, var_1965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_2444_cast_fp16, var_1972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_2444_cast_fp16, var_1979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2448_cast_fp16, var_1986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2448_cast_fp16, var_1993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2448_cast_fp16, var_2000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2448_cast_fp16, var_2007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2452_cast_fp16, var_2014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2452_cast_fp16, var_2021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2452_cast_fp16, var_2028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2452_cast_fp16, var_2035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2456_cast_fp16, var_2042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2456_cast_fp16, var_2049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2456_cast_fp16, var_2056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2456_cast_fp16, var_2063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2460_cast_fp16, var_2070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2460_cast_fp16, var_2077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2460_cast_fp16, var_2084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2460_cast_fp16, var_2091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2464_cast_fp16, var_2098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2464_cast_fp16, var_2105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2464_cast_fp16, var_2112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2464_cast_fp16, var_2119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2468_cast_fp16, var_2126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2468_cast_fp16, var_2133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2468_cast_fp16, var_2140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2468_cast_fp16, var_2147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2472_cast_fp16, var_2154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2472_cast_fp16, var_2161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2472_cast_fp16, var_2168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2472_cast_fp16, var_2175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2476_cast_fp16, var_2182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2476_cast_fp16, var_2189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2476_cast_fp16, var_2196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2476_cast_fp16, var_2203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_2480_cast_fp16, var_2210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_2480_cast_fp16, var_2217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_2480_cast_fp16, var_2224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_2480_cast_fp16, var_2231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_2484_cast_fp16, var_2238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_2484_cast_fp16, var_2245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_2484_cast_fp16, var_2252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_2484_cast_fp16, var_2259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_2488_cast_fp16, var_2266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_2488_cast_fp16, var_2273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_2488_cast_fp16, var_2280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_2488_cast_fp16, var_2287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_2492_cast_fp16, var_2294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_2492_cast_fp16, var_2301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_2492_cast_fp16, var_2308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_2492_cast_fp16, var_2315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_2496_cast_fp16, var_2322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_2496_cast_fp16, var_2329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_2496_cast_fp16, var_2336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_2496_cast_fp16, var_2343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_2500_cast_fp16, var_2350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_2500_cast_fp16, var_2357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_2500_cast_fp16, var_2364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_2500_cast_fp16, var_2371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_2504_cast_fp16, var_2378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_2504_cast_fp16, var_2385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_2504_cast_fp16, var_2392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_2504_cast_fp16, var_2399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_2508_cast_fp16, var_2406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_2508_cast_fp16, var_2413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_2508_cast_fp16, var_2420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_2508_cast_fp16, var_2427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + fp16 var_2749_to_fp16 = const()[name = string("op_2749_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_2749_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_2751_to_fp16 = const()[name = string("op_2751_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_2751_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_2753_to_fp16 = const()[name = string("op_2753_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_2753_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_2755_to_fp16 = const()[name = string("op_2755_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_2755_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_2757_to_fp16 = const()[name = string("op_2757_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_2757_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_2759_to_fp16 = const()[name = string("op_2759_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_2759_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_2761_to_fp16 = const()[name = string("op_2761_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_2761_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_2763_to_fp16 = const()[name = string("op_2763_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_2763_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_2765_to_fp16 = const()[name = string("op_2765_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_2765_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_2767_to_fp16 = const()[name = string("op_2767_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_2767_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_2769_to_fp16 = const()[name = string("op_2769_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_2769_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_2771_to_fp16 = const()[name = string("op_2771_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_2771_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_2773_to_fp16 = const()[name = string("op_2773_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_2773_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_2775_to_fp16 = const()[name = string("op_2775_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_2775_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_2777_to_fp16 = const()[name = string("op_2777_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_2777_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_2779_to_fp16 = const()[name = string("op_2779_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_2779_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + fp16 var_2781_to_fp16 = const()[name = string("op_2781_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2781_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2783_to_fp16 = const()[name = string("op_2783_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2783_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2785_to_fp16 = const()[name = string("op_2785_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2785_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2787_to_fp16 = const()[name = string("op_2787_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2787_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2789_to_fp16 = const()[name = string("op_2789_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2789_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2791_to_fp16 = const()[name = string("op_2791_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2791_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2793_to_fp16 = const()[name = string("op_2793_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2793_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2795_to_fp16 = const()[name = string("op_2795_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2795_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2797_to_fp16 = const()[name = string("op_2797_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2797_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2799_to_fp16 = const()[name = string("op_2799_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2799_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2801_to_fp16 = const()[name = string("op_2801_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2801_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2803_to_fp16 = const()[name = string("op_2803_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2803_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2805_to_fp16 = const()[name = string("op_2805_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2805_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2807_to_fp16 = const()[name = string("op_2807_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2807_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2809_to_fp16 = const()[name = string("op_2809_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2809_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2811_to_fp16 = const()[name = string("op_2811_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2811_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2813_to_fp16 = const()[name = string("op_2813_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2813_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2815_to_fp16 = const()[name = string("op_2815_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2815_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2817_to_fp16 = const()[name = string("op_2817_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2817_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2819_to_fp16 = const()[name = string("op_2819_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2819_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2821_to_fp16 = const()[name = string("op_2821_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2821_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2823_to_fp16 = const()[name = string("op_2823_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2823_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2825_to_fp16 = const()[name = string("op_2825_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2825_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2827_to_fp16 = const()[name = string("op_2827_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2827_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2829_to_fp16 = const()[name = string("op_2829_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2829_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2831_to_fp16 = const()[name = string("op_2831_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2831_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2833_to_fp16 = const()[name = string("op_2833_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2833_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2835_to_fp16 = const()[name = string("op_2835_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2835_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2837_to_fp16 = const()[name = string("op_2837_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2837_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2839_to_fp16 = const()[name = string("op_2839_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2839_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2841_to_fp16 = const()[name = string("op_2841_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2841_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2843_to_fp16 = const()[name = string("op_2843_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2843_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + fp16 var_2845_to_fp16 = const()[name = string("op_2845_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_2845_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_2847_to_fp16 = const()[name = string("op_2847_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_2847_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_2849_to_fp16 = const()[name = string("op_2849_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_2849_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_2851_to_fp16 = const()[name = string("op_2851_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_2851_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_2853_to_fp16 = const()[name = string("op_2853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_2853_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_2855_to_fp16 = const()[name = string("op_2855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_2855_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_2857_to_fp16 = const()[name = string("op_2857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_2857_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_2859_to_fp16 = const()[name = string("op_2859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_2859_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_2861_to_fp16 = const()[name = string("op_2861_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_2861_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_2863_to_fp16 = const()[name = string("op_2863_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_2863_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_2865_to_fp16 = const()[name = string("op_2865_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_2865_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_2867_to_fp16 = const()[name = string("op_2867_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_2867_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_2869_to_fp16 = const()[name = string("op_2869_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_2869_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_2871_to_fp16 = const()[name = string("op_2871_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_2871_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_2873_to_fp16 = const()[name = string("op_2873_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_2873_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_2875_to_fp16 = const()[name = string("op_2875_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_2875_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + fp16 var_2877_to_fp16 = const()[name = string("op_2877_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_2877_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_2879_to_fp16 = const()[name = string("op_2879_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_2879_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_2881_to_fp16 = const()[name = string("op_2881_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_2881_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_2883_to_fp16 = const()[name = string("op_2883_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_2883_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_2885_to_fp16 = const()[name = string("op_2885_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_2885_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_2887_to_fp16 = const()[name = string("op_2887_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_2887_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_2889_to_fp16 = const()[name = string("op_2889_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_2889_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_2891_to_fp16 = const()[name = string("op_2891_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_2891_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_2893_to_fp16 = const()[name = string("op_2893_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_2893_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_2895_to_fp16 = const()[name = string("op_2895_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_2895_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_2897_to_fp16 = const()[name = string("op_2897_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_2897_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_2899_to_fp16 = const()[name = string("op_2899_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_2899_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_2901_to_fp16 = const()[name = string("op_2901_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_2901_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_2903_to_fp16 = const()[name = string("op_2903_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_2903_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_2905_to_fp16 = const()[name = string("op_2905_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_2905_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_2907_to_fp16 = const()[name = string("op_2907_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_2907_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + tensor var_2909_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_161_cast_fp16)[name = string("op_2909_cast_fp16")]; + tensor var_2910_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_163_cast_fp16)[name = string("op_2910_cast_fp16")]; + tensor var_2911_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_165_cast_fp16)[name = string("op_2911_cast_fp16")]; + tensor var_2912_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_167_cast_fp16)[name = string("op_2912_cast_fp16")]; + tensor var_2913_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_169_cast_fp16)[name = string("op_2913_cast_fp16")]; + tensor var_2914_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_171_cast_fp16)[name = string("op_2914_cast_fp16")]; + tensor var_2915_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_173_cast_fp16)[name = string("op_2915_cast_fp16")]; + tensor var_2916_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_175_cast_fp16)[name = string("op_2916_cast_fp16")]; + tensor var_2917_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_177_cast_fp16)[name = string("op_2917_cast_fp16")]; + tensor var_2918_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_179_cast_fp16)[name = string("op_2918_cast_fp16")]; + tensor var_2919_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_181_cast_fp16)[name = string("op_2919_cast_fp16")]; + tensor var_2920_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_183_cast_fp16)[name = string("op_2920_cast_fp16")]; + tensor var_2921_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_185_cast_fp16)[name = string("op_2921_cast_fp16")]; + tensor var_2922_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_187_cast_fp16)[name = string("op_2922_cast_fp16")]; + tensor var_2923_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_189_cast_fp16)[name = string("op_2923_cast_fp16")]; + tensor var_2924_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_191_cast_fp16)[name = string("op_2924_cast_fp16")]; + tensor var_2925_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_193_cast_fp16)[name = string("op_2925_cast_fp16")]; + tensor var_2926_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_195_cast_fp16)[name = string("op_2926_cast_fp16")]; + tensor var_2927_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_197_cast_fp16)[name = string("op_2927_cast_fp16")]; + tensor var_2928_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_199_cast_fp16)[name = string("op_2928_cast_fp16")]; + tensor var_2929_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_201_cast_fp16)[name = string("op_2929_cast_fp16")]; + tensor var_2930_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_203_cast_fp16)[name = string("op_2930_cast_fp16")]; + tensor var_2931_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_205_cast_fp16)[name = string("op_2931_cast_fp16")]; + tensor var_2932_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_207_cast_fp16)[name = string("op_2932_cast_fp16")]; + tensor var_2933_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_209_cast_fp16)[name = string("op_2933_cast_fp16")]; + tensor var_2934_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_211_cast_fp16)[name = string("op_2934_cast_fp16")]; + tensor var_2935_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_213_cast_fp16)[name = string("op_2935_cast_fp16")]; + tensor var_2936_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_215_cast_fp16)[name = string("op_2936_cast_fp16")]; + tensor var_2937_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_217_cast_fp16)[name = string("op_2937_cast_fp16")]; + tensor var_2938_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_219_cast_fp16)[name = string("op_2938_cast_fp16")]; + tensor var_2939_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_221_cast_fp16)[name = string("op_2939_cast_fp16")]; + tensor var_2940_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_223_cast_fp16)[name = string("op_2940_cast_fp16")]; + tensor var_2941_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_225_cast_fp16)[name = string("op_2941_cast_fp16")]; + tensor var_2942_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_227_cast_fp16)[name = string("op_2942_cast_fp16")]; + tensor var_2943_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_229_cast_fp16)[name = string("op_2943_cast_fp16")]; + tensor var_2944_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_231_cast_fp16)[name = string("op_2944_cast_fp16")]; + tensor var_2945_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_233_cast_fp16)[name = string("op_2945_cast_fp16")]; + tensor var_2946_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_235_cast_fp16)[name = string("op_2946_cast_fp16")]; + tensor var_2947_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_237_cast_fp16)[name = string("op_2947_cast_fp16")]; + tensor var_2948_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_239_cast_fp16)[name = string("op_2948_cast_fp16")]; + tensor var_2949_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_241_cast_fp16)[name = string("op_2949_cast_fp16")]; + tensor var_2950_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_243_cast_fp16)[name = string("op_2950_cast_fp16")]; + tensor var_2951_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_245_cast_fp16)[name = string("op_2951_cast_fp16")]; + tensor var_2952_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_247_cast_fp16)[name = string("op_2952_cast_fp16")]; + tensor var_2953_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_249_cast_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2954_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_251_cast_fp16)[name = string("op_2954_cast_fp16")]; + tensor var_2955_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_253_cast_fp16)[name = string("op_2955_cast_fp16")]; + tensor var_2956_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_255_cast_fp16)[name = string("op_2956_cast_fp16")]; + tensor var_2957_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_257_cast_fp16)[name = string("op_2957_cast_fp16")]; + tensor var_2958_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_259_cast_fp16)[name = string("op_2958_cast_fp16")]; + tensor var_2959_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_261_cast_fp16)[name = string("op_2959_cast_fp16")]; + tensor var_2960_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_263_cast_fp16)[name = string("op_2960_cast_fp16")]; + tensor var_2961_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_265_cast_fp16)[name = string("op_2961_cast_fp16")]; + tensor var_2962_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_267_cast_fp16)[name = string("op_2962_cast_fp16")]; + tensor var_2963_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_269_cast_fp16)[name = string("op_2963_cast_fp16")]; + tensor var_2964_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_271_cast_fp16)[name = string("op_2964_cast_fp16")]; + tensor var_2965_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_273_cast_fp16)[name = string("op_2965_cast_fp16")]; + tensor var_2966_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_275_cast_fp16)[name = string("op_2966_cast_fp16")]; + tensor var_2967_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_277_cast_fp16)[name = string("op_2967_cast_fp16")]; + tensor var_2968_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_279_cast_fp16)[name = string("op_2968_cast_fp16")]; + tensor var_2969_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_281_cast_fp16)[name = string("op_2969_cast_fp16")]; + tensor var_2970_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_283_cast_fp16)[name = string("op_2970_cast_fp16")]; + tensor var_2971_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_285_cast_fp16)[name = string("op_2971_cast_fp16")]; + tensor var_2972_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_287_cast_fp16)[name = string("op_2972_cast_fp16")]; + tensor var_2973_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_289_cast_fp16)[name = string("op_2973_cast_fp16")]; + tensor var_2974_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_291_cast_fp16)[name = string("op_2974_cast_fp16")]; + tensor var_2975_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_293_cast_fp16)[name = string("op_2975_cast_fp16")]; + tensor var_2976_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_295_cast_fp16)[name = string("op_2976_cast_fp16")]; + tensor var_2977_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_297_cast_fp16)[name = string("op_2977_cast_fp16")]; + tensor var_2978_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_299_cast_fp16)[name = string("op_2978_cast_fp16")]; + tensor var_2979_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_301_cast_fp16)[name = string("op_2979_cast_fp16")]; + tensor var_2980_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_303_cast_fp16)[name = string("op_2980_cast_fp16")]; + tensor var_2981_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_305_cast_fp16)[name = string("op_2981_cast_fp16")]; + tensor var_2982_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_307_cast_fp16)[name = string("op_2982_cast_fp16")]; + tensor var_2983_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_309_cast_fp16)[name = string("op_2983_cast_fp16")]; + tensor var_2984_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_311_cast_fp16)[name = string("op_2984_cast_fp16")]; + tensor var_2985_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_313_cast_fp16)[name = string("op_2985_cast_fp16")]; + tensor var_2986_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_315_cast_fp16)[name = string("op_2986_cast_fp16")]; + tensor var_2987_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_317_cast_fp16)[name = string("op_2987_cast_fp16")]; + tensor var_2988_cast_fp16 = softmax(axis = var_1734, x = aw_chunk_319_cast_fp16)[name = string("op_2988_cast_fp16")]; + string var_2990_equation_0 = const()[name = string("op_2990_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2990_cast_fp16 = einsum(equation = var_2990_equation_0, values = (var_2510_cast_fp16, var_2909_cast_fp16))[name = string("op_2990_cast_fp16")]; + string var_2992_equation_0 = const()[name = string("op_2992_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2992_cast_fp16 = einsum(equation = var_2992_equation_0, values = (var_2510_cast_fp16, var_2910_cast_fp16))[name = string("op_2992_cast_fp16")]; + string var_2994_equation_0 = const()[name = string("op_2994_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2994_cast_fp16 = einsum(equation = var_2994_equation_0, values = (var_2510_cast_fp16, var_2911_cast_fp16))[name = string("op_2994_cast_fp16")]; + string var_2996_equation_0 = const()[name = string("op_2996_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2996_cast_fp16 = einsum(equation = var_2996_equation_0, values = (var_2510_cast_fp16, var_2912_cast_fp16))[name = string("op_2996_cast_fp16")]; + string var_2998_equation_0 = const()[name = string("op_2998_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2998_cast_fp16 = einsum(equation = var_2998_equation_0, values = (var_2514_cast_fp16, var_2913_cast_fp16))[name = string("op_2998_cast_fp16")]; + string var_3000_equation_0 = const()[name = string("op_3000_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3000_cast_fp16 = einsum(equation = var_3000_equation_0, values = (var_2514_cast_fp16, var_2914_cast_fp16))[name = string("op_3000_cast_fp16")]; + string var_3002_equation_0 = const()[name = string("op_3002_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3002_cast_fp16 = einsum(equation = var_3002_equation_0, values = (var_2514_cast_fp16, var_2915_cast_fp16))[name = string("op_3002_cast_fp16")]; + string var_3004_equation_0 = const()[name = string("op_3004_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3004_cast_fp16 = einsum(equation = var_3004_equation_0, values = (var_2514_cast_fp16, var_2916_cast_fp16))[name = string("op_3004_cast_fp16")]; + string var_3006_equation_0 = const()[name = string("op_3006_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3006_cast_fp16 = einsum(equation = var_3006_equation_0, values = (var_2518_cast_fp16, var_2917_cast_fp16))[name = string("op_3006_cast_fp16")]; + string var_3008_equation_0 = const()[name = string("op_3008_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3008_cast_fp16 = einsum(equation = var_3008_equation_0, values = (var_2518_cast_fp16, var_2918_cast_fp16))[name = string("op_3008_cast_fp16")]; + string var_3010_equation_0 = const()[name = string("op_3010_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3010_cast_fp16 = einsum(equation = var_3010_equation_0, values = (var_2518_cast_fp16, var_2919_cast_fp16))[name = string("op_3010_cast_fp16")]; + string var_3012_equation_0 = const()[name = string("op_3012_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3012_cast_fp16 = einsum(equation = var_3012_equation_0, values = (var_2518_cast_fp16, var_2920_cast_fp16))[name = string("op_3012_cast_fp16")]; + string var_3014_equation_0 = const()[name = string("op_3014_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3014_cast_fp16 = einsum(equation = var_3014_equation_0, values = (var_2522_cast_fp16, var_2921_cast_fp16))[name = string("op_3014_cast_fp16")]; + string var_3016_equation_0 = const()[name = string("op_3016_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3016_cast_fp16 = einsum(equation = var_3016_equation_0, values = (var_2522_cast_fp16, var_2922_cast_fp16))[name = string("op_3016_cast_fp16")]; + string var_3018_equation_0 = const()[name = string("op_3018_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3018_cast_fp16 = einsum(equation = var_3018_equation_0, values = (var_2522_cast_fp16, var_2923_cast_fp16))[name = string("op_3018_cast_fp16")]; + string var_3020_equation_0 = const()[name = string("op_3020_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3020_cast_fp16 = einsum(equation = var_3020_equation_0, values = (var_2522_cast_fp16, var_2924_cast_fp16))[name = string("op_3020_cast_fp16")]; + string var_3022_equation_0 = const()[name = string("op_3022_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3022_cast_fp16 = einsum(equation = var_3022_equation_0, values = (var_2526_cast_fp16, var_2925_cast_fp16))[name = string("op_3022_cast_fp16")]; + string var_3024_equation_0 = const()[name = string("op_3024_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3024_cast_fp16 = einsum(equation = var_3024_equation_0, values = (var_2526_cast_fp16, var_2926_cast_fp16))[name = string("op_3024_cast_fp16")]; + string var_3026_equation_0 = const()[name = string("op_3026_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3026_cast_fp16 = einsum(equation = var_3026_equation_0, values = (var_2526_cast_fp16, var_2927_cast_fp16))[name = string("op_3026_cast_fp16")]; + string var_3028_equation_0 = const()[name = string("op_3028_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3028_cast_fp16 = einsum(equation = var_3028_equation_0, values = (var_2526_cast_fp16, var_2928_cast_fp16))[name = string("op_3028_cast_fp16")]; + string var_3030_equation_0 = const()[name = string("op_3030_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3030_cast_fp16 = einsum(equation = var_3030_equation_0, values = (var_2530_cast_fp16, var_2929_cast_fp16))[name = string("op_3030_cast_fp16")]; + string var_3032_equation_0 = const()[name = string("op_3032_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3032_cast_fp16 = einsum(equation = var_3032_equation_0, values = (var_2530_cast_fp16, var_2930_cast_fp16))[name = string("op_3032_cast_fp16")]; + string var_3034_equation_0 = const()[name = string("op_3034_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3034_cast_fp16 = einsum(equation = var_3034_equation_0, values = (var_2530_cast_fp16, var_2931_cast_fp16))[name = string("op_3034_cast_fp16")]; + string var_3036_equation_0 = const()[name = string("op_3036_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3036_cast_fp16 = einsum(equation = var_3036_equation_0, values = (var_2530_cast_fp16, var_2932_cast_fp16))[name = string("op_3036_cast_fp16")]; + string var_3038_equation_0 = const()[name = string("op_3038_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3038_cast_fp16 = einsum(equation = var_3038_equation_0, values = (var_2534_cast_fp16, var_2933_cast_fp16))[name = string("op_3038_cast_fp16")]; + string var_3040_equation_0 = const()[name = string("op_3040_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3040_cast_fp16 = einsum(equation = var_3040_equation_0, values = (var_2534_cast_fp16, var_2934_cast_fp16))[name = string("op_3040_cast_fp16")]; + string var_3042_equation_0 = const()[name = string("op_3042_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3042_cast_fp16 = einsum(equation = var_3042_equation_0, values = (var_2534_cast_fp16, var_2935_cast_fp16))[name = string("op_3042_cast_fp16")]; + string var_3044_equation_0 = const()[name = string("op_3044_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3044_cast_fp16 = einsum(equation = var_3044_equation_0, values = (var_2534_cast_fp16, var_2936_cast_fp16))[name = string("op_3044_cast_fp16")]; + string var_3046_equation_0 = const()[name = string("op_3046_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3046_cast_fp16 = einsum(equation = var_3046_equation_0, values = (var_2538_cast_fp16, var_2937_cast_fp16))[name = string("op_3046_cast_fp16")]; + string var_3048_equation_0 = const()[name = string("op_3048_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3048_cast_fp16 = einsum(equation = var_3048_equation_0, values = (var_2538_cast_fp16, var_2938_cast_fp16))[name = string("op_3048_cast_fp16")]; + string var_3050_equation_0 = const()[name = string("op_3050_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3050_cast_fp16 = einsum(equation = var_3050_equation_0, values = (var_2538_cast_fp16, var_2939_cast_fp16))[name = string("op_3050_cast_fp16")]; + string var_3052_equation_0 = const()[name = string("op_3052_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3052_cast_fp16 = einsum(equation = var_3052_equation_0, values = (var_2538_cast_fp16, var_2940_cast_fp16))[name = string("op_3052_cast_fp16")]; + string var_3054_equation_0 = const()[name = string("op_3054_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3054_cast_fp16 = einsum(equation = var_3054_equation_0, values = (var_2542_cast_fp16, var_2941_cast_fp16))[name = string("op_3054_cast_fp16")]; + string var_3056_equation_0 = const()[name = string("op_3056_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3056_cast_fp16 = einsum(equation = var_3056_equation_0, values = (var_2542_cast_fp16, var_2942_cast_fp16))[name = string("op_3056_cast_fp16")]; + string var_3058_equation_0 = const()[name = string("op_3058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3058_cast_fp16 = einsum(equation = var_3058_equation_0, values = (var_2542_cast_fp16, var_2943_cast_fp16))[name = string("op_3058_cast_fp16")]; + string var_3060_equation_0 = const()[name = string("op_3060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3060_cast_fp16 = einsum(equation = var_3060_equation_0, values = (var_2542_cast_fp16, var_2944_cast_fp16))[name = string("op_3060_cast_fp16")]; + string var_3062_equation_0 = const()[name = string("op_3062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3062_cast_fp16 = einsum(equation = var_3062_equation_0, values = (var_2546_cast_fp16, var_2945_cast_fp16))[name = string("op_3062_cast_fp16")]; + string var_3064_equation_0 = const()[name = string("op_3064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3064_cast_fp16 = einsum(equation = var_3064_equation_0, values = (var_2546_cast_fp16, var_2946_cast_fp16))[name = string("op_3064_cast_fp16")]; + string var_3066_equation_0 = const()[name = string("op_3066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3066_cast_fp16 = einsum(equation = var_3066_equation_0, values = (var_2546_cast_fp16, var_2947_cast_fp16))[name = string("op_3066_cast_fp16")]; + string var_3068_equation_0 = const()[name = string("op_3068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3068_cast_fp16 = einsum(equation = var_3068_equation_0, values = (var_2546_cast_fp16, var_2948_cast_fp16))[name = string("op_3068_cast_fp16")]; + string var_3070_equation_0 = const()[name = string("op_3070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3070_cast_fp16 = einsum(equation = var_3070_equation_0, values = (var_2550_cast_fp16, var_2949_cast_fp16))[name = string("op_3070_cast_fp16")]; + string var_3072_equation_0 = const()[name = string("op_3072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3072_cast_fp16 = einsum(equation = var_3072_equation_0, values = (var_2550_cast_fp16, var_2950_cast_fp16))[name = string("op_3072_cast_fp16")]; + string var_3074_equation_0 = const()[name = string("op_3074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3074_cast_fp16 = einsum(equation = var_3074_equation_0, values = (var_2550_cast_fp16, var_2951_cast_fp16))[name = string("op_3074_cast_fp16")]; + string var_3076_equation_0 = const()[name = string("op_3076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3076_cast_fp16 = einsum(equation = var_3076_equation_0, values = (var_2550_cast_fp16, var_2952_cast_fp16))[name = string("op_3076_cast_fp16")]; + string var_3078_equation_0 = const()[name = string("op_3078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3078_cast_fp16 = einsum(equation = var_3078_equation_0, values = (var_2554_cast_fp16, var_2953_cast_fp16))[name = string("op_3078_cast_fp16")]; + string var_3080_equation_0 = const()[name = string("op_3080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3080_cast_fp16 = einsum(equation = var_3080_equation_0, values = (var_2554_cast_fp16, var_2954_cast_fp16))[name = string("op_3080_cast_fp16")]; + string var_3082_equation_0 = const()[name = string("op_3082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3082_cast_fp16 = einsum(equation = var_3082_equation_0, values = (var_2554_cast_fp16, var_2955_cast_fp16))[name = string("op_3082_cast_fp16")]; + string var_3084_equation_0 = const()[name = string("op_3084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3084_cast_fp16 = einsum(equation = var_3084_equation_0, values = (var_2554_cast_fp16, var_2956_cast_fp16))[name = string("op_3084_cast_fp16")]; + string var_3086_equation_0 = const()[name = string("op_3086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3086_cast_fp16 = einsum(equation = var_3086_equation_0, values = (var_2558_cast_fp16, var_2957_cast_fp16))[name = string("op_3086_cast_fp16")]; + string var_3088_equation_0 = const()[name = string("op_3088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3088_cast_fp16 = einsum(equation = var_3088_equation_0, values = (var_2558_cast_fp16, var_2958_cast_fp16))[name = string("op_3088_cast_fp16")]; + string var_3090_equation_0 = const()[name = string("op_3090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3090_cast_fp16 = einsum(equation = var_3090_equation_0, values = (var_2558_cast_fp16, var_2959_cast_fp16))[name = string("op_3090_cast_fp16")]; + string var_3092_equation_0 = const()[name = string("op_3092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3092_cast_fp16 = einsum(equation = var_3092_equation_0, values = (var_2558_cast_fp16, var_2960_cast_fp16))[name = string("op_3092_cast_fp16")]; + string var_3094_equation_0 = const()[name = string("op_3094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3094_cast_fp16 = einsum(equation = var_3094_equation_0, values = (var_2562_cast_fp16, var_2961_cast_fp16))[name = string("op_3094_cast_fp16")]; + string var_3096_equation_0 = const()[name = string("op_3096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3096_cast_fp16 = einsum(equation = var_3096_equation_0, values = (var_2562_cast_fp16, var_2962_cast_fp16))[name = string("op_3096_cast_fp16")]; + string var_3098_equation_0 = const()[name = string("op_3098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3098_cast_fp16 = einsum(equation = var_3098_equation_0, values = (var_2562_cast_fp16, var_2963_cast_fp16))[name = string("op_3098_cast_fp16")]; + string var_3100_equation_0 = const()[name = string("op_3100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3100_cast_fp16 = einsum(equation = var_3100_equation_0, values = (var_2562_cast_fp16, var_2964_cast_fp16))[name = string("op_3100_cast_fp16")]; + string var_3102_equation_0 = const()[name = string("op_3102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3102_cast_fp16 = einsum(equation = var_3102_equation_0, values = (var_2566_cast_fp16, var_2965_cast_fp16))[name = string("op_3102_cast_fp16")]; + string var_3104_equation_0 = const()[name = string("op_3104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3104_cast_fp16 = einsum(equation = var_3104_equation_0, values = (var_2566_cast_fp16, var_2966_cast_fp16))[name = string("op_3104_cast_fp16")]; + string var_3106_equation_0 = const()[name = string("op_3106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3106_cast_fp16 = einsum(equation = var_3106_equation_0, values = (var_2566_cast_fp16, var_2967_cast_fp16))[name = string("op_3106_cast_fp16")]; + string var_3108_equation_0 = const()[name = string("op_3108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3108_cast_fp16 = einsum(equation = var_3108_equation_0, values = (var_2566_cast_fp16, var_2968_cast_fp16))[name = string("op_3108_cast_fp16")]; + string var_3110_equation_0 = const()[name = string("op_3110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3110_cast_fp16 = einsum(equation = var_3110_equation_0, values = (var_2570_cast_fp16, var_2969_cast_fp16))[name = string("op_3110_cast_fp16")]; + string var_3112_equation_0 = const()[name = string("op_3112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3112_cast_fp16 = einsum(equation = var_3112_equation_0, values = (var_2570_cast_fp16, var_2970_cast_fp16))[name = string("op_3112_cast_fp16")]; + string var_3114_equation_0 = const()[name = string("op_3114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3114_cast_fp16 = einsum(equation = var_3114_equation_0, values = (var_2570_cast_fp16, var_2971_cast_fp16))[name = string("op_3114_cast_fp16")]; + string var_3116_equation_0 = const()[name = string("op_3116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3116_cast_fp16 = einsum(equation = var_3116_equation_0, values = (var_2570_cast_fp16, var_2972_cast_fp16))[name = string("op_3116_cast_fp16")]; + string var_3118_equation_0 = const()[name = string("op_3118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3118_cast_fp16 = einsum(equation = var_3118_equation_0, values = (var_2574_cast_fp16, var_2973_cast_fp16))[name = string("op_3118_cast_fp16")]; + string var_3120_equation_0 = const()[name = string("op_3120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3120_cast_fp16 = einsum(equation = var_3120_equation_0, values = (var_2574_cast_fp16, var_2974_cast_fp16))[name = string("op_3120_cast_fp16")]; + string var_3122_equation_0 = const()[name = string("op_3122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3122_cast_fp16 = einsum(equation = var_3122_equation_0, values = (var_2574_cast_fp16, var_2975_cast_fp16))[name = string("op_3122_cast_fp16")]; + string var_3124_equation_0 = const()[name = string("op_3124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3124_cast_fp16 = einsum(equation = var_3124_equation_0, values = (var_2574_cast_fp16, var_2976_cast_fp16))[name = string("op_3124_cast_fp16")]; + string var_3126_equation_0 = const()[name = string("op_3126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3126_cast_fp16 = einsum(equation = var_3126_equation_0, values = (var_2578_cast_fp16, var_2977_cast_fp16))[name = string("op_3126_cast_fp16")]; + string var_3128_equation_0 = const()[name = string("op_3128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3128_cast_fp16 = einsum(equation = var_3128_equation_0, values = (var_2578_cast_fp16, var_2978_cast_fp16))[name = string("op_3128_cast_fp16")]; + string var_3130_equation_0 = const()[name = string("op_3130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3130_cast_fp16 = einsum(equation = var_3130_equation_0, values = (var_2578_cast_fp16, var_2979_cast_fp16))[name = string("op_3130_cast_fp16")]; + string var_3132_equation_0 = const()[name = string("op_3132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3132_cast_fp16 = einsum(equation = var_3132_equation_0, values = (var_2578_cast_fp16, var_2980_cast_fp16))[name = string("op_3132_cast_fp16")]; + string var_3134_equation_0 = const()[name = string("op_3134_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3134_cast_fp16 = einsum(equation = var_3134_equation_0, values = (var_2582_cast_fp16, var_2981_cast_fp16))[name = string("op_3134_cast_fp16")]; + string var_3136_equation_0 = const()[name = string("op_3136_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3136_cast_fp16 = einsum(equation = var_3136_equation_0, values = (var_2582_cast_fp16, var_2982_cast_fp16))[name = string("op_3136_cast_fp16")]; + string var_3138_equation_0 = const()[name = string("op_3138_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3138_cast_fp16 = einsum(equation = var_3138_equation_0, values = (var_2582_cast_fp16, var_2983_cast_fp16))[name = string("op_3138_cast_fp16")]; + string var_3140_equation_0 = const()[name = string("op_3140_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3140_cast_fp16 = einsum(equation = var_3140_equation_0, values = (var_2582_cast_fp16, var_2984_cast_fp16))[name = string("op_3140_cast_fp16")]; + string var_3142_equation_0 = const()[name = string("op_3142_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3142_cast_fp16 = einsum(equation = var_3142_equation_0, values = (var_2586_cast_fp16, var_2985_cast_fp16))[name = string("op_3142_cast_fp16")]; + string var_3144_equation_0 = const()[name = string("op_3144_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3144_cast_fp16 = einsum(equation = var_3144_equation_0, values = (var_2586_cast_fp16, var_2986_cast_fp16))[name = string("op_3144_cast_fp16")]; + string var_3146_equation_0 = const()[name = string("op_3146_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3146_cast_fp16 = einsum(equation = var_3146_equation_0, values = (var_2586_cast_fp16, var_2987_cast_fp16))[name = string("op_3146_cast_fp16")]; + string var_3148_equation_0 = const()[name = string("op_3148_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3148_cast_fp16 = einsum(equation = var_3148_equation_0, values = (var_2586_cast_fp16, var_2988_cast_fp16))[name = string("op_3148_cast_fp16")]; + bool var_3150_interleave_0 = const()[name = string("op_3150_interleave_0"), val = bool(false)]; + tensor var_3150_cast_fp16 = concat(axis = var_1709, interleave = var_3150_interleave_0, values = (var_2990_cast_fp16, var_2992_cast_fp16, var_2994_cast_fp16, var_2996_cast_fp16))[name = string("op_3150_cast_fp16")]; + bool var_3152_interleave_0 = const()[name = string("op_3152_interleave_0"), val = bool(false)]; + tensor var_3152_cast_fp16 = concat(axis = var_1709, interleave = var_3152_interleave_0, values = (var_2998_cast_fp16, var_3000_cast_fp16, var_3002_cast_fp16, var_3004_cast_fp16))[name = string("op_3152_cast_fp16")]; + bool var_3154_interleave_0 = const()[name = string("op_3154_interleave_0"), val = bool(false)]; + tensor var_3154_cast_fp16 = concat(axis = var_1709, interleave = var_3154_interleave_0, values = (var_3006_cast_fp16, var_3008_cast_fp16, var_3010_cast_fp16, var_3012_cast_fp16))[name = string("op_3154_cast_fp16")]; + bool var_3156_interleave_0 = const()[name = string("op_3156_interleave_0"), val = bool(false)]; + tensor var_3156_cast_fp16 = concat(axis = var_1709, interleave = var_3156_interleave_0, values = (var_3014_cast_fp16, var_3016_cast_fp16, var_3018_cast_fp16, var_3020_cast_fp16))[name = string("op_3156_cast_fp16")]; + bool var_3158_interleave_0 = const()[name = string("op_3158_interleave_0"), val = bool(false)]; + tensor var_3158_cast_fp16 = concat(axis = var_1709, interleave = var_3158_interleave_0, values = (var_3022_cast_fp16, var_3024_cast_fp16, var_3026_cast_fp16, var_3028_cast_fp16))[name = string("op_3158_cast_fp16")]; + bool var_3160_interleave_0 = const()[name = string("op_3160_interleave_0"), val = bool(false)]; + tensor var_3160_cast_fp16 = concat(axis = var_1709, interleave = var_3160_interleave_0, values = (var_3030_cast_fp16, var_3032_cast_fp16, var_3034_cast_fp16, var_3036_cast_fp16))[name = string("op_3160_cast_fp16")]; + bool var_3162_interleave_0 = const()[name = string("op_3162_interleave_0"), val = bool(false)]; + tensor var_3162_cast_fp16 = concat(axis = var_1709, interleave = var_3162_interleave_0, values = (var_3038_cast_fp16, var_3040_cast_fp16, var_3042_cast_fp16, var_3044_cast_fp16))[name = string("op_3162_cast_fp16")]; + bool var_3164_interleave_0 = const()[name = string("op_3164_interleave_0"), val = bool(false)]; + tensor var_3164_cast_fp16 = concat(axis = var_1709, interleave = var_3164_interleave_0, values = (var_3046_cast_fp16, var_3048_cast_fp16, var_3050_cast_fp16, var_3052_cast_fp16))[name = string("op_3164_cast_fp16")]; + bool var_3166_interleave_0 = const()[name = string("op_3166_interleave_0"), val = bool(false)]; + tensor var_3166_cast_fp16 = concat(axis = var_1709, interleave = var_3166_interleave_0, values = (var_3054_cast_fp16, var_3056_cast_fp16, var_3058_cast_fp16, var_3060_cast_fp16))[name = string("op_3166_cast_fp16")]; + bool var_3168_interleave_0 = const()[name = string("op_3168_interleave_0"), val = bool(false)]; + tensor var_3168_cast_fp16 = concat(axis = var_1709, interleave = var_3168_interleave_0, values = (var_3062_cast_fp16, var_3064_cast_fp16, var_3066_cast_fp16, var_3068_cast_fp16))[name = string("op_3168_cast_fp16")]; + bool var_3170_interleave_0 = const()[name = string("op_3170_interleave_0"), val = bool(false)]; + tensor var_3170_cast_fp16 = concat(axis = var_1709, interleave = var_3170_interleave_0, values = (var_3070_cast_fp16, var_3072_cast_fp16, var_3074_cast_fp16, var_3076_cast_fp16))[name = string("op_3170_cast_fp16")]; + bool var_3172_interleave_0 = const()[name = string("op_3172_interleave_0"), val = bool(false)]; + tensor var_3172_cast_fp16 = concat(axis = var_1709, interleave = var_3172_interleave_0, values = (var_3078_cast_fp16, var_3080_cast_fp16, var_3082_cast_fp16, var_3084_cast_fp16))[name = string("op_3172_cast_fp16")]; + bool var_3174_interleave_0 = const()[name = string("op_3174_interleave_0"), val = bool(false)]; + tensor var_3174_cast_fp16 = concat(axis = var_1709, interleave = var_3174_interleave_0, values = (var_3086_cast_fp16, var_3088_cast_fp16, var_3090_cast_fp16, var_3092_cast_fp16))[name = string("op_3174_cast_fp16")]; + bool var_3176_interleave_0 = const()[name = string("op_3176_interleave_0"), val = bool(false)]; + tensor var_3176_cast_fp16 = concat(axis = var_1709, interleave = var_3176_interleave_0, values = (var_3094_cast_fp16, var_3096_cast_fp16, var_3098_cast_fp16, var_3100_cast_fp16))[name = string("op_3176_cast_fp16")]; + bool var_3178_interleave_0 = const()[name = string("op_3178_interleave_0"), val = bool(false)]; + tensor var_3178_cast_fp16 = concat(axis = var_1709, interleave = var_3178_interleave_0, values = (var_3102_cast_fp16, var_3104_cast_fp16, var_3106_cast_fp16, var_3108_cast_fp16))[name = string("op_3178_cast_fp16")]; + bool var_3180_interleave_0 = const()[name = string("op_3180_interleave_0"), val = bool(false)]; + tensor var_3180_cast_fp16 = concat(axis = var_1709, interleave = var_3180_interleave_0, values = (var_3110_cast_fp16, var_3112_cast_fp16, var_3114_cast_fp16, var_3116_cast_fp16))[name = string("op_3180_cast_fp16")]; + bool var_3182_interleave_0 = const()[name = string("op_3182_interleave_0"), val = bool(false)]; + tensor var_3182_cast_fp16 = concat(axis = var_1709, interleave = var_3182_interleave_0, values = (var_3118_cast_fp16, var_3120_cast_fp16, var_3122_cast_fp16, var_3124_cast_fp16))[name = string("op_3182_cast_fp16")]; + bool var_3184_interleave_0 = const()[name = string("op_3184_interleave_0"), val = bool(false)]; + tensor var_3184_cast_fp16 = concat(axis = var_1709, interleave = var_3184_interleave_0, values = (var_3126_cast_fp16, var_3128_cast_fp16, var_3130_cast_fp16, var_3132_cast_fp16))[name = string("op_3184_cast_fp16")]; + bool var_3186_interleave_0 = const()[name = string("op_3186_interleave_0"), val = bool(false)]; + tensor var_3186_cast_fp16 = concat(axis = var_1709, interleave = var_3186_interleave_0, values = (var_3134_cast_fp16, var_3136_cast_fp16, var_3138_cast_fp16, var_3140_cast_fp16))[name = string("op_3186_cast_fp16")]; + bool var_3188_interleave_0 = const()[name = string("op_3188_interleave_0"), val = bool(false)]; + tensor var_3188_cast_fp16 = concat(axis = var_1709, interleave = var_3188_interleave_0, values = (var_3142_cast_fp16, var_3144_cast_fp16, var_3146_cast_fp16, var_3148_cast_fp16))[name = string("op_3188_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_1734, interleave = input_9_interleave_0, values = (var_3150_cast_fp16, var_3152_cast_fp16, var_3154_cast_fp16, var_3156_cast_fp16, var_3158_cast_fp16, var_3160_cast_fp16, var_3162_cast_fp16, var_3164_cast_fp16, var_3166_cast_fp16, var_3168_cast_fp16, var_3170_cast_fp16, var_3172_cast_fp16, var_3174_cast_fp16, var_3176_cast_fp16, var_3178_cast_fp16, var_3180_cast_fp16, var_3182_cast_fp16, var_3184_cast_fp16, var_3186_cast_fp16, var_3188_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63858560)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67135424)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_3207_to_fp16 = const()[name = string("op_3207_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_3207_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67138048)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67140672)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67143296)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80250560)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80260864)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93368128)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_3236 = const()[name = string("op_3236"), val = int32(3)]; + int32 var_3261 = const()[name = string("op_3261"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_3278_to_fp16 = const()[name = string("op_3278_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_3278_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93370752)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93373376)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93376000)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96652864)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96655488)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99932352)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103209216)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_3316_begin_0 = const()[name = string("op_3316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3316_end_0 = const()[name = string("op_3316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3316_end_mask_0 = const()[name = string("op_3316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3316_cast_fp16 = slice_by_index(begin = var_3316_begin_0, end = var_3316_end_0, end_mask = var_3316_end_mask_0, x = query_5_cast_fp16)[name = string("op_3316_cast_fp16")]; + tensor var_3320_begin_0 = const()[name = string("op_3320_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3320_end_0 = const()[name = string("op_3320_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3320_end_mask_0 = const()[name = string("op_3320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3320_cast_fp16 = slice_by_index(begin = var_3320_begin_0, end = var_3320_end_0, end_mask = var_3320_end_mask_0, x = query_5_cast_fp16)[name = string("op_3320_cast_fp16")]; + tensor var_3324_begin_0 = const()[name = string("op_3324_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3324_end_0 = const()[name = string("op_3324_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3324_end_mask_0 = const()[name = string("op_3324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3324_cast_fp16 = slice_by_index(begin = var_3324_begin_0, end = var_3324_end_0, end_mask = var_3324_end_mask_0, x = query_5_cast_fp16)[name = string("op_3324_cast_fp16")]; + tensor var_3328_begin_0 = const()[name = string("op_3328_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3328_end_0 = const()[name = string("op_3328_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3328_end_mask_0 = const()[name = string("op_3328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3328_cast_fp16 = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = query_5_cast_fp16)[name = string("op_3328_cast_fp16")]; + tensor var_3332_begin_0 = const()[name = string("op_3332_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3332_end_0 = const()[name = string("op_3332_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3332_end_mask_0 = const()[name = string("op_3332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3332_cast_fp16 = slice_by_index(begin = var_3332_begin_0, end = var_3332_end_0, end_mask = var_3332_end_mask_0, x = query_5_cast_fp16)[name = string("op_3332_cast_fp16")]; + tensor var_3336_begin_0 = const()[name = string("op_3336_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3336_end_0 = const()[name = string("op_3336_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3336_end_mask_0 = const()[name = string("op_3336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3336_cast_fp16 = slice_by_index(begin = var_3336_begin_0, end = var_3336_end_0, end_mask = var_3336_end_mask_0, x = query_5_cast_fp16)[name = string("op_3336_cast_fp16")]; + tensor var_3340_begin_0 = const()[name = string("op_3340_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3340_end_0 = const()[name = string("op_3340_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3340_end_mask_0 = const()[name = string("op_3340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3340_cast_fp16 = slice_by_index(begin = var_3340_begin_0, end = var_3340_end_0, end_mask = var_3340_end_mask_0, x = query_5_cast_fp16)[name = string("op_3340_cast_fp16")]; + tensor var_3344_begin_0 = const()[name = string("op_3344_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3344_end_0 = const()[name = string("op_3344_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3344_end_mask_0 = const()[name = string("op_3344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3344_cast_fp16 = slice_by_index(begin = var_3344_begin_0, end = var_3344_end_0, end_mask = var_3344_end_mask_0, x = query_5_cast_fp16)[name = string("op_3344_cast_fp16")]; + tensor var_3348_begin_0 = const()[name = string("op_3348_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3348_end_0 = const()[name = string("op_3348_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3348_end_mask_0 = const()[name = string("op_3348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3348_cast_fp16 = slice_by_index(begin = var_3348_begin_0, end = var_3348_end_0, end_mask = var_3348_end_mask_0, x = query_5_cast_fp16)[name = string("op_3348_cast_fp16")]; + tensor var_3352_begin_0 = const()[name = string("op_3352_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3352_end_0 = const()[name = string("op_3352_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3352_end_mask_0 = const()[name = string("op_3352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3352_cast_fp16 = slice_by_index(begin = var_3352_begin_0, end = var_3352_end_0, end_mask = var_3352_end_mask_0, x = query_5_cast_fp16)[name = string("op_3352_cast_fp16")]; + tensor var_3356_begin_0 = const()[name = string("op_3356_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3356_end_0 = const()[name = string("op_3356_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3356_end_mask_0 = const()[name = string("op_3356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3356_cast_fp16 = slice_by_index(begin = var_3356_begin_0, end = var_3356_end_0, end_mask = var_3356_end_mask_0, x = query_5_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3360_begin_0 = const()[name = string("op_3360_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3360_end_0 = const()[name = string("op_3360_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3360_end_mask_0 = const()[name = string("op_3360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3360_cast_fp16 = slice_by_index(begin = var_3360_begin_0, end = var_3360_end_0, end_mask = var_3360_end_mask_0, x = query_5_cast_fp16)[name = string("op_3360_cast_fp16")]; + tensor var_3364_begin_0 = const()[name = string("op_3364_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3364_end_0 = const()[name = string("op_3364_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_3364_end_mask_0 = const()[name = string("op_3364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3364_cast_fp16 = slice_by_index(begin = var_3364_begin_0, end = var_3364_end_0, end_mask = var_3364_end_mask_0, x = query_5_cast_fp16)[name = string("op_3364_cast_fp16")]; + tensor var_3368_begin_0 = const()[name = string("op_3368_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3368_end_0 = const()[name = string("op_3368_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_3368_end_mask_0 = const()[name = string("op_3368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3368_cast_fp16 = slice_by_index(begin = var_3368_begin_0, end = var_3368_end_0, end_mask = var_3368_end_mask_0, x = query_5_cast_fp16)[name = string("op_3368_cast_fp16")]; + tensor var_3372_begin_0 = const()[name = string("op_3372_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3372_end_0 = const()[name = string("op_3372_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_3372_end_mask_0 = const()[name = string("op_3372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3372_cast_fp16 = slice_by_index(begin = var_3372_begin_0, end = var_3372_end_0, end_mask = var_3372_end_mask_0, x = query_5_cast_fp16)[name = string("op_3372_cast_fp16")]; + tensor var_3376_begin_0 = const()[name = string("op_3376_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3376_end_0 = const()[name = string("op_3376_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_3376_end_mask_0 = const()[name = string("op_3376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3376_cast_fp16 = slice_by_index(begin = var_3376_begin_0, end = var_3376_end_0, end_mask = var_3376_end_mask_0, x = query_5_cast_fp16)[name = string("op_3376_cast_fp16")]; + tensor var_3380_begin_0 = const()[name = string("op_3380_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3380_end_0 = const()[name = string("op_3380_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_3380_end_mask_0 = const()[name = string("op_3380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3380_cast_fp16 = slice_by_index(begin = var_3380_begin_0, end = var_3380_end_0, end_mask = var_3380_end_mask_0, x = query_5_cast_fp16)[name = string("op_3380_cast_fp16")]; + tensor var_3384_begin_0 = const()[name = string("op_3384_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3384_end_0 = const()[name = string("op_3384_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_3384_end_mask_0 = const()[name = string("op_3384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3384_cast_fp16 = slice_by_index(begin = var_3384_begin_0, end = var_3384_end_0, end_mask = var_3384_end_mask_0, x = query_5_cast_fp16)[name = string("op_3384_cast_fp16")]; + tensor var_3388_begin_0 = const()[name = string("op_3388_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3388_end_0 = const()[name = string("op_3388_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_3388_end_mask_0 = const()[name = string("op_3388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3388_cast_fp16 = slice_by_index(begin = var_3388_begin_0, end = var_3388_end_0, end_mask = var_3388_end_mask_0, x = query_5_cast_fp16)[name = string("op_3388_cast_fp16")]; + tensor var_3392_begin_0 = const()[name = string("op_3392_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3392_end_0 = const()[name = string("op_3392_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_3392_end_mask_0 = const()[name = string("op_3392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3392_cast_fp16 = slice_by_index(begin = var_3392_begin_0, end = var_3392_end_0, end_mask = var_3392_end_mask_0, x = query_5_cast_fp16)[name = string("op_3392_cast_fp16")]; + tensor var_3401_begin_0 = const()[name = string("op_3401_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3401_end_0 = const()[name = string("op_3401_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3401_end_mask_0 = const()[name = string("op_3401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3401_cast_fp16 = slice_by_index(begin = var_3401_begin_0, end = var_3401_end_0, end_mask = var_3401_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3401_cast_fp16")]; + tensor var_3408_begin_0 = const()[name = string("op_3408_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3408_end_0 = const()[name = string("op_3408_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3408_end_mask_0 = const()[name = string("op_3408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3408_cast_fp16 = slice_by_index(begin = var_3408_begin_0, end = var_3408_end_0, end_mask = var_3408_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3408_cast_fp16")]; + tensor var_3415_begin_0 = const()[name = string("op_3415_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3415_end_0 = const()[name = string("op_3415_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3415_end_mask_0 = const()[name = string("op_3415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3415_cast_fp16 = slice_by_index(begin = var_3415_begin_0, end = var_3415_end_0, end_mask = var_3415_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3415_cast_fp16")]; + tensor var_3422_begin_0 = const()[name = string("op_3422_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3422_end_0 = const()[name = string("op_3422_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3422_end_mask_0 = const()[name = string("op_3422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3422_cast_fp16 = slice_by_index(begin = var_3422_begin_0, end = var_3422_end_0, end_mask = var_3422_end_mask_0, x = var_3316_cast_fp16)[name = string("op_3422_cast_fp16")]; + tensor var_3429_begin_0 = const()[name = string("op_3429_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3429_end_0 = const()[name = string("op_3429_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3429_end_mask_0 = const()[name = string("op_3429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3429_cast_fp16 = slice_by_index(begin = var_3429_begin_0, end = var_3429_end_0, end_mask = var_3429_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3429_cast_fp16")]; + tensor var_3436_begin_0 = const()[name = string("op_3436_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3436_end_0 = const()[name = string("op_3436_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3436_end_mask_0 = const()[name = string("op_3436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3436_cast_fp16 = slice_by_index(begin = var_3436_begin_0, end = var_3436_end_0, end_mask = var_3436_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3436_cast_fp16")]; + tensor var_3443_begin_0 = const()[name = string("op_3443_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3443_end_0 = const()[name = string("op_3443_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3443_end_mask_0 = const()[name = string("op_3443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3443_cast_fp16 = slice_by_index(begin = var_3443_begin_0, end = var_3443_end_0, end_mask = var_3443_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3443_cast_fp16")]; + tensor var_3450_begin_0 = const()[name = string("op_3450_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3450_end_0 = const()[name = string("op_3450_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3450_end_mask_0 = const()[name = string("op_3450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3450_cast_fp16 = slice_by_index(begin = var_3450_begin_0, end = var_3450_end_0, end_mask = var_3450_end_mask_0, x = var_3320_cast_fp16)[name = string("op_3450_cast_fp16")]; + tensor var_3457_begin_0 = const()[name = string("op_3457_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3457_end_0 = const()[name = string("op_3457_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3457_end_mask_0 = const()[name = string("op_3457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3457_cast_fp16 = slice_by_index(begin = var_3457_begin_0, end = var_3457_end_0, end_mask = var_3457_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3457_cast_fp16")]; + tensor var_3464_begin_0 = const()[name = string("op_3464_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3464_end_0 = const()[name = string("op_3464_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3464_end_mask_0 = const()[name = string("op_3464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3464_cast_fp16 = slice_by_index(begin = var_3464_begin_0, end = var_3464_end_0, end_mask = var_3464_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3464_cast_fp16")]; + tensor var_3471_begin_0 = const()[name = string("op_3471_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3471_end_0 = const()[name = string("op_3471_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3471_end_mask_0 = const()[name = string("op_3471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3471_cast_fp16 = slice_by_index(begin = var_3471_begin_0, end = var_3471_end_0, end_mask = var_3471_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3471_cast_fp16")]; + tensor var_3478_begin_0 = const()[name = string("op_3478_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3478_end_0 = const()[name = string("op_3478_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3478_end_mask_0 = const()[name = string("op_3478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3478_cast_fp16 = slice_by_index(begin = var_3478_begin_0, end = var_3478_end_0, end_mask = var_3478_end_mask_0, x = var_3324_cast_fp16)[name = string("op_3478_cast_fp16")]; + tensor var_3485_begin_0 = const()[name = string("op_3485_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3485_end_0 = const()[name = string("op_3485_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3485_end_mask_0 = const()[name = string("op_3485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3485_cast_fp16 = slice_by_index(begin = var_3485_begin_0, end = var_3485_end_0, end_mask = var_3485_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3485_cast_fp16")]; + tensor var_3492_begin_0 = const()[name = string("op_3492_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3492_end_0 = const()[name = string("op_3492_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3492_end_mask_0 = const()[name = string("op_3492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3492_cast_fp16 = slice_by_index(begin = var_3492_begin_0, end = var_3492_end_0, end_mask = var_3492_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3492_cast_fp16")]; + tensor var_3499_begin_0 = const()[name = string("op_3499_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3499_end_0 = const()[name = string("op_3499_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3499_end_mask_0 = const()[name = string("op_3499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3499_cast_fp16 = slice_by_index(begin = var_3499_begin_0, end = var_3499_end_0, end_mask = var_3499_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3499_cast_fp16")]; + tensor var_3506_begin_0 = const()[name = string("op_3506_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3506_end_0 = const()[name = string("op_3506_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3506_end_mask_0 = const()[name = string("op_3506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3506_cast_fp16 = slice_by_index(begin = var_3506_begin_0, end = var_3506_end_0, end_mask = var_3506_end_mask_0, x = var_3328_cast_fp16)[name = string("op_3506_cast_fp16")]; + tensor var_3513_begin_0 = const()[name = string("op_3513_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3513_end_0 = const()[name = string("op_3513_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3513_end_mask_0 = const()[name = string("op_3513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3513_cast_fp16 = slice_by_index(begin = var_3513_begin_0, end = var_3513_end_0, end_mask = var_3513_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3513_cast_fp16")]; + tensor var_3520_begin_0 = const()[name = string("op_3520_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3520_end_0 = const()[name = string("op_3520_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3520_end_mask_0 = const()[name = string("op_3520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3520_cast_fp16 = slice_by_index(begin = var_3520_begin_0, end = var_3520_end_0, end_mask = var_3520_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3520_cast_fp16")]; + tensor var_3527_begin_0 = const()[name = string("op_3527_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3527_end_0 = const()[name = string("op_3527_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3527_end_mask_0 = const()[name = string("op_3527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3527_cast_fp16 = slice_by_index(begin = var_3527_begin_0, end = var_3527_end_0, end_mask = var_3527_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3527_cast_fp16")]; + tensor var_3534_begin_0 = const()[name = string("op_3534_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3534_end_0 = const()[name = string("op_3534_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3534_end_mask_0 = const()[name = string("op_3534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3534_cast_fp16 = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = var_3332_cast_fp16)[name = string("op_3534_cast_fp16")]; + tensor var_3541_begin_0 = const()[name = string("op_3541_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3541_end_0 = const()[name = string("op_3541_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3541_end_mask_0 = const()[name = string("op_3541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3541_cast_fp16 = slice_by_index(begin = var_3541_begin_0, end = var_3541_end_0, end_mask = var_3541_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3541_cast_fp16")]; + tensor var_3548_begin_0 = const()[name = string("op_3548_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3548_end_0 = const()[name = string("op_3548_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3548_end_mask_0 = const()[name = string("op_3548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3548_cast_fp16 = slice_by_index(begin = var_3548_begin_0, end = var_3548_end_0, end_mask = var_3548_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3548_cast_fp16")]; + tensor var_3555_begin_0 = const()[name = string("op_3555_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3555_end_0 = const()[name = string("op_3555_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3555_end_mask_0 = const()[name = string("op_3555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3555_cast_fp16 = slice_by_index(begin = var_3555_begin_0, end = var_3555_end_0, end_mask = var_3555_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3555_cast_fp16")]; + tensor var_3562_begin_0 = const()[name = string("op_3562_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3562_end_0 = const()[name = string("op_3562_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3562_end_mask_0 = const()[name = string("op_3562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3562_cast_fp16 = slice_by_index(begin = var_3562_begin_0, end = var_3562_end_0, end_mask = var_3562_end_mask_0, x = var_3336_cast_fp16)[name = string("op_3562_cast_fp16")]; + tensor var_3569_begin_0 = const()[name = string("op_3569_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3569_end_0 = const()[name = string("op_3569_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3569_end_mask_0 = const()[name = string("op_3569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3569_cast_fp16 = slice_by_index(begin = var_3569_begin_0, end = var_3569_end_0, end_mask = var_3569_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3569_cast_fp16")]; + tensor var_3576_begin_0 = const()[name = string("op_3576_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3576_end_0 = const()[name = string("op_3576_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3576_end_mask_0 = const()[name = string("op_3576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3576_cast_fp16 = slice_by_index(begin = var_3576_begin_0, end = var_3576_end_0, end_mask = var_3576_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3576_cast_fp16")]; + tensor var_3583_begin_0 = const()[name = string("op_3583_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3583_end_0 = const()[name = string("op_3583_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3583_end_mask_0 = const()[name = string("op_3583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3583_cast_fp16 = slice_by_index(begin = var_3583_begin_0, end = var_3583_end_0, end_mask = var_3583_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3583_cast_fp16")]; + tensor var_3590_begin_0 = const()[name = string("op_3590_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3590_end_0 = const()[name = string("op_3590_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3590_end_mask_0 = const()[name = string("op_3590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3590_cast_fp16 = slice_by_index(begin = var_3590_begin_0, end = var_3590_end_0, end_mask = var_3590_end_mask_0, x = var_3340_cast_fp16)[name = string("op_3590_cast_fp16")]; + tensor var_3597_begin_0 = const()[name = string("op_3597_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3597_end_0 = const()[name = string("op_3597_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3597_end_mask_0 = const()[name = string("op_3597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3597_cast_fp16 = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3597_cast_fp16")]; + tensor var_3604_begin_0 = const()[name = string("op_3604_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3604_end_0 = const()[name = string("op_3604_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3604_end_mask_0 = const()[name = string("op_3604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3604_cast_fp16 = slice_by_index(begin = var_3604_begin_0, end = var_3604_end_0, end_mask = var_3604_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3604_cast_fp16")]; + tensor var_3611_begin_0 = const()[name = string("op_3611_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3611_end_0 = const()[name = string("op_3611_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3611_end_mask_0 = const()[name = string("op_3611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3611_cast_fp16 = slice_by_index(begin = var_3611_begin_0, end = var_3611_end_0, end_mask = var_3611_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3611_cast_fp16")]; + tensor var_3618_begin_0 = const()[name = string("op_3618_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3618_end_0 = const()[name = string("op_3618_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3618_end_mask_0 = const()[name = string("op_3618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3618_cast_fp16 = slice_by_index(begin = var_3618_begin_0, end = var_3618_end_0, end_mask = var_3618_end_mask_0, x = var_3344_cast_fp16)[name = string("op_3618_cast_fp16")]; + tensor var_3625_begin_0 = const()[name = string("op_3625_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3625_end_0 = const()[name = string("op_3625_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3625_end_mask_0 = const()[name = string("op_3625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3625_cast_fp16 = slice_by_index(begin = var_3625_begin_0, end = var_3625_end_0, end_mask = var_3625_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3625_cast_fp16")]; + tensor var_3632_begin_0 = const()[name = string("op_3632_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3632_end_0 = const()[name = string("op_3632_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3632_end_mask_0 = const()[name = string("op_3632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3632_cast_fp16 = slice_by_index(begin = var_3632_begin_0, end = var_3632_end_0, end_mask = var_3632_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3632_cast_fp16")]; + tensor var_3639_begin_0 = const()[name = string("op_3639_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3639_end_0 = const()[name = string("op_3639_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3639_end_mask_0 = const()[name = string("op_3639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3639_cast_fp16 = slice_by_index(begin = var_3639_begin_0, end = var_3639_end_0, end_mask = var_3639_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3639_cast_fp16")]; + tensor var_3646_begin_0 = const()[name = string("op_3646_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3646_end_0 = const()[name = string("op_3646_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3646_end_mask_0 = const()[name = string("op_3646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3646_cast_fp16 = slice_by_index(begin = var_3646_begin_0, end = var_3646_end_0, end_mask = var_3646_end_mask_0, x = var_3348_cast_fp16)[name = string("op_3646_cast_fp16")]; + tensor var_3653_begin_0 = const()[name = string("op_3653_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3653_end_0 = const()[name = string("op_3653_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3653_end_mask_0 = const()[name = string("op_3653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3653_cast_fp16 = slice_by_index(begin = var_3653_begin_0, end = var_3653_end_0, end_mask = var_3653_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3653_cast_fp16")]; + tensor var_3660_begin_0 = const()[name = string("op_3660_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3660_end_0 = const()[name = string("op_3660_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3660_end_mask_0 = const()[name = string("op_3660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3660_cast_fp16 = slice_by_index(begin = var_3660_begin_0, end = var_3660_end_0, end_mask = var_3660_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3660_cast_fp16")]; + tensor var_3667_begin_0 = const()[name = string("op_3667_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3667_end_0 = const()[name = string("op_3667_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3667_end_mask_0 = const()[name = string("op_3667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3667_cast_fp16 = slice_by_index(begin = var_3667_begin_0, end = var_3667_end_0, end_mask = var_3667_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3667_cast_fp16")]; + tensor var_3674_begin_0 = const()[name = string("op_3674_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3674_end_0 = const()[name = string("op_3674_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3674_end_mask_0 = const()[name = string("op_3674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3674_cast_fp16 = slice_by_index(begin = var_3674_begin_0, end = var_3674_end_0, end_mask = var_3674_end_mask_0, x = var_3352_cast_fp16)[name = string("op_3674_cast_fp16")]; + tensor var_3681_begin_0 = const()[name = string("op_3681_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3681_end_0 = const()[name = string("op_3681_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3681_end_mask_0 = const()[name = string("op_3681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3681_cast_fp16 = slice_by_index(begin = var_3681_begin_0, end = var_3681_end_0, end_mask = var_3681_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3681_cast_fp16")]; + tensor var_3688_begin_0 = const()[name = string("op_3688_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3688_end_0 = const()[name = string("op_3688_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3688_end_mask_0 = const()[name = string("op_3688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3688_cast_fp16 = slice_by_index(begin = var_3688_begin_0, end = var_3688_end_0, end_mask = var_3688_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3688_cast_fp16")]; + tensor var_3695_begin_0 = const()[name = string("op_3695_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3695_end_0 = const()[name = string("op_3695_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3695_end_mask_0 = const()[name = string("op_3695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3695_cast_fp16 = slice_by_index(begin = var_3695_begin_0, end = var_3695_end_0, end_mask = var_3695_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3695_cast_fp16")]; + tensor var_3702_begin_0 = const()[name = string("op_3702_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3702_end_0 = const()[name = string("op_3702_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3702_end_mask_0 = const()[name = string("op_3702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3702_cast_fp16 = slice_by_index(begin = var_3702_begin_0, end = var_3702_end_0, end_mask = var_3702_end_mask_0, x = var_3356_cast_fp16)[name = string("op_3702_cast_fp16")]; + tensor var_3709_begin_0 = const()[name = string("op_3709_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3709_end_0 = const()[name = string("op_3709_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3709_end_mask_0 = const()[name = string("op_3709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3709_cast_fp16 = slice_by_index(begin = var_3709_begin_0, end = var_3709_end_0, end_mask = var_3709_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3709_cast_fp16")]; + tensor var_3716_begin_0 = const()[name = string("op_3716_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3716_end_0 = const()[name = string("op_3716_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3716_end_mask_0 = const()[name = string("op_3716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3716_cast_fp16 = slice_by_index(begin = var_3716_begin_0, end = var_3716_end_0, end_mask = var_3716_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3716_cast_fp16")]; + tensor var_3723_begin_0 = const()[name = string("op_3723_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3723_end_0 = const()[name = string("op_3723_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3723_end_mask_0 = const()[name = string("op_3723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3723_cast_fp16 = slice_by_index(begin = var_3723_begin_0, end = var_3723_end_0, end_mask = var_3723_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3723_cast_fp16")]; + tensor var_3730_begin_0 = const()[name = string("op_3730_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3730_end_0 = const()[name = string("op_3730_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3730_end_mask_0 = const()[name = string("op_3730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3730_cast_fp16 = slice_by_index(begin = var_3730_begin_0, end = var_3730_end_0, end_mask = var_3730_end_mask_0, x = var_3360_cast_fp16)[name = string("op_3730_cast_fp16")]; + tensor var_3737_begin_0 = const()[name = string("op_3737_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3737_end_0 = const()[name = string("op_3737_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3737_end_mask_0 = const()[name = string("op_3737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3737_cast_fp16 = slice_by_index(begin = var_3737_begin_0, end = var_3737_end_0, end_mask = var_3737_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3737_cast_fp16")]; + tensor var_3744_begin_0 = const()[name = string("op_3744_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3744_end_0 = const()[name = string("op_3744_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3744_end_mask_0 = const()[name = string("op_3744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3744_cast_fp16 = slice_by_index(begin = var_3744_begin_0, end = var_3744_end_0, end_mask = var_3744_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3744_cast_fp16")]; + tensor var_3751_begin_0 = const()[name = string("op_3751_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3751_end_0 = const()[name = string("op_3751_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3751_end_mask_0 = const()[name = string("op_3751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3751_cast_fp16 = slice_by_index(begin = var_3751_begin_0, end = var_3751_end_0, end_mask = var_3751_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3751_cast_fp16")]; + tensor var_3758_begin_0 = const()[name = string("op_3758_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3758_end_0 = const()[name = string("op_3758_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3758_end_mask_0 = const()[name = string("op_3758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3758_cast_fp16 = slice_by_index(begin = var_3758_begin_0, end = var_3758_end_0, end_mask = var_3758_end_mask_0, x = var_3364_cast_fp16)[name = string("op_3758_cast_fp16")]; + tensor var_3765_begin_0 = const()[name = string("op_3765_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3765_end_0 = const()[name = string("op_3765_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3765_end_mask_0 = const()[name = string("op_3765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3765_cast_fp16 = slice_by_index(begin = var_3765_begin_0, end = var_3765_end_0, end_mask = var_3765_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3765_cast_fp16")]; + tensor var_3772_begin_0 = const()[name = string("op_3772_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3772_end_0 = const()[name = string("op_3772_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3772_end_mask_0 = const()[name = string("op_3772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3772_cast_fp16 = slice_by_index(begin = var_3772_begin_0, end = var_3772_end_0, end_mask = var_3772_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3772_cast_fp16")]; + tensor var_3779_begin_0 = const()[name = string("op_3779_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3779_end_0 = const()[name = string("op_3779_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3779_end_mask_0 = const()[name = string("op_3779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3779_cast_fp16 = slice_by_index(begin = var_3779_begin_0, end = var_3779_end_0, end_mask = var_3779_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3779_cast_fp16")]; + tensor var_3786_begin_0 = const()[name = string("op_3786_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3786_end_0 = const()[name = string("op_3786_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3786_end_mask_0 = const()[name = string("op_3786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3786_cast_fp16 = slice_by_index(begin = var_3786_begin_0, end = var_3786_end_0, end_mask = var_3786_end_mask_0, x = var_3368_cast_fp16)[name = string("op_3786_cast_fp16")]; + tensor var_3793_begin_0 = const()[name = string("op_3793_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3793_end_0 = const()[name = string("op_3793_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3793_end_mask_0 = const()[name = string("op_3793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3793_cast_fp16 = slice_by_index(begin = var_3793_begin_0, end = var_3793_end_0, end_mask = var_3793_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3793_cast_fp16")]; + tensor var_3800_begin_0 = const()[name = string("op_3800_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3800_end_0 = const()[name = string("op_3800_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3800_end_mask_0 = const()[name = string("op_3800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3800_cast_fp16 = slice_by_index(begin = var_3800_begin_0, end = var_3800_end_0, end_mask = var_3800_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3800_cast_fp16")]; + tensor var_3807_begin_0 = const()[name = string("op_3807_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3807_end_0 = const()[name = string("op_3807_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3807_end_mask_0 = const()[name = string("op_3807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3807_cast_fp16 = slice_by_index(begin = var_3807_begin_0, end = var_3807_end_0, end_mask = var_3807_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3807_cast_fp16")]; + tensor var_3814_begin_0 = const()[name = string("op_3814_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3814_end_0 = const()[name = string("op_3814_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3814_end_mask_0 = const()[name = string("op_3814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3814_cast_fp16 = slice_by_index(begin = var_3814_begin_0, end = var_3814_end_0, end_mask = var_3814_end_mask_0, x = var_3372_cast_fp16)[name = string("op_3814_cast_fp16")]; + tensor var_3821_begin_0 = const()[name = string("op_3821_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3821_end_0 = const()[name = string("op_3821_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3821_end_mask_0 = const()[name = string("op_3821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3821_cast_fp16 = slice_by_index(begin = var_3821_begin_0, end = var_3821_end_0, end_mask = var_3821_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3821_cast_fp16")]; + tensor var_3828_begin_0 = const()[name = string("op_3828_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3828_end_0 = const()[name = string("op_3828_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3828_end_mask_0 = const()[name = string("op_3828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3828_cast_fp16 = slice_by_index(begin = var_3828_begin_0, end = var_3828_end_0, end_mask = var_3828_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3828_cast_fp16")]; + tensor var_3835_begin_0 = const()[name = string("op_3835_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3835_end_0 = const()[name = string("op_3835_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3835_end_mask_0 = const()[name = string("op_3835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3835_cast_fp16 = slice_by_index(begin = var_3835_begin_0, end = var_3835_end_0, end_mask = var_3835_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3835_cast_fp16")]; + tensor var_3842_begin_0 = const()[name = string("op_3842_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3842_end_0 = const()[name = string("op_3842_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3842_end_mask_0 = const()[name = string("op_3842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3842_cast_fp16 = slice_by_index(begin = var_3842_begin_0, end = var_3842_end_0, end_mask = var_3842_end_mask_0, x = var_3376_cast_fp16)[name = string("op_3842_cast_fp16")]; + tensor var_3849_begin_0 = const()[name = string("op_3849_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3849_end_0 = const()[name = string("op_3849_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3849_end_mask_0 = const()[name = string("op_3849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3849_cast_fp16 = slice_by_index(begin = var_3849_begin_0, end = var_3849_end_0, end_mask = var_3849_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3849_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = string("op_3856_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3856_end_0 = const()[name = string("op_3856_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3856_end_mask_0 = const()[name = string("op_3856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3856_cast_fp16")]; + tensor var_3863_begin_0 = const()[name = string("op_3863_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3863_end_0 = const()[name = string("op_3863_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3863_end_mask_0 = const()[name = string("op_3863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3863_cast_fp16 = slice_by_index(begin = var_3863_begin_0, end = var_3863_end_0, end_mask = var_3863_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3863_cast_fp16")]; + tensor var_3870_begin_0 = const()[name = string("op_3870_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3870_end_0 = const()[name = string("op_3870_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3870_end_mask_0 = const()[name = string("op_3870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3870_cast_fp16 = slice_by_index(begin = var_3870_begin_0, end = var_3870_end_0, end_mask = var_3870_end_mask_0, x = var_3380_cast_fp16)[name = string("op_3870_cast_fp16")]; + tensor var_3877_begin_0 = const()[name = string("op_3877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3877_end_0 = const()[name = string("op_3877_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3877_end_mask_0 = const()[name = string("op_3877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3877_cast_fp16 = slice_by_index(begin = var_3877_begin_0, end = var_3877_end_0, end_mask = var_3877_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3877_cast_fp16")]; + tensor var_3884_begin_0 = const()[name = string("op_3884_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3884_end_0 = const()[name = string("op_3884_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3884_end_mask_0 = const()[name = string("op_3884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3884_cast_fp16 = slice_by_index(begin = var_3884_begin_0, end = var_3884_end_0, end_mask = var_3884_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3884_cast_fp16")]; + tensor var_3891_begin_0 = const()[name = string("op_3891_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3891_end_0 = const()[name = string("op_3891_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3891_end_mask_0 = const()[name = string("op_3891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3891_cast_fp16 = slice_by_index(begin = var_3891_begin_0, end = var_3891_end_0, end_mask = var_3891_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3891_cast_fp16")]; + tensor var_3898_begin_0 = const()[name = string("op_3898_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3898_end_0 = const()[name = string("op_3898_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3898_end_mask_0 = const()[name = string("op_3898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3898_cast_fp16 = slice_by_index(begin = var_3898_begin_0, end = var_3898_end_0, end_mask = var_3898_end_mask_0, x = var_3384_cast_fp16)[name = string("op_3898_cast_fp16")]; + tensor var_3905_begin_0 = const()[name = string("op_3905_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3905_end_0 = const()[name = string("op_3905_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3905_end_mask_0 = const()[name = string("op_3905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3905_cast_fp16 = slice_by_index(begin = var_3905_begin_0, end = var_3905_end_0, end_mask = var_3905_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3905_cast_fp16")]; + tensor var_3912_begin_0 = const()[name = string("op_3912_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3912_end_0 = const()[name = string("op_3912_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3912_end_mask_0 = const()[name = string("op_3912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3912_cast_fp16 = slice_by_index(begin = var_3912_begin_0, end = var_3912_end_0, end_mask = var_3912_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3912_cast_fp16")]; + tensor var_3919_begin_0 = const()[name = string("op_3919_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3919_end_0 = const()[name = string("op_3919_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3919_end_mask_0 = const()[name = string("op_3919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3919_cast_fp16 = slice_by_index(begin = var_3919_begin_0, end = var_3919_end_0, end_mask = var_3919_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3919_cast_fp16")]; + tensor var_3926_begin_0 = const()[name = string("op_3926_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3926_end_0 = const()[name = string("op_3926_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3926_end_mask_0 = const()[name = string("op_3926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3926_cast_fp16 = slice_by_index(begin = var_3926_begin_0, end = var_3926_end_0, end_mask = var_3926_end_mask_0, x = var_3388_cast_fp16)[name = string("op_3926_cast_fp16")]; + tensor var_3933_begin_0 = const()[name = string("op_3933_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3933_end_0 = const()[name = string("op_3933_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3933_end_mask_0 = const()[name = string("op_3933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3933_cast_fp16 = slice_by_index(begin = var_3933_begin_0, end = var_3933_end_0, end_mask = var_3933_end_mask_0, x = var_3392_cast_fp16)[name = string("op_3933_cast_fp16")]; + tensor var_3940_begin_0 = const()[name = string("op_3940_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3940_end_0 = const()[name = string("op_3940_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3940_end_mask_0 = const()[name = string("op_3940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3940_cast_fp16 = slice_by_index(begin = var_3940_begin_0, end = var_3940_end_0, end_mask = var_3940_end_mask_0, x = var_3392_cast_fp16)[name = string("op_3940_cast_fp16")]; + tensor var_3947_begin_0 = const()[name = string("op_3947_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3947_end_0 = const()[name = string("op_3947_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3947_end_mask_0 = const()[name = string("op_3947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3947_cast_fp16 = slice_by_index(begin = var_3947_begin_0, end = var_3947_end_0, end_mask = var_3947_end_mask_0, x = var_3392_cast_fp16)[name = string("op_3947_cast_fp16")]; + tensor var_3954_begin_0 = const()[name = string("op_3954_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3954_end_0 = const()[name = string("op_3954_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3954_end_mask_0 = const()[name = string("op_3954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3954_cast_fp16 = slice_by_index(begin = var_3954_begin_0, end = var_3954_end_0, end_mask = var_3954_end_mask_0, x = var_3392_cast_fp16)[name = string("op_3954_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3959_begin_0 = const()[name = string("op_3959_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3959_end_0 = const()[name = string("op_3959_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3959_end_mask_0 = const()[name = string("op_3959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_29")]; + tensor var_3959_cast_fp16 = slice_by_index(begin = var_3959_begin_0, end = var_3959_end_0, end_mask = var_3959_end_mask_0, x = k_5_cast_fp16)[name = string("op_3959_cast_fp16")]; + tensor var_3963_begin_0 = const()[name = string("op_3963_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3963_end_0 = const()[name = string("op_3963_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3963_end_mask_0 = const()[name = string("op_3963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3963_cast_fp16 = slice_by_index(begin = var_3963_begin_0, end = var_3963_end_0, end_mask = var_3963_end_mask_0, x = k_5_cast_fp16)[name = string("op_3963_cast_fp16")]; + tensor var_3967_begin_0 = const()[name = string("op_3967_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3967_end_0 = const()[name = string("op_3967_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3967_end_mask_0 = const()[name = string("op_3967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3967_cast_fp16 = slice_by_index(begin = var_3967_begin_0, end = var_3967_end_0, end_mask = var_3967_end_mask_0, x = k_5_cast_fp16)[name = string("op_3967_cast_fp16")]; + tensor var_3971_begin_0 = const()[name = string("op_3971_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3971_end_0 = const()[name = string("op_3971_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3971_end_mask_0 = const()[name = string("op_3971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3971_cast_fp16 = slice_by_index(begin = var_3971_begin_0, end = var_3971_end_0, end_mask = var_3971_end_mask_0, x = k_5_cast_fp16)[name = string("op_3971_cast_fp16")]; + tensor var_3975_begin_0 = const()[name = string("op_3975_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3975_end_0 = const()[name = string("op_3975_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3975_end_mask_0 = const()[name = string("op_3975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3975_cast_fp16 = slice_by_index(begin = var_3975_begin_0, end = var_3975_end_0, end_mask = var_3975_end_mask_0, x = k_5_cast_fp16)[name = string("op_3975_cast_fp16")]; + tensor var_3979_begin_0 = const()[name = string("op_3979_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3979_end_0 = const()[name = string("op_3979_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3979_end_mask_0 = const()[name = string("op_3979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3979_cast_fp16 = slice_by_index(begin = var_3979_begin_0, end = var_3979_end_0, end_mask = var_3979_end_mask_0, x = k_5_cast_fp16)[name = string("op_3979_cast_fp16")]; + tensor var_3983_begin_0 = const()[name = string("op_3983_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3983_end_0 = const()[name = string("op_3983_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3983_end_mask_0 = const()[name = string("op_3983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3983_cast_fp16 = slice_by_index(begin = var_3983_begin_0, end = var_3983_end_0, end_mask = var_3983_end_mask_0, x = k_5_cast_fp16)[name = string("op_3983_cast_fp16")]; + tensor var_3987_begin_0 = const()[name = string("op_3987_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3987_end_0 = const()[name = string("op_3987_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3987_end_mask_0 = const()[name = string("op_3987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3987_cast_fp16 = slice_by_index(begin = var_3987_begin_0, end = var_3987_end_0, end_mask = var_3987_end_mask_0, x = k_5_cast_fp16)[name = string("op_3987_cast_fp16")]; + tensor var_3991_begin_0 = const()[name = string("op_3991_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3991_end_0 = const()[name = string("op_3991_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_3991_end_mask_0 = const()[name = string("op_3991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3991_cast_fp16 = slice_by_index(begin = var_3991_begin_0, end = var_3991_end_0, end_mask = var_3991_end_mask_0, x = k_5_cast_fp16)[name = string("op_3991_cast_fp16")]; + tensor var_3995_begin_0 = const()[name = string("op_3995_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3995_end_0 = const()[name = string("op_3995_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_3995_end_mask_0 = const()[name = string("op_3995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3995_cast_fp16 = slice_by_index(begin = var_3995_begin_0, end = var_3995_end_0, end_mask = var_3995_end_mask_0, x = k_5_cast_fp16)[name = string("op_3995_cast_fp16")]; + tensor var_3999_begin_0 = const()[name = string("op_3999_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3999_end_0 = const()[name = string("op_3999_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_3999_end_mask_0 = const()[name = string("op_3999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3999_cast_fp16 = slice_by_index(begin = var_3999_begin_0, end = var_3999_end_0, end_mask = var_3999_end_mask_0, x = k_5_cast_fp16)[name = string("op_3999_cast_fp16")]; + tensor var_4003_begin_0 = const()[name = string("op_4003_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4003_end_0 = const()[name = string("op_4003_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_4003_end_mask_0 = const()[name = string("op_4003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4003_cast_fp16 = slice_by_index(begin = var_4003_begin_0, end = var_4003_end_0, end_mask = var_4003_end_mask_0, x = k_5_cast_fp16)[name = string("op_4003_cast_fp16")]; + tensor var_4007_begin_0 = const()[name = string("op_4007_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4007_end_0 = const()[name = string("op_4007_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_4007_end_mask_0 = const()[name = string("op_4007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4007_cast_fp16 = slice_by_index(begin = var_4007_begin_0, end = var_4007_end_0, end_mask = var_4007_end_mask_0, x = k_5_cast_fp16)[name = string("op_4007_cast_fp16")]; + tensor var_4011_begin_0 = const()[name = string("op_4011_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4011_end_0 = const()[name = string("op_4011_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_4011_end_mask_0 = const()[name = string("op_4011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4011_cast_fp16 = slice_by_index(begin = var_4011_begin_0, end = var_4011_end_0, end_mask = var_4011_end_mask_0, x = k_5_cast_fp16)[name = string("op_4011_cast_fp16")]; + tensor var_4015_begin_0 = const()[name = string("op_4015_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4015_end_0 = const()[name = string("op_4015_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_4015_end_mask_0 = const()[name = string("op_4015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4015_cast_fp16 = slice_by_index(begin = var_4015_begin_0, end = var_4015_end_0, end_mask = var_4015_end_mask_0, x = k_5_cast_fp16)[name = string("op_4015_cast_fp16")]; + tensor var_4019_begin_0 = const()[name = string("op_4019_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4019_end_0 = const()[name = string("op_4019_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_4019_end_mask_0 = const()[name = string("op_4019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4019_cast_fp16 = slice_by_index(begin = var_4019_begin_0, end = var_4019_end_0, end_mask = var_4019_end_mask_0, x = k_5_cast_fp16)[name = string("op_4019_cast_fp16")]; + tensor var_4023_begin_0 = const()[name = string("op_4023_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4023_end_0 = const()[name = string("op_4023_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_4023_end_mask_0 = const()[name = string("op_4023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4023_cast_fp16 = slice_by_index(begin = var_4023_begin_0, end = var_4023_end_0, end_mask = var_4023_end_mask_0, x = k_5_cast_fp16)[name = string("op_4023_cast_fp16")]; + tensor var_4027_begin_0 = const()[name = string("op_4027_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4027_end_0 = const()[name = string("op_4027_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_4027_end_mask_0 = const()[name = string("op_4027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4027_cast_fp16 = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = k_5_cast_fp16)[name = string("op_4027_cast_fp16")]; + tensor var_4031_begin_0 = const()[name = string("op_4031_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4031_end_0 = const()[name = string("op_4031_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_4031_end_mask_0 = const()[name = string("op_4031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4031_cast_fp16 = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = k_5_cast_fp16)[name = string("op_4031_cast_fp16")]; + tensor var_4035_begin_0 = const()[name = string("op_4035_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4035_end_0 = const()[name = string("op_4035_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_4035_end_mask_0 = const()[name = string("op_4035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4035_cast_fp16 = slice_by_index(begin = var_4035_begin_0, end = var_4035_end_0, end_mask = var_4035_end_mask_0, x = k_5_cast_fp16)[name = string("op_4035_cast_fp16")]; + tensor var_4037_begin_0 = const()[name = string("op_4037_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4037_end_0 = const()[name = string("op_4037_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4037_end_mask_0 = const()[name = string("op_4037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4037_cast_fp16 = slice_by_index(begin = var_4037_begin_0, end = var_4037_end_0, end_mask = var_4037_end_mask_0, x = value_5_cast_fp16)[name = string("op_4037_cast_fp16")]; + tensor var_4041_begin_0 = const()[name = string("op_4041_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4041_end_0 = const()[name = string("op_4041_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4041_end_mask_0 = const()[name = string("op_4041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4041_cast_fp16 = slice_by_index(begin = var_4041_begin_0, end = var_4041_end_0, end_mask = var_4041_end_mask_0, x = value_5_cast_fp16)[name = string("op_4041_cast_fp16")]; + tensor var_4045_begin_0 = const()[name = string("op_4045_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4045_end_0 = const()[name = string("op_4045_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4045_end_mask_0 = const()[name = string("op_4045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4045_cast_fp16 = slice_by_index(begin = var_4045_begin_0, end = var_4045_end_0, end_mask = var_4045_end_mask_0, x = value_5_cast_fp16)[name = string("op_4045_cast_fp16")]; + tensor var_4049_begin_0 = const()[name = string("op_4049_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4049_end_0 = const()[name = string("op_4049_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4049_end_mask_0 = const()[name = string("op_4049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4049_cast_fp16 = slice_by_index(begin = var_4049_begin_0, end = var_4049_end_0, end_mask = var_4049_end_mask_0, x = value_5_cast_fp16)[name = string("op_4049_cast_fp16")]; + tensor var_4053_begin_0 = const()[name = string("op_4053_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4053_end_0 = const()[name = string("op_4053_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4053_end_mask_0 = const()[name = string("op_4053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4053_cast_fp16 = slice_by_index(begin = var_4053_begin_0, end = var_4053_end_0, end_mask = var_4053_end_mask_0, x = value_5_cast_fp16)[name = string("op_4053_cast_fp16")]; + tensor var_4057_begin_0 = const()[name = string("op_4057_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4057_end_0 = const()[name = string("op_4057_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4057_end_mask_0 = const()[name = string("op_4057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4057_cast_fp16 = slice_by_index(begin = var_4057_begin_0, end = var_4057_end_0, end_mask = var_4057_end_mask_0, x = value_5_cast_fp16)[name = string("op_4057_cast_fp16")]; + tensor var_4061_begin_0 = const()[name = string("op_4061_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4061_end_0 = const()[name = string("op_4061_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4061_end_mask_0 = const()[name = string("op_4061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4061_cast_fp16 = slice_by_index(begin = var_4061_begin_0, end = var_4061_end_0, end_mask = var_4061_end_mask_0, x = value_5_cast_fp16)[name = string("op_4061_cast_fp16")]; + tensor var_4065_begin_0 = const()[name = string("op_4065_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4065_end_0 = const()[name = string("op_4065_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4065_end_mask_0 = const()[name = string("op_4065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4065_cast_fp16 = slice_by_index(begin = var_4065_begin_0, end = var_4065_end_0, end_mask = var_4065_end_mask_0, x = value_5_cast_fp16)[name = string("op_4065_cast_fp16")]; + tensor var_4069_begin_0 = const()[name = string("op_4069_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4069_end_0 = const()[name = string("op_4069_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4069_end_mask_0 = const()[name = string("op_4069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4069_cast_fp16 = slice_by_index(begin = var_4069_begin_0, end = var_4069_end_0, end_mask = var_4069_end_mask_0, x = value_5_cast_fp16)[name = string("op_4069_cast_fp16")]; + tensor var_4073_begin_0 = const()[name = string("op_4073_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4073_end_0 = const()[name = string("op_4073_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4073_end_mask_0 = const()[name = string("op_4073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4073_cast_fp16 = slice_by_index(begin = var_4073_begin_0, end = var_4073_end_0, end_mask = var_4073_end_mask_0, x = value_5_cast_fp16)[name = string("op_4073_cast_fp16")]; + tensor var_4077_begin_0 = const()[name = string("op_4077_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4077_end_0 = const()[name = string("op_4077_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4077_end_mask_0 = const()[name = string("op_4077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4077_cast_fp16 = slice_by_index(begin = var_4077_begin_0, end = var_4077_end_0, end_mask = var_4077_end_mask_0, x = value_5_cast_fp16)[name = string("op_4077_cast_fp16")]; + tensor var_4081_begin_0 = const()[name = string("op_4081_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4081_end_0 = const()[name = string("op_4081_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4081_end_mask_0 = const()[name = string("op_4081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4081_cast_fp16 = slice_by_index(begin = var_4081_begin_0, end = var_4081_end_0, end_mask = var_4081_end_mask_0, x = value_5_cast_fp16)[name = string("op_4081_cast_fp16")]; + tensor var_4085_begin_0 = const()[name = string("op_4085_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4085_end_0 = const()[name = string("op_4085_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_4085_end_mask_0 = const()[name = string("op_4085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4085_cast_fp16 = slice_by_index(begin = var_4085_begin_0, end = var_4085_end_0, end_mask = var_4085_end_mask_0, x = value_5_cast_fp16)[name = string("op_4085_cast_fp16")]; + tensor var_4089_begin_0 = const()[name = string("op_4089_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4089_end_0 = const()[name = string("op_4089_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_4089_end_mask_0 = const()[name = string("op_4089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4089_cast_fp16 = slice_by_index(begin = var_4089_begin_0, end = var_4089_end_0, end_mask = var_4089_end_mask_0, x = value_5_cast_fp16)[name = string("op_4089_cast_fp16")]; + tensor var_4093_begin_0 = const()[name = string("op_4093_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4093_end_0 = const()[name = string("op_4093_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_4093_end_mask_0 = const()[name = string("op_4093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4093_cast_fp16 = slice_by_index(begin = var_4093_begin_0, end = var_4093_end_0, end_mask = var_4093_end_mask_0, x = value_5_cast_fp16)[name = string("op_4093_cast_fp16")]; + tensor var_4097_begin_0 = const()[name = string("op_4097_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4097_end_0 = const()[name = string("op_4097_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_4097_end_mask_0 = const()[name = string("op_4097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4097_cast_fp16 = slice_by_index(begin = var_4097_begin_0, end = var_4097_end_0, end_mask = var_4097_end_mask_0, x = value_5_cast_fp16)[name = string("op_4097_cast_fp16")]; + tensor var_4101_begin_0 = const()[name = string("op_4101_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4101_end_0 = const()[name = string("op_4101_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_4101_end_mask_0 = const()[name = string("op_4101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4101_cast_fp16 = slice_by_index(begin = var_4101_begin_0, end = var_4101_end_0, end_mask = var_4101_end_mask_0, x = value_5_cast_fp16)[name = string("op_4101_cast_fp16")]; + tensor var_4105_begin_0 = const()[name = string("op_4105_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4105_end_0 = const()[name = string("op_4105_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_4105_end_mask_0 = const()[name = string("op_4105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4105_cast_fp16 = slice_by_index(begin = var_4105_begin_0, end = var_4105_end_0, end_mask = var_4105_end_mask_0, x = value_5_cast_fp16)[name = string("op_4105_cast_fp16")]; + tensor var_4109_begin_0 = const()[name = string("op_4109_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4109_end_0 = const()[name = string("op_4109_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_4109_end_mask_0 = const()[name = string("op_4109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4109_cast_fp16 = slice_by_index(begin = var_4109_begin_0, end = var_4109_end_0, end_mask = var_4109_end_mask_0, x = value_5_cast_fp16)[name = string("op_4109_cast_fp16")]; + tensor var_4113_begin_0 = const()[name = string("op_4113_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4113_end_0 = const()[name = string("op_4113_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_4113_end_mask_0 = const()[name = string("op_4113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4113_cast_fp16 = slice_by_index(begin = var_4113_begin_0, end = var_4113_end_0, end_mask = var_4113_end_mask_0, x = value_5_cast_fp16)[name = string("op_4113_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3959_cast_fp16, var_3401_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3959_cast_fp16, var_3408_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3959_cast_fp16, var_3415_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3959_cast_fp16, var_3422_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3963_cast_fp16, var_3429_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3963_cast_fp16, var_3436_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3963_cast_fp16, var_3443_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3963_cast_fp16, var_3450_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3967_cast_fp16, var_3457_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3967_cast_fp16, var_3464_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3967_cast_fp16, var_3471_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3967_cast_fp16, var_3478_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3971_cast_fp16, var_3485_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3971_cast_fp16, var_3492_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3971_cast_fp16, var_3499_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3971_cast_fp16, var_3506_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3975_cast_fp16, var_3513_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3975_cast_fp16, var_3520_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3975_cast_fp16, var_3527_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3975_cast_fp16, var_3534_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3979_cast_fp16, var_3541_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3979_cast_fp16, var_3548_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3979_cast_fp16, var_3555_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3979_cast_fp16, var_3562_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3983_cast_fp16, var_3569_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3983_cast_fp16, var_3576_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3983_cast_fp16, var_3583_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3983_cast_fp16, var_3590_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3987_cast_fp16, var_3597_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3987_cast_fp16, var_3604_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3987_cast_fp16, var_3611_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_383_equation_0, values = (var_3987_cast_fp16, var_3618_cast_fp16))[name = string("_SplitHeadsQ__mh_w_383_cast_fp16")]; + string _SplitHeadsQ__mh_w_385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_385_equation_0, values = (var_3991_cast_fp16, var_3625_cast_fp16))[name = string("_SplitHeadsQ__mh_w_385_cast_fp16")]; + string _SplitHeadsQ__mh_w_387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_387_equation_0, values = (var_3991_cast_fp16, var_3632_cast_fp16))[name = string("_SplitHeadsQ__mh_w_387_cast_fp16")]; + string _SplitHeadsQ__mh_w_389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_389_equation_0, values = (var_3991_cast_fp16, var_3639_cast_fp16))[name = string("_SplitHeadsQ__mh_w_389_cast_fp16")]; + string _SplitHeadsQ__mh_w_391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_391_equation_0, values = (var_3991_cast_fp16, var_3646_cast_fp16))[name = string("_SplitHeadsQ__mh_w_391_cast_fp16")]; + string _SplitHeadsQ__mh_w_393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_393_equation_0, values = (var_3995_cast_fp16, var_3653_cast_fp16))[name = string("_SplitHeadsQ__mh_w_393_cast_fp16")]; + string _SplitHeadsQ__mh_w_395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_395_equation_0, values = (var_3995_cast_fp16, var_3660_cast_fp16))[name = string("_SplitHeadsQ__mh_w_395_cast_fp16")]; + string _SplitHeadsQ__mh_w_397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_397_equation_0, values = (var_3995_cast_fp16, var_3667_cast_fp16))[name = string("_SplitHeadsQ__mh_w_397_cast_fp16")]; + string _SplitHeadsQ__mh_w_399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_399_equation_0, values = (var_3995_cast_fp16, var_3674_cast_fp16))[name = string("_SplitHeadsQ__mh_w_399_cast_fp16")]; + string _SplitHeadsQ__mh_w_401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_401_equation_0, values = (var_3999_cast_fp16, var_3681_cast_fp16))[name = string("_SplitHeadsQ__mh_w_401_cast_fp16")]; + string _SplitHeadsQ__mh_w_403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_403_equation_0, values = (var_3999_cast_fp16, var_3688_cast_fp16))[name = string("_SplitHeadsQ__mh_w_403_cast_fp16")]; + string _SplitHeadsQ__mh_w_405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_405_equation_0, values = (var_3999_cast_fp16, var_3695_cast_fp16))[name = string("_SplitHeadsQ__mh_w_405_cast_fp16")]; + string _SplitHeadsQ__mh_w_407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_407_equation_0, values = (var_3999_cast_fp16, var_3702_cast_fp16))[name = string("_SplitHeadsQ__mh_w_407_cast_fp16")]; + string _SplitHeadsQ__mh_w_409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_409_equation_0, values = (var_4003_cast_fp16, var_3709_cast_fp16))[name = string("_SplitHeadsQ__mh_w_409_cast_fp16")]; + string _SplitHeadsQ__mh_w_411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_411_equation_0, values = (var_4003_cast_fp16, var_3716_cast_fp16))[name = string("_SplitHeadsQ__mh_w_411_cast_fp16")]; + string _SplitHeadsQ__mh_w_413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_413_equation_0, values = (var_4003_cast_fp16, var_3723_cast_fp16))[name = string("_SplitHeadsQ__mh_w_413_cast_fp16")]; + string _SplitHeadsQ__mh_w_415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_415_equation_0, values = (var_4003_cast_fp16, var_3730_cast_fp16))[name = string("_SplitHeadsQ__mh_w_415_cast_fp16")]; + string _SplitHeadsQ__mh_w_417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_417_equation_0, values = (var_4007_cast_fp16, var_3737_cast_fp16))[name = string("_SplitHeadsQ__mh_w_417_cast_fp16")]; + string _SplitHeadsQ__mh_w_419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_419_equation_0, values = (var_4007_cast_fp16, var_3744_cast_fp16))[name = string("_SplitHeadsQ__mh_w_419_cast_fp16")]; + string _SplitHeadsQ__mh_w_421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_421_equation_0, values = (var_4007_cast_fp16, var_3751_cast_fp16))[name = string("_SplitHeadsQ__mh_w_421_cast_fp16")]; + string _SplitHeadsQ__mh_w_423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_423_equation_0, values = (var_4007_cast_fp16, var_3758_cast_fp16))[name = string("_SplitHeadsQ__mh_w_423_cast_fp16")]; + string _SplitHeadsQ__mh_w_425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_425_equation_0, values = (var_4011_cast_fp16, var_3765_cast_fp16))[name = string("_SplitHeadsQ__mh_w_425_cast_fp16")]; + string _SplitHeadsQ__mh_w_427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_427_equation_0, values = (var_4011_cast_fp16, var_3772_cast_fp16))[name = string("_SplitHeadsQ__mh_w_427_cast_fp16")]; + string _SplitHeadsQ__mh_w_429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_429_equation_0, values = (var_4011_cast_fp16, var_3779_cast_fp16))[name = string("_SplitHeadsQ__mh_w_429_cast_fp16")]; + string _SplitHeadsQ__mh_w_431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_431_equation_0, values = (var_4011_cast_fp16, var_3786_cast_fp16))[name = string("_SplitHeadsQ__mh_w_431_cast_fp16")]; + string _SplitHeadsQ__mh_w_433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_433_equation_0, values = (var_4015_cast_fp16, var_3793_cast_fp16))[name = string("_SplitHeadsQ__mh_w_433_cast_fp16")]; + string _SplitHeadsQ__mh_w_435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_435_equation_0, values = (var_4015_cast_fp16, var_3800_cast_fp16))[name = string("_SplitHeadsQ__mh_w_435_cast_fp16")]; + string _SplitHeadsQ__mh_w_437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_437_equation_0, values = (var_4015_cast_fp16, var_3807_cast_fp16))[name = string("_SplitHeadsQ__mh_w_437_cast_fp16")]; + string _SplitHeadsQ__mh_w_439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_439_equation_0, values = (var_4015_cast_fp16, var_3814_cast_fp16))[name = string("_SplitHeadsQ__mh_w_439_cast_fp16")]; + string _SplitHeadsQ__mh_w_441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_441_equation_0, values = (var_4019_cast_fp16, var_3821_cast_fp16))[name = string("_SplitHeadsQ__mh_w_441_cast_fp16")]; + string _SplitHeadsQ__mh_w_443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_443_equation_0, values = (var_4019_cast_fp16, var_3828_cast_fp16))[name = string("_SplitHeadsQ__mh_w_443_cast_fp16")]; + string _SplitHeadsQ__mh_w_445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_445_equation_0, values = (var_4019_cast_fp16, var_3835_cast_fp16))[name = string("_SplitHeadsQ__mh_w_445_cast_fp16")]; + string _SplitHeadsQ__mh_w_447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_447_equation_0, values = (var_4019_cast_fp16, var_3842_cast_fp16))[name = string("_SplitHeadsQ__mh_w_447_cast_fp16")]; + string _SplitHeadsQ__mh_w_449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_449_equation_0, values = (var_4023_cast_fp16, var_3849_cast_fp16))[name = string("_SplitHeadsQ__mh_w_449_cast_fp16")]; + string _SplitHeadsQ__mh_w_451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_451_equation_0, values = (var_4023_cast_fp16, var_3856_cast_fp16))[name = string("_SplitHeadsQ__mh_w_451_cast_fp16")]; + string _SplitHeadsQ__mh_w_453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_453_equation_0, values = (var_4023_cast_fp16, var_3863_cast_fp16))[name = string("_SplitHeadsQ__mh_w_453_cast_fp16")]; + string _SplitHeadsQ__mh_w_455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_455_equation_0, values = (var_4023_cast_fp16, var_3870_cast_fp16))[name = string("_SplitHeadsQ__mh_w_455_cast_fp16")]; + string _SplitHeadsQ__mh_w_457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_457_equation_0, values = (var_4027_cast_fp16, var_3877_cast_fp16))[name = string("_SplitHeadsQ__mh_w_457_cast_fp16")]; + string _SplitHeadsQ__mh_w_459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_459_equation_0, values = (var_4027_cast_fp16, var_3884_cast_fp16))[name = string("_SplitHeadsQ__mh_w_459_cast_fp16")]; + string _SplitHeadsQ__mh_w_461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_461_equation_0, values = (var_4027_cast_fp16, var_3891_cast_fp16))[name = string("_SplitHeadsQ__mh_w_461_cast_fp16")]; + string _SplitHeadsQ__mh_w_463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_463_equation_0, values = (var_4027_cast_fp16, var_3898_cast_fp16))[name = string("_SplitHeadsQ__mh_w_463_cast_fp16")]; + string _SplitHeadsQ__mh_w_465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_465_equation_0, values = (var_4031_cast_fp16, var_3905_cast_fp16))[name = string("_SplitHeadsQ__mh_w_465_cast_fp16")]; + string _SplitHeadsQ__mh_w_467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_467_equation_0, values = (var_4031_cast_fp16, var_3912_cast_fp16))[name = string("_SplitHeadsQ__mh_w_467_cast_fp16")]; + string _SplitHeadsQ__mh_w_469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_469_equation_0, values = (var_4031_cast_fp16, var_3919_cast_fp16))[name = string("_SplitHeadsQ__mh_w_469_cast_fp16")]; + string _SplitHeadsQ__mh_w_471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_471_equation_0, values = (var_4031_cast_fp16, var_3926_cast_fp16))[name = string("_SplitHeadsQ__mh_w_471_cast_fp16")]; + string _SplitHeadsQ__mh_w_473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_473_equation_0, values = (var_4035_cast_fp16, var_3933_cast_fp16))[name = string("_SplitHeadsQ__mh_w_473_cast_fp16")]; + string _SplitHeadsQ__mh_w_475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_475_equation_0, values = (var_4035_cast_fp16, var_3940_cast_fp16))[name = string("_SplitHeadsQ__mh_w_475_cast_fp16")]; + string _SplitHeadsQ__mh_w_477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_477_equation_0, values = (var_4035_cast_fp16, var_3947_cast_fp16))[name = string("_SplitHeadsQ__mh_w_477_cast_fp16")]; + string _SplitHeadsQ__mh_w_479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_479_equation_0, values = (var_4035_cast_fp16, var_3954_cast_fp16))[name = string("_SplitHeadsQ__mh_w_479_cast_fp16")]; + fp16 var_4276_to_fp16 = const()[name = string("op_4276_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_4276_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_4278_to_fp16 = const()[name = string("op_4278_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_4278_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_4280_to_fp16 = const()[name = string("op_4280_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_4280_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_4282_to_fp16 = const()[name = string("op_4282_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_4282_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_4284_to_fp16 = const()[name = string("op_4284_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_4284_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_4286_to_fp16 = const()[name = string("op_4286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_4286_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_4288_to_fp16 = const()[name = string("op_4288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_4288_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_4290_to_fp16 = const()[name = string("op_4290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_4290_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_4292_to_fp16 = const()[name = string("op_4292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_4292_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_4294_to_fp16 = const()[name = string("op_4294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_4294_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_4296_to_fp16 = const()[name = string("op_4296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_4296_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_4298_to_fp16 = const()[name = string("op_4298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_4298_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_4300_to_fp16 = const()[name = string("op_4300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_4300_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_4302_to_fp16 = const()[name = string("op_4302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_4302_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_4304_to_fp16 = const()[name = string("op_4304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_4304_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_4306_to_fp16 = const()[name = string("op_4306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_4306_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_4308_to_fp16 = const()[name = string("op_4308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_4308_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_4310_to_fp16 = const()[name = string("op_4310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_4310_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_4312_to_fp16 = const()[name = string("op_4312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_4312_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_4314_to_fp16 = const()[name = string("op_4314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_4314_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_4316_to_fp16 = const()[name = string("op_4316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_4316_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_4318_to_fp16 = const()[name = string("op_4318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_4318_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_4320_to_fp16 = const()[name = string("op_4320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_4320_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_4322_to_fp16 = const()[name = string("op_4322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_4322_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_4324_to_fp16 = const()[name = string("op_4324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_4324_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_4326_to_fp16 = const()[name = string("op_4326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_4326_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_4328_to_fp16 = const()[name = string("op_4328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_4328_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_4330_to_fp16 = const()[name = string("op_4330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_4330_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_4332_to_fp16 = const()[name = string("op_4332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_4332_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_4334_to_fp16 = const()[name = string("op_4334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_4334_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_4336_to_fp16 = const()[name = string("op_4336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_4336_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_4338_to_fp16 = const()[name = string("op_4338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_383_cast_fp16, y = var_4338_to_fp16)[name = string("aw_chunk_383_cast_fp16")]; + fp16 var_4340_to_fp16 = const()[name = string("op_4340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_385_cast_fp16, y = var_4340_to_fp16)[name = string("aw_chunk_385_cast_fp16")]; + fp16 var_4342_to_fp16 = const()[name = string("op_4342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_387_cast_fp16, y = var_4342_to_fp16)[name = string("aw_chunk_387_cast_fp16")]; + fp16 var_4344_to_fp16 = const()[name = string("op_4344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_389_cast_fp16, y = var_4344_to_fp16)[name = string("aw_chunk_389_cast_fp16")]; + fp16 var_4346_to_fp16 = const()[name = string("op_4346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_391_cast_fp16, y = var_4346_to_fp16)[name = string("aw_chunk_391_cast_fp16")]; + fp16 var_4348_to_fp16 = const()[name = string("op_4348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_393_cast_fp16, y = var_4348_to_fp16)[name = string("aw_chunk_393_cast_fp16")]; + fp16 var_4350_to_fp16 = const()[name = string("op_4350_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_395_cast_fp16, y = var_4350_to_fp16)[name = string("aw_chunk_395_cast_fp16")]; + fp16 var_4352_to_fp16 = const()[name = string("op_4352_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_397_cast_fp16, y = var_4352_to_fp16)[name = string("aw_chunk_397_cast_fp16")]; + fp16 var_4354_to_fp16 = const()[name = string("op_4354_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_399_cast_fp16, y = var_4354_to_fp16)[name = string("aw_chunk_399_cast_fp16")]; + fp16 var_4356_to_fp16 = const()[name = string("op_4356_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_401_cast_fp16, y = var_4356_to_fp16)[name = string("aw_chunk_401_cast_fp16")]; + fp16 var_4358_to_fp16 = const()[name = string("op_4358_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_403_cast_fp16, y = var_4358_to_fp16)[name = string("aw_chunk_403_cast_fp16")]; + fp16 var_4360_to_fp16 = const()[name = string("op_4360_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_405_cast_fp16, y = var_4360_to_fp16)[name = string("aw_chunk_405_cast_fp16")]; + fp16 var_4362_to_fp16 = const()[name = string("op_4362_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_407_cast_fp16, y = var_4362_to_fp16)[name = string("aw_chunk_407_cast_fp16")]; + fp16 var_4364_to_fp16 = const()[name = string("op_4364_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_409_cast_fp16, y = var_4364_to_fp16)[name = string("aw_chunk_409_cast_fp16")]; + fp16 var_4366_to_fp16 = const()[name = string("op_4366_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_411_cast_fp16, y = var_4366_to_fp16)[name = string("aw_chunk_411_cast_fp16")]; + fp16 var_4368_to_fp16 = const()[name = string("op_4368_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_413_cast_fp16, y = var_4368_to_fp16)[name = string("aw_chunk_413_cast_fp16")]; + fp16 var_4370_to_fp16 = const()[name = string("op_4370_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_415_cast_fp16, y = var_4370_to_fp16)[name = string("aw_chunk_415_cast_fp16")]; + fp16 var_4372_to_fp16 = const()[name = string("op_4372_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_417_cast_fp16, y = var_4372_to_fp16)[name = string("aw_chunk_417_cast_fp16")]; + fp16 var_4374_to_fp16 = const()[name = string("op_4374_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_419_cast_fp16, y = var_4374_to_fp16)[name = string("aw_chunk_419_cast_fp16")]; + fp16 var_4376_to_fp16 = const()[name = string("op_4376_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_421_cast_fp16, y = var_4376_to_fp16)[name = string("aw_chunk_421_cast_fp16")]; + fp16 var_4378_to_fp16 = const()[name = string("op_4378_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_423_cast_fp16, y = var_4378_to_fp16)[name = string("aw_chunk_423_cast_fp16")]; + fp16 var_4380_to_fp16 = const()[name = string("op_4380_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_425_cast_fp16, y = var_4380_to_fp16)[name = string("aw_chunk_425_cast_fp16")]; + fp16 var_4382_to_fp16 = const()[name = string("op_4382_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_427_cast_fp16, y = var_4382_to_fp16)[name = string("aw_chunk_427_cast_fp16")]; + fp16 var_4384_to_fp16 = const()[name = string("op_4384_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_429_cast_fp16, y = var_4384_to_fp16)[name = string("aw_chunk_429_cast_fp16")]; + fp16 var_4386_to_fp16 = const()[name = string("op_4386_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_431_cast_fp16, y = var_4386_to_fp16)[name = string("aw_chunk_431_cast_fp16")]; + fp16 var_4388_to_fp16 = const()[name = string("op_4388_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_433_cast_fp16, y = var_4388_to_fp16)[name = string("aw_chunk_433_cast_fp16")]; + fp16 var_4390_to_fp16 = const()[name = string("op_4390_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_435_cast_fp16, y = var_4390_to_fp16)[name = string("aw_chunk_435_cast_fp16")]; + fp16 var_4392_to_fp16 = const()[name = string("op_4392_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_437_cast_fp16, y = var_4392_to_fp16)[name = string("aw_chunk_437_cast_fp16")]; + fp16 var_4394_to_fp16 = const()[name = string("op_4394_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_439_cast_fp16, y = var_4394_to_fp16)[name = string("aw_chunk_439_cast_fp16")]; + fp16 var_4396_to_fp16 = const()[name = string("op_4396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_441_cast_fp16, y = var_4396_to_fp16)[name = string("aw_chunk_441_cast_fp16")]; + fp16 var_4398_to_fp16 = const()[name = string("op_4398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_443_cast_fp16, y = var_4398_to_fp16)[name = string("aw_chunk_443_cast_fp16")]; + fp16 var_4400_to_fp16 = const()[name = string("op_4400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_445_cast_fp16, y = var_4400_to_fp16)[name = string("aw_chunk_445_cast_fp16")]; + fp16 var_4402_to_fp16 = const()[name = string("op_4402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_447_cast_fp16, y = var_4402_to_fp16)[name = string("aw_chunk_447_cast_fp16")]; + fp16 var_4404_to_fp16 = const()[name = string("op_4404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_449_cast_fp16, y = var_4404_to_fp16)[name = string("aw_chunk_449_cast_fp16")]; + fp16 var_4406_to_fp16 = const()[name = string("op_4406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_451_cast_fp16, y = var_4406_to_fp16)[name = string("aw_chunk_451_cast_fp16")]; + fp16 var_4408_to_fp16 = const()[name = string("op_4408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_453_cast_fp16, y = var_4408_to_fp16)[name = string("aw_chunk_453_cast_fp16")]; + fp16 var_4410_to_fp16 = const()[name = string("op_4410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_455_cast_fp16, y = var_4410_to_fp16)[name = string("aw_chunk_455_cast_fp16")]; + fp16 var_4412_to_fp16 = const()[name = string("op_4412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_457_cast_fp16, y = var_4412_to_fp16)[name = string("aw_chunk_457_cast_fp16")]; + fp16 var_4414_to_fp16 = const()[name = string("op_4414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_459_cast_fp16, y = var_4414_to_fp16)[name = string("aw_chunk_459_cast_fp16")]; + fp16 var_4416_to_fp16 = const()[name = string("op_4416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_461_cast_fp16, y = var_4416_to_fp16)[name = string("aw_chunk_461_cast_fp16")]; + fp16 var_4418_to_fp16 = const()[name = string("op_4418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_463_cast_fp16, y = var_4418_to_fp16)[name = string("aw_chunk_463_cast_fp16")]; + fp16 var_4420_to_fp16 = const()[name = string("op_4420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_465_cast_fp16, y = var_4420_to_fp16)[name = string("aw_chunk_465_cast_fp16")]; + fp16 var_4422_to_fp16 = const()[name = string("op_4422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_467_cast_fp16, y = var_4422_to_fp16)[name = string("aw_chunk_467_cast_fp16")]; + fp16 var_4424_to_fp16 = const()[name = string("op_4424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_469_cast_fp16, y = var_4424_to_fp16)[name = string("aw_chunk_469_cast_fp16")]; + fp16 var_4426_to_fp16 = const()[name = string("op_4426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_471_cast_fp16, y = var_4426_to_fp16)[name = string("aw_chunk_471_cast_fp16")]; + fp16 var_4428_to_fp16 = const()[name = string("op_4428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_473_cast_fp16, y = var_4428_to_fp16)[name = string("aw_chunk_473_cast_fp16")]; + fp16 var_4430_to_fp16 = const()[name = string("op_4430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_475_cast_fp16, y = var_4430_to_fp16)[name = string("aw_chunk_475_cast_fp16")]; + fp16 var_4432_to_fp16 = const()[name = string("op_4432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_477_cast_fp16, y = var_4432_to_fp16)[name = string("aw_chunk_477_cast_fp16")]; + fp16 var_4434_to_fp16 = const()[name = string("op_4434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_479_cast_fp16, y = var_4434_to_fp16)[name = string("aw_chunk_479_cast_fp16")]; + tensor var_4436_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_321_cast_fp16)[name = string("op_4436_cast_fp16")]; + tensor var_4437_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_323_cast_fp16)[name = string("op_4437_cast_fp16")]; + tensor var_4438_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_325_cast_fp16)[name = string("op_4438_cast_fp16")]; + tensor var_4439_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_327_cast_fp16)[name = string("op_4439_cast_fp16")]; + tensor var_4440_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_329_cast_fp16)[name = string("op_4440_cast_fp16")]; + tensor var_4441_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_331_cast_fp16)[name = string("op_4441_cast_fp16")]; + tensor var_4442_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_333_cast_fp16)[name = string("op_4442_cast_fp16")]; + tensor var_4443_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_335_cast_fp16)[name = string("op_4443_cast_fp16")]; + tensor var_4444_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_337_cast_fp16)[name = string("op_4444_cast_fp16")]; + tensor var_4445_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_339_cast_fp16)[name = string("op_4445_cast_fp16")]; + tensor var_4446_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_341_cast_fp16)[name = string("op_4446_cast_fp16")]; + tensor var_4447_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_343_cast_fp16)[name = string("op_4447_cast_fp16")]; + tensor var_4448_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_345_cast_fp16)[name = string("op_4448_cast_fp16")]; + tensor var_4449_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_347_cast_fp16)[name = string("op_4449_cast_fp16")]; + tensor var_4450_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_349_cast_fp16)[name = string("op_4450_cast_fp16")]; + tensor var_4451_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_351_cast_fp16)[name = string("op_4451_cast_fp16")]; + tensor var_4452_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_353_cast_fp16)[name = string("op_4452_cast_fp16")]; + tensor var_4453_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_355_cast_fp16)[name = string("op_4453_cast_fp16")]; + tensor var_4454_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_357_cast_fp16)[name = string("op_4454_cast_fp16")]; + tensor var_4455_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_359_cast_fp16)[name = string("op_4455_cast_fp16")]; + tensor var_4456_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_361_cast_fp16)[name = string("op_4456_cast_fp16")]; + tensor var_4457_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_363_cast_fp16)[name = string("op_4457_cast_fp16")]; + tensor var_4458_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_365_cast_fp16)[name = string("op_4458_cast_fp16")]; + tensor var_4459_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_367_cast_fp16)[name = string("op_4459_cast_fp16")]; + tensor var_4460_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_369_cast_fp16)[name = string("op_4460_cast_fp16")]; + tensor var_4461_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_371_cast_fp16)[name = string("op_4461_cast_fp16")]; + tensor var_4462_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_373_cast_fp16)[name = string("op_4462_cast_fp16")]; + tensor var_4463_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_375_cast_fp16)[name = string("op_4463_cast_fp16")]; + tensor var_4464_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_377_cast_fp16)[name = string("op_4464_cast_fp16")]; + tensor var_4465_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_379_cast_fp16)[name = string("op_4465_cast_fp16")]; + tensor var_4466_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_381_cast_fp16)[name = string("op_4466_cast_fp16")]; + tensor var_4467_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_383_cast_fp16)[name = string("op_4467_cast_fp16")]; + tensor var_4468_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_385_cast_fp16)[name = string("op_4468_cast_fp16")]; + tensor var_4469_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_387_cast_fp16)[name = string("op_4469_cast_fp16")]; + tensor var_4470_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_389_cast_fp16)[name = string("op_4470_cast_fp16")]; + tensor var_4471_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_391_cast_fp16)[name = string("op_4471_cast_fp16")]; + tensor var_4472_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_393_cast_fp16)[name = string("op_4472_cast_fp16")]; + tensor var_4473_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_395_cast_fp16)[name = string("op_4473_cast_fp16")]; + tensor var_4474_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_397_cast_fp16)[name = string("op_4474_cast_fp16")]; + tensor var_4475_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_399_cast_fp16)[name = string("op_4475_cast_fp16")]; + tensor var_4476_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_401_cast_fp16)[name = string("op_4476_cast_fp16")]; + tensor var_4477_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_403_cast_fp16)[name = string("op_4477_cast_fp16")]; + tensor var_4478_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_405_cast_fp16)[name = string("op_4478_cast_fp16")]; + tensor var_4479_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_407_cast_fp16)[name = string("op_4479_cast_fp16")]; + tensor var_4480_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_409_cast_fp16)[name = string("op_4480_cast_fp16")]; + tensor var_4481_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_411_cast_fp16)[name = string("op_4481_cast_fp16")]; + tensor var_4482_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_413_cast_fp16)[name = string("op_4482_cast_fp16")]; + tensor var_4483_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_415_cast_fp16)[name = string("op_4483_cast_fp16")]; + tensor var_4484_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_417_cast_fp16)[name = string("op_4484_cast_fp16")]; + tensor var_4485_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_419_cast_fp16)[name = string("op_4485_cast_fp16")]; + tensor var_4486_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_421_cast_fp16)[name = string("op_4486_cast_fp16")]; + tensor var_4487_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_423_cast_fp16)[name = string("op_4487_cast_fp16")]; + tensor var_4488_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_425_cast_fp16)[name = string("op_4488_cast_fp16")]; + tensor var_4489_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_427_cast_fp16)[name = string("op_4489_cast_fp16")]; + tensor var_4490_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_429_cast_fp16)[name = string("op_4490_cast_fp16")]; + tensor var_4491_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_431_cast_fp16)[name = string("op_4491_cast_fp16")]; + tensor var_4492_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_433_cast_fp16)[name = string("op_4492_cast_fp16")]; + tensor var_4493_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_435_cast_fp16)[name = string("op_4493_cast_fp16")]; + tensor var_4494_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_437_cast_fp16)[name = string("op_4494_cast_fp16")]; + tensor var_4495_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_439_cast_fp16)[name = string("op_4495_cast_fp16")]; + tensor var_4496_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_441_cast_fp16)[name = string("op_4496_cast_fp16")]; + tensor var_4497_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_443_cast_fp16)[name = string("op_4497_cast_fp16")]; + tensor var_4498_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_445_cast_fp16)[name = string("op_4498_cast_fp16")]; + tensor var_4499_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_447_cast_fp16)[name = string("op_4499_cast_fp16")]; + tensor var_4500_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_449_cast_fp16)[name = string("op_4500_cast_fp16")]; + tensor var_4501_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_451_cast_fp16)[name = string("op_4501_cast_fp16")]; + tensor var_4502_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_453_cast_fp16)[name = string("op_4502_cast_fp16")]; + tensor var_4503_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_455_cast_fp16)[name = string("op_4503_cast_fp16")]; + tensor var_4504_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_457_cast_fp16)[name = string("op_4504_cast_fp16")]; + tensor var_4505_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_459_cast_fp16)[name = string("op_4505_cast_fp16")]; + tensor var_4506_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_461_cast_fp16)[name = string("op_4506_cast_fp16")]; + tensor var_4507_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_463_cast_fp16)[name = string("op_4507_cast_fp16")]; + tensor var_4508_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_465_cast_fp16)[name = string("op_4508_cast_fp16")]; + tensor var_4509_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_467_cast_fp16)[name = string("op_4509_cast_fp16")]; + tensor var_4510_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_469_cast_fp16)[name = string("op_4510_cast_fp16")]; + tensor var_4511_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_471_cast_fp16)[name = string("op_4511_cast_fp16")]; + tensor var_4512_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_473_cast_fp16)[name = string("op_4512_cast_fp16")]; + tensor var_4513_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_475_cast_fp16)[name = string("op_4513_cast_fp16")]; + tensor var_4514_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_477_cast_fp16)[name = string("op_4514_cast_fp16")]; + tensor var_4515_cast_fp16 = softmax(axis = var_3261, x = aw_chunk_479_cast_fp16)[name = string("op_4515_cast_fp16")]; + string var_4517_equation_0 = const()[name = string("op_4517_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4517_cast_fp16 = einsum(equation = var_4517_equation_0, values = (var_4037_cast_fp16, var_4436_cast_fp16))[name = string("op_4517_cast_fp16")]; + string var_4519_equation_0 = const()[name = string("op_4519_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4519_cast_fp16 = einsum(equation = var_4519_equation_0, values = (var_4037_cast_fp16, var_4437_cast_fp16))[name = string("op_4519_cast_fp16")]; + string var_4521_equation_0 = const()[name = string("op_4521_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4521_cast_fp16 = einsum(equation = var_4521_equation_0, values = (var_4037_cast_fp16, var_4438_cast_fp16))[name = string("op_4521_cast_fp16")]; + string var_4523_equation_0 = const()[name = string("op_4523_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4523_cast_fp16 = einsum(equation = var_4523_equation_0, values = (var_4037_cast_fp16, var_4439_cast_fp16))[name = string("op_4523_cast_fp16")]; + string var_4525_equation_0 = const()[name = string("op_4525_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4525_cast_fp16 = einsum(equation = var_4525_equation_0, values = (var_4041_cast_fp16, var_4440_cast_fp16))[name = string("op_4525_cast_fp16")]; + string var_4527_equation_0 = const()[name = string("op_4527_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4527_cast_fp16 = einsum(equation = var_4527_equation_0, values = (var_4041_cast_fp16, var_4441_cast_fp16))[name = string("op_4527_cast_fp16")]; + string var_4529_equation_0 = const()[name = string("op_4529_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4529_cast_fp16 = einsum(equation = var_4529_equation_0, values = (var_4041_cast_fp16, var_4442_cast_fp16))[name = string("op_4529_cast_fp16")]; + string var_4531_equation_0 = const()[name = string("op_4531_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4531_cast_fp16 = einsum(equation = var_4531_equation_0, values = (var_4041_cast_fp16, var_4443_cast_fp16))[name = string("op_4531_cast_fp16")]; + string var_4533_equation_0 = const()[name = string("op_4533_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4533_cast_fp16 = einsum(equation = var_4533_equation_0, values = (var_4045_cast_fp16, var_4444_cast_fp16))[name = string("op_4533_cast_fp16")]; + string var_4535_equation_0 = const()[name = string("op_4535_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4535_cast_fp16 = einsum(equation = var_4535_equation_0, values = (var_4045_cast_fp16, var_4445_cast_fp16))[name = string("op_4535_cast_fp16")]; + string var_4537_equation_0 = const()[name = string("op_4537_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4537_cast_fp16 = einsum(equation = var_4537_equation_0, values = (var_4045_cast_fp16, var_4446_cast_fp16))[name = string("op_4537_cast_fp16")]; + string var_4539_equation_0 = const()[name = string("op_4539_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4539_cast_fp16 = einsum(equation = var_4539_equation_0, values = (var_4045_cast_fp16, var_4447_cast_fp16))[name = string("op_4539_cast_fp16")]; + string var_4541_equation_0 = const()[name = string("op_4541_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4541_cast_fp16 = einsum(equation = var_4541_equation_0, values = (var_4049_cast_fp16, var_4448_cast_fp16))[name = string("op_4541_cast_fp16")]; + string var_4543_equation_0 = const()[name = string("op_4543_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4543_cast_fp16 = einsum(equation = var_4543_equation_0, values = (var_4049_cast_fp16, var_4449_cast_fp16))[name = string("op_4543_cast_fp16")]; + string var_4545_equation_0 = const()[name = string("op_4545_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4545_cast_fp16 = einsum(equation = var_4545_equation_0, values = (var_4049_cast_fp16, var_4450_cast_fp16))[name = string("op_4545_cast_fp16")]; + string var_4547_equation_0 = const()[name = string("op_4547_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4547_cast_fp16 = einsum(equation = var_4547_equation_0, values = (var_4049_cast_fp16, var_4451_cast_fp16))[name = string("op_4547_cast_fp16")]; + string var_4549_equation_0 = const()[name = string("op_4549_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4549_cast_fp16 = einsum(equation = var_4549_equation_0, values = (var_4053_cast_fp16, var_4452_cast_fp16))[name = string("op_4549_cast_fp16")]; + string var_4551_equation_0 = const()[name = string("op_4551_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4551_cast_fp16 = einsum(equation = var_4551_equation_0, values = (var_4053_cast_fp16, var_4453_cast_fp16))[name = string("op_4551_cast_fp16")]; + string var_4553_equation_0 = const()[name = string("op_4553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4553_cast_fp16 = einsum(equation = var_4553_equation_0, values = (var_4053_cast_fp16, var_4454_cast_fp16))[name = string("op_4553_cast_fp16")]; + string var_4555_equation_0 = const()[name = string("op_4555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4555_cast_fp16 = einsum(equation = var_4555_equation_0, values = (var_4053_cast_fp16, var_4455_cast_fp16))[name = string("op_4555_cast_fp16")]; + string var_4557_equation_0 = const()[name = string("op_4557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4557_cast_fp16 = einsum(equation = var_4557_equation_0, values = (var_4057_cast_fp16, var_4456_cast_fp16))[name = string("op_4557_cast_fp16")]; + string var_4559_equation_0 = const()[name = string("op_4559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4559_cast_fp16 = einsum(equation = var_4559_equation_0, values = (var_4057_cast_fp16, var_4457_cast_fp16))[name = string("op_4559_cast_fp16")]; + string var_4561_equation_0 = const()[name = string("op_4561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4561_cast_fp16 = einsum(equation = var_4561_equation_0, values = (var_4057_cast_fp16, var_4458_cast_fp16))[name = string("op_4561_cast_fp16")]; + string var_4563_equation_0 = const()[name = string("op_4563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4563_cast_fp16 = einsum(equation = var_4563_equation_0, values = (var_4057_cast_fp16, var_4459_cast_fp16))[name = string("op_4563_cast_fp16")]; + string var_4565_equation_0 = const()[name = string("op_4565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4565_cast_fp16 = einsum(equation = var_4565_equation_0, values = (var_4061_cast_fp16, var_4460_cast_fp16))[name = string("op_4565_cast_fp16")]; + string var_4567_equation_0 = const()[name = string("op_4567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4567_cast_fp16 = einsum(equation = var_4567_equation_0, values = (var_4061_cast_fp16, var_4461_cast_fp16))[name = string("op_4567_cast_fp16")]; + string var_4569_equation_0 = const()[name = string("op_4569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4569_cast_fp16 = einsum(equation = var_4569_equation_0, values = (var_4061_cast_fp16, var_4462_cast_fp16))[name = string("op_4569_cast_fp16")]; + string var_4571_equation_0 = const()[name = string("op_4571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4571_cast_fp16 = einsum(equation = var_4571_equation_0, values = (var_4061_cast_fp16, var_4463_cast_fp16))[name = string("op_4571_cast_fp16")]; + string var_4573_equation_0 = const()[name = string("op_4573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4573_cast_fp16 = einsum(equation = var_4573_equation_0, values = (var_4065_cast_fp16, var_4464_cast_fp16))[name = string("op_4573_cast_fp16")]; + string var_4575_equation_0 = const()[name = string("op_4575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4575_cast_fp16 = einsum(equation = var_4575_equation_0, values = (var_4065_cast_fp16, var_4465_cast_fp16))[name = string("op_4575_cast_fp16")]; + string var_4577_equation_0 = const()[name = string("op_4577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4577_cast_fp16 = einsum(equation = var_4577_equation_0, values = (var_4065_cast_fp16, var_4466_cast_fp16))[name = string("op_4577_cast_fp16")]; + string var_4579_equation_0 = const()[name = string("op_4579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4579_cast_fp16 = einsum(equation = var_4579_equation_0, values = (var_4065_cast_fp16, var_4467_cast_fp16))[name = string("op_4579_cast_fp16")]; + string var_4581_equation_0 = const()[name = string("op_4581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4581_cast_fp16 = einsum(equation = var_4581_equation_0, values = (var_4069_cast_fp16, var_4468_cast_fp16))[name = string("op_4581_cast_fp16")]; + string var_4583_equation_0 = const()[name = string("op_4583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4583_cast_fp16 = einsum(equation = var_4583_equation_0, values = (var_4069_cast_fp16, var_4469_cast_fp16))[name = string("op_4583_cast_fp16")]; + string var_4585_equation_0 = const()[name = string("op_4585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4585_cast_fp16 = einsum(equation = var_4585_equation_0, values = (var_4069_cast_fp16, var_4470_cast_fp16))[name = string("op_4585_cast_fp16")]; + string var_4587_equation_0 = const()[name = string("op_4587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4587_cast_fp16 = einsum(equation = var_4587_equation_0, values = (var_4069_cast_fp16, var_4471_cast_fp16))[name = string("op_4587_cast_fp16")]; + string var_4589_equation_0 = const()[name = string("op_4589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4589_cast_fp16 = einsum(equation = var_4589_equation_0, values = (var_4073_cast_fp16, var_4472_cast_fp16))[name = string("op_4589_cast_fp16")]; + string var_4591_equation_0 = const()[name = string("op_4591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4591_cast_fp16 = einsum(equation = var_4591_equation_0, values = (var_4073_cast_fp16, var_4473_cast_fp16))[name = string("op_4591_cast_fp16")]; + string var_4593_equation_0 = const()[name = string("op_4593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4593_cast_fp16 = einsum(equation = var_4593_equation_0, values = (var_4073_cast_fp16, var_4474_cast_fp16))[name = string("op_4593_cast_fp16")]; + string var_4595_equation_0 = const()[name = string("op_4595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4595_cast_fp16 = einsum(equation = var_4595_equation_0, values = (var_4073_cast_fp16, var_4475_cast_fp16))[name = string("op_4595_cast_fp16")]; + string var_4597_equation_0 = const()[name = string("op_4597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4597_cast_fp16 = einsum(equation = var_4597_equation_0, values = (var_4077_cast_fp16, var_4476_cast_fp16))[name = string("op_4597_cast_fp16")]; + string var_4599_equation_0 = const()[name = string("op_4599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4599_cast_fp16 = einsum(equation = var_4599_equation_0, values = (var_4077_cast_fp16, var_4477_cast_fp16))[name = string("op_4599_cast_fp16")]; + string var_4601_equation_0 = const()[name = string("op_4601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4601_cast_fp16 = einsum(equation = var_4601_equation_0, values = (var_4077_cast_fp16, var_4478_cast_fp16))[name = string("op_4601_cast_fp16")]; + string var_4603_equation_0 = const()[name = string("op_4603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4603_cast_fp16 = einsum(equation = var_4603_equation_0, values = (var_4077_cast_fp16, var_4479_cast_fp16))[name = string("op_4603_cast_fp16")]; + string var_4605_equation_0 = const()[name = string("op_4605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4605_cast_fp16 = einsum(equation = var_4605_equation_0, values = (var_4081_cast_fp16, var_4480_cast_fp16))[name = string("op_4605_cast_fp16")]; + string var_4607_equation_0 = const()[name = string("op_4607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4607_cast_fp16 = einsum(equation = var_4607_equation_0, values = (var_4081_cast_fp16, var_4481_cast_fp16))[name = string("op_4607_cast_fp16")]; + string var_4609_equation_0 = const()[name = string("op_4609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4609_cast_fp16 = einsum(equation = var_4609_equation_0, values = (var_4081_cast_fp16, var_4482_cast_fp16))[name = string("op_4609_cast_fp16")]; + string var_4611_equation_0 = const()[name = string("op_4611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4611_cast_fp16 = einsum(equation = var_4611_equation_0, values = (var_4081_cast_fp16, var_4483_cast_fp16))[name = string("op_4611_cast_fp16")]; + string var_4613_equation_0 = const()[name = string("op_4613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4613_cast_fp16 = einsum(equation = var_4613_equation_0, values = (var_4085_cast_fp16, var_4484_cast_fp16))[name = string("op_4613_cast_fp16")]; + string var_4615_equation_0 = const()[name = string("op_4615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4615_cast_fp16 = einsum(equation = var_4615_equation_0, values = (var_4085_cast_fp16, var_4485_cast_fp16))[name = string("op_4615_cast_fp16")]; + string var_4617_equation_0 = const()[name = string("op_4617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4617_cast_fp16 = einsum(equation = var_4617_equation_0, values = (var_4085_cast_fp16, var_4486_cast_fp16))[name = string("op_4617_cast_fp16")]; + string var_4619_equation_0 = const()[name = string("op_4619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4619_cast_fp16 = einsum(equation = var_4619_equation_0, values = (var_4085_cast_fp16, var_4487_cast_fp16))[name = string("op_4619_cast_fp16")]; + string var_4621_equation_0 = const()[name = string("op_4621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4621_cast_fp16 = einsum(equation = var_4621_equation_0, values = (var_4089_cast_fp16, var_4488_cast_fp16))[name = string("op_4621_cast_fp16")]; + string var_4623_equation_0 = const()[name = string("op_4623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4623_cast_fp16 = einsum(equation = var_4623_equation_0, values = (var_4089_cast_fp16, var_4489_cast_fp16))[name = string("op_4623_cast_fp16")]; + string var_4625_equation_0 = const()[name = string("op_4625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4625_cast_fp16 = einsum(equation = var_4625_equation_0, values = (var_4089_cast_fp16, var_4490_cast_fp16))[name = string("op_4625_cast_fp16")]; + string var_4627_equation_0 = const()[name = string("op_4627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4627_cast_fp16 = einsum(equation = var_4627_equation_0, values = (var_4089_cast_fp16, var_4491_cast_fp16))[name = string("op_4627_cast_fp16")]; + string var_4629_equation_0 = const()[name = string("op_4629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4629_cast_fp16 = einsum(equation = var_4629_equation_0, values = (var_4093_cast_fp16, var_4492_cast_fp16))[name = string("op_4629_cast_fp16")]; + string var_4631_equation_0 = const()[name = string("op_4631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4631_cast_fp16 = einsum(equation = var_4631_equation_0, values = (var_4093_cast_fp16, var_4493_cast_fp16))[name = string("op_4631_cast_fp16")]; + string var_4633_equation_0 = const()[name = string("op_4633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4633_cast_fp16 = einsum(equation = var_4633_equation_0, values = (var_4093_cast_fp16, var_4494_cast_fp16))[name = string("op_4633_cast_fp16")]; + string var_4635_equation_0 = const()[name = string("op_4635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4635_cast_fp16 = einsum(equation = var_4635_equation_0, values = (var_4093_cast_fp16, var_4495_cast_fp16))[name = string("op_4635_cast_fp16")]; + string var_4637_equation_0 = const()[name = string("op_4637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4637_cast_fp16 = einsum(equation = var_4637_equation_0, values = (var_4097_cast_fp16, var_4496_cast_fp16))[name = string("op_4637_cast_fp16")]; + string var_4639_equation_0 = const()[name = string("op_4639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4639_cast_fp16 = einsum(equation = var_4639_equation_0, values = (var_4097_cast_fp16, var_4497_cast_fp16))[name = string("op_4639_cast_fp16")]; + string var_4641_equation_0 = const()[name = string("op_4641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4641_cast_fp16 = einsum(equation = var_4641_equation_0, values = (var_4097_cast_fp16, var_4498_cast_fp16))[name = string("op_4641_cast_fp16")]; + string var_4643_equation_0 = const()[name = string("op_4643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4643_cast_fp16 = einsum(equation = var_4643_equation_0, values = (var_4097_cast_fp16, var_4499_cast_fp16))[name = string("op_4643_cast_fp16")]; + string var_4645_equation_0 = const()[name = string("op_4645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4645_cast_fp16 = einsum(equation = var_4645_equation_0, values = (var_4101_cast_fp16, var_4500_cast_fp16))[name = string("op_4645_cast_fp16")]; + string var_4647_equation_0 = const()[name = string("op_4647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4647_cast_fp16 = einsum(equation = var_4647_equation_0, values = (var_4101_cast_fp16, var_4501_cast_fp16))[name = string("op_4647_cast_fp16")]; + string var_4649_equation_0 = const()[name = string("op_4649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4649_cast_fp16 = einsum(equation = var_4649_equation_0, values = (var_4101_cast_fp16, var_4502_cast_fp16))[name = string("op_4649_cast_fp16")]; + string var_4651_equation_0 = const()[name = string("op_4651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4651_cast_fp16 = einsum(equation = var_4651_equation_0, values = (var_4101_cast_fp16, var_4503_cast_fp16))[name = string("op_4651_cast_fp16")]; + string var_4653_equation_0 = const()[name = string("op_4653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4653_cast_fp16 = einsum(equation = var_4653_equation_0, values = (var_4105_cast_fp16, var_4504_cast_fp16))[name = string("op_4653_cast_fp16")]; + string var_4655_equation_0 = const()[name = string("op_4655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4655_cast_fp16 = einsum(equation = var_4655_equation_0, values = (var_4105_cast_fp16, var_4505_cast_fp16))[name = string("op_4655_cast_fp16")]; + string var_4657_equation_0 = const()[name = string("op_4657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4657_cast_fp16 = einsum(equation = var_4657_equation_0, values = (var_4105_cast_fp16, var_4506_cast_fp16))[name = string("op_4657_cast_fp16")]; + string var_4659_equation_0 = const()[name = string("op_4659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4659_cast_fp16 = einsum(equation = var_4659_equation_0, values = (var_4105_cast_fp16, var_4507_cast_fp16))[name = string("op_4659_cast_fp16")]; + string var_4661_equation_0 = const()[name = string("op_4661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4661_cast_fp16 = einsum(equation = var_4661_equation_0, values = (var_4109_cast_fp16, var_4508_cast_fp16))[name = string("op_4661_cast_fp16")]; + string var_4663_equation_0 = const()[name = string("op_4663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4663_cast_fp16 = einsum(equation = var_4663_equation_0, values = (var_4109_cast_fp16, var_4509_cast_fp16))[name = string("op_4663_cast_fp16")]; + string var_4665_equation_0 = const()[name = string("op_4665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4665_cast_fp16 = einsum(equation = var_4665_equation_0, values = (var_4109_cast_fp16, var_4510_cast_fp16))[name = string("op_4665_cast_fp16")]; + string var_4667_equation_0 = const()[name = string("op_4667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4667_cast_fp16 = einsum(equation = var_4667_equation_0, values = (var_4109_cast_fp16, var_4511_cast_fp16))[name = string("op_4667_cast_fp16")]; + string var_4669_equation_0 = const()[name = string("op_4669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4669_cast_fp16 = einsum(equation = var_4669_equation_0, values = (var_4113_cast_fp16, var_4512_cast_fp16))[name = string("op_4669_cast_fp16")]; + string var_4671_equation_0 = const()[name = string("op_4671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4671_cast_fp16 = einsum(equation = var_4671_equation_0, values = (var_4113_cast_fp16, var_4513_cast_fp16))[name = string("op_4671_cast_fp16")]; + string var_4673_equation_0 = const()[name = string("op_4673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4673_cast_fp16 = einsum(equation = var_4673_equation_0, values = (var_4113_cast_fp16, var_4514_cast_fp16))[name = string("op_4673_cast_fp16")]; + string var_4675_equation_0 = const()[name = string("op_4675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4675_cast_fp16 = einsum(equation = var_4675_equation_0, values = (var_4113_cast_fp16, var_4515_cast_fp16))[name = string("op_4675_cast_fp16")]; + bool var_4677_interleave_0 = const()[name = string("op_4677_interleave_0"), val = bool(false)]; + tensor var_4677_cast_fp16 = concat(axis = var_3236, interleave = var_4677_interleave_0, values = (var_4517_cast_fp16, var_4519_cast_fp16, var_4521_cast_fp16, var_4523_cast_fp16))[name = string("op_4677_cast_fp16")]; + bool var_4679_interleave_0 = const()[name = string("op_4679_interleave_0"), val = bool(false)]; + tensor var_4679_cast_fp16 = concat(axis = var_3236, interleave = var_4679_interleave_0, values = (var_4525_cast_fp16, var_4527_cast_fp16, var_4529_cast_fp16, var_4531_cast_fp16))[name = string("op_4679_cast_fp16")]; + bool var_4681_interleave_0 = const()[name = string("op_4681_interleave_0"), val = bool(false)]; + tensor var_4681_cast_fp16 = concat(axis = var_3236, interleave = var_4681_interleave_0, values = (var_4533_cast_fp16, var_4535_cast_fp16, var_4537_cast_fp16, var_4539_cast_fp16))[name = string("op_4681_cast_fp16")]; + bool var_4683_interleave_0 = const()[name = string("op_4683_interleave_0"), val = bool(false)]; + tensor var_4683_cast_fp16 = concat(axis = var_3236, interleave = var_4683_interleave_0, values = (var_4541_cast_fp16, var_4543_cast_fp16, var_4545_cast_fp16, var_4547_cast_fp16))[name = string("op_4683_cast_fp16")]; + bool var_4685_interleave_0 = const()[name = string("op_4685_interleave_0"), val = bool(false)]; + tensor var_4685_cast_fp16 = concat(axis = var_3236, interleave = var_4685_interleave_0, values = (var_4549_cast_fp16, var_4551_cast_fp16, var_4553_cast_fp16, var_4555_cast_fp16))[name = string("op_4685_cast_fp16")]; + bool var_4687_interleave_0 = const()[name = string("op_4687_interleave_0"), val = bool(false)]; + tensor var_4687_cast_fp16 = concat(axis = var_3236, interleave = var_4687_interleave_0, values = (var_4557_cast_fp16, var_4559_cast_fp16, var_4561_cast_fp16, var_4563_cast_fp16))[name = string("op_4687_cast_fp16")]; + bool var_4689_interleave_0 = const()[name = string("op_4689_interleave_0"), val = bool(false)]; + tensor var_4689_cast_fp16 = concat(axis = var_3236, interleave = var_4689_interleave_0, values = (var_4565_cast_fp16, var_4567_cast_fp16, var_4569_cast_fp16, var_4571_cast_fp16))[name = string("op_4689_cast_fp16")]; + bool var_4691_interleave_0 = const()[name = string("op_4691_interleave_0"), val = bool(false)]; + tensor var_4691_cast_fp16 = concat(axis = var_3236, interleave = var_4691_interleave_0, values = (var_4573_cast_fp16, var_4575_cast_fp16, var_4577_cast_fp16, var_4579_cast_fp16))[name = string("op_4691_cast_fp16")]; + bool var_4693_interleave_0 = const()[name = string("op_4693_interleave_0"), val = bool(false)]; + tensor var_4693_cast_fp16 = concat(axis = var_3236, interleave = var_4693_interleave_0, values = (var_4581_cast_fp16, var_4583_cast_fp16, var_4585_cast_fp16, var_4587_cast_fp16))[name = string("op_4693_cast_fp16")]; + bool var_4695_interleave_0 = const()[name = string("op_4695_interleave_0"), val = bool(false)]; + tensor var_4695_cast_fp16 = concat(axis = var_3236, interleave = var_4695_interleave_0, values = (var_4589_cast_fp16, var_4591_cast_fp16, var_4593_cast_fp16, var_4595_cast_fp16))[name = string("op_4695_cast_fp16")]; + bool var_4697_interleave_0 = const()[name = string("op_4697_interleave_0"), val = bool(false)]; + tensor var_4697_cast_fp16 = concat(axis = var_3236, interleave = var_4697_interleave_0, values = (var_4597_cast_fp16, var_4599_cast_fp16, var_4601_cast_fp16, var_4603_cast_fp16))[name = string("op_4697_cast_fp16")]; + bool var_4699_interleave_0 = const()[name = string("op_4699_interleave_0"), val = bool(false)]; + tensor var_4699_cast_fp16 = concat(axis = var_3236, interleave = var_4699_interleave_0, values = (var_4605_cast_fp16, var_4607_cast_fp16, var_4609_cast_fp16, var_4611_cast_fp16))[name = string("op_4699_cast_fp16")]; + bool var_4701_interleave_0 = const()[name = string("op_4701_interleave_0"), val = bool(false)]; + tensor var_4701_cast_fp16 = concat(axis = var_3236, interleave = var_4701_interleave_0, values = (var_4613_cast_fp16, var_4615_cast_fp16, var_4617_cast_fp16, var_4619_cast_fp16))[name = string("op_4701_cast_fp16")]; + bool var_4703_interleave_0 = const()[name = string("op_4703_interleave_0"), val = bool(false)]; + tensor var_4703_cast_fp16 = concat(axis = var_3236, interleave = var_4703_interleave_0, values = (var_4621_cast_fp16, var_4623_cast_fp16, var_4625_cast_fp16, var_4627_cast_fp16))[name = string("op_4703_cast_fp16")]; + bool var_4705_interleave_0 = const()[name = string("op_4705_interleave_0"), val = bool(false)]; + tensor var_4705_cast_fp16 = concat(axis = var_3236, interleave = var_4705_interleave_0, values = (var_4629_cast_fp16, var_4631_cast_fp16, var_4633_cast_fp16, var_4635_cast_fp16))[name = string("op_4705_cast_fp16")]; + bool var_4707_interleave_0 = const()[name = string("op_4707_interleave_0"), val = bool(false)]; + tensor var_4707_cast_fp16 = concat(axis = var_3236, interleave = var_4707_interleave_0, values = (var_4637_cast_fp16, var_4639_cast_fp16, var_4641_cast_fp16, var_4643_cast_fp16))[name = string("op_4707_cast_fp16")]; + bool var_4709_interleave_0 = const()[name = string("op_4709_interleave_0"), val = bool(false)]; + tensor var_4709_cast_fp16 = concat(axis = var_3236, interleave = var_4709_interleave_0, values = (var_4645_cast_fp16, var_4647_cast_fp16, var_4649_cast_fp16, var_4651_cast_fp16))[name = string("op_4709_cast_fp16")]; + bool var_4711_interleave_0 = const()[name = string("op_4711_interleave_0"), val = bool(false)]; + tensor var_4711_cast_fp16 = concat(axis = var_3236, interleave = var_4711_interleave_0, values = (var_4653_cast_fp16, var_4655_cast_fp16, var_4657_cast_fp16, var_4659_cast_fp16))[name = string("op_4711_cast_fp16")]; + bool var_4713_interleave_0 = const()[name = string("op_4713_interleave_0"), val = bool(false)]; + tensor var_4713_cast_fp16 = concat(axis = var_3236, interleave = var_4713_interleave_0, values = (var_4661_cast_fp16, var_4663_cast_fp16, var_4665_cast_fp16, var_4667_cast_fp16))[name = string("op_4713_cast_fp16")]; + bool var_4715_interleave_0 = const()[name = string("op_4715_interleave_0"), val = bool(false)]; + tensor var_4715_cast_fp16 = concat(axis = var_3236, interleave = var_4715_interleave_0, values = (var_4669_cast_fp16, var_4671_cast_fp16, var_4673_cast_fp16, var_4675_cast_fp16))[name = string("op_4715_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_3261, interleave = input_17_interleave_0, values = (var_4677_cast_fp16, var_4679_cast_fp16, var_4681_cast_fp16, var_4683_cast_fp16, var_4685_cast_fp16, var_4687_cast_fp16, var_4689_cast_fp16, var_4691_cast_fp16, var_4693_cast_fp16, var_4695_cast_fp16, var_4697_cast_fp16, var_4699_cast_fp16, var_4701_cast_fp16, var_4703_cast_fp16, var_4705_cast_fp16, var_4707_cast_fp16, var_4709_cast_fp16, var_4711_cast_fp16, var_4713_cast_fp16, var_4715_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103211840)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106488704)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_4734_to_fp16 = const()[name = string("op_4734_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_4734_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106491328)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106493952)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106496576)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119603840)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119614144)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132721408)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_4763 = const()[name = string("op_4763"), val = int32(3)]; + int32 var_4788 = const()[name = string("op_4788"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_4805_to_fp16 = const()[name = string("op_4805_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_4805_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132724032)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132726656)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132729280)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136006144)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136008768)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139285632)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142562496)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_4843_begin_0 = const()[name = string("op_4843_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4843_end_0 = const()[name = string("op_4843_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4843_end_mask_0 = const()[name = string("op_4843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4843_cast_fp16 = slice_by_index(begin = var_4843_begin_0, end = var_4843_end_0, end_mask = var_4843_end_mask_0, x = query_7_cast_fp16)[name = string("op_4843_cast_fp16")]; + tensor var_4847_begin_0 = const()[name = string("op_4847_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4847_end_0 = const()[name = string("op_4847_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4847_end_mask_0 = const()[name = string("op_4847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4847_cast_fp16 = slice_by_index(begin = var_4847_begin_0, end = var_4847_end_0, end_mask = var_4847_end_mask_0, x = query_7_cast_fp16)[name = string("op_4847_cast_fp16")]; + tensor var_4851_begin_0 = const()[name = string("op_4851_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4851_end_0 = const()[name = string("op_4851_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4851_end_mask_0 = const()[name = string("op_4851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4851_cast_fp16 = slice_by_index(begin = var_4851_begin_0, end = var_4851_end_0, end_mask = var_4851_end_mask_0, x = query_7_cast_fp16)[name = string("op_4851_cast_fp16")]; + tensor var_4855_begin_0 = const()[name = string("op_4855_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4855_end_0 = const()[name = string("op_4855_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4855_end_mask_0 = const()[name = string("op_4855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4855_cast_fp16 = slice_by_index(begin = var_4855_begin_0, end = var_4855_end_0, end_mask = var_4855_end_mask_0, x = query_7_cast_fp16)[name = string("op_4855_cast_fp16")]; + tensor var_4859_begin_0 = const()[name = string("op_4859_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4859_end_0 = const()[name = string("op_4859_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4859_end_mask_0 = const()[name = string("op_4859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4859_cast_fp16 = slice_by_index(begin = var_4859_begin_0, end = var_4859_end_0, end_mask = var_4859_end_mask_0, x = query_7_cast_fp16)[name = string("op_4859_cast_fp16")]; + tensor var_4863_begin_0 = const()[name = string("op_4863_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4863_end_0 = const()[name = string("op_4863_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4863_end_mask_0 = const()[name = string("op_4863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4863_cast_fp16 = slice_by_index(begin = var_4863_begin_0, end = var_4863_end_0, end_mask = var_4863_end_mask_0, x = query_7_cast_fp16)[name = string("op_4863_cast_fp16")]; + tensor var_4867_begin_0 = const()[name = string("op_4867_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4867_end_0 = const()[name = string("op_4867_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4867_end_mask_0 = const()[name = string("op_4867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4867_cast_fp16 = slice_by_index(begin = var_4867_begin_0, end = var_4867_end_0, end_mask = var_4867_end_mask_0, x = query_7_cast_fp16)[name = string("op_4867_cast_fp16")]; + tensor var_4871_begin_0 = const()[name = string("op_4871_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4871_end_0 = const()[name = string("op_4871_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4871_end_mask_0 = const()[name = string("op_4871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4871_cast_fp16 = slice_by_index(begin = var_4871_begin_0, end = var_4871_end_0, end_mask = var_4871_end_mask_0, x = query_7_cast_fp16)[name = string("op_4871_cast_fp16")]; + tensor var_4875_begin_0 = const()[name = string("op_4875_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4875_end_0 = const()[name = string("op_4875_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4875_end_mask_0 = const()[name = string("op_4875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4875_cast_fp16 = slice_by_index(begin = var_4875_begin_0, end = var_4875_end_0, end_mask = var_4875_end_mask_0, x = query_7_cast_fp16)[name = string("op_4875_cast_fp16")]; + tensor var_4879_begin_0 = const()[name = string("op_4879_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4879_end_0 = const()[name = string("op_4879_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4879_end_mask_0 = const()[name = string("op_4879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4879_cast_fp16 = slice_by_index(begin = var_4879_begin_0, end = var_4879_end_0, end_mask = var_4879_end_mask_0, x = query_7_cast_fp16)[name = string("op_4879_cast_fp16")]; + tensor var_4883_begin_0 = const()[name = string("op_4883_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4883_end_0 = const()[name = string("op_4883_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4883_end_mask_0 = const()[name = string("op_4883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4883_cast_fp16 = slice_by_index(begin = var_4883_begin_0, end = var_4883_end_0, end_mask = var_4883_end_mask_0, x = query_7_cast_fp16)[name = string("op_4883_cast_fp16")]; + tensor var_4887_begin_0 = const()[name = string("op_4887_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4887_end_0 = const()[name = string("op_4887_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4887_end_mask_0 = const()[name = string("op_4887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4887_cast_fp16 = slice_by_index(begin = var_4887_begin_0, end = var_4887_end_0, end_mask = var_4887_end_mask_0, x = query_7_cast_fp16)[name = string("op_4887_cast_fp16")]; + tensor var_4891_begin_0 = const()[name = string("op_4891_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4891_end_0 = const()[name = string("op_4891_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_4891_end_mask_0 = const()[name = string("op_4891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4891_cast_fp16 = slice_by_index(begin = var_4891_begin_0, end = var_4891_end_0, end_mask = var_4891_end_mask_0, x = query_7_cast_fp16)[name = string("op_4891_cast_fp16")]; + tensor var_4895_begin_0 = const()[name = string("op_4895_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4895_end_0 = const()[name = string("op_4895_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_4895_end_mask_0 = const()[name = string("op_4895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4895_cast_fp16 = slice_by_index(begin = var_4895_begin_0, end = var_4895_end_0, end_mask = var_4895_end_mask_0, x = query_7_cast_fp16)[name = string("op_4895_cast_fp16")]; + tensor var_4899_begin_0 = const()[name = string("op_4899_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4899_end_0 = const()[name = string("op_4899_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_4899_end_mask_0 = const()[name = string("op_4899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4899_cast_fp16 = slice_by_index(begin = var_4899_begin_0, end = var_4899_end_0, end_mask = var_4899_end_mask_0, x = query_7_cast_fp16)[name = string("op_4899_cast_fp16")]; + tensor var_4903_begin_0 = const()[name = string("op_4903_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4903_end_0 = const()[name = string("op_4903_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_4903_end_mask_0 = const()[name = string("op_4903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4903_cast_fp16 = slice_by_index(begin = var_4903_begin_0, end = var_4903_end_0, end_mask = var_4903_end_mask_0, x = query_7_cast_fp16)[name = string("op_4903_cast_fp16")]; + tensor var_4907_begin_0 = const()[name = string("op_4907_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4907_end_0 = const()[name = string("op_4907_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_4907_end_mask_0 = const()[name = string("op_4907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4907_cast_fp16 = slice_by_index(begin = var_4907_begin_0, end = var_4907_end_0, end_mask = var_4907_end_mask_0, x = query_7_cast_fp16)[name = string("op_4907_cast_fp16")]; + tensor var_4911_begin_0 = const()[name = string("op_4911_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4911_end_0 = const()[name = string("op_4911_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_4911_end_mask_0 = const()[name = string("op_4911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4911_cast_fp16 = slice_by_index(begin = var_4911_begin_0, end = var_4911_end_0, end_mask = var_4911_end_mask_0, x = query_7_cast_fp16)[name = string("op_4911_cast_fp16")]; + tensor var_4915_begin_0 = const()[name = string("op_4915_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4915_end_0 = const()[name = string("op_4915_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_4915_end_mask_0 = const()[name = string("op_4915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4915_cast_fp16 = slice_by_index(begin = var_4915_begin_0, end = var_4915_end_0, end_mask = var_4915_end_mask_0, x = query_7_cast_fp16)[name = string("op_4915_cast_fp16")]; + tensor var_4919_begin_0 = const()[name = string("op_4919_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4919_end_0 = const()[name = string("op_4919_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_4919_end_mask_0 = const()[name = string("op_4919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4919_cast_fp16 = slice_by_index(begin = var_4919_begin_0, end = var_4919_end_0, end_mask = var_4919_end_mask_0, x = query_7_cast_fp16)[name = string("op_4919_cast_fp16")]; + tensor var_4928_begin_0 = const()[name = string("op_4928_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4928_end_0 = const()[name = string("op_4928_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4928_end_mask_0 = const()[name = string("op_4928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4928_cast_fp16 = slice_by_index(begin = var_4928_begin_0, end = var_4928_end_0, end_mask = var_4928_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4928_cast_fp16")]; + tensor var_4935_begin_0 = const()[name = string("op_4935_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4935_end_0 = const()[name = string("op_4935_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4935_end_mask_0 = const()[name = string("op_4935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4935_cast_fp16 = slice_by_index(begin = var_4935_begin_0, end = var_4935_end_0, end_mask = var_4935_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4935_cast_fp16")]; + tensor var_4942_begin_0 = const()[name = string("op_4942_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4942_end_0 = const()[name = string("op_4942_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4942_end_mask_0 = const()[name = string("op_4942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4942_cast_fp16 = slice_by_index(begin = var_4942_begin_0, end = var_4942_end_0, end_mask = var_4942_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4942_cast_fp16")]; + tensor var_4949_begin_0 = const()[name = string("op_4949_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4949_end_0 = const()[name = string("op_4949_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4949_end_mask_0 = const()[name = string("op_4949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4949_cast_fp16 = slice_by_index(begin = var_4949_begin_0, end = var_4949_end_0, end_mask = var_4949_end_mask_0, x = var_4843_cast_fp16)[name = string("op_4949_cast_fp16")]; + tensor var_4956_begin_0 = const()[name = string("op_4956_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4956_end_0 = const()[name = string("op_4956_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4956_end_mask_0 = const()[name = string("op_4956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4956_cast_fp16 = slice_by_index(begin = var_4956_begin_0, end = var_4956_end_0, end_mask = var_4956_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4956_cast_fp16")]; + tensor var_4963_begin_0 = const()[name = string("op_4963_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4963_end_0 = const()[name = string("op_4963_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4963_end_mask_0 = const()[name = string("op_4963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4963_cast_fp16 = slice_by_index(begin = var_4963_begin_0, end = var_4963_end_0, end_mask = var_4963_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4963_cast_fp16")]; + tensor var_4970_begin_0 = const()[name = string("op_4970_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4970_end_0 = const()[name = string("op_4970_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4970_end_mask_0 = const()[name = string("op_4970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4970_cast_fp16 = slice_by_index(begin = var_4970_begin_0, end = var_4970_end_0, end_mask = var_4970_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4970_cast_fp16")]; + tensor var_4977_begin_0 = const()[name = string("op_4977_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4977_end_0 = const()[name = string("op_4977_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4977_end_mask_0 = const()[name = string("op_4977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4977_cast_fp16 = slice_by_index(begin = var_4977_begin_0, end = var_4977_end_0, end_mask = var_4977_end_mask_0, x = var_4847_cast_fp16)[name = string("op_4977_cast_fp16")]; + tensor var_4984_begin_0 = const()[name = string("op_4984_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4984_end_0 = const()[name = string("op_4984_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4984_end_mask_0 = const()[name = string("op_4984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4984_cast_fp16 = slice_by_index(begin = var_4984_begin_0, end = var_4984_end_0, end_mask = var_4984_end_mask_0, x = var_4851_cast_fp16)[name = string("op_4984_cast_fp16")]; + tensor var_4991_begin_0 = const()[name = string("op_4991_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4991_end_0 = const()[name = string("op_4991_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4991_end_mask_0 = const()[name = string("op_4991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4991_cast_fp16 = slice_by_index(begin = var_4991_begin_0, end = var_4991_end_0, end_mask = var_4991_end_mask_0, x = var_4851_cast_fp16)[name = string("op_4991_cast_fp16")]; + tensor var_4998_begin_0 = const()[name = string("op_4998_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4998_end_0 = const()[name = string("op_4998_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4998_end_mask_0 = const()[name = string("op_4998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4998_cast_fp16 = slice_by_index(begin = var_4998_begin_0, end = var_4998_end_0, end_mask = var_4998_end_mask_0, x = var_4851_cast_fp16)[name = string("op_4998_cast_fp16")]; + tensor var_5005_begin_0 = const()[name = string("op_5005_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5005_end_0 = const()[name = string("op_5005_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5005_end_mask_0 = const()[name = string("op_5005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5005_cast_fp16 = slice_by_index(begin = var_5005_begin_0, end = var_5005_end_0, end_mask = var_5005_end_mask_0, x = var_4851_cast_fp16)[name = string("op_5005_cast_fp16")]; + tensor var_5012_begin_0 = const()[name = string("op_5012_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5012_end_0 = const()[name = string("op_5012_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5012_end_mask_0 = const()[name = string("op_5012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5012_cast_fp16 = slice_by_index(begin = var_5012_begin_0, end = var_5012_end_0, end_mask = var_5012_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5012_cast_fp16")]; + tensor var_5019_begin_0 = const()[name = string("op_5019_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5019_end_0 = const()[name = string("op_5019_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5019_end_mask_0 = const()[name = string("op_5019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5019_cast_fp16 = slice_by_index(begin = var_5019_begin_0, end = var_5019_end_0, end_mask = var_5019_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5019_cast_fp16")]; + tensor var_5026_begin_0 = const()[name = string("op_5026_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5026_end_0 = const()[name = string("op_5026_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5026_end_mask_0 = const()[name = string("op_5026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5026_cast_fp16 = slice_by_index(begin = var_5026_begin_0, end = var_5026_end_0, end_mask = var_5026_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5026_cast_fp16")]; + tensor var_5033_begin_0 = const()[name = string("op_5033_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5033_end_0 = const()[name = string("op_5033_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5033_end_mask_0 = const()[name = string("op_5033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5033_cast_fp16 = slice_by_index(begin = var_5033_begin_0, end = var_5033_end_0, end_mask = var_5033_end_mask_0, x = var_4855_cast_fp16)[name = string("op_5033_cast_fp16")]; + tensor var_5040_begin_0 = const()[name = string("op_5040_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5040_end_0 = const()[name = string("op_5040_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5040_end_mask_0 = const()[name = string("op_5040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5040_cast_fp16 = slice_by_index(begin = var_5040_begin_0, end = var_5040_end_0, end_mask = var_5040_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5040_cast_fp16")]; + tensor var_5047_begin_0 = const()[name = string("op_5047_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5047_end_0 = const()[name = string("op_5047_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5047_end_mask_0 = const()[name = string("op_5047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5047_cast_fp16 = slice_by_index(begin = var_5047_begin_0, end = var_5047_end_0, end_mask = var_5047_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5047_cast_fp16")]; + tensor var_5054_begin_0 = const()[name = string("op_5054_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5054_end_0 = const()[name = string("op_5054_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5054_end_mask_0 = const()[name = string("op_5054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5054_cast_fp16 = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5054_cast_fp16")]; + tensor var_5061_begin_0 = const()[name = string("op_5061_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5061_end_0 = const()[name = string("op_5061_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5061_end_mask_0 = const()[name = string("op_5061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5061_cast_fp16 = slice_by_index(begin = var_5061_begin_0, end = var_5061_end_0, end_mask = var_5061_end_mask_0, x = var_4859_cast_fp16)[name = string("op_5061_cast_fp16")]; + tensor var_5068_begin_0 = const()[name = string("op_5068_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5068_end_0 = const()[name = string("op_5068_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5068_end_mask_0 = const()[name = string("op_5068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5068_cast_fp16 = slice_by_index(begin = var_5068_begin_0, end = var_5068_end_0, end_mask = var_5068_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5068_cast_fp16")]; + tensor var_5075_begin_0 = const()[name = string("op_5075_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5075_end_0 = const()[name = string("op_5075_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5075_end_mask_0 = const()[name = string("op_5075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5075_cast_fp16 = slice_by_index(begin = var_5075_begin_0, end = var_5075_end_0, end_mask = var_5075_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5075_cast_fp16")]; + tensor var_5082_begin_0 = const()[name = string("op_5082_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5082_end_0 = const()[name = string("op_5082_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5082_end_mask_0 = const()[name = string("op_5082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5082_cast_fp16 = slice_by_index(begin = var_5082_begin_0, end = var_5082_end_0, end_mask = var_5082_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5082_cast_fp16")]; + tensor var_5089_begin_0 = const()[name = string("op_5089_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5089_end_0 = const()[name = string("op_5089_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5089_end_mask_0 = const()[name = string("op_5089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5089_cast_fp16 = slice_by_index(begin = var_5089_begin_0, end = var_5089_end_0, end_mask = var_5089_end_mask_0, x = var_4863_cast_fp16)[name = string("op_5089_cast_fp16")]; + tensor var_5096_begin_0 = const()[name = string("op_5096_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5096_end_0 = const()[name = string("op_5096_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5096_end_mask_0 = const()[name = string("op_5096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5096_cast_fp16 = slice_by_index(begin = var_5096_begin_0, end = var_5096_end_0, end_mask = var_5096_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5096_cast_fp16")]; + tensor var_5103_begin_0 = const()[name = string("op_5103_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5103_end_0 = const()[name = string("op_5103_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5103_end_mask_0 = const()[name = string("op_5103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5103_cast_fp16 = slice_by_index(begin = var_5103_begin_0, end = var_5103_end_0, end_mask = var_5103_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5103_cast_fp16")]; + tensor var_5110_begin_0 = const()[name = string("op_5110_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5110_end_0 = const()[name = string("op_5110_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5110_end_mask_0 = const()[name = string("op_5110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5110_cast_fp16 = slice_by_index(begin = var_5110_begin_0, end = var_5110_end_0, end_mask = var_5110_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5110_cast_fp16")]; + tensor var_5117_begin_0 = const()[name = string("op_5117_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5117_end_0 = const()[name = string("op_5117_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5117_end_mask_0 = const()[name = string("op_5117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5117_cast_fp16 = slice_by_index(begin = var_5117_begin_0, end = var_5117_end_0, end_mask = var_5117_end_mask_0, x = var_4867_cast_fp16)[name = string("op_5117_cast_fp16")]; + tensor var_5124_begin_0 = const()[name = string("op_5124_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5124_end_0 = const()[name = string("op_5124_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5124_end_mask_0 = const()[name = string("op_5124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5124_cast_fp16 = slice_by_index(begin = var_5124_begin_0, end = var_5124_end_0, end_mask = var_5124_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5124_cast_fp16")]; + tensor var_5131_begin_0 = const()[name = string("op_5131_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5131_end_0 = const()[name = string("op_5131_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5131_end_mask_0 = const()[name = string("op_5131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5131_cast_fp16 = slice_by_index(begin = var_5131_begin_0, end = var_5131_end_0, end_mask = var_5131_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5131_cast_fp16")]; + tensor var_5138_begin_0 = const()[name = string("op_5138_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5138_end_0 = const()[name = string("op_5138_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5138_end_mask_0 = const()[name = string("op_5138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5138_cast_fp16 = slice_by_index(begin = var_5138_begin_0, end = var_5138_end_0, end_mask = var_5138_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5138_cast_fp16")]; + tensor var_5145_begin_0 = const()[name = string("op_5145_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5145_end_0 = const()[name = string("op_5145_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5145_end_mask_0 = const()[name = string("op_5145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5145_cast_fp16 = slice_by_index(begin = var_5145_begin_0, end = var_5145_end_0, end_mask = var_5145_end_mask_0, x = var_4871_cast_fp16)[name = string("op_5145_cast_fp16")]; + tensor var_5152_begin_0 = const()[name = string("op_5152_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5152_end_0 = const()[name = string("op_5152_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5152_end_mask_0 = const()[name = string("op_5152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5152_cast_fp16 = slice_by_index(begin = var_5152_begin_0, end = var_5152_end_0, end_mask = var_5152_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5152_cast_fp16")]; + tensor var_5159_begin_0 = const()[name = string("op_5159_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5159_end_0 = const()[name = string("op_5159_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5159_end_mask_0 = const()[name = string("op_5159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5159_cast_fp16 = slice_by_index(begin = var_5159_begin_0, end = var_5159_end_0, end_mask = var_5159_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5159_cast_fp16")]; + tensor var_5166_begin_0 = const()[name = string("op_5166_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5166_end_0 = const()[name = string("op_5166_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5166_end_mask_0 = const()[name = string("op_5166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5166_cast_fp16 = slice_by_index(begin = var_5166_begin_0, end = var_5166_end_0, end_mask = var_5166_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5166_cast_fp16")]; + tensor var_5173_begin_0 = const()[name = string("op_5173_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5173_end_0 = const()[name = string("op_5173_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5173_end_mask_0 = const()[name = string("op_5173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5173_cast_fp16 = slice_by_index(begin = var_5173_begin_0, end = var_5173_end_0, end_mask = var_5173_end_mask_0, x = var_4875_cast_fp16)[name = string("op_5173_cast_fp16")]; + tensor var_5180_begin_0 = const()[name = string("op_5180_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5180_end_0 = const()[name = string("op_5180_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5180_end_mask_0 = const()[name = string("op_5180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5180_cast_fp16 = slice_by_index(begin = var_5180_begin_0, end = var_5180_end_0, end_mask = var_5180_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5180_cast_fp16")]; + tensor var_5187_begin_0 = const()[name = string("op_5187_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5187_end_0 = const()[name = string("op_5187_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5187_end_mask_0 = const()[name = string("op_5187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5187_cast_fp16 = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5187_cast_fp16")]; + tensor var_5194_begin_0 = const()[name = string("op_5194_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5194_end_0 = const()[name = string("op_5194_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5194_end_mask_0 = const()[name = string("op_5194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5194_cast_fp16 = slice_by_index(begin = var_5194_begin_0, end = var_5194_end_0, end_mask = var_5194_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5194_cast_fp16")]; + tensor var_5201_begin_0 = const()[name = string("op_5201_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5201_end_0 = const()[name = string("op_5201_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5201_end_mask_0 = const()[name = string("op_5201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5201_cast_fp16 = slice_by_index(begin = var_5201_begin_0, end = var_5201_end_0, end_mask = var_5201_end_mask_0, x = var_4879_cast_fp16)[name = string("op_5201_cast_fp16")]; + tensor var_5208_begin_0 = const()[name = string("op_5208_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5208_end_0 = const()[name = string("op_5208_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5208_end_mask_0 = const()[name = string("op_5208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5208_cast_fp16 = slice_by_index(begin = var_5208_begin_0, end = var_5208_end_0, end_mask = var_5208_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5208_cast_fp16")]; + tensor var_5215_begin_0 = const()[name = string("op_5215_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5215_end_0 = const()[name = string("op_5215_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5215_end_mask_0 = const()[name = string("op_5215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5215_cast_fp16 = slice_by_index(begin = var_5215_begin_0, end = var_5215_end_0, end_mask = var_5215_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5215_cast_fp16")]; + tensor var_5222_begin_0 = const()[name = string("op_5222_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5222_end_0 = const()[name = string("op_5222_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5222_end_mask_0 = const()[name = string("op_5222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5222_cast_fp16 = slice_by_index(begin = var_5222_begin_0, end = var_5222_end_0, end_mask = var_5222_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5222_cast_fp16")]; + tensor var_5229_begin_0 = const()[name = string("op_5229_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5229_end_0 = const()[name = string("op_5229_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5229_end_mask_0 = const()[name = string("op_5229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5229_cast_fp16 = slice_by_index(begin = var_5229_begin_0, end = var_5229_end_0, end_mask = var_5229_end_mask_0, x = var_4883_cast_fp16)[name = string("op_5229_cast_fp16")]; + tensor var_5236_begin_0 = const()[name = string("op_5236_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5236_end_0 = const()[name = string("op_5236_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5236_end_mask_0 = const()[name = string("op_5236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5236_cast_fp16 = slice_by_index(begin = var_5236_begin_0, end = var_5236_end_0, end_mask = var_5236_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5236_cast_fp16")]; + tensor var_5243_begin_0 = const()[name = string("op_5243_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5243_end_0 = const()[name = string("op_5243_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5243_end_mask_0 = const()[name = string("op_5243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5243_cast_fp16 = slice_by_index(begin = var_5243_begin_0, end = var_5243_end_0, end_mask = var_5243_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5243_cast_fp16")]; + tensor var_5250_begin_0 = const()[name = string("op_5250_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5250_end_0 = const()[name = string("op_5250_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5250_end_mask_0 = const()[name = string("op_5250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5250_cast_fp16 = slice_by_index(begin = var_5250_begin_0, end = var_5250_end_0, end_mask = var_5250_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5250_cast_fp16")]; + tensor var_5257_begin_0 = const()[name = string("op_5257_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5257_end_0 = const()[name = string("op_5257_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5257_end_mask_0 = const()[name = string("op_5257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5257_cast_fp16 = slice_by_index(begin = var_5257_begin_0, end = var_5257_end_0, end_mask = var_5257_end_mask_0, x = var_4887_cast_fp16)[name = string("op_5257_cast_fp16")]; + tensor var_5264_begin_0 = const()[name = string("op_5264_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5264_end_0 = const()[name = string("op_5264_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5264_end_mask_0 = const()[name = string("op_5264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5264_cast_fp16 = slice_by_index(begin = var_5264_begin_0, end = var_5264_end_0, end_mask = var_5264_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5264_cast_fp16")]; + tensor var_5271_begin_0 = const()[name = string("op_5271_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5271_end_0 = const()[name = string("op_5271_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5271_end_mask_0 = const()[name = string("op_5271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5271_cast_fp16 = slice_by_index(begin = var_5271_begin_0, end = var_5271_end_0, end_mask = var_5271_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5271_cast_fp16")]; + tensor var_5278_begin_0 = const()[name = string("op_5278_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5278_end_0 = const()[name = string("op_5278_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5278_end_mask_0 = const()[name = string("op_5278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5278_cast_fp16 = slice_by_index(begin = var_5278_begin_0, end = var_5278_end_0, end_mask = var_5278_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5278_cast_fp16")]; + tensor var_5285_begin_0 = const()[name = string("op_5285_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5285_end_0 = const()[name = string("op_5285_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5285_end_mask_0 = const()[name = string("op_5285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5285_cast_fp16 = slice_by_index(begin = var_5285_begin_0, end = var_5285_end_0, end_mask = var_5285_end_mask_0, x = var_4891_cast_fp16)[name = string("op_5285_cast_fp16")]; + tensor var_5292_begin_0 = const()[name = string("op_5292_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5292_end_0 = const()[name = string("op_5292_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5292_end_mask_0 = const()[name = string("op_5292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5292_cast_fp16 = slice_by_index(begin = var_5292_begin_0, end = var_5292_end_0, end_mask = var_5292_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5292_cast_fp16")]; + tensor var_5299_begin_0 = const()[name = string("op_5299_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5299_end_0 = const()[name = string("op_5299_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5299_end_mask_0 = const()[name = string("op_5299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5299_cast_fp16 = slice_by_index(begin = var_5299_begin_0, end = var_5299_end_0, end_mask = var_5299_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5299_cast_fp16")]; + tensor var_5306_begin_0 = const()[name = string("op_5306_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5306_end_0 = const()[name = string("op_5306_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5306_end_mask_0 = const()[name = string("op_5306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5306_cast_fp16 = slice_by_index(begin = var_5306_begin_0, end = var_5306_end_0, end_mask = var_5306_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5306_cast_fp16")]; + tensor var_5313_begin_0 = const()[name = string("op_5313_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5313_end_0 = const()[name = string("op_5313_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5313_end_mask_0 = const()[name = string("op_5313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5313_cast_fp16 = slice_by_index(begin = var_5313_begin_0, end = var_5313_end_0, end_mask = var_5313_end_mask_0, x = var_4895_cast_fp16)[name = string("op_5313_cast_fp16")]; + tensor var_5320_begin_0 = const()[name = string("op_5320_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5320_end_0 = const()[name = string("op_5320_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5320_end_mask_0 = const()[name = string("op_5320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5320_cast_fp16 = slice_by_index(begin = var_5320_begin_0, end = var_5320_end_0, end_mask = var_5320_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5320_cast_fp16")]; + tensor var_5327_begin_0 = const()[name = string("op_5327_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5327_end_0 = const()[name = string("op_5327_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5327_end_mask_0 = const()[name = string("op_5327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5327_cast_fp16 = slice_by_index(begin = var_5327_begin_0, end = var_5327_end_0, end_mask = var_5327_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5327_cast_fp16")]; + tensor var_5334_begin_0 = const()[name = string("op_5334_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5334_end_0 = const()[name = string("op_5334_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5334_end_mask_0 = const()[name = string("op_5334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5334_cast_fp16 = slice_by_index(begin = var_5334_begin_0, end = var_5334_end_0, end_mask = var_5334_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5334_cast_fp16")]; + tensor var_5341_begin_0 = const()[name = string("op_5341_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5341_end_0 = const()[name = string("op_5341_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5341_end_mask_0 = const()[name = string("op_5341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5341_cast_fp16 = slice_by_index(begin = var_5341_begin_0, end = var_5341_end_0, end_mask = var_5341_end_mask_0, x = var_4899_cast_fp16)[name = string("op_5341_cast_fp16")]; + tensor var_5348_begin_0 = const()[name = string("op_5348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5348_end_0 = const()[name = string("op_5348_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5348_end_mask_0 = const()[name = string("op_5348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5348_cast_fp16 = slice_by_index(begin = var_5348_begin_0, end = var_5348_end_0, end_mask = var_5348_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5348_cast_fp16")]; + tensor var_5355_begin_0 = const()[name = string("op_5355_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5355_end_0 = const()[name = string("op_5355_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5355_end_mask_0 = const()[name = string("op_5355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5355_cast_fp16 = slice_by_index(begin = var_5355_begin_0, end = var_5355_end_0, end_mask = var_5355_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5355_cast_fp16")]; + tensor var_5362_begin_0 = const()[name = string("op_5362_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5362_end_0 = const()[name = string("op_5362_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5362_end_mask_0 = const()[name = string("op_5362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5362_cast_fp16 = slice_by_index(begin = var_5362_begin_0, end = var_5362_end_0, end_mask = var_5362_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5362_cast_fp16")]; + tensor var_5369_begin_0 = const()[name = string("op_5369_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5369_end_0 = const()[name = string("op_5369_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5369_end_mask_0 = const()[name = string("op_5369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5369_cast_fp16 = slice_by_index(begin = var_5369_begin_0, end = var_5369_end_0, end_mask = var_5369_end_mask_0, x = var_4903_cast_fp16)[name = string("op_5369_cast_fp16")]; + tensor var_5376_begin_0 = const()[name = string("op_5376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5376_end_0 = const()[name = string("op_5376_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5376_end_mask_0 = const()[name = string("op_5376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5376_cast_fp16 = slice_by_index(begin = var_5376_begin_0, end = var_5376_end_0, end_mask = var_5376_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5376_cast_fp16")]; + tensor var_5383_begin_0 = const()[name = string("op_5383_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5383_end_0 = const()[name = string("op_5383_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5383_end_mask_0 = const()[name = string("op_5383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5383_cast_fp16 = slice_by_index(begin = var_5383_begin_0, end = var_5383_end_0, end_mask = var_5383_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5383_cast_fp16")]; + tensor var_5390_begin_0 = const()[name = string("op_5390_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5390_end_0 = const()[name = string("op_5390_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5390_end_mask_0 = const()[name = string("op_5390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5390_cast_fp16 = slice_by_index(begin = var_5390_begin_0, end = var_5390_end_0, end_mask = var_5390_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5390_cast_fp16")]; + tensor var_5397_begin_0 = const()[name = string("op_5397_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5397_end_0 = const()[name = string("op_5397_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5397_end_mask_0 = const()[name = string("op_5397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5397_cast_fp16 = slice_by_index(begin = var_5397_begin_0, end = var_5397_end_0, end_mask = var_5397_end_mask_0, x = var_4907_cast_fp16)[name = string("op_5397_cast_fp16")]; + tensor var_5404_begin_0 = const()[name = string("op_5404_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5404_end_0 = const()[name = string("op_5404_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5404_end_mask_0 = const()[name = string("op_5404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5404_cast_fp16 = slice_by_index(begin = var_5404_begin_0, end = var_5404_end_0, end_mask = var_5404_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5404_cast_fp16")]; + tensor var_5411_begin_0 = const()[name = string("op_5411_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5411_end_0 = const()[name = string("op_5411_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5411_end_mask_0 = const()[name = string("op_5411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5411_cast_fp16 = slice_by_index(begin = var_5411_begin_0, end = var_5411_end_0, end_mask = var_5411_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5411_cast_fp16")]; + tensor var_5418_begin_0 = const()[name = string("op_5418_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5418_end_0 = const()[name = string("op_5418_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5418_end_mask_0 = const()[name = string("op_5418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5418_cast_fp16 = slice_by_index(begin = var_5418_begin_0, end = var_5418_end_0, end_mask = var_5418_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5418_cast_fp16")]; + tensor var_5425_begin_0 = const()[name = string("op_5425_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5425_end_0 = const()[name = string("op_5425_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5425_end_mask_0 = const()[name = string("op_5425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5425_cast_fp16 = slice_by_index(begin = var_5425_begin_0, end = var_5425_end_0, end_mask = var_5425_end_mask_0, x = var_4911_cast_fp16)[name = string("op_5425_cast_fp16")]; + tensor var_5432_begin_0 = const()[name = string("op_5432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5432_end_0 = const()[name = string("op_5432_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5432_end_mask_0 = const()[name = string("op_5432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5432_cast_fp16 = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5432_cast_fp16")]; + tensor var_5439_begin_0 = const()[name = string("op_5439_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5439_end_0 = const()[name = string("op_5439_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5439_end_mask_0 = const()[name = string("op_5439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5439_cast_fp16 = slice_by_index(begin = var_5439_begin_0, end = var_5439_end_0, end_mask = var_5439_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5439_cast_fp16")]; + tensor var_5446_begin_0 = const()[name = string("op_5446_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5446_end_0 = const()[name = string("op_5446_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5446_end_mask_0 = const()[name = string("op_5446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5446_cast_fp16 = slice_by_index(begin = var_5446_begin_0, end = var_5446_end_0, end_mask = var_5446_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5446_cast_fp16")]; + tensor var_5453_begin_0 = const()[name = string("op_5453_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5453_end_0 = const()[name = string("op_5453_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5453_end_mask_0 = const()[name = string("op_5453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5453_cast_fp16 = slice_by_index(begin = var_5453_begin_0, end = var_5453_end_0, end_mask = var_5453_end_mask_0, x = var_4915_cast_fp16)[name = string("op_5453_cast_fp16")]; + tensor var_5460_begin_0 = const()[name = string("op_5460_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5460_end_0 = const()[name = string("op_5460_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5460_end_mask_0 = const()[name = string("op_5460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5460_cast_fp16 = slice_by_index(begin = var_5460_begin_0, end = var_5460_end_0, end_mask = var_5460_end_mask_0, x = var_4919_cast_fp16)[name = string("op_5460_cast_fp16")]; + tensor var_5467_begin_0 = const()[name = string("op_5467_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5467_end_0 = const()[name = string("op_5467_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5467_end_mask_0 = const()[name = string("op_5467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5467_cast_fp16 = slice_by_index(begin = var_5467_begin_0, end = var_5467_end_0, end_mask = var_5467_end_mask_0, x = var_4919_cast_fp16)[name = string("op_5467_cast_fp16")]; + tensor var_5474_begin_0 = const()[name = string("op_5474_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5474_end_0 = const()[name = string("op_5474_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5474_end_mask_0 = const()[name = string("op_5474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5474_cast_fp16 = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = var_4919_cast_fp16)[name = string("op_5474_cast_fp16")]; + tensor var_5481_begin_0 = const()[name = string("op_5481_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5481_end_0 = const()[name = string("op_5481_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5481_end_mask_0 = const()[name = string("op_5481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5481_cast_fp16 = slice_by_index(begin = var_5481_begin_0, end = var_5481_end_0, end_mask = var_5481_end_mask_0, x = var_4919_cast_fp16)[name = string("op_5481_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5486_begin_0 = const()[name = string("op_5486_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5486_end_0 = const()[name = string("op_5486_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_5486_end_mask_0 = const()[name = string("op_5486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_28")]; + tensor var_5486_cast_fp16 = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = k_7_cast_fp16)[name = string("op_5486_cast_fp16")]; + tensor var_5490_begin_0 = const()[name = string("op_5490_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5490_end_0 = const()[name = string("op_5490_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_5490_end_mask_0 = const()[name = string("op_5490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5490_cast_fp16 = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = k_7_cast_fp16)[name = string("op_5490_cast_fp16")]; + tensor var_5494_begin_0 = const()[name = string("op_5494_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5494_end_0 = const()[name = string("op_5494_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_5494_end_mask_0 = const()[name = string("op_5494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5494_cast_fp16 = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = k_7_cast_fp16)[name = string("op_5494_cast_fp16")]; + tensor var_5498_begin_0 = const()[name = string("op_5498_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5498_end_0 = const()[name = string("op_5498_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_5498_end_mask_0 = const()[name = string("op_5498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = k_7_cast_fp16)[name = string("op_5498_cast_fp16")]; + tensor var_5502_begin_0 = const()[name = string("op_5502_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5502_end_0 = const()[name = string("op_5502_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_5502_end_mask_0 = const()[name = string("op_5502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5502_cast_fp16 = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = k_7_cast_fp16)[name = string("op_5502_cast_fp16")]; + tensor var_5506_begin_0 = const()[name = string("op_5506_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5506_end_0 = const()[name = string("op_5506_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_5506_end_mask_0 = const()[name = string("op_5506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5506_cast_fp16 = slice_by_index(begin = var_5506_begin_0, end = var_5506_end_0, end_mask = var_5506_end_mask_0, x = k_7_cast_fp16)[name = string("op_5506_cast_fp16")]; + tensor var_5510_begin_0 = const()[name = string("op_5510_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5510_end_0 = const()[name = string("op_5510_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_5510_end_mask_0 = const()[name = string("op_5510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5510_cast_fp16 = slice_by_index(begin = var_5510_begin_0, end = var_5510_end_0, end_mask = var_5510_end_mask_0, x = k_7_cast_fp16)[name = string("op_5510_cast_fp16")]; + tensor var_5514_begin_0 = const()[name = string("op_5514_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5514_end_0 = const()[name = string("op_5514_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_5514_end_mask_0 = const()[name = string("op_5514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5514_cast_fp16 = slice_by_index(begin = var_5514_begin_0, end = var_5514_end_0, end_mask = var_5514_end_mask_0, x = k_7_cast_fp16)[name = string("op_5514_cast_fp16")]; + tensor var_5518_begin_0 = const()[name = string("op_5518_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5518_end_0 = const()[name = string("op_5518_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_5518_end_mask_0 = const()[name = string("op_5518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5518_cast_fp16 = slice_by_index(begin = var_5518_begin_0, end = var_5518_end_0, end_mask = var_5518_end_mask_0, x = k_7_cast_fp16)[name = string("op_5518_cast_fp16")]; + tensor var_5522_begin_0 = const()[name = string("op_5522_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5522_end_0 = const()[name = string("op_5522_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_5522_end_mask_0 = const()[name = string("op_5522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5522_cast_fp16 = slice_by_index(begin = var_5522_begin_0, end = var_5522_end_0, end_mask = var_5522_end_mask_0, x = k_7_cast_fp16)[name = string("op_5522_cast_fp16")]; + tensor var_5526_begin_0 = const()[name = string("op_5526_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5526_end_0 = const()[name = string("op_5526_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_5526_end_mask_0 = const()[name = string("op_5526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5526_cast_fp16 = slice_by_index(begin = var_5526_begin_0, end = var_5526_end_0, end_mask = var_5526_end_mask_0, x = k_7_cast_fp16)[name = string("op_5526_cast_fp16")]; + tensor var_5530_begin_0 = const()[name = string("op_5530_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5530_end_0 = const()[name = string("op_5530_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_5530_end_mask_0 = const()[name = string("op_5530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5530_cast_fp16 = slice_by_index(begin = var_5530_begin_0, end = var_5530_end_0, end_mask = var_5530_end_mask_0, x = k_7_cast_fp16)[name = string("op_5530_cast_fp16")]; + tensor var_5534_begin_0 = const()[name = string("op_5534_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5534_end_0 = const()[name = string("op_5534_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_5534_end_mask_0 = const()[name = string("op_5534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5534_cast_fp16 = slice_by_index(begin = var_5534_begin_0, end = var_5534_end_0, end_mask = var_5534_end_mask_0, x = k_7_cast_fp16)[name = string("op_5534_cast_fp16")]; + tensor var_5538_begin_0 = const()[name = string("op_5538_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5538_end_0 = const()[name = string("op_5538_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_5538_end_mask_0 = const()[name = string("op_5538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5538_cast_fp16 = slice_by_index(begin = var_5538_begin_0, end = var_5538_end_0, end_mask = var_5538_end_mask_0, x = k_7_cast_fp16)[name = string("op_5538_cast_fp16")]; + tensor var_5542_begin_0 = const()[name = string("op_5542_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5542_end_0 = const()[name = string("op_5542_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_5542_end_mask_0 = const()[name = string("op_5542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5542_cast_fp16 = slice_by_index(begin = var_5542_begin_0, end = var_5542_end_0, end_mask = var_5542_end_mask_0, x = k_7_cast_fp16)[name = string("op_5542_cast_fp16")]; + tensor var_5546_begin_0 = const()[name = string("op_5546_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5546_end_0 = const()[name = string("op_5546_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_5546_end_mask_0 = const()[name = string("op_5546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5546_cast_fp16 = slice_by_index(begin = var_5546_begin_0, end = var_5546_end_0, end_mask = var_5546_end_mask_0, x = k_7_cast_fp16)[name = string("op_5546_cast_fp16")]; + tensor var_5550_begin_0 = const()[name = string("op_5550_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5550_end_0 = const()[name = string("op_5550_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_5550_end_mask_0 = const()[name = string("op_5550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5550_cast_fp16 = slice_by_index(begin = var_5550_begin_0, end = var_5550_end_0, end_mask = var_5550_end_mask_0, x = k_7_cast_fp16)[name = string("op_5550_cast_fp16")]; + tensor var_5554_begin_0 = const()[name = string("op_5554_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5554_end_0 = const()[name = string("op_5554_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_5554_end_mask_0 = const()[name = string("op_5554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5554_cast_fp16 = slice_by_index(begin = var_5554_begin_0, end = var_5554_end_0, end_mask = var_5554_end_mask_0, x = k_7_cast_fp16)[name = string("op_5554_cast_fp16")]; + tensor var_5558_begin_0 = const()[name = string("op_5558_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5558_end_0 = const()[name = string("op_5558_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_5558_end_mask_0 = const()[name = string("op_5558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5558_cast_fp16 = slice_by_index(begin = var_5558_begin_0, end = var_5558_end_0, end_mask = var_5558_end_mask_0, x = k_7_cast_fp16)[name = string("op_5558_cast_fp16")]; + tensor var_5562_begin_0 = const()[name = string("op_5562_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5562_end_0 = const()[name = string("op_5562_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_5562_end_mask_0 = const()[name = string("op_5562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5562_cast_fp16 = slice_by_index(begin = var_5562_begin_0, end = var_5562_end_0, end_mask = var_5562_end_mask_0, x = k_7_cast_fp16)[name = string("op_5562_cast_fp16")]; + tensor var_5564_begin_0 = const()[name = string("op_5564_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5564_end_0 = const()[name = string("op_5564_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5564_end_mask_0 = const()[name = string("op_5564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5564_cast_fp16 = slice_by_index(begin = var_5564_begin_0, end = var_5564_end_0, end_mask = var_5564_end_mask_0, x = value_7_cast_fp16)[name = string("op_5564_cast_fp16")]; + tensor var_5568_begin_0 = const()[name = string("op_5568_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5568_end_0 = const()[name = string("op_5568_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5568_end_mask_0 = const()[name = string("op_5568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5568_cast_fp16 = slice_by_index(begin = var_5568_begin_0, end = var_5568_end_0, end_mask = var_5568_end_mask_0, x = value_7_cast_fp16)[name = string("op_5568_cast_fp16")]; + tensor var_5572_begin_0 = const()[name = string("op_5572_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5572_end_0 = const()[name = string("op_5572_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5572_end_mask_0 = const()[name = string("op_5572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5572_cast_fp16 = slice_by_index(begin = var_5572_begin_0, end = var_5572_end_0, end_mask = var_5572_end_mask_0, x = value_7_cast_fp16)[name = string("op_5572_cast_fp16")]; + tensor var_5576_begin_0 = const()[name = string("op_5576_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5576_end_0 = const()[name = string("op_5576_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5576_end_mask_0 = const()[name = string("op_5576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5576_cast_fp16 = slice_by_index(begin = var_5576_begin_0, end = var_5576_end_0, end_mask = var_5576_end_mask_0, x = value_7_cast_fp16)[name = string("op_5576_cast_fp16")]; + tensor var_5580_begin_0 = const()[name = string("op_5580_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5580_end_0 = const()[name = string("op_5580_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5580_end_mask_0 = const()[name = string("op_5580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5580_cast_fp16 = slice_by_index(begin = var_5580_begin_0, end = var_5580_end_0, end_mask = var_5580_end_mask_0, x = value_7_cast_fp16)[name = string("op_5580_cast_fp16")]; + tensor var_5584_begin_0 = const()[name = string("op_5584_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5584_end_0 = const()[name = string("op_5584_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5584_end_mask_0 = const()[name = string("op_5584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5584_cast_fp16 = slice_by_index(begin = var_5584_begin_0, end = var_5584_end_0, end_mask = var_5584_end_mask_0, x = value_7_cast_fp16)[name = string("op_5584_cast_fp16")]; + tensor var_5588_begin_0 = const()[name = string("op_5588_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5588_end_0 = const()[name = string("op_5588_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5588_end_mask_0 = const()[name = string("op_5588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5588_cast_fp16 = slice_by_index(begin = var_5588_begin_0, end = var_5588_end_0, end_mask = var_5588_end_mask_0, x = value_7_cast_fp16)[name = string("op_5588_cast_fp16")]; + tensor var_5592_begin_0 = const()[name = string("op_5592_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5592_end_0 = const()[name = string("op_5592_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5592_end_mask_0 = const()[name = string("op_5592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5592_cast_fp16 = slice_by_index(begin = var_5592_begin_0, end = var_5592_end_0, end_mask = var_5592_end_mask_0, x = value_7_cast_fp16)[name = string("op_5592_cast_fp16")]; + tensor var_5596_begin_0 = const()[name = string("op_5596_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5596_end_0 = const()[name = string("op_5596_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5596_end_mask_0 = const()[name = string("op_5596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5596_cast_fp16 = slice_by_index(begin = var_5596_begin_0, end = var_5596_end_0, end_mask = var_5596_end_mask_0, x = value_7_cast_fp16)[name = string("op_5596_cast_fp16")]; + tensor var_5600_begin_0 = const()[name = string("op_5600_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5600_end_0 = const()[name = string("op_5600_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5600_end_mask_0 = const()[name = string("op_5600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5600_cast_fp16 = slice_by_index(begin = var_5600_begin_0, end = var_5600_end_0, end_mask = var_5600_end_mask_0, x = value_7_cast_fp16)[name = string("op_5600_cast_fp16")]; + tensor var_5604_begin_0 = const()[name = string("op_5604_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5604_end_0 = const()[name = string("op_5604_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5604_end_mask_0 = const()[name = string("op_5604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5604_cast_fp16 = slice_by_index(begin = var_5604_begin_0, end = var_5604_end_0, end_mask = var_5604_end_mask_0, x = value_7_cast_fp16)[name = string("op_5604_cast_fp16")]; + tensor var_5608_begin_0 = const()[name = string("op_5608_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5608_end_0 = const()[name = string("op_5608_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5608_end_mask_0 = const()[name = string("op_5608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5608_cast_fp16 = slice_by_index(begin = var_5608_begin_0, end = var_5608_end_0, end_mask = var_5608_end_mask_0, x = value_7_cast_fp16)[name = string("op_5608_cast_fp16")]; + tensor var_5612_begin_0 = const()[name = string("op_5612_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5612_end_0 = const()[name = string("op_5612_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_5612_end_mask_0 = const()[name = string("op_5612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5612_cast_fp16 = slice_by_index(begin = var_5612_begin_0, end = var_5612_end_0, end_mask = var_5612_end_mask_0, x = value_7_cast_fp16)[name = string("op_5612_cast_fp16")]; + tensor var_5616_begin_0 = const()[name = string("op_5616_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5616_end_0 = const()[name = string("op_5616_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_5616_end_mask_0 = const()[name = string("op_5616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5616_cast_fp16 = slice_by_index(begin = var_5616_begin_0, end = var_5616_end_0, end_mask = var_5616_end_mask_0, x = value_7_cast_fp16)[name = string("op_5616_cast_fp16")]; + tensor var_5620_begin_0 = const()[name = string("op_5620_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5620_end_0 = const()[name = string("op_5620_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_5620_end_mask_0 = const()[name = string("op_5620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5620_cast_fp16 = slice_by_index(begin = var_5620_begin_0, end = var_5620_end_0, end_mask = var_5620_end_mask_0, x = value_7_cast_fp16)[name = string("op_5620_cast_fp16")]; + tensor var_5624_begin_0 = const()[name = string("op_5624_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5624_end_0 = const()[name = string("op_5624_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_5624_end_mask_0 = const()[name = string("op_5624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5624_cast_fp16 = slice_by_index(begin = var_5624_begin_0, end = var_5624_end_0, end_mask = var_5624_end_mask_0, x = value_7_cast_fp16)[name = string("op_5624_cast_fp16")]; + tensor var_5628_begin_0 = const()[name = string("op_5628_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5628_end_0 = const()[name = string("op_5628_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_5628_end_mask_0 = const()[name = string("op_5628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5628_cast_fp16 = slice_by_index(begin = var_5628_begin_0, end = var_5628_end_0, end_mask = var_5628_end_mask_0, x = value_7_cast_fp16)[name = string("op_5628_cast_fp16")]; + tensor var_5632_begin_0 = const()[name = string("op_5632_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5632_end_0 = const()[name = string("op_5632_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_5632_end_mask_0 = const()[name = string("op_5632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5632_cast_fp16 = slice_by_index(begin = var_5632_begin_0, end = var_5632_end_0, end_mask = var_5632_end_mask_0, x = value_7_cast_fp16)[name = string("op_5632_cast_fp16")]; + tensor var_5636_begin_0 = const()[name = string("op_5636_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5636_end_0 = const()[name = string("op_5636_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_5636_end_mask_0 = const()[name = string("op_5636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5636_cast_fp16 = slice_by_index(begin = var_5636_begin_0, end = var_5636_end_0, end_mask = var_5636_end_mask_0, x = value_7_cast_fp16)[name = string("op_5636_cast_fp16")]; + tensor var_5640_begin_0 = const()[name = string("op_5640_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5640_end_0 = const()[name = string("op_5640_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_5640_end_mask_0 = const()[name = string("op_5640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5640_cast_fp16 = slice_by_index(begin = var_5640_begin_0, end = var_5640_end_0, end_mask = var_5640_end_mask_0, x = value_7_cast_fp16)[name = string("op_5640_cast_fp16")]; + string _SplitHeadsQ__mh_w_481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_481_equation_0, values = (var_5486_cast_fp16, var_4928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_481_cast_fp16")]; + string _SplitHeadsQ__mh_w_483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_483_equation_0, values = (var_5486_cast_fp16, var_4935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_483_cast_fp16")]; + string _SplitHeadsQ__mh_w_485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_485_equation_0, values = (var_5486_cast_fp16, var_4942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_485_cast_fp16")]; + string _SplitHeadsQ__mh_w_487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_487_equation_0, values = (var_5486_cast_fp16, var_4949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_487_cast_fp16")]; + string _SplitHeadsQ__mh_w_489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_489_equation_0, values = (var_5490_cast_fp16, var_4956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_489_cast_fp16")]; + string _SplitHeadsQ__mh_w_491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_491_equation_0, values = (var_5490_cast_fp16, var_4963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_491_cast_fp16")]; + string _SplitHeadsQ__mh_w_493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_493_equation_0, values = (var_5490_cast_fp16, var_4970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_493_cast_fp16")]; + string _SplitHeadsQ__mh_w_495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_495_equation_0, values = (var_5490_cast_fp16, var_4977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_495_cast_fp16")]; + string _SplitHeadsQ__mh_w_497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_497_equation_0, values = (var_5494_cast_fp16, var_4984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_497_cast_fp16")]; + string _SplitHeadsQ__mh_w_499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_499_equation_0, values = (var_5494_cast_fp16, var_4991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_499_cast_fp16")]; + string _SplitHeadsQ__mh_w_501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_501_equation_0, values = (var_5494_cast_fp16, var_4998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_501_cast_fp16")]; + string _SplitHeadsQ__mh_w_503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_503_equation_0, values = (var_5494_cast_fp16, var_5005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_503_cast_fp16")]; + string _SplitHeadsQ__mh_w_505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_505_equation_0, values = (var_5498_cast_fp16, var_5012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_505_cast_fp16")]; + string _SplitHeadsQ__mh_w_507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_507_equation_0, values = (var_5498_cast_fp16, var_5019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_507_cast_fp16")]; + string _SplitHeadsQ__mh_w_509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_509_equation_0, values = (var_5498_cast_fp16, var_5026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_509_cast_fp16")]; + string _SplitHeadsQ__mh_w_511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_511_equation_0, values = (var_5498_cast_fp16, var_5033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_511_cast_fp16")]; + string _SplitHeadsQ__mh_w_513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_513_equation_0, values = (var_5502_cast_fp16, var_5040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_513_cast_fp16")]; + string _SplitHeadsQ__mh_w_515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_515_equation_0, values = (var_5502_cast_fp16, var_5047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_515_cast_fp16")]; + string _SplitHeadsQ__mh_w_517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_517_equation_0, values = (var_5502_cast_fp16, var_5054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_517_cast_fp16")]; + string _SplitHeadsQ__mh_w_519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_519_equation_0, values = (var_5502_cast_fp16, var_5061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_521_equation_0, values = (var_5506_cast_fp16, var_5068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_521_cast_fp16")]; + string _SplitHeadsQ__mh_w_523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_523_equation_0, values = (var_5506_cast_fp16, var_5075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_523_cast_fp16")]; + string _SplitHeadsQ__mh_w_525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_525_equation_0, values = (var_5506_cast_fp16, var_5082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_525_cast_fp16")]; + string _SplitHeadsQ__mh_w_527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_527_equation_0, values = (var_5506_cast_fp16, var_5089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_527_cast_fp16")]; + string _SplitHeadsQ__mh_w_529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_529_equation_0, values = (var_5510_cast_fp16, var_5096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_529_cast_fp16")]; + string _SplitHeadsQ__mh_w_531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_531_equation_0, values = (var_5510_cast_fp16, var_5103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_531_cast_fp16")]; + string _SplitHeadsQ__mh_w_533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_533_equation_0, values = (var_5510_cast_fp16, var_5110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_533_cast_fp16")]; + string _SplitHeadsQ__mh_w_535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_535_equation_0, values = (var_5510_cast_fp16, var_5117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_535_cast_fp16")]; + string _SplitHeadsQ__mh_w_537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_537_equation_0, values = (var_5514_cast_fp16, var_5124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_537_cast_fp16")]; + string _SplitHeadsQ__mh_w_539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_539_equation_0, values = (var_5514_cast_fp16, var_5131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_539_cast_fp16")]; + string _SplitHeadsQ__mh_w_541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_541_equation_0, values = (var_5514_cast_fp16, var_5138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_541_cast_fp16")]; + string _SplitHeadsQ__mh_w_543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_543_equation_0, values = (var_5514_cast_fp16, var_5145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_543_cast_fp16")]; + string _SplitHeadsQ__mh_w_545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_545_equation_0, values = (var_5518_cast_fp16, var_5152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_545_cast_fp16")]; + string _SplitHeadsQ__mh_w_547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_547_equation_0, values = (var_5518_cast_fp16, var_5159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_547_cast_fp16")]; + string _SplitHeadsQ__mh_w_549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_549_equation_0, values = (var_5518_cast_fp16, var_5166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_549_cast_fp16")]; + string _SplitHeadsQ__mh_w_551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_551_equation_0, values = (var_5518_cast_fp16, var_5173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_551_cast_fp16")]; + string _SplitHeadsQ__mh_w_553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_553_equation_0, values = (var_5522_cast_fp16, var_5180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_553_cast_fp16")]; + string _SplitHeadsQ__mh_w_555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_555_equation_0, values = (var_5522_cast_fp16, var_5187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_555_cast_fp16")]; + string _SplitHeadsQ__mh_w_557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_557_equation_0, values = (var_5522_cast_fp16, var_5194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_557_cast_fp16")]; + string _SplitHeadsQ__mh_w_559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_559_equation_0, values = (var_5522_cast_fp16, var_5201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_559_cast_fp16")]; + string _SplitHeadsQ__mh_w_561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_561_equation_0, values = (var_5526_cast_fp16, var_5208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_561_cast_fp16")]; + string _SplitHeadsQ__mh_w_563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_563_equation_0, values = (var_5526_cast_fp16, var_5215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_563_cast_fp16")]; + string _SplitHeadsQ__mh_w_565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_565_equation_0, values = (var_5526_cast_fp16, var_5222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_565_cast_fp16")]; + string _SplitHeadsQ__mh_w_567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_567_equation_0, values = (var_5526_cast_fp16, var_5229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_567_cast_fp16")]; + string _SplitHeadsQ__mh_w_569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_569_equation_0, values = (var_5530_cast_fp16, var_5236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_569_cast_fp16")]; + string _SplitHeadsQ__mh_w_571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_571_equation_0, values = (var_5530_cast_fp16, var_5243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_571_cast_fp16")]; + string _SplitHeadsQ__mh_w_573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_573_equation_0, values = (var_5530_cast_fp16, var_5250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_573_cast_fp16")]; + string _SplitHeadsQ__mh_w_575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_575_equation_0, values = (var_5530_cast_fp16, var_5257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_575_cast_fp16")]; + string _SplitHeadsQ__mh_w_577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_577_equation_0, values = (var_5534_cast_fp16, var_5264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_577_cast_fp16")]; + string _SplitHeadsQ__mh_w_579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_579_equation_0, values = (var_5534_cast_fp16, var_5271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_579_cast_fp16")]; + string _SplitHeadsQ__mh_w_581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_581_equation_0, values = (var_5534_cast_fp16, var_5278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_581_cast_fp16")]; + string _SplitHeadsQ__mh_w_583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_583_equation_0, values = (var_5534_cast_fp16, var_5285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_583_cast_fp16")]; + string _SplitHeadsQ__mh_w_585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_585_equation_0, values = (var_5538_cast_fp16, var_5292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_585_cast_fp16")]; + string _SplitHeadsQ__mh_w_587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_587_equation_0, values = (var_5538_cast_fp16, var_5299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_587_cast_fp16")]; + string _SplitHeadsQ__mh_w_589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_589_equation_0, values = (var_5538_cast_fp16, var_5306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_589_cast_fp16")]; + string _SplitHeadsQ__mh_w_591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_591_equation_0, values = (var_5538_cast_fp16, var_5313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_591_cast_fp16")]; + string _SplitHeadsQ__mh_w_593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_593_equation_0, values = (var_5542_cast_fp16, var_5320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_593_cast_fp16")]; + string _SplitHeadsQ__mh_w_595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_595_equation_0, values = (var_5542_cast_fp16, var_5327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_595_cast_fp16")]; + string _SplitHeadsQ__mh_w_597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_597_equation_0, values = (var_5542_cast_fp16, var_5334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_597_cast_fp16")]; + string _SplitHeadsQ__mh_w_599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_599_equation_0, values = (var_5542_cast_fp16, var_5341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_599_cast_fp16")]; + string _SplitHeadsQ__mh_w_601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_601_equation_0, values = (var_5546_cast_fp16, var_5348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_601_cast_fp16")]; + string _SplitHeadsQ__mh_w_603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_603_equation_0, values = (var_5546_cast_fp16, var_5355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_603_cast_fp16")]; + string _SplitHeadsQ__mh_w_605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_605_equation_0, values = (var_5546_cast_fp16, var_5362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_605_cast_fp16")]; + string _SplitHeadsQ__mh_w_607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_607_equation_0, values = (var_5546_cast_fp16, var_5369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_607_cast_fp16")]; + string _SplitHeadsQ__mh_w_609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_609_equation_0, values = (var_5550_cast_fp16, var_5376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_609_cast_fp16")]; + string _SplitHeadsQ__mh_w_611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_611_equation_0, values = (var_5550_cast_fp16, var_5383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_611_cast_fp16")]; + string _SplitHeadsQ__mh_w_613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_613_equation_0, values = (var_5550_cast_fp16, var_5390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_613_cast_fp16")]; + string _SplitHeadsQ__mh_w_615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_615_equation_0, values = (var_5550_cast_fp16, var_5397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_615_cast_fp16")]; + string _SplitHeadsQ__mh_w_617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_617_equation_0, values = (var_5554_cast_fp16, var_5404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_617_cast_fp16")]; + string _SplitHeadsQ__mh_w_619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_619_equation_0, values = (var_5554_cast_fp16, var_5411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_619_cast_fp16")]; + string _SplitHeadsQ__mh_w_621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_621_equation_0, values = (var_5554_cast_fp16, var_5418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_621_cast_fp16")]; + string _SplitHeadsQ__mh_w_623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_623_equation_0, values = (var_5554_cast_fp16, var_5425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_623_cast_fp16")]; + string _SplitHeadsQ__mh_w_625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_625_equation_0, values = (var_5558_cast_fp16, var_5432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_625_cast_fp16")]; + string _SplitHeadsQ__mh_w_627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_627_equation_0, values = (var_5558_cast_fp16, var_5439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_627_cast_fp16")]; + string _SplitHeadsQ__mh_w_629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_629_equation_0, values = (var_5558_cast_fp16, var_5446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_629_cast_fp16")]; + string _SplitHeadsQ__mh_w_631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_631_equation_0, values = (var_5558_cast_fp16, var_5453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_631_cast_fp16")]; + string _SplitHeadsQ__mh_w_633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_633_equation_0, values = (var_5562_cast_fp16, var_5460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_633_cast_fp16")]; + string _SplitHeadsQ__mh_w_635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_635_equation_0, values = (var_5562_cast_fp16, var_5467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_635_cast_fp16")]; + string _SplitHeadsQ__mh_w_637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_637_equation_0, values = (var_5562_cast_fp16, var_5474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_637_cast_fp16")]; + string _SplitHeadsQ__mh_w_639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_639_equation_0, values = (var_5562_cast_fp16, var_5481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_639_cast_fp16")]; + fp16 var_5803_to_fp16 = const()[name = string("op_5803_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_481_cast_fp16, y = var_5803_to_fp16)[name = string("aw_chunk_481_cast_fp16")]; + fp16 var_5805_to_fp16 = const()[name = string("op_5805_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_483_cast_fp16, y = var_5805_to_fp16)[name = string("aw_chunk_483_cast_fp16")]; + fp16 var_5807_to_fp16 = const()[name = string("op_5807_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_485_cast_fp16, y = var_5807_to_fp16)[name = string("aw_chunk_485_cast_fp16")]; + fp16 var_5809_to_fp16 = const()[name = string("op_5809_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_487_cast_fp16, y = var_5809_to_fp16)[name = string("aw_chunk_487_cast_fp16")]; + fp16 var_5811_to_fp16 = const()[name = string("op_5811_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_489_cast_fp16, y = var_5811_to_fp16)[name = string("aw_chunk_489_cast_fp16")]; + fp16 var_5813_to_fp16 = const()[name = string("op_5813_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_491_cast_fp16, y = var_5813_to_fp16)[name = string("aw_chunk_491_cast_fp16")]; + fp16 var_5815_to_fp16 = const()[name = string("op_5815_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_493_cast_fp16, y = var_5815_to_fp16)[name = string("aw_chunk_493_cast_fp16")]; + fp16 var_5817_to_fp16 = const()[name = string("op_5817_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_495_cast_fp16, y = var_5817_to_fp16)[name = string("aw_chunk_495_cast_fp16")]; + fp16 var_5819_to_fp16 = const()[name = string("op_5819_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_497_cast_fp16, y = var_5819_to_fp16)[name = string("aw_chunk_497_cast_fp16")]; + fp16 var_5821_to_fp16 = const()[name = string("op_5821_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_499_cast_fp16, y = var_5821_to_fp16)[name = string("aw_chunk_499_cast_fp16")]; + fp16 var_5823_to_fp16 = const()[name = string("op_5823_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_501_cast_fp16, y = var_5823_to_fp16)[name = string("aw_chunk_501_cast_fp16")]; + fp16 var_5825_to_fp16 = const()[name = string("op_5825_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_503_cast_fp16, y = var_5825_to_fp16)[name = string("aw_chunk_503_cast_fp16")]; + fp16 var_5827_to_fp16 = const()[name = string("op_5827_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_505_cast_fp16, y = var_5827_to_fp16)[name = string("aw_chunk_505_cast_fp16")]; + fp16 var_5829_to_fp16 = const()[name = string("op_5829_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_507_cast_fp16, y = var_5829_to_fp16)[name = string("aw_chunk_507_cast_fp16")]; + fp16 var_5831_to_fp16 = const()[name = string("op_5831_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_509_cast_fp16, y = var_5831_to_fp16)[name = string("aw_chunk_509_cast_fp16")]; + fp16 var_5833_to_fp16 = const()[name = string("op_5833_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_511_cast_fp16, y = var_5833_to_fp16)[name = string("aw_chunk_511_cast_fp16")]; + fp16 var_5835_to_fp16 = const()[name = string("op_5835_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_513_cast_fp16, y = var_5835_to_fp16)[name = string("aw_chunk_513_cast_fp16")]; + fp16 var_5837_to_fp16 = const()[name = string("op_5837_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_515_cast_fp16, y = var_5837_to_fp16)[name = string("aw_chunk_515_cast_fp16")]; + fp16 var_5839_to_fp16 = const()[name = string("op_5839_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_517_cast_fp16, y = var_5839_to_fp16)[name = string("aw_chunk_517_cast_fp16")]; + fp16 var_5841_to_fp16 = const()[name = string("op_5841_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_519_cast_fp16, y = var_5841_to_fp16)[name = string("aw_chunk_519_cast_fp16")]; + fp16 var_5843_to_fp16 = const()[name = string("op_5843_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_521_cast_fp16, y = var_5843_to_fp16)[name = string("aw_chunk_521_cast_fp16")]; + fp16 var_5845_to_fp16 = const()[name = string("op_5845_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_523_cast_fp16, y = var_5845_to_fp16)[name = string("aw_chunk_523_cast_fp16")]; + fp16 var_5847_to_fp16 = const()[name = string("op_5847_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_525_cast_fp16, y = var_5847_to_fp16)[name = string("aw_chunk_525_cast_fp16")]; + fp16 var_5849_to_fp16 = const()[name = string("op_5849_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_527_cast_fp16, y = var_5849_to_fp16)[name = string("aw_chunk_527_cast_fp16")]; + fp16 var_5851_to_fp16 = const()[name = string("op_5851_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_529_cast_fp16, y = var_5851_to_fp16)[name = string("aw_chunk_529_cast_fp16")]; + fp16 var_5853_to_fp16 = const()[name = string("op_5853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_531_cast_fp16, y = var_5853_to_fp16)[name = string("aw_chunk_531_cast_fp16")]; + fp16 var_5855_to_fp16 = const()[name = string("op_5855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_533_cast_fp16, y = var_5855_to_fp16)[name = string("aw_chunk_533_cast_fp16")]; + fp16 var_5857_to_fp16 = const()[name = string("op_5857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_535_cast_fp16, y = var_5857_to_fp16)[name = string("aw_chunk_535_cast_fp16")]; + fp16 var_5859_to_fp16 = const()[name = string("op_5859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_537_cast_fp16, y = var_5859_to_fp16)[name = string("aw_chunk_537_cast_fp16")]; + fp16 var_5861_to_fp16 = const()[name = string("op_5861_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_539_cast_fp16, y = var_5861_to_fp16)[name = string("aw_chunk_539_cast_fp16")]; + fp16 var_5863_to_fp16 = const()[name = string("op_5863_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_541_cast_fp16, y = var_5863_to_fp16)[name = string("aw_chunk_541_cast_fp16")]; + fp16 var_5865_to_fp16 = const()[name = string("op_5865_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_543_cast_fp16, y = var_5865_to_fp16)[name = string("aw_chunk_543_cast_fp16")]; + fp16 var_5867_to_fp16 = const()[name = string("op_5867_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_545_cast_fp16, y = var_5867_to_fp16)[name = string("aw_chunk_545_cast_fp16")]; + fp16 var_5869_to_fp16 = const()[name = string("op_5869_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_547_cast_fp16, y = var_5869_to_fp16)[name = string("aw_chunk_547_cast_fp16")]; + fp16 var_5871_to_fp16 = const()[name = string("op_5871_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_549_cast_fp16, y = var_5871_to_fp16)[name = string("aw_chunk_549_cast_fp16")]; + fp16 var_5873_to_fp16 = const()[name = string("op_5873_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_551_cast_fp16, y = var_5873_to_fp16)[name = string("aw_chunk_551_cast_fp16")]; + fp16 var_5875_to_fp16 = const()[name = string("op_5875_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_553_cast_fp16, y = var_5875_to_fp16)[name = string("aw_chunk_553_cast_fp16")]; + fp16 var_5877_to_fp16 = const()[name = string("op_5877_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_555_cast_fp16, y = var_5877_to_fp16)[name = string("aw_chunk_555_cast_fp16")]; + fp16 var_5879_to_fp16 = const()[name = string("op_5879_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_557_cast_fp16, y = var_5879_to_fp16)[name = string("aw_chunk_557_cast_fp16")]; + fp16 var_5881_to_fp16 = const()[name = string("op_5881_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_559_cast_fp16, y = var_5881_to_fp16)[name = string("aw_chunk_559_cast_fp16")]; + fp16 var_5883_to_fp16 = const()[name = string("op_5883_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_561_cast_fp16, y = var_5883_to_fp16)[name = string("aw_chunk_561_cast_fp16")]; + fp16 var_5885_to_fp16 = const()[name = string("op_5885_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_563_cast_fp16, y = var_5885_to_fp16)[name = string("aw_chunk_563_cast_fp16")]; + fp16 var_5887_to_fp16 = const()[name = string("op_5887_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_565_cast_fp16, y = var_5887_to_fp16)[name = string("aw_chunk_565_cast_fp16")]; + fp16 var_5889_to_fp16 = const()[name = string("op_5889_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_567_cast_fp16, y = var_5889_to_fp16)[name = string("aw_chunk_567_cast_fp16")]; + fp16 var_5891_to_fp16 = const()[name = string("op_5891_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_569_cast_fp16, y = var_5891_to_fp16)[name = string("aw_chunk_569_cast_fp16")]; + fp16 var_5893_to_fp16 = const()[name = string("op_5893_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_571_cast_fp16, y = var_5893_to_fp16)[name = string("aw_chunk_571_cast_fp16")]; + fp16 var_5895_to_fp16 = const()[name = string("op_5895_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_573_cast_fp16, y = var_5895_to_fp16)[name = string("aw_chunk_573_cast_fp16")]; + fp16 var_5897_to_fp16 = const()[name = string("op_5897_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_575_cast_fp16, y = var_5897_to_fp16)[name = string("aw_chunk_575_cast_fp16")]; + fp16 var_5899_to_fp16 = const()[name = string("op_5899_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_577_cast_fp16, y = var_5899_to_fp16)[name = string("aw_chunk_577_cast_fp16")]; + fp16 var_5901_to_fp16 = const()[name = string("op_5901_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_579_cast_fp16, y = var_5901_to_fp16)[name = string("aw_chunk_579_cast_fp16")]; + fp16 var_5903_to_fp16 = const()[name = string("op_5903_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_581_cast_fp16, y = var_5903_to_fp16)[name = string("aw_chunk_581_cast_fp16")]; + fp16 var_5905_to_fp16 = const()[name = string("op_5905_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_583_cast_fp16, y = var_5905_to_fp16)[name = string("aw_chunk_583_cast_fp16")]; + fp16 var_5907_to_fp16 = const()[name = string("op_5907_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_585_cast_fp16, y = var_5907_to_fp16)[name = string("aw_chunk_585_cast_fp16")]; + fp16 var_5909_to_fp16 = const()[name = string("op_5909_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_587_cast_fp16, y = var_5909_to_fp16)[name = string("aw_chunk_587_cast_fp16")]; + fp16 var_5911_to_fp16 = const()[name = string("op_5911_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_589_cast_fp16, y = var_5911_to_fp16)[name = string("aw_chunk_589_cast_fp16")]; + fp16 var_5913_to_fp16 = const()[name = string("op_5913_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_591_cast_fp16, y = var_5913_to_fp16)[name = string("aw_chunk_591_cast_fp16")]; + fp16 var_5915_to_fp16 = const()[name = string("op_5915_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_593_cast_fp16, y = var_5915_to_fp16)[name = string("aw_chunk_593_cast_fp16")]; + fp16 var_5917_to_fp16 = const()[name = string("op_5917_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_595_cast_fp16, y = var_5917_to_fp16)[name = string("aw_chunk_595_cast_fp16")]; + fp16 var_5919_to_fp16 = const()[name = string("op_5919_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_597_cast_fp16, y = var_5919_to_fp16)[name = string("aw_chunk_597_cast_fp16")]; + fp16 var_5921_to_fp16 = const()[name = string("op_5921_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_599_cast_fp16, y = var_5921_to_fp16)[name = string("aw_chunk_599_cast_fp16")]; + fp16 var_5923_to_fp16 = const()[name = string("op_5923_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_601_cast_fp16, y = var_5923_to_fp16)[name = string("aw_chunk_601_cast_fp16")]; + fp16 var_5925_to_fp16 = const()[name = string("op_5925_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_603_cast_fp16, y = var_5925_to_fp16)[name = string("aw_chunk_603_cast_fp16")]; + fp16 var_5927_to_fp16 = const()[name = string("op_5927_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_605_cast_fp16, y = var_5927_to_fp16)[name = string("aw_chunk_605_cast_fp16")]; + fp16 var_5929_to_fp16 = const()[name = string("op_5929_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_607_cast_fp16, y = var_5929_to_fp16)[name = string("aw_chunk_607_cast_fp16")]; + fp16 var_5931_to_fp16 = const()[name = string("op_5931_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_609_cast_fp16, y = var_5931_to_fp16)[name = string("aw_chunk_609_cast_fp16")]; + fp16 var_5933_to_fp16 = const()[name = string("op_5933_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_611_cast_fp16, y = var_5933_to_fp16)[name = string("aw_chunk_611_cast_fp16")]; + fp16 var_5935_to_fp16 = const()[name = string("op_5935_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_613_cast_fp16, y = var_5935_to_fp16)[name = string("aw_chunk_613_cast_fp16")]; + fp16 var_5937_to_fp16 = const()[name = string("op_5937_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_615_cast_fp16, y = var_5937_to_fp16)[name = string("aw_chunk_615_cast_fp16")]; + fp16 var_5939_to_fp16 = const()[name = string("op_5939_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_617_cast_fp16, y = var_5939_to_fp16)[name = string("aw_chunk_617_cast_fp16")]; + fp16 var_5941_to_fp16 = const()[name = string("op_5941_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_619_cast_fp16, y = var_5941_to_fp16)[name = string("aw_chunk_619_cast_fp16")]; + fp16 var_5943_to_fp16 = const()[name = string("op_5943_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_621_cast_fp16, y = var_5943_to_fp16)[name = string("aw_chunk_621_cast_fp16")]; + fp16 var_5945_to_fp16 = const()[name = string("op_5945_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_623_cast_fp16, y = var_5945_to_fp16)[name = string("aw_chunk_623_cast_fp16")]; + fp16 var_5947_to_fp16 = const()[name = string("op_5947_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_625_cast_fp16, y = var_5947_to_fp16)[name = string("aw_chunk_625_cast_fp16")]; + fp16 var_5949_to_fp16 = const()[name = string("op_5949_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_627_cast_fp16, y = var_5949_to_fp16)[name = string("aw_chunk_627_cast_fp16")]; + fp16 var_5951_to_fp16 = const()[name = string("op_5951_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_629_cast_fp16, y = var_5951_to_fp16)[name = string("aw_chunk_629_cast_fp16")]; + fp16 var_5953_to_fp16 = const()[name = string("op_5953_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_631_cast_fp16, y = var_5953_to_fp16)[name = string("aw_chunk_631_cast_fp16")]; + fp16 var_5955_to_fp16 = const()[name = string("op_5955_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_633_cast_fp16, y = var_5955_to_fp16)[name = string("aw_chunk_633_cast_fp16")]; + fp16 var_5957_to_fp16 = const()[name = string("op_5957_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_635_cast_fp16, y = var_5957_to_fp16)[name = string("aw_chunk_635_cast_fp16")]; + fp16 var_5959_to_fp16 = const()[name = string("op_5959_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_637_cast_fp16, y = var_5959_to_fp16)[name = string("aw_chunk_637_cast_fp16")]; + fp16 var_5961_to_fp16 = const()[name = string("op_5961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_639_cast_fp16, y = var_5961_to_fp16)[name = string("aw_chunk_639_cast_fp16")]; + tensor var_5963_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_481_cast_fp16)[name = string("op_5963_cast_fp16")]; + tensor var_5964_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_483_cast_fp16)[name = string("op_5964_cast_fp16")]; + tensor var_5965_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_485_cast_fp16)[name = string("op_5965_cast_fp16")]; + tensor var_5966_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_487_cast_fp16)[name = string("op_5966_cast_fp16")]; + tensor var_5967_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_489_cast_fp16)[name = string("op_5967_cast_fp16")]; + tensor var_5968_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_491_cast_fp16)[name = string("op_5968_cast_fp16")]; + tensor var_5969_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_493_cast_fp16)[name = string("op_5969_cast_fp16")]; + tensor var_5970_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_495_cast_fp16)[name = string("op_5970_cast_fp16")]; + tensor var_5971_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_497_cast_fp16)[name = string("op_5971_cast_fp16")]; + tensor var_5972_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_499_cast_fp16)[name = string("op_5972_cast_fp16")]; + tensor var_5973_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_501_cast_fp16)[name = string("op_5973_cast_fp16")]; + tensor var_5974_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_503_cast_fp16)[name = string("op_5974_cast_fp16")]; + tensor var_5975_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_505_cast_fp16)[name = string("op_5975_cast_fp16")]; + tensor var_5976_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_507_cast_fp16)[name = string("op_5976_cast_fp16")]; + tensor var_5977_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_509_cast_fp16)[name = string("op_5977_cast_fp16")]; + tensor var_5978_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_511_cast_fp16)[name = string("op_5978_cast_fp16")]; + tensor var_5979_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_513_cast_fp16)[name = string("op_5979_cast_fp16")]; + tensor var_5980_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_515_cast_fp16)[name = string("op_5980_cast_fp16")]; + tensor var_5981_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_517_cast_fp16)[name = string("op_5981_cast_fp16")]; + tensor var_5982_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_519_cast_fp16)[name = string("op_5982_cast_fp16")]; + tensor var_5983_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_521_cast_fp16)[name = string("op_5983_cast_fp16")]; + tensor var_5984_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_523_cast_fp16)[name = string("op_5984_cast_fp16")]; + tensor var_5985_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_525_cast_fp16)[name = string("op_5985_cast_fp16")]; + tensor var_5986_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_527_cast_fp16)[name = string("op_5986_cast_fp16")]; + tensor var_5987_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_529_cast_fp16)[name = string("op_5987_cast_fp16")]; + tensor var_5988_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_531_cast_fp16)[name = string("op_5988_cast_fp16")]; + tensor var_5989_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_533_cast_fp16)[name = string("op_5989_cast_fp16")]; + tensor var_5990_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_535_cast_fp16)[name = string("op_5990_cast_fp16")]; + tensor var_5991_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_537_cast_fp16)[name = string("op_5991_cast_fp16")]; + tensor var_5992_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_539_cast_fp16)[name = string("op_5992_cast_fp16")]; + tensor var_5993_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_541_cast_fp16)[name = string("op_5993_cast_fp16")]; + tensor var_5994_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_543_cast_fp16)[name = string("op_5994_cast_fp16")]; + tensor var_5995_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_545_cast_fp16)[name = string("op_5995_cast_fp16")]; + tensor var_5996_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_547_cast_fp16)[name = string("op_5996_cast_fp16")]; + tensor var_5997_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_549_cast_fp16)[name = string("op_5997_cast_fp16")]; + tensor var_5998_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_551_cast_fp16)[name = string("op_5998_cast_fp16")]; + tensor var_5999_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_553_cast_fp16)[name = string("op_5999_cast_fp16")]; + tensor var_6000_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_555_cast_fp16)[name = string("op_6000_cast_fp16")]; + tensor var_6001_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_557_cast_fp16)[name = string("op_6001_cast_fp16")]; + tensor var_6002_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_559_cast_fp16)[name = string("op_6002_cast_fp16")]; + tensor var_6003_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_561_cast_fp16)[name = string("op_6003_cast_fp16")]; + tensor var_6004_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_563_cast_fp16)[name = string("op_6004_cast_fp16")]; + tensor var_6005_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_565_cast_fp16)[name = string("op_6005_cast_fp16")]; + tensor var_6006_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_567_cast_fp16)[name = string("op_6006_cast_fp16")]; + tensor var_6007_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_569_cast_fp16)[name = string("op_6007_cast_fp16")]; + tensor var_6008_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_571_cast_fp16)[name = string("op_6008_cast_fp16")]; + tensor var_6009_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_573_cast_fp16)[name = string("op_6009_cast_fp16")]; + tensor var_6010_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_575_cast_fp16)[name = string("op_6010_cast_fp16")]; + tensor var_6011_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_577_cast_fp16)[name = string("op_6011_cast_fp16")]; + tensor var_6012_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_579_cast_fp16)[name = string("op_6012_cast_fp16")]; + tensor var_6013_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_581_cast_fp16)[name = string("op_6013_cast_fp16")]; + tensor var_6014_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_583_cast_fp16)[name = string("op_6014_cast_fp16")]; + tensor var_6015_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_585_cast_fp16)[name = string("op_6015_cast_fp16")]; + tensor var_6016_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_587_cast_fp16)[name = string("op_6016_cast_fp16")]; + tensor var_6017_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_589_cast_fp16)[name = string("op_6017_cast_fp16")]; + tensor var_6018_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_591_cast_fp16)[name = string("op_6018_cast_fp16")]; + tensor var_6019_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_593_cast_fp16)[name = string("op_6019_cast_fp16")]; + tensor var_6020_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_595_cast_fp16)[name = string("op_6020_cast_fp16")]; + tensor var_6021_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_597_cast_fp16)[name = string("op_6021_cast_fp16")]; + tensor var_6022_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_599_cast_fp16)[name = string("op_6022_cast_fp16")]; + tensor var_6023_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_601_cast_fp16)[name = string("op_6023_cast_fp16")]; + tensor var_6024_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_603_cast_fp16)[name = string("op_6024_cast_fp16")]; + tensor var_6025_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_605_cast_fp16)[name = string("op_6025_cast_fp16")]; + tensor var_6026_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_607_cast_fp16)[name = string("op_6026_cast_fp16")]; + tensor var_6027_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_609_cast_fp16)[name = string("op_6027_cast_fp16")]; + tensor var_6028_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_611_cast_fp16)[name = string("op_6028_cast_fp16")]; + tensor var_6029_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_613_cast_fp16)[name = string("op_6029_cast_fp16")]; + tensor var_6030_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_615_cast_fp16)[name = string("op_6030_cast_fp16")]; + tensor var_6031_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_617_cast_fp16)[name = string("op_6031_cast_fp16")]; + tensor var_6032_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_619_cast_fp16)[name = string("op_6032_cast_fp16")]; + tensor var_6033_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_621_cast_fp16)[name = string("op_6033_cast_fp16")]; + tensor var_6034_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_623_cast_fp16)[name = string("op_6034_cast_fp16")]; + tensor var_6035_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_625_cast_fp16)[name = string("op_6035_cast_fp16")]; + tensor var_6036_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_627_cast_fp16)[name = string("op_6036_cast_fp16")]; + tensor var_6037_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_629_cast_fp16)[name = string("op_6037_cast_fp16")]; + tensor var_6038_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_631_cast_fp16)[name = string("op_6038_cast_fp16")]; + tensor var_6039_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_633_cast_fp16)[name = string("op_6039_cast_fp16")]; + tensor var_6040_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_635_cast_fp16)[name = string("op_6040_cast_fp16")]; + tensor var_6041_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_637_cast_fp16)[name = string("op_6041_cast_fp16")]; + tensor var_6042_cast_fp16 = softmax(axis = var_4788, x = aw_chunk_639_cast_fp16)[name = string("op_6042_cast_fp16")]; + string var_6044_equation_0 = const()[name = string("op_6044_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6044_cast_fp16 = einsum(equation = var_6044_equation_0, values = (var_5564_cast_fp16, var_5963_cast_fp16))[name = string("op_6044_cast_fp16")]; + string var_6046_equation_0 = const()[name = string("op_6046_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6046_cast_fp16 = einsum(equation = var_6046_equation_0, values = (var_5564_cast_fp16, var_5964_cast_fp16))[name = string("op_6046_cast_fp16")]; + string var_6048_equation_0 = const()[name = string("op_6048_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6048_cast_fp16 = einsum(equation = var_6048_equation_0, values = (var_5564_cast_fp16, var_5965_cast_fp16))[name = string("op_6048_cast_fp16")]; + string var_6050_equation_0 = const()[name = string("op_6050_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6050_cast_fp16 = einsum(equation = var_6050_equation_0, values = (var_5564_cast_fp16, var_5966_cast_fp16))[name = string("op_6050_cast_fp16")]; + string var_6052_equation_0 = const()[name = string("op_6052_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6052_cast_fp16 = einsum(equation = var_6052_equation_0, values = (var_5568_cast_fp16, var_5967_cast_fp16))[name = string("op_6052_cast_fp16")]; + string var_6054_equation_0 = const()[name = string("op_6054_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6054_cast_fp16 = einsum(equation = var_6054_equation_0, values = (var_5568_cast_fp16, var_5968_cast_fp16))[name = string("op_6054_cast_fp16")]; + string var_6056_equation_0 = const()[name = string("op_6056_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6056_cast_fp16 = einsum(equation = var_6056_equation_0, values = (var_5568_cast_fp16, var_5969_cast_fp16))[name = string("op_6056_cast_fp16")]; + string var_6058_equation_0 = const()[name = string("op_6058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6058_cast_fp16 = einsum(equation = var_6058_equation_0, values = (var_5568_cast_fp16, var_5970_cast_fp16))[name = string("op_6058_cast_fp16")]; + string var_6060_equation_0 = const()[name = string("op_6060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6060_cast_fp16 = einsum(equation = var_6060_equation_0, values = (var_5572_cast_fp16, var_5971_cast_fp16))[name = string("op_6060_cast_fp16")]; + string var_6062_equation_0 = const()[name = string("op_6062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6062_cast_fp16 = einsum(equation = var_6062_equation_0, values = (var_5572_cast_fp16, var_5972_cast_fp16))[name = string("op_6062_cast_fp16")]; + string var_6064_equation_0 = const()[name = string("op_6064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6064_cast_fp16 = einsum(equation = var_6064_equation_0, values = (var_5572_cast_fp16, var_5973_cast_fp16))[name = string("op_6064_cast_fp16")]; + string var_6066_equation_0 = const()[name = string("op_6066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6066_cast_fp16 = einsum(equation = var_6066_equation_0, values = (var_5572_cast_fp16, var_5974_cast_fp16))[name = string("op_6066_cast_fp16")]; + string var_6068_equation_0 = const()[name = string("op_6068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6068_cast_fp16 = einsum(equation = var_6068_equation_0, values = (var_5576_cast_fp16, var_5975_cast_fp16))[name = string("op_6068_cast_fp16")]; + string var_6070_equation_0 = const()[name = string("op_6070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6070_cast_fp16 = einsum(equation = var_6070_equation_0, values = (var_5576_cast_fp16, var_5976_cast_fp16))[name = string("op_6070_cast_fp16")]; + string var_6072_equation_0 = const()[name = string("op_6072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6072_cast_fp16 = einsum(equation = var_6072_equation_0, values = (var_5576_cast_fp16, var_5977_cast_fp16))[name = string("op_6072_cast_fp16")]; + string var_6074_equation_0 = const()[name = string("op_6074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6074_cast_fp16 = einsum(equation = var_6074_equation_0, values = (var_5576_cast_fp16, var_5978_cast_fp16))[name = string("op_6074_cast_fp16")]; + string var_6076_equation_0 = const()[name = string("op_6076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6076_cast_fp16 = einsum(equation = var_6076_equation_0, values = (var_5580_cast_fp16, var_5979_cast_fp16))[name = string("op_6076_cast_fp16")]; + string var_6078_equation_0 = const()[name = string("op_6078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6078_cast_fp16 = einsum(equation = var_6078_equation_0, values = (var_5580_cast_fp16, var_5980_cast_fp16))[name = string("op_6078_cast_fp16")]; + string var_6080_equation_0 = const()[name = string("op_6080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6080_cast_fp16 = einsum(equation = var_6080_equation_0, values = (var_5580_cast_fp16, var_5981_cast_fp16))[name = string("op_6080_cast_fp16")]; + string var_6082_equation_0 = const()[name = string("op_6082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6082_cast_fp16 = einsum(equation = var_6082_equation_0, values = (var_5580_cast_fp16, var_5982_cast_fp16))[name = string("op_6082_cast_fp16")]; + string var_6084_equation_0 = const()[name = string("op_6084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6084_cast_fp16 = einsum(equation = var_6084_equation_0, values = (var_5584_cast_fp16, var_5983_cast_fp16))[name = string("op_6084_cast_fp16")]; + string var_6086_equation_0 = const()[name = string("op_6086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6086_cast_fp16 = einsum(equation = var_6086_equation_0, values = (var_5584_cast_fp16, var_5984_cast_fp16))[name = string("op_6086_cast_fp16")]; + string var_6088_equation_0 = const()[name = string("op_6088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6088_cast_fp16 = einsum(equation = var_6088_equation_0, values = (var_5584_cast_fp16, var_5985_cast_fp16))[name = string("op_6088_cast_fp16")]; + string var_6090_equation_0 = const()[name = string("op_6090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6090_cast_fp16 = einsum(equation = var_6090_equation_0, values = (var_5584_cast_fp16, var_5986_cast_fp16))[name = string("op_6090_cast_fp16")]; + string var_6092_equation_0 = const()[name = string("op_6092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6092_cast_fp16 = einsum(equation = var_6092_equation_0, values = (var_5588_cast_fp16, var_5987_cast_fp16))[name = string("op_6092_cast_fp16")]; + string var_6094_equation_0 = const()[name = string("op_6094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6094_cast_fp16 = einsum(equation = var_6094_equation_0, values = (var_5588_cast_fp16, var_5988_cast_fp16))[name = string("op_6094_cast_fp16")]; + string var_6096_equation_0 = const()[name = string("op_6096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6096_cast_fp16 = einsum(equation = var_6096_equation_0, values = (var_5588_cast_fp16, var_5989_cast_fp16))[name = string("op_6096_cast_fp16")]; + string var_6098_equation_0 = const()[name = string("op_6098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6098_cast_fp16 = einsum(equation = var_6098_equation_0, values = (var_5588_cast_fp16, var_5990_cast_fp16))[name = string("op_6098_cast_fp16")]; + string var_6100_equation_0 = const()[name = string("op_6100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6100_cast_fp16 = einsum(equation = var_6100_equation_0, values = (var_5592_cast_fp16, var_5991_cast_fp16))[name = string("op_6100_cast_fp16")]; + string var_6102_equation_0 = const()[name = string("op_6102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6102_cast_fp16 = einsum(equation = var_6102_equation_0, values = (var_5592_cast_fp16, var_5992_cast_fp16))[name = string("op_6102_cast_fp16")]; + string var_6104_equation_0 = const()[name = string("op_6104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6104_cast_fp16 = einsum(equation = var_6104_equation_0, values = (var_5592_cast_fp16, var_5993_cast_fp16))[name = string("op_6104_cast_fp16")]; + string var_6106_equation_0 = const()[name = string("op_6106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6106_cast_fp16 = einsum(equation = var_6106_equation_0, values = (var_5592_cast_fp16, var_5994_cast_fp16))[name = string("op_6106_cast_fp16")]; + string var_6108_equation_0 = const()[name = string("op_6108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6108_cast_fp16 = einsum(equation = var_6108_equation_0, values = (var_5596_cast_fp16, var_5995_cast_fp16))[name = string("op_6108_cast_fp16")]; + string var_6110_equation_0 = const()[name = string("op_6110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6110_cast_fp16 = einsum(equation = var_6110_equation_0, values = (var_5596_cast_fp16, var_5996_cast_fp16))[name = string("op_6110_cast_fp16")]; + string var_6112_equation_0 = const()[name = string("op_6112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6112_cast_fp16 = einsum(equation = var_6112_equation_0, values = (var_5596_cast_fp16, var_5997_cast_fp16))[name = string("op_6112_cast_fp16")]; + string var_6114_equation_0 = const()[name = string("op_6114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6114_cast_fp16 = einsum(equation = var_6114_equation_0, values = (var_5596_cast_fp16, var_5998_cast_fp16))[name = string("op_6114_cast_fp16")]; + string var_6116_equation_0 = const()[name = string("op_6116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6116_cast_fp16 = einsum(equation = var_6116_equation_0, values = (var_5600_cast_fp16, var_5999_cast_fp16))[name = string("op_6116_cast_fp16")]; + string var_6118_equation_0 = const()[name = string("op_6118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6118_cast_fp16 = einsum(equation = var_6118_equation_0, values = (var_5600_cast_fp16, var_6000_cast_fp16))[name = string("op_6118_cast_fp16")]; + string var_6120_equation_0 = const()[name = string("op_6120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6120_cast_fp16 = einsum(equation = var_6120_equation_0, values = (var_5600_cast_fp16, var_6001_cast_fp16))[name = string("op_6120_cast_fp16")]; + string var_6122_equation_0 = const()[name = string("op_6122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6122_cast_fp16 = einsum(equation = var_6122_equation_0, values = (var_5600_cast_fp16, var_6002_cast_fp16))[name = string("op_6122_cast_fp16")]; + string var_6124_equation_0 = const()[name = string("op_6124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6124_cast_fp16 = einsum(equation = var_6124_equation_0, values = (var_5604_cast_fp16, var_6003_cast_fp16))[name = string("op_6124_cast_fp16")]; + string var_6126_equation_0 = const()[name = string("op_6126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6126_cast_fp16 = einsum(equation = var_6126_equation_0, values = (var_5604_cast_fp16, var_6004_cast_fp16))[name = string("op_6126_cast_fp16")]; + string var_6128_equation_0 = const()[name = string("op_6128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6128_cast_fp16 = einsum(equation = var_6128_equation_0, values = (var_5604_cast_fp16, var_6005_cast_fp16))[name = string("op_6128_cast_fp16")]; + string var_6130_equation_0 = const()[name = string("op_6130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6130_cast_fp16 = einsum(equation = var_6130_equation_0, values = (var_5604_cast_fp16, var_6006_cast_fp16))[name = string("op_6130_cast_fp16")]; + string var_6132_equation_0 = const()[name = string("op_6132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6132_cast_fp16 = einsum(equation = var_6132_equation_0, values = (var_5608_cast_fp16, var_6007_cast_fp16))[name = string("op_6132_cast_fp16")]; + string var_6134_equation_0 = const()[name = string("op_6134_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6134_cast_fp16 = einsum(equation = var_6134_equation_0, values = (var_5608_cast_fp16, var_6008_cast_fp16))[name = string("op_6134_cast_fp16")]; + string var_6136_equation_0 = const()[name = string("op_6136_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6136_cast_fp16 = einsum(equation = var_6136_equation_0, values = (var_5608_cast_fp16, var_6009_cast_fp16))[name = string("op_6136_cast_fp16")]; + string var_6138_equation_0 = const()[name = string("op_6138_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6138_cast_fp16 = einsum(equation = var_6138_equation_0, values = (var_5608_cast_fp16, var_6010_cast_fp16))[name = string("op_6138_cast_fp16")]; + string var_6140_equation_0 = const()[name = string("op_6140_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6140_cast_fp16 = einsum(equation = var_6140_equation_0, values = (var_5612_cast_fp16, var_6011_cast_fp16))[name = string("op_6140_cast_fp16")]; + string var_6142_equation_0 = const()[name = string("op_6142_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6142_cast_fp16 = einsum(equation = var_6142_equation_0, values = (var_5612_cast_fp16, var_6012_cast_fp16))[name = string("op_6142_cast_fp16")]; + string var_6144_equation_0 = const()[name = string("op_6144_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6144_cast_fp16 = einsum(equation = var_6144_equation_0, values = (var_5612_cast_fp16, var_6013_cast_fp16))[name = string("op_6144_cast_fp16")]; + string var_6146_equation_0 = const()[name = string("op_6146_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6146_cast_fp16 = einsum(equation = var_6146_equation_0, values = (var_5612_cast_fp16, var_6014_cast_fp16))[name = string("op_6146_cast_fp16")]; + string var_6148_equation_0 = const()[name = string("op_6148_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6148_cast_fp16 = einsum(equation = var_6148_equation_0, values = (var_5616_cast_fp16, var_6015_cast_fp16))[name = string("op_6148_cast_fp16")]; + string var_6150_equation_0 = const()[name = string("op_6150_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6150_cast_fp16 = einsum(equation = var_6150_equation_0, values = (var_5616_cast_fp16, var_6016_cast_fp16))[name = string("op_6150_cast_fp16")]; + string var_6152_equation_0 = const()[name = string("op_6152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6152_cast_fp16 = einsum(equation = var_6152_equation_0, values = (var_5616_cast_fp16, var_6017_cast_fp16))[name = string("op_6152_cast_fp16")]; + string var_6154_equation_0 = const()[name = string("op_6154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6154_cast_fp16 = einsum(equation = var_6154_equation_0, values = (var_5616_cast_fp16, var_6018_cast_fp16))[name = string("op_6154_cast_fp16")]; + string var_6156_equation_0 = const()[name = string("op_6156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6156_cast_fp16 = einsum(equation = var_6156_equation_0, values = (var_5620_cast_fp16, var_6019_cast_fp16))[name = string("op_6156_cast_fp16")]; + string var_6158_equation_0 = const()[name = string("op_6158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6158_cast_fp16 = einsum(equation = var_6158_equation_0, values = (var_5620_cast_fp16, var_6020_cast_fp16))[name = string("op_6158_cast_fp16")]; + string var_6160_equation_0 = const()[name = string("op_6160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6160_cast_fp16 = einsum(equation = var_6160_equation_0, values = (var_5620_cast_fp16, var_6021_cast_fp16))[name = string("op_6160_cast_fp16")]; + string var_6162_equation_0 = const()[name = string("op_6162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6162_cast_fp16 = einsum(equation = var_6162_equation_0, values = (var_5620_cast_fp16, var_6022_cast_fp16))[name = string("op_6162_cast_fp16")]; + string var_6164_equation_0 = const()[name = string("op_6164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6164_cast_fp16 = einsum(equation = var_6164_equation_0, values = (var_5624_cast_fp16, var_6023_cast_fp16))[name = string("op_6164_cast_fp16")]; + string var_6166_equation_0 = const()[name = string("op_6166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6166_cast_fp16 = einsum(equation = var_6166_equation_0, values = (var_5624_cast_fp16, var_6024_cast_fp16))[name = string("op_6166_cast_fp16")]; + string var_6168_equation_0 = const()[name = string("op_6168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6168_cast_fp16 = einsum(equation = var_6168_equation_0, values = (var_5624_cast_fp16, var_6025_cast_fp16))[name = string("op_6168_cast_fp16")]; + string var_6170_equation_0 = const()[name = string("op_6170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6170_cast_fp16 = einsum(equation = var_6170_equation_0, values = (var_5624_cast_fp16, var_6026_cast_fp16))[name = string("op_6170_cast_fp16")]; + string var_6172_equation_0 = const()[name = string("op_6172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6172_cast_fp16 = einsum(equation = var_6172_equation_0, values = (var_5628_cast_fp16, var_6027_cast_fp16))[name = string("op_6172_cast_fp16")]; + string var_6174_equation_0 = const()[name = string("op_6174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6174_cast_fp16 = einsum(equation = var_6174_equation_0, values = (var_5628_cast_fp16, var_6028_cast_fp16))[name = string("op_6174_cast_fp16")]; + string var_6176_equation_0 = const()[name = string("op_6176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6176_cast_fp16 = einsum(equation = var_6176_equation_0, values = (var_5628_cast_fp16, var_6029_cast_fp16))[name = string("op_6176_cast_fp16")]; + string var_6178_equation_0 = const()[name = string("op_6178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6178_cast_fp16 = einsum(equation = var_6178_equation_0, values = (var_5628_cast_fp16, var_6030_cast_fp16))[name = string("op_6178_cast_fp16")]; + string var_6180_equation_0 = const()[name = string("op_6180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6180_cast_fp16 = einsum(equation = var_6180_equation_0, values = (var_5632_cast_fp16, var_6031_cast_fp16))[name = string("op_6180_cast_fp16")]; + string var_6182_equation_0 = const()[name = string("op_6182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6182_cast_fp16 = einsum(equation = var_6182_equation_0, values = (var_5632_cast_fp16, var_6032_cast_fp16))[name = string("op_6182_cast_fp16")]; + string var_6184_equation_0 = const()[name = string("op_6184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6184_cast_fp16 = einsum(equation = var_6184_equation_0, values = (var_5632_cast_fp16, var_6033_cast_fp16))[name = string("op_6184_cast_fp16")]; + string var_6186_equation_0 = const()[name = string("op_6186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6186_cast_fp16 = einsum(equation = var_6186_equation_0, values = (var_5632_cast_fp16, var_6034_cast_fp16))[name = string("op_6186_cast_fp16")]; + string var_6188_equation_0 = const()[name = string("op_6188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6188_cast_fp16 = einsum(equation = var_6188_equation_0, values = (var_5636_cast_fp16, var_6035_cast_fp16))[name = string("op_6188_cast_fp16")]; + string var_6190_equation_0 = const()[name = string("op_6190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6190_cast_fp16 = einsum(equation = var_6190_equation_0, values = (var_5636_cast_fp16, var_6036_cast_fp16))[name = string("op_6190_cast_fp16")]; + string var_6192_equation_0 = const()[name = string("op_6192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6192_cast_fp16 = einsum(equation = var_6192_equation_0, values = (var_5636_cast_fp16, var_6037_cast_fp16))[name = string("op_6192_cast_fp16")]; + string var_6194_equation_0 = const()[name = string("op_6194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6194_cast_fp16 = einsum(equation = var_6194_equation_0, values = (var_5636_cast_fp16, var_6038_cast_fp16))[name = string("op_6194_cast_fp16")]; + string var_6196_equation_0 = const()[name = string("op_6196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6196_cast_fp16 = einsum(equation = var_6196_equation_0, values = (var_5640_cast_fp16, var_6039_cast_fp16))[name = string("op_6196_cast_fp16")]; + string var_6198_equation_0 = const()[name = string("op_6198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6198_cast_fp16 = einsum(equation = var_6198_equation_0, values = (var_5640_cast_fp16, var_6040_cast_fp16))[name = string("op_6198_cast_fp16")]; + string var_6200_equation_0 = const()[name = string("op_6200_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6200_cast_fp16 = einsum(equation = var_6200_equation_0, values = (var_5640_cast_fp16, var_6041_cast_fp16))[name = string("op_6200_cast_fp16")]; + string var_6202_equation_0 = const()[name = string("op_6202_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6202_cast_fp16 = einsum(equation = var_6202_equation_0, values = (var_5640_cast_fp16, var_6042_cast_fp16))[name = string("op_6202_cast_fp16")]; + bool var_6204_interleave_0 = const()[name = string("op_6204_interleave_0"), val = bool(false)]; + tensor var_6204_cast_fp16 = concat(axis = var_4763, interleave = var_6204_interleave_0, values = (var_6044_cast_fp16, var_6046_cast_fp16, var_6048_cast_fp16, var_6050_cast_fp16))[name = string("op_6204_cast_fp16")]; + bool var_6206_interleave_0 = const()[name = string("op_6206_interleave_0"), val = bool(false)]; + tensor var_6206_cast_fp16 = concat(axis = var_4763, interleave = var_6206_interleave_0, values = (var_6052_cast_fp16, var_6054_cast_fp16, var_6056_cast_fp16, var_6058_cast_fp16))[name = string("op_6206_cast_fp16")]; + bool var_6208_interleave_0 = const()[name = string("op_6208_interleave_0"), val = bool(false)]; + tensor var_6208_cast_fp16 = concat(axis = var_4763, interleave = var_6208_interleave_0, values = (var_6060_cast_fp16, var_6062_cast_fp16, var_6064_cast_fp16, var_6066_cast_fp16))[name = string("op_6208_cast_fp16")]; + bool var_6210_interleave_0 = const()[name = string("op_6210_interleave_0"), val = bool(false)]; + tensor var_6210_cast_fp16 = concat(axis = var_4763, interleave = var_6210_interleave_0, values = (var_6068_cast_fp16, var_6070_cast_fp16, var_6072_cast_fp16, var_6074_cast_fp16))[name = string("op_6210_cast_fp16")]; + bool var_6212_interleave_0 = const()[name = string("op_6212_interleave_0"), val = bool(false)]; + tensor var_6212_cast_fp16 = concat(axis = var_4763, interleave = var_6212_interleave_0, values = (var_6076_cast_fp16, var_6078_cast_fp16, var_6080_cast_fp16, var_6082_cast_fp16))[name = string("op_6212_cast_fp16")]; + bool var_6214_interleave_0 = const()[name = string("op_6214_interleave_0"), val = bool(false)]; + tensor var_6214_cast_fp16 = concat(axis = var_4763, interleave = var_6214_interleave_0, values = (var_6084_cast_fp16, var_6086_cast_fp16, var_6088_cast_fp16, var_6090_cast_fp16))[name = string("op_6214_cast_fp16")]; + bool var_6216_interleave_0 = const()[name = string("op_6216_interleave_0"), val = bool(false)]; + tensor var_6216_cast_fp16 = concat(axis = var_4763, interleave = var_6216_interleave_0, values = (var_6092_cast_fp16, var_6094_cast_fp16, var_6096_cast_fp16, var_6098_cast_fp16))[name = string("op_6216_cast_fp16")]; + bool var_6218_interleave_0 = const()[name = string("op_6218_interleave_0"), val = bool(false)]; + tensor var_6218_cast_fp16 = concat(axis = var_4763, interleave = var_6218_interleave_0, values = (var_6100_cast_fp16, var_6102_cast_fp16, var_6104_cast_fp16, var_6106_cast_fp16))[name = string("op_6218_cast_fp16")]; + bool var_6220_interleave_0 = const()[name = string("op_6220_interleave_0"), val = bool(false)]; + tensor var_6220_cast_fp16 = concat(axis = var_4763, interleave = var_6220_interleave_0, values = (var_6108_cast_fp16, var_6110_cast_fp16, var_6112_cast_fp16, var_6114_cast_fp16))[name = string("op_6220_cast_fp16")]; + bool var_6222_interleave_0 = const()[name = string("op_6222_interleave_0"), val = bool(false)]; + tensor var_6222_cast_fp16 = concat(axis = var_4763, interleave = var_6222_interleave_0, values = (var_6116_cast_fp16, var_6118_cast_fp16, var_6120_cast_fp16, var_6122_cast_fp16))[name = string("op_6222_cast_fp16")]; + bool var_6224_interleave_0 = const()[name = string("op_6224_interleave_0"), val = bool(false)]; + tensor var_6224_cast_fp16 = concat(axis = var_4763, interleave = var_6224_interleave_0, values = (var_6124_cast_fp16, var_6126_cast_fp16, var_6128_cast_fp16, var_6130_cast_fp16))[name = string("op_6224_cast_fp16")]; + bool var_6226_interleave_0 = const()[name = string("op_6226_interleave_0"), val = bool(false)]; + tensor var_6226_cast_fp16 = concat(axis = var_4763, interleave = var_6226_interleave_0, values = (var_6132_cast_fp16, var_6134_cast_fp16, var_6136_cast_fp16, var_6138_cast_fp16))[name = string("op_6226_cast_fp16")]; + bool var_6228_interleave_0 = const()[name = string("op_6228_interleave_0"), val = bool(false)]; + tensor var_6228_cast_fp16 = concat(axis = var_4763, interleave = var_6228_interleave_0, values = (var_6140_cast_fp16, var_6142_cast_fp16, var_6144_cast_fp16, var_6146_cast_fp16))[name = string("op_6228_cast_fp16")]; + bool var_6230_interleave_0 = const()[name = string("op_6230_interleave_0"), val = bool(false)]; + tensor var_6230_cast_fp16 = concat(axis = var_4763, interleave = var_6230_interleave_0, values = (var_6148_cast_fp16, var_6150_cast_fp16, var_6152_cast_fp16, var_6154_cast_fp16))[name = string("op_6230_cast_fp16")]; + bool var_6232_interleave_0 = const()[name = string("op_6232_interleave_0"), val = bool(false)]; + tensor var_6232_cast_fp16 = concat(axis = var_4763, interleave = var_6232_interleave_0, values = (var_6156_cast_fp16, var_6158_cast_fp16, var_6160_cast_fp16, var_6162_cast_fp16))[name = string("op_6232_cast_fp16")]; + bool var_6234_interleave_0 = const()[name = string("op_6234_interleave_0"), val = bool(false)]; + tensor var_6234_cast_fp16 = concat(axis = var_4763, interleave = var_6234_interleave_0, values = (var_6164_cast_fp16, var_6166_cast_fp16, var_6168_cast_fp16, var_6170_cast_fp16))[name = string("op_6234_cast_fp16")]; + bool var_6236_interleave_0 = const()[name = string("op_6236_interleave_0"), val = bool(false)]; + tensor var_6236_cast_fp16 = concat(axis = var_4763, interleave = var_6236_interleave_0, values = (var_6172_cast_fp16, var_6174_cast_fp16, var_6176_cast_fp16, var_6178_cast_fp16))[name = string("op_6236_cast_fp16")]; + bool var_6238_interleave_0 = const()[name = string("op_6238_interleave_0"), val = bool(false)]; + tensor var_6238_cast_fp16 = concat(axis = var_4763, interleave = var_6238_interleave_0, values = (var_6180_cast_fp16, var_6182_cast_fp16, var_6184_cast_fp16, var_6186_cast_fp16))[name = string("op_6238_cast_fp16")]; + bool var_6240_interleave_0 = const()[name = string("op_6240_interleave_0"), val = bool(false)]; + tensor var_6240_cast_fp16 = concat(axis = var_4763, interleave = var_6240_interleave_0, values = (var_6188_cast_fp16, var_6190_cast_fp16, var_6192_cast_fp16, var_6194_cast_fp16))[name = string("op_6240_cast_fp16")]; + bool var_6242_interleave_0 = const()[name = string("op_6242_interleave_0"), val = bool(false)]; + tensor var_6242_cast_fp16 = concat(axis = var_4763, interleave = var_6242_interleave_0, values = (var_6196_cast_fp16, var_6198_cast_fp16, var_6200_cast_fp16, var_6202_cast_fp16))[name = string("op_6242_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_4788, interleave = input_25_interleave_0, values = (var_6204_cast_fp16, var_6206_cast_fp16, var_6208_cast_fp16, var_6210_cast_fp16, var_6212_cast_fp16, var_6214_cast_fp16, var_6216_cast_fp16, var_6218_cast_fp16, var_6220_cast_fp16, var_6222_cast_fp16, var_6224_cast_fp16, var_6226_cast_fp16, var_6228_cast_fp16, var_6230_cast_fp16, var_6232_cast_fp16, var_6234_cast_fp16, var_6236_cast_fp16, var_6238_cast_fp16, var_6240_cast_fp16, var_6242_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142565120)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145841984)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_6261_to_fp16 = const()[name = string("op_6261_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_6261_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145844608)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145847232)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145849856)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158957120)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158967424)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172074688)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_6290 = const()[name = string("op_6290"), val = int32(3)]; + int32 var_6315 = const()[name = string("op_6315"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_6332_to_fp16 = const()[name = string("op_6332_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_6332_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172077312)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172079936)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172082560)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175359424)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175362048)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178638912)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181915776)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_6370_begin_0 = const()[name = string("op_6370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6370_end_0 = const()[name = string("op_6370_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6370_end_mask_0 = const()[name = string("op_6370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6370_cast_fp16 = slice_by_index(begin = var_6370_begin_0, end = var_6370_end_0, end_mask = var_6370_end_mask_0, x = query_9_cast_fp16)[name = string("op_6370_cast_fp16")]; + tensor var_6374_begin_0 = const()[name = string("op_6374_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6374_end_0 = const()[name = string("op_6374_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6374_end_mask_0 = const()[name = string("op_6374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6374_cast_fp16 = slice_by_index(begin = var_6374_begin_0, end = var_6374_end_0, end_mask = var_6374_end_mask_0, x = query_9_cast_fp16)[name = string("op_6374_cast_fp16")]; + tensor var_6378_begin_0 = const()[name = string("op_6378_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6378_end_0 = const()[name = string("op_6378_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6378_end_mask_0 = const()[name = string("op_6378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6378_cast_fp16 = slice_by_index(begin = var_6378_begin_0, end = var_6378_end_0, end_mask = var_6378_end_mask_0, x = query_9_cast_fp16)[name = string("op_6378_cast_fp16")]; + tensor var_6382_begin_0 = const()[name = string("op_6382_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6382_end_0 = const()[name = string("op_6382_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6382_end_mask_0 = const()[name = string("op_6382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6382_cast_fp16 = slice_by_index(begin = var_6382_begin_0, end = var_6382_end_0, end_mask = var_6382_end_mask_0, x = query_9_cast_fp16)[name = string("op_6382_cast_fp16")]; + tensor var_6386_begin_0 = const()[name = string("op_6386_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6386_end_0 = const()[name = string("op_6386_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6386_end_mask_0 = const()[name = string("op_6386_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6386_cast_fp16 = slice_by_index(begin = var_6386_begin_0, end = var_6386_end_0, end_mask = var_6386_end_mask_0, x = query_9_cast_fp16)[name = string("op_6386_cast_fp16")]; + tensor var_6390_begin_0 = const()[name = string("op_6390_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6390_end_0 = const()[name = string("op_6390_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6390_end_mask_0 = const()[name = string("op_6390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6390_cast_fp16 = slice_by_index(begin = var_6390_begin_0, end = var_6390_end_0, end_mask = var_6390_end_mask_0, x = query_9_cast_fp16)[name = string("op_6390_cast_fp16")]; + tensor var_6394_begin_0 = const()[name = string("op_6394_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6394_end_0 = const()[name = string("op_6394_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6394_end_mask_0 = const()[name = string("op_6394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6394_cast_fp16 = slice_by_index(begin = var_6394_begin_0, end = var_6394_end_0, end_mask = var_6394_end_mask_0, x = query_9_cast_fp16)[name = string("op_6394_cast_fp16")]; + tensor var_6398_begin_0 = const()[name = string("op_6398_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6398_end_0 = const()[name = string("op_6398_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6398_end_mask_0 = const()[name = string("op_6398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6398_cast_fp16 = slice_by_index(begin = var_6398_begin_0, end = var_6398_end_0, end_mask = var_6398_end_mask_0, x = query_9_cast_fp16)[name = string("op_6398_cast_fp16")]; + tensor var_6402_begin_0 = const()[name = string("op_6402_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6402_end_0 = const()[name = string("op_6402_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6402_end_mask_0 = const()[name = string("op_6402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6402_cast_fp16 = slice_by_index(begin = var_6402_begin_0, end = var_6402_end_0, end_mask = var_6402_end_mask_0, x = query_9_cast_fp16)[name = string("op_6402_cast_fp16")]; + tensor var_6406_begin_0 = const()[name = string("op_6406_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6406_end_0 = const()[name = string("op_6406_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6406_end_mask_0 = const()[name = string("op_6406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6406_cast_fp16 = slice_by_index(begin = var_6406_begin_0, end = var_6406_end_0, end_mask = var_6406_end_mask_0, x = query_9_cast_fp16)[name = string("op_6406_cast_fp16")]; + tensor var_6410_begin_0 = const()[name = string("op_6410_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6410_end_0 = const()[name = string("op_6410_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6410_end_mask_0 = const()[name = string("op_6410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6410_cast_fp16 = slice_by_index(begin = var_6410_begin_0, end = var_6410_end_0, end_mask = var_6410_end_mask_0, x = query_9_cast_fp16)[name = string("op_6410_cast_fp16")]; + tensor var_6414_begin_0 = const()[name = string("op_6414_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6414_end_0 = const()[name = string("op_6414_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6414_end_mask_0 = const()[name = string("op_6414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6414_cast_fp16 = slice_by_index(begin = var_6414_begin_0, end = var_6414_end_0, end_mask = var_6414_end_mask_0, x = query_9_cast_fp16)[name = string("op_6414_cast_fp16")]; + tensor var_6418_begin_0 = const()[name = string("op_6418_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6418_end_0 = const()[name = string("op_6418_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_6418_end_mask_0 = const()[name = string("op_6418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6418_cast_fp16 = slice_by_index(begin = var_6418_begin_0, end = var_6418_end_0, end_mask = var_6418_end_mask_0, x = query_9_cast_fp16)[name = string("op_6418_cast_fp16")]; + tensor var_6422_begin_0 = const()[name = string("op_6422_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6422_end_0 = const()[name = string("op_6422_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_6422_end_mask_0 = const()[name = string("op_6422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6422_cast_fp16 = slice_by_index(begin = var_6422_begin_0, end = var_6422_end_0, end_mask = var_6422_end_mask_0, x = query_9_cast_fp16)[name = string("op_6422_cast_fp16")]; + tensor var_6426_begin_0 = const()[name = string("op_6426_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6426_end_0 = const()[name = string("op_6426_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_6426_end_mask_0 = const()[name = string("op_6426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6426_cast_fp16 = slice_by_index(begin = var_6426_begin_0, end = var_6426_end_0, end_mask = var_6426_end_mask_0, x = query_9_cast_fp16)[name = string("op_6426_cast_fp16")]; + tensor var_6430_begin_0 = const()[name = string("op_6430_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6430_end_0 = const()[name = string("op_6430_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_6430_end_mask_0 = const()[name = string("op_6430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6430_cast_fp16 = slice_by_index(begin = var_6430_begin_0, end = var_6430_end_0, end_mask = var_6430_end_mask_0, x = query_9_cast_fp16)[name = string("op_6430_cast_fp16")]; + tensor var_6434_begin_0 = const()[name = string("op_6434_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6434_end_0 = const()[name = string("op_6434_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_6434_end_mask_0 = const()[name = string("op_6434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6434_cast_fp16 = slice_by_index(begin = var_6434_begin_0, end = var_6434_end_0, end_mask = var_6434_end_mask_0, x = query_9_cast_fp16)[name = string("op_6434_cast_fp16")]; + tensor var_6438_begin_0 = const()[name = string("op_6438_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6438_end_0 = const()[name = string("op_6438_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_6438_end_mask_0 = const()[name = string("op_6438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6438_cast_fp16 = slice_by_index(begin = var_6438_begin_0, end = var_6438_end_0, end_mask = var_6438_end_mask_0, x = query_9_cast_fp16)[name = string("op_6438_cast_fp16")]; + tensor var_6442_begin_0 = const()[name = string("op_6442_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6442_end_0 = const()[name = string("op_6442_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_6442_end_mask_0 = const()[name = string("op_6442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6442_cast_fp16 = slice_by_index(begin = var_6442_begin_0, end = var_6442_end_0, end_mask = var_6442_end_mask_0, x = query_9_cast_fp16)[name = string("op_6442_cast_fp16")]; + tensor var_6446_begin_0 = const()[name = string("op_6446_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6446_end_0 = const()[name = string("op_6446_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_6446_end_mask_0 = const()[name = string("op_6446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6446_cast_fp16 = slice_by_index(begin = var_6446_begin_0, end = var_6446_end_0, end_mask = var_6446_end_mask_0, x = query_9_cast_fp16)[name = string("op_6446_cast_fp16")]; + tensor var_6455_begin_0 = const()[name = string("op_6455_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6455_end_0 = const()[name = string("op_6455_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6455_end_mask_0 = const()[name = string("op_6455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6455_cast_fp16 = slice_by_index(begin = var_6455_begin_0, end = var_6455_end_0, end_mask = var_6455_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6455_cast_fp16")]; + tensor var_6462_begin_0 = const()[name = string("op_6462_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6462_end_0 = const()[name = string("op_6462_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6462_end_mask_0 = const()[name = string("op_6462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6462_cast_fp16 = slice_by_index(begin = var_6462_begin_0, end = var_6462_end_0, end_mask = var_6462_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6462_cast_fp16")]; + tensor var_6469_begin_0 = const()[name = string("op_6469_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6469_end_0 = const()[name = string("op_6469_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6469_end_mask_0 = const()[name = string("op_6469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6469_cast_fp16 = slice_by_index(begin = var_6469_begin_0, end = var_6469_end_0, end_mask = var_6469_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6469_cast_fp16")]; + tensor var_6476_begin_0 = const()[name = string("op_6476_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6476_end_0 = const()[name = string("op_6476_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6476_end_mask_0 = const()[name = string("op_6476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6476_cast_fp16 = slice_by_index(begin = var_6476_begin_0, end = var_6476_end_0, end_mask = var_6476_end_mask_0, x = var_6370_cast_fp16)[name = string("op_6476_cast_fp16")]; + tensor var_6483_begin_0 = const()[name = string("op_6483_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6483_end_0 = const()[name = string("op_6483_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6483_end_mask_0 = const()[name = string("op_6483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6483_cast_fp16 = slice_by_index(begin = var_6483_begin_0, end = var_6483_end_0, end_mask = var_6483_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6483_cast_fp16")]; + tensor var_6490_begin_0 = const()[name = string("op_6490_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6490_end_0 = const()[name = string("op_6490_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6490_end_mask_0 = const()[name = string("op_6490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6490_cast_fp16 = slice_by_index(begin = var_6490_begin_0, end = var_6490_end_0, end_mask = var_6490_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6490_cast_fp16")]; + tensor var_6497_begin_0 = const()[name = string("op_6497_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6497_end_0 = const()[name = string("op_6497_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6497_end_mask_0 = const()[name = string("op_6497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6497_cast_fp16 = slice_by_index(begin = var_6497_begin_0, end = var_6497_end_0, end_mask = var_6497_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6497_cast_fp16")]; + tensor var_6504_begin_0 = const()[name = string("op_6504_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6504_end_0 = const()[name = string("op_6504_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6504_end_mask_0 = const()[name = string("op_6504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6504_cast_fp16 = slice_by_index(begin = var_6504_begin_0, end = var_6504_end_0, end_mask = var_6504_end_mask_0, x = var_6374_cast_fp16)[name = string("op_6504_cast_fp16")]; + tensor var_6511_begin_0 = const()[name = string("op_6511_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6511_end_0 = const()[name = string("op_6511_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6511_end_mask_0 = const()[name = string("op_6511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6511_cast_fp16 = slice_by_index(begin = var_6511_begin_0, end = var_6511_end_0, end_mask = var_6511_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6511_cast_fp16")]; + tensor var_6518_begin_0 = const()[name = string("op_6518_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6518_end_0 = const()[name = string("op_6518_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6518_end_mask_0 = const()[name = string("op_6518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6518_cast_fp16 = slice_by_index(begin = var_6518_begin_0, end = var_6518_end_0, end_mask = var_6518_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6518_cast_fp16")]; + tensor var_6525_begin_0 = const()[name = string("op_6525_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6525_end_0 = const()[name = string("op_6525_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6525_end_mask_0 = const()[name = string("op_6525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6525_cast_fp16 = slice_by_index(begin = var_6525_begin_0, end = var_6525_end_0, end_mask = var_6525_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6525_cast_fp16")]; + tensor var_6532_begin_0 = const()[name = string("op_6532_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6532_end_0 = const()[name = string("op_6532_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6532_end_mask_0 = const()[name = string("op_6532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6532_cast_fp16 = slice_by_index(begin = var_6532_begin_0, end = var_6532_end_0, end_mask = var_6532_end_mask_0, x = var_6378_cast_fp16)[name = string("op_6532_cast_fp16")]; + tensor var_6539_begin_0 = const()[name = string("op_6539_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6539_end_0 = const()[name = string("op_6539_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6539_end_mask_0 = const()[name = string("op_6539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6539_cast_fp16 = slice_by_index(begin = var_6539_begin_0, end = var_6539_end_0, end_mask = var_6539_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6539_cast_fp16")]; + tensor var_6546_begin_0 = const()[name = string("op_6546_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6546_end_0 = const()[name = string("op_6546_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6546_end_mask_0 = const()[name = string("op_6546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6546_cast_fp16 = slice_by_index(begin = var_6546_begin_0, end = var_6546_end_0, end_mask = var_6546_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6546_cast_fp16")]; + tensor var_6553_begin_0 = const()[name = string("op_6553_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6553_end_0 = const()[name = string("op_6553_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6553_end_mask_0 = const()[name = string("op_6553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6553_cast_fp16 = slice_by_index(begin = var_6553_begin_0, end = var_6553_end_0, end_mask = var_6553_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6553_cast_fp16")]; + tensor var_6560_begin_0 = const()[name = string("op_6560_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6560_end_0 = const()[name = string("op_6560_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6560_end_mask_0 = const()[name = string("op_6560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6560_cast_fp16 = slice_by_index(begin = var_6560_begin_0, end = var_6560_end_0, end_mask = var_6560_end_mask_0, x = var_6382_cast_fp16)[name = string("op_6560_cast_fp16")]; + tensor var_6567_begin_0 = const()[name = string("op_6567_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6567_end_0 = const()[name = string("op_6567_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6567_end_mask_0 = const()[name = string("op_6567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6567_cast_fp16 = slice_by_index(begin = var_6567_begin_0, end = var_6567_end_0, end_mask = var_6567_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6567_cast_fp16")]; + tensor var_6574_begin_0 = const()[name = string("op_6574_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6574_end_0 = const()[name = string("op_6574_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6574_end_mask_0 = const()[name = string("op_6574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6574_cast_fp16 = slice_by_index(begin = var_6574_begin_0, end = var_6574_end_0, end_mask = var_6574_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6574_cast_fp16")]; + tensor var_6581_begin_0 = const()[name = string("op_6581_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6581_end_0 = const()[name = string("op_6581_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6581_end_mask_0 = const()[name = string("op_6581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6581_cast_fp16 = slice_by_index(begin = var_6581_begin_0, end = var_6581_end_0, end_mask = var_6581_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6581_cast_fp16")]; + tensor var_6588_begin_0 = const()[name = string("op_6588_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6588_end_0 = const()[name = string("op_6588_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6588_end_mask_0 = const()[name = string("op_6588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6588_cast_fp16 = slice_by_index(begin = var_6588_begin_0, end = var_6588_end_0, end_mask = var_6588_end_mask_0, x = var_6386_cast_fp16)[name = string("op_6588_cast_fp16")]; + tensor var_6595_begin_0 = const()[name = string("op_6595_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6595_end_0 = const()[name = string("op_6595_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6595_end_mask_0 = const()[name = string("op_6595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6595_cast_fp16 = slice_by_index(begin = var_6595_begin_0, end = var_6595_end_0, end_mask = var_6595_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6595_cast_fp16")]; + tensor var_6602_begin_0 = const()[name = string("op_6602_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6602_end_0 = const()[name = string("op_6602_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6602_end_mask_0 = const()[name = string("op_6602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6602_cast_fp16 = slice_by_index(begin = var_6602_begin_0, end = var_6602_end_0, end_mask = var_6602_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6602_cast_fp16")]; + tensor var_6609_begin_0 = const()[name = string("op_6609_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6609_end_0 = const()[name = string("op_6609_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6609_end_mask_0 = const()[name = string("op_6609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6609_cast_fp16 = slice_by_index(begin = var_6609_begin_0, end = var_6609_end_0, end_mask = var_6609_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6609_cast_fp16")]; + tensor var_6616_begin_0 = const()[name = string("op_6616_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6616_end_0 = const()[name = string("op_6616_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6616_end_mask_0 = const()[name = string("op_6616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6616_cast_fp16 = slice_by_index(begin = var_6616_begin_0, end = var_6616_end_0, end_mask = var_6616_end_mask_0, x = var_6390_cast_fp16)[name = string("op_6616_cast_fp16")]; + tensor var_6623_begin_0 = const()[name = string("op_6623_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6623_end_0 = const()[name = string("op_6623_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6623_end_mask_0 = const()[name = string("op_6623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6623_cast_fp16 = slice_by_index(begin = var_6623_begin_0, end = var_6623_end_0, end_mask = var_6623_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6623_cast_fp16")]; + tensor var_6630_begin_0 = const()[name = string("op_6630_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6630_end_0 = const()[name = string("op_6630_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6630_end_mask_0 = const()[name = string("op_6630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6630_cast_fp16 = slice_by_index(begin = var_6630_begin_0, end = var_6630_end_0, end_mask = var_6630_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6630_cast_fp16")]; + tensor var_6637_begin_0 = const()[name = string("op_6637_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6637_end_0 = const()[name = string("op_6637_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6637_end_mask_0 = const()[name = string("op_6637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6637_cast_fp16 = slice_by_index(begin = var_6637_begin_0, end = var_6637_end_0, end_mask = var_6637_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6637_cast_fp16")]; + tensor var_6644_begin_0 = const()[name = string("op_6644_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6644_end_0 = const()[name = string("op_6644_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6644_end_mask_0 = const()[name = string("op_6644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6644_cast_fp16 = slice_by_index(begin = var_6644_begin_0, end = var_6644_end_0, end_mask = var_6644_end_mask_0, x = var_6394_cast_fp16)[name = string("op_6644_cast_fp16")]; + tensor var_6651_begin_0 = const()[name = string("op_6651_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6651_end_0 = const()[name = string("op_6651_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6651_end_mask_0 = const()[name = string("op_6651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6651_cast_fp16 = slice_by_index(begin = var_6651_begin_0, end = var_6651_end_0, end_mask = var_6651_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6651_cast_fp16")]; + tensor var_6658_begin_0 = const()[name = string("op_6658_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6658_end_0 = const()[name = string("op_6658_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6658_end_mask_0 = const()[name = string("op_6658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6658_cast_fp16 = slice_by_index(begin = var_6658_begin_0, end = var_6658_end_0, end_mask = var_6658_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6658_cast_fp16")]; + tensor var_6665_begin_0 = const()[name = string("op_6665_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6665_end_0 = const()[name = string("op_6665_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6665_end_mask_0 = const()[name = string("op_6665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6665_cast_fp16 = slice_by_index(begin = var_6665_begin_0, end = var_6665_end_0, end_mask = var_6665_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6665_cast_fp16")]; + tensor var_6672_begin_0 = const()[name = string("op_6672_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6672_end_0 = const()[name = string("op_6672_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6672_end_mask_0 = const()[name = string("op_6672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6672_cast_fp16 = slice_by_index(begin = var_6672_begin_0, end = var_6672_end_0, end_mask = var_6672_end_mask_0, x = var_6398_cast_fp16)[name = string("op_6672_cast_fp16")]; + tensor var_6679_begin_0 = const()[name = string("op_6679_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6679_end_0 = const()[name = string("op_6679_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6679_end_mask_0 = const()[name = string("op_6679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6679_cast_fp16 = slice_by_index(begin = var_6679_begin_0, end = var_6679_end_0, end_mask = var_6679_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6679_cast_fp16")]; + tensor var_6686_begin_0 = const()[name = string("op_6686_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6686_end_0 = const()[name = string("op_6686_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6686_end_mask_0 = const()[name = string("op_6686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6686_cast_fp16 = slice_by_index(begin = var_6686_begin_0, end = var_6686_end_0, end_mask = var_6686_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6686_cast_fp16")]; + tensor var_6693_begin_0 = const()[name = string("op_6693_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6693_end_0 = const()[name = string("op_6693_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6693_end_mask_0 = const()[name = string("op_6693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6693_cast_fp16 = slice_by_index(begin = var_6693_begin_0, end = var_6693_end_0, end_mask = var_6693_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6693_cast_fp16")]; + tensor var_6700_begin_0 = const()[name = string("op_6700_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6700_end_0 = const()[name = string("op_6700_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6700_end_mask_0 = const()[name = string("op_6700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6700_cast_fp16 = slice_by_index(begin = var_6700_begin_0, end = var_6700_end_0, end_mask = var_6700_end_mask_0, x = var_6402_cast_fp16)[name = string("op_6700_cast_fp16")]; + tensor var_6707_begin_0 = const()[name = string("op_6707_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6707_end_0 = const()[name = string("op_6707_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6707_end_mask_0 = const()[name = string("op_6707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6707_cast_fp16 = slice_by_index(begin = var_6707_begin_0, end = var_6707_end_0, end_mask = var_6707_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6707_cast_fp16")]; + tensor var_6714_begin_0 = const()[name = string("op_6714_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6714_end_0 = const()[name = string("op_6714_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6714_end_mask_0 = const()[name = string("op_6714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6714_cast_fp16 = slice_by_index(begin = var_6714_begin_0, end = var_6714_end_0, end_mask = var_6714_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6714_cast_fp16")]; + tensor var_6721_begin_0 = const()[name = string("op_6721_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6721_end_0 = const()[name = string("op_6721_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6721_end_mask_0 = const()[name = string("op_6721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6721_cast_fp16 = slice_by_index(begin = var_6721_begin_0, end = var_6721_end_0, end_mask = var_6721_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6721_cast_fp16")]; + tensor var_6728_begin_0 = const()[name = string("op_6728_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6728_end_0 = const()[name = string("op_6728_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6728_end_mask_0 = const()[name = string("op_6728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6728_cast_fp16 = slice_by_index(begin = var_6728_begin_0, end = var_6728_end_0, end_mask = var_6728_end_mask_0, x = var_6406_cast_fp16)[name = string("op_6728_cast_fp16")]; + tensor var_6735_begin_0 = const()[name = string("op_6735_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6735_end_0 = const()[name = string("op_6735_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6735_end_mask_0 = const()[name = string("op_6735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6735_cast_fp16 = slice_by_index(begin = var_6735_begin_0, end = var_6735_end_0, end_mask = var_6735_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6735_cast_fp16")]; + tensor var_6742_begin_0 = const()[name = string("op_6742_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6742_end_0 = const()[name = string("op_6742_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6742_end_mask_0 = const()[name = string("op_6742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6742_cast_fp16 = slice_by_index(begin = var_6742_begin_0, end = var_6742_end_0, end_mask = var_6742_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6742_cast_fp16")]; + tensor var_6749_begin_0 = const()[name = string("op_6749_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6749_end_0 = const()[name = string("op_6749_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6749_end_mask_0 = const()[name = string("op_6749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6749_cast_fp16 = slice_by_index(begin = var_6749_begin_0, end = var_6749_end_0, end_mask = var_6749_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6749_cast_fp16")]; + tensor var_6756_begin_0 = const()[name = string("op_6756_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6756_end_0 = const()[name = string("op_6756_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6756_end_mask_0 = const()[name = string("op_6756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6756_cast_fp16 = slice_by_index(begin = var_6756_begin_0, end = var_6756_end_0, end_mask = var_6756_end_mask_0, x = var_6410_cast_fp16)[name = string("op_6756_cast_fp16")]; + tensor var_6763_begin_0 = const()[name = string("op_6763_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6763_end_0 = const()[name = string("op_6763_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6763_end_mask_0 = const()[name = string("op_6763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6763_cast_fp16 = slice_by_index(begin = var_6763_begin_0, end = var_6763_end_0, end_mask = var_6763_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6763_cast_fp16")]; + tensor var_6770_begin_0 = const()[name = string("op_6770_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6770_end_0 = const()[name = string("op_6770_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6770_end_mask_0 = const()[name = string("op_6770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6770_cast_fp16 = slice_by_index(begin = var_6770_begin_0, end = var_6770_end_0, end_mask = var_6770_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6770_cast_fp16")]; + tensor var_6777_begin_0 = const()[name = string("op_6777_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6777_end_0 = const()[name = string("op_6777_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6777_end_mask_0 = const()[name = string("op_6777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6777_cast_fp16 = slice_by_index(begin = var_6777_begin_0, end = var_6777_end_0, end_mask = var_6777_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6777_cast_fp16")]; + tensor var_6784_begin_0 = const()[name = string("op_6784_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6784_end_0 = const()[name = string("op_6784_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6784_end_mask_0 = const()[name = string("op_6784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6784_cast_fp16 = slice_by_index(begin = var_6784_begin_0, end = var_6784_end_0, end_mask = var_6784_end_mask_0, x = var_6414_cast_fp16)[name = string("op_6784_cast_fp16")]; + tensor var_6791_begin_0 = const()[name = string("op_6791_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6791_end_0 = const()[name = string("op_6791_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6791_end_mask_0 = const()[name = string("op_6791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6791_cast_fp16 = slice_by_index(begin = var_6791_begin_0, end = var_6791_end_0, end_mask = var_6791_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6791_cast_fp16")]; + tensor var_6798_begin_0 = const()[name = string("op_6798_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6798_end_0 = const()[name = string("op_6798_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6798_end_mask_0 = const()[name = string("op_6798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6798_cast_fp16 = slice_by_index(begin = var_6798_begin_0, end = var_6798_end_0, end_mask = var_6798_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6798_cast_fp16")]; + tensor var_6805_begin_0 = const()[name = string("op_6805_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6805_end_0 = const()[name = string("op_6805_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6805_end_mask_0 = const()[name = string("op_6805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6805_cast_fp16 = slice_by_index(begin = var_6805_begin_0, end = var_6805_end_0, end_mask = var_6805_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6805_cast_fp16")]; + tensor var_6812_begin_0 = const()[name = string("op_6812_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6812_end_0 = const()[name = string("op_6812_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6812_end_mask_0 = const()[name = string("op_6812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6812_cast_fp16 = slice_by_index(begin = var_6812_begin_0, end = var_6812_end_0, end_mask = var_6812_end_mask_0, x = var_6418_cast_fp16)[name = string("op_6812_cast_fp16")]; + tensor var_6819_begin_0 = const()[name = string("op_6819_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6819_end_0 = const()[name = string("op_6819_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6819_end_mask_0 = const()[name = string("op_6819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6819_cast_fp16 = slice_by_index(begin = var_6819_begin_0, end = var_6819_end_0, end_mask = var_6819_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6819_cast_fp16")]; + tensor var_6826_begin_0 = const()[name = string("op_6826_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6826_end_0 = const()[name = string("op_6826_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6826_end_mask_0 = const()[name = string("op_6826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6826_cast_fp16 = slice_by_index(begin = var_6826_begin_0, end = var_6826_end_0, end_mask = var_6826_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6826_cast_fp16")]; + tensor var_6833_begin_0 = const()[name = string("op_6833_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6833_end_0 = const()[name = string("op_6833_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6833_end_mask_0 = const()[name = string("op_6833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6833_cast_fp16 = slice_by_index(begin = var_6833_begin_0, end = var_6833_end_0, end_mask = var_6833_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6833_cast_fp16")]; + tensor var_6840_begin_0 = const()[name = string("op_6840_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6840_end_0 = const()[name = string("op_6840_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6840_end_mask_0 = const()[name = string("op_6840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6840_cast_fp16 = slice_by_index(begin = var_6840_begin_0, end = var_6840_end_0, end_mask = var_6840_end_mask_0, x = var_6422_cast_fp16)[name = string("op_6840_cast_fp16")]; + tensor var_6847_begin_0 = const()[name = string("op_6847_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6847_end_0 = const()[name = string("op_6847_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6847_end_mask_0 = const()[name = string("op_6847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6847_cast_fp16 = slice_by_index(begin = var_6847_begin_0, end = var_6847_end_0, end_mask = var_6847_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6847_cast_fp16")]; + tensor var_6854_begin_0 = const()[name = string("op_6854_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6854_end_0 = const()[name = string("op_6854_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6854_end_mask_0 = const()[name = string("op_6854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6854_cast_fp16 = slice_by_index(begin = var_6854_begin_0, end = var_6854_end_0, end_mask = var_6854_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6854_cast_fp16")]; + tensor var_6861_begin_0 = const()[name = string("op_6861_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6861_end_0 = const()[name = string("op_6861_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6861_end_mask_0 = const()[name = string("op_6861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6861_cast_fp16 = slice_by_index(begin = var_6861_begin_0, end = var_6861_end_0, end_mask = var_6861_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6861_cast_fp16")]; + tensor var_6868_begin_0 = const()[name = string("op_6868_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6868_end_0 = const()[name = string("op_6868_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6868_end_mask_0 = const()[name = string("op_6868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6868_cast_fp16 = slice_by_index(begin = var_6868_begin_0, end = var_6868_end_0, end_mask = var_6868_end_mask_0, x = var_6426_cast_fp16)[name = string("op_6868_cast_fp16")]; + tensor var_6875_begin_0 = const()[name = string("op_6875_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6875_end_0 = const()[name = string("op_6875_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6875_end_mask_0 = const()[name = string("op_6875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6875_cast_fp16 = slice_by_index(begin = var_6875_begin_0, end = var_6875_end_0, end_mask = var_6875_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6875_cast_fp16")]; + tensor var_6882_begin_0 = const()[name = string("op_6882_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6882_end_0 = const()[name = string("op_6882_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6882_end_mask_0 = const()[name = string("op_6882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6882_cast_fp16 = slice_by_index(begin = var_6882_begin_0, end = var_6882_end_0, end_mask = var_6882_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6882_cast_fp16")]; + tensor var_6889_begin_0 = const()[name = string("op_6889_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6889_end_0 = const()[name = string("op_6889_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6889_end_mask_0 = const()[name = string("op_6889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6889_cast_fp16 = slice_by_index(begin = var_6889_begin_0, end = var_6889_end_0, end_mask = var_6889_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6889_cast_fp16")]; + tensor var_6896_begin_0 = const()[name = string("op_6896_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6896_end_0 = const()[name = string("op_6896_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6896_end_mask_0 = const()[name = string("op_6896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6896_cast_fp16 = slice_by_index(begin = var_6896_begin_0, end = var_6896_end_0, end_mask = var_6896_end_mask_0, x = var_6430_cast_fp16)[name = string("op_6896_cast_fp16")]; + tensor var_6903_begin_0 = const()[name = string("op_6903_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6903_end_0 = const()[name = string("op_6903_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6903_end_mask_0 = const()[name = string("op_6903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6903_cast_fp16 = slice_by_index(begin = var_6903_begin_0, end = var_6903_end_0, end_mask = var_6903_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6903_cast_fp16")]; + tensor var_6910_begin_0 = const()[name = string("op_6910_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6910_end_0 = const()[name = string("op_6910_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6910_end_mask_0 = const()[name = string("op_6910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6910_cast_fp16 = slice_by_index(begin = var_6910_begin_0, end = var_6910_end_0, end_mask = var_6910_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6910_cast_fp16")]; + tensor var_6917_begin_0 = const()[name = string("op_6917_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6917_end_0 = const()[name = string("op_6917_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6917_end_mask_0 = const()[name = string("op_6917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6917_cast_fp16 = slice_by_index(begin = var_6917_begin_0, end = var_6917_end_0, end_mask = var_6917_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6917_cast_fp16")]; + tensor var_6924_begin_0 = const()[name = string("op_6924_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6924_end_0 = const()[name = string("op_6924_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6924_end_mask_0 = const()[name = string("op_6924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6924_cast_fp16 = slice_by_index(begin = var_6924_begin_0, end = var_6924_end_0, end_mask = var_6924_end_mask_0, x = var_6434_cast_fp16)[name = string("op_6924_cast_fp16")]; + tensor var_6931_begin_0 = const()[name = string("op_6931_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6931_end_0 = const()[name = string("op_6931_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6931_end_mask_0 = const()[name = string("op_6931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6931_cast_fp16 = slice_by_index(begin = var_6931_begin_0, end = var_6931_end_0, end_mask = var_6931_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6931_cast_fp16")]; + tensor var_6938_begin_0 = const()[name = string("op_6938_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6938_end_0 = const()[name = string("op_6938_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6938_end_mask_0 = const()[name = string("op_6938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6938_cast_fp16 = slice_by_index(begin = var_6938_begin_0, end = var_6938_end_0, end_mask = var_6938_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6938_cast_fp16")]; + tensor var_6945_begin_0 = const()[name = string("op_6945_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6945_end_0 = const()[name = string("op_6945_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6945_end_mask_0 = const()[name = string("op_6945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6945_cast_fp16 = slice_by_index(begin = var_6945_begin_0, end = var_6945_end_0, end_mask = var_6945_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6945_cast_fp16")]; + tensor var_6952_begin_0 = const()[name = string("op_6952_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6952_end_0 = const()[name = string("op_6952_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6952_end_mask_0 = const()[name = string("op_6952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6952_cast_fp16 = slice_by_index(begin = var_6952_begin_0, end = var_6952_end_0, end_mask = var_6952_end_mask_0, x = var_6438_cast_fp16)[name = string("op_6952_cast_fp16")]; + tensor var_6959_begin_0 = const()[name = string("op_6959_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6959_end_0 = const()[name = string("op_6959_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6959_end_mask_0 = const()[name = string("op_6959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6959_cast_fp16")]; + tensor var_6966_begin_0 = const()[name = string("op_6966_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6966_end_0 = const()[name = string("op_6966_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6966_end_mask_0 = const()[name = string("op_6966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6966_cast_fp16 = slice_by_index(begin = var_6966_begin_0, end = var_6966_end_0, end_mask = var_6966_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6966_cast_fp16")]; + tensor var_6973_begin_0 = const()[name = string("op_6973_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6973_end_0 = const()[name = string("op_6973_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6973_end_mask_0 = const()[name = string("op_6973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6973_cast_fp16 = slice_by_index(begin = var_6973_begin_0, end = var_6973_end_0, end_mask = var_6973_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6973_cast_fp16")]; + tensor var_6980_begin_0 = const()[name = string("op_6980_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6980_end_0 = const()[name = string("op_6980_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6980_end_mask_0 = const()[name = string("op_6980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6980_cast_fp16 = slice_by_index(begin = var_6980_begin_0, end = var_6980_end_0, end_mask = var_6980_end_mask_0, x = var_6442_cast_fp16)[name = string("op_6980_cast_fp16")]; + tensor var_6987_begin_0 = const()[name = string("op_6987_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6987_end_0 = const()[name = string("op_6987_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6987_end_mask_0 = const()[name = string("op_6987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6987_cast_fp16 = slice_by_index(begin = var_6987_begin_0, end = var_6987_end_0, end_mask = var_6987_end_mask_0, x = var_6446_cast_fp16)[name = string("op_6987_cast_fp16")]; + tensor var_6994_begin_0 = const()[name = string("op_6994_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6994_end_0 = const()[name = string("op_6994_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6994_end_mask_0 = const()[name = string("op_6994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6994_cast_fp16 = slice_by_index(begin = var_6994_begin_0, end = var_6994_end_0, end_mask = var_6994_end_mask_0, x = var_6446_cast_fp16)[name = string("op_6994_cast_fp16")]; + tensor var_7001_begin_0 = const()[name = string("op_7001_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7001_end_0 = const()[name = string("op_7001_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7001_end_mask_0 = const()[name = string("op_7001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7001_cast_fp16 = slice_by_index(begin = var_7001_begin_0, end = var_7001_end_0, end_mask = var_7001_end_mask_0, x = var_6446_cast_fp16)[name = string("op_7001_cast_fp16")]; + tensor var_7008_begin_0 = const()[name = string("op_7008_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7008_end_0 = const()[name = string("op_7008_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7008_end_mask_0 = const()[name = string("op_7008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7008_cast_fp16 = slice_by_index(begin = var_7008_begin_0, end = var_7008_end_0, end_mask = var_7008_end_mask_0, x = var_6446_cast_fp16)[name = string("op_7008_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7013_begin_0 = const()[name = string("op_7013_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7013_end_0 = const()[name = string("op_7013_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_7013_end_mask_0 = const()[name = string("op_7013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_27")]; + tensor var_7013_cast_fp16 = slice_by_index(begin = var_7013_begin_0, end = var_7013_end_0, end_mask = var_7013_end_mask_0, x = k_9_cast_fp16)[name = string("op_7013_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = string("op_7017_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7017_end_0 = const()[name = string("op_7017_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_7017_end_mask_0 = const()[name = string("op_7017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = k_9_cast_fp16)[name = string("op_7017_cast_fp16")]; + tensor var_7021_begin_0 = const()[name = string("op_7021_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7021_end_0 = const()[name = string("op_7021_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_7021_end_mask_0 = const()[name = string("op_7021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7021_cast_fp16 = slice_by_index(begin = var_7021_begin_0, end = var_7021_end_0, end_mask = var_7021_end_mask_0, x = k_9_cast_fp16)[name = string("op_7021_cast_fp16")]; + tensor var_7025_begin_0 = const()[name = string("op_7025_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7025_end_0 = const()[name = string("op_7025_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_7025_end_mask_0 = const()[name = string("op_7025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7025_cast_fp16 = slice_by_index(begin = var_7025_begin_0, end = var_7025_end_0, end_mask = var_7025_end_mask_0, x = k_9_cast_fp16)[name = string("op_7025_cast_fp16")]; + tensor var_7029_begin_0 = const()[name = string("op_7029_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7029_end_0 = const()[name = string("op_7029_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_7029_end_mask_0 = const()[name = string("op_7029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7029_cast_fp16 = slice_by_index(begin = var_7029_begin_0, end = var_7029_end_0, end_mask = var_7029_end_mask_0, x = k_9_cast_fp16)[name = string("op_7029_cast_fp16")]; + tensor var_7033_begin_0 = const()[name = string("op_7033_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7033_end_0 = const()[name = string("op_7033_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_7033_end_mask_0 = const()[name = string("op_7033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7033_cast_fp16 = slice_by_index(begin = var_7033_begin_0, end = var_7033_end_0, end_mask = var_7033_end_mask_0, x = k_9_cast_fp16)[name = string("op_7033_cast_fp16")]; + tensor var_7037_begin_0 = const()[name = string("op_7037_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7037_end_0 = const()[name = string("op_7037_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_7037_end_mask_0 = const()[name = string("op_7037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7037_cast_fp16 = slice_by_index(begin = var_7037_begin_0, end = var_7037_end_0, end_mask = var_7037_end_mask_0, x = k_9_cast_fp16)[name = string("op_7037_cast_fp16")]; + tensor var_7041_begin_0 = const()[name = string("op_7041_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7041_end_0 = const()[name = string("op_7041_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_7041_end_mask_0 = const()[name = string("op_7041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7041_cast_fp16 = slice_by_index(begin = var_7041_begin_0, end = var_7041_end_0, end_mask = var_7041_end_mask_0, x = k_9_cast_fp16)[name = string("op_7041_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = string("op_7045_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7045_end_0 = const()[name = string("op_7045_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_7045_end_mask_0 = const()[name = string("op_7045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = k_9_cast_fp16)[name = string("op_7045_cast_fp16")]; + tensor var_7049_begin_0 = const()[name = string("op_7049_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7049_end_0 = const()[name = string("op_7049_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_7049_end_mask_0 = const()[name = string("op_7049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7049_cast_fp16 = slice_by_index(begin = var_7049_begin_0, end = var_7049_end_0, end_mask = var_7049_end_mask_0, x = k_9_cast_fp16)[name = string("op_7049_cast_fp16")]; + tensor var_7053_begin_0 = const()[name = string("op_7053_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7053_end_0 = const()[name = string("op_7053_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_7053_end_mask_0 = const()[name = string("op_7053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7053_cast_fp16 = slice_by_index(begin = var_7053_begin_0, end = var_7053_end_0, end_mask = var_7053_end_mask_0, x = k_9_cast_fp16)[name = string("op_7053_cast_fp16")]; + tensor var_7057_begin_0 = const()[name = string("op_7057_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7057_end_0 = const()[name = string("op_7057_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_7057_end_mask_0 = const()[name = string("op_7057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7057_cast_fp16 = slice_by_index(begin = var_7057_begin_0, end = var_7057_end_0, end_mask = var_7057_end_mask_0, x = k_9_cast_fp16)[name = string("op_7057_cast_fp16")]; + tensor var_7061_begin_0 = const()[name = string("op_7061_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7061_end_0 = const()[name = string("op_7061_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_7061_end_mask_0 = const()[name = string("op_7061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7061_cast_fp16 = slice_by_index(begin = var_7061_begin_0, end = var_7061_end_0, end_mask = var_7061_end_mask_0, x = k_9_cast_fp16)[name = string("op_7061_cast_fp16")]; + tensor var_7065_begin_0 = const()[name = string("op_7065_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7065_end_0 = const()[name = string("op_7065_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_7065_end_mask_0 = const()[name = string("op_7065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7065_cast_fp16 = slice_by_index(begin = var_7065_begin_0, end = var_7065_end_0, end_mask = var_7065_end_mask_0, x = k_9_cast_fp16)[name = string("op_7065_cast_fp16")]; + tensor var_7069_begin_0 = const()[name = string("op_7069_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7069_end_0 = const()[name = string("op_7069_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_7069_end_mask_0 = const()[name = string("op_7069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7069_cast_fp16 = slice_by_index(begin = var_7069_begin_0, end = var_7069_end_0, end_mask = var_7069_end_mask_0, x = k_9_cast_fp16)[name = string("op_7069_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = string("op_7073_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7073_end_0 = const()[name = string("op_7073_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_7073_end_mask_0 = const()[name = string("op_7073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = k_9_cast_fp16)[name = string("op_7073_cast_fp16")]; + tensor var_7077_begin_0 = const()[name = string("op_7077_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7077_end_0 = const()[name = string("op_7077_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_7077_end_mask_0 = const()[name = string("op_7077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7077_cast_fp16 = slice_by_index(begin = var_7077_begin_0, end = var_7077_end_0, end_mask = var_7077_end_mask_0, x = k_9_cast_fp16)[name = string("op_7077_cast_fp16")]; + tensor var_7081_begin_0 = const()[name = string("op_7081_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7081_end_0 = const()[name = string("op_7081_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_7081_end_mask_0 = const()[name = string("op_7081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7081_cast_fp16 = slice_by_index(begin = var_7081_begin_0, end = var_7081_end_0, end_mask = var_7081_end_mask_0, x = k_9_cast_fp16)[name = string("op_7081_cast_fp16")]; + tensor var_7085_begin_0 = const()[name = string("op_7085_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7085_end_0 = const()[name = string("op_7085_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_7085_end_mask_0 = const()[name = string("op_7085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7085_cast_fp16 = slice_by_index(begin = var_7085_begin_0, end = var_7085_end_0, end_mask = var_7085_end_mask_0, x = k_9_cast_fp16)[name = string("op_7085_cast_fp16")]; + tensor var_7089_begin_0 = const()[name = string("op_7089_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7089_end_0 = const()[name = string("op_7089_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_7089_end_mask_0 = const()[name = string("op_7089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7089_cast_fp16 = slice_by_index(begin = var_7089_begin_0, end = var_7089_end_0, end_mask = var_7089_end_mask_0, x = k_9_cast_fp16)[name = string("op_7089_cast_fp16")]; + tensor var_7091_begin_0 = const()[name = string("op_7091_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7091_end_0 = const()[name = string("op_7091_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7091_end_mask_0 = const()[name = string("op_7091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7091_cast_fp16 = slice_by_index(begin = var_7091_begin_0, end = var_7091_end_0, end_mask = var_7091_end_mask_0, x = value_9_cast_fp16)[name = string("op_7091_cast_fp16")]; + tensor var_7095_begin_0 = const()[name = string("op_7095_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7095_end_0 = const()[name = string("op_7095_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7095_end_mask_0 = const()[name = string("op_7095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7095_cast_fp16 = slice_by_index(begin = var_7095_begin_0, end = var_7095_end_0, end_mask = var_7095_end_mask_0, x = value_9_cast_fp16)[name = string("op_7095_cast_fp16")]; + tensor var_7099_begin_0 = const()[name = string("op_7099_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7099_end_0 = const()[name = string("op_7099_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7099_end_mask_0 = const()[name = string("op_7099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7099_cast_fp16 = slice_by_index(begin = var_7099_begin_0, end = var_7099_end_0, end_mask = var_7099_end_mask_0, x = value_9_cast_fp16)[name = string("op_7099_cast_fp16")]; + tensor var_7103_begin_0 = const()[name = string("op_7103_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7103_end_0 = const()[name = string("op_7103_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7103_end_mask_0 = const()[name = string("op_7103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7103_cast_fp16 = slice_by_index(begin = var_7103_begin_0, end = var_7103_end_0, end_mask = var_7103_end_mask_0, x = value_9_cast_fp16)[name = string("op_7103_cast_fp16")]; + tensor var_7107_begin_0 = const()[name = string("op_7107_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7107_end_0 = const()[name = string("op_7107_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7107_end_mask_0 = const()[name = string("op_7107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7107_cast_fp16 = slice_by_index(begin = var_7107_begin_0, end = var_7107_end_0, end_mask = var_7107_end_mask_0, x = value_9_cast_fp16)[name = string("op_7107_cast_fp16")]; + tensor var_7111_begin_0 = const()[name = string("op_7111_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7111_end_0 = const()[name = string("op_7111_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7111_end_mask_0 = const()[name = string("op_7111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7111_cast_fp16 = slice_by_index(begin = var_7111_begin_0, end = var_7111_end_0, end_mask = var_7111_end_mask_0, x = value_9_cast_fp16)[name = string("op_7111_cast_fp16")]; + tensor var_7115_begin_0 = const()[name = string("op_7115_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7115_end_0 = const()[name = string("op_7115_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7115_end_mask_0 = const()[name = string("op_7115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7115_cast_fp16 = slice_by_index(begin = var_7115_begin_0, end = var_7115_end_0, end_mask = var_7115_end_mask_0, x = value_9_cast_fp16)[name = string("op_7115_cast_fp16")]; + tensor var_7119_begin_0 = const()[name = string("op_7119_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7119_end_0 = const()[name = string("op_7119_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7119_end_mask_0 = const()[name = string("op_7119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7119_cast_fp16 = slice_by_index(begin = var_7119_begin_0, end = var_7119_end_0, end_mask = var_7119_end_mask_0, x = value_9_cast_fp16)[name = string("op_7119_cast_fp16")]; + tensor var_7123_begin_0 = const()[name = string("op_7123_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7123_end_0 = const()[name = string("op_7123_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7123_end_mask_0 = const()[name = string("op_7123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7123_cast_fp16 = slice_by_index(begin = var_7123_begin_0, end = var_7123_end_0, end_mask = var_7123_end_mask_0, x = value_9_cast_fp16)[name = string("op_7123_cast_fp16")]; + tensor var_7127_begin_0 = const()[name = string("op_7127_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7127_end_0 = const()[name = string("op_7127_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7127_end_mask_0 = const()[name = string("op_7127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7127_cast_fp16 = slice_by_index(begin = var_7127_begin_0, end = var_7127_end_0, end_mask = var_7127_end_mask_0, x = value_9_cast_fp16)[name = string("op_7127_cast_fp16")]; + tensor var_7131_begin_0 = const()[name = string("op_7131_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7131_end_0 = const()[name = string("op_7131_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7131_end_mask_0 = const()[name = string("op_7131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7131_cast_fp16 = slice_by_index(begin = var_7131_begin_0, end = var_7131_end_0, end_mask = var_7131_end_mask_0, x = value_9_cast_fp16)[name = string("op_7131_cast_fp16")]; + tensor var_7135_begin_0 = const()[name = string("op_7135_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7135_end_0 = const()[name = string("op_7135_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7135_end_mask_0 = const()[name = string("op_7135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7135_cast_fp16 = slice_by_index(begin = var_7135_begin_0, end = var_7135_end_0, end_mask = var_7135_end_mask_0, x = value_9_cast_fp16)[name = string("op_7135_cast_fp16")]; + tensor var_7139_begin_0 = const()[name = string("op_7139_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7139_end_0 = const()[name = string("op_7139_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_7139_end_mask_0 = const()[name = string("op_7139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7139_cast_fp16 = slice_by_index(begin = var_7139_begin_0, end = var_7139_end_0, end_mask = var_7139_end_mask_0, x = value_9_cast_fp16)[name = string("op_7139_cast_fp16")]; + tensor var_7143_begin_0 = const()[name = string("op_7143_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7143_end_0 = const()[name = string("op_7143_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_7143_end_mask_0 = const()[name = string("op_7143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7143_cast_fp16 = slice_by_index(begin = var_7143_begin_0, end = var_7143_end_0, end_mask = var_7143_end_mask_0, x = value_9_cast_fp16)[name = string("op_7143_cast_fp16")]; + tensor var_7147_begin_0 = const()[name = string("op_7147_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7147_end_0 = const()[name = string("op_7147_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_7147_end_mask_0 = const()[name = string("op_7147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7147_cast_fp16 = slice_by_index(begin = var_7147_begin_0, end = var_7147_end_0, end_mask = var_7147_end_mask_0, x = value_9_cast_fp16)[name = string("op_7147_cast_fp16")]; + tensor var_7151_begin_0 = const()[name = string("op_7151_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7151_end_0 = const()[name = string("op_7151_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_7151_end_mask_0 = const()[name = string("op_7151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7151_cast_fp16 = slice_by_index(begin = var_7151_begin_0, end = var_7151_end_0, end_mask = var_7151_end_mask_0, x = value_9_cast_fp16)[name = string("op_7151_cast_fp16")]; + tensor var_7155_begin_0 = const()[name = string("op_7155_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7155_end_0 = const()[name = string("op_7155_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_7155_end_mask_0 = const()[name = string("op_7155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7155_cast_fp16 = slice_by_index(begin = var_7155_begin_0, end = var_7155_end_0, end_mask = var_7155_end_mask_0, x = value_9_cast_fp16)[name = string("op_7155_cast_fp16")]; + tensor var_7159_begin_0 = const()[name = string("op_7159_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7159_end_0 = const()[name = string("op_7159_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_7159_end_mask_0 = const()[name = string("op_7159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7159_cast_fp16 = slice_by_index(begin = var_7159_begin_0, end = var_7159_end_0, end_mask = var_7159_end_mask_0, x = value_9_cast_fp16)[name = string("op_7159_cast_fp16")]; + tensor var_7163_begin_0 = const()[name = string("op_7163_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7163_end_0 = const()[name = string("op_7163_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_7163_end_mask_0 = const()[name = string("op_7163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7163_cast_fp16 = slice_by_index(begin = var_7163_begin_0, end = var_7163_end_0, end_mask = var_7163_end_mask_0, x = value_9_cast_fp16)[name = string("op_7163_cast_fp16")]; + tensor var_7167_begin_0 = const()[name = string("op_7167_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7167_end_0 = const()[name = string("op_7167_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_7167_end_mask_0 = const()[name = string("op_7167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7167_cast_fp16 = slice_by_index(begin = var_7167_begin_0, end = var_7167_end_0, end_mask = var_7167_end_mask_0, x = value_9_cast_fp16)[name = string("op_7167_cast_fp16")]; + string _SplitHeadsQ__mh_w_641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_641_equation_0, values = (var_7013_cast_fp16, var_6455_cast_fp16))[name = string("_SplitHeadsQ__mh_w_641_cast_fp16")]; + string _SplitHeadsQ__mh_w_643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_643_equation_0, values = (var_7013_cast_fp16, var_6462_cast_fp16))[name = string("_SplitHeadsQ__mh_w_643_cast_fp16")]; + string _SplitHeadsQ__mh_w_645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_645_equation_0, values = (var_7013_cast_fp16, var_6469_cast_fp16))[name = string("_SplitHeadsQ__mh_w_645_cast_fp16")]; + string _SplitHeadsQ__mh_w_647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_647_equation_0, values = (var_7013_cast_fp16, var_6476_cast_fp16))[name = string("_SplitHeadsQ__mh_w_647_cast_fp16")]; + string _SplitHeadsQ__mh_w_649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_649_equation_0, values = (var_7017_cast_fp16, var_6483_cast_fp16))[name = string("_SplitHeadsQ__mh_w_649_cast_fp16")]; + string _SplitHeadsQ__mh_w_651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_651_equation_0, values = (var_7017_cast_fp16, var_6490_cast_fp16))[name = string("_SplitHeadsQ__mh_w_651_cast_fp16")]; + string _SplitHeadsQ__mh_w_653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_653_equation_0, values = (var_7017_cast_fp16, var_6497_cast_fp16))[name = string("_SplitHeadsQ__mh_w_653_cast_fp16")]; + string _SplitHeadsQ__mh_w_655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_655_equation_0, values = (var_7017_cast_fp16, var_6504_cast_fp16))[name = string("_SplitHeadsQ__mh_w_655_cast_fp16")]; + string _SplitHeadsQ__mh_w_657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_657_equation_0, values = (var_7021_cast_fp16, var_6511_cast_fp16))[name = string("_SplitHeadsQ__mh_w_657_cast_fp16")]; + string _SplitHeadsQ__mh_w_659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_659_equation_0, values = (var_7021_cast_fp16, var_6518_cast_fp16))[name = string("_SplitHeadsQ__mh_w_659_cast_fp16")]; + string _SplitHeadsQ__mh_w_661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_661_equation_0, values = (var_7021_cast_fp16, var_6525_cast_fp16))[name = string("_SplitHeadsQ__mh_w_661_cast_fp16")]; + string _SplitHeadsQ__mh_w_663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_663_equation_0, values = (var_7021_cast_fp16, var_6532_cast_fp16))[name = string("_SplitHeadsQ__mh_w_663_cast_fp16")]; + string _SplitHeadsQ__mh_w_665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_665_equation_0, values = (var_7025_cast_fp16, var_6539_cast_fp16))[name = string("_SplitHeadsQ__mh_w_665_cast_fp16")]; + string _SplitHeadsQ__mh_w_667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_667_equation_0, values = (var_7025_cast_fp16, var_6546_cast_fp16))[name = string("_SplitHeadsQ__mh_w_667_cast_fp16")]; + string _SplitHeadsQ__mh_w_669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_669_equation_0, values = (var_7025_cast_fp16, var_6553_cast_fp16))[name = string("_SplitHeadsQ__mh_w_669_cast_fp16")]; + string _SplitHeadsQ__mh_w_671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_671_equation_0, values = (var_7025_cast_fp16, var_6560_cast_fp16))[name = string("_SplitHeadsQ__mh_w_671_cast_fp16")]; + string _SplitHeadsQ__mh_w_673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_673_equation_0, values = (var_7029_cast_fp16, var_6567_cast_fp16))[name = string("_SplitHeadsQ__mh_w_673_cast_fp16")]; + string _SplitHeadsQ__mh_w_675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_675_equation_0, values = (var_7029_cast_fp16, var_6574_cast_fp16))[name = string("_SplitHeadsQ__mh_w_675_cast_fp16")]; + string _SplitHeadsQ__mh_w_677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_677_equation_0, values = (var_7029_cast_fp16, var_6581_cast_fp16))[name = string("_SplitHeadsQ__mh_w_677_cast_fp16")]; + string _SplitHeadsQ__mh_w_679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_679_equation_0, values = (var_7029_cast_fp16, var_6588_cast_fp16))[name = string("_SplitHeadsQ__mh_w_679_cast_fp16")]; + string _SplitHeadsQ__mh_w_681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_681_equation_0, values = (var_7033_cast_fp16, var_6595_cast_fp16))[name = string("_SplitHeadsQ__mh_w_681_cast_fp16")]; + string _SplitHeadsQ__mh_w_683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_683_equation_0, values = (var_7033_cast_fp16, var_6602_cast_fp16))[name = string("_SplitHeadsQ__mh_w_683_cast_fp16")]; + string _SplitHeadsQ__mh_w_685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_685_equation_0, values = (var_7033_cast_fp16, var_6609_cast_fp16))[name = string("_SplitHeadsQ__mh_w_685_cast_fp16")]; + string _SplitHeadsQ__mh_w_687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_687_equation_0, values = (var_7033_cast_fp16, var_6616_cast_fp16))[name = string("_SplitHeadsQ__mh_w_687_cast_fp16")]; + string _SplitHeadsQ__mh_w_689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_689_equation_0, values = (var_7037_cast_fp16, var_6623_cast_fp16))[name = string("_SplitHeadsQ__mh_w_689_cast_fp16")]; + string _SplitHeadsQ__mh_w_691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_691_equation_0, values = (var_7037_cast_fp16, var_6630_cast_fp16))[name = string("_SplitHeadsQ__mh_w_691_cast_fp16")]; + string _SplitHeadsQ__mh_w_693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_693_equation_0, values = (var_7037_cast_fp16, var_6637_cast_fp16))[name = string("_SplitHeadsQ__mh_w_693_cast_fp16")]; + string _SplitHeadsQ__mh_w_695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_695_equation_0, values = (var_7037_cast_fp16, var_6644_cast_fp16))[name = string("_SplitHeadsQ__mh_w_695_cast_fp16")]; + string _SplitHeadsQ__mh_w_697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_697_equation_0, values = (var_7041_cast_fp16, var_6651_cast_fp16))[name = string("_SplitHeadsQ__mh_w_697_cast_fp16")]; + string _SplitHeadsQ__mh_w_699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_699_equation_0, values = (var_7041_cast_fp16, var_6658_cast_fp16))[name = string("_SplitHeadsQ__mh_w_699_cast_fp16")]; + string _SplitHeadsQ__mh_w_701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_701_equation_0, values = (var_7041_cast_fp16, var_6665_cast_fp16))[name = string("_SplitHeadsQ__mh_w_701_cast_fp16")]; + string _SplitHeadsQ__mh_w_703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_703_equation_0, values = (var_7041_cast_fp16, var_6672_cast_fp16))[name = string("_SplitHeadsQ__mh_w_703_cast_fp16")]; + string _SplitHeadsQ__mh_w_705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_705_equation_0, values = (var_7045_cast_fp16, var_6679_cast_fp16))[name = string("_SplitHeadsQ__mh_w_705_cast_fp16")]; + string _SplitHeadsQ__mh_w_707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_707_equation_0, values = (var_7045_cast_fp16, var_6686_cast_fp16))[name = string("_SplitHeadsQ__mh_w_707_cast_fp16")]; + string _SplitHeadsQ__mh_w_709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_709_equation_0, values = (var_7045_cast_fp16, var_6693_cast_fp16))[name = string("_SplitHeadsQ__mh_w_709_cast_fp16")]; + string _SplitHeadsQ__mh_w_711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_711_equation_0, values = (var_7045_cast_fp16, var_6700_cast_fp16))[name = string("_SplitHeadsQ__mh_w_711_cast_fp16")]; + string _SplitHeadsQ__mh_w_713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_713_equation_0, values = (var_7049_cast_fp16, var_6707_cast_fp16))[name = string("_SplitHeadsQ__mh_w_713_cast_fp16")]; + string _SplitHeadsQ__mh_w_715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_715_equation_0, values = (var_7049_cast_fp16, var_6714_cast_fp16))[name = string("_SplitHeadsQ__mh_w_715_cast_fp16")]; + string _SplitHeadsQ__mh_w_717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_717_equation_0, values = (var_7049_cast_fp16, var_6721_cast_fp16))[name = string("_SplitHeadsQ__mh_w_717_cast_fp16")]; + string _SplitHeadsQ__mh_w_719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_719_equation_0, values = (var_7049_cast_fp16, var_6728_cast_fp16))[name = string("_SplitHeadsQ__mh_w_719_cast_fp16")]; + string _SplitHeadsQ__mh_w_721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_721_equation_0, values = (var_7053_cast_fp16, var_6735_cast_fp16))[name = string("_SplitHeadsQ__mh_w_721_cast_fp16")]; + string _SplitHeadsQ__mh_w_723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_723_equation_0, values = (var_7053_cast_fp16, var_6742_cast_fp16))[name = string("_SplitHeadsQ__mh_w_723_cast_fp16")]; + string _SplitHeadsQ__mh_w_725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_725_equation_0, values = (var_7053_cast_fp16, var_6749_cast_fp16))[name = string("_SplitHeadsQ__mh_w_725_cast_fp16")]; + string _SplitHeadsQ__mh_w_727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_727_equation_0, values = (var_7053_cast_fp16, var_6756_cast_fp16))[name = string("_SplitHeadsQ__mh_w_727_cast_fp16")]; + string _SplitHeadsQ__mh_w_729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_729_equation_0, values = (var_7057_cast_fp16, var_6763_cast_fp16))[name = string("_SplitHeadsQ__mh_w_729_cast_fp16")]; + string _SplitHeadsQ__mh_w_731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_731_equation_0, values = (var_7057_cast_fp16, var_6770_cast_fp16))[name = string("_SplitHeadsQ__mh_w_731_cast_fp16")]; + string _SplitHeadsQ__mh_w_733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_733_equation_0, values = (var_7057_cast_fp16, var_6777_cast_fp16))[name = string("_SplitHeadsQ__mh_w_733_cast_fp16")]; + string _SplitHeadsQ__mh_w_735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_735_equation_0, values = (var_7057_cast_fp16, var_6784_cast_fp16))[name = string("_SplitHeadsQ__mh_w_735_cast_fp16")]; + string _SplitHeadsQ__mh_w_737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_737_equation_0, values = (var_7061_cast_fp16, var_6791_cast_fp16))[name = string("_SplitHeadsQ__mh_w_737_cast_fp16")]; + string _SplitHeadsQ__mh_w_739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_739_equation_0, values = (var_7061_cast_fp16, var_6798_cast_fp16))[name = string("_SplitHeadsQ__mh_w_739_cast_fp16")]; + string _SplitHeadsQ__mh_w_741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_741_equation_0, values = (var_7061_cast_fp16, var_6805_cast_fp16))[name = string("_SplitHeadsQ__mh_w_741_cast_fp16")]; + string _SplitHeadsQ__mh_w_743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_743_equation_0, values = (var_7061_cast_fp16, var_6812_cast_fp16))[name = string("_SplitHeadsQ__mh_w_743_cast_fp16")]; + string _SplitHeadsQ__mh_w_745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_745_equation_0, values = (var_7065_cast_fp16, var_6819_cast_fp16))[name = string("_SplitHeadsQ__mh_w_745_cast_fp16")]; + string _SplitHeadsQ__mh_w_747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_747_equation_0, values = (var_7065_cast_fp16, var_6826_cast_fp16))[name = string("_SplitHeadsQ__mh_w_747_cast_fp16")]; + string _SplitHeadsQ__mh_w_749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_749_equation_0, values = (var_7065_cast_fp16, var_6833_cast_fp16))[name = string("_SplitHeadsQ__mh_w_749_cast_fp16")]; + string _SplitHeadsQ__mh_w_751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_751_equation_0, values = (var_7065_cast_fp16, var_6840_cast_fp16))[name = string("_SplitHeadsQ__mh_w_751_cast_fp16")]; + string _SplitHeadsQ__mh_w_753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_753_equation_0, values = (var_7069_cast_fp16, var_6847_cast_fp16))[name = string("_SplitHeadsQ__mh_w_753_cast_fp16")]; + string _SplitHeadsQ__mh_w_755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_755_equation_0, values = (var_7069_cast_fp16, var_6854_cast_fp16))[name = string("_SplitHeadsQ__mh_w_755_cast_fp16")]; + string _SplitHeadsQ__mh_w_757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_757_equation_0, values = (var_7069_cast_fp16, var_6861_cast_fp16))[name = string("_SplitHeadsQ__mh_w_757_cast_fp16")]; + string _SplitHeadsQ__mh_w_759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_759_equation_0, values = (var_7069_cast_fp16, var_6868_cast_fp16))[name = string("_SplitHeadsQ__mh_w_759_cast_fp16")]; + string _SplitHeadsQ__mh_w_761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_761_equation_0, values = (var_7073_cast_fp16, var_6875_cast_fp16))[name = string("_SplitHeadsQ__mh_w_761_cast_fp16")]; + string _SplitHeadsQ__mh_w_763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_763_equation_0, values = (var_7073_cast_fp16, var_6882_cast_fp16))[name = string("_SplitHeadsQ__mh_w_763_cast_fp16")]; + string _SplitHeadsQ__mh_w_765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_765_equation_0, values = (var_7073_cast_fp16, var_6889_cast_fp16))[name = string("_SplitHeadsQ__mh_w_765_cast_fp16")]; + string _SplitHeadsQ__mh_w_767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_767_equation_0, values = (var_7073_cast_fp16, var_6896_cast_fp16))[name = string("_SplitHeadsQ__mh_w_767_cast_fp16")]; + string _SplitHeadsQ__mh_w_769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_769_equation_0, values = (var_7077_cast_fp16, var_6903_cast_fp16))[name = string("_SplitHeadsQ__mh_w_769_cast_fp16")]; + string _SplitHeadsQ__mh_w_771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_771_equation_0, values = (var_7077_cast_fp16, var_6910_cast_fp16))[name = string("_SplitHeadsQ__mh_w_771_cast_fp16")]; + string _SplitHeadsQ__mh_w_773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_773_equation_0, values = (var_7077_cast_fp16, var_6917_cast_fp16))[name = string("_SplitHeadsQ__mh_w_773_cast_fp16")]; + string _SplitHeadsQ__mh_w_775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_775_equation_0, values = (var_7077_cast_fp16, var_6924_cast_fp16))[name = string("_SplitHeadsQ__mh_w_775_cast_fp16")]; + string _SplitHeadsQ__mh_w_777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_777_equation_0, values = (var_7081_cast_fp16, var_6931_cast_fp16))[name = string("_SplitHeadsQ__mh_w_777_cast_fp16")]; + string _SplitHeadsQ__mh_w_779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_779_equation_0, values = (var_7081_cast_fp16, var_6938_cast_fp16))[name = string("_SplitHeadsQ__mh_w_779_cast_fp16")]; + string _SplitHeadsQ__mh_w_781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_781_equation_0, values = (var_7081_cast_fp16, var_6945_cast_fp16))[name = string("_SplitHeadsQ__mh_w_781_cast_fp16")]; + string _SplitHeadsQ__mh_w_783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_783_equation_0, values = (var_7081_cast_fp16, var_6952_cast_fp16))[name = string("_SplitHeadsQ__mh_w_783_cast_fp16")]; + string _SplitHeadsQ__mh_w_785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_785_equation_0, values = (var_7085_cast_fp16, var_6959_cast_fp16))[name = string("_SplitHeadsQ__mh_w_785_cast_fp16")]; + string _SplitHeadsQ__mh_w_787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_787_equation_0, values = (var_7085_cast_fp16, var_6966_cast_fp16))[name = string("_SplitHeadsQ__mh_w_787_cast_fp16")]; + string _SplitHeadsQ__mh_w_789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_789_equation_0, values = (var_7085_cast_fp16, var_6973_cast_fp16))[name = string("_SplitHeadsQ__mh_w_789_cast_fp16")]; + string _SplitHeadsQ__mh_w_791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_791_equation_0, values = (var_7085_cast_fp16, var_6980_cast_fp16))[name = string("_SplitHeadsQ__mh_w_791_cast_fp16")]; + string _SplitHeadsQ__mh_w_793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_793_equation_0, values = (var_7089_cast_fp16, var_6987_cast_fp16))[name = string("_SplitHeadsQ__mh_w_793_cast_fp16")]; + string _SplitHeadsQ__mh_w_795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_795_equation_0, values = (var_7089_cast_fp16, var_6994_cast_fp16))[name = string("_SplitHeadsQ__mh_w_795_cast_fp16")]; + string _SplitHeadsQ__mh_w_797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_797_equation_0, values = (var_7089_cast_fp16, var_7001_cast_fp16))[name = string("_SplitHeadsQ__mh_w_797_cast_fp16")]; + string _SplitHeadsQ__mh_w_799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_799_equation_0, values = (var_7089_cast_fp16, var_7008_cast_fp16))[name = string("_SplitHeadsQ__mh_w_799_cast_fp16")]; + fp16 var_7330_to_fp16 = const()[name = string("op_7330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_641_cast_fp16, y = var_7330_to_fp16)[name = string("aw_chunk_641_cast_fp16")]; + fp16 var_7332_to_fp16 = const()[name = string("op_7332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_643_cast_fp16, y = var_7332_to_fp16)[name = string("aw_chunk_643_cast_fp16")]; + fp16 var_7334_to_fp16 = const()[name = string("op_7334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_645_cast_fp16, y = var_7334_to_fp16)[name = string("aw_chunk_645_cast_fp16")]; + fp16 var_7336_to_fp16 = const()[name = string("op_7336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_647_cast_fp16, y = var_7336_to_fp16)[name = string("aw_chunk_647_cast_fp16")]; + fp16 var_7338_to_fp16 = const()[name = string("op_7338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_649_cast_fp16, y = var_7338_to_fp16)[name = string("aw_chunk_649_cast_fp16")]; + fp16 var_7340_to_fp16 = const()[name = string("op_7340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_651_cast_fp16, y = var_7340_to_fp16)[name = string("aw_chunk_651_cast_fp16")]; + fp16 var_7342_to_fp16 = const()[name = string("op_7342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_653_cast_fp16, y = var_7342_to_fp16)[name = string("aw_chunk_653_cast_fp16")]; + fp16 var_7344_to_fp16 = const()[name = string("op_7344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_655_cast_fp16, y = var_7344_to_fp16)[name = string("aw_chunk_655_cast_fp16")]; + fp16 var_7346_to_fp16 = const()[name = string("op_7346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_657_cast_fp16, y = var_7346_to_fp16)[name = string("aw_chunk_657_cast_fp16")]; + fp16 var_7348_to_fp16 = const()[name = string("op_7348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_659_cast_fp16, y = var_7348_to_fp16)[name = string("aw_chunk_659_cast_fp16")]; + fp16 var_7350_to_fp16 = const()[name = string("op_7350_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_661_cast_fp16, y = var_7350_to_fp16)[name = string("aw_chunk_661_cast_fp16")]; + fp16 var_7352_to_fp16 = const()[name = string("op_7352_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_663_cast_fp16, y = var_7352_to_fp16)[name = string("aw_chunk_663_cast_fp16")]; + fp16 var_7354_to_fp16 = const()[name = string("op_7354_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_665_cast_fp16, y = var_7354_to_fp16)[name = string("aw_chunk_665_cast_fp16")]; + fp16 var_7356_to_fp16 = const()[name = string("op_7356_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_667_cast_fp16, y = var_7356_to_fp16)[name = string("aw_chunk_667_cast_fp16")]; + fp16 var_7358_to_fp16 = const()[name = string("op_7358_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_669_cast_fp16, y = var_7358_to_fp16)[name = string("aw_chunk_669_cast_fp16")]; + fp16 var_7360_to_fp16 = const()[name = string("op_7360_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_671_cast_fp16, y = var_7360_to_fp16)[name = string("aw_chunk_671_cast_fp16")]; + fp16 var_7362_to_fp16 = const()[name = string("op_7362_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_673_cast_fp16, y = var_7362_to_fp16)[name = string("aw_chunk_673_cast_fp16")]; + fp16 var_7364_to_fp16 = const()[name = string("op_7364_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_675_cast_fp16, y = var_7364_to_fp16)[name = string("aw_chunk_675_cast_fp16")]; + fp16 var_7366_to_fp16 = const()[name = string("op_7366_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_677_cast_fp16, y = var_7366_to_fp16)[name = string("aw_chunk_677_cast_fp16")]; + fp16 var_7368_to_fp16 = const()[name = string("op_7368_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_679_cast_fp16, y = var_7368_to_fp16)[name = string("aw_chunk_679_cast_fp16")]; + fp16 var_7370_to_fp16 = const()[name = string("op_7370_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_681_cast_fp16, y = var_7370_to_fp16)[name = string("aw_chunk_681_cast_fp16")]; + fp16 var_7372_to_fp16 = const()[name = string("op_7372_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_683_cast_fp16, y = var_7372_to_fp16)[name = string("aw_chunk_683_cast_fp16")]; + fp16 var_7374_to_fp16 = const()[name = string("op_7374_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_685_cast_fp16, y = var_7374_to_fp16)[name = string("aw_chunk_685_cast_fp16")]; + fp16 var_7376_to_fp16 = const()[name = string("op_7376_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_687_cast_fp16, y = var_7376_to_fp16)[name = string("aw_chunk_687_cast_fp16")]; + fp16 var_7378_to_fp16 = const()[name = string("op_7378_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_689_cast_fp16, y = var_7378_to_fp16)[name = string("aw_chunk_689_cast_fp16")]; + fp16 var_7380_to_fp16 = const()[name = string("op_7380_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_691_cast_fp16, y = var_7380_to_fp16)[name = string("aw_chunk_691_cast_fp16")]; + fp16 var_7382_to_fp16 = const()[name = string("op_7382_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_693_cast_fp16, y = var_7382_to_fp16)[name = string("aw_chunk_693_cast_fp16")]; + fp16 var_7384_to_fp16 = const()[name = string("op_7384_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_695_cast_fp16, y = var_7384_to_fp16)[name = string("aw_chunk_695_cast_fp16")]; + fp16 var_7386_to_fp16 = const()[name = string("op_7386_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_697_cast_fp16, y = var_7386_to_fp16)[name = string("aw_chunk_697_cast_fp16")]; + fp16 var_7388_to_fp16 = const()[name = string("op_7388_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_699_cast_fp16, y = var_7388_to_fp16)[name = string("aw_chunk_699_cast_fp16")]; + fp16 var_7390_to_fp16 = const()[name = string("op_7390_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_701_cast_fp16, y = var_7390_to_fp16)[name = string("aw_chunk_701_cast_fp16")]; + fp16 var_7392_to_fp16 = const()[name = string("op_7392_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_703_cast_fp16, y = var_7392_to_fp16)[name = string("aw_chunk_703_cast_fp16")]; + fp16 var_7394_to_fp16 = const()[name = string("op_7394_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_705_cast_fp16, y = var_7394_to_fp16)[name = string("aw_chunk_705_cast_fp16")]; + fp16 var_7396_to_fp16 = const()[name = string("op_7396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_707_cast_fp16, y = var_7396_to_fp16)[name = string("aw_chunk_707_cast_fp16")]; + fp16 var_7398_to_fp16 = const()[name = string("op_7398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_709_cast_fp16, y = var_7398_to_fp16)[name = string("aw_chunk_709_cast_fp16")]; + fp16 var_7400_to_fp16 = const()[name = string("op_7400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_711_cast_fp16, y = var_7400_to_fp16)[name = string("aw_chunk_711_cast_fp16")]; + fp16 var_7402_to_fp16 = const()[name = string("op_7402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_713_cast_fp16, y = var_7402_to_fp16)[name = string("aw_chunk_713_cast_fp16")]; + fp16 var_7404_to_fp16 = const()[name = string("op_7404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_715_cast_fp16, y = var_7404_to_fp16)[name = string("aw_chunk_715_cast_fp16")]; + fp16 var_7406_to_fp16 = const()[name = string("op_7406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_717_cast_fp16, y = var_7406_to_fp16)[name = string("aw_chunk_717_cast_fp16")]; + fp16 var_7408_to_fp16 = const()[name = string("op_7408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_719_cast_fp16, y = var_7408_to_fp16)[name = string("aw_chunk_719_cast_fp16")]; + fp16 var_7410_to_fp16 = const()[name = string("op_7410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_721_cast_fp16, y = var_7410_to_fp16)[name = string("aw_chunk_721_cast_fp16")]; + fp16 var_7412_to_fp16 = const()[name = string("op_7412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_723_cast_fp16, y = var_7412_to_fp16)[name = string("aw_chunk_723_cast_fp16")]; + fp16 var_7414_to_fp16 = const()[name = string("op_7414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_725_cast_fp16, y = var_7414_to_fp16)[name = string("aw_chunk_725_cast_fp16")]; + fp16 var_7416_to_fp16 = const()[name = string("op_7416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_727_cast_fp16, y = var_7416_to_fp16)[name = string("aw_chunk_727_cast_fp16")]; + fp16 var_7418_to_fp16 = const()[name = string("op_7418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_729_cast_fp16, y = var_7418_to_fp16)[name = string("aw_chunk_729_cast_fp16")]; + fp16 var_7420_to_fp16 = const()[name = string("op_7420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_731_cast_fp16, y = var_7420_to_fp16)[name = string("aw_chunk_731_cast_fp16")]; + fp16 var_7422_to_fp16 = const()[name = string("op_7422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_733_cast_fp16, y = var_7422_to_fp16)[name = string("aw_chunk_733_cast_fp16")]; + fp16 var_7424_to_fp16 = const()[name = string("op_7424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_735_cast_fp16, y = var_7424_to_fp16)[name = string("aw_chunk_735_cast_fp16")]; + fp16 var_7426_to_fp16 = const()[name = string("op_7426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_737_cast_fp16, y = var_7426_to_fp16)[name = string("aw_chunk_737_cast_fp16")]; + fp16 var_7428_to_fp16 = const()[name = string("op_7428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_739_cast_fp16, y = var_7428_to_fp16)[name = string("aw_chunk_739_cast_fp16")]; + fp16 var_7430_to_fp16 = const()[name = string("op_7430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_741_cast_fp16, y = var_7430_to_fp16)[name = string("aw_chunk_741_cast_fp16")]; + fp16 var_7432_to_fp16 = const()[name = string("op_7432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_743_cast_fp16, y = var_7432_to_fp16)[name = string("aw_chunk_743_cast_fp16")]; + fp16 var_7434_to_fp16 = const()[name = string("op_7434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_745_cast_fp16, y = var_7434_to_fp16)[name = string("aw_chunk_745_cast_fp16")]; + fp16 var_7436_to_fp16 = const()[name = string("op_7436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_747_cast_fp16, y = var_7436_to_fp16)[name = string("aw_chunk_747_cast_fp16")]; + fp16 var_7438_to_fp16 = const()[name = string("op_7438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_749_cast_fp16, y = var_7438_to_fp16)[name = string("aw_chunk_749_cast_fp16")]; + fp16 var_7440_to_fp16 = const()[name = string("op_7440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_751_cast_fp16, y = var_7440_to_fp16)[name = string("aw_chunk_751_cast_fp16")]; + fp16 var_7442_to_fp16 = const()[name = string("op_7442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_753_cast_fp16, y = var_7442_to_fp16)[name = string("aw_chunk_753_cast_fp16")]; + fp16 var_7444_to_fp16 = const()[name = string("op_7444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_755_cast_fp16, y = var_7444_to_fp16)[name = string("aw_chunk_755_cast_fp16")]; + fp16 var_7446_to_fp16 = const()[name = string("op_7446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_757_cast_fp16, y = var_7446_to_fp16)[name = string("aw_chunk_757_cast_fp16")]; + fp16 var_7448_to_fp16 = const()[name = string("op_7448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_759_cast_fp16, y = var_7448_to_fp16)[name = string("aw_chunk_759_cast_fp16")]; + fp16 var_7450_to_fp16 = const()[name = string("op_7450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_761_cast_fp16, y = var_7450_to_fp16)[name = string("aw_chunk_761_cast_fp16")]; + fp16 var_7452_to_fp16 = const()[name = string("op_7452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_763_cast_fp16, y = var_7452_to_fp16)[name = string("aw_chunk_763_cast_fp16")]; + fp16 var_7454_to_fp16 = const()[name = string("op_7454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_765_cast_fp16, y = var_7454_to_fp16)[name = string("aw_chunk_765_cast_fp16")]; + fp16 var_7456_to_fp16 = const()[name = string("op_7456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_767_cast_fp16, y = var_7456_to_fp16)[name = string("aw_chunk_767_cast_fp16")]; + fp16 var_7458_to_fp16 = const()[name = string("op_7458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_769_cast_fp16, y = var_7458_to_fp16)[name = string("aw_chunk_769_cast_fp16")]; + fp16 var_7460_to_fp16 = const()[name = string("op_7460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_771_cast_fp16, y = var_7460_to_fp16)[name = string("aw_chunk_771_cast_fp16")]; + fp16 var_7462_to_fp16 = const()[name = string("op_7462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_773_cast_fp16, y = var_7462_to_fp16)[name = string("aw_chunk_773_cast_fp16")]; + fp16 var_7464_to_fp16 = const()[name = string("op_7464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_775_cast_fp16, y = var_7464_to_fp16)[name = string("aw_chunk_775_cast_fp16")]; + fp16 var_7466_to_fp16 = const()[name = string("op_7466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_777_cast_fp16, y = var_7466_to_fp16)[name = string("aw_chunk_777_cast_fp16")]; + fp16 var_7468_to_fp16 = const()[name = string("op_7468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_779_cast_fp16, y = var_7468_to_fp16)[name = string("aw_chunk_779_cast_fp16")]; + fp16 var_7470_to_fp16 = const()[name = string("op_7470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_781_cast_fp16, y = var_7470_to_fp16)[name = string("aw_chunk_781_cast_fp16")]; + fp16 var_7472_to_fp16 = const()[name = string("op_7472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_783_cast_fp16, y = var_7472_to_fp16)[name = string("aw_chunk_783_cast_fp16")]; + fp16 var_7474_to_fp16 = const()[name = string("op_7474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_785_cast_fp16, y = var_7474_to_fp16)[name = string("aw_chunk_785_cast_fp16")]; + fp16 var_7476_to_fp16 = const()[name = string("op_7476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_787_cast_fp16, y = var_7476_to_fp16)[name = string("aw_chunk_787_cast_fp16")]; + fp16 var_7478_to_fp16 = const()[name = string("op_7478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_789_cast_fp16, y = var_7478_to_fp16)[name = string("aw_chunk_789_cast_fp16")]; + fp16 var_7480_to_fp16 = const()[name = string("op_7480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_791_cast_fp16, y = var_7480_to_fp16)[name = string("aw_chunk_791_cast_fp16")]; + fp16 var_7482_to_fp16 = const()[name = string("op_7482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_793_cast_fp16, y = var_7482_to_fp16)[name = string("aw_chunk_793_cast_fp16")]; + fp16 var_7484_to_fp16 = const()[name = string("op_7484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_795_cast_fp16, y = var_7484_to_fp16)[name = string("aw_chunk_795_cast_fp16")]; + fp16 var_7486_to_fp16 = const()[name = string("op_7486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_797_cast_fp16, y = var_7486_to_fp16)[name = string("aw_chunk_797_cast_fp16")]; + fp16 var_7488_to_fp16 = const()[name = string("op_7488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_799_cast_fp16, y = var_7488_to_fp16)[name = string("aw_chunk_799_cast_fp16")]; + tensor var_7490_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_641_cast_fp16)[name = string("op_7490_cast_fp16")]; + tensor var_7491_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_643_cast_fp16)[name = string("op_7491_cast_fp16")]; + tensor var_7492_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_645_cast_fp16)[name = string("op_7492_cast_fp16")]; + tensor var_7493_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_647_cast_fp16)[name = string("op_7493_cast_fp16")]; + tensor var_7494_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_649_cast_fp16)[name = string("op_7494_cast_fp16")]; + tensor var_7495_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_651_cast_fp16)[name = string("op_7495_cast_fp16")]; + tensor var_7496_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_653_cast_fp16)[name = string("op_7496_cast_fp16")]; + tensor var_7497_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_655_cast_fp16)[name = string("op_7497_cast_fp16")]; + tensor var_7498_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_657_cast_fp16)[name = string("op_7498_cast_fp16")]; + tensor var_7499_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_659_cast_fp16)[name = string("op_7499_cast_fp16")]; + tensor var_7500_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_661_cast_fp16)[name = string("op_7500_cast_fp16")]; + tensor var_7501_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_663_cast_fp16)[name = string("op_7501_cast_fp16")]; + tensor var_7502_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_665_cast_fp16)[name = string("op_7502_cast_fp16")]; + tensor var_7503_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_667_cast_fp16)[name = string("op_7503_cast_fp16")]; + tensor var_7504_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_669_cast_fp16)[name = string("op_7504_cast_fp16")]; + tensor var_7505_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_671_cast_fp16)[name = string("op_7505_cast_fp16")]; + tensor var_7506_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_673_cast_fp16)[name = string("op_7506_cast_fp16")]; + tensor var_7507_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_675_cast_fp16)[name = string("op_7507_cast_fp16")]; + tensor var_7508_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_677_cast_fp16)[name = string("op_7508_cast_fp16")]; + tensor var_7509_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_679_cast_fp16)[name = string("op_7509_cast_fp16")]; + tensor var_7510_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_681_cast_fp16)[name = string("op_7510_cast_fp16")]; + tensor var_7511_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_683_cast_fp16)[name = string("op_7511_cast_fp16")]; + tensor var_7512_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_685_cast_fp16)[name = string("op_7512_cast_fp16")]; + tensor var_7513_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_687_cast_fp16)[name = string("op_7513_cast_fp16")]; + tensor var_7514_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_689_cast_fp16)[name = string("op_7514_cast_fp16")]; + tensor var_7515_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_691_cast_fp16)[name = string("op_7515_cast_fp16")]; + tensor var_7516_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_693_cast_fp16)[name = string("op_7516_cast_fp16")]; + tensor var_7517_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_695_cast_fp16)[name = string("op_7517_cast_fp16")]; + tensor var_7518_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_697_cast_fp16)[name = string("op_7518_cast_fp16")]; + tensor var_7519_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_699_cast_fp16)[name = string("op_7519_cast_fp16")]; + tensor var_7520_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_701_cast_fp16)[name = string("op_7520_cast_fp16")]; + tensor var_7521_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_703_cast_fp16)[name = string("op_7521_cast_fp16")]; + tensor var_7522_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_705_cast_fp16)[name = string("op_7522_cast_fp16")]; + tensor var_7523_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_707_cast_fp16)[name = string("op_7523_cast_fp16")]; + tensor var_7524_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_709_cast_fp16)[name = string("op_7524_cast_fp16")]; + tensor var_7525_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_711_cast_fp16)[name = string("op_7525_cast_fp16")]; + tensor var_7526_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_713_cast_fp16)[name = string("op_7526_cast_fp16")]; + tensor var_7527_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_715_cast_fp16)[name = string("op_7527_cast_fp16")]; + tensor var_7528_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_717_cast_fp16)[name = string("op_7528_cast_fp16")]; + tensor var_7529_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_719_cast_fp16)[name = string("op_7529_cast_fp16")]; + tensor var_7530_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_721_cast_fp16)[name = string("op_7530_cast_fp16")]; + tensor var_7531_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_723_cast_fp16)[name = string("op_7531_cast_fp16")]; + tensor var_7532_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_725_cast_fp16)[name = string("op_7532_cast_fp16")]; + tensor var_7533_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_727_cast_fp16)[name = string("op_7533_cast_fp16")]; + tensor var_7534_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_729_cast_fp16)[name = string("op_7534_cast_fp16")]; + tensor var_7535_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_731_cast_fp16)[name = string("op_7535_cast_fp16")]; + tensor var_7536_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_733_cast_fp16)[name = string("op_7536_cast_fp16")]; + tensor var_7537_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_735_cast_fp16)[name = string("op_7537_cast_fp16")]; + tensor var_7538_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_737_cast_fp16)[name = string("op_7538_cast_fp16")]; + tensor var_7539_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_739_cast_fp16)[name = string("op_7539_cast_fp16")]; + tensor var_7540_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_741_cast_fp16)[name = string("op_7540_cast_fp16")]; + tensor var_7541_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_743_cast_fp16)[name = string("op_7541_cast_fp16")]; + tensor var_7542_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_745_cast_fp16)[name = string("op_7542_cast_fp16")]; + tensor var_7543_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_747_cast_fp16)[name = string("op_7543_cast_fp16")]; + tensor var_7544_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_749_cast_fp16)[name = string("op_7544_cast_fp16")]; + tensor var_7545_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_751_cast_fp16)[name = string("op_7545_cast_fp16")]; + tensor var_7546_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_753_cast_fp16)[name = string("op_7546_cast_fp16")]; + tensor var_7547_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_755_cast_fp16)[name = string("op_7547_cast_fp16")]; + tensor var_7548_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_757_cast_fp16)[name = string("op_7548_cast_fp16")]; + tensor var_7549_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_759_cast_fp16)[name = string("op_7549_cast_fp16")]; + tensor var_7550_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_761_cast_fp16)[name = string("op_7550_cast_fp16")]; + tensor var_7551_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_763_cast_fp16)[name = string("op_7551_cast_fp16")]; + tensor var_7552_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_765_cast_fp16)[name = string("op_7552_cast_fp16")]; + tensor var_7553_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_767_cast_fp16)[name = string("op_7553_cast_fp16")]; + tensor var_7554_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_769_cast_fp16)[name = string("op_7554_cast_fp16")]; + tensor var_7555_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_771_cast_fp16)[name = string("op_7555_cast_fp16")]; + tensor var_7556_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_773_cast_fp16)[name = string("op_7556_cast_fp16")]; + tensor var_7557_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_775_cast_fp16)[name = string("op_7557_cast_fp16")]; + tensor var_7558_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_777_cast_fp16)[name = string("op_7558_cast_fp16")]; + tensor var_7559_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_779_cast_fp16)[name = string("op_7559_cast_fp16")]; + tensor var_7560_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_781_cast_fp16)[name = string("op_7560_cast_fp16")]; + tensor var_7561_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_783_cast_fp16)[name = string("op_7561_cast_fp16")]; + tensor var_7562_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_785_cast_fp16)[name = string("op_7562_cast_fp16")]; + tensor var_7563_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_787_cast_fp16)[name = string("op_7563_cast_fp16")]; + tensor var_7564_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_789_cast_fp16)[name = string("op_7564_cast_fp16")]; + tensor var_7565_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_791_cast_fp16)[name = string("op_7565_cast_fp16")]; + tensor var_7566_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_793_cast_fp16)[name = string("op_7566_cast_fp16")]; + tensor var_7567_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_795_cast_fp16)[name = string("op_7567_cast_fp16")]; + tensor var_7568_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_797_cast_fp16)[name = string("op_7568_cast_fp16")]; + tensor var_7569_cast_fp16 = softmax(axis = var_6315, x = aw_chunk_799_cast_fp16)[name = string("op_7569_cast_fp16")]; + string var_7571_equation_0 = const()[name = string("op_7571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7571_cast_fp16 = einsum(equation = var_7571_equation_0, values = (var_7091_cast_fp16, var_7490_cast_fp16))[name = string("op_7571_cast_fp16")]; + string var_7573_equation_0 = const()[name = string("op_7573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7573_cast_fp16 = einsum(equation = var_7573_equation_0, values = (var_7091_cast_fp16, var_7491_cast_fp16))[name = string("op_7573_cast_fp16")]; + string var_7575_equation_0 = const()[name = string("op_7575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7575_cast_fp16 = einsum(equation = var_7575_equation_0, values = (var_7091_cast_fp16, var_7492_cast_fp16))[name = string("op_7575_cast_fp16")]; + string var_7577_equation_0 = const()[name = string("op_7577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7577_cast_fp16 = einsum(equation = var_7577_equation_0, values = (var_7091_cast_fp16, var_7493_cast_fp16))[name = string("op_7577_cast_fp16")]; + string var_7579_equation_0 = const()[name = string("op_7579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7579_cast_fp16 = einsum(equation = var_7579_equation_0, values = (var_7095_cast_fp16, var_7494_cast_fp16))[name = string("op_7579_cast_fp16")]; + string var_7581_equation_0 = const()[name = string("op_7581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7581_cast_fp16 = einsum(equation = var_7581_equation_0, values = (var_7095_cast_fp16, var_7495_cast_fp16))[name = string("op_7581_cast_fp16")]; + string var_7583_equation_0 = const()[name = string("op_7583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7583_cast_fp16 = einsum(equation = var_7583_equation_0, values = (var_7095_cast_fp16, var_7496_cast_fp16))[name = string("op_7583_cast_fp16")]; + string var_7585_equation_0 = const()[name = string("op_7585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7585_cast_fp16 = einsum(equation = var_7585_equation_0, values = (var_7095_cast_fp16, var_7497_cast_fp16))[name = string("op_7585_cast_fp16")]; + string var_7587_equation_0 = const()[name = string("op_7587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7587_cast_fp16 = einsum(equation = var_7587_equation_0, values = (var_7099_cast_fp16, var_7498_cast_fp16))[name = string("op_7587_cast_fp16")]; + string var_7589_equation_0 = const()[name = string("op_7589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7589_cast_fp16 = einsum(equation = var_7589_equation_0, values = (var_7099_cast_fp16, var_7499_cast_fp16))[name = string("op_7589_cast_fp16")]; + string var_7591_equation_0 = const()[name = string("op_7591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7591_cast_fp16 = einsum(equation = var_7591_equation_0, values = (var_7099_cast_fp16, var_7500_cast_fp16))[name = string("op_7591_cast_fp16")]; + string var_7593_equation_0 = const()[name = string("op_7593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7593_cast_fp16 = einsum(equation = var_7593_equation_0, values = (var_7099_cast_fp16, var_7501_cast_fp16))[name = string("op_7593_cast_fp16")]; + string var_7595_equation_0 = const()[name = string("op_7595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7595_cast_fp16 = einsum(equation = var_7595_equation_0, values = (var_7103_cast_fp16, var_7502_cast_fp16))[name = string("op_7595_cast_fp16")]; + string var_7597_equation_0 = const()[name = string("op_7597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7597_cast_fp16 = einsum(equation = var_7597_equation_0, values = (var_7103_cast_fp16, var_7503_cast_fp16))[name = string("op_7597_cast_fp16")]; + string var_7599_equation_0 = const()[name = string("op_7599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7599_cast_fp16 = einsum(equation = var_7599_equation_0, values = (var_7103_cast_fp16, var_7504_cast_fp16))[name = string("op_7599_cast_fp16")]; + string var_7601_equation_0 = const()[name = string("op_7601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7601_cast_fp16 = einsum(equation = var_7601_equation_0, values = (var_7103_cast_fp16, var_7505_cast_fp16))[name = string("op_7601_cast_fp16")]; + string var_7603_equation_0 = const()[name = string("op_7603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7603_cast_fp16 = einsum(equation = var_7603_equation_0, values = (var_7107_cast_fp16, var_7506_cast_fp16))[name = string("op_7603_cast_fp16")]; + string var_7605_equation_0 = const()[name = string("op_7605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7605_cast_fp16 = einsum(equation = var_7605_equation_0, values = (var_7107_cast_fp16, var_7507_cast_fp16))[name = string("op_7605_cast_fp16")]; + string var_7607_equation_0 = const()[name = string("op_7607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7607_cast_fp16 = einsum(equation = var_7607_equation_0, values = (var_7107_cast_fp16, var_7508_cast_fp16))[name = string("op_7607_cast_fp16")]; + string var_7609_equation_0 = const()[name = string("op_7609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7609_cast_fp16 = einsum(equation = var_7609_equation_0, values = (var_7107_cast_fp16, var_7509_cast_fp16))[name = string("op_7609_cast_fp16")]; + string var_7611_equation_0 = const()[name = string("op_7611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7611_cast_fp16 = einsum(equation = var_7611_equation_0, values = (var_7111_cast_fp16, var_7510_cast_fp16))[name = string("op_7611_cast_fp16")]; + string var_7613_equation_0 = const()[name = string("op_7613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7613_cast_fp16 = einsum(equation = var_7613_equation_0, values = (var_7111_cast_fp16, var_7511_cast_fp16))[name = string("op_7613_cast_fp16")]; + string var_7615_equation_0 = const()[name = string("op_7615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7615_cast_fp16 = einsum(equation = var_7615_equation_0, values = (var_7111_cast_fp16, var_7512_cast_fp16))[name = string("op_7615_cast_fp16")]; + string var_7617_equation_0 = const()[name = string("op_7617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7617_cast_fp16 = einsum(equation = var_7617_equation_0, values = (var_7111_cast_fp16, var_7513_cast_fp16))[name = string("op_7617_cast_fp16")]; + string var_7619_equation_0 = const()[name = string("op_7619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7619_cast_fp16 = einsum(equation = var_7619_equation_0, values = (var_7115_cast_fp16, var_7514_cast_fp16))[name = string("op_7619_cast_fp16")]; + string var_7621_equation_0 = const()[name = string("op_7621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7621_cast_fp16 = einsum(equation = var_7621_equation_0, values = (var_7115_cast_fp16, var_7515_cast_fp16))[name = string("op_7621_cast_fp16")]; + string var_7623_equation_0 = const()[name = string("op_7623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7623_cast_fp16 = einsum(equation = var_7623_equation_0, values = (var_7115_cast_fp16, var_7516_cast_fp16))[name = string("op_7623_cast_fp16")]; + string var_7625_equation_0 = const()[name = string("op_7625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7625_cast_fp16 = einsum(equation = var_7625_equation_0, values = (var_7115_cast_fp16, var_7517_cast_fp16))[name = string("op_7625_cast_fp16")]; + string var_7627_equation_0 = const()[name = string("op_7627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7627_cast_fp16 = einsum(equation = var_7627_equation_0, values = (var_7119_cast_fp16, var_7518_cast_fp16))[name = string("op_7627_cast_fp16")]; + string var_7629_equation_0 = const()[name = string("op_7629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7629_cast_fp16 = einsum(equation = var_7629_equation_0, values = (var_7119_cast_fp16, var_7519_cast_fp16))[name = string("op_7629_cast_fp16")]; + string var_7631_equation_0 = const()[name = string("op_7631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7631_cast_fp16 = einsum(equation = var_7631_equation_0, values = (var_7119_cast_fp16, var_7520_cast_fp16))[name = string("op_7631_cast_fp16")]; + string var_7633_equation_0 = const()[name = string("op_7633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7633_cast_fp16 = einsum(equation = var_7633_equation_0, values = (var_7119_cast_fp16, var_7521_cast_fp16))[name = string("op_7633_cast_fp16")]; + string var_7635_equation_0 = const()[name = string("op_7635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7635_cast_fp16 = einsum(equation = var_7635_equation_0, values = (var_7123_cast_fp16, var_7522_cast_fp16))[name = string("op_7635_cast_fp16")]; + string var_7637_equation_0 = const()[name = string("op_7637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7637_cast_fp16 = einsum(equation = var_7637_equation_0, values = (var_7123_cast_fp16, var_7523_cast_fp16))[name = string("op_7637_cast_fp16")]; + string var_7639_equation_0 = const()[name = string("op_7639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7639_cast_fp16 = einsum(equation = var_7639_equation_0, values = (var_7123_cast_fp16, var_7524_cast_fp16))[name = string("op_7639_cast_fp16")]; + string var_7641_equation_0 = const()[name = string("op_7641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7641_cast_fp16 = einsum(equation = var_7641_equation_0, values = (var_7123_cast_fp16, var_7525_cast_fp16))[name = string("op_7641_cast_fp16")]; + string var_7643_equation_0 = const()[name = string("op_7643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7643_cast_fp16 = einsum(equation = var_7643_equation_0, values = (var_7127_cast_fp16, var_7526_cast_fp16))[name = string("op_7643_cast_fp16")]; + string var_7645_equation_0 = const()[name = string("op_7645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7645_cast_fp16 = einsum(equation = var_7645_equation_0, values = (var_7127_cast_fp16, var_7527_cast_fp16))[name = string("op_7645_cast_fp16")]; + string var_7647_equation_0 = const()[name = string("op_7647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7647_cast_fp16 = einsum(equation = var_7647_equation_0, values = (var_7127_cast_fp16, var_7528_cast_fp16))[name = string("op_7647_cast_fp16")]; + string var_7649_equation_0 = const()[name = string("op_7649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7649_cast_fp16 = einsum(equation = var_7649_equation_0, values = (var_7127_cast_fp16, var_7529_cast_fp16))[name = string("op_7649_cast_fp16")]; + string var_7651_equation_0 = const()[name = string("op_7651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7651_cast_fp16 = einsum(equation = var_7651_equation_0, values = (var_7131_cast_fp16, var_7530_cast_fp16))[name = string("op_7651_cast_fp16")]; + string var_7653_equation_0 = const()[name = string("op_7653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7653_cast_fp16 = einsum(equation = var_7653_equation_0, values = (var_7131_cast_fp16, var_7531_cast_fp16))[name = string("op_7653_cast_fp16")]; + string var_7655_equation_0 = const()[name = string("op_7655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7655_cast_fp16 = einsum(equation = var_7655_equation_0, values = (var_7131_cast_fp16, var_7532_cast_fp16))[name = string("op_7655_cast_fp16")]; + string var_7657_equation_0 = const()[name = string("op_7657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7657_cast_fp16 = einsum(equation = var_7657_equation_0, values = (var_7131_cast_fp16, var_7533_cast_fp16))[name = string("op_7657_cast_fp16")]; + string var_7659_equation_0 = const()[name = string("op_7659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7659_cast_fp16 = einsum(equation = var_7659_equation_0, values = (var_7135_cast_fp16, var_7534_cast_fp16))[name = string("op_7659_cast_fp16")]; + string var_7661_equation_0 = const()[name = string("op_7661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7661_cast_fp16 = einsum(equation = var_7661_equation_0, values = (var_7135_cast_fp16, var_7535_cast_fp16))[name = string("op_7661_cast_fp16")]; + string var_7663_equation_0 = const()[name = string("op_7663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7663_cast_fp16 = einsum(equation = var_7663_equation_0, values = (var_7135_cast_fp16, var_7536_cast_fp16))[name = string("op_7663_cast_fp16")]; + string var_7665_equation_0 = const()[name = string("op_7665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7665_cast_fp16 = einsum(equation = var_7665_equation_0, values = (var_7135_cast_fp16, var_7537_cast_fp16))[name = string("op_7665_cast_fp16")]; + string var_7667_equation_0 = const()[name = string("op_7667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7667_cast_fp16 = einsum(equation = var_7667_equation_0, values = (var_7139_cast_fp16, var_7538_cast_fp16))[name = string("op_7667_cast_fp16")]; + string var_7669_equation_0 = const()[name = string("op_7669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7669_cast_fp16 = einsum(equation = var_7669_equation_0, values = (var_7139_cast_fp16, var_7539_cast_fp16))[name = string("op_7669_cast_fp16")]; + string var_7671_equation_0 = const()[name = string("op_7671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7671_cast_fp16 = einsum(equation = var_7671_equation_0, values = (var_7139_cast_fp16, var_7540_cast_fp16))[name = string("op_7671_cast_fp16")]; + string var_7673_equation_0 = const()[name = string("op_7673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7673_cast_fp16 = einsum(equation = var_7673_equation_0, values = (var_7139_cast_fp16, var_7541_cast_fp16))[name = string("op_7673_cast_fp16")]; + string var_7675_equation_0 = const()[name = string("op_7675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7675_cast_fp16 = einsum(equation = var_7675_equation_0, values = (var_7143_cast_fp16, var_7542_cast_fp16))[name = string("op_7675_cast_fp16")]; + string var_7677_equation_0 = const()[name = string("op_7677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7677_cast_fp16 = einsum(equation = var_7677_equation_0, values = (var_7143_cast_fp16, var_7543_cast_fp16))[name = string("op_7677_cast_fp16")]; + string var_7679_equation_0 = const()[name = string("op_7679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7679_cast_fp16 = einsum(equation = var_7679_equation_0, values = (var_7143_cast_fp16, var_7544_cast_fp16))[name = string("op_7679_cast_fp16")]; + string var_7681_equation_0 = const()[name = string("op_7681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7681_cast_fp16 = einsum(equation = var_7681_equation_0, values = (var_7143_cast_fp16, var_7545_cast_fp16))[name = string("op_7681_cast_fp16")]; + string var_7683_equation_0 = const()[name = string("op_7683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7683_cast_fp16 = einsum(equation = var_7683_equation_0, values = (var_7147_cast_fp16, var_7546_cast_fp16))[name = string("op_7683_cast_fp16")]; + string var_7685_equation_0 = const()[name = string("op_7685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7685_cast_fp16 = einsum(equation = var_7685_equation_0, values = (var_7147_cast_fp16, var_7547_cast_fp16))[name = string("op_7685_cast_fp16")]; + string var_7687_equation_0 = const()[name = string("op_7687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7687_cast_fp16 = einsum(equation = var_7687_equation_0, values = (var_7147_cast_fp16, var_7548_cast_fp16))[name = string("op_7687_cast_fp16")]; + string var_7689_equation_0 = const()[name = string("op_7689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7689_cast_fp16 = einsum(equation = var_7689_equation_0, values = (var_7147_cast_fp16, var_7549_cast_fp16))[name = string("op_7689_cast_fp16")]; + string var_7691_equation_0 = const()[name = string("op_7691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7691_cast_fp16 = einsum(equation = var_7691_equation_0, values = (var_7151_cast_fp16, var_7550_cast_fp16))[name = string("op_7691_cast_fp16")]; + string var_7693_equation_0 = const()[name = string("op_7693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7693_cast_fp16 = einsum(equation = var_7693_equation_0, values = (var_7151_cast_fp16, var_7551_cast_fp16))[name = string("op_7693_cast_fp16")]; + string var_7695_equation_0 = const()[name = string("op_7695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7695_cast_fp16 = einsum(equation = var_7695_equation_0, values = (var_7151_cast_fp16, var_7552_cast_fp16))[name = string("op_7695_cast_fp16")]; + string var_7697_equation_0 = const()[name = string("op_7697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7697_cast_fp16 = einsum(equation = var_7697_equation_0, values = (var_7151_cast_fp16, var_7553_cast_fp16))[name = string("op_7697_cast_fp16")]; + string var_7699_equation_0 = const()[name = string("op_7699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7699_cast_fp16 = einsum(equation = var_7699_equation_0, values = (var_7155_cast_fp16, var_7554_cast_fp16))[name = string("op_7699_cast_fp16")]; + string var_7701_equation_0 = const()[name = string("op_7701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7701_cast_fp16 = einsum(equation = var_7701_equation_0, values = (var_7155_cast_fp16, var_7555_cast_fp16))[name = string("op_7701_cast_fp16")]; + string var_7703_equation_0 = const()[name = string("op_7703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7703_cast_fp16 = einsum(equation = var_7703_equation_0, values = (var_7155_cast_fp16, var_7556_cast_fp16))[name = string("op_7703_cast_fp16")]; + string var_7705_equation_0 = const()[name = string("op_7705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7705_cast_fp16 = einsum(equation = var_7705_equation_0, values = (var_7155_cast_fp16, var_7557_cast_fp16))[name = string("op_7705_cast_fp16")]; + string var_7707_equation_0 = const()[name = string("op_7707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7707_cast_fp16 = einsum(equation = var_7707_equation_0, values = (var_7159_cast_fp16, var_7558_cast_fp16))[name = string("op_7707_cast_fp16")]; + string var_7709_equation_0 = const()[name = string("op_7709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7709_cast_fp16 = einsum(equation = var_7709_equation_0, values = (var_7159_cast_fp16, var_7559_cast_fp16))[name = string("op_7709_cast_fp16")]; + string var_7711_equation_0 = const()[name = string("op_7711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7711_cast_fp16 = einsum(equation = var_7711_equation_0, values = (var_7159_cast_fp16, var_7560_cast_fp16))[name = string("op_7711_cast_fp16")]; + string var_7713_equation_0 = const()[name = string("op_7713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7713_cast_fp16 = einsum(equation = var_7713_equation_0, values = (var_7159_cast_fp16, var_7561_cast_fp16))[name = string("op_7713_cast_fp16")]; + string var_7715_equation_0 = const()[name = string("op_7715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7715_cast_fp16 = einsum(equation = var_7715_equation_0, values = (var_7163_cast_fp16, var_7562_cast_fp16))[name = string("op_7715_cast_fp16")]; + string var_7717_equation_0 = const()[name = string("op_7717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7717_cast_fp16 = einsum(equation = var_7717_equation_0, values = (var_7163_cast_fp16, var_7563_cast_fp16))[name = string("op_7717_cast_fp16")]; + string var_7719_equation_0 = const()[name = string("op_7719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7719_cast_fp16 = einsum(equation = var_7719_equation_0, values = (var_7163_cast_fp16, var_7564_cast_fp16))[name = string("op_7719_cast_fp16")]; + string var_7721_equation_0 = const()[name = string("op_7721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7721_cast_fp16 = einsum(equation = var_7721_equation_0, values = (var_7163_cast_fp16, var_7565_cast_fp16))[name = string("op_7721_cast_fp16")]; + string var_7723_equation_0 = const()[name = string("op_7723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7723_cast_fp16 = einsum(equation = var_7723_equation_0, values = (var_7167_cast_fp16, var_7566_cast_fp16))[name = string("op_7723_cast_fp16")]; + string var_7725_equation_0 = const()[name = string("op_7725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7725_cast_fp16 = einsum(equation = var_7725_equation_0, values = (var_7167_cast_fp16, var_7567_cast_fp16))[name = string("op_7725_cast_fp16")]; + string var_7727_equation_0 = const()[name = string("op_7727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7727_cast_fp16 = einsum(equation = var_7727_equation_0, values = (var_7167_cast_fp16, var_7568_cast_fp16))[name = string("op_7727_cast_fp16")]; + string var_7729_equation_0 = const()[name = string("op_7729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7729_cast_fp16 = einsum(equation = var_7729_equation_0, values = (var_7167_cast_fp16, var_7569_cast_fp16))[name = string("op_7729_cast_fp16")]; + bool var_7731_interleave_0 = const()[name = string("op_7731_interleave_0"), val = bool(false)]; + tensor var_7731_cast_fp16 = concat(axis = var_6290, interleave = var_7731_interleave_0, values = (var_7571_cast_fp16, var_7573_cast_fp16, var_7575_cast_fp16, var_7577_cast_fp16))[name = string("op_7731_cast_fp16")]; + bool var_7733_interleave_0 = const()[name = string("op_7733_interleave_0"), val = bool(false)]; + tensor var_7733_cast_fp16 = concat(axis = var_6290, interleave = var_7733_interleave_0, values = (var_7579_cast_fp16, var_7581_cast_fp16, var_7583_cast_fp16, var_7585_cast_fp16))[name = string("op_7733_cast_fp16")]; + bool var_7735_interleave_0 = const()[name = string("op_7735_interleave_0"), val = bool(false)]; + tensor var_7735_cast_fp16 = concat(axis = var_6290, interleave = var_7735_interleave_0, values = (var_7587_cast_fp16, var_7589_cast_fp16, var_7591_cast_fp16, var_7593_cast_fp16))[name = string("op_7735_cast_fp16")]; + bool var_7737_interleave_0 = const()[name = string("op_7737_interleave_0"), val = bool(false)]; + tensor var_7737_cast_fp16 = concat(axis = var_6290, interleave = var_7737_interleave_0, values = (var_7595_cast_fp16, var_7597_cast_fp16, var_7599_cast_fp16, var_7601_cast_fp16))[name = string("op_7737_cast_fp16")]; + bool var_7739_interleave_0 = const()[name = string("op_7739_interleave_0"), val = bool(false)]; + tensor var_7739_cast_fp16 = concat(axis = var_6290, interleave = var_7739_interleave_0, values = (var_7603_cast_fp16, var_7605_cast_fp16, var_7607_cast_fp16, var_7609_cast_fp16))[name = string("op_7739_cast_fp16")]; + bool var_7741_interleave_0 = const()[name = string("op_7741_interleave_0"), val = bool(false)]; + tensor var_7741_cast_fp16 = concat(axis = var_6290, interleave = var_7741_interleave_0, values = (var_7611_cast_fp16, var_7613_cast_fp16, var_7615_cast_fp16, var_7617_cast_fp16))[name = string("op_7741_cast_fp16")]; + bool var_7743_interleave_0 = const()[name = string("op_7743_interleave_0"), val = bool(false)]; + tensor var_7743_cast_fp16 = concat(axis = var_6290, interleave = var_7743_interleave_0, values = (var_7619_cast_fp16, var_7621_cast_fp16, var_7623_cast_fp16, var_7625_cast_fp16))[name = string("op_7743_cast_fp16")]; + bool var_7745_interleave_0 = const()[name = string("op_7745_interleave_0"), val = bool(false)]; + tensor var_7745_cast_fp16 = concat(axis = var_6290, interleave = var_7745_interleave_0, values = (var_7627_cast_fp16, var_7629_cast_fp16, var_7631_cast_fp16, var_7633_cast_fp16))[name = string("op_7745_cast_fp16")]; + bool var_7747_interleave_0 = const()[name = string("op_7747_interleave_0"), val = bool(false)]; + tensor var_7747_cast_fp16 = concat(axis = var_6290, interleave = var_7747_interleave_0, values = (var_7635_cast_fp16, var_7637_cast_fp16, var_7639_cast_fp16, var_7641_cast_fp16))[name = string("op_7747_cast_fp16")]; + bool var_7749_interleave_0 = const()[name = string("op_7749_interleave_0"), val = bool(false)]; + tensor var_7749_cast_fp16 = concat(axis = var_6290, interleave = var_7749_interleave_0, values = (var_7643_cast_fp16, var_7645_cast_fp16, var_7647_cast_fp16, var_7649_cast_fp16))[name = string("op_7749_cast_fp16")]; + bool var_7751_interleave_0 = const()[name = string("op_7751_interleave_0"), val = bool(false)]; + tensor var_7751_cast_fp16 = concat(axis = var_6290, interleave = var_7751_interleave_0, values = (var_7651_cast_fp16, var_7653_cast_fp16, var_7655_cast_fp16, var_7657_cast_fp16))[name = string("op_7751_cast_fp16")]; + bool var_7753_interleave_0 = const()[name = string("op_7753_interleave_0"), val = bool(false)]; + tensor var_7753_cast_fp16 = concat(axis = var_6290, interleave = var_7753_interleave_0, values = (var_7659_cast_fp16, var_7661_cast_fp16, var_7663_cast_fp16, var_7665_cast_fp16))[name = string("op_7753_cast_fp16")]; + bool var_7755_interleave_0 = const()[name = string("op_7755_interleave_0"), val = bool(false)]; + tensor var_7755_cast_fp16 = concat(axis = var_6290, interleave = var_7755_interleave_0, values = (var_7667_cast_fp16, var_7669_cast_fp16, var_7671_cast_fp16, var_7673_cast_fp16))[name = string("op_7755_cast_fp16")]; + bool var_7757_interleave_0 = const()[name = string("op_7757_interleave_0"), val = bool(false)]; + tensor var_7757_cast_fp16 = concat(axis = var_6290, interleave = var_7757_interleave_0, values = (var_7675_cast_fp16, var_7677_cast_fp16, var_7679_cast_fp16, var_7681_cast_fp16))[name = string("op_7757_cast_fp16")]; + bool var_7759_interleave_0 = const()[name = string("op_7759_interleave_0"), val = bool(false)]; + tensor var_7759_cast_fp16 = concat(axis = var_6290, interleave = var_7759_interleave_0, values = (var_7683_cast_fp16, var_7685_cast_fp16, var_7687_cast_fp16, var_7689_cast_fp16))[name = string("op_7759_cast_fp16")]; + bool var_7761_interleave_0 = const()[name = string("op_7761_interleave_0"), val = bool(false)]; + tensor var_7761_cast_fp16 = concat(axis = var_6290, interleave = var_7761_interleave_0, values = (var_7691_cast_fp16, var_7693_cast_fp16, var_7695_cast_fp16, var_7697_cast_fp16))[name = string("op_7761_cast_fp16")]; + bool var_7763_interleave_0 = const()[name = string("op_7763_interleave_0"), val = bool(false)]; + tensor var_7763_cast_fp16 = concat(axis = var_6290, interleave = var_7763_interleave_0, values = (var_7699_cast_fp16, var_7701_cast_fp16, var_7703_cast_fp16, var_7705_cast_fp16))[name = string("op_7763_cast_fp16")]; + bool var_7765_interleave_0 = const()[name = string("op_7765_interleave_0"), val = bool(false)]; + tensor var_7765_cast_fp16 = concat(axis = var_6290, interleave = var_7765_interleave_0, values = (var_7707_cast_fp16, var_7709_cast_fp16, var_7711_cast_fp16, var_7713_cast_fp16))[name = string("op_7765_cast_fp16")]; + bool var_7767_interleave_0 = const()[name = string("op_7767_interleave_0"), val = bool(false)]; + tensor var_7767_cast_fp16 = concat(axis = var_6290, interleave = var_7767_interleave_0, values = (var_7715_cast_fp16, var_7717_cast_fp16, var_7719_cast_fp16, var_7721_cast_fp16))[name = string("op_7767_cast_fp16")]; + bool var_7769_interleave_0 = const()[name = string("op_7769_interleave_0"), val = bool(false)]; + tensor var_7769_cast_fp16 = concat(axis = var_6290, interleave = var_7769_interleave_0, values = (var_7723_cast_fp16, var_7725_cast_fp16, var_7727_cast_fp16, var_7729_cast_fp16))[name = string("op_7769_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_6315, interleave = input_33_interleave_0, values = (var_7731_cast_fp16, var_7733_cast_fp16, var_7735_cast_fp16, var_7737_cast_fp16, var_7739_cast_fp16, var_7741_cast_fp16, var_7743_cast_fp16, var_7745_cast_fp16, var_7747_cast_fp16, var_7749_cast_fp16, var_7751_cast_fp16, var_7753_cast_fp16, var_7755_cast_fp16, var_7757_cast_fp16, var_7759_cast_fp16, var_7761_cast_fp16, var_7763_cast_fp16, var_7765_cast_fp16, var_7767_cast_fp16, var_7769_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181918400)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185195264)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_7788_to_fp16 = const()[name = string("op_7788_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_7788_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185197888)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185200512)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185203136)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198310400)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198320704)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211427968)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_7817 = const()[name = string("op_7817"), val = int32(3)]; + int32 var_7842 = const()[name = string("op_7842"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_7859_to_fp16 = const()[name = string("op_7859_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_7859_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211430592)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211433216)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211435840)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214712704)))]; + tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; + string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; + tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; + tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; + int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214715328)))]; + tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; + string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; + tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; + tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; + int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217992192)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221269056)))]; + tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_7897_begin_0 = const()[name = string("op_7897_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7897_end_0 = const()[name = string("op_7897_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7897_end_mask_0 = const()[name = string("op_7897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7897_cast_fp16 = slice_by_index(begin = var_7897_begin_0, end = var_7897_end_0, end_mask = var_7897_end_mask_0, x = query_11_cast_fp16)[name = string("op_7897_cast_fp16")]; + tensor var_7901_begin_0 = const()[name = string("op_7901_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7901_end_0 = const()[name = string("op_7901_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7901_end_mask_0 = const()[name = string("op_7901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7901_cast_fp16 = slice_by_index(begin = var_7901_begin_0, end = var_7901_end_0, end_mask = var_7901_end_mask_0, x = query_11_cast_fp16)[name = string("op_7901_cast_fp16")]; + tensor var_7905_begin_0 = const()[name = string("op_7905_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7905_end_0 = const()[name = string("op_7905_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7905_end_mask_0 = const()[name = string("op_7905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7905_cast_fp16 = slice_by_index(begin = var_7905_begin_0, end = var_7905_end_0, end_mask = var_7905_end_mask_0, x = query_11_cast_fp16)[name = string("op_7905_cast_fp16")]; + tensor var_7909_begin_0 = const()[name = string("op_7909_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7909_end_0 = const()[name = string("op_7909_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7909_end_mask_0 = const()[name = string("op_7909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7909_cast_fp16 = slice_by_index(begin = var_7909_begin_0, end = var_7909_end_0, end_mask = var_7909_end_mask_0, x = query_11_cast_fp16)[name = string("op_7909_cast_fp16")]; + tensor var_7913_begin_0 = const()[name = string("op_7913_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7913_end_0 = const()[name = string("op_7913_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7913_end_mask_0 = const()[name = string("op_7913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7913_cast_fp16 = slice_by_index(begin = var_7913_begin_0, end = var_7913_end_0, end_mask = var_7913_end_mask_0, x = query_11_cast_fp16)[name = string("op_7913_cast_fp16")]; + tensor var_7917_begin_0 = const()[name = string("op_7917_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7917_end_0 = const()[name = string("op_7917_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7917_end_mask_0 = const()[name = string("op_7917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7917_cast_fp16 = slice_by_index(begin = var_7917_begin_0, end = var_7917_end_0, end_mask = var_7917_end_mask_0, x = query_11_cast_fp16)[name = string("op_7917_cast_fp16")]; + tensor var_7921_begin_0 = const()[name = string("op_7921_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7921_end_0 = const()[name = string("op_7921_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7921_end_mask_0 = const()[name = string("op_7921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7921_cast_fp16 = slice_by_index(begin = var_7921_begin_0, end = var_7921_end_0, end_mask = var_7921_end_mask_0, x = query_11_cast_fp16)[name = string("op_7921_cast_fp16")]; + tensor var_7925_begin_0 = const()[name = string("op_7925_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7925_end_0 = const()[name = string("op_7925_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7925_end_mask_0 = const()[name = string("op_7925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7925_cast_fp16 = slice_by_index(begin = var_7925_begin_0, end = var_7925_end_0, end_mask = var_7925_end_mask_0, x = query_11_cast_fp16)[name = string("op_7925_cast_fp16")]; + tensor var_7929_begin_0 = const()[name = string("op_7929_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7929_end_0 = const()[name = string("op_7929_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7929_end_mask_0 = const()[name = string("op_7929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7929_cast_fp16 = slice_by_index(begin = var_7929_begin_0, end = var_7929_end_0, end_mask = var_7929_end_mask_0, x = query_11_cast_fp16)[name = string("op_7929_cast_fp16")]; + tensor var_7933_begin_0 = const()[name = string("op_7933_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7933_end_0 = const()[name = string("op_7933_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7933_end_mask_0 = const()[name = string("op_7933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7933_cast_fp16 = slice_by_index(begin = var_7933_begin_0, end = var_7933_end_0, end_mask = var_7933_end_mask_0, x = query_11_cast_fp16)[name = string("op_7933_cast_fp16")]; + tensor var_7937_begin_0 = const()[name = string("op_7937_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7937_end_0 = const()[name = string("op_7937_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7937_end_mask_0 = const()[name = string("op_7937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7937_cast_fp16 = slice_by_index(begin = var_7937_begin_0, end = var_7937_end_0, end_mask = var_7937_end_mask_0, x = query_11_cast_fp16)[name = string("op_7937_cast_fp16")]; + tensor var_7941_begin_0 = const()[name = string("op_7941_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7941_end_0 = const()[name = string("op_7941_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7941_end_mask_0 = const()[name = string("op_7941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7941_cast_fp16 = slice_by_index(begin = var_7941_begin_0, end = var_7941_end_0, end_mask = var_7941_end_mask_0, x = query_11_cast_fp16)[name = string("op_7941_cast_fp16")]; + tensor var_7945_begin_0 = const()[name = string("op_7945_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7945_end_0 = const()[name = string("op_7945_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_7945_end_mask_0 = const()[name = string("op_7945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7945_cast_fp16 = slice_by_index(begin = var_7945_begin_0, end = var_7945_end_0, end_mask = var_7945_end_mask_0, x = query_11_cast_fp16)[name = string("op_7945_cast_fp16")]; + tensor var_7949_begin_0 = const()[name = string("op_7949_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7949_end_0 = const()[name = string("op_7949_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_7949_end_mask_0 = const()[name = string("op_7949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7949_cast_fp16 = slice_by_index(begin = var_7949_begin_0, end = var_7949_end_0, end_mask = var_7949_end_mask_0, x = query_11_cast_fp16)[name = string("op_7949_cast_fp16")]; + tensor var_7953_begin_0 = const()[name = string("op_7953_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7953_end_0 = const()[name = string("op_7953_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_7953_end_mask_0 = const()[name = string("op_7953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7953_cast_fp16 = slice_by_index(begin = var_7953_begin_0, end = var_7953_end_0, end_mask = var_7953_end_mask_0, x = query_11_cast_fp16)[name = string("op_7953_cast_fp16")]; + tensor var_7957_begin_0 = const()[name = string("op_7957_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7957_end_0 = const()[name = string("op_7957_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_7957_end_mask_0 = const()[name = string("op_7957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7957_cast_fp16 = slice_by_index(begin = var_7957_begin_0, end = var_7957_end_0, end_mask = var_7957_end_mask_0, x = query_11_cast_fp16)[name = string("op_7957_cast_fp16")]; + tensor var_7961_begin_0 = const()[name = string("op_7961_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7961_end_0 = const()[name = string("op_7961_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_7961_end_mask_0 = const()[name = string("op_7961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7961_cast_fp16 = slice_by_index(begin = var_7961_begin_0, end = var_7961_end_0, end_mask = var_7961_end_mask_0, x = query_11_cast_fp16)[name = string("op_7961_cast_fp16")]; + tensor var_7965_begin_0 = const()[name = string("op_7965_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7965_end_0 = const()[name = string("op_7965_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_7965_end_mask_0 = const()[name = string("op_7965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7965_cast_fp16 = slice_by_index(begin = var_7965_begin_0, end = var_7965_end_0, end_mask = var_7965_end_mask_0, x = query_11_cast_fp16)[name = string("op_7965_cast_fp16")]; + tensor var_7969_begin_0 = const()[name = string("op_7969_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7969_end_0 = const()[name = string("op_7969_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_7969_end_mask_0 = const()[name = string("op_7969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7969_cast_fp16 = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = query_11_cast_fp16)[name = string("op_7969_cast_fp16")]; + tensor var_7973_begin_0 = const()[name = string("op_7973_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7973_end_0 = const()[name = string("op_7973_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_7973_end_mask_0 = const()[name = string("op_7973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7973_cast_fp16 = slice_by_index(begin = var_7973_begin_0, end = var_7973_end_0, end_mask = var_7973_end_mask_0, x = query_11_cast_fp16)[name = string("op_7973_cast_fp16")]; + tensor var_7982_begin_0 = const()[name = string("op_7982_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7982_end_0 = const()[name = string("op_7982_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7982_end_mask_0 = const()[name = string("op_7982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7982_cast_fp16 = slice_by_index(begin = var_7982_begin_0, end = var_7982_end_0, end_mask = var_7982_end_mask_0, x = var_7897_cast_fp16)[name = string("op_7982_cast_fp16")]; + tensor var_7989_begin_0 = const()[name = string("op_7989_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7989_end_0 = const()[name = string("op_7989_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7989_end_mask_0 = const()[name = string("op_7989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7989_cast_fp16 = slice_by_index(begin = var_7989_begin_0, end = var_7989_end_0, end_mask = var_7989_end_mask_0, x = var_7897_cast_fp16)[name = string("op_7989_cast_fp16")]; + tensor var_7996_begin_0 = const()[name = string("op_7996_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7996_end_0 = const()[name = string("op_7996_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7996_end_mask_0 = const()[name = string("op_7996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7996_cast_fp16 = slice_by_index(begin = var_7996_begin_0, end = var_7996_end_0, end_mask = var_7996_end_mask_0, x = var_7897_cast_fp16)[name = string("op_7996_cast_fp16")]; + tensor var_8003_begin_0 = const()[name = string("op_8003_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8003_end_0 = const()[name = string("op_8003_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8003_end_mask_0 = const()[name = string("op_8003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8003_cast_fp16 = slice_by_index(begin = var_8003_begin_0, end = var_8003_end_0, end_mask = var_8003_end_mask_0, x = var_7897_cast_fp16)[name = string("op_8003_cast_fp16")]; + tensor var_8010_begin_0 = const()[name = string("op_8010_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8010_end_0 = const()[name = string("op_8010_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8010_end_mask_0 = const()[name = string("op_8010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8010_cast_fp16 = slice_by_index(begin = var_8010_begin_0, end = var_8010_end_0, end_mask = var_8010_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8010_cast_fp16")]; + tensor var_8017_begin_0 = const()[name = string("op_8017_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8017_end_0 = const()[name = string("op_8017_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8017_end_mask_0 = const()[name = string("op_8017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8017_cast_fp16 = slice_by_index(begin = var_8017_begin_0, end = var_8017_end_0, end_mask = var_8017_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8017_cast_fp16")]; + tensor var_8024_begin_0 = const()[name = string("op_8024_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8024_end_0 = const()[name = string("op_8024_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8024_end_mask_0 = const()[name = string("op_8024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8024_cast_fp16 = slice_by_index(begin = var_8024_begin_0, end = var_8024_end_0, end_mask = var_8024_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8024_cast_fp16")]; + tensor var_8031_begin_0 = const()[name = string("op_8031_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8031_end_0 = const()[name = string("op_8031_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8031_end_mask_0 = const()[name = string("op_8031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8031_cast_fp16 = slice_by_index(begin = var_8031_begin_0, end = var_8031_end_0, end_mask = var_8031_end_mask_0, x = var_7901_cast_fp16)[name = string("op_8031_cast_fp16")]; + tensor var_8038_begin_0 = const()[name = string("op_8038_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8038_end_0 = const()[name = string("op_8038_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8038_end_mask_0 = const()[name = string("op_8038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8038_cast_fp16 = slice_by_index(begin = var_8038_begin_0, end = var_8038_end_0, end_mask = var_8038_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8038_cast_fp16")]; + tensor var_8045_begin_0 = const()[name = string("op_8045_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8045_end_0 = const()[name = string("op_8045_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8045_end_mask_0 = const()[name = string("op_8045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8045_cast_fp16 = slice_by_index(begin = var_8045_begin_0, end = var_8045_end_0, end_mask = var_8045_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8045_cast_fp16")]; + tensor var_8052_begin_0 = const()[name = string("op_8052_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8052_end_0 = const()[name = string("op_8052_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8052_end_mask_0 = const()[name = string("op_8052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8052_cast_fp16 = slice_by_index(begin = var_8052_begin_0, end = var_8052_end_0, end_mask = var_8052_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8052_cast_fp16")]; + tensor var_8059_begin_0 = const()[name = string("op_8059_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8059_end_0 = const()[name = string("op_8059_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8059_end_mask_0 = const()[name = string("op_8059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8059_cast_fp16 = slice_by_index(begin = var_8059_begin_0, end = var_8059_end_0, end_mask = var_8059_end_mask_0, x = var_7905_cast_fp16)[name = string("op_8059_cast_fp16")]; + tensor var_8066_begin_0 = const()[name = string("op_8066_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8066_end_0 = const()[name = string("op_8066_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8066_end_mask_0 = const()[name = string("op_8066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8066_cast_fp16 = slice_by_index(begin = var_8066_begin_0, end = var_8066_end_0, end_mask = var_8066_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8066_cast_fp16")]; + tensor var_8073_begin_0 = const()[name = string("op_8073_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8073_end_0 = const()[name = string("op_8073_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8073_end_mask_0 = const()[name = string("op_8073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8073_cast_fp16 = slice_by_index(begin = var_8073_begin_0, end = var_8073_end_0, end_mask = var_8073_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8073_cast_fp16")]; + tensor var_8080_begin_0 = const()[name = string("op_8080_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8080_end_0 = const()[name = string("op_8080_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8080_end_mask_0 = const()[name = string("op_8080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8080_cast_fp16 = slice_by_index(begin = var_8080_begin_0, end = var_8080_end_0, end_mask = var_8080_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8080_cast_fp16")]; + tensor var_8087_begin_0 = const()[name = string("op_8087_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8087_end_0 = const()[name = string("op_8087_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8087_end_mask_0 = const()[name = string("op_8087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8087_cast_fp16 = slice_by_index(begin = var_8087_begin_0, end = var_8087_end_0, end_mask = var_8087_end_mask_0, x = var_7909_cast_fp16)[name = string("op_8087_cast_fp16")]; + tensor var_8094_begin_0 = const()[name = string("op_8094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8094_end_0 = const()[name = string("op_8094_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8094_end_mask_0 = const()[name = string("op_8094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8094_cast_fp16 = slice_by_index(begin = var_8094_begin_0, end = var_8094_end_0, end_mask = var_8094_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8094_cast_fp16")]; + tensor var_8101_begin_0 = const()[name = string("op_8101_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8101_end_0 = const()[name = string("op_8101_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8101_end_mask_0 = const()[name = string("op_8101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8101_cast_fp16 = slice_by_index(begin = var_8101_begin_0, end = var_8101_end_0, end_mask = var_8101_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8101_cast_fp16")]; + tensor var_8108_begin_0 = const()[name = string("op_8108_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8108_end_0 = const()[name = string("op_8108_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8108_end_mask_0 = const()[name = string("op_8108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8108_cast_fp16 = slice_by_index(begin = var_8108_begin_0, end = var_8108_end_0, end_mask = var_8108_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8108_cast_fp16")]; + tensor var_8115_begin_0 = const()[name = string("op_8115_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8115_end_0 = const()[name = string("op_8115_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8115_end_mask_0 = const()[name = string("op_8115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8115_cast_fp16 = slice_by_index(begin = var_8115_begin_0, end = var_8115_end_0, end_mask = var_8115_end_mask_0, x = var_7913_cast_fp16)[name = string("op_8115_cast_fp16")]; + tensor var_8122_begin_0 = const()[name = string("op_8122_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8122_end_0 = const()[name = string("op_8122_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8122_end_mask_0 = const()[name = string("op_8122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8122_cast_fp16 = slice_by_index(begin = var_8122_begin_0, end = var_8122_end_0, end_mask = var_8122_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8122_cast_fp16")]; + tensor var_8129_begin_0 = const()[name = string("op_8129_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8129_end_0 = const()[name = string("op_8129_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8129_end_mask_0 = const()[name = string("op_8129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8129_cast_fp16 = slice_by_index(begin = var_8129_begin_0, end = var_8129_end_0, end_mask = var_8129_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8129_cast_fp16")]; + tensor var_8136_begin_0 = const()[name = string("op_8136_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8136_end_0 = const()[name = string("op_8136_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8136_end_mask_0 = const()[name = string("op_8136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8136_cast_fp16 = slice_by_index(begin = var_8136_begin_0, end = var_8136_end_0, end_mask = var_8136_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8136_cast_fp16")]; + tensor var_8143_begin_0 = const()[name = string("op_8143_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8143_end_0 = const()[name = string("op_8143_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8143_end_mask_0 = const()[name = string("op_8143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8143_cast_fp16 = slice_by_index(begin = var_8143_begin_0, end = var_8143_end_0, end_mask = var_8143_end_mask_0, x = var_7917_cast_fp16)[name = string("op_8143_cast_fp16")]; + tensor var_8150_begin_0 = const()[name = string("op_8150_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8150_end_0 = const()[name = string("op_8150_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8150_end_mask_0 = const()[name = string("op_8150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8150_cast_fp16 = slice_by_index(begin = var_8150_begin_0, end = var_8150_end_0, end_mask = var_8150_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8150_cast_fp16")]; + tensor var_8157_begin_0 = const()[name = string("op_8157_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8157_end_0 = const()[name = string("op_8157_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8157_end_mask_0 = const()[name = string("op_8157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8157_cast_fp16 = slice_by_index(begin = var_8157_begin_0, end = var_8157_end_0, end_mask = var_8157_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8157_cast_fp16")]; + tensor var_8164_begin_0 = const()[name = string("op_8164_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8164_end_0 = const()[name = string("op_8164_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8164_end_mask_0 = const()[name = string("op_8164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8164_cast_fp16 = slice_by_index(begin = var_8164_begin_0, end = var_8164_end_0, end_mask = var_8164_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8164_cast_fp16")]; + tensor var_8171_begin_0 = const()[name = string("op_8171_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8171_end_0 = const()[name = string("op_8171_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8171_end_mask_0 = const()[name = string("op_8171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8171_cast_fp16 = slice_by_index(begin = var_8171_begin_0, end = var_8171_end_0, end_mask = var_8171_end_mask_0, x = var_7921_cast_fp16)[name = string("op_8171_cast_fp16")]; + tensor var_8178_begin_0 = const()[name = string("op_8178_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8178_end_0 = const()[name = string("op_8178_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8178_end_mask_0 = const()[name = string("op_8178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8178_cast_fp16 = slice_by_index(begin = var_8178_begin_0, end = var_8178_end_0, end_mask = var_8178_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8178_cast_fp16")]; + tensor var_8185_begin_0 = const()[name = string("op_8185_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8185_end_0 = const()[name = string("op_8185_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8185_end_mask_0 = const()[name = string("op_8185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8185_cast_fp16 = slice_by_index(begin = var_8185_begin_0, end = var_8185_end_0, end_mask = var_8185_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8185_cast_fp16")]; + tensor var_8192_begin_0 = const()[name = string("op_8192_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8192_end_0 = const()[name = string("op_8192_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8192_end_mask_0 = const()[name = string("op_8192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8192_cast_fp16 = slice_by_index(begin = var_8192_begin_0, end = var_8192_end_0, end_mask = var_8192_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8192_cast_fp16")]; + tensor var_8199_begin_0 = const()[name = string("op_8199_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8199_end_0 = const()[name = string("op_8199_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8199_end_mask_0 = const()[name = string("op_8199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8199_cast_fp16 = slice_by_index(begin = var_8199_begin_0, end = var_8199_end_0, end_mask = var_8199_end_mask_0, x = var_7925_cast_fp16)[name = string("op_8199_cast_fp16")]; + tensor var_8206_begin_0 = const()[name = string("op_8206_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8206_end_0 = const()[name = string("op_8206_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8206_end_mask_0 = const()[name = string("op_8206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8206_cast_fp16 = slice_by_index(begin = var_8206_begin_0, end = var_8206_end_0, end_mask = var_8206_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8206_cast_fp16")]; + tensor var_8213_begin_0 = const()[name = string("op_8213_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8213_end_0 = const()[name = string("op_8213_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8213_end_mask_0 = const()[name = string("op_8213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8213_cast_fp16 = slice_by_index(begin = var_8213_begin_0, end = var_8213_end_0, end_mask = var_8213_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8213_cast_fp16")]; + tensor var_8220_begin_0 = const()[name = string("op_8220_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8220_end_0 = const()[name = string("op_8220_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8220_end_mask_0 = const()[name = string("op_8220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8220_cast_fp16 = slice_by_index(begin = var_8220_begin_0, end = var_8220_end_0, end_mask = var_8220_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8220_cast_fp16")]; + tensor var_8227_begin_0 = const()[name = string("op_8227_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8227_end_0 = const()[name = string("op_8227_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8227_end_mask_0 = const()[name = string("op_8227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8227_cast_fp16 = slice_by_index(begin = var_8227_begin_0, end = var_8227_end_0, end_mask = var_8227_end_mask_0, x = var_7929_cast_fp16)[name = string("op_8227_cast_fp16")]; + tensor var_8234_begin_0 = const()[name = string("op_8234_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8234_end_0 = const()[name = string("op_8234_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8234_end_mask_0 = const()[name = string("op_8234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8234_cast_fp16 = slice_by_index(begin = var_8234_begin_0, end = var_8234_end_0, end_mask = var_8234_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8234_cast_fp16")]; + tensor var_8241_begin_0 = const()[name = string("op_8241_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8241_end_0 = const()[name = string("op_8241_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8241_end_mask_0 = const()[name = string("op_8241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8241_cast_fp16 = slice_by_index(begin = var_8241_begin_0, end = var_8241_end_0, end_mask = var_8241_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8241_cast_fp16")]; + tensor var_8248_begin_0 = const()[name = string("op_8248_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8248_end_0 = const()[name = string("op_8248_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8248_end_mask_0 = const()[name = string("op_8248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8248_cast_fp16 = slice_by_index(begin = var_8248_begin_0, end = var_8248_end_0, end_mask = var_8248_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8248_cast_fp16")]; + tensor var_8255_begin_0 = const()[name = string("op_8255_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8255_end_0 = const()[name = string("op_8255_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8255_end_mask_0 = const()[name = string("op_8255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8255_cast_fp16 = slice_by_index(begin = var_8255_begin_0, end = var_8255_end_0, end_mask = var_8255_end_mask_0, x = var_7933_cast_fp16)[name = string("op_8255_cast_fp16")]; + tensor var_8262_begin_0 = const()[name = string("op_8262_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8262_end_0 = const()[name = string("op_8262_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8262_end_mask_0 = const()[name = string("op_8262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8262_cast_fp16 = slice_by_index(begin = var_8262_begin_0, end = var_8262_end_0, end_mask = var_8262_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8262_cast_fp16")]; + tensor var_8269_begin_0 = const()[name = string("op_8269_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8269_end_0 = const()[name = string("op_8269_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8269_end_mask_0 = const()[name = string("op_8269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8269_cast_fp16 = slice_by_index(begin = var_8269_begin_0, end = var_8269_end_0, end_mask = var_8269_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8269_cast_fp16")]; + tensor var_8276_begin_0 = const()[name = string("op_8276_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8276_end_0 = const()[name = string("op_8276_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8276_end_mask_0 = const()[name = string("op_8276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8276_cast_fp16 = slice_by_index(begin = var_8276_begin_0, end = var_8276_end_0, end_mask = var_8276_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8276_cast_fp16")]; + tensor var_8283_begin_0 = const()[name = string("op_8283_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8283_end_0 = const()[name = string("op_8283_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8283_end_mask_0 = const()[name = string("op_8283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8283_cast_fp16 = slice_by_index(begin = var_8283_begin_0, end = var_8283_end_0, end_mask = var_8283_end_mask_0, x = var_7937_cast_fp16)[name = string("op_8283_cast_fp16")]; + tensor var_8290_begin_0 = const()[name = string("op_8290_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8290_end_0 = const()[name = string("op_8290_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8290_end_mask_0 = const()[name = string("op_8290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8290_cast_fp16 = slice_by_index(begin = var_8290_begin_0, end = var_8290_end_0, end_mask = var_8290_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8290_cast_fp16")]; + tensor var_8297_begin_0 = const()[name = string("op_8297_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8297_end_0 = const()[name = string("op_8297_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8297_end_mask_0 = const()[name = string("op_8297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8297_cast_fp16 = slice_by_index(begin = var_8297_begin_0, end = var_8297_end_0, end_mask = var_8297_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8297_cast_fp16")]; + tensor var_8304_begin_0 = const()[name = string("op_8304_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8304_end_0 = const()[name = string("op_8304_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8304_end_mask_0 = const()[name = string("op_8304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8304_cast_fp16 = slice_by_index(begin = var_8304_begin_0, end = var_8304_end_0, end_mask = var_8304_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8304_cast_fp16")]; + tensor var_8311_begin_0 = const()[name = string("op_8311_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8311_end_0 = const()[name = string("op_8311_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8311_end_mask_0 = const()[name = string("op_8311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8311_cast_fp16 = slice_by_index(begin = var_8311_begin_0, end = var_8311_end_0, end_mask = var_8311_end_mask_0, x = var_7941_cast_fp16)[name = string("op_8311_cast_fp16")]; + tensor var_8318_begin_0 = const()[name = string("op_8318_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8318_end_0 = const()[name = string("op_8318_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8318_end_mask_0 = const()[name = string("op_8318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8318_cast_fp16 = slice_by_index(begin = var_8318_begin_0, end = var_8318_end_0, end_mask = var_8318_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8318_cast_fp16")]; + tensor var_8325_begin_0 = const()[name = string("op_8325_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8325_end_0 = const()[name = string("op_8325_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8325_end_mask_0 = const()[name = string("op_8325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8325_cast_fp16 = slice_by_index(begin = var_8325_begin_0, end = var_8325_end_0, end_mask = var_8325_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8325_cast_fp16")]; + tensor var_8332_begin_0 = const()[name = string("op_8332_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8332_end_0 = const()[name = string("op_8332_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8332_end_mask_0 = const()[name = string("op_8332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8332_cast_fp16 = slice_by_index(begin = var_8332_begin_0, end = var_8332_end_0, end_mask = var_8332_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8332_cast_fp16")]; + tensor var_8339_begin_0 = const()[name = string("op_8339_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8339_end_0 = const()[name = string("op_8339_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8339_end_mask_0 = const()[name = string("op_8339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8339_cast_fp16 = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = var_7945_cast_fp16)[name = string("op_8339_cast_fp16")]; + tensor var_8346_begin_0 = const()[name = string("op_8346_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8346_end_0 = const()[name = string("op_8346_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8346_end_mask_0 = const()[name = string("op_8346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8346_cast_fp16 = slice_by_index(begin = var_8346_begin_0, end = var_8346_end_0, end_mask = var_8346_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8346_cast_fp16")]; + tensor var_8353_begin_0 = const()[name = string("op_8353_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8353_end_0 = const()[name = string("op_8353_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8353_end_mask_0 = const()[name = string("op_8353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8353_cast_fp16 = slice_by_index(begin = var_8353_begin_0, end = var_8353_end_0, end_mask = var_8353_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8353_cast_fp16")]; + tensor var_8360_begin_0 = const()[name = string("op_8360_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8360_end_0 = const()[name = string("op_8360_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8360_end_mask_0 = const()[name = string("op_8360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8360_cast_fp16 = slice_by_index(begin = var_8360_begin_0, end = var_8360_end_0, end_mask = var_8360_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8360_cast_fp16")]; + tensor var_8367_begin_0 = const()[name = string("op_8367_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8367_end_0 = const()[name = string("op_8367_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8367_end_mask_0 = const()[name = string("op_8367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8367_cast_fp16 = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = var_7949_cast_fp16)[name = string("op_8367_cast_fp16")]; + tensor var_8374_begin_0 = const()[name = string("op_8374_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8374_end_0 = const()[name = string("op_8374_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8374_end_mask_0 = const()[name = string("op_8374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8374_cast_fp16 = slice_by_index(begin = var_8374_begin_0, end = var_8374_end_0, end_mask = var_8374_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8374_cast_fp16")]; + tensor var_8381_begin_0 = const()[name = string("op_8381_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8381_end_0 = const()[name = string("op_8381_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8381_end_mask_0 = const()[name = string("op_8381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8381_cast_fp16 = slice_by_index(begin = var_8381_begin_0, end = var_8381_end_0, end_mask = var_8381_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8381_cast_fp16")]; + tensor var_8388_begin_0 = const()[name = string("op_8388_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8388_end_0 = const()[name = string("op_8388_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8388_end_mask_0 = const()[name = string("op_8388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8388_cast_fp16 = slice_by_index(begin = var_8388_begin_0, end = var_8388_end_0, end_mask = var_8388_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8388_cast_fp16")]; + tensor var_8395_begin_0 = const()[name = string("op_8395_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8395_end_0 = const()[name = string("op_8395_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8395_end_mask_0 = const()[name = string("op_8395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8395_cast_fp16 = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = var_7953_cast_fp16)[name = string("op_8395_cast_fp16")]; + tensor var_8402_begin_0 = const()[name = string("op_8402_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8402_end_0 = const()[name = string("op_8402_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8402_end_mask_0 = const()[name = string("op_8402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8402_cast_fp16 = slice_by_index(begin = var_8402_begin_0, end = var_8402_end_0, end_mask = var_8402_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8402_cast_fp16")]; + tensor var_8409_begin_0 = const()[name = string("op_8409_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8409_end_0 = const()[name = string("op_8409_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8409_end_mask_0 = const()[name = string("op_8409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8409_cast_fp16 = slice_by_index(begin = var_8409_begin_0, end = var_8409_end_0, end_mask = var_8409_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8409_cast_fp16")]; + tensor var_8416_begin_0 = const()[name = string("op_8416_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8416_end_0 = const()[name = string("op_8416_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8416_end_mask_0 = const()[name = string("op_8416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8416_cast_fp16 = slice_by_index(begin = var_8416_begin_0, end = var_8416_end_0, end_mask = var_8416_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8416_cast_fp16")]; + tensor var_8423_begin_0 = const()[name = string("op_8423_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8423_end_0 = const()[name = string("op_8423_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8423_end_mask_0 = const()[name = string("op_8423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8423_cast_fp16 = slice_by_index(begin = var_8423_begin_0, end = var_8423_end_0, end_mask = var_8423_end_mask_0, x = var_7957_cast_fp16)[name = string("op_8423_cast_fp16")]; + tensor var_8430_begin_0 = const()[name = string("op_8430_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8430_end_0 = const()[name = string("op_8430_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8430_end_mask_0 = const()[name = string("op_8430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8430_cast_fp16 = slice_by_index(begin = var_8430_begin_0, end = var_8430_end_0, end_mask = var_8430_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8430_cast_fp16")]; + tensor var_8437_begin_0 = const()[name = string("op_8437_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8437_end_0 = const()[name = string("op_8437_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8437_end_mask_0 = const()[name = string("op_8437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8437_cast_fp16 = slice_by_index(begin = var_8437_begin_0, end = var_8437_end_0, end_mask = var_8437_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8437_cast_fp16")]; + tensor var_8444_begin_0 = const()[name = string("op_8444_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8444_end_0 = const()[name = string("op_8444_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8444_end_mask_0 = const()[name = string("op_8444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8444_cast_fp16 = slice_by_index(begin = var_8444_begin_0, end = var_8444_end_0, end_mask = var_8444_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8444_cast_fp16")]; + tensor var_8451_begin_0 = const()[name = string("op_8451_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8451_end_0 = const()[name = string("op_8451_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8451_end_mask_0 = const()[name = string("op_8451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8451_cast_fp16 = slice_by_index(begin = var_8451_begin_0, end = var_8451_end_0, end_mask = var_8451_end_mask_0, x = var_7961_cast_fp16)[name = string("op_8451_cast_fp16")]; + tensor var_8458_begin_0 = const()[name = string("op_8458_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8458_end_0 = const()[name = string("op_8458_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8458_end_mask_0 = const()[name = string("op_8458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8458_cast_fp16 = slice_by_index(begin = var_8458_begin_0, end = var_8458_end_0, end_mask = var_8458_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8458_cast_fp16")]; + tensor var_8465_begin_0 = const()[name = string("op_8465_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8465_end_0 = const()[name = string("op_8465_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8465_end_mask_0 = const()[name = string("op_8465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8465_cast_fp16 = slice_by_index(begin = var_8465_begin_0, end = var_8465_end_0, end_mask = var_8465_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8465_cast_fp16")]; + tensor var_8472_begin_0 = const()[name = string("op_8472_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8472_end_0 = const()[name = string("op_8472_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8472_end_mask_0 = const()[name = string("op_8472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8472_cast_fp16 = slice_by_index(begin = var_8472_begin_0, end = var_8472_end_0, end_mask = var_8472_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8472_cast_fp16")]; + tensor var_8479_begin_0 = const()[name = string("op_8479_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8479_end_0 = const()[name = string("op_8479_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8479_end_mask_0 = const()[name = string("op_8479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8479_cast_fp16 = slice_by_index(begin = var_8479_begin_0, end = var_8479_end_0, end_mask = var_8479_end_mask_0, x = var_7965_cast_fp16)[name = string("op_8479_cast_fp16")]; + tensor var_8486_begin_0 = const()[name = string("op_8486_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8486_end_0 = const()[name = string("op_8486_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8486_end_mask_0 = const()[name = string("op_8486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8486_cast_fp16 = slice_by_index(begin = var_8486_begin_0, end = var_8486_end_0, end_mask = var_8486_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8486_cast_fp16")]; + tensor var_8493_begin_0 = const()[name = string("op_8493_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8493_end_0 = const()[name = string("op_8493_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8493_end_mask_0 = const()[name = string("op_8493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8493_cast_fp16 = slice_by_index(begin = var_8493_begin_0, end = var_8493_end_0, end_mask = var_8493_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8493_cast_fp16")]; + tensor var_8500_begin_0 = const()[name = string("op_8500_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8500_end_0 = const()[name = string("op_8500_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8500_end_mask_0 = const()[name = string("op_8500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8500_cast_fp16 = slice_by_index(begin = var_8500_begin_0, end = var_8500_end_0, end_mask = var_8500_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8500_cast_fp16")]; + tensor var_8507_begin_0 = const()[name = string("op_8507_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8507_end_0 = const()[name = string("op_8507_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8507_end_mask_0 = const()[name = string("op_8507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8507_cast_fp16 = slice_by_index(begin = var_8507_begin_0, end = var_8507_end_0, end_mask = var_8507_end_mask_0, x = var_7969_cast_fp16)[name = string("op_8507_cast_fp16")]; + tensor var_8514_begin_0 = const()[name = string("op_8514_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8514_end_0 = const()[name = string("op_8514_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8514_end_mask_0 = const()[name = string("op_8514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8514_cast_fp16 = slice_by_index(begin = var_8514_begin_0, end = var_8514_end_0, end_mask = var_8514_end_mask_0, x = var_7973_cast_fp16)[name = string("op_8514_cast_fp16")]; + tensor var_8521_begin_0 = const()[name = string("op_8521_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8521_end_0 = const()[name = string("op_8521_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8521_end_mask_0 = const()[name = string("op_8521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8521_cast_fp16 = slice_by_index(begin = var_8521_begin_0, end = var_8521_end_0, end_mask = var_8521_end_mask_0, x = var_7973_cast_fp16)[name = string("op_8521_cast_fp16")]; + tensor var_8528_begin_0 = const()[name = string("op_8528_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8528_end_0 = const()[name = string("op_8528_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8528_end_mask_0 = const()[name = string("op_8528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8528_cast_fp16 = slice_by_index(begin = var_8528_begin_0, end = var_8528_end_0, end_mask = var_8528_end_mask_0, x = var_7973_cast_fp16)[name = string("op_8528_cast_fp16")]; + tensor var_8535_begin_0 = const()[name = string("op_8535_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8535_end_0 = const()[name = string("op_8535_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8535_end_mask_0 = const()[name = string("op_8535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8535_cast_fp16 = slice_by_index(begin = var_8535_begin_0, end = var_8535_end_0, end_mask = var_8535_end_mask_0, x = var_7973_cast_fp16)[name = string("op_8535_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8540_begin_0 = const()[name = string("op_8540_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8540_end_0 = const()[name = string("op_8540_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_8540_end_mask_0 = const()[name = string("op_8540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = string("transpose_26")]; + tensor var_8540_cast_fp16 = slice_by_index(begin = var_8540_begin_0, end = var_8540_end_0, end_mask = var_8540_end_mask_0, x = k_11_cast_fp16)[name = string("op_8540_cast_fp16")]; + tensor var_8544_begin_0 = const()[name = string("op_8544_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8544_end_0 = const()[name = string("op_8544_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_8544_end_mask_0 = const()[name = string("op_8544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8544_cast_fp16 = slice_by_index(begin = var_8544_begin_0, end = var_8544_end_0, end_mask = var_8544_end_mask_0, x = k_11_cast_fp16)[name = string("op_8544_cast_fp16")]; + tensor var_8548_begin_0 = const()[name = string("op_8548_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8548_end_0 = const()[name = string("op_8548_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_8548_end_mask_0 = const()[name = string("op_8548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8548_cast_fp16 = slice_by_index(begin = var_8548_begin_0, end = var_8548_end_0, end_mask = var_8548_end_mask_0, x = k_11_cast_fp16)[name = string("op_8548_cast_fp16")]; + tensor var_8552_begin_0 = const()[name = string("op_8552_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8552_end_0 = const()[name = string("op_8552_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_8552_end_mask_0 = const()[name = string("op_8552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8552_cast_fp16 = slice_by_index(begin = var_8552_begin_0, end = var_8552_end_0, end_mask = var_8552_end_mask_0, x = k_11_cast_fp16)[name = string("op_8552_cast_fp16")]; + tensor var_8556_begin_0 = const()[name = string("op_8556_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8556_end_0 = const()[name = string("op_8556_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_8556_end_mask_0 = const()[name = string("op_8556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8556_cast_fp16 = slice_by_index(begin = var_8556_begin_0, end = var_8556_end_0, end_mask = var_8556_end_mask_0, x = k_11_cast_fp16)[name = string("op_8556_cast_fp16")]; + tensor var_8560_begin_0 = const()[name = string("op_8560_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8560_end_0 = const()[name = string("op_8560_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_8560_end_mask_0 = const()[name = string("op_8560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8560_cast_fp16 = slice_by_index(begin = var_8560_begin_0, end = var_8560_end_0, end_mask = var_8560_end_mask_0, x = k_11_cast_fp16)[name = string("op_8560_cast_fp16")]; + tensor var_8564_begin_0 = const()[name = string("op_8564_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8564_end_0 = const()[name = string("op_8564_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_8564_end_mask_0 = const()[name = string("op_8564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8564_cast_fp16 = slice_by_index(begin = var_8564_begin_0, end = var_8564_end_0, end_mask = var_8564_end_mask_0, x = k_11_cast_fp16)[name = string("op_8564_cast_fp16")]; + tensor var_8568_begin_0 = const()[name = string("op_8568_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8568_end_0 = const()[name = string("op_8568_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_8568_end_mask_0 = const()[name = string("op_8568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8568_cast_fp16 = slice_by_index(begin = var_8568_begin_0, end = var_8568_end_0, end_mask = var_8568_end_mask_0, x = k_11_cast_fp16)[name = string("op_8568_cast_fp16")]; + tensor var_8572_begin_0 = const()[name = string("op_8572_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8572_end_0 = const()[name = string("op_8572_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_8572_end_mask_0 = const()[name = string("op_8572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8572_cast_fp16 = slice_by_index(begin = var_8572_begin_0, end = var_8572_end_0, end_mask = var_8572_end_mask_0, x = k_11_cast_fp16)[name = string("op_8572_cast_fp16")]; + tensor var_8576_begin_0 = const()[name = string("op_8576_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8576_end_0 = const()[name = string("op_8576_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_8576_end_mask_0 = const()[name = string("op_8576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8576_cast_fp16 = slice_by_index(begin = var_8576_begin_0, end = var_8576_end_0, end_mask = var_8576_end_mask_0, x = k_11_cast_fp16)[name = string("op_8576_cast_fp16")]; + tensor var_8580_begin_0 = const()[name = string("op_8580_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8580_end_0 = const()[name = string("op_8580_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_8580_end_mask_0 = const()[name = string("op_8580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8580_cast_fp16 = slice_by_index(begin = var_8580_begin_0, end = var_8580_end_0, end_mask = var_8580_end_mask_0, x = k_11_cast_fp16)[name = string("op_8580_cast_fp16")]; + tensor var_8584_begin_0 = const()[name = string("op_8584_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8584_end_0 = const()[name = string("op_8584_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_8584_end_mask_0 = const()[name = string("op_8584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8584_cast_fp16 = slice_by_index(begin = var_8584_begin_0, end = var_8584_end_0, end_mask = var_8584_end_mask_0, x = k_11_cast_fp16)[name = string("op_8584_cast_fp16")]; + tensor var_8588_begin_0 = const()[name = string("op_8588_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8588_end_0 = const()[name = string("op_8588_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_8588_end_mask_0 = const()[name = string("op_8588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8588_cast_fp16 = slice_by_index(begin = var_8588_begin_0, end = var_8588_end_0, end_mask = var_8588_end_mask_0, x = k_11_cast_fp16)[name = string("op_8588_cast_fp16")]; + tensor var_8592_begin_0 = const()[name = string("op_8592_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8592_end_0 = const()[name = string("op_8592_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_8592_end_mask_0 = const()[name = string("op_8592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8592_cast_fp16 = slice_by_index(begin = var_8592_begin_0, end = var_8592_end_0, end_mask = var_8592_end_mask_0, x = k_11_cast_fp16)[name = string("op_8592_cast_fp16")]; + tensor var_8596_begin_0 = const()[name = string("op_8596_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8596_end_0 = const()[name = string("op_8596_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_8596_end_mask_0 = const()[name = string("op_8596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8596_cast_fp16 = slice_by_index(begin = var_8596_begin_0, end = var_8596_end_0, end_mask = var_8596_end_mask_0, x = k_11_cast_fp16)[name = string("op_8596_cast_fp16")]; + tensor var_8600_begin_0 = const()[name = string("op_8600_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8600_end_0 = const()[name = string("op_8600_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_8600_end_mask_0 = const()[name = string("op_8600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8600_cast_fp16 = slice_by_index(begin = var_8600_begin_0, end = var_8600_end_0, end_mask = var_8600_end_mask_0, x = k_11_cast_fp16)[name = string("op_8600_cast_fp16")]; + tensor var_8604_begin_0 = const()[name = string("op_8604_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8604_end_0 = const()[name = string("op_8604_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_8604_end_mask_0 = const()[name = string("op_8604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8604_cast_fp16 = slice_by_index(begin = var_8604_begin_0, end = var_8604_end_0, end_mask = var_8604_end_mask_0, x = k_11_cast_fp16)[name = string("op_8604_cast_fp16")]; + tensor var_8608_begin_0 = const()[name = string("op_8608_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8608_end_0 = const()[name = string("op_8608_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_8608_end_mask_0 = const()[name = string("op_8608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8608_cast_fp16 = slice_by_index(begin = var_8608_begin_0, end = var_8608_end_0, end_mask = var_8608_end_mask_0, x = k_11_cast_fp16)[name = string("op_8608_cast_fp16")]; + tensor var_8612_begin_0 = const()[name = string("op_8612_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8612_end_0 = const()[name = string("op_8612_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_8612_end_mask_0 = const()[name = string("op_8612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8612_cast_fp16 = slice_by_index(begin = var_8612_begin_0, end = var_8612_end_0, end_mask = var_8612_end_mask_0, x = k_11_cast_fp16)[name = string("op_8612_cast_fp16")]; + tensor var_8616_begin_0 = const()[name = string("op_8616_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8616_end_0 = const()[name = string("op_8616_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_8616_end_mask_0 = const()[name = string("op_8616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8616_cast_fp16 = slice_by_index(begin = var_8616_begin_0, end = var_8616_end_0, end_mask = var_8616_end_mask_0, x = k_11_cast_fp16)[name = string("op_8616_cast_fp16")]; + tensor var_8618_begin_0 = const()[name = string("op_8618_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8618_end_0 = const()[name = string("op_8618_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8618_end_mask_0 = const()[name = string("op_8618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8618_cast_fp16 = slice_by_index(begin = var_8618_begin_0, end = var_8618_end_0, end_mask = var_8618_end_mask_0, x = value_11_cast_fp16)[name = string("op_8618_cast_fp16")]; + tensor var_8622_begin_0 = const()[name = string("op_8622_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8622_end_0 = const()[name = string("op_8622_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8622_end_mask_0 = const()[name = string("op_8622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8622_cast_fp16 = slice_by_index(begin = var_8622_begin_0, end = var_8622_end_0, end_mask = var_8622_end_mask_0, x = value_11_cast_fp16)[name = string("op_8622_cast_fp16")]; + tensor var_8626_begin_0 = const()[name = string("op_8626_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8626_end_0 = const()[name = string("op_8626_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8626_end_mask_0 = const()[name = string("op_8626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8626_cast_fp16 = slice_by_index(begin = var_8626_begin_0, end = var_8626_end_0, end_mask = var_8626_end_mask_0, x = value_11_cast_fp16)[name = string("op_8626_cast_fp16")]; + tensor var_8630_begin_0 = const()[name = string("op_8630_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8630_end_0 = const()[name = string("op_8630_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8630_end_mask_0 = const()[name = string("op_8630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8630_cast_fp16 = slice_by_index(begin = var_8630_begin_0, end = var_8630_end_0, end_mask = var_8630_end_mask_0, x = value_11_cast_fp16)[name = string("op_8630_cast_fp16")]; + tensor var_8634_begin_0 = const()[name = string("op_8634_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8634_end_0 = const()[name = string("op_8634_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8634_end_mask_0 = const()[name = string("op_8634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8634_cast_fp16 = slice_by_index(begin = var_8634_begin_0, end = var_8634_end_0, end_mask = var_8634_end_mask_0, x = value_11_cast_fp16)[name = string("op_8634_cast_fp16")]; + tensor var_8638_begin_0 = const()[name = string("op_8638_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8638_end_0 = const()[name = string("op_8638_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8638_end_mask_0 = const()[name = string("op_8638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8638_cast_fp16 = slice_by_index(begin = var_8638_begin_0, end = var_8638_end_0, end_mask = var_8638_end_mask_0, x = value_11_cast_fp16)[name = string("op_8638_cast_fp16")]; + tensor var_8642_begin_0 = const()[name = string("op_8642_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8642_end_0 = const()[name = string("op_8642_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8642_end_mask_0 = const()[name = string("op_8642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8642_cast_fp16 = slice_by_index(begin = var_8642_begin_0, end = var_8642_end_0, end_mask = var_8642_end_mask_0, x = value_11_cast_fp16)[name = string("op_8642_cast_fp16")]; + tensor var_8646_begin_0 = const()[name = string("op_8646_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8646_end_0 = const()[name = string("op_8646_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8646_end_mask_0 = const()[name = string("op_8646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8646_cast_fp16 = slice_by_index(begin = var_8646_begin_0, end = var_8646_end_0, end_mask = var_8646_end_mask_0, x = value_11_cast_fp16)[name = string("op_8646_cast_fp16")]; + tensor var_8650_begin_0 = const()[name = string("op_8650_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8650_end_0 = const()[name = string("op_8650_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8650_end_mask_0 = const()[name = string("op_8650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8650_cast_fp16 = slice_by_index(begin = var_8650_begin_0, end = var_8650_end_0, end_mask = var_8650_end_mask_0, x = value_11_cast_fp16)[name = string("op_8650_cast_fp16")]; + tensor var_8654_begin_0 = const()[name = string("op_8654_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8654_end_0 = const()[name = string("op_8654_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8654_end_mask_0 = const()[name = string("op_8654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8654_cast_fp16 = slice_by_index(begin = var_8654_begin_0, end = var_8654_end_0, end_mask = var_8654_end_mask_0, x = value_11_cast_fp16)[name = string("op_8654_cast_fp16")]; + tensor var_8658_begin_0 = const()[name = string("op_8658_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8658_end_0 = const()[name = string("op_8658_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8658_end_mask_0 = const()[name = string("op_8658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8658_cast_fp16 = slice_by_index(begin = var_8658_begin_0, end = var_8658_end_0, end_mask = var_8658_end_mask_0, x = value_11_cast_fp16)[name = string("op_8658_cast_fp16")]; + tensor var_8662_begin_0 = const()[name = string("op_8662_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8662_end_0 = const()[name = string("op_8662_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8662_end_mask_0 = const()[name = string("op_8662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8662_cast_fp16 = slice_by_index(begin = var_8662_begin_0, end = var_8662_end_0, end_mask = var_8662_end_mask_0, x = value_11_cast_fp16)[name = string("op_8662_cast_fp16")]; + tensor var_8666_begin_0 = const()[name = string("op_8666_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8666_end_0 = const()[name = string("op_8666_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_8666_end_mask_0 = const()[name = string("op_8666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8666_cast_fp16 = slice_by_index(begin = var_8666_begin_0, end = var_8666_end_0, end_mask = var_8666_end_mask_0, x = value_11_cast_fp16)[name = string("op_8666_cast_fp16")]; + tensor var_8670_begin_0 = const()[name = string("op_8670_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8670_end_0 = const()[name = string("op_8670_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_8670_end_mask_0 = const()[name = string("op_8670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8670_cast_fp16 = slice_by_index(begin = var_8670_begin_0, end = var_8670_end_0, end_mask = var_8670_end_mask_0, x = value_11_cast_fp16)[name = string("op_8670_cast_fp16")]; + tensor var_8674_begin_0 = const()[name = string("op_8674_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8674_end_0 = const()[name = string("op_8674_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_8674_end_mask_0 = const()[name = string("op_8674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8674_cast_fp16 = slice_by_index(begin = var_8674_begin_0, end = var_8674_end_0, end_mask = var_8674_end_mask_0, x = value_11_cast_fp16)[name = string("op_8674_cast_fp16")]; + tensor var_8678_begin_0 = const()[name = string("op_8678_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8678_end_0 = const()[name = string("op_8678_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_8678_end_mask_0 = const()[name = string("op_8678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8678_cast_fp16 = slice_by_index(begin = var_8678_begin_0, end = var_8678_end_0, end_mask = var_8678_end_mask_0, x = value_11_cast_fp16)[name = string("op_8678_cast_fp16")]; + tensor var_8682_begin_0 = const()[name = string("op_8682_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8682_end_0 = const()[name = string("op_8682_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_8682_end_mask_0 = const()[name = string("op_8682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8682_cast_fp16 = slice_by_index(begin = var_8682_begin_0, end = var_8682_end_0, end_mask = var_8682_end_mask_0, x = value_11_cast_fp16)[name = string("op_8682_cast_fp16")]; + tensor var_8686_begin_0 = const()[name = string("op_8686_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8686_end_0 = const()[name = string("op_8686_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_8686_end_mask_0 = const()[name = string("op_8686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8686_cast_fp16 = slice_by_index(begin = var_8686_begin_0, end = var_8686_end_0, end_mask = var_8686_end_mask_0, x = value_11_cast_fp16)[name = string("op_8686_cast_fp16")]; + tensor var_8690_begin_0 = const()[name = string("op_8690_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8690_end_0 = const()[name = string("op_8690_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_8690_end_mask_0 = const()[name = string("op_8690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8690_cast_fp16 = slice_by_index(begin = var_8690_begin_0, end = var_8690_end_0, end_mask = var_8690_end_mask_0, x = value_11_cast_fp16)[name = string("op_8690_cast_fp16")]; + tensor var_8694_begin_0 = const()[name = string("op_8694_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8694_end_0 = const()[name = string("op_8694_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_8694_end_mask_0 = const()[name = string("op_8694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8694_cast_fp16 = slice_by_index(begin = var_8694_begin_0, end = var_8694_end_0, end_mask = var_8694_end_mask_0, x = value_11_cast_fp16)[name = string("op_8694_cast_fp16")]; + string _SplitHeadsQ__mh_w_801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_801_equation_0, values = (var_8540_cast_fp16, var_7982_cast_fp16))[name = string("_SplitHeadsQ__mh_w_801_cast_fp16")]; + string _SplitHeadsQ__mh_w_803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_803_equation_0, values = (var_8540_cast_fp16, var_7989_cast_fp16))[name = string("_SplitHeadsQ__mh_w_803_cast_fp16")]; + string _SplitHeadsQ__mh_w_805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_805_equation_0, values = (var_8540_cast_fp16, var_7996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_805_cast_fp16")]; + string _SplitHeadsQ__mh_w_807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_807_equation_0, values = (var_8540_cast_fp16, var_8003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_807_cast_fp16")]; + string _SplitHeadsQ__mh_w_809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_809_equation_0, values = (var_8544_cast_fp16, var_8010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_809_cast_fp16")]; + string _SplitHeadsQ__mh_w_811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_811_equation_0, values = (var_8544_cast_fp16, var_8017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_811_cast_fp16")]; + string _SplitHeadsQ__mh_w_813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_813_equation_0, values = (var_8544_cast_fp16, var_8024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_813_cast_fp16")]; + string _SplitHeadsQ__mh_w_815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_815_equation_0, values = (var_8544_cast_fp16, var_8031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_815_cast_fp16")]; + string _SplitHeadsQ__mh_w_817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_817_equation_0, values = (var_8548_cast_fp16, var_8038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_817_cast_fp16")]; + string _SplitHeadsQ__mh_w_819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_819_equation_0, values = (var_8548_cast_fp16, var_8045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_819_cast_fp16")]; + string _SplitHeadsQ__mh_w_821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_821_equation_0, values = (var_8548_cast_fp16, var_8052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_821_cast_fp16")]; + string _SplitHeadsQ__mh_w_823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_823_equation_0, values = (var_8548_cast_fp16, var_8059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_823_cast_fp16")]; + string _SplitHeadsQ__mh_w_825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_825_equation_0, values = (var_8552_cast_fp16, var_8066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_825_cast_fp16")]; + string _SplitHeadsQ__mh_w_827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_827_equation_0, values = (var_8552_cast_fp16, var_8073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_827_cast_fp16")]; + string _SplitHeadsQ__mh_w_829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_829_equation_0, values = (var_8552_cast_fp16, var_8080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_829_cast_fp16")]; + string _SplitHeadsQ__mh_w_831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_831_equation_0, values = (var_8552_cast_fp16, var_8087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_831_cast_fp16")]; + string _SplitHeadsQ__mh_w_833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_833_equation_0, values = (var_8556_cast_fp16, var_8094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_833_cast_fp16")]; + string _SplitHeadsQ__mh_w_835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_835_equation_0, values = (var_8556_cast_fp16, var_8101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_835_cast_fp16")]; + string _SplitHeadsQ__mh_w_837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_837_equation_0, values = (var_8556_cast_fp16, var_8108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_837_cast_fp16")]; + string _SplitHeadsQ__mh_w_839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_839_equation_0, values = (var_8556_cast_fp16, var_8115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_839_cast_fp16")]; + string _SplitHeadsQ__mh_w_841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_841_equation_0, values = (var_8560_cast_fp16, var_8122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_841_cast_fp16")]; + string _SplitHeadsQ__mh_w_843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_843_equation_0, values = (var_8560_cast_fp16, var_8129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_843_cast_fp16")]; + string _SplitHeadsQ__mh_w_845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_845_equation_0, values = (var_8560_cast_fp16, var_8136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_845_cast_fp16")]; + string _SplitHeadsQ__mh_w_847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_847_equation_0, values = (var_8560_cast_fp16, var_8143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_847_cast_fp16")]; + string _SplitHeadsQ__mh_w_849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_849_equation_0, values = (var_8564_cast_fp16, var_8150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_849_cast_fp16")]; + string _SplitHeadsQ__mh_w_851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_851_equation_0, values = (var_8564_cast_fp16, var_8157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_851_cast_fp16")]; + string _SplitHeadsQ__mh_w_853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_853_equation_0, values = (var_8564_cast_fp16, var_8164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_853_cast_fp16")]; + string _SplitHeadsQ__mh_w_855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_855_equation_0, values = (var_8564_cast_fp16, var_8171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_855_cast_fp16")]; + string _SplitHeadsQ__mh_w_857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_857_equation_0, values = (var_8568_cast_fp16, var_8178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_857_cast_fp16")]; + string _SplitHeadsQ__mh_w_859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_859_equation_0, values = (var_8568_cast_fp16, var_8185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_859_cast_fp16")]; + string _SplitHeadsQ__mh_w_861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_861_equation_0, values = (var_8568_cast_fp16, var_8192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_861_cast_fp16")]; + string _SplitHeadsQ__mh_w_863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_863_equation_0, values = (var_8568_cast_fp16, var_8199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_863_cast_fp16")]; + string _SplitHeadsQ__mh_w_865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_865_equation_0, values = (var_8572_cast_fp16, var_8206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_865_cast_fp16")]; + string _SplitHeadsQ__mh_w_867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_867_equation_0, values = (var_8572_cast_fp16, var_8213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_867_cast_fp16")]; + string _SplitHeadsQ__mh_w_869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_869_equation_0, values = (var_8572_cast_fp16, var_8220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_869_cast_fp16")]; + string _SplitHeadsQ__mh_w_871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_871_equation_0, values = (var_8572_cast_fp16, var_8227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_871_cast_fp16")]; + string _SplitHeadsQ__mh_w_873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_873_equation_0, values = (var_8576_cast_fp16, var_8234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_873_cast_fp16")]; + string _SplitHeadsQ__mh_w_875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_875_equation_0, values = (var_8576_cast_fp16, var_8241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_875_cast_fp16")]; + string _SplitHeadsQ__mh_w_877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_877_equation_0, values = (var_8576_cast_fp16, var_8248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_877_cast_fp16")]; + string _SplitHeadsQ__mh_w_879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_879_equation_0, values = (var_8576_cast_fp16, var_8255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_879_cast_fp16")]; + string _SplitHeadsQ__mh_w_881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_881_equation_0, values = (var_8580_cast_fp16, var_8262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_881_cast_fp16")]; + string _SplitHeadsQ__mh_w_883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_883_equation_0, values = (var_8580_cast_fp16, var_8269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_883_cast_fp16")]; + string _SplitHeadsQ__mh_w_885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_885_equation_0, values = (var_8580_cast_fp16, var_8276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_885_cast_fp16")]; + string _SplitHeadsQ__mh_w_887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_887_equation_0, values = (var_8580_cast_fp16, var_8283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_887_cast_fp16")]; + string _SplitHeadsQ__mh_w_889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_889_equation_0, values = (var_8584_cast_fp16, var_8290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_889_cast_fp16")]; + string _SplitHeadsQ__mh_w_891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_891_equation_0, values = (var_8584_cast_fp16, var_8297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_891_cast_fp16")]; + string _SplitHeadsQ__mh_w_893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_893_equation_0, values = (var_8584_cast_fp16, var_8304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_893_cast_fp16")]; + string _SplitHeadsQ__mh_w_895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_895_equation_0, values = (var_8584_cast_fp16, var_8311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_895_cast_fp16")]; + string _SplitHeadsQ__mh_w_897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_897_equation_0, values = (var_8588_cast_fp16, var_8318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_897_cast_fp16")]; + string _SplitHeadsQ__mh_w_899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_899_equation_0, values = (var_8588_cast_fp16, var_8325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_899_cast_fp16")]; + string _SplitHeadsQ__mh_w_901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_901_equation_0, values = (var_8588_cast_fp16, var_8332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_901_cast_fp16")]; + string _SplitHeadsQ__mh_w_903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_903_equation_0, values = (var_8588_cast_fp16, var_8339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_903_cast_fp16")]; + string _SplitHeadsQ__mh_w_905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_905_equation_0, values = (var_8592_cast_fp16, var_8346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_905_cast_fp16")]; + string _SplitHeadsQ__mh_w_907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_907_equation_0, values = (var_8592_cast_fp16, var_8353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_907_cast_fp16")]; + string _SplitHeadsQ__mh_w_909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_909_equation_0, values = (var_8592_cast_fp16, var_8360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_909_cast_fp16")]; + string _SplitHeadsQ__mh_w_911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_911_equation_0, values = (var_8592_cast_fp16, var_8367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_911_cast_fp16")]; + string _SplitHeadsQ__mh_w_913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_913_equation_0, values = (var_8596_cast_fp16, var_8374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_913_cast_fp16")]; + string _SplitHeadsQ__mh_w_915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_915_equation_0, values = (var_8596_cast_fp16, var_8381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_915_cast_fp16")]; + string _SplitHeadsQ__mh_w_917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_917_equation_0, values = (var_8596_cast_fp16, var_8388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_917_cast_fp16")]; + string _SplitHeadsQ__mh_w_919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_919_equation_0, values = (var_8596_cast_fp16, var_8395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_919_cast_fp16")]; + string _SplitHeadsQ__mh_w_921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_921_equation_0, values = (var_8600_cast_fp16, var_8402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_921_cast_fp16")]; + string _SplitHeadsQ__mh_w_923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_923_equation_0, values = (var_8600_cast_fp16, var_8409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_923_cast_fp16")]; + string _SplitHeadsQ__mh_w_925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_925_equation_0, values = (var_8600_cast_fp16, var_8416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_925_cast_fp16")]; + string _SplitHeadsQ__mh_w_927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_927_equation_0, values = (var_8600_cast_fp16, var_8423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_927_cast_fp16")]; + string _SplitHeadsQ__mh_w_929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_929_equation_0, values = (var_8604_cast_fp16, var_8430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_929_cast_fp16")]; + string _SplitHeadsQ__mh_w_931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_931_equation_0, values = (var_8604_cast_fp16, var_8437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_931_cast_fp16")]; + string _SplitHeadsQ__mh_w_933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_933_equation_0, values = (var_8604_cast_fp16, var_8444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_933_cast_fp16")]; + string _SplitHeadsQ__mh_w_935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_935_equation_0, values = (var_8604_cast_fp16, var_8451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_935_cast_fp16")]; + string _SplitHeadsQ__mh_w_937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_937_equation_0, values = (var_8608_cast_fp16, var_8458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_937_cast_fp16")]; + string _SplitHeadsQ__mh_w_939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_939_equation_0, values = (var_8608_cast_fp16, var_8465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_939_cast_fp16")]; + string _SplitHeadsQ__mh_w_941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_941_equation_0, values = (var_8608_cast_fp16, var_8472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_941_cast_fp16")]; + string _SplitHeadsQ__mh_w_943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_943_equation_0, values = (var_8608_cast_fp16, var_8479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_943_cast_fp16")]; + string _SplitHeadsQ__mh_w_945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_945_equation_0, values = (var_8612_cast_fp16, var_8486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_945_cast_fp16")]; + string _SplitHeadsQ__mh_w_947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_947_equation_0, values = (var_8612_cast_fp16, var_8493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_947_cast_fp16")]; + string _SplitHeadsQ__mh_w_949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_949_equation_0, values = (var_8612_cast_fp16, var_8500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_949_cast_fp16")]; + string _SplitHeadsQ__mh_w_951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_951_equation_0, values = (var_8612_cast_fp16, var_8507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_951_cast_fp16")]; + string _SplitHeadsQ__mh_w_953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_953_equation_0, values = (var_8616_cast_fp16, var_8514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_953_cast_fp16")]; + string _SplitHeadsQ__mh_w_955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_955_equation_0, values = (var_8616_cast_fp16, var_8521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_955_cast_fp16")]; + string _SplitHeadsQ__mh_w_957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_957_equation_0, values = (var_8616_cast_fp16, var_8528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_957_cast_fp16")]; + string _SplitHeadsQ__mh_w_959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_959_equation_0, values = (var_8616_cast_fp16, var_8535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_959_cast_fp16")]; + fp16 var_8857_to_fp16 = const()[name = string("op_8857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_801_cast_fp16, y = var_8857_to_fp16)[name = string("aw_chunk_801_cast_fp16")]; + fp16 var_8859_to_fp16 = const()[name = string("op_8859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_803_cast_fp16, y = var_8859_to_fp16)[name = string("aw_chunk_803_cast_fp16")]; + fp16 var_8861_to_fp16 = const()[name = string("op_8861_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_805_cast_fp16, y = var_8861_to_fp16)[name = string("aw_chunk_805_cast_fp16")]; + fp16 var_8863_to_fp16 = const()[name = string("op_8863_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_807_cast_fp16, y = var_8863_to_fp16)[name = string("aw_chunk_807_cast_fp16")]; + fp16 var_8865_to_fp16 = const()[name = string("op_8865_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_809_cast_fp16, y = var_8865_to_fp16)[name = string("aw_chunk_809_cast_fp16")]; + fp16 var_8867_to_fp16 = const()[name = string("op_8867_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_811_cast_fp16, y = var_8867_to_fp16)[name = string("aw_chunk_811_cast_fp16")]; + fp16 var_8869_to_fp16 = const()[name = string("op_8869_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_813_cast_fp16, y = var_8869_to_fp16)[name = string("aw_chunk_813_cast_fp16")]; + fp16 var_8871_to_fp16 = const()[name = string("op_8871_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_815_cast_fp16, y = var_8871_to_fp16)[name = string("aw_chunk_815_cast_fp16")]; + fp16 var_8873_to_fp16 = const()[name = string("op_8873_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_817_cast_fp16, y = var_8873_to_fp16)[name = string("aw_chunk_817_cast_fp16")]; + fp16 var_8875_to_fp16 = const()[name = string("op_8875_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_819_cast_fp16, y = var_8875_to_fp16)[name = string("aw_chunk_819_cast_fp16")]; + fp16 var_8877_to_fp16 = const()[name = string("op_8877_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_821_cast_fp16, y = var_8877_to_fp16)[name = string("aw_chunk_821_cast_fp16")]; + fp16 var_8879_to_fp16 = const()[name = string("op_8879_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_823_cast_fp16, y = var_8879_to_fp16)[name = string("aw_chunk_823_cast_fp16")]; + fp16 var_8881_to_fp16 = const()[name = string("op_8881_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_825_cast_fp16, y = var_8881_to_fp16)[name = string("aw_chunk_825_cast_fp16")]; + fp16 var_8883_to_fp16 = const()[name = string("op_8883_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_827_cast_fp16, y = var_8883_to_fp16)[name = string("aw_chunk_827_cast_fp16")]; + fp16 var_8885_to_fp16 = const()[name = string("op_8885_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_829_cast_fp16, y = var_8885_to_fp16)[name = string("aw_chunk_829_cast_fp16")]; + fp16 var_8887_to_fp16 = const()[name = string("op_8887_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_831_cast_fp16, y = var_8887_to_fp16)[name = string("aw_chunk_831_cast_fp16")]; + fp16 var_8889_to_fp16 = const()[name = string("op_8889_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_833_cast_fp16, y = var_8889_to_fp16)[name = string("aw_chunk_833_cast_fp16")]; + fp16 var_8891_to_fp16 = const()[name = string("op_8891_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_835_cast_fp16, y = var_8891_to_fp16)[name = string("aw_chunk_835_cast_fp16")]; + fp16 var_8893_to_fp16 = const()[name = string("op_8893_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_837_cast_fp16, y = var_8893_to_fp16)[name = string("aw_chunk_837_cast_fp16")]; + fp16 var_8895_to_fp16 = const()[name = string("op_8895_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_839_cast_fp16, y = var_8895_to_fp16)[name = string("aw_chunk_839_cast_fp16")]; + fp16 var_8897_to_fp16 = const()[name = string("op_8897_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_841_cast_fp16, y = var_8897_to_fp16)[name = string("aw_chunk_841_cast_fp16")]; + fp16 var_8899_to_fp16 = const()[name = string("op_8899_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_843_cast_fp16, y = var_8899_to_fp16)[name = string("aw_chunk_843_cast_fp16")]; + fp16 var_8901_to_fp16 = const()[name = string("op_8901_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_845_cast_fp16, y = var_8901_to_fp16)[name = string("aw_chunk_845_cast_fp16")]; + fp16 var_8903_to_fp16 = const()[name = string("op_8903_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_847_cast_fp16, y = var_8903_to_fp16)[name = string("aw_chunk_847_cast_fp16")]; + fp16 var_8905_to_fp16 = const()[name = string("op_8905_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_849_cast_fp16, y = var_8905_to_fp16)[name = string("aw_chunk_849_cast_fp16")]; + fp16 var_8907_to_fp16 = const()[name = string("op_8907_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_851_cast_fp16, y = var_8907_to_fp16)[name = string("aw_chunk_851_cast_fp16")]; + fp16 var_8909_to_fp16 = const()[name = string("op_8909_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_853_cast_fp16, y = var_8909_to_fp16)[name = string("aw_chunk_853_cast_fp16")]; + fp16 var_8911_to_fp16 = const()[name = string("op_8911_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_855_cast_fp16, y = var_8911_to_fp16)[name = string("aw_chunk_855_cast_fp16")]; + fp16 var_8913_to_fp16 = const()[name = string("op_8913_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_857_cast_fp16, y = var_8913_to_fp16)[name = string("aw_chunk_857_cast_fp16")]; + fp16 var_8915_to_fp16 = const()[name = string("op_8915_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_859_cast_fp16, y = var_8915_to_fp16)[name = string("aw_chunk_859_cast_fp16")]; + fp16 var_8917_to_fp16 = const()[name = string("op_8917_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_861_cast_fp16, y = var_8917_to_fp16)[name = string("aw_chunk_861_cast_fp16")]; + fp16 var_8919_to_fp16 = const()[name = string("op_8919_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_863_cast_fp16, y = var_8919_to_fp16)[name = string("aw_chunk_863_cast_fp16")]; + fp16 var_8921_to_fp16 = const()[name = string("op_8921_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_865_cast_fp16, y = var_8921_to_fp16)[name = string("aw_chunk_865_cast_fp16")]; + fp16 var_8923_to_fp16 = const()[name = string("op_8923_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_867_cast_fp16, y = var_8923_to_fp16)[name = string("aw_chunk_867_cast_fp16")]; + fp16 var_8925_to_fp16 = const()[name = string("op_8925_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_869_cast_fp16, y = var_8925_to_fp16)[name = string("aw_chunk_869_cast_fp16")]; + fp16 var_8927_to_fp16 = const()[name = string("op_8927_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_871_cast_fp16, y = var_8927_to_fp16)[name = string("aw_chunk_871_cast_fp16")]; + fp16 var_8929_to_fp16 = const()[name = string("op_8929_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_873_cast_fp16, y = var_8929_to_fp16)[name = string("aw_chunk_873_cast_fp16")]; + fp16 var_8931_to_fp16 = const()[name = string("op_8931_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_875_cast_fp16, y = var_8931_to_fp16)[name = string("aw_chunk_875_cast_fp16")]; + fp16 var_8933_to_fp16 = const()[name = string("op_8933_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_877_cast_fp16, y = var_8933_to_fp16)[name = string("aw_chunk_877_cast_fp16")]; + fp16 var_8935_to_fp16 = const()[name = string("op_8935_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_879_cast_fp16, y = var_8935_to_fp16)[name = string("aw_chunk_879_cast_fp16")]; + fp16 var_8937_to_fp16 = const()[name = string("op_8937_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_881_cast_fp16, y = var_8937_to_fp16)[name = string("aw_chunk_881_cast_fp16")]; + fp16 var_8939_to_fp16 = const()[name = string("op_8939_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_883_cast_fp16, y = var_8939_to_fp16)[name = string("aw_chunk_883_cast_fp16")]; + fp16 var_8941_to_fp16 = const()[name = string("op_8941_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_885_cast_fp16, y = var_8941_to_fp16)[name = string("aw_chunk_885_cast_fp16")]; + fp16 var_8943_to_fp16 = const()[name = string("op_8943_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_887_cast_fp16, y = var_8943_to_fp16)[name = string("aw_chunk_887_cast_fp16")]; + fp16 var_8945_to_fp16 = const()[name = string("op_8945_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_889_cast_fp16, y = var_8945_to_fp16)[name = string("aw_chunk_889_cast_fp16")]; + fp16 var_8947_to_fp16 = const()[name = string("op_8947_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_891_cast_fp16, y = var_8947_to_fp16)[name = string("aw_chunk_891_cast_fp16")]; + fp16 var_8949_to_fp16 = const()[name = string("op_8949_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_893_cast_fp16, y = var_8949_to_fp16)[name = string("aw_chunk_893_cast_fp16")]; + fp16 var_8951_to_fp16 = const()[name = string("op_8951_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_895_cast_fp16, y = var_8951_to_fp16)[name = string("aw_chunk_895_cast_fp16")]; + fp16 var_8953_to_fp16 = const()[name = string("op_8953_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_897_cast_fp16, y = var_8953_to_fp16)[name = string("aw_chunk_897_cast_fp16")]; + fp16 var_8955_to_fp16 = const()[name = string("op_8955_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_899_cast_fp16, y = var_8955_to_fp16)[name = string("aw_chunk_899_cast_fp16")]; + fp16 var_8957_to_fp16 = const()[name = string("op_8957_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_901_cast_fp16, y = var_8957_to_fp16)[name = string("aw_chunk_901_cast_fp16")]; + fp16 var_8959_to_fp16 = const()[name = string("op_8959_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_903_cast_fp16, y = var_8959_to_fp16)[name = string("aw_chunk_903_cast_fp16")]; + fp16 var_8961_to_fp16 = const()[name = string("op_8961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_905_cast_fp16, y = var_8961_to_fp16)[name = string("aw_chunk_905_cast_fp16")]; + fp16 var_8963_to_fp16 = const()[name = string("op_8963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_907_cast_fp16, y = var_8963_to_fp16)[name = string("aw_chunk_907_cast_fp16")]; + fp16 var_8965_to_fp16 = const()[name = string("op_8965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_909_cast_fp16, y = var_8965_to_fp16)[name = string("aw_chunk_909_cast_fp16")]; + fp16 var_8967_to_fp16 = const()[name = string("op_8967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_911_cast_fp16, y = var_8967_to_fp16)[name = string("aw_chunk_911_cast_fp16")]; + fp16 var_8969_to_fp16 = const()[name = string("op_8969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_913_cast_fp16, y = var_8969_to_fp16)[name = string("aw_chunk_913_cast_fp16")]; + fp16 var_8971_to_fp16 = const()[name = string("op_8971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_915_cast_fp16, y = var_8971_to_fp16)[name = string("aw_chunk_915_cast_fp16")]; + fp16 var_8973_to_fp16 = const()[name = string("op_8973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_917_cast_fp16, y = var_8973_to_fp16)[name = string("aw_chunk_917_cast_fp16")]; + fp16 var_8975_to_fp16 = const()[name = string("op_8975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_919_cast_fp16, y = var_8975_to_fp16)[name = string("aw_chunk_919_cast_fp16")]; + fp16 var_8977_to_fp16 = const()[name = string("op_8977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_921_cast_fp16, y = var_8977_to_fp16)[name = string("aw_chunk_921_cast_fp16")]; + fp16 var_8979_to_fp16 = const()[name = string("op_8979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_923_cast_fp16, y = var_8979_to_fp16)[name = string("aw_chunk_923_cast_fp16")]; + fp16 var_8981_to_fp16 = const()[name = string("op_8981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_925_cast_fp16, y = var_8981_to_fp16)[name = string("aw_chunk_925_cast_fp16")]; + fp16 var_8983_to_fp16 = const()[name = string("op_8983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_927_cast_fp16, y = var_8983_to_fp16)[name = string("aw_chunk_927_cast_fp16")]; + fp16 var_8985_to_fp16 = const()[name = string("op_8985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_929_cast_fp16, y = var_8985_to_fp16)[name = string("aw_chunk_929_cast_fp16")]; + fp16 var_8987_to_fp16 = const()[name = string("op_8987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_931_cast_fp16, y = var_8987_to_fp16)[name = string("aw_chunk_931_cast_fp16")]; + fp16 var_8989_to_fp16 = const()[name = string("op_8989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_933_cast_fp16, y = var_8989_to_fp16)[name = string("aw_chunk_933_cast_fp16")]; + fp16 var_8991_to_fp16 = const()[name = string("op_8991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_935_cast_fp16, y = var_8991_to_fp16)[name = string("aw_chunk_935_cast_fp16")]; + fp16 var_8993_to_fp16 = const()[name = string("op_8993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_937_cast_fp16, y = var_8993_to_fp16)[name = string("aw_chunk_937_cast_fp16")]; + fp16 var_8995_to_fp16 = const()[name = string("op_8995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_939_cast_fp16, y = var_8995_to_fp16)[name = string("aw_chunk_939_cast_fp16")]; + fp16 var_8997_to_fp16 = const()[name = string("op_8997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_941_cast_fp16, y = var_8997_to_fp16)[name = string("aw_chunk_941_cast_fp16")]; + fp16 var_8999_to_fp16 = const()[name = string("op_8999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_943_cast_fp16, y = var_8999_to_fp16)[name = string("aw_chunk_943_cast_fp16")]; + fp16 var_9001_to_fp16 = const()[name = string("op_9001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_945_cast_fp16, y = var_9001_to_fp16)[name = string("aw_chunk_945_cast_fp16")]; + fp16 var_9003_to_fp16 = const()[name = string("op_9003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_947_cast_fp16, y = var_9003_to_fp16)[name = string("aw_chunk_947_cast_fp16")]; + fp16 var_9005_to_fp16 = const()[name = string("op_9005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_949_cast_fp16, y = var_9005_to_fp16)[name = string("aw_chunk_949_cast_fp16")]; + fp16 var_9007_to_fp16 = const()[name = string("op_9007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_951_cast_fp16, y = var_9007_to_fp16)[name = string("aw_chunk_951_cast_fp16")]; + fp16 var_9009_to_fp16 = const()[name = string("op_9009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_953_cast_fp16, y = var_9009_to_fp16)[name = string("aw_chunk_953_cast_fp16")]; + fp16 var_9011_to_fp16 = const()[name = string("op_9011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_955_cast_fp16, y = var_9011_to_fp16)[name = string("aw_chunk_955_cast_fp16")]; + fp16 var_9013_to_fp16 = const()[name = string("op_9013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_957_cast_fp16, y = var_9013_to_fp16)[name = string("aw_chunk_957_cast_fp16")]; + fp16 var_9015_to_fp16 = const()[name = string("op_9015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_959_cast_fp16, y = var_9015_to_fp16)[name = string("aw_chunk_959_cast_fp16")]; + tensor var_9017_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_801_cast_fp16)[name = string("op_9017_cast_fp16")]; + tensor var_9018_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_803_cast_fp16)[name = string("op_9018_cast_fp16")]; + tensor var_9019_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_805_cast_fp16)[name = string("op_9019_cast_fp16")]; + tensor var_9020_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_807_cast_fp16)[name = string("op_9020_cast_fp16")]; + tensor var_9021_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_809_cast_fp16)[name = string("op_9021_cast_fp16")]; + tensor var_9022_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_811_cast_fp16)[name = string("op_9022_cast_fp16")]; + tensor var_9023_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_813_cast_fp16)[name = string("op_9023_cast_fp16")]; + tensor var_9024_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_815_cast_fp16)[name = string("op_9024_cast_fp16")]; + tensor var_9025_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_817_cast_fp16)[name = string("op_9025_cast_fp16")]; + tensor var_9026_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_819_cast_fp16)[name = string("op_9026_cast_fp16")]; + tensor var_9027_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_821_cast_fp16)[name = string("op_9027_cast_fp16")]; + tensor var_9028_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_823_cast_fp16)[name = string("op_9028_cast_fp16")]; + tensor var_9029_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_825_cast_fp16)[name = string("op_9029_cast_fp16")]; + tensor var_9030_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_827_cast_fp16)[name = string("op_9030_cast_fp16")]; + tensor var_9031_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_829_cast_fp16)[name = string("op_9031_cast_fp16")]; + tensor var_9032_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_831_cast_fp16)[name = string("op_9032_cast_fp16")]; + tensor var_9033_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_833_cast_fp16)[name = string("op_9033_cast_fp16")]; + tensor var_9034_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_835_cast_fp16)[name = string("op_9034_cast_fp16")]; + tensor var_9035_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_837_cast_fp16)[name = string("op_9035_cast_fp16")]; + tensor var_9036_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_839_cast_fp16)[name = string("op_9036_cast_fp16")]; + tensor var_9037_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_841_cast_fp16)[name = string("op_9037_cast_fp16")]; + tensor var_9038_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_843_cast_fp16)[name = string("op_9038_cast_fp16")]; + tensor var_9039_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_845_cast_fp16)[name = string("op_9039_cast_fp16")]; + tensor var_9040_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_847_cast_fp16)[name = string("op_9040_cast_fp16")]; + tensor var_9041_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_849_cast_fp16)[name = string("op_9041_cast_fp16")]; + tensor var_9042_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_851_cast_fp16)[name = string("op_9042_cast_fp16")]; + tensor var_9043_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_853_cast_fp16)[name = string("op_9043_cast_fp16")]; + tensor var_9044_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_855_cast_fp16)[name = string("op_9044_cast_fp16")]; + tensor var_9045_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_857_cast_fp16)[name = string("op_9045_cast_fp16")]; + tensor var_9046_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_859_cast_fp16)[name = string("op_9046_cast_fp16")]; + tensor var_9047_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_861_cast_fp16)[name = string("op_9047_cast_fp16")]; + tensor var_9048_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_863_cast_fp16)[name = string("op_9048_cast_fp16")]; + tensor var_9049_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_865_cast_fp16)[name = string("op_9049_cast_fp16")]; + tensor var_9050_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_867_cast_fp16)[name = string("op_9050_cast_fp16")]; + tensor var_9051_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_869_cast_fp16)[name = string("op_9051_cast_fp16")]; + tensor var_9052_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_871_cast_fp16)[name = string("op_9052_cast_fp16")]; + tensor var_9053_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_873_cast_fp16)[name = string("op_9053_cast_fp16")]; + tensor var_9054_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_875_cast_fp16)[name = string("op_9054_cast_fp16")]; + tensor var_9055_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_877_cast_fp16)[name = string("op_9055_cast_fp16")]; + tensor var_9056_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_879_cast_fp16)[name = string("op_9056_cast_fp16")]; + tensor var_9057_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_881_cast_fp16)[name = string("op_9057_cast_fp16")]; + tensor var_9058_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_883_cast_fp16)[name = string("op_9058_cast_fp16")]; + tensor var_9059_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_885_cast_fp16)[name = string("op_9059_cast_fp16")]; + tensor var_9060_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_887_cast_fp16)[name = string("op_9060_cast_fp16")]; + tensor var_9061_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_889_cast_fp16)[name = string("op_9061_cast_fp16")]; + tensor var_9062_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_891_cast_fp16)[name = string("op_9062_cast_fp16")]; + tensor var_9063_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_893_cast_fp16)[name = string("op_9063_cast_fp16")]; + tensor var_9064_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_895_cast_fp16)[name = string("op_9064_cast_fp16")]; + tensor var_9065_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_897_cast_fp16)[name = string("op_9065_cast_fp16")]; + tensor var_9066_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_899_cast_fp16)[name = string("op_9066_cast_fp16")]; + tensor var_9067_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_901_cast_fp16)[name = string("op_9067_cast_fp16")]; + tensor var_9068_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_903_cast_fp16)[name = string("op_9068_cast_fp16")]; + tensor var_9069_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_905_cast_fp16)[name = string("op_9069_cast_fp16")]; + tensor var_9070_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_907_cast_fp16)[name = string("op_9070_cast_fp16")]; + tensor var_9071_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_909_cast_fp16)[name = string("op_9071_cast_fp16")]; + tensor var_9072_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_911_cast_fp16)[name = string("op_9072_cast_fp16")]; + tensor var_9073_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_913_cast_fp16)[name = string("op_9073_cast_fp16")]; + tensor var_9074_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_915_cast_fp16)[name = string("op_9074_cast_fp16")]; + tensor var_9075_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_917_cast_fp16)[name = string("op_9075_cast_fp16")]; + tensor var_9076_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_919_cast_fp16)[name = string("op_9076_cast_fp16")]; + tensor var_9077_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_921_cast_fp16)[name = string("op_9077_cast_fp16")]; + tensor var_9078_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_923_cast_fp16)[name = string("op_9078_cast_fp16")]; + tensor var_9079_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_925_cast_fp16)[name = string("op_9079_cast_fp16")]; + tensor var_9080_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_927_cast_fp16)[name = string("op_9080_cast_fp16")]; + tensor var_9081_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_929_cast_fp16)[name = string("op_9081_cast_fp16")]; + tensor var_9082_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_931_cast_fp16)[name = string("op_9082_cast_fp16")]; + tensor var_9083_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_933_cast_fp16)[name = string("op_9083_cast_fp16")]; + tensor var_9084_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_935_cast_fp16)[name = string("op_9084_cast_fp16")]; + tensor var_9085_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_937_cast_fp16)[name = string("op_9085_cast_fp16")]; + tensor var_9086_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_939_cast_fp16)[name = string("op_9086_cast_fp16")]; + tensor var_9087_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_941_cast_fp16)[name = string("op_9087_cast_fp16")]; + tensor var_9088_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_943_cast_fp16)[name = string("op_9088_cast_fp16")]; + tensor var_9089_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_945_cast_fp16)[name = string("op_9089_cast_fp16")]; + tensor var_9090_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_947_cast_fp16)[name = string("op_9090_cast_fp16")]; + tensor var_9091_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_949_cast_fp16)[name = string("op_9091_cast_fp16")]; + tensor var_9092_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_951_cast_fp16)[name = string("op_9092_cast_fp16")]; + tensor var_9093_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_953_cast_fp16)[name = string("op_9093_cast_fp16")]; + tensor var_9094_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_955_cast_fp16)[name = string("op_9094_cast_fp16")]; + tensor var_9095_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_957_cast_fp16)[name = string("op_9095_cast_fp16")]; + tensor var_9096_cast_fp16 = softmax(axis = var_7842, x = aw_chunk_959_cast_fp16)[name = string("op_9096_cast_fp16")]; + string var_9098_equation_0 = const()[name = string("op_9098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9098_cast_fp16 = einsum(equation = var_9098_equation_0, values = (var_8618_cast_fp16, var_9017_cast_fp16))[name = string("op_9098_cast_fp16")]; + string var_9100_equation_0 = const()[name = string("op_9100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9100_cast_fp16 = einsum(equation = var_9100_equation_0, values = (var_8618_cast_fp16, var_9018_cast_fp16))[name = string("op_9100_cast_fp16")]; + string var_9102_equation_0 = const()[name = string("op_9102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9102_cast_fp16 = einsum(equation = var_9102_equation_0, values = (var_8618_cast_fp16, var_9019_cast_fp16))[name = string("op_9102_cast_fp16")]; + string var_9104_equation_0 = const()[name = string("op_9104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9104_cast_fp16 = einsum(equation = var_9104_equation_0, values = (var_8618_cast_fp16, var_9020_cast_fp16))[name = string("op_9104_cast_fp16")]; + string var_9106_equation_0 = const()[name = string("op_9106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9106_cast_fp16 = einsum(equation = var_9106_equation_0, values = (var_8622_cast_fp16, var_9021_cast_fp16))[name = string("op_9106_cast_fp16")]; + string var_9108_equation_0 = const()[name = string("op_9108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9108_cast_fp16 = einsum(equation = var_9108_equation_0, values = (var_8622_cast_fp16, var_9022_cast_fp16))[name = string("op_9108_cast_fp16")]; + string var_9110_equation_0 = const()[name = string("op_9110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9110_cast_fp16 = einsum(equation = var_9110_equation_0, values = (var_8622_cast_fp16, var_9023_cast_fp16))[name = string("op_9110_cast_fp16")]; + string var_9112_equation_0 = const()[name = string("op_9112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9112_cast_fp16 = einsum(equation = var_9112_equation_0, values = (var_8622_cast_fp16, var_9024_cast_fp16))[name = string("op_9112_cast_fp16")]; + string var_9114_equation_0 = const()[name = string("op_9114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9114_cast_fp16 = einsum(equation = var_9114_equation_0, values = (var_8626_cast_fp16, var_9025_cast_fp16))[name = string("op_9114_cast_fp16")]; + string var_9116_equation_0 = const()[name = string("op_9116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9116_cast_fp16 = einsum(equation = var_9116_equation_0, values = (var_8626_cast_fp16, var_9026_cast_fp16))[name = string("op_9116_cast_fp16")]; + string var_9118_equation_0 = const()[name = string("op_9118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9118_cast_fp16 = einsum(equation = var_9118_equation_0, values = (var_8626_cast_fp16, var_9027_cast_fp16))[name = string("op_9118_cast_fp16")]; + string var_9120_equation_0 = const()[name = string("op_9120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9120_cast_fp16 = einsum(equation = var_9120_equation_0, values = (var_8626_cast_fp16, var_9028_cast_fp16))[name = string("op_9120_cast_fp16")]; + string var_9122_equation_0 = const()[name = string("op_9122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9122_cast_fp16 = einsum(equation = var_9122_equation_0, values = (var_8630_cast_fp16, var_9029_cast_fp16))[name = string("op_9122_cast_fp16")]; + string var_9124_equation_0 = const()[name = string("op_9124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9124_cast_fp16 = einsum(equation = var_9124_equation_0, values = (var_8630_cast_fp16, var_9030_cast_fp16))[name = string("op_9124_cast_fp16")]; + string var_9126_equation_0 = const()[name = string("op_9126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9126_cast_fp16 = einsum(equation = var_9126_equation_0, values = (var_8630_cast_fp16, var_9031_cast_fp16))[name = string("op_9126_cast_fp16")]; + string var_9128_equation_0 = const()[name = string("op_9128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9128_cast_fp16 = einsum(equation = var_9128_equation_0, values = (var_8630_cast_fp16, var_9032_cast_fp16))[name = string("op_9128_cast_fp16")]; + string var_9130_equation_0 = const()[name = string("op_9130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9130_cast_fp16 = einsum(equation = var_9130_equation_0, values = (var_8634_cast_fp16, var_9033_cast_fp16))[name = string("op_9130_cast_fp16")]; + string var_9132_equation_0 = const()[name = string("op_9132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9132_cast_fp16 = einsum(equation = var_9132_equation_0, values = (var_8634_cast_fp16, var_9034_cast_fp16))[name = string("op_9132_cast_fp16")]; + string var_9134_equation_0 = const()[name = string("op_9134_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9134_cast_fp16 = einsum(equation = var_9134_equation_0, values = (var_8634_cast_fp16, var_9035_cast_fp16))[name = string("op_9134_cast_fp16")]; + string var_9136_equation_0 = const()[name = string("op_9136_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9136_cast_fp16 = einsum(equation = var_9136_equation_0, values = (var_8634_cast_fp16, var_9036_cast_fp16))[name = string("op_9136_cast_fp16")]; + string var_9138_equation_0 = const()[name = string("op_9138_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9138_cast_fp16 = einsum(equation = var_9138_equation_0, values = (var_8638_cast_fp16, var_9037_cast_fp16))[name = string("op_9138_cast_fp16")]; + string var_9140_equation_0 = const()[name = string("op_9140_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9140_cast_fp16 = einsum(equation = var_9140_equation_0, values = (var_8638_cast_fp16, var_9038_cast_fp16))[name = string("op_9140_cast_fp16")]; + string var_9142_equation_0 = const()[name = string("op_9142_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9142_cast_fp16 = einsum(equation = var_9142_equation_0, values = (var_8638_cast_fp16, var_9039_cast_fp16))[name = string("op_9142_cast_fp16")]; + string var_9144_equation_0 = const()[name = string("op_9144_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9144_cast_fp16 = einsum(equation = var_9144_equation_0, values = (var_8638_cast_fp16, var_9040_cast_fp16))[name = string("op_9144_cast_fp16")]; + string var_9146_equation_0 = const()[name = string("op_9146_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9146_cast_fp16 = einsum(equation = var_9146_equation_0, values = (var_8642_cast_fp16, var_9041_cast_fp16))[name = string("op_9146_cast_fp16")]; + string var_9148_equation_0 = const()[name = string("op_9148_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9148_cast_fp16 = einsum(equation = var_9148_equation_0, values = (var_8642_cast_fp16, var_9042_cast_fp16))[name = string("op_9148_cast_fp16")]; + string var_9150_equation_0 = const()[name = string("op_9150_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9150_cast_fp16 = einsum(equation = var_9150_equation_0, values = (var_8642_cast_fp16, var_9043_cast_fp16))[name = string("op_9150_cast_fp16")]; + string var_9152_equation_0 = const()[name = string("op_9152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9152_cast_fp16 = einsum(equation = var_9152_equation_0, values = (var_8642_cast_fp16, var_9044_cast_fp16))[name = string("op_9152_cast_fp16")]; + string var_9154_equation_0 = const()[name = string("op_9154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9154_cast_fp16 = einsum(equation = var_9154_equation_0, values = (var_8646_cast_fp16, var_9045_cast_fp16))[name = string("op_9154_cast_fp16")]; + string var_9156_equation_0 = const()[name = string("op_9156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9156_cast_fp16 = einsum(equation = var_9156_equation_0, values = (var_8646_cast_fp16, var_9046_cast_fp16))[name = string("op_9156_cast_fp16")]; + string var_9158_equation_0 = const()[name = string("op_9158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9158_cast_fp16 = einsum(equation = var_9158_equation_0, values = (var_8646_cast_fp16, var_9047_cast_fp16))[name = string("op_9158_cast_fp16")]; + string var_9160_equation_0 = const()[name = string("op_9160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9160_cast_fp16 = einsum(equation = var_9160_equation_0, values = (var_8646_cast_fp16, var_9048_cast_fp16))[name = string("op_9160_cast_fp16")]; + string var_9162_equation_0 = const()[name = string("op_9162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9162_cast_fp16 = einsum(equation = var_9162_equation_0, values = (var_8650_cast_fp16, var_9049_cast_fp16))[name = string("op_9162_cast_fp16")]; + string var_9164_equation_0 = const()[name = string("op_9164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9164_cast_fp16 = einsum(equation = var_9164_equation_0, values = (var_8650_cast_fp16, var_9050_cast_fp16))[name = string("op_9164_cast_fp16")]; + string var_9166_equation_0 = const()[name = string("op_9166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9166_cast_fp16 = einsum(equation = var_9166_equation_0, values = (var_8650_cast_fp16, var_9051_cast_fp16))[name = string("op_9166_cast_fp16")]; + string var_9168_equation_0 = const()[name = string("op_9168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9168_cast_fp16 = einsum(equation = var_9168_equation_0, values = (var_8650_cast_fp16, var_9052_cast_fp16))[name = string("op_9168_cast_fp16")]; + string var_9170_equation_0 = const()[name = string("op_9170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9170_cast_fp16 = einsum(equation = var_9170_equation_0, values = (var_8654_cast_fp16, var_9053_cast_fp16))[name = string("op_9170_cast_fp16")]; + string var_9172_equation_0 = const()[name = string("op_9172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9172_cast_fp16 = einsum(equation = var_9172_equation_0, values = (var_8654_cast_fp16, var_9054_cast_fp16))[name = string("op_9172_cast_fp16")]; + string var_9174_equation_0 = const()[name = string("op_9174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9174_cast_fp16 = einsum(equation = var_9174_equation_0, values = (var_8654_cast_fp16, var_9055_cast_fp16))[name = string("op_9174_cast_fp16")]; + string var_9176_equation_0 = const()[name = string("op_9176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9176_cast_fp16 = einsum(equation = var_9176_equation_0, values = (var_8654_cast_fp16, var_9056_cast_fp16))[name = string("op_9176_cast_fp16")]; + string var_9178_equation_0 = const()[name = string("op_9178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9178_cast_fp16 = einsum(equation = var_9178_equation_0, values = (var_8658_cast_fp16, var_9057_cast_fp16))[name = string("op_9178_cast_fp16")]; + string var_9180_equation_0 = const()[name = string("op_9180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9180_cast_fp16 = einsum(equation = var_9180_equation_0, values = (var_8658_cast_fp16, var_9058_cast_fp16))[name = string("op_9180_cast_fp16")]; + string var_9182_equation_0 = const()[name = string("op_9182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9182_cast_fp16 = einsum(equation = var_9182_equation_0, values = (var_8658_cast_fp16, var_9059_cast_fp16))[name = string("op_9182_cast_fp16")]; + string var_9184_equation_0 = const()[name = string("op_9184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9184_cast_fp16 = einsum(equation = var_9184_equation_0, values = (var_8658_cast_fp16, var_9060_cast_fp16))[name = string("op_9184_cast_fp16")]; + string var_9186_equation_0 = const()[name = string("op_9186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9186_cast_fp16 = einsum(equation = var_9186_equation_0, values = (var_8662_cast_fp16, var_9061_cast_fp16))[name = string("op_9186_cast_fp16")]; + string var_9188_equation_0 = const()[name = string("op_9188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9188_cast_fp16 = einsum(equation = var_9188_equation_0, values = (var_8662_cast_fp16, var_9062_cast_fp16))[name = string("op_9188_cast_fp16")]; + string var_9190_equation_0 = const()[name = string("op_9190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9190_cast_fp16 = einsum(equation = var_9190_equation_0, values = (var_8662_cast_fp16, var_9063_cast_fp16))[name = string("op_9190_cast_fp16")]; + string var_9192_equation_0 = const()[name = string("op_9192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9192_cast_fp16 = einsum(equation = var_9192_equation_0, values = (var_8662_cast_fp16, var_9064_cast_fp16))[name = string("op_9192_cast_fp16")]; + string var_9194_equation_0 = const()[name = string("op_9194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9194_cast_fp16 = einsum(equation = var_9194_equation_0, values = (var_8666_cast_fp16, var_9065_cast_fp16))[name = string("op_9194_cast_fp16")]; + string var_9196_equation_0 = const()[name = string("op_9196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9196_cast_fp16 = einsum(equation = var_9196_equation_0, values = (var_8666_cast_fp16, var_9066_cast_fp16))[name = string("op_9196_cast_fp16")]; + string var_9198_equation_0 = const()[name = string("op_9198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9198_cast_fp16 = einsum(equation = var_9198_equation_0, values = (var_8666_cast_fp16, var_9067_cast_fp16))[name = string("op_9198_cast_fp16")]; + string var_9200_equation_0 = const()[name = string("op_9200_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9200_cast_fp16 = einsum(equation = var_9200_equation_0, values = (var_8666_cast_fp16, var_9068_cast_fp16))[name = string("op_9200_cast_fp16")]; + string var_9202_equation_0 = const()[name = string("op_9202_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9202_cast_fp16 = einsum(equation = var_9202_equation_0, values = (var_8670_cast_fp16, var_9069_cast_fp16))[name = string("op_9202_cast_fp16")]; + string var_9204_equation_0 = const()[name = string("op_9204_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9204_cast_fp16 = einsum(equation = var_9204_equation_0, values = (var_8670_cast_fp16, var_9070_cast_fp16))[name = string("op_9204_cast_fp16")]; + string var_9206_equation_0 = const()[name = string("op_9206_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9206_cast_fp16 = einsum(equation = var_9206_equation_0, values = (var_8670_cast_fp16, var_9071_cast_fp16))[name = string("op_9206_cast_fp16")]; + string var_9208_equation_0 = const()[name = string("op_9208_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9208_cast_fp16 = einsum(equation = var_9208_equation_0, values = (var_8670_cast_fp16, var_9072_cast_fp16))[name = string("op_9208_cast_fp16")]; + string var_9210_equation_0 = const()[name = string("op_9210_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9210_cast_fp16 = einsum(equation = var_9210_equation_0, values = (var_8674_cast_fp16, var_9073_cast_fp16))[name = string("op_9210_cast_fp16")]; + string var_9212_equation_0 = const()[name = string("op_9212_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9212_cast_fp16 = einsum(equation = var_9212_equation_0, values = (var_8674_cast_fp16, var_9074_cast_fp16))[name = string("op_9212_cast_fp16")]; + string var_9214_equation_0 = const()[name = string("op_9214_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9214_cast_fp16 = einsum(equation = var_9214_equation_0, values = (var_8674_cast_fp16, var_9075_cast_fp16))[name = string("op_9214_cast_fp16")]; + string var_9216_equation_0 = const()[name = string("op_9216_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9216_cast_fp16 = einsum(equation = var_9216_equation_0, values = (var_8674_cast_fp16, var_9076_cast_fp16))[name = string("op_9216_cast_fp16")]; + string var_9218_equation_0 = const()[name = string("op_9218_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9218_cast_fp16 = einsum(equation = var_9218_equation_0, values = (var_8678_cast_fp16, var_9077_cast_fp16))[name = string("op_9218_cast_fp16")]; + string var_9220_equation_0 = const()[name = string("op_9220_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9220_cast_fp16 = einsum(equation = var_9220_equation_0, values = (var_8678_cast_fp16, var_9078_cast_fp16))[name = string("op_9220_cast_fp16")]; + string var_9222_equation_0 = const()[name = string("op_9222_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9222_cast_fp16 = einsum(equation = var_9222_equation_0, values = (var_8678_cast_fp16, var_9079_cast_fp16))[name = string("op_9222_cast_fp16")]; + string var_9224_equation_0 = const()[name = string("op_9224_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9224_cast_fp16 = einsum(equation = var_9224_equation_0, values = (var_8678_cast_fp16, var_9080_cast_fp16))[name = string("op_9224_cast_fp16")]; + string var_9226_equation_0 = const()[name = string("op_9226_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9226_cast_fp16 = einsum(equation = var_9226_equation_0, values = (var_8682_cast_fp16, var_9081_cast_fp16))[name = string("op_9226_cast_fp16")]; + string var_9228_equation_0 = const()[name = string("op_9228_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9228_cast_fp16 = einsum(equation = var_9228_equation_0, values = (var_8682_cast_fp16, var_9082_cast_fp16))[name = string("op_9228_cast_fp16")]; + string var_9230_equation_0 = const()[name = string("op_9230_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9230_cast_fp16 = einsum(equation = var_9230_equation_0, values = (var_8682_cast_fp16, var_9083_cast_fp16))[name = string("op_9230_cast_fp16")]; + string var_9232_equation_0 = const()[name = string("op_9232_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9232_cast_fp16 = einsum(equation = var_9232_equation_0, values = (var_8682_cast_fp16, var_9084_cast_fp16))[name = string("op_9232_cast_fp16")]; + string var_9234_equation_0 = const()[name = string("op_9234_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9234_cast_fp16 = einsum(equation = var_9234_equation_0, values = (var_8686_cast_fp16, var_9085_cast_fp16))[name = string("op_9234_cast_fp16")]; + string var_9236_equation_0 = const()[name = string("op_9236_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9236_cast_fp16 = einsum(equation = var_9236_equation_0, values = (var_8686_cast_fp16, var_9086_cast_fp16))[name = string("op_9236_cast_fp16")]; + string var_9238_equation_0 = const()[name = string("op_9238_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9238_cast_fp16 = einsum(equation = var_9238_equation_0, values = (var_8686_cast_fp16, var_9087_cast_fp16))[name = string("op_9238_cast_fp16")]; + string var_9240_equation_0 = const()[name = string("op_9240_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9240_cast_fp16 = einsum(equation = var_9240_equation_0, values = (var_8686_cast_fp16, var_9088_cast_fp16))[name = string("op_9240_cast_fp16")]; + string var_9242_equation_0 = const()[name = string("op_9242_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9242_cast_fp16 = einsum(equation = var_9242_equation_0, values = (var_8690_cast_fp16, var_9089_cast_fp16))[name = string("op_9242_cast_fp16")]; + string var_9244_equation_0 = const()[name = string("op_9244_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9244_cast_fp16 = einsum(equation = var_9244_equation_0, values = (var_8690_cast_fp16, var_9090_cast_fp16))[name = string("op_9244_cast_fp16")]; + string var_9246_equation_0 = const()[name = string("op_9246_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9246_cast_fp16 = einsum(equation = var_9246_equation_0, values = (var_8690_cast_fp16, var_9091_cast_fp16))[name = string("op_9246_cast_fp16")]; + string var_9248_equation_0 = const()[name = string("op_9248_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9248_cast_fp16 = einsum(equation = var_9248_equation_0, values = (var_8690_cast_fp16, var_9092_cast_fp16))[name = string("op_9248_cast_fp16")]; + string var_9250_equation_0 = const()[name = string("op_9250_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9250_cast_fp16 = einsum(equation = var_9250_equation_0, values = (var_8694_cast_fp16, var_9093_cast_fp16))[name = string("op_9250_cast_fp16")]; + string var_9252_equation_0 = const()[name = string("op_9252_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9252_cast_fp16 = einsum(equation = var_9252_equation_0, values = (var_8694_cast_fp16, var_9094_cast_fp16))[name = string("op_9252_cast_fp16")]; + string var_9254_equation_0 = const()[name = string("op_9254_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9254_cast_fp16 = einsum(equation = var_9254_equation_0, values = (var_8694_cast_fp16, var_9095_cast_fp16))[name = string("op_9254_cast_fp16")]; + string var_9256_equation_0 = const()[name = string("op_9256_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9256_cast_fp16 = einsum(equation = var_9256_equation_0, values = (var_8694_cast_fp16, var_9096_cast_fp16))[name = string("op_9256_cast_fp16")]; + bool var_9258_interleave_0 = const()[name = string("op_9258_interleave_0"), val = bool(false)]; + tensor var_9258_cast_fp16 = concat(axis = var_7817, interleave = var_9258_interleave_0, values = (var_9098_cast_fp16, var_9100_cast_fp16, var_9102_cast_fp16, var_9104_cast_fp16))[name = string("op_9258_cast_fp16")]; + bool var_9260_interleave_0 = const()[name = string("op_9260_interleave_0"), val = bool(false)]; + tensor var_9260_cast_fp16 = concat(axis = var_7817, interleave = var_9260_interleave_0, values = (var_9106_cast_fp16, var_9108_cast_fp16, var_9110_cast_fp16, var_9112_cast_fp16))[name = string("op_9260_cast_fp16")]; + bool var_9262_interleave_0 = const()[name = string("op_9262_interleave_0"), val = bool(false)]; + tensor var_9262_cast_fp16 = concat(axis = var_7817, interleave = var_9262_interleave_0, values = (var_9114_cast_fp16, var_9116_cast_fp16, var_9118_cast_fp16, var_9120_cast_fp16))[name = string("op_9262_cast_fp16")]; + bool var_9264_interleave_0 = const()[name = string("op_9264_interleave_0"), val = bool(false)]; + tensor var_9264_cast_fp16 = concat(axis = var_7817, interleave = var_9264_interleave_0, values = (var_9122_cast_fp16, var_9124_cast_fp16, var_9126_cast_fp16, var_9128_cast_fp16))[name = string("op_9264_cast_fp16")]; + bool var_9266_interleave_0 = const()[name = string("op_9266_interleave_0"), val = bool(false)]; + tensor var_9266_cast_fp16 = concat(axis = var_7817, interleave = var_9266_interleave_0, values = (var_9130_cast_fp16, var_9132_cast_fp16, var_9134_cast_fp16, var_9136_cast_fp16))[name = string("op_9266_cast_fp16")]; + bool var_9268_interleave_0 = const()[name = string("op_9268_interleave_0"), val = bool(false)]; + tensor var_9268_cast_fp16 = concat(axis = var_7817, interleave = var_9268_interleave_0, values = (var_9138_cast_fp16, var_9140_cast_fp16, var_9142_cast_fp16, var_9144_cast_fp16))[name = string("op_9268_cast_fp16")]; + bool var_9270_interleave_0 = const()[name = string("op_9270_interleave_0"), val = bool(false)]; + tensor var_9270_cast_fp16 = concat(axis = var_7817, interleave = var_9270_interleave_0, values = (var_9146_cast_fp16, var_9148_cast_fp16, var_9150_cast_fp16, var_9152_cast_fp16))[name = string("op_9270_cast_fp16")]; + bool var_9272_interleave_0 = const()[name = string("op_9272_interleave_0"), val = bool(false)]; + tensor var_9272_cast_fp16 = concat(axis = var_7817, interleave = var_9272_interleave_0, values = (var_9154_cast_fp16, var_9156_cast_fp16, var_9158_cast_fp16, var_9160_cast_fp16))[name = string("op_9272_cast_fp16")]; + bool var_9274_interleave_0 = const()[name = string("op_9274_interleave_0"), val = bool(false)]; + tensor var_9274_cast_fp16 = concat(axis = var_7817, interleave = var_9274_interleave_0, values = (var_9162_cast_fp16, var_9164_cast_fp16, var_9166_cast_fp16, var_9168_cast_fp16))[name = string("op_9274_cast_fp16")]; + bool var_9276_interleave_0 = const()[name = string("op_9276_interleave_0"), val = bool(false)]; + tensor var_9276_cast_fp16 = concat(axis = var_7817, interleave = var_9276_interleave_0, values = (var_9170_cast_fp16, var_9172_cast_fp16, var_9174_cast_fp16, var_9176_cast_fp16))[name = string("op_9276_cast_fp16")]; + bool var_9278_interleave_0 = const()[name = string("op_9278_interleave_0"), val = bool(false)]; + tensor var_9278_cast_fp16 = concat(axis = var_7817, interleave = var_9278_interleave_0, values = (var_9178_cast_fp16, var_9180_cast_fp16, var_9182_cast_fp16, var_9184_cast_fp16))[name = string("op_9278_cast_fp16")]; + bool var_9280_interleave_0 = const()[name = string("op_9280_interleave_0"), val = bool(false)]; + tensor var_9280_cast_fp16 = concat(axis = var_7817, interleave = var_9280_interleave_0, values = (var_9186_cast_fp16, var_9188_cast_fp16, var_9190_cast_fp16, var_9192_cast_fp16))[name = string("op_9280_cast_fp16")]; + bool var_9282_interleave_0 = const()[name = string("op_9282_interleave_0"), val = bool(false)]; + tensor var_9282_cast_fp16 = concat(axis = var_7817, interleave = var_9282_interleave_0, values = (var_9194_cast_fp16, var_9196_cast_fp16, var_9198_cast_fp16, var_9200_cast_fp16))[name = string("op_9282_cast_fp16")]; + bool var_9284_interleave_0 = const()[name = string("op_9284_interleave_0"), val = bool(false)]; + tensor var_9284_cast_fp16 = concat(axis = var_7817, interleave = var_9284_interleave_0, values = (var_9202_cast_fp16, var_9204_cast_fp16, var_9206_cast_fp16, var_9208_cast_fp16))[name = string("op_9284_cast_fp16")]; + bool var_9286_interleave_0 = const()[name = string("op_9286_interleave_0"), val = bool(false)]; + tensor var_9286_cast_fp16 = concat(axis = var_7817, interleave = var_9286_interleave_0, values = (var_9210_cast_fp16, var_9212_cast_fp16, var_9214_cast_fp16, var_9216_cast_fp16))[name = string("op_9286_cast_fp16")]; + bool var_9288_interleave_0 = const()[name = string("op_9288_interleave_0"), val = bool(false)]; + tensor var_9288_cast_fp16 = concat(axis = var_7817, interleave = var_9288_interleave_0, values = (var_9218_cast_fp16, var_9220_cast_fp16, var_9222_cast_fp16, var_9224_cast_fp16))[name = string("op_9288_cast_fp16")]; + bool var_9290_interleave_0 = const()[name = string("op_9290_interleave_0"), val = bool(false)]; + tensor var_9290_cast_fp16 = concat(axis = var_7817, interleave = var_9290_interleave_0, values = (var_9226_cast_fp16, var_9228_cast_fp16, var_9230_cast_fp16, var_9232_cast_fp16))[name = string("op_9290_cast_fp16")]; + bool var_9292_interleave_0 = const()[name = string("op_9292_interleave_0"), val = bool(false)]; + tensor var_9292_cast_fp16 = concat(axis = var_7817, interleave = var_9292_interleave_0, values = (var_9234_cast_fp16, var_9236_cast_fp16, var_9238_cast_fp16, var_9240_cast_fp16))[name = string("op_9292_cast_fp16")]; + bool var_9294_interleave_0 = const()[name = string("op_9294_interleave_0"), val = bool(false)]; + tensor var_9294_cast_fp16 = concat(axis = var_7817, interleave = var_9294_interleave_0, values = (var_9242_cast_fp16, var_9244_cast_fp16, var_9246_cast_fp16, var_9248_cast_fp16))[name = string("op_9294_cast_fp16")]; + bool var_9296_interleave_0 = const()[name = string("op_9296_interleave_0"), val = bool(false)]; + tensor var_9296_cast_fp16 = concat(axis = var_7817, interleave = var_9296_interleave_0, values = (var_9250_cast_fp16, var_9252_cast_fp16, var_9254_cast_fp16, var_9256_cast_fp16))[name = string("op_9296_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_7842, interleave = input_41_interleave_0, values = (var_9258_cast_fp16, var_9260_cast_fp16, var_9262_cast_fp16, var_9264_cast_fp16, var_9266_cast_fp16, var_9268_cast_fp16, var_9270_cast_fp16, var_9272_cast_fp16, var_9274_cast_fp16, var_9276_cast_fp16, var_9278_cast_fp16, var_9280_cast_fp16, var_9282_cast_fp16, var_9284_cast_fp16, var_9286_cast_fp16, var_9288_cast_fp16, var_9290_cast_fp16, var_9292_cast_fp16, var_9294_cast_fp16, var_9296_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; + tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; + tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; + int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221271680)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224548544)))]; + tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_9315_to_fp16 = const()[name = string("op_9315_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_9315_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224551168)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224553792)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224556416)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237663680)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237673984)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250781248)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; + int32 var_9344 = const()[name = string("op_9344"), val = int32(3)]; + int32 var_9369 = const()[name = string("op_9369"), val = int32(1)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_9386_to_fp16 = const()[name = string("op_9386_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_9386_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250783872)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250786496)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250789120)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254065984)))]; + tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; + string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; + tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; + tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; + int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(254068608)))]; + tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; + string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; + tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; + tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; + int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257345472)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260622336)))]; + tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_9424_begin_0 = const()[name = string("op_9424_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9424_end_0 = const()[name = string("op_9424_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9424_end_mask_0 = const()[name = string("op_9424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9424_cast_fp16 = slice_by_index(begin = var_9424_begin_0, end = var_9424_end_0, end_mask = var_9424_end_mask_0, x = query_13_cast_fp16)[name = string("op_9424_cast_fp16")]; + tensor var_9428_begin_0 = const()[name = string("op_9428_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9428_end_0 = const()[name = string("op_9428_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9428_end_mask_0 = const()[name = string("op_9428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9428_cast_fp16 = slice_by_index(begin = var_9428_begin_0, end = var_9428_end_0, end_mask = var_9428_end_mask_0, x = query_13_cast_fp16)[name = string("op_9428_cast_fp16")]; + tensor var_9432_begin_0 = const()[name = string("op_9432_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9432_end_0 = const()[name = string("op_9432_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9432_end_mask_0 = const()[name = string("op_9432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9432_cast_fp16 = slice_by_index(begin = var_9432_begin_0, end = var_9432_end_0, end_mask = var_9432_end_mask_0, x = query_13_cast_fp16)[name = string("op_9432_cast_fp16")]; + tensor var_9436_begin_0 = const()[name = string("op_9436_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9436_end_0 = const()[name = string("op_9436_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9436_end_mask_0 = const()[name = string("op_9436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9436_cast_fp16 = slice_by_index(begin = var_9436_begin_0, end = var_9436_end_0, end_mask = var_9436_end_mask_0, x = query_13_cast_fp16)[name = string("op_9436_cast_fp16")]; + tensor var_9440_begin_0 = const()[name = string("op_9440_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9440_end_0 = const()[name = string("op_9440_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9440_end_mask_0 = const()[name = string("op_9440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9440_cast_fp16 = slice_by_index(begin = var_9440_begin_0, end = var_9440_end_0, end_mask = var_9440_end_mask_0, x = query_13_cast_fp16)[name = string("op_9440_cast_fp16")]; + tensor var_9444_begin_0 = const()[name = string("op_9444_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9444_end_0 = const()[name = string("op_9444_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9444_end_mask_0 = const()[name = string("op_9444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9444_cast_fp16 = slice_by_index(begin = var_9444_begin_0, end = var_9444_end_0, end_mask = var_9444_end_mask_0, x = query_13_cast_fp16)[name = string("op_9444_cast_fp16")]; + tensor var_9448_begin_0 = const()[name = string("op_9448_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9448_end_0 = const()[name = string("op_9448_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9448_end_mask_0 = const()[name = string("op_9448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9448_cast_fp16 = slice_by_index(begin = var_9448_begin_0, end = var_9448_end_0, end_mask = var_9448_end_mask_0, x = query_13_cast_fp16)[name = string("op_9448_cast_fp16")]; + tensor var_9452_begin_0 = const()[name = string("op_9452_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9452_end_0 = const()[name = string("op_9452_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9452_end_mask_0 = const()[name = string("op_9452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9452_cast_fp16 = slice_by_index(begin = var_9452_begin_0, end = var_9452_end_0, end_mask = var_9452_end_mask_0, x = query_13_cast_fp16)[name = string("op_9452_cast_fp16")]; + tensor var_9456_begin_0 = const()[name = string("op_9456_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9456_end_0 = const()[name = string("op_9456_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9456_end_mask_0 = const()[name = string("op_9456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9456_cast_fp16 = slice_by_index(begin = var_9456_begin_0, end = var_9456_end_0, end_mask = var_9456_end_mask_0, x = query_13_cast_fp16)[name = string("op_9456_cast_fp16")]; + tensor var_9460_begin_0 = const()[name = string("op_9460_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9460_end_0 = const()[name = string("op_9460_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9460_end_mask_0 = const()[name = string("op_9460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9460_cast_fp16 = slice_by_index(begin = var_9460_begin_0, end = var_9460_end_0, end_mask = var_9460_end_mask_0, x = query_13_cast_fp16)[name = string("op_9460_cast_fp16")]; + tensor var_9464_begin_0 = const()[name = string("op_9464_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9464_end_0 = const()[name = string("op_9464_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9464_end_mask_0 = const()[name = string("op_9464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9464_cast_fp16 = slice_by_index(begin = var_9464_begin_0, end = var_9464_end_0, end_mask = var_9464_end_mask_0, x = query_13_cast_fp16)[name = string("op_9464_cast_fp16")]; + tensor var_9468_begin_0 = const()[name = string("op_9468_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9468_end_0 = const()[name = string("op_9468_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9468_end_mask_0 = const()[name = string("op_9468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9468_cast_fp16 = slice_by_index(begin = var_9468_begin_0, end = var_9468_end_0, end_mask = var_9468_end_mask_0, x = query_13_cast_fp16)[name = string("op_9468_cast_fp16")]; + tensor var_9472_begin_0 = const()[name = string("op_9472_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9472_end_0 = const()[name = string("op_9472_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_9472_end_mask_0 = const()[name = string("op_9472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9472_cast_fp16 = slice_by_index(begin = var_9472_begin_0, end = var_9472_end_0, end_mask = var_9472_end_mask_0, x = query_13_cast_fp16)[name = string("op_9472_cast_fp16")]; + tensor var_9476_begin_0 = const()[name = string("op_9476_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9476_end_0 = const()[name = string("op_9476_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_9476_end_mask_0 = const()[name = string("op_9476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9476_cast_fp16 = slice_by_index(begin = var_9476_begin_0, end = var_9476_end_0, end_mask = var_9476_end_mask_0, x = query_13_cast_fp16)[name = string("op_9476_cast_fp16")]; + tensor var_9480_begin_0 = const()[name = string("op_9480_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9480_end_0 = const()[name = string("op_9480_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_9480_end_mask_0 = const()[name = string("op_9480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9480_cast_fp16 = slice_by_index(begin = var_9480_begin_0, end = var_9480_end_0, end_mask = var_9480_end_mask_0, x = query_13_cast_fp16)[name = string("op_9480_cast_fp16")]; + tensor var_9484_begin_0 = const()[name = string("op_9484_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9484_end_0 = const()[name = string("op_9484_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_9484_end_mask_0 = const()[name = string("op_9484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9484_cast_fp16 = slice_by_index(begin = var_9484_begin_0, end = var_9484_end_0, end_mask = var_9484_end_mask_0, x = query_13_cast_fp16)[name = string("op_9484_cast_fp16")]; + tensor var_9488_begin_0 = const()[name = string("op_9488_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9488_end_0 = const()[name = string("op_9488_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_9488_end_mask_0 = const()[name = string("op_9488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9488_cast_fp16 = slice_by_index(begin = var_9488_begin_0, end = var_9488_end_0, end_mask = var_9488_end_mask_0, x = query_13_cast_fp16)[name = string("op_9488_cast_fp16")]; + tensor var_9492_begin_0 = const()[name = string("op_9492_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9492_end_0 = const()[name = string("op_9492_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_9492_end_mask_0 = const()[name = string("op_9492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9492_cast_fp16 = slice_by_index(begin = var_9492_begin_0, end = var_9492_end_0, end_mask = var_9492_end_mask_0, x = query_13_cast_fp16)[name = string("op_9492_cast_fp16")]; + tensor var_9496_begin_0 = const()[name = string("op_9496_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9496_end_0 = const()[name = string("op_9496_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_9496_end_mask_0 = const()[name = string("op_9496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9496_cast_fp16 = slice_by_index(begin = var_9496_begin_0, end = var_9496_end_0, end_mask = var_9496_end_mask_0, x = query_13_cast_fp16)[name = string("op_9496_cast_fp16")]; + tensor var_9500_begin_0 = const()[name = string("op_9500_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9500_end_0 = const()[name = string("op_9500_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_9500_end_mask_0 = const()[name = string("op_9500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9500_cast_fp16 = slice_by_index(begin = var_9500_begin_0, end = var_9500_end_0, end_mask = var_9500_end_mask_0, x = query_13_cast_fp16)[name = string("op_9500_cast_fp16")]; + tensor var_9509_begin_0 = const()[name = string("op_9509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9509_end_0 = const()[name = string("op_9509_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9509_end_mask_0 = const()[name = string("op_9509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9509_cast_fp16 = slice_by_index(begin = var_9509_begin_0, end = var_9509_end_0, end_mask = var_9509_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9509_cast_fp16")]; + tensor var_9516_begin_0 = const()[name = string("op_9516_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9516_end_0 = const()[name = string("op_9516_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9516_end_mask_0 = const()[name = string("op_9516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9516_cast_fp16 = slice_by_index(begin = var_9516_begin_0, end = var_9516_end_0, end_mask = var_9516_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9516_cast_fp16")]; + tensor var_9523_begin_0 = const()[name = string("op_9523_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9523_end_0 = const()[name = string("op_9523_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9523_end_mask_0 = const()[name = string("op_9523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9523_cast_fp16 = slice_by_index(begin = var_9523_begin_0, end = var_9523_end_0, end_mask = var_9523_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9523_cast_fp16")]; + tensor var_9530_begin_0 = const()[name = string("op_9530_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9530_end_0 = const()[name = string("op_9530_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9530_end_mask_0 = const()[name = string("op_9530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9530_cast_fp16 = slice_by_index(begin = var_9530_begin_0, end = var_9530_end_0, end_mask = var_9530_end_mask_0, x = var_9424_cast_fp16)[name = string("op_9530_cast_fp16")]; + tensor var_9537_begin_0 = const()[name = string("op_9537_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9537_end_0 = const()[name = string("op_9537_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9537_end_mask_0 = const()[name = string("op_9537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9537_cast_fp16 = slice_by_index(begin = var_9537_begin_0, end = var_9537_end_0, end_mask = var_9537_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9537_cast_fp16")]; + tensor var_9544_begin_0 = const()[name = string("op_9544_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9544_end_0 = const()[name = string("op_9544_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9544_end_mask_0 = const()[name = string("op_9544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9544_cast_fp16 = slice_by_index(begin = var_9544_begin_0, end = var_9544_end_0, end_mask = var_9544_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9544_cast_fp16")]; + tensor var_9551_begin_0 = const()[name = string("op_9551_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9551_end_0 = const()[name = string("op_9551_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9551_end_mask_0 = const()[name = string("op_9551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9551_cast_fp16 = slice_by_index(begin = var_9551_begin_0, end = var_9551_end_0, end_mask = var_9551_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9551_cast_fp16")]; + tensor var_9558_begin_0 = const()[name = string("op_9558_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9558_end_0 = const()[name = string("op_9558_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9558_end_mask_0 = const()[name = string("op_9558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9558_cast_fp16 = slice_by_index(begin = var_9558_begin_0, end = var_9558_end_0, end_mask = var_9558_end_mask_0, x = var_9428_cast_fp16)[name = string("op_9558_cast_fp16")]; + tensor var_9565_begin_0 = const()[name = string("op_9565_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9565_end_0 = const()[name = string("op_9565_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9565_end_mask_0 = const()[name = string("op_9565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9565_cast_fp16 = slice_by_index(begin = var_9565_begin_0, end = var_9565_end_0, end_mask = var_9565_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9565_cast_fp16")]; + tensor var_9572_begin_0 = const()[name = string("op_9572_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9572_end_0 = const()[name = string("op_9572_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9572_end_mask_0 = const()[name = string("op_9572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9572_cast_fp16 = slice_by_index(begin = var_9572_begin_0, end = var_9572_end_0, end_mask = var_9572_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9572_cast_fp16")]; + tensor var_9579_begin_0 = const()[name = string("op_9579_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9579_end_0 = const()[name = string("op_9579_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9579_end_mask_0 = const()[name = string("op_9579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9579_cast_fp16 = slice_by_index(begin = var_9579_begin_0, end = var_9579_end_0, end_mask = var_9579_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9579_cast_fp16")]; + tensor var_9586_begin_0 = const()[name = string("op_9586_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9586_end_0 = const()[name = string("op_9586_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9586_end_mask_0 = const()[name = string("op_9586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9586_cast_fp16 = slice_by_index(begin = var_9586_begin_0, end = var_9586_end_0, end_mask = var_9586_end_mask_0, x = var_9432_cast_fp16)[name = string("op_9586_cast_fp16")]; + tensor var_9593_begin_0 = const()[name = string("op_9593_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9593_end_0 = const()[name = string("op_9593_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9593_end_mask_0 = const()[name = string("op_9593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9593_cast_fp16 = slice_by_index(begin = var_9593_begin_0, end = var_9593_end_0, end_mask = var_9593_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9593_cast_fp16")]; + tensor var_9600_begin_0 = const()[name = string("op_9600_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9600_end_0 = const()[name = string("op_9600_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9600_end_mask_0 = const()[name = string("op_9600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9600_cast_fp16 = slice_by_index(begin = var_9600_begin_0, end = var_9600_end_0, end_mask = var_9600_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9600_cast_fp16")]; + tensor var_9607_begin_0 = const()[name = string("op_9607_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9607_end_0 = const()[name = string("op_9607_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9607_end_mask_0 = const()[name = string("op_9607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9607_cast_fp16 = slice_by_index(begin = var_9607_begin_0, end = var_9607_end_0, end_mask = var_9607_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9607_cast_fp16")]; + tensor var_9614_begin_0 = const()[name = string("op_9614_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9614_end_0 = const()[name = string("op_9614_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9614_end_mask_0 = const()[name = string("op_9614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9614_cast_fp16 = slice_by_index(begin = var_9614_begin_0, end = var_9614_end_0, end_mask = var_9614_end_mask_0, x = var_9436_cast_fp16)[name = string("op_9614_cast_fp16")]; + tensor var_9621_begin_0 = const()[name = string("op_9621_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9621_end_0 = const()[name = string("op_9621_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9621_end_mask_0 = const()[name = string("op_9621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9621_cast_fp16 = slice_by_index(begin = var_9621_begin_0, end = var_9621_end_0, end_mask = var_9621_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9621_cast_fp16")]; + tensor var_9628_begin_0 = const()[name = string("op_9628_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9628_end_0 = const()[name = string("op_9628_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9628_end_mask_0 = const()[name = string("op_9628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9628_cast_fp16 = slice_by_index(begin = var_9628_begin_0, end = var_9628_end_0, end_mask = var_9628_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9628_cast_fp16")]; + tensor var_9635_begin_0 = const()[name = string("op_9635_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9635_end_0 = const()[name = string("op_9635_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9635_end_mask_0 = const()[name = string("op_9635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9635_cast_fp16 = slice_by_index(begin = var_9635_begin_0, end = var_9635_end_0, end_mask = var_9635_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9635_cast_fp16")]; + tensor var_9642_begin_0 = const()[name = string("op_9642_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9642_end_0 = const()[name = string("op_9642_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9642_end_mask_0 = const()[name = string("op_9642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9642_cast_fp16 = slice_by_index(begin = var_9642_begin_0, end = var_9642_end_0, end_mask = var_9642_end_mask_0, x = var_9440_cast_fp16)[name = string("op_9642_cast_fp16")]; + tensor var_9649_begin_0 = const()[name = string("op_9649_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9649_end_0 = const()[name = string("op_9649_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9649_end_mask_0 = const()[name = string("op_9649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9649_cast_fp16 = slice_by_index(begin = var_9649_begin_0, end = var_9649_end_0, end_mask = var_9649_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9649_cast_fp16")]; + tensor var_9656_begin_0 = const()[name = string("op_9656_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9656_end_0 = const()[name = string("op_9656_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9656_end_mask_0 = const()[name = string("op_9656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9656_cast_fp16 = slice_by_index(begin = var_9656_begin_0, end = var_9656_end_0, end_mask = var_9656_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9656_cast_fp16")]; + tensor var_9663_begin_0 = const()[name = string("op_9663_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9663_end_0 = const()[name = string("op_9663_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9663_end_mask_0 = const()[name = string("op_9663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9663_cast_fp16 = slice_by_index(begin = var_9663_begin_0, end = var_9663_end_0, end_mask = var_9663_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9663_cast_fp16")]; + tensor var_9670_begin_0 = const()[name = string("op_9670_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9670_end_0 = const()[name = string("op_9670_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9670_end_mask_0 = const()[name = string("op_9670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9670_cast_fp16 = slice_by_index(begin = var_9670_begin_0, end = var_9670_end_0, end_mask = var_9670_end_mask_0, x = var_9444_cast_fp16)[name = string("op_9670_cast_fp16")]; + tensor var_9677_begin_0 = const()[name = string("op_9677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9677_end_0 = const()[name = string("op_9677_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9677_end_mask_0 = const()[name = string("op_9677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9677_cast_fp16 = slice_by_index(begin = var_9677_begin_0, end = var_9677_end_0, end_mask = var_9677_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9677_cast_fp16")]; + tensor var_9684_begin_0 = const()[name = string("op_9684_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9684_end_0 = const()[name = string("op_9684_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9684_end_mask_0 = const()[name = string("op_9684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9684_cast_fp16 = slice_by_index(begin = var_9684_begin_0, end = var_9684_end_0, end_mask = var_9684_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9684_cast_fp16")]; + tensor var_9691_begin_0 = const()[name = string("op_9691_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9691_end_0 = const()[name = string("op_9691_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9691_end_mask_0 = const()[name = string("op_9691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9691_cast_fp16 = slice_by_index(begin = var_9691_begin_0, end = var_9691_end_0, end_mask = var_9691_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9691_cast_fp16")]; + tensor var_9698_begin_0 = const()[name = string("op_9698_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9698_end_0 = const()[name = string("op_9698_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9698_end_mask_0 = const()[name = string("op_9698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9698_cast_fp16 = slice_by_index(begin = var_9698_begin_0, end = var_9698_end_0, end_mask = var_9698_end_mask_0, x = var_9448_cast_fp16)[name = string("op_9698_cast_fp16")]; + tensor var_9705_begin_0 = const()[name = string("op_9705_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9705_end_0 = const()[name = string("op_9705_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9705_end_mask_0 = const()[name = string("op_9705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9705_cast_fp16 = slice_by_index(begin = var_9705_begin_0, end = var_9705_end_0, end_mask = var_9705_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9705_cast_fp16")]; + tensor var_9712_begin_0 = const()[name = string("op_9712_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9712_end_0 = const()[name = string("op_9712_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9712_end_mask_0 = const()[name = string("op_9712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9712_cast_fp16 = slice_by_index(begin = var_9712_begin_0, end = var_9712_end_0, end_mask = var_9712_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9712_cast_fp16")]; + tensor var_9719_begin_0 = const()[name = string("op_9719_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9719_end_0 = const()[name = string("op_9719_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9719_end_mask_0 = const()[name = string("op_9719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9719_cast_fp16 = slice_by_index(begin = var_9719_begin_0, end = var_9719_end_0, end_mask = var_9719_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9719_cast_fp16")]; + tensor var_9726_begin_0 = const()[name = string("op_9726_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9726_end_0 = const()[name = string("op_9726_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9726_end_mask_0 = const()[name = string("op_9726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9726_cast_fp16 = slice_by_index(begin = var_9726_begin_0, end = var_9726_end_0, end_mask = var_9726_end_mask_0, x = var_9452_cast_fp16)[name = string("op_9726_cast_fp16")]; + tensor var_9733_begin_0 = const()[name = string("op_9733_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9733_end_0 = const()[name = string("op_9733_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9733_end_mask_0 = const()[name = string("op_9733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9733_cast_fp16 = slice_by_index(begin = var_9733_begin_0, end = var_9733_end_0, end_mask = var_9733_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9733_cast_fp16")]; + tensor var_9740_begin_0 = const()[name = string("op_9740_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9740_end_0 = const()[name = string("op_9740_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9740_end_mask_0 = const()[name = string("op_9740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9740_cast_fp16 = slice_by_index(begin = var_9740_begin_0, end = var_9740_end_0, end_mask = var_9740_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9740_cast_fp16")]; + tensor var_9747_begin_0 = const()[name = string("op_9747_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9747_end_0 = const()[name = string("op_9747_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9747_end_mask_0 = const()[name = string("op_9747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9747_cast_fp16 = slice_by_index(begin = var_9747_begin_0, end = var_9747_end_0, end_mask = var_9747_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9747_cast_fp16")]; + tensor var_9754_begin_0 = const()[name = string("op_9754_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9754_end_0 = const()[name = string("op_9754_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9754_end_mask_0 = const()[name = string("op_9754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9754_cast_fp16 = slice_by_index(begin = var_9754_begin_0, end = var_9754_end_0, end_mask = var_9754_end_mask_0, x = var_9456_cast_fp16)[name = string("op_9754_cast_fp16")]; + tensor var_9761_begin_0 = const()[name = string("op_9761_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9761_end_0 = const()[name = string("op_9761_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9761_end_mask_0 = const()[name = string("op_9761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9761_cast_fp16 = slice_by_index(begin = var_9761_begin_0, end = var_9761_end_0, end_mask = var_9761_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9761_cast_fp16")]; + tensor var_9768_begin_0 = const()[name = string("op_9768_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9768_end_0 = const()[name = string("op_9768_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9768_end_mask_0 = const()[name = string("op_9768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9768_cast_fp16 = slice_by_index(begin = var_9768_begin_0, end = var_9768_end_0, end_mask = var_9768_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9768_cast_fp16")]; + tensor var_9775_begin_0 = const()[name = string("op_9775_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9775_end_0 = const()[name = string("op_9775_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9775_end_mask_0 = const()[name = string("op_9775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9775_cast_fp16 = slice_by_index(begin = var_9775_begin_0, end = var_9775_end_0, end_mask = var_9775_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9775_cast_fp16")]; + tensor var_9782_begin_0 = const()[name = string("op_9782_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9782_end_0 = const()[name = string("op_9782_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9782_end_mask_0 = const()[name = string("op_9782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9782_cast_fp16 = slice_by_index(begin = var_9782_begin_0, end = var_9782_end_0, end_mask = var_9782_end_mask_0, x = var_9460_cast_fp16)[name = string("op_9782_cast_fp16")]; + tensor var_9789_begin_0 = const()[name = string("op_9789_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9789_end_0 = const()[name = string("op_9789_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9789_end_mask_0 = const()[name = string("op_9789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9789_cast_fp16 = slice_by_index(begin = var_9789_begin_0, end = var_9789_end_0, end_mask = var_9789_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9789_cast_fp16")]; + tensor var_9796_begin_0 = const()[name = string("op_9796_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9796_end_0 = const()[name = string("op_9796_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9796_end_mask_0 = const()[name = string("op_9796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9796_cast_fp16 = slice_by_index(begin = var_9796_begin_0, end = var_9796_end_0, end_mask = var_9796_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9796_cast_fp16")]; + tensor var_9803_begin_0 = const()[name = string("op_9803_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9803_end_0 = const()[name = string("op_9803_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9803_end_mask_0 = const()[name = string("op_9803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9803_cast_fp16 = slice_by_index(begin = var_9803_begin_0, end = var_9803_end_0, end_mask = var_9803_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9803_cast_fp16")]; + tensor var_9810_begin_0 = const()[name = string("op_9810_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9810_end_0 = const()[name = string("op_9810_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9810_end_mask_0 = const()[name = string("op_9810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9810_cast_fp16 = slice_by_index(begin = var_9810_begin_0, end = var_9810_end_0, end_mask = var_9810_end_mask_0, x = var_9464_cast_fp16)[name = string("op_9810_cast_fp16")]; + tensor var_9817_begin_0 = const()[name = string("op_9817_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9817_end_0 = const()[name = string("op_9817_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9817_end_mask_0 = const()[name = string("op_9817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9817_cast_fp16 = slice_by_index(begin = var_9817_begin_0, end = var_9817_end_0, end_mask = var_9817_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9817_cast_fp16")]; + tensor var_9824_begin_0 = const()[name = string("op_9824_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9824_end_0 = const()[name = string("op_9824_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9824_end_mask_0 = const()[name = string("op_9824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9824_cast_fp16 = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9824_cast_fp16")]; + tensor var_9831_begin_0 = const()[name = string("op_9831_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9831_end_0 = const()[name = string("op_9831_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9831_end_mask_0 = const()[name = string("op_9831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9831_cast_fp16 = slice_by_index(begin = var_9831_begin_0, end = var_9831_end_0, end_mask = var_9831_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9831_cast_fp16")]; + tensor var_9838_begin_0 = const()[name = string("op_9838_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9838_end_0 = const()[name = string("op_9838_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9838_end_mask_0 = const()[name = string("op_9838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9838_cast_fp16 = slice_by_index(begin = var_9838_begin_0, end = var_9838_end_0, end_mask = var_9838_end_mask_0, x = var_9468_cast_fp16)[name = string("op_9838_cast_fp16")]; + tensor var_9845_begin_0 = const()[name = string("op_9845_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9845_end_0 = const()[name = string("op_9845_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9845_end_mask_0 = const()[name = string("op_9845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9845_cast_fp16 = slice_by_index(begin = var_9845_begin_0, end = var_9845_end_0, end_mask = var_9845_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9845_cast_fp16")]; + tensor var_9852_begin_0 = const()[name = string("op_9852_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9852_end_0 = const()[name = string("op_9852_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9852_end_mask_0 = const()[name = string("op_9852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9852_cast_fp16 = slice_by_index(begin = var_9852_begin_0, end = var_9852_end_0, end_mask = var_9852_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9852_cast_fp16")]; + tensor var_9859_begin_0 = const()[name = string("op_9859_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9859_end_0 = const()[name = string("op_9859_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9859_end_mask_0 = const()[name = string("op_9859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9859_cast_fp16 = slice_by_index(begin = var_9859_begin_0, end = var_9859_end_0, end_mask = var_9859_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9859_cast_fp16")]; + tensor var_9866_begin_0 = const()[name = string("op_9866_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9866_end_0 = const()[name = string("op_9866_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9866_end_mask_0 = const()[name = string("op_9866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9866_cast_fp16 = slice_by_index(begin = var_9866_begin_0, end = var_9866_end_0, end_mask = var_9866_end_mask_0, x = var_9472_cast_fp16)[name = string("op_9866_cast_fp16")]; + tensor var_9873_begin_0 = const()[name = string("op_9873_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9873_end_0 = const()[name = string("op_9873_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9873_end_mask_0 = const()[name = string("op_9873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9873_cast_fp16 = slice_by_index(begin = var_9873_begin_0, end = var_9873_end_0, end_mask = var_9873_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9873_cast_fp16")]; + tensor var_9880_begin_0 = const()[name = string("op_9880_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9880_end_0 = const()[name = string("op_9880_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9880_end_mask_0 = const()[name = string("op_9880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9880_cast_fp16 = slice_by_index(begin = var_9880_begin_0, end = var_9880_end_0, end_mask = var_9880_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9880_cast_fp16")]; + tensor var_9887_begin_0 = const()[name = string("op_9887_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9887_end_0 = const()[name = string("op_9887_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9887_end_mask_0 = const()[name = string("op_9887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9887_cast_fp16 = slice_by_index(begin = var_9887_begin_0, end = var_9887_end_0, end_mask = var_9887_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9887_cast_fp16")]; + tensor var_9894_begin_0 = const()[name = string("op_9894_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9894_end_0 = const()[name = string("op_9894_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9894_end_mask_0 = const()[name = string("op_9894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9894_cast_fp16 = slice_by_index(begin = var_9894_begin_0, end = var_9894_end_0, end_mask = var_9894_end_mask_0, x = var_9476_cast_fp16)[name = string("op_9894_cast_fp16")]; + tensor var_9901_begin_0 = const()[name = string("op_9901_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9901_end_0 = const()[name = string("op_9901_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9901_end_mask_0 = const()[name = string("op_9901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9901_cast_fp16 = slice_by_index(begin = var_9901_begin_0, end = var_9901_end_0, end_mask = var_9901_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9901_cast_fp16")]; + tensor var_9908_begin_0 = const()[name = string("op_9908_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9908_end_0 = const()[name = string("op_9908_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9908_end_mask_0 = const()[name = string("op_9908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9908_cast_fp16 = slice_by_index(begin = var_9908_begin_0, end = var_9908_end_0, end_mask = var_9908_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9908_cast_fp16")]; + tensor var_9915_begin_0 = const()[name = string("op_9915_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9915_end_0 = const()[name = string("op_9915_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9915_end_mask_0 = const()[name = string("op_9915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9915_cast_fp16 = slice_by_index(begin = var_9915_begin_0, end = var_9915_end_0, end_mask = var_9915_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9915_cast_fp16")]; + tensor var_9922_begin_0 = const()[name = string("op_9922_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9922_end_0 = const()[name = string("op_9922_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9922_end_mask_0 = const()[name = string("op_9922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9922_cast_fp16 = slice_by_index(begin = var_9922_begin_0, end = var_9922_end_0, end_mask = var_9922_end_mask_0, x = var_9480_cast_fp16)[name = string("op_9922_cast_fp16")]; + tensor var_9929_begin_0 = const()[name = string("op_9929_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9929_end_0 = const()[name = string("op_9929_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9929_end_mask_0 = const()[name = string("op_9929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9929_cast_fp16 = slice_by_index(begin = var_9929_begin_0, end = var_9929_end_0, end_mask = var_9929_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9929_cast_fp16")]; + tensor var_9936_begin_0 = const()[name = string("op_9936_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9936_end_0 = const()[name = string("op_9936_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9936_end_mask_0 = const()[name = string("op_9936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9936_cast_fp16 = slice_by_index(begin = var_9936_begin_0, end = var_9936_end_0, end_mask = var_9936_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9936_cast_fp16")]; + tensor var_9943_begin_0 = const()[name = string("op_9943_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9943_end_0 = const()[name = string("op_9943_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9943_end_mask_0 = const()[name = string("op_9943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9943_cast_fp16 = slice_by_index(begin = var_9943_begin_0, end = var_9943_end_0, end_mask = var_9943_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9943_cast_fp16")]; + tensor var_9950_begin_0 = const()[name = string("op_9950_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9950_end_0 = const()[name = string("op_9950_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9950_end_mask_0 = const()[name = string("op_9950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9950_cast_fp16 = slice_by_index(begin = var_9950_begin_0, end = var_9950_end_0, end_mask = var_9950_end_mask_0, x = var_9484_cast_fp16)[name = string("op_9950_cast_fp16")]; + tensor var_9957_begin_0 = const()[name = string("op_9957_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9957_end_0 = const()[name = string("op_9957_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9957_end_mask_0 = const()[name = string("op_9957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9957_cast_fp16 = slice_by_index(begin = var_9957_begin_0, end = var_9957_end_0, end_mask = var_9957_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9957_cast_fp16")]; + tensor var_9964_begin_0 = const()[name = string("op_9964_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9964_end_0 = const()[name = string("op_9964_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9964_end_mask_0 = const()[name = string("op_9964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9964_cast_fp16 = slice_by_index(begin = var_9964_begin_0, end = var_9964_end_0, end_mask = var_9964_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9964_cast_fp16")]; + tensor var_9971_begin_0 = const()[name = string("op_9971_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9971_end_0 = const()[name = string("op_9971_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9971_end_mask_0 = const()[name = string("op_9971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9971_cast_fp16 = slice_by_index(begin = var_9971_begin_0, end = var_9971_end_0, end_mask = var_9971_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9971_cast_fp16")]; + tensor var_9978_begin_0 = const()[name = string("op_9978_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9978_end_0 = const()[name = string("op_9978_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9978_end_mask_0 = const()[name = string("op_9978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9978_cast_fp16 = slice_by_index(begin = var_9978_begin_0, end = var_9978_end_0, end_mask = var_9978_end_mask_0, x = var_9488_cast_fp16)[name = string("op_9978_cast_fp16")]; + tensor var_9985_begin_0 = const()[name = string("op_9985_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9985_end_0 = const()[name = string("op_9985_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9985_end_mask_0 = const()[name = string("op_9985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9985_cast_fp16 = slice_by_index(begin = var_9985_begin_0, end = var_9985_end_0, end_mask = var_9985_end_mask_0, x = var_9492_cast_fp16)[name = string("op_9985_cast_fp16")]; + tensor var_9992_begin_0 = const()[name = string("op_9992_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9992_end_0 = const()[name = string("op_9992_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9992_end_mask_0 = const()[name = string("op_9992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9992_cast_fp16 = slice_by_index(begin = var_9992_begin_0, end = var_9992_end_0, end_mask = var_9992_end_mask_0, x = var_9492_cast_fp16)[name = string("op_9992_cast_fp16")]; + tensor var_9999_begin_0 = const()[name = string("op_9999_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9999_end_0 = const()[name = string("op_9999_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9999_end_mask_0 = const()[name = string("op_9999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9999_cast_fp16 = slice_by_index(begin = var_9999_begin_0, end = var_9999_end_0, end_mask = var_9999_end_mask_0, x = var_9492_cast_fp16)[name = string("op_9999_cast_fp16")]; + tensor var_10006_begin_0 = const()[name = string("op_10006_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10006_end_0 = const()[name = string("op_10006_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10006_end_mask_0 = const()[name = string("op_10006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10006_cast_fp16 = slice_by_index(begin = var_10006_begin_0, end = var_10006_end_0, end_mask = var_10006_end_mask_0, x = var_9492_cast_fp16)[name = string("op_10006_cast_fp16")]; + tensor var_10013_begin_0 = const()[name = string("op_10013_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10013_end_0 = const()[name = string("op_10013_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10013_end_mask_0 = const()[name = string("op_10013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10013_cast_fp16 = slice_by_index(begin = var_10013_begin_0, end = var_10013_end_0, end_mask = var_10013_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10013_cast_fp16")]; + tensor var_10020_begin_0 = const()[name = string("op_10020_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10020_end_0 = const()[name = string("op_10020_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10020_end_mask_0 = const()[name = string("op_10020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10020_cast_fp16 = slice_by_index(begin = var_10020_begin_0, end = var_10020_end_0, end_mask = var_10020_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10020_cast_fp16")]; + tensor var_10027_begin_0 = const()[name = string("op_10027_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10027_end_0 = const()[name = string("op_10027_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10027_end_mask_0 = const()[name = string("op_10027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10027_cast_fp16 = slice_by_index(begin = var_10027_begin_0, end = var_10027_end_0, end_mask = var_10027_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10027_cast_fp16")]; + tensor var_10034_begin_0 = const()[name = string("op_10034_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10034_end_0 = const()[name = string("op_10034_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10034_end_mask_0 = const()[name = string("op_10034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10034_cast_fp16 = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = var_9496_cast_fp16)[name = string("op_10034_cast_fp16")]; + tensor var_10041_begin_0 = const()[name = string("op_10041_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10041_end_0 = const()[name = string("op_10041_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10041_end_mask_0 = const()[name = string("op_10041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10041_cast_fp16 = slice_by_index(begin = var_10041_begin_0, end = var_10041_end_0, end_mask = var_10041_end_mask_0, x = var_9500_cast_fp16)[name = string("op_10041_cast_fp16")]; + tensor var_10048_begin_0 = const()[name = string("op_10048_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10048_end_0 = const()[name = string("op_10048_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10048_end_mask_0 = const()[name = string("op_10048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10048_cast_fp16 = slice_by_index(begin = var_10048_begin_0, end = var_10048_end_0, end_mask = var_10048_end_mask_0, x = var_9500_cast_fp16)[name = string("op_10048_cast_fp16")]; + tensor var_10055_begin_0 = const()[name = string("op_10055_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10055_end_0 = const()[name = string("op_10055_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10055_end_mask_0 = const()[name = string("op_10055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10055_cast_fp16 = slice_by_index(begin = var_10055_begin_0, end = var_10055_end_0, end_mask = var_10055_end_mask_0, x = var_9500_cast_fp16)[name = string("op_10055_cast_fp16")]; + tensor var_10062_begin_0 = const()[name = string("op_10062_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10062_end_0 = const()[name = string("op_10062_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10062_end_mask_0 = const()[name = string("op_10062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10062_cast_fp16 = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = var_9500_cast_fp16)[name = string("op_10062_cast_fp16")]; + tensor k_13_perm_0 = const()[name = string("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10067_begin_0 = const()[name = string("op_10067_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10067_end_0 = const()[name = string("op_10067_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_10067_end_mask_0 = const()[name = string("op_10067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_13_cast_fp16 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = string("transpose_25")]; + tensor var_10067_cast_fp16 = slice_by_index(begin = var_10067_begin_0, end = var_10067_end_0, end_mask = var_10067_end_mask_0, x = k_13_cast_fp16)[name = string("op_10067_cast_fp16")]; + tensor var_10071_begin_0 = const()[name = string("op_10071_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10071_end_0 = const()[name = string("op_10071_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_10071_end_mask_0 = const()[name = string("op_10071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10071_cast_fp16 = slice_by_index(begin = var_10071_begin_0, end = var_10071_end_0, end_mask = var_10071_end_mask_0, x = k_13_cast_fp16)[name = string("op_10071_cast_fp16")]; + tensor var_10075_begin_0 = const()[name = string("op_10075_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10075_end_0 = const()[name = string("op_10075_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_10075_end_mask_0 = const()[name = string("op_10075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10075_cast_fp16 = slice_by_index(begin = var_10075_begin_0, end = var_10075_end_0, end_mask = var_10075_end_mask_0, x = k_13_cast_fp16)[name = string("op_10075_cast_fp16")]; + tensor var_10079_begin_0 = const()[name = string("op_10079_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10079_end_0 = const()[name = string("op_10079_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_10079_end_mask_0 = const()[name = string("op_10079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10079_cast_fp16 = slice_by_index(begin = var_10079_begin_0, end = var_10079_end_0, end_mask = var_10079_end_mask_0, x = k_13_cast_fp16)[name = string("op_10079_cast_fp16")]; + tensor var_10083_begin_0 = const()[name = string("op_10083_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10083_end_0 = const()[name = string("op_10083_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_10083_end_mask_0 = const()[name = string("op_10083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10083_cast_fp16 = slice_by_index(begin = var_10083_begin_0, end = var_10083_end_0, end_mask = var_10083_end_mask_0, x = k_13_cast_fp16)[name = string("op_10083_cast_fp16")]; + tensor var_10087_begin_0 = const()[name = string("op_10087_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10087_end_0 = const()[name = string("op_10087_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_10087_end_mask_0 = const()[name = string("op_10087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10087_cast_fp16 = slice_by_index(begin = var_10087_begin_0, end = var_10087_end_0, end_mask = var_10087_end_mask_0, x = k_13_cast_fp16)[name = string("op_10087_cast_fp16")]; + tensor var_10091_begin_0 = const()[name = string("op_10091_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10091_end_0 = const()[name = string("op_10091_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_10091_end_mask_0 = const()[name = string("op_10091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10091_cast_fp16 = slice_by_index(begin = var_10091_begin_0, end = var_10091_end_0, end_mask = var_10091_end_mask_0, x = k_13_cast_fp16)[name = string("op_10091_cast_fp16")]; + tensor var_10095_begin_0 = const()[name = string("op_10095_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10095_end_0 = const()[name = string("op_10095_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_10095_end_mask_0 = const()[name = string("op_10095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10095_cast_fp16 = slice_by_index(begin = var_10095_begin_0, end = var_10095_end_0, end_mask = var_10095_end_mask_0, x = k_13_cast_fp16)[name = string("op_10095_cast_fp16")]; + tensor var_10099_begin_0 = const()[name = string("op_10099_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10099_end_0 = const()[name = string("op_10099_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_10099_end_mask_0 = const()[name = string("op_10099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10099_cast_fp16 = slice_by_index(begin = var_10099_begin_0, end = var_10099_end_0, end_mask = var_10099_end_mask_0, x = k_13_cast_fp16)[name = string("op_10099_cast_fp16")]; + tensor var_10103_begin_0 = const()[name = string("op_10103_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10103_end_0 = const()[name = string("op_10103_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_10103_end_mask_0 = const()[name = string("op_10103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10103_cast_fp16 = slice_by_index(begin = var_10103_begin_0, end = var_10103_end_0, end_mask = var_10103_end_mask_0, x = k_13_cast_fp16)[name = string("op_10103_cast_fp16")]; + tensor var_10107_begin_0 = const()[name = string("op_10107_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10107_end_0 = const()[name = string("op_10107_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_10107_end_mask_0 = const()[name = string("op_10107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10107_cast_fp16 = slice_by_index(begin = var_10107_begin_0, end = var_10107_end_0, end_mask = var_10107_end_mask_0, x = k_13_cast_fp16)[name = string("op_10107_cast_fp16")]; + tensor var_10111_begin_0 = const()[name = string("op_10111_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10111_end_0 = const()[name = string("op_10111_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_10111_end_mask_0 = const()[name = string("op_10111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10111_cast_fp16 = slice_by_index(begin = var_10111_begin_0, end = var_10111_end_0, end_mask = var_10111_end_mask_0, x = k_13_cast_fp16)[name = string("op_10111_cast_fp16")]; + tensor var_10115_begin_0 = const()[name = string("op_10115_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10115_end_0 = const()[name = string("op_10115_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_10115_end_mask_0 = const()[name = string("op_10115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10115_cast_fp16 = slice_by_index(begin = var_10115_begin_0, end = var_10115_end_0, end_mask = var_10115_end_mask_0, x = k_13_cast_fp16)[name = string("op_10115_cast_fp16")]; + tensor var_10119_begin_0 = const()[name = string("op_10119_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10119_end_0 = const()[name = string("op_10119_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_10119_end_mask_0 = const()[name = string("op_10119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10119_cast_fp16 = slice_by_index(begin = var_10119_begin_0, end = var_10119_end_0, end_mask = var_10119_end_mask_0, x = k_13_cast_fp16)[name = string("op_10119_cast_fp16")]; + tensor var_10123_begin_0 = const()[name = string("op_10123_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10123_end_0 = const()[name = string("op_10123_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_10123_end_mask_0 = const()[name = string("op_10123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10123_cast_fp16 = slice_by_index(begin = var_10123_begin_0, end = var_10123_end_0, end_mask = var_10123_end_mask_0, x = k_13_cast_fp16)[name = string("op_10123_cast_fp16")]; + tensor var_10127_begin_0 = const()[name = string("op_10127_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10127_end_0 = const()[name = string("op_10127_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_10127_end_mask_0 = const()[name = string("op_10127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10127_cast_fp16 = slice_by_index(begin = var_10127_begin_0, end = var_10127_end_0, end_mask = var_10127_end_mask_0, x = k_13_cast_fp16)[name = string("op_10127_cast_fp16")]; + tensor var_10131_begin_0 = const()[name = string("op_10131_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10131_end_0 = const()[name = string("op_10131_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_10131_end_mask_0 = const()[name = string("op_10131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10131_cast_fp16 = slice_by_index(begin = var_10131_begin_0, end = var_10131_end_0, end_mask = var_10131_end_mask_0, x = k_13_cast_fp16)[name = string("op_10131_cast_fp16")]; + tensor var_10135_begin_0 = const()[name = string("op_10135_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10135_end_0 = const()[name = string("op_10135_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_10135_end_mask_0 = const()[name = string("op_10135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10135_cast_fp16 = slice_by_index(begin = var_10135_begin_0, end = var_10135_end_0, end_mask = var_10135_end_mask_0, x = k_13_cast_fp16)[name = string("op_10135_cast_fp16")]; + tensor var_10139_begin_0 = const()[name = string("op_10139_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10139_end_0 = const()[name = string("op_10139_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_10139_end_mask_0 = const()[name = string("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = k_13_cast_fp16)[name = string("op_10139_cast_fp16")]; + tensor var_10143_begin_0 = const()[name = string("op_10143_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10143_end_0 = const()[name = string("op_10143_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_10143_end_mask_0 = const()[name = string("op_10143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10143_cast_fp16 = slice_by_index(begin = var_10143_begin_0, end = var_10143_end_0, end_mask = var_10143_end_mask_0, x = k_13_cast_fp16)[name = string("op_10143_cast_fp16")]; + tensor var_10145_begin_0 = const()[name = string("op_10145_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10145_end_0 = const()[name = string("op_10145_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10145_end_mask_0 = const()[name = string("op_10145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10145_cast_fp16 = slice_by_index(begin = var_10145_begin_0, end = var_10145_end_0, end_mask = var_10145_end_mask_0, x = value_13_cast_fp16)[name = string("op_10145_cast_fp16")]; + tensor var_10149_begin_0 = const()[name = string("op_10149_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10149_end_0 = const()[name = string("op_10149_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10149_end_mask_0 = const()[name = string("op_10149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10149_cast_fp16 = slice_by_index(begin = var_10149_begin_0, end = var_10149_end_0, end_mask = var_10149_end_mask_0, x = value_13_cast_fp16)[name = string("op_10149_cast_fp16")]; + tensor var_10153_begin_0 = const()[name = string("op_10153_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10153_end_0 = const()[name = string("op_10153_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10153_end_mask_0 = const()[name = string("op_10153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10153_cast_fp16 = slice_by_index(begin = var_10153_begin_0, end = var_10153_end_0, end_mask = var_10153_end_mask_0, x = value_13_cast_fp16)[name = string("op_10153_cast_fp16")]; + tensor var_10157_begin_0 = const()[name = string("op_10157_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10157_end_0 = const()[name = string("op_10157_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10157_end_mask_0 = const()[name = string("op_10157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10157_cast_fp16 = slice_by_index(begin = var_10157_begin_0, end = var_10157_end_0, end_mask = var_10157_end_mask_0, x = value_13_cast_fp16)[name = string("op_10157_cast_fp16")]; + tensor var_10161_begin_0 = const()[name = string("op_10161_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10161_end_0 = const()[name = string("op_10161_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10161_end_mask_0 = const()[name = string("op_10161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10161_cast_fp16 = slice_by_index(begin = var_10161_begin_0, end = var_10161_end_0, end_mask = var_10161_end_mask_0, x = value_13_cast_fp16)[name = string("op_10161_cast_fp16")]; + tensor var_10165_begin_0 = const()[name = string("op_10165_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10165_end_0 = const()[name = string("op_10165_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10165_end_mask_0 = const()[name = string("op_10165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10165_cast_fp16 = slice_by_index(begin = var_10165_begin_0, end = var_10165_end_0, end_mask = var_10165_end_mask_0, x = value_13_cast_fp16)[name = string("op_10165_cast_fp16")]; + tensor var_10169_begin_0 = const()[name = string("op_10169_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10169_end_0 = const()[name = string("op_10169_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10169_end_mask_0 = const()[name = string("op_10169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10169_cast_fp16 = slice_by_index(begin = var_10169_begin_0, end = var_10169_end_0, end_mask = var_10169_end_mask_0, x = value_13_cast_fp16)[name = string("op_10169_cast_fp16")]; + tensor var_10173_begin_0 = const()[name = string("op_10173_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10173_end_0 = const()[name = string("op_10173_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10173_end_mask_0 = const()[name = string("op_10173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10173_cast_fp16 = slice_by_index(begin = var_10173_begin_0, end = var_10173_end_0, end_mask = var_10173_end_mask_0, x = value_13_cast_fp16)[name = string("op_10173_cast_fp16")]; + tensor var_10177_begin_0 = const()[name = string("op_10177_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10177_end_0 = const()[name = string("op_10177_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10177_end_mask_0 = const()[name = string("op_10177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10177_cast_fp16 = slice_by_index(begin = var_10177_begin_0, end = var_10177_end_0, end_mask = var_10177_end_mask_0, x = value_13_cast_fp16)[name = string("op_10177_cast_fp16")]; + tensor var_10181_begin_0 = const()[name = string("op_10181_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10181_end_0 = const()[name = string("op_10181_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10181_end_mask_0 = const()[name = string("op_10181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10181_cast_fp16 = slice_by_index(begin = var_10181_begin_0, end = var_10181_end_0, end_mask = var_10181_end_mask_0, x = value_13_cast_fp16)[name = string("op_10181_cast_fp16")]; + tensor var_10185_begin_0 = const()[name = string("op_10185_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10185_end_0 = const()[name = string("op_10185_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10185_end_mask_0 = const()[name = string("op_10185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10185_cast_fp16 = slice_by_index(begin = var_10185_begin_0, end = var_10185_end_0, end_mask = var_10185_end_mask_0, x = value_13_cast_fp16)[name = string("op_10185_cast_fp16")]; + tensor var_10189_begin_0 = const()[name = string("op_10189_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10189_end_0 = const()[name = string("op_10189_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10189_end_mask_0 = const()[name = string("op_10189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10189_cast_fp16 = slice_by_index(begin = var_10189_begin_0, end = var_10189_end_0, end_mask = var_10189_end_mask_0, x = value_13_cast_fp16)[name = string("op_10189_cast_fp16")]; + tensor var_10193_begin_0 = const()[name = string("op_10193_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10193_end_0 = const()[name = string("op_10193_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_10193_end_mask_0 = const()[name = string("op_10193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10193_cast_fp16 = slice_by_index(begin = var_10193_begin_0, end = var_10193_end_0, end_mask = var_10193_end_mask_0, x = value_13_cast_fp16)[name = string("op_10193_cast_fp16")]; + tensor var_10197_begin_0 = const()[name = string("op_10197_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10197_end_0 = const()[name = string("op_10197_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_10197_end_mask_0 = const()[name = string("op_10197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10197_cast_fp16 = slice_by_index(begin = var_10197_begin_0, end = var_10197_end_0, end_mask = var_10197_end_mask_0, x = value_13_cast_fp16)[name = string("op_10197_cast_fp16")]; + tensor var_10201_begin_0 = const()[name = string("op_10201_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10201_end_0 = const()[name = string("op_10201_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_10201_end_mask_0 = const()[name = string("op_10201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10201_cast_fp16 = slice_by_index(begin = var_10201_begin_0, end = var_10201_end_0, end_mask = var_10201_end_mask_0, x = value_13_cast_fp16)[name = string("op_10201_cast_fp16")]; + tensor var_10205_begin_0 = const()[name = string("op_10205_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10205_end_0 = const()[name = string("op_10205_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_10205_end_mask_0 = const()[name = string("op_10205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10205_cast_fp16 = slice_by_index(begin = var_10205_begin_0, end = var_10205_end_0, end_mask = var_10205_end_mask_0, x = value_13_cast_fp16)[name = string("op_10205_cast_fp16")]; + tensor var_10209_begin_0 = const()[name = string("op_10209_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10209_end_0 = const()[name = string("op_10209_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_10209_end_mask_0 = const()[name = string("op_10209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10209_cast_fp16 = slice_by_index(begin = var_10209_begin_0, end = var_10209_end_0, end_mask = var_10209_end_mask_0, x = value_13_cast_fp16)[name = string("op_10209_cast_fp16")]; + tensor var_10213_begin_0 = const()[name = string("op_10213_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10213_end_0 = const()[name = string("op_10213_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_10213_end_mask_0 = const()[name = string("op_10213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10213_cast_fp16 = slice_by_index(begin = var_10213_begin_0, end = var_10213_end_0, end_mask = var_10213_end_mask_0, x = value_13_cast_fp16)[name = string("op_10213_cast_fp16")]; + tensor var_10217_begin_0 = const()[name = string("op_10217_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10217_end_0 = const()[name = string("op_10217_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_10217_end_mask_0 = const()[name = string("op_10217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10217_cast_fp16 = slice_by_index(begin = var_10217_begin_0, end = var_10217_end_0, end_mask = var_10217_end_mask_0, x = value_13_cast_fp16)[name = string("op_10217_cast_fp16")]; + tensor var_10221_begin_0 = const()[name = string("op_10221_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10221_end_0 = const()[name = string("op_10221_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_10221_end_mask_0 = const()[name = string("op_10221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10221_cast_fp16 = slice_by_index(begin = var_10221_begin_0, end = var_10221_end_0, end_mask = var_10221_end_mask_0, x = value_13_cast_fp16)[name = string("op_10221_cast_fp16")]; + string _SplitHeadsQ__mh_w_961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_961_equation_0, values = (var_10067_cast_fp16, var_9509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_961_cast_fp16")]; + string _SplitHeadsQ__mh_w_963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_963_equation_0, values = (var_10067_cast_fp16, var_9516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_963_cast_fp16")]; + string _SplitHeadsQ__mh_w_965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_965_equation_0, values = (var_10067_cast_fp16, var_9523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_965_cast_fp16")]; + string _SplitHeadsQ__mh_w_967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_967_equation_0, values = (var_10067_cast_fp16, var_9530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_967_cast_fp16")]; + string _SplitHeadsQ__mh_w_969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_969_equation_0, values = (var_10071_cast_fp16, var_9537_cast_fp16))[name = string("_SplitHeadsQ__mh_w_969_cast_fp16")]; + string _SplitHeadsQ__mh_w_971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_971_equation_0, values = (var_10071_cast_fp16, var_9544_cast_fp16))[name = string("_SplitHeadsQ__mh_w_971_cast_fp16")]; + string _SplitHeadsQ__mh_w_973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_973_equation_0, values = (var_10071_cast_fp16, var_9551_cast_fp16))[name = string("_SplitHeadsQ__mh_w_973_cast_fp16")]; + string _SplitHeadsQ__mh_w_975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_975_equation_0, values = (var_10071_cast_fp16, var_9558_cast_fp16))[name = string("_SplitHeadsQ__mh_w_975_cast_fp16")]; + string _SplitHeadsQ__mh_w_977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_977_equation_0, values = (var_10075_cast_fp16, var_9565_cast_fp16))[name = string("_SplitHeadsQ__mh_w_977_cast_fp16")]; + string _SplitHeadsQ__mh_w_979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_979_equation_0, values = (var_10075_cast_fp16, var_9572_cast_fp16))[name = string("_SplitHeadsQ__mh_w_979_cast_fp16")]; + string _SplitHeadsQ__mh_w_981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_981_equation_0, values = (var_10075_cast_fp16, var_9579_cast_fp16))[name = string("_SplitHeadsQ__mh_w_981_cast_fp16")]; + string _SplitHeadsQ__mh_w_983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_983_equation_0, values = (var_10075_cast_fp16, var_9586_cast_fp16))[name = string("_SplitHeadsQ__mh_w_983_cast_fp16")]; + string _SplitHeadsQ__mh_w_985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_985_equation_0, values = (var_10079_cast_fp16, var_9593_cast_fp16))[name = string("_SplitHeadsQ__mh_w_985_cast_fp16")]; + string _SplitHeadsQ__mh_w_987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_987_equation_0, values = (var_10079_cast_fp16, var_9600_cast_fp16))[name = string("_SplitHeadsQ__mh_w_987_cast_fp16")]; + string _SplitHeadsQ__mh_w_989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_989_equation_0, values = (var_10079_cast_fp16, var_9607_cast_fp16))[name = string("_SplitHeadsQ__mh_w_989_cast_fp16")]; + string _SplitHeadsQ__mh_w_991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_991_equation_0, values = (var_10079_cast_fp16, var_9614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_991_cast_fp16")]; + string _SplitHeadsQ__mh_w_993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_993_equation_0, values = (var_10083_cast_fp16, var_9621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_993_cast_fp16")]; + string _SplitHeadsQ__mh_w_995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_995_equation_0, values = (var_10083_cast_fp16, var_9628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_995_cast_fp16")]; + string _SplitHeadsQ__mh_w_997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_997_equation_0, values = (var_10083_cast_fp16, var_9635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_997_cast_fp16")]; + string _SplitHeadsQ__mh_w_999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_999_equation_0, values = (var_10083_cast_fp16, var_9642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_999_cast_fp16")]; + string _SplitHeadsQ__mh_w_1001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1001_equation_0, values = (var_10087_cast_fp16, var_9649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1001_cast_fp16")]; + string _SplitHeadsQ__mh_w_1003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1003_equation_0, values = (var_10087_cast_fp16, var_9656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1003_cast_fp16")]; + string _SplitHeadsQ__mh_w_1005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1005_equation_0, values = (var_10087_cast_fp16, var_9663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1005_cast_fp16")]; + string _SplitHeadsQ__mh_w_1007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1007_equation_0, values = (var_10087_cast_fp16, var_9670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1007_cast_fp16")]; + string _SplitHeadsQ__mh_w_1009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1009_equation_0, values = (var_10091_cast_fp16, var_9677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1009_cast_fp16")]; + string _SplitHeadsQ__mh_w_1011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1011_equation_0, values = (var_10091_cast_fp16, var_9684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1011_cast_fp16")]; + string _SplitHeadsQ__mh_w_1013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1013_equation_0, values = (var_10091_cast_fp16, var_9691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1013_cast_fp16")]; + string _SplitHeadsQ__mh_w_1015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1015_equation_0, values = (var_10091_cast_fp16, var_9698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1015_cast_fp16")]; + string _SplitHeadsQ__mh_w_1017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1017_equation_0, values = (var_10095_cast_fp16, var_9705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1017_cast_fp16")]; + string _SplitHeadsQ__mh_w_1019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1019_equation_0, values = (var_10095_cast_fp16, var_9712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1019_cast_fp16")]; + string _SplitHeadsQ__mh_w_1021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1021_equation_0, values = (var_10095_cast_fp16, var_9719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1021_cast_fp16")]; + string _SplitHeadsQ__mh_w_1023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1023_equation_0, values = (var_10095_cast_fp16, var_9726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1023_cast_fp16")]; + string _SplitHeadsQ__mh_w_1025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1025_equation_0, values = (var_10099_cast_fp16, var_9733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1025_cast_fp16")]; + string _SplitHeadsQ__mh_w_1027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1027_equation_0, values = (var_10099_cast_fp16, var_9740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1027_cast_fp16")]; + string _SplitHeadsQ__mh_w_1029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1029_equation_0, values = (var_10099_cast_fp16, var_9747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1029_cast_fp16")]; + string _SplitHeadsQ__mh_w_1031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1031_equation_0, values = (var_10099_cast_fp16, var_9754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1031_cast_fp16")]; + string _SplitHeadsQ__mh_w_1033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1033_equation_0, values = (var_10103_cast_fp16, var_9761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1033_cast_fp16")]; + string _SplitHeadsQ__mh_w_1035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1035_equation_0, values = (var_10103_cast_fp16, var_9768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1035_cast_fp16")]; + string _SplitHeadsQ__mh_w_1037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1037_equation_0, values = (var_10103_cast_fp16, var_9775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1037_cast_fp16")]; + string _SplitHeadsQ__mh_w_1039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1039_equation_0, values = (var_10103_cast_fp16, var_9782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1039_cast_fp16")]; + string _SplitHeadsQ__mh_w_1041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1041_equation_0, values = (var_10107_cast_fp16, var_9789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1041_cast_fp16")]; + string _SplitHeadsQ__mh_w_1043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1043_equation_0, values = (var_10107_cast_fp16, var_9796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1043_cast_fp16")]; + string _SplitHeadsQ__mh_w_1045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1045_equation_0, values = (var_10107_cast_fp16, var_9803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1045_cast_fp16")]; + string _SplitHeadsQ__mh_w_1047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1047_equation_0, values = (var_10107_cast_fp16, var_9810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1047_cast_fp16")]; + string _SplitHeadsQ__mh_w_1049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1049_equation_0, values = (var_10111_cast_fp16, var_9817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1049_cast_fp16")]; + string _SplitHeadsQ__mh_w_1051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1051_equation_0, values = (var_10111_cast_fp16, var_9824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1051_cast_fp16")]; + string _SplitHeadsQ__mh_w_1053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1053_equation_0, values = (var_10111_cast_fp16, var_9831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1053_cast_fp16")]; + string _SplitHeadsQ__mh_w_1055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1055_equation_0, values = (var_10111_cast_fp16, var_9838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1055_cast_fp16")]; + string _SplitHeadsQ__mh_w_1057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1057_equation_0, values = (var_10115_cast_fp16, var_9845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1057_cast_fp16")]; + string _SplitHeadsQ__mh_w_1059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1059_equation_0, values = (var_10115_cast_fp16, var_9852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1059_cast_fp16")]; + string _SplitHeadsQ__mh_w_1061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1061_equation_0, values = (var_10115_cast_fp16, var_9859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1061_cast_fp16")]; + string _SplitHeadsQ__mh_w_1063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1063_equation_0, values = (var_10115_cast_fp16, var_9866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1063_cast_fp16")]; + string _SplitHeadsQ__mh_w_1065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1065_equation_0, values = (var_10119_cast_fp16, var_9873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1065_cast_fp16")]; + string _SplitHeadsQ__mh_w_1067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1067_equation_0, values = (var_10119_cast_fp16, var_9880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1067_cast_fp16")]; + string _SplitHeadsQ__mh_w_1069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1069_equation_0, values = (var_10119_cast_fp16, var_9887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1069_cast_fp16")]; + string _SplitHeadsQ__mh_w_1071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1071_equation_0, values = (var_10119_cast_fp16, var_9894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1071_cast_fp16")]; + string _SplitHeadsQ__mh_w_1073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1073_equation_0, values = (var_10123_cast_fp16, var_9901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1073_cast_fp16")]; + string _SplitHeadsQ__mh_w_1075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1075_equation_0, values = (var_10123_cast_fp16, var_9908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1075_cast_fp16")]; + string _SplitHeadsQ__mh_w_1077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1077_equation_0, values = (var_10123_cast_fp16, var_9915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1077_cast_fp16")]; + string _SplitHeadsQ__mh_w_1079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1079_equation_0, values = (var_10123_cast_fp16, var_9922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1079_cast_fp16")]; + string _SplitHeadsQ__mh_w_1081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1081_equation_0, values = (var_10127_cast_fp16, var_9929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1081_cast_fp16")]; + string _SplitHeadsQ__mh_w_1083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1083_equation_0, values = (var_10127_cast_fp16, var_9936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1083_cast_fp16")]; + string _SplitHeadsQ__mh_w_1085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1085_equation_0, values = (var_10127_cast_fp16, var_9943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1085_cast_fp16")]; + string _SplitHeadsQ__mh_w_1087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1087_equation_0, values = (var_10127_cast_fp16, var_9950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1087_cast_fp16")]; + string _SplitHeadsQ__mh_w_1089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1089_equation_0, values = (var_10131_cast_fp16, var_9957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1089_cast_fp16")]; + string _SplitHeadsQ__mh_w_1091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1091_equation_0, values = (var_10131_cast_fp16, var_9964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1091_cast_fp16")]; + string _SplitHeadsQ__mh_w_1093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1093_equation_0, values = (var_10131_cast_fp16, var_9971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1093_cast_fp16")]; + string _SplitHeadsQ__mh_w_1095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1095_equation_0, values = (var_10131_cast_fp16, var_9978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1095_cast_fp16")]; + string _SplitHeadsQ__mh_w_1097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1097_equation_0, values = (var_10135_cast_fp16, var_9985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1097_cast_fp16")]; + string _SplitHeadsQ__mh_w_1099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1099_equation_0, values = (var_10135_cast_fp16, var_9992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1099_cast_fp16")]; + string _SplitHeadsQ__mh_w_1101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1101_equation_0, values = (var_10135_cast_fp16, var_9999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1101_cast_fp16")]; + string _SplitHeadsQ__mh_w_1103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1103_equation_0, values = (var_10135_cast_fp16, var_10006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1103_cast_fp16")]; + string _SplitHeadsQ__mh_w_1105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1105_equation_0, values = (var_10139_cast_fp16, var_10013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1105_cast_fp16")]; + string _SplitHeadsQ__mh_w_1107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1107_equation_0, values = (var_10139_cast_fp16, var_10020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1107_cast_fp16")]; + string _SplitHeadsQ__mh_w_1109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1109_equation_0, values = (var_10139_cast_fp16, var_10027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1109_cast_fp16")]; + string _SplitHeadsQ__mh_w_1111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1111_equation_0, values = (var_10139_cast_fp16, var_10034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1111_cast_fp16")]; + string _SplitHeadsQ__mh_w_1113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1113_equation_0, values = (var_10143_cast_fp16, var_10041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1113_cast_fp16")]; + string _SplitHeadsQ__mh_w_1115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1115_equation_0, values = (var_10143_cast_fp16, var_10048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1115_cast_fp16")]; + string _SplitHeadsQ__mh_w_1117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1117_equation_0, values = (var_10143_cast_fp16, var_10055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1117_cast_fp16")]; + string _SplitHeadsQ__mh_w_1119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1119_equation_0, values = (var_10143_cast_fp16, var_10062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1119_cast_fp16")]; + fp16 var_10384_to_fp16 = const()[name = string("op_10384_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_961_cast_fp16, y = var_10384_to_fp16)[name = string("aw_chunk_961_cast_fp16")]; + fp16 var_10386_to_fp16 = const()[name = string("op_10386_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_963_cast_fp16, y = var_10386_to_fp16)[name = string("aw_chunk_963_cast_fp16")]; + fp16 var_10388_to_fp16 = const()[name = string("op_10388_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_965_cast_fp16, y = var_10388_to_fp16)[name = string("aw_chunk_965_cast_fp16")]; + fp16 var_10390_to_fp16 = const()[name = string("op_10390_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_967_cast_fp16, y = var_10390_to_fp16)[name = string("aw_chunk_967_cast_fp16")]; + fp16 var_10392_to_fp16 = const()[name = string("op_10392_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_969_cast_fp16, y = var_10392_to_fp16)[name = string("aw_chunk_969_cast_fp16")]; + fp16 var_10394_to_fp16 = const()[name = string("op_10394_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_971_cast_fp16, y = var_10394_to_fp16)[name = string("aw_chunk_971_cast_fp16")]; + fp16 var_10396_to_fp16 = const()[name = string("op_10396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_973_cast_fp16, y = var_10396_to_fp16)[name = string("aw_chunk_973_cast_fp16")]; + fp16 var_10398_to_fp16 = const()[name = string("op_10398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_975_cast_fp16, y = var_10398_to_fp16)[name = string("aw_chunk_975_cast_fp16")]; + fp16 var_10400_to_fp16 = const()[name = string("op_10400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_977_cast_fp16, y = var_10400_to_fp16)[name = string("aw_chunk_977_cast_fp16")]; + fp16 var_10402_to_fp16 = const()[name = string("op_10402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_979_cast_fp16, y = var_10402_to_fp16)[name = string("aw_chunk_979_cast_fp16")]; + fp16 var_10404_to_fp16 = const()[name = string("op_10404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_981_cast_fp16, y = var_10404_to_fp16)[name = string("aw_chunk_981_cast_fp16")]; + fp16 var_10406_to_fp16 = const()[name = string("op_10406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_983_cast_fp16, y = var_10406_to_fp16)[name = string("aw_chunk_983_cast_fp16")]; + fp16 var_10408_to_fp16 = const()[name = string("op_10408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_985_cast_fp16, y = var_10408_to_fp16)[name = string("aw_chunk_985_cast_fp16")]; + fp16 var_10410_to_fp16 = const()[name = string("op_10410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_987_cast_fp16, y = var_10410_to_fp16)[name = string("aw_chunk_987_cast_fp16")]; + fp16 var_10412_to_fp16 = const()[name = string("op_10412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_989_cast_fp16, y = var_10412_to_fp16)[name = string("aw_chunk_989_cast_fp16")]; + fp16 var_10414_to_fp16 = const()[name = string("op_10414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_991_cast_fp16, y = var_10414_to_fp16)[name = string("aw_chunk_991_cast_fp16")]; + fp16 var_10416_to_fp16 = const()[name = string("op_10416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_993_cast_fp16, y = var_10416_to_fp16)[name = string("aw_chunk_993_cast_fp16")]; + fp16 var_10418_to_fp16 = const()[name = string("op_10418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_995_cast_fp16, y = var_10418_to_fp16)[name = string("aw_chunk_995_cast_fp16")]; + fp16 var_10420_to_fp16 = const()[name = string("op_10420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_997_cast_fp16, y = var_10420_to_fp16)[name = string("aw_chunk_997_cast_fp16")]; + fp16 var_10422_to_fp16 = const()[name = string("op_10422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_999_cast_fp16, y = var_10422_to_fp16)[name = string("aw_chunk_999_cast_fp16")]; + fp16 var_10424_to_fp16 = const()[name = string("op_10424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1001_cast_fp16, y = var_10424_to_fp16)[name = string("aw_chunk_1001_cast_fp16")]; + fp16 var_10426_to_fp16 = const()[name = string("op_10426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1003_cast_fp16, y = var_10426_to_fp16)[name = string("aw_chunk_1003_cast_fp16")]; + fp16 var_10428_to_fp16 = const()[name = string("op_10428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1005_cast_fp16, y = var_10428_to_fp16)[name = string("aw_chunk_1005_cast_fp16")]; + fp16 var_10430_to_fp16 = const()[name = string("op_10430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1007_cast_fp16, y = var_10430_to_fp16)[name = string("aw_chunk_1007_cast_fp16")]; + fp16 var_10432_to_fp16 = const()[name = string("op_10432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1009_cast_fp16, y = var_10432_to_fp16)[name = string("aw_chunk_1009_cast_fp16")]; + fp16 var_10434_to_fp16 = const()[name = string("op_10434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1011_cast_fp16, y = var_10434_to_fp16)[name = string("aw_chunk_1011_cast_fp16")]; + fp16 var_10436_to_fp16 = const()[name = string("op_10436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1013_cast_fp16, y = var_10436_to_fp16)[name = string("aw_chunk_1013_cast_fp16")]; + fp16 var_10438_to_fp16 = const()[name = string("op_10438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1015_cast_fp16, y = var_10438_to_fp16)[name = string("aw_chunk_1015_cast_fp16")]; + fp16 var_10440_to_fp16 = const()[name = string("op_10440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1017_cast_fp16, y = var_10440_to_fp16)[name = string("aw_chunk_1017_cast_fp16")]; + fp16 var_10442_to_fp16 = const()[name = string("op_10442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1019_cast_fp16, y = var_10442_to_fp16)[name = string("aw_chunk_1019_cast_fp16")]; + fp16 var_10444_to_fp16 = const()[name = string("op_10444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1021_cast_fp16, y = var_10444_to_fp16)[name = string("aw_chunk_1021_cast_fp16")]; + fp16 var_10446_to_fp16 = const()[name = string("op_10446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1023_cast_fp16, y = var_10446_to_fp16)[name = string("aw_chunk_1023_cast_fp16")]; + fp16 var_10448_to_fp16 = const()[name = string("op_10448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1025_cast_fp16, y = var_10448_to_fp16)[name = string("aw_chunk_1025_cast_fp16")]; + fp16 var_10450_to_fp16 = const()[name = string("op_10450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1027_cast_fp16, y = var_10450_to_fp16)[name = string("aw_chunk_1027_cast_fp16")]; + fp16 var_10452_to_fp16 = const()[name = string("op_10452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1029_cast_fp16, y = var_10452_to_fp16)[name = string("aw_chunk_1029_cast_fp16")]; + fp16 var_10454_to_fp16 = const()[name = string("op_10454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1031_cast_fp16, y = var_10454_to_fp16)[name = string("aw_chunk_1031_cast_fp16")]; + fp16 var_10456_to_fp16 = const()[name = string("op_10456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1033_cast_fp16, y = var_10456_to_fp16)[name = string("aw_chunk_1033_cast_fp16")]; + fp16 var_10458_to_fp16 = const()[name = string("op_10458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1035_cast_fp16, y = var_10458_to_fp16)[name = string("aw_chunk_1035_cast_fp16")]; + fp16 var_10460_to_fp16 = const()[name = string("op_10460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1037_cast_fp16, y = var_10460_to_fp16)[name = string("aw_chunk_1037_cast_fp16")]; + fp16 var_10462_to_fp16 = const()[name = string("op_10462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1039_cast_fp16, y = var_10462_to_fp16)[name = string("aw_chunk_1039_cast_fp16")]; + fp16 var_10464_to_fp16 = const()[name = string("op_10464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1041_cast_fp16, y = var_10464_to_fp16)[name = string("aw_chunk_1041_cast_fp16")]; + fp16 var_10466_to_fp16 = const()[name = string("op_10466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1043_cast_fp16, y = var_10466_to_fp16)[name = string("aw_chunk_1043_cast_fp16")]; + fp16 var_10468_to_fp16 = const()[name = string("op_10468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1045_cast_fp16, y = var_10468_to_fp16)[name = string("aw_chunk_1045_cast_fp16")]; + fp16 var_10470_to_fp16 = const()[name = string("op_10470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1047_cast_fp16, y = var_10470_to_fp16)[name = string("aw_chunk_1047_cast_fp16")]; + fp16 var_10472_to_fp16 = const()[name = string("op_10472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1049_cast_fp16, y = var_10472_to_fp16)[name = string("aw_chunk_1049_cast_fp16")]; + fp16 var_10474_to_fp16 = const()[name = string("op_10474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1051_cast_fp16, y = var_10474_to_fp16)[name = string("aw_chunk_1051_cast_fp16")]; + fp16 var_10476_to_fp16 = const()[name = string("op_10476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1053_cast_fp16, y = var_10476_to_fp16)[name = string("aw_chunk_1053_cast_fp16")]; + fp16 var_10478_to_fp16 = const()[name = string("op_10478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1055_cast_fp16, y = var_10478_to_fp16)[name = string("aw_chunk_1055_cast_fp16")]; + fp16 var_10480_to_fp16 = const()[name = string("op_10480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1057_cast_fp16, y = var_10480_to_fp16)[name = string("aw_chunk_1057_cast_fp16")]; + fp16 var_10482_to_fp16 = const()[name = string("op_10482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1059_cast_fp16, y = var_10482_to_fp16)[name = string("aw_chunk_1059_cast_fp16")]; + fp16 var_10484_to_fp16 = const()[name = string("op_10484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1061_cast_fp16, y = var_10484_to_fp16)[name = string("aw_chunk_1061_cast_fp16")]; + fp16 var_10486_to_fp16 = const()[name = string("op_10486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1063_cast_fp16, y = var_10486_to_fp16)[name = string("aw_chunk_1063_cast_fp16")]; + fp16 var_10488_to_fp16 = const()[name = string("op_10488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1065_cast_fp16, y = var_10488_to_fp16)[name = string("aw_chunk_1065_cast_fp16")]; + fp16 var_10490_to_fp16 = const()[name = string("op_10490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1067_cast_fp16, y = var_10490_to_fp16)[name = string("aw_chunk_1067_cast_fp16")]; + fp16 var_10492_to_fp16 = const()[name = string("op_10492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1069_cast_fp16, y = var_10492_to_fp16)[name = string("aw_chunk_1069_cast_fp16")]; + fp16 var_10494_to_fp16 = const()[name = string("op_10494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1071_cast_fp16, y = var_10494_to_fp16)[name = string("aw_chunk_1071_cast_fp16")]; + fp16 var_10496_to_fp16 = const()[name = string("op_10496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1073_cast_fp16, y = var_10496_to_fp16)[name = string("aw_chunk_1073_cast_fp16")]; + fp16 var_10498_to_fp16 = const()[name = string("op_10498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1075_cast_fp16, y = var_10498_to_fp16)[name = string("aw_chunk_1075_cast_fp16")]; + fp16 var_10500_to_fp16 = const()[name = string("op_10500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1077_cast_fp16, y = var_10500_to_fp16)[name = string("aw_chunk_1077_cast_fp16")]; + fp16 var_10502_to_fp16 = const()[name = string("op_10502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1079_cast_fp16, y = var_10502_to_fp16)[name = string("aw_chunk_1079_cast_fp16")]; + fp16 var_10504_to_fp16 = const()[name = string("op_10504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1081_cast_fp16, y = var_10504_to_fp16)[name = string("aw_chunk_1081_cast_fp16")]; + fp16 var_10506_to_fp16 = const()[name = string("op_10506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1083_cast_fp16, y = var_10506_to_fp16)[name = string("aw_chunk_1083_cast_fp16")]; + fp16 var_10508_to_fp16 = const()[name = string("op_10508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1085_cast_fp16, y = var_10508_to_fp16)[name = string("aw_chunk_1085_cast_fp16")]; + fp16 var_10510_to_fp16 = const()[name = string("op_10510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1087_cast_fp16, y = var_10510_to_fp16)[name = string("aw_chunk_1087_cast_fp16")]; + fp16 var_10512_to_fp16 = const()[name = string("op_10512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1089_cast_fp16, y = var_10512_to_fp16)[name = string("aw_chunk_1089_cast_fp16")]; + fp16 var_10514_to_fp16 = const()[name = string("op_10514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1091_cast_fp16, y = var_10514_to_fp16)[name = string("aw_chunk_1091_cast_fp16")]; + fp16 var_10516_to_fp16 = const()[name = string("op_10516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1093_cast_fp16, y = var_10516_to_fp16)[name = string("aw_chunk_1093_cast_fp16")]; + fp16 var_10518_to_fp16 = const()[name = string("op_10518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1095_cast_fp16, y = var_10518_to_fp16)[name = string("aw_chunk_1095_cast_fp16")]; + fp16 var_10520_to_fp16 = const()[name = string("op_10520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1097_cast_fp16, y = var_10520_to_fp16)[name = string("aw_chunk_1097_cast_fp16")]; + fp16 var_10522_to_fp16 = const()[name = string("op_10522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1099_cast_fp16, y = var_10522_to_fp16)[name = string("aw_chunk_1099_cast_fp16")]; + fp16 var_10524_to_fp16 = const()[name = string("op_10524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1101_cast_fp16, y = var_10524_to_fp16)[name = string("aw_chunk_1101_cast_fp16")]; + fp16 var_10526_to_fp16 = const()[name = string("op_10526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1103_cast_fp16, y = var_10526_to_fp16)[name = string("aw_chunk_1103_cast_fp16")]; + fp16 var_10528_to_fp16 = const()[name = string("op_10528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1105_cast_fp16, y = var_10528_to_fp16)[name = string("aw_chunk_1105_cast_fp16")]; + fp16 var_10530_to_fp16 = const()[name = string("op_10530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1107_cast_fp16, y = var_10530_to_fp16)[name = string("aw_chunk_1107_cast_fp16")]; + fp16 var_10532_to_fp16 = const()[name = string("op_10532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1109_cast_fp16, y = var_10532_to_fp16)[name = string("aw_chunk_1109_cast_fp16")]; + fp16 var_10534_to_fp16 = const()[name = string("op_10534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1111_cast_fp16, y = var_10534_to_fp16)[name = string("aw_chunk_1111_cast_fp16")]; + fp16 var_10536_to_fp16 = const()[name = string("op_10536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1113_cast_fp16, y = var_10536_to_fp16)[name = string("aw_chunk_1113_cast_fp16")]; + fp16 var_10538_to_fp16 = const()[name = string("op_10538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1115_cast_fp16, y = var_10538_to_fp16)[name = string("aw_chunk_1115_cast_fp16")]; + fp16 var_10540_to_fp16 = const()[name = string("op_10540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1117_cast_fp16, y = var_10540_to_fp16)[name = string("aw_chunk_1117_cast_fp16")]; + fp16 var_10542_to_fp16 = const()[name = string("op_10542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1119_cast_fp16, y = var_10542_to_fp16)[name = string("aw_chunk_1119_cast_fp16")]; + tensor var_10544_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_961_cast_fp16)[name = string("op_10544_cast_fp16")]; + tensor var_10545_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_963_cast_fp16)[name = string("op_10545_cast_fp16")]; + tensor var_10546_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_965_cast_fp16)[name = string("op_10546_cast_fp16")]; + tensor var_10547_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_967_cast_fp16)[name = string("op_10547_cast_fp16")]; + tensor var_10548_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_969_cast_fp16)[name = string("op_10548_cast_fp16")]; + tensor var_10549_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_971_cast_fp16)[name = string("op_10549_cast_fp16")]; + tensor var_10550_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_973_cast_fp16)[name = string("op_10550_cast_fp16")]; + tensor var_10551_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_975_cast_fp16)[name = string("op_10551_cast_fp16")]; + tensor var_10552_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_977_cast_fp16)[name = string("op_10552_cast_fp16")]; + tensor var_10553_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_979_cast_fp16)[name = string("op_10553_cast_fp16")]; + tensor var_10554_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_981_cast_fp16)[name = string("op_10554_cast_fp16")]; + tensor var_10555_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_983_cast_fp16)[name = string("op_10555_cast_fp16")]; + tensor var_10556_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_985_cast_fp16)[name = string("op_10556_cast_fp16")]; + tensor var_10557_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_987_cast_fp16)[name = string("op_10557_cast_fp16")]; + tensor var_10558_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_989_cast_fp16)[name = string("op_10558_cast_fp16")]; + tensor var_10559_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_991_cast_fp16)[name = string("op_10559_cast_fp16")]; + tensor var_10560_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_993_cast_fp16)[name = string("op_10560_cast_fp16")]; + tensor var_10561_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_995_cast_fp16)[name = string("op_10561_cast_fp16")]; + tensor var_10562_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_997_cast_fp16)[name = string("op_10562_cast_fp16")]; + tensor var_10563_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_999_cast_fp16)[name = string("op_10563_cast_fp16")]; + tensor var_10564_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1001_cast_fp16)[name = string("op_10564_cast_fp16")]; + tensor var_10565_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1003_cast_fp16)[name = string("op_10565_cast_fp16")]; + tensor var_10566_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1005_cast_fp16)[name = string("op_10566_cast_fp16")]; + tensor var_10567_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1007_cast_fp16)[name = string("op_10567_cast_fp16")]; + tensor var_10568_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1009_cast_fp16)[name = string("op_10568_cast_fp16")]; + tensor var_10569_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1011_cast_fp16)[name = string("op_10569_cast_fp16")]; + tensor var_10570_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1013_cast_fp16)[name = string("op_10570_cast_fp16")]; + tensor var_10571_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1015_cast_fp16)[name = string("op_10571_cast_fp16")]; + tensor var_10572_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1017_cast_fp16)[name = string("op_10572_cast_fp16")]; + tensor var_10573_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1019_cast_fp16)[name = string("op_10573_cast_fp16")]; + tensor var_10574_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1021_cast_fp16)[name = string("op_10574_cast_fp16")]; + tensor var_10575_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1023_cast_fp16)[name = string("op_10575_cast_fp16")]; + tensor var_10576_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1025_cast_fp16)[name = string("op_10576_cast_fp16")]; + tensor var_10577_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1027_cast_fp16)[name = string("op_10577_cast_fp16")]; + tensor var_10578_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1029_cast_fp16)[name = string("op_10578_cast_fp16")]; + tensor var_10579_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1031_cast_fp16)[name = string("op_10579_cast_fp16")]; + tensor var_10580_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1033_cast_fp16)[name = string("op_10580_cast_fp16")]; + tensor var_10581_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1035_cast_fp16)[name = string("op_10581_cast_fp16")]; + tensor var_10582_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1037_cast_fp16)[name = string("op_10582_cast_fp16")]; + tensor var_10583_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1039_cast_fp16)[name = string("op_10583_cast_fp16")]; + tensor var_10584_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1041_cast_fp16)[name = string("op_10584_cast_fp16")]; + tensor var_10585_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1043_cast_fp16)[name = string("op_10585_cast_fp16")]; + tensor var_10586_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1045_cast_fp16)[name = string("op_10586_cast_fp16")]; + tensor var_10587_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1047_cast_fp16)[name = string("op_10587_cast_fp16")]; + tensor var_10588_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1049_cast_fp16)[name = string("op_10588_cast_fp16")]; + tensor var_10589_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1051_cast_fp16)[name = string("op_10589_cast_fp16")]; + tensor var_10590_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1053_cast_fp16)[name = string("op_10590_cast_fp16")]; + tensor var_10591_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1055_cast_fp16)[name = string("op_10591_cast_fp16")]; + tensor var_10592_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1057_cast_fp16)[name = string("op_10592_cast_fp16")]; + tensor var_10593_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1059_cast_fp16)[name = string("op_10593_cast_fp16")]; + tensor var_10594_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1061_cast_fp16)[name = string("op_10594_cast_fp16")]; + tensor var_10595_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1063_cast_fp16)[name = string("op_10595_cast_fp16")]; + tensor var_10596_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1065_cast_fp16)[name = string("op_10596_cast_fp16")]; + tensor var_10597_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1067_cast_fp16)[name = string("op_10597_cast_fp16")]; + tensor var_10598_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1069_cast_fp16)[name = string("op_10598_cast_fp16")]; + tensor var_10599_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1071_cast_fp16)[name = string("op_10599_cast_fp16")]; + tensor var_10600_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1073_cast_fp16)[name = string("op_10600_cast_fp16")]; + tensor var_10601_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1075_cast_fp16)[name = string("op_10601_cast_fp16")]; + tensor var_10602_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1077_cast_fp16)[name = string("op_10602_cast_fp16")]; + tensor var_10603_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1079_cast_fp16)[name = string("op_10603_cast_fp16")]; + tensor var_10604_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1081_cast_fp16)[name = string("op_10604_cast_fp16")]; + tensor var_10605_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1083_cast_fp16)[name = string("op_10605_cast_fp16")]; + tensor var_10606_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1085_cast_fp16)[name = string("op_10606_cast_fp16")]; + tensor var_10607_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1087_cast_fp16)[name = string("op_10607_cast_fp16")]; + tensor var_10608_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1089_cast_fp16)[name = string("op_10608_cast_fp16")]; + tensor var_10609_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1091_cast_fp16)[name = string("op_10609_cast_fp16")]; + tensor var_10610_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1093_cast_fp16)[name = string("op_10610_cast_fp16")]; + tensor var_10611_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1095_cast_fp16)[name = string("op_10611_cast_fp16")]; + tensor var_10612_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1097_cast_fp16)[name = string("op_10612_cast_fp16")]; + tensor var_10613_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1099_cast_fp16)[name = string("op_10613_cast_fp16")]; + tensor var_10614_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1101_cast_fp16)[name = string("op_10614_cast_fp16")]; + tensor var_10615_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1103_cast_fp16)[name = string("op_10615_cast_fp16")]; + tensor var_10616_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1105_cast_fp16)[name = string("op_10616_cast_fp16")]; + tensor var_10617_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1107_cast_fp16)[name = string("op_10617_cast_fp16")]; + tensor var_10618_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1109_cast_fp16)[name = string("op_10618_cast_fp16")]; + tensor var_10619_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1111_cast_fp16)[name = string("op_10619_cast_fp16")]; + tensor var_10620_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1113_cast_fp16)[name = string("op_10620_cast_fp16")]; + tensor var_10621_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1115_cast_fp16)[name = string("op_10621_cast_fp16")]; + tensor var_10622_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1117_cast_fp16)[name = string("op_10622_cast_fp16")]; + tensor var_10623_cast_fp16 = softmax(axis = var_9369, x = aw_chunk_1119_cast_fp16)[name = string("op_10623_cast_fp16")]; + string var_10625_equation_0 = const()[name = string("op_10625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10625_cast_fp16 = einsum(equation = var_10625_equation_0, values = (var_10145_cast_fp16, var_10544_cast_fp16))[name = string("op_10625_cast_fp16")]; + string var_10627_equation_0 = const()[name = string("op_10627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10627_cast_fp16 = einsum(equation = var_10627_equation_0, values = (var_10145_cast_fp16, var_10545_cast_fp16))[name = string("op_10627_cast_fp16")]; + string var_10629_equation_0 = const()[name = string("op_10629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10629_cast_fp16 = einsum(equation = var_10629_equation_0, values = (var_10145_cast_fp16, var_10546_cast_fp16))[name = string("op_10629_cast_fp16")]; + string var_10631_equation_0 = const()[name = string("op_10631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10631_cast_fp16 = einsum(equation = var_10631_equation_0, values = (var_10145_cast_fp16, var_10547_cast_fp16))[name = string("op_10631_cast_fp16")]; + string var_10633_equation_0 = const()[name = string("op_10633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10633_cast_fp16 = einsum(equation = var_10633_equation_0, values = (var_10149_cast_fp16, var_10548_cast_fp16))[name = string("op_10633_cast_fp16")]; + string var_10635_equation_0 = const()[name = string("op_10635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10635_cast_fp16 = einsum(equation = var_10635_equation_0, values = (var_10149_cast_fp16, var_10549_cast_fp16))[name = string("op_10635_cast_fp16")]; + string var_10637_equation_0 = const()[name = string("op_10637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10637_cast_fp16 = einsum(equation = var_10637_equation_0, values = (var_10149_cast_fp16, var_10550_cast_fp16))[name = string("op_10637_cast_fp16")]; + string var_10639_equation_0 = const()[name = string("op_10639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10639_cast_fp16 = einsum(equation = var_10639_equation_0, values = (var_10149_cast_fp16, var_10551_cast_fp16))[name = string("op_10639_cast_fp16")]; + string var_10641_equation_0 = const()[name = string("op_10641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10641_cast_fp16 = einsum(equation = var_10641_equation_0, values = (var_10153_cast_fp16, var_10552_cast_fp16))[name = string("op_10641_cast_fp16")]; + string var_10643_equation_0 = const()[name = string("op_10643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10643_cast_fp16 = einsum(equation = var_10643_equation_0, values = (var_10153_cast_fp16, var_10553_cast_fp16))[name = string("op_10643_cast_fp16")]; + string var_10645_equation_0 = const()[name = string("op_10645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10645_cast_fp16 = einsum(equation = var_10645_equation_0, values = (var_10153_cast_fp16, var_10554_cast_fp16))[name = string("op_10645_cast_fp16")]; + string var_10647_equation_0 = const()[name = string("op_10647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10647_cast_fp16 = einsum(equation = var_10647_equation_0, values = (var_10153_cast_fp16, var_10555_cast_fp16))[name = string("op_10647_cast_fp16")]; + string var_10649_equation_0 = const()[name = string("op_10649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10649_cast_fp16 = einsum(equation = var_10649_equation_0, values = (var_10157_cast_fp16, var_10556_cast_fp16))[name = string("op_10649_cast_fp16")]; + string var_10651_equation_0 = const()[name = string("op_10651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10651_cast_fp16 = einsum(equation = var_10651_equation_0, values = (var_10157_cast_fp16, var_10557_cast_fp16))[name = string("op_10651_cast_fp16")]; + string var_10653_equation_0 = const()[name = string("op_10653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10653_cast_fp16 = einsum(equation = var_10653_equation_0, values = (var_10157_cast_fp16, var_10558_cast_fp16))[name = string("op_10653_cast_fp16")]; + string var_10655_equation_0 = const()[name = string("op_10655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10655_cast_fp16 = einsum(equation = var_10655_equation_0, values = (var_10157_cast_fp16, var_10559_cast_fp16))[name = string("op_10655_cast_fp16")]; + string var_10657_equation_0 = const()[name = string("op_10657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10657_cast_fp16 = einsum(equation = var_10657_equation_0, values = (var_10161_cast_fp16, var_10560_cast_fp16))[name = string("op_10657_cast_fp16")]; + string var_10659_equation_0 = const()[name = string("op_10659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10659_cast_fp16 = einsum(equation = var_10659_equation_0, values = (var_10161_cast_fp16, var_10561_cast_fp16))[name = string("op_10659_cast_fp16")]; + string var_10661_equation_0 = const()[name = string("op_10661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10661_cast_fp16 = einsum(equation = var_10661_equation_0, values = (var_10161_cast_fp16, var_10562_cast_fp16))[name = string("op_10661_cast_fp16")]; + string var_10663_equation_0 = const()[name = string("op_10663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10663_cast_fp16 = einsum(equation = var_10663_equation_0, values = (var_10161_cast_fp16, var_10563_cast_fp16))[name = string("op_10663_cast_fp16")]; + string var_10665_equation_0 = const()[name = string("op_10665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10665_cast_fp16 = einsum(equation = var_10665_equation_0, values = (var_10165_cast_fp16, var_10564_cast_fp16))[name = string("op_10665_cast_fp16")]; + string var_10667_equation_0 = const()[name = string("op_10667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10667_cast_fp16 = einsum(equation = var_10667_equation_0, values = (var_10165_cast_fp16, var_10565_cast_fp16))[name = string("op_10667_cast_fp16")]; + string var_10669_equation_0 = const()[name = string("op_10669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10669_cast_fp16 = einsum(equation = var_10669_equation_0, values = (var_10165_cast_fp16, var_10566_cast_fp16))[name = string("op_10669_cast_fp16")]; + string var_10671_equation_0 = const()[name = string("op_10671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10671_cast_fp16 = einsum(equation = var_10671_equation_0, values = (var_10165_cast_fp16, var_10567_cast_fp16))[name = string("op_10671_cast_fp16")]; + string var_10673_equation_0 = const()[name = string("op_10673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10673_cast_fp16 = einsum(equation = var_10673_equation_0, values = (var_10169_cast_fp16, var_10568_cast_fp16))[name = string("op_10673_cast_fp16")]; + string var_10675_equation_0 = const()[name = string("op_10675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10675_cast_fp16 = einsum(equation = var_10675_equation_0, values = (var_10169_cast_fp16, var_10569_cast_fp16))[name = string("op_10675_cast_fp16")]; + string var_10677_equation_0 = const()[name = string("op_10677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10677_cast_fp16 = einsum(equation = var_10677_equation_0, values = (var_10169_cast_fp16, var_10570_cast_fp16))[name = string("op_10677_cast_fp16")]; + string var_10679_equation_0 = const()[name = string("op_10679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10679_cast_fp16 = einsum(equation = var_10679_equation_0, values = (var_10169_cast_fp16, var_10571_cast_fp16))[name = string("op_10679_cast_fp16")]; + string var_10681_equation_0 = const()[name = string("op_10681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10681_cast_fp16 = einsum(equation = var_10681_equation_0, values = (var_10173_cast_fp16, var_10572_cast_fp16))[name = string("op_10681_cast_fp16")]; + string var_10683_equation_0 = const()[name = string("op_10683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10683_cast_fp16 = einsum(equation = var_10683_equation_0, values = (var_10173_cast_fp16, var_10573_cast_fp16))[name = string("op_10683_cast_fp16")]; + string var_10685_equation_0 = const()[name = string("op_10685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10685_cast_fp16 = einsum(equation = var_10685_equation_0, values = (var_10173_cast_fp16, var_10574_cast_fp16))[name = string("op_10685_cast_fp16")]; + string var_10687_equation_0 = const()[name = string("op_10687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10687_cast_fp16 = einsum(equation = var_10687_equation_0, values = (var_10173_cast_fp16, var_10575_cast_fp16))[name = string("op_10687_cast_fp16")]; + string var_10689_equation_0 = const()[name = string("op_10689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10689_cast_fp16 = einsum(equation = var_10689_equation_0, values = (var_10177_cast_fp16, var_10576_cast_fp16))[name = string("op_10689_cast_fp16")]; + string var_10691_equation_0 = const()[name = string("op_10691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10691_cast_fp16 = einsum(equation = var_10691_equation_0, values = (var_10177_cast_fp16, var_10577_cast_fp16))[name = string("op_10691_cast_fp16")]; + string var_10693_equation_0 = const()[name = string("op_10693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10693_cast_fp16 = einsum(equation = var_10693_equation_0, values = (var_10177_cast_fp16, var_10578_cast_fp16))[name = string("op_10693_cast_fp16")]; + string var_10695_equation_0 = const()[name = string("op_10695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10695_cast_fp16 = einsum(equation = var_10695_equation_0, values = (var_10177_cast_fp16, var_10579_cast_fp16))[name = string("op_10695_cast_fp16")]; + string var_10697_equation_0 = const()[name = string("op_10697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10697_cast_fp16 = einsum(equation = var_10697_equation_0, values = (var_10181_cast_fp16, var_10580_cast_fp16))[name = string("op_10697_cast_fp16")]; + string var_10699_equation_0 = const()[name = string("op_10699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10699_cast_fp16 = einsum(equation = var_10699_equation_0, values = (var_10181_cast_fp16, var_10581_cast_fp16))[name = string("op_10699_cast_fp16")]; + string var_10701_equation_0 = const()[name = string("op_10701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10701_cast_fp16 = einsum(equation = var_10701_equation_0, values = (var_10181_cast_fp16, var_10582_cast_fp16))[name = string("op_10701_cast_fp16")]; + string var_10703_equation_0 = const()[name = string("op_10703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10703_cast_fp16 = einsum(equation = var_10703_equation_0, values = (var_10181_cast_fp16, var_10583_cast_fp16))[name = string("op_10703_cast_fp16")]; + string var_10705_equation_0 = const()[name = string("op_10705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10705_cast_fp16 = einsum(equation = var_10705_equation_0, values = (var_10185_cast_fp16, var_10584_cast_fp16))[name = string("op_10705_cast_fp16")]; + string var_10707_equation_0 = const()[name = string("op_10707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10707_cast_fp16 = einsum(equation = var_10707_equation_0, values = (var_10185_cast_fp16, var_10585_cast_fp16))[name = string("op_10707_cast_fp16")]; + string var_10709_equation_0 = const()[name = string("op_10709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10709_cast_fp16 = einsum(equation = var_10709_equation_0, values = (var_10185_cast_fp16, var_10586_cast_fp16))[name = string("op_10709_cast_fp16")]; + string var_10711_equation_0 = const()[name = string("op_10711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10711_cast_fp16 = einsum(equation = var_10711_equation_0, values = (var_10185_cast_fp16, var_10587_cast_fp16))[name = string("op_10711_cast_fp16")]; + string var_10713_equation_0 = const()[name = string("op_10713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10713_cast_fp16 = einsum(equation = var_10713_equation_0, values = (var_10189_cast_fp16, var_10588_cast_fp16))[name = string("op_10713_cast_fp16")]; + string var_10715_equation_0 = const()[name = string("op_10715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10715_cast_fp16 = einsum(equation = var_10715_equation_0, values = (var_10189_cast_fp16, var_10589_cast_fp16))[name = string("op_10715_cast_fp16")]; + string var_10717_equation_0 = const()[name = string("op_10717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10717_cast_fp16 = einsum(equation = var_10717_equation_0, values = (var_10189_cast_fp16, var_10590_cast_fp16))[name = string("op_10717_cast_fp16")]; + string var_10719_equation_0 = const()[name = string("op_10719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10719_cast_fp16 = einsum(equation = var_10719_equation_0, values = (var_10189_cast_fp16, var_10591_cast_fp16))[name = string("op_10719_cast_fp16")]; + string var_10721_equation_0 = const()[name = string("op_10721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10721_cast_fp16 = einsum(equation = var_10721_equation_0, values = (var_10193_cast_fp16, var_10592_cast_fp16))[name = string("op_10721_cast_fp16")]; + string var_10723_equation_0 = const()[name = string("op_10723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10723_cast_fp16 = einsum(equation = var_10723_equation_0, values = (var_10193_cast_fp16, var_10593_cast_fp16))[name = string("op_10723_cast_fp16")]; + string var_10725_equation_0 = const()[name = string("op_10725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10725_cast_fp16 = einsum(equation = var_10725_equation_0, values = (var_10193_cast_fp16, var_10594_cast_fp16))[name = string("op_10725_cast_fp16")]; + string var_10727_equation_0 = const()[name = string("op_10727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10727_cast_fp16 = einsum(equation = var_10727_equation_0, values = (var_10193_cast_fp16, var_10595_cast_fp16))[name = string("op_10727_cast_fp16")]; + string var_10729_equation_0 = const()[name = string("op_10729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10729_cast_fp16 = einsum(equation = var_10729_equation_0, values = (var_10197_cast_fp16, var_10596_cast_fp16))[name = string("op_10729_cast_fp16")]; + string var_10731_equation_0 = const()[name = string("op_10731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10731_cast_fp16 = einsum(equation = var_10731_equation_0, values = (var_10197_cast_fp16, var_10597_cast_fp16))[name = string("op_10731_cast_fp16")]; + string var_10733_equation_0 = const()[name = string("op_10733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10733_cast_fp16 = einsum(equation = var_10733_equation_0, values = (var_10197_cast_fp16, var_10598_cast_fp16))[name = string("op_10733_cast_fp16")]; + string var_10735_equation_0 = const()[name = string("op_10735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10735_cast_fp16 = einsum(equation = var_10735_equation_0, values = (var_10197_cast_fp16, var_10599_cast_fp16))[name = string("op_10735_cast_fp16")]; + string var_10737_equation_0 = const()[name = string("op_10737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10737_cast_fp16 = einsum(equation = var_10737_equation_0, values = (var_10201_cast_fp16, var_10600_cast_fp16))[name = string("op_10737_cast_fp16")]; + string var_10739_equation_0 = const()[name = string("op_10739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10739_cast_fp16 = einsum(equation = var_10739_equation_0, values = (var_10201_cast_fp16, var_10601_cast_fp16))[name = string("op_10739_cast_fp16")]; + string var_10741_equation_0 = const()[name = string("op_10741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10741_cast_fp16 = einsum(equation = var_10741_equation_0, values = (var_10201_cast_fp16, var_10602_cast_fp16))[name = string("op_10741_cast_fp16")]; + string var_10743_equation_0 = const()[name = string("op_10743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10743_cast_fp16 = einsum(equation = var_10743_equation_0, values = (var_10201_cast_fp16, var_10603_cast_fp16))[name = string("op_10743_cast_fp16")]; + string var_10745_equation_0 = const()[name = string("op_10745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10745_cast_fp16 = einsum(equation = var_10745_equation_0, values = (var_10205_cast_fp16, var_10604_cast_fp16))[name = string("op_10745_cast_fp16")]; + string var_10747_equation_0 = const()[name = string("op_10747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10747_cast_fp16 = einsum(equation = var_10747_equation_0, values = (var_10205_cast_fp16, var_10605_cast_fp16))[name = string("op_10747_cast_fp16")]; + string var_10749_equation_0 = const()[name = string("op_10749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10749_cast_fp16 = einsum(equation = var_10749_equation_0, values = (var_10205_cast_fp16, var_10606_cast_fp16))[name = string("op_10749_cast_fp16")]; + string var_10751_equation_0 = const()[name = string("op_10751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10751_cast_fp16 = einsum(equation = var_10751_equation_0, values = (var_10205_cast_fp16, var_10607_cast_fp16))[name = string("op_10751_cast_fp16")]; + string var_10753_equation_0 = const()[name = string("op_10753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10753_cast_fp16 = einsum(equation = var_10753_equation_0, values = (var_10209_cast_fp16, var_10608_cast_fp16))[name = string("op_10753_cast_fp16")]; + string var_10755_equation_0 = const()[name = string("op_10755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10755_cast_fp16 = einsum(equation = var_10755_equation_0, values = (var_10209_cast_fp16, var_10609_cast_fp16))[name = string("op_10755_cast_fp16")]; + string var_10757_equation_0 = const()[name = string("op_10757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10757_cast_fp16 = einsum(equation = var_10757_equation_0, values = (var_10209_cast_fp16, var_10610_cast_fp16))[name = string("op_10757_cast_fp16")]; + string var_10759_equation_0 = const()[name = string("op_10759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10759_cast_fp16 = einsum(equation = var_10759_equation_0, values = (var_10209_cast_fp16, var_10611_cast_fp16))[name = string("op_10759_cast_fp16")]; + string var_10761_equation_0 = const()[name = string("op_10761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10761_cast_fp16 = einsum(equation = var_10761_equation_0, values = (var_10213_cast_fp16, var_10612_cast_fp16))[name = string("op_10761_cast_fp16")]; + string var_10763_equation_0 = const()[name = string("op_10763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10763_cast_fp16 = einsum(equation = var_10763_equation_0, values = (var_10213_cast_fp16, var_10613_cast_fp16))[name = string("op_10763_cast_fp16")]; + string var_10765_equation_0 = const()[name = string("op_10765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10765_cast_fp16 = einsum(equation = var_10765_equation_0, values = (var_10213_cast_fp16, var_10614_cast_fp16))[name = string("op_10765_cast_fp16")]; + string var_10767_equation_0 = const()[name = string("op_10767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10767_cast_fp16 = einsum(equation = var_10767_equation_0, values = (var_10213_cast_fp16, var_10615_cast_fp16))[name = string("op_10767_cast_fp16")]; + string var_10769_equation_0 = const()[name = string("op_10769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10769_cast_fp16 = einsum(equation = var_10769_equation_0, values = (var_10217_cast_fp16, var_10616_cast_fp16))[name = string("op_10769_cast_fp16")]; + string var_10771_equation_0 = const()[name = string("op_10771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10771_cast_fp16 = einsum(equation = var_10771_equation_0, values = (var_10217_cast_fp16, var_10617_cast_fp16))[name = string("op_10771_cast_fp16")]; + string var_10773_equation_0 = const()[name = string("op_10773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10773_cast_fp16 = einsum(equation = var_10773_equation_0, values = (var_10217_cast_fp16, var_10618_cast_fp16))[name = string("op_10773_cast_fp16")]; + string var_10775_equation_0 = const()[name = string("op_10775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10775_cast_fp16 = einsum(equation = var_10775_equation_0, values = (var_10217_cast_fp16, var_10619_cast_fp16))[name = string("op_10775_cast_fp16")]; + string var_10777_equation_0 = const()[name = string("op_10777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10777_cast_fp16 = einsum(equation = var_10777_equation_0, values = (var_10221_cast_fp16, var_10620_cast_fp16))[name = string("op_10777_cast_fp16")]; + string var_10779_equation_0 = const()[name = string("op_10779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10779_cast_fp16 = einsum(equation = var_10779_equation_0, values = (var_10221_cast_fp16, var_10621_cast_fp16))[name = string("op_10779_cast_fp16")]; + string var_10781_equation_0 = const()[name = string("op_10781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10781_cast_fp16 = einsum(equation = var_10781_equation_0, values = (var_10221_cast_fp16, var_10622_cast_fp16))[name = string("op_10781_cast_fp16")]; + string var_10783_equation_0 = const()[name = string("op_10783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10783_cast_fp16 = einsum(equation = var_10783_equation_0, values = (var_10221_cast_fp16, var_10623_cast_fp16))[name = string("op_10783_cast_fp16")]; + bool var_10785_interleave_0 = const()[name = string("op_10785_interleave_0"), val = bool(false)]; + tensor var_10785_cast_fp16 = concat(axis = var_9344, interleave = var_10785_interleave_0, values = (var_10625_cast_fp16, var_10627_cast_fp16, var_10629_cast_fp16, var_10631_cast_fp16))[name = string("op_10785_cast_fp16")]; + bool var_10787_interleave_0 = const()[name = string("op_10787_interleave_0"), val = bool(false)]; + tensor var_10787_cast_fp16 = concat(axis = var_9344, interleave = var_10787_interleave_0, values = (var_10633_cast_fp16, var_10635_cast_fp16, var_10637_cast_fp16, var_10639_cast_fp16))[name = string("op_10787_cast_fp16")]; + bool var_10789_interleave_0 = const()[name = string("op_10789_interleave_0"), val = bool(false)]; + tensor var_10789_cast_fp16 = concat(axis = var_9344, interleave = var_10789_interleave_0, values = (var_10641_cast_fp16, var_10643_cast_fp16, var_10645_cast_fp16, var_10647_cast_fp16))[name = string("op_10789_cast_fp16")]; + bool var_10791_interleave_0 = const()[name = string("op_10791_interleave_0"), val = bool(false)]; + tensor var_10791_cast_fp16 = concat(axis = var_9344, interleave = var_10791_interleave_0, values = (var_10649_cast_fp16, var_10651_cast_fp16, var_10653_cast_fp16, var_10655_cast_fp16))[name = string("op_10791_cast_fp16")]; + bool var_10793_interleave_0 = const()[name = string("op_10793_interleave_0"), val = bool(false)]; + tensor var_10793_cast_fp16 = concat(axis = var_9344, interleave = var_10793_interleave_0, values = (var_10657_cast_fp16, var_10659_cast_fp16, var_10661_cast_fp16, var_10663_cast_fp16))[name = string("op_10793_cast_fp16")]; + bool var_10795_interleave_0 = const()[name = string("op_10795_interleave_0"), val = bool(false)]; + tensor var_10795_cast_fp16 = concat(axis = var_9344, interleave = var_10795_interleave_0, values = (var_10665_cast_fp16, var_10667_cast_fp16, var_10669_cast_fp16, var_10671_cast_fp16))[name = string("op_10795_cast_fp16")]; + bool var_10797_interleave_0 = const()[name = string("op_10797_interleave_0"), val = bool(false)]; + tensor var_10797_cast_fp16 = concat(axis = var_9344, interleave = var_10797_interleave_0, values = (var_10673_cast_fp16, var_10675_cast_fp16, var_10677_cast_fp16, var_10679_cast_fp16))[name = string("op_10797_cast_fp16")]; + bool var_10799_interleave_0 = const()[name = string("op_10799_interleave_0"), val = bool(false)]; + tensor var_10799_cast_fp16 = concat(axis = var_9344, interleave = var_10799_interleave_0, values = (var_10681_cast_fp16, var_10683_cast_fp16, var_10685_cast_fp16, var_10687_cast_fp16))[name = string("op_10799_cast_fp16")]; + bool var_10801_interleave_0 = const()[name = string("op_10801_interleave_0"), val = bool(false)]; + tensor var_10801_cast_fp16 = concat(axis = var_9344, interleave = var_10801_interleave_0, values = (var_10689_cast_fp16, var_10691_cast_fp16, var_10693_cast_fp16, var_10695_cast_fp16))[name = string("op_10801_cast_fp16")]; + bool var_10803_interleave_0 = const()[name = string("op_10803_interleave_0"), val = bool(false)]; + tensor var_10803_cast_fp16 = concat(axis = var_9344, interleave = var_10803_interleave_0, values = (var_10697_cast_fp16, var_10699_cast_fp16, var_10701_cast_fp16, var_10703_cast_fp16))[name = string("op_10803_cast_fp16")]; + bool var_10805_interleave_0 = const()[name = string("op_10805_interleave_0"), val = bool(false)]; + tensor var_10805_cast_fp16 = concat(axis = var_9344, interleave = var_10805_interleave_0, values = (var_10705_cast_fp16, var_10707_cast_fp16, var_10709_cast_fp16, var_10711_cast_fp16))[name = string("op_10805_cast_fp16")]; + bool var_10807_interleave_0 = const()[name = string("op_10807_interleave_0"), val = bool(false)]; + tensor var_10807_cast_fp16 = concat(axis = var_9344, interleave = var_10807_interleave_0, values = (var_10713_cast_fp16, var_10715_cast_fp16, var_10717_cast_fp16, var_10719_cast_fp16))[name = string("op_10807_cast_fp16")]; + bool var_10809_interleave_0 = const()[name = string("op_10809_interleave_0"), val = bool(false)]; + tensor var_10809_cast_fp16 = concat(axis = var_9344, interleave = var_10809_interleave_0, values = (var_10721_cast_fp16, var_10723_cast_fp16, var_10725_cast_fp16, var_10727_cast_fp16))[name = string("op_10809_cast_fp16")]; + bool var_10811_interleave_0 = const()[name = string("op_10811_interleave_0"), val = bool(false)]; + tensor var_10811_cast_fp16 = concat(axis = var_9344, interleave = var_10811_interleave_0, values = (var_10729_cast_fp16, var_10731_cast_fp16, var_10733_cast_fp16, var_10735_cast_fp16))[name = string("op_10811_cast_fp16")]; + bool var_10813_interleave_0 = const()[name = string("op_10813_interleave_0"), val = bool(false)]; + tensor var_10813_cast_fp16 = concat(axis = var_9344, interleave = var_10813_interleave_0, values = (var_10737_cast_fp16, var_10739_cast_fp16, var_10741_cast_fp16, var_10743_cast_fp16))[name = string("op_10813_cast_fp16")]; + bool var_10815_interleave_0 = const()[name = string("op_10815_interleave_0"), val = bool(false)]; + tensor var_10815_cast_fp16 = concat(axis = var_9344, interleave = var_10815_interleave_0, values = (var_10745_cast_fp16, var_10747_cast_fp16, var_10749_cast_fp16, var_10751_cast_fp16))[name = string("op_10815_cast_fp16")]; + bool var_10817_interleave_0 = const()[name = string("op_10817_interleave_0"), val = bool(false)]; + tensor var_10817_cast_fp16 = concat(axis = var_9344, interleave = var_10817_interleave_0, values = (var_10753_cast_fp16, var_10755_cast_fp16, var_10757_cast_fp16, var_10759_cast_fp16))[name = string("op_10817_cast_fp16")]; + bool var_10819_interleave_0 = const()[name = string("op_10819_interleave_0"), val = bool(false)]; + tensor var_10819_cast_fp16 = concat(axis = var_9344, interleave = var_10819_interleave_0, values = (var_10761_cast_fp16, var_10763_cast_fp16, var_10765_cast_fp16, var_10767_cast_fp16))[name = string("op_10819_cast_fp16")]; + bool var_10821_interleave_0 = const()[name = string("op_10821_interleave_0"), val = bool(false)]; + tensor var_10821_cast_fp16 = concat(axis = var_9344, interleave = var_10821_interleave_0, values = (var_10769_cast_fp16, var_10771_cast_fp16, var_10773_cast_fp16, var_10775_cast_fp16))[name = string("op_10821_cast_fp16")]; + bool var_10823_interleave_0 = const()[name = string("op_10823_interleave_0"), val = bool(false)]; + tensor var_10823_cast_fp16 = concat(axis = var_9344, interleave = var_10823_interleave_0, values = (var_10777_cast_fp16, var_10779_cast_fp16, var_10781_cast_fp16, var_10783_cast_fp16))[name = string("op_10823_cast_fp16")]; + bool input_49_interleave_0 = const()[name = string("input_49_interleave_0"), val = bool(false)]; + tensor input_49_cast_fp16 = concat(axis = var_9369, interleave = input_49_interleave_0, values = (var_10785_cast_fp16, var_10787_cast_fp16, var_10789_cast_fp16, var_10791_cast_fp16, var_10793_cast_fp16, var_10795_cast_fp16, var_10797_cast_fp16, var_10799_cast_fp16, var_10801_cast_fp16, var_10803_cast_fp16, var_10805_cast_fp16, var_10807_cast_fp16, var_10809_cast_fp16, var_10811_cast_fp16, var_10813_cast_fp16, var_10815_cast_fp16, var_10817_cast_fp16, var_10819_cast_fp16, var_10821_cast_fp16, var_10823_cast_fp16))[name = string("input_49_cast_fp16")]; + string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; + tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; + tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; + int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260624960)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263901824)))]; + tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_10842_to_fp16 = const()[name = string("op_10842_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_10842_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263904448)))]; + tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263907072)))]; + fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; + string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; + tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; + tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; + int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263909696)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277016960)))]; + tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; + string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; + tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277027264)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290134528)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; + int32 var_10871 = const()[name = string("op_10871"), val = int32(3)]; + int32 var_10896 = const()[name = string("op_10896"), val = int32(1)]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_10913_to_fp16 = const()[name = string("op_10913_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_10913_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290137152)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290139776)))]; + fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290142400)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293419264)))]; + tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; + string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; + tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; + tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; + int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293421888)))]; + tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; + string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; + tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; + tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; + int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296698752)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299975616)))]; + tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_10951_begin_0 = const()[name = string("op_10951_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10951_end_0 = const()[name = string("op_10951_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10951_end_mask_0 = const()[name = string("op_10951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10951_cast_fp16 = slice_by_index(begin = var_10951_begin_0, end = var_10951_end_0, end_mask = var_10951_end_mask_0, x = query_15_cast_fp16)[name = string("op_10951_cast_fp16")]; + tensor var_10955_begin_0 = const()[name = string("op_10955_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10955_end_0 = const()[name = string("op_10955_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10955_end_mask_0 = const()[name = string("op_10955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10955_cast_fp16 = slice_by_index(begin = var_10955_begin_0, end = var_10955_end_0, end_mask = var_10955_end_mask_0, x = query_15_cast_fp16)[name = string("op_10955_cast_fp16")]; + tensor var_10959_begin_0 = const()[name = string("op_10959_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10959_end_0 = const()[name = string("op_10959_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10959_end_mask_0 = const()[name = string("op_10959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10959_cast_fp16 = slice_by_index(begin = var_10959_begin_0, end = var_10959_end_0, end_mask = var_10959_end_mask_0, x = query_15_cast_fp16)[name = string("op_10959_cast_fp16")]; + tensor var_10963_begin_0 = const()[name = string("op_10963_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10963_end_0 = const()[name = string("op_10963_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10963_end_mask_0 = const()[name = string("op_10963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10963_cast_fp16 = slice_by_index(begin = var_10963_begin_0, end = var_10963_end_0, end_mask = var_10963_end_mask_0, x = query_15_cast_fp16)[name = string("op_10963_cast_fp16")]; + tensor var_10967_begin_0 = const()[name = string("op_10967_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10967_end_0 = const()[name = string("op_10967_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10967_end_mask_0 = const()[name = string("op_10967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10967_cast_fp16 = slice_by_index(begin = var_10967_begin_0, end = var_10967_end_0, end_mask = var_10967_end_mask_0, x = query_15_cast_fp16)[name = string("op_10967_cast_fp16")]; + tensor var_10971_begin_0 = const()[name = string("op_10971_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10971_end_0 = const()[name = string("op_10971_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10971_end_mask_0 = const()[name = string("op_10971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10971_cast_fp16 = slice_by_index(begin = var_10971_begin_0, end = var_10971_end_0, end_mask = var_10971_end_mask_0, x = query_15_cast_fp16)[name = string("op_10971_cast_fp16")]; + tensor var_10975_begin_0 = const()[name = string("op_10975_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10975_end_0 = const()[name = string("op_10975_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10975_end_mask_0 = const()[name = string("op_10975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10975_cast_fp16 = slice_by_index(begin = var_10975_begin_0, end = var_10975_end_0, end_mask = var_10975_end_mask_0, x = query_15_cast_fp16)[name = string("op_10975_cast_fp16")]; + tensor var_10979_begin_0 = const()[name = string("op_10979_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10979_end_0 = const()[name = string("op_10979_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10979_end_mask_0 = const()[name = string("op_10979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = query_15_cast_fp16)[name = string("op_10979_cast_fp16")]; + tensor var_10983_begin_0 = const()[name = string("op_10983_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10983_end_0 = const()[name = string("op_10983_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10983_end_mask_0 = const()[name = string("op_10983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10983_cast_fp16 = slice_by_index(begin = var_10983_begin_0, end = var_10983_end_0, end_mask = var_10983_end_mask_0, x = query_15_cast_fp16)[name = string("op_10983_cast_fp16")]; + tensor var_10987_begin_0 = const()[name = string("op_10987_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10987_end_0 = const()[name = string("op_10987_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10987_end_mask_0 = const()[name = string("op_10987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10987_cast_fp16 = slice_by_index(begin = var_10987_begin_0, end = var_10987_end_0, end_mask = var_10987_end_mask_0, x = query_15_cast_fp16)[name = string("op_10987_cast_fp16")]; + tensor var_10991_begin_0 = const()[name = string("op_10991_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10991_end_0 = const()[name = string("op_10991_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10991_end_mask_0 = const()[name = string("op_10991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10991_cast_fp16 = slice_by_index(begin = var_10991_begin_0, end = var_10991_end_0, end_mask = var_10991_end_mask_0, x = query_15_cast_fp16)[name = string("op_10991_cast_fp16")]; + tensor var_10995_begin_0 = const()[name = string("op_10995_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10995_end_0 = const()[name = string("op_10995_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10995_end_mask_0 = const()[name = string("op_10995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10995_cast_fp16 = slice_by_index(begin = var_10995_begin_0, end = var_10995_end_0, end_mask = var_10995_end_mask_0, x = query_15_cast_fp16)[name = string("op_10995_cast_fp16")]; + tensor var_10999_begin_0 = const()[name = string("op_10999_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10999_end_0 = const()[name = string("op_10999_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_10999_end_mask_0 = const()[name = string("op_10999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10999_cast_fp16 = slice_by_index(begin = var_10999_begin_0, end = var_10999_end_0, end_mask = var_10999_end_mask_0, x = query_15_cast_fp16)[name = string("op_10999_cast_fp16")]; + tensor var_11003_begin_0 = const()[name = string("op_11003_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11003_end_0 = const()[name = string("op_11003_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_11003_end_mask_0 = const()[name = string("op_11003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11003_cast_fp16 = slice_by_index(begin = var_11003_begin_0, end = var_11003_end_0, end_mask = var_11003_end_mask_0, x = query_15_cast_fp16)[name = string("op_11003_cast_fp16")]; + tensor var_11007_begin_0 = const()[name = string("op_11007_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11007_end_0 = const()[name = string("op_11007_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_11007_end_mask_0 = const()[name = string("op_11007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = query_15_cast_fp16)[name = string("op_11007_cast_fp16")]; + tensor var_11011_begin_0 = const()[name = string("op_11011_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11011_end_0 = const()[name = string("op_11011_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_11011_end_mask_0 = const()[name = string("op_11011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11011_cast_fp16 = slice_by_index(begin = var_11011_begin_0, end = var_11011_end_0, end_mask = var_11011_end_mask_0, x = query_15_cast_fp16)[name = string("op_11011_cast_fp16")]; + tensor var_11015_begin_0 = const()[name = string("op_11015_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11015_end_0 = const()[name = string("op_11015_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_11015_end_mask_0 = const()[name = string("op_11015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11015_cast_fp16 = slice_by_index(begin = var_11015_begin_0, end = var_11015_end_0, end_mask = var_11015_end_mask_0, x = query_15_cast_fp16)[name = string("op_11015_cast_fp16")]; + tensor var_11019_begin_0 = const()[name = string("op_11019_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11019_end_0 = const()[name = string("op_11019_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_11019_end_mask_0 = const()[name = string("op_11019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11019_cast_fp16 = slice_by_index(begin = var_11019_begin_0, end = var_11019_end_0, end_mask = var_11019_end_mask_0, x = query_15_cast_fp16)[name = string("op_11019_cast_fp16")]; + tensor var_11023_begin_0 = const()[name = string("op_11023_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11023_end_0 = const()[name = string("op_11023_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_11023_end_mask_0 = const()[name = string("op_11023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11023_cast_fp16 = slice_by_index(begin = var_11023_begin_0, end = var_11023_end_0, end_mask = var_11023_end_mask_0, x = query_15_cast_fp16)[name = string("op_11023_cast_fp16")]; + tensor var_11027_begin_0 = const()[name = string("op_11027_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11027_end_0 = const()[name = string("op_11027_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_11027_end_mask_0 = const()[name = string("op_11027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11027_cast_fp16 = slice_by_index(begin = var_11027_begin_0, end = var_11027_end_0, end_mask = var_11027_end_mask_0, x = query_15_cast_fp16)[name = string("op_11027_cast_fp16")]; + tensor var_11036_begin_0 = const()[name = string("op_11036_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11036_end_0 = const()[name = string("op_11036_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11036_end_mask_0 = const()[name = string("op_11036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11036_cast_fp16 = slice_by_index(begin = var_11036_begin_0, end = var_11036_end_0, end_mask = var_11036_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11036_cast_fp16")]; + tensor var_11043_begin_0 = const()[name = string("op_11043_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11043_end_0 = const()[name = string("op_11043_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11043_end_mask_0 = const()[name = string("op_11043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11043_cast_fp16 = slice_by_index(begin = var_11043_begin_0, end = var_11043_end_0, end_mask = var_11043_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11043_cast_fp16")]; + tensor var_11050_begin_0 = const()[name = string("op_11050_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11050_end_0 = const()[name = string("op_11050_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11050_end_mask_0 = const()[name = string("op_11050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11050_cast_fp16 = slice_by_index(begin = var_11050_begin_0, end = var_11050_end_0, end_mask = var_11050_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11050_cast_fp16")]; + tensor var_11057_begin_0 = const()[name = string("op_11057_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11057_end_0 = const()[name = string("op_11057_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11057_end_mask_0 = const()[name = string("op_11057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11057_cast_fp16 = slice_by_index(begin = var_11057_begin_0, end = var_11057_end_0, end_mask = var_11057_end_mask_0, x = var_10951_cast_fp16)[name = string("op_11057_cast_fp16")]; + tensor var_11064_begin_0 = const()[name = string("op_11064_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11064_end_0 = const()[name = string("op_11064_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11064_end_mask_0 = const()[name = string("op_11064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11064_cast_fp16 = slice_by_index(begin = var_11064_begin_0, end = var_11064_end_0, end_mask = var_11064_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11064_cast_fp16")]; + tensor var_11071_begin_0 = const()[name = string("op_11071_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11071_end_0 = const()[name = string("op_11071_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11071_end_mask_0 = const()[name = string("op_11071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11071_cast_fp16 = slice_by_index(begin = var_11071_begin_0, end = var_11071_end_0, end_mask = var_11071_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11071_cast_fp16")]; + tensor var_11078_begin_0 = const()[name = string("op_11078_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11078_end_0 = const()[name = string("op_11078_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11078_end_mask_0 = const()[name = string("op_11078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11078_cast_fp16 = slice_by_index(begin = var_11078_begin_0, end = var_11078_end_0, end_mask = var_11078_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11078_cast_fp16")]; + tensor var_11085_begin_0 = const()[name = string("op_11085_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11085_end_0 = const()[name = string("op_11085_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11085_end_mask_0 = const()[name = string("op_11085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11085_cast_fp16 = slice_by_index(begin = var_11085_begin_0, end = var_11085_end_0, end_mask = var_11085_end_mask_0, x = var_10955_cast_fp16)[name = string("op_11085_cast_fp16")]; + tensor var_11092_begin_0 = const()[name = string("op_11092_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11092_end_0 = const()[name = string("op_11092_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11092_end_mask_0 = const()[name = string("op_11092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11092_cast_fp16 = slice_by_index(begin = var_11092_begin_0, end = var_11092_end_0, end_mask = var_11092_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11092_cast_fp16")]; + tensor var_11099_begin_0 = const()[name = string("op_11099_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11099_end_0 = const()[name = string("op_11099_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11099_end_mask_0 = const()[name = string("op_11099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11099_cast_fp16 = slice_by_index(begin = var_11099_begin_0, end = var_11099_end_0, end_mask = var_11099_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11099_cast_fp16")]; + tensor var_11106_begin_0 = const()[name = string("op_11106_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11106_end_0 = const()[name = string("op_11106_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11106_end_mask_0 = const()[name = string("op_11106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11106_cast_fp16 = slice_by_index(begin = var_11106_begin_0, end = var_11106_end_0, end_mask = var_11106_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11106_cast_fp16")]; + tensor var_11113_begin_0 = const()[name = string("op_11113_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11113_end_0 = const()[name = string("op_11113_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11113_end_mask_0 = const()[name = string("op_11113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11113_cast_fp16 = slice_by_index(begin = var_11113_begin_0, end = var_11113_end_0, end_mask = var_11113_end_mask_0, x = var_10959_cast_fp16)[name = string("op_11113_cast_fp16")]; + tensor var_11120_begin_0 = const()[name = string("op_11120_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11120_end_0 = const()[name = string("op_11120_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11120_end_mask_0 = const()[name = string("op_11120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11120_cast_fp16 = slice_by_index(begin = var_11120_begin_0, end = var_11120_end_0, end_mask = var_11120_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11120_cast_fp16")]; + tensor var_11127_begin_0 = const()[name = string("op_11127_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11127_end_0 = const()[name = string("op_11127_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11127_end_mask_0 = const()[name = string("op_11127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11127_cast_fp16 = slice_by_index(begin = var_11127_begin_0, end = var_11127_end_0, end_mask = var_11127_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11127_cast_fp16")]; + tensor var_11134_begin_0 = const()[name = string("op_11134_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11134_end_0 = const()[name = string("op_11134_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11134_end_mask_0 = const()[name = string("op_11134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11134_cast_fp16 = slice_by_index(begin = var_11134_begin_0, end = var_11134_end_0, end_mask = var_11134_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11134_cast_fp16")]; + tensor var_11141_begin_0 = const()[name = string("op_11141_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11141_end_0 = const()[name = string("op_11141_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11141_end_mask_0 = const()[name = string("op_11141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11141_cast_fp16 = slice_by_index(begin = var_11141_begin_0, end = var_11141_end_0, end_mask = var_11141_end_mask_0, x = var_10963_cast_fp16)[name = string("op_11141_cast_fp16")]; + tensor var_11148_begin_0 = const()[name = string("op_11148_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11148_end_0 = const()[name = string("op_11148_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11148_end_mask_0 = const()[name = string("op_11148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11148_cast_fp16 = slice_by_index(begin = var_11148_begin_0, end = var_11148_end_0, end_mask = var_11148_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11148_cast_fp16")]; + tensor var_11155_begin_0 = const()[name = string("op_11155_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11155_end_0 = const()[name = string("op_11155_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11155_end_mask_0 = const()[name = string("op_11155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11155_cast_fp16 = slice_by_index(begin = var_11155_begin_0, end = var_11155_end_0, end_mask = var_11155_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11155_cast_fp16")]; + tensor var_11162_begin_0 = const()[name = string("op_11162_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11162_end_0 = const()[name = string("op_11162_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11162_end_mask_0 = const()[name = string("op_11162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11162_cast_fp16 = slice_by_index(begin = var_11162_begin_0, end = var_11162_end_0, end_mask = var_11162_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11162_cast_fp16")]; + tensor var_11169_begin_0 = const()[name = string("op_11169_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11169_end_0 = const()[name = string("op_11169_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11169_end_mask_0 = const()[name = string("op_11169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11169_cast_fp16 = slice_by_index(begin = var_11169_begin_0, end = var_11169_end_0, end_mask = var_11169_end_mask_0, x = var_10967_cast_fp16)[name = string("op_11169_cast_fp16")]; + tensor var_11176_begin_0 = const()[name = string("op_11176_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11176_end_0 = const()[name = string("op_11176_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11176_end_mask_0 = const()[name = string("op_11176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11176_cast_fp16 = slice_by_index(begin = var_11176_begin_0, end = var_11176_end_0, end_mask = var_11176_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11176_cast_fp16")]; + tensor var_11183_begin_0 = const()[name = string("op_11183_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11183_end_0 = const()[name = string("op_11183_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11183_end_mask_0 = const()[name = string("op_11183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11183_cast_fp16 = slice_by_index(begin = var_11183_begin_0, end = var_11183_end_0, end_mask = var_11183_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11183_cast_fp16")]; + tensor var_11190_begin_0 = const()[name = string("op_11190_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11190_end_0 = const()[name = string("op_11190_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11190_end_mask_0 = const()[name = string("op_11190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11190_cast_fp16 = slice_by_index(begin = var_11190_begin_0, end = var_11190_end_0, end_mask = var_11190_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11190_cast_fp16")]; + tensor var_11197_begin_0 = const()[name = string("op_11197_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11197_end_0 = const()[name = string("op_11197_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11197_end_mask_0 = const()[name = string("op_11197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11197_cast_fp16 = slice_by_index(begin = var_11197_begin_0, end = var_11197_end_0, end_mask = var_11197_end_mask_0, x = var_10971_cast_fp16)[name = string("op_11197_cast_fp16")]; + tensor var_11204_begin_0 = const()[name = string("op_11204_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11204_end_0 = const()[name = string("op_11204_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11204_end_mask_0 = const()[name = string("op_11204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11204_cast_fp16 = slice_by_index(begin = var_11204_begin_0, end = var_11204_end_0, end_mask = var_11204_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11204_cast_fp16")]; + tensor var_11211_begin_0 = const()[name = string("op_11211_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11211_end_0 = const()[name = string("op_11211_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11211_end_mask_0 = const()[name = string("op_11211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11211_cast_fp16 = slice_by_index(begin = var_11211_begin_0, end = var_11211_end_0, end_mask = var_11211_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11211_cast_fp16")]; + tensor var_11218_begin_0 = const()[name = string("op_11218_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11218_end_0 = const()[name = string("op_11218_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11218_end_mask_0 = const()[name = string("op_11218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11218_cast_fp16 = slice_by_index(begin = var_11218_begin_0, end = var_11218_end_0, end_mask = var_11218_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11218_cast_fp16")]; + tensor var_11225_begin_0 = const()[name = string("op_11225_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11225_end_0 = const()[name = string("op_11225_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11225_end_mask_0 = const()[name = string("op_11225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11225_cast_fp16 = slice_by_index(begin = var_11225_begin_0, end = var_11225_end_0, end_mask = var_11225_end_mask_0, x = var_10975_cast_fp16)[name = string("op_11225_cast_fp16")]; + tensor var_11232_begin_0 = const()[name = string("op_11232_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11232_end_0 = const()[name = string("op_11232_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11232_end_mask_0 = const()[name = string("op_11232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11232_cast_fp16 = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11232_cast_fp16")]; + tensor var_11239_begin_0 = const()[name = string("op_11239_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11239_end_0 = const()[name = string("op_11239_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11239_end_mask_0 = const()[name = string("op_11239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11239_cast_fp16 = slice_by_index(begin = var_11239_begin_0, end = var_11239_end_0, end_mask = var_11239_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11239_cast_fp16")]; + tensor var_11246_begin_0 = const()[name = string("op_11246_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11246_end_0 = const()[name = string("op_11246_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11246_end_mask_0 = const()[name = string("op_11246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11246_cast_fp16 = slice_by_index(begin = var_11246_begin_0, end = var_11246_end_0, end_mask = var_11246_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11246_cast_fp16")]; + tensor var_11253_begin_0 = const()[name = string("op_11253_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11253_end_0 = const()[name = string("op_11253_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11253_end_mask_0 = const()[name = string("op_11253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11253_cast_fp16 = slice_by_index(begin = var_11253_begin_0, end = var_11253_end_0, end_mask = var_11253_end_mask_0, x = var_10979_cast_fp16)[name = string("op_11253_cast_fp16")]; + tensor var_11260_begin_0 = const()[name = string("op_11260_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11260_end_0 = const()[name = string("op_11260_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11260_end_mask_0 = const()[name = string("op_11260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11260_cast_fp16 = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11260_cast_fp16")]; + tensor var_11267_begin_0 = const()[name = string("op_11267_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11267_end_0 = const()[name = string("op_11267_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11267_end_mask_0 = const()[name = string("op_11267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11267_cast_fp16 = slice_by_index(begin = var_11267_begin_0, end = var_11267_end_0, end_mask = var_11267_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11267_cast_fp16")]; + tensor var_11274_begin_0 = const()[name = string("op_11274_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11274_end_0 = const()[name = string("op_11274_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11274_end_mask_0 = const()[name = string("op_11274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11274_cast_fp16 = slice_by_index(begin = var_11274_begin_0, end = var_11274_end_0, end_mask = var_11274_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11274_cast_fp16")]; + tensor var_11281_begin_0 = const()[name = string("op_11281_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11281_end_0 = const()[name = string("op_11281_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11281_end_mask_0 = const()[name = string("op_11281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11281_cast_fp16 = slice_by_index(begin = var_11281_begin_0, end = var_11281_end_0, end_mask = var_11281_end_mask_0, x = var_10983_cast_fp16)[name = string("op_11281_cast_fp16")]; + tensor var_11288_begin_0 = const()[name = string("op_11288_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11288_end_0 = const()[name = string("op_11288_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11288_end_mask_0 = const()[name = string("op_11288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11288_cast_fp16 = slice_by_index(begin = var_11288_begin_0, end = var_11288_end_0, end_mask = var_11288_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11288_cast_fp16")]; + tensor var_11295_begin_0 = const()[name = string("op_11295_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11295_end_0 = const()[name = string("op_11295_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11295_end_mask_0 = const()[name = string("op_11295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11295_cast_fp16 = slice_by_index(begin = var_11295_begin_0, end = var_11295_end_0, end_mask = var_11295_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11295_cast_fp16")]; + tensor var_11302_begin_0 = const()[name = string("op_11302_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11302_end_0 = const()[name = string("op_11302_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11302_end_mask_0 = const()[name = string("op_11302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11302_cast_fp16 = slice_by_index(begin = var_11302_begin_0, end = var_11302_end_0, end_mask = var_11302_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11302_cast_fp16")]; + tensor var_11309_begin_0 = const()[name = string("op_11309_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11309_end_0 = const()[name = string("op_11309_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11309_end_mask_0 = const()[name = string("op_11309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11309_cast_fp16 = slice_by_index(begin = var_11309_begin_0, end = var_11309_end_0, end_mask = var_11309_end_mask_0, x = var_10987_cast_fp16)[name = string("op_11309_cast_fp16")]; + tensor var_11316_begin_0 = const()[name = string("op_11316_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11316_end_0 = const()[name = string("op_11316_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11316_end_mask_0 = const()[name = string("op_11316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11316_cast_fp16 = slice_by_index(begin = var_11316_begin_0, end = var_11316_end_0, end_mask = var_11316_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11316_cast_fp16")]; + tensor var_11323_begin_0 = const()[name = string("op_11323_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11323_end_0 = const()[name = string("op_11323_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11323_end_mask_0 = const()[name = string("op_11323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11323_cast_fp16 = slice_by_index(begin = var_11323_begin_0, end = var_11323_end_0, end_mask = var_11323_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11323_cast_fp16")]; + tensor var_11330_begin_0 = const()[name = string("op_11330_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11330_end_0 = const()[name = string("op_11330_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11330_end_mask_0 = const()[name = string("op_11330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11330_cast_fp16 = slice_by_index(begin = var_11330_begin_0, end = var_11330_end_0, end_mask = var_11330_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11330_cast_fp16")]; + tensor var_11337_begin_0 = const()[name = string("op_11337_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11337_end_0 = const()[name = string("op_11337_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11337_end_mask_0 = const()[name = string("op_11337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11337_cast_fp16 = slice_by_index(begin = var_11337_begin_0, end = var_11337_end_0, end_mask = var_11337_end_mask_0, x = var_10991_cast_fp16)[name = string("op_11337_cast_fp16")]; + tensor var_11344_begin_0 = const()[name = string("op_11344_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11344_end_0 = const()[name = string("op_11344_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11344_end_mask_0 = const()[name = string("op_11344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11344_cast_fp16 = slice_by_index(begin = var_11344_begin_0, end = var_11344_end_0, end_mask = var_11344_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11344_cast_fp16")]; + tensor var_11351_begin_0 = const()[name = string("op_11351_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11351_end_0 = const()[name = string("op_11351_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11351_end_mask_0 = const()[name = string("op_11351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11351_cast_fp16 = slice_by_index(begin = var_11351_begin_0, end = var_11351_end_0, end_mask = var_11351_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11351_cast_fp16")]; + tensor var_11358_begin_0 = const()[name = string("op_11358_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11358_end_0 = const()[name = string("op_11358_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11358_end_mask_0 = const()[name = string("op_11358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11358_cast_fp16 = slice_by_index(begin = var_11358_begin_0, end = var_11358_end_0, end_mask = var_11358_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11358_cast_fp16")]; + tensor var_11365_begin_0 = const()[name = string("op_11365_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11365_end_0 = const()[name = string("op_11365_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11365_end_mask_0 = const()[name = string("op_11365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11365_cast_fp16 = slice_by_index(begin = var_11365_begin_0, end = var_11365_end_0, end_mask = var_11365_end_mask_0, x = var_10995_cast_fp16)[name = string("op_11365_cast_fp16")]; + tensor var_11372_begin_0 = const()[name = string("op_11372_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11372_end_0 = const()[name = string("op_11372_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11372_end_mask_0 = const()[name = string("op_11372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11372_cast_fp16 = slice_by_index(begin = var_11372_begin_0, end = var_11372_end_0, end_mask = var_11372_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11372_cast_fp16")]; + tensor var_11379_begin_0 = const()[name = string("op_11379_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11379_end_0 = const()[name = string("op_11379_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11379_end_mask_0 = const()[name = string("op_11379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11379_cast_fp16 = slice_by_index(begin = var_11379_begin_0, end = var_11379_end_0, end_mask = var_11379_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11379_cast_fp16")]; + tensor var_11386_begin_0 = const()[name = string("op_11386_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11386_end_0 = const()[name = string("op_11386_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11386_end_mask_0 = const()[name = string("op_11386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11386_cast_fp16 = slice_by_index(begin = var_11386_begin_0, end = var_11386_end_0, end_mask = var_11386_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11386_cast_fp16")]; + tensor var_11393_begin_0 = const()[name = string("op_11393_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11393_end_0 = const()[name = string("op_11393_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11393_end_mask_0 = const()[name = string("op_11393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11393_cast_fp16 = slice_by_index(begin = var_11393_begin_0, end = var_11393_end_0, end_mask = var_11393_end_mask_0, x = var_10999_cast_fp16)[name = string("op_11393_cast_fp16")]; + tensor var_11400_begin_0 = const()[name = string("op_11400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11400_end_0 = const()[name = string("op_11400_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11400_end_mask_0 = const()[name = string("op_11400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11400_cast_fp16 = slice_by_index(begin = var_11400_begin_0, end = var_11400_end_0, end_mask = var_11400_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11400_cast_fp16")]; + tensor var_11407_begin_0 = const()[name = string("op_11407_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11407_end_0 = const()[name = string("op_11407_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11407_end_mask_0 = const()[name = string("op_11407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11407_cast_fp16 = slice_by_index(begin = var_11407_begin_0, end = var_11407_end_0, end_mask = var_11407_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11407_cast_fp16")]; + tensor var_11414_begin_0 = const()[name = string("op_11414_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11414_end_0 = const()[name = string("op_11414_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11414_end_mask_0 = const()[name = string("op_11414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11414_cast_fp16 = slice_by_index(begin = var_11414_begin_0, end = var_11414_end_0, end_mask = var_11414_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11414_cast_fp16")]; + tensor var_11421_begin_0 = const()[name = string("op_11421_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11421_end_0 = const()[name = string("op_11421_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11421_end_mask_0 = const()[name = string("op_11421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11421_cast_fp16 = slice_by_index(begin = var_11421_begin_0, end = var_11421_end_0, end_mask = var_11421_end_mask_0, x = var_11003_cast_fp16)[name = string("op_11421_cast_fp16")]; + tensor var_11428_begin_0 = const()[name = string("op_11428_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11428_end_0 = const()[name = string("op_11428_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11428_end_mask_0 = const()[name = string("op_11428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11428_cast_fp16 = slice_by_index(begin = var_11428_begin_0, end = var_11428_end_0, end_mask = var_11428_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11428_cast_fp16")]; + tensor var_11435_begin_0 = const()[name = string("op_11435_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11435_end_0 = const()[name = string("op_11435_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11435_end_mask_0 = const()[name = string("op_11435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11435_cast_fp16 = slice_by_index(begin = var_11435_begin_0, end = var_11435_end_0, end_mask = var_11435_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11435_cast_fp16")]; + tensor var_11442_begin_0 = const()[name = string("op_11442_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11442_end_0 = const()[name = string("op_11442_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11442_end_mask_0 = const()[name = string("op_11442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11442_cast_fp16 = slice_by_index(begin = var_11442_begin_0, end = var_11442_end_0, end_mask = var_11442_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11442_cast_fp16")]; + tensor var_11449_begin_0 = const()[name = string("op_11449_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11449_end_0 = const()[name = string("op_11449_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11449_end_mask_0 = const()[name = string("op_11449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11449_cast_fp16 = slice_by_index(begin = var_11449_begin_0, end = var_11449_end_0, end_mask = var_11449_end_mask_0, x = var_11007_cast_fp16)[name = string("op_11449_cast_fp16")]; + tensor var_11456_begin_0 = const()[name = string("op_11456_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11456_end_0 = const()[name = string("op_11456_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11456_end_mask_0 = const()[name = string("op_11456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11456_cast_fp16 = slice_by_index(begin = var_11456_begin_0, end = var_11456_end_0, end_mask = var_11456_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11456_cast_fp16")]; + tensor var_11463_begin_0 = const()[name = string("op_11463_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11463_end_0 = const()[name = string("op_11463_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11463_end_mask_0 = const()[name = string("op_11463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11463_cast_fp16 = slice_by_index(begin = var_11463_begin_0, end = var_11463_end_0, end_mask = var_11463_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11463_cast_fp16")]; + tensor var_11470_begin_0 = const()[name = string("op_11470_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11470_end_0 = const()[name = string("op_11470_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11470_end_mask_0 = const()[name = string("op_11470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11470_cast_fp16 = slice_by_index(begin = var_11470_begin_0, end = var_11470_end_0, end_mask = var_11470_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11470_cast_fp16")]; + tensor var_11477_begin_0 = const()[name = string("op_11477_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11477_end_0 = const()[name = string("op_11477_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11477_end_mask_0 = const()[name = string("op_11477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11477_cast_fp16 = slice_by_index(begin = var_11477_begin_0, end = var_11477_end_0, end_mask = var_11477_end_mask_0, x = var_11011_cast_fp16)[name = string("op_11477_cast_fp16")]; + tensor var_11484_begin_0 = const()[name = string("op_11484_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11484_end_0 = const()[name = string("op_11484_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11484_end_mask_0 = const()[name = string("op_11484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11484_cast_fp16 = slice_by_index(begin = var_11484_begin_0, end = var_11484_end_0, end_mask = var_11484_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11484_cast_fp16")]; + tensor var_11491_begin_0 = const()[name = string("op_11491_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11491_end_0 = const()[name = string("op_11491_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11491_end_mask_0 = const()[name = string("op_11491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11491_cast_fp16 = slice_by_index(begin = var_11491_begin_0, end = var_11491_end_0, end_mask = var_11491_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11491_cast_fp16")]; + tensor var_11498_begin_0 = const()[name = string("op_11498_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11498_end_0 = const()[name = string("op_11498_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11498_end_mask_0 = const()[name = string("op_11498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11498_cast_fp16 = slice_by_index(begin = var_11498_begin_0, end = var_11498_end_0, end_mask = var_11498_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11498_cast_fp16")]; + tensor var_11505_begin_0 = const()[name = string("op_11505_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11505_end_0 = const()[name = string("op_11505_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11505_end_mask_0 = const()[name = string("op_11505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11505_cast_fp16 = slice_by_index(begin = var_11505_begin_0, end = var_11505_end_0, end_mask = var_11505_end_mask_0, x = var_11015_cast_fp16)[name = string("op_11505_cast_fp16")]; + tensor var_11512_begin_0 = const()[name = string("op_11512_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11512_end_0 = const()[name = string("op_11512_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11512_end_mask_0 = const()[name = string("op_11512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11512_cast_fp16 = slice_by_index(begin = var_11512_begin_0, end = var_11512_end_0, end_mask = var_11512_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11512_cast_fp16")]; + tensor var_11519_begin_0 = const()[name = string("op_11519_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11519_end_0 = const()[name = string("op_11519_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11519_end_mask_0 = const()[name = string("op_11519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11519_cast_fp16 = slice_by_index(begin = var_11519_begin_0, end = var_11519_end_0, end_mask = var_11519_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11519_cast_fp16")]; + tensor var_11526_begin_0 = const()[name = string("op_11526_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11526_end_0 = const()[name = string("op_11526_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11526_end_mask_0 = const()[name = string("op_11526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11526_cast_fp16 = slice_by_index(begin = var_11526_begin_0, end = var_11526_end_0, end_mask = var_11526_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11526_cast_fp16")]; + tensor var_11533_begin_0 = const()[name = string("op_11533_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11533_end_0 = const()[name = string("op_11533_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11533_end_mask_0 = const()[name = string("op_11533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11533_cast_fp16 = slice_by_index(begin = var_11533_begin_0, end = var_11533_end_0, end_mask = var_11533_end_mask_0, x = var_11019_cast_fp16)[name = string("op_11533_cast_fp16")]; + tensor var_11540_begin_0 = const()[name = string("op_11540_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11540_end_0 = const()[name = string("op_11540_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11540_end_mask_0 = const()[name = string("op_11540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11540_cast_fp16 = slice_by_index(begin = var_11540_begin_0, end = var_11540_end_0, end_mask = var_11540_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11540_cast_fp16")]; + tensor var_11547_begin_0 = const()[name = string("op_11547_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11547_end_0 = const()[name = string("op_11547_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11547_end_mask_0 = const()[name = string("op_11547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11547_cast_fp16 = slice_by_index(begin = var_11547_begin_0, end = var_11547_end_0, end_mask = var_11547_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11547_cast_fp16")]; + tensor var_11554_begin_0 = const()[name = string("op_11554_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11554_end_0 = const()[name = string("op_11554_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11554_end_mask_0 = const()[name = string("op_11554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11554_cast_fp16 = slice_by_index(begin = var_11554_begin_0, end = var_11554_end_0, end_mask = var_11554_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11554_cast_fp16")]; + tensor var_11561_begin_0 = const()[name = string("op_11561_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11561_end_0 = const()[name = string("op_11561_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11561_end_mask_0 = const()[name = string("op_11561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11561_cast_fp16 = slice_by_index(begin = var_11561_begin_0, end = var_11561_end_0, end_mask = var_11561_end_mask_0, x = var_11023_cast_fp16)[name = string("op_11561_cast_fp16")]; + tensor var_11568_begin_0 = const()[name = string("op_11568_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11568_end_0 = const()[name = string("op_11568_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11568_end_mask_0 = const()[name = string("op_11568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11568_cast_fp16 = slice_by_index(begin = var_11568_begin_0, end = var_11568_end_0, end_mask = var_11568_end_mask_0, x = var_11027_cast_fp16)[name = string("op_11568_cast_fp16")]; + tensor var_11575_begin_0 = const()[name = string("op_11575_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11575_end_0 = const()[name = string("op_11575_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11575_end_mask_0 = const()[name = string("op_11575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11575_cast_fp16 = slice_by_index(begin = var_11575_begin_0, end = var_11575_end_0, end_mask = var_11575_end_mask_0, x = var_11027_cast_fp16)[name = string("op_11575_cast_fp16")]; + tensor var_11582_begin_0 = const()[name = string("op_11582_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11582_end_0 = const()[name = string("op_11582_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11582_end_mask_0 = const()[name = string("op_11582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11582_cast_fp16 = slice_by_index(begin = var_11582_begin_0, end = var_11582_end_0, end_mask = var_11582_end_mask_0, x = var_11027_cast_fp16)[name = string("op_11582_cast_fp16")]; + tensor var_11589_begin_0 = const()[name = string("op_11589_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11589_end_0 = const()[name = string("op_11589_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11589_end_mask_0 = const()[name = string("op_11589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11589_cast_fp16 = slice_by_index(begin = var_11589_begin_0, end = var_11589_end_0, end_mask = var_11589_end_mask_0, x = var_11027_cast_fp16)[name = string("op_11589_cast_fp16")]; + tensor k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11594_begin_0 = const()[name = string("op_11594_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11594_end_0 = const()[name = string("op_11594_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_11594_end_mask_0 = const()[name = string("op_11594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = string("transpose_24")]; + tensor var_11594_cast_fp16 = slice_by_index(begin = var_11594_begin_0, end = var_11594_end_0, end_mask = var_11594_end_mask_0, x = k_15_cast_fp16)[name = string("op_11594_cast_fp16")]; + tensor var_11598_begin_0 = const()[name = string("op_11598_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11598_end_0 = const()[name = string("op_11598_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_11598_end_mask_0 = const()[name = string("op_11598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11598_cast_fp16 = slice_by_index(begin = var_11598_begin_0, end = var_11598_end_0, end_mask = var_11598_end_mask_0, x = k_15_cast_fp16)[name = string("op_11598_cast_fp16")]; + tensor var_11602_begin_0 = const()[name = string("op_11602_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11602_end_0 = const()[name = string("op_11602_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_11602_end_mask_0 = const()[name = string("op_11602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11602_cast_fp16 = slice_by_index(begin = var_11602_begin_0, end = var_11602_end_0, end_mask = var_11602_end_mask_0, x = k_15_cast_fp16)[name = string("op_11602_cast_fp16")]; + tensor var_11606_begin_0 = const()[name = string("op_11606_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11606_end_0 = const()[name = string("op_11606_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_11606_end_mask_0 = const()[name = string("op_11606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11606_cast_fp16 = slice_by_index(begin = var_11606_begin_0, end = var_11606_end_0, end_mask = var_11606_end_mask_0, x = k_15_cast_fp16)[name = string("op_11606_cast_fp16")]; + tensor var_11610_begin_0 = const()[name = string("op_11610_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11610_end_0 = const()[name = string("op_11610_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_11610_end_mask_0 = const()[name = string("op_11610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11610_cast_fp16 = slice_by_index(begin = var_11610_begin_0, end = var_11610_end_0, end_mask = var_11610_end_mask_0, x = k_15_cast_fp16)[name = string("op_11610_cast_fp16")]; + tensor var_11614_begin_0 = const()[name = string("op_11614_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11614_end_0 = const()[name = string("op_11614_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_11614_end_mask_0 = const()[name = string("op_11614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11614_cast_fp16 = slice_by_index(begin = var_11614_begin_0, end = var_11614_end_0, end_mask = var_11614_end_mask_0, x = k_15_cast_fp16)[name = string("op_11614_cast_fp16")]; + tensor var_11618_begin_0 = const()[name = string("op_11618_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11618_end_0 = const()[name = string("op_11618_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_11618_end_mask_0 = const()[name = string("op_11618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11618_cast_fp16 = slice_by_index(begin = var_11618_begin_0, end = var_11618_end_0, end_mask = var_11618_end_mask_0, x = k_15_cast_fp16)[name = string("op_11618_cast_fp16")]; + tensor var_11622_begin_0 = const()[name = string("op_11622_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11622_end_0 = const()[name = string("op_11622_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_11622_end_mask_0 = const()[name = string("op_11622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11622_cast_fp16 = slice_by_index(begin = var_11622_begin_0, end = var_11622_end_0, end_mask = var_11622_end_mask_0, x = k_15_cast_fp16)[name = string("op_11622_cast_fp16")]; + tensor var_11626_begin_0 = const()[name = string("op_11626_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11626_end_0 = const()[name = string("op_11626_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_11626_end_mask_0 = const()[name = string("op_11626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11626_cast_fp16 = slice_by_index(begin = var_11626_begin_0, end = var_11626_end_0, end_mask = var_11626_end_mask_0, x = k_15_cast_fp16)[name = string("op_11626_cast_fp16")]; + tensor var_11630_begin_0 = const()[name = string("op_11630_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11630_end_0 = const()[name = string("op_11630_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_11630_end_mask_0 = const()[name = string("op_11630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11630_cast_fp16 = slice_by_index(begin = var_11630_begin_0, end = var_11630_end_0, end_mask = var_11630_end_mask_0, x = k_15_cast_fp16)[name = string("op_11630_cast_fp16")]; + tensor var_11634_begin_0 = const()[name = string("op_11634_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11634_end_0 = const()[name = string("op_11634_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_11634_end_mask_0 = const()[name = string("op_11634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11634_cast_fp16 = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = k_15_cast_fp16)[name = string("op_11634_cast_fp16")]; + tensor var_11638_begin_0 = const()[name = string("op_11638_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11638_end_0 = const()[name = string("op_11638_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_11638_end_mask_0 = const()[name = string("op_11638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11638_cast_fp16 = slice_by_index(begin = var_11638_begin_0, end = var_11638_end_0, end_mask = var_11638_end_mask_0, x = k_15_cast_fp16)[name = string("op_11638_cast_fp16")]; + tensor var_11642_begin_0 = const()[name = string("op_11642_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_11642_end_0 = const()[name = string("op_11642_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_11642_end_mask_0 = const()[name = string("op_11642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11642_cast_fp16 = slice_by_index(begin = var_11642_begin_0, end = var_11642_end_0, end_mask = var_11642_end_mask_0, x = k_15_cast_fp16)[name = string("op_11642_cast_fp16")]; + tensor var_11646_begin_0 = const()[name = string("op_11646_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_11646_end_0 = const()[name = string("op_11646_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_11646_end_mask_0 = const()[name = string("op_11646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11646_cast_fp16 = slice_by_index(begin = var_11646_begin_0, end = var_11646_end_0, end_mask = var_11646_end_mask_0, x = k_15_cast_fp16)[name = string("op_11646_cast_fp16")]; + tensor var_11650_begin_0 = const()[name = string("op_11650_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_11650_end_0 = const()[name = string("op_11650_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_11650_end_mask_0 = const()[name = string("op_11650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11650_cast_fp16 = slice_by_index(begin = var_11650_begin_0, end = var_11650_end_0, end_mask = var_11650_end_mask_0, x = k_15_cast_fp16)[name = string("op_11650_cast_fp16")]; + tensor var_11654_begin_0 = const()[name = string("op_11654_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_11654_end_0 = const()[name = string("op_11654_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_11654_end_mask_0 = const()[name = string("op_11654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11654_cast_fp16 = slice_by_index(begin = var_11654_begin_0, end = var_11654_end_0, end_mask = var_11654_end_mask_0, x = k_15_cast_fp16)[name = string("op_11654_cast_fp16")]; + tensor var_11658_begin_0 = const()[name = string("op_11658_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_11658_end_0 = const()[name = string("op_11658_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_11658_end_mask_0 = const()[name = string("op_11658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11658_cast_fp16 = slice_by_index(begin = var_11658_begin_0, end = var_11658_end_0, end_mask = var_11658_end_mask_0, x = k_15_cast_fp16)[name = string("op_11658_cast_fp16")]; + tensor var_11662_begin_0 = const()[name = string("op_11662_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_11662_end_0 = const()[name = string("op_11662_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_11662_end_mask_0 = const()[name = string("op_11662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11662_cast_fp16 = slice_by_index(begin = var_11662_begin_0, end = var_11662_end_0, end_mask = var_11662_end_mask_0, x = k_15_cast_fp16)[name = string("op_11662_cast_fp16")]; + tensor var_11666_begin_0 = const()[name = string("op_11666_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_11666_end_0 = const()[name = string("op_11666_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_11666_end_mask_0 = const()[name = string("op_11666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11666_cast_fp16 = slice_by_index(begin = var_11666_begin_0, end = var_11666_end_0, end_mask = var_11666_end_mask_0, x = k_15_cast_fp16)[name = string("op_11666_cast_fp16")]; + tensor var_11670_begin_0 = const()[name = string("op_11670_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_11670_end_0 = const()[name = string("op_11670_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_11670_end_mask_0 = const()[name = string("op_11670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11670_cast_fp16 = slice_by_index(begin = var_11670_begin_0, end = var_11670_end_0, end_mask = var_11670_end_mask_0, x = k_15_cast_fp16)[name = string("op_11670_cast_fp16")]; + tensor var_11672_begin_0 = const()[name = string("op_11672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11672_end_0 = const()[name = string("op_11672_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11672_end_mask_0 = const()[name = string("op_11672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11672_cast_fp16 = slice_by_index(begin = var_11672_begin_0, end = var_11672_end_0, end_mask = var_11672_end_mask_0, x = value_15_cast_fp16)[name = string("op_11672_cast_fp16")]; + tensor var_11676_begin_0 = const()[name = string("op_11676_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11676_end_0 = const()[name = string("op_11676_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_11676_end_mask_0 = const()[name = string("op_11676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11676_cast_fp16 = slice_by_index(begin = var_11676_begin_0, end = var_11676_end_0, end_mask = var_11676_end_mask_0, x = value_15_cast_fp16)[name = string("op_11676_cast_fp16")]; + tensor var_11680_begin_0 = const()[name = string("op_11680_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11680_end_0 = const()[name = string("op_11680_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_11680_end_mask_0 = const()[name = string("op_11680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11680_cast_fp16 = slice_by_index(begin = var_11680_begin_0, end = var_11680_end_0, end_mask = var_11680_end_mask_0, x = value_15_cast_fp16)[name = string("op_11680_cast_fp16")]; + tensor var_11684_begin_0 = const()[name = string("op_11684_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11684_end_0 = const()[name = string("op_11684_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_11684_end_mask_0 = const()[name = string("op_11684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11684_cast_fp16 = slice_by_index(begin = var_11684_begin_0, end = var_11684_end_0, end_mask = var_11684_end_mask_0, x = value_15_cast_fp16)[name = string("op_11684_cast_fp16")]; + tensor var_11688_begin_0 = const()[name = string("op_11688_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11688_end_0 = const()[name = string("op_11688_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_11688_end_mask_0 = const()[name = string("op_11688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11688_cast_fp16 = slice_by_index(begin = var_11688_begin_0, end = var_11688_end_0, end_mask = var_11688_end_mask_0, x = value_15_cast_fp16)[name = string("op_11688_cast_fp16")]; + tensor var_11692_begin_0 = const()[name = string("op_11692_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11692_end_0 = const()[name = string("op_11692_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_11692_end_mask_0 = const()[name = string("op_11692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11692_cast_fp16 = slice_by_index(begin = var_11692_begin_0, end = var_11692_end_0, end_mask = var_11692_end_mask_0, x = value_15_cast_fp16)[name = string("op_11692_cast_fp16")]; + tensor var_11696_begin_0 = const()[name = string("op_11696_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11696_end_0 = const()[name = string("op_11696_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_11696_end_mask_0 = const()[name = string("op_11696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11696_cast_fp16 = slice_by_index(begin = var_11696_begin_0, end = var_11696_end_0, end_mask = var_11696_end_mask_0, x = value_15_cast_fp16)[name = string("op_11696_cast_fp16")]; + tensor var_11700_begin_0 = const()[name = string("op_11700_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11700_end_0 = const()[name = string("op_11700_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_11700_end_mask_0 = const()[name = string("op_11700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11700_cast_fp16 = slice_by_index(begin = var_11700_begin_0, end = var_11700_end_0, end_mask = var_11700_end_mask_0, x = value_15_cast_fp16)[name = string("op_11700_cast_fp16")]; + tensor var_11704_begin_0 = const()[name = string("op_11704_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11704_end_0 = const()[name = string("op_11704_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_11704_end_mask_0 = const()[name = string("op_11704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11704_cast_fp16 = slice_by_index(begin = var_11704_begin_0, end = var_11704_end_0, end_mask = var_11704_end_mask_0, x = value_15_cast_fp16)[name = string("op_11704_cast_fp16")]; + tensor var_11708_begin_0 = const()[name = string("op_11708_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11708_end_0 = const()[name = string("op_11708_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_11708_end_mask_0 = const()[name = string("op_11708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11708_cast_fp16 = slice_by_index(begin = var_11708_begin_0, end = var_11708_end_0, end_mask = var_11708_end_mask_0, x = value_15_cast_fp16)[name = string("op_11708_cast_fp16")]; + tensor var_11712_begin_0 = const()[name = string("op_11712_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11712_end_0 = const()[name = string("op_11712_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_11712_end_mask_0 = const()[name = string("op_11712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11712_cast_fp16 = slice_by_index(begin = var_11712_begin_0, end = var_11712_end_0, end_mask = var_11712_end_mask_0, x = value_15_cast_fp16)[name = string("op_11712_cast_fp16")]; + tensor var_11716_begin_0 = const()[name = string("op_11716_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11716_end_0 = const()[name = string("op_11716_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_11716_end_mask_0 = const()[name = string("op_11716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11716_cast_fp16 = slice_by_index(begin = var_11716_begin_0, end = var_11716_end_0, end_mask = var_11716_end_mask_0, x = value_15_cast_fp16)[name = string("op_11716_cast_fp16")]; + tensor var_11720_begin_0 = const()[name = string("op_11720_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11720_end_0 = const()[name = string("op_11720_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_11720_end_mask_0 = const()[name = string("op_11720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11720_cast_fp16 = slice_by_index(begin = var_11720_begin_0, end = var_11720_end_0, end_mask = var_11720_end_mask_0, x = value_15_cast_fp16)[name = string("op_11720_cast_fp16")]; + tensor var_11724_begin_0 = const()[name = string("op_11724_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11724_end_0 = const()[name = string("op_11724_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_11724_end_mask_0 = const()[name = string("op_11724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11724_cast_fp16 = slice_by_index(begin = var_11724_begin_0, end = var_11724_end_0, end_mask = var_11724_end_mask_0, x = value_15_cast_fp16)[name = string("op_11724_cast_fp16")]; + tensor var_11728_begin_0 = const()[name = string("op_11728_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11728_end_0 = const()[name = string("op_11728_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_11728_end_mask_0 = const()[name = string("op_11728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11728_cast_fp16 = slice_by_index(begin = var_11728_begin_0, end = var_11728_end_0, end_mask = var_11728_end_mask_0, x = value_15_cast_fp16)[name = string("op_11728_cast_fp16")]; + tensor var_11732_begin_0 = const()[name = string("op_11732_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11732_end_0 = const()[name = string("op_11732_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_11732_end_mask_0 = const()[name = string("op_11732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11732_cast_fp16 = slice_by_index(begin = var_11732_begin_0, end = var_11732_end_0, end_mask = var_11732_end_mask_0, x = value_15_cast_fp16)[name = string("op_11732_cast_fp16")]; + tensor var_11736_begin_0 = const()[name = string("op_11736_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11736_end_0 = const()[name = string("op_11736_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_11736_end_mask_0 = const()[name = string("op_11736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11736_cast_fp16 = slice_by_index(begin = var_11736_begin_0, end = var_11736_end_0, end_mask = var_11736_end_mask_0, x = value_15_cast_fp16)[name = string("op_11736_cast_fp16")]; + tensor var_11740_begin_0 = const()[name = string("op_11740_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11740_end_0 = const()[name = string("op_11740_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_11740_end_mask_0 = const()[name = string("op_11740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11740_cast_fp16 = slice_by_index(begin = var_11740_begin_0, end = var_11740_end_0, end_mask = var_11740_end_mask_0, x = value_15_cast_fp16)[name = string("op_11740_cast_fp16")]; + tensor var_11744_begin_0 = const()[name = string("op_11744_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11744_end_0 = const()[name = string("op_11744_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_11744_end_mask_0 = const()[name = string("op_11744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11744_cast_fp16 = slice_by_index(begin = var_11744_begin_0, end = var_11744_end_0, end_mask = var_11744_end_mask_0, x = value_15_cast_fp16)[name = string("op_11744_cast_fp16")]; + tensor var_11748_begin_0 = const()[name = string("op_11748_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11748_end_0 = const()[name = string("op_11748_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_11748_end_mask_0 = const()[name = string("op_11748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11748_cast_fp16 = slice_by_index(begin = var_11748_begin_0, end = var_11748_end_0, end_mask = var_11748_end_mask_0, x = value_15_cast_fp16)[name = string("op_11748_cast_fp16")]; + string _SplitHeadsQ__mh_w_1121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1121_equation_0, values = (var_11594_cast_fp16, var_11036_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1121_cast_fp16")]; + string _SplitHeadsQ__mh_w_1123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1123_equation_0, values = (var_11594_cast_fp16, var_11043_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1123_cast_fp16")]; + string _SplitHeadsQ__mh_w_1125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1125_equation_0, values = (var_11594_cast_fp16, var_11050_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1125_cast_fp16")]; + string _SplitHeadsQ__mh_w_1127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1127_equation_0, values = (var_11594_cast_fp16, var_11057_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1127_cast_fp16")]; + string _SplitHeadsQ__mh_w_1129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1129_equation_0, values = (var_11598_cast_fp16, var_11064_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1129_cast_fp16")]; + string _SplitHeadsQ__mh_w_1131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1131_equation_0, values = (var_11598_cast_fp16, var_11071_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1131_cast_fp16")]; + string _SplitHeadsQ__mh_w_1133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1133_equation_0, values = (var_11598_cast_fp16, var_11078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1133_cast_fp16")]; + string _SplitHeadsQ__mh_w_1135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1135_equation_0, values = (var_11598_cast_fp16, var_11085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1135_cast_fp16")]; + string _SplitHeadsQ__mh_w_1137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1137_equation_0, values = (var_11602_cast_fp16, var_11092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1137_cast_fp16")]; + string _SplitHeadsQ__mh_w_1139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1139_equation_0, values = (var_11602_cast_fp16, var_11099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1139_cast_fp16")]; + string _SplitHeadsQ__mh_w_1141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1141_equation_0, values = (var_11602_cast_fp16, var_11106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1141_cast_fp16")]; + string _SplitHeadsQ__mh_w_1143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1143_equation_0, values = (var_11602_cast_fp16, var_11113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1143_cast_fp16")]; + string _SplitHeadsQ__mh_w_1145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1145_equation_0, values = (var_11606_cast_fp16, var_11120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1145_cast_fp16")]; + string _SplitHeadsQ__mh_w_1147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1147_equation_0, values = (var_11606_cast_fp16, var_11127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1147_cast_fp16")]; + string _SplitHeadsQ__mh_w_1149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1149_equation_0, values = (var_11606_cast_fp16, var_11134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1149_cast_fp16")]; + string _SplitHeadsQ__mh_w_1151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1151_equation_0, values = (var_11606_cast_fp16, var_11141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1151_cast_fp16")]; + string _SplitHeadsQ__mh_w_1153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1153_equation_0, values = (var_11610_cast_fp16, var_11148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1153_cast_fp16")]; + string _SplitHeadsQ__mh_w_1155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1155_equation_0, values = (var_11610_cast_fp16, var_11155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1155_cast_fp16")]; + string _SplitHeadsQ__mh_w_1157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1157_equation_0, values = (var_11610_cast_fp16, var_11162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1157_cast_fp16")]; + string _SplitHeadsQ__mh_w_1159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1159_equation_0, values = (var_11610_cast_fp16, var_11169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1159_cast_fp16")]; + string _SplitHeadsQ__mh_w_1161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1161_equation_0, values = (var_11614_cast_fp16, var_11176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1161_cast_fp16")]; + string _SplitHeadsQ__mh_w_1163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1163_equation_0, values = (var_11614_cast_fp16, var_11183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1163_cast_fp16")]; + string _SplitHeadsQ__mh_w_1165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1165_equation_0, values = (var_11614_cast_fp16, var_11190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1165_cast_fp16")]; + string _SplitHeadsQ__mh_w_1167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1167_equation_0, values = (var_11614_cast_fp16, var_11197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1167_cast_fp16")]; + string _SplitHeadsQ__mh_w_1169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1169_equation_0, values = (var_11618_cast_fp16, var_11204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1169_cast_fp16")]; + string _SplitHeadsQ__mh_w_1171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1171_equation_0, values = (var_11618_cast_fp16, var_11211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1171_cast_fp16")]; + string _SplitHeadsQ__mh_w_1173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1173_equation_0, values = (var_11618_cast_fp16, var_11218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1173_cast_fp16")]; + string _SplitHeadsQ__mh_w_1175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1175_equation_0, values = (var_11618_cast_fp16, var_11225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1175_cast_fp16")]; + string _SplitHeadsQ__mh_w_1177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1177_equation_0, values = (var_11622_cast_fp16, var_11232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1177_cast_fp16")]; + string _SplitHeadsQ__mh_w_1179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1179_equation_0, values = (var_11622_cast_fp16, var_11239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1179_cast_fp16")]; + string _SplitHeadsQ__mh_w_1181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1181_equation_0, values = (var_11622_cast_fp16, var_11246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1181_cast_fp16")]; + string _SplitHeadsQ__mh_w_1183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1183_equation_0, values = (var_11622_cast_fp16, var_11253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1183_cast_fp16")]; + string _SplitHeadsQ__mh_w_1185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1185_equation_0, values = (var_11626_cast_fp16, var_11260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1185_cast_fp16")]; + string _SplitHeadsQ__mh_w_1187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1187_equation_0, values = (var_11626_cast_fp16, var_11267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1187_cast_fp16")]; + string _SplitHeadsQ__mh_w_1189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1189_equation_0, values = (var_11626_cast_fp16, var_11274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1189_cast_fp16")]; + string _SplitHeadsQ__mh_w_1191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1191_equation_0, values = (var_11626_cast_fp16, var_11281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1191_cast_fp16")]; + string _SplitHeadsQ__mh_w_1193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1193_equation_0, values = (var_11630_cast_fp16, var_11288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1193_cast_fp16")]; + string _SplitHeadsQ__mh_w_1195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1195_equation_0, values = (var_11630_cast_fp16, var_11295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1195_cast_fp16")]; + string _SplitHeadsQ__mh_w_1197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1197_equation_0, values = (var_11630_cast_fp16, var_11302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1197_cast_fp16")]; + string _SplitHeadsQ__mh_w_1199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1199_equation_0, values = (var_11630_cast_fp16, var_11309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1199_cast_fp16")]; + string _SplitHeadsQ__mh_w_1201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1201_equation_0, values = (var_11634_cast_fp16, var_11316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1201_cast_fp16")]; + string _SplitHeadsQ__mh_w_1203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1203_equation_0, values = (var_11634_cast_fp16, var_11323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1203_cast_fp16")]; + string _SplitHeadsQ__mh_w_1205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1205_equation_0, values = (var_11634_cast_fp16, var_11330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1205_cast_fp16")]; + string _SplitHeadsQ__mh_w_1207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1207_equation_0, values = (var_11634_cast_fp16, var_11337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1207_cast_fp16")]; + string _SplitHeadsQ__mh_w_1209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1209_equation_0, values = (var_11638_cast_fp16, var_11344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1209_cast_fp16")]; + string _SplitHeadsQ__mh_w_1211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1211_equation_0, values = (var_11638_cast_fp16, var_11351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1211_cast_fp16")]; + string _SplitHeadsQ__mh_w_1213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1213_equation_0, values = (var_11638_cast_fp16, var_11358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1213_cast_fp16")]; + string _SplitHeadsQ__mh_w_1215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1215_equation_0, values = (var_11638_cast_fp16, var_11365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1215_cast_fp16")]; + string _SplitHeadsQ__mh_w_1217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1217_equation_0, values = (var_11642_cast_fp16, var_11372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1217_cast_fp16")]; + string _SplitHeadsQ__mh_w_1219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1219_equation_0, values = (var_11642_cast_fp16, var_11379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1219_cast_fp16")]; + string _SplitHeadsQ__mh_w_1221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1221_equation_0, values = (var_11642_cast_fp16, var_11386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1221_cast_fp16")]; + string _SplitHeadsQ__mh_w_1223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1223_equation_0, values = (var_11642_cast_fp16, var_11393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1223_cast_fp16")]; + string _SplitHeadsQ__mh_w_1225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1225_equation_0, values = (var_11646_cast_fp16, var_11400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1225_cast_fp16")]; + string _SplitHeadsQ__mh_w_1227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1227_equation_0, values = (var_11646_cast_fp16, var_11407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1227_cast_fp16")]; + string _SplitHeadsQ__mh_w_1229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1229_equation_0, values = (var_11646_cast_fp16, var_11414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1229_cast_fp16")]; + string _SplitHeadsQ__mh_w_1231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1231_equation_0, values = (var_11646_cast_fp16, var_11421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1231_cast_fp16")]; + string _SplitHeadsQ__mh_w_1233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1233_equation_0, values = (var_11650_cast_fp16, var_11428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1233_cast_fp16")]; + string _SplitHeadsQ__mh_w_1235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1235_equation_0, values = (var_11650_cast_fp16, var_11435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1235_cast_fp16")]; + string _SplitHeadsQ__mh_w_1237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1237_equation_0, values = (var_11650_cast_fp16, var_11442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1237_cast_fp16")]; + string _SplitHeadsQ__mh_w_1239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1239_equation_0, values = (var_11650_cast_fp16, var_11449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1239_cast_fp16")]; + string _SplitHeadsQ__mh_w_1241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1241_equation_0, values = (var_11654_cast_fp16, var_11456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1241_cast_fp16")]; + string _SplitHeadsQ__mh_w_1243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1243_equation_0, values = (var_11654_cast_fp16, var_11463_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1243_cast_fp16")]; + string _SplitHeadsQ__mh_w_1245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1245_equation_0, values = (var_11654_cast_fp16, var_11470_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1245_cast_fp16")]; + string _SplitHeadsQ__mh_w_1247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1247_equation_0, values = (var_11654_cast_fp16, var_11477_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1247_cast_fp16")]; + string _SplitHeadsQ__mh_w_1249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1249_equation_0, values = (var_11658_cast_fp16, var_11484_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1249_cast_fp16")]; + string _SplitHeadsQ__mh_w_1251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1251_equation_0, values = (var_11658_cast_fp16, var_11491_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1251_cast_fp16")]; + string _SplitHeadsQ__mh_w_1253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1253_equation_0, values = (var_11658_cast_fp16, var_11498_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1253_cast_fp16")]; + string _SplitHeadsQ__mh_w_1255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1255_equation_0, values = (var_11658_cast_fp16, var_11505_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1255_cast_fp16")]; + string _SplitHeadsQ__mh_w_1257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1257_equation_0, values = (var_11662_cast_fp16, var_11512_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1257_cast_fp16")]; + string _SplitHeadsQ__mh_w_1259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1259_equation_0, values = (var_11662_cast_fp16, var_11519_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1259_cast_fp16")]; + string _SplitHeadsQ__mh_w_1261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1261_equation_0, values = (var_11662_cast_fp16, var_11526_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1261_cast_fp16")]; + string _SplitHeadsQ__mh_w_1263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1263_equation_0, values = (var_11662_cast_fp16, var_11533_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1263_cast_fp16")]; + string _SplitHeadsQ__mh_w_1265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1265_equation_0, values = (var_11666_cast_fp16, var_11540_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1265_cast_fp16")]; + string _SplitHeadsQ__mh_w_1267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1267_equation_0, values = (var_11666_cast_fp16, var_11547_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1267_cast_fp16")]; + string _SplitHeadsQ__mh_w_1269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1269_equation_0, values = (var_11666_cast_fp16, var_11554_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1269_cast_fp16")]; + string _SplitHeadsQ__mh_w_1271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1271_equation_0, values = (var_11666_cast_fp16, var_11561_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1271_cast_fp16")]; + string _SplitHeadsQ__mh_w_1273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1273_equation_0, values = (var_11670_cast_fp16, var_11568_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1273_cast_fp16")]; + string _SplitHeadsQ__mh_w_1275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1275_equation_0, values = (var_11670_cast_fp16, var_11575_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1275_cast_fp16")]; + string _SplitHeadsQ__mh_w_1277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1277_equation_0, values = (var_11670_cast_fp16, var_11582_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1277_cast_fp16")]; + string _SplitHeadsQ__mh_w_1279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1279_equation_0, values = (var_11670_cast_fp16, var_11589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1279_cast_fp16")]; + fp16 var_11911_to_fp16 = const()[name = string("op_11911_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1121_cast_fp16, y = var_11911_to_fp16)[name = string("aw_chunk_1121_cast_fp16")]; + fp16 var_11913_to_fp16 = const()[name = string("op_11913_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1123_cast_fp16, y = var_11913_to_fp16)[name = string("aw_chunk_1123_cast_fp16")]; + fp16 var_11915_to_fp16 = const()[name = string("op_11915_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1125_cast_fp16, y = var_11915_to_fp16)[name = string("aw_chunk_1125_cast_fp16")]; + fp16 var_11917_to_fp16 = const()[name = string("op_11917_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1127_cast_fp16, y = var_11917_to_fp16)[name = string("aw_chunk_1127_cast_fp16")]; + fp16 var_11919_to_fp16 = const()[name = string("op_11919_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1129_cast_fp16, y = var_11919_to_fp16)[name = string("aw_chunk_1129_cast_fp16")]; + fp16 var_11921_to_fp16 = const()[name = string("op_11921_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1131_cast_fp16, y = var_11921_to_fp16)[name = string("aw_chunk_1131_cast_fp16")]; + fp16 var_11923_to_fp16 = const()[name = string("op_11923_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1133_cast_fp16, y = var_11923_to_fp16)[name = string("aw_chunk_1133_cast_fp16")]; + fp16 var_11925_to_fp16 = const()[name = string("op_11925_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1135_cast_fp16, y = var_11925_to_fp16)[name = string("aw_chunk_1135_cast_fp16")]; + fp16 var_11927_to_fp16 = const()[name = string("op_11927_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1137_cast_fp16, y = var_11927_to_fp16)[name = string("aw_chunk_1137_cast_fp16")]; + fp16 var_11929_to_fp16 = const()[name = string("op_11929_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1139_cast_fp16, y = var_11929_to_fp16)[name = string("aw_chunk_1139_cast_fp16")]; + fp16 var_11931_to_fp16 = const()[name = string("op_11931_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1141_cast_fp16, y = var_11931_to_fp16)[name = string("aw_chunk_1141_cast_fp16")]; + fp16 var_11933_to_fp16 = const()[name = string("op_11933_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1143_cast_fp16, y = var_11933_to_fp16)[name = string("aw_chunk_1143_cast_fp16")]; + fp16 var_11935_to_fp16 = const()[name = string("op_11935_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1145_cast_fp16, y = var_11935_to_fp16)[name = string("aw_chunk_1145_cast_fp16")]; + fp16 var_11937_to_fp16 = const()[name = string("op_11937_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1147_cast_fp16, y = var_11937_to_fp16)[name = string("aw_chunk_1147_cast_fp16")]; + fp16 var_11939_to_fp16 = const()[name = string("op_11939_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1149_cast_fp16, y = var_11939_to_fp16)[name = string("aw_chunk_1149_cast_fp16")]; + fp16 var_11941_to_fp16 = const()[name = string("op_11941_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1151_cast_fp16, y = var_11941_to_fp16)[name = string("aw_chunk_1151_cast_fp16")]; + fp16 var_11943_to_fp16 = const()[name = string("op_11943_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1153_cast_fp16, y = var_11943_to_fp16)[name = string("aw_chunk_1153_cast_fp16")]; + fp16 var_11945_to_fp16 = const()[name = string("op_11945_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1155_cast_fp16, y = var_11945_to_fp16)[name = string("aw_chunk_1155_cast_fp16")]; + fp16 var_11947_to_fp16 = const()[name = string("op_11947_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1157_cast_fp16, y = var_11947_to_fp16)[name = string("aw_chunk_1157_cast_fp16")]; + fp16 var_11949_to_fp16 = const()[name = string("op_11949_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1159_cast_fp16, y = var_11949_to_fp16)[name = string("aw_chunk_1159_cast_fp16")]; + fp16 var_11951_to_fp16 = const()[name = string("op_11951_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1161_cast_fp16, y = var_11951_to_fp16)[name = string("aw_chunk_1161_cast_fp16")]; + fp16 var_11953_to_fp16 = const()[name = string("op_11953_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1163_cast_fp16, y = var_11953_to_fp16)[name = string("aw_chunk_1163_cast_fp16")]; + fp16 var_11955_to_fp16 = const()[name = string("op_11955_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1165_cast_fp16, y = var_11955_to_fp16)[name = string("aw_chunk_1165_cast_fp16")]; + fp16 var_11957_to_fp16 = const()[name = string("op_11957_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1167_cast_fp16, y = var_11957_to_fp16)[name = string("aw_chunk_1167_cast_fp16")]; + fp16 var_11959_to_fp16 = const()[name = string("op_11959_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1169_cast_fp16, y = var_11959_to_fp16)[name = string("aw_chunk_1169_cast_fp16")]; + fp16 var_11961_to_fp16 = const()[name = string("op_11961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1171_cast_fp16, y = var_11961_to_fp16)[name = string("aw_chunk_1171_cast_fp16")]; + fp16 var_11963_to_fp16 = const()[name = string("op_11963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1173_cast_fp16, y = var_11963_to_fp16)[name = string("aw_chunk_1173_cast_fp16")]; + fp16 var_11965_to_fp16 = const()[name = string("op_11965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1175_cast_fp16, y = var_11965_to_fp16)[name = string("aw_chunk_1175_cast_fp16")]; + fp16 var_11967_to_fp16 = const()[name = string("op_11967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1177_cast_fp16, y = var_11967_to_fp16)[name = string("aw_chunk_1177_cast_fp16")]; + fp16 var_11969_to_fp16 = const()[name = string("op_11969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1179_cast_fp16, y = var_11969_to_fp16)[name = string("aw_chunk_1179_cast_fp16")]; + fp16 var_11971_to_fp16 = const()[name = string("op_11971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1181_cast_fp16, y = var_11971_to_fp16)[name = string("aw_chunk_1181_cast_fp16")]; + fp16 var_11973_to_fp16 = const()[name = string("op_11973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1183_cast_fp16, y = var_11973_to_fp16)[name = string("aw_chunk_1183_cast_fp16")]; + fp16 var_11975_to_fp16 = const()[name = string("op_11975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1185_cast_fp16, y = var_11975_to_fp16)[name = string("aw_chunk_1185_cast_fp16")]; + fp16 var_11977_to_fp16 = const()[name = string("op_11977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1187_cast_fp16, y = var_11977_to_fp16)[name = string("aw_chunk_1187_cast_fp16")]; + fp16 var_11979_to_fp16 = const()[name = string("op_11979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1189_cast_fp16, y = var_11979_to_fp16)[name = string("aw_chunk_1189_cast_fp16")]; + fp16 var_11981_to_fp16 = const()[name = string("op_11981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1191_cast_fp16, y = var_11981_to_fp16)[name = string("aw_chunk_1191_cast_fp16")]; + fp16 var_11983_to_fp16 = const()[name = string("op_11983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1193_cast_fp16, y = var_11983_to_fp16)[name = string("aw_chunk_1193_cast_fp16")]; + fp16 var_11985_to_fp16 = const()[name = string("op_11985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1195_cast_fp16, y = var_11985_to_fp16)[name = string("aw_chunk_1195_cast_fp16")]; + fp16 var_11987_to_fp16 = const()[name = string("op_11987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1197_cast_fp16, y = var_11987_to_fp16)[name = string("aw_chunk_1197_cast_fp16")]; + fp16 var_11989_to_fp16 = const()[name = string("op_11989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1199_cast_fp16, y = var_11989_to_fp16)[name = string("aw_chunk_1199_cast_fp16")]; + fp16 var_11991_to_fp16 = const()[name = string("op_11991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1201_cast_fp16, y = var_11991_to_fp16)[name = string("aw_chunk_1201_cast_fp16")]; + fp16 var_11993_to_fp16 = const()[name = string("op_11993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1203_cast_fp16, y = var_11993_to_fp16)[name = string("aw_chunk_1203_cast_fp16")]; + fp16 var_11995_to_fp16 = const()[name = string("op_11995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1205_cast_fp16, y = var_11995_to_fp16)[name = string("aw_chunk_1205_cast_fp16")]; + fp16 var_11997_to_fp16 = const()[name = string("op_11997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1207_cast_fp16, y = var_11997_to_fp16)[name = string("aw_chunk_1207_cast_fp16")]; + fp16 var_11999_to_fp16 = const()[name = string("op_11999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1209_cast_fp16, y = var_11999_to_fp16)[name = string("aw_chunk_1209_cast_fp16")]; + fp16 var_12001_to_fp16 = const()[name = string("op_12001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1211_cast_fp16, y = var_12001_to_fp16)[name = string("aw_chunk_1211_cast_fp16")]; + fp16 var_12003_to_fp16 = const()[name = string("op_12003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1213_cast_fp16, y = var_12003_to_fp16)[name = string("aw_chunk_1213_cast_fp16")]; + fp16 var_12005_to_fp16 = const()[name = string("op_12005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1215_cast_fp16, y = var_12005_to_fp16)[name = string("aw_chunk_1215_cast_fp16")]; + fp16 var_12007_to_fp16 = const()[name = string("op_12007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1217_cast_fp16, y = var_12007_to_fp16)[name = string("aw_chunk_1217_cast_fp16")]; + fp16 var_12009_to_fp16 = const()[name = string("op_12009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1219_cast_fp16, y = var_12009_to_fp16)[name = string("aw_chunk_1219_cast_fp16")]; + fp16 var_12011_to_fp16 = const()[name = string("op_12011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1221_cast_fp16, y = var_12011_to_fp16)[name = string("aw_chunk_1221_cast_fp16")]; + fp16 var_12013_to_fp16 = const()[name = string("op_12013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1223_cast_fp16, y = var_12013_to_fp16)[name = string("aw_chunk_1223_cast_fp16")]; + fp16 var_12015_to_fp16 = const()[name = string("op_12015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1225_cast_fp16, y = var_12015_to_fp16)[name = string("aw_chunk_1225_cast_fp16")]; + fp16 var_12017_to_fp16 = const()[name = string("op_12017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1227_cast_fp16, y = var_12017_to_fp16)[name = string("aw_chunk_1227_cast_fp16")]; + fp16 var_12019_to_fp16 = const()[name = string("op_12019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1229_cast_fp16, y = var_12019_to_fp16)[name = string("aw_chunk_1229_cast_fp16")]; + fp16 var_12021_to_fp16 = const()[name = string("op_12021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1231_cast_fp16, y = var_12021_to_fp16)[name = string("aw_chunk_1231_cast_fp16")]; + fp16 var_12023_to_fp16 = const()[name = string("op_12023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1233_cast_fp16, y = var_12023_to_fp16)[name = string("aw_chunk_1233_cast_fp16")]; + fp16 var_12025_to_fp16 = const()[name = string("op_12025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1235_cast_fp16, y = var_12025_to_fp16)[name = string("aw_chunk_1235_cast_fp16")]; + fp16 var_12027_to_fp16 = const()[name = string("op_12027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1237_cast_fp16, y = var_12027_to_fp16)[name = string("aw_chunk_1237_cast_fp16")]; + fp16 var_12029_to_fp16 = const()[name = string("op_12029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1239_cast_fp16, y = var_12029_to_fp16)[name = string("aw_chunk_1239_cast_fp16")]; + fp16 var_12031_to_fp16 = const()[name = string("op_12031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1241_cast_fp16, y = var_12031_to_fp16)[name = string("aw_chunk_1241_cast_fp16")]; + fp16 var_12033_to_fp16 = const()[name = string("op_12033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1243_cast_fp16, y = var_12033_to_fp16)[name = string("aw_chunk_1243_cast_fp16")]; + fp16 var_12035_to_fp16 = const()[name = string("op_12035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1245_cast_fp16, y = var_12035_to_fp16)[name = string("aw_chunk_1245_cast_fp16")]; + fp16 var_12037_to_fp16 = const()[name = string("op_12037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1247_cast_fp16, y = var_12037_to_fp16)[name = string("aw_chunk_1247_cast_fp16")]; + fp16 var_12039_to_fp16 = const()[name = string("op_12039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1249_cast_fp16, y = var_12039_to_fp16)[name = string("aw_chunk_1249_cast_fp16")]; + fp16 var_12041_to_fp16 = const()[name = string("op_12041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1251_cast_fp16, y = var_12041_to_fp16)[name = string("aw_chunk_1251_cast_fp16")]; + fp16 var_12043_to_fp16 = const()[name = string("op_12043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1253_cast_fp16, y = var_12043_to_fp16)[name = string("aw_chunk_1253_cast_fp16")]; + fp16 var_12045_to_fp16 = const()[name = string("op_12045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1255_cast_fp16, y = var_12045_to_fp16)[name = string("aw_chunk_1255_cast_fp16")]; + fp16 var_12047_to_fp16 = const()[name = string("op_12047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1257_cast_fp16, y = var_12047_to_fp16)[name = string("aw_chunk_1257_cast_fp16")]; + fp16 var_12049_to_fp16 = const()[name = string("op_12049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1259_cast_fp16, y = var_12049_to_fp16)[name = string("aw_chunk_1259_cast_fp16")]; + fp16 var_12051_to_fp16 = const()[name = string("op_12051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1261_cast_fp16, y = var_12051_to_fp16)[name = string("aw_chunk_1261_cast_fp16")]; + fp16 var_12053_to_fp16 = const()[name = string("op_12053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1263_cast_fp16, y = var_12053_to_fp16)[name = string("aw_chunk_1263_cast_fp16")]; + fp16 var_12055_to_fp16 = const()[name = string("op_12055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1265_cast_fp16, y = var_12055_to_fp16)[name = string("aw_chunk_1265_cast_fp16")]; + fp16 var_12057_to_fp16 = const()[name = string("op_12057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1267_cast_fp16, y = var_12057_to_fp16)[name = string("aw_chunk_1267_cast_fp16")]; + fp16 var_12059_to_fp16 = const()[name = string("op_12059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1269_cast_fp16, y = var_12059_to_fp16)[name = string("aw_chunk_1269_cast_fp16")]; + fp16 var_12061_to_fp16 = const()[name = string("op_12061_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1271_cast_fp16, y = var_12061_to_fp16)[name = string("aw_chunk_1271_cast_fp16")]; + fp16 var_12063_to_fp16 = const()[name = string("op_12063_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1273_cast_fp16, y = var_12063_to_fp16)[name = string("aw_chunk_1273_cast_fp16")]; + fp16 var_12065_to_fp16 = const()[name = string("op_12065_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1275_cast_fp16, y = var_12065_to_fp16)[name = string("aw_chunk_1275_cast_fp16")]; + fp16 var_12067_to_fp16 = const()[name = string("op_12067_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1277_cast_fp16, y = var_12067_to_fp16)[name = string("aw_chunk_1277_cast_fp16")]; + fp16 var_12069_to_fp16 = const()[name = string("op_12069_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1279_cast_fp16, y = var_12069_to_fp16)[name = string("aw_chunk_1279_cast_fp16")]; + tensor var_12071_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1121_cast_fp16)[name = string("op_12071_cast_fp16")]; + tensor var_12072_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1123_cast_fp16)[name = string("op_12072_cast_fp16")]; + tensor var_12073_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1125_cast_fp16)[name = string("op_12073_cast_fp16")]; + tensor var_12074_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1127_cast_fp16)[name = string("op_12074_cast_fp16")]; + tensor var_12075_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1129_cast_fp16)[name = string("op_12075_cast_fp16")]; + tensor var_12076_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1131_cast_fp16)[name = string("op_12076_cast_fp16")]; + tensor var_12077_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1133_cast_fp16)[name = string("op_12077_cast_fp16")]; + tensor var_12078_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1135_cast_fp16)[name = string("op_12078_cast_fp16")]; + tensor var_12079_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1137_cast_fp16)[name = string("op_12079_cast_fp16")]; + tensor var_12080_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1139_cast_fp16)[name = string("op_12080_cast_fp16")]; + tensor var_12081_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1141_cast_fp16)[name = string("op_12081_cast_fp16")]; + tensor var_12082_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1143_cast_fp16)[name = string("op_12082_cast_fp16")]; + tensor var_12083_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1145_cast_fp16)[name = string("op_12083_cast_fp16")]; + tensor var_12084_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1147_cast_fp16)[name = string("op_12084_cast_fp16")]; + tensor var_12085_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1149_cast_fp16)[name = string("op_12085_cast_fp16")]; + tensor var_12086_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1151_cast_fp16)[name = string("op_12086_cast_fp16")]; + tensor var_12087_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1153_cast_fp16)[name = string("op_12087_cast_fp16")]; + tensor var_12088_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1155_cast_fp16)[name = string("op_12088_cast_fp16")]; + tensor var_12089_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1157_cast_fp16)[name = string("op_12089_cast_fp16")]; + tensor var_12090_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1159_cast_fp16)[name = string("op_12090_cast_fp16")]; + tensor var_12091_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1161_cast_fp16)[name = string("op_12091_cast_fp16")]; + tensor var_12092_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1163_cast_fp16)[name = string("op_12092_cast_fp16")]; + tensor var_12093_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1165_cast_fp16)[name = string("op_12093_cast_fp16")]; + tensor var_12094_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1167_cast_fp16)[name = string("op_12094_cast_fp16")]; + tensor var_12095_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1169_cast_fp16)[name = string("op_12095_cast_fp16")]; + tensor var_12096_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1171_cast_fp16)[name = string("op_12096_cast_fp16")]; + tensor var_12097_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1173_cast_fp16)[name = string("op_12097_cast_fp16")]; + tensor var_12098_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1175_cast_fp16)[name = string("op_12098_cast_fp16")]; + tensor var_12099_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1177_cast_fp16)[name = string("op_12099_cast_fp16")]; + tensor var_12100_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1179_cast_fp16)[name = string("op_12100_cast_fp16")]; + tensor var_12101_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1181_cast_fp16)[name = string("op_12101_cast_fp16")]; + tensor var_12102_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1183_cast_fp16)[name = string("op_12102_cast_fp16")]; + tensor var_12103_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1185_cast_fp16)[name = string("op_12103_cast_fp16")]; + tensor var_12104_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1187_cast_fp16)[name = string("op_12104_cast_fp16")]; + tensor var_12105_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1189_cast_fp16)[name = string("op_12105_cast_fp16")]; + tensor var_12106_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1191_cast_fp16)[name = string("op_12106_cast_fp16")]; + tensor var_12107_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1193_cast_fp16)[name = string("op_12107_cast_fp16")]; + tensor var_12108_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1195_cast_fp16)[name = string("op_12108_cast_fp16")]; + tensor var_12109_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1197_cast_fp16)[name = string("op_12109_cast_fp16")]; + tensor var_12110_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1199_cast_fp16)[name = string("op_12110_cast_fp16")]; + tensor var_12111_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1201_cast_fp16)[name = string("op_12111_cast_fp16")]; + tensor var_12112_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1203_cast_fp16)[name = string("op_12112_cast_fp16")]; + tensor var_12113_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1205_cast_fp16)[name = string("op_12113_cast_fp16")]; + tensor var_12114_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1207_cast_fp16)[name = string("op_12114_cast_fp16")]; + tensor var_12115_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1209_cast_fp16)[name = string("op_12115_cast_fp16")]; + tensor var_12116_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1211_cast_fp16)[name = string("op_12116_cast_fp16")]; + tensor var_12117_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1213_cast_fp16)[name = string("op_12117_cast_fp16")]; + tensor var_12118_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1215_cast_fp16)[name = string("op_12118_cast_fp16")]; + tensor var_12119_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1217_cast_fp16)[name = string("op_12119_cast_fp16")]; + tensor var_12120_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1219_cast_fp16)[name = string("op_12120_cast_fp16")]; + tensor var_12121_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1221_cast_fp16)[name = string("op_12121_cast_fp16")]; + tensor var_12122_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1223_cast_fp16)[name = string("op_12122_cast_fp16")]; + tensor var_12123_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1225_cast_fp16)[name = string("op_12123_cast_fp16")]; + tensor var_12124_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1227_cast_fp16)[name = string("op_12124_cast_fp16")]; + tensor var_12125_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1229_cast_fp16)[name = string("op_12125_cast_fp16")]; + tensor var_12126_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1231_cast_fp16)[name = string("op_12126_cast_fp16")]; + tensor var_12127_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1233_cast_fp16)[name = string("op_12127_cast_fp16")]; + tensor var_12128_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1235_cast_fp16)[name = string("op_12128_cast_fp16")]; + tensor var_12129_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1237_cast_fp16)[name = string("op_12129_cast_fp16")]; + tensor var_12130_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1239_cast_fp16)[name = string("op_12130_cast_fp16")]; + tensor var_12131_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1241_cast_fp16)[name = string("op_12131_cast_fp16")]; + tensor var_12132_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1243_cast_fp16)[name = string("op_12132_cast_fp16")]; + tensor var_12133_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1245_cast_fp16)[name = string("op_12133_cast_fp16")]; + tensor var_12134_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1247_cast_fp16)[name = string("op_12134_cast_fp16")]; + tensor var_12135_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1249_cast_fp16)[name = string("op_12135_cast_fp16")]; + tensor var_12136_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1251_cast_fp16)[name = string("op_12136_cast_fp16")]; + tensor var_12137_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1253_cast_fp16)[name = string("op_12137_cast_fp16")]; + tensor var_12138_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1255_cast_fp16)[name = string("op_12138_cast_fp16")]; + tensor var_12139_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1257_cast_fp16)[name = string("op_12139_cast_fp16")]; + tensor var_12140_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1259_cast_fp16)[name = string("op_12140_cast_fp16")]; + tensor var_12141_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1261_cast_fp16)[name = string("op_12141_cast_fp16")]; + tensor var_12142_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1263_cast_fp16)[name = string("op_12142_cast_fp16")]; + tensor var_12143_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1265_cast_fp16)[name = string("op_12143_cast_fp16")]; + tensor var_12144_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1267_cast_fp16)[name = string("op_12144_cast_fp16")]; + tensor var_12145_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1269_cast_fp16)[name = string("op_12145_cast_fp16")]; + tensor var_12146_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1271_cast_fp16)[name = string("op_12146_cast_fp16")]; + tensor var_12147_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1273_cast_fp16)[name = string("op_12147_cast_fp16")]; + tensor var_12148_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1275_cast_fp16)[name = string("op_12148_cast_fp16")]; + tensor var_12149_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1277_cast_fp16)[name = string("op_12149_cast_fp16")]; + tensor var_12150_cast_fp16 = softmax(axis = var_10896, x = aw_chunk_1279_cast_fp16)[name = string("op_12150_cast_fp16")]; + string var_12152_equation_0 = const()[name = string("op_12152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12152_cast_fp16 = einsum(equation = var_12152_equation_0, values = (var_11672_cast_fp16, var_12071_cast_fp16))[name = string("op_12152_cast_fp16")]; + string var_12154_equation_0 = const()[name = string("op_12154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12154_cast_fp16 = einsum(equation = var_12154_equation_0, values = (var_11672_cast_fp16, var_12072_cast_fp16))[name = string("op_12154_cast_fp16")]; + string var_12156_equation_0 = const()[name = string("op_12156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12156_cast_fp16 = einsum(equation = var_12156_equation_0, values = (var_11672_cast_fp16, var_12073_cast_fp16))[name = string("op_12156_cast_fp16")]; + string var_12158_equation_0 = const()[name = string("op_12158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12158_cast_fp16 = einsum(equation = var_12158_equation_0, values = (var_11672_cast_fp16, var_12074_cast_fp16))[name = string("op_12158_cast_fp16")]; + string var_12160_equation_0 = const()[name = string("op_12160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12160_cast_fp16 = einsum(equation = var_12160_equation_0, values = (var_11676_cast_fp16, var_12075_cast_fp16))[name = string("op_12160_cast_fp16")]; + string var_12162_equation_0 = const()[name = string("op_12162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12162_cast_fp16 = einsum(equation = var_12162_equation_0, values = (var_11676_cast_fp16, var_12076_cast_fp16))[name = string("op_12162_cast_fp16")]; + string var_12164_equation_0 = const()[name = string("op_12164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12164_cast_fp16 = einsum(equation = var_12164_equation_0, values = (var_11676_cast_fp16, var_12077_cast_fp16))[name = string("op_12164_cast_fp16")]; + string var_12166_equation_0 = const()[name = string("op_12166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12166_cast_fp16 = einsum(equation = var_12166_equation_0, values = (var_11676_cast_fp16, var_12078_cast_fp16))[name = string("op_12166_cast_fp16")]; + string var_12168_equation_0 = const()[name = string("op_12168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12168_cast_fp16 = einsum(equation = var_12168_equation_0, values = (var_11680_cast_fp16, var_12079_cast_fp16))[name = string("op_12168_cast_fp16")]; + string var_12170_equation_0 = const()[name = string("op_12170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12170_cast_fp16 = einsum(equation = var_12170_equation_0, values = (var_11680_cast_fp16, var_12080_cast_fp16))[name = string("op_12170_cast_fp16")]; + string var_12172_equation_0 = const()[name = string("op_12172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12172_cast_fp16 = einsum(equation = var_12172_equation_0, values = (var_11680_cast_fp16, var_12081_cast_fp16))[name = string("op_12172_cast_fp16")]; + string var_12174_equation_0 = const()[name = string("op_12174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12174_cast_fp16 = einsum(equation = var_12174_equation_0, values = (var_11680_cast_fp16, var_12082_cast_fp16))[name = string("op_12174_cast_fp16")]; + string var_12176_equation_0 = const()[name = string("op_12176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12176_cast_fp16 = einsum(equation = var_12176_equation_0, values = (var_11684_cast_fp16, var_12083_cast_fp16))[name = string("op_12176_cast_fp16")]; + string var_12178_equation_0 = const()[name = string("op_12178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12178_cast_fp16 = einsum(equation = var_12178_equation_0, values = (var_11684_cast_fp16, var_12084_cast_fp16))[name = string("op_12178_cast_fp16")]; + string var_12180_equation_0 = const()[name = string("op_12180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12180_cast_fp16 = einsum(equation = var_12180_equation_0, values = (var_11684_cast_fp16, var_12085_cast_fp16))[name = string("op_12180_cast_fp16")]; + string var_12182_equation_0 = const()[name = string("op_12182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12182_cast_fp16 = einsum(equation = var_12182_equation_0, values = (var_11684_cast_fp16, var_12086_cast_fp16))[name = string("op_12182_cast_fp16")]; + string var_12184_equation_0 = const()[name = string("op_12184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12184_cast_fp16 = einsum(equation = var_12184_equation_0, values = (var_11688_cast_fp16, var_12087_cast_fp16))[name = string("op_12184_cast_fp16")]; + string var_12186_equation_0 = const()[name = string("op_12186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12186_cast_fp16 = einsum(equation = var_12186_equation_0, values = (var_11688_cast_fp16, var_12088_cast_fp16))[name = string("op_12186_cast_fp16")]; + string var_12188_equation_0 = const()[name = string("op_12188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12188_cast_fp16 = einsum(equation = var_12188_equation_0, values = (var_11688_cast_fp16, var_12089_cast_fp16))[name = string("op_12188_cast_fp16")]; + string var_12190_equation_0 = const()[name = string("op_12190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12190_cast_fp16 = einsum(equation = var_12190_equation_0, values = (var_11688_cast_fp16, var_12090_cast_fp16))[name = string("op_12190_cast_fp16")]; + string var_12192_equation_0 = const()[name = string("op_12192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12192_cast_fp16 = einsum(equation = var_12192_equation_0, values = (var_11692_cast_fp16, var_12091_cast_fp16))[name = string("op_12192_cast_fp16")]; + string var_12194_equation_0 = const()[name = string("op_12194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12194_cast_fp16 = einsum(equation = var_12194_equation_0, values = (var_11692_cast_fp16, var_12092_cast_fp16))[name = string("op_12194_cast_fp16")]; + string var_12196_equation_0 = const()[name = string("op_12196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12196_cast_fp16 = einsum(equation = var_12196_equation_0, values = (var_11692_cast_fp16, var_12093_cast_fp16))[name = string("op_12196_cast_fp16")]; + string var_12198_equation_0 = const()[name = string("op_12198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12198_cast_fp16 = einsum(equation = var_12198_equation_0, values = (var_11692_cast_fp16, var_12094_cast_fp16))[name = string("op_12198_cast_fp16")]; + string var_12200_equation_0 = const()[name = string("op_12200_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12200_cast_fp16 = einsum(equation = var_12200_equation_0, values = (var_11696_cast_fp16, var_12095_cast_fp16))[name = string("op_12200_cast_fp16")]; + string var_12202_equation_0 = const()[name = string("op_12202_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12202_cast_fp16 = einsum(equation = var_12202_equation_0, values = (var_11696_cast_fp16, var_12096_cast_fp16))[name = string("op_12202_cast_fp16")]; + string var_12204_equation_0 = const()[name = string("op_12204_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12204_cast_fp16 = einsum(equation = var_12204_equation_0, values = (var_11696_cast_fp16, var_12097_cast_fp16))[name = string("op_12204_cast_fp16")]; + string var_12206_equation_0 = const()[name = string("op_12206_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12206_cast_fp16 = einsum(equation = var_12206_equation_0, values = (var_11696_cast_fp16, var_12098_cast_fp16))[name = string("op_12206_cast_fp16")]; + string var_12208_equation_0 = const()[name = string("op_12208_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12208_cast_fp16 = einsum(equation = var_12208_equation_0, values = (var_11700_cast_fp16, var_12099_cast_fp16))[name = string("op_12208_cast_fp16")]; + string var_12210_equation_0 = const()[name = string("op_12210_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12210_cast_fp16 = einsum(equation = var_12210_equation_0, values = (var_11700_cast_fp16, var_12100_cast_fp16))[name = string("op_12210_cast_fp16")]; + string var_12212_equation_0 = const()[name = string("op_12212_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12212_cast_fp16 = einsum(equation = var_12212_equation_0, values = (var_11700_cast_fp16, var_12101_cast_fp16))[name = string("op_12212_cast_fp16")]; + string var_12214_equation_0 = const()[name = string("op_12214_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12214_cast_fp16 = einsum(equation = var_12214_equation_0, values = (var_11700_cast_fp16, var_12102_cast_fp16))[name = string("op_12214_cast_fp16")]; + string var_12216_equation_0 = const()[name = string("op_12216_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12216_cast_fp16 = einsum(equation = var_12216_equation_0, values = (var_11704_cast_fp16, var_12103_cast_fp16))[name = string("op_12216_cast_fp16")]; + string var_12218_equation_0 = const()[name = string("op_12218_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12218_cast_fp16 = einsum(equation = var_12218_equation_0, values = (var_11704_cast_fp16, var_12104_cast_fp16))[name = string("op_12218_cast_fp16")]; + string var_12220_equation_0 = const()[name = string("op_12220_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12220_cast_fp16 = einsum(equation = var_12220_equation_0, values = (var_11704_cast_fp16, var_12105_cast_fp16))[name = string("op_12220_cast_fp16")]; + string var_12222_equation_0 = const()[name = string("op_12222_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12222_cast_fp16 = einsum(equation = var_12222_equation_0, values = (var_11704_cast_fp16, var_12106_cast_fp16))[name = string("op_12222_cast_fp16")]; + string var_12224_equation_0 = const()[name = string("op_12224_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12224_cast_fp16 = einsum(equation = var_12224_equation_0, values = (var_11708_cast_fp16, var_12107_cast_fp16))[name = string("op_12224_cast_fp16")]; + string var_12226_equation_0 = const()[name = string("op_12226_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12226_cast_fp16 = einsum(equation = var_12226_equation_0, values = (var_11708_cast_fp16, var_12108_cast_fp16))[name = string("op_12226_cast_fp16")]; + string var_12228_equation_0 = const()[name = string("op_12228_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12228_cast_fp16 = einsum(equation = var_12228_equation_0, values = (var_11708_cast_fp16, var_12109_cast_fp16))[name = string("op_12228_cast_fp16")]; + string var_12230_equation_0 = const()[name = string("op_12230_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12230_cast_fp16 = einsum(equation = var_12230_equation_0, values = (var_11708_cast_fp16, var_12110_cast_fp16))[name = string("op_12230_cast_fp16")]; + string var_12232_equation_0 = const()[name = string("op_12232_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12232_cast_fp16 = einsum(equation = var_12232_equation_0, values = (var_11712_cast_fp16, var_12111_cast_fp16))[name = string("op_12232_cast_fp16")]; + string var_12234_equation_0 = const()[name = string("op_12234_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12234_cast_fp16 = einsum(equation = var_12234_equation_0, values = (var_11712_cast_fp16, var_12112_cast_fp16))[name = string("op_12234_cast_fp16")]; + string var_12236_equation_0 = const()[name = string("op_12236_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12236_cast_fp16 = einsum(equation = var_12236_equation_0, values = (var_11712_cast_fp16, var_12113_cast_fp16))[name = string("op_12236_cast_fp16")]; + string var_12238_equation_0 = const()[name = string("op_12238_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12238_cast_fp16 = einsum(equation = var_12238_equation_0, values = (var_11712_cast_fp16, var_12114_cast_fp16))[name = string("op_12238_cast_fp16")]; + string var_12240_equation_0 = const()[name = string("op_12240_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12240_cast_fp16 = einsum(equation = var_12240_equation_0, values = (var_11716_cast_fp16, var_12115_cast_fp16))[name = string("op_12240_cast_fp16")]; + string var_12242_equation_0 = const()[name = string("op_12242_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12242_cast_fp16 = einsum(equation = var_12242_equation_0, values = (var_11716_cast_fp16, var_12116_cast_fp16))[name = string("op_12242_cast_fp16")]; + string var_12244_equation_0 = const()[name = string("op_12244_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12244_cast_fp16 = einsum(equation = var_12244_equation_0, values = (var_11716_cast_fp16, var_12117_cast_fp16))[name = string("op_12244_cast_fp16")]; + string var_12246_equation_0 = const()[name = string("op_12246_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12246_cast_fp16 = einsum(equation = var_12246_equation_0, values = (var_11716_cast_fp16, var_12118_cast_fp16))[name = string("op_12246_cast_fp16")]; + string var_12248_equation_0 = const()[name = string("op_12248_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12248_cast_fp16 = einsum(equation = var_12248_equation_0, values = (var_11720_cast_fp16, var_12119_cast_fp16))[name = string("op_12248_cast_fp16")]; + string var_12250_equation_0 = const()[name = string("op_12250_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12250_cast_fp16 = einsum(equation = var_12250_equation_0, values = (var_11720_cast_fp16, var_12120_cast_fp16))[name = string("op_12250_cast_fp16")]; + string var_12252_equation_0 = const()[name = string("op_12252_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12252_cast_fp16 = einsum(equation = var_12252_equation_0, values = (var_11720_cast_fp16, var_12121_cast_fp16))[name = string("op_12252_cast_fp16")]; + string var_12254_equation_0 = const()[name = string("op_12254_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12254_cast_fp16 = einsum(equation = var_12254_equation_0, values = (var_11720_cast_fp16, var_12122_cast_fp16))[name = string("op_12254_cast_fp16")]; + string var_12256_equation_0 = const()[name = string("op_12256_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12256_cast_fp16 = einsum(equation = var_12256_equation_0, values = (var_11724_cast_fp16, var_12123_cast_fp16))[name = string("op_12256_cast_fp16")]; + string var_12258_equation_0 = const()[name = string("op_12258_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12258_cast_fp16 = einsum(equation = var_12258_equation_0, values = (var_11724_cast_fp16, var_12124_cast_fp16))[name = string("op_12258_cast_fp16")]; + string var_12260_equation_0 = const()[name = string("op_12260_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12260_cast_fp16 = einsum(equation = var_12260_equation_0, values = (var_11724_cast_fp16, var_12125_cast_fp16))[name = string("op_12260_cast_fp16")]; + string var_12262_equation_0 = const()[name = string("op_12262_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12262_cast_fp16 = einsum(equation = var_12262_equation_0, values = (var_11724_cast_fp16, var_12126_cast_fp16))[name = string("op_12262_cast_fp16")]; + string var_12264_equation_0 = const()[name = string("op_12264_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12264_cast_fp16 = einsum(equation = var_12264_equation_0, values = (var_11728_cast_fp16, var_12127_cast_fp16))[name = string("op_12264_cast_fp16")]; + string var_12266_equation_0 = const()[name = string("op_12266_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12266_cast_fp16 = einsum(equation = var_12266_equation_0, values = (var_11728_cast_fp16, var_12128_cast_fp16))[name = string("op_12266_cast_fp16")]; + string var_12268_equation_0 = const()[name = string("op_12268_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12268_cast_fp16 = einsum(equation = var_12268_equation_0, values = (var_11728_cast_fp16, var_12129_cast_fp16))[name = string("op_12268_cast_fp16")]; + string var_12270_equation_0 = const()[name = string("op_12270_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12270_cast_fp16 = einsum(equation = var_12270_equation_0, values = (var_11728_cast_fp16, var_12130_cast_fp16))[name = string("op_12270_cast_fp16")]; + string var_12272_equation_0 = const()[name = string("op_12272_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12272_cast_fp16 = einsum(equation = var_12272_equation_0, values = (var_11732_cast_fp16, var_12131_cast_fp16))[name = string("op_12272_cast_fp16")]; + string var_12274_equation_0 = const()[name = string("op_12274_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12274_cast_fp16 = einsum(equation = var_12274_equation_0, values = (var_11732_cast_fp16, var_12132_cast_fp16))[name = string("op_12274_cast_fp16")]; + string var_12276_equation_0 = const()[name = string("op_12276_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12276_cast_fp16 = einsum(equation = var_12276_equation_0, values = (var_11732_cast_fp16, var_12133_cast_fp16))[name = string("op_12276_cast_fp16")]; + string var_12278_equation_0 = const()[name = string("op_12278_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12278_cast_fp16 = einsum(equation = var_12278_equation_0, values = (var_11732_cast_fp16, var_12134_cast_fp16))[name = string("op_12278_cast_fp16")]; + string var_12280_equation_0 = const()[name = string("op_12280_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12280_cast_fp16 = einsum(equation = var_12280_equation_0, values = (var_11736_cast_fp16, var_12135_cast_fp16))[name = string("op_12280_cast_fp16")]; + string var_12282_equation_0 = const()[name = string("op_12282_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12282_cast_fp16 = einsum(equation = var_12282_equation_0, values = (var_11736_cast_fp16, var_12136_cast_fp16))[name = string("op_12282_cast_fp16")]; + string var_12284_equation_0 = const()[name = string("op_12284_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12284_cast_fp16 = einsum(equation = var_12284_equation_0, values = (var_11736_cast_fp16, var_12137_cast_fp16))[name = string("op_12284_cast_fp16")]; + string var_12286_equation_0 = const()[name = string("op_12286_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12286_cast_fp16 = einsum(equation = var_12286_equation_0, values = (var_11736_cast_fp16, var_12138_cast_fp16))[name = string("op_12286_cast_fp16")]; + string var_12288_equation_0 = const()[name = string("op_12288_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12288_cast_fp16 = einsum(equation = var_12288_equation_0, values = (var_11740_cast_fp16, var_12139_cast_fp16))[name = string("op_12288_cast_fp16")]; + string var_12290_equation_0 = const()[name = string("op_12290_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12290_cast_fp16 = einsum(equation = var_12290_equation_0, values = (var_11740_cast_fp16, var_12140_cast_fp16))[name = string("op_12290_cast_fp16")]; + string var_12292_equation_0 = const()[name = string("op_12292_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12292_cast_fp16 = einsum(equation = var_12292_equation_0, values = (var_11740_cast_fp16, var_12141_cast_fp16))[name = string("op_12292_cast_fp16")]; + string var_12294_equation_0 = const()[name = string("op_12294_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12294_cast_fp16 = einsum(equation = var_12294_equation_0, values = (var_11740_cast_fp16, var_12142_cast_fp16))[name = string("op_12294_cast_fp16")]; + string var_12296_equation_0 = const()[name = string("op_12296_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12296_cast_fp16 = einsum(equation = var_12296_equation_0, values = (var_11744_cast_fp16, var_12143_cast_fp16))[name = string("op_12296_cast_fp16")]; + string var_12298_equation_0 = const()[name = string("op_12298_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12298_cast_fp16 = einsum(equation = var_12298_equation_0, values = (var_11744_cast_fp16, var_12144_cast_fp16))[name = string("op_12298_cast_fp16")]; + string var_12300_equation_0 = const()[name = string("op_12300_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12300_cast_fp16 = einsum(equation = var_12300_equation_0, values = (var_11744_cast_fp16, var_12145_cast_fp16))[name = string("op_12300_cast_fp16")]; + string var_12302_equation_0 = const()[name = string("op_12302_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12302_cast_fp16 = einsum(equation = var_12302_equation_0, values = (var_11744_cast_fp16, var_12146_cast_fp16))[name = string("op_12302_cast_fp16")]; + string var_12304_equation_0 = const()[name = string("op_12304_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12304_cast_fp16 = einsum(equation = var_12304_equation_0, values = (var_11748_cast_fp16, var_12147_cast_fp16))[name = string("op_12304_cast_fp16")]; + string var_12306_equation_0 = const()[name = string("op_12306_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12306_cast_fp16 = einsum(equation = var_12306_equation_0, values = (var_11748_cast_fp16, var_12148_cast_fp16))[name = string("op_12306_cast_fp16")]; + string var_12308_equation_0 = const()[name = string("op_12308_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12308_cast_fp16 = einsum(equation = var_12308_equation_0, values = (var_11748_cast_fp16, var_12149_cast_fp16))[name = string("op_12308_cast_fp16")]; + string var_12310_equation_0 = const()[name = string("op_12310_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_12310_cast_fp16 = einsum(equation = var_12310_equation_0, values = (var_11748_cast_fp16, var_12150_cast_fp16))[name = string("op_12310_cast_fp16")]; + bool var_12312_interleave_0 = const()[name = string("op_12312_interleave_0"), val = bool(false)]; + tensor var_12312_cast_fp16 = concat(axis = var_10871, interleave = var_12312_interleave_0, values = (var_12152_cast_fp16, var_12154_cast_fp16, var_12156_cast_fp16, var_12158_cast_fp16))[name = string("op_12312_cast_fp16")]; + bool var_12314_interleave_0 = const()[name = string("op_12314_interleave_0"), val = bool(false)]; + tensor var_12314_cast_fp16 = concat(axis = var_10871, interleave = var_12314_interleave_0, values = (var_12160_cast_fp16, var_12162_cast_fp16, var_12164_cast_fp16, var_12166_cast_fp16))[name = string("op_12314_cast_fp16")]; + bool var_12316_interleave_0 = const()[name = string("op_12316_interleave_0"), val = bool(false)]; + tensor var_12316_cast_fp16 = concat(axis = var_10871, interleave = var_12316_interleave_0, values = (var_12168_cast_fp16, var_12170_cast_fp16, var_12172_cast_fp16, var_12174_cast_fp16))[name = string("op_12316_cast_fp16")]; + bool var_12318_interleave_0 = const()[name = string("op_12318_interleave_0"), val = bool(false)]; + tensor var_12318_cast_fp16 = concat(axis = var_10871, interleave = var_12318_interleave_0, values = (var_12176_cast_fp16, var_12178_cast_fp16, var_12180_cast_fp16, var_12182_cast_fp16))[name = string("op_12318_cast_fp16")]; + bool var_12320_interleave_0 = const()[name = string("op_12320_interleave_0"), val = bool(false)]; + tensor var_12320_cast_fp16 = concat(axis = var_10871, interleave = var_12320_interleave_0, values = (var_12184_cast_fp16, var_12186_cast_fp16, var_12188_cast_fp16, var_12190_cast_fp16))[name = string("op_12320_cast_fp16")]; + bool var_12322_interleave_0 = const()[name = string("op_12322_interleave_0"), val = bool(false)]; + tensor var_12322_cast_fp16 = concat(axis = var_10871, interleave = var_12322_interleave_0, values = (var_12192_cast_fp16, var_12194_cast_fp16, var_12196_cast_fp16, var_12198_cast_fp16))[name = string("op_12322_cast_fp16")]; + bool var_12324_interleave_0 = const()[name = string("op_12324_interleave_0"), val = bool(false)]; + tensor var_12324_cast_fp16 = concat(axis = var_10871, interleave = var_12324_interleave_0, values = (var_12200_cast_fp16, var_12202_cast_fp16, var_12204_cast_fp16, var_12206_cast_fp16))[name = string("op_12324_cast_fp16")]; + bool var_12326_interleave_0 = const()[name = string("op_12326_interleave_0"), val = bool(false)]; + tensor var_12326_cast_fp16 = concat(axis = var_10871, interleave = var_12326_interleave_0, values = (var_12208_cast_fp16, var_12210_cast_fp16, var_12212_cast_fp16, var_12214_cast_fp16))[name = string("op_12326_cast_fp16")]; + bool var_12328_interleave_0 = const()[name = string("op_12328_interleave_0"), val = bool(false)]; + tensor var_12328_cast_fp16 = concat(axis = var_10871, interleave = var_12328_interleave_0, values = (var_12216_cast_fp16, var_12218_cast_fp16, var_12220_cast_fp16, var_12222_cast_fp16))[name = string("op_12328_cast_fp16")]; + bool var_12330_interleave_0 = const()[name = string("op_12330_interleave_0"), val = bool(false)]; + tensor var_12330_cast_fp16 = concat(axis = var_10871, interleave = var_12330_interleave_0, values = (var_12224_cast_fp16, var_12226_cast_fp16, var_12228_cast_fp16, var_12230_cast_fp16))[name = string("op_12330_cast_fp16")]; + bool var_12332_interleave_0 = const()[name = string("op_12332_interleave_0"), val = bool(false)]; + tensor var_12332_cast_fp16 = concat(axis = var_10871, interleave = var_12332_interleave_0, values = (var_12232_cast_fp16, var_12234_cast_fp16, var_12236_cast_fp16, var_12238_cast_fp16))[name = string("op_12332_cast_fp16")]; + bool var_12334_interleave_0 = const()[name = string("op_12334_interleave_0"), val = bool(false)]; + tensor var_12334_cast_fp16 = concat(axis = var_10871, interleave = var_12334_interleave_0, values = (var_12240_cast_fp16, var_12242_cast_fp16, var_12244_cast_fp16, var_12246_cast_fp16))[name = string("op_12334_cast_fp16")]; + bool var_12336_interleave_0 = const()[name = string("op_12336_interleave_0"), val = bool(false)]; + tensor var_12336_cast_fp16 = concat(axis = var_10871, interleave = var_12336_interleave_0, values = (var_12248_cast_fp16, var_12250_cast_fp16, var_12252_cast_fp16, var_12254_cast_fp16))[name = string("op_12336_cast_fp16")]; + bool var_12338_interleave_0 = const()[name = string("op_12338_interleave_0"), val = bool(false)]; + tensor var_12338_cast_fp16 = concat(axis = var_10871, interleave = var_12338_interleave_0, values = (var_12256_cast_fp16, var_12258_cast_fp16, var_12260_cast_fp16, var_12262_cast_fp16))[name = string("op_12338_cast_fp16")]; + bool var_12340_interleave_0 = const()[name = string("op_12340_interleave_0"), val = bool(false)]; + tensor var_12340_cast_fp16 = concat(axis = var_10871, interleave = var_12340_interleave_0, values = (var_12264_cast_fp16, var_12266_cast_fp16, var_12268_cast_fp16, var_12270_cast_fp16))[name = string("op_12340_cast_fp16")]; + bool var_12342_interleave_0 = const()[name = string("op_12342_interleave_0"), val = bool(false)]; + tensor var_12342_cast_fp16 = concat(axis = var_10871, interleave = var_12342_interleave_0, values = (var_12272_cast_fp16, var_12274_cast_fp16, var_12276_cast_fp16, var_12278_cast_fp16))[name = string("op_12342_cast_fp16")]; + bool var_12344_interleave_0 = const()[name = string("op_12344_interleave_0"), val = bool(false)]; + tensor var_12344_cast_fp16 = concat(axis = var_10871, interleave = var_12344_interleave_0, values = (var_12280_cast_fp16, var_12282_cast_fp16, var_12284_cast_fp16, var_12286_cast_fp16))[name = string("op_12344_cast_fp16")]; + bool var_12346_interleave_0 = const()[name = string("op_12346_interleave_0"), val = bool(false)]; + tensor var_12346_cast_fp16 = concat(axis = var_10871, interleave = var_12346_interleave_0, values = (var_12288_cast_fp16, var_12290_cast_fp16, var_12292_cast_fp16, var_12294_cast_fp16))[name = string("op_12346_cast_fp16")]; + bool var_12348_interleave_0 = const()[name = string("op_12348_interleave_0"), val = bool(false)]; + tensor var_12348_cast_fp16 = concat(axis = var_10871, interleave = var_12348_interleave_0, values = (var_12296_cast_fp16, var_12298_cast_fp16, var_12300_cast_fp16, var_12302_cast_fp16))[name = string("op_12348_cast_fp16")]; + bool var_12350_interleave_0 = const()[name = string("op_12350_interleave_0"), val = bool(false)]; + tensor var_12350_cast_fp16 = concat(axis = var_10871, interleave = var_12350_interleave_0, values = (var_12304_cast_fp16, var_12306_cast_fp16, var_12308_cast_fp16, var_12310_cast_fp16))[name = string("op_12350_cast_fp16")]; + bool input_57_interleave_0 = const()[name = string("input_57_interleave_0"), val = bool(false)]; + tensor input_57_cast_fp16 = concat(axis = var_10896, interleave = input_57_interleave_0, values = (var_12312_cast_fp16, var_12314_cast_fp16, var_12316_cast_fp16, var_12318_cast_fp16, var_12320_cast_fp16, var_12322_cast_fp16, var_12324_cast_fp16, var_12326_cast_fp16, var_12328_cast_fp16, var_12330_cast_fp16, var_12332_cast_fp16, var_12334_cast_fp16, var_12336_cast_fp16, var_12338_cast_fp16, var_12340_cast_fp16, var_12342_cast_fp16, var_12344_cast_fp16, var_12346_cast_fp16, var_12348_cast_fp16, var_12350_cast_fp16))[name = string("input_57_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299978240)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303255104)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_12369_to_fp16 = const()[name = string("op_12369_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_12369_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303257728)))]; + tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303260352)))]; + fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303262976)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316370240)))]; + tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; + tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316380544)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329487808)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; + int32 var_12398 = const()[name = string("op_12398"), val = int32(3)]; + int32 var_12423 = const()[name = string("op_12423"), val = int32(1)]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_12440_to_fp16 = const()[name = string("op_12440_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_12440_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329490432)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329493056)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329495680)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332772544)))]; + tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; + string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; + tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; + tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; + int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332775168)))]; + tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; + string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; + tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; + tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; + int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336052032)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339328896)))]; + tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_12478_begin_0 = const()[name = string("op_12478_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12478_end_0 = const()[name = string("op_12478_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12478_end_mask_0 = const()[name = string("op_12478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12478_cast_fp16 = slice_by_index(begin = var_12478_begin_0, end = var_12478_end_0, end_mask = var_12478_end_mask_0, x = query_17_cast_fp16)[name = string("op_12478_cast_fp16")]; + tensor var_12482_begin_0 = const()[name = string("op_12482_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12482_end_0 = const()[name = string("op_12482_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_12482_end_mask_0 = const()[name = string("op_12482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12482_cast_fp16 = slice_by_index(begin = var_12482_begin_0, end = var_12482_end_0, end_mask = var_12482_end_mask_0, x = query_17_cast_fp16)[name = string("op_12482_cast_fp16")]; + tensor var_12486_begin_0 = const()[name = string("op_12486_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12486_end_0 = const()[name = string("op_12486_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_12486_end_mask_0 = const()[name = string("op_12486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12486_cast_fp16 = slice_by_index(begin = var_12486_begin_0, end = var_12486_end_0, end_mask = var_12486_end_mask_0, x = query_17_cast_fp16)[name = string("op_12486_cast_fp16")]; + tensor var_12490_begin_0 = const()[name = string("op_12490_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12490_end_0 = const()[name = string("op_12490_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_12490_end_mask_0 = const()[name = string("op_12490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12490_cast_fp16 = slice_by_index(begin = var_12490_begin_0, end = var_12490_end_0, end_mask = var_12490_end_mask_0, x = query_17_cast_fp16)[name = string("op_12490_cast_fp16")]; + tensor var_12494_begin_0 = const()[name = string("op_12494_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12494_end_0 = const()[name = string("op_12494_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_12494_end_mask_0 = const()[name = string("op_12494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12494_cast_fp16 = slice_by_index(begin = var_12494_begin_0, end = var_12494_end_0, end_mask = var_12494_end_mask_0, x = query_17_cast_fp16)[name = string("op_12494_cast_fp16")]; + tensor var_12498_begin_0 = const()[name = string("op_12498_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12498_end_0 = const()[name = string("op_12498_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_12498_end_mask_0 = const()[name = string("op_12498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12498_cast_fp16 = slice_by_index(begin = var_12498_begin_0, end = var_12498_end_0, end_mask = var_12498_end_mask_0, x = query_17_cast_fp16)[name = string("op_12498_cast_fp16")]; + tensor var_12502_begin_0 = const()[name = string("op_12502_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12502_end_0 = const()[name = string("op_12502_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_12502_end_mask_0 = const()[name = string("op_12502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12502_cast_fp16 = slice_by_index(begin = var_12502_begin_0, end = var_12502_end_0, end_mask = var_12502_end_mask_0, x = query_17_cast_fp16)[name = string("op_12502_cast_fp16")]; + tensor var_12506_begin_0 = const()[name = string("op_12506_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12506_end_0 = const()[name = string("op_12506_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_12506_end_mask_0 = const()[name = string("op_12506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12506_cast_fp16 = slice_by_index(begin = var_12506_begin_0, end = var_12506_end_0, end_mask = var_12506_end_mask_0, x = query_17_cast_fp16)[name = string("op_12506_cast_fp16")]; + tensor var_12510_begin_0 = const()[name = string("op_12510_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12510_end_0 = const()[name = string("op_12510_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_12510_end_mask_0 = const()[name = string("op_12510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12510_cast_fp16 = slice_by_index(begin = var_12510_begin_0, end = var_12510_end_0, end_mask = var_12510_end_mask_0, x = query_17_cast_fp16)[name = string("op_12510_cast_fp16")]; + tensor var_12514_begin_0 = const()[name = string("op_12514_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12514_end_0 = const()[name = string("op_12514_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_12514_end_mask_0 = const()[name = string("op_12514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12514_cast_fp16 = slice_by_index(begin = var_12514_begin_0, end = var_12514_end_0, end_mask = var_12514_end_mask_0, x = query_17_cast_fp16)[name = string("op_12514_cast_fp16")]; + tensor var_12518_begin_0 = const()[name = string("op_12518_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12518_end_0 = const()[name = string("op_12518_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_12518_end_mask_0 = const()[name = string("op_12518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12518_cast_fp16 = slice_by_index(begin = var_12518_begin_0, end = var_12518_end_0, end_mask = var_12518_end_mask_0, x = query_17_cast_fp16)[name = string("op_12518_cast_fp16")]; + tensor var_12522_begin_0 = const()[name = string("op_12522_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12522_end_0 = const()[name = string("op_12522_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_12522_end_mask_0 = const()[name = string("op_12522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12522_cast_fp16 = slice_by_index(begin = var_12522_begin_0, end = var_12522_end_0, end_mask = var_12522_end_mask_0, x = query_17_cast_fp16)[name = string("op_12522_cast_fp16")]; + tensor var_12526_begin_0 = const()[name = string("op_12526_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12526_end_0 = const()[name = string("op_12526_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_12526_end_mask_0 = const()[name = string("op_12526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12526_cast_fp16 = slice_by_index(begin = var_12526_begin_0, end = var_12526_end_0, end_mask = var_12526_end_mask_0, x = query_17_cast_fp16)[name = string("op_12526_cast_fp16")]; + tensor var_12530_begin_0 = const()[name = string("op_12530_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12530_end_0 = const()[name = string("op_12530_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_12530_end_mask_0 = const()[name = string("op_12530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12530_cast_fp16 = slice_by_index(begin = var_12530_begin_0, end = var_12530_end_0, end_mask = var_12530_end_mask_0, x = query_17_cast_fp16)[name = string("op_12530_cast_fp16")]; + tensor var_12534_begin_0 = const()[name = string("op_12534_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12534_end_0 = const()[name = string("op_12534_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_12534_end_mask_0 = const()[name = string("op_12534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12534_cast_fp16 = slice_by_index(begin = var_12534_begin_0, end = var_12534_end_0, end_mask = var_12534_end_mask_0, x = query_17_cast_fp16)[name = string("op_12534_cast_fp16")]; + tensor var_12538_begin_0 = const()[name = string("op_12538_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12538_end_0 = const()[name = string("op_12538_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_12538_end_mask_0 = const()[name = string("op_12538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12538_cast_fp16 = slice_by_index(begin = var_12538_begin_0, end = var_12538_end_0, end_mask = var_12538_end_mask_0, x = query_17_cast_fp16)[name = string("op_12538_cast_fp16")]; + tensor var_12542_begin_0 = const()[name = string("op_12542_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12542_end_0 = const()[name = string("op_12542_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_12542_end_mask_0 = const()[name = string("op_12542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12542_cast_fp16 = slice_by_index(begin = var_12542_begin_0, end = var_12542_end_0, end_mask = var_12542_end_mask_0, x = query_17_cast_fp16)[name = string("op_12542_cast_fp16")]; + tensor var_12546_begin_0 = const()[name = string("op_12546_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12546_end_0 = const()[name = string("op_12546_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_12546_end_mask_0 = const()[name = string("op_12546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12546_cast_fp16 = slice_by_index(begin = var_12546_begin_0, end = var_12546_end_0, end_mask = var_12546_end_mask_0, x = query_17_cast_fp16)[name = string("op_12546_cast_fp16")]; + tensor var_12550_begin_0 = const()[name = string("op_12550_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12550_end_0 = const()[name = string("op_12550_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_12550_end_mask_0 = const()[name = string("op_12550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12550_cast_fp16 = slice_by_index(begin = var_12550_begin_0, end = var_12550_end_0, end_mask = var_12550_end_mask_0, x = query_17_cast_fp16)[name = string("op_12550_cast_fp16")]; + tensor var_12554_begin_0 = const()[name = string("op_12554_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12554_end_0 = const()[name = string("op_12554_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_12554_end_mask_0 = const()[name = string("op_12554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12554_cast_fp16 = slice_by_index(begin = var_12554_begin_0, end = var_12554_end_0, end_mask = var_12554_end_mask_0, x = query_17_cast_fp16)[name = string("op_12554_cast_fp16")]; + tensor var_12563_begin_0 = const()[name = string("op_12563_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12563_end_0 = const()[name = string("op_12563_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12563_end_mask_0 = const()[name = string("op_12563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12563_cast_fp16 = slice_by_index(begin = var_12563_begin_0, end = var_12563_end_0, end_mask = var_12563_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12563_cast_fp16")]; + tensor var_12570_begin_0 = const()[name = string("op_12570_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12570_end_0 = const()[name = string("op_12570_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12570_end_mask_0 = const()[name = string("op_12570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12570_cast_fp16 = slice_by_index(begin = var_12570_begin_0, end = var_12570_end_0, end_mask = var_12570_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12570_cast_fp16")]; + tensor var_12577_begin_0 = const()[name = string("op_12577_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12577_end_0 = const()[name = string("op_12577_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12577_end_mask_0 = const()[name = string("op_12577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12577_cast_fp16 = slice_by_index(begin = var_12577_begin_0, end = var_12577_end_0, end_mask = var_12577_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12577_cast_fp16")]; + tensor var_12584_begin_0 = const()[name = string("op_12584_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12584_end_0 = const()[name = string("op_12584_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12584_end_mask_0 = const()[name = string("op_12584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12584_cast_fp16 = slice_by_index(begin = var_12584_begin_0, end = var_12584_end_0, end_mask = var_12584_end_mask_0, x = var_12478_cast_fp16)[name = string("op_12584_cast_fp16")]; + tensor var_12591_begin_0 = const()[name = string("op_12591_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12591_end_0 = const()[name = string("op_12591_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12591_end_mask_0 = const()[name = string("op_12591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12591_cast_fp16 = slice_by_index(begin = var_12591_begin_0, end = var_12591_end_0, end_mask = var_12591_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12591_cast_fp16")]; + tensor var_12598_begin_0 = const()[name = string("op_12598_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12598_end_0 = const()[name = string("op_12598_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12598_end_mask_0 = const()[name = string("op_12598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12598_cast_fp16 = slice_by_index(begin = var_12598_begin_0, end = var_12598_end_0, end_mask = var_12598_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12598_cast_fp16")]; + tensor var_12605_begin_0 = const()[name = string("op_12605_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12605_end_0 = const()[name = string("op_12605_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12605_end_mask_0 = const()[name = string("op_12605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12605_cast_fp16 = slice_by_index(begin = var_12605_begin_0, end = var_12605_end_0, end_mask = var_12605_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12605_cast_fp16")]; + tensor var_12612_begin_0 = const()[name = string("op_12612_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12612_end_0 = const()[name = string("op_12612_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12612_end_mask_0 = const()[name = string("op_12612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12612_cast_fp16 = slice_by_index(begin = var_12612_begin_0, end = var_12612_end_0, end_mask = var_12612_end_mask_0, x = var_12482_cast_fp16)[name = string("op_12612_cast_fp16")]; + tensor var_12619_begin_0 = const()[name = string("op_12619_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12619_end_0 = const()[name = string("op_12619_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12619_end_mask_0 = const()[name = string("op_12619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12619_cast_fp16 = slice_by_index(begin = var_12619_begin_0, end = var_12619_end_0, end_mask = var_12619_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12619_cast_fp16")]; + tensor var_12626_begin_0 = const()[name = string("op_12626_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12626_end_0 = const()[name = string("op_12626_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12626_end_mask_0 = const()[name = string("op_12626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12626_cast_fp16 = slice_by_index(begin = var_12626_begin_0, end = var_12626_end_0, end_mask = var_12626_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12626_cast_fp16")]; + tensor var_12633_begin_0 = const()[name = string("op_12633_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12633_end_0 = const()[name = string("op_12633_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12633_end_mask_0 = const()[name = string("op_12633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12633_cast_fp16 = slice_by_index(begin = var_12633_begin_0, end = var_12633_end_0, end_mask = var_12633_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12633_cast_fp16")]; + tensor var_12640_begin_0 = const()[name = string("op_12640_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12640_end_0 = const()[name = string("op_12640_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12640_end_mask_0 = const()[name = string("op_12640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12640_cast_fp16 = slice_by_index(begin = var_12640_begin_0, end = var_12640_end_0, end_mask = var_12640_end_mask_0, x = var_12486_cast_fp16)[name = string("op_12640_cast_fp16")]; + tensor var_12647_begin_0 = const()[name = string("op_12647_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12647_end_0 = const()[name = string("op_12647_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12647_end_mask_0 = const()[name = string("op_12647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12647_cast_fp16 = slice_by_index(begin = var_12647_begin_0, end = var_12647_end_0, end_mask = var_12647_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12647_cast_fp16")]; + tensor var_12654_begin_0 = const()[name = string("op_12654_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12654_end_0 = const()[name = string("op_12654_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12654_end_mask_0 = const()[name = string("op_12654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12654_cast_fp16 = slice_by_index(begin = var_12654_begin_0, end = var_12654_end_0, end_mask = var_12654_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12654_cast_fp16")]; + tensor var_12661_begin_0 = const()[name = string("op_12661_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12661_end_0 = const()[name = string("op_12661_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12661_end_mask_0 = const()[name = string("op_12661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12661_cast_fp16 = slice_by_index(begin = var_12661_begin_0, end = var_12661_end_0, end_mask = var_12661_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12661_cast_fp16")]; + tensor var_12668_begin_0 = const()[name = string("op_12668_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12668_end_0 = const()[name = string("op_12668_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12668_end_mask_0 = const()[name = string("op_12668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12668_cast_fp16 = slice_by_index(begin = var_12668_begin_0, end = var_12668_end_0, end_mask = var_12668_end_mask_0, x = var_12490_cast_fp16)[name = string("op_12668_cast_fp16")]; + tensor var_12675_begin_0 = const()[name = string("op_12675_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12675_end_0 = const()[name = string("op_12675_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12675_end_mask_0 = const()[name = string("op_12675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12675_cast_fp16 = slice_by_index(begin = var_12675_begin_0, end = var_12675_end_0, end_mask = var_12675_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12675_cast_fp16")]; + tensor var_12682_begin_0 = const()[name = string("op_12682_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12682_end_0 = const()[name = string("op_12682_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12682_end_mask_0 = const()[name = string("op_12682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12682_cast_fp16 = slice_by_index(begin = var_12682_begin_0, end = var_12682_end_0, end_mask = var_12682_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12682_cast_fp16")]; + tensor var_12689_begin_0 = const()[name = string("op_12689_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12689_end_0 = const()[name = string("op_12689_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12689_end_mask_0 = const()[name = string("op_12689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12689_cast_fp16 = slice_by_index(begin = var_12689_begin_0, end = var_12689_end_0, end_mask = var_12689_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12689_cast_fp16")]; + tensor var_12696_begin_0 = const()[name = string("op_12696_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12696_end_0 = const()[name = string("op_12696_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12696_end_mask_0 = const()[name = string("op_12696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12696_cast_fp16 = slice_by_index(begin = var_12696_begin_0, end = var_12696_end_0, end_mask = var_12696_end_mask_0, x = var_12494_cast_fp16)[name = string("op_12696_cast_fp16")]; + tensor var_12703_begin_0 = const()[name = string("op_12703_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12703_end_0 = const()[name = string("op_12703_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12703_end_mask_0 = const()[name = string("op_12703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12703_cast_fp16 = slice_by_index(begin = var_12703_begin_0, end = var_12703_end_0, end_mask = var_12703_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12703_cast_fp16")]; + tensor var_12710_begin_0 = const()[name = string("op_12710_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12710_end_0 = const()[name = string("op_12710_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12710_end_mask_0 = const()[name = string("op_12710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12710_cast_fp16 = slice_by_index(begin = var_12710_begin_0, end = var_12710_end_0, end_mask = var_12710_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12710_cast_fp16")]; + tensor var_12717_begin_0 = const()[name = string("op_12717_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12717_end_0 = const()[name = string("op_12717_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12717_end_mask_0 = const()[name = string("op_12717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12717_cast_fp16 = slice_by_index(begin = var_12717_begin_0, end = var_12717_end_0, end_mask = var_12717_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12717_cast_fp16")]; + tensor var_12724_begin_0 = const()[name = string("op_12724_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12724_end_0 = const()[name = string("op_12724_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12724_end_mask_0 = const()[name = string("op_12724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12724_cast_fp16 = slice_by_index(begin = var_12724_begin_0, end = var_12724_end_0, end_mask = var_12724_end_mask_0, x = var_12498_cast_fp16)[name = string("op_12724_cast_fp16")]; + tensor var_12731_begin_0 = const()[name = string("op_12731_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12731_end_0 = const()[name = string("op_12731_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12731_end_mask_0 = const()[name = string("op_12731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12731_cast_fp16 = slice_by_index(begin = var_12731_begin_0, end = var_12731_end_0, end_mask = var_12731_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12731_cast_fp16")]; + tensor var_12738_begin_0 = const()[name = string("op_12738_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12738_end_0 = const()[name = string("op_12738_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12738_end_mask_0 = const()[name = string("op_12738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12738_cast_fp16 = slice_by_index(begin = var_12738_begin_0, end = var_12738_end_0, end_mask = var_12738_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12738_cast_fp16")]; + tensor var_12745_begin_0 = const()[name = string("op_12745_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12745_end_0 = const()[name = string("op_12745_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12745_end_mask_0 = const()[name = string("op_12745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12745_cast_fp16 = slice_by_index(begin = var_12745_begin_0, end = var_12745_end_0, end_mask = var_12745_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12745_cast_fp16")]; + tensor var_12752_begin_0 = const()[name = string("op_12752_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12752_end_0 = const()[name = string("op_12752_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12752_end_mask_0 = const()[name = string("op_12752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12752_cast_fp16 = slice_by_index(begin = var_12752_begin_0, end = var_12752_end_0, end_mask = var_12752_end_mask_0, x = var_12502_cast_fp16)[name = string("op_12752_cast_fp16")]; + tensor var_12759_begin_0 = const()[name = string("op_12759_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12759_end_0 = const()[name = string("op_12759_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12759_end_mask_0 = const()[name = string("op_12759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12759_cast_fp16 = slice_by_index(begin = var_12759_begin_0, end = var_12759_end_0, end_mask = var_12759_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12759_cast_fp16")]; + tensor var_12766_begin_0 = const()[name = string("op_12766_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12766_end_0 = const()[name = string("op_12766_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12766_end_mask_0 = const()[name = string("op_12766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12766_cast_fp16 = slice_by_index(begin = var_12766_begin_0, end = var_12766_end_0, end_mask = var_12766_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12766_cast_fp16")]; + tensor var_12773_begin_0 = const()[name = string("op_12773_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12773_end_0 = const()[name = string("op_12773_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12773_end_mask_0 = const()[name = string("op_12773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12773_cast_fp16 = slice_by_index(begin = var_12773_begin_0, end = var_12773_end_0, end_mask = var_12773_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12773_cast_fp16")]; + tensor var_12780_begin_0 = const()[name = string("op_12780_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12780_end_0 = const()[name = string("op_12780_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12780_end_mask_0 = const()[name = string("op_12780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12780_cast_fp16 = slice_by_index(begin = var_12780_begin_0, end = var_12780_end_0, end_mask = var_12780_end_mask_0, x = var_12506_cast_fp16)[name = string("op_12780_cast_fp16")]; + tensor var_12787_begin_0 = const()[name = string("op_12787_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12787_end_0 = const()[name = string("op_12787_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12787_end_mask_0 = const()[name = string("op_12787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12787_cast_fp16 = slice_by_index(begin = var_12787_begin_0, end = var_12787_end_0, end_mask = var_12787_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12787_cast_fp16")]; + tensor var_12794_begin_0 = const()[name = string("op_12794_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12794_end_0 = const()[name = string("op_12794_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12794_end_mask_0 = const()[name = string("op_12794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12794_cast_fp16 = slice_by_index(begin = var_12794_begin_0, end = var_12794_end_0, end_mask = var_12794_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12794_cast_fp16")]; + tensor var_12801_begin_0 = const()[name = string("op_12801_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12801_end_0 = const()[name = string("op_12801_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12801_end_mask_0 = const()[name = string("op_12801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12801_cast_fp16 = slice_by_index(begin = var_12801_begin_0, end = var_12801_end_0, end_mask = var_12801_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12801_cast_fp16")]; + tensor var_12808_begin_0 = const()[name = string("op_12808_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12808_end_0 = const()[name = string("op_12808_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12808_end_mask_0 = const()[name = string("op_12808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12808_cast_fp16 = slice_by_index(begin = var_12808_begin_0, end = var_12808_end_0, end_mask = var_12808_end_mask_0, x = var_12510_cast_fp16)[name = string("op_12808_cast_fp16")]; + tensor var_12815_begin_0 = const()[name = string("op_12815_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12815_end_0 = const()[name = string("op_12815_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12815_end_mask_0 = const()[name = string("op_12815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12815_cast_fp16 = slice_by_index(begin = var_12815_begin_0, end = var_12815_end_0, end_mask = var_12815_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12815_cast_fp16")]; + tensor var_12822_begin_0 = const()[name = string("op_12822_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12822_end_0 = const()[name = string("op_12822_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12822_end_mask_0 = const()[name = string("op_12822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12822_cast_fp16 = slice_by_index(begin = var_12822_begin_0, end = var_12822_end_0, end_mask = var_12822_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12822_cast_fp16")]; + tensor var_12829_begin_0 = const()[name = string("op_12829_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12829_end_0 = const()[name = string("op_12829_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12829_end_mask_0 = const()[name = string("op_12829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12829_cast_fp16 = slice_by_index(begin = var_12829_begin_0, end = var_12829_end_0, end_mask = var_12829_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12829_cast_fp16")]; + tensor var_12836_begin_0 = const()[name = string("op_12836_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12836_end_0 = const()[name = string("op_12836_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12836_end_mask_0 = const()[name = string("op_12836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12836_cast_fp16 = slice_by_index(begin = var_12836_begin_0, end = var_12836_end_0, end_mask = var_12836_end_mask_0, x = var_12514_cast_fp16)[name = string("op_12836_cast_fp16")]; + tensor var_12843_begin_0 = const()[name = string("op_12843_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12843_end_0 = const()[name = string("op_12843_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12843_end_mask_0 = const()[name = string("op_12843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12843_cast_fp16 = slice_by_index(begin = var_12843_begin_0, end = var_12843_end_0, end_mask = var_12843_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12843_cast_fp16")]; + tensor var_12850_begin_0 = const()[name = string("op_12850_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12850_end_0 = const()[name = string("op_12850_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12850_end_mask_0 = const()[name = string("op_12850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12850_cast_fp16 = slice_by_index(begin = var_12850_begin_0, end = var_12850_end_0, end_mask = var_12850_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12850_cast_fp16")]; + tensor var_12857_begin_0 = const()[name = string("op_12857_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12857_end_0 = const()[name = string("op_12857_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12857_end_mask_0 = const()[name = string("op_12857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12857_cast_fp16 = slice_by_index(begin = var_12857_begin_0, end = var_12857_end_0, end_mask = var_12857_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12857_cast_fp16")]; + tensor var_12864_begin_0 = const()[name = string("op_12864_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12864_end_0 = const()[name = string("op_12864_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12864_end_mask_0 = const()[name = string("op_12864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12864_cast_fp16 = slice_by_index(begin = var_12864_begin_0, end = var_12864_end_0, end_mask = var_12864_end_mask_0, x = var_12518_cast_fp16)[name = string("op_12864_cast_fp16")]; + tensor var_12871_begin_0 = const()[name = string("op_12871_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12871_end_0 = const()[name = string("op_12871_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12871_end_mask_0 = const()[name = string("op_12871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12871_cast_fp16 = slice_by_index(begin = var_12871_begin_0, end = var_12871_end_0, end_mask = var_12871_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12871_cast_fp16")]; + tensor var_12878_begin_0 = const()[name = string("op_12878_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12878_end_0 = const()[name = string("op_12878_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12878_end_mask_0 = const()[name = string("op_12878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12878_cast_fp16 = slice_by_index(begin = var_12878_begin_0, end = var_12878_end_0, end_mask = var_12878_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12878_cast_fp16")]; + tensor var_12885_begin_0 = const()[name = string("op_12885_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12885_end_0 = const()[name = string("op_12885_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12885_end_mask_0 = const()[name = string("op_12885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12885_cast_fp16 = slice_by_index(begin = var_12885_begin_0, end = var_12885_end_0, end_mask = var_12885_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12885_cast_fp16")]; + tensor var_12892_begin_0 = const()[name = string("op_12892_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12892_end_0 = const()[name = string("op_12892_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12892_end_mask_0 = const()[name = string("op_12892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12892_cast_fp16 = slice_by_index(begin = var_12892_begin_0, end = var_12892_end_0, end_mask = var_12892_end_mask_0, x = var_12522_cast_fp16)[name = string("op_12892_cast_fp16")]; + tensor var_12899_begin_0 = const()[name = string("op_12899_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12899_end_0 = const()[name = string("op_12899_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12899_end_mask_0 = const()[name = string("op_12899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12899_cast_fp16 = slice_by_index(begin = var_12899_begin_0, end = var_12899_end_0, end_mask = var_12899_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12899_cast_fp16")]; + tensor var_12906_begin_0 = const()[name = string("op_12906_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12906_end_0 = const()[name = string("op_12906_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12906_end_mask_0 = const()[name = string("op_12906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12906_cast_fp16 = slice_by_index(begin = var_12906_begin_0, end = var_12906_end_0, end_mask = var_12906_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12906_cast_fp16")]; + tensor var_12913_begin_0 = const()[name = string("op_12913_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12913_end_0 = const()[name = string("op_12913_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12913_end_mask_0 = const()[name = string("op_12913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12913_cast_fp16 = slice_by_index(begin = var_12913_begin_0, end = var_12913_end_0, end_mask = var_12913_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12913_cast_fp16")]; + tensor var_12920_begin_0 = const()[name = string("op_12920_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12920_end_0 = const()[name = string("op_12920_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12920_end_mask_0 = const()[name = string("op_12920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12920_cast_fp16 = slice_by_index(begin = var_12920_begin_0, end = var_12920_end_0, end_mask = var_12920_end_mask_0, x = var_12526_cast_fp16)[name = string("op_12920_cast_fp16")]; + tensor var_12927_begin_0 = const()[name = string("op_12927_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12927_end_0 = const()[name = string("op_12927_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12927_end_mask_0 = const()[name = string("op_12927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12927_cast_fp16 = slice_by_index(begin = var_12927_begin_0, end = var_12927_end_0, end_mask = var_12927_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12927_cast_fp16")]; + tensor var_12934_begin_0 = const()[name = string("op_12934_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12934_end_0 = const()[name = string("op_12934_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12934_end_mask_0 = const()[name = string("op_12934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12934_cast_fp16 = slice_by_index(begin = var_12934_begin_0, end = var_12934_end_0, end_mask = var_12934_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12934_cast_fp16")]; + tensor var_12941_begin_0 = const()[name = string("op_12941_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12941_end_0 = const()[name = string("op_12941_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12941_end_mask_0 = const()[name = string("op_12941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12941_cast_fp16 = slice_by_index(begin = var_12941_begin_0, end = var_12941_end_0, end_mask = var_12941_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12941_cast_fp16")]; + tensor var_12948_begin_0 = const()[name = string("op_12948_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12948_end_0 = const()[name = string("op_12948_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12948_end_mask_0 = const()[name = string("op_12948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12948_cast_fp16 = slice_by_index(begin = var_12948_begin_0, end = var_12948_end_0, end_mask = var_12948_end_mask_0, x = var_12530_cast_fp16)[name = string("op_12948_cast_fp16")]; + tensor var_12955_begin_0 = const()[name = string("op_12955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12955_end_0 = const()[name = string("op_12955_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12955_end_mask_0 = const()[name = string("op_12955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12955_cast_fp16 = slice_by_index(begin = var_12955_begin_0, end = var_12955_end_0, end_mask = var_12955_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12955_cast_fp16")]; + tensor var_12962_begin_0 = const()[name = string("op_12962_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12962_end_0 = const()[name = string("op_12962_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12962_end_mask_0 = const()[name = string("op_12962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12962_cast_fp16 = slice_by_index(begin = var_12962_begin_0, end = var_12962_end_0, end_mask = var_12962_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12962_cast_fp16")]; + tensor var_12969_begin_0 = const()[name = string("op_12969_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12969_end_0 = const()[name = string("op_12969_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12969_end_mask_0 = const()[name = string("op_12969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12969_cast_fp16 = slice_by_index(begin = var_12969_begin_0, end = var_12969_end_0, end_mask = var_12969_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12969_cast_fp16")]; + tensor var_12976_begin_0 = const()[name = string("op_12976_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12976_end_0 = const()[name = string("op_12976_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12976_end_mask_0 = const()[name = string("op_12976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12976_cast_fp16 = slice_by_index(begin = var_12976_begin_0, end = var_12976_end_0, end_mask = var_12976_end_mask_0, x = var_12534_cast_fp16)[name = string("op_12976_cast_fp16")]; + tensor var_12983_begin_0 = const()[name = string("op_12983_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12983_end_0 = const()[name = string("op_12983_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12983_end_mask_0 = const()[name = string("op_12983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12983_cast_fp16 = slice_by_index(begin = var_12983_begin_0, end = var_12983_end_0, end_mask = var_12983_end_mask_0, x = var_12538_cast_fp16)[name = string("op_12983_cast_fp16")]; + tensor var_12990_begin_0 = const()[name = string("op_12990_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12990_end_0 = const()[name = string("op_12990_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12990_end_mask_0 = const()[name = string("op_12990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12990_cast_fp16 = slice_by_index(begin = var_12990_begin_0, end = var_12990_end_0, end_mask = var_12990_end_mask_0, x = var_12538_cast_fp16)[name = string("op_12990_cast_fp16")]; + tensor var_12997_begin_0 = const()[name = string("op_12997_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12997_end_0 = const()[name = string("op_12997_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12997_end_mask_0 = const()[name = string("op_12997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12997_cast_fp16 = slice_by_index(begin = var_12997_begin_0, end = var_12997_end_0, end_mask = var_12997_end_mask_0, x = var_12538_cast_fp16)[name = string("op_12997_cast_fp16")]; + tensor var_13004_begin_0 = const()[name = string("op_13004_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13004_end_0 = const()[name = string("op_13004_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13004_end_mask_0 = const()[name = string("op_13004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13004_cast_fp16 = slice_by_index(begin = var_13004_begin_0, end = var_13004_end_0, end_mask = var_13004_end_mask_0, x = var_12538_cast_fp16)[name = string("op_13004_cast_fp16")]; + tensor var_13011_begin_0 = const()[name = string("op_13011_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13011_end_0 = const()[name = string("op_13011_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13011_end_mask_0 = const()[name = string("op_13011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13011_cast_fp16 = slice_by_index(begin = var_13011_begin_0, end = var_13011_end_0, end_mask = var_13011_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13011_cast_fp16")]; + tensor var_13018_begin_0 = const()[name = string("op_13018_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13018_end_0 = const()[name = string("op_13018_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13018_end_mask_0 = const()[name = string("op_13018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13018_cast_fp16 = slice_by_index(begin = var_13018_begin_0, end = var_13018_end_0, end_mask = var_13018_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13018_cast_fp16")]; + tensor var_13025_begin_0 = const()[name = string("op_13025_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13025_end_0 = const()[name = string("op_13025_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13025_end_mask_0 = const()[name = string("op_13025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13025_cast_fp16 = slice_by_index(begin = var_13025_begin_0, end = var_13025_end_0, end_mask = var_13025_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13025_cast_fp16")]; + tensor var_13032_begin_0 = const()[name = string("op_13032_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13032_end_0 = const()[name = string("op_13032_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13032_end_mask_0 = const()[name = string("op_13032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13032_cast_fp16 = slice_by_index(begin = var_13032_begin_0, end = var_13032_end_0, end_mask = var_13032_end_mask_0, x = var_12542_cast_fp16)[name = string("op_13032_cast_fp16")]; + tensor var_13039_begin_0 = const()[name = string("op_13039_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13039_end_0 = const()[name = string("op_13039_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13039_end_mask_0 = const()[name = string("op_13039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13039_cast_fp16 = slice_by_index(begin = var_13039_begin_0, end = var_13039_end_0, end_mask = var_13039_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13039_cast_fp16")]; + tensor var_13046_begin_0 = const()[name = string("op_13046_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13046_end_0 = const()[name = string("op_13046_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13046_end_mask_0 = const()[name = string("op_13046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13046_cast_fp16 = slice_by_index(begin = var_13046_begin_0, end = var_13046_end_0, end_mask = var_13046_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13046_cast_fp16")]; + tensor var_13053_begin_0 = const()[name = string("op_13053_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13053_end_0 = const()[name = string("op_13053_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13053_end_mask_0 = const()[name = string("op_13053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13053_cast_fp16 = slice_by_index(begin = var_13053_begin_0, end = var_13053_end_0, end_mask = var_13053_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13053_cast_fp16")]; + tensor var_13060_begin_0 = const()[name = string("op_13060_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13060_end_0 = const()[name = string("op_13060_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13060_end_mask_0 = const()[name = string("op_13060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13060_cast_fp16 = slice_by_index(begin = var_13060_begin_0, end = var_13060_end_0, end_mask = var_13060_end_mask_0, x = var_12546_cast_fp16)[name = string("op_13060_cast_fp16")]; + tensor var_13067_begin_0 = const()[name = string("op_13067_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13067_end_0 = const()[name = string("op_13067_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13067_end_mask_0 = const()[name = string("op_13067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13067_cast_fp16 = slice_by_index(begin = var_13067_begin_0, end = var_13067_end_0, end_mask = var_13067_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13067_cast_fp16")]; + tensor var_13074_begin_0 = const()[name = string("op_13074_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13074_end_0 = const()[name = string("op_13074_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13074_end_mask_0 = const()[name = string("op_13074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13074_cast_fp16 = slice_by_index(begin = var_13074_begin_0, end = var_13074_end_0, end_mask = var_13074_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13074_cast_fp16")]; + tensor var_13081_begin_0 = const()[name = string("op_13081_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13081_end_0 = const()[name = string("op_13081_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13081_end_mask_0 = const()[name = string("op_13081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13081_cast_fp16 = slice_by_index(begin = var_13081_begin_0, end = var_13081_end_0, end_mask = var_13081_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13081_cast_fp16")]; + tensor var_13088_begin_0 = const()[name = string("op_13088_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13088_end_0 = const()[name = string("op_13088_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13088_end_mask_0 = const()[name = string("op_13088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13088_cast_fp16 = slice_by_index(begin = var_13088_begin_0, end = var_13088_end_0, end_mask = var_13088_end_mask_0, x = var_12550_cast_fp16)[name = string("op_13088_cast_fp16")]; + tensor var_13095_begin_0 = const()[name = string("op_13095_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13095_end_0 = const()[name = string("op_13095_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13095_end_mask_0 = const()[name = string("op_13095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13095_cast_fp16 = slice_by_index(begin = var_13095_begin_0, end = var_13095_end_0, end_mask = var_13095_end_mask_0, x = var_12554_cast_fp16)[name = string("op_13095_cast_fp16")]; + tensor var_13102_begin_0 = const()[name = string("op_13102_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13102_end_0 = const()[name = string("op_13102_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13102_end_mask_0 = const()[name = string("op_13102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13102_cast_fp16 = slice_by_index(begin = var_13102_begin_0, end = var_13102_end_0, end_mask = var_13102_end_mask_0, x = var_12554_cast_fp16)[name = string("op_13102_cast_fp16")]; + tensor var_13109_begin_0 = const()[name = string("op_13109_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13109_end_0 = const()[name = string("op_13109_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13109_end_mask_0 = const()[name = string("op_13109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13109_cast_fp16 = slice_by_index(begin = var_13109_begin_0, end = var_13109_end_0, end_mask = var_13109_end_mask_0, x = var_12554_cast_fp16)[name = string("op_13109_cast_fp16")]; + tensor var_13116_begin_0 = const()[name = string("op_13116_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13116_end_0 = const()[name = string("op_13116_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13116_end_mask_0 = const()[name = string("op_13116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13116_cast_fp16 = slice_by_index(begin = var_13116_begin_0, end = var_13116_end_0, end_mask = var_13116_end_mask_0, x = var_12554_cast_fp16)[name = string("op_13116_cast_fp16")]; + tensor k_17_perm_0 = const()[name = string("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13121_begin_0 = const()[name = string("op_13121_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13121_end_0 = const()[name = string("op_13121_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_13121_end_mask_0 = const()[name = string("op_13121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_17_cast_fp16 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = string("transpose_23")]; + tensor var_13121_cast_fp16 = slice_by_index(begin = var_13121_begin_0, end = var_13121_end_0, end_mask = var_13121_end_mask_0, x = k_17_cast_fp16)[name = string("op_13121_cast_fp16")]; + tensor var_13125_begin_0 = const()[name = string("op_13125_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13125_end_0 = const()[name = string("op_13125_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_13125_end_mask_0 = const()[name = string("op_13125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13125_cast_fp16 = slice_by_index(begin = var_13125_begin_0, end = var_13125_end_0, end_mask = var_13125_end_mask_0, x = k_17_cast_fp16)[name = string("op_13125_cast_fp16")]; + tensor var_13129_begin_0 = const()[name = string("op_13129_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13129_end_0 = const()[name = string("op_13129_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_13129_end_mask_0 = const()[name = string("op_13129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13129_cast_fp16 = slice_by_index(begin = var_13129_begin_0, end = var_13129_end_0, end_mask = var_13129_end_mask_0, x = k_17_cast_fp16)[name = string("op_13129_cast_fp16")]; + tensor var_13133_begin_0 = const()[name = string("op_13133_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13133_end_0 = const()[name = string("op_13133_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_13133_end_mask_0 = const()[name = string("op_13133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13133_cast_fp16 = slice_by_index(begin = var_13133_begin_0, end = var_13133_end_0, end_mask = var_13133_end_mask_0, x = k_17_cast_fp16)[name = string("op_13133_cast_fp16")]; + tensor var_13137_begin_0 = const()[name = string("op_13137_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13137_end_0 = const()[name = string("op_13137_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_13137_end_mask_0 = const()[name = string("op_13137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13137_cast_fp16 = slice_by_index(begin = var_13137_begin_0, end = var_13137_end_0, end_mask = var_13137_end_mask_0, x = k_17_cast_fp16)[name = string("op_13137_cast_fp16")]; + tensor var_13141_begin_0 = const()[name = string("op_13141_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_13141_end_0 = const()[name = string("op_13141_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_13141_end_mask_0 = const()[name = string("op_13141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13141_cast_fp16 = slice_by_index(begin = var_13141_begin_0, end = var_13141_end_0, end_mask = var_13141_end_mask_0, x = k_17_cast_fp16)[name = string("op_13141_cast_fp16")]; + tensor var_13145_begin_0 = const()[name = string("op_13145_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_13145_end_0 = const()[name = string("op_13145_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_13145_end_mask_0 = const()[name = string("op_13145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13145_cast_fp16 = slice_by_index(begin = var_13145_begin_0, end = var_13145_end_0, end_mask = var_13145_end_mask_0, x = k_17_cast_fp16)[name = string("op_13145_cast_fp16")]; + tensor var_13149_begin_0 = const()[name = string("op_13149_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_13149_end_0 = const()[name = string("op_13149_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_13149_end_mask_0 = const()[name = string("op_13149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13149_cast_fp16 = slice_by_index(begin = var_13149_begin_0, end = var_13149_end_0, end_mask = var_13149_end_mask_0, x = k_17_cast_fp16)[name = string("op_13149_cast_fp16")]; + tensor var_13153_begin_0 = const()[name = string("op_13153_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13153_end_0 = const()[name = string("op_13153_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_13153_end_mask_0 = const()[name = string("op_13153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13153_cast_fp16 = slice_by_index(begin = var_13153_begin_0, end = var_13153_end_0, end_mask = var_13153_end_mask_0, x = k_17_cast_fp16)[name = string("op_13153_cast_fp16")]; + tensor var_13157_begin_0 = const()[name = string("op_13157_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_13157_end_0 = const()[name = string("op_13157_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_13157_end_mask_0 = const()[name = string("op_13157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13157_cast_fp16 = slice_by_index(begin = var_13157_begin_0, end = var_13157_end_0, end_mask = var_13157_end_mask_0, x = k_17_cast_fp16)[name = string("op_13157_cast_fp16")]; + tensor var_13161_begin_0 = const()[name = string("op_13161_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_13161_end_0 = const()[name = string("op_13161_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_13161_end_mask_0 = const()[name = string("op_13161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13161_cast_fp16 = slice_by_index(begin = var_13161_begin_0, end = var_13161_end_0, end_mask = var_13161_end_mask_0, x = k_17_cast_fp16)[name = string("op_13161_cast_fp16")]; + tensor var_13165_begin_0 = const()[name = string("op_13165_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_13165_end_0 = const()[name = string("op_13165_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_13165_end_mask_0 = const()[name = string("op_13165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13165_cast_fp16 = slice_by_index(begin = var_13165_begin_0, end = var_13165_end_0, end_mask = var_13165_end_mask_0, x = k_17_cast_fp16)[name = string("op_13165_cast_fp16")]; + tensor var_13169_begin_0 = const()[name = string("op_13169_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_13169_end_0 = const()[name = string("op_13169_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_13169_end_mask_0 = const()[name = string("op_13169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13169_cast_fp16 = slice_by_index(begin = var_13169_begin_0, end = var_13169_end_0, end_mask = var_13169_end_mask_0, x = k_17_cast_fp16)[name = string("op_13169_cast_fp16")]; + tensor var_13173_begin_0 = const()[name = string("op_13173_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_13173_end_0 = const()[name = string("op_13173_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_13173_end_mask_0 = const()[name = string("op_13173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13173_cast_fp16 = slice_by_index(begin = var_13173_begin_0, end = var_13173_end_0, end_mask = var_13173_end_mask_0, x = k_17_cast_fp16)[name = string("op_13173_cast_fp16")]; + tensor var_13177_begin_0 = const()[name = string("op_13177_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_13177_end_0 = const()[name = string("op_13177_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_13177_end_mask_0 = const()[name = string("op_13177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13177_cast_fp16 = slice_by_index(begin = var_13177_begin_0, end = var_13177_end_0, end_mask = var_13177_end_mask_0, x = k_17_cast_fp16)[name = string("op_13177_cast_fp16")]; + tensor var_13181_begin_0 = const()[name = string("op_13181_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_13181_end_0 = const()[name = string("op_13181_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_13181_end_mask_0 = const()[name = string("op_13181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13181_cast_fp16 = slice_by_index(begin = var_13181_begin_0, end = var_13181_end_0, end_mask = var_13181_end_mask_0, x = k_17_cast_fp16)[name = string("op_13181_cast_fp16")]; + tensor var_13185_begin_0 = const()[name = string("op_13185_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13185_end_0 = const()[name = string("op_13185_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_13185_end_mask_0 = const()[name = string("op_13185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13185_cast_fp16 = slice_by_index(begin = var_13185_begin_0, end = var_13185_end_0, end_mask = var_13185_end_mask_0, x = k_17_cast_fp16)[name = string("op_13185_cast_fp16")]; + tensor var_13189_begin_0 = const()[name = string("op_13189_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_13189_end_0 = const()[name = string("op_13189_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_13189_end_mask_0 = const()[name = string("op_13189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13189_cast_fp16 = slice_by_index(begin = var_13189_begin_0, end = var_13189_end_0, end_mask = var_13189_end_mask_0, x = k_17_cast_fp16)[name = string("op_13189_cast_fp16")]; + tensor var_13193_begin_0 = const()[name = string("op_13193_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_13193_end_0 = const()[name = string("op_13193_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_13193_end_mask_0 = const()[name = string("op_13193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13193_cast_fp16 = slice_by_index(begin = var_13193_begin_0, end = var_13193_end_0, end_mask = var_13193_end_mask_0, x = k_17_cast_fp16)[name = string("op_13193_cast_fp16")]; + tensor var_13197_begin_0 = const()[name = string("op_13197_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_13197_end_0 = const()[name = string("op_13197_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_13197_end_mask_0 = const()[name = string("op_13197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13197_cast_fp16 = slice_by_index(begin = var_13197_begin_0, end = var_13197_end_0, end_mask = var_13197_end_mask_0, x = k_17_cast_fp16)[name = string("op_13197_cast_fp16")]; + tensor var_13199_begin_0 = const()[name = string("op_13199_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13199_end_0 = const()[name = string("op_13199_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13199_end_mask_0 = const()[name = string("op_13199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13199_cast_fp16 = slice_by_index(begin = var_13199_begin_0, end = var_13199_end_0, end_mask = var_13199_end_mask_0, x = value_17_cast_fp16)[name = string("op_13199_cast_fp16")]; + tensor var_13203_begin_0 = const()[name = string("op_13203_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13203_end_0 = const()[name = string("op_13203_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_13203_end_mask_0 = const()[name = string("op_13203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13203_cast_fp16 = slice_by_index(begin = var_13203_begin_0, end = var_13203_end_0, end_mask = var_13203_end_mask_0, x = value_17_cast_fp16)[name = string("op_13203_cast_fp16")]; + tensor var_13207_begin_0 = const()[name = string("op_13207_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13207_end_0 = const()[name = string("op_13207_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_13207_end_mask_0 = const()[name = string("op_13207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13207_cast_fp16 = slice_by_index(begin = var_13207_begin_0, end = var_13207_end_0, end_mask = var_13207_end_mask_0, x = value_17_cast_fp16)[name = string("op_13207_cast_fp16")]; + tensor var_13211_begin_0 = const()[name = string("op_13211_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13211_end_0 = const()[name = string("op_13211_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_13211_end_mask_0 = const()[name = string("op_13211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13211_cast_fp16 = slice_by_index(begin = var_13211_begin_0, end = var_13211_end_0, end_mask = var_13211_end_mask_0, x = value_17_cast_fp16)[name = string("op_13211_cast_fp16")]; + tensor var_13215_begin_0 = const()[name = string("op_13215_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13215_end_0 = const()[name = string("op_13215_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_13215_end_mask_0 = const()[name = string("op_13215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13215_cast_fp16 = slice_by_index(begin = var_13215_begin_0, end = var_13215_end_0, end_mask = var_13215_end_mask_0, x = value_17_cast_fp16)[name = string("op_13215_cast_fp16")]; + tensor var_13219_begin_0 = const()[name = string("op_13219_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13219_end_0 = const()[name = string("op_13219_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_13219_end_mask_0 = const()[name = string("op_13219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13219_cast_fp16 = slice_by_index(begin = var_13219_begin_0, end = var_13219_end_0, end_mask = var_13219_end_mask_0, x = value_17_cast_fp16)[name = string("op_13219_cast_fp16")]; + tensor var_13223_begin_0 = const()[name = string("op_13223_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13223_end_0 = const()[name = string("op_13223_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_13223_end_mask_0 = const()[name = string("op_13223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13223_cast_fp16 = slice_by_index(begin = var_13223_begin_0, end = var_13223_end_0, end_mask = var_13223_end_mask_0, x = value_17_cast_fp16)[name = string("op_13223_cast_fp16")]; + tensor var_13227_begin_0 = const()[name = string("op_13227_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13227_end_0 = const()[name = string("op_13227_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_13227_end_mask_0 = const()[name = string("op_13227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13227_cast_fp16 = slice_by_index(begin = var_13227_begin_0, end = var_13227_end_0, end_mask = var_13227_end_mask_0, x = value_17_cast_fp16)[name = string("op_13227_cast_fp16")]; + tensor var_13231_begin_0 = const()[name = string("op_13231_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13231_end_0 = const()[name = string("op_13231_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_13231_end_mask_0 = const()[name = string("op_13231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13231_cast_fp16 = slice_by_index(begin = var_13231_begin_0, end = var_13231_end_0, end_mask = var_13231_end_mask_0, x = value_17_cast_fp16)[name = string("op_13231_cast_fp16")]; + tensor var_13235_begin_0 = const()[name = string("op_13235_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13235_end_0 = const()[name = string("op_13235_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_13235_end_mask_0 = const()[name = string("op_13235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13235_cast_fp16 = slice_by_index(begin = var_13235_begin_0, end = var_13235_end_0, end_mask = var_13235_end_mask_0, x = value_17_cast_fp16)[name = string("op_13235_cast_fp16")]; + tensor var_13239_begin_0 = const()[name = string("op_13239_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13239_end_0 = const()[name = string("op_13239_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_13239_end_mask_0 = const()[name = string("op_13239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13239_cast_fp16 = slice_by_index(begin = var_13239_begin_0, end = var_13239_end_0, end_mask = var_13239_end_mask_0, x = value_17_cast_fp16)[name = string("op_13239_cast_fp16")]; + tensor var_13243_begin_0 = const()[name = string("op_13243_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13243_end_0 = const()[name = string("op_13243_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_13243_end_mask_0 = const()[name = string("op_13243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13243_cast_fp16 = slice_by_index(begin = var_13243_begin_0, end = var_13243_end_0, end_mask = var_13243_end_mask_0, x = value_17_cast_fp16)[name = string("op_13243_cast_fp16")]; + tensor var_13247_begin_0 = const()[name = string("op_13247_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13247_end_0 = const()[name = string("op_13247_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_13247_end_mask_0 = const()[name = string("op_13247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13247_cast_fp16 = slice_by_index(begin = var_13247_begin_0, end = var_13247_end_0, end_mask = var_13247_end_mask_0, x = value_17_cast_fp16)[name = string("op_13247_cast_fp16")]; + tensor var_13251_begin_0 = const()[name = string("op_13251_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13251_end_0 = const()[name = string("op_13251_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_13251_end_mask_0 = const()[name = string("op_13251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13251_cast_fp16 = slice_by_index(begin = var_13251_begin_0, end = var_13251_end_0, end_mask = var_13251_end_mask_0, x = value_17_cast_fp16)[name = string("op_13251_cast_fp16")]; + tensor var_13255_begin_0 = const()[name = string("op_13255_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13255_end_0 = const()[name = string("op_13255_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_13255_end_mask_0 = const()[name = string("op_13255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13255_cast_fp16 = slice_by_index(begin = var_13255_begin_0, end = var_13255_end_0, end_mask = var_13255_end_mask_0, x = value_17_cast_fp16)[name = string("op_13255_cast_fp16")]; + tensor var_13259_begin_0 = const()[name = string("op_13259_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13259_end_0 = const()[name = string("op_13259_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_13259_end_mask_0 = const()[name = string("op_13259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13259_cast_fp16 = slice_by_index(begin = var_13259_begin_0, end = var_13259_end_0, end_mask = var_13259_end_mask_0, x = value_17_cast_fp16)[name = string("op_13259_cast_fp16")]; + tensor var_13263_begin_0 = const()[name = string("op_13263_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13263_end_0 = const()[name = string("op_13263_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_13263_end_mask_0 = const()[name = string("op_13263_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13263_cast_fp16 = slice_by_index(begin = var_13263_begin_0, end = var_13263_end_0, end_mask = var_13263_end_mask_0, x = value_17_cast_fp16)[name = string("op_13263_cast_fp16")]; + tensor var_13267_begin_0 = const()[name = string("op_13267_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13267_end_0 = const()[name = string("op_13267_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_13267_end_mask_0 = const()[name = string("op_13267_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13267_cast_fp16 = slice_by_index(begin = var_13267_begin_0, end = var_13267_end_0, end_mask = var_13267_end_mask_0, x = value_17_cast_fp16)[name = string("op_13267_cast_fp16")]; + tensor var_13271_begin_0 = const()[name = string("op_13271_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13271_end_0 = const()[name = string("op_13271_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_13271_end_mask_0 = const()[name = string("op_13271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13271_cast_fp16 = slice_by_index(begin = var_13271_begin_0, end = var_13271_end_0, end_mask = var_13271_end_mask_0, x = value_17_cast_fp16)[name = string("op_13271_cast_fp16")]; + tensor var_13275_begin_0 = const()[name = string("op_13275_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13275_end_0 = const()[name = string("op_13275_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_13275_end_mask_0 = const()[name = string("op_13275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13275_cast_fp16 = slice_by_index(begin = var_13275_begin_0, end = var_13275_end_0, end_mask = var_13275_end_mask_0, x = value_17_cast_fp16)[name = string("op_13275_cast_fp16")]; + string _SplitHeadsQ__mh_w_1281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1281_equation_0, values = (var_13121_cast_fp16, var_12563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1281_cast_fp16")]; + string _SplitHeadsQ__mh_w_1283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1283_equation_0, values = (var_13121_cast_fp16, var_12570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1283_cast_fp16")]; + string _SplitHeadsQ__mh_w_1285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1285_equation_0, values = (var_13121_cast_fp16, var_12577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1285_cast_fp16")]; + string _SplitHeadsQ__mh_w_1287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1287_equation_0, values = (var_13121_cast_fp16, var_12584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1287_cast_fp16")]; + string _SplitHeadsQ__mh_w_1289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1289_equation_0, values = (var_13125_cast_fp16, var_12591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1289_cast_fp16")]; + string _SplitHeadsQ__mh_w_1291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1291_equation_0, values = (var_13125_cast_fp16, var_12598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1291_cast_fp16")]; + string _SplitHeadsQ__mh_w_1293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1293_equation_0, values = (var_13125_cast_fp16, var_12605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1293_cast_fp16")]; + string _SplitHeadsQ__mh_w_1295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1295_equation_0, values = (var_13125_cast_fp16, var_12612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1295_cast_fp16")]; + string _SplitHeadsQ__mh_w_1297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1297_equation_0, values = (var_13129_cast_fp16, var_12619_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1297_cast_fp16")]; + string _SplitHeadsQ__mh_w_1299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1299_equation_0, values = (var_13129_cast_fp16, var_12626_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1299_cast_fp16")]; + string _SplitHeadsQ__mh_w_1301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1301_equation_0, values = (var_13129_cast_fp16, var_12633_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1301_cast_fp16")]; + string _SplitHeadsQ__mh_w_1303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1303_equation_0, values = (var_13129_cast_fp16, var_12640_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1303_cast_fp16")]; + string _SplitHeadsQ__mh_w_1305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1305_equation_0, values = (var_13133_cast_fp16, var_12647_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1305_cast_fp16")]; + string _SplitHeadsQ__mh_w_1307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1307_equation_0, values = (var_13133_cast_fp16, var_12654_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1307_cast_fp16")]; + string _SplitHeadsQ__mh_w_1309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1309_equation_0, values = (var_13133_cast_fp16, var_12661_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1309_cast_fp16")]; + string _SplitHeadsQ__mh_w_1311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1311_equation_0, values = (var_13133_cast_fp16, var_12668_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1311_cast_fp16")]; + string _SplitHeadsQ__mh_w_1313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1313_equation_0, values = (var_13137_cast_fp16, var_12675_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1313_cast_fp16")]; + string _SplitHeadsQ__mh_w_1315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1315_equation_0, values = (var_13137_cast_fp16, var_12682_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1315_cast_fp16")]; + string _SplitHeadsQ__mh_w_1317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1317_equation_0, values = (var_13137_cast_fp16, var_12689_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1317_cast_fp16")]; + string _SplitHeadsQ__mh_w_1319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1319_equation_0, values = (var_13137_cast_fp16, var_12696_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1319_cast_fp16")]; + string _SplitHeadsQ__mh_w_1321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1321_equation_0, values = (var_13141_cast_fp16, var_12703_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1321_cast_fp16")]; + string _SplitHeadsQ__mh_w_1323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1323_equation_0, values = (var_13141_cast_fp16, var_12710_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1323_cast_fp16")]; + string _SplitHeadsQ__mh_w_1325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1325_equation_0, values = (var_13141_cast_fp16, var_12717_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1325_cast_fp16")]; + string _SplitHeadsQ__mh_w_1327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1327_equation_0, values = (var_13141_cast_fp16, var_12724_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1327_cast_fp16")]; + string _SplitHeadsQ__mh_w_1329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1329_equation_0, values = (var_13145_cast_fp16, var_12731_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1329_cast_fp16")]; + string _SplitHeadsQ__mh_w_1331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1331_equation_0, values = (var_13145_cast_fp16, var_12738_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1331_cast_fp16")]; + string _SplitHeadsQ__mh_w_1333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1333_equation_0, values = (var_13145_cast_fp16, var_12745_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1333_cast_fp16")]; + string _SplitHeadsQ__mh_w_1335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1335_equation_0, values = (var_13145_cast_fp16, var_12752_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1335_cast_fp16")]; + string _SplitHeadsQ__mh_w_1337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1337_equation_0, values = (var_13149_cast_fp16, var_12759_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1337_cast_fp16")]; + string _SplitHeadsQ__mh_w_1339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1339_equation_0, values = (var_13149_cast_fp16, var_12766_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1339_cast_fp16")]; + string _SplitHeadsQ__mh_w_1341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1341_equation_0, values = (var_13149_cast_fp16, var_12773_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1341_cast_fp16")]; + string _SplitHeadsQ__mh_w_1343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1343_equation_0, values = (var_13149_cast_fp16, var_12780_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1343_cast_fp16")]; + string _SplitHeadsQ__mh_w_1345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1345_equation_0, values = (var_13153_cast_fp16, var_12787_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1345_cast_fp16")]; + string _SplitHeadsQ__mh_w_1347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1347_equation_0, values = (var_13153_cast_fp16, var_12794_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1347_cast_fp16")]; + string _SplitHeadsQ__mh_w_1349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1349_equation_0, values = (var_13153_cast_fp16, var_12801_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1349_cast_fp16")]; + string _SplitHeadsQ__mh_w_1351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1351_equation_0, values = (var_13153_cast_fp16, var_12808_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1351_cast_fp16")]; + string _SplitHeadsQ__mh_w_1353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1353_equation_0, values = (var_13157_cast_fp16, var_12815_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1353_cast_fp16")]; + string _SplitHeadsQ__mh_w_1355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1355_equation_0, values = (var_13157_cast_fp16, var_12822_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1355_cast_fp16")]; + string _SplitHeadsQ__mh_w_1357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1357_equation_0, values = (var_13157_cast_fp16, var_12829_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1357_cast_fp16")]; + string _SplitHeadsQ__mh_w_1359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1359_equation_0, values = (var_13157_cast_fp16, var_12836_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1359_cast_fp16")]; + string _SplitHeadsQ__mh_w_1361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1361_equation_0, values = (var_13161_cast_fp16, var_12843_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1361_cast_fp16")]; + string _SplitHeadsQ__mh_w_1363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1363_equation_0, values = (var_13161_cast_fp16, var_12850_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1363_cast_fp16")]; + string _SplitHeadsQ__mh_w_1365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1365_equation_0, values = (var_13161_cast_fp16, var_12857_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1365_cast_fp16")]; + string _SplitHeadsQ__mh_w_1367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1367_equation_0, values = (var_13161_cast_fp16, var_12864_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1367_cast_fp16")]; + string _SplitHeadsQ__mh_w_1369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1369_equation_0, values = (var_13165_cast_fp16, var_12871_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1369_cast_fp16")]; + string _SplitHeadsQ__mh_w_1371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1371_equation_0, values = (var_13165_cast_fp16, var_12878_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1371_cast_fp16")]; + string _SplitHeadsQ__mh_w_1373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1373_equation_0, values = (var_13165_cast_fp16, var_12885_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1373_cast_fp16")]; + string _SplitHeadsQ__mh_w_1375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1375_equation_0, values = (var_13165_cast_fp16, var_12892_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1375_cast_fp16")]; + string _SplitHeadsQ__mh_w_1377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1377_equation_0, values = (var_13169_cast_fp16, var_12899_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1377_cast_fp16")]; + string _SplitHeadsQ__mh_w_1379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1379_equation_0, values = (var_13169_cast_fp16, var_12906_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1379_cast_fp16")]; + string _SplitHeadsQ__mh_w_1381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1381_equation_0, values = (var_13169_cast_fp16, var_12913_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1381_cast_fp16")]; + string _SplitHeadsQ__mh_w_1383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1383_equation_0, values = (var_13169_cast_fp16, var_12920_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1383_cast_fp16")]; + string _SplitHeadsQ__mh_w_1385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1385_equation_0, values = (var_13173_cast_fp16, var_12927_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1385_cast_fp16")]; + string _SplitHeadsQ__mh_w_1387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1387_equation_0, values = (var_13173_cast_fp16, var_12934_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1387_cast_fp16")]; + string _SplitHeadsQ__mh_w_1389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1389_equation_0, values = (var_13173_cast_fp16, var_12941_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1389_cast_fp16")]; + string _SplitHeadsQ__mh_w_1391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1391_equation_0, values = (var_13173_cast_fp16, var_12948_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1391_cast_fp16")]; + string _SplitHeadsQ__mh_w_1393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1393_equation_0, values = (var_13177_cast_fp16, var_12955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1393_cast_fp16")]; + string _SplitHeadsQ__mh_w_1395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1395_equation_0, values = (var_13177_cast_fp16, var_12962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1395_cast_fp16")]; + string _SplitHeadsQ__mh_w_1397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1397_equation_0, values = (var_13177_cast_fp16, var_12969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1397_cast_fp16")]; + string _SplitHeadsQ__mh_w_1399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1399_equation_0, values = (var_13177_cast_fp16, var_12976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1399_cast_fp16")]; + string _SplitHeadsQ__mh_w_1401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1401_equation_0, values = (var_13181_cast_fp16, var_12983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1401_cast_fp16")]; + string _SplitHeadsQ__mh_w_1403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1403_equation_0, values = (var_13181_cast_fp16, var_12990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1403_cast_fp16")]; + string _SplitHeadsQ__mh_w_1405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1405_equation_0, values = (var_13181_cast_fp16, var_12997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1405_cast_fp16")]; + string _SplitHeadsQ__mh_w_1407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1407_equation_0, values = (var_13181_cast_fp16, var_13004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1407_cast_fp16")]; + string _SplitHeadsQ__mh_w_1409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1409_equation_0, values = (var_13185_cast_fp16, var_13011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1409_cast_fp16")]; + string _SplitHeadsQ__mh_w_1411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1411_equation_0, values = (var_13185_cast_fp16, var_13018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1411_cast_fp16")]; + string _SplitHeadsQ__mh_w_1413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1413_equation_0, values = (var_13185_cast_fp16, var_13025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1413_cast_fp16")]; + string _SplitHeadsQ__mh_w_1415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1415_equation_0, values = (var_13185_cast_fp16, var_13032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1415_cast_fp16")]; + string _SplitHeadsQ__mh_w_1417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1417_equation_0, values = (var_13189_cast_fp16, var_13039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1417_cast_fp16")]; + string _SplitHeadsQ__mh_w_1419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1419_equation_0, values = (var_13189_cast_fp16, var_13046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1419_cast_fp16")]; + string _SplitHeadsQ__mh_w_1421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1421_equation_0, values = (var_13189_cast_fp16, var_13053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1421_cast_fp16")]; + string _SplitHeadsQ__mh_w_1423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1423_equation_0, values = (var_13189_cast_fp16, var_13060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1423_cast_fp16")]; + string _SplitHeadsQ__mh_w_1425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1425_equation_0, values = (var_13193_cast_fp16, var_13067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1425_cast_fp16")]; + string _SplitHeadsQ__mh_w_1427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1427_equation_0, values = (var_13193_cast_fp16, var_13074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1427_cast_fp16")]; + string _SplitHeadsQ__mh_w_1429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1429_equation_0, values = (var_13193_cast_fp16, var_13081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1429_cast_fp16")]; + string _SplitHeadsQ__mh_w_1431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1431_equation_0, values = (var_13193_cast_fp16, var_13088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1431_cast_fp16")]; + string _SplitHeadsQ__mh_w_1433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1433_equation_0, values = (var_13197_cast_fp16, var_13095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1433_cast_fp16")]; + string _SplitHeadsQ__mh_w_1435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1435_equation_0, values = (var_13197_cast_fp16, var_13102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1435_cast_fp16")]; + string _SplitHeadsQ__mh_w_1437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1437_equation_0, values = (var_13197_cast_fp16, var_13109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1437_cast_fp16")]; + string _SplitHeadsQ__mh_w_1439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1439_equation_0, values = (var_13197_cast_fp16, var_13116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1439_cast_fp16")]; + fp16 var_13438_to_fp16 = const()[name = string("op_13438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1281_cast_fp16, y = var_13438_to_fp16)[name = string("aw_chunk_1281_cast_fp16")]; + fp16 var_13440_to_fp16 = const()[name = string("op_13440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1283_cast_fp16, y = var_13440_to_fp16)[name = string("aw_chunk_1283_cast_fp16")]; + fp16 var_13442_to_fp16 = const()[name = string("op_13442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1285_cast_fp16, y = var_13442_to_fp16)[name = string("aw_chunk_1285_cast_fp16")]; + fp16 var_13444_to_fp16 = const()[name = string("op_13444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1287_cast_fp16, y = var_13444_to_fp16)[name = string("aw_chunk_1287_cast_fp16")]; + fp16 var_13446_to_fp16 = const()[name = string("op_13446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1289_cast_fp16, y = var_13446_to_fp16)[name = string("aw_chunk_1289_cast_fp16")]; + fp16 var_13448_to_fp16 = const()[name = string("op_13448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1291_cast_fp16, y = var_13448_to_fp16)[name = string("aw_chunk_1291_cast_fp16")]; + fp16 var_13450_to_fp16 = const()[name = string("op_13450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1293_cast_fp16, y = var_13450_to_fp16)[name = string("aw_chunk_1293_cast_fp16")]; + fp16 var_13452_to_fp16 = const()[name = string("op_13452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1295_cast_fp16, y = var_13452_to_fp16)[name = string("aw_chunk_1295_cast_fp16")]; + fp16 var_13454_to_fp16 = const()[name = string("op_13454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1297_cast_fp16, y = var_13454_to_fp16)[name = string("aw_chunk_1297_cast_fp16")]; + fp16 var_13456_to_fp16 = const()[name = string("op_13456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1299_cast_fp16, y = var_13456_to_fp16)[name = string("aw_chunk_1299_cast_fp16")]; + fp16 var_13458_to_fp16 = const()[name = string("op_13458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1301_cast_fp16, y = var_13458_to_fp16)[name = string("aw_chunk_1301_cast_fp16")]; + fp16 var_13460_to_fp16 = const()[name = string("op_13460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1303_cast_fp16, y = var_13460_to_fp16)[name = string("aw_chunk_1303_cast_fp16")]; + fp16 var_13462_to_fp16 = const()[name = string("op_13462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1305_cast_fp16, y = var_13462_to_fp16)[name = string("aw_chunk_1305_cast_fp16")]; + fp16 var_13464_to_fp16 = const()[name = string("op_13464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1307_cast_fp16, y = var_13464_to_fp16)[name = string("aw_chunk_1307_cast_fp16")]; + fp16 var_13466_to_fp16 = const()[name = string("op_13466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1309_cast_fp16, y = var_13466_to_fp16)[name = string("aw_chunk_1309_cast_fp16")]; + fp16 var_13468_to_fp16 = const()[name = string("op_13468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1311_cast_fp16, y = var_13468_to_fp16)[name = string("aw_chunk_1311_cast_fp16")]; + fp16 var_13470_to_fp16 = const()[name = string("op_13470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1313_cast_fp16, y = var_13470_to_fp16)[name = string("aw_chunk_1313_cast_fp16")]; + fp16 var_13472_to_fp16 = const()[name = string("op_13472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1315_cast_fp16, y = var_13472_to_fp16)[name = string("aw_chunk_1315_cast_fp16")]; + fp16 var_13474_to_fp16 = const()[name = string("op_13474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1317_cast_fp16, y = var_13474_to_fp16)[name = string("aw_chunk_1317_cast_fp16")]; + fp16 var_13476_to_fp16 = const()[name = string("op_13476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1319_cast_fp16, y = var_13476_to_fp16)[name = string("aw_chunk_1319_cast_fp16")]; + fp16 var_13478_to_fp16 = const()[name = string("op_13478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1321_cast_fp16, y = var_13478_to_fp16)[name = string("aw_chunk_1321_cast_fp16")]; + fp16 var_13480_to_fp16 = const()[name = string("op_13480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1323_cast_fp16, y = var_13480_to_fp16)[name = string("aw_chunk_1323_cast_fp16")]; + fp16 var_13482_to_fp16 = const()[name = string("op_13482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1325_cast_fp16, y = var_13482_to_fp16)[name = string("aw_chunk_1325_cast_fp16")]; + fp16 var_13484_to_fp16 = const()[name = string("op_13484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1327_cast_fp16, y = var_13484_to_fp16)[name = string("aw_chunk_1327_cast_fp16")]; + fp16 var_13486_to_fp16 = const()[name = string("op_13486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1329_cast_fp16, y = var_13486_to_fp16)[name = string("aw_chunk_1329_cast_fp16")]; + fp16 var_13488_to_fp16 = const()[name = string("op_13488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1331_cast_fp16, y = var_13488_to_fp16)[name = string("aw_chunk_1331_cast_fp16")]; + fp16 var_13490_to_fp16 = const()[name = string("op_13490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1333_cast_fp16, y = var_13490_to_fp16)[name = string("aw_chunk_1333_cast_fp16")]; + fp16 var_13492_to_fp16 = const()[name = string("op_13492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1335_cast_fp16, y = var_13492_to_fp16)[name = string("aw_chunk_1335_cast_fp16")]; + fp16 var_13494_to_fp16 = const()[name = string("op_13494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1337_cast_fp16, y = var_13494_to_fp16)[name = string("aw_chunk_1337_cast_fp16")]; + fp16 var_13496_to_fp16 = const()[name = string("op_13496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1339_cast_fp16, y = var_13496_to_fp16)[name = string("aw_chunk_1339_cast_fp16")]; + fp16 var_13498_to_fp16 = const()[name = string("op_13498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1341_cast_fp16, y = var_13498_to_fp16)[name = string("aw_chunk_1341_cast_fp16")]; + fp16 var_13500_to_fp16 = const()[name = string("op_13500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1343_cast_fp16, y = var_13500_to_fp16)[name = string("aw_chunk_1343_cast_fp16")]; + fp16 var_13502_to_fp16 = const()[name = string("op_13502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1345_cast_fp16, y = var_13502_to_fp16)[name = string("aw_chunk_1345_cast_fp16")]; + fp16 var_13504_to_fp16 = const()[name = string("op_13504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1347_cast_fp16, y = var_13504_to_fp16)[name = string("aw_chunk_1347_cast_fp16")]; + fp16 var_13506_to_fp16 = const()[name = string("op_13506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1349_cast_fp16, y = var_13506_to_fp16)[name = string("aw_chunk_1349_cast_fp16")]; + fp16 var_13508_to_fp16 = const()[name = string("op_13508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1351_cast_fp16, y = var_13508_to_fp16)[name = string("aw_chunk_1351_cast_fp16")]; + fp16 var_13510_to_fp16 = const()[name = string("op_13510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1353_cast_fp16, y = var_13510_to_fp16)[name = string("aw_chunk_1353_cast_fp16")]; + fp16 var_13512_to_fp16 = const()[name = string("op_13512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1355_cast_fp16, y = var_13512_to_fp16)[name = string("aw_chunk_1355_cast_fp16")]; + fp16 var_13514_to_fp16 = const()[name = string("op_13514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1357_cast_fp16, y = var_13514_to_fp16)[name = string("aw_chunk_1357_cast_fp16")]; + fp16 var_13516_to_fp16 = const()[name = string("op_13516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1359_cast_fp16, y = var_13516_to_fp16)[name = string("aw_chunk_1359_cast_fp16")]; + fp16 var_13518_to_fp16 = const()[name = string("op_13518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1361_cast_fp16, y = var_13518_to_fp16)[name = string("aw_chunk_1361_cast_fp16")]; + fp16 var_13520_to_fp16 = const()[name = string("op_13520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1363_cast_fp16, y = var_13520_to_fp16)[name = string("aw_chunk_1363_cast_fp16")]; + fp16 var_13522_to_fp16 = const()[name = string("op_13522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1365_cast_fp16, y = var_13522_to_fp16)[name = string("aw_chunk_1365_cast_fp16")]; + fp16 var_13524_to_fp16 = const()[name = string("op_13524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1367_cast_fp16, y = var_13524_to_fp16)[name = string("aw_chunk_1367_cast_fp16")]; + fp16 var_13526_to_fp16 = const()[name = string("op_13526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1369_cast_fp16, y = var_13526_to_fp16)[name = string("aw_chunk_1369_cast_fp16")]; + fp16 var_13528_to_fp16 = const()[name = string("op_13528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1371_cast_fp16, y = var_13528_to_fp16)[name = string("aw_chunk_1371_cast_fp16")]; + fp16 var_13530_to_fp16 = const()[name = string("op_13530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1373_cast_fp16, y = var_13530_to_fp16)[name = string("aw_chunk_1373_cast_fp16")]; + fp16 var_13532_to_fp16 = const()[name = string("op_13532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1375_cast_fp16, y = var_13532_to_fp16)[name = string("aw_chunk_1375_cast_fp16")]; + fp16 var_13534_to_fp16 = const()[name = string("op_13534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1377_cast_fp16, y = var_13534_to_fp16)[name = string("aw_chunk_1377_cast_fp16")]; + fp16 var_13536_to_fp16 = const()[name = string("op_13536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1379_cast_fp16, y = var_13536_to_fp16)[name = string("aw_chunk_1379_cast_fp16")]; + fp16 var_13538_to_fp16 = const()[name = string("op_13538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1381_cast_fp16, y = var_13538_to_fp16)[name = string("aw_chunk_1381_cast_fp16")]; + fp16 var_13540_to_fp16 = const()[name = string("op_13540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1383_cast_fp16, y = var_13540_to_fp16)[name = string("aw_chunk_1383_cast_fp16")]; + fp16 var_13542_to_fp16 = const()[name = string("op_13542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1385_cast_fp16, y = var_13542_to_fp16)[name = string("aw_chunk_1385_cast_fp16")]; + fp16 var_13544_to_fp16 = const()[name = string("op_13544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1387_cast_fp16, y = var_13544_to_fp16)[name = string("aw_chunk_1387_cast_fp16")]; + fp16 var_13546_to_fp16 = const()[name = string("op_13546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1389_cast_fp16, y = var_13546_to_fp16)[name = string("aw_chunk_1389_cast_fp16")]; + fp16 var_13548_to_fp16 = const()[name = string("op_13548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1391_cast_fp16, y = var_13548_to_fp16)[name = string("aw_chunk_1391_cast_fp16")]; + fp16 var_13550_to_fp16 = const()[name = string("op_13550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1393_cast_fp16, y = var_13550_to_fp16)[name = string("aw_chunk_1393_cast_fp16")]; + fp16 var_13552_to_fp16 = const()[name = string("op_13552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1395_cast_fp16, y = var_13552_to_fp16)[name = string("aw_chunk_1395_cast_fp16")]; + fp16 var_13554_to_fp16 = const()[name = string("op_13554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1397_cast_fp16, y = var_13554_to_fp16)[name = string("aw_chunk_1397_cast_fp16")]; + fp16 var_13556_to_fp16 = const()[name = string("op_13556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1399_cast_fp16, y = var_13556_to_fp16)[name = string("aw_chunk_1399_cast_fp16")]; + fp16 var_13558_to_fp16 = const()[name = string("op_13558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1401_cast_fp16, y = var_13558_to_fp16)[name = string("aw_chunk_1401_cast_fp16")]; + fp16 var_13560_to_fp16 = const()[name = string("op_13560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1403_cast_fp16, y = var_13560_to_fp16)[name = string("aw_chunk_1403_cast_fp16")]; + fp16 var_13562_to_fp16 = const()[name = string("op_13562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1405_cast_fp16, y = var_13562_to_fp16)[name = string("aw_chunk_1405_cast_fp16")]; + fp16 var_13564_to_fp16 = const()[name = string("op_13564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1407_cast_fp16, y = var_13564_to_fp16)[name = string("aw_chunk_1407_cast_fp16")]; + fp16 var_13566_to_fp16 = const()[name = string("op_13566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1409_cast_fp16, y = var_13566_to_fp16)[name = string("aw_chunk_1409_cast_fp16")]; + fp16 var_13568_to_fp16 = const()[name = string("op_13568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1411_cast_fp16, y = var_13568_to_fp16)[name = string("aw_chunk_1411_cast_fp16")]; + fp16 var_13570_to_fp16 = const()[name = string("op_13570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1413_cast_fp16, y = var_13570_to_fp16)[name = string("aw_chunk_1413_cast_fp16")]; + fp16 var_13572_to_fp16 = const()[name = string("op_13572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1415_cast_fp16, y = var_13572_to_fp16)[name = string("aw_chunk_1415_cast_fp16")]; + fp16 var_13574_to_fp16 = const()[name = string("op_13574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1417_cast_fp16, y = var_13574_to_fp16)[name = string("aw_chunk_1417_cast_fp16")]; + fp16 var_13576_to_fp16 = const()[name = string("op_13576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1419_cast_fp16, y = var_13576_to_fp16)[name = string("aw_chunk_1419_cast_fp16")]; + fp16 var_13578_to_fp16 = const()[name = string("op_13578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1421_cast_fp16, y = var_13578_to_fp16)[name = string("aw_chunk_1421_cast_fp16")]; + fp16 var_13580_to_fp16 = const()[name = string("op_13580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1423_cast_fp16, y = var_13580_to_fp16)[name = string("aw_chunk_1423_cast_fp16")]; + fp16 var_13582_to_fp16 = const()[name = string("op_13582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1425_cast_fp16, y = var_13582_to_fp16)[name = string("aw_chunk_1425_cast_fp16")]; + fp16 var_13584_to_fp16 = const()[name = string("op_13584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1427_cast_fp16, y = var_13584_to_fp16)[name = string("aw_chunk_1427_cast_fp16")]; + fp16 var_13586_to_fp16 = const()[name = string("op_13586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1429_cast_fp16, y = var_13586_to_fp16)[name = string("aw_chunk_1429_cast_fp16")]; + fp16 var_13588_to_fp16 = const()[name = string("op_13588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1431_cast_fp16, y = var_13588_to_fp16)[name = string("aw_chunk_1431_cast_fp16")]; + fp16 var_13590_to_fp16 = const()[name = string("op_13590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1433_cast_fp16, y = var_13590_to_fp16)[name = string("aw_chunk_1433_cast_fp16")]; + fp16 var_13592_to_fp16 = const()[name = string("op_13592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1435_cast_fp16, y = var_13592_to_fp16)[name = string("aw_chunk_1435_cast_fp16")]; + fp16 var_13594_to_fp16 = const()[name = string("op_13594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1437_cast_fp16, y = var_13594_to_fp16)[name = string("aw_chunk_1437_cast_fp16")]; + fp16 var_13596_to_fp16 = const()[name = string("op_13596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1439_cast_fp16, y = var_13596_to_fp16)[name = string("aw_chunk_1439_cast_fp16")]; + tensor var_13598_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1281_cast_fp16)[name = string("op_13598_cast_fp16")]; + tensor var_13599_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1283_cast_fp16)[name = string("op_13599_cast_fp16")]; + tensor var_13600_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1285_cast_fp16)[name = string("op_13600_cast_fp16")]; + tensor var_13601_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1287_cast_fp16)[name = string("op_13601_cast_fp16")]; + tensor var_13602_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1289_cast_fp16)[name = string("op_13602_cast_fp16")]; + tensor var_13603_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1291_cast_fp16)[name = string("op_13603_cast_fp16")]; + tensor var_13604_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1293_cast_fp16)[name = string("op_13604_cast_fp16")]; + tensor var_13605_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1295_cast_fp16)[name = string("op_13605_cast_fp16")]; + tensor var_13606_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1297_cast_fp16)[name = string("op_13606_cast_fp16")]; + tensor var_13607_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1299_cast_fp16)[name = string("op_13607_cast_fp16")]; + tensor var_13608_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1301_cast_fp16)[name = string("op_13608_cast_fp16")]; + tensor var_13609_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1303_cast_fp16)[name = string("op_13609_cast_fp16")]; + tensor var_13610_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1305_cast_fp16)[name = string("op_13610_cast_fp16")]; + tensor var_13611_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1307_cast_fp16)[name = string("op_13611_cast_fp16")]; + tensor var_13612_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1309_cast_fp16)[name = string("op_13612_cast_fp16")]; + tensor var_13613_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1311_cast_fp16)[name = string("op_13613_cast_fp16")]; + tensor var_13614_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1313_cast_fp16)[name = string("op_13614_cast_fp16")]; + tensor var_13615_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1315_cast_fp16)[name = string("op_13615_cast_fp16")]; + tensor var_13616_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1317_cast_fp16)[name = string("op_13616_cast_fp16")]; + tensor var_13617_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1319_cast_fp16)[name = string("op_13617_cast_fp16")]; + tensor var_13618_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1321_cast_fp16)[name = string("op_13618_cast_fp16")]; + tensor var_13619_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1323_cast_fp16)[name = string("op_13619_cast_fp16")]; + tensor var_13620_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1325_cast_fp16)[name = string("op_13620_cast_fp16")]; + tensor var_13621_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1327_cast_fp16)[name = string("op_13621_cast_fp16")]; + tensor var_13622_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1329_cast_fp16)[name = string("op_13622_cast_fp16")]; + tensor var_13623_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1331_cast_fp16)[name = string("op_13623_cast_fp16")]; + tensor var_13624_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1333_cast_fp16)[name = string("op_13624_cast_fp16")]; + tensor var_13625_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1335_cast_fp16)[name = string("op_13625_cast_fp16")]; + tensor var_13626_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1337_cast_fp16)[name = string("op_13626_cast_fp16")]; + tensor var_13627_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1339_cast_fp16)[name = string("op_13627_cast_fp16")]; + tensor var_13628_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1341_cast_fp16)[name = string("op_13628_cast_fp16")]; + tensor var_13629_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1343_cast_fp16)[name = string("op_13629_cast_fp16")]; + tensor var_13630_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1345_cast_fp16)[name = string("op_13630_cast_fp16")]; + tensor var_13631_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1347_cast_fp16)[name = string("op_13631_cast_fp16")]; + tensor var_13632_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1349_cast_fp16)[name = string("op_13632_cast_fp16")]; + tensor var_13633_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1351_cast_fp16)[name = string("op_13633_cast_fp16")]; + tensor var_13634_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1353_cast_fp16)[name = string("op_13634_cast_fp16")]; + tensor var_13635_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1355_cast_fp16)[name = string("op_13635_cast_fp16")]; + tensor var_13636_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1357_cast_fp16)[name = string("op_13636_cast_fp16")]; + tensor var_13637_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1359_cast_fp16)[name = string("op_13637_cast_fp16")]; + tensor var_13638_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1361_cast_fp16)[name = string("op_13638_cast_fp16")]; + tensor var_13639_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1363_cast_fp16)[name = string("op_13639_cast_fp16")]; + tensor var_13640_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1365_cast_fp16)[name = string("op_13640_cast_fp16")]; + tensor var_13641_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1367_cast_fp16)[name = string("op_13641_cast_fp16")]; + tensor var_13642_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1369_cast_fp16)[name = string("op_13642_cast_fp16")]; + tensor var_13643_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1371_cast_fp16)[name = string("op_13643_cast_fp16")]; + tensor var_13644_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1373_cast_fp16)[name = string("op_13644_cast_fp16")]; + tensor var_13645_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1375_cast_fp16)[name = string("op_13645_cast_fp16")]; + tensor var_13646_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1377_cast_fp16)[name = string("op_13646_cast_fp16")]; + tensor var_13647_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1379_cast_fp16)[name = string("op_13647_cast_fp16")]; + tensor var_13648_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1381_cast_fp16)[name = string("op_13648_cast_fp16")]; + tensor var_13649_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1383_cast_fp16)[name = string("op_13649_cast_fp16")]; + tensor var_13650_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1385_cast_fp16)[name = string("op_13650_cast_fp16")]; + tensor var_13651_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1387_cast_fp16)[name = string("op_13651_cast_fp16")]; + tensor var_13652_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1389_cast_fp16)[name = string("op_13652_cast_fp16")]; + tensor var_13653_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1391_cast_fp16)[name = string("op_13653_cast_fp16")]; + tensor var_13654_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1393_cast_fp16)[name = string("op_13654_cast_fp16")]; + tensor var_13655_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1395_cast_fp16)[name = string("op_13655_cast_fp16")]; + tensor var_13656_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1397_cast_fp16)[name = string("op_13656_cast_fp16")]; + tensor var_13657_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1399_cast_fp16)[name = string("op_13657_cast_fp16")]; + tensor var_13658_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1401_cast_fp16)[name = string("op_13658_cast_fp16")]; + tensor var_13659_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1403_cast_fp16)[name = string("op_13659_cast_fp16")]; + tensor var_13660_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1405_cast_fp16)[name = string("op_13660_cast_fp16")]; + tensor var_13661_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1407_cast_fp16)[name = string("op_13661_cast_fp16")]; + tensor var_13662_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1409_cast_fp16)[name = string("op_13662_cast_fp16")]; + tensor var_13663_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1411_cast_fp16)[name = string("op_13663_cast_fp16")]; + tensor var_13664_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1413_cast_fp16)[name = string("op_13664_cast_fp16")]; + tensor var_13665_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1415_cast_fp16)[name = string("op_13665_cast_fp16")]; + tensor var_13666_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1417_cast_fp16)[name = string("op_13666_cast_fp16")]; + tensor var_13667_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1419_cast_fp16)[name = string("op_13667_cast_fp16")]; + tensor var_13668_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1421_cast_fp16)[name = string("op_13668_cast_fp16")]; + tensor var_13669_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1423_cast_fp16)[name = string("op_13669_cast_fp16")]; + tensor var_13670_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1425_cast_fp16)[name = string("op_13670_cast_fp16")]; + tensor var_13671_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1427_cast_fp16)[name = string("op_13671_cast_fp16")]; + tensor var_13672_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1429_cast_fp16)[name = string("op_13672_cast_fp16")]; + tensor var_13673_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1431_cast_fp16)[name = string("op_13673_cast_fp16")]; + tensor var_13674_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1433_cast_fp16)[name = string("op_13674_cast_fp16")]; + tensor var_13675_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1435_cast_fp16)[name = string("op_13675_cast_fp16")]; + tensor var_13676_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1437_cast_fp16)[name = string("op_13676_cast_fp16")]; + tensor var_13677_cast_fp16 = softmax(axis = var_12423, x = aw_chunk_1439_cast_fp16)[name = string("op_13677_cast_fp16")]; + string var_13679_equation_0 = const()[name = string("op_13679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13679_cast_fp16 = einsum(equation = var_13679_equation_0, values = (var_13199_cast_fp16, var_13598_cast_fp16))[name = string("op_13679_cast_fp16")]; + string var_13681_equation_0 = const()[name = string("op_13681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13681_cast_fp16 = einsum(equation = var_13681_equation_0, values = (var_13199_cast_fp16, var_13599_cast_fp16))[name = string("op_13681_cast_fp16")]; + string var_13683_equation_0 = const()[name = string("op_13683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13683_cast_fp16 = einsum(equation = var_13683_equation_0, values = (var_13199_cast_fp16, var_13600_cast_fp16))[name = string("op_13683_cast_fp16")]; + string var_13685_equation_0 = const()[name = string("op_13685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13685_cast_fp16 = einsum(equation = var_13685_equation_0, values = (var_13199_cast_fp16, var_13601_cast_fp16))[name = string("op_13685_cast_fp16")]; + string var_13687_equation_0 = const()[name = string("op_13687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13687_cast_fp16 = einsum(equation = var_13687_equation_0, values = (var_13203_cast_fp16, var_13602_cast_fp16))[name = string("op_13687_cast_fp16")]; + string var_13689_equation_0 = const()[name = string("op_13689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13689_cast_fp16 = einsum(equation = var_13689_equation_0, values = (var_13203_cast_fp16, var_13603_cast_fp16))[name = string("op_13689_cast_fp16")]; + string var_13691_equation_0 = const()[name = string("op_13691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13691_cast_fp16 = einsum(equation = var_13691_equation_0, values = (var_13203_cast_fp16, var_13604_cast_fp16))[name = string("op_13691_cast_fp16")]; + string var_13693_equation_0 = const()[name = string("op_13693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13693_cast_fp16 = einsum(equation = var_13693_equation_0, values = (var_13203_cast_fp16, var_13605_cast_fp16))[name = string("op_13693_cast_fp16")]; + string var_13695_equation_0 = const()[name = string("op_13695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13695_cast_fp16 = einsum(equation = var_13695_equation_0, values = (var_13207_cast_fp16, var_13606_cast_fp16))[name = string("op_13695_cast_fp16")]; + string var_13697_equation_0 = const()[name = string("op_13697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13697_cast_fp16 = einsum(equation = var_13697_equation_0, values = (var_13207_cast_fp16, var_13607_cast_fp16))[name = string("op_13697_cast_fp16")]; + string var_13699_equation_0 = const()[name = string("op_13699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13699_cast_fp16 = einsum(equation = var_13699_equation_0, values = (var_13207_cast_fp16, var_13608_cast_fp16))[name = string("op_13699_cast_fp16")]; + string var_13701_equation_0 = const()[name = string("op_13701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13701_cast_fp16 = einsum(equation = var_13701_equation_0, values = (var_13207_cast_fp16, var_13609_cast_fp16))[name = string("op_13701_cast_fp16")]; + string var_13703_equation_0 = const()[name = string("op_13703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13703_cast_fp16 = einsum(equation = var_13703_equation_0, values = (var_13211_cast_fp16, var_13610_cast_fp16))[name = string("op_13703_cast_fp16")]; + string var_13705_equation_0 = const()[name = string("op_13705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13705_cast_fp16 = einsum(equation = var_13705_equation_0, values = (var_13211_cast_fp16, var_13611_cast_fp16))[name = string("op_13705_cast_fp16")]; + string var_13707_equation_0 = const()[name = string("op_13707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13707_cast_fp16 = einsum(equation = var_13707_equation_0, values = (var_13211_cast_fp16, var_13612_cast_fp16))[name = string("op_13707_cast_fp16")]; + string var_13709_equation_0 = const()[name = string("op_13709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13709_cast_fp16 = einsum(equation = var_13709_equation_0, values = (var_13211_cast_fp16, var_13613_cast_fp16))[name = string("op_13709_cast_fp16")]; + string var_13711_equation_0 = const()[name = string("op_13711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13711_cast_fp16 = einsum(equation = var_13711_equation_0, values = (var_13215_cast_fp16, var_13614_cast_fp16))[name = string("op_13711_cast_fp16")]; + string var_13713_equation_0 = const()[name = string("op_13713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13713_cast_fp16 = einsum(equation = var_13713_equation_0, values = (var_13215_cast_fp16, var_13615_cast_fp16))[name = string("op_13713_cast_fp16")]; + string var_13715_equation_0 = const()[name = string("op_13715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13715_cast_fp16 = einsum(equation = var_13715_equation_0, values = (var_13215_cast_fp16, var_13616_cast_fp16))[name = string("op_13715_cast_fp16")]; + string var_13717_equation_0 = const()[name = string("op_13717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13717_cast_fp16 = einsum(equation = var_13717_equation_0, values = (var_13215_cast_fp16, var_13617_cast_fp16))[name = string("op_13717_cast_fp16")]; + string var_13719_equation_0 = const()[name = string("op_13719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13719_cast_fp16 = einsum(equation = var_13719_equation_0, values = (var_13219_cast_fp16, var_13618_cast_fp16))[name = string("op_13719_cast_fp16")]; + string var_13721_equation_0 = const()[name = string("op_13721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13721_cast_fp16 = einsum(equation = var_13721_equation_0, values = (var_13219_cast_fp16, var_13619_cast_fp16))[name = string("op_13721_cast_fp16")]; + string var_13723_equation_0 = const()[name = string("op_13723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13723_cast_fp16 = einsum(equation = var_13723_equation_0, values = (var_13219_cast_fp16, var_13620_cast_fp16))[name = string("op_13723_cast_fp16")]; + string var_13725_equation_0 = const()[name = string("op_13725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13725_cast_fp16 = einsum(equation = var_13725_equation_0, values = (var_13219_cast_fp16, var_13621_cast_fp16))[name = string("op_13725_cast_fp16")]; + string var_13727_equation_0 = const()[name = string("op_13727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13727_cast_fp16 = einsum(equation = var_13727_equation_0, values = (var_13223_cast_fp16, var_13622_cast_fp16))[name = string("op_13727_cast_fp16")]; + string var_13729_equation_0 = const()[name = string("op_13729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13729_cast_fp16 = einsum(equation = var_13729_equation_0, values = (var_13223_cast_fp16, var_13623_cast_fp16))[name = string("op_13729_cast_fp16")]; + string var_13731_equation_0 = const()[name = string("op_13731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13731_cast_fp16 = einsum(equation = var_13731_equation_0, values = (var_13223_cast_fp16, var_13624_cast_fp16))[name = string("op_13731_cast_fp16")]; + string var_13733_equation_0 = const()[name = string("op_13733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13733_cast_fp16 = einsum(equation = var_13733_equation_0, values = (var_13223_cast_fp16, var_13625_cast_fp16))[name = string("op_13733_cast_fp16")]; + string var_13735_equation_0 = const()[name = string("op_13735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13735_cast_fp16 = einsum(equation = var_13735_equation_0, values = (var_13227_cast_fp16, var_13626_cast_fp16))[name = string("op_13735_cast_fp16")]; + string var_13737_equation_0 = const()[name = string("op_13737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13737_cast_fp16 = einsum(equation = var_13737_equation_0, values = (var_13227_cast_fp16, var_13627_cast_fp16))[name = string("op_13737_cast_fp16")]; + string var_13739_equation_0 = const()[name = string("op_13739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13739_cast_fp16 = einsum(equation = var_13739_equation_0, values = (var_13227_cast_fp16, var_13628_cast_fp16))[name = string("op_13739_cast_fp16")]; + string var_13741_equation_0 = const()[name = string("op_13741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13741_cast_fp16 = einsum(equation = var_13741_equation_0, values = (var_13227_cast_fp16, var_13629_cast_fp16))[name = string("op_13741_cast_fp16")]; + string var_13743_equation_0 = const()[name = string("op_13743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13743_cast_fp16 = einsum(equation = var_13743_equation_0, values = (var_13231_cast_fp16, var_13630_cast_fp16))[name = string("op_13743_cast_fp16")]; + string var_13745_equation_0 = const()[name = string("op_13745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13745_cast_fp16 = einsum(equation = var_13745_equation_0, values = (var_13231_cast_fp16, var_13631_cast_fp16))[name = string("op_13745_cast_fp16")]; + string var_13747_equation_0 = const()[name = string("op_13747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13747_cast_fp16 = einsum(equation = var_13747_equation_0, values = (var_13231_cast_fp16, var_13632_cast_fp16))[name = string("op_13747_cast_fp16")]; + string var_13749_equation_0 = const()[name = string("op_13749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13749_cast_fp16 = einsum(equation = var_13749_equation_0, values = (var_13231_cast_fp16, var_13633_cast_fp16))[name = string("op_13749_cast_fp16")]; + string var_13751_equation_0 = const()[name = string("op_13751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13751_cast_fp16 = einsum(equation = var_13751_equation_0, values = (var_13235_cast_fp16, var_13634_cast_fp16))[name = string("op_13751_cast_fp16")]; + string var_13753_equation_0 = const()[name = string("op_13753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13753_cast_fp16 = einsum(equation = var_13753_equation_0, values = (var_13235_cast_fp16, var_13635_cast_fp16))[name = string("op_13753_cast_fp16")]; + string var_13755_equation_0 = const()[name = string("op_13755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13755_cast_fp16 = einsum(equation = var_13755_equation_0, values = (var_13235_cast_fp16, var_13636_cast_fp16))[name = string("op_13755_cast_fp16")]; + string var_13757_equation_0 = const()[name = string("op_13757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13757_cast_fp16 = einsum(equation = var_13757_equation_0, values = (var_13235_cast_fp16, var_13637_cast_fp16))[name = string("op_13757_cast_fp16")]; + string var_13759_equation_0 = const()[name = string("op_13759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13759_cast_fp16 = einsum(equation = var_13759_equation_0, values = (var_13239_cast_fp16, var_13638_cast_fp16))[name = string("op_13759_cast_fp16")]; + string var_13761_equation_0 = const()[name = string("op_13761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13761_cast_fp16 = einsum(equation = var_13761_equation_0, values = (var_13239_cast_fp16, var_13639_cast_fp16))[name = string("op_13761_cast_fp16")]; + string var_13763_equation_0 = const()[name = string("op_13763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13763_cast_fp16 = einsum(equation = var_13763_equation_0, values = (var_13239_cast_fp16, var_13640_cast_fp16))[name = string("op_13763_cast_fp16")]; + string var_13765_equation_0 = const()[name = string("op_13765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13765_cast_fp16 = einsum(equation = var_13765_equation_0, values = (var_13239_cast_fp16, var_13641_cast_fp16))[name = string("op_13765_cast_fp16")]; + string var_13767_equation_0 = const()[name = string("op_13767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13767_cast_fp16 = einsum(equation = var_13767_equation_0, values = (var_13243_cast_fp16, var_13642_cast_fp16))[name = string("op_13767_cast_fp16")]; + string var_13769_equation_0 = const()[name = string("op_13769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13769_cast_fp16 = einsum(equation = var_13769_equation_0, values = (var_13243_cast_fp16, var_13643_cast_fp16))[name = string("op_13769_cast_fp16")]; + string var_13771_equation_0 = const()[name = string("op_13771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13771_cast_fp16 = einsum(equation = var_13771_equation_0, values = (var_13243_cast_fp16, var_13644_cast_fp16))[name = string("op_13771_cast_fp16")]; + string var_13773_equation_0 = const()[name = string("op_13773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13773_cast_fp16 = einsum(equation = var_13773_equation_0, values = (var_13243_cast_fp16, var_13645_cast_fp16))[name = string("op_13773_cast_fp16")]; + string var_13775_equation_0 = const()[name = string("op_13775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13775_cast_fp16 = einsum(equation = var_13775_equation_0, values = (var_13247_cast_fp16, var_13646_cast_fp16))[name = string("op_13775_cast_fp16")]; + string var_13777_equation_0 = const()[name = string("op_13777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13777_cast_fp16 = einsum(equation = var_13777_equation_0, values = (var_13247_cast_fp16, var_13647_cast_fp16))[name = string("op_13777_cast_fp16")]; + string var_13779_equation_0 = const()[name = string("op_13779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13779_cast_fp16 = einsum(equation = var_13779_equation_0, values = (var_13247_cast_fp16, var_13648_cast_fp16))[name = string("op_13779_cast_fp16")]; + string var_13781_equation_0 = const()[name = string("op_13781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13781_cast_fp16 = einsum(equation = var_13781_equation_0, values = (var_13247_cast_fp16, var_13649_cast_fp16))[name = string("op_13781_cast_fp16")]; + string var_13783_equation_0 = const()[name = string("op_13783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13783_cast_fp16 = einsum(equation = var_13783_equation_0, values = (var_13251_cast_fp16, var_13650_cast_fp16))[name = string("op_13783_cast_fp16")]; + string var_13785_equation_0 = const()[name = string("op_13785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13785_cast_fp16 = einsum(equation = var_13785_equation_0, values = (var_13251_cast_fp16, var_13651_cast_fp16))[name = string("op_13785_cast_fp16")]; + string var_13787_equation_0 = const()[name = string("op_13787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13787_cast_fp16 = einsum(equation = var_13787_equation_0, values = (var_13251_cast_fp16, var_13652_cast_fp16))[name = string("op_13787_cast_fp16")]; + string var_13789_equation_0 = const()[name = string("op_13789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13789_cast_fp16 = einsum(equation = var_13789_equation_0, values = (var_13251_cast_fp16, var_13653_cast_fp16))[name = string("op_13789_cast_fp16")]; + string var_13791_equation_0 = const()[name = string("op_13791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13791_cast_fp16 = einsum(equation = var_13791_equation_0, values = (var_13255_cast_fp16, var_13654_cast_fp16))[name = string("op_13791_cast_fp16")]; + string var_13793_equation_0 = const()[name = string("op_13793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13793_cast_fp16 = einsum(equation = var_13793_equation_0, values = (var_13255_cast_fp16, var_13655_cast_fp16))[name = string("op_13793_cast_fp16")]; + string var_13795_equation_0 = const()[name = string("op_13795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13795_cast_fp16 = einsum(equation = var_13795_equation_0, values = (var_13255_cast_fp16, var_13656_cast_fp16))[name = string("op_13795_cast_fp16")]; + string var_13797_equation_0 = const()[name = string("op_13797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13797_cast_fp16 = einsum(equation = var_13797_equation_0, values = (var_13255_cast_fp16, var_13657_cast_fp16))[name = string("op_13797_cast_fp16")]; + string var_13799_equation_0 = const()[name = string("op_13799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13799_cast_fp16 = einsum(equation = var_13799_equation_0, values = (var_13259_cast_fp16, var_13658_cast_fp16))[name = string("op_13799_cast_fp16")]; + string var_13801_equation_0 = const()[name = string("op_13801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13801_cast_fp16 = einsum(equation = var_13801_equation_0, values = (var_13259_cast_fp16, var_13659_cast_fp16))[name = string("op_13801_cast_fp16")]; + string var_13803_equation_0 = const()[name = string("op_13803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13803_cast_fp16 = einsum(equation = var_13803_equation_0, values = (var_13259_cast_fp16, var_13660_cast_fp16))[name = string("op_13803_cast_fp16")]; + string var_13805_equation_0 = const()[name = string("op_13805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13805_cast_fp16 = einsum(equation = var_13805_equation_0, values = (var_13259_cast_fp16, var_13661_cast_fp16))[name = string("op_13805_cast_fp16")]; + string var_13807_equation_0 = const()[name = string("op_13807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13807_cast_fp16 = einsum(equation = var_13807_equation_0, values = (var_13263_cast_fp16, var_13662_cast_fp16))[name = string("op_13807_cast_fp16")]; + string var_13809_equation_0 = const()[name = string("op_13809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13809_cast_fp16 = einsum(equation = var_13809_equation_0, values = (var_13263_cast_fp16, var_13663_cast_fp16))[name = string("op_13809_cast_fp16")]; + string var_13811_equation_0 = const()[name = string("op_13811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13811_cast_fp16 = einsum(equation = var_13811_equation_0, values = (var_13263_cast_fp16, var_13664_cast_fp16))[name = string("op_13811_cast_fp16")]; + string var_13813_equation_0 = const()[name = string("op_13813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13813_cast_fp16 = einsum(equation = var_13813_equation_0, values = (var_13263_cast_fp16, var_13665_cast_fp16))[name = string("op_13813_cast_fp16")]; + string var_13815_equation_0 = const()[name = string("op_13815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13815_cast_fp16 = einsum(equation = var_13815_equation_0, values = (var_13267_cast_fp16, var_13666_cast_fp16))[name = string("op_13815_cast_fp16")]; + string var_13817_equation_0 = const()[name = string("op_13817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13817_cast_fp16 = einsum(equation = var_13817_equation_0, values = (var_13267_cast_fp16, var_13667_cast_fp16))[name = string("op_13817_cast_fp16")]; + string var_13819_equation_0 = const()[name = string("op_13819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13819_cast_fp16 = einsum(equation = var_13819_equation_0, values = (var_13267_cast_fp16, var_13668_cast_fp16))[name = string("op_13819_cast_fp16")]; + string var_13821_equation_0 = const()[name = string("op_13821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13821_cast_fp16 = einsum(equation = var_13821_equation_0, values = (var_13267_cast_fp16, var_13669_cast_fp16))[name = string("op_13821_cast_fp16")]; + string var_13823_equation_0 = const()[name = string("op_13823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13823_cast_fp16 = einsum(equation = var_13823_equation_0, values = (var_13271_cast_fp16, var_13670_cast_fp16))[name = string("op_13823_cast_fp16")]; + string var_13825_equation_0 = const()[name = string("op_13825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13825_cast_fp16 = einsum(equation = var_13825_equation_0, values = (var_13271_cast_fp16, var_13671_cast_fp16))[name = string("op_13825_cast_fp16")]; + string var_13827_equation_0 = const()[name = string("op_13827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13827_cast_fp16 = einsum(equation = var_13827_equation_0, values = (var_13271_cast_fp16, var_13672_cast_fp16))[name = string("op_13827_cast_fp16")]; + string var_13829_equation_0 = const()[name = string("op_13829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13829_cast_fp16 = einsum(equation = var_13829_equation_0, values = (var_13271_cast_fp16, var_13673_cast_fp16))[name = string("op_13829_cast_fp16")]; + string var_13831_equation_0 = const()[name = string("op_13831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13831_cast_fp16 = einsum(equation = var_13831_equation_0, values = (var_13275_cast_fp16, var_13674_cast_fp16))[name = string("op_13831_cast_fp16")]; + string var_13833_equation_0 = const()[name = string("op_13833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13833_cast_fp16 = einsum(equation = var_13833_equation_0, values = (var_13275_cast_fp16, var_13675_cast_fp16))[name = string("op_13833_cast_fp16")]; + string var_13835_equation_0 = const()[name = string("op_13835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13835_cast_fp16 = einsum(equation = var_13835_equation_0, values = (var_13275_cast_fp16, var_13676_cast_fp16))[name = string("op_13835_cast_fp16")]; + string var_13837_equation_0 = const()[name = string("op_13837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_13837_cast_fp16 = einsum(equation = var_13837_equation_0, values = (var_13275_cast_fp16, var_13677_cast_fp16))[name = string("op_13837_cast_fp16")]; + bool var_13839_interleave_0 = const()[name = string("op_13839_interleave_0"), val = bool(false)]; + tensor var_13839_cast_fp16 = concat(axis = var_12398, interleave = var_13839_interleave_0, values = (var_13679_cast_fp16, var_13681_cast_fp16, var_13683_cast_fp16, var_13685_cast_fp16))[name = string("op_13839_cast_fp16")]; + bool var_13841_interleave_0 = const()[name = string("op_13841_interleave_0"), val = bool(false)]; + tensor var_13841_cast_fp16 = concat(axis = var_12398, interleave = var_13841_interleave_0, values = (var_13687_cast_fp16, var_13689_cast_fp16, var_13691_cast_fp16, var_13693_cast_fp16))[name = string("op_13841_cast_fp16")]; + bool var_13843_interleave_0 = const()[name = string("op_13843_interleave_0"), val = bool(false)]; + tensor var_13843_cast_fp16 = concat(axis = var_12398, interleave = var_13843_interleave_0, values = (var_13695_cast_fp16, var_13697_cast_fp16, var_13699_cast_fp16, var_13701_cast_fp16))[name = string("op_13843_cast_fp16")]; + bool var_13845_interleave_0 = const()[name = string("op_13845_interleave_0"), val = bool(false)]; + tensor var_13845_cast_fp16 = concat(axis = var_12398, interleave = var_13845_interleave_0, values = (var_13703_cast_fp16, var_13705_cast_fp16, var_13707_cast_fp16, var_13709_cast_fp16))[name = string("op_13845_cast_fp16")]; + bool var_13847_interleave_0 = const()[name = string("op_13847_interleave_0"), val = bool(false)]; + tensor var_13847_cast_fp16 = concat(axis = var_12398, interleave = var_13847_interleave_0, values = (var_13711_cast_fp16, var_13713_cast_fp16, var_13715_cast_fp16, var_13717_cast_fp16))[name = string("op_13847_cast_fp16")]; + bool var_13849_interleave_0 = const()[name = string("op_13849_interleave_0"), val = bool(false)]; + tensor var_13849_cast_fp16 = concat(axis = var_12398, interleave = var_13849_interleave_0, values = (var_13719_cast_fp16, var_13721_cast_fp16, var_13723_cast_fp16, var_13725_cast_fp16))[name = string("op_13849_cast_fp16")]; + bool var_13851_interleave_0 = const()[name = string("op_13851_interleave_0"), val = bool(false)]; + tensor var_13851_cast_fp16 = concat(axis = var_12398, interleave = var_13851_interleave_0, values = (var_13727_cast_fp16, var_13729_cast_fp16, var_13731_cast_fp16, var_13733_cast_fp16))[name = string("op_13851_cast_fp16")]; + bool var_13853_interleave_0 = const()[name = string("op_13853_interleave_0"), val = bool(false)]; + tensor var_13853_cast_fp16 = concat(axis = var_12398, interleave = var_13853_interleave_0, values = (var_13735_cast_fp16, var_13737_cast_fp16, var_13739_cast_fp16, var_13741_cast_fp16))[name = string("op_13853_cast_fp16")]; + bool var_13855_interleave_0 = const()[name = string("op_13855_interleave_0"), val = bool(false)]; + tensor var_13855_cast_fp16 = concat(axis = var_12398, interleave = var_13855_interleave_0, values = (var_13743_cast_fp16, var_13745_cast_fp16, var_13747_cast_fp16, var_13749_cast_fp16))[name = string("op_13855_cast_fp16")]; + bool var_13857_interleave_0 = const()[name = string("op_13857_interleave_0"), val = bool(false)]; + tensor var_13857_cast_fp16 = concat(axis = var_12398, interleave = var_13857_interleave_0, values = (var_13751_cast_fp16, var_13753_cast_fp16, var_13755_cast_fp16, var_13757_cast_fp16))[name = string("op_13857_cast_fp16")]; + bool var_13859_interleave_0 = const()[name = string("op_13859_interleave_0"), val = bool(false)]; + tensor var_13859_cast_fp16 = concat(axis = var_12398, interleave = var_13859_interleave_0, values = (var_13759_cast_fp16, var_13761_cast_fp16, var_13763_cast_fp16, var_13765_cast_fp16))[name = string("op_13859_cast_fp16")]; + bool var_13861_interleave_0 = const()[name = string("op_13861_interleave_0"), val = bool(false)]; + tensor var_13861_cast_fp16 = concat(axis = var_12398, interleave = var_13861_interleave_0, values = (var_13767_cast_fp16, var_13769_cast_fp16, var_13771_cast_fp16, var_13773_cast_fp16))[name = string("op_13861_cast_fp16")]; + bool var_13863_interleave_0 = const()[name = string("op_13863_interleave_0"), val = bool(false)]; + tensor var_13863_cast_fp16 = concat(axis = var_12398, interleave = var_13863_interleave_0, values = (var_13775_cast_fp16, var_13777_cast_fp16, var_13779_cast_fp16, var_13781_cast_fp16))[name = string("op_13863_cast_fp16")]; + bool var_13865_interleave_0 = const()[name = string("op_13865_interleave_0"), val = bool(false)]; + tensor var_13865_cast_fp16 = concat(axis = var_12398, interleave = var_13865_interleave_0, values = (var_13783_cast_fp16, var_13785_cast_fp16, var_13787_cast_fp16, var_13789_cast_fp16))[name = string("op_13865_cast_fp16")]; + bool var_13867_interleave_0 = const()[name = string("op_13867_interleave_0"), val = bool(false)]; + tensor var_13867_cast_fp16 = concat(axis = var_12398, interleave = var_13867_interleave_0, values = (var_13791_cast_fp16, var_13793_cast_fp16, var_13795_cast_fp16, var_13797_cast_fp16))[name = string("op_13867_cast_fp16")]; + bool var_13869_interleave_0 = const()[name = string("op_13869_interleave_0"), val = bool(false)]; + tensor var_13869_cast_fp16 = concat(axis = var_12398, interleave = var_13869_interleave_0, values = (var_13799_cast_fp16, var_13801_cast_fp16, var_13803_cast_fp16, var_13805_cast_fp16))[name = string("op_13869_cast_fp16")]; + bool var_13871_interleave_0 = const()[name = string("op_13871_interleave_0"), val = bool(false)]; + tensor var_13871_cast_fp16 = concat(axis = var_12398, interleave = var_13871_interleave_0, values = (var_13807_cast_fp16, var_13809_cast_fp16, var_13811_cast_fp16, var_13813_cast_fp16))[name = string("op_13871_cast_fp16")]; + bool var_13873_interleave_0 = const()[name = string("op_13873_interleave_0"), val = bool(false)]; + tensor var_13873_cast_fp16 = concat(axis = var_12398, interleave = var_13873_interleave_0, values = (var_13815_cast_fp16, var_13817_cast_fp16, var_13819_cast_fp16, var_13821_cast_fp16))[name = string("op_13873_cast_fp16")]; + bool var_13875_interleave_0 = const()[name = string("op_13875_interleave_0"), val = bool(false)]; + tensor var_13875_cast_fp16 = concat(axis = var_12398, interleave = var_13875_interleave_0, values = (var_13823_cast_fp16, var_13825_cast_fp16, var_13827_cast_fp16, var_13829_cast_fp16))[name = string("op_13875_cast_fp16")]; + bool var_13877_interleave_0 = const()[name = string("op_13877_interleave_0"), val = bool(false)]; + tensor var_13877_cast_fp16 = concat(axis = var_12398, interleave = var_13877_interleave_0, values = (var_13831_cast_fp16, var_13833_cast_fp16, var_13835_cast_fp16, var_13837_cast_fp16))[name = string("op_13877_cast_fp16")]; + bool input_65_interleave_0 = const()[name = string("input_65_interleave_0"), val = bool(false)]; + tensor input_65_cast_fp16 = concat(axis = var_12423, interleave = input_65_interleave_0, values = (var_13839_cast_fp16, var_13841_cast_fp16, var_13843_cast_fp16, var_13845_cast_fp16, var_13847_cast_fp16, var_13849_cast_fp16, var_13851_cast_fp16, var_13853_cast_fp16, var_13855_cast_fp16, var_13857_cast_fp16, var_13859_cast_fp16, var_13861_cast_fp16, var_13863_cast_fp16, var_13865_cast_fp16, var_13867_cast_fp16, var_13869_cast_fp16, var_13871_cast_fp16, var_13873_cast_fp16, var_13875_cast_fp16, var_13877_cast_fp16))[name = string("input_65_cast_fp16")]; + string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; + tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; + tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; + int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339331520)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342608384)))]; + tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_13896_to_fp16 = const()[name = string("op_13896_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_13896_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342611008)))]; + tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342613632)))]; + fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; + tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; + tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; + int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342616256)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355723520)))]; + tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; + tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355733824)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368841088)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; + int32 var_13925 = const()[name = string("op_13925"), val = int32(3)]; + int32 var_13950 = const()[name = string("op_13950"), val = int32(1)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_13967_to_fp16 = const()[name = string("op_13967_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_13967_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368843712)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368846336)))]; + fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368848960)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372125824)))]; + tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; + string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; + tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; + tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; + int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372128448)))]; + tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; + string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; + tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; + tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; + int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375405312)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378682176)))]; + tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_14005_begin_0 = const()[name = string("op_14005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14005_end_0 = const()[name = string("op_14005_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14005_end_mask_0 = const()[name = string("op_14005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14005_cast_fp16 = slice_by_index(begin = var_14005_begin_0, end = var_14005_end_0, end_mask = var_14005_end_mask_0, x = query_19_cast_fp16)[name = string("op_14005_cast_fp16")]; + tensor var_14009_begin_0 = const()[name = string("op_14009_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14009_end_0 = const()[name = string("op_14009_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_14009_end_mask_0 = const()[name = string("op_14009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14009_cast_fp16 = slice_by_index(begin = var_14009_begin_0, end = var_14009_end_0, end_mask = var_14009_end_mask_0, x = query_19_cast_fp16)[name = string("op_14009_cast_fp16")]; + tensor var_14013_begin_0 = const()[name = string("op_14013_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14013_end_0 = const()[name = string("op_14013_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_14013_end_mask_0 = const()[name = string("op_14013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14013_cast_fp16 = slice_by_index(begin = var_14013_begin_0, end = var_14013_end_0, end_mask = var_14013_end_mask_0, x = query_19_cast_fp16)[name = string("op_14013_cast_fp16")]; + tensor var_14017_begin_0 = const()[name = string("op_14017_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14017_end_0 = const()[name = string("op_14017_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_14017_end_mask_0 = const()[name = string("op_14017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14017_cast_fp16 = slice_by_index(begin = var_14017_begin_0, end = var_14017_end_0, end_mask = var_14017_end_mask_0, x = query_19_cast_fp16)[name = string("op_14017_cast_fp16")]; + tensor var_14021_begin_0 = const()[name = string("op_14021_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14021_end_0 = const()[name = string("op_14021_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_14021_end_mask_0 = const()[name = string("op_14021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14021_cast_fp16 = slice_by_index(begin = var_14021_begin_0, end = var_14021_end_0, end_mask = var_14021_end_mask_0, x = query_19_cast_fp16)[name = string("op_14021_cast_fp16")]; + tensor var_14025_begin_0 = const()[name = string("op_14025_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14025_end_0 = const()[name = string("op_14025_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_14025_end_mask_0 = const()[name = string("op_14025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14025_cast_fp16 = slice_by_index(begin = var_14025_begin_0, end = var_14025_end_0, end_mask = var_14025_end_mask_0, x = query_19_cast_fp16)[name = string("op_14025_cast_fp16")]; + tensor var_14029_begin_0 = const()[name = string("op_14029_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14029_end_0 = const()[name = string("op_14029_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_14029_end_mask_0 = const()[name = string("op_14029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14029_cast_fp16 = slice_by_index(begin = var_14029_begin_0, end = var_14029_end_0, end_mask = var_14029_end_mask_0, x = query_19_cast_fp16)[name = string("op_14029_cast_fp16")]; + tensor var_14033_begin_0 = const()[name = string("op_14033_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14033_end_0 = const()[name = string("op_14033_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_14033_end_mask_0 = const()[name = string("op_14033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14033_cast_fp16 = slice_by_index(begin = var_14033_begin_0, end = var_14033_end_0, end_mask = var_14033_end_mask_0, x = query_19_cast_fp16)[name = string("op_14033_cast_fp16")]; + tensor var_14037_begin_0 = const()[name = string("op_14037_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14037_end_0 = const()[name = string("op_14037_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_14037_end_mask_0 = const()[name = string("op_14037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14037_cast_fp16 = slice_by_index(begin = var_14037_begin_0, end = var_14037_end_0, end_mask = var_14037_end_mask_0, x = query_19_cast_fp16)[name = string("op_14037_cast_fp16")]; + tensor var_14041_begin_0 = const()[name = string("op_14041_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14041_end_0 = const()[name = string("op_14041_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_14041_end_mask_0 = const()[name = string("op_14041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14041_cast_fp16 = slice_by_index(begin = var_14041_begin_0, end = var_14041_end_0, end_mask = var_14041_end_mask_0, x = query_19_cast_fp16)[name = string("op_14041_cast_fp16")]; + tensor var_14045_begin_0 = const()[name = string("op_14045_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14045_end_0 = const()[name = string("op_14045_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_14045_end_mask_0 = const()[name = string("op_14045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14045_cast_fp16 = slice_by_index(begin = var_14045_begin_0, end = var_14045_end_0, end_mask = var_14045_end_mask_0, x = query_19_cast_fp16)[name = string("op_14045_cast_fp16")]; + tensor var_14049_begin_0 = const()[name = string("op_14049_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14049_end_0 = const()[name = string("op_14049_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_14049_end_mask_0 = const()[name = string("op_14049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14049_cast_fp16 = slice_by_index(begin = var_14049_begin_0, end = var_14049_end_0, end_mask = var_14049_end_mask_0, x = query_19_cast_fp16)[name = string("op_14049_cast_fp16")]; + tensor var_14053_begin_0 = const()[name = string("op_14053_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14053_end_0 = const()[name = string("op_14053_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_14053_end_mask_0 = const()[name = string("op_14053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14053_cast_fp16 = slice_by_index(begin = var_14053_begin_0, end = var_14053_end_0, end_mask = var_14053_end_mask_0, x = query_19_cast_fp16)[name = string("op_14053_cast_fp16")]; + tensor var_14057_begin_0 = const()[name = string("op_14057_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14057_end_0 = const()[name = string("op_14057_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_14057_end_mask_0 = const()[name = string("op_14057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14057_cast_fp16 = slice_by_index(begin = var_14057_begin_0, end = var_14057_end_0, end_mask = var_14057_end_mask_0, x = query_19_cast_fp16)[name = string("op_14057_cast_fp16")]; + tensor var_14061_begin_0 = const()[name = string("op_14061_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14061_end_0 = const()[name = string("op_14061_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_14061_end_mask_0 = const()[name = string("op_14061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14061_cast_fp16 = slice_by_index(begin = var_14061_begin_0, end = var_14061_end_0, end_mask = var_14061_end_mask_0, x = query_19_cast_fp16)[name = string("op_14061_cast_fp16")]; + tensor var_14065_begin_0 = const()[name = string("op_14065_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14065_end_0 = const()[name = string("op_14065_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_14065_end_mask_0 = const()[name = string("op_14065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14065_cast_fp16 = slice_by_index(begin = var_14065_begin_0, end = var_14065_end_0, end_mask = var_14065_end_mask_0, x = query_19_cast_fp16)[name = string("op_14065_cast_fp16")]; + tensor var_14069_begin_0 = const()[name = string("op_14069_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14069_end_0 = const()[name = string("op_14069_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_14069_end_mask_0 = const()[name = string("op_14069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14069_cast_fp16 = slice_by_index(begin = var_14069_begin_0, end = var_14069_end_0, end_mask = var_14069_end_mask_0, x = query_19_cast_fp16)[name = string("op_14069_cast_fp16")]; + tensor var_14073_begin_0 = const()[name = string("op_14073_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14073_end_0 = const()[name = string("op_14073_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_14073_end_mask_0 = const()[name = string("op_14073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14073_cast_fp16 = slice_by_index(begin = var_14073_begin_0, end = var_14073_end_0, end_mask = var_14073_end_mask_0, x = query_19_cast_fp16)[name = string("op_14073_cast_fp16")]; + tensor var_14077_begin_0 = const()[name = string("op_14077_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14077_end_0 = const()[name = string("op_14077_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_14077_end_mask_0 = const()[name = string("op_14077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14077_cast_fp16 = slice_by_index(begin = var_14077_begin_0, end = var_14077_end_0, end_mask = var_14077_end_mask_0, x = query_19_cast_fp16)[name = string("op_14077_cast_fp16")]; + tensor var_14081_begin_0 = const()[name = string("op_14081_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14081_end_0 = const()[name = string("op_14081_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_14081_end_mask_0 = const()[name = string("op_14081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14081_cast_fp16 = slice_by_index(begin = var_14081_begin_0, end = var_14081_end_0, end_mask = var_14081_end_mask_0, x = query_19_cast_fp16)[name = string("op_14081_cast_fp16")]; + tensor var_14090_begin_0 = const()[name = string("op_14090_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14090_end_0 = const()[name = string("op_14090_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14090_end_mask_0 = const()[name = string("op_14090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14090_cast_fp16 = slice_by_index(begin = var_14090_begin_0, end = var_14090_end_0, end_mask = var_14090_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14090_cast_fp16")]; + tensor var_14097_begin_0 = const()[name = string("op_14097_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14097_end_0 = const()[name = string("op_14097_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14097_end_mask_0 = const()[name = string("op_14097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14097_cast_fp16 = slice_by_index(begin = var_14097_begin_0, end = var_14097_end_0, end_mask = var_14097_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14097_cast_fp16")]; + tensor var_14104_begin_0 = const()[name = string("op_14104_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14104_end_0 = const()[name = string("op_14104_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14104_end_mask_0 = const()[name = string("op_14104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14104_cast_fp16 = slice_by_index(begin = var_14104_begin_0, end = var_14104_end_0, end_mask = var_14104_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14104_cast_fp16")]; + tensor var_14111_begin_0 = const()[name = string("op_14111_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14111_end_0 = const()[name = string("op_14111_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14111_end_mask_0 = const()[name = string("op_14111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14111_cast_fp16 = slice_by_index(begin = var_14111_begin_0, end = var_14111_end_0, end_mask = var_14111_end_mask_0, x = var_14005_cast_fp16)[name = string("op_14111_cast_fp16")]; + tensor var_14118_begin_0 = const()[name = string("op_14118_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14118_end_0 = const()[name = string("op_14118_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14118_end_mask_0 = const()[name = string("op_14118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14118_cast_fp16 = slice_by_index(begin = var_14118_begin_0, end = var_14118_end_0, end_mask = var_14118_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14118_cast_fp16")]; + tensor var_14125_begin_0 = const()[name = string("op_14125_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14125_end_0 = const()[name = string("op_14125_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14125_end_mask_0 = const()[name = string("op_14125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14125_cast_fp16 = slice_by_index(begin = var_14125_begin_0, end = var_14125_end_0, end_mask = var_14125_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14125_cast_fp16")]; + tensor var_14132_begin_0 = const()[name = string("op_14132_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14132_end_0 = const()[name = string("op_14132_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14132_end_mask_0 = const()[name = string("op_14132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14132_cast_fp16 = slice_by_index(begin = var_14132_begin_0, end = var_14132_end_0, end_mask = var_14132_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14132_cast_fp16")]; + tensor var_14139_begin_0 = const()[name = string("op_14139_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14139_end_0 = const()[name = string("op_14139_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14139_end_mask_0 = const()[name = string("op_14139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14139_cast_fp16 = slice_by_index(begin = var_14139_begin_0, end = var_14139_end_0, end_mask = var_14139_end_mask_0, x = var_14009_cast_fp16)[name = string("op_14139_cast_fp16")]; + tensor var_14146_begin_0 = const()[name = string("op_14146_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14146_end_0 = const()[name = string("op_14146_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14146_end_mask_0 = const()[name = string("op_14146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14146_cast_fp16 = slice_by_index(begin = var_14146_begin_0, end = var_14146_end_0, end_mask = var_14146_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14146_cast_fp16")]; + tensor var_14153_begin_0 = const()[name = string("op_14153_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14153_end_0 = const()[name = string("op_14153_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14153_end_mask_0 = const()[name = string("op_14153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14153_cast_fp16 = slice_by_index(begin = var_14153_begin_0, end = var_14153_end_0, end_mask = var_14153_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14153_cast_fp16")]; + tensor var_14160_begin_0 = const()[name = string("op_14160_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14160_end_0 = const()[name = string("op_14160_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14160_end_mask_0 = const()[name = string("op_14160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14160_cast_fp16 = slice_by_index(begin = var_14160_begin_0, end = var_14160_end_0, end_mask = var_14160_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14160_cast_fp16")]; + tensor var_14167_begin_0 = const()[name = string("op_14167_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14167_end_0 = const()[name = string("op_14167_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14167_end_mask_0 = const()[name = string("op_14167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14167_cast_fp16 = slice_by_index(begin = var_14167_begin_0, end = var_14167_end_0, end_mask = var_14167_end_mask_0, x = var_14013_cast_fp16)[name = string("op_14167_cast_fp16")]; + tensor var_14174_begin_0 = const()[name = string("op_14174_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14174_end_0 = const()[name = string("op_14174_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14174_end_mask_0 = const()[name = string("op_14174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14174_cast_fp16 = slice_by_index(begin = var_14174_begin_0, end = var_14174_end_0, end_mask = var_14174_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14174_cast_fp16")]; + tensor var_14181_begin_0 = const()[name = string("op_14181_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14181_end_0 = const()[name = string("op_14181_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14181_end_mask_0 = const()[name = string("op_14181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14181_cast_fp16 = slice_by_index(begin = var_14181_begin_0, end = var_14181_end_0, end_mask = var_14181_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14181_cast_fp16")]; + tensor var_14188_begin_0 = const()[name = string("op_14188_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14188_end_0 = const()[name = string("op_14188_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14188_end_mask_0 = const()[name = string("op_14188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14188_cast_fp16 = slice_by_index(begin = var_14188_begin_0, end = var_14188_end_0, end_mask = var_14188_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14188_cast_fp16")]; + tensor var_14195_begin_0 = const()[name = string("op_14195_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14195_end_0 = const()[name = string("op_14195_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14195_end_mask_0 = const()[name = string("op_14195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14195_cast_fp16 = slice_by_index(begin = var_14195_begin_0, end = var_14195_end_0, end_mask = var_14195_end_mask_0, x = var_14017_cast_fp16)[name = string("op_14195_cast_fp16")]; + tensor var_14202_begin_0 = const()[name = string("op_14202_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14202_end_0 = const()[name = string("op_14202_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14202_end_mask_0 = const()[name = string("op_14202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14202_cast_fp16 = slice_by_index(begin = var_14202_begin_0, end = var_14202_end_0, end_mask = var_14202_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14202_cast_fp16")]; + tensor var_14209_begin_0 = const()[name = string("op_14209_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14209_end_0 = const()[name = string("op_14209_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14209_end_mask_0 = const()[name = string("op_14209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14209_cast_fp16 = slice_by_index(begin = var_14209_begin_0, end = var_14209_end_0, end_mask = var_14209_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14209_cast_fp16")]; + tensor var_14216_begin_0 = const()[name = string("op_14216_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14216_end_0 = const()[name = string("op_14216_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14216_end_mask_0 = const()[name = string("op_14216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14216_cast_fp16 = slice_by_index(begin = var_14216_begin_0, end = var_14216_end_0, end_mask = var_14216_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14216_cast_fp16")]; + tensor var_14223_begin_0 = const()[name = string("op_14223_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14223_end_0 = const()[name = string("op_14223_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14223_end_mask_0 = const()[name = string("op_14223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14223_cast_fp16 = slice_by_index(begin = var_14223_begin_0, end = var_14223_end_0, end_mask = var_14223_end_mask_0, x = var_14021_cast_fp16)[name = string("op_14223_cast_fp16")]; + tensor var_14230_begin_0 = const()[name = string("op_14230_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14230_end_0 = const()[name = string("op_14230_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14230_end_mask_0 = const()[name = string("op_14230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14230_cast_fp16 = slice_by_index(begin = var_14230_begin_0, end = var_14230_end_0, end_mask = var_14230_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14230_cast_fp16")]; + tensor var_14237_begin_0 = const()[name = string("op_14237_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14237_end_0 = const()[name = string("op_14237_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14237_end_mask_0 = const()[name = string("op_14237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14237_cast_fp16 = slice_by_index(begin = var_14237_begin_0, end = var_14237_end_0, end_mask = var_14237_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14237_cast_fp16")]; + tensor var_14244_begin_0 = const()[name = string("op_14244_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14244_end_0 = const()[name = string("op_14244_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14244_end_mask_0 = const()[name = string("op_14244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14244_cast_fp16 = slice_by_index(begin = var_14244_begin_0, end = var_14244_end_0, end_mask = var_14244_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14244_cast_fp16")]; + tensor var_14251_begin_0 = const()[name = string("op_14251_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14251_end_0 = const()[name = string("op_14251_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14251_end_mask_0 = const()[name = string("op_14251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14251_cast_fp16 = slice_by_index(begin = var_14251_begin_0, end = var_14251_end_0, end_mask = var_14251_end_mask_0, x = var_14025_cast_fp16)[name = string("op_14251_cast_fp16")]; + tensor var_14258_begin_0 = const()[name = string("op_14258_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14258_end_0 = const()[name = string("op_14258_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14258_end_mask_0 = const()[name = string("op_14258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14258_cast_fp16 = slice_by_index(begin = var_14258_begin_0, end = var_14258_end_0, end_mask = var_14258_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14258_cast_fp16")]; + tensor var_14265_begin_0 = const()[name = string("op_14265_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14265_end_0 = const()[name = string("op_14265_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14265_end_mask_0 = const()[name = string("op_14265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14265_cast_fp16 = slice_by_index(begin = var_14265_begin_0, end = var_14265_end_0, end_mask = var_14265_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14265_cast_fp16")]; + tensor var_14272_begin_0 = const()[name = string("op_14272_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14272_end_0 = const()[name = string("op_14272_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14272_end_mask_0 = const()[name = string("op_14272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14272_cast_fp16 = slice_by_index(begin = var_14272_begin_0, end = var_14272_end_0, end_mask = var_14272_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14272_cast_fp16")]; + tensor var_14279_begin_0 = const()[name = string("op_14279_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14279_end_0 = const()[name = string("op_14279_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14279_end_mask_0 = const()[name = string("op_14279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14279_cast_fp16 = slice_by_index(begin = var_14279_begin_0, end = var_14279_end_0, end_mask = var_14279_end_mask_0, x = var_14029_cast_fp16)[name = string("op_14279_cast_fp16")]; + tensor var_14286_begin_0 = const()[name = string("op_14286_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14286_end_0 = const()[name = string("op_14286_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14286_end_mask_0 = const()[name = string("op_14286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14286_cast_fp16 = slice_by_index(begin = var_14286_begin_0, end = var_14286_end_0, end_mask = var_14286_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14286_cast_fp16")]; + tensor var_14293_begin_0 = const()[name = string("op_14293_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14293_end_0 = const()[name = string("op_14293_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14293_end_mask_0 = const()[name = string("op_14293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14293_cast_fp16 = slice_by_index(begin = var_14293_begin_0, end = var_14293_end_0, end_mask = var_14293_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14293_cast_fp16")]; + tensor var_14300_begin_0 = const()[name = string("op_14300_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14300_end_0 = const()[name = string("op_14300_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14300_end_mask_0 = const()[name = string("op_14300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14300_cast_fp16 = slice_by_index(begin = var_14300_begin_0, end = var_14300_end_0, end_mask = var_14300_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14300_cast_fp16")]; + tensor var_14307_begin_0 = const()[name = string("op_14307_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14307_end_0 = const()[name = string("op_14307_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14307_end_mask_0 = const()[name = string("op_14307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14307_cast_fp16 = slice_by_index(begin = var_14307_begin_0, end = var_14307_end_0, end_mask = var_14307_end_mask_0, x = var_14033_cast_fp16)[name = string("op_14307_cast_fp16")]; + tensor var_14314_begin_0 = const()[name = string("op_14314_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14314_end_0 = const()[name = string("op_14314_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14314_end_mask_0 = const()[name = string("op_14314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14314_cast_fp16 = slice_by_index(begin = var_14314_begin_0, end = var_14314_end_0, end_mask = var_14314_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14314_cast_fp16")]; + tensor var_14321_begin_0 = const()[name = string("op_14321_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14321_end_0 = const()[name = string("op_14321_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14321_end_mask_0 = const()[name = string("op_14321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14321_cast_fp16 = slice_by_index(begin = var_14321_begin_0, end = var_14321_end_0, end_mask = var_14321_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14321_cast_fp16")]; + tensor var_14328_begin_0 = const()[name = string("op_14328_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14328_end_0 = const()[name = string("op_14328_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14328_end_mask_0 = const()[name = string("op_14328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14328_cast_fp16 = slice_by_index(begin = var_14328_begin_0, end = var_14328_end_0, end_mask = var_14328_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14328_cast_fp16")]; + tensor var_14335_begin_0 = const()[name = string("op_14335_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14335_end_0 = const()[name = string("op_14335_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14335_end_mask_0 = const()[name = string("op_14335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14335_cast_fp16 = slice_by_index(begin = var_14335_begin_0, end = var_14335_end_0, end_mask = var_14335_end_mask_0, x = var_14037_cast_fp16)[name = string("op_14335_cast_fp16")]; + tensor var_14342_begin_0 = const()[name = string("op_14342_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14342_end_0 = const()[name = string("op_14342_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14342_end_mask_0 = const()[name = string("op_14342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14342_cast_fp16 = slice_by_index(begin = var_14342_begin_0, end = var_14342_end_0, end_mask = var_14342_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14342_cast_fp16")]; + tensor var_14349_begin_0 = const()[name = string("op_14349_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14349_end_0 = const()[name = string("op_14349_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14349_end_mask_0 = const()[name = string("op_14349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14349_cast_fp16 = slice_by_index(begin = var_14349_begin_0, end = var_14349_end_0, end_mask = var_14349_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14349_cast_fp16")]; + tensor var_14356_begin_0 = const()[name = string("op_14356_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14356_end_0 = const()[name = string("op_14356_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14356_end_mask_0 = const()[name = string("op_14356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14356_cast_fp16 = slice_by_index(begin = var_14356_begin_0, end = var_14356_end_0, end_mask = var_14356_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14356_cast_fp16")]; + tensor var_14363_begin_0 = const()[name = string("op_14363_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14363_end_0 = const()[name = string("op_14363_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14363_end_mask_0 = const()[name = string("op_14363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14363_cast_fp16 = slice_by_index(begin = var_14363_begin_0, end = var_14363_end_0, end_mask = var_14363_end_mask_0, x = var_14041_cast_fp16)[name = string("op_14363_cast_fp16")]; + tensor var_14370_begin_0 = const()[name = string("op_14370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14370_end_0 = const()[name = string("op_14370_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14370_end_mask_0 = const()[name = string("op_14370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14370_cast_fp16 = slice_by_index(begin = var_14370_begin_0, end = var_14370_end_0, end_mask = var_14370_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14370_cast_fp16")]; + tensor var_14377_begin_0 = const()[name = string("op_14377_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14377_end_0 = const()[name = string("op_14377_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14377_end_mask_0 = const()[name = string("op_14377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14377_cast_fp16 = slice_by_index(begin = var_14377_begin_0, end = var_14377_end_0, end_mask = var_14377_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14377_cast_fp16")]; + tensor var_14384_begin_0 = const()[name = string("op_14384_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14384_end_0 = const()[name = string("op_14384_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14384_end_mask_0 = const()[name = string("op_14384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14384_cast_fp16 = slice_by_index(begin = var_14384_begin_0, end = var_14384_end_0, end_mask = var_14384_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14384_cast_fp16")]; + tensor var_14391_begin_0 = const()[name = string("op_14391_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14391_end_0 = const()[name = string("op_14391_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14391_end_mask_0 = const()[name = string("op_14391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14391_cast_fp16 = slice_by_index(begin = var_14391_begin_0, end = var_14391_end_0, end_mask = var_14391_end_mask_0, x = var_14045_cast_fp16)[name = string("op_14391_cast_fp16")]; + tensor var_14398_begin_0 = const()[name = string("op_14398_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14398_end_0 = const()[name = string("op_14398_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14398_end_mask_0 = const()[name = string("op_14398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14398_cast_fp16 = slice_by_index(begin = var_14398_begin_0, end = var_14398_end_0, end_mask = var_14398_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14398_cast_fp16")]; + tensor var_14405_begin_0 = const()[name = string("op_14405_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14405_end_0 = const()[name = string("op_14405_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14405_end_mask_0 = const()[name = string("op_14405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14405_cast_fp16 = slice_by_index(begin = var_14405_begin_0, end = var_14405_end_0, end_mask = var_14405_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14405_cast_fp16")]; + tensor var_14412_begin_0 = const()[name = string("op_14412_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14412_end_0 = const()[name = string("op_14412_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14412_end_mask_0 = const()[name = string("op_14412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14412_cast_fp16 = slice_by_index(begin = var_14412_begin_0, end = var_14412_end_0, end_mask = var_14412_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14412_cast_fp16")]; + tensor var_14419_begin_0 = const()[name = string("op_14419_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14419_end_0 = const()[name = string("op_14419_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14419_end_mask_0 = const()[name = string("op_14419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14419_cast_fp16 = slice_by_index(begin = var_14419_begin_0, end = var_14419_end_0, end_mask = var_14419_end_mask_0, x = var_14049_cast_fp16)[name = string("op_14419_cast_fp16")]; + tensor var_14426_begin_0 = const()[name = string("op_14426_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14426_end_0 = const()[name = string("op_14426_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14426_end_mask_0 = const()[name = string("op_14426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14426_cast_fp16 = slice_by_index(begin = var_14426_begin_0, end = var_14426_end_0, end_mask = var_14426_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14426_cast_fp16")]; + tensor var_14433_begin_0 = const()[name = string("op_14433_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14433_end_0 = const()[name = string("op_14433_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14433_end_mask_0 = const()[name = string("op_14433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14433_cast_fp16 = slice_by_index(begin = var_14433_begin_0, end = var_14433_end_0, end_mask = var_14433_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14433_cast_fp16")]; + tensor var_14440_begin_0 = const()[name = string("op_14440_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14440_end_0 = const()[name = string("op_14440_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14440_end_mask_0 = const()[name = string("op_14440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14440_cast_fp16 = slice_by_index(begin = var_14440_begin_0, end = var_14440_end_0, end_mask = var_14440_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14440_cast_fp16")]; + tensor var_14447_begin_0 = const()[name = string("op_14447_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14447_end_0 = const()[name = string("op_14447_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14447_end_mask_0 = const()[name = string("op_14447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14447_cast_fp16 = slice_by_index(begin = var_14447_begin_0, end = var_14447_end_0, end_mask = var_14447_end_mask_0, x = var_14053_cast_fp16)[name = string("op_14447_cast_fp16")]; + tensor var_14454_begin_0 = const()[name = string("op_14454_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14454_end_0 = const()[name = string("op_14454_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14454_end_mask_0 = const()[name = string("op_14454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14454_cast_fp16 = slice_by_index(begin = var_14454_begin_0, end = var_14454_end_0, end_mask = var_14454_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14454_cast_fp16")]; + tensor var_14461_begin_0 = const()[name = string("op_14461_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14461_end_0 = const()[name = string("op_14461_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14461_end_mask_0 = const()[name = string("op_14461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14461_cast_fp16 = slice_by_index(begin = var_14461_begin_0, end = var_14461_end_0, end_mask = var_14461_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14461_cast_fp16")]; + tensor var_14468_begin_0 = const()[name = string("op_14468_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14468_end_0 = const()[name = string("op_14468_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14468_end_mask_0 = const()[name = string("op_14468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14468_cast_fp16 = slice_by_index(begin = var_14468_begin_0, end = var_14468_end_0, end_mask = var_14468_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14468_cast_fp16")]; + tensor var_14475_begin_0 = const()[name = string("op_14475_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14475_end_0 = const()[name = string("op_14475_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14475_end_mask_0 = const()[name = string("op_14475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14475_cast_fp16 = slice_by_index(begin = var_14475_begin_0, end = var_14475_end_0, end_mask = var_14475_end_mask_0, x = var_14057_cast_fp16)[name = string("op_14475_cast_fp16")]; + tensor var_14482_begin_0 = const()[name = string("op_14482_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14482_end_0 = const()[name = string("op_14482_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14482_end_mask_0 = const()[name = string("op_14482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14482_cast_fp16 = slice_by_index(begin = var_14482_begin_0, end = var_14482_end_0, end_mask = var_14482_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14482_cast_fp16")]; + tensor var_14489_begin_0 = const()[name = string("op_14489_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14489_end_0 = const()[name = string("op_14489_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14489_end_mask_0 = const()[name = string("op_14489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14489_cast_fp16 = slice_by_index(begin = var_14489_begin_0, end = var_14489_end_0, end_mask = var_14489_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14489_cast_fp16")]; + tensor var_14496_begin_0 = const()[name = string("op_14496_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14496_end_0 = const()[name = string("op_14496_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14496_end_mask_0 = const()[name = string("op_14496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14496_cast_fp16 = slice_by_index(begin = var_14496_begin_0, end = var_14496_end_0, end_mask = var_14496_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14496_cast_fp16")]; + tensor var_14503_begin_0 = const()[name = string("op_14503_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14503_end_0 = const()[name = string("op_14503_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14503_end_mask_0 = const()[name = string("op_14503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14503_cast_fp16 = slice_by_index(begin = var_14503_begin_0, end = var_14503_end_0, end_mask = var_14503_end_mask_0, x = var_14061_cast_fp16)[name = string("op_14503_cast_fp16")]; + tensor var_14510_begin_0 = const()[name = string("op_14510_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14510_end_0 = const()[name = string("op_14510_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14510_end_mask_0 = const()[name = string("op_14510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14510_cast_fp16 = slice_by_index(begin = var_14510_begin_0, end = var_14510_end_0, end_mask = var_14510_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14510_cast_fp16")]; + tensor var_14517_begin_0 = const()[name = string("op_14517_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14517_end_0 = const()[name = string("op_14517_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14517_end_mask_0 = const()[name = string("op_14517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14517_cast_fp16 = slice_by_index(begin = var_14517_begin_0, end = var_14517_end_0, end_mask = var_14517_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14517_cast_fp16")]; + tensor var_14524_begin_0 = const()[name = string("op_14524_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14524_end_0 = const()[name = string("op_14524_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14524_end_mask_0 = const()[name = string("op_14524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14524_cast_fp16 = slice_by_index(begin = var_14524_begin_0, end = var_14524_end_0, end_mask = var_14524_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14524_cast_fp16")]; + tensor var_14531_begin_0 = const()[name = string("op_14531_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14531_end_0 = const()[name = string("op_14531_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14531_end_mask_0 = const()[name = string("op_14531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14531_cast_fp16 = slice_by_index(begin = var_14531_begin_0, end = var_14531_end_0, end_mask = var_14531_end_mask_0, x = var_14065_cast_fp16)[name = string("op_14531_cast_fp16")]; + tensor var_14538_begin_0 = const()[name = string("op_14538_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14538_end_0 = const()[name = string("op_14538_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14538_end_mask_0 = const()[name = string("op_14538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14538_cast_fp16 = slice_by_index(begin = var_14538_begin_0, end = var_14538_end_0, end_mask = var_14538_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14538_cast_fp16")]; + tensor var_14545_begin_0 = const()[name = string("op_14545_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14545_end_0 = const()[name = string("op_14545_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14545_end_mask_0 = const()[name = string("op_14545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14545_cast_fp16 = slice_by_index(begin = var_14545_begin_0, end = var_14545_end_0, end_mask = var_14545_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14545_cast_fp16")]; + tensor var_14552_begin_0 = const()[name = string("op_14552_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14552_end_0 = const()[name = string("op_14552_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14552_end_mask_0 = const()[name = string("op_14552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14552_cast_fp16 = slice_by_index(begin = var_14552_begin_0, end = var_14552_end_0, end_mask = var_14552_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14552_cast_fp16")]; + tensor var_14559_begin_0 = const()[name = string("op_14559_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14559_end_0 = const()[name = string("op_14559_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14559_end_mask_0 = const()[name = string("op_14559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14559_cast_fp16 = slice_by_index(begin = var_14559_begin_0, end = var_14559_end_0, end_mask = var_14559_end_mask_0, x = var_14069_cast_fp16)[name = string("op_14559_cast_fp16")]; + tensor var_14566_begin_0 = const()[name = string("op_14566_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14566_end_0 = const()[name = string("op_14566_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14566_end_mask_0 = const()[name = string("op_14566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14566_cast_fp16 = slice_by_index(begin = var_14566_begin_0, end = var_14566_end_0, end_mask = var_14566_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14566_cast_fp16")]; + tensor var_14573_begin_0 = const()[name = string("op_14573_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14573_end_0 = const()[name = string("op_14573_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14573_end_mask_0 = const()[name = string("op_14573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14573_cast_fp16 = slice_by_index(begin = var_14573_begin_0, end = var_14573_end_0, end_mask = var_14573_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14573_cast_fp16")]; + tensor var_14580_begin_0 = const()[name = string("op_14580_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14580_end_0 = const()[name = string("op_14580_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14580_end_mask_0 = const()[name = string("op_14580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14580_cast_fp16 = slice_by_index(begin = var_14580_begin_0, end = var_14580_end_0, end_mask = var_14580_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14580_cast_fp16")]; + tensor var_14587_begin_0 = const()[name = string("op_14587_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14587_end_0 = const()[name = string("op_14587_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14587_end_mask_0 = const()[name = string("op_14587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14587_cast_fp16 = slice_by_index(begin = var_14587_begin_0, end = var_14587_end_0, end_mask = var_14587_end_mask_0, x = var_14073_cast_fp16)[name = string("op_14587_cast_fp16")]; + tensor var_14594_begin_0 = const()[name = string("op_14594_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14594_end_0 = const()[name = string("op_14594_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14594_end_mask_0 = const()[name = string("op_14594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14594_cast_fp16 = slice_by_index(begin = var_14594_begin_0, end = var_14594_end_0, end_mask = var_14594_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14594_cast_fp16")]; + tensor var_14601_begin_0 = const()[name = string("op_14601_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14601_end_0 = const()[name = string("op_14601_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14601_end_mask_0 = const()[name = string("op_14601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14601_cast_fp16 = slice_by_index(begin = var_14601_begin_0, end = var_14601_end_0, end_mask = var_14601_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14601_cast_fp16")]; + tensor var_14608_begin_0 = const()[name = string("op_14608_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14608_end_0 = const()[name = string("op_14608_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14608_end_mask_0 = const()[name = string("op_14608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14608_cast_fp16 = slice_by_index(begin = var_14608_begin_0, end = var_14608_end_0, end_mask = var_14608_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14608_cast_fp16")]; + tensor var_14615_begin_0 = const()[name = string("op_14615_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14615_end_0 = const()[name = string("op_14615_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14615_end_mask_0 = const()[name = string("op_14615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14615_cast_fp16 = slice_by_index(begin = var_14615_begin_0, end = var_14615_end_0, end_mask = var_14615_end_mask_0, x = var_14077_cast_fp16)[name = string("op_14615_cast_fp16")]; + tensor var_14622_begin_0 = const()[name = string("op_14622_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14622_end_0 = const()[name = string("op_14622_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14622_end_mask_0 = const()[name = string("op_14622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14622_cast_fp16 = slice_by_index(begin = var_14622_begin_0, end = var_14622_end_0, end_mask = var_14622_end_mask_0, x = var_14081_cast_fp16)[name = string("op_14622_cast_fp16")]; + tensor var_14629_begin_0 = const()[name = string("op_14629_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14629_end_0 = const()[name = string("op_14629_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14629_end_mask_0 = const()[name = string("op_14629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14629_cast_fp16 = slice_by_index(begin = var_14629_begin_0, end = var_14629_end_0, end_mask = var_14629_end_mask_0, x = var_14081_cast_fp16)[name = string("op_14629_cast_fp16")]; + tensor var_14636_begin_0 = const()[name = string("op_14636_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14636_end_0 = const()[name = string("op_14636_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14636_end_mask_0 = const()[name = string("op_14636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14636_cast_fp16 = slice_by_index(begin = var_14636_begin_0, end = var_14636_end_0, end_mask = var_14636_end_mask_0, x = var_14081_cast_fp16)[name = string("op_14636_cast_fp16")]; + tensor var_14643_begin_0 = const()[name = string("op_14643_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14643_end_0 = const()[name = string("op_14643_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14643_end_mask_0 = const()[name = string("op_14643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14643_cast_fp16 = slice_by_index(begin = var_14643_begin_0, end = var_14643_end_0, end_mask = var_14643_end_mask_0, x = var_14081_cast_fp16)[name = string("op_14643_cast_fp16")]; + tensor k_19_perm_0 = const()[name = string("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14648_begin_0 = const()[name = string("op_14648_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14648_end_0 = const()[name = string("op_14648_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_14648_end_mask_0 = const()[name = string("op_14648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = string("transpose_22")]; + tensor var_14648_cast_fp16 = slice_by_index(begin = var_14648_begin_0, end = var_14648_end_0, end_mask = var_14648_end_mask_0, x = k_19_cast_fp16)[name = string("op_14648_cast_fp16")]; + tensor var_14652_begin_0 = const()[name = string("op_14652_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14652_end_0 = const()[name = string("op_14652_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_14652_end_mask_0 = const()[name = string("op_14652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14652_cast_fp16 = slice_by_index(begin = var_14652_begin_0, end = var_14652_end_0, end_mask = var_14652_end_mask_0, x = k_19_cast_fp16)[name = string("op_14652_cast_fp16")]; + tensor var_14656_begin_0 = const()[name = string("op_14656_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14656_end_0 = const()[name = string("op_14656_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_14656_end_mask_0 = const()[name = string("op_14656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14656_cast_fp16 = slice_by_index(begin = var_14656_begin_0, end = var_14656_end_0, end_mask = var_14656_end_mask_0, x = k_19_cast_fp16)[name = string("op_14656_cast_fp16")]; + tensor var_14660_begin_0 = const()[name = string("op_14660_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14660_end_0 = const()[name = string("op_14660_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_14660_end_mask_0 = const()[name = string("op_14660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14660_cast_fp16 = slice_by_index(begin = var_14660_begin_0, end = var_14660_end_0, end_mask = var_14660_end_mask_0, x = k_19_cast_fp16)[name = string("op_14660_cast_fp16")]; + tensor var_14664_begin_0 = const()[name = string("op_14664_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14664_end_0 = const()[name = string("op_14664_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_14664_end_mask_0 = const()[name = string("op_14664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14664_cast_fp16 = slice_by_index(begin = var_14664_begin_0, end = var_14664_end_0, end_mask = var_14664_end_mask_0, x = k_19_cast_fp16)[name = string("op_14664_cast_fp16")]; + tensor var_14668_begin_0 = const()[name = string("op_14668_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14668_end_0 = const()[name = string("op_14668_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_14668_end_mask_0 = const()[name = string("op_14668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14668_cast_fp16 = slice_by_index(begin = var_14668_begin_0, end = var_14668_end_0, end_mask = var_14668_end_mask_0, x = k_19_cast_fp16)[name = string("op_14668_cast_fp16")]; + tensor var_14672_begin_0 = const()[name = string("op_14672_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14672_end_0 = const()[name = string("op_14672_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_14672_end_mask_0 = const()[name = string("op_14672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14672_cast_fp16 = slice_by_index(begin = var_14672_begin_0, end = var_14672_end_0, end_mask = var_14672_end_mask_0, x = k_19_cast_fp16)[name = string("op_14672_cast_fp16")]; + tensor var_14676_begin_0 = const()[name = string("op_14676_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14676_end_0 = const()[name = string("op_14676_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_14676_end_mask_0 = const()[name = string("op_14676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14676_cast_fp16 = slice_by_index(begin = var_14676_begin_0, end = var_14676_end_0, end_mask = var_14676_end_mask_0, x = k_19_cast_fp16)[name = string("op_14676_cast_fp16")]; + tensor var_14680_begin_0 = const()[name = string("op_14680_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14680_end_0 = const()[name = string("op_14680_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_14680_end_mask_0 = const()[name = string("op_14680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14680_cast_fp16 = slice_by_index(begin = var_14680_begin_0, end = var_14680_end_0, end_mask = var_14680_end_mask_0, x = k_19_cast_fp16)[name = string("op_14680_cast_fp16")]; + tensor var_14684_begin_0 = const()[name = string("op_14684_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14684_end_0 = const()[name = string("op_14684_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_14684_end_mask_0 = const()[name = string("op_14684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14684_cast_fp16 = slice_by_index(begin = var_14684_begin_0, end = var_14684_end_0, end_mask = var_14684_end_mask_0, x = k_19_cast_fp16)[name = string("op_14684_cast_fp16")]; + tensor var_14688_begin_0 = const()[name = string("op_14688_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14688_end_0 = const()[name = string("op_14688_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_14688_end_mask_0 = const()[name = string("op_14688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14688_cast_fp16 = slice_by_index(begin = var_14688_begin_0, end = var_14688_end_0, end_mask = var_14688_end_mask_0, x = k_19_cast_fp16)[name = string("op_14688_cast_fp16")]; + tensor var_14692_begin_0 = const()[name = string("op_14692_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14692_end_0 = const()[name = string("op_14692_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_14692_end_mask_0 = const()[name = string("op_14692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14692_cast_fp16 = slice_by_index(begin = var_14692_begin_0, end = var_14692_end_0, end_mask = var_14692_end_mask_0, x = k_19_cast_fp16)[name = string("op_14692_cast_fp16")]; + tensor var_14696_begin_0 = const()[name = string("op_14696_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14696_end_0 = const()[name = string("op_14696_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_14696_end_mask_0 = const()[name = string("op_14696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14696_cast_fp16 = slice_by_index(begin = var_14696_begin_0, end = var_14696_end_0, end_mask = var_14696_end_mask_0, x = k_19_cast_fp16)[name = string("op_14696_cast_fp16")]; + tensor var_14700_begin_0 = const()[name = string("op_14700_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14700_end_0 = const()[name = string("op_14700_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_14700_end_mask_0 = const()[name = string("op_14700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14700_cast_fp16 = slice_by_index(begin = var_14700_begin_0, end = var_14700_end_0, end_mask = var_14700_end_mask_0, x = k_19_cast_fp16)[name = string("op_14700_cast_fp16")]; + tensor var_14704_begin_0 = const()[name = string("op_14704_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14704_end_0 = const()[name = string("op_14704_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_14704_end_mask_0 = const()[name = string("op_14704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14704_cast_fp16 = slice_by_index(begin = var_14704_begin_0, end = var_14704_end_0, end_mask = var_14704_end_mask_0, x = k_19_cast_fp16)[name = string("op_14704_cast_fp16")]; + tensor var_14708_begin_0 = const()[name = string("op_14708_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14708_end_0 = const()[name = string("op_14708_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_14708_end_mask_0 = const()[name = string("op_14708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14708_cast_fp16 = slice_by_index(begin = var_14708_begin_0, end = var_14708_end_0, end_mask = var_14708_end_mask_0, x = k_19_cast_fp16)[name = string("op_14708_cast_fp16")]; + tensor var_14712_begin_0 = const()[name = string("op_14712_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14712_end_0 = const()[name = string("op_14712_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_14712_end_mask_0 = const()[name = string("op_14712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14712_cast_fp16 = slice_by_index(begin = var_14712_begin_0, end = var_14712_end_0, end_mask = var_14712_end_mask_0, x = k_19_cast_fp16)[name = string("op_14712_cast_fp16")]; + tensor var_14716_begin_0 = const()[name = string("op_14716_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14716_end_0 = const()[name = string("op_14716_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_14716_end_mask_0 = const()[name = string("op_14716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14716_cast_fp16 = slice_by_index(begin = var_14716_begin_0, end = var_14716_end_0, end_mask = var_14716_end_mask_0, x = k_19_cast_fp16)[name = string("op_14716_cast_fp16")]; + tensor var_14720_begin_0 = const()[name = string("op_14720_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14720_end_0 = const()[name = string("op_14720_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_14720_end_mask_0 = const()[name = string("op_14720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14720_cast_fp16 = slice_by_index(begin = var_14720_begin_0, end = var_14720_end_0, end_mask = var_14720_end_mask_0, x = k_19_cast_fp16)[name = string("op_14720_cast_fp16")]; + tensor var_14724_begin_0 = const()[name = string("op_14724_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14724_end_0 = const()[name = string("op_14724_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_14724_end_mask_0 = const()[name = string("op_14724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14724_cast_fp16 = slice_by_index(begin = var_14724_begin_0, end = var_14724_end_0, end_mask = var_14724_end_mask_0, x = k_19_cast_fp16)[name = string("op_14724_cast_fp16")]; + tensor var_14726_begin_0 = const()[name = string("op_14726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14726_end_0 = const()[name = string("op_14726_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14726_end_mask_0 = const()[name = string("op_14726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14726_cast_fp16 = slice_by_index(begin = var_14726_begin_0, end = var_14726_end_0, end_mask = var_14726_end_mask_0, x = value_19_cast_fp16)[name = string("op_14726_cast_fp16")]; + tensor var_14730_begin_0 = const()[name = string("op_14730_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14730_end_0 = const()[name = string("op_14730_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_14730_end_mask_0 = const()[name = string("op_14730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14730_cast_fp16 = slice_by_index(begin = var_14730_begin_0, end = var_14730_end_0, end_mask = var_14730_end_mask_0, x = value_19_cast_fp16)[name = string("op_14730_cast_fp16")]; + tensor var_14734_begin_0 = const()[name = string("op_14734_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14734_end_0 = const()[name = string("op_14734_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_14734_end_mask_0 = const()[name = string("op_14734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14734_cast_fp16 = slice_by_index(begin = var_14734_begin_0, end = var_14734_end_0, end_mask = var_14734_end_mask_0, x = value_19_cast_fp16)[name = string("op_14734_cast_fp16")]; + tensor var_14738_begin_0 = const()[name = string("op_14738_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14738_end_0 = const()[name = string("op_14738_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_14738_end_mask_0 = const()[name = string("op_14738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14738_cast_fp16 = slice_by_index(begin = var_14738_begin_0, end = var_14738_end_0, end_mask = var_14738_end_mask_0, x = value_19_cast_fp16)[name = string("op_14738_cast_fp16")]; + tensor var_14742_begin_0 = const()[name = string("op_14742_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14742_end_0 = const()[name = string("op_14742_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_14742_end_mask_0 = const()[name = string("op_14742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14742_cast_fp16 = slice_by_index(begin = var_14742_begin_0, end = var_14742_end_0, end_mask = var_14742_end_mask_0, x = value_19_cast_fp16)[name = string("op_14742_cast_fp16")]; + tensor var_14746_begin_0 = const()[name = string("op_14746_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14746_end_0 = const()[name = string("op_14746_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_14746_end_mask_0 = const()[name = string("op_14746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14746_cast_fp16 = slice_by_index(begin = var_14746_begin_0, end = var_14746_end_0, end_mask = var_14746_end_mask_0, x = value_19_cast_fp16)[name = string("op_14746_cast_fp16")]; + tensor var_14750_begin_0 = const()[name = string("op_14750_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14750_end_0 = const()[name = string("op_14750_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_14750_end_mask_0 = const()[name = string("op_14750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14750_cast_fp16 = slice_by_index(begin = var_14750_begin_0, end = var_14750_end_0, end_mask = var_14750_end_mask_0, x = value_19_cast_fp16)[name = string("op_14750_cast_fp16")]; + tensor var_14754_begin_0 = const()[name = string("op_14754_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14754_end_0 = const()[name = string("op_14754_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_14754_end_mask_0 = const()[name = string("op_14754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14754_cast_fp16 = slice_by_index(begin = var_14754_begin_0, end = var_14754_end_0, end_mask = var_14754_end_mask_0, x = value_19_cast_fp16)[name = string("op_14754_cast_fp16")]; + tensor var_14758_begin_0 = const()[name = string("op_14758_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14758_end_0 = const()[name = string("op_14758_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_14758_end_mask_0 = const()[name = string("op_14758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14758_cast_fp16 = slice_by_index(begin = var_14758_begin_0, end = var_14758_end_0, end_mask = var_14758_end_mask_0, x = value_19_cast_fp16)[name = string("op_14758_cast_fp16")]; + tensor var_14762_begin_0 = const()[name = string("op_14762_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14762_end_0 = const()[name = string("op_14762_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_14762_end_mask_0 = const()[name = string("op_14762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14762_cast_fp16 = slice_by_index(begin = var_14762_begin_0, end = var_14762_end_0, end_mask = var_14762_end_mask_0, x = value_19_cast_fp16)[name = string("op_14762_cast_fp16")]; + tensor var_14766_begin_0 = const()[name = string("op_14766_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14766_end_0 = const()[name = string("op_14766_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_14766_end_mask_0 = const()[name = string("op_14766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14766_cast_fp16 = slice_by_index(begin = var_14766_begin_0, end = var_14766_end_0, end_mask = var_14766_end_mask_0, x = value_19_cast_fp16)[name = string("op_14766_cast_fp16")]; + tensor var_14770_begin_0 = const()[name = string("op_14770_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14770_end_0 = const()[name = string("op_14770_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_14770_end_mask_0 = const()[name = string("op_14770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14770_cast_fp16 = slice_by_index(begin = var_14770_begin_0, end = var_14770_end_0, end_mask = var_14770_end_mask_0, x = value_19_cast_fp16)[name = string("op_14770_cast_fp16")]; + tensor var_14774_begin_0 = const()[name = string("op_14774_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14774_end_0 = const()[name = string("op_14774_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_14774_end_mask_0 = const()[name = string("op_14774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14774_cast_fp16 = slice_by_index(begin = var_14774_begin_0, end = var_14774_end_0, end_mask = var_14774_end_mask_0, x = value_19_cast_fp16)[name = string("op_14774_cast_fp16")]; + tensor var_14778_begin_0 = const()[name = string("op_14778_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14778_end_0 = const()[name = string("op_14778_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_14778_end_mask_0 = const()[name = string("op_14778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14778_cast_fp16 = slice_by_index(begin = var_14778_begin_0, end = var_14778_end_0, end_mask = var_14778_end_mask_0, x = value_19_cast_fp16)[name = string("op_14778_cast_fp16")]; + tensor var_14782_begin_0 = const()[name = string("op_14782_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14782_end_0 = const()[name = string("op_14782_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_14782_end_mask_0 = const()[name = string("op_14782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14782_cast_fp16 = slice_by_index(begin = var_14782_begin_0, end = var_14782_end_0, end_mask = var_14782_end_mask_0, x = value_19_cast_fp16)[name = string("op_14782_cast_fp16")]; + tensor var_14786_begin_0 = const()[name = string("op_14786_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14786_end_0 = const()[name = string("op_14786_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_14786_end_mask_0 = const()[name = string("op_14786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14786_cast_fp16 = slice_by_index(begin = var_14786_begin_0, end = var_14786_end_0, end_mask = var_14786_end_mask_0, x = value_19_cast_fp16)[name = string("op_14786_cast_fp16")]; + tensor var_14790_begin_0 = const()[name = string("op_14790_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14790_end_0 = const()[name = string("op_14790_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_14790_end_mask_0 = const()[name = string("op_14790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14790_cast_fp16 = slice_by_index(begin = var_14790_begin_0, end = var_14790_end_0, end_mask = var_14790_end_mask_0, x = value_19_cast_fp16)[name = string("op_14790_cast_fp16")]; + tensor var_14794_begin_0 = const()[name = string("op_14794_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14794_end_0 = const()[name = string("op_14794_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_14794_end_mask_0 = const()[name = string("op_14794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14794_cast_fp16 = slice_by_index(begin = var_14794_begin_0, end = var_14794_end_0, end_mask = var_14794_end_mask_0, x = value_19_cast_fp16)[name = string("op_14794_cast_fp16")]; + tensor var_14798_begin_0 = const()[name = string("op_14798_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14798_end_0 = const()[name = string("op_14798_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_14798_end_mask_0 = const()[name = string("op_14798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14798_cast_fp16 = slice_by_index(begin = var_14798_begin_0, end = var_14798_end_0, end_mask = var_14798_end_mask_0, x = value_19_cast_fp16)[name = string("op_14798_cast_fp16")]; + tensor var_14802_begin_0 = const()[name = string("op_14802_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14802_end_0 = const()[name = string("op_14802_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_14802_end_mask_0 = const()[name = string("op_14802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14802_cast_fp16 = slice_by_index(begin = var_14802_begin_0, end = var_14802_end_0, end_mask = var_14802_end_mask_0, x = value_19_cast_fp16)[name = string("op_14802_cast_fp16")]; + string _SplitHeadsQ__mh_w_1441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1441_equation_0, values = (var_14648_cast_fp16, var_14090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1441_cast_fp16")]; + string _SplitHeadsQ__mh_w_1443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1443_equation_0, values = (var_14648_cast_fp16, var_14097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1443_cast_fp16")]; + string _SplitHeadsQ__mh_w_1445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1445_equation_0, values = (var_14648_cast_fp16, var_14104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1445_cast_fp16")]; + string _SplitHeadsQ__mh_w_1447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1447_equation_0, values = (var_14648_cast_fp16, var_14111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1447_cast_fp16")]; + string _SplitHeadsQ__mh_w_1449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1449_equation_0, values = (var_14652_cast_fp16, var_14118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1449_cast_fp16")]; + string _SplitHeadsQ__mh_w_1451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1451_equation_0, values = (var_14652_cast_fp16, var_14125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1451_cast_fp16")]; + string _SplitHeadsQ__mh_w_1453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1453_equation_0, values = (var_14652_cast_fp16, var_14132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1453_cast_fp16")]; + string _SplitHeadsQ__mh_w_1455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1455_equation_0, values = (var_14652_cast_fp16, var_14139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1455_cast_fp16")]; + string _SplitHeadsQ__mh_w_1457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1457_equation_0, values = (var_14656_cast_fp16, var_14146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1457_cast_fp16")]; + string _SplitHeadsQ__mh_w_1459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1459_equation_0, values = (var_14656_cast_fp16, var_14153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1459_cast_fp16")]; + string _SplitHeadsQ__mh_w_1461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1461_equation_0, values = (var_14656_cast_fp16, var_14160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1461_cast_fp16")]; + string _SplitHeadsQ__mh_w_1463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1463_equation_0, values = (var_14656_cast_fp16, var_14167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1463_cast_fp16")]; + string _SplitHeadsQ__mh_w_1465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1465_equation_0, values = (var_14660_cast_fp16, var_14174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1465_cast_fp16")]; + string _SplitHeadsQ__mh_w_1467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1467_equation_0, values = (var_14660_cast_fp16, var_14181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1467_cast_fp16")]; + string _SplitHeadsQ__mh_w_1469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1469_equation_0, values = (var_14660_cast_fp16, var_14188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1469_cast_fp16")]; + string _SplitHeadsQ__mh_w_1471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1471_equation_0, values = (var_14660_cast_fp16, var_14195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1471_cast_fp16")]; + string _SplitHeadsQ__mh_w_1473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1473_equation_0, values = (var_14664_cast_fp16, var_14202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1473_cast_fp16")]; + string _SplitHeadsQ__mh_w_1475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1475_equation_0, values = (var_14664_cast_fp16, var_14209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1475_cast_fp16")]; + string _SplitHeadsQ__mh_w_1477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1477_equation_0, values = (var_14664_cast_fp16, var_14216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1477_cast_fp16")]; + string _SplitHeadsQ__mh_w_1479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1479_equation_0, values = (var_14664_cast_fp16, var_14223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1479_cast_fp16")]; + string _SplitHeadsQ__mh_w_1481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1481_equation_0, values = (var_14668_cast_fp16, var_14230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1481_cast_fp16")]; + string _SplitHeadsQ__mh_w_1483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1483_equation_0, values = (var_14668_cast_fp16, var_14237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1483_cast_fp16")]; + string _SplitHeadsQ__mh_w_1485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1485_equation_0, values = (var_14668_cast_fp16, var_14244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1485_cast_fp16")]; + string _SplitHeadsQ__mh_w_1487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1487_equation_0, values = (var_14668_cast_fp16, var_14251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1487_cast_fp16")]; + string _SplitHeadsQ__mh_w_1489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1489_equation_0, values = (var_14672_cast_fp16, var_14258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1489_cast_fp16")]; + string _SplitHeadsQ__mh_w_1491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1491_equation_0, values = (var_14672_cast_fp16, var_14265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1491_cast_fp16")]; + string _SplitHeadsQ__mh_w_1493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1493_equation_0, values = (var_14672_cast_fp16, var_14272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1493_cast_fp16")]; + string _SplitHeadsQ__mh_w_1495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1495_equation_0, values = (var_14672_cast_fp16, var_14279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1495_cast_fp16")]; + string _SplitHeadsQ__mh_w_1497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1497_equation_0, values = (var_14676_cast_fp16, var_14286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1497_cast_fp16")]; + string _SplitHeadsQ__mh_w_1499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1499_equation_0, values = (var_14676_cast_fp16, var_14293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1499_cast_fp16")]; + string _SplitHeadsQ__mh_w_1501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1501_equation_0, values = (var_14676_cast_fp16, var_14300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1501_cast_fp16")]; + string _SplitHeadsQ__mh_w_1503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1503_equation_0, values = (var_14676_cast_fp16, var_14307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1503_cast_fp16")]; + string _SplitHeadsQ__mh_w_1505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1505_equation_0, values = (var_14680_cast_fp16, var_14314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1505_cast_fp16")]; + string _SplitHeadsQ__mh_w_1507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1507_equation_0, values = (var_14680_cast_fp16, var_14321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1507_cast_fp16")]; + string _SplitHeadsQ__mh_w_1509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1509_equation_0, values = (var_14680_cast_fp16, var_14328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1509_cast_fp16")]; + string _SplitHeadsQ__mh_w_1511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1511_equation_0, values = (var_14680_cast_fp16, var_14335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1511_cast_fp16")]; + string _SplitHeadsQ__mh_w_1513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1513_equation_0, values = (var_14684_cast_fp16, var_14342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1513_cast_fp16")]; + string _SplitHeadsQ__mh_w_1515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1515_equation_0, values = (var_14684_cast_fp16, var_14349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1515_cast_fp16")]; + string _SplitHeadsQ__mh_w_1517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1517_equation_0, values = (var_14684_cast_fp16, var_14356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1517_cast_fp16")]; + string _SplitHeadsQ__mh_w_1519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1519_equation_0, values = (var_14684_cast_fp16, var_14363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1519_cast_fp16")]; + string _SplitHeadsQ__mh_w_1521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1521_equation_0, values = (var_14688_cast_fp16, var_14370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1521_cast_fp16")]; + string _SplitHeadsQ__mh_w_1523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1523_equation_0, values = (var_14688_cast_fp16, var_14377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1523_cast_fp16")]; + string _SplitHeadsQ__mh_w_1525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1525_equation_0, values = (var_14688_cast_fp16, var_14384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1525_cast_fp16")]; + string _SplitHeadsQ__mh_w_1527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1527_equation_0, values = (var_14688_cast_fp16, var_14391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1527_cast_fp16")]; + string _SplitHeadsQ__mh_w_1529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1529_equation_0, values = (var_14692_cast_fp16, var_14398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1529_cast_fp16")]; + string _SplitHeadsQ__mh_w_1531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1531_equation_0, values = (var_14692_cast_fp16, var_14405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1531_cast_fp16")]; + string _SplitHeadsQ__mh_w_1533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1533_equation_0, values = (var_14692_cast_fp16, var_14412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1533_cast_fp16")]; + string _SplitHeadsQ__mh_w_1535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1535_equation_0, values = (var_14692_cast_fp16, var_14419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1535_cast_fp16")]; + string _SplitHeadsQ__mh_w_1537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1537_equation_0, values = (var_14696_cast_fp16, var_14426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1537_cast_fp16")]; + string _SplitHeadsQ__mh_w_1539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1539_equation_0, values = (var_14696_cast_fp16, var_14433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1539_cast_fp16")]; + string _SplitHeadsQ__mh_w_1541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1541_equation_0, values = (var_14696_cast_fp16, var_14440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1541_cast_fp16")]; + string _SplitHeadsQ__mh_w_1543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1543_equation_0, values = (var_14696_cast_fp16, var_14447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1543_cast_fp16")]; + string _SplitHeadsQ__mh_w_1545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1545_equation_0, values = (var_14700_cast_fp16, var_14454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1545_cast_fp16")]; + string _SplitHeadsQ__mh_w_1547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1547_equation_0, values = (var_14700_cast_fp16, var_14461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1547_cast_fp16")]; + string _SplitHeadsQ__mh_w_1549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1549_equation_0, values = (var_14700_cast_fp16, var_14468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1549_cast_fp16")]; + string _SplitHeadsQ__mh_w_1551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1551_equation_0, values = (var_14700_cast_fp16, var_14475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1551_cast_fp16")]; + string _SplitHeadsQ__mh_w_1553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1553_equation_0, values = (var_14704_cast_fp16, var_14482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1553_cast_fp16")]; + string _SplitHeadsQ__mh_w_1555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1555_equation_0, values = (var_14704_cast_fp16, var_14489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1555_cast_fp16")]; + string _SplitHeadsQ__mh_w_1557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1557_equation_0, values = (var_14704_cast_fp16, var_14496_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1557_cast_fp16")]; + string _SplitHeadsQ__mh_w_1559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1559_equation_0, values = (var_14704_cast_fp16, var_14503_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1559_cast_fp16")]; + string _SplitHeadsQ__mh_w_1561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1561_equation_0, values = (var_14708_cast_fp16, var_14510_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1561_cast_fp16")]; + string _SplitHeadsQ__mh_w_1563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1563_equation_0, values = (var_14708_cast_fp16, var_14517_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1563_cast_fp16")]; + string _SplitHeadsQ__mh_w_1565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1565_equation_0, values = (var_14708_cast_fp16, var_14524_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1565_cast_fp16")]; + string _SplitHeadsQ__mh_w_1567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1567_equation_0, values = (var_14708_cast_fp16, var_14531_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1567_cast_fp16")]; + string _SplitHeadsQ__mh_w_1569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1569_equation_0, values = (var_14712_cast_fp16, var_14538_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1569_cast_fp16")]; + string _SplitHeadsQ__mh_w_1571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1571_equation_0, values = (var_14712_cast_fp16, var_14545_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1571_cast_fp16")]; + string _SplitHeadsQ__mh_w_1573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1573_equation_0, values = (var_14712_cast_fp16, var_14552_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1573_cast_fp16")]; + string _SplitHeadsQ__mh_w_1575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1575_equation_0, values = (var_14712_cast_fp16, var_14559_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1575_cast_fp16")]; + string _SplitHeadsQ__mh_w_1577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1577_equation_0, values = (var_14716_cast_fp16, var_14566_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1577_cast_fp16")]; + string _SplitHeadsQ__mh_w_1579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1579_equation_0, values = (var_14716_cast_fp16, var_14573_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1579_cast_fp16")]; + string _SplitHeadsQ__mh_w_1581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1581_equation_0, values = (var_14716_cast_fp16, var_14580_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1581_cast_fp16")]; + string _SplitHeadsQ__mh_w_1583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1583_equation_0, values = (var_14716_cast_fp16, var_14587_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1583_cast_fp16")]; + string _SplitHeadsQ__mh_w_1585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1585_equation_0, values = (var_14720_cast_fp16, var_14594_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1585_cast_fp16")]; + string _SplitHeadsQ__mh_w_1587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1587_equation_0, values = (var_14720_cast_fp16, var_14601_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1587_cast_fp16")]; + string _SplitHeadsQ__mh_w_1589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1589_equation_0, values = (var_14720_cast_fp16, var_14608_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1589_cast_fp16")]; + string _SplitHeadsQ__mh_w_1591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1591_equation_0, values = (var_14720_cast_fp16, var_14615_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1591_cast_fp16")]; + string _SplitHeadsQ__mh_w_1593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1593_equation_0, values = (var_14724_cast_fp16, var_14622_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1593_cast_fp16")]; + string _SplitHeadsQ__mh_w_1595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1595_equation_0, values = (var_14724_cast_fp16, var_14629_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1595_cast_fp16")]; + string _SplitHeadsQ__mh_w_1597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1597_equation_0, values = (var_14724_cast_fp16, var_14636_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1597_cast_fp16")]; + string _SplitHeadsQ__mh_w_1599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1599_equation_0, values = (var_14724_cast_fp16, var_14643_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1599_cast_fp16")]; + fp16 var_14965_to_fp16 = const()[name = string("op_14965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1441_cast_fp16, y = var_14965_to_fp16)[name = string("aw_chunk_1441_cast_fp16")]; + fp16 var_14967_to_fp16 = const()[name = string("op_14967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1443_cast_fp16, y = var_14967_to_fp16)[name = string("aw_chunk_1443_cast_fp16")]; + fp16 var_14969_to_fp16 = const()[name = string("op_14969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1445_cast_fp16, y = var_14969_to_fp16)[name = string("aw_chunk_1445_cast_fp16")]; + fp16 var_14971_to_fp16 = const()[name = string("op_14971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1447_cast_fp16, y = var_14971_to_fp16)[name = string("aw_chunk_1447_cast_fp16")]; + fp16 var_14973_to_fp16 = const()[name = string("op_14973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1449_cast_fp16, y = var_14973_to_fp16)[name = string("aw_chunk_1449_cast_fp16")]; + fp16 var_14975_to_fp16 = const()[name = string("op_14975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1451_cast_fp16, y = var_14975_to_fp16)[name = string("aw_chunk_1451_cast_fp16")]; + fp16 var_14977_to_fp16 = const()[name = string("op_14977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1453_cast_fp16, y = var_14977_to_fp16)[name = string("aw_chunk_1453_cast_fp16")]; + fp16 var_14979_to_fp16 = const()[name = string("op_14979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1455_cast_fp16, y = var_14979_to_fp16)[name = string("aw_chunk_1455_cast_fp16")]; + fp16 var_14981_to_fp16 = const()[name = string("op_14981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1457_cast_fp16, y = var_14981_to_fp16)[name = string("aw_chunk_1457_cast_fp16")]; + fp16 var_14983_to_fp16 = const()[name = string("op_14983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1459_cast_fp16, y = var_14983_to_fp16)[name = string("aw_chunk_1459_cast_fp16")]; + fp16 var_14985_to_fp16 = const()[name = string("op_14985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1461_cast_fp16, y = var_14985_to_fp16)[name = string("aw_chunk_1461_cast_fp16")]; + fp16 var_14987_to_fp16 = const()[name = string("op_14987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1463_cast_fp16, y = var_14987_to_fp16)[name = string("aw_chunk_1463_cast_fp16")]; + fp16 var_14989_to_fp16 = const()[name = string("op_14989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1465_cast_fp16, y = var_14989_to_fp16)[name = string("aw_chunk_1465_cast_fp16")]; + fp16 var_14991_to_fp16 = const()[name = string("op_14991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1467_cast_fp16, y = var_14991_to_fp16)[name = string("aw_chunk_1467_cast_fp16")]; + fp16 var_14993_to_fp16 = const()[name = string("op_14993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1469_cast_fp16, y = var_14993_to_fp16)[name = string("aw_chunk_1469_cast_fp16")]; + fp16 var_14995_to_fp16 = const()[name = string("op_14995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1471_cast_fp16, y = var_14995_to_fp16)[name = string("aw_chunk_1471_cast_fp16")]; + fp16 var_14997_to_fp16 = const()[name = string("op_14997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1473_cast_fp16, y = var_14997_to_fp16)[name = string("aw_chunk_1473_cast_fp16")]; + fp16 var_14999_to_fp16 = const()[name = string("op_14999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1475_cast_fp16, y = var_14999_to_fp16)[name = string("aw_chunk_1475_cast_fp16")]; + fp16 var_15001_to_fp16 = const()[name = string("op_15001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1477_cast_fp16, y = var_15001_to_fp16)[name = string("aw_chunk_1477_cast_fp16")]; + fp16 var_15003_to_fp16 = const()[name = string("op_15003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1479_cast_fp16, y = var_15003_to_fp16)[name = string("aw_chunk_1479_cast_fp16")]; + fp16 var_15005_to_fp16 = const()[name = string("op_15005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1481_cast_fp16, y = var_15005_to_fp16)[name = string("aw_chunk_1481_cast_fp16")]; + fp16 var_15007_to_fp16 = const()[name = string("op_15007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1483_cast_fp16, y = var_15007_to_fp16)[name = string("aw_chunk_1483_cast_fp16")]; + fp16 var_15009_to_fp16 = const()[name = string("op_15009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1485_cast_fp16, y = var_15009_to_fp16)[name = string("aw_chunk_1485_cast_fp16")]; + fp16 var_15011_to_fp16 = const()[name = string("op_15011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1487_cast_fp16, y = var_15011_to_fp16)[name = string("aw_chunk_1487_cast_fp16")]; + fp16 var_15013_to_fp16 = const()[name = string("op_15013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1489_cast_fp16, y = var_15013_to_fp16)[name = string("aw_chunk_1489_cast_fp16")]; + fp16 var_15015_to_fp16 = const()[name = string("op_15015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1491_cast_fp16, y = var_15015_to_fp16)[name = string("aw_chunk_1491_cast_fp16")]; + fp16 var_15017_to_fp16 = const()[name = string("op_15017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1493_cast_fp16, y = var_15017_to_fp16)[name = string("aw_chunk_1493_cast_fp16")]; + fp16 var_15019_to_fp16 = const()[name = string("op_15019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1495_cast_fp16, y = var_15019_to_fp16)[name = string("aw_chunk_1495_cast_fp16")]; + fp16 var_15021_to_fp16 = const()[name = string("op_15021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1497_cast_fp16, y = var_15021_to_fp16)[name = string("aw_chunk_1497_cast_fp16")]; + fp16 var_15023_to_fp16 = const()[name = string("op_15023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1499_cast_fp16, y = var_15023_to_fp16)[name = string("aw_chunk_1499_cast_fp16")]; + fp16 var_15025_to_fp16 = const()[name = string("op_15025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1501_cast_fp16, y = var_15025_to_fp16)[name = string("aw_chunk_1501_cast_fp16")]; + fp16 var_15027_to_fp16 = const()[name = string("op_15027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1503_cast_fp16, y = var_15027_to_fp16)[name = string("aw_chunk_1503_cast_fp16")]; + fp16 var_15029_to_fp16 = const()[name = string("op_15029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1505_cast_fp16, y = var_15029_to_fp16)[name = string("aw_chunk_1505_cast_fp16")]; + fp16 var_15031_to_fp16 = const()[name = string("op_15031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1507_cast_fp16, y = var_15031_to_fp16)[name = string("aw_chunk_1507_cast_fp16")]; + fp16 var_15033_to_fp16 = const()[name = string("op_15033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1509_cast_fp16, y = var_15033_to_fp16)[name = string("aw_chunk_1509_cast_fp16")]; + fp16 var_15035_to_fp16 = const()[name = string("op_15035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1511_cast_fp16, y = var_15035_to_fp16)[name = string("aw_chunk_1511_cast_fp16")]; + fp16 var_15037_to_fp16 = const()[name = string("op_15037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1513_cast_fp16, y = var_15037_to_fp16)[name = string("aw_chunk_1513_cast_fp16")]; + fp16 var_15039_to_fp16 = const()[name = string("op_15039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1515_cast_fp16, y = var_15039_to_fp16)[name = string("aw_chunk_1515_cast_fp16")]; + fp16 var_15041_to_fp16 = const()[name = string("op_15041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1517_cast_fp16, y = var_15041_to_fp16)[name = string("aw_chunk_1517_cast_fp16")]; + fp16 var_15043_to_fp16 = const()[name = string("op_15043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1519_cast_fp16, y = var_15043_to_fp16)[name = string("aw_chunk_1519_cast_fp16")]; + fp16 var_15045_to_fp16 = const()[name = string("op_15045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1521_cast_fp16, y = var_15045_to_fp16)[name = string("aw_chunk_1521_cast_fp16")]; + fp16 var_15047_to_fp16 = const()[name = string("op_15047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1523_cast_fp16, y = var_15047_to_fp16)[name = string("aw_chunk_1523_cast_fp16")]; + fp16 var_15049_to_fp16 = const()[name = string("op_15049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1525_cast_fp16, y = var_15049_to_fp16)[name = string("aw_chunk_1525_cast_fp16")]; + fp16 var_15051_to_fp16 = const()[name = string("op_15051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1527_cast_fp16, y = var_15051_to_fp16)[name = string("aw_chunk_1527_cast_fp16")]; + fp16 var_15053_to_fp16 = const()[name = string("op_15053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1529_cast_fp16, y = var_15053_to_fp16)[name = string("aw_chunk_1529_cast_fp16")]; + fp16 var_15055_to_fp16 = const()[name = string("op_15055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1531_cast_fp16, y = var_15055_to_fp16)[name = string("aw_chunk_1531_cast_fp16")]; + fp16 var_15057_to_fp16 = const()[name = string("op_15057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1533_cast_fp16, y = var_15057_to_fp16)[name = string("aw_chunk_1533_cast_fp16")]; + fp16 var_15059_to_fp16 = const()[name = string("op_15059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1535_cast_fp16, y = var_15059_to_fp16)[name = string("aw_chunk_1535_cast_fp16")]; + fp16 var_15061_to_fp16 = const()[name = string("op_15061_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1537_cast_fp16, y = var_15061_to_fp16)[name = string("aw_chunk_1537_cast_fp16")]; + fp16 var_15063_to_fp16 = const()[name = string("op_15063_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1539_cast_fp16, y = var_15063_to_fp16)[name = string("aw_chunk_1539_cast_fp16")]; + fp16 var_15065_to_fp16 = const()[name = string("op_15065_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1541_cast_fp16, y = var_15065_to_fp16)[name = string("aw_chunk_1541_cast_fp16")]; + fp16 var_15067_to_fp16 = const()[name = string("op_15067_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1543_cast_fp16, y = var_15067_to_fp16)[name = string("aw_chunk_1543_cast_fp16")]; + fp16 var_15069_to_fp16 = const()[name = string("op_15069_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1545_cast_fp16, y = var_15069_to_fp16)[name = string("aw_chunk_1545_cast_fp16")]; + fp16 var_15071_to_fp16 = const()[name = string("op_15071_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1547_cast_fp16, y = var_15071_to_fp16)[name = string("aw_chunk_1547_cast_fp16")]; + fp16 var_15073_to_fp16 = const()[name = string("op_15073_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1549_cast_fp16, y = var_15073_to_fp16)[name = string("aw_chunk_1549_cast_fp16")]; + fp16 var_15075_to_fp16 = const()[name = string("op_15075_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1551_cast_fp16, y = var_15075_to_fp16)[name = string("aw_chunk_1551_cast_fp16")]; + fp16 var_15077_to_fp16 = const()[name = string("op_15077_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1553_cast_fp16, y = var_15077_to_fp16)[name = string("aw_chunk_1553_cast_fp16")]; + fp16 var_15079_to_fp16 = const()[name = string("op_15079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1555_cast_fp16, y = var_15079_to_fp16)[name = string("aw_chunk_1555_cast_fp16")]; + fp16 var_15081_to_fp16 = const()[name = string("op_15081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1557_cast_fp16, y = var_15081_to_fp16)[name = string("aw_chunk_1557_cast_fp16")]; + fp16 var_15083_to_fp16 = const()[name = string("op_15083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1559_cast_fp16, y = var_15083_to_fp16)[name = string("aw_chunk_1559_cast_fp16")]; + fp16 var_15085_to_fp16 = const()[name = string("op_15085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1561_cast_fp16, y = var_15085_to_fp16)[name = string("aw_chunk_1561_cast_fp16")]; + fp16 var_15087_to_fp16 = const()[name = string("op_15087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1563_cast_fp16, y = var_15087_to_fp16)[name = string("aw_chunk_1563_cast_fp16")]; + fp16 var_15089_to_fp16 = const()[name = string("op_15089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1565_cast_fp16, y = var_15089_to_fp16)[name = string("aw_chunk_1565_cast_fp16")]; + fp16 var_15091_to_fp16 = const()[name = string("op_15091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1567_cast_fp16, y = var_15091_to_fp16)[name = string("aw_chunk_1567_cast_fp16")]; + fp16 var_15093_to_fp16 = const()[name = string("op_15093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1569_cast_fp16, y = var_15093_to_fp16)[name = string("aw_chunk_1569_cast_fp16")]; + fp16 var_15095_to_fp16 = const()[name = string("op_15095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1571_cast_fp16, y = var_15095_to_fp16)[name = string("aw_chunk_1571_cast_fp16")]; + fp16 var_15097_to_fp16 = const()[name = string("op_15097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1573_cast_fp16, y = var_15097_to_fp16)[name = string("aw_chunk_1573_cast_fp16")]; + fp16 var_15099_to_fp16 = const()[name = string("op_15099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1575_cast_fp16, y = var_15099_to_fp16)[name = string("aw_chunk_1575_cast_fp16")]; + fp16 var_15101_to_fp16 = const()[name = string("op_15101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1577_cast_fp16, y = var_15101_to_fp16)[name = string("aw_chunk_1577_cast_fp16")]; + fp16 var_15103_to_fp16 = const()[name = string("op_15103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1579_cast_fp16, y = var_15103_to_fp16)[name = string("aw_chunk_1579_cast_fp16")]; + fp16 var_15105_to_fp16 = const()[name = string("op_15105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1581_cast_fp16, y = var_15105_to_fp16)[name = string("aw_chunk_1581_cast_fp16")]; + fp16 var_15107_to_fp16 = const()[name = string("op_15107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1583_cast_fp16, y = var_15107_to_fp16)[name = string("aw_chunk_1583_cast_fp16")]; + fp16 var_15109_to_fp16 = const()[name = string("op_15109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1585_cast_fp16, y = var_15109_to_fp16)[name = string("aw_chunk_1585_cast_fp16")]; + fp16 var_15111_to_fp16 = const()[name = string("op_15111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1587_cast_fp16, y = var_15111_to_fp16)[name = string("aw_chunk_1587_cast_fp16")]; + fp16 var_15113_to_fp16 = const()[name = string("op_15113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1589_cast_fp16, y = var_15113_to_fp16)[name = string("aw_chunk_1589_cast_fp16")]; + fp16 var_15115_to_fp16 = const()[name = string("op_15115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1591_cast_fp16, y = var_15115_to_fp16)[name = string("aw_chunk_1591_cast_fp16")]; + fp16 var_15117_to_fp16 = const()[name = string("op_15117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1593_cast_fp16, y = var_15117_to_fp16)[name = string("aw_chunk_1593_cast_fp16")]; + fp16 var_15119_to_fp16 = const()[name = string("op_15119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1595_cast_fp16, y = var_15119_to_fp16)[name = string("aw_chunk_1595_cast_fp16")]; + fp16 var_15121_to_fp16 = const()[name = string("op_15121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1597_cast_fp16, y = var_15121_to_fp16)[name = string("aw_chunk_1597_cast_fp16")]; + fp16 var_15123_to_fp16 = const()[name = string("op_15123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1599_cast_fp16, y = var_15123_to_fp16)[name = string("aw_chunk_1599_cast_fp16")]; + tensor var_15125_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1441_cast_fp16)[name = string("op_15125_cast_fp16")]; + tensor var_15126_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1443_cast_fp16)[name = string("op_15126_cast_fp16")]; + tensor var_15127_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1445_cast_fp16)[name = string("op_15127_cast_fp16")]; + tensor var_15128_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1447_cast_fp16)[name = string("op_15128_cast_fp16")]; + tensor var_15129_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1449_cast_fp16)[name = string("op_15129_cast_fp16")]; + tensor var_15130_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1451_cast_fp16)[name = string("op_15130_cast_fp16")]; + tensor var_15131_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1453_cast_fp16)[name = string("op_15131_cast_fp16")]; + tensor var_15132_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1455_cast_fp16)[name = string("op_15132_cast_fp16")]; + tensor var_15133_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1457_cast_fp16)[name = string("op_15133_cast_fp16")]; + tensor var_15134_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1459_cast_fp16)[name = string("op_15134_cast_fp16")]; + tensor var_15135_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1461_cast_fp16)[name = string("op_15135_cast_fp16")]; + tensor var_15136_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1463_cast_fp16)[name = string("op_15136_cast_fp16")]; + tensor var_15137_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1465_cast_fp16)[name = string("op_15137_cast_fp16")]; + tensor var_15138_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1467_cast_fp16)[name = string("op_15138_cast_fp16")]; + tensor var_15139_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1469_cast_fp16)[name = string("op_15139_cast_fp16")]; + tensor var_15140_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1471_cast_fp16)[name = string("op_15140_cast_fp16")]; + tensor var_15141_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1473_cast_fp16)[name = string("op_15141_cast_fp16")]; + tensor var_15142_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1475_cast_fp16)[name = string("op_15142_cast_fp16")]; + tensor var_15143_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1477_cast_fp16)[name = string("op_15143_cast_fp16")]; + tensor var_15144_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1479_cast_fp16)[name = string("op_15144_cast_fp16")]; + tensor var_15145_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1481_cast_fp16)[name = string("op_15145_cast_fp16")]; + tensor var_15146_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1483_cast_fp16)[name = string("op_15146_cast_fp16")]; + tensor var_15147_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1485_cast_fp16)[name = string("op_15147_cast_fp16")]; + tensor var_15148_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1487_cast_fp16)[name = string("op_15148_cast_fp16")]; + tensor var_15149_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1489_cast_fp16)[name = string("op_15149_cast_fp16")]; + tensor var_15150_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1491_cast_fp16)[name = string("op_15150_cast_fp16")]; + tensor var_15151_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1493_cast_fp16)[name = string("op_15151_cast_fp16")]; + tensor var_15152_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1495_cast_fp16)[name = string("op_15152_cast_fp16")]; + tensor var_15153_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1497_cast_fp16)[name = string("op_15153_cast_fp16")]; + tensor var_15154_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1499_cast_fp16)[name = string("op_15154_cast_fp16")]; + tensor var_15155_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1501_cast_fp16)[name = string("op_15155_cast_fp16")]; + tensor var_15156_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1503_cast_fp16)[name = string("op_15156_cast_fp16")]; + tensor var_15157_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1505_cast_fp16)[name = string("op_15157_cast_fp16")]; + tensor var_15158_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1507_cast_fp16)[name = string("op_15158_cast_fp16")]; + tensor var_15159_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1509_cast_fp16)[name = string("op_15159_cast_fp16")]; + tensor var_15160_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1511_cast_fp16)[name = string("op_15160_cast_fp16")]; + tensor var_15161_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1513_cast_fp16)[name = string("op_15161_cast_fp16")]; + tensor var_15162_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1515_cast_fp16)[name = string("op_15162_cast_fp16")]; + tensor var_15163_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1517_cast_fp16)[name = string("op_15163_cast_fp16")]; + tensor var_15164_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1519_cast_fp16)[name = string("op_15164_cast_fp16")]; + tensor var_15165_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1521_cast_fp16)[name = string("op_15165_cast_fp16")]; + tensor var_15166_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1523_cast_fp16)[name = string("op_15166_cast_fp16")]; + tensor var_15167_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1525_cast_fp16)[name = string("op_15167_cast_fp16")]; + tensor var_15168_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1527_cast_fp16)[name = string("op_15168_cast_fp16")]; + tensor var_15169_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1529_cast_fp16)[name = string("op_15169_cast_fp16")]; + tensor var_15170_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1531_cast_fp16)[name = string("op_15170_cast_fp16")]; + tensor var_15171_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1533_cast_fp16)[name = string("op_15171_cast_fp16")]; + tensor var_15172_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1535_cast_fp16)[name = string("op_15172_cast_fp16")]; + tensor var_15173_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1537_cast_fp16)[name = string("op_15173_cast_fp16")]; + tensor var_15174_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1539_cast_fp16)[name = string("op_15174_cast_fp16")]; + tensor var_15175_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1541_cast_fp16)[name = string("op_15175_cast_fp16")]; + tensor var_15176_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1543_cast_fp16)[name = string("op_15176_cast_fp16")]; + tensor var_15177_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1545_cast_fp16)[name = string("op_15177_cast_fp16")]; + tensor var_15178_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1547_cast_fp16)[name = string("op_15178_cast_fp16")]; + tensor var_15179_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1549_cast_fp16)[name = string("op_15179_cast_fp16")]; + tensor var_15180_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1551_cast_fp16)[name = string("op_15180_cast_fp16")]; + tensor var_15181_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1553_cast_fp16)[name = string("op_15181_cast_fp16")]; + tensor var_15182_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1555_cast_fp16)[name = string("op_15182_cast_fp16")]; + tensor var_15183_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1557_cast_fp16)[name = string("op_15183_cast_fp16")]; + tensor var_15184_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1559_cast_fp16)[name = string("op_15184_cast_fp16")]; + tensor var_15185_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1561_cast_fp16)[name = string("op_15185_cast_fp16")]; + tensor var_15186_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1563_cast_fp16)[name = string("op_15186_cast_fp16")]; + tensor var_15187_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1565_cast_fp16)[name = string("op_15187_cast_fp16")]; + tensor var_15188_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1567_cast_fp16)[name = string("op_15188_cast_fp16")]; + tensor var_15189_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1569_cast_fp16)[name = string("op_15189_cast_fp16")]; + tensor var_15190_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1571_cast_fp16)[name = string("op_15190_cast_fp16")]; + tensor var_15191_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1573_cast_fp16)[name = string("op_15191_cast_fp16")]; + tensor var_15192_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1575_cast_fp16)[name = string("op_15192_cast_fp16")]; + tensor var_15193_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1577_cast_fp16)[name = string("op_15193_cast_fp16")]; + tensor var_15194_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1579_cast_fp16)[name = string("op_15194_cast_fp16")]; + tensor var_15195_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1581_cast_fp16)[name = string("op_15195_cast_fp16")]; + tensor var_15196_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1583_cast_fp16)[name = string("op_15196_cast_fp16")]; + tensor var_15197_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1585_cast_fp16)[name = string("op_15197_cast_fp16")]; + tensor var_15198_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1587_cast_fp16)[name = string("op_15198_cast_fp16")]; + tensor var_15199_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1589_cast_fp16)[name = string("op_15199_cast_fp16")]; + tensor var_15200_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1591_cast_fp16)[name = string("op_15200_cast_fp16")]; + tensor var_15201_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1593_cast_fp16)[name = string("op_15201_cast_fp16")]; + tensor var_15202_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1595_cast_fp16)[name = string("op_15202_cast_fp16")]; + tensor var_15203_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1597_cast_fp16)[name = string("op_15203_cast_fp16")]; + tensor var_15204_cast_fp16 = softmax(axis = var_13950, x = aw_chunk_1599_cast_fp16)[name = string("op_15204_cast_fp16")]; + string var_15206_equation_0 = const()[name = string("op_15206_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15206_cast_fp16 = einsum(equation = var_15206_equation_0, values = (var_14726_cast_fp16, var_15125_cast_fp16))[name = string("op_15206_cast_fp16")]; + string var_15208_equation_0 = const()[name = string("op_15208_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15208_cast_fp16 = einsum(equation = var_15208_equation_0, values = (var_14726_cast_fp16, var_15126_cast_fp16))[name = string("op_15208_cast_fp16")]; + string var_15210_equation_0 = const()[name = string("op_15210_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15210_cast_fp16 = einsum(equation = var_15210_equation_0, values = (var_14726_cast_fp16, var_15127_cast_fp16))[name = string("op_15210_cast_fp16")]; + string var_15212_equation_0 = const()[name = string("op_15212_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15212_cast_fp16 = einsum(equation = var_15212_equation_0, values = (var_14726_cast_fp16, var_15128_cast_fp16))[name = string("op_15212_cast_fp16")]; + string var_15214_equation_0 = const()[name = string("op_15214_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15214_cast_fp16 = einsum(equation = var_15214_equation_0, values = (var_14730_cast_fp16, var_15129_cast_fp16))[name = string("op_15214_cast_fp16")]; + string var_15216_equation_0 = const()[name = string("op_15216_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15216_cast_fp16 = einsum(equation = var_15216_equation_0, values = (var_14730_cast_fp16, var_15130_cast_fp16))[name = string("op_15216_cast_fp16")]; + string var_15218_equation_0 = const()[name = string("op_15218_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15218_cast_fp16 = einsum(equation = var_15218_equation_0, values = (var_14730_cast_fp16, var_15131_cast_fp16))[name = string("op_15218_cast_fp16")]; + string var_15220_equation_0 = const()[name = string("op_15220_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15220_cast_fp16 = einsum(equation = var_15220_equation_0, values = (var_14730_cast_fp16, var_15132_cast_fp16))[name = string("op_15220_cast_fp16")]; + string var_15222_equation_0 = const()[name = string("op_15222_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15222_cast_fp16 = einsum(equation = var_15222_equation_0, values = (var_14734_cast_fp16, var_15133_cast_fp16))[name = string("op_15222_cast_fp16")]; + string var_15224_equation_0 = const()[name = string("op_15224_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15224_cast_fp16 = einsum(equation = var_15224_equation_0, values = (var_14734_cast_fp16, var_15134_cast_fp16))[name = string("op_15224_cast_fp16")]; + string var_15226_equation_0 = const()[name = string("op_15226_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15226_cast_fp16 = einsum(equation = var_15226_equation_0, values = (var_14734_cast_fp16, var_15135_cast_fp16))[name = string("op_15226_cast_fp16")]; + string var_15228_equation_0 = const()[name = string("op_15228_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15228_cast_fp16 = einsum(equation = var_15228_equation_0, values = (var_14734_cast_fp16, var_15136_cast_fp16))[name = string("op_15228_cast_fp16")]; + string var_15230_equation_0 = const()[name = string("op_15230_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15230_cast_fp16 = einsum(equation = var_15230_equation_0, values = (var_14738_cast_fp16, var_15137_cast_fp16))[name = string("op_15230_cast_fp16")]; + string var_15232_equation_0 = const()[name = string("op_15232_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15232_cast_fp16 = einsum(equation = var_15232_equation_0, values = (var_14738_cast_fp16, var_15138_cast_fp16))[name = string("op_15232_cast_fp16")]; + string var_15234_equation_0 = const()[name = string("op_15234_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15234_cast_fp16 = einsum(equation = var_15234_equation_0, values = (var_14738_cast_fp16, var_15139_cast_fp16))[name = string("op_15234_cast_fp16")]; + string var_15236_equation_0 = const()[name = string("op_15236_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15236_cast_fp16 = einsum(equation = var_15236_equation_0, values = (var_14738_cast_fp16, var_15140_cast_fp16))[name = string("op_15236_cast_fp16")]; + string var_15238_equation_0 = const()[name = string("op_15238_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15238_cast_fp16 = einsum(equation = var_15238_equation_0, values = (var_14742_cast_fp16, var_15141_cast_fp16))[name = string("op_15238_cast_fp16")]; + string var_15240_equation_0 = const()[name = string("op_15240_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15240_cast_fp16 = einsum(equation = var_15240_equation_0, values = (var_14742_cast_fp16, var_15142_cast_fp16))[name = string("op_15240_cast_fp16")]; + string var_15242_equation_0 = const()[name = string("op_15242_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15242_cast_fp16 = einsum(equation = var_15242_equation_0, values = (var_14742_cast_fp16, var_15143_cast_fp16))[name = string("op_15242_cast_fp16")]; + string var_15244_equation_0 = const()[name = string("op_15244_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15244_cast_fp16 = einsum(equation = var_15244_equation_0, values = (var_14742_cast_fp16, var_15144_cast_fp16))[name = string("op_15244_cast_fp16")]; + string var_15246_equation_0 = const()[name = string("op_15246_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15246_cast_fp16 = einsum(equation = var_15246_equation_0, values = (var_14746_cast_fp16, var_15145_cast_fp16))[name = string("op_15246_cast_fp16")]; + string var_15248_equation_0 = const()[name = string("op_15248_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15248_cast_fp16 = einsum(equation = var_15248_equation_0, values = (var_14746_cast_fp16, var_15146_cast_fp16))[name = string("op_15248_cast_fp16")]; + string var_15250_equation_0 = const()[name = string("op_15250_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15250_cast_fp16 = einsum(equation = var_15250_equation_0, values = (var_14746_cast_fp16, var_15147_cast_fp16))[name = string("op_15250_cast_fp16")]; + string var_15252_equation_0 = const()[name = string("op_15252_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15252_cast_fp16 = einsum(equation = var_15252_equation_0, values = (var_14746_cast_fp16, var_15148_cast_fp16))[name = string("op_15252_cast_fp16")]; + string var_15254_equation_0 = const()[name = string("op_15254_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15254_cast_fp16 = einsum(equation = var_15254_equation_0, values = (var_14750_cast_fp16, var_15149_cast_fp16))[name = string("op_15254_cast_fp16")]; + string var_15256_equation_0 = const()[name = string("op_15256_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15256_cast_fp16 = einsum(equation = var_15256_equation_0, values = (var_14750_cast_fp16, var_15150_cast_fp16))[name = string("op_15256_cast_fp16")]; + string var_15258_equation_0 = const()[name = string("op_15258_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15258_cast_fp16 = einsum(equation = var_15258_equation_0, values = (var_14750_cast_fp16, var_15151_cast_fp16))[name = string("op_15258_cast_fp16")]; + string var_15260_equation_0 = const()[name = string("op_15260_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15260_cast_fp16 = einsum(equation = var_15260_equation_0, values = (var_14750_cast_fp16, var_15152_cast_fp16))[name = string("op_15260_cast_fp16")]; + string var_15262_equation_0 = const()[name = string("op_15262_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15262_cast_fp16 = einsum(equation = var_15262_equation_0, values = (var_14754_cast_fp16, var_15153_cast_fp16))[name = string("op_15262_cast_fp16")]; + string var_15264_equation_0 = const()[name = string("op_15264_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15264_cast_fp16 = einsum(equation = var_15264_equation_0, values = (var_14754_cast_fp16, var_15154_cast_fp16))[name = string("op_15264_cast_fp16")]; + string var_15266_equation_0 = const()[name = string("op_15266_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15266_cast_fp16 = einsum(equation = var_15266_equation_0, values = (var_14754_cast_fp16, var_15155_cast_fp16))[name = string("op_15266_cast_fp16")]; + string var_15268_equation_0 = const()[name = string("op_15268_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15268_cast_fp16 = einsum(equation = var_15268_equation_0, values = (var_14754_cast_fp16, var_15156_cast_fp16))[name = string("op_15268_cast_fp16")]; + string var_15270_equation_0 = const()[name = string("op_15270_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15270_cast_fp16 = einsum(equation = var_15270_equation_0, values = (var_14758_cast_fp16, var_15157_cast_fp16))[name = string("op_15270_cast_fp16")]; + string var_15272_equation_0 = const()[name = string("op_15272_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15272_cast_fp16 = einsum(equation = var_15272_equation_0, values = (var_14758_cast_fp16, var_15158_cast_fp16))[name = string("op_15272_cast_fp16")]; + string var_15274_equation_0 = const()[name = string("op_15274_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15274_cast_fp16 = einsum(equation = var_15274_equation_0, values = (var_14758_cast_fp16, var_15159_cast_fp16))[name = string("op_15274_cast_fp16")]; + string var_15276_equation_0 = const()[name = string("op_15276_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15276_cast_fp16 = einsum(equation = var_15276_equation_0, values = (var_14758_cast_fp16, var_15160_cast_fp16))[name = string("op_15276_cast_fp16")]; + string var_15278_equation_0 = const()[name = string("op_15278_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15278_cast_fp16 = einsum(equation = var_15278_equation_0, values = (var_14762_cast_fp16, var_15161_cast_fp16))[name = string("op_15278_cast_fp16")]; + string var_15280_equation_0 = const()[name = string("op_15280_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15280_cast_fp16 = einsum(equation = var_15280_equation_0, values = (var_14762_cast_fp16, var_15162_cast_fp16))[name = string("op_15280_cast_fp16")]; + string var_15282_equation_0 = const()[name = string("op_15282_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15282_cast_fp16 = einsum(equation = var_15282_equation_0, values = (var_14762_cast_fp16, var_15163_cast_fp16))[name = string("op_15282_cast_fp16")]; + string var_15284_equation_0 = const()[name = string("op_15284_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15284_cast_fp16 = einsum(equation = var_15284_equation_0, values = (var_14762_cast_fp16, var_15164_cast_fp16))[name = string("op_15284_cast_fp16")]; + string var_15286_equation_0 = const()[name = string("op_15286_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15286_cast_fp16 = einsum(equation = var_15286_equation_0, values = (var_14766_cast_fp16, var_15165_cast_fp16))[name = string("op_15286_cast_fp16")]; + string var_15288_equation_0 = const()[name = string("op_15288_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15288_cast_fp16 = einsum(equation = var_15288_equation_0, values = (var_14766_cast_fp16, var_15166_cast_fp16))[name = string("op_15288_cast_fp16")]; + string var_15290_equation_0 = const()[name = string("op_15290_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15290_cast_fp16 = einsum(equation = var_15290_equation_0, values = (var_14766_cast_fp16, var_15167_cast_fp16))[name = string("op_15290_cast_fp16")]; + string var_15292_equation_0 = const()[name = string("op_15292_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15292_cast_fp16 = einsum(equation = var_15292_equation_0, values = (var_14766_cast_fp16, var_15168_cast_fp16))[name = string("op_15292_cast_fp16")]; + string var_15294_equation_0 = const()[name = string("op_15294_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15294_cast_fp16 = einsum(equation = var_15294_equation_0, values = (var_14770_cast_fp16, var_15169_cast_fp16))[name = string("op_15294_cast_fp16")]; + string var_15296_equation_0 = const()[name = string("op_15296_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15296_cast_fp16 = einsum(equation = var_15296_equation_0, values = (var_14770_cast_fp16, var_15170_cast_fp16))[name = string("op_15296_cast_fp16")]; + string var_15298_equation_0 = const()[name = string("op_15298_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15298_cast_fp16 = einsum(equation = var_15298_equation_0, values = (var_14770_cast_fp16, var_15171_cast_fp16))[name = string("op_15298_cast_fp16")]; + string var_15300_equation_0 = const()[name = string("op_15300_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15300_cast_fp16 = einsum(equation = var_15300_equation_0, values = (var_14770_cast_fp16, var_15172_cast_fp16))[name = string("op_15300_cast_fp16")]; + string var_15302_equation_0 = const()[name = string("op_15302_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15302_cast_fp16 = einsum(equation = var_15302_equation_0, values = (var_14774_cast_fp16, var_15173_cast_fp16))[name = string("op_15302_cast_fp16")]; + string var_15304_equation_0 = const()[name = string("op_15304_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15304_cast_fp16 = einsum(equation = var_15304_equation_0, values = (var_14774_cast_fp16, var_15174_cast_fp16))[name = string("op_15304_cast_fp16")]; + string var_15306_equation_0 = const()[name = string("op_15306_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15306_cast_fp16 = einsum(equation = var_15306_equation_0, values = (var_14774_cast_fp16, var_15175_cast_fp16))[name = string("op_15306_cast_fp16")]; + string var_15308_equation_0 = const()[name = string("op_15308_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15308_cast_fp16 = einsum(equation = var_15308_equation_0, values = (var_14774_cast_fp16, var_15176_cast_fp16))[name = string("op_15308_cast_fp16")]; + string var_15310_equation_0 = const()[name = string("op_15310_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15310_cast_fp16 = einsum(equation = var_15310_equation_0, values = (var_14778_cast_fp16, var_15177_cast_fp16))[name = string("op_15310_cast_fp16")]; + string var_15312_equation_0 = const()[name = string("op_15312_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15312_cast_fp16 = einsum(equation = var_15312_equation_0, values = (var_14778_cast_fp16, var_15178_cast_fp16))[name = string("op_15312_cast_fp16")]; + string var_15314_equation_0 = const()[name = string("op_15314_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15314_cast_fp16 = einsum(equation = var_15314_equation_0, values = (var_14778_cast_fp16, var_15179_cast_fp16))[name = string("op_15314_cast_fp16")]; + string var_15316_equation_0 = const()[name = string("op_15316_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15316_cast_fp16 = einsum(equation = var_15316_equation_0, values = (var_14778_cast_fp16, var_15180_cast_fp16))[name = string("op_15316_cast_fp16")]; + string var_15318_equation_0 = const()[name = string("op_15318_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15318_cast_fp16 = einsum(equation = var_15318_equation_0, values = (var_14782_cast_fp16, var_15181_cast_fp16))[name = string("op_15318_cast_fp16")]; + string var_15320_equation_0 = const()[name = string("op_15320_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15320_cast_fp16 = einsum(equation = var_15320_equation_0, values = (var_14782_cast_fp16, var_15182_cast_fp16))[name = string("op_15320_cast_fp16")]; + string var_15322_equation_0 = const()[name = string("op_15322_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15322_cast_fp16 = einsum(equation = var_15322_equation_0, values = (var_14782_cast_fp16, var_15183_cast_fp16))[name = string("op_15322_cast_fp16")]; + string var_15324_equation_0 = const()[name = string("op_15324_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15324_cast_fp16 = einsum(equation = var_15324_equation_0, values = (var_14782_cast_fp16, var_15184_cast_fp16))[name = string("op_15324_cast_fp16")]; + string var_15326_equation_0 = const()[name = string("op_15326_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15326_cast_fp16 = einsum(equation = var_15326_equation_0, values = (var_14786_cast_fp16, var_15185_cast_fp16))[name = string("op_15326_cast_fp16")]; + string var_15328_equation_0 = const()[name = string("op_15328_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15328_cast_fp16 = einsum(equation = var_15328_equation_0, values = (var_14786_cast_fp16, var_15186_cast_fp16))[name = string("op_15328_cast_fp16")]; + string var_15330_equation_0 = const()[name = string("op_15330_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15330_cast_fp16 = einsum(equation = var_15330_equation_0, values = (var_14786_cast_fp16, var_15187_cast_fp16))[name = string("op_15330_cast_fp16")]; + string var_15332_equation_0 = const()[name = string("op_15332_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15332_cast_fp16 = einsum(equation = var_15332_equation_0, values = (var_14786_cast_fp16, var_15188_cast_fp16))[name = string("op_15332_cast_fp16")]; + string var_15334_equation_0 = const()[name = string("op_15334_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15334_cast_fp16 = einsum(equation = var_15334_equation_0, values = (var_14790_cast_fp16, var_15189_cast_fp16))[name = string("op_15334_cast_fp16")]; + string var_15336_equation_0 = const()[name = string("op_15336_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15336_cast_fp16 = einsum(equation = var_15336_equation_0, values = (var_14790_cast_fp16, var_15190_cast_fp16))[name = string("op_15336_cast_fp16")]; + string var_15338_equation_0 = const()[name = string("op_15338_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15338_cast_fp16 = einsum(equation = var_15338_equation_0, values = (var_14790_cast_fp16, var_15191_cast_fp16))[name = string("op_15338_cast_fp16")]; + string var_15340_equation_0 = const()[name = string("op_15340_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15340_cast_fp16 = einsum(equation = var_15340_equation_0, values = (var_14790_cast_fp16, var_15192_cast_fp16))[name = string("op_15340_cast_fp16")]; + string var_15342_equation_0 = const()[name = string("op_15342_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15342_cast_fp16 = einsum(equation = var_15342_equation_0, values = (var_14794_cast_fp16, var_15193_cast_fp16))[name = string("op_15342_cast_fp16")]; + string var_15344_equation_0 = const()[name = string("op_15344_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15344_cast_fp16 = einsum(equation = var_15344_equation_0, values = (var_14794_cast_fp16, var_15194_cast_fp16))[name = string("op_15344_cast_fp16")]; + string var_15346_equation_0 = const()[name = string("op_15346_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15346_cast_fp16 = einsum(equation = var_15346_equation_0, values = (var_14794_cast_fp16, var_15195_cast_fp16))[name = string("op_15346_cast_fp16")]; + string var_15348_equation_0 = const()[name = string("op_15348_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15348_cast_fp16 = einsum(equation = var_15348_equation_0, values = (var_14794_cast_fp16, var_15196_cast_fp16))[name = string("op_15348_cast_fp16")]; + string var_15350_equation_0 = const()[name = string("op_15350_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15350_cast_fp16 = einsum(equation = var_15350_equation_0, values = (var_14798_cast_fp16, var_15197_cast_fp16))[name = string("op_15350_cast_fp16")]; + string var_15352_equation_0 = const()[name = string("op_15352_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15352_cast_fp16 = einsum(equation = var_15352_equation_0, values = (var_14798_cast_fp16, var_15198_cast_fp16))[name = string("op_15352_cast_fp16")]; + string var_15354_equation_0 = const()[name = string("op_15354_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15354_cast_fp16 = einsum(equation = var_15354_equation_0, values = (var_14798_cast_fp16, var_15199_cast_fp16))[name = string("op_15354_cast_fp16")]; + string var_15356_equation_0 = const()[name = string("op_15356_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15356_cast_fp16 = einsum(equation = var_15356_equation_0, values = (var_14798_cast_fp16, var_15200_cast_fp16))[name = string("op_15356_cast_fp16")]; + string var_15358_equation_0 = const()[name = string("op_15358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15358_cast_fp16 = einsum(equation = var_15358_equation_0, values = (var_14802_cast_fp16, var_15201_cast_fp16))[name = string("op_15358_cast_fp16")]; + string var_15360_equation_0 = const()[name = string("op_15360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15360_cast_fp16 = einsum(equation = var_15360_equation_0, values = (var_14802_cast_fp16, var_15202_cast_fp16))[name = string("op_15360_cast_fp16")]; + string var_15362_equation_0 = const()[name = string("op_15362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15362_cast_fp16 = einsum(equation = var_15362_equation_0, values = (var_14802_cast_fp16, var_15203_cast_fp16))[name = string("op_15362_cast_fp16")]; + string var_15364_equation_0 = const()[name = string("op_15364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_15364_cast_fp16 = einsum(equation = var_15364_equation_0, values = (var_14802_cast_fp16, var_15204_cast_fp16))[name = string("op_15364_cast_fp16")]; + bool var_15366_interleave_0 = const()[name = string("op_15366_interleave_0"), val = bool(false)]; + tensor var_15366_cast_fp16 = concat(axis = var_13925, interleave = var_15366_interleave_0, values = (var_15206_cast_fp16, var_15208_cast_fp16, var_15210_cast_fp16, var_15212_cast_fp16))[name = string("op_15366_cast_fp16")]; + bool var_15368_interleave_0 = const()[name = string("op_15368_interleave_0"), val = bool(false)]; + tensor var_15368_cast_fp16 = concat(axis = var_13925, interleave = var_15368_interleave_0, values = (var_15214_cast_fp16, var_15216_cast_fp16, var_15218_cast_fp16, var_15220_cast_fp16))[name = string("op_15368_cast_fp16")]; + bool var_15370_interleave_0 = const()[name = string("op_15370_interleave_0"), val = bool(false)]; + tensor var_15370_cast_fp16 = concat(axis = var_13925, interleave = var_15370_interleave_0, values = (var_15222_cast_fp16, var_15224_cast_fp16, var_15226_cast_fp16, var_15228_cast_fp16))[name = string("op_15370_cast_fp16")]; + bool var_15372_interleave_0 = const()[name = string("op_15372_interleave_0"), val = bool(false)]; + tensor var_15372_cast_fp16 = concat(axis = var_13925, interleave = var_15372_interleave_0, values = (var_15230_cast_fp16, var_15232_cast_fp16, var_15234_cast_fp16, var_15236_cast_fp16))[name = string("op_15372_cast_fp16")]; + bool var_15374_interleave_0 = const()[name = string("op_15374_interleave_0"), val = bool(false)]; + tensor var_15374_cast_fp16 = concat(axis = var_13925, interleave = var_15374_interleave_0, values = (var_15238_cast_fp16, var_15240_cast_fp16, var_15242_cast_fp16, var_15244_cast_fp16))[name = string("op_15374_cast_fp16")]; + bool var_15376_interleave_0 = const()[name = string("op_15376_interleave_0"), val = bool(false)]; + tensor var_15376_cast_fp16 = concat(axis = var_13925, interleave = var_15376_interleave_0, values = (var_15246_cast_fp16, var_15248_cast_fp16, var_15250_cast_fp16, var_15252_cast_fp16))[name = string("op_15376_cast_fp16")]; + bool var_15378_interleave_0 = const()[name = string("op_15378_interleave_0"), val = bool(false)]; + tensor var_15378_cast_fp16 = concat(axis = var_13925, interleave = var_15378_interleave_0, values = (var_15254_cast_fp16, var_15256_cast_fp16, var_15258_cast_fp16, var_15260_cast_fp16))[name = string("op_15378_cast_fp16")]; + bool var_15380_interleave_0 = const()[name = string("op_15380_interleave_0"), val = bool(false)]; + tensor var_15380_cast_fp16 = concat(axis = var_13925, interleave = var_15380_interleave_0, values = (var_15262_cast_fp16, var_15264_cast_fp16, var_15266_cast_fp16, var_15268_cast_fp16))[name = string("op_15380_cast_fp16")]; + bool var_15382_interleave_0 = const()[name = string("op_15382_interleave_0"), val = bool(false)]; + tensor var_15382_cast_fp16 = concat(axis = var_13925, interleave = var_15382_interleave_0, values = (var_15270_cast_fp16, var_15272_cast_fp16, var_15274_cast_fp16, var_15276_cast_fp16))[name = string("op_15382_cast_fp16")]; + bool var_15384_interleave_0 = const()[name = string("op_15384_interleave_0"), val = bool(false)]; + tensor var_15384_cast_fp16 = concat(axis = var_13925, interleave = var_15384_interleave_0, values = (var_15278_cast_fp16, var_15280_cast_fp16, var_15282_cast_fp16, var_15284_cast_fp16))[name = string("op_15384_cast_fp16")]; + bool var_15386_interleave_0 = const()[name = string("op_15386_interleave_0"), val = bool(false)]; + tensor var_15386_cast_fp16 = concat(axis = var_13925, interleave = var_15386_interleave_0, values = (var_15286_cast_fp16, var_15288_cast_fp16, var_15290_cast_fp16, var_15292_cast_fp16))[name = string("op_15386_cast_fp16")]; + bool var_15388_interleave_0 = const()[name = string("op_15388_interleave_0"), val = bool(false)]; + tensor var_15388_cast_fp16 = concat(axis = var_13925, interleave = var_15388_interleave_0, values = (var_15294_cast_fp16, var_15296_cast_fp16, var_15298_cast_fp16, var_15300_cast_fp16))[name = string("op_15388_cast_fp16")]; + bool var_15390_interleave_0 = const()[name = string("op_15390_interleave_0"), val = bool(false)]; + tensor var_15390_cast_fp16 = concat(axis = var_13925, interleave = var_15390_interleave_0, values = (var_15302_cast_fp16, var_15304_cast_fp16, var_15306_cast_fp16, var_15308_cast_fp16))[name = string("op_15390_cast_fp16")]; + bool var_15392_interleave_0 = const()[name = string("op_15392_interleave_0"), val = bool(false)]; + tensor var_15392_cast_fp16 = concat(axis = var_13925, interleave = var_15392_interleave_0, values = (var_15310_cast_fp16, var_15312_cast_fp16, var_15314_cast_fp16, var_15316_cast_fp16))[name = string("op_15392_cast_fp16")]; + bool var_15394_interleave_0 = const()[name = string("op_15394_interleave_0"), val = bool(false)]; + tensor var_15394_cast_fp16 = concat(axis = var_13925, interleave = var_15394_interleave_0, values = (var_15318_cast_fp16, var_15320_cast_fp16, var_15322_cast_fp16, var_15324_cast_fp16))[name = string("op_15394_cast_fp16")]; + bool var_15396_interleave_0 = const()[name = string("op_15396_interleave_0"), val = bool(false)]; + tensor var_15396_cast_fp16 = concat(axis = var_13925, interleave = var_15396_interleave_0, values = (var_15326_cast_fp16, var_15328_cast_fp16, var_15330_cast_fp16, var_15332_cast_fp16))[name = string("op_15396_cast_fp16")]; + bool var_15398_interleave_0 = const()[name = string("op_15398_interleave_0"), val = bool(false)]; + tensor var_15398_cast_fp16 = concat(axis = var_13925, interleave = var_15398_interleave_0, values = (var_15334_cast_fp16, var_15336_cast_fp16, var_15338_cast_fp16, var_15340_cast_fp16))[name = string("op_15398_cast_fp16")]; + bool var_15400_interleave_0 = const()[name = string("op_15400_interleave_0"), val = bool(false)]; + tensor var_15400_cast_fp16 = concat(axis = var_13925, interleave = var_15400_interleave_0, values = (var_15342_cast_fp16, var_15344_cast_fp16, var_15346_cast_fp16, var_15348_cast_fp16))[name = string("op_15400_cast_fp16")]; + bool var_15402_interleave_0 = const()[name = string("op_15402_interleave_0"), val = bool(false)]; + tensor var_15402_cast_fp16 = concat(axis = var_13925, interleave = var_15402_interleave_0, values = (var_15350_cast_fp16, var_15352_cast_fp16, var_15354_cast_fp16, var_15356_cast_fp16))[name = string("op_15402_cast_fp16")]; + bool var_15404_interleave_0 = const()[name = string("op_15404_interleave_0"), val = bool(false)]; + tensor var_15404_cast_fp16 = concat(axis = var_13925, interleave = var_15404_interleave_0, values = (var_15358_cast_fp16, var_15360_cast_fp16, var_15362_cast_fp16, var_15364_cast_fp16))[name = string("op_15404_cast_fp16")]; + bool input_73_interleave_0 = const()[name = string("input_73_interleave_0"), val = bool(false)]; + tensor input_73_cast_fp16 = concat(axis = var_13950, interleave = input_73_interleave_0, values = (var_15366_cast_fp16, var_15368_cast_fp16, var_15370_cast_fp16, var_15372_cast_fp16, var_15374_cast_fp16, var_15376_cast_fp16, var_15378_cast_fp16, var_15380_cast_fp16, var_15382_cast_fp16, var_15384_cast_fp16, var_15386_cast_fp16, var_15388_cast_fp16, var_15390_cast_fp16, var_15392_cast_fp16, var_15394_cast_fp16, var_15396_cast_fp16, var_15398_cast_fp16, var_15400_cast_fp16, var_15402_cast_fp16, var_15404_cast_fp16))[name = string("input_73_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378684800)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381961664)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_15423_to_fp16 = const()[name = string("op_15423_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_15423_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381964288)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381966912)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381969536)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395076800)))]; + tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395087104)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408194368)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; + int32 var_15452 = const()[name = string("op_15452"), val = int32(3)]; + int32 var_15477 = const()[name = string("op_15477"), val = int32(1)]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_15494_to_fp16 = const()[name = string("op_15494_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_15494_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408196992)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408199616)))]; + fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408202240)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411479104)))]; + tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; + string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; + tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; + tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; + int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411481728)))]; + tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; + string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; + tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; + tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; + int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414758592)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418035456)))]; + tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_15532_begin_0 = const()[name = string("op_15532_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15532_end_0 = const()[name = string("op_15532_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15532_end_mask_0 = const()[name = string("op_15532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15532_cast_fp16 = slice_by_index(begin = var_15532_begin_0, end = var_15532_end_0, end_mask = var_15532_end_mask_0, x = query_21_cast_fp16)[name = string("op_15532_cast_fp16")]; + tensor var_15536_begin_0 = const()[name = string("op_15536_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15536_end_0 = const()[name = string("op_15536_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_15536_end_mask_0 = const()[name = string("op_15536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15536_cast_fp16 = slice_by_index(begin = var_15536_begin_0, end = var_15536_end_0, end_mask = var_15536_end_mask_0, x = query_21_cast_fp16)[name = string("op_15536_cast_fp16")]; + tensor var_15540_begin_0 = const()[name = string("op_15540_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15540_end_0 = const()[name = string("op_15540_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_15540_end_mask_0 = const()[name = string("op_15540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15540_cast_fp16 = slice_by_index(begin = var_15540_begin_0, end = var_15540_end_0, end_mask = var_15540_end_mask_0, x = query_21_cast_fp16)[name = string("op_15540_cast_fp16")]; + tensor var_15544_begin_0 = const()[name = string("op_15544_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15544_end_0 = const()[name = string("op_15544_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_15544_end_mask_0 = const()[name = string("op_15544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15544_cast_fp16 = slice_by_index(begin = var_15544_begin_0, end = var_15544_end_0, end_mask = var_15544_end_mask_0, x = query_21_cast_fp16)[name = string("op_15544_cast_fp16")]; + tensor var_15548_begin_0 = const()[name = string("op_15548_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15548_end_0 = const()[name = string("op_15548_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_15548_end_mask_0 = const()[name = string("op_15548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15548_cast_fp16 = slice_by_index(begin = var_15548_begin_0, end = var_15548_end_0, end_mask = var_15548_end_mask_0, x = query_21_cast_fp16)[name = string("op_15548_cast_fp16")]; + tensor var_15552_begin_0 = const()[name = string("op_15552_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15552_end_0 = const()[name = string("op_15552_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_15552_end_mask_0 = const()[name = string("op_15552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15552_cast_fp16 = slice_by_index(begin = var_15552_begin_0, end = var_15552_end_0, end_mask = var_15552_end_mask_0, x = query_21_cast_fp16)[name = string("op_15552_cast_fp16")]; + tensor var_15556_begin_0 = const()[name = string("op_15556_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15556_end_0 = const()[name = string("op_15556_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_15556_end_mask_0 = const()[name = string("op_15556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15556_cast_fp16 = slice_by_index(begin = var_15556_begin_0, end = var_15556_end_0, end_mask = var_15556_end_mask_0, x = query_21_cast_fp16)[name = string("op_15556_cast_fp16")]; + tensor var_15560_begin_0 = const()[name = string("op_15560_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15560_end_0 = const()[name = string("op_15560_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_15560_end_mask_0 = const()[name = string("op_15560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15560_cast_fp16 = slice_by_index(begin = var_15560_begin_0, end = var_15560_end_0, end_mask = var_15560_end_mask_0, x = query_21_cast_fp16)[name = string("op_15560_cast_fp16")]; + tensor var_15564_begin_0 = const()[name = string("op_15564_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15564_end_0 = const()[name = string("op_15564_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_15564_end_mask_0 = const()[name = string("op_15564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15564_cast_fp16 = slice_by_index(begin = var_15564_begin_0, end = var_15564_end_0, end_mask = var_15564_end_mask_0, x = query_21_cast_fp16)[name = string("op_15564_cast_fp16")]; + tensor var_15568_begin_0 = const()[name = string("op_15568_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15568_end_0 = const()[name = string("op_15568_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_15568_end_mask_0 = const()[name = string("op_15568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15568_cast_fp16 = slice_by_index(begin = var_15568_begin_0, end = var_15568_end_0, end_mask = var_15568_end_mask_0, x = query_21_cast_fp16)[name = string("op_15568_cast_fp16")]; + tensor var_15572_begin_0 = const()[name = string("op_15572_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15572_end_0 = const()[name = string("op_15572_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_15572_end_mask_0 = const()[name = string("op_15572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15572_cast_fp16 = slice_by_index(begin = var_15572_begin_0, end = var_15572_end_0, end_mask = var_15572_end_mask_0, x = query_21_cast_fp16)[name = string("op_15572_cast_fp16")]; + tensor var_15576_begin_0 = const()[name = string("op_15576_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15576_end_0 = const()[name = string("op_15576_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_15576_end_mask_0 = const()[name = string("op_15576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15576_cast_fp16 = slice_by_index(begin = var_15576_begin_0, end = var_15576_end_0, end_mask = var_15576_end_mask_0, x = query_21_cast_fp16)[name = string("op_15576_cast_fp16")]; + tensor var_15580_begin_0 = const()[name = string("op_15580_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15580_end_0 = const()[name = string("op_15580_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_15580_end_mask_0 = const()[name = string("op_15580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15580_cast_fp16 = slice_by_index(begin = var_15580_begin_0, end = var_15580_end_0, end_mask = var_15580_end_mask_0, x = query_21_cast_fp16)[name = string("op_15580_cast_fp16")]; + tensor var_15584_begin_0 = const()[name = string("op_15584_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15584_end_0 = const()[name = string("op_15584_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_15584_end_mask_0 = const()[name = string("op_15584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15584_cast_fp16 = slice_by_index(begin = var_15584_begin_0, end = var_15584_end_0, end_mask = var_15584_end_mask_0, x = query_21_cast_fp16)[name = string("op_15584_cast_fp16")]; + tensor var_15588_begin_0 = const()[name = string("op_15588_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15588_end_0 = const()[name = string("op_15588_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_15588_end_mask_0 = const()[name = string("op_15588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15588_cast_fp16 = slice_by_index(begin = var_15588_begin_0, end = var_15588_end_0, end_mask = var_15588_end_mask_0, x = query_21_cast_fp16)[name = string("op_15588_cast_fp16")]; + tensor var_15592_begin_0 = const()[name = string("op_15592_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15592_end_0 = const()[name = string("op_15592_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_15592_end_mask_0 = const()[name = string("op_15592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15592_cast_fp16 = slice_by_index(begin = var_15592_begin_0, end = var_15592_end_0, end_mask = var_15592_end_mask_0, x = query_21_cast_fp16)[name = string("op_15592_cast_fp16")]; + tensor var_15596_begin_0 = const()[name = string("op_15596_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15596_end_0 = const()[name = string("op_15596_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_15596_end_mask_0 = const()[name = string("op_15596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15596_cast_fp16 = slice_by_index(begin = var_15596_begin_0, end = var_15596_end_0, end_mask = var_15596_end_mask_0, x = query_21_cast_fp16)[name = string("op_15596_cast_fp16")]; + tensor var_15600_begin_0 = const()[name = string("op_15600_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15600_end_0 = const()[name = string("op_15600_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_15600_end_mask_0 = const()[name = string("op_15600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15600_cast_fp16 = slice_by_index(begin = var_15600_begin_0, end = var_15600_end_0, end_mask = var_15600_end_mask_0, x = query_21_cast_fp16)[name = string("op_15600_cast_fp16")]; + tensor var_15604_begin_0 = const()[name = string("op_15604_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15604_end_0 = const()[name = string("op_15604_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_15604_end_mask_0 = const()[name = string("op_15604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15604_cast_fp16 = slice_by_index(begin = var_15604_begin_0, end = var_15604_end_0, end_mask = var_15604_end_mask_0, x = query_21_cast_fp16)[name = string("op_15604_cast_fp16")]; + tensor var_15608_begin_0 = const()[name = string("op_15608_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15608_end_0 = const()[name = string("op_15608_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_15608_end_mask_0 = const()[name = string("op_15608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15608_cast_fp16 = slice_by_index(begin = var_15608_begin_0, end = var_15608_end_0, end_mask = var_15608_end_mask_0, x = query_21_cast_fp16)[name = string("op_15608_cast_fp16")]; + tensor var_15617_begin_0 = const()[name = string("op_15617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15617_end_0 = const()[name = string("op_15617_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15617_end_mask_0 = const()[name = string("op_15617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15617_cast_fp16 = slice_by_index(begin = var_15617_begin_0, end = var_15617_end_0, end_mask = var_15617_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15617_cast_fp16")]; + tensor var_15624_begin_0 = const()[name = string("op_15624_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15624_end_0 = const()[name = string("op_15624_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15624_end_mask_0 = const()[name = string("op_15624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15624_cast_fp16 = slice_by_index(begin = var_15624_begin_0, end = var_15624_end_0, end_mask = var_15624_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15624_cast_fp16")]; + tensor var_15631_begin_0 = const()[name = string("op_15631_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15631_end_0 = const()[name = string("op_15631_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15631_end_mask_0 = const()[name = string("op_15631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15631_cast_fp16 = slice_by_index(begin = var_15631_begin_0, end = var_15631_end_0, end_mask = var_15631_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15631_cast_fp16")]; + tensor var_15638_begin_0 = const()[name = string("op_15638_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15638_end_0 = const()[name = string("op_15638_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15638_end_mask_0 = const()[name = string("op_15638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15638_cast_fp16 = slice_by_index(begin = var_15638_begin_0, end = var_15638_end_0, end_mask = var_15638_end_mask_0, x = var_15532_cast_fp16)[name = string("op_15638_cast_fp16")]; + tensor var_15645_begin_0 = const()[name = string("op_15645_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15645_end_0 = const()[name = string("op_15645_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15645_end_mask_0 = const()[name = string("op_15645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15645_cast_fp16 = slice_by_index(begin = var_15645_begin_0, end = var_15645_end_0, end_mask = var_15645_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15645_cast_fp16")]; + tensor var_15652_begin_0 = const()[name = string("op_15652_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15652_end_0 = const()[name = string("op_15652_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15652_end_mask_0 = const()[name = string("op_15652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15652_cast_fp16 = slice_by_index(begin = var_15652_begin_0, end = var_15652_end_0, end_mask = var_15652_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15652_cast_fp16")]; + tensor var_15659_begin_0 = const()[name = string("op_15659_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15659_end_0 = const()[name = string("op_15659_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15659_end_mask_0 = const()[name = string("op_15659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15659_cast_fp16 = slice_by_index(begin = var_15659_begin_0, end = var_15659_end_0, end_mask = var_15659_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15659_cast_fp16")]; + tensor var_15666_begin_0 = const()[name = string("op_15666_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15666_end_0 = const()[name = string("op_15666_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15666_end_mask_0 = const()[name = string("op_15666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15666_cast_fp16 = slice_by_index(begin = var_15666_begin_0, end = var_15666_end_0, end_mask = var_15666_end_mask_0, x = var_15536_cast_fp16)[name = string("op_15666_cast_fp16")]; + tensor var_15673_begin_0 = const()[name = string("op_15673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15673_end_0 = const()[name = string("op_15673_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15673_end_mask_0 = const()[name = string("op_15673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15673_cast_fp16 = slice_by_index(begin = var_15673_begin_0, end = var_15673_end_0, end_mask = var_15673_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15673_cast_fp16")]; + tensor var_15680_begin_0 = const()[name = string("op_15680_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15680_end_0 = const()[name = string("op_15680_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15680_end_mask_0 = const()[name = string("op_15680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15680_cast_fp16 = slice_by_index(begin = var_15680_begin_0, end = var_15680_end_0, end_mask = var_15680_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15680_cast_fp16")]; + tensor var_15687_begin_0 = const()[name = string("op_15687_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15687_end_0 = const()[name = string("op_15687_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15687_end_mask_0 = const()[name = string("op_15687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15687_cast_fp16 = slice_by_index(begin = var_15687_begin_0, end = var_15687_end_0, end_mask = var_15687_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15687_cast_fp16")]; + tensor var_15694_begin_0 = const()[name = string("op_15694_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15694_end_0 = const()[name = string("op_15694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15694_end_mask_0 = const()[name = string("op_15694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15694_cast_fp16 = slice_by_index(begin = var_15694_begin_0, end = var_15694_end_0, end_mask = var_15694_end_mask_0, x = var_15540_cast_fp16)[name = string("op_15694_cast_fp16")]; + tensor var_15701_begin_0 = const()[name = string("op_15701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15701_end_0 = const()[name = string("op_15701_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15701_end_mask_0 = const()[name = string("op_15701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15701_cast_fp16 = slice_by_index(begin = var_15701_begin_0, end = var_15701_end_0, end_mask = var_15701_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15701_cast_fp16")]; + tensor var_15708_begin_0 = const()[name = string("op_15708_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15708_end_0 = const()[name = string("op_15708_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15708_end_mask_0 = const()[name = string("op_15708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15708_cast_fp16 = slice_by_index(begin = var_15708_begin_0, end = var_15708_end_0, end_mask = var_15708_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15708_cast_fp16")]; + tensor var_15715_begin_0 = const()[name = string("op_15715_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15715_end_0 = const()[name = string("op_15715_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15715_end_mask_0 = const()[name = string("op_15715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15715_cast_fp16 = slice_by_index(begin = var_15715_begin_0, end = var_15715_end_0, end_mask = var_15715_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15715_cast_fp16")]; + tensor var_15722_begin_0 = const()[name = string("op_15722_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15722_end_0 = const()[name = string("op_15722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15722_end_mask_0 = const()[name = string("op_15722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15722_cast_fp16 = slice_by_index(begin = var_15722_begin_0, end = var_15722_end_0, end_mask = var_15722_end_mask_0, x = var_15544_cast_fp16)[name = string("op_15722_cast_fp16")]; + tensor var_15729_begin_0 = const()[name = string("op_15729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15729_end_0 = const()[name = string("op_15729_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15729_end_mask_0 = const()[name = string("op_15729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15729_cast_fp16 = slice_by_index(begin = var_15729_begin_0, end = var_15729_end_0, end_mask = var_15729_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15729_cast_fp16")]; + tensor var_15736_begin_0 = const()[name = string("op_15736_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15736_end_0 = const()[name = string("op_15736_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15736_end_mask_0 = const()[name = string("op_15736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15736_cast_fp16 = slice_by_index(begin = var_15736_begin_0, end = var_15736_end_0, end_mask = var_15736_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15736_cast_fp16")]; + tensor var_15743_begin_0 = const()[name = string("op_15743_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15743_end_0 = const()[name = string("op_15743_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15743_end_mask_0 = const()[name = string("op_15743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15743_cast_fp16 = slice_by_index(begin = var_15743_begin_0, end = var_15743_end_0, end_mask = var_15743_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15743_cast_fp16")]; + tensor var_15750_begin_0 = const()[name = string("op_15750_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15750_end_0 = const()[name = string("op_15750_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15750_end_mask_0 = const()[name = string("op_15750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15750_cast_fp16 = slice_by_index(begin = var_15750_begin_0, end = var_15750_end_0, end_mask = var_15750_end_mask_0, x = var_15548_cast_fp16)[name = string("op_15750_cast_fp16")]; + tensor var_15757_begin_0 = const()[name = string("op_15757_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15757_end_0 = const()[name = string("op_15757_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15757_end_mask_0 = const()[name = string("op_15757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15757_cast_fp16 = slice_by_index(begin = var_15757_begin_0, end = var_15757_end_0, end_mask = var_15757_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15757_cast_fp16")]; + tensor var_15764_begin_0 = const()[name = string("op_15764_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15764_end_0 = const()[name = string("op_15764_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15764_end_mask_0 = const()[name = string("op_15764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15764_cast_fp16 = slice_by_index(begin = var_15764_begin_0, end = var_15764_end_0, end_mask = var_15764_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15764_cast_fp16")]; + tensor var_15771_begin_0 = const()[name = string("op_15771_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15771_end_0 = const()[name = string("op_15771_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15771_end_mask_0 = const()[name = string("op_15771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15771_cast_fp16 = slice_by_index(begin = var_15771_begin_0, end = var_15771_end_0, end_mask = var_15771_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15771_cast_fp16")]; + tensor var_15778_begin_0 = const()[name = string("op_15778_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15778_end_0 = const()[name = string("op_15778_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15778_end_mask_0 = const()[name = string("op_15778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15778_cast_fp16 = slice_by_index(begin = var_15778_begin_0, end = var_15778_end_0, end_mask = var_15778_end_mask_0, x = var_15552_cast_fp16)[name = string("op_15778_cast_fp16")]; + tensor var_15785_begin_0 = const()[name = string("op_15785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15785_end_0 = const()[name = string("op_15785_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15785_end_mask_0 = const()[name = string("op_15785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15785_cast_fp16 = slice_by_index(begin = var_15785_begin_0, end = var_15785_end_0, end_mask = var_15785_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15785_cast_fp16")]; + tensor var_15792_begin_0 = const()[name = string("op_15792_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15792_end_0 = const()[name = string("op_15792_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15792_end_mask_0 = const()[name = string("op_15792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15792_cast_fp16 = slice_by_index(begin = var_15792_begin_0, end = var_15792_end_0, end_mask = var_15792_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15792_cast_fp16")]; + tensor var_15799_begin_0 = const()[name = string("op_15799_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15799_end_0 = const()[name = string("op_15799_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15799_end_mask_0 = const()[name = string("op_15799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15799_cast_fp16 = slice_by_index(begin = var_15799_begin_0, end = var_15799_end_0, end_mask = var_15799_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15799_cast_fp16")]; + tensor var_15806_begin_0 = const()[name = string("op_15806_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15806_end_0 = const()[name = string("op_15806_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15806_end_mask_0 = const()[name = string("op_15806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15806_cast_fp16 = slice_by_index(begin = var_15806_begin_0, end = var_15806_end_0, end_mask = var_15806_end_mask_0, x = var_15556_cast_fp16)[name = string("op_15806_cast_fp16")]; + tensor var_15813_begin_0 = const()[name = string("op_15813_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15813_end_0 = const()[name = string("op_15813_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15813_end_mask_0 = const()[name = string("op_15813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15813_cast_fp16 = slice_by_index(begin = var_15813_begin_0, end = var_15813_end_0, end_mask = var_15813_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15813_cast_fp16")]; + tensor var_15820_begin_0 = const()[name = string("op_15820_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15820_end_0 = const()[name = string("op_15820_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15820_end_mask_0 = const()[name = string("op_15820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15820_cast_fp16 = slice_by_index(begin = var_15820_begin_0, end = var_15820_end_0, end_mask = var_15820_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15820_cast_fp16")]; + tensor var_15827_begin_0 = const()[name = string("op_15827_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15827_end_0 = const()[name = string("op_15827_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15827_end_mask_0 = const()[name = string("op_15827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15827_cast_fp16 = slice_by_index(begin = var_15827_begin_0, end = var_15827_end_0, end_mask = var_15827_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15827_cast_fp16")]; + tensor var_15834_begin_0 = const()[name = string("op_15834_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15834_end_0 = const()[name = string("op_15834_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15834_end_mask_0 = const()[name = string("op_15834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15834_cast_fp16 = slice_by_index(begin = var_15834_begin_0, end = var_15834_end_0, end_mask = var_15834_end_mask_0, x = var_15560_cast_fp16)[name = string("op_15834_cast_fp16")]; + tensor var_15841_begin_0 = const()[name = string("op_15841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15841_end_0 = const()[name = string("op_15841_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15841_end_mask_0 = const()[name = string("op_15841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15841_cast_fp16 = slice_by_index(begin = var_15841_begin_0, end = var_15841_end_0, end_mask = var_15841_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15841_cast_fp16")]; + tensor var_15848_begin_0 = const()[name = string("op_15848_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15848_end_0 = const()[name = string("op_15848_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15848_end_mask_0 = const()[name = string("op_15848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15848_cast_fp16 = slice_by_index(begin = var_15848_begin_0, end = var_15848_end_0, end_mask = var_15848_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15848_cast_fp16")]; + tensor var_15855_begin_0 = const()[name = string("op_15855_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15855_end_0 = const()[name = string("op_15855_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15855_end_mask_0 = const()[name = string("op_15855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15855_cast_fp16 = slice_by_index(begin = var_15855_begin_0, end = var_15855_end_0, end_mask = var_15855_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15855_cast_fp16")]; + tensor var_15862_begin_0 = const()[name = string("op_15862_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15862_end_0 = const()[name = string("op_15862_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15862_end_mask_0 = const()[name = string("op_15862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15862_cast_fp16 = slice_by_index(begin = var_15862_begin_0, end = var_15862_end_0, end_mask = var_15862_end_mask_0, x = var_15564_cast_fp16)[name = string("op_15862_cast_fp16")]; + tensor var_15869_begin_0 = const()[name = string("op_15869_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15869_end_0 = const()[name = string("op_15869_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15869_end_mask_0 = const()[name = string("op_15869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15869_cast_fp16 = slice_by_index(begin = var_15869_begin_0, end = var_15869_end_0, end_mask = var_15869_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15869_cast_fp16")]; + tensor var_15876_begin_0 = const()[name = string("op_15876_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15876_end_0 = const()[name = string("op_15876_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15876_end_mask_0 = const()[name = string("op_15876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15876_cast_fp16 = slice_by_index(begin = var_15876_begin_0, end = var_15876_end_0, end_mask = var_15876_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15876_cast_fp16")]; + tensor var_15883_begin_0 = const()[name = string("op_15883_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15883_end_0 = const()[name = string("op_15883_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15883_end_mask_0 = const()[name = string("op_15883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15883_cast_fp16 = slice_by_index(begin = var_15883_begin_0, end = var_15883_end_0, end_mask = var_15883_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15883_cast_fp16")]; + tensor var_15890_begin_0 = const()[name = string("op_15890_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15890_end_0 = const()[name = string("op_15890_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15890_end_mask_0 = const()[name = string("op_15890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15890_cast_fp16 = slice_by_index(begin = var_15890_begin_0, end = var_15890_end_0, end_mask = var_15890_end_mask_0, x = var_15568_cast_fp16)[name = string("op_15890_cast_fp16")]; + tensor var_15897_begin_0 = const()[name = string("op_15897_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15897_end_0 = const()[name = string("op_15897_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15897_end_mask_0 = const()[name = string("op_15897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15897_cast_fp16 = slice_by_index(begin = var_15897_begin_0, end = var_15897_end_0, end_mask = var_15897_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15897_cast_fp16")]; + tensor var_15904_begin_0 = const()[name = string("op_15904_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15904_end_0 = const()[name = string("op_15904_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15904_end_mask_0 = const()[name = string("op_15904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15904_cast_fp16 = slice_by_index(begin = var_15904_begin_0, end = var_15904_end_0, end_mask = var_15904_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15904_cast_fp16")]; + tensor var_15911_begin_0 = const()[name = string("op_15911_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15911_end_0 = const()[name = string("op_15911_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15911_end_mask_0 = const()[name = string("op_15911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15911_cast_fp16 = slice_by_index(begin = var_15911_begin_0, end = var_15911_end_0, end_mask = var_15911_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15911_cast_fp16")]; + tensor var_15918_begin_0 = const()[name = string("op_15918_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15918_end_0 = const()[name = string("op_15918_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15918_end_mask_0 = const()[name = string("op_15918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15918_cast_fp16 = slice_by_index(begin = var_15918_begin_0, end = var_15918_end_0, end_mask = var_15918_end_mask_0, x = var_15572_cast_fp16)[name = string("op_15918_cast_fp16")]; + tensor var_15925_begin_0 = const()[name = string("op_15925_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15925_end_0 = const()[name = string("op_15925_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15925_end_mask_0 = const()[name = string("op_15925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15925_cast_fp16 = slice_by_index(begin = var_15925_begin_0, end = var_15925_end_0, end_mask = var_15925_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15925_cast_fp16")]; + tensor var_15932_begin_0 = const()[name = string("op_15932_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15932_end_0 = const()[name = string("op_15932_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15932_end_mask_0 = const()[name = string("op_15932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15932_cast_fp16 = slice_by_index(begin = var_15932_begin_0, end = var_15932_end_0, end_mask = var_15932_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15932_cast_fp16")]; + tensor var_15939_begin_0 = const()[name = string("op_15939_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15939_end_0 = const()[name = string("op_15939_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15939_end_mask_0 = const()[name = string("op_15939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15939_cast_fp16 = slice_by_index(begin = var_15939_begin_0, end = var_15939_end_0, end_mask = var_15939_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15939_cast_fp16")]; + tensor var_15946_begin_0 = const()[name = string("op_15946_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15946_end_0 = const()[name = string("op_15946_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15946_end_mask_0 = const()[name = string("op_15946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15946_cast_fp16 = slice_by_index(begin = var_15946_begin_0, end = var_15946_end_0, end_mask = var_15946_end_mask_0, x = var_15576_cast_fp16)[name = string("op_15946_cast_fp16")]; + tensor var_15953_begin_0 = const()[name = string("op_15953_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15953_end_0 = const()[name = string("op_15953_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15953_end_mask_0 = const()[name = string("op_15953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15953_cast_fp16 = slice_by_index(begin = var_15953_begin_0, end = var_15953_end_0, end_mask = var_15953_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15953_cast_fp16")]; + tensor var_15960_begin_0 = const()[name = string("op_15960_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15960_end_0 = const()[name = string("op_15960_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15960_end_mask_0 = const()[name = string("op_15960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15960_cast_fp16 = slice_by_index(begin = var_15960_begin_0, end = var_15960_end_0, end_mask = var_15960_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15960_cast_fp16")]; + tensor var_15967_begin_0 = const()[name = string("op_15967_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15967_end_0 = const()[name = string("op_15967_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15967_end_mask_0 = const()[name = string("op_15967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15967_cast_fp16 = slice_by_index(begin = var_15967_begin_0, end = var_15967_end_0, end_mask = var_15967_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15967_cast_fp16")]; + tensor var_15974_begin_0 = const()[name = string("op_15974_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15974_end_0 = const()[name = string("op_15974_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15974_end_mask_0 = const()[name = string("op_15974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15974_cast_fp16 = slice_by_index(begin = var_15974_begin_0, end = var_15974_end_0, end_mask = var_15974_end_mask_0, x = var_15580_cast_fp16)[name = string("op_15974_cast_fp16")]; + tensor var_15981_begin_0 = const()[name = string("op_15981_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15981_end_0 = const()[name = string("op_15981_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15981_end_mask_0 = const()[name = string("op_15981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15981_cast_fp16 = slice_by_index(begin = var_15981_begin_0, end = var_15981_end_0, end_mask = var_15981_end_mask_0, x = var_15584_cast_fp16)[name = string("op_15981_cast_fp16")]; + tensor var_15988_begin_0 = const()[name = string("op_15988_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15988_end_0 = const()[name = string("op_15988_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15988_end_mask_0 = const()[name = string("op_15988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15988_cast_fp16 = slice_by_index(begin = var_15988_begin_0, end = var_15988_end_0, end_mask = var_15988_end_mask_0, x = var_15584_cast_fp16)[name = string("op_15988_cast_fp16")]; + tensor var_15995_begin_0 = const()[name = string("op_15995_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15995_end_0 = const()[name = string("op_15995_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15995_end_mask_0 = const()[name = string("op_15995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15995_cast_fp16 = slice_by_index(begin = var_15995_begin_0, end = var_15995_end_0, end_mask = var_15995_end_mask_0, x = var_15584_cast_fp16)[name = string("op_15995_cast_fp16")]; + tensor var_16002_begin_0 = const()[name = string("op_16002_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16002_end_0 = const()[name = string("op_16002_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16002_end_mask_0 = const()[name = string("op_16002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16002_cast_fp16 = slice_by_index(begin = var_16002_begin_0, end = var_16002_end_0, end_mask = var_16002_end_mask_0, x = var_15584_cast_fp16)[name = string("op_16002_cast_fp16")]; + tensor var_16009_begin_0 = const()[name = string("op_16009_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16009_end_0 = const()[name = string("op_16009_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16009_end_mask_0 = const()[name = string("op_16009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16009_cast_fp16 = slice_by_index(begin = var_16009_begin_0, end = var_16009_end_0, end_mask = var_16009_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16009_cast_fp16")]; + tensor var_16016_begin_0 = const()[name = string("op_16016_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16016_end_0 = const()[name = string("op_16016_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16016_end_mask_0 = const()[name = string("op_16016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16016_cast_fp16 = slice_by_index(begin = var_16016_begin_0, end = var_16016_end_0, end_mask = var_16016_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16016_cast_fp16")]; + tensor var_16023_begin_0 = const()[name = string("op_16023_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16023_end_0 = const()[name = string("op_16023_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16023_end_mask_0 = const()[name = string("op_16023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16023_cast_fp16 = slice_by_index(begin = var_16023_begin_0, end = var_16023_end_0, end_mask = var_16023_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16023_cast_fp16")]; + tensor var_16030_begin_0 = const()[name = string("op_16030_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16030_end_0 = const()[name = string("op_16030_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16030_end_mask_0 = const()[name = string("op_16030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16030_cast_fp16 = slice_by_index(begin = var_16030_begin_0, end = var_16030_end_0, end_mask = var_16030_end_mask_0, x = var_15588_cast_fp16)[name = string("op_16030_cast_fp16")]; + tensor var_16037_begin_0 = const()[name = string("op_16037_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16037_end_0 = const()[name = string("op_16037_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16037_end_mask_0 = const()[name = string("op_16037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16037_cast_fp16 = slice_by_index(begin = var_16037_begin_0, end = var_16037_end_0, end_mask = var_16037_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16037_cast_fp16")]; + tensor var_16044_begin_0 = const()[name = string("op_16044_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16044_end_0 = const()[name = string("op_16044_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16044_end_mask_0 = const()[name = string("op_16044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16044_cast_fp16 = slice_by_index(begin = var_16044_begin_0, end = var_16044_end_0, end_mask = var_16044_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16044_cast_fp16")]; + tensor var_16051_begin_0 = const()[name = string("op_16051_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16051_end_0 = const()[name = string("op_16051_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16051_end_mask_0 = const()[name = string("op_16051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16051_cast_fp16 = slice_by_index(begin = var_16051_begin_0, end = var_16051_end_0, end_mask = var_16051_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16051_cast_fp16")]; + tensor var_16058_begin_0 = const()[name = string("op_16058_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16058_end_0 = const()[name = string("op_16058_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16058_end_mask_0 = const()[name = string("op_16058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16058_cast_fp16 = slice_by_index(begin = var_16058_begin_0, end = var_16058_end_0, end_mask = var_16058_end_mask_0, x = var_15592_cast_fp16)[name = string("op_16058_cast_fp16")]; + tensor var_16065_begin_0 = const()[name = string("op_16065_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16065_end_0 = const()[name = string("op_16065_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16065_end_mask_0 = const()[name = string("op_16065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16065_cast_fp16 = slice_by_index(begin = var_16065_begin_0, end = var_16065_end_0, end_mask = var_16065_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16065_cast_fp16")]; + tensor var_16072_begin_0 = const()[name = string("op_16072_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16072_end_0 = const()[name = string("op_16072_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16072_end_mask_0 = const()[name = string("op_16072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16072_cast_fp16 = slice_by_index(begin = var_16072_begin_0, end = var_16072_end_0, end_mask = var_16072_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16072_cast_fp16")]; + tensor var_16079_begin_0 = const()[name = string("op_16079_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16079_end_0 = const()[name = string("op_16079_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16079_end_mask_0 = const()[name = string("op_16079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16079_cast_fp16 = slice_by_index(begin = var_16079_begin_0, end = var_16079_end_0, end_mask = var_16079_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16079_cast_fp16")]; + tensor var_16086_begin_0 = const()[name = string("op_16086_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16086_end_0 = const()[name = string("op_16086_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16086_end_mask_0 = const()[name = string("op_16086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16086_cast_fp16 = slice_by_index(begin = var_16086_begin_0, end = var_16086_end_0, end_mask = var_16086_end_mask_0, x = var_15596_cast_fp16)[name = string("op_16086_cast_fp16")]; + tensor var_16093_begin_0 = const()[name = string("op_16093_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16093_end_0 = const()[name = string("op_16093_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16093_end_mask_0 = const()[name = string("op_16093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16093_cast_fp16 = slice_by_index(begin = var_16093_begin_0, end = var_16093_end_0, end_mask = var_16093_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16093_cast_fp16")]; + tensor var_16100_begin_0 = const()[name = string("op_16100_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16100_end_0 = const()[name = string("op_16100_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16100_end_mask_0 = const()[name = string("op_16100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16100_cast_fp16 = slice_by_index(begin = var_16100_begin_0, end = var_16100_end_0, end_mask = var_16100_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16100_cast_fp16")]; + tensor var_16107_begin_0 = const()[name = string("op_16107_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16107_end_0 = const()[name = string("op_16107_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16107_end_mask_0 = const()[name = string("op_16107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16107_cast_fp16 = slice_by_index(begin = var_16107_begin_0, end = var_16107_end_0, end_mask = var_16107_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16107_cast_fp16")]; + tensor var_16114_begin_0 = const()[name = string("op_16114_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16114_end_0 = const()[name = string("op_16114_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16114_end_mask_0 = const()[name = string("op_16114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16114_cast_fp16 = slice_by_index(begin = var_16114_begin_0, end = var_16114_end_0, end_mask = var_16114_end_mask_0, x = var_15600_cast_fp16)[name = string("op_16114_cast_fp16")]; + tensor var_16121_begin_0 = const()[name = string("op_16121_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16121_end_0 = const()[name = string("op_16121_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16121_end_mask_0 = const()[name = string("op_16121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16121_cast_fp16 = slice_by_index(begin = var_16121_begin_0, end = var_16121_end_0, end_mask = var_16121_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16121_cast_fp16")]; + tensor var_16128_begin_0 = const()[name = string("op_16128_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16128_end_0 = const()[name = string("op_16128_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16128_end_mask_0 = const()[name = string("op_16128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16128_cast_fp16 = slice_by_index(begin = var_16128_begin_0, end = var_16128_end_0, end_mask = var_16128_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16128_cast_fp16")]; + tensor var_16135_begin_0 = const()[name = string("op_16135_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16135_end_0 = const()[name = string("op_16135_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16135_end_mask_0 = const()[name = string("op_16135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16135_cast_fp16 = slice_by_index(begin = var_16135_begin_0, end = var_16135_end_0, end_mask = var_16135_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16135_cast_fp16")]; + tensor var_16142_begin_0 = const()[name = string("op_16142_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16142_end_0 = const()[name = string("op_16142_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16142_end_mask_0 = const()[name = string("op_16142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16142_cast_fp16 = slice_by_index(begin = var_16142_begin_0, end = var_16142_end_0, end_mask = var_16142_end_mask_0, x = var_15604_cast_fp16)[name = string("op_16142_cast_fp16")]; + tensor var_16149_begin_0 = const()[name = string("op_16149_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16149_end_0 = const()[name = string("op_16149_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16149_end_mask_0 = const()[name = string("op_16149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16149_cast_fp16 = slice_by_index(begin = var_16149_begin_0, end = var_16149_end_0, end_mask = var_16149_end_mask_0, x = var_15608_cast_fp16)[name = string("op_16149_cast_fp16")]; + tensor var_16156_begin_0 = const()[name = string("op_16156_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16156_end_0 = const()[name = string("op_16156_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16156_end_mask_0 = const()[name = string("op_16156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16156_cast_fp16 = slice_by_index(begin = var_16156_begin_0, end = var_16156_end_0, end_mask = var_16156_end_mask_0, x = var_15608_cast_fp16)[name = string("op_16156_cast_fp16")]; + tensor var_16163_begin_0 = const()[name = string("op_16163_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16163_end_0 = const()[name = string("op_16163_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16163_end_mask_0 = const()[name = string("op_16163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16163_cast_fp16 = slice_by_index(begin = var_16163_begin_0, end = var_16163_end_0, end_mask = var_16163_end_mask_0, x = var_15608_cast_fp16)[name = string("op_16163_cast_fp16")]; + tensor var_16170_begin_0 = const()[name = string("op_16170_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16170_end_0 = const()[name = string("op_16170_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16170_end_mask_0 = const()[name = string("op_16170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16170_cast_fp16 = slice_by_index(begin = var_16170_begin_0, end = var_16170_end_0, end_mask = var_16170_end_mask_0, x = var_15608_cast_fp16)[name = string("op_16170_cast_fp16")]; + tensor k_21_perm_0 = const()[name = string("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_16175_begin_0 = const()[name = string("op_16175_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16175_end_0 = const()[name = string("op_16175_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_16175_end_mask_0 = const()[name = string("op_16175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = string("transpose_21")]; + tensor var_16175_cast_fp16 = slice_by_index(begin = var_16175_begin_0, end = var_16175_end_0, end_mask = var_16175_end_mask_0, x = k_21_cast_fp16)[name = string("op_16175_cast_fp16")]; + tensor var_16179_begin_0 = const()[name = string("op_16179_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_16179_end_0 = const()[name = string("op_16179_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_16179_end_mask_0 = const()[name = string("op_16179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16179_cast_fp16 = slice_by_index(begin = var_16179_begin_0, end = var_16179_end_0, end_mask = var_16179_end_mask_0, x = k_21_cast_fp16)[name = string("op_16179_cast_fp16")]; + tensor var_16183_begin_0 = const()[name = string("op_16183_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_16183_end_0 = const()[name = string("op_16183_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_16183_end_mask_0 = const()[name = string("op_16183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16183_cast_fp16 = slice_by_index(begin = var_16183_begin_0, end = var_16183_end_0, end_mask = var_16183_end_mask_0, x = k_21_cast_fp16)[name = string("op_16183_cast_fp16")]; + tensor var_16187_begin_0 = const()[name = string("op_16187_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_16187_end_0 = const()[name = string("op_16187_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_16187_end_mask_0 = const()[name = string("op_16187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16187_cast_fp16 = slice_by_index(begin = var_16187_begin_0, end = var_16187_end_0, end_mask = var_16187_end_mask_0, x = k_21_cast_fp16)[name = string("op_16187_cast_fp16")]; + tensor var_16191_begin_0 = const()[name = string("op_16191_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_16191_end_0 = const()[name = string("op_16191_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_16191_end_mask_0 = const()[name = string("op_16191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16191_cast_fp16 = slice_by_index(begin = var_16191_begin_0, end = var_16191_end_0, end_mask = var_16191_end_mask_0, x = k_21_cast_fp16)[name = string("op_16191_cast_fp16")]; + tensor var_16195_begin_0 = const()[name = string("op_16195_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_16195_end_0 = const()[name = string("op_16195_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_16195_end_mask_0 = const()[name = string("op_16195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16195_cast_fp16 = slice_by_index(begin = var_16195_begin_0, end = var_16195_end_0, end_mask = var_16195_end_mask_0, x = k_21_cast_fp16)[name = string("op_16195_cast_fp16")]; + tensor var_16199_begin_0 = const()[name = string("op_16199_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_16199_end_0 = const()[name = string("op_16199_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_16199_end_mask_0 = const()[name = string("op_16199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16199_cast_fp16 = slice_by_index(begin = var_16199_begin_0, end = var_16199_end_0, end_mask = var_16199_end_mask_0, x = k_21_cast_fp16)[name = string("op_16199_cast_fp16")]; + tensor var_16203_begin_0 = const()[name = string("op_16203_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_16203_end_0 = const()[name = string("op_16203_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_16203_end_mask_0 = const()[name = string("op_16203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16203_cast_fp16 = slice_by_index(begin = var_16203_begin_0, end = var_16203_end_0, end_mask = var_16203_end_mask_0, x = k_21_cast_fp16)[name = string("op_16203_cast_fp16")]; + tensor var_16207_begin_0 = const()[name = string("op_16207_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_16207_end_0 = const()[name = string("op_16207_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_16207_end_mask_0 = const()[name = string("op_16207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16207_cast_fp16 = slice_by_index(begin = var_16207_begin_0, end = var_16207_end_0, end_mask = var_16207_end_mask_0, x = k_21_cast_fp16)[name = string("op_16207_cast_fp16")]; + tensor var_16211_begin_0 = const()[name = string("op_16211_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_16211_end_0 = const()[name = string("op_16211_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_16211_end_mask_0 = const()[name = string("op_16211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16211_cast_fp16 = slice_by_index(begin = var_16211_begin_0, end = var_16211_end_0, end_mask = var_16211_end_mask_0, x = k_21_cast_fp16)[name = string("op_16211_cast_fp16")]; + tensor var_16215_begin_0 = const()[name = string("op_16215_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_16215_end_0 = const()[name = string("op_16215_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_16215_end_mask_0 = const()[name = string("op_16215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16215_cast_fp16 = slice_by_index(begin = var_16215_begin_0, end = var_16215_end_0, end_mask = var_16215_end_mask_0, x = k_21_cast_fp16)[name = string("op_16215_cast_fp16")]; + tensor var_16219_begin_0 = const()[name = string("op_16219_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_16219_end_0 = const()[name = string("op_16219_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_16219_end_mask_0 = const()[name = string("op_16219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16219_cast_fp16 = slice_by_index(begin = var_16219_begin_0, end = var_16219_end_0, end_mask = var_16219_end_mask_0, x = k_21_cast_fp16)[name = string("op_16219_cast_fp16")]; + tensor var_16223_begin_0 = const()[name = string("op_16223_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_16223_end_0 = const()[name = string("op_16223_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_16223_end_mask_0 = const()[name = string("op_16223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16223_cast_fp16 = slice_by_index(begin = var_16223_begin_0, end = var_16223_end_0, end_mask = var_16223_end_mask_0, x = k_21_cast_fp16)[name = string("op_16223_cast_fp16")]; + tensor var_16227_begin_0 = const()[name = string("op_16227_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_16227_end_0 = const()[name = string("op_16227_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_16227_end_mask_0 = const()[name = string("op_16227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16227_cast_fp16 = slice_by_index(begin = var_16227_begin_0, end = var_16227_end_0, end_mask = var_16227_end_mask_0, x = k_21_cast_fp16)[name = string("op_16227_cast_fp16")]; + tensor var_16231_begin_0 = const()[name = string("op_16231_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_16231_end_0 = const()[name = string("op_16231_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_16231_end_mask_0 = const()[name = string("op_16231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16231_cast_fp16 = slice_by_index(begin = var_16231_begin_0, end = var_16231_end_0, end_mask = var_16231_end_mask_0, x = k_21_cast_fp16)[name = string("op_16231_cast_fp16")]; + tensor var_16235_begin_0 = const()[name = string("op_16235_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_16235_end_0 = const()[name = string("op_16235_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_16235_end_mask_0 = const()[name = string("op_16235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16235_cast_fp16 = slice_by_index(begin = var_16235_begin_0, end = var_16235_end_0, end_mask = var_16235_end_mask_0, x = k_21_cast_fp16)[name = string("op_16235_cast_fp16")]; + tensor var_16239_begin_0 = const()[name = string("op_16239_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_16239_end_0 = const()[name = string("op_16239_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_16239_end_mask_0 = const()[name = string("op_16239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16239_cast_fp16 = slice_by_index(begin = var_16239_begin_0, end = var_16239_end_0, end_mask = var_16239_end_mask_0, x = k_21_cast_fp16)[name = string("op_16239_cast_fp16")]; + tensor var_16243_begin_0 = const()[name = string("op_16243_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_16243_end_0 = const()[name = string("op_16243_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_16243_end_mask_0 = const()[name = string("op_16243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16243_cast_fp16 = slice_by_index(begin = var_16243_begin_0, end = var_16243_end_0, end_mask = var_16243_end_mask_0, x = k_21_cast_fp16)[name = string("op_16243_cast_fp16")]; + tensor var_16247_begin_0 = const()[name = string("op_16247_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_16247_end_0 = const()[name = string("op_16247_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_16247_end_mask_0 = const()[name = string("op_16247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16247_cast_fp16 = slice_by_index(begin = var_16247_begin_0, end = var_16247_end_0, end_mask = var_16247_end_mask_0, x = k_21_cast_fp16)[name = string("op_16247_cast_fp16")]; + tensor var_16251_begin_0 = const()[name = string("op_16251_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_16251_end_0 = const()[name = string("op_16251_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_16251_end_mask_0 = const()[name = string("op_16251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16251_cast_fp16 = slice_by_index(begin = var_16251_begin_0, end = var_16251_end_0, end_mask = var_16251_end_mask_0, x = k_21_cast_fp16)[name = string("op_16251_cast_fp16")]; + tensor var_16253_begin_0 = const()[name = string("op_16253_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16253_end_0 = const()[name = string("op_16253_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16253_end_mask_0 = const()[name = string("op_16253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16253_cast_fp16 = slice_by_index(begin = var_16253_begin_0, end = var_16253_end_0, end_mask = var_16253_end_mask_0, x = value_21_cast_fp16)[name = string("op_16253_cast_fp16")]; + tensor var_16257_begin_0 = const()[name = string("op_16257_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16257_end_0 = const()[name = string("op_16257_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_16257_end_mask_0 = const()[name = string("op_16257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16257_cast_fp16 = slice_by_index(begin = var_16257_begin_0, end = var_16257_end_0, end_mask = var_16257_end_mask_0, x = value_21_cast_fp16)[name = string("op_16257_cast_fp16")]; + tensor var_16261_begin_0 = const()[name = string("op_16261_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16261_end_0 = const()[name = string("op_16261_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_16261_end_mask_0 = const()[name = string("op_16261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16261_cast_fp16 = slice_by_index(begin = var_16261_begin_0, end = var_16261_end_0, end_mask = var_16261_end_mask_0, x = value_21_cast_fp16)[name = string("op_16261_cast_fp16")]; + tensor var_16265_begin_0 = const()[name = string("op_16265_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16265_end_0 = const()[name = string("op_16265_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_16265_end_mask_0 = const()[name = string("op_16265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16265_cast_fp16 = slice_by_index(begin = var_16265_begin_0, end = var_16265_end_0, end_mask = var_16265_end_mask_0, x = value_21_cast_fp16)[name = string("op_16265_cast_fp16")]; + tensor var_16269_begin_0 = const()[name = string("op_16269_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16269_end_0 = const()[name = string("op_16269_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_16269_end_mask_0 = const()[name = string("op_16269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16269_cast_fp16 = slice_by_index(begin = var_16269_begin_0, end = var_16269_end_0, end_mask = var_16269_end_mask_0, x = value_21_cast_fp16)[name = string("op_16269_cast_fp16")]; + tensor var_16273_begin_0 = const()[name = string("op_16273_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16273_end_0 = const()[name = string("op_16273_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_16273_end_mask_0 = const()[name = string("op_16273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16273_cast_fp16 = slice_by_index(begin = var_16273_begin_0, end = var_16273_end_0, end_mask = var_16273_end_mask_0, x = value_21_cast_fp16)[name = string("op_16273_cast_fp16")]; + tensor var_16277_begin_0 = const()[name = string("op_16277_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16277_end_0 = const()[name = string("op_16277_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_16277_end_mask_0 = const()[name = string("op_16277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16277_cast_fp16 = slice_by_index(begin = var_16277_begin_0, end = var_16277_end_0, end_mask = var_16277_end_mask_0, x = value_21_cast_fp16)[name = string("op_16277_cast_fp16")]; + tensor var_16281_begin_0 = const()[name = string("op_16281_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16281_end_0 = const()[name = string("op_16281_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_16281_end_mask_0 = const()[name = string("op_16281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16281_cast_fp16 = slice_by_index(begin = var_16281_begin_0, end = var_16281_end_0, end_mask = var_16281_end_mask_0, x = value_21_cast_fp16)[name = string("op_16281_cast_fp16")]; + tensor var_16285_begin_0 = const()[name = string("op_16285_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16285_end_0 = const()[name = string("op_16285_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_16285_end_mask_0 = const()[name = string("op_16285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16285_cast_fp16 = slice_by_index(begin = var_16285_begin_0, end = var_16285_end_0, end_mask = var_16285_end_mask_0, x = value_21_cast_fp16)[name = string("op_16285_cast_fp16")]; + tensor var_16289_begin_0 = const()[name = string("op_16289_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16289_end_0 = const()[name = string("op_16289_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_16289_end_mask_0 = const()[name = string("op_16289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16289_cast_fp16 = slice_by_index(begin = var_16289_begin_0, end = var_16289_end_0, end_mask = var_16289_end_mask_0, x = value_21_cast_fp16)[name = string("op_16289_cast_fp16")]; + tensor var_16293_begin_0 = const()[name = string("op_16293_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16293_end_0 = const()[name = string("op_16293_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_16293_end_mask_0 = const()[name = string("op_16293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16293_cast_fp16 = slice_by_index(begin = var_16293_begin_0, end = var_16293_end_0, end_mask = var_16293_end_mask_0, x = value_21_cast_fp16)[name = string("op_16293_cast_fp16")]; + tensor var_16297_begin_0 = const()[name = string("op_16297_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16297_end_0 = const()[name = string("op_16297_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_16297_end_mask_0 = const()[name = string("op_16297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16297_cast_fp16 = slice_by_index(begin = var_16297_begin_0, end = var_16297_end_0, end_mask = var_16297_end_mask_0, x = value_21_cast_fp16)[name = string("op_16297_cast_fp16")]; + tensor var_16301_begin_0 = const()[name = string("op_16301_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16301_end_0 = const()[name = string("op_16301_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_16301_end_mask_0 = const()[name = string("op_16301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16301_cast_fp16 = slice_by_index(begin = var_16301_begin_0, end = var_16301_end_0, end_mask = var_16301_end_mask_0, x = value_21_cast_fp16)[name = string("op_16301_cast_fp16")]; + tensor var_16305_begin_0 = const()[name = string("op_16305_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16305_end_0 = const()[name = string("op_16305_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_16305_end_mask_0 = const()[name = string("op_16305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16305_cast_fp16 = slice_by_index(begin = var_16305_begin_0, end = var_16305_end_0, end_mask = var_16305_end_mask_0, x = value_21_cast_fp16)[name = string("op_16305_cast_fp16")]; + tensor var_16309_begin_0 = const()[name = string("op_16309_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16309_end_0 = const()[name = string("op_16309_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_16309_end_mask_0 = const()[name = string("op_16309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16309_cast_fp16 = slice_by_index(begin = var_16309_begin_0, end = var_16309_end_0, end_mask = var_16309_end_mask_0, x = value_21_cast_fp16)[name = string("op_16309_cast_fp16")]; + tensor var_16313_begin_0 = const()[name = string("op_16313_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16313_end_0 = const()[name = string("op_16313_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_16313_end_mask_0 = const()[name = string("op_16313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16313_cast_fp16 = slice_by_index(begin = var_16313_begin_0, end = var_16313_end_0, end_mask = var_16313_end_mask_0, x = value_21_cast_fp16)[name = string("op_16313_cast_fp16")]; + tensor var_16317_begin_0 = const()[name = string("op_16317_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16317_end_0 = const()[name = string("op_16317_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_16317_end_mask_0 = const()[name = string("op_16317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16317_cast_fp16 = slice_by_index(begin = var_16317_begin_0, end = var_16317_end_0, end_mask = var_16317_end_mask_0, x = value_21_cast_fp16)[name = string("op_16317_cast_fp16")]; + tensor var_16321_begin_0 = const()[name = string("op_16321_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16321_end_0 = const()[name = string("op_16321_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_16321_end_mask_0 = const()[name = string("op_16321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16321_cast_fp16 = slice_by_index(begin = var_16321_begin_0, end = var_16321_end_0, end_mask = var_16321_end_mask_0, x = value_21_cast_fp16)[name = string("op_16321_cast_fp16")]; + tensor var_16325_begin_0 = const()[name = string("op_16325_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16325_end_0 = const()[name = string("op_16325_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_16325_end_mask_0 = const()[name = string("op_16325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16325_cast_fp16 = slice_by_index(begin = var_16325_begin_0, end = var_16325_end_0, end_mask = var_16325_end_mask_0, x = value_21_cast_fp16)[name = string("op_16325_cast_fp16")]; + tensor var_16329_begin_0 = const()[name = string("op_16329_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16329_end_0 = const()[name = string("op_16329_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_16329_end_mask_0 = const()[name = string("op_16329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16329_cast_fp16 = slice_by_index(begin = var_16329_begin_0, end = var_16329_end_0, end_mask = var_16329_end_mask_0, x = value_21_cast_fp16)[name = string("op_16329_cast_fp16")]; + string _SplitHeadsQ__mh_w_1601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1601_equation_0, values = (var_16175_cast_fp16, var_15617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1601_cast_fp16")]; + string _SplitHeadsQ__mh_w_1603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1603_equation_0, values = (var_16175_cast_fp16, var_15624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1603_cast_fp16")]; + string _SplitHeadsQ__mh_w_1605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1605_equation_0, values = (var_16175_cast_fp16, var_15631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1605_cast_fp16")]; + string _SplitHeadsQ__mh_w_1607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1607_equation_0, values = (var_16175_cast_fp16, var_15638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1607_cast_fp16")]; + string _SplitHeadsQ__mh_w_1609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1609_equation_0, values = (var_16179_cast_fp16, var_15645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1609_cast_fp16")]; + string _SplitHeadsQ__mh_w_1611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1611_equation_0, values = (var_16179_cast_fp16, var_15652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1611_cast_fp16")]; + string _SplitHeadsQ__mh_w_1613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1613_equation_0, values = (var_16179_cast_fp16, var_15659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1613_cast_fp16")]; + string _SplitHeadsQ__mh_w_1615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1615_equation_0, values = (var_16179_cast_fp16, var_15666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1615_cast_fp16")]; + string _SplitHeadsQ__mh_w_1617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1617_equation_0, values = (var_16183_cast_fp16, var_15673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1617_cast_fp16")]; + string _SplitHeadsQ__mh_w_1619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1619_equation_0, values = (var_16183_cast_fp16, var_15680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1619_cast_fp16")]; + string _SplitHeadsQ__mh_w_1621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1621_equation_0, values = (var_16183_cast_fp16, var_15687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1621_cast_fp16")]; + string _SplitHeadsQ__mh_w_1623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1623_equation_0, values = (var_16183_cast_fp16, var_15694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1623_cast_fp16")]; + string _SplitHeadsQ__mh_w_1625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1625_equation_0, values = (var_16187_cast_fp16, var_15701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1625_cast_fp16")]; + string _SplitHeadsQ__mh_w_1627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1627_equation_0, values = (var_16187_cast_fp16, var_15708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1627_cast_fp16")]; + string _SplitHeadsQ__mh_w_1629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1629_equation_0, values = (var_16187_cast_fp16, var_15715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1629_cast_fp16")]; + string _SplitHeadsQ__mh_w_1631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1631_equation_0, values = (var_16187_cast_fp16, var_15722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1631_cast_fp16")]; + string _SplitHeadsQ__mh_w_1633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1633_equation_0, values = (var_16191_cast_fp16, var_15729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1633_cast_fp16")]; + string _SplitHeadsQ__mh_w_1635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1635_equation_0, values = (var_16191_cast_fp16, var_15736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1635_cast_fp16")]; + string _SplitHeadsQ__mh_w_1637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1637_equation_0, values = (var_16191_cast_fp16, var_15743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1637_cast_fp16")]; + string _SplitHeadsQ__mh_w_1639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1639_equation_0, values = (var_16191_cast_fp16, var_15750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1639_cast_fp16")]; + string _SplitHeadsQ__mh_w_1641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1641_equation_0, values = (var_16195_cast_fp16, var_15757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1641_cast_fp16")]; + string _SplitHeadsQ__mh_w_1643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1643_equation_0, values = (var_16195_cast_fp16, var_15764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1643_cast_fp16")]; + string _SplitHeadsQ__mh_w_1645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1645_equation_0, values = (var_16195_cast_fp16, var_15771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1645_cast_fp16")]; + string _SplitHeadsQ__mh_w_1647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1647_equation_0, values = (var_16195_cast_fp16, var_15778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1647_cast_fp16")]; + string _SplitHeadsQ__mh_w_1649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1649_equation_0, values = (var_16199_cast_fp16, var_15785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1649_cast_fp16")]; + string _SplitHeadsQ__mh_w_1651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1651_equation_0, values = (var_16199_cast_fp16, var_15792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1651_cast_fp16")]; + string _SplitHeadsQ__mh_w_1653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1653_equation_0, values = (var_16199_cast_fp16, var_15799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1653_cast_fp16")]; + string _SplitHeadsQ__mh_w_1655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1655_equation_0, values = (var_16199_cast_fp16, var_15806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1655_cast_fp16")]; + string _SplitHeadsQ__mh_w_1657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1657_equation_0, values = (var_16203_cast_fp16, var_15813_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1657_cast_fp16")]; + string _SplitHeadsQ__mh_w_1659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1659_equation_0, values = (var_16203_cast_fp16, var_15820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1659_cast_fp16")]; + string _SplitHeadsQ__mh_w_1661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1661_equation_0, values = (var_16203_cast_fp16, var_15827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1661_cast_fp16")]; + string _SplitHeadsQ__mh_w_1663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1663_equation_0, values = (var_16203_cast_fp16, var_15834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1663_cast_fp16")]; + string _SplitHeadsQ__mh_w_1665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1665_equation_0, values = (var_16207_cast_fp16, var_15841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1665_cast_fp16")]; + string _SplitHeadsQ__mh_w_1667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1667_equation_0, values = (var_16207_cast_fp16, var_15848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1667_cast_fp16")]; + string _SplitHeadsQ__mh_w_1669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1669_equation_0, values = (var_16207_cast_fp16, var_15855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1669_cast_fp16")]; + string _SplitHeadsQ__mh_w_1671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1671_equation_0, values = (var_16207_cast_fp16, var_15862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1671_cast_fp16")]; + string _SplitHeadsQ__mh_w_1673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1673_equation_0, values = (var_16211_cast_fp16, var_15869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1673_cast_fp16")]; + string _SplitHeadsQ__mh_w_1675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1675_equation_0, values = (var_16211_cast_fp16, var_15876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1675_cast_fp16")]; + string _SplitHeadsQ__mh_w_1677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1677_equation_0, values = (var_16211_cast_fp16, var_15883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1677_cast_fp16")]; + string _SplitHeadsQ__mh_w_1679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1679_equation_0, values = (var_16211_cast_fp16, var_15890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1679_cast_fp16")]; + string _SplitHeadsQ__mh_w_1681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1681_equation_0, values = (var_16215_cast_fp16, var_15897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1681_cast_fp16")]; + string _SplitHeadsQ__mh_w_1683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1683_equation_0, values = (var_16215_cast_fp16, var_15904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1683_cast_fp16")]; + string _SplitHeadsQ__mh_w_1685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1685_equation_0, values = (var_16215_cast_fp16, var_15911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1685_cast_fp16")]; + string _SplitHeadsQ__mh_w_1687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1687_equation_0, values = (var_16215_cast_fp16, var_15918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1687_cast_fp16")]; + string _SplitHeadsQ__mh_w_1689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1689_equation_0, values = (var_16219_cast_fp16, var_15925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1689_cast_fp16")]; + string _SplitHeadsQ__mh_w_1691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1691_equation_0, values = (var_16219_cast_fp16, var_15932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1691_cast_fp16")]; + string _SplitHeadsQ__mh_w_1693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1693_equation_0, values = (var_16219_cast_fp16, var_15939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1693_cast_fp16")]; + string _SplitHeadsQ__mh_w_1695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1695_equation_0, values = (var_16219_cast_fp16, var_15946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1695_cast_fp16")]; + string _SplitHeadsQ__mh_w_1697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1697_equation_0, values = (var_16223_cast_fp16, var_15953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1697_cast_fp16")]; + string _SplitHeadsQ__mh_w_1699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1699_equation_0, values = (var_16223_cast_fp16, var_15960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1699_cast_fp16")]; + string _SplitHeadsQ__mh_w_1701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1701_equation_0, values = (var_16223_cast_fp16, var_15967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1701_cast_fp16")]; + string _SplitHeadsQ__mh_w_1703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1703_equation_0, values = (var_16223_cast_fp16, var_15974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1703_cast_fp16")]; + string _SplitHeadsQ__mh_w_1705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1705_equation_0, values = (var_16227_cast_fp16, var_15981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1705_cast_fp16")]; + string _SplitHeadsQ__mh_w_1707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1707_equation_0, values = (var_16227_cast_fp16, var_15988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1707_cast_fp16")]; + string _SplitHeadsQ__mh_w_1709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1709_equation_0, values = (var_16227_cast_fp16, var_15995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1709_cast_fp16")]; + string _SplitHeadsQ__mh_w_1711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1711_equation_0, values = (var_16227_cast_fp16, var_16002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1711_cast_fp16")]; + string _SplitHeadsQ__mh_w_1713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1713_equation_0, values = (var_16231_cast_fp16, var_16009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1713_cast_fp16")]; + string _SplitHeadsQ__mh_w_1715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1715_equation_0, values = (var_16231_cast_fp16, var_16016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1715_cast_fp16")]; + string _SplitHeadsQ__mh_w_1717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1717_equation_0, values = (var_16231_cast_fp16, var_16023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1717_cast_fp16")]; + string _SplitHeadsQ__mh_w_1719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1719_equation_0, values = (var_16231_cast_fp16, var_16030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1719_cast_fp16")]; + string _SplitHeadsQ__mh_w_1721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1721_equation_0, values = (var_16235_cast_fp16, var_16037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1721_cast_fp16")]; + string _SplitHeadsQ__mh_w_1723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1723_equation_0, values = (var_16235_cast_fp16, var_16044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1723_cast_fp16")]; + string _SplitHeadsQ__mh_w_1725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1725_equation_0, values = (var_16235_cast_fp16, var_16051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1725_cast_fp16")]; + string _SplitHeadsQ__mh_w_1727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1727_equation_0, values = (var_16235_cast_fp16, var_16058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1727_cast_fp16")]; + string _SplitHeadsQ__mh_w_1729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1729_equation_0, values = (var_16239_cast_fp16, var_16065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1729_cast_fp16")]; + string _SplitHeadsQ__mh_w_1731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1731_equation_0, values = (var_16239_cast_fp16, var_16072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1731_cast_fp16")]; + string _SplitHeadsQ__mh_w_1733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1733_equation_0, values = (var_16239_cast_fp16, var_16079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1733_cast_fp16")]; + string _SplitHeadsQ__mh_w_1735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1735_equation_0, values = (var_16239_cast_fp16, var_16086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1735_cast_fp16")]; + string _SplitHeadsQ__mh_w_1737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1737_equation_0, values = (var_16243_cast_fp16, var_16093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1737_cast_fp16")]; + string _SplitHeadsQ__mh_w_1739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1739_equation_0, values = (var_16243_cast_fp16, var_16100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1739_cast_fp16")]; + string _SplitHeadsQ__mh_w_1741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1741_equation_0, values = (var_16243_cast_fp16, var_16107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1741_cast_fp16")]; + string _SplitHeadsQ__mh_w_1743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1743_equation_0, values = (var_16243_cast_fp16, var_16114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1743_cast_fp16")]; + string _SplitHeadsQ__mh_w_1745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1745_equation_0, values = (var_16247_cast_fp16, var_16121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1745_cast_fp16")]; + string _SplitHeadsQ__mh_w_1747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1747_equation_0, values = (var_16247_cast_fp16, var_16128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1747_cast_fp16")]; + string _SplitHeadsQ__mh_w_1749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1749_equation_0, values = (var_16247_cast_fp16, var_16135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1749_cast_fp16")]; + string _SplitHeadsQ__mh_w_1751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1751_equation_0, values = (var_16247_cast_fp16, var_16142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1751_cast_fp16")]; + string _SplitHeadsQ__mh_w_1753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1753_equation_0, values = (var_16251_cast_fp16, var_16149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1753_cast_fp16")]; + string _SplitHeadsQ__mh_w_1755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1755_equation_0, values = (var_16251_cast_fp16, var_16156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1755_cast_fp16")]; + string _SplitHeadsQ__mh_w_1757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1757_equation_0, values = (var_16251_cast_fp16, var_16163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1757_cast_fp16")]; + string _SplitHeadsQ__mh_w_1759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1759_equation_0, values = (var_16251_cast_fp16, var_16170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1759_cast_fp16")]; + fp16 var_16492_to_fp16 = const()[name = string("op_16492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1601_cast_fp16, y = var_16492_to_fp16)[name = string("aw_chunk_1601_cast_fp16")]; + fp16 var_16494_to_fp16 = const()[name = string("op_16494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1603_cast_fp16, y = var_16494_to_fp16)[name = string("aw_chunk_1603_cast_fp16")]; + fp16 var_16496_to_fp16 = const()[name = string("op_16496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1605_cast_fp16, y = var_16496_to_fp16)[name = string("aw_chunk_1605_cast_fp16")]; + fp16 var_16498_to_fp16 = const()[name = string("op_16498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1607_cast_fp16, y = var_16498_to_fp16)[name = string("aw_chunk_1607_cast_fp16")]; + fp16 var_16500_to_fp16 = const()[name = string("op_16500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1609_cast_fp16, y = var_16500_to_fp16)[name = string("aw_chunk_1609_cast_fp16")]; + fp16 var_16502_to_fp16 = const()[name = string("op_16502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1611_cast_fp16, y = var_16502_to_fp16)[name = string("aw_chunk_1611_cast_fp16")]; + fp16 var_16504_to_fp16 = const()[name = string("op_16504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1613_cast_fp16, y = var_16504_to_fp16)[name = string("aw_chunk_1613_cast_fp16")]; + fp16 var_16506_to_fp16 = const()[name = string("op_16506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1615_cast_fp16, y = var_16506_to_fp16)[name = string("aw_chunk_1615_cast_fp16")]; + fp16 var_16508_to_fp16 = const()[name = string("op_16508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1617_cast_fp16, y = var_16508_to_fp16)[name = string("aw_chunk_1617_cast_fp16")]; + fp16 var_16510_to_fp16 = const()[name = string("op_16510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1619_cast_fp16, y = var_16510_to_fp16)[name = string("aw_chunk_1619_cast_fp16")]; + fp16 var_16512_to_fp16 = const()[name = string("op_16512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1621_cast_fp16, y = var_16512_to_fp16)[name = string("aw_chunk_1621_cast_fp16")]; + fp16 var_16514_to_fp16 = const()[name = string("op_16514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1623_cast_fp16, y = var_16514_to_fp16)[name = string("aw_chunk_1623_cast_fp16")]; + fp16 var_16516_to_fp16 = const()[name = string("op_16516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1625_cast_fp16, y = var_16516_to_fp16)[name = string("aw_chunk_1625_cast_fp16")]; + fp16 var_16518_to_fp16 = const()[name = string("op_16518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1627_cast_fp16, y = var_16518_to_fp16)[name = string("aw_chunk_1627_cast_fp16")]; + fp16 var_16520_to_fp16 = const()[name = string("op_16520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1629_cast_fp16, y = var_16520_to_fp16)[name = string("aw_chunk_1629_cast_fp16")]; + fp16 var_16522_to_fp16 = const()[name = string("op_16522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1631_cast_fp16, y = var_16522_to_fp16)[name = string("aw_chunk_1631_cast_fp16")]; + fp16 var_16524_to_fp16 = const()[name = string("op_16524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1633_cast_fp16, y = var_16524_to_fp16)[name = string("aw_chunk_1633_cast_fp16")]; + fp16 var_16526_to_fp16 = const()[name = string("op_16526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1635_cast_fp16, y = var_16526_to_fp16)[name = string("aw_chunk_1635_cast_fp16")]; + fp16 var_16528_to_fp16 = const()[name = string("op_16528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1637_cast_fp16, y = var_16528_to_fp16)[name = string("aw_chunk_1637_cast_fp16")]; + fp16 var_16530_to_fp16 = const()[name = string("op_16530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1639_cast_fp16, y = var_16530_to_fp16)[name = string("aw_chunk_1639_cast_fp16")]; + fp16 var_16532_to_fp16 = const()[name = string("op_16532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1641_cast_fp16, y = var_16532_to_fp16)[name = string("aw_chunk_1641_cast_fp16")]; + fp16 var_16534_to_fp16 = const()[name = string("op_16534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1643_cast_fp16, y = var_16534_to_fp16)[name = string("aw_chunk_1643_cast_fp16")]; + fp16 var_16536_to_fp16 = const()[name = string("op_16536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1645_cast_fp16, y = var_16536_to_fp16)[name = string("aw_chunk_1645_cast_fp16")]; + fp16 var_16538_to_fp16 = const()[name = string("op_16538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1647_cast_fp16, y = var_16538_to_fp16)[name = string("aw_chunk_1647_cast_fp16")]; + fp16 var_16540_to_fp16 = const()[name = string("op_16540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1649_cast_fp16, y = var_16540_to_fp16)[name = string("aw_chunk_1649_cast_fp16")]; + fp16 var_16542_to_fp16 = const()[name = string("op_16542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1651_cast_fp16, y = var_16542_to_fp16)[name = string("aw_chunk_1651_cast_fp16")]; + fp16 var_16544_to_fp16 = const()[name = string("op_16544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1653_cast_fp16, y = var_16544_to_fp16)[name = string("aw_chunk_1653_cast_fp16")]; + fp16 var_16546_to_fp16 = const()[name = string("op_16546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1655_cast_fp16, y = var_16546_to_fp16)[name = string("aw_chunk_1655_cast_fp16")]; + fp16 var_16548_to_fp16 = const()[name = string("op_16548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1657_cast_fp16, y = var_16548_to_fp16)[name = string("aw_chunk_1657_cast_fp16")]; + fp16 var_16550_to_fp16 = const()[name = string("op_16550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1659_cast_fp16, y = var_16550_to_fp16)[name = string("aw_chunk_1659_cast_fp16")]; + fp16 var_16552_to_fp16 = const()[name = string("op_16552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1661_cast_fp16, y = var_16552_to_fp16)[name = string("aw_chunk_1661_cast_fp16")]; + fp16 var_16554_to_fp16 = const()[name = string("op_16554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1663_cast_fp16, y = var_16554_to_fp16)[name = string("aw_chunk_1663_cast_fp16")]; + fp16 var_16556_to_fp16 = const()[name = string("op_16556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1665_cast_fp16, y = var_16556_to_fp16)[name = string("aw_chunk_1665_cast_fp16")]; + fp16 var_16558_to_fp16 = const()[name = string("op_16558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1667_cast_fp16, y = var_16558_to_fp16)[name = string("aw_chunk_1667_cast_fp16")]; + fp16 var_16560_to_fp16 = const()[name = string("op_16560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1669_cast_fp16, y = var_16560_to_fp16)[name = string("aw_chunk_1669_cast_fp16")]; + fp16 var_16562_to_fp16 = const()[name = string("op_16562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1671_cast_fp16, y = var_16562_to_fp16)[name = string("aw_chunk_1671_cast_fp16")]; + fp16 var_16564_to_fp16 = const()[name = string("op_16564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1673_cast_fp16, y = var_16564_to_fp16)[name = string("aw_chunk_1673_cast_fp16")]; + fp16 var_16566_to_fp16 = const()[name = string("op_16566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1675_cast_fp16, y = var_16566_to_fp16)[name = string("aw_chunk_1675_cast_fp16")]; + fp16 var_16568_to_fp16 = const()[name = string("op_16568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1677_cast_fp16, y = var_16568_to_fp16)[name = string("aw_chunk_1677_cast_fp16")]; + fp16 var_16570_to_fp16 = const()[name = string("op_16570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1679_cast_fp16, y = var_16570_to_fp16)[name = string("aw_chunk_1679_cast_fp16")]; + fp16 var_16572_to_fp16 = const()[name = string("op_16572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1681_cast_fp16, y = var_16572_to_fp16)[name = string("aw_chunk_1681_cast_fp16")]; + fp16 var_16574_to_fp16 = const()[name = string("op_16574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1683_cast_fp16, y = var_16574_to_fp16)[name = string("aw_chunk_1683_cast_fp16")]; + fp16 var_16576_to_fp16 = const()[name = string("op_16576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1685_cast_fp16, y = var_16576_to_fp16)[name = string("aw_chunk_1685_cast_fp16")]; + fp16 var_16578_to_fp16 = const()[name = string("op_16578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1687_cast_fp16, y = var_16578_to_fp16)[name = string("aw_chunk_1687_cast_fp16")]; + fp16 var_16580_to_fp16 = const()[name = string("op_16580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1689_cast_fp16, y = var_16580_to_fp16)[name = string("aw_chunk_1689_cast_fp16")]; + fp16 var_16582_to_fp16 = const()[name = string("op_16582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1691_cast_fp16, y = var_16582_to_fp16)[name = string("aw_chunk_1691_cast_fp16")]; + fp16 var_16584_to_fp16 = const()[name = string("op_16584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1693_cast_fp16, y = var_16584_to_fp16)[name = string("aw_chunk_1693_cast_fp16")]; + fp16 var_16586_to_fp16 = const()[name = string("op_16586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1695_cast_fp16, y = var_16586_to_fp16)[name = string("aw_chunk_1695_cast_fp16")]; + fp16 var_16588_to_fp16 = const()[name = string("op_16588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1697_cast_fp16, y = var_16588_to_fp16)[name = string("aw_chunk_1697_cast_fp16")]; + fp16 var_16590_to_fp16 = const()[name = string("op_16590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1699_cast_fp16, y = var_16590_to_fp16)[name = string("aw_chunk_1699_cast_fp16")]; + fp16 var_16592_to_fp16 = const()[name = string("op_16592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1701_cast_fp16, y = var_16592_to_fp16)[name = string("aw_chunk_1701_cast_fp16")]; + fp16 var_16594_to_fp16 = const()[name = string("op_16594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1703_cast_fp16, y = var_16594_to_fp16)[name = string("aw_chunk_1703_cast_fp16")]; + fp16 var_16596_to_fp16 = const()[name = string("op_16596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1705_cast_fp16, y = var_16596_to_fp16)[name = string("aw_chunk_1705_cast_fp16")]; + fp16 var_16598_to_fp16 = const()[name = string("op_16598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1707_cast_fp16, y = var_16598_to_fp16)[name = string("aw_chunk_1707_cast_fp16")]; + fp16 var_16600_to_fp16 = const()[name = string("op_16600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1709_cast_fp16, y = var_16600_to_fp16)[name = string("aw_chunk_1709_cast_fp16")]; + fp16 var_16602_to_fp16 = const()[name = string("op_16602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1711_cast_fp16, y = var_16602_to_fp16)[name = string("aw_chunk_1711_cast_fp16")]; + fp16 var_16604_to_fp16 = const()[name = string("op_16604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1713_cast_fp16, y = var_16604_to_fp16)[name = string("aw_chunk_1713_cast_fp16")]; + fp16 var_16606_to_fp16 = const()[name = string("op_16606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1715_cast_fp16, y = var_16606_to_fp16)[name = string("aw_chunk_1715_cast_fp16")]; + fp16 var_16608_to_fp16 = const()[name = string("op_16608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1717_cast_fp16, y = var_16608_to_fp16)[name = string("aw_chunk_1717_cast_fp16")]; + fp16 var_16610_to_fp16 = const()[name = string("op_16610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1719_cast_fp16, y = var_16610_to_fp16)[name = string("aw_chunk_1719_cast_fp16")]; + fp16 var_16612_to_fp16 = const()[name = string("op_16612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1721_cast_fp16, y = var_16612_to_fp16)[name = string("aw_chunk_1721_cast_fp16")]; + fp16 var_16614_to_fp16 = const()[name = string("op_16614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1723_cast_fp16, y = var_16614_to_fp16)[name = string("aw_chunk_1723_cast_fp16")]; + fp16 var_16616_to_fp16 = const()[name = string("op_16616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1725_cast_fp16, y = var_16616_to_fp16)[name = string("aw_chunk_1725_cast_fp16")]; + fp16 var_16618_to_fp16 = const()[name = string("op_16618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1727_cast_fp16, y = var_16618_to_fp16)[name = string("aw_chunk_1727_cast_fp16")]; + fp16 var_16620_to_fp16 = const()[name = string("op_16620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1729_cast_fp16, y = var_16620_to_fp16)[name = string("aw_chunk_1729_cast_fp16")]; + fp16 var_16622_to_fp16 = const()[name = string("op_16622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1731_cast_fp16, y = var_16622_to_fp16)[name = string("aw_chunk_1731_cast_fp16")]; + fp16 var_16624_to_fp16 = const()[name = string("op_16624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1733_cast_fp16, y = var_16624_to_fp16)[name = string("aw_chunk_1733_cast_fp16")]; + fp16 var_16626_to_fp16 = const()[name = string("op_16626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1735_cast_fp16, y = var_16626_to_fp16)[name = string("aw_chunk_1735_cast_fp16")]; + fp16 var_16628_to_fp16 = const()[name = string("op_16628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1737_cast_fp16, y = var_16628_to_fp16)[name = string("aw_chunk_1737_cast_fp16")]; + fp16 var_16630_to_fp16 = const()[name = string("op_16630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1739_cast_fp16, y = var_16630_to_fp16)[name = string("aw_chunk_1739_cast_fp16")]; + fp16 var_16632_to_fp16 = const()[name = string("op_16632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1741_cast_fp16, y = var_16632_to_fp16)[name = string("aw_chunk_1741_cast_fp16")]; + fp16 var_16634_to_fp16 = const()[name = string("op_16634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1743_cast_fp16, y = var_16634_to_fp16)[name = string("aw_chunk_1743_cast_fp16")]; + fp16 var_16636_to_fp16 = const()[name = string("op_16636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1745_cast_fp16, y = var_16636_to_fp16)[name = string("aw_chunk_1745_cast_fp16")]; + fp16 var_16638_to_fp16 = const()[name = string("op_16638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1747_cast_fp16, y = var_16638_to_fp16)[name = string("aw_chunk_1747_cast_fp16")]; + fp16 var_16640_to_fp16 = const()[name = string("op_16640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1749_cast_fp16, y = var_16640_to_fp16)[name = string("aw_chunk_1749_cast_fp16")]; + fp16 var_16642_to_fp16 = const()[name = string("op_16642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1751_cast_fp16, y = var_16642_to_fp16)[name = string("aw_chunk_1751_cast_fp16")]; + fp16 var_16644_to_fp16 = const()[name = string("op_16644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1753_cast_fp16, y = var_16644_to_fp16)[name = string("aw_chunk_1753_cast_fp16")]; + fp16 var_16646_to_fp16 = const()[name = string("op_16646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1755_cast_fp16, y = var_16646_to_fp16)[name = string("aw_chunk_1755_cast_fp16")]; + fp16 var_16648_to_fp16 = const()[name = string("op_16648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1757_cast_fp16, y = var_16648_to_fp16)[name = string("aw_chunk_1757_cast_fp16")]; + fp16 var_16650_to_fp16 = const()[name = string("op_16650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1759_cast_fp16, y = var_16650_to_fp16)[name = string("aw_chunk_1759_cast_fp16")]; + tensor var_16652_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1601_cast_fp16)[name = string("op_16652_cast_fp16")]; + tensor var_16653_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1603_cast_fp16)[name = string("op_16653_cast_fp16")]; + tensor var_16654_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1605_cast_fp16)[name = string("op_16654_cast_fp16")]; + tensor var_16655_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1607_cast_fp16)[name = string("op_16655_cast_fp16")]; + tensor var_16656_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1609_cast_fp16)[name = string("op_16656_cast_fp16")]; + tensor var_16657_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1611_cast_fp16)[name = string("op_16657_cast_fp16")]; + tensor var_16658_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1613_cast_fp16)[name = string("op_16658_cast_fp16")]; + tensor var_16659_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1615_cast_fp16)[name = string("op_16659_cast_fp16")]; + tensor var_16660_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1617_cast_fp16)[name = string("op_16660_cast_fp16")]; + tensor var_16661_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1619_cast_fp16)[name = string("op_16661_cast_fp16")]; + tensor var_16662_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1621_cast_fp16)[name = string("op_16662_cast_fp16")]; + tensor var_16663_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1623_cast_fp16)[name = string("op_16663_cast_fp16")]; + tensor var_16664_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1625_cast_fp16)[name = string("op_16664_cast_fp16")]; + tensor var_16665_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1627_cast_fp16)[name = string("op_16665_cast_fp16")]; + tensor var_16666_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1629_cast_fp16)[name = string("op_16666_cast_fp16")]; + tensor var_16667_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1631_cast_fp16)[name = string("op_16667_cast_fp16")]; + tensor var_16668_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1633_cast_fp16)[name = string("op_16668_cast_fp16")]; + tensor var_16669_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1635_cast_fp16)[name = string("op_16669_cast_fp16")]; + tensor var_16670_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1637_cast_fp16)[name = string("op_16670_cast_fp16")]; + tensor var_16671_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1639_cast_fp16)[name = string("op_16671_cast_fp16")]; + tensor var_16672_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1641_cast_fp16)[name = string("op_16672_cast_fp16")]; + tensor var_16673_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1643_cast_fp16)[name = string("op_16673_cast_fp16")]; + tensor var_16674_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1645_cast_fp16)[name = string("op_16674_cast_fp16")]; + tensor var_16675_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1647_cast_fp16)[name = string("op_16675_cast_fp16")]; + tensor var_16676_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1649_cast_fp16)[name = string("op_16676_cast_fp16")]; + tensor var_16677_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1651_cast_fp16)[name = string("op_16677_cast_fp16")]; + tensor var_16678_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1653_cast_fp16)[name = string("op_16678_cast_fp16")]; + tensor var_16679_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1655_cast_fp16)[name = string("op_16679_cast_fp16")]; + tensor var_16680_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1657_cast_fp16)[name = string("op_16680_cast_fp16")]; + tensor var_16681_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1659_cast_fp16)[name = string("op_16681_cast_fp16")]; + tensor var_16682_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1661_cast_fp16)[name = string("op_16682_cast_fp16")]; + tensor var_16683_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1663_cast_fp16)[name = string("op_16683_cast_fp16")]; + tensor var_16684_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1665_cast_fp16)[name = string("op_16684_cast_fp16")]; + tensor var_16685_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1667_cast_fp16)[name = string("op_16685_cast_fp16")]; + tensor var_16686_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1669_cast_fp16)[name = string("op_16686_cast_fp16")]; + tensor var_16687_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1671_cast_fp16)[name = string("op_16687_cast_fp16")]; + tensor var_16688_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1673_cast_fp16)[name = string("op_16688_cast_fp16")]; + tensor var_16689_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1675_cast_fp16)[name = string("op_16689_cast_fp16")]; + tensor var_16690_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1677_cast_fp16)[name = string("op_16690_cast_fp16")]; + tensor var_16691_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1679_cast_fp16)[name = string("op_16691_cast_fp16")]; + tensor var_16692_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1681_cast_fp16)[name = string("op_16692_cast_fp16")]; + tensor var_16693_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1683_cast_fp16)[name = string("op_16693_cast_fp16")]; + tensor var_16694_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1685_cast_fp16)[name = string("op_16694_cast_fp16")]; + tensor var_16695_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1687_cast_fp16)[name = string("op_16695_cast_fp16")]; + tensor var_16696_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1689_cast_fp16)[name = string("op_16696_cast_fp16")]; + tensor var_16697_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1691_cast_fp16)[name = string("op_16697_cast_fp16")]; + tensor var_16698_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1693_cast_fp16)[name = string("op_16698_cast_fp16")]; + tensor var_16699_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1695_cast_fp16)[name = string("op_16699_cast_fp16")]; + tensor var_16700_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1697_cast_fp16)[name = string("op_16700_cast_fp16")]; + tensor var_16701_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1699_cast_fp16)[name = string("op_16701_cast_fp16")]; + tensor var_16702_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1701_cast_fp16)[name = string("op_16702_cast_fp16")]; + tensor var_16703_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1703_cast_fp16)[name = string("op_16703_cast_fp16")]; + tensor var_16704_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1705_cast_fp16)[name = string("op_16704_cast_fp16")]; + tensor var_16705_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1707_cast_fp16)[name = string("op_16705_cast_fp16")]; + tensor var_16706_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1709_cast_fp16)[name = string("op_16706_cast_fp16")]; + tensor var_16707_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1711_cast_fp16)[name = string("op_16707_cast_fp16")]; + tensor var_16708_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1713_cast_fp16)[name = string("op_16708_cast_fp16")]; + tensor var_16709_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1715_cast_fp16)[name = string("op_16709_cast_fp16")]; + tensor var_16710_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1717_cast_fp16)[name = string("op_16710_cast_fp16")]; + tensor var_16711_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1719_cast_fp16)[name = string("op_16711_cast_fp16")]; + tensor var_16712_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1721_cast_fp16)[name = string("op_16712_cast_fp16")]; + tensor var_16713_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1723_cast_fp16)[name = string("op_16713_cast_fp16")]; + tensor var_16714_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1725_cast_fp16)[name = string("op_16714_cast_fp16")]; + tensor var_16715_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1727_cast_fp16)[name = string("op_16715_cast_fp16")]; + tensor var_16716_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1729_cast_fp16)[name = string("op_16716_cast_fp16")]; + tensor var_16717_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1731_cast_fp16)[name = string("op_16717_cast_fp16")]; + tensor var_16718_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1733_cast_fp16)[name = string("op_16718_cast_fp16")]; + tensor var_16719_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1735_cast_fp16)[name = string("op_16719_cast_fp16")]; + tensor var_16720_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1737_cast_fp16)[name = string("op_16720_cast_fp16")]; + tensor var_16721_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1739_cast_fp16)[name = string("op_16721_cast_fp16")]; + tensor var_16722_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1741_cast_fp16)[name = string("op_16722_cast_fp16")]; + tensor var_16723_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1743_cast_fp16)[name = string("op_16723_cast_fp16")]; + tensor var_16724_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1745_cast_fp16)[name = string("op_16724_cast_fp16")]; + tensor var_16725_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1747_cast_fp16)[name = string("op_16725_cast_fp16")]; + tensor var_16726_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1749_cast_fp16)[name = string("op_16726_cast_fp16")]; + tensor var_16727_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1751_cast_fp16)[name = string("op_16727_cast_fp16")]; + tensor var_16728_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1753_cast_fp16)[name = string("op_16728_cast_fp16")]; + tensor var_16729_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1755_cast_fp16)[name = string("op_16729_cast_fp16")]; + tensor var_16730_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1757_cast_fp16)[name = string("op_16730_cast_fp16")]; + tensor var_16731_cast_fp16 = softmax(axis = var_15477, x = aw_chunk_1759_cast_fp16)[name = string("op_16731_cast_fp16")]; + string var_16733_equation_0 = const()[name = string("op_16733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16733_cast_fp16 = einsum(equation = var_16733_equation_0, values = (var_16253_cast_fp16, var_16652_cast_fp16))[name = string("op_16733_cast_fp16")]; + string var_16735_equation_0 = const()[name = string("op_16735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16735_cast_fp16 = einsum(equation = var_16735_equation_0, values = (var_16253_cast_fp16, var_16653_cast_fp16))[name = string("op_16735_cast_fp16")]; + string var_16737_equation_0 = const()[name = string("op_16737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16737_cast_fp16 = einsum(equation = var_16737_equation_0, values = (var_16253_cast_fp16, var_16654_cast_fp16))[name = string("op_16737_cast_fp16")]; + string var_16739_equation_0 = const()[name = string("op_16739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16739_cast_fp16 = einsum(equation = var_16739_equation_0, values = (var_16253_cast_fp16, var_16655_cast_fp16))[name = string("op_16739_cast_fp16")]; + string var_16741_equation_0 = const()[name = string("op_16741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16741_cast_fp16 = einsum(equation = var_16741_equation_0, values = (var_16257_cast_fp16, var_16656_cast_fp16))[name = string("op_16741_cast_fp16")]; + string var_16743_equation_0 = const()[name = string("op_16743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16743_cast_fp16 = einsum(equation = var_16743_equation_0, values = (var_16257_cast_fp16, var_16657_cast_fp16))[name = string("op_16743_cast_fp16")]; + string var_16745_equation_0 = const()[name = string("op_16745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16745_cast_fp16 = einsum(equation = var_16745_equation_0, values = (var_16257_cast_fp16, var_16658_cast_fp16))[name = string("op_16745_cast_fp16")]; + string var_16747_equation_0 = const()[name = string("op_16747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16747_cast_fp16 = einsum(equation = var_16747_equation_0, values = (var_16257_cast_fp16, var_16659_cast_fp16))[name = string("op_16747_cast_fp16")]; + string var_16749_equation_0 = const()[name = string("op_16749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16749_cast_fp16 = einsum(equation = var_16749_equation_0, values = (var_16261_cast_fp16, var_16660_cast_fp16))[name = string("op_16749_cast_fp16")]; + string var_16751_equation_0 = const()[name = string("op_16751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16751_cast_fp16 = einsum(equation = var_16751_equation_0, values = (var_16261_cast_fp16, var_16661_cast_fp16))[name = string("op_16751_cast_fp16")]; + string var_16753_equation_0 = const()[name = string("op_16753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16753_cast_fp16 = einsum(equation = var_16753_equation_0, values = (var_16261_cast_fp16, var_16662_cast_fp16))[name = string("op_16753_cast_fp16")]; + string var_16755_equation_0 = const()[name = string("op_16755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16755_cast_fp16 = einsum(equation = var_16755_equation_0, values = (var_16261_cast_fp16, var_16663_cast_fp16))[name = string("op_16755_cast_fp16")]; + string var_16757_equation_0 = const()[name = string("op_16757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16757_cast_fp16 = einsum(equation = var_16757_equation_0, values = (var_16265_cast_fp16, var_16664_cast_fp16))[name = string("op_16757_cast_fp16")]; + string var_16759_equation_0 = const()[name = string("op_16759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16759_cast_fp16 = einsum(equation = var_16759_equation_0, values = (var_16265_cast_fp16, var_16665_cast_fp16))[name = string("op_16759_cast_fp16")]; + string var_16761_equation_0 = const()[name = string("op_16761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16761_cast_fp16 = einsum(equation = var_16761_equation_0, values = (var_16265_cast_fp16, var_16666_cast_fp16))[name = string("op_16761_cast_fp16")]; + string var_16763_equation_0 = const()[name = string("op_16763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16763_cast_fp16 = einsum(equation = var_16763_equation_0, values = (var_16265_cast_fp16, var_16667_cast_fp16))[name = string("op_16763_cast_fp16")]; + string var_16765_equation_0 = const()[name = string("op_16765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16765_cast_fp16 = einsum(equation = var_16765_equation_0, values = (var_16269_cast_fp16, var_16668_cast_fp16))[name = string("op_16765_cast_fp16")]; + string var_16767_equation_0 = const()[name = string("op_16767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16767_cast_fp16 = einsum(equation = var_16767_equation_0, values = (var_16269_cast_fp16, var_16669_cast_fp16))[name = string("op_16767_cast_fp16")]; + string var_16769_equation_0 = const()[name = string("op_16769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16769_cast_fp16 = einsum(equation = var_16769_equation_0, values = (var_16269_cast_fp16, var_16670_cast_fp16))[name = string("op_16769_cast_fp16")]; + string var_16771_equation_0 = const()[name = string("op_16771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16771_cast_fp16 = einsum(equation = var_16771_equation_0, values = (var_16269_cast_fp16, var_16671_cast_fp16))[name = string("op_16771_cast_fp16")]; + string var_16773_equation_0 = const()[name = string("op_16773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16773_cast_fp16 = einsum(equation = var_16773_equation_0, values = (var_16273_cast_fp16, var_16672_cast_fp16))[name = string("op_16773_cast_fp16")]; + string var_16775_equation_0 = const()[name = string("op_16775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16775_cast_fp16 = einsum(equation = var_16775_equation_0, values = (var_16273_cast_fp16, var_16673_cast_fp16))[name = string("op_16775_cast_fp16")]; + string var_16777_equation_0 = const()[name = string("op_16777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16777_cast_fp16 = einsum(equation = var_16777_equation_0, values = (var_16273_cast_fp16, var_16674_cast_fp16))[name = string("op_16777_cast_fp16")]; + string var_16779_equation_0 = const()[name = string("op_16779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16779_cast_fp16 = einsum(equation = var_16779_equation_0, values = (var_16273_cast_fp16, var_16675_cast_fp16))[name = string("op_16779_cast_fp16")]; + string var_16781_equation_0 = const()[name = string("op_16781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16781_cast_fp16 = einsum(equation = var_16781_equation_0, values = (var_16277_cast_fp16, var_16676_cast_fp16))[name = string("op_16781_cast_fp16")]; + string var_16783_equation_0 = const()[name = string("op_16783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16783_cast_fp16 = einsum(equation = var_16783_equation_0, values = (var_16277_cast_fp16, var_16677_cast_fp16))[name = string("op_16783_cast_fp16")]; + string var_16785_equation_0 = const()[name = string("op_16785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16785_cast_fp16 = einsum(equation = var_16785_equation_0, values = (var_16277_cast_fp16, var_16678_cast_fp16))[name = string("op_16785_cast_fp16")]; + string var_16787_equation_0 = const()[name = string("op_16787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16787_cast_fp16 = einsum(equation = var_16787_equation_0, values = (var_16277_cast_fp16, var_16679_cast_fp16))[name = string("op_16787_cast_fp16")]; + string var_16789_equation_0 = const()[name = string("op_16789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16789_cast_fp16 = einsum(equation = var_16789_equation_0, values = (var_16281_cast_fp16, var_16680_cast_fp16))[name = string("op_16789_cast_fp16")]; + string var_16791_equation_0 = const()[name = string("op_16791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16791_cast_fp16 = einsum(equation = var_16791_equation_0, values = (var_16281_cast_fp16, var_16681_cast_fp16))[name = string("op_16791_cast_fp16")]; + string var_16793_equation_0 = const()[name = string("op_16793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16793_cast_fp16 = einsum(equation = var_16793_equation_0, values = (var_16281_cast_fp16, var_16682_cast_fp16))[name = string("op_16793_cast_fp16")]; + string var_16795_equation_0 = const()[name = string("op_16795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16795_cast_fp16 = einsum(equation = var_16795_equation_0, values = (var_16281_cast_fp16, var_16683_cast_fp16))[name = string("op_16795_cast_fp16")]; + string var_16797_equation_0 = const()[name = string("op_16797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16797_cast_fp16 = einsum(equation = var_16797_equation_0, values = (var_16285_cast_fp16, var_16684_cast_fp16))[name = string("op_16797_cast_fp16")]; + string var_16799_equation_0 = const()[name = string("op_16799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16799_cast_fp16 = einsum(equation = var_16799_equation_0, values = (var_16285_cast_fp16, var_16685_cast_fp16))[name = string("op_16799_cast_fp16")]; + string var_16801_equation_0 = const()[name = string("op_16801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16801_cast_fp16 = einsum(equation = var_16801_equation_0, values = (var_16285_cast_fp16, var_16686_cast_fp16))[name = string("op_16801_cast_fp16")]; + string var_16803_equation_0 = const()[name = string("op_16803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16803_cast_fp16 = einsum(equation = var_16803_equation_0, values = (var_16285_cast_fp16, var_16687_cast_fp16))[name = string("op_16803_cast_fp16")]; + string var_16805_equation_0 = const()[name = string("op_16805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16805_cast_fp16 = einsum(equation = var_16805_equation_0, values = (var_16289_cast_fp16, var_16688_cast_fp16))[name = string("op_16805_cast_fp16")]; + string var_16807_equation_0 = const()[name = string("op_16807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16807_cast_fp16 = einsum(equation = var_16807_equation_0, values = (var_16289_cast_fp16, var_16689_cast_fp16))[name = string("op_16807_cast_fp16")]; + string var_16809_equation_0 = const()[name = string("op_16809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16809_cast_fp16 = einsum(equation = var_16809_equation_0, values = (var_16289_cast_fp16, var_16690_cast_fp16))[name = string("op_16809_cast_fp16")]; + string var_16811_equation_0 = const()[name = string("op_16811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16811_cast_fp16 = einsum(equation = var_16811_equation_0, values = (var_16289_cast_fp16, var_16691_cast_fp16))[name = string("op_16811_cast_fp16")]; + string var_16813_equation_0 = const()[name = string("op_16813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16813_cast_fp16 = einsum(equation = var_16813_equation_0, values = (var_16293_cast_fp16, var_16692_cast_fp16))[name = string("op_16813_cast_fp16")]; + string var_16815_equation_0 = const()[name = string("op_16815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16815_cast_fp16 = einsum(equation = var_16815_equation_0, values = (var_16293_cast_fp16, var_16693_cast_fp16))[name = string("op_16815_cast_fp16")]; + string var_16817_equation_0 = const()[name = string("op_16817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16817_cast_fp16 = einsum(equation = var_16817_equation_0, values = (var_16293_cast_fp16, var_16694_cast_fp16))[name = string("op_16817_cast_fp16")]; + string var_16819_equation_0 = const()[name = string("op_16819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16819_cast_fp16 = einsum(equation = var_16819_equation_0, values = (var_16293_cast_fp16, var_16695_cast_fp16))[name = string("op_16819_cast_fp16")]; + string var_16821_equation_0 = const()[name = string("op_16821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16821_cast_fp16 = einsum(equation = var_16821_equation_0, values = (var_16297_cast_fp16, var_16696_cast_fp16))[name = string("op_16821_cast_fp16")]; + string var_16823_equation_0 = const()[name = string("op_16823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16823_cast_fp16 = einsum(equation = var_16823_equation_0, values = (var_16297_cast_fp16, var_16697_cast_fp16))[name = string("op_16823_cast_fp16")]; + string var_16825_equation_0 = const()[name = string("op_16825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16825_cast_fp16 = einsum(equation = var_16825_equation_0, values = (var_16297_cast_fp16, var_16698_cast_fp16))[name = string("op_16825_cast_fp16")]; + string var_16827_equation_0 = const()[name = string("op_16827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16827_cast_fp16 = einsum(equation = var_16827_equation_0, values = (var_16297_cast_fp16, var_16699_cast_fp16))[name = string("op_16827_cast_fp16")]; + string var_16829_equation_0 = const()[name = string("op_16829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16829_cast_fp16 = einsum(equation = var_16829_equation_0, values = (var_16301_cast_fp16, var_16700_cast_fp16))[name = string("op_16829_cast_fp16")]; + string var_16831_equation_0 = const()[name = string("op_16831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16831_cast_fp16 = einsum(equation = var_16831_equation_0, values = (var_16301_cast_fp16, var_16701_cast_fp16))[name = string("op_16831_cast_fp16")]; + string var_16833_equation_0 = const()[name = string("op_16833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16833_cast_fp16 = einsum(equation = var_16833_equation_0, values = (var_16301_cast_fp16, var_16702_cast_fp16))[name = string("op_16833_cast_fp16")]; + string var_16835_equation_0 = const()[name = string("op_16835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16835_cast_fp16 = einsum(equation = var_16835_equation_0, values = (var_16301_cast_fp16, var_16703_cast_fp16))[name = string("op_16835_cast_fp16")]; + string var_16837_equation_0 = const()[name = string("op_16837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16837_cast_fp16 = einsum(equation = var_16837_equation_0, values = (var_16305_cast_fp16, var_16704_cast_fp16))[name = string("op_16837_cast_fp16")]; + string var_16839_equation_0 = const()[name = string("op_16839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16839_cast_fp16 = einsum(equation = var_16839_equation_0, values = (var_16305_cast_fp16, var_16705_cast_fp16))[name = string("op_16839_cast_fp16")]; + string var_16841_equation_0 = const()[name = string("op_16841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16841_cast_fp16 = einsum(equation = var_16841_equation_0, values = (var_16305_cast_fp16, var_16706_cast_fp16))[name = string("op_16841_cast_fp16")]; + string var_16843_equation_0 = const()[name = string("op_16843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16843_cast_fp16 = einsum(equation = var_16843_equation_0, values = (var_16305_cast_fp16, var_16707_cast_fp16))[name = string("op_16843_cast_fp16")]; + string var_16845_equation_0 = const()[name = string("op_16845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16845_cast_fp16 = einsum(equation = var_16845_equation_0, values = (var_16309_cast_fp16, var_16708_cast_fp16))[name = string("op_16845_cast_fp16")]; + string var_16847_equation_0 = const()[name = string("op_16847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16847_cast_fp16 = einsum(equation = var_16847_equation_0, values = (var_16309_cast_fp16, var_16709_cast_fp16))[name = string("op_16847_cast_fp16")]; + string var_16849_equation_0 = const()[name = string("op_16849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16849_cast_fp16 = einsum(equation = var_16849_equation_0, values = (var_16309_cast_fp16, var_16710_cast_fp16))[name = string("op_16849_cast_fp16")]; + string var_16851_equation_0 = const()[name = string("op_16851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16851_cast_fp16 = einsum(equation = var_16851_equation_0, values = (var_16309_cast_fp16, var_16711_cast_fp16))[name = string("op_16851_cast_fp16")]; + string var_16853_equation_0 = const()[name = string("op_16853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16853_cast_fp16 = einsum(equation = var_16853_equation_0, values = (var_16313_cast_fp16, var_16712_cast_fp16))[name = string("op_16853_cast_fp16")]; + string var_16855_equation_0 = const()[name = string("op_16855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16855_cast_fp16 = einsum(equation = var_16855_equation_0, values = (var_16313_cast_fp16, var_16713_cast_fp16))[name = string("op_16855_cast_fp16")]; + string var_16857_equation_0 = const()[name = string("op_16857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16857_cast_fp16 = einsum(equation = var_16857_equation_0, values = (var_16313_cast_fp16, var_16714_cast_fp16))[name = string("op_16857_cast_fp16")]; + string var_16859_equation_0 = const()[name = string("op_16859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16859_cast_fp16 = einsum(equation = var_16859_equation_0, values = (var_16313_cast_fp16, var_16715_cast_fp16))[name = string("op_16859_cast_fp16")]; + string var_16861_equation_0 = const()[name = string("op_16861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16861_cast_fp16 = einsum(equation = var_16861_equation_0, values = (var_16317_cast_fp16, var_16716_cast_fp16))[name = string("op_16861_cast_fp16")]; + string var_16863_equation_0 = const()[name = string("op_16863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16863_cast_fp16 = einsum(equation = var_16863_equation_0, values = (var_16317_cast_fp16, var_16717_cast_fp16))[name = string("op_16863_cast_fp16")]; + string var_16865_equation_0 = const()[name = string("op_16865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16865_cast_fp16 = einsum(equation = var_16865_equation_0, values = (var_16317_cast_fp16, var_16718_cast_fp16))[name = string("op_16865_cast_fp16")]; + string var_16867_equation_0 = const()[name = string("op_16867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16867_cast_fp16 = einsum(equation = var_16867_equation_0, values = (var_16317_cast_fp16, var_16719_cast_fp16))[name = string("op_16867_cast_fp16")]; + string var_16869_equation_0 = const()[name = string("op_16869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16869_cast_fp16 = einsum(equation = var_16869_equation_0, values = (var_16321_cast_fp16, var_16720_cast_fp16))[name = string("op_16869_cast_fp16")]; + string var_16871_equation_0 = const()[name = string("op_16871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16871_cast_fp16 = einsum(equation = var_16871_equation_0, values = (var_16321_cast_fp16, var_16721_cast_fp16))[name = string("op_16871_cast_fp16")]; + string var_16873_equation_0 = const()[name = string("op_16873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16873_cast_fp16 = einsum(equation = var_16873_equation_0, values = (var_16321_cast_fp16, var_16722_cast_fp16))[name = string("op_16873_cast_fp16")]; + string var_16875_equation_0 = const()[name = string("op_16875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16875_cast_fp16 = einsum(equation = var_16875_equation_0, values = (var_16321_cast_fp16, var_16723_cast_fp16))[name = string("op_16875_cast_fp16")]; + string var_16877_equation_0 = const()[name = string("op_16877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16877_cast_fp16 = einsum(equation = var_16877_equation_0, values = (var_16325_cast_fp16, var_16724_cast_fp16))[name = string("op_16877_cast_fp16")]; + string var_16879_equation_0 = const()[name = string("op_16879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16879_cast_fp16 = einsum(equation = var_16879_equation_0, values = (var_16325_cast_fp16, var_16725_cast_fp16))[name = string("op_16879_cast_fp16")]; + string var_16881_equation_0 = const()[name = string("op_16881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16881_cast_fp16 = einsum(equation = var_16881_equation_0, values = (var_16325_cast_fp16, var_16726_cast_fp16))[name = string("op_16881_cast_fp16")]; + string var_16883_equation_0 = const()[name = string("op_16883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16883_cast_fp16 = einsum(equation = var_16883_equation_0, values = (var_16325_cast_fp16, var_16727_cast_fp16))[name = string("op_16883_cast_fp16")]; + string var_16885_equation_0 = const()[name = string("op_16885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16885_cast_fp16 = einsum(equation = var_16885_equation_0, values = (var_16329_cast_fp16, var_16728_cast_fp16))[name = string("op_16885_cast_fp16")]; + string var_16887_equation_0 = const()[name = string("op_16887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16887_cast_fp16 = einsum(equation = var_16887_equation_0, values = (var_16329_cast_fp16, var_16729_cast_fp16))[name = string("op_16887_cast_fp16")]; + string var_16889_equation_0 = const()[name = string("op_16889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16889_cast_fp16 = einsum(equation = var_16889_equation_0, values = (var_16329_cast_fp16, var_16730_cast_fp16))[name = string("op_16889_cast_fp16")]; + string var_16891_equation_0 = const()[name = string("op_16891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_16891_cast_fp16 = einsum(equation = var_16891_equation_0, values = (var_16329_cast_fp16, var_16731_cast_fp16))[name = string("op_16891_cast_fp16")]; + bool var_16893_interleave_0 = const()[name = string("op_16893_interleave_0"), val = bool(false)]; + tensor var_16893_cast_fp16 = concat(axis = var_15452, interleave = var_16893_interleave_0, values = (var_16733_cast_fp16, var_16735_cast_fp16, var_16737_cast_fp16, var_16739_cast_fp16))[name = string("op_16893_cast_fp16")]; + bool var_16895_interleave_0 = const()[name = string("op_16895_interleave_0"), val = bool(false)]; + tensor var_16895_cast_fp16 = concat(axis = var_15452, interleave = var_16895_interleave_0, values = (var_16741_cast_fp16, var_16743_cast_fp16, var_16745_cast_fp16, var_16747_cast_fp16))[name = string("op_16895_cast_fp16")]; + bool var_16897_interleave_0 = const()[name = string("op_16897_interleave_0"), val = bool(false)]; + tensor var_16897_cast_fp16 = concat(axis = var_15452, interleave = var_16897_interleave_0, values = (var_16749_cast_fp16, var_16751_cast_fp16, var_16753_cast_fp16, var_16755_cast_fp16))[name = string("op_16897_cast_fp16")]; + bool var_16899_interleave_0 = const()[name = string("op_16899_interleave_0"), val = bool(false)]; + tensor var_16899_cast_fp16 = concat(axis = var_15452, interleave = var_16899_interleave_0, values = (var_16757_cast_fp16, var_16759_cast_fp16, var_16761_cast_fp16, var_16763_cast_fp16))[name = string("op_16899_cast_fp16")]; + bool var_16901_interleave_0 = const()[name = string("op_16901_interleave_0"), val = bool(false)]; + tensor var_16901_cast_fp16 = concat(axis = var_15452, interleave = var_16901_interleave_0, values = (var_16765_cast_fp16, var_16767_cast_fp16, var_16769_cast_fp16, var_16771_cast_fp16))[name = string("op_16901_cast_fp16")]; + bool var_16903_interleave_0 = const()[name = string("op_16903_interleave_0"), val = bool(false)]; + tensor var_16903_cast_fp16 = concat(axis = var_15452, interleave = var_16903_interleave_0, values = (var_16773_cast_fp16, var_16775_cast_fp16, var_16777_cast_fp16, var_16779_cast_fp16))[name = string("op_16903_cast_fp16")]; + bool var_16905_interleave_0 = const()[name = string("op_16905_interleave_0"), val = bool(false)]; + tensor var_16905_cast_fp16 = concat(axis = var_15452, interleave = var_16905_interleave_0, values = (var_16781_cast_fp16, var_16783_cast_fp16, var_16785_cast_fp16, var_16787_cast_fp16))[name = string("op_16905_cast_fp16")]; + bool var_16907_interleave_0 = const()[name = string("op_16907_interleave_0"), val = bool(false)]; + tensor var_16907_cast_fp16 = concat(axis = var_15452, interleave = var_16907_interleave_0, values = (var_16789_cast_fp16, var_16791_cast_fp16, var_16793_cast_fp16, var_16795_cast_fp16))[name = string("op_16907_cast_fp16")]; + bool var_16909_interleave_0 = const()[name = string("op_16909_interleave_0"), val = bool(false)]; + tensor var_16909_cast_fp16 = concat(axis = var_15452, interleave = var_16909_interleave_0, values = (var_16797_cast_fp16, var_16799_cast_fp16, var_16801_cast_fp16, var_16803_cast_fp16))[name = string("op_16909_cast_fp16")]; + bool var_16911_interleave_0 = const()[name = string("op_16911_interleave_0"), val = bool(false)]; + tensor var_16911_cast_fp16 = concat(axis = var_15452, interleave = var_16911_interleave_0, values = (var_16805_cast_fp16, var_16807_cast_fp16, var_16809_cast_fp16, var_16811_cast_fp16))[name = string("op_16911_cast_fp16")]; + bool var_16913_interleave_0 = const()[name = string("op_16913_interleave_0"), val = bool(false)]; + tensor var_16913_cast_fp16 = concat(axis = var_15452, interleave = var_16913_interleave_0, values = (var_16813_cast_fp16, var_16815_cast_fp16, var_16817_cast_fp16, var_16819_cast_fp16))[name = string("op_16913_cast_fp16")]; + bool var_16915_interleave_0 = const()[name = string("op_16915_interleave_0"), val = bool(false)]; + tensor var_16915_cast_fp16 = concat(axis = var_15452, interleave = var_16915_interleave_0, values = (var_16821_cast_fp16, var_16823_cast_fp16, var_16825_cast_fp16, var_16827_cast_fp16))[name = string("op_16915_cast_fp16")]; + bool var_16917_interleave_0 = const()[name = string("op_16917_interleave_0"), val = bool(false)]; + tensor var_16917_cast_fp16 = concat(axis = var_15452, interleave = var_16917_interleave_0, values = (var_16829_cast_fp16, var_16831_cast_fp16, var_16833_cast_fp16, var_16835_cast_fp16))[name = string("op_16917_cast_fp16")]; + bool var_16919_interleave_0 = const()[name = string("op_16919_interleave_0"), val = bool(false)]; + tensor var_16919_cast_fp16 = concat(axis = var_15452, interleave = var_16919_interleave_0, values = (var_16837_cast_fp16, var_16839_cast_fp16, var_16841_cast_fp16, var_16843_cast_fp16))[name = string("op_16919_cast_fp16")]; + bool var_16921_interleave_0 = const()[name = string("op_16921_interleave_0"), val = bool(false)]; + tensor var_16921_cast_fp16 = concat(axis = var_15452, interleave = var_16921_interleave_0, values = (var_16845_cast_fp16, var_16847_cast_fp16, var_16849_cast_fp16, var_16851_cast_fp16))[name = string("op_16921_cast_fp16")]; + bool var_16923_interleave_0 = const()[name = string("op_16923_interleave_0"), val = bool(false)]; + tensor var_16923_cast_fp16 = concat(axis = var_15452, interleave = var_16923_interleave_0, values = (var_16853_cast_fp16, var_16855_cast_fp16, var_16857_cast_fp16, var_16859_cast_fp16))[name = string("op_16923_cast_fp16")]; + bool var_16925_interleave_0 = const()[name = string("op_16925_interleave_0"), val = bool(false)]; + tensor var_16925_cast_fp16 = concat(axis = var_15452, interleave = var_16925_interleave_0, values = (var_16861_cast_fp16, var_16863_cast_fp16, var_16865_cast_fp16, var_16867_cast_fp16))[name = string("op_16925_cast_fp16")]; + bool var_16927_interleave_0 = const()[name = string("op_16927_interleave_0"), val = bool(false)]; + tensor var_16927_cast_fp16 = concat(axis = var_15452, interleave = var_16927_interleave_0, values = (var_16869_cast_fp16, var_16871_cast_fp16, var_16873_cast_fp16, var_16875_cast_fp16))[name = string("op_16927_cast_fp16")]; + bool var_16929_interleave_0 = const()[name = string("op_16929_interleave_0"), val = bool(false)]; + tensor var_16929_cast_fp16 = concat(axis = var_15452, interleave = var_16929_interleave_0, values = (var_16877_cast_fp16, var_16879_cast_fp16, var_16881_cast_fp16, var_16883_cast_fp16))[name = string("op_16929_cast_fp16")]; + bool var_16931_interleave_0 = const()[name = string("op_16931_interleave_0"), val = bool(false)]; + tensor var_16931_cast_fp16 = concat(axis = var_15452, interleave = var_16931_interleave_0, values = (var_16885_cast_fp16, var_16887_cast_fp16, var_16889_cast_fp16, var_16891_cast_fp16))[name = string("op_16931_cast_fp16")]; + bool input_81_interleave_0 = const()[name = string("input_81_interleave_0"), val = bool(false)]; + tensor input_81_cast_fp16 = concat(axis = var_15477, interleave = input_81_interleave_0, values = (var_16893_cast_fp16, var_16895_cast_fp16, var_16897_cast_fp16, var_16899_cast_fp16, var_16901_cast_fp16, var_16903_cast_fp16, var_16905_cast_fp16, var_16907_cast_fp16, var_16909_cast_fp16, var_16911_cast_fp16, var_16913_cast_fp16, var_16915_cast_fp16, var_16917_cast_fp16, var_16919_cast_fp16, var_16921_cast_fp16, var_16923_cast_fp16, var_16925_cast_fp16, var_16927_cast_fp16, var_16929_cast_fp16, var_16931_cast_fp16))[name = string("input_81_cast_fp16")]; + string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; + tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; + tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; + int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418038080)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421314944)))]; + tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_16950_to_fp16 = const()[name = string("op_16950_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_16950_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421317568)))]; + tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421320192)))]; + fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; + string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; + tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; + tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; + int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421322816)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434430080)))]; + tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; + tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434440384)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447547648)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; + int32 var_16979 = const()[name = string("op_16979"), val = int32(3)]; + int32 var_17004 = const()[name = string("op_17004"), val = int32(1)]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_17021_to_fp16 = const()[name = string("op_17021_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_17021_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447550272)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447552896)))]; + fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; + string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; + tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; + tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; + int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447555520)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450832384)))]; + tensor query_23_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_23_cast_fp16")]; + string key_23_pad_type_0 = const()[name = string("key_23_pad_type_0"), val = string("valid")]; + tensor key_23_strides_0 = const()[name = string("key_23_strides_0"), val = tensor([1, 1])]; + tensor key_23_pad_0 = const()[name = string("key_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_23_dilations_0 = const()[name = string("key_23_dilations_0"), val = tensor([1, 1])]; + int32 key_23_groups_0 = const()[name = string("key_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450835008)))]; + tensor key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_23_cast_fp16")]; + string value_23_pad_type_0 = const()[name = string("value_23_pad_type_0"), val = string("valid")]; + tensor value_23_strides_0 = const()[name = string("value_23_strides_0"), val = tensor([1, 1])]; + tensor value_23_pad_0 = const()[name = string("value_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_23_dilations_0 = const()[name = string("value_23_dilations_0"), val = tensor([1, 1])]; + int32 value_23_groups_0 = const()[name = string("value_23_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454111872)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457388736)))]; + tensor value_23_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_23_cast_fp16")]; + tensor var_17059_begin_0 = const()[name = string("op_17059_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17059_end_0 = const()[name = string("op_17059_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17059_end_mask_0 = const()[name = string("op_17059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17059_cast_fp16 = slice_by_index(begin = var_17059_begin_0, end = var_17059_end_0, end_mask = var_17059_end_mask_0, x = query_23_cast_fp16)[name = string("op_17059_cast_fp16")]; + tensor var_17063_begin_0 = const()[name = string("op_17063_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17063_end_0 = const()[name = string("op_17063_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_17063_end_mask_0 = const()[name = string("op_17063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17063_cast_fp16 = slice_by_index(begin = var_17063_begin_0, end = var_17063_end_0, end_mask = var_17063_end_mask_0, x = query_23_cast_fp16)[name = string("op_17063_cast_fp16")]; + tensor var_17067_begin_0 = const()[name = string("op_17067_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17067_end_0 = const()[name = string("op_17067_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_17067_end_mask_0 = const()[name = string("op_17067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17067_cast_fp16 = slice_by_index(begin = var_17067_begin_0, end = var_17067_end_0, end_mask = var_17067_end_mask_0, x = query_23_cast_fp16)[name = string("op_17067_cast_fp16")]; + tensor var_17071_begin_0 = const()[name = string("op_17071_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17071_end_0 = const()[name = string("op_17071_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_17071_end_mask_0 = const()[name = string("op_17071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17071_cast_fp16 = slice_by_index(begin = var_17071_begin_0, end = var_17071_end_0, end_mask = var_17071_end_mask_0, x = query_23_cast_fp16)[name = string("op_17071_cast_fp16")]; + tensor var_17075_begin_0 = const()[name = string("op_17075_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17075_end_0 = const()[name = string("op_17075_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_17075_end_mask_0 = const()[name = string("op_17075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17075_cast_fp16 = slice_by_index(begin = var_17075_begin_0, end = var_17075_end_0, end_mask = var_17075_end_mask_0, x = query_23_cast_fp16)[name = string("op_17075_cast_fp16")]; + tensor var_17079_begin_0 = const()[name = string("op_17079_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17079_end_0 = const()[name = string("op_17079_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_17079_end_mask_0 = const()[name = string("op_17079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17079_cast_fp16 = slice_by_index(begin = var_17079_begin_0, end = var_17079_end_0, end_mask = var_17079_end_mask_0, x = query_23_cast_fp16)[name = string("op_17079_cast_fp16")]; + tensor var_17083_begin_0 = const()[name = string("op_17083_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17083_end_0 = const()[name = string("op_17083_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_17083_end_mask_0 = const()[name = string("op_17083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17083_cast_fp16 = slice_by_index(begin = var_17083_begin_0, end = var_17083_end_0, end_mask = var_17083_end_mask_0, x = query_23_cast_fp16)[name = string("op_17083_cast_fp16")]; + tensor var_17087_begin_0 = const()[name = string("op_17087_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17087_end_0 = const()[name = string("op_17087_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_17087_end_mask_0 = const()[name = string("op_17087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17087_cast_fp16 = slice_by_index(begin = var_17087_begin_0, end = var_17087_end_0, end_mask = var_17087_end_mask_0, x = query_23_cast_fp16)[name = string("op_17087_cast_fp16")]; + tensor var_17091_begin_0 = const()[name = string("op_17091_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17091_end_0 = const()[name = string("op_17091_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_17091_end_mask_0 = const()[name = string("op_17091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17091_cast_fp16 = slice_by_index(begin = var_17091_begin_0, end = var_17091_end_0, end_mask = var_17091_end_mask_0, x = query_23_cast_fp16)[name = string("op_17091_cast_fp16")]; + tensor var_17095_begin_0 = const()[name = string("op_17095_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17095_end_0 = const()[name = string("op_17095_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_17095_end_mask_0 = const()[name = string("op_17095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17095_cast_fp16 = slice_by_index(begin = var_17095_begin_0, end = var_17095_end_0, end_mask = var_17095_end_mask_0, x = query_23_cast_fp16)[name = string("op_17095_cast_fp16")]; + tensor var_17099_begin_0 = const()[name = string("op_17099_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17099_end_0 = const()[name = string("op_17099_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_17099_end_mask_0 = const()[name = string("op_17099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17099_cast_fp16 = slice_by_index(begin = var_17099_begin_0, end = var_17099_end_0, end_mask = var_17099_end_mask_0, x = query_23_cast_fp16)[name = string("op_17099_cast_fp16")]; + tensor var_17103_begin_0 = const()[name = string("op_17103_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17103_end_0 = const()[name = string("op_17103_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_17103_end_mask_0 = const()[name = string("op_17103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17103_cast_fp16 = slice_by_index(begin = var_17103_begin_0, end = var_17103_end_0, end_mask = var_17103_end_mask_0, x = query_23_cast_fp16)[name = string("op_17103_cast_fp16")]; + tensor var_17107_begin_0 = const()[name = string("op_17107_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17107_end_0 = const()[name = string("op_17107_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_17107_end_mask_0 = const()[name = string("op_17107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17107_cast_fp16 = slice_by_index(begin = var_17107_begin_0, end = var_17107_end_0, end_mask = var_17107_end_mask_0, x = query_23_cast_fp16)[name = string("op_17107_cast_fp16")]; + tensor var_17111_begin_0 = const()[name = string("op_17111_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17111_end_0 = const()[name = string("op_17111_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_17111_end_mask_0 = const()[name = string("op_17111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17111_cast_fp16 = slice_by_index(begin = var_17111_begin_0, end = var_17111_end_0, end_mask = var_17111_end_mask_0, x = query_23_cast_fp16)[name = string("op_17111_cast_fp16")]; + tensor var_17115_begin_0 = const()[name = string("op_17115_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17115_end_0 = const()[name = string("op_17115_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_17115_end_mask_0 = const()[name = string("op_17115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17115_cast_fp16 = slice_by_index(begin = var_17115_begin_0, end = var_17115_end_0, end_mask = var_17115_end_mask_0, x = query_23_cast_fp16)[name = string("op_17115_cast_fp16")]; + tensor var_17119_begin_0 = const()[name = string("op_17119_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17119_end_0 = const()[name = string("op_17119_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_17119_end_mask_0 = const()[name = string("op_17119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17119_cast_fp16 = slice_by_index(begin = var_17119_begin_0, end = var_17119_end_0, end_mask = var_17119_end_mask_0, x = query_23_cast_fp16)[name = string("op_17119_cast_fp16")]; + tensor var_17123_begin_0 = const()[name = string("op_17123_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17123_end_0 = const()[name = string("op_17123_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_17123_end_mask_0 = const()[name = string("op_17123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17123_cast_fp16 = slice_by_index(begin = var_17123_begin_0, end = var_17123_end_0, end_mask = var_17123_end_mask_0, x = query_23_cast_fp16)[name = string("op_17123_cast_fp16")]; + tensor var_17127_begin_0 = const()[name = string("op_17127_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17127_end_0 = const()[name = string("op_17127_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_17127_end_mask_0 = const()[name = string("op_17127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17127_cast_fp16 = slice_by_index(begin = var_17127_begin_0, end = var_17127_end_0, end_mask = var_17127_end_mask_0, x = query_23_cast_fp16)[name = string("op_17127_cast_fp16")]; + tensor var_17131_begin_0 = const()[name = string("op_17131_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17131_end_0 = const()[name = string("op_17131_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_17131_end_mask_0 = const()[name = string("op_17131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17131_cast_fp16 = slice_by_index(begin = var_17131_begin_0, end = var_17131_end_0, end_mask = var_17131_end_mask_0, x = query_23_cast_fp16)[name = string("op_17131_cast_fp16")]; + tensor var_17135_begin_0 = const()[name = string("op_17135_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17135_end_0 = const()[name = string("op_17135_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_17135_end_mask_0 = const()[name = string("op_17135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17135_cast_fp16 = slice_by_index(begin = var_17135_begin_0, end = var_17135_end_0, end_mask = var_17135_end_mask_0, x = query_23_cast_fp16)[name = string("op_17135_cast_fp16")]; + tensor var_17144_begin_0 = const()[name = string("op_17144_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17144_end_0 = const()[name = string("op_17144_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17144_end_mask_0 = const()[name = string("op_17144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17144_cast_fp16 = slice_by_index(begin = var_17144_begin_0, end = var_17144_end_0, end_mask = var_17144_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17144_cast_fp16")]; + tensor var_17151_begin_0 = const()[name = string("op_17151_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17151_end_0 = const()[name = string("op_17151_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17151_end_mask_0 = const()[name = string("op_17151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17151_cast_fp16 = slice_by_index(begin = var_17151_begin_0, end = var_17151_end_0, end_mask = var_17151_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17151_cast_fp16")]; + tensor var_17158_begin_0 = const()[name = string("op_17158_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17158_end_0 = const()[name = string("op_17158_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17158_end_mask_0 = const()[name = string("op_17158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17158_cast_fp16 = slice_by_index(begin = var_17158_begin_0, end = var_17158_end_0, end_mask = var_17158_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17158_cast_fp16")]; + tensor var_17165_begin_0 = const()[name = string("op_17165_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17165_end_0 = const()[name = string("op_17165_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17165_end_mask_0 = const()[name = string("op_17165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17165_cast_fp16 = slice_by_index(begin = var_17165_begin_0, end = var_17165_end_0, end_mask = var_17165_end_mask_0, x = var_17059_cast_fp16)[name = string("op_17165_cast_fp16")]; + tensor var_17172_begin_0 = const()[name = string("op_17172_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17172_end_0 = const()[name = string("op_17172_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17172_end_mask_0 = const()[name = string("op_17172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17172_cast_fp16 = slice_by_index(begin = var_17172_begin_0, end = var_17172_end_0, end_mask = var_17172_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17172_cast_fp16")]; + tensor var_17179_begin_0 = const()[name = string("op_17179_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17179_end_0 = const()[name = string("op_17179_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17179_end_mask_0 = const()[name = string("op_17179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17179_cast_fp16 = slice_by_index(begin = var_17179_begin_0, end = var_17179_end_0, end_mask = var_17179_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17179_cast_fp16")]; + tensor var_17186_begin_0 = const()[name = string("op_17186_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17186_end_0 = const()[name = string("op_17186_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17186_end_mask_0 = const()[name = string("op_17186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17186_cast_fp16 = slice_by_index(begin = var_17186_begin_0, end = var_17186_end_0, end_mask = var_17186_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17186_cast_fp16")]; + tensor var_17193_begin_0 = const()[name = string("op_17193_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17193_end_0 = const()[name = string("op_17193_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17193_end_mask_0 = const()[name = string("op_17193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17193_cast_fp16 = slice_by_index(begin = var_17193_begin_0, end = var_17193_end_0, end_mask = var_17193_end_mask_0, x = var_17063_cast_fp16)[name = string("op_17193_cast_fp16")]; + tensor var_17200_begin_0 = const()[name = string("op_17200_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17200_end_0 = const()[name = string("op_17200_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17200_end_mask_0 = const()[name = string("op_17200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17200_cast_fp16 = slice_by_index(begin = var_17200_begin_0, end = var_17200_end_0, end_mask = var_17200_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17200_cast_fp16")]; + tensor var_17207_begin_0 = const()[name = string("op_17207_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17207_end_0 = const()[name = string("op_17207_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17207_end_mask_0 = const()[name = string("op_17207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17207_cast_fp16 = slice_by_index(begin = var_17207_begin_0, end = var_17207_end_0, end_mask = var_17207_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17207_cast_fp16")]; + tensor var_17214_begin_0 = const()[name = string("op_17214_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17214_end_0 = const()[name = string("op_17214_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17214_end_mask_0 = const()[name = string("op_17214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17214_cast_fp16 = slice_by_index(begin = var_17214_begin_0, end = var_17214_end_0, end_mask = var_17214_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17214_cast_fp16")]; + tensor var_17221_begin_0 = const()[name = string("op_17221_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17221_end_0 = const()[name = string("op_17221_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17221_end_mask_0 = const()[name = string("op_17221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17221_cast_fp16 = slice_by_index(begin = var_17221_begin_0, end = var_17221_end_0, end_mask = var_17221_end_mask_0, x = var_17067_cast_fp16)[name = string("op_17221_cast_fp16")]; + tensor var_17228_begin_0 = const()[name = string("op_17228_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17228_end_0 = const()[name = string("op_17228_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17228_end_mask_0 = const()[name = string("op_17228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17228_cast_fp16 = slice_by_index(begin = var_17228_begin_0, end = var_17228_end_0, end_mask = var_17228_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17228_cast_fp16")]; + tensor var_17235_begin_0 = const()[name = string("op_17235_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17235_end_0 = const()[name = string("op_17235_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17235_end_mask_0 = const()[name = string("op_17235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17235_cast_fp16 = slice_by_index(begin = var_17235_begin_0, end = var_17235_end_0, end_mask = var_17235_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17235_cast_fp16")]; + tensor var_17242_begin_0 = const()[name = string("op_17242_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17242_end_0 = const()[name = string("op_17242_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17242_end_mask_0 = const()[name = string("op_17242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17242_cast_fp16 = slice_by_index(begin = var_17242_begin_0, end = var_17242_end_0, end_mask = var_17242_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17242_cast_fp16")]; + tensor var_17249_begin_0 = const()[name = string("op_17249_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17249_end_0 = const()[name = string("op_17249_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17249_end_mask_0 = const()[name = string("op_17249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17249_cast_fp16 = slice_by_index(begin = var_17249_begin_0, end = var_17249_end_0, end_mask = var_17249_end_mask_0, x = var_17071_cast_fp16)[name = string("op_17249_cast_fp16")]; + tensor var_17256_begin_0 = const()[name = string("op_17256_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17256_end_0 = const()[name = string("op_17256_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17256_end_mask_0 = const()[name = string("op_17256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17256_cast_fp16 = slice_by_index(begin = var_17256_begin_0, end = var_17256_end_0, end_mask = var_17256_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17256_cast_fp16")]; + tensor var_17263_begin_0 = const()[name = string("op_17263_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17263_end_0 = const()[name = string("op_17263_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17263_end_mask_0 = const()[name = string("op_17263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17263_cast_fp16 = slice_by_index(begin = var_17263_begin_0, end = var_17263_end_0, end_mask = var_17263_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17263_cast_fp16")]; + tensor var_17270_begin_0 = const()[name = string("op_17270_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17270_end_0 = const()[name = string("op_17270_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17270_end_mask_0 = const()[name = string("op_17270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17270_cast_fp16 = slice_by_index(begin = var_17270_begin_0, end = var_17270_end_0, end_mask = var_17270_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17270_cast_fp16")]; + tensor var_17277_begin_0 = const()[name = string("op_17277_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17277_end_0 = const()[name = string("op_17277_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17277_end_mask_0 = const()[name = string("op_17277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17277_cast_fp16 = slice_by_index(begin = var_17277_begin_0, end = var_17277_end_0, end_mask = var_17277_end_mask_0, x = var_17075_cast_fp16)[name = string("op_17277_cast_fp16")]; + tensor var_17284_begin_0 = const()[name = string("op_17284_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17284_end_0 = const()[name = string("op_17284_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17284_end_mask_0 = const()[name = string("op_17284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17284_cast_fp16 = slice_by_index(begin = var_17284_begin_0, end = var_17284_end_0, end_mask = var_17284_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17284_cast_fp16")]; + tensor var_17291_begin_0 = const()[name = string("op_17291_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17291_end_0 = const()[name = string("op_17291_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17291_end_mask_0 = const()[name = string("op_17291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17291_cast_fp16 = slice_by_index(begin = var_17291_begin_0, end = var_17291_end_0, end_mask = var_17291_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17291_cast_fp16")]; + tensor var_17298_begin_0 = const()[name = string("op_17298_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17298_end_0 = const()[name = string("op_17298_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17298_end_mask_0 = const()[name = string("op_17298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17298_cast_fp16 = slice_by_index(begin = var_17298_begin_0, end = var_17298_end_0, end_mask = var_17298_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17298_cast_fp16")]; + tensor var_17305_begin_0 = const()[name = string("op_17305_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17305_end_0 = const()[name = string("op_17305_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17305_end_mask_0 = const()[name = string("op_17305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17305_cast_fp16 = slice_by_index(begin = var_17305_begin_0, end = var_17305_end_0, end_mask = var_17305_end_mask_0, x = var_17079_cast_fp16)[name = string("op_17305_cast_fp16")]; + tensor var_17312_begin_0 = const()[name = string("op_17312_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17312_end_0 = const()[name = string("op_17312_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17312_end_mask_0 = const()[name = string("op_17312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17312_cast_fp16 = slice_by_index(begin = var_17312_begin_0, end = var_17312_end_0, end_mask = var_17312_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17312_cast_fp16")]; + tensor var_17319_begin_0 = const()[name = string("op_17319_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17319_end_0 = const()[name = string("op_17319_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17319_end_mask_0 = const()[name = string("op_17319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17319_cast_fp16 = slice_by_index(begin = var_17319_begin_0, end = var_17319_end_0, end_mask = var_17319_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17319_cast_fp16")]; + tensor var_17326_begin_0 = const()[name = string("op_17326_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17326_end_0 = const()[name = string("op_17326_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17326_end_mask_0 = const()[name = string("op_17326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17326_cast_fp16 = slice_by_index(begin = var_17326_begin_0, end = var_17326_end_0, end_mask = var_17326_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17326_cast_fp16")]; + tensor var_17333_begin_0 = const()[name = string("op_17333_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17333_end_0 = const()[name = string("op_17333_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17333_end_mask_0 = const()[name = string("op_17333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17333_cast_fp16 = slice_by_index(begin = var_17333_begin_0, end = var_17333_end_0, end_mask = var_17333_end_mask_0, x = var_17083_cast_fp16)[name = string("op_17333_cast_fp16")]; + tensor var_17340_begin_0 = const()[name = string("op_17340_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17340_end_0 = const()[name = string("op_17340_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17340_end_mask_0 = const()[name = string("op_17340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17340_cast_fp16 = slice_by_index(begin = var_17340_begin_0, end = var_17340_end_0, end_mask = var_17340_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17340_cast_fp16")]; + tensor var_17347_begin_0 = const()[name = string("op_17347_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17347_end_0 = const()[name = string("op_17347_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17347_end_mask_0 = const()[name = string("op_17347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17347_cast_fp16 = slice_by_index(begin = var_17347_begin_0, end = var_17347_end_0, end_mask = var_17347_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17347_cast_fp16")]; + tensor var_17354_begin_0 = const()[name = string("op_17354_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17354_end_0 = const()[name = string("op_17354_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17354_end_mask_0 = const()[name = string("op_17354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17354_cast_fp16 = slice_by_index(begin = var_17354_begin_0, end = var_17354_end_0, end_mask = var_17354_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17354_cast_fp16")]; + tensor var_17361_begin_0 = const()[name = string("op_17361_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17361_end_0 = const()[name = string("op_17361_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17361_end_mask_0 = const()[name = string("op_17361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17361_cast_fp16 = slice_by_index(begin = var_17361_begin_0, end = var_17361_end_0, end_mask = var_17361_end_mask_0, x = var_17087_cast_fp16)[name = string("op_17361_cast_fp16")]; + tensor var_17368_begin_0 = const()[name = string("op_17368_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17368_end_0 = const()[name = string("op_17368_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17368_end_mask_0 = const()[name = string("op_17368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17368_cast_fp16 = slice_by_index(begin = var_17368_begin_0, end = var_17368_end_0, end_mask = var_17368_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17368_cast_fp16")]; + tensor var_17375_begin_0 = const()[name = string("op_17375_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17375_end_0 = const()[name = string("op_17375_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17375_end_mask_0 = const()[name = string("op_17375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17375_cast_fp16 = slice_by_index(begin = var_17375_begin_0, end = var_17375_end_0, end_mask = var_17375_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17375_cast_fp16")]; + tensor var_17382_begin_0 = const()[name = string("op_17382_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17382_end_0 = const()[name = string("op_17382_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17382_end_mask_0 = const()[name = string("op_17382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17382_cast_fp16 = slice_by_index(begin = var_17382_begin_0, end = var_17382_end_0, end_mask = var_17382_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17382_cast_fp16")]; + tensor var_17389_begin_0 = const()[name = string("op_17389_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17389_end_0 = const()[name = string("op_17389_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17389_end_mask_0 = const()[name = string("op_17389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17389_cast_fp16 = slice_by_index(begin = var_17389_begin_0, end = var_17389_end_0, end_mask = var_17389_end_mask_0, x = var_17091_cast_fp16)[name = string("op_17389_cast_fp16")]; + tensor var_17396_begin_0 = const()[name = string("op_17396_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17396_end_0 = const()[name = string("op_17396_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17396_end_mask_0 = const()[name = string("op_17396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17396_cast_fp16 = slice_by_index(begin = var_17396_begin_0, end = var_17396_end_0, end_mask = var_17396_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17396_cast_fp16")]; + tensor var_17403_begin_0 = const()[name = string("op_17403_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17403_end_0 = const()[name = string("op_17403_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17403_end_mask_0 = const()[name = string("op_17403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17403_cast_fp16 = slice_by_index(begin = var_17403_begin_0, end = var_17403_end_0, end_mask = var_17403_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17403_cast_fp16")]; + tensor var_17410_begin_0 = const()[name = string("op_17410_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17410_end_0 = const()[name = string("op_17410_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17410_end_mask_0 = const()[name = string("op_17410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17410_cast_fp16 = slice_by_index(begin = var_17410_begin_0, end = var_17410_end_0, end_mask = var_17410_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17410_cast_fp16")]; + tensor var_17417_begin_0 = const()[name = string("op_17417_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17417_end_0 = const()[name = string("op_17417_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17417_end_mask_0 = const()[name = string("op_17417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17417_cast_fp16 = slice_by_index(begin = var_17417_begin_0, end = var_17417_end_0, end_mask = var_17417_end_mask_0, x = var_17095_cast_fp16)[name = string("op_17417_cast_fp16")]; + tensor var_17424_begin_0 = const()[name = string("op_17424_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17424_end_0 = const()[name = string("op_17424_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17424_end_mask_0 = const()[name = string("op_17424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17424_cast_fp16 = slice_by_index(begin = var_17424_begin_0, end = var_17424_end_0, end_mask = var_17424_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17424_cast_fp16")]; + tensor var_17431_begin_0 = const()[name = string("op_17431_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17431_end_0 = const()[name = string("op_17431_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17431_end_mask_0 = const()[name = string("op_17431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17431_cast_fp16 = slice_by_index(begin = var_17431_begin_0, end = var_17431_end_0, end_mask = var_17431_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17431_cast_fp16")]; + tensor var_17438_begin_0 = const()[name = string("op_17438_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17438_end_0 = const()[name = string("op_17438_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17438_end_mask_0 = const()[name = string("op_17438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17438_cast_fp16 = slice_by_index(begin = var_17438_begin_0, end = var_17438_end_0, end_mask = var_17438_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17438_cast_fp16")]; + tensor var_17445_begin_0 = const()[name = string("op_17445_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17445_end_0 = const()[name = string("op_17445_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17445_end_mask_0 = const()[name = string("op_17445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17445_cast_fp16 = slice_by_index(begin = var_17445_begin_0, end = var_17445_end_0, end_mask = var_17445_end_mask_0, x = var_17099_cast_fp16)[name = string("op_17445_cast_fp16")]; + tensor var_17452_begin_0 = const()[name = string("op_17452_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17452_end_0 = const()[name = string("op_17452_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17452_end_mask_0 = const()[name = string("op_17452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17452_cast_fp16 = slice_by_index(begin = var_17452_begin_0, end = var_17452_end_0, end_mask = var_17452_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17452_cast_fp16")]; + tensor var_17459_begin_0 = const()[name = string("op_17459_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17459_end_0 = const()[name = string("op_17459_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17459_end_mask_0 = const()[name = string("op_17459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17459_cast_fp16 = slice_by_index(begin = var_17459_begin_0, end = var_17459_end_0, end_mask = var_17459_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17459_cast_fp16")]; + tensor var_17466_begin_0 = const()[name = string("op_17466_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17466_end_0 = const()[name = string("op_17466_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17466_end_mask_0 = const()[name = string("op_17466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17466_cast_fp16 = slice_by_index(begin = var_17466_begin_0, end = var_17466_end_0, end_mask = var_17466_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17466_cast_fp16")]; + tensor var_17473_begin_0 = const()[name = string("op_17473_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17473_end_0 = const()[name = string("op_17473_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17473_end_mask_0 = const()[name = string("op_17473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17473_cast_fp16 = slice_by_index(begin = var_17473_begin_0, end = var_17473_end_0, end_mask = var_17473_end_mask_0, x = var_17103_cast_fp16)[name = string("op_17473_cast_fp16")]; + tensor var_17480_begin_0 = const()[name = string("op_17480_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17480_end_0 = const()[name = string("op_17480_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17480_end_mask_0 = const()[name = string("op_17480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17480_cast_fp16 = slice_by_index(begin = var_17480_begin_0, end = var_17480_end_0, end_mask = var_17480_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17480_cast_fp16")]; + tensor var_17487_begin_0 = const()[name = string("op_17487_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17487_end_0 = const()[name = string("op_17487_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17487_end_mask_0 = const()[name = string("op_17487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17487_cast_fp16 = slice_by_index(begin = var_17487_begin_0, end = var_17487_end_0, end_mask = var_17487_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17487_cast_fp16")]; + tensor var_17494_begin_0 = const()[name = string("op_17494_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17494_end_0 = const()[name = string("op_17494_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17494_end_mask_0 = const()[name = string("op_17494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17494_cast_fp16 = slice_by_index(begin = var_17494_begin_0, end = var_17494_end_0, end_mask = var_17494_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17494_cast_fp16")]; + tensor var_17501_begin_0 = const()[name = string("op_17501_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17501_end_0 = const()[name = string("op_17501_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17501_end_mask_0 = const()[name = string("op_17501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17501_cast_fp16 = slice_by_index(begin = var_17501_begin_0, end = var_17501_end_0, end_mask = var_17501_end_mask_0, x = var_17107_cast_fp16)[name = string("op_17501_cast_fp16")]; + tensor var_17508_begin_0 = const()[name = string("op_17508_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17508_end_0 = const()[name = string("op_17508_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17508_end_mask_0 = const()[name = string("op_17508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17508_cast_fp16 = slice_by_index(begin = var_17508_begin_0, end = var_17508_end_0, end_mask = var_17508_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17508_cast_fp16")]; + tensor var_17515_begin_0 = const()[name = string("op_17515_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17515_end_0 = const()[name = string("op_17515_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17515_end_mask_0 = const()[name = string("op_17515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17515_cast_fp16 = slice_by_index(begin = var_17515_begin_0, end = var_17515_end_0, end_mask = var_17515_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17515_cast_fp16")]; + tensor var_17522_begin_0 = const()[name = string("op_17522_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17522_end_0 = const()[name = string("op_17522_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17522_end_mask_0 = const()[name = string("op_17522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17522_cast_fp16 = slice_by_index(begin = var_17522_begin_0, end = var_17522_end_0, end_mask = var_17522_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17522_cast_fp16")]; + tensor var_17529_begin_0 = const()[name = string("op_17529_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17529_end_0 = const()[name = string("op_17529_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17529_end_mask_0 = const()[name = string("op_17529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17529_cast_fp16 = slice_by_index(begin = var_17529_begin_0, end = var_17529_end_0, end_mask = var_17529_end_mask_0, x = var_17111_cast_fp16)[name = string("op_17529_cast_fp16")]; + tensor var_17536_begin_0 = const()[name = string("op_17536_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17536_end_0 = const()[name = string("op_17536_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17536_end_mask_0 = const()[name = string("op_17536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17536_cast_fp16 = slice_by_index(begin = var_17536_begin_0, end = var_17536_end_0, end_mask = var_17536_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17536_cast_fp16")]; + tensor var_17543_begin_0 = const()[name = string("op_17543_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17543_end_0 = const()[name = string("op_17543_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17543_end_mask_0 = const()[name = string("op_17543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17543_cast_fp16 = slice_by_index(begin = var_17543_begin_0, end = var_17543_end_0, end_mask = var_17543_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17543_cast_fp16")]; + tensor var_17550_begin_0 = const()[name = string("op_17550_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17550_end_0 = const()[name = string("op_17550_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17550_end_mask_0 = const()[name = string("op_17550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17550_cast_fp16 = slice_by_index(begin = var_17550_begin_0, end = var_17550_end_0, end_mask = var_17550_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17550_cast_fp16")]; + tensor var_17557_begin_0 = const()[name = string("op_17557_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17557_end_0 = const()[name = string("op_17557_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17557_end_mask_0 = const()[name = string("op_17557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17557_cast_fp16 = slice_by_index(begin = var_17557_begin_0, end = var_17557_end_0, end_mask = var_17557_end_mask_0, x = var_17115_cast_fp16)[name = string("op_17557_cast_fp16")]; + tensor var_17564_begin_0 = const()[name = string("op_17564_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17564_end_0 = const()[name = string("op_17564_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17564_end_mask_0 = const()[name = string("op_17564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17564_cast_fp16 = slice_by_index(begin = var_17564_begin_0, end = var_17564_end_0, end_mask = var_17564_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17564_cast_fp16")]; + tensor var_17571_begin_0 = const()[name = string("op_17571_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17571_end_0 = const()[name = string("op_17571_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17571_end_mask_0 = const()[name = string("op_17571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17571_cast_fp16 = slice_by_index(begin = var_17571_begin_0, end = var_17571_end_0, end_mask = var_17571_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17571_cast_fp16")]; + tensor var_17578_begin_0 = const()[name = string("op_17578_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17578_end_0 = const()[name = string("op_17578_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17578_end_mask_0 = const()[name = string("op_17578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17578_cast_fp16 = slice_by_index(begin = var_17578_begin_0, end = var_17578_end_0, end_mask = var_17578_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17578_cast_fp16")]; + tensor var_17585_begin_0 = const()[name = string("op_17585_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17585_end_0 = const()[name = string("op_17585_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17585_end_mask_0 = const()[name = string("op_17585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17585_cast_fp16 = slice_by_index(begin = var_17585_begin_0, end = var_17585_end_0, end_mask = var_17585_end_mask_0, x = var_17119_cast_fp16)[name = string("op_17585_cast_fp16")]; + tensor var_17592_begin_0 = const()[name = string("op_17592_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17592_end_0 = const()[name = string("op_17592_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17592_end_mask_0 = const()[name = string("op_17592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17592_cast_fp16 = slice_by_index(begin = var_17592_begin_0, end = var_17592_end_0, end_mask = var_17592_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17592_cast_fp16")]; + tensor var_17599_begin_0 = const()[name = string("op_17599_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17599_end_0 = const()[name = string("op_17599_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17599_end_mask_0 = const()[name = string("op_17599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17599_cast_fp16 = slice_by_index(begin = var_17599_begin_0, end = var_17599_end_0, end_mask = var_17599_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17599_cast_fp16")]; + tensor var_17606_begin_0 = const()[name = string("op_17606_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17606_end_0 = const()[name = string("op_17606_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17606_end_mask_0 = const()[name = string("op_17606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17606_cast_fp16 = slice_by_index(begin = var_17606_begin_0, end = var_17606_end_0, end_mask = var_17606_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17606_cast_fp16")]; + tensor var_17613_begin_0 = const()[name = string("op_17613_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17613_end_0 = const()[name = string("op_17613_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17613_end_mask_0 = const()[name = string("op_17613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17613_cast_fp16 = slice_by_index(begin = var_17613_begin_0, end = var_17613_end_0, end_mask = var_17613_end_mask_0, x = var_17123_cast_fp16)[name = string("op_17613_cast_fp16")]; + tensor var_17620_begin_0 = const()[name = string("op_17620_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17620_end_0 = const()[name = string("op_17620_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17620_end_mask_0 = const()[name = string("op_17620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17620_cast_fp16 = slice_by_index(begin = var_17620_begin_0, end = var_17620_end_0, end_mask = var_17620_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17620_cast_fp16")]; + tensor var_17627_begin_0 = const()[name = string("op_17627_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17627_end_0 = const()[name = string("op_17627_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17627_end_mask_0 = const()[name = string("op_17627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17627_cast_fp16 = slice_by_index(begin = var_17627_begin_0, end = var_17627_end_0, end_mask = var_17627_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17627_cast_fp16")]; + tensor var_17634_begin_0 = const()[name = string("op_17634_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17634_end_0 = const()[name = string("op_17634_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17634_end_mask_0 = const()[name = string("op_17634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17634_cast_fp16 = slice_by_index(begin = var_17634_begin_0, end = var_17634_end_0, end_mask = var_17634_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17634_cast_fp16")]; + tensor var_17641_begin_0 = const()[name = string("op_17641_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17641_end_0 = const()[name = string("op_17641_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17641_end_mask_0 = const()[name = string("op_17641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17641_cast_fp16 = slice_by_index(begin = var_17641_begin_0, end = var_17641_end_0, end_mask = var_17641_end_mask_0, x = var_17127_cast_fp16)[name = string("op_17641_cast_fp16")]; + tensor var_17648_begin_0 = const()[name = string("op_17648_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17648_end_0 = const()[name = string("op_17648_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17648_end_mask_0 = const()[name = string("op_17648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17648_cast_fp16 = slice_by_index(begin = var_17648_begin_0, end = var_17648_end_0, end_mask = var_17648_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17648_cast_fp16")]; + tensor var_17655_begin_0 = const()[name = string("op_17655_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17655_end_0 = const()[name = string("op_17655_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17655_end_mask_0 = const()[name = string("op_17655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17655_cast_fp16 = slice_by_index(begin = var_17655_begin_0, end = var_17655_end_0, end_mask = var_17655_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17655_cast_fp16")]; + tensor var_17662_begin_0 = const()[name = string("op_17662_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17662_end_0 = const()[name = string("op_17662_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17662_end_mask_0 = const()[name = string("op_17662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17662_cast_fp16 = slice_by_index(begin = var_17662_begin_0, end = var_17662_end_0, end_mask = var_17662_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17662_cast_fp16")]; + tensor var_17669_begin_0 = const()[name = string("op_17669_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17669_end_0 = const()[name = string("op_17669_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17669_end_mask_0 = const()[name = string("op_17669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17669_cast_fp16 = slice_by_index(begin = var_17669_begin_0, end = var_17669_end_0, end_mask = var_17669_end_mask_0, x = var_17131_cast_fp16)[name = string("op_17669_cast_fp16")]; + tensor var_17676_begin_0 = const()[name = string("op_17676_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17676_end_0 = const()[name = string("op_17676_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17676_end_mask_0 = const()[name = string("op_17676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17676_cast_fp16 = slice_by_index(begin = var_17676_begin_0, end = var_17676_end_0, end_mask = var_17676_end_mask_0, x = var_17135_cast_fp16)[name = string("op_17676_cast_fp16")]; + tensor var_17683_begin_0 = const()[name = string("op_17683_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17683_end_0 = const()[name = string("op_17683_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17683_end_mask_0 = const()[name = string("op_17683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17683_cast_fp16 = slice_by_index(begin = var_17683_begin_0, end = var_17683_end_0, end_mask = var_17683_end_mask_0, x = var_17135_cast_fp16)[name = string("op_17683_cast_fp16")]; + tensor var_17690_begin_0 = const()[name = string("op_17690_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17690_end_0 = const()[name = string("op_17690_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17690_end_mask_0 = const()[name = string("op_17690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17690_cast_fp16 = slice_by_index(begin = var_17690_begin_0, end = var_17690_end_0, end_mask = var_17690_end_mask_0, x = var_17135_cast_fp16)[name = string("op_17690_cast_fp16")]; + tensor var_17697_begin_0 = const()[name = string("op_17697_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17697_end_0 = const()[name = string("op_17697_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17697_end_mask_0 = const()[name = string("op_17697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17697_cast_fp16 = slice_by_index(begin = var_17697_begin_0, end = var_17697_end_0, end_mask = var_17697_end_mask_0, x = var_17135_cast_fp16)[name = string("op_17697_cast_fp16")]; + tensor k_23_perm_0 = const()[name = string("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17702_begin_0 = const()[name = string("op_17702_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17702_end_0 = const()[name = string("op_17702_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_17702_end_mask_0 = const()[name = string("op_17702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = key_23_cast_fp16)[name = string("transpose_20")]; + tensor var_17702_cast_fp16 = slice_by_index(begin = var_17702_begin_0, end = var_17702_end_0, end_mask = var_17702_end_mask_0, x = k_23_cast_fp16)[name = string("op_17702_cast_fp16")]; + tensor var_17706_begin_0 = const()[name = string("op_17706_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17706_end_0 = const()[name = string("op_17706_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_17706_end_mask_0 = const()[name = string("op_17706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17706_cast_fp16 = slice_by_index(begin = var_17706_begin_0, end = var_17706_end_0, end_mask = var_17706_end_mask_0, x = k_23_cast_fp16)[name = string("op_17706_cast_fp16")]; + tensor var_17710_begin_0 = const()[name = string("op_17710_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17710_end_0 = const()[name = string("op_17710_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_17710_end_mask_0 = const()[name = string("op_17710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17710_cast_fp16 = slice_by_index(begin = var_17710_begin_0, end = var_17710_end_0, end_mask = var_17710_end_mask_0, x = k_23_cast_fp16)[name = string("op_17710_cast_fp16")]; + tensor var_17714_begin_0 = const()[name = string("op_17714_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17714_end_0 = const()[name = string("op_17714_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_17714_end_mask_0 = const()[name = string("op_17714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17714_cast_fp16 = slice_by_index(begin = var_17714_begin_0, end = var_17714_end_0, end_mask = var_17714_end_mask_0, x = k_23_cast_fp16)[name = string("op_17714_cast_fp16")]; + tensor var_17718_begin_0 = const()[name = string("op_17718_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17718_end_0 = const()[name = string("op_17718_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_17718_end_mask_0 = const()[name = string("op_17718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17718_cast_fp16 = slice_by_index(begin = var_17718_begin_0, end = var_17718_end_0, end_mask = var_17718_end_mask_0, x = k_23_cast_fp16)[name = string("op_17718_cast_fp16")]; + tensor var_17722_begin_0 = const()[name = string("op_17722_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17722_end_0 = const()[name = string("op_17722_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_17722_end_mask_0 = const()[name = string("op_17722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17722_cast_fp16 = slice_by_index(begin = var_17722_begin_0, end = var_17722_end_0, end_mask = var_17722_end_mask_0, x = k_23_cast_fp16)[name = string("op_17722_cast_fp16")]; + tensor var_17726_begin_0 = const()[name = string("op_17726_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17726_end_0 = const()[name = string("op_17726_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_17726_end_mask_0 = const()[name = string("op_17726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17726_cast_fp16 = slice_by_index(begin = var_17726_begin_0, end = var_17726_end_0, end_mask = var_17726_end_mask_0, x = k_23_cast_fp16)[name = string("op_17726_cast_fp16")]; + tensor var_17730_begin_0 = const()[name = string("op_17730_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17730_end_0 = const()[name = string("op_17730_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_17730_end_mask_0 = const()[name = string("op_17730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17730_cast_fp16 = slice_by_index(begin = var_17730_begin_0, end = var_17730_end_0, end_mask = var_17730_end_mask_0, x = k_23_cast_fp16)[name = string("op_17730_cast_fp16")]; + tensor var_17734_begin_0 = const()[name = string("op_17734_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17734_end_0 = const()[name = string("op_17734_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_17734_end_mask_0 = const()[name = string("op_17734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17734_cast_fp16 = slice_by_index(begin = var_17734_begin_0, end = var_17734_end_0, end_mask = var_17734_end_mask_0, x = k_23_cast_fp16)[name = string("op_17734_cast_fp16")]; + tensor var_17738_begin_0 = const()[name = string("op_17738_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17738_end_0 = const()[name = string("op_17738_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_17738_end_mask_0 = const()[name = string("op_17738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17738_cast_fp16 = slice_by_index(begin = var_17738_begin_0, end = var_17738_end_0, end_mask = var_17738_end_mask_0, x = k_23_cast_fp16)[name = string("op_17738_cast_fp16")]; + tensor var_17742_begin_0 = const()[name = string("op_17742_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17742_end_0 = const()[name = string("op_17742_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_17742_end_mask_0 = const()[name = string("op_17742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17742_cast_fp16 = slice_by_index(begin = var_17742_begin_0, end = var_17742_end_0, end_mask = var_17742_end_mask_0, x = k_23_cast_fp16)[name = string("op_17742_cast_fp16")]; + tensor var_17746_begin_0 = const()[name = string("op_17746_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17746_end_0 = const()[name = string("op_17746_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_17746_end_mask_0 = const()[name = string("op_17746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17746_cast_fp16 = slice_by_index(begin = var_17746_begin_0, end = var_17746_end_0, end_mask = var_17746_end_mask_0, x = k_23_cast_fp16)[name = string("op_17746_cast_fp16")]; + tensor var_17750_begin_0 = const()[name = string("op_17750_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17750_end_0 = const()[name = string("op_17750_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_17750_end_mask_0 = const()[name = string("op_17750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17750_cast_fp16 = slice_by_index(begin = var_17750_begin_0, end = var_17750_end_0, end_mask = var_17750_end_mask_0, x = k_23_cast_fp16)[name = string("op_17750_cast_fp16")]; + tensor var_17754_begin_0 = const()[name = string("op_17754_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17754_end_0 = const()[name = string("op_17754_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_17754_end_mask_0 = const()[name = string("op_17754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17754_cast_fp16 = slice_by_index(begin = var_17754_begin_0, end = var_17754_end_0, end_mask = var_17754_end_mask_0, x = k_23_cast_fp16)[name = string("op_17754_cast_fp16")]; + tensor var_17758_begin_0 = const()[name = string("op_17758_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17758_end_0 = const()[name = string("op_17758_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_17758_end_mask_0 = const()[name = string("op_17758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17758_cast_fp16 = slice_by_index(begin = var_17758_begin_0, end = var_17758_end_0, end_mask = var_17758_end_mask_0, x = k_23_cast_fp16)[name = string("op_17758_cast_fp16")]; + tensor var_17762_begin_0 = const()[name = string("op_17762_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17762_end_0 = const()[name = string("op_17762_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_17762_end_mask_0 = const()[name = string("op_17762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17762_cast_fp16 = slice_by_index(begin = var_17762_begin_0, end = var_17762_end_0, end_mask = var_17762_end_mask_0, x = k_23_cast_fp16)[name = string("op_17762_cast_fp16")]; + tensor var_17766_begin_0 = const()[name = string("op_17766_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17766_end_0 = const()[name = string("op_17766_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_17766_end_mask_0 = const()[name = string("op_17766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17766_cast_fp16 = slice_by_index(begin = var_17766_begin_0, end = var_17766_end_0, end_mask = var_17766_end_mask_0, x = k_23_cast_fp16)[name = string("op_17766_cast_fp16")]; + tensor var_17770_begin_0 = const()[name = string("op_17770_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17770_end_0 = const()[name = string("op_17770_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_17770_end_mask_0 = const()[name = string("op_17770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17770_cast_fp16 = slice_by_index(begin = var_17770_begin_0, end = var_17770_end_0, end_mask = var_17770_end_mask_0, x = k_23_cast_fp16)[name = string("op_17770_cast_fp16")]; + tensor var_17774_begin_0 = const()[name = string("op_17774_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17774_end_0 = const()[name = string("op_17774_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_17774_end_mask_0 = const()[name = string("op_17774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17774_cast_fp16 = slice_by_index(begin = var_17774_begin_0, end = var_17774_end_0, end_mask = var_17774_end_mask_0, x = k_23_cast_fp16)[name = string("op_17774_cast_fp16")]; + tensor var_17778_begin_0 = const()[name = string("op_17778_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17778_end_0 = const()[name = string("op_17778_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_17778_end_mask_0 = const()[name = string("op_17778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17778_cast_fp16 = slice_by_index(begin = var_17778_begin_0, end = var_17778_end_0, end_mask = var_17778_end_mask_0, x = k_23_cast_fp16)[name = string("op_17778_cast_fp16")]; + tensor var_17780_begin_0 = const()[name = string("op_17780_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17780_end_0 = const()[name = string("op_17780_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17780_end_mask_0 = const()[name = string("op_17780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17780_cast_fp16 = slice_by_index(begin = var_17780_begin_0, end = var_17780_end_0, end_mask = var_17780_end_mask_0, x = value_23_cast_fp16)[name = string("op_17780_cast_fp16")]; + tensor var_17784_begin_0 = const()[name = string("op_17784_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17784_end_0 = const()[name = string("op_17784_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_17784_end_mask_0 = const()[name = string("op_17784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17784_cast_fp16 = slice_by_index(begin = var_17784_begin_0, end = var_17784_end_0, end_mask = var_17784_end_mask_0, x = value_23_cast_fp16)[name = string("op_17784_cast_fp16")]; + tensor var_17788_begin_0 = const()[name = string("op_17788_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17788_end_0 = const()[name = string("op_17788_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_17788_end_mask_0 = const()[name = string("op_17788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17788_cast_fp16 = slice_by_index(begin = var_17788_begin_0, end = var_17788_end_0, end_mask = var_17788_end_mask_0, x = value_23_cast_fp16)[name = string("op_17788_cast_fp16")]; + tensor var_17792_begin_0 = const()[name = string("op_17792_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17792_end_0 = const()[name = string("op_17792_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_17792_end_mask_0 = const()[name = string("op_17792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17792_cast_fp16 = slice_by_index(begin = var_17792_begin_0, end = var_17792_end_0, end_mask = var_17792_end_mask_0, x = value_23_cast_fp16)[name = string("op_17792_cast_fp16")]; + tensor var_17796_begin_0 = const()[name = string("op_17796_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17796_end_0 = const()[name = string("op_17796_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_17796_end_mask_0 = const()[name = string("op_17796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17796_cast_fp16 = slice_by_index(begin = var_17796_begin_0, end = var_17796_end_0, end_mask = var_17796_end_mask_0, x = value_23_cast_fp16)[name = string("op_17796_cast_fp16")]; + tensor var_17800_begin_0 = const()[name = string("op_17800_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17800_end_0 = const()[name = string("op_17800_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_17800_end_mask_0 = const()[name = string("op_17800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17800_cast_fp16 = slice_by_index(begin = var_17800_begin_0, end = var_17800_end_0, end_mask = var_17800_end_mask_0, x = value_23_cast_fp16)[name = string("op_17800_cast_fp16")]; + tensor var_17804_begin_0 = const()[name = string("op_17804_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17804_end_0 = const()[name = string("op_17804_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_17804_end_mask_0 = const()[name = string("op_17804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17804_cast_fp16 = slice_by_index(begin = var_17804_begin_0, end = var_17804_end_0, end_mask = var_17804_end_mask_0, x = value_23_cast_fp16)[name = string("op_17804_cast_fp16")]; + tensor var_17808_begin_0 = const()[name = string("op_17808_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17808_end_0 = const()[name = string("op_17808_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_17808_end_mask_0 = const()[name = string("op_17808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17808_cast_fp16 = slice_by_index(begin = var_17808_begin_0, end = var_17808_end_0, end_mask = var_17808_end_mask_0, x = value_23_cast_fp16)[name = string("op_17808_cast_fp16")]; + tensor var_17812_begin_0 = const()[name = string("op_17812_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17812_end_0 = const()[name = string("op_17812_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_17812_end_mask_0 = const()[name = string("op_17812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17812_cast_fp16 = slice_by_index(begin = var_17812_begin_0, end = var_17812_end_0, end_mask = var_17812_end_mask_0, x = value_23_cast_fp16)[name = string("op_17812_cast_fp16")]; + tensor var_17816_begin_0 = const()[name = string("op_17816_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17816_end_0 = const()[name = string("op_17816_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_17816_end_mask_0 = const()[name = string("op_17816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17816_cast_fp16 = slice_by_index(begin = var_17816_begin_0, end = var_17816_end_0, end_mask = var_17816_end_mask_0, x = value_23_cast_fp16)[name = string("op_17816_cast_fp16")]; + tensor var_17820_begin_0 = const()[name = string("op_17820_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17820_end_0 = const()[name = string("op_17820_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_17820_end_mask_0 = const()[name = string("op_17820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17820_cast_fp16 = slice_by_index(begin = var_17820_begin_0, end = var_17820_end_0, end_mask = var_17820_end_mask_0, x = value_23_cast_fp16)[name = string("op_17820_cast_fp16")]; + tensor var_17824_begin_0 = const()[name = string("op_17824_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17824_end_0 = const()[name = string("op_17824_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_17824_end_mask_0 = const()[name = string("op_17824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17824_cast_fp16 = slice_by_index(begin = var_17824_begin_0, end = var_17824_end_0, end_mask = var_17824_end_mask_0, x = value_23_cast_fp16)[name = string("op_17824_cast_fp16")]; + tensor var_17828_begin_0 = const()[name = string("op_17828_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17828_end_0 = const()[name = string("op_17828_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_17828_end_mask_0 = const()[name = string("op_17828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17828_cast_fp16 = slice_by_index(begin = var_17828_begin_0, end = var_17828_end_0, end_mask = var_17828_end_mask_0, x = value_23_cast_fp16)[name = string("op_17828_cast_fp16")]; + tensor var_17832_begin_0 = const()[name = string("op_17832_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17832_end_0 = const()[name = string("op_17832_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_17832_end_mask_0 = const()[name = string("op_17832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17832_cast_fp16 = slice_by_index(begin = var_17832_begin_0, end = var_17832_end_0, end_mask = var_17832_end_mask_0, x = value_23_cast_fp16)[name = string("op_17832_cast_fp16")]; + tensor var_17836_begin_0 = const()[name = string("op_17836_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17836_end_0 = const()[name = string("op_17836_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_17836_end_mask_0 = const()[name = string("op_17836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17836_cast_fp16 = slice_by_index(begin = var_17836_begin_0, end = var_17836_end_0, end_mask = var_17836_end_mask_0, x = value_23_cast_fp16)[name = string("op_17836_cast_fp16")]; + tensor var_17840_begin_0 = const()[name = string("op_17840_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17840_end_0 = const()[name = string("op_17840_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_17840_end_mask_0 = const()[name = string("op_17840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17840_cast_fp16 = slice_by_index(begin = var_17840_begin_0, end = var_17840_end_0, end_mask = var_17840_end_mask_0, x = value_23_cast_fp16)[name = string("op_17840_cast_fp16")]; + tensor var_17844_begin_0 = const()[name = string("op_17844_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17844_end_0 = const()[name = string("op_17844_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_17844_end_mask_0 = const()[name = string("op_17844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17844_cast_fp16 = slice_by_index(begin = var_17844_begin_0, end = var_17844_end_0, end_mask = var_17844_end_mask_0, x = value_23_cast_fp16)[name = string("op_17844_cast_fp16")]; + tensor var_17848_begin_0 = const()[name = string("op_17848_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17848_end_0 = const()[name = string("op_17848_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_17848_end_mask_0 = const()[name = string("op_17848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17848_cast_fp16 = slice_by_index(begin = var_17848_begin_0, end = var_17848_end_0, end_mask = var_17848_end_mask_0, x = value_23_cast_fp16)[name = string("op_17848_cast_fp16")]; + tensor var_17852_begin_0 = const()[name = string("op_17852_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17852_end_0 = const()[name = string("op_17852_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_17852_end_mask_0 = const()[name = string("op_17852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17852_cast_fp16 = slice_by_index(begin = var_17852_begin_0, end = var_17852_end_0, end_mask = var_17852_end_mask_0, x = value_23_cast_fp16)[name = string("op_17852_cast_fp16")]; + tensor var_17856_begin_0 = const()[name = string("op_17856_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17856_end_0 = const()[name = string("op_17856_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_17856_end_mask_0 = const()[name = string("op_17856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17856_cast_fp16 = slice_by_index(begin = var_17856_begin_0, end = var_17856_end_0, end_mask = var_17856_end_mask_0, x = value_23_cast_fp16)[name = string("op_17856_cast_fp16")]; + string _SplitHeadsQ__mh_w_1761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1761_equation_0, values = (var_17702_cast_fp16, var_17144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1761_cast_fp16")]; + string _SplitHeadsQ__mh_w_1763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1763_equation_0, values = (var_17702_cast_fp16, var_17151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1763_cast_fp16")]; + string _SplitHeadsQ__mh_w_1765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1765_equation_0, values = (var_17702_cast_fp16, var_17158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1765_cast_fp16")]; + string _SplitHeadsQ__mh_w_1767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1767_equation_0, values = (var_17702_cast_fp16, var_17165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1767_cast_fp16")]; + string _SplitHeadsQ__mh_w_1769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1769_equation_0, values = (var_17706_cast_fp16, var_17172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1769_cast_fp16")]; + string _SplitHeadsQ__mh_w_1771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1771_equation_0, values = (var_17706_cast_fp16, var_17179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1771_cast_fp16")]; + string _SplitHeadsQ__mh_w_1773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1773_equation_0, values = (var_17706_cast_fp16, var_17186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1773_cast_fp16")]; + string _SplitHeadsQ__mh_w_1775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1775_equation_0, values = (var_17706_cast_fp16, var_17193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1775_cast_fp16")]; + string _SplitHeadsQ__mh_w_1777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1777_equation_0, values = (var_17710_cast_fp16, var_17200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1777_cast_fp16")]; + string _SplitHeadsQ__mh_w_1779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1779_equation_0, values = (var_17710_cast_fp16, var_17207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1779_cast_fp16")]; + string _SplitHeadsQ__mh_w_1781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1781_equation_0, values = (var_17710_cast_fp16, var_17214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1781_cast_fp16")]; + string _SplitHeadsQ__mh_w_1783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1783_equation_0, values = (var_17710_cast_fp16, var_17221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1783_cast_fp16")]; + string _SplitHeadsQ__mh_w_1785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1785_equation_0, values = (var_17714_cast_fp16, var_17228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1785_cast_fp16")]; + string _SplitHeadsQ__mh_w_1787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1787_equation_0, values = (var_17714_cast_fp16, var_17235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1787_cast_fp16")]; + string _SplitHeadsQ__mh_w_1789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1789_equation_0, values = (var_17714_cast_fp16, var_17242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1789_cast_fp16")]; + string _SplitHeadsQ__mh_w_1791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1791_equation_0, values = (var_17714_cast_fp16, var_17249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1791_cast_fp16")]; + string _SplitHeadsQ__mh_w_1793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1793_equation_0, values = (var_17718_cast_fp16, var_17256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1793_cast_fp16")]; + string _SplitHeadsQ__mh_w_1795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1795_equation_0, values = (var_17718_cast_fp16, var_17263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1795_cast_fp16")]; + string _SplitHeadsQ__mh_w_1797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1797_equation_0, values = (var_17718_cast_fp16, var_17270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1797_cast_fp16")]; + string _SplitHeadsQ__mh_w_1799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1799_equation_0, values = (var_17718_cast_fp16, var_17277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1799_cast_fp16")]; + string _SplitHeadsQ__mh_w_1801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1801_equation_0, values = (var_17722_cast_fp16, var_17284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1801_cast_fp16")]; + string _SplitHeadsQ__mh_w_1803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1803_equation_0, values = (var_17722_cast_fp16, var_17291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1803_cast_fp16")]; + string _SplitHeadsQ__mh_w_1805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1805_equation_0, values = (var_17722_cast_fp16, var_17298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1805_cast_fp16")]; + string _SplitHeadsQ__mh_w_1807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1807_equation_0, values = (var_17722_cast_fp16, var_17305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1807_cast_fp16")]; + string _SplitHeadsQ__mh_w_1809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1809_equation_0, values = (var_17726_cast_fp16, var_17312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1809_cast_fp16")]; + string _SplitHeadsQ__mh_w_1811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1811_equation_0, values = (var_17726_cast_fp16, var_17319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1811_cast_fp16")]; + string _SplitHeadsQ__mh_w_1813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1813_equation_0, values = (var_17726_cast_fp16, var_17326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1813_cast_fp16")]; + string _SplitHeadsQ__mh_w_1815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1815_equation_0, values = (var_17726_cast_fp16, var_17333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1815_cast_fp16")]; + string _SplitHeadsQ__mh_w_1817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1817_equation_0, values = (var_17730_cast_fp16, var_17340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1817_cast_fp16")]; + string _SplitHeadsQ__mh_w_1819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1819_equation_0, values = (var_17730_cast_fp16, var_17347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1819_cast_fp16")]; + string _SplitHeadsQ__mh_w_1821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1821_equation_0, values = (var_17730_cast_fp16, var_17354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1821_cast_fp16")]; + string _SplitHeadsQ__mh_w_1823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1823_equation_0, values = (var_17730_cast_fp16, var_17361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1823_cast_fp16")]; + string _SplitHeadsQ__mh_w_1825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1825_equation_0, values = (var_17734_cast_fp16, var_17368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1825_cast_fp16")]; + string _SplitHeadsQ__mh_w_1827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1827_equation_0, values = (var_17734_cast_fp16, var_17375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1827_cast_fp16")]; + string _SplitHeadsQ__mh_w_1829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1829_equation_0, values = (var_17734_cast_fp16, var_17382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1829_cast_fp16")]; + string _SplitHeadsQ__mh_w_1831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1831_equation_0, values = (var_17734_cast_fp16, var_17389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1831_cast_fp16")]; + string _SplitHeadsQ__mh_w_1833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1833_equation_0, values = (var_17738_cast_fp16, var_17396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1833_cast_fp16")]; + string _SplitHeadsQ__mh_w_1835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1835_equation_0, values = (var_17738_cast_fp16, var_17403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1835_cast_fp16")]; + string _SplitHeadsQ__mh_w_1837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1837_equation_0, values = (var_17738_cast_fp16, var_17410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1837_cast_fp16")]; + string _SplitHeadsQ__mh_w_1839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1839_equation_0, values = (var_17738_cast_fp16, var_17417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1839_cast_fp16")]; + string _SplitHeadsQ__mh_w_1841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1841_equation_0, values = (var_17742_cast_fp16, var_17424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1841_cast_fp16")]; + string _SplitHeadsQ__mh_w_1843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1843_equation_0, values = (var_17742_cast_fp16, var_17431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1843_cast_fp16")]; + string _SplitHeadsQ__mh_w_1845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1845_equation_0, values = (var_17742_cast_fp16, var_17438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1845_cast_fp16")]; + string _SplitHeadsQ__mh_w_1847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1847_equation_0, values = (var_17742_cast_fp16, var_17445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1847_cast_fp16")]; + string _SplitHeadsQ__mh_w_1849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1849_equation_0, values = (var_17746_cast_fp16, var_17452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1849_cast_fp16")]; + string _SplitHeadsQ__mh_w_1851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1851_equation_0, values = (var_17746_cast_fp16, var_17459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1851_cast_fp16")]; + string _SplitHeadsQ__mh_w_1853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1853_equation_0, values = (var_17746_cast_fp16, var_17466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1853_cast_fp16")]; + string _SplitHeadsQ__mh_w_1855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1855_equation_0, values = (var_17746_cast_fp16, var_17473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1855_cast_fp16")]; + string _SplitHeadsQ__mh_w_1857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1857_equation_0, values = (var_17750_cast_fp16, var_17480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1857_cast_fp16")]; + string _SplitHeadsQ__mh_w_1859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1859_equation_0, values = (var_17750_cast_fp16, var_17487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1859_cast_fp16")]; + string _SplitHeadsQ__mh_w_1861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1861_equation_0, values = (var_17750_cast_fp16, var_17494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1861_cast_fp16")]; + string _SplitHeadsQ__mh_w_1863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1863_equation_0, values = (var_17750_cast_fp16, var_17501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1863_cast_fp16")]; + string _SplitHeadsQ__mh_w_1865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1865_equation_0, values = (var_17754_cast_fp16, var_17508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1865_cast_fp16")]; + string _SplitHeadsQ__mh_w_1867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1867_equation_0, values = (var_17754_cast_fp16, var_17515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1867_cast_fp16")]; + string _SplitHeadsQ__mh_w_1869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1869_equation_0, values = (var_17754_cast_fp16, var_17522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1869_cast_fp16")]; + string _SplitHeadsQ__mh_w_1871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1871_equation_0, values = (var_17754_cast_fp16, var_17529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1871_cast_fp16")]; + string _SplitHeadsQ__mh_w_1873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1873_equation_0, values = (var_17758_cast_fp16, var_17536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1873_cast_fp16")]; + string _SplitHeadsQ__mh_w_1875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1875_equation_0, values = (var_17758_cast_fp16, var_17543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1875_cast_fp16")]; + string _SplitHeadsQ__mh_w_1877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1877_equation_0, values = (var_17758_cast_fp16, var_17550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1877_cast_fp16")]; + string _SplitHeadsQ__mh_w_1879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1879_equation_0, values = (var_17758_cast_fp16, var_17557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1879_cast_fp16")]; + string _SplitHeadsQ__mh_w_1881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1881_equation_0, values = (var_17762_cast_fp16, var_17564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1881_cast_fp16")]; + string _SplitHeadsQ__mh_w_1883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1883_equation_0, values = (var_17762_cast_fp16, var_17571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1883_cast_fp16")]; + string _SplitHeadsQ__mh_w_1885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1885_equation_0, values = (var_17762_cast_fp16, var_17578_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1885_cast_fp16")]; + string _SplitHeadsQ__mh_w_1887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1887_equation_0, values = (var_17762_cast_fp16, var_17585_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1887_cast_fp16")]; + string _SplitHeadsQ__mh_w_1889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1889_equation_0, values = (var_17766_cast_fp16, var_17592_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1889_cast_fp16")]; + string _SplitHeadsQ__mh_w_1891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1891_equation_0, values = (var_17766_cast_fp16, var_17599_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1891_cast_fp16")]; + string _SplitHeadsQ__mh_w_1893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1893_equation_0, values = (var_17766_cast_fp16, var_17606_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1893_cast_fp16")]; + string _SplitHeadsQ__mh_w_1895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1895_equation_0, values = (var_17766_cast_fp16, var_17613_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1895_cast_fp16")]; + string _SplitHeadsQ__mh_w_1897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1897_equation_0, values = (var_17770_cast_fp16, var_17620_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1897_cast_fp16")]; + string _SplitHeadsQ__mh_w_1899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1899_equation_0, values = (var_17770_cast_fp16, var_17627_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1899_cast_fp16")]; + string _SplitHeadsQ__mh_w_1901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1901_equation_0, values = (var_17770_cast_fp16, var_17634_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1901_cast_fp16")]; + string _SplitHeadsQ__mh_w_1903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1903_equation_0, values = (var_17770_cast_fp16, var_17641_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1903_cast_fp16")]; + string _SplitHeadsQ__mh_w_1905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1905_equation_0, values = (var_17774_cast_fp16, var_17648_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1905_cast_fp16")]; + string _SplitHeadsQ__mh_w_1907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1907_equation_0, values = (var_17774_cast_fp16, var_17655_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1907_cast_fp16")]; + string _SplitHeadsQ__mh_w_1909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1909_equation_0, values = (var_17774_cast_fp16, var_17662_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1909_cast_fp16")]; + string _SplitHeadsQ__mh_w_1911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1911_equation_0, values = (var_17774_cast_fp16, var_17669_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1911_cast_fp16")]; + string _SplitHeadsQ__mh_w_1913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1913_equation_0, values = (var_17778_cast_fp16, var_17676_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1913_cast_fp16")]; + string _SplitHeadsQ__mh_w_1915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1915_equation_0, values = (var_17778_cast_fp16, var_17683_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1915_cast_fp16")]; + string _SplitHeadsQ__mh_w_1917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1917_equation_0, values = (var_17778_cast_fp16, var_17690_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1917_cast_fp16")]; + string _SplitHeadsQ__mh_w_1919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1919_equation_0, values = (var_17778_cast_fp16, var_17697_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1919_cast_fp16")]; + fp16 var_18019_to_fp16 = const()[name = string("op_18019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1761_cast_fp16, y = var_18019_to_fp16)[name = string("aw_chunk_1761_cast_fp16")]; + fp16 var_18021_to_fp16 = const()[name = string("op_18021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1763_cast_fp16, y = var_18021_to_fp16)[name = string("aw_chunk_1763_cast_fp16")]; + fp16 var_18023_to_fp16 = const()[name = string("op_18023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1765_cast_fp16, y = var_18023_to_fp16)[name = string("aw_chunk_1765_cast_fp16")]; + fp16 var_18025_to_fp16 = const()[name = string("op_18025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1767_cast_fp16, y = var_18025_to_fp16)[name = string("aw_chunk_1767_cast_fp16")]; + fp16 var_18027_to_fp16 = const()[name = string("op_18027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1769_cast_fp16, y = var_18027_to_fp16)[name = string("aw_chunk_1769_cast_fp16")]; + fp16 var_18029_to_fp16 = const()[name = string("op_18029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1771_cast_fp16, y = var_18029_to_fp16)[name = string("aw_chunk_1771_cast_fp16")]; + fp16 var_18031_to_fp16 = const()[name = string("op_18031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1773_cast_fp16, y = var_18031_to_fp16)[name = string("aw_chunk_1773_cast_fp16")]; + fp16 var_18033_to_fp16 = const()[name = string("op_18033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1775_cast_fp16, y = var_18033_to_fp16)[name = string("aw_chunk_1775_cast_fp16")]; + fp16 var_18035_to_fp16 = const()[name = string("op_18035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1777_cast_fp16, y = var_18035_to_fp16)[name = string("aw_chunk_1777_cast_fp16")]; + fp16 var_18037_to_fp16 = const()[name = string("op_18037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1779_cast_fp16, y = var_18037_to_fp16)[name = string("aw_chunk_1779_cast_fp16")]; + fp16 var_18039_to_fp16 = const()[name = string("op_18039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1781_cast_fp16, y = var_18039_to_fp16)[name = string("aw_chunk_1781_cast_fp16")]; + fp16 var_18041_to_fp16 = const()[name = string("op_18041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1783_cast_fp16, y = var_18041_to_fp16)[name = string("aw_chunk_1783_cast_fp16")]; + fp16 var_18043_to_fp16 = const()[name = string("op_18043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1785_cast_fp16, y = var_18043_to_fp16)[name = string("aw_chunk_1785_cast_fp16")]; + fp16 var_18045_to_fp16 = const()[name = string("op_18045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1787_cast_fp16, y = var_18045_to_fp16)[name = string("aw_chunk_1787_cast_fp16")]; + fp16 var_18047_to_fp16 = const()[name = string("op_18047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1789_cast_fp16, y = var_18047_to_fp16)[name = string("aw_chunk_1789_cast_fp16")]; + fp16 var_18049_to_fp16 = const()[name = string("op_18049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1791_cast_fp16, y = var_18049_to_fp16)[name = string("aw_chunk_1791_cast_fp16")]; + fp16 var_18051_to_fp16 = const()[name = string("op_18051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1793_cast_fp16, y = var_18051_to_fp16)[name = string("aw_chunk_1793_cast_fp16")]; + fp16 var_18053_to_fp16 = const()[name = string("op_18053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1795_cast_fp16, y = var_18053_to_fp16)[name = string("aw_chunk_1795_cast_fp16")]; + fp16 var_18055_to_fp16 = const()[name = string("op_18055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1797_cast_fp16, y = var_18055_to_fp16)[name = string("aw_chunk_1797_cast_fp16")]; + fp16 var_18057_to_fp16 = const()[name = string("op_18057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1799_cast_fp16, y = var_18057_to_fp16)[name = string("aw_chunk_1799_cast_fp16")]; + fp16 var_18059_to_fp16 = const()[name = string("op_18059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1801_cast_fp16, y = var_18059_to_fp16)[name = string("aw_chunk_1801_cast_fp16")]; + fp16 var_18061_to_fp16 = const()[name = string("op_18061_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1803_cast_fp16, y = var_18061_to_fp16)[name = string("aw_chunk_1803_cast_fp16")]; + fp16 var_18063_to_fp16 = const()[name = string("op_18063_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1805_cast_fp16, y = var_18063_to_fp16)[name = string("aw_chunk_1805_cast_fp16")]; + fp16 var_18065_to_fp16 = const()[name = string("op_18065_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1807_cast_fp16, y = var_18065_to_fp16)[name = string("aw_chunk_1807_cast_fp16")]; + fp16 var_18067_to_fp16 = const()[name = string("op_18067_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1809_cast_fp16, y = var_18067_to_fp16)[name = string("aw_chunk_1809_cast_fp16")]; + fp16 var_18069_to_fp16 = const()[name = string("op_18069_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1811_cast_fp16, y = var_18069_to_fp16)[name = string("aw_chunk_1811_cast_fp16")]; + fp16 var_18071_to_fp16 = const()[name = string("op_18071_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1813_cast_fp16, y = var_18071_to_fp16)[name = string("aw_chunk_1813_cast_fp16")]; + fp16 var_18073_to_fp16 = const()[name = string("op_18073_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1815_cast_fp16, y = var_18073_to_fp16)[name = string("aw_chunk_1815_cast_fp16")]; + fp16 var_18075_to_fp16 = const()[name = string("op_18075_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1817_cast_fp16, y = var_18075_to_fp16)[name = string("aw_chunk_1817_cast_fp16")]; + fp16 var_18077_to_fp16 = const()[name = string("op_18077_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1819_cast_fp16, y = var_18077_to_fp16)[name = string("aw_chunk_1819_cast_fp16")]; + fp16 var_18079_to_fp16 = const()[name = string("op_18079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1821_cast_fp16, y = var_18079_to_fp16)[name = string("aw_chunk_1821_cast_fp16")]; + fp16 var_18081_to_fp16 = const()[name = string("op_18081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1823_cast_fp16, y = var_18081_to_fp16)[name = string("aw_chunk_1823_cast_fp16")]; + fp16 var_18083_to_fp16 = const()[name = string("op_18083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1825_cast_fp16, y = var_18083_to_fp16)[name = string("aw_chunk_1825_cast_fp16")]; + fp16 var_18085_to_fp16 = const()[name = string("op_18085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1827_cast_fp16, y = var_18085_to_fp16)[name = string("aw_chunk_1827_cast_fp16")]; + fp16 var_18087_to_fp16 = const()[name = string("op_18087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1829_cast_fp16, y = var_18087_to_fp16)[name = string("aw_chunk_1829_cast_fp16")]; + fp16 var_18089_to_fp16 = const()[name = string("op_18089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1831_cast_fp16, y = var_18089_to_fp16)[name = string("aw_chunk_1831_cast_fp16")]; + fp16 var_18091_to_fp16 = const()[name = string("op_18091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1833_cast_fp16, y = var_18091_to_fp16)[name = string("aw_chunk_1833_cast_fp16")]; + fp16 var_18093_to_fp16 = const()[name = string("op_18093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1835_cast_fp16, y = var_18093_to_fp16)[name = string("aw_chunk_1835_cast_fp16")]; + fp16 var_18095_to_fp16 = const()[name = string("op_18095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1837_cast_fp16, y = var_18095_to_fp16)[name = string("aw_chunk_1837_cast_fp16")]; + fp16 var_18097_to_fp16 = const()[name = string("op_18097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1839_cast_fp16, y = var_18097_to_fp16)[name = string("aw_chunk_1839_cast_fp16")]; + fp16 var_18099_to_fp16 = const()[name = string("op_18099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1841_cast_fp16, y = var_18099_to_fp16)[name = string("aw_chunk_1841_cast_fp16")]; + fp16 var_18101_to_fp16 = const()[name = string("op_18101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1843_cast_fp16, y = var_18101_to_fp16)[name = string("aw_chunk_1843_cast_fp16")]; + fp16 var_18103_to_fp16 = const()[name = string("op_18103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1845_cast_fp16, y = var_18103_to_fp16)[name = string("aw_chunk_1845_cast_fp16")]; + fp16 var_18105_to_fp16 = const()[name = string("op_18105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1847_cast_fp16, y = var_18105_to_fp16)[name = string("aw_chunk_1847_cast_fp16")]; + fp16 var_18107_to_fp16 = const()[name = string("op_18107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1849_cast_fp16, y = var_18107_to_fp16)[name = string("aw_chunk_1849_cast_fp16")]; + fp16 var_18109_to_fp16 = const()[name = string("op_18109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1851_cast_fp16, y = var_18109_to_fp16)[name = string("aw_chunk_1851_cast_fp16")]; + fp16 var_18111_to_fp16 = const()[name = string("op_18111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1853_cast_fp16, y = var_18111_to_fp16)[name = string("aw_chunk_1853_cast_fp16")]; + fp16 var_18113_to_fp16 = const()[name = string("op_18113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1855_cast_fp16, y = var_18113_to_fp16)[name = string("aw_chunk_1855_cast_fp16")]; + fp16 var_18115_to_fp16 = const()[name = string("op_18115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1857_cast_fp16, y = var_18115_to_fp16)[name = string("aw_chunk_1857_cast_fp16")]; + fp16 var_18117_to_fp16 = const()[name = string("op_18117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1859_cast_fp16, y = var_18117_to_fp16)[name = string("aw_chunk_1859_cast_fp16")]; + fp16 var_18119_to_fp16 = const()[name = string("op_18119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1861_cast_fp16, y = var_18119_to_fp16)[name = string("aw_chunk_1861_cast_fp16")]; + fp16 var_18121_to_fp16 = const()[name = string("op_18121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1863_cast_fp16, y = var_18121_to_fp16)[name = string("aw_chunk_1863_cast_fp16")]; + fp16 var_18123_to_fp16 = const()[name = string("op_18123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1865_cast_fp16, y = var_18123_to_fp16)[name = string("aw_chunk_1865_cast_fp16")]; + fp16 var_18125_to_fp16 = const()[name = string("op_18125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1867_cast_fp16, y = var_18125_to_fp16)[name = string("aw_chunk_1867_cast_fp16")]; + fp16 var_18127_to_fp16 = const()[name = string("op_18127_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1869_cast_fp16, y = var_18127_to_fp16)[name = string("aw_chunk_1869_cast_fp16")]; + fp16 var_18129_to_fp16 = const()[name = string("op_18129_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1871_cast_fp16, y = var_18129_to_fp16)[name = string("aw_chunk_1871_cast_fp16")]; + fp16 var_18131_to_fp16 = const()[name = string("op_18131_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1873_cast_fp16, y = var_18131_to_fp16)[name = string("aw_chunk_1873_cast_fp16")]; + fp16 var_18133_to_fp16 = const()[name = string("op_18133_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1875_cast_fp16, y = var_18133_to_fp16)[name = string("aw_chunk_1875_cast_fp16")]; + fp16 var_18135_to_fp16 = const()[name = string("op_18135_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1877_cast_fp16, y = var_18135_to_fp16)[name = string("aw_chunk_1877_cast_fp16")]; + fp16 var_18137_to_fp16 = const()[name = string("op_18137_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1879_cast_fp16, y = var_18137_to_fp16)[name = string("aw_chunk_1879_cast_fp16")]; + fp16 var_18139_to_fp16 = const()[name = string("op_18139_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1881_cast_fp16, y = var_18139_to_fp16)[name = string("aw_chunk_1881_cast_fp16")]; + fp16 var_18141_to_fp16 = const()[name = string("op_18141_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1883_cast_fp16, y = var_18141_to_fp16)[name = string("aw_chunk_1883_cast_fp16")]; + fp16 var_18143_to_fp16 = const()[name = string("op_18143_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1885_cast_fp16, y = var_18143_to_fp16)[name = string("aw_chunk_1885_cast_fp16")]; + fp16 var_18145_to_fp16 = const()[name = string("op_18145_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1887_cast_fp16, y = var_18145_to_fp16)[name = string("aw_chunk_1887_cast_fp16")]; + fp16 var_18147_to_fp16 = const()[name = string("op_18147_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1889_cast_fp16, y = var_18147_to_fp16)[name = string("aw_chunk_1889_cast_fp16")]; + fp16 var_18149_to_fp16 = const()[name = string("op_18149_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1891_cast_fp16, y = var_18149_to_fp16)[name = string("aw_chunk_1891_cast_fp16")]; + fp16 var_18151_to_fp16 = const()[name = string("op_18151_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1893_cast_fp16, y = var_18151_to_fp16)[name = string("aw_chunk_1893_cast_fp16")]; + fp16 var_18153_to_fp16 = const()[name = string("op_18153_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1895_cast_fp16, y = var_18153_to_fp16)[name = string("aw_chunk_1895_cast_fp16")]; + fp16 var_18155_to_fp16 = const()[name = string("op_18155_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1897_cast_fp16, y = var_18155_to_fp16)[name = string("aw_chunk_1897_cast_fp16")]; + fp16 var_18157_to_fp16 = const()[name = string("op_18157_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1899_cast_fp16, y = var_18157_to_fp16)[name = string("aw_chunk_1899_cast_fp16")]; + fp16 var_18159_to_fp16 = const()[name = string("op_18159_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1901_cast_fp16, y = var_18159_to_fp16)[name = string("aw_chunk_1901_cast_fp16")]; + fp16 var_18161_to_fp16 = const()[name = string("op_18161_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1903_cast_fp16, y = var_18161_to_fp16)[name = string("aw_chunk_1903_cast_fp16")]; + fp16 var_18163_to_fp16 = const()[name = string("op_18163_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1905_cast_fp16, y = var_18163_to_fp16)[name = string("aw_chunk_1905_cast_fp16")]; + fp16 var_18165_to_fp16 = const()[name = string("op_18165_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1907_cast_fp16, y = var_18165_to_fp16)[name = string("aw_chunk_1907_cast_fp16")]; + fp16 var_18167_to_fp16 = const()[name = string("op_18167_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1909_cast_fp16, y = var_18167_to_fp16)[name = string("aw_chunk_1909_cast_fp16")]; + fp16 var_18169_to_fp16 = const()[name = string("op_18169_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1911_cast_fp16, y = var_18169_to_fp16)[name = string("aw_chunk_1911_cast_fp16")]; + fp16 var_18171_to_fp16 = const()[name = string("op_18171_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1913_cast_fp16, y = var_18171_to_fp16)[name = string("aw_chunk_1913_cast_fp16")]; + fp16 var_18173_to_fp16 = const()[name = string("op_18173_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1915_cast_fp16, y = var_18173_to_fp16)[name = string("aw_chunk_1915_cast_fp16")]; + fp16 var_18175_to_fp16 = const()[name = string("op_18175_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1917_cast_fp16, y = var_18175_to_fp16)[name = string("aw_chunk_1917_cast_fp16")]; + fp16 var_18177_to_fp16 = const()[name = string("op_18177_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1919_cast_fp16, y = var_18177_to_fp16)[name = string("aw_chunk_1919_cast_fp16")]; + tensor var_18179_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1761_cast_fp16)[name = string("op_18179_cast_fp16")]; + tensor var_18180_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1763_cast_fp16)[name = string("op_18180_cast_fp16")]; + tensor var_18181_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1765_cast_fp16)[name = string("op_18181_cast_fp16")]; + tensor var_18182_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1767_cast_fp16)[name = string("op_18182_cast_fp16")]; + tensor var_18183_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1769_cast_fp16)[name = string("op_18183_cast_fp16")]; + tensor var_18184_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1771_cast_fp16)[name = string("op_18184_cast_fp16")]; + tensor var_18185_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1773_cast_fp16)[name = string("op_18185_cast_fp16")]; + tensor var_18186_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1775_cast_fp16)[name = string("op_18186_cast_fp16")]; + tensor var_18187_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1777_cast_fp16)[name = string("op_18187_cast_fp16")]; + tensor var_18188_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1779_cast_fp16)[name = string("op_18188_cast_fp16")]; + tensor var_18189_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1781_cast_fp16)[name = string("op_18189_cast_fp16")]; + tensor var_18190_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1783_cast_fp16)[name = string("op_18190_cast_fp16")]; + tensor var_18191_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1785_cast_fp16)[name = string("op_18191_cast_fp16")]; + tensor var_18192_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1787_cast_fp16)[name = string("op_18192_cast_fp16")]; + tensor var_18193_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1789_cast_fp16)[name = string("op_18193_cast_fp16")]; + tensor var_18194_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1791_cast_fp16)[name = string("op_18194_cast_fp16")]; + tensor var_18195_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1793_cast_fp16)[name = string("op_18195_cast_fp16")]; + tensor var_18196_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1795_cast_fp16)[name = string("op_18196_cast_fp16")]; + tensor var_18197_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1797_cast_fp16)[name = string("op_18197_cast_fp16")]; + tensor var_18198_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1799_cast_fp16)[name = string("op_18198_cast_fp16")]; + tensor var_18199_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1801_cast_fp16)[name = string("op_18199_cast_fp16")]; + tensor var_18200_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1803_cast_fp16)[name = string("op_18200_cast_fp16")]; + tensor var_18201_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1805_cast_fp16)[name = string("op_18201_cast_fp16")]; + tensor var_18202_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1807_cast_fp16)[name = string("op_18202_cast_fp16")]; + tensor var_18203_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1809_cast_fp16)[name = string("op_18203_cast_fp16")]; + tensor var_18204_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1811_cast_fp16)[name = string("op_18204_cast_fp16")]; + tensor var_18205_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1813_cast_fp16)[name = string("op_18205_cast_fp16")]; + tensor var_18206_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1815_cast_fp16)[name = string("op_18206_cast_fp16")]; + tensor var_18207_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1817_cast_fp16)[name = string("op_18207_cast_fp16")]; + tensor var_18208_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1819_cast_fp16)[name = string("op_18208_cast_fp16")]; + tensor var_18209_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1821_cast_fp16)[name = string("op_18209_cast_fp16")]; + tensor var_18210_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1823_cast_fp16)[name = string("op_18210_cast_fp16")]; + tensor var_18211_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1825_cast_fp16)[name = string("op_18211_cast_fp16")]; + tensor var_18212_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1827_cast_fp16)[name = string("op_18212_cast_fp16")]; + tensor var_18213_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1829_cast_fp16)[name = string("op_18213_cast_fp16")]; + tensor var_18214_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1831_cast_fp16)[name = string("op_18214_cast_fp16")]; + tensor var_18215_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1833_cast_fp16)[name = string("op_18215_cast_fp16")]; + tensor var_18216_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1835_cast_fp16)[name = string("op_18216_cast_fp16")]; + tensor var_18217_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1837_cast_fp16)[name = string("op_18217_cast_fp16")]; + tensor var_18218_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1839_cast_fp16)[name = string("op_18218_cast_fp16")]; + tensor var_18219_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1841_cast_fp16)[name = string("op_18219_cast_fp16")]; + tensor var_18220_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1843_cast_fp16)[name = string("op_18220_cast_fp16")]; + tensor var_18221_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1845_cast_fp16)[name = string("op_18221_cast_fp16")]; + tensor var_18222_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1847_cast_fp16)[name = string("op_18222_cast_fp16")]; + tensor var_18223_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1849_cast_fp16)[name = string("op_18223_cast_fp16")]; + tensor var_18224_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1851_cast_fp16)[name = string("op_18224_cast_fp16")]; + tensor var_18225_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1853_cast_fp16)[name = string("op_18225_cast_fp16")]; + tensor var_18226_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1855_cast_fp16)[name = string("op_18226_cast_fp16")]; + tensor var_18227_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1857_cast_fp16)[name = string("op_18227_cast_fp16")]; + tensor var_18228_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1859_cast_fp16)[name = string("op_18228_cast_fp16")]; + tensor var_18229_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1861_cast_fp16)[name = string("op_18229_cast_fp16")]; + tensor var_18230_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1863_cast_fp16)[name = string("op_18230_cast_fp16")]; + tensor var_18231_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1865_cast_fp16)[name = string("op_18231_cast_fp16")]; + tensor var_18232_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1867_cast_fp16)[name = string("op_18232_cast_fp16")]; + tensor var_18233_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1869_cast_fp16)[name = string("op_18233_cast_fp16")]; + tensor var_18234_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1871_cast_fp16)[name = string("op_18234_cast_fp16")]; + tensor var_18235_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1873_cast_fp16)[name = string("op_18235_cast_fp16")]; + tensor var_18236_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1875_cast_fp16)[name = string("op_18236_cast_fp16")]; + tensor var_18237_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1877_cast_fp16)[name = string("op_18237_cast_fp16")]; + tensor var_18238_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1879_cast_fp16)[name = string("op_18238_cast_fp16")]; + tensor var_18239_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1881_cast_fp16)[name = string("op_18239_cast_fp16")]; + tensor var_18240_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1883_cast_fp16)[name = string("op_18240_cast_fp16")]; + tensor var_18241_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1885_cast_fp16)[name = string("op_18241_cast_fp16")]; + tensor var_18242_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1887_cast_fp16)[name = string("op_18242_cast_fp16")]; + tensor var_18243_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1889_cast_fp16)[name = string("op_18243_cast_fp16")]; + tensor var_18244_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1891_cast_fp16)[name = string("op_18244_cast_fp16")]; + tensor var_18245_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1893_cast_fp16)[name = string("op_18245_cast_fp16")]; + tensor var_18246_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1895_cast_fp16)[name = string("op_18246_cast_fp16")]; + tensor var_18247_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1897_cast_fp16)[name = string("op_18247_cast_fp16")]; + tensor var_18248_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1899_cast_fp16)[name = string("op_18248_cast_fp16")]; + tensor var_18249_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1901_cast_fp16)[name = string("op_18249_cast_fp16")]; + tensor var_18250_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1903_cast_fp16)[name = string("op_18250_cast_fp16")]; + tensor var_18251_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1905_cast_fp16)[name = string("op_18251_cast_fp16")]; + tensor var_18252_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1907_cast_fp16)[name = string("op_18252_cast_fp16")]; + tensor var_18253_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1909_cast_fp16)[name = string("op_18253_cast_fp16")]; + tensor var_18254_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1911_cast_fp16)[name = string("op_18254_cast_fp16")]; + tensor var_18255_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1913_cast_fp16)[name = string("op_18255_cast_fp16")]; + tensor var_18256_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1915_cast_fp16)[name = string("op_18256_cast_fp16")]; + tensor var_18257_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1917_cast_fp16)[name = string("op_18257_cast_fp16")]; + tensor var_18258_cast_fp16 = softmax(axis = var_17004, x = aw_chunk_1919_cast_fp16)[name = string("op_18258_cast_fp16")]; + string var_18260_equation_0 = const()[name = string("op_18260_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18260_cast_fp16 = einsum(equation = var_18260_equation_0, values = (var_17780_cast_fp16, var_18179_cast_fp16))[name = string("op_18260_cast_fp16")]; + string var_18262_equation_0 = const()[name = string("op_18262_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18262_cast_fp16 = einsum(equation = var_18262_equation_0, values = (var_17780_cast_fp16, var_18180_cast_fp16))[name = string("op_18262_cast_fp16")]; + string var_18264_equation_0 = const()[name = string("op_18264_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18264_cast_fp16 = einsum(equation = var_18264_equation_0, values = (var_17780_cast_fp16, var_18181_cast_fp16))[name = string("op_18264_cast_fp16")]; + string var_18266_equation_0 = const()[name = string("op_18266_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18266_cast_fp16 = einsum(equation = var_18266_equation_0, values = (var_17780_cast_fp16, var_18182_cast_fp16))[name = string("op_18266_cast_fp16")]; + string var_18268_equation_0 = const()[name = string("op_18268_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18268_cast_fp16 = einsum(equation = var_18268_equation_0, values = (var_17784_cast_fp16, var_18183_cast_fp16))[name = string("op_18268_cast_fp16")]; + string var_18270_equation_0 = const()[name = string("op_18270_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18270_cast_fp16 = einsum(equation = var_18270_equation_0, values = (var_17784_cast_fp16, var_18184_cast_fp16))[name = string("op_18270_cast_fp16")]; + string var_18272_equation_0 = const()[name = string("op_18272_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18272_cast_fp16 = einsum(equation = var_18272_equation_0, values = (var_17784_cast_fp16, var_18185_cast_fp16))[name = string("op_18272_cast_fp16")]; + string var_18274_equation_0 = const()[name = string("op_18274_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18274_cast_fp16 = einsum(equation = var_18274_equation_0, values = (var_17784_cast_fp16, var_18186_cast_fp16))[name = string("op_18274_cast_fp16")]; + string var_18276_equation_0 = const()[name = string("op_18276_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18276_cast_fp16 = einsum(equation = var_18276_equation_0, values = (var_17788_cast_fp16, var_18187_cast_fp16))[name = string("op_18276_cast_fp16")]; + string var_18278_equation_0 = const()[name = string("op_18278_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18278_cast_fp16 = einsum(equation = var_18278_equation_0, values = (var_17788_cast_fp16, var_18188_cast_fp16))[name = string("op_18278_cast_fp16")]; + string var_18280_equation_0 = const()[name = string("op_18280_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18280_cast_fp16 = einsum(equation = var_18280_equation_0, values = (var_17788_cast_fp16, var_18189_cast_fp16))[name = string("op_18280_cast_fp16")]; + string var_18282_equation_0 = const()[name = string("op_18282_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18282_cast_fp16 = einsum(equation = var_18282_equation_0, values = (var_17788_cast_fp16, var_18190_cast_fp16))[name = string("op_18282_cast_fp16")]; + string var_18284_equation_0 = const()[name = string("op_18284_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18284_cast_fp16 = einsum(equation = var_18284_equation_0, values = (var_17792_cast_fp16, var_18191_cast_fp16))[name = string("op_18284_cast_fp16")]; + string var_18286_equation_0 = const()[name = string("op_18286_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18286_cast_fp16 = einsum(equation = var_18286_equation_0, values = (var_17792_cast_fp16, var_18192_cast_fp16))[name = string("op_18286_cast_fp16")]; + string var_18288_equation_0 = const()[name = string("op_18288_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18288_cast_fp16 = einsum(equation = var_18288_equation_0, values = (var_17792_cast_fp16, var_18193_cast_fp16))[name = string("op_18288_cast_fp16")]; + string var_18290_equation_0 = const()[name = string("op_18290_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18290_cast_fp16 = einsum(equation = var_18290_equation_0, values = (var_17792_cast_fp16, var_18194_cast_fp16))[name = string("op_18290_cast_fp16")]; + string var_18292_equation_0 = const()[name = string("op_18292_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18292_cast_fp16 = einsum(equation = var_18292_equation_0, values = (var_17796_cast_fp16, var_18195_cast_fp16))[name = string("op_18292_cast_fp16")]; + string var_18294_equation_0 = const()[name = string("op_18294_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18294_cast_fp16 = einsum(equation = var_18294_equation_0, values = (var_17796_cast_fp16, var_18196_cast_fp16))[name = string("op_18294_cast_fp16")]; + string var_18296_equation_0 = const()[name = string("op_18296_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18296_cast_fp16 = einsum(equation = var_18296_equation_0, values = (var_17796_cast_fp16, var_18197_cast_fp16))[name = string("op_18296_cast_fp16")]; + string var_18298_equation_0 = const()[name = string("op_18298_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18298_cast_fp16 = einsum(equation = var_18298_equation_0, values = (var_17796_cast_fp16, var_18198_cast_fp16))[name = string("op_18298_cast_fp16")]; + string var_18300_equation_0 = const()[name = string("op_18300_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18300_cast_fp16 = einsum(equation = var_18300_equation_0, values = (var_17800_cast_fp16, var_18199_cast_fp16))[name = string("op_18300_cast_fp16")]; + string var_18302_equation_0 = const()[name = string("op_18302_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18302_cast_fp16 = einsum(equation = var_18302_equation_0, values = (var_17800_cast_fp16, var_18200_cast_fp16))[name = string("op_18302_cast_fp16")]; + string var_18304_equation_0 = const()[name = string("op_18304_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18304_cast_fp16 = einsum(equation = var_18304_equation_0, values = (var_17800_cast_fp16, var_18201_cast_fp16))[name = string("op_18304_cast_fp16")]; + string var_18306_equation_0 = const()[name = string("op_18306_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18306_cast_fp16 = einsum(equation = var_18306_equation_0, values = (var_17800_cast_fp16, var_18202_cast_fp16))[name = string("op_18306_cast_fp16")]; + string var_18308_equation_0 = const()[name = string("op_18308_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18308_cast_fp16 = einsum(equation = var_18308_equation_0, values = (var_17804_cast_fp16, var_18203_cast_fp16))[name = string("op_18308_cast_fp16")]; + string var_18310_equation_0 = const()[name = string("op_18310_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18310_cast_fp16 = einsum(equation = var_18310_equation_0, values = (var_17804_cast_fp16, var_18204_cast_fp16))[name = string("op_18310_cast_fp16")]; + string var_18312_equation_0 = const()[name = string("op_18312_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18312_cast_fp16 = einsum(equation = var_18312_equation_0, values = (var_17804_cast_fp16, var_18205_cast_fp16))[name = string("op_18312_cast_fp16")]; + string var_18314_equation_0 = const()[name = string("op_18314_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18314_cast_fp16 = einsum(equation = var_18314_equation_0, values = (var_17804_cast_fp16, var_18206_cast_fp16))[name = string("op_18314_cast_fp16")]; + string var_18316_equation_0 = const()[name = string("op_18316_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18316_cast_fp16 = einsum(equation = var_18316_equation_0, values = (var_17808_cast_fp16, var_18207_cast_fp16))[name = string("op_18316_cast_fp16")]; + string var_18318_equation_0 = const()[name = string("op_18318_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18318_cast_fp16 = einsum(equation = var_18318_equation_0, values = (var_17808_cast_fp16, var_18208_cast_fp16))[name = string("op_18318_cast_fp16")]; + string var_18320_equation_0 = const()[name = string("op_18320_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18320_cast_fp16 = einsum(equation = var_18320_equation_0, values = (var_17808_cast_fp16, var_18209_cast_fp16))[name = string("op_18320_cast_fp16")]; + string var_18322_equation_0 = const()[name = string("op_18322_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18322_cast_fp16 = einsum(equation = var_18322_equation_0, values = (var_17808_cast_fp16, var_18210_cast_fp16))[name = string("op_18322_cast_fp16")]; + string var_18324_equation_0 = const()[name = string("op_18324_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18324_cast_fp16 = einsum(equation = var_18324_equation_0, values = (var_17812_cast_fp16, var_18211_cast_fp16))[name = string("op_18324_cast_fp16")]; + string var_18326_equation_0 = const()[name = string("op_18326_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18326_cast_fp16 = einsum(equation = var_18326_equation_0, values = (var_17812_cast_fp16, var_18212_cast_fp16))[name = string("op_18326_cast_fp16")]; + string var_18328_equation_0 = const()[name = string("op_18328_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18328_cast_fp16 = einsum(equation = var_18328_equation_0, values = (var_17812_cast_fp16, var_18213_cast_fp16))[name = string("op_18328_cast_fp16")]; + string var_18330_equation_0 = const()[name = string("op_18330_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18330_cast_fp16 = einsum(equation = var_18330_equation_0, values = (var_17812_cast_fp16, var_18214_cast_fp16))[name = string("op_18330_cast_fp16")]; + string var_18332_equation_0 = const()[name = string("op_18332_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18332_cast_fp16 = einsum(equation = var_18332_equation_0, values = (var_17816_cast_fp16, var_18215_cast_fp16))[name = string("op_18332_cast_fp16")]; + string var_18334_equation_0 = const()[name = string("op_18334_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18334_cast_fp16 = einsum(equation = var_18334_equation_0, values = (var_17816_cast_fp16, var_18216_cast_fp16))[name = string("op_18334_cast_fp16")]; + string var_18336_equation_0 = const()[name = string("op_18336_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18336_cast_fp16 = einsum(equation = var_18336_equation_0, values = (var_17816_cast_fp16, var_18217_cast_fp16))[name = string("op_18336_cast_fp16")]; + string var_18338_equation_0 = const()[name = string("op_18338_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18338_cast_fp16 = einsum(equation = var_18338_equation_0, values = (var_17816_cast_fp16, var_18218_cast_fp16))[name = string("op_18338_cast_fp16")]; + string var_18340_equation_0 = const()[name = string("op_18340_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18340_cast_fp16 = einsum(equation = var_18340_equation_0, values = (var_17820_cast_fp16, var_18219_cast_fp16))[name = string("op_18340_cast_fp16")]; + string var_18342_equation_0 = const()[name = string("op_18342_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18342_cast_fp16 = einsum(equation = var_18342_equation_0, values = (var_17820_cast_fp16, var_18220_cast_fp16))[name = string("op_18342_cast_fp16")]; + string var_18344_equation_0 = const()[name = string("op_18344_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18344_cast_fp16 = einsum(equation = var_18344_equation_0, values = (var_17820_cast_fp16, var_18221_cast_fp16))[name = string("op_18344_cast_fp16")]; + string var_18346_equation_0 = const()[name = string("op_18346_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18346_cast_fp16 = einsum(equation = var_18346_equation_0, values = (var_17820_cast_fp16, var_18222_cast_fp16))[name = string("op_18346_cast_fp16")]; + string var_18348_equation_0 = const()[name = string("op_18348_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18348_cast_fp16 = einsum(equation = var_18348_equation_0, values = (var_17824_cast_fp16, var_18223_cast_fp16))[name = string("op_18348_cast_fp16")]; + string var_18350_equation_0 = const()[name = string("op_18350_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18350_cast_fp16 = einsum(equation = var_18350_equation_0, values = (var_17824_cast_fp16, var_18224_cast_fp16))[name = string("op_18350_cast_fp16")]; + string var_18352_equation_0 = const()[name = string("op_18352_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18352_cast_fp16 = einsum(equation = var_18352_equation_0, values = (var_17824_cast_fp16, var_18225_cast_fp16))[name = string("op_18352_cast_fp16")]; + string var_18354_equation_0 = const()[name = string("op_18354_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18354_cast_fp16 = einsum(equation = var_18354_equation_0, values = (var_17824_cast_fp16, var_18226_cast_fp16))[name = string("op_18354_cast_fp16")]; + string var_18356_equation_0 = const()[name = string("op_18356_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18356_cast_fp16 = einsum(equation = var_18356_equation_0, values = (var_17828_cast_fp16, var_18227_cast_fp16))[name = string("op_18356_cast_fp16")]; + string var_18358_equation_0 = const()[name = string("op_18358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18358_cast_fp16 = einsum(equation = var_18358_equation_0, values = (var_17828_cast_fp16, var_18228_cast_fp16))[name = string("op_18358_cast_fp16")]; + string var_18360_equation_0 = const()[name = string("op_18360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18360_cast_fp16 = einsum(equation = var_18360_equation_0, values = (var_17828_cast_fp16, var_18229_cast_fp16))[name = string("op_18360_cast_fp16")]; + string var_18362_equation_0 = const()[name = string("op_18362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18362_cast_fp16 = einsum(equation = var_18362_equation_0, values = (var_17828_cast_fp16, var_18230_cast_fp16))[name = string("op_18362_cast_fp16")]; + string var_18364_equation_0 = const()[name = string("op_18364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18364_cast_fp16 = einsum(equation = var_18364_equation_0, values = (var_17832_cast_fp16, var_18231_cast_fp16))[name = string("op_18364_cast_fp16")]; + string var_18366_equation_0 = const()[name = string("op_18366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18366_cast_fp16 = einsum(equation = var_18366_equation_0, values = (var_17832_cast_fp16, var_18232_cast_fp16))[name = string("op_18366_cast_fp16")]; + string var_18368_equation_0 = const()[name = string("op_18368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18368_cast_fp16 = einsum(equation = var_18368_equation_0, values = (var_17832_cast_fp16, var_18233_cast_fp16))[name = string("op_18368_cast_fp16")]; + string var_18370_equation_0 = const()[name = string("op_18370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18370_cast_fp16 = einsum(equation = var_18370_equation_0, values = (var_17832_cast_fp16, var_18234_cast_fp16))[name = string("op_18370_cast_fp16")]; + string var_18372_equation_0 = const()[name = string("op_18372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18372_cast_fp16 = einsum(equation = var_18372_equation_0, values = (var_17836_cast_fp16, var_18235_cast_fp16))[name = string("op_18372_cast_fp16")]; + string var_18374_equation_0 = const()[name = string("op_18374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18374_cast_fp16 = einsum(equation = var_18374_equation_0, values = (var_17836_cast_fp16, var_18236_cast_fp16))[name = string("op_18374_cast_fp16")]; + string var_18376_equation_0 = const()[name = string("op_18376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18376_cast_fp16 = einsum(equation = var_18376_equation_0, values = (var_17836_cast_fp16, var_18237_cast_fp16))[name = string("op_18376_cast_fp16")]; + string var_18378_equation_0 = const()[name = string("op_18378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18378_cast_fp16 = einsum(equation = var_18378_equation_0, values = (var_17836_cast_fp16, var_18238_cast_fp16))[name = string("op_18378_cast_fp16")]; + string var_18380_equation_0 = const()[name = string("op_18380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18380_cast_fp16 = einsum(equation = var_18380_equation_0, values = (var_17840_cast_fp16, var_18239_cast_fp16))[name = string("op_18380_cast_fp16")]; + string var_18382_equation_0 = const()[name = string("op_18382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18382_cast_fp16 = einsum(equation = var_18382_equation_0, values = (var_17840_cast_fp16, var_18240_cast_fp16))[name = string("op_18382_cast_fp16")]; + string var_18384_equation_0 = const()[name = string("op_18384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18384_cast_fp16 = einsum(equation = var_18384_equation_0, values = (var_17840_cast_fp16, var_18241_cast_fp16))[name = string("op_18384_cast_fp16")]; + string var_18386_equation_0 = const()[name = string("op_18386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18386_cast_fp16 = einsum(equation = var_18386_equation_0, values = (var_17840_cast_fp16, var_18242_cast_fp16))[name = string("op_18386_cast_fp16")]; + string var_18388_equation_0 = const()[name = string("op_18388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18388_cast_fp16 = einsum(equation = var_18388_equation_0, values = (var_17844_cast_fp16, var_18243_cast_fp16))[name = string("op_18388_cast_fp16")]; + string var_18390_equation_0 = const()[name = string("op_18390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18390_cast_fp16 = einsum(equation = var_18390_equation_0, values = (var_17844_cast_fp16, var_18244_cast_fp16))[name = string("op_18390_cast_fp16")]; + string var_18392_equation_0 = const()[name = string("op_18392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18392_cast_fp16 = einsum(equation = var_18392_equation_0, values = (var_17844_cast_fp16, var_18245_cast_fp16))[name = string("op_18392_cast_fp16")]; + string var_18394_equation_0 = const()[name = string("op_18394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18394_cast_fp16 = einsum(equation = var_18394_equation_0, values = (var_17844_cast_fp16, var_18246_cast_fp16))[name = string("op_18394_cast_fp16")]; + string var_18396_equation_0 = const()[name = string("op_18396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18396_cast_fp16 = einsum(equation = var_18396_equation_0, values = (var_17848_cast_fp16, var_18247_cast_fp16))[name = string("op_18396_cast_fp16")]; + string var_18398_equation_0 = const()[name = string("op_18398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18398_cast_fp16 = einsum(equation = var_18398_equation_0, values = (var_17848_cast_fp16, var_18248_cast_fp16))[name = string("op_18398_cast_fp16")]; + string var_18400_equation_0 = const()[name = string("op_18400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18400_cast_fp16 = einsum(equation = var_18400_equation_0, values = (var_17848_cast_fp16, var_18249_cast_fp16))[name = string("op_18400_cast_fp16")]; + string var_18402_equation_0 = const()[name = string("op_18402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18402_cast_fp16 = einsum(equation = var_18402_equation_0, values = (var_17848_cast_fp16, var_18250_cast_fp16))[name = string("op_18402_cast_fp16")]; + string var_18404_equation_0 = const()[name = string("op_18404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18404_cast_fp16 = einsum(equation = var_18404_equation_0, values = (var_17852_cast_fp16, var_18251_cast_fp16))[name = string("op_18404_cast_fp16")]; + string var_18406_equation_0 = const()[name = string("op_18406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18406_cast_fp16 = einsum(equation = var_18406_equation_0, values = (var_17852_cast_fp16, var_18252_cast_fp16))[name = string("op_18406_cast_fp16")]; + string var_18408_equation_0 = const()[name = string("op_18408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18408_cast_fp16 = einsum(equation = var_18408_equation_0, values = (var_17852_cast_fp16, var_18253_cast_fp16))[name = string("op_18408_cast_fp16")]; + string var_18410_equation_0 = const()[name = string("op_18410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18410_cast_fp16 = einsum(equation = var_18410_equation_0, values = (var_17852_cast_fp16, var_18254_cast_fp16))[name = string("op_18410_cast_fp16")]; + string var_18412_equation_0 = const()[name = string("op_18412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18412_cast_fp16 = einsum(equation = var_18412_equation_0, values = (var_17856_cast_fp16, var_18255_cast_fp16))[name = string("op_18412_cast_fp16")]; + string var_18414_equation_0 = const()[name = string("op_18414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18414_cast_fp16 = einsum(equation = var_18414_equation_0, values = (var_17856_cast_fp16, var_18256_cast_fp16))[name = string("op_18414_cast_fp16")]; + string var_18416_equation_0 = const()[name = string("op_18416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18416_cast_fp16 = einsum(equation = var_18416_equation_0, values = (var_17856_cast_fp16, var_18257_cast_fp16))[name = string("op_18416_cast_fp16")]; + string var_18418_equation_0 = const()[name = string("op_18418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_18418_cast_fp16 = einsum(equation = var_18418_equation_0, values = (var_17856_cast_fp16, var_18258_cast_fp16))[name = string("op_18418_cast_fp16")]; + bool var_18420_interleave_0 = const()[name = string("op_18420_interleave_0"), val = bool(false)]; + tensor var_18420_cast_fp16 = concat(axis = var_16979, interleave = var_18420_interleave_0, values = (var_18260_cast_fp16, var_18262_cast_fp16, var_18264_cast_fp16, var_18266_cast_fp16))[name = string("op_18420_cast_fp16")]; + bool var_18422_interleave_0 = const()[name = string("op_18422_interleave_0"), val = bool(false)]; + tensor var_18422_cast_fp16 = concat(axis = var_16979, interleave = var_18422_interleave_0, values = (var_18268_cast_fp16, var_18270_cast_fp16, var_18272_cast_fp16, var_18274_cast_fp16))[name = string("op_18422_cast_fp16")]; + bool var_18424_interleave_0 = const()[name = string("op_18424_interleave_0"), val = bool(false)]; + tensor var_18424_cast_fp16 = concat(axis = var_16979, interleave = var_18424_interleave_0, values = (var_18276_cast_fp16, var_18278_cast_fp16, var_18280_cast_fp16, var_18282_cast_fp16))[name = string("op_18424_cast_fp16")]; + bool var_18426_interleave_0 = const()[name = string("op_18426_interleave_0"), val = bool(false)]; + tensor var_18426_cast_fp16 = concat(axis = var_16979, interleave = var_18426_interleave_0, values = (var_18284_cast_fp16, var_18286_cast_fp16, var_18288_cast_fp16, var_18290_cast_fp16))[name = string("op_18426_cast_fp16")]; + bool var_18428_interleave_0 = const()[name = string("op_18428_interleave_0"), val = bool(false)]; + tensor var_18428_cast_fp16 = concat(axis = var_16979, interleave = var_18428_interleave_0, values = (var_18292_cast_fp16, var_18294_cast_fp16, var_18296_cast_fp16, var_18298_cast_fp16))[name = string("op_18428_cast_fp16")]; + bool var_18430_interleave_0 = const()[name = string("op_18430_interleave_0"), val = bool(false)]; + tensor var_18430_cast_fp16 = concat(axis = var_16979, interleave = var_18430_interleave_0, values = (var_18300_cast_fp16, var_18302_cast_fp16, var_18304_cast_fp16, var_18306_cast_fp16))[name = string("op_18430_cast_fp16")]; + bool var_18432_interleave_0 = const()[name = string("op_18432_interleave_0"), val = bool(false)]; + tensor var_18432_cast_fp16 = concat(axis = var_16979, interleave = var_18432_interleave_0, values = (var_18308_cast_fp16, var_18310_cast_fp16, var_18312_cast_fp16, var_18314_cast_fp16))[name = string("op_18432_cast_fp16")]; + bool var_18434_interleave_0 = const()[name = string("op_18434_interleave_0"), val = bool(false)]; + tensor var_18434_cast_fp16 = concat(axis = var_16979, interleave = var_18434_interleave_0, values = (var_18316_cast_fp16, var_18318_cast_fp16, var_18320_cast_fp16, var_18322_cast_fp16))[name = string("op_18434_cast_fp16")]; + bool var_18436_interleave_0 = const()[name = string("op_18436_interleave_0"), val = bool(false)]; + tensor var_18436_cast_fp16 = concat(axis = var_16979, interleave = var_18436_interleave_0, values = (var_18324_cast_fp16, var_18326_cast_fp16, var_18328_cast_fp16, var_18330_cast_fp16))[name = string("op_18436_cast_fp16")]; + bool var_18438_interleave_0 = const()[name = string("op_18438_interleave_0"), val = bool(false)]; + tensor var_18438_cast_fp16 = concat(axis = var_16979, interleave = var_18438_interleave_0, values = (var_18332_cast_fp16, var_18334_cast_fp16, var_18336_cast_fp16, var_18338_cast_fp16))[name = string("op_18438_cast_fp16")]; + bool var_18440_interleave_0 = const()[name = string("op_18440_interleave_0"), val = bool(false)]; + tensor var_18440_cast_fp16 = concat(axis = var_16979, interleave = var_18440_interleave_0, values = (var_18340_cast_fp16, var_18342_cast_fp16, var_18344_cast_fp16, var_18346_cast_fp16))[name = string("op_18440_cast_fp16")]; + bool var_18442_interleave_0 = const()[name = string("op_18442_interleave_0"), val = bool(false)]; + tensor var_18442_cast_fp16 = concat(axis = var_16979, interleave = var_18442_interleave_0, values = (var_18348_cast_fp16, var_18350_cast_fp16, var_18352_cast_fp16, var_18354_cast_fp16))[name = string("op_18442_cast_fp16")]; + bool var_18444_interleave_0 = const()[name = string("op_18444_interleave_0"), val = bool(false)]; + tensor var_18444_cast_fp16 = concat(axis = var_16979, interleave = var_18444_interleave_0, values = (var_18356_cast_fp16, var_18358_cast_fp16, var_18360_cast_fp16, var_18362_cast_fp16))[name = string("op_18444_cast_fp16")]; + bool var_18446_interleave_0 = const()[name = string("op_18446_interleave_0"), val = bool(false)]; + tensor var_18446_cast_fp16 = concat(axis = var_16979, interleave = var_18446_interleave_0, values = (var_18364_cast_fp16, var_18366_cast_fp16, var_18368_cast_fp16, var_18370_cast_fp16))[name = string("op_18446_cast_fp16")]; + bool var_18448_interleave_0 = const()[name = string("op_18448_interleave_0"), val = bool(false)]; + tensor var_18448_cast_fp16 = concat(axis = var_16979, interleave = var_18448_interleave_0, values = (var_18372_cast_fp16, var_18374_cast_fp16, var_18376_cast_fp16, var_18378_cast_fp16))[name = string("op_18448_cast_fp16")]; + bool var_18450_interleave_0 = const()[name = string("op_18450_interleave_0"), val = bool(false)]; + tensor var_18450_cast_fp16 = concat(axis = var_16979, interleave = var_18450_interleave_0, values = (var_18380_cast_fp16, var_18382_cast_fp16, var_18384_cast_fp16, var_18386_cast_fp16))[name = string("op_18450_cast_fp16")]; + bool var_18452_interleave_0 = const()[name = string("op_18452_interleave_0"), val = bool(false)]; + tensor var_18452_cast_fp16 = concat(axis = var_16979, interleave = var_18452_interleave_0, values = (var_18388_cast_fp16, var_18390_cast_fp16, var_18392_cast_fp16, var_18394_cast_fp16))[name = string("op_18452_cast_fp16")]; + bool var_18454_interleave_0 = const()[name = string("op_18454_interleave_0"), val = bool(false)]; + tensor var_18454_cast_fp16 = concat(axis = var_16979, interleave = var_18454_interleave_0, values = (var_18396_cast_fp16, var_18398_cast_fp16, var_18400_cast_fp16, var_18402_cast_fp16))[name = string("op_18454_cast_fp16")]; + bool var_18456_interleave_0 = const()[name = string("op_18456_interleave_0"), val = bool(false)]; + tensor var_18456_cast_fp16 = concat(axis = var_16979, interleave = var_18456_interleave_0, values = (var_18404_cast_fp16, var_18406_cast_fp16, var_18408_cast_fp16, var_18410_cast_fp16))[name = string("op_18456_cast_fp16")]; + bool var_18458_interleave_0 = const()[name = string("op_18458_interleave_0"), val = bool(false)]; + tensor var_18458_cast_fp16 = concat(axis = var_16979, interleave = var_18458_interleave_0, values = (var_18412_cast_fp16, var_18414_cast_fp16, var_18416_cast_fp16, var_18418_cast_fp16))[name = string("op_18458_cast_fp16")]; + bool input_89_interleave_0 = const()[name = string("input_89_interleave_0"), val = bool(false)]; + tensor input_89_cast_fp16 = concat(axis = var_17004, interleave = input_89_interleave_0, values = (var_18420_cast_fp16, var_18422_cast_fp16, var_18424_cast_fp16, var_18426_cast_fp16, var_18428_cast_fp16, var_18430_cast_fp16, var_18432_cast_fp16, var_18434_cast_fp16, var_18436_cast_fp16, var_18438_cast_fp16, var_18440_cast_fp16, var_18442_cast_fp16, var_18444_cast_fp16, var_18446_cast_fp16, var_18448_cast_fp16, var_18450_cast_fp16, var_18452_cast_fp16, var_18454_cast_fp16, var_18456_cast_fp16, var_18458_cast_fp16))[name = string("input_89_cast_fp16")]; + string obj_47_pad_type_0 = const()[name = string("obj_47_pad_type_0"), val = string("valid")]; + tensor obj_47_strides_0 = const()[name = string("obj_47_strides_0"), val = tensor([1, 1])]; + tensor obj_47_pad_0 = const()[name = string("obj_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_47_dilations_0 = const()[name = string("obj_47_dilations_0"), val = tensor([1, 1])]; + int32 obj_47_groups_0 = const()[name = string("obj_47_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457391360)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460668224)))]; + tensor obj_47_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_47_dilations_0, groups = obj_47_groups_0, pad = obj_47_pad_0, pad_type = obj_47_pad_type_0, strides = obj_47_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_47_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_47_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_18477_to_fp16 = const()[name = string("op_18477_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_18477_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460670848)))]; + tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460673472)))]; + fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460676096)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473783360)))]; + tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; + tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + string hidden_states_27_pad_type_0 = const()[name = string("hidden_states_27_pad_type_0"), val = string("valid")]; + tensor hidden_states_27_strides_0 = const()[name = string("hidden_states_27_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_27_pad_0 = const()[name = string("hidden_states_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_27_dilations_0 = const()[name = string("hidden_states_27_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_27_groups_0 = const()[name = string("hidden_states_27_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(473793664)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486900928)))]; + tensor hidden_states_27_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_27_dilations_0, groups = hidden_states_27_groups_0, pad = hidden_states_27_pad_0, pad_type = hidden_states_27_pad_type_0, strides = hidden_states_27_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_27_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_27_cast_fp16)[name = string("inputs_49_cast_fp16")]; + int32 var_18506 = const()[name = string("op_18506"), val = int32(3)]; + int32 var_18531 = const()[name = string("op_18531"), val = int32(1)]; + tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; + fp16 var_18548_to_fp16 = const()[name = string("op_18548_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_18548_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; + tensor obj_49_gamma_0_to_fp16 = const()[name = string("obj_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486903552)))]; + tensor obj_49_beta_0_to_fp16 = const()[name = string("obj_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486906176)))]; + fp16 obj_49_epsilon_0_to_fp16 = const()[name = string("obj_49_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_49_cast_fp16 = batch_norm(beta = obj_49_beta_0_to_fp16, epsilon = obj_49_epsilon_0_to_fp16, gamma = obj_49_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_49_cast_fp16")]; + string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; + tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; + tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; + int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(486908800)))]; + tensor layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490185664)))]; + tensor query_25_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("query_25_cast_fp16")]; + string key_25_pad_type_0 = const()[name = string("key_25_pad_type_0"), val = string("valid")]; + tensor key_25_strides_0 = const()[name = string("key_25_strides_0"), val = tensor([1, 1])]; + tensor key_25_pad_0 = const()[name = string("key_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_25_dilations_0 = const()[name = string("key_25_dilations_0"), val = tensor([1, 1])]; + int32 key_25_groups_0 = const()[name = string("key_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490188288)))]; + tensor key_25_cast_fp16 = conv(dilations = key_25_dilations_0, groups = key_25_groups_0, pad = key_25_pad_0, pad_type = key_25_pad_type_0, strides = key_25_strides_0, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("key_25_cast_fp16")]; + string value_25_pad_type_0 = const()[name = string("value_25_pad_type_0"), val = string("valid")]; + tensor value_25_strides_0 = const()[name = string("value_25_strides_0"), val = tensor([1, 1])]; + tensor value_25_pad_0 = const()[name = string("value_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_25_dilations_0 = const()[name = string("value_25_dilations_0"), val = tensor([1, 1])]; + int32 value_25_groups_0 = const()[name = string("value_25_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(493465152)))]; + tensor layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496742016)))]; + tensor value_25_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_bias_to_fp16, dilations = value_25_dilations_0, groups = value_25_groups_0, pad = value_25_pad_0, pad_type = value_25_pad_type_0, strides = value_25_strides_0, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = obj_49_cast_fp16)[name = string("value_25_cast_fp16")]; + tensor var_18586_begin_0 = const()[name = string("op_18586_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18586_end_0 = const()[name = string("op_18586_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18586_end_mask_0 = const()[name = string("op_18586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18586_cast_fp16 = slice_by_index(begin = var_18586_begin_0, end = var_18586_end_0, end_mask = var_18586_end_mask_0, x = query_25_cast_fp16)[name = string("op_18586_cast_fp16")]; + tensor var_18590_begin_0 = const()[name = string("op_18590_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18590_end_0 = const()[name = string("op_18590_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_18590_end_mask_0 = const()[name = string("op_18590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18590_cast_fp16 = slice_by_index(begin = var_18590_begin_0, end = var_18590_end_0, end_mask = var_18590_end_mask_0, x = query_25_cast_fp16)[name = string("op_18590_cast_fp16")]; + tensor var_18594_begin_0 = const()[name = string("op_18594_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18594_end_0 = const()[name = string("op_18594_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_18594_end_mask_0 = const()[name = string("op_18594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18594_cast_fp16 = slice_by_index(begin = var_18594_begin_0, end = var_18594_end_0, end_mask = var_18594_end_mask_0, x = query_25_cast_fp16)[name = string("op_18594_cast_fp16")]; + tensor var_18598_begin_0 = const()[name = string("op_18598_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18598_end_0 = const()[name = string("op_18598_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_18598_end_mask_0 = const()[name = string("op_18598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18598_cast_fp16 = slice_by_index(begin = var_18598_begin_0, end = var_18598_end_0, end_mask = var_18598_end_mask_0, x = query_25_cast_fp16)[name = string("op_18598_cast_fp16")]; + tensor var_18602_begin_0 = const()[name = string("op_18602_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18602_end_0 = const()[name = string("op_18602_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_18602_end_mask_0 = const()[name = string("op_18602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18602_cast_fp16 = slice_by_index(begin = var_18602_begin_0, end = var_18602_end_0, end_mask = var_18602_end_mask_0, x = query_25_cast_fp16)[name = string("op_18602_cast_fp16")]; + tensor var_18606_begin_0 = const()[name = string("op_18606_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18606_end_0 = const()[name = string("op_18606_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_18606_end_mask_0 = const()[name = string("op_18606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18606_cast_fp16 = slice_by_index(begin = var_18606_begin_0, end = var_18606_end_0, end_mask = var_18606_end_mask_0, x = query_25_cast_fp16)[name = string("op_18606_cast_fp16")]; + tensor var_18610_begin_0 = const()[name = string("op_18610_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18610_end_0 = const()[name = string("op_18610_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_18610_end_mask_0 = const()[name = string("op_18610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18610_cast_fp16 = slice_by_index(begin = var_18610_begin_0, end = var_18610_end_0, end_mask = var_18610_end_mask_0, x = query_25_cast_fp16)[name = string("op_18610_cast_fp16")]; + tensor var_18614_begin_0 = const()[name = string("op_18614_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18614_end_0 = const()[name = string("op_18614_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_18614_end_mask_0 = const()[name = string("op_18614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18614_cast_fp16 = slice_by_index(begin = var_18614_begin_0, end = var_18614_end_0, end_mask = var_18614_end_mask_0, x = query_25_cast_fp16)[name = string("op_18614_cast_fp16")]; + tensor var_18618_begin_0 = const()[name = string("op_18618_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18618_end_0 = const()[name = string("op_18618_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_18618_end_mask_0 = const()[name = string("op_18618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18618_cast_fp16 = slice_by_index(begin = var_18618_begin_0, end = var_18618_end_0, end_mask = var_18618_end_mask_0, x = query_25_cast_fp16)[name = string("op_18618_cast_fp16")]; + tensor var_18622_begin_0 = const()[name = string("op_18622_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18622_end_0 = const()[name = string("op_18622_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_18622_end_mask_0 = const()[name = string("op_18622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18622_cast_fp16 = slice_by_index(begin = var_18622_begin_0, end = var_18622_end_0, end_mask = var_18622_end_mask_0, x = query_25_cast_fp16)[name = string("op_18622_cast_fp16")]; + tensor var_18626_begin_0 = const()[name = string("op_18626_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18626_end_0 = const()[name = string("op_18626_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_18626_end_mask_0 = const()[name = string("op_18626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18626_cast_fp16 = slice_by_index(begin = var_18626_begin_0, end = var_18626_end_0, end_mask = var_18626_end_mask_0, x = query_25_cast_fp16)[name = string("op_18626_cast_fp16")]; + tensor var_18630_begin_0 = const()[name = string("op_18630_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18630_end_0 = const()[name = string("op_18630_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_18630_end_mask_0 = const()[name = string("op_18630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18630_cast_fp16 = slice_by_index(begin = var_18630_begin_0, end = var_18630_end_0, end_mask = var_18630_end_mask_0, x = query_25_cast_fp16)[name = string("op_18630_cast_fp16")]; + tensor var_18634_begin_0 = const()[name = string("op_18634_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18634_end_0 = const()[name = string("op_18634_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_18634_end_mask_0 = const()[name = string("op_18634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18634_cast_fp16 = slice_by_index(begin = var_18634_begin_0, end = var_18634_end_0, end_mask = var_18634_end_mask_0, x = query_25_cast_fp16)[name = string("op_18634_cast_fp16")]; + tensor var_18638_begin_0 = const()[name = string("op_18638_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18638_end_0 = const()[name = string("op_18638_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_18638_end_mask_0 = const()[name = string("op_18638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18638_cast_fp16 = slice_by_index(begin = var_18638_begin_0, end = var_18638_end_0, end_mask = var_18638_end_mask_0, x = query_25_cast_fp16)[name = string("op_18638_cast_fp16")]; + tensor var_18642_begin_0 = const()[name = string("op_18642_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18642_end_0 = const()[name = string("op_18642_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_18642_end_mask_0 = const()[name = string("op_18642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18642_cast_fp16 = slice_by_index(begin = var_18642_begin_0, end = var_18642_end_0, end_mask = var_18642_end_mask_0, x = query_25_cast_fp16)[name = string("op_18642_cast_fp16")]; + tensor var_18646_begin_0 = const()[name = string("op_18646_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18646_end_0 = const()[name = string("op_18646_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_18646_end_mask_0 = const()[name = string("op_18646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18646_cast_fp16 = slice_by_index(begin = var_18646_begin_0, end = var_18646_end_0, end_mask = var_18646_end_mask_0, x = query_25_cast_fp16)[name = string("op_18646_cast_fp16")]; + tensor var_18650_begin_0 = const()[name = string("op_18650_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18650_end_0 = const()[name = string("op_18650_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_18650_end_mask_0 = const()[name = string("op_18650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18650_cast_fp16 = slice_by_index(begin = var_18650_begin_0, end = var_18650_end_0, end_mask = var_18650_end_mask_0, x = query_25_cast_fp16)[name = string("op_18650_cast_fp16")]; + tensor var_18654_begin_0 = const()[name = string("op_18654_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18654_end_0 = const()[name = string("op_18654_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_18654_end_mask_0 = const()[name = string("op_18654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18654_cast_fp16 = slice_by_index(begin = var_18654_begin_0, end = var_18654_end_0, end_mask = var_18654_end_mask_0, x = query_25_cast_fp16)[name = string("op_18654_cast_fp16")]; + tensor var_18658_begin_0 = const()[name = string("op_18658_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18658_end_0 = const()[name = string("op_18658_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_18658_end_mask_0 = const()[name = string("op_18658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18658_cast_fp16 = slice_by_index(begin = var_18658_begin_0, end = var_18658_end_0, end_mask = var_18658_end_mask_0, x = query_25_cast_fp16)[name = string("op_18658_cast_fp16")]; + tensor var_18662_begin_0 = const()[name = string("op_18662_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18662_end_0 = const()[name = string("op_18662_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_18662_end_mask_0 = const()[name = string("op_18662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18662_cast_fp16 = slice_by_index(begin = var_18662_begin_0, end = var_18662_end_0, end_mask = var_18662_end_mask_0, x = query_25_cast_fp16)[name = string("op_18662_cast_fp16")]; + tensor var_18671_begin_0 = const()[name = string("op_18671_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18671_end_0 = const()[name = string("op_18671_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18671_end_mask_0 = const()[name = string("op_18671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18671_cast_fp16 = slice_by_index(begin = var_18671_begin_0, end = var_18671_end_0, end_mask = var_18671_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18671_cast_fp16")]; + tensor var_18678_begin_0 = const()[name = string("op_18678_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18678_end_0 = const()[name = string("op_18678_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18678_end_mask_0 = const()[name = string("op_18678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18678_cast_fp16 = slice_by_index(begin = var_18678_begin_0, end = var_18678_end_0, end_mask = var_18678_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18678_cast_fp16")]; + tensor var_18685_begin_0 = const()[name = string("op_18685_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18685_end_0 = const()[name = string("op_18685_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18685_end_mask_0 = const()[name = string("op_18685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18685_cast_fp16 = slice_by_index(begin = var_18685_begin_0, end = var_18685_end_0, end_mask = var_18685_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18685_cast_fp16")]; + tensor var_18692_begin_0 = const()[name = string("op_18692_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18692_end_0 = const()[name = string("op_18692_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18692_end_mask_0 = const()[name = string("op_18692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18692_cast_fp16 = slice_by_index(begin = var_18692_begin_0, end = var_18692_end_0, end_mask = var_18692_end_mask_0, x = var_18586_cast_fp16)[name = string("op_18692_cast_fp16")]; + tensor var_18699_begin_0 = const()[name = string("op_18699_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18699_end_0 = const()[name = string("op_18699_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18699_end_mask_0 = const()[name = string("op_18699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18699_cast_fp16 = slice_by_index(begin = var_18699_begin_0, end = var_18699_end_0, end_mask = var_18699_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18699_cast_fp16")]; + tensor var_18706_begin_0 = const()[name = string("op_18706_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18706_end_0 = const()[name = string("op_18706_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18706_end_mask_0 = const()[name = string("op_18706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18706_cast_fp16 = slice_by_index(begin = var_18706_begin_0, end = var_18706_end_0, end_mask = var_18706_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18706_cast_fp16")]; + tensor var_18713_begin_0 = const()[name = string("op_18713_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18713_end_0 = const()[name = string("op_18713_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18713_end_mask_0 = const()[name = string("op_18713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18713_cast_fp16 = slice_by_index(begin = var_18713_begin_0, end = var_18713_end_0, end_mask = var_18713_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18713_cast_fp16")]; + tensor var_18720_begin_0 = const()[name = string("op_18720_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18720_end_0 = const()[name = string("op_18720_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18720_end_mask_0 = const()[name = string("op_18720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18720_cast_fp16 = slice_by_index(begin = var_18720_begin_0, end = var_18720_end_0, end_mask = var_18720_end_mask_0, x = var_18590_cast_fp16)[name = string("op_18720_cast_fp16")]; + tensor var_18727_begin_0 = const()[name = string("op_18727_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18727_end_0 = const()[name = string("op_18727_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18727_end_mask_0 = const()[name = string("op_18727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18727_cast_fp16 = slice_by_index(begin = var_18727_begin_0, end = var_18727_end_0, end_mask = var_18727_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18727_cast_fp16")]; + tensor var_18734_begin_0 = const()[name = string("op_18734_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18734_end_0 = const()[name = string("op_18734_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18734_end_mask_0 = const()[name = string("op_18734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18734_cast_fp16 = slice_by_index(begin = var_18734_begin_0, end = var_18734_end_0, end_mask = var_18734_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18734_cast_fp16")]; + tensor var_18741_begin_0 = const()[name = string("op_18741_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18741_end_0 = const()[name = string("op_18741_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18741_end_mask_0 = const()[name = string("op_18741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18741_cast_fp16 = slice_by_index(begin = var_18741_begin_0, end = var_18741_end_0, end_mask = var_18741_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18741_cast_fp16")]; + tensor var_18748_begin_0 = const()[name = string("op_18748_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18748_end_0 = const()[name = string("op_18748_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18748_end_mask_0 = const()[name = string("op_18748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18748_cast_fp16 = slice_by_index(begin = var_18748_begin_0, end = var_18748_end_0, end_mask = var_18748_end_mask_0, x = var_18594_cast_fp16)[name = string("op_18748_cast_fp16")]; + tensor var_18755_begin_0 = const()[name = string("op_18755_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18755_end_0 = const()[name = string("op_18755_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18755_end_mask_0 = const()[name = string("op_18755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18755_cast_fp16 = slice_by_index(begin = var_18755_begin_0, end = var_18755_end_0, end_mask = var_18755_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18755_cast_fp16")]; + tensor var_18762_begin_0 = const()[name = string("op_18762_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18762_end_0 = const()[name = string("op_18762_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18762_end_mask_0 = const()[name = string("op_18762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18762_cast_fp16 = slice_by_index(begin = var_18762_begin_0, end = var_18762_end_0, end_mask = var_18762_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18762_cast_fp16")]; + tensor var_18769_begin_0 = const()[name = string("op_18769_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18769_end_0 = const()[name = string("op_18769_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18769_end_mask_0 = const()[name = string("op_18769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18769_cast_fp16 = slice_by_index(begin = var_18769_begin_0, end = var_18769_end_0, end_mask = var_18769_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18769_cast_fp16")]; + tensor var_18776_begin_0 = const()[name = string("op_18776_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18776_end_0 = const()[name = string("op_18776_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18776_end_mask_0 = const()[name = string("op_18776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18776_cast_fp16 = slice_by_index(begin = var_18776_begin_0, end = var_18776_end_0, end_mask = var_18776_end_mask_0, x = var_18598_cast_fp16)[name = string("op_18776_cast_fp16")]; + tensor var_18783_begin_0 = const()[name = string("op_18783_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18783_end_0 = const()[name = string("op_18783_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18783_end_mask_0 = const()[name = string("op_18783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18783_cast_fp16 = slice_by_index(begin = var_18783_begin_0, end = var_18783_end_0, end_mask = var_18783_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18783_cast_fp16")]; + tensor var_18790_begin_0 = const()[name = string("op_18790_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18790_end_0 = const()[name = string("op_18790_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18790_end_mask_0 = const()[name = string("op_18790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18790_cast_fp16 = slice_by_index(begin = var_18790_begin_0, end = var_18790_end_0, end_mask = var_18790_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18790_cast_fp16")]; + tensor var_18797_begin_0 = const()[name = string("op_18797_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18797_end_0 = const()[name = string("op_18797_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18797_end_mask_0 = const()[name = string("op_18797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18797_cast_fp16 = slice_by_index(begin = var_18797_begin_0, end = var_18797_end_0, end_mask = var_18797_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18797_cast_fp16")]; + tensor var_18804_begin_0 = const()[name = string("op_18804_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18804_end_0 = const()[name = string("op_18804_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18804_end_mask_0 = const()[name = string("op_18804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18804_cast_fp16 = slice_by_index(begin = var_18804_begin_0, end = var_18804_end_0, end_mask = var_18804_end_mask_0, x = var_18602_cast_fp16)[name = string("op_18804_cast_fp16")]; + tensor var_18811_begin_0 = const()[name = string("op_18811_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18811_end_0 = const()[name = string("op_18811_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18811_end_mask_0 = const()[name = string("op_18811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18811_cast_fp16 = slice_by_index(begin = var_18811_begin_0, end = var_18811_end_0, end_mask = var_18811_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18811_cast_fp16")]; + tensor var_18818_begin_0 = const()[name = string("op_18818_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18818_end_0 = const()[name = string("op_18818_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18818_end_mask_0 = const()[name = string("op_18818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18818_cast_fp16 = slice_by_index(begin = var_18818_begin_0, end = var_18818_end_0, end_mask = var_18818_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18818_cast_fp16")]; + tensor var_18825_begin_0 = const()[name = string("op_18825_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18825_end_0 = const()[name = string("op_18825_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18825_end_mask_0 = const()[name = string("op_18825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18825_cast_fp16 = slice_by_index(begin = var_18825_begin_0, end = var_18825_end_0, end_mask = var_18825_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18825_cast_fp16")]; + tensor var_18832_begin_0 = const()[name = string("op_18832_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18832_end_0 = const()[name = string("op_18832_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18832_end_mask_0 = const()[name = string("op_18832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18832_cast_fp16 = slice_by_index(begin = var_18832_begin_0, end = var_18832_end_0, end_mask = var_18832_end_mask_0, x = var_18606_cast_fp16)[name = string("op_18832_cast_fp16")]; + tensor var_18839_begin_0 = const()[name = string("op_18839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18839_end_0 = const()[name = string("op_18839_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18839_end_mask_0 = const()[name = string("op_18839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18839_cast_fp16 = slice_by_index(begin = var_18839_begin_0, end = var_18839_end_0, end_mask = var_18839_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18839_cast_fp16")]; + tensor var_18846_begin_0 = const()[name = string("op_18846_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18846_end_0 = const()[name = string("op_18846_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18846_end_mask_0 = const()[name = string("op_18846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18846_cast_fp16 = slice_by_index(begin = var_18846_begin_0, end = var_18846_end_0, end_mask = var_18846_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18846_cast_fp16")]; + tensor var_18853_begin_0 = const()[name = string("op_18853_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18853_end_0 = const()[name = string("op_18853_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18853_end_mask_0 = const()[name = string("op_18853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18853_cast_fp16 = slice_by_index(begin = var_18853_begin_0, end = var_18853_end_0, end_mask = var_18853_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18853_cast_fp16")]; + tensor var_18860_begin_0 = const()[name = string("op_18860_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18860_end_0 = const()[name = string("op_18860_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18860_end_mask_0 = const()[name = string("op_18860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18860_cast_fp16 = slice_by_index(begin = var_18860_begin_0, end = var_18860_end_0, end_mask = var_18860_end_mask_0, x = var_18610_cast_fp16)[name = string("op_18860_cast_fp16")]; + tensor var_18867_begin_0 = const()[name = string("op_18867_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18867_end_0 = const()[name = string("op_18867_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18867_end_mask_0 = const()[name = string("op_18867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18867_cast_fp16 = slice_by_index(begin = var_18867_begin_0, end = var_18867_end_0, end_mask = var_18867_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18867_cast_fp16")]; + tensor var_18874_begin_0 = const()[name = string("op_18874_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18874_end_0 = const()[name = string("op_18874_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18874_end_mask_0 = const()[name = string("op_18874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18874_cast_fp16 = slice_by_index(begin = var_18874_begin_0, end = var_18874_end_0, end_mask = var_18874_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18874_cast_fp16")]; + tensor var_18881_begin_0 = const()[name = string("op_18881_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18881_end_0 = const()[name = string("op_18881_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18881_end_mask_0 = const()[name = string("op_18881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18881_cast_fp16 = slice_by_index(begin = var_18881_begin_0, end = var_18881_end_0, end_mask = var_18881_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18881_cast_fp16")]; + tensor var_18888_begin_0 = const()[name = string("op_18888_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18888_end_0 = const()[name = string("op_18888_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18888_end_mask_0 = const()[name = string("op_18888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18888_cast_fp16 = slice_by_index(begin = var_18888_begin_0, end = var_18888_end_0, end_mask = var_18888_end_mask_0, x = var_18614_cast_fp16)[name = string("op_18888_cast_fp16")]; + tensor var_18895_begin_0 = const()[name = string("op_18895_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18895_end_0 = const()[name = string("op_18895_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18895_end_mask_0 = const()[name = string("op_18895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18895_cast_fp16 = slice_by_index(begin = var_18895_begin_0, end = var_18895_end_0, end_mask = var_18895_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18895_cast_fp16")]; + tensor var_18902_begin_0 = const()[name = string("op_18902_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18902_end_0 = const()[name = string("op_18902_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18902_end_mask_0 = const()[name = string("op_18902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18902_cast_fp16 = slice_by_index(begin = var_18902_begin_0, end = var_18902_end_0, end_mask = var_18902_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18902_cast_fp16")]; + tensor var_18909_begin_0 = const()[name = string("op_18909_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18909_end_0 = const()[name = string("op_18909_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18909_end_mask_0 = const()[name = string("op_18909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18909_cast_fp16 = slice_by_index(begin = var_18909_begin_0, end = var_18909_end_0, end_mask = var_18909_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18909_cast_fp16")]; + tensor var_18916_begin_0 = const()[name = string("op_18916_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18916_end_0 = const()[name = string("op_18916_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18916_end_mask_0 = const()[name = string("op_18916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18916_cast_fp16 = slice_by_index(begin = var_18916_begin_0, end = var_18916_end_0, end_mask = var_18916_end_mask_0, x = var_18618_cast_fp16)[name = string("op_18916_cast_fp16")]; + tensor var_18923_begin_0 = const()[name = string("op_18923_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18923_end_0 = const()[name = string("op_18923_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18923_end_mask_0 = const()[name = string("op_18923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18923_cast_fp16 = slice_by_index(begin = var_18923_begin_0, end = var_18923_end_0, end_mask = var_18923_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18923_cast_fp16")]; + tensor var_18930_begin_0 = const()[name = string("op_18930_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18930_end_0 = const()[name = string("op_18930_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18930_end_mask_0 = const()[name = string("op_18930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18930_cast_fp16 = slice_by_index(begin = var_18930_begin_0, end = var_18930_end_0, end_mask = var_18930_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18930_cast_fp16")]; + tensor var_18937_begin_0 = const()[name = string("op_18937_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18937_end_0 = const()[name = string("op_18937_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18937_end_mask_0 = const()[name = string("op_18937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18937_cast_fp16 = slice_by_index(begin = var_18937_begin_0, end = var_18937_end_0, end_mask = var_18937_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18937_cast_fp16")]; + tensor var_18944_begin_0 = const()[name = string("op_18944_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18944_end_0 = const()[name = string("op_18944_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18944_end_mask_0 = const()[name = string("op_18944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18944_cast_fp16 = slice_by_index(begin = var_18944_begin_0, end = var_18944_end_0, end_mask = var_18944_end_mask_0, x = var_18622_cast_fp16)[name = string("op_18944_cast_fp16")]; + tensor var_18951_begin_0 = const()[name = string("op_18951_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18951_end_0 = const()[name = string("op_18951_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18951_end_mask_0 = const()[name = string("op_18951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18951_cast_fp16 = slice_by_index(begin = var_18951_begin_0, end = var_18951_end_0, end_mask = var_18951_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18951_cast_fp16")]; + tensor var_18958_begin_0 = const()[name = string("op_18958_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18958_end_0 = const()[name = string("op_18958_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18958_end_mask_0 = const()[name = string("op_18958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18958_cast_fp16 = slice_by_index(begin = var_18958_begin_0, end = var_18958_end_0, end_mask = var_18958_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18958_cast_fp16")]; + tensor var_18965_begin_0 = const()[name = string("op_18965_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18965_end_0 = const()[name = string("op_18965_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18965_end_mask_0 = const()[name = string("op_18965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18965_cast_fp16 = slice_by_index(begin = var_18965_begin_0, end = var_18965_end_0, end_mask = var_18965_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18965_cast_fp16")]; + tensor var_18972_begin_0 = const()[name = string("op_18972_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18972_end_0 = const()[name = string("op_18972_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18972_end_mask_0 = const()[name = string("op_18972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18972_cast_fp16 = slice_by_index(begin = var_18972_begin_0, end = var_18972_end_0, end_mask = var_18972_end_mask_0, x = var_18626_cast_fp16)[name = string("op_18972_cast_fp16")]; + tensor var_18979_begin_0 = const()[name = string("op_18979_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18979_end_0 = const()[name = string("op_18979_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18979_end_mask_0 = const()[name = string("op_18979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18979_cast_fp16 = slice_by_index(begin = var_18979_begin_0, end = var_18979_end_0, end_mask = var_18979_end_mask_0, x = var_18630_cast_fp16)[name = string("op_18979_cast_fp16")]; + tensor var_18986_begin_0 = const()[name = string("op_18986_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18986_end_0 = const()[name = string("op_18986_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18986_end_mask_0 = const()[name = string("op_18986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18986_cast_fp16 = slice_by_index(begin = var_18986_begin_0, end = var_18986_end_0, end_mask = var_18986_end_mask_0, x = var_18630_cast_fp16)[name = string("op_18986_cast_fp16")]; + tensor var_18993_begin_0 = const()[name = string("op_18993_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18993_end_0 = const()[name = string("op_18993_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18993_end_mask_0 = const()[name = string("op_18993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18993_cast_fp16 = slice_by_index(begin = var_18993_begin_0, end = var_18993_end_0, end_mask = var_18993_end_mask_0, x = var_18630_cast_fp16)[name = string("op_18993_cast_fp16")]; + tensor var_19000_begin_0 = const()[name = string("op_19000_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19000_end_0 = const()[name = string("op_19000_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19000_end_mask_0 = const()[name = string("op_19000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19000_cast_fp16 = slice_by_index(begin = var_19000_begin_0, end = var_19000_end_0, end_mask = var_19000_end_mask_0, x = var_18630_cast_fp16)[name = string("op_19000_cast_fp16")]; + tensor var_19007_begin_0 = const()[name = string("op_19007_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19007_end_0 = const()[name = string("op_19007_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19007_end_mask_0 = const()[name = string("op_19007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19007_cast_fp16 = slice_by_index(begin = var_19007_begin_0, end = var_19007_end_0, end_mask = var_19007_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19007_cast_fp16")]; + tensor var_19014_begin_0 = const()[name = string("op_19014_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19014_end_0 = const()[name = string("op_19014_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19014_end_mask_0 = const()[name = string("op_19014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19014_cast_fp16 = slice_by_index(begin = var_19014_begin_0, end = var_19014_end_0, end_mask = var_19014_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19014_cast_fp16")]; + tensor var_19021_begin_0 = const()[name = string("op_19021_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19021_end_0 = const()[name = string("op_19021_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19021_end_mask_0 = const()[name = string("op_19021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19021_cast_fp16 = slice_by_index(begin = var_19021_begin_0, end = var_19021_end_0, end_mask = var_19021_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19021_cast_fp16")]; + tensor var_19028_begin_0 = const()[name = string("op_19028_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19028_end_0 = const()[name = string("op_19028_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19028_end_mask_0 = const()[name = string("op_19028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19028_cast_fp16 = slice_by_index(begin = var_19028_begin_0, end = var_19028_end_0, end_mask = var_19028_end_mask_0, x = var_18634_cast_fp16)[name = string("op_19028_cast_fp16")]; + tensor var_19035_begin_0 = const()[name = string("op_19035_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19035_end_0 = const()[name = string("op_19035_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19035_end_mask_0 = const()[name = string("op_19035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19035_cast_fp16 = slice_by_index(begin = var_19035_begin_0, end = var_19035_end_0, end_mask = var_19035_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19035_cast_fp16")]; + tensor var_19042_begin_0 = const()[name = string("op_19042_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19042_end_0 = const()[name = string("op_19042_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19042_end_mask_0 = const()[name = string("op_19042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19042_cast_fp16 = slice_by_index(begin = var_19042_begin_0, end = var_19042_end_0, end_mask = var_19042_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19042_cast_fp16")]; + tensor var_19049_begin_0 = const()[name = string("op_19049_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19049_end_0 = const()[name = string("op_19049_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19049_end_mask_0 = const()[name = string("op_19049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19049_cast_fp16 = slice_by_index(begin = var_19049_begin_0, end = var_19049_end_0, end_mask = var_19049_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19049_cast_fp16")]; + tensor var_19056_begin_0 = const()[name = string("op_19056_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19056_end_0 = const()[name = string("op_19056_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19056_end_mask_0 = const()[name = string("op_19056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19056_cast_fp16 = slice_by_index(begin = var_19056_begin_0, end = var_19056_end_0, end_mask = var_19056_end_mask_0, x = var_18638_cast_fp16)[name = string("op_19056_cast_fp16")]; + tensor var_19063_begin_0 = const()[name = string("op_19063_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19063_end_0 = const()[name = string("op_19063_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19063_end_mask_0 = const()[name = string("op_19063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19063_cast_fp16 = slice_by_index(begin = var_19063_begin_0, end = var_19063_end_0, end_mask = var_19063_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19063_cast_fp16")]; + tensor var_19070_begin_0 = const()[name = string("op_19070_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19070_end_0 = const()[name = string("op_19070_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19070_end_mask_0 = const()[name = string("op_19070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19070_cast_fp16 = slice_by_index(begin = var_19070_begin_0, end = var_19070_end_0, end_mask = var_19070_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19070_cast_fp16")]; + tensor var_19077_begin_0 = const()[name = string("op_19077_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19077_end_0 = const()[name = string("op_19077_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19077_end_mask_0 = const()[name = string("op_19077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19077_cast_fp16 = slice_by_index(begin = var_19077_begin_0, end = var_19077_end_0, end_mask = var_19077_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19077_cast_fp16")]; + tensor var_19084_begin_0 = const()[name = string("op_19084_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19084_end_0 = const()[name = string("op_19084_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19084_end_mask_0 = const()[name = string("op_19084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19084_cast_fp16 = slice_by_index(begin = var_19084_begin_0, end = var_19084_end_0, end_mask = var_19084_end_mask_0, x = var_18642_cast_fp16)[name = string("op_19084_cast_fp16")]; + tensor var_19091_begin_0 = const()[name = string("op_19091_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19091_end_0 = const()[name = string("op_19091_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19091_end_mask_0 = const()[name = string("op_19091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19091_cast_fp16 = slice_by_index(begin = var_19091_begin_0, end = var_19091_end_0, end_mask = var_19091_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19091_cast_fp16")]; + tensor var_19098_begin_0 = const()[name = string("op_19098_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19098_end_0 = const()[name = string("op_19098_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19098_end_mask_0 = const()[name = string("op_19098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19098_cast_fp16 = slice_by_index(begin = var_19098_begin_0, end = var_19098_end_0, end_mask = var_19098_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19098_cast_fp16")]; + tensor var_19105_begin_0 = const()[name = string("op_19105_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19105_end_0 = const()[name = string("op_19105_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19105_end_mask_0 = const()[name = string("op_19105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19105_cast_fp16 = slice_by_index(begin = var_19105_begin_0, end = var_19105_end_0, end_mask = var_19105_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19105_cast_fp16")]; + tensor var_19112_begin_0 = const()[name = string("op_19112_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19112_end_0 = const()[name = string("op_19112_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19112_end_mask_0 = const()[name = string("op_19112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19112_cast_fp16 = slice_by_index(begin = var_19112_begin_0, end = var_19112_end_0, end_mask = var_19112_end_mask_0, x = var_18646_cast_fp16)[name = string("op_19112_cast_fp16")]; + tensor var_19119_begin_0 = const()[name = string("op_19119_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19119_end_0 = const()[name = string("op_19119_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19119_end_mask_0 = const()[name = string("op_19119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19119_cast_fp16 = slice_by_index(begin = var_19119_begin_0, end = var_19119_end_0, end_mask = var_19119_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19119_cast_fp16")]; + tensor var_19126_begin_0 = const()[name = string("op_19126_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19126_end_0 = const()[name = string("op_19126_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19126_end_mask_0 = const()[name = string("op_19126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19126_cast_fp16 = slice_by_index(begin = var_19126_begin_0, end = var_19126_end_0, end_mask = var_19126_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19126_cast_fp16")]; + tensor var_19133_begin_0 = const()[name = string("op_19133_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19133_end_0 = const()[name = string("op_19133_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19133_end_mask_0 = const()[name = string("op_19133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19133_cast_fp16 = slice_by_index(begin = var_19133_begin_0, end = var_19133_end_0, end_mask = var_19133_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19133_cast_fp16")]; + tensor var_19140_begin_0 = const()[name = string("op_19140_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19140_end_0 = const()[name = string("op_19140_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19140_end_mask_0 = const()[name = string("op_19140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19140_cast_fp16 = slice_by_index(begin = var_19140_begin_0, end = var_19140_end_0, end_mask = var_19140_end_mask_0, x = var_18650_cast_fp16)[name = string("op_19140_cast_fp16")]; + tensor var_19147_begin_0 = const()[name = string("op_19147_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19147_end_0 = const()[name = string("op_19147_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19147_end_mask_0 = const()[name = string("op_19147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19147_cast_fp16 = slice_by_index(begin = var_19147_begin_0, end = var_19147_end_0, end_mask = var_19147_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19147_cast_fp16")]; + tensor var_19154_begin_0 = const()[name = string("op_19154_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19154_end_0 = const()[name = string("op_19154_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19154_end_mask_0 = const()[name = string("op_19154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19154_cast_fp16 = slice_by_index(begin = var_19154_begin_0, end = var_19154_end_0, end_mask = var_19154_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19154_cast_fp16")]; + tensor var_19161_begin_0 = const()[name = string("op_19161_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19161_end_0 = const()[name = string("op_19161_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19161_end_mask_0 = const()[name = string("op_19161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19161_cast_fp16 = slice_by_index(begin = var_19161_begin_0, end = var_19161_end_0, end_mask = var_19161_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19161_cast_fp16")]; + tensor var_19168_begin_0 = const()[name = string("op_19168_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19168_end_0 = const()[name = string("op_19168_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19168_end_mask_0 = const()[name = string("op_19168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19168_cast_fp16 = slice_by_index(begin = var_19168_begin_0, end = var_19168_end_0, end_mask = var_19168_end_mask_0, x = var_18654_cast_fp16)[name = string("op_19168_cast_fp16")]; + tensor var_19175_begin_0 = const()[name = string("op_19175_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19175_end_0 = const()[name = string("op_19175_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19175_end_mask_0 = const()[name = string("op_19175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19175_cast_fp16 = slice_by_index(begin = var_19175_begin_0, end = var_19175_end_0, end_mask = var_19175_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19175_cast_fp16")]; + tensor var_19182_begin_0 = const()[name = string("op_19182_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19182_end_0 = const()[name = string("op_19182_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19182_end_mask_0 = const()[name = string("op_19182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19182_cast_fp16 = slice_by_index(begin = var_19182_begin_0, end = var_19182_end_0, end_mask = var_19182_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19182_cast_fp16")]; + tensor var_19189_begin_0 = const()[name = string("op_19189_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19189_end_0 = const()[name = string("op_19189_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19189_end_mask_0 = const()[name = string("op_19189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19189_cast_fp16 = slice_by_index(begin = var_19189_begin_0, end = var_19189_end_0, end_mask = var_19189_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19189_cast_fp16")]; + tensor var_19196_begin_0 = const()[name = string("op_19196_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19196_end_0 = const()[name = string("op_19196_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19196_end_mask_0 = const()[name = string("op_19196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19196_cast_fp16 = slice_by_index(begin = var_19196_begin_0, end = var_19196_end_0, end_mask = var_19196_end_mask_0, x = var_18658_cast_fp16)[name = string("op_19196_cast_fp16")]; + tensor var_19203_begin_0 = const()[name = string("op_19203_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19203_end_0 = const()[name = string("op_19203_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_19203_end_mask_0 = const()[name = string("op_19203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19203_cast_fp16 = slice_by_index(begin = var_19203_begin_0, end = var_19203_end_0, end_mask = var_19203_end_mask_0, x = var_18662_cast_fp16)[name = string("op_19203_cast_fp16")]; + tensor var_19210_begin_0 = const()[name = string("op_19210_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_19210_end_0 = const()[name = string("op_19210_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_19210_end_mask_0 = const()[name = string("op_19210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19210_cast_fp16 = slice_by_index(begin = var_19210_begin_0, end = var_19210_end_0, end_mask = var_19210_end_mask_0, x = var_18662_cast_fp16)[name = string("op_19210_cast_fp16")]; + tensor var_19217_begin_0 = const()[name = string("op_19217_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_19217_end_0 = const()[name = string("op_19217_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_19217_end_mask_0 = const()[name = string("op_19217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19217_cast_fp16 = slice_by_index(begin = var_19217_begin_0, end = var_19217_end_0, end_mask = var_19217_end_mask_0, x = var_18662_cast_fp16)[name = string("op_19217_cast_fp16")]; + tensor var_19224_begin_0 = const()[name = string("op_19224_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_19224_end_0 = const()[name = string("op_19224_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19224_end_mask_0 = const()[name = string("op_19224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19224_cast_fp16 = slice_by_index(begin = var_19224_begin_0, end = var_19224_end_0, end_mask = var_19224_end_mask_0, x = var_18662_cast_fp16)[name = string("op_19224_cast_fp16")]; + tensor k_25_perm_0 = const()[name = string("k_25_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_19229_begin_0 = const()[name = string("op_19229_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19229_end_0 = const()[name = string("op_19229_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_19229_end_mask_0 = const()[name = string("op_19229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_25_cast_fp16 = transpose(perm = k_25_perm_0, x = key_25_cast_fp16)[name = string("transpose_19")]; + tensor var_19229_cast_fp16 = slice_by_index(begin = var_19229_begin_0, end = var_19229_end_0, end_mask = var_19229_end_mask_0, x = k_25_cast_fp16)[name = string("op_19229_cast_fp16")]; + tensor var_19233_begin_0 = const()[name = string("op_19233_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_19233_end_0 = const()[name = string("op_19233_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_19233_end_mask_0 = const()[name = string("op_19233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19233_cast_fp16 = slice_by_index(begin = var_19233_begin_0, end = var_19233_end_0, end_mask = var_19233_end_mask_0, x = k_25_cast_fp16)[name = string("op_19233_cast_fp16")]; + tensor var_19237_begin_0 = const()[name = string("op_19237_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_19237_end_0 = const()[name = string("op_19237_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_19237_end_mask_0 = const()[name = string("op_19237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19237_cast_fp16 = slice_by_index(begin = var_19237_begin_0, end = var_19237_end_0, end_mask = var_19237_end_mask_0, x = k_25_cast_fp16)[name = string("op_19237_cast_fp16")]; + tensor var_19241_begin_0 = const()[name = string("op_19241_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_19241_end_0 = const()[name = string("op_19241_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_19241_end_mask_0 = const()[name = string("op_19241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19241_cast_fp16 = slice_by_index(begin = var_19241_begin_0, end = var_19241_end_0, end_mask = var_19241_end_mask_0, x = k_25_cast_fp16)[name = string("op_19241_cast_fp16")]; + tensor var_19245_begin_0 = const()[name = string("op_19245_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_19245_end_0 = const()[name = string("op_19245_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_19245_end_mask_0 = const()[name = string("op_19245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19245_cast_fp16 = slice_by_index(begin = var_19245_begin_0, end = var_19245_end_0, end_mask = var_19245_end_mask_0, x = k_25_cast_fp16)[name = string("op_19245_cast_fp16")]; + tensor var_19249_begin_0 = const()[name = string("op_19249_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_19249_end_0 = const()[name = string("op_19249_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_19249_end_mask_0 = const()[name = string("op_19249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19249_cast_fp16 = slice_by_index(begin = var_19249_begin_0, end = var_19249_end_0, end_mask = var_19249_end_mask_0, x = k_25_cast_fp16)[name = string("op_19249_cast_fp16")]; + tensor var_19253_begin_0 = const()[name = string("op_19253_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_19253_end_0 = const()[name = string("op_19253_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_19253_end_mask_0 = const()[name = string("op_19253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19253_cast_fp16 = slice_by_index(begin = var_19253_begin_0, end = var_19253_end_0, end_mask = var_19253_end_mask_0, x = k_25_cast_fp16)[name = string("op_19253_cast_fp16")]; + tensor var_19257_begin_0 = const()[name = string("op_19257_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_19257_end_0 = const()[name = string("op_19257_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_19257_end_mask_0 = const()[name = string("op_19257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19257_cast_fp16 = slice_by_index(begin = var_19257_begin_0, end = var_19257_end_0, end_mask = var_19257_end_mask_0, x = k_25_cast_fp16)[name = string("op_19257_cast_fp16")]; + tensor var_19261_begin_0 = const()[name = string("op_19261_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_19261_end_0 = const()[name = string("op_19261_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_19261_end_mask_0 = const()[name = string("op_19261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19261_cast_fp16 = slice_by_index(begin = var_19261_begin_0, end = var_19261_end_0, end_mask = var_19261_end_mask_0, x = k_25_cast_fp16)[name = string("op_19261_cast_fp16")]; + tensor var_19265_begin_0 = const()[name = string("op_19265_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_19265_end_0 = const()[name = string("op_19265_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_19265_end_mask_0 = const()[name = string("op_19265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19265_cast_fp16 = slice_by_index(begin = var_19265_begin_0, end = var_19265_end_0, end_mask = var_19265_end_mask_0, x = k_25_cast_fp16)[name = string("op_19265_cast_fp16")]; + tensor var_19269_begin_0 = const()[name = string("op_19269_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_19269_end_0 = const()[name = string("op_19269_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_19269_end_mask_0 = const()[name = string("op_19269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19269_cast_fp16 = slice_by_index(begin = var_19269_begin_0, end = var_19269_end_0, end_mask = var_19269_end_mask_0, x = k_25_cast_fp16)[name = string("op_19269_cast_fp16")]; + tensor var_19273_begin_0 = const()[name = string("op_19273_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_19273_end_0 = const()[name = string("op_19273_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_19273_end_mask_0 = const()[name = string("op_19273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19273_cast_fp16 = slice_by_index(begin = var_19273_begin_0, end = var_19273_end_0, end_mask = var_19273_end_mask_0, x = k_25_cast_fp16)[name = string("op_19273_cast_fp16")]; + tensor var_19277_begin_0 = const()[name = string("op_19277_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_19277_end_0 = const()[name = string("op_19277_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_19277_end_mask_0 = const()[name = string("op_19277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19277_cast_fp16 = slice_by_index(begin = var_19277_begin_0, end = var_19277_end_0, end_mask = var_19277_end_mask_0, x = k_25_cast_fp16)[name = string("op_19277_cast_fp16")]; + tensor var_19281_begin_0 = const()[name = string("op_19281_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_19281_end_0 = const()[name = string("op_19281_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_19281_end_mask_0 = const()[name = string("op_19281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19281_cast_fp16 = slice_by_index(begin = var_19281_begin_0, end = var_19281_end_0, end_mask = var_19281_end_mask_0, x = k_25_cast_fp16)[name = string("op_19281_cast_fp16")]; + tensor var_19285_begin_0 = const()[name = string("op_19285_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_19285_end_0 = const()[name = string("op_19285_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_19285_end_mask_0 = const()[name = string("op_19285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19285_cast_fp16 = slice_by_index(begin = var_19285_begin_0, end = var_19285_end_0, end_mask = var_19285_end_mask_0, x = k_25_cast_fp16)[name = string("op_19285_cast_fp16")]; + tensor var_19289_begin_0 = const()[name = string("op_19289_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_19289_end_0 = const()[name = string("op_19289_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_19289_end_mask_0 = const()[name = string("op_19289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19289_cast_fp16 = slice_by_index(begin = var_19289_begin_0, end = var_19289_end_0, end_mask = var_19289_end_mask_0, x = k_25_cast_fp16)[name = string("op_19289_cast_fp16")]; + tensor var_19293_begin_0 = const()[name = string("op_19293_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19293_end_0 = const()[name = string("op_19293_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_19293_end_mask_0 = const()[name = string("op_19293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19293_cast_fp16 = slice_by_index(begin = var_19293_begin_0, end = var_19293_end_0, end_mask = var_19293_end_mask_0, x = k_25_cast_fp16)[name = string("op_19293_cast_fp16")]; + tensor var_19297_begin_0 = const()[name = string("op_19297_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19297_end_0 = const()[name = string("op_19297_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_19297_end_mask_0 = const()[name = string("op_19297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19297_cast_fp16 = slice_by_index(begin = var_19297_begin_0, end = var_19297_end_0, end_mask = var_19297_end_mask_0, x = k_25_cast_fp16)[name = string("op_19297_cast_fp16")]; + tensor var_19301_begin_0 = const()[name = string("op_19301_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19301_end_0 = const()[name = string("op_19301_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_19301_end_mask_0 = const()[name = string("op_19301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19301_cast_fp16 = slice_by_index(begin = var_19301_begin_0, end = var_19301_end_0, end_mask = var_19301_end_mask_0, x = k_25_cast_fp16)[name = string("op_19301_cast_fp16")]; + tensor var_19305_begin_0 = const()[name = string("op_19305_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19305_end_0 = const()[name = string("op_19305_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_19305_end_mask_0 = const()[name = string("op_19305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19305_cast_fp16 = slice_by_index(begin = var_19305_begin_0, end = var_19305_end_0, end_mask = var_19305_end_mask_0, x = k_25_cast_fp16)[name = string("op_19305_cast_fp16")]; + tensor var_19307_begin_0 = const()[name = string("op_19307_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19307_end_0 = const()[name = string("op_19307_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19307_end_mask_0 = const()[name = string("op_19307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19307_cast_fp16 = slice_by_index(begin = var_19307_begin_0, end = var_19307_end_0, end_mask = var_19307_end_mask_0, x = value_25_cast_fp16)[name = string("op_19307_cast_fp16")]; + tensor var_19311_begin_0 = const()[name = string("op_19311_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19311_end_0 = const()[name = string("op_19311_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_19311_end_mask_0 = const()[name = string("op_19311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19311_cast_fp16 = slice_by_index(begin = var_19311_begin_0, end = var_19311_end_0, end_mask = var_19311_end_mask_0, x = value_25_cast_fp16)[name = string("op_19311_cast_fp16")]; + tensor var_19315_begin_0 = const()[name = string("op_19315_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19315_end_0 = const()[name = string("op_19315_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_19315_end_mask_0 = const()[name = string("op_19315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19315_cast_fp16 = slice_by_index(begin = var_19315_begin_0, end = var_19315_end_0, end_mask = var_19315_end_mask_0, x = value_25_cast_fp16)[name = string("op_19315_cast_fp16")]; + tensor var_19319_begin_0 = const()[name = string("op_19319_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19319_end_0 = const()[name = string("op_19319_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_19319_end_mask_0 = const()[name = string("op_19319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19319_cast_fp16 = slice_by_index(begin = var_19319_begin_0, end = var_19319_end_0, end_mask = var_19319_end_mask_0, x = value_25_cast_fp16)[name = string("op_19319_cast_fp16")]; + tensor var_19323_begin_0 = const()[name = string("op_19323_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19323_end_0 = const()[name = string("op_19323_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_19323_end_mask_0 = const()[name = string("op_19323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19323_cast_fp16 = slice_by_index(begin = var_19323_begin_0, end = var_19323_end_0, end_mask = var_19323_end_mask_0, x = value_25_cast_fp16)[name = string("op_19323_cast_fp16")]; + tensor var_19327_begin_0 = const()[name = string("op_19327_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19327_end_0 = const()[name = string("op_19327_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_19327_end_mask_0 = const()[name = string("op_19327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19327_cast_fp16 = slice_by_index(begin = var_19327_begin_0, end = var_19327_end_0, end_mask = var_19327_end_mask_0, x = value_25_cast_fp16)[name = string("op_19327_cast_fp16")]; + tensor var_19331_begin_0 = const()[name = string("op_19331_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19331_end_0 = const()[name = string("op_19331_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_19331_end_mask_0 = const()[name = string("op_19331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19331_cast_fp16 = slice_by_index(begin = var_19331_begin_0, end = var_19331_end_0, end_mask = var_19331_end_mask_0, x = value_25_cast_fp16)[name = string("op_19331_cast_fp16")]; + tensor var_19335_begin_0 = const()[name = string("op_19335_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19335_end_0 = const()[name = string("op_19335_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_19335_end_mask_0 = const()[name = string("op_19335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19335_cast_fp16 = slice_by_index(begin = var_19335_begin_0, end = var_19335_end_0, end_mask = var_19335_end_mask_0, x = value_25_cast_fp16)[name = string("op_19335_cast_fp16")]; + tensor var_19339_begin_0 = const()[name = string("op_19339_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19339_end_0 = const()[name = string("op_19339_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_19339_end_mask_0 = const()[name = string("op_19339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19339_cast_fp16 = slice_by_index(begin = var_19339_begin_0, end = var_19339_end_0, end_mask = var_19339_end_mask_0, x = value_25_cast_fp16)[name = string("op_19339_cast_fp16")]; + tensor var_19343_begin_0 = const()[name = string("op_19343_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19343_end_0 = const()[name = string("op_19343_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_19343_end_mask_0 = const()[name = string("op_19343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19343_cast_fp16 = slice_by_index(begin = var_19343_begin_0, end = var_19343_end_0, end_mask = var_19343_end_mask_0, x = value_25_cast_fp16)[name = string("op_19343_cast_fp16")]; + tensor var_19347_begin_0 = const()[name = string("op_19347_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19347_end_0 = const()[name = string("op_19347_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_19347_end_mask_0 = const()[name = string("op_19347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19347_cast_fp16 = slice_by_index(begin = var_19347_begin_0, end = var_19347_end_0, end_mask = var_19347_end_mask_0, x = value_25_cast_fp16)[name = string("op_19347_cast_fp16")]; + tensor var_19351_begin_0 = const()[name = string("op_19351_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19351_end_0 = const()[name = string("op_19351_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_19351_end_mask_0 = const()[name = string("op_19351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19351_cast_fp16 = slice_by_index(begin = var_19351_begin_0, end = var_19351_end_0, end_mask = var_19351_end_mask_0, x = value_25_cast_fp16)[name = string("op_19351_cast_fp16")]; + tensor var_19355_begin_0 = const()[name = string("op_19355_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19355_end_0 = const()[name = string("op_19355_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_19355_end_mask_0 = const()[name = string("op_19355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19355_cast_fp16 = slice_by_index(begin = var_19355_begin_0, end = var_19355_end_0, end_mask = var_19355_end_mask_0, x = value_25_cast_fp16)[name = string("op_19355_cast_fp16")]; + tensor var_19359_begin_0 = const()[name = string("op_19359_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19359_end_0 = const()[name = string("op_19359_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_19359_end_mask_0 = const()[name = string("op_19359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19359_cast_fp16 = slice_by_index(begin = var_19359_begin_0, end = var_19359_end_0, end_mask = var_19359_end_mask_0, x = value_25_cast_fp16)[name = string("op_19359_cast_fp16")]; + tensor var_19363_begin_0 = const()[name = string("op_19363_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19363_end_0 = const()[name = string("op_19363_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_19363_end_mask_0 = const()[name = string("op_19363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19363_cast_fp16 = slice_by_index(begin = var_19363_begin_0, end = var_19363_end_0, end_mask = var_19363_end_mask_0, x = value_25_cast_fp16)[name = string("op_19363_cast_fp16")]; + tensor var_19367_begin_0 = const()[name = string("op_19367_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19367_end_0 = const()[name = string("op_19367_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_19367_end_mask_0 = const()[name = string("op_19367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19367_cast_fp16 = slice_by_index(begin = var_19367_begin_0, end = var_19367_end_0, end_mask = var_19367_end_mask_0, x = value_25_cast_fp16)[name = string("op_19367_cast_fp16")]; + tensor var_19371_begin_0 = const()[name = string("op_19371_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19371_end_0 = const()[name = string("op_19371_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_19371_end_mask_0 = const()[name = string("op_19371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19371_cast_fp16 = slice_by_index(begin = var_19371_begin_0, end = var_19371_end_0, end_mask = var_19371_end_mask_0, x = value_25_cast_fp16)[name = string("op_19371_cast_fp16")]; + tensor var_19375_begin_0 = const()[name = string("op_19375_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19375_end_0 = const()[name = string("op_19375_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_19375_end_mask_0 = const()[name = string("op_19375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19375_cast_fp16 = slice_by_index(begin = var_19375_begin_0, end = var_19375_end_0, end_mask = var_19375_end_mask_0, x = value_25_cast_fp16)[name = string("op_19375_cast_fp16")]; + tensor var_19379_begin_0 = const()[name = string("op_19379_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19379_end_0 = const()[name = string("op_19379_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_19379_end_mask_0 = const()[name = string("op_19379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19379_cast_fp16 = slice_by_index(begin = var_19379_begin_0, end = var_19379_end_0, end_mask = var_19379_end_mask_0, x = value_25_cast_fp16)[name = string("op_19379_cast_fp16")]; + tensor var_19383_begin_0 = const()[name = string("op_19383_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19383_end_0 = const()[name = string("op_19383_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_19383_end_mask_0 = const()[name = string("op_19383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19383_cast_fp16 = slice_by_index(begin = var_19383_begin_0, end = var_19383_end_0, end_mask = var_19383_end_mask_0, x = value_25_cast_fp16)[name = string("op_19383_cast_fp16")]; + string _SplitHeadsQ__mh_w_1921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1921_equation_0, values = (var_19229_cast_fp16, var_18671_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1921_cast_fp16")]; + string _SplitHeadsQ__mh_w_1923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1923_equation_0, values = (var_19229_cast_fp16, var_18678_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1923_cast_fp16")]; + string _SplitHeadsQ__mh_w_1925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1925_equation_0, values = (var_19229_cast_fp16, var_18685_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1925_cast_fp16")]; + string _SplitHeadsQ__mh_w_1927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1927_equation_0, values = (var_19229_cast_fp16, var_18692_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1927_cast_fp16")]; + string _SplitHeadsQ__mh_w_1929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1929_equation_0, values = (var_19233_cast_fp16, var_18699_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1929_cast_fp16")]; + string _SplitHeadsQ__mh_w_1931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1931_equation_0, values = (var_19233_cast_fp16, var_18706_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1931_cast_fp16")]; + string _SplitHeadsQ__mh_w_1933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1933_equation_0, values = (var_19233_cast_fp16, var_18713_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1933_cast_fp16")]; + string _SplitHeadsQ__mh_w_1935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1935_equation_0, values = (var_19233_cast_fp16, var_18720_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1935_cast_fp16")]; + string _SplitHeadsQ__mh_w_1937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1937_equation_0, values = (var_19237_cast_fp16, var_18727_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1937_cast_fp16")]; + string _SplitHeadsQ__mh_w_1939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1939_equation_0, values = (var_19237_cast_fp16, var_18734_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1939_cast_fp16")]; + string _SplitHeadsQ__mh_w_1941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1941_equation_0, values = (var_19237_cast_fp16, var_18741_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1941_cast_fp16")]; + string _SplitHeadsQ__mh_w_1943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1943_equation_0, values = (var_19237_cast_fp16, var_18748_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1943_cast_fp16")]; + string _SplitHeadsQ__mh_w_1945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1945_equation_0, values = (var_19241_cast_fp16, var_18755_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1945_cast_fp16")]; + string _SplitHeadsQ__mh_w_1947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1947_equation_0, values = (var_19241_cast_fp16, var_18762_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1947_cast_fp16")]; + string _SplitHeadsQ__mh_w_1949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1949_equation_0, values = (var_19241_cast_fp16, var_18769_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1949_cast_fp16")]; + string _SplitHeadsQ__mh_w_1951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1951_equation_0, values = (var_19241_cast_fp16, var_18776_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1951_cast_fp16")]; + string _SplitHeadsQ__mh_w_1953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1953_equation_0, values = (var_19245_cast_fp16, var_18783_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1953_cast_fp16")]; + string _SplitHeadsQ__mh_w_1955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1955_equation_0, values = (var_19245_cast_fp16, var_18790_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1955_cast_fp16")]; + string _SplitHeadsQ__mh_w_1957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1957_equation_0, values = (var_19245_cast_fp16, var_18797_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1957_cast_fp16")]; + string _SplitHeadsQ__mh_w_1959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1959_equation_0, values = (var_19245_cast_fp16, var_18804_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1959_cast_fp16")]; + string _SplitHeadsQ__mh_w_1961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1961_equation_0, values = (var_19249_cast_fp16, var_18811_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1961_cast_fp16")]; + string _SplitHeadsQ__mh_w_1963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1963_equation_0, values = (var_19249_cast_fp16, var_18818_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1963_cast_fp16")]; + string _SplitHeadsQ__mh_w_1965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1965_equation_0, values = (var_19249_cast_fp16, var_18825_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1965_cast_fp16")]; + string _SplitHeadsQ__mh_w_1967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1967_equation_0, values = (var_19249_cast_fp16, var_18832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1967_cast_fp16")]; + string _SplitHeadsQ__mh_w_1969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1969_equation_0, values = (var_19253_cast_fp16, var_18839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1969_cast_fp16")]; + string _SplitHeadsQ__mh_w_1971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1971_equation_0, values = (var_19253_cast_fp16, var_18846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1971_cast_fp16")]; + string _SplitHeadsQ__mh_w_1973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1973_equation_0, values = (var_19253_cast_fp16, var_18853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1973_cast_fp16")]; + string _SplitHeadsQ__mh_w_1975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1975_equation_0, values = (var_19253_cast_fp16, var_18860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1975_cast_fp16")]; + string _SplitHeadsQ__mh_w_1977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1977_equation_0, values = (var_19257_cast_fp16, var_18867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1977_cast_fp16")]; + string _SplitHeadsQ__mh_w_1979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1979_equation_0, values = (var_19257_cast_fp16, var_18874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1979_cast_fp16")]; + string _SplitHeadsQ__mh_w_1981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1981_equation_0, values = (var_19257_cast_fp16, var_18881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1981_cast_fp16")]; + string _SplitHeadsQ__mh_w_1983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1983_equation_0, values = (var_19257_cast_fp16, var_18888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1983_cast_fp16")]; + string _SplitHeadsQ__mh_w_1985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1985_equation_0, values = (var_19261_cast_fp16, var_18895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1985_cast_fp16")]; + string _SplitHeadsQ__mh_w_1987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1987_equation_0, values = (var_19261_cast_fp16, var_18902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1987_cast_fp16")]; + string _SplitHeadsQ__mh_w_1989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1989_equation_0, values = (var_19261_cast_fp16, var_18909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1989_cast_fp16")]; + string _SplitHeadsQ__mh_w_1991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1991_equation_0, values = (var_19261_cast_fp16, var_18916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1991_cast_fp16")]; + string _SplitHeadsQ__mh_w_1993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1993_equation_0, values = (var_19265_cast_fp16, var_18923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1993_cast_fp16")]; + string _SplitHeadsQ__mh_w_1995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1995_equation_0, values = (var_19265_cast_fp16, var_18930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1995_cast_fp16")]; + string _SplitHeadsQ__mh_w_1997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1997_equation_0, values = (var_19265_cast_fp16, var_18937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1997_cast_fp16")]; + string _SplitHeadsQ__mh_w_1999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1999_equation_0, values = (var_19265_cast_fp16, var_18944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1999_cast_fp16")]; + string _SplitHeadsQ__mh_w_2001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2001_equation_0, values = (var_19269_cast_fp16, var_18951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2001_cast_fp16")]; + string _SplitHeadsQ__mh_w_2003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2003_equation_0, values = (var_19269_cast_fp16, var_18958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2003_cast_fp16")]; + string _SplitHeadsQ__mh_w_2005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2005_equation_0, values = (var_19269_cast_fp16, var_18965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2005_cast_fp16")]; + string _SplitHeadsQ__mh_w_2007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2007_equation_0, values = (var_19269_cast_fp16, var_18972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2007_cast_fp16")]; + string _SplitHeadsQ__mh_w_2009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2009_equation_0, values = (var_19273_cast_fp16, var_18979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2009_cast_fp16")]; + string _SplitHeadsQ__mh_w_2011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2011_equation_0, values = (var_19273_cast_fp16, var_18986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2011_cast_fp16")]; + string _SplitHeadsQ__mh_w_2013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2013_equation_0, values = (var_19273_cast_fp16, var_18993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2013_cast_fp16")]; + string _SplitHeadsQ__mh_w_2015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2015_equation_0, values = (var_19273_cast_fp16, var_19000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2015_cast_fp16")]; + string _SplitHeadsQ__mh_w_2017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2017_equation_0, values = (var_19277_cast_fp16, var_19007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2017_cast_fp16")]; + string _SplitHeadsQ__mh_w_2019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2019_equation_0, values = (var_19277_cast_fp16, var_19014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2019_cast_fp16")]; + string _SplitHeadsQ__mh_w_2021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2021_equation_0, values = (var_19277_cast_fp16, var_19021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2021_cast_fp16")]; + string _SplitHeadsQ__mh_w_2023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2023_equation_0, values = (var_19277_cast_fp16, var_19028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2023_cast_fp16")]; + string _SplitHeadsQ__mh_w_2025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2025_equation_0, values = (var_19281_cast_fp16, var_19035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2025_cast_fp16")]; + string _SplitHeadsQ__mh_w_2027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2027_equation_0, values = (var_19281_cast_fp16, var_19042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2027_cast_fp16")]; + string _SplitHeadsQ__mh_w_2029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2029_equation_0, values = (var_19281_cast_fp16, var_19049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2029_cast_fp16")]; + string _SplitHeadsQ__mh_w_2031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2031_equation_0, values = (var_19281_cast_fp16, var_19056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2031_cast_fp16")]; + string _SplitHeadsQ__mh_w_2033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2033_equation_0, values = (var_19285_cast_fp16, var_19063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2033_cast_fp16")]; + string _SplitHeadsQ__mh_w_2035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2035_equation_0, values = (var_19285_cast_fp16, var_19070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2035_cast_fp16")]; + string _SplitHeadsQ__mh_w_2037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2037_equation_0, values = (var_19285_cast_fp16, var_19077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2037_cast_fp16")]; + string _SplitHeadsQ__mh_w_2039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2039_equation_0, values = (var_19285_cast_fp16, var_19084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2039_cast_fp16")]; + string _SplitHeadsQ__mh_w_2041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2041_equation_0, values = (var_19289_cast_fp16, var_19091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2041_cast_fp16")]; + string _SplitHeadsQ__mh_w_2043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2043_equation_0, values = (var_19289_cast_fp16, var_19098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2043_cast_fp16")]; + string _SplitHeadsQ__mh_w_2045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2045_equation_0, values = (var_19289_cast_fp16, var_19105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2045_cast_fp16")]; + string _SplitHeadsQ__mh_w_2047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2047_equation_0, values = (var_19289_cast_fp16, var_19112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2047_cast_fp16")]; + string _SplitHeadsQ__mh_w_2049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2049_equation_0, values = (var_19293_cast_fp16, var_19119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2049_cast_fp16")]; + string _SplitHeadsQ__mh_w_2051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2051_equation_0, values = (var_19293_cast_fp16, var_19126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2051_cast_fp16")]; + string _SplitHeadsQ__mh_w_2053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2053_equation_0, values = (var_19293_cast_fp16, var_19133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2053_cast_fp16")]; + string _SplitHeadsQ__mh_w_2055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2055_equation_0, values = (var_19293_cast_fp16, var_19140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2055_cast_fp16")]; + string _SplitHeadsQ__mh_w_2057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2057_equation_0, values = (var_19297_cast_fp16, var_19147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2057_cast_fp16")]; + string _SplitHeadsQ__mh_w_2059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2059_equation_0, values = (var_19297_cast_fp16, var_19154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2059_cast_fp16")]; + string _SplitHeadsQ__mh_w_2061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2061_equation_0, values = (var_19297_cast_fp16, var_19161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2061_cast_fp16")]; + string _SplitHeadsQ__mh_w_2063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2063_equation_0, values = (var_19297_cast_fp16, var_19168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2063_cast_fp16")]; + string _SplitHeadsQ__mh_w_2065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2065_equation_0, values = (var_19301_cast_fp16, var_19175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2065_cast_fp16")]; + string _SplitHeadsQ__mh_w_2067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2067_equation_0, values = (var_19301_cast_fp16, var_19182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2067_cast_fp16")]; + string _SplitHeadsQ__mh_w_2069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2069_equation_0, values = (var_19301_cast_fp16, var_19189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2069_cast_fp16")]; + string _SplitHeadsQ__mh_w_2071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2071_equation_0, values = (var_19301_cast_fp16, var_19196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2071_cast_fp16")]; + string _SplitHeadsQ__mh_w_2073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2073_equation_0, values = (var_19305_cast_fp16, var_19203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2073_cast_fp16")]; + string _SplitHeadsQ__mh_w_2075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2075_equation_0, values = (var_19305_cast_fp16, var_19210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2075_cast_fp16")]; + string _SplitHeadsQ__mh_w_2077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2077_equation_0, values = (var_19305_cast_fp16, var_19217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2077_cast_fp16")]; + string _SplitHeadsQ__mh_w_2079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2079_equation_0, values = (var_19305_cast_fp16, var_19224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2079_cast_fp16")]; + fp16 var_19546_to_fp16 = const()[name = string("op_19546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1921_cast_fp16, y = var_19546_to_fp16)[name = string("aw_chunk_1921_cast_fp16")]; + fp16 var_19548_to_fp16 = const()[name = string("op_19548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1923_cast_fp16, y = var_19548_to_fp16)[name = string("aw_chunk_1923_cast_fp16")]; + fp16 var_19550_to_fp16 = const()[name = string("op_19550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1925_cast_fp16, y = var_19550_to_fp16)[name = string("aw_chunk_1925_cast_fp16")]; + fp16 var_19552_to_fp16 = const()[name = string("op_19552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1927_cast_fp16, y = var_19552_to_fp16)[name = string("aw_chunk_1927_cast_fp16")]; + fp16 var_19554_to_fp16 = const()[name = string("op_19554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1929_cast_fp16, y = var_19554_to_fp16)[name = string("aw_chunk_1929_cast_fp16")]; + fp16 var_19556_to_fp16 = const()[name = string("op_19556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1931_cast_fp16, y = var_19556_to_fp16)[name = string("aw_chunk_1931_cast_fp16")]; + fp16 var_19558_to_fp16 = const()[name = string("op_19558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1933_cast_fp16, y = var_19558_to_fp16)[name = string("aw_chunk_1933_cast_fp16")]; + fp16 var_19560_to_fp16 = const()[name = string("op_19560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1935_cast_fp16, y = var_19560_to_fp16)[name = string("aw_chunk_1935_cast_fp16")]; + fp16 var_19562_to_fp16 = const()[name = string("op_19562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1937_cast_fp16, y = var_19562_to_fp16)[name = string("aw_chunk_1937_cast_fp16")]; + fp16 var_19564_to_fp16 = const()[name = string("op_19564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1939_cast_fp16, y = var_19564_to_fp16)[name = string("aw_chunk_1939_cast_fp16")]; + fp16 var_19566_to_fp16 = const()[name = string("op_19566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1941_cast_fp16, y = var_19566_to_fp16)[name = string("aw_chunk_1941_cast_fp16")]; + fp16 var_19568_to_fp16 = const()[name = string("op_19568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1943_cast_fp16, y = var_19568_to_fp16)[name = string("aw_chunk_1943_cast_fp16")]; + fp16 var_19570_to_fp16 = const()[name = string("op_19570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1945_cast_fp16, y = var_19570_to_fp16)[name = string("aw_chunk_1945_cast_fp16")]; + fp16 var_19572_to_fp16 = const()[name = string("op_19572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1947_cast_fp16, y = var_19572_to_fp16)[name = string("aw_chunk_1947_cast_fp16")]; + fp16 var_19574_to_fp16 = const()[name = string("op_19574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1949_cast_fp16, y = var_19574_to_fp16)[name = string("aw_chunk_1949_cast_fp16")]; + fp16 var_19576_to_fp16 = const()[name = string("op_19576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1951_cast_fp16, y = var_19576_to_fp16)[name = string("aw_chunk_1951_cast_fp16")]; + fp16 var_19578_to_fp16 = const()[name = string("op_19578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1953_cast_fp16, y = var_19578_to_fp16)[name = string("aw_chunk_1953_cast_fp16")]; + fp16 var_19580_to_fp16 = const()[name = string("op_19580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1955_cast_fp16, y = var_19580_to_fp16)[name = string("aw_chunk_1955_cast_fp16")]; + fp16 var_19582_to_fp16 = const()[name = string("op_19582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1957_cast_fp16, y = var_19582_to_fp16)[name = string("aw_chunk_1957_cast_fp16")]; + fp16 var_19584_to_fp16 = const()[name = string("op_19584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1959_cast_fp16, y = var_19584_to_fp16)[name = string("aw_chunk_1959_cast_fp16")]; + fp16 var_19586_to_fp16 = const()[name = string("op_19586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1961_cast_fp16, y = var_19586_to_fp16)[name = string("aw_chunk_1961_cast_fp16")]; + fp16 var_19588_to_fp16 = const()[name = string("op_19588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1963_cast_fp16, y = var_19588_to_fp16)[name = string("aw_chunk_1963_cast_fp16")]; + fp16 var_19590_to_fp16 = const()[name = string("op_19590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1965_cast_fp16, y = var_19590_to_fp16)[name = string("aw_chunk_1965_cast_fp16")]; + fp16 var_19592_to_fp16 = const()[name = string("op_19592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1967_cast_fp16, y = var_19592_to_fp16)[name = string("aw_chunk_1967_cast_fp16")]; + fp16 var_19594_to_fp16 = const()[name = string("op_19594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1969_cast_fp16, y = var_19594_to_fp16)[name = string("aw_chunk_1969_cast_fp16")]; + fp16 var_19596_to_fp16 = const()[name = string("op_19596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1971_cast_fp16, y = var_19596_to_fp16)[name = string("aw_chunk_1971_cast_fp16")]; + fp16 var_19598_to_fp16 = const()[name = string("op_19598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1973_cast_fp16, y = var_19598_to_fp16)[name = string("aw_chunk_1973_cast_fp16")]; + fp16 var_19600_to_fp16 = const()[name = string("op_19600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1975_cast_fp16, y = var_19600_to_fp16)[name = string("aw_chunk_1975_cast_fp16")]; + fp16 var_19602_to_fp16 = const()[name = string("op_19602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1977_cast_fp16, y = var_19602_to_fp16)[name = string("aw_chunk_1977_cast_fp16")]; + fp16 var_19604_to_fp16 = const()[name = string("op_19604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1979_cast_fp16, y = var_19604_to_fp16)[name = string("aw_chunk_1979_cast_fp16")]; + fp16 var_19606_to_fp16 = const()[name = string("op_19606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1981_cast_fp16, y = var_19606_to_fp16)[name = string("aw_chunk_1981_cast_fp16")]; + fp16 var_19608_to_fp16 = const()[name = string("op_19608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1983_cast_fp16, y = var_19608_to_fp16)[name = string("aw_chunk_1983_cast_fp16")]; + fp16 var_19610_to_fp16 = const()[name = string("op_19610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1985_cast_fp16, y = var_19610_to_fp16)[name = string("aw_chunk_1985_cast_fp16")]; + fp16 var_19612_to_fp16 = const()[name = string("op_19612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1987_cast_fp16, y = var_19612_to_fp16)[name = string("aw_chunk_1987_cast_fp16")]; + fp16 var_19614_to_fp16 = const()[name = string("op_19614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1989_cast_fp16, y = var_19614_to_fp16)[name = string("aw_chunk_1989_cast_fp16")]; + fp16 var_19616_to_fp16 = const()[name = string("op_19616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1991_cast_fp16, y = var_19616_to_fp16)[name = string("aw_chunk_1991_cast_fp16")]; + fp16 var_19618_to_fp16 = const()[name = string("op_19618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1993_cast_fp16, y = var_19618_to_fp16)[name = string("aw_chunk_1993_cast_fp16")]; + fp16 var_19620_to_fp16 = const()[name = string("op_19620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1995_cast_fp16, y = var_19620_to_fp16)[name = string("aw_chunk_1995_cast_fp16")]; + fp16 var_19622_to_fp16 = const()[name = string("op_19622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1997_cast_fp16, y = var_19622_to_fp16)[name = string("aw_chunk_1997_cast_fp16")]; + fp16 var_19624_to_fp16 = const()[name = string("op_19624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1999_cast_fp16, y = var_19624_to_fp16)[name = string("aw_chunk_1999_cast_fp16")]; + fp16 var_19626_to_fp16 = const()[name = string("op_19626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2001_cast_fp16, y = var_19626_to_fp16)[name = string("aw_chunk_2001_cast_fp16")]; + fp16 var_19628_to_fp16 = const()[name = string("op_19628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2003_cast_fp16, y = var_19628_to_fp16)[name = string("aw_chunk_2003_cast_fp16")]; + fp16 var_19630_to_fp16 = const()[name = string("op_19630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2005_cast_fp16, y = var_19630_to_fp16)[name = string("aw_chunk_2005_cast_fp16")]; + fp16 var_19632_to_fp16 = const()[name = string("op_19632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2007_cast_fp16, y = var_19632_to_fp16)[name = string("aw_chunk_2007_cast_fp16")]; + fp16 var_19634_to_fp16 = const()[name = string("op_19634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2009_cast_fp16, y = var_19634_to_fp16)[name = string("aw_chunk_2009_cast_fp16")]; + fp16 var_19636_to_fp16 = const()[name = string("op_19636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2011_cast_fp16, y = var_19636_to_fp16)[name = string("aw_chunk_2011_cast_fp16")]; + fp16 var_19638_to_fp16 = const()[name = string("op_19638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2013_cast_fp16, y = var_19638_to_fp16)[name = string("aw_chunk_2013_cast_fp16")]; + fp16 var_19640_to_fp16 = const()[name = string("op_19640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2015_cast_fp16, y = var_19640_to_fp16)[name = string("aw_chunk_2015_cast_fp16")]; + fp16 var_19642_to_fp16 = const()[name = string("op_19642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2017_cast_fp16, y = var_19642_to_fp16)[name = string("aw_chunk_2017_cast_fp16")]; + fp16 var_19644_to_fp16 = const()[name = string("op_19644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2019_cast_fp16, y = var_19644_to_fp16)[name = string("aw_chunk_2019_cast_fp16")]; + fp16 var_19646_to_fp16 = const()[name = string("op_19646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2021_cast_fp16, y = var_19646_to_fp16)[name = string("aw_chunk_2021_cast_fp16")]; + fp16 var_19648_to_fp16 = const()[name = string("op_19648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2023_cast_fp16, y = var_19648_to_fp16)[name = string("aw_chunk_2023_cast_fp16")]; + fp16 var_19650_to_fp16 = const()[name = string("op_19650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2025_cast_fp16, y = var_19650_to_fp16)[name = string("aw_chunk_2025_cast_fp16")]; + fp16 var_19652_to_fp16 = const()[name = string("op_19652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2027_cast_fp16, y = var_19652_to_fp16)[name = string("aw_chunk_2027_cast_fp16")]; + fp16 var_19654_to_fp16 = const()[name = string("op_19654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2029_cast_fp16, y = var_19654_to_fp16)[name = string("aw_chunk_2029_cast_fp16")]; + fp16 var_19656_to_fp16 = const()[name = string("op_19656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2031_cast_fp16, y = var_19656_to_fp16)[name = string("aw_chunk_2031_cast_fp16")]; + fp16 var_19658_to_fp16 = const()[name = string("op_19658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2033_cast_fp16, y = var_19658_to_fp16)[name = string("aw_chunk_2033_cast_fp16")]; + fp16 var_19660_to_fp16 = const()[name = string("op_19660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2035_cast_fp16, y = var_19660_to_fp16)[name = string("aw_chunk_2035_cast_fp16")]; + fp16 var_19662_to_fp16 = const()[name = string("op_19662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2037_cast_fp16, y = var_19662_to_fp16)[name = string("aw_chunk_2037_cast_fp16")]; + fp16 var_19664_to_fp16 = const()[name = string("op_19664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2039_cast_fp16, y = var_19664_to_fp16)[name = string("aw_chunk_2039_cast_fp16")]; + fp16 var_19666_to_fp16 = const()[name = string("op_19666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2041_cast_fp16, y = var_19666_to_fp16)[name = string("aw_chunk_2041_cast_fp16")]; + fp16 var_19668_to_fp16 = const()[name = string("op_19668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2043_cast_fp16, y = var_19668_to_fp16)[name = string("aw_chunk_2043_cast_fp16")]; + fp16 var_19670_to_fp16 = const()[name = string("op_19670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2045_cast_fp16, y = var_19670_to_fp16)[name = string("aw_chunk_2045_cast_fp16")]; + fp16 var_19672_to_fp16 = const()[name = string("op_19672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2047_cast_fp16, y = var_19672_to_fp16)[name = string("aw_chunk_2047_cast_fp16")]; + fp16 var_19674_to_fp16 = const()[name = string("op_19674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2049_cast_fp16, y = var_19674_to_fp16)[name = string("aw_chunk_2049_cast_fp16")]; + fp16 var_19676_to_fp16 = const()[name = string("op_19676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2051_cast_fp16, y = var_19676_to_fp16)[name = string("aw_chunk_2051_cast_fp16")]; + fp16 var_19678_to_fp16 = const()[name = string("op_19678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2053_cast_fp16, y = var_19678_to_fp16)[name = string("aw_chunk_2053_cast_fp16")]; + fp16 var_19680_to_fp16 = const()[name = string("op_19680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2055_cast_fp16, y = var_19680_to_fp16)[name = string("aw_chunk_2055_cast_fp16")]; + fp16 var_19682_to_fp16 = const()[name = string("op_19682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2057_cast_fp16, y = var_19682_to_fp16)[name = string("aw_chunk_2057_cast_fp16")]; + fp16 var_19684_to_fp16 = const()[name = string("op_19684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2059_cast_fp16, y = var_19684_to_fp16)[name = string("aw_chunk_2059_cast_fp16")]; + fp16 var_19686_to_fp16 = const()[name = string("op_19686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2061_cast_fp16, y = var_19686_to_fp16)[name = string("aw_chunk_2061_cast_fp16")]; + fp16 var_19688_to_fp16 = const()[name = string("op_19688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2063_cast_fp16, y = var_19688_to_fp16)[name = string("aw_chunk_2063_cast_fp16")]; + fp16 var_19690_to_fp16 = const()[name = string("op_19690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2065_cast_fp16, y = var_19690_to_fp16)[name = string("aw_chunk_2065_cast_fp16")]; + fp16 var_19692_to_fp16 = const()[name = string("op_19692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2067_cast_fp16, y = var_19692_to_fp16)[name = string("aw_chunk_2067_cast_fp16")]; + fp16 var_19694_to_fp16 = const()[name = string("op_19694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2069_cast_fp16, y = var_19694_to_fp16)[name = string("aw_chunk_2069_cast_fp16")]; + fp16 var_19696_to_fp16 = const()[name = string("op_19696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2071_cast_fp16, y = var_19696_to_fp16)[name = string("aw_chunk_2071_cast_fp16")]; + fp16 var_19698_to_fp16 = const()[name = string("op_19698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2073_cast_fp16, y = var_19698_to_fp16)[name = string("aw_chunk_2073_cast_fp16")]; + fp16 var_19700_to_fp16 = const()[name = string("op_19700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2075_cast_fp16, y = var_19700_to_fp16)[name = string("aw_chunk_2075_cast_fp16")]; + fp16 var_19702_to_fp16 = const()[name = string("op_19702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2077_cast_fp16, y = var_19702_to_fp16)[name = string("aw_chunk_2077_cast_fp16")]; + fp16 var_19704_to_fp16 = const()[name = string("op_19704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2079_cast_fp16, y = var_19704_to_fp16)[name = string("aw_chunk_2079_cast_fp16")]; + tensor var_19706_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1921_cast_fp16)[name = string("op_19706_cast_fp16")]; + tensor var_19707_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1923_cast_fp16)[name = string("op_19707_cast_fp16")]; + tensor var_19708_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1925_cast_fp16)[name = string("op_19708_cast_fp16")]; + tensor var_19709_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1927_cast_fp16)[name = string("op_19709_cast_fp16")]; + tensor var_19710_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1929_cast_fp16)[name = string("op_19710_cast_fp16")]; + tensor var_19711_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1931_cast_fp16)[name = string("op_19711_cast_fp16")]; + tensor var_19712_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1933_cast_fp16)[name = string("op_19712_cast_fp16")]; + tensor var_19713_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1935_cast_fp16)[name = string("op_19713_cast_fp16")]; + tensor var_19714_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1937_cast_fp16)[name = string("op_19714_cast_fp16")]; + tensor var_19715_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1939_cast_fp16)[name = string("op_19715_cast_fp16")]; + tensor var_19716_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1941_cast_fp16)[name = string("op_19716_cast_fp16")]; + tensor var_19717_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1943_cast_fp16)[name = string("op_19717_cast_fp16")]; + tensor var_19718_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1945_cast_fp16)[name = string("op_19718_cast_fp16")]; + tensor var_19719_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1947_cast_fp16)[name = string("op_19719_cast_fp16")]; + tensor var_19720_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1949_cast_fp16)[name = string("op_19720_cast_fp16")]; + tensor var_19721_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1951_cast_fp16)[name = string("op_19721_cast_fp16")]; + tensor var_19722_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1953_cast_fp16)[name = string("op_19722_cast_fp16")]; + tensor var_19723_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1955_cast_fp16)[name = string("op_19723_cast_fp16")]; + tensor var_19724_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1957_cast_fp16)[name = string("op_19724_cast_fp16")]; + tensor var_19725_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1959_cast_fp16)[name = string("op_19725_cast_fp16")]; + tensor var_19726_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1961_cast_fp16)[name = string("op_19726_cast_fp16")]; + tensor var_19727_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1963_cast_fp16)[name = string("op_19727_cast_fp16")]; + tensor var_19728_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1965_cast_fp16)[name = string("op_19728_cast_fp16")]; + tensor var_19729_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1967_cast_fp16)[name = string("op_19729_cast_fp16")]; + tensor var_19730_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1969_cast_fp16)[name = string("op_19730_cast_fp16")]; + tensor var_19731_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1971_cast_fp16)[name = string("op_19731_cast_fp16")]; + tensor var_19732_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1973_cast_fp16)[name = string("op_19732_cast_fp16")]; + tensor var_19733_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1975_cast_fp16)[name = string("op_19733_cast_fp16")]; + tensor var_19734_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1977_cast_fp16)[name = string("op_19734_cast_fp16")]; + tensor var_19735_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1979_cast_fp16)[name = string("op_19735_cast_fp16")]; + tensor var_19736_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1981_cast_fp16)[name = string("op_19736_cast_fp16")]; + tensor var_19737_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1983_cast_fp16)[name = string("op_19737_cast_fp16")]; + tensor var_19738_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1985_cast_fp16)[name = string("op_19738_cast_fp16")]; + tensor var_19739_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1987_cast_fp16)[name = string("op_19739_cast_fp16")]; + tensor var_19740_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1989_cast_fp16)[name = string("op_19740_cast_fp16")]; + tensor var_19741_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1991_cast_fp16)[name = string("op_19741_cast_fp16")]; + tensor var_19742_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1993_cast_fp16)[name = string("op_19742_cast_fp16")]; + tensor var_19743_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1995_cast_fp16)[name = string("op_19743_cast_fp16")]; + tensor var_19744_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1997_cast_fp16)[name = string("op_19744_cast_fp16")]; + tensor var_19745_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_1999_cast_fp16)[name = string("op_19745_cast_fp16")]; + tensor var_19746_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2001_cast_fp16)[name = string("op_19746_cast_fp16")]; + tensor var_19747_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2003_cast_fp16)[name = string("op_19747_cast_fp16")]; + tensor var_19748_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2005_cast_fp16)[name = string("op_19748_cast_fp16")]; + tensor var_19749_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2007_cast_fp16)[name = string("op_19749_cast_fp16")]; + tensor var_19750_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2009_cast_fp16)[name = string("op_19750_cast_fp16")]; + tensor var_19751_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2011_cast_fp16)[name = string("op_19751_cast_fp16")]; + tensor var_19752_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2013_cast_fp16)[name = string("op_19752_cast_fp16")]; + tensor var_19753_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2015_cast_fp16)[name = string("op_19753_cast_fp16")]; + tensor var_19754_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2017_cast_fp16)[name = string("op_19754_cast_fp16")]; + tensor var_19755_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2019_cast_fp16)[name = string("op_19755_cast_fp16")]; + tensor var_19756_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2021_cast_fp16)[name = string("op_19756_cast_fp16")]; + tensor var_19757_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2023_cast_fp16)[name = string("op_19757_cast_fp16")]; + tensor var_19758_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2025_cast_fp16)[name = string("op_19758_cast_fp16")]; + tensor var_19759_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2027_cast_fp16)[name = string("op_19759_cast_fp16")]; + tensor var_19760_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2029_cast_fp16)[name = string("op_19760_cast_fp16")]; + tensor var_19761_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2031_cast_fp16)[name = string("op_19761_cast_fp16")]; + tensor var_19762_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2033_cast_fp16)[name = string("op_19762_cast_fp16")]; + tensor var_19763_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2035_cast_fp16)[name = string("op_19763_cast_fp16")]; + tensor var_19764_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2037_cast_fp16)[name = string("op_19764_cast_fp16")]; + tensor var_19765_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2039_cast_fp16)[name = string("op_19765_cast_fp16")]; + tensor var_19766_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2041_cast_fp16)[name = string("op_19766_cast_fp16")]; + tensor var_19767_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2043_cast_fp16)[name = string("op_19767_cast_fp16")]; + tensor var_19768_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2045_cast_fp16)[name = string("op_19768_cast_fp16")]; + tensor var_19769_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2047_cast_fp16)[name = string("op_19769_cast_fp16")]; + tensor var_19770_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2049_cast_fp16)[name = string("op_19770_cast_fp16")]; + tensor var_19771_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2051_cast_fp16)[name = string("op_19771_cast_fp16")]; + tensor var_19772_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2053_cast_fp16)[name = string("op_19772_cast_fp16")]; + tensor var_19773_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2055_cast_fp16)[name = string("op_19773_cast_fp16")]; + tensor var_19774_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2057_cast_fp16)[name = string("op_19774_cast_fp16")]; + tensor var_19775_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2059_cast_fp16)[name = string("op_19775_cast_fp16")]; + tensor var_19776_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2061_cast_fp16)[name = string("op_19776_cast_fp16")]; + tensor var_19777_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2063_cast_fp16)[name = string("op_19777_cast_fp16")]; + tensor var_19778_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2065_cast_fp16)[name = string("op_19778_cast_fp16")]; + tensor var_19779_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2067_cast_fp16)[name = string("op_19779_cast_fp16")]; + tensor var_19780_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2069_cast_fp16)[name = string("op_19780_cast_fp16")]; + tensor var_19781_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2071_cast_fp16)[name = string("op_19781_cast_fp16")]; + tensor var_19782_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2073_cast_fp16)[name = string("op_19782_cast_fp16")]; + tensor var_19783_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2075_cast_fp16)[name = string("op_19783_cast_fp16")]; + tensor var_19784_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2077_cast_fp16)[name = string("op_19784_cast_fp16")]; + tensor var_19785_cast_fp16 = softmax(axis = var_18531, x = aw_chunk_2079_cast_fp16)[name = string("op_19785_cast_fp16")]; + string var_19787_equation_0 = const()[name = string("op_19787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19787_cast_fp16 = einsum(equation = var_19787_equation_0, values = (var_19307_cast_fp16, var_19706_cast_fp16))[name = string("op_19787_cast_fp16")]; + string var_19789_equation_0 = const()[name = string("op_19789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19789_cast_fp16 = einsum(equation = var_19789_equation_0, values = (var_19307_cast_fp16, var_19707_cast_fp16))[name = string("op_19789_cast_fp16")]; + string var_19791_equation_0 = const()[name = string("op_19791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19791_cast_fp16 = einsum(equation = var_19791_equation_0, values = (var_19307_cast_fp16, var_19708_cast_fp16))[name = string("op_19791_cast_fp16")]; + string var_19793_equation_0 = const()[name = string("op_19793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19793_cast_fp16 = einsum(equation = var_19793_equation_0, values = (var_19307_cast_fp16, var_19709_cast_fp16))[name = string("op_19793_cast_fp16")]; + string var_19795_equation_0 = const()[name = string("op_19795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19795_cast_fp16 = einsum(equation = var_19795_equation_0, values = (var_19311_cast_fp16, var_19710_cast_fp16))[name = string("op_19795_cast_fp16")]; + string var_19797_equation_0 = const()[name = string("op_19797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19797_cast_fp16 = einsum(equation = var_19797_equation_0, values = (var_19311_cast_fp16, var_19711_cast_fp16))[name = string("op_19797_cast_fp16")]; + string var_19799_equation_0 = const()[name = string("op_19799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19799_cast_fp16 = einsum(equation = var_19799_equation_0, values = (var_19311_cast_fp16, var_19712_cast_fp16))[name = string("op_19799_cast_fp16")]; + string var_19801_equation_0 = const()[name = string("op_19801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19801_cast_fp16 = einsum(equation = var_19801_equation_0, values = (var_19311_cast_fp16, var_19713_cast_fp16))[name = string("op_19801_cast_fp16")]; + string var_19803_equation_0 = const()[name = string("op_19803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19803_cast_fp16 = einsum(equation = var_19803_equation_0, values = (var_19315_cast_fp16, var_19714_cast_fp16))[name = string("op_19803_cast_fp16")]; + string var_19805_equation_0 = const()[name = string("op_19805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19805_cast_fp16 = einsum(equation = var_19805_equation_0, values = (var_19315_cast_fp16, var_19715_cast_fp16))[name = string("op_19805_cast_fp16")]; + string var_19807_equation_0 = const()[name = string("op_19807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19807_cast_fp16 = einsum(equation = var_19807_equation_0, values = (var_19315_cast_fp16, var_19716_cast_fp16))[name = string("op_19807_cast_fp16")]; + string var_19809_equation_0 = const()[name = string("op_19809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19809_cast_fp16 = einsum(equation = var_19809_equation_0, values = (var_19315_cast_fp16, var_19717_cast_fp16))[name = string("op_19809_cast_fp16")]; + string var_19811_equation_0 = const()[name = string("op_19811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19811_cast_fp16 = einsum(equation = var_19811_equation_0, values = (var_19319_cast_fp16, var_19718_cast_fp16))[name = string("op_19811_cast_fp16")]; + string var_19813_equation_0 = const()[name = string("op_19813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19813_cast_fp16 = einsum(equation = var_19813_equation_0, values = (var_19319_cast_fp16, var_19719_cast_fp16))[name = string("op_19813_cast_fp16")]; + string var_19815_equation_0 = const()[name = string("op_19815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19815_cast_fp16 = einsum(equation = var_19815_equation_0, values = (var_19319_cast_fp16, var_19720_cast_fp16))[name = string("op_19815_cast_fp16")]; + string var_19817_equation_0 = const()[name = string("op_19817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19817_cast_fp16 = einsum(equation = var_19817_equation_0, values = (var_19319_cast_fp16, var_19721_cast_fp16))[name = string("op_19817_cast_fp16")]; + string var_19819_equation_0 = const()[name = string("op_19819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19819_cast_fp16 = einsum(equation = var_19819_equation_0, values = (var_19323_cast_fp16, var_19722_cast_fp16))[name = string("op_19819_cast_fp16")]; + string var_19821_equation_0 = const()[name = string("op_19821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19821_cast_fp16 = einsum(equation = var_19821_equation_0, values = (var_19323_cast_fp16, var_19723_cast_fp16))[name = string("op_19821_cast_fp16")]; + string var_19823_equation_0 = const()[name = string("op_19823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19823_cast_fp16 = einsum(equation = var_19823_equation_0, values = (var_19323_cast_fp16, var_19724_cast_fp16))[name = string("op_19823_cast_fp16")]; + string var_19825_equation_0 = const()[name = string("op_19825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19825_cast_fp16 = einsum(equation = var_19825_equation_0, values = (var_19323_cast_fp16, var_19725_cast_fp16))[name = string("op_19825_cast_fp16")]; + string var_19827_equation_0 = const()[name = string("op_19827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19827_cast_fp16 = einsum(equation = var_19827_equation_0, values = (var_19327_cast_fp16, var_19726_cast_fp16))[name = string("op_19827_cast_fp16")]; + string var_19829_equation_0 = const()[name = string("op_19829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19829_cast_fp16 = einsum(equation = var_19829_equation_0, values = (var_19327_cast_fp16, var_19727_cast_fp16))[name = string("op_19829_cast_fp16")]; + string var_19831_equation_0 = const()[name = string("op_19831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19831_cast_fp16 = einsum(equation = var_19831_equation_0, values = (var_19327_cast_fp16, var_19728_cast_fp16))[name = string("op_19831_cast_fp16")]; + string var_19833_equation_0 = const()[name = string("op_19833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19833_cast_fp16 = einsum(equation = var_19833_equation_0, values = (var_19327_cast_fp16, var_19729_cast_fp16))[name = string("op_19833_cast_fp16")]; + string var_19835_equation_0 = const()[name = string("op_19835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19835_cast_fp16 = einsum(equation = var_19835_equation_0, values = (var_19331_cast_fp16, var_19730_cast_fp16))[name = string("op_19835_cast_fp16")]; + string var_19837_equation_0 = const()[name = string("op_19837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19837_cast_fp16 = einsum(equation = var_19837_equation_0, values = (var_19331_cast_fp16, var_19731_cast_fp16))[name = string("op_19837_cast_fp16")]; + string var_19839_equation_0 = const()[name = string("op_19839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19839_cast_fp16 = einsum(equation = var_19839_equation_0, values = (var_19331_cast_fp16, var_19732_cast_fp16))[name = string("op_19839_cast_fp16")]; + string var_19841_equation_0 = const()[name = string("op_19841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19841_cast_fp16 = einsum(equation = var_19841_equation_0, values = (var_19331_cast_fp16, var_19733_cast_fp16))[name = string("op_19841_cast_fp16")]; + string var_19843_equation_0 = const()[name = string("op_19843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19843_cast_fp16 = einsum(equation = var_19843_equation_0, values = (var_19335_cast_fp16, var_19734_cast_fp16))[name = string("op_19843_cast_fp16")]; + string var_19845_equation_0 = const()[name = string("op_19845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19845_cast_fp16 = einsum(equation = var_19845_equation_0, values = (var_19335_cast_fp16, var_19735_cast_fp16))[name = string("op_19845_cast_fp16")]; + string var_19847_equation_0 = const()[name = string("op_19847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19847_cast_fp16 = einsum(equation = var_19847_equation_0, values = (var_19335_cast_fp16, var_19736_cast_fp16))[name = string("op_19847_cast_fp16")]; + string var_19849_equation_0 = const()[name = string("op_19849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19849_cast_fp16 = einsum(equation = var_19849_equation_0, values = (var_19335_cast_fp16, var_19737_cast_fp16))[name = string("op_19849_cast_fp16")]; + string var_19851_equation_0 = const()[name = string("op_19851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19851_cast_fp16 = einsum(equation = var_19851_equation_0, values = (var_19339_cast_fp16, var_19738_cast_fp16))[name = string("op_19851_cast_fp16")]; + string var_19853_equation_0 = const()[name = string("op_19853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19853_cast_fp16 = einsum(equation = var_19853_equation_0, values = (var_19339_cast_fp16, var_19739_cast_fp16))[name = string("op_19853_cast_fp16")]; + string var_19855_equation_0 = const()[name = string("op_19855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19855_cast_fp16 = einsum(equation = var_19855_equation_0, values = (var_19339_cast_fp16, var_19740_cast_fp16))[name = string("op_19855_cast_fp16")]; + string var_19857_equation_0 = const()[name = string("op_19857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19857_cast_fp16 = einsum(equation = var_19857_equation_0, values = (var_19339_cast_fp16, var_19741_cast_fp16))[name = string("op_19857_cast_fp16")]; + string var_19859_equation_0 = const()[name = string("op_19859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19859_cast_fp16 = einsum(equation = var_19859_equation_0, values = (var_19343_cast_fp16, var_19742_cast_fp16))[name = string("op_19859_cast_fp16")]; + string var_19861_equation_0 = const()[name = string("op_19861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19861_cast_fp16 = einsum(equation = var_19861_equation_0, values = (var_19343_cast_fp16, var_19743_cast_fp16))[name = string("op_19861_cast_fp16")]; + string var_19863_equation_0 = const()[name = string("op_19863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19863_cast_fp16 = einsum(equation = var_19863_equation_0, values = (var_19343_cast_fp16, var_19744_cast_fp16))[name = string("op_19863_cast_fp16")]; + string var_19865_equation_0 = const()[name = string("op_19865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19865_cast_fp16 = einsum(equation = var_19865_equation_0, values = (var_19343_cast_fp16, var_19745_cast_fp16))[name = string("op_19865_cast_fp16")]; + string var_19867_equation_0 = const()[name = string("op_19867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19867_cast_fp16 = einsum(equation = var_19867_equation_0, values = (var_19347_cast_fp16, var_19746_cast_fp16))[name = string("op_19867_cast_fp16")]; + string var_19869_equation_0 = const()[name = string("op_19869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19869_cast_fp16 = einsum(equation = var_19869_equation_0, values = (var_19347_cast_fp16, var_19747_cast_fp16))[name = string("op_19869_cast_fp16")]; + string var_19871_equation_0 = const()[name = string("op_19871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19871_cast_fp16 = einsum(equation = var_19871_equation_0, values = (var_19347_cast_fp16, var_19748_cast_fp16))[name = string("op_19871_cast_fp16")]; + string var_19873_equation_0 = const()[name = string("op_19873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19873_cast_fp16 = einsum(equation = var_19873_equation_0, values = (var_19347_cast_fp16, var_19749_cast_fp16))[name = string("op_19873_cast_fp16")]; + string var_19875_equation_0 = const()[name = string("op_19875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19875_cast_fp16 = einsum(equation = var_19875_equation_0, values = (var_19351_cast_fp16, var_19750_cast_fp16))[name = string("op_19875_cast_fp16")]; + string var_19877_equation_0 = const()[name = string("op_19877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19877_cast_fp16 = einsum(equation = var_19877_equation_0, values = (var_19351_cast_fp16, var_19751_cast_fp16))[name = string("op_19877_cast_fp16")]; + string var_19879_equation_0 = const()[name = string("op_19879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19879_cast_fp16 = einsum(equation = var_19879_equation_0, values = (var_19351_cast_fp16, var_19752_cast_fp16))[name = string("op_19879_cast_fp16")]; + string var_19881_equation_0 = const()[name = string("op_19881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19881_cast_fp16 = einsum(equation = var_19881_equation_0, values = (var_19351_cast_fp16, var_19753_cast_fp16))[name = string("op_19881_cast_fp16")]; + string var_19883_equation_0 = const()[name = string("op_19883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19883_cast_fp16 = einsum(equation = var_19883_equation_0, values = (var_19355_cast_fp16, var_19754_cast_fp16))[name = string("op_19883_cast_fp16")]; + string var_19885_equation_0 = const()[name = string("op_19885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19885_cast_fp16 = einsum(equation = var_19885_equation_0, values = (var_19355_cast_fp16, var_19755_cast_fp16))[name = string("op_19885_cast_fp16")]; + string var_19887_equation_0 = const()[name = string("op_19887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19887_cast_fp16 = einsum(equation = var_19887_equation_0, values = (var_19355_cast_fp16, var_19756_cast_fp16))[name = string("op_19887_cast_fp16")]; + string var_19889_equation_0 = const()[name = string("op_19889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19889_cast_fp16 = einsum(equation = var_19889_equation_0, values = (var_19355_cast_fp16, var_19757_cast_fp16))[name = string("op_19889_cast_fp16")]; + string var_19891_equation_0 = const()[name = string("op_19891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19891_cast_fp16 = einsum(equation = var_19891_equation_0, values = (var_19359_cast_fp16, var_19758_cast_fp16))[name = string("op_19891_cast_fp16")]; + string var_19893_equation_0 = const()[name = string("op_19893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19893_cast_fp16 = einsum(equation = var_19893_equation_0, values = (var_19359_cast_fp16, var_19759_cast_fp16))[name = string("op_19893_cast_fp16")]; + string var_19895_equation_0 = const()[name = string("op_19895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19895_cast_fp16 = einsum(equation = var_19895_equation_0, values = (var_19359_cast_fp16, var_19760_cast_fp16))[name = string("op_19895_cast_fp16")]; + string var_19897_equation_0 = const()[name = string("op_19897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19897_cast_fp16 = einsum(equation = var_19897_equation_0, values = (var_19359_cast_fp16, var_19761_cast_fp16))[name = string("op_19897_cast_fp16")]; + string var_19899_equation_0 = const()[name = string("op_19899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19899_cast_fp16 = einsum(equation = var_19899_equation_0, values = (var_19363_cast_fp16, var_19762_cast_fp16))[name = string("op_19899_cast_fp16")]; + string var_19901_equation_0 = const()[name = string("op_19901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19901_cast_fp16 = einsum(equation = var_19901_equation_0, values = (var_19363_cast_fp16, var_19763_cast_fp16))[name = string("op_19901_cast_fp16")]; + string var_19903_equation_0 = const()[name = string("op_19903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19903_cast_fp16 = einsum(equation = var_19903_equation_0, values = (var_19363_cast_fp16, var_19764_cast_fp16))[name = string("op_19903_cast_fp16")]; + string var_19905_equation_0 = const()[name = string("op_19905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19905_cast_fp16 = einsum(equation = var_19905_equation_0, values = (var_19363_cast_fp16, var_19765_cast_fp16))[name = string("op_19905_cast_fp16")]; + string var_19907_equation_0 = const()[name = string("op_19907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19907_cast_fp16 = einsum(equation = var_19907_equation_0, values = (var_19367_cast_fp16, var_19766_cast_fp16))[name = string("op_19907_cast_fp16")]; + string var_19909_equation_0 = const()[name = string("op_19909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19909_cast_fp16 = einsum(equation = var_19909_equation_0, values = (var_19367_cast_fp16, var_19767_cast_fp16))[name = string("op_19909_cast_fp16")]; + string var_19911_equation_0 = const()[name = string("op_19911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19911_cast_fp16 = einsum(equation = var_19911_equation_0, values = (var_19367_cast_fp16, var_19768_cast_fp16))[name = string("op_19911_cast_fp16")]; + string var_19913_equation_0 = const()[name = string("op_19913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19913_cast_fp16 = einsum(equation = var_19913_equation_0, values = (var_19367_cast_fp16, var_19769_cast_fp16))[name = string("op_19913_cast_fp16")]; + string var_19915_equation_0 = const()[name = string("op_19915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19915_cast_fp16 = einsum(equation = var_19915_equation_0, values = (var_19371_cast_fp16, var_19770_cast_fp16))[name = string("op_19915_cast_fp16")]; + string var_19917_equation_0 = const()[name = string("op_19917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19917_cast_fp16 = einsum(equation = var_19917_equation_0, values = (var_19371_cast_fp16, var_19771_cast_fp16))[name = string("op_19917_cast_fp16")]; + string var_19919_equation_0 = const()[name = string("op_19919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19919_cast_fp16 = einsum(equation = var_19919_equation_0, values = (var_19371_cast_fp16, var_19772_cast_fp16))[name = string("op_19919_cast_fp16")]; + string var_19921_equation_0 = const()[name = string("op_19921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19921_cast_fp16 = einsum(equation = var_19921_equation_0, values = (var_19371_cast_fp16, var_19773_cast_fp16))[name = string("op_19921_cast_fp16")]; + string var_19923_equation_0 = const()[name = string("op_19923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19923_cast_fp16 = einsum(equation = var_19923_equation_0, values = (var_19375_cast_fp16, var_19774_cast_fp16))[name = string("op_19923_cast_fp16")]; + string var_19925_equation_0 = const()[name = string("op_19925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19925_cast_fp16 = einsum(equation = var_19925_equation_0, values = (var_19375_cast_fp16, var_19775_cast_fp16))[name = string("op_19925_cast_fp16")]; + string var_19927_equation_0 = const()[name = string("op_19927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19927_cast_fp16 = einsum(equation = var_19927_equation_0, values = (var_19375_cast_fp16, var_19776_cast_fp16))[name = string("op_19927_cast_fp16")]; + string var_19929_equation_0 = const()[name = string("op_19929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19929_cast_fp16 = einsum(equation = var_19929_equation_0, values = (var_19375_cast_fp16, var_19777_cast_fp16))[name = string("op_19929_cast_fp16")]; + string var_19931_equation_0 = const()[name = string("op_19931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19931_cast_fp16 = einsum(equation = var_19931_equation_0, values = (var_19379_cast_fp16, var_19778_cast_fp16))[name = string("op_19931_cast_fp16")]; + string var_19933_equation_0 = const()[name = string("op_19933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19933_cast_fp16 = einsum(equation = var_19933_equation_0, values = (var_19379_cast_fp16, var_19779_cast_fp16))[name = string("op_19933_cast_fp16")]; + string var_19935_equation_0 = const()[name = string("op_19935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19935_cast_fp16 = einsum(equation = var_19935_equation_0, values = (var_19379_cast_fp16, var_19780_cast_fp16))[name = string("op_19935_cast_fp16")]; + string var_19937_equation_0 = const()[name = string("op_19937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19937_cast_fp16 = einsum(equation = var_19937_equation_0, values = (var_19379_cast_fp16, var_19781_cast_fp16))[name = string("op_19937_cast_fp16")]; + string var_19939_equation_0 = const()[name = string("op_19939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19939_cast_fp16 = einsum(equation = var_19939_equation_0, values = (var_19383_cast_fp16, var_19782_cast_fp16))[name = string("op_19939_cast_fp16")]; + string var_19941_equation_0 = const()[name = string("op_19941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19941_cast_fp16 = einsum(equation = var_19941_equation_0, values = (var_19383_cast_fp16, var_19783_cast_fp16))[name = string("op_19941_cast_fp16")]; + string var_19943_equation_0 = const()[name = string("op_19943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19943_cast_fp16 = einsum(equation = var_19943_equation_0, values = (var_19383_cast_fp16, var_19784_cast_fp16))[name = string("op_19943_cast_fp16")]; + string var_19945_equation_0 = const()[name = string("op_19945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_19945_cast_fp16 = einsum(equation = var_19945_equation_0, values = (var_19383_cast_fp16, var_19785_cast_fp16))[name = string("op_19945_cast_fp16")]; + bool var_19947_interleave_0 = const()[name = string("op_19947_interleave_0"), val = bool(false)]; + tensor var_19947_cast_fp16 = concat(axis = var_18506, interleave = var_19947_interleave_0, values = (var_19787_cast_fp16, var_19789_cast_fp16, var_19791_cast_fp16, var_19793_cast_fp16))[name = string("op_19947_cast_fp16")]; + bool var_19949_interleave_0 = const()[name = string("op_19949_interleave_0"), val = bool(false)]; + tensor var_19949_cast_fp16 = concat(axis = var_18506, interleave = var_19949_interleave_0, values = (var_19795_cast_fp16, var_19797_cast_fp16, var_19799_cast_fp16, var_19801_cast_fp16))[name = string("op_19949_cast_fp16")]; + bool var_19951_interleave_0 = const()[name = string("op_19951_interleave_0"), val = bool(false)]; + tensor var_19951_cast_fp16 = concat(axis = var_18506, interleave = var_19951_interleave_0, values = (var_19803_cast_fp16, var_19805_cast_fp16, var_19807_cast_fp16, var_19809_cast_fp16))[name = string("op_19951_cast_fp16")]; + bool var_19953_interleave_0 = const()[name = string("op_19953_interleave_0"), val = bool(false)]; + tensor var_19953_cast_fp16 = concat(axis = var_18506, interleave = var_19953_interleave_0, values = (var_19811_cast_fp16, var_19813_cast_fp16, var_19815_cast_fp16, var_19817_cast_fp16))[name = string("op_19953_cast_fp16")]; + bool var_19955_interleave_0 = const()[name = string("op_19955_interleave_0"), val = bool(false)]; + tensor var_19955_cast_fp16 = concat(axis = var_18506, interleave = var_19955_interleave_0, values = (var_19819_cast_fp16, var_19821_cast_fp16, var_19823_cast_fp16, var_19825_cast_fp16))[name = string("op_19955_cast_fp16")]; + bool var_19957_interleave_0 = const()[name = string("op_19957_interleave_0"), val = bool(false)]; + tensor var_19957_cast_fp16 = concat(axis = var_18506, interleave = var_19957_interleave_0, values = (var_19827_cast_fp16, var_19829_cast_fp16, var_19831_cast_fp16, var_19833_cast_fp16))[name = string("op_19957_cast_fp16")]; + bool var_19959_interleave_0 = const()[name = string("op_19959_interleave_0"), val = bool(false)]; + tensor var_19959_cast_fp16 = concat(axis = var_18506, interleave = var_19959_interleave_0, values = (var_19835_cast_fp16, var_19837_cast_fp16, var_19839_cast_fp16, var_19841_cast_fp16))[name = string("op_19959_cast_fp16")]; + bool var_19961_interleave_0 = const()[name = string("op_19961_interleave_0"), val = bool(false)]; + tensor var_19961_cast_fp16 = concat(axis = var_18506, interleave = var_19961_interleave_0, values = (var_19843_cast_fp16, var_19845_cast_fp16, var_19847_cast_fp16, var_19849_cast_fp16))[name = string("op_19961_cast_fp16")]; + bool var_19963_interleave_0 = const()[name = string("op_19963_interleave_0"), val = bool(false)]; + tensor var_19963_cast_fp16 = concat(axis = var_18506, interleave = var_19963_interleave_0, values = (var_19851_cast_fp16, var_19853_cast_fp16, var_19855_cast_fp16, var_19857_cast_fp16))[name = string("op_19963_cast_fp16")]; + bool var_19965_interleave_0 = const()[name = string("op_19965_interleave_0"), val = bool(false)]; + tensor var_19965_cast_fp16 = concat(axis = var_18506, interleave = var_19965_interleave_0, values = (var_19859_cast_fp16, var_19861_cast_fp16, var_19863_cast_fp16, var_19865_cast_fp16))[name = string("op_19965_cast_fp16")]; + bool var_19967_interleave_0 = const()[name = string("op_19967_interleave_0"), val = bool(false)]; + tensor var_19967_cast_fp16 = concat(axis = var_18506, interleave = var_19967_interleave_0, values = (var_19867_cast_fp16, var_19869_cast_fp16, var_19871_cast_fp16, var_19873_cast_fp16))[name = string("op_19967_cast_fp16")]; + bool var_19969_interleave_0 = const()[name = string("op_19969_interleave_0"), val = bool(false)]; + tensor var_19969_cast_fp16 = concat(axis = var_18506, interleave = var_19969_interleave_0, values = (var_19875_cast_fp16, var_19877_cast_fp16, var_19879_cast_fp16, var_19881_cast_fp16))[name = string("op_19969_cast_fp16")]; + bool var_19971_interleave_0 = const()[name = string("op_19971_interleave_0"), val = bool(false)]; + tensor var_19971_cast_fp16 = concat(axis = var_18506, interleave = var_19971_interleave_0, values = (var_19883_cast_fp16, var_19885_cast_fp16, var_19887_cast_fp16, var_19889_cast_fp16))[name = string("op_19971_cast_fp16")]; + bool var_19973_interleave_0 = const()[name = string("op_19973_interleave_0"), val = bool(false)]; + tensor var_19973_cast_fp16 = concat(axis = var_18506, interleave = var_19973_interleave_0, values = (var_19891_cast_fp16, var_19893_cast_fp16, var_19895_cast_fp16, var_19897_cast_fp16))[name = string("op_19973_cast_fp16")]; + bool var_19975_interleave_0 = const()[name = string("op_19975_interleave_0"), val = bool(false)]; + tensor var_19975_cast_fp16 = concat(axis = var_18506, interleave = var_19975_interleave_0, values = (var_19899_cast_fp16, var_19901_cast_fp16, var_19903_cast_fp16, var_19905_cast_fp16))[name = string("op_19975_cast_fp16")]; + bool var_19977_interleave_0 = const()[name = string("op_19977_interleave_0"), val = bool(false)]; + tensor var_19977_cast_fp16 = concat(axis = var_18506, interleave = var_19977_interleave_0, values = (var_19907_cast_fp16, var_19909_cast_fp16, var_19911_cast_fp16, var_19913_cast_fp16))[name = string("op_19977_cast_fp16")]; + bool var_19979_interleave_0 = const()[name = string("op_19979_interleave_0"), val = bool(false)]; + tensor var_19979_cast_fp16 = concat(axis = var_18506, interleave = var_19979_interleave_0, values = (var_19915_cast_fp16, var_19917_cast_fp16, var_19919_cast_fp16, var_19921_cast_fp16))[name = string("op_19979_cast_fp16")]; + bool var_19981_interleave_0 = const()[name = string("op_19981_interleave_0"), val = bool(false)]; + tensor var_19981_cast_fp16 = concat(axis = var_18506, interleave = var_19981_interleave_0, values = (var_19923_cast_fp16, var_19925_cast_fp16, var_19927_cast_fp16, var_19929_cast_fp16))[name = string("op_19981_cast_fp16")]; + bool var_19983_interleave_0 = const()[name = string("op_19983_interleave_0"), val = bool(false)]; + tensor var_19983_cast_fp16 = concat(axis = var_18506, interleave = var_19983_interleave_0, values = (var_19931_cast_fp16, var_19933_cast_fp16, var_19935_cast_fp16, var_19937_cast_fp16))[name = string("op_19983_cast_fp16")]; + bool var_19985_interleave_0 = const()[name = string("op_19985_interleave_0"), val = bool(false)]; + tensor var_19985_cast_fp16 = concat(axis = var_18506, interleave = var_19985_interleave_0, values = (var_19939_cast_fp16, var_19941_cast_fp16, var_19943_cast_fp16, var_19945_cast_fp16))[name = string("op_19985_cast_fp16")]; + bool input_97_interleave_0 = const()[name = string("input_97_interleave_0"), val = bool(false)]; + tensor input_97_cast_fp16 = concat(axis = var_18531, interleave = input_97_interleave_0, values = (var_19947_cast_fp16, var_19949_cast_fp16, var_19951_cast_fp16, var_19953_cast_fp16, var_19955_cast_fp16, var_19957_cast_fp16, var_19959_cast_fp16, var_19961_cast_fp16, var_19963_cast_fp16, var_19965_cast_fp16, var_19967_cast_fp16, var_19969_cast_fp16, var_19971_cast_fp16, var_19973_cast_fp16, var_19975_cast_fp16, var_19977_cast_fp16, var_19979_cast_fp16, var_19981_cast_fp16, var_19983_cast_fp16, var_19985_cast_fp16))[name = string("input_97_cast_fp16")]; + string obj_51_pad_type_0 = const()[name = string("obj_51_pad_type_0"), val = string("valid")]; + tensor obj_51_strides_0 = const()[name = string("obj_51_strides_0"), val = tensor([1, 1])]; + tensor obj_51_pad_0 = const()[name = string("obj_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_51_dilations_0 = const()[name = string("obj_51_dilations_0"), val = tensor([1, 1])]; + int32 obj_51_groups_0 = const()[name = string("obj_51_groups_0"), val = int32(1)]; + tensor layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496744640)))]; + tensor layers_12_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500021504)))]; + tensor obj_51_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_bias_to_fp16, dilations = obj_51_dilations_0, groups = obj_51_groups_0, pad = obj_51_pad_0, pad_type = obj_51_pad_type_0, strides = obj_51_strides_0, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("obj_51_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_51_cast_fp16)[name = string("inputs_51_cast_fp16")]; + tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; + fp16 var_20004_to_fp16 = const()[name = string("op_20004_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_20004_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; + tensor input_99_gamma_0_to_fp16 = const()[name = string("input_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500024128)))]; + tensor input_99_beta_0_to_fp16 = const()[name = string("input_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500026752)))]; + fp16 input_99_epsilon_0_to_fp16 = const()[name = string("input_99_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_99_cast_fp16 = batch_norm(beta = input_99_beta_0_to_fp16, epsilon = input_99_epsilon_0_to_fp16, gamma = input_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("input_99_cast_fp16")]; + string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")]; + tensor input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor([1, 1])]; + tensor input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor([1, 1])]; + int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1)]; + tensor layers_12_fc1_weight_to_fp16 = const()[name = string("layers_12_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(500029376)))]; + tensor layers_12_fc1_bias_to_fp16 = const()[name = string("layers_12_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513136640)))]; + tensor input_101_cast_fp16 = conv(bias = layers_12_fc1_bias_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = layers_12_fc1_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")]; + string input_103_mode_0 = const()[name = string("input_103_mode_0"), val = string("EXACT")]; + tensor input_103_cast_fp16 = gelu(mode = input_103_mode_0, x = input_101_cast_fp16)[name = string("input_103_cast_fp16")]; + string hidden_states_29_pad_type_0 = const()[name = string("hidden_states_29_pad_type_0"), val = string("valid")]; + tensor hidden_states_29_strides_0 = const()[name = string("hidden_states_29_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_29_pad_0 = const()[name = string("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_29_dilations_0 = const()[name = string("hidden_states_29_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_29_groups_0 = const()[name = string("hidden_states_29_groups_0"), val = int32(1)]; + tensor layers_12_fc2_weight_to_fp16 = const()[name = string("layers_12_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513146944)))]; + tensor layers_12_fc2_bias_to_fp16 = const()[name = string("layers_12_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526254208)))]; + tensor hidden_states_29_cast_fp16 = conv(bias = layers_12_fc2_bias_to_fp16, dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = layers_12_fc2_weight_to_fp16, x = input_103_cast_fp16)[name = string("hidden_states_29_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = hidden_states_29_cast_fp16)[name = string("inputs_53_cast_fp16")]; + int32 var_20033 = const()[name = string("op_20033"), val = int32(3)]; + int32 var_20058 = const()[name = string("op_20058"), val = int32(1)]; + tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; + fp16 var_20075_to_fp16 = const()[name = string("op_20075_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_20075_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; + tensor obj_53_gamma_0_to_fp16 = const()[name = string("obj_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526256832)))]; + tensor obj_53_beta_0_to_fp16 = const()[name = string("obj_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526259456)))]; + fp16 obj_53_epsilon_0_to_fp16 = const()[name = string("obj_53_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_53_cast_fp16 = batch_norm(beta = obj_53_beta_0_to_fp16, epsilon = obj_53_epsilon_0_to_fp16, gamma = obj_53_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("obj_53_cast_fp16")]; + string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; + tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; + tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; + int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(526262080)))]; + tensor layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529538944)))]; + tensor query_27_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("query_27_cast_fp16")]; + string key_27_pad_type_0 = const()[name = string("key_27_pad_type_0"), val = string("valid")]; + tensor key_27_strides_0 = const()[name = string("key_27_strides_0"), val = tensor([1, 1])]; + tensor key_27_pad_0 = const()[name = string("key_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_27_dilations_0 = const()[name = string("key_27_dilations_0"), val = tensor([1, 1])]; + int32 key_27_groups_0 = const()[name = string("key_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(529541568)))]; + tensor key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("key_27_cast_fp16")]; + string value_27_pad_type_0 = const()[name = string("value_27_pad_type_0"), val = string("valid")]; + tensor value_27_strides_0 = const()[name = string("value_27_strides_0"), val = tensor([1, 1])]; + tensor value_27_pad_0 = const()[name = string("value_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_27_dilations_0 = const()[name = string("value_27_dilations_0"), val = tensor([1, 1])]; + int32 value_27_groups_0 = const()[name = string("value_27_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(532818432)))]; + tensor layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536095296)))]; + tensor value_27_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = obj_53_cast_fp16)[name = string("value_27_cast_fp16")]; + tensor var_20113_begin_0 = const()[name = string("op_20113_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20113_end_0 = const()[name = string("op_20113_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20113_end_mask_0 = const()[name = string("op_20113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20113_cast_fp16 = slice_by_index(begin = var_20113_begin_0, end = var_20113_end_0, end_mask = var_20113_end_mask_0, x = query_27_cast_fp16)[name = string("op_20113_cast_fp16")]; + tensor var_20117_begin_0 = const()[name = string("op_20117_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20117_end_0 = const()[name = string("op_20117_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_20117_end_mask_0 = const()[name = string("op_20117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20117_cast_fp16 = slice_by_index(begin = var_20117_begin_0, end = var_20117_end_0, end_mask = var_20117_end_mask_0, x = query_27_cast_fp16)[name = string("op_20117_cast_fp16")]; + tensor var_20121_begin_0 = const()[name = string("op_20121_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20121_end_0 = const()[name = string("op_20121_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_20121_end_mask_0 = const()[name = string("op_20121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20121_cast_fp16 = slice_by_index(begin = var_20121_begin_0, end = var_20121_end_0, end_mask = var_20121_end_mask_0, x = query_27_cast_fp16)[name = string("op_20121_cast_fp16")]; + tensor var_20125_begin_0 = const()[name = string("op_20125_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20125_end_0 = const()[name = string("op_20125_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_20125_end_mask_0 = const()[name = string("op_20125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20125_cast_fp16 = slice_by_index(begin = var_20125_begin_0, end = var_20125_end_0, end_mask = var_20125_end_mask_0, x = query_27_cast_fp16)[name = string("op_20125_cast_fp16")]; + tensor var_20129_begin_0 = const()[name = string("op_20129_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20129_end_0 = const()[name = string("op_20129_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_20129_end_mask_0 = const()[name = string("op_20129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20129_cast_fp16 = slice_by_index(begin = var_20129_begin_0, end = var_20129_end_0, end_mask = var_20129_end_mask_0, x = query_27_cast_fp16)[name = string("op_20129_cast_fp16")]; + tensor var_20133_begin_0 = const()[name = string("op_20133_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20133_end_0 = const()[name = string("op_20133_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_20133_end_mask_0 = const()[name = string("op_20133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20133_cast_fp16 = slice_by_index(begin = var_20133_begin_0, end = var_20133_end_0, end_mask = var_20133_end_mask_0, x = query_27_cast_fp16)[name = string("op_20133_cast_fp16")]; + tensor var_20137_begin_0 = const()[name = string("op_20137_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20137_end_0 = const()[name = string("op_20137_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_20137_end_mask_0 = const()[name = string("op_20137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20137_cast_fp16 = slice_by_index(begin = var_20137_begin_0, end = var_20137_end_0, end_mask = var_20137_end_mask_0, x = query_27_cast_fp16)[name = string("op_20137_cast_fp16")]; + tensor var_20141_begin_0 = const()[name = string("op_20141_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20141_end_0 = const()[name = string("op_20141_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_20141_end_mask_0 = const()[name = string("op_20141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20141_cast_fp16 = slice_by_index(begin = var_20141_begin_0, end = var_20141_end_0, end_mask = var_20141_end_mask_0, x = query_27_cast_fp16)[name = string("op_20141_cast_fp16")]; + tensor var_20145_begin_0 = const()[name = string("op_20145_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20145_end_0 = const()[name = string("op_20145_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_20145_end_mask_0 = const()[name = string("op_20145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20145_cast_fp16 = slice_by_index(begin = var_20145_begin_0, end = var_20145_end_0, end_mask = var_20145_end_mask_0, x = query_27_cast_fp16)[name = string("op_20145_cast_fp16")]; + tensor var_20149_begin_0 = const()[name = string("op_20149_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20149_end_0 = const()[name = string("op_20149_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_20149_end_mask_0 = const()[name = string("op_20149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20149_cast_fp16 = slice_by_index(begin = var_20149_begin_0, end = var_20149_end_0, end_mask = var_20149_end_mask_0, x = query_27_cast_fp16)[name = string("op_20149_cast_fp16")]; + tensor var_20153_begin_0 = const()[name = string("op_20153_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20153_end_0 = const()[name = string("op_20153_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_20153_end_mask_0 = const()[name = string("op_20153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20153_cast_fp16 = slice_by_index(begin = var_20153_begin_0, end = var_20153_end_0, end_mask = var_20153_end_mask_0, x = query_27_cast_fp16)[name = string("op_20153_cast_fp16")]; + tensor var_20157_begin_0 = const()[name = string("op_20157_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20157_end_0 = const()[name = string("op_20157_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_20157_end_mask_0 = const()[name = string("op_20157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20157_cast_fp16 = slice_by_index(begin = var_20157_begin_0, end = var_20157_end_0, end_mask = var_20157_end_mask_0, x = query_27_cast_fp16)[name = string("op_20157_cast_fp16")]; + tensor var_20161_begin_0 = const()[name = string("op_20161_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20161_end_0 = const()[name = string("op_20161_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_20161_end_mask_0 = const()[name = string("op_20161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20161_cast_fp16 = slice_by_index(begin = var_20161_begin_0, end = var_20161_end_0, end_mask = var_20161_end_mask_0, x = query_27_cast_fp16)[name = string("op_20161_cast_fp16")]; + tensor var_20165_begin_0 = const()[name = string("op_20165_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20165_end_0 = const()[name = string("op_20165_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_20165_end_mask_0 = const()[name = string("op_20165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20165_cast_fp16 = slice_by_index(begin = var_20165_begin_0, end = var_20165_end_0, end_mask = var_20165_end_mask_0, x = query_27_cast_fp16)[name = string("op_20165_cast_fp16")]; + tensor var_20169_begin_0 = const()[name = string("op_20169_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20169_end_0 = const()[name = string("op_20169_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_20169_end_mask_0 = const()[name = string("op_20169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20169_cast_fp16 = slice_by_index(begin = var_20169_begin_0, end = var_20169_end_0, end_mask = var_20169_end_mask_0, x = query_27_cast_fp16)[name = string("op_20169_cast_fp16")]; + tensor var_20173_begin_0 = const()[name = string("op_20173_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20173_end_0 = const()[name = string("op_20173_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_20173_end_mask_0 = const()[name = string("op_20173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20173_cast_fp16 = slice_by_index(begin = var_20173_begin_0, end = var_20173_end_0, end_mask = var_20173_end_mask_0, x = query_27_cast_fp16)[name = string("op_20173_cast_fp16")]; + tensor var_20177_begin_0 = const()[name = string("op_20177_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20177_end_0 = const()[name = string("op_20177_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_20177_end_mask_0 = const()[name = string("op_20177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20177_cast_fp16 = slice_by_index(begin = var_20177_begin_0, end = var_20177_end_0, end_mask = var_20177_end_mask_0, x = query_27_cast_fp16)[name = string("op_20177_cast_fp16")]; + tensor var_20181_begin_0 = const()[name = string("op_20181_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20181_end_0 = const()[name = string("op_20181_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_20181_end_mask_0 = const()[name = string("op_20181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20181_cast_fp16 = slice_by_index(begin = var_20181_begin_0, end = var_20181_end_0, end_mask = var_20181_end_mask_0, x = query_27_cast_fp16)[name = string("op_20181_cast_fp16")]; + tensor var_20185_begin_0 = const()[name = string("op_20185_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20185_end_0 = const()[name = string("op_20185_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_20185_end_mask_0 = const()[name = string("op_20185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20185_cast_fp16 = slice_by_index(begin = var_20185_begin_0, end = var_20185_end_0, end_mask = var_20185_end_mask_0, x = query_27_cast_fp16)[name = string("op_20185_cast_fp16")]; + tensor var_20189_begin_0 = const()[name = string("op_20189_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20189_end_0 = const()[name = string("op_20189_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_20189_end_mask_0 = const()[name = string("op_20189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20189_cast_fp16 = slice_by_index(begin = var_20189_begin_0, end = var_20189_end_0, end_mask = var_20189_end_mask_0, x = query_27_cast_fp16)[name = string("op_20189_cast_fp16")]; + tensor var_20198_begin_0 = const()[name = string("op_20198_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20198_end_0 = const()[name = string("op_20198_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20198_end_mask_0 = const()[name = string("op_20198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20198_cast_fp16 = slice_by_index(begin = var_20198_begin_0, end = var_20198_end_0, end_mask = var_20198_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20198_cast_fp16")]; + tensor var_20205_begin_0 = const()[name = string("op_20205_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20205_end_0 = const()[name = string("op_20205_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20205_end_mask_0 = const()[name = string("op_20205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20205_cast_fp16 = slice_by_index(begin = var_20205_begin_0, end = var_20205_end_0, end_mask = var_20205_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20205_cast_fp16")]; + tensor var_20212_begin_0 = const()[name = string("op_20212_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20212_end_0 = const()[name = string("op_20212_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20212_end_mask_0 = const()[name = string("op_20212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20212_cast_fp16 = slice_by_index(begin = var_20212_begin_0, end = var_20212_end_0, end_mask = var_20212_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20212_cast_fp16")]; + tensor var_20219_begin_0 = const()[name = string("op_20219_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20219_end_0 = const()[name = string("op_20219_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20219_end_mask_0 = const()[name = string("op_20219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20219_cast_fp16 = slice_by_index(begin = var_20219_begin_0, end = var_20219_end_0, end_mask = var_20219_end_mask_0, x = var_20113_cast_fp16)[name = string("op_20219_cast_fp16")]; + tensor var_20226_begin_0 = const()[name = string("op_20226_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20226_end_0 = const()[name = string("op_20226_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20226_end_mask_0 = const()[name = string("op_20226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20226_cast_fp16 = slice_by_index(begin = var_20226_begin_0, end = var_20226_end_0, end_mask = var_20226_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20226_cast_fp16")]; + tensor var_20233_begin_0 = const()[name = string("op_20233_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20233_end_0 = const()[name = string("op_20233_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20233_end_mask_0 = const()[name = string("op_20233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20233_cast_fp16 = slice_by_index(begin = var_20233_begin_0, end = var_20233_end_0, end_mask = var_20233_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20233_cast_fp16")]; + tensor var_20240_begin_0 = const()[name = string("op_20240_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20240_end_0 = const()[name = string("op_20240_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20240_end_mask_0 = const()[name = string("op_20240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20240_cast_fp16 = slice_by_index(begin = var_20240_begin_0, end = var_20240_end_0, end_mask = var_20240_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20240_cast_fp16")]; + tensor var_20247_begin_0 = const()[name = string("op_20247_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20247_end_0 = const()[name = string("op_20247_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20247_end_mask_0 = const()[name = string("op_20247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20247_cast_fp16 = slice_by_index(begin = var_20247_begin_0, end = var_20247_end_0, end_mask = var_20247_end_mask_0, x = var_20117_cast_fp16)[name = string("op_20247_cast_fp16")]; + tensor var_20254_begin_0 = const()[name = string("op_20254_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20254_end_0 = const()[name = string("op_20254_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20254_end_mask_0 = const()[name = string("op_20254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20254_cast_fp16 = slice_by_index(begin = var_20254_begin_0, end = var_20254_end_0, end_mask = var_20254_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20254_cast_fp16")]; + tensor var_20261_begin_0 = const()[name = string("op_20261_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20261_end_0 = const()[name = string("op_20261_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20261_end_mask_0 = const()[name = string("op_20261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20261_cast_fp16 = slice_by_index(begin = var_20261_begin_0, end = var_20261_end_0, end_mask = var_20261_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20261_cast_fp16")]; + tensor var_20268_begin_0 = const()[name = string("op_20268_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20268_end_0 = const()[name = string("op_20268_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20268_end_mask_0 = const()[name = string("op_20268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20268_cast_fp16 = slice_by_index(begin = var_20268_begin_0, end = var_20268_end_0, end_mask = var_20268_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20268_cast_fp16")]; + tensor var_20275_begin_0 = const()[name = string("op_20275_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20275_end_0 = const()[name = string("op_20275_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20275_end_mask_0 = const()[name = string("op_20275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20275_cast_fp16 = slice_by_index(begin = var_20275_begin_0, end = var_20275_end_0, end_mask = var_20275_end_mask_0, x = var_20121_cast_fp16)[name = string("op_20275_cast_fp16")]; + tensor var_20282_begin_0 = const()[name = string("op_20282_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20282_end_0 = const()[name = string("op_20282_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20282_end_mask_0 = const()[name = string("op_20282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20282_cast_fp16 = slice_by_index(begin = var_20282_begin_0, end = var_20282_end_0, end_mask = var_20282_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20282_cast_fp16")]; + tensor var_20289_begin_0 = const()[name = string("op_20289_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20289_end_0 = const()[name = string("op_20289_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20289_end_mask_0 = const()[name = string("op_20289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20289_cast_fp16 = slice_by_index(begin = var_20289_begin_0, end = var_20289_end_0, end_mask = var_20289_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20289_cast_fp16")]; + tensor var_20296_begin_0 = const()[name = string("op_20296_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20296_end_0 = const()[name = string("op_20296_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20296_end_mask_0 = const()[name = string("op_20296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20296_cast_fp16 = slice_by_index(begin = var_20296_begin_0, end = var_20296_end_0, end_mask = var_20296_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20296_cast_fp16")]; + tensor var_20303_begin_0 = const()[name = string("op_20303_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20303_end_0 = const()[name = string("op_20303_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20303_end_mask_0 = const()[name = string("op_20303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20303_cast_fp16 = slice_by_index(begin = var_20303_begin_0, end = var_20303_end_0, end_mask = var_20303_end_mask_0, x = var_20125_cast_fp16)[name = string("op_20303_cast_fp16")]; + tensor var_20310_begin_0 = const()[name = string("op_20310_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20310_end_0 = const()[name = string("op_20310_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20310_end_mask_0 = const()[name = string("op_20310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20310_cast_fp16 = slice_by_index(begin = var_20310_begin_0, end = var_20310_end_0, end_mask = var_20310_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20310_cast_fp16")]; + tensor var_20317_begin_0 = const()[name = string("op_20317_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20317_end_0 = const()[name = string("op_20317_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20317_end_mask_0 = const()[name = string("op_20317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20317_cast_fp16 = slice_by_index(begin = var_20317_begin_0, end = var_20317_end_0, end_mask = var_20317_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20317_cast_fp16")]; + tensor var_20324_begin_0 = const()[name = string("op_20324_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20324_end_0 = const()[name = string("op_20324_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20324_end_mask_0 = const()[name = string("op_20324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20324_cast_fp16 = slice_by_index(begin = var_20324_begin_0, end = var_20324_end_0, end_mask = var_20324_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20324_cast_fp16")]; + tensor var_20331_begin_0 = const()[name = string("op_20331_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20331_end_0 = const()[name = string("op_20331_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20331_end_mask_0 = const()[name = string("op_20331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20331_cast_fp16 = slice_by_index(begin = var_20331_begin_0, end = var_20331_end_0, end_mask = var_20331_end_mask_0, x = var_20129_cast_fp16)[name = string("op_20331_cast_fp16")]; + tensor var_20338_begin_0 = const()[name = string("op_20338_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20338_end_0 = const()[name = string("op_20338_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20338_end_mask_0 = const()[name = string("op_20338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20338_cast_fp16 = slice_by_index(begin = var_20338_begin_0, end = var_20338_end_0, end_mask = var_20338_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20338_cast_fp16")]; + tensor var_20345_begin_0 = const()[name = string("op_20345_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20345_end_0 = const()[name = string("op_20345_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20345_end_mask_0 = const()[name = string("op_20345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20345_cast_fp16 = slice_by_index(begin = var_20345_begin_0, end = var_20345_end_0, end_mask = var_20345_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20345_cast_fp16")]; + tensor var_20352_begin_0 = const()[name = string("op_20352_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20352_end_0 = const()[name = string("op_20352_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20352_end_mask_0 = const()[name = string("op_20352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20352_cast_fp16 = slice_by_index(begin = var_20352_begin_0, end = var_20352_end_0, end_mask = var_20352_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20352_cast_fp16")]; + tensor var_20359_begin_0 = const()[name = string("op_20359_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20359_end_0 = const()[name = string("op_20359_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20359_end_mask_0 = const()[name = string("op_20359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20359_cast_fp16 = slice_by_index(begin = var_20359_begin_0, end = var_20359_end_0, end_mask = var_20359_end_mask_0, x = var_20133_cast_fp16)[name = string("op_20359_cast_fp16")]; + tensor var_20366_begin_0 = const()[name = string("op_20366_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20366_end_0 = const()[name = string("op_20366_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20366_end_mask_0 = const()[name = string("op_20366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20366_cast_fp16 = slice_by_index(begin = var_20366_begin_0, end = var_20366_end_0, end_mask = var_20366_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20366_cast_fp16")]; + tensor var_20373_begin_0 = const()[name = string("op_20373_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20373_end_0 = const()[name = string("op_20373_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20373_end_mask_0 = const()[name = string("op_20373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20373_cast_fp16 = slice_by_index(begin = var_20373_begin_0, end = var_20373_end_0, end_mask = var_20373_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20373_cast_fp16")]; + tensor var_20380_begin_0 = const()[name = string("op_20380_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20380_end_0 = const()[name = string("op_20380_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20380_end_mask_0 = const()[name = string("op_20380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20380_cast_fp16 = slice_by_index(begin = var_20380_begin_0, end = var_20380_end_0, end_mask = var_20380_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20380_cast_fp16")]; + tensor var_20387_begin_0 = const()[name = string("op_20387_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20387_end_0 = const()[name = string("op_20387_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20387_end_mask_0 = const()[name = string("op_20387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20387_cast_fp16 = slice_by_index(begin = var_20387_begin_0, end = var_20387_end_0, end_mask = var_20387_end_mask_0, x = var_20137_cast_fp16)[name = string("op_20387_cast_fp16")]; + tensor var_20394_begin_0 = const()[name = string("op_20394_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20394_end_0 = const()[name = string("op_20394_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20394_end_mask_0 = const()[name = string("op_20394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20394_cast_fp16 = slice_by_index(begin = var_20394_begin_0, end = var_20394_end_0, end_mask = var_20394_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20394_cast_fp16")]; + tensor var_20401_begin_0 = const()[name = string("op_20401_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20401_end_0 = const()[name = string("op_20401_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20401_end_mask_0 = const()[name = string("op_20401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20401_cast_fp16 = slice_by_index(begin = var_20401_begin_0, end = var_20401_end_0, end_mask = var_20401_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20401_cast_fp16")]; + tensor var_20408_begin_0 = const()[name = string("op_20408_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20408_end_0 = const()[name = string("op_20408_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20408_end_mask_0 = const()[name = string("op_20408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20408_cast_fp16 = slice_by_index(begin = var_20408_begin_0, end = var_20408_end_0, end_mask = var_20408_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20408_cast_fp16")]; + tensor var_20415_begin_0 = const()[name = string("op_20415_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20415_end_0 = const()[name = string("op_20415_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20415_end_mask_0 = const()[name = string("op_20415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20415_cast_fp16 = slice_by_index(begin = var_20415_begin_0, end = var_20415_end_0, end_mask = var_20415_end_mask_0, x = var_20141_cast_fp16)[name = string("op_20415_cast_fp16")]; + tensor var_20422_begin_0 = const()[name = string("op_20422_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20422_end_0 = const()[name = string("op_20422_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20422_end_mask_0 = const()[name = string("op_20422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20422_cast_fp16 = slice_by_index(begin = var_20422_begin_0, end = var_20422_end_0, end_mask = var_20422_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20422_cast_fp16")]; + tensor var_20429_begin_0 = const()[name = string("op_20429_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20429_end_0 = const()[name = string("op_20429_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20429_end_mask_0 = const()[name = string("op_20429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20429_cast_fp16 = slice_by_index(begin = var_20429_begin_0, end = var_20429_end_0, end_mask = var_20429_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20429_cast_fp16")]; + tensor var_20436_begin_0 = const()[name = string("op_20436_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20436_end_0 = const()[name = string("op_20436_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20436_end_mask_0 = const()[name = string("op_20436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20436_cast_fp16 = slice_by_index(begin = var_20436_begin_0, end = var_20436_end_0, end_mask = var_20436_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20436_cast_fp16")]; + tensor var_20443_begin_0 = const()[name = string("op_20443_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20443_end_0 = const()[name = string("op_20443_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20443_end_mask_0 = const()[name = string("op_20443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20443_cast_fp16 = slice_by_index(begin = var_20443_begin_0, end = var_20443_end_0, end_mask = var_20443_end_mask_0, x = var_20145_cast_fp16)[name = string("op_20443_cast_fp16")]; + tensor var_20450_begin_0 = const()[name = string("op_20450_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20450_end_0 = const()[name = string("op_20450_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20450_end_mask_0 = const()[name = string("op_20450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20450_cast_fp16 = slice_by_index(begin = var_20450_begin_0, end = var_20450_end_0, end_mask = var_20450_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20450_cast_fp16")]; + tensor var_20457_begin_0 = const()[name = string("op_20457_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20457_end_0 = const()[name = string("op_20457_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20457_end_mask_0 = const()[name = string("op_20457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20457_cast_fp16 = slice_by_index(begin = var_20457_begin_0, end = var_20457_end_0, end_mask = var_20457_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20457_cast_fp16")]; + tensor var_20464_begin_0 = const()[name = string("op_20464_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20464_end_0 = const()[name = string("op_20464_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20464_end_mask_0 = const()[name = string("op_20464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20464_cast_fp16 = slice_by_index(begin = var_20464_begin_0, end = var_20464_end_0, end_mask = var_20464_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20464_cast_fp16")]; + tensor var_20471_begin_0 = const()[name = string("op_20471_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20471_end_0 = const()[name = string("op_20471_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20471_end_mask_0 = const()[name = string("op_20471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20471_cast_fp16 = slice_by_index(begin = var_20471_begin_0, end = var_20471_end_0, end_mask = var_20471_end_mask_0, x = var_20149_cast_fp16)[name = string("op_20471_cast_fp16")]; + tensor var_20478_begin_0 = const()[name = string("op_20478_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20478_end_0 = const()[name = string("op_20478_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20478_end_mask_0 = const()[name = string("op_20478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20478_cast_fp16 = slice_by_index(begin = var_20478_begin_0, end = var_20478_end_0, end_mask = var_20478_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20478_cast_fp16")]; + tensor var_20485_begin_0 = const()[name = string("op_20485_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20485_end_0 = const()[name = string("op_20485_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20485_end_mask_0 = const()[name = string("op_20485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20485_cast_fp16 = slice_by_index(begin = var_20485_begin_0, end = var_20485_end_0, end_mask = var_20485_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20485_cast_fp16")]; + tensor var_20492_begin_0 = const()[name = string("op_20492_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20492_end_0 = const()[name = string("op_20492_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20492_end_mask_0 = const()[name = string("op_20492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20492_cast_fp16 = slice_by_index(begin = var_20492_begin_0, end = var_20492_end_0, end_mask = var_20492_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20492_cast_fp16")]; + tensor var_20499_begin_0 = const()[name = string("op_20499_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20499_end_0 = const()[name = string("op_20499_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20499_end_mask_0 = const()[name = string("op_20499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20499_cast_fp16 = slice_by_index(begin = var_20499_begin_0, end = var_20499_end_0, end_mask = var_20499_end_mask_0, x = var_20153_cast_fp16)[name = string("op_20499_cast_fp16")]; + tensor var_20506_begin_0 = const()[name = string("op_20506_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20506_end_0 = const()[name = string("op_20506_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20506_end_mask_0 = const()[name = string("op_20506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20506_cast_fp16 = slice_by_index(begin = var_20506_begin_0, end = var_20506_end_0, end_mask = var_20506_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20506_cast_fp16")]; + tensor var_20513_begin_0 = const()[name = string("op_20513_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20513_end_0 = const()[name = string("op_20513_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20513_end_mask_0 = const()[name = string("op_20513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20513_cast_fp16 = slice_by_index(begin = var_20513_begin_0, end = var_20513_end_0, end_mask = var_20513_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20513_cast_fp16")]; + tensor var_20520_begin_0 = const()[name = string("op_20520_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20520_end_0 = const()[name = string("op_20520_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20520_end_mask_0 = const()[name = string("op_20520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20520_cast_fp16 = slice_by_index(begin = var_20520_begin_0, end = var_20520_end_0, end_mask = var_20520_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20520_cast_fp16")]; + tensor var_20527_begin_0 = const()[name = string("op_20527_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20527_end_0 = const()[name = string("op_20527_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20527_end_mask_0 = const()[name = string("op_20527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20527_cast_fp16 = slice_by_index(begin = var_20527_begin_0, end = var_20527_end_0, end_mask = var_20527_end_mask_0, x = var_20157_cast_fp16)[name = string("op_20527_cast_fp16")]; + tensor var_20534_begin_0 = const()[name = string("op_20534_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20534_end_0 = const()[name = string("op_20534_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20534_end_mask_0 = const()[name = string("op_20534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20534_cast_fp16 = slice_by_index(begin = var_20534_begin_0, end = var_20534_end_0, end_mask = var_20534_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20534_cast_fp16")]; + tensor var_20541_begin_0 = const()[name = string("op_20541_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20541_end_0 = const()[name = string("op_20541_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20541_end_mask_0 = const()[name = string("op_20541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20541_cast_fp16 = slice_by_index(begin = var_20541_begin_0, end = var_20541_end_0, end_mask = var_20541_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20541_cast_fp16")]; + tensor var_20548_begin_0 = const()[name = string("op_20548_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20548_end_0 = const()[name = string("op_20548_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20548_end_mask_0 = const()[name = string("op_20548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20548_cast_fp16 = slice_by_index(begin = var_20548_begin_0, end = var_20548_end_0, end_mask = var_20548_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20548_cast_fp16")]; + tensor var_20555_begin_0 = const()[name = string("op_20555_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20555_end_0 = const()[name = string("op_20555_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20555_end_mask_0 = const()[name = string("op_20555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20555_cast_fp16 = slice_by_index(begin = var_20555_begin_0, end = var_20555_end_0, end_mask = var_20555_end_mask_0, x = var_20161_cast_fp16)[name = string("op_20555_cast_fp16")]; + tensor var_20562_begin_0 = const()[name = string("op_20562_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20562_end_0 = const()[name = string("op_20562_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20562_end_mask_0 = const()[name = string("op_20562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20562_cast_fp16 = slice_by_index(begin = var_20562_begin_0, end = var_20562_end_0, end_mask = var_20562_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20562_cast_fp16")]; + tensor var_20569_begin_0 = const()[name = string("op_20569_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20569_end_0 = const()[name = string("op_20569_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20569_end_mask_0 = const()[name = string("op_20569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20569_cast_fp16 = slice_by_index(begin = var_20569_begin_0, end = var_20569_end_0, end_mask = var_20569_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20569_cast_fp16")]; + tensor var_20576_begin_0 = const()[name = string("op_20576_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20576_end_0 = const()[name = string("op_20576_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20576_end_mask_0 = const()[name = string("op_20576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20576_cast_fp16 = slice_by_index(begin = var_20576_begin_0, end = var_20576_end_0, end_mask = var_20576_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20576_cast_fp16")]; + tensor var_20583_begin_0 = const()[name = string("op_20583_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20583_end_0 = const()[name = string("op_20583_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20583_end_mask_0 = const()[name = string("op_20583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20583_cast_fp16 = slice_by_index(begin = var_20583_begin_0, end = var_20583_end_0, end_mask = var_20583_end_mask_0, x = var_20165_cast_fp16)[name = string("op_20583_cast_fp16")]; + tensor var_20590_begin_0 = const()[name = string("op_20590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20590_end_0 = const()[name = string("op_20590_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20590_end_mask_0 = const()[name = string("op_20590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20590_cast_fp16 = slice_by_index(begin = var_20590_begin_0, end = var_20590_end_0, end_mask = var_20590_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20590_cast_fp16")]; + tensor var_20597_begin_0 = const()[name = string("op_20597_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20597_end_0 = const()[name = string("op_20597_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20597_end_mask_0 = const()[name = string("op_20597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20597_cast_fp16 = slice_by_index(begin = var_20597_begin_0, end = var_20597_end_0, end_mask = var_20597_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20597_cast_fp16")]; + tensor var_20604_begin_0 = const()[name = string("op_20604_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20604_end_0 = const()[name = string("op_20604_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20604_end_mask_0 = const()[name = string("op_20604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20604_cast_fp16 = slice_by_index(begin = var_20604_begin_0, end = var_20604_end_0, end_mask = var_20604_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20604_cast_fp16")]; + tensor var_20611_begin_0 = const()[name = string("op_20611_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20611_end_0 = const()[name = string("op_20611_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20611_end_mask_0 = const()[name = string("op_20611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20611_cast_fp16 = slice_by_index(begin = var_20611_begin_0, end = var_20611_end_0, end_mask = var_20611_end_mask_0, x = var_20169_cast_fp16)[name = string("op_20611_cast_fp16")]; + tensor var_20618_begin_0 = const()[name = string("op_20618_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20618_end_0 = const()[name = string("op_20618_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20618_end_mask_0 = const()[name = string("op_20618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20618_cast_fp16 = slice_by_index(begin = var_20618_begin_0, end = var_20618_end_0, end_mask = var_20618_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20618_cast_fp16")]; + tensor var_20625_begin_0 = const()[name = string("op_20625_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20625_end_0 = const()[name = string("op_20625_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20625_end_mask_0 = const()[name = string("op_20625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20625_cast_fp16 = slice_by_index(begin = var_20625_begin_0, end = var_20625_end_0, end_mask = var_20625_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20625_cast_fp16")]; + tensor var_20632_begin_0 = const()[name = string("op_20632_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20632_end_0 = const()[name = string("op_20632_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20632_end_mask_0 = const()[name = string("op_20632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20632_cast_fp16 = slice_by_index(begin = var_20632_begin_0, end = var_20632_end_0, end_mask = var_20632_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20632_cast_fp16")]; + tensor var_20639_begin_0 = const()[name = string("op_20639_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20639_end_0 = const()[name = string("op_20639_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20639_end_mask_0 = const()[name = string("op_20639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20639_cast_fp16 = slice_by_index(begin = var_20639_begin_0, end = var_20639_end_0, end_mask = var_20639_end_mask_0, x = var_20173_cast_fp16)[name = string("op_20639_cast_fp16")]; + tensor var_20646_begin_0 = const()[name = string("op_20646_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20646_end_0 = const()[name = string("op_20646_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20646_end_mask_0 = const()[name = string("op_20646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20646_cast_fp16 = slice_by_index(begin = var_20646_begin_0, end = var_20646_end_0, end_mask = var_20646_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20646_cast_fp16")]; + tensor var_20653_begin_0 = const()[name = string("op_20653_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20653_end_0 = const()[name = string("op_20653_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20653_end_mask_0 = const()[name = string("op_20653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20653_cast_fp16 = slice_by_index(begin = var_20653_begin_0, end = var_20653_end_0, end_mask = var_20653_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20653_cast_fp16")]; + tensor var_20660_begin_0 = const()[name = string("op_20660_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20660_end_0 = const()[name = string("op_20660_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20660_end_mask_0 = const()[name = string("op_20660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20660_cast_fp16 = slice_by_index(begin = var_20660_begin_0, end = var_20660_end_0, end_mask = var_20660_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20660_cast_fp16")]; + tensor var_20667_begin_0 = const()[name = string("op_20667_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20667_end_0 = const()[name = string("op_20667_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20667_end_mask_0 = const()[name = string("op_20667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20667_cast_fp16 = slice_by_index(begin = var_20667_begin_0, end = var_20667_end_0, end_mask = var_20667_end_mask_0, x = var_20177_cast_fp16)[name = string("op_20667_cast_fp16")]; + tensor var_20674_begin_0 = const()[name = string("op_20674_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20674_end_0 = const()[name = string("op_20674_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20674_end_mask_0 = const()[name = string("op_20674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20674_cast_fp16 = slice_by_index(begin = var_20674_begin_0, end = var_20674_end_0, end_mask = var_20674_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20674_cast_fp16")]; + tensor var_20681_begin_0 = const()[name = string("op_20681_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20681_end_0 = const()[name = string("op_20681_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20681_end_mask_0 = const()[name = string("op_20681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20681_cast_fp16 = slice_by_index(begin = var_20681_begin_0, end = var_20681_end_0, end_mask = var_20681_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20681_cast_fp16")]; + tensor var_20688_begin_0 = const()[name = string("op_20688_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20688_end_0 = const()[name = string("op_20688_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20688_end_mask_0 = const()[name = string("op_20688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20688_cast_fp16 = slice_by_index(begin = var_20688_begin_0, end = var_20688_end_0, end_mask = var_20688_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20688_cast_fp16")]; + tensor var_20695_begin_0 = const()[name = string("op_20695_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20695_end_0 = const()[name = string("op_20695_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20695_end_mask_0 = const()[name = string("op_20695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20695_cast_fp16 = slice_by_index(begin = var_20695_begin_0, end = var_20695_end_0, end_mask = var_20695_end_mask_0, x = var_20181_cast_fp16)[name = string("op_20695_cast_fp16")]; + tensor var_20702_begin_0 = const()[name = string("op_20702_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20702_end_0 = const()[name = string("op_20702_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20702_end_mask_0 = const()[name = string("op_20702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20702_cast_fp16 = slice_by_index(begin = var_20702_begin_0, end = var_20702_end_0, end_mask = var_20702_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20702_cast_fp16")]; + tensor var_20709_begin_0 = const()[name = string("op_20709_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20709_end_0 = const()[name = string("op_20709_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20709_end_mask_0 = const()[name = string("op_20709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20709_cast_fp16 = slice_by_index(begin = var_20709_begin_0, end = var_20709_end_0, end_mask = var_20709_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20709_cast_fp16")]; + tensor var_20716_begin_0 = const()[name = string("op_20716_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20716_end_0 = const()[name = string("op_20716_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20716_end_mask_0 = const()[name = string("op_20716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20716_cast_fp16 = slice_by_index(begin = var_20716_begin_0, end = var_20716_end_0, end_mask = var_20716_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20716_cast_fp16")]; + tensor var_20723_begin_0 = const()[name = string("op_20723_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20723_end_0 = const()[name = string("op_20723_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20723_end_mask_0 = const()[name = string("op_20723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20723_cast_fp16 = slice_by_index(begin = var_20723_begin_0, end = var_20723_end_0, end_mask = var_20723_end_mask_0, x = var_20185_cast_fp16)[name = string("op_20723_cast_fp16")]; + tensor var_20730_begin_0 = const()[name = string("op_20730_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20730_end_0 = const()[name = string("op_20730_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20730_end_mask_0 = const()[name = string("op_20730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20730_cast_fp16 = slice_by_index(begin = var_20730_begin_0, end = var_20730_end_0, end_mask = var_20730_end_mask_0, x = var_20189_cast_fp16)[name = string("op_20730_cast_fp16")]; + tensor var_20737_begin_0 = const()[name = string("op_20737_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20737_end_0 = const()[name = string("op_20737_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20737_end_mask_0 = const()[name = string("op_20737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20737_cast_fp16 = slice_by_index(begin = var_20737_begin_0, end = var_20737_end_0, end_mask = var_20737_end_mask_0, x = var_20189_cast_fp16)[name = string("op_20737_cast_fp16")]; + tensor var_20744_begin_0 = const()[name = string("op_20744_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20744_end_0 = const()[name = string("op_20744_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20744_end_mask_0 = const()[name = string("op_20744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20744_cast_fp16 = slice_by_index(begin = var_20744_begin_0, end = var_20744_end_0, end_mask = var_20744_end_mask_0, x = var_20189_cast_fp16)[name = string("op_20744_cast_fp16")]; + tensor var_20751_begin_0 = const()[name = string("op_20751_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20751_end_0 = const()[name = string("op_20751_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20751_end_mask_0 = const()[name = string("op_20751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20751_cast_fp16 = slice_by_index(begin = var_20751_begin_0, end = var_20751_end_0, end_mask = var_20751_end_mask_0, x = var_20189_cast_fp16)[name = string("op_20751_cast_fp16")]; + tensor k_27_perm_0 = const()[name = string("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20756_begin_0 = const()[name = string("op_20756_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20756_end_0 = const()[name = string("op_20756_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_20756_end_mask_0 = const()[name = string("op_20756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = key_27_cast_fp16)[name = string("transpose_18")]; + tensor var_20756_cast_fp16 = slice_by_index(begin = var_20756_begin_0, end = var_20756_end_0, end_mask = var_20756_end_mask_0, x = k_27_cast_fp16)[name = string("op_20756_cast_fp16")]; + tensor var_20760_begin_0 = const()[name = string("op_20760_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20760_end_0 = const()[name = string("op_20760_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_20760_end_mask_0 = const()[name = string("op_20760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20760_cast_fp16 = slice_by_index(begin = var_20760_begin_0, end = var_20760_end_0, end_mask = var_20760_end_mask_0, x = k_27_cast_fp16)[name = string("op_20760_cast_fp16")]; + tensor var_20764_begin_0 = const()[name = string("op_20764_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20764_end_0 = const()[name = string("op_20764_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_20764_end_mask_0 = const()[name = string("op_20764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20764_cast_fp16 = slice_by_index(begin = var_20764_begin_0, end = var_20764_end_0, end_mask = var_20764_end_mask_0, x = k_27_cast_fp16)[name = string("op_20764_cast_fp16")]; + tensor var_20768_begin_0 = const()[name = string("op_20768_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20768_end_0 = const()[name = string("op_20768_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_20768_end_mask_0 = const()[name = string("op_20768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20768_cast_fp16 = slice_by_index(begin = var_20768_begin_0, end = var_20768_end_0, end_mask = var_20768_end_mask_0, x = k_27_cast_fp16)[name = string("op_20768_cast_fp16")]; + tensor var_20772_begin_0 = const()[name = string("op_20772_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20772_end_0 = const()[name = string("op_20772_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_20772_end_mask_0 = const()[name = string("op_20772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20772_cast_fp16 = slice_by_index(begin = var_20772_begin_0, end = var_20772_end_0, end_mask = var_20772_end_mask_0, x = k_27_cast_fp16)[name = string("op_20772_cast_fp16")]; + tensor var_20776_begin_0 = const()[name = string("op_20776_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20776_end_0 = const()[name = string("op_20776_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_20776_end_mask_0 = const()[name = string("op_20776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20776_cast_fp16 = slice_by_index(begin = var_20776_begin_0, end = var_20776_end_0, end_mask = var_20776_end_mask_0, x = k_27_cast_fp16)[name = string("op_20776_cast_fp16")]; + tensor var_20780_begin_0 = const()[name = string("op_20780_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20780_end_0 = const()[name = string("op_20780_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_20780_end_mask_0 = const()[name = string("op_20780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20780_cast_fp16 = slice_by_index(begin = var_20780_begin_0, end = var_20780_end_0, end_mask = var_20780_end_mask_0, x = k_27_cast_fp16)[name = string("op_20780_cast_fp16")]; + tensor var_20784_begin_0 = const()[name = string("op_20784_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20784_end_0 = const()[name = string("op_20784_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_20784_end_mask_0 = const()[name = string("op_20784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20784_cast_fp16 = slice_by_index(begin = var_20784_begin_0, end = var_20784_end_0, end_mask = var_20784_end_mask_0, x = k_27_cast_fp16)[name = string("op_20784_cast_fp16")]; + tensor var_20788_begin_0 = const()[name = string("op_20788_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20788_end_0 = const()[name = string("op_20788_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_20788_end_mask_0 = const()[name = string("op_20788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20788_cast_fp16 = slice_by_index(begin = var_20788_begin_0, end = var_20788_end_0, end_mask = var_20788_end_mask_0, x = k_27_cast_fp16)[name = string("op_20788_cast_fp16")]; + tensor var_20792_begin_0 = const()[name = string("op_20792_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20792_end_0 = const()[name = string("op_20792_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_20792_end_mask_0 = const()[name = string("op_20792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20792_cast_fp16 = slice_by_index(begin = var_20792_begin_0, end = var_20792_end_0, end_mask = var_20792_end_mask_0, x = k_27_cast_fp16)[name = string("op_20792_cast_fp16")]; + tensor var_20796_begin_0 = const()[name = string("op_20796_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20796_end_0 = const()[name = string("op_20796_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_20796_end_mask_0 = const()[name = string("op_20796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20796_cast_fp16 = slice_by_index(begin = var_20796_begin_0, end = var_20796_end_0, end_mask = var_20796_end_mask_0, x = k_27_cast_fp16)[name = string("op_20796_cast_fp16")]; + tensor var_20800_begin_0 = const()[name = string("op_20800_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20800_end_0 = const()[name = string("op_20800_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_20800_end_mask_0 = const()[name = string("op_20800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20800_cast_fp16 = slice_by_index(begin = var_20800_begin_0, end = var_20800_end_0, end_mask = var_20800_end_mask_0, x = k_27_cast_fp16)[name = string("op_20800_cast_fp16")]; + tensor var_20804_begin_0 = const()[name = string("op_20804_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20804_end_0 = const()[name = string("op_20804_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_20804_end_mask_0 = const()[name = string("op_20804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20804_cast_fp16 = slice_by_index(begin = var_20804_begin_0, end = var_20804_end_0, end_mask = var_20804_end_mask_0, x = k_27_cast_fp16)[name = string("op_20804_cast_fp16")]; + tensor var_20808_begin_0 = const()[name = string("op_20808_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20808_end_0 = const()[name = string("op_20808_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_20808_end_mask_0 = const()[name = string("op_20808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20808_cast_fp16 = slice_by_index(begin = var_20808_begin_0, end = var_20808_end_0, end_mask = var_20808_end_mask_0, x = k_27_cast_fp16)[name = string("op_20808_cast_fp16")]; + tensor var_20812_begin_0 = const()[name = string("op_20812_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20812_end_0 = const()[name = string("op_20812_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_20812_end_mask_0 = const()[name = string("op_20812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20812_cast_fp16 = slice_by_index(begin = var_20812_begin_0, end = var_20812_end_0, end_mask = var_20812_end_mask_0, x = k_27_cast_fp16)[name = string("op_20812_cast_fp16")]; + tensor var_20816_begin_0 = const()[name = string("op_20816_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20816_end_0 = const()[name = string("op_20816_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_20816_end_mask_0 = const()[name = string("op_20816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20816_cast_fp16 = slice_by_index(begin = var_20816_begin_0, end = var_20816_end_0, end_mask = var_20816_end_mask_0, x = k_27_cast_fp16)[name = string("op_20816_cast_fp16")]; + tensor var_20820_begin_0 = const()[name = string("op_20820_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20820_end_0 = const()[name = string("op_20820_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_20820_end_mask_0 = const()[name = string("op_20820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20820_cast_fp16 = slice_by_index(begin = var_20820_begin_0, end = var_20820_end_0, end_mask = var_20820_end_mask_0, x = k_27_cast_fp16)[name = string("op_20820_cast_fp16")]; + tensor var_20824_begin_0 = const()[name = string("op_20824_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20824_end_0 = const()[name = string("op_20824_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_20824_end_mask_0 = const()[name = string("op_20824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20824_cast_fp16 = slice_by_index(begin = var_20824_begin_0, end = var_20824_end_0, end_mask = var_20824_end_mask_0, x = k_27_cast_fp16)[name = string("op_20824_cast_fp16")]; + tensor var_20828_begin_0 = const()[name = string("op_20828_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20828_end_0 = const()[name = string("op_20828_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_20828_end_mask_0 = const()[name = string("op_20828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20828_cast_fp16 = slice_by_index(begin = var_20828_begin_0, end = var_20828_end_0, end_mask = var_20828_end_mask_0, x = k_27_cast_fp16)[name = string("op_20828_cast_fp16")]; + tensor var_20832_begin_0 = const()[name = string("op_20832_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20832_end_0 = const()[name = string("op_20832_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_20832_end_mask_0 = const()[name = string("op_20832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20832_cast_fp16 = slice_by_index(begin = var_20832_begin_0, end = var_20832_end_0, end_mask = var_20832_end_mask_0, x = k_27_cast_fp16)[name = string("op_20832_cast_fp16")]; + tensor var_20834_begin_0 = const()[name = string("op_20834_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20834_end_0 = const()[name = string("op_20834_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20834_end_mask_0 = const()[name = string("op_20834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20834_cast_fp16 = slice_by_index(begin = var_20834_begin_0, end = var_20834_end_0, end_mask = var_20834_end_mask_0, x = value_27_cast_fp16)[name = string("op_20834_cast_fp16")]; + tensor var_20838_begin_0 = const()[name = string("op_20838_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20838_end_0 = const()[name = string("op_20838_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_20838_end_mask_0 = const()[name = string("op_20838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20838_cast_fp16 = slice_by_index(begin = var_20838_begin_0, end = var_20838_end_0, end_mask = var_20838_end_mask_0, x = value_27_cast_fp16)[name = string("op_20838_cast_fp16")]; + tensor var_20842_begin_0 = const()[name = string("op_20842_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20842_end_0 = const()[name = string("op_20842_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_20842_end_mask_0 = const()[name = string("op_20842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20842_cast_fp16 = slice_by_index(begin = var_20842_begin_0, end = var_20842_end_0, end_mask = var_20842_end_mask_0, x = value_27_cast_fp16)[name = string("op_20842_cast_fp16")]; + tensor var_20846_begin_0 = const()[name = string("op_20846_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20846_end_0 = const()[name = string("op_20846_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_20846_end_mask_0 = const()[name = string("op_20846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20846_cast_fp16 = slice_by_index(begin = var_20846_begin_0, end = var_20846_end_0, end_mask = var_20846_end_mask_0, x = value_27_cast_fp16)[name = string("op_20846_cast_fp16")]; + tensor var_20850_begin_0 = const()[name = string("op_20850_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20850_end_0 = const()[name = string("op_20850_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_20850_end_mask_0 = const()[name = string("op_20850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20850_cast_fp16 = slice_by_index(begin = var_20850_begin_0, end = var_20850_end_0, end_mask = var_20850_end_mask_0, x = value_27_cast_fp16)[name = string("op_20850_cast_fp16")]; + tensor var_20854_begin_0 = const()[name = string("op_20854_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20854_end_0 = const()[name = string("op_20854_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_20854_end_mask_0 = const()[name = string("op_20854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20854_cast_fp16 = slice_by_index(begin = var_20854_begin_0, end = var_20854_end_0, end_mask = var_20854_end_mask_0, x = value_27_cast_fp16)[name = string("op_20854_cast_fp16")]; + tensor var_20858_begin_0 = const()[name = string("op_20858_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20858_end_0 = const()[name = string("op_20858_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_20858_end_mask_0 = const()[name = string("op_20858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20858_cast_fp16 = slice_by_index(begin = var_20858_begin_0, end = var_20858_end_0, end_mask = var_20858_end_mask_0, x = value_27_cast_fp16)[name = string("op_20858_cast_fp16")]; + tensor var_20862_begin_0 = const()[name = string("op_20862_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20862_end_0 = const()[name = string("op_20862_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_20862_end_mask_0 = const()[name = string("op_20862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20862_cast_fp16 = slice_by_index(begin = var_20862_begin_0, end = var_20862_end_0, end_mask = var_20862_end_mask_0, x = value_27_cast_fp16)[name = string("op_20862_cast_fp16")]; + tensor var_20866_begin_0 = const()[name = string("op_20866_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20866_end_0 = const()[name = string("op_20866_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_20866_end_mask_0 = const()[name = string("op_20866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20866_cast_fp16 = slice_by_index(begin = var_20866_begin_0, end = var_20866_end_0, end_mask = var_20866_end_mask_0, x = value_27_cast_fp16)[name = string("op_20866_cast_fp16")]; + tensor var_20870_begin_0 = const()[name = string("op_20870_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20870_end_0 = const()[name = string("op_20870_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_20870_end_mask_0 = const()[name = string("op_20870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20870_cast_fp16 = slice_by_index(begin = var_20870_begin_0, end = var_20870_end_0, end_mask = var_20870_end_mask_0, x = value_27_cast_fp16)[name = string("op_20870_cast_fp16")]; + tensor var_20874_begin_0 = const()[name = string("op_20874_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20874_end_0 = const()[name = string("op_20874_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_20874_end_mask_0 = const()[name = string("op_20874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20874_cast_fp16 = slice_by_index(begin = var_20874_begin_0, end = var_20874_end_0, end_mask = var_20874_end_mask_0, x = value_27_cast_fp16)[name = string("op_20874_cast_fp16")]; + tensor var_20878_begin_0 = const()[name = string("op_20878_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20878_end_0 = const()[name = string("op_20878_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_20878_end_mask_0 = const()[name = string("op_20878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20878_cast_fp16 = slice_by_index(begin = var_20878_begin_0, end = var_20878_end_0, end_mask = var_20878_end_mask_0, x = value_27_cast_fp16)[name = string("op_20878_cast_fp16")]; + tensor var_20882_begin_0 = const()[name = string("op_20882_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20882_end_0 = const()[name = string("op_20882_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_20882_end_mask_0 = const()[name = string("op_20882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20882_cast_fp16 = slice_by_index(begin = var_20882_begin_0, end = var_20882_end_0, end_mask = var_20882_end_mask_0, x = value_27_cast_fp16)[name = string("op_20882_cast_fp16")]; + tensor var_20886_begin_0 = const()[name = string("op_20886_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20886_end_0 = const()[name = string("op_20886_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_20886_end_mask_0 = const()[name = string("op_20886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20886_cast_fp16 = slice_by_index(begin = var_20886_begin_0, end = var_20886_end_0, end_mask = var_20886_end_mask_0, x = value_27_cast_fp16)[name = string("op_20886_cast_fp16")]; + tensor var_20890_begin_0 = const()[name = string("op_20890_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20890_end_0 = const()[name = string("op_20890_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_20890_end_mask_0 = const()[name = string("op_20890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20890_cast_fp16 = slice_by_index(begin = var_20890_begin_0, end = var_20890_end_0, end_mask = var_20890_end_mask_0, x = value_27_cast_fp16)[name = string("op_20890_cast_fp16")]; + tensor var_20894_begin_0 = const()[name = string("op_20894_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20894_end_0 = const()[name = string("op_20894_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_20894_end_mask_0 = const()[name = string("op_20894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20894_cast_fp16 = slice_by_index(begin = var_20894_begin_0, end = var_20894_end_0, end_mask = var_20894_end_mask_0, x = value_27_cast_fp16)[name = string("op_20894_cast_fp16")]; + tensor var_20898_begin_0 = const()[name = string("op_20898_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20898_end_0 = const()[name = string("op_20898_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_20898_end_mask_0 = const()[name = string("op_20898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20898_cast_fp16 = slice_by_index(begin = var_20898_begin_0, end = var_20898_end_0, end_mask = var_20898_end_mask_0, x = value_27_cast_fp16)[name = string("op_20898_cast_fp16")]; + tensor var_20902_begin_0 = const()[name = string("op_20902_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20902_end_0 = const()[name = string("op_20902_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_20902_end_mask_0 = const()[name = string("op_20902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20902_cast_fp16 = slice_by_index(begin = var_20902_begin_0, end = var_20902_end_0, end_mask = var_20902_end_mask_0, x = value_27_cast_fp16)[name = string("op_20902_cast_fp16")]; + tensor var_20906_begin_0 = const()[name = string("op_20906_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20906_end_0 = const()[name = string("op_20906_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_20906_end_mask_0 = const()[name = string("op_20906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20906_cast_fp16 = slice_by_index(begin = var_20906_begin_0, end = var_20906_end_0, end_mask = var_20906_end_mask_0, x = value_27_cast_fp16)[name = string("op_20906_cast_fp16")]; + tensor var_20910_begin_0 = const()[name = string("op_20910_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20910_end_0 = const()[name = string("op_20910_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_20910_end_mask_0 = const()[name = string("op_20910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20910_cast_fp16 = slice_by_index(begin = var_20910_begin_0, end = var_20910_end_0, end_mask = var_20910_end_mask_0, x = value_27_cast_fp16)[name = string("op_20910_cast_fp16")]; + string _SplitHeadsQ__mh_w_2081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2081_equation_0, values = (var_20756_cast_fp16, var_20198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2081_cast_fp16")]; + string _SplitHeadsQ__mh_w_2083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2083_equation_0, values = (var_20756_cast_fp16, var_20205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2083_cast_fp16")]; + string _SplitHeadsQ__mh_w_2085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2085_equation_0, values = (var_20756_cast_fp16, var_20212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2085_cast_fp16")]; + string _SplitHeadsQ__mh_w_2087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2087_equation_0, values = (var_20756_cast_fp16, var_20219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2087_cast_fp16")]; + string _SplitHeadsQ__mh_w_2089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2089_equation_0, values = (var_20760_cast_fp16, var_20226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2089_cast_fp16")]; + string _SplitHeadsQ__mh_w_2091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2091_equation_0, values = (var_20760_cast_fp16, var_20233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2091_cast_fp16")]; + string _SplitHeadsQ__mh_w_2093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2093_equation_0, values = (var_20760_cast_fp16, var_20240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2093_cast_fp16")]; + string _SplitHeadsQ__mh_w_2095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2095_equation_0, values = (var_20760_cast_fp16, var_20247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2095_cast_fp16")]; + string _SplitHeadsQ__mh_w_2097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2097_equation_0, values = (var_20764_cast_fp16, var_20254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2097_cast_fp16")]; + string _SplitHeadsQ__mh_w_2099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2099_equation_0, values = (var_20764_cast_fp16, var_20261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2099_cast_fp16")]; + string _SplitHeadsQ__mh_w_2101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2101_equation_0, values = (var_20764_cast_fp16, var_20268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2101_cast_fp16")]; + string _SplitHeadsQ__mh_w_2103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2103_equation_0, values = (var_20764_cast_fp16, var_20275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2103_cast_fp16")]; + string _SplitHeadsQ__mh_w_2105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2105_equation_0, values = (var_20768_cast_fp16, var_20282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2105_cast_fp16")]; + string _SplitHeadsQ__mh_w_2107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2107_equation_0, values = (var_20768_cast_fp16, var_20289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2107_cast_fp16")]; + string _SplitHeadsQ__mh_w_2109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2109_equation_0, values = (var_20768_cast_fp16, var_20296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2109_cast_fp16")]; + string _SplitHeadsQ__mh_w_2111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2111_equation_0, values = (var_20768_cast_fp16, var_20303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2111_cast_fp16")]; + string _SplitHeadsQ__mh_w_2113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2113_equation_0, values = (var_20772_cast_fp16, var_20310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2113_cast_fp16")]; + string _SplitHeadsQ__mh_w_2115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2115_equation_0, values = (var_20772_cast_fp16, var_20317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2115_cast_fp16")]; + string _SplitHeadsQ__mh_w_2117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2117_equation_0, values = (var_20772_cast_fp16, var_20324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2117_cast_fp16")]; + string _SplitHeadsQ__mh_w_2119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2119_equation_0, values = (var_20772_cast_fp16, var_20331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2119_cast_fp16")]; + string _SplitHeadsQ__mh_w_2121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2121_equation_0, values = (var_20776_cast_fp16, var_20338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2121_cast_fp16")]; + string _SplitHeadsQ__mh_w_2123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2123_equation_0, values = (var_20776_cast_fp16, var_20345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2123_cast_fp16")]; + string _SplitHeadsQ__mh_w_2125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2125_equation_0, values = (var_20776_cast_fp16, var_20352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2125_cast_fp16")]; + string _SplitHeadsQ__mh_w_2127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2127_equation_0, values = (var_20776_cast_fp16, var_20359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2127_cast_fp16")]; + string _SplitHeadsQ__mh_w_2129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2129_equation_0, values = (var_20780_cast_fp16, var_20366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2129_cast_fp16")]; + string _SplitHeadsQ__mh_w_2131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2131_equation_0, values = (var_20780_cast_fp16, var_20373_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2131_cast_fp16")]; + string _SplitHeadsQ__mh_w_2133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2133_equation_0, values = (var_20780_cast_fp16, var_20380_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2133_cast_fp16")]; + string _SplitHeadsQ__mh_w_2135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2135_equation_0, values = (var_20780_cast_fp16, var_20387_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2135_cast_fp16")]; + string _SplitHeadsQ__mh_w_2137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2137_equation_0, values = (var_20784_cast_fp16, var_20394_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2137_cast_fp16")]; + string _SplitHeadsQ__mh_w_2139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2139_equation_0, values = (var_20784_cast_fp16, var_20401_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2139_cast_fp16")]; + string _SplitHeadsQ__mh_w_2141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2141_equation_0, values = (var_20784_cast_fp16, var_20408_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2141_cast_fp16")]; + string _SplitHeadsQ__mh_w_2143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2143_equation_0, values = (var_20784_cast_fp16, var_20415_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2143_cast_fp16")]; + string _SplitHeadsQ__mh_w_2145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2145_equation_0, values = (var_20788_cast_fp16, var_20422_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2145_cast_fp16")]; + string _SplitHeadsQ__mh_w_2147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2147_equation_0, values = (var_20788_cast_fp16, var_20429_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2147_cast_fp16")]; + string _SplitHeadsQ__mh_w_2149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2149_equation_0, values = (var_20788_cast_fp16, var_20436_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2149_cast_fp16")]; + string _SplitHeadsQ__mh_w_2151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2151_equation_0, values = (var_20788_cast_fp16, var_20443_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2151_cast_fp16")]; + string _SplitHeadsQ__mh_w_2153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2153_equation_0, values = (var_20792_cast_fp16, var_20450_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2153_cast_fp16")]; + string _SplitHeadsQ__mh_w_2155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2155_equation_0, values = (var_20792_cast_fp16, var_20457_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2155_cast_fp16")]; + string _SplitHeadsQ__mh_w_2157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2157_equation_0, values = (var_20792_cast_fp16, var_20464_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2157_cast_fp16")]; + string _SplitHeadsQ__mh_w_2159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2159_equation_0, values = (var_20792_cast_fp16, var_20471_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2159_cast_fp16")]; + string _SplitHeadsQ__mh_w_2161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2161_equation_0, values = (var_20796_cast_fp16, var_20478_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2161_cast_fp16")]; + string _SplitHeadsQ__mh_w_2163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2163_equation_0, values = (var_20796_cast_fp16, var_20485_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2163_cast_fp16")]; + string _SplitHeadsQ__mh_w_2165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2165_equation_0, values = (var_20796_cast_fp16, var_20492_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2165_cast_fp16")]; + string _SplitHeadsQ__mh_w_2167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2167_equation_0, values = (var_20796_cast_fp16, var_20499_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2167_cast_fp16")]; + string _SplitHeadsQ__mh_w_2169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2169_equation_0, values = (var_20800_cast_fp16, var_20506_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2169_cast_fp16")]; + string _SplitHeadsQ__mh_w_2171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2171_equation_0, values = (var_20800_cast_fp16, var_20513_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2171_cast_fp16")]; + string _SplitHeadsQ__mh_w_2173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2173_equation_0, values = (var_20800_cast_fp16, var_20520_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2173_cast_fp16")]; + string _SplitHeadsQ__mh_w_2175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2175_equation_0, values = (var_20800_cast_fp16, var_20527_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2175_cast_fp16")]; + string _SplitHeadsQ__mh_w_2177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2177_equation_0, values = (var_20804_cast_fp16, var_20534_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2177_cast_fp16")]; + string _SplitHeadsQ__mh_w_2179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2179_equation_0, values = (var_20804_cast_fp16, var_20541_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2179_cast_fp16")]; + string _SplitHeadsQ__mh_w_2181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2181_equation_0, values = (var_20804_cast_fp16, var_20548_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2181_cast_fp16")]; + string _SplitHeadsQ__mh_w_2183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2183_equation_0, values = (var_20804_cast_fp16, var_20555_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2183_cast_fp16")]; + string _SplitHeadsQ__mh_w_2185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2185_equation_0, values = (var_20808_cast_fp16, var_20562_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2185_cast_fp16")]; + string _SplitHeadsQ__mh_w_2187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2187_equation_0, values = (var_20808_cast_fp16, var_20569_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2187_cast_fp16")]; + string _SplitHeadsQ__mh_w_2189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2189_equation_0, values = (var_20808_cast_fp16, var_20576_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2189_cast_fp16")]; + string _SplitHeadsQ__mh_w_2191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2191_equation_0, values = (var_20808_cast_fp16, var_20583_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2191_cast_fp16")]; + string _SplitHeadsQ__mh_w_2193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2193_equation_0, values = (var_20812_cast_fp16, var_20590_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2193_cast_fp16")]; + string _SplitHeadsQ__mh_w_2195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2195_equation_0, values = (var_20812_cast_fp16, var_20597_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2195_cast_fp16")]; + string _SplitHeadsQ__mh_w_2197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2197_equation_0, values = (var_20812_cast_fp16, var_20604_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2197_cast_fp16")]; + string _SplitHeadsQ__mh_w_2199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2199_equation_0, values = (var_20812_cast_fp16, var_20611_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2199_cast_fp16")]; + string _SplitHeadsQ__mh_w_2201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2201_equation_0, values = (var_20816_cast_fp16, var_20618_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2201_cast_fp16")]; + string _SplitHeadsQ__mh_w_2203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2203_equation_0, values = (var_20816_cast_fp16, var_20625_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2203_cast_fp16")]; + string _SplitHeadsQ__mh_w_2205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2205_equation_0, values = (var_20816_cast_fp16, var_20632_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2205_cast_fp16")]; + string _SplitHeadsQ__mh_w_2207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2207_equation_0, values = (var_20816_cast_fp16, var_20639_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2207_cast_fp16")]; + string _SplitHeadsQ__mh_w_2209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2209_equation_0, values = (var_20820_cast_fp16, var_20646_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2209_cast_fp16")]; + string _SplitHeadsQ__mh_w_2211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2211_equation_0, values = (var_20820_cast_fp16, var_20653_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2211_cast_fp16")]; + string _SplitHeadsQ__mh_w_2213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2213_equation_0, values = (var_20820_cast_fp16, var_20660_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2213_cast_fp16")]; + string _SplitHeadsQ__mh_w_2215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2215_equation_0, values = (var_20820_cast_fp16, var_20667_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2215_cast_fp16")]; + string _SplitHeadsQ__mh_w_2217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2217_equation_0, values = (var_20824_cast_fp16, var_20674_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2217_cast_fp16")]; + string _SplitHeadsQ__mh_w_2219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2219_equation_0, values = (var_20824_cast_fp16, var_20681_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2219_cast_fp16")]; + string _SplitHeadsQ__mh_w_2221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2221_equation_0, values = (var_20824_cast_fp16, var_20688_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2221_cast_fp16")]; + string _SplitHeadsQ__mh_w_2223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2223_equation_0, values = (var_20824_cast_fp16, var_20695_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2223_cast_fp16")]; + string _SplitHeadsQ__mh_w_2225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2225_equation_0, values = (var_20828_cast_fp16, var_20702_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2225_cast_fp16")]; + string _SplitHeadsQ__mh_w_2227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2227_equation_0, values = (var_20828_cast_fp16, var_20709_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2227_cast_fp16")]; + string _SplitHeadsQ__mh_w_2229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2229_equation_0, values = (var_20828_cast_fp16, var_20716_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2229_cast_fp16")]; + string _SplitHeadsQ__mh_w_2231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2231_equation_0, values = (var_20828_cast_fp16, var_20723_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2231_cast_fp16")]; + string _SplitHeadsQ__mh_w_2233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2233_equation_0, values = (var_20832_cast_fp16, var_20730_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2233_cast_fp16")]; + string _SplitHeadsQ__mh_w_2235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2235_equation_0, values = (var_20832_cast_fp16, var_20737_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2235_cast_fp16")]; + string _SplitHeadsQ__mh_w_2237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2237_equation_0, values = (var_20832_cast_fp16, var_20744_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2237_cast_fp16")]; + string _SplitHeadsQ__mh_w_2239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2239_equation_0, values = (var_20832_cast_fp16, var_20751_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2239_cast_fp16")]; + fp16 var_21073_to_fp16 = const()[name = string("op_21073_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2081_cast_fp16, y = var_21073_to_fp16)[name = string("aw_chunk_2081_cast_fp16")]; + fp16 var_21075_to_fp16 = const()[name = string("op_21075_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2083_cast_fp16, y = var_21075_to_fp16)[name = string("aw_chunk_2083_cast_fp16")]; + fp16 var_21077_to_fp16 = const()[name = string("op_21077_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2085_cast_fp16, y = var_21077_to_fp16)[name = string("aw_chunk_2085_cast_fp16")]; + fp16 var_21079_to_fp16 = const()[name = string("op_21079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2087_cast_fp16, y = var_21079_to_fp16)[name = string("aw_chunk_2087_cast_fp16")]; + fp16 var_21081_to_fp16 = const()[name = string("op_21081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2089_cast_fp16, y = var_21081_to_fp16)[name = string("aw_chunk_2089_cast_fp16")]; + fp16 var_21083_to_fp16 = const()[name = string("op_21083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2091_cast_fp16, y = var_21083_to_fp16)[name = string("aw_chunk_2091_cast_fp16")]; + fp16 var_21085_to_fp16 = const()[name = string("op_21085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2093_cast_fp16, y = var_21085_to_fp16)[name = string("aw_chunk_2093_cast_fp16")]; + fp16 var_21087_to_fp16 = const()[name = string("op_21087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2095_cast_fp16, y = var_21087_to_fp16)[name = string("aw_chunk_2095_cast_fp16")]; + fp16 var_21089_to_fp16 = const()[name = string("op_21089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2097_cast_fp16, y = var_21089_to_fp16)[name = string("aw_chunk_2097_cast_fp16")]; + fp16 var_21091_to_fp16 = const()[name = string("op_21091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2099_cast_fp16, y = var_21091_to_fp16)[name = string("aw_chunk_2099_cast_fp16")]; + fp16 var_21093_to_fp16 = const()[name = string("op_21093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2101_cast_fp16, y = var_21093_to_fp16)[name = string("aw_chunk_2101_cast_fp16")]; + fp16 var_21095_to_fp16 = const()[name = string("op_21095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2103_cast_fp16, y = var_21095_to_fp16)[name = string("aw_chunk_2103_cast_fp16")]; + fp16 var_21097_to_fp16 = const()[name = string("op_21097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2105_cast_fp16, y = var_21097_to_fp16)[name = string("aw_chunk_2105_cast_fp16")]; + fp16 var_21099_to_fp16 = const()[name = string("op_21099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2107_cast_fp16, y = var_21099_to_fp16)[name = string("aw_chunk_2107_cast_fp16")]; + fp16 var_21101_to_fp16 = const()[name = string("op_21101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2109_cast_fp16, y = var_21101_to_fp16)[name = string("aw_chunk_2109_cast_fp16")]; + fp16 var_21103_to_fp16 = const()[name = string("op_21103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2111_cast_fp16, y = var_21103_to_fp16)[name = string("aw_chunk_2111_cast_fp16")]; + fp16 var_21105_to_fp16 = const()[name = string("op_21105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2113_cast_fp16, y = var_21105_to_fp16)[name = string("aw_chunk_2113_cast_fp16")]; + fp16 var_21107_to_fp16 = const()[name = string("op_21107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2115_cast_fp16, y = var_21107_to_fp16)[name = string("aw_chunk_2115_cast_fp16")]; + fp16 var_21109_to_fp16 = const()[name = string("op_21109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2117_cast_fp16, y = var_21109_to_fp16)[name = string("aw_chunk_2117_cast_fp16")]; + fp16 var_21111_to_fp16 = const()[name = string("op_21111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2119_cast_fp16, y = var_21111_to_fp16)[name = string("aw_chunk_2119_cast_fp16")]; + fp16 var_21113_to_fp16 = const()[name = string("op_21113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2121_cast_fp16, y = var_21113_to_fp16)[name = string("aw_chunk_2121_cast_fp16")]; + fp16 var_21115_to_fp16 = const()[name = string("op_21115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2123_cast_fp16, y = var_21115_to_fp16)[name = string("aw_chunk_2123_cast_fp16")]; + fp16 var_21117_to_fp16 = const()[name = string("op_21117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2125_cast_fp16, y = var_21117_to_fp16)[name = string("aw_chunk_2125_cast_fp16")]; + fp16 var_21119_to_fp16 = const()[name = string("op_21119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2127_cast_fp16, y = var_21119_to_fp16)[name = string("aw_chunk_2127_cast_fp16")]; + fp16 var_21121_to_fp16 = const()[name = string("op_21121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2129_cast_fp16, y = var_21121_to_fp16)[name = string("aw_chunk_2129_cast_fp16")]; + fp16 var_21123_to_fp16 = const()[name = string("op_21123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2131_cast_fp16, y = var_21123_to_fp16)[name = string("aw_chunk_2131_cast_fp16")]; + fp16 var_21125_to_fp16 = const()[name = string("op_21125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2133_cast_fp16, y = var_21125_to_fp16)[name = string("aw_chunk_2133_cast_fp16")]; + fp16 var_21127_to_fp16 = const()[name = string("op_21127_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2135_cast_fp16, y = var_21127_to_fp16)[name = string("aw_chunk_2135_cast_fp16")]; + fp16 var_21129_to_fp16 = const()[name = string("op_21129_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2137_cast_fp16, y = var_21129_to_fp16)[name = string("aw_chunk_2137_cast_fp16")]; + fp16 var_21131_to_fp16 = const()[name = string("op_21131_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2139_cast_fp16, y = var_21131_to_fp16)[name = string("aw_chunk_2139_cast_fp16")]; + fp16 var_21133_to_fp16 = const()[name = string("op_21133_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2141_cast_fp16, y = var_21133_to_fp16)[name = string("aw_chunk_2141_cast_fp16")]; + fp16 var_21135_to_fp16 = const()[name = string("op_21135_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2143_cast_fp16, y = var_21135_to_fp16)[name = string("aw_chunk_2143_cast_fp16")]; + fp16 var_21137_to_fp16 = const()[name = string("op_21137_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2145_cast_fp16, y = var_21137_to_fp16)[name = string("aw_chunk_2145_cast_fp16")]; + fp16 var_21139_to_fp16 = const()[name = string("op_21139_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2147_cast_fp16, y = var_21139_to_fp16)[name = string("aw_chunk_2147_cast_fp16")]; + fp16 var_21141_to_fp16 = const()[name = string("op_21141_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2149_cast_fp16, y = var_21141_to_fp16)[name = string("aw_chunk_2149_cast_fp16")]; + fp16 var_21143_to_fp16 = const()[name = string("op_21143_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2151_cast_fp16, y = var_21143_to_fp16)[name = string("aw_chunk_2151_cast_fp16")]; + fp16 var_21145_to_fp16 = const()[name = string("op_21145_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2153_cast_fp16, y = var_21145_to_fp16)[name = string("aw_chunk_2153_cast_fp16")]; + fp16 var_21147_to_fp16 = const()[name = string("op_21147_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2155_cast_fp16, y = var_21147_to_fp16)[name = string("aw_chunk_2155_cast_fp16")]; + fp16 var_21149_to_fp16 = const()[name = string("op_21149_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2157_cast_fp16, y = var_21149_to_fp16)[name = string("aw_chunk_2157_cast_fp16")]; + fp16 var_21151_to_fp16 = const()[name = string("op_21151_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2159_cast_fp16, y = var_21151_to_fp16)[name = string("aw_chunk_2159_cast_fp16")]; + fp16 var_21153_to_fp16 = const()[name = string("op_21153_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2161_cast_fp16, y = var_21153_to_fp16)[name = string("aw_chunk_2161_cast_fp16")]; + fp16 var_21155_to_fp16 = const()[name = string("op_21155_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2163_cast_fp16, y = var_21155_to_fp16)[name = string("aw_chunk_2163_cast_fp16")]; + fp16 var_21157_to_fp16 = const()[name = string("op_21157_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2165_cast_fp16, y = var_21157_to_fp16)[name = string("aw_chunk_2165_cast_fp16")]; + fp16 var_21159_to_fp16 = const()[name = string("op_21159_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2167_cast_fp16, y = var_21159_to_fp16)[name = string("aw_chunk_2167_cast_fp16")]; + fp16 var_21161_to_fp16 = const()[name = string("op_21161_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2169_cast_fp16, y = var_21161_to_fp16)[name = string("aw_chunk_2169_cast_fp16")]; + fp16 var_21163_to_fp16 = const()[name = string("op_21163_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2171_cast_fp16, y = var_21163_to_fp16)[name = string("aw_chunk_2171_cast_fp16")]; + fp16 var_21165_to_fp16 = const()[name = string("op_21165_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2173_cast_fp16, y = var_21165_to_fp16)[name = string("aw_chunk_2173_cast_fp16")]; + fp16 var_21167_to_fp16 = const()[name = string("op_21167_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2175_cast_fp16, y = var_21167_to_fp16)[name = string("aw_chunk_2175_cast_fp16")]; + fp16 var_21169_to_fp16 = const()[name = string("op_21169_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2177_cast_fp16, y = var_21169_to_fp16)[name = string("aw_chunk_2177_cast_fp16")]; + fp16 var_21171_to_fp16 = const()[name = string("op_21171_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2179_cast_fp16, y = var_21171_to_fp16)[name = string("aw_chunk_2179_cast_fp16")]; + fp16 var_21173_to_fp16 = const()[name = string("op_21173_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2181_cast_fp16, y = var_21173_to_fp16)[name = string("aw_chunk_2181_cast_fp16")]; + fp16 var_21175_to_fp16 = const()[name = string("op_21175_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2183_cast_fp16, y = var_21175_to_fp16)[name = string("aw_chunk_2183_cast_fp16")]; + fp16 var_21177_to_fp16 = const()[name = string("op_21177_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2185_cast_fp16, y = var_21177_to_fp16)[name = string("aw_chunk_2185_cast_fp16")]; + fp16 var_21179_to_fp16 = const()[name = string("op_21179_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2187_cast_fp16, y = var_21179_to_fp16)[name = string("aw_chunk_2187_cast_fp16")]; + fp16 var_21181_to_fp16 = const()[name = string("op_21181_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2189_cast_fp16, y = var_21181_to_fp16)[name = string("aw_chunk_2189_cast_fp16")]; + fp16 var_21183_to_fp16 = const()[name = string("op_21183_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2191_cast_fp16, y = var_21183_to_fp16)[name = string("aw_chunk_2191_cast_fp16")]; + fp16 var_21185_to_fp16 = const()[name = string("op_21185_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2193_cast_fp16, y = var_21185_to_fp16)[name = string("aw_chunk_2193_cast_fp16")]; + fp16 var_21187_to_fp16 = const()[name = string("op_21187_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2195_cast_fp16, y = var_21187_to_fp16)[name = string("aw_chunk_2195_cast_fp16")]; + fp16 var_21189_to_fp16 = const()[name = string("op_21189_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2197_cast_fp16, y = var_21189_to_fp16)[name = string("aw_chunk_2197_cast_fp16")]; + fp16 var_21191_to_fp16 = const()[name = string("op_21191_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2199_cast_fp16, y = var_21191_to_fp16)[name = string("aw_chunk_2199_cast_fp16")]; + fp16 var_21193_to_fp16 = const()[name = string("op_21193_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2201_cast_fp16, y = var_21193_to_fp16)[name = string("aw_chunk_2201_cast_fp16")]; + fp16 var_21195_to_fp16 = const()[name = string("op_21195_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2203_cast_fp16, y = var_21195_to_fp16)[name = string("aw_chunk_2203_cast_fp16")]; + fp16 var_21197_to_fp16 = const()[name = string("op_21197_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2205_cast_fp16, y = var_21197_to_fp16)[name = string("aw_chunk_2205_cast_fp16")]; + fp16 var_21199_to_fp16 = const()[name = string("op_21199_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2207_cast_fp16, y = var_21199_to_fp16)[name = string("aw_chunk_2207_cast_fp16")]; + fp16 var_21201_to_fp16 = const()[name = string("op_21201_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2209_cast_fp16, y = var_21201_to_fp16)[name = string("aw_chunk_2209_cast_fp16")]; + fp16 var_21203_to_fp16 = const()[name = string("op_21203_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2211_cast_fp16, y = var_21203_to_fp16)[name = string("aw_chunk_2211_cast_fp16")]; + fp16 var_21205_to_fp16 = const()[name = string("op_21205_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2213_cast_fp16, y = var_21205_to_fp16)[name = string("aw_chunk_2213_cast_fp16")]; + fp16 var_21207_to_fp16 = const()[name = string("op_21207_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2215_cast_fp16, y = var_21207_to_fp16)[name = string("aw_chunk_2215_cast_fp16")]; + fp16 var_21209_to_fp16 = const()[name = string("op_21209_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2217_cast_fp16, y = var_21209_to_fp16)[name = string("aw_chunk_2217_cast_fp16")]; + fp16 var_21211_to_fp16 = const()[name = string("op_21211_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2219_cast_fp16, y = var_21211_to_fp16)[name = string("aw_chunk_2219_cast_fp16")]; + fp16 var_21213_to_fp16 = const()[name = string("op_21213_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2221_cast_fp16, y = var_21213_to_fp16)[name = string("aw_chunk_2221_cast_fp16")]; + fp16 var_21215_to_fp16 = const()[name = string("op_21215_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2223_cast_fp16, y = var_21215_to_fp16)[name = string("aw_chunk_2223_cast_fp16")]; + fp16 var_21217_to_fp16 = const()[name = string("op_21217_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2225_cast_fp16, y = var_21217_to_fp16)[name = string("aw_chunk_2225_cast_fp16")]; + fp16 var_21219_to_fp16 = const()[name = string("op_21219_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2227_cast_fp16, y = var_21219_to_fp16)[name = string("aw_chunk_2227_cast_fp16")]; + fp16 var_21221_to_fp16 = const()[name = string("op_21221_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2229_cast_fp16, y = var_21221_to_fp16)[name = string("aw_chunk_2229_cast_fp16")]; + fp16 var_21223_to_fp16 = const()[name = string("op_21223_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2231_cast_fp16, y = var_21223_to_fp16)[name = string("aw_chunk_2231_cast_fp16")]; + fp16 var_21225_to_fp16 = const()[name = string("op_21225_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2233_cast_fp16, y = var_21225_to_fp16)[name = string("aw_chunk_2233_cast_fp16")]; + fp16 var_21227_to_fp16 = const()[name = string("op_21227_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2235_cast_fp16, y = var_21227_to_fp16)[name = string("aw_chunk_2235_cast_fp16")]; + fp16 var_21229_to_fp16 = const()[name = string("op_21229_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2237_cast_fp16, y = var_21229_to_fp16)[name = string("aw_chunk_2237_cast_fp16")]; + fp16 var_21231_to_fp16 = const()[name = string("op_21231_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2239_cast_fp16, y = var_21231_to_fp16)[name = string("aw_chunk_2239_cast_fp16")]; + tensor var_21233_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2081_cast_fp16)[name = string("op_21233_cast_fp16")]; + tensor var_21234_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2083_cast_fp16)[name = string("op_21234_cast_fp16")]; + tensor var_21235_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2085_cast_fp16)[name = string("op_21235_cast_fp16")]; + tensor var_21236_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2087_cast_fp16)[name = string("op_21236_cast_fp16")]; + tensor var_21237_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2089_cast_fp16)[name = string("op_21237_cast_fp16")]; + tensor var_21238_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2091_cast_fp16)[name = string("op_21238_cast_fp16")]; + tensor var_21239_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2093_cast_fp16)[name = string("op_21239_cast_fp16")]; + tensor var_21240_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2095_cast_fp16)[name = string("op_21240_cast_fp16")]; + tensor var_21241_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2097_cast_fp16)[name = string("op_21241_cast_fp16")]; + tensor var_21242_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2099_cast_fp16)[name = string("op_21242_cast_fp16")]; + tensor var_21243_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2101_cast_fp16)[name = string("op_21243_cast_fp16")]; + tensor var_21244_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2103_cast_fp16)[name = string("op_21244_cast_fp16")]; + tensor var_21245_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2105_cast_fp16)[name = string("op_21245_cast_fp16")]; + tensor var_21246_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2107_cast_fp16)[name = string("op_21246_cast_fp16")]; + tensor var_21247_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2109_cast_fp16)[name = string("op_21247_cast_fp16")]; + tensor var_21248_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2111_cast_fp16)[name = string("op_21248_cast_fp16")]; + tensor var_21249_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2113_cast_fp16)[name = string("op_21249_cast_fp16")]; + tensor var_21250_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2115_cast_fp16)[name = string("op_21250_cast_fp16")]; + tensor var_21251_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2117_cast_fp16)[name = string("op_21251_cast_fp16")]; + tensor var_21252_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2119_cast_fp16)[name = string("op_21252_cast_fp16")]; + tensor var_21253_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2121_cast_fp16)[name = string("op_21253_cast_fp16")]; + tensor var_21254_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2123_cast_fp16)[name = string("op_21254_cast_fp16")]; + tensor var_21255_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2125_cast_fp16)[name = string("op_21255_cast_fp16")]; + tensor var_21256_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2127_cast_fp16)[name = string("op_21256_cast_fp16")]; + tensor var_21257_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2129_cast_fp16)[name = string("op_21257_cast_fp16")]; + tensor var_21258_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2131_cast_fp16)[name = string("op_21258_cast_fp16")]; + tensor var_21259_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2133_cast_fp16)[name = string("op_21259_cast_fp16")]; + tensor var_21260_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2135_cast_fp16)[name = string("op_21260_cast_fp16")]; + tensor var_21261_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2137_cast_fp16)[name = string("op_21261_cast_fp16")]; + tensor var_21262_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2139_cast_fp16)[name = string("op_21262_cast_fp16")]; + tensor var_21263_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2141_cast_fp16)[name = string("op_21263_cast_fp16")]; + tensor var_21264_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2143_cast_fp16)[name = string("op_21264_cast_fp16")]; + tensor var_21265_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2145_cast_fp16)[name = string("op_21265_cast_fp16")]; + tensor var_21266_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2147_cast_fp16)[name = string("op_21266_cast_fp16")]; + tensor var_21267_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2149_cast_fp16)[name = string("op_21267_cast_fp16")]; + tensor var_21268_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2151_cast_fp16)[name = string("op_21268_cast_fp16")]; + tensor var_21269_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2153_cast_fp16)[name = string("op_21269_cast_fp16")]; + tensor var_21270_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2155_cast_fp16)[name = string("op_21270_cast_fp16")]; + tensor var_21271_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2157_cast_fp16)[name = string("op_21271_cast_fp16")]; + tensor var_21272_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2159_cast_fp16)[name = string("op_21272_cast_fp16")]; + tensor var_21273_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2161_cast_fp16)[name = string("op_21273_cast_fp16")]; + tensor var_21274_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2163_cast_fp16)[name = string("op_21274_cast_fp16")]; + tensor var_21275_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2165_cast_fp16)[name = string("op_21275_cast_fp16")]; + tensor var_21276_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2167_cast_fp16)[name = string("op_21276_cast_fp16")]; + tensor var_21277_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2169_cast_fp16)[name = string("op_21277_cast_fp16")]; + tensor var_21278_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2171_cast_fp16)[name = string("op_21278_cast_fp16")]; + tensor var_21279_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2173_cast_fp16)[name = string("op_21279_cast_fp16")]; + tensor var_21280_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2175_cast_fp16)[name = string("op_21280_cast_fp16")]; + tensor var_21281_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2177_cast_fp16)[name = string("op_21281_cast_fp16")]; + tensor var_21282_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2179_cast_fp16)[name = string("op_21282_cast_fp16")]; + tensor var_21283_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2181_cast_fp16)[name = string("op_21283_cast_fp16")]; + tensor var_21284_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2183_cast_fp16)[name = string("op_21284_cast_fp16")]; + tensor var_21285_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2185_cast_fp16)[name = string("op_21285_cast_fp16")]; + tensor var_21286_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2187_cast_fp16)[name = string("op_21286_cast_fp16")]; + tensor var_21287_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2189_cast_fp16)[name = string("op_21287_cast_fp16")]; + tensor var_21288_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2191_cast_fp16)[name = string("op_21288_cast_fp16")]; + tensor var_21289_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2193_cast_fp16)[name = string("op_21289_cast_fp16")]; + tensor var_21290_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2195_cast_fp16)[name = string("op_21290_cast_fp16")]; + tensor var_21291_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2197_cast_fp16)[name = string("op_21291_cast_fp16")]; + tensor var_21292_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2199_cast_fp16)[name = string("op_21292_cast_fp16")]; + tensor var_21293_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2201_cast_fp16)[name = string("op_21293_cast_fp16")]; + tensor var_21294_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2203_cast_fp16)[name = string("op_21294_cast_fp16")]; + tensor var_21295_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2205_cast_fp16)[name = string("op_21295_cast_fp16")]; + tensor var_21296_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2207_cast_fp16)[name = string("op_21296_cast_fp16")]; + tensor var_21297_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2209_cast_fp16)[name = string("op_21297_cast_fp16")]; + tensor var_21298_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2211_cast_fp16)[name = string("op_21298_cast_fp16")]; + tensor var_21299_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2213_cast_fp16)[name = string("op_21299_cast_fp16")]; + tensor var_21300_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2215_cast_fp16)[name = string("op_21300_cast_fp16")]; + tensor var_21301_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2217_cast_fp16)[name = string("op_21301_cast_fp16")]; + tensor var_21302_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2219_cast_fp16)[name = string("op_21302_cast_fp16")]; + tensor var_21303_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2221_cast_fp16)[name = string("op_21303_cast_fp16")]; + tensor var_21304_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2223_cast_fp16)[name = string("op_21304_cast_fp16")]; + tensor var_21305_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2225_cast_fp16)[name = string("op_21305_cast_fp16")]; + tensor var_21306_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2227_cast_fp16)[name = string("op_21306_cast_fp16")]; + tensor var_21307_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2229_cast_fp16)[name = string("op_21307_cast_fp16")]; + tensor var_21308_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2231_cast_fp16)[name = string("op_21308_cast_fp16")]; + tensor var_21309_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2233_cast_fp16)[name = string("op_21309_cast_fp16")]; + tensor var_21310_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2235_cast_fp16)[name = string("op_21310_cast_fp16")]; + tensor var_21311_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2237_cast_fp16)[name = string("op_21311_cast_fp16")]; + tensor var_21312_cast_fp16 = softmax(axis = var_20058, x = aw_chunk_2239_cast_fp16)[name = string("op_21312_cast_fp16")]; + string var_21314_equation_0 = const()[name = string("op_21314_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21314_cast_fp16 = einsum(equation = var_21314_equation_0, values = (var_20834_cast_fp16, var_21233_cast_fp16))[name = string("op_21314_cast_fp16")]; + string var_21316_equation_0 = const()[name = string("op_21316_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21316_cast_fp16 = einsum(equation = var_21316_equation_0, values = (var_20834_cast_fp16, var_21234_cast_fp16))[name = string("op_21316_cast_fp16")]; + string var_21318_equation_0 = const()[name = string("op_21318_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21318_cast_fp16 = einsum(equation = var_21318_equation_0, values = (var_20834_cast_fp16, var_21235_cast_fp16))[name = string("op_21318_cast_fp16")]; + string var_21320_equation_0 = const()[name = string("op_21320_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21320_cast_fp16 = einsum(equation = var_21320_equation_0, values = (var_20834_cast_fp16, var_21236_cast_fp16))[name = string("op_21320_cast_fp16")]; + string var_21322_equation_0 = const()[name = string("op_21322_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21322_cast_fp16 = einsum(equation = var_21322_equation_0, values = (var_20838_cast_fp16, var_21237_cast_fp16))[name = string("op_21322_cast_fp16")]; + string var_21324_equation_0 = const()[name = string("op_21324_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21324_cast_fp16 = einsum(equation = var_21324_equation_0, values = (var_20838_cast_fp16, var_21238_cast_fp16))[name = string("op_21324_cast_fp16")]; + string var_21326_equation_0 = const()[name = string("op_21326_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21326_cast_fp16 = einsum(equation = var_21326_equation_0, values = (var_20838_cast_fp16, var_21239_cast_fp16))[name = string("op_21326_cast_fp16")]; + string var_21328_equation_0 = const()[name = string("op_21328_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21328_cast_fp16 = einsum(equation = var_21328_equation_0, values = (var_20838_cast_fp16, var_21240_cast_fp16))[name = string("op_21328_cast_fp16")]; + string var_21330_equation_0 = const()[name = string("op_21330_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21330_cast_fp16 = einsum(equation = var_21330_equation_0, values = (var_20842_cast_fp16, var_21241_cast_fp16))[name = string("op_21330_cast_fp16")]; + string var_21332_equation_0 = const()[name = string("op_21332_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21332_cast_fp16 = einsum(equation = var_21332_equation_0, values = (var_20842_cast_fp16, var_21242_cast_fp16))[name = string("op_21332_cast_fp16")]; + string var_21334_equation_0 = const()[name = string("op_21334_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21334_cast_fp16 = einsum(equation = var_21334_equation_0, values = (var_20842_cast_fp16, var_21243_cast_fp16))[name = string("op_21334_cast_fp16")]; + string var_21336_equation_0 = const()[name = string("op_21336_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21336_cast_fp16 = einsum(equation = var_21336_equation_0, values = (var_20842_cast_fp16, var_21244_cast_fp16))[name = string("op_21336_cast_fp16")]; + string var_21338_equation_0 = const()[name = string("op_21338_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21338_cast_fp16 = einsum(equation = var_21338_equation_0, values = (var_20846_cast_fp16, var_21245_cast_fp16))[name = string("op_21338_cast_fp16")]; + string var_21340_equation_0 = const()[name = string("op_21340_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21340_cast_fp16 = einsum(equation = var_21340_equation_0, values = (var_20846_cast_fp16, var_21246_cast_fp16))[name = string("op_21340_cast_fp16")]; + string var_21342_equation_0 = const()[name = string("op_21342_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21342_cast_fp16 = einsum(equation = var_21342_equation_0, values = (var_20846_cast_fp16, var_21247_cast_fp16))[name = string("op_21342_cast_fp16")]; + string var_21344_equation_0 = const()[name = string("op_21344_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21344_cast_fp16 = einsum(equation = var_21344_equation_0, values = (var_20846_cast_fp16, var_21248_cast_fp16))[name = string("op_21344_cast_fp16")]; + string var_21346_equation_0 = const()[name = string("op_21346_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21346_cast_fp16 = einsum(equation = var_21346_equation_0, values = (var_20850_cast_fp16, var_21249_cast_fp16))[name = string("op_21346_cast_fp16")]; + string var_21348_equation_0 = const()[name = string("op_21348_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21348_cast_fp16 = einsum(equation = var_21348_equation_0, values = (var_20850_cast_fp16, var_21250_cast_fp16))[name = string("op_21348_cast_fp16")]; + string var_21350_equation_0 = const()[name = string("op_21350_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21350_cast_fp16 = einsum(equation = var_21350_equation_0, values = (var_20850_cast_fp16, var_21251_cast_fp16))[name = string("op_21350_cast_fp16")]; + string var_21352_equation_0 = const()[name = string("op_21352_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21352_cast_fp16 = einsum(equation = var_21352_equation_0, values = (var_20850_cast_fp16, var_21252_cast_fp16))[name = string("op_21352_cast_fp16")]; + string var_21354_equation_0 = const()[name = string("op_21354_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21354_cast_fp16 = einsum(equation = var_21354_equation_0, values = (var_20854_cast_fp16, var_21253_cast_fp16))[name = string("op_21354_cast_fp16")]; + string var_21356_equation_0 = const()[name = string("op_21356_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21356_cast_fp16 = einsum(equation = var_21356_equation_0, values = (var_20854_cast_fp16, var_21254_cast_fp16))[name = string("op_21356_cast_fp16")]; + string var_21358_equation_0 = const()[name = string("op_21358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21358_cast_fp16 = einsum(equation = var_21358_equation_0, values = (var_20854_cast_fp16, var_21255_cast_fp16))[name = string("op_21358_cast_fp16")]; + string var_21360_equation_0 = const()[name = string("op_21360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21360_cast_fp16 = einsum(equation = var_21360_equation_0, values = (var_20854_cast_fp16, var_21256_cast_fp16))[name = string("op_21360_cast_fp16")]; + string var_21362_equation_0 = const()[name = string("op_21362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21362_cast_fp16 = einsum(equation = var_21362_equation_0, values = (var_20858_cast_fp16, var_21257_cast_fp16))[name = string("op_21362_cast_fp16")]; + string var_21364_equation_0 = const()[name = string("op_21364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21364_cast_fp16 = einsum(equation = var_21364_equation_0, values = (var_20858_cast_fp16, var_21258_cast_fp16))[name = string("op_21364_cast_fp16")]; + string var_21366_equation_0 = const()[name = string("op_21366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21366_cast_fp16 = einsum(equation = var_21366_equation_0, values = (var_20858_cast_fp16, var_21259_cast_fp16))[name = string("op_21366_cast_fp16")]; + string var_21368_equation_0 = const()[name = string("op_21368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21368_cast_fp16 = einsum(equation = var_21368_equation_0, values = (var_20858_cast_fp16, var_21260_cast_fp16))[name = string("op_21368_cast_fp16")]; + string var_21370_equation_0 = const()[name = string("op_21370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21370_cast_fp16 = einsum(equation = var_21370_equation_0, values = (var_20862_cast_fp16, var_21261_cast_fp16))[name = string("op_21370_cast_fp16")]; + string var_21372_equation_0 = const()[name = string("op_21372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21372_cast_fp16 = einsum(equation = var_21372_equation_0, values = (var_20862_cast_fp16, var_21262_cast_fp16))[name = string("op_21372_cast_fp16")]; + string var_21374_equation_0 = const()[name = string("op_21374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21374_cast_fp16 = einsum(equation = var_21374_equation_0, values = (var_20862_cast_fp16, var_21263_cast_fp16))[name = string("op_21374_cast_fp16")]; + string var_21376_equation_0 = const()[name = string("op_21376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21376_cast_fp16 = einsum(equation = var_21376_equation_0, values = (var_20862_cast_fp16, var_21264_cast_fp16))[name = string("op_21376_cast_fp16")]; + string var_21378_equation_0 = const()[name = string("op_21378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21378_cast_fp16 = einsum(equation = var_21378_equation_0, values = (var_20866_cast_fp16, var_21265_cast_fp16))[name = string("op_21378_cast_fp16")]; + string var_21380_equation_0 = const()[name = string("op_21380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21380_cast_fp16 = einsum(equation = var_21380_equation_0, values = (var_20866_cast_fp16, var_21266_cast_fp16))[name = string("op_21380_cast_fp16")]; + string var_21382_equation_0 = const()[name = string("op_21382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21382_cast_fp16 = einsum(equation = var_21382_equation_0, values = (var_20866_cast_fp16, var_21267_cast_fp16))[name = string("op_21382_cast_fp16")]; + string var_21384_equation_0 = const()[name = string("op_21384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21384_cast_fp16 = einsum(equation = var_21384_equation_0, values = (var_20866_cast_fp16, var_21268_cast_fp16))[name = string("op_21384_cast_fp16")]; + string var_21386_equation_0 = const()[name = string("op_21386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21386_cast_fp16 = einsum(equation = var_21386_equation_0, values = (var_20870_cast_fp16, var_21269_cast_fp16))[name = string("op_21386_cast_fp16")]; + string var_21388_equation_0 = const()[name = string("op_21388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21388_cast_fp16 = einsum(equation = var_21388_equation_0, values = (var_20870_cast_fp16, var_21270_cast_fp16))[name = string("op_21388_cast_fp16")]; + string var_21390_equation_0 = const()[name = string("op_21390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21390_cast_fp16 = einsum(equation = var_21390_equation_0, values = (var_20870_cast_fp16, var_21271_cast_fp16))[name = string("op_21390_cast_fp16")]; + string var_21392_equation_0 = const()[name = string("op_21392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21392_cast_fp16 = einsum(equation = var_21392_equation_0, values = (var_20870_cast_fp16, var_21272_cast_fp16))[name = string("op_21392_cast_fp16")]; + string var_21394_equation_0 = const()[name = string("op_21394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21394_cast_fp16 = einsum(equation = var_21394_equation_0, values = (var_20874_cast_fp16, var_21273_cast_fp16))[name = string("op_21394_cast_fp16")]; + string var_21396_equation_0 = const()[name = string("op_21396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21396_cast_fp16 = einsum(equation = var_21396_equation_0, values = (var_20874_cast_fp16, var_21274_cast_fp16))[name = string("op_21396_cast_fp16")]; + string var_21398_equation_0 = const()[name = string("op_21398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21398_cast_fp16 = einsum(equation = var_21398_equation_0, values = (var_20874_cast_fp16, var_21275_cast_fp16))[name = string("op_21398_cast_fp16")]; + string var_21400_equation_0 = const()[name = string("op_21400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21400_cast_fp16 = einsum(equation = var_21400_equation_0, values = (var_20874_cast_fp16, var_21276_cast_fp16))[name = string("op_21400_cast_fp16")]; + string var_21402_equation_0 = const()[name = string("op_21402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21402_cast_fp16 = einsum(equation = var_21402_equation_0, values = (var_20878_cast_fp16, var_21277_cast_fp16))[name = string("op_21402_cast_fp16")]; + string var_21404_equation_0 = const()[name = string("op_21404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21404_cast_fp16 = einsum(equation = var_21404_equation_0, values = (var_20878_cast_fp16, var_21278_cast_fp16))[name = string("op_21404_cast_fp16")]; + string var_21406_equation_0 = const()[name = string("op_21406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21406_cast_fp16 = einsum(equation = var_21406_equation_0, values = (var_20878_cast_fp16, var_21279_cast_fp16))[name = string("op_21406_cast_fp16")]; + string var_21408_equation_0 = const()[name = string("op_21408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21408_cast_fp16 = einsum(equation = var_21408_equation_0, values = (var_20878_cast_fp16, var_21280_cast_fp16))[name = string("op_21408_cast_fp16")]; + string var_21410_equation_0 = const()[name = string("op_21410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21410_cast_fp16 = einsum(equation = var_21410_equation_0, values = (var_20882_cast_fp16, var_21281_cast_fp16))[name = string("op_21410_cast_fp16")]; + string var_21412_equation_0 = const()[name = string("op_21412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21412_cast_fp16 = einsum(equation = var_21412_equation_0, values = (var_20882_cast_fp16, var_21282_cast_fp16))[name = string("op_21412_cast_fp16")]; + string var_21414_equation_0 = const()[name = string("op_21414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21414_cast_fp16 = einsum(equation = var_21414_equation_0, values = (var_20882_cast_fp16, var_21283_cast_fp16))[name = string("op_21414_cast_fp16")]; + string var_21416_equation_0 = const()[name = string("op_21416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21416_cast_fp16 = einsum(equation = var_21416_equation_0, values = (var_20882_cast_fp16, var_21284_cast_fp16))[name = string("op_21416_cast_fp16")]; + string var_21418_equation_0 = const()[name = string("op_21418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21418_cast_fp16 = einsum(equation = var_21418_equation_0, values = (var_20886_cast_fp16, var_21285_cast_fp16))[name = string("op_21418_cast_fp16")]; + string var_21420_equation_0 = const()[name = string("op_21420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21420_cast_fp16 = einsum(equation = var_21420_equation_0, values = (var_20886_cast_fp16, var_21286_cast_fp16))[name = string("op_21420_cast_fp16")]; + string var_21422_equation_0 = const()[name = string("op_21422_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21422_cast_fp16 = einsum(equation = var_21422_equation_0, values = (var_20886_cast_fp16, var_21287_cast_fp16))[name = string("op_21422_cast_fp16")]; + string var_21424_equation_0 = const()[name = string("op_21424_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21424_cast_fp16 = einsum(equation = var_21424_equation_0, values = (var_20886_cast_fp16, var_21288_cast_fp16))[name = string("op_21424_cast_fp16")]; + string var_21426_equation_0 = const()[name = string("op_21426_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21426_cast_fp16 = einsum(equation = var_21426_equation_0, values = (var_20890_cast_fp16, var_21289_cast_fp16))[name = string("op_21426_cast_fp16")]; + string var_21428_equation_0 = const()[name = string("op_21428_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21428_cast_fp16 = einsum(equation = var_21428_equation_0, values = (var_20890_cast_fp16, var_21290_cast_fp16))[name = string("op_21428_cast_fp16")]; + string var_21430_equation_0 = const()[name = string("op_21430_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21430_cast_fp16 = einsum(equation = var_21430_equation_0, values = (var_20890_cast_fp16, var_21291_cast_fp16))[name = string("op_21430_cast_fp16")]; + string var_21432_equation_0 = const()[name = string("op_21432_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21432_cast_fp16 = einsum(equation = var_21432_equation_0, values = (var_20890_cast_fp16, var_21292_cast_fp16))[name = string("op_21432_cast_fp16")]; + string var_21434_equation_0 = const()[name = string("op_21434_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21434_cast_fp16 = einsum(equation = var_21434_equation_0, values = (var_20894_cast_fp16, var_21293_cast_fp16))[name = string("op_21434_cast_fp16")]; + string var_21436_equation_0 = const()[name = string("op_21436_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21436_cast_fp16 = einsum(equation = var_21436_equation_0, values = (var_20894_cast_fp16, var_21294_cast_fp16))[name = string("op_21436_cast_fp16")]; + string var_21438_equation_0 = const()[name = string("op_21438_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21438_cast_fp16 = einsum(equation = var_21438_equation_0, values = (var_20894_cast_fp16, var_21295_cast_fp16))[name = string("op_21438_cast_fp16")]; + string var_21440_equation_0 = const()[name = string("op_21440_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21440_cast_fp16 = einsum(equation = var_21440_equation_0, values = (var_20894_cast_fp16, var_21296_cast_fp16))[name = string("op_21440_cast_fp16")]; + string var_21442_equation_0 = const()[name = string("op_21442_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21442_cast_fp16 = einsum(equation = var_21442_equation_0, values = (var_20898_cast_fp16, var_21297_cast_fp16))[name = string("op_21442_cast_fp16")]; + string var_21444_equation_0 = const()[name = string("op_21444_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21444_cast_fp16 = einsum(equation = var_21444_equation_0, values = (var_20898_cast_fp16, var_21298_cast_fp16))[name = string("op_21444_cast_fp16")]; + string var_21446_equation_0 = const()[name = string("op_21446_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21446_cast_fp16 = einsum(equation = var_21446_equation_0, values = (var_20898_cast_fp16, var_21299_cast_fp16))[name = string("op_21446_cast_fp16")]; + string var_21448_equation_0 = const()[name = string("op_21448_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21448_cast_fp16 = einsum(equation = var_21448_equation_0, values = (var_20898_cast_fp16, var_21300_cast_fp16))[name = string("op_21448_cast_fp16")]; + string var_21450_equation_0 = const()[name = string("op_21450_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21450_cast_fp16 = einsum(equation = var_21450_equation_0, values = (var_20902_cast_fp16, var_21301_cast_fp16))[name = string("op_21450_cast_fp16")]; + string var_21452_equation_0 = const()[name = string("op_21452_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21452_cast_fp16 = einsum(equation = var_21452_equation_0, values = (var_20902_cast_fp16, var_21302_cast_fp16))[name = string("op_21452_cast_fp16")]; + string var_21454_equation_0 = const()[name = string("op_21454_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21454_cast_fp16 = einsum(equation = var_21454_equation_0, values = (var_20902_cast_fp16, var_21303_cast_fp16))[name = string("op_21454_cast_fp16")]; + string var_21456_equation_0 = const()[name = string("op_21456_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21456_cast_fp16 = einsum(equation = var_21456_equation_0, values = (var_20902_cast_fp16, var_21304_cast_fp16))[name = string("op_21456_cast_fp16")]; + string var_21458_equation_0 = const()[name = string("op_21458_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21458_cast_fp16 = einsum(equation = var_21458_equation_0, values = (var_20906_cast_fp16, var_21305_cast_fp16))[name = string("op_21458_cast_fp16")]; + string var_21460_equation_0 = const()[name = string("op_21460_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21460_cast_fp16 = einsum(equation = var_21460_equation_0, values = (var_20906_cast_fp16, var_21306_cast_fp16))[name = string("op_21460_cast_fp16")]; + string var_21462_equation_0 = const()[name = string("op_21462_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21462_cast_fp16 = einsum(equation = var_21462_equation_0, values = (var_20906_cast_fp16, var_21307_cast_fp16))[name = string("op_21462_cast_fp16")]; + string var_21464_equation_0 = const()[name = string("op_21464_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21464_cast_fp16 = einsum(equation = var_21464_equation_0, values = (var_20906_cast_fp16, var_21308_cast_fp16))[name = string("op_21464_cast_fp16")]; + string var_21466_equation_0 = const()[name = string("op_21466_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21466_cast_fp16 = einsum(equation = var_21466_equation_0, values = (var_20910_cast_fp16, var_21309_cast_fp16))[name = string("op_21466_cast_fp16")]; + string var_21468_equation_0 = const()[name = string("op_21468_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21468_cast_fp16 = einsum(equation = var_21468_equation_0, values = (var_20910_cast_fp16, var_21310_cast_fp16))[name = string("op_21468_cast_fp16")]; + string var_21470_equation_0 = const()[name = string("op_21470_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21470_cast_fp16 = einsum(equation = var_21470_equation_0, values = (var_20910_cast_fp16, var_21311_cast_fp16))[name = string("op_21470_cast_fp16")]; + string var_21472_equation_0 = const()[name = string("op_21472_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_21472_cast_fp16 = einsum(equation = var_21472_equation_0, values = (var_20910_cast_fp16, var_21312_cast_fp16))[name = string("op_21472_cast_fp16")]; + bool var_21474_interleave_0 = const()[name = string("op_21474_interleave_0"), val = bool(false)]; + tensor var_21474_cast_fp16 = concat(axis = var_20033, interleave = var_21474_interleave_0, values = (var_21314_cast_fp16, var_21316_cast_fp16, var_21318_cast_fp16, var_21320_cast_fp16))[name = string("op_21474_cast_fp16")]; + bool var_21476_interleave_0 = const()[name = string("op_21476_interleave_0"), val = bool(false)]; + tensor var_21476_cast_fp16 = concat(axis = var_20033, interleave = var_21476_interleave_0, values = (var_21322_cast_fp16, var_21324_cast_fp16, var_21326_cast_fp16, var_21328_cast_fp16))[name = string("op_21476_cast_fp16")]; + bool var_21478_interleave_0 = const()[name = string("op_21478_interleave_0"), val = bool(false)]; + tensor var_21478_cast_fp16 = concat(axis = var_20033, interleave = var_21478_interleave_0, values = (var_21330_cast_fp16, var_21332_cast_fp16, var_21334_cast_fp16, var_21336_cast_fp16))[name = string("op_21478_cast_fp16")]; + bool var_21480_interleave_0 = const()[name = string("op_21480_interleave_0"), val = bool(false)]; + tensor var_21480_cast_fp16 = concat(axis = var_20033, interleave = var_21480_interleave_0, values = (var_21338_cast_fp16, var_21340_cast_fp16, var_21342_cast_fp16, var_21344_cast_fp16))[name = string("op_21480_cast_fp16")]; + bool var_21482_interleave_0 = const()[name = string("op_21482_interleave_0"), val = bool(false)]; + tensor var_21482_cast_fp16 = concat(axis = var_20033, interleave = var_21482_interleave_0, values = (var_21346_cast_fp16, var_21348_cast_fp16, var_21350_cast_fp16, var_21352_cast_fp16))[name = string("op_21482_cast_fp16")]; + bool var_21484_interleave_0 = const()[name = string("op_21484_interleave_0"), val = bool(false)]; + tensor var_21484_cast_fp16 = concat(axis = var_20033, interleave = var_21484_interleave_0, values = (var_21354_cast_fp16, var_21356_cast_fp16, var_21358_cast_fp16, var_21360_cast_fp16))[name = string("op_21484_cast_fp16")]; + bool var_21486_interleave_0 = const()[name = string("op_21486_interleave_0"), val = bool(false)]; + tensor var_21486_cast_fp16 = concat(axis = var_20033, interleave = var_21486_interleave_0, values = (var_21362_cast_fp16, var_21364_cast_fp16, var_21366_cast_fp16, var_21368_cast_fp16))[name = string("op_21486_cast_fp16")]; + bool var_21488_interleave_0 = const()[name = string("op_21488_interleave_0"), val = bool(false)]; + tensor var_21488_cast_fp16 = concat(axis = var_20033, interleave = var_21488_interleave_0, values = (var_21370_cast_fp16, var_21372_cast_fp16, var_21374_cast_fp16, var_21376_cast_fp16))[name = string("op_21488_cast_fp16")]; + bool var_21490_interleave_0 = const()[name = string("op_21490_interleave_0"), val = bool(false)]; + tensor var_21490_cast_fp16 = concat(axis = var_20033, interleave = var_21490_interleave_0, values = (var_21378_cast_fp16, var_21380_cast_fp16, var_21382_cast_fp16, var_21384_cast_fp16))[name = string("op_21490_cast_fp16")]; + bool var_21492_interleave_0 = const()[name = string("op_21492_interleave_0"), val = bool(false)]; + tensor var_21492_cast_fp16 = concat(axis = var_20033, interleave = var_21492_interleave_0, values = (var_21386_cast_fp16, var_21388_cast_fp16, var_21390_cast_fp16, var_21392_cast_fp16))[name = string("op_21492_cast_fp16")]; + bool var_21494_interleave_0 = const()[name = string("op_21494_interleave_0"), val = bool(false)]; + tensor var_21494_cast_fp16 = concat(axis = var_20033, interleave = var_21494_interleave_0, values = (var_21394_cast_fp16, var_21396_cast_fp16, var_21398_cast_fp16, var_21400_cast_fp16))[name = string("op_21494_cast_fp16")]; + bool var_21496_interleave_0 = const()[name = string("op_21496_interleave_0"), val = bool(false)]; + tensor var_21496_cast_fp16 = concat(axis = var_20033, interleave = var_21496_interleave_0, values = (var_21402_cast_fp16, var_21404_cast_fp16, var_21406_cast_fp16, var_21408_cast_fp16))[name = string("op_21496_cast_fp16")]; + bool var_21498_interleave_0 = const()[name = string("op_21498_interleave_0"), val = bool(false)]; + tensor var_21498_cast_fp16 = concat(axis = var_20033, interleave = var_21498_interleave_0, values = (var_21410_cast_fp16, var_21412_cast_fp16, var_21414_cast_fp16, var_21416_cast_fp16))[name = string("op_21498_cast_fp16")]; + bool var_21500_interleave_0 = const()[name = string("op_21500_interleave_0"), val = bool(false)]; + tensor var_21500_cast_fp16 = concat(axis = var_20033, interleave = var_21500_interleave_0, values = (var_21418_cast_fp16, var_21420_cast_fp16, var_21422_cast_fp16, var_21424_cast_fp16))[name = string("op_21500_cast_fp16")]; + bool var_21502_interleave_0 = const()[name = string("op_21502_interleave_0"), val = bool(false)]; + tensor var_21502_cast_fp16 = concat(axis = var_20033, interleave = var_21502_interleave_0, values = (var_21426_cast_fp16, var_21428_cast_fp16, var_21430_cast_fp16, var_21432_cast_fp16))[name = string("op_21502_cast_fp16")]; + bool var_21504_interleave_0 = const()[name = string("op_21504_interleave_0"), val = bool(false)]; + tensor var_21504_cast_fp16 = concat(axis = var_20033, interleave = var_21504_interleave_0, values = (var_21434_cast_fp16, var_21436_cast_fp16, var_21438_cast_fp16, var_21440_cast_fp16))[name = string("op_21504_cast_fp16")]; + bool var_21506_interleave_0 = const()[name = string("op_21506_interleave_0"), val = bool(false)]; + tensor var_21506_cast_fp16 = concat(axis = var_20033, interleave = var_21506_interleave_0, values = (var_21442_cast_fp16, var_21444_cast_fp16, var_21446_cast_fp16, var_21448_cast_fp16))[name = string("op_21506_cast_fp16")]; + bool var_21508_interleave_0 = const()[name = string("op_21508_interleave_0"), val = bool(false)]; + tensor var_21508_cast_fp16 = concat(axis = var_20033, interleave = var_21508_interleave_0, values = (var_21450_cast_fp16, var_21452_cast_fp16, var_21454_cast_fp16, var_21456_cast_fp16))[name = string("op_21508_cast_fp16")]; + bool var_21510_interleave_0 = const()[name = string("op_21510_interleave_0"), val = bool(false)]; + tensor var_21510_cast_fp16 = concat(axis = var_20033, interleave = var_21510_interleave_0, values = (var_21458_cast_fp16, var_21460_cast_fp16, var_21462_cast_fp16, var_21464_cast_fp16))[name = string("op_21510_cast_fp16")]; + bool var_21512_interleave_0 = const()[name = string("op_21512_interleave_0"), val = bool(false)]; + tensor var_21512_cast_fp16 = concat(axis = var_20033, interleave = var_21512_interleave_0, values = (var_21466_cast_fp16, var_21468_cast_fp16, var_21470_cast_fp16, var_21472_cast_fp16))[name = string("op_21512_cast_fp16")]; + bool input_105_interleave_0 = const()[name = string("input_105_interleave_0"), val = bool(false)]; + tensor input_105_cast_fp16 = concat(axis = var_20058, interleave = input_105_interleave_0, values = (var_21474_cast_fp16, var_21476_cast_fp16, var_21478_cast_fp16, var_21480_cast_fp16, var_21482_cast_fp16, var_21484_cast_fp16, var_21486_cast_fp16, var_21488_cast_fp16, var_21490_cast_fp16, var_21492_cast_fp16, var_21494_cast_fp16, var_21496_cast_fp16, var_21498_cast_fp16, var_21500_cast_fp16, var_21502_cast_fp16, var_21504_cast_fp16, var_21506_cast_fp16, var_21508_cast_fp16, var_21510_cast_fp16, var_21512_cast_fp16))[name = string("input_105_cast_fp16")]; + string obj_55_pad_type_0 = const()[name = string("obj_55_pad_type_0"), val = string("valid")]; + tensor obj_55_strides_0 = const()[name = string("obj_55_strides_0"), val = tensor([1, 1])]; + tensor obj_55_pad_0 = const()[name = string("obj_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_55_dilations_0 = const()[name = string("obj_55_dilations_0"), val = tensor([1, 1])]; + int32 obj_55_groups_0 = const()[name = string("obj_55_groups_0"), val = int32(1)]; + tensor layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(536097920)))]; + tensor layers_13_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539374784)))]; + tensor obj_55_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_bias_to_fp16, dilations = obj_55_dilations_0, groups = obj_55_groups_0, pad = obj_55_pad_0, pad_type = obj_55_pad_type_0, strides = obj_55_strides_0, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("obj_55_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_55_cast_fp16)[name = string("inputs_55_cast_fp16")]; + tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; + fp16 var_21531_to_fp16 = const()[name = string("op_21531_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_21531_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; + tensor input_107_gamma_0_to_fp16 = const()[name = string("input_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539377408)))]; + tensor input_107_beta_0_to_fp16 = const()[name = string("input_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539380032)))]; + fp16 input_107_epsilon_0_to_fp16 = const()[name = string("input_107_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_107_cast_fp16 = batch_norm(beta = input_107_beta_0_to_fp16, epsilon = input_107_epsilon_0_to_fp16, gamma = input_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("input_107_cast_fp16")]; + string input_109_pad_type_0 = const()[name = string("input_109_pad_type_0"), val = string("valid")]; + tensor input_109_strides_0 = const()[name = string("input_109_strides_0"), val = tensor([1, 1])]; + tensor input_109_pad_0 = const()[name = string("input_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_109_dilations_0 = const()[name = string("input_109_dilations_0"), val = tensor([1, 1])]; + int32 input_109_groups_0 = const()[name = string("input_109_groups_0"), val = int32(1)]; + tensor layers_13_fc1_weight_to_fp16 = const()[name = string("layers_13_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539382656)))]; + tensor layers_13_fc1_bias_to_fp16 = const()[name = string("layers_13_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552489920)))]; + tensor input_109_cast_fp16 = conv(bias = layers_13_fc1_bias_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = layers_13_fc1_weight_to_fp16, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")]; + string input_111_mode_0 = const()[name = string("input_111_mode_0"), val = string("EXACT")]; + tensor input_111_cast_fp16 = gelu(mode = input_111_mode_0, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")]; + string hidden_states_31_pad_type_0 = const()[name = string("hidden_states_31_pad_type_0"), val = string("valid")]; + tensor hidden_states_31_strides_0 = const()[name = string("hidden_states_31_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_31_pad_0 = const()[name = string("hidden_states_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_31_dilations_0 = const()[name = string("hidden_states_31_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_31_groups_0 = const()[name = string("hidden_states_31_groups_0"), val = int32(1)]; + tensor layers_13_fc2_weight_to_fp16 = const()[name = string("layers_13_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(552500224)))]; + tensor layers_13_fc2_bias_to_fp16 = const()[name = string("layers_13_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565607488)))]; + tensor hidden_states_31_cast_fp16 = conv(bias = layers_13_fc2_bias_to_fp16, dilations = hidden_states_31_dilations_0, groups = hidden_states_31_groups_0, pad = hidden_states_31_pad_0, pad_type = hidden_states_31_pad_type_0, strides = hidden_states_31_strides_0, weight = layers_13_fc2_weight_to_fp16, x = input_111_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = hidden_states_31_cast_fp16)[name = string("inputs_57_cast_fp16")]; + int32 var_21560 = const()[name = string("op_21560"), val = int32(3)]; + int32 var_21585 = const()[name = string("op_21585"), val = int32(1)]; + tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; + fp16 var_21602_to_fp16 = const()[name = string("op_21602_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_21602_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; + tensor obj_57_gamma_0_to_fp16 = const()[name = string("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565610112)))]; + tensor obj_57_beta_0_to_fp16 = const()[name = string("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565612736)))]; + fp16 obj_57_epsilon_0_to_fp16 = const()[name = string("obj_57_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_57_cast_fp16")]; + string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; + tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; + tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; + int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(565615360)))]; + tensor layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568892224)))]; + tensor query_29_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("query_29_cast_fp16")]; + string key_29_pad_type_0 = const()[name = string("key_29_pad_type_0"), val = string("valid")]; + tensor key_29_strides_0 = const()[name = string("key_29_strides_0"), val = tensor([1, 1])]; + tensor key_29_pad_0 = const()[name = string("key_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_29_dilations_0 = const()[name = string("key_29_dilations_0"), val = tensor([1, 1])]; + int32 key_29_groups_0 = const()[name = string("key_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568894848)))]; + tensor key_29_cast_fp16 = conv(dilations = key_29_dilations_0, groups = key_29_groups_0, pad = key_29_pad_0, pad_type = key_29_pad_type_0, strides = key_29_strides_0, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("key_29_cast_fp16")]; + string value_29_pad_type_0 = const()[name = string("value_29_pad_type_0"), val = string("valid")]; + tensor value_29_strides_0 = const()[name = string("value_29_strides_0"), val = tensor([1, 1])]; + tensor value_29_pad_0 = const()[name = string("value_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_29_dilations_0 = const()[name = string("value_29_dilations_0"), val = tensor([1, 1])]; + int32 value_29_groups_0 = const()[name = string("value_29_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572171712)))]; + tensor layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575448576)))]; + tensor value_29_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_bias_to_fp16, dilations = value_29_dilations_0, groups = value_29_groups_0, pad = value_29_pad_0, pad_type = value_29_pad_type_0, strides = value_29_strides_0, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = string("value_29_cast_fp16")]; + tensor var_21640_begin_0 = const()[name = string("op_21640_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21640_end_0 = const()[name = string("op_21640_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21640_end_mask_0 = const()[name = string("op_21640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21640_cast_fp16 = slice_by_index(begin = var_21640_begin_0, end = var_21640_end_0, end_mask = var_21640_end_mask_0, x = query_29_cast_fp16)[name = string("op_21640_cast_fp16")]; + tensor var_21644_begin_0 = const()[name = string("op_21644_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21644_end_0 = const()[name = string("op_21644_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_21644_end_mask_0 = const()[name = string("op_21644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21644_cast_fp16 = slice_by_index(begin = var_21644_begin_0, end = var_21644_end_0, end_mask = var_21644_end_mask_0, x = query_29_cast_fp16)[name = string("op_21644_cast_fp16")]; + tensor var_21648_begin_0 = const()[name = string("op_21648_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21648_end_0 = const()[name = string("op_21648_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_21648_end_mask_0 = const()[name = string("op_21648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21648_cast_fp16 = slice_by_index(begin = var_21648_begin_0, end = var_21648_end_0, end_mask = var_21648_end_mask_0, x = query_29_cast_fp16)[name = string("op_21648_cast_fp16")]; + tensor var_21652_begin_0 = const()[name = string("op_21652_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21652_end_0 = const()[name = string("op_21652_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_21652_end_mask_0 = const()[name = string("op_21652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21652_cast_fp16 = slice_by_index(begin = var_21652_begin_0, end = var_21652_end_0, end_mask = var_21652_end_mask_0, x = query_29_cast_fp16)[name = string("op_21652_cast_fp16")]; + tensor var_21656_begin_0 = const()[name = string("op_21656_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21656_end_0 = const()[name = string("op_21656_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_21656_end_mask_0 = const()[name = string("op_21656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21656_cast_fp16 = slice_by_index(begin = var_21656_begin_0, end = var_21656_end_0, end_mask = var_21656_end_mask_0, x = query_29_cast_fp16)[name = string("op_21656_cast_fp16")]; + tensor var_21660_begin_0 = const()[name = string("op_21660_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21660_end_0 = const()[name = string("op_21660_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_21660_end_mask_0 = const()[name = string("op_21660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21660_cast_fp16 = slice_by_index(begin = var_21660_begin_0, end = var_21660_end_0, end_mask = var_21660_end_mask_0, x = query_29_cast_fp16)[name = string("op_21660_cast_fp16")]; + tensor var_21664_begin_0 = const()[name = string("op_21664_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21664_end_0 = const()[name = string("op_21664_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_21664_end_mask_0 = const()[name = string("op_21664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21664_cast_fp16 = slice_by_index(begin = var_21664_begin_0, end = var_21664_end_0, end_mask = var_21664_end_mask_0, x = query_29_cast_fp16)[name = string("op_21664_cast_fp16")]; + tensor var_21668_begin_0 = const()[name = string("op_21668_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21668_end_0 = const()[name = string("op_21668_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_21668_end_mask_0 = const()[name = string("op_21668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21668_cast_fp16 = slice_by_index(begin = var_21668_begin_0, end = var_21668_end_0, end_mask = var_21668_end_mask_0, x = query_29_cast_fp16)[name = string("op_21668_cast_fp16")]; + tensor var_21672_begin_0 = const()[name = string("op_21672_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21672_end_0 = const()[name = string("op_21672_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_21672_end_mask_0 = const()[name = string("op_21672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21672_cast_fp16 = slice_by_index(begin = var_21672_begin_0, end = var_21672_end_0, end_mask = var_21672_end_mask_0, x = query_29_cast_fp16)[name = string("op_21672_cast_fp16")]; + tensor var_21676_begin_0 = const()[name = string("op_21676_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21676_end_0 = const()[name = string("op_21676_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_21676_end_mask_0 = const()[name = string("op_21676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21676_cast_fp16 = slice_by_index(begin = var_21676_begin_0, end = var_21676_end_0, end_mask = var_21676_end_mask_0, x = query_29_cast_fp16)[name = string("op_21676_cast_fp16")]; + tensor var_21680_begin_0 = const()[name = string("op_21680_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21680_end_0 = const()[name = string("op_21680_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_21680_end_mask_0 = const()[name = string("op_21680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21680_cast_fp16 = slice_by_index(begin = var_21680_begin_0, end = var_21680_end_0, end_mask = var_21680_end_mask_0, x = query_29_cast_fp16)[name = string("op_21680_cast_fp16")]; + tensor var_21684_begin_0 = const()[name = string("op_21684_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21684_end_0 = const()[name = string("op_21684_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_21684_end_mask_0 = const()[name = string("op_21684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21684_cast_fp16 = slice_by_index(begin = var_21684_begin_0, end = var_21684_end_0, end_mask = var_21684_end_mask_0, x = query_29_cast_fp16)[name = string("op_21684_cast_fp16")]; + tensor var_21688_begin_0 = const()[name = string("op_21688_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21688_end_0 = const()[name = string("op_21688_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_21688_end_mask_0 = const()[name = string("op_21688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21688_cast_fp16 = slice_by_index(begin = var_21688_begin_0, end = var_21688_end_0, end_mask = var_21688_end_mask_0, x = query_29_cast_fp16)[name = string("op_21688_cast_fp16")]; + tensor var_21692_begin_0 = const()[name = string("op_21692_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21692_end_0 = const()[name = string("op_21692_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_21692_end_mask_0 = const()[name = string("op_21692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21692_cast_fp16 = slice_by_index(begin = var_21692_begin_0, end = var_21692_end_0, end_mask = var_21692_end_mask_0, x = query_29_cast_fp16)[name = string("op_21692_cast_fp16")]; + tensor var_21696_begin_0 = const()[name = string("op_21696_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21696_end_0 = const()[name = string("op_21696_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_21696_end_mask_0 = const()[name = string("op_21696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21696_cast_fp16 = slice_by_index(begin = var_21696_begin_0, end = var_21696_end_0, end_mask = var_21696_end_mask_0, x = query_29_cast_fp16)[name = string("op_21696_cast_fp16")]; + tensor var_21700_begin_0 = const()[name = string("op_21700_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21700_end_0 = const()[name = string("op_21700_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_21700_end_mask_0 = const()[name = string("op_21700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21700_cast_fp16 = slice_by_index(begin = var_21700_begin_0, end = var_21700_end_0, end_mask = var_21700_end_mask_0, x = query_29_cast_fp16)[name = string("op_21700_cast_fp16")]; + tensor var_21704_begin_0 = const()[name = string("op_21704_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21704_end_0 = const()[name = string("op_21704_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_21704_end_mask_0 = const()[name = string("op_21704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21704_cast_fp16 = slice_by_index(begin = var_21704_begin_0, end = var_21704_end_0, end_mask = var_21704_end_mask_0, x = query_29_cast_fp16)[name = string("op_21704_cast_fp16")]; + tensor var_21708_begin_0 = const()[name = string("op_21708_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21708_end_0 = const()[name = string("op_21708_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_21708_end_mask_0 = const()[name = string("op_21708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21708_cast_fp16 = slice_by_index(begin = var_21708_begin_0, end = var_21708_end_0, end_mask = var_21708_end_mask_0, x = query_29_cast_fp16)[name = string("op_21708_cast_fp16")]; + tensor var_21712_begin_0 = const()[name = string("op_21712_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21712_end_0 = const()[name = string("op_21712_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_21712_end_mask_0 = const()[name = string("op_21712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21712_cast_fp16 = slice_by_index(begin = var_21712_begin_0, end = var_21712_end_0, end_mask = var_21712_end_mask_0, x = query_29_cast_fp16)[name = string("op_21712_cast_fp16")]; + tensor var_21716_begin_0 = const()[name = string("op_21716_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21716_end_0 = const()[name = string("op_21716_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_21716_end_mask_0 = const()[name = string("op_21716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21716_cast_fp16 = slice_by_index(begin = var_21716_begin_0, end = var_21716_end_0, end_mask = var_21716_end_mask_0, x = query_29_cast_fp16)[name = string("op_21716_cast_fp16")]; + tensor var_21725_begin_0 = const()[name = string("op_21725_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21725_end_0 = const()[name = string("op_21725_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21725_end_mask_0 = const()[name = string("op_21725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21725_cast_fp16 = slice_by_index(begin = var_21725_begin_0, end = var_21725_end_0, end_mask = var_21725_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21725_cast_fp16")]; + tensor var_21732_begin_0 = const()[name = string("op_21732_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21732_end_0 = const()[name = string("op_21732_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21732_end_mask_0 = const()[name = string("op_21732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21732_cast_fp16 = slice_by_index(begin = var_21732_begin_0, end = var_21732_end_0, end_mask = var_21732_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21732_cast_fp16")]; + tensor var_21739_begin_0 = const()[name = string("op_21739_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21739_end_0 = const()[name = string("op_21739_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21739_end_mask_0 = const()[name = string("op_21739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21739_cast_fp16 = slice_by_index(begin = var_21739_begin_0, end = var_21739_end_0, end_mask = var_21739_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21739_cast_fp16")]; + tensor var_21746_begin_0 = const()[name = string("op_21746_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21746_end_0 = const()[name = string("op_21746_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21746_end_mask_0 = const()[name = string("op_21746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21746_cast_fp16 = slice_by_index(begin = var_21746_begin_0, end = var_21746_end_0, end_mask = var_21746_end_mask_0, x = var_21640_cast_fp16)[name = string("op_21746_cast_fp16")]; + tensor var_21753_begin_0 = const()[name = string("op_21753_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21753_end_0 = const()[name = string("op_21753_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21753_end_mask_0 = const()[name = string("op_21753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21753_cast_fp16 = slice_by_index(begin = var_21753_begin_0, end = var_21753_end_0, end_mask = var_21753_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21753_cast_fp16")]; + tensor var_21760_begin_0 = const()[name = string("op_21760_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21760_end_0 = const()[name = string("op_21760_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21760_end_mask_0 = const()[name = string("op_21760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21760_cast_fp16 = slice_by_index(begin = var_21760_begin_0, end = var_21760_end_0, end_mask = var_21760_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21760_cast_fp16")]; + tensor var_21767_begin_0 = const()[name = string("op_21767_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21767_end_0 = const()[name = string("op_21767_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21767_end_mask_0 = const()[name = string("op_21767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21767_cast_fp16 = slice_by_index(begin = var_21767_begin_0, end = var_21767_end_0, end_mask = var_21767_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21767_cast_fp16")]; + tensor var_21774_begin_0 = const()[name = string("op_21774_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21774_end_0 = const()[name = string("op_21774_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21774_end_mask_0 = const()[name = string("op_21774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21774_cast_fp16 = slice_by_index(begin = var_21774_begin_0, end = var_21774_end_0, end_mask = var_21774_end_mask_0, x = var_21644_cast_fp16)[name = string("op_21774_cast_fp16")]; + tensor var_21781_begin_0 = const()[name = string("op_21781_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21781_end_0 = const()[name = string("op_21781_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21781_end_mask_0 = const()[name = string("op_21781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21781_cast_fp16 = slice_by_index(begin = var_21781_begin_0, end = var_21781_end_0, end_mask = var_21781_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21781_cast_fp16")]; + tensor var_21788_begin_0 = const()[name = string("op_21788_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21788_end_0 = const()[name = string("op_21788_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21788_end_mask_0 = const()[name = string("op_21788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21788_cast_fp16 = slice_by_index(begin = var_21788_begin_0, end = var_21788_end_0, end_mask = var_21788_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21788_cast_fp16")]; + tensor var_21795_begin_0 = const()[name = string("op_21795_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21795_end_0 = const()[name = string("op_21795_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21795_end_mask_0 = const()[name = string("op_21795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21795_cast_fp16 = slice_by_index(begin = var_21795_begin_0, end = var_21795_end_0, end_mask = var_21795_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21795_cast_fp16")]; + tensor var_21802_begin_0 = const()[name = string("op_21802_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21802_end_0 = const()[name = string("op_21802_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21802_end_mask_0 = const()[name = string("op_21802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21802_cast_fp16 = slice_by_index(begin = var_21802_begin_0, end = var_21802_end_0, end_mask = var_21802_end_mask_0, x = var_21648_cast_fp16)[name = string("op_21802_cast_fp16")]; + tensor var_21809_begin_0 = const()[name = string("op_21809_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21809_end_0 = const()[name = string("op_21809_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21809_end_mask_0 = const()[name = string("op_21809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21809_cast_fp16 = slice_by_index(begin = var_21809_begin_0, end = var_21809_end_0, end_mask = var_21809_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21809_cast_fp16")]; + tensor var_21816_begin_0 = const()[name = string("op_21816_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21816_end_0 = const()[name = string("op_21816_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21816_end_mask_0 = const()[name = string("op_21816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21816_cast_fp16 = slice_by_index(begin = var_21816_begin_0, end = var_21816_end_0, end_mask = var_21816_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21816_cast_fp16")]; + tensor var_21823_begin_0 = const()[name = string("op_21823_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21823_end_0 = const()[name = string("op_21823_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21823_end_mask_0 = const()[name = string("op_21823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21823_cast_fp16 = slice_by_index(begin = var_21823_begin_0, end = var_21823_end_0, end_mask = var_21823_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21823_cast_fp16")]; + tensor var_21830_begin_0 = const()[name = string("op_21830_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21830_end_0 = const()[name = string("op_21830_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21830_end_mask_0 = const()[name = string("op_21830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21830_cast_fp16 = slice_by_index(begin = var_21830_begin_0, end = var_21830_end_0, end_mask = var_21830_end_mask_0, x = var_21652_cast_fp16)[name = string("op_21830_cast_fp16")]; + tensor var_21837_begin_0 = const()[name = string("op_21837_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21837_end_0 = const()[name = string("op_21837_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21837_end_mask_0 = const()[name = string("op_21837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21837_cast_fp16 = slice_by_index(begin = var_21837_begin_0, end = var_21837_end_0, end_mask = var_21837_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21837_cast_fp16")]; + tensor var_21844_begin_0 = const()[name = string("op_21844_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21844_end_0 = const()[name = string("op_21844_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21844_end_mask_0 = const()[name = string("op_21844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21844_cast_fp16 = slice_by_index(begin = var_21844_begin_0, end = var_21844_end_0, end_mask = var_21844_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21844_cast_fp16")]; + tensor var_21851_begin_0 = const()[name = string("op_21851_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21851_end_0 = const()[name = string("op_21851_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21851_end_mask_0 = const()[name = string("op_21851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21851_cast_fp16 = slice_by_index(begin = var_21851_begin_0, end = var_21851_end_0, end_mask = var_21851_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21851_cast_fp16")]; + tensor var_21858_begin_0 = const()[name = string("op_21858_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21858_end_0 = const()[name = string("op_21858_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21858_end_mask_0 = const()[name = string("op_21858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21858_cast_fp16 = slice_by_index(begin = var_21858_begin_0, end = var_21858_end_0, end_mask = var_21858_end_mask_0, x = var_21656_cast_fp16)[name = string("op_21858_cast_fp16")]; + tensor var_21865_begin_0 = const()[name = string("op_21865_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21865_end_0 = const()[name = string("op_21865_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21865_end_mask_0 = const()[name = string("op_21865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21865_cast_fp16 = slice_by_index(begin = var_21865_begin_0, end = var_21865_end_0, end_mask = var_21865_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21865_cast_fp16")]; + tensor var_21872_begin_0 = const()[name = string("op_21872_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21872_end_0 = const()[name = string("op_21872_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21872_end_mask_0 = const()[name = string("op_21872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21872_cast_fp16 = slice_by_index(begin = var_21872_begin_0, end = var_21872_end_0, end_mask = var_21872_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21872_cast_fp16")]; + tensor var_21879_begin_0 = const()[name = string("op_21879_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21879_end_0 = const()[name = string("op_21879_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21879_end_mask_0 = const()[name = string("op_21879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21879_cast_fp16 = slice_by_index(begin = var_21879_begin_0, end = var_21879_end_0, end_mask = var_21879_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21879_cast_fp16")]; + tensor var_21886_begin_0 = const()[name = string("op_21886_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21886_end_0 = const()[name = string("op_21886_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21886_end_mask_0 = const()[name = string("op_21886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21886_cast_fp16 = slice_by_index(begin = var_21886_begin_0, end = var_21886_end_0, end_mask = var_21886_end_mask_0, x = var_21660_cast_fp16)[name = string("op_21886_cast_fp16")]; + tensor var_21893_begin_0 = const()[name = string("op_21893_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21893_end_0 = const()[name = string("op_21893_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21893_end_mask_0 = const()[name = string("op_21893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21893_cast_fp16 = slice_by_index(begin = var_21893_begin_0, end = var_21893_end_0, end_mask = var_21893_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21893_cast_fp16")]; + tensor var_21900_begin_0 = const()[name = string("op_21900_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21900_end_0 = const()[name = string("op_21900_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21900_end_mask_0 = const()[name = string("op_21900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21900_cast_fp16 = slice_by_index(begin = var_21900_begin_0, end = var_21900_end_0, end_mask = var_21900_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21900_cast_fp16")]; + tensor var_21907_begin_0 = const()[name = string("op_21907_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21907_end_0 = const()[name = string("op_21907_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21907_end_mask_0 = const()[name = string("op_21907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21907_cast_fp16 = slice_by_index(begin = var_21907_begin_0, end = var_21907_end_0, end_mask = var_21907_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21907_cast_fp16")]; + tensor var_21914_begin_0 = const()[name = string("op_21914_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21914_end_0 = const()[name = string("op_21914_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21914_end_mask_0 = const()[name = string("op_21914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21914_cast_fp16 = slice_by_index(begin = var_21914_begin_0, end = var_21914_end_0, end_mask = var_21914_end_mask_0, x = var_21664_cast_fp16)[name = string("op_21914_cast_fp16")]; + tensor var_21921_begin_0 = const()[name = string("op_21921_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21921_end_0 = const()[name = string("op_21921_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21921_end_mask_0 = const()[name = string("op_21921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21921_cast_fp16 = slice_by_index(begin = var_21921_begin_0, end = var_21921_end_0, end_mask = var_21921_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21921_cast_fp16")]; + tensor var_21928_begin_0 = const()[name = string("op_21928_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21928_end_0 = const()[name = string("op_21928_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21928_end_mask_0 = const()[name = string("op_21928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21928_cast_fp16 = slice_by_index(begin = var_21928_begin_0, end = var_21928_end_0, end_mask = var_21928_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21928_cast_fp16")]; + tensor var_21935_begin_0 = const()[name = string("op_21935_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21935_end_0 = const()[name = string("op_21935_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21935_end_mask_0 = const()[name = string("op_21935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21935_cast_fp16 = slice_by_index(begin = var_21935_begin_0, end = var_21935_end_0, end_mask = var_21935_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21935_cast_fp16")]; + tensor var_21942_begin_0 = const()[name = string("op_21942_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21942_end_0 = const()[name = string("op_21942_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21942_end_mask_0 = const()[name = string("op_21942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21942_cast_fp16 = slice_by_index(begin = var_21942_begin_0, end = var_21942_end_0, end_mask = var_21942_end_mask_0, x = var_21668_cast_fp16)[name = string("op_21942_cast_fp16")]; + tensor var_21949_begin_0 = const()[name = string("op_21949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21949_end_0 = const()[name = string("op_21949_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21949_end_mask_0 = const()[name = string("op_21949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21949_cast_fp16 = slice_by_index(begin = var_21949_begin_0, end = var_21949_end_0, end_mask = var_21949_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21949_cast_fp16")]; + tensor var_21956_begin_0 = const()[name = string("op_21956_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21956_end_0 = const()[name = string("op_21956_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21956_end_mask_0 = const()[name = string("op_21956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21956_cast_fp16 = slice_by_index(begin = var_21956_begin_0, end = var_21956_end_0, end_mask = var_21956_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21956_cast_fp16")]; + tensor var_21963_begin_0 = const()[name = string("op_21963_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21963_end_0 = const()[name = string("op_21963_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21963_end_mask_0 = const()[name = string("op_21963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21963_cast_fp16 = slice_by_index(begin = var_21963_begin_0, end = var_21963_end_0, end_mask = var_21963_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21963_cast_fp16")]; + tensor var_21970_begin_0 = const()[name = string("op_21970_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21970_end_0 = const()[name = string("op_21970_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21970_end_mask_0 = const()[name = string("op_21970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21970_cast_fp16 = slice_by_index(begin = var_21970_begin_0, end = var_21970_end_0, end_mask = var_21970_end_mask_0, x = var_21672_cast_fp16)[name = string("op_21970_cast_fp16")]; + tensor var_21977_begin_0 = const()[name = string("op_21977_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21977_end_0 = const()[name = string("op_21977_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21977_end_mask_0 = const()[name = string("op_21977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21977_cast_fp16 = slice_by_index(begin = var_21977_begin_0, end = var_21977_end_0, end_mask = var_21977_end_mask_0, x = var_21676_cast_fp16)[name = string("op_21977_cast_fp16")]; + tensor var_21984_begin_0 = const()[name = string("op_21984_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21984_end_0 = const()[name = string("op_21984_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21984_end_mask_0 = const()[name = string("op_21984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21984_cast_fp16 = slice_by_index(begin = var_21984_begin_0, end = var_21984_end_0, end_mask = var_21984_end_mask_0, x = var_21676_cast_fp16)[name = string("op_21984_cast_fp16")]; + tensor var_21991_begin_0 = const()[name = string("op_21991_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21991_end_0 = const()[name = string("op_21991_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21991_end_mask_0 = const()[name = string("op_21991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21991_cast_fp16 = slice_by_index(begin = var_21991_begin_0, end = var_21991_end_0, end_mask = var_21991_end_mask_0, x = var_21676_cast_fp16)[name = string("op_21991_cast_fp16")]; + tensor var_21998_begin_0 = const()[name = string("op_21998_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21998_end_0 = const()[name = string("op_21998_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21998_end_mask_0 = const()[name = string("op_21998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21998_cast_fp16 = slice_by_index(begin = var_21998_begin_0, end = var_21998_end_0, end_mask = var_21998_end_mask_0, x = var_21676_cast_fp16)[name = string("op_21998_cast_fp16")]; + tensor var_22005_begin_0 = const()[name = string("op_22005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22005_end_0 = const()[name = string("op_22005_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22005_end_mask_0 = const()[name = string("op_22005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22005_cast_fp16 = slice_by_index(begin = var_22005_begin_0, end = var_22005_end_0, end_mask = var_22005_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22005_cast_fp16")]; + tensor var_22012_begin_0 = const()[name = string("op_22012_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22012_end_0 = const()[name = string("op_22012_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22012_end_mask_0 = const()[name = string("op_22012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22012_cast_fp16 = slice_by_index(begin = var_22012_begin_0, end = var_22012_end_0, end_mask = var_22012_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22012_cast_fp16")]; + tensor var_22019_begin_0 = const()[name = string("op_22019_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22019_end_0 = const()[name = string("op_22019_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22019_end_mask_0 = const()[name = string("op_22019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22019_cast_fp16 = slice_by_index(begin = var_22019_begin_0, end = var_22019_end_0, end_mask = var_22019_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22019_cast_fp16")]; + tensor var_22026_begin_0 = const()[name = string("op_22026_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22026_end_0 = const()[name = string("op_22026_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22026_end_mask_0 = const()[name = string("op_22026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22026_cast_fp16 = slice_by_index(begin = var_22026_begin_0, end = var_22026_end_0, end_mask = var_22026_end_mask_0, x = var_21680_cast_fp16)[name = string("op_22026_cast_fp16")]; + tensor var_22033_begin_0 = const()[name = string("op_22033_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22033_end_0 = const()[name = string("op_22033_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22033_end_mask_0 = const()[name = string("op_22033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22033_cast_fp16 = slice_by_index(begin = var_22033_begin_0, end = var_22033_end_0, end_mask = var_22033_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22033_cast_fp16")]; + tensor var_22040_begin_0 = const()[name = string("op_22040_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22040_end_0 = const()[name = string("op_22040_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22040_end_mask_0 = const()[name = string("op_22040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22040_cast_fp16 = slice_by_index(begin = var_22040_begin_0, end = var_22040_end_0, end_mask = var_22040_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22040_cast_fp16")]; + tensor var_22047_begin_0 = const()[name = string("op_22047_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22047_end_0 = const()[name = string("op_22047_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22047_end_mask_0 = const()[name = string("op_22047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22047_cast_fp16 = slice_by_index(begin = var_22047_begin_0, end = var_22047_end_0, end_mask = var_22047_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22047_cast_fp16")]; + tensor var_22054_begin_0 = const()[name = string("op_22054_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22054_end_0 = const()[name = string("op_22054_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22054_end_mask_0 = const()[name = string("op_22054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22054_cast_fp16 = slice_by_index(begin = var_22054_begin_0, end = var_22054_end_0, end_mask = var_22054_end_mask_0, x = var_21684_cast_fp16)[name = string("op_22054_cast_fp16")]; + tensor var_22061_begin_0 = const()[name = string("op_22061_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22061_end_0 = const()[name = string("op_22061_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22061_end_mask_0 = const()[name = string("op_22061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22061_cast_fp16 = slice_by_index(begin = var_22061_begin_0, end = var_22061_end_0, end_mask = var_22061_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22061_cast_fp16")]; + tensor var_22068_begin_0 = const()[name = string("op_22068_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22068_end_0 = const()[name = string("op_22068_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22068_end_mask_0 = const()[name = string("op_22068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22068_cast_fp16 = slice_by_index(begin = var_22068_begin_0, end = var_22068_end_0, end_mask = var_22068_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22068_cast_fp16")]; + tensor var_22075_begin_0 = const()[name = string("op_22075_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22075_end_0 = const()[name = string("op_22075_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22075_end_mask_0 = const()[name = string("op_22075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22075_cast_fp16 = slice_by_index(begin = var_22075_begin_0, end = var_22075_end_0, end_mask = var_22075_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22075_cast_fp16")]; + tensor var_22082_begin_0 = const()[name = string("op_22082_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22082_end_0 = const()[name = string("op_22082_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22082_end_mask_0 = const()[name = string("op_22082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22082_cast_fp16 = slice_by_index(begin = var_22082_begin_0, end = var_22082_end_0, end_mask = var_22082_end_mask_0, x = var_21688_cast_fp16)[name = string("op_22082_cast_fp16")]; + tensor var_22089_begin_0 = const()[name = string("op_22089_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22089_end_0 = const()[name = string("op_22089_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22089_end_mask_0 = const()[name = string("op_22089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22089_cast_fp16 = slice_by_index(begin = var_22089_begin_0, end = var_22089_end_0, end_mask = var_22089_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22089_cast_fp16")]; + tensor var_22096_begin_0 = const()[name = string("op_22096_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22096_end_0 = const()[name = string("op_22096_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22096_end_mask_0 = const()[name = string("op_22096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22096_cast_fp16 = slice_by_index(begin = var_22096_begin_0, end = var_22096_end_0, end_mask = var_22096_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22096_cast_fp16")]; + tensor var_22103_begin_0 = const()[name = string("op_22103_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22103_end_0 = const()[name = string("op_22103_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22103_end_mask_0 = const()[name = string("op_22103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22103_cast_fp16 = slice_by_index(begin = var_22103_begin_0, end = var_22103_end_0, end_mask = var_22103_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22103_cast_fp16")]; + tensor var_22110_begin_0 = const()[name = string("op_22110_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22110_end_0 = const()[name = string("op_22110_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22110_end_mask_0 = const()[name = string("op_22110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22110_cast_fp16 = slice_by_index(begin = var_22110_begin_0, end = var_22110_end_0, end_mask = var_22110_end_mask_0, x = var_21692_cast_fp16)[name = string("op_22110_cast_fp16")]; + tensor var_22117_begin_0 = const()[name = string("op_22117_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22117_end_0 = const()[name = string("op_22117_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22117_end_mask_0 = const()[name = string("op_22117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22117_cast_fp16 = slice_by_index(begin = var_22117_begin_0, end = var_22117_end_0, end_mask = var_22117_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22117_cast_fp16")]; + tensor var_22124_begin_0 = const()[name = string("op_22124_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22124_end_0 = const()[name = string("op_22124_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22124_end_mask_0 = const()[name = string("op_22124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22124_cast_fp16 = slice_by_index(begin = var_22124_begin_0, end = var_22124_end_0, end_mask = var_22124_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22124_cast_fp16")]; + tensor var_22131_begin_0 = const()[name = string("op_22131_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22131_end_0 = const()[name = string("op_22131_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22131_end_mask_0 = const()[name = string("op_22131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22131_cast_fp16 = slice_by_index(begin = var_22131_begin_0, end = var_22131_end_0, end_mask = var_22131_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22131_cast_fp16")]; + tensor var_22138_begin_0 = const()[name = string("op_22138_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22138_end_0 = const()[name = string("op_22138_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22138_end_mask_0 = const()[name = string("op_22138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22138_cast_fp16 = slice_by_index(begin = var_22138_begin_0, end = var_22138_end_0, end_mask = var_22138_end_mask_0, x = var_21696_cast_fp16)[name = string("op_22138_cast_fp16")]; + tensor var_22145_begin_0 = const()[name = string("op_22145_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22145_end_0 = const()[name = string("op_22145_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22145_end_mask_0 = const()[name = string("op_22145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22145_cast_fp16 = slice_by_index(begin = var_22145_begin_0, end = var_22145_end_0, end_mask = var_22145_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22145_cast_fp16")]; + tensor var_22152_begin_0 = const()[name = string("op_22152_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22152_end_0 = const()[name = string("op_22152_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22152_end_mask_0 = const()[name = string("op_22152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22152_cast_fp16 = slice_by_index(begin = var_22152_begin_0, end = var_22152_end_0, end_mask = var_22152_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22152_cast_fp16")]; + tensor var_22159_begin_0 = const()[name = string("op_22159_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22159_end_0 = const()[name = string("op_22159_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22159_end_mask_0 = const()[name = string("op_22159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22159_cast_fp16 = slice_by_index(begin = var_22159_begin_0, end = var_22159_end_0, end_mask = var_22159_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22159_cast_fp16")]; + tensor var_22166_begin_0 = const()[name = string("op_22166_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22166_end_0 = const()[name = string("op_22166_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22166_end_mask_0 = const()[name = string("op_22166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22166_cast_fp16 = slice_by_index(begin = var_22166_begin_0, end = var_22166_end_0, end_mask = var_22166_end_mask_0, x = var_21700_cast_fp16)[name = string("op_22166_cast_fp16")]; + tensor var_22173_begin_0 = const()[name = string("op_22173_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22173_end_0 = const()[name = string("op_22173_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22173_end_mask_0 = const()[name = string("op_22173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22173_cast_fp16 = slice_by_index(begin = var_22173_begin_0, end = var_22173_end_0, end_mask = var_22173_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22173_cast_fp16")]; + tensor var_22180_begin_0 = const()[name = string("op_22180_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22180_end_0 = const()[name = string("op_22180_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22180_end_mask_0 = const()[name = string("op_22180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22180_cast_fp16 = slice_by_index(begin = var_22180_begin_0, end = var_22180_end_0, end_mask = var_22180_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22180_cast_fp16")]; + tensor var_22187_begin_0 = const()[name = string("op_22187_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22187_end_0 = const()[name = string("op_22187_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22187_end_mask_0 = const()[name = string("op_22187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22187_cast_fp16 = slice_by_index(begin = var_22187_begin_0, end = var_22187_end_0, end_mask = var_22187_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22187_cast_fp16")]; + tensor var_22194_begin_0 = const()[name = string("op_22194_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22194_end_0 = const()[name = string("op_22194_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22194_end_mask_0 = const()[name = string("op_22194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22194_cast_fp16 = slice_by_index(begin = var_22194_begin_0, end = var_22194_end_0, end_mask = var_22194_end_mask_0, x = var_21704_cast_fp16)[name = string("op_22194_cast_fp16")]; + tensor var_22201_begin_0 = const()[name = string("op_22201_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22201_end_0 = const()[name = string("op_22201_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22201_end_mask_0 = const()[name = string("op_22201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22201_cast_fp16 = slice_by_index(begin = var_22201_begin_0, end = var_22201_end_0, end_mask = var_22201_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22201_cast_fp16")]; + tensor var_22208_begin_0 = const()[name = string("op_22208_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22208_end_0 = const()[name = string("op_22208_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22208_end_mask_0 = const()[name = string("op_22208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22208_cast_fp16 = slice_by_index(begin = var_22208_begin_0, end = var_22208_end_0, end_mask = var_22208_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22208_cast_fp16")]; + tensor var_22215_begin_0 = const()[name = string("op_22215_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22215_end_0 = const()[name = string("op_22215_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22215_end_mask_0 = const()[name = string("op_22215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22215_cast_fp16 = slice_by_index(begin = var_22215_begin_0, end = var_22215_end_0, end_mask = var_22215_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22215_cast_fp16")]; + tensor var_22222_begin_0 = const()[name = string("op_22222_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22222_end_0 = const()[name = string("op_22222_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22222_end_mask_0 = const()[name = string("op_22222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22222_cast_fp16 = slice_by_index(begin = var_22222_begin_0, end = var_22222_end_0, end_mask = var_22222_end_mask_0, x = var_21708_cast_fp16)[name = string("op_22222_cast_fp16")]; + tensor var_22229_begin_0 = const()[name = string("op_22229_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22229_end_0 = const()[name = string("op_22229_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22229_end_mask_0 = const()[name = string("op_22229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22229_cast_fp16 = slice_by_index(begin = var_22229_begin_0, end = var_22229_end_0, end_mask = var_22229_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22229_cast_fp16")]; + tensor var_22236_begin_0 = const()[name = string("op_22236_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22236_end_0 = const()[name = string("op_22236_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22236_end_mask_0 = const()[name = string("op_22236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22236_cast_fp16 = slice_by_index(begin = var_22236_begin_0, end = var_22236_end_0, end_mask = var_22236_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22236_cast_fp16")]; + tensor var_22243_begin_0 = const()[name = string("op_22243_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22243_end_0 = const()[name = string("op_22243_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22243_end_mask_0 = const()[name = string("op_22243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22243_cast_fp16 = slice_by_index(begin = var_22243_begin_0, end = var_22243_end_0, end_mask = var_22243_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22243_cast_fp16")]; + tensor var_22250_begin_0 = const()[name = string("op_22250_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22250_end_0 = const()[name = string("op_22250_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22250_end_mask_0 = const()[name = string("op_22250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22250_cast_fp16 = slice_by_index(begin = var_22250_begin_0, end = var_22250_end_0, end_mask = var_22250_end_mask_0, x = var_21712_cast_fp16)[name = string("op_22250_cast_fp16")]; + tensor var_22257_begin_0 = const()[name = string("op_22257_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22257_end_0 = const()[name = string("op_22257_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22257_end_mask_0 = const()[name = string("op_22257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22257_cast_fp16 = slice_by_index(begin = var_22257_begin_0, end = var_22257_end_0, end_mask = var_22257_end_mask_0, x = var_21716_cast_fp16)[name = string("op_22257_cast_fp16")]; + tensor var_22264_begin_0 = const()[name = string("op_22264_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22264_end_0 = const()[name = string("op_22264_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22264_end_mask_0 = const()[name = string("op_22264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22264_cast_fp16 = slice_by_index(begin = var_22264_begin_0, end = var_22264_end_0, end_mask = var_22264_end_mask_0, x = var_21716_cast_fp16)[name = string("op_22264_cast_fp16")]; + tensor var_22271_begin_0 = const()[name = string("op_22271_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22271_end_0 = const()[name = string("op_22271_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22271_end_mask_0 = const()[name = string("op_22271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22271_cast_fp16 = slice_by_index(begin = var_22271_begin_0, end = var_22271_end_0, end_mask = var_22271_end_mask_0, x = var_21716_cast_fp16)[name = string("op_22271_cast_fp16")]; + tensor var_22278_begin_0 = const()[name = string("op_22278_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22278_end_0 = const()[name = string("op_22278_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22278_end_mask_0 = const()[name = string("op_22278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22278_cast_fp16 = slice_by_index(begin = var_22278_begin_0, end = var_22278_end_0, end_mask = var_22278_end_mask_0, x = var_21716_cast_fp16)[name = string("op_22278_cast_fp16")]; + tensor k_29_perm_0 = const()[name = string("k_29_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22283_begin_0 = const()[name = string("op_22283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22283_end_0 = const()[name = string("op_22283_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_22283_end_mask_0 = const()[name = string("op_22283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_29_cast_fp16 = transpose(perm = k_29_perm_0, x = key_29_cast_fp16)[name = string("transpose_17")]; + tensor var_22283_cast_fp16 = slice_by_index(begin = var_22283_begin_0, end = var_22283_end_0, end_mask = var_22283_end_mask_0, x = k_29_cast_fp16)[name = string("op_22283_cast_fp16")]; + tensor var_22287_begin_0 = const()[name = string("op_22287_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22287_end_0 = const()[name = string("op_22287_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_22287_end_mask_0 = const()[name = string("op_22287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22287_cast_fp16 = slice_by_index(begin = var_22287_begin_0, end = var_22287_end_0, end_mask = var_22287_end_mask_0, x = k_29_cast_fp16)[name = string("op_22287_cast_fp16")]; + tensor var_22291_begin_0 = const()[name = string("op_22291_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22291_end_0 = const()[name = string("op_22291_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_22291_end_mask_0 = const()[name = string("op_22291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22291_cast_fp16 = slice_by_index(begin = var_22291_begin_0, end = var_22291_end_0, end_mask = var_22291_end_mask_0, x = k_29_cast_fp16)[name = string("op_22291_cast_fp16")]; + tensor var_22295_begin_0 = const()[name = string("op_22295_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22295_end_0 = const()[name = string("op_22295_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_22295_end_mask_0 = const()[name = string("op_22295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22295_cast_fp16 = slice_by_index(begin = var_22295_begin_0, end = var_22295_end_0, end_mask = var_22295_end_mask_0, x = k_29_cast_fp16)[name = string("op_22295_cast_fp16")]; + tensor var_22299_begin_0 = const()[name = string("op_22299_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22299_end_0 = const()[name = string("op_22299_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_22299_end_mask_0 = const()[name = string("op_22299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22299_cast_fp16 = slice_by_index(begin = var_22299_begin_0, end = var_22299_end_0, end_mask = var_22299_end_mask_0, x = k_29_cast_fp16)[name = string("op_22299_cast_fp16")]; + tensor var_22303_begin_0 = const()[name = string("op_22303_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22303_end_0 = const()[name = string("op_22303_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_22303_end_mask_0 = const()[name = string("op_22303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22303_cast_fp16 = slice_by_index(begin = var_22303_begin_0, end = var_22303_end_0, end_mask = var_22303_end_mask_0, x = k_29_cast_fp16)[name = string("op_22303_cast_fp16")]; + tensor var_22307_begin_0 = const()[name = string("op_22307_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22307_end_0 = const()[name = string("op_22307_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_22307_end_mask_0 = const()[name = string("op_22307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22307_cast_fp16 = slice_by_index(begin = var_22307_begin_0, end = var_22307_end_0, end_mask = var_22307_end_mask_0, x = k_29_cast_fp16)[name = string("op_22307_cast_fp16")]; + tensor var_22311_begin_0 = const()[name = string("op_22311_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22311_end_0 = const()[name = string("op_22311_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_22311_end_mask_0 = const()[name = string("op_22311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22311_cast_fp16 = slice_by_index(begin = var_22311_begin_0, end = var_22311_end_0, end_mask = var_22311_end_mask_0, x = k_29_cast_fp16)[name = string("op_22311_cast_fp16")]; + tensor var_22315_begin_0 = const()[name = string("op_22315_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22315_end_0 = const()[name = string("op_22315_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_22315_end_mask_0 = const()[name = string("op_22315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22315_cast_fp16 = slice_by_index(begin = var_22315_begin_0, end = var_22315_end_0, end_mask = var_22315_end_mask_0, x = k_29_cast_fp16)[name = string("op_22315_cast_fp16")]; + tensor var_22319_begin_0 = const()[name = string("op_22319_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22319_end_0 = const()[name = string("op_22319_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_22319_end_mask_0 = const()[name = string("op_22319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22319_cast_fp16 = slice_by_index(begin = var_22319_begin_0, end = var_22319_end_0, end_mask = var_22319_end_mask_0, x = k_29_cast_fp16)[name = string("op_22319_cast_fp16")]; + tensor var_22323_begin_0 = const()[name = string("op_22323_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22323_end_0 = const()[name = string("op_22323_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_22323_end_mask_0 = const()[name = string("op_22323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22323_cast_fp16 = slice_by_index(begin = var_22323_begin_0, end = var_22323_end_0, end_mask = var_22323_end_mask_0, x = k_29_cast_fp16)[name = string("op_22323_cast_fp16")]; + tensor var_22327_begin_0 = const()[name = string("op_22327_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22327_end_0 = const()[name = string("op_22327_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_22327_end_mask_0 = const()[name = string("op_22327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22327_cast_fp16 = slice_by_index(begin = var_22327_begin_0, end = var_22327_end_0, end_mask = var_22327_end_mask_0, x = k_29_cast_fp16)[name = string("op_22327_cast_fp16")]; + tensor var_22331_begin_0 = const()[name = string("op_22331_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22331_end_0 = const()[name = string("op_22331_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_22331_end_mask_0 = const()[name = string("op_22331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22331_cast_fp16 = slice_by_index(begin = var_22331_begin_0, end = var_22331_end_0, end_mask = var_22331_end_mask_0, x = k_29_cast_fp16)[name = string("op_22331_cast_fp16")]; + tensor var_22335_begin_0 = const()[name = string("op_22335_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22335_end_0 = const()[name = string("op_22335_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_22335_end_mask_0 = const()[name = string("op_22335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22335_cast_fp16 = slice_by_index(begin = var_22335_begin_0, end = var_22335_end_0, end_mask = var_22335_end_mask_0, x = k_29_cast_fp16)[name = string("op_22335_cast_fp16")]; + tensor var_22339_begin_0 = const()[name = string("op_22339_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22339_end_0 = const()[name = string("op_22339_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_22339_end_mask_0 = const()[name = string("op_22339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22339_cast_fp16 = slice_by_index(begin = var_22339_begin_0, end = var_22339_end_0, end_mask = var_22339_end_mask_0, x = k_29_cast_fp16)[name = string("op_22339_cast_fp16")]; + tensor var_22343_begin_0 = const()[name = string("op_22343_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22343_end_0 = const()[name = string("op_22343_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_22343_end_mask_0 = const()[name = string("op_22343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22343_cast_fp16 = slice_by_index(begin = var_22343_begin_0, end = var_22343_end_0, end_mask = var_22343_end_mask_0, x = k_29_cast_fp16)[name = string("op_22343_cast_fp16")]; + tensor var_22347_begin_0 = const()[name = string("op_22347_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22347_end_0 = const()[name = string("op_22347_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_22347_end_mask_0 = const()[name = string("op_22347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22347_cast_fp16 = slice_by_index(begin = var_22347_begin_0, end = var_22347_end_0, end_mask = var_22347_end_mask_0, x = k_29_cast_fp16)[name = string("op_22347_cast_fp16")]; + tensor var_22351_begin_0 = const()[name = string("op_22351_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22351_end_0 = const()[name = string("op_22351_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_22351_end_mask_0 = const()[name = string("op_22351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22351_cast_fp16 = slice_by_index(begin = var_22351_begin_0, end = var_22351_end_0, end_mask = var_22351_end_mask_0, x = k_29_cast_fp16)[name = string("op_22351_cast_fp16")]; + tensor var_22355_begin_0 = const()[name = string("op_22355_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22355_end_0 = const()[name = string("op_22355_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_22355_end_mask_0 = const()[name = string("op_22355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22355_cast_fp16 = slice_by_index(begin = var_22355_begin_0, end = var_22355_end_0, end_mask = var_22355_end_mask_0, x = k_29_cast_fp16)[name = string("op_22355_cast_fp16")]; + tensor var_22359_begin_0 = const()[name = string("op_22359_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22359_end_0 = const()[name = string("op_22359_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_22359_end_mask_0 = const()[name = string("op_22359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22359_cast_fp16 = slice_by_index(begin = var_22359_begin_0, end = var_22359_end_0, end_mask = var_22359_end_mask_0, x = k_29_cast_fp16)[name = string("op_22359_cast_fp16")]; + tensor var_22361_begin_0 = const()[name = string("op_22361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22361_end_0 = const()[name = string("op_22361_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22361_end_mask_0 = const()[name = string("op_22361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22361_cast_fp16 = slice_by_index(begin = var_22361_begin_0, end = var_22361_end_0, end_mask = var_22361_end_mask_0, x = value_29_cast_fp16)[name = string("op_22361_cast_fp16")]; + tensor var_22365_begin_0 = const()[name = string("op_22365_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22365_end_0 = const()[name = string("op_22365_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_22365_end_mask_0 = const()[name = string("op_22365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22365_cast_fp16 = slice_by_index(begin = var_22365_begin_0, end = var_22365_end_0, end_mask = var_22365_end_mask_0, x = value_29_cast_fp16)[name = string("op_22365_cast_fp16")]; + tensor var_22369_begin_0 = const()[name = string("op_22369_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22369_end_0 = const()[name = string("op_22369_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_22369_end_mask_0 = const()[name = string("op_22369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22369_cast_fp16 = slice_by_index(begin = var_22369_begin_0, end = var_22369_end_0, end_mask = var_22369_end_mask_0, x = value_29_cast_fp16)[name = string("op_22369_cast_fp16")]; + tensor var_22373_begin_0 = const()[name = string("op_22373_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22373_end_0 = const()[name = string("op_22373_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_22373_end_mask_0 = const()[name = string("op_22373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22373_cast_fp16 = slice_by_index(begin = var_22373_begin_0, end = var_22373_end_0, end_mask = var_22373_end_mask_0, x = value_29_cast_fp16)[name = string("op_22373_cast_fp16")]; + tensor var_22377_begin_0 = const()[name = string("op_22377_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22377_end_0 = const()[name = string("op_22377_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_22377_end_mask_0 = const()[name = string("op_22377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22377_cast_fp16 = slice_by_index(begin = var_22377_begin_0, end = var_22377_end_0, end_mask = var_22377_end_mask_0, x = value_29_cast_fp16)[name = string("op_22377_cast_fp16")]; + tensor var_22381_begin_0 = const()[name = string("op_22381_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22381_end_0 = const()[name = string("op_22381_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_22381_end_mask_0 = const()[name = string("op_22381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22381_cast_fp16 = slice_by_index(begin = var_22381_begin_0, end = var_22381_end_0, end_mask = var_22381_end_mask_0, x = value_29_cast_fp16)[name = string("op_22381_cast_fp16")]; + tensor var_22385_begin_0 = const()[name = string("op_22385_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22385_end_0 = const()[name = string("op_22385_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_22385_end_mask_0 = const()[name = string("op_22385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22385_cast_fp16 = slice_by_index(begin = var_22385_begin_0, end = var_22385_end_0, end_mask = var_22385_end_mask_0, x = value_29_cast_fp16)[name = string("op_22385_cast_fp16")]; + tensor var_22389_begin_0 = const()[name = string("op_22389_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22389_end_0 = const()[name = string("op_22389_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_22389_end_mask_0 = const()[name = string("op_22389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22389_cast_fp16 = slice_by_index(begin = var_22389_begin_0, end = var_22389_end_0, end_mask = var_22389_end_mask_0, x = value_29_cast_fp16)[name = string("op_22389_cast_fp16")]; + tensor var_22393_begin_0 = const()[name = string("op_22393_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22393_end_0 = const()[name = string("op_22393_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_22393_end_mask_0 = const()[name = string("op_22393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22393_cast_fp16 = slice_by_index(begin = var_22393_begin_0, end = var_22393_end_0, end_mask = var_22393_end_mask_0, x = value_29_cast_fp16)[name = string("op_22393_cast_fp16")]; + tensor var_22397_begin_0 = const()[name = string("op_22397_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22397_end_0 = const()[name = string("op_22397_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_22397_end_mask_0 = const()[name = string("op_22397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22397_cast_fp16 = slice_by_index(begin = var_22397_begin_0, end = var_22397_end_0, end_mask = var_22397_end_mask_0, x = value_29_cast_fp16)[name = string("op_22397_cast_fp16")]; + tensor var_22401_begin_0 = const()[name = string("op_22401_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22401_end_0 = const()[name = string("op_22401_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_22401_end_mask_0 = const()[name = string("op_22401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22401_cast_fp16 = slice_by_index(begin = var_22401_begin_0, end = var_22401_end_0, end_mask = var_22401_end_mask_0, x = value_29_cast_fp16)[name = string("op_22401_cast_fp16")]; + tensor var_22405_begin_0 = const()[name = string("op_22405_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22405_end_0 = const()[name = string("op_22405_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_22405_end_mask_0 = const()[name = string("op_22405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22405_cast_fp16 = slice_by_index(begin = var_22405_begin_0, end = var_22405_end_0, end_mask = var_22405_end_mask_0, x = value_29_cast_fp16)[name = string("op_22405_cast_fp16")]; + tensor var_22409_begin_0 = const()[name = string("op_22409_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22409_end_0 = const()[name = string("op_22409_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_22409_end_mask_0 = const()[name = string("op_22409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22409_cast_fp16 = slice_by_index(begin = var_22409_begin_0, end = var_22409_end_0, end_mask = var_22409_end_mask_0, x = value_29_cast_fp16)[name = string("op_22409_cast_fp16")]; + tensor var_22413_begin_0 = const()[name = string("op_22413_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22413_end_0 = const()[name = string("op_22413_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_22413_end_mask_0 = const()[name = string("op_22413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22413_cast_fp16 = slice_by_index(begin = var_22413_begin_0, end = var_22413_end_0, end_mask = var_22413_end_mask_0, x = value_29_cast_fp16)[name = string("op_22413_cast_fp16")]; + tensor var_22417_begin_0 = const()[name = string("op_22417_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22417_end_0 = const()[name = string("op_22417_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_22417_end_mask_0 = const()[name = string("op_22417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22417_cast_fp16 = slice_by_index(begin = var_22417_begin_0, end = var_22417_end_0, end_mask = var_22417_end_mask_0, x = value_29_cast_fp16)[name = string("op_22417_cast_fp16")]; + tensor var_22421_begin_0 = const()[name = string("op_22421_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22421_end_0 = const()[name = string("op_22421_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_22421_end_mask_0 = const()[name = string("op_22421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22421_cast_fp16 = slice_by_index(begin = var_22421_begin_0, end = var_22421_end_0, end_mask = var_22421_end_mask_0, x = value_29_cast_fp16)[name = string("op_22421_cast_fp16")]; + tensor var_22425_begin_0 = const()[name = string("op_22425_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22425_end_0 = const()[name = string("op_22425_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_22425_end_mask_0 = const()[name = string("op_22425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22425_cast_fp16 = slice_by_index(begin = var_22425_begin_0, end = var_22425_end_0, end_mask = var_22425_end_mask_0, x = value_29_cast_fp16)[name = string("op_22425_cast_fp16")]; + tensor var_22429_begin_0 = const()[name = string("op_22429_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22429_end_0 = const()[name = string("op_22429_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_22429_end_mask_0 = const()[name = string("op_22429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22429_cast_fp16 = slice_by_index(begin = var_22429_begin_0, end = var_22429_end_0, end_mask = var_22429_end_mask_0, x = value_29_cast_fp16)[name = string("op_22429_cast_fp16")]; + tensor var_22433_begin_0 = const()[name = string("op_22433_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22433_end_0 = const()[name = string("op_22433_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_22433_end_mask_0 = const()[name = string("op_22433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22433_cast_fp16 = slice_by_index(begin = var_22433_begin_0, end = var_22433_end_0, end_mask = var_22433_end_mask_0, x = value_29_cast_fp16)[name = string("op_22433_cast_fp16")]; + tensor var_22437_begin_0 = const()[name = string("op_22437_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22437_end_0 = const()[name = string("op_22437_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_22437_end_mask_0 = const()[name = string("op_22437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22437_cast_fp16 = slice_by_index(begin = var_22437_begin_0, end = var_22437_end_0, end_mask = var_22437_end_mask_0, x = value_29_cast_fp16)[name = string("op_22437_cast_fp16")]; + string _SplitHeadsQ__mh_w_2241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2241_equation_0, values = (var_22283_cast_fp16, var_21725_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2241_cast_fp16")]; + string _SplitHeadsQ__mh_w_2243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2243_equation_0, values = (var_22283_cast_fp16, var_21732_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2243_cast_fp16")]; + string _SplitHeadsQ__mh_w_2245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2245_equation_0, values = (var_22283_cast_fp16, var_21739_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2245_cast_fp16")]; + string _SplitHeadsQ__mh_w_2247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2247_equation_0, values = (var_22283_cast_fp16, var_21746_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2247_cast_fp16")]; + string _SplitHeadsQ__mh_w_2249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2249_equation_0, values = (var_22287_cast_fp16, var_21753_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2249_cast_fp16")]; + string _SplitHeadsQ__mh_w_2251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2251_equation_0, values = (var_22287_cast_fp16, var_21760_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2251_cast_fp16")]; + string _SplitHeadsQ__mh_w_2253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2253_equation_0, values = (var_22287_cast_fp16, var_21767_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2253_cast_fp16")]; + string _SplitHeadsQ__mh_w_2255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2255_equation_0, values = (var_22287_cast_fp16, var_21774_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2255_cast_fp16")]; + string _SplitHeadsQ__mh_w_2257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2257_equation_0, values = (var_22291_cast_fp16, var_21781_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2257_cast_fp16")]; + string _SplitHeadsQ__mh_w_2259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2259_equation_0, values = (var_22291_cast_fp16, var_21788_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2259_cast_fp16")]; + string _SplitHeadsQ__mh_w_2261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2261_equation_0, values = (var_22291_cast_fp16, var_21795_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2261_cast_fp16")]; + string _SplitHeadsQ__mh_w_2263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2263_equation_0, values = (var_22291_cast_fp16, var_21802_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2263_cast_fp16")]; + string _SplitHeadsQ__mh_w_2265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2265_equation_0, values = (var_22295_cast_fp16, var_21809_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2265_cast_fp16")]; + string _SplitHeadsQ__mh_w_2267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2267_equation_0, values = (var_22295_cast_fp16, var_21816_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2267_cast_fp16")]; + string _SplitHeadsQ__mh_w_2269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2269_equation_0, values = (var_22295_cast_fp16, var_21823_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2269_cast_fp16")]; + string _SplitHeadsQ__mh_w_2271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2271_equation_0, values = (var_22295_cast_fp16, var_21830_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2271_cast_fp16")]; + string _SplitHeadsQ__mh_w_2273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2273_equation_0, values = (var_22299_cast_fp16, var_21837_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2273_cast_fp16")]; + string _SplitHeadsQ__mh_w_2275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2275_equation_0, values = (var_22299_cast_fp16, var_21844_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2275_cast_fp16")]; + string _SplitHeadsQ__mh_w_2277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2277_equation_0, values = (var_22299_cast_fp16, var_21851_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2277_cast_fp16")]; + string _SplitHeadsQ__mh_w_2279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2279_equation_0, values = (var_22299_cast_fp16, var_21858_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2279_cast_fp16")]; + string _SplitHeadsQ__mh_w_2281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2281_equation_0, values = (var_22303_cast_fp16, var_21865_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2281_cast_fp16")]; + string _SplitHeadsQ__mh_w_2283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2283_equation_0, values = (var_22303_cast_fp16, var_21872_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2283_cast_fp16")]; + string _SplitHeadsQ__mh_w_2285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2285_equation_0, values = (var_22303_cast_fp16, var_21879_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2285_cast_fp16")]; + string _SplitHeadsQ__mh_w_2287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2287_equation_0, values = (var_22303_cast_fp16, var_21886_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2287_cast_fp16")]; + string _SplitHeadsQ__mh_w_2289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2289_equation_0, values = (var_22307_cast_fp16, var_21893_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2289_cast_fp16")]; + string _SplitHeadsQ__mh_w_2291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2291_equation_0, values = (var_22307_cast_fp16, var_21900_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2291_cast_fp16")]; + string _SplitHeadsQ__mh_w_2293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2293_equation_0, values = (var_22307_cast_fp16, var_21907_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2293_cast_fp16")]; + string _SplitHeadsQ__mh_w_2295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2295_equation_0, values = (var_22307_cast_fp16, var_21914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2295_cast_fp16")]; + string _SplitHeadsQ__mh_w_2297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2297_equation_0, values = (var_22311_cast_fp16, var_21921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2297_cast_fp16")]; + string _SplitHeadsQ__mh_w_2299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2299_equation_0, values = (var_22311_cast_fp16, var_21928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2299_cast_fp16")]; + string _SplitHeadsQ__mh_w_2301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2301_equation_0, values = (var_22311_cast_fp16, var_21935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2301_cast_fp16")]; + string _SplitHeadsQ__mh_w_2303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2303_equation_0, values = (var_22311_cast_fp16, var_21942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2303_cast_fp16")]; + string _SplitHeadsQ__mh_w_2305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2305_equation_0, values = (var_22315_cast_fp16, var_21949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2305_cast_fp16")]; + string _SplitHeadsQ__mh_w_2307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2307_equation_0, values = (var_22315_cast_fp16, var_21956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2307_cast_fp16")]; + string _SplitHeadsQ__mh_w_2309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2309_equation_0, values = (var_22315_cast_fp16, var_21963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2309_cast_fp16")]; + string _SplitHeadsQ__mh_w_2311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2311_equation_0, values = (var_22315_cast_fp16, var_21970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2311_cast_fp16")]; + string _SplitHeadsQ__mh_w_2313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2313_equation_0, values = (var_22319_cast_fp16, var_21977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2313_cast_fp16")]; + string _SplitHeadsQ__mh_w_2315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2315_equation_0, values = (var_22319_cast_fp16, var_21984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2315_cast_fp16")]; + string _SplitHeadsQ__mh_w_2317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2317_equation_0, values = (var_22319_cast_fp16, var_21991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2317_cast_fp16")]; + string _SplitHeadsQ__mh_w_2319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2319_equation_0, values = (var_22319_cast_fp16, var_21998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2319_cast_fp16")]; + string _SplitHeadsQ__mh_w_2321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2321_equation_0, values = (var_22323_cast_fp16, var_22005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2321_cast_fp16")]; + string _SplitHeadsQ__mh_w_2323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2323_equation_0, values = (var_22323_cast_fp16, var_22012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2323_cast_fp16")]; + string _SplitHeadsQ__mh_w_2325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2325_equation_0, values = (var_22323_cast_fp16, var_22019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2325_cast_fp16")]; + string _SplitHeadsQ__mh_w_2327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2327_equation_0, values = (var_22323_cast_fp16, var_22026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2327_cast_fp16")]; + string _SplitHeadsQ__mh_w_2329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2329_equation_0, values = (var_22327_cast_fp16, var_22033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2329_cast_fp16")]; + string _SplitHeadsQ__mh_w_2331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2331_equation_0, values = (var_22327_cast_fp16, var_22040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2331_cast_fp16")]; + string _SplitHeadsQ__mh_w_2333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2333_equation_0, values = (var_22327_cast_fp16, var_22047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2333_cast_fp16")]; + string _SplitHeadsQ__mh_w_2335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2335_equation_0, values = (var_22327_cast_fp16, var_22054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2335_cast_fp16")]; + string _SplitHeadsQ__mh_w_2337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2337_equation_0, values = (var_22331_cast_fp16, var_22061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2337_cast_fp16")]; + string _SplitHeadsQ__mh_w_2339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2339_equation_0, values = (var_22331_cast_fp16, var_22068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2339_cast_fp16")]; + string _SplitHeadsQ__mh_w_2341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2341_equation_0, values = (var_22331_cast_fp16, var_22075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2341_cast_fp16")]; + string _SplitHeadsQ__mh_w_2343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2343_equation_0, values = (var_22331_cast_fp16, var_22082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2343_cast_fp16")]; + string _SplitHeadsQ__mh_w_2345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2345_equation_0, values = (var_22335_cast_fp16, var_22089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2345_cast_fp16")]; + string _SplitHeadsQ__mh_w_2347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2347_equation_0, values = (var_22335_cast_fp16, var_22096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2347_cast_fp16")]; + string _SplitHeadsQ__mh_w_2349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2349_equation_0, values = (var_22335_cast_fp16, var_22103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2349_cast_fp16")]; + string _SplitHeadsQ__mh_w_2351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2351_equation_0, values = (var_22335_cast_fp16, var_22110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2351_cast_fp16")]; + string _SplitHeadsQ__mh_w_2353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2353_equation_0, values = (var_22339_cast_fp16, var_22117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2353_cast_fp16")]; + string _SplitHeadsQ__mh_w_2355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2355_equation_0, values = (var_22339_cast_fp16, var_22124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2355_cast_fp16")]; + string _SplitHeadsQ__mh_w_2357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2357_equation_0, values = (var_22339_cast_fp16, var_22131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2357_cast_fp16")]; + string _SplitHeadsQ__mh_w_2359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2359_equation_0, values = (var_22339_cast_fp16, var_22138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2359_cast_fp16")]; + string _SplitHeadsQ__mh_w_2361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2361_equation_0, values = (var_22343_cast_fp16, var_22145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2361_cast_fp16")]; + string _SplitHeadsQ__mh_w_2363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2363_equation_0, values = (var_22343_cast_fp16, var_22152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2363_cast_fp16")]; + string _SplitHeadsQ__mh_w_2365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2365_equation_0, values = (var_22343_cast_fp16, var_22159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2365_cast_fp16")]; + string _SplitHeadsQ__mh_w_2367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2367_equation_0, values = (var_22343_cast_fp16, var_22166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2367_cast_fp16")]; + string _SplitHeadsQ__mh_w_2369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2369_equation_0, values = (var_22347_cast_fp16, var_22173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2369_cast_fp16")]; + string _SplitHeadsQ__mh_w_2371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2371_equation_0, values = (var_22347_cast_fp16, var_22180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2371_cast_fp16")]; + string _SplitHeadsQ__mh_w_2373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2373_equation_0, values = (var_22347_cast_fp16, var_22187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2373_cast_fp16")]; + string _SplitHeadsQ__mh_w_2375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2375_equation_0, values = (var_22347_cast_fp16, var_22194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2375_cast_fp16")]; + string _SplitHeadsQ__mh_w_2377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2377_equation_0, values = (var_22351_cast_fp16, var_22201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2377_cast_fp16")]; + string _SplitHeadsQ__mh_w_2379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2379_equation_0, values = (var_22351_cast_fp16, var_22208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2379_cast_fp16")]; + string _SplitHeadsQ__mh_w_2381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2381_equation_0, values = (var_22351_cast_fp16, var_22215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2381_cast_fp16")]; + string _SplitHeadsQ__mh_w_2383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2383_equation_0, values = (var_22351_cast_fp16, var_22222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2383_cast_fp16")]; + string _SplitHeadsQ__mh_w_2385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2385_equation_0, values = (var_22355_cast_fp16, var_22229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2385_cast_fp16")]; + string _SplitHeadsQ__mh_w_2387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2387_equation_0, values = (var_22355_cast_fp16, var_22236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2387_cast_fp16")]; + string _SplitHeadsQ__mh_w_2389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2389_equation_0, values = (var_22355_cast_fp16, var_22243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2389_cast_fp16")]; + string _SplitHeadsQ__mh_w_2391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2391_equation_0, values = (var_22355_cast_fp16, var_22250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2391_cast_fp16")]; + string _SplitHeadsQ__mh_w_2393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2393_equation_0, values = (var_22359_cast_fp16, var_22257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2393_cast_fp16")]; + string _SplitHeadsQ__mh_w_2395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2395_equation_0, values = (var_22359_cast_fp16, var_22264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2395_cast_fp16")]; + string _SplitHeadsQ__mh_w_2397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2397_equation_0, values = (var_22359_cast_fp16, var_22271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2397_cast_fp16")]; + string _SplitHeadsQ__mh_w_2399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2399_equation_0, values = (var_22359_cast_fp16, var_22278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2399_cast_fp16")]; + fp16 var_22600_to_fp16 = const()[name = string("op_22600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2241_cast_fp16, y = var_22600_to_fp16)[name = string("aw_chunk_2241_cast_fp16")]; + fp16 var_22602_to_fp16 = const()[name = string("op_22602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2243_cast_fp16, y = var_22602_to_fp16)[name = string("aw_chunk_2243_cast_fp16")]; + fp16 var_22604_to_fp16 = const()[name = string("op_22604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2245_cast_fp16, y = var_22604_to_fp16)[name = string("aw_chunk_2245_cast_fp16")]; + fp16 var_22606_to_fp16 = const()[name = string("op_22606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2247_cast_fp16, y = var_22606_to_fp16)[name = string("aw_chunk_2247_cast_fp16")]; + fp16 var_22608_to_fp16 = const()[name = string("op_22608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2249_cast_fp16, y = var_22608_to_fp16)[name = string("aw_chunk_2249_cast_fp16")]; + fp16 var_22610_to_fp16 = const()[name = string("op_22610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2251_cast_fp16, y = var_22610_to_fp16)[name = string("aw_chunk_2251_cast_fp16")]; + fp16 var_22612_to_fp16 = const()[name = string("op_22612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2253_cast_fp16, y = var_22612_to_fp16)[name = string("aw_chunk_2253_cast_fp16")]; + fp16 var_22614_to_fp16 = const()[name = string("op_22614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2255_cast_fp16, y = var_22614_to_fp16)[name = string("aw_chunk_2255_cast_fp16")]; + fp16 var_22616_to_fp16 = const()[name = string("op_22616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2257_cast_fp16, y = var_22616_to_fp16)[name = string("aw_chunk_2257_cast_fp16")]; + fp16 var_22618_to_fp16 = const()[name = string("op_22618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2259_cast_fp16, y = var_22618_to_fp16)[name = string("aw_chunk_2259_cast_fp16")]; + fp16 var_22620_to_fp16 = const()[name = string("op_22620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2261_cast_fp16, y = var_22620_to_fp16)[name = string("aw_chunk_2261_cast_fp16")]; + fp16 var_22622_to_fp16 = const()[name = string("op_22622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2263_cast_fp16, y = var_22622_to_fp16)[name = string("aw_chunk_2263_cast_fp16")]; + fp16 var_22624_to_fp16 = const()[name = string("op_22624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2265_cast_fp16, y = var_22624_to_fp16)[name = string("aw_chunk_2265_cast_fp16")]; + fp16 var_22626_to_fp16 = const()[name = string("op_22626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2267_cast_fp16, y = var_22626_to_fp16)[name = string("aw_chunk_2267_cast_fp16")]; + fp16 var_22628_to_fp16 = const()[name = string("op_22628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2269_cast_fp16, y = var_22628_to_fp16)[name = string("aw_chunk_2269_cast_fp16")]; + fp16 var_22630_to_fp16 = const()[name = string("op_22630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2271_cast_fp16, y = var_22630_to_fp16)[name = string("aw_chunk_2271_cast_fp16")]; + fp16 var_22632_to_fp16 = const()[name = string("op_22632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2273_cast_fp16, y = var_22632_to_fp16)[name = string("aw_chunk_2273_cast_fp16")]; + fp16 var_22634_to_fp16 = const()[name = string("op_22634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2275_cast_fp16, y = var_22634_to_fp16)[name = string("aw_chunk_2275_cast_fp16")]; + fp16 var_22636_to_fp16 = const()[name = string("op_22636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2277_cast_fp16, y = var_22636_to_fp16)[name = string("aw_chunk_2277_cast_fp16")]; + fp16 var_22638_to_fp16 = const()[name = string("op_22638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2279_cast_fp16, y = var_22638_to_fp16)[name = string("aw_chunk_2279_cast_fp16")]; + fp16 var_22640_to_fp16 = const()[name = string("op_22640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2281_cast_fp16, y = var_22640_to_fp16)[name = string("aw_chunk_2281_cast_fp16")]; + fp16 var_22642_to_fp16 = const()[name = string("op_22642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2283_cast_fp16, y = var_22642_to_fp16)[name = string("aw_chunk_2283_cast_fp16")]; + fp16 var_22644_to_fp16 = const()[name = string("op_22644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2285_cast_fp16, y = var_22644_to_fp16)[name = string("aw_chunk_2285_cast_fp16")]; + fp16 var_22646_to_fp16 = const()[name = string("op_22646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2287_cast_fp16, y = var_22646_to_fp16)[name = string("aw_chunk_2287_cast_fp16")]; + fp16 var_22648_to_fp16 = const()[name = string("op_22648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2289_cast_fp16, y = var_22648_to_fp16)[name = string("aw_chunk_2289_cast_fp16")]; + fp16 var_22650_to_fp16 = const()[name = string("op_22650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2291_cast_fp16, y = var_22650_to_fp16)[name = string("aw_chunk_2291_cast_fp16")]; + fp16 var_22652_to_fp16 = const()[name = string("op_22652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2293_cast_fp16, y = var_22652_to_fp16)[name = string("aw_chunk_2293_cast_fp16")]; + fp16 var_22654_to_fp16 = const()[name = string("op_22654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2295_cast_fp16, y = var_22654_to_fp16)[name = string("aw_chunk_2295_cast_fp16")]; + fp16 var_22656_to_fp16 = const()[name = string("op_22656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2297_cast_fp16, y = var_22656_to_fp16)[name = string("aw_chunk_2297_cast_fp16")]; + fp16 var_22658_to_fp16 = const()[name = string("op_22658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2299_cast_fp16, y = var_22658_to_fp16)[name = string("aw_chunk_2299_cast_fp16")]; + fp16 var_22660_to_fp16 = const()[name = string("op_22660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2301_cast_fp16, y = var_22660_to_fp16)[name = string("aw_chunk_2301_cast_fp16")]; + fp16 var_22662_to_fp16 = const()[name = string("op_22662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2303_cast_fp16, y = var_22662_to_fp16)[name = string("aw_chunk_2303_cast_fp16")]; + fp16 var_22664_to_fp16 = const()[name = string("op_22664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2305_cast_fp16, y = var_22664_to_fp16)[name = string("aw_chunk_2305_cast_fp16")]; + fp16 var_22666_to_fp16 = const()[name = string("op_22666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2307_cast_fp16, y = var_22666_to_fp16)[name = string("aw_chunk_2307_cast_fp16")]; + fp16 var_22668_to_fp16 = const()[name = string("op_22668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2309_cast_fp16, y = var_22668_to_fp16)[name = string("aw_chunk_2309_cast_fp16")]; + fp16 var_22670_to_fp16 = const()[name = string("op_22670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2311_cast_fp16, y = var_22670_to_fp16)[name = string("aw_chunk_2311_cast_fp16")]; + fp16 var_22672_to_fp16 = const()[name = string("op_22672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2313_cast_fp16, y = var_22672_to_fp16)[name = string("aw_chunk_2313_cast_fp16")]; + fp16 var_22674_to_fp16 = const()[name = string("op_22674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2315_cast_fp16, y = var_22674_to_fp16)[name = string("aw_chunk_2315_cast_fp16")]; + fp16 var_22676_to_fp16 = const()[name = string("op_22676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2317_cast_fp16, y = var_22676_to_fp16)[name = string("aw_chunk_2317_cast_fp16")]; + fp16 var_22678_to_fp16 = const()[name = string("op_22678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2319_cast_fp16, y = var_22678_to_fp16)[name = string("aw_chunk_2319_cast_fp16")]; + fp16 var_22680_to_fp16 = const()[name = string("op_22680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2321_cast_fp16, y = var_22680_to_fp16)[name = string("aw_chunk_2321_cast_fp16")]; + fp16 var_22682_to_fp16 = const()[name = string("op_22682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2323_cast_fp16, y = var_22682_to_fp16)[name = string("aw_chunk_2323_cast_fp16")]; + fp16 var_22684_to_fp16 = const()[name = string("op_22684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2325_cast_fp16, y = var_22684_to_fp16)[name = string("aw_chunk_2325_cast_fp16")]; + fp16 var_22686_to_fp16 = const()[name = string("op_22686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2327_cast_fp16, y = var_22686_to_fp16)[name = string("aw_chunk_2327_cast_fp16")]; + fp16 var_22688_to_fp16 = const()[name = string("op_22688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2329_cast_fp16, y = var_22688_to_fp16)[name = string("aw_chunk_2329_cast_fp16")]; + fp16 var_22690_to_fp16 = const()[name = string("op_22690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2331_cast_fp16, y = var_22690_to_fp16)[name = string("aw_chunk_2331_cast_fp16")]; + fp16 var_22692_to_fp16 = const()[name = string("op_22692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2333_cast_fp16, y = var_22692_to_fp16)[name = string("aw_chunk_2333_cast_fp16")]; + fp16 var_22694_to_fp16 = const()[name = string("op_22694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2335_cast_fp16, y = var_22694_to_fp16)[name = string("aw_chunk_2335_cast_fp16")]; + fp16 var_22696_to_fp16 = const()[name = string("op_22696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2337_cast_fp16, y = var_22696_to_fp16)[name = string("aw_chunk_2337_cast_fp16")]; + fp16 var_22698_to_fp16 = const()[name = string("op_22698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2339_cast_fp16, y = var_22698_to_fp16)[name = string("aw_chunk_2339_cast_fp16")]; + fp16 var_22700_to_fp16 = const()[name = string("op_22700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2341_cast_fp16, y = var_22700_to_fp16)[name = string("aw_chunk_2341_cast_fp16")]; + fp16 var_22702_to_fp16 = const()[name = string("op_22702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2343_cast_fp16, y = var_22702_to_fp16)[name = string("aw_chunk_2343_cast_fp16")]; + fp16 var_22704_to_fp16 = const()[name = string("op_22704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2345_cast_fp16, y = var_22704_to_fp16)[name = string("aw_chunk_2345_cast_fp16")]; + fp16 var_22706_to_fp16 = const()[name = string("op_22706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2347_cast_fp16, y = var_22706_to_fp16)[name = string("aw_chunk_2347_cast_fp16")]; + fp16 var_22708_to_fp16 = const()[name = string("op_22708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2349_cast_fp16, y = var_22708_to_fp16)[name = string("aw_chunk_2349_cast_fp16")]; + fp16 var_22710_to_fp16 = const()[name = string("op_22710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2351_cast_fp16, y = var_22710_to_fp16)[name = string("aw_chunk_2351_cast_fp16")]; + fp16 var_22712_to_fp16 = const()[name = string("op_22712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2353_cast_fp16, y = var_22712_to_fp16)[name = string("aw_chunk_2353_cast_fp16")]; + fp16 var_22714_to_fp16 = const()[name = string("op_22714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2355_cast_fp16, y = var_22714_to_fp16)[name = string("aw_chunk_2355_cast_fp16")]; + fp16 var_22716_to_fp16 = const()[name = string("op_22716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2357_cast_fp16, y = var_22716_to_fp16)[name = string("aw_chunk_2357_cast_fp16")]; + fp16 var_22718_to_fp16 = const()[name = string("op_22718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2359_cast_fp16, y = var_22718_to_fp16)[name = string("aw_chunk_2359_cast_fp16")]; + fp16 var_22720_to_fp16 = const()[name = string("op_22720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2361_cast_fp16, y = var_22720_to_fp16)[name = string("aw_chunk_2361_cast_fp16")]; + fp16 var_22722_to_fp16 = const()[name = string("op_22722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2363_cast_fp16, y = var_22722_to_fp16)[name = string("aw_chunk_2363_cast_fp16")]; + fp16 var_22724_to_fp16 = const()[name = string("op_22724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2365_cast_fp16, y = var_22724_to_fp16)[name = string("aw_chunk_2365_cast_fp16")]; + fp16 var_22726_to_fp16 = const()[name = string("op_22726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2367_cast_fp16, y = var_22726_to_fp16)[name = string("aw_chunk_2367_cast_fp16")]; + fp16 var_22728_to_fp16 = const()[name = string("op_22728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2369_cast_fp16, y = var_22728_to_fp16)[name = string("aw_chunk_2369_cast_fp16")]; + fp16 var_22730_to_fp16 = const()[name = string("op_22730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2371_cast_fp16, y = var_22730_to_fp16)[name = string("aw_chunk_2371_cast_fp16")]; + fp16 var_22732_to_fp16 = const()[name = string("op_22732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2373_cast_fp16, y = var_22732_to_fp16)[name = string("aw_chunk_2373_cast_fp16")]; + fp16 var_22734_to_fp16 = const()[name = string("op_22734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2375_cast_fp16, y = var_22734_to_fp16)[name = string("aw_chunk_2375_cast_fp16")]; + fp16 var_22736_to_fp16 = const()[name = string("op_22736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2377_cast_fp16, y = var_22736_to_fp16)[name = string("aw_chunk_2377_cast_fp16")]; + fp16 var_22738_to_fp16 = const()[name = string("op_22738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2379_cast_fp16, y = var_22738_to_fp16)[name = string("aw_chunk_2379_cast_fp16")]; + fp16 var_22740_to_fp16 = const()[name = string("op_22740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2381_cast_fp16, y = var_22740_to_fp16)[name = string("aw_chunk_2381_cast_fp16")]; + fp16 var_22742_to_fp16 = const()[name = string("op_22742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2383_cast_fp16, y = var_22742_to_fp16)[name = string("aw_chunk_2383_cast_fp16")]; + fp16 var_22744_to_fp16 = const()[name = string("op_22744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2385_cast_fp16, y = var_22744_to_fp16)[name = string("aw_chunk_2385_cast_fp16")]; + fp16 var_22746_to_fp16 = const()[name = string("op_22746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2387_cast_fp16, y = var_22746_to_fp16)[name = string("aw_chunk_2387_cast_fp16")]; + fp16 var_22748_to_fp16 = const()[name = string("op_22748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2389_cast_fp16, y = var_22748_to_fp16)[name = string("aw_chunk_2389_cast_fp16")]; + fp16 var_22750_to_fp16 = const()[name = string("op_22750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2391_cast_fp16, y = var_22750_to_fp16)[name = string("aw_chunk_2391_cast_fp16")]; + fp16 var_22752_to_fp16 = const()[name = string("op_22752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2393_cast_fp16, y = var_22752_to_fp16)[name = string("aw_chunk_2393_cast_fp16")]; + fp16 var_22754_to_fp16 = const()[name = string("op_22754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2395_cast_fp16, y = var_22754_to_fp16)[name = string("aw_chunk_2395_cast_fp16")]; + fp16 var_22756_to_fp16 = const()[name = string("op_22756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2397_cast_fp16, y = var_22756_to_fp16)[name = string("aw_chunk_2397_cast_fp16")]; + fp16 var_22758_to_fp16 = const()[name = string("op_22758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2399_cast_fp16, y = var_22758_to_fp16)[name = string("aw_chunk_2399_cast_fp16")]; + tensor var_22760_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2241_cast_fp16)[name = string("op_22760_cast_fp16")]; + tensor var_22761_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2243_cast_fp16)[name = string("op_22761_cast_fp16")]; + tensor var_22762_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2245_cast_fp16)[name = string("op_22762_cast_fp16")]; + tensor var_22763_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2247_cast_fp16)[name = string("op_22763_cast_fp16")]; + tensor var_22764_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2249_cast_fp16)[name = string("op_22764_cast_fp16")]; + tensor var_22765_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2251_cast_fp16)[name = string("op_22765_cast_fp16")]; + tensor var_22766_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2253_cast_fp16)[name = string("op_22766_cast_fp16")]; + tensor var_22767_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2255_cast_fp16)[name = string("op_22767_cast_fp16")]; + tensor var_22768_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2257_cast_fp16)[name = string("op_22768_cast_fp16")]; + tensor var_22769_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2259_cast_fp16)[name = string("op_22769_cast_fp16")]; + tensor var_22770_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2261_cast_fp16)[name = string("op_22770_cast_fp16")]; + tensor var_22771_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2263_cast_fp16)[name = string("op_22771_cast_fp16")]; + tensor var_22772_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2265_cast_fp16)[name = string("op_22772_cast_fp16")]; + tensor var_22773_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2267_cast_fp16)[name = string("op_22773_cast_fp16")]; + tensor var_22774_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2269_cast_fp16)[name = string("op_22774_cast_fp16")]; + tensor var_22775_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2271_cast_fp16)[name = string("op_22775_cast_fp16")]; + tensor var_22776_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2273_cast_fp16)[name = string("op_22776_cast_fp16")]; + tensor var_22777_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2275_cast_fp16)[name = string("op_22777_cast_fp16")]; + tensor var_22778_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2277_cast_fp16)[name = string("op_22778_cast_fp16")]; + tensor var_22779_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2279_cast_fp16)[name = string("op_22779_cast_fp16")]; + tensor var_22780_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2281_cast_fp16)[name = string("op_22780_cast_fp16")]; + tensor var_22781_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2283_cast_fp16)[name = string("op_22781_cast_fp16")]; + tensor var_22782_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2285_cast_fp16)[name = string("op_22782_cast_fp16")]; + tensor var_22783_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2287_cast_fp16)[name = string("op_22783_cast_fp16")]; + tensor var_22784_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2289_cast_fp16)[name = string("op_22784_cast_fp16")]; + tensor var_22785_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2291_cast_fp16)[name = string("op_22785_cast_fp16")]; + tensor var_22786_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2293_cast_fp16)[name = string("op_22786_cast_fp16")]; + tensor var_22787_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2295_cast_fp16)[name = string("op_22787_cast_fp16")]; + tensor var_22788_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2297_cast_fp16)[name = string("op_22788_cast_fp16")]; + tensor var_22789_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2299_cast_fp16)[name = string("op_22789_cast_fp16")]; + tensor var_22790_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2301_cast_fp16)[name = string("op_22790_cast_fp16")]; + tensor var_22791_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2303_cast_fp16)[name = string("op_22791_cast_fp16")]; + tensor var_22792_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2305_cast_fp16)[name = string("op_22792_cast_fp16")]; + tensor var_22793_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2307_cast_fp16)[name = string("op_22793_cast_fp16")]; + tensor var_22794_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2309_cast_fp16)[name = string("op_22794_cast_fp16")]; + tensor var_22795_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2311_cast_fp16)[name = string("op_22795_cast_fp16")]; + tensor var_22796_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2313_cast_fp16)[name = string("op_22796_cast_fp16")]; + tensor var_22797_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2315_cast_fp16)[name = string("op_22797_cast_fp16")]; + tensor var_22798_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2317_cast_fp16)[name = string("op_22798_cast_fp16")]; + tensor var_22799_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2319_cast_fp16)[name = string("op_22799_cast_fp16")]; + tensor var_22800_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2321_cast_fp16)[name = string("op_22800_cast_fp16")]; + tensor var_22801_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2323_cast_fp16)[name = string("op_22801_cast_fp16")]; + tensor var_22802_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2325_cast_fp16)[name = string("op_22802_cast_fp16")]; + tensor var_22803_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2327_cast_fp16)[name = string("op_22803_cast_fp16")]; + tensor var_22804_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2329_cast_fp16)[name = string("op_22804_cast_fp16")]; + tensor var_22805_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2331_cast_fp16)[name = string("op_22805_cast_fp16")]; + tensor var_22806_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2333_cast_fp16)[name = string("op_22806_cast_fp16")]; + tensor var_22807_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2335_cast_fp16)[name = string("op_22807_cast_fp16")]; + tensor var_22808_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2337_cast_fp16)[name = string("op_22808_cast_fp16")]; + tensor var_22809_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2339_cast_fp16)[name = string("op_22809_cast_fp16")]; + tensor var_22810_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2341_cast_fp16)[name = string("op_22810_cast_fp16")]; + tensor var_22811_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2343_cast_fp16)[name = string("op_22811_cast_fp16")]; + tensor var_22812_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2345_cast_fp16)[name = string("op_22812_cast_fp16")]; + tensor var_22813_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2347_cast_fp16)[name = string("op_22813_cast_fp16")]; + tensor var_22814_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2349_cast_fp16)[name = string("op_22814_cast_fp16")]; + tensor var_22815_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2351_cast_fp16)[name = string("op_22815_cast_fp16")]; + tensor var_22816_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2353_cast_fp16)[name = string("op_22816_cast_fp16")]; + tensor var_22817_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2355_cast_fp16)[name = string("op_22817_cast_fp16")]; + tensor var_22818_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2357_cast_fp16)[name = string("op_22818_cast_fp16")]; + tensor var_22819_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2359_cast_fp16)[name = string("op_22819_cast_fp16")]; + tensor var_22820_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2361_cast_fp16)[name = string("op_22820_cast_fp16")]; + tensor var_22821_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2363_cast_fp16)[name = string("op_22821_cast_fp16")]; + tensor var_22822_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2365_cast_fp16)[name = string("op_22822_cast_fp16")]; + tensor var_22823_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2367_cast_fp16)[name = string("op_22823_cast_fp16")]; + tensor var_22824_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2369_cast_fp16)[name = string("op_22824_cast_fp16")]; + tensor var_22825_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2371_cast_fp16)[name = string("op_22825_cast_fp16")]; + tensor var_22826_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2373_cast_fp16)[name = string("op_22826_cast_fp16")]; + tensor var_22827_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2375_cast_fp16)[name = string("op_22827_cast_fp16")]; + tensor var_22828_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2377_cast_fp16)[name = string("op_22828_cast_fp16")]; + tensor var_22829_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2379_cast_fp16)[name = string("op_22829_cast_fp16")]; + tensor var_22830_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2381_cast_fp16)[name = string("op_22830_cast_fp16")]; + tensor var_22831_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2383_cast_fp16)[name = string("op_22831_cast_fp16")]; + tensor var_22832_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2385_cast_fp16)[name = string("op_22832_cast_fp16")]; + tensor var_22833_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2387_cast_fp16)[name = string("op_22833_cast_fp16")]; + tensor var_22834_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2389_cast_fp16)[name = string("op_22834_cast_fp16")]; + tensor var_22835_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2391_cast_fp16)[name = string("op_22835_cast_fp16")]; + tensor var_22836_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2393_cast_fp16)[name = string("op_22836_cast_fp16")]; + tensor var_22837_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2395_cast_fp16)[name = string("op_22837_cast_fp16")]; + tensor var_22838_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2397_cast_fp16)[name = string("op_22838_cast_fp16")]; + tensor var_22839_cast_fp16 = softmax(axis = var_21585, x = aw_chunk_2399_cast_fp16)[name = string("op_22839_cast_fp16")]; + string var_22841_equation_0 = const()[name = string("op_22841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22841_cast_fp16 = einsum(equation = var_22841_equation_0, values = (var_22361_cast_fp16, var_22760_cast_fp16))[name = string("op_22841_cast_fp16")]; + string var_22843_equation_0 = const()[name = string("op_22843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22843_cast_fp16 = einsum(equation = var_22843_equation_0, values = (var_22361_cast_fp16, var_22761_cast_fp16))[name = string("op_22843_cast_fp16")]; + string var_22845_equation_0 = const()[name = string("op_22845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22845_cast_fp16 = einsum(equation = var_22845_equation_0, values = (var_22361_cast_fp16, var_22762_cast_fp16))[name = string("op_22845_cast_fp16")]; + string var_22847_equation_0 = const()[name = string("op_22847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22847_cast_fp16 = einsum(equation = var_22847_equation_0, values = (var_22361_cast_fp16, var_22763_cast_fp16))[name = string("op_22847_cast_fp16")]; + string var_22849_equation_0 = const()[name = string("op_22849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22849_cast_fp16 = einsum(equation = var_22849_equation_0, values = (var_22365_cast_fp16, var_22764_cast_fp16))[name = string("op_22849_cast_fp16")]; + string var_22851_equation_0 = const()[name = string("op_22851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22851_cast_fp16 = einsum(equation = var_22851_equation_0, values = (var_22365_cast_fp16, var_22765_cast_fp16))[name = string("op_22851_cast_fp16")]; + string var_22853_equation_0 = const()[name = string("op_22853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22853_cast_fp16 = einsum(equation = var_22853_equation_0, values = (var_22365_cast_fp16, var_22766_cast_fp16))[name = string("op_22853_cast_fp16")]; + string var_22855_equation_0 = const()[name = string("op_22855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22855_cast_fp16 = einsum(equation = var_22855_equation_0, values = (var_22365_cast_fp16, var_22767_cast_fp16))[name = string("op_22855_cast_fp16")]; + string var_22857_equation_0 = const()[name = string("op_22857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22857_cast_fp16 = einsum(equation = var_22857_equation_0, values = (var_22369_cast_fp16, var_22768_cast_fp16))[name = string("op_22857_cast_fp16")]; + string var_22859_equation_0 = const()[name = string("op_22859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22859_cast_fp16 = einsum(equation = var_22859_equation_0, values = (var_22369_cast_fp16, var_22769_cast_fp16))[name = string("op_22859_cast_fp16")]; + string var_22861_equation_0 = const()[name = string("op_22861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22861_cast_fp16 = einsum(equation = var_22861_equation_0, values = (var_22369_cast_fp16, var_22770_cast_fp16))[name = string("op_22861_cast_fp16")]; + string var_22863_equation_0 = const()[name = string("op_22863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22863_cast_fp16 = einsum(equation = var_22863_equation_0, values = (var_22369_cast_fp16, var_22771_cast_fp16))[name = string("op_22863_cast_fp16")]; + string var_22865_equation_0 = const()[name = string("op_22865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22865_cast_fp16 = einsum(equation = var_22865_equation_0, values = (var_22373_cast_fp16, var_22772_cast_fp16))[name = string("op_22865_cast_fp16")]; + string var_22867_equation_0 = const()[name = string("op_22867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22867_cast_fp16 = einsum(equation = var_22867_equation_0, values = (var_22373_cast_fp16, var_22773_cast_fp16))[name = string("op_22867_cast_fp16")]; + string var_22869_equation_0 = const()[name = string("op_22869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22869_cast_fp16 = einsum(equation = var_22869_equation_0, values = (var_22373_cast_fp16, var_22774_cast_fp16))[name = string("op_22869_cast_fp16")]; + string var_22871_equation_0 = const()[name = string("op_22871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22871_cast_fp16 = einsum(equation = var_22871_equation_0, values = (var_22373_cast_fp16, var_22775_cast_fp16))[name = string("op_22871_cast_fp16")]; + string var_22873_equation_0 = const()[name = string("op_22873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22873_cast_fp16 = einsum(equation = var_22873_equation_0, values = (var_22377_cast_fp16, var_22776_cast_fp16))[name = string("op_22873_cast_fp16")]; + string var_22875_equation_0 = const()[name = string("op_22875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22875_cast_fp16 = einsum(equation = var_22875_equation_0, values = (var_22377_cast_fp16, var_22777_cast_fp16))[name = string("op_22875_cast_fp16")]; + string var_22877_equation_0 = const()[name = string("op_22877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22877_cast_fp16 = einsum(equation = var_22877_equation_0, values = (var_22377_cast_fp16, var_22778_cast_fp16))[name = string("op_22877_cast_fp16")]; + string var_22879_equation_0 = const()[name = string("op_22879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22879_cast_fp16 = einsum(equation = var_22879_equation_0, values = (var_22377_cast_fp16, var_22779_cast_fp16))[name = string("op_22879_cast_fp16")]; + string var_22881_equation_0 = const()[name = string("op_22881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22881_cast_fp16 = einsum(equation = var_22881_equation_0, values = (var_22381_cast_fp16, var_22780_cast_fp16))[name = string("op_22881_cast_fp16")]; + string var_22883_equation_0 = const()[name = string("op_22883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22883_cast_fp16 = einsum(equation = var_22883_equation_0, values = (var_22381_cast_fp16, var_22781_cast_fp16))[name = string("op_22883_cast_fp16")]; + string var_22885_equation_0 = const()[name = string("op_22885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22885_cast_fp16 = einsum(equation = var_22885_equation_0, values = (var_22381_cast_fp16, var_22782_cast_fp16))[name = string("op_22885_cast_fp16")]; + string var_22887_equation_0 = const()[name = string("op_22887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22887_cast_fp16 = einsum(equation = var_22887_equation_0, values = (var_22381_cast_fp16, var_22783_cast_fp16))[name = string("op_22887_cast_fp16")]; + string var_22889_equation_0 = const()[name = string("op_22889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22889_cast_fp16 = einsum(equation = var_22889_equation_0, values = (var_22385_cast_fp16, var_22784_cast_fp16))[name = string("op_22889_cast_fp16")]; + string var_22891_equation_0 = const()[name = string("op_22891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22891_cast_fp16 = einsum(equation = var_22891_equation_0, values = (var_22385_cast_fp16, var_22785_cast_fp16))[name = string("op_22891_cast_fp16")]; + string var_22893_equation_0 = const()[name = string("op_22893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22893_cast_fp16 = einsum(equation = var_22893_equation_0, values = (var_22385_cast_fp16, var_22786_cast_fp16))[name = string("op_22893_cast_fp16")]; + string var_22895_equation_0 = const()[name = string("op_22895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22895_cast_fp16 = einsum(equation = var_22895_equation_0, values = (var_22385_cast_fp16, var_22787_cast_fp16))[name = string("op_22895_cast_fp16")]; + string var_22897_equation_0 = const()[name = string("op_22897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22897_cast_fp16 = einsum(equation = var_22897_equation_0, values = (var_22389_cast_fp16, var_22788_cast_fp16))[name = string("op_22897_cast_fp16")]; + string var_22899_equation_0 = const()[name = string("op_22899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22899_cast_fp16 = einsum(equation = var_22899_equation_0, values = (var_22389_cast_fp16, var_22789_cast_fp16))[name = string("op_22899_cast_fp16")]; + string var_22901_equation_0 = const()[name = string("op_22901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22901_cast_fp16 = einsum(equation = var_22901_equation_0, values = (var_22389_cast_fp16, var_22790_cast_fp16))[name = string("op_22901_cast_fp16")]; + string var_22903_equation_0 = const()[name = string("op_22903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22903_cast_fp16 = einsum(equation = var_22903_equation_0, values = (var_22389_cast_fp16, var_22791_cast_fp16))[name = string("op_22903_cast_fp16")]; + string var_22905_equation_0 = const()[name = string("op_22905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22905_cast_fp16 = einsum(equation = var_22905_equation_0, values = (var_22393_cast_fp16, var_22792_cast_fp16))[name = string("op_22905_cast_fp16")]; + string var_22907_equation_0 = const()[name = string("op_22907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22907_cast_fp16 = einsum(equation = var_22907_equation_0, values = (var_22393_cast_fp16, var_22793_cast_fp16))[name = string("op_22907_cast_fp16")]; + string var_22909_equation_0 = const()[name = string("op_22909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22909_cast_fp16 = einsum(equation = var_22909_equation_0, values = (var_22393_cast_fp16, var_22794_cast_fp16))[name = string("op_22909_cast_fp16")]; + string var_22911_equation_0 = const()[name = string("op_22911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22911_cast_fp16 = einsum(equation = var_22911_equation_0, values = (var_22393_cast_fp16, var_22795_cast_fp16))[name = string("op_22911_cast_fp16")]; + string var_22913_equation_0 = const()[name = string("op_22913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22913_cast_fp16 = einsum(equation = var_22913_equation_0, values = (var_22397_cast_fp16, var_22796_cast_fp16))[name = string("op_22913_cast_fp16")]; + string var_22915_equation_0 = const()[name = string("op_22915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22915_cast_fp16 = einsum(equation = var_22915_equation_0, values = (var_22397_cast_fp16, var_22797_cast_fp16))[name = string("op_22915_cast_fp16")]; + string var_22917_equation_0 = const()[name = string("op_22917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22917_cast_fp16 = einsum(equation = var_22917_equation_0, values = (var_22397_cast_fp16, var_22798_cast_fp16))[name = string("op_22917_cast_fp16")]; + string var_22919_equation_0 = const()[name = string("op_22919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22919_cast_fp16 = einsum(equation = var_22919_equation_0, values = (var_22397_cast_fp16, var_22799_cast_fp16))[name = string("op_22919_cast_fp16")]; + string var_22921_equation_0 = const()[name = string("op_22921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22921_cast_fp16 = einsum(equation = var_22921_equation_0, values = (var_22401_cast_fp16, var_22800_cast_fp16))[name = string("op_22921_cast_fp16")]; + string var_22923_equation_0 = const()[name = string("op_22923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22923_cast_fp16 = einsum(equation = var_22923_equation_0, values = (var_22401_cast_fp16, var_22801_cast_fp16))[name = string("op_22923_cast_fp16")]; + string var_22925_equation_0 = const()[name = string("op_22925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22925_cast_fp16 = einsum(equation = var_22925_equation_0, values = (var_22401_cast_fp16, var_22802_cast_fp16))[name = string("op_22925_cast_fp16")]; + string var_22927_equation_0 = const()[name = string("op_22927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22927_cast_fp16 = einsum(equation = var_22927_equation_0, values = (var_22401_cast_fp16, var_22803_cast_fp16))[name = string("op_22927_cast_fp16")]; + string var_22929_equation_0 = const()[name = string("op_22929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22929_cast_fp16 = einsum(equation = var_22929_equation_0, values = (var_22405_cast_fp16, var_22804_cast_fp16))[name = string("op_22929_cast_fp16")]; + string var_22931_equation_0 = const()[name = string("op_22931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22931_cast_fp16 = einsum(equation = var_22931_equation_0, values = (var_22405_cast_fp16, var_22805_cast_fp16))[name = string("op_22931_cast_fp16")]; + string var_22933_equation_0 = const()[name = string("op_22933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22933_cast_fp16 = einsum(equation = var_22933_equation_0, values = (var_22405_cast_fp16, var_22806_cast_fp16))[name = string("op_22933_cast_fp16")]; + string var_22935_equation_0 = const()[name = string("op_22935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22935_cast_fp16 = einsum(equation = var_22935_equation_0, values = (var_22405_cast_fp16, var_22807_cast_fp16))[name = string("op_22935_cast_fp16")]; + string var_22937_equation_0 = const()[name = string("op_22937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22937_cast_fp16 = einsum(equation = var_22937_equation_0, values = (var_22409_cast_fp16, var_22808_cast_fp16))[name = string("op_22937_cast_fp16")]; + string var_22939_equation_0 = const()[name = string("op_22939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22939_cast_fp16 = einsum(equation = var_22939_equation_0, values = (var_22409_cast_fp16, var_22809_cast_fp16))[name = string("op_22939_cast_fp16")]; + string var_22941_equation_0 = const()[name = string("op_22941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22941_cast_fp16 = einsum(equation = var_22941_equation_0, values = (var_22409_cast_fp16, var_22810_cast_fp16))[name = string("op_22941_cast_fp16")]; + string var_22943_equation_0 = const()[name = string("op_22943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22943_cast_fp16 = einsum(equation = var_22943_equation_0, values = (var_22409_cast_fp16, var_22811_cast_fp16))[name = string("op_22943_cast_fp16")]; + string var_22945_equation_0 = const()[name = string("op_22945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22945_cast_fp16 = einsum(equation = var_22945_equation_0, values = (var_22413_cast_fp16, var_22812_cast_fp16))[name = string("op_22945_cast_fp16")]; + string var_22947_equation_0 = const()[name = string("op_22947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22947_cast_fp16 = einsum(equation = var_22947_equation_0, values = (var_22413_cast_fp16, var_22813_cast_fp16))[name = string("op_22947_cast_fp16")]; + string var_22949_equation_0 = const()[name = string("op_22949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22949_cast_fp16 = einsum(equation = var_22949_equation_0, values = (var_22413_cast_fp16, var_22814_cast_fp16))[name = string("op_22949_cast_fp16")]; + string var_22951_equation_0 = const()[name = string("op_22951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22951_cast_fp16 = einsum(equation = var_22951_equation_0, values = (var_22413_cast_fp16, var_22815_cast_fp16))[name = string("op_22951_cast_fp16")]; + string var_22953_equation_0 = const()[name = string("op_22953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22953_cast_fp16 = einsum(equation = var_22953_equation_0, values = (var_22417_cast_fp16, var_22816_cast_fp16))[name = string("op_22953_cast_fp16")]; + string var_22955_equation_0 = const()[name = string("op_22955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22955_cast_fp16 = einsum(equation = var_22955_equation_0, values = (var_22417_cast_fp16, var_22817_cast_fp16))[name = string("op_22955_cast_fp16")]; + string var_22957_equation_0 = const()[name = string("op_22957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22957_cast_fp16 = einsum(equation = var_22957_equation_0, values = (var_22417_cast_fp16, var_22818_cast_fp16))[name = string("op_22957_cast_fp16")]; + string var_22959_equation_0 = const()[name = string("op_22959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22959_cast_fp16 = einsum(equation = var_22959_equation_0, values = (var_22417_cast_fp16, var_22819_cast_fp16))[name = string("op_22959_cast_fp16")]; + string var_22961_equation_0 = const()[name = string("op_22961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22961_cast_fp16 = einsum(equation = var_22961_equation_0, values = (var_22421_cast_fp16, var_22820_cast_fp16))[name = string("op_22961_cast_fp16")]; + string var_22963_equation_0 = const()[name = string("op_22963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22963_cast_fp16 = einsum(equation = var_22963_equation_0, values = (var_22421_cast_fp16, var_22821_cast_fp16))[name = string("op_22963_cast_fp16")]; + string var_22965_equation_0 = const()[name = string("op_22965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22965_cast_fp16 = einsum(equation = var_22965_equation_0, values = (var_22421_cast_fp16, var_22822_cast_fp16))[name = string("op_22965_cast_fp16")]; + string var_22967_equation_0 = const()[name = string("op_22967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22967_cast_fp16 = einsum(equation = var_22967_equation_0, values = (var_22421_cast_fp16, var_22823_cast_fp16))[name = string("op_22967_cast_fp16")]; + string var_22969_equation_0 = const()[name = string("op_22969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22969_cast_fp16 = einsum(equation = var_22969_equation_0, values = (var_22425_cast_fp16, var_22824_cast_fp16))[name = string("op_22969_cast_fp16")]; + string var_22971_equation_0 = const()[name = string("op_22971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22971_cast_fp16 = einsum(equation = var_22971_equation_0, values = (var_22425_cast_fp16, var_22825_cast_fp16))[name = string("op_22971_cast_fp16")]; + string var_22973_equation_0 = const()[name = string("op_22973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22973_cast_fp16 = einsum(equation = var_22973_equation_0, values = (var_22425_cast_fp16, var_22826_cast_fp16))[name = string("op_22973_cast_fp16")]; + string var_22975_equation_0 = const()[name = string("op_22975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22975_cast_fp16 = einsum(equation = var_22975_equation_0, values = (var_22425_cast_fp16, var_22827_cast_fp16))[name = string("op_22975_cast_fp16")]; + string var_22977_equation_0 = const()[name = string("op_22977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22977_cast_fp16 = einsum(equation = var_22977_equation_0, values = (var_22429_cast_fp16, var_22828_cast_fp16))[name = string("op_22977_cast_fp16")]; + string var_22979_equation_0 = const()[name = string("op_22979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22979_cast_fp16 = einsum(equation = var_22979_equation_0, values = (var_22429_cast_fp16, var_22829_cast_fp16))[name = string("op_22979_cast_fp16")]; + string var_22981_equation_0 = const()[name = string("op_22981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22981_cast_fp16 = einsum(equation = var_22981_equation_0, values = (var_22429_cast_fp16, var_22830_cast_fp16))[name = string("op_22981_cast_fp16")]; + string var_22983_equation_0 = const()[name = string("op_22983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22983_cast_fp16 = einsum(equation = var_22983_equation_0, values = (var_22429_cast_fp16, var_22831_cast_fp16))[name = string("op_22983_cast_fp16")]; + string var_22985_equation_0 = const()[name = string("op_22985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22985_cast_fp16 = einsum(equation = var_22985_equation_0, values = (var_22433_cast_fp16, var_22832_cast_fp16))[name = string("op_22985_cast_fp16")]; + string var_22987_equation_0 = const()[name = string("op_22987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22987_cast_fp16 = einsum(equation = var_22987_equation_0, values = (var_22433_cast_fp16, var_22833_cast_fp16))[name = string("op_22987_cast_fp16")]; + string var_22989_equation_0 = const()[name = string("op_22989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22989_cast_fp16 = einsum(equation = var_22989_equation_0, values = (var_22433_cast_fp16, var_22834_cast_fp16))[name = string("op_22989_cast_fp16")]; + string var_22991_equation_0 = const()[name = string("op_22991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22991_cast_fp16 = einsum(equation = var_22991_equation_0, values = (var_22433_cast_fp16, var_22835_cast_fp16))[name = string("op_22991_cast_fp16")]; + string var_22993_equation_0 = const()[name = string("op_22993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22993_cast_fp16 = einsum(equation = var_22993_equation_0, values = (var_22437_cast_fp16, var_22836_cast_fp16))[name = string("op_22993_cast_fp16")]; + string var_22995_equation_0 = const()[name = string("op_22995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22995_cast_fp16 = einsum(equation = var_22995_equation_0, values = (var_22437_cast_fp16, var_22837_cast_fp16))[name = string("op_22995_cast_fp16")]; + string var_22997_equation_0 = const()[name = string("op_22997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22997_cast_fp16 = einsum(equation = var_22997_equation_0, values = (var_22437_cast_fp16, var_22838_cast_fp16))[name = string("op_22997_cast_fp16")]; + string var_22999_equation_0 = const()[name = string("op_22999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_22999_cast_fp16 = einsum(equation = var_22999_equation_0, values = (var_22437_cast_fp16, var_22839_cast_fp16))[name = string("op_22999_cast_fp16")]; + bool var_23001_interleave_0 = const()[name = string("op_23001_interleave_0"), val = bool(false)]; + tensor var_23001_cast_fp16 = concat(axis = var_21560, interleave = var_23001_interleave_0, values = (var_22841_cast_fp16, var_22843_cast_fp16, var_22845_cast_fp16, var_22847_cast_fp16))[name = string("op_23001_cast_fp16")]; + bool var_23003_interleave_0 = const()[name = string("op_23003_interleave_0"), val = bool(false)]; + tensor var_23003_cast_fp16 = concat(axis = var_21560, interleave = var_23003_interleave_0, values = (var_22849_cast_fp16, var_22851_cast_fp16, var_22853_cast_fp16, var_22855_cast_fp16))[name = string("op_23003_cast_fp16")]; + bool var_23005_interleave_0 = const()[name = string("op_23005_interleave_0"), val = bool(false)]; + tensor var_23005_cast_fp16 = concat(axis = var_21560, interleave = var_23005_interleave_0, values = (var_22857_cast_fp16, var_22859_cast_fp16, var_22861_cast_fp16, var_22863_cast_fp16))[name = string("op_23005_cast_fp16")]; + bool var_23007_interleave_0 = const()[name = string("op_23007_interleave_0"), val = bool(false)]; + tensor var_23007_cast_fp16 = concat(axis = var_21560, interleave = var_23007_interleave_0, values = (var_22865_cast_fp16, var_22867_cast_fp16, var_22869_cast_fp16, var_22871_cast_fp16))[name = string("op_23007_cast_fp16")]; + bool var_23009_interleave_0 = const()[name = string("op_23009_interleave_0"), val = bool(false)]; + tensor var_23009_cast_fp16 = concat(axis = var_21560, interleave = var_23009_interleave_0, values = (var_22873_cast_fp16, var_22875_cast_fp16, var_22877_cast_fp16, var_22879_cast_fp16))[name = string("op_23009_cast_fp16")]; + bool var_23011_interleave_0 = const()[name = string("op_23011_interleave_0"), val = bool(false)]; + tensor var_23011_cast_fp16 = concat(axis = var_21560, interleave = var_23011_interleave_0, values = (var_22881_cast_fp16, var_22883_cast_fp16, var_22885_cast_fp16, var_22887_cast_fp16))[name = string("op_23011_cast_fp16")]; + bool var_23013_interleave_0 = const()[name = string("op_23013_interleave_0"), val = bool(false)]; + tensor var_23013_cast_fp16 = concat(axis = var_21560, interleave = var_23013_interleave_0, values = (var_22889_cast_fp16, var_22891_cast_fp16, var_22893_cast_fp16, var_22895_cast_fp16))[name = string("op_23013_cast_fp16")]; + bool var_23015_interleave_0 = const()[name = string("op_23015_interleave_0"), val = bool(false)]; + tensor var_23015_cast_fp16 = concat(axis = var_21560, interleave = var_23015_interleave_0, values = (var_22897_cast_fp16, var_22899_cast_fp16, var_22901_cast_fp16, var_22903_cast_fp16))[name = string("op_23015_cast_fp16")]; + bool var_23017_interleave_0 = const()[name = string("op_23017_interleave_0"), val = bool(false)]; + tensor var_23017_cast_fp16 = concat(axis = var_21560, interleave = var_23017_interleave_0, values = (var_22905_cast_fp16, var_22907_cast_fp16, var_22909_cast_fp16, var_22911_cast_fp16))[name = string("op_23017_cast_fp16")]; + bool var_23019_interleave_0 = const()[name = string("op_23019_interleave_0"), val = bool(false)]; + tensor var_23019_cast_fp16 = concat(axis = var_21560, interleave = var_23019_interleave_0, values = (var_22913_cast_fp16, var_22915_cast_fp16, var_22917_cast_fp16, var_22919_cast_fp16))[name = string("op_23019_cast_fp16")]; + bool var_23021_interleave_0 = const()[name = string("op_23021_interleave_0"), val = bool(false)]; + tensor var_23021_cast_fp16 = concat(axis = var_21560, interleave = var_23021_interleave_0, values = (var_22921_cast_fp16, var_22923_cast_fp16, var_22925_cast_fp16, var_22927_cast_fp16))[name = string("op_23021_cast_fp16")]; + bool var_23023_interleave_0 = const()[name = string("op_23023_interleave_0"), val = bool(false)]; + tensor var_23023_cast_fp16 = concat(axis = var_21560, interleave = var_23023_interleave_0, values = (var_22929_cast_fp16, var_22931_cast_fp16, var_22933_cast_fp16, var_22935_cast_fp16))[name = string("op_23023_cast_fp16")]; + bool var_23025_interleave_0 = const()[name = string("op_23025_interleave_0"), val = bool(false)]; + tensor var_23025_cast_fp16 = concat(axis = var_21560, interleave = var_23025_interleave_0, values = (var_22937_cast_fp16, var_22939_cast_fp16, var_22941_cast_fp16, var_22943_cast_fp16))[name = string("op_23025_cast_fp16")]; + bool var_23027_interleave_0 = const()[name = string("op_23027_interleave_0"), val = bool(false)]; + tensor var_23027_cast_fp16 = concat(axis = var_21560, interleave = var_23027_interleave_0, values = (var_22945_cast_fp16, var_22947_cast_fp16, var_22949_cast_fp16, var_22951_cast_fp16))[name = string("op_23027_cast_fp16")]; + bool var_23029_interleave_0 = const()[name = string("op_23029_interleave_0"), val = bool(false)]; + tensor var_23029_cast_fp16 = concat(axis = var_21560, interleave = var_23029_interleave_0, values = (var_22953_cast_fp16, var_22955_cast_fp16, var_22957_cast_fp16, var_22959_cast_fp16))[name = string("op_23029_cast_fp16")]; + bool var_23031_interleave_0 = const()[name = string("op_23031_interleave_0"), val = bool(false)]; + tensor var_23031_cast_fp16 = concat(axis = var_21560, interleave = var_23031_interleave_0, values = (var_22961_cast_fp16, var_22963_cast_fp16, var_22965_cast_fp16, var_22967_cast_fp16))[name = string("op_23031_cast_fp16")]; + bool var_23033_interleave_0 = const()[name = string("op_23033_interleave_0"), val = bool(false)]; + tensor var_23033_cast_fp16 = concat(axis = var_21560, interleave = var_23033_interleave_0, values = (var_22969_cast_fp16, var_22971_cast_fp16, var_22973_cast_fp16, var_22975_cast_fp16))[name = string("op_23033_cast_fp16")]; + bool var_23035_interleave_0 = const()[name = string("op_23035_interleave_0"), val = bool(false)]; + tensor var_23035_cast_fp16 = concat(axis = var_21560, interleave = var_23035_interleave_0, values = (var_22977_cast_fp16, var_22979_cast_fp16, var_22981_cast_fp16, var_22983_cast_fp16))[name = string("op_23035_cast_fp16")]; + bool var_23037_interleave_0 = const()[name = string("op_23037_interleave_0"), val = bool(false)]; + tensor var_23037_cast_fp16 = concat(axis = var_21560, interleave = var_23037_interleave_0, values = (var_22985_cast_fp16, var_22987_cast_fp16, var_22989_cast_fp16, var_22991_cast_fp16))[name = string("op_23037_cast_fp16")]; + bool var_23039_interleave_0 = const()[name = string("op_23039_interleave_0"), val = bool(false)]; + tensor var_23039_cast_fp16 = concat(axis = var_21560, interleave = var_23039_interleave_0, values = (var_22993_cast_fp16, var_22995_cast_fp16, var_22997_cast_fp16, var_22999_cast_fp16))[name = string("op_23039_cast_fp16")]; + bool input_113_interleave_0 = const()[name = string("input_113_interleave_0"), val = bool(false)]; + tensor input_113_cast_fp16 = concat(axis = var_21585, interleave = input_113_interleave_0, values = (var_23001_cast_fp16, var_23003_cast_fp16, var_23005_cast_fp16, var_23007_cast_fp16, var_23009_cast_fp16, var_23011_cast_fp16, var_23013_cast_fp16, var_23015_cast_fp16, var_23017_cast_fp16, var_23019_cast_fp16, var_23021_cast_fp16, var_23023_cast_fp16, var_23025_cast_fp16, var_23027_cast_fp16, var_23029_cast_fp16, var_23031_cast_fp16, var_23033_cast_fp16, var_23035_cast_fp16, var_23037_cast_fp16, var_23039_cast_fp16))[name = string("input_113_cast_fp16")]; + string obj_59_pad_type_0 = const()[name = string("obj_59_pad_type_0"), val = string("valid")]; + tensor obj_59_strides_0 = const()[name = string("obj_59_strides_0"), val = tensor([1, 1])]; + tensor obj_59_pad_0 = const()[name = string("obj_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_59_dilations_0 = const()[name = string("obj_59_dilations_0"), val = tensor([1, 1])]; + int32 obj_59_groups_0 = const()[name = string("obj_59_groups_0"), val = int32(1)]; + tensor layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(575451200)))]; + tensor layers_14_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578728064)))]; + tensor obj_59_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_bias_to_fp16, dilations = obj_59_dilations_0, groups = obj_59_groups_0, pad = obj_59_pad_0, pad_type = obj_59_pad_type_0, strides = obj_59_strides_0, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_59_cast_fp16)[name = string("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; + fp16 var_23058_to_fp16 = const()[name = string("op_23058_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_23058_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578730688)))]; + tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578733312)))]; + fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor layers_14_fc1_weight_to_fp16 = const()[name = string("layers_14_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(578735936)))]; + tensor layers_14_fc1_bias_to_fp16 = const()[name = string("layers_14_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591843200)))]; + tensor input_117_cast_fp16 = conv(bias = layers_14_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_14_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")]; + tensor input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")]; + string hidden_states_33_pad_type_0 = const()[name = string("hidden_states_33_pad_type_0"), val = string("valid")]; + tensor hidden_states_33_strides_0 = const()[name = string("hidden_states_33_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_33_pad_0 = const()[name = string("hidden_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_33_dilations_0 = const()[name = string("hidden_states_33_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_33_groups_0 = const()[name = string("hidden_states_33_groups_0"), val = int32(1)]; + tensor layers_14_fc2_weight_to_fp16 = const()[name = string("layers_14_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591853504)))]; + tensor layers_14_fc2_bias_to_fp16 = const()[name = string("layers_14_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604960768)))]; + tensor hidden_states_33_cast_fp16 = conv(bias = layers_14_fc2_bias_to_fp16, dilations = hidden_states_33_dilations_0, groups = hidden_states_33_groups_0, pad = hidden_states_33_pad_0, pad_type = hidden_states_33_pad_type_0, strides = hidden_states_33_strides_0, weight = layers_14_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = string("hidden_states_33_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_33_cast_fp16)[name = string("inputs_61_cast_fp16")]; + int32 var_23087 = const()[name = string("op_23087"), val = int32(3)]; + int32 var_23112 = const()[name = string("op_23112"), val = int32(1)]; + tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; + fp16 var_23129_to_fp16 = const()[name = string("op_23129_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_23129_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604963392)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604966016)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; + tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; + tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; + int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(604968640)))]; + tensor layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608245504)))]; + tensor query_31_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_31_cast_fp16")]; + string key_31_pad_type_0 = const()[name = string("key_31_pad_type_0"), val = string("valid")]; + tensor key_31_strides_0 = const()[name = string("key_31_strides_0"), val = tensor([1, 1])]; + tensor key_31_pad_0 = const()[name = string("key_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_31_dilations_0 = const()[name = string("key_31_dilations_0"), val = tensor([1, 1])]; + int32 key_31_groups_0 = const()[name = string("key_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608248128)))]; + tensor key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("key_31_cast_fp16")]; + string value_31_pad_type_0 = const()[name = string("value_31_pad_type_0"), val = string("valid")]; + tensor value_31_strides_0 = const()[name = string("value_31_strides_0"), val = tensor([1, 1])]; + tensor value_31_pad_0 = const()[name = string("value_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_31_dilations_0 = const()[name = string("value_31_dilations_0"), val = tensor([1, 1])]; + int32 value_31_groups_0 = const()[name = string("value_31_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(611524992)))]; + tensor layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614801856)))]; + tensor value_31_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("value_31_cast_fp16")]; + tensor var_23167_begin_0 = const()[name = string("op_23167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23167_end_0 = const()[name = string("op_23167_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23167_end_mask_0 = const()[name = string("op_23167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23167_cast_fp16 = slice_by_index(begin = var_23167_begin_0, end = var_23167_end_0, end_mask = var_23167_end_mask_0, x = query_31_cast_fp16)[name = string("op_23167_cast_fp16")]; + tensor var_23171_begin_0 = const()[name = string("op_23171_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23171_end_0 = const()[name = string("op_23171_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_23171_end_mask_0 = const()[name = string("op_23171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23171_cast_fp16 = slice_by_index(begin = var_23171_begin_0, end = var_23171_end_0, end_mask = var_23171_end_mask_0, x = query_31_cast_fp16)[name = string("op_23171_cast_fp16")]; + tensor var_23175_begin_0 = const()[name = string("op_23175_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23175_end_0 = const()[name = string("op_23175_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_23175_end_mask_0 = const()[name = string("op_23175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23175_cast_fp16 = slice_by_index(begin = var_23175_begin_0, end = var_23175_end_0, end_mask = var_23175_end_mask_0, x = query_31_cast_fp16)[name = string("op_23175_cast_fp16")]; + tensor var_23179_begin_0 = const()[name = string("op_23179_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23179_end_0 = const()[name = string("op_23179_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_23179_end_mask_0 = const()[name = string("op_23179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23179_cast_fp16 = slice_by_index(begin = var_23179_begin_0, end = var_23179_end_0, end_mask = var_23179_end_mask_0, x = query_31_cast_fp16)[name = string("op_23179_cast_fp16")]; + tensor var_23183_begin_0 = const()[name = string("op_23183_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23183_end_0 = const()[name = string("op_23183_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_23183_end_mask_0 = const()[name = string("op_23183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23183_cast_fp16 = slice_by_index(begin = var_23183_begin_0, end = var_23183_end_0, end_mask = var_23183_end_mask_0, x = query_31_cast_fp16)[name = string("op_23183_cast_fp16")]; + tensor var_23187_begin_0 = const()[name = string("op_23187_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23187_end_0 = const()[name = string("op_23187_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_23187_end_mask_0 = const()[name = string("op_23187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23187_cast_fp16 = slice_by_index(begin = var_23187_begin_0, end = var_23187_end_0, end_mask = var_23187_end_mask_0, x = query_31_cast_fp16)[name = string("op_23187_cast_fp16")]; + tensor var_23191_begin_0 = const()[name = string("op_23191_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23191_end_0 = const()[name = string("op_23191_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_23191_end_mask_0 = const()[name = string("op_23191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23191_cast_fp16 = slice_by_index(begin = var_23191_begin_0, end = var_23191_end_0, end_mask = var_23191_end_mask_0, x = query_31_cast_fp16)[name = string("op_23191_cast_fp16")]; + tensor var_23195_begin_0 = const()[name = string("op_23195_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23195_end_0 = const()[name = string("op_23195_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_23195_end_mask_0 = const()[name = string("op_23195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23195_cast_fp16 = slice_by_index(begin = var_23195_begin_0, end = var_23195_end_0, end_mask = var_23195_end_mask_0, x = query_31_cast_fp16)[name = string("op_23195_cast_fp16")]; + tensor var_23199_begin_0 = const()[name = string("op_23199_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23199_end_0 = const()[name = string("op_23199_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_23199_end_mask_0 = const()[name = string("op_23199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23199_cast_fp16 = slice_by_index(begin = var_23199_begin_0, end = var_23199_end_0, end_mask = var_23199_end_mask_0, x = query_31_cast_fp16)[name = string("op_23199_cast_fp16")]; + tensor var_23203_begin_0 = const()[name = string("op_23203_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23203_end_0 = const()[name = string("op_23203_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_23203_end_mask_0 = const()[name = string("op_23203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23203_cast_fp16 = slice_by_index(begin = var_23203_begin_0, end = var_23203_end_0, end_mask = var_23203_end_mask_0, x = query_31_cast_fp16)[name = string("op_23203_cast_fp16")]; + tensor var_23207_begin_0 = const()[name = string("op_23207_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23207_end_0 = const()[name = string("op_23207_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_23207_end_mask_0 = const()[name = string("op_23207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23207_cast_fp16 = slice_by_index(begin = var_23207_begin_0, end = var_23207_end_0, end_mask = var_23207_end_mask_0, x = query_31_cast_fp16)[name = string("op_23207_cast_fp16")]; + tensor var_23211_begin_0 = const()[name = string("op_23211_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23211_end_0 = const()[name = string("op_23211_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_23211_end_mask_0 = const()[name = string("op_23211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23211_cast_fp16 = slice_by_index(begin = var_23211_begin_0, end = var_23211_end_0, end_mask = var_23211_end_mask_0, x = query_31_cast_fp16)[name = string("op_23211_cast_fp16")]; + tensor var_23215_begin_0 = const()[name = string("op_23215_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23215_end_0 = const()[name = string("op_23215_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_23215_end_mask_0 = const()[name = string("op_23215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23215_cast_fp16 = slice_by_index(begin = var_23215_begin_0, end = var_23215_end_0, end_mask = var_23215_end_mask_0, x = query_31_cast_fp16)[name = string("op_23215_cast_fp16")]; + tensor var_23219_begin_0 = const()[name = string("op_23219_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23219_end_0 = const()[name = string("op_23219_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_23219_end_mask_0 = const()[name = string("op_23219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23219_cast_fp16 = slice_by_index(begin = var_23219_begin_0, end = var_23219_end_0, end_mask = var_23219_end_mask_0, x = query_31_cast_fp16)[name = string("op_23219_cast_fp16")]; + tensor var_23223_begin_0 = const()[name = string("op_23223_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23223_end_0 = const()[name = string("op_23223_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_23223_end_mask_0 = const()[name = string("op_23223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23223_cast_fp16 = slice_by_index(begin = var_23223_begin_0, end = var_23223_end_0, end_mask = var_23223_end_mask_0, x = query_31_cast_fp16)[name = string("op_23223_cast_fp16")]; + tensor var_23227_begin_0 = const()[name = string("op_23227_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23227_end_0 = const()[name = string("op_23227_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_23227_end_mask_0 = const()[name = string("op_23227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23227_cast_fp16 = slice_by_index(begin = var_23227_begin_0, end = var_23227_end_0, end_mask = var_23227_end_mask_0, x = query_31_cast_fp16)[name = string("op_23227_cast_fp16")]; + tensor var_23231_begin_0 = const()[name = string("op_23231_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23231_end_0 = const()[name = string("op_23231_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_23231_end_mask_0 = const()[name = string("op_23231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23231_cast_fp16 = slice_by_index(begin = var_23231_begin_0, end = var_23231_end_0, end_mask = var_23231_end_mask_0, x = query_31_cast_fp16)[name = string("op_23231_cast_fp16")]; + tensor var_23235_begin_0 = const()[name = string("op_23235_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23235_end_0 = const()[name = string("op_23235_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_23235_end_mask_0 = const()[name = string("op_23235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23235_cast_fp16 = slice_by_index(begin = var_23235_begin_0, end = var_23235_end_0, end_mask = var_23235_end_mask_0, x = query_31_cast_fp16)[name = string("op_23235_cast_fp16")]; + tensor var_23239_begin_0 = const()[name = string("op_23239_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23239_end_0 = const()[name = string("op_23239_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_23239_end_mask_0 = const()[name = string("op_23239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23239_cast_fp16 = slice_by_index(begin = var_23239_begin_0, end = var_23239_end_0, end_mask = var_23239_end_mask_0, x = query_31_cast_fp16)[name = string("op_23239_cast_fp16")]; + tensor var_23243_begin_0 = const()[name = string("op_23243_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23243_end_0 = const()[name = string("op_23243_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_23243_end_mask_0 = const()[name = string("op_23243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23243_cast_fp16 = slice_by_index(begin = var_23243_begin_0, end = var_23243_end_0, end_mask = var_23243_end_mask_0, x = query_31_cast_fp16)[name = string("op_23243_cast_fp16")]; + tensor var_23252_begin_0 = const()[name = string("op_23252_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23252_end_0 = const()[name = string("op_23252_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23252_end_mask_0 = const()[name = string("op_23252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23252_cast_fp16 = slice_by_index(begin = var_23252_begin_0, end = var_23252_end_0, end_mask = var_23252_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23252_cast_fp16")]; + tensor var_23259_begin_0 = const()[name = string("op_23259_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23259_end_0 = const()[name = string("op_23259_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23259_end_mask_0 = const()[name = string("op_23259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23259_cast_fp16 = slice_by_index(begin = var_23259_begin_0, end = var_23259_end_0, end_mask = var_23259_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23259_cast_fp16")]; + tensor var_23266_begin_0 = const()[name = string("op_23266_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23266_end_0 = const()[name = string("op_23266_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23266_end_mask_0 = const()[name = string("op_23266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23266_cast_fp16 = slice_by_index(begin = var_23266_begin_0, end = var_23266_end_0, end_mask = var_23266_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23266_cast_fp16")]; + tensor var_23273_begin_0 = const()[name = string("op_23273_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23273_end_0 = const()[name = string("op_23273_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23273_end_mask_0 = const()[name = string("op_23273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23273_cast_fp16 = slice_by_index(begin = var_23273_begin_0, end = var_23273_end_0, end_mask = var_23273_end_mask_0, x = var_23167_cast_fp16)[name = string("op_23273_cast_fp16")]; + tensor var_23280_begin_0 = const()[name = string("op_23280_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23280_end_0 = const()[name = string("op_23280_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23280_end_mask_0 = const()[name = string("op_23280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23280_cast_fp16 = slice_by_index(begin = var_23280_begin_0, end = var_23280_end_0, end_mask = var_23280_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23280_cast_fp16")]; + tensor var_23287_begin_0 = const()[name = string("op_23287_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23287_end_0 = const()[name = string("op_23287_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23287_end_mask_0 = const()[name = string("op_23287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23287_cast_fp16 = slice_by_index(begin = var_23287_begin_0, end = var_23287_end_0, end_mask = var_23287_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23287_cast_fp16")]; + tensor var_23294_begin_0 = const()[name = string("op_23294_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23294_end_0 = const()[name = string("op_23294_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23294_end_mask_0 = const()[name = string("op_23294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23294_cast_fp16 = slice_by_index(begin = var_23294_begin_0, end = var_23294_end_0, end_mask = var_23294_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23294_cast_fp16")]; + tensor var_23301_begin_0 = const()[name = string("op_23301_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23301_end_0 = const()[name = string("op_23301_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23301_end_mask_0 = const()[name = string("op_23301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23301_cast_fp16 = slice_by_index(begin = var_23301_begin_0, end = var_23301_end_0, end_mask = var_23301_end_mask_0, x = var_23171_cast_fp16)[name = string("op_23301_cast_fp16")]; + tensor var_23308_begin_0 = const()[name = string("op_23308_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23308_end_0 = const()[name = string("op_23308_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23308_end_mask_0 = const()[name = string("op_23308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23308_cast_fp16 = slice_by_index(begin = var_23308_begin_0, end = var_23308_end_0, end_mask = var_23308_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23308_cast_fp16")]; + tensor var_23315_begin_0 = const()[name = string("op_23315_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23315_end_0 = const()[name = string("op_23315_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23315_end_mask_0 = const()[name = string("op_23315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23315_cast_fp16 = slice_by_index(begin = var_23315_begin_0, end = var_23315_end_0, end_mask = var_23315_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23315_cast_fp16")]; + tensor var_23322_begin_0 = const()[name = string("op_23322_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23322_end_0 = const()[name = string("op_23322_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23322_end_mask_0 = const()[name = string("op_23322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23322_cast_fp16 = slice_by_index(begin = var_23322_begin_0, end = var_23322_end_0, end_mask = var_23322_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23322_cast_fp16")]; + tensor var_23329_begin_0 = const()[name = string("op_23329_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23329_end_0 = const()[name = string("op_23329_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23329_end_mask_0 = const()[name = string("op_23329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23329_cast_fp16 = slice_by_index(begin = var_23329_begin_0, end = var_23329_end_0, end_mask = var_23329_end_mask_0, x = var_23175_cast_fp16)[name = string("op_23329_cast_fp16")]; + tensor var_23336_begin_0 = const()[name = string("op_23336_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23336_end_0 = const()[name = string("op_23336_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23336_end_mask_0 = const()[name = string("op_23336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23336_cast_fp16 = slice_by_index(begin = var_23336_begin_0, end = var_23336_end_0, end_mask = var_23336_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23336_cast_fp16")]; + tensor var_23343_begin_0 = const()[name = string("op_23343_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23343_end_0 = const()[name = string("op_23343_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23343_end_mask_0 = const()[name = string("op_23343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23343_cast_fp16 = slice_by_index(begin = var_23343_begin_0, end = var_23343_end_0, end_mask = var_23343_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23343_cast_fp16")]; + tensor var_23350_begin_0 = const()[name = string("op_23350_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23350_end_0 = const()[name = string("op_23350_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23350_end_mask_0 = const()[name = string("op_23350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23350_cast_fp16 = slice_by_index(begin = var_23350_begin_0, end = var_23350_end_0, end_mask = var_23350_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23350_cast_fp16")]; + tensor var_23357_begin_0 = const()[name = string("op_23357_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23357_end_0 = const()[name = string("op_23357_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23357_end_mask_0 = const()[name = string("op_23357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23357_cast_fp16 = slice_by_index(begin = var_23357_begin_0, end = var_23357_end_0, end_mask = var_23357_end_mask_0, x = var_23179_cast_fp16)[name = string("op_23357_cast_fp16")]; + tensor var_23364_begin_0 = const()[name = string("op_23364_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23364_end_0 = const()[name = string("op_23364_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23364_end_mask_0 = const()[name = string("op_23364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23364_cast_fp16 = slice_by_index(begin = var_23364_begin_0, end = var_23364_end_0, end_mask = var_23364_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23364_cast_fp16")]; + tensor var_23371_begin_0 = const()[name = string("op_23371_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23371_end_0 = const()[name = string("op_23371_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23371_end_mask_0 = const()[name = string("op_23371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23371_cast_fp16 = slice_by_index(begin = var_23371_begin_0, end = var_23371_end_0, end_mask = var_23371_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23371_cast_fp16")]; + tensor var_23378_begin_0 = const()[name = string("op_23378_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23378_end_0 = const()[name = string("op_23378_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23378_end_mask_0 = const()[name = string("op_23378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23378_cast_fp16 = slice_by_index(begin = var_23378_begin_0, end = var_23378_end_0, end_mask = var_23378_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23378_cast_fp16")]; + tensor var_23385_begin_0 = const()[name = string("op_23385_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23385_end_0 = const()[name = string("op_23385_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23385_end_mask_0 = const()[name = string("op_23385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23385_cast_fp16 = slice_by_index(begin = var_23385_begin_0, end = var_23385_end_0, end_mask = var_23385_end_mask_0, x = var_23183_cast_fp16)[name = string("op_23385_cast_fp16")]; + tensor var_23392_begin_0 = const()[name = string("op_23392_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23392_end_0 = const()[name = string("op_23392_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23392_end_mask_0 = const()[name = string("op_23392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23392_cast_fp16 = slice_by_index(begin = var_23392_begin_0, end = var_23392_end_0, end_mask = var_23392_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23392_cast_fp16")]; + tensor var_23399_begin_0 = const()[name = string("op_23399_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23399_end_0 = const()[name = string("op_23399_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23399_end_mask_0 = const()[name = string("op_23399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23399_cast_fp16 = slice_by_index(begin = var_23399_begin_0, end = var_23399_end_0, end_mask = var_23399_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23399_cast_fp16")]; + tensor var_23406_begin_0 = const()[name = string("op_23406_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23406_end_0 = const()[name = string("op_23406_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23406_end_mask_0 = const()[name = string("op_23406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23406_cast_fp16 = slice_by_index(begin = var_23406_begin_0, end = var_23406_end_0, end_mask = var_23406_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23406_cast_fp16")]; + tensor var_23413_begin_0 = const()[name = string("op_23413_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23413_end_0 = const()[name = string("op_23413_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23413_end_mask_0 = const()[name = string("op_23413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23413_cast_fp16 = slice_by_index(begin = var_23413_begin_0, end = var_23413_end_0, end_mask = var_23413_end_mask_0, x = var_23187_cast_fp16)[name = string("op_23413_cast_fp16")]; + tensor var_23420_begin_0 = const()[name = string("op_23420_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23420_end_0 = const()[name = string("op_23420_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23420_end_mask_0 = const()[name = string("op_23420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23420_cast_fp16 = slice_by_index(begin = var_23420_begin_0, end = var_23420_end_0, end_mask = var_23420_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23420_cast_fp16")]; + tensor var_23427_begin_0 = const()[name = string("op_23427_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23427_end_0 = const()[name = string("op_23427_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23427_end_mask_0 = const()[name = string("op_23427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23427_cast_fp16 = slice_by_index(begin = var_23427_begin_0, end = var_23427_end_0, end_mask = var_23427_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23427_cast_fp16")]; + tensor var_23434_begin_0 = const()[name = string("op_23434_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23434_end_0 = const()[name = string("op_23434_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23434_end_mask_0 = const()[name = string("op_23434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23434_cast_fp16 = slice_by_index(begin = var_23434_begin_0, end = var_23434_end_0, end_mask = var_23434_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23434_cast_fp16")]; + tensor var_23441_begin_0 = const()[name = string("op_23441_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23441_end_0 = const()[name = string("op_23441_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23441_end_mask_0 = const()[name = string("op_23441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23441_cast_fp16 = slice_by_index(begin = var_23441_begin_0, end = var_23441_end_0, end_mask = var_23441_end_mask_0, x = var_23191_cast_fp16)[name = string("op_23441_cast_fp16")]; + tensor var_23448_begin_0 = const()[name = string("op_23448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23448_end_0 = const()[name = string("op_23448_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23448_end_mask_0 = const()[name = string("op_23448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23448_cast_fp16 = slice_by_index(begin = var_23448_begin_0, end = var_23448_end_0, end_mask = var_23448_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23448_cast_fp16")]; + tensor var_23455_begin_0 = const()[name = string("op_23455_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23455_end_0 = const()[name = string("op_23455_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23455_end_mask_0 = const()[name = string("op_23455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23455_cast_fp16 = slice_by_index(begin = var_23455_begin_0, end = var_23455_end_0, end_mask = var_23455_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23455_cast_fp16")]; + tensor var_23462_begin_0 = const()[name = string("op_23462_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23462_end_0 = const()[name = string("op_23462_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23462_end_mask_0 = const()[name = string("op_23462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23462_cast_fp16 = slice_by_index(begin = var_23462_begin_0, end = var_23462_end_0, end_mask = var_23462_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23462_cast_fp16")]; + tensor var_23469_begin_0 = const()[name = string("op_23469_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23469_end_0 = const()[name = string("op_23469_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23469_end_mask_0 = const()[name = string("op_23469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23469_cast_fp16 = slice_by_index(begin = var_23469_begin_0, end = var_23469_end_0, end_mask = var_23469_end_mask_0, x = var_23195_cast_fp16)[name = string("op_23469_cast_fp16")]; + tensor var_23476_begin_0 = const()[name = string("op_23476_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23476_end_0 = const()[name = string("op_23476_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23476_end_mask_0 = const()[name = string("op_23476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23476_cast_fp16 = slice_by_index(begin = var_23476_begin_0, end = var_23476_end_0, end_mask = var_23476_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23476_cast_fp16")]; + tensor var_23483_begin_0 = const()[name = string("op_23483_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23483_end_0 = const()[name = string("op_23483_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23483_end_mask_0 = const()[name = string("op_23483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23483_cast_fp16 = slice_by_index(begin = var_23483_begin_0, end = var_23483_end_0, end_mask = var_23483_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23483_cast_fp16")]; + tensor var_23490_begin_0 = const()[name = string("op_23490_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23490_end_0 = const()[name = string("op_23490_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23490_end_mask_0 = const()[name = string("op_23490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23490_cast_fp16 = slice_by_index(begin = var_23490_begin_0, end = var_23490_end_0, end_mask = var_23490_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23490_cast_fp16")]; + tensor var_23497_begin_0 = const()[name = string("op_23497_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23497_end_0 = const()[name = string("op_23497_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23497_end_mask_0 = const()[name = string("op_23497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23497_cast_fp16 = slice_by_index(begin = var_23497_begin_0, end = var_23497_end_0, end_mask = var_23497_end_mask_0, x = var_23199_cast_fp16)[name = string("op_23497_cast_fp16")]; + tensor var_23504_begin_0 = const()[name = string("op_23504_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23504_end_0 = const()[name = string("op_23504_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23504_end_mask_0 = const()[name = string("op_23504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23504_cast_fp16 = slice_by_index(begin = var_23504_begin_0, end = var_23504_end_0, end_mask = var_23504_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23504_cast_fp16")]; + tensor var_23511_begin_0 = const()[name = string("op_23511_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23511_end_0 = const()[name = string("op_23511_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23511_end_mask_0 = const()[name = string("op_23511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23511_cast_fp16 = slice_by_index(begin = var_23511_begin_0, end = var_23511_end_0, end_mask = var_23511_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23511_cast_fp16")]; + tensor var_23518_begin_0 = const()[name = string("op_23518_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23518_end_0 = const()[name = string("op_23518_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23518_end_mask_0 = const()[name = string("op_23518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23518_cast_fp16 = slice_by_index(begin = var_23518_begin_0, end = var_23518_end_0, end_mask = var_23518_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23518_cast_fp16")]; + tensor var_23525_begin_0 = const()[name = string("op_23525_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23525_end_0 = const()[name = string("op_23525_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23525_end_mask_0 = const()[name = string("op_23525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23525_cast_fp16 = slice_by_index(begin = var_23525_begin_0, end = var_23525_end_0, end_mask = var_23525_end_mask_0, x = var_23203_cast_fp16)[name = string("op_23525_cast_fp16")]; + tensor var_23532_begin_0 = const()[name = string("op_23532_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23532_end_0 = const()[name = string("op_23532_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23532_end_mask_0 = const()[name = string("op_23532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23532_cast_fp16 = slice_by_index(begin = var_23532_begin_0, end = var_23532_end_0, end_mask = var_23532_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23532_cast_fp16")]; + tensor var_23539_begin_0 = const()[name = string("op_23539_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23539_end_0 = const()[name = string("op_23539_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23539_end_mask_0 = const()[name = string("op_23539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23539_cast_fp16 = slice_by_index(begin = var_23539_begin_0, end = var_23539_end_0, end_mask = var_23539_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23539_cast_fp16")]; + tensor var_23546_begin_0 = const()[name = string("op_23546_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23546_end_0 = const()[name = string("op_23546_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23546_end_mask_0 = const()[name = string("op_23546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23546_cast_fp16 = slice_by_index(begin = var_23546_begin_0, end = var_23546_end_0, end_mask = var_23546_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23546_cast_fp16")]; + tensor var_23553_begin_0 = const()[name = string("op_23553_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23553_end_0 = const()[name = string("op_23553_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23553_end_mask_0 = const()[name = string("op_23553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23553_cast_fp16 = slice_by_index(begin = var_23553_begin_0, end = var_23553_end_0, end_mask = var_23553_end_mask_0, x = var_23207_cast_fp16)[name = string("op_23553_cast_fp16")]; + tensor var_23560_begin_0 = const()[name = string("op_23560_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23560_end_0 = const()[name = string("op_23560_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23560_end_mask_0 = const()[name = string("op_23560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23560_cast_fp16 = slice_by_index(begin = var_23560_begin_0, end = var_23560_end_0, end_mask = var_23560_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23560_cast_fp16")]; + tensor var_23567_begin_0 = const()[name = string("op_23567_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23567_end_0 = const()[name = string("op_23567_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23567_end_mask_0 = const()[name = string("op_23567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23567_cast_fp16 = slice_by_index(begin = var_23567_begin_0, end = var_23567_end_0, end_mask = var_23567_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23567_cast_fp16")]; + tensor var_23574_begin_0 = const()[name = string("op_23574_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23574_end_0 = const()[name = string("op_23574_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23574_end_mask_0 = const()[name = string("op_23574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23574_cast_fp16 = slice_by_index(begin = var_23574_begin_0, end = var_23574_end_0, end_mask = var_23574_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23574_cast_fp16")]; + tensor var_23581_begin_0 = const()[name = string("op_23581_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23581_end_0 = const()[name = string("op_23581_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23581_end_mask_0 = const()[name = string("op_23581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23581_cast_fp16 = slice_by_index(begin = var_23581_begin_0, end = var_23581_end_0, end_mask = var_23581_end_mask_0, x = var_23211_cast_fp16)[name = string("op_23581_cast_fp16")]; + tensor var_23588_begin_0 = const()[name = string("op_23588_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23588_end_0 = const()[name = string("op_23588_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23588_end_mask_0 = const()[name = string("op_23588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23588_cast_fp16 = slice_by_index(begin = var_23588_begin_0, end = var_23588_end_0, end_mask = var_23588_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23588_cast_fp16")]; + tensor var_23595_begin_0 = const()[name = string("op_23595_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23595_end_0 = const()[name = string("op_23595_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23595_end_mask_0 = const()[name = string("op_23595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23595_cast_fp16 = slice_by_index(begin = var_23595_begin_0, end = var_23595_end_0, end_mask = var_23595_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23595_cast_fp16")]; + tensor var_23602_begin_0 = const()[name = string("op_23602_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23602_end_0 = const()[name = string("op_23602_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23602_end_mask_0 = const()[name = string("op_23602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23602_cast_fp16 = slice_by_index(begin = var_23602_begin_0, end = var_23602_end_0, end_mask = var_23602_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23602_cast_fp16")]; + tensor var_23609_begin_0 = const()[name = string("op_23609_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23609_end_0 = const()[name = string("op_23609_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23609_end_mask_0 = const()[name = string("op_23609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23609_cast_fp16 = slice_by_index(begin = var_23609_begin_0, end = var_23609_end_0, end_mask = var_23609_end_mask_0, x = var_23215_cast_fp16)[name = string("op_23609_cast_fp16")]; + tensor var_23616_begin_0 = const()[name = string("op_23616_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23616_end_0 = const()[name = string("op_23616_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23616_end_mask_0 = const()[name = string("op_23616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23616_cast_fp16 = slice_by_index(begin = var_23616_begin_0, end = var_23616_end_0, end_mask = var_23616_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23616_cast_fp16")]; + tensor var_23623_begin_0 = const()[name = string("op_23623_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23623_end_0 = const()[name = string("op_23623_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23623_end_mask_0 = const()[name = string("op_23623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23623_cast_fp16 = slice_by_index(begin = var_23623_begin_0, end = var_23623_end_0, end_mask = var_23623_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23623_cast_fp16")]; + tensor var_23630_begin_0 = const()[name = string("op_23630_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23630_end_0 = const()[name = string("op_23630_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23630_end_mask_0 = const()[name = string("op_23630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23630_cast_fp16 = slice_by_index(begin = var_23630_begin_0, end = var_23630_end_0, end_mask = var_23630_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23630_cast_fp16")]; + tensor var_23637_begin_0 = const()[name = string("op_23637_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23637_end_0 = const()[name = string("op_23637_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23637_end_mask_0 = const()[name = string("op_23637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23637_cast_fp16 = slice_by_index(begin = var_23637_begin_0, end = var_23637_end_0, end_mask = var_23637_end_mask_0, x = var_23219_cast_fp16)[name = string("op_23637_cast_fp16")]; + tensor var_23644_begin_0 = const()[name = string("op_23644_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23644_end_0 = const()[name = string("op_23644_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23644_end_mask_0 = const()[name = string("op_23644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23644_cast_fp16 = slice_by_index(begin = var_23644_begin_0, end = var_23644_end_0, end_mask = var_23644_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23644_cast_fp16")]; + tensor var_23651_begin_0 = const()[name = string("op_23651_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23651_end_0 = const()[name = string("op_23651_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23651_end_mask_0 = const()[name = string("op_23651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23651_cast_fp16 = slice_by_index(begin = var_23651_begin_0, end = var_23651_end_0, end_mask = var_23651_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23651_cast_fp16")]; + tensor var_23658_begin_0 = const()[name = string("op_23658_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23658_end_0 = const()[name = string("op_23658_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23658_end_mask_0 = const()[name = string("op_23658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23658_cast_fp16 = slice_by_index(begin = var_23658_begin_0, end = var_23658_end_0, end_mask = var_23658_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23658_cast_fp16")]; + tensor var_23665_begin_0 = const()[name = string("op_23665_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23665_end_0 = const()[name = string("op_23665_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23665_end_mask_0 = const()[name = string("op_23665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23665_cast_fp16 = slice_by_index(begin = var_23665_begin_0, end = var_23665_end_0, end_mask = var_23665_end_mask_0, x = var_23223_cast_fp16)[name = string("op_23665_cast_fp16")]; + tensor var_23672_begin_0 = const()[name = string("op_23672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23672_end_0 = const()[name = string("op_23672_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23672_end_mask_0 = const()[name = string("op_23672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23672_cast_fp16 = slice_by_index(begin = var_23672_begin_0, end = var_23672_end_0, end_mask = var_23672_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23672_cast_fp16")]; + tensor var_23679_begin_0 = const()[name = string("op_23679_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23679_end_0 = const()[name = string("op_23679_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23679_end_mask_0 = const()[name = string("op_23679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23679_cast_fp16 = slice_by_index(begin = var_23679_begin_0, end = var_23679_end_0, end_mask = var_23679_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23679_cast_fp16")]; + tensor var_23686_begin_0 = const()[name = string("op_23686_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23686_end_0 = const()[name = string("op_23686_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23686_end_mask_0 = const()[name = string("op_23686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23686_cast_fp16 = slice_by_index(begin = var_23686_begin_0, end = var_23686_end_0, end_mask = var_23686_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23686_cast_fp16")]; + tensor var_23693_begin_0 = const()[name = string("op_23693_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23693_end_0 = const()[name = string("op_23693_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23693_end_mask_0 = const()[name = string("op_23693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23693_cast_fp16 = slice_by_index(begin = var_23693_begin_0, end = var_23693_end_0, end_mask = var_23693_end_mask_0, x = var_23227_cast_fp16)[name = string("op_23693_cast_fp16")]; + tensor var_23700_begin_0 = const()[name = string("op_23700_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23700_end_0 = const()[name = string("op_23700_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23700_end_mask_0 = const()[name = string("op_23700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23700_cast_fp16 = slice_by_index(begin = var_23700_begin_0, end = var_23700_end_0, end_mask = var_23700_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23700_cast_fp16")]; + tensor var_23707_begin_0 = const()[name = string("op_23707_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23707_end_0 = const()[name = string("op_23707_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23707_end_mask_0 = const()[name = string("op_23707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23707_cast_fp16 = slice_by_index(begin = var_23707_begin_0, end = var_23707_end_0, end_mask = var_23707_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23707_cast_fp16")]; + tensor var_23714_begin_0 = const()[name = string("op_23714_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23714_end_0 = const()[name = string("op_23714_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23714_end_mask_0 = const()[name = string("op_23714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23714_cast_fp16 = slice_by_index(begin = var_23714_begin_0, end = var_23714_end_0, end_mask = var_23714_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23714_cast_fp16")]; + tensor var_23721_begin_0 = const()[name = string("op_23721_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23721_end_0 = const()[name = string("op_23721_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23721_end_mask_0 = const()[name = string("op_23721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23721_cast_fp16 = slice_by_index(begin = var_23721_begin_0, end = var_23721_end_0, end_mask = var_23721_end_mask_0, x = var_23231_cast_fp16)[name = string("op_23721_cast_fp16")]; + tensor var_23728_begin_0 = const()[name = string("op_23728_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23728_end_0 = const()[name = string("op_23728_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23728_end_mask_0 = const()[name = string("op_23728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23728_cast_fp16 = slice_by_index(begin = var_23728_begin_0, end = var_23728_end_0, end_mask = var_23728_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23728_cast_fp16")]; + tensor var_23735_begin_0 = const()[name = string("op_23735_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23735_end_0 = const()[name = string("op_23735_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23735_end_mask_0 = const()[name = string("op_23735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23735_cast_fp16 = slice_by_index(begin = var_23735_begin_0, end = var_23735_end_0, end_mask = var_23735_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23735_cast_fp16")]; + tensor var_23742_begin_0 = const()[name = string("op_23742_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23742_end_0 = const()[name = string("op_23742_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23742_end_mask_0 = const()[name = string("op_23742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23742_cast_fp16 = slice_by_index(begin = var_23742_begin_0, end = var_23742_end_0, end_mask = var_23742_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23742_cast_fp16")]; + tensor var_23749_begin_0 = const()[name = string("op_23749_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23749_end_0 = const()[name = string("op_23749_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23749_end_mask_0 = const()[name = string("op_23749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23749_cast_fp16 = slice_by_index(begin = var_23749_begin_0, end = var_23749_end_0, end_mask = var_23749_end_mask_0, x = var_23235_cast_fp16)[name = string("op_23749_cast_fp16")]; + tensor var_23756_begin_0 = const()[name = string("op_23756_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23756_end_0 = const()[name = string("op_23756_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23756_end_mask_0 = const()[name = string("op_23756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23756_cast_fp16 = slice_by_index(begin = var_23756_begin_0, end = var_23756_end_0, end_mask = var_23756_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23756_cast_fp16")]; + tensor var_23763_begin_0 = const()[name = string("op_23763_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23763_end_0 = const()[name = string("op_23763_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23763_end_mask_0 = const()[name = string("op_23763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23763_cast_fp16 = slice_by_index(begin = var_23763_begin_0, end = var_23763_end_0, end_mask = var_23763_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23763_cast_fp16")]; + tensor var_23770_begin_0 = const()[name = string("op_23770_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23770_end_0 = const()[name = string("op_23770_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23770_end_mask_0 = const()[name = string("op_23770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23770_cast_fp16 = slice_by_index(begin = var_23770_begin_0, end = var_23770_end_0, end_mask = var_23770_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23770_cast_fp16")]; + tensor var_23777_begin_0 = const()[name = string("op_23777_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23777_end_0 = const()[name = string("op_23777_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23777_end_mask_0 = const()[name = string("op_23777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23777_cast_fp16 = slice_by_index(begin = var_23777_begin_0, end = var_23777_end_0, end_mask = var_23777_end_mask_0, x = var_23239_cast_fp16)[name = string("op_23777_cast_fp16")]; + tensor var_23784_begin_0 = const()[name = string("op_23784_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23784_end_0 = const()[name = string("op_23784_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23784_end_mask_0 = const()[name = string("op_23784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23784_cast_fp16 = slice_by_index(begin = var_23784_begin_0, end = var_23784_end_0, end_mask = var_23784_end_mask_0, x = var_23243_cast_fp16)[name = string("op_23784_cast_fp16")]; + tensor var_23791_begin_0 = const()[name = string("op_23791_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23791_end_0 = const()[name = string("op_23791_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23791_end_mask_0 = const()[name = string("op_23791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23791_cast_fp16 = slice_by_index(begin = var_23791_begin_0, end = var_23791_end_0, end_mask = var_23791_end_mask_0, x = var_23243_cast_fp16)[name = string("op_23791_cast_fp16")]; + tensor var_23798_begin_0 = const()[name = string("op_23798_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23798_end_0 = const()[name = string("op_23798_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23798_end_mask_0 = const()[name = string("op_23798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23798_cast_fp16 = slice_by_index(begin = var_23798_begin_0, end = var_23798_end_0, end_mask = var_23798_end_mask_0, x = var_23243_cast_fp16)[name = string("op_23798_cast_fp16")]; + tensor var_23805_begin_0 = const()[name = string("op_23805_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23805_end_0 = const()[name = string("op_23805_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23805_end_mask_0 = const()[name = string("op_23805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23805_cast_fp16 = slice_by_index(begin = var_23805_begin_0, end = var_23805_end_0, end_mask = var_23805_end_mask_0, x = var_23243_cast_fp16)[name = string("op_23805_cast_fp16")]; + tensor k_31_perm_0 = const()[name = string("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23810_begin_0 = const()[name = string("op_23810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23810_end_0 = const()[name = string("op_23810_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_23810_end_mask_0 = const()[name = string("op_23810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_31_cast_fp16 = transpose(perm = k_31_perm_0, x = key_31_cast_fp16)[name = string("transpose_16")]; + tensor var_23810_cast_fp16 = slice_by_index(begin = var_23810_begin_0, end = var_23810_end_0, end_mask = var_23810_end_mask_0, x = k_31_cast_fp16)[name = string("op_23810_cast_fp16")]; + tensor var_23814_begin_0 = const()[name = string("op_23814_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23814_end_0 = const()[name = string("op_23814_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_23814_end_mask_0 = const()[name = string("op_23814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23814_cast_fp16 = slice_by_index(begin = var_23814_begin_0, end = var_23814_end_0, end_mask = var_23814_end_mask_0, x = k_31_cast_fp16)[name = string("op_23814_cast_fp16")]; + tensor var_23818_begin_0 = const()[name = string("op_23818_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23818_end_0 = const()[name = string("op_23818_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_23818_end_mask_0 = const()[name = string("op_23818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23818_cast_fp16 = slice_by_index(begin = var_23818_begin_0, end = var_23818_end_0, end_mask = var_23818_end_mask_0, x = k_31_cast_fp16)[name = string("op_23818_cast_fp16")]; + tensor var_23822_begin_0 = const()[name = string("op_23822_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23822_end_0 = const()[name = string("op_23822_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_23822_end_mask_0 = const()[name = string("op_23822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23822_cast_fp16 = slice_by_index(begin = var_23822_begin_0, end = var_23822_end_0, end_mask = var_23822_end_mask_0, x = k_31_cast_fp16)[name = string("op_23822_cast_fp16")]; + tensor var_23826_begin_0 = const()[name = string("op_23826_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23826_end_0 = const()[name = string("op_23826_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_23826_end_mask_0 = const()[name = string("op_23826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23826_cast_fp16 = slice_by_index(begin = var_23826_begin_0, end = var_23826_end_0, end_mask = var_23826_end_mask_0, x = k_31_cast_fp16)[name = string("op_23826_cast_fp16")]; + tensor var_23830_begin_0 = const()[name = string("op_23830_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23830_end_0 = const()[name = string("op_23830_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_23830_end_mask_0 = const()[name = string("op_23830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23830_cast_fp16 = slice_by_index(begin = var_23830_begin_0, end = var_23830_end_0, end_mask = var_23830_end_mask_0, x = k_31_cast_fp16)[name = string("op_23830_cast_fp16")]; + tensor var_23834_begin_0 = const()[name = string("op_23834_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23834_end_0 = const()[name = string("op_23834_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_23834_end_mask_0 = const()[name = string("op_23834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23834_cast_fp16 = slice_by_index(begin = var_23834_begin_0, end = var_23834_end_0, end_mask = var_23834_end_mask_0, x = k_31_cast_fp16)[name = string("op_23834_cast_fp16")]; + tensor var_23838_begin_0 = const()[name = string("op_23838_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23838_end_0 = const()[name = string("op_23838_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_23838_end_mask_0 = const()[name = string("op_23838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23838_cast_fp16 = slice_by_index(begin = var_23838_begin_0, end = var_23838_end_0, end_mask = var_23838_end_mask_0, x = k_31_cast_fp16)[name = string("op_23838_cast_fp16")]; + tensor var_23842_begin_0 = const()[name = string("op_23842_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23842_end_0 = const()[name = string("op_23842_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_23842_end_mask_0 = const()[name = string("op_23842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23842_cast_fp16 = slice_by_index(begin = var_23842_begin_0, end = var_23842_end_0, end_mask = var_23842_end_mask_0, x = k_31_cast_fp16)[name = string("op_23842_cast_fp16")]; + tensor var_23846_begin_0 = const()[name = string("op_23846_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23846_end_0 = const()[name = string("op_23846_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_23846_end_mask_0 = const()[name = string("op_23846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23846_cast_fp16 = slice_by_index(begin = var_23846_begin_0, end = var_23846_end_0, end_mask = var_23846_end_mask_0, x = k_31_cast_fp16)[name = string("op_23846_cast_fp16")]; + tensor var_23850_begin_0 = const()[name = string("op_23850_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23850_end_0 = const()[name = string("op_23850_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_23850_end_mask_0 = const()[name = string("op_23850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23850_cast_fp16 = slice_by_index(begin = var_23850_begin_0, end = var_23850_end_0, end_mask = var_23850_end_mask_0, x = k_31_cast_fp16)[name = string("op_23850_cast_fp16")]; + tensor var_23854_begin_0 = const()[name = string("op_23854_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23854_end_0 = const()[name = string("op_23854_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_23854_end_mask_0 = const()[name = string("op_23854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23854_cast_fp16 = slice_by_index(begin = var_23854_begin_0, end = var_23854_end_0, end_mask = var_23854_end_mask_0, x = k_31_cast_fp16)[name = string("op_23854_cast_fp16")]; + tensor var_23858_begin_0 = const()[name = string("op_23858_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23858_end_0 = const()[name = string("op_23858_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_23858_end_mask_0 = const()[name = string("op_23858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23858_cast_fp16 = slice_by_index(begin = var_23858_begin_0, end = var_23858_end_0, end_mask = var_23858_end_mask_0, x = k_31_cast_fp16)[name = string("op_23858_cast_fp16")]; + tensor var_23862_begin_0 = const()[name = string("op_23862_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23862_end_0 = const()[name = string("op_23862_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_23862_end_mask_0 = const()[name = string("op_23862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23862_cast_fp16 = slice_by_index(begin = var_23862_begin_0, end = var_23862_end_0, end_mask = var_23862_end_mask_0, x = k_31_cast_fp16)[name = string("op_23862_cast_fp16")]; + tensor var_23866_begin_0 = const()[name = string("op_23866_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23866_end_0 = const()[name = string("op_23866_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_23866_end_mask_0 = const()[name = string("op_23866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23866_cast_fp16 = slice_by_index(begin = var_23866_begin_0, end = var_23866_end_0, end_mask = var_23866_end_mask_0, x = k_31_cast_fp16)[name = string("op_23866_cast_fp16")]; + tensor var_23870_begin_0 = const()[name = string("op_23870_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23870_end_0 = const()[name = string("op_23870_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_23870_end_mask_0 = const()[name = string("op_23870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23870_cast_fp16 = slice_by_index(begin = var_23870_begin_0, end = var_23870_end_0, end_mask = var_23870_end_mask_0, x = k_31_cast_fp16)[name = string("op_23870_cast_fp16")]; + tensor var_23874_begin_0 = const()[name = string("op_23874_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23874_end_0 = const()[name = string("op_23874_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_23874_end_mask_0 = const()[name = string("op_23874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23874_cast_fp16 = slice_by_index(begin = var_23874_begin_0, end = var_23874_end_0, end_mask = var_23874_end_mask_0, x = k_31_cast_fp16)[name = string("op_23874_cast_fp16")]; + tensor var_23878_begin_0 = const()[name = string("op_23878_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23878_end_0 = const()[name = string("op_23878_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_23878_end_mask_0 = const()[name = string("op_23878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23878_cast_fp16 = slice_by_index(begin = var_23878_begin_0, end = var_23878_end_0, end_mask = var_23878_end_mask_0, x = k_31_cast_fp16)[name = string("op_23878_cast_fp16")]; + tensor var_23882_begin_0 = const()[name = string("op_23882_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23882_end_0 = const()[name = string("op_23882_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_23882_end_mask_0 = const()[name = string("op_23882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23882_cast_fp16 = slice_by_index(begin = var_23882_begin_0, end = var_23882_end_0, end_mask = var_23882_end_mask_0, x = k_31_cast_fp16)[name = string("op_23882_cast_fp16")]; + tensor var_23886_begin_0 = const()[name = string("op_23886_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23886_end_0 = const()[name = string("op_23886_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_23886_end_mask_0 = const()[name = string("op_23886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23886_cast_fp16 = slice_by_index(begin = var_23886_begin_0, end = var_23886_end_0, end_mask = var_23886_end_mask_0, x = k_31_cast_fp16)[name = string("op_23886_cast_fp16")]; + tensor var_23888_begin_0 = const()[name = string("op_23888_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23888_end_0 = const()[name = string("op_23888_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23888_end_mask_0 = const()[name = string("op_23888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23888_cast_fp16 = slice_by_index(begin = var_23888_begin_0, end = var_23888_end_0, end_mask = var_23888_end_mask_0, x = value_31_cast_fp16)[name = string("op_23888_cast_fp16")]; + tensor var_23892_begin_0 = const()[name = string("op_23892_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23892_end_0 = const()[name = string("op_23892_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_23892_end_mask_0 = const()[name = string("op_23892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23892_cast_fp16 = slice_by_index(begin = var_23892_begin_0, end = var_23892_end_0, end_mask = var_23892_end_mask_0, x = value_31_cast_fp16)[name = string("op_23892_cast_fp16")]; + tensor var_23896_begin_0 = const()[name = string("op_23896_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23896_end_0 = const()[name = string("op_23896_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_23896_end_mask_0 = const()[name = string("op_23896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23896_cast_fp16 = slice_by_index(begin = var_23896_begin_0, end = var_23896_end_0, end_mask = var_23896_end_mask_0, x = value_31_cast_fp16)[name = string("op_23896_cast_fp16")]; + tensor var_23900_begin_0 = const()[name = string("op_23900_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23900_end_0 = const()[name = string("op_23900_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_23900_end_mask_0 = const()[name = string("op_23900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23900_cast_fp16 = slice_by_index(begin = var_23900_begin_0, end = var_23900_end_0, end_mask = var_23900_end_mask_0, x = value_31_cast_fp16)[name = string("op_23900_cast_fp16")]; + tensor var_23904_begin_0 = const()[name = string("op_23904_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23904_end_0 = const()[name = string("op_23904_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_23904_end_mask_0 = const()[name = string("op_23904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23904_cast_fp16 = slice_by_index(begin = var_23904_begin_0, end = var_23904_end_0, end_mask = var_23904_end_mask_0, x = value_31_cast_fp16)[name = string("op_23904_cast_fp16")]; + tensor var_23908_begin_0 = const()[name = string("op_23908_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23908_end_0 = const()[name = string("op_23908_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_23908_end_mask_0 = const()[name = string("op_23908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23908_cast_fp16 = slice_by_index(begin = var_23908_begin_0, end = var_23908_end_0, end_mask = var_23908_end_mask_0, x = value_31_cast_fp16)[name = string("op_23908_cast_fp16")]; + tensor var_23912_begin_0 = const()[name = string("op_23912_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23912_end_0 = const()[name = string("op_23912_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_23912_end_mask_0 = const()[name = string("op_23912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23912_cast_fp16 = slice_by_index(begin = var_23912_begin_0, end = var_23912_end_0, end_mask = var_23912_end_mask_0, x = value_31_cast_fp16)[name = string("op_23912_cast_fp16")]; + tensor var_23916_begin_0 = const()[name = string("op_23916_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23916_end_0 = const()[name = string("op_23916_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_23916_end_mask_0 = const()[name = string("op_23916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23916_cast_fp16 = slice_by_index(begin = var_23916_begin_0, end = var_23916_end_0, end_mask = var_23916_end_mask_0, x = value_31_cast_fp16)[name = string("op_23916_cast_fp16")]; + tensor var_23920_begin_0 = const()[name = string("op_23920_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23920_end_0 = const()[name = string("op_23920_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_23920_end_mask_0 = const()[name = string("op_23920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23920_cast_fp16 = slice_by_index(begin = var_23920_begin_0, end = var_23920_end_0, end_mask = var_23920_end_mask_0, x = value_31_cast_fp16)[name = string("op_23920_cast_fp16")]; + tensor var_23924_begin_0 = const()[name = string("op_23924_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23924_end_0 = const()[name = string("op_23924_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_23924_end_mask_0 = const()[name = string("op_23924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23924_cast_fp16 = slice_by_index(begin = var_23924_begin_0, end = var_23924_end_0, end_mask = var_23924_end_mask_0, x = value_31_cast_fp16)[name = string("op_23924_cast_fp16")]; + tensor var_23928_begin_0 = const()[name = string("op_23928_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23928_end_0 = const()[name = string("op_23928_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_23928_end_mask_0 = const()[name = string("op_23928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23928_cast_fp16 = slice_by_index(begin = var_23928_begin_0, end = var_23928_end_0, end_mask = var_23928_end_mask_0, x = value_31_cast_fp16)[name = string("op_23928_cast_fp16")]; + tensor var_23932_begin_0 = const()[name = string("op_23932_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23932_end_0 = const()[name = string("op_23932_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_23932_end_mask_0 = const()[name = string("op_23932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23932_cast_fp16 = slice_by_index(begin = var_23932_begin_0, end = var_23932_end_0, end_mask = var_23932_end_mask_0, x = value_31_cast_fp16)[name = string("op_23932_cast_fp16")]; + tensor var_23936_begin_0 = const()[name = string("op_23936_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23936_end_0 = const()[name = string("op_23936_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_23936_end_mask_0 = const()[name = string("op_23936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23936_cast_fp16 = slice_by_index(begin = var_23936_begin_0, end = var_23936_end_0, end_mask = var_23936_end_mask_0, x = value_31_cast_fp16)[name = string("op_23936_cast_fp16")]; + tensor var_23940_begin_0 = const()[name = string("op_23940_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23940_end_0 = const()[name = string("op_23940_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_23940_end_mask_0 = const()[name = string("op_23940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23940_cast_fp16 = slice_by_index(begin = var_23940_begin_0, end = var_23940_end_0, end_mask = var_23940_end_mask_0, x = value_31_cast_fp16)[name = string("op_23940_cast_fp16")]; + tensor var_23944_begin_0 = const()[name = string("op_23944_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23944_end_0 = const()[name = string("op_23944_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_23944_end_mask_0 = const()[name = string("op_23944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23944_cast_fp16 = slice_by_index(begin = var_23944_begin_0, end = var_23944_end_0, end_mask = var_23944_end_mask_0, x = value_31_cast_fp16)[name = string("op_23944_cast_fp16")]; + tensor var_23948_begin_0 = const()[name = string("op_23948_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23948_end_0 = const()[name = string("op_23948_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_23948_end_mask_0 = const()[name = string("op_23948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23948_cast_fp16 = slice_by_index(begin = var_23948_begin_0, end = var_23948_end_0, end_mask = var_23948_end_mask_0, x = value_31_cast_fp16)[name = string("op_23948_cast_fp16")]; + tensor var_23952_begin_0 = const()[name = string("op_23952_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23952_end_0 = const()[name = string("op_23952_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_23952_end_mask_0 = const()[name = string("op_23952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23952_cast_fp16 = slice_by_index(begin = var_23952_begin_0, end = var_23952_end_0, end_mask = var_23952_end_mask_0, x = value_31_cast_fp16)[name = string("op_23952_cast_fp16")]; + tensor var_23956_begin_0 = const()[name = string("op_23956_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23956_end_0 = const()[name = string("op_23956_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_23956_end_mask_0 = const()[name = string("op_23956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23956_cast_fp16 = slice_by_index(begin = var_23956_begin_0, end = var_23956_end_0, end_mask = var_23956_end_mask_0, x = value_31_cast_fp16)[name = string("op_23956_cast_fp16")]; + tensor var_23960_begin_0 = const()[name = string("op_23960_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23960_end_0 = const()[name = string("op_23960_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_23960_end_mask_0 = const()[name = string("op_23960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23960_cast_fp16 = slice_by_index(begin = var_23960_begin_0, end = var_23960_end_0, end_mask = var_23960_end_mask_0, x = value_31_cast_fp16)[name = string("op_23960_cast_fp16")]; + tensor var_23964_begin_0 = const()[name = string("op_23964_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23964_end_0 = const()[name = string("op_23964_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_23964_end_mask_0 = const()[name = string("op_23964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23964_cast_fp16 = slice_by_index(begin = var_23964_begin_0, end = var_23964_end_0, end_mask = var_23964_end_mask_0, x = value_31_cast_fp16)[name = string("op_23964_cast_fp16")]; + string _SplitHeadsQ__mh_w_2401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2401_equation_0, values = (var_23810_cast_fp16, var_23252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2401_cast_fp16")]; + string _SplitHeadsQ__mh_w_2403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2403_equation_0, values = (var_23810_cast_fp16, var_23259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2403_cast_fp16")]; + string _SplitHeadsQ__mh_w_2405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2405_equation_0, values = (var_23810_cast_fp16, var_23266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2405_cast_fp16")]; + string _SplitHeadsQ__mh_w_2407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2407_equation_0, values = (var_23810_cast_fp16, var_23273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2407_cast_fp16")]; + string _SplitHeadsQ__mh_w_2409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2409_equation_0, values = (var_23814_cast_fp16, var_23280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2409_cast_fp16")]; + string _SplitHeadsQ__mh_w_2411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2411_equation_0, values = (var_23814_cast_fp16, var_23287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2411_cast_fp16")]; + string _SplitHeadsQ__mh_w_2413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2413_equation_0, values = (var_23814_cast_fp16, var_23294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2413_cast_fp16")]; + string _SplitHeadsQ__mh_w_2415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2415_equation_0, values = (var_23814_cast_fp16, var_23301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2415_cast_fp16")]; + string _SplitHeadsQ__mh_w_2417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2417_equation_0, values = (var_23818_cast_fp16, var_23308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2417_cast_fp16")]; + string _SplitHeadsQ__mh_w_2419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2419_equation_0, values = (var_23818_cast_fp16, var_23315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2419_cast_fp16")]; + string _SplitHeadsQ__mh_w_2421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2421_equation_0, values = (var_23818_cast_fp16, var_23322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2421_cast_fp16")]; + string _SplitHeadsQ__mh_w_2423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2423_equation_0, values = (var_23818_cast_fp16, var_23329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2423_cast_fp16")]; + string _SplitHeadsQ__mh_w_2425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2425_equation_0, values = (var_23822_cast_fp16, var_23336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2425_cast_fp16")]; + string _SplitHeadsQ__mh_w_2427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2427_equation_0, values = (var_23822_cast_fp16, var_23343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2427_cast_fp16")]; + string _SplitHeadsQ__mh_w_2429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2429_equation_0, values = (var_23822_cast_fp16, var_23350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2429_cast_fp16")]; + string _SplitHeadsQ__mh_w_2431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2431_equation_0, values = (var_23822_cast_fp16, var_23357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2431_cast_fp16")]; + string _SplitHeadsQ__mh_w_2433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2433_equation_0, values = (var_23826_cast_fp16, var_23364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2433_cast_fp16")]; + string _SplitHeadsQ__mh_w_2435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2435_equation_0, values = (var_23826_cast_fp16, var_23371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2435_cast_fp16")]; + string _SplitHeadsQ__mh_w_2437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2437_equation_0, values = (var_23826_cast_fp16, var_23378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2437_cast_fp16")]; + string _SplitHeadsQ__mh_w_2439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2439_equation_0, values = (var_23826_cast_fp16, var_23385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2439_cast_fp16")]; + string _SplitHeadsQ__mh_w_2441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2441_equation_0, values = (var_23830_cast_fp16, var_23392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2441_cast_fp16")]; + string _SplitHeadsQ__mh_w_2443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2443_equation_0, values = (var_23830_cast_fp16, var_23399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2443_cast_fp16")]; + string _SplitHeadsQ__mh_w_2445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2445_equation_0, values = (var_23830_cast_fp16, var_23406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2445_cast_fp16")]; + string _SplitHeadsQ__mh_w_2447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2447_equation_0, values = (var_23830_cast_fp16, var_23413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2447_cast_fp16")]; + string _SplitHeadsQ__mh_w_2449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2449_equation_0, values = (var_23834_cast_fp16, var_23420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2449_cast_fp16")]; + string _SplitHeadsQ__mh_w_2451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2451_equation_0, values = (var_23834_cast_fp16, var_23427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2451_cast_fp16")]; + string _SplitHeadsQ__mh_w_2453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2453_equation_0, values = (var_23834_cast_fp16, var_23434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2453_cast_fp16")]; + string _SplitHeadsQ__mh_w_2455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2455_equation_0, values = (var_23834_cast_fp16, var_23441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2455_cast_fp16")]; + string _SplitHeadsQ__mh_w_2457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2457_equation_0, values = (var_23838_cast_fp16, var_23448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2457_cast_fp16")]; + string _SplitHeadsQ__mh_w_2459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2459_equation_0, values = (var_23838_cast_fp16, var_23455_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2459_cast_fp16")]; + string _SplitHeadsQ__mh_w_2461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2461_equation_0, values = (var_23838_cast_fp16, var_23462_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2461_cast_fp16")]; + string _SplitHeadsQ__mh_w_2463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2463_equation_0, values = (var_23838_cast_fp16, var_23469_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2463_cast_fp16")]; + string _SplitHeadsQ__mh_w_2465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2465_equation_0, values = (var_23842_cast_fp16, var_23476_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2465_cast_fp16")]; + string _SplitHeadsQ__mh_w_2467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2467_equation_0, values = (var_23842_cast_fp16, var_23483_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2467_cast_fp16")]; + string _SplitHeadsQ__mh_w_2469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2469_equation_0, values = (var_23842_cast_fp16, var_23490_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2469_cast_fp16")]; + string _SplitHeadsQ__mh_w_2471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2471_equation_0, values = (var_23842_cast_fp16, var_23497_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2471_cast_fp16")]; + string _SplitHeadsQ__mh_w_2473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2473_equation_0, values = (var_23846_cast_fp16, var_23504_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2473_cast_fp16")]; + string _SplitHeadsQ__mh_w_2475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2475_equation_0, values = (var_23846_cast_fp16, var_23511_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2475_cast_fp16")]; + string _SplitHeadsQ__mh_w_2477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2477_equation_0, values = (var_23846_cast_fp16, var_23518_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2477_cast_fp16")]; + string _SplitHeadsQ__mh_w_2479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2479_equation_0, values = (var_23846_cast_fp16, var_23525_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2479_cast_fp16")]; + string _SplitHeadsQ__mh_w_2481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2481_equation_0, values = (var_23850_cast_fp16, var_23532_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2481_cast_fp16")]; + string _SplitHeadsQ__mh_w_2483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2483_equation_0, values = (var_23850_cast_fp16, var_23539_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2483_cast_fp16")]; + string _SplitHeadsQ__mh_w_2485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2485_equation_0, values = (var_23850_cast_fp16, var_23546_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2485_cast_fp16")]; + string _SplitHeadsQ__mh_w_2487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2487_equation_0, values = (var_23850_cast_fp16, var_23553_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2487_cast_fp16")]; + string _SplitHeadsQ__mh_w_2489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2489_equation_0, values = (var_23854_cast_fp16, var_23560_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2489_cast_fp16")]; + string _SplitHeadsQ__mh_w_2491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2491_equation_0, values = (var_23854_cast_fp16, var_23567_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2491_cast_fp16")]; + string _SplitHeadsQ__mh_w_2493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2493_equation_0, values = (var_23854_cast_fp16, var_23574_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2493_cast_fp16")]; + string _SplitHeadsQ__mh_w_2495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2495_equation_0, values = (var_23854_cast_fp16, var_23581_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2495_cast_fp16")]; + string _SplitHeadsQ__mh_w_2497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2497_equation_0, values = (var_23858_cast_fp16, var_23588_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2497_cast_fp16")]; + string _SplitHeadsQ__mh_w_2499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2499_equation_0, values = (var_23858_cast_fp16, var_23595_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2499_cast_fp16")]; + string _SplitHeadsQ__mh_w_2501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2501_equation_0, values = (var_23858_cast_fp16, var_23602_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2501_cast_fp16")]; + string _SplitHeadsQ__mh_w_2503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2503_equation_0, values = (var_23858_cast_fp16, var_23609_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2503_cast_fp16")]; + string _SplitHeadsQ__mh_w_2505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2505_equation_0, values = (var_23862_cast_fp16, var_23616_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2505_cast_fp16")]; + string _SplitHeadsQ__mh_w_2507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2507_equation_0, values = (var_23862_cast_fp16, var_23623_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2507_cast_fp16")]; + string _SplitHeadsQ__mh_w_2509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2509_equation_0, values = (var_23862_cast_fp16, var_23630_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2509_cast_fp16")]; + string _SplitHeadsQ__mh_w_2511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2511_equation_0, values = (var_23862_cast_fp16, var_23637_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2511_cast_fp16")]; + string _SplitHeadsQ__mh_w_2513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2513_equation_0, values = (var_23866_cast_fp16, var_23644_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2513_cast_fp16")]; + string _SplitHeadsQ__mh_w_2515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2515_equation_0, values = (var_23866_cast_fp16, var_23651_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2515_cast_fp16")]; + string _SplitHeadsQ__mh_w_2517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2517_equation_0, values = (var_23866_cast_fp16, var_23658_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2517_cast_fp16")]; + string _SplitHeadsQ__mh_w_2519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2519_equation_0, values = (var_23866_cast_fp16, var_23665_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2519_cast_fp16")]; + string _SplitHeadsQ__mh_w_2521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2521_equation_0, values = (var_23870_cast_fp16, var_23672_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2521_cast_fp16")]; + string _SplitHeadsQ__mh_w_2523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2523_equation_0, values = (var_23870_cast_fp16, var_23679_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2523_cast_fp16")]; + string _SplitHeadsQ__mh_w_2525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2525_equation_0, values = (var_23870_cast_fp16, var_23686_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2525_cast_fp16")]; + string _SplitHeadsQ__mh_w_2527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2527_equation_0, values = (var_23870_cast_fp16, var_23693_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2527_cast_fp16")]; + string _SplitHeadsQ__mh_w_2529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2529_equation_0, values = (var_23874_cast_fp16, var_23700_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2529_cast_fp16")]; + string _SplitHeadsQ__mh_w_2531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2531_equation_0, values = (var_23874_cast_fp16, var_23707_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2531_cast_fp16")]; + string _SplitHeadsQ__mh_w_2533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2533_equation_0, values = (var_23874_cast_fp16, var_23714_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2533_cast_fp16")]; + string _SplitHeadsQ__mh_w_2535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2535_equation_0, values = (var_23874_cast_fp16, var_23721_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2535_cast_fp16")]; + string _SplitHeadsQ__mh_w_2537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2537_equation_0, values = (var_23878_cast_fp16, var_23728_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2537_cast_fp16")]; + string _SplitHeadsQ__mh_w_2539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2539_equation_0, values = (var_23878_cast_fp16, var_23735_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2539_cast_fp16")]; + string _SplitHeadsQ__mh_w_2541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2541_equation_0, values = (var_23878_cast_fp16, var_23742_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2541_cast_fp16")]; + string _SplitHeadsQ__mh_w_2543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2543_equation_0, values = (var_23878_cast_fp16, var_23749_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2543_cast_fp16")]; + string _SplitHeadsQ__mh_w_2545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2545_equation_0, values = (var_23882_cast_fp16, var_23756_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2545_cast_fp16")]; + string _SplitHeadsQ__mh_w_2547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2547_equation_0, values = (var_23882_cast_fp16, var_23763_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2547_cast_fp16")]; + string _SplitHeadsQ__mh_w_2549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2549_equation_0, values = (var_23882_cast_fp16, var_23770_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2549_cast_fp16")]; + string _SplitHeadsQ__mh_w_2551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2551_equation_0, values = (var_23882_cast_fp16, var_23777_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2551_cast_fp16")]; + string _SplitHeadsQ__mh_w_2553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2553_equation_0, values = (var_23886_cast_fp16, var_23784_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2553_cast_fp16")]; + string _SplitHeadsQ__mh_w_2555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2555_equation_0, values = (var_23886_cast_fp16, var_23791_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2555_cast_fp16")]; + string _SplitHeadsQ__mh_w_2557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2557_equation_0, values = (var_23886_cast_fp16, var_23798_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2557_cast_fp16")]; + string _SplitHeadsQ__mh_w_2559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2559_equation_0, values = (var_23886_cast_fp16, var_23805_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2559_cast_fp16")]; + fp16 var_24127_to_fp16 = const()[name = string("op_24127_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2401_cast_fp16, y = var_24127_to_fp16)[name = string("aw_chunk_2401_cast_fp16")]; + fp16 var_24129_to_fp16 = const()[name = string("op_24129_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2403_cast_fp16, y = var_24129_to_fp16)[name = string("aw_chunk_2403_cast_fp16")]; + fp16 var_24131_to_fp16 = const()[name = string("op_24131_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2405_cast_fp16, y = var_24131_to_fp16)[name = string("aw_chunk_2405_cast_fp16")]; + fp16 var_24133_to_fp16 = const()[name = string("op_24133_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2407_cast_fp16, y = var_24133_to_fp16)[name = string("aw_chunk_2407_cast_fp16")]; + fp16 var_24135_to_fp16 = const()[name = string("op_24135_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2409_cast_fp16, y = var_24135_to_fp16)[name = string("aw_chunk_2409_cast_fp16")]; + fp16 var_24137_to_fp16 = const()[name = string("op_24137_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2411_cast_fp16, y = var_24137_to_fp16)[name = string("aw_chunk_2411_cast_fp16")]; + fp16 var_24139_to_fp16 = const()[name = string("op_24139_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2413_cast_fp16, y = var_24139_to_fp16)[name = string("aw_chunk_2413_cast_fp16")]; + fp16 var_24141_to_fp16 = const()[name = string("op_24141_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2415_cast_fp16, y = var_24141_to_fp16)[name = string("aw_chunk_2415_cast_fp16")]; + fp16 var_24143_to_fp16 = const()[name = string("op_24143_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2417_cast_fp16, y = var_24143_to_fp16)[name = string("aw_chunk_2417_cast_fp16")]; + fp16 var_24145_to_fp16 = const()[name = string("op_24145_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2419_cast_fp16, y = var_24145_to_fp16)[name = string("aw_chunk_2419_cast_fp16")]; + fp16 var_24147_to_fp16 = const()[name = string("op_24147_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2421_cast_fp16, y = var_24147_to_fp16)[name = string("aw_chunk_2421_cast_fp16")]; + fp16 var_24149_to_fp16 = const()[name = string("op_24149_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2423_cast_fp16, y = var_24149_to_fp16)[name = string("aw_chunk_2423_cast_fp16")]; + fp16 var_24151_to_fp16 = const()[name = string("op_24151_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2425_cast_fp16, y = var_24151_to_fp16)[name = string("aw_chunk_2425_cast_fp16")]; + fp16 var_24153_to_fp16 = const()[name = string("op_24153_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2427_cast_fp16, y = var_24153_to_fp16)[name = string("aw_chunk_2427_cast_fp16")]; + fp16 var_24155_to_fp16 = const()[name = string("op_24155_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2429_cast_fp16, y = var_24155_to_fp16)[name = string("aw_chunk_2429_cast_fp16")]; + fp16 var_24157_to_fp16 = const()[name = string("op_24157_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2431_cast_fp16, y = var_24157_to_fp16)[name = string("aw_chunk_2431_cast_fp16")]; + fp16 var_24159_to_fp16 = const()[name = string("op_24159_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2433_cast_fp16, y = var_24159_to_fp16)[name = string("aw_chunk_2433_cast_fp16")]; + fp16 var_24161_to_fp16 = const()[name = string("op_24161_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2435_cast_fp16, y = var_24161_to_fp16)[name = string("aw_chunk_2435_cast_fp16")]; + fp16 var_24163_to_fp16 = const()[name = string("op_24163_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2437_cast_fp16, y = var_24163_to_fp16)[name = string("aw_chunk_2437_cast_fp16")]; + fp16 var_24165_to_fp16 = const()[name = string("op_24165_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2439_cast_fp16, y = var_24165_to_fp16)[name = string("aw_chunk_2439_cast_fp16")]; + fp16 var_24167_to_fp16 = const()[name = string("op_24167_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2441_cast_fp16, y = var_24167_to_fp16)[name = string("aw_chunk_2441_cast_fp16")]; + fp16 var_24169_to_fp16 = const()[name = string("op_24169_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2443_cast_fp16, y = var_24169_to_fp16)[name = string("aw_chunk_2443_cast_fp16")]; + fp16 var_24171_to_fp16 = const()[name = string("op_24171_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2445_cast_fp16, y = var_24171_to_fp16)[name = string("aw_chunk_2445_cast_fp16")]; + fp16 var_24173_to_fp16 = const()[name = string("op_24173_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2447_cast_fp16, y = var_24173_to_fp16)[name = string("aw_chunk_2447_cast_fp16")]; + fp16 var_24175_to_fp16 = const()[name = string("op_24175_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2449_cast_fp16, y = var_24175_to_fp16)[name = string("aw_chunk_2449_cast_fp16")]; + fp16 var_24177_to_fp16 = const()[name = string("op_24177_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2451_cast_fp16, y = var_24177_to_fp16)[name = string("aw_chunk_2451_cast_fp16")]; + fp16 var_24179_to_fp16 = const()[name = string("op_24179_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2453_cast_fp16, y = var_24179_to_fp16)[name = string("aw_chunk_2453_cast_fp16")]; + fp16 var_24181_to_fp16 = const()[name = string("op_24181_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2455_cast_fp16, y = var_24181_to_fp16)[name = string("aw_chunk_2455_cast_fp16")]; + fp16 var_24183_to_fp16 = const()[name = string("op_24183_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2457_cast_fp16, y = var_24183_to_fp16)[name = string("aw_chunk_2457_cast_fp16")]; + fp16 var_24185_to_fp16 = const()[name = string("op_24185_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2459_cast_fp16, y = var_24185_to_fp16)[name = string("aw_chunk_2459_cast_fp16")]; + fp16 var_24187_to_fp16 = const()[name = string("op_24187_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2461_cast_fp16, y = var_24187_to_fp16)[name = string("aw_chunk_2461_cast_fp16")]; + fp16 var_24189_to_fp16 = const()[name = string("op_24189_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2463_cast_fp16, y = var_24189_to_fp16)[name = string("aw_chunk_2463_cast_fp16")]; + fp16 var_24191_to_fp16 = const()[name = string("op_24191_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2465_cast_fp16, y = var_24191_to_fp16)[name = string("aw_chunk_2465_cast_fp16")]; + fp16 var_24193_to_fp16 = const()[name = string("op_24193_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2467_cast_fp16, y = var_24193_to_fp16)[name = string("aw_chunk_2467_cast_fp16")]; + fp16 var_24195_to_fp16 = const()[name = string("op_24195_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2469_cast_fp16, y = var_24195_to_fp16)[name = string("aw_chunk_2469_cast_fp16")]; + fp16 var_24197_to_fp16 = const()[name = string("op_24197_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2471_cast_fp16, y = var_24197_to_fp16)[name = string("aw_chunk_2471_cast_fp16")]; + fp16 var_24199_to_fp16 = const()[name = string("op_24199_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2473_cast_fp16, y = var_24199_to_fp16)[name = string("aw_chunk_2473_cast_fp16")]; + fp16 var_24201_to_fp16 = const()[name = string("op_24201_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2475_cast_fp16, y = var_24201_to_fp16)[name = string("aw_chunk_2475_cast_fp16")]; + fp16 var_24203_to_fp16 = const()[name = string("op_24203_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2477_cast_fp16, y = var_24203_to_fp16)[name = string("aw_chunk_2477_cast_fp16")]; + fp16 var_24205_to_fp16 = const()[name = string("op_24205_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2479_cast_fp16, y = var_24205_to_fp16)[name = string("aw_chunk_2479_cast_fp16")]; + fp16 var_24207_to_fp16 = const()[name = string("op_24207_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2481_cast_fp16, y = var_24207_to_fp16)[name = string("aw_chunk_2481_cast_fp16")]; + fp16 var_24209_to_fp16 = const()[name = string("op_24209_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2483_cast_fp16, y = var_24209_to_fp16)[name = string("aw_chunk_2483_cast_fp16")]; + fp16 var_24211_to_fp16 = const()[name = string("op_24211_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2485_cast_fp16, y = var_24211_to_fp16)[name = string("aw_chunk_2485_cast_fp16")]; + fp16 var_24213_to_fp16 = const()[name = string("op_24213_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2487_cast_fp16, y = var_24213_to_fp16)[name = string("aw_chunk_2487_cast_fp16")]; + fp16 var_24215_to_fp16 = const()[name = string("op_24215_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2489_cast_fp16, y = var_24215_to_fp16)[name = string("aw_chunk_2489_cast_fp16")]; + fp16 var_24217_to_fp16 = const()[name = string("op_24217_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2491_cast_fp16, y = var_24217_to_fp16)[name = string("aw_chunk_2491_cast_fp16")]; + fp16 var_24219_to_fp16 = const()[name = string("op_24219_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2493_cast_fp16, y = var_24219_to_fp16)[name = string("aw_chunk_2493_cast_fp16")]; + fp16 var_24221_to_fp16 = const()[name = string("op_24221_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2495_cast_fp16, y = var_24221_to_fp16)[name = string("aw_chunk_2495_cast_fp16")]; + fp16 var_24223_to_fp16 = const()[name = string("op_24223_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2497_cast_fp16, y = var_24223_to_fp16)[name = string("aw_chunk_2497_cast_fp16")]; + fp16 var_24225_to_fp16 = const()[name = string("op_24225_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2499_cast_fp16, y = var_24225_to_fp16)[name = string("aw_chunk_2499_cast_fp16")]; + fp16 var_24227_to_fp16 = const()[name = string("op_24227_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2501_cast_fp16, y = var_24227_to_fp16)[name = string("aw_chunk_2501_cast_fp16")]; + fp16 var_24229_to_fp16 = const()[name = string("op_24229_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2503_cast_fp16, y = var_24229_to_fp16)[name = string("aw_chunk_2503_cast_fp16")]; + fp16 var_24231_to_fp16 = const()[name = string("op_24231_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2505_cast_fp16, y = var_24231_to_fp16)[name = string("aw_chunk_2505_cast_fp16")]; + fp16 var_24233_to_fp16 = const()[name = string("op_24233_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2507_cast_fp16, y = var_24233_to_fp16)[name = string("aw_chunk_2507_cast_fp16")]; + fp16 var_24235_to_fp16 = const()[name = string("op_24235_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2509_cast_fp16, y = var_24235_to_fp16)[name = string("aw_chunk_2509_cast_fp16")]; + fp16 var_24237_to_fp16 = const()[name = string("op_24237_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2511_cast_fp16, y = var_24237_to_fp16)[name = string("aw_chunk_2511_cast_fp16")]; + fp16 var_24239_to_fp16 = const()[name = string("op_24239_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2513_cast_fp16, y = var_24239_to_fp16)[name = string("aw_chunk_2513_cast_fp16")]; + fp16 var_24241_to_fp16 = const()[name = string("op_24241_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2515_cast_fp16, y = var_24241_to_fp16)[name = string("aw_chunk_2515_cast_fp16")]; + fp16 var_24243_to_fp16 = const()[name = string("op_24243_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2517_cast_fp16, y = var_24243_to_fp16)[name = string("aw_chunk_2517_cast_fp16")]; + fp16 var_24245_to_fp16 = const()[name = string("op_24245_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2519_cast_fp16, y = var_24245_to_fp16)[name = string("aw_chunk_2519_cast_fp16")]; + fp16 var_24247_to_fp16 = const()[name = string("op_24247_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2521_cast_fp16, y = var_24247_to_fp16)[name = string("aw_chunk_2521_cast_fp16")]; + fp16 var_24249_to_fp16 = const()[name = string("op_24249_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2523_cast_fp16, y = var_24249_to_fp16)[name = string("aw_chunk_2523_cast_fp16")]; + fp16 var_24251_to_fp16 = const()[name = string("op_24251_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2525_cast_fp16, y = var_24251_to_fp16)[name = string("aw_chunk_2525_cast_fp16")]; + fp16 var_24253_to_fp16 = const()[name = string("op_24253_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2527_cast_fp16, y = var_24253_to_fp16)[name = string("aw_chunk_2527_cast_fp16")]; + fp16 var_24255_to_fp16 = const()[name = string("op_24255_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2529_cast_fp16, y = var_24255_to_fp16)[name = string("aw_chunk_2529_cast_fp16")]; + fp16 var_24257_to_fp16 = const()[name = string("op_24257_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2531_cast_fp16, y = var_24257_to_fp16)[name = string("aw_chunk_2531_cast_fp16")]; + fp16 var_24259_to_fp16 = const()[name = string("op_24259_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2533_cast_fp16, y = var_24259_to_fp16)[name = string("aw_chunk_2533_cast_fp16")]; + fp16 var_24261_to_fp16 = const()[name = string("op_24261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2535_cast_fp16, y = var_24261_to_fp16)[name = string("aw_chunk_2535_cast_fp16")]; + fp16 var_24263_to_fp16 = const()[name = string("op_24263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2537_cast_fp16, y = var_24263_to_fp16)[name = string("aw_chunk_2537_cast_fp16")]; + fp16 var_24265_to_fp16 = const()[name = string("op_24265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2539_cast_fp16, y = var_24265_to_fp16)[name = string("aw_chunk_2539_cast_fp16")]; + fp16 var_24267_to_fp16 = const()[name = string("op_24267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2541_cast_fp16, y = var_24267_to_fp16)[name = string("aw_chunk_2541_cast_fp16")]; + fp16 var_24269_to_fp16 = const()[name = string("op_24269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2543_cast_fp16, y = var_24269_to_fp16)[name = string("aw_chunk_2543_cast_fp16")]; + fp16 var_24271_to_fp16 = const()[name = string("op_24271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2545_cast_fp16, y = var_24271_to_fp16)[name = string("aw_chunk_2545_cast_fp16")]; + fp16 var_24273_to_fp16 = const()[name = string("op_24273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2547_cast_fp16, y = var_24273_to_fp16)[name = string("aw_chunk_2547_cast_fp16")]; + fp16 var_24275_to_fp16 = const()[name = string("op_24275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2549_cast_fp16, y = var_24275_to_fp16)[name = string("aw_chunk_2549_cast_fp16")]; + fp16 var_24277_to_fp16 = const()[name = string("op_24277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2551_cast_fp16, y = var_24277_to_fp16)[name = string("aw_chunk_2551_cast_fp16")]; + fp16 var_24279_to_fp16 = const()[name = string("op_24279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2553_cast_fp16, y = var_24279_to_fp16)[name = string("aw_chunk_2553_cast_fp16")]; + fp16 var_24281_to_fp16 = const()[name = string("op_24281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2555_cast_fp16, y = var_24281_to_fp16)[name = string("aw_chunk_2555_cast_fp16")]; + fp16 var_24283_to_fp16 = const()[name = string("op_24283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2557_cast_fp16, y = var_24283_to_fp16)[name = string("aw_chunk_2557_cast_fp16")]; + fp16 var_24285_to_fp16 = const()[name = string("op_24285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2559_cast_fp16, y = var_24285_to_fp16)[name = string("aw_chunk_2559_cast_fp16")]; + tensor var_24287_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2401_cast_fp16)[name = string("op_24287_cast_fp16")]; + tensor var_24288_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2403_cast_fp16)[name = string("op_24288_cast_fp16")]; + tensor var_24289_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2405_cast_fp16)[name = string("op_24289_cast_fp16")]; + tensor var_24290_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2407_cast_fp16)[name = string("op_24290_cast_fp16")]; + tensor var_24291_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2409_cast_fp16)[name = string("op_24291_cast_fp16")]; + tensor var_24292_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2411_cast_fp16)[name = string("op_24292_cast_fp16")]; + tensor var_24293_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2413_cast_fp16)[name = string("op_24293_cast_fp16")]; + tensor var_24294_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2415_cast_fp16)[name = string("op_24294_cast_fp16")]; + tensor var_24295_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2417_cast_fp16)[name = string("op_24295_cast_fp16")]; + tensor var_24296_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2419_cast_fp16)[name = string("op_24296_cast_fp16")]; + tensor var_24297_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2421_cast_fp16)[name = string("op_24297_cast_fp16")]; + tensor var_24298_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2423_cast_fp16)[name = string("op_24298_cast_fp16")]; + tensor var_24299_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2425_cast_fp16)[name = string("op_24299_cast_fp16")]; + tensor var_24300_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2427_cast_fp16)[name = string("op_24300_cast_fp16")]; + tensor var_24301_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2429_cast_fp16)[name = string("op_24301_cast_fp16")]; + tensor var_24302_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2431_cast_fp16)[name = string("op_24302_cast_fp16")]; + tensor var_24303_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2433_cast_fp16)[name = string("op_24303_cast_fp16")]; + tensor var_24304_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2435_cast_fp16)[name = string("op_24304_cast_fp16")]; + tensor var_24305_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2437_cast_fp16)[name = string("op_24305_cast_fp16")]; + tensor var_24306_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2439_cast_fp16)[name = string("op_24306_cast_fp16")]; + tensor var_24307_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2441_cast_fp16)[name = string("op_24307_cast_fp16")]; + tensor var_24308_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2443_cast_fp16)[name = string("op_24308_cast_fp16")]; + tensor var_24309_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2445_cast_fp16)[name = string("op_24309_cast_fp16")]; + tensor var_24310_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2447_cast_fp16)[name = string("op_24310_cast_fp16")]; + tensor var_24311_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2449_cast_fp16)[name = string("op_24311_cast_fp16")]; + tensor var_24312_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2451_cast_fp16)[name = string("op_24312_cast_fp16")]; + tensor var_24313_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2453_cast_fp16)[name = string("op_24313_cast_fp16")]; + tensor var_24314_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2455_cast_fp16)[name = string("op_24314_cast_fp16")]; + tensor var_24315_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2457_cast_fp16)[name = string("op_24315_cast_fp16")]; + tensor var_24316_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2459_cast_fp16)[name = string("op_24316_cast_fp16")]; + tensor var_24317_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2461_cast_fp16)[name = string("op_24317_cast_fp16")]; + tensor var_24318_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2463_cast_fp16)[name = string("op_24318_cast_fp16")]; + tensor var_24319_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2465_cast_fp16)[name = string("op_24319_cast_fp16")]; + tensor var_24320_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2467_cast_fp16)[name = string("op_24320_cast_fp16")]; + tensor var_24321_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2469_cast_fp16)[name = string("op_24321_cast_fp16")]; + tensor var_24322_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2471_cast_fp16)[name = string("op_24322_cast_fp16")]; + tensor var_24323_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2473_cast_fp16)[name = string("op_24323_cast_fp16")]; + tensor var_24324_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2475_cast_fp16)[name = string("op_24324_cast_fp16")]; + tensor var_24325_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2477_cast_fp16)[name = string("op_24325_cast_fp16")]; + tensor var_24326_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2479_cast_fp16)[name = string("op_24326_cast_fp16")]; + tensor var_24327_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2481_cast_fp16)[name = string("op_24327_cast_fp16")]; + tensor var_24328_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2483_cast_fp16)[name = string("op_24328_cast_fp16")]; + tensor var_24329_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2485_cast_fp16)[name = string("op_24329_cast_fp16")]; + tensor var_24330_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2487_cast_fp16)[name = string("op_24330_cast_fp16")]; + tensor var_24331_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2489_cast_fp16)[name = string("op_24331_cast_fp16")]; + tensor var_24332_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2491_cast_fp16)[name = string("op_24332_cast_fp16")]; + tensor var_24333_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2493_cast_fp16)[name = string("op_24333_cast_fp16")]; + tensor var_24334_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2495_cast_fp16)[name = string("op_24334_cast_fp16")]; + tensor var_24335_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2497_cast_fp16)[name = string("op_24335_cast_fp16")]; + tensor var_24336_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2499_cast_fp16)[name = string("op_24336_cast_fp16")]; + tensor var_24337_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2501_cast_fp16)[name = string("op_24337_cast_fp16")]; + tensor var_24338_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2503_cast_fp16)[name = string("op_24338_cast_fp16")]; + tensor var_24339_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2505_cast_fp16)[name = string("op_24339_cast_fp16")]; + tensor var_24340_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2507_cast_fp16)[name = string("op_24340_cast_fp16")]; + tensor var_24341_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2509_cast_fp16)[name = string("op_24341_cast_fp16")]; + tensor var_24342_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2511_cast_fp16)[name = string("op_24342_cast_fp16")]; + tensor var_24343_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2513_cast_fp16)[name = string("op_24343_cast_fp16")]; + tensor var_24344_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2515_cast_fp16)[name = string("op_24344_cast_fp16")]; + tensor var_24345_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2517_cast_fp16)[name = string("op_24345_cast_fp16")]; + tensor var_24346_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2519_cast_fp16)[name = string("op_24346_cast_fp16")]; + tensor var_24347_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2521_cast_fp16)[name = string("op_24347_cast_fp16")]; + tensor var_24348_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2523_cast_fp16)[name = string("op_24348_cast_fp16")]; + tensor var_24349_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2525_cast_fp16)[name = string("op_24349_cast_fp16")]; + tensor var_24350_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2527_cast_fp16)[name = string("op_24350_cast_fp16")]; + tensor var_24351_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2529_cast_fp16)[name = string("op_24351_cast_fp16")]; + tensor var_24352_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2531_cast_fp16)[name = string("op_24352_cast_fp16")]; + tensor var_24353_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2533_cast_fp16)[name = string("op_24353_cast_fp16")]; + tensor var_24354_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2535_cast_fp16)[name = string("op_24354_cast_fp16")]; + tensor var_24355_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2537_cast_fp16)[name = string("op_24355_cast_fp16")]; + tensor var_24356_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2539_cast_fp16)[name = string("op_24356_cast_fp16")]; + tensor var_24357_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2541_cast_fp16)[name = string("op_24357_cast_fp16")]; + tensor var_24358_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2543_cast_fp16)[name = string("op_24358_cast_fp16")]; + tensor var_24359_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2545_cast_fp16)[name = string("op_24359_cast_fp16")]; + tensor var_24360_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2547_cast_fp16)[name = string("op_24360_cast_fp16")]; + tensor var_24361_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2549_cast_fp16)[name = string("op_24361_cast_fp16")]; + tensor var_24362_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2551_cast_fp16)[name = string("op_24362_cast_fp16")]; + tensor var_24363_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2553_cast_fp16)[name = string("op_24363_cast_fp16")]; + tensor var_24364_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2555_cast_fp16)[name = string("op_24364_cast_fp16")]; + tensor var_24365_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2557_cast_fp16)[name = string("op_24365_cast_fp16")]; + tensor var_24366_cast_fp16 = softmax(axis = var_23112, x = aw_chunk_2559_cast_fp16)[name = string("op_24366_cast_fp16")]; + string var_24368_equation_0 = const()[name = string("op_24368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24368_cast_fp16 = einsum(equation = var_24368_equation_0, values = (var_23888_cast_fp16, var_24287_cast_fp16))[name = string("op_24368_cast_fp16")]; + string var_24370_equation_0 = const()[name = string("op_24370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24370_cast_fp16 = einsum(equation = var_24370_equation_0, values = (var_23888_cast_fp16, var_24288_cast_fp16))[name = string("op_24370_cast_fp16")]; + string var_24372_equation_0 = const()[name = string("op_24372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24372_cast_fp16 = einsum(equation = var_24372_equation_0, values = (var_23888_cast_fp16, var_24289_cast_fp16))[name = string("op_24372_cast_fp16")]; + string var_24374_equation_0 = const()[name = string("op_24374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24374_cast_fp16 = einsum(equation = var_24374_equation_0, values = (var_23888_cast_fp16, var_24290_cast_fp16))[name = string("op_24374_cast_fp16")]; + string var_24376_equation_0 = const()[name = string("op_24376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24376_cast_fp16 = einsum(equation = var_24376_equation_0, values = (var_23892_cast_fp16, var_24291_cast_fp16))[name = string("op_24376_cast_fp16")]; + string var_24378_equation_0 = const()[name = string("op_24378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24378_cast_fp16 = einsum(equation = var_24378_equation_0, values = (var_23892_cast_fp16, var_24292_cast_fp16))[name = string("op_24378_cast_fp16")]; + string var_24380_equation_0 = const()[name = string("op_24380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24380_cast_fp16 = einsum(equation = var_24380_equation_0, values = (var_23892_cast_fp16, var_24293_cast_fp16))[name = string("op_24380_cast_fp16")]; + string var_24382_equation_0 = const()[name = string("op_24382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24382_cast_fp16 = einsum(equation = var_24382_equation_0, values = (var_23892_cast_fp16, var_24294_cast_fp16))[name = string("op_24382_cast_fp16")]; + string var_24384_equation_0 = const()[name = string("op_24384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24384_cast_fp16 = einsum(equation = var_24384_equation_0, values = (var_23896_cast_fp16, var_24295_cast_fp16))[name = string("op_24384_cast_fp16")]; + string var_24386_equation_0 = const()[name = string("op_24386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24386_cast_fp16 = einsum(equation = var_24386_equation_0, values = (var_23896_cast_fp16, var_24296_cast_fp16))[name = string("op_24386_cast_fp16")]; + string var_24388_equation_0 = const()[name = string("op_24388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24388_cast_fp16 = einsum(equation = var_24388_equation_0, values = (var_23896_cast_fp16, var_24297_cast_fp16))[name = string("op_24388_cast_fp16")]; + string var_24390_equation_0 = const()[name = string("op_24390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24390_cast_fp16 = einsum(equation = var_24390_equation_0, values = (var_23896_cast_fp16, var_24298_cast_fp16))[name = string("op_24390_cast_fp16")]; + string var_24392_equation_0 = const()[name = string("op_24392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24392_cast_fp16 = einsum(equation = var_24392_equation_0, values = (var_23900_cast_fp16, var_24299_cast_fp16))[name = string("op_24392_cast_fp16")]; + string var_24394_equation_0 = const()[name = string("op_24394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24394_cast_fp16 = einsum(equation = var_24394_equation_0, values = (var_23900_cast_fp16, var_24300_cast_fp16))[name = string("op_24394_cast_fp16")]; + string var_24396_equation_0 = const()[name = string("op_24396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24396_cast_fp16 = einsum(equation = var_24396_equation_0, values = (var_23900_cast_fp16, var_24301_cast_fp16))[name = string("op_24396_cast_fp16")]; + string var_24398_equation_0 = const()[name = string("op_24398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24398_cast_fp16 = einsum(equation = var_24398_equation_0, values = (var_23900_cast_fp16, var_24302_cast_fp16))[name = string("op_24398_cast_fp16")]; + string var_24400_equation_0 = const()[name = string("op_24400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24400_cast_fp16 = einsum(equation = var_24400_equation_0, values = (var_23904_cast_fp16, var_24303_cast_fp16))[name = string("op_24400_cast_fp16")]; + string var_24402_equation_0 = const()[name = string("op_24402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24402_cast_fp16 = einsum(equation = var_24402_equation_0, values = (var_23904_cast_fp16, var_24304_cast_fp16))[name = string("op_24402_cast_fp16")]; + string var_24404_equation_0 = const()[name = string("op_24404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24404_cast_fp16 = einsum(equation = var_24404_equation_0, values = (var_23904_cast_fp16, var_24305_cast_fp16))[name = string("op_24404_cast_fp16")]; + string var_24406_equation_0 = const()[name = string("op_24406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24406_cast_fp16 = einsum(equation = var_24406_equation_0, values = (var_23904_cast_fp16, var_24306_cast_fp16))[name = string("op_24406_cast_fp16")]; + string var_24408_equation_0 = const()[name = string("op_24408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24408_cast_fp16 = einsum(equation = var_24408_equation_0, values = (var_23908_cast_fp16, var_24307_cast_fp16))[name = string("op_24408_cast_fp16")]; + string var_24410_equation_0 = const()[name = string("op_24410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24410_cast_fp16 = einsum(equation = var_24410_equation_0, values = (var_23908_cast_fp16, var_24308_cast_fp16))[name = string("op_24410_cast_fp16")]; + string var_24412_equation_0 = const()[name = string("op_24412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24412_cast_fp16 = einsum(equation = var_24412_equation_0, values = (var_23908_cast_fp16, var_24309_cast_fp16))[name = string("op_24412_cast_fp16")]; + string var_24414_equation_0 = const()[name = string("op_24414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24414_cast_fp16 = einsum(equation = var_24414_equation_0, values = (var_23908_cast_fp16, var_24310_cast_fp16))[name = string("op_24414_cast_fp16")]; + string var_24416_equation_0 = const()[name = string("op_24416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24416_cast_fp16 = einsum(equation = var_24416_equation_0, values = (var_23912_cast_fp16, var_24311_cast_fp16))[name = string("op_24416_cast_fp16")]; + string var_24418_equation_0 = const()[name = string("op_24418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24418_cast_fp16 = einsum(equation = var_24418_equation_0, values = (var_23912_cast_fp16, var_24312_cast_fp16))[name = string("op_24418_cast_fp16")]; + string var_24420_equation_0 = const()[name = string("op_24420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24420_cast_fp16 = einsum(equation = var_24420_equation_0, values = (var_23912_cast_fp16, var_24313_cast_fp16))[name = string("op_24420_cast_fp16")]; + string var_24422_equation_0 = const()[name = string("op_24422_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24422_cast_fp16 = einsum(equation = var_24422_equation_0, values = (var_23912_cast_fp16, var_24314_cast_fp16))[name = string("op_24422_cast_fp16")]; + string var_24424_equation_0 = const()[name = string("op_24424_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24424_cast_fp16 = einsum(equation = var_24424_equation_0, values = (var_23916_cast_fp16, var_24315_cast_fp16))[name = string("op_24424_cast_fp16")]; + string var_24426_equation_0 = const()[name = string("op_24426_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24426_cast_fp16 = einsum(equation = var_24426_equation_0, values = (var_23916_cast_fp16, var_24316_cast_fp16))[name = string("op_24426_cast_fp16")]; + string var_24428_equation_0 = const()[name = string("op_24428_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24428_cast_fp16 = einsum(equation = var_24428_equation_0, values = (var_23916_cast_fp16, var_24317_cast_fp16))[name = string("op_24428_cast_fp16")]; + string var_24430_equation_0 = const()[name = string("op_24430_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24430_cast_fp16 = einsum(equation = var_24430_equation_0, values = (var_23916_cast_fp16, var_24318_cast_fp16))[name = string("op_24430_cast_fp16")]; + string var_24432_equation_0 = const()[name = string("op_24432_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24432_cast_fp16 = einsum(equation = var_24432_equation_0, values = (var_23920_cast_fp16, var_24319_cast_fp16))[name = string("op_24432_cast_fp16")]; + string var_24434_equation_0 = const()[name = string("op_24434_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24434_cast_fp16 = einsum(equation = var_24434_equation_0, values = (var_23920_cast_fp16, var_24320_cast_fp16))[name = string("op_24434_cast_fp16")]; + string var_24436_equation_0 = const()[name = string("op_24436_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24436_cast_fp16 = einsum(equation = var_24436_equation_0, values = (var_23920_cast_fp16, var_24321_cast_fp16))[name = string("op_24436_cast_fp16")]; + string var_24438_equation_0 = const()[name = string("op_24438_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24438_cast_fp16 = einsum(equation = var_24438_equation_0, values = (var_23920_cast_fp16, var_24322_cast_fp16))[name = string("op_24438_cast_fp16")]; + string var_24440_equation_0 = const()[name = string("op_24440_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24440_cast_fp16 = einsum(equation = var_24440_equation_0, values = (var_23924_cast_fp16, var_24323_cast_fp16))[name = string("op_24440_cast_fp16")]; + string var_24442_equation_0 = const()[name = string("op_24442_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24442_cast_fp16 = einsum(equation = var_24442_equation_0, values = (var_23924_cast_fp16, var_24324_cast_fp16))[name = string("op_24442_cast_fp16")]; + string var_24444_equation_0 = const()[name = string("op_24444_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24444_cast_fp16 = einsum(equation = var_24444_equation_0, values = (var_23924_cast_fp16, var_24325_cast_fp16))[name = string("op_24444_cast_fp16")]; + string var_24446_equation_0 = const()[name = string("op_24446_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24446_cast_fp16 = einsum(equation = var_24446_equation_0, values = (var_23924_cast_fp16, var_24326_cast_fp16))[name = string("op_24446_cast_fp16")]; + string var_24448_equation_0 = const()[name = string("op_24448_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24448_cast_fp16 = einsum(equation = var_24448_equation_0, values = (var_23928_cast_fp16, var_24327_cast_fp16))[name = string("op_24448_cast_fp16")]; + string var_24450_equation_0 = const()[name = string("op_24450_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24450_cast_fp16 = einsum(equation = var_24450_equation_0, values = (var_23928_cast_fp16, var_24328_cast_fp16))[name = string("op_24450_cast_fp16")]; + string var_24452_equation_0 = const()[name = string("op_24452_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24452_cast_fp16 = einsum(equation = var_24452_equation_0, values = (var_23928_cast_fp16, var_24329_cast_fp16))[name = string("op_24452_cast_fp16")]; + string var_24454_equation_0 = const()[name = string("op_24454_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24454_cast_fp16 = einsum(equation = var_24454_equation_0, values = (var_23928_cast_fp16, var_24330_cast_fp16))[name = string("op_24454_cast_fp16")]; + string var_24456_equation_0 = const()[name = string("op_24456_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24456_cast_fp16 = einsum(equation = var_24456_equation_0, values = (var_23932_cast_fp16, var_24331_cast_fp16))[name = string("op_24456_cast_fp16")]; + string var_24458_equation_0 = const()[name = string("op_24458_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24458_cast_fp16 = einsum(equation = var_24458_equation_0, values = (var_23932_cast_fp16, var_24332_cast_fp16))[name = string("op_24458_cast_fp16")]; + string var_24460_equation_0 = const()[name = string("op_24460_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24460_cast_fp16 = einsum(equation = var_24460_equation_0, values = (var_23932_cast_fp16, var_24333_cast_fp16))[name = string("op_24460_cast_fp16")]; + string var_24462_equation_0 = const()[name = string("op_24462_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24462_cast_fp16 = einsum(equation = var_24462_equation_0, values = (var_23932_cast_fp16, var_24334_cast_fp16))[name = string("op_24462_cast_fp16")]; + string var_24464_equation_0 = const()[name = string("op_24464_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24464_cast_fp16 = einsum(equation = var_24464_equation_0, values = (var_23936_cast_fp16, var_24335_cast_fp16))[name = string("op_24464_cast_fp16")]; + string var_24466_equation_0 = const()[name = string("op_24466_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24466_cast_fp16 = einsum(equation = var_24466_equation_0, values = (var_23936_cast_fp16, var_24336_cast_fp16))[name = string("op_24466_cast_fp16")]; + string var_24468_equation_0 = const()[name = string("op_24468_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24468_cast_fp16 = einsum(equation = var_24468_equation_0, values = (var_23936_cast_fp16, var_24337_cast_fp16))[name = string("op_24468_cast_fp16")]; + string var_24470_equation_0 = const()[name = string("op_24470_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24470_cast_fp16 = einsum(equation = var_24470_equation_0, values = (var_23936_cast_fp16, var_24338_cast_fp16))[name = string("op_24470_cast_fp16")]; + string var_24472_equation_0 = const()[name = string("op_24472_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24472_cast_fp16 = einsum(equation = var_24472_equation_0, values = (var_23940_cast_fp16, var_24339_cast_fp16))[name = string("op_24472_cast_fp16")]; + string var_24474_equation_0 = const()[name = string("op_24474_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24474_cast_fp16 = einsum(equation = var_24474_equation_0, values = (var_23940_cast_fp16, var_24340_cast_fp16))[name = string("op_24474_cast_fp16")]; + string var_24476_equation_0 = const()[name = string("op_24476_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24476_cast_fp16 = einsum(equation = var_24476_equation_0, values = (var_23940_cast_fp16, var_24341_cast_fp16))[name = string("op_24476_cast_fp16")]; + string var_24478_equation_0 = const()[name = string("op_24478_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24478_cast_fp16 = einsum(equation = var_24478_equation_0, values = (var_23940_cast_fp16, var_24342_cast_fp16))[name = string("op_24478_cast_fp16")]; + string var_24480_equation_0 = const()[name = string("op_24480_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24480_cast_fp16 = einsum(equation = var_24480_equation_0, values = (var_23944_cast_fp16, var_24343_cast_fp16))[name = string("op_24480_cast_fp16")]; + string var_24482_equation_0 = const()[name = string("op_24482_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24482_cast_fp16 = einsum(equation = var_24482_equation_0, values = (var_23944_cast_fp16, var_24344_cast_fp16))[name = string("op_24482_cast_fp16")]; + string var_24484_equation_0 = const()[name = string("op_24484_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24484_cast_fp16 = einsum(equation = var_24484_equation_0, values = (var_23944_cast_fp16, var_24345_cast_fp16))[name = string("op_24484_cast_fp16")]; + string var_24486_equation_0 = const()[name = string("op_24486_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24486_cast_fp16 = einsum(equation = var_24486_equation_0, values = (var_23944_cast_fp16, var_24346_cast_fp16))[name = string("op_24486_cast_fp16")]; + string var_24488_equation_0 = const()[name = string("op_24488_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24488_cast_fp16 = einsum(equation = var_24488_equation_0, values = (var_23948_cast_fp16, var_24347_cast_fp16))[name = string("op_24488_cast_fp16")]; + string var_24490_equation_0 = const()[name = string("op_24490_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24490_cast_fp16 = einsum(equation = var_24490_equation_0, values = (var_23948_cast_fp16, var_24348_cast_fp16))[name = string("op_24490_cast_fp16")]; + string var_24492_equation_0 = const()[name = string("op_24492_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24492_cast_fp16 = einsum(equation = var_24492_equation_0, values = (var_23948_cast_fp16, var_24349_cast_fp16))[name = string("op_24492_cast_fp16")]; + string var_24494_equation_0 = const()[name = string("op_24494_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24494_cast_fp16 = einsum(equation = var_24494_equation_0, values = (var_23948_cast_fp16, var_24350_cast_fp16))[name = string("op_24494_cast_fp16")]; + string var_24496_equation_0 = const()[name = string("op_24496_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24496_cast_fp16 = einsum(equation = var_24496_equation_0, values = (var_23952_cast_fp16, var_24351_cast_fp16))[name = string("op_24496_cast_fp16")]; + string var_24498_equation_0 = const()[name = string("op_24498_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24498_cast_fp16 = einsum(equation = var_24498_equation_0, values = (var_23952_cast_fp16, var_24352_cast_fp16))[name = string("op_24498_cast_fp16")]; + string var_24500_equation_0 = const()[name = string("op_24500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24500_cast_fp16 = einsum(equation = var_24500_equation_0, values = (var_23952_cast_fp16, var_24353_cast_fp16))[name = string("op_24500_cast_fp16")]; + string var_24502_equation_0 = const()[name = string("op_24502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24502_cast_fp16 = einsum(equation = var_24502_equation_0, values = (var_23952_cast_fp16, var_24354_cast_fp16))[name = string("op_24502_cast_fp16")]; + string var_24504_equation_0 = const()[name = string("op_24504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24504_cast_fp16 = einsum(equation = var_24504_equation_0, values = (var_23956_cast_fp16, var_24355_cast_fp16))[name = string("op_24504_cast_fp16")]; + string var_24506_equation_0 = const()[name = string("op_24506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24506_cast_fp16 = einsum(equation = var_24506_equation_0, values = (var_23956_cast_fp16, var_24356_cast_fp16))[name = string("op_24506_cast_fp16")]; + string var_24508_equation_0 = const()[name = string("op_24508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24508_cast_fp16 = einsum(equation = var_24508_equation_0, values = (var_23956_cast_fp16, var_24357_cast_fp16))[name = string("op_24508_cast_fp16")]; + string var_24510_equation_0 = const()[name = string("op_24510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24510_cast_fp16 = einsum(equation = var_24510_equation_0, values = (var_23956_cast_fp16, var_24358_cast_fp16))[name = string("op_24510_cast_fp16")]; + string var_24512_equation_0 = const()[name = string("op_24512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24512_cast_fp16 = einsum(equation = var_24512_equation_0, values = (var_23960_cast_fp16, var_24359_cast_fp16))[name = string("op_24512_cast_fp16")]; + string var_24514_equation_0 = const()[name = string("op_24514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24514_cast_fp16 = einsum(equation = var_24514_equation_0, values = (var_23960_cast_fp16, var_24360_cast_fp16))[name = string("op_24514_cast_fp16")]; + string var_24516_equation_0 = const()[name = string("op_24516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24516_cast_fp16 = einsum(equation = var_24516_equation_0, values = (var_23960_cast_fp16, var_24361_cast_fp16))[name = string("op_24516_cast_fp16")]; + string var_24518_equation_0 = const()[name = string("op_24518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24518_cast_fp16 = einsum(equation = var_24518_equation_0, values = (var_23960_cast_fp16, var_24362_cast_fp16))[name = string("op_24518_cast_fp16")]; + string var_24520_equation_0 = const()[name = string("op_24520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24520_cast_fp16 = einsum(equation = var_24520_equation_0, values = (var_23964_cast_fp16, var_24363_cast_fp16))[name = string("op_24520_cast_fp16")]; + string var_24522_equation_0 = const()[name = string("op_24522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24522_cast_fp16 = einsum(equation = var_24522_equation_0, values = (var_23964_cast_fp16, var_24364_cast_fp16))[name = string("op_24522_cast_fp16")]; + string var_24524_equation_0 = const()[name = string("op_24524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24524_cast_fp16 = einsum(equation = var_24524_equation_0, values = (var_23964_cast_fp16, var_24365_cast_fp16))[name = string("op_24524_cast_fp16")]; + string var_24526_equation_0 = const()[name = string("op_24526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_24526_cast_fp16 = einsum(equation = var_24526_equation_0, values = (var_23964_cast_fp16, var_24366_cast_fp16))[name = string("op_24526_cast_fp16")]; + bool var_24528_interleave_0 = const()[name = string("op_24528_interleave_0"), val = bool(false)]; + tensor var_24528_cast_fp16 = concat(axis = var_23087, interleave = var_24528_interleave_0, values = (var_24368_cast_fp16, var_24370_cast_fp16, var_24372_cast_fp16, var_24374_cast_fp16))[name = string("op_24528_cast_fp16")]; + bool var_24530_interleave_0 = const()[name = string("op_24530_interleave_0"), val = bool(false)]; + tensor var_24530_cast_fp16 = concat(axis = var_23087, interleave = var_24530_interleave_0, values = (var_24376_cast_fp16, var_24378_cast_fp16, var_24380_cast_fp16, var_24382_cast_fp16))[name = string("op_24530_cast_fp16")]; + bool var_24532_interleave_0 = const()[name = string("op_24532_interleave_0"), val = bool(false)]; + tensor var_24532_cast_fp16 = concat(axis = var_23087, interleave = var_24532_interleave_0, values = (var_24384_cast_fp16, var_24386_cast_fp16, var_24388_cast_fp16, var_24390_cast_fp16))[name = string("op_24532_cast_fp16")]; + bool var_24534_interleave_0 = const()[name = string("op_24534_interleave_0"), val = bool(false)]; + tensor var_24534_cast_fp16 = concat(axis = var_23087, interleave = var_24534_interleave_0, values = (var_24392_cast_fp16, var_24394_cast_fp16, var_24396_cast_fp16, var_24398_cast_fp16))[name = string("op_24534_cast_fp16")]; + bool var_24536_interleave_0 = const()[name = string("op_24536_interleave_0"), val = bool(false)]; + tensor var_24536_cast_fp16 = concat(axis = var_23087, interleave = var_24536_interleave_0, values = (var_24400_cast_fp16, var_24402_cast_fp16, var_24404_cast_fp16, var_24406_cast_fp16))[name = string("op_24536_cast_fp16")]; + bool var_24538_interleave_0 = const()[name = string("op_24538_interleave_0"), val = bool(false)]; + tensor var_24538_cast_fp16 = concat(axis = var_23087, interleave = var_24538_interleave_0, values = (var_24408_cast_fp16, var_24410_cast_fp16, var_24412_cast_fp16, var_24414_cast_fp16))[name = string("op_24538_cast_fp16")]; + bool var_24540_interleave_0 = const()[name = string("op_24540_interleave_0"), val = bool(false)]; + tensor var_24540_cast_fp16 = concat(axis = var_23087, interleave = var_24540_interleave_0, values = (var_24416_cast_fp16, var_24418_cast_fp16, var_24420_cast_fp16, var_24422_cast_fp16))[name = string("op_24540_cast_fp16")]; + bool var_24542_interleave_0 = const()[name = string("op_24542_interleave_0"), val = bool(false)]; + tensor var_24542_cast_fp16 = concat(axis = var_23087, interleave = var_24542_interleave_0, values = (var_24424_cast_fp16, var_24426_cast_fp16, var_24428_cast_fp16, var_24430_cast_fp16))[name = string("op_24542_cast_fp16")]; + bool var_24544_interleave_0 = const()[name = string("op_24544_interleave_0"), val = bool(false)]; + tensor var_24544_cast_fp16 = concat(axis = var_23087, interleave = var_24544_interleave_0, values = (var_24432_cast_fp16, var_24434_cast_fp16, var_24436_cast_fp16, var_24438_cast_fp16))[name = string("op_24544_cast_fp16")]; + bool var_24546_interleave_0 = const()[name = string("op_24546_interleave_0"), val = bool(false)]; + tensor var_24546_cast_fp16 = concat(axis = var_23087, interleave = var_24546_interleave_0, values = (var_24440_cast_fp16, var_24442_cast_fp16, var_24444_cast_fp16, var_24446_cast_fp16))[name = string("op_24546_cast_fp16")]; + bool var_24548_interleave_0 = const()[name = string("op_24548_interleave_0"), val = bool(false)]; + tensor var_24548_cast_fp16 = concat(axis = var_23087, interleave = var_24548_interleave_0, values = (var_24448_cast_fp16, var_24450_cast_fp16, var_24452_cast_fp16, var_24454_cast_fp16))[name = string("op_24548_cast_fp16")]; + bool var_24550_interleave_0 = const()[name = string("op_24550_interleave_0"), val = bool(false)]; + tensor var_24550_cast_fp16 = concat(axis = var_23087, interleave = var_24550_interleave_0, values = (var_24456_cast_fp16, var_24458_cast_fp16, var_24460_cast_fp16, var_24462_cast_fp16))[name = string("op_24550_cast_fp16")]; + bool var_24552_interleave_0 = const()[name = string("op_24552_interleave_0"), val = bool(false)]; + tensor var_24552_cast_fp16 = concat(axis = var_23087, interleave = var_24552_interleave_0, values = (var_24464_cast_fp16, var_24466_cast_fp16, var_24468_cast_fp16, var_24470_cast_fp16))[name = string("op_24552_cast_fp16")]; + bool var_24554_interleave_0 = const()[name = string("op_24554_interleave_0"), val = bool(false)]; + tensor var_24554_cast_fp16 = concat(axis = var_23087, interleave = var_24554_interleave_0, values = (var_24472_cast_fp16, var_24474_cast_fp16, var_24476_cast_fp16, var_24478_cast_fp16))[name = string("op_24554_cast_fp16")]; + bool var_24556_interleave_0 = const()[name = string("op_24556_interleave_0"), val = bool(false)]; + tensor var_24556_cast_fp16 = concat(axis = var_23087, interleave = var_24556_interleave_0, values = (var_24480_cast_fp16, var_24482_cast_fp16, var_24484_cast_fp16, var_24486_cast_fp16))[name = string("op_24556_cast_fp16")]; + bool var_24558_interleave_0 = const()[name = string("op_24558_interleave_0"), val = bool(false)]; + tensor var_24558_cast_fp16 = concat(axis = var_23087, interleave = var_24558_interleave_0, values = (var_24488_cast_fp16, var_24490_cast_fp16, var_24492_cast_fp16, var_24494_cast_fp16))[name = string("op_24558_cast_fp16")]; + bool var_24560_interleave_0 = const()[name = string("op_24560_interleave_0"), val = bool(false)]; + tensor var_24560_cast_fp16 = concat(axis = var_23087, interleave = var_24560_interleave_0, values = (var_24496_cast_fp16, var_24498_cast_fp16, var_24500_cast_fp16, var_24502_cast_fp16))[name = string("op_24560_cast_fp16")]; + bool var_24562_interleave_0 = const()[name = string("op_24562_interleave_0"), val = bool(false)]; + tensor var_24562_cast_fp16 = concat(axis = var_23087, interleave = var_24562_interleave_0, values = (var_24504_cast_fp16, var_24506_cast_fp16, var_24508_cast_fp16, var_24510_cast_fp16))[name = string("op_24562_cast_fp16")]; + bool var_24564_interleave_0 = const()[name = string("op_24564_interleave_0"), val = bool(false)]; + tensor var_24564_cast_fp16 = concat(axis = var_23087, interleave = var_24564_interleave_0, values = (var_24512_cast_fp16, var_24514_cast_fp16, var_24516_cast_fp16, var_24518_cast_fp16))[name = string("op_24564_cast_fp16")]; + bool var_24566_interleave_0 = const()[name = string("op_24566_interleave_0"), val = bool(false)]; + tensor var_24566_cast_fp16 = concat(axis = var_23087, interleave = var_24566_interleave_0, values = (var_24520_cast_fp16, var_24522_cast_fp16, var_24524_cast_fp16, var_24526_cast_fp16))[name = string("op_24566_cast_fp16")]; + bool input_121_interleave_0 = const()[name = string("input_121_interleave_0"), val = bool(false)]; + tensor input_121_cast_fp16 = concat(axis = var_23112, interleave = input_121_interleave_0, values = (var_24528_cast_fp16, var_24530_cast_fp16, var_24532_cast_fp16, var_24534_cast_fp16, var_24536_cast_fp16, var_24538_cast_fp16, var_24540_cast_fp16, var_24542_cast_fp16, var_24544_cast_fp16, var_24546_cast_fp16, var_24548_cast_fp16, var_24550_cast_fp16, var_24552_cast_fp16, var_24554_cast_fp16, var_24556_cast_fp16, var_24558_cast_fp16, var_24560_cast_fp16, var_24562_cast_fp16, var_24564_cast_fp16, var_24566_cast_fp16))[name = string("input_121_cast_fp16")]; + string obj_63_pad_type_0 = const()[name = string("obj_63_pad_type_0"), val = string("valid")]; + tensor obj_63_strides_0 = const()[name = string("obj_63_strides_0"), val = tensor([1, 1])]; + tensor obj_63_pad_0 = const()[name = string("obj_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_63_dilations_0 = const()[name = string("obj_63_dilations_0"), val = tensor([1, 1])]; + int32 obj_63_groups_0 = const()[name = string("obj_63_groups_0"), val = int32(1)]; + tensor layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614804480)))]; + tensor layers_15_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618081344)))]; + tensor obj_63_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_121_cast_fp16)[name = string("obj_63_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_63_cast_fp16)[name = string("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; + fp16 var_24585_to_fp16 = const()[name = string("op_24585_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_24585_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; + tensor input_123_gamma_0_to_fp16 = const()[name = string("input_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618083968)))]; + tensor input_123_beta_0_to_fp16 = const()[name = string("input_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618086592)))]; + fp16 input_123_epsilon_0_to_fp16 = const()[name = string("input_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_123_cast_fp16 = batch_norm(beta = input_123_beta_0_to_fp16, epsilon = input_123_epsilon_0_to_fp16, gamma = input_123_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("input_123_cast_fp16")]; + string input_125_pad_type_0 = const()[name = string("input_125_pad_type_0"), val = string("valid")]; + tensor input_125_strides_0 = const()[name = string("input_125_strides_0"), val = tensor([1, 1])]; + tensor input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_125_dilations_0 = const()[name = string("input_125_dilations_0"), val = tensor([1, 1])]; + int32 input_125_groups_0 = const()[name = string("input_125_groups_0"), val = int32(1)]; + tensor layers_15_fc1_weight_to_fp16 = const()[name = string("layers_15_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(618089216)))]; + tensor layers_15_fc1_bias_to_fp16 = const()[name = string("layers_15_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631196480)))]; + tensor input_125_cast_fp16 = conv(bias = layers_15_fc1_bias_to_fp16, dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = layers_15_fc1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")]; + string input_127_mode_0 = const()[name = string("input_127_mode_0"), val = string("EXACT")]; + tensor input_127_cast_fp16 = gelu(mode = input_127_mode_0, x = input_125_cast_fp16)[name = string("input_127_cast_fp16")]; + string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")]; + tensor hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)]; + tensor layers_15_fc2_weight_to_fp16 = const()[name = string("layers_15_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631206784)))]; + tensor layers_15_fc2_bias_to_fp16 = const()[name = string("layers_15_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644314048)))]; + tensor hidden_states_35_cast_fp16 = conv(bias = layers_15_fc2_bias_to_fp16, dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = layers_15_fc2_weight_to_fp16, x = input_127_cast_fp16)[name = string("hidden_states_35_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = hidden_states_35_cast_fp16)[name = string("inputs_65_cast_fp16")]; + int32 var_24614 = const()[name = string("op_24614"), val = int32(3)]; + int32 var_24639 = const()[name = string("op_24639"), val = int32(1)]; + tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; + fp16 var_24656_to_fp16 = const()[name = string("op_24656_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_24656_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; + tensor obj_65_gamma_0_to_fp16 = const()[name = string("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644316672)))]; + tensor obj_65_beta_0_to_fp16 = const()[name = string("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644319296)))]; + fp16 obj_65_epsilon_0_to_fp16 = const()[name = string("obj_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("obj_65_cast_fp16")]; + string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; + tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; + tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; + int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644321920)))]; + tensor layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647598784)))]; + tensor query_33_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("query_33_cast_fp16")]; + string key_33_pad_type_0 = const()[name = string("key_33_pad_type_0"), val = string("valid")]; + tensor key_33_strides_0 = const()[name = string("key_33_strides_0"), val = tensor([1, 1])]; + tensor key_33_pad_0 = const()[name = string("key_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_33_dilations_0 = const()[name = string("key_33_dilations_0"), val = tensor([1, 1])]; + int32 key_33_groups_0 = const()[name = string("key_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(647601408)))]; + tensor key_33_cast_fp16 = conv(dilations = key_33_dilations_0, groups = key_33_groups_0, pad = key_33_pad_0, pad_type = key_33_pad_type_0, strides = key_33_strides_0, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("key_33_cast_fp16")]; + string value_33_pad_type_0 = const()[name = string("value_33_pad_type_0"), val = string("valid")]; + tensor value_33_strides_0 = const()[name = string("value_33_strides_0"), val = tensor([1, 1])]; + tensor value_33_pad_0 = const()[name = string("value_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_33_dilations_0 = const()[name = string("value_33_dilations_0"), val = tensor([1, 1])]; + int32 value_33_groups_0 = const()[name = string("value_33_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(650878272)))]; + tensor layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654155136)))]; + tensor value_33_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_bias_to_fp16, dilations = value_33_dilations_0, groups = value_33_groups_0, pad = value_33_pad_0, pad_type = value_33_pad_type_0, strides = value_33_strides_0, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = string("value_33_cast_fp16")]; + tensor var_24694_begin_0 = const()[name = string("op_24694_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24694_end_0 = const()[name = string("op_24694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24694_end_mask_0 = const()[name = string("op_24694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24694_cast_fp16 = slice_by_index(begin = var_24694_begin_0, end = var_24694_end_0, end_mask = var_24694_end_mask_0, x = query_33_cast_fp16)[name = string("op_24694_cast_fp16")]; + tensor var_24698_begin_0 = const()[name = string("op_24698_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24698_end_0 = const()[name = string("op_24698_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_24698_end_mask_0 = const()[name = string("op_24698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24698_cast_fp16 = slice_by_index(begin = var_24698_begin_0, end = var_24698_end_0, end_mask = var_24698_end_mask_0, x = query_33_cast_fp16)[name = string("op_24698_cast_fp16")]; + tensor var_24702_begin_0 = const()[name = string("op_24702_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24702_end_0 = const()[name = string("op_24702_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_24702_end_mask_0 = const()[name = string("op_24702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24702_cast_fp16 = slice_by_index(begin = var_24702_begin_0, end = var_24702_end_0, end_mask = var_24702_end_mask_0, x = query_33_cast_fp16)[name = string("op_24702_cast_fp16")]; + tensor var_24706_begin_0 = const()[name = string("op_24706_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24706_end_0 = const()[name = string("op_24706_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_24706_end_mask_0 = const()[name = string("op_24706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24706_cast_fp16 = slice_by_index(begin = var_24706_begin_0, end = var_24706_end_0, end_mask = var_24706_end_mask_0, x = query_33_cast_fp16)[name = string("op_24706_cast_fp16")]; + tensor var_24710_begin_0 = const()[name = string("op_24710_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24710_end_0 = const()[name = string("op_24710_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_24710_end_mask_0 = const()[name = string("op_24710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24710_cast_fp16 = slice_by_index(begin = var_24710_begin_0, end = var_24710_end_0, end_mask = var_24710_end_mask_0, x = query_33_cast_fp16)[name = string("op_24710_cast_fp16")]; + tensor var_24714_begin_0 = const()[name = string("op_24714_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24714_end_0 = const()[name = string("op_24714_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_24714_end_mask_0 = const()[name = string("op_24714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24714_cast_fp16 = slice_by_index(begin = var_24714_begin_0, end = var_24714_end_0, end_mask = var_24714_end_mask_0, x = query_33_cast_fp16)[name = string("op_24714_cast_fp16")]; + tensor var_24718_begin_0 = const()[name = string("op_24718_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24718_end_0 = const()[name = string("op_24718_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_24718_end_mask_0 = const()[name = string("op_24718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24718_cast_fp16 = slice_by_index(begin = var_24718_begin_0, end = var_24718_end_0, end_mask = var_24718_end_mask_0, x = query_33_cast_fp16)[name = string("op_24718_cast_fp16")]; + tensor var_24722_begin_0 = const()[name = string("op_24722_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24722_end_0 = const()[name = string("op_24722_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_24722_end_mask_0 = const()[name = string("op_24722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24722_cast_fp16 = slice_by_index(begin = var_24722_begin_0, end = var_24722_end_0, end_mask = var_24722_end_mask_0, x = query_33_cast_fp16)[name = string("op_24722_cast_fp16")]; + tensor var_24726_begin_0 = const()[name = string("op_24726_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24726_end_0 = const()[name = string("op_24726_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_24726_end_mask_0 = const()[name = string("op_24726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24726_cast_fp16 = slice_by_index(begin = var_24726_begin_0, end = var_24726_end_0, end_mask = var_24726_end_mask_0, x = query_33_cast_fp16)[name = string("op_24726_cast_fp16")]; + tensor var_24730_begin_0 = const()[name = string("op_24730_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24730_end_0 = const()[name = string("op_24730_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_24730_end_mask_0 = const()[name = string("op_24730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24730_cast_fp16 = slice_by_index(begin = var_24730_begin_0, end = var_24730_end_0, end_mask = var_24730_end_mask_0, x = query_33_cast_fp16)[name = string("op_24730_cast_fp16")]; + tensor var_24734_begin_0 = const()[name = string("op_24734_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24734_end_0 = const()[name = string("op_24734_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_24734_end_mask_0 = const()[name = string("op_24734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24734_cast_fp16 = slice_by_index(begin = var_24734_begin_0, end = var_24734_end_0, end_mask = var_24734_end_mask_0, x = query_33_cast_fp16)[name = string("op_24734_cast_fp16")]; + tensor var_24738_begin_0 = const()[name = string("op_24738_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24738_end_0 = const()[name = string("op_24738_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_24738_end_mask_0 = const()[name = string("op_24738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24738_cast_fp16 = slice_by_index(begin = var_24738_begin_0, end = var_24738_end_0, end_mask = var_24738_end_mask_0, x = query_33_cast_fp16)[name = string("op_24738_cast_fp16")]; + tensor var_24742_begin_0 = const()[name = string("op_24742_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24742_end_0 = const()[name = string("op_24742_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_24742_end_mask_0 = const()[name = string("op_24742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24742_cast_fp16 = slice_by_index(begin = var_24742_begin_0, end = var_24742_end_0, end_mask = var_24742_end_mask_0, x = query_33_cast_fp16)[name = string("op_24742_cast_fp16")]; + tensor var_24746_begin_0 = const()[name = string("op_24746_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24746_end_0 = const()[name = string("op_24746_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_24746_end_mask_0 = const()[name = string("op_24746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24746_cast_fp16 = slice_by_index(begin = var_24746_begin_0, end = var_24746_end_0, end_mask = var_24746_end_mask_0, x = query_33_cast_fp16)[name = string("op_24746_cast_fp16")]; + tensor var_24750_begin_0 = const()[name = string("op_24750_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24750_end_0 = const()[name = string("op_24750_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_24750_end_mask_0 = const()[name = string("op_24750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24750_cast_fp16 = slice_by_index(begin = var_24750_begin_0, end = var_24750_end_0, end_mask = var_24750_end_mask_0, x = query_33_cast_fp16)[name = string("op_24750_cast_fp16")]; + tensor var_24754_begin_0 = const()[name = string("op_24754_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24754_end_0 = const()[name = string("op_24754_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_24754_end_mask_0 = const()[name = string("op_24754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24754_cast_fp16 = slice_by_index(begin = var_24754_begin_0, end = var_24754_end_0, end_mask = var_24754_end_mask_0, x = query_33_cast_fp16)[name = string("op_24754_cast_fp16")]; + tensor var_24758_begin_0 = const()[name = string("op_24758_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24758_end_0 = const()[name = string("op_24758_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_24758_end_mask_0 = const()[name = string("op_24758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24758_cast_fp16 = slice_by_index(begin = var_24758_begin_0, end = var_24758_end_0, end_mask = var_24758_end_mask_0, x = query_33_cast_fp16)[name = string("op_24758_cast_fp16")]; + tensor var_24762_begin_0 = const()[name = string("op_24762_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24762_end_0 = const()[name = string("op_24762_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_24762_end_mask_0 = const()[name = string("op_24762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24762_cast_fp16 = slice_by_index(begin = var_24762_begin_0, end = var_24762_end_0, end_mask = var_24762_end_mask_0, x = query_33_cast_fp16)[name = string("op_24762_cast_fp16")]; + tensor var_24766_begin_0 = const()[name = string("op_24766_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24766_end_0 = const()[name = string("op_24766_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_24766_end_mask_0 = const()[name = string("op_24766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24766_cast_fp16 = slice_by_index(begin = var_24766_begin_0, end = var_24766_end_0, end_mask = var_24766_end_mask_0, x = query_33_cast_fp16)[name = string("op_24766_cast_fp16")]; + tensor var_24770_begin_0 = const()[name = string("op_24770_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24770_end_0 = const()[name = string("op_24770_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_24770_end_mask_0 = const()[name = string("op_24770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24770_cast_fp16 = slice_by_index(begin = var_24770_begin_0, end = var_24770_end_0, end_mask = var_24770_end_mask_0, x = query_33_cast_fp16)[name = string("op_24770_cast_fp16")]; + tensor var_24779_begin_0 = const()[name = string("op_24779_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24779_end_0 = const()[name = string("op_24779_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24779_end_mask_0 = const()[name = string("op_24779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24779_cast_fp16 = slice_by_index(begin = var_24779_begin_0, end = var_24779_end_0, end_mask = var_24779_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24779_cast_fp16")]; + tensor var_24786_begin_0 = const()[name = string("op_24786_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24786_end_0 = const()[name = string("op_24786_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24786_end_mask_0 = const()[name = string("op_24786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24786_cast_fp16 = slice_by_index(begin = var_24786_begin_0, end = var_24786_end_0, end_mask = var_24786_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24786_cast_fp16")]; + tensor var_24793_begin_0 = const()[name = string("op_24793_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24793_end_0 = const()[name = string("op_24793_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24793_end_mask_0 = const()[name = string("op_24793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24793_cast_fp16 = slice_by_index(begin = var_24793_begin_0, end = var_24793_end_0, end_mask = var_24793_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24793_cast_fp16")]; + tensor var_24800_begin_0 = const()[name = string("op_24800_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24800_end_0 = const()[name = string("op_24800_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24800_end_mask_0 = const()[name = string("op_24800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24800_cast_fp16 = slice_by_index(begin = var_24800_begin_0, end = var_24800_end_0, end_mask = var_24800_end_mask_0, x = var_24694_cast_fp16)[name = string("op_24800_cast_fp16")]; + tensor var_24807_begin_0 = const()[name = string("op_24807_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24807_end_0 = const()[name = string("op_24807_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24807_end_mask_0 = const()[name = string("op_24807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24807_cast_fp16 = slice_by_index(begin = var_24807_begin_0, end = var_24807_end_0, end_mask = var_24807_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24807_cast_fp16")]; + tensor var_24814_begin_0 = const()[name = string("op_24814_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24814_end_0 = const()[name = string("op_24814_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24814_end_mask_0 = const()[name = string("op_24814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24814_cast_fp16 = slice_by_index(begin = var_24814_begin_0, end = var_24814_end_0, end_mask = var_24814_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24814_cast_fp16")]; + tensor var_24821_begin_0 = const()[name = string("op_24821_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24821_end_0 = const()[name = string("op_24821_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24821_end_mask_0 = const()[name = string("op_24821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24821_cast_fp16 = slice_by_index(begin = var_24821_begin_0, end = var_24821_end_0, end_mask = var_24821_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24821_cast_fp16")]; + tensor var_24828_begin_0 = const()[name = string("op_24828_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24828_end_0 = const()[name = string("op_24828_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24828_end_mask_0 = const()[name = string("op_24828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24828_cast_fp16 = slice_by_index(begin = var_24828_begin_0, end = var_24828_end_0, end_mask = var_24828_end_mask_0, x = var_24698_cast_fp16)[name = string("op_24828_cast_fp16")]; + tensor var_24835_begin_0 = const()[name = string("op_24835_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24835_end_0 = const()[name = string("op_24835_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24835_end_mask_0 = const()[name = string("op_24835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24835_cast_fp16 = slice_by_index(begin = var_24835_begin_0, end = var_24835_end_0, end_mask = var_24835_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24835_cast_fp16")]; + tensor var_24842_begin_0 = const()[name = string("op_24842_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24842_end_0 = const()[name = string("op_24842_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24842_end_mask_0 = const()[name = string("op_24842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24842_cast_fp16 = slice_by_index(begin = var_24842_begin_0, end = var_24842_end_0, end_mask = var_24842_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24842_cast_fp16")]; + tensor var_24849_begin_0 = const()[name = string("op_24849_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24849_end_0 = const()[name = string("op_24849_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24849_end_mask_0 = const()[name = string("op_24849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24849_cast_fp16 = slice_by_index(begin = var_24849_begin_0, end = var_24849_end_0, end_mask = var_24849_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24849_cast_fp16")]; + tensor var_24856_begin_0 = const()[name = string("op_24856_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24856_end_0 = const()[name = string("op_24856_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24856_end_mask_0 = const()[name = string("op_24856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24856_cast_fp16 = slice_by_index(begin = var_24856_begin_0, end = var_24856_end_0, end_mask = var_24856_end_mask_0, x = var_24702_cast_fp16)[name = string("op_24856_cast_fp16")]; + tensor var_24863_begin_0 = const()[name = string("op_24863_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24863_end_0 = const()[name = string("op_24863_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24863_end_mask_0 = const()[name = string("op_24863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24863_cast_fp16 = slice_by_index(begin = var_24863_begin_0, end = var_24863_end_0, end_mask = var_24863_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24863_cast_fp16")]; + tensor var_24870_begin_0 = const()[name = string("op_24870_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24870_end_0 = const()[name = string("op_24870_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24870_end_mask_0 = const()[name = string("op_24870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24870_cast_fp16 = slice_by_index(begin = var_24870_begin_0, end = var_24870_end_0, end_mask = var_24870_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24870_cast_fp16")]; + tensor var_24877_begin_0 = const()[name = string("op_24877_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24877_end_0 = const()[name = string("op_24877_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24877_end_mask_0 = const()[name = string("op_24877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24877_cast_fp16 = slice_by_index(begin = var_24877_begin_0, end = var_24877_end_0, end_mask = var_24877_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24877_cast_fp16")]; + tensor var_24884_begin_0 = const()[name = string("op_24884_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24884_end_0 = const()[name = string("op_24884_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24884_end_mask_0 = const()[name = string("op_24884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24884_cast_fp16 = slice_by_index(begin = var_24884_begin_0, end = var_24884_end_0, end_mask = var_24884_end_mask_0, x = var_24706_cast_fp16)[name = string("op_24884_cast_fp16")]; + tensor var_24891_begin_0 = const()[name = string("op_24891_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24891_end_0 = const()[name = string("op_24891_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24891_end_mask_0 = const()[name = string("op_24891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24891_cast_fp16 = slice_by_index(begin = var_24891_begin_0, end = var_24891_end_0, end_mask = var_24891_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24891_cast_fp16")]; + tensor var_24898_begin_0 = const()[name = string("op_24898_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24898_end_0 = const()[name = string("op_24898_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24898_end_mask_0 = const()[name = string("op_24898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24898_cast_fp16 = slice_by_index(begin = var_24898_begin_0, end = var_24898_end_0, end_mask = var_24898_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24898_cast_fp16")]; + tensor var_24905_begin_0 = const()[name = string("op_24905_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24905_end_0 = const()[name = string("op_24905_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24905_end_mask_0 = const()[name = string("op_24905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24905_cast_fp16 = slice_by_index(begin = var_24905_begin_0, end = var_24905_end_0, end_mask = var_24905_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24905_cast_fp16")]; + tensor var_24912_begin_0 = const()[name = string("op_24912_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24912_end_0 = const()[name = string("op_24912_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24912_end_mask_0 = const()[name = string("op_24912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24912_cast_fp16 = slice_by_index(begin = var_24912_begin_0, end = var_24912_end_0, end_mask = var_24912_end_mask_0, x = var_24710_cast_fp16)[name = string("op_24912_cast_fp16")]; + tensor var_24919_begin_0 = const()[name = string("op_24919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24919_end_0 = const()[name = string("op_24919_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24919_end_mask_0 = const()[name = string("op_24919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24919_cast_fp16 = slice_by_index(begin = var_24919_begin_0, end = var_24919_end_0, end_mask = var_24919_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24919_cast_fp16")]; + tensor var_24926_begin_0 = const()[name = string("op_24926_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24926_end_0 = const()[name = string("op_24926_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24926_end_mask_0 = const()[name = string("op_24926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24926_cast_fp16 = slice_by_index(begin = var_24926_begin_0, end = var_24926_end_0, end_mask = var_24926_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24926_cast_fp16")]; + tensor var_24933_begin_0 = const()[name = string("op_24933_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24933_end_0 = const()[name = string("op_24933_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24933_end_mask_0 = const()[name = string("op_24933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24933_cast_fp16 = slice_by_index(begin = var_24933_begin_0, end = var_24933_end_0, end_mask = var_24933_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24933_cast_fp16")]; + tensor var_24940_begin_0 = const()[name = string("op_24940_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24940_end_0 = const()[name = string("op_24940_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24940_end_mask_0 = const()[name = string("op_24940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24940_cast_fp16 = slice_by_index(begin = var_24940_begin_0, end = var_24940_end_0, end_mask = var_24940_end_mask_0, x = var_24714_cast_fp16)[name = string("op_24940_cast_fp16")]; + tensor var_24947_begin_0 = const()[name = string("op_24947_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24947_end_0 = const()[name = string("op_24947_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24947_end_mask_0 = const()[name = string("op_24947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24947_cast_fp16 = slice_by_index(begin = var_24947_begin_0, end = var_24947_end_0, end_mask = var_24947_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24947_cast_fp16")]; + tensor var_24954_begin_0 = const()[name = string("op_24954_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24954_end_0 = const()[name = string("op_24954_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24954_end_mask_0 = const()[name = string("op_24954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24954_cast_fp16 = slice_by_index(begin = var_24954_begin_0, end = var_24954_end_0, end_mask = var_24954_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24954_cast_fp16")]; + tensor var_24961_begin_0 = const()[name = string("op_24961_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24961_end_0 = const()[name = string("op_24961_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24961_end_mask_0 = const()[name = string("op_24961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24961_cast_fp16 = slice_by_index(begin = var_24961_begin_0, end = var_24961_end_0, end_mask = var_24961_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24961_cast_fp16")]; + tensor var_24968_begin_0 = const()[name = string("op_24968_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24968_end_0 = const()[name = string("op_24968_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24968_end_mask_0 = const()[name = string("op_24968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24968_cast_fp16 = slice_by_index(begin = var_24968_begin_0, end = var_24968_end_0, end_mask = var_24968_end_mask_0, x = var_24718_cast_fp16)[name = string("op_24968_cast_fp16")]; + tensor var_24975_begin_0 = const()[name = string("op_24975_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24975_end_0 = const()[name = string("op_24975_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24975_end_mask_0 = const()[name = string("op_24975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24975_cast_fp16 = slice_by_index(begin = var_24975_begin_0, end = var_24975_end_0, end_mask = var_24975_end_mask_0, x = var_24722_cast_fp16)[name = string("op_24975_cast_fp16")]; + tensor var_24982_begin_0 = const()[name = string("op_24982_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24982_end_0 = const()[name = string("op_24982_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24982_end_mask_0 = const()[name = string("op_24982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24982_cast_fp16 = slice_by_index(begin = var_24982_begin_0, end = var_24982_end_0, end_mask = var_24982_end_mask_0, x = var_24722_cast_fp16)[name = string("op_24982_cast_fp16")]; + tensor var_24989_begin_0 = const()[name = string("op_24989_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24989_end_0 = const()[name = string("op_24989_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24989_end_mask_0 = const()[name = string("op_24989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24989_cast_fp16 = slice_by_index(begin = var_24989_begin_0, end = var_24989_end_0, end_mask = var_24989_end_mask_0, x = var_24722_cast_fp16)[name = string("op_24989_cast_fp16")]; + tensor var_24996_begin_0 = const()[name = string("op_24996_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24996_end_0 = const()[name = string("op_24996_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24996_end_mask_0 = const()[name = string("op_24996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24996_cast_fp16 = slice_by_index(begin = var_24996_begin_0, end = var_24996_end_0, end_mask = var_24996_end_mask_0, x = var_24722_cast_fp16)[name = string("op_24996_cast_fp16")]; + tensor var_25003_begin_0 = const()[name = string("op_25003_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25003_end_0 = const()[name = string("op_25003_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25003_end_mask_0 = const()[name = string("op_25003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25003_cast_fp16 = slice_by_index(begin = var_25003_begin_0, end = var_25003_end_0, end_mask = var_25003_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25003_cast_fp16")]; + tensor var_25010_begin_0 = const()[name = string("op_25010_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25010_end_0 = const()[name = string("op_25010_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25010_end_mask_0 = const()[name = string("op_25010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25010_cast_fp16 = slice_by_index(begin = var_25010_begin_0, end = var_25010_end_0, end_mask = var_25010_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25010_cast_fp16")]; + tensor var_25017_begin_0 = const()[name = string("op_25017_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25017_end_0 = const()[name = string("op_25017_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25017_end_mask_0 = const()[name = string("op_25017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25017_cast_fp16 = slice_by_index(begin = var_25017_begin_0, end = var_25017_end_0, end_mask = var_25017_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25017_cast_fp16")]; + tensor var_25024_begin_0 = const()[name = string("op_25024_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25024_end_0 = const()[name = string("op_25024_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25024_end_mask_0 = const()[name = string("op_25024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25024_cast_fp16 = slice_by_index(begin = var_25024_begin_0, end = var_25024_end_0, end_mask = var_25024_end_mask_0, x = var_24726_cast_fp16)[name = string("op_25024_cast_fp16")]; + tensor var_25031_begin_0 = const()[name = string("op_25031_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25031_end_0 = const()[name = string("op_25031_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25031_end_mask_0 = const()[name = string("op_25031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25031_cast_fp16 = slice_by_index(begin = var_25031_begin_0, end = var_25031_end_0, end_mask = var_25031_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25031_cast_fp16")]; + tensor var_25038_begin_0 = const()[name = string("op_25038_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25038_end_0 = const()[name = string("op_25038_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25038_end_mask_0 = const()[name = string("op_25038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25038_cast_fp16 = slice_by_index(begin = var_25038_begin_0, end = var_25038_end_0, end_mask = var_25038_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25038_cast_fp16")]; + tensor var_25045_begin_0 = const()[name = string("op_25045_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25045_end_0 = const()[name = string("op_25045_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25045_end_mask_0 = const()[name = string("op_25045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25045_cast_fp16 = slice_by_index(begin = var_25045_begin_0, end = var_25045_end_0, end_mask = var_25045_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25045_cast_fp16")]; + tensor var_25052_begin_0 = const()[name = string("op_25052_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25052_end_0 = const()[name = string("op_25052_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25052_end_mask_0 = const()[name = string("op_25052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25052_cast_fp16 = slice_by_index(begin = var_25052_begin_0, end = var_25052_end_0, end_mask = var_25052_end_mask_0, x = var_24730_cast_fp16)[name = string("op_25052_cast_fp16")]; + tensor var_25059_begin_0 = const()[name = string("op_25059_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25059_end_0 = const()[name = string("op_25059_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25059_end_mask_0 = const()[name = string("op_25059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25059_cast_fp16 = slice_by_index(begin = var_25059_begin_0, end = var_25059_end_0, end_mask = var_25059_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25059_cast_fp16")]; + tensor var_25066_begin_0 = const()[name = string("op_25066_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25066_end_0 = const()[name = string("op_25066_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25066_end_mask_0 = const()[name = string("op_25066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25066_cast_fp16 = slice_by_index(begin = var_25066_begin_0, end = var_25066_end_0, end_mask = var_25066_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25066_cast_fp16")]; + tensor var_25073_begin_0 = const()[name = string("op_25073_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25073_end_0 = const()[name = string("op_25073_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25073_end_mask_0 = const()[name = string("op_25073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25073_cast_fp16 = slice_by_index(begin = var_25073_begin_0, end = var_25073_end_0, end_mask = var_25073_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25073_cast_fp16")]; + tensor var_25080_begin_0 = const()[name = string("op_25080_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25080_end_0 = const()[name = string("op_25080_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25080_end_mask_0 = const()[name = string("op_25080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25080_cast_fp16 = slice_by_index(begin = var_25080_begin_0, end = var_25080_end_0, end_mask = var_25080_end_mask_0, x = var_24734_cast_fp16)[name = string("op_25080_cast_fp16")]; + tensor var_25087_begin_0 = const()[name = string("op_25087_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25087_end_0 = const()[name = string("op_25087_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25087_end_mask_0 = const()[name = string("op_25087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25087_cast_fp16 = slice_by_index(begin = var_25087_begin_0, end = var_25087_end_0, end_mask = var_25087_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25087_cast_fp16")]; + tensor var_25094_begin_0 = const()[name = string("op_25094_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25094_end_0 = const()[name = string("op_25094_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25094_end_mask_0 = const()[name = string("op_25094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25094_cast_fp16 = slice_by_index(begin = var_25094_begin_0, end = var_25094_end_0, end_mask = var_25094_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25094_cast_fp16")]; + tensor var_25101_begin_0 = const()[name = string("op_25101_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25101_end_0 = const()[name = string("op_25101_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25101_end_mask_0 = const()[name = string("op_25101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25101_cast_fp16 = slice_by_index(begin = var_25101_begin_0, end = var_25101_end_0, end_mask = var_25101_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25101_cast_fp16")]; + tensor var_25108_begin_0 = const()[name = string("op_25108_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25108_end_0 = const()[name = string("op_25108_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25108_end_mask_0 = const()[name = string("op_25108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25108_cast_fp16 = slice_by_index(begin = var_25108_begin_0, end = var_25108_end_0, end_mask = var_25108_end_mask_0, x = var_24738_cast_fp16)[name = string("op_25108_cast_fp16")]; + tensor var_25115_begin_0 = const()[name = string("op_25115_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25115_end_0 = const()[name = string("op_25115_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25115_end_mask_0 = const()[name = string("op_25115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25115_cast_fp16 = slice_by_index(begin = var_25115_begin_0, end = var_25115_end_0, end_mask = var_25115_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25115_cast_fp16")]; + tensor var_25122_begin_0 = const()[name = string("op_25122_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25122_end_0 = const()[name = string("op_25122_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25122_end_mask_0 = const()[name = string("op_25122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25122_cast_fp16 = slice_by_index(begin = var_25122_begin_0, end = var_25122_end_0, end_mask = var_25122_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25122_cast_fp16")]; + tensor var_25129_begin_0 = const()[name = string("op_25129_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25129_end_0 = const()[name = string("op_25129_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25129_end_mask_0 = const()[name = string("op_25129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25129_cast_fp16 = slice_by_index(begin = var_25129_begin_0, end = var_25129_end_0, end_mask = var_25129_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25129_cast_fp16")]; + tensor var_25136_begin_0 = const()[name = string("op_25136_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25136_end_0 = const()[name = string("op_25136_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25136_end_mask_0 = const()[name = string("op_25136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25136_cast_fp16 = slice_by_index(begin = var_25136_begin_0, end = var_25136_end_0, end_mask = var_25136_end_mask_0, x = var_24742_cast_fp16)[name = string("op_25136_cast_fp16")]; + tensor var_25143_begin_0 = const()[name = string("op_25143_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25143_end_0 = const()[name = string("op_25143_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25143_end_mask_0 = const()[name = string("op_25143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25143_cast_fp16 = slice_by_index(begin = var_25143_begin_0, end = var_25143_end_0, end_mask = var_25143_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25143_cast_fp16")]; + tensor var_25150_begin_0 = const()[name = string("op_25150_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25150_end_0 = const()[name = string("op_25150_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25150_end_mask_0 = const()[name = string("op_25150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25150_cast_fp16 = slice_by_index(begin = var_25150_begin_0, end = var_25150_end_0, end_mask = var_25150_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25150_cast_fp16")]; + tensor var_25157_begin_0 = const()[name = string("op_25157_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25157_end_0 = const()[name = string("op_25157_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25157_end_mask_0 = const()[name = string("op_25157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25157_cast_fp16 = slice_by_index(begin = var_25157_begin_0, end = var_25157_end_0, end_mask = var_25157_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25157_cast_fp16")]; + tensor var_25164_begin_0 = const()[name = string("op_25164_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25164_end_0 = const()[name = string("op_25164_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25164_end_mask_0 = const()[name = string("op_25164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25164_cast_fp16 = slice_by_index(begin = var_25164_begin_0, end = var_25164_end_0, end_mask = var_25164_end_mask_0, x = var_24746_cast_fp16)[name = string("op_25164_cast_fp16")]; + tensor var_25171_begin_0 = const()[name = string("op_25171_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25171_end_0 = const()[name = string("op_25171_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25171_end_mask_0 = const()[name = string("op_25171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25171_cast_fp16 = slice_by_index(begin = var_25171_begin_0, end = var_25171_end_0, end_mask = var_25171_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25171_cast_fp16")]; + tensor var_25178_begin_0 = const()[name = string("op_25178_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25178_end_0 = const()[name = string("op_25178_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25178_end_mask_0 = const()[name = string("op_25178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25178_cast_fp16 = slice_by_index(begin = var_25178_begin_0, end = var_25178_end_0, end_mask = var_25178_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25178_cast_fp16")]; + tensor var_25185_begin_0 = const()[name = string("op_25185_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25185_end_0 = const()[name = string("op_25185_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25185_end_mask_0 = const()[name = string("op_25185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25185_cast_fp16 = slice_by_index(begin = var_25185_begin_0, end = var_25185_end_0, end_mask = var_25185_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25185_cast_fp16")]; + tensor var_25192_begin_0 = const()[name = string("op_25192_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25192_end_0 = const()[name = string("op_25192_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25192_end_mask_0 = const()[name = string("op_25192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25192_cast_fp16 = slice_by_index(begin = var_25192_begin_0, end = var_25192_end_0, end_mask = var_25192_end_mask_0, x = var_24750_cast_fp16)[name = string("op_25192_cast_fp16")]; + tensor var_25199_begin_0 = const()[name = string("op_25199_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25199_end_0 = const()[name = string("op_25199_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25199_end_mask_0 = const()[name = string("op_25199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25199_cast_fp16 = slice_by_index(begin = var_25199_begin_0, end = var_25199_end_0, end_mask = var_25199_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25199_cast_fp16")]; + tensor var_25206_begin_0 = const()[name = string("op_25206_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25206_end_0 = const()[name = string("op_25206_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25206_end_mask_0 = const()[name = string("op_25206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25206_cast_fp16 = slice_by_index(begin = var_25206_begin_0, end = var_25206_end_0, end_mask = var_25206_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25206_cast_fp16")]; + tensor var_25213_begin_0 = const()[name = string("op_25213_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25213_end_0 = const()[name = string("op_25213_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25213_end_mask_0 = const()[name = string("op_25213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25213_cast_fp16 = slice_by_index(begin = var_25213_begin_0, end = var_25213_end_0, end_mask = var_25213_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25213_cast_fp16")]; + tensor var_25220_begin_0 = const()[name = string("op_25220_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25220_end_0 = const()[name = string("op_25220_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25220_end_mask_0 = const()[name = string("op_25220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25220_cast_fp16 = slice_by_index(begin = var_25220_begin_0, end = var_25220_end_0, end_mask = var_25220_end_mask_0, x = var_24754_cast_fp16)[name = string("op_25220_cast_fp16")]; + tensor var_25227_begin_0 = const()[name = string("op_25227_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25227_end_0 = const()[name = string("op_25227_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25227_end_mask_0 = const()[name = string("op_25227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25227_cast_fp16 = slice_by_index(begin = var_25227_begin_0, end = var_25227_end_0, end_mask = var_25227_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25227_cast_fp16")]; + tensor var_25234_begin_0 = const()[name = string("op_25234_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25234_end_0 = const()[name = string("op_25234_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25234_end_mask_0 = const()[name = string("op_25234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25234_cast_fp16 = slice_by_index(begin = var_25234_begin_0, end = var_25234_end_0, end_mask = var_25234_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25234_cast_fp16")]; + tensor var_25241_begin_0 = const()[name = string("op_25241_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25241_end_0 = const()[name = string("op_25241_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25241_end_mask_0 = const()[name = string("op_25241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25241_cast_fp16 = slice_by_index(begin = var_25241_begin_0, end = var_25241_end_0, end_mask = var_25241_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25241_cast_fp16")]; + tensor var_25248_begin_0 = const()[name = string("op_25248_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25248_end_0 = const()[name = string("op_25248_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25248_end_mask_0 = const()[name = string("op_25248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25248_cast_fp16 = slice_by_index(begin = var_25248_begin_0, end = var_25248_end_0, end_mask = var_25248_end_mask_0, x = var_24758_cast_fp16)[name = string("op_25248_cast_fp16")]; + tensor var_25255_begin_0 = const()[name = string("op_25255_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25255_end_0 = const()[name = string("op_25255_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25255_end_mask_0 = const()[name = string("op_25255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25255_cast_fp16 = slice_by_index(begin = var_25255_begin_0, end = var_25255_end_0, end_mask = var_25255_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25255_cast_fp16")]; + tensor var_25262_begin_0 = const()[name = string("op_25262_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25262_end_0 = const()[name = string("op_25262_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25262_end_mask_0 = const()[name = string("op_25262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25262_cast_fp16 = slice_by_index(begin = var_25262_begin_0, end = var_25262_end_0, end_mask = var_25262_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25262_cast_fp16")]; + tensor var_25269_begin_0 = const()[name = string("op_25269_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25269_end_0 = const()[name = string("op_25269_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25269_end_mask_0 = const()[name = string("op_25269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25269_cast_fp16 = slice_by_index(begin = var_25269_begin_0, end = var_25269_end_0, end_mask = var_25269_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25269_cast_fp16")]; + tensor var_25276_begin_0 = const()[name = string("op_25276_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25276_end_0 = const()[name = string("op_25276_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25276_end_mask_0 = const()[name = string("op_25276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25276_cast_fp16 = slice_by_index(begin = var_25276_begin_0, end = var_25276_end_0, end_mask = var_25276_end_mask_0, x = var_24762_cast_fp16)[name = string("op_25276_cast_fp16")]; + tensor var_25283_begin_0 = const()[name = string("op_25283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25283_end_0 = const()[name = string("op_25283_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25283_end_mask_0 = const()[name = string("op_25283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25283_cast_fp16 = slice_by_index(begin = var_25283_begin_0, end = var_25283_end_0, end_mask = var_25283_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25283_cast_fp16")]; + tensor var_25290_begin_0 = const()[name = string("op_25290_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25290_end_0 = const()[name = string("op_25290_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25290_end_mask_0 = const()[name = string("op_25290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25290_cast_fp16 = slice_by_index(begin = var_25290_begin_0, end = var_25290_end_0, end_mask = var_25290_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25290_cast_fp16")]; + tensor var_25297_begin_0 = const()[name = string("op_25297_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25297_end_0 = const()[name = string("op_25297_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25297_end_mask_0 = const()[name = string("op_25297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25297_cast_fp16 = slice_by_index(begin = var_25297_begin_0, end = var_25297_end_0, end_mask = var_25297_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25297_cast_fp16")]; + tensor var_25304_begin_0 = const()[name = string("op_25304_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25304_end_0 = const()[name = string("op_25304_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25304_end_mask_0 = const()[name = string("op_25304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25304_cast_fp16 = slice_by_index(begin = var_25304_begin_0, end = var_25304_end_0, end_mask = var_25304_end_mask_0, x = var_24766_cast_fp16)[name = string("op_25304_cast_fp16")]; + tensor var_25311_begin_0 = const()[name = string("op_25311_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25311_end_0 = const()[name = string("op_25311_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25311_end_mask_0 = const()[name = string("op_25311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25311_cast_fp16 = slice_by_index(begin = var_25311_begin_0, end = var_25311_end_0, end_mask = var_25311_end_mask_0, x = var_24770_cast_fp16)[name = string("op_25311_cast_fp16")]; + tensor var_25318_begin_0 = const()[name = string("op_25318_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25318_end_0 = const()[name = string("op_25318_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25318_end_mask_0 = const()[name = string("op_25318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25318_cast_fp16 = slice_by_index(begin = var_25318_begin_0, end = var_25318_end_0, end_mask = var_25318_end_mask_0, x = var_24770_cast_fp16)[name = string("op_25318_cast_fp16")]; + tensor var_25325_begin_0 = const()[name = string("op_25325_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25325_end_0 = const()[name = string("op_25325_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25325_end_mask_0 = const()[name = string("op_25325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25325_cast_fp16 = slice_by_index(begin = var_25325_begin_0, end = var_25325_end_0, end_mask = var_25325_end_mask_0, x = var_24770_cast_fp16)[name = string("op_25325_cast_fp16")]; + tensor var_25332_begin_0 = const()[name = string("op_25332_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25332_end_0 = const()[name = string("op_25332_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25332_end_mask_0 = const()[name = string("op_25332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25332_cast_fp16 = slice_by_index(begin = var_25332_begin_0, end = var_25332_end_0, end_mask = var_25332_end_mask_0, x = var_24770_cast_fp16)[name = string("op_25332_cast_fp16")]; + tensor k_33_perm_0 = const()[name = string("k_33_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25337_begin_0 = const()[name = string("op_25337_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25337_end_0 = const()[name = string("op_25337_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_25337_end_mask_0 = const()[name = string("op_25337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_33_cast_fp16 = transpose(perm = k_33_perm_0, x = key_33_cast_fp16)[name = string("transpose_15")]; + tensor var_25337_cast_fp16 = slice_by_index(begin = var_25337_begin_0, end = var_25337_end_0, end_mask = var_25337_end_mask_0, x = k_33_cast_fp16)[name = string("op_25337_cast_fp16")]; + tensor var_25341_begin_0 = const()[name = string("op_25341_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25341_end_0 = const()[name = string("op_25341_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_25341_end_mask_0 = const()[name = string("op_25341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25341_cast_fp16 = slice_by_index(begin = var_25341_begin_0, end = var_25341_end_0, end_mask = var_25341_end_mask_0, x = k_33_cast_fp16)[name = string("op_25341_cast_fp16")]; + tensor var_25345_begin_0 = const()[name = string("op_25345_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25345_end_0 = const()[name = string("op_25345_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_25345_end_mask_0 = const()[name = string("op_25345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25345_cast_fp16 = slice_by_index(begin = var_25345_begin_0, end = var_25345_end_0, end_mask = var_25345_end_mask_0, x = k_33_cast_fp16)[name = string("op_25345_cast_fp16")]; + tensor var_25349_begin_0 = const()[name = string("op_25349_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25349_end_0 = const()[name = string("op_25349_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_25349_end_mask_0 = const()[name = string("op_25349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25349_cast_fp16 = slice_by_index(begin = var_25349_begin_0, end = var_25349_end_0, end_mask = var_25349_end_mask_0, x = k_33_cast_fp16)[name = string("op_25349_cast_fp16")]; + tensor var_25353_begin_0 = const()[name = string("op_25353_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25353_end_0 = const()[name = string("op_25353_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_25353_end_mask_0 = const()[name = string("op_25353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25353_cast_fp16 = slice_by_index(begin = var_25353_begin_0, end = var_25353_end_0, end_mask = var_25353_end_mask_0, x = k_33_cast_fp16)[name = string("op_25353_cast_fp16")]; + tensor var_25357_begin_0 = const()[name = string("op_25357_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25357_end_0 = const()[name = string("op_25357_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_25357_end_mask_0 = const()[name = string("op_25357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25357_cast_fp16 = slice_by_index(begin = var_25357_begin_0, end = var_25357_end_0, end_mask = var_25357_end_mask_0, x = k_33_cast_fp16)[name = string("op_25357_cast_fp16")]; + tensor var_25361_begin_0 = const()[name = string("op_25361_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25361_end_0 = const()[name = string("op_25361_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_25361_end_mask_0 = const()[name = string("op_25361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25361_cast_fp16 = slice_by_index(begin = var_25361_begin_0, end = var_25361_end_0, end_mask = var_25361_end_mask_0, x = k_33_cast_fp16)[name = string("op_25361_cast_fp16")]; + tensor var_25365_begin_0 = const()[name = string("op_25365_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25365_end_0 = const()[name = string("op_25365_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_25365_end_mask_0 = const()[name = string("op_25365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25365_cast_fp16 = slice_by_index(begin = var_25365_begin_0, end = var_25365_end_0, end_mask = var_25365_end_mask_0, x = k_33_cast_fp16)[name = string("op_25365_cast_fp16")]; + tensor var_25369_begin_0 = const()[name = string("op_25369_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25369_end_0 = const()[name = string("op_25369_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_25369_end_mask_0 = const()[name = string("op_25369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25369_cast_fp16 = slice_by_index(begin = var_25369_begin_0, end = var_25369_end_0, end_mask = var_25369_end_mask_0, x = k_33_cast_fp16)[name = string("op_25369_cast_fp16")]; + tensor var_25373_begin_0 = const()[name = string("op_25373_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25373_end_0 = const()[name = string("op_25373_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_25373_end_mask_0 = const()[name = string("op_25373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25373_cast_fp16 = slice_by_index(begin = var_25373_begin_0, end = var_25373_end_0, end_mask = var_25373_end_mask_0, x = k_33_cast_fp16)[name = string("op_25373_cast_fp16")]; + tensor var_25377_begin_0 = const()[name = string("op_25377_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25377_end_0 = const()[name = string("op_25377_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_25377_end_mask_0 = const()[name = string("op_25377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25377_cast_fp16 = slice_by_index(begin = var_25377_begin_0, end = var_25377_end_0, end_mask = var_25377_end_mask_0, x = k_33_cast_fp16)[name = string("op_25377_cast_fp16")]; + tensor var_25381_begin_0 = const()[name = string("op_25381_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25381_end_0 = const()[name = string("op_25381_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_25381_end_mask_0 = const()[name = string("op_25381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25381_cast_fp16 = slice_by_index(begin = var_25381_begin_0, end = var_25381_end_0, end_mask = var_25381_end_mask_0, x = k_33_cast_fp16)[name = string("op_25381_cast_fp16")]; + tensor var_25385_begin_0 = const()[name = string("op_25385_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25385_end_0 = const()[name = string("op_25385_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_25385_end_mask_0 = const()[name = string("op_25385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25385_cast_fp16 = slice_by_index(begin = var_25385_begin_0, end = var_25385_end_0, end_mask = var_25385_end_mask_0, x = k_33_cast_fp16)[name = string("op_25385_cast_fp16")]; + tensor var_25389_begin_0 = const()[name = string("op_25389_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25389_end_0 = const()[name = string("op_25389_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_25389_end_mask_0 = const()[name = string("op_25389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25389_cast_fp16 = slice_by_index(begin = var_25389_begin_0, end = var_25389_end_0, end_mask = var_25389_end_mask_0, x = k_33_cast_fp16)[name = string("op_25389_cast_fp16")]; + tensor var_25393_begin_0 = const()[name = string("op_25393_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25393_end_0 = const()[name = string("op_25393_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_25393_end_mask_0 = const()[name = string("op_25393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25393_cast_fp16 = slice_by_index(begin = var_25393_begin_0, end = var_25393_end_0, end_mask = var_25393_end_mask_0, x = k_33_cast_fp16)[name = string("op_25393_cast_fp16")]; + tensor var_25397_begin_0 = const()[name = string("op_25397_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25397_end_0 = const()[name = string("op_25397_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_25397_end_mask_0 = const()[name = string("op_25397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25397_cast_fp16 = slice_by_index(begin = var_25397_begin_0, end = var_25397_end_0, end_mask = var_25397_end_mask_0, x = k_33_cast_fp16)[name = string("op_25397_cast_fp16")]; + tensor var_25401_begin_0 = const()[name = string("op_25401_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25401_end_0 = const()[name = string("op_25401_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_25401_end_mask_0 = const()[name = string("op_25401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25401_cast_fp16 = slice_by_index(begin = var_25401_begin_0, end = var_25401_end_0, end_mask = var_25401_end_mask_0, x = k_33_cast_fp16)[name = string("op_25401_cast_fp16")]; + tensor var_25405_begin_0 = const()[name = string("op_25405_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25405_end_0 = const()[name = string("op_25405_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_25405_end_mask_0 = const()[name = string("op_25405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25405_cast_fp16 = slice_by_index(begin = var_25405_begin_0, end = var_25405_end_0, end_mask = var_25405_end_mask_0, x = k_33_cast_fp16)[name = string("op_25405_cast_fp16")]; + tensor var_25409_begin_0 = const()[name = string("op_25409_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25409_end_0 = const()[name = string("op_25409_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_25409_end_mask_0 = const()[name = string("op_25409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25409_cast_fp16 = slice_by_index(begin = var_25409_begin_0, end = var_25409_end_0, end_mask = var_25409_end_mask_0, x = k_33_cast_fp16)[name = string("op_25409_cast_fp16")]; + tensor var_25413_begin_0 = const()[name = string("op_25413_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25413_end_0 = const()[name = string("op_25413_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_25413_end_mask_0 = const()[name = string("op_25413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25413_cast_fp16 = slice_by_index(begin = var_25413_begin_0, end = var_25413_end_0, end_mask = var_25413_end_mask_0, x = k_33_cast_fp16)[name = string("op_25413_cast_fp16")]; + tensor var_25415_begin_0 = const()[name = string("op_25415_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25415_end_0 = const()[name = string("op_25415_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25415_end_mask_0 = const()[name = string("op_25415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25415_cast_fp16 = slice_by_index(begin = var_25415_begin_0, end = var_25415_end_0, end_mask = var_25415_end_mask_0, x = value_33_cast_fp16)[name = string("op_25415_cast_fp16")]; + tensor var_25419_begin_0 = const()[name = string("op_25419_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25419_end_0 = const()[name = string("op_25419_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_25419_end_mask_0 = const()[name = string("op_25419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25419_cast_fp16 = slice_by_index(begin = var_25419_begin_0, end = var_25419_end_0, end_mask = var_25419_end_mask_0, x = value_33_cast_fp16)[name = string("op_25419_cast_fp16")]; + tensor var_25423_begin_0 = const()[name = string("op_25423_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25423_end_0 = const()[name = string("op_25423_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_25423_end_mask_0 = const()[name = string("op_25423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25423_cast_fp16 = slice_by_index(begin = var_25423_begin_0, end = var_25423_end_0, end_mask = var_25423_end_mask_0, x = value_33_cast_fp16)[name = string("op_25423_cast_fp16")]; + tensor var_25427_begin_0 = const()[name = string("op_25427_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25427_end_0 = const()[name = string("op_25427_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_25427_end_mask_0 = const()[name = string("op_25427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25427_cast_fp16 = slice_by_index(begin = var_25427_begin_0, end = var_25427_end_0, end_mask = var_25427_end_mask_0, x = value_33_cast_fp16)[name = string("op_25427_cast_fp16")]; + tensor var_25431_begin_0 = const()[name = string("op_25431_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25431_end_0 = const()[name = string("op_25431_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_25431_end_mask_0 = const()[name = string("op_25431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25431_cast_fp16 = slice_by_index(begin = var_25431_begin_0, end = var_25431_end_0, end_mask = var_25431_end_mask_0, x = value_33_cast_fp16)[name = string("op_25431_cast_fp16")]; + tensor var_25435_begin_0 = const()[name = string("op_25435_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25435_end_0 = const()[name = string("op_25435_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_25435_end_mask_0 = const()[name = string("op_25435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25435_cast_fp16 = slice_by_index(begin = var_25435_begin_0, end = var_25435_end_0, end_mask = var_25435_end_mask_0, x = value_33_cast_fp16)[name = string("op_25435_cast_fp16")]; + tensor var_25439_begin_0 = const()[name = string("op_25439_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25439_end_0 = const()[name = string("op_25439_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_25439_end_mask_0 = const()[name = string("op_25439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25439_cast_fp16 = slice_by_index(begin = var_25439_begin_0, end = var_25439_end_0, end_mask = var_25439_end_mask_0, x = value_33_cast_fp16)[name = string("op_25439_cast_fp16")]; + tensor var_25443_begin_0 = const()[name = string("op_25443_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25443_end_0 = const()[name = string("op_25443_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_25443_end_mask_0 = const()[name = string("op_25443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25443_cast_fp16 = slice_by_index(begin = var_25443_begin_0, end = var_25443_end_0, end_mask = var_25443_end_mask_0, x = value_33_cast_fp16)[name = string("op_25443_cast_fp16")]; + tensor var_25447_begin_0 = const()[name = string("op_25447_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25447_end_0 = const()[name = string("op_25447_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_25447_end_mask_0 = const()[name = string("op_25447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25447_cast_fp16 = slice_by_index(begin = var_25447_begin_0, end = var_25447_end_0, end_mask = var_25447_end_mask_0, x = value_33_cast_fp16)[name = string("op_25447_cast_fp16")]; + tensor var_25451_begin_0 = const()[name = string("op_25451_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25451_end_0 = const()[name = string("op_25451_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_25451_end_mask_0 = const()[name = string("op_25451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25451_cast_fp16 = slice_by_index(begin = var_25451_begin_0, end = var_25451_end_0, end_mask = var_25451_end_mask_0, x = value_33_cast_fp16)[name = string("op_25451_cast_fp16")]; + tensor var_25455_begin_0 = const()[name = string("op_25455_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25455_end_0 = const()[name = string("op_25455_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_25455_end_mask_0 = const()[name = string("op_25455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25455_cast_fp16 = slice_by_index(begin = var_25455_begin_0, end = var_25455_end_0, end_mask = var_25455_end_mask_0, x = value_33_cast_fp16)[name = string("op_25455_cast_fp16")]; + tensor var_25459_begin_0 = const()[name = string("op_25459_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25459_end_0 = const()[name = string("op_25459_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_25459_end_mask_0 = const()[name = string("op_25459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25459_cast_fp16 = slice_by_index(begin = var_25459_begin_0, end = var_25459_end_0, end_mask = var_25459_end_mask_0, x = value_33_cast_fp16)[name = string("op_25459_cast_fp16")]; + tensor var_25463_begin_0 = const()[name = string("op_25463_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25463_end_0 = const()[name = string("op_25463_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_25463_end_mask_0 = const()[name = string("op_25463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25463_cast_fp16 = slice_by_index(begin = var_25463_begin_0, end = var_25463_end_0, end_mask = var_25463_end_mask_0, x = value_33_cast_fp16)[name = string("op_25463_cast_fp16")]; + tensor var_25467_begin_0 = const()[name = string("op_25467_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25467_end_0 = const()[name = string("op_25467_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_25467_end_mask_0 = const()[name = string("op_25467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25467_cast_fp16 = slice_by_index(begin = var_25467_begin_0, end = var_25467_end_0, end_mask = var_25467_end_mask_0, x = value_33_cast_fp16)[name = string("op_25467_cast_fp16")]; + tensor var_25471_begin_0 = const()[name = string("op_25471_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25471_end_0 = const()[name = string("op_25471_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_25471_end_mask_0 = const()[name = string("op_25471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25471_cast_fp16 = slice_by_index(begin = var_25471_begin_0, end = var_25471_end_0, end_mask = var_25471_end_mask_0, x = value_33_cast_fp16)[name = string("op_25471_cast_fp16")]; + tensor var_25475_begin_0 = const()[name = string("op_25475_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25475_end_0 = const()[name = string("op_25475_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_25475_end_mask_0 = const()[name = string("op_25475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25475_cast_fp16 = slice_by_index(begin = var_25475_begin_0, end = var_25475_end_0, end_mask = var_25475_end_mask_0, x = value_33_cast_fp16)[name = string("op_25475_cast_fp16")]; + tensor var_25479_begin_0 = const()[name = string("op_25479_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25479_end_0 = const()[name = string("op_25479_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_25479_end_mask_0 = const()[name = string("op_25479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25479_cast_fp16 = slice_by_index(begin = var_25479_begin_0, end = var_25479_end_0, end_mask = var_25479_end_mask_0, x = value_33_cast_fp16)[name = string("op_25479_cast_fp16")]; + tensor var_25483_begin_0 = const()[name = string("op_25483_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25483_end_0 = const()[name = string("op_25483_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_25483_end_mask_0 = const()[name = string("op_25483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25483_cast_fp16 = slice_by_index(begin = var_25483_begin_0, end = var_25483_end_0, end_mask = var_25483_end_mask_0, x = value_33_cast_fp16)[name = string("op_25483_cast_fp16")]; + tensor var_25487_begin_0 = const()[name = string("op_25487_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25487_end_0 = const()[name = string("op_25487_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_25487_end_mask_0 = const()[name = string("op_25487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25487_cast_fp16 = slice_by_index(begin = var_25487_begin_0, end = var_25487_end_0, end_mask = var_25487_end_mask_0, x = value_33_cast_fp16)[name = string("op_25487_cast_fp16")]; + tensor var_25491_begin_0 = const()[name = string("op_25491_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25491_end_0 = const()[name = string("op_25491_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_25491_end_mask_0 = const()[name = string("op_25491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25491_cast_fp16 = slice_by_index(begin = var_25491_begin_0, end = var_25491_end_0, end_mask = var_25491_end_mask_0, x = value_33_cast_fp16)[name = string("op_25491_cast_fp16")]; + string _SplitHeadsQ__mh_w_2561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2561_equation_0, values = (var_25337_cast_fp16, var_24779_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2561_cast_fp16")]; + string _SplitHeadsQ__mh_w_2563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2563_equation_0, values = (var_25337_cast_fp16, var_24786_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2563_cast_fp16")]; + string _SplitHeadsQ__mh_w_2565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2565_equation_0, values = (var_25337_cast_fp16, var_24793_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2565_cast_fp16")]; + string _SplitHeadsQ__mh_w_2567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2567_equation_0, values = (var_25337_cast_fp16, var_24800_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2567_cast_fp16")]; + string _SplitHeadsQ__mh_w_2569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2569_equation_0, values = (var_25341_cast_fp16, var_24807_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2569_cast_fp16")]; + string _SplitHeadsQ__mh_w_2571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2571_equation_0, values = (var_25341_cast_fp16, var_24814_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2571_cast_fp16")]; + string _SplitHeadsQ__mh_w_2573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2573_equation_0, values = (var_25341_cast_fp16, var_24821_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2573_cast_fp16")]; + string _SplitHeadsQ__mh_w_2575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2575_equation_0, values = (var_25341_cast_fp16, var_24828_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2575_cast_fp16")]; + string _SplitHeadsQ__mh_w_2577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2577_equation_0, values = (var_25345_cast_fp16, var_24835_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2577_cast_fp16")]; + string _SplitHeadsQ__mh_w_2579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2579_equation_0, values = (var_25345_cast_fp16, var_24842_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2579_cast_fp16")]; + string _SplitHeadsQ__mh_w_2581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2581_equation_0, values = (var_25345_cast_fp16, var_24849_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2581_cast_fp16")]; + string _SplitHeadsQ__mh_w_2583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2583_equation_0, values = (var_25345_cast_fp16, var_24856_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2583_cast_fp16")]; + string _SplitHeadsQ__mh_w_2585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2585_equation_0, values = (var_25349_cast_fp16, var_24863_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2585_cast_fp16")]; + string _SplitHeadsQ__mh_w_2587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2587_equation_0, values = (var_25349_cast_fp16, var_24870_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2587_cast_fp16")]; + string _SplitHeadsQ__mh_w_2589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2589_equation_0, values = (var_25349_cast_fp16, var_24877_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2589_cast_fp16")]; + string _SplitHeadsQ__mh_w_2591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2591_equation_0, values = (var_25349_cast_fp16, var_24884_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2591_cast_fp16")]; + string _SplitHeadsQ__mh_w_2593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2593_equation_0, values = (var_25353_cast_fp16, var_24891_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2593_cast_fp16")]; + string _SplitHeadsQ__mh_w_2595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2595_equation_0, values = (var_25353_cast_fp16, var_24898_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2595_cast_fp16")]; + string _SplitHeadsQ__mh_w_2597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2597_equation_0, values = (var_25353_cast_fp16, var_24905_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2597_cast_fp16")]; + string _SplitHeadsQ__mh_w_2599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2599_equation_0, values = (var_25353_cast_fp16, var_24912_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2599_cast_fp16")]; + string _SplitHeadsQ__mh_w_2601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2601_equation_0, values = (var_25357_cast_fp16, var_24919_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2601_cast_fp16")]; + string _SplitHeadsQ__mh_w_2603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2603_equation_0, values = (var_25357_cast_fp16, var_24926_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2603_cast_fp16")]; + string _SplitHeadsQ__mh_w_2605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2605_equation_0, values = (var_25357_cast_fp16, var_24933_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2605_cast_fp16")]; + string _SplitHeadsQ__mh_w_2607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2607_equation_0, values = (var_25357_cast_fp16, var_24940_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2607_cast_fp16")]; + string _SplitHeadsQ__mh_w_2609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2609_equation_0, values = (var_25361_cast_fp16, var_24947_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2609_cast_fp16")]; + string _SplitHeadsQ__mh_w_2611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2611_equation_0, values = (var_25361_cast_fp16, var_24954_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2611_cast_fp16")]; + string _SplitHeadsQ__mh_w_2613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2613_equation_0, values = (var_25361_cast_fp16, var_24961_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2613_cast_fp16")]; + string _SplitHeadsQ__mh_w_2615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2615_equation_0, values = (var_25361_cast_fp16, var_24968_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2615_cast_fp16")]; + string _SplitHeadsQ__mh_w_2617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2617_equation_0, values = (var_25365_cast_fp16, var_24975_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2617_cast_fp16")]; + string _SplitHeadsQ__mh_w_2619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2619_equation_0, values = (var_25365_cast_fp16, var_24982_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2619_cast_fp16")]; + string _SplitHeadsQ__mh_w_2621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2621_equation_0, values = (var_25365_cast_fp16, var_24989_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2621_cast_fp16")]; + string _SplitHeadsQ__mh_w_2623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2623_equation_0, values = (var_25365_cast_fp16, var_24996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2623_cast_fp16")]; + string _SplitHeadsQ__mh_w_2625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2625_equation_0, values = (var_25369_cast_fp16, var_25003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2625_cast_fp16")]; + string _SplitHeadsQ__mh_w_2627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2627_equation_0, values = (var_25369_cast_fp16, var_25010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2627_cast_fp16")]; + string _SplitHeadsQ__mh_w_2629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2629_equation_0, values = (var_25369_cast_fp16, var_25017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2629_cast_fp16")]; + string _SplitHeadsQ__mh_w_2631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2631_equation_0, values = (var_25369_cast_fp16, var_25024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2631_cast_fp16")]; + string _SplitHeadsQ__mh_w_2633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2633_equation_0, values = (var_25373_cast_fp16, var_25031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2633_cast_fp16")]; + string _SplitHeadsQ__mh_w_2635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2635_equation_0, values = (var_25373_cast_fp16, var_25038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2635_cast_fp16")]; + string _SplitHeadsQ__mh_w_2637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2637_equation_0, values = (var_25373_cast_fp16, var_25045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2637_cast_fp16")]; + string _SplitHeadsQ__mh_w_2639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2639_equation_0, values = (var_25373_cast_fp16, var_25052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2639_cast_fp16")]; + string _SplitHeadsQ__mh_w_2641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2641_equation_0, values = (var_25377_cast_fp16, var_25059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2641_cast_fp16")]; + string _SplitHeadsQ__mh_w_2643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2643_equation_0, values = (var_25377_cast_fp16, var_25066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2643_cast_fp16")]; + string _SplitHeadsQ__mh_w_2645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2645_equation_0, values = (var_25377_cast_fp16, var_25073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2645_cast_fp16")]; + string _SplitHeadsQ__mh_w_2647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2647_equation_0, values = (var_25377_cast_fp16, var_25080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2647_cast_fp16")]; + string _SplitHeadsQ__mh_w_2649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2649_equation_0, values = (var_25381_cast_fp16, var_25087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2649_cast_fp16")]; + string _SplitHeadsQ__mh_w_2651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2651_equation_0, values = (var_25381_cast_fp16, var_25094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2651_cast_fp16")]; + string _SplitHeadsQ__mh_w_2653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2653_equation_0, values = (var_25381_cast_fp16, var_25101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2653_cast_fp16")]; + string _SplitHeadsQ__mh_w_2655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2655_equation_0, values = (var_25381_cast_fp16, var_25108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2655_cast_fp16")]; + string _SplitHeadsQ__mh_w_2657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2657_equation_0, values = (var_25385_cast_fp16, var_25115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2657_cast_fp16")]; + string _SplitHeadsQ__mh_w_2659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2659_equation_0, values = (var_25385_cast_fp16, var_25122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2659_cast_fp16")]; + string _SplitHeadsQ__mh_w_2661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2661_equation_0, values = (var_25385_cast_fp16, var_25129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2661_cast_fp16")]; + string _SplitHeadsQ__mh_w_2663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2663_equation_0, values = (var_25385_cast_fp16, var_25136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2663_cast_fp16")]; + string _SplitHeadsQ__mh_w_2665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2665_equation_0, values = (var_25389_cast_fp16, var_25143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2665_cast_fp16")]; + string _SplitHeadsQ__mh_w_2667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2667_equation_0, values = (var_25389_cast_fp16, var_25150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2667_cast_fp16")]; + string _SplitHeadsQ__mh_w_2669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2669_equation_0, values = (var_25389_cast_fp16, var_25157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2669_cast_fp16")]; + string _SplitHeadsQ__mh_w_2671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2671_equation_0, values = (var_25389_cast_fp16, var_25164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2671_cast_fp16")]; + string _SplitHeadsQ__mh_w_2673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2673_equation_0, values = (var_25393_cast_fp16, var_25171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2673_cast_fp16")]; + string _SplitHeadsQ__mh_w_2675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2675_equation_0, values = (var_25393_cast_fp16, var_25178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2675_cast_fp16")]; + string _SplitHeadsQ__mh_w_2677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2677_equation_0, values = (var_25393_cast_fp16, var_25185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2677_cast_fp16")]; + string _SplitHeadsQ__mh_w_2679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2679_equation_0, values = (var_25393_cast_fp16, var_25192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2679_cast_fp16")]; + string _SplitHeadsQ__mh_w_2681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2681_equation_0, values = (var_25397_cast_fp16, var_25199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2681_cast_fp16")]; + string _SplitHeadsQ__mh_w_2683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2683_equation_0, values = (var_25397_cast_fp16, var_25206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2683_cast_fp16")]; + string _SplitHeadsQ__mh_w_2685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2685_equation_0, values = (var_25397_cast_fp16, var_25213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2685_cast_fp16")]; + string _SplitHeadsQ__mh_w_2687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2687_equation_0, values = (var_25397_cast_fp16, var_25220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2687_cast_fp16")]; + string _SplitHeadsQ__mh_w_2689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2689_equation_0, values = (var_25401_cast_fp16, var_25227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2689_cast_fp16")]; + string _SplitHeadsQ__mh_w_2691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2691_equation_0, values = (var_25401_cast_fp16, var_25234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2691_cast_fp16")]; + string _SplitHeadsQ__mh_w_2693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2693_equation_0, values = (var_25401_cast_fp16, var_25241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2693_cast_fp16")]; + string _SplitHeadsQ__mh_w_2695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2695_equation_0, values = (var_25401_cast_fp16, var_25248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2695_cast_fp16")]; + string _SplitHeadsQ__mh_w_2697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2697_equation_0, values = (var_25405_cast_fp16, var_25255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2697_cast_fp16")]; + string _SplitHeadsQ__mh_w_2699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2699_equation_0, values = (var_25405_cast_fp16, var_25262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2699_cast_fp16")]; + string _SplitHeadsQ__mh_w_2701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2701_equation_0, values = (var_25405_cast_fp16, var_25269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2701_cast_fp16")]; + string _SplitHeadsQ__mh_w_2703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2703_equation_0, values = (var_25405_cast_fp16, var_25276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2703_cast_fp16")]; + string _SplitHeadsQ__mh_w_2705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2705_equation_0, values = (var_25409_cast_fp16, var_25283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2705_cast_fp16")]; + string _SplitHeadsQ__mh_w_2707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2707_equation_0, values = (var_25409_cast_fp16, var_25290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2707_cast_fp16")]; + string _SplitHeadsQ__mh_w_2709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2709_equation_0, values = (var_25409_cast_fp16, var_25297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2709_cast_fp16")]; + string _SplitHeadsQ__mh_w_2711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2711_equation_0, values = (var_25409_cast_fp16, var_25304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2711_cast_fp16")]; + string _SplitHeadsQ__mh_w_2713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2713_equation_0, values = (var_25413_cast_fp16, var_25311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2713_cast_fp16")]; + string _SplitHeadsQ__mh_w_2715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2715_equation_0, values = (var_25413_cast_fp16, var_25318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2715_cast_fp16")]; + string _SplitHeadsQ__mh_w_2717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2717_equation_0, values = (var_25413_cast_fp16, var_25325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2717_cast_fp16")]; + string _SplitHeadsQ__mh_w_2719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2719_equation_0, values = (var_25413_cast_fp16, var_25332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2719_cast_fp16")]; + fp16 var_25654_to_fp16 = const()[name = string("op_25654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2561_cast_fp16, y = var_25654_to_fp16)[name = string("aw_chunk_2561_cast_fp16")]; + fp16 var_25656_to_fp16 = const()[name = string("op_25656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2563_cast_fp16, y = var_25656_to_fp16)[name = string("aw_chunk_2563_cast_fp16")]; + fp16 var_25658_to_fp16 = const()[name = string("op_25658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2565_cast_fp16, y = var_25658_to_fp16)[name = string("aw_chunk_2565_cast_fp16")]; + fp16 var_25660_to_fp16 = const()[name = string("op_25660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2567_cast_fp16, y = var_25660_to_fp16)[name = string("aw_chunk_2567_cast_fp16")]; + fp16 var_25662_to_fp16 = const()[name = string("op_25662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2569_cast_fp16, y = var_25662_to_fp16)[name = string("aw_chunk_2569_cast_fp16")]; + fp16 var_25664_to_fp16 = const()[name = string("op_25664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2571_cast_fp16, y = var_25664_to_fp16)[name = string("aw_chunk_2571_cast_fp16")]; + fp16 var_25666_to_fp16 = const()[name = string("op_25666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2573_cast_fp16, y = var_25666_to_fp16)[name = string("aw_chunk_2573_cast_fp16")]; + fp16 var_25668_to_fp16 = const()[name = string("op_25668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2575_cast_fp16, y = var_25668_to_fp16)[name = string("aw_chunk_2575_cast_fp16")]; + fp16 var_25670_to_fp16 = const()[name = string("op_25670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2577_cast_fp16, y = var_25670_to_fp16)[name = string("aw_chunk_2577_cast_fp16")]; + fp16 var_25672_to_fp16 = const()[name = string("op_25672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2579_cast_fp16, y = var_25672_to_fp16)[name = string("aw_chunk_2579_cast_fp16")]; + fp16 var_25674_to_fp16 = const()[name = string("op_25674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2581_cast_fp16, y = var_25674_to_fp16)[name = string("aw_chunk_2581_cast_fp16")]; + fp16 var_25676_to_fp16 = const()[name = string("op_25676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2583_cast_fp16, y = var_25676_to_fp16)[name = string("aw_chunk_2583_cast_fp16")]; + fp16 var_25678_to_fp16 = const()[name = string("op_25678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2585_cast_fp16, y = var_25678_to_fp16)[name = string("aw_chunk_2585_cast_fp16")]; + fp16 var_25680_to_fp16 = const()[name = string("op_25680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2587_cast_fp16, y = var_25680_to_fp16)[name = string("aw_chunk_2587_cast_fp16")]; + fp16 var_25682_to_fp16 = const()[name = string("op_25682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2589_cast_fp16, y = var_25682_to_fp16)[name = string("aw_chunk_2589_cast_fp16")]; + fp16 var_25684_to_fp16 = const()[name = string("op_25684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2591_cast_fp16, y = var_25684_to_fp16)[name = string("aw_chunk_2591_cast_fp16")]; + fp16 var_25686_to_fp16 = const()[name = string("op_25686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2593_cast_fp16, y = var_25686_to_fp16)[name = string("aw_chunk_2593_cast_fp16")]; + fp16 var_25688_to_fp16 = const()[name = string("op_25688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2595_cast_fp16, y = var_25688_to_fp16)[name = string("aw_chunk_2595_cast_fp16")]; + fp16 var_25690_to_fp16 = const()[name = string("op_25690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2597_cast_fp16, y = var_25690_to_fp16)[name = string("aw_chunk_2597_cast_fp16")]; + fp16 var_25692_to_fp16 = const()[name = string("op_25692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2599_cast_fp16, y = var_25692_to_fp16)[name = string("aw_chunk_2599_cast_fp16")]; + fp16 var_25694_to_fp16 = const()[name = string("op_25694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2601_cast_fp16, y = var_25694_to_fp16)[name = string("aw_chunk_2601_cast_fp16")]; + fp16 var_25696_to_fp16 = const()[name = string("op_25696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2603_cast_fp16, y = var_25696_to_fp16)[name = string("aw_chunk_2603_cast_fp16")]; + fp16 var_25698_to_fp16 = const()[name = string("op_25698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2605_cast_fp16, y = var_25698_to_fp16)[name = string("aw_chunk_2605_cast_fp16")]; + fp16 var_25700_to_fp16 = const()[name = string("op_25700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2607_cast_fp16, y = var_25700_to_fp16)[name = string("aw_chunk_2607_cast_fp16")]; + fp16 var_25702_to_fp16 = const()[name = string("op_25702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2609_cast_fp16, y = var_25702_to_fp16)[name = string("aw_chunk_2609_cast_fp16")]; + fp16 var_25704_to_fp16 = const()[name = string("op_25704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2611_cast_fp16, y = var_25704_to_fp16)[name = string("aw_chunk_2611_cast_fp16")]; + fp16 var_25706_to_fp16 = const()[name = string("op_25706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2613_cast_fp16, y = var_25706_to_fp16)[name = string("aw_chunk_2613_cast_fp16")]; + fp16 var_25708_to_fp16 = const()[name = string("op_25708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2615_cast_fp16, y = var_25708_to_fp16)[name = string("aw_chunk_2615_cast_fp16")]; + fp16 var_25710_to_fp16 = const()[name = string("op_25710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2617_cast_fp16, y = var_25710_to_fp16)[name = string("aw_chunk_2617_cast_fp16")]; + fp16 var_25712_to_fp16 = const()[name = string("op_25712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2619_cast_fp16, y = var_25712_to_fp16)[name = string("aw_chunk_2619_cast_fp16")]; + fp16 var_25714_to_fp16 = const()[name = string("op_25714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2621_cast_fp16, y = var_25714_to_fp16)[name = string("aw_chunk_2621_cast_fp16")]; + fp16 var_25716_to_fp16 = const()[name = string("op_25716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2623_cast_fp16, y = var_25716_to_fp16)[name = string("aw_chunk_2623_cast_fp16")]; + fp16 var_25718_to_fp16 = const()[name = string("op_25718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2625_cast_fp16, y = var_25718_to_fp16)[name = string("aw_chunk_2625_cast_fp16")]; + fp16 var_25720_to_fp16 = const()[name = string("op_25720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2627_cast_fp16, y = var_25720_to_fp16)[name = string("aw_chunk_2627_cast_fp16")]; + fp16 var_25722_to_fp16 = const()[name = string("op_25722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2629_cast_fp16, y = var_25722_to_fp16)[name = string("aw_chunk_2629_cast_fp16")]; + fp16 var_25724_to_fp16 = const()[name = string("op_25724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2631_cast_fp16, y = var_25724_to_fp16)[name = string("aw_chunk_2631_cast_fp16")]; + fp16 var_25726_to_fp16 = const()[name = string("op_25726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2633_cast_fp16, y = var_25726_to_fp16)[name = string("aw_chunk_2633_cast_fp16")]; + fp16 var_25728_to_fp16 = const()[name = string("op_25728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2635_cast_fp16, y = var_25728_to_fp16)[name = string("aw_chunk_2635_cast_fp16")]; + fp16 var_25730_to_fp16 = const()[name = string("op_25730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2637_cast_fp16, y = var_25730_to_fp16)[name = string("aw_chunk_2637_cast_fp16")]; + fp16 var_25732_to_fp16 = const()[name = string("op_25732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2639_cast_fp16, y = var_25732_to_fp16)[name = string("aw_chunk_2639_cast_fp16")]; + fp16 var_25734_to_fp16 = const()[name = string("op_25734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2641_cast_fp16, y = var_25734_to_fp16)[name = string("aw_chunk_2641_cast_fp16")]; + fp16 var_25736_to_fp16 = const()[name = string("op_25736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2643_cast_fp16, y = var_25736_to_fp16)[name = string("aw_chunk_2643_cast_fp16")]; + fp16 var_25738_to_fp16 = const()[name = string("op_25738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2645_cast_fp16, y = var_25738_to_fp16)[name = string("aw_chunk_2645_cast_fp16")]; + fp16 var_25740_to_fp16 = const()[name = string("op_25740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2647_cast_fp16, y = var_25740_to_fp16)[name = string("aw_chunk_2647_cast_fp16")]; + fp16 var_25742_to_fp16 = const()[name = string("op_25742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2649_cast_fp16, y = var_25742_to_fp16)[name = string("aw_chunk_2649_cast_fp16")]; + fp16 var_25744_to_fp16 = const()[name = string("op_25744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2651_cast_fp16, y = var_25744_to_fp16)[name = string("aw_chunk_2651_cast_fp16")]; + fp16 var_25746_to_fp16 = const()[name = string("op_25746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2653_cast_fp16, y = var_25746_to_fp16)[name = string("aw_chunk_2653_cast_fp16")]; + fp16 var_25748_to_fp16 = const()[name = string("op_25748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2655_cast_fp16, y = var_25748_to_fp16)[name = string("aw_chunk_2655_cast_fp16")]; + fp16 var_25750_to_fp16 = const()[name = string("op_25750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2657_cast_fp16, y = var_25750_to_fp16)[name = string("aw_chunk_2657_cast_fp16")]; + fp16 var_25752_to_fp16 = const()[name = string("op_25752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2659_cast_fp16, y = var_25752_to_fp16)[name = string("aw_chunk_2659_cast_fp16")]; + fp16 var_25754_to_fp16 = const()[name = string("op_25754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2661_cast_fp16, y = var_25754_to_fp16)[name = string("aw_chunk_2661_cast_fp16")]; + fp16 var_25756_to_fp16 = const()[name = string("op_25756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2663_cast_fp16, y = var_25756_to_fp16)[name = string("aw_chunk_2663_cast_fp16")]; + fp16 var_25758_to_fp16 = const()[name = string("op_25758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2665_cast_fp16, y = var_25758_to_fp16)[name = string("aw_chunk_2665_cast_fp16")]; + fp16 var_25760_to_fp16 = const()[name = string("op_25760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2667_cast_fp16, y = var_25760_to_fp16)[name = string("aw_chunk_2667_cast_fp16")]; + fp16 var_25762_to_fp16 = const()[name = string("op_25762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2669_cast_fp16, y = var_25762_to_fp16)[name = string("aw_chunk_2669_cast_fp16")]; + fp16 var_25764_to_fp16 = const()[name = string("op_25764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2671_cast_fp16, y = var_25764_to_fp16)[name = string("aw_chunk_2671_cast_fp16")]; + fp16 var_25766_to_fp16 = const()[name = string("op_25766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2673_cast_fp16, y = var_25766_to_fp16)[name = string("aw_chunk_2673_cast_fp16")]; + fp16 var_25768_to_fp16 = const()[name = string("op_25768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2675_cast_fp16, y = var_25768_to_fp16)[name = string("aw_chunk_2675_cast_fp16")]; + fp16 var_25770_to_fp16 = const()[name = string("op_25770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2677_cast_fp16, y = var_25770_to_fp16)[name = string("aw_chunk_2677_cast_fp16")]; + fp16 var_25772_to_fp16 = const()[name = string("op_25772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2679_cast_fp16, y = var_25772_to_fp16)[name = string("aw_chunk_2679_cast_fp16")]; + fp16 var_25774_to_fp16 = const()[name = string("op_25774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2681_cast_fp16, y = var_25774_to_fp16)[name = string("aw_chunk_2681_cast_fp16")]; + fp16 var_25776_to_fp16 = const()[name = string("op_25776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2683_cast_fp16, y = var_25776_to_fp16)[name = string("aw_chunk_2683_cast_fp16")]; + fp16 var_25778_to_fp16 = const()[name = string("op_25778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2685_cast_fp16, y = var_25778_to_fp16)[name = string("aw_chunk_2685_cast_fp16")]; + fp16 var_25780_to_fp16 = const()[name = string("op_25780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2687_cast_fp16, y = var_25780_to_fp16)[name = string("aw_chunk_2687_cast_fp16")]; + fp16 var_25782_to_fp16 = const()[name = string("op_25782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2689_cast_fp16, y = var_25782_to_fp16)[name = string("aw_chunk_2689_cast_fp16")]; + fp16 var_25784_to_fp16 = const()[name = string("op_25784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2691_cast_fp16, y = var_25784_to_fp16)[name = string("aw_chunk_2691_cast_fp16")]; + fp16 var_25786_to_fp16 = const()[name = string("op_25786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2693_cast_fp16, y = var_25786_to_fp16)[name = string("aw_chunk_2693_cast_fp16")]; + fp16 var_25788_to_fp16 = const()[name = string("op_25788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2695_cast_fp16, y = var_25788_to_fp16)[name = string("aw_chunk_2695_cast_fp16")]; + fp16 var_25790_to_fp16 = const()[name = string("op_25790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2697_cast_fp16, y = var_25790_to_fp16)[name = string("aw_chunk_2697_cast_fp16")]; + fp16 var_25792_to_fp16 = const()[name = string("op_25792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2699_cast_fp16, y = var_25792_to_fp16)[name = string("aw_chunk_2699_cast_fp16")]; + fp16 var_25794_to_fp16 = const()[name = string("op_25794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2701_cast_fp16, y = var_25794_to_fp16)[name = string("aw_chunk_2701_cast_fp16")]; + fp16 var_25796_to_fp16 = const()[name = string("op_25796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2703_cast_fp16, y = var_25796_to_fp16)[name = string("aw_chunk_2703_cast_fp16")]; + fp16 var_25798_to_fp16 = const()[name = string("op_25798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2705_cast_fp16, y = var_25798_to_fp16)[name = string("aw_chunk_2705_cast_fp16")]; + fp16 var_25800_to_fp16 = const()[name = string("op_25800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2707_cast_fp16, y = var_25800_to_fp16)[name = string("aw_chunk_2707_cast_fp16")]; + fp16 var_25802_to_fp16 = const()[name = string("op_25802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2709_cast_fp16, y = var_25802_to_fp16)[name = string("aw_chunk_2709_cast_fp16")]; + fp16 var_25804_to_fp16 = const()[name = string("op_25804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2711_cast_fp16, y = var_25804_to_fp16)[name = string("aw_chunk_2711_cast_fp16")]; + fp16 var_25806_to_fp16 = const()[name = string("op_25806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2713_cast_fp16, y = var_25806_to_fp16)[name = string("aw_chunk_2713_cast_fp16")]; + fp16 var_25808_to_fp16 = const()[name = string("op_25808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2715_cast_fp16, y = var_25808_to_fp16)[name = string("aw_chunk_2715_cast_fp16")]; + fp16 var_25810_to_fp16 = const()[name = string("op_25810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2717_cast_fp16, y = var_25810_to_fp16)[name = string("aw_chunk_2717_cast_fp16")]; + fp16 var_25812_to_fp16 = const()[name = string("op_25812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2719_cast_fp16, y = var_25812_to_fp16)[name = string("aw_chunk_2719_cast_fp16")]; + tensor var_25814_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2561_cast_fp16)[name = string("op_25814_cast_fp16")]; + tensor var_25815_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2563_cast_fp16)[name = string("op_25815_cast_fp16")]; + tensor var_25816_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2565_cast_fp16)[name = string("op_25816_cast_fp16")]; + tensor var_25817_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2567_cast_fp16)[name = string("op_25817_cast_fp16")]; + tensor var_25818_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2569_cast_fp16)[name = string("op_25818_cast_fp16")]; + tensor var_25819_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2571_cast_fp16)[name = string("op_25819_cast_fp16")]; + tensor var_25820_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2573_cast_fp16)[name = string("op_25820_cast_fp16")]; + tensor var_25821_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2575_cast_fp16)[name = string("op_25821_cast_fp16")]; + tensor var_25822_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2577_cast_fp16)[name = string("op_25822_cast_fp16")]; + tensor var_25823_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2579_cast_fp16)[name = string("op_25823_cast_fp16")]; + tensor var_25824_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2581_cast_fp16)[name = string("op_25824_cast_fp16")]; + tensor var_25825_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2583_cast_fp16)[name = string("op_25825_cast_fp16")]; + tensor var_25826_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2585_cast_fp16)[name = string("op_25826_cast_fp16")]; + tensor var_25827_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2587_cast_fp16)[name = string("op_25827_cast_fp16")]; + tensor var_25828_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2589_cast_fp16)[name = string("op_25828_cast_fp16")]; + tensor var_25829_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2591_cast_fp16)[name = string("op_25829_cast_fp16")]; + tensor var_25830_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2593_cast_fp16)[name = string("op_25830_cast_fp16")]; + tensor var_25831_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2595_cast_fp16)[name = string("op_25831_cast_fp16")]; + tensor var_25832_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2597_cast_fp16)[name = string("op_25832_cast_fp16")]; + tensor var_25833_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2599_cast_fp16)[name = string("op_25833_cast_fp16")]; + tensor var_25834_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2601_cast_fp16)[name = string("op_25834_cast_fp16")]; + tensor var_25835_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2603_cast_fp16)[name = string("op_25835_cast_fp16")]; + tensor var_25836_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2605_cast_fp16)[name = string("op_25836_cast_fp16")]; + tensor var_25837_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2607_cast_fp16)[name = string("op_25837_cast_fp16")]; + tensor var_25838_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2609_cast_fp16)[name = string("op_25838_cast_fp16")]; + tensor var_25839_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2611_cast_fp16)[name = string("op_25839_cast_fp16")]; + tensor var_25840_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2613_cast_fp16)[name = string("op_25840_cast_fp16")]; + tensor var_25841_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2615_cast_fp16)[name = string("op_25841_cast_fp16")]; + tensor var_25842_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2617_cast_fp16)[name = string("op_25842_cast_fp16")]; + tensor var_25843_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2619_cast_fp16)[name = string("op_25843_cast_fp16")]; + tensor var_25844_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2621_cast_fp16)[name = string("op_25844_cast_fp16")]; + tensor var_25845_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2623_cast_fp16)[name = string("op_25845_cast_fp16")]; + tensor var_25846_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2625_cast_fp16)[name = string("op_25846_cast_fp16")]; + tensor var_25847_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2627_cast_fp16)[name = string("op_25847_cast_fp16")]; + tensor var_25848_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2629_cast_fp16)[name = string("op_25848_cast_fp16")]; + tensor var_25849_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2631_cast_fp16)[name = string("op_25849_cast_fp16")]; + tensor var_25850_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2633_cast_fp16)[name = string("op_25850_cast_fp16")]; + tensor var_25851_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2635_cast_fp16)[name = string("op_25851_cast_fp16")]; + tensor var_25852_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2637_cast_fp16)[name = string("op_25852_cast_fp16")]; + tensor var_25853_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2639_cast_fp16)[name = string("op_25853_cast_fp16")]; + tensor var_25854_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2641_cast_fp16)[name = string("op_25854_cast_fp16")]; + tensor var_25855_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2643_cast_fp16)[name = string("op_25855_cast_fp16")]; + tensor var_25856_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2645_cast_fp16)[name = string("op_25856_cast_fp16")]; + tensor var_25857_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2647_cast_fp16)[name = string("op_25857_cast_fp16")]; + tensor var_25858_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2649_cast_fp16)[name = string("op_25858_cast_fp16")]; + tensor var_25859_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2651_cast_fp16)[name = string("op_25859_cast_fp16")]; + tensor var_25860_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2653_cast_fp16)[name = string("op_25860_cast_fp16")]; + tensor var_25861_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2655_cast_fp16)[name = string("op_25861_cast_fp16")]; + tensor var_25862_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2657_cast_fp16)[name = string("op_25862_cast_fp16")]; + tensor var_25863_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2659_cast_fp16)[name = string("op_25863_cast_fp16")]; + tensor var_25864_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2661_cast_fp16)[name = string("op_25864_cast_fp16")]; + tensor var_25865_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2663_cast_fp16)[name = string("op_25865_cast_fp16")]; + tensor var_25866_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2665_cast_fp16)[name = string("op_25866_cast_fp16")]; + tensor var_25867_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2667_cast_fp16)[name = string("op_25867_cast_fp16")]; + tensor var_25868_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2669_cast_fp16)[name = string("op_25868_cast_fp16")]; + tensor var_25869_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2671_cast_fp16)[name = string("op_25869_cast_fp16")]; + tensor var_25870_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2673_cast_fp16)[name = string("op_25870_cast_fp16")]; + tensor var_25871_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2675_cast_fp16)[name = string("op_25871_cast_fp16")]; + tensor var_25872_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2677_cast_fp16)[name = string("op_25872_cast_fp16")]; + tensor var_25873_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2679_cast_fp16)[name = string("op_25873_cast_fp16")]; + tensor var_25874_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2681_cast_fp16)[name = string("op_25874_cast_fp16")]; + tensor var_25875_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2683_cast_fp16)[name = string("op_25875_cast_fp16")]; + tensor var_25876_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2685_cast_fp16)[name = string("op_25876_cast_fp16")]; + tensor var_25877_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2687_cast_fp16)[name = string("op_25877_cast_fp16")]; + tensor var_25878_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2689_cast_fp16)[name = string("op_25878_cast_fp16")]; + tensor var_25879_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2691_cast_fp16)[name = string("op_25879_cast_fp16")]; + tensor var_25880_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2693_cast_fp16)[name = string("op_25880_cast_fp16")]; + tensor var_25881_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2695_cast_fp16)[name = string("op_25881_cast_fp16")]; + tensor var_25882_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2697_cast_fp16)[name = string("op_25882_cast_fp16")]; + tensor var_25883_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2699_cast_fp16)[name = string("op_25883_cast_fp16")]; + tensor var_25884_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2701_cast_fp16)[name = string("op_25884_cast_fp16")]; + tensor var_25885_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2703_cast_fp16)[name = string("op_25885_cast_fp16")]; + tensor var_25886_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2705_cast_fp16)[name = string("op_25886_cast_fp16")]; + tensor var_25887_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2707_cast_fp16)[name = string("op_25887_cast_fp16")]; + tensor var_25888_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2709_cast_fp16)[name = string("op_25888_cast_fp16")]; + tensor var_25889_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2711_cast_fp16)[name = string("op_25889_cast_fp16")]; + tensor var_25890_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2713_cast_fp16)[name = string("op_25890_cast_fp16")]; + tensor var_25891_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2715_cast_fp16)[name = string("op_25891_cast_fp16")]; + tensor var_25892_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2717_cast_fp16)[name = string("op_25892_cast_fp16")]; + tensor var_25893_cast_fp16 = softmax(axis = var_24639, x = aw_chunk_2719_cast_fp16)[name = string("op_25893_cast_fp16")]; + string var_25895_equation_0 = const()[name = string("op_25895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25895_cast_fp16 = einsum(equation = var_25895_equation_0, values = (var_25415_cast_fp16, var_25814_cast_fp16))[name = string("op_25895_cast_fp16")]; + string var_25897_equation_0 = const()[name = string("op_25897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25897_cast_fp16 = einsum(equation = var_25897_equation_0, values = (var_25415_cast_fp16, var_25815_cast_fp16))[name = string("op_25897_cast_fp16")]; + string var_25899_equation_0 = const()[name = string("op_25899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25899_cast_fp16 = einsum(equation = var_25899_equation_0, values = (var_25415_cast_fp16, var_25816_cast_fp16))[name = string("op_25899_cast_fp16")]; + string var_25901_equation_0 = const()[name = string("op_25901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25901_cast_fp16 = einsum(equation = var_25901_equation_0, values = (var_25415_cast_fp16, var_25817_cast_fp16))[name = string("op_25901_cast_fp16")]; + string var_25903_equation_0 = const()[name = string("op_25903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25903_cast_fp16 = einsum(equation = var_25903_equation_0, values = (var_25419_cast_fp16, var_25818_cast_fp16))[name = string("op_25903_cast_fp16")]; + string var_25905_equation_0 = const()[name = string("op_25905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25905_cast_fp16 = einsum(equation = var_25905_equation_0, values = (var_25419_cast_fp16, var_25819_cast_fp16))[name = string("op_25905_cast_fp16")]; + string var_25907_equation_0 = const()[name = string("op_25907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25907_cast_fp16 = einsum(equation = var_25907_equation_0, values = (var_25419_cast_fp16, var_25820_cast_fp16))[name = string("op_25907_cast_fp16")]; + string var_25909_equation_0 = const()[name = string("op_25909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25909_cast_fp16 = einsum(equation = var_25909_equation_0, values = (var_25419_cast_fp16, var_25821_cast_fp16))[name = string("op_25909_cast_fp16")]; + string var_25911_equation_0 = const()[name = string("op_25911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25911_cast_fp16 = einsum(equation = var_25911_equation_0, values = (var_25423_cast_fp16, var_25822_cast_fp16))[name = string("op_25911_cast_fp16")]; + string var_25913_equation_0 = const()[name = string("op_25913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25913_cast_fp16 = einsum(equation = var_25913_equation_0, values = (var_25423_cast_fp16, var_25823_cast_fp16))[name = string("op_25913_cast_fp16")]; + string var_25915_equation_0 = const()[name = string("op_25915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25915_cast_fp16 = einsum(equation = var_25915_equation_0, values = (var_25423_cast_fp16, var_25824_cast_fp16))[name = string("op_25915_cast_fp16")]; + string var_25917_equation_0 = const()[name = string("op_25917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25917_cast_fp16 = einsum(equation = var_25917_equation_0, values = (var_25423_cast_fp16, var_25825_cast_fp16))[name = string("op_25917_cast_fp16")]; + string var_25919_equation_0 = const()[name = string("op_25919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25919_cast_fp16 = einsum(equation = var_25919_equation_0, values = (var_25427_cast_fp16, var_25826_cast_fp16))[name = string("op_25919_cast_fp16")]; + string var_25921_equation_0 = const()[name = string("op_25921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25921_cast_fp16 = einsum(equation = var_25921_equation_0, values = (var_25427_cast_fp16, var_25827_cast_fp16))[name = string("op_25921_cast_fp16")]; + string var_25923_equation_0 = const()[name = string("op_25923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25923_cast_fp16 = einsum(equation = var_25923_equation_0, values = (var_25427_cast_fp16, var_25828_cast_fp16))[name = string("op_25923_cast_fp16")]; + string var_25925_equation_0 = const()[name = string("op_25925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25925_cast_fp16 = einsum(equation = var_25925_equation_0, values = (var_25427_cast_fp16, var_25829_cast_fp16))[name = string("op_25925_cast_fp16")]; + string var_25927_equation_0 = const()[name = string("op_25927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25927_cast_fp16 = einsum(equation = var_25927_equation_0, values = (var_25431_cast_fp16, var_25830_cast_fp16))[name = string("op_25927_cast_fp16")]; + string var_25929_equation_0 = const()[name = string("op_25929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25929_cast_fp16 = einsum(equation = var_25929_equation_0, values = (var_25431_cast_fp16, var_25831_cast_fp16))[name = string("op_25929_cast_fp16")]; + string var_25931_equation_0 = const()[name = string("op_25931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25931_cast_fp16 = einsum(equation = var_25931_equation_0, values = (var_25431_cast_fp16, var_25832_cast_fp16))[name = string("op_25931_cast_fp16")]; + string var_25933_equation_0 = const()[name = string("op_25933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25933_cast_fp16 = einsum(equation = var_25933_equation_0, values = (var_25431_cast_fp16, var_25833_cast_fp16))[name = string("op_25933_cast_fp16")]; + string var_25935_equation_0 = const()[name = string("op_25935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25935_cast_fp16 = einsum(equation = var_25935_equation_0, values = (var_25435_cast_fp16, var_25834_cast_fp16))[name = string("op_25935_cast_fp16")]; + string var_25937_equation_0 = const()[name = string("op_25937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25937_cast_fp16 = einsum(equation = var_25937_equation_0, values = (var_25435_cast_fp16, var_25835_cast_fp16))[name = string("op_25937_cast_fp16")]; + string var_25939_equation_0 = const()[name = string("op_25939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25939_cast_fp16 = einsum(equation = var_25939_equation_0, values = (var_25435_cast_fp16, var_25836_cast_fp16))[name = string("op_25939_cast_fp16")]; + string var_25941_equation_0 = const()[name = string("op_25941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25941_cast_fp16 = einsum(equation = var_25941_equation_0, values = (var_25435_cast_fp16, var_25837_cast_fp16))[name = string("op_25941_cast_fp16")]; + string var_25943_equation_0 = const()[name = string("op_25943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25943_cast_fp16 = einsum(equation = var_25943_equation_0, values = (var_25439_cast_fp16, var_25838_cast_fp16))[name = string("op_25943_cast_fp16")]; + string var_25945_equation_0 = const()[name = string("op_25945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25945_cast_fp16 = einsum(equation = var_25945_equation_0, values = (var_25439_cast_fp16, var_25839_cast_fp16))[name = string("op_25945_cast_fp16")]; + string var_25947_equation_0 = const()[name = string("op_25947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25947_cast_fp16 = einsum(equation = var_25947_equation_0, values = (var_25439_cast_fp16, var_25840_cast_fp16))[name = string("op_25947_cast_fp16")]; + string var_25949_equation_0 = const()[name = string("op_25949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25949_cast_fp16 = einsum(equation = var_25949_equation_0, values = (var_25439_cast_fp16, var_25841_cast_fp16))[name = string("op_25949_cast_fp16")]; + string var_25951_equation_0 = const()[name = string("op_25951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25951_cast_fp16 = einsum(equation = var_25951_equation_0, values = (var_25443_cast_fp16, var_25842_cast_fp16))[name = string("op_25951_cast_fp16")]; + string var_25953_equation_0 = const()[name = string("op_25953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25953_cast_fp16 = einsum(equation = var_25953_equation_0, values = (var_25443_cast_fp16, var_25843_cast_fp16))[name = string("op_25953_cast_fp16")]; + string var_25955_equation_0 = const()[name = string("op_25955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25955_cast_fp16 = einsum(equation = var_25955_equation_0, values = (var_25443_cast_fp16, var_25844_cast_fp16))[name = string("op_25955_cast_fp16")]; + string var_25957_equation_0 = const()[name = string("op_25957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25957_cast_fp16 = einsum(equation = var_25957_equation_0, values = (var_25443_cast_fp16, var_25845_cast_fp16))[name = string("op_25957_cast_fp16")]; + string var_25959_equation_0 = const()[name = string("op_25959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25959_cast_fp16 = einsum(equation = var_25959_equation_0, values = (var_25447_cast_fp16, var_25846_cast_fp16))[name = string("op_25959_cast_fp16")]; + string var_25961_equation_0 = const()[name = string("op_25961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25961_cast_fp16 = einsum(equation = var_25961_equation_0, values = (var_25447_cast_fp16, var_25847_cast_fp16))[name = string("op_25961_cast_fp16")]; + string var_25963_equation_0 = const()[name = string("op_25963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25963_cast_fp16 = einsum(equation = var_25963_equation_0, values = (var_25447_cast_fp16, var_25848_cast_fp16))[name = string("op_25963_cast_fp16")]; + string var_25965_equation_0 = const()[name = string("op_25965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25965_cast_fp16 = einsum(equation = var_25965_equation_0, values = (var_25447_cast_fp16, var_25849_cast_fp16))[name = string("op_25965_cast_fp16")]; + string var_25967_equation_0 = const()[name = string("op_25967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25967_cast_fp16 = einsum(equation = var_25967_equation_0, values = (var_25451_cast_fp16, var_25850_cast_fp16))[name = string("op_25967_cast_fp16")]; + string var_25969_equation_0 = const()[name = string("op_25969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25969_cast_fp16 = einsum(equation = var_25969_equation_0, values = (var_25451_cast_fp16, var_25851_cast_fp16))[name = string("op_25969_cast_fp16")]; + string var_25971_equation_0 = const()[name = string("op_25971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25971_cast_fp16 = einsum(equation = var_25971_equation_0, values = (var_25451_cast_fp16, var_25852_cast_fp16))[name = string("op_25971_cast_fp16")]; + string var_25973_equation_0 = const()[name = string("op_25973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25973_cast_fp16 = einsum(equation = var_25973_equation_0, values = (var_25451_cast_fp16, var_25853_cast_fp16))[name = string("op_25973_cast_fp16")]; + string var_25975_equation_0 = const()[name = string("op_25975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25975_cast_fp16 = einsum(equation = var_25975_equation_0, values = (var_25455_cast_fp16, var_25854_cast_fp16))[name = string("op_25975_cast_fp16")]; + string var_25977_equation_0 = const()[name = string("op_25977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25977_cast_fp16 = einsum(equation = var_25977_equation_0, values = (var_25455_cast_fp16, var_25855_cast_fp16))[name = string("op_25977_cast_fp16")]; + string var_25979_equation_0 = const()[name = string("op_25979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25979_cast_fp16 = einsum(equation = var_25979_equation_0, values = (var_25455_cast_fp16, var_25856_cast_fp16))[name = string("op_25979_cast_fp16")]; + string var_25981_equation_0 = const()[name = string("op_25981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25981_cast_fp16 = einsum(equation = var_25981_equation_0, values = (var_25455_cast_fp16, var_25857_cast_fp16))[name = string("op_25981_cast_fp16")]; + string var_25983_equation_0 = const()[name = string("op_25983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25983_cast_fp16 = einsum(equation = var_25983_equation_0, values = (var_25459_cast_fp16, var_25858_cast_fp16))[name = string("op_25983_cast_fp16")]; + string var_25985_equation_0 = const()[name = string("op_25985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25985_cast_fp16 = einsum(equation = var_25985_equation_0, values = (var_25459_cast_fp16, var_25859_cast_fp16))[name = string("op_25985_cast_fp16")]; + string var_25987_equation_0 = const()[name = string("op_25987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25987_cast_fp16 = einsum(equation = var_25987_equation_0, values = (var_25459_cast_fp16, var_25860_cast_fp16))[name = string("op_25987_cast_fp16")]; + string var_25989_equation_0 = const()[name = string("op_25989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25989_cast_fp16 = einsum(equation = var_25989_equation_0, values = (var_25459_cast_fp16, var_25861_cast_fp16))[name = string("op_25989_cast_fp16")]; + string var_25991_equation_0 = const()[name = string("op_25991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25991_cast_fp16 = einsum(equation = var_25991_equation_0, values = (var_25463_cast_fp16, var_25862_cast_fp16))[name = string("op_25991_cast_fp16")]; + string var_25993_equation_0 = const()[name = string("op_25993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25993_cast_fp16 = einsum(equation = var_25993_equation_0, values = (var_25463_cast_fp16, var_25863_cast_fp16))[name = string("op_25993_cast_fp16")]; + string var_25995_equation_0 = const()[name = string("op_25995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25995_cast_fp16 = einsum(equation = var_25995_equation_0, values = (var_25463_cast_fp16, var_25864_cast_fp16))[name = string("op_25995_cast_fp16")]; + string var_25997_equation_0 = const()[name = string("op_25997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25997_cast_fp16 = einsum(equation = var_25997_equation_0, values = (var_25463_cast_fp16, var_25865_cast_fp16))[name = string("op_25997_cast_fp16")]; + string var_25999_equation_0 = const()[name = string("op_25999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_25999_cast_fp16 = einsum(equation = var_25999_equation_0, values = (var_25467_cast_fp16, var_25866_cast_fp16))[name = string("op_25999_cast_fp16")]; + string var_26001_equation_0 = const()[name = string("op_26001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26001_cast_fp16 = einsum(equation = var_26001_equation_0, values = (var_25467_cast_fp16, var_25867_cast_fp16))[name = string("op_26001_cast_fp16")]; + string var_26003_equation_0 = const()[name = string("op_26003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26003_cast_fp16 = einsum(equation = var_26003_equation_0, values = (var_25467_cast_fp16, var_25868_cast_fp16))[name = string("op_26003_cast_fp16")]; + string var_26005_equation_0 = const()[name = string("op_26005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26005_cast_fp16 = einsum(equation = var_26005_equation_0, values = (var_25467_cast_fp16, var_25869_cast_fp16))[name = string("op_26005_cast_fp16")]; + string var_26007_equation_0 = const()[name = string("op_26007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26007_cast_fp16 = einsum(equation = var_26007_equation_0, values = (var_25471_cast_fp16, var_25870_cast_fp16))[name = string("op_26007_cast_fp16")]; + string var_26009_equation_0 = const()[name = string("op_26009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26009_cast_fp16 = einsum(equation = var_26009_equation_0, values = (var_25471_cast_fp16, var_25871_cast_fp16))[name = string("op_26009_cast_fp16")]; + string var_26011_equation_0 = const()[name = string("op_26011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26011_cast_fp16 = einsum(equation = var_26011_equation_0, values = (var_25471_cast_fp16, var_25872_cast_fp16))[name = string("op_26011_cast_fp16")]; + string var_26013_equation_0 = const()[name = string("op_26013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26013_cast_fp16 = einsum(equation = var_26013_equation_0, values = (var_25471_cast_fp16, var_25873_cast_fp16))[name = string("op_26013_cast_fp16")]; + string var_26015_equation_0 = const()[name = string("op_26015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26015_cast_fp16 = einsum(equation = var_26015_equation_0, values = (var_25475_cast_fp16, var_25874_cast_fp16))[name = string("op_26015_cast_fp16")]; + string var_26017_equation_0 = const()[name = string("op_26017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26017_cast_fp16 = einsum(equation = var_26017_equation_0, values = (var_25475_cast_fp16, var_25875_cast_fp16))[name = string("op_26017_cast_fp16")]; + string var_26019_equation_0 = const()[name = string("op_26019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26019_cast_fp16 = einsum(equation = var_26019_equation_0, values = (var_25475_cast_fp16, var_25876_cast_fp16))[name = string("op_26019_cast_fp16")]; + string var_26021_equation_0 = const()[name = string("op_26021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26021_cast_fp16 = einsum(equation = var_26021_equation_0, values = (var_25475_cast_fp16, var_25877_cast_fp16))[name = string("op_26021_cast_fp16")]; + string var_26023_equation_0 = const()[name = string("op_26023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26023_cast_fp16 = einsum(equation = var_26023_equation_0, values = (var_25479_cast_fp16, var_25878_cast_fp16))[name = string("op_26023_cast_fp16")]; + string var_26025_equation_0 = const()[name = string("op_26025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26025_cast_fp16 = einsum(equation = var_26025_equation_0, values = (var_25479_cast_fp16, var_25879_cast_fp16))[name = string("op_26025_cast_fp16")]; + string var_26027_equation_0 = const()[name = string("op_26027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26027_cast_fp16 = einsum(equation = var_26027_equation_0, values = (var_25479_cast_fp16, var_25880_cast_fp16))[name = string("op_26027_cast_fp16")]; + string var_26029_equation_0 = const()[name = string("op_26029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26029_cast_fp16 = einsum(equation = var_26029_equation_0, values = (var_25479_cast_fp16, var_25881_cast_fp16))[name = string("op_26029_cast_fp16")]; + string var_26031_equation_0 = const()[name = string("op_26031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26031_cast_fp16 = einsum(equation = var_26031_equation_0, values = (var_25483_cast_fp16, var_25882_cast_fp16))[name = string("op_26031_cast_fp16")]; + string var_26033_equation_0 = const()[name = string("op_26033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26033_cast_fp16 = einsum(equation = var_26033_equation_0, values = (var_25483_cast_fp16, var_25883_cast_fp16))[name = string("op_26033_cast_fp16")]; + string var_26035_equation_0 = const()[name = string("op_26035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26035_cast_fp16 = einsum(equation = var_26035_equation_0, values = (var_25483_cast_fp16, var_25884_cast_fp16))[name = string("op_26035_cast_fp16")]; + string var_26037_equation_0 = const()[name = string("op_26037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26037_cast_fp16 = einsum(equation = var_26037_equation_0, values = (var_25483_cast_fp16, var_25885_cast_fp16))[name = string("op_26037_cast_fp16")]; + string var_26039_equation_0 = const()[name = string("op_26039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26039_cast_fp16 = einsum(equation = var_26039_equation_0, values = (var_25487_cast_fp16, var_25886_cast_fp16))[name = string("op_26039_cast_fp16")]; + string var_26041_equation_0 = const()[name = string("op_26041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26041_cast_fp16 = einsum(equation = var_26041_equation_0, values = (var_25487_cast_fp16, var_25887_cast_fp16))[name = string("op_26041_cast_fp16")]; + string var_26043_equation_0 = const()[name = string("op_26043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26043_cast_fp16 = einsum(equation = var_26043_equation_0, values = (var_25487_cast_fp16, var_25888_cast_fp16))[name = string("op_26043_cast_fp16")]; + string var_26045_equation_0 = const()[name = string("op_26045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26045_cast_fp16 = einsum(equation = var_26045_equation_0, values = (var_25487_cast_fp16, var_25889_cast_fp16))[name = string("op_26045_cast_fp16")]; + string var_26047_equation_0 = const()[name = string("op_26047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26047_cast_fp16 = einsum(equation = var_26047_equation_0, values = (var_25491_cast_fp16, var_25890_cast_fp16))[name = string("op_26047_cast_fp16")]; + string var_26049_equation_0 = const()[name = string("op_26049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26049_cast_fp16 = einsum(equation = var_26049_equation_0, values = (var_25491_cast_fp16, var_25891_cast_fp16))[name = string("op_26049_cast_fp16")]; + string var_26051_equation_0 = const()[name = string("op_26051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26051_cast_fp16 = einsum(equation = var_26051_equation_0, values = (var_25491_cast_fp16, var_25892_cast_fp16))[name = string("op_26051_cast_fp16")]; + string var_26053_equation_0 = const()[name = string("op_26053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_26053_cast_fp16 = einsum(equation = var_26053_equation_0, values = (var_25491_cast_fp16, var_25893_cast_fp16))[name = string("op_26053_cast_fp16")]; + bool var_26055_interleave_0 = const()[name = string("op_26055_interleave_0"), val = bool(false)]; + tensor var_26055_cast_fp16 = concat(axis = var_24614, interleave = var_26055_interleave_0, values = (var_25895_cast_fp16, var_25897_cast_fp16, var_25899_cast_fp16, var_25901_cast_fp16))[name = string("op_26055_cast_fp16")]; + bool var_26057_interleave_0 = const()[name = string("op_26057_interleave_0"), val = bool(false)]; + tensor var_26057_cast_fp16 = concat(axis = var_24614, interleave = var_26057_interleave_0, values = (var_25903_cast_fp16, var_25905_cast_fp16, var_25907_cast_fp16, var_25909_cast_fp16))[name = string("op_26057_cast_fp16")]; + bool var_26059_interleave_0 = const()[name = string("op_26059_interleave_0"), val = bool(false)]; + tensor var_26059_cast_fp16 = concat(axis = var_24614, interleave = var_26059_interleave_0, values = (var_25911_cast_fp16, var_25913_cast_fp16, var_25915_cast_fp16, var_25917_cast_fp16))[name = string("op_26059_cast_fp16")]; + bool var_26061_interleave_0 = const()[name = string("op_26061_interleave_0"), val = bool(false)]; + tensor var_26061_cast_fp16 = concat(axis = var_24614, interleave = var_26061_interleave_0, values = (var_25919_cast_fp16, var_25921_cast_fp16, var_25923_cast_fp16, var_25925_cast_fp16))[name = string("op_26061_cast_fp16")]; + bool var_26063_interleave_0 = const()[name = string("op_26063_interleave_0"), val = bool(false)]; + tensor var_26063_cast_fp16 = concat(axis = var_24614, interleave = var_26063_interleave_0, values = (var_25927_cast_fp16, var_25929_cast_fp16, var_25931_cast_fp16, var_25933_cast_fp16))[name = string("op_26063_cast_fp16")]; + bool var_26065_interleave_0 = const()[name = string("op_26065_interleave_0"), val = bool(false)]; + tensor var_26065_cast_fp16 = concat(axis = var_24614, interleave = var_26065_interleave_0, values = (var_25935_cast_fp16, var_25937_cast_fp16, var_25939_cast_fp16, var_25941_cast_fp16))[name = string("op_26065_cast_fp16")]; + bool var_26067_interleave_0 = const()[name = string("op_26067_interleave_0"), val = bool(false)]; + tensor var_26067_cast_fp16 = concat(axis = var_24614, interleave = var_26067_interleave_0, values = (var_25943_cast_fp16, var_25945_cast_fp16, var_25947_cast_fp16, var_25949_cast_fp16))[name = string("op_26067_cast_fp16")]; + bool var_26069_interleave_0 = const()[name = string("op_26069_interleave_0"), val = bool(false)]; + tensor var_26069_cast_fp16 = concat(axis = var_24614, interleave = var_26069_interleave_0, values = (var_25951_cast_fp16, var_25953_cast_fp16, var_25955_cast_fp16, var_25957_cast_fp16))[name = string("op_26069_cast_fp16")]; + bool var_26071_interleave_0 = const()[name = string("op_26071_interleave_0"), val = bool(false)]; + tensor var_26071_cast_fp16 = concat(axis = var_24614, interleave = var_26071_interleave_0, values = (var_25959_cast_fp16, var_25961_cast_fp16, var_25963_cast_fp16, var_25965_cast_fp16))[name = string("op_26071_cast_fp16")]; + bool var_26073_interleave_0 = const()[name = string("op_26073_interleave_0"), val = bool(false)]; + tensor var_26073_cast_fp16 = concat(axis = var_24614, interleave = var_26073_interleave_0, values = (var_25967_cast_fp16, var_25969_cast_fp16, var_25971_cast_fp16, var_25973_cast_fp16))[name = string("op_26073_cast_fp16")]; + bool var_26075_interleave_0 = const()[name = string("op_26075_interleave_0"), val = bool(false)]; + tensor var_26075_cast_fp16 = concat(axis = var_24614, interleave = var_26075_interleave_0, values = (var_25975_cast_fp16, var_25977_cast_fp16, var_25979_cast_fp16, var_25981_cast_fp16))[name = string("op_26075_cast_fp16")]; + bool var_26077_interleave_0 = const()[name = string("op_26077_interleave_0"), val = bool(false)]; + tensor var_26077_cast_fp16 = concat(axis = var_24614, interleave = var_26077_interleave_0, values = (var_25983_cast_fp16, var_25985_cast_fp16, var_25987_cast_fp16, var_25989_cast_fp16))[name = string("op_26077_cast_fp16")]; + bool var_26079_interleave_0 = const()[name = string("op_26079_interleave_0"), val = bool(false)]; + tensor var_26079_cast_fp16 = concat(axis = var_24614, interleave = var_26079_interleave_0, values = (var_25991_cast_fp16, var_25993_cast_fp16, var_25995_cast_fp16, var_25997_cast_fp16))[name = string("op_26079_cast_fp16")]; + bool var_26081_interleave_0 = const()[name = string("op_26081_interleave_0"), val = bool(false)]; + tensor var_26081_cast_fp16 = concat(axis = var_24614, interleave = var_26081_interleave_0, values = (var_25999_cast_fp16, var_26001_cast_fp16, var_26003_cast_fp16, var_26005_cast_fp16))[name = string("op_26081_cast_fp16")]; + bool var_26083_interleave_0 = const()[name = string("op_26083_interleave_0"), val = bool(false)]; + tensor var_26083_cast_fp16 = concat(axis = var_24614, interleave = var_26083_interleave_0, values = (var_26007_cast_fp16, var_26009_cast_fp16, var_26011_cast_fp16, var_26013_cast_fp16))[name = string("op_26083_cast_fp16")]; + bool var_26085_interleave_0 = const()[name = string("op_26085_interleave_0"), val = bool(false)]; + tensor var_26085_cast_fp16 = concat(axis = var_24614, interleave = var_26085_interleave_0, values = (var_26015_cast_fp16, var_26017_cast_fp16, var_26019_cast_fp16, var_26021_cast_fp16))[name = string("op_26085_cast_fp16")]; + bool var_26087_interleave_0 = const()[name = string("op_26087_interleave_0"), val = bool(false)]; + tensor var_26087_cast_fp16 = concat(axis = var_24614, interleave = var_26087_interleave_0, values = (var_26023_cast_fp16, var_26025_cast_fp16, var_26027_cast_fp16, var_26029_cast_fp16))[name = string("op_26087_cast_fp16")]; + bool var_26089_interleave_0 = const()[name = string("op_26089_interleave_0"), val = bool(false)]; + tensor var_26089_cast_fp16 = concat(axis = var_24614, interleave = var_26089_interleave_0, values = (var_26031_cast_fp16, var_26033_cast_fp16, var_26035_cast_fp16, var_26037_cast_fp16))[name = string("op_26089_cast_fp16")]; + bool var_26091_interleave_0 = const()[name = string("op_26091_interleave_0"), val = bool(false)]; + tensor var_26091_cast_fp16 = concat(axis = var_24614, interleave = var_26091_interleave_0, values = (var_26039_cast_fp16, var_26041_cast_fp16, var_26043_cast_fp16, var_26045_cast_fp16))[name = string("op_26091_cast_fp16")]; + bool var_26093_interleave_0 = const()[name = string("op_26093_interleave_0"), val = bool(false)]; + tensor var_26093_cast_fp16 = concat(axis = var_24614, interleave = var_26093_interleave_0, values = (var_26047_cast_fp16, var_26049_cast_fp16, var_26051_cast_fp16, var_26053_cast_fp16))[name = string("op_26093_cast_fp16")]; + bool input_129_interleave_0 = const()[name = string("input_129_interleave_0"), val = bool(false)]; + tensor input_129_cast_fp16 = concat(axis = var_24639, interleave = input_129_interleave_0, values = (var_26055_cast_fp16, var_26057_cast_fp16, var_26059_cast_fp16, var_26061_cast_fp16, var_26063_cast_fp16, var_26065_cast_fp16, var_26067_cast_fp16, var_26069_cast_fp16, var_26071_cast_fp16, var_26073_cast_fp16, var_26075_cast_fp16, var_26077_cast_fp16, var_26079_cast_fp16, var_26081_cast_fp16, var_26083_cast_fp16, var_26085_cast_fp16, var_26087_cast_fp16, var_26089_cast_fp16, var_26091_cast_fp16, var_26093_cast_fp16))[name = string("input_129_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(654157760)))]; + tensor layers_16_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657434624)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_129_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_67_cast_fp16")]; + tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; + fp16 var_26112_to_fp16 = const()[name = string("op_26112_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_26112_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; + tensor input_131_gamma_0_to_fp16 = const()[name = string("input_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657437248)))]; + tensor input_131_beta_0_to_fp16 = const()[name = string("input_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657439872)))]; + fp16 input_131_epsilon_0_to_fp16 = const()[name = string("input_131_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("input_131_cast_fp16")]; + string input_133_pad_type_0 = const()[name = string("input_133_pad_type_0"), val = string("valid")]; + tensor input_133_strides_0 = const()[name = string("input_133_strides_0"), val = tensor([1, 1])]; + tensor input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_133_dilations_0 = const()[name = string("input_133_dilations_0"), val = tensor([1, 1])]; + int32 input_133_groups_0 = const()[name = string("input_133_groups_0"), val = int32(1)]; + tensor layers_16_fc1_weight_to_fp16 = const()[name = string("layers_16_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(657442496)))]; + tensor layers_16_fc1_bias_to_fp16 = const()[name = string("layers_16_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670549760)))]; + tensor input_133_cast_fp16 = conv(bias = layers_16_fc1_bias_to_fp16, dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = layers_16_fc1_weight_to_fp16, x = input_131_cast_fp16)[name = string("input_133_cast_fp16")]; + string input_135_mode_0 = const()[name = string("input_135_mode_0"), val = string("EXACT")]; + tensor input_135_cast_fp16 = gelu(mode = input_135_mode_0, x = input_133_cast_fp16)[name = string("input_135_cast_fp16")]; + string hidden_states_37_pad_type_0 = const()[name = string("hidden_states_37_pad_type_0"), val = string("valid")]; + tensor hidden_states_37_strides_0 = const()[name = string("hidden_states_37_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_37_pad_0 = const()[name = string("hidden_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_37_dilations_0 = const()[name = string("hidden_states_37_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_37_groups_0 = const()[name = string("hidden_states_37_groups_0"), val = int32(1)]; + tensor layers_16_fc2_weight_to_fp16 = const()[name = string("layers_16_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(670560064)))]; + tensor layers_16_fc2_bias_to_fp16 = const()[name = string("layers_16_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683667328)))]; + tensor hidden_states_37_cast_fp16 = conv(bias = layers_16_fc2_bias_to_fp16, dilations = hidden_states_37_dilations_0, groups = hidden_states_37_groups_0, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = hidden_states_37_strides_0, weight = layers_16_fc2_weight_to_fp16, x = input_135_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = hidden_states_37_cast_fp16)[name = string("inputs_69_cast_fp16")]; + int32 var_26141 = const()[name = string("op_26141"), val = int32(3)]; + int32 var_26166 = const()[name = string("op_26166"), val = int32(1)]; + tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; + fp16 var_26183_to_fp16 = const()[name = string("op_26183_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_26183_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683669952)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683672576)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; + tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; + tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; + int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(683675200)))]; + tensor layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686952064)))]; + tensor query_35_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_35_cast_fp16")]; + string key_35_pad_type_0 = const()[name = string("key_35_pad_type_0"), val = string("valid")]; + tensor key_35_strides_0 = const()[name = string("key_35_strides_0"), val = tensor([1, 1])]; + tensor key_35_pad_0 = const()[name = string("key_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_35_dilations_0 = const()[name = string("key_35_dilations_0"), val = tensor([1, 1])]; + int32 key_35_groups_0 = const()[name = string("key_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(686954688)))]; + tensor key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("key_35_cast_fp16")]; + string value_35_pad_type_0 = const()[name = string("value_35_pad_type_0"), val = string("valid")]; + tensor value_35_strides_0 = const()[name = string("value_35_strides_0"), val = tensor([1, 1])]; + tensor value_35_pad_0 = const()[name = string("value_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_35_dilations_0 = const()[name = string("value_35_dilations_0"), val = tensor([1, 1])]; + int32 value_35_groups_0 = const()[name = string("value_35_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(690231552)))]; + tensor layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693508416)))]; + tensor value_35_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("value_35_cast_fp16")]; + tensor var_26221_begin_0 = const()[name = string("op_26221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26221_end_0 = const()[name = string("op_26221_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26221_end_mask_0 = const()[name = string("op_26221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26221_cast_fp16 = slice_by_index(begin = var_26221_begin_0, end = var_26221_end_0, end_mask = var_26221_end_mask_0, x = query_35_cast_fp16)[name = string("op_26221_cast_fp16")]; + tensor var_26225_begin_0 = const()[name = string("op_26225_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26225_end_0 = const()[name = string("op_26225_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_26225_end_mask_0 = const()[name = string("op_26225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26225_cast_fp16 = slice_by_index(begin = var_26225_begin_0, end = var_26225_end_0, end_mask = var_26225_end_mask_0, x = query_35_cast_fp16)[name = string("op_26225_cast_fp16")]; + tensor var_26229_begin_0 = const()[name = string("op_26229_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26229_end_0 = const()[name = string("op_26229_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_26229_end_mask_0 = const()[name = string("op_26229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26229_cast_fp16 = slice_by_index(begin = var_26229_begin_0, end = var_26229_end_0, end_mask = var_26229_end_mask_0, x = query_35_cast_fp16)[name = string("op_26229_cast_fp16")]; + tensor var_26233_begin_0 = const()[name = string("op_26233_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26233_end_0 = const()[name = string("op_26233_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_26233_end_mask_0 = const()[name = string("op_26233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26233_cast_fp16 = slice_by_index(begin = var_26233_begin_0, end = var_26233_end_0, end_mask = var_26233_end_mask_0, x = query_35_cast_fp16)[name = string("op_26233_cast_fp16")]; + tensor var_26237_begin_0 = const()[name = string("op_26237_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26237_end_0 = const()[name = string("op_26237_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_26237_end_mask_0 = const()[name = string("op_26237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26237_cast_fp16 = slice_by_index(begin = var_26237_begin_0, end = var_26237_end_0, end_mask = var_26237_end_mask_0, x = query_35_cast_fp16)[name = string("op_26237_cast_fp16")]; + tensor var_26241_begin_0 = const()[name = string("op_26241_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26241_end_0 = const()[name = string("op_26241_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_26241_end_mask_0 = const()[name = string("op_26241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26241_cast_fp16 = slice_by_index(begin = var_26241_begin_0, end = var_26241_end_0, end_mask = var_26241_end_mask_0, x = query_35_cast_fp16)[name = string("op_26241_cast_fp16")]; + tensor var_26245_begin_0 = const()[name = string("op_26245_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26245_end_0 = const()[name = string("op_26245_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_26245_end_mask_0 = const()[name = string("op_26245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26245_cast_fp16 = slice_by_index(begin = var_26245_begin_0, end = var_26245_end_0, end_mask = var_26245_end_mask_0, x = query_35_cast_fp16)[name = string("op_26245_cast_fp16")]; + tensor var_26249_begin_0 = const()[name = string("op_26249_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26249_end_0 = const()[name = string("op_26249_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_26249_end_mask_0 = const()[name = string("op_26249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26249_cast_fp16 = slice_by_index(begin = var_26249_begin_0, end = var_26249_end_0, end_mask = var_26249_end_mask_0, x = query_35_cast_fp16)[name = string("op_26249_cast_fp16")]; + tensor var_26253_begin_0 = const()[name = string("op_26253_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26253_end_0 = const()[name = string("op_26253_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_26253_end_mask_0 = const()[name = string("op_26253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26253_cast_fp16 = slice_by_index(begin = var_26253_begin_0, end = var_26253_end_0, end_mask = var_26253_end_mask_0, x = query_35_cast_fp16)[name = string("op_26253_cast_fp16")]; + tensor var_26257_begin_0 = const()[name = string("op_26257_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26257_end_0 = const()[name = string("op_26257_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_26257_end_mask_0 = const()[name = string("op_26257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26257_cast_fp16 = slice_by_index(begin = var_26257_begin_0, end = var_26257_end_0, end_mask = var_26257_end_mask_0, x = query_35_cast_fp16)[name = string("op_26257_cast_fp16")]; + tensor var_26261_begin_0 = const()[name = string("op_26261_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26261_end_0 = const()[name = string("op_26261_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_26261_end_mask_0 = const()[name = string("op_26261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26261_cast_fp16 = slice_by_index(begin = var_26261_begin_0, end = var_26261_end_0, end_mask = var_26261_end_mask_0, x = query_35_cast_fp16)[name = string("op_26261_cast_fp16")]; + tensor var_26265_begin_0 = const()[name = string("op_26265_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26265_end_0 = const()[name = string("op_26265_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_26265_end_mask_0 = const()[name = string("op_26265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26265_cast_fp16 = slice_by_index(begin = var_26265_begin_0, end = var_26265_end_0, end_mask = var_26265_end_mask_0, x = query_35_cast_fp16)[name = string("op_26265_cast_fp16")]; + tensor var_26269_begin_0 = const()[name = string("op_26269_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26269_end_0 = const()[name = string("op_26269_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_26269_end_mask_0 = const()[name = string("op_26269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26269_cast_fp16 = slice_by_index(begin = var_26269_begin_0, end = var_26269_end_0, end_mask = var_26269_end_mask_0, x = query_35_cast_fp16)[name = string("op_26269_cast_fp16")]; + tensor var_26273_begin_0 = const()[name = string("op_26273_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26273_end_0 = const()[name = string("op_26273_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_26273_end_mask_0 = const()[name = string("op_26273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26273_cast_fp16 = slice_by_index(begin = var_26273_begin_0, end = var_26273_end_0, end_mask = var_26273_end_mask_0, x = query_35_cast_fp16)[name = string("op_26273_cast_fp16")]; + tensor var_26277_begin_0 = const()[name = string("op_26277_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26277_end_0 = const()[name = string("op_26277_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_26277_end_mask_0 = const()[name = string("op_26277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26277_cast_fp16 = slice_by_index(begin = var_26277_begin_0, end = var_26277_end_0, end_mask = var_26277_end_mask_0, x = query_35_cast_fp16)[name = string("op_26277_cast_fp16")]; + tensor var_26281_begin_0 = const()[name = string("op_26281_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26281_end_0 = const()[name = string("op_26281_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_26281_end_mask_0 = const()[name = string("op_26281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26281_cast_fp16 = slice_by_index(begin = var_26281_begin_0, end = var_26281_end_0, end_mask = var_26281_end_mask_0, x = query_35_cast_fp16)[name = string("op_26281_cast_fp16")]; + tensor var_26285_begin_0 = const()[name = string("op_26285_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26285_end_0 = const()[name = string("op_26285_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_26285_end_mask_0 = const()[name = string("op_26285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26285_cast_fp16 = slice_by_index(begin = var_26285_begin_0, end = var_26285_end_0, end_mask = var_26285_end_mask_0, x = query_35_cast_fp16)[name = string("op_26285_cast_fp16")]; + tensor var_26289_begin_0 = const()[name = string("op_26289_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26289_end_0 = const()[name = string("op_26289_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_26289_end_mask_0 = const()[name = string("op_26289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26289_cast_fp16 = slice_by_index(begin = var_26289_begin_0, end = var_26289_end_0, end_mask = var_26289_end_mask_0, x = query_35_cast_fp16)[name = string("op_26289_cast_fp16")]; + tensor var_26293_begin_0 = const()[name = string("op_26293_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26293_end_0 = const()[name = string("op_26293_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_26293_end_mask_0 = const()[name = string("op_26293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26293_cast_fp16 = slice_by_index(begin = var_26293_begin_0, end = var_26293_end_0, end_mask = var_26293_end_mask_0, x = query_35_cast_fp16)[name = string("op_26293_cast_fp16")]; + tensor var_26297_begin_0 = const()[name = string("op_26297_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26297_end_0 = const()[name = string("op_26297_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_26297_end_mask_0 = const()[name = string("op_26297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26297_cast_fp16 = slice_by_index(begin = var_26297_begin_0, end = var_26297_end_0, end_mask = var_26297_end_mask_0, x = query_35_cast_fp16)[name = string("op_26297_cast_fp16")]; + tensor var_26306_begin_0 = const()[name = string("op_26306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26306_end_0 = const()[name = string("op_26306_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26306_end_mask_0 = const()[name = string("op_26306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26306_cast_fp16 = slice_by_index(begin = var_26306_begin_0, end = var_26306_end_0, end_mask = var_26306_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26306_cast_fp16")]; + tensor var_26313_begin_0 = const()[name = string("op_26313_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26313_end_0 = const()[name = string("op_26313_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26313_end_mask_0 = const()[name = string("op_26313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26313_cast_fp16 = slice_by_index(begin = var_26313_begin_0, end = var_26313_end_0, end_mask = var_26313_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26313_cast_fp16")]; + tensor var_26320_begin_0 = const()[name = string("op_26320_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26320_end_0 = const()[name = string("op_26320_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26320_end_mask_0 = const()[name = string("op_26320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26320_cast_fp16 = slice_by_index(begin = var_26320_begin_0, end = var_26320_end_0, end_mask = var_26320_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26320_cast_fp16")]; + tensor var_26327_begin_0 = const()[name = string("op_26327_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26327_end_0 = const()[name = string("op_26327_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26327_end_mask_0 = const()[name = string("op_26327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26327_cast_fp16 = slice_by_index(begin = var_26327_begin_0, end = var_26327_end_0, end_mask = var_26327_end_mask_0, x = var_26221_cast_fp16)[name = string("op_26327_cast_fp16")]; + tensor var_26334_begin_0 = const()[name = string("op_26334_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26334_end_0 = const()[name = string("op_26334_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26334_end_mask_0 = const()[name = string("op_26334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26334_cast_fp16 = slice_by_index(begin = var_26334_begin_0, end = var_26334_end_0, end_mask = var_26334_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26334_cast_fp16")]; + tensor var_26341_begin_0 = const()[name = string("op_26341_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26341_end_0 = const()[name = string("op_26341_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26341_end_mask_0 = const()[name = string("op_26341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26341_cast_fp16 = slice_by_index(begin = var_26341_begin_0, end = var_26341_end_0, end_mask = var_26341_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26341_cast_fp16")]; + tensor var_26348_begin_0 = const()[name = string("op_26348_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26348_end_0 = const()[name = string("op_26348_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26348_end_mask_0 = const()[name = string("op_26348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26348_cast_fp16 = slice_by_index(begin = var_26348_begin_0, end = var_26348_end_0, end_mask = var_26348_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26348_cast_fp16")]; + tensor var_26355_begin_0 = const()[name = string("op_26355_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26355_end_0 = const()[name = string("op_26355_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26355_end_mask_0 = const()[name = string("op_26355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26355_cast_fp16 = slice_by_index(begin = var_26355_begin_0, end = var_26355_end_0, end_mask = var_26355_end_mask_0, x = var_26225_cast_fp16)[name = string("op_26355_cast_fp16")]; + tensor var_26362_begin_0 = const()[name = string("op_26362_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26362_end_0 = const()[name = string("op_26362_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26362_end_mask_0 = const()[name = string("op_26362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26362_cast_fp16 = slice_by_index(begin = var_26362_begin_0, end = var_26362_end_0, end_mask = var_26362_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26362_cast_fp16")]; + tensor var_26369_begin_0 = const()[name = string("op_26369_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26369_end_0 = const()[name = string("op_26369_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26369_end_mask_0 = const()[name = string("op_26369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26369_cast_fp16 = slice_by_index(begin = var_26369_begin_0, end = var_26369_end_0, end_mask = var_26369_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26369_cast_fp16")]; + tensor var_26376_begin_0 = const()[name = string("op_26376_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26376_end_0 = const()[name = string("op_26376_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26376_end_mask_0 = const()[name = string("op_26376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26376_cast_fp16 = slice_by_index(begin = var_26376_begin_0, end = var_26376_end_0, end_mask = var_26376_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26376_cast_fp16")]; + tensor var_26383_begin_0 = const()[name = string("op_26383_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26383_end_0 = const()[name = string("op_26383_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26383_end_mask_0 = const()[name = string("op_26383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26383_cast_fp16 = slice_by_index(begin = var_26383_begin_0, end = var_26383_end_0, end_mask = var_26383_end_mask_0, x = var_26229_cast_fp16)[name = string("op_26383_cast_fp16")]; + tensor var_26390_begin_0 = const()[name = string("op_26390_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26390_end_0 = const()[name = string("op_26390_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26390_end_mask_0 = const()[name = string("op_26390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26390_cast_fp16 = slice_by_index(begin = var_26390_begin_0, end = var_26390_end_0, end_mask = var_26390_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26390_cast_fp16")]; + tensor var_26397_begin_0 = const()[name = string("op_26397_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26397_end_0 = const()[name = string("op_26397_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26397_end_mask_0 = const()[name = string("op_26397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26397_cast_fp16 = slice_by_index(begin = var_26397_begin_0, end = var_26397_end_0, end_mask = var_26397_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26397_cast_fp16")]; + tensor var_26404_begin_0 = const()[name = string("op_26404_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26404_end_0 = const()[name = string("op_26404_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26404_end_mask_0 = const()[name = string("op_26404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26404_cast_fp16 = slice_by_index(begin = var_26404_begin_0, end = var_26404_end_0, end_mask = var_26404_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26404_cast_fp16")]; + tensor var_26411_begin_0 = const()[name = string("op_26411_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26411_end_0 = const()[name = string("op_26411_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26411_end_mask_0 = const()[name = string("op_26411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26411_cast_fp16 = slice_by_index(begin = var_26411_begin_0, end = var_26411_end_0, end_mask = var_26411_end_mask_0, x = var_26233_cast_fp16)[name = string("op_26411_cast_fp16")]; + tensor var_26418_begin_0 = const()[name = string("op_26418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26418_end_0 = const()[name = string("op_26418_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26418_end_mask_0 = const()[name = string("op_26418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26418_cast_fp16 = slice_by_index(begin = var_26418_begin_0, end = var_26418_end_0, end_mask = var_26418_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26418_cast_fp16")]; + tensor var_26425_begin_0 = const()[name = string("op_26425_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26425_end_0 = const()[name = string("op_26425_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26425_end_mask_0 = const()[name = string("op_26425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26425_cast_fp16 = slice_by_index(begin = var_26425_begin_0, end = var_26425_end_0, end_mask = var_26425_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26425_cast_fp16")]; + tensor var_26432_begin_0 = const()[name = string("op_26432_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26432_end_0 = const()[name = string("op_26432_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26432_end_mask_0 = const()[name = string("op_26432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26432_cast_fp16 = slice_by_index(begin = var_26432_begin_0, end = var_26432_end_0, end_mask = var_26432_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26432_cast_fp16")]; + tensor var_26439_begin_0 = const()[name = string("op_26439_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26439_end_0 = const()[name = string("op_26439_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26439_end_mask_0 = const()[name = string("op_26439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26439_cast_fp16 = slice_by_index(begin = var_26439_begin_0, end = var_26439_end_0, end_mask = var_26439_end_mask_0, x = var_26237_cast_fp16)[name = string("op_26439_cast_fp16")]; + tensor var_26446_begin_0 = const()[name = string("op_26446_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26446_end_0 = const()[name = string("op_26446_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26446_end_mask_0 = const()[name = string("op_26446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26446_cast_fp16 = slice_by_index(begin = var_26446_begin_0, end = var_26446_end_0, end_mask = var_26446_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26446_cast_fp16")]; + tensor var_26453_begin_0 = const()[name = string("op_26453_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26453_end_0 = const()[name = string("op_26453_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26453_end_mask_0 = const()[name = string("op_26453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26453_cast_fp16 = slice_by_index(begin = var_26453_begin_0, end = var_26453_end_0, end_mask = var_26453_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26453_cast_fp16")]; + tensor var_26460_begin_0 = const()[name = string("op_26460_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26460_end_0 = const()[name = string("op_26460_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26460_end_mask_0 = const()[name = string("op_26460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26460_cast_fp16 = slice_by_index(begin = var_26460_begin_0, end = var_26460_end_0, end_mask = var_26460_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26460_cast_fp16")]; + tensor var_26467_begin_0 = const()[name = string("op_26467_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26467_end_0 = const()[name = string("op_26467_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26467_end_mask_0 = const()[name = string("op_26467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26467_cast_fp16 = slice_by_index(begin = var_26467_begin_0, end = var_26467_end_0, end_mask = var_26467_end_mask_0, x = var_26241_cast_fp16)[name = string("op_26467_cast_fp16")]; + tensor var_26474_begin_0 = const()[name = string("op_26474_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26474_end_0 = const()[name = string("op_26474_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26474_end_mask_0 = const()[name = string("op_26474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26474_cast_fp16 = slice_by_index(begin = var_26474_begin_0, end = var_26474_end_0, end_mask = var_26474_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26474_cast_fp16")]; + tensor var_26481_begin_0 = const()[name = string("op_26481_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26481_end_0 = const()[name = string("op_26481_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26481_end_mask_0 = const()[name = string("op_26481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26481_cast_fp16 = slice_by_index(begin = var_26481_begin_0, end = var_26481_end_0, end_mask = var_26481_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26481_cast_fp16")]; + tensor var_26488_begin_0 = const()[name = string("op_26488_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26488_end_0 = const()[name = string("op_26488_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26488_end_mask_0 = const()[name = string("op_26488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26488_cast_fp16 = slice_by_index(begin = var_26488_begin_0, end = var_26488_end_0, end_mask = var_26488_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26488_cast_fp16")]; + tensor var_26495_begin_0 = const()[name = string("op_26495_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26495_end_0 = const()[name = string("op_26495_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26495_end_mask_0 = const()[name = string("op_26495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26495_cast_fp16 = slice_by_index(begin = var_26495_begin_0, end = var_26495_end_0, end_mask = var_26495_end_mask_0, x = var_26245_cast_fp16)[name = string("op_26495_cast_fp16")]; + tensor var_26502_begin_0 = const()[name = string("op_26502_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26502_end_0 = const()[name = string("op_26502_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26502_end_mask_0 = const()[name = string("op_26502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26502_cast_fp16 = slice_by_index(begin = var_26502_begin_0, end = var_26502_end_0, end_mask = var_26502_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26502_cast_fp16")]; + tensor var_26509_begin_0 = const()[name = string("op_26509_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26509_end_0 = const()[name = string("op_26509_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26509_end_mask_0 = const()[name = string("op_26509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26509_cast_fp16 = slice_by_index(begin = var_26509_begin_0, end = var_26509_end_0, end_mask = var_26509_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26509_cast_fp16")]; + tensor var_26516_begin_0 = const()[name = string("op_26516_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26516_end_0 = const()[name = string("op_26516_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26516_end_mask_0 = const()[name = string("op_26516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26516_cast_fp16 = slice_by_index(begin = var_26516_begin_0, end = var_26516_end_0, end_mask = var_26516_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26516_cast_fp16")]; + tensor var_26523_begin_0 = const()[name = string("op_26523_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26523_end_0 = const()[name = string("op_26523_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26523_end_mask_0 = const()[name = string("op_26523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26523_cast_fp16 = slice_by_index(begin = var_26523_begin_0, end = var_26523_end_0, end_mask = var_26523_end_mask_0, x = var_26249_cast_fp16)[name = string("op_26523_cast_fp16")]; + tensor var_26530_begin_0 = const()[name = string("op_26530_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26530_end_0 = const()[name = string("op_26530_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26530_end_mask_0 = const()[name = string("op_26530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26530_cast_fp16 = slice_by_index(begin = var_26530_begin_0, end = var_26530_end_0, end_mask = var_26530_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26530_cast_fp16")]; + tensor var_26537_begin_0 = const()[name = string("op_26537_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26537_end_0 = const()[name = string("op_26537_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26537_end_mask_0 = const()[name = string("op_26537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26537_cast_fp16 = slice_by_index(begin = var_26537_begin_0, end = var_26537_end_0, end_mask = var_26537_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26537_cast_fp16")]; + tensor var_26544_begin_0 = const()[name = string("op_26544_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26544_end_0 = const()[name = string("op_26544_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26544_end_mask_0 = const()[name = string("op_26544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26544_cast_fp16 = slice_by_index(begin = var_26544_begin_0, end = var_26544_end_0, end_mask = var_26544_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26544_cast_fp16")]; + tensor var_26551_begin_0 = const()[name = string("op_26551_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26551_end_0 = const()[name = string("op_26551_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26551_end_mask_0 = const()[name = string("op_26551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26551_cast_fp16 = slice_by_index(begin = var_26551_begin_0, end = var_26551_end_0, end_mask = var_26551_end_mask_0, x = var_26253_cast_fp16)[name = string("op_26551_cast_fp16")]; + tensor var_26558_begin_0 = const()[name = string("op_26558_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26558_end_0 = const()[name = string("op_26558_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26558_end_mask_0 = const()[name = string("op_26558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26558_cast_fp16 = slice_by_index(begin = var_26558_begin_0, end = var_26558_end_0, end_mask = var_26558_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26558_cast_fp16")]; + tensor var_26565_begin_0 = const()[name = string("op_26565_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26565_end_0 = const()[name = string("op_26565_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26565_end_mask_0 = const()[name = string("op_26565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26565_cast_fp16 = slice_by_index(begin = var_26565_begin_0, end = var_26565_end_0, end_mask = var_26565_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26565_cast_fp16")]; + tensor var_26572_begin_0 = const()[name = string("op_26572_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26572_end_0 = const()[name = string("op_26572_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26572_end_mask_0 = const()[name = string("op_26572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26572_cast_fp16 = slice_by_index(begin = var_26572_begin_0, end = var_26572_end_0, end_mask = var_26572_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26572_cast_fp16")]; + tensor var_26579_begin_0 = const()[name = string("op_26579_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26579_end_0 = const()[name = string("op_26579_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26579_end_mask_0 = const()[name = string("op_26579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26579_cast_fp16 = slice_by_index(begin = var_26579_begin_0, end = var_26579_end_0, end_mask = var_26579_end_mask_0, x = var_26257_cast_fp16)[name = string("op_26579_cast_fp16")]; + tensor var_26586_begin_0 = const()[name = string("op_26586_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26586_end_0 = const()[name = string("op_26586_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26586_end_mask_0 = const()[name = string("op_26586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26586_cast_fp16 = slice_by_index(begin = var_26586_begin_0, end = var_26586_end_0, end_mask = var_26586_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26586_cast_fp16")]; + tensor var_26593_begin_0 = const()[name = string("op_26593_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26593_end_0 = const()[name = string("op_26593_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26593_end_mask_0 = const()[name = string("op_26593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26593_cast_fp16 = slice_by_index(begin = var_26593_begin_0, end = var_26593_end_0, end_mask = var_26593_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26593_cast_fp16")]; + tensor var_26600_begin_0 = const()[name = string("op_26600_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26600_end_0 = const()[name = string("op_26600_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26600_end_mask_0 = const()[name = string("op_26600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26600_cast_fp16 = slice_by_index(begin = var_26600_begin_0, end = var_26600_end_0, end_mask = var_26600_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26600_cast_fp16")]; + tensor var_26607_begin_0 = const()[name = string("op_26607_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26607_end_0 = const()[name = string("op_26607_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26607_end_mask_0 = const()[name = string("op_26607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26607_cast_fp16 = slice_by_index(begin = var_26607_begin_0, end = var_26607_end_0, end_mask = var_26607_end_mask_0, x = var_26261_cast_fp16)[name = string("op_26607_cast_fp16")]; + tensor var_26614_begin_0 = const()[name = string("op_26614_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26614_end_0 = const()[name = string("op_26614_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26614_end_mask_0 = const()[name = string("op_26614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26614_cast_fp16 = slice_by_index(begin = var_26614_begin_0, end = var_26614_end_0, end_mask = var_26614_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26614_cast_fp16")]; + tensor var_26621_begin_0 = const()[name = string("op_26621_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26621_end_0 = const()[name = string("op_26621_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26621_end_mask_0 = const()[name = string("op_26621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26621_cast_fp16 = slice_by_index(begin = var_26621_begin_0, end = var_26621_end_0, end_mask = var_26621_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26621_cast_fp16")]; + tensor var_26628_begin_0 = const()[name = string("op_26628_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26628_end_0 = const()[name = string("op_26628_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26628_end_mask_0 = const()[name = string("op_26628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26628_cast_fp16 = slice_by_index(begin = var_26628_begin_0, end = var_26628_end_0, end_mask = var_26628_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26628_cast_fp16")]; + tensor var_26635_begin_0 = const()[name = string("op_26635_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26635_end_0 = const()[name = string("op_26635_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26635_end_mask_0 = const()[name = string("op_26635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26635_cast_fp16 = slice_by_index(begin = var_26635_begin_0, end = var_26635_end_0, end_mask = var_26635_end_mask_0, x = var_26265_cast_fp16)[name = string("op_26635_cast_fp16")]; + tensor var_26642_begin_0 = const()[name = string("op_26642_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26642_end_0 = const()[name = string("op_26642_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26642_end_mask_0 = const()[name = string("op_26642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26642_cast_fp16 = slice_by_index(begin = var_26642_begin_0, end = var_26642_end_0, end_mask = var_26642_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26642_cast_fp16")]; + tensor var_26649_begin_0 = const()[name = string("op_26649_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26649_end_0 = const()[name = string("op_26649_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26649_end_mask_0 = const()[name = string("op_26649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26649_cast_fp16 = slice_by_index(begin = var_26649_begin_0, end = var_26649_end_0, end_mask = var_26649_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26649_cast_fp16")]; + tensor var_26656_begin_0 = const()[name = string("op_26656_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26656_end_0 = const()[name = string("op_26656_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26656_end_mask_0 = const()[name = string("op_26656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26656_cast_fp16 = slice_by_index(begin = var_26656_begin_0, end = var_26656_end_0, end_mask = var_26656_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26656_cast_fp16")]; + tensor var_26663_begin_0 = const()[name = string("op_26663_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26663_end_0 = const()[name = string("op_26663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26663_end_mask_0 = const()[name = string("op_26663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26663_cast_fp16 = slice_by_index(begin = var_26663_begin_0, end = var_26663_end_0, end_mask = var_26663_end_mask_0, x = var_26269_cast_fp16)[name = string("op_26663_cast_fp16")]; + tensor var_26670_begin_0 = const()[name = string("op_26670_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26670_end_0 = const()[name = string("op_26670_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26670_end_mask_0 = const()[name = string("op_26670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26670_cast_fp16 = slice_by_index(begin = var_26670_begin_0, end = var_26670_end_0, end_mask = var_26670_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26670_cast_fp16")]; + tensor var_26677_begin_0 = const()[name = string("op_26677_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26677_end_0 = const()[name = string("op_26677_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26677_end_mask_0 = const()[name = string("op_26677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26677_cast_fp16 = slice_by_index(begin = var_26677_begin_0, end = var_26677_end_0, end_mask = var_26677_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26677_cast_fp16")]; + tensor var_26684_begin_0 = const()[name = string("op_26684_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26684_end_0 = const()[name = string("op_26684_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26684_end_mask_0 = const()[name = string("op_26684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26684_cast_fp16 = slice_by_index(begin = var_26684_begin_0, end = var_26684_end_0, end_mask = var_26684_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26684_cast_fp16")]; + tensor var_26691_begin_0 = const()[name = string("op_26691_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26691_end_0 = const()[name = string("op_26691_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26691_end_mask_0 = const()[name = string("op_26691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26691_cast_fp16 = slice_by_index(begin = var_26691_begin_0, end = var_26691_end_0, end_mask = var_26691_end_mask_0, x = var_26273_cast_fp16)[name = string("op_26691_cast_fp16")]; + tensor var_26698_begin_0 = const()[name = string("op_26698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26698_end_0 = const()[name = string("op_26698_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26698_end_mask_0 = const()[name = string("op_26698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26698_cast_fp16 = slice_by_index(begin = var_26698_begin_0, end = var_26698_end_0, end_mask = var_26698_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26698_cast_fp16")]; + tensor var_26705_begin_0 = const()[name = string("op_26705_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26705_end_0 = const()[name = string("op_26705_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26705_end_mask_0 = const()[name = string("op_26705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26705_cast_fp16 = slice_by_index(begin = var_26705_begin_0, end = var_26705_end_0, end_mask = var_26705_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26705_cast_fp16")]; + tensor var_26712_begin_0 = const()[name = string("op_26712_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26712_end_0 = const()[name = string("op_26712_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26712_end_mask_0 = const()[name = string("op_26712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26712_cast_fp16 = slice_by_index(begin = var_26712_begin_0, end = var_26712_end_0, end_mask = var_26712_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26712_cast_fp16")]; + tensor var_26719_begin_0 = const()[name = string("op_26719_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26719_end_0 = const()[name = string("op_26719_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26719_end_mask_0 = const()[name = string("op_26719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26719_cast_fp16 = slice_by_index(begin = var_26719_begin_0, end = var_26719_end_0, end_mask = var_26719_end_mask_0, x = var_26277_cast_fp16)[name = string("op_26719_cast_fp16")]; + tensor var_26726_begin_0 = const()[name = string("op_26726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26726_end_0 = const()[name = string("op_26726_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26726_end_mask_0 = const()[name = string("op_26726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26726_cast_fp16 = slice_by_index(begin = var_26726_begin_0, end = var_26726_end_0, end_mask = var_26726_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26726_cast_fp16")]; + tensor var_26733_begin_0 = const()[name = string("op_26733_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26733_end_0 = const()[name = string("op_26733_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26733_end_mask_0 = const()[name = string("op_26733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26733_cast_fp16 = slice_by_index(begin = var_26733_begin_0, end = var_26733_end_0, end_mask = var_26733_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26733_cast_fp16")]; + tensor var_26740_begin_0 = const()[name = string("op_26740_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26740_end_0 = const()[name = string("op_26740_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26740_end_mask_0 = const()[name = string("op_26740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26740_cast_fp16 = slice_by_index(begin = var_26740_begin_0, end = var_26740_end_0, end_mask = var_26740_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26740_cast_fp16")]; + tensor var_26747_begin_0 = const()[name = string("op_26747_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26747_end_0 = const()[name = string("op_26747_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26747_end_mask_0 = const()[name = string("op_26747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26747_cast_fp16 = slice_by_index(begin = var_26747_begin_0, end = var_26747_end_0, end_mask = var_26747_end_mask_0, x = var_26281_cast_fp16)[name = string("op_26747_cast_fp16")]; + tensor var_26754_begin_0 = const()[name = string("op_26754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26754_end_0 = const()[name = string("op_26754_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26754_end_mask_0 = const()[name = string("op_26754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26754_cast_fp16 = slice_by_index(begin = var_26754_begin_0, end = var_26754_end_0, end_mask = var_26754_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26754_cast_fp16")]; + tensor var_26761_begin_0 = const()[name = string("op_26761_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26761_end_0 = const()[name = string("op_26761_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26761_end_mask_0 = const()[name = string("op_26761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26761_cast_fp16 = slice_by_index(begin = var_26761_begin_0, end = var_26761_end_0, end_mask = var_26761_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26761_cast_fp16")]; + tensor var_26768_begin_0 = const()[name = string("op_26768_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26768_end_0 = const()[name = string("op_26768_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26768_end_mask_0 = const()[name = string("op_26768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26768_cast_fp16 = slice_by_index(begin = var_26768_begin_0, end = var_26768_end_0, end_mask = var_26768_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26768_cast_fp16")]; + tensor var_26775_begin_0 = const()[name = string("op_26775_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26775_end_0 = const()[name = string("op_26775_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26775_end_mask_0 = const()[name = string("op_26775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26775_cast_fp16 = slice_by_index(begin = var_26775_begin_0, end = var_26775_end_0, end_mask = var_26775_end_mask_0, x = var_26285_cast_fp16)[name = string("op_26775_cast_fp16")]; + tensor var_26782_begin_0 = const()[name = string("op_26782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26782_end_0 = const()[name = string("op_26782_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26782_end_mask_0 = const()[name = string("op_26782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26782_cast_fp16 = slice_by_index(begin = var_26782_begin_0, end = var_26782_end_0, end_mask = var_26782_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26782_cast_fp16")]; + tensor var_26789_begin_0 = const()[name = string("op_26789_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26789_end_0 = const()[name = string("op_26789_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26789_end_mask_0 = const()[name = string("op_26789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26789_cast_fp16 = slice_by_index(begin = var_26789_begin_0, end = var_26789_end_0, end_mask = var_26789_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26789_cast_fp16")]; + tensor var_26796_begin_0 = const()[name = string("op_26796_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26796_end_0 = const()[name = string("op_26796_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26796_end_mask_0 = const()[name = string("op_26796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26796_cast_fp16 = slice_by_index(begin = var_26796_begin_0, end = var_26796_end_0, end_mask = var_26796_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26796_cast_fp16")]; + tensor var_26803_begin_0 = const()[name = string("op_26803_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26803_end_0 = const()[name = string("op_26803_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26803_end_mask_0 = const()[name = string("op_26803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26803_cast_fp16 = slice_by_index(begin = var_26803_begin_0, end = var_26803_end_0, end_mask = var_26803_end_mask_0, x = var_26289_cast_fp16)[name = string("op_26803_cast_fp16")]; + tensor var_26810_begin_0 = const()[name = string("op_26810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26810_end_0 = const()[name = string("op_26810_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26810_end_mask_0 = const()[name = string("op_26810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26810_cast_fp16 = slice_by_index(begin = var_26810_begin_0, end = var_26810_end_0, end_mask = var_26810_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26810_cast_fp16")]; + tensor var_26817_begin_0 = const()[name = string("op_26817_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26817_end_0 = const()[name = string("op_26817_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26817_end_mask_0 = const()[name = string("op_26817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26817_cast_fp16 = slice_by_index(begin = var_26817_begin_0, end = var_26817_end_0, end_mask = var_26817_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26817_cast_fp16")]; + tensor var_26824_begin_0 = const()[name = string("op_26824_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26824_end_0 = const()[name = string("op_26824_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26824_end_mask_0 = const()[name = string("op_26824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26824_cast_fp16 = slice_by_index(begin = var_26824_begin_0, end = var_26824_end_0, end_mask = var_26824_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26824_cast_fp16")]; + tensor var_26831_begin_0 = const()[name = string("op_26831_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26831_end_0 = const()[name = string("op_26831_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26831_end_mask_0 = const()[name = string("op_26831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26831_cast_fp16 = slice_by_index(begin = var_26831_begin_0, end = var_26831_end_0, end_mask = var_26831_end_mask_0, x = var_26293_cast_fp16)[name = string("op_26831_cast_fp16")]; + tensor var_26838_begin_0 = const()[name = string("op_26838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26838_end_0 = const()[name = string("op_26838_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26838_end_mask_0 = const()[name = string("op_26838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26838_cast_fp16 = slice_by_index(begin = var_26838_begin_0, end = var_26838_end_0, end_mask = var_26838_end_mask_0, x = var_26297_cast_fp16)[name = string("op_26838_cast_fp16")]; + tensor var_26845_begin_0 = const()[name = string("op_26845_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26845_end_0 = const()[name = string("op_26845_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26845_end_mask_0 = const()[name = string("op_26845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26845_cast_fp16 = slice_by_index(begin = var_26845_begin_0, end = var_26845_end_0, end_mask = var_26845_end_mask_0, x = var_26297_cast_fp16)[name = string("op_26845_cast_fp16")]; + tensor var_26852_begin_0 = const()[name = string("op_26852_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26852_end_0 = const()[name = string("op_26852_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26852_end_mask_0 = const()[name = string("op_26852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26852_cast_fp16 = slice_by_index(begin = var_26852_begin_0, end = var_26852_end_0, end_mask = var_26852_end_mask_0, x = var_26297_cast_fp16)[name = string("op_26852_cast_fp16")]; + tensor var_26859_begin_0 = const()[name = string("op_26859_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26859_end_0 = const()[name = string("op_26859_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26859_end_mask_0 = const()[name = string("op_26859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26859_cast_fp16 = slice_by_index(begin = var_26859_begin_0, end = var_26859_end_0, end_mask = var_26859_end_mask_0, x = var_26297_cast_fp16)[name = string("op_26859_cast_fp16")]; + tensor k_35_perm_0 = const()[name = string("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_26864_begin_0 = const()[name = string("op_26864_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26864_end_0 = const()[name = string("op_26864_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_26864_end_mask_0 = const()[name = string("op_26864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_35_cast_fp16 = transpose(perm = k_35_perm_0, x = key_35_cast_fp16)[name = string("transpose_14")]; + tensor var_26864_cast_fp16 = slice_by_index(begin = var_26864_begin_0, end = var_26864_end_0, end_mask = var_26864_end_mask_0, x = k_35_cast_fp16)[name = string("op_26864_cast_fp16")]; + tensor var_26868_begin_0 = const()[name = string("op_26868_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_26868_end_0 = const()[name = string("op_26868_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_26868_end_mask_0 = const()[name = string("op_26868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26868_cast_fp16 = slice_by_index(begin = var_26868_begin_0, end = var_26868_end_0, end_mask = var_26868_end_mask_0, x = k_35_cast_fp16)[name = string("op_26868_cast_fp16")]; + tensor var_26872_begin_0 = const()[name = string("op_26872_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_26872_end_0 = const()[name = string("op_26872_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_26872_end_mask_0 = const()[name = string("op_26872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26872_cast_fp16 = slice_by_index(begin = var_26872_begin_0, end = var_26872_end_0, end_mask = var_26872_end_mask_0, x = k_35_cast_fp16)[name = string("op_26872_cast_fp16")]; + tensor var_26876_begin_0 = const()[name = string("op_26876_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_26876_end_0 = const()[name = string("op_26876_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_26876_end_mask_0 = const()[name = string("op_26876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26876_cast_fp16 = slice_by_index(begin = var_26876_begin_0, end = var_26876_end_0, end_mask = var_26876_end_mask_0, x = k_35_cast_fp16)[name = string("op_26876_cast_fp16")]; + tensor var_26880_begin_0 = const()[name = string("op_26880_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_26880_end_0 = const()[name = string("op_26880_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_26880_end_mask_0 = const()[name = string("op_26880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26880_cast_fp16 = slice_by_index(begin = var_26880_begin_0, end = var_26880_end_0, end_mask = var_26880_end_mask_0, x = k_35_cast_fp16)[name = string("op_26880_cast_fp16")]; + tensor var_26884_begin_0 = const()[name = string("op_26884_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_26884_end_0 = const()[name = string("op_26884_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_26884_end_mask_0 = const()[name = string("op_26884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26884_cast_fp16 = slice_by_index(begin = var_26884_begin_0, end = var_26884_end_0, end_mask = var_26884_end_mask_0, x = k_35_cast_fp16)[name = string("op_26884_cast_fp16")]; + tensor var_26888_begin_0 = const()[name = string("op_26888_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_26888_end_0 = const()[name = string("op_26888_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_26888_end_mask_0 = const()[name = string("op_26888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26888_cast_fp16 = slice_by_index(begin = var_26888_begin_0, end = var_26888_end_0, end_mask = var_26888_end_mask_0, x = k_35_cast_fp16)[name = string("op_26888_cast_fp16")]; + tensor var_26892_begin_0 = const()[name = string("op_26892_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_26892_end_0 = const()[name = string("op_26892_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_26892_end_mask_0 = const()[name = string("op_26892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26892_cast_fp16 = slice_by_index(begin = var_26892_begin_0, end = var_26892_end_0, end_mask = var_26892_end_mask_0, x = k_35_cast_fp16)[name = string("op_26892_cast_fp16")]; + tensor var_26896_begin_0 = const()[name = string("op_26896_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_26896_end_0 = const()[name = string("op_26896_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_26896_end_mask_0 = const()[name = string("op_26896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26896_cast_fp16 = slice_by_index(begin = var_26896_begin_0, end = var_26896_end_0, end_mask = var_26896_end_mask_0, x = k_35_cast_fp16)[name = string("op_26896_cast_fp16")]; + tensor var_26900_begin_0 = const()[name = string("op_26900_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_26900_end_0 = const()[name = string("op_26900_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_26900_end_mask_0 = const()[name = string("op_26900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26900_cast_fp16 = slice_by_index(begin = var_26900_begin_0, end = var_26900_end_0, end_mask = var_26900_end_mask_0, x = k_35_cast_fp16)[name = string("op_26900_cast_fp16")]; + tensor var_26904_begin_0 = const()[name = string("op_26904_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_26904_end_0 = const()[name = string("op_26904_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_26904_end_mask_0 = const()[name = string("op_26904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26904_cast_fp16 = slice_by_index(begin = var_26904_begin_0, end = var_26904_end_0, end_mask = var_26904_end_mask_0, x = k_35_cast_fp16)[name = string("op_26904_cast_fp16")]; + tensor var_26908_begin_0 = const()[name = string("op_26908_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_26908_end_0 = const()[name = string("op_26908_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_26908_end_mask_0 = const()[name = string("op_26908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26908_cast_fp16 = slice_by_index(begin = var_26908_begin_0, end = var_26908_end_0, end_mask = var_26908_end_mask_0, x = k_35_cast_fp16)[name = string("op_26908_cast_fp16")]; + tensor var_26912_begin_0 = const()[name = string("op_26912_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_26912_end_0 = const()[name = string("op_26912_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_26912_end_mask_0 = const()[name = string("op_26912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26912_cast_fp16 = slice_by_index(begin = var_26912_begin_0, end = var_26912_end_0, end_mask = var_26912_end_mask_0, x = k_35_cast_fp16)[name = string("op_26912_cast_fp16")]; + tensor var_26916_begin_0 = const()[name = string("op_26916_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_26916_end_0 = const()[name = string("op_26916_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_26916_end_mask_0 = const()[name = string("op_26916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26916_cast_fp16 = slice_by_index(begin = var_26916_begin_0, end = var_26916_end_0, end_mask = var_26916_end_mask_0, x = k_35_cast_fp16)[name = string("op_26916_cast_fp16")]; + tensor var_26920_begin_0 = const()[name = string("op_26920_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_26920_end_0 = const()[name = string("op_26920_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_26920_end_mask_0 = const()[name = string("op_26920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26920_cast_fp16 = slice_by_index(begin = var_26920_begin_0, end = var_26920_end_0, end_mask = var_26920_end_mask_0, x = k_35_cast_fp16)[name = string("op_26920_cast_fp16")]; + tensor var_26924_begin_0 = const()[name = string("op_26924_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_26924_end_0 = const()[name = string("op_26924_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_26924_end_mask_0 = const()[name = string("op_26924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26924_cast_fp16 = slice_by_index(begin = var_26924_begin_0, end = var_26924_end_0, end_mask = var_26924_end_mask_0, x = k_35_cast_fp16)[name = string("op_26924_cast_fp16")]; + tensor var_26928_begin_0 = const()[name = string("op_26928_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_26928_end_0 = const()[name = string("op_26928_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_26928_end_mask_0 = const()[name = string("op_26928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26928_cast_fp16 = slice_by_index(begin = var_26928_begin_0, end = var_26928_end_0, end_mask = var_26928_end_mask_0, x = k_35_cast_fp16)[name = string("op_26928_cast_fp16")]; + tensor var_26932_begin_0 = const()[name = string("op_26932_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_26932_end_0 = const()[name = string("op_26932_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_26932_end_mask_0 = const()[name = string("op_26932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26932_cast_fp16 = slice_by_index(begin = var_26932_begin_0, end = var_26932_end_0, end_mask = var_26932_end_mask_0, x = k_35_cast_fp16)[name = string("op_26932_cast_fp16")]; + tensor var_26936_begin_0 = const()[name = string("op_26936_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_26936_end_0 = const()[name = string("op_26936_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_26936_end_mask_0 = const()[name = string("op_26936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26936_cast_fp16 = slice_by_index(begin = var_26936_begin_0, end = var_26936_end_0, end_mask = var_26936_end_mask_0, x = k_35_cast_fp16)[name = string("op_26936_cast_fp16")]; + tensor var_26940_begin_0 = const()[name = string("op_26940_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_26940_end_0 = const()[name = string("op_26940_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_26940_end_mask_0 = const()[name = string("op_26940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26940_cast_fp16 = slice_by_index(begin = var_26940_begin_0, end = var_26940_end_0, end_mask = var_26940_end_mask_0, x = k_35_cast_fp16)[name = string("op_26940_cast_fp16")]; + tensor var_26942_begin_0 = const()[name = string("op_26942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26942_end_0 = const()[name = string("op_26942_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26942_end_mask_0 = const()[name = string("op_26942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26942_cast_fp16 = slice_by_index(begin = var_26942_begin_0, end = var_26942_end_0, end_mask = var_26942_end_mask_0, x = value_35_cast_fp16)[name = string("op_26942_cast_fp16")]; + tensor var_26946_begin_0 = const()[name = string("op_26946_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26946_end_0 = const()[name = string("op_26946_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_26946_end_mask_0 = const()[name = string("op_26946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26946_cast_fp16 = slice_by_index(begin = var_26946_begin_0, end = var_26946_end_0, end_mask = var_26946_end_mask_0, x = value_35_cast_fp16)[name = string("op_26946_cast_fp16")]; + tensor var_26950_begin_0 = const()[name = string("op_26950_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26950_end_0 = const()[name = string("op_26950_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_26950_end_mask_0 = const()[name = string("op_26950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26950_cast_fp16 = slice_by_index(begin = var_26950_begin_0, end = var_26950_end_0, end_mask = var_26950_end_mask_0, x = value_35_cast_fp16)[name = string("op_26950_cast_fp16")]; + tensor var_26954_begin_0 = const()[name = string("op_26954_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26954_end_0 = const()[name = string("op_26954_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_26954_end_mask_0 = const()[name = string("op_26954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26954_cast_fp16 = slice_by_index(begin = var_26954_begin_0, end = var_26954_end_0, end_mask = var_26954_end_mask_0, x = value_35_cast_fp16)[name = string("op_26954_cast_fp16")]; + tensor var_26958_begin_0 = const()[name = string("op_26958_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26958_end_0 = const()[name = string("op_26958_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_26958_end_mask_0 = const()[name = string("op_26958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26958_cast_fp16 = slice_by_index(begin = var_26958_begin_0, end = var_26958_end_0, end_mask = var_26958_end_mask_0, x = value_35_cast_fp16)[name = string("op_26958_cast_fp16")]; + tensor var_26962_begin_0 = const()[name = string("op_26962_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26962_end_0 = const()[name = string("op_26962_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_26962_end_mask_0 = const()[name = string("op_26962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26962_cast_fp16 = slice_by_index(begin = var_26962_begin_0, end = var_26962_end_0, end_mask = var_26962_end_mask_0, x = value_35_cast_fp16)[name = string("op_26962_cast_fp16")]; + tensor var_26966_begin_0 = const()[name = string("op_26966_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26966_end_0 = const()[name = string("op_26966_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_26966_end_mask_0 = const()[name = string("op_26966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26966_cast_fp16 = slice_by_index(begin = var_26966_begin_0, end = var_26966_end_0, end_mask = var_26966_end_mask_0, x = value_35_cast_fp16)[name = string("op_26966_cast_fp16")]; + tensor var_26970_begin_0 = const()[name = string("op_26970_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26970_end_0 = const()[name = string("op_26970_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_26970_end_mask_0 = const()[name = string("op_26970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26970_cast_fp16 = slice_by_index(begin = var_26970_begin_0, end = var_26970_end_0, end_mask = var_26970_end_mask_0, x = value_35_cast_fp16)[name = string("op_26970_cast_fp16")]; + tensor var_26974_begin_0 = const()[name = string("op_26974_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26974_end_0 = const()[name = string("op_26974_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_26974_end_mask_0 = const()[name = string("op_26974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26974_cast_fp16 = slice_by_index(begin = var_26974_begin_0, end = var_26974_end_0, end_mask = var_26974_end_mask_0, x = value_35_cast_fp16)[name = string("op_26974_cast_fp16")]; + tensor var_26978_begin_0 = const()[name = string("op_26978_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26978_end_0 = const()[name = string("op_26978_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_26978_end_mask_0 = const()[name = string("op_26978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26978_cast_fp16 = slice_by_index(begin = var_26978_begin_0, end = var_26978_end_0, end_mask = var_26978_end_mask_0, x = value_35_cast_fp16)[name = string("op_26978_cast_fp16")]; + tensor var_26982_begin_0 = const()[name = string("op_26982_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26982_end_0 = const()[name = string("op_26982_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_26982_end_mask_0 = const()[name = string("op_26982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26982_cast_fp16 = slice_by_index(begin = var_26982_begin_0, end = var_26982_end_0, end_mask = var_26982_end_mask_0, x = value_35_cast_fp16)[name = string("op_26982_cast_fp16")]; + tensor var_26986_begin_0 = const()[name = string("op_26986_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26986_end_0 = const()[name = string("op_26986_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_26986_end_mask_0 = const()[name = string("op_26986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26986_cast_fp16 = slice_by_index(begin = var_26986_begin_0, end = var_26986_end_0, end_mask = var_26986_end_mask_0, x = value_35_cast_fp16)[name = string("op_26986_cast_fp16")]; + tensor var_26990_begin_0 = const()[name = string("op_26990_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26990_end_0 = const()[name = string("op_26990_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_26990_end_mask_0 = const()[name = string("op_26990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26990_cast_fp16 = slice_by_index(begin = var_26990_begin_0, end = var_26990_end_0, end_mask = var_26990_end_mask_0, x = value_35_cast_fp16)[name = string("op_26990_cast_fp16")]; + tensor var_26994_begin_0 = const()[name = string("op_26994_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26994_end_0 = const()[name = string("op_26994_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_26994_end_mask_0 = const()[name = string("op_26994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26994_cast_fp16 = slice_by_index(begin = var_26994_begin_0, end = var_26994_end_0, end_mask = var_26994_end_mask_0, x = value_35_cast_fp16)[name = string("op_26994_cast_fp16")]; + tensor var_26998_begin_0 = const()[name = string("op_26998_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26998_end_0 = const()[name = string("op_26998_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_26998_end_mask_0 = const()[name = string("op_26998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26998_cast_fp16 = slice_by_index(begin = var_26998_begin_0, end = var_26998_end_0, end_mask = var_26998_end_mask_0, x = value_35_cast_fp16)[name = string("op_26998_cast_fp16")]; + tensor var_27002_begin_0 = const()[name = string("op_27002_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27002_end_0 = const()[name = string("op_27002_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_27002_end_mask_0 = const()[name = string("op_27002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27002_cast_fp16 = slice_by_index(begin = var_27002_begin_0, end = var_27002_end_0, end_mask = var_27002_end_mask_0, x = value_35_cast_fp16)[name = string("op_27002_cast_fp16")]; + tensor var_27006_begin_0 = const()[name = string("op_27006_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27006_end_0 = const()[name = string("op_27006_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_27006_end_mask_0 = const()[name = string("op_27006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27006_cast_fp16 = slice_by_index(begin = var_27006_begin_0, end = var_27006_end_0, end_mask = var_27006_end_mask_0, x = value_35_cast_fp16)[name = string("op_27006_cast_fp16")]; + tensor var_27010_begin_0 = const()[name = string("op_27010_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27010_end_0 = const()[name = string("op_27010_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_27010_end_mask_0 = const()[name = string("op_27010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27010_cast_fp16 = slice_by_index(begin = var_27010_begin_0, end = var_27010_end_0, end_mask = var_27010_end_mask_0, x = value_35_cast_fp16)[name = string("op_27010_cast_fp16")]; + tensor var_27014_begin_0 = const()[name = string("op_27014_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27014_end_0 = const()[name = string("op_27014_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_27014_end_mask_0 = const()[name = string("op_27014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27014_cast_fp16 = slice_by_index(begin = var_27014_begin_0, end = var_27014_end_0, end_mask = var_27014_end_mask_0, x = value_35_cast_fp16)[name = string("op_27014_cast_fp16")]; + tensor var_27018_begin_0 = const()[name = string("op_27018_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27018_end_0 = const()[name = string("op_27018_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_27018_end_mask_0 = const()[name = string("op_27018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27018_cast_fp16 = slice_by_index(begin = var_27018_begin_0, end = var_27018_end_0, end_mask = var_27018_end_mask_0, x = value_35_cast_fp16)[name = string("op_27018_cast_fp16")]; + string _SplitHeadsQ__mh_w_2721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2721_equation_0, values = (var_26864_cast_fp16, var_26306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2721_cast_fp16")]; + string _SplitHeadsQ__mh_w_2723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2723_equation_0, values = (var_26864_cast_fp16, var_26313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2723_cast_fp16")]; + string _SplitHeadsQ__mh_w_2725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2725_equation_0, values = (var_26864_cast_fp16, var_26320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2725_cast_fp16")]; + string _SplitHeadsQ__mh_w_2727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2727_equation_0, values = (var_26864_cast_fp16, var_26327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2727_cast_fp16")]; + string _SplitHeadsQ__mh_w_2729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2729_equation_0, values = (var_26868_cast_fp16, var_26334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2729_cast_fp16")]; + string _SplitHeadsQ__mh_w_2731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2731_equation_0, values = (var_26868_cast_fp16, var_26341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2731_cast_fp16")]; + string _SplitHeadsQ__mh_w_2733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2733_equation_0, values = (var_26868_cast_fp16, var_26348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2733_cast_fp16")]; + string _SplitHeadsQ__mh_w_2735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2735_equation_0, values = (var_26868_cast_fp16, var_26355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2735_cast_fp16")]; + string _SplitHeadsQ__mh_w_2737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2737_equation_0, values = (var_26872_cast_fp16, var_26362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2737_cast_fp16")]; + string _SplitHeadsQ__mh_w_2739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2739_equation_0, values = (var_26872_cast_fp16, var_26369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2739_cast_fp16")]; + string _SplitHeadsQ__mh_w_2741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2741_equation_0, values = (var_26872_cast_fp16, var_26376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2741_cast_fp16")]; + string _SplitHeadsQ__mh_w_2743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2743_equation_0, values = (var_26872_cast_fp16, var_26383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2743_cast_fp16")]; + string _SplitHeadsQ__mh_w_2745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2745_equation_0, values = (var_26876_cast_fp16, var_26390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2745_cast_fp16")]; + string _SplitHeadsQ__mh_w_2747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2747_equation_0, values = (var_26876_cast_fp16, var_26397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2747_cast_fp16")]; + string _SplitHeadsQ__mh_w_2749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2749_equation_0, values = (var_26876_cast_fp16, var_26404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2749_cast_fp16")]; + string _SplitHeadsQ__mh_w_2751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2751_equation_0, values = (var_26876_cast_fp16, var_26411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2751_cast_fp16")]; + string _SplitHeadsQ__mh_w_2753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2753_equation_0, values = (var_26880_cast_fp16, var_26418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2753_cast_fp16")]; + string _SplitHeadsQ__mh_w_2755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2755_equation_0, values = (var_26880_cast_fp16, var_26425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2755_cast_fp16")]; + string _SplitHeadsQ__mh_w_2757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2757_equation_0, values = (var_26880_cast_fp16, var_26432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2757_cast_fp16")]; + string _SplitHeadsQ__mh_w_2759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2759_equation_0, values = (var_26880_cast_fp16, var_26439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2759_cast_fp16")]; + string _SplitHeadsQ__mh_w_2761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2761_equation_0, values = (var_26884_cast_fp16, var_26446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2761_cast_fp16")]; + string _SplitHeadsQ__mh_w_2763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2763_equation_0, values = (var_26884_cast_fp16, var_26453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2763_cast_fp16")]; + string _SplitHeadsQ__mh_w_2765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2765_equation_0, values = (var_26884_cast_fp16, var_26460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2765_cast_fp16")]; + string _SplitHeadsQ__mh_w_2767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2767_equation_0, values = (var_26884_cast_fp16, var_26467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2767_cast_fp16")]; + string _SplitHeadsQ__mh_w_2769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2769_equation_0, values = (var_26888_cast_fp16, var_26474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2769_cast_fp16")]; + string _SplitHeadsQ__mh_w_2771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2771_equation_0, values = (var_26888_cast_fp16, var_26481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2771_cast_fp16")]; + string _SplitHeadsQ__mh_w_2773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2773_equation_0, values = (var_26888_cast_fp16, var_26488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2773_cast_fp16")]; + string _SplitHeadsQ__mh_w_2775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2775_equation_0, values = (var_26888_cast_fp16, var_26495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2775_cast_fp16")]; + string _SplitHeadsQ__mh_w_2777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2777_equation_0, values = (var_26892_cast_fp16, var_26502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2777_cast_fp16")]; + string _SplitHeadsQ__mh_w_2779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2779_equation_0, values = (var_26892_cast_fp16, var_26509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2779_cast_fp16")]; + string _SplitHeadsQ__mh_w_2781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2781_equation_0, values = (var_26892_cast_fp16, var_26516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2781_cast_fp16")]; + string _SplitHeadsQ__mh_w_2783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2783_equation_0, values = (var_26892_cast_fp16, var_26523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2783_cast_fp16")]; + string _SplitHeadsQ__mh_w_2785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2785_equation_0, values = (var_26896_cast_fp16, var_26530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2785_cast_fp16")]; + string _SplitHeadsQ__mh_w_2787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2787_equation_0, values = (var_26896_cast_fp16, var_26537_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2787_cast_fp16")]; + string _SplitHeadsQ__mh_w_2789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2789_equation_0, values = (var_26896_cast_fp16, var_26544_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2789_cast_fp16")]; + string _SplitHeadsQ__mh_w_2791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2791_equation_0, values = (var_26896_cast_fp16, var_26551_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2791_cast_fp16")]; + string _SplitHeadsQ__mh_w_2793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2793_equation_0, values = (var_26900_cast_fp16, var_26558_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2793_cast_fp16")]; + string _SplitHeadsQ__mh_w_2795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2795_equation_0, values = (var_26900_cast_fp16, var_26565_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2795_cast_fp16")]; + string _SplitHeadsQ__mh_w_2797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2797_equation_0, values = (var_26900_cast_fp16, var_26572_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2797_cast_fp16")]; + string _SplitHeadsQ__mh_w_2799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2799_equation_0, values = (var_26900_cast_fp16, var_26579_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2799_cast_fp16")]; + string _SplitHeadsQ__mh_w_2801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2801_equation_0, values = (var_26904_cast_fp16, var_26586_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2801_cast_fp16")]; + string _SplitHeadsQ__mh_w_2803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2803_equation_0, values = (var_26904_cast_fp16, var_26593_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2803_cast_fp16")]; + string _SplitHeadsQ__mh_w_2805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2805_equation_0, values = (var_26904_cast_fp16, var_26600_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2805_cast_fp16")]; + string _SplitHeadsQ__mh_w_2807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2807_equation_0, values = (var_26904_cast_fp16, var_26607_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2807_cast_fp16")]; + string _SplitHeadsQ__mh_w_2809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2809_equation_0, values = (var_26908_cast_fp16, var_26614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2809_cast_fp16")]; + string _SplitHeadsQ__mh_w_2811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2811_equation_0, values = (var_26908_cast_fp16, var_26621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2811_cast_fp16")]; + string _SplitHeadsQ__mh_w_2813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2813_equation_0, values = (var_26908_cast_fp16, var_26628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2813_cast_fp16")]; + string _SplitHeadsQ__mh_w_2815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2815_equation_0, values = (var_26908_cast_fp16, var_26635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2815_cast_fp16")]; + string _SplitHeadsQ__mh_w_2817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2817_equation_0, values = (var_26912_cast_fp16, var_26642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2817_cast_fp16")]; + string _SplitHeadsQ__mh_w_2819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2819_equation_0, values = (var_26912_cast_fp16, var_26649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2819_cast_fp16")]; + string _SplitHeadsQ__mh_w_2821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2821_equation_0, values = (var_26912_cast_fp16, var_26656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2821_cast_fp16")]; + string _SplitHeadsQ__mh_w_2823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2823_equation_0, values = (var_26912_cast_fp16, var_26663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2823_cast_fp16")]; + string _SplitHeadsQ__mh_w_2825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2825_equation_0, values = (var_26916_cast_fp16, var_26670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2825_cast_fp16")]; + string _SplitHeadsQ__mh_w_2827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2827_equation_0, values = (var_26916_cast_fp16, var_26677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2827_cast_fp16")]; + string _SplitHeadsQ__mh_w_2829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2829_equation_0, values = (var_26916_cast_fp16, var_26684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2829_cast_fp16")]; + string _SplitHeadsQ__mh_w_2831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2831_equation_0, values = (var_26916_cast_fp16, var_26691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2831_cast_fp16")]; + string _SplitHeadsQ__mh_w_2833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2833_equation_0, values = (var_26920_cast_fp16, var_26698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2833_cast_fp16")]; + string _SplitHeadsQ__mh_w_2835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2835_equation_0, values = (var_26920_cast_fp16, var_26705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2835_cast_fp16")]; + string _SplitHeadsQ__mh_w_2837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2837_equation_0, values = (var_26920_cast_fp16, var_26712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2837_cast_fp16")]; + string _SplitHeadsQ__mh_w_2839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2839_equation_0, values = (var_26920_cast_fp16, var_26719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2839_cast_fp16")]; + string _SplitHeadsQ__mh_w_2841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2841_equation_0, values = (var_26924_cast_fp16, var_26726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2841_cast_fp16")]; + string _SplitHeadsQ__mh_w_2843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2843_equation_0, values = (var_26924_cast_fp16, var_26733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2843_cast_fp16")]; + string _SplitHeadsQ__mh_w_2845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2845_equation_0, values = (var_26924_cast_fp16, var_26740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2845_cast_fp16")]; + string _SplitHeadsQ__mh_w_2847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2847_equation_0, values = (var_26924_cast_fp16, var_26747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2847_cast_fp16")]; + string _SplitHeadsQ__mh_w_2849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2849_equation_0, values = (var_26928_cast_fp16, var_26754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2849_cast_fp16")]; + string _SplitHeadsQ__mh_w_2851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2851_equation_0, values = (var_26928_cast_fp16, var_26761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2851_cast_fp16")]; + string _SplitHeadsQ__mh_w_2853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2853_equation_0, values = (var_26928_cast_fp16, var_26768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2853_cast_fp16")]; + string _SplitHeadsQ__mh_w_2855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2855_equation_0, values = (var_26928_cast_fp16, var_26775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2855_cast_fp16")]; + string _SplitHeadsQ__mh_w_2857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2857_equation_0, values = (var_26932_cast_fp16, var_26782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2857_cast_fp16")]; + string _SplitHeadsQ__mh_w_2859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2859_equation_0, values = (var_26932_cast_fp16, var_26789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2859_cast_fp16")]; + string _SplitHeadsQ__mh_w_2861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2861_equation_0, values = (var_26932_cast_fp16, var_26796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2861_cast_fp16")]; + string _SplitHeadsQ__mh_w_2863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2863_equation_0, values = (var_26932_cast_fp16, var_26803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2863_cast_fp16")]; + string _SplitHeadsQ__mh_w_2865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2865_equation_0, values = (var_26936_cast_fp16, var_26810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2865_cast_fp16")]; + string _SplitHeadsQ__mh_w_2867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2867_equation_0, values = (var_26936_cast_fp16, var_26817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2867_cast_fp16")]; + string _SplitHeadsQ__mh_w_2869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2869_equation_0, values = (var_26936_cast_fp16, var_26824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2869_cast_fp16")]; + string _SplitHeadsQ__mh_w_2871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2871_equation_0, values = (var_26936_cast_fp16, var_26831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2871_cast_fp16")]; + string _SplitHeadsQ__mh_w_2873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2873_equation_0, values = (var_26940_cast_fp16, var_26838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2873_cast_fp16")]; + string _SplitHeadsQ__mh_w_2875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2875_equation_0, values = (var_26940_cast_fp16, var_26845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2875_cast_fp16")]; + string _SplitHeadsQ__mh_w_2877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2877_equation_0, values = (var_26940_cast_fp16, var_26852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2877_cast_fp16")]; + string _SplitHeadsQ__mh_w_2879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2879_equation_0, values = (var_26940_cast_fp16, var_26859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2879_cast_fp16")]; + fp16 var_27181_to_fp16 = const()[name = string("op_27181_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2721_cast_fp16, y = var_27181_to_fp16)[name = string("aw_chunk_2721_cast_fp16")]; + fp16 var_27183_to_fp16 = const()[name = string("op_27183_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2723_cast_fp16, y = var_27183_to_fp16)[name = string("aw_chunk_2723_cast_fp16")]; + fp16 var_27185_to_fp16 = const()[name = string("op_27185_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2725_cast_fp16, y = var_27185_to_fp16)[name = string("aw_chunk_2725_cast_fp16")]; + fp16 var_27187_to_fp16 = const()[name = string("op_27187_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2727_cast_fp16, y = var_27187_to_fp16)[name = string("aw_chunk_2727_cast_fp16")]; + fp16 var_27189_to_fp16 = const()[name = string("op_27189_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2729_cast_fp16, y = var_27189_to_fp16)[name = string("aw_chunk_2729_cast_fp16")]; + fp16 var_27191_to_fp16 = const()[name = string("op_27191_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2731_cast_fp16, y = var_27191_to_fp16)[name = string("aw_chunk_2731_cast_fp16")]; + fp16 var_27193_to_fp16 = const()[name = string("op_27193_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2733_cast_fp16, y = var_27193_to_fp16)[name = string("aw_chunk_2733_cast_fp16")]; + fp16 var_27195_to_fp16 = const()[name = string("op_27195_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2735_cast_fp16, y = var_27195_to_fp16)[name = string("aw_chunk_2735_cast_fp16")]; + fp16 var_27197_to_fp16 = const()[name = string("op_27197_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2737_cast_fp16, y = var_27197_to_fp16)[name = string("aw_chunk_2737_cast_fp16")]; + fp16 var_27199_to_fp16 = const()[name = string("op_27199_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2739_cast_fp16, y = var_27199_to_fp16)[name = string("aw_chunk_2739_cast_fp16")]; + fp16 var_27201_to_fp16 = const()[name = string("op_27201_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2741_cast_fp16, y = var_27201_to_fp16)[name = string("aw_chunk_2741_cast_fp16")]; + fp16 var_27203_to_fp16 = const()[name = string("op_27203_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2743_cast_fp16, y = var_27203_to_fp16)[name = string("aw_chunk_2743_cast_fp16")]; + fp16 var_27205_to_fp16 = const()[name = string("op_27205_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2745_cast_fp16, y = var_27205_to_fp16)[name = string("aw_chunk_2745_cast_fp16")]; + fp16 var_27207_to_fp16 = const()[name = string("op_27207_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2747_cast_fp16, y = var_27207_to_fp16)[name = string("aw_chunk_2747_cast_fp16")]; + fp16 var_27209_to_fp16 = const()[name = string("op_27209_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2749_cast_fp16, y = var_27209_to_fp16)[name = string("aw_chunk_2749_cast_fp16")]; + fp16 var_27211_to_fp16 = const()[name = string("op_27211_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2751_cast_fp16, y = var_27211_to_fp16)[name = string("aw_chunk_2751_cast_fp16")]; + fp16 var_27213_to_fp16 = const()[name = string("op_27213_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2753_cast_fp16, y = var_27213_to_fp16)[name = string("aw_chunk_2753_cast_fp16")]; + fp16 var_27215_to_fp16 = const()[name = string("op_27215_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2755_cast_fp16, y = var_27215_to_fp16)[name = string("aw_chunk_2755_cast_fp16")]; + fp16 var_27217_to_fp16 = const()[name = string("op_27217_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2757_cast_fp16, y = var_27217_to_fp16)[name = string("aw_chunk_2757_cast_fp16")]; + fp16 var_27219_to_fp16 = const()[name = string("op_27219_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2759_cast_fp16, y = var_27219_to_fp16)[name = string("aw_chunk_2759_cast_fp16")]; + fp16 var_27221_to_fp16 = const()[name = string("op_27221_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2761_cast_fp16, y = var_27221_to_fp16)[name = string("aw_chunk_2761_cast_fp16")]; + fp16 var_27223_to_fp16 = const()[name = string("op_27223_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2763_cast_fp16, y = var_27223_to_fp16)[name = string("aw_chunk_2763_cast_fp16")]; + fp16 var_27225_to_fp16 = const()[name = string("op_27225_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2765_cast_fp16, y = var_27225_to_fp16)[name = string("aw_chunk_2765_cast_fp16")]; + fp16 var_27227_to_fp16 = const()[name = string("op_27227_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2767_cast_fp16, y = var_27227_to_fp16)[name = string("aw_chunk_2767_cast_fp16")]; + fp16 var_27229_to_fp16 = const()[name = string("op_27229_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2769_cast_fp16, y = var_27229_to_fp16)[name = string("aw_chunk_2769_cast_fp16")]; + fp16 var_27231_to_fp16 = const()[name = string("op_27231_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2771_cast_fp16, y = var_27231_to_fp16)[name = string("aw_chunk_2771_cast_fp16")]; + fp16 var_27233_to_fp16 = const()[name = string("op_27233_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2773_cast_fp16, y = var_27233_to_fp16)[name = string("aw_chunk_2773_cast_fp16")]; + fp16 var_27235_to_fp16 = const()[name = string("op_27235_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2775_cast_fp16, y = var_27235_to_fp16)[name = string("aw_chunk_2775_cast_fp16")]; + fp16 var_27237_to_fp16 = const()[name = string("op_27237_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2777_cast_fp16, y = var_27237_to_fp16)[name = string("aw_chunk_2777_cast_fp16")]; + fp16 var_27239_to_fp16 = const()[name = string("op_27239_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2779_cast_fp16, y = var_27239_to_fp16)[name = string("aw_chunk_2779_cast_fp16")]; + fp16 var_27241_to_fp16 = const()[name = string("op_27241_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2781_cast_fp16, y = var_27241_to_fp16)[name = string("aw_chunk_2781_cast_fp16")]; + fp16 var_27243_to_fp16 = const()[name = string("op_27243_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2783_cast_fp16, y = var_27243_to_fp16)[name = string("aw_chunk_2783_cast_fp16")]; + fp16 var_27245_to_fp16 = const()[name = string("op_27245_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2785_cast_fp16, y = var_27245_to_fp16)[name = string("aw_chunk_2785_cast_fp16")]; + fp16 var_27247_to_fp16 = const()[name = string("op_27247_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2787_cast_fp16, y = var_27247_to_fp16)[name = string("aw_chunk_2787_cast_fp16")]; + fp16 var_27249_to_fp16 = const()[name = string("op_27249_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2789_cast_fp16, y = var_27249_to_fp16)[name = string("aw_chunk_2789_cast_fp16")]; + fp16 var_27251_to_fp16 = const()[name = string("op_27251_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2791_cast_fp16, y = var_27251_to_fp16)[name = string("aw_chunk_2791_cast_fp16")]; + fp16 var_27253_to_fp16 = const()[name = string("op_27253_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2793_cast_fp16, y = var_27253_to_fp16)[name = string("aw_chunk_2793_cast_fp16")]; + fp16 var_27255_to_fp16 = const()[name = string("op_27255_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2795_cast_fp16, y = var_27255_to_fp16)[name = string("aw_chunk_2795_cast_fp16")]; + fp16 var_27257_to_fp16 = const()[name = string("op_27257_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2797_cast_fp16, y = var_27257_to_fp16)[name = string("aw_chunk_2797_cast_fp16")]; + fp16 var_27259_to_fp16 = const()[name = string("op_27259_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2799_cast_fp16, y = var_27259_to_fp16)[name = string("aw_chunk_2799_cast_fp16")]; + fp16 var_27261_to_fp16 = const()[name = string("op_27261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2801_cast_fp16, y = var_27261_to_fp16)[name = string("aw_chunk_2801_cast_fp16")]; + fp16 var_27263_to_fp16 = const()[name = string("op_27263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2803_cast_fp16, y = var_27263_to_fp16)[name = string("aw_chunk_2803_cast_fp16")]; + fp16 var_27265_to_fp16 = const()[name = string("op_27265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2805_cast_fp16, y = var_27265_to_fp16)[name = string("aw_chunk_2805_cast_fp16")]; + fp16 var_27267_to_fp16 = const()[name = string("op_27267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2807_cast_fp16, y = var_27267_to_fp16)[name = string("aw_chunk_2807_cast_fp16")]; + fp16 var_27269_to_fp16 = const()[name = string("op_27269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2809_cast_fp16, y = var_27269_to_fp16)[name = string("aw_chunk_2809_cast_fp16")]; + fp16 var_27271_to_fp16 = const()[name = string("op_27271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2811_cast_fp16, y = var_27271_to_fp16)[name = string("aw_chunk_2811_cast_fp16")]; + fp16 var_27273_to_fp16 = const()[name = string("op_27273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2813_cast_fp16, y = var_27273_to_fp16)[name = string("aw_chunk_2813_cast_fp16")]; + fp16 var_27275_to_fp16 = const()[name = string("op_27275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2815_cast_fp16, y = var_27275_to_fp16)[name = string("aw_chunk_2815_cast_fp16")]; + fp16 var_27277_to_fp16 = const()[name = string("op_27277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2817_cast_fp16, y = var_27277_to_fp16)[name = string("aw_chunk_2817_cast_fp16")]; + fp16 var_27279_to_fp16 = const()[name = string("op_27279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2819_cast_fp16, y = var_27279_to_fp16)[name = string("aw_chunk_2819_cast_fp16")]; + fp16 var_27281_to_fp16 = const()[name = string("op_27281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2821_cast_fp16, y = var_27281_to_fp16)[name = string("aw_chunk_2821_cast_fp16")]; + fp16 var_27283_to_fp16 = const()[name = string("op_27283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2823_cast_fp16, y = var_27283_to_fp16)[name = string("aw_chunk_2823_cast_fp16")]; + fp16 var_27285_to_fp16 = const()[name = string("op_27285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2825_cast_fp16, y = var_27285_to_fp16)[name = string("aw_chunk_2825_cast_fp16")]; + fp16 var_27287_to_fp16 = const()[name = string("op_27287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2827_cast_fp16, y = var_27287_to_fp16)[name = string("aw_chunk_2827_cast_fp16")]; + fp16 var_27289_to_fp16 = const()[name = string("op_27289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2829_cast_fp16, y = var_27289_to_fp16)[name = string("aw_chunk_2829_cast_fp16")]; + fp16 var_27291_to_fp16 = const()[name = string("op_27291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2831_cast_fp16, y = var_27291_to_fp16)[name = string("aw_chunk_2831_cast_fp16")]; + fp16 var_27293_to_fp16 = const()[name = string("op_27293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2833_cast_fp16, y = var_27293_to_fp16)[name = string("aw_chunk_2833_cast_fp16")]; + fp16 var_27295_to_fp16 = const()[name = string("op_27295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2835_cast_fp16, y = var_27295_to_fp16)[name = string("aw_chunk_2835_cast_fp16")]; + fp16 var_27297_to_fp16 = const()[name = string("op_27297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2837_cast_fp16, y = var_27297_to_fp16)[name = string("aw_chunk_2837_cast_fp16")]; + fp16 var_27299_to_fp16 = const()[name = string("op_27299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2839_cast_fp16, y = var_27299_to_fp16)[name = string("aw_chunk_2839_cast_fp16")]; + fp16 var_27301_to_fp16 = const()[name = string("op_27301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2841_cast_fp16, y = var_27301_to_fp16)[name = string("aw_chunk_2841_cast_fp16")]; + fp16 var_27303_to_fp16 = const()[name = string("op_27303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2843_cast_fp16, y = var_27303_to_fp16)[name = string("aw_chunk_2843_cast_fp16")]; + fp16 var_27305_to_fp16 = const()[name = string("op_27305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2845_cast_fp16, y = var_27305_to_fp16)[name = string("aw_chunk_2845_cast_fp16")]; + fp16 var_27307_to_fp16 = const()[name = string("op_27307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2847_cast_fp16, y = var_27307_to_fp16)[name = string("aw_chunk_2847_cast_fp16")]; + fp16 var_27309_to_fp16 = const()[name = string("op_27309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2849_cast_fp16, y = var_27309_to_fp16)[name = string("aw_chunk_2849_cast_fp16")]; + fp16 var_27311_to_fp16 = const()[name = string("op_27311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2851_cast_fp16, y = var_27311_to_fp16)[name = string("aw_chunk_2851_cast_fp16")]; + fp16 var_27313_to_fp16 = const()[name = string("op_27313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2853_cast_fp16, y = var_27313_to_fp16)[name = string("aw_chunk_2853_cast_fp16")]; + fp16 var_27315_to_fp16 = const()[name = string("op_27315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2855_cast_fp16, y = var_27315_to_fp16)[name = string("aw_chunk_2855_cast_fp16")]; + fp16 var_27317_to_fp16 = const()[name = string("op_27317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2857_cast_fp16, y = var_27317_to_fp16)[name = string("aw_chunk_2857_cast_fp16")]; + fp16 var_27319_to_fp16 = const()[name = string("op_27319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2859_cast_fp16, y = var_27319_to_fp16)[name = string("aw_chunk_2859_cast_fp16")]; + fp16 var_27321_to_fp16 = const()[name = string("op_27321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2861_cast_fp16, y = var_27321_to_fp16)[name = string("aw_chunk_2861_cast_fp16")]; + fp16 var_27323_to_fp16 = const()[name = string("op_27323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2863_cast_fp16, y = var_27323_to_fp16)[name = string("aw_chunk_2863_cast_fp16")]; + fp16 var_27325_to_fp16 = const()[name = string("op_27325_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2865_cast_fp16, y = var_27325_to_fp16)[name = string("aw_chunk_2865_cast_fp16")]; + fp16 var_27327_to_fp16 = const()[name = string("op_27327_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2867_cast_fp16, y = var_27327_to_fp16)[name = string("aw_chunk_2867_cast_fp16")]; + fp16 var_27329_to_fp16 = const()[name = string("op_27329_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2869_cast_fp16, y = var_27329_to_fp16)[name = string("aw_chunk_2869_cast_fp16")]; + fp16 var_27331_to_fp16 = const()[name = string("op_27331_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2871_cast_fp16, y = var_27331_to_fp16)[name = string("aw_chunk_2871_cast_fp16")]; + fp16 var_27333_to_fp16 = const()[name = string("op_27333_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2873_cast_fp16, y = var_27333_to_fp16)[name = string("aw_chunk_2873_cast_fp16")]; + fp16 var_27335_to_fp16 = const()[name = string("op_27335_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2875_cast_fp16, y = var_27335_to_fp16)[name = string("aw_chunk_2875_cast_fp16")]; + fp16 var_27337_to_fp16 = const()[name = string("op_27337_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2877_cast_fp16, y = var_27337_to_fp16)[name = string("aw_chunk_2877_cast_fp16")]; + fp16 var_27339_to_fp16 = const()[name = string("op_27339_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2879_cast_fp16, y = var_27339_to_fp16)[name = string("aw_chunk_2879_cast_fp16")]; + tensor var_27341_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2721_cast_fp16)[name = string("op_27341_cast_fp16")]; + tensor var_27342_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2723_cast_fp16)[name = string("op_27342_cast_fp16")]; + tensor var_27343_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2725_cast_fp16)[name = string("op_27343_cast_fp16")]; + tensor var_27344_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2727_cast_fp16)[name = string("op_27344_cast_fp16")]; + tensor var_27345_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2729_cast_fp16)[name = string("op_27345_cast_fp16")]; + tensor var_27346_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2731_cast_fp16)[name = string("op_27346_cast_fp16")]; + tensor var_27347_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2733_cast_fp16)[name = string("op_27347_cast_fp16")]; + tensor var_27348_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2735_cast_fp16)[name = string("op_27348_cast_fp16")]; + tensor var_27349_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2737_cast_fp16)[name = string("op_27349_cast_fp16")]; + tensor var_27350_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2739_cast_fp16)[name = string("op_27350_cast_fp16")]; + tensor var_27351_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2741_cast_fp16)[name = string("op_27351_cast_fp16")]; + tensor var_27352_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2743_cast_fp16)[name = string("op_27352_cast_fp16")]; + tensor var_27353_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2745_cast_fp16)[name = string("op_27353_cast_fp16")]; + tensor var_27354_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2747_cast_fp16)[name = string("op_27354_cast_fp16")]; + tensor var_27355_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2749_cast_fp16)[name = string("op_27355_cast_fp16")]; + tensor var_27356_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2751_cast_fp16)[name = string("op_27356_cast_fp16")]; + tensor var_27357_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2753_cast_fp16)[name = string("op_27357_cast_fp16")]; + tensor var_27358_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2755_cast_fp16)[name = string("op_27358_cast_fp16")]; + tensor var_27359_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2757_cast_fp16)[name = string("op_27359_cast_fp16")]; + tensor var_27360_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2759_cast_fp16)[name = string("op_27360_cast_fp16")]; + tensor var_27361_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2761_cast_fp16)[name = string("op_27361_cast_fp16")]; + tensor var_27362_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2763_cast_fp16)[name = string("op_27362_cast_fp16")]; + tensor var_27363_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2765_cast_fp16)[name = string("op_27363_cast_fp16")]; + tensor var_27364_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2767_cast_fp16)[name = string("op_27364_cast_fp16")]; + tensor var_27365_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2769_cast_fp16)[name = string("op_27365_cast_fp16")]; + tensor var_27366_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2771_cast_fp16)[name = string("op_27366_cast_fp16")]; + tensor var_27367_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2773_cast_fp16)[name = string("op_27367_cast_fp16")]; + tensor var_27368_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2775_cast_fp16)[name = string("op_27368_cast_fp16")]; + tensor var_27369_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2777_cast_fp16)[name = string("op_27369_cast_fp16")]; + tensor var_27370_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2779_cast_fp16)[name = string("op_27370_cast_fp16")]; + tensor var_27371_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2781_cast_fp16)[name = string("op_27371_cast_fp16")]; + tensor var_27372_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2783_cast_fp16)[name = string("op_27372_cast_fp16")]; + tensor var_27373_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2785_cast_fp16)[name = string("op_27373_cast_fp16")]; + tensor var_27374_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2787_cast_fp16)[name = string("op_27374_cast_fp16")]; + tensor var_27375_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2789_cast_fp16)[name = string("op_27375_cast_fp16")]; + tensor var_27376_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2791_cast_fp16)[name = string("op_27376_cast_fp16")]; + tensor var_27377_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2793_cast_fp16)[name = string("op_27377_cast_fp16")]; + tensor var_27378_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2795_cast_fp16)[name = string("op_27378_cast_fp16")]; + tensor var_27379_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2797_cast_fp16)[name = string("op_27379_cast_fp16")]; + tensor var_27380_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2799_cast_fp16)[name = string("op_27380_cast_fp16")]; + tensor var_27381_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2801_cast_fp16)[name = string("op_27381_cast_fp16")]; + tensor var_27382_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2803_cast_fp16)[name = string("op_27382_cast_fp16")]; + tensor var_27383_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2805_cast_fp16)[name = string("op_27383_cast_fp16")]; + tensor var_27384_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2807_cast_fp16)[name = string("op_27384_cast_fp16")]; + tensor var_27385_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2809_cast_fp16)[name = string("op_27385_cast_fp16")]; + tensor var_27386_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2811_cast_fp16)[name = string("op_27386_cast_fp16")]; + tensor var_27387_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2813_cast_fp16)[name = string("op_27387_cast_fp16")]; + tensor var_27388_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2815_cast_fp16)[name = string("op_27388_cast_fp16")]; + tensor var_27389_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2817_cast_fp16)[name = string("op_27389_cast_fp16")]; + tensor var_27390_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2819_cast_fp16)[name = string("op_27390_cast_fp16")]; + tensor var_27391_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2821_cast_fp16)[name = string("op_27391_cast_fp16")]; + tensor var_27392_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2823_cast_fp16)[name = string("op_27392_cast_fp16")]; + tensor var_27393_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2825_cast_fp16)[name = string("op_27393_cast_fp16")]; + tensor var_27394_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2827_cast_fp16)[name = string("op_27394_cast_fp16")]; + tensor var_27395_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2829_cast_fp16)[name = string("op_27395_cast_fp16")]; + tensor var_27396_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2831_cast_fp16)[name = string("op_27396_cast_fp16")]; + tensor var_27397_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2833_cast_fp16)[name = string("op_27397_cast_fp16")]; + tensor var_27398_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2835_cast_fp16)[name = string("op_27398_cast_fp16")]; + tensor var_27399_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2837_cast_fp16)[name = string("op_27399_cast_fp16")]; + tensor var_27400_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2839_cast_fp16)[name = string("op_27400_cast_fp16")]; + tensor var_27401_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2841_cast_fp16)[name = string("op_27401_cast_fp16")]; + tensor var_27402_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2843_cast_fp16)[name = string("op_27402_cast_fp16")]; + tensor var_27403_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2845_cast_fp16)[name = string("op_27403_cast_fp16")]; + tensor var_27404_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2847_cast_fp16)[name = string("op_27404_cast_fp16")]; + tensor var_27405_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2849_cast_fp16)[name = string("op_27405_cast_fp16")]; + tensor var_27406_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2851_cast_fp16)[name = string("op_27406_cast_fp16")]; + tensor var_27407_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2853_cast_fp16)[name = string("op_27407_cast_fp16")]; + tensor var_27408_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2855_cast_fp16)[name = string("op_27408_cast_fp16")]; + tensor var_27409_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2857_cast_fp16)[name = string("op_27409_cast_fp16")]; + tensor var_27410_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2859_cast_fp16)[name = string("op_27410_cast_fp16")]; + tensor var_27411_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2861_cast_fp16)[name = string("op_27411_cast_fp16")]; + tensor var_27412_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2863_cast_fp16)[name = string("op_27412_cast_fp16")]; + tensor var_27413_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2865_cast_fp16)[name = string("op_27413_cast_fp16")]; + tensor var_27414_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2867_cast_fp16)[name = string("op_27414_cast_fp16")]; + tensor var_27415_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2869_cast_fp16)[name = string("op_27415_cast_fp16")]; + tensor var_27416_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2871_cast_fp16)[name = string("op_27416_cast_fp16")]; + tensor var_27417_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2873_cast_fp16)[name = string("op_27417_cast_fp16")]; + tensor var_27418_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2875_cast_fp16)[name = string("op_27418_cast_fp16")]; + tensor var_27419_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2877_cast_fp16)[name = string("op_27419_cast_fp16")]; + tensor var_27420_cast_fp16 = softmax(axis = var_26166, x = aw_chunk_2879_cast_fp16)[name = string("op_27420_cast_fp16")]; + string var_27422_equation_0 = const()[name = string("op_27422_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27422_cast_fp16 = einsum(equation = var_27422_equation_0, values = (var_26942_cast_fp16, var_27341_cast_fp16))[name = string("op_27422_cast_fp16")]; + string var_27424_equation_0 = const()[name = string("op_27424_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27424_cast_fp16 = einsum(equation = var_27424_equation_0, values = (var_26942_cast_fp16, var_27342_cast_fp16))[name = string("op_27424_cast_fp16")]; + string var_27426_equation_0 = const()[name = string("op_27426_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27426_cast_fp16 = einsum(equation = var_27426_equation_0, values = (var_26942_cast_fp16, var_27343_cast_fp16))[name = string("op_27426_cast_fp16")]; + string var_27428_equation_0 = const()[name = string("op_27428_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27428_cast_fp16 = einsum(equation = var_27428_equation_0, values = (var_26942_cast_fp16, var_27344_cast_fp16))[name = string("op_27428_cast_fp16")]; + string var_27430_equation_0 = const()[name = string("op_27430_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27430_cast_fp16 = einsum(equation = var_27430_equation_0, values = (var_26946_cast_fp16, var_27345_cast_fp16))[name = string("op_27430_cast_fp16")]; + string var_27432_equation_0 = const()[name = string("op_27432_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27432_cast_fp16 = einsum(equation = var_27432_equation_0, values = (var_26946_cast_fp16, var_27346_cast_fp16))[name = string("op_27432_cast_fp16")]; + string var_27434_equation_0 = const()[name = string("op_27434_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27434_cast_fp16 = einsum(equation = var_27434_equation_0, values = (var_26946_cast_fp16, var_27347_cast_fp16))[name = string("op_27434_cast_fp16")]; + string var_27436_equation_0 = const()[name = string("op_27436_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27436_cast_fp16 = einsum(equation = var_27436_equation_0, values = (var_26946_cast_fp16, var_27348_cast_fp16))[name = string("op_27436_cast_fp16")]; + string var_27438_equation_0 = const()[name = string("op_27438_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27438_cast_fp16 = einsum(equation = var_27438_equation_0, values = (var_26950_cast_fp16, var_27349_cast_fp16))[name = string("op_27438_cast_fp16")]; + string var_27440_equation_0 = const()[name = string("op_27440_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27440_cast_fp16 = einsum(equation = var_27440_equation_0, values = (var_26950_cast_fp16, var_27350_cast_fp16))[name = string("op_27440_cast_fp16")]; + string var_27442_equation_0 = const()[name = string("op_27442_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27442_cast_fp16 = einsum(equation = var_27442_equation_0, values = (var_26950_cast_fp16, var_27351_cast_fp16))[name = string("op_27442_cast_fp16")]; + string var_27444_equation_0 = const()[name = string("op_27444_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27444_cast_fp16 = einsum(equation = var_27444_equation_0, values = (var_26950_cast_fp16, var_27352_cast_fp16))[name = string("op_27444_cast_fp16")]; + string var_27446_equation_0 = const()[name = string("op_27446_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27446_cast_fp16 = einsum(equation = var_27446_equation_0, values = (var_26954_cast_fp16, var_27353_cast_fp16))[name = string("op_27446_cast_fp16")]; + string var_27448_equation_0 = const()[name = string("op_27448_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27448_cast_fp16 = einsum(equation = var_27448_equation_0, values = (var_26954_cast_fp16, var_27354_cast_fp16))[name = string("op_27448_cast_fp16")]; + string var_27450_equation_0 = const()[name = string("op_27450_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27450_cast_fp16 = einsum(equation = var_27450_equation_0, values = (var_26954_cast_fp16, var_27355_cast_fp16))[name = string("op_27450_cast_fp16")]; + string var_27452_equation_0 = const()[name = string("op_27452_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27452_cast_fp16 = einsum(equation = var_27452_equation_0, values = (var_26954_cast_fp16, var_27356_cast_fp16))[name = string("op_27452_cast_fp16")]; + string var_27454_equation_0 = const()[name = string("op_27454_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27454_cast_fp16 = einsum(equation = var_27454_equation_0, values = (var_26958_cast_fp16, var_27357_cast_fp16))[name = string("op_27454_cast_fp16")]; + string var_27456_equation_0 = const()[name = string("op_27456_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27456_cast_fp16 = einsum(equation = var_27456_equation_0, values = (var_26958_cast_fp16, var_27358_cast_fp16))[name = string("op_27456_cast_fp16")]; + string var_27458_equation_0 = const()[name = string("op_27458_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27458_cast_fp16 = einsum(equation = var_27458_equation_0, values = (var_26958_cast_fp16, var_27359_cast_fp16))[name = string("op_27458_cast_fp16")]; + string var_27460_equation_0 = const()[name = string("op_27460_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27460_cast_fp16 = einsum(equation = var_27460_equation_0, values = (var_26958_cast_fp16, var_27360_cast_fp16))[name = string("op_27460_cast_fp16")]; + string var_27462_equation_0 = const()[name = string("op_27462_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27462_cast_fp16 = einsum(equation = var_27462_equation_0, values = (var_26962_cast_fp16, var_27361_cast_fp16))[name = string("op_27462_cast_fp16")]; + string var_27464_equation_0 = const()[name = string("op_27464_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27464_cast_fp16 = einsum(equation = var_27464_equation_0, values = (var_26962_cast_fp16, var_27362_cast_fp16))[name = string("op_27464_cast_fp16")]; + string var_27466_equation_0 = const()[name = string("op_27466_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27466_cast_fp16 = einsum(equation = var_27466_equation_0, values = (var_26962_cast_fp16, var_27363_cast_fp16))[name = string("op_27466_cast_fp16")]; + string var_27468_equation_0 = const()[name = string("op_27468_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27468_cast_fp16 = einsum(equation = var_27468_equation_0, values = (var_26962_cast_fp16, var_27364_cast_fp16))[name = string("op_27468_cast_fp16")]; + string var_27470_equation_0 = const()[name = string("op_27470_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27470_cast_fp16 = einsum(equation = var_27470_equation_0, values = (var_26966_cast_fp16, var_27365_cast_fp16))[name = string("op_27470_cast_fp16")]; + string var_27472_equation_0 = const()[name = string("op_27472_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27472_cast_fp16 = einsum(equation = var_27472_equation_0, values = (var_26966_cast_fp16, var_27366_cast_fp16))[name = string("op_27472_cast_fp16")]; + string var_27474_equation_0 = const()[name = string("op_27474_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27474_cast_fp16 = einsum(equation = var_27474_equation_0, values = (var_26966_cast_fp16, var_27367_cast_fp16))[name = string("op_27474_cast_fp16")]; + string var_27476_equation_0 = const()[name = string("op_27476_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27476_cast_fp16 = einsum(equation = var_27476_equation_0, values = (var_26966_cast_fp16, var_27368_cast_fp16))[name = string("op_27476_cast_fp16")]; + string var_27478_equation_0 = const()[name = string("op_27478_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27478_cast_fp16 = einsum(equation = var_27478_equation_0, values = (var_26970_cast_fp16, var_27369_cast_fp16))[name = string("op_27478_cast_fp16")]; + string var_27480_equation_0 = const()[name = string("op_27480_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27480_cast_fp16 = einsum(equation = var_27480_equation_0, values = (var_26970_cast_fp16, var_27370_cast_fp16))[name = string("op_27480_cast_fp16")]; + string var_27482_equation_0 = const()[name = string("op_27482_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27482_cast_fp16 = einsum(equation = var_27482_equation_0, values = (var_26970_cast_fp16, var_27371_cast_fp16))[name = string("op_27482_cast_fp16")]; + string var_27484_equation_0 = const()[name = string("op_27484_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27484_cast_fp16 = einsum(equation = var_27484_equation_0, values = (var_26970_cast_fp16, var_27372_cast_fp16))[name = string("op_27484_cast_fp16")]; + string var_27486_equation_0 = const()[name = string("op_27486_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27486_cast_fp16 = einsum(equation = var_27486_equation_0, values = (var_26974_cast_fp16, var_27373_cast_fp16))[name = string("op_27486_cast_fp16")]; + string var_27488_equation_0 = const()[name = string("op_27488_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27488_cast_fp16 = einsum(equation = var_27488_equation_0, values = (var_26974_cast_fp16, var_27374_cast_fp16))[name = string("op_27488_cast_fp16")]; + string var_27490_equation_0 = const()[name = string("op_27490_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27490_cast_fp16 = einsum(equation = var_27490_equation_0, values = (var_26974_cast_fp16, var_27375_cast_fp16))[name = string("op_27490_cast_fp16")]; + string var_27492_equation_0 = const()[name = string("op_27492_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27492_cast_fp16 = einsum(equation = var_27492_equation_0, values = (var_26974_cast_fp16, var_27376_cast_fp16))[name = string("op_27492_cast_fp16")]; + string var_27494_equation_0 = const()[name = string("op_27494_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27494_cast_fp16 = einsum(equation = var_27494_equation_0, values = (var_26978_cast_fp16, var_27377_cast_fp16))[name = string("op_27494_cast_fp16")]; + string var_27496_equation_0 = const()[name = string("op_27496_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27496_cast_fp16 = einsum(equation = var_27496_equation_0, values = (var_26978_cast_fp16, var_27378_cast_fp16))[name = string("op_27496_cast_fp16")]; + string var_27498_equation_0 = const()[name = string("op_27498_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27498_cast_fp16 = einsum(equation = var_27498_equation_0, values = (var_26978_cast_fp16, var_27379_cast_fp16))[name = string("op_27498_cast_fp16")]; + string var_27500_equation_0 = const()[name = string("op_27500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27500_cast_fp16 = einsum(equation = var_27500_equation_0, values = (var_26978_cast_fp16, var_27380_cast_fp16))[name = string("op_27500_cast_fp16")]; + string var_27502_equation_0 = const()[name = string("op_27502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27502_cast_fp16 = einsum(equation = var_27502_equation_0, values = (var_26982_cast_fp16, var_27381_cast_fp16))[name = string("op_27502_cast_fp16")]; + string var_27504_equation_0 = const()[name = string("op_27504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27504_cast_fp16 = einsum(equation = var_27504_equation_0, values = (var_26982_cast_fp16, var_27382_cast_fp16))[name = string("op_27504_cast_fp16")]; + string var_27506_equation_0 = const()[name = string("op_27506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27506_cast_fp16 = einsum(equation = var_27506_equation_0, values = (var_26982_cast_fp16, var_27383_cast_fp16))[name = string("op_27506_cast_fp16")]; + string var_27508_equation_0 = const()[name = string("op_27508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27508_cast_fp16 = einsum(equation = var_27508_equation_0, values = (var_26982_cast_fp16, var_27384_cast_fp16))[name = string("op_27508_cast_fp16")]; + string var_27510_equation_0 = const()[name = string("op_27510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27510_cast_fp16 = einsum(equation = var_27510_equation_0, values = (var_26986_cast_fp16, var_27385_cast_fp16))[name = string("op_27510_cast_fp16")]; + string var_27512_equation_0 = const()[name = string("op_27512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27512_cast_fp16 = einsum(equation = var_27512_equation_0, values = (var_26986_cast_fp16, var_27386_cast_fp16))[name = string("op_27512_cast_fp16")]; + string var_27514_equation_0 = const()[name = string("op_27514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27514_cast_fp16 = einsum(equation = var_27514_equation_0, values = (var_26986_cast_fp16, var_27387_cast_fp16))[name = string("op_27514_cast_fp16")]; + string var_27516_equation_0 = const()[name = string("op_27516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27516_cast_fp16 = einsum(equation = var_27516_equation_0, values = (var_26986_cast_fp16, var_27388_cast_fp16))[name = string("op_27516_cast_fp16")]; + string var_27518_equation_0 = const()[name = string("op_27518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27518_cast_fp16 = einsum(equation = var_27518_equation_0, values = (var_26990_cast_fp16, var_27389_cast_fp16))[name = string("op_27518_cast_fp16")]; + string var_27520_equation_0 = const()[name = string("op_27520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27520_cast_fp16 = einsum(equation = var_27520_equation_0, values = (var_26990_cast_fp16, var_27390_cast_fp16))[name = string("op_27520_cast_fp16")]; + string var_27522_equation_0 = const()[name = string("op_27522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27522_cast_fp16 = einsum(equation = var_27522_equation_0, values = (var_26990_cast_fp16, var_27391_cast_fp16))[name = string("op_27522_cast_fp16")]; + string var_27524_equation_0 = const()[name = string("op_27524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27524_cast_fp16 = einsum(equation = var_27524_equation_0, values = (var_26990_cast_fp16, var_27392_cast_fp16))[name = string("op_27524_cast_fp16")]; + string var_27526_equation_0 = const()[name = string("op_27526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27526_cast_fp16 = einsum(equation = var_27526_equation_0, values = (var_26994_cast_fp16, var_27393_cast_fp16))[name = string("op_27526_cast_fp16")]; + string var_27528_equation_0 = const()[name = string("op_27528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27528_cast_fp16 = einsum(equation = var_27528_equation_0, values = (var_26994_cast_fp16, var_27394_cast_fp16))[name = string("op_27528_cast_fp16")]; + string var_27530_equation_0 = const()[name = string("op_27530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27530_cast_fp16 = einsum(equation = var_27530_equation_0, values = (var_26994_cast_fp16, var_27395_cast_fp16))[name = string("op_27530_cast_fp16")]; + string var_27532_equation_0 = const()[name = string("op_27532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27532_cast_fp16 = einsum(equation = var_27532_equation_0, values = (var_26994_cast_fp16, var_27396_cast_fp16))[name = string("op_27532_cast_fp16")]; + string var_27534_equation_0 = const()[name = string("op_27534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27534_cast_fp16 = einsum(equation = var_27534_equation_0, values = (var_26998_cast_fp16, var_27397_cast_fp16))[name = string("op_27534_cast_fp16")]; + string var_27536_equation_0 = const()[name = string("op_27536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27536_cast_fp16 = einsum(equation = var_27536_equation_0, values = (var_26998_cast_fp16, var_27398_cast_fp16))[name = string("op_27536_cast_fp16")]; + string var_27538_equation_0 = const()[name = string("op_27538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27538_cast_fp16 = einsum(equation = var_27538_equation_0, values = (var_26998_cast_fp16, var_27399_cast_fp16))[name = string("op_27538_cast_fp16")]; + string var_27540_equation_0 = const()[name = string("op_27540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27540_cast_fp16 = einsum(equation = var_27540_equation_0, values = (var_26998_cast_fp16, var_27400_cast_fp16))[name = string("op_27540_cast_fp16")]; + string var_27542_equation_0 = const()[name = string("op_27542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27542_cast_fp16 = einsum(equation = var_27542_equation_0, values = (var_27002_cast_fp16, var_27401_cast_fp16))[name = string("op_27542_cast_fp16")]; + string var_27544_equation_0 = const()[name = string("op_27544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27544_cast_fp16 = einsum(equation = var_27544_equation_0, values = (var_27002_cast_fp16, var_27402_cast_fp16))[name = string("op_27544_cast_fp16")]; + string var_27546_equation_0 = const()[name = string("op_27546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27546_cast_fp16 = einsum(equation = var_27546_equation_0, values = (var_27002_cast_fp16, var_27403_cast_fp16))[name = string("op_27546_cast_fp16")]; + string var_27548_equation_0 = const()[name = string("op_27548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27548_cast_fp16 = einsum(equation = var_27548_equation_0, values = (var_27002_cast_fp16, var_27404_cast_fp16))[name = string("op_27548_cast_fp16")]; + string var_27550_equation_0 = const()[name = string("op_27550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27550_cast_fp16 = einsum(equation = var_27550_equation_0, values = (var_27006_cast_fp16, var_27405_cast_fp16))[name = string("op_27550_cast_fp16")]; + string var_27552_equation_0 = const()[name = string("op_27552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27552_cast_fp16 = einsum(equation = var_27552_equation_0, values = (var_27006_cast_fp16, var_27406_cast_fp16))[name = string("op_27552_cast_fp16")]; + string var_27554_equation_0 = const()[name = string("op_27554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27554_cast_fp16 = einsum(equation = var_27554_equation_0, values = (var_27006_cast_fp16, var_27407_cast_fp16))[name = string("op_27554_cast_fp16")]; + string var_27556_equation_0 = const()[name = string("op_27556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27556_cast_fp16 = einsum(equation = var_27556_equation_0, values = (var_27006_cast_fp16, var_27408_cast_fp16))[name = string("op_27556_cast_fp16")]; + string var_27558_equation_0 = const()[name = string("op_27558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27558_cast_fp16 = einsum(equation = var_27558_equation_0, values = (var_27010_cast_fp16, var_27409_cast_fp16))[name = string("op_27558_cast_fp16")]; + string var_27560_equation_0 = const()[name = string("op_27560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27560_cast_fp16 = einsum(equation = var_27560_equation_0, values = (var_27010_cast_fp16, var_27410_cast_fp16))[name = string("op_27560_cast_fp16")]; + string var_27562_equation_0 = const()[name = string("op_27562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27562_cast_fp16 = einsum(equation = var_27562_equation_0, values = (var_27010_cast_fp16, var_27411_cast_fp16))[name = string("op_27562_cast_fp16")]; + string var_27564_equation_0 = const()[name = string("op_27564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27564_cast_fp16 = einsum(equation = var_27564_equation_0, values = (var_27010_cast_fp16, var_27412_cast_fp16))[name = string("op_27564_cast_fp16")]; + string var_27566_equation_0 = const()[name = string("op_27566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27566_cast_fp16 = einsum(equation = var_27566_equation_0, values = (var_27014_cast_fp16, var_27413_cast_fp16))[name = string("op_27566_cast_fp16")]; + string var_27568_equation_0 = const()[name = string("op_27568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27568_cast_fp16 = einsum(equation = var_27568_equation_0, values = (var_27014_cast_fp16, var_27414_cast_fp16))[name = string("op_27568_cast_fp16")]; + string var_27570_equation_0 = const()[name = string("op_27570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27570_cast_fp16 = einsum(equation = var_27570_equation_0, values = (var_27014_cast_fp16, var_27415_cast_fp16))[name = string("op_27570_cast_fp16")]; + string var_27572_equation_0 = const()[name = string("op_27572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27572_cast_fp16 = einsum(equation = var_27572_equation_0, values = (var_27014_cast_fp16, var_27416_cast_fp16))[name = string("op_27572_cast_fp16")]; + string var_27574_equation_0 = const()[name = string("op_27574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27574_cast_fp16 = einsum(equation = var_27574_equation_0, values = (var_27018_cast_fp16, var_27417_cast_fp16))[name = string("op_27574_cast_fp16")]; + string var_27576_equation_0 = const()[name = string("op_27576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27576_cast_fp16 = einsum(equation = var_27576_equation_0, values = (var_27018_cast_fp16, var_27418_cast_fp16))[name = string("op_27576_cast_fp16")]; + string var_27578_equation_0 = const()[name = string("op_27578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27578_cast_fp16 = einsum(equation = var_27578_equation_0, values = (var_27018_cast_fp16, var_27419_cast_fp16))[name = string("op_27578_cast_fp16")]; + string var_27580_equation_0 = const()[name = string("op_27580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_27580_cast_fp16 = einsum(equation = var_27580_equation_0, values = (var_27018_cast_fp16, var_27420_cast_fp16))[name = string("op_27580_cast_fp16")]; + bool var_27582_interleave_0 = const()[name = string("op_27582_interleave_0"), val = bool(false)]; + tensor var_27582_cast_fp16 = concat(axis = var_26141, interleave = var_27582_interleave_0, values = (var_27422_cast_fp16, var_27424_cast_fp16, var_27426_cast_fp16, var_27428_cast_fp16))[name = string("op_27582_cast_fp16")]; + bool var_27584_interleave_0 = const()[name = string("op_27584_interleave_0"), val = bool(false)]; + tensor var_27584_cast_fp16 = concat(axis = var_26141, interleave = var_27584_interleave_0, values = (var_27430_cast_fp16, var_27432_cast_fp16, var_27434_cast_fp16, var_27436_cast_fp16))[name = string("op_27584_cast_fp16")]; + bool var_27586_interleave_0 = const()[name = string("op_27586_interleave_0"), val = bool(false)]; + tensor var_27586_cast_fp16 = concat(axis = var_26141, interleave = var_27586_interleave_0, values = (var_27438_cast_fp16, var_27440_cast_fp16, var_27442_cast_fp16, var_27444_cast_fp16))[name = string("op_27586_cast_fp16")]; + bool var_27588_interleave_0 = const()[name = string("op_27588_interleave_0"), val = bool(false)]; + tensor var_27588_cast_fp16 = concat(axis = var_26141, interleave = var_27588_interleave_0, values = (var_27446_cast_fp16, var_27448_cast_fp16, var_27450_cast_fp16, var_27452_cast_fp16))[name = string("op_27588_cast_fp16")]; + bool var_27590_interleave_0 = const()[name = string("op_27590_interleave_0"), val = bool(false)]; + tensor var_27590_cast_fp16 = concat(axis = var_26141, interleave = var_27590_interleave_0, values = (var_27454_cast_fp16, var_27456_cast_fp16, var_27458_cast_fp16, var_27460_cast_fp16))[name = string("op_27590_cast_fp16")]; + bool var_27592_interleave_0 = const()[name = string("op_27592_interleave_0"), val = bool(false)]; + tensor var_27592_cast_fp16 = concat(axis = var_26141, interleave = var_27592_interleave_0, values = (var_27462_cast_fp16, var_27464_cast_fp16, var_27466_cast_fp16, var_27468_cast_fp16))[name = string("op_27592_cast_fp16")]; + bool var_27594_interleave_0 = const()[name = string("op_27594_interleave_0"), val = bool(false)]; + tensor var_27594_cast_fp16 = concat(axis = var_26141, interleave = var_27594_interleave_0, values = (var_27470_cast_fp16, var_27472_cast_fp16, var_27474_cast_fp16, var_27476_cast_fp16))[name = string("op_27594_cast_fp16")]; + bool var_27596_interleave_0 = const()[name = string("op_27596_interleave_0"), val = bool(false)]; + tensor var_27596_cast_fp16 = concat(axis = var_26141, interleave = var_27596_interleave_0, values = (var_27478_cast_fp16, var_27480_cast_fp16, var_27482_cast_fp16, var_27484_cast_fp16))[name = string("op_27596_cast_fp16")]; + bool var_27598_interleave_0 = const()[name = string("op_27598_interleave_0"), val = bool(false)]; + tensor var_27598_cast_fp16 = concat(axis = var_26141, interleave = var_27598_interleave_0, values = (var_27486_cast_fp16, var_27488_cast_fp16, var_27490_cast_fp16, var_27492_cast_fp16))[name = string("op_27598_cast_fp16")]; + bool var_27600_interleave_0 = const()[name = string("op_27600_interleave_0"), val = bool(false)]; + tensor var_27600_cast_fp16 = concat(axis = var_26141, interleave = var_27600_interleave_0, values = (var_27494_cast_fp16, var_27496_cast_fp16, var_27498_cast_fp16, var_27500_cast_fp16))[name = string("op_27600_cast_fp16")]; + bool var_27602_interleave_0 = const()[name = string("op_27602_interleave_0"), val = bool(false)]; + tensor var_27602_cast_fp16 = concat(axis = var_26141, interleave = var_27602_interleave_0, values = (var_27502_cast_fp16, var_27504_cast_fp16, var_27506_cast_fp16, var_27508_cast_fp16))[name = string("op_27602_cast_fp16")]; + bool var_27604_interleave_0 = const()[name = string("op_27604_interleave_0"), val = bool(false)]; + tensor var_27604_cast_fp16 = concat(axis = var_26141, interleave = var_27604_interleave_0, values = (var_27510_cast_fp16, var_27512_cast_fp16, var_27514_cast_fp16, var_27516_cast_fp16))[name = string("op_27604_cast_fp16")]; + bool var_27606_interleave_0 = const()[name = string("op_27606_interleave_0"), val = bool(false)]; + tensor var_27606_cast_fp16 = concat(axis = var_26141, interleave = var_27606_interleave_0, values = (var_27518_cast_fp16, var_27520_cast_fp16, var_27522_cast_fp16, var_27524_cast_fp16))[name = string("op_27606_cast_fp16")]; + bool var_27608_interleave_0 = const()[name = string("op_27608_interleave_0"), val = bool(false)]; + tensor var_27608_cast_fp16 = concat(axis = var_26141, interleave = var_27608_interleave_0, values = (var_27526_cast_fp16, var_27528_cast_fp16, var_27530_cast_fp16, var_27532_cast_fp16))[name = string("op_27608_cast_fp16")]; + bool var_27610_interleave_0 = const()[name = string("op_27610_interleave_0"), val = bool(false)]; + tensor var_27610_cast_fp16 = concat(axis = var_26141, interleave = var_27610_interleave_0, values = (var_27534_cast_fp16, var_27536_cast_fp16, var_27538_cast_fp16, var_27540_cast_fp16))[name = string("op_27610_cast_fp16")]; + bool var_27612_interleave_0 = const()[name = string("op_27612_interleave_0"), val = bool(false)]; + tensor var_27612_cast_fp16 = concat(axis = var_26141, interleave = var_27612_interleave_0, values = (var_27542_cast_fp16, var_27544_cast_fp16, var_27546_cast_fp16, var_27548_cast_fp16))[name = string("op_27612_cast_fp16")]; + bool var_27614_interleave_0 = const()[name = string("op_27614_interleave_0"), val = bool(false)]; + tensor var_27614_cast_fp16 = concat(axis = var_26141, interleave = var_27614_interleave_0, values = (var_27550_cast_fp16, var_27552_cast_fp16, var_27554_cast_fp16, var_27556_cast_fp16))[name = string("op_27614_cast_fp16")]; + bool var_27616_interleave_0 = const()[name = string("op_27616_interleave_0"), val = bool(false)]; + tensor var_27616_cast_fp16 = concat(axis = var_26141, interleave = var_27616_interleave_0, values = (var_27558_cast_fp16, var_27560_cast_fp16, var_27562_cast_fp16, var_27564_cast_fp16))[name = string("op_27616_cast_fp16")]; + bool var_27618_interleave_0 = const()[name = string("op_27618_interleave_0"), val = bool(false)]; + tensor var_27618_cast_fp16 = concat(axis = var_26141, interleave = var_27618_interleave_0, values = (var_27566_cast_fp16, var_27568_cast_fp16, var_27570_cast_fp16, var_27572_cast_fp16))[name = string("op_27618_cast_fp16")]; + bool var_27620_interleave_0 = const()[name = string("op_27620_interleave_0"), val = bool(false)]; + tensor var_27620_cast_fp16 = concat(axis = var_26141, interleave = var_27620_interleave_0, values = (var_27574_cast_fp16, var_27576_cast_fp16, var_27578_cast_fp16, var_27580_cast_fp16))[name = string("op_27620_cast_fp16")]; + bool input_137_interleave_0 = const()[name = string("input_137_interleave_0"), val = bool(false)]; + tensor input_137_cast_fp16 = concat(axis = var_26166, interleave = input_137_interleave_0, values = (var_27582_cast_fp16, var_27584_cast_fp16, var_27586_cast_fp16, var_27588_cast_fp16, var_27590_cast_fp16, var_27592_cast_fp16, var_27594_cast_fp16, var_27596_cast_fp16, var_27598_cast_fp16, var_27600_cast_fp16, var_27602_cast_fp16, var_27604_cast_fp16, var_27606_cast_fp16, var_27608_cast_fp16, var_27610_cast_fp16, var_27612_cast_fp16, var_27614_cast_fp16, var_27616_cast_fp16, var_27618_cast_fp16, var_27620_cast_fp16))[name = string("input_137_cast_fp16")]; + string obj_71_pad_type_0 = const()[name = string("obj_71_pad_type_0"), val = string("valid")]; + tensor obj_71_strides_0 = const()[name = string("obj_71_strides_0"), val = tensor([1, 1])]; + tensor obj_71_pad_0 = const()[name = string("obj_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_71_dilations_0 = const()[name = string("obj_71_dilations_0"), val = tensor([1, 1])]; + int32 obj_71_groups_0 = const()[name = string("obj_71_groups_0"), val = int32(1)]; + tensor layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(693511040)))]; + tensor layers_17_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696787904)))]; + tensor obj_71_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_bias_to_fp16, dilations = obj_71_dilations_0, groups = obj_71_groups_0, pad = obj_71_pad_0, pad_type = obj_71_pad_type_0, strides = obj_71_strides_0, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("obj_71_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_71_cast_fp16)[name = string("inputs_71_cast_fp16")]; + tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; + fp16 var_27639_to_fp16 = const()[name = string("op_27639_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_27639_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; + tensor input_139_gamma_0_to_fp16 = const()[name = string("input_139_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696790528)))]; + tensor input_139_beta_0_to_fp16 = const()[name = string("input_139_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696793152)))]; + fp16 input_139_epsilon_0_to_fp16 = const()[name = string("input_139_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_139_cast_fp16 = batch_norm(beta = input_139_beta_0_to_fp16, epsilon = input_139_epsilon_0_to_fp16, gamma = input_139_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_139_cast_fp16")]; + string input_141_pad_type_0 = const()[name = string("input_141_pad_type_0"), val = string("valid")]; + tensor input_141_strides_0 = const()[name = string("input_141_strides_0"), val = tensor([1, 1])]; + tensor input_141_pad_0 = const()[name = string("input_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_141_dilations_0 = const()[name = string("input_141_dilations_0"), val = tensor([1, 1])]; + int32 input_141_groups_0 = const()[name = string("input_141_groups_0"), val = int32(1)]; + tensor layers_17_fc1_weight_to_fp16 = const()[name = string("layers_17_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(696795776)))]; + tensor layers_17_fc1_bias_to_fp16 = const()[name = string("layers_17_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709903040)))]; + tensor input_141_cast_fp16 = conv(bias = layers_17_fc1_bias_to_fp16, dilations = input_141_dilations_0, groups = input_141_groups_0, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = input_141_strides_0, weight = layers_17_fc1_weight_to_fp16, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")]; + string input_143_mode_0 = const()[name = string("input_143_mode_0"), val = string("EXACT")]; + tensor input_143_cast_fp16 = gelu(mode = input_143_mode_0, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")]; + string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")]; + tensor hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)]; + tensor layers_17_fc2_weight_to_fp16 = const()[name = string("layers_17_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709913344)))]; + tensor layers_17_fc2_bias_to_fp16 = const()[name = string("layers_17_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723020608)))]; + tensor hidden_states_39_cast_fp16 = conv(bias = layers_17_fc2_bias_to_fp16, dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = layers_17_fc2_weight_to_fp16, x = input_143_cast_fp16)[name = string("hidden_states_39_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("inputs_73_cast_fp16")]; + int32 var_27668 = const()[name = string("op_27668"), val = int32(3)]; + int32 var_27693 = const()[name = string("op_27693"), val = int32(1)]; + tensor out_73_axes_0 = const()[name = string("out_73_axes_0"), val = tensor([1])]; + fp16 var_27710_to_fp16 = const()[name = string("op_27710_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_27710_to_fp16, x = inputs_73_cast_fp16)[name = string("out_73_cast_fp16")]; + tensor obj_73_gamma_0_to_fp16 = const()[name = string("obj_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723023232)))]; + tensor obj_73_beta_0_to_fp16 = const()[name = string("obj_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723025856)))]; + fp16 obj_73_epsilon_0_to_fp16 = const()[name = string("obj_73_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_73_cast_fp16 = batch_norm(beta = obj_73_beta_0_to_fp16, epsilon = obj_73_epsilon_0_to_fp16, gamma = obj_73_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = string("obj_73_cast_fp16")]; + string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; + tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; + tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; + int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723028480)))]; + tensor layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726305344)))]; + tensor query_37_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("query_37_cast_fp16")]; + string key_37_pad_type_0 = const()[name = string("key_37_pad_type_0"), val = string("valid")]; + tensor key_37_strides_0 = const()[name = string("key_37_strides_0"), val = tensor([1, 1])]; + tensor key_37_pad_0 = const()[name = string("key_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_37_dilations_0 = const()[name = string("key_37_dilations_0"), val = tensor([1, 1])]; + int32 key_37_groups_0 = const()[name = string("key_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(726307968)))]; + tensor key_37_cast_fp16 = conv(dilations = key_37_dilations_0, groups = key_37_groups_0, pad = key_37_pad_0, pad_type = key_37_pad_type_0, strides = key_37_strides_0, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("key_37_cast_fp16")]; + string value_37_pad_type_0 = const()[name = string("value_37_pad_type_0"), val = string("valid")]; + tensor value_37_strides_0 = const()[name = string("value_37_strides_0"), val = tensor([1, 1])]; + tensor value_37_pad_0 = const()[name = string("value_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_37_dilations_0 = const()[name = string("value_37_dilations_0"), val = tensor([1, 1])]; + int32 value_37_groups_0 = const()[name = string("value_37_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(729584832)))]; + tensor layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732861696)))]; + tensor value_37_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_bias_to_fp16, dilations = value_37_dilations_0, groups = value_37_groups_0, pad = value_37_pad_0, pad_type = value_37_pad_type_0, strides = value_37_strides_0, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = obj_73_cast_fp16)[name = string("value_37_cast_fp16")]; + tensor var_27748_begin_0 = const()[name = string("op_27748_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27748_end_0 = const()[name = string("op_27748_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27748_end_mask_0 = const()[name = string("op_27748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27748_cast_fp16 = slice_by_index(begin = var_27748_begin_0, end = var_27748_end_0, end_mask = var_27748_end_mask_0, x = query_37_cast_fp16)[name = string("op_27748_cast_fp16")]; + tensor var_27752_begin_0 = const()[name = string("op_27752_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27752_end_0 = const()[name = string("op_27752_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_27752_end_mask_0 = const()[name = string("op_27752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27752_cast_fp16 = slice_by_index(begin = var_27752_begin_0, end = var_27752_end_0, end_mask = var_27752_end_mask_0, x = query_37_cast_fp16)[name = string("op_27752_cast_fp16")]; + tensor var_27756_begin_0 = const()[name = string("op_27756_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27756_end_0 = const()[name = string("op_27756_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_27756_end_mask_0 = const()[name = string("op_27756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27756_cast_fp16 = slice_by_index(begin = var_27756_begin_0, end = var_27756_end_0, end_mask = var_27756_end_mask_0, x = query_37_cast_fp16)[name = string("op_27756_cast_fp16")]; + tensor var_27760_begin_0 = const()[name = string("op_27760_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27760_end_0 = const()[name = string("op_27760_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_27760_end_mask_0 = const()[name = string("op_27760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27760_cast_fp16 = slice_by_index(begin = var_27760_begin_0, end = var_27760_end_0, end_mask = var_27760_end_mask_0, x = query_37_cast_fp16)[name = string("op_27760_cast_fp16")]; + tensor var_27764_begin_0 = const()[name = string("op_27764_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27764_end_0 = const()[name = string("op_27764_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_27764_end_mask_0 = const()[name = string("op_27764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27764_cast_fp16 = slice_by_index(begin = var_27764_begin_0, end = var_27764_end_0, end_mask = var_27764_end_mask_0, x = query_37_cast_fp16)[name = string("op_27764_cast_fp16")]; + tensor var_27768_begin_0 = const()[name = string("op_27768_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27768_end_0 = const()[name = string("op_27768_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_27768_end_mask_0 = const()[name = string("op_27768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27768_cast_fp16 = slice_by_index(begin = var_27768_begin_0, end = var_27768_end_0, end_mask = var_27768_end_mask_0, x = query_37_cast_fp16)[name = string("op_27768_cast_fp16")]; + tensor var_27772_begin_0 = const()[name = string("op_27772_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27772_end_0 = const()[name = string("op_27772_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_27772_end_mask_0 = const()[name = string("op_27772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27772_cast_fp16 = slice_by_index(begin = var_27772_begin_0, end = var_27772_end_0, end_mask = var_27772_end_mask_0, x = query_37_cast_fp16)[name = string("op_27772_cast_fp16")]; + tensor var_27776_begin_0 = const()[name = string("op_27776_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27776_end_0 = const()[name = string("op_27776_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_27776_end_mask_0 = const()[name = string("op_27776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27776_cast_fp16 = slice_by_index(begin = var_27776_begin_0, end = var_27776_end_0, end_mask = var_27776_end_mask_0, x = query_37_cast_fp16)[name = string("op_27776_cast_fp16")]; + tensor var_27780_begin_0 = const()[name = string("op_27780_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27780_end_0 = const()[name = string("op_27780_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_27780_end_mask_0 = const()[name = string("op_27780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27780_cast_fp16 = slice_by_index(begin = var_27780_begin_0, end = var_27780_end_0, end_mask = var_27780_end_mask_0, x = query_37_cast_fp16)[name = string("op_27780_cast_fp16")]; + tensor var_27784_begin_0 = const()[name = string("op_27784_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27784_end_0 = const()[name = string("op_27784_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_27784_end_mask_0 = const()[name = string("op_27784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27784_cast_fp16 = slice_by_index(begin = var_27784_begin_0, end = var_27784_end_0, end_mask = var_27784_end_mask_0, x = query_37_cast_fp16)[name = string("op_27784_cast_fp16")]; + tensor var_27788_begin_0 = const()[name = string("op_27788_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27788_end_0 = const()[name = string("op_27788_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_27788_end_mask_0 = const()[name = string("op_27788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27788_cast_fp16 = slice_by_index(begin = var_27788_begin_0, end = var_27788_end_0, end_mask = var_27788_end_mask_0, x = query_37_cast_fp16)[name = string("op_27788_cast_fp16")]; + tensor var_27792_begin_0 = const()[name = string("op_27792_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27792_end_0 = const()[name = string("op_27792_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_27792_end_mask_0 = const()[name = string("op_27792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27792_cast_fp16 = slice_by_index(begin = var_27792_begin_0, end = var_27792_end_0, end_mask = var_27792_end_mask_0, x = query_37_cast_fp16)[name = string("op_27792_cast_fp16")]; + tensor var_27796_begin_0 = const()[name = string("op_27796_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27796_end_0 = const()[name = string("op_27796_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_27796_end_mask_0 = const()[name = string("op_27796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27796_cast_fp16 = slice_by_index(begin = var_27796_begin_0, end = var_27796_end_0, end_mask = var_27796_end_mask_0, x = query_37_cast_fp16)[name = string("op_27796_cast_fp16")]; + tensor var_27800_begin_0 = const()[name = string("op_27800_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27800_end_0 = const()[name = string("op_27800_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_27800_end_mask_0 = const()[name = string("op_27800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27800_cast_fp16 = slice_by_index(begin = var_27800_begin_0, end = var_27800_end_0, end_mask = var_27800_end_mask_0, x = query_37_cast_fp16)[name = string("op_27800_cast_fp16")]; + tensor var_27804_begin_0 = const()[name = string("op_27804_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27804_end_0 = const()[name = string("op_27804_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_27804_end_mask_0 = const()[name = string("op_27804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27804_cast_fp16 = slice_by_index(begin = var_27804_begin_0, end = var_27804_end_0, end_mask = var_27804_end_mask_0, x = query_37_cast_fp16)[name = string("op_27804_cast_fp16")]; + tensor var_27808_begin_0 = const()[name = string("op_27808_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27808_end_0 = const()[name = string("op_27808_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_27808_end_mask_0 = const()[name = string("op_27808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27808_cast_fp16 = slice_by_index(begin = var_27808_begin_0, end = var_27808_end_0, end_mask = var_27808_end_mask_0, x = query_37_cast_fp16)[name = string("op_27808_cast_fp16")]; + tensor var_27812_begin_0 = const()[name = string("op_27812_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27812_end_0 = const()[name = string("op_27812_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_27812_end_mask_0 = const()[name = string("op_27812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27812_cast_fp16 = slice_by_index(begin = var_27812_begin_0, end = var_27812_end_0, end_mask = var_27812_end_mask_0, x = query_37_cast_fp16)[name = string("op_27812_cast_fp16")]; + tensor var_27816_begin_0 = const()[name = string("op_27816_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27816_end_0 = const()[name = string("op_27816_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_27816_end_mask_0 = const()[name = string("op_27816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27816_cast_fp16 = slice_by_index(begin = var_27816_begin_0, end = var_27816_end_0, end_mask = var_27816_end_mask_0, x = query_37_cast_fp16)[name = string("op_27816_cast_fp16")]; + tensor var_27820_begin_0 = const()[name = string("op_27820_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27820_end_0 = const()[name = string("op_27820_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_27820_end_mask_0 = const()[name = string("op_27820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27820_cast_fp16 = slice_by_index(begin = var_27820_begin_0, end = var_27820_end_0, end_mask = var_27820_end_mask_0, x = query_37_cast_fp16)[name = string("op_27820_cast_fp16")]; + tensor var_27824_begin_0 = const()[name = string("op_27824_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27824_end_0 = const()[name = string("op_27824_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_27824_end_mask_0 = const()[name = string("op_27824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27824_cast_fp16 = slice_by_index(begin = var_27824_begin_0, end = var_27824_end_0, end_mask = var_27824_end_mask_0, x = query_37_cast_fp16)[name = string("op_27824_cast_fp16")]; + tensor var_27833_begin_0 = const()[name = string("op_27833_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27833_end_0 = const()[name = string("op_27833_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27833_end_mask_0 = const()[name = string("op_27833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27833_cast_fp16 = slice_by_index(begin = var_27833_begin_0, end = var_27833_end_0, end_mask = var_27833_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27833_cast_fp16")]; + tensor var_27840_begin_0 = const()[name = string("op_27840_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27840_end_0 = const()[name = string("op_27840_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27840_end_mask_0 = const()[name = string("op_27840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27840_cast_fp16 = slice_by_index(begin = var_27840_begin_0, end = var_27840_end_0, end_mask = var_27840_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27840_cast_fp16")]; + tensor var_27847_begin_0 = const()[name = string("op_27847_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27847_end_0 = const()[name = string("op_27847_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27847_end_mask_0 = const()[name = string("op_27847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27847_cast_fp16 = slice_by_index(begin = var_27847_begin_0, end = var_27847_end_0, end_mask = var_27847_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27847_cast_fp16")]; + tensor var_27854_begin_0 = const()[name = string("op_27854_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27854_end_0 = const()[name = string("op_27854_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27854_end_mask_0 = const()[name = string("op_27854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27854_cast_fp16 = slice_by_index(begin = var_27854_begin_0, end = var_27854_end_0, end_mask = var_27854_end_mask_0, x = var_27748_cast_fp16)[name = string("op_27854_cast_fp16")]; + tensor var_27861_begin_0 = const()[name = string("op_27861_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27861_end_0 = const()[name = string("op_27861_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27861_end_mask_0 = const()[name = string("op_27861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27861_cast_fp16 = slice_by_index(begin = var_27861_begin_0, end = var_27861_end_0, end_mask = var_27861_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27861_cast_fp16")]; + tensor var_27868_begin_0 = const()[name = string("op_27868_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27868_end_0 = const()[name = string("op_27868_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27868_end_mask_0 = const()[name = string("op_27868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27868_cast_fp16 = slice_by_index(begin = var_27868_begin_0, end = var_27868_end_0, end_mask = var_27868_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27868_cast_fp16")]; + tensor var_27875_begin_0 = const()[name = string("op_27875_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27875_end_0 = const()[name = string("op_27875_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27875_end_mask_0 = const()[name = string("op_27875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27875_cast_fp16 = slice_by_index(begin = var_27875_begin_0, end = var_27875_end_0, end_mask = var_27875_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27875_cast_fp16")]; + tensor var_27882_begin_0 = const()[name = string("op_27882_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27882_end_0 = const()[name = string("op_27882_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27882_end_mask_0 = const()[name = string("op_27882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27882_cast_fp16 = slice_by_index(begin = var_27882_begin_0, end = var_27882_end_0, end_mask = var_27882_end_mask_0, x = var_27752_cast_fp16)[name = string("op_27882_cast_fp16")]; + tensor var_27889_begin_0 = const()[name = string("op_27889_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27889_end_0 = const()[name = string("op_27889_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27889_end_mask_0 = const()[name = string("op_27889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27889_cast_fp16 = slice_by_index(begin = var_27889_begin_0, end = var_27889_end_0, end_mask = var_27889_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27889_cast_fp16")]; + tensor var_27896_begin_0 = const()[name = string("op_27896_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27896_end_0 = const()[name = string("op_27896_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27896_end_mask_0 = const()[name = string("op_27896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27896_cast_fp16 = slice_by_index(begin = var_27896_begin_0, end = var_27896_end_0, end_mask = var_27896_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27896_cast_fp16")]; + tensor var_27903_begin_0 = const()[name = string("op_27903_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27903_end_0 = const()[name = string("op_27903_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27903_end_mask_0 = const()[name = string("op_27903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27903_cast_fp16 = slice_by_index(begin = var_27903_begin_0, end = var_27903_end_0, end_mask = var_27903_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27903_cast_fp16")]; + tensor var_27910_begin_0 = const()[name = string("op_27910_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27910_end_0 = const()[name = string("op_27910_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27910_end_mask_0 = const()[name = string("op_27910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27910_cast_fp16 = slice_by_index(begin = var_27910_begin_0, end = var_27910_end_0, end_mask = var_27910_end_mask_0, x = var_27756_cast_fp16)[name = string("op_27910_cast_fp16")]; + tensor var_27917_begin_0 = const()[name = string("op_27917_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27917_end_0 = const()[name = string("op_27917_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27917_end_mask_0 = const()[name = string("op_27917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27917_cast_fp16 = slice_by_index(begin = var_27917_begin_0, end = var_27917_end_0, end_mask = var_27917_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27917_cast_fp16")]; + tensor var_27924_begin_0 = const()[name = string("op_27924_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27924_end_0 = const()[name = string("op_27924_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27924_end_mask_0 = const()[name = string("op_27924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27924_cast_fp16 = slice_by_index(begin = var_27924_begin_0, end = var_27924_end_0, end_mask = var_27924_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27924_cast_fp16")]; + tensor var_27931_begin_0 = const()[name = string("op_27931_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27931_end_0 = const()[name = string("op_27931_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27931_end_mask_0 = const()[name = string("op_27931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27931_cast_fp16 = slice_by_index(begin = var_27931_begin_0, end = var_27931_end_0, end_mask = var_27931_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27931_cast_fp16")]; + tensor var_27938_begin_0 = const()[name = string("op_27938_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27938_end_0 = const()[name = string("op_27938_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27938_end_mask_0 = const()[name = string("op_27938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27938_cast_fp16 = slice_by_index(begin = var_27938_begin_0, end = var_27938_end_0, end_mask = var_27938_end_mask_0, x = var_27760_cast_fp16)[name = string("op_27938_cast_fp16")]; + tensor var_27945_begin_0 = const()[name = string("op_27945_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27945_end_0 = const()[name = string("op_27945_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27945_end_mask_0 = const()[name = string("op_27945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27945_cast_fp16 = slice_by_index(begin = var_27945_begin_0, end = var_27945_end_0, end_mask = var_27945_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27945_cast_fp16")]; + tensor var_27952_begin_0 = const()[name = string("op_27952_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27952_end_0 = const()[name = string("op_27952_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27952_end_mask_0 = const()[name = string("op_27952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27952_cast_fp16 = slice_by_index(begin = var_27952_begin_0, end = var_27952_end_0, end_mask = var_27952_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27952_cast_fp16")]; + tensor var_27959_begin_0 = const()[name = string("op_27959_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27959_end_0 = const()[name = string("op_27959_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27959_end_mask_0 = const()[name = string("op_27959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27959_cast_fp16 = slice_by_index(begin = var_27959_begin_0, end = var_27959_end_0, end_mask = var_27959_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27959_cast_fp16")]; + tensor var_27966_begin_0 = const()[name = string("op_27966_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27966_end_0 = const()[name = string("op_27966_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27966_end_mask_0 = const()[name = string("op_27966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27966_cast_fp16 = slice_by_index(begin = var_27966_begin_0, end = var_27966_end_0, end_mask = var_27966_end_mask_0, x = var_27764_cast_fp16)[name = string("op_27966_cast_fp16")]; + tensor var_27973_begin_0 = const()[name = string("op_27973_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27973_end_0 = const()[name = string("op_27973_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27973_end_mask_0 = const()[name = string("op_27973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27973_cast_fp16 = slice_by_index(begin = var_27973_begin_0, end = var_27973_end_0, end_mask = var_27973_end_mask_0, x = var_27768_cast_fp16)[name = string("op_27973_cast_fp16")]; + tensor var_27980_begin_0 = const()[name = string("op_27980_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27980_end_0 = const()[name = string("op_27980_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27980_end_mask_0 = const()[name = string("op_27980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27980_cast_fp16 = slice_by_index(begin = var_27980_begin_0, end = var_27980_end_0, end_mask = var_27980_end_mask_0, x = var_27768_cast_fp16)[name = string("op_27980_cast_fp16")]; + tensor var_27987_begin_0 = const()[name = string("op_27987_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27987_end_0 = const()[name = string("op_27987_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27987_end_mask_0 = const()[name = string("op_27987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27987_cast_fp16 = slice_by_index(begin = var_27987_begin_0, end = var_27987_end_0, end_mask = var_27987_end_mask_0, x = var_27768_cast_fp16)[name = string("op_27987_cast_fp16")]; + tensor var_27994_begin_0 = const()[name = string("op_27994_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27994_end_0 = const()[name = string("op_27994_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27994_end_mask_0 = const()[name = string("op_27994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27994_cast_fp16 = slice_by_index(begin = var_27994_begin_0, end = var_27994_end_0, end_mask = var_27994_end_mask_0, x = var_27768_cast_fp16)[name = string("op_27994_cast_fp16")]; + tensor var_28001_begin_0 = const()[name = string("op_28001_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28001_end_0 = const()[name = string("op_28001_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28001_end_mask_0 = const()[name = string("op_28001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28001_cast_fp16 = slice_by_index(begin = var_28001_begin_0, end = var_28001_end_0, end_mask = var_28001_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28001_cast_fp16")]; + tensor var_28008_begin_0 = const()[name = string("op_28008_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28008_end_0 = const()[name = string("op_28008_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28008_end_mask_0 = const()[name = string("op_28008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28008_cast_fp16 = slice_by_index(begin = var_28008_begin_0, end = var_28008_end_0, end_mask = var_28008_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28008_cast_fp16")]; + tensor var_28015_begin_0 = const()[name = string("op_28015_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28015_end_0 = const()[name = string("op_28015_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28015_end_mask_0 = const()[name = string("op_28015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28015_cast_fp16 = slice_by_index(begin = var_28015_begin_0, end = var_28015_end_0, end_mask = var_28015_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28015_cast_fp16")]; + tensor var_28022_begin_0 = const()[name = string("op_28022_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28022_end_0 = const()[name = string("op_28022_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28022_end_mask_0 = const()[name = string("op_28022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28022_cast_fp16 = slice_by_index(begin = var_28022_begin_0, end = var_28022_end_0, end_mask = var_28022_end_mask_0, x = var_27772_cast_fp16)[name = string("op_28022_cast_fp16")]; + tensor var_28029_begin_0 = const()[name = string("op_28029_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28029_end_0 = const()[name = string("op_28029_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28029_end_mask_0 = const()[name = string("op_28029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28029_cast_fp16 = slice_by_index(begin = var_28029_begin_0, end = var_28029_end_0, end_mask = var_28029_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28029_cast_fp16")]; + tensor var_28036_begin_0 = const()[name = string("op_28036_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28036_end_0 = const()[name = string("op_28036_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28036_end_mask_0 = const()[name = string("op_28036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28036_cast_fp16 = slice_by_index(begin = var_28036_begin_0, end = var_28036_end_0, end_mask = var_28036_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28036_cast_fp16")]; + tensor var_28043_begin_0 = const()[name = string("op_28043_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28043_end_0 = const()[name = string("op_28043_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28043_end_mask_0 = const()[name = string("op_28043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28043_cast_fp16 = slice_by_index(begin = var_28043_begin_0, end = var_28043_end_0, end_mask = var_28043_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28043_cast_fp16")]; + tensor var_28050_begin_0 = const()[name = string("op_28050_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28050_end_0 = const()[name = string("op_28050_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28050_end_mask_0 = const()[name = string("op_28050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28050_cast_fp16 = slice_by_index(begin = var_28050_begin_0, end = var_28050_end_0, end_mask = var_28050_end_mask_0, x = var_27776_cast_fp16)[name = string("op_28050_cast_fp16")]; + tensor var_28057_begin_0 = const()[name = string("op_28057_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28057_end_0 = const()[name = string("op_28057_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28057_end_mask_0 = const()[name = string("op_28057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28057_cast_fp16 = slice_by_index(begin = var_28057_begin_0, end = var_28057_end_0, end_mask = var_28057_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28057_cast_fp16")]; + tensor var_28064_begin_0 = const()[name = string("op_28064_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28064_end_0 = const()[name = string("op_28064_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28064_end_mask_0 = const()[name = string("op_28064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28064_cast_fp16 = slice_by_index(begin = var_28064_begin_0, end = var_28064_end_0, end_mask = var_28064_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28064_cast_fp16")]; + tensor var_28071_begin_0 = const()[name = string("op_28071_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28071_end_0 = const()[name = string("op_28071_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28071_end_mask_0 = const()[name = string("op_28071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28071_cast_fp16 = slice_by_index(begin = var_28071_begin_0, end = var_28071_end_0, end_mask = var_28071_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28071_cast_fp16")]; + tensor var_28078_begin_0 = const()[name = string("op_28078_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28078_end_0 = const()[name = string("op_28078_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28078_end_mask_0 = const()[name = string("op_28078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28078_cast_fp16 = slice_by_index(begin = var_28078_begin_0, end = var_28078_end_0, end_mask = var_28078_end_mask_0, x = var_27780_cast_fp16)[name = string("op_28078_cast_fp16")]; + tensor var_28085_begin_0 = const()[name = string("op_28085_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28085_end_0 = const()[name = string("op_28085_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28085_end_mask_0 = const()[name = string("op_28085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28085_cast_fp16 = slice_by_index(begin = var_28085_begin_0, end = var_28085_end_0, end_mask = var_28085_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28085_cast_fp16")]; + tensor var_28092_begin_0 = const()[name = string("op_28092_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28092_end_0 = const()[name = string("op_28092_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28092_end_mask_0 = const()[name = string("op_28092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28092_cast_fp16 = slice_by_index(begin = var_28092_begin_0, end = var_28092_end_0, end_mask = var_28092_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28092_cast_fp16")]; + tensor var_28099_begin_0 = const()[name = string("op_28099_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28099_end_0 = const()[name = string("op_28099_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28099_end_mask_0 = const()[name = string("op_28099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28099_cast_fp16 = slice_by_index(begin = var_28099_begin_0, end = var_28099_end_0, end_mask = var_28099_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28099_cast_fp16")]; + tensor var_28106_begin_0 = const()[name = string("op_28106_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28106_end_0 = const()[name = string("op_28106_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28106_end_mask_0 = const()[name = string("op_28106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28106_cast_fp16 = slice_by_index(begin = var_28106_begin_0, end = var_28106_end_0, end_mask = var_28106_end_mask_0, x = var_27784_cast_fp16)[name = string("op_28106_cast_fp16")]; + tensor var_28113_begin_0 = const()[name = string("op_28113_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28113_end_0 = const()[name = string("op_28113_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28113_end_mask_0 = const()[name = string("op_28113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28113_cast_fp16 = slice_by_index(begin = var_28113_begin_0, end = var_28113_end_0, end_mask = var_28113_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28113_cast_fp16")]; + tensor var_28120_begin_0 = const()[name = string("op_28120_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28120_end_0 = const()[name = string("op_28120_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28120_end_mask_0 = const()[name = string("op_28120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28120_cast_fp16 = slice_by_index(begin = var_28120_begin_0, end = var_28120_end_0, end_mask = var_28120_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28120_cast_fp16")]; + tensor var_28127_begin_0 = const()[name = string("op_28127_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28127_end_0 = const()[name = string("op_28127_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28127_end_mask_0 = const()[name = string("op_28127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28127_cast_fp16 = slice_by_index(begin = var_28127_begin_0, end = var_28127_end_0, end_mask = var_28127_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28127_cast_fp16")]; + tensor var_28134_begin_0 = const()[name = string("op_28134_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28134_end_0 = const()[name = string("op_28134_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28134_end_mask_0 = const()[name = string("op_28134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28134_cast_fp16 = slice_by_index(begin = var_28134_begin_0, end = var_28134_end_0, end_mask = var_28134_end_mask_0, x = var_27788_cast_fp16)[name = string("op_28134_cast_fp16")]; + tensor var_28141_begin_0 = const()[name = string("op_28141_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28141_end_0 = const()[name = string("op_28141_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28141_end_mask_0 = const()[name = string("op_28141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28141_cast_fp16 = slice_by_index(begin = var_28141_begin_0, end = var_28141_end_0, end_mask = var_28141_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28141_cast_fp16")]; + tensor var_28148_begin_0 = const()[name = string("op_28148_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28148_end_0 = const()[name = string("op_28148_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28148_end_mask_0 = const()[name = string("op_28148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28148_cast_fp16 = slice_by_index(begin = var_28148_begin_0, end = var_28148_end_0, end_mask = var_28148_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28148_cast_fp16")]; + tensor var_28155_begin_0 = const()[name = string("op_28155_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28155_end_0 = const()[name = string("op_28155_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28155_end_mask_0 = const()[name = string("op_28155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28155_cast_fp16 = slice_by_index(begin = var_28155_begin_0, end = var_28155_end_0, end_mask = var_28155_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28155_cast_fp16")]; + tensor var_28162_begin_0 = const()[name = string("op_28162_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28162_end_0 = const()[name = string("op_28162_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28162_end_mask_0 = const()[name = string("op_28162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28162_cast_fp16 = slice_by_index(begin = var_28162_begin_0, end = var_28162_end_0, end_mask = var_28162_end_mask_0, x = var_27792_cast_fp16)[name = string("op_28162_cast_fp16")]; + tensor var_28169_begin_0 = const()[name = string("op_28169_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28169_end_0 = const()[name = string("op_28169_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28169_end_mask_0 = const()[name = string("op_28169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28169_cast_fp16 = slice_by_index(begin = var_28169_begin_0, end = var_28169_end_0, end_mask = var_28169_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28169_cast_fp16")]; + tensor var_28176_begin_0 = const()[name = string("op_28176_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28176_end_0 = const()[name = string("op_28176_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28176_end_mask_0 = const()[name = string("op_28176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28176_cast_fp16 = slice_by_index(begin = var_28176_begin_0, end = var_28176_end_0, end_mask = var_28176_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28176_cast_fp16")]; + tensor var_28183_begin_0 = const()[name = string("op_28183_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28183_end_0 = const()[name = string("op_28183_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28183_end_mask_0 = const()[name = string("op_28183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28183_cast_fp16 = slice_by_index(begin = var_28183_begin_0, end = var_28183_end_0, end_mask = var_28183_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28183_cast_fp16")]; + tensor var_28190_begin_0 = const()[name = string("op_28190_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28190_end_0 = const()[name = string("op_28190_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28190_end_mask_0 = const()[name = string("op_28190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28190_cast_fp16 = slice_by_index(begin = var_28190_begin_0, end = var_28190_end_0, end_mask = var_28190_end_mask_0, x = var_27796_cast_fp16)[name = string("op_28190_cast_fp16")]; + tensor var_28197_begin_0 = const()[name = string("op_28197_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28197_end_0 = const()[name = string("op_28197_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28197_end_mask_0 = const()[name = string("op_28197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28197_cast_fp16 = slice_by_index(begin = var_28197_begin_0, end = var_28197_end_0, end_mask = var_28197_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28197_cast_fp16")]; + tensor var_28204_begin_0 = const()[name = string("op_28204_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28204_end_0 = const()[name = string("op_28204_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28204_end_mask_0 = const()[name = string("op_28204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28204_cast_fp16 = slice_by_index(begin = var_28204_begin_0, end = var_28204_end_0, end_mask = var_28204_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28204_cast_fp16")]; + tensor var_28211_begin_0 = const()[name = string("op_28211_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28211_end_0 = const()[name = string("op_28211_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28211_end_mask_0 = const()[name = string("op_28211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28211_cast_fp16 = slice_by_index(begin = var_28211_begin_0, end = var_28211_end_0, end_mask = var_28211_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28211_cast_fp16")]; + tensor var_28218_begin_0 = const()[name = string("op_28218_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28218_end_0 = const()[name = string("op_28218_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28218_end_mask_0 = const()[name = string("op_28218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28218_cast_fp16 = slice_by_index(begin = var_28218_begin_0, end = var_28218_end_0, end_mask = var_28218_end_mask_0, x = var_27800_cast_fp16)[name = string("op_28218_cast_fp16")]; + tensor var_28225_begin_0 = const()[name = string("op_28225_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28225_end_0 = const()[name = string("op_28225_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28225_end_mask_0 = const()[name = string("op_28225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28225_cast_fp16 = slice_by_index(begin = var_28225_begin_0, end = var_28225_end_0, end_mask = var_28225_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28225_cast_fp16")]; + tensor var_28232_begin_0 = const()[name = string("op_28232_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28232_end_0 = const()[name = string("op_28232_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28232_end_mask_0 = const()[name = string("op_28232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28232_cast_fp16 = slice_by_index(begin = var_28232_begin_0, end = var_28232_end_0, end_mask = var_28232_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28232_cast_fp16")]; + tensor var_28239_begin_0 = const()[name = string("op_28239_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28239_end_0 = const()[name = string("op_28239_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28239_end_mask_0 = const()[name = string("op_28239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28239_cast_fp16 = slice_by_index(begin = var_28239_begin_0, end = var_28239_end_0, end_mask = var_28239_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28239_cast_fp16")]; + tensor var_28246_begin_0 = const()[name = string("op_28246_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28246_end_0 = const()[name = string("op_28246_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28246_end_mask_0 = const()[name = string("op_28246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28246_cast_fp16 = slice_by_index(begin = var_28246_begin_0, end = var_28246_end_0, end_mask = var_28246_end_mask_0, x = var_27804_cast_fp16)[name = string("op_28246_cast_fp16")]; + tensor var_28253_begin_0 = const()[name = string("op_28253_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28253_end_0 = const()[name = string("op_28253_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28253_end_mask_0 = const()[name = string("op_28253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28253_cast_fp16 = slice_by_index(begin = var_28253_begin_0, end = var_28253_end_0, end_mask = var_28253_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28253_cast_fp16")]; + tensor var_28260_begin_0 = const()[name = string("op_28260_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28260_end_0 = const()[name = string("op_28260_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28260_end_mask_0 = const()[name = string("op_28260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28260_cast_fp16 = slice_by_index(begin = var_28260_begin_0, end = var_28260_end_0, end_mask = var_28260_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28260_cast_fp16")]; + tensor var_28267_begin_0 = const()[name = string("op_28267_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28267_end_0 = const()[name = string("op_28267_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28267_end_mask_0 = const()[name = string("op_28267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28267_cast_fp16 = slice_by_index(begin = var_28267_begin_0, end = var_28267_end_0, end_mask = var_28267_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28267_cast_fp16")]; + tensor var_28274_begin_0 = const()[name = string("op_28274_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28274_end_0 = const()[name = string("op_28274_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28274_end_mask_0 = const()[name = string("op_28274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28274_cast_fp16 = slice_by_index(begin = var_28274_begin_0, end = var_28274_end_0, end_mask = var_28274_end_mask_0, x = var_27808_cast_fp16)[name = string("op_28274_cast_fp16")]; + tensor var_28281_begin_0 = const()[name = string("op_28281_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28281_end_0 = const()[name = string("op_28281_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28281_end_mask_0 = const()[name = string("op_28281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28281_cast_fp16 = slice_by_index(begin = var_28281_begin_0, end = var_28281_end_0, end_mask = var_28281_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28281_cast_fp16")]; + tensor var_28288_begin_0 = const()[name = string("op_28288_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28288_end_0 = const()[name = string("op_28288_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28288_end_mask_0 = const()[name = string("op_28288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28288_cast_fp16 = slice_by_index(begin = var_28288_begin_0, end = var_28288_end_0, end_mask = var_28288_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28288_cast_fp16")]; + tensor var_28295_begin_0 = const()[name = string("op_28295_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28295_end_0 = const()[name = string("op_28295_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28295_end_mask_0 = const()[name = string("op_28295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28295_cast_fp16 = slice_by_index(begin = var_28295_begin_0, end = var_28295_end_0, end_mask = var_28295_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28295_cast_fp16")]; + tensor var_28302_begin_0 = const()[name = string("op_28302_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28302_end_0 = const()[name = string("op_28302_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28302_end_mask_0 = const()[name = string("op_28302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28302_cast_fp16 = slice_by_index(begin = var_28302_begin_0, end = var_28302_end_0, end_mask = var_28302_end_mask_0, x = var_27812_cast_fp16)[name = string("op_28302_cast_fp16")]; + tensor var_28309_begin_0 = const()[name = string("op_28309_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28309_end_0 = const()[name = string("op_28309_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28309_end_mask_0 = const()[name = string("op_28309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28309_cast_fp16 = slice_by_index(begin = var_28309_begin_0, end = var_28309_end_0, end_mask = var_28309_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28309_cast_fp16")]; + tensor var_28316_begin_0 = const()[name = string("op_28316_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28316_end_0 = const()[name = string("op_28316_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28316_end_mask_0 = const()[name = string("op_28316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28316_cast_fp16 = slice_by_index(begin = var_28316_begin_0, end = var_28316_end_0, end_mask = var_28316_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28316_cast_fp16")]; + tensor var_28323_begin_0 = const()[name = string("op_28323_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28323_end_0 = const()[name = string("op_28323_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28323_end_mask_0 = const()[name = string("op_28323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28323_cast_fp16 = slice_by_index(begin = var_28323_begin_0, end = var_28323_end_0, end_mask = var_28323_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28323_cast_fp16")]; + tensor var_28330_begin_0 = const()[name = string("op_28330_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28330_end_0 = const()[name = string("op_28330_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28330_end_mask_0 = const()[name = string("op_28330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28330_cast_fp16 = slice_by_index(begin = var_28330_begin_0, end = var_28330_end_0, end_mask = var_28330_end_mask_0, x = var_27816_cast_fp16)[name = string("op_28330_cast_fp16")]; + tensor var_28337_begin_0 = const()[name = string("op_28337_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28337_end_0 = const()[name = string("op_28337_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28337_end_mask_0 = const()[name = string("op_28337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28337_cast_fp16 = slice_by_index(begin = var_28337_begin_0, end = var_28337_end_0, end_mask = var_28337_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28337_cast_fp16")]; + tensor var_28344_begin_0 = const()[name = string("op_28344_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28344_end_0 = const()[name = string("op_28344_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28344_end_mask_0 = const()[name = string("op_28344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28344_cast_fp16 = slice_by_index(begin = var_28344_begin_0, end = var_28344_end_0, end_mask = var_28344_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28344_cast_fp16")]; + tensor var_28351_begin_0 = const()[name = string("op_28351_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28351_end_0 = const()[name = string("op_28351_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28351_end_mask_0 = const()[name = string("op_28351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28351_cast_fp16 = slice_by_index(begin = var_28351_begin_0, end = var_28351_end_0, end_mask = var_28351_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28351_cast_fp16")]; + tensor var_28358_begin_0 = const()[name = string("op_28358_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28358_end_0 = const()[name = string("op_28358_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28358_end_mask_0 = const()[name = string("op_28358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28358_cast_fp16 = slice_by_index(begin = var_28358_begin_0, end = var_28358_end_0, end_mask = var_28358_end_mask_0, x = var_27820_cast_fp16)[name = string("op_28358_cast_fp16")]; + tensor var_28365_begin_0 = const()[name = string("op_28365_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28365_end_0 = const()[name = string("op_28365_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28365_end_mask_0 = const()[name = string("op_28365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28365_cast_fp16 = slice_by_index(begin = var_28365_begin_0, end = var_28365_end_0, end_mask = var_28365_end_mask_0, x = var_27824_cast_fp16)[name = string("op_28365_cast_fp16")]; + tensor var_28372_begin_0 = const()[name = string("op_28372_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28372_end_0 = const()[name = string("op_28372_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28372_end_mask_0 = const()[name = string("op_28372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28372_cast_fp16 = slice_by_index(begin = var_28372_begin_0, end = var_28372_end_0, end_mask = var_28372_end_mask_0, x = var_27824_cast_fp16)[name = string("op_28372_cast_fp16")]; + tensor var_28379_begin_0 = const()[name = string("op_28379_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28379_end_0 = const()[name = string("op_28379_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28379_end_mask_0 = const()[name = string("op_28379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28379_cast_fp16 = slice_by_index(begin = var_28379_begin_0, end = var_28379_end_0, end_mask = var_28379_end_mask_0, x = var_27824_cast_fp16)[name = string("op_28379_cast_fp16")]; + tensor var_28386_begin_0 = const()[name = string("op_28386_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28386_end_0 = const()[name = string("op_28386_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28386_end_mask_0 = const()[name = string("op_28386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28386_cast_fp16 = slice_by_index(begin = var_28386_begin_0, end = var_28386_end_0, end_mask = var_28386_end_mask_0, x = var_27824_cast_fp16)[name = string("op_28386_cast_fp16")]; + tensor k_37_perm_0 = const()[name = string("k_37_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28391_begin_0 = const()[name = string("op_28391_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28391_end_0 = const()[name = string("op_28391_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_28391_end_mask_0 = const()[name = string("op_28391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_37_cast_fp16 = transpose(perm = k_37_perm_0, x = key_37_cast_fp16)[name = string("transpose_13")]; + tensor var_28391_cast_fp16 = slice_by_index(begin = var_28391_begin_0, end = var_28391_end_0, end_mask = var_28391_end_mask_0, x = k_37_cast_fp16)[name = string("op_28391_cast_fp16")]; + tensor var_28395_begin_0 = const()[name = string("op_28395_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28395_end_0 = const()[name = string("op_28395_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_28395_end_mask_0 = const()[name = string("op_28395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28395_cast_fp16 = slice_by_index(begin = var_28395_begin_0, end = var_28395_end_0, end_mask = var_28395_end_mask_0, x = k_37_cast_fp16)[name = string("op_28395_cast_fp16")]; + tensor var_28399_begin_0 = const()[name = string("op_28399_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28399_end_0 = const()[name = string("op_28399_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_28399_end_mask_0 = const()[name = string("op_28399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28399_cast_fp16 = slice_by_index(begin = var_28399_begin_0, end = var_28399_end_0, end_mask = var_28399_end_mask_0, x = k_37_cast_fp16)[name = string("op_28399_cast_fp16")]; + tensor var_28403_begin_0 = const()[name = string("op_28403_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28403_end_0 = const()[name = string("op_28403_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_28403_end_mask_0 = const()[name = string("op_28403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28403_cast_fp16 = slice_by_index(begin = var_28403_begin_0, end = var_28403_end_0, end_mask = var_28403_end_mask_0, x = k_37_cast_fp16)[name = string("op_28403_cast_fp16")]; + tensor var_28407_begin_0 = const()[name = string("op_28407_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28407_end_0 = const()[name = string("op_28407_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_28407_end_mask_0 = const()[name = string("op_28407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28407_cast_fp16 = slice_by_index(begin = var_28407_begin_0, end = var_28407_end_0, end_mask = var_28407_end_mask_0, x = k_37_cast_fp16)[name = string("op_28407_cast_fp16")]; + tensor var_28411_begin_0 = const()[name = string("op_28411_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28411_end_0 = const()[name = string("op_28411_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_28411_end_mask_0 = const()[name = string("op_28411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28411_cast_fp16 = slice_by_index(begin = var_28411_begin_0, end = var_28411_end_0, end_mask = var_28411_end_mask_0, x = k_37_cast_fp16)[name = string("op_28411_cast_fp16")]; + tensor var_28415_begin_0 = const()[name = string("op_28415_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28415_end_0 = const()[name = string("op_28415_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_28415_end_mask_0 = const()[name = string("op_28415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28415_cast_fp16 = slice_by_index(begin = var_28415_begin_0, end = var_28415_end_0, end_mask = var_28415_end_mask_0, x = k_37_cast_fp16)[name = string("op_28415_cast_fp16")]; + tensor var_28419_begin_0 = const()[name = string("op_28419_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28419_end_0 = const()[name = string("op_28419_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_28419_end_mask_0 = const()[name = string("op_28419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28419_cast_fp16 = slice_by_index(begin = var_28419_begin_0, end = var_28419_end_0, end_mask = var_28419_end_mask_0, x = k_37_cast_fp16)[name = string("op_28419_cast_fp16")]; + tensor var_28423_begin_0 = const()[name = string("op_28423_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28423_end_0 = const()[name = string("op_28423_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_28423_end_mask_0 = const()[name = string("op_28423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28423_cast_fp16 = slice_by_index(begin = var_28423_begin_0, end = var_28423_end_0, end_mask = var_28423_end_mask_0, x = k_37_cast_fp16)[name = string("op_28423_cast_fp16")]; + tensor var_28427_begin_0 = const()[name = string("op_28427_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28427_end_0 = const()[name = string("op_28427_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_28427_end_mask_0 = const()[name = string("op_28427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28427_cast_fp16 = slice_by_index(begin = var_28427_begin_0, end = var_28427_end_0, end_mask = var_28427_end_mask_0, x = k_37_cast_fp16)[name = string("op_28427_cast_fp16")]; + tensor var_28431_begin_0 = const()[name = string("op_28431_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28431_end_0 = const()[name = string("op_28431_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_28431_end_mask_0 = const()[name = string("op_28431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28431_cast_fp16 = slice_by_index(begin = var_28431_begin_0, end = var_28431_end_0, end_mask = var_28431_end_mask_0, x = k_37_cast_fp16)[name = string("op_28431_cast_fp16")]; + tensor var_28435_begin_0 = const()[name = string("op_28435_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28435_end_0 = const()[name = string("op_28435_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_28435_end_mask_0 = const()[name = string("op_28435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28435_cast_fp16 = slice_by_index(begin = var_28435_begin_0, end = var_28435_end_0, end_mask = var_28435_end_mask_0, x = k_37_cast_fp16)[name = string("op_28435_cast_fp16")]; + tensor var_28439_begin_0 = const()[name = string("op_28439_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28439_end_0 = const()[name = string("op_28439_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_28439_end_mask_0 = const()[name = string("op_28439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28439_cast_fp16 = slice_by_index(begin = var_28439_begin_0, end = var_28439_end_0, end_mask = var_28439_end_mask_0, x = k_37_cast_fp16)[name = string("op_28439_cast_fp16")]; + tensor var_28443_begin_0 = const()[name = string("op_28443_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28443_end_0 = const()[name = string("op_28443_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_28443_end_mask_0 = const()[name = string("op_28443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28443_cast_fp16 = slice_by_index(begin = var_28443_begin_0, end = var_28443_end_0, end_mask = var_28443_end_mask_0, x = k_37_cast_fp16)[name = string("op_28443_cast_fp16")]; + tensor var_28447_begin_0 = const()[name = string("op_28447_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28447_end_0 = const()[name = string("op_28447_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_28447_end_mask_0 = const()[name = string("op_28447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28447_cast_fp16 = slice_by_index(begin = var_28447_begin_0, end = var_28447_end_0, end_mask = var_28447_end_mask_0, x = k_37_cast_fp16)[name = string("op_28447_cast_fp16")]; + tensor var_28451_begin_0 = const()[name = string("op_28451_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28451_end_0 = const()[name = string("op_28451_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_28451_end_mask_0 = const()[name = string("op_28451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28451_cast_fp16 = slice_by_index(begin = var_28451_begin_0, end = var_28451_end_0, end_mask = var_28451_end_mask_0, x = k_37_cast_fp16)[name = string("op_28451_cast_fp16")]; + tensor var_28455_begin_0 = const()[name = string("op_28455_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28455_end_0 = const()[name = string("op_28455_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_28455_end_mask_0 = const()[name = string("op_28455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28455_cast_fp16 = slice_by_index(begin = var_28455_begin_0, end = var_28455_end_0, end_mask = var_28455_end_mask_0, x = k_37_cast_fp16)[name = string("op_28455_cast_fp16")]; + tensor var_28459_begin_0 = const()[name = string("op_28459_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28459_end_0 = const()[name = string("op_28459_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_28459_end_mask_0 = const()[name = string("op_28459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28459_cast_fp16 = slice_by_index(begin = var_28459_begin_0, end = var_28459_end_0, end_mask = var_28459_end_mask_0, x = k_37_cast_fp16)[name = string("op_28459_cast_fp16")]; + tensor var_28463_begin_0 = const()[name = string("op_28463_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28463_end_0 = const()[name = string("op_28463_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_28463_end_mask_0 = const()[name = string("op_28463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28463_cast_fp16 = slice_by_index(begin = var_28463_begin_0, end = var_28463_end_0, end_mask = var_28463_end_mask_0, x = k_37_cast_fp16)[name = string("op_28463_cast_fp16")]; + tensor var_28467_begin_0 = const()[name = string("op_28467_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28467_end_0 = const()[name = string("op_28467_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_28467_end_mask_0 = const()[name = string("op_28467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28467_cast_fp16 = slice_by_index(begin = var_28467_begin_0, end = var_28467_end_0, end_mask = var_28467_end_mask_0, x = k_37_cast_fp16)[name = string("op_28467_cast_fp16")]; + tensor var_28469_begin_0 = const()[name = string("op_28469_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28469_end_0 = const()[name = string("op_28469_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28469_end_mask_0 = const()[name = string("op_28469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28469_cast_fp16 = slice_by_index(begin = var_28469_begin_0, end = var_28469_end_0, end_mask = var_28469_end_mask_0, x = value_37_cast_fp16)[name = string("op_28469_cast_fp16")]; + tensor var_28473_begin_0 = const()[name = string("op_28473_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28473_end_0 = const()[name = string("op_28473_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_28473_end_mask_0 = const()[name = string("op_28473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28473_cast_fp16 = slice_by_index(begin = var_28473_begin_0, end = var_28473_end_0, end_mask = var_28473_end_mask_0, x = value_37_cast_fp16)[name = string("op_28473_cast_fp16")]; + tensor var_28477_begin_0 = const()[name = string("op_28477_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28477_end_0 = const()[name = string("op_28477_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_28477_end_mask_0 = const()[name = string("op_28477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28477_cast_fp16 = slice_by_index(begin = var_28477_begin_0, end = var_28477_end_0, end_mask = var_28477_end_mask_0, x = value_37_cast_fp16)[name = string("op_28477_cast_fp16")]; + tensor var_28481_begin_0 = const()[name = string("op_28481_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28481_end_0 = const()[name = string("op_28481_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_28481_end_mask_0 = const()[name = string("op_28481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28481_cast_fp16 = slice_by_index(begin = var_28481_begin_0, end = var_28481_end_0, end_mask = var_28481_end_mask_0, x = value_37_cast_fp16)[name = string("op_28481_cast_fp16")]; + tensor var_28485_begin_0 = const()[name = string("op_28485_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28485_end_0 = const()[name = string("op_28485_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_28485_end_mask_0 = const()[name = string("op_28485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28485_cast_fp16 = slice_by_index(begin = var_28485_begin_0, end = var_28485_end_0, end_mask = var_28485_end_mask_0, x = value_37_cast_fp16)[name = string("op_28485_cast_fp16")]; + tensor var_28489_begin_0 = const()[name = string("op_28489_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28489_end_0 = const()[name = string("op_28489_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_28489_end_mask_0 = const()[name = string("op_28489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28489_cast_fp16 = slice_by_index(begin = var_28489_begin_0, end = var_28489_end_0, end_mask = var_28489_end_mask_0, x = value_37_cast_fp16)[name = string("op_28489_cast_fp16")]; + tensor var_28493_begin_0 = const()[name = string("op_28493_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28493_end_0 = const()[name = string("op_28493_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_28493_end_mask_0 = const()[name = string("op_28493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28493_cast_fp16 = slice_by_index(begin = var_28493_begin_0, end = var_28493_end_0, end_mask = var_28493_end_mask_0, x = value_37_cast_fp16)[name = string("op_28493_cast_fp16")]; + tensor var_28497_begin_0 = const()[name = string("op_28497_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28497_end_0 = const()[name = string("op_28497_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_28497_end_mask_0 = const()[name = string("op_28497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28497_cast_fp16 = slice_by_index(begin = var_28497_begin_0, end = var_28497_end_0, end_mask = var_28497_end_mask_0, x = value_37_cast_fp16)[name = string("op_28497_cast_fp16")]; + tensor var_28501_begin_0 = const()[name = string("op_28501_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28501_end_0 = const()[name = string("op_28501_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_28501_end_mask_0 = const()[name = string("op_28501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28501_cast_fp16 = slice_by_index(begin = var_28501_begin_0, end = var_28501_end_0, end_mask = var_28501_end_mask_0, x = value_37_cast_fp16)[name = string("op_28501_cast_fp16")]; + tensor var_28505_begin_0 = const()[name = string("op_28505_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28505_end_0 = const()[name = string("op_28505_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_28505_end_mask_0 = const()[name = string("op_28505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28505_cast_fp16 = slice_by_index(begin = var_28505_begin_0, end = var_28505_end_0, end_mask = var_28505_end_mask_0, x = value_37_cast_fp16)[name = string("op_28505_cast_fp16")]; + tensor var_28509_begin_0 = const()[name = string("op_28509_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28509_end_0 = const()[name = string("op_28509_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_28509_end_mask_0 = const()[name = string("op_28509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28509_cast_fp16 = slice_by_index(begin = var_28509_begin_0, end = var_28509_end_0, end_mask = var_28509_end_mask_0, x = value_37_cast_fp16)[name = string("op_28509_cast_fp16")]; + tensor var_28513_begin_0 = const()[name = string("op_28513_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28513_end_0 = const()[name = string("op_28513_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_28513_end_mask_0 = const()[name = string("op_28513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28513_cast_fp16 = slice_by_index(begin = var_28513_begin_0, end = var_28513_end_0, end_mask = var_28513_end_mask_0, x = value_37_cast_fp16)[name = string("op_28513_cast_fp16")]; + tensor var_28517_begin_0 = const()[name = string("op_28517_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28517_end_0 = const()[name = string("op_28517_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_28517_end_mask_0 = const()[name = string("op_28517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28517_cast_fp16 = slice_by_index(begin = var_28517_begin_0, end = var_28517_end_0, end_mask = var_28517_end_mask_0, x = value_37_cast_fp16)[name = string("op_28517_cast_fp16")]; + tensor var_28521_begin_0 = const()[name = string("op_28521_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28521_end_0 = const()[name = string("op_28521_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_28521_end_mask_0 = const()[name = string("op_28521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28521_cast_fp16 = slice_by_index(begin = var_28521_begin_0, end = var_28521_end_0, end_mask = var_28521_end_mask_0, x = value_37_cast_fp16)[name = string("op_28521_cast_fp16")]; + tensor var_28525_begin_0 = const()[name = string("op_28525_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28525_end_0 = const()[name = string("op_28525_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_28525_end_mask_0 = const()[name = string("op_28525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28525_cast_fp16 = slice_by_index(begin = var_28525_begin_0, end = var_28525_end_0, end_mask = var_28525_end_mask_0, x = value_37_cast_fp16)[name = string("op_28525_cast_fp16")]; + tensor var_28529_begin_0 = const()[name = string("op_28529_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28529_end_0 = const()[name = string("op_28529_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_28529_end_mask_0 = const()[name = string("op_28529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28529_cast_fp16 = slice_by_index(begin = var_28529_begin_0, end = var_28529_end_0, end_mask = var_28529_end_mask_0, x = value_37_cast_fp16)[name = string("op_28529_cast_fp16")]; + tensor var_28533_begin_0 = const()[name = string("op_28533_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28533_end_0 = const()[name = string("op_28533_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_28533_end_mask_0 = const()[name = string("op_28533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28533_cast_fp16 = slice_by_index(begin = var_28533_begin_0, end = var_28533_end_0, end_mask = var_28533_end_mask_0, x = value_37_cast_fp16)[name = string("op_28533_cast_fp16")]; + tensor var_28537_begin_0 = const()[name = string("op_28537_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28537_end_0 = const()[name = string("op_28537_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_28537_end_mask_0 = const()[name = string("op_28537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28537_cast_fp16 = slice_by_index(begin = var_28537_begin_0, end = var_28537_end_0, end_mask = var_28537_end_mask_0, x = value_37_cast_fp16)[name = string("op_28537_cast_fp16")]; + tensor var_28541_begin_0 = const()[name = string("op_28541_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28541_end_0 = const()[name = string("op_28541_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_28541_end_mask_0 = const()[name = string("op_28541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28541_cast_fp16 = slice_by_index(begin = var_28541_begin_0, end = var_28541_end_0, end_mask = var_28541_end_mask_0, x = value_37_cast_fp16)[name = string("op_28541_cast_fp16")]; + tensor var_28545_begin_0 = const()[name = string("op_28545_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28545_end_0 = const()[name = string("op_28545_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_28545_end_mask_0 = const()[name = string("op_28545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28545_cast_fp16 = slice_by_index(begin = var_28545_begin_0, end = var_28545_end_0, end_mask = var_28545_end_mask_0, x = value_37_cast_fp16)[name = string("op_28545_cast_fp16")]; + string _SplitHeadsQ__mh_w_2881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2881_equation_0, values = (var_28391_cast_fp16, var_27833_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2881_cast_fp16")]; + string _SplitHeadsQ__mh_w_2883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2883_equation_0, values = (var_28391_cast_fp16, var_27840_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2883_cast_fp16")]; + string _SplitHeadsQ__mh_w_2885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2885_equation_0, values = (var_28391_cast_fp16, var_27847_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2885_cast_fp16")]; + string _SplitHeadsQ__mh_w_2887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2887_equation_0, values = (var_28391_cast_fp16, var_27854_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2887_cast_fp16")]; + string _SplitHeadsQ__mh_w_2889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2889_equation_0, values = (var_28395_cast_fp16, var_27861_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2889_cast_fp16")]; + string _SplitHeadsQ__mh_w_2891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2891_equation_0, values = (var_28395_cast_fp16, var_27868_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2891_cast_fp16")]; + string _SplitHeadsQ__mh_w_2893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2893_equation_0, values = (var_28395_cast_fp16, var_27875_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2893_cast_fp16")]; + string _SplitHeadsQ__mh_w_2895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2895_equation_0, values = (var_28395_cast_fp16, var_27882_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2895_cast_fp16")]; + string _SplitHeadsQ__mh_w_2897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2897_equation_0, values = (var_28399_cast_fp16, var_27889_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2897_cast_fp16")]; + string _SplitHeadsQ__mh_w_2899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2899_equation_0, values = (var_28399_cast_fp16, var_27896_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2899_cast_fp16")]; + string _SplitHeadsQ__mh_w_2901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2901_equation_0, values = (var_28399_cast_fp16, var_27903_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2901_cast_fp16")]; + string _SplitHeadsQ__mh_w_2903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2903_equation_0, values = (var_28399_cast_fp16, var_27910_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2903_cast_fp16")]; + string _SplitHeadsQ__mh_w_2905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2905_equation_0, values = (var_28403_cast_fp16, var_27917_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2905_cast_fp16")]; + string _SplitHeadsQ__mh_w_2907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2907_equation_0, values = (var_28403_cast_fp16, var_27924_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2907_cast_fp16")]; + string _SplitHeadsQ__mh_w_2909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2909_equation_0, values = (var_28403_cast_fp16, var_27931_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2909_cast_fp16")]; + string _SplitHeadsQ__mh_w_2911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2911_equation_0, values = (var_28403_cast_fp16, var_27938_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2911_cast_fp16")]; + string _SplitHeadsQ__mh_w_2913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2913_equation_0, values = (var_28407_cast_fp16, var_27945_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2913_cast_fp16")]; + string _SplitHeadsQ__mh_w_2915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2915_equation_0, values = (var_28407_cast_fp16, var_27952_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2915_cast_fp16")]; + string _SplitHeadsQ__mh_w_2917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2917_equation_0, values = (var_28407_cast_fp16, var_27959_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2917_cast_fp16")]; + string _SplitHeadsQ__mh_w_2919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2919_equation_0, values = (var_28407_cast_fp16, var_27966_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2919_cast_fp16")]; + string _SplitHeadsQ__mh_w_2921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2921_equation_0, values = (var_28411_cast_fp16, var_27973_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2921_cast_fp16")]; + string _SplitHeadsQ__mh_w_2923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2923_equation_0, values = (var_28411_cast_fp16, var_27980_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2923_cast_fp16")]; + string _SplitHeadsQ__mh_w_2925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2925_equation_0, values = (var_28411_cast_fp16, var_27987_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2925_cast_fp16")]; + string _SplitHeadsQ__mh_w_2927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2927_equation_0, values = (var_28411_cast_fp16, var_27994_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2927_cast_fp16")]; + string _SplitHeadsQ__mh_w_2929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2929_equation_0, values = (var_28415_cast_fp16, var_28001_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2929_cast_fp16")]; + string _SplitHeadsQ__mh_w_2931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2931_equation_0, values = (var_28415_cast_fp16, var_28008_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2931_cast_fp16")]; + string _SplitHeadsQ__mh_w_2933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2933_equation_0, values = (var_28415_cast_fp16, var_28015_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2933_cast_fp16")]; + string _SplitHeadsQ__mh_w_2935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2935_equation_0, values = (var_28415_cast_fp16, var_28022_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2935_cast_fp16")]; + string _SplitHeadsQ__mh_w_2937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2937_equation_0, values = (var_28419_cast_fp16, var_28029_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2937_cast_fp16")]; + string _SplitHeadsQ__mh_w_2939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2939_equation_0, values = (var_28419_cast_fp16, var_28036_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2939_cast_fp16")]; + string _SplitHeadsQ__mh_w_2941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2941_equation_0, values = (var_28419_cast_fp16, var_28043_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2941_cast_fp16")]; + string _SplitHeadsQ__mh_w_2943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2943_equation_0, values = (var_28419_cast_fp16, var_28050_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2943_cast_fp16")]; + string _SplitHeadsQ__mh_w_2945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2945_equation_0, values = (var_28423_cast_fp16, var_28057_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2945_cast_fp16")]; + string _SplitHeadsQ__mh_w_2947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2947_equation_0, values = (var_28423_cast_fp16, var_28064_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2947_cast_fp16")]; + string _SplitHeadsQ__mh_w_2949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2949_equation_0, values = (var_28423_cast_fp16, var_28071_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2949_cast_fp16")]; + string _SplitHeadsQ__mh_w_2951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2951_equation_0, values = (var_28423_cast_fp16, var_28078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2951_cast_fp16")]; + string _SplitHeadsQ__mh_w_2953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2953_equation_0, values = (var_28427_cast_fp16, var_28085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2953_cast_fp16")]; + string _SplitHeadsQ__mh_w_2955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2955_equation_0, values = (var_28427_cast_fp16, var_28092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2955_cast_fp16")]; + string _SplitHeadsQ__mh_w_2957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2957_equation_0, values = (var_28427_cast_fp16, var_28099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2957_cast_fp16")]; + string _SplitHeadsQ__mh_w_2959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2959_equation_0, values = (var_28427_cast_fp16, var_28106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2959_cast_fp16")]; + string _SplitHeadsQ__mh_w_2961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2961_equation_0, values = (var_28431_cast_fp16, var_28113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2961_cast_fp16")]; + string _SplitHeadsQ__mh_w_2963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2963_equation_0, values = (var_28431_cast_fp16, var_28120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2963_cast_fp16")]; + string _SplitHeadsQ__mh_w_2965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2965_equation_0, values = (var_28431_cast_fp16, var_28127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2965_cast_fp16")]; + string _SplitHeadsQ__mh_w_2967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2967_equation_0, values = (var_28431_cast_fp16, var_28134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2967_cast_fp16")]; + string _SplitHeadsQ__mh_w_2969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2969_equation_0, values = (var_28435_cast_fp16, var_28141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2969_cast_fp16")]; + string _SplitHeadsQ__mh_w_2971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2971_equation_0, values = (var_28435_cast_fp16, var_28148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2971_cast_fp16")]; + string _SplitHeadsQ__mh_w_2973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2973_equation_0, values = (var_28435_cast_fp16, var_28155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2973_cast_fp16")]; + string _SplitHeadsQ__mh_w_2975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2975_equation_0, values = (var_28435_cast_fp16, var_28162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2975_cast_fp16")]; + string _SplitHeadsQ__mh_w_2977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2977_equation_0, values = (var_28439_cast_fp16, var_28169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2977_cast_fp16")]; + string _SplitHeadsQ__mh_w_2979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2979_equation_0, values = (var_28439_cast_fp16, var_28176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2979_cast_fp16")]; + string _SplitHeadsQ__mh_w_2981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2981_equation_0, values = (var_28439_cast_fp16, var_28183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2981_cast_fp16")]; + string _SplitHeadsQ__mh_w_2983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2983_equation_0, values = (var_28439_cast_fp16, var_28190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2983_cast_fp16")]; + string _SplitHeadsQ__mh_w_2985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2985_equation_0, values = (var_28443_cast_fp16, var_28197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2985_cast_fp16")]; + string _SplitHeadsQ__mh_w_2987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2987_equation_0, values = (var_28443_cast_fp16, var_28204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2987_cast_fp16")]; + string _SplitHeadsQ__mh_w_2989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2989_equation_0, values = (var_28443_cast_fp16, var_28211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2989_cast_fp16")]; + string _SplitHeadsQ__mh_w_2991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2991_equation_0, values = (var_28443_cast_fp16, var_28218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2991_cast_fp16")]; + string _SplitHeadsQ__mh_w_2993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2993_equation_0, values = (var_28447_cast_fp16, var_28225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2993_cast_fp16")]; + string _SplitHeadsQ__mh_w_2995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2995_equation_0, values = (var_28447_cast_fp16, var_28232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2995_cast_fp16")]; + string _SplitHeadsQ__mh_w_2997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2997_equation_0, values = (var_28447_cast_fp16, var_28239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2997_cast_fp16")]; + string _SplitHeadsQ__mh_w_2999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_2999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_2999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_2999_equation_0, values = (var_28447_cast_fp16, var_28246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_2999_cast_fp16")]; + string _SplitHeadsQ__mh_w_3001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3001_equation_0, values = (var_28451_cast_fp16, var_28253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3001_cast_fp16")]; + string _SplitHeadsQ__mh_w_3003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3003_equation_0, values = (var_28451_cast_fp16, var_28260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3003_cast_fp16")]; + string _SplitHeadsQ__mh_w_3005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3005_equation_0, values = (var_28451_cast_fp16, var_28267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3005_cast_fp16")]; + string _SplitHeadsQ__mh_w_3007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3007_equation_0, values = (var_28451_cast_fp16, var_28274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3007_cast_fp16")]; + string _SplitHeadsQ__mh_w_3009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3009_equation_0, values = (var_28455_cast_fp16, var_28281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3009_cast_fp16")]; + string _SplitHeadsQ__mh_w_3011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3011_equation_0, values = (var_28455_cast_fp16, var_28288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3011_cast_fp16")]; + string _SplitHeadsQ__mh_w_3013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3013_equation_0, values = (var_28455_cast_fp16, var_28295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3013_cast_fp16")]; + string _SplitHeadsQ__mh_w_3015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3015_equation_0, values = (var_28455_cast_fp16, var_28302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3015_cast_fp16")]; + string _SplitHeadsQ__mh_w_3017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3017_equation_0, values = (var_28459_cast_fp16, var_28309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3017_cast_fp16")]; + string _SplitHeadsQ__mh_w_3019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3019_equation_0, values = (var_28459_cast_fp16, var_28316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3019_cast_fp16")]; + string _SplitHeadsQ__mh_w_3021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3021_equation_0, values = (var_28459_cast_fp16, var_28323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3021_cast_fp16")]; + string _SplitHeadsQ__mh_w_3023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3023_equation_0, values = (var_28459_cast_fp16, var_28330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3023_cast_fp16")]; + string _SplitHeadsQ__mh_w_3025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3025_equation_0, values = (var_28463_cast_fp16, var_28337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3025_cast_fp16")]; + string _SplitHeadsQ__mh_w_3027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3027_equation_0, values = (var_28463_cast_fp16, var_28344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3027_cast_fp16")]; + string _SplitHeadsQ__mh_w_3029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3029_equation_0, values = (var_28463_cast_fp16, var_28351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3029_cast_fp16")]; + string _SplitHeadsQ__mh_w_3031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3031_equation_0, values = (var_28463_cast_fp16, var_28358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3031_cast_fp16")]; + string _SplitHeadsQ__mh_w_3033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3033_equation_0, values = (var_28467_cast_fp16, var_28365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3033_cast_fp16")]; + string _SplitHeadsQ__mh_w_3035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3035_equation_0, values = (var_28467_cast_fp16, var_28372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3035_cast_fp16")]; + string _SplitHeadsQ__mh_w_3037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3037_equation_0, values = (var_28467_cast_fp16, var_28379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3037_cast_fp16")]; + string _SplitHeadsQ__mh_w_3039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3039_equation_0, values = (var_28467_cast_fp16, var_28386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3039_cast_fp16")]; + fp16 var_28708_to_fp16 = const()[name = string("op_28708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2881_cast_fp16, y = var_28708_to_fp16)[name = string("aw_chunk_2881_cast_fp16")]; + fp16 var_28710_to_fp16 = const()[name = string("op_28710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2883_cast_fp16, y = var_28710_to_fp16)[name = string("aw_chunk_2883_cast_fp16")]; + fp16 var_28712_to_fp16 = const()[name = string("op_28712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2885_cast_fp16, y = var_28712_to_fp16)[name = string("aw_chunk_2885_cast_fp16")]; + fp16 var_28714_to_fp16 = const()[name = string("op_28714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2887_cast_fp16, y = var_28714_to_fp16)[name = string("aw_chunk_2887_cast_fp16")]; + fp16 var_28716_to_fp16 = const()[name = string("op_28716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2889_cast_fp16, y = var_28716_to_fp16)[name = string("aw_chunk_2889_cast_fp16")]; + fp16 var_28718_to_fp16 = const()[name = string("op_28718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2891_cast_fp16, y = var_28718_to_fp16)[name = string("aw_chunk_2891_cast_fp16")]; + fp16 var_28720_to_fp16 = const()[name = string("op_28720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2893_cast_fp16, y = var_28720_to_fp16)[name = string("aw_chunk_2893_cast_fp16")]; + fp16 var_28722_to_fp16 = const()[name = string("op_28722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2895_cast_fp16, y = var_28722_to_fp16)[name = string("aw_chunk_2895_cast_fp16")]; + fp16 var_28724_to_fp16 = const()[name = string("op_28724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2897_cast_fp16, y = var_28724_to_fp16)[name = string("aw_chunk_2897_cast_fp16")]; + fp16 var_28726_to_fp16 = const()[name = string("op_28726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2899_cast_fp16, y = var_28726_to_fp16)[name = string("aw_chunk_2899_cast_fp16")]; + fp16 var_28728_to_fp16 = const()[name = string("op_28728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2901_cast_fp16, y = var_28728_to_fp16)[name = string("aw_chunk_2901_cast_fp16")]; + fp16 var_28730_to_fp16 = const()[name = string("op_28730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2903_cast_fp16, y = var_28730_to_fp16)[name = string("aw_chunk_2903_cast_fp16")]; + fp16 var_28732_to_fp16 = const()[name = string("op_28732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2905_cast_fp16, y = var_28732_to_fp16)[name = string("aw_chunk_2905_cast_fp16")]; + fp16 var_28734_to_fp16 = const()[name = string("op_28734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2907_cast_fp16, y = var_28734_to_fp16)[name = string("aw_chunk_2907_cast_fp16")]; + fp16 var_28736_to_fp16 = const()[name = string("op_28736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2909_cast_fp16, y = var_28736_to_fp16)[name = string("aw_chunk_2909_cast_fp16")]; + fp16 var_28738_to_fp16 = const()[name = string("op_28738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2911_cast_fp16, y = var_28738_to_fp16)[name = string("aw_chunk_2911_cast_fp16")]; + fp16 var_28740_to_fp16 = const()[name = string("op_28740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2913_cast_fp16, y = var_28740_to_fp16)[name = string("aw_chunk_2913_cast_fp16")]; + fp16 var_28742_to_fp16 = const()[name = string("op_28742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2915_cast_fp16, y = var_28742_to_fp16)[name = string("aw_chunk_2915_cast_fp16")]; + fp16 var_28744_to_fp16 = const()[name = string("op_28744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2917_cast_fp16, y = var_28744_to_fp16)[name = string("aw_chunk_2917_cast_fp16")]; + fp16 var_28746_to_fp16 = const()[name = string("op_28746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2919_cast_fp16, y = var_28746_to_fp16)[name = string("aw_chunk_2919_cast_fp16")]; + fp16 var_28748_to_fp16 = const()[name = string("op_28748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2921_cast_fp16, y = var_28748_to_fp16)[name = string("aw_chunk_2921_cast_fp16")]; + fp16 var_28750_to_fp16 = const()[name = string("op_28750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2923_cast_fp16, y = var_28750_to_fp16)[name = string("aw_chunk_2923_cast_fp16")]; + fp16 var_28752_to_fp16 = const()[name = string("op_28752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2925_cast_fp16, y = var_28752_to_fp16)[name = string("aw_chunk_2925_cast_fp16")]; + fp16 var_28754_to_fp16 = const()[name = string("op_28754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2927_cast_fp16, y = var_28754_to_fp16)[name = string("aw_chunk_2927_cast_fp16")]; + fp16 var_28756_to_fp16 = const()[name = string("op_28756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2929_cast_fp16, y = var_28756_to_fp16)[name = string("aw_chunk_2929_cast_fp16")]; + fp16 var_28758_to_fp16 = const()[name = string("op_28758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2931_cast_fp16, y = var_28758_to_fp16)[name = string("aw_chunk_2931_cast_fp16")]; + fp16 var_28760_to_fp16 = const()[name = string("op_28760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2933_cast_fp16, y = var_28760_to_fp16)[name = string("aw_chunk_2933_cast_fp16")]; + fp16 var_28762_to_fp16 = const()[name = string("op_28762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2935_cast_fp16, y = var_28762_to_fp16)[name = string("aw_chunk_2935_cast_fp16")]; + fp16 var_28764_to_fp16 = const()[name = string("op_28764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2937_cast_fp16, y = var_28764_to_fp16)[name = string("aw_chunk_2937_cast_fp16")]; + fp16 var_28766_to_fp16 = const()[name = string("op_28766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2939_cast_fp16, y = var_28766_to_fp16)[name = string("aw_chunk_2939_cast_fp16")]; + fp16 var_28768_to_fp16 = const()[name = string("op_28768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2941_cast_fp16, y = var_28768_to_fp16)[name = string("aw_chunk_2941_cast_fp16")]; + fp16 var_28770_to_fp16 = const()[name = string("op_28770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2943_cast_fp16, y = var_28770_to_fp16)[name = string("aw_chunk_2943_cast_fp16")]; + fp16 var_28772_to_fp16 = const()[name = string("op_28772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2945_cast_fp16, y = var_28772_to_fp16)[name = string("aw_chunk_2945_cast_fp16")]; + fp16 var_28774_to_fp16 = const()[name = string("op_28774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2947_cast_fp16, y = var_28774_to_fp16)[name = string("aw_chunk_2947_cast_fp16")]; + fp16 var_28776_to_fp16 = const()[name = string("op_28776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2949_cast_fp16, y = var_28776_to_fp16)[name = string("aw_chunk_2949_cast_fp16")]; + fp16 var_28778_to_fp16 = const()[name = string("op_28778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2951_cast_fp16, y = var_28778_to_fp16)[name = string("aw_chunk_2951_cast_fp16")]; + fp16 var_28780_to_fp16 = const()[name = string("op_28780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2953_cast_fp16, y = var_28780_to_fp16)[name = string("aw_chunk_2953_cast_fp16")]; + fp16 var_28782_to_fp16 = const()[name = string("op_28782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2955_cast_fp16, y = var_28782_to_fp16)[name = string("aw_chunk_2955_cast_fp16")]; + fp16 var_28784_to_fp16 = const()[name = string("op_28784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2957_cast_fp16, y = var_28784_to_fp16)[name = string("aw_chunk_2957_cast_fp16")]; + fp16 var_28786_to_fp16 = const()[name = string("op_28786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2959_cast_fp16, y = var_28786_to_fp16)[name = string("aw_chunk_2959_cast_fp16")]; + fp16 var_28788_to_fp16 = const()[name = string("op_28788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2961_cast_fp16, y = var_28788_to_fp16)[name = string("aw_chunk_2961_cast_fp16")]; + fp16 var_28790_to_fp16 = const()[name = string("op_28790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2963_cast_fp16, y = var_28790_to_fp16)[name = string("aw_chunk_2963_cast_fp16")]; + fp16 var_28792_to_fp16 = const()[name = string("op_28792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2965_cast_fp16, y = var_28792_to_fp16)[name = string("aw_chunk_2965_cast_fp16")]; + fp16 var_28794_to_fp16 = const()[name = string("op_28794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2967_cast_fp16, y = var_28794_to_fp16)[name = string("aw_chunk_2967_cast_fp16")]; + fp16 var_28796_to_fp16 = const()[name = string("op_28796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2969_cast_fp16, y = var_28796_to_fp16)[name = string("aw_chunk_2969_cast_fp16")]; + fp16 var_28798_to_fp16 = const()[name = string("op_28798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2971_cast_fp16, y = var_28798_to_fp16)[name = string("aw_chunk_2971_cast_fp16")]; + fp16 var_28800_to_fp16 = const()[name = string("op_28800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2973_cast_fp16, y = var_28800_to_fp16)[name = string("aw_chunk_2973_cast_fp16")]; + fp16 var_28802_to_fp16 = const()[name = string("op_28802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2975_cast_fp16, y = var_28802_to_fp16)[name = string("aw_chunk_2975_cast_fp16")]; + fp16 var_28804_to_fp16 = const()[name = string("op_28804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2977_cast_fp16, y = var_28804_to_fp16)[name = string("aw_chunk_2977_cast_fp16")]; + fp16 var_28806_to_fp16 = const()[name = string("op_28806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2979_cast_fp16, y = var_28806_to_fp16)[name = string("aw_chunk_2979_cast_fp16")]; + fp16 var_28808_to_fp16 = const()[name = string("op_28808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2981_cast_fp16, y = var_28808_to_fp16)[name = string("aw_chunk_2981_cast_fp16")]; + fp16 var_28810_to_fp16 = const()[name = string("op_28810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2983_cast_fp16, y = var_28810_to_fp16)[name = string("aw_chunk_2983_cast_fp16")]; + fp16 var_28812_to_fp16 = const()[name = string("op_28812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2985_cast_fp16, y = var_28812_to_fp16)[name = string("aw_chunk_2985_cast_fp16")]; + fp16 var_28814_to_fp16 = const()[name = string("op_28814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2987_cast_fp16, y = var_28814_to_fp16)[name = string("aw_chunk_2987_cast_fp16")]; + fp16 var_28816_to_fp16 = const()[name = string("op_28816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2989_cast_fp16, y = var_28816_to_fp16)[name = string("aw_chunk_2989_cast_fp16")]; + fp16 var_28818_to_fp16 = const()[name = string("op_28818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2991_cast_fp16, y = var_28818_to_fp16)[name = string("aw_chunk_2991_cast_fp16")]; + fp16 var_28820_to_fp16 = const()[name = string("op_28820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2993_cast_fp16, y = var_28820_to_fp16)[name = string("aw_chunk_2993_cast_fp16")]; + fp16 var_28822_to_fp16 = const()[name = string("op_28822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2995_cast_fp16, y = var_28822_to_fp16)[name = string("aw_chunk_2995_cast_fp16")]; + fp16 var_28824_to_fp16 = const()[name = string("op_28824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2997_cast_fp16, y = var_28824_to_fp16)[name = string("aw_chunk_2997_cast_fp16")]; + fp16 var_28826_to_fp16 = const()[name = string("op_28826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_2999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_2999_cast_fp16, y = var_28826_to_fp16)[name = string("aw_chunk_2999_cast_fp16")]; + fp16 var_28828_to_fp16 = const()[name = string("op_28828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3001_cast_fp16, y = var_28828_to_fp16)[name = string("aw_chunk_3001_cast_fp16")]; + fp16 var_28830_to_fp16 = const()[name = string("op_28830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3003_cast_fp16, y = var_28830_to_fp16)[name = string("aw_chunk_3003_cast_fp16")]; + fp16 var_28832_to_fp16 = const()[name = string("op_28832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3005_cast_fp16, y = var_28832_to_fp16)[name = string("aw_chunk_3005_cast_fp16")]; + fp16 var_28834_to_fp16 = const()[name = string("op_28834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3007_cast_fp16, y = var_28834_to_fp16)[name = string("aw_chunk_3007_cast_fp16")]; + fp16 var_28836_to_fp16 = const()[name = string("op_28836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3009_cast_fp16, y = var_28836_to_fp16)[name = string("aw_chunk_3009_cast_fp16")]; + fp16 var_28838_to_fp16 = const()[name = string("op_28838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3011_cast_fp16, y = var_28838_to_fp16)[name = string("aw_chunk_3011_cast_fp16")]; + fp16 var_28840_to_fp16 = const()[name = string("op_28840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3013_cast_fp16, y = var_28840_to_fp16)[name = string("aw_chunk_3013_cast_fp16")]; + fp16 var_28842_to_fp16 = const()[name = string("op_28842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3015_cast_fp16, y = var_28842_to_fp16)[name = string("aw_chunk_3015_cast_fp16")]; + fp16 var_28844_to_fp16 = const()[name = string("op_28844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3017_cast_fp16, y = var_28844_to_fp16)[name = string("aw_chunk_3017_cast_fp16")]; + fp16 var_28846_to_fp16 = const()[name = string("op_28846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3019_cast_fp16, y = var_28846_to_fp16)[name = string("aw_chunk_3019_cast_fp16")]; + fp16 var_28848_to_fp16 = const()[name = string("op_28848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3021_cast_fp16, y = var_28848_to_fp16)[name = string("aw_chunk_3021_cast_fp16")]; + fp16 var_28850_to_fp16 = const()[name = string("op_28850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3023_cast_fp16, y = var_28850_to_fp16)[name = string("aw_chunk_3023_cast_fp16")]; + fp16 var_28852_to_fp16 = const()[name = string("op_28852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3025_cast_fp16, y = var_28852_to_fp16)[name = string("aw_chunk_3025_cast_fp16")]; + fp16 var_28854_to_fp16 = const()[name = string("op_28854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3027_cast_fp16, y = var_28854_to_fp16)[name = string("aw_chunk_3027_cast_fp16")]; + fp16 var_28856_to_fp16 = const()[name = string("op_28856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3029_cast_fp16, y = var_28856_to_fp16)[name = string("aw_chunk_3029_cast_fp16")]; + fp16 var_28858_to_fp16 = const()[name = string("op_28858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3031_cast_fp16, y = var_28858_to_fp16)[name = string("aw_chunk_3031_cast_fp16")]; + fp16 var_28860_to_fp16 = const()[name = string("op_28860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3033_cast_fp16, y = var_28860_to_fp16)[name = string("aw_chunk_3033_cast_fp16")]; + fp16 var_28862_to_fp16 = const()[name = string("op_28862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3035_cast_fp16, y = var_28862_to_fp16)[name = string("aw_chunk_3035_cast_fp16")]; + fp16 var_28864_to_fp16 = const()[name = string("op_28864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3037_cast_fp16, y = var_28864_to_fp16)[name = string("aw_chunk_3037_cast_fp16")]; + fp16 var_28866_to_fp16 = const()[name = string("op_28866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3039_cast_fp16, y = var_28866_to_fp16)[name = string("aw_chunk_3039_cast_fp16")]; + tensor var_28868_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2881_cast_fp16)[name = string("op_28868_cast_fp16")]; + tensor var_28869_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2883_cast_fp16)[name = string("op_28869_cast_fp16")]; + tensor var_28870_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2885_cast_fp16)[name = string("op_28870_cast_fp16")]; + tensor var_28871_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2887_cast_fp16)[name = string("op_28871_cast_fp16")]; + tensor var_28872_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2889_cast_fp16)[name = string("op_28872_cast_fp16")]; + tensor var_28873_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2891_cast_fp16)[name = string("op_28873_cast_fp16")]; + tensor var_28874_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2893_cast_fp16)[name = string("op_28874_cast_fp16")]; + tensor var_28875_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2895_cast_fp16)[name = string("op_28875_cast_fp16")]; + tensor var_28876_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2897_cast_fp16)[name = string("op_28876_cast_fp16")]; + tensor var_28877_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2899_cast_fp16)[name = string("op_28877_cast_fp16")]; + tensor var_28878_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2901_cast_fp16)[name = string("op_28878_cast_fp16")]; + tensor var_28879_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2903_cast_fp16)[name = string("op_28879_cast_fp16")]; + tensor var_28880_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2905_cast_fp16)[name = string("op_28880_cast_fp16")]; + tensor var_28881_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2907_cast_fp16)[name = string("op_28881_cast_fp16")]; + tensor var_28882_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2909_cast_fp16)[name = string("op_28882_cast_fp16")]; + tensor var_28883_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2911_cast_fp16)[name = string("op_28883_cast_fp16")]; + tensor var_28884_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2913_cast_fp16)[name = string("op_28884_cast_fp16")]; + tensor var_28885_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2915_cast_fp16)[name = string("op_28885_cast_fp16")]; + tensor var_28886_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2917_cast_fp16)[name = string("op_28886_cast_fp16")]; + tensor var_28887_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2919_cast_fp16)[name = string("op_28887_cast_fp16")]; + tensor var_28888_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2921_cast_fp16)[name = string("op_28888_cast_fp16")]; + tensor var_28889_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2923_cast_fp16)[name = string("op_28889_cast_fp16")]; + tensor var_28890_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2925_cast_fp16)[name = string("op_28890_cast_fp16")]; + tensor var_28891_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2927_cast_fp16)[name = string("op_28891_cast_fp16")]; + tensor var_28892_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2929_cast_fp16)[name = string("op_28892_cast_fp16")]; + tensor var_28893_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2931_cast_fp16)[name = string("op_28893_cast_fp16")]; + tensor var_28894_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2933_cast_fp16)[name = string("op_28894_cast_fp16")]; + tensor var_28895_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2935_cast_fp16)[name = string("op_28895_cast_fp16")]; + tensor var_28896_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2937_cast_fp16)[name = string("op_28896_cast_fp16")]; + tensor var_28897_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2939_cast_fp16)[name = string("op_28897_cast_fp16")]; + tensor var_28898_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2941_cast_fp16)[name = string("op_28898_cast_fp16")]; + tensor var_28899_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2943_cast_fp16)[name = string("op_28899_cast_fp16")]; + tensor var_28900_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2945_cast_fp16)[name = string("op_28900_cast_fp16")]; + tensor var_28901_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2947_cast_fp16)[name = string("op_28901_cast_fp16")]; + tensor var_28902_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2949_cast_fp16)[name = string("op_28902_cast_fp16")]; + tensor var_28903_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2951_cast_fp16)[name = string("op_28903_cast_fp16")]; + tensor var_28904_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2953_cast_fp16)[name = string("op_28904_cast_fp16")]; + tensor var_28905_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2955_cast_fp16)[name = string("op_28905_cast_fp16")]; + tensor var_28906_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2957_cast_fp16)[name = string("op_28906_cast_fp16")]; + tensor var_28907_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2959_cast_fp16)[name = string("op_28907_cast_fp16")]; + tensor var_28908_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2961_cast_fp16)[name = string("op_28908_cast_fp16")]; + tensor var_28909_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2963_cast_fp16)[name = string("op_28909_cast_fp16")]; + tensor var_28910_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2965_cast_fp16)[name = string("op_28910_cast_fp16")]; + tensor var_28911_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2967_cast_fp16)[name = string("op_28911_cast_fp16")]; + tensor var_28912_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2969_cast_fp16)[name = string("op_28912_cast_fp16")]; + tensor var_28913_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2971_cast_fp16)[name = string("op_28913_cast_fp16")]; + tensor var_28914_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2973_cast_fp16)[name = string("op_28914_cast_fp16")]; + tensor var_28915_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2975_cast_fp16)[name = string("op_28915_cast_fp16")]; + tensor var_28916_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2977_cast_fp16)[name = string("op_28916_cast_fp16")]; + tensor var_28917_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2979_cast_fp16)[name = string("op_28917_cast_fp16")]; + tensor var_28918_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2981_cast_fp16)[name = string("op_28918_cast_fp16")]; + tensor var_28919_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2983_cast_fp16)[name = string("op_28919_cast_fp16")]; + tensor var_28920_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2985_cast_fp16)[name = string("op_28920_cast_fp16")]; + tensor var_28921_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2987_cast_fp16)[name = string("op_28921_cast_fp16")]; + tensor var_28922_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2989_cast_fp16)[name = string("op_28922_cast_fp16")]; + tensor var_28923_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2991_cast_fp16)[name = string("op_28923_cast_fp16")]; + tensor var_28924_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2993_cast_fp16)[name = string("op_28924_cast_fp16")]; + tensor var_28925_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2995_cast_fp16)[name = string("op_28925_cast_fp16")]; + tensor var_28926_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2997_cast_fp16)[name = string("op_28926_cast_fp16")]; + tensor var_28927_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_2999_cast_fp16)[name = string("op_28927_cast_fp16")]; + tensor var_28928_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3001_cast_fp16)[name = string("op_28928_cast_fp16")]; + tensor var_28929_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3003_cast_fp16)[name = string("op_28929_cast_fp16")]; + tensor var_28930_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3005_cast_fp16)[name = string("op_28930_cast_fp16")]; + tensor var_28931_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3007_cast_fp16)[name = string("op_28931_cast_fp16")]; + tensor var_28932_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3009_cast_fp16)[name = string("op_28932_cast_fp16")]; + tensor var_28933_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3011_cast_fp16)[name = string("op_28933_cast_fp16")]; + tensor var_28934_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3013_cast_fp16)[name = string("op_28934_cast_fp16")]; + tensor var_28935_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3015_cast_fp16)[name = string("op_28935_cast_fp16")]; + tensor var_28936_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3017_cast_fp16)[name = string("op_28936_cast_fp16")]; + tensor var_28937_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3019_cast_fp16)[name = string("op_28937_cast_fp16")]; + tensor var_28938_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3021_cast_fp16)[name = string("op_28938_cast_fp16")]; + tensor var_28939_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3023_cast_fp16)[name = string("op_28939_cast_fp16")]; + tensor var_28940_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3025_cast_fp16)[name = string("op_28940_cast_fp16")]; + tensor var_28941_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3027_cast_fp16)[name = string("op_28941_cast_fp16")]; + tensor var_28942_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3029_cast_fp16)[name = string("op_28942_cast_fp16")]; + tensor var_28943_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3031_cast_fp16)[name = string("op_28943_cast_fp16")]; + tensor var_28944_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3033_cast_fp16)[name = string("op_28944_cast_fp16")]; + tensor var_28945_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3035_cast_fp16)[name = string("op_28945_cast_fp16")]; + tensor var_28946_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3037_cast_fp16)[name = string("op_28946_cast_fp16")]; + tensor var_28947_cast_fp16 = softmax(axis = var_27693, x = aw_chunk_3039_cast_fp16)[name = string("op_28947_cast_fp16")]; + string var_28949_equation_0 = const()[name = string("op_28949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28949_cast_fp16 = einsum(equation = var_28949_equation_0, values = (var_28469_cast_fp16, var_28868_cast_fp16))[name = string("op_28949_cast_fp16")]; + string var_28951_equation_0 = const()[name = string("op_28951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28951_cast_fp16 = einsum(equation = var_28951_equation_0, values = (var_28469_cast_fp16, var_28869_cast_fp16))[name = string("op_28951_cast_fp16")]; + string var_28953_equation_0 = const()[name = string("op_28953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28953_cast_fp16 = einsum(equation = var_28953_equation_0, values = (var_28469_cast_fp16, var_28870_cast_fp16))[name = string("op_28953_cast_fp16")]; + string var_28955_equation_0 = const()[name = string("op_28955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28955_cast_fp16 = einsum(equation = var_28955_equation_0, values = (var_28469_cast_fp16, var_28871_cast_fp16))[name = string("op_28955_cast_fp16")]; + string var_28957_equation_0 = const()[name = string("op_28957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28957_cast_fp16 = einsum(equation = var_28957_equation_0, values = (var_28473_cast_fp16, var_28872_cast_fp16))[name = string("op_28957_cast_fp16")]; + string var_28959_equation_0 = const()[name = string("op_28959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28959_cast_fp16 = einsum(equation = var_28959_equation_0, values = (var_28473_cast_fp16, var_28873_cast_fp16))[name = string("op_28959_cast_fp16")]; + string var_28961_equation_0 = const()[name = string("op_28961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28961_cast_fp16 = einsum(equation = var_28961_equation_0, values = (var_28473_cast_fp16, var_28874_cast_fp16))[name = string("op_28961_cast_fp16")]; + string var_28963_equation_0 = const()[name = string("op_28963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28963_cast_fp16 = einsum(equation = var_28963_equation_0, values = (var_28473_cast_fp16, var_28875_cast_fp16))[name = string("op_28963_cast_fp16")]; + string var_28965_equation_0 = const()[name = string("op_28965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28965_cast_fp16 = einsum(equation = var_28965_equation_0, values = (var_28477_cast_fp16, var_28876_cast_fp16))[name = string("op_28965_cast_fp16")]; + string var_28967_equation_0 = const()[name = string("op_28967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28967_cast_fp16 = einsum(equation = var_28967_equation_0, values = (var_28477_cast_fp16, var_28877_cast_fp16))[name = string("op_28967_cast_fp16")]; + string var_28969_equation_0 = const()[name = string("op_28969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28969_cast_fp16 = einsum(equation = var_28969_equation_0, values = (var_28477_cast_fp16, var_28878_cast_fp16))[name = string("op_28969_cast_fp16")]; + string var_28971_equation_0 = const()[name = string("op_28971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28971_cast_fp16 = einsum(equation = var_28971_equation_0, values = (var_28477_cast_fp16, var_28879_cast_fp16))[name = string("op_28971_cast_fp16")]; + string var_28973_equation_0 = const()[name = string("op_28973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28973_cast_fp16 = einsum(equation = var_28973_equation_0, values = (var_28481_cast_fp16, var_28880_cast_fp16))[name = string("op_28973_cast_fp16")]; + string var_28975_equation_0 = const()[name = string("op_28975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28975_cast_fp16 = einsum(equation = var_28975_equation_0, values = (var_28481_cast_fp16, var_28881_cast_fp16))[name = string("op_28975_cast_fp16")]; + string var_28977_equation_0 = const()[name = string("op_28977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28977_cast_fp16 = einsum(equation = var_28977_equation_0, values = (var_28481_cast_fp16, var_28882_cast_fp16))[name = string("op_28977_cast_fp16")]; + string var_28979_equation_0 = const()[name = string("op_28979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28979_cast_fp16 = einsum(equation = var_28979_equation_0, values = (var_28481_cast_fp16, var_28883_cast_fp16))[name = string("op_28979_cast_fp16")]; + string var_28981_equation_0 = const()[name = string("op_28981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28981_cast_fp16 = einsum(equation = var_28981_equation_0, values = (var_28485_cast_fp16, var_28884_cast_fp16))[name = string("op_28981_cast_fp16")]; + string var_28983_equation_0 = const()[name = string("op_28983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28983_cast_fp16 = einsum(equation = var_28983_equation_0, values = (var_28485_cast_fp16, var_28885_cast_fp16))[name = string("op_28983_cast_fp16")]; + string var_28985_equation_0 = const()[name = string("op_28985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28985_cast_fp16 = einsum(equation = var_28985_equation_0, values = (var_28485_cast_fp16, var_28886_cast_fp16))[name = string("op_28985_cast_fp16")]; + string var_28987_equation_0 = const()[name = string("op_28987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28987_cast_fp16 = einsum(equation = var_28987_equation_0, values = (var_28485_cast_fp16, var_28887_cast_fp16))[name = string("op_28987_cast_fp16")]; + string var_28989_equation_0 = const()[name = string("op_28989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28989_cast_fp16 = einsum(equation = var_28989_equation_0, values = (var_28489_cast_fp16, var_28888_cast_fp16))[name = string("op_28989_cast_fp16")]; + string var_28991_equation_0 = const()[name = string("op_28991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28991_cast_fp16 = einsum(equation = var_28991_equation_0, values = (var_28489_cast_fp16, var_28889_cast_fp16))[name = string("op_28991_cast_fp16")]; + string var_28993_equation_0 = const()[name = string("op_28993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28993_cast_fp16 = einsum(equation = var_28993_equation_0, values = (var_28489_cast_fp16, var_28890_cast_fp16))[name = string("op_28993_cast_fp16")]; + string var_28995_equation_0 = const()[name = string("op_28995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28995_cast_fp16 = einsum(equation = var_28995_equation_0, values = (var_28489_cast_fp16, var_28891_cast_fp16))[name = string("op_28995_cast_fp16")]; + string var_28997_equation_0 = const()[name = string("op_28997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28997_cast_fp16 = einsum(equation = var_28997_equation_0, values = (var_28493_cast_fp16, var_28892_cast_fp16))[name = string("op_28997_cast_fp16")]; + string var_28999_equation_0 = const()[name = string("op_28999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_28999_cast_fp16 = einsum(equation = var_28999_equation_0, values = (var_28493_cast_fp16, var_28893_cast_fp16))[name = string("op_28999_cast_fp16")]; + string var_29001_equation_0 = const()[name = string("op_29001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29001_cast_fp16 = einsum(equation = var_29001_equation_0, values = (var_28493_cast_fp16, var_28894_cast_fp16))[name = string("op_29001_cast_fp16")]; + string var_29003_equation_0 = const()[name = string("op_29003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29003_cast_fp16 = einsum(equation = var_29003_equation_0, values = (var_28493_cast_fp16, var_28895_cast_fp16))[name = string("op_29003_cast_fp16")]; + string var_29005_equation_0 = const()[name = string("op_29005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29005_cast_fp16 = einsum(equation = var_29005_equation_0, values = (var_28497_cast_fp16, var_28896_cast_fp16))[name = string("op_29005_cast_fp16")]; + string var_29007_equation_0 = const()[name = string("op_29007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29007_cast_fp16 = einsum(equation = var_29007_equation_0, values = (var_28497_cast_fp16, var_28897_cast_fp16))[name = string("op_29007_cast_fp16")]; + string var_29009_equation_0 = const()[name = string("op_29009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29009_cast_fp16 = einsum(equation = var_29009_equation_0, values = (var_28497_cast_fp16, var_28898_cast_fp16))[name = string("op_29009_cast_fp16")]; + string var_29011_equation_0 = const()[name = string("op_29011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29011_cast_fp16 = einsum(equation = var_29011_equation_0, values = (var_28497_cast_fp16, var_28899_cast_fp16))[name = string("op_29011_cast_fp16")]; + string var_29013_equation_0 = const()[name = string("op_29013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29013_cast_fp16 = einsum(equation = var_29013_equation_0, values = (var_28501_cast_fp16, var_28900_cast_fp16))[name = string("op_29013_cast_fp16")]; + string var_29015_equation_0 = const()[name = string("op_29015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29015_cast_fp16 = einsum(equation = var_29015_equation_0, values = (var_28501_cast_fp16, var_28901_cast_fp16))[name = string("op_29015_cast_fp16")]; + string var_29017_equation_0 = const()[name = string("op_29017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29017_cast_fp16 = einsum(equation = var_29017_equation_0, values = (var_28501_cast_fp16, var_28902_cast_fp16))[name = string("op_29017_cast_fp16")]; + string var_29019_equation_0 = const()[name = string("op_29019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29019_cast_fp16 = einsum(equation = var_29019_equation_0, values = (var_28501_cast_fp16, var_28903_cast_fp16))[name = string("op_29019_cast_fp16")]; + string var_29021_equation_0 = const()[name = string("op_29021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29021_cast_fp16 = einsum(equation = var_29021_equation_0, values = (var_28505_cast_fp16, var_28904_cast_fp16))[name = string("op_29021_cast_fp16")]; + string var_29023_equation_0 = const()[name = string("op_29023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29023_cast_fp16 = einsum(equation = var_29023_equation_0, values = (var_28505_cast_fp16, var_28905_cast_fp16))[name = string("op_29023_cast_fp16")]; + string var_29025_equation_0 = const()[name = string("op_29025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29025_cast_fp16 = einsum(equation = var_29025_equation_0, values = (var_28505_cast_fp16, var_28906_cast_fp16))[name = string("op_29025_cast_fp16")]; + string var_29027_equation_0 = const()[name = string("op_29027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29027_cast_fp16 = einsum(equation = var_29027_equation_0, values = (var_28505_cast_fp16, var_28907_cast_fp16))[name = string("op_29027_cast_fp16")]; + string var_29029_equation_0 = const()[name = string("op_29029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29029_cast_fp16 = einsum(equation = var_29029_equation_0, values = (var_28509_cast_fp16, var_28908_cast_fp16))[name = string("op_29029_cast_fp16")]; + string var_29031_equation_0 = const()[name = string("op_29031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29031_cast_fp16 = einsum(equation = var_29031_equation_0, values = (var_28509_cast_fp16, var_28909_cast_fp16))[name = string("op_29031_cast_fp16")]; + string var_29033_equation_0 = const()[name = string("op_29033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29033_cast_fp16 = einsum(equation = var_29033_equation_0, values = (var_28509_cast_fp16, var_28910_cast_fp16))[name = string("op_29033_cast_fp16")]; + string var_29035_equation_0 = const()[name = string("op_29035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29035_cast_fp16 = einsum(equation = var_29035_equation_0, values = (var_28509_cast_fp16, var_28911_cast_fp16))[name = string("op_29035_cast_fp16")]; + string var_29037_equation_0 = const()[name = string("op_29037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29037_cast_fp16 = einsum(equation = var_29037_equation_0, values = (var_28513_cast_fp16, var_28912_cast_fp16))[name = string("op_29037_cast_fp16")]; + string var_29039_equation_0 = const()[name = string("op_29039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29039_cast_fp16 = einsum(equation = var_29039_equation_0, values = (var_28513_cast_fp16, var_28913_cast_fp16))[name = string("op_29039_cast_fp16")]; + string var_29041_equation_0 = const()[name = string("op_29041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29041_cast_fp16 = einsum(equation = var_29041_equation_0, values = (var_28513_cast_fp16, var_28914_cast_fp16))[name = string("op_29041_cast_fp16")]; + string var_29043_equation_0 = const()[name = string("op_29043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29043_cast_fp16 = einsum(equation = var_29043_equation_0, values = (var_28513_cast_fp16, var_28915_cast_fp16))[name = string("op_29043_cast_fp16")]; + string var_29045_equation_0 = const()[name = string("op_29045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29045_cast_fp16 = einsum(equation = var_29045_equation_0, values = (var_28517_cast_fp16, var_28916_cast_fp16))[name = string("op_29045_cast_fp16")]; + string var_29047_equation_0 = const()[name = string("op_29047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29047_cast_fp16 = einsum(equation = var_29047_equation_0, values = (var_28517_cast_fp16, var_28917_cast_fp16))[name = string("op_29047_cast_fp16")]; + string var_29049_equation_0 = const()[name = string("op_29049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29049_cast_fp16 = einsum(equation = var_29049_equation_0, values = (var_28517_cast_fp16, var_28918_cast_fp16))[name = string("op_29049_cast_fp16")]; + string var_29051_equation_0 = const()[name = string("op_29051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29051_cast_fp16 = einsum(equation = var_29051_equation_0, values = (var_28517_cast_fp16, var_28919_cast_fp16))[name = string("op_29051_cast_fp16")]; + string var_29053_equation_0 = const()[name = string("op_29053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29053_cast_fp16 = einsum(equation = var_29053_equation_0, values = (var_28521_cast_fp16, var_28920_cast_fp16))[name = string("op_29053_cast_fp16")]; + string var_29055_equation_0 = const()[name = string("op_29055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29055_cast_fp16 = einsum(equation = var_29055_equation_0, values = (var_28521_cast_fp16, var_28921_cast_fp16))[name = string("op_29055_cast_fp16")]; + string var_29057_equation_0 = const()[name = string("op_29057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29057_cast_fp16 = einsum(equation = var_29057_equation_0, values = (var_28521_cast_fp16, var_28922_cast_fp16))[name = string("op_29057_cast_fp16")]; + string var_29059_equation_0 = const()[name = string("op_29059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29059_cast_fp16 = einsum(equation = var_29059_equation_0, values = (var_28521_cast_fp16, var_28923_cast_fp16))[name = string("op_29059_cast_fp16")]; + string var_29061_equation_0 = const()[name = string("op_29061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29061_cast_fp16 = einsum(equation = var_29061_equation_0, values = (var_28525_cast_fp16, var_28924_cast_fp16))[name = string("op_29061_cast_fp16")]; + string var_29063_equation_0 = const()[name = string("op_29063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29063_cast_fp16 = einsum(equation = var_29063_equation_0, values = (var_28525_cast_fp16, var_28925_cast_fp16))[name = string("op_29063_cast_fp16")]; + string var_29065_equation_0 = const()[name = string("op_29065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29065_cast_fp16 = einsum(equation = var_29065_equation_0, values = (var_28525_cast_fp16, var_28926_cast_fp16))[name = string("op_29065_cast_fp16")]; + string var_29067_equation_0 = const()[name = string("op_29067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29067_cast_fp16 = einsum(equation = var_29067_equation_0, values = (var_28525_cast_fp16, var_28927_cast_fp16))[name = string("op_29067_cast_fp16")]; + string var_29069_equation_0 = const()[name = string("op_29069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29069_cast_fp16 = einsum(equation = var_29069_equation_0, values = (var_28529_cast_fp16, var_28928_cast_fp16))[name = string("op_29069_cast_fp16")]; + string var_29071_equation_0 = const()[name = string("op_29071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29071_cast_fp16 = einsum(equation = var_29071_equation_0, values = (var_28529_cast_fp16, var_28929_cast_fp16))[name = string("op_29071_cast_fp16")]; + string var_29073_equation_0 = const()[name = string("op_29073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29073_cast_fp16 = einsum(equation = var_29073_equation_0, values = (var_28529_cast_fp16, var_28930_cast_fp16))[name = string("op_29073_cast_fp16")]; + string var_29075_equation_0 = const()[name = string("op_29075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29075_cast_fp16 = einsum(equation = var_29075_equation_0, values = (var_28529_cast_fp16, var_28931_cast_fp16))[name = string("op_29075_cast_fp16")]; + string var_29077_equation_0 = const()[name = string("op_29077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29077_cast_fp16 = einsum(equation = var_29077_equation_0, values = (var_28533_cast_fp16, var_28932_cast_fp16))[name = string("op_29077_cast_fp16")]; + string var_29079_equation_0 = const()[name = string("op_29079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29079_cast_fp16 = einsum(equation = var_29079_equation_0, values = (var_28533_cast_fp16, var_28933_cast_fp16))[name = string("op_29079_cast_fp16")]; + string var_29081_equation_0 = const()[name = string("op_29081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29081_cast_fp16 = einsum(equation = var_29081_equation_0, values = (var_28533_cast_fp16, var_28934_cast_fp16))[name = string("op_29081_cast_fp16")]; + string var_29083_equation_0 = const()[name = string("op_29083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29083_cast_fp16 = einsum(equation = var_29083_equation_0, values = (var_28533_cast_fp16, var_28935_cast_fp16))[name = string("op_29083_cast_fp16")]; + string var_29085_equation_0 = const()[name = string("op_29085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29085_cast_fp16 = einsum(equation = var_29085_equation_0, values = (var_28537_cast_fp16, var_28936_cast_fp16))[name = string("op_29085_cast_fp16")]; + string var_29087_equation_0 = const()[name = string("op_29087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29087_cast_fp16 = einsum(equation = var_29087_equation_0, values = (var_28537_cast_fp16, var_28937_cast_fp16))[name = string("op_29087_cast_fp16")]; + string var_29089_equation_0 = const()[name = string("op_29089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29089_cast_fp16 = einsum(equation = var_29089_equation_0, values = (var_28537_cast_fp16, var_28938_cast_fp16))[name = string("op_29089_cast_fp16")]; + string var_29091_equation_0 = const()[name = string("op_29091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29091_cast_fp16 = einsum(equation = var_29091_equation_0, values = (var_28537_cast_fp16, var_28939_cast_fp16))[name = string("op_29091_cast_fp16")]; + string var_29093_equation_0 = const()[name = string("op_29093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29093_cast_fp16 = einsum(equation = var_29093_equation_0, values = (var_28541_cast_fp16, var_28940_cast_fp16))[name = string("op_29093_cast_fp16")]; + string var_29095_equation_0 = const()[name = string("op_29095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29095_cast_fp16 = einsum(equation = var_29095_equation_0, values = (var_28541_cast_fp16, var_28941_cast_fp16))[name = string("op_29095_cast_fp16")]; + string var_29097_equation_0 = const()[name = string("op_29097_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29097_cast_fp16 = einsum(equation = var_29097_equation_0, values = (var_28541_cast_fp16, var_28942_cast_fp16))[name = string("op_29097_cast_fp16")]; + string var_29099_equation_0 = const()[name = string("op_29099_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29099_cast_fp16 = einsum(equation = var_29099_equation_0, values = (var_28541_cast_fp16, var_28943_cast_fp16))[name = string("op_29099_cast_fp16")]; + string var_29101_equation_0 = const()[name = string("op_29101_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29101_cast_fp16 = einsum(equation = var_29101_equation_0, values = (var_28545_cast_fp16, var_28944_cast_fp16))[name = string("op_29101_cast_fp16")]; + string var_29103_equation_0 = const()[name = string("op_29103_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29103_cast_fp16 = einsum(equation = var_29103_equation_0, values = (var_28545_cast_fp16, var_28945_cast_fp16))[name = string("op_29103_cast_fp16")]; + string var_29105_equation_0 = const()[name = string("op_29105_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29105_cast_fp16 = einsum(equation = var_29105_equation_0, values = (var_28545_cast_fp16, var_28946_cast_fp16))[name = string("op_29105_cast_fp16")]; + string var_29107_equation_0 = const()[name = string("op_29107_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_29107_cast_fp16 = einsum(equation = var_29107_equation_0, values = (var_28545_cast_fp16, var_28947_cast_fp16))[name = string("op_29107_cast_fp16")]; + bool var_29109_interleave_0 = const()[name = string("op_29109_interleave_0"), val = bool(false)]; + tensor var_29109_cast_fp16 = concat(axis = var_27668, interleave = var_29109_interleave_0, values = (var_28949_cast_fp16, var_28951_cast_fp16, var_28953_cast_fp16, var_28955_cast_fp16))[name = string("op_29109_cast_fp16")]; + bool var_29111_interleave_0 = const()[name = string("op_29111_interleave_0"), val = bool(false)]; + tensor var_29111_cast_fp16 = concat(axis = var_27668, interleave = var_29111_interleave_0, values = (var_28957_cast_fp16, var_28959_cast_fp16, var_28961_cast_fp16, var_28963_cast_fp16))[name = string("op_29111_cast_fp16")]; + bool var_29113_interleave_0 = const()[name = string("op_29113_interleave_0"), val = bool(false)]; + tensor var_29113_cast_fp16 = concat(axis = var_27668, interleave = var_29113_interleave_0, values = (var_28965_cast_fp16, var_28967_cast_fp16, var_28969_cast_fp16, var_28971_cast_fp16))[name = string("op_29113_cast_fp16")]; + bool var_29115_interleave_0 = const()[name = string("op_29115_interleave_0"), val = bool(false)]; + tensor var_29115_cast_fp16 = concat(axis = var_27668, interleave = var_29115_interleave_0, values = (var_28973_cast_fp16, var_28975_cast_fp16, var_28977_cast_fp16, var_28979_cast_fp16))[name = string("op_29115_cast_fp16")]; + bool var_29117_interleave_0 = const()[name = string("op_29117_interleave_0"), val = bool(false)]; + tensor var_29117_cast_fp16 = concat(axis = var_27668, interleave = var_29117_interleave_0, values = (var_28981_cast_fp16, var_28983_cast_fp16, var_28985_cast_fp16, var_28987_cast_fp16))[name = string("op_29117_cast_fp16")]; + bool var_29119_interleave_0 = const()[name = string("op_29119_interleave_0"), val = bool(false)]; + tensor var_29119_cast_fp16 = concat(axis = var_27668, interleave = var_29119_interleave_0, values = (var_28989_cast_fp16, var_28991_cast_fp16, var_28993_cast_fp16, var_28995_cast_fp16))[name = string("op_29119_cast_fp16")]; + bool var_29121_interleave_0 = const()[name = string("op_29121_interleave_0"), val = bool(false)]; + tensor var_29121_cast_fp16 = concat(axis = var_27668, interleave = var_29121_interleave_0, values = (var_28997_cast_fp16, var_28999_cast_fp16, var_29001_cast_fp16, var_29003_cast_fp16))[name = string("op_29121_cast_fp16")]; + bool var_29123_interleave_0 = const()[name = string("op_29123_interleave_0"), val = bool(false)]; + tensor var_29123_cast_fp16 = concat(axis = var_27668, interleave = var_29123_interleave_0, values = (var_29005_cast_fp16, var_29007_cast_fp16, var_29009_cast_fp16, var_29011_cast_fp16))[name = string("op_29123_cast_fp16")]; + bool var_29125_interleave_0 = const()[name = string("op_29125_interleave_0"), val = bool(false)]; + tensor var_29125_cast_fp16 = concat(axis = var_27668, interleave = var_29125_interleave_0, values = (var_29013_cast_fp16, var_29015_cast_fp16, var_29017_cast_fp16, var_29019_cast_fp16))[name = string("op_29125_cast_fp16")]; + bool var_29127_interleave_0 = const()[name = string("op_29127_interleave_0"), val = bool(false)]; + tensor var_29127_cast_fp16 = concat(axis = var_27668, interleave = var_29127_interleave_0, values = (var_29021_cast_fp16, var_29023_cast_fp16, var_29025_cast_fp16, var_29027_cast_fp16))[name = string("op_29127_cast_fp16")]; + bool var_29129_interleave_0 = const()[name = string("op_29129_interleave_0"), val = bool(false)]; + tensor var_29129_cast_fp16 = concat(axis = var_27668, interleave = var_29129_interleave_0, values = (var_29029_cast_fp16, var_29031_cast_fp16, var_29033_cast_fp16, var_29035_cast_fp16))[name = string("op_29129_cast_fp16")]; + bool var_29131_interleave_0 = const()[name = string("op_29131_interleave_0"), val = bool(false)]; + tensor var_29131_cast_fp16 = concat(axis = var_27668, interleave = var_29131_interleave_0, values = (var_29037_cast_fp16, var_29039_cast_fp16, var_29041_cast_fp16, var_29043_cast_fp16))[name = string("op_29131_cast_fp16")]; + bool var_29133_interleave_0 = const()[name = string("op_29133_interleave_0"), val = bool(false)]; + tensor var_29133_cast_fp16 = concat(axis = var_27668, interleave = var_29133_interleave_0, values = (var_29045_cast_fp16, var_29047_cast_fp16, var_29049_cast_fp16, var_29051_cast_fp16))[name = string("op_29133_cast_fp16")]; + bool var_29135_interleave_0 = const()[name = string("op_29135_interleave_0"), val = bool(false)]; + tensor var_29135_cast_fp16 = concat(axis = var_27668, interleave = var_29135_interleave_0, values = (var_29053_cast_fp16, var_29055_cast_fp16, var_29057_cast_fp16, var_29059_cast_fp16))[name = string("op_29135_cast_fp16")]; + bool var_29137_interleave_0 = const()[name = string("op_29137_interleave_0"), val = bool(false)]; + tensor var_29137_cast_fp16 = concat(axis = var_27668, interleave = var_29137_interleave_0, values = (var_29061_cast_fp16, var_29063_cast_fp16, var_29065_cast_fp16, var_29067_cast_fp16))[name = string("op_29137_cast_fp16")]; + bool var_29139_interleave_0 = const()[name = string("op_29139_interleave_0"), val = bool(false)]; + tensor var_29139_cast_fp16 = concat(axis = var_27668, interleave = var_29139_interleave_0, values = (var_29069_cast_fp16, var_29071_cast_fp16, var_29073_cast_fp16, var_29075_cast_fp16))[name = string("op_29139_cast_fp16")]; + bool var_29141_interleave_0 = const()[name = string("op_29141_interleave_0"), val = bool(false)]; + tensor var_29141_cast_fp16 = concat(axis = var_27668, interleave = var_29141_interleave_0, values = (var_29077_cast_fp16, var_29079_cast_fp16, var_29081_cast_fp16, var_29083_cast_fp16))[name = string("op_29141_cast_fp16")]; + bool var_29143_interleave_0 = const()[name = string("op_29143_interleave_0"), val = bool(false)]; + tensor var_29143_cast_fp16 = concat(axis = var_27668, interleave = var_29143_interleave_0, values = (var_29085_cast_fp16, var_29087_cast_fp16, var_29089_cast_fp16, var_29091_cast_fp16))[name = string("op_29143_cast_fp16")]; + bool var_29145_interleave_0 = const()[name = string("op_29145_interleave_0"), val = bool(false)]; + tensor var_29145_cast_fp16 = concat(axis = var_27668, interleave = var_29145_interleave_0, values = (var_29093_cast_fp16, var_29095_cast_fp16, var_29097_cast_fp16, var_29099_cast_fp16))[name = string("op_29145_cast_fp16")]; + bool var_29147_interleave_0 = const()[name = string("op_29147_interleave_0"), val = bool(false)]; + tensor var_29147_cast_fp16 = concat(axis = var_27668, interleave = var_29147_interleave_0, values = (var_29101_cast_fp16, var_29103_cast_fp16, var_29105_cast_fp16, var_29107_cast_fp16))[name = string("op_29147_cast_fp16")]; + bool input_145_interleave_0 = const()[name = string("input_145_interleave_0"), val = bool(false)]; + tensor input_145_cast_fp16 = concat(axis = var_27693, interleave = input_145_interleave_0, values = (var_29109_cast_fp16, var_29111_cast_fp16, var_29113_cast_fp16, var_29115_cast_fp16, var_29117_cast_fp16, var_29119_cast_fp16, var_29121_cast_fp16, var_29123_cast_fp16, var_29125_cast_fp16, var_29127_cast_fp16, var_29129_cast_fp16, var_29131_cast_fp16, var_29133_cast_fp16, var_29135_cast_fp16, var_29137_cast_fp16, var_29139_cast_fp16, var_29141_cast_fp16, var_29143_cast_fp16, var_29145_cast_fp16, var_29147_cast_fp16))[name = string("input_145_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732864320)))]; + tensor layers_18_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736141184)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_75_cast_fp16")]; + tensor out_75_axes_0 = const()[name = string("out_75_axes_0"), val = tensor([1])]; + fp16 var_29166_to_fp16 = const()[name = string("op_29166_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_29166_to_fp16, x = inputs_75_cast_fp16)[name = string("out_75_cast_fp16")]; + tensor input_147_gamma_0_to_fp16 = const()[name = string("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736143808)))]; + tensor input_147_beta_0_to_fp16 = const()[name = string("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736146432)))]; + fp16 input_147_epsilon_0_to_fp16 = const()[name = string("input_147_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = string("input_147_cast_fp16")]; + string input_149_pad_type_0 = const()[name = string("input_149_pad_type_0"), val = string("valid")]; + tensor input_149_strides_0 = const()[name = string("input_149_strides_0"), val = tensor([1, 1])]; + tensor input_149_pad_0 = const()[name = string("input_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_149_dilations_0 = const()[name = string("input_149_dilations_0"), val = tensor([1, 1])]; + int32 input_149_groups_0 = const()[name = string("input_149_groups_0"), val = int32(1)]; + tensor layers_18_fc1_weight_to_fp16 = const()[name = string("layers_18_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(736149056)))]; + tensor layers_18_fc1_bias_to_fp16 = const()[name = string("layers_18_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749256320)))]; + tensor input_149_cast_fp16 = conv(bias = layers_18_fc1_bias_to_fp16, dilations = input_149_dilations_0, groups = input_149_groups_0, pad = input_149_pad_0, pad_type = input_149_pad_type_0, strides = input_149_strides_0, weight = layers_18_fc1_weight_to_fp16, x = input_147_cast_fp16)[name = string("input_149_cast_fp16")]; + string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("EXACT")]; + tensor input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")]; + string hidden_states_41_pad_type_0 = const()[name = string("hidden_states_41_pad_type_0"), val = string("valid")]; + tensor hidden_states_41_strides_0 = const()[name = string("hidden_states_41_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_41_pad_0 = const()[name = string("hidden_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_41_dilations_0 = const()[name = string("hidden_states_41_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_41_groups_0 = const()[name = string("hidden_states_41_groups_0"), val = int32(1)]; + tensor layers_18_fc2_weight_to_fp16 = const()[name = string("layers_18_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(749266624)))]; + tensor layers_18_fc2_bias_to_fp16 = const()[name = string("layers_18_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762373888)))]; + tensor hidden_states_41_cast_fp16 = conv(bias = layers_18_fc2_bias_to_fp16, dilations = hidden_states_41_dilations_0, groups = hidden_states_41_groups_0, pad = hidden_states_41_pad_0, pad_type = hidden_states_41_pad_type_0, strides = hidden_states_41_strides_0, weight = layers_18_fc2_weight_to_fp16, x = input_151_cast_fp16)[name = string("hidden_states_41_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = hidden_states_41_cast_fp16)[name = string("inputs_77_cast_fp16")]; + int32 var_29195 = const()[name = string("op_29195"), val = int32(3)]; + int32 var_29220 = const()[name = string("op_29220"), val = int32(1)]; + tensor out_77_axes_0 = const()[name = string("out_77_axes_0"), val = tensor([1])]; + fp16 var_29237_to_fp16 = const()[name = string("op_29237_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_29237_to_fp16, x = inputs_77_cast_fp16)[name = string("out_77_cast_fp16")]; + tensor obj_77_gamma_0_to_fp16 = const()[name = string("obj_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762376512)))]; + tensor obj_77_beta_0_to_fp16 = const()[name = string("obj_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762379136)))]; + fp16 obj_77_epsilon_0_to_fp16 = const()[name = string("obj_77_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_77_cast_fp16 = batch_norm(beta = obj_77_beta_0_to_fp16, epsilon = obj_77_epsilon_0_to_fp16, gamma = obj_77_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = string("obj_77_cast_fp16")]; + string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; + tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; + tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; + int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(762381760)))]; + tensor layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765658624)))]; + tensor query_39_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("query_39_cast_fp16")]; + string key_39_pad_type_0 = const()[name = string("key_39_pad_type_0"), val = string("valid")]; + tensor key_39_strides_0 = const()[name = string("key_39_strides_0"), val = tensor([1, 1])]; + tensor key_39_pad_0 = const()[name = string("key_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_39_dilations_0 = const()[name = string("key_39_dilations_0"), val = tensor([1, 1])]; + int32 key_39_groups_0 = const()[name = string("key_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765661248)))]; + tensor key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("key_39_cast_fp16")]; + string value_39_pad_type_0 = const()[name = string("value_39_pad_type_0"), val = string("valid")]; + tensor value_39_strides_0 = const()[name = string("value_39_strides_0"), val = tensor([1, 1])]; + tensor value_39_pad_0 = const()[name = string("value_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_39_dilations_0 = const()[name = string("value_39_dilations_0"), val = tensor([1, 1])]; + int32 value_39_groups_0 = const()[name = string("value_39_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(768938112)))]; + tensor layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772214976)))]; + tensor value_39_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = obj_77_cast_fp16)[name = string("value_39_cast_fp16")]; + tensor var_29275_begin_0 = const()[name = string("op_29275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29275_end_0 = const()[name = string("op_29275_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29275_end_mask_0 = const()[name = string("op_29275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29275_cast_fp16 = slice_by_index(begin = var_29275_begin_0, end = var_29275_end_0, end_mask = var_29275_end_mask_0, x = query_39_cast_fp16)[name = string("op_29275_cast_fp16")]; + tensor var_29279_begin_0 = const()[name = string("op_29279_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29279_end_0 = const()[name = string("op_29279_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_29279_end_mask_0 = const()[name = string("op_29279_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29279_cast_fp16 = slice_by_index(begin = var_29279_begin_0, end = var_29279_end_0, end_mask = var_29279_end_mask_0, x = query_39_cast_fp16)[name = string("op_29279_cast_fp16")]; + tensor var_29283_begin_0 = const()[name = string("op_29283_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29283_end_0 = const()[name = string("op_29283_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_29283_end_mask_0 = const()[name = string("op_29283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29283_cast_fp16 = slice_by_index(begin = var_29283_begin_0, end = var_29283_end_0, end_mask = var_29283_end_mask_0, x = query_39_cast_fp16)[name = string("op_29283_cast_fp16")]; + tensor var_29287_begin_0 = const()[name = string("op_29287_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29287_end_0 = const()[name = string("op_29287_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_29287_end_mask_0 = const()[name = string("op_29287_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29287_cast_fp16 = slice_by_index(begin = var_29287_begin_0, end = var_29287_end_0, end_mask = var_29287_end_mask_0, x = query_39_cast_fp16)[name = string("op_29287_cast_fp16")]; + tensor var_29291_begin_0 = const()[name = string("op_29291_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29291_end_0 = const()[name = string("op_29291_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_29291_end_mask_0 = const()[name = string("op_29291_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29291_cast_fp16 = slice_by_index(begin = var_29291_begin_0, end = var_29291_end_0, end_mask = var_29291_end_mask_0, x = query_39_cast_fp16)[name = string("op_29291_cast_fp16")]; + tensor var_29295_begin_0 = const()[name = string("op_29295_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29295_end_0 = const()[name = string("op_29295_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_29295_end_mask_0 = const()[name = string("op_29295_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29295_cast_fp16 = slice_by_index(begin = var_29295_begin_0, end = var_29295_end_0, end_mask = var_29295_end_mask_0, x = query_39_cast_fp16)[name = string("op_29295_cast_fp16")]; + tensor var_29299_begin_0 = const()[name = string("op_29299_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29299_end_0 = const()[name = string("op_29299_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_29299_end_mask_0 = const()[name = string("op_29299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29299_cast_fp16 = slice_by_index(begin = var_29299_begin_0, end = var_29299_end_0, end_mask = var_29299_end_mask_0, x = query_39_cast_fp16)[name = string("op_29299_cast_fp16")]; + tensor var_29303_begin_0 = const()[name = string("op_29303_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29303_end_0 = const()[name = string("op_29303_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_29303_end_mask_0 = const()[name = string("op_29303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29303_cast_fp16 = slice_by_index(begin = var_29303_begin_0, end = var_29303_end_0, end_mask = var_29303_end_mask_0, x = query_39_cast_fp16)[name = string("op_29303_cast_fp16")]; + tensor var_29307_begin_0 = const()[name = string("op_29307_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29307_end_0 = const()[name = string("op_29307_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_29307_end_mask_0 = const()[name = string("op_29307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29307_cast_fp16 = slice_by_index(begin = var_29307_begin_0, end = var_29307_end_0, end_mask = var_29307_end_mask_0, x = query_39_cast_fp16)[name = string("op_29307_cast_fp16")]; + tensor var_29311_begin_0 = const()[name = string("op_29311_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29311_end_0 = const()[name = string("op_29311_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_29311_end_mask_0 = const()[name = string("op_29311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29311_cast_fp16 = slice_by_index(begin = var_29311_begin_0, end = var_29311_end_0, end_mask = var_29311_end_mask_0, x = query_39_cast_fp16)[name = string("op_29311_cast_fp16")]; + tensor var_29315_begin_0 = const()[name = string("op_29315_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29315_end_0 = const()[name = string("op_29315_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_29315_end_mask_0 = const()[name = string("op_29315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29315_cast_fp16 = slice_by_index(begin = var_29315_begin_0, end = var_29315_end_0, end_mask = var_29315_end_mask_0, x = query_39_cast_fp16)[name = string("op_29315_cast_fp16")]; + tensor var_29319_begin_0 = const()[name = string("op_29319_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29319_end_0 = const()[name = string("op_29319_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_29319_end_mask_0 = const()[name = string("op_29319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29319_cast_fp16 = slice_by_index(begin = var_29319_begin_0, end = var_29319_end_0, end_mask = var_29319_end_mask_0, x = query_39_cast_fp16)[name = string("op_29319_cast_fp16")]; + tensor var_29323_begin_0 = const()[name = string("op_29323_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29323_end_0 = const()[name = string("op_29323_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_29323_end_mask_0 = const()[name = string("op_29323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29323_cast_fp16 = slice_by_index(begin = var_29323_begin_0, end = var_29323_end_0, end_mask = var_29323_end_mask_0, x = query_39_cast_fp16)[name = string("op_29323_cast_fp16")]; + tensor var_29327_begin_0 = const()[name = string("op_29327_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29327_end_0 = const()[name = string("op_29327_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_29327_end_mask_0 = const()[name = string("op_29327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29327_cast_fp16 = slice_by_index(begin = var_29327_begin_0, end = var_29327_end_0, end_mask = var_29327_end_mask_0, x = query_39_cast_fp16)[name = string("op_29327_cast_fp16")]; + tensor var_29331_begin_0 = const()[name = string("op_29331_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29331_end_0 = const()[name = string("op_29331_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_29331_end_mask_0 = const()[name = string("op_29331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29331_cast_fp16 = slice_by_index(begin = var_29331_begin_0, end = var_29331_end_0, end_mask = var_29331_end_mask_0, x = query_39_cast_fp16)[name = string("op_29331_cast_fp16")]; + tensor var_29335_begin_0 = const()[name = string("op_29335_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29335_end_0 = const()[name = string("op_29335_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_29335_end_mask_0 = const()[name = string("op_29335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29335_cast_fp16 = slice_by_index(begin = var_29335_begin_0, end = var_29335_end_0, end_mask = var_29335_end_mask_0, x = query_39_cast_fp16)[name = string("op_29335_cast_fp16")]; + tensor var_29339_begin_0 = const()[name = string("op_29339_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29339_end_0 = const()[name = string("op_29339_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_29339_end_mask_0 = const()[name = string("op_29339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29339_cast_fp16 = slice_by_index(begin = var_29339_begin_0, end = var_29339_end_0, end_mask = var_29339_end_mask_0, x = query_39_cast_fp16)[name = string("op_29339_cast_fp16")]; + tensor var_29343_begin_0 = const()[name = string("op_29343_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29343_end_0 = const()[name = string("op_29343_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_29343_end_mask_0 = const()[name = string("op_29343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29343_cast_fp16 = slice_by_index(begin = var_29343_begin_0, end = var_29343_end_0, end_mask = var_29343_end_mask_0, x = query_39_cast_fp16)[name = string("op_29343_cast_fp16")]; + tensor var_29347_begin_0 = const()[name = string("op_29347_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29347_end_0 = const()[name = string("op_29347_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_29347_end_mask_0 = const()[name = string("op_29347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29347_cast_fp16 = slice_by_index(begin = var_29347_begin_0, end = var_29347_end_0, end_mask = var_29347_end_mask_0, x = query_39_cast_fp16)[name = string("op_29347_cast_fp16")]; + tensor var_29351_begin_0 = const()[name = string("op_29351_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29351_end_0 = const()[name = string("op_29351_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_29351_end_mask_0 = const()[name = string("op_29351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29351_cast_fp16 = slice_by_index(begin = var_29351_begin_0, end = var_29351_end_0, end_mask = var_29351_end_mask_0, x = query_39_cast_fp16)[name = string("op_29351_cast_fp16")]; + tensor var_29360_begin_0 = const()[name = string("op_29360_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29360_end_0 = const()[name = string("op_29360_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29360_end_mask_0 = const()[name = string("op_29360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29360_cast_fp16 = slice_by_index(begin = var_29360_begin_0, end = var_29360_end_0, end_mask = var_29360_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29360_cast_fp16")]; + tensor var_29367_begin_0 = const()[name = string("op_29367_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29367_end_0 = const()[name = string("op_29367_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29367_end_mask_0 = const()[name = string("op_29367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29367_cast_fp16 = slice_by_index(begin = var_29367_begin_0, end = var_29367_end_0, end_mask = var_29367_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29367_cast_fp16")]; + tensor var_29374_begin_0 = const()[name = string("op_29374_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29374_end_0 = const()[name = string("op_29374_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29374_end_mask_0 = const()[name = string("op_29374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29374_cast_fp16 = slice_by_index(begin = var_29374_begin_0, end = var_29374_end_0, end_mask = var_29374_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29374_cast_fp16")]; + tensor var_29381_begin_0 = const()[name = string("op_29381_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29381_end_0 = const()[name = string("op_29381_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29381_end_mask_0 = const()[name = string("op_29381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29381_cast_fp16 = slice_by_index(begin = var_29381_begin_0, end = var_29381_end_0, end_mask = var_29381_end_mask_0, x = var_29275_cast_fp16)[name = string("op_29381_cast_fp16")]; + tensor var_29388_begin_0 = const()[name = string("op_29388_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29388_end_0 = const()[name = string("op_29388_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29388_end_mask_0 = const()[name = string("op_29388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29388_cast_fp16 = slice_by_index(begin = var_29388_begin_0, end = var_29388_end_0, end_mask = var_29388_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29388_cast_fp16")]; + tensor var_29395_begin_0 = const()[name = string("op_29395_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29395_end_0 = const()[name = string("op_29395_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29395_end_mask_0 = const()[name = string("op_29395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29395_cast_fp16 = slice_by_index(begin = var_29395_begin_0, end = var_29395_end_0, end_mask = var_29395_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29395_cast_fp16")]; + tensor var_29402_begin_0 = const()[name = string("op_29402_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29402_end_0 = const()[name = string("op_29402_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29402_end_mask_0 = const()[name = string("op_29402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29402_cast_fp16 = slice_by_index(begin = var_29402_begin_0, end = var_29402_end_0, end_mask = var_29402_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29402_cast_fp16")]; + tensor var_29409_begin_0 = const()[name = string("op_29409_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29409_end_0 = const()[name = string("op_29409_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29409_end_mask_0 = const()[name = string("op_29409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29409_cast_fp16 = slice_by_index(begin = var_29409_begin_0, end = var_29409_end_0, end_mask = var_29409_end_mask_0, x = var_29279_cast_fp16)[name = string("op_29409_cast_fp16")]; + tensor var_29416_begin_0 = const()[name = string("op_29416_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29416_end_0 = const()[name = string("op_29416_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29416_end_mask_0 = const()[name = string("op_29416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29416_cast_fp16 = slice_by_index(begin = var_29416_begin_0, end = var_29416_end_0, end_mask = var_29416_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29416_cast_fp16")]; + tensor var_29423_begin_0 = const()[name = string("op_29423_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29423_end_0 = const()[name = string("op_29423_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29423_end_mask_0 = const()[name = string("op_29423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29423_cast_fp16 = slice_by_index(begin = var_29423_begin_0, end = var_29423_end_0, end_mask = var_29423_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29423_cast_fp16")]; + tensor var_29430_begin_0 = const()[name = string("op_29430_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29430_end_0 = const()[name = string("op_29430_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29430_end_mask_0 = const()[name = string("op_29430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29430_cast_fp16 = slice_by_index(begin = var_29430_begin_0, end = var_29430_end_0, end_mask = var_29430_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29430_cast_fp16")]; + tensor var_29437_begin_0 = const()[name = string("op_29437_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29437_end_0 = const()[name = string("op_29437_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29437_end_mask_0 = const()[name = string("op_29437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29437_cast_fp16 = slice_by_index(begin = var_29437_begin_0, end = var_29437_end_0, end_mask = var_29437_end_mask_0, x = var_29283_cast_fp16)[name = string("op_29437_cast_fp16")]; + tensor var_29444_begin_0 = const()[name = string("op_29444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29444_end_0 = const()[name = string("op_29444_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29444_end_mask_0 = const()[name = string("op_29444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29444_cast_fp16 = slice_by_index(begin = var_29444_begin_0, end = var_29444_end_0, end_mask = var_29444_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29444_cast_fp16")]; + tensor var_29451_begin_0 = const()[name = string("op_29451_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29451_end_0 = const()[name = string("op_29451_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29451_end_mask_0 = const()[name = string("op_29451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29451_cast_fp16 = slice_by_index(begin = var_29451_begin_0, end = var_29451_end_0, end_mask = var_29451_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29451_cast_fp16")]; + tensor var_29458_begin_0 = const()[name = string("op_29458_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29458_end_0 = const()[name = string("op_29458_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29458_end_mask_0 = const()[name = string("op_29458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29458_cast_fp16 = slice_by_index(begin = var_29458_begin_0, end = var_29458_end_0, end_mask = var_29458_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29458_cast_fp16")]; + tensor var_29465_begin_0 = const()[name = string("op_29465_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29465_end_0 = const()[name = string("op_29465_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29465_end_mask_0 = const()[name = string("op_29465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29465_cast_fp16 = slice_by_index(begin = var_29465_begin_0, end = var_29465_end_0, end_mask = var_29465_end_mask_0, x = var_29287_cast_fp16)[name = string("op_29465_cast_fp16")]; + tensor var_29472_begin_0 = const()[name = string("op_29472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29472_end_0 = const()[name = string("op_29472_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29472_end_mask_0 = const()[name = string("op_29472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29472_cast_fp16 = slice_by_index(begin = var_29472_begin_0, end = var_29472_end_0, end_mask = var_29472_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29472_cast_fp16")]; + tensor var_29479_begin_0 = const()[name = string("op_29479_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29479_end_0 = const()[name = string("op_29479_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29479_end_mask_0 = const()[name = string("op_29479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29479_cast_fp16 = slice_by_index(begin = var_29479_begin_0, end = var_29479_end_0, end_mask = var_29479_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29479_cast_fp16")]; + tensor var_29486_begin_0 = const()[name = string("op_29486_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29486_end_0 = const()[name = string("op_29486_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29486_end_mask_0 = const()[name = string("op_29486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29486_cast_fp16 = slice_by_index(begin = var_29486_begin_0, end = var_29486_end_0, end_mask = var_29486_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29486_cast_fp16")]; + tensor var_29493_begin_0 = const()[name = string("op_29493_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29493_end_0 = const()[name = string("op_29493_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29493_end_mask_0 = const()[name = string("op_29493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29493_cast_fp16 = slice_by_index(begin = var_29493_begin_0, end = var_29493_end_0, end_mask = var_29493_end_mask_0, x = var_29291_cast_fp16)[name = string("op_29493_cast_fp16")]; + tensor var_29500_begin_0 = const()[name = string("op_29500_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29500_end_0 = const()[name = string("op_29500_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29500_end_mask_0 = const()[name = string("op_29500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29500_cast_fp16 = slice_by_index(begin = var_29500_begin_0, end = var_29500_end_0, end_mask = var_29500_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29500_cast_fp16")]; + tensor var_29507_begin_0 = const()[name = string("op_29507_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29507_end_0 = const()[name = string("op_29507_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29507_end_mask_0 = const()[name = string("op_29507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29507_cast_fp16 = slice_by_index(begin = var_29507_begin_0, end = var_29507_end_0, end_mask = var_29507_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29507_cast_fp16")]; + tensor var_29514_begin_0 = const()[name = string("op_29514_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29514_end_0 = const()[name = string("op_29514_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29514_end_mask_0 = const()[name = string("op_29514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29514_cast_fp16 = slice_by_index(begin = var_29514_begin_0, end = var_29514_end_0, end_mask = var_29514_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29514_cast_fp16")]; + tensor var_29521_begin_0 = const()[name = string("op_29521_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29521_end_0 = const()[name = string("op_29521_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29521_end_mask_0 = const()[name = string("op_29521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29521_cast_fp16 = slice_by_index(begin = var_29521_begin_0, end = var_29521_end_0, end_mask = var_29521_end_mask_0, x = var_29295_cast_fp16)[name = string("op_29521_cast_fp16")]; + tensor var_29528_begin_0 = const()[name = string("op_29528_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29528_end_0 = const()[name = string("op_29528_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29528_end_mask_0 = const()[name = string("op_29528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29528_cast_fp16 = slice_by_index(begin = var_29528_begin_0, end = var_29528_end_0, end_mask = var_29528_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29528_cast_fp16")]; + tensor var_29535_begin_0 = const()[name = string("op_29535_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29535_end_0 = const()[name = string("op_29535_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29535_end_mask_0 = const()[name = string("op_29535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29535_cast_fp16 = slice_by_index(begin = var_29535_begin_0, end = var_29535_end_0, end_mask = var_29535_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29535_cast_fp16")]; + tensor var_29542_begin_0 = const()[name = string("op_29542_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29542_end_0 = const()[name = string("op_29542_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29542_end_mask_0 = const()[name = string("op_29542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29542_cast_fp16 = slice_by_index(begin = var_29542_begin_0, end = var_29542_end_0, end_mask = var_29542_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29542_cast_fp16")]; + tensor var_29549_begin_0 = const()[name = string("op_29549_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29549_end_0 = const()[name = string("op_29549_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29549_end_mask_0 = const()[name = string("op_29549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29549_cast_fp16 = slice_by_index(begin = var_29549_begin_0, end = var_29549_end_0, end_mask = var_29549_end_mask_0, x = var_29299_cast_fp16)[name = string("op_29549_cast_fp16")]; + tensor var_29556_begin_0 = const()[name = string("op_29556_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29556_end_0 = const()[name = string("op_29556_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29556_end_mask_0 = const()[name = string("op_29556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29556_cast_fp16 = slice_by_index(begin = var_29556_begin_0, end = var_29556_end_0, end_mask = var_29556_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29556_cast_fp16")]; + tensor var_29563_begin_0 = const()[name = string("op_29563_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29563_end_0 = const()[name = string("op_29563_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29563_end_mask_0 = const()[name = string("op_29563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29563_cast_fp16 = slice_by_index(begin = var_29563_begin_0, end = var_29563_end_0, end_mask = var_29563_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29563_cast_fp16")]; + tensor var_29570_begin_0 = const()[name = string("op_29570_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29570_end_0 = const()[name = string("op_29570_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29570_end_mask_0 = const()[name = string("op_29570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29570_cast_fp16 = slice_by_index(begin = var_29570_begin_0, end = var_29570_end_0, end_mask = var_29570_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29570_cast_fp16")]; + tensor var_29577_begin_0 = const()[name = string("op_29577_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29577_end_0 = const()[name = string("op_29577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29577_end_mask_0 = const()[name = string("op_29577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29577_cast_fp16 = slice_by_index(begin = var_29577_begin_0, end = var_29577_end_0, end_mask = var_29577_end_mask_0, x = var_29303_cast_fp16)[name = string("op_29577_cast_fp16")]; + tensor var_29584_begin_0 = const()[name = string("op_29584_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29584_end_0 = const()[name = string("op_29584_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29584_end_mask_0 = const()[name = string("op_29584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29584_cast_fp16 = slice_by_index(begin = var_29584_begin_0, end = var_29584_end_0, end_mask = var_29584_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29584_cast_fp16")]; + tensor var_29591_begin_0 = const()[name = string("op_29591_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29591_end_0 = const()[name = string("op_29591_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29591_end_mask_0 = const()[name = string("op_29591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29591_cast_fp16 = slice_by_index(begin = var_29591_begin_0, end = var_29591_end_0, end_mask = var_29591_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29591_cast_fp16")]; + tensor var_29598_begin_0 = const()[name = string("op_29598_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29598_end_0 = const()[name = string("op_29598_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29598_end_mask_0 = const()[name = string("op_29598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29598_cast_fp16 = slice_by_index(begin = var_29598_begin_0, end = var_29598_end_0, end_mask = var_29598_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29598_cast_fp16")]; + tensor var_29605_begin_0 = const()[name = string("op_29605_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29605_end_0 = const()[name = string("op_29605_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29605_end_mask_0 = const()[name = string("op_29605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29605_cast_fp16 = slice_by_index(begin = var_29605_begin_0, end = var_29605_end_0, end_mask = var_29605_end_mask_0, x = var_29307_cast_fp16)[name = string("op_29605_cast_fp16")]; + tensor var_29612_begin_0 = const()[name = string("op_29612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29612_end_0 = const()[name = string("op_29612_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29612_end_mask_0 = const()[name = string("op_29612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29612_cast_fp16 = slice_by_index(begin = var_29612_begin_0, end = var_29612_end_0, end_mask = var_29612_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29612_cast_fp16")]; + tensor var_29619_begin_0 = const()[name = string("op_29619_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29619_end_0 = const()[name = string("op_29619_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29619_end_mask_0 = const()[name = string("op_29619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29619_cast_fp16 = slice_by_index(begin = var_29619_begin_0, end = var_29619_end_0, end_mask = var_29619_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29619_cast_fp16")]; + tensor var_29626_begin_0 = const()[name = string("op_29626_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29626_end_0 = const()[name = string("op_29626_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29626_end_mask_0 = const()[name = string("op_29626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29626_cast_fp16 = slice_by_index(begin = var_29626_begin_0, end = var_29626_end_0, end_mask = var_29626_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29626_cast_fp16")]; + tensor var_29633_begin_0 = const()[name = string("op_29633_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29633_end_0 = const()[name = string("op_29633_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29633_end_mask_0 = const()[name = string("op_29633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29633_cast_fp16 = slice_by_index(begin = var_29633_begin_0, end = var_29633_end_0, end_mask = var_29633_end_mask_0, x = var_29311_cast_fp16)[name = string("op_29633_cast_fp16")]; + tensor var_29640_begin_0 = const()[name = string("op_29640_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29640_end_0 = const()[name = string("op_29640_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29640_end_mask_0 = const()[name = string("op_29640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29640_cast_fp16 = slice_by_index(begin = var_29640_begin_0, end = var_29640_end_0, end_mask = var_29640_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29640_cast_fp16")]; + tensor var_29647_begin_0 = const()[name = string("op_29647_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29647_end_0 = const()[name = string("op_29647_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29647_end_mask_0 = const()[name = string("op_29647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29647_cast_fp16 = slice_by_index(begin = var_29647_begin_0, end = var_29647_end_0, end_mask = var_29647_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29647_cast_fp16")]; + tensor var_29654_begin_0 = const()[name = string("op_29654_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29654_end_0 = const()[name = string("op_29654_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29654_end_mask_0 = const()[name = string("op_29654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29654_cast_fp16 = slice_by_index(begin = var_29654_begin_0, end = var_29654_end_0, end_mask = var_29654_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29654_cast_fp16")]; + tensor var_29661_begin_0 = const()[name = string("op_29661_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29661_end_0 = const()[name = string("op_29661_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29661_end_mask_0 = const()[name = string("op_29661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29661_cast_fp16 = slice_by_index(begin = var_29661_begin_0, end = var_29661_end_0, end_mask = var_29661_end_mask_0, x = var_29315_cast_fp16)[name = string("op_29661_cast_fp16")]; + tensor var_29668_begin_0 = const()[name = string("op_29668_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29668_end_0 = const()[name = string("op_29668_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29668_end_mask_0 = const()[name = string("op_29668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29668_cast_fp16 = slice_by_index(begin = var_29668_begin_0, end = var_29668_end_0, end_mask = var_29668_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29668_cast_fp16")]; + tensor var_29675_begin_0 = const()[name = string("op_29675_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29675_end_0 = const()[name = string("op_29675_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29675_end_mask_0 = const()[name = string("op_29675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29675_cast_fp16 = slice_by_index(begin = var_29675_begin_0, end = var_29675_end_0, end_mask = var_29675_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29675_cast_fp16")]; + tensor var_29682_begin_0 = const()[name = string("op_29682_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29682_end_0 = const()[name = string("op_29682_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29682_end_mask_0 = const()[name = string("op_29682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29682_cast_fp16 = slice_by_index(begin = var_29682_begin_0, end = var_29682_end_0, end_mask = var_29682_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29682_cast_fp16")]; + tensor var_29689_begin_0 = const()[name = string("op_29689_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29689_end_0 = const()[name = string("op_29689_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29689_end_mask_0 = const()[name = string("op_29689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29689_cast_fp16 = slice_by_index(begin = var_29689_begin_0, end = var_29689_end_0, end_mask = var_29689_end_mask_0, x = var_29319_cast_fp16)[name = string("op_29689_cast_fp16")]; + tensor var_29696_begin_0 = const()[name = string("op_29696_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29696_end_0 = const()[name = string("op_29696_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29696_end_mask_0 = const()[name = string("op_29696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29696_cast_fp16 = slice_by_index(begin = var_29696_begin_0, end = var_29696_end_0, end_mask = var_29696_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29696_cast_fp16")]; + tensor var_29703_begin_0 = const()[name = string("op_29703_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29703_end_0 = const()[name = string("op_29703_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29703_end_mask_0 = const()[name = string("op_29703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29703_cast_fp16 = slice_by_index(begin = var_29703_begin_0, end = var_29703_end_0, end_mask = var_29703_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29703_cast_fp16")]; + tensor var_29710_begin_0 = const()[name = string("op_29710_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29710_end_0 = const()[name = string("op_29710_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29710_end_mask_0 = const()[name = string("op_29710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29710_cast_fp16 = slice_by_index(begin = var_29710_begin_0, end = var_29710_end_0, end_mask = var_29710_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29710_cast_fp16")]; + tensor var_29717_begin_0 = const()[name = string("op_29717_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29717_end_0 = const()[name = string("op_29717_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29717_end_mask_0 = const()[name = string("op_29717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29717_cast_fp16 = slice_by_index(begin = var_29717_begin_0, end = var_29717_end_0, end_mask = var_29717_end_mask_0, x = var_29323_cast_fp16)[name = string("op_29717_cast_fp16")]; + tensor var_29724_begin_0 = const()[name = string("op_29724_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29724_end_0 = const()[name = string("op_29724_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29724_end_mask_0 = const()[name = string("op_29724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29724_cast_fp16 = slice_by_index(begin = var_29724_begin_0, end = var_29724_end_0, end_mask = var_29724_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29724_cast_fp16")]; + tensor var_29731_begin_0 = const()[name = string("op_29731_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29731_end_0 = const()[name = string("op_29731_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29731_end_mask_0 = const()[name = string("op_29731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29731_cast_fp16 = slice_by_index(begin = var_29731_begin_0, end = var_29731_end_0, end_mask = var_29731_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29731_cast_fp16")]; + tensor var_29738_begin_0 = const()[name = string("op_29738_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29738_end_0 = const()[name = string("op_29738_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29738_end_mask_0 = const()[name = string("op_29738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29738_cast_fp16 = slice_by_index(begin = var_29738_begin_0, end = var_29738_end_0, end_mask = var_29738_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29738_cast_fp16")]; + tensor var_29745_begin_0 = const()[name = string("op_29745_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29745_end_0 = const()[name = string("op_29745_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29745_end_mask_0 = const()[name = string("op_29745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29745_cast_fp16 = slice_by_index(begin = var_29745_begin_0, end = var_29745_end_0, end_mask = var_29745_end_mask_0, x = var_29327_cast_fp16)[name = string("op_29745_cast_fp16")]; + tensor var_29752_begin_0 = const()[name = string("op_29752_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29752_end_0 = const()[name = string("op_29752_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29752_end_mask_0 = const()[name = string("op_29752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29752_cast_fp16 = slice_by_index(begin = var_29752_begin_0, end = var_29752_end_0, end_mask = var_29752_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29752_cast_fp16")]; + tensor var_29759_begin_0 = const()[name = string("op_29759_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29759_end_0 = const()[name = string("op_29759_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29759_end_mask_0 = const()[name = string("op_29759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29759_cast_fp16 = slice_by_index(begin = var_29759_begin_0, end = var_29759_end_0, end_mask = var_29759_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29759_cast_fp16")]; + tensor var_29766_begin_0 = const()[name = string("op_29766_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29766_end_0 = const()[name = string("op_29766_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29766_end_mask_0 = const()[name = string("op_29766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29766_cast_fp16 = slice_by_index(begin = var_29766_begin_0, end = var_29766_end_0, end_mask = var_29766_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29766_cast_fp16")]; + tensor var_29773_begin_0 = const()[name = string("op_29773_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29773_end_0 = const()[name = string("op_29773_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29773_end_mask_0 = const()[name = string("op_29773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29773_cast_fp16 = slice_by_index(begin = var_29773_begin_0, end = var_29773_end_0, end_mask = var_29773_end_mask_0, x = var_29331_cast_fp16)[name = string("op_29773_cast_fp16")]; + tensor var_29780_begin_0 = const()[name = string("op_29780_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29780_end_0 = const()[name = string("op_29780_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29780_end_mask_0 = const()[name = string("op_29780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29780_cast_fp16 = slice_by_index(begin = var_29780_begin_0, end = var_29780_end_0, end_mask = var_29780_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29780_cast_fp16")]; + tensor var_29787_begin_0 = const()[name = string("op_29787_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29787_end_0 = const()[name = string("op_29787_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29787_end_mask_0 = const()[name = string("op_29787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29787_cast_fp16 = slice_by_index(begin = var_29787_begin_0, end = var_29787_end_0, end_mask = var_29787_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29787_cast_fp16")]; + tensor var_29794_begin_0 = const()[name = string("op_29794_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29794_end_0 = const()[name = string("op_29794_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29794_end_mask_0 = const()[name = string("op_29794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29794_cast_fp16 = slice_by_index(begin = var_29794_begin_0, end = var_29794_end_0, end_mask = var_29794_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29794_cast_fp16")]; + tensor var_29801_begin_0 = const()[name = string("op_29801_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29801_end_0 = const()[name = string("op_29801_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29801_end_mask_0 = const()[name = string("op_29801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29801_cast_fp16 = slice_by_index(begin = var_29801_begin_0, end = var_29801_end_0, end_mask = var_29801_end_mask_0, x = var_29335_cast_fp16)[name = string("op_29801_cast_fp16")]; + tensor var_29808_begin_0 = const()[name = string("op_29808_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29808_end_0 = const()[name = string("op_29808_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29808_end_mask_0 = const()[name = string("op_29808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29808_cast_fp16 = slice_by_index(begin = var_29808_begin_0, end = var_29808_end_0, end_mask = var_29808_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29808_cast_fp16")]; + tensor var_29815_begin_0 = const()[name = string("op_29815_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29815_end_0 = const()[name = string("op_29815_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29815_end_mask_0 = const()[name = string("op_29815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29815_cast_fp16 = slice_by_index(begin = var_29815_begin_0, end = var_29815_end_0, end_mask = var_29815_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29815_cast_fp16")]; + tensor var_29822_begin_0 = const()[name = string("op_29822_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29822_end_0 = const()[name = string("op_29822_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29822_end_mask_0 = const()[name = string("op_29822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29822_cast_fp16 = slice_by_index(begin = var_29822_begin_0, end = var_29822_end_0, end_mask = var_29822_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29822_cast_fp16")]; + tensor var_29829_begin_0 = const()[name = string("op_29829_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29829_end_0 = const()[name = string("op_29829_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29829_end_mask_0 = const()[name = string("op_29829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29829_cast_fp16 = slice_by_index(begin = var_29829_begin_0, end = var_29829_end_0, end_mask = var_29829_end_mask_0, x = var_29339_cast_fp16)[name = string("op_29829_cast_fp16")]; + tensor var_29836_begin_0 = const()[name = string("op_29836_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29836_end_0 = const()[name = string("op_29836_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29836_end_mask_0 = const()[name = string("op_29836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29836_cast_fp16 = slice_by_index(begin = var_29836_begin_0, end = var_29836_end_0, end_mask = var_29836_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29836_cast_fp16")]; + tensor var_29843_begin_0 = const()[name = string("op_29843_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29843_end_0 = const()[name = string("op_29843_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29843_end_mask_0 = const()[name = string("op_29843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29843_cast_fp16 = slice_by_index(begin = var_29843_begin_0, end = var_29843_end_0, end_mask = var_29843_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29843_cast_fp16")]; + tensor var_29850_begin_0 = const()[name = string("op_29850_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29850_end_0 = const()[name = string("op_29850_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29850_end_mask_0 = const()[name = string("op_29850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29850_cast_fp16 = slice_by_index(begin = var_29850_begin_0, end = var_29850_end_0, end_mask = var_29850_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29850_cast_fp16")]; + tensor var_29857_begin_0 = const()[name = string("op_29857_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29857_end_0 = const()[name = string("op_29857_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29857_end_mask_0 = const()[name = string("op_29857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29857_cast_fp16 = slice_by_index(begin = var_29857_begin_0, end = var_29857_end_0, end_mask = var_29857_end_mask_0, x = var_29343_cast_fp16)[name = string("op_29857_cast_fp16")]; + tensor var_29864_begin_0 = const()[name = string("op_29864_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29864_end_0 = const()[name = string("op_29864_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29864_end_mask_0 = const()[name = string("op_29864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29864_cast_fp16 = slice_by_index(begin = var_29864_begin_0, end = var_29864_end_0, end_mask = var_29864_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29864_cast_fp16")]; + tensor var_29871_begin_0 = const()[name = string("op_29871_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29871_end_0 = const()[name = string("op_29871_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29871_end_mask_0 = const()[name = string("op_29871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29871_cast_fp16 = slice_by_index(begin = var_29871_begin_0, end = var_29871_end_0, end_mask = var_29871_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29871_cast_fp16")]; + tensor var_29878_begin_0 = const()[name = string("op_29878_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29878_end_0 = const()[name = string("op_29878_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29878_end_mask_0 = const()[name = string("op_29878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29878_cast_fp16 = slice_by_index(begin = var_29878_begin_0, end = var_29878_end_0, end_mask = var_29878_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29878_cast_fp16")]; + tensor var_29885_begin_0 = const()[name = string("op_29885_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29885_end_0 = const()[name = string("op_29885_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29885_end_mask_0 = const()[name = string("op_29885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29885_cast_fp16 = slice_by_index(begin = var_29885_begin_0, end = var_29885_end_0, end_mask = var_29885_end_mask_0, x = var_29347_cast_fp16)[name = string("op_29885_cast_fp16")]; + tensor var_29892_begin_0 = const()[name = string("op_29892_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29892_end_0 = const()[name = string("op_29892_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29892_end_mask_0 = const()[name = string("op_29892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29892_cast_fp16 = slice_by_index(begin = var_29892_begin_0, end = var_29892_end_0, end_mask = var_29892_end_mask_0, x = var_29351_cast_fp16)[name = string("op_29892_cast_fp16")]; + tensor var_29899_begin_0 = const()[name = string("op_29899_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29899_end_0 = const()[name = string("op_29899_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29899_end_mask_0 = const()[name = string("op_29899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29899_cast_fp16 = slice_by_index(begin = var_29899_begin_0, end = var_29899_end_0, end_mask = var_29899_end_mask_0, x = var_29351_cast_fp16)[name = string("op_29899_cast_fp16")]; + tensor var_29906_begin_0 = const()[name = string("op_29906_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29906_end_0 = const()[name = string("op_29906_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29906_end_mask_0 = const()[name = string("op_29906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29906_cast_fp16 = slice_by_index(begin = var_29906_begin_0, end = var_29906_end_0, end_mask = var_29906_end_mask_0, x = var_29351_cast_fp16)[name = string("op_29906_cast_fp16")]; + tensor var_29913_begin_0 = const()[name = string("op_29913_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29913_end_0 = const()[name = string("op_29913_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29913_end_mask_0 = const()[name = string("op_29913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29913_cast_fp16 = slice_by_index(begin = var_29913_begin_0, end = var_29913_end_0, end_mask = var_29913_end_mask_0, x = var_29351_cast_fp16)[name = string("op_29913_cast_fp16")]; + tensor k_39_perm_0 = const()[name = string("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_29918_begin_0 = const()[name = string("op_29918_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29918_end_0 = const()[name = string("op_29918_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_29918_end_mask_0 = const()[name = string("op_29918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = key_39_cast_fp16)[name = string("transpose_12")]; + tensor var_29918_cast_fp16 = slice_by_index(begin = var_29918_begin_0, end = var_29918_end_0, end_mask = var_29918_end_mask_0, x = k_39_cast_fp16)[name = string("op_29918_cast_fp16")]; + tensor var_29922_begin_0 = const()[name = string("op_29922_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_29922_end_0 = const()[name = string("op_29922_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_29922_end_mask_0 = const()[name = string("op_29922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29922_cast_fp16 = slice_by_index(begin = var_29922_begin_0, end = var_29922_end_0, end_mask = var_29922_end_mask_0, x = k_39_cast_fp16)[name = string("op_29922_cast_fp16")]; + tensor var_29926_begin_0 = const()[name = string("op_29926_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_29926_end_0 = const()[name = string("op_29926_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_29926_end_mask_0 = const()[name = string("op_29926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29926_cast_fp16 = slice_by_index(begin = var_29926_begin_0, end = var_29926_end_0, end_mask = var_29926_end_mask_0, x = k_39_cast_fp16)[name = string("op_29926_cast_fp16")]; + tensor var_29930_begin_0 = const()[name = string("op_29930_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_29930_end_0 = const()[name = string("op_29930_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_29930_end_mask_0 = const()[name = string("op_29930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29930_cast_fp16 = slice_by_index(begin = var_29930_begin_0, end = var_29930_end_0, end_mask = var_29930_end_mask_0, x = k_39_cast_fp16)[name = string("op_29930_cast_fp16")]; + tensor var_29934_begin_0 = const()[name = string("op_29934_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_29934_end_0 = const()[name = string("op_29934_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_29934_end_mask_0 = const()[name = string("op_29934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29934_cast_fp16 = slice_by_index(begin = var_29934_begin_0, end = var_29934_end_0, end_mask = var_29934_end_mask_0, x = k_39_cast_fp16)[name = string("op_29934_cast_fp16")]; + tensor var_29938_begin_0 = const()[name = string("op_29938_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_29938_end_0 = const()[name = string("op_29938_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_29938_end_mask_0 = const()[name = string("op_29938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29938_cast_fp16 = slice_by_index(begin = var_29938_begin_0, end = var_29938_end_0, end_mask = var_29938_end_mask_0, x = k_39_cast_fp16)[name = string("op_29938_cast_fp16")]; + tensor var_29942_begin_0 = const()[name = string("op_29942_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_29942_end_0 = const()[name = string("op_29942_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_29942_end_mask_0 = const()[name = string("op_29942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29942_cast_fp16 = slice_by_index(begin = var_29942_begin_0, end = var_29942_end_0, end_mask = var_29942_end_mask_0, x = k_39_cast_fp16)[name = string("op_29942_cast_fp16")]; + tensor var_29946_begin_0 = const()[name = string("op_29946_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_29946_end_0 = const()[name = string("op_29946_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_29946_end_mask_0 = const()[name = string("op_29946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29946_cast_fp16 = slice_by_index(begin = var_29946_begin_0, end = var_29946_end_0, end_mask = var_29946_end_mask_0, x = k_39_cast_fp16)[name = string("op_29946_cast_fp16")]; + tensor var_29950_begin_0 = const()[name = string("op_29950_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_29950_end_0 = const()[name = string("op_29950_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_29950_end_mask_0 = const()[name = string("op_29950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29950_cast_fp16 = slice_by_index(begin = var_29950_begin_0, end = var_29950_end_0, end_mask = var_29950_end_mask_0, x = k_39_cast_fp16)[name = string("op_29950_cast_fp16")]; + tensor var_29954_begin_0 = const()[name = string("op_29954_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_29954_end_0 = const()[name = string("op_29954_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_29954_end_mask_0 = const()[name = string("op_29954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29954_cast_fp16 = slice_by_index(begin = var_29954_begin_0, end = var_29954_end_0, end_mask = var_29954_end_mask_0, x = k_39_cast_fp16)[name = string("op_29954_cast_fp16")]; + tensor var_29958_begin_0 = const()[name = string("op_29958_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_29958_end_0 = const()[name = string("op_29958_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_29958_end_mask_0 = const()[name = string("op_29958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29958_cast_fp16 = slice_by_index(begin = var_29958_begin_0, end = var_29958_end_0, end_mask = var_29958_end_mask_0, x = k_39_cast_fp16)[name = string("op_29958_cast_fp16")]; + tensor var_29962_begin_0 = const()[name = string("op_29962_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_29962_end_0 = const()[name = string("op_29962_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_29962_end_mask_0 = const()[name = string("op_29962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29962_cast_fp16 = slice_by_index(begin = var_29962_begin_0, end = var_29962_end_0, end_mask = var_29962_end_mask_0, x = k_39_cast_fp16)[name = string("op_29962_cast_fp16")]; + tensor var_29966_begin_0 = const()[name = string("op_29966_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_29966_end_0 = const()[name = string("op_29966_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_29966_end_mask_0 = const()[name = string("op_29966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29966_cast_fp16 = slice_by_index(begin = var_29966_begin_0, end = var_29966_end_0, end_mask = var_29966_end_mask_0, x = k_39_cast_fp16)[name = string("op_29966_cast_fp16")]; + tensor var_29970_begin_0 = const()[name = string("op_29970_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_29970_end_0 = const()[name = string("op_29970_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_29970_end_mask_0 = const()[name = string("op_29970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29970_cast_fp16 = slice_by_index(begin = var_29970_begin_0, end = var_29970_end_0, end_mask = var_29970_end_mask_0, x = k_39_cast_fp16)[name = string("op_29970_cast_fp16")]; + tensor var_29974_begin_0 = const()[name = string("op_29974_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_29974_end_0 = const()[name = string("op_29974_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_29974_end_mask_0 = const()[name = string("op_29974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29974_cast_fp16 = slice_by_index(begin = var_29974_begin_0, end = var_29974_end_0, end_mask = var_29974_end_mask_0, x = k_39_cast_fp16)[name = string("op_29974_cast_fp16")]; + tensor var_29978_begin_0 = const()[name = string("op_29978_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_29978_end_0 = const()[name = string("op_29978_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_29978_end_mask_0 = const()[name = string("op_29978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29978_cast_fp16 = slice_by_index(begin = var_29978_begin_0, end = var_29978_end_0, end_mask = var_29978_end_mask_0, x = k_39_cast_fp16)[name = string("op_29978_cast_fp16")]; + tensor var_29982_begin_0 = const()[name = string("op_29982_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_29982_end_0 = const()[name = string("op_29982_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_29982_end_mask_0 = const()[name = string("op_29982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29982_cast_fp16 = slice_by_index(begin = var_29982_begin_0, end = var_29982_end_0, end_mask = var_29982_end_mask_0, x = k_39_cast_fp16)[name = string("op_29982_cast_fp16")]; + tensor var_29986_begin_0 = const()[name = string("op_29986_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_29986_end_0 = const()[name = string("op_29986_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_29986_end_mask_0 = const()[name = string("op_29986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29986_cast_fp16 = slice_by_index(begin = var_29986_begin_0, end = var_29986_end_0, end_mask = var_29986_end_mask_0, x = k_39_cast_fp16)[name = string("op_29986_cast_fp16")]; + tensor var_29990_begin_0 = const()[name = string("op_29990_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_29990_end_0 = const()[name = string("op_29990_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_29990_end_mask_0 = const()[name = string("op_29990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29990_cast_fp16 = slice_by_index(begin = var_29990_begin_0, end = var_29990_end_0, end_mask = var_29990_end_mask_0, x = k_39_cast_fp16)[name = string("op_29990_cast_fp16")]; + tensor var_29994_begin_0 = const()[name = string("op_29994_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_29994_end_0 = const()[name = string("op_29994_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_29994_end_mask_0 = const()[name = string("op_29994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29994_cast_fp16 = slice_by_index(begin = var_29994_begin_0, end = var_29994_end_0, end_mask = var_29994_end_mask_0, x = k_39_cast_fp16)[name = string("op_29994_cast_fp16")]; + tensor var_29996_begin_0 = const()[name = string("op_29996_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29996_end_0 = const()[name = string("op_29996_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29996_end_mask_0 = const()[name = string("op_29996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29996_cast_fp16 = slice_by_index(begin = var_29996_begin_0, end = var_29996_end_0, end_mask = var_29996_end_mask_0, x = value_39_cast_fp16)[name = string("op_29996_cast_fp16")]; + tensor var_30000_begin_0 = const()[name = string("op_30000_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30000_end_0 = const()[name = string("op_30000_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_30000_end_mask_0 = const()[name = string("op_30000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30000_cast_fp16 = slice_by_index(begin = var_30000_begin_0, end = var_30000_end_0, end_mask = var_30000_end_mask_0, x = value_39_cast_fp16)[name = string("op_30000_cast_fp16")]; + tensor var_30004_begin_0 = const()[name = string("op_30004_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30004_end_0 = const()[name = string("op_30004_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_30004_end_mask_0 = const()[name = string("op_30004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30004_cast_fp16 = slice_by_index(begin = var_30004_begin_0, end = var_30004_end_0, end_mask = var_30004_end_mask_0, x = value_39_cast_fp16)[name = string("op_30004_cast_fp16")]; + tensor var_30008_begin_0 = const()[name = string("op_30008_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30008_end_0 = const()[name = string("op_30008_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_30008_end_mask_0 = const()[name = string("op_30008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30008_cast_fp16 = slice_by_index(begin = var_30008_begin_0, end = var_30008_end_0, end_mask = var_30008_end_mask_0, x = value_39_cast_fp16)[name = string("op_30008_cast_fp16")]; + tensor var_30012_begin_0 = const()[name = string("op_30012_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30012_end_0 = const()[name = string("op_30012_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_30012_end_mask_0 = const()[name = string("op_30012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30012_cast_fp16 = slice_by_index(begin = var_30012_begin_0, end = var_30012_end_0, end_mask = var_30012_end_mask_0, x = value_39_cast_fp16)[name = string("op_30012_cast_fp16")]; + tensor var_30016_begin_0 = const()[name = string("op_30016_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30016_end_0 = const()[name = string("op_30016_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_30016_end_mask_0 = const()[name = string("op_30016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30016_cast_fp16 = slice_by_index(begin = var_30016_begin_0, end = var_30016_end_0, end_mask = var_30016_end_mask_0, x = value_39_cast_fp16)[name = string("op_30016_cast_fp16")]; + tensor var_30020_begin_0 = const()[name = string("op_30020_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30020_end_0 = const()[name = string("op_30020_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_30020_end_mask_0 = const()[name = string("op_30020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30020_cast_fp16 = slice_by_index(begin = var_30020_begin_0, end = var_30020_end_0, end_mask = var_30020_end_mask_0, x = value_39_cast_fp16)[name = string("op_30020_cast_fp16")]; + tensor var_30024_begin_0 = const()[name = string("op_30024_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30024_end_0 = const()[name = string("op_30024_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_30024_end_mask_0 = const()[name = string("op_30024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30024_cast_fp16 = slice_by_index(begin = var_30024_begin_0, end = var_30024_end_0, end_mask = var_30024_end_mask_0, x = value_39_cast_fp16)[name = string("op_30024_cast_fp16")]; + tensor var_30028_begin_0 = const()[name = string("op_30028_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30028_end_0 = const()[name = string("op_30028_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_30028_end_mask_0 = const()[name = string("op_30028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30028_cast_fp16 = slice_by_index(begin = var_30028_begin_0, end = var_30028_end_0, end_mask = var_30028_end_mask_0, x = value_39_cast_fp16)[name = string("op_30028_cast_fp16")]; + tensor var_30032_begin_0 = const()[name = string("op_30032_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30032_end_0 = const()[name = string("op_30032_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_30032_end_mask_0 = const()[name = string("op_30032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30032_cast_fp16 = slice_by_index(begin = var_30032_begin_0, end = var_30032_end_0, end_mask = var_30032_end_mask_0, x = value_39_cast_fp16)[name = string("op_30032_cast_fp16")]; + tensor var_30036_begin_0 = const()[name = string("op_30036_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30036_end_0 = const()[name = string("op_30036_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_30036_end_mask_0 = const()[name = string("op_30036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30036_cast_fp16 = slice_by_index(begin = var_30036_begin_0, end = var_30036_end_0, end_mask = var_30036_end_mask_0, x = value_39_cast_fp16)[name = string("op_30036_cast_fp16")]; + tensor var_30040_begin_0 = const()[name = string("op_30040_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30040_end_0 = const()[name = string("op_30040_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_30040_end_mask_0 = const()[name = string("op_30040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30040_cast_fp16 = slice_by_index(begin = var_30040_begin_0, end = var_30040_end_0, end_mask = var_30040_end_mask_0, x = value_39_cast_fp16)[name = string("op_30040_cast_fp16")]; + tensor var_30044_begin_0 = const()[name = string("op_30044_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30044_end_0 = const()[name = string("op_30044_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_30044_end_mask_0 = const()[name = string("op_30044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30044_cast_fp16 = slice_by_index(begin = var_30044_begin_0, end = var_30044_end_0, end_mask = var_30044_end_mask_0, x = value_39_cast_fp16)[name = string("op_30044_cast_fp16")]; + tensor var_30048_begin_0 = const()[name = string("op_30048_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30048_end_0 = const()[name = string("op_30048_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_30048_end_mask_0 = const()[name = string("op_30048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30048_cast_fp16 = slice_by_index(begin = var_30048_begin_0, end = var_30048_end_0, end_mask = var_30048_end_mask_0, x = value_39_cast_fp16)[name = string("op_30048_cast_fp16")]; + tensor var_30052_begin_0 = const()[name = string("op_30052_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30052_end_0 = const()[name = string("op_30052_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_30052_end_mask_0 = const()[name = string("op_30052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30052_cast_fp16 = slice_by_index(begin = var_30052_begin_0, end = var_30052_end_0, end_mask = var_30052_end_mask_0, x = value_39_cast_fp16)[name = string("op_30052_cast_fp16")]; + tensor var_30056_begin_0 = const()[name = string("op_30056_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30056_end_0 = const()[name = string("op_30056_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_30056_end_mask_0 = const()[name = string("op_30056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30056_cast_fp16 = slice_by_index(begin = var_30056_begin_0, end = var_30056_end_0, end_mask = var_30056_end_mask_0, x = value_39_cast_fp16)[name = string("op_30056_cast_fp16")]; + tensor var_30060_begin_0 = const()[name = string("op_30060_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30060_end_0 = const()[name = string("op_30060_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_30060_end_mask_0 = const()[name = string("op_30060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30060_cast_fp16 = slice_by_index(begin = var_30060_begin_0, end = var_30060_end_0, end_mask = var_30060_end_mask_0, x = value_39_cast_fp16)[name = string("op_30060_cast_fp16")]; + tensor var_30064_begin_0 = const()[name = string("op_30064_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30064_end_0 = const()[name = string("op_30064_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_30064_end_mask_0 = const()[name = string("op_30064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30064_cast_fp16 = slice_by_index(begin = var_30064_begin_0, end = var_30064_end_0, end_mask = var_30064_end_mask_0, x = value_39_cast_fp16)[name = string("op_30064_cast_fp16")]; + tensor var_30068_begin_0 = const()[name = string("op_30068_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30068_end_0 = const()[name = string("op_30068_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_30068_end_mask_0 = const()[name = string("op_30068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30068_cast_fp16 = slice_by_index(begin = var_30068_begin_0, end = var_30068_end_0, end_mask = var_30068_end_mask_0, x = value_39_cast_fp16)[name = string("op_30068_cast_fp16")]; + tensor var_30072_begin_0 = const()[name = string("op_30072_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30072_end_0 = const()[name = string("op_30072_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_30072_end_mask_0 = const()[name = string("op_30072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30072_cast_fp16 = slice_by_index(begin = var_30072_begin_0, end = var_30072_end_0, end_mask = var_30072_end_mask_0, x = value_39_cast_fp16)[name = string("op_30072_cast_fp16")]; + string _SplitHeadsQ__mh_w_3041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3041_equation_0, values = (var_29918_cast_fp16, var_29360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3041_cast_fp16")]; + string _SplitHeadsQ__mh_w_3043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3043_equation_0, values = (var_29918_cast_fp16, var_29367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3043_cast_fp16")]; + string _SplitHeadsQ__mh_w_3045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3045_equation_0, values = (var_29918_cast_fp16, var_29374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3045_cast_fp16")]; + string _SplitHeadsQ__mh_w_3047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3047_equation_0, values = (var_29918_cast_fp16, var_29381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3047_cast_fp16")]; + string _SplitHeadsQ__mh_w_3049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3049_equation_0, values = (var_29922_cast_fp16, var_29388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3049_cast_fp16")]; + string _SplitHeadsQ__mh_w_3051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3051_equation_0, values = (var_29922_cast_fp16, var_29395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3051_cast_fp16")]; + string _SplitHeadsQ__mh_w_3053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3053_equation_0, values = (var_29922_cast_fp16, var_29402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3053_cast_fp16")]; + string _SplitHeadsQ__mh_w_3055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3055_equation_0, values = (var_29922_cast_fp16, var_29409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3055_cast_fp16")]; + string _SplitHeadsQ__mh_w_3057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3057_equation_0, values = (var_29926_cast_fp16, var_29416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3057_cast_fp16")]; + string _SplitHeadsQ__mh_w_3059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3059_equation_0, values = (var_29926_cast_fp16, var_29423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3059_cast_fp16")]; + string _SplitHeadsQ__mh_w_3061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3061_equation_0, values = (var_29926_cast_fp16, var_29430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3061_cast_fp16")]; + string _SplitHeadsQ__mh_w_3063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3063_equation_0, values = (var_29926_cast_fp16, var_29437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3063_cast_fp16")]; + string _SplitHeadsQ__mh_w_3065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3065_equation_0, values = (var_29930_cast_fp16, var_29444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3065_cast_fp16")]; + string _SplitHeadsQ__mh_w_3067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3067_equation_0, values = (var_29930_cast_fp16, var_29451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3067_cast_fp16")]; + string _SplitHeadsQ__mh_w_3069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3069_equation_0, values = (var_29930_cast_fp16, var_29458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3069_cast_fp16")]; + string _SplitHeadsQ__mh_w_3071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3071_equation_0, values = (var_29930_cast_fp16, var_29465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3071_cast_fp16")]; + string _SplitHeadsQ__mh_w_3073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3073_equation_0, values = (var_29934_cast_fp16, var_29472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3073_cast_fp16")]; + string _SplitHeadsQ__mh_w_3075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3075_equation_0, values = (var_29934_cast_fp16, var_29479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3075_cast_fp16")]; + string _SplitHeadsQ__mh_w_3077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3077_equation_0, values = (var_29934_cast_fp16, var_29486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3077_cast_fp16")]; + string _SplitHeadsQ__mh_w_3079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3079_equation_0, values = (var_29934_cast_fp16, var_29493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3079_cast_fp16")]; + string _SplitHeadsQ__mh_w_3081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3081_equation_0, values = (var_29938_cast_fp16, var_29500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3081_cast_fp16")]; + string _SplitHeadsQ__mh_w_3083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3083_equation_0, values = (var_29938_cast_fp16, var_29507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3083_cast_fp16")]; + string _SplitHeadsQ__mh_w_3085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3085_equation_0, values = (var_29938_cast_fp16, var_29514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3085_cast_fp16")]; + string _SplitHeadsQ__mh_w_3087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3087_equation_0, values = (var_29938_cast_fp16, var_29521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3087_cast_fp16")]; + string _SplitHeadsQ__mh_w_3089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3089_equation_0, values = (var_29942_cast_fp16, var_29528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3089_cast_fp16")]; + string _SplitHeadsQ__mh_w_3091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3091_equation_0, values = (var_29942_cast_fp16, var_29535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3091_cast_fp16")]; + string _SplitHeadsQ__mh_w_3093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3093_equation_0, values = (var_29942_cast_fp16, var_29542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3093_cast_fp16")]; + string _SplitHeadsQ__mh_w_3095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3095_equation_0, values = (var_29942_cast_fp16, var_29549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3095_cast_fp16")]; + string _SplitHeadsQ__mh_w_3097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3097_equation_0, values = (var_29946_cast_fp16, var_29556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3097_cast_fp16")]; + string _SplitHeadsQ__mh_w_3099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3099_equation_0, values = (var_29946_cast_fp16, var_29563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3099_cast_fp16")]; + string _SplitHeadsQ__mh_w_3101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3101_equation_0, values = (var_29946_cast_fp16, var_29570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3101_cast_fp16")]; + string _SplitHeadsQ__mh_w_3103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3103_equation_0, values = (var_29946_cast_fp16, var_29577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3103_cast_fp16")]; + string _SplitHeadsQ__mh_w_3105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3105_equation_0, values = (var_29950_cast_fp16, var_29584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3105_cast_fp16")]; + string _SplitHeadsQ__mh_w_3107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3107_equation_0, values = (var_29950_cast_fp16, var_29591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3107_cast_fp16")]; + string _SplitHeadsQ__mh_w_3109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3109_equation_0, values = (var_29950_cast_fp16, var_29598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3109_cast_fp16")]; + string _SplitHeadsQ__mh_w_3111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3111_equation_0, values = (var_29950_cast_fp16, var_29605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3111_cast_fp16")]; + string _SplitHeadsQ__mh_w_3113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3113_equation_0, values = (var_29954_cast_fp16, var_29612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3113_cast_fp16")]; + string _SplitHeadsQ__mh_w_3115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3115_equation_0, values = (var_29954_cast_fp16, var_29619_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3115_cast_fp16")]; + string _SplitHeadsQ__mh_w_3117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3117_equation_0, values = (var_29954_cast_fp16, var_29626_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3117_cast_fp16")]; + string _SplitHeadsQ__mh_w_3119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3119_equation_0, values = (var_29954_cast_fp16, var_29633_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3119_cast_fp16")]; + string _SplitHeadsQ__mh_w_3121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3121_equation_0, values = (var_29958_cast_fp16, var_29640_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3121_cast_fp16")]; + string _SplitHeadsQ__mh_w_3123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3123_equation_0, values = (var_29958_cast_fp16, var_29647_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3123_cast_fp16")]; + string _SplitHeadsQ__mh_w_3125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3125_equation_0, values = (var_29958_cast_fp16, var_29654_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3125_cast_fp16")]; + string _SplitHeadsQ__mh_w_3127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3127_equation_0, values = (var_29958_cast_fp16, var_29661_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3127_cast_fp16")]; + string _SplitHeadsQ__mh_w_3129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3129_equation_0, values = (var_29962_cast_fp16, var_29668_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3129_cast_fp16")]; + string _SplitHeadsQ__mh_w_3131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3131_equation_0, values = (var_29962_cast_fp16, var_29675_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3131_cast_fp16")]; + string _SplitHeadsQ__mh_w_3133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3133_equation_0, values = (var_29962_cast_fp16, var_29682_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3133_cast_fp16")]; + string _SplitHeadsQ__mh_w_3135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3135_equation_0, values = (var_29962_cast_fp16, var_29689_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3135_cast_fp16")]; + string _SplitHeadsQ__mh_w_3137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3137_equation_0, values = (var_29966_cast_fp16, var_29696_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3137_cast_fp16")]; + string _SplitHeadsQ__mh_w_3139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3139_equation_0, values = (var_29966_cast_fp16, var_29703_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3139_cast_fp16")]; + string _SplitHeadsQ__mh_w_3141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3141_equation_0, values = (var_29966_cast_fp16, var_29710_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3141_cast_fp16")]; + string _SplitHeadsQ__mh_w_3143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3143_equation_0, values = (var_29966_cast_fp16, var_29717_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3143_cast_fp16")]; + string _SplitHeadsQ__mh_w_3145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3145_equation_0, values = (var_29970_cast_fp16, var_29724_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3145_cast_fp16")]; + string _SplitHeadsQ__mh_w_3147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3147_equation_0, values = (var_29970_cast_fp16, var_29731_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3147_cast_fp16")]; + string _SplitHeadsQ__mh_w_3149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3149_equation_0, values = (var_29970_cast_fp16, var_29738_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3149_cast_fp16")]; + string _SplitHeadsQ__mh_w_3151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3151_equation_0, values = (var_29970_cast_fp16, var_29745_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3151_cast_fp16")]; + string _SplitHeadsQ__mh_w_3153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3153_equation_0, values = (var_29974_cast_fp16, var_29752_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3153_cast_fp16")]; + string _SplitHeadsQ__mh_w_3155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3155_equation_0, values = (var_29974_cast_fp16, var_29759_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3155_cast_fp16")]; + string _SplitHeadsQ__mh_w_3157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3157_equation_0, values = (var_29974_cast_fp16, var_29766_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3157_cast_fp16")]; + string _SplitHeadsQ__mh_w_3159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3159_equation_0, values = (var_29974_cast_fp16, var_29773_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3159_cast_fp16")]; + string _SplitHeadsQ__mh_w_3161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3161_equation_0, values = (var_29978_cast_fp16, var_29780_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3161_cast_fp16")]; + string _SplitHeadsQ__mh_w_3163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3163_equation_0, values = (var_29978_cast_fp16, var_29787_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3163_cast_fp16")]; + string _SplitHeadsQ__mh_w_3165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3165_equation_0, values = (var_29978_cast_fp16, var_29794_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3165_cast_fp16")]; + string _SplitHeadsQ__mh_w_3167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3167_equation_0, values = (var_29978_cast_fp16, var_29801_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3167_cast_fp16")]; + string _SplitHeadsQ__mh_w_3169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3169_equation_0, values = (var_29982_cast_fp16, var_29808_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3169_cast_fp16")]; + string _SplitHeadsQ__mh_w_3171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3171_equation_0, values = (var_29982_cast_fp16, var_29815_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3171_cast_fp16")]; + string _SplitHeadsQ__mh_w_3173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3173_equation_0, values = (var_29982_cast_fp16, var_29822_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3173_cast_fp16")]; + string _SplitHeadsQ__mh_w_3175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3175_equation_0, values = (var_29982_cast_fp16, var_29829_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3175_cast_fp16")]; + string _SplitHeadsQ__mh_w_3177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3177_equation_0, values = (var_29986_cast_fp16, var_29836_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3177_cast_fp16")]; + string _SplitHeadsQ__mh_w_3179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3179_equation_0, values = (var_29986_cast_fp16, var_29843_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3179_cast_fp16")]; + string _SplitHeadsQ__mh_w_3181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3181_equation_0, values = (var_29986_cast_fp16, var_29850_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3181_cast_fp16")]; + string _SplitHeadsQ__mh_w_3183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3183_equation_0, values = (var_29986_cast_fp16, var_29857_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3183_cast_fp16")]; + string _SplitHeadsQ__mh_w_3185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3185_equation_0, values = (var_29990_cast_fp16, var_29864_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3185_cast_fp16")]; + string _SplitHeadsQ__mh_w_3187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3187_equation_0, values = (var_29990_cast_fp16, var_29871_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3187_cast_fp16")]; + string _SplitHeadsQ__mh_w_3189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3189_equation_0, values = (var_29990_cast_fp16, var_29878_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3189_cast_fp16")]; + string _SplitHeadsQ__mh_w_3191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3191_equation_0, values = (var_29990_cast_fp16, var_29885_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3191_cast_fp16")]; + string _SplitHeadsQ__mh_w_3193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3193_equation_0, values = (var_29994_cast_fp16, var_29892_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3193_cast_fp16")]; + string _SplitHeadsQ__mh_w_3195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3195_equation_0, values = (var_29994_cast_fp16, var_29899_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3195_cast_fp16")]; + string _SplitHeadsQ__mh_w_3197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3197_equation_0, values = (var_29994_cast_fp16, var_29906_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3197_cast_fp16")]; + string _SplitHeadsQ__mh_w_3199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3199_equation_0, values = (var_29994_cast_fp16, var_29913_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3199_cast_fp16")]; + fp16 var_30235_to_fp16 = const()[name = string("op_30235_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3041_cast_fp16, y = var_30235_to_fp16)[name = string("aw_chunk_3041_cast_fp16")]; + fp16 var_30237_to_fp16 = const()[name = string("op_30237_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3043_cast_fp16, y = var_30237_to_fp16)[name = string("aw_chunk_3043_cast_fp16")]; + fp16 var_30239_to_fp16 = const()[name = string("op_30239_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3045_cast_fp16, y = var_30239_to_fp16)[name = string("aw_chunk_3045_cast_fp16")]; + fp16 var_30241_to_fp16 = const()[name = string("op_30241_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3047_cast_fp16, y = var_30241_to_fp16)[name = string("aw_chunk_3047_cast_fp16")]; + fp16 var_30243_to_fp16 = const()[name = string("op_30243_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3049_cast_fp16, y = var_30243_to_fp16)[name = string("aw_chunk_3049_cast_fp16")]; + fp16 var_30245_to_fp16 = const()[name = string("op_30245_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3051_cast_fp16, y = var_30245_to_fp16)[name = string("aw_chunk_3051_cast_fp16")]; + fp16 var_30247_to_fp16 = const()[name = string("op_30247_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3053_cast_fp16, y = var_30247_to_fp16)[name = string("aw_chunk_3053_cast_fp16")]; + fp16 var_30249_to_fp16 = const()[name = string("op_30249_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3055_cast_fp16, y = var_30249_to_fp16)[name = string("aw_chunk_3055_cast_fp16")]; + fp16 var_30251_to_fp16 = const()[name = string("op_30251_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3057_cast_fp16, y = var_30251_to_fp16)[name = string("aw_chunk_3057_cast_fp16")]; + fp16 var_30253_to_fp16 = const()[name = string("op_30253_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3059_cast_fp16, y = var_30253_to_fp16)[name = string("aw_chunk_3059_cast_fp16")]; + fp16 var_30255_to_fp16 = const()[name = string("op_30255_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3061_cast_fp16, y = var_30255_to_fp16)[name = string("aw_chunk_3061_cast_fp16")]; + fp16 var_30257_to_fp16 = const()[name = string("op_30257_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3063_cast_fp16, y = var_30257_to_fp16)[name = string("aw_chunk_3063_cast_fp16")]; + fp16 var_30259_to_fp16 = const()[name = string("op_30259_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3065_cast_fp16, y = var_30259_to_fp16)[name = string("aw_chunk_3065_cast_fp16")]; + fp16 var_30261_to_fp16 = const()[name = string("op_30261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3067_cast_fp16, y = var_30261_to_fp16)[name = string("aw_chunk_3067_cast_fp16")]; + fp16 var_30263_to_fp16 = const()[name = string("op_30263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3069_cast_fp16, y = var_30263_to_fp16)[name = string("aw_chunk_3069_cast_fp16")]; + fp16 var_30265_to_fp16 = const()[name = string("op_30265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3071_cast_fp16, y = var_30265_to_fp16)[name = string("aw_chunk_3071_cast_fp16")]; + fp16 var_30267_to_fp16 = const()[name = string("op_30267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3073_cast_fp16, y = var_30267_to_fp16)[name = string("aw_chunk_3073_cast_fp16")]; + fp16 var_30269_to_fp16 = const()[name = string("op_30269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3075_cast_fp16, y = var_30269_to_fp16)[name = string("aw_chunk_3075_cast_fp16")]; + fp16 var_30271_to_fp16 = const()[name = string("op_30271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3077_cast_fp16, y = var_30271_to_fp16)[name = string("aw_chunk_3077_cast_fp16")]; + fp16 var_30273_to_fp16 = const()[name = string("op_30273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3079_cast_fp16, y = var_30273_to_fp16)[name = string("aw_chunk_3079_cast_fp16")]; + fp16 var_30275_to_fp16 = const()[name = string("op_30275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3081_cast_fp16, y = var_30275_to_fp16)[name = string("aw_chunk_3081_cast_fp16")]; + fp16 var_30277_to_fp16 = const()[name = string("op_30277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3083_cast_fp16, y = var_30277_to_fp16)[name = string("aw_chunk_3083_cast_fp16")]; + fp16 var_30279_to_fp16 = const()[name = string("op_30279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3085_cast_fp16, y = var_30279_to_fp16)[name = string("aw_chunk_3085_cast_fp16")]; + fp16 var_30281_to_fp16 = const()[name = string("op_30281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3087_cast_fp16, y = var_30281_to_fp16)[name = string("aw_chunk_3087_cast_fp16")]; + fp16 var_30283_to_fp16 = const()[name = string("op_30283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3089_cast_fp16, y = var_30283_to_fp16)[name = string("aw_chunk_3089_cast_fp16")]; + fp16 var_30285_to_fp16 = const()[name = string("op_30285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3091_cast_fp16, y = var_30285_to_fp16)[name = string("aw_chunk_3091_cast_fp16")]; + fp16 var_30287_to_fp16 = const()[name = string("op_30287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3093_cast_fp16, y = var_30287_to_fp16)[name = string("aw_chunk_3093_cast_fp16")]; + fp16 var_30289_to_fp16 = const()[name = string("op_30289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3095_cast_fp16, y = var_30289_to_fp16)[name = string("aw_chunk_3095_cast_fp16")]; + fp16 var_30291_to_fp16 = const()[name = string("op_30291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3097_cast_fp16, y = var_30291_to_fp16)[name = string("aw_chunk_3097_cast_fp16")]; + fp16 var_30293_to_fp16 = const()[name = string("op_30293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3099_cast_fp16, y = var_30293_to_fp16)[name = string("aw_chunk_3099_cast_fp16")]; + fp16 var_30295_to_fp16 = const()[name = string("op_30295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3101_cast_fp16, y = var_30295_to_fp16)[name = string("aw_chunk_3101_cast_fp16")]; + fp16 var_30297_to_fp16 = const()[name = string("op_30297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3103_cast_fp16, y = var_30297_to_fp16)[name = string("aw_chunk_3103_cast_fp16")]; + fp16 var_30299_to_fp16 = const()[name = string("op_30299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3105_cast_fp16, y = var_30299_to_fp16)[name = string("aw_chunk_3105_cast_fp16")]; + fp16 var_30301_to_fp16 = const()[name = string("op_30301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3107_cast_fp16, y = var_30301_to_fp16)[name = string("aw_chunk_3107_cast_fp16")]; + fp16 var_30303_to_fp16 = const()[name = string("op_30303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3109_cast_fp16, y = var_30303_to_fp16)[name = string("aw_chunk_3109_cast_fp16")]; + fp16 var_30305_to_fp16 = const()[name = string("op_30305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3111_cast_fp16, y = var_30305_to_fp16)[name = string("aw_chunk_3111_cast_fp16")]; + fp16 var_30307_to_fp16 = const()[name = string("op_30307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3113_cast_fp16, y = var_30307_to_fp16)[name = string("aw_chunk_3113_cast_fp16")]; + fp16 var_30309_to_fp16 = const()[name = string("op_30309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3115_cast_fp16, y = var_30309_to_fp16)[name = string("aw_chunk_3115_cast_fp16")]; + fp16 var_30311_to_fp16 = const()[name = string("op_30311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3117_cast_fp16, y = var_30311_to_fp16)[name = string("aw_chunk_3117_cast_fp16")]; + fp16 var_30313_to_fp16 = const()[name = string("op_30313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3119_cast_fp16, y = var_30313_to_fp16)[name = string("aw_chunk_3119_cast_fp16")]; + fp16 var_30315_to_fp16 = const()[name = string("op_30315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3121_cast_fp16, y = var_30315_to_fp16)[name = string("aw_chunk_3121_cast_fp16")]; + fp16 var_30317_to_fp16 = const()[name = string("op_30317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3123_cast_fp16, y = var_30317_to_fp16)[name = string("aw_chunk_3123_cast_fp16")]; + fp16 var_30319_to_fp16 = const()[name = string("op_30319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3125_cast_fp16, y = var_30319_to_fp16)[name = string("aw_chunk_3125_cast_fp16")]; + fp16 var_30321_to_fp16 = const()[name = string("op_30321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3127_cast_fp16, y = var_30321_to_fp16)[name = string("aw_chunk_3127_cast_fp16")]; + fp16 var_30323_to_fp16 = const()[name = string("op_30323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3129_cast_fp16, y = var_30323_to_fp16)[name = string("aw_chunk_3129_cast_fp16")]; + fp16 var_30325_to_fp16 = const()[name = string("op_30325_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3131_cast_fp16, y = var_30325_to_fp16)[name = string("aw_chunk_3131_cast_fp16")]; + fp16 var_30327_to_fp16 = const()[name = string("op_30327_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3133_cast_fp16, y = var_30327_to_fp16)[name = string("aw_chunk_3133_cast_fp16")]; + fp16 var_30329_to_fp16 = const()[name = string("op_30329_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3135_cast_fp16, y = var_30329_to_fp16)[name = string("aw_chunk_3135_cast_fp16")]; + fp16 var_30331_to_fp16 = const()[name = string("op_30331_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3137_cast_fp16, y = var_30331_to_fp16)[name = string("aw_chunk_3137_cast_fp16")]; + fp16 var_30333_to_fp16 = const()[name = string("op_30333_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3139_cast_fp16, y = var_30333_to_fp16)[name = string("aw_chunk_3139_cast_fp16")]; + fp16 var_30335_to_fp16 = const()[name = string("op_30335_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3141_cast_fp16, y = var_30335_to_fp16)[name = string("aw_chunk_3141_cast_fp16")]; + fp16 var_30337_to_fp16 = const()[name = string("op_30337_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3143_cast_fp16, y = var_30337_to_fp16)[name = string("aw_chunk_3143_cast_fp16")]; + fp16 var_30339_to_fp16 = const()[name = string("op_30339_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3145_cast_fp16, y = var_30339_to_fp16)[name = string("aw_chunk_3145_cast_fp16")]; + fp16 var_30341_to_fp16 = const()[name = string("op_30341_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3147_cast_fp16, y = var_30341_to_fp16)[name = string("aw_chunk_3147_cast_fp16")]; + fp16 var_30343_to_fp16 = const()[name = string("op_30343_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3149_cast_fp16, y = var_30343_to_fp16)[name = string("aw_chunk_3149_cast_fp16")]; + fp16 var_30345_to_fp16 = const()[name = string("op_30345_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3151_cast_fp16, y = var_30345_to_fp16)[name = string("aw_chunk_3151_cast_fp16")]; + fp16 var_30347_to_fp16 = const()[name = string("op_30347_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3153_cast_fp16, y = var_30347_to_fp16)[name = string("aw_chunk_3153_cast_fp16")]; + fp16 var_30349_to_fp16 = const()[name = string("op_30349_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3155_cast_fp16, y = var_30349_to_fp16)[name = string("aw_chunk_3155_cast_fp16")]; + fp16 var_30351_to_fp16 = const()[name = string("op_30351_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3157_cast_fp16, y = var_30351_to_fp16)[name = string("aw_chunk_3157_cast_fp16")]; + fp16 var_30353_to_fp16 = const()[name = string("op_30353_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3159_cast_fp16, y = var_30353_to_fp16)[name = string("aw_chunk_3159_cast_fp16")]; + fp16 var_30355_to_fp16 = const()[name = string("op_30355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3161_cast_fp16, y = var_30355_to_fp16)[name = string("aw_chunk_3161_cast_fp16")]; + fp16 var_30357_to_fp16 = const()[name = string("op_30357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3163_cast_fp16, y = var_30357_to_fp16)[name = string("aw_chunk_3163_cast_fp16")]; + fp16 var_30359_to_fp16 = const()[name = string("op_30359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3165_cast_fp16, y = var_30359_to_fp16)[name = string("aw_chunk_3165_cast_fp16")]; + fp16 var_30361_to_fp16 = const()[name = string("op_30361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3167_cast_fp16, y = var_30361_to_fp16)[name = string("aw_chunk_3167_cast_fp16")]; + fp16 var_30363_to_fp16 = const()[name = string("op_30363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3169_cast_fp16, y = var_30363_to_fp16)[name = string("aw_chunk_3169_cast_fp16")]; + fp16 var_30365_to_fp16 = const()[name = string("op_30365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3171_cast_fp16, y = var_30365_to_fp16)[name = string("aw_chunk_3171_cast_fp16")]; + fp16 var_30367_to_fp16 = const()[name = string("op_30367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3173_cast_fp16, y = var_30367_to_fp16)[name = string("aw_chunk_3173_cast_fp16")]; + fp16 var_30369_to_fp16 = const()[name = string("op_30369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3175_cast_fp16, y = var_30369_to_fp16)[name = string("aw_chunk_3175_cast_fp16")]; + fp16 var_30371_to_fp16 = const()[name = string("op_30371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3177_cast_fp16, y = var_30371_to_fp16)[name = string("aw_chunk_3177_cast_fp16")]; + fp16 var_30373_to_fp16 = const()[name = string("op_30373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3179_cast_fp16, y = var_30373_to_fp16)[name = string("aw_chunk_3179_cast_fp16")]; + fp16 var_30375_to_fp16 = const()[name = string("op_30375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3181_cast_fp16, y = var_30375_to_fp16)[name = string("aw_chunk_3181_cast_fp16")]; + fp16 var_30377_to_fp16 = const()[name = string("op_30377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3183_cast_fp16, y = var_30377_to_fp16)[name = string("aw_chunk_3183_cast_fp16")]; + fp16 var_30379_to_fp16 = const()[name = string("op_30379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3185_cast_fp16, y = var_30379_to_fp16)[name = string("aw_chunk_3185_cast_fp16")]; + fp16 var_30381_to_fp16 = const()[name = string("op_30381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3187_cast_fp16, y = var_30381_to_fp16)[name = string("aw_chunk_3187_cast_fp16")]; + fp16 var_30383_to_fp16 = const()[name = string("op_30383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3189_cast_fp16, y = var_30383_to_fp16)[name = string("aw_chunk_3189_cast_fp16")]; + fp16 var_30385_to_fp16 = const()[name = string("op_30385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3191_cast_fp16, y = var_30385_to_fp16)[name = string("aw_chunk_3191_cast_fp16")]; + fp16 var_30387_to_fp16 = const()[name = string("op_30387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3193_cast_fp16, y = var_30387_to_fp16)[name = string("aw_chunk_3193_cast_fp16")]; + fp16 var_30389_to_fp16 = const()[name = string("op_30389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3195_cast_fp16, y = var_30389_to_fp16)[name = string("aw_chunk_3195_cast_fp16")]; + fp16 var_30391_to_fp16 = const()[name = string("op_30391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3197_cast_fp16, y = var_30391_to_fp16)[name = string("aw_chunk_3197_cast_fp16")]; + fp16 var_30393_to_fp16 = const()[name = string("op_30393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3199_cast_fp16, y = var_30393_to_fp16)[name = string("aw_chunk_3199_cast_fp16")]; + tensor var_30395_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3041_cast_fp16)[name = string("op_30395_cast_fp16")]; + tensor var_30396_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3043_cast_fp16)[name = string("op_30396_cast_fp16")]; + tensor var_30397_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3045_cast_fp16)[name = string("op_30397_cast_fp16")]; + tensor var_30398_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3047_cast_fp16)[name = string("op_30398_cast_fp16")]; + tensor var_30399_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3049_cast_fp16)[name = string("op_30399_cast_fp16")]; + tensor var_30400_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3051_cast_fp16)[name = string("op_30400_cast_fp16")]; + tensor var_30401_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3053_cast_fp16)[name = string("op_30401_cast_fp16")]; + tensor var_30402_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3055_cast_fp16)[name = string("op_30402_cast_fp16")]; + tensor var_30403_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3057_cast_fp16)[name = string("op_30403_cast_fp16")]; + tensor var_30404_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3059_cast_fp16)[name = string("op_30404_cast_fp16")]; + tensor var_30405_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3061_cast_fp16)[name = string("op_30405_cast_fp16")]; + tensor var_30406_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3063_cast_fp16)[name = string("op_30406_cast_fp16")]; + tensor var_30407_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3065_cast_fp16)[name = string("op_30407_cast_fp16")]; + tensor var_30408_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3067_cast_fp16)[name = string("op_30408_cast_fp16")]; + tensor var_30409_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3069_cast_fp16)[name = string("op_30409_cast_fp16")]; + tensor var_30410_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3071_cast_fp16)[name = string("op_30410_cast_fp16")]; + tensor var_30411_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3073_cast_fp16)[name = string("op_30411_cast_fp16")]; + tensor var_30412_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3075_cast_fp16)[name = string("op_30412_cast_fp16")]; + tensor var_30413_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3077_cast_fp16)[name = string("op_30413_cast_fp16")]; + tensor var_30414_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3079_cast_fp16)[name = string("op_30414_cast_fp16")]; + tensor var_30415_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3081_cast_fp16)[name = string("op_30415_cast_fp16")]; + tensor var_30416_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3083_cast_fp16)[name = string("op_30416_cast_fp16")]; + tensor var_30417_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3085_cast_fp16)[name = string("op_30417_cast_fp16")]; + tensor var_30418_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3087_cast_fp16)[name = string("op_30418_cast_fp16")]; + tensor var_30419_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3089_cast_fp16)[name = string("op_30419_cast_fp16")]; + tensor var_30420_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3091_cast_fp16)[name = string("op_30420_cast_fp16")]; + tensor var_30421_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3093_cast_fp16)[name = string("op_30421_cast_fp16")]; + tensor var_30422_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3095_cast_fp16)[name = string("op_30422_cast_fp16")]; + tensor var_30423_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3097_cast_fp16)[name = string("op_30423_cast_fp16")]; + tensor var_30424_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3099_cast_fp16)[name = string("op_30424_cast_fp16")]; + tensor var_30425_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3101_cast_fp16)[name = string("op_30425_cast_fp16")]; + tensor var_30426_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3103_cast_fp16)[name = string("op_30426_cast_fp16")]; + tensor var_30427_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3105_cast_fp16)[name = string("op_30427_cast_fp16")]; + tensor var_30428_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3107_cast_fp16)[name = string("op_30428_cast_fp16")]; + tensor var_30429_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3109_cast_fp16)[name = string("op_30429_cast_fp16")]; + tensor var_30430_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3111_cast_fp16)[name = string("op_30430_cast_fp16")]; + tensor var_30431_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3113_cast_fp16)[name = string("op_30431_cast_fp16")]; + tensor var_30432_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3115_cast_fp16)[name = string("op_30432_cast_fp16")]; + tensor var_30433_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3117_cast_fp16)[name = string("op_30433_cast_fp16")]; + tensor var_30434_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3119_cast_fp16)[name = string("op_30434_cast_fp16")]; + tensor var_30435_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3121_cast_fp16)[name = string("op_30435_cast_fp16")]; + tensor var_30436_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3123_cast_fp16)[name = string("op_30436_cast_fp16")]; + tensor var_30437_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3125_cast_fp16)[name = string("op_30437_cast_fp16")]; + tensor var_30438_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3127_cast_fp16)[name = string("op_30438_cast_fp16")]; + tensor var_30439_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3129_cast_fp16)[name = string("op_30439_cast_fp16")]; + tensor var_30440_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3131_cast_fp16)[name = string("op_30440_cast_fp16")]; + tensor var_30441_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3133_cast_fp16)[name = string("op_30441_cast_fp16")]; + tensor var_30442_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3135_cast_fp16)[name = string("op_30442_cast_fp16")]; + tensor var_30443_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3137_cast_fp16)[name = string("op_30443_cast_fp16")]; + tensor var_30444_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3139_cast_fp16)[name = string("op_30444_cast_fp16")]; + tensor var_30445_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3141_cast_fp16)[name = string("op_30445_cast_fp16")]; + tensor var_30446_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3143_cast_fp16)[name = string("op_30446_cast_fp16")]; + tensor var_30447_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3145_cast_fp16)[name = string("op_30447_cast_fp16")]; + tensor var_30448_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3147_cast_fp16)[name = string("op_30448_cast_fp16")]; + tensor var_30449_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3149_cast_fp16)[name = string("op_30449_cast_fp16")]; + tensor var_30450_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3151_cast_fp16)[name = string("op_30450_cast_fp16")]; + tensor var_30451_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3153_cast_fp16)[name = string("op_30451_cast_fp16")]; + tensor var_30452_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3155_cast_fp16)[name = string("op_30452_cast_fp16")]; + tensor var_30453_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3157_cast_fp16)[name = string("op_30453_cast_fp16")]; + tensor var_30454_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3159_cast_fp16)[name = string("op_30454_cast_fp16")]; + tensor var_30455_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3161_cast_fp16)[name = string("op_30455_cast_fp16")]; + tensor var_30456_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3163_cast_fp16)[name = string("op_30456_cast_fp16")]; + tensor var_30457_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3165_cast_fp16)[name = string("op_30457_cast_fp16")]; + tensor var_30458_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3167_cast_fp16)[name = string("op_30458_cast_fp16")]; + tensor var_30459_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3169_cast_fp16)[name = string("op_30459_cast_fp16")]; + tensor var_30460_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3171_cast_fp16)[name = string("op_30460_cast_fp16")]; + tensor var_30461_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3173_cast_fp16)[name = string("op_30461_cast_fp16")]; + tensor var_30462_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3175_cast_fp16)[name = string("op_30462_cast_fp16")]; + tensor var_30463_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3177_cast_fp16)[name = string("op_30463_cast_fp16")]; + tensor var_30464_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3179_cast_fp16)[name = string("op_30464_cast_fp16")]; + tensor var_30465_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3181_cast_fp16)[name = string("op_30465_cast_fp16")]; + tensor var_30466_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3183_cast_fp16)[name = string("op_30466_cast_fp16")]; + tensor var_30467_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3185_cast_fp16)[name = string("op_30467_cast_fp16")]; + tensor var_30468_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3187_cast_fp16)[name = string("op_30468_cast_fp16")]; + tensor var_30469_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3189_cast_fp16)[name = string("op_30469_cast_fp16")]; + tensor var_30470_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3191_cast_fp16)[name = string("op_30470_cast_fp16")]; + tensor var_30471_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3193_cast_fp16)[name = string("op_30471_cast_fp16")]; + tensor var_30472_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3195_cast_fp16)[name = string("op_30472_cast_fp16")]; + tensor var_30473_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3197_cast_fp16)[name = string("op_30473_cast_fp16")]; + tensor var_30474_cast_fp16 = softmax(axis = var_29220, x = aw_chunk_3199_cast_fp16)[name = string("op_30474_cast_fp16")]; + string var_30476_equation_0 = const()[name = string("op_30476_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30476_cast_fp16 = einsum(equation = var_30476_equation_0, values = (var_29996_cast_fp16, var_30395_cast_fp16))[name = string("op_30476_cast_fp16")]; + string var_30478_equation_0 = const()[name = string("op_30478_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30478_cast_fp16 = einsum(equation = var_30478_equation_0, values = (var_29996_cast_fp16, var_30396_cast_fp16))[name = string("op_30478_cast_fp16")]; + string var_30480_equation_0 = const()[name = string("op_30480_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30480_cast_fp16 = einsum(equation = var_30480_equation_0, values = (var_29996_cast_fp16, var_30397_cast_fp16))[name = string("op_30480_cast_fp16")]; + string var_30482_equation_0 = const()[name = string("op_30482_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30482_cast_fp16 = einsum(equation = var_30482_equation_0, values = (var_29996_cast_fp16, var_30398_cast_fp16))[name = string("op_30482_cast_fp16")]; + string var_30484_equation_0 = const()[name = string("op_30484_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30484_cast_fp16 = einsum(equation = var_30484_equation_0, values = (var_30000_cast_fp16, var_30399_cast_fp16))[name = string("op_30484_cast_fp16")]; + string var_30486_equation_0 = const()[name = string("op_30486_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30486_cast_fp16 = einsum(equation = var_30486_equation_0, values = (var_30000_cast_fp16, var_30400_cast_fp16))[name = string("op_30486_cast_fp16")]; + string var_30488_equation_0 = const()[name = string("op_30488_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30488_cast_fp16 = einsum(equation = var_30488_equation_0, values = (var_30000_cast_fp16, var_30401_cast_fp16))[name = string("op_30488_cast_fp16")]; + string var_30490_equation_0 = const()[name = string("op_30490_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30490_cast_fp16 = einsum(equation = var_30490_equation_0, values = (var_30000_cast_fp16, var_30402_cast_fp16))[name = string("op_30490_cast_fp16")]; + string var_30492_equation_0 = const()[name = string("op_30492_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30492_cast_fp16 = einsum(equation = var_30492_equation_0, values = (var_30004_cast_fp16, var_30403_cast_fp16))[name = string("op_30492_cast_fp16")]; + string var_30494_equation_0 = const()[name = string("op_30494_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30494_cast_fp16 = einsum(equation = var_30494_equation_0, values = (var_30004_cast_fp16, var_30404_cast_fp16))[name = string("op_30494_cast_fp16")]; + string var_30496_equation_0 = const()[name = string("op_30496_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30496_cast_fp16 = einsum(equation = var_30496_equation_0, values = (var_30004_cast_fp16, var_30405_cast_fp16))[name = string("op_30496_cast_fp16")]; + string var_30498_equation_0 = const()[name = string("op_30498_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30498_cast_fp16 = einsum(equation = var_30498_equation_0, values = (var_30004_cast_fp16, var_30406_cast_fp16))[name = string("op_30498_cast_fp16")]; + string var_30500_equation_0 = const()[name = string("op_30500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30500_cast_fp16 = einsum(equation = var_30500_equation_0, values = (var_30008_cast_fp16, var_30407_cast_fp16))[name = string("op_30500_cast_fp16")]; + string var_30502_equation_0 = const()[name = string("op_30502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30502_cast_fp16 = einsum(equation = var_30502_equation_0, values = (var_30008_cast_fp16, var_30408_cast_fp16))[name = string("op_30502_cast_fp16")]; + string var_30504_equation_0 = const()[name = string("op_30504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30504_cast_fp16 = einsum(equation = var_30504_equation_0, values = (var_30008_cast_fp16, var_30409_cast_fp16))[name = string("op_30504_cast_fp16")]; + string var_30506_equation_0 = const()[name = string("op_30506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30506_cast_fp16 = einsum(equation = var_30506_equation_0, values = (var_30008_cast_fp16, var_30410_cast_fp16))[name = string("op_30506_cast_fp16")]; + string var_30508_equation_0 = const()[name = string("op_30508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30508_cast_fp16 = einsum(equation = var_30508_equation_0, values = (var_30012_cast_fp16, var_30411_cast_fp16))[name = string("op_30508_cast_fp16")]; + string var_30510_equation_0 = const()[name = string("op_30510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30510_cast_fp16 = einsum(equation = var_30510_equation_0, values = (var_30012_cast_fp16, var_30412_cast_fp16))[name = string("op_30510_cast_fp16")]; + string var_30512_equation_0 = const()[name = string("op_30512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30512_cast_fp16 = einsum(equation = var_30512_equation_0, values = (var_30012_cast_fp16, var_30413_cast_fp16))[name = string("op_30512_cast_fp16")]; + string var_30514_equation_0 = const()[name = string("op_30514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30514_cast_fp16 = einsum(equation = var_30514_equation_0, values = (var_30012_cast_fp16, var_30414_cast_fp16))[name = string("op_30514_cast_fp16")]; + string var_30516_equation_0 = const()[name = string("op_30516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30516_cast_fp16 = einsum(equation = var_30516_equation_0, values = (var_30016_cast_fp16, var_30415_cast_fp16))[name = string("op_30516_cast_fp16")]; + string var_30518_equation_0 = const()[name = string("op_30518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30518_cast_fp16 = einsum(equation = var_30518_equation_0, values = (var_30016_cast_fp16, var_30416_cast_fp16))[name = string("op_30518_cast_fp16")]; + string var_30520_equation_0 = const()[name = string("op_30520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30520_cast_fp16 = einsum(equation = var_30520_equation_0, values = (var_30016_cast_fp16, var_30417_cast_fp16))[name = string("op_30520_cast_fp16")]; + string var_30522_equation_0 = const()[name = string("op_30522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30522_cast_fp16 = einsum(equation = var_30522_equation_0, values = (var_30016_cast_fp16, var_30418_cast_fp16))[name = string("op_30522_cast_fp16")]; + string var_30524_equation_0 = const()[name = string("op_30524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30524_cast_fp16 = einsum(equation = var_30524_equation_0, values = (var_30020_cast_fp16, var_30419_cast_fp16))[name = string("op_30524_cast_fp16")]; + string var_30526_equation_0 = const()[name = string("op_30526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30526_cast_fp16 = einsum(equation = var_30526_equation_0, values = (var_30020_cast_fp16, var_30420_cast_fp16))[name = string("op_30526_cast_fp16")]; + string var_30528_equation_0 = const()[name = string("op_30528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30528_cast_fp16 = einsum(equation = var_30528_equation_0, values = (var_30020_cast_fp16, var_30421_cast_fp16))[name = string("op_30528_cast_fp16")]; + string var_30530_equation_0 = const()[name = string("op_30530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30530_cast_fp16 = einsum(equation = var_30530_equation_0, values = (var_30020_cast_fp16, var_30422_cast_fp16))[name = string("op_30530_cast_fp16")]; + string var_30532_equation_0 = const()[name = string("op_30532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30532_cast_fp16 = einsum(equation = var_30532_equation_0, values = (var_30024_cast_fp16, var_30423_cast_fp16))[name = string("op_30532_cast_fp16")]; + string var_30534_equation_0 = const()[name = string("op_30534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30534_cast_fp16 = einsum(equation = var_30534_equation_0, values = (var_30024_cast_fp16, var_30424_cast_fp16))[name = string("op_30534_cast_fp16")]; + string var_30536_equation_0 = const()[name = string("op_30536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30536_cast_fp16 = einsum(equation = var_30536_equation_0, values = (var_30024_cast_fp16, var_30425_cast_fp16))[name = string("op_30536_cast_fp16")]; + string var_30538_equation_0 = const()[name = string("op_30538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30538_cast_fp16 = einsum(equation = var_30538_equation_0, values = (var_30024_cast_fp16, var_30426_cast_fp16))[name = string("op_30538_cast_fp16")]; + string var_30540_equation_0 = const()[name = string("op_30540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30540_cast_fp16 = einsum(equation = var_30540_equation_0, values = (var_30028_cast_fp16, var_30427_cast_fp16))[name = string("op_30540_cast_fp16")]; + string var_30542_equation_0 = const()[name = string("op_30542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30542_cast_fp16 = einsum(equation = var_30542_equation_0, values = (var_30028_cast_fp16, var_30428_cast_fp16))[name = string("op_30542_cast_fp16")]; + string var_30544_equation_0 = const()[name = string("op_30544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30544_cast_fp16 = einsum(equation = var_30544_equation_0, values = (var_30028_cast_fp16, var_30429_cast_fp16))[name = string("op_30544_cast_fp16")]; + string var_30546_equation_0 = const()[name = string("op_30546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30546_cast_fp16 = einsum(equation = var_30546_equation_0, values = (var_30028_cast_fp16, var_30430_cast_fp16))[name = string("op_30546_cast_fp16")]; + string var_30548_equation_0 = const()[name = string("op_30548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30548_cast_fp16 = einsum(equation = var_30548_equation_0, values = (var_30032_cast_fp16, var_30431_cast_fp16))[name = string("op_30548_cast_fp16")]; + string var_30550_equation_0 = const()[name = string("op_30550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30550_cast_fp16 = einsum(equation = var_30550_equation_0, values = (var_30032_cast_fp16, var_30432_cast_fp16))[name = string("op_30550_cast_fp16")]; + string var_30552_equation_0 = const()[name = string("op_30552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30552_cast_fp16 = einsum(equation = var_30552_equation_0, values = (var_30032_cast_fp16, var_30433_cast_fp16))[name = string("op_30552_cast_fp16")]; + string var_30554_equation_0 = const()[name = string("op_30554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30554_cast_fp16 = einsum(equation = var_30554_equation_0, values = (var_30032_cast_fp16, var_30434_cast_fp16))[name = string("op_30554_cast_fp16")]; + string var_30556_equation_0 = const()[name = string("op_30556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30556_cast_fp16 = einsum(equation = var_30556_equation_0, values = (var_30036_cast_fp16, var_30435_cast_fp16))[name = string("op_30556_cast_fp16")]; + string var_30558_equation_0 = const()[name = string("op_30558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30558_cast_fp16 = einsum(equation = var_30558_equation_0, values = (var_30036_cast_fp16, var_30436_cast_fp16))[name = string("op_30558_cast_fp16")]; + string var_30560_equation_0 = const()[name = string("op_30560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30560_cast_fp16 = einsum(equation = var_30560_equation_0, values = (var_30036_cast_fp16, var_30437_cast_fp16))[name = string("op_30560_cast_fp16")]; + string var_30562_equation_0 = const()[name = string("op_30562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30562_cast_fp16 = einsum(equation = var_30562_equation_0, values = (var_30036_cast_fp16, var_30438_cast_fp16))[name = string("op_30562_cast_fp16")]; + string var_30564_equation_0 = const()[name = string("op_30564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30564_cast_fp16 = einsum(equation = var_30564_equation_0, values = (var_30040_cast_fp16, var_30439_cast_fp16))[name = string("op_30564_cast_fp16")]; + string var_30566_equation_0 = const()[name = string("op_30566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30566_cast_fp16 = einsum(equation = var_30566_equation_0, values = (var_30040_cast_fp16, var_30440_cast_fp16))[name = string("op_30566_cast_fp16")]; + string var_30568_equation_0 = const()[name = string("op_30568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30568_cast_fp16 = einsum(equation = var_30568_equation_0, values = (var_30040_cast_fp16, var_30441_cast_fp16))[name = string("op_30568_cast_fp16")]; + string var_30570_equation_0 = const()[name = string("op_30570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30570_cast_fp16 = einsum(equation = var_30570_equation_0, values = (var_30040_cast_fp16, var_30442_cast_fp16))[name = string("op_30570_cast_fp16")]; + string var_30572_equation_0 = const()[name = string("op_30572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30572_cast_fp16 = einsum(equation = var_30572_equation_0, values = (var_30044_cast_fp16, var_30443_cast_fp16))[name = string("op_30572_cast_fp16")]; + string var_30574_equation_0 = const()[name = string("op_30574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30574_cast_fp16 = einsum(equation = var_30574_equation_0, values = (var_30044_cast_fp16, var_30444_cast_fp16))[name = string("op_30574_cast_fp16")]; + string var_30576_equation_0 = const()[name = string("op_30576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30576_cast_fp16 = einsum(equation = var_30576_equation_0, values = (var_30044_cast_fp16, var_30445_cast_fp16))[name = string("op_30576_cast_fp16")]; + string var_30578_equation_0 = const()[name = string("op_30578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30578_cast_fp16 = einsum(equation = var_30578_equation_0, values = (var_30044_cast_fp16, var_30446_cast_fp16))[name = string("op_30578_cast_fp16")]; + string var_30580_equation_0 = const()[name = string("op_30580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30580_cast_fp16 = einsum(equation = var_30580_equation_0, values = (var_30048_cast_fp16, var_30447_cast_fp16))[name = string("op_30580_cast_fp16")]; + string var_30582_equation_0 = const()[name = string("op_30582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30582_cast_fp16 = einsum(equation = var_30582_equation_0, values = (var_30048_cast_fp16, var_30448_cast_fp16))[name = string("op_30582_cast_fp16")]; + string var_30584_equation_0 = const()[name = string("op_30584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30584_cast_fp16 = einsum(equation = var_30584_equation_0, values = (var_30048_cast_fp16, var_30449_cast_fp16))[name = string("op_30584_cast_fp16")]; + string var_30586_equation_0 = const()[name = string("op_30586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30586_cast_fp16 = einsum(equation = var_30586_equation_0, values = (var_30048_cast_fp16, var_30450_cast_fp16))[name = string("op_30586_cast_fp16")]; + string var_30588_equation_0 = const()[name = string("op_30588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30588_cast_fp16 = einsum(equation = var_30588_equation_0, values = (var_30052_cast_fp16, var_30451_cast_fp16))[name = string("op_30588_cast_fp16")]; + string var_30590_equation_0 = const()[name = string("op_30590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30590_cast_fp16 = einsum(equation = var_30590_equation_0, values = (var_30052_cast_fp16, var_30452_cast_fp16))[name = string("op_30590_cast_fp16")]; + string var_30592_equation_0 = const()[name = string("op_30592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30592_cast_fp16 = einsum(equation = var_30592_equation_0, values = (var_30052_cast_fp16, var_30453_cast_fp16))[name = string("op_30592_cast_fp16")]; + string var_30594_equation_0 = const()[name = string("op_30594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30594_cast_fp16 = einsum(equation = var_30594_equation_0, values = (var_30052_cast_fp16, var_30454_cast_fp16))[name = string("op_30594_cast_fp16")]; + string var_30596_equation_0 = const()[name = string("op_30596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30596_cast_fp16 = einsum(equation = var_30596_equation_0, values = (var_30056_cast_fp16, var_30455_cast_fp16))[name = string("op_30596_cast_fp16")]; + string var_30598_equation_0 = const()[name = string("op_30598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30598_cast_fp16 = einsum(equation = var_30598_equation_0, values = (var_30056_cast_fp16, var_30456_cast_fp16))[name = string("op_30598_cast_fp16")]; + string var_30600_equation_0 = const()[name = string("op_30600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30600_cast_fp16 = einsum(equation = var_30600_equation_0, values = (var_30056_cast_fp16, var_30457_cast_fp16))[name = string("op_30600_cast_fp16")]; + string var_30602_equation_0 = const()[name = string("op_30602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30602_cast_fp16 = einsum(equation = var_30602_equation_0, values = (var_30056_cast_fp16, var_30458_cast_fp16))[name = string("op_30602_cast_fp16")]; + string var_30604_equation_0 = const()[name = string("op_30604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30604_cast_fp16 = einsum(equation = var_30604_equation_0, values = (var_30060_cast_fp16, var_30459_cast_fp16))[name = string("op_30604_cast_fp16")]; + string var_30606_equation_0 = const()[name = string("op_30606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30606_cast_fp16 = einsum(equation = var_30606_equation_0, values = (var_30060_cast_fp16, var_30460_cast_fp16))[name = string("op_30606_cast_fp16")]; + string var_30608_equation_0 = const()[name = string("op_30608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30608_cast_fp16 = einsum(equation = var_30608_equation_0, values = (var_30060_cast_fp16, var_30461_cast_fp16))[name = string("op_30608_cast_fp16")]; + string var_30610_equation_0 = const()[name = string("op_30610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30610_cast_fp16 = einsum(equation = var_30610_equation_0, values = (var_30060_cast_fp16, var_30462_cast_fp16))[name = string("op_30610_cast_fp16")]; + string var_30612_equation_0 = const()[name = string("op_30612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30612_cast_fp16 = einsum(equation = var_30612_equation_0, values = (var_30064_cast_fp16, var_30463_cast_fp16))[name = string("op_30612_cast_fp16")]; + string var_30614_equation_0 = const()[name = string("op_30614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30614_cast_fp16 = einsum(equation = var_30614_equation_0, values = (var_30064_cast_fp16, var_30464_cast_fp16))[name = string("op_30614_cast_fp16")]; + string var_30616_equation_0 = const()[name = string("op_30616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30616_cast_fp16 = einsum(equation = var_30616_equation_0, values = (var_30064_cast_fp16, var_30465_cast_fp16))[name = string("op_30616_cast_fp16")]; + string var_30618_equation_0 = const()[name = string("op_30618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30618_cast_fp16 = einsum(equation = var_30618_equation_0, values = (var_30064_cast_fp16, var_30466_cast_fp16))[name = string("op_30618_cast_fp16")]; + string var_30620_equation_0 = const()[name = string("op_30620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30620_cast_fp16 = einsum(equation = var_30620_equation_0, values = (var_30068_cast_fp16, var_30467_cast_fp16))[name = string("op_30620_cast_fp16")]; + string var_30622_equation_0 = const()[name = string("op_30622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30622_cast_fp16 = einsum(equation = var_30622_equation_0, values = (var_30068_cast_fp16, var_30468_cast_fp16))[name = string("op_30622_cast_fp16")]; + string var_30624_equation_0 = const()[name = string("op_30624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30624_cast_fp16 = einsum(equation = var_30624_equation_0, values = (var_30068_cast_fp16, var_30469_cast_fp16))[name = string("op_30624_cast_fp16")]; + string var_30626_equation_0 = const()[name = string("op_30626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30626_cast_fp16 = einsum(equation = var_30626_equation_0, values = (var_30068_cast_fp16, var_30470_cast_fp16))[name = string("op_30626_cast_fp16")]; + string var_30628_equation_0 = const()[name = string("op_30628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30628_cast_fp16 = einsum(equation = var_30628_equation_0, values = (var_30072_cast_fp16, var_30471_cast_fp16))[name = string("op_30628_cast_fp16")]; + string var_30630_equation_0 = const()[name = string("op_30630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30630_cast_fp16 = einsum(equation = var_30630_equation_0, values = (var_30072_cast_fp16, var_30472_cast_fp16))[name = string("op_30630_cast_fp16")]; + string var_30632_equation_0 = const()[name = string("op_30632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30632_cast_fp16 = einsum(equation = var_30632_equation_0, values = (var_30072_cast_fp16, var_30473_cast_fp16))[name = string("op_30632_cast_fp16")]; + string var_30634_equation_0 = const()[name = string("op_30634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_30634_cast_fp16 = einsum(equation = var_30634_equation_0, values = (var_30072_cast_fp16, var_30474_cast_fp16))[name = string("op_30634_cast_fp16")]; + bool var_30636_interleave_0 = const()[name = string("op_30636_interleave_0"), val = bool(false)]; + tensor var_30636_cast_fp16 = concat(axis = var_29195, interleave = var_30636_interleave_0, values = (var_30476_cast_fp16, var_30478_cast_fp16, var_30480_cast_fp16, var_30482_cast_fp16))[name = string("op_30636_cast_fp16")]; + bool var_30638_interleave_0 = const()[name = string("op_30638_interleave_0"), val = bool(false)]; + tensor var_30638_cast_fp16 = concat(axis = var_29195, interleave = var_30638_interleave_0, values = (var_30484_cast_fp16, var_30486_cast_fp16, var_30488_cast_fp16, var_30490_cast_fp16))[name = string("op_30638_cast_fp16")]; + bool var_30640_interleave_0 = const()[name = string("op_30640_interleave_0"), val = bool(false)]; + tensor var_30640_cast_fp16 = concat(axis = var_29195, interleave = var_30640_interleave_0, values = (var_30492_cast_fp16, var_30494_cast_fp16, var_30496_cast_fp16, var_30498_cast_fp16))[name = string("op_30640_cast_fp16")]; + bool var_30642_interleave_0 = const()[name = string("op_30642_interleave_0"), val = bool(false)]; + tensor var_30642_cast_fp16 = concat(axis = var_29195, interleave = var_30642_interleave_0, values = (var_30500_cast_fp16, var_30502_cast_fp16, var_30504_cast_fp16, var_30506_cast_fp16))[name = string("op_30642_cast_fp16")]; + bool var_30644_interleave_0 = const()[name = string("op_30644_interleave_0"), val = bool(false)]; + tensor var_30644_cast_fp16 = concat(axis = var_29195, interleave = var_30644_interleave_0, values = (var_30508_cast_fp16, var_30510_cast_fp16, var_30512_cast_fp16, var_30514_cast_fp16))[name = string("op_30644_cast_fp16")]; + bool var_30646_interleave_0 = const()[name = string("op_30646_interleave_0"), val = bool(false)]; + tensor var_30646_cast_fp16 = concat(axis = var_29195, interleave = var_30646_interleave_0, values = (var_30516_cast_fp16, var_30518_cast_fp16, var_30520_cast_fp16, var_30522_cast_fp16))[name = string("op_30646_cast_fp16")]; + bool var_30648_interleave_0 = const()[name = string("op_30648_interleave_0"), val = bool(false)]; + tensor var_30648_cast_fp16 = concat(axis = var_29195, interleave = var_30648_interleave_0, values = (var_30524_cast_fp16, var_30526_cast_fp16, var_30528_cast_fp16, var_30530_cast_fp16))[name = string("op_30648_cast_fp16")]; + bool var_30650_interleave_0 = const()[name = string("op_30650_interleave_0"), val = bool(false)]; + tensor var_30650_cast_fp16 = concat(axis = var_29195, interleave = var_30650_interleave_0, values = (var_30532_cast_fp16, var_30534_cast_fp16, var_30536_cast_fp16, var_30538_cast_fp16))[name = string("op_30650_cast_fp16")]; + bool var_30652_interleave_0 = const()[name = string("op_30652_interleave_0"), val = bool(false)]; + tensor var_30652_cast_fp16 = concat(axis = var_29195, interleave = var_30652_interleave_0, values = (var_30540_cast_fp16, var_30542_cast_fp16, var_30544_cast_fp16, var_30546_cast_fp16))[name = string("op_30652_cast_fp16")]; + bool var_30654_interleave_0 = const()[name = string("op_30654_interleave_0"), val = bool(false)]; + tensor var_30654_cast_fp16 = concat(axis = var_29195, interleave = var_30654_interleave_0, values = (var_30548_cast_fp16, var_30550_cast_fp16, var_30552_cast_fp16, var_30554_cast_fp16))[name = string("op_30654_cast_fp16")]; + bool var_30656_interleave_0 = const()[name = string("op_30656_interleave_0"), val = bool(false)]; + tensor var_30656_cast_fp16 = concat(axis = var_29195, interleave = var_30656_interleave_0, values = (var_30556_cast_fp16, var_30558_cast_fp16, var_30560_cast_fp16, var_30562_cast_fp16))[name = string("op_30656_cast_fp16")]; + bool var_30658_interleave_0 = const()[name = string("op_30658_interleave_0"), val = bool(false)]; + tensor var_30658_cast_fp16 = concat(axis = var_29195, interleave = var_30658_interleave_0, values = (var_30564_cast_fp16, var_30566_cast_fp16, var_30568_cast_fp16, var_30570_cast_fp16))[name = string("op_30658_cast_fp16")]; + bool var_30660_interleave_0 = const()[name = string("op_30660_interleave_0"), val = bool(false)]; + tensor var_30660_cast_fp16 = concat(axis = var_29195, interleave = var_30660_interleave_0, values = (var_30572_cast_fp16, var_30574_cast_fp16, var_30576_cast_fp16, var_30578_cast_fp16))[name = string("op_30660_cast_fp16")]; + bool var_30662_interleave_0 = const()[name = string("op_30662_interleave_0"), val = bool(false)]; + tensor var_30662_cast_fp16 = concat(axis = var_29195, interleave = var_30662_interleave_0, values = (var_30580_cast_fp16, var_30582_cast_fp16, var_30584_cast_fp16, var_30586_cast_fp16))[name = string("op_30662_cast_fp16")]; + bool var_30664_interleave_0 = const()[name = string("op_30664_interleave_0"), val = bool(false)]; + tensor var_30664_cast_fp16 = concat(axis = var_29195, interleave = var_30664_interleave_0, values = (var_30588_cast_fp16, var_30590_cast_fp16, var_30592_cast_fp16, var_30594_cast_fp16))[name = string("op_30664_cast_fp16")]; + bool var_30666_interleave_0 = const()[name = string("op_30666_interleave_0"), val = bool(false)]; + tensor var_30666_cast_fp16 = concat(axis = var_29195, interleave = var_30666_interleave_0, values = (var_30596_cast_fp16, var_30598_cast_fp16, var_30600_cast_fp16, var_30602_cast_fp16))[name = string("op_30666_cast_fp16")]; + bool var_30668_interleave_0 = const()[name = string("op_30668_interleave_0"), val = bool(false)]; + tensor var_30668_cast_fp16 = concat(axis = var_29195, interleave = var_30668_interleave_0, values = (var_30604_cast_fp16, var_30606_cast_fp16, var_30608_cast_fp16, var_30610_cast_fp16))[name = string("op_30668_cast_fp16")]; + bool var_30670_interleave_0 = const()[name = string("op_30670_interleave_0"), val = bool(false)]; + tensor var_30670_cast_fp16 = concat(axis = var_29195, interleave = var_30670_interleave_0, values = (var_30612_cast_fp16, var_30614_cast_fp16, var_30616_cast_fp16, var_30618_cast_fp16))[name = string("op_30670_cast_fp16")]; + bool var_30672_interleave_0 = const()[name = string("op_30672_interleave_0"), val = bool(false)]; + tensor var_30672_cast_fp16 = concat(axis = var_29195, interleave = var_30672_interleave_0, values = (var_30620_cast_fp16, var_30622_cast_fp16, var_30624_cast_fp16, var_30626_cast_fp16))[name = string("op_30672_cast_fp16")]; + bool var_30674_interleave_0 = const()[name = string("op_30674_interleave_0"), val = bool(false)]; + tensor var_30674_cast_fp16 = concat(axis = var_29195, interleave = var_30674_interleave_0, values = (var_30628_cast_fp16, var_30630_cast_fp16, var_30632_cast_fp16, var_30634_cast_fp16))[name = string("op_30674_cast_fp16")]; + bool input_153_interleave_0 = const()[name = string("input_153_interleave_0"), val = bool(false)]; + tensor input_153_cast_fp16 = concat(axis = var_29220, interleave = input_153_interleave_0, values = (var_30636_cast_fp16, var_30638_cast_fp16, var_30640_cast_fp16, var_30642_cast_fp16, var_30644_cast_fp16, var_30646_cast_fp16, var_30648_cast_fp16, var_30650_cast_fp16, var_30652_cast_fp16, var_30654_cast_fp16, var_30656_cast_fp16, var_30658_cast_fp16, var_30660_cast_fp16, var_30662_cast_fp16, var_30664_cast_fp16, var_30666_cast_fp16, var_30668_cast_fp16, var_30670_cast_fp16, var_30672_cast_fp16, var_30674_cast_fp16))[name = string("input_153_cast_fp16")]; + string obj_79_pad_type_0 = const()[name = string("obj_79_pad_type_0"), val = string("valid")]; + tensor obj_79_strides_0 = const()[name = string("obj_79_strides_0"), val = tensor([1, 1])]; + tensor obj_79_pad_0 = const()[name = string("obj_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_79_dilations_0 = const()[name = string("obj_79_dilations_0"), val = tensor([1, 1])]; + int32 obj_79_groups_0 = const()[name = string("obj_79_groups_0"), val = int32(1)]; + tensor layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(772217600)))]; + tensor layers_19_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775494464)))]; + tensor obj_79_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_bias_to_fp16, dilations = obj_79_dilations_0, groups = obj_79_groups_0, pad = obj_79_pad_0, pad_type = obj_79_pad_type_0, strides = obj_79_strides_0, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = string("obj_79_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = obj_79_cast_fp16)[name = string("inputs_79_cast_fp16")]; + tensor out_79_axes_0 = const()[name = string("out_79_axes_0"), val = tensor([1])]; + fp16 var_30693_to_fp16 = const()[name = string("op_30693_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_30693_to_fp16, x = inputs_79_cast_fp16)[name = string("out_79_cast_fp16")]; + tensor input_155_gamma_0_to_fp16 = const()[name = string("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775497088)))]; + tensor input_155_beta_0_to_fp16 = const()[name = string("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775499712)))]; + fp16 input_155_epsilon_0_to_fp16 = const()[name = string("input_155_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = string("input_155_cast_fp16")]; + string input_157_pad_type_0 = const()[name = string("input_157_pad_type_0"), val = string("valid")]; + tensor input_157_strides_0 = const()[name = string("input_157_strides_0"), val = tensor([1, 1])]; + tensor input_157_pad_0 = const()[name = string("input_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_157_dilations_0 = const()[name = string("input_157_dilations_0"), val = tensor([1, 1])]; + int32 input_157_groups_0 = const()[name = string("input_157_groups_0"), val = int32(1)]; + tensor layers_19_fc1_weight_to_fp16 = const()[name = string("layers_19_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(775502336)))]; + tensor layers_19_fc1_bias_to_fp16 = const()[name = string("layers_19_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788609600)))]; + tensor input_157_cast_fp16 = conv(bias = layers_19_fc1_bias_to_fp16, dilations = input_157_dilations_0, groups = input_157_groups_0, pad = input_157_pad_0, pad_type = input_157_pad_type_0, strides = input_157_strides_0, weight = layers_19_fc1_weight_to_fp16, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")]; + string input_159_mode_0 = const()[name = string("input_159_mode_0"), val = string("EXACT")]; + tensor input_159_cast_fp16 = gelu(mode = input_159_mode_0, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")]; + string hidden_states_43_pad_type_0 = const()[name = string("hidden_states_43_pad_type_0"), val = string("valid")]; + tensor hidden_states_43_strides_0 = const()[name = string("hidden_states_43_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_43_pad_0 = const()[name = string("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_43_dilations_0 = const()[name = string("hidden_states_43_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_43_groups_0 = const()[name = string("hidden_states_43_groups_0"), val = int32(1)]; + tensor layers_19_fc2_weight_to_fp16 = const()[name = string("layers_19_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(788619904)))]; + tensor layers_19_fc2_bias_to_fp16 = const()[name = string("layers_19_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801727168)))]; + tensor hidden_states_43_cast_fp16 = conv(bias = layers_19_fc2_bias_to_fp16, dilations = hidden_states_43_dilations_0, groups = hidden_states_43_groups_0, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = hidden_states_43_strides_0, weight = layers_19_fc2_weight_to_fp16, x = input_159_cast_fp16)[name = string("hidden_states_43_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = hidden_states_43_cast_fp16)[name = string("inputs_81_cast_fp16")]; + int32 var_30722 = const()[name = string("op_30722"), val = int32(3)]; + int32 var_30747 = const()[name = string("op_30747"), val = int32(1)]; + tensor out_81_axes_0 = const()[name = string("out_81_axes_0"), val = tensor([1])]; + fp16 var_30764_to_fp16 = const()[name = string("op_30764_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_30764_to_fp16, x = inputs_81_cast_fp16)[name = string("out_81_cast_fp16")]; + tensor obj_81_gamma_0_to_fp16 = const()[name = string("obj_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801729792)))]; + tensor obj_81_beta_0_to_fp16 = const()[name = string("obj_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801732416)))]; + fp16 obj_81_epsilon_0_to_fp16 = const()[name = string("obj_81_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_81_cast_fp16 = batch_norm(beta = obj_81_beta_0_to_fp16, epsilon = obj_81_epsilon_0_to_fp16, gamma = obj_81_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = string("obj_81_cast_fp16")]; + string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; + tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; + tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; + int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(801735040)))]; + tensor layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805011904)))]; + tensor query_41_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("query_41_cast_fp16")]; + string key_41_pad_type_0 = const()[name = string("key_41_pad_type_0"), val = string("valid")]; + tensor key_41_strides_0 = const()[name = string("key_41_strides_0"), val = tensor([1, 1])]; + tensor key_41_pad_0 = const()[name = string("key_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_41_dilations_0 = const()[name = string("key_41_dilations_0"), val = tensor([1, 1])]; + int32 key_41_groups_0 = const()[name = string("key_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(805014528)))]; + tensor key_41_cast_fp16 = conv(dilations = key_41_dilations_0, groups = key_41_groups_0, pad = key_41_pad_0, pad_type = key_41_pad_type_0, strides = key_41_strides_0, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("key_41_cast_fp16")]; + string value_41_pad_type_0 = const()[name = string("value_41_pad_type_0"), val = string("valid")]; + tensor value_41_strides_0 = const()[name = string("value_41_strides_0"), val = tensor([1, 1])]; + tensor value_41_pad_0 = const()[name = string("value_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_41_dilations_0 = const()[name = string("value_41_dilations_0"), val = tensor([1, 1])]; + int32 value_41_groups_0 = const()[name = string("value_41_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(808291392)))]; + tensor layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811568256)))]; + tensor value_41_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_bias_to_fp16, dilations = value_41_dilations_0, groups = value_41_groups_0, pad = value_41_pad_0, pad_type = value_41_pad_type_0, strides = value_41_strides_0, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = obj_81_cast_fp16)[name = string("value_41_cast_fp16")]; + tensor var_30802_begin_0 = const()[name = string("op_30802_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30802_end_0 = const()[name = string("op_30802_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30802_end_mask_0 = const()[name = string("op_30802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30802_cast_fp16 = slice_by_index(begin = var_30802_begin_0, end = var_30802_end_0, end_mask = var_30802_end_mask_0, x = query_41_cast_fp16)[name = string("op_30802_cast_fp16")]; + tensor var_30806_begin_0 = const()[name = string("op_30806_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30806_end_0 = const()[name = string("op_30806_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_30806_end_mask_0 = const()[name = string("op_30806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30806_cast_fp16 = slice_by_index(begin = var_30806_begin_0, end = var_30806_end_0, end_mask = var_30806_end_mask_0, x = query_41_cast_fp16)[name = string("op_30806_cast_fp16")]; + tensor var_30810_begin_0 = const()[name = string("op_30810_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30810_end_0 = const()[name = string("op_30810_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_30810_end_mask_0 = const()[name = string("op_30810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30810_cast_fp16 = slice_by_index(begin = var_30810_begin_0, end = var_30810_end_0, end_mask = var_30810_end_mask_0, x = query_41_cast_fp16)[name = string("op_30810_cast_fp16")]; + tensor var_30814_begin_0 = const()[name = string("op_30814_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30814_end_0 = const()[name = string("op_30814_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_30814_end_mask_0 = const()[name = string("op_30814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30814_cast_fp16 = slice_by_index(begin = var_30814_begin_0, end = var_30814_end_0, end_mask = var_30814_end_mask_0, x = query_41_cast_fp16)[name = string("op_30814_cast_fp16")]; + tensor var_30818_begin_0 = const()[name = string("op_30818_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30818_end_0 = const()[name = string("op_30818_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_30818_end_mask_0 = const()[name = string("op_30818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30818_cast_fp16 = slice_by_index(begin = var_30818_begin_0, end = var_30818_end_0, end_mask = var_30818_end_mask_0, x = query_41_cast_fp16)[name = string("op_30818_cast_fp16")]; + tensor var_30822_begin_0 = const()[name = string("op_30822_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30822_end_0 = const()[name = string("op_30822_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_30822_end_mask_0 = const()[name = string("op_30822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30822_cast_fp16 = slice_by_index(begin = var_30822_begin_0, end = var_30822_end_0, end_mask = var_30822_end_mask_0, x = query_41_cast_fp16)[name = string("op_30822_cast_fp16")]; + tensor var_30826_begin_0 = const()[name = string("op_30826_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30826_end_0 = const()[name = string("op_30826_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_30826_end_mask_0 = const()[name = string("op_30826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30826_cast_fp16 = slice_by_index(begin = var_30826_begin_0, end = var_30826_end_0, end_mask = var_30826_end_mask_0, x = query_41_cast_fp16)[name = string("op_30826_cast_fp16")]; + tensor var_30830_begin_0 = const()[name = string("op_30830_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30830_end_0 = const()[name = string("op_30830_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_30830_end_mask_0 = const()[name = string("op_30830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30830_cast_fp16 = slice_by_index(begin = var_30830_begin_0, end = var_30830_end_0, end_mask = var_30830_end_mask_0, x = query_41_cast_fp16)[name = string("op_30830_cast_fp16")]; + tensor var_30834_begin_0 = const()[name = string("op_30834_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30834_end_0 = const()[name = string("op_30834_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_30834_end_mask_0 = const()[name = string("op_30834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30834_cast_fp16 = slice_by_index(begin = var_30834_begin_0, end = var_30834_end_0, end_mask = var_30834_end_mask_0, x = query_41_cast_fp16)[name = string("op_30834_cast_fp16")]; + tensor var_30838_begin_0 = const()[name = string("op_30838_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30838_end_0 = const()[name = string("op_30838_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_30838_end_mask_0 = const()[name = string("op_30838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30838_cast_fp16 = slice_by_index(begin = var_30838_begin_0, end = var_30838_end_0, end_mask = var_30838_end_mask_0, x = query_41_cast_fp16)[name = string("op_30838_cast_fp16")]; + tensor var_30842_begin_0 = const()[name = string("op_30842_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30842_end_0 = const()[name = string("op_30842_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_30842_end_mask_0 = const()[name = string("op_30842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30842_cast_fp16 = slice_by_index(begin = var_30842_begin_0, end = var_30842_end_0, end_mask = var_30842_end_mask_0, x = query_41_cast_fp16)[name = string("op_30842_cast_fp16")]; + tensor var_30846_begin_0 = const()[name = string("op_30846_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30846_end_0 = const()[name = string("op_30846_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_30846_end_mask_0 = const()[name = string("op_30846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30846_cast_fp16 = slice_by_index(begin = var_30846_begin_0, end = var_30846_end_0, end_mask = var_30846_end_mask_0, x = query_41_cast_fp16)[name = string("op_30846_cast_fp16")]; + tensor var_30850_begin_0 = const()[name = string("op_30850_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30850_end_0 = const()[name = string("op_30850_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_30850_end_mask_0 = const()[name = string("op_30850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30850_cast_fp16 = slice_by_index(begin = var_30850_begin_0, end = var_30850_end_0, end_mask = var_30850_end_mask_0, x = query_41_cast_fp16)[name = string("op_30850_cast_fp16")]; + tensor var_30854_begin_0 = const()[name = string("op_30854_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30854_end_0 = const()[name = string("op_30854_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_30854_end_mask_0 = const()[name = string("op_30854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30854_cast_fp16 = slice_by_index(begin = var_30854_begin_0, end = var_30854_end_0, end_mask = var_30854_end_mask_0, x = query_41_cast_fp16)[name = string("op_30854_cast_fp16")]; + tensor var_30858_begin_0 = const()[name = string("op_30858_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30858_end_0 = const()[name = string("op_30858_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_30858_end_mask_0 = const()[name = string("op_30858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30858_cast_fp16 = slice_by_index(begin = var_30858_begin_0, end = var_30858_end_0, end_mask = var_30858_end_mask_0, x = query_41_cast_fp16)[name = string("op_30858_cast_fp16")]; + tensor var_30862_begin_0 = const()[name = string("op_30862_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30862_end_0 = const()[name = string("op_30862_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_30862_end_mask_0 = const()[name = string("op_30862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30862_cast_fp16 = slice_by_index(begin = var_30862_begin_0, end = var_30862_end_0, end_mask = var_30862_end_mask_0, x = query_41_cast_fp16)[name = string("op_30862_cast_fp16")]; + tensor var_30866_begin_0 = const()[name = string("op_30866_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30866_end_0 = const()[name = string("op_30866_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_30866_end_mask_0 = const()[name = string("op_30866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30866_cast_fp16 = slice_by_index(begin = var_30866_begin_0, end = var_30866_end_0, end_mask = var_30866_end_mask_0, x = query_41_cast_fp16)[name = string("op_30866_cast_fp16")]; + tensor var_30870_begin_0 = const()[name = string("op_30870_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30870_end_0 = const()[name = string("op_30870_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_30870_end_mask_0 = const()[name = string("op_30870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30870_cast_fp16 = slice_by_index(begin = var_30870_begin_0, end = var_30870_end_0, end_mask = var_30870_end_mask_0, x = query_41_cast_fp16)[name = string("op_30870_cast_fp16")]; + tensor var_30874_begin_0 = const()[name = string("op_30874_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30874_end_0 = const()[name = string("op_30874_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_30874_end_mask_0 = const()[name = string("op_30874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30874_cast_fp16 = slice_by_index(begin = var_30874_begin_0, end = var_30874_end_0, end_mask = var_30874_end_mask_0, x = query_41_cast_fp16)[name = string("op_30874_cast_fp16")]; + tensor var_30878_begin_0 = const()[name = string("op_30878_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30878_end_0 = const()[name = string("op_30878_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_30878_end_mask_0 = const()[name = string("op_30878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30878_cast_fp16 = slice_by_index(begin = var_30878_begin_0, end = var_30878_end_0, end_mask = var_30878_end_mask_0, x = query_41_cast_fp16)[name = string("op_30878_cast_fp16")]; + tensor var_30887_begin_0 = const()[name = string("op_30887_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30887_end_0 = const()[name = string("op_30887_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30887_end_mask_0 = const()[name = string("op_30887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30887_cast_fp16 = slice_by_index(begin = var_30887_begin_0, end = var_30887_end_0, end_mask = var_30887_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30887_cast_fp16")]; + tensor var_30894_begin_0 = const()[name = string("op_30894_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30894_end_0 = const()[name = string("op_30894_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30894_end_mask_0 = const()[name = string("op_30894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30894_cast_fp16 = slice_by_index(begin = var_30894_begin_0, end = var_30894_end_0, end_mask = var_30894_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30894_cast_fp16")]; + tensor var_30901_begin_0 = const()[name = string("op_30901_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30901_end_0 = const()[name = string("op_30901_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30901_end_mask_0 = const()[name = string("op_30901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30901_cast_fp16 = slice_by_index(begin = var_30901_begin_0, end = var_30901_end_0, end_mask = var_30901_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30901_cast_fp16")]; + tensor var_30908_begin_0 = const()[name = string("op_30908_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30908_end_0 = const()[name = string("op_30908_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30908_end_mask_0 = const()[name = string("op_30908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30908_cast_fp16 = slice_by_index(begin = var_30908_begin_0, end = var_30908_end_0, end_mask = var_30908_end_mask_0, x = var_30802_cast_fp16)[name = string("op_30908_cast_fp16")]; + tensor var_30915_begin_0 = const()[name = string("op_30915_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30915_end_0 = const()[name = string("op_30915_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30915_end_mask_0 = const()[name = string("op_30915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30915_cast_fp16 = slice_by_index(begin = var_30915_begin_0, end = var_30915_end_0, end_mask = var_30915_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30915_cast_fp16")]; + tensor var_30922_begin_0 = const()[name = string("op_30922_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30922_end_0 = const()[name = string("op_30922_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30922_end_mask_0 = const()[name = string("op_30922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30922_cast_fp16 = slice_by_index(begin = var_30922_begin_0, end = var_30922_end_0, end_mask = var_30922_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30922_cast_fp16")]; + tensor var_30929_begin_0 = const()[name = string("op_30929_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30929_end_0 = const()[name = string("op_30929_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30929_end_mask_0 = const()[name = string("op_30929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30929_cast_fp16 = slice_by_index(begin = var_30929_begin_0, end = var_30929_end_0, end_mask = var_30929_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30929_cast_fp16")]; + tensor var_30936_begin_0 = const()[name = string("op_30936_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30936_end_0 = const()[name = string("op_30936_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30936_end_mask_0 = const()[name = string("op_30936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30936_cast_fp16 = slice_by_index(begin = var_30936_begin_0, end = var_30936_end_0, end_mask = var_30936_end_mask_0, x = var_30806_cast_fp16)[name = string("op_30936_cast_fp16")]; + tensor var_30943_begin_0 = const()[name = string("op_30943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30943_end_0 = const()[name = string("op_30943_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30943_end_mask_0 = const()[name = string("op_30943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30943_cast_fp16 = slice_by_index(begin = var_30943_begin_0, end = var_30943_end_0, end_mask = var_30943_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30943_cast_fp16")]; + tensor var_30950_begin_0 = const()[name = string("op_30950_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30950_end_0 = const()[name = string("op_30950_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30950_end_mask_0 = const()[name = string("op_30950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30950_cast_fp16 = slice_by_index(begin = var_30950_begin_0, end = var_30950_end_0, end_mask = var_30950_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30950_cast_fp16")]; + tensor var_30957_begin_0 = const()[name = string("op_30957_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30957_end_0 = const()[name = string("op_30957_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30957_end_mask_0 = const()[name = string("op_30957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30957_cast_fp16 = slice_by_index(begin = var_30957_begin_0, end = var_30957_end_0, end_mask = var_30957_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30957_cast_fp16")]; + tensor var_30964_begin_0 = const()[name = string("op_30964_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30964_end_0 = const()[name = string("op_30964_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30964_end_mask_0 = const()[name = string("op_30964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30964_cast_fp16 = slice_by_index(begin = var_30964_begin_0, end = var_30964_end_0, end_mask = var_30964_end_mask_0, x = var_30810_cast_fp16)[name = string("op_30964_cast_fp16")]; + tensor var_30971_begin_0 = const()[name = string("op_30971_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30971_end_0 = const()[name = string("op_30971_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30971_end_mask_0 = const()[name = string("op_30971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30971_cast_fp16 = slice_by_index(begin = var_30971_begin_0, end = var_30971_end_0, end_mask = var_30971_end_mask_0, x = var_30814_cast_fp16)[name = string("op_30971_cast_fp16")]; + tensor var_30978_begin_0 = const()[name = string("op_30978_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30978_end_0 = const()[name = string("op_30978_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30978_end_mask_0 = const()[name = string("op_30978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30978_cast_fp16 = slice_by_index(begin = var_30978_begin_0, end = var_30978_end_0, end_mask = var_30978_end_mask_0, x = var_30814_cast_fp16)[name = string("op_30978_cast_fp16")]; + tensor var_30985_begin_0 = const()[name = string("op_30985_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30985_end_0 = const()[name = string("op_30985_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30985_end_mask_0 = const()[name = string("op_30985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30985_cast_fp16 = slice_by_index(begin = var_30985_begin_0, end = var_30985_end_0, end_mask = var_30985_end_mask_0, x = var_30814_cast_fp16)[name = string("op_30985_cast_fp16")]; + tensor var_30992_begin_0 = const()[name = string("op_30992_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30992_end_0 = const()[name = string("op_30992_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30992_end_mask_0 = const()[name = string("op_30992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30992_cast_fp16 = slice_by_index(begin = var_30992_begin_0, end = var_30992_end_0, end_mask = var_30992_end_mask_0, x = var_30814_cast_fp16)[name = string("op_30992_cast_fp16")]; + tensor var_30999_begin_0 = const()[name = string("op_30999_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30999_end_0 = const()[name = string("op_30999_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30999_end_mask_0 = const()[name = string("op_30999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30999_cast_fp16 = slice_by_index(begin = var_30999_begin_0, end = var_30999_end_0, end_mask = var_30999_end_mask_0, x = var_30818_cast_fp16)[name = string("op_30999_cast_fp16")]; + tensor var_31006_begin_0 = const()[name = string("op_31006_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31006_end_0 = const()[name = string("op_31006_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31006_end_mask_0 = const()[name = string("op_31006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31006_cast_fp16 = slice_by_index(begin = var_31006_begin_0, end = var_31006_end_0, end_mask = var_31006_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31006_cast_fp16")]; + tensor var_31013_begin_0 = const()[name = string("op_31013_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31013_end_0 = const()[name = string("op_31013_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31013_end_mask_0 = const()[name = string("op_31013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31013_cast_fp16 = slice_by_index(begin = var_31013_begin_0, end = var_31013_end_0, end_mask = var_31013_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31013_cast_fp16")]; + tensor var_31020_begin_0 = const()[name = string("op_31020_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31020_end_0 = const()[name = string("op_31020_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31020_end_mask_0 = const()[name = string("op_31020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31020_cast_fp16 = slice_by_index(begin = var_31020_begin_0, end = var_31020_end_0, end_mask = var_31020_end_mask_0, x = var_30818_cast_fp16)[name = string("op_31020_cast_fp16")]; + tensor var_31027_begin_0 = const()[name = string("op_31027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31027_end_0 = const()[name = string("op_31027_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31027_end_mask_0 = const()[name = string("op_31027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31027_cast_fp16 = slice_by_index(begin = var_31027_begin_0, end = var_31027_end_0, end_mask = var_31027_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31027_cast_fp16")]; + tensor var_31034_begin_0 = const()[name = string("op_31034_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31034_end_0 = const()[name = string("op_31034_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31034_end_mask_0 = const()[name = string("op_31034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31034_cast_fp16 = slice_by_index(begin = var_31034_begin_0, end = var_31034_end_0, end_mask = var_31034_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31034_cast_fp16")]; + tensor var_31041_begin_0 = const()[name = string("op_31041_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31041_end_0 = const()[name = string("op_31041_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31041_end_mask_0 = const()[name = string("op_31041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31041_cast_fp16 = slice_by_index(begin = var_31041_begin_0, end = var_31041_end_0, end_mask = var_31041_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31041_cast_fp16")]; + tensor var_31048_begin_0 = const()[name = string("op_31048_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31048_end_0 = const()[name = string("op_31048_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31048_end_mask_0 = const()[name = string("op_31048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31048_cast_fp16 = slice_by_index(begin = var_31048_begin_0, end = var_31048_end_0, end_mask = var_31048_end_mask_0, x = var_30822_cast_fp16)[name = string("op_31048_cast_fp16")]; + tensor var_31055_begin_0 = const()[name = string("op_31055_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31055_end_0 = const()[name = string("op_31055_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31055_end_mask_0 = const()[name = string("op_31055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31055_cast_fp16 = slice_by_index(begin = var_31055_begin_0, end = var_31055_end_0, end_mask = var_31055_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31055_cast_fp16")]; + tensor var_31062_begin_0 = const()[name = string("op_31062_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31062_end_0 = const()[name = string("op_31062_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31062_end_mask_0 = const()[name = string("op_31062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31062_cast_fp16 = slice_by_index(begin = var_31062_begin_0, end = var_31062_end_0, end_mask = var_31062_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31062_cast_fp16")]; + tensor var_31069_begin_0 = const()[name = string("op_31069_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31069_end_0 = const()[name = string("op_31069_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31069_end_mask_0 = const()[name = string("op_31069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31069_cast_fp16 = slice_by_index(begin = var_31069_begin_0, end = var_31069_end_0, end_mask = var_31069_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31069_cast_fp16")]; + tensor var_31076_begin_0 = const()[name = string("op_31076_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31076_end_0 = const()[name = string("op_31076_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31076_end_mask_0 = const()[name = string("op_31076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31076_cast_fp16 = slice_by_index(begin = var_31076_begin_0, end = var_31076_end_0, end_mask = var_31076_end_mask_0, x = var_30826_cast_fp16)[name = string("op_31076_cast_fp16")]; + tensor var_31083_begin_0 = const()[name = string("op_31083_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31083_end_0 = const()[name = string("op_31083_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31083_end_mask_0 = const()[name = string("op_31083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31083_cast_fp16 = slice_by_index(begin = var_31083_begin_0, end = var_31083_end_0, end_mask = var_31083_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31083_cast_fp16")]; + tensor var_31090_begin_0 = const()[name = string("op_31090_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31090_end_0 = const()[name = string("op_31090_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31090_end_mask_0 = const()[name = string("op_31090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31090_cast_fp16 = slice_by_index(begin = var_31090_begin_0, end = var_31090_end_0, end_mask = var_31090_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31090_cast_fp16")]; + tensor var_31097_begin_0 = const()[name = string("op_31097_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31097_end_0 = const()[name = string("op_31097_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31097_end_mask_0 = const()[name = string("op_31097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31097_cast_fp16 = slice_by_index(begin = var_31097_begin_0, end = var_31097_end_0, end_mask = var_31097_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31097_cast_fp16")]; + tensor var_31104_begin_0 = const()[name = string("op_31104_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31104_end_0 = const()[name = string("op_31104_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31104_end_mask_0 = const()[name = string("op_31104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31104_cast_fp16 = slice_by_index(begin = var_31104_begin_0, end = var_31104_end_0, end_mask = var_31104_end_mask_0, x = var_30830_cast_fp16)[name = string("op_31104_cast_fp16")]; + tensor var_31111_begin_0 = const()[name = string("op_31111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31111_end_0 = const()[name = string("op_31111_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31111_end_mask_0 = const()[name = string("op_31111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31111_cast_fp16 = slice_by_index(begin = var_31111_begin_0, end = var_31111_end_0, end_mask = var_31111_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31111_cast_fp16")]; + tensor var_31118_begin_0 = const()[name = string("op_31118_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31118_end_0 = const()[name = string("op_31118_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31118_end_mask_0 = const()[name = string("op_31118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31118_cast_fp16 = slice_by_index(begin = var_31118_begin_0, end = var_31118_end_0, end_mask = var_31118_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31118_cast_fp16")]; + tensor var_31125_begin_0 = const()[name = string("op_31125_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31125_end_0 = const()[name = string("op_31125_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31125_end_mask_0 = const()[name = string("op_31125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31125_cast_fp16 = slice_by_index(begin = var_31125_begin_0, end = var_31125_end_0, end_mask = var_31125_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31125_cast_fp16")]; + tensor var_31132_begin_0 = const()[name = string("op_31132_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31132_end_0 = const()[name = string("op_31132_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31132_end_mask_0 = const()[name = string("op_31132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31132_cast_fp16 = slice_by_index(begin = var_31132_begin_0, end = var_31132_end_0, end_mask = var_31132_end_mask_0, x = var_30834_cast_fp16)[name = string("op_31132_cast_fp16")]; + tensor var_31139_begin_0 = const()[name = string("op_31139_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31139_end_0 = const()[name = string("op_31139_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31139_end_mask_0 = const()[name = string("op_31139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31139_cast_fp16 = slice_by_index(begin = var_31139_begin_0, end = var_31139_end_0, end_mask = var_31139_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31139_cast_fp16")]; + tensor var_31146_begin_0 = const()[name = string("op_31146_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31146_end_0 = const()[name = string("op_31146_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31146_end_mask_0 = const()[name = string("op_31146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31146_cast_fp16 = slice_by_index(begin = var_31146_begin_0, end = var_31146_end_0, end_mask = var_31146_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31146_cast_fp16")]; + tensor var_31153_begin_0 = const()[name = string("op_31153_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31153_end_0 = const()[name = string("op_31153_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31153_end_mask_0 = const()[name = string("op_31153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31153_cast_fp16 = slice_by_index(begin = var_31153_begin_0, end = var_31153_end_0, end_mask = var_31153_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31153_cast_fp16")]; + tensor var_31160_begin_0 = const()[name = string("op_31160_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31160_end_0 = const()[name = string("op_31160_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31160_end_mask_0 = const()[name = string("op_31160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31160_cast_fp16 = slice_by_index(begin = var_31160_begin_0, end = var_31160_end_0, end_mask = var_31160_end_mask_0, x = var_30838_cast_fp16)[name = string("op_31160_cast_fp16")]; + tensor var_31167_begin_0 = const()[name = string("op_31167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31167_end_0 = const()[name = string("op_31167_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31167_end_mask_0 = const()[name = string("op_31167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31167_cast_fp16 = slice_by_index(begin = var_31167_begin_0, end = var_31167_end_0, end_mask = var_31167_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31167_cast_fp16")]; + tensor var_31174_begin_0 = const()[name = string("op_31174_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31174_end_0 = const()[name = string("op_31174_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31174_end_mask_0 = const()[name = string("op_31174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31174_cast_fp16 = slice_by_index(begin = var_31174_begin_0, end = var_31174_end_0, end_mask = var_31174_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31174_cast_fp16")]; + tensor var_31181_begin_0 = const()[name = string("op_31181_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31181_end_0 = const()[name = string("op_31181_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31181_end_mask_0 = const()[name = string("op_31181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31181_cast_fp16 = slice_by_index(begin = var_31181_begin_0, end = var_31181_end_0, end_mask = var_31181_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31181_cast_fp16")]; + tensor var_31188_begin_0 = const()[name = string("op_31188_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31188_end_0 = const()[name = string("op_31188_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31188_end_mask_0 = const()[name = string("op_31188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31188_cast_fp16 = slice_by_index(begin = var_31188_begin_0, end = var_31188_end_0, end_mask = var_31188_end_mask_0, x = var_30842_cast_fp16)[name = string("op_31188_cast_fp16")]; + tensor var_31195_begin_0 = const()[name = string("op_31195_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31195_end_0 = const()[name = string("op_31195_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31195_end_mask_0 = const()[name = string("op_31195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31195_cast_fp16 = slice_by_index(begin = var_31195_begin_0, end = var_31195_end_0, end_mask = var_31195_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31195_cast_fp16")]; + tensor var_31202_begin_0 = const()[name = string("op_31202_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31202_end_0 = const()[name = string("op_31202_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31202_end_mask_0 = const()[name = string("op_31202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31202_cast_fp16 = slice_by_index(begin = var_31202_begin_0, end = var_31202_end_0, end_mask = var_31202_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31202_cast_fp16")]; + tensor var_31209_begin_0 = const()[name = string("op_31209_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31209_end_0 = const()[name = string("op_31209_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31209_end_mask_0 = const()[name = string("op_31209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31209_cast_fp16 = slice_by_index(begin = var_31209_begin_0, end = var_31209_end_0, end_mask = var_31209_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31209_cast_fp16")]; + tensor var_31216_begin_0 = const()[name = string("op_31216_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31216_end_0 = const()[name = string("op_31216_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31216_end_mask_0 = const()[name = string("op_31216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31216_cast_fp16 = slice_by_index(begin = var_31216_begin_0, end = var_31216_end_0, end_mask = var_31216_end_mask_0, x = var_30846_cast_fp16)[name = string("op_31216_cast_fp16")]; + tensor var_31223_begin_0 = const()[name = string("op_31223_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31223_end_0 = const()[name = string("op_31223_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31223_end_mask_0 = const()[name = string("op_31223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31223_cast_fp16 = slice_by_index(begin = var_31223_begin_0, end = var_31223_end_0, end_mask = var_31223_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31223_cast_fp16")]; + tensor var_31230_begin_0 = const()[name = string("op_31230_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31230_end_0 = const()[name = string("op_31230_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31230_end_mask_0 = const()[name = string("op_31230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31230_cast_fp16 = slice_by_index(begin = var_31230_begin_0, end = var_31230_end_0, end_mask = var_31230_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31230_cast_fp16")]; + tensor var_31237_begin_0 = const()[name = string("op_31237_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31237_end_0 = const()[name = string("op_31237_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31237_end_mask_0 = const()[name = string("op_31237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31237_cast_fp16 = slice_by_index(begin = var_31237_begin_0, end = var_31237_end_0, end_mask = var_31237_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31237_cast_fp16")]; + tensor var_31244_begin_0 = const()[name = string("op_31244_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31244_end_0 = const()[name = string("op_31244_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31244_end_mask_0 = const()[name = string("op_31244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31244_cast_fp16 = slice_by_index(begin = var_31244_begin_0, end = var_31244_end_0, end_mask = var_31244_end_mask_0, x = var_30850_cast_fp16)[name = string("op_31244_cast_fp16")]; + tensor var_31251_begin_0 = const()[name = string("op_31251_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31251_end_0 = const()[name = string("op_31251_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31251_end_mask_0 = const()[name = string("op_31251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31251_cast_fp16 = slice_by_index(begin = var_31251_begin_0, end = var_31251_end_0, end_mask = var_31251_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31251_cast_fp16")]; + tensor var_31258_begin_0 = const()[name = string("op_31258_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31258_end_0 = const()[name = string("op_31258_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31258_end_mask_0 = const()[name = string("op_31258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31258_cast_fp16 = slice_by_index(begin = var_31258_begin_0, end = var_31258_end_0, end_mask = var_31258_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31258_cast_fp16")]; + tensor var_31265_begin_0 = const()[name = string("op_31265_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31265_end_0 = const()[name = string("op_31265_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31265_end_mask_0 = const()[name = string("op_31265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31265_cast_fp16 = slice_by_index(begin = var_31265_begin_0, end = var_31265_end_0, end_mask = var_31265_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31265_cast_fp16")]; + tensor var_31272_begin_0 = const()[name = string("op_31272_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31272_end_0 = const()[name = string("op_31272_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31272_end_mask_0 = const()[name = string("op_31272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31272_cast_fp16 = slice_by_index(begin = var_31272_begin_0, end = var_31272_end_0, end_mask = var_31272_end_mask_0, x = var_30854_cast_fp16)[name = string("op_31272_cast_fp16")]; + tensor var_31279_begin_0 = const()[name = string("op_31279_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31279_end_0 = const()[name = string("op_31279_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31279_end_mask_0 = const()[name = string("op_31279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31279_cast_fp16 = slice_by_index(begin = var_31279_begin_0, end = var_31279_end_0, end_mask = var_31279_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31279_cast_fp16")]; + tensor var_31286_begin_0 = const()[name = string("op_31286_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31286_end_0 = const()[name = string("op_31286_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31286_end_mask_0 = const()[name = string("op_31286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31286_cast_fp16 = slice_by_index(begin = var_31286_begin_0, end = var_31286_end_0, end_mask = var_31286_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31286_cast_fp16")]; + tensor var_31293_begin_0 = const()[name = string("op_31293_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31293_end_0 = const()[name = string("op_31293_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31293_end_mask_0 = const()[name = string("op_31293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31293_cast_fp16 = slice_by_index(begin = var_31293_begin_0, end = var_31293_end_0, end_mask = var_31293_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31293_cast_fp16")]; + tensor var_31300_begin_0 = const()[name = string("op_31300_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31300_end_0 = const()[name = string("op_31300_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31300_end_mask_0 = const()[name = string("op_31300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31300_cast_fp16 = slice_by_index(begin = var_31300_begin_0, end = var_31300_end_0, end_mask = var_31300_end_mask_0, x = var_30858_cast_fp16)[name = string("op_31300_cast_fp16")]; + tensor var_31307_begin_0 = const()[name = string("op_31307_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31307_end_0 = const()[name = string("op_31307_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31307_end_mask_0 = const()[name = string("op_31307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31307_cast_fp16 = slice_by_index(begin = var_31307_begin_0, end = var_31307_end_0, end_mask = var_31307_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31307_cast_fp16")]; + tensor var_31314_begin_0 = const()[name = string("op_31314_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31314_end_0 = const()[name = string("op_31314_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31314_end_mask_0 = const()[name = string("op_31314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31314_cast_fp16 = slice_by_index(begin = var_31314_begin_0, end = var_31314_end_0, end_mask = var_31314_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31314_cast_fp16")]; + tensor var_31321_begin_0 = const()[name = string("op_31321_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31321_end_0 = const()[name = string("op_31321_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31321_end_mask_0 = const()[name = string("op_31321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31321_cast_fp16 = slice_by_index(begin = var_31321_begin_0, end = var_31321_end_0, end_mask = var_31321_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31321_cast_fp16")]; + tensor var_31328_begin_0 = const()[name = string("op_31328_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31328_end_0 = const()[name = string("op_31328_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31328_end_mask_0 = const()[name = string("op_31328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31328_cast_fp16 = slice_by_index(begin = var_31328_begin_0, end = var_31328_end_0, end_mask = var_31328_end_mask_0, x = var_30862_cast_fp16)[name = string("op_31328_cast_fp16")]; + tensor var_31335_begin_0 = const()[name = string("op_31335_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31335_end_0 = const()[name = string("op_31335_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31335_end_mask_0 = const()[name = string("op_31335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31335_cast_fp16 = slice_by_index(begin = var_31335_begin_0, end = var_31335_end_0, end_mask = var_31335_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31335_cast_fp16")]; + tensor var_31342_begin_0 = const()[name = string("op_31342_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31342_end_0 = const()[name = string("op_31342_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31342_end_mask_0 = const()[name = string("op_31342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31342_cast_fp16 = slice_by_index(begin = var_31342_begin_0, end = var_31342_end_0, end_mask = var_31342_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31342_cast_fp16")]; + tensor var_31349_begin_0 = const()[name = string("op_31349_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31349_end_0 = const()[name = string("op_31349_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31349_end_mask_0 = const()[name = string("op_31349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31349_cast_fp16 = slice_by_index(begin = var_31349_begin_0, end = var_31349_end_0, end_mask = var_31349_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31349_cast_fp16")]; + tensor var_31356_begin_0 = const()[name = string("op_31356_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31356_end_0 = const()[name = string("op_31356_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31356_end_mask_0 = const()[name = string("op_31356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31356_cast_fp16 = slice_by_index(begin = var_31356_begin_0, end = var_31356_end_0, end_mask = var_31356_end_mask_0, x = var_30866_cast_fp16)[name = string("op_31356_cast_fp16")]; + tensor var_31363_begin_0 = const()[name = string("op_31363_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31363_end_0 = const()[name = string("op_31363_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31363_end_mask_0 = const()[name = string("op_31363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31363_cast_fp16 = slice_by_index(begin = var_31363_begin_0, end = var_31363_end_0, end_mask = var_31363_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31363_cast_fp16")]; + tensor var_31370_begin_0 = const()[name = string("op_31370_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31370_end_0 = const()[name = string("op_31370_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31370_end_mask_0 = const()[name = string("op_31370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31370_cast_fp16 = slice_by_index(begin = var_31370_begin_0, end = var_31370_end_0, end_mask = var_31370_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31370_cast_fp16")]; + tensor var_31377_begin_0 = const()[name = string("op_31377_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31377_end_0 = const()[name = string("op_31377_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31377_end_mask_0 = const()[name = string("op_31377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31377_cast_fp16 = slice_by_index(begin = var_31377_begin_0, end = var_31377_end_0, end_mask = var_31377_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31377_cast_fp16")]; + tensor var_31384_begin_0 = const()[name = string("op_31384_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31384_end_0 = const()[name = string("op_31384_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31384_end_mask_0 = const()[name = string("op_31384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31384_cast_fp16 = slice_by_index(begin = var_31384_begin_0, end = var_31384_end_0, end_mask = var_31384_end_mask_0, x = var_30870_cast_fp16)[name = string("op_31384_cast_fp16")]; + tensor var_31391_begin_0 = const()[name = string("op_31391_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31391_end_0 = const()[name = string("op_31391_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31391_end_mask_0 = const()[name = string("op_31391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31391_cast_fp16 = slice_by_index(begin = var_31391_begin_0, end = var_31391_end_0, end_mask = var_31391_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31391_cast_fp16")]; + tensor var_31398_begin_0 = const()[name = string("op_31398_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31398_end_0 = const()[name = string("op_31398_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31398_end_mask_0 = const()[name = string("op_31398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31398_cast_fp16 = slice_by_index(begin = var_31398_begin_0, end = var_31398_end_0, end_mask = var_31398_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31398_cast_fp16")]; + tensor var_31405_begin_0 = const()[name = string("op_31405_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31405_end_0 = const()[name = string("op_31405_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31405_end_mask_0 = const()[name = string("op_31405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31405_cast_fp16 = slice_by_index(begin = var_31405_begin_0, end = var_31405_end_0, end_mask = var_31405_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31405_cast_fp16")]; + tensor var_31412_begin_0 = const()[name = string("op_31412_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31412_end_0 = const()[name = string("op_31412_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31412_end_mask_0 = const()[name = string("op_31412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31412_cast_fp16 = slice_by_index(begin = var_31412_begin_0, end = var_31412_end_0, end_mask = var_31412_end_mask_0, x = var_30874_cast_fp16)[name = string("op_31412_cast_fp16")]; + tensor var_31419_begin_0 = const()[name = string("op_31419_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31419_end_0 = const()[name = string("op_31419_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31419_end_mask_0 = const()[name = string("op_31419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31419_cast_fp16 = slice_by_index(begin = var_31419_begin_0, end = var_31419_end_0, end_mask = var_31419_end_mask_0, x = var_30878_cast_fp16)[name = string("op_31419_cast_fp16")]; + tensor var_31426_begin_0 = const()[name = string("op_31426_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31426_end_0 = const()[name = string("op_31426_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31426_end_mask_0 = const()[name = string("op_31426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31426_cast_fp16 = slice_by_index(begin = var_31426_begin_0, end = var_31426_end_0, end_mask = var_31426_end_mask_0, x = var_30878_cast_fp16)[name = string("op_31426_cast_fp16")]; + tensor var_31433_begin_0 = const()[name = string("op_31433_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31433_end_0 = const()[name = string("op_31433_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31433_end_mask_0 = const()[name = string("op_31433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31433_cast_fp16 = slice_by_index(begin = var_31433_begin_0, end = var_31433_end_0, end_mask = var_31433_end_mask_0, x = var_30878_cast_fp16)[name = string("op_31433_cast_fp16")]; + tensor var_31440_begin_0 = const()[name = string("op_31440_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31440_end_0 = const()[name = string("op_31440_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31440_end_mask_0 = const()[name = string("op_31440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31440_cast_fp16 = slice_by_index(begin = var_31440_begin_0, end = var_31440_end_0, end_mask = var_31440_end_mask_0, x = var_30878_cast_fp16)[name = string("op_31440_cast_fp16")]; + tensor k_41_perm_0 = const()[name = string("k_41_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_31445_begin_0 = const()[name = string("op_31445_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31445_end_0 = const()[name = string("op_31445_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_31445_end_mask_0 = const()[name = string("op_31445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_41_cast_fp16 = transpose(perm = k_41_perm_0, x = key_41_cast_fp16)[name = string("transpose_11")]; + tensor var_31445_cast_fp16 = slice_by_index(begin = var_31445_begin_0, end = var_31445_end_0, end_mask = var_31445_end_mask_0, x = k_41_cast_fp16)[name = string("op_31445_cast_fp16")]; + tensor var_31449_begin_0 = const()[name = string("op_31449_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_31449_end_0 = const()[name = string("op_31449_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_31449_end_mask_0 = const()[name = string("op_31449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31449_cast_fp16 = slice_by_index(begin = var_31449_begin_0, end = var_31449_end_0, end_mask = var_31449_end_mask_0, x = k_41_cast_fp16)[name = string("op_31449_cast_fp16")]; + tensor var_31453_begin_0 = const()[name = string("op_31453_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_31453_end_0 = const()[name = string("op_31453_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_31453_end_mask_0 = const()[name = string("op_31453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31453_cast_fp16 = slice_by_index(begin = var_31453_begin_0, end = var_31453_end_0, end_mask = var_31453_end_mask_0, x = k_41_cast_fp16)[name = string("op_31453_cast_fp16")]; + tensor var_31457_begin_0 = const()[name = string("op_31457_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_31457_end_0 = const()[name = string("op_31457_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_31457_end_mask_0 = const()[name = string("op_31457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31457_cast_fp16 = slice_by_index(begin = var_31457_begin_0, end = var_31457_end_0, end_mask = var_31457_end_mask_0, x = k_41_cast_fp16)[name = string("op_31457_cast_fp16")]; + tensor var_31461_begin_0 = const()[name = string("op_31461_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_31461_end_0 = const()[name = string("op_31461_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_31461_end_mask_0 = const()[name = string("op_31461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31461_cast_fp16 = slice_by_index(begin = var_31461_begin_0, end = var_31461_end_0, end_mask = var_31461_end_mask_0, x = k_41_cast_fp16)[name = string("op_31461_cast_fp16")]; + tensor var_31465_begin_0 = const()[name = string("op_31465_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_31465_end_0 = const()[name = string("op_31465_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_31465_end_mask_0 = const()[name = string("op_31465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31465_cast_fp16 = slice_by_index(begin = var_31465_begin_0, end = var_31465_end_0, end_mask = var_31465_end_mask_0, x = k_41_cast_fp16)[name = string("op_31465_cast_fp16")]; + tensor var_31469_begin_0 = const()[name = string("op_31469_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_31469_end_0 = const()[name = string("op_31469_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_31469_end_mask_0 = const()[name = string("op_31469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31469_cast_fp16 = slice_by_index(begin = var_31469_begin_0, end = var_31469_end_0, end_mask = var_31469_end_mask_0, x = k_41_cast_fp16)[name = string("op_31469_cast_fp16")]; + tensor var_31473_begin_0 = const()[name = string("op_31473_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_31473_end_0 = const()[name = string("op_31473_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_31473_end_mask_0 = const()[name = string("op_31473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31473_cast_fp16 = slice_by_index(begin = var_31473_begin_0, end = var_31473_end_0, end_mask = var_31473_end_mask_0, x = k_41_cast_fp16)[name = string("op_31473_cast_fp16")]; + tensor var_31477_begin_0 = const()[name = string("op_31477_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_31477_end_0 = const()[name = string("op_31477_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_31477_end_mask_0 = const()[name = string("op_31477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31477_cast_fp16 = slice_by_index(begin = var_31477_begin_0, end = var_31477_end_0, end_mask = var_31477_end_mask_0, x = k_41_cast_fp16)[name = string("op_31477_cast_fp16")]; + tensor var_31481_begin_0 = const()[name = string("op_31481_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_31481_end_0 = const()[name = string("op_31481_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_31481_end_mask_0 = const()[name = string("op_31481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31481_cast_fp16 = slice_by_index(begin = var_31481_begin_0, end = var_31481_end_0, end_mask = var_31481_end_mask_0, x = k_41_cast_fp16)[name = string("op_31481_cast_fp16")]; + tensor var_31485_begin_0 = const()[name = string("op_31485_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_31485_end_0 = const()[name = string("op_31485_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_31485_end_mask_0 = const()[name = string("op_31485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31485_cast_fp16 = slice_by_index(begin = var_31485_begin_0, end = var_31485_end_0, end_mask = var_31485_end_mask_0, x = k_41_cast_fp16)[name = string("op_31485_cast_fp16")]; + tensor var_31489_begin_0 = const()[name = string("op_31489_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_31489_end_0 = const()[name = string("op_31489_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_31489_end_mask_0 = const()[name = string("op_31489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31489_cast_fp16 = slice_by_index(begin = var_31489_begin_0, end = var_31489_end_0, end_mask = var_31489_end_mask_0, x = k_41_cast_fp16)[name = string("op_31489_cast_fp16")]; + tensor var_31493_begin_0 = const()[name = string("op_31493_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_31493_end_0 = const()[name = string("op_31493_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_31493_end_mask_0 = const()[name = string("op_31493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31493_cast_fp16 = slice_by_index(begin = var_31493_begin_0, end = var_31493_end_0, end_mask = var_31493_end_mask_0, x = k_41_cast_fp16)[name = string("op_31493_cast_fp16")]; + tensor var_31497_begin_0 = const()[name = string("op_31497_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_31497_end_0 = const()[name = string("op_31497_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_31497_end_mask_0 = const()[name = string("op_31497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31497_cast_fp16 = slice_by_index(begin = var_31497_begin_0, end = var_31497_end_0, end_mask = var_31497_end_mask_0, x = k_41_cast_fp16)[name = string("op_31497_cast_fp16")]; + tensor var_31501_begin_0 = const()[name = string("op_31501_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_31501_end_0 = const()[name = string("op_31501_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_31501_end_mask_0 = const()[name = string("op_31501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31501_cast_fp16 = slice_by_index(begin = var_31501_begin_0, end = var_31501_end_0, end_mask = var_31501_end_mask_0, x = k_41_cast_fp16)[name = string("op_31501_cast_fp16")]; + tensor var_31505_begin_0 = const()[name = string("op_31505_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_31505_end_0 = const()[name = string("op_31505_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_31505_end_mask_0 = const()[name = string("op_31505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31505_cast_fp16 = slice_by_index(begin = var_31505_begin_0, end = var_31505_end_0, end_mask = var_31505_end_mask_0, x = k_41_cast_fp16)[name = string("op_31505_cast_fp16")]; + tensor var_31509_begin_0 = const()[name = string("op_31509_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_31509_end_0 = const()[name = string("op_31509_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_31509_end_mask_0 = const()[name = string("op_31509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31509_cast_fp16 = slice_by_index(begin = var_31509_begin_0, end = var_31509_end_0, end_mask = var_31509_end_mask_0, x = k_41_cast_fp16)[name = string("op_31509_cast_fp16")]; + tensor var_31513_begin_0 = const()[name = string("op_31513_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_31513_end_0 = const()[name = string("op_31513_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_31513_end_mask_0 = const()[name = string("op_31513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31513_cast_fp16 = slice_by_index(begin = var_31513_begin_0, end = var_31513_end_0, end_mask = var_31513_end_mask_0, x = k_41_cast_fp16)[name = string("op_31513_cast_fp16")]; + tensor var_31517_begin_0 = const()[name = string("op_31517_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_31517_end_0 = const()[name = string("op_31517_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_31517_end_mask_0 = const()[name = string("op_31517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31517_cast_fp16 = slice_by_index(begin = var_31517_begin_0, end = var_31517_end_0, end_mask = var_31517_end_mask_0, x = k_41_cast_fp16)[name = string("op_31517_cast_fp16")]; + tensor var_31521_begin_0 = const()[name = string("op_31521_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_31521_end_0 = const()[name = string("op_31521_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_31521_end_mask_0 = const()[name = string("op_31521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31521_cast_fp16 = slice_by_index(begin = var_31521_begin_0, end = var_31521_end_0, end_mask = var_31521_end_mask_0, x = k_41_cast_fp16)[name = string("op_31521_cast_fp16")]; + tensor var_31523_begin_0 = const()[name = string("op_31523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31523_end_0 = const()[name = string("op_31523_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31523_end_mask_0 = const()[name = string("op_31523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31523_cast_fp16 = slice_by_index(begin = var_31523_begin_0, end = var_31523_end_0, end_mask = var_31523_end_mask_0, x = value_41_cast_fp16)[name = string("op_31523_cast_fp16")]; + tensor var_31527_begin_0 = const()[name = string("op_31527_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31527_end_0 = const()[name = string("op_31527_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_31527_end_mask_0 = const()[name = string("op_31527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31527_cast_fp16 = slice_by_index(begin = var_31527_begin_0, end = var_31527_end_0, end_mask = var_31527_end_mask_0, x = value_41_cast_fp16)[name = string("op_31527_cast_fp16")]; + tensor var_31531_begin_0 = const()[name = string("op_31531_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31531_end_0 = const()[name = string("op_31531_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_31531_end_mask_0 = const()[name = string("op_31531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31531_cast_fp16 = slice_by_index(begin = var_31531_begin_0, end = var_31531_end_0, end_mask = var_31531_end_mask_0, x = value_41_cast_fp16)[name = string("op_31531_cast_fp16")]; + tensor var_31535_begin_0 = const()[name = string("op_31535_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31535_end_0 = const()[name = string("op_31535_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_31535_end_mask_0 = const()[name = string("op_31535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31535_cast_fp16 = slice_by_index(begin = var_31535_begin_0, end = var_31535_end_0, end_mask = var_31535_end_mask_0, x = value_41_cast_fp16)[name = string("op_31535_cast_fp16")]; + tensor var_31539_begin_0 = const()[name = string("op_31539_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31539_end_0 = const()[name = string("op_31539_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_31539_end_mask_0 = const()[name = string("op_31539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31539_cast_fp16 = slice_by_index(begin = var_31539_begin_0, end = var_31539_end_0, end_mask = var_31539_end_mask_0, x = value_41_cast_fp16)[name = string("op_31539_cast_fp16")]; + tensor var_31543_begin_0 = const()[name = string("op_31543_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31543_end_0 = const()[name = string("op_31543_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_31543_end_mask_0 = const()[name = string("op_31543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31543_cast_fp16 = slice_by_index(begin = var_31543_begin_0, end = var_31543_end_0, end_mask = var_31543_end_mask_0, x = value_41_cast_fp16)[name = string("op_31543_cast_fp16")]; + tensor var_31547_begin_0 = const()[name = string("op_31547_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31547_end_0 = const()[name = string("op_31547_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_31547_end_mask_0 = const()[name = string("op_31547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31547_cast_fp16 = slice_by_index(begin = var_31547_begin_0, end = var_31547_end_0, end_mask = var_31547_end_mask_0, x = value_41_cast_fp16)[name = string("op_31547_cast_fp16")]; + tensor var_31551_begin_0 = const()[name = string("op_31551_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31551_end_0 = const()[name = string("op_31551_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_31551_end_mask_0 = const()[name = string("op_31551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31551_cast_fp16 = slice_by_index(begin = var_31551_begin_0, end = var_31551_end_0, end_mask = var_31551_end_mask_0, x = value_41_cast_fp16)[name = string("op_31551_cast_fp16")]; + tensor var_31555_begin_0 = const()[name = string("op_31555_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31555_end_0 = const()[name = string("op_31555_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_31555_end_mask_0 = const()[name = string("op_31555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31555_cast_fp16 = slice_by_index(begin = var_31555_begin_0, end = var_31555_end_0, end_mask = var_31555_end_mask_0, x = value_41_cast_fp16)[name = string("op_31555_cast_fp16")]; + tensor var_31559_begin_0 = const()[name = string("op_31559_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31559_end_0 = const()[name = string("op_31559_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_31559_end_mask_0 = const()[name = string("op_31559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31559_cast_fp16 = slice_by_index(begin = var_31559_begin_0, end = var_31559_end_0, end_mask = var_31559_end_mask_0, x = value_41_cast_fp16)[name = string("op_31559_cast_fp16")]; + tensor var_31563_begin_0 = const()[name = string("op_31563_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31563_end_0 = const()[name = string("op_31563_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_31563_end_mask_0 = const()[name = string("op_31563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31563_cast_fp16 = slice_by_index(begin = var_31563_begin_0, end = var_31563_end_0, end_mask = var_31563_end_mask_0, x = value_41_cast_fp16)[name = string("op_31563_cast_fp16")]; + tensor var_31567_begin_0 = const()[name = string("op_31567_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31567_end_0 = const()[name = string("op_31567_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_31567_end_mask_0 = const()[name = string("op_31567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31567_cast_fp16 = slice_by_index(begin = var_31567_begin_0, end = var_31567_end_0, end_mask = var_31567_end_mask_0, x = value_41_cast_fp16)[name = string("op_31567_cast_fp16")]; + tensor var_31571_begin_0 = const()[name = string("op_31571_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31571_end_0 = const()[name = string("op_31571_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_31571_end_mask_0 = const()[name = string("op_31571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31571_cast_fp16 = slice_by_index(begin = var_31571_begin_0, end = var_31571_end_0, end_mask = var_31571_end_mask_0, x = value_41_cast_fp16)[name = string("op_31571_cast_fp16")]; + tensor var_31575_begin_0 = const()[name = string("op_31575_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31575_end_0 = const()[name = string("op_31575_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_31575_end_mask_0 = const()[name = string("op_31575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31575_cast_fp16 = slice_by_index(begin = var_31575_begin_0, end = var_31575_end_0, end_mask = var_31575_end_mask_0, x = value_41_cast_fp16)[name = string("op_31575_cast_fp16")]; + tensor var_31579_begin_0 = const()[name = string("op_31579_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31579_end_0 = const()[name = string("op_31579_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_31579_end_mask_0 = const()[name = string("op_31579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31579_cast_fp16 = slice_by_index(begin = var_31579_begin_0, end = var_31579_end_0, end_mask = var_31579_end_mask_0, x = value_41_cast_fp16)[name = string("op_31579_cast_fp16")]; + tensor var_31583_begin_0 = const()[name = string("op_31583_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31583_end_0 = const()[name = string("op_31583_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_31583_end_mask_0 = const()[name = string("op_31583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31583_cast_fp16 = slice_by_index(begin = var_31583_begin_0, end = var_31583_end_0, end_mask = var_31583_end_mask_0, x = value_41_cast_fp16)[name = string("op_31583_cast_fp16")]; + tensor var_31587_begin_0 = const()[name = string("op_31587_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31587_end_0 = const()[name = string("op_31587_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_31587_end_mask_0 = const()[name = string("op_31587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31587_cast_fp16 = slice_by_index(begin = var_31587_begin_0, end = var_31587_end_0, end_mask = var_31587_end_mask_0, x = value_41_cast_fp16)[name = string("op_31587_cast_fp16")]; + tensor var_31591_begin_0 = const()[name = string("op_31591_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31591_end_0 = const()[name = string("op_31591_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_31591_end_mask_0 = const()[name = string("op_31591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31591_cast_fp16 = slice_by_index(begin = var_31591_begin_0, end = var_31591_end_0, end_mask = var_31591_end_mask_0, x = value_41_cast_fp16)[name = string("op_31591_cast_fp16")]; + tensor var_31595_begin_0 = const()[name = string("op_31595_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31595_end_0 = const()[name = string("op_31595_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_31595_end_mask_0 = const()[name = string("op_31595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31595_cast_fp16 = slice_by_index(begin = var_31595_begin_0, end = var_31595_end_0, end_mask = var_31595_end_mask_0, x = value_41_cast_fp16)[name = string("op_31595_cast_fp16")]; + tensor var_31599_begin_0 = const()[name = string("op_31599_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31599_end_0 = const()[name = string("op_31599_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_31599_end_mask_0 = const()[name = string("op_31599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31599_cast_fp16 = slice_by_index(begin = var_31599_begin_0, end = var_31599_end_0, end_mask = var_31599_end_mask_0, x = value_41_cast_fp16)[name = string("op_31599_cast_fp16")]; + string _SplitHeadsQ__mh_w_3201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3201_equation_0, values = (var_31445_cast_fp16, var_30887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3201_cast_fp16")]; + string _SplitHeadsQ__mh_w_3203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3203_equation_0, values = (var_31445_cast_fp16, var_30894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3203_cast_fp16")]; + string _SplitHeadsQ__mh_w_3205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3205_equation_0, values = (var_31445_cast_fp16, var_30901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3205_cast_fp16")]; + string _SplitHeadsQ__mh_w_3207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3207_equation_0, values = (var_31445_cast_fp16, var_30908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3207_cast_fp16")]; + string _SplitHeadsQ__mh_w_3209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3209_equation_0, values = (var_31449_cast_fp16, var_30915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3209_cast_fp16")]; + string _SplitHeadsQ__mh_w_3211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3211_equation_0, values = (var_31449_cast_fp16, var_30922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3211_cast_fp16")]; + string _SplitHeadsQ__mh_w_3213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3213_equation_0, values = (var_31449_cast_fp16, var_30929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3213_cast_fp16")]; + string _SplitHeadsQ__mh_w_3215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3215_equation_0, values = (var_31449_cast_fp16, var_30936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3215_cast_fp16")]; + string _SplitHeadsQ__mh_w_3217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3217_equation_0, values = (var_31453_cast_fp16, var_30943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3217_cast_fp16")]; + string _SplitHeadsQ__mh_w_3219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3219_equation_0, values = (var_31453_cast_fp16, var_30950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3219_cast_fp16")]; + string _SplitHeadsQ__mh_w_3221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3221_equation_0, values = (var_31453_cast_fp16, var_30957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3221_cast_fp16")]; + string _SplitHeadsQ__mh_w_3223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3223_equation_0, values = (var_31453_cast_fp16, var_30964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3223_cast_fp16")]; + string _SplitHeadsQ__mh_w_3225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3225_equation_0, values = (var_31457_cast_fp16, var_30971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3225_cast_fp16")]; + string _SplitHeadsQ__mh_w_3227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3227_equation_0, values = (var_31457_cast_fp16, var_30978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3227_cast_fp16")]; + string _SplitHeadsQ__mh_w_3229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3229_equation_0, values = (var_31457_cast_fp16, var_30985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3229_cast_fp16")]; + string _SplitHeadsQ__mh_w_3231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3231_equation_0, values = (var_31457_cast_fp16, var_30992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3231_cast_fp16")]; + string _SplitHeadsQ__mh_w_3233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3233_equation_0, values = (var_31461_cast_fp16, var_30999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3233_cast_fp16")]; + string _SplitHeadsQ__mh_w_3235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3235_equation_0, values = (var_31461_cast_fp16, var_31006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3235_cast_fp16")]; + string _SplitHeadsQ__mh_w_3237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3237_equation_0, values = (var_31461_cast_fp16, var_31013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3237_cast_fp16")]; + string _SplitHeadsQ__mh_w_3239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3239_equation_0, values = (var_31461_cast_fp16, var_31020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3239_cast_fp16")]; + string _SplitHeadsQ__mh_w_3241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3241_equation_0, values = (var_31465_cast_fp16, var_31027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3241_cast_fp16")]; + string _SplitHeadsQ__mh_w_3243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3243_equation_0, values = (var_31465_cast_fp16, var_31034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3243_cast_fp16")]; + string _SplitHeadsQ__mh_w_3245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3245_equation_0, values = (var_31465_cast_fp16, var_31041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3245_cast_fp16")]; + string _SplitHeadsQ__mh_w_3247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3247_equation_0, values = (var_31465_cast_fp16, var_31048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3247_cast_fp16")]; + string _SplitHeadsQ__mh_w_3249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3249_equation_0, values = (var_31469_cast_fp16, var_31055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3249_cast_fp16")]; + string _SplitHeadsQ__mh_w_3251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3251_equation_0, values = (var_31469_cast_fp16, var_31062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3251_cast_fp16")]; + string _SplitHeadsQ__mh_w_3253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3253_equation_0, values = (var_31469_cast_fp16, var_31069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3253_cast_fp16")]; + string _SplitHeadsQ__mh_w_3255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3255_equation_0, values = (var_31469_cast_fp16, var_31076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3255_cast_fp16")]; + string _SplitHeadsQ__mh_w_3257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3257_equation_0, values = (var_31473_cast_fp16, var_31083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3257_cast_fp16")]; + string _SplitHeadsQ__mh_w_3259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3259_equation_0, values = (var_31473_cast_fp16, var_31090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3259_cast_fp16")]; + string _SplitHeadsQ__mh_w_3261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3261_equation_0, values = (var_31473_cast_fp16, var_31097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3261_cast_fp16")]; + string _SplitHeadsQ__mh_w_3263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3263_equation_0, values = (var_31473_cast_fp16, var_31104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3263_cast_fp16")]; + string _SplitHeadsQ__mh_w_3265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3265_equation_0, values = (var_31477_cast_fp16, var_31111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3265_cast_fp16")]; + string _SplitHeadsQ__mh_w_3267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3267_equation_0, values = (var_31477_cast_fp16, var_31118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3267_cast_fp16")]; + string _SplitHeadsQ__mh_w_3269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3269_equation_0, values = (var_31477_cast_fp16, var_31125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3269_cast_fp16")]; + string _SplitHeadsQ__mh_w_3271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3271_equation_0, values = (var_31477_cast_fp16, var_31132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3271_cast_fp16")]; + string _SplitHeadsQ__mh_w_3273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3273_equation_0, values = (var_31481_cast_fp16, var_31139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3273_cast_fp16")]; + string _SplitHeadsQ__mh_w_3275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3275_equation_0, values = (var_31481_cast_fp16, var_31146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3275_cast_fp16")]; + string _SplitHeadsQ__mh_w_3277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3277_equation_0, values = (var_31481_cast_fp16, var_31153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3277_cast_fp16")]; + string _SplitHeadsQ__mh_w_3279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3279_equation_0, values = (var_31481_cast_fp16, var_31160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3279_cast_fp16")]; + string _SplitHeadsQ__mh_w_3281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3281_equation_0, values = (var_31485_cast_fp16, var_31167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3281_cast_fp16")]; + string _SplitHeadsQ__mh_w_3283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3283_equation_0, values = (var_31485_cast_fp16, var_31174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3283_cast_fp16")]; + string _SplitHeadsQ__mh_w_3285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3285_equation_0, values = (var_31485_cast_fp16, var_31181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3285_cast_fp16")]; + string _SplitHeadsQ__mh_w_3287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3287_equation_0, values = (var_31485_cast_fp16, var_31188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3287_cast_fp16")]; + string _SplitHeadsQ__mh_w_3289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3289_equation_0, values = (var_31489_cast_fp16, var_31195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3289_cast_fp16")]; + string _SplitHeadsQ__mh_w_3291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3291_equation_0, values = (var_31489_cast_fp16, var_31202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3291_cast_fp16")]; + string _SplitHeadsQ__mh_w_3293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3293_equation_0, values = (var_31489_cast_fp16, var_31209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3293_cast_fp16")]; + string _SplitHeadsQ__mh_w_3295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3295_equation_0, values = (var_31489_cast_fp16, var_31216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3295_cast_fp16")]; + string _SplitHeadsQ__mh_w_3297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3297_equation_0, values = (var_31493_cast_fp16, var_31223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3297_cast_fp16")]; + string _SplitHeadsQ__mh_w_3299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3299_equation_0, values = (var_31493_cast_fp16, var_31230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3299_cast_fp16")]; + string _SplitHeadsQ__mh_w_3301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3301_equation_0, values = (var_31493_cast_fp16, var_31237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3301_cast_fp16")]; + string _SplitHeadsQ__mh_w_3303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3303_equation_0, values = (var_31493_cast_fp16, var_31244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3303_cast_fp16")]; + string _SplitHeadsQ__mh_w_3305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3305_equation_0, values = (var_31497_cast_fp16, var_31251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3305_cast_fp16")]; + string _SplitHeadsQ__mh_w_3307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3307_equation_0, values = (var_31497_cast_fp16, var_31258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3307_cast_fp16")]; + string _SplitHeadsQ__mh_w_3309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3309_equation_0, values = (var_31497_cast_fp16, var_31265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3309_cast_fp16")]; + string _SplitHeadsQ__mh_w_3311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3311_equation_0, values = (var_31497_cast_fp16, var_31272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3311_cast_fp16")]; + string _SplitHeadsQ__mh_w_3313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3313_equation_0, values = (var_31501_cast_fp16, var_31279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3313_cast_fp16")]; + string _SplitHeadsQ__mh_w_3315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3315_equation_0, values = (var_31501_cast_fp16, var_31286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3315_cast_fp16")]; + string _SplitHeadsQ__mh_w_3317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3317_equation_0, values = (var_31501_cast_fp16, var_31293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3317_cast_fp16")]; + string _SplitHeadsQ__mh_w_3319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3319_equation_0, values = (var_31501_cast_fp16, var_31300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3319_cast_fp16")]; + string _SplitHeadsQ__mh_w_3321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3321_equation_0, values = (var_31505_cast_fp16, var_31307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3321_cast_fp16")]; + string _SplitHeadsQ__mh_w_3323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3323_equation_0, values = (var_31505_cast_fp16, var_31314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3323_cast_fp16")]; + string _SplitHeadsQ__mh_w_3325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3325_equation_0, values = (var_31505_cast_fp16, var_31321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3325_cast_fp16")]; + string _SplitHeadsQ__mh_w_3327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3327_equation_0, values = (var_31505_cast_fp16, var_31328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3327_cast_fp16")]; + string _SplitHeadsQ__mh_w_3329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3329_equation_0, values = (var_31509_cast_fp16, var_31335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3329_cast_fp16")]; + string _SplitHeadsQ__mh_w_3331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3331_equation_0, values = (var_31509_cast_fp16, var_31342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3331_cast_fp16")]; + string _SplitHeadsQ__mh_w_3333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3333_equation_0, values = (var_31509_cast_fp16, var_31349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3333_cast_fp16")]; + string _SplitHeadsQ__mh_w_3335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3335_equation_0, values = (var_31509_cast_fp16, var_31356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3335_cast_fp16")]; + string _SplitHeadsQ__mh_w_3337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3337_equation_0, values = (var_31513_cast_fp16, var_31363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3337_cast_fp16")]; + string _SplitHeadsQ__mh_w_3339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3339_equation_0, values = (var_31513_cast_fp16, var_31370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3339_cast_fp16")]; + string _SplitHeadsQ__mh_w_3341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3341_equation_0, values = (var_31513_cast_fp16, var_31377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3341_cast_fp16")]; + string _SplitHeadsQ__mh_w_3343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3343_equation_0, values = (var_31513_cast_fp16, var_31384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3343_cast_fp16")]; + string _SplitHeadsQ__mh_w_3345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3345_equation_0, values = (var_31517_cast_fp16, var_31391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3345_cast_fp16")]; + string _SplitHeadsQ__mh_w_3347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3347_equation_0, values = (var_31517_cast_fp16, var_31398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3347_cast_fp16")]; + string _SplitHeadsQ__mh_w_3349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3349_equation_0, values = (var_31517_cast_fp16, var_31405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3349_cast_fp16")]; + string _SplitHeadsQ__mh_w_3351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3351_equation_0, values = (var_31517_cast_fp16, var_31412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3351_cast_fp16")]; + string _SplitHeadsQ__mh_w_3353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3353_equation_0, values = (var_31521_cast_fp16, var_31419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3353_cast_fp16")]; + string _SplitHeadsQ__mh_w_3355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3355_equation_0, values = (var_31521_cast_fp16, var_31426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3355_cast_fp16")]; + string _SplitHeadsQ__mh_w_3357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3357_equation_0, values = (var_31521_cast_fp16, var_31433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3357_cast_fp16")]; + string _SplitHeadsQ__mh_w_3359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3359_equation_0, values = (var_31521_cast_fp16, var_31440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3359_cast_fp16")]; + fp16 var_31762_to_fp16 = const()[name = string("op_31762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3201_cast_fp16, y = var_31762_to_fp16)[name = string("aw_chunk_3201_cast_fp16")]; + fp16 var_31764_to_fp16 = const()[name = string("op_31764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3203_cast_fp16, y = var_31764_to_fp16)[name = string("aw_chunk_3203_cast_fp16")]; + fp16 var_31766_to_fp16 = const()[name = string("op_31766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3205_cast_fp16, y = var_31766_to_fp16)[name = string("aw_chunk_3205_cast_fp16")]; + fp16 var_31768_to_fp16 = const()[name = string("op_31768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3207_cast_fp16, y = var_31768_to_fp16)[name = string("aw_chunk_3207_cast_fp16")]; + fp16 var_31770_to_fp16 = const()[name = string("op_31770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3209_cast_fp16, y = var_31770_to_fp16)[name = string("aw_chunk_3209_cast_fp16")]; + fp16 var_31772_to_fp16 = const()[name = string("op_31772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3211_cast_fp16, y = var_31772_to_fp16)[name = string("aw_chunk_3211_cast_fp16")]; + fp16 var_31774_to_fp16 = const()[name = string("op_31774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3213_cast_fp16, y = var_31774_to_fp16)[name = string("aw_chunk_3213_cast_fp16")]; + fp16 var_31776_to_fp16 = const()[name = string("op_31776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3215_cast_fp16, y = var_31776_to_fp16)[name = string("aw_chunk_3215_cast_fp16")]; + fp16 var_31778_to_fp16 = const()[name = string("op_31778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3217_cast_fp16, y = var_31778_to_fp16)[name = string("aw_chunk_3217_cast_fp16")]; + fp16 var_31780_to_fp16 = const()[name = string("op_31780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3219_cast_fp16, y = var_31780_to_fp16)[name = string("aw_chunk_3219_cast_fp16")]; + fp16 var_31782_to_fp16 = const()[name = string("op_31782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3221_cast_fp16, y = var_31782_to_fp16)[name = string("aw_chunk_3221_cast_fp16")]; + fp16 var_31784_to_fp16 = const()[name = string("op_31784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3223_cast_fp16, y = var_31784_to_fp16)[name = string("aw_chunk_3223_cast_fp16")]; + fp16 var_31786_to_fp16 = const()[name = string("op_31786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3225_cast_fp16, y = var_31786_to_fp16)[name = string("aw_chunk_3225_cast_fp16")]; + fp16 var_31788_to_fp16 = const()[name = string("op_31788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3227_cast_fp16, y = var_31788_to_fp16)[name = string("aw_chunk_3227_cast_fp16")]; + fp16 var_31790_to_fp16 = const()[name = string("op_31790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3229_cast_fp16, y = var_31790_to_fp16)[name = string("aw_chunk_3229_cast_fp16")]; + fp16 var_31792_to_fp16 = const()[name = string("op_31792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3231_cast_fp16, y = var_31792_to_fp16)[name = string("aw_chunk_3231_cast_fp16")]; + fp16 var_31794_to_fp16 = const()[name = string("op_31794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3233_cast_fp16, y = var_31794_to_fp16)[name = string("aw_chunk_3233_cast_fp16")]; + fp16 var_31796_to_fp16 = const()[name = string("op_31796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3235_cast_fp16, y = var_31796_to_fp16)[name = string("aw_chunk_3235_cast_fp16")]; + fp16 var_31798_to_fp16 = const()[name = string("op_31798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3237_cast_fp16, y = var_31798_to_fp16)[name = string("aw_chunk_3237_cast_fp16")]; + fp16 var_31800_to_fp16 = const()[name = string("op_31800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3239_cast_fp16, y = var_31800_to_fp16)[name = string("aw_chunk_3239_cast_fp16")]; + fp16 var_31802_to_fp16 = const()[name = string("op_31802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3241_cast_fp16, y = var_31802_to_fp16)[name = string("aw_chunk_3241_cast_fp16")]; + fp16 var_31804_to_fp16 = const()[name = string("op_31804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3243_cast_fp16, y = var_31804_to_fp16)[name = string("aw_chunk_3243_cast_fp16")]; + fp16 var_31806_to_fp16 = const()[name = string("op_31806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3245_cast_fp16, y = var_31806_to_fp16)[name = string("aw_chunk_3245_cast_fp16")]; + fp16 var_31808_to_fp16 = const()[name = string("op_31808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3247_cast_fp16, y = var_31808_to_fp16)[name = string("aw_chunk_3247_cast_fp16")]; + fp16 var_31810_to_fp16 = const()[name = string("op_31810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3249_cast_fp16, y = var_31810_to_fp16)[name = string("aw_chunk_3249_cast_fp16")]; + fp16 var_31812_to_fp16 = const()[name = string("op_31812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3251_cast_fp16, y = var_31812_to_fp16)[name = string("aw_chunk_3251_cast_fp16")]; + fp16 var_31814_to_fp16 = const()[name = string("op_31814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3253_cast_fp16, y = var_31814_to_fp16)[name = string("aw_chunk_3253_cast_fp16")]; + fp16 var_31816_to_fp16 = const()[name = string("op_31816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3255_cast_fp16, y = var_31816_to_fp16)[name = string("aw_chunk_3255_cast_fp16")]; + fp16 var_31818_to_fp16 = const()[name = string("op_31818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3257_cast_fp16, y = var_31818_to_fp16)[name = string("aw_chunk_3257_cast_fp16")]; + fp16 var_31820_to_fp16 = const()[name = string("op_31820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3259_cast_fp16, y = var_31820_to_fp16)[name = string("aw_chunk_3259_cast_fp16")]; + fp16 var_31822_to_fp16 = const()[name = string("op_31822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3261_cast_fp16, y = var_31822_to_fp16)[name = string("aw_chunk_3261_cast_fp16")]; + fp16 var_31824_to_fp16 = const()[name = string("op_31824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3263_cast_fp16, y = var_31824_to_fp16)[name = string("aw_chunk_3263_cast_fp16")]; + fp16 var_31826_to_fp16 = const()[name = string("op_31826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3265_cast_fp16, y = var_31826_to_fp16)[name = string("aw_chunk_3265_cast_fp16")]; + fp16 var_31828_to_fp16 = const()[name = string("op_31828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3267_cast_fp16, y = var_31828_to_fp16)[name = string("aw_chunk_3267_cast_fp16")]; + fp16 var_31830_to_fp16 = const()[name = string("op_31830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3269_cast_fp16, y = var_31830_to_fp16)[name = string("aw_chunk_3269_cast_fp16")]; + fp16 var_31832_to_fp16 = const()[name = string("op_31832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3271_cast_fp16, y = var_31832_to_fp16)[name = string("aw_chunk_3271_cast_fp16")]; + fp16 var_31834_to_fp16 = const()[name = string("op_31834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3273_cast_fp16, y = var_31834_to_fp16)[name = string("aw_chunk_3273_cast_fp16")]; + fp16 var_31836_to_fp16 = const()[name = string("op_31836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3275_cast_fp16, y = var_31836_to_fp16)[name = string("aw_chunk_3275_cast_fp16")]; + fp16 var_31838_to_fp16 = const()[name = string("op_31838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3277_cast_fp16, y = var_31838_to_fp16)[name = string("aw_chunk_3277_cast_fp16")]; + fp16 var_31840_to_fp16 = const()[name = string("op_31840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3279_cast_fp16, y = var_31840_to_fp16)[name = string("aw_chunk_3279_cast_fp16")]; + fp16 var_31842_to_fp16 = const()[name = string("op_31842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3281_cast_fp16, y = var_31842_to_fp16)[name = string("aw_chunk_3281_cast_fp16")]; + fp16 var_31844_to_fp16 = const()[name = string("op_31844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3283_cast_fp16, y = var_31844_to_fp16)[name = string("aw_chunk_3283_cast_fp16")]; + fp16 var_31846_to_fp16 = const()[name = string("op_31846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3285_cast_fp16, y = var_31846_to_fp16)[name = string("aw_chunk_3285_cast_fp16")]; + fp16 var_31848_to_fp16 = const()[name = string("op_31848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3287_cast_fp16, y = var_31848_to_fp16)[name = string("aw_chunk_3287_cast_fp16")]; + fp16 var_31850_to_fp16 = const()[name = string("op_31850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3289_cast_fp16, y = var_31850_to_fp16)[name = string("aw_chunk_3289_cast_fp16")]; + fp16 var_31852_to_fp16 = const()[name = string("op_31852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3291_cast_fp16, y = var_31852_to_fp16)[name = string("aw_chunk_3291_cast_fp16")]; + fp16 var_31854_to_fp16 = const()[name = string("op_31854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3293_cast_fp16, y = var_31854_to_fp16)[name = string("aw_chunk_3293_cast_fp16")]; + fp16 var_31856_to_fp16 = const()[name = string("op_31856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3295_cast_fp16, y = var_31856_to_fp16)[name = string("aw_chunk_3295_cast_fp16")]; + fp16 var_31858_to_fp16 = const()[name = string("op_31858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3297_cast_fp16, y = var_31858_to_fp16)[name = string("aw_chunk_3297_cast_fp16")]; + fp16 var_31860_to_fp16 = const()[name = string("op_31860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3299_cast_fp16, y = var_31860_to_fp16)[name = string("aw_chunk_3299_cast_fp16")]; + fp16 var_31862_to_fp16 = const()[name = string("op_31862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3301_cast_fp16, y = var_31862_to_fp16)[name = string("aw_chunk_3301_cast_fp16")]; + fp16 var_31864_to_fp16 = const()[name = string("op_31864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3303_cast_fp16, y = var_31864_to_fp16)[name = string("aw_chunk_3303_cast_fp16")]; + fp16 var_31866_to_fp16 = const()[name = string("op_31866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3305_cast_fp16, y = var_31866_to_fp16)[name = string("aw_chunk_3305_cast_fp16")]; + fp16 var_31868_to_fp16 = const()[name = string("op_31868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3307_cast_fp16, y = var_31868_to_fp16)[name = string("aw_chunk_3307_cast_fp16")]; + fp16 var_31870_to_fp16 = const()[name = string("op_31870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3309_cast_fp16, y = var_31870_to_fp16)[name = string("aw_chunk_3309_cast_fp16")]; + fp16 var_31872_to_fp16 = const()[name = string("op_31872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3311_cast_fp16, y = var_31872_to_fp16)[name = string("aw_chunk_3311_cast_fp16")]; + fp16 var_31874_to_fp16 = const()[name = string("op_31874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3313_cast_fp16, y = var_31874_to_fp16)[name = string("aw_chunk_3313_cast_fp16")]; + fp16 var_31876_to_fp16 = const()[name = string("op_31876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3315_cast_fp16, y = var_31876_to_fp16)[name = string("aw_chunk_3315_cast_fp16")]; + fp16 var_31878_to_fp16 = const()[name = string("op_31878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3317_cast_fp16, y = var_31878_to_fp16)[name = string("aw_chunk_3317_cast_fp16")]; + fp16 var_31880_to_fp16 = const()[name = string("op_31880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3319_cast_fp16, y = var_31880_to_fp16)[name = string("aw_chunk_3319_cast_fp16")]; + fp16 var_31882_to_fp16 = const()[name = string("op_31882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3321_cast_fp16, y = var_31882_to_fp16)[name = string("aw_chunk_3321_cast_fp16")]; + fp16 var_31884_to_fp16 = const()[name = string("op_31884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3323_cast_fp16, y = var_31884_to_fp16)[name = string("aw_chunk_3323_cast_fp16")]; + fp16 var_31886_to_fp16 = const()[name = string("op_31886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3325_cast_fp16, y = var_31886_to_fp16)[name = string("aw_chunk_3325_cast_fp16")]; + fp16 var_31888_to_fp16 = const()[name = string("op_31888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3327_cast_fp16, y = var_31888_to_fp16)[name = string("aw_chunk_3327_cast_fp16")]; + fp16 var_31890_to_fp16 = const()[name = string("op_31890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3329_cast_fp16, y = var_31890_to_fp16)[name = string("aw_chunk_3329_cast_fp16")]; + fp16 var_31892_to_fp16 = const()[name = string("op_31892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3331_cast_fp16, y = var_31892_to_fp16)[name = string("aw_chunk_3331_cast_fp16")]; + fp16 var_31894_to_fp16 = const()[name = string("op_31894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3333_cast_fp16, y = var_31894_to_fp16)[name = string("aw_chunk_3333_cast_fp16")]; + fp16 var_31896_to_fp16 = const()[name = string("op_31896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3335_cast_fp16, y = var_31896_to_fp16)[name = string("aw_chunk_3335_cast_fp16")]; + fp16 var_31898_to_fp16 = const()[name = string("op_31898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3337_cast_fp16, y = var_31898_to_fp16)[name = string("aw_chunk_3337_cast_fp16")]; + fp16 var_31900_to_fp16 = const()[name = string("op_31900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3339_cast_fp16, y = var_31900_to_fp16)[name = string("aw_chunk_3339_cast_fp16")]; + fp16 var_31902_to_fp16 = const()[name = string("op_31902_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3341_cast_fp16, y = var_31902_to_fp16)[name = string("aw_chunk_3341_cast_fp16")]; + fp16 var_31904_to_fp16 = const()[name = string("op_31904_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3343_cast_fp16, y = var_31904_to_fp16)[name = string("aw_chunk_3343_cast_fp16")]; + fp16 var_31906_to_fp16 = const()[name = string("op_31906_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3345_cast_fp16, y = var_31906_to_fp16)[name = string("aw_chunk_3345_cast_fp16")]; + fp16 var_31908_to_fp16 = const()[name = string("op_31908_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3347_cast_fp16, y = var_31908_to_fp16)[name = string("aw_chunk_3347_cast_fp16")]; + fp16 var_31910_to_fp16 = const()[name = string("op_31910_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3349_cast_fp16, y = var_31910_to_fp16)[name = string("aw_chunk_3349_cast_fp16")]; + fp16 var_31912_to_fp16 = const()[name = string("op_31912_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3351_cast_fp16, y = var_31912_to_fp16)[name = string("aw_chunk_3351_cast_fp16")]; + fp16 var_31914_to_fp16 = const()[name = string("op_31914_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3353_cast_fp16, y = var_31914_to_fp16)[name = string("aw_chunk_3353_cast_fp16")]; + fp16 var_31916_to_fp16 = const()[name = string("op_31916_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3355_cast_fp16, y = var_31916_to_fp16)[name = string("aw_chunk_3355_cast_fp16")]; + fp16 var_31918_to_fp16 = const()[name = string("op_31918_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3357_cast_fp16, y = var_31918_to_fp16)[name = string("aw_chunk_3357_cast_fp16")]; + fp16 var_31920_to_fp16 = const()[name = string("op_31920_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3359_cast_fp16, y = var_31920_to_fp16)[name = string("aw_chunk_3359_cast_fp16")]; + tensor var_31922_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3201_cast_fp16)[name = string("op_31922_cast_fp16")]; + tensor var_31923_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3203_cast_fp16)[name = string("op_31923_cast_fp16")]; + tensor var_31924_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3205_cast_fp16)[name = string("op_31924_cast_fp16")]; + tensor var_31925_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3207_cast_fp16)[name = string("op_31925_cast_fp16")]; + tensor var_31926_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3209_cast_fp16)[name = string("op_31926_cast_fp16")]; + tensor var_31927_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3211_cast_fp16)[name = string("op_31927_cast_fp16")]; + tensor var_31928_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3213_cast_fp16)[name = string("op_31928_cast_fp16")]; + tensor var_31929_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3215_cast_fp16)[name = string("op_31929_cast_fp16")]; + tensor var_31930_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3217_cast_fp16)[name = string("op_31930_cast_fp16")]; + tensor var_31931_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3219_cast_fp16)[name = string("op_31931_cast_fp16")]; + tensor var_31932_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3221_cast_fp16)[name = string("op_31932_cast_fp16")]; + tensor var_31933_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3223_cast_fp16)[name = string("op_31933_cast_fp16")]; + tensor var_31934_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3225_cast_fp16)[name = string("op_31934_cast_fp16")]; + tensor var_31935_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3227_cast_fp16)[name = string("op_31935_cast_fp16")]; + tensor var_31936_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3229_cast_fp16)[name = string("op_31936_cast_fp16")]; + tensor var_31937_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3231_cast_fp16)[name = string("op_31937_cast_fp16")]; + tensor var_31938_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3233_cast_fp16)[name = string("op_31938_cast_fp16")]; + tensor var_31939_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3235_cast_fp16)[name = string("op_31939_cast_fp16")]; + tensor var_31940_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3237_cast_fp16)[name = string("op_31940_cast_fp16")]; + tensor var_31941_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3239_cast_fp16)[name = string("op_31941_cast_fp16")]; + tensor var_31942_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3241_cast_fp16)[name = string("op_31942_cast_fp16")]; + tensor var_31943_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3243_cast_fp16)[name = string("op_31943_cast_fp16")]; + tensor var_31944_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3245_cast_fp16)[name = string("op_31944_cast_fp16")]; + tensor var_31945_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3247_cast_fp16)[name = string("op_31945_cast_fp16")]; + tensor var_31946_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3249_cast_fp16)[name = string("op_31946_cast_fp16")]; + tensor var_31947_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3251_cast_fp16)[name = string("op_31947_cast_fp16")]; + tensor var_31948_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3253_cast_fp16)[name = string("op_31948_cast_fp16")]; + tensor var_31949_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3255_cast_fp16)[name = string("op_31949_cast_fp16")]; + tensor var_31950_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3257_cast_fp16)[name = string("op_31950_cast_fp16")]; + tensor var_31951_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3259_cast_fp16)[name = string("op_31951_cast_fp16")]; + tensor var_31952_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3261_cast_fp16)[name = string("op_31952_cast_fp16")]; + tensor var_31953_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3263_cast_fp16)[name = string("op_31953_cast_fp16")]; + tensor var_31954_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3265_cast_fp16)[name = string("op_31954_cast_fp16")]; + tensor var_31955_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3267_cast_fp16)[name = string("op_31955_cast_fp16")]; + tensor var_31956_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3269_cast_fp16)[name = string("op_31956_cast_fp16")]; + tensor var_31957_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3271_cast_fp16)[name = string("op_31957_cast_fp16")]; + tensor var_31958_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3273_cast_fp16)[name = string("op_31958_cast_fp16")]; + tensor var_31959_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3275_cast_fp16)[name = string("op_31959_cast_fp16")]; + tensor var_31960_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3277_cast_fp16)[name = string("op_31960_cast_fp16")]; + tensor var_31961_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3279_cast_fp16)[name = string("op_31961_cast_fp16")]; + tensor var_31962_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3281_cast_fp16)[name = string("op_31962_cast_fp16")]; + tensor var_31963_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3283_cast_fp16)[name = string("op_31963_cast_fp16")]; + tensor var_31964_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3285_cast_fp16)[name = string("op_31964_cast_fp16")]; + tensor var_31965_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3287_cast_fp16)[name = string("op_31965_cast_fp16")]; + tensor var_31966_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3289_cast_fp16)[name = string("op_31966_cast_fp16")]; + tensor var_31967_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3291_cast_fp16)[name = string("op_31967_cast_fp16")]; + tensor var_31968_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3293_cast_fp16)[name = string("op_31968_cast_fp16")]; + tensor var_31969_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3295_cast_fp16)[name = string("op_31969_cast_fp16")]; + tensor var_31970_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3297_cast_fp16)[name = string("op_31970_cast_fp16")]; + tensor var_31971_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3299_cast_fp16)[name = string("op_31971_cast_fp16")]; + tensor var_31972_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3301_cast_fp16)[name = string("op_31972_cast_fp16")]; + tensor var_31973_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3303_cast_fp16)[name = string("op_31973_cast_fp16")]; + tensor var_31974_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3305_cast_fp16)[name = string("op_31974_cast_fp16")]; + tensor var_31975_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3307_cast_fp16)[name = string("op_31975_cast_fp16")]; + tensor var_31976_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3309_cast_fp16)[name = string("op_31976_cast_fp16")]; + tensor var_31977_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3311_cast_fp16)[name = string("op_31977_cast_fp16")]; + tensor var_31978_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3313_cast_fp16)[name = string("op_31978_cast_fp16")]; + tensor var_31979_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3315_cast_fp16)[name = string("op_31979_cast_fp16")]; + tensor var_31980_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3317_cast_fp16)[name = string("op_31980_cast_fp16")]; + tensor var_31981_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3319_cast_fp16)[name = string("op_31981_cast_fp16")]; + tensor var_31982_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3321_cast_fp16)[name = string("op_31982_cast_fp16")]; + tensor var_31983_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3323_cast_fp16)[name = string("op_31983_cast_fp16")]; + tensor var_31984_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3325_cast_fp16)[name = string("op_31984_cast_fp16")]; + tensor var_31985_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3327_cast_fp16)[name = string("op_31985_cast_fp16")]; + tensor var_31986_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3329_cast_fp16)[name = string("op_31986_cast_fp16")]; + tensor var_31987_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3331_cast_fp16)[name = string("op_31987_cast_fp16")]; + tensor var_31988_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3333_cast_fp16)[name = string("op_31988_cast_fp16")]; + tensor var_31989_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3335_cast_fp16)[name = string("op_31989_cast_fp16")]; + tensor var_31990_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3337_cast_fp16)[name = string("op_31990_cast_fp16")]; + tensor var_31991_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3339_cast_fp16)[name = string("op_31991_cast_fp16")]; + tensor var_31992_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3341_cast_fp16)[name = string("op_31992_cast_fp16")]; + tensor var_31993_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3343_cast_fp16)[name = string("op_31993_cast_fp16")]; + tensor var_31994_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3345_cast_fp16)[name = string("op_31994_cast_fp16")]; + tensor var_31995_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3347_cast_fp16)[name = string("op_31995_cast_fp16")]; + tensor var_31996_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3349_cast_fp16)[name = string("op_31996_cast_fp16")]; + tensor var_31997_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3351_cast_fp16)[name = string("op_31997_cast_fp16")]; + tensor var_31998_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3353_cast_fp16)[name = string("op_31998_cast_fp16")]; + tensor var_31999_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3355_cast_fp16)[name = string("op_31999_cast_fp16")]; + tensor var_32000_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3357_cast_fp16)[name = string("op_32000_cast_fp16")]; + tensor var_32001_cast_fp16 = softmax(axis = var_30747, x = aw_chunk_3359_cast_fp16)[name = string("op_32001_cast_fp16")]; + string var_32003_equation_0 = const()[name = string("op_32003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32003_cast_fp16 = einsum(equation = var_32003_equation_0, values = (var_31523_cast_fp16, var_31922_cast_fp16))[name = string("op_32003_cast_fp16")]; + string var_32005_equation_0 = const()[name = string("op_32005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32005_cast_fp16 = einsum(equation = var_32005_equation_0, values = (var_31523_cast_fp16, var_31923_cast_fp16))[name = string("op_32005_cast_fp16")]; + string var_32007_equation_0 = const()[name = string("op_32007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32007_cast_fp16 = einsum(equation = var_32007_equation_0, values = (var_31523_cast_fp16, var_31924_cast_fp16))[name = string("op_32007_cast_fp16")]; + string var_32009_equation_0 = const()[name = string("op_32009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32009_cast_fp16 = einsum(equation = var_32009_equation_0, values = (var_31523_cast_fp16, var_31925_cast_fp16))[name = string("op_32009_cast_fp16")]; + string var_32011_equation_0 = const()[name = string("op_32011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32011_cast_fp16 = einsum(equation = var_32011_equation_0, values = (var_31527_cast_fp16, var_31926_cast_fp16))[name = string("op_32011_cast_fp16")]; + string var_32013_equation_0 = const()[name = string("op_32013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32013_cast_fp16 = einsum(equation = var_32013_equation_0, values = (var_31527_cast_fp16, var_31927_cast_fp16))[name = string("op_32013_cast_fp16")]; + string var_32015_equation_0 = const()[name = string("op_32015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32015_cast_fp16 = einsum(equation = var_32015_equation_0, values = (var_31527_cast_fp16, var_31928_cast_fp16))[name = string("op_32015_cast_fp16")]; + string var_32017_equation_0 = const()[name = string("op_32017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32017_cast_fp16 = einsum(equation = var_32017_equation_0, values = (var_31527_cast_fp16, var_31929_cast_fp16))[name = string("op_32017_cast_fp16")]; + string var_32019_equation_0 = const()[name = string("op_32019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32019_cast_fp16 = einsum(equation = var_32019_equation_0, values = (var_31531_cast_fp16, var_31930_cast_fp16))[name = string("op_32019_cast_fp16")]; + string var_32021_equation_0 = const()[name = string("op_32021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32021_cast_fp16 = einsum(equation = var_32021_equation_0, values = (var_31531_cast_fp16, var_31931_cast_fp16))[name = string("op_32021_cast_fp16")]; + string var_32023_equation_0 = const()[name = string("op_32023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32023_cast_fp16 = einsum(equation = var_32023_equation_0, values = (var_31531_cast_fp16, var_31932_cast_fp16))[name = string("op_32023_cast_fp16")]; + string var_32025_equation_0 = const()[name = string("op_32025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32025_cast_fp16 = einsum(equation = var_32025_equation_0, values = (var_31531_cast_fp16, var_31933_cast_fp16))[name = string("op_32025_cast_fp16")]; + string var_32027_equation_0 = const()[name = string("op_32027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32027_cast_fp16 = einsum(equation = var_32027_equation_0, values = (var_31535_cast_fp16, var_31934_cast_fp16))[name = string("op_32027_cast_fp16")]; + string var_32029_equation_0 = const()[name = string("op_32029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32029_cast_fp16 = einsum(equation = var_32029_equation_0, values = (var_31535_cast_fp16, var_31935_cast_fp16))[name = string("op_32029_cast_fp16")]; + string var_32031_equation_0 = const()[name = string("op_32031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32031_cast_fp16 = einsum(equation = var_32031_equation_0, values = (var_31535_cast_fp16, var_31936_cast_fp16))[name = string("op_32031_cast_fp16")]; + string var_32033_equation_0 = const()[name = string("op_32033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32033_cast_fp16 = einsum(equation = var_32033_equation_0, values = (var_31535_cast_fp16, var_31937_cast_fp16))[name = string("op_32033_cast_fp16")]; + string var_32035_equation_0 = const()[name = string("op_32035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32035_cast_fp16 = einsum(equation = var_32035_equation_0, values = (var_31539_cast_fp16, var_31938_cast_fp16))[name = string("op_32035_cast_fp16")]; + string var_32037_equation_0 = const()[name = string("op_32037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32037_cast_fp16 = einsum(equation = var_32037_equation_0, values = (var_31539_cast_fp16, var_31939_cast_fp16))[name = string("op_32037_cast_fp16")]; + string var_32039_equation_0 = const()[name = string("op_32039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32039_cast_fp16 = einsum(equation = var_32039_equation_0, values = (var_31539_cast_fp16, var_31940_cast_fp16))[name = string("op_32039_cast_fp16")]; + string var_32041_equation_0 = const()[name = string("op_32041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32041_cast_fp16 = einsum(equation = var_32041_equation_0, values = (var_31539_cast_fp16, var_31941_cast_fp16))[name = string("op_32041_cast_fp16")]; + string var_32043_equation_0 = const()[name = string("op_32043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32043_cast_fp16 = einsum(equation = var_32043_equation_0, values = (var_31543_cast_fp16, var_31942_cast_fp16))[name = string("op_32043_cast_fp16")]; + string var_32045_equation_0 = const()[name = string("op_32045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32045_cast_fp16 = einsum(equation = var_32045_equation_0, values = (var_31543_cast_fp16, var_31943_cast_fp16))[name = string("op_32045_cast_fp16")]; + string var_32047_equation_0 = const()[name = string("op_32047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32047_cast_fp16 = einsum(equation = var_32047_equation_0, values = (var_31543_cast_fp16, var_31944_cast_fp16))[name = string("op_32047_cast_fp16")]; + string var_32049_equation_0 = const()[name = string("op_32049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32049_cast_fp16 = einsum(equation = var_32049_equation_0, values = (var_31543_cast_fp16, var_31945_cast_fp16))[name = string("op_32049_cast_fp16")]; + string var_32051_equation_0 = const()[name = string("op_32051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32051_cast_fp16 = einsum(equation = var_32051_equation_0, values = (var_31547_cast_fp16, var_31946_cast_fp16))[name = string("op_32051_cast_fp16")]; + string var_32053_equation_0 = const()[name = string("op_32053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32053_cast_fp16 = einsum(equation = var_32053_equation_0, values = (var_31547_cast_fp16, var_31947_cast_fp16))[name = string("op_32053_cast_fp16")]; + string var_32055_equation_0 = const()[name = string("op_32055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32055_cast_fp16 = einsum(equation = var_32055_equation_0, values = (var_31547_cast_fp16, var_31948_cast_fp16))[name = string("op_32055_cast_fp16")]; + string var_32057_equation_0 = const()[name = string("op_32057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32057_cast_fp16 = einsum(equation = var_32057_equation_0, values = (var_31547_cast_fp16, var_31949_cast_fp16))[name = string("op_32057_cast_fp16")]; + string var_32059_equation_0 = const()[name = string("op_32059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32059_cast_fp16 = einsum(equation = var_32059_equation_0, values = (var_31551_cast_fp16, var_31950_cast_fp16))[name = string("op_32059_cast_fp16")]; + string var_32061_equation_0 = const()[name = string("op_32061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32061_cast_fp16 = einsum(equation = var_32061_equation_0, values = (var_31551_cast_fp16, var_31951_cast_fp16))[name = string("op_32061_cast_fp16")]; + string var_32063_equation_0 = const()[name = string("op_32063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32063_cast_fp16 = einsum(equation = var_32063_equation_0, values = (var_31551_cast_fp16, var_31952_cast_fp16))[name = string("op_32063_cast_fp16")]; + string var_32065_equation_0 = const()[name = string("op_32065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32065_cast_fp16 = einsum(equation = var_32065_equation_0, values = (var_31551_cast_fp16, var_31953_cast_fp16))[name = string("op_32065_cast_fp16")]; + string var_32067_equation_0 = const()[name = string("op_32067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32067_cast_fp16 = einsum(equation = var_32067_equation_0, values = (var_31555_cast_fp16, var_31954_cast_fp16))[name = string("op_32067_cast_fp16")]; + string var_32069_equation_0 = const()[name = string("op_32069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32069_cast_fp16 = einsum(equation = var_32069_equation_0, values = (var_31555_cast_fp16, var_31955_cast_fp16))[name = string("op_32069_cast_fp16")]; + string var_32071_equation_0 = const()[name = string("op_32071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32071_cast_fp16 = einsum(equation = var_32071_equation_0, values = (var_31555_cast_fp16, var_31956_cast_fp16))[name = string("op_32071_cast_fp16")]; + string var_32073_equation_0 = const()[name = string("op_32073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32073_cast_fp16 = einsum(equation = var_32073_equation_0, values = (var_31555_cast_fp16, var_31957_cast_fp16))[name = string("op_32073_cast_fp16")]; + string var_32075_equation_0 = const()[name = string("op_32075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32075_cast_fp16 = einsum(equation = var_32075_equation_0, values = (var_31559_cast_fp16, var_31958_cast_fp16))[name = string("op_32075_cast_fp16")]; + string var_32077_equation_0 = const()[name = string("op_32077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32077_cast_fp16 = einsum(equation = var_32077_equation_0, values = (var_31559_cast_fp16, var_31959_cast_fp16))[name = string("op_32077_cast_fp16")]; + string var_32079_equation_0 = const()[name = string("op_32079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32079_cast_fp16 = einsum(equation = var_32079_equation_0, values = (var_31559_cast_fp16, var_31960_cast_fp16))[name = string("op_32079_cast_fp16")]; + string var_32081_equation_0 = const()[name = string("op_32081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32081_cast_fp16 = einsum(equation = var_32081_equation_0, values = (var_31559_cast_fp16, var_31961_cast_fp16))[name = string("op_32081_cast_fp16")]; + string var_32083_equation_0 = const()[name = string("op_32083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32083_cast_fp16 = einsum(equation = var_32083_equation_0, values = (var_31563_cast_fp16, var_31962_cast_fp16))[name = string("op_32083_cast_fp16")]; + string var_32085_equation_0 = const()[name = string("op_32085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32085_cast_fp16 = einsum(equation = var_32085_equation_0, values = (var_31563_cast_fp16, var_31963_cast_fp16))[name = string("op_32085_cast_fp16")]; + string var_32087_equation_0 = const()[name = string("op_32087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32087_cast_fp16 = einsum(equation = var_32087_equation_0, values = (var_31563_cast_fp16, var_31964_cast_fp16))[name = string("op_32087_cast_fp16")]; + string var_32089_equation_0 = const()[name = string("op_32089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32089_cast_fp16 = einsum(equation = var_32089_equation_0, values = (var_31563_cast_fp16, var_31965_cast_fp16))[name = string("op_32089_cast_fp16")]; + string var_32091_equation_0 = const()[name = string("op_32091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32091_cast_fp16 = einsum(equation = var_32091_equation_0, values = (var_31567_cast_fp16, var_31966_cast_fp16))[name = string("op_32091_cast_fp16")]; + string var_32093_equation_0 = const()[name = string("op_32093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32093_cast_fp16 = einsum(equation = var_32093_equation_0, values = (var_31567_cast_fp16, var_31967_cast_fp16))[name = string("op_32093_cast_fp16")]; + string var_32095_equation_0 = const()[name = string("op_32095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32095_cast_fp16 = einsum(equation = var_32095_equation_0, values = (var_31567_cast_fp16, var_31968_cast_fp16))[name = string("op_32095_cast_fp16")]; + string var_32097_equation_0 = const()[name = string("op_32097_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32097_cast_fp16 = einsum(equation = var_32097_equation_0, values = (var_31567_cast_fp16, var_31969_cast_fp16))[name = string("op_32097_cast_fp16")]; + string var_32099_equation_0 = const()[name = string("op_32099_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32099_cast_fp16 = einsum(equation = var_32099_equation_0, values = (var_31571_cast_fp16, var_31970_cast_fp16))[name = string("op_32099_cast_fp16")]; + string var_32101_equation_0 = const()[name = string("op_32101_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32101_cast_fp16 = einsum(equation = var_32101_equation_0, values = (var_31571_cast_fp16, var_31971_cast_fp16))[name = string("op_32101_cast_fp16")]; + string var_32103_equation_0 = const()[name = string("op_32103_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32103_cast_fp16 = einsum(equation = var_32103_equation_0, values = (var_31571_cast_fp16, var_31972_cast_fp16))[name = string("op_32103_cast_fp16")]; + string var_32105_equation_0 = const()[name = string("op_32105_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32105_cast_fp16 = einsum(equation = var_32105_equation_0, values = (var_31571_cast_fp16, var_31973_cast_fp16))[name = string("op_32105_cast_fp16")]; + string var_32107_equation_0 = const()[name = string("op_32107_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32107_cast_fp16 = einsum(equation = var_32107_equation_0, values = (var_31575_cast_fp16, var_31974_cast_fp16))[name = string("op_32107_cast_fp16")]; + string var_32109_equation_0 = const()[name = string("op_32109_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32109_cast_fp16 = einsum(equation = var_32109_equation_0, values = (var_31575_cast_fp16, var_31975_cast_fp16))[name = string("op_32109_cast_fp16")]; + string var_32111_equation_0 = const()[name = string("op_32111_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32111_cast_fp16 = einsum(equation = var_32111_equation_0, values = (var_31575_cast_fp16, var_31976_cast_fp16))[name = string("op_32111_cast_fp16")]; + string var_32113_equation_0 = const()[name = string("op_32113_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32113_cast_fp16 = einsum(equation = var_32113_equation_0, values = (var_31575_cast_fp16, var_31977_cast_fp16))[name = string("op_32113_cast_fp16")]; + string var_32115_equation_0 = const()[name = string("op_32115_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32115_cast_fp16 = einsum(equation = var_32115_equation_0, values = (var_31579_cast_fp16, var_31978_cast_fp16))[name = string("op_32115_cast_fp16")]; + string var_32117_equation_0 = const()[name = string("op_32117_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32117_cast_fp16 = einsum(equation = var_32117_equation_0, values = (var_31579_cast_fp16, var_31979_cast_fp16))[name = string("op_32117_cast_fp16")]; + string var_32119_equation_0 = const()[name = string("op_32119_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32119_cast_fp16 = einsum(equation = var_32119_equation_0, values = (var_31579_cast_fp16, var_31980_cast_fp16))[name = string("op_32119_cast_fp16")]; + string var_32121_equation_0 = const()[name = string("op_32121_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32121_cast_fp16 = einsum(equation = var_32121_equation_0, values = (var_31579_cast_fp16, var_31981_cast_fp16))[name = string("op_32121_cast_fp16")]; + string var_32123_equation_0 = const()[name = string("op_32123_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32123_cast_fp16 = einsum(equation = var_32123_equation_0, values = (var_31583_cast_fp16, var_31982_cast_fp16))[name = string("op_32123_cast_fp16")]; + string var_32125_equation_0 = const()[name = string("op_32125_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32125_cast_fp16 = einsum(equation = var_32125_equation_0, values = (var_31583_cast_fp16, var_31983_cast_fp16))[name = string("op_32125_cast_fp16")]; + string var_32127_equation_0 = const()[name = string("op_32127_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32127_cast_fp16 = einsum(equation = var_32127_equation_0, values = (var_31583_cast_fp16, var_31984_cast_fp16))[name = string("op_32127_cast_fp16")]; + string var_32129_equation_0 = const()[name = string("op_32129_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32129_cast_fp16 = einsum(equation = var_32129_equation_0, values = (var_31583_cast_fp16, var_31985_cast_fp16))[name = string("op_32129_cast_fp16")]; + string var_32131_equation_0 = const()[name = string("op_32131_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32131_cast_fp16 = einsum(equation = var_32131_equation_0, values = (var_31587_cast_fp16, var_31986_cast_fp16))[name = string("op_32131_cast_fp16")]; + string var_32133_equation_0 = const()[name = string("op_32133_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32133_cast_fp16 = einsum(equation = var_32133_equation_0, values = (var_31587_cast_fp16, var_31987_cast_fp16))[name = string("op_32133_cast_fp16")]; + string var_32135_equation_0 = const()[name = string("op_32135_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32135_cast_fp16 = einsum(equation = var_32135_equation_0, values = (var_31587_cast_fp16, var_31988_cast_fp16))[name = string("op_32135_cast_fp16")]; + string var_32137_equation_0 = const()[name = string("op_32137_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32137_cast_fp16 = einsum(equation = var_32137_equation_0, values = (var_31587_cast_fp16, var_31989_cast_fp16))[name = string("op_32137_cast_fp16")]; + string var_32139_equation_0 = const()[name = string("op_32139_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32139_cast_fp16 = einsum(equation = var_32139_equation_0, values = (var_31591_cast_fp16, var_31990_cast_fp16))[name = string("op_32139_cast_fp16")]; + string var_32141_equation_0 = const()[name = string("op_32141_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32141_cast_fp16 = einsum(equation = var_32141_equation_0, values = (var_31591_cast_fp16, var_31991_cast_fp16))[name = string("op_32141_cast_fp16")]; + string var_32143_equation_0 = const()[name = string("op_32143_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32143_cast_fp16 = einsum(equation = var_32143_equation_0, values = (var_31591_cast_fp16, var_31992_cast_fp16))[name = string("op_32143_cast_fp16")]; + string var_32145_equation_0 = const()[name = string("op_32145_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32145_cast_fp16 = einsum(equation = var_32145_equation_0, values = (var_31591_cast_fp16, var_31993_cast_fp16))[name = string("op_32145_cast_fp16")]; + string var_32147_equation_0 = const()[name = string("op_32147_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32147_cast_fp16 = einsum(equation = var_32147_equation_0, values = (var_31595_cast_fp16, var_31994_cast_fp16))[name = string("op_32147_cast_fp16")]; + string var_32149_equation_0 = const()[name = string("op_32149_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32149_cast_fp16 = einsum(equation = var_32149_equation_0, values = (var_31595_cast_fp16, var_31995_cast_fp16))[name = string("op_32149_cast_fp16")]; + string var_32151_equation_0 = const()[name = string("op_32151_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32151_cast_fp16 = einsum(equation = var_32151_equation_0, values = (var_31595_cast_fp16, var_31996_cast_fp16))[name = string("op_32151_cast_fp16")]; + string var_32153_equation_0 = const()[name = string("op_32153_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32153_cast_fp16 = einsum(equation = var_32153_equation_0, values = (var_31595_cast_fp16, var_31997_cast_fp16))[name = string("op_32153_cast_fp16")]; + string var_32155_equation_0 = const()[name = string("op_32155_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32155_cast_fp16 = einsum(equation = var_32155_equation_0, values = (var_31599_cast_fp16, var_31998_cast_fp16))[name = string("op_32155_cast_fp16")]; + string var_32157_equation_0 = const()[name = string("op_32157_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32157_cast_fp16 = einsum(equation = var_32157_equation_0, values = (var_31599_cast_fp16, var_31999_cast_fp16))[name = string("op_32157_cast_fp16")]; + string var_32159_equation_0 = const()[name = string("op_32159_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32159_cast_fp16 = einsum(equation = var_32159_equation_0, values = (var_31599_cast_fp16, var_32000_cast_fp16))[name = string("op_32159_cast_fp16")]; + string var_32161_equation_0 = const()[name = string("op_32161_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_32161_cast_fp16 = einsum(equation = var_32161_equation_0, values = (var_31599_cast_fp16, var_32001_cast_fp16))[name = string("op_32161_cast_fp16")]; + bool var_32163_interleave_0 = const()[name = string("op_32163_interleave_0"), val = bool(false)]; + tensor var_32163_cast_fp16 = concat(axis = var_30722, interleave = var_32163_interleave_0, values = (var_32003_cast_fp16, var_32005_cast_fp16, var_32007_cast_fp16, var_32009_cast_fp16))[name = string("op_32163_cast_fp16")]; + bool var_32165_interleave_0 = const()[name = string("op_32165_interleave_0"), val = bool(false)]; + tensor var_32165_cast_fp16 = concat(axis = var_30722, interleave = var_32165_interleave_0, values = (var_32011_cast_fp16, var_32013_cast_fp16, var_32015_cast_fp16, var_32017_cast_fp16))[name = string("op_32165_cast_fp16")]; + bool var_32167_interleave_0 = const()[name = string("op_32167_interleave_0"), val = bool(false)]; + tensor var_32167_cast_fp16 = concat(axis = var_30722, interleave = var_32167_interleave_0, values = (var_32019_cast_fp16, var_32021_cast_fp16, var_32023_cast_fp16, var_32025_cast_fp16))[name = string("op_32167_cast_fp16")]; + bool var_32169_interleave_0 = const()[name = string("op_32169_interleave_0"), val = bool(false)]; + tensor var_32169_cast_fp16 = concat(axis = var_30722, interleave = var_32169_interleave_0, values = (var_32027_cast_fp16, var_32029_cast_fp16, var_32031_cast_fp16, var_32033_cast_fp16))[name = string("op_32169_cast_fp16")]; + bool var_32171_interleave_0 = const()[name = string("op_32171_interleave_0"), val = bool(false)]; + tensor var_32171_cast_fp16 = concat(axis = var_30722, interleave = var_32171_interleave_0, values = (var_32035_cast_fp16, var_32037_cast_fp16, var_32039_cast_fp16, var_32041_cast_fp16))[name = string("op_32171_cast_fp16")]; + bool var_32173_interleave_0 = const()[name = string("op_32173_interleave_0"), val = bool(false)]; + tensor var_32173_cast_fp16 = concat(axis = var_30722, interleave = var_32173_interleave_0, values = (var_32043_cast_fp16, var_32045_cast_fp16, var_32047_cast_fp16, var_32049_cast_fp16))[name = string("op_32173_cast_fp16")]; + bool var_32175_interleave_0 = const()[name = string("op_32175_interleave_0"), val = bool(false)]; + tensor var_32175_cast_fp16 = concat(axis = var_30722, interleave = var_32175_interleave_0, values = (var_32051_cast_fp16, var_32053_cast_fp16, var_32055_cast_fp16, var_32057_cast_fp16))[name = string("op_32175_cast_fp16")]; + bool var_32177_interleave_0 = const()[name = string("op_32177_interleave_0"), val = bool(false)]; + tensor var_32177_cast_fp16 = concat(axis = var_30722, interleave = var_32177_interleave_0, values = (var_32059_cast_fp16, var_32061_cast_fp16, var_32063_cast_fp16, var_32065_cast_fp16))[name = string("op_32177_cast_fp16")]; + bool var_32179_interleave_0 = const()[name = string("op_32179_interleave_0"), val = bool(false)]; + tensor var_32179_cast_fp16 = concat(axis = var_30722, interleave = var_32179_interleave_0, values = (var_32067_cast_fp16, var_32069_cast_fp16, var_32071_cast_fp16, var_32073_cast_fp16))[name = string("op_32179_cast_fp16")]; + bool var_32181_interleave_0 = const()[name = string("op_32181_interleave_0"), val = bool(false)]; + tensor var_32181_cast_fp16 = concat(axis = var_30722, interleave = var_32181_interleave_0, values = (var_32075_cast_fp16, var_32077_cast_fp16, var_32079_cast_fp16, var_32081_cast_fp16))[name = string("op_32181_cast_fp16")]; + bool var_32183_interleave_0 = const()[name = string("op_32183_interleave_0"), val = bool(false)]; + tensor var_32183_cast_fp16 = concat(axis = var_30722, interleave = var_32183_interleave_0, values = (var_32083_cast_fp16, var_32085_cast_fp16, var_32087_cast_fp16, var_32089_cast_fp16))[name = string("op_32183_cast_fp16")]; + bool var_32185_interleave_0 = const()[name = string("op_32185_interleave_0"), val = bool(false)]; + tensor var_32185_cast_fp16 = concat(axis = var_30722, interleave = var_32185_interleave_0, values = (var_32091_cast_fp16, var_32093_cast_fp16, var_32095_cast_fp16, var_32097_cast_fp16))[name = string("op_32185_cast_fp16")]; + bool var_32187_interleave_0 = const()[name = string("op_32187_interleave_0"), val = bool(false)]; + tensor var_32187_cast_fp16 = concat(axis = var_30722, interleave = var_32187_interleave_0, values = (var_32099_cast_fp16, var_32101_cast_fp16, var_32103_cast_fp16, var_32105_cast_fp16))[name = string("op_32187_cast_fp16")]; + bool var_32189_interleave_0 = const()[name = string("op_32189_interleave_0"), val = bool(false)]; + tensor var_32189_cast_fp16 = concat(axis = var_30722, interleave = var_32189_interleave_0, values = (var_32107_cast_fp16, var_32109_cast_fp16, var_32111_cast_fp16, var_32113_cast_fp16))[name = string("op_32189_cast_fp16")]; + bool var_32191_interleave_0 = const()[name = string("op_32191_interleave_0"), val = bool(false)]; + tensor var_32191_cast_fp16 = concat(axis = var_30722, interleave = var_32191_interleave_0, values = (var_32115_cast_fp16, var_32117_cast_fp16, var_32119_cast_fp16, var_32121_cast_fp16))[name = string("op_32191_cast_fp16")]; + bool var_32193_interleave_0 = const()[name = string("op_32193_interleave_0"), val = bool(false)]; + tensor var_32193_cast_fp16 = concat(axis = var_30722, interleave = var_32193_interleave_0, values = (var_32123_cast_fp16, var_32125_cast_fp16, var_32127_cast_fp16, var_32129_cast_fp16))[name = string("op_32193_cast_fp16")]; + bool var_32195_interleave_0 = const()[name = string("op_32195_interleave_0"), val = bool(false)]; + tensor var_32195_cast_fp16 = concat(axis = var_30722, interleave = var_32195_interleave_0, values = (var_32131_cast_fp16, var_32133_cast_fp16, var_32135_cast_fp16, var_32137_cast_fp16))[name = string("op_32195_cast_fp16")]; + bool var_32197_interleave_0 = const()[name = string("op_32197_interleave_0"), val = bool(false)]; + tensor var_32197_cast_fp16 = concat(axis = var_30722, interleave = var_32197_interleave_0, values = (var_32139_cast_fp16, var_32141_cast_fp16, var_32143_cast_fp16, var_32145_cast_fp16))[name = string("op_32197_cast_fp16")]; + bool var_32199_interleave_0 = const()[name = string("op_32199_interleave_0"), val = bool(false)]; + tensor var_32199_cast_fp16 = concat(axis = var_30722, interleave = var_32199_interleave_0, values = (var_32147_cast_fp16, var_32149_cast_fp16, var_32151_cast_fp16, var_32153_cast_fp16))[name = string("op_32199_cast_fp16")]; + bool var_32201_interleave_0 = const()[name = string("op_32201_interleave_0"), val = bool(false)]; + tensor var_32201_cast_fp16 = concat(axis = var_30722, interleave = var_32201_interleave_0, values = (var_32155_cast_fp16, var_32157_cast_fp16, var_32159_cast_fp16, var_32161_cast_fp16))[name = string("op_32201_cast_fp16")]; + bool input_161_interleave_0 = const()[name = string("input_161_interleave_0"), val = bool(false)]; + tensor input_161_cast_fp16 = concat(axis = var_30747, interleave = input_161_interleave_0, values = (var_32163_cast_fp16, var_32165_cast_fp16, var_32167_cast_fp16, var_32169_cast_fp16, var_32171_cast_fp16, var_32173_cast_fp16, var_32175_cast_fp16, var_32177_cast_fp16, var_32179_cast_fp16, var_32181_cast_fp16, var_32183_cast_fp16, var_32185_cast_fp16, var_32187_cast_fp16, var_32189_cast_fp16, var_32191_cast_fp16, var_32193_cast_fp16, var_32195_cast_fp16, var_32197_cast_fp16, var_32199_cast_fp16, var_32201_cast_fp16))[name = string("input_161_cast_fp16")]; + string obj_83_pad_type_0 = const()[name = string("obj_83_pad_type_0"), val = string("valid")]; + tensor obj_83_strides_0 = const()[name = string("obj_83_strides_0"), val = tensor([1, 1])]; + tensor obj_83_pad_0 = const()[name = string("obj_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_83_dilations_0 = const()[name = string("obj_83_dilations_0"), val = tensor([1, 1])]; + int32 obj_83_groups_0 = const()[name = string("obj_83_groups_0"), val = int32(1)]; + tensor layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811570880)))]; + tensor layers_20_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814847744)))]; + tensor obj_83_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_bias_to_fp16, dilations = obj_83_dilations_0, groups = obj_83_groups_0, pad = obj_83_pad_0, pad_type = obj_83_pad_type_0, strides = obj_83_strides_0, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_161_cast_fp16)[name = string("obj_83_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_83_cast_fp16)[name = string("inputs_83_cast_fp16")]; + tensor out_83_axes_0 = const()[name = string("out_83_axes_0"), val = tensor([1])]; + fp16 var_32220_to_fp16 = const()[name = string("op_32220_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_32220_to_fp16, x = inputs_83_cast_fp16)[name = string("out_83_cast_fp16")]; + tensor input_163_gamma_0_to_fp16 = const()[name = string("input_163_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814850368)))]; + tensor input_163_beta_0_to_fp16 = const()[name = string("input_163_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814852992)))]; + fp16 input_163_epsilon_0_to_fp16 = const()[name = string("input_163_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_163_cast_fp16 = batch_norm(beta = input_163_beta_0_to_fp16, epsilon = input_163_epsilon_0_to_fp16, gamma = input_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = string("input_163_cast_fp16")]; + string input_165_pad_type_0 = const()[name = string("input_165_pad_type_0"), val = string("valid")]; + tensor input_165_strides_0 = const()[name = string("input_165_strides_0"), val = tensor([1, 1])]; + tensor input_165_pad_0 = const()[name = string("input_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_165_dilations_0 = const()[name = string("input_165_dilations_0"), val = tensor([1, 1])]; + int32 input_165_groups_0 = const()[name = string("input_165_groups_0"), val = int32(1)]; + tensor layers_20_fc1_weight_to_fp16 = const()[name = string("layers_20_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(814855616)))]; + tensor layers_20_fc1_bias_to_fp16 = const()[name = string("layers_20_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827962880)))]; + tensor input_165_cast_fp16 = conv(bias = layers_20_fc1_bias_to_fp16, dilations = input_165_dilations_0, groups = input_165_groups_0, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = input_165_strides_0, weight = layers_20_fc1_weight_to_fp16, x = input_163_cast_fp16)[name = string("input_165_cast_fp16")]; + string input_167_mode_0 = const()[name = string("input_167_mode_0"), val = string("EXACT")]; + tensor input_167_cast_fp16 = gelu(mode = input_167_mode_0, x = input_165_cast_fp16)[name = string("input_167_cast_fp16")]; + string hidden_states_45_pad_type_0 = const()[name = string("hidden_states_45_pad_type_0"), val = string("valid")]; + tensor hidden_states_45_strides_0 = const()[name = string("hidden_states_45_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_45_pad_0 = const()[name = string("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_45_dilations_0 = const()[name = string("hidden_states_45_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_45_groups_0 = const()[name = string("hidden_states_45_groups_0"), val = int32(1)]; + tensor layers_20_fc2_weight_to_fp16 = const()[name = string("layers_20_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(827973184)))]; + tensor layers_20_fc2_bias_to_fp16 = const()[name = string("layers_20_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841080448)))]; + tensor hidden_states_45_cast_fp16 = conv(bias = layers_20_fc2_bias_to_fp16, dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = layers_20_fc2_weight_to_fp16, x = input_167_cast_fp16)[name = string("hidden_states_45_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_45_cast_fp16)[name = string("inputs_85_cast_fp16")]; + int32 var_32249 = const()[name = string("op_32249"), val = int32(3)]; + int32 var_32274 = const()[name = string("op_32274"), val = int32(1)]; + tensor out_85_axes_0 = const()[name = string("out_85_axes_0"), val = tensor([1])]; + fp16 var_32291_to_fp16 = const()[name = string("op_32291_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_32291_to_fp16, x = inputs_85_cast_fp16)[name = string("out_85_cast_fp16")]; + tensor obj_85_gamma_0_to_fp16 = const()[name = string("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841083072)))]; + tensor obj_85_beta_0_to_fp16 = const()[name = string("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841085696)))]; + fp16 obj_85_epsilon_0_to_fp16 = const()[name = string("obj_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = string("obj_85_cast_fp16")]; + string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; + tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; + tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; + int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841088320)))]; + tensor layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844365184)))]; + tensor query_43_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("query_43_cast_fp16")]; + string key_43_pad_type_0 = const()[name = string("key_43_pad_type_0"), val = string("valid")]; + tensor key_43_strides_0 = const()[name = string("key_43_strides_0"), val = tensor([1, 1])]; + tensor key_43_pad_0 = const()[name = string("key_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_43_dilations_0 = const()[name = string("key_43_dilations_0"), val = tensor([1, 1])]; + int32 key_43_groups_0 = const()[name = string("key_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(844367808)))]; + tensor key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("key_43_cast_fp16")]; + string value_43_pad_type_0 = const()[name = string("value_43_pad_type_0"), val = string("valid")]; + tensor value_43_strides_0 = const()[name = string("value_43_strides_0"), val = tensor([1, 1])]; + tensor value_43_pad_0 = const()[name = string("value_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_43_dilations_0 = const()[name = string("value_43_dilations_0"), val = tensor([1, 1])]; + int32 value_43_groups_0 = const()[name = string("value_43_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(847644672)))]; + tensor layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850921536)))]; + tensor value_43_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = string("value_43_cast_fp16")]; + tensor var_32329_begin_0 = const()[name = string("op_32329_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32329_end_0 = const()[name = string("op_32329_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32329_end_mask_0 = const()[name = string("op_32329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32329_cast_fp16 = slice_by_index(begin = var_32329_begin_0, end = var_32329_end_0, end_mask = var_32329_end_mask_0, x = query_43_cast_fp16)[name = string("op_32329_cast_fp16")]; + tensor var_32333_begin_0 = const()[name = string("op_32333_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32333_end_0 = const()[name = string("op_32333_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_32333_end_mask_0 = const()[name = string("op_32333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32333_cast_fp16 = slice_by_index(begin = var_32333_begin_0, end = var_32333_end_0, end_mask = var_32333_end_mask_0, x = query_43_cast_fp16)[name = string("op_32333_cast_fp16")]; + tensor var_32337_begin_0 = const()[name = string("op_32337_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32337_end_0 = const()[name = string("op_32337_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_32337_end_mask_0 = const()[name = string("op_32337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32337_cast_fp16 = slice_by_index(begin = var_32337_begin_0, end = var_32337_end_0, end_mask = var_32337_end_mask_0, x = query_43_cast_fp16)[name = string("op_32337_cast_fp16")]; + tensor var_32341_begin_0 = const()[name = string("op_32341_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32341_end_0 = const()[name = string("op_32341_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_32341_end_mask_0 = const()[name = string("op_32341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32341_cast_fp16 = slice_by_index(begin = var_32341_begin_0, end = var_32341_end_0, end_mask = var_32341_end_mask_0, x = query_43_cast_fp16)[name = string("op_32341_cast_fp16")]; + tensor var_32345_begin_0 = const()[name = string("op_32345_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32345_end_0 = const()[name = string("op_32345_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_32345_end_mask_0 = const()[name = string("op_32345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32345_cast_fp16 = slice_by_index(begin = var_32345_begin_0, end = var_32345_end_0, end_mask = var_32345_end_mask_0, x = query_43_cast_fp16)[name = string("op_32345_cast_fp16")]; + tensor var_32349_begin_0 = const()[name = string("op_32349_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32349_end_0 = const()[name = string("op_32349_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_32349_end_mask_0 = const()[name = string("op_32349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32349_cast_fp16 = slice_by_index(begin = var_32349_begin_0, end = var_32349_end_0, end_mask = var_32349_end_mask_0, x = query_43_cast_fp16)[name = string("op_32349_cast_fp16")]; + tensor var_32353_begin_0 = const()[name = string("op_32353_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32353_end_0 = const()[name = string("op_32353_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_32353_end_mask_0 = const()[name = string("op_32353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32353_cast_fp16 = slice_by_index(begin = var_32353_begin_0, end = var_32353_end_0, end_mask = var_32353_end_mask_0, x = query_43_cast_fp16)[name = string("op_32353_cast_fp16")]; + tensor var_32357_begin_0 = const()[name = string("op_32357_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32357_end_0 = const()[name = string("op_32357_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_32357_end_mask_0 = const()[name = string("op_32357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32357_cast_fp16 = slice_by_index(begin = var_32357_begin_0, end = var_32357_end_0, end_mask = var_32357_end_mask_0, x = query_43_cast_fp16)[name = string("op_32357_cast_fp16")]; + tensor var_32361_begin_0 = const()[name = string("op_32361_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32361_end_0 = const()[name = string("op_32361_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_32361_end_mask_0 = const()[name = string("op_32361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32361_cast_fp16 = slice_by_index(begin = var_32361_begin_0, end = var_32361_end_0, end_mask = var_32361_end_mask_0, x = query_43_cast_fp16)[name = string("op_32361_cast_fp16")]; + tensor var_32365_begin_0 = const()[name = string("op_32365_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32365_end_0 = const()[name = string("op_32365_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_32365_end_mask_0 = const()[name = string("op_32365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32365_cast_fp16 = slice_by_index(begin = var_32365_begin_0, end = var_32365_end_0, end_mask = var_32365_end_mask_0, x = query_43_cast_fp16)[name = string("op_32365_cast_fp16")]; + tensor var_32369_begin_0 = const()[name = string("op_32369_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32369_end_0 = const()[name = string("op_32369_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_32369_end_mask_0 = const()[name = string("op_32369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32369_cast_fp16 = slice_by_index(begin = var_32369_begin_0, end = var_32369_end_0, end_mask = var_32369_end_mask_0, x = query_43_cast_fp16)[name = string("op_32369_cast_fp16")]; + tensor var_32373_begin_0 = const()[name = string("op_32373_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32373_end_0 = const()[name = string("op_32373_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_32373_end_mask_0 = const()[name = string("op_32373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32373_cast_fp16 = slice_by_index(begin = var_32373_begin_0, end = var_32373_end_0, end_mask = var_32373_end_mask_0, x = query_43_cast_fp16)[name = string("op_32373_cast_fp16")]; + tensor var_32377_begin_0 = const()[name = string("op_32377_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32377_end_0 = const()[name = string("op_32377_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_32377_end_mask_0 = const()[name = string("op_32377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32377_cast_fp16 = slice_by_index(begin = var_32377_begin_0, end = var_32377_end_0, end_mask = var_32377_end_mask_0, x = query_43_cast_fp16)[name = string("op_32377_cast_fp16")]; + tensor var_32381_begin_0 = const()[name = string("op_32381_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32381_end_0 = const()[name = string("op_32381_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_32381_end_mask_0 = const()[name = string("op_32381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32381_cast_fp16 = slice_by_index(begin = var_32381_begin_0, end = var_32381_end_0, end_mask = var_32381_end_mask_0, x = query_43_cast_fp16)[name = string("op_32381_cast_fp16")]; + tensor var_32385_begin_0 = const()[name = string("op_32385_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32385_end_0 = const()[name = string("op_32385_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_32385_end_mask_0 = const()[name = string("op_32385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32385_cast_fp16 = slice_by_index(begin = var_32385_begin_0, end = var_32385_end_0, end_mask = var_32385_end_mask_0, x = query_43_cast_fp16)[name = string("op_32385_cast_fp16")]; + tensor var_32389_begin_0 = const()[name = string("op_32389_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32389_end_0 = const()[name = string("op_32389_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_32389_end_mask_0 = const()[name = string("op_32389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32389_cast_fp16 = slice_by_index(begin = var_32389_begin_0, end = var_32389_end_0, end_mask = var_32389_end_mask_0, x = query_43_cast_fp16)[name = string("op_32389_cast_fp16")]; + tensor var_32393_begin_0 = const()[name = string("op_32393_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32393_end_0 = const()[name = string("op_32393_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_32393_end_mask_0 = const()[name = string("op_32393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32393_cast_fp16 = slice_by_index(begin = var_32393_begin_0, end = var_32393_end_0, end_mask = var_32393_end_mask_0, x = query_43_cast_fp16)[name = string("op_32393_cast_fp16")]; + tensor var_32397_begin_0 = const()[name = string("op_32397_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32397_end_0 = const()[name = string("op_32397_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_32397_end_mask_0 = const()[name = string("op_32397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32397_cast_fp16 = slice_by_index(begin = var_32397_begin_0, end = var_32397_end_0, end_mask = var_32397_end_mask_0, x = query_43_cast_fp16)[name = string("op_32397_cast_fp16")]; + tensor var_32401_begin_0 = const()[name = string("op_32401_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32401_end_0 = const()[name = string("op_32401_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_32401_end_mask_0 = const()[name = string("op_32401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32401_cast_fp16 = slice_by_index(begin = var_32401_begin_0, end = var_32401_end_0, end_mask = var_32401_end_mask_0, x = query_43_cast_fp16)[name = string("op_32401_cast_fp16")]; + tensor var_32405_begin_0 = const()[name = string("op_32405_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32405_end_0 = const()[name = string("op_32405_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_32405_end_mask_0 = const()[name = string("op_32405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32405_cast_fp16 = slice_by_index(begin = var_32405_begin_0, end = var_32405_end_0, end_mask = var_32405_end_mask_0, x = query_43_cast_fp16)[name = string("op_32405_cast_fp16")]; + tensor var_32414_begin_0 = const()[name = string("op_32414_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32414_end_0 = const()[name = string("op_32414_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32414_end_mask_0 = const()[name = string("op_32414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32414_cast_fp16 = slice_by_index(begin = var_32414_begin_0, end = var_32414_end_0, end_mask = var_32414_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32414_cast_fp16")]; + tensor var_32421_begin_0 = const()[name = string("op_32421_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32421_end_0 = const()[name = string("op_32421_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32421_end_mask_0 = const()[name = string("op_32421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32421_cast_fp16 = slice_by_index(begin = var_32421_begin_0, end = var_32421_end_0, end_mask = var_32421_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32421_cast_fp16")]; + tensor var_32428_begin_0 = const()[name = string("op_32428_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32428_end_0 = const()[name = string("op_32428_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32428_end_mask_0 = const()[name = string("op_32428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32428_cast_fp16 = slice_by_index(begin = var_32428_begin_0, end = var_32428_end_0, end_mask = var_32428_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32428_cast_fp16")]; + tensor var_32435_begin_0 = const()[name = string("op_32435_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32435_end_0 = const()[name = string("op_32435_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32435_end_mask_0 = const()[name = string("op_32435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32435_cast_fp16 = slice_by_index(begin = var_32435_begin_0, end = var_32435_end_0, end_mask = var_32435_end_mask_0, x = var_32329_cast_fp16)[name = string("op_32435_cast_fp16")]; + tensor var_32442_begin_0 = const()[name = string("op_32442_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32442_end_0 = const()[name = string("op_32442_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32442_end_mask_0 = const()[name = string("op_32442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32442_cast_fp16 = slice_by_index(begin = var_32442_begin_0, end = var_32442_end_0, end_mask = var_32442_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32442_cast_fp16")]; + tensor var_32449_begin_0 = const()[name = string("op_32449_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32449_end_0 = const()[name = string("op_32449_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32449_end_mask_0 = const()[name = string("op_32449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32449_cast_fp16 = slice_by_index(begin = var_32449_begin_0, end = var_32449_end_0, end_mask = var_32449_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32449_cast_fp16")]; + tensor var_32456_begin_0 = const()[name = string("op_32456_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32456_end_0 = const()[name = string("op_32456_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32456_end_mask_0 = const()[name = string("op_32456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32456_cast_fp16 = slice_by_index(begin = var_32456_begin_0, end = var_32456_end_0, end_mask = var_32456_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32456_cast_fp16")]; + tensor var_32463_begin_0 = const()[name = string("op_32463_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32463_end_0 = const()[name = string("op_32463_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32463_end_mask_0 = const()[name = string("op_32463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32463_cast_fp16 = slice_by_index(begin = var_32463_begin_0, end = var_32463_end_0, end_mask = var_32463_end_mask_0, x = var_32333_cast_fp16)[name = string("op_32463_cast_fp16")]; + tensor var_32470_begin_0 = const()[name = string("op_32470_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32470_end_0 = const()[name = string("op_32470_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32470_end_mask_0 = const()[name = string("op_32470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32470_cast_fp16 = slice_by_index(begin = var_32470_begin_0, end = var_32470_end_0, end_mask = var_32470_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32470_cast_fp16")]; + tensor var_32477_begin_0 = const()[name = string("op_32477_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32477_end_0 = const()[name = string("op_32477_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32477_end_mask_0 = const()[name = string("op_32477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32477_cast_fp16 = slice_by_index(begin = var_32477_begin_0, end = var_32477_end_0, end_mask = var_32477_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32477_cast_fp16")]; + tensor var_32484_begin_0 = const()[name = string("op_32484_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32484_end_0 = const()[name = string("op_32484_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32484_end_mask_0 = const()[name = string("op_32484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32484_cast_fp16 = slice_by_index(begin = var_32484_begin_0, end = var_32484_end_0, end_mask = var_32484_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32484_cast_fp16")]; + tensor var_32491_begin_0 = const()[name = string("op_32491_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32491_end_0 = const()[name = string("op_32491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32491_end_mask_0 = const()[name = string("op_32491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32491_cast_fp16 = slice_by_index(begin = var_32491_begin_0, end = var_32491_end_0, end_mask = var_32491_end_mask_0, x = var_32337_cast_fp16)[name = string("op_32491_cast_fp16")]; + tensor var_32498_begin_0 = const()[name = string("op_32498_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32498_end_0 = const()[name = string("op_32498_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32498_end_mask_0 = const()[name = string("op_32498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32498_cast_fp16 = slice_by_index(begin = var_32498_begin_0, end = var_32498_end_0, end_mask = var_32498_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32498_cast_fp16")]; + tensor var_32505_begin_0 = const()[name = string("op_32505_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32505_end_0 = const()[name = string("op_32505_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32505_end_mask_0 = const()[name = string("op_32505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32505_cast_fp16 = slice_by_index(begin = var_32505_begin_0, end = var_32505_end_0, end_mask = var_32505_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32505_cast_fp16")]; + tensor var_32512_begin_0 = const()[name = string("op_32512_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32512_end_0 = const()[name = string("op_32512_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32512_end_mask_0 = const()[name = string("op_32512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32512_cast_fp16 = slice_by_index(begin = var_32512_begin_0, end = var_32512_end_0, end_mask = var_32512_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32512_cast_fp16")]; + tensor var_32519_begin_0 = const()[name = string("op_32519_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32519_end_0 = const()[name = string("op_32519_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32519_end_mask_0 = const()[name = string("op_32519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32519_cast_fp16 = slice_by_index(begin = var_32519_begin_0, end = var_32519_end_0, end_mask = var_32519_end_mask_0, x = var_32341_cast_fp16)[name = string("op_32519_cast_fp16")]; + tensor var_32526_begin_0 = const()[name = string("op_32526_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32526_end_0 = const()[name = string("op_32526_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32526_end_mask_0 = const()[name = string("op_32526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32526_cast_fp16 = slice_by_index(begin = var_32526_begin_0, end = var_32526_end_0, end_mask = var_32526_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32526_cast_fp16")]; + tensor var_32533_begin_0 = const()[name = string("op_32533_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32533_end_0 = const()[name = string("op_32533_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32533_end_mask_0 = const()[name = string("op_32533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32533_cast_fp16 = slice_by_index(begin = var_32533_begin_0, end = var_32533_end_0, end_mask = var_32533_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32533_cast_fp16")]; + tensor var_32540_begin_0 = const()[name = string("op_32540_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32540_end_0 = const()[name = string("op_32540_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32540_end_mask_0 = const()[name = string("op_32540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32540_cast_fp16 = slice_by_index(begin = var_32540_begin_0, end = var_32540_end_0, end_mask = var_32540_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32540_cast_fp16")]; + tensor var_32547_begin_0 = const()[name = string("op_32547_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32547_end_0 = const()[name = string("op_32547_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32547_end_mask_0 = const()[name = string("op_32547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32547_cast_fp16 = slice_by_index(begin = var_32547_begin_0, end = var_32547_end_0, end_mask = var_32547_end_mask_0, x = var_32345_cast_fp16)[name = string("op_32547_cast_fp16")]; + tensor var_32554_begin_0 = const()[name = string("op_32554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32554_end_0 = const()[name = string("op_32554_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32554_end_mask_0 = const()[name = string("op_32554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32554_cast_fp16 = slice_by_index(begin = var_32554_begin_0, end = var_32554_end_0, end_mask = var_32554_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32554_cast_fp16")]; + tensor var_32561_begin_0 = const()[name = string("op_32561_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32561_end_0 = const()[name = string("op_32561_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32561_end_mask_0 = const()[name = string("op_32561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32561_cast_fp16 = slice_by_index(begin = var_32561_begin_0, end = var_32561_end_0, end_mask = var_32561_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32561_cast_fp16")]; + tensor var_32568_begin_0 = const()[name = string("op_32568_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32568_end_0 = const()[name = string("op_32568_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32568_end_mask_0 = const()[name = string("op_32568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32568_cast_fp16 = slice_by_index(begin = var_32568_begin_0, end = var_32568_end_0, end_mask = var_32568_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32568_cast_fp16")]; + tensor var_32575_begin_0 = const()[name = string("op_32575_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32575_end_0 = const()[name = string("op_32575_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32575_end_mask_0 = const()[name = string("op_32575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32575_cast_fp16 = slice_by_index(begin = var_32575_begin_0, end = var_32575_end_0, end_mask = var_32575_end_mask_0, x = var_32349_cast_fp16)[name = string("op_32575_cast_fp16")]; + tensor var_32582_begin_0 = const()[name = string("op_32582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32582_end_0 = const()[name = string("op_32582_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32582_end_mask_0 = const()[name = string("op_32582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32582_cast_fp16 = slice_by_index(begin = var_32582_begin_0, end = var_32582_end_0, end_mask = var_32582_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32582_cast_fp16")]; + tensor var_32589_begin_0 = const()[name = string("op_32589_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32589_end_0 = const()[name = string("op_32589_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32589_end_mask_0 = const()[name = string("op_32589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32589_cast_fp16 = slice_by_index(begin = var_32589_begin_0, end = var_32589_end_0, end_mask = var_32589_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32589_cast_fp16")]; + tensor var_32596_begin_0 = const()[name = string("op_32596_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32596_end_0 = const()[name = string("op_32596_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32596_end_mask_0 = const()[name = string("op_32596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32596_cast_fp16 = slice_by_index(begin = var_32596_begin_0, end = var_32596_end_0, end_mask = var_32596_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32596_cast_fp16")]; + tensor var_32603_begin_0 = const()[name = string("op_32603_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32603_end_0 = const()[name = string("op_32603_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32603_end_mask_0 = const()[name = string("op_32603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32603_cast_fp16 = slice_by_index(begin = var_32603_begin_0, end = var_32603_end_0, end_mask = var_32603_end_mask_0, x = var_32353_cast_fp16)[name = string("op_32603_cast_fp16")]; + tensor var_32610_begin_0 = const()[name = string("op_32610_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32610_end_0 = const()[name = string("op_32610_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32610_end_mask_0 = const()[name = string("op_32610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32610_cast_fp16 = slice_by_index(begin = var_32610_begin_0, end = var_32610_end_0, end_mask = var_32610_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32610_cast_fp16")]; + tensor var_32617_begin_0 = const()[name = string("op_32617_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32617_end_0 = const()[name = string("op_32617_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32617_end_mask_0 = const()[name = string("op_32617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32617_cast_fp16 = slice_by_index(begin = var_32617_begin_0, end = var_32617_end_0, end_mask = var_32617_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32617_cast_fp16")]; + tensor var_32624_begin_0 = const()[name = string("op_32624_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32624_end_0 = const()[name = string("op_32624_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32624_end_mask_0 = const()[name = string("op_32624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32624_cast_fp16 = slice_by_index(begin = var_32624_begin_0, end = var_32624_end_0, end_mask = var_32624_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32624_cast_fp16")]; + tensor var_32631_begin_0 = const()[name = string("op_32631_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32631_end_0 = const()[name = string("op_32631_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32631_end_mask_0 = const()[name = string("op_32631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32631_cast_fp16 = slice_by_index(begin = var_32631_begin_0, end = var_32631_end_0, end_mask = var_32631_end_mask_0, x = var_32357_cast_fp16)[name = string("op_32631_cast_fp16")]; + tensor var_32638_begin_0 = const()[name = string("op_32638_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32638_end_0 = const()[name = string("op_32638_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32638_end_mask_0 = const()[name = string("op_32638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32638_cast_fp16 = slice_by_index(begin = var_32638_begin_0, end = var_32638_end_0, end_mask = var_32638_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32638_cast_fp16")]; + tensor var_32645_begin_0 = const()[name = string("op_32645_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32645_end_0 = const()[name = string("op_32645_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32645_end_mask_0 = const()[name = string("op_32645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32645_cast_fp16 = slice_by_index(begin = var_32645_begin_0, end = var_32645_end_0, end_mask = var_32645_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32645_cast_fp16")]; + tensor var_32652_begin_0 = const()[name = string("op_32652_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32652_end_0 = const()[name = string("op_32652_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32652_end_mask_0 = const()[name = string("op_32652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32652_cast_fp16 = slice_by_index(begin = var_32652_begin_0, end = var_32652_end_0, end_mask = var_32652_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32652_cast_fp16")]; + tensor var_32659_begin_0 = const()[name = string("op_32659_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32659_end_0 = const()[name = string("op_32659_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32659_end_mask_0 = const()[name = string("op_32659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32659_cast_fp16 = slice_by_index(begin = var_32659_begin_0, end = var_32659_end_0, end_mask = var_32659_end_mask_0, x = var_32361_cast_fp16)[name = string("op_32659_cast_fp16")]; + tensor var_32666_begin_0 = const()[name = string("op_32666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32666_end_0 = const()[name = string("op_32666_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32666_end_mask_0 = const()[name = string("op_32666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32666_cast_fp16 = slice_by_index(begin = var_32666_begin_0, end = var_32666_end_0, end_mask = var_32666_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32666_cast_fp16")]; + tensor var_32673_begin_0 = const()[name = string("op_32673_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32673_end_0 = const()[name = string("op_32673_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32673_end_mask_0 = const()[name = string("op_32673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32673_cast_fp16 = slice_by_index(begin = var_32673_begin_0, end = var_32673_end_0, end_mask = var_32673_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32673_cast_fp16")]; + tensor var_32680_begin_0 = const()[name = string("op_32680_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32680_end_0 = const()[name = string("op_32680_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32680_end_mask_0 = const()[name = string("op_32680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32680_cast_fp16 = slice_by_index(begin = var_32680_begin_0, end = var_32680_end_0, end_mask = var_32680_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32680_cast_fp16")]; + tensor var_32687_begin_0 = const()[name = string("op_32687_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32687_end_0 = const()[name = string("op_32687_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32687_end_mask_0 = const()[name = string("op_32687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32687_cast_fp16 = slice_by_index(begin = var_32687_begin_0, end = var_32687_end_0, end_mask = var_32687_end_mask_0, x = var_32365_cast_fp16)[name = string("op_32687_cast_fp16")]; + tensor var_32694_begin_0 = const()[name = string("op_32694_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32694_end_0 = const()[name = string("op_32694_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32694_end_mask_0 = const()[name = string("op_32694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32694_cast_fp16 = slice_by_index(begin = var_32694_begin_0, end = var_32694_end_0, end_mask = var_32694_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32694_cast_fp16")]; + tensor var_32701_begin_0 = const()[name = string("op_32701_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32701_end_0 = const()[name = string("op_32701_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32701_end_mask_0 = const()[name = string("op_32701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32701_cast_fp16 = slice_by_index(begin = var_32701_begin_0, end = var_32701_end_0, end_mask = var_32701_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32701_cast_fp16")]; + tensor var_32708_begin_0 = const()[name = string("op_32708_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32708_end_0 = const()[name = string("op_32708_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32708_end_mask_0 = const()[name = string("op_32708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32708_cast_fp16 = slice_by_index(begin = var_32708_begin_0, end = var_32708_end_0, end_mask = var_32708_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32708_cast_fp16")]; + tensor var_32715_begin_0 = const()[name = string("op_32715_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32715_end_0 = const()[name = string("op_32715_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32715_end_mask_0 = const()[name = string("op_32715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32715_cast_fp16 = slice_by_index(begin = var_32715_begin_0, end = var_32715_end_0, end_mask = var_32715_end_mask_0, x = var_32369_cast_fp16)[name = string("op_32715_cast_fp16")]; + tensor var_32722_begin_0 = const()[name = string("op_32722_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32722_end_0 = const()[name = string("op_32722_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32722_end_mask_0 = const()[name = string("op_32722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32722_cast_fp16 = slice_by_index(begin = var_32722_begin_0, end = var_32722_end_0, end_mask = var_32722_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32722_cast_fp16")]; + tensor var_32729_begin_0 = const()[name = string("op_32729_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32729_end_0 = const()[name = string("op_32729_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32729_end_mask_0 = const()[name = string("op_32729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32729_cast_fp16 = slice_by_index(begin = var_32729_begin_0, end = var_32729_end_0, end_mask = var_32729_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32729_cast_fp16")]; + tensor var_32736_begin_0 = const()[name = string("op_32736_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32736_end_0 = const()[name = string("op_32736_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32736_end_mask_0 = const()[name = string("op_32736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32736_cast_fp16 = slice_by_index(begin = var_32736_begin_0, end = var_32736_end_0, end_mask = var_32736_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32736_cast_fp16")]; + tensor var_32743_begin_0 = const()[name = string("op_32743_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32743_end_0 = const()[name = string("op_32743_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32743_end_mask_0 = const()[name = string("op_32743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32743_cast_fp16 = slice_by_index(begin = var_32743_begin_0, end = var_32743_end_0, end_mask = var_32743_end_mask_0, x = var_32373_cast_fp16)[name = string("op_32743_cast_fp16")]; + tensor var_32750_begin_0 = const()[name = string("op_32750_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32750_end_0 = const()[name = string("op_32750_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32750_end_mask_0 = const()[name = string("op_32750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32750_cast_fp16 = slice_by_index(begin = var_32750_begin_0, end = var_32750_end_0, end_mask = var_32750_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32750_cast_fp16")]; + tensor var_32757_begin_0 = const()[name = string("op_32757_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32757_end_0 = const()[name = string("op_32757_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32757_end_mask_0 = const()[name = string("op_32757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32757_cast_fp16 = slice_by_index(begin = var_32757_begin_0, end = var_32757_end_0, end_mask = var_32757_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32757_cast_fp16")]; + tensor var_32764_begin_0 = const()[name = string("op_32764_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32764_end_0 = const()[name = string("op_32764_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32764_end_mask_0 = const()[name = string("op_32764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32764_cast_fp16 = slice_by_index(begin = var_32764_begin_0, end = var_32764_end_0, end_mask = var_32764_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32764_cast_fp16")]; + tensor var_32771_begin_0 = const()[name = string("op_32771_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32771_end_0 = const()[name = string("op_32771_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32771_end_mask_0 = const()[name = string("op_32771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32771_cast_fp16 = slice_by_index(begin = var_32771_begin_0, end = var_32771_end_0, end_mask = var_32771_end_mask_0, x = var_32377_cast_fp16)[name = string("op_32771_cast_fp16")]; + tensor var_32778_begin_0 = const()[name = string("op_32778_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32778_end_0 = const()[name = string("op_32778_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32778_end_mask_0 = const()[name = string("op_32778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32778_cast_fp16 = slice_by_index(begin = var_32778_begin_0, end = var_32778_end_0, end_mask = var_32778_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32778_cast_fp16")]; + tensor var_32785_begin_0 = const()[name = string("op_32785_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32785_end_0 = const()[name = string("op_32785_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32785_end_mask_0 = const()[name = string("op_32785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32785_cast_fp16 = slice_by_index(begin = var_32785_begin_0, end = var_32785_end_0, end_mask = var_32785_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32785_cast_fp16")]; + tensor var_32792_begin_0 = const()[name = string("op_32792_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32792_end_0 = const()[name = string("op_32792_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32792_end_mask_0 = const()[name = string("op_32792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32792_cast_fp16 = slice_by_index(begin = var_32792_begin_0, end = var_32792_end_0, end_mask = var_32792_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32792_cast_fp16")]; + tensor var_32799_begin_0 = const()[name = string("op_32799_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32799_end_0 = const()[name = string("op_32799_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32799_end_mask_0 = const()[name = string("op_32799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32799_cast_fp16 = slice_by_index(begin = var_32799_begin_0, end = var_32799_end_0, end_mask = var_32799_end_mask_0, x = var_32381_cast_fp16)[name = string("op_32799_cast_fp16")]; + tensor var_32806_begin_0 = const()[name = string("op_32806_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32806_end_0 = const()[name = string("op_32806_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32806_end_mask_0 = const()[name = string("op_32806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32806_cast_fp16 = slice_by_index(begin = var_32806_begin_0, end = var_32806_end_0, end_mask = var_32806_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32806_cast_fp16")]; + tensor var_32813_begin_0 = const()[name = string("op_32813_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32813_end_0 = const()[name = string("op_32813_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32813_end_mask_0 = const()[name = string("op_32813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32813_cast_fp16 = slice_by_index(begin = var_32813_begin_0, end = var_32813_end_0, end_mask = var_32813_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32813_cast_fp16")]; + tensor var_32820_begin_0 = const()[name = string("op_32820_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32820_end_0 = const()[name = string("op_32820_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32820_end_mask_0 = const()[name = string("op_32820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32820_cast_fp16 = slice_by_index(begin = var_32820_begin_0, end = var_32820_end_0, end_mask = var_32820_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32820_cast_fp16")]; + tensor var_32827_begin_0 = const()[name = string("op_32827_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32827_end_0 = const()[name = string("op_32827_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32827_end_mask_0 = const()[name = string("op_32827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32827_cast_fp16 = slice_by_index(begin = var_32827_begin_0, end = var_32827_end_0, end_mask = var_32827_end_mask_0, x = var_32385_cast_fp16)[name = string("op_32827_cast_fp16")]; + tensor var_32834_begin_0 = const()[name = string("op_32834_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32834_end_0 = const()[name = string("op_32834_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32834_end_mask_0 = const()[name = string("op_32834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32834_cast_fp16 = slice_by_index(begin = var_32834_begin_0, end = var_32834_end_0, end_mask = var_32834_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32834_cast_fp16")]; + tensor var_32841_begin_0 = const()[name = string("op_32841_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32841_end_0 = const()[name = string("op_32841_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32841_end_mask_0 = const()[name = string("op_32841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32841_cast_fp16 = slice_by_index(begin = var_32841_begin_0, end = var_32841_end_0, end_mask = var_32841_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32841_cast_fp16")]; + tensor var_32848_begin_0 = const()[name = string("op_32848_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32848_end_0 = const()[name = string("op_32848_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32848_end_mask_0 = const()[name = string("op_32848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32848_cast_fp16 = slice_by_index(begin = var_32848_begin_0, end = var_32848_end_0, end_mask = var_32848_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32848_cast_fp16")]; + tensor var_32855_begin_0 = const()[name = string("op_32855_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32855_end_0 = const()[name = string("op_32855_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32855_end_mask_0 = const()[name = string("op_32855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32855_cast_fp16 = slice_by_index(begin = var_32855_begin_0, end = var_32855_end_0, end_mask = var_32855_end_mask_0, x = var_32389_cast_fp16)[name = string("op_32855_cast_fp16")]; + tensor var_32862_begin_0 = const()[name = string("op_32862_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32862_end_0 = const()[name = string("op_32862_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32862_end_mask_0 = const()[name = string("op_32862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32862_cast_fp16 = slice_by_index(begin = var_32862_begin_0, end = var_32862_end_0, end_mask = var_32862_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32862_cast_fp16")]; + tensor var_32869_begin_0 = const()[name = string("op_32869_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32869_end_0 = const()[name = string("op_32869_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32869_end_mask_0 = const()[name = string("op_32869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32869_cast_fp16 = slice_by_index(begin = var_32869_begin_0, end = var_32869_end_0, end_mask = var_32869_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32869_cast_fp16")]; + tensor var_32876_begin_0 = const()[name = string("op_32876_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32876_end_0 = const()[name = string("op_32876_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32876_end_mask_0 = const()[name = string("op_32876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32876_cast_fp16 = slice_by_index(begin = var_32876_begin_0, end = var_32876_end_0, end_mask = var_32876_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32876_cast_fp16")]; + tensor var_32883_begin_0 = const()[name = string("op_32883_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32883_end_0 = const()[name = string("op_32883_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32883_end_mask_0 = const()[name = string("op_32883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32883_cast_fp16 = slice_by_index(begin = var_32883_begin_0, end = var_32883_end_0, end_mask = var_32883_end_mask_0, x = var_32393_cast_fp16)[name = string("op_32883_cast_fp16")]; + tensor var_32890_begin_0 = const()[name = string("op_32890_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32890_end_0 = const()[name = string("op_32890_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32890_end_mask_0 = const()[name = string("op_32890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32890_cast_fp16 = slice_by_index(begin = var_32890_begin_0, end = var_32890_end_0, end_mask = var_32890_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32890_cast_fp16")]; + tensor var_32897_begin_0 = const()[name = string("op_32897_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32897_end_0 = const()[name = string("op_32897_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32897_end_mask_0 = const()[name = string("op_32897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32897_cast_fp16 = slice_by_index(begin = var_32897_begin_0, end = var_32897_end_0, end_mask = var_32897_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32897_cast_fp16")]; + tensor var_32904_begin_0 = const()[name = string("op_32904_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32904_end_0 = const()[name = string("op_32904_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32904_end_mask_0 = const()[name = string("op_32904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32904_cast_fp16 = slice_by_index(begin = var_32904_begin_0, end = var_32904_end_0, end_mask = var_32904_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32904_cast_fp16")]; + tensor var_32911_begin_0 = const()[name = string("op_32911_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32911_end_0 = const()[name = string("op_32911_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32911_end_mask_0 = const()[name = string("op_32911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32911_cast_fp16 = slice_by_index(begin = var_32911_begin_0, end = var_32911_end_0, end_mask = var_32911_end_mask_0, x = var_32397_cast_fp16)[name = string("op_32911_cast_fp16")]; + tensor var_32918_begin_0 = const()[name = string("op_32918_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32918_end_0 = const()[name = string("op_32918_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32918_end_mask_0 = const()[name = string("op_32918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32918_cast_fp16 = slice_by_index(begin = var_32918_begin_0, end = var_32918_end_0, end_mask = var_32918_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32918_cast_fp16")]; + tensor var_32925_begin_0 = const()[name = string("op_32925_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32925_end_0 = const()[name = string("op_32925_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32925_end_mask_0 = const()[name = string("op_32925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32925_cast_fp16 = slice_by_index(begin = var_32925_begin_0, end = var_32925_end_0, end_mask = var_32925_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32925_cast_fp16")]; + tensor var_32932_begin_0 = const()[name = string("op_32932_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32932_end_0 = const()[name = string("op_32932_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32932_end_mask_0 = const()[name = string("op_32932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32932_cast_fp16 = slice_by_index(begin = var_32932_begin_0, end = var_32932_end_0, end_mask = var_32932_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32932_cast_fp16")]; + tensor var_32939_begin_0 = const()[name = string("op_32939_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32939_end_0 = const()[name = string("op_32939_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32939_end_mask_0 = const()[name = string("op_32939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32939_cast_fp16 = slice_by_index(begin = var_32939_begin_0, end = var_32939_end_0, end_mask = var_32939_end_mask_0, x = var_32401_cast_fp16)[name = string("op_32939_cast_fp16")]; + tensor var_32946_begin_0 = const()[name = string("op_32946_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32946_end_0 = const()[name = string("op_32946_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_32946_end_mask_0 = const()[name = string("op_32946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32946_cast_fp16 = slice_by_index(begin = var_32946_begin_0, end = var_32946_end_0, end_mask = var_32946_end_mask_0, x = var_32405_cast_fp16)[name = string("op_32946_cast_fp16")]; + tensor var_32953_begin_0 = const()[name = string("op_32953_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_32953_end_0 = const()[name = string("op_32953_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_32953_end_mask_0 = const()[name = string("op_32953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32953_cast_fp16 = slice_by_index(begin = var_32953_begin_0, end = var_32953_end_0, end_mask = var_32953_end_mask_0, x = var_32405_cast_fp16)[name = string("op_32953_cast_fp16")]; + tensor var_32960_begin_0 = const()[name = string("op_32960_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32960_end_0 = const()[name = string("op_32960_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32960_end_mask_0 = const()[name = string("op_32960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32960_cast_fp16 = slice_by_index(begin = var_32960_begin_0, end = var_32960_end_0, end_mask = var_32960_end_mask_0, x = var_32405_cast_fp16)[name = string("op_32960_cast_fp16")]; + tensor var_32967_begin_0 = const()[name = string("op_32967_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32967_end_0 = const()[name = string("op_32967_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32967_end_mask_0 = const()[name = string("op_32967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32967_cast_fp16 = slice_by_index(begin = var_32967_begin_0, end = var_32967_end_0, end_mask = var_32967_end_mask_0, x = var_32405_cast_fp16)[name = string("op_32967_cast_fp16")]; + tensor k_43_perm_0 = const()[name = string("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_32972_begin_0 = const()[name = string("op_32972_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32972_end_0 = const()[name = string("op_32972_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_32972_end_mask_0 = const()[name = string("op_32972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_43_cast_fp16 = transpose(perm = k_43_perm_0, x = key_43_cast_fp16)[name = string("transpose_10")]; + tensor var_32972_cast_fp16 = slice_by_index(begin = var_32972_begin_0, end = var_32972_end_0, end_mask = var_32972_end_mask_0, x = k_43_cast_fp16)[name = string("op_32972_cast_fp16")]; + tensor var_32976_begin_0 = const()[name = string("op_32976_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_32976_end_0 = const()[name = string("op_32976_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_32976_end_mask_0 = const()[name = string("op_32976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32976_cast_fp16 = slice_by_index(begin = var_32976_begin_0, end = var_32976_end_0, end_mask = var_32976_end_mask_0, x = k_43_cast_fp16)[name = string("op_32976_cast_fp16")]; + tensor var_32980_begin_0 = const()[name = string("op_32980_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_32980_end_0 = const()[name = string("op_32980_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_32980_end_mask_0 = const()[name = string("op_32980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32980_cast_fp16 = slice_by_index(begin = var_32980_begin_0, end = var_32980_end_0, end_mask = var_32980_end_mask_0, x = k_43_cast_fp16)[name = string("op_32980_cast_fp16")]; + tensor var_32984_begin_0 = const()[name = string("op_32984_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_32984_end_0 = const()[name = string("op_32984_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_32984_end_mask_0 = const()[name = string("op_32984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32984_cast_fp16 = slice_by_index(begin = var_32984_begin_0, end = var_32984_end_0, end_mask = var_32984_end_mask_0, x = k_43_cast_fp16)[name = string("op_32984_cast_fp16")]; + tensor var_32988_begin_0 = const()[name = string("op_32988_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_32988_end_0 = const()[name = string("op_32988_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_32988_end_mask_0 = const()[name = string("op_32988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32988_cast_fp16 = slice_by_index(begin = var_32988_begin_0, end = var_32988_end_0, end_mask = var_32988_end_mask_0, x = k_43_cast_fp16)[name = string("op_32988_cast_fp16")]; + tensor var_32992_begin_0 = const()[name = string("op_32992_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_32992_end_0 = const()[name = string("op_32992_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_32992_end_mask_0 = const()[name = string("op_32992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32992_cast_fp16 = slice_by_index(begin = var_32992_begin_0, end = var_32992_end_0, end_mask = var_32992_end_mask_0, x = k_43_cast_fp16)[name = string("op_32992_cast_fp16")]; + tensor var_32996_begin_0 = const()[name = string("op_32996_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_32996_end_0 = const()[name = string("op_32996_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_32996_end_mask_0 = const()[name = string("op_32996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32996_cast_fp16 = slice_by_index(begin = var_32996_begin_0, end = var_32996_end_0, end_mask = var_32996_end_mask_0, x = k_43_cast_fp16)[name = string("op_32996_cast_fp16")]; + tensor var_33000_begin_0 = const()[name = string("op_33000_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_33000_end_0 = const()[name = string("op_33000_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_33000_end_mask_0 = const()[name = string("op_33000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33000_cast_fp16 = slice_by_index(begin = var_33000_begin_0, end = var_33000_end_0, end_mask = var_33000_end_mask_0, x = k_43_cast_fp16)[name = string("op_33000_cast_fp16")]; + tensor var_33004_begin_0 = const()[name = string("op_33004_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_33004_end_0 = const()[name = string("op_33004_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_33004_end_mask_0 = const()[name = string("op_33004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33004_cast_fp16 = slice_by_index(begin = var_33004_begin_0, end = var_33004_end_0, end_mask = var_33004_end_mask_0, x = k_43_cast_fp16)[name = string("op_33004_cast_fp16")]; + tensor var_33008_begin_0 = const()[name = string("op_33008_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_33008_end_0 = const()[name = string("op_33008_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_33008_end_mask_0 = const()[name = string("op_33008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33008_cast_fp16 = slice_by_index(begin = var_33008_begin_0, end = var_33008_end_0, end_mask = var_33008_end_mask_0, x = k_43_cast_fp16)[name = string("op_33008_cast_fp16")]; + tensor var_33012_begin_0 = const()[name = string("op_33012_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_33012_end_0 = const()[name = string("op_33012_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_33012_end_mask_0 = const()[name = string("op_33012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33012_cast_fp16 = slice_by_index(begin = var_33012_begin_0, end = var_33012_end_0, end_mask = var_33012_end_mask_0, x = k_43_cast_fp16)[name = string("op_33012_cast_fp16")]; + tensor var_33016_begin_0 = const()[name = string("op_33016_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_33016_end_0 = const()[name = string("op_33016_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_33016_end_mask_0 = const()[name = string("op_33016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33016_cast_fp16 = slice_by_index(begin = var_33016_begin_0, end = var_33016_end_0, end_mask = var_33016_end_mask_0, x = k_43_cast_fp16)[name = string("op_33016_cast_fp16")]; + tensor var_33020_begin_0 = const()[name = string("op_33020_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_33020_end_0 = const()[name = string("op_33020_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_33020_end_mask_0 = const()[name = string("op_33020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33020_cast_fp16 = slice_by_index(begin = var_33020_begin_0, end = var_33020_end_0, end_mask = var_33020_end_mask_0, x = k_43_cast_fp16)[name = string("op_33020_cast_fp16")]; + tensor var_33024_begin_0 = const()[name = string("op_33024_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_33024_end_0 = const()[name = string("op_33024_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_33024_end_mask_0 = const()[name = string("op_33024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33024_cast_fp16 = slice_by_index(begin = var_33024_begin_0, end = var_33024_end_0, end_mask = var_33024_end_mask_0, x = k_43_cast_fp16)[name = string("op_33024_cast_fp16")]; + tensor var_33028_begin_0 = const()[name = string("op_33028_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_33028_end_0 = const()[name = string("op_33028_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_33028_end_mask_0 = const()[name = string("op_33028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33028_cast_fp16 = slice_by_index(begin = var_33028_begin_0, end = var_33028_end_0, end_mask = var_33028_end_mask_0, x = k_43_cast_fp16)[name = string("op_33028_cast_fp16")]; + tensor var_33032_begin_0 = const()[name = string("op_33032_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_33032_end_0 = const()[name = string("op_33032_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_33032_end_mask_0 = const()[name = string("op_33032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33032_cast_fp16 = slice_by_index(begin = var_33032_begin_0, end = var_33032_end_0, end_mask = var_33032_end_mask_0, x = k_43_cast_fp16)[name = string("op_33032_cast_fp16")]; + tensor var_33036_begin_0 = const()[name = string("op_33036_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_33036_end_0 = const()[name = string("op_33036_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_33036_end_mask_0 = const()[name = string("op_33036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33036_cast_fp16 = slice_by_index(begin = var_33036_begin_0, end = var_33036_end_0, end_mask = var_33036_end_mask_0, x = k_43_cast_fp16)[name = string("op_33036_cast_fp16")]; + tensor var_33040_begin_0 = const()[name = string("op_33040_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_33040_end_0 = const()[name = string("op_33040_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_33040_end_mask_0 = const()[name = string("op_33040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33040_cast_fp16 = slice_by_index(begin = var_33040_begin_0, end = var_33040_end_0, end_mask = var_33040_end_mask_0, x = k_43_cast_fp16)[name = string("op_33040_cast_fp16")]; + tensor var_33044_begin_0 = const()[name = string("op_33044_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_33044_end_0 = const()[name = string("op_33044_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_33044_end_mask_0 = const()[name = string("op_33044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33044_cast_fp16 = slice_by_index(begin = var_33044_begin_0, end = var_33044_end_0, end_mask = var_33044_end_mask_0, x = k_43_cast_fp16)[name = string("op_33044_cast_fp16")]; + tensor var_33048_begin_0 = const()[name = string("op_33048_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_33048_end_0 = const()[name = string("op_33048_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_33048_end_mask_0 = const()[name = string("op_33048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33048_cast_fp16 = slice_by_index(begin = var_33048_begin_0, end = var_33048_end_0, end_mask = var_33048_end_mask_0, x = k_43_cast_fp16)[name = string("op_33048_cast_fp16")]; + tensor var_33050_begin_0 = const()[name = string("op_33050_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33050_end_0 = const()[name = string("op_33050_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33050_end_mask_0 = const()[name = string("op_33050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33050_cast_fp16 = slice_by_index(begin = var_33050_begin_0, end = var_33050_end_0, end_mask = var_33050_end_mask_0, x = value_43_cast_fp16)[name = string("op_33050_cast_fp16")]; + tensor var_33054_begin_0 = const()[name = string("op_33054_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33054_end_0 = const()[name = string("op_33054_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_33054_end_mask_0 = const()[name = string("op_33054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33054_cast_fp16 = slice_by_index(begin = var_33054_begin_0, end = var_33054_end_0, end_mask = var_33054_end_mask_0, x = value_43_cast_fp16)[name = string("op_33054_cast_fp16")]; + tensor var_33058_begin_0 = const()[name = string("op_33058_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33058_end_0 = const()[name = string("op_33058_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_33058_end_mask_0 = const()[name = string("op_33058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33058_cast_fp16 = slice_by_index(begin = var_33058_begin_0, end = var_33058_end_0, end_mask = var_33058_end_mask_0, x = value_43_cast_fp16)[name = string("op_33058_cast_fp16")]; + tensor var_33062_begin_0 = const()[name = string("op_33062_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33062_end_0 = const()[name = string("op_33062_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_33062_end_mask_0 = const()[name = string("op_33062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33062_cast_fp16 = slice_by_index(begin = var_33062_begin_0, end = var_33062_end_0, end_mask = var_33062_end_mask_0, x = value_43_cast_fp16)[name = string("op_33062_cast_fp16")]; + tensor var_33066_begin_0 = const()[name = string("op_33066_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33066_end_0 = const()[name = string("op_33066_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_33066_end_mask_0 = const()[name = string("op_33066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33066_cast_fp16 = slice_by_index(begin = var_33066_begin_0, end = var_33066_end_0, end_mask = var_33066_end_mask_0, x = value_43_cast_fp16)[name = string("op_33066_cast_fp16")]; + tensor var_33070_begin_0 = const()[name = string("op_33070_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33070_end_0 = const()[name = string("op_33070_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_33070_end_mask_0 = const()[name = string("op_33070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33070_cast_fp16 = slice_by_index(begin = var_33070_begin_0, end = var_33070_end_0, end_mask = var_33070_end_mask_0, x = value_43_cast_fp16)[name = string("op_33070_cast_fp16")]; + tensor var_33074_begin_0 = const()[name = string("op_33074_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33074_end_0 = const()[name = string("op_33074_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_33074_end_mask_0 = const()[name = string("op_33074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33074_cast_fp16 = slice_by_index(begin = var_33074_begin_0, end = var_33074_end_0, end_mask = var_33074_end_mask_0, x = value_43_cast_fp16)[name = string("op_33074_cast_fp16")]; + tensor var_33078_begin_0 = const()[name = string("op_33078_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33078_end_0 = const()[name = string("op_33078_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_33078_end_mask_0 = const()[name = string("op_33078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33078_cast_fp16 = slice_by_index(begin = var_33078_begin_0, end = var_33078_end_0, end_mask = var_33078_end_mask_0, x = value_43_cast_fp16)[name = string("op_33078_cast_fp16")]; + tensor var_33082_begin_0 = const()[name = string("op_33082_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33082_end_0 = const()[name = string("op_33082_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_33082_end_mask_0 = const()[name = string("op_33082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33082_cast_fp16 = slice_by_index(begin = var_33082_begin_0, end = var_33082_end_0, end_mask = var_33082_end_mask_0, x = value_43_cast_fp16)[name = string("op_33082_cast_fp16")]; + tensor var_33086_begin_0 = const()[name = string("op_33086_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33086_end_0 = const()[name = string("op_33086_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_33086_end_mask_0 = const()[name = string("op_33086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33086_cast_fp16 = slice_by_index(begin = var_33086_begin_0, end = var_33086_end_0, end_mask = var_33086_end_mask_0, x = value_43_cast_fp16)[name = string("op_33086_cast_fp16")]; + tensor var_33090_begin_0 = const()[name = string("op_33090_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33090_end_0 = const()[name = string("op_33090_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_33090_end_mask_0 = const()[name = string("op_33090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33090_cast_fp16 = slice_by_index(begin = var_33090_begin_0, end = var_33090_end_0, end_mask = var_33090_end_mask_0, x = value_43_cast_fp16)[name = string("op_33090_cast_fp16")]; + tensor var_33094_begin_0 = const()[name = string("op_33094_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33094_end_0 = const()[name = string("op_33094_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_33094_end_mask_0 = const()[name = string("op_33094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33094_cast_fp16 = slice_by_index(begin = var_33094_begin_0, end = var_33094_end_0, end_mask = var_33094_end_mask_0, x = value_43_cast_fp16)[name = string("op_33094_cast_fp16")]; + tensor var_33098_begin_0 = const()[name = string("op_33098_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33098_end_0 = const()[name = string("op_33098_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_33098_end_mask_0 = const()[name = string("op_33098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33098_cast_fp16 = slice_by_index(begin = var_33098_begin_0, end = var_33098_end_0, end_mask = var_33098_end_mask_0, x = value_43_cast_fp16)[name = string("op_33098_cast_fp16")]; + tensor var_33102_begin_0 = const()[name = string("op_33102_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33102_end_0 = const()[name = string("op_33102_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_33102_end_mask_0 = const()[name = string("op_33102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33102_cast_fp16 = slice_by_index(begin = var_33102_begin_0, end = var_33102_end_0, end_mask = var_33102_end_mask_0, x = value_43_cast_fp16)[name = string("op_33102_cast_fp16")]; + tensor var_33106_begin_0 = const()[name = string("op_33106_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33106_end_0 = const()[name = string("op_33106_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_33106_end_mask_0 = const()[name = string("op_33106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33106_cast_fp16 = slice_by_index(begin = var_33106_begin_0, end = var_33106_end_0, end_mask = var_33106_end_mask_0, x = value_43_cast_fp16)[name = string("op_33106_cast_fp16")]; + tensor var_33110_begin_0 = const()[name = string("op_33110_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33110_end_0 = const()[name = string("op_33110_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_33110_end_mask_0 = const()[name = string("op_33110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33110_cast_fp16 = slice_by_index(begin = var_33110_begin_0, end = var_33110_end_0, end_mask = var_33110_end_mask_0, x = value_43_cast_fp16)[name = string("op_33110_cast_fp16")]; + tensor var_33114_begin_0 = const()[name = string("op_33114_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33114_end_0 = const()[name = string("op_33114_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_33114_end_mask_0 = const()[name = string("op_33114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33114_cast_fp16 = slice_by_index(begin = var_33114_begin_0, end = var_33114_end_0, end_mask = var_33114_end_mask_0, x = value_43_cast_fp16)[name = string("op_33114_cast_fp16")]; + tensor var_33118_begin_0 = const()[name = string("op_33118_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33118_end_0 = const()[name = string("op_33118_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_33118_end_mask_0 = const()[name = string("op_33118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33118_cast_fp16 = slice_by_index(begin = var_33118_begin_0, end = var_33118_end_0, end_mask = var_33118_end_mask_0, x = value_43_cast_fp16)[name = string("op_33118_cast_fp16")]; + tensor var_33122_begin_0 = const()[name = string("op_33122_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33122_end_0 = const()[name = string("op_33122_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_33122_end_mask_0 = const()[name = string("op_33122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33122_cast_fp16 = slice_by_index(begin = var_33122_begin_0, end = var_33122_end_0, end_mask = var_33122_end_mask_0, x = value_43_cast_fp16)[name = string("op_33122_cast_fp16")]; + tensor var_33126_begin_0 = const()[name = string("op_33126_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33126_end_0 = const()[name = string("op_33126_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_33126_end_mask_0 = const()[name = string("op_33126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33126_cast_fp16 = slice_by_index(begin = var_33126_begin_0, end = var_33126_end_0, end_mask = var_33126_end_mask_0, x = value_43_cast_fp16)[name = string("op_33126_cast_fp16")]; + string _SplitHeadsQ__mh_w_3361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3361_equation_0, values = (var_32972_cast_fp16, var_32414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3361_cast_fp16")]; + string _SplitHeadsQ__mh_w_3363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3363_equation_0, values = (var_32972_cast_fp16, var_32421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3363_cast_fp16")]; + string _SplitHeadsQ__mh_w_3365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3365_equation_0, values = (var_32972_cast_fp16, var_32428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3365_cast_fp16")]; + string _SplitHeadsQ__mh_w_3367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3367_equation_0, values = (var_32972_cast_fp16, var_32435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3367_cast_fp16")]; + string _SplitHeadsQ__mh_w_3369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3369_equation_0, values = (var_32976_cast_fp16, var_32442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3369_cast_fp16")]; + string _SplitHeadsQ__mh_w_3371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3371_equation_0, values = (var_32976_cast_fp16, var_32449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3371_cast_fp16")]; + string _SplitHeadsQ__mh_w_3373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3373_equation_0, values = (var_32976_cast_fp16, var_32456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3373_cast_fp16")]; + string _SplitHeadsQ__mh_w_3375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3375_equation_0, values = (var_32976_cast_fp16, var_32463_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3375_cast_fp16")]; + string _SplitHeadsQ__mh_w_3377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3377_equation_0, values = (var_32980_cast_fp16, var_32470_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3377_cast_fp16")]; + string _SplitHeadsQ__mh_w_3379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3379_equation_0, values = (var_32980_cast_fp16, var_32477_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3379_cast_fp16")]; + string _SplitHeadsQ__mh_w_3381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3381_equation_0, values = (var_32980_cast_fp16, var_32484_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3381_cast_fp16")]; + string _SplitHeadsQ__mh_w_3383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3383_equation_0, values = (var_32980_cast_fp16, var_32491_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3383_cast_fp16")]; + string _SplitHeadsQ__mh_w_3385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3385_equation_0, values = (var_32984_cast_fp16, var_32498_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3385_cast_fp16")]; + string _SplitHeadsQ__mh_w_3387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3387_equation_0, values = (var_32984_cast_fp16, var_32505_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3387_cast_fp16")]; + string _SplitHeadsQ__mh_w_3389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3389_equation_0, values = (var_32984_cast_fp16, var_32512_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3389_cast_fp16")]; + string _SplitHeadsQ__mh_w_3391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3391_equation_0, values = (var_32984_cast_fp16, var_32519_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3391_cast_fp16")]; + string _SplitHeadsQ__mh_w_3393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3393_equation_0, values = (var_32988_cast_fp16, var_32526_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3393_cast_fp16")]; + string _SplitHeadsQ__mh_w_3395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3395_equation_0, values = (var_32988_cast_fp16, var_32533_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3395_cast_fp16")]; + string _SplitHeadsQ__mh_w_3397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3397_equation_0, values = (var_32988_cast_fp16, var_32540_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3397_cast_fp16")]; + string _SplitHeadsQ__mh_w_3399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3399_equation_0, values = (var_32988_cast_fp16, var_32547_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3399_cast_fp16")]; + string _SplitHeadsQ__mh_w_3401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3401_equation_0, values = (var_32992_cast_fp16, var_32554_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3401_cast_fp16")]; + string _SplitHeadsQ__mh_w_3403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3403_equation_0, values = (var_32992_cast_fp16, var_32561_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3403_cast_fp16")]; + string _SplitHeadsQ__mh_w_3405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3405_equation_0, values = (var_32992_cast_fp16, var_32568_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3405_cast_fp16")]; + string _SplitHeadsQ__mh_w_3407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3407_equation_0, values = (var_32992_cast_fp16, var_32575_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3407_cast_fp16")]; + string _SplitHeadsQ__mh_w_3409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3409_equation_0, values = (var_32996_cast_fp16, var_32582_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3409_cast_fp16")]; + string _SplitHeadsQ__mh_w_3411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3411_equation_0, values = (var_32996_cast_fp16, var_32589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3411_cast_fp16")]; + string _SplitHeadsQ__mh_w_3413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3413_equation_0, values = (var_32996_cast_fp16, var_32596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3413_cast_fp16")]; + string _SplitHeadsQ__mh_w_3415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3415_equation_0, values = (var_32996_cast_fp16, var_32603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3415_cast_fp16")]; + string _SplitHeadsQ__mh_w_3417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3417_equation_0, values = (var_33000_cast_fp16, var_32610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3417_cast_fp16")]; + string _SplitHeadsQ__mh_w_3419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3419_equation_0, values = (var_33000_cast_fp16, var_32617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3419_cast_fp16")]; + string _SplitHeadsQ__mh_w_3421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3421_equation_0, values = (var_33000_cast_fp16, var_32624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3421_cast_fp16")]; + string _SplitHeadsQ__mh_w_3423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3423_equation_0, values = (var_33000_cast_fp16, var_32631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3423_cast_fp16")]; + string _SplitHeadsQ__mh_w_3425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3425_equation_0, values = (var_33004_cast_fp16, var_32638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3425_cast_fp16")]; + string _SplitHeadsQ__mh_w_3427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3427_equation_0, values = (var_33004_cast_fp16, var_32645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3427_cast_fp16")]; + string _SplitHeadsQ__mh_w_3429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3429_equation_0, values = (var_33004_cast_fp16, var_32652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3429_cast_fp16")]; + string _SplitHeadsQ__mh_w_3431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3431_equation_0, values = (var_33004_cast_fp16, var_32659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3431_cast_fp16")]; + string _SplitHeadsQ__mh_w_3433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3433_equation_0, values = (var_33008_cast_fp16, var_32666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3433_cast_fp16")]; + string _SplitHeadsQ__mh_w_3435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3435_equation_0, values = (var_33008_cast_fp16, var_32673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3435_cast_fp16")]; + string _SplitHeadsQ__mh_w_3437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3437_equation_0, values = (var_33008_cast_fp16, var_32680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3437_cast_fp16")]; + string _SplitHeadsQ__mh_w_3439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3439_equation_0, values = (var_33008_cast_fp16, var_32687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3439_cast_fp16")]; + string _SplitHeadsQ__mh_w_3441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3441_equation_0, values = (var_33012_cast_fp16, var_32694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3441_cast_fp16")]; + string _SplitHeadsQ__mh_w_3443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3443_equation_0, values = (var_33012_cast_fp16, var_32701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3443_cast_fp16")]; + string _SplitHeadsQ__mh_w_3445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3445_equation_0, values = (var_33012_cast_fp16, var_32708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3445_cast_fp16")]; + string _SplitHeadsQ__mh_w_3447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3447_equation_0, values = (var_33012_cast_fp16, var_32715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3447_cast_fp16")]; + string _SplitHeadsQ__mh_w_3449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3449_equation_0, values = (var_33016_cast_fp16, var_32722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3449_cast_fp16")]; + string _SplitHeadsQ__mh_w_3451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3451_equation_0, values = (var_33016_cast_fp16, var_32729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3451_cast_fp16")]; + string _SplitHeadsQ__mh_w_3453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3453_equation_0, values = (var_33016_cast_fp16, var_32736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3453_cast_fp16")]; + string _SplitHeadsQ__mh_w_3455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3455_equation_0, values = (var_33016_cast_fp16, var_32743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3455_cast_fp16")]; + string _SplitHeadsQ__mh_w_3457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3457_equation_0, values = (var_33020_cast_fp16, var_32750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3457_cast_fp16")]; + string _SplitHeadsQ__mh_w_3459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3459_equation_0, values = (var_33020_cast_fp16, var_32757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3459_cast_fp16")]; + string _SplitHeadsQ__mh_w_3461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3461_equation_0, values = (var_33020_cast_fp16, var_32764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3461_cast_fp16")]; + string _SplitHeadsQ__mh_w_3463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3463_equation_0, values = (var_33020_cast_fp16, var_32771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3463_cast_fp16")]; + string _SplitHeadsQ__mh_w_3465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3465_equation_0, values = (var_33024_cast_fp16, var_32778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3465_cast_fp16")]; + string _SplitHeadsQ__mh_w_3467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3467_equation_0, values = (var_33024_cast_fp16, var_32785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3467_cast_fp16")]; + string _SplitHeadsQ__mh_w_3469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3469_equation_0, values = (var_33024_cast_fp16, var_32792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3469_cast_fp16")]; + string _SplitHeadsQ__mh_w_3471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3471_equation_0, values = (var_33024_cast_fp16, var_32799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3471_cast_fp16")]; + string _SplitHeadsQ__mh_w_3473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3473_equation_0, values = (var_33028_cast_fp16, var_32806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3473_cast_fp16")]; + string _SplitHeadsQ__mh_w_3475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3475_equation_0, values = (var_33028_cast_fp16, var_32813_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3475_cast_fp16")]; + string _SplitHeadsQ__mh_w_3477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3477_equation_0, values = (var_33028_cast_fp16, var_32820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3477_cast_fp16")]; + string _SplitHeadsQ__mh_w_3479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3479_equation_0, values = (var_33028_cast_fp16, var_32827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3479_cast_fp16")]; + string _SplitHeadsQ__mh_w_3481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3481_equation_0, values = (var_33032_cast_fp16, var_32834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3481_cast_fp16")]; + string _SplitHeadsQ__mh_w_3483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3483_equation_0, values = (var_33032_cast_fp16, var_32841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3483_cast_fp16")]; + string _SplitHeadsQ__mh_w_3485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3485_equation_0, values = (var_33032_cast_fp16, var_32848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3485_cast_fp16")]; + string _SplitHeadsQ__mh_w_3487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3487_equation_0, values = (var_33032_cast_fp16, var_32855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3487_cast_fp16")]; + string _SplitHeadsQ__mh_w_3489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3489_equation_0, values = (var_33036_cast_fp16, var_32862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3489_cast_fp16")]; + string _SplitHeadsQ__mh_w_3491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3491_equation_0, values = (var_33036_cast_fp16, var_32869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3491_cast_fp16")]; + string _SplitHeadsQ__mh_w_3493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3493_equation_0, values = (var_33036_cast_fp16, var_32876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3493_cast_fp16")]; + string _SplitHeadsQ__mh_w_3495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3495_equation_0, values = (var_33036_cast_fp16, var_32883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3495_cast_fp16")]; + string _SplitHeadsQ__mh_w_3497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3497_equation_0, values = (var_33040_cast_fp16, var_32890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3497_cast_fp16")]; + string _SplitHeadsQ__mh_w_3499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3499_equation_0, values = (var_33040_cast_fp16, var_32897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3499_cast_fp16")]; + string _SplitHeadsQ__mh_w_3501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3501_equation_0, values = (var_33040_cast_fp16, var_32904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3501_cast_fp16")]; + string _SplitHeadsQ__mh_w_3503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3503_equation_0, values = (var_33040_cast_fp16, var_32911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3503_cast_fp16")]; + string _SplitHeadsQ__mh_w_3505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3505_equation_0, values = (var_33044_cast_fp16, var_32918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3505_cast_fp16")]; + string _SplitHeadsQ__mh_w_3507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3507_equation_0, values = (var_33044_cast_fp16, var_32925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3507_cast_fp16")]; + string _SplitHeadsQ__mh_w_3509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3509_equation_0, values = (var_33044_cast_fp16, var_32932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3509_cast_fp16")]; + string _SplitHeadsQ__mh_w_3511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3511_equation_0, values = (var_33044_cast_fp16, var_32939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3511_cast_fp16")]; + string _SplitHeadsQ__mh_w_3513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3513_equation_0, values = (var_33048_cast_fp16, var_32946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3513_cast_fp16")]; + string _SplitHeadsQ__mh_w_3515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3515_equation_0, values = (var_33048_cast_fp16, var_32953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3515_cast_fp16")]; + string _SplitHeadsQ__mh_w_3517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3517_equation_0, values = (var_33048_cast_fp16, var_32960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3517_cast_fp16")]; + string _SplitHeadsQ__mh_w_3519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3519_equation_0, values = (var_33048_cast_fp16, var_32967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3519_cast_fp16")]; + fp16 var_33289_to_fp16 = const()[name = string("op_33289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3361_cast_fp16, y = var_33289_to_fp16)[name = string("aw_chunk_3361_cast_fp16")]; + fp16 var_33291_to_fp16 = const()[name = string("op_33291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3363_cast_fp16, y = var_33291_to_fp16)[name = string("aw_chunk_3363_cast_fp16")]; + fp16 var_33293_to_fp16 = const()[name = string("op_33293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3365_cast_fp16, y = var_33293_to_fp16)[name = string("aw_chunk_3365_cast_fp16")]; + fp16 var_33295_to_fp16 = const()[name = string("op_33295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3367_cast_fp16, y = var_33295_to_fp16)[name = string("aw_chunk_3367_cast_fp16")]; + fp16 var_33297_to_fp16 = const()[name = string("op_33297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3369_cast_fp16, y = var_33297_to_fp16)[name = string("aw_chunk_3369_cast_fp16")]; + fp16 var_33299_to_fp16 = const()[name = string("op_33299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3371_cast_fp16, y = var_33299_to_fp16)[name = string("aw_chunk_3371_cast_fp16")]; + fp16 var_33301_to_fp16 = const()[name = string("op_33301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3373_cast_fp16, y = var_33301_to_fp16)[name = string("aw_chunk_3373_cast_fp16")]; + fp16 var_33303_to_fp16 = const()[name = string("op_33303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3375_cast_fp16, y = var_33303_to_fp16)[name = string("aw_chunk_3375_cast_fp16")]; + fp16 var_33305_to_fp16 = const()[name = string("op_33305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3377_cast_fp16, y = var_33305_to_fp16)[name = string("aw_chunk_3377_cast_fp16")]; + fp16 var_33307_to_fp16 = const()[name = string("op_33307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3379_cast_fp16, y = var_33307_to_fp16)[name = string("aw_chunk_3379_cast_fp16")]; + fp16 var_33309_to_fp16 = const()[name = string("op_33309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3381_cast_fp16, y = var_33309_to_fp16)[name = string("aw_chunk_3381_cast_fp16")]; + fp16 var_33311_to_fp16 = const()[name = string("op_33311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3383_cast_fp16, y = var_33311_to_fp16)[name = string("aw_chunk_3383_cast_fp16")]; + fp16 var_33313_to_fp16 = const()[name = string("op_33313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3385_cast_fp16, y = var_33313_to_fp16)[name = string("aw_chunk_3385_cast_fp16")]; + fp16 var_33315_to_fp16 = const()[name = string("op_33315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3387_cast_fp16, y = var_33315_to_fp16)[name = string("aw_chunk_3387_cast_fp16")]; + fp16 var_33317_to_fp16 = const()[name = string("op_33317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3389_cast_fp16, y = var_33317_to_fp16)[name = string("aw_chunk_3389_cast_fp16")]; + fp16 var_33319_to_fp16 = const()[name = string("op_33319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3391_cast_fp16, y = var_33319_to_fp16)[name = string("aw_chunk_3391_cast_fp16")]; + fp16 var_33321_to_fp16 = const()[name = string("op_33321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3393_cast_fp16, y = var_33321_to_fp16)[name = string("aw_chunk_3393_cast_fp16")]; + fp16 var_33323_to_fp16 = const()[name = string("op_33323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3395_cast_fp16, y = var_33323_to_fp16)[name = string("aw_chunk_3395_cast_fp16")]; + fp16 var_33325_to_fp16 = const()[name = string("op_33325_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3397_cast_fp16, y = var_33325_to_fp16)[name = string("aw_chunk_3397_cast_fp16")]; + fp16 var_33327_to_fp16 = const()[name = string("op_33327_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3399_cast_fp16, y = var_33327_to_fp16)[name = string("aw_chunk_3399_cast_fp16")]; + fp16 var_33329_to_fp16 = const()[name = string("op_33329_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3401_cast_fp16, y = var_33329_to_fp16)[name = string("aw_chunk_3401_cast_fp16")]; + fp16 var_33331_to_fp16 = const()[name = string("op_33331_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3403_cast_fp16, y = var_33331_to_fp16)[name = string("aw_chunk_3403_cast_fp16")]; + fp16 var_33333_to_fp16 = const()[name = string("op_33333_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3405_cast_fp16, y = var_33333_to_fp16)[name = string("aw_chunk_3405_cast_fp16")]; + fp16 var_33335_to_fp16 = const()[name = string("op_33335_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3407_cast_fp16, y = var_33335_to_fp16)[name = string("aw_chunk_3407_cast_fp16")]; + fp16 var_33337_to_fp16 = const()[name = string("op_33337_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3409_cast_fp16, y = var_33337_to_fp16)[name = string("aw_chunk_3409_cast_fp16")]; + fp16 var_33339_to_fp16 = const()[name = string("op_33339_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3411_cast_fp16, y = var_33339_to_fp16)[name = string("aw_chunk_3411_cast_fp16")]; + fp16 var_33341_to_fp16 = const()[name = string("op_33341_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3413_cast_fp16, y = var_33341_to_fp16)[name = string("aw_chunk_3413_cast_fp16")]; + fp16 var_33343_to_fp16 = const()[name = string("op_33343_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3415_cast_fp16, y = var_33343_to_fp16)[name = string("aw_chunk_3415_cast_fp16")]; + fp16 var_33345_to_fp16 = const()[name = string("op_33345_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3417_cast_fp16, y = var_33345_to_fp16)[name = string("aw_chunk_3417_cast_fp16")]; + fp16 var_33347_to_fp16 = const()[name = string("op_33347_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3419_cast_fp16, y = var_33347_to_fp16)[name = string("aw_chunk_3419_cast_fp16")]; + fp16 var_33349_to_fp16 = const()[name = string("op_33349_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3421_cast_fp16, y = var_33349_to_fp16)[name = string("aw_chunk_3421_cast_fp16")]; + fp16 var_33351_to_fp16 = const()[name = string("op_33351_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3423_cast_fp16, y = var_33351_to_fp16)[name = string("aw_chunk_3423_cast_fp16")]; + fp16 var_33353_to_fp16 = const()[name = string("op_33353_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3425_cast_fp16, y = var_33353_to_fp16)[name = string("aw_chunk_3425_cast_fp16")]; + fp16 var_33355_to_fp16 = const()[name = string("op_33355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3427_cast_fp16, y = var_33355_to_fp16)[name = string("aw_chunk_3427_cast_fp16")]; + fp16 var_33357_to_fp16 = const()[name = string("op_33357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3429_cast_fp16, y = var_33357_to_fp16)[name = string("aw_chunk_3429_cast_fp16")]; + fp16 var_33359_to_fp16 = const()[name = string("op_33359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3431_cast_fp16, y = var_33359_to_fp16)[name = string("aw_chunk_3431_cast_fp16")]; + fp16 var_33361_to_fp16 = const()[name = string("op_33361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3433_cast_fp16, y = var_33361_to_fp16)[name = string("aw_chunk_3433_cast_fp16")]; + fp16 var_33363_to_fp16 = const()[name = string("op_33363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3435_cast_fp16, y = var_33363_to_fp16)[name = string("aw_chunk_3435_cast_fp16")]; + fp16 var_33365_to_fp16 = const()[name = string("op_33365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3437_cast_fp16, y = var_33365_to_fp16)[name = string("aw_chunk_3437_cast_fp16")]; + fp16 var_33367_to_fp16 = const()[name = string("op_33367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3439_cast_fp16, y = var_33367_to_fp16)[name = string("aw_chunk_3439_cast_fp16")]; + fp16 var_33369_to_fp16 = const()[name = string("op_33369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3441_cast_fp16, y = var_33369_to_fp16)[name = string("aw_chunk_3441_cast_fp16")]; + fp16 var_33371_to_fp16 = const()[name = string("op_33371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3443_cast_fp16, y = var_33371_to_fp16)[name = string("aw_chunk_3443_cast_fp16")]; + fp16 var_33373_to_fp16 = const()[name = string("op_33373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3445_cast_fp16, y = var_33373_to_fp16)[name = string("aw_chunk_3445_cast_fp16")]; + fp16 var_33375_to_fp16 = const()[name = string("op_33375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3447_cast_fp16, y = var_33375_to_fp16)[name = string("aw_chunk_3447_cast_fp16")]; + fp16 var_33377_to_fp16 = const()[name = string("op_33377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3449_cast_fp16, y = var_33377_to_fp16)[name = string("aw_chunk_3449_cast_fp16")]; + fp16 var_33379_to_fp16 = const()[name = string("op_33379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3451_cast_fp16, y = var_33379_to_fp16)[name = string("aw_chunk_3451_cast_fp16")]; + fp16 var_33381_to_fp16 = const()[name = string("op_33381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3453_cast_fp16, y = var_33381_to_fp16)[name = string("aw_chunk_3453_cast_fp16")]; + fp16 var_33383_to_fp16 = const()[name = string("op_33383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3455_cast_fp16, y = var_33383_to_fp16)[name = string("aw_chunk_3455_cast_fp16")]; + fp16 var_33385_to_fp16 = const()[name = string("op_33385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3457_cast_fp16, y = var_33385_to_fp16)[name = string("aw_chunk_3457_cast_fp16")]; + fp16 var_33387_to_fp16 = const()[name = string("op_33387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3459_cast_fp16, y = var_33387_to_fp16)[name = string("aw_chunk_3459_cast_fp16")]; + fp16 var_33389_to_fp16 = const()[name = string("op_33389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3461_cast_fp16, y = var_33389_to_fp16)[name = string("aw_chunk_3461_cast_fp16")]; + fp16 var_33391_to_fp16 = const()[name = string("op_33391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3463_cast_fp16, y = var_33391_to_fp16)[name = string("aw_chunk_3463_cast_fp16")]; + fp16 var_33393_to_fp16 = const()[name = string("op_33393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3465_cast_fp16, y = var_33393_to_fp16)[name = string("aw_chunk_3465_cast_fp16")]; + fp16 var_33395_to_fp16 = const()[name = string("op_33395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3467_cast_fp16, y = var_33395_to_fp16)[name = string("aw_chunk_3467_cast_fp16")]; + fp16 var_33397_to_fp16 = const()[name = string("op_33397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3469_cast_fp16, y = var_33397_to_fp16)[name = string("aw_chunk_3469_cast_fp16")]; + fp16 var_33399_to_fp16 = const()[name = string("op_33399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3471_cast_fp16, y = var_33399_to_fp16)[name = string("aw_chunk_3471_cast_fp16")]; + fp16 var_33401_to_fp16 = const()[name = string("op_33401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3473_cast_fp16, y = var_33401_to_fp16)[name = string("aw_chunk_3473_cast_fp16")]; + fp16 var_33403_to_fp16 = const()[name = string("op_33403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3475_cast_fp16, y = var_33403_to_fp16)[name = string("aw_chunk_3475_cast_fp16")]; + fp16 var_33405_to_fp16 = const()[name = string("op_33405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3477_cast_fp16, y = var_33405_to_fp16)[name = string("aw_chunk_3477_cast_fp16")]; + fp16 var_33407_to_fp16 = const()[name = string("op_33407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3479_cast_fp16, y = var_33407_to_fp16)[name = string("aw_chunk_3479_cast_fp16")]; + fp16 var_33409_to_fp16 = const()[name = string("op_33409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3481_cast_fp16, y = var_33409_to_fp16)[name = string("aw_chunk_3481_cast_fp16")]; + fp16 var_33411_to_fp16 = const()[name = string("op_33411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3483_cast_fp16, y = var_33411_to_fp16)[name = string("aw_chunk_3483_cast_fp16")]; + fp16 var_33413_to_fp16 = const()[name = string("op_33413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3485_cast_fp16, y = var_33413_to_fp16)[name = string("aw_chunk_3485_cast_fp16")]; + fp16 var_33415_to_fp16 = const()[name = string("op_33415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3487_cast_fp16, y = var_33415_to_fp16)[name = string("aw_chunk_3487_cast_fp16")]; + fp16 var_33417_to_fp16 = const()[name = string("op_33417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3489_cast_fp16, y = var_33417_to_fp16)[name = string("aw_chunk_3489_cast_fp16")]; + fp16 var_33419_to_fp16 = const()[name = string("op_33419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3491_cast_fp16, y = var_33419_to_fp16)[name = string("aw_chunk_3491_cast_fp16")]; + fp16 var_33421_to_fp16 = const()[name = string("op_33421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3493_cast_fp16, y = var_33421_to_fp16)[name = string("aw_chunk_3493_cast_fp16")]; + fp16 var_33423_to_fp16 = const()[name = string("op_33423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3495_cast_fp16, y = var_33423_to_fp16)[name = string("aw_chunk_3495_cast_fp16")]; + fp16 var_33425_to_fp16 = const()[name = string("op_33425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3497_cast_fp16, y = var_33425_to_fp16)[name = string("aw_chunk_3497_cast_fp16")]; + fp16 var_33427_to_fp16 = const()[name = string("op_33427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3499_cast_fp16, y = var_33427_to_fp16)[name = string("aw_chunk_3499_cast_fp16")]; + fp16 var_33429_to_fp16 = const()[name = string("op_33429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3501_cast_fp16, y = var_33429_to_fp16)[name = string("aw_chunk_3501_cast_fp16")]; + fp16 var_33431_to_fp16 = const()[name = string("op_33431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3503_cast_fp16, y = var_33431_to_fp16)[name = string("aw_chunk_3503_cast_fp16")]; + fp16 var_33433_to_fp16 = const()[name = string("op_33433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3505_cast_fp16, y = var_33433_to_fp16)[name = string("aw_chunk_3505_cast_fp16")]; + fp16 var_33435_to_fp16 = const()[name = string("op_33435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3507_cast_fp16, y = var_33435_to_fp16)[name = string("aw_chunk_3507_cast_fp16")]; + fp16 var_33437_to_fp16 = const()[name = string("op_33437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3509_cast_fp16, y = var_33437_to_fp16)[name = string("aw_chunk_3509_cast_fp16")]; + fp16 var_33439_to_fp16 = const()[name = string("op_33439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3511_cast_fp16, y = var_33439_to_fp16)[name = string("aw_chunk_3511_cast_fp16")]; + fp16 var_33441_to_fp16 = const()[name = string("op_33441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3513_cast_fp16, y = var_33441_to_fp16)[name = string("aw_chunk_3513_cast_fp16")]; + fp16 var_33443_to_fp16 = const()[name = string("op_33443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3515_cast_fp16, y = var_33443_to_fp16)[name = string("aw_chunk_3515_cast_fp16")]; + fp16 var_33445_to_fp16 = const()[name = string("op_33445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3517_cast_fp16, y = var_33445_to_fp16)[name = string("aw_chunk_3517_cast_fp16")]; + fp16 var_33447_to_fp16 = const()[name = string("op_33447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3519_cast_fp16, y = var_33447_to_fp16)[name = string("aw_chunk_3519_cast_fp16")]; + tensor var_33449_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3361_cast_fp16)[name = string("op_33449_cast_fp16")]; + tensor var_33450_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3363_cast_fp16)[name = string("op_33450_cast_fp16")]; + tensor var_33451_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3365_cast_fp16)[name = string("op_33451_cast_fp16")]; + tensor var_33452_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3367_cast_fp16)[name = string("op_33452_cast_fp16")]; + tensor var_33453_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3369_cast_fp16)[name = string("op_33453_cast_fp16")]; + tensor var_33454_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3371_cast_fp16)[name = string("op_33454_cast_fp16")]; + tensor var_33455_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3373_cast_fp16)[name = string("op_33455_cast_fp16")]; + tensor var_33456_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3375_cast_fp16)[name = string("op_33456_cast_fp16")]; + tensor var_33457_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3377_cast_fp16)[name = string("op_33457_cast_fp16")]; + tensor var_33458_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3379_cast_fp16)[name = string("op_33458_cast_fp16")]; + tensor var_33459_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3381_cast_fp16)[name = string("op_33459_cast_fp16")]; + tensor var_33460_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3383_cast_fp16)[name = string("op_33460_cast_fp16")]; + tensor var_33461_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3385_cast_fp16)[name = string("op_33461_cast_fp16")]; + tensor var_33462_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3387_cast_fp16)[name = string("op_33462_cast_fp16")]; + tensor var_33463_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3389_cast_fp16)[name = string("op_33463_cast_fp16")]; + tensor var_33464_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3391_cast_fp16)[name = string("op_33464_cast_fp16")]; + tensor var_33465_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3393_cast_fp16)[name = string("op_33465_cast_fp16")]; + tensor var_33466_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3395_cast_fp16)[name = string("op_33466_cast_fp16")]; + tensor var_33467_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3397_cast_fp16)[name = string("op_33467_cast_fp16")]; + tensor var_33468_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3399_cast_fp16)[name = string("op_33468_cast_fp16")]; + tensor var_33469_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3401_cast_fp16)[name = string("op_33469_cast_fp16")]; + tensor var_33470_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3403_cast_fp16)[name = string("op_33470_cast_fp16")]; + tensor var_33471_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3405_cast_fp16)[name = string("op_33471_cast_fp16")]; + tensor var_33472_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3407_cast_fp16)[name = string("op_33472_cast_fp16")]; + tensor var_33473_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3409_cast_fp16)[name = string("op_33473_cast_fp16")]; + tensor var_33474_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3411_cast_fp16)[name = string("op_33474_cast_fp16")]; + tensor var_33475_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3413_cast_fp16)[name = string("op_33475_cast_fp16")]; + tensor var_33476_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3415_cast_fp16)[name = string("op_33476_cast_fp16")]; + tensor var_33477_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3417_cast_fp16)[name = string("op_33477_cast_fp16")]; + tensor var_33478_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3419_cast_fp16)[name = string("op_33478_cast_fp16")]; + tensor var_33479_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3421_cast_fp16)[name = string("op_33479_cast_fp16")]; + tensor var_33480_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3423_cast_fp16)[name = string("op_33480_cast_fp16")]; + tensor var_33481_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3425_cast_fp16)[name = string("op_33481_cast_fp16")]; + tensor var_33482_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3427_cast_fp16)[name = string("op_33482_cast_fp16")]; + tensor var_33483_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3429_cast_fp16)[name = string("op_33483_cast_fp16")]; + tensor var_33484_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3431_cast_fp16)[name = string("op_33484_cast_fp16")]; + tensor var_33485_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3433_cast_fp16)[name = string("op_33485_cast_fp16")]; + tensor var_33486_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3435_cast_fp16)[name = string("op_33486_cast_fp16")]; + tensor var_33487_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3437_cast_fp16)[name = string("op_33487_cast_fp16")]; + tensor var_33488_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3439_cast_fp16)[name = string("op_33488_cast_fp16")]; + tensor var_33489_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3441_cast_fp16)[name = string("op_33489_cast_fp16")]; + tensor var_33490_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3443_cast_fp16)[name = string("op_33490_cast_fp16")]; + tensor var_33491_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3445_cast_fp16)[name = string("op_33491_cast_fp16")]; + tensor var_33492_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3447_cast_fp16)[name = string("op_33492_cast_fp16")]; + tensor var_33493_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3449_cast_fp16)[name = string("op_33493_cast_fp16")]; + tensor var_33494_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3451_cast_fp16)[name = string("op_33494_cast_fp16")]; + tensor var_33495_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3453_cast_fp16)[name = string("op_33495_cast_fp16")]; + tensor var_33496_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3455_cast_fp16)[name = string("op_33496_cast_fp16")]; + tensor var_33497_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3457_cast_fp16)[name = string("op_33497_cast_fp16")]; + tensor var_33498_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3459_cast_fp16)[name = string("op_33498_cast_fp16")]; + tensor var_33499_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3461_cast_fp16)[name = string("op_33499_cast_fp16")]; + tensor var_33500_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3463_cast_fp16)[name = string("op_33500_cast_fp16")]; + tensor var_33501_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3465_cast_fp16)[name = string("op_33501_cast_fp16")]; + tensor var_33502_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3467_cast_fp16)[name = string("op_33502_cast_fp16")]; + tensor var_33503_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3469_cast_fp16)[name = string("op_33503_cast_fp16")]; + tensor var_33504_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3471_cast_fp16)[name = string("op_33504_cast_fp16")]; + tensor var_33505_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3473_cast_fp16)[name = string("op_33505_cast_fp16")]; + tensor var_33506_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3475_cast_fp16)[name = string("op_33506_cast_fp16")]; + tensor var_33507_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3477_cast_fp16)[name = string("op_33507_cast_fp16")]; + tensor var_33508_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3479_cast_fp16)[name = string("op_33508_cast_fp16")]; + tensor var_33509_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3481_cast_fp16)[name = string("op_33509_cast_fp16")]; + tensor var_33510_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3483_cast_fp16)[name = string("op_33510_cast_fp16")]; + tensor var_33511_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3485_cast_fp16)[name = string("op_33511_cast_fp16")]; + tensor var_33512_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3487_cast_fp16)[name = string("op_33512_cast_fp16")]; + tensor var_33513_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3489_cast_fp16)[name = string("op_33513_cast_fp16")]; + tensor var_33514_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3491_cast_fp16)[name = string("op_33514_cast_fp16")]; + tensor var_33515_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3493_cast_fp16)[name = string("op_33515_cast_fp16")]; + tensor var_33516_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3495_cast_fp16)[name = string("op_33516_cast_fp16")]; + tensor var_33517_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3497_cast_fp16)[name = string("op_33517_cast_fp16")]; + tensor var_33518_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3499_cast_fp16)[name = string("op_33518_cast_fp16")]; + tensor var_33519_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3501_cast_fp16)[name = string("op_33519_cast_fp16")]; + tensor var_33520_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3503_cast_fp16)[name = string("op_33520_cast_fp16")]; + tensor var_33521_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3505_cast_fp16)[name = string("op_33521_cast_fp16")]; + tensor var_33522_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3507_cast_fp16)[name = string("op_33522_cast_fp16")]; + tensor var_33523_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3509_cast_fp16)[name = string("op_33523_cast_fp16")]; + tensor var_33524_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3511_cast_fp16)[name = string("op_33524_cast_fp16")]; + tensor var_33525_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3513_cast_fp16)[name = string("op_33525_cast_fp16")]; + tensor var_33526_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3515_cast_fp16)[name = string("op_33526_cast_fp16")]; + tensor var_33527_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3517_cast_fp16)[name = string("op_33527_cast_fp16")]; + tensor var_33528_cast_fp16 = softmax(axis = var_32274, x = aw_chunk_3519_cast_fp16)[name = string("op_33528_cast_fp16")]; + string var_33530_equation_0 = const()[name = string("op_33530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33530_cast_fp16 = einsum(equation = var_33530_equation_0, values = (var_33050_cast_fp16, var_33449_cast_fp16))[name = string("op_33530_cast_fp16")]; + string var_33532_equation_0 = const()[name = string("op_33532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33532_cast_fp16 = einsum(equation = var_33532_equation_0, values = (var_33050_cast_fp16, var_33450_cast_fp16))[name = string("op_33532_cast_fp16")]; + string var_33534_equation_0 = const()[name = string("op_33534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33534_cast_fp16 = einsum(equation = var_33534_equation_0, values = (var_33050_cast_fp16, var_33451_cast_fp16))[name = string("op_33534_cast_fp16")]; + string var_33536_equation_0 = const()[name = string("op_33536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33536_cast_fp16 = einsum(equation = var_33536_equation_0, values = (var_33050_cast_fp16, var_33452_cast_fp16))[name = string("op_33536_cast_fp16")]; + string var_33538_equation_0 = const()[name = string("op_33538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33538_cast_fp16 = einsum(equation = var_33538_equation_0, values = (var_33054_cast_fp16, var_33453_cast_fp16))[name = string("op_33538_cast_fp16")]; + string var_33540_equation_0 = const()[name = string("op_33540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33540_cast_fp16 = einsum(equation = var_33540_equation_0, values = (var_33054_cast_fp16, var_33454_cast_fp16))[name = string("op_33540_cast_fp16")]; + string var_33542_equation_0 = const()[name = string("op_33542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33542_cast_fp16 = einsum(equation = var_33542_equation_0, values = (var_33054_cast_fp16, var_33455_cast_fp16))[name = string("op_33542_cast_fp16")]; + string var_33544_equation_0 = const()[name = string("op_33544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33544_cast_fp16 = einsum(equation = var_33544_equation_0, values = (var_33054_cast_fp16, var_33456_cast_fp16))[name = string("op_33544_cast_fp16")]; + string var_33546_equation_0 = const()[name = string("op_33546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33546_cast_fp16 = einsum(equation = var_33546_equation_0, values = (var_33058_cast_fp16, var_33457_cast_fp16))[name = string("op_33546_cast_fp16")]; + string var_33548_equation_0 = const()[name = string("op_33548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33548_cast_fp16 = einsum(equation = var_33548_equation_0, values = (var_33058_cast_fp16, var_33458_cast_fp16))[name = string("op_33548_cast_fp16")]; + string var_33550_equation_0 = const()[name = string("op_33550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33550_cast_fp16 = einsum(equation = var_33550_equation_0, values = (var_33058_cast_fp16, var_33459_cast_fp16))[name = string("op_33550_cast_fp16")]; + string var_33552_equation_0 = const()[name = string("op_33552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33552_cast_fp16 = einsum(equation = var_33552_equation_0, values = (var_33058_cast_fp16, var_33460_cast_fp16))[name = string("op_33552_cast_fp16")]; + string var_33554_equation_0 = const()[name = string("op_33554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33554_cast_fp16 = einsum(equation = var_33554_equation_0, values = (var_33062_cast_fp16, var_33461_cast_fp16))[name = string("op_33554_cast_fp16")]; + string var_33556_equation_0 = const()[name = string("op_33556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33556_cast_fp16 = einsum(equation = var_33556_equation_0, values = (var_33062_cast_fp16, var_33462_cast_fp16))[name = string("op_33556_cast_fp16")]; + string var_33558_equation_0 = const()[name = string("op_33558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33558_cast_fp16 = einsum(equation = var_33558_equation_0, values = (var_33062_cast_fp16, var_33463_cast_fp16))[name = string("op_33558_cast_fp16")]; + string var_33560_equation_0 = const()[name = string("op_33560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33560_cast_fp16 = einsum(equation = var_33560_equation_0, values = (var_33062_cast_fp16, var_33464_cast_fp16))[name = string("op_33560_cast_fp16")]; + string var_33562_equation_0 = const()[name = string("op_33562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33562_cast_fp16 = einsum(equation = var_33562_equation_0, values = (var_33066_cast_fp16, var_33465_cast_fp16))[name = string("op_33562_cast_fp16")]; + string var_33564_equation_0 = const()[name = string("op_33564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33564_cast_fp16 = einsum(equation = var_33564_equation_0, values = (var_33066_cast_fp16, var_33466_cast_fp16))[name = string("op_33564_cast_fp16")]; + string var_33566_equation_0 = const()[name = string("op_33566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33566_cast_fp16 = einsum(equation = var_33566_equation_0, values = (var_33066_cast_fp16, var_33467_cast_fp16))[name = string("op_33566_cast_fp16")]; + string var_33568_equation_0 = const()[name = string("op_33568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33568_cast_fp16 = einsum(equation = var_33568_equation_0, values = (var_33066_cast_fp16, var_33468_cast_fp16))[name = string("op_33568_cast_fp16")]; + string var_33570_equation_0 = const()[name = string("op_33570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33570_cast_fp16 = einsum(equation = var_33570_equation_0, values = (var_33070_cast_fp16, var_33469_cast_fp16))[name = string("op_33570_cast_fp16")]; + string var_33572_equation_0 = const()[name = string("op_33572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33572_cast_fp16 = einsum(equation = var_33572_equation_0, values = (var_33070_cast_fp16, var_33470_cast_fp16))[name = string("op_33572_cast_fp16")]; + string var_33574_equation_0 = const()[name = string("op_33574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33574_cast_fp16 = einsum(equation = var_33574_equation_0, values = (var_33070_cast_fp16, var_33471_cast_fp16))[name = string("op_33574_cast_fp16")]; + string var_33576_equation_0 = const()[name = string("op_33576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33576_cast_fp16 = einsum(equation = var_33576_equation_0, values = (var_33070_cast_fp16, var_33472_cast_fp16))[name = string("op_33576_cast_fp16")]; + string var_33578_equation_0 = const()[name = string("op_33578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33578_cast_fp16 = einsum(equation = var_33578_equation_0, values = (var_33074_cast_fp16, var_33473_cast_fp16))[name = string("op_33578_cast_fp16")]; + string var_33580_equation_0 = const()[name = string("op_33580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33580_cast_fp16 = einsum(equation = var_33580_equation_0, values = (var_33074_cast_fp16, var_33474_cast_fp16))[name = string("op_33580_cast_fp16")]; + string var_33582_equation_0 = const()[name = string("op_33582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33582_cast_fp16 = einsum(equation = var_33582_equation_0, values = (var_33074_cast_fp16, var_33475_cast_fp16))[name = string("op_33582_cast_fp16")]; + string var_33584_equation_0 = const()[name = string("op_33584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33584_cast_fp16 = einsum(equation = var_33584_equation_0, values = (var_33074_cast_fp16, var_33476_cast_fp16))[name = string("op_33584_cast_fp16")]; + string var_33586_equation_0 = const()[name = string("op_33586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33586_cast_fp16 = einsum(equation = var_33586_equation_0, values = (var_33078_cast_fp16, var_33477_cast_fp16))[name = string("op_33586_cast_fp16")]; + string var_33588_equation_0 = const()[name = string("op_33588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33588_cast_fp16 = einsum(equation = var_33588_equation_0, values = (var_33078_cast_fp16, var_33478_cast_fp16))[name = string("op_33588_cast_fp16")]; + string var_33590_equation_0 = const()[name = string("op_33590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33590_cast_fp16 = einsum(equation = var_33590_equation_0, values = (var_33078_cast_fp16, var_33479_cast_fp16))[name = string("op_33590_cast_fp16")]; + string var_33592_equation_0 = const()[name = string("op_33592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33592_cast_fp16 = einsum(equation = var_33592_equation_0, values = (var_33078_cast_fp16, var_33480_cast_fp16))[name = string("op_33592_cast_fp16")]; + string var_33594_equation_0 = const()[name = string("op_33594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33594_cast_fp16 = einsum(equation = var_33594_equation_0, values = (var_33082_cast_fp16, var_33481_cast_fp16))[name = string("op_33594_cast_fp16")]; + string var_33596_equation_0 = const()[name = string("op_33596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33596_cast_fp16 = einsum(equation = var_33596_equation_0, values = (var_33082_cast_fp16, var_33482_cast_fp16))[name = string("op_33596_cast_fp16")]; + string var_33598_equation_0 = const()[name = string("op_33598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33598_cast_fp16 = einsum(equation = var_33598_equation_0, values = (var_33082_cast_fp16, var_33483_cast_fp16))[name = string("op_33598_cast_fp16")]; + string var_33600_equation_0 = const()[name = string("op_33600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33600_cast_fp16 = einsum(equation = var_33600_equation_0, values = (var_33082_cast_fp16, var_33484_cast_fp16))[name = string("op_33600_cast_fp16")]; + string var_33602_equation_0 = const()[name = string("op_33602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33602_cast_fp16 = einsum(equation = var_33602_equation_0, values = (var_33086_cast_fp16, var_33485_cast_fp16))[name = string("op_33602_cast_fp16")]; + string var_33604_equation_0 = const()[name = string("op_33604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33604_cast_fp16 = einsum(equation = var_33604_equation_0, values = (var_33086_cast_fp16, var_33486_cast_fp16))[name = string("op_33604_cast_fp16")]; + string var_33606_equation_0 = const()[name = string("op_33606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33606_cast_fp16 = einsum(equation = var_33606_equation_0, values = (var_33086_cast_fp16, var_33487_cast_fp16))[name = string("op_33606_cast_fp16")]; + string var_33608_equation_0 = const()[name = string("op_33608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33608_cast_fp16 = einsum(equation = var_33608_equation_0, values = (var_33086_cast_fp16, var_33488_cast_fp16))[name = string("op_33608_cast_fp16")]; + string var_33610_equation_0 = const()[name = string("op_33610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33610_cast_fp16 = einsum(equation = var_33610_equation_0, values = (var_33090_cast_fp16, var_33489_cast_fp16))[name = string("op_33610_cast_fp16")]; + string var_33612_equation_0 = const()[name = string("op_33612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33612_cast_fp16 = einsum(equation = var_33612_equation_0, values = (var_33090_cast_fp16, var_33490_cast_fp16))[name = string("op_33612_cast_fp16")]; + string var_33614_equation_0 = const()[name = string("op_33614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33614_cast_fp16 = einsum(equation = var_33614_equation_0, values = (var_33090_cast_fp16, var_33491_cast_fp16))[name = string("op_33614_cast_fp16")]; + string var_33616_equation_0 = const()[name = string("op_33616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33616_cast_fp16 = einsum(equation = var_33616_equation_0, values = (var_33090_cast_fp16, var_33492_cast_fp16))[name = string("op_33616_cast_fp16")]; + string var_33618_equation_0 = const()[name = string("op_33618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33618_cast_fp16 = einsum(equation = var_33618_equation_0, values = (var_33094_cast_fp16, var_33493_cast_fp16))[name = string("op_33618_cast_fp16")]; + string var_33620_equation_0 = const()[name = string("op_33620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33620_cast_fp16 = einsum(equation = var_33620_equation_0, values = (var_33094_cast_fp16, var_33494_cast_fp16))[name = string("op_33620_cast_fp16")]; + string var_33622_equation_0 = const()[name = string("op_33622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33622_cast_fp16 = einsum(equation = var_33622_equation_0, values = (var_33094_cast_fp16, var_33495_cast_fp16))[name = string("op_33622_cast_fp16")]; + string var_33624_equation_0 = const()[name = string("op_33624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33624_cast_fp16 = einsum(equation = var_33624_equation_0, values = (var_33094_cast_fp16, var_33496_cast_fp16))[name = string("op_33624_cast_fp16")]; + string var_33626_equation_0 = const()[name = string("op_33626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33626_cast_fp16 = einsum(equation = var_33626_equation_0, values = (var_33098_cast_fp16, var_33497_cast_fp16))[name = string("op_33626_cast_fp16")]; + string var_33628_equation_0 = const()[name = string("op_33628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33628_cast_fp16 = einsum(equation = var_33628_equation_0, values = (var_33098_cast_fp16, var_33498_cast_fp16))[name = string("op_33628_cast_fp16")]; + string var_33630_equation_0 = const()[name = string("op_33630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33630_cast_fp16 = einsum(equation = var_33630_equation_0, values = (var_33098_cast_fp16, var_33499_cast_fp16))[name = string("op_33630_cast_fp16")]; + string var_33632_equation_0 = const()[name = string("op_33632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33632_cast_fp16 = einsum(equation = var_33632_equation_0, values = (var_33098_cast_fp16, var_33500_cast_fp16))[name = string("op_33632_cast_fp16")]; + string var_33634_equation_0 = const()[name = string("op_33634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33634_cast_fp16 = einsum(equation = var_33634_equation_0, values = (var_33102_cast_fp16, var_33501_cast_fp16))[name = string("op_33634_cast_fp16")]; + string var_33636_equation_0 = const()[name = string("op_33636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33636_cast_fp16 = einsum(equation = var_33636_equation_0, values = (var_33102_cast_fp16, var_33502_cast_fp16))[name = string("op_33636_cast_fp16")]; + string var_33638_equation_0 = const()[name = string("op_33638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33638_cast_fp16 = einsum(equation = var_33638_equation_0, values = (var_33102_cast_fp16, var_33503_cast_fp16))[name = string("op_33638_cast_fp16")]; + string var_33640_equation_0 = const()[name = string("op_33640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33640_cast_fp16 = einsum(equation = var_33640_equation_0, values = (var_33102_cast_fp16, var_33504_cast_fp16))[name = string("op_33640_cast_fp16")]; + string var_33642_equation_0 = const()[name = string("op_33642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33642_cast_fp16 = einsum(equation = var_33642_equation_0, values = (var_33106_cast_fp16, var_33505_cast_fp16))[name = string("op_33642_cast_fp16")]; + string var_33644_equation_0 = const()[name = string("op_33644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33644_cast_fp16 = einsum(equation = var_33644_equation_0, values = (var_33106_cast_fp16, var_33506_cast_fp16))[name = string("op_33644_cast_fp16")]; + string var_33646_equation_0 = const()[name = string("op_33646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33646_cast_fp16 = einsum(equation = var_33646_equation_0, values = (var_33106_cast_fp16, var_33507_cast_fp16))[name = string("op_33646_cast_fp16")]; + string var_33648_equation_0 = const()[name = string("op_33648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33648_cast_fp16 = einsum(equation = var_33648_equation_0, values = (var_33106_cast_fp16, var_33508_cast_fp16))[name = string("op_33648_cast_fp16")]; + string var_33650_equation_0 = const()[name = string("op_33650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33650_cast_fp16 = einsum(equation = var_33650_equation_0, values = (var_33110_cast_fp16, var_33509_cast_fp16))[name = string("op_33650_cast_fp16")]; + string var_33652_equation_0 = const()[name = string("op_33652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33652_cast_fp16 = einsum(equation = var_33652_equation_0, values = (var_33110_cast_fp16, var_33510_cast_fp16))[name = string("op_33652_cast_fp16")]; + string var_33654_equation_0 = const()[name = string("op_33654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33654_cast_fp16 = einsum(equation = var_33654_equation_0, values = (var_33110_cast_fp16, var_33511_cast_fp16))[name = string("op_33654_cast_fp16")]; + string var_33656_equation_0 = const()[name = string("op_33656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33656_cast_fp16 = einsum(equation = var_33656_equation_0, values = (var_33110_cast_fp16, var_33512_cast_fp16))[name = string("op_33656_cast_fp16")]; + string var_33658_equation_0 = const()[name = string("op_33658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33658_cast_fp16 = einsum(equation = var_33658_equation_0, values = (var_33114_cast_fp16, var_33513_cast_fp16))[name = string("op_33658_cast_fp16")]; + string var_33660_equation_0 = const()[name = string("op_33660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33660_cast_fp16 = einsum(equation = var_33660_equation_0, values = (var_33114_cast_fp16, var_33514_cast_fp16))[name = string("op_33660_cast_fp16")]; + string var_33662_equation_0 = const()[name = string("op_33662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33662_cast_fp16 = einsum(equation = var_33662_equation_0, values = (var_33114_cast_fp16, var_33515_cast_fp16))[name = string("op_33662_cast_fp16")]; + string var_33664_equation_0 = const()[name = string("op_33664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33664_cast_fp16 = einsum(equation = var_33664_equation_0, values = (var_33114_cast_fp16, var_33516_cast_fp16))[name = string("op_33664_cast_fp16")]; + string var_33666_equation_0 = const()[name = string("op_33666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33666_cast_fp16 = einsum(equation = var_33666_equation_0, values = (var_33118_cast_fp16, var_33517_cast_fp16))[name = string("op_33666_cast_fp16")]; + string var_33668_equation_0 = const()[name = string("op_33668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33668_cast_fp16 = einsum(equation = var_33668_equation_0, values = (var_33118_cast_fp16, var_33518_cast_fp16))[name = string("op_33668_cast_fp16")]; + string var_33670_equation_0 = const()[name = string("op_33670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33670_cast_fp16 = einsum(equation = var_33670_equation_0, values = (var_33118_cast_fp16, var_33519_cast_fp16))[name = string("op_33670_cast_fp16")]; + string var_33672_equation_0 = const()[name = string("op_33672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33672_cast_fp16 = einsum(equation = var_33672_equation_0, values = (var_33118_cast_fp16, var_33520_cast_fp16))[name = string("op_33672_cast_fp16")]; + string var_33674_equation_0 = const()[name = string("op_33674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33674_cast_fp16 = einsum(equation = var_33674_equation_0, values = (var_33122_cast_fp16, var_33521_cast_fp16))[name = string("op_33674_cast_fp16")]; + string var_33676_equation_0 = const()[name = string("op_33676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33676_cast_fp16 = einsum(equation = var_33676_equation_0, values = (var_33122_cast_fp16, var_33522_cast_fp16))[name = string("op_33676_cast_fp16")]; + string var_33678_equation_0 = const()[name = string("op_33678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33678_cast_fp16 = einsum(equation = var_33678_equation_0, values = (var_33122_cast_fp16, var_33523_cast_fp16))[name = string("op_33678_cast_fp16")]; + string var_33680_equation_0 = const()[name = string("op_33680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33680_cast_fp16 = einsum(equation = var_33680_equation_0, values = (var_33122_cast_fp16, var_33524_cast_fp16))[name = string("op_33680_cast_fp16")]; + string var_33682_equation_0 = const()[name = string("op_33682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33682_cast_fp16 = einsum(equation = var_33682_equation_0, values = (var_33126_cast_fp16, var_33525_cast_fp16))[name = string("op_33682_cast_fp16")]; + string var_33684_equation_0 = const()[name = string("op_33684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33684_cast_fp16 = einsum(equation = var_33684_equation_0, values = (var_33126_cast_fp16, var_33526_cast_fp16))[name = string("op_33684_cast_fp16")]; + string var_33686_equation_0 = const()[name = string("op_33686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33686_cast_fp16 = einsum(equation = var_33686_equation_0, values = (var_33126_cast_fp16, var_33527_cast_fp16))[name = string("op_33686_cast_fp16")]; + string var_33688_equation_0 = const()[name = string("op_33688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_33688_cast_fp16 = einsum(equation = var_33688_equation_0, values = (var_33126_cast_fp16, var_33528_cast_fp16))[name = string("op_33688_cast_fp16")]; + bool var_33690_interleave_0 = const()[name = string("op_33690_interleave_0"), val = bool(false)]; + tensor var_33690_cast_fp16 = concat(axis = var_32249, interleave = var_33690_interleave_0, values = (var_33530_cast_fp16, var_33532_cast_fp16, var_33534_cast_fp16, var_33536_cast_fp16))[name = string("op_33690_cast_fp16")]; + bool var_33692_interleave_0 = const()[name = string("op_33692_interleave_0"), val = bool(false)]; + tensor var_33692_cast_fp16 = concat(axis = var_32249, interleave = var_33692_interleave_0, values = (var_33538_cast_fp16, var_33540_cast_fp16, var_33542_cast_fp16, var_33544_cast_fp16))[name = string("op_33692_cast_fp16")]; + bool var_33694_interleave_0 = const()[name = string("op_33694_interleave_0"), val = bool(false)]; + tensor var_33694_cast_fp16 = concat(axis = var_32249, interleave = var_33694_interleave_0, values = (var_33546_cast_fp16, var_33548_cast_fp16, var_33550_cast_fp16, var_33552_cast_fp16))[name = string("op_33694_cast_fp16")]; + bool var_33696_interleave_0 = const()[name = string("op_33696_interleave_0"), val = bool(false)]; + tensor var_33696_cast_fp16 = concat(axis = var_32249, interleave = var_33696_interleave_0, values = (var_33554_cast_fp16, var_33556_cast_fp16, var_33558_cast_fp16, var_33560_cast_fp16))[name = string("op_33696_cast_fp16")]; + bool var_33698_interleave_0 = const()[name = string("op_33698_interleave_0"), val = bool(false)]; + tensor var_33698_cast_fp16 = concat(axis = var_32249, interleave = var_33698_interleave_0, values = (var_33562_cast_fp16, var_33564_cast_fp16, var_33566_cast_fp16, var_33568_cast_fp16))[name = string("op_33698_cast_fp16")]; + bool var_33700_interleave_0 = const()[name = string("op_33700_interleave_0"), val = bool(false)]; + tensor var_33700_cast_fp16 = concat(axis = var_32249, interleave = var_33700_interleave_0, values = (var_33570_cast_fp16, var_33572_cast_fp16, var_33574_cast_fp16, var_33576_cast_fp16))[name = string("op_33700_cast_fp16")]; + bool var_33702_interleave_0 = const()[name = string("op_33702_interleave_0"), val = bool(false)]; + tensor var_33702_cast_fp16 = concat(axis = var_32249, interleave = var_33702_interleave_0, values = (var_33578_cast_fp16, var_33580_cast_fp16, var_33582_cast_fp16, var_33584_cast_fp16))[name = string("op_33702_cast_fp16")]; + bool var_33704_interleave_0 = const()[name = string("op_33704_interleave_0"), val = bool(false)]; + tensor var_33704_cast_fp16 = concat(axis = var_32249, interleave = var_33704_interleave_0, values = (var_33586_cast_fp16, var_33588_cast_fp16, var_33590_cast_fp16, var_33592_cast_fp16))[name = string("op_33704_cast_fp16")]; + bool var_33706_interleave_0 = const()[name = string("op_33706_interleave_0"), val = bool(false)]; + tensor var_33706_cast_fp16 = concat(axis = var_32249, interleave = var_33706_interleave_0, values = (var_33594_cast_fp16, var_33596_cast_fp16, var_33598_cast_fp16, var_33600_cast_fp16))[name = string("op_33706_cast_fp16")]; + bool var_33708_interleave_0 = const()[name = string("op_33708_interleave_0"), val = bool(false)]; + tensor var_33708_cast_fp16 = concat(axis = var_32249, interleave = var_33708_interleave_0, values = (var_33602_cast_fp16, var_33604_cast_fp16, var_33606_cast_fp16, var_33608_cast_fp16))[name = string("op_33708_cast_fp16")]; + bool var_33710_interleave_0 = const()[name = string("op_33710_interleave_0"), val = bool(false)]; + tensor var_33710_cast_fp16 = concat(axis = var_32249, interleave = var_33710_interleave_0, values = (var_33610_cast_fp16, var_33612_cast_fp16, var_33614_cast_fp16, var_33616_cast_fp16))[name = string("op_33710_cast_fp16")]; + bool var_33712_interleave_0 = const()[name = string("op_33712_interleave_0"), val = bool(false)]; + tensor var_33712_cast_fp16 = concat(axis = var_32249, interleave = var_33712_interleave_0, values = (var_33618_cast_fp16, var_33620_cast_fp16, var_33622_cast_fp16, var_33624_cast_fp16))[name = string("op_33712_cast_fp16")]; + bool var_33714_interleave_0 = const()[name = string("op_33714_interleave_0"), val = bool(false)]; + tensor var_33714_cast_fp16 = concat(axis = var_32249, interleave = var_33714_interleave_0, values = (var_33626_cast_fp16, var_33628_cast_fp16, var_33630_cast_fp16, var_33632_cast_fp16))[name = string("op_33714_cast_fp16")]; + bool var_33716_interleave_0 = const()[name = string("op_33716_interleave_0"), val = bool(false)]; + tensor var_33716_cast_fp16 = concat(axis = var_32249, interleave = var_33716_interleave_0, values = (var_33634_cast_fp16, var_33636_cast_fp16, var_33638_cast_fp16, var_33640_cast_fp16))[name = string("op_33716_cast_fp16")]; + bool var_33718_interleave_0 = const()[name = string("op_33718_interleave_0"), val = bool(false)]; + tensor var_33718_cast_fp16 = concat(axis = var_32249, interleave = var_33718_interleave_0, values = (var_33642_cast_fp16, var_33644_cast_fp16, var_33646_cast_fp16, var_33648_cast_fp16))[name = string("op_33718_cast_fp16")]; + bool var_33720_interleave_0 = const()[name = string("op_33720_interleave_0"), val = bool(false)]; + tensor var_33720_cast_fp16 = concat(axis = var_32249, interleave = var_33720_interleave_0, values = (var_33650_cast_fp16, var_33652_cast_fp16, var_33654_cast_fp16, var_33656_cast_fp16))[name = string("op_33720_cast_fp16")]; + bool var_33722_interleave_0 = const()[name = string("op_33722_interleave_0"), val = bool(false)]; + tensor var_33722_cast_fp16 = concat(axis = var_32249, interleave = var_33722_interleave_0, values = (var_33658_cast_fp16, var_33660_cast_fp16, var_33662_cast_fp16, var_33664_cast_fp16))[name = string("op_33722_cast_fp16")]; + bool var_33724_interleave_0 = const()[name = string("op_33724_interleave_0"), val = bool(false)]; + tensor var_33724_cast_fp16 = concat(axis = var_32249, interleave = var_33724_interleave_0, values = (var_33666_cast_fp16, var_33668_cast_fp16, var_33670_cast_fp16, var_33672_cast_fp16))[name = string("op_33724_cast_fp16")]; + bool var_33726_interleave_0 = const()[name = string("op_33726_interleave_0"), val = bool(false)]; + tensor var_33726_cast_fp16 = concat(axis = var_32249, interleave = var_33726_interleave_0, values = (var_33674_cast_fp16, var_33676_cast_fp16, var_33678_cast_fp16, var_33680_cast_fp16))[name = string("op_33726_cast_fp16")]; + bool var_33728_interleave_0 = const()[name = string("op_33728_interleave_0"), val = bool(false)]; + tensor var_33728_cast_fp16 = concat(axis = var_32249, interleave = var_33728_interleave_0, values = (var_33682_cast_fp16, var_33684_cast_fp16, var_33686_cast_fp16, var_33688_cast_fp16))[name = string("op_33728_cast_fp16")]; + bool input_169_interleave_0 = const()[name = string("input_169_interleave_0"), val = bool(false)]; + tensor input_169_cast_fp16 = concat(axis = var_32274, interleave = input_169_interleave_0, values = (var_33690_cast_fp16, var_33692_cast_fp16, var_33694_cast_fp16, var_33696_cast_fp16, var_33698_cast_fp16, var_33700_cast_fp16, var_33702_cast_fp16, var_33704_cast_fp16, var_33706_cast_fp16, var_33708_cast_fp16, var_33710_cast_fp16, var_33712_cast_fp16, var_33714_cast_fp16, var_33716_cast_fp16, var_33718_cast_fp16, var_33720_cast_fp16, var_33722_cast_fp16, var_33724_cast_fp16, var_33726_cast_fp16, var_33728_cast_fp16))[name = string("input_169_cast_fp16")]; + string obj_87_pad_type_0 = const()[name = string("obj_87_pad_type_0"), val = string("valid")]; + tensor obj_87_strides_0 = const()[name = string("obj_87_strides_0"), val = tensor([1, 1])]; + tensor obj_87_pad_0 = const()[name = string("obj_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_87_dilations_0 = const()[name = string("obj_87_dilations_0"), val = tensor([1, 1])]; + int32 obj_87_groups_0 = const()[name = string("obj_87_groups_0"), val = int32(1)]; + tensor layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(850924160)))]; + tensor layers_21_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854201024)))]; + tensor obj_87_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_bias_to_fp16, dilations = obj_87_dilations_0, groups = obj_87_groups_0, pad = obj_87_pad_0, pad_type = obj_87_pad_type_0, strides = obj_87_strides_0, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_169_cast_fp16)[name = string("obj_87_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_87_cast_fp16)[name = string("inputs_87_cast_fp16")]; + tensor out_87_axes_0 = const()[name = string("out_87_axes_0"), val = tensor([1])]; + fp16 var_33747_to_fp16 = const()[name = string("op_33747_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_33747_to_fp16, x = inputs_87_cast_fp16)[name = string("out_87_cast_fp16")]; + tensor input_171_gamma_0_to_fp16 = const()[name = string("input_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854203648)))]; + tensor input_171_beta_0_to_fp16 = const()[name = string("input_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854206272)))]; + fp16 input_171_epsilon_0_to_fp16 = const()[name = string("input_171_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_171_cast_fp16 = batch_norm(beta = input_171_beta_0_to_fp16, epsilon = input_171_epsilon_0_to_fp16, gamma = input_171_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = string("input_171_cast_fp16")]; + string input_173_pad_type_0 = const()[name = string("input_173_pad_type_0"), val = string("valid")]; + tensor input_173_strides_0 = const()[name = string("input_173_strides_0"), val = tensor([1, 1])]; + tensor input_173_pad_0 = const()[name = string("input_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_173_dilations_0 = const()[name = string("input_173_dilations_0"), val = tensor([1, 1])]; + int32 input_173_groups_0 = const()[name = string("input_173_groups_0"), val = int32(1)]; + tensor layers_21_fc1_weight_to_fp16 = const()[name = string("layers_21_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(854208896)))]; + tensor layers_21_fc1_bias_to_fp16 = const()[name = string("layers_21_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867316160)))]; + tensor input_173_cast_fp16 = conv(bias = layers_21_fc1_bias_to_fp16, dilations = input_173_dilations_0, groups = input_173_groups_0, pad = input_173_pad_0, pad_type = input_173_pad_type_0, strides = input_173_strides_0, weight = layers_21_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = string("input_173_cast_fp16")]; + string input_175_mode_0 = const()[name = string("input_175_mode_0"), val = string("EXACT")]; + tensor input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")]; + string hidden_states_47_pad_type_0 = const()[name = string("hidden_states_47_pad_type_0"), val = string("valid")]; + tensor hidden_states_47_strides_0 = const()[name = string("hidden_states_47_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_47_pad_0 = const()[name = string("hidden_states_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_47_dilations_0 = const()[name = string("hidden_states_47_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_47_groups_0 = const()[name = string("hidden_states_47_groups_0"), val = int32(1)]; + tensor layers_21_fc2_weight_to_fp16 = const()[name = string("layers_21_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(867326464)))]; + tensor layers_21_fc2_bias_to_fp16 = const()[name = string("layers_21_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880433728)))]; + tensor hidden_states_47_cast_fp16 = conv(bias = layers_21_fc2_bias_to_fp16, dilations = hidden_states_47_dilations_0, groups = hidden_states_47_groups_0, pad = hidden_states_47_pad_0, pad_type = hidden_states_47_pad_type_0, strides = hidden_states_47_strides_0, weight = layers_21_fc2_weight_to_fp16, x = input_175_cast_fp16)[name = string("hidden_states_47_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = hidden_states_47_cast_fp16)[name = string("inputs_89_cast_fp16")]; + int32 var_33776 = const()[name = string("op_33776"), val = int32(3)]; + int32 var_33801 = const()[name = string("op_33801"), val = int32(1)]; + tensor out_89_axes_0 = const()[name = string("out_89_axes_0"), val = tensor([1])]; + fp16 var_33818_to_fp16 = const()[name = string("op_33818_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_33818_to_fp16, x = inputs_89_cast_fp16)[name = string("out_89_cast_fp16")]; + tensor obj_89_gamma_0_to_fp16 = const()[name = string("obj_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880436352)))]; + tensor obj_89_beta_0_to_fp16 = const()[name = string("obj_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880438976)))]; + fp16 obj_89_epsilon_0_to_fp16 = const()[name = string("obj_89_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_89_cast_fp16 = batch_norm(beta = obj_89_beta_0_to_fp16, epsilon = obj_89_epsilon_0_to_fp16, gamma = obj_89_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = string("obj_89_cast_fp16")]; + string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; + tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; + tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; + int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(880441600)))]; + tensor layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883718464)))]; + tensor query_45_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("query_45_cast_fp16")]; + string key_45_pad_type_0 = const()[name = string("key_45_pad_type_0"), val = string("valid")]; + tensor key_45_strides_0 = const()[name = string("key_45_strides_0"), val = tensor([1, 1])]; + tensor key_45_pad_0 = const()[name = string("key_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_45_dilations_0 = const()[name = string("key_45_dilations_0"), val = tensor([1, 1])]; + int32 key_45_groups_0 = const()[name = string("key_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883721088)))]; + tensor key_45_cast_fp16 = conv(dilations = key_45_dilations_0, groups = key_45_groups_0, pad = key_45_pad_0, pad_type = key_45_pad_type_0, strides = key_45_strides_0, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("key_45_cast_fp16")]; + string value_45_pad_type_0 = const()[name = string("value_45_pad_type_0"), val = string("valid")]; + tensor value_45_strides_0 = const()[name = string("value_45_strides_0"), val = tensor([1, 1])]; + tensor value_45_pad_0 = const()[name = string("value_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_45_dilations_0 = const()[name = string("value_45_dilations_0"), val = tensor([1, 1])]; + int32 value_45_groups_0 = const()[name = string("value_45_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(886997952)))]; + tensor layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890274816)))]; + tensor value_45_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_bias_to_fp16, dilations = value_45_dilations_0, groups = value_45_groups_0, pad = value_45_pad_0, pad_type = value_45_pad_type_0, strides = value_45_strides_0, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = obj_89_cast_fp16)[name = string("value_45_cast_fp16")]; + tensor var_33856_begin_0 = const()[name = string("op_33856_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33856_end_0 = const()[name = string("op_33856_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33856_end_mask_0 = const()[name = string("op_33856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33856_cast_fp16 = slice_by_index(begin = var_33856_begin_0, end = var_33856_end_0, end_mask = var_33856_end_mask_0, x = query_45_cast_fp16)[name = string("op_33856_cast_fp16")]; + tensor var_33860_begin_0 = const()[name = string("op_33860_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33860_end_0 = const()[name = string("op_33860_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_33860_end_mask_0 = const()[name = string("op_33860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33860_cast_fp16 = slice_by_index(begin = var_33860_begin_0, end = var_33860_end_0, end_mask = var_33860_end_mask_0, x = query_45_cast_fp16)[name = string("op_33860_cast_fp16")]; + tensor var_33864_begin_0 = const()[name = string("op_33864_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33864_end_0 = const()[name = string("op_33864_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_33864_end_mask_0 = const()[name = string("op_33864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33864_cast_fp16 = slice_by_index(begin = var_33864_begin_0, end = var_33864_end_0, end_mask = var_33864_end_mask_0, x = query_45_cast_fp16)[name = string("op_33864_cast_fp16")]; + tensor var_33868_begin_0 = const()[name = string("op_33868_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33868_end_0 = const()[name = string("op_33868_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_33868_end_mask_0 = const()[name = string("op_33868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33868_cast_fp16 = slice_by_index(begin = var_33868_begin_0, end = var_33868_end_0, end_mask = var_33868_end_mask_0, x = query_45_cast_fp16)[name = string("op_33868_cast_fp16")]; + tensor var_33872_begin_0 = const()[name = string("op_33872_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33872_end_0 = const()[name = string("op_33872_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_33872_end_mask_0 = const()[name = string("op_33872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33872_cast_fp16 = slice_by_index(begin = var_33872_begin_0, end = var_33872_end_0, end_mask = var_33872_end_mask_0, x = query_45_cast_fp16)[name = string("op_33872_cast_fp16")]; + tensor var_33876_begin_0 = const()[name = string("op_33876_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33876_end_0 = const()[name = string("op_33876_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_33876_end_mask_0 = const()[name = string("op_33876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33876_cast_fp16 = slice_by_index(begin = var_33876_begin_0, end = var_33876_end_0, end_mask = var_33876_end_mask_0, x = query_45_cast_fp16)[name = string("op_33876_cast_fp16")]; + tensor var_33880_begin_0 = const()[name = string("op_33880_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33880_end_0 = const()[name = string("op_33880_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_33880_end_mask_0 = const()[name = string("op_33880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33880_cast_fp16 = slice_by_index(begin = var_33880_begin_0, end = var_33880_end_0, end_mask = var_33880_end_mask_0, x = query_45_cast_fp16)[name = string("op_33880_cast_fp16")]; + tensor var_33884_begin_0 = const()[name = string("op_33884_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33884_end_0 = const()[name = string("op_33884_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_33884_end_mask_0 = const()[name = string("op_33884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33884_cast_fp16 = slice_by_index(begin = var_33884_begin_0, end = var_33884_end_0, end_mask = var_33884_end_mask_0, x = query_45_cast_fp16)[name = string("op_33884_cast_fp16")]; + tensor var_33888_begin_0 = const()[name = string("op_33888_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33888_end_0 = const()[name = string("op_33888_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_33888_end_mask_0 = const()[name = string("op_33888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33888_cast_fp16 = slice_by_index(begin = var_33888_begin_0, end = var_33888_end_0, end_mask = var_33888_end_mask_0, x = query_45_cast_fp16)[name = string("op_33888_cast_fp16")]; + tensor var_33892_begin_0 = const()[name = string("op_33892_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33892_end_0 = const()[name = string("op_33892_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_33892_end_mask_0 = const()[name = string("op_33892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33892_cast_fp16 = slice_by_index(begin = var_33892_begin_0, end = var_33892_end_0, end_mask = var_33892_end_mask_0, x = query_45_cast_fp16)[name = string("op_33892_cast_fp16")]; + tensor var_33896_begin_0 = const()[name = string("op_33896_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33896_end_0 = const()[name = string("op_33896_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_33896_end_mask_0 = const()[name = string("op_33896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33896_cast_fp16 = slice_by_index(begin = var_33896_begin_0, end = var_33896_end_0, end_mask = var_33896_end_mask_0, x = query_45_cast_fp16)[name = string("op_33896_cast_fp16")]; + tensor var_33900_begin_0 = const()[name = string("op_33900_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33900_end_0 = const()[name = string("op_33900_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_33900_end_mask_0 = const()[name = string("op_33900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33900_cast_fp16 = slice_by_index(begin = var_33900_begin_0, end = var_33900_end_0, end_mask = var_33900_end_mask_0, x = query_45_cast_fp16)[name = string("op_33900_cast_fp16")]; + tensor var_33904_begin_0 = const()[name = string("op_33904_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33904_end_0 = const()[name = string("op_33904_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_33904_end_mask_0 = const()[name = string("op_33904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33904_cast_fp16 = slice_by_index(begin = var_33904_begin_0, end = var_33904_end_0, end_mask = var_33904_end_mask_0, x = query_45_cast_fp16)[name = string("op_33904_cast_fp16")]; + tensor var_33908_begin_0 = const()[name = string("op_33908_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33908_end_0 = const()[name = string("op_33908_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_33908_end_mask_0 = const()[name = string("op_33908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33908_cast_fp16 = slice_by_index(begin = var_33908_begin_0, end = var_33908_end_0, end_mask = var_33908_end_mask_0, x = query_45_cast_fp16)[name = string("op_33908_cast_fp16")]; + tensor var_33912_begin_0 = const()[name = string("op_33912_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33912_end_0 = const()[name = string("op_33912_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_33912_end_mask_0 = const()[name = string("op_33912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33912_cast_fp16 = slice_by_index(begin = var_33912_begin_0, end = var_33912_end_0, end_mask = var_33912_end_mask_0, x = query_45_cast_fp16)[name = string("op_33912_cast_fp16")]; + tensor var_33916_begin_0 = const()[name = string("op_33916_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33916_end_0 = const()[name = string("op_33916_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_33916_end_mask_0 = const()[name = string("op_33916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33916_cast_fp16 = slice_by_index(begin = var_33916_begin_0, end = var_33916_end_0, end_mask = var_33916_end_mask_0, x = query_45_cast_fp16)[name = string("op_33916_cast_fp16")]; + tensor var_33920_begin_0 = const()[name = string("op_33920_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33920_end_0 = const()[name = string("op_33920_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_33920_end_mask_0 = const()[name = string("op_33920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33920_cast_fp16 = slice_by_index(begin = var_33920_begin_0, end = var_33920_end_0, end_mask = var_33920_end_mask_0, x = query_45_cast_fp16)[name = string("op_33920_cast_fp16")]; + tensor var_33924_begin_0 = const()[name = string("op_33924_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33924_end_0 = const()[name = string("op_33924_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_33924_end_mask_0 = const()[name = string("op_33924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33924_cast_fp16 = slice_by_index(begin = var_33924_begin_0, end = var_33924_end_0, end_mask = var_33924_end_mask_0, x = query_45_cast_fp16)[name = string("op_33924_cast_fp16")]; + tensor var_33928_begin_0 = const()[name = string("op_33928_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33928_end_0 = const()[name = string("op_33928_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_33928_end_mask_0 = const()[name = string("op_33928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33928_cast_fp16 = slice_by_index(begin = var_33928_begin_0, end = var_33928_end_0, end_mask = var_33928_end_mask_0, x = query_45_cast_fp16)[name = string("op_33928_cast_fp16")]; + tensor var_33932_begin_0 = const()[name = string("op_33932_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33932_end_0 = const()[name = string("op_33932_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_33932_end_mask_0 = const()[name = string("op_33932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33932_cast_fp16 = slice_by_index(begin = var_33932_begin_0, end = var_33932_end_0, end_mask = var_33932_end_mask_0, x = query_45_cast_fp16)[name = string("op_33932_cast_fp16")]; + tensor var_33941_begin_0 = const()[name = string("op_33941_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33941_end_0 = const()[name = string("op_33941_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33941_end_mask_0 = const()[name = string("op_33941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33941_cast_fp16 = slice_by_index(begin = var_33941_begin_0, end = var_33941_end_0, end_mask = var_33941_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33941_cast_fp16")]; + tensor var_33948_begin_0 = const()[name = string("op_33948_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33948_end_0 = const()[name = string("op_33948_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33948_end_mask_0 = const()[name = string("op_33948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33948_cast_fp16 = slice_by_index(begin = var_33948_begin_0, end = var_33948_end_0, end_mask = var_33948_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33948_cast_fp16")]; + tensor var_33955_begin_0 = const()[name = string("op_33955_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33955_end_0 = const()[name = string("op_33955_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33955_end_mask_0 = const()[name = string("op_33955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33955_cast_fp16 = slice_by_index(begin = var_33955_begin_0, end = var_33955_end_0, end_mask = var_33955_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33955_cast_fp16")]; + tensor var_33962_begin_0 = const()[name = string("op_33962_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33962_end_0 = const()[name = string("op_33962_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33962_end_mask_0 = const()[name = string("op_33962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33962_cast_fp16 = slice_by_index(begin = var_33962_begin_0, end = var_33962_end_0, end_mask = var_33962_end_mask_0, x = var_33856_cast_fp16)[name = string("op_33962_cast_fp16")]; + tensor var_33969_begin_0 = const()[name = string("op_33969_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33969_end_0 = const()[name = string("op_33969_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33969_end_mask_0 = const()[name = string("op_33969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33969_cast_fp16 = slice_by_index(begin = var_33969_begin_0, end = var_33969_end_0, end_mask = var_33969_end_mask_0, x = var_33860_cast_fp16)[name = string("op_33969_cast_fp16")]; + tensor var_33976_begin_0 = const()[name = string("op_33976_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33976_end_0 = const()[name = string("op_33976_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33976_end_mask_0 = const()[name = string("op_33976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33976_cast_fp16 = slice_by_index(begin = var_33976_begin_0, end = var_33976_end_0, end_mask = var_33976_end_mask_0, x = var_33860_cast_fp16)[name = string("op_33976_cast_fp16")]; + tensor var_33983_begin_0 = const()[name = string("op_33983_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33983_end_0 = const()[name = string("op_33983_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33983_end_mask_0 = const()[name = string("op_33983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33983_cast_fp16 = slice_by_index(begin = var_33983_begin_0, end = var_33983_end_0, end_mask = var_33983_end_mask_0, x = var_33860_cast_fp16)[name = string("op_33983_cast_fp16")]; + tensor var_33990_begin_0 = const()[name = string("op_33990_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33990_end_0 = const()[name = string("op_33990_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33990_end_mask_0 = const()[name = string("op_33990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33990_cast_fp16 = slice_by_index(begin = var_33990_begin_0, end = var_33990_end_0, end_mask = var_33990_end_mask_0, x = var_33860_cast_fp16)[name = string("op_33990_cast_fp16")]; + tensor var_33997_begin_0 = const()[name = string("op_33997_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33997_end_0 = const()[name = string("op_33997_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33997_end_mask_0 = const()[name = string("op_33997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33997_cast_fp16 = slice_by_index(begin = var_33997_begin_0, end = var_33997_end_0, end_mask = var_33997_end_mask_0, x = var_33864_cast_fp16)[name = string("op_33997_cast_fp16")]; + tensor var_34004_begin_0 = const()[name = string("op_34004_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34004_end_0 = const()[name = string("op_34004_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34004_end_mask_0 = const()[name = string("op_34004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34004_cast_fp16 = slice_by_index(begin = var_34004_begin_0, end = var_34004_end_0, end_mask = var_34004_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34004_cast_fp16")]; + tensor var_34011_begin_0 = const()[name = string("op_34011_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34011_end_0 = const()[name = string("op_34011_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34011_end_mask_0 = const()[name = string("op_34011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34011_cast_fp16 = slice_by_index(begin = var_34011_begin_0, end = var_34011_end_0, end_mask = var_34011_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34011_cast_fp16")]; + tensor var_34018_begin_0 = const()[name = string("op_34018_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34018_end_0 = const()[name = string("op_34018_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34018_end_mask_0 = const()[name = string("op_34018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34018_cast_fp16 = slice_by_index(begin = var_34018_begin_0, end = var_34018_end_0, end_mask = var_34018_end_mask_0, x = var_33864_cast_fp16)[name = string("op_34018_cast_fp16")]; + tensor var_34025_begin_0 = const()[name = string("op_34025_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34025_end_0 = const()[name = string("op_34025_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34025_end_mask_0 = const()[name = string("op_34025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34025_cast_fp16 = slice_by_index(begin = var_34025_begin_0, end = var_34025_end_0, end_mask = var_34025_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34025_cast_fp16")]; + tensor var_34032_begin_0 = const()[name = string("op_34032_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34032_end_0 = const()[name = string("op_34032_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34032_end_mask_0 = const()[name = string("op_34032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34032_cast_fp16 = slice_by_index(begin = var_34032_begin_0, end = var_34032_end_0, end_mask = var_34032_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34032_cast_fp16")]; + tensor var_34039_begin_0 = const()[name = string("op_34039_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34039_end_0 = const()[name = string("op_34039_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34039_end_mask_0 = const()[name = string("op_34039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34039_cast_fp16 = slice_by_index(begin = var_34039_begin_0, end = var_34039_end_0, end_mask = var_34039_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34039_cast_fp16")]; + tensor var_34046_begin_0 = const()[name = string("op_34046_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34046_end_0 = const()[name = string("op_34046_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34046_end_mask_0 = const()[name = string("op_34046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34046_cast_fp16 = slice_by_index(begin = var_34046_begin_0, end = var_34046_end_0, end_mask = var_34046_end_mask_0, x = var_33868_cast_fp16)[name = string("op_34046_cast_fp16")]; + tensor var_34053_begin_0 = const()[name = string("op_34053_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34053_end_0 = const()[name = string("op_34053_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34053_end_mask_0 = const()[name = string("op_34053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34053_cast_fp16 = slice_by_index(begin = var_34053_begin_0, end = var_34053_end_0, end_mask = var_34053_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34053_cast_fp16")]; + tensor var_34060_begin_0 = const()[name = string("op_34060_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34060_end_0 = const()[name = string("op_34060_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34060_end_mask_0 = const()[name = string("op_34060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34060_cast_fp16 = slice_by_index(begin = var_34060_begin_0, end = var_34060_end_0, end_mask = var_34060_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34060_cast_fp16")]; + tensor var_34067_begin_0 = const()[name = string("op_34067_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34067_end_0 = const()[name = string("op_34067_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34067_end_mask_0 = const()[name = string("op_34067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34067_cast_fp16 = slice_by_index(begin = var_34067_begin_0, end = var_34067_end_0, end_mask = var_34067_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34067_cast_fp16")]; + tensor var_34074_begin_0 = const()[name = string("op_34074_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34074_end_0 = const()[name = string("op_34074_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34074_end_mask_0 = const()[name = string("op_34074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34074_cast_fp16 = slice_by_index(begin = var_34074_begin_0, end = var_34074_end_0, end_mask = var_34074_end_mask_0, x = var_33872_cast_fp16)[name = string("op_34074_cast_fp16")]; + tensor var_34081_begin_0 = const()[name = string("op_34081_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34081_end_0 = const()[name = string("op_34081_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34081_end_mask_0 = const()[name = string("op_34081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34081_cast_fp16 = slice_by_index(begin = var_34081_begin_0, end = var_34081_end_0, end_mask = var_34081_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34081_cast_fp16")]; + tensor var_34088_begin_0 = const()[name = string("op_34088_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34088_end_0 = const()[name = string("op_34088_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34088_end_mask_0 = const()[name = string("op_34088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34088_cast_fp16 = slice_by_index(begin = var_34088_begin_0, end = var_34088_end_0, end_mask = var_34088_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34088_cast_fp16")]; + tensor var_34095_begin_0 = const()[name = string("op_34095_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34095_end_0 = const()[name = string("op_34095_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34095_end_mask_0 = const()[name = string("op_34095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34095_cast_fp16 = slice_by_index(begin = var_34095_begin_0, end = var_34095_end_0, end_mask = var_34095_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34095_cast_fp16")]; + tensor var_34102_begin_0 = const()[name = string("op_34102_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34102_end_0 = const()[name = string("op_34102_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34102_end_mask_0 = const()[name = string("op_34102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34102_cast_fp16 = slice_by_index(begin = var_34102_begin_0, end = var_34102_end_0, end_mask = var_34102_end_mask_0, x = var_33876_cast_fp16)[name = string("op_34102_cast_fp16")]; + tensor var_34109_begin_0 = const()[name = string("op_34109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34109_end_0 = const()[name = string("op_34109_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34109_end_mask_0 = const()[name = string("op_34109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34109_cast_fp16 = slice_by_index(begin = var_34109_begin_0, end = var_34109_end_0, end_mask = var_34109_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34109_cast_fp16")]; + tensor var_34116_begin_0 = const()[name = string("op_34116_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34116_end_0 = const()[name = string("op_34116_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34116_end_mask_0 = const()[name = string("op_34116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34116_cast_fp16 = slice_by_index(begin = var_34116_begin_0, end = var_34116_end_0, end_mask = var_34116_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34116_cast_fp16")]; + tensor var_34123_begin_0 = const()[name = string("op_34123_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34123_end_0 = const()[name = string("op_34123_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34123_end_mask_0 = const()[name = string("op_34123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34123_cast_fp16 = slice_by_index(begin = var_34123_begin_0, end = var_34123_end_0, end_mask = var_34123_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34123_cast_fp16")]; + tensor var_34130_begin_0 = const()[name = string("op_34130_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34130_end_0 = const()[name = string("op_34130_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34130_end_mask_0 = const()[name = string("op_34130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34130_cast_fp16 = slice_by_index(begin = var_34130_begin_0, end = var_34130_end_0, end_mask = var_34130_end_mask_0, x = var_33880_cast_fp16)[name = string("op_34130_cast_fp16")]; + tensor var_34137_begin_0 = const()[name = string("op_34137_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34137_end_0 = const()[name = string("op_34137_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34137_end_mask_0 = const()[name = string("op_34137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34137_cast_fp16 = slice_by_index(begin = var_34137_begin_0, end = var_34137_end_0, end_mask = var_34137_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34137_cast_fp16")]; + tensor var_34144_begin_0 = const()[name = string("op_34144_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34144_end_0 = const()[name = string("op_34144_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34144_end_mask_0 = const()[name = string("op_34144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34144_cast_fp16 = slice_by_index(begin = var_34144_begin_0, end = var_34144_end_0, end_mask = var_34144_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34144_cast_fp16")]; + tensor var_34151_begin_0 = const()[name = string("op_34151_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34151_end_0 = const()[name = string("op_34151_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34151_end_mask_0 = const()[name = string("op_34151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34151_cast_fp16 = slice_by_index(begin = var_34151_begin_0, end = var_34151_end_0, end_mask = var_34151_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34151_cast_fp16")]; + tensor var_34158_begin_0 = const()[name = string("op_34158_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34158_end_0 = const()[name = string("op_34158_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34158_end_mask_0 = const()[name = string("op_34158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34158_cast_fp16 = slice_by_index(begin = var_34158_begin_0, end = var_34158_end_0, end_mask = var_34158_end_mask_0, x = var_33884_cast_fp16)[name = string("op_34158_cast_fp16")]; + tensor var_34165_begin_0 = const()[name = string("op_34165_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34165_end_0 = const()[name = string("op_34165_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34165_end_mask_0 = const()[name = string("op_34165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34165_cast_fp16 = slice_by_index(begin = var_34165_begin_0, end = var_34165_end_0, end_mask = var_34165_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34165_cast_fp16")]; + tensor var_34172_begin_0 = const()[name = string("op_34172_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34172_end_0 = const()[name = string("op_34172_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34172_end_mask_0 = const()[name = string("op_34172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34172_cast_fp16 = slice_by_index(begin = var_34172_begin_0, end = var_34172_end_0, end_mask = var_34172_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34172_cast_fp16")]; + tensor var_34179_begin_0 = const()[name = string("op_34179_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34179_end_0 = const()[name = string("op_34179_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34179_end_mask_0 = const()[name = string("op_34179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34179_cast_fp16 = slice_by_index(begin = var_34179_begin_0, end = var_34179_end_0, end_mask = var_34179_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34179_cast_fp16")]; + tensor var_34186_begin_0 = const()[name = string("op_34186_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34186_end_0 = const()[name = string("op_34186_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34186_end_mask_0 = const()[name = string("op_34186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34186_cast_fp16 = slice_by_index(begin = var_34186_begin_0, end = var_34186_end_0, end_mask = var_34186_end_mask_0, x = var_33888_cast_fp16)[name = string("op_34186_cast_fp16")]; + tensor var_34193_begin_0 = const()[name = string("op_34193_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34193_end_0 = const()[name = string("op_34193_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34193_end_mask_0 = const()[name = string("op_34193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34193_cast_fp16 = slice_by_index(begin = var_34193_begin_0, end = var_34193_end_0, end_mask = var_34193_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34193_cast_fp16")]; + tensor var_34200_begin_0 = const()[name = string("op_34200_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34200_end_0 = const()[name = string("op_34200_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34200_end_mask_0 = const()[name = string("op_34200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34200_cast_fp16 = slice_by_index(begin = var_34200_begin_0, end = var_34200_end_0, end_mask = var_34200_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34200_cast_fp16")]; + tensor var_34207_begin_0 = const()[name = string("op_34207_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34207_end_0 = const()[name = string("op_34207_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34207_end_mask_0 = const()[name = string("op_34207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34207_cast_fp16 = slice_by_index(begin = var_34207_begin_0, end = var_34207_end_0, end_mask = var_34207_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34207_cast_fp16")]; + tensor var_34214_begin_0 = const()[name = string("op_34214_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34214_end_0 = const()[name = string("op_34214_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34214_end_mask_0 = const()[name = string("op_34214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34214_cast_fp16 = slice_by_index(begin = var_34214_begin_0, end = var_34214_end_0, end_mask = var_34214_end_mask_0, x = var_33892_cast_fp16)[name = string("op_34214_cast_fp16")]; + tensor var_34221_begin_0 = const()[name = string("op_34221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34221_end_0 = const()[name = string("op_34221_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34221_end_mask_0 = const()[name = string("op_34221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34221_cast_fp16 = slice_by_index(begin = var_34221_begin_0, end = var_34221_end_0, end_mask = var_34221_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34221_cast_fp16")]; + tensor var_34228_begin_0 = const()[name = string("op_34228_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34228_end_0 = const()[name = string("op_34228_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34228_end_mask_0 = const()[name = string("op_34228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34228_cast_fp16 = slice_by_index(begin = var_34228_begin_0, end = var_34228_end_0, end_mask = var_34228_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34228_cast_fp16")]; + tensor var_34235_begin_0 = const()[name = string("op_34235_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34235_end_0 = const()[name = string("op_34235_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34235_end_mask_0 = const()[name = string("op_34235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34235_cast_fp16 = slice_by_index(begin = var_34235_begin_0, end = var_34235_end_0, end_mask = var_34235_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34235_cast_fp16")]; + tensor var_34242_begin_0 = const()[name = string("op_34242_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34242_end_0 = const()[name = string("op_34242_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34242_end_mask_0 = const()[name = string("op_34242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34242_cast_fp16 = slice_by_index(begin = var_34242_begin_0, end = var_34242_end_0, end_mask = var_34242_end_mask_0, x = var_33896_cast_fp16)[name = string("op_34242_cast_fp16")]; + tensor var_34249_begin_0 = const()[name = string("op_34249_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34249_end_0 = const()[name = string("op_34249_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34249_end_mask_0 = const()[name = string("op_34249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34249_cast_fp16 = slice_by_index(begin = var_34249_begin_0, end = var_34249_end_0, end_mask = var_34249_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34249_cast_fp16")]; + tensor var_34256_begin_0 = const()[name = string("op_34256_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34256_end_0 = const()[name = string("op_34256_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34256_end_mask_0 = const()[name = string("op_34256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34256_cast_fp16 = slice_by_index(begin = var_34256_begin_0, end = var_34256_end_0, end_mask = var_34256_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34256_cast_fp16")]; + tensor var_34263_begin_0 = const()[name = string("op_34263_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34263_end_0 = const()[name = string("op_34263_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34263_end_mask_0 = const()[name = string("op_34263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34263_cast_fp16 = slice_by_index(begin = var_34263_begin_0, end = var_34263_end_0, end_mask = var_34263_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34263_cast_fp16")]; + tensor var_34270_begin_0 = const()[name = string("op_34270_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34270_end_0 = const()[name = string("op_34270_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34270_end_mask_0 = const()[name = string("op_34270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34270_cast_fp16 = slice_by_index(begin = var_34270_begin_0, end = var_34270_end_0, end_mask = var_34270_end_mask_0, x = var_33900_cast_fp16)[name = string("op_34270_cast_fp16")]; + tensor var_34277_begin_0 = const()[name = string("op_34277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34277_end_0 = const()[name = string("op_34277_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34277_end_mask_0 = const()[name = string("op_34277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34277_cast_fp16 = slice_by_index(begin = var_34277_begin_0, end = var_34277_end_0, end_mask = var_34277_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34277_cast_fp16")]; + tensor var_34284_begin_0 = const()[name = string("op_34284_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34284_end_0 = const()[name = string("op_34284_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34284_end_mask_0 = const()[name = string("op_34284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34284_cast_fp16 = slice_by_index(begin = var_34284_begin_0, end = var_34284_end_0, end_mask = var_34284_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34284_cast_fp16")]; + tensor var_34291_begin_0 = const()[name = string("op_34291_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34291_end_0 = const()[name = string("op_34291_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34291_end_mask_0 = const()[name = string("op_34291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34291_cast_fp16 = slice_by_index(begin = var_34291_begin_0, end = var_34291_end_0, end_mask = var_34291_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34291_cast_fp16")]; + tensor var_34298_begin_0 = const()[name = string("op_34298_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34298_end_0 = const()[name = string("op_34298_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34298_end_mask_0 = const()[name = string("op_34298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34298_cast_fp16 = slice_by_index(begin = var_34298_begin_0, end = var_34298_end_0, end_mask = var_34298_end_mask_0, x = var_33904_cast_fp16)[name = string("op_34298_cast_fp16")]; + tensor var_34305_begin_0 = const()[name = string("op_34305_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34305_end_0 = const()[name = string("op_34305_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34305_end_mask_0 = const()[name = string("op_34305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34305_cast_fp16 = slice_by_index(begin = var_34305_begin_0, end = var_34305_end_0, end_mask = var_34305_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34305_cast_fp16")]; + tensor var_34312_begin_0 = const()[name = string("op_34312_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34312_end_0 = const()[name = string("op_34312_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34312_end_mask_0 = const()[name = string("op_34312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34312_cast_fp16 = slice_by_index(begin = var_34312_begin_0, end = var_34312_end_0, end_mask = var_34312_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34312_cast_fp16")]; + tensor var_34319_begin_0 = const()[name = string("op_34319_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34319_end_0 = const()[name = string("op_34319_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34319_end_mask_0 = const()[name = string("op_34319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34319_cast_fp16 = slice_by_index(begin = var_34319_begin_0, end = var_34319_end_0, end_mask = var_34319_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34319_cast_fp16")]; + tensor var_34326_begin_0 = const()[name = string("op_34326_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34326_end_0 = const()[name = string("op_34326_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34326_end_mask_0 = const()[name = string("op_34326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34326_cast_fp16 = slice_by_index(begin = var_34326_begin_0, end = var_34326_end_0, end_mask = var_34326_end_mask_0, x = var_33908_cast_fp16)[name = string("op_34326_cast_fp16")]; + tensor var_34333_begin_0 = const()[name = string("op_34333_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34333_end_0 = const()[name = string("op_34333_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34333_end_mask_0 = const()[name = string("op_34333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34333_cast_fp16 = slice_by_index(begin = var_34333_begin_0, end = var_34333_end_0, end_mask = var_34333_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34333_cast_fp16")]; + tensor var_34340_begin_0 = const()[name = string("op_34340_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34340_end_0 = const()[name = string("op_34340_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34340_end_mask_0 = const()[name = string("op_34340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34340_cast_fp16 = slice_by_index(begin = var_34340_begin_0, end = var_34340_end_0, end_mask = var_34340_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34340_cast_fp16")]; + tensor var_34347_begin_0 = const()[name = string("op_34347_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34347_end_0 = const()[name = string("op_34347_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34347_end_mask_0 = const()[name = string("op_34347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34347_cast_fp16 = slice_by_index(begin = var_34347_begin_0, end = var_34347_end_0, end_mask = var_34347_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34347_cast_fp16")]; + tensor var_34354_begin_0 = const()[name = string("op_34354_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34354_end_0 = const()[name = string("op_34354_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34354_end_mask_0 = const()[name = string("op_34354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34354_cast_fp16 = slice_by_index(begin = var_34354_begin_0, end = var_34354_end_0, end_mask = var_34354_end_mask_0, x = var_33912_cast_fp16)[name = string("op_34354_cast_fp16")]; + tensor var_34361_begin_0 = const()[name = string("op_34361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34361_end_0 = const()[name = string("op_34361_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34361_end_mask_0 = const()[name = string("op_34361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34361_cast_fp16 = slice_by_index(begin = var_34361_begin_0, end = var_34361_end_0, end_mask = var_34361_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34361_cast_fp16")]; + tensor var_34368_begin_0 = const()[name = string("op_34368_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34368_end_0 = const()[name = string("op_34368_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34368_end_mask_0 = const()[name = string("op_34368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34368_cast_fp16 = slice_by_index(begin = var_34368_begin_0, end = var_34368_end_0, end_mask = var_34368_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34368_cast_fp16")]; + tensor var_34375_begin_0 = const()[name = string("op_34375_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34375_end_0 = const()[name = string("op_34375_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34375_end_mask_0 = const()[name = string("op_34375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34375_cast_fp16 = slice_by_index(begin = var_34375_begin_0, end = var_34375_end_0, end_mask = var_34375_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34375_cast_fp16")]; + tensor var_34382_begin_0 = const()[name = string("op_34382_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34382_end_0 = const()[name = string("op_34382_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34382_end_mask_0 = const()[name = string("op_34382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34382_cast_fp16 = slice_by_index(begin = var_34382_begin_0, end = var_34382_end_0, end_mask = var_34382_end_mask_0, x = var_33916_cast_fp16)[name = string("op_34382_cast_fp16")]; + tensor var_34389_begin_0 = const()[name = string("op_34389_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34389_end_0 = const()[name = string("op_34389_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34389_end_mask_0 = const()[name = string("op_34389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34389_cast_fp16 = slice_by_index(begin = var_34389_begin_0, end = var_34389_end_0, end_mask = var_34389_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34389_cast_fp16")]; + tensor var_34396_begin_0 = const()[name = string("op_34396_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34396_end_0 = const()[name = string("op_34396_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34396_end_mask_0 = const()[name = string("op_34396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34396_cast_fp16 = slice_by_index(begin = var_34396_begin_0, end = var_34396_end_0, end_mask = var_34396_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34396_cast_fp16")]; + tensor var_34403_begin_0 = const()[name = string("op_34403_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34403_end_0 = const()[name = string("op_34403_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34403_end_mask_0 = const()[name = string("op_34403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34403_cast_fp16 = slice_by_index(begin = var_34403_begin_0, end = var_34403_end_0, end_mask = var_34403_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34403_cast_fp16")]; + tensor var_34410_begin_0 = const()[name = string("op_34410_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34410_end_0 = const()[name = string("op_34410_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34410_end_mask_0 = const()[name = string("op_34410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34410_cast_fp16 = slice_by_index(begin = var_34410_begin_0, end = var_34410_end_0, end_mask = var_34410_end_mask_0, x = var_33920_cast_fp16)[name = string("op_34410_cast_fp16")]; + tensor var_34417_begin_0 = const()[name = string("op_34417_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34417_end_0 = const()[name = string("op_34417_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34417_end_mask_0 = const()[name = string("op_34417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34417_cast_fp16 = slice_by_index(begin = var_34417_begin_0, end = var_34417_end_0, end_mask = var_34417_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34417_cast_fp16")]; + tensor var_34424_begin_0 = const()[name = string("op_34424_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34424_end_0 = const()[name = string("op_34424_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34424_end_mask_0 = const()[name = string("op_34424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34424_cast_fp16 = slice_by_index(begin = var_34424_begin_0, end = var_34424_end_0, end_mask = var_34424_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34424_cast_fp16")]; + tensor var_34431_begin_0 = const()[name = string("op_34431_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34431_end_0 = const()[name = string("op_34431_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34431_end_mask_0 = const()[name = string("op_34431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34431_cast_fp16 = slice_by_index(begin = var_34431_begin_0, end = var_34431_end_0, end_mask = var_34431_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34431_cast_fp16")]; + tensor var_34438_begin_0 = const()[name = string("op_34438_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34438_end_0 = const()[name = string("op_34438_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34438_end_mask_0 = const()[name = string("op_34438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34438_cast_fp16 = slice_by_index(begin = var_34438_begin_0, end = var_34438_end_0, end_mask = var_34438_end_mask_0, x = var_33924_cast_fp16)[name = string("op_34438_cast_fp16")]; + tensor var_34445_begin_0 = const()[name = string("op_34445_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34445_end_0 = const()[name = string("op_34445_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34445_end_mask_0 = const()[name = string("op_34445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34445_cast_fp16 = slice_by_index(begin = var_34445_begin_0, end = var_34445_end_0, end_mask = var_34445_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34445_cast_fp16")]; + tensor var_34452_begin_0 = const()[name = string("op_34452_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34452_end_0 = const()[name = string("op_34452_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34452_end_mask_0 = const()[name = string("op_34452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34452_cast_fp16 = slice_by_index(begin = var_34452_begin_0, end = var_34452_end_0, end_mask = var_34452_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34452_cast_fp16")]; + tensor var_34459_begin_0 = const()[name = string("op_34459_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34459_end_0 = const()[name = string("op_34459_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34459_end_mask_0 = const()[name = string("op_34459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34459_cast_fp16 = slice_by_index(begin = var_34459_begin_0, end = var_34459_end_0, end_mask = var_34459_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34459_cast_fp16")]; + tensor var_34466_begin_0 = const()[name = string("op_34466_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34466_end_0 = const()[name = string("op_34466_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34466_end_mask_0 = const()[name = string("op_34466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34466_cast_fp16 = slice_by_index(begin = var_34466_begin_0, end = var_34466_end_0, end_mask = var_34466_end_mask_0, x = var_33928_cast_fp16)[name = string("op_34466_cast_fp16")]; + tensor var_34473_begin_0 = const()[name = string("op_34473_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34473_end_0 = const()[name = string("op_34473_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34473_end_mask_0 = const()[name = string("op_34473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34473_cast_fp16 = slice_by_index(begin = var_34473_begin_0, end = var_34473_end_0, end_mask = var_34473_end_mask_0, x = var_33932_cast_fp16)[name = string("op_34473_cast_fp16")]; + tensor var_34480_begin_0 = const()[name = string("op_34480_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34480_end_0 = const()[name = string("op_34480_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34480_end_mask_0 = const()[name = string("op_34480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34480_cast_fp16 = slice_by_index(begin = var_34480_begin_0, end = var_34480_end_0, end_mask = var_34480_end_mask_0, x = var_33932_cast_fp16)[name = string("op_34480_cast_fp16")]; + tensor var_34487_begin_0 = const()[name = string("op_34487_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34487_end_0 = const()[name = string("op_34487_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34487_end_mask_0 = const()[name = string("op_34487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34487_cast_fp16 = slice_by_index(begin = var_34487_begin_0, end = var_34487_end_0, end_mask = var_34487_end_mask_0, x = var_33932_cast_fp16)[name = string("op_34487_cast_fp16")]; + tensor var_34494_begin_0 = const()[name = string("op_34494_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34494_end_0 = const()[name = string("op_34494_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34494_end_mask_0 = const()[name = string("op_34494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34494_cast_fp16 = slice_by_index(begin = var_34494_begin_0, end = var_34494_end_0, end_mask = var_34494_end_mask_0, x = var_33932_cast_fp16)[name = string("op_34494_cast_fp16")]; + tensor k_45_perm_0 = const()[name = string("k_45_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_34499_begin_0 = const()[name = string("op_34499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34499_end_0 = const()[name = string("op_34499_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_34499_end_mask_0 = const()[name = string("op_34499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_45_cast_fp16 = transpose(perm = k_45_perm_0, x = key_45_cast_fp16)[name = string("transpose_9")]; + tensor var_34499_cast_fp16 = slice_by_index(begin = var_34499_begin_0, end = var_34499_end_0, end_mask = var_34499_end_mask_0, x = k_45_cast_fp16)[name = string("op_34499_cast_fp16")]; + tensor var_34503_begin_0 = const()[name = string("op_34503_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_34503_end_0 = const()[name = string("op_34503_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_34503_end_mask_0 = const()[name = string("op_34503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34503_cast_fp16 = slice_by_index(begin = var_34503_begin_0, end = var_34503_end_0, end_mask = var_34503_end_mask_0, x = k_45_cast_fp16)[name = string("op_34503_cast_fp16")]; + tensor var_34507_begin_0 = const()[name = string("op_34507_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_34507_end_0 = const()[name = string("op_34507_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_34507_end_mask_0 = const()[name = string("op_34507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34507_cast_fp16 = slice_by_index(begin = var_34507_begin_0, end = var_34507_end_0, end_mask = var_34507_end_mask_0, x = k_45_cast_fp16)[name = string("op_34507_cast_fp16")]; + tensor var_34511_begin_0 = const()[name = string("op_34511_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_34511_end_0 = const()[name = string("op_34511_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_34511_end_mask_0 = const()[name = string("op_34511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34511_cast_fp16 = slice_by_index(begin = var_34511_begin_0, end = var_34511_end_0, end_mask = var_34511_end_mask_0, x = k_45_cast_fp16)[name = string("op_34511_cast_fp16")]; + tensor var_34515_begin_0 = const()[name = string("op_34515_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_34515_end_0 = const()[name = string("op_34515_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_34515_end_mask_0 = const()[name = string("op_34515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34515_cast_fp16 = slice_by_index(begin = var_34515_begin_0, end = var_34515_end_0, end_mask = var_34515_end_mask_0, x = k_45_cast_fp16)[name = string("op_34515_cast_fp16")]; + tensor var_34519_begin_0 = const()[name = string("op_34519_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_34519_end_0 = const()[name = string("op_34519_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_34519_end_mask_0 = const()[name = string("op_34519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34519_cast_fp16 = slice_by_index(begin = var_34519_begin_0, end = var_34519_end_0, end_mask = var_34519_end_mask_0, x = k_45_cast_fp16)[name = string("op_34519_cast_fp16")]; + tensor var_34523_begin_0 = const()[name = string("op_34523_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_34523_end_0 = const()[name = string("op_34523_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_34523_end_mask_0 = const()[name = string("op_34523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34523_cast_fp16 = slice_by_index(begin = var_34523_begin_0, end = var_34523_end_0, end_mask = var_34523_end_mask_0, x = k_45_cast_fp16)[name = string("op_34523_cast_fp16")]; + tensor var_34527_begin_0 = const()[name = string("op_34527_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_34527_end_0 = const()[name = string("op_34527_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_34527_end_mask_0 = const()[name = string("op_34527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34527_cast_fp16 = slice_by_index(begin = var_34527_begin_0, end = var_34527_end_0, end_mask = var_34527_end_mask_0, x = k_45_cast_fp16)[name = string("op_34527_cast_fp16")]; + tensor var_34531_begin_0 = const()[name = string("op_34531_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_34531_end_0 = const()[name = string("op_34531_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_34531_end_mask_0 = const()[name = string("op_34531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34531_cast_fp16 = slice_by_index(begin = var_34531_begin_0, end = var_34531_end_0, end_mask = var_34531_end_mask_0, x = k_45_cast_fp16)[name = string("op_34531_cast_fp16")]; + tensor var_34535_begin_0 = const()[name = string("op_34535_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_34535_end_0 = const()[name = string("op_34535_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_34535_end_mask_0 = const()[name = string("op_34535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34535_cast_fp16 = slice_by_index(begin = var_34535_begin_0, end = var_34535_end_0, end_mask = var_34535_end_mask_0, x = k_45_cast_fp16)[name = string("op_34535_cast_fp16")]; + tensor var_34539_begin_0 = const()[name = string("op_34539_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_34539_end_0 = const()[name = string("op_34539_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_34539_end_mask_0 = const()[name = string("op_34539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34539_cast_fp16 = slice_by_index(begin = var_34539_begin_0, end = var_34539_end_0, end_mask = var_34539_end_mask_0, x = k_45_cast_fp16)[name = string("op_34539_cast_fp16")]; + tensor var_34543_begin_0 = const()[name = string("op_34543_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_34543_end_0 = const()[name = string("op_34543_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_34543_end_mask_0 = const()[name = string("op_34543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34543_cast_fp16 = slice_by_index(begin = var_34543_begin_0, end = var_34543_end_0, end_mask = var_34543_end_mask_0, x = k_45_cast_fp16)[name = string("op_34543_cast_fp16")]; + tensor var_34547_begin_0 = const()[name = string("op_34547_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_34547_end_0 = const()[name = string("op_34547_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_34547_end_mask_0 = const()[name = string("op_34547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34547_cast_fp16 = slice_by_index(begin = var_34547_begin_0, end = var_34547_end_0, end_mask = var_34547_end_mask_0, x = k_45_cast_fp16)[name = string("op_34547_cast_fp16")]; + tensor var_34551_begin_0 = const()[name = string("op_34551_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_34551_end_0 = const()[name = string("op_34551_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_34551_end_mask_0 = const()[name = string("op_34551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34551_cast_fp16 = slice_by_index(begin = var_34551_begin_0, end = var_34551_end_0, end_mask = var_34551_end_mask_0, x = k_45_cast_fp16)[name = string("op_34551_cast_fp16")]; + tensor var_34555_begin_0 = const()[name = string("op_34555_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_34555_end_0 = const()[name = string("op_34555_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_34555_end_mask_0 = const()[name = string("op_34555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34555_cast_fp16 = slice_by_index(begin = var_34555_begin_0, end = var_34555_end_0, end_mask = var_34555_end_mask_0, x = k_45_cast_fp16)[name = string("op_34555_cast_fp16")]; + tensor var_34559_begin_0 = const()[name = string("op_34559_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_34559_end_0 = const()[name = string("op_34559_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_34559_end_mask_0 = const()[name = string("op_34559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34559_cast_fp16 = slice_by_index(begin = var_34559_begin_0, end = var_34559_end_0, end_mask = var_34559_end_mask_0, x = k_45_cast_fp16)[name = string("op_34559_cast_fp16")]; + tensor var_34563_begin_0 = const()[name = string("op_34563_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_34563_end_0 = const()[name = string("op_34563_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_34563_end_mask_0 = const()[name = string("op_34563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34563_cast_fp16 = slice_by_index(begin = var_34563_begin_0, end = var_34563_end_0, end_mask = var_34563_end_mask_0, x = k_45_cast_fp16)[name = string("op_34563_cast_fp16")]; + tensor var_34567_begin_0 = const()[name = string("op_34567_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_34567_end_0 = const()[name = string("op_34567_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_34567_end_mask_0 = const()[name = string("op_34567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34567_cast_fp16 = slice_by_index(begin = var_34567_begin_0, end = var_34567_end_0, end_mask = var_34567_end_mask_0, x = k_45_cast_fp16)[name = string("op_34567_cast_fp16")]; + tensor var_34571_begin_0 = const()[name = string("op_34571_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_34571_end_0 = const()[name = string("op_34571_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_34571_end_mask_0 = const()[name = string("op_34571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34571_cast_fp16 = slice_by_index(begin = var_34571_begin_0, end = var_34571_end_0, end_mask = var_34571_end_mask_0, x = k_45_cast_fp16)[name = string("op_34571_cast_fp16")]; + tensor var_34575_begin_0 = const()[name = string("op_34575_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_34575_end_0 = const()[name = string("op_34575_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_34575_end_mask_0 = const()[name = string("op_34575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34575_cast_fp16 = slice_by_index(begin = var_34575_begin_0, end = var_34575_end_0, end_mask = var_34575_end_mask_0, x = k_45_cast_fp16)[name = string("op_34575_cast_fp16")]; + tensor var_34577_begin_0 = const()[name = string("op_34577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34577_end_0 = const()[name = string("op_34577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34577_end_mask_0 = const()[name = string("op_34577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34577_cast_fp16 = slice_by_index(begin = var_34577_begin_0, end = var_34577_end_0, end_mask = var_34577_end_mask_0, x = value_45_cast_fp16)[name = string("op_34577_cast_fp16")]; + tensor var_34581_begin_0 = const()[name = string("op_34581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34581_end_0 = const()[name = string("op_34581_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_34581_end_mask_0 = const()[name = string("op_34581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34581_cast_fp16 = slice_by_index(begin = var_34581_begin_0, end = var_34581_end_0, end_mask = var_34581_end_mask_0, x = value_45_cast_fp16)[name = string("op_34581_cast_fp16")]; + tensor var_34585_begin_0 = const()[name = string("op_34585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34585_end_0 = const()[name = string("op_34585_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_34585_end_mask_0 = const()[name = string("op_34585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34585_cast_fp16 = slice_by_index(begin = var_34585_begin_0, end = var_34585_end_0, end_mask = var_34585_end_mask_0, x = value_45_cast_fp16)[name = string("op_34585_cast_fp16")]; + tensor var_34589_begin_0 = const()[name = string("op_34589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34589_end_0 = const()[name = string("op_34589_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_34589_end_mask_0 = const()[name = string("op_34589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34589_cast_fp16 = slice_by_index(begin = var_34589_begin_0, end = var_34589_end_0, end_mask = var_34589_end_mask_0, x = value_45_cast_fp16)[name = string("op_34589_cast_fp16")]; + tensor var_34593_begin_0 = const()[name = string("op_34593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34593_end_0 = const()[name = string("op_34593_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_34593_end_mask_0 = const()[name = string("op_34593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34593_cast_fp16 = slice_by_index(begin = var_34593_begin_0, end = var_34593_end_0, end_mask = var_34593_end_mask_0, x = value_45_cast_fp16)[name = string("op_34593_cast_fp16")]; + tensor var_34597_begin_0 = const()[name = string("op_34597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34597_end_0 = const()[name = string("op_34597_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_34597_end_mask_0 = const()[name = string("op_34597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34597_cast_fp16 = slice_by_index(begin = var_34597_begin_0, end = var_34597_end_0, end_mask = var_34597_end_mask_0, x = value_45_cast_fp16)[name = string("op_34597_cast_fp16")]; + tensor var_34601_begin_0 = const()[name = string("op_34601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34601_end_0 = const()[name = string("op_34601_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_34601_end_mask_0 = const()[name = string("op_34601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34601_cast_fp16 = slice_by_index(begin = var_34601_begin_0, end = var_34601_end_0, end_mask = var_34601_end_mask_0, x = value_45_cast_fp16)[name = string("op_34601_cast_fp16")]; + tensor var_34605_begin_0 = const()[name = string("op_34605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34605_end_0 = const()[name = string("op_34605_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_34605_end_mask_0 = const()[name = string("op_34605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34605_cast_fp16 = slice_by_index(begin = var_34605_begin_0, end = var_34605_end_0, end_mask = var_34605_end_mask_0, x = value_45_cast_fp16)[name = string("op_34605_cast_fp16")]; + tensor var_34609_begin_0 = const()[name = string("op_34609_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34609_end_0 = const()[name = string("op_34609_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_34609_end_mask_0 = const()[name = string("op_34609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34609_cast_fp16 = slice_by_index(begin = var_34609_begin_0, end = var_34609_end_0, end_mask = var_34609_end_mask_0, x = value_45_cast_fp16)[name = string("op_34609_cast_fp16")]; + tensor var_34613_begin_0 = const()[name = string("op_34613_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34613_end_0 = const()[name = string("op_34613_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_34613_end_mask_0 = const()[name = string("op_34613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34613_cast_fp16 = slice_by_index(begin = var_34613_begin_0, end = var_34613_end_0, end_mask = var_34613_end_mask_0, x = value_45_cast_fp16)[name = string("op_34613_cast_fp16")]; + tensor var_34617_begin_0 = const()[name = string("op_34617_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34617_end_0 = const()[name = string("op_34617_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_34617_end_mask_0 = const()[name = string("op_34617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34617_cast_fp16 = slice_by_index(begin = var_34617_begin_0, end = var_34617_end_0, end_mask = var_34617_end_mask_0, x = value_45_cast_fp16)[name = string("op_34617_cast_fp16")]; + tensor var_34621_begin_0 = const()[name = string("op_34621_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34621_end_0 = const()[name = string("op_34621_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_34621_end_mask_0 = const()[name = string("op_34621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34621_cast_fp16 = slice_by_index(begin = var_34621_begin_0, end = var_34621_end_0, end_mask = var_34621_end_mask_0, x = value_45_cast_fp16)[name = string("op_34621_cast_fp16")]; + tensor var_34625_begin_0 = const()[name = string("op_34625_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34625_end_0 = const()[name = string("op_34625_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_34625_end_mask_0 = const()[name = string("op_34625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34625_cast_fp16 = slice_by_index(begin = var_34625_begin_0, end = var_34625_end_0, end_mask = var_34625_end_mask_0, x = value_45_cast_fp16)[name = string("op_34625_cast_fp16")]; + tensor var_34629_begin_0 = const()[name = string("op_34629_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34629_end_0 = const()[name = string("op_34629_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_34629_end_mask_0 = const()[name = string("op_34629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34629_cast_fp16 = slice_by_index(begin = var_34629_begin_0, end = var_34629_end_0, end_mask = var_34629_end_mask_0, x = value_45_cast_fp16)[name = string("op_34629_cast_fp16")]; + tensor var_34633_begin_0 = const()[name = string("op_34633_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34633_end_0 = const()[name = string("op_34633_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_34633_end_mask_0 = const()[name = string("op_34633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34633_cast_fp16 = slice_by_index(begin = var_34633_begin_0, end = var_34633_end_0, end_mask = var_34633_end_mask_0, x = value_45_cast_fp16)[name = string("op_34633_cast_fp16")]; + tensor var_34637_begin_0 = const()[name = string("op_34637_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34637_end_0 = const()[name = string("op_34637_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_34637_end_mask_0 = const()[name = string("op_34637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34637_cast_fp16 = slice_by_index(begin = var_34637_begin_0, end = var_34637_end_0, end_mask = var_34637_end_mask_0, x = value_45_cast_fp16)[name = string("op_34637_cast_fp16")]; + tensor var_34641_begin_0 = const()[name = string("op_34641_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34641_end_0 = const()[name = string("op_34641_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_34641_end_mask_0 = const()[name = string("op_34641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34641_cast_fp16 = slice_by_index(begin = var_34641_begin_0, end = var_34641_end_0, end_mask = var_34641_end_mask_0, x = value_45_cast_fp16)[name = string("op_34641_cast_fp16")]; + tensor var_34645_begin_0 = const()[name = string("op_34645_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34645_end_0 = const()[name = string("op_34645_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_34645_end_mask_0 = const()[name = string("op_34645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34645_cast_fp16 = slice_by_index(begin = var_34645_begin_0, end = var_34645_end_0, end_mask = var_34645_end_mask_0, x = value_45_cast_fp16)[name = string("op_34645_cast_fp16")]; + tensor var_34649_begin_0 = const()[name = string("op_34649_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34649_end_0 = const()[name = string("op_34649_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_34649_end_mask_0 = const()[name = string("op_34649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34649_cast_fp16 = slice_by_index(begin = var_34649_begin_0, end = var_34649_end_0, end_mask = var_34649_end_mask_0, x = value_45_cast_fp16)[name = string("op_34649_cast_fp16")]; + tensor var_34653_begin_0 = const()[name = string("op_34653_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34653_end_0 = const()[name = string("op_34653_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_34653_end_mask_0 = const()[name = string("op_34653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34653_cast_fp16 = slice_by_index(begin = var_34653_begin_0, end = var_34653_end_0, end_mask = var_34653_end_mask_0, x = value_45_cast_fp16)[name = string("op_34653_cast_fp16")]; + string _SplitHeadsQ__mh_w_3521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3521_equation_0, values = (var_34499_cast_fp16, var_33941_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3521_cast_fp16")]; + string _SplitHeadsQ__mh_w_3523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3523_equation_0, values = (var_34499_cast_fp16, var_33948_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3523_cast_fp16")]; + string _SplitHeadsQ__mh_w_3525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3525_equation_0, values = (var_34499_cast_fp16, var_33955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3525_cast_fp16")]; + string _SplitHeadsQ__mh_w_3527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3527_equation_0, values = (var_34499_cast_fp16, var_33962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3527_cast_fp16")]; + string _SplitHeadsQ__mh_w_3529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3529_equation_0, values = (var_34503_cast_fp16, var_33969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3529_cast_fp16")]; + string _SplitHeadsQ__mh_w_3531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3531_equation_0, values = (var_34503_cast_fp16, var_33976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3531_cast_fp16")]; + string _SplitHeadsQ__mh_w_3533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3533_equation_0, values = (var_34503_cast_fp16, var_33983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3533_cast_fp16")]; + string _SplitHeadsQ__mh_w_3535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3535_equation_0, values = (var_34503_cast_fp16, var_33990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3535_cast_fp16")]; + string _SplitHeadsQ__mh_w_3537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3537_equation_0, values = (var_34507_cast_fp16, var_33997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3537_cast_fp16")]; + string _SplitHeadsQ__mh_w_3539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3539_equation_0, values = (var_34507_cast_fp16, var_34004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3539_cast_fp16")]; + string _SplitHeadsQ__mh_w_3541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3541_equation_0, values = (var_34507_cast_fp16, var_34011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3541_cast_fp16")]; + string _SplitHeadsQ__mh_w_3543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3543_equation_0, values = (var_34507_cast_fp16, var_34018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3543_cast_fp16")]; + string _SplitHeadsQ__mh_w_3545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3545_equation_0, values = (var_34511_cast_fp16, var_34025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3545_cast_fp16")]; + string _SplitHeadsQ__mh_w_3547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3547_equation_0, values = (var_34511_cast_fp16, var_34032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3547_cast_fp16")]; + string _SplitHeadsQ__mh_w_3549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3549_equation_0, values = (var_34511_cast_fp16, var_34039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3549_cast_fp16")]; + string _SplitHeadsQ__mh_w_3551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3551_equation_0, values = (var_34511_cast_fp16, var_34046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3551_cast_fp16")]; + string _SplitHeadsQ__mh_w_3553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3553_equation_0, values = (var_34515_cast_fp16, var_34053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3553_cast_fp16")]; + string _SplitHeadsQ__mh_w_3555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3555_equation_0, values = (var_34515_cast_fp16, var_34060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3555_cast_fp16")]; + string _SplitHeadsQ__mh_w_3557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3557_equation_0, values = (var_34515_cast_fp16, var_34067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3557_cast_fp16")]; + string _SplitHeadsQ__mh_w_3559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3559_equation_0, values = (var_34515_cast_fp16, var_34074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3559_cast_fp16")]; + string _SplitHeadsQ__mh_w_3561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3561_equation_0, values = (var_34519_cast_fp16, var_34081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3561_cast_fp16")]; + string _SplitHeadsQ__mh_w_3563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3563_equation_0, values = (var_34519_cast_fp16, var_34088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3563_cast_fp16")]; + string _SplitHeadsQ__mh_w_3565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3565_equation_0, values = (var_34519_cast_fp16, var_34095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3565_cast_fp16")]; + string _SplitHeadsQ__mh_w_3567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3567_equation_0, values = (var_34519_cast_fp16, var_34102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3567_cast_fp16")]; + string _SplitHeadsQ__mh_w_3569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3569_equation_0, values = (var_34523_cast_fp16, var_34109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3569_cast_fp16")]; + string _SplitHeadsQ__mh_w_3571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3571_equation_0, values = (var_34523_cast_fp16, var_34116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3571_cast_fp16")]; + string _SplitHeadsQ__mh_w_3573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3573_equation_0, values = (var_34523_cast_fp16, var_34123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3573_cast_fp16")]; + string _SplitHeadsQ__mh_w_3575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3575_equation_0, values = (var_34523_cast_fp16, var_34130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3575_cast_fp16")]; + string _SplitHeadsQ__mh_w_3577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3577_equation_0, values = (var_34527_cast_fp16, var_34137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3577_cast_fp16")]; + string _SplitHeadsQ__mh_w_3579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3579_equation_0, values = (var_34527_cast_fp16, var_34144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3579_cast_fp16")]; + string _SplitHeadsQ__mh_w_3581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3581_equation_0, values = (var_34527_cast_fp16, var_34151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3581_cast_fp16")]; + string _SplitHeadsQ__mh_w_3583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3583_equation_0, values = (var_34527_cast_fp16, var_34158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3583_cast_fp16")]; + string _SplitHeadsQ__mh_w_3585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3585_equation_0, values = (var_34531_cast_fp16, var_34165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3585_cast_fp16")]; + string _SplitHeadsQ__mh_w_3587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3587_equation_0, values = (var_34531_cast_fp16, var_34172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3587_cast_fp16")]; + string _SplitHeadsQ__mh_w_3589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3589_equation_0, values = (var_34531_cast_fp16, var_34179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3589_cast_fp16")]; + string _SplitHeadsQ__mh_w_3591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3591_equation_0, values = (var_34531_cast_fp16, var_34186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3591_cast_fp16")]; + string _SplitHeadsQ__mh_w_3593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3593_equation_0, values = (var_34535_cast_fp16, var_34193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3593_cast_fp16")]; + string _SplitHeadsQ__mh_w_3595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3595_equation_0, values = (var_34535_cast_fp16, var_34200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3595_cast_fp16")]; + string _SplitHeadsQ__mh_w_3597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3597_equation_0, values = (var_34535_cast_fp16, var_34207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3597_cast_fp16")]; + string _SplitHeadsQ__mh_w_3599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3599_equation_0, values = (var_34535_cast_fp16, var_34214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3599_cast_fp16")]; + string _SplitHeadsQ__mh_w_3601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3601_equation_0, values = (var_34539_cast_fp16, var_34221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3601_cast_fp16")]; + string _SplitHeadsQ__mh_w_3603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3603_equation_0, values = (var_34539_cast_fp16, var_34228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3603_cast_fp16")]; + string _SplitHeadsQ__mh_w_3605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3605_equation_0, values = (var_34539_cast_fp16, var_34235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3605_cast_fp16")]; + string _SplitHeadsQ__mh_w_3607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3607_equation_0, values = (var_34539_cast_fp16, var_34242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3607_cast_fp16")]; + string _SplitHeadsQ__mh_w_3609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3609_equation_0, values = (var_34543_cast_fp16, var_34249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3609_cast_fp16")]; + string _SplitHeadsQ__mh_w_3611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3611_equation_0, values = (var_34543_cast_fp16, var_34256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3611_cast_fp16")]; + string _SplitHeadsQ__mh_w_3613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3613_equation_0, values = (var_34543_cast_fp16, var_34263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3613_cast_fp16")]; + string _SplitHeadsQ__mh_w_3615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3615_equation_0, values = (var_34543_cast_fp16, var_34270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3615_cast_fp16")]; + string _SplitHeadsQ__mh_w_3617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3617_equation_0, values = (var_34547_cast_fp16, var_34277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3617_cast_fp16")]; + string _SplitHeadsQ__mh_w_3619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3619_equation_0, values = (var_34547_cast_fp16, var_34284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3619_cast_fp16")]; + string _SplitHeadsQ__mh_w_3621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3621_equation_0, values = (var_34547_cast_fp16, var_34291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3621_cast_fp16")]; + string _SplitHeadsQ__mh_w_3623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3623_equation_0, values = (var_34547_cast_fp16, var_34298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3623_cast_fp16")]; + string _SplitHeadsQ__mh_w_3625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3625_equation_0, values = (var_34551_cast_fp16, var_34305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3625_cast_fp16")]; + string _SplitHeadsQ__mh_w_3627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3627_equation_0, values = (var_34551_cast_fp16, var_34312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3627_cast_fp16")]; + string _SplitHeadsQ__mh_w_3629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3629_equation_0, values = (var_34551_cast_fp16, var_34319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3629_cast_fp16")]; + string _SplitHeadsQ__mh_w_3631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3631_equation_0, values = (var_34551_cast_fp16, var_34326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3631_cast_fp16")]; + string _SplitHeadsQ__mh_w_3633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3633_equation_0, values = (var_34555_cast_fp16, var_34333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3633_cast_fp16")]; + string _SplitHeadsQ__mh_w_3635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3635_equation_0, values = (var_34555_cast_fp16, var_34340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3635_cast_fp16")]; + string _SplitHeadsQ__mh_w_3637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3637_equation_0, values = (var_34555_cast_fp16, var_34347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3637_cast_fp16")]; + string _SplitHeadsQ__mh_w_3639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3639_equation_0, values = (var_34555_cast_fp16, var_34354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3639_cast_fp16")]; + string _SplitHeadsQ__mh_w_3641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3641_equation_0, values = (var_34559_cast_fp16, var_34361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3641_cast_fp16")]; + string _SplitHeadsQ__mh_w_3643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3643_equation_0, values = (var_34559_cast_fp16, var_34368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3643_cast_fp16")]; + string _SplitHeadsQ__mh_w_3645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3645_equation_0, values = (var_34559_cast_fp16, var_34375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3645_cast_fp16")]; + string _SplitHeadsQ__mh_w_3647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3647_equation_0, values = (var_34559_cast_fp16, var_34382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3647_cast_fp16")]; + string _SplitHeadsQ__mh_w_3649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3649_equation_0, values = (var_34563_cast_fp16, var_34389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3649_cast_fp16")]; + string _SplitHeadsQ__mh_w_3651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3651_equation_0, values = (var_34563_cast_fp16, var_34396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3651_cast_fp16")]; + string _SplitHeadsQ__mh_w_3653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3653_equation_0, values = (var_34563_cast_fp16, var_34403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3653_cast_fp16")]; + string _SplitHeadsQ__mh_w_3655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3655_equation_0, values = (var_34563_cast_fp16, var_34410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3655_cast_fp16")]; + string _SplitHeadsQ__mh_w_3657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3657_equation_0, values = (var_34567_cast_fp16, var_34417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3657_cast_fp16")]; + string _SplitHeadsQ__mh_w_3659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3659_equation_0, values = (var_34567_cast_fp16, var_34424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3659_cast_fp16")]; + string _SplitHeadsQ__mh_w_3661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3661_equation_0, values = (var_34567_cast_fp16, var_34431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3661_cast_fp16")]; + string _SplitHeadsQ__mh_w_3663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3663_equation_0, values = (var_34567_cast_fp16, var_34438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3663_cast_fp16")]; + string _SplitHeadsQ__mh_w_3665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3665_equation_0, values = (var_34571_cast_fp16, var_34445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3665_cast_fp16")]; + string _SplitHeadsQ__mh_w_3667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3667_equation_0, values = (var_34571_cast_fp16, var_34452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3667_cast_fp16")]; + string _SplitHeadsQ__mh_w_3669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3669_equation_0, values = (var_34571_cast_fp16, var_34459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3669_cast_fp16")]; + string _SplitHeadsQ__mh_w_3671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3671_equation_0, values = (var_34571_cast_fp16, var_34466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3671_cast_fp16")]; + string _SplitHeadsQ__mh_w_3673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3673_equation_0, values = (var_34575_cast_fp16, var_34473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3673_cast_fp16")]; + string _SplitHeadsQ__mh_w_3675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3675_equation_0, values = (var_34575_cast_fp16, var_34480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3675_cast_fp16")]; + string _SplitHeadsQ__mh_w_3677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3677_equation_0, values = (var_34575_cast_fp16, var_34487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3677_cast_fp16")]; + string _SplitHeadsQ__mh_w_3679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3679_equation_0, values = (var_34575_cast_fp16, var_34494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3679_cast_fp16")]; + fp16 var_34816_to_fp16 = const()[name = string("op_34816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3521_cast_fp16, y = var_34816_to_fp16)[name = string("aw_chunk_3521_cast_fp16")]; + fp16 var_34818_to_fp16 = const()[name = string("op_34818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3523_cast_fp16, y = var_34818_to_fp16)[name = string("aw_chunk_3523_cast_fp16")]; + fp16 var_34820_to_fp16 = const()[name = string("op_34820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3525_cast_fp16, y = var_34820_to_fp16)[name = string("aw_chunk_3525_cast_fp16")]; + fp16 var_34822_to_fp16 = const()[name = string("op_34822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3527_cast_fp16, y = var_34822_to_fp16)[name = string("aw_chunk_3527_cast_fp16")]; + fp16 var_34824_to_fp16 = const()[name = string("op_34824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3529_cast_fp16, y = var_34824_to_fp16)[name = string("aw_chunk_3529_cast_fp16")]; + fp16 var_34826_to_fp16 = const()[name = string("op_34826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3531_cast_fp16, y = var_34826_to_fp16)[name = string("aw_chunk_3531_cast_fp16")]; + fp16 var_34828_to_fp16 = const()[name = string("op_34828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3533_cast_fp16, y = var_34828_to_fp16)[name = string("aw_chunk_3533_cast_fp16")]; + fp16 var_34830_to_fp16 = const()[name = string("op_34830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3535_cast_fp16, y = var_34830_to_fp16)[name = string("aw_chunk_3535_cast_fp16")]; + fp16 var_34832_to_fp16 = const()[name = string("op_34832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3537_cast_fp16, y = var_34832_to_fp16)[name = string("aw_chunk_3537_cast_fp16")]; + fp16 var_34834_to_fp16 = const()[name = string("op_34834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3539_cast_fp16, y = var_34834_to_fp16)[name = string("aw_chunk_3539_cast_fp16")]; + fp16 var_34836_to_fp16 = const()[name = string("op_34836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3541_cast_fp16, y = var_34836_to_fp16)[name = string("aw_chunk_3541_cast_fp16")]; + fp16 var_34838_to_fp16 = const()[name = string("op_34838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3543_cast_fp16, y = var_34838_to_fp16)[name = string("aw_chunk_3543_cast_fp16")]; + fp16 var_34840_to_fp16 = const()[name = string("op_34840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3545_cast_fp16, y = var_34840_to_fp16)[name = string("aw_chunk_3545_cast_fp16")]; + fp16 var_34842_to_fp16 = const()[name = string("op_34842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3547_cast_fp16, y = var_34842_to_fp16)[name = string("aw_chunk_3547_cast_fp16")]; + fp16 var_34844_to_fp16 = const()[name = string("op_34844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3549_cast_fp16, y = var_34844_to_fp16)[name = string("aw_chunk_3549_cast_fp16")]; + fp16 var_34846_to_fp16 = const()[name = string("op_34846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3551_cast_fp16, y = var_34846_to_fp16)[name = string("aw_chunk_3551_cast_fp16")]; + fp16 var_34848_to_fp16 = const()[name = string("op_34848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3553_cast_fp16, y = var_34848_to_fp16)[name = string("aw_chunk_3553_cast_fp16")]; + fp16 var_34850_to_fp16 = const()[name = string("op_34850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3555_cast_fp16, y = var_34850_to_fp16)[name = string("aw_chunk_3555_cast_fp16")]; + fp16 var_34852_to_fp16 = const()[name = string("op_34852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3557_cast_fp16, y = var_34852_to_fp16)[name = string("aw_chunk_3557_cast_fp16")]; + fp16 var_34854_to_fp16 = const()[name = string("op_34854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3559_cast_fp16, y = var_34854_to_fp16)[name = string("aw_chunk_3559_cast_fp16")]; + fp16 var_34856_to_fp16 = const()[name = string("op_34856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3561_cast_fp16, y = var_34856_to_fp16)[name = string("aw_chunk_3561_cast_fp16")]; + fp16 var_34858_to_fp16 = const()[name = string("op_34858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3563_cast_fp16, y = var_34858_to_fp16)[name = string("aw_chunk_3563_cast_fp16")]; + fp16 var_34860_to_fp16 = const()[name = string("op_34860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3565_cast_fp16, y = var_34860_to_fp16)[name = string("aw_chunk_3565_cast_fp16")]; + fp16 var_34862_to_fp16 = const()[name = string("op_34862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3567_cast_fp16, y = var_34862_to_fp16)[name = string("aw_chunk_3567_cast_fp16")]; + fp16 var_34864_to_fp16 = const()[name = string("op_34864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3569_cast_fp16, y = var_34864_to_fp16)[name = string("aw_chunk_3569_cast_fp16")]; + fp16 var_34866_to_fp16 = const()[name = string("op_34866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3571_cast_fp16, y = var_34866_to_fp16)[name = string("aw_chunk_3571_cast_fp16")]; + fp16 var_34868_to_fp16 = const()[name = string("op_34868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3573_cast_fp16, y = var_34868_to_fp16)[name = string("aw_chunk_3573_cast_fp16")]; + fp16 var_34870_to_fp16 = const()[name = string("op_34870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3575_cast_fp16, y = var_34870_to_fp16)[name = string("aw_chunk_3575_cast_fp16")]; + fp16 var_34872_to_fp16 = const()[name = string("op_34872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3577_cast_fp16, y = var_34872_to_fp16)[name = string("aw_chunk_3577_cast_fp16")]; + fp16 var_34874_to_fp16 = const()[name = string("op_34874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3579_cast_fp16, y = var_34874_to_fp16)[name = string("aw_chunk_3579_cast_fp16")]; + fp16 var_34876_to_fp16 = const()[name = string("op_34876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3581_cast_fp16, y = var_34876_to_fp16)[name = string("aw_chunk_3581_cast_fp16")]; + fp16 var_34878_to_fp16 = const()[name = string("op_34878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3583_cast_fp16, y = var_34878_to_fp16)[name = string("aw_chunk_3583_cast_fp16")]; + fp16 var_34880_to_fp16 = const()[name = string("op_34880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3585_cast_fp16, y = var_34880_to_fp16)[name = string("aw_chunk_3585_cast_fp16")]; + fp16 var_34882_to_fp16 = const()[name = string("op_34882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3587_cast_fp16, y = var_34882_to_fp16)[name = string("aw_chunk_3587_cast_fp16")]; + fp16 var_34884_to_fp16 = const()[name = string("op_34884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3589_cast_fp16, y = var_34884_to_fp16)[name = string("aw_chunk_3589_cast_fp16")]; + fp16 var_34886_to_fp16 = const()[name = string("op_34886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3591_cast_fp16, y = var_34886_to_fp16)[name = string("aw_chunk_3591_cast_fp16")]; + fp16 var_34888_to_fp16 = const()[name = string("op_34888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3593_cast_fp16, y = var_34888_to_fp16)[name = string("aw_chunk_3593_cast_fp16")]; + fp16 var_34890_to_fp16 = const()[name = string("op_34890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3595_cast_fp16, y = var_34890_to_fp16)[name = string("aw_chunk_3595_cast_fp16")]; + fp16 var_34892_to_fp16 = const()[name = string("op_34892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3597_cast_fp16, y = var_34892_to_fp16)[name = string("aw_chunk_3597_cast_fp16")]; + fp16 var_34894_to_fp16 = const()[name = string("op_34894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3599_cast_fp16, y = var_34894_to_fp16)[name = string("aw_chunk_3599_cast_fp16")]; + fp16 var_34896_to_fp16 = const()[name = string("op_34896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3601_cast_fp16, y = var_34896_to_fp16)[name = string("aw_chunk_3601_cast_fp16")]; + fp16 var_34898_to_fp16 = const()[name = string("op_34898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3603_cast_fp16, y = var_34898_to_fp16)[name = string("aw_chunk_3603_cast_fp16")]; + fp16 var_34900_to_fp16 = const()[name = string("op_34900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3605_cast_fp16, y = var_34900_to_fp16)[name = string("aw_chunk_3605_cast_fp16")]; + fp16 var_34902_to_fp16 = const()[name = string("op_34902_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3607_cast_fp16, y = var_34902_to_fp16)[name = string("aw_chunk_3607_cast_fp16")]; + fp16 var_34904_to_fp16 = const()[name = string("op_34904_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3609_cast_fp16, y = var_34904_to_fp16)[name = string("aw_chunk_3609_cast_fp16")]; + fp16 var_34906_to_fp16 = const()[name = string("op_34906_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3611_cast_fp16, y = var_34906_to_fp16)[name = string("aw_chunk_3611_cast_fp16")]; + fp16 var_34908_to_fp16 = const()[name = string("op_34908_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3613_cast_fp16, y = var_34908_to_fp16)[name = string("aw_chunk_3613_cast_fp16")]; + fp16 var_34910_to_fp16 = const()[name = string("op_34910_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3615_cast_fp16, y = var_34910_to_fp16)[name = string("aw_chunk_3615_cast_fp16")]; + fp16 var_34912_to_fp16 = const()[name = string("op_34912_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3617_cast_fp16, y = var_34912_to_fp16)[name = string("aw_chunk_3617_cast_fp16")]; + fp16 var_34914_to_fp16 = const()[name = string("op_34914_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3619_cast_fp16, y = var_34914_to_fp16)[name = string("aw_chunk_3619_cast_fp16")]; + fp16 var_34916_to_fp16 = const()[name = string("op_34916_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3621_cast_fp16, y = var_34916_to_fp16)[name = string("aw_chunk_3621_cast_fp16")]; + fp16 var_34918_to_fp16 = const()[name = string("op_34918_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3623_cast_fp16, y = var_34918_to_fp16)[name = string("aw_chunk_3623_cast_fp16")]; + fp16 var_34920_to_fp16 = const()[name = string("op_34920_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3625_cast_fp16, y = var_34920_to_fp16)[name = string("aw_chunk_3625_cast_fp16")]; + fp16 var_34922_to_fp16 = const()[name = string("op_34922_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3627_cast_fp16, y = var_34922_to_fp16)[name = string("aw_chunk_3627_cast_fp16")]; + fp16 var_34924_to_fp16 = const()[name = string("op_34924_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3629_cast_fp16, y = var_34924_to_fp16)[name = string("aw_chunk_3629_cast_fp16")]; + fp16 var_34926_to_fp16 = const()[name = string("op_34926_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3631_cast_fp16, y = var_34926_to_fp16)[name = string("aw_chunk_3631_cast_fp16")]; + fp16 var_34928_to_fp16 = const()[name = string("op_34928_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3633_cast_fp16, y = var_34928_to_fp16)[name = string("aw_chunk_3633_cast_fp16")]; + fp16 var_34930_to_fp16 = const()[name = string("op_34930_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3635_cast_fp16, y = var_34930_to_fp16)[name = string("aw_chunk_3635_cast_fp16")]; + fp16 var_34932_to_fp16 = const()[name = string("op_34932_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3637_cast_fp16, y = var_34932_to_fp16)[name = string("aw_chunk_3637_cast_fp16")]; + fp16 var_34934_to_fp16 = const()[name = string("op_34934_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3639_cast_fp16, y = var_34934_to_fp16)[name = string("aw_chunk_3639_cast_fp16")]; + fp16 var_34936_to_fp16 = const()[name = string("op_34936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3641_cast_fp16, y = var_34936_to_fp16)[name = string("aw_chunk_3641_cast_fp16")]; + fp16 var_34938_to_fp16 = const()[name = string("op_34938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3643_cast_fp16, y = var_34938_to_fp16)[name = string("aw_chunk_3643_cast_fp16")]; + fp16 var_34940_to_fp16 = const()[name = string("op_34940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3645_cast_fp16, y = var_34940_to_fp16)[name = string("aw_chunk_3645_cast_fp16")]; + fp16 var_34942_to_fp16 = const()[name = string("op_34942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3647_cast_fp16, y = var_34942_to_fp16)[name = string("aw_chunk_3647_cast_fp16")]; + fp16 var_34944_to_fp16 = const()[name = string("op_34944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3649_cast_fp16, y = var_34944_to_fp16)[name = string("aw_chunk_3649_cast_fp16")]; + fp16 var_34946_to_fp16 = const()[name = string("op_34946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3651_cast_fp16, y = var_34946_to_fp16)[name = string("aw_chunk_3651_cast_fp16")]; + fp16 var_34948_to_fp16 = const()[name = string("op_34948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3653_cast_fp16, y = var_34948_to_fp16)[name = string("aw_chunk_3653_cast_fp16")]; + fp16 var_34950_to_fp16 = const()[name = string("op_34950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3655_cast_fp16, y = var_34950_to_fp16)[name = string("aw_chunk_3655_cast_fp16")]; + fp16 var_34952_to_fp16 = const()[name = string("op_34952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3657_cast_fp16, y = var_34952_to_fp16)[name = string("aw_chunk_3657_cast_fp16")]; + fp16 var_34954_to_fp16 = const()[name = string("op_34954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3659_cast_fp16, y = var_34954_to_fp16)[name = string("aw_chunk_3659_cast_fp16")]; + fp16 var_34956_to_fp16 = const()[name = string("op_34956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3661_cast_fp16, y = var_34956_to_fp16)[name = string("aw_chunk_3661_cast_fp16")]; + fp16 var_34958_to_fp16 = const()[name = string("op_34958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3663_cast_fp16, y = var_34958_to_fp16)[name = string("aw_chunk_3663_cast_fp16")]; + fp16 var_34960_to_fp16 = const()[name = string("op_34960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3665_cast_fp16, y = var_34960_to_fp16)[name = string("aw_chunk_3665_cast_fp16")]; + fp16 var_34962_to_fp16 = const()[name = string("op_34962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3667_cast_fp16, y = var_34962_to_fp16)[name = string("aw_chunk_3667_cast_fp16")]; + fp16 var_34964_to_fp16 = const()[name = string("op_34964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3669_cast_fp16, y = var_34964_to_fp16)[name = string("aw_chunk_3669_cast_fp16")]; + fp16 var_34966_to_fp16 = const()[name = string("op_34966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3671_cast_fp16, y = var_34966_to_fp16)[name = string("aw_chunk_3671_cast_fp16")]; + fp16 var_34968_to_fp16 = const()[name = string("op_34968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3673_cast_fp16, y = var_34968_to_fp16)[name = string("aw_chunk_3673_cast_fp16")]; + fp16 var_34970_to_fp16 = const()[name = string("op_34970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3675_cast_fp16, y = var_34970_to_fp16)[name = string("aw_chunk_3675_cast_fp16")]; + fp16 var_34972_to_fp16 = const()[name = string("op_34972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3677_cast_fp16, y = var_34972_to_fp16)[name = string("aw_chunk_3677_cast_fp16")]; + fp16 var_34974_to_fp16 = const()[name = string("op_34974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3679_cast_fp16, y = var_34974_to_fp16)[name = string("aw_chunk_3679_cast_fp16")]; + tensor var_34976_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3521_cast_fp16)[name = string("op_34976_cast_fp16")]; + tensor var_34977_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3523_cast_fp16)[name = string("op_34977_cast_fp16")]; + tensor var_34978_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3525_cast_fp16)[name = string("op_34978_cast_fp16")]; + tensor var_34979_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3527_cast_fp16)[name = string("op_34979_cast_fp16")]; + tensor var_34980_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3529_cast_fp16)[name = string("op_34980_cast_fp16")]; + tensor var_34981_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3531_cast_fp16)[name = string("op_34981_cast_fp16")]; + tensor var_34982_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3533_cast_fp16)[name = string("op_34982_cast_fp16")]; + tensor var_34983_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3535_cast_fp16)[name = string("op_34983_cast_fp16")]; + tensor var_34984_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3537_cast_fp16)[name = string("op_34984_cast_fp16")]; + tensor var_34985_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3539_cast_fp16)[name = string("op_34985_cast_fp16")]; + tensor var_34986_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3541_cast_fp16)[name = string("op_34986_cast_fp16")]; + tensor var_34987_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3543_cast_fp16)[name = string("op_34987_cast_fp16")]; + tensor var_34988_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3545_cast_fp16)[name = string("op_34988_cast_fp16")]; + tensor var_34989_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3547_cast_fp16)[name = string("op_34989_cast_fp16")]; + tensor var_34990_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3549_cast_fp16)[name = string("op_34990_cast_fp16")]; + tensor var_34991_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3551_cast_fp16)[name = string("op_34991_cast_fp16")]; + tensor var_34992_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3553_cast_fp16)[name = string("op_34992_cast_fp16")]; + tensor var_34993_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3555_cast_fp16)[name = string("op_34993_cast_fp16")]; + tensor var_34994_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3557_cast_fp16)[name = string("op_34994_cast_fp16")]; + tensor var_34995_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3559_cast_fp16)[name = string("op_34995_cast_fp16")]; + tensor var_34996_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3561_cast_fp16)[name = string("op_34996_cast_fp16")]; + tensor var_34997_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3563_cast_fp16)[name = string("op_34997_cast_fp16")]; + tensor var_34998_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3565_cast_fp16)[name = string("op_34998_cast_fp16")]; + tensor var_34999_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3567_cast_fp16)[name = string("op_34999_cast_fp16")]; + tensor var_35000_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3569_cast_fp16)[name = string("op_35000_cast_fp16")]; + tensor var_35001_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3571_cast_fp16)[name = string("op_35001_cast_fp16")]; + tensor var_35002_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3573_cast_fp16)[name = string("op_35002_cast_fp16")]; + tensor var_35003_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3575_cast_fp16)[name = string("op_35003_cast_fp16")]; + tensor var_35004_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3577_cast_fp16)[name = string("op_35004_cast_fp16")]; + tensor var_35005_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3579_cast_fp16)[name = string("op_35005_cast_fp16")]; + tensor var_35006_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3581_cast_fp16)[name = string("op_35006_cast_fp16")]; + tensor var_35007_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3583_cast_fp16)[name = string("op_35007_cast_fp16")]; + tensor var_35008_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3585_cast_fp16)[name = string("op_35008_cast_fp16")]; + tensor var_35009_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3587_cast_fp16)[name = string("op_35009_cast_fp16")]; + tensor var_35010_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3589_cast_fp16)[name = string("op_35010_cast_fp16")]; + tensor var_35011_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3591_cast_fp16)[name = string("op_35011_cast_fp16")]; + tensor var_35012_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3593_cast_fp16)[name = string("op_35012_cast_fp16")]; + tensor var_35013_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3595_cast_fp16)[name = string("op_35013_cast_fp16")]; + tensor var_35014_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3597_cast_fp16)[name = string("op_35014_cast_fp16")]; + tensor var_35015_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3599_cast_fp16)[name = string("op_35015_cast_fp16")]; + tensor var_35016_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3601_cast_fp16)[name = string("op_35016_cast_fp16")]; + tensor var_35017_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3603_cast_fp16)[name = string("op_35017_cast_fp16")]; + tensor var_35018_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3605_cast_fp16)[name = string("op_35018_cast_fp16")]; + tensor var_35019_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3607_cast_fp16)[name = string("op_35019_cast_fp16")]; + tensor var_35020_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3609_cast_fp16)[name = string("op_35020_cast_fp16")]; + tensor var_35021_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3611_cast_fp16)[name = string("op_35021_cast_fp16")]; + tensor var_35022_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3613_cast_fp16)[name = string("op_35022_cast_fp16")]; + tensor var_35023_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3615_cast_fp16)[name = string("op_35023_cast_fp16")]; + tensor var_35024_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3617_cast_fp16)[name = string("op_35024_cast_fp16")]; + tensor var_35025_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3619_cast_fp16)[name = string("op_35025_cast_fp16")]; + tensor var_35026_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3621_cast_fp16)[name = string("op_35026_cast_fp16")]; + tensor var_35027_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3623_cast_fp16)[name = string("op_35027_cast_fp16")]; + tensor var_35028_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3625_cast_fp16)[name = string("op_35028_cast_fp16")]; + tensor var_35029_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3627_cast_fp16)[name = string("op_35029_cast_fp16")]; + tensor var_35030_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3629_cast_fp16)[name = string("op_35030_cast_fp16")]; + tensor var_35031_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3631_cast_fp16)[name = string("op_35031_cast_fp16")]; + tensor var_35032_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3633_cast_fp16)[name = string("op_35032_cast_fp16")]; + tensor var_35033_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3635_cast_fp16)[name = string("op_35033_cast_fp16")]; + tensor var_35034_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3637_cast_fp16)[name = string("op_35034_cast_fp16")]; + tensor var_35035_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3639_cast_fp16)[name = string("op_35035_cast_fp16")]; + tensor var_35036_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3641_cast_fp16)[name = string("op_35036_cast_fp16")]; + tensor var_35037_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3643_cast_fp16)[name = string("op_35037_cast_fp16")]; + tensor var_35038_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3645_cast_fp16)[name = string("op_35038_cast_fp16")]; + tensor var_35039_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3647_cast_fp16)[name = string("op_35039_cast_fp16")]; + tensor var_35040_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3649_cast_fp16)[name = string("op_35040_cast_fp16")]; + tensor var_35041_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3651_cast_fp16)[name = string("op_35041_cast_fp16")]; + tensor var_35042_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3653_cast_fp16)[name = string("op_35042_cast_fp16")]; + tensor var_35043_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3655_cast_fp16)[name = string("op_35043_cast_fp16")]; + tensor var_35044_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3657_cast_fp16)[name = string("op_35044_cast_fp16")]; + tensor var_35045_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3659_cast_fp16)[name = string("op_35045_cast_fp16")]; + tensor var_35046_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3661_cast_fp16)[name = string("op_35046_cast_fp16")]; + tensor var_35047_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3663_cast_fp16)[name = string("op_35047_cast_fp16")]; + tensor var_35048_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3665_cast_fp16)[name = string("op_35048_cast_fp16")]; + tensor var_35049_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3667_cast_fp16)[name = string("op_35049_cast_fp16")]; + tensor var_35050_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3669_cast_fp16)[name = string("op_35050_cast_fp16")]; + tensor var_35051_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3671_cast_fp16)[name = string("op_35051_cast_fp16")]; + tensor var_35052_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3673_cast_fp16)[name = string("op_35052_cast_fp16")]; + tensor var_35053_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3675_cast_fp16)[name = string("op_35053_cast_fp16")]; + tensor var_35054_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3677_cast_fp16)[name = string("op_35054_cast_fp16")]; + tensor var_35055_cast_fp16 = softmax(axis = var_33801, x = aw_chunk_3679_cast_fp16)[name = string("op_35055_cast_fp16")]; + string var_35057_equation_0 = const()[name = string("op_35057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35057_cast_fp16 = einsum(equation = var_35057_equation_0, values = (var_34577_cast_fp16, var_34976_cast_fp16))[name = string("op_35057_cast_fp16")]; + string var_35059_equation_0 = const()[name = string("op_35059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35059_cast_fp16 = einsum(equation = var_35059_equation_0, values = (var_34577_cast_fp16, var_34977_cast_fp16))[name = string("op_35059_cast_fp16")]; + string var_35061_equation_0 = const()[name = string("op_35061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35061_cast_fp16 = einsum(equation = var_35061_equation_0, values = (var_34577_cast_fp16, var_34978_cast_fp16))[name = string("op_35061_cast_fp16")]; + string var_35063_equation_0 = const()[name = string("op_35063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35063_cast_fp16 = einsum(equation = var_35063_equation_0, values = (var_34577_cast_fp16, var_34979_cast_fp16))[name = string("op_35063_cast_fp16")]; + string var_35065_equation_0 = const()[name = string("op_35065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35065_cast_fp16 = einsum(equation = var_35065_equation_0, values = (var_34581_cast_fp16, var_34980_cast_fp16))[name = string("op_35065_cast_fp16")]; + string var_35067_equation_0 = const()[name = string("op_35067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35067_cast_fp16 = einsum(equation = var_35067_equation_0, values = (var_34581_cast_fp16, var_34981_cast_fp16))[name = string("op_35067_cast_fp16")]; + string var_35069_equation_0 = const()[name = string("op_35069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35069_cast_fp16 = einsum(equation = var_35069_equation_0, values = (var_34581_cast_fp16, var_34982_cast_fp16))[name = string("op_35069_cast_fp16")]; + string var_35071_equation_0 = const()[name = string("op_35071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35071_cast_fp16 = einsum(equation = var_35071_equation_0, values = (var_34581_cast_fp16, var_34983_cast_fp16))[name = string("op_35071_cast_fp16")]; + string var_35073_equation_0 = const()[name = string("op_35073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35073_cast_fp16 = einsum(equation = var_35073_equation_0, values = (var_34585_cast_fp16, var_34984_cast_fp16))[name = string("op_35073_cast_fp16")]; + string var_35075_equation_0 = const()[name = string("op_35075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35075_cast_fp16 = einsum(equation = var_35075_equation_0, values = (var_34585_cast_fp16, var_34985_cast_fp16))[name = string("op_35075_cast_fp16")]; + string var_35077_equation_0 = const()[name = string("op_35077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35077_cast_fp16 = einsum(equation = var_35077_equation_0, values = (var_34585_cast_fp16, var_34986_cast_fp16))[name = string("op_35077_cast_fp16")]; + string var_35079_equation_0 = const()[name = string("op_35079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35079_cast_fp16 = einsum(equation = var_35079_equation_0, values = (var_34585_cast_fp16, var_34987_cast_fp16))[name = string("op_35079_cast_fp16")]; + string var_35081_equation_0 = const()[name = string("op_35081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35081_cast_fp16 = einsum(equation = var_35081_equation_0, values = (var_34589_cast_fp16, var_34988_cast_fp16))[name = string("op_35081_cast_fp16")]; + string var_35083_equation_0 = const()[name = string("op_35083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35083_cast_fp16 = einsum(equation = var_35083_equation_0, values = (var_34589_cast_fp16, var_34989_cast_fp16))[name = string("op_35083_cast_fp16")]; + string var_35085_equation_0 = const()[name = string("op_35085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35085_cast_fp16 = einsum(equation = var_35085_equation_0, values = (var_34589_cast_fp16, var_34990_cast_fp16))[name = string("op_35085_cast_fp16")]; + string var_35087_equation_0 = const()[name = string("op_35087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35087_cast_fp16 = einsum(equation = var_35087_equation_0, values = (var_34589_cast_fp16, var_34991_cast_fp16))[name = string("op_35087_cast_fp16")]; + string var_35089_equation_0 = const()[name = string("op_35089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35089_cast_fp16 = einsum(equation = var_35089_equation_0, values = (var_34593_cast_fp16, var_34992_cast_fp16))[name = string("op_35089_cast_fp16")]; + string var_35091_equation_0 = const()[name = string("op_35091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35091_cast_fp16 = einsum(equation = var_35091_equation_0, values = (var_34593_cast_fp16, var_34993_cast_fp16))[name = string("op_35091_cast_fp16")]; + string var_35093_equation_0 = const()[name = string("op_35093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35093_cast_fp16 = einsum(equation = var_35093_equation_0, values = (var_34593_cast_fp16, var_34994_cast_fp16))[name = string("op_35093_cast_fp16")]; + string var_35095_equation_0 = const()[name = string("op_35095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35095_cast_fp16 = einsum(equation = var_35095_equation_0, values = (var_34593_cast_fp16, var_34995_cast_fp16))[name = string("op_35095_cast_fp16")]; + string var_35097_equation_0 = const()[name = string("op_35097_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35097_cast_fp16 = einsum(equation = var_35097_equation_0, values = (var_34597_cast_fp16, var_34996_cast_fp16))[name = string("op_35097_cast_fp16")]; + string var_35099_equation_0 = const()[name = string("op_35099_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35099_cast_fp16 = einsum(equation = var_35099_equation_0, values = (var_34597_cast_fp16, var_34997_cast_fp16))[name = string("op_35099_cast_fp16")]; + string var_35101_equation_0 = const()[name = string("op_35101_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35101_cast_fp16 = einsum(equation = var_35101_equation_0, values = (var_34597_cast_fp16, var_34998_cast_fp16))[name = string("op_35101_cast_fp16")]; + string var_35103_equation_0 = const()[name = string("op_35103_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35103_cast_fp16 = einsum(equation = var_35103_equation_0, values = (var_34597_cast_fp16, var_34999_cast_fp16))[name = string("op_35103_cast_fp16")]; + string var_35105_equation_0 = const()[name = string("op_35105_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35105_cast_fp16 = einsum(equation = var_35105_equation_0, values = (var_34601_cast_fp16, var_35000_cast_fp16))[name = string("op_35105_cast_fp16")]; + string var_35107_equation_0 = const()[name = string("op_35107_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35107_cast_fp16 = einsum(equation = var_35107_equation_0, values = (var_34601_cast_fp16, var_35001_cast_fp16))[name = string("op_35107_cast_fp16")]; + string var_35109_equation_0 = const()[name = string("op_35109_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35109_cast_fp16 = einsum(equation = var_35109_equation_0, values = (var_34601_cast_fp16, var_35002_cast_fp16))[name = string("op_35109_cast_fp16")]; + string var_35111_equation_0 = const()[name = string("op_35111_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35111_cast_fp16 = einsum(equation = var_35111_equation_0, values = (var_34601_cast_fp16, var_35003_cast_fp16))[name = string("op_35111_cast_fp16")]; + string var_35113_equation_0 = const()[name = string("op_35113_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35113_cast_fp16 = einsum(equation = var_35113_equation_0, values = (var_34605_cast_fp16, var_35004_cast_fp16))[name = string("op_35113_cast_fp16")]; + string var_35115_equation_0 = const()[name = string("op_35115_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35115_cast_fp16 = einsum(equation = var_35115_equation_0, values = (var_34605_cast_fp16, var_35005_cast_fp16))[name = string("op_35115_cast_fp16")]; + string var_35117_equation_0 = const()[name = string("op_35117_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35117_cast_fp16 = einsum(equation = var_35117_equation_0, values = (var_34605_cast_fp16, var_35006_cast_fp16))[name = string("op_35117_cast_fp16")]; + string var_35119_equation_0 = const()[name = string("op_35119_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35119_cast_fp16 = einsum(equation = var_35119_equation_0, values = (var_34605_cast_fp16, var_35007_cast_fp16))[name = string("op_35119_cast_fp16")]; + string var_35121_equation_0 = const()[name = string("op_35121_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35121_cast_fp16 = einsum(equation = var_35121_equation_0, values = (var_34609_cast_fp16, var_35008_cast_fp16))[name = string("op_35121_cast_fp16")]; + string var_35123_equation_0 = const()[name = string("op_35123_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35123_cast_fp16 = einsum(equation = var_35123_equation_0, values = (var_34609_cast_fp16, var_35009_cast_fp16))[name = string("op_35123_cast_fp16")]; + string var_35125_equation_0 = const()[name = string("op_35125_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35125_cast_fp16 = einsum(equation = var_35125_equation_0, values = (var_34609_cast_fp16, var_35010_cast_fp16))[name = string("op_35125_cast_fp16")]; + string var_35127_equation_0 = const()[name = string("op_35127_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35127_cast_fp16 = einsum(equation = var_35127_equation_0, values = (var_34609_cast_fp16, var_35011_cast_fp16))[name = string("op_35127_cast_fp16")]; + string var_35129_equation_0 = const()[name = string("op_35129_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35129_cast_fp16 = einsum(equation = var_35129_equation_0, values = (var_34613_cast_fp16, var_35012_cast_fp16))[name = string("op_35129_cast_fp16")]; + string var_35131_equation_0 = const()[name = string("op_35131_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35131_cast_fp16 = einsum(equation = var_35131_equation_0, values = (var_34613_cast_fp16, var_35013_cast_fp16))[name = string("op_35131_cast_fp16")]; + string var_35133_equation_0 = const()[name = string("op_35133_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35133_cast_fp16 = einsum(equation = var_35133_equation_0, values = (var_34613_cast_fp16, var_35014_cast_fp16))[name = string("op_35133_cast_fp16")]; + string var_35135_equation_0 = const()[name = string("op_35135_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35135_cast_fp16 = einsum(equation = var_35135_equation_0, values = (var_34613_cast_fp16, var_35015_cast_fp16))[name = string("op_35135_cast_fp16")]; + string var_35137_equation_0 = const()[name = string("op_35137_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35137_cast_fp16 = einsum(equation = var_35137_equation_0, values = (var_34617_cast_fp16, var_35016_cast_fp16))[name = string("op_35137_cast_fp16")]; + string var_35139_equation_0 = const()[name = string("op_35139_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35139_cast_fp16 = einsum(equation = var_35139_equation_0, values = (var_34617_cast_fp16, var_35017_cast_fp16))[name = string("op_35139_cast_fp16")]; + string var_35141_equation_0 = const()[name = string("op_35141_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35141_cast_fp16 = einsum(equation = var_35141_equation_0, values = (var_34617_cast_fp16, var_35018_cast_fp16))[name = string("op_35141_cast_fp16")]; + string var_35143_equation_0 = const()[name = string("op_35143_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35143_cast_fp16 = einsum(equation = var_35143_equation_0, values = (var_34617_cast_fp16, var_35019_cast_fp16))[name = string("op_35143_cast_fp16")]; + string var_35145_equation_0 = const()[name = string("op_35145_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35145_cast_fp16 = einsum(equation = var_35145_equation_0, values = (var_34621_cast_fp16, var_35020_cast_fp16))[name = string("op_35145_cast_fp16")]; + string var_35147_equation_0 = const()[name = string("op_35147_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35147_cast_fp16 = einsum(equation = var_35147_equation_0, values = (var_34621_cast_fp16, var_35021_cast_fp16))[name = string("op_35147_cast_fp16")]; + string var_35149_equation_0 = const()[name = string("op_35149_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35149_cast_fp16 = einsum(equation = var_35149_equation_0, values = (var_34621_cast_fp16, var_35022_cast_fp16))[name = string("op_35149_cast_fp16")]; + string var_35151_equation_0 = const()[name = string("op_35151_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35151_cast_fp16 = einsum(equation = var_35151_equation_0, values = (var_34621_cast_fp16, var_35023_cast_fp16))[name = string("op_35151_cast_fp16")]; + string var_35153_equation_0 = const()[name = string("op_35153_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35153_cast_fp16 = einsum(equation = var_35153_equation_0, values = (var_34625_cast_fp16, var_35024_cast_fp16))[name = string("op_35153_cast_fp16")]; + string var_35155_equation_0 = const()[name = string("op_35155_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35155_cast_fp16 = einsum(equation = var_35155_equation_0, values = (var_34625_cast_fp16, var_35025_cast_fp16))[name = string("op_35155_cast_fp16")]; + string var_35157_equation_0 = const()[name = string("op_35157_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35157_cast_fp16 = einsum(equation = var_35157_equation_0, values = (var_34625_cast_fp16, var_35026_cast_fp16))[name = string("op_35157_cast_fp16")]; + string var_35159_equation_0 = const()[name = string("op_35159_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35159_cast_fp16 = einsum(equation = var_35159_equation_0, values = (var_34625_cast_fp16, var_35027_cast_fp16))[name = string("op_35159_cast_fp16")]; + string var_35161_equation_0 = const()[name = string("op_35161_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35161_cast_fp16 = einsum(equation = var_35161_equation_0, values = (var_34629_cast_fp16, var_35028_cast_fp16))[name = string("op_35161_cast_fp16")]; + string var_35163_equation_0 = const()[name = string("op_35163_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35163_cast_fp16 = einsum(equation = var_35163_equation_0, values = (var_34629_cast_fp16, var_35029_cast_fp16))[name = string("op_35163_cast_fp16")]; + string var_35165_equation_0 = const()[name = string("op_35165_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35165_cast_fp16 = einsum(equation = var_35165_equation_0, values = (var_34629_cast_fp16, var_35030_cast_fp16))[name = string("op_35165_cast_fp16")]; + string var_35167_equation_0 = const()[name = string("op_35167_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35167_cast_fp16 = einsum(equation = var_35167_equation_0, values = (var_34629_cast_fp16, var_35031_cast_fp16))[name = string("op_35167_cast_fp16")]; + string var_35169_equation_0 = const()[name = string("op_35169_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35169_cast_fp16 = einsum(equation = var_35169_equation_0, values = (var_34633_cast_fp16, var_35032_cast_fp16))[name = string("op_35169_cast_fp16")]; + string var_35171_equation_0 = const()[name = string("op_35171_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35171_cast_fp16 = einsum(equation = var_35171_equation_0, values = (var_34633_cast_fp16, var_35033_cast_fp16))[name = string("op_35171_cast_fp16")]; + string var_35173_equation_0 = const()[name = string("op_35173_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35173_cast_fp16 = einsum(equation = var_35173_equation_0, values = (var_34633_cast_fp16, var_35034_cast_fp16))[name = string("op_35173_cast_fp16")]; + string var_35175_equation_0 = const()[name = string("op_35175_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35175_cast_fp16 = einsum(equation = var_35175_equation_0, values = (var_34633_cast_fp16, var_35035_cast_fp16))[name = string("op_35175_cast_fp16")]; + string var_35177_equation_0 = const()[name = string("op_35177_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35177_cast_fp16 = einsum(equation = var_35177_equation_0, values = (var_34637_cast_fp16, var_35036_cast_fp16))[name = string("op_35177_cast_fp16")]; + string var_35179_equation_0 = const()[name = string("op_35179_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35179_cast_fp16 = einsum(equation = var_35179_equation_0, values = (var_34637_cast_fp16, var_35037_cast_fp16))[name = string("op_35179_cast_fp16")]; + string var_35181_equation_0 = const()[name = string("op_35181_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35181_cast_fp16 = einsum(equation = var_35181_equation_0, values = (var_34637_cast_fp16, var_35038_cast_fp16))[name = string("op_35181_cast_fp16")]; + string var_35183_equation_0 = const()[name = string("op_35183_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35183_cast_fp16 = einsum(equation = var_35183_equation_0, values = (var_34637_cast_fp16, var_35039_cast_fp16))[name = string("op_35183_cast_fp16")]; + string var_35185_equation_0 = const()[name = string("op_35185_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35185_cast_fp16 = einsum(equation = var_35185_equation_0, values = (var_34641_cast_fp16, var_35040_cast_fp16))[name = string("op_35185_cast_fp16")]; + string var_35187_equation_0 = const()[name = string("op_35187_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35187_cast_fp16 = einsum(equation = var_35187_equation_0, values = (var_34641_cast_fp16, var_35041_cast_fp16))[name = string("op_35187_cast_fp16")]; + string var_35189_equation_0 = const()[name = string("op_35189_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35189_cast_fp16 = einsum(equation = var_35189_equation_0, values = (var_34641_cast_fp16, var_35042_cast_fp16))[name = string("op_35189_cast_fp16")]; + string var_35191_equation_0 = const()[name = string("op_35191_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35191_cast_fp16 = einsum(equation = var_35191_equation_0, values = (var_34641_cast_fp16, var_35043_cast_fp16))[name = string("op_35191_cast_fp16")]; + string var_35193_equation_0 = const()[name = string("op_35193_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35193_cast_fp16 = einsum(equation = var_35193_equation_0, values = (var_34645_cast_fp16, var_35044_cast_fp16))[name = string("op_35193_cast_fp16")]; + string var_35195_equation_0 = const()[name = string("op_35195_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35195_cast_fp16 = einsum(equation = var_35195_equation_0, values = (var_34645_cast_fp16, var_35045_cast_fp16))[name = string("op_35195_cast_fp16")]; + string var_35197_equation_0 = const()[name = string("op_35197_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35197_cast_fp16 = einsum(equation = var_35197_equation_0, values = (var_34645_cast_fp16, var_35046_cast_fp16))[name = string("op_35197_cast_fp16")]; + string var_35199_equation_0 = const()[name = string("op_35199_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35199_cast_fp16 = einsum(equation = var_35199_equation_0, values = (var_34645_cast_fp16, var_35047_cast_fp16))[name = string("op_35199_cast_fp16")]; + string var_35201_equation_0 = const()[name = string("op_35201_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35201_cast_fp16 = einsum(equation = var_35201_equation_0, values = (var_34649_cast_fp16, var_35048_cast_fp16))[name = string("op_35201_cast_fp16")]; + string var_35203_equation_0 = const()[name = string("op_35203_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35203_cast_fp16 = einsum(equation = var_35203_equation_0, values = (var_34649_cast_fp16, var_35049_cast_fp16))[name = string("op_35203_cast_fp16")]; + string var_35205_equation_0 = const()[name = string("op_35205_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35205_cast_fp16 = einsum(equation = var_35205_equation_0, values = (var_34649_cast_fp16, var_35050_cast_fp16))[name = string("op_35205_cast_fp16")]; + string var_35207_equation_0 = const()[name = string("op_35207_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35207_cast_fp16 = einsum(equation = var_35207_equation_0, values = (var_34649_cast_fp16, var_35051_cast_fp16))[name = string("op_35207_cast_fp16")]; + string var_35209_equation_0 = const()[name = string("op_35209_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35209_cast_fp16 = einsum(equation = var_35209_equation_0, values = (var_34653_cast_fp16, var_35052_cast_fp16))[name = string("op_35209_cast_fp16")]; + string var_35211_equation_0 = const()[name = string("op_35211_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35211_cast_fp16 = einsum(equation = var_35211_equation_0, values = (var_34653_cast_fp16, var_35053_cast_fp16))[name = string("op_35211_cast_fp16")]; + string var_35213_equation_0 = const()[name = string("op_35213_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35213_cast_fp16 = einsum(equation = var_35213_equation_0, values = (var_34653_cast_fp16, var_35054_cast_fp16))[name = string("op_35213_cast_fp16")]; + string var_35215_equation_0 = const()[name = string("op_35215_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_35215_cast_fp16 = einsum(equation = var_35215_equation_0, values = (var_34653_cast_fp16, var_35055_cast_fp16))[name = string("op_35215_cast_fp16")]; + bool var_35217_interleave_0 = const()[name = string("op_35217_interleave_0"), val = bool(false)]; + tensor var_35217_cast_fp16 = concat(axis = var_33776, interleave = var_35217_interleave_0, values = (var_35057_cast_fp16, var_35059_cast_fp16, var_35061_cast_fp16, var_35063_cast_fp16))[name = string("op_35217_cast_fp16")]; + bool var_35219_interleave_0 = const()[name = string("op_35219_interleave_0"), val = bool(false)]; + tensor var_35219_cast_fp16 = concat(axis = var_33776, interleave = var_35219_interleave_0, values = (var_35065_cast_fp16, var_35067_cast_fp16, var_35069_cast_fp16, var_35071_cast_fp16))[name = string("op_35219_cast_fp16")]; + bool var_35221_interleave_0 = const()[name = string("op_35221_interleave_0"), val = bool(false)]; + tensor var_35221_cast_fp16 = concat(axis = var_33776, interleave = var_35221_interleave_0, values = (var_35073_cast_fp16, var_35075_cast_fp16, var_35077_cast_fp16, var_35079_cast_fp16))[name = string("op_35221_cast_fp16")]; + bool var_35223_interleave_0 = const()[name = string("op_35223_interleave_0"), val = bool(false)]; + tensor var_35223_cast_fp16 = concat(axis = var_33776, interleave = var_35223_interleave_0, values = (var_35081_cast_fp16, var_35083_cast_fp16, var_35085_cast_fp16, var_35087_cast_fp16))[name = string("op_35223_cast_fp16")]; + bool var_35225_interleave_0 = const()[name = string("op_35225_interleave_0"), val = bool(false)]; + tensor var_35225_cast_fp16 = concat(axis = var_33776, interleave = var_35225_interleave_0, values = (var_35089_cast_fp16, var_35091_cast_fp16, var_35093_cast_fp16, var_35095_cast_fp16))[name = string("op_35225_cast_fp16")]; + bool var_35227_interleave_0 = const()[name = string("op_35227_interleave_0"), val = bool(false)]; + tensor var_35227_cast_fp16 = concat(axis = var_33776, interleave = var_35227_interleave_0, values = (var_35097_cast_fp16, var_35099_cast_fp16, var_35101_cast_fp16, var_35103_cast_fp16))[name = string("op_35227_cast_fp16")]; + bool var_35229_interleave_0 = const()[name = string("op_35229_interleave_0"), val = bool(false)]; + tensor var_35229_cast_fp16 = concat(axis = var_33776, interleave = var_35229_interleave_0, values = (var_35105_cast_fp16, var_35107_cast_fp16, var_35109_cast_fp16, var_35111_cast_fp16))[name = string("op_35229_cast_fp16")]; + bool var_35231_interleave_0 = const()[name = string("op_35231_interleave_0"), val = bool(false)]; + tensor var_35231_cast_fp16 = concat(axis = var_33776, interleave = var_35231_interleave_0, values = (var_35113_cast_fp16, var_35115_cast_fp16, var_35117_cast_fp16, var_35119_cast_fp16))[name = string("op_35231_cast_fp16")]; + bool var_35233_interleave_0 = const()[name = string("op_35233_interleave_0"), val = bool(false)]; + tensor var_35233_cast_fp16 = concat(axis = var_33776, interleave = var_35233_interleave_0, values = (var_35121_cast_fp16, var_35123_cast_fp16, var_35125_cast_fp16, var_35127_cast_fp16))[name = string("op_35233_cast_fp16")]; + bool var_35235_interleave_0 = const()[name = string("op_35235_interleave_0"), val = bool(false)]; + tensor var_35235_cast_fp16 = concat(axis = var_33776, interleave = var_35235_interleave_0, values = (var_35129_cast_fp16, var_35131_cast_fp16, var_35133_cast_fp16, var_35135_cast_fp16))[name = string("op_35235_cast_fp16")]; + bool var_35237_interleave_0 = const()[name = string("op_35237_interleave_0"), val = bool(false)]; + tensor var_35237_cast_fp16 = concat(axis = var_33776, interleave = var_35237_interleave_0, values = (var_35137_cast_fp16, var_35139_cast_fp16, var_35141_cast_fp16, var_35143_cast_fp16))[name = string("op_35237_cast_fp16")]; + bool var_35239_interleave_0 = const()[name = string("op_35239_interleave_0"), val = bool(false)]; + tensor var_35239_cast_fp16 = concat(axis = var_33776, interleave = var_35239_interleave_0, values = (var_35145_cast_fp16, var_35147_cast_fp16, var_35149_cast_fp16, var_35151_cast_fp16))[name = string("op_35239_cast_fp16")]; + bool var_35241_interleave_0 = const()[name = string("op_35241_interleave_0"), val = bool(false)]; + tensor var_35241_cast_fp16 = concat(axis = var_33776, interleave = var_35241_interleave_0, values = (var_35153_cast_fp16, var_35155_cast_fp16, var_35157_cast_fp16, var_35159_cast_fp16))[name = string("op_35241_cast_fp16")]; + bool var_35243_interleave_0 = const()[name = string("op_35243_interleave_0"), val = bool(false)]; + tensor var_35243_cast_fp16 = concat(axis = var_33776, interleave = var_35243_interleave_0, values = (var_35161_cast_fp16, var_35163_cast_fp16, var_35165_cast_fp16, var_35167_cast_fp16))[name = string("op_35243_cast_fp16")]; + bool var_35245_interleave_0 = const()[name = string("op_35245_interleave_0"), val = bool(false)]; + tensor var_35245_cast_fp16 = concat(axis = var_33776, interleave = var_35245_interleave_0, values = (var_35169_cast_fp16, var_35171_cast_fp16, var_35173_cast_fp16, var_35175_cast_fp16))[name = string("op_35245_cast_fp16")]; + bool var_35247_interleave_0 = const()[name = string("op_35247_interleave_0"), val = bool(false)]; + tensor var_35247_cast_fp16 = concat(axis = var_33776, interleave = var_35247_interleave_0, values = (var_35177_cast_fp16, var_35179_cast_fp16, var_35181_cast_fp16, var_35183_cast_fp16))[name = string("op_35247_cast_fp16")]; + bool var_35249_interleave_0 = const()[name = string("op_35249_interleave_0"), val = bool(false)]; + tensor var_35249_cast_fp16 = concat(axis = var_33776, interleave = var_35249_interleave_0, values = (var_35185_cast_fp16, var_35187_cast_fp16, var_35189_cast_fp16, var_35191_cast_fp16))[name = string("op_35249_cast_fp16")]; + bool var_35251_interleave_0 = const()[name = string("op_35251_interleave_0"), val = bool(false)]; + tensor var_35251_cast_fp16 = concat(axis = var_33776, interleave = var_35251_interleave_0, values = (var_35193_cast_fp16, var_35195_cast_fp16, var_35197_cast_fp16, var_35199_cast_fp16))[name = string("op_35251_cast_fp16")]; + bool var_35253_interleave_0 = const()[name = string("op_35253_interleave_0"), val = bool(false)]; + tensor var_35253_cast_fp16 = concat(axis = var_33776, interleave = var_35253_interleave_0, values = (var_35201_cast_fp16, var_35203_cast_fp16, var_35205_cast_fp16, var_35207_cast_fp16))[name = string("op_35253_cast_fp16")]; + bool var_35255_interleave_0 = const()[name = string("op_35255_interleave_0"), val = bool(false)]; + tensor var_35255_cast_fp16 = concat(axis = var_33776, interleave = var_35255_interleave_0, values = (var_35209_cast_fp16, var_35211_cast_fp16, var_35213_cast_fp16, var_35215_cast_fp16))[name = string("op_35255_cast_fp16")]; + bool input_177_interleave_0 = const()[name = string("input_177_interleave_0"), val = bool(false)]; + tensor input_177_cast_fp16 = concat(axis = var_33801, interleave = input_177_interleave_0, values = (var_35217_cast_fp16, var_35219_cast_fp16, var_35221_cast_fp16, var_35223_cast_fp16, var_35225_cast_fp16, var_35227_cast_fp16, var_35229_cast_fp16, var_35231_cast_fp16, var_35233_cast_fp16, var_35235_cast_fp16, var_35237_cast_fp16, var_35239_cast_fp16, var_35241_cast_fp16, var_35243_cast_fp16, var_35245_cast_fp16, var_35247_cast_fp16, var_35249_cast_fp16, var_35251_cast_fp16, var_35253_cast_fp16, var_35255_cast_fp16))[name = string("input_177_cast_fp16")]; + string obj_91_pad_type_0 = const()[name = string("obj_91_pad_type_0"), val = string("valid")]; + tensor obj_91_strides_0 = const()[name = string("obj_91_strides_0"), val = tensor([1, 1])]; + tensor obj_91_pad_0 = const()[name = string("obj_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_91_dilations_0 = const()[name = string("obj_91_dilations_0"), val = tensor([1, 1])]; + int32 obj_91_groups_0 = const()[name = string("obj_91_groups_0"), val = int32(1)]; + tensor layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(890277440)))]; + tensor layers_22_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893554304)))]; + tensor obj_91_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_177_cast_fp16)[name = string("obj_91_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = obj_91_cast_fp16)[name = string("inputs_91_cast_fp16")]; + tensor out_91_axes_0 = const()[name = string("out_91_axes_0"), val = tensor([1])]; + fp16 var_35274_to_fp16 = const()[name = string("op_35274_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_35274_to_fp16, x = inputs_91_cast_fp16)[name = string("out_91_cast_fp16")]; + tensor input_179_gamma_0_to_fp16 = const()[name = string("input_179_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893556928)))]; + tensor input_179_beta_0_to_fp16 = const()[name = string("input_179_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893559552)))]; + fp16 input_179_epsilon_0_to_fp16 = const()[name = string("input_179_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_179_cast_fp16 = batch_norm(beta = input_179_beta_0_to_fp16, epsilon = input_179_epsilon_0_to_fp16, gamma = input_179_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = string("input_179_cast_fp16")]; + string input_181_pad_type_0 = const()[name = string("input_181_pad_type_0"), val = string("valid")]; + tensor input_181_strides_0 = const()[name = string("input_181_strides_0"), val = tensor([1, 1])]; + tensor input_181_pad_0 = const()[name = string("input_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_181_dilations_0 = const()[name = string("input_181_dilations_0"), val = tensor([1, 1])]; + int32 input_181_groups_0 = const()[name = string("input_181_groups_0"), val = int32(1)]; + tensor layers_22_fc1_weight_to_fp16 = const()[name = string("layers_22_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(893562176)))]; + tensor layers_22_fc1_bias_to_fp16 = const()[name = string("layers_22_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906669440)))]; + tensor input_181_cast_fp16 = conv(bias = layers_22_fc1_bias_to_fp16, dilations = input_181_dilations_0, groups = input_181_groups_0, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = input_181_strides_0, weight = layers_22_fc1_weight_to_fp16, x = input_179_cast_fp16)[name = string("input_181_cast_fp16")]; + string input_183_mode_0 = const()[name = string("input_183_mode_0"), val = string("EXACT")]; + tensor input_183_cast_fp16 = gelu(mode = input_183_mode_0, x = input_181_cast_fp16)[name = string("input_183_cast_fp16")]; + string hidden_states_49_pad_type_0 = const()[name = string("hidden_states_49_pad_type_0"), val = string("valid")]; + tensor hidden_states_49_strides_0 = const()[name = string("hidden_states_49_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_49_pad_0 = const()[name = string("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_49_dilations_0 = const()[name = string("hidden_states_49_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_49_groups_0 = const()[name = string("hidden_states_49_groups_0"), val = int32(1)]; + tensor layers_22_fc2_weight_to_fp16 = const()[name = string("layers_22_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(906679744)))]; + tensor layers_22_fc2_bias_to_fp16 = const()[name = string("layers_22_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919787008)))]; + tensor hidden_states_49_cast_fp16 = conv(bias = layers_22_fc2_bias_to_fp16, dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = layers_22_fc2_weight_to_fp16, x = input_183_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("inputs_93_cast_fp16")]; + int32 var_35303 = const()[name = string("op_35303"), val = int32(3)]; + int32 var_35328 = const()[name = string("op_35328"), val = int32(1)]; + tensor out_93_axes_0 = const()[name = string("out_93_axes_0"), val = tensor([1])]; + fp16 var_35345_to_fp16 = const()[name = string("op_35345_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_35345_to_fp16, x = inputs_93_cast_fp16)[name = string("out_93_cast_fp16")]; + tensor obj_93_gamma_0_to_fp16 = const()[name = string("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919789632)))]; + tensor obj_93_beta_0_to_fp16 = const()[name = string("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919792256)))]; + fp16 obj_93_epsilon_0_to_fp16 = const()[name = string("obj_93_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = string("obj_93_cast_fp16")]; + string query_47_pad_type_0 = const()[name = string("query_47_pad_type_0"), val = string("valid")]; + tensor query_47_strides_0 = const()[name = string("query_47_strides_0"), val = tensor([1, 1])]; + tensor query_47_pad_0 = const()[name = string("query_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_47_dilations_0 = const()[name = string("query_47_dilations_0"), val = tensor([1, 1])]; + int32 query_47_groups_0 = const()[name = string("query_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(919794880)))]; + tensor layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923071744)))]; + tensor query_47_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_bias_to_fp16, dilations = query_47_dilations_0, groups = query_47_groups_0, pad = query_47_pad_0, pad_type = query_47_pad_type_0, strides = query_47_strides_0, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("query_47_cast_fp16")]; + string key_47_pad_type_0 = const()[name = string("key_47_pad_type_0"), val = string("valid")]; + tensor key_47_strides_0 = const()[name = string("key_47_strides_0"), val = tensor([1, 1])]; + tensor key_47_pad_0 = const()[name = string("key_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_47_dilations_0 = const()[name = string("key_47_dilations_0"), val = tensor([1, 1])]; + int32 key_47_groups_0 = const()[name = string("key_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(923074368)))]; + tensor key_47_cast_fp16 = conv(dilations = key_47_dilations_0, groups = key_47_groups_0, pad = key_47_pad_0, pad_type = key_47_pad_type_0, strides = key_47_strides_0, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("key_47_cast_fp16")]; + string value_47_pad_type_0 = const()[name = string("value_47_pad_type_0"), val = string("valid")]; + tensor value_47_strides_0 = const()[name = string("value_47_strides_0"), val = tensor([1, 1])]; + tensor value_47_pad_0 = const()[name = string("value_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_47_dilations_0 = const()[name = string("value_47_dilations_0"), val = tensor([1, 1])]; + int32 value_47_groups_0 = const()[name = string("value_47_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(926351232)))]; + tensor layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929628096)))]; + tensor value_47_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_bias_to_fp16, dilations = value_47_dilations_0, groups = value_47_groups_0, pad = value_47_pad_0, pad_type = value_47_pad_type_0, strides = value_47_strides_0, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = string("value_47_cast_fp16")]; + tensor var_35383_begin_0 = const()[name = string("op_35383_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35383_end_0 = const()[name = string("op_35383_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35383_end_mask_0 = const()[name = string("op_35383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35383_cast_fp16 = slice_by_index(begin = var_35383_begin_0, end = var_35383_end_0, end_mask = var_35383_end_mask_0, x = query_47_cast_fp16)[name = string("op_35383_cast_fp16")]; + tensor var_35387_begin_0 = const()[name = string("op_35387_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35387_end_0 = const()[name = string("op_35387_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_35387_end_mask_0 = const()[name = string("op_35387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35387_cast_fp16 = slice_by_index(begin = var_35387_begin_0, end = var_35387_end_0, end_mask = var_35387_end_mask_0, x = query_47_cast_fp16)[name = string("op_35387_cast_fp16")]; + tensor var_35391_begin_0 = const()[name = string("op_35391_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35391_end_0 = const()[name = string("op_35391_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_35391_end_mask_0 = const()[name = string("op_35391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35391_cast_fp16 = slice_by_index(begin = var_35391_begin_0, end = var_35391_end_0, end_mask = var_35391_end_mask_0, x = query_47_cast_fp16)[name = string("op_35391_cast_fp16")]; + tensor var_35395_begin_0 = const()[name = string("op_35395_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35395_end_0 = const()[name = string("op_35395_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_35395_end_mask_0 = const()[name = string("op_35395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35395_cast_fp16 = slice_by_index(begin = var_35395_begin_0, end = var_35395_end_0, end_mask = var_35395_end_mask_0, x = query_47_cast_fp16)[name = string("op_35395_cast_fp16")]; + tensor var_35399_begin_0 = const()[name = string("op_35399_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35399_end_0 = const()[name = string("op_35399_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_35399_end_mask_0 = const()[name = string("op_35399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35399_cast_fp16 = slice_by_index(begin = var_35399_begin_0, end = var_35399_end_0, end_mask = var_35399_end_mask_0, x = query_47_cast_fp16)[name = string("op_35399_cast_fp16")]; + tensor var_35403_begin_0 = const()[name = string("op_35403_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35403_end_0 = const()[name = string("op_35403_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_35403_end_mask_0 = const()[name = string("op_35403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35403_cast_fp16 = slice_by_index(begin = var_35403_begin_0, end = var_35403_end_0, end_mask = var_35403_end_mask_0, x = query_47_cast_fp16)[name = string("op_35403_cast_fp16")]; + tensor var_35407_begin_0 = const()[name = string("op_35407_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35407_end_0 = const()[name = string("op_35407_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_35407_end_mask_0 = const()[name = string("op_35407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35407_cast_fp16 = slice_by_index(begin = var_35407_begin_0, end = var_35407_end_0, end_mask = var_35407_end_mask_0, x = query_47_cast_fp16)[name = string("op_35407_cast_fp16")]; + tensor var_35411_begin_0 = const()[name = string("op_35411_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35411_end_0 = const()[name = string("op_35411_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_35411_end_mask_0 = const()[name = string("op_35411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35411_cast_fp16 = slice_by_index(begin = var_35411_begin_0, end = var_35411_end_0, end_mask = var_35411_end_mask_0, x = query_47_cast_fp16)[name = string("op_35411_cast_fp16")]; + tensor var_35415_begin_0 = const()[name = string("op_35415_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35415_end_0 = const()[name = string("op_35415_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_35415_end_mask_0 = const()[name = string("op_35415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35415_cast_fp16 = slice_by_index(begin = var_35415_begin_0, end = var_35415_end_0, end_mask = var_35415_end_mask_0, x = query_47_cast_fp16)[name = string("op_35415_cast_fp16")]; + tensor var_35419_begin_0 = const()[name = string("op_35419_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35419_end_0 = const()[name = string("op_35419_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_35419_end_mask_0 = const()[name = string("op_35419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35419_cast_fp16 = slice_by_index(begin = var_35419_begin_0, end = var_35419_end_0, end_mask = var_35419_end_mask_0, x = query_47_cast_fp16)[name = string("op_35419_cast_fp16")]; + tensor var_35423_begin_0 = const()[name = string("op_35423_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35423_end_0 = const()[name = string("op_35423_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_35423_end_mask_0 = const()[name = string("op_35423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35423_cast_fp16 = slice_by_index(begin = var_35423_begin_0, end = var_35423_end_0, end_mask = var_35423_end_mask_0, x = query_47_cast_fp16)[name = string("op_35423_cast_fp16")]; + tensor var_35427_begin_0 = const()[name = string("op_35427_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35427_end_0 = const()[name = string("op_35427_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_35427_end_mask_0 = const()[name = string("op_35427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35427_cast_fp16 = slice_by_index(begin = var_35427_begin_0, end = var_35427_end_0, end_mask = var_35427_end_mask_0, x = query_47_cast_fp16)[name = string("op_35427_cast_fp16")]; + tensor var_35431_begin_0 = const()[name = string("op_35431_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35431_end_0 = const()[name = string("op_35431_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_35431_end_mask_0 = const()[name = string("op_35431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35431_cast_fp16 = slice_by_index(begin = var_35431_begin_0, end = var_35431_end_0, end_mask = var_35431_end_mask_0, x = query_47_cast_fp16)[name = string("op_35431_cast_fp16")]; + tensor var_35435_begin_0 = const()[name = string("op_35435_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35435_end_0 = const()[name = string("op_35435_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_35435_end_mask_0 = const()[name = string("op_35435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35435_cast_fp16 = slice_by_index(begin = var_35435_begin_0, end = var_35435_end_0, end_mask = var_35435_end_mask_0, x = query_47_cast_fp16)[name = string("op_35435_cast_fp16")]; + tensor var_35439_begin_0 = const()[name = string("op_35439_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35439_end_0 = const()[name = string("op_35439_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_35439_end_mask_0 = const()[name = string("op_35439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35439_cast_fp16 = slice_by_index(begin = var_35439_begin_0, end = var_35439_end_0, end_mask = var_35439_end_mask_0, x = query_47_cast_fp16)[name = string("op_35439_cast_fp16")]; + tensor var_35443_begin_0 = const()[name = string("op_35443_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35443_end_0 = const()[name = string("op_35443_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_35443_end_mask_0 = const()[name = string("op_35443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35443_cast_fp16 = slice_by_index(begin = var_35443_begin_0, end = var_35443_end_0, end_mask = var_35443_end_mask_0, x = query_47_cast_fp16)[name = string("op_35443_cast_fp16")]; + tensor var_35447_begin_0 = const()[name = string("op_35447_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35447_end_0 = const()[name = string("op_35447_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_35447_end_mask_0 = const()[name = string("op_35447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35447_cast_fp16 = slice_by_index(begin = var_35447_begin_0, end = var_35447_end_0, end_mask = var_35447_end_mask_0, x = query_47_cast_fp16)[name = string("op_35447_cast_fp16")]; + tensor var_35451_begin_0 = const()[name = string("op_35451_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35451_end_0 = const()[name = string("op_35451_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_35451_end_mask_0 = const()[name = string("op_35451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35451_cast_fp16 = slice_by_index(begin = var_35451_begin_0, end = var_35451_end_0, end_mask = var_35451_end_mask_0, x = query_47_cast_fp16)[name = string("op_35451_cast_fp16")]; + tensor var_35455_begin_0 = const()[name = string("op_35455_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35455_end_0 = const()[name = string("op_35455_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_35455_end_mask_0 = const()[name = string("op_35455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35455_cast_fp16 = slice_by_index(begin = var_35455_begin_0, end = var_35455_end_0, end_mask = var_35455_end_mask_0, x = query_47_cast_fp16)[name = string("op_35455_cast_fp16")]; + tensor var_35459_begin_0 = const()[name = string("op_35459_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35459_end_0 = const()[name = string("op_35459_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_35459_end_mask_0 = const()[name = string("op_35459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35459_cast_fp16 = slice_by_index(begin = var_35459_begin_0, end = var_35459_end_0, end_mask = var_35459_end_mask_0, x = query_47_cast_fp16)[name = string("op_35459_cast_fp16")]; + tensor var_35468_begin_0 = const()[name = string("op_35468_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35468_end_0 = const()[name = string("op_35468_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35468_end_mask_0 = const()[name = string("op_35468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35468_cast_fp16 = slice_by_index(begin = var_35468_begin_0, end = var_35468_end_0, end_mask = var_35468_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35468_cast_fp16")]; + tensor var_35475_begin_0 = const()[name = string("op_35475_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35475_end_0 = const()[name = string("op_35475_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35475_end_mask_0 = const()[name = string("op_35475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35475_cast_fp16 = slice_by_index(begin = var_35475_begin_0, end = var_35475_end_0, end_mask = var_35475_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35475_cast_fp16")]; + tensor var_35482_begin_0 = const()[name = string("op_35482_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35482_end_0 = const()[name = string("op_35482_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35482_end_mask_0 = const()[name = string("op_35482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35482_cast_fp16 = slice_by_index(begin = var_35482_begin_0, end = var_35482_end_0, end_mask = var_35482_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35482_cast_fp16")]; + tensor var_35489_begin_0 = const()[name = string("op_35489_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35489_end_0 = const()[name = string("op_35489_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35489_end_mask_0 = const()[name = string("op_35489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35489_cast_fp16 = slice_by_index(begin = var_35489_begin_0, end = var_35489_end_0, end_mask = var_35489_end_mask_0, x = var_35383_cast_fp16)[name = string("op_35489_cast_fp16")]; + tensor var_35496_begin_0 = const()[name = string("op_35496_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35496_end_0 = const()[name = string("op_35496_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35496_end_mask_0 = const()[name = string("op_35496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35496_cast_fp16 = slice_by_index(begin = var_35496_begin_0, end = var_35496_end_0, end_mask = var_35496_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35496_cast_fp16")]; + tensor var_35503_begin_0 = const()[name = string("op_35503_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35503_end_0 = const()[name = string("op_35503_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35503_end_mask_0 = const()[name = string("op_35503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35503_cast_fp16 = slice_by_index(begin = var_35503_begin_0, end = var_35503_end_0, end_mask = var_35503_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35503_cast_fp16")]; + tensor var_35510_begin_0 = const()[name = string("op_35510_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35510_end_0 = const()[name = string("op_35510_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35510_end_mask_0 = const()[name = string("op_35510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35510_cast_fp16 = slice_by_index(begin = var_35510_begin_0, end = var_35510_end_0, end_mask = var_35510_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35510_cast_fp16")]; + tensor var_35517_begin_0 = const()[name = string("op_35517_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35517_end_0 = const()[name = string("op_35517_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35517_end_mask_0 = const()[name = string("op_35517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35517_cast_fp16 = slice_by_index(begin = var_35517_begin_0, end = var_35517_end_0, end_mask = var_35517_end_mask_0, x = var_35387_cast_fp16)[name = string("op_35517_cast_fp16")]; + tensor var_35524_begin_0 = const()[name = string("op_35524_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35524_end_0 = const()[name = string("op_35524_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35524_end_mask_0 = const()[name = string("op_35524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35524_cast_fp16 = slice_by_index(begin = var_35524_begin_0, end = var_35524_end_0, end_mask = var_35524_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35524_cast_fp16")]; + tensor var_35531_begin_0 = const()[name = string("op_35531_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35531_end_0 = const()[name = string("op_35531_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35531_end_mask_0 = const()[name = string("op_35531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35531_cast_fp16 = slice_by_index(begin = var_35531_begin_0, end = var_35531_end_0, end_mask = var_35531_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35531_cast_fp16")]; + tensor var_35538_begin_0 = const()[name = string("op_35538_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35538_end_0 = const()[name = string("op_35538_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35538_end_mask_0 = const()[name = string("op_35538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35538_cast_fp16 = slice_by_index(begin = var_35538_begin_0, end = var_35538_end_0, end_mask = var_35538_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35538_cast_fp16")]; + tensor var_35545_begin_0 = const()[name = string("op_35545_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35545_end_0 = const()[name = string("op_35545_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35545_end_mask_0 = const()[name = string("op_35545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35545_cast_fp16 = slice_by_index(begin = var_35545_begin_0, end = var_35545_end_0, end_mask = var_35545_end_mask_0, x = var_35391_cast_fp16)[name = string("op_35545_cast_fp16")]; + tensor var_35552_begin_0 = const()[name = string("op_35552_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35552_end_0 = const()[name = string("op_35552_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35552_end_mask_0 = const()[name = string("op_35552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35552_cast_fp16 = slice_by_index(begin = var_35552_begin_0, end = var_35552_end_0, end_mask = var_35552_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35552_cast_fp16")]; + tensor var_35559_begin_0 = const()[name = string("op_35559_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35559_end_0 = const()[name = string("op_35559_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35559_end_mask_0 = const()[name = string("op_35559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35559_cast_fp16 = slice_by_index(begin = var_35559_begin_0, end = var_35559_end_0, end_mask = var_35559_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35559_cast_fp16")]; + tensor var_35566_begin_0 = const()[name = string("op_35566_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35566_end_0 = const()[name = string("op_35566_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35566_end_mask_0 = const()[name = string("op_35566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35566_cast_fp16 = slice_by_index(begin = var_35566_begin_0, end = var_35566_end_0, end_mask = var_35566_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35566_cast_fp16")]; + tensor var_35573_begin_0 = const()[name = string("op_35573_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35573_end_0 = const()[name = string("op_35573_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35573_end_mask_0 = const()[name = string("op_35573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35573_cast_fp16 = slice_by_index(begin = var_35573_begin_0, end = var_35573_end_0, end_mask = var_35573_end_mask_0, x = var_35395_cast_fp16)[name = string("op_35573_cast_fp16")]; + tensor var_35580_begin_0 = const()[name = string("op_35580_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35580_end_0 = const()[name = string("op_35580_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35580_end_mask_0 = const()[name = string("op_35580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35580_cast_fp16 = slice_by_index(begin = var_35580_begin_0, end = var_35580_end_0, end_mask = var_35580_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35580_cast_fp16")]; + tensor var_35587_begin_0 = const()[name = string("op_35587_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35587_end_0 = const()[name = string("op_35587_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35587_end_mask_0 = const()[name = string("op_35587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35587_cast_fp16 = slice_by_index(begin = var_35587_begin_0, end = var_35587_end_0, end_mask = var_35587_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35587_cast_fp16")]; + tensor var_35594_begin_0 = const()[name = string("op_35594_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35594_end_0 = const()[name = string("op_35594_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35594_end_mask_0 = const()[name = string("op_35594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35594_cast_fp16 = slice_by_index(begin = var_35594_begin_0, end = var_35594_end_0, end_mask = var_35594_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35594_cast_fp16")]; + tensor var_35601_begin_0 = const()[name = string("op_35601_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35601_end_0 = const()[name = string("op_35601_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35601_end_mask_0 = const()[name = string("op_35601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35601_cast_fp16 = slice_by_index(begin = var_35601_begin_0, end = var_35601_end_0, end_mask = var_35601_end_mask_0, x = var_35399_cast_fp16)[name = string("op_35601_cast_fp16")]; + tensor var_35608_begin_0 = const()[name = string("op_35608_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35608_end_0 = const()[name = string("op_35608_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35608_end_mask_0 = const()[name = string("op_35608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35608_cast_fp16 = slice_by_index(begin = var_35608_begin_0, end = var_35608_end_0, end_mask = var_35608_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35608_cast_fp16")]; + tensor var_35615_begin_0 = const()[name = string("op_35615_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35615_end_0 = const()[name = string("op_35615_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35615_end_mask_0 = const()[name = string("op_35615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35615_cast_fp16 = slice_by_index(begin = var_35615_begin_0, end = var_35615_end_0, end_mask = var_35615_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35615_cast_fp16")]; + tensor var_35622_begin_0 = const()[name = string("op_35622_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35622_end_0 = const()[name = string("op_35622_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35622_end_mask_0 = const()[name = string("op_35622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35622_cast_fp16 = slice_by_index(begin = var_35622_begin_0, end = var_35622_end_0, end_mask = var_35622_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35622_cast_fp16")]; + tensor var_35629_begin_0 = const()[name = string("op_35629_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35629_end_0 = const()[name = string("op_35629_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35629_end_mask_0 = const()[name = string("op_35629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35629_cast_fp16 = slice_by_index(begin = var_35629_begin_0, end = var_35629_end_0, end_mask = var_35629_end_mask_0, x = var_35403_cast_fp16)[name = string("op_35629_cast_fp16")]; + tensor var_35636_begin_0 = const()[name = string("op_35636_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35636_end_0 = const()[name = string("op_35636_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35636_end_mask_0 = const()[name = string("op_35636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35636_cast_fp16 = slice_by_index(begin = var_35636_begin_0, end = var_35636_end_0, end_mask = var_35636_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35636_cast_fp16")]; + tensor var_35643_begin_0 = const()[name = string("op_35643_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35643_end_0 = const()[name = string("op_35643_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35643_end_mask_0 = const()[name = string("op_35643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35643_cast_fp16 = slice_by_index(begin = var_35643_begin_0, end = var_35643_end_0, end_mask = var_35643_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35643_cast_fp16")]; + tensor var_35650_begin_0 = const()[name = string("op_35650_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35650_end_0 = const()[name = string("op_35650_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35650_end_mask_0 = const()[name = string("op_35650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35650_cast_fp16 = slice_by_index(begin = var_35650_begin_0, end = var_35650_end_0, end_mask = var_35650_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35650_cast_fp16")]; + tensor var_35657_begin_0 = const()[name = string("op_35657_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35657_end_0 = const()[name = string("op_35657_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35657_end_mask_0 = const()[name = string("op_35657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35657_cast_fp16 = slice_by_index(begin = var_35657_begin_0, end = var_35657_end_0, end_mask = var_35657_end_mask_0, x = var_35407_cast_fp16)[name = string("op_35657_cast_fp16")]; + tensor var_35664_begin_0 = const()[name = string("op_35664_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35664_end_0 = const()[name = string("op_35664_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35664_end_mask_0 = const()[name = string("op_35664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35664_cast_fp16 = slice_by_index(begin = var_35664_begin_0, end = var_35664_end_0, end_mask = var_35664_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35664_cast_fp16")]; + tensor var_35671_begin_0 = const()[name = string("op_35671_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35671_end_0 = const()[name = string("op_35671_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35671_end_mask_0 = const()[name = string("op_35671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35671_cast_fp16 = slice_by_index(begin = var_35671_begin_0, end = var_35671_end_0, end_mask = var_35671_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35671_cast_fp16")]; + tensor var_35678_begin_0 = const()[name = string("op_35678_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35678_end_0 = const()[name = string("op_35678_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35678_end_mask_0 = const()[name = string("op_35678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35678_cast_fp16 = slice_by_index(begin = var_35678_begin_0, end = var_35678_end_0, end_mask = var_35678_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35678_cast_fp16")]; + tensor var_35685_begin_0 = const()[name = string("op_35685_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35685_end_0 = const()[name = string("op_35685_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35685_end_mask_0 = const()[name = string("op_35685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35685_cast_fp16 = slice_by_index(begin = var_35685_begin_0, end = var_35685_end_0, end_mask = var_35685_end_mask_0, x = var_35411_cast_fp16)[name = string("op_35685_cast_fp16")]; + tensor var_35692_begin_0 = const()[name = string("op_35692_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35692_end_0 = const()[name = string("op_35692_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35692_end_mask_0 = const()[name = string("op_35692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35692_cast_fp16 = slice_by_index(begin = var_35692_begin_0, end = var_35692_end_0, end_mask = var_35692_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35692_cast_fp16")]; + tensor var_35699_begin_0 = const()[name = string("op_35699_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35699_end_0 = const()[name = string("op_35699_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35699_end_mask_0 = const()[name = string("op_35699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35699_cast_fp16 = slice_by_index(begin = var_35699_begin_0, end = var_35699_end_0, end_mask = var_35699_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35699_cast_fp16")]; + tensor var_35706_begin_0 = const()[name = string("op_35706_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35706_end_0 = const()[name = string("op_35706_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35706_end_mask_0 = const()[name = string("op_35706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35706_cast_fp16 = slice_by_index(begin = var_35706_begin_0, end = var_35706_end_0, end_mask = var_35706_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35706_cast_fp16")]; + tensor var_35713_begin_0 = const()[name = string("op_35713_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35713_end_0 = const()[name = string("op_35713_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35713_end_mask_0 = const()[name = string("op_35713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35713_cast_fp16 = slice_by_index(begin = var_35713_begin_0, end = var_35713_end_0, end_mask = var_35713_end_mask_0, x = var_35415_cast_fp16)[name = string("op_35713_cast_fp16")]; + tensor var_35720_begin_0 = const()[name = string("op_35720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35720_end_0 = const()[name = string("op_35720_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35720_end_mask_0 = const()[name = string("op_35720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35720_cast_fp16 = slice_by_index(begin = var_35720_begin_0, end = var_35720_end_0, end_mask = var_35720_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35720_cast_fp16")]; + tensor var_35727_begin_0 = const()[name = string("op_35727_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35727_end_0 = const()[name = string("op_35727_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35727_end_mask_0 = const()[name = string("op_35727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35727_cast_fp16 = slice_by_index(begin = var_35727_begin_0, end = var_35727_end_0, end_mask = var_35727_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35727_cast_fp16")]; + tensor var_35734_begin_0 = const()[name = string("op_35734_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35734_end_0 = const()[name = string("op_35734_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35734_end_mask_0 = const()[name = string("op_35734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35734_cast_fp16 = slice_by_index(begin = var_35734_begin_0, end = var_35734_end_0, end_mask = var_35734_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35734_cast_fp16")]; + tensor var_35741_begin_0 = const()[name = string("op_35741_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35741_end_0 = const()[name = string("op_35741_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35741_end_mask_0 = const()[name = string("op_35741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35741_cast_fp16 = slice_by_index(begin = var_35741_begin_0, end = var_35741_end_0, end_mask = var_35741_end_mask_0, x = var_35419_cast_fp16)[name = string("op_35741_cast_fp16")]; + tensor var_35748_begin_0 = const()[name = string("op_35748_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35748_end_0 = const()[name = string("op_35748_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35748_end_mask_0 = const()[name = string("op_35748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35748_cast_fp16 = slice_by_index(begin = var_35748_begin_0, end = var_35748_end_0, end_mask = var_35748_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35748_cast_fp16")]; + tensor var_35755_begin_0 = const()[name = string("op_35755_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35755_end_0 = const()[name = string("op_35755_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35755_end_mask_0 = const()[name = string("op_35755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35755_cast_fp16 = slice_by_index(begin = var_35755_begin_0, end = var_35755_end_0, end_mask = var_35755_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35755_cast_fp16")]; + tensor var_35762_begin_0 = const()[name = string("op_35762_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35762_end_0 = const()[name = string("op_35762_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35762_end_mask_0 = const()[name = string("op_35762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35762_cast_fp16 = slice_by_index(begin = var_35762_begin_0, end = var_35762_end_0, end_mask = var_35762_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35762_cast_fp16")]; + tensor var_35769_begin_0 = const()[name = string("op_35769_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35769_end_0 = const()[name = string("op_35769_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35769_end_mask_0 = const()[name = string("op_35769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35769_cast_fp16 = slice_by_index(begin = var_35769_begin_0, end = var_35769_end_0, end_mask = var_35769_end_mask_0, x = var_35423_cast_fp16)[name = string("op_35769_cast_fp16")]; + tensor var_35776_begin_0 = const()[name = string("op_35776_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35776_end_0 = const()[name = string("op_35776_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35776_end_mask_0 = const()[name = string("op_35776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35776_cast_fp16 = slice_by_index(begin = var_35776_begin_0, end = var_35776_end_0, end_mask = var_35776_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35776_cast_fp16")]; + tensor var_35783_begin_0 = const()[name = string("op_35783_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35783_end_0 = const()[name = string("op_35783_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35783_end_mask_0 = const()[name = string("op_35783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35783_cast_fp16 = slice_by_index(begin = var_35783_begin_0, end = var_35783_end_0, end_mask = var_35783_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35783_cast_fp16")]; + tensor var_35790_begin_0 = const()[name = string("op_35790_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35790_end_0 = const()[name = string("op_35790_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35790_end_mask_0 = const()[name = string("op_35790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35790_cast_fp16 = slice_by_index(begin = var_35790_begin_0, end = var_35790_end_0, end_mask = var_35790_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35790_cast_fp16")]; + tensor var_35797_begin_0 = const()[name = string("op_35797_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35797_end_0 = const()[name = string("op_35797_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35797_end_mask_0 = const()[name = string("op_35797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35797_cast_fp16 = slice_by_index(begin = var_35797_begin_0, end = var_35797_end_0, end_mask = var_35797_end_mask_0, x = var_35427_cast_fp16)[name = string("op_35797_cast_fp16")]; + tensor var_35804_begin_0 = const()[name = string("op_35804_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35804_end_0 = const()[name = string("op_35804_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35804_end_mask_0 = const()[name = string("op_35804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35804_cast_fp16 = slice_by_index(begin = var_35804_begin_0, end = var_35804_end_0, end_mask = var_35804_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35804_cast_fp16")]; + tensor var_35811_begin_0 = const()[name = string("op_35811_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35811_end_0 = const()[name = string("op_35811_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35811_end_mask_0 = const()[name = string("op_35811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35811_cast_fp16 = slice_by_index(begin = var_35811_begin_0, end = var_35811_end_0, end_mask = var_35811_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35811_cast_fp16")]; + tensor var_35818_begin_0 = const()[name = string("op_35818_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35818_end_0 = const()[name = string("op_35818_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35818_end_mask_0 = const()[name = string("op_35818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35818_cast_fp16 = slice_by_index(begin = var_35818_begin_0, end = var_35818_end_0, end_mask = var_35818_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35818_cast_fp16")]; + tensor var_35825_begin_0 = const()[name = string("op_35825_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35825_end_0 = const()[name = string("op_35825_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35825_end_mask_0 = const()[name = string("op_35825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35825_cast_fp16 = slice_by_index(begin = var_35825_begin_0, end = var_35825_end_0, end_mask = var_35825_end_mask_0, x = var_35431_cast_fp16)[name = string("op_35825_cast_fp16")]; + tensor var_35832_begin_0 = const()[name = string("op_35832_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35832_end_0 = const()[name = string("op_35832_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35832_end_mask_0 = const()[name = string("op_35832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35832_cast_fp16 = slice_by_index(begin = var_35832_begin_0, end = var_35832_end_0, end_mask = var_35832_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35832_cast_fp16")]; + tensor var_35839_begin_0 = const()[name = string("op_35839_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35839_end_0 = const()[name = string("op_35839_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35839_end_mask_0 = const()[name = string("op_35839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35839_cast_fp16 = slice_by_index(begin = var_35839_begin_0, end = var_35839_end_0, end_mask = var_35839_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35839_cast_fp16")]; + tensor var_35846_begin_0 = const()[name = string("op_35846_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35846_end_0 = const()[name = string("op_35846_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35846_end_mask_0 = const()[name = string("op_35846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35846_cast_fp16 = slice_by_index(begin = var_35846_begin_0, end = var_35846_end_0, end_mask = var_35846_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35846_cast_fp16")]; + tensor var_35853_begin_0 = const()[name = string("op_35853_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35853_end_0 = const()[name = string("op_35853_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35853_end_mask_0 = const()[name = string("op_35853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35853_cast_fp16 = slice_by_index(begin = var_35853_begin_0, end = var_35853_end_0, end_mask = var_35853_end_mask_0, x = var_35435_cast_fp16)[name = string("op_35853_cast_fp16")]; + tensor var_35860_begin_0 = const()[name = string("op_35860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35860_end_0 = const()[name = string("op_35860_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35860_end_mask_0 = const()[name = string("op_35860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35860_cast_fp16 = slice_by_index(begin = var_35860_begin_0, end = var_35860_end_0, end_mask = var_35860_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35860_cast_fp16")]; + tensor var_35867_begin_0 = const()[name = string("op_35867_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35867_end_0 = const()[name = string("op_35867_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35867_end_mask_0 = const()[name = string("op_35867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35867_cast_fp16 = slice_by_index(begin = var_35867_begin_0, end = var_35867_end_0, end_mask = var_35867_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35867_cast_fp16")]; + tensor var_35874_begin_0 = const()[name = string("op_35874_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35874_end_0 = const()[name = string("op_35874_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35874_end_mask_0 = const()[name = string("op_35874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35874_cast_fp16 = slice_by_index(begin = var_35874_begin_0, end = var_35874_end_0, end_mask = var_35874_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35874_cast_fp16")]; + tensor var_35881_begin_0 = const()[name = string("op_35881_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35881_end_0 = const()[name = string("op_35881_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35881_end_mask_0 = const()[name = string("op_35881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35881_cast_fp16 = slice_by_index(begin = var_35881_begin_0, end = var_35881_end_0, end_mask = var_35881_end_mask_0, x = var_35439_cast_fp16)[name = string("op_35881_cast_fp16")]; + tensor var_35888_begin_0 = const()[name = string("op_35888_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35888_end_0 = const()[name = string("op_35888_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35888_end_mask_0 = const()[name = string("op_35888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35888_cast_fp16 = slice_by_index(begin = var_35888_begin_0, end = var_35888_end_0, end_mask = var_35888_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35888_cast_fp16")]; + tensor var_35895_begin_0 = const()[name = string("op_35895_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35895_end_0 = const()[name = string("op_35895_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35895_end_mask_0 = const()[name = string("op_35895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35895_cast_fp16 = slice_by_index(begin = var_35895_begin_0, end = var_35895_end_0, end_mask = var_35895_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35895_cast_fp16")]; + tensor var_35902_begin_0 = const()[name = string("op_35902_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35902_end_0 = const()[name = string("op_35902_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35902_end_mask_0 = const()[name = string("op_35902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35902_cast_fp16 = slice_by_index(begin = var_35902_begin_0, end = var_35902_end_0, end_mask = var_35902_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35902_cast_fp16")]; + tensor var_35909_begin_0 = const()[name = string("op_35909_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35909_end_0 = const()[name = string("op_35909_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35909_end_mask_0 = const()[name = string("op_35909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35909_cast_fp16 = slice_by_index(begin = var_35909_begin_0, end = var_35909_end_0, end_mask = var_35909_end_mask_0, x = var_35443_cast_fp16)[name = string("op_35909_cast_fp16")]; + tensor var_35916_begin_0 = const()[name = string("op_35916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35916_end_0 = const()[name = string("op_35916_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35916_end_mask_0 = const()[name = string("op_35916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35916_cast_fp16 = slice_by_index(begin = var_35916_begin_0, end = var_35916_end_0, end_mask = var_35916_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35916_cast_fp16")]; + tensor var_35923_begin_0 = const()[name = string("op_35923_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35923_end_0 = const()[name = string("op_35923_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35923_end_mask_0 = const()[name = string("op_35923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35923_cast_fp16 = slice_by_index(begin = var_35923_begin_0, end = var_35923_end_0, end_mask = var_35923_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35923_cast_fp16")]; + tensor var_35930_begin_0 = const()[name = string("op_35930_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35930_end_0 = const()[name = string("op_35930_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35930_end_mask_0 = const()[name = string("op_35930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35930_cast_fp16 = slice_by_index(begin = var_35930_begin_0, end = var_35930_end_0, end_mask = var_35930_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35930_cast_fp16")]; + tensor var_35937_begin_0 = const()[name = string("op_35937_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35937_end_0 = const()[name = string("op_35937_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35937_end_mask_0 = const()[name = string("op_35937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35937_cast_fp16 = slice_by_index(begin = var_35937_begin_0, end = var_35937_end_0, end_mask = var_35937_end_mask_0, x = var_35447_cast_fp16)[name = string("op_35937_cast_fp16")]; + tensor var_35944_begin_0 = const()[name = string("op_35944_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35944_end_0 = const()[name = string("op_35944_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35944_end_mask_0 = const()[name = string("op_35944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35944_cast_fp16 = slice_by_index(begin = var_35944_begin_0, end = var_35944_end_0, end_mask = var_35944_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35944_cast_fp16")]; + tensor var_35951_begin_0 = const()[name = string("op_35951_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35951_end_0 = const()[name = string("op_35951_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35951_end_mask_0 = const()[name = string("op_35951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35951_cast_fp16 = slice_by_index(begin = var_35951_begin_0, end = var_35951_end_0, end_mask = var_35951_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35951_cast_fp16")]; + tensor var_35958_begin_0 = const()[name = string("op_35958_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35958_end_0 = const()[name = string("op_35958_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35958_end_mask_0 = const()[name = string("op_35958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35958_cast_fp16 = slice_by_index(begin = var_35958_begin_0, end = var_35958_end_0, end_mask = var_35958_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35958_cast_fp16")]; + tensor var_35965_begin_0 = const()[name = string("op_35965_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35965_end_0 = const()[name = string("op_35965_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35965_end_mask_0 = const()[name = string("op_35965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35965_cast_fp16 = slice_by_index(begin = var_35965_begin_0, end = var_35965_end_0, end_mask = var_35965_end_mask_0, x = var_35451_cast_fp16)[name = string("op_35965_cast_fp16")]; + tensor var_35972_begin_0 = const()[name = string("op_35972_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35972_end_0 = const()[name = string("op_35972_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35972_end_mask_0 = const()[name = string("op_35972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35972_cast_fp16 = slice_by_index(begin = var_35972_begin_0, end = var_35972_end_0, end_mask = var_35972_end_mask_0, x = var_35455_cast_fp16)[name = string("op_35972_cast_fp16")]; + tensor var_35979_begin_0 = const()[name = string("op_35979_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35979_end_0 = const()[name = string("op_35979_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35979_end_mask_0 = const()[name = string("op_35979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35979_cast_fp16 = slice_by_index(begin = var_35979_begin_0, end = var_35979_end_0, end_mask = var_35979_end_mask_0, x = var_35455_cast_fp16)[name = string("op_35979_cast_fp16")]; + tensor var_35986_begin_0 = const()[name = string("op_35986_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35986_end_0 = const()[name = string("op_35986_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35986_end_mask_0 = const()[name = string("op_35986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35986_cast_fp16 = slice_by_index(begin = var_35986_begin_0, end = var_35986_end_0, end_mask = var_35986_end_mask_0, x = var_35455_cast_fp16)[name = string("op_35986_cast_fp16")]; + tensor var_35993_begin_0 = const()[name = string("op_35993_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35993_end_0 = const()[name = string("op_35993_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35993_end_mask_0 = const()[name = string("op_35993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35993_cast_fp16 = slice_by_index(begin = var_35993_begin_0, end = var_35993_end_0, end_mask = var_35993_end_mask_0, x = var_35455_cast_fp16)[name = string("op_35993_cast_fp16")]; + tensor var_36000_begin_0 = const()[name = string("op_36000_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36000_end_0 = const()[name = string("op_36000_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36000_end_mask_0 = const()[name = string("op_36000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36000_cast_fp16 = slice_by_index(begin = var_36000_begin_0, end = var_36000_end_0, end_mask = var_36000_end_mask_0, x = var_35459_cast_fp16)[name = string("op_36000_cast_fp16")]; + tensor var_36007_begin_0 = const()[name = string("op_36007_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36007_end_0 = const()[name = string("op_36007_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36007_end_mask_0 = const()[name = string("op_36007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36007_cast_fp16 = slice_by_index(begin = var_36007_begin_0, end = var_36007_end_0, end_mask = var_36007_end_mask_0, x = var_35459_cast_fp16)[name = string("op_36007_cast_fp16")]; + tensor var_36014_begin_0 = const()[name = string("op_36014_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36014_end_0 = const()[name = string("op_36014_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36014_end_mask_0 = const()[name = string("op_36014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36014_cast_fp16 = slice_by_index(begin = var_36014_begin_0, end = var_36014_end_0, end_mask = var_36014_end_mask_0, x = var_35459_cast_fp16)[name = string("op_36014_cast_fp16")]; + tensor var_36021_begin_0 = const()[name = string("op_36021_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36021_end_0 = const()[name = string("op_36021_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36021_end_mask_0 = const()[name = string("op_36021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36021_cast_fp16 = slice_by_index(begin = var_36021_begin_0, end = var_36021_end_0, end_mask = var_36021_end_mask_0, x = var_35459_cast_fp16)[name = string("op_36021_cast_fp16")]; + tensor k_47_perm_0 = const()[name = string("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_36026_begin_0 = const()[name = string("op_36026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36026_end_0 = const()[name = string("op_36026_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_36026_end_mask_0 = const()[name = string("op_36026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_47_cast_fp16 = transpose(perm = k_47_perm_0, x = key_47_cast_fp16)[name = string("transpose_8")]; + tensor var_36026_cast_fp16 = slice_by_index(begin = var_36026_begin_0, end = var_36026_end_0, end_mask = var_36026_end_mask_0, x = k_47_cast_fp16)[name = string("op_36026_cast_fp16")]; + tensor var_36030_begin_0 = const()[name = string("op_36030_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_36030_end_0 = const()[name = string("op_36030_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_36030_end_mask_0 = const()[name = string("op_36030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36030_cast_fp16 = slice_by_index(begin = var_36030_begin_0, end = var_36030_end_0, end_mask = var_36030_end_mask_0, x = k_47_cast_fp16)[name = string("op_36030_cast_fp16")]; + tensor var_36034_begin_0 = const()[name = string("op_36034_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_36034_end_0 = const()[name = string("op_36034_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_36034_end_mask_0 = const()[name = string("op_36034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36034_cast_fp16 = slice_by_index(begin = var_36034_begin_0, end = var_36034_end_0, end_mask = var_36034_end_mask_0, x = k_47_cast_fp16)[name = string("op_36034_cast_fp16")]; + tensor var_36038_begin_0 = const()[name = string("op_36038_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_36038_end_0 = const()[name = string("op_36038_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_36038_end_mask_0 = const()[name = string("op_36038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36038_cast_fp16 = slice_by_index(begin = var_36038_begin_0, end = var_36038_end_0, end_mask = var_36038_end_mask_0, x = k_47_cast_fp16)[name = string("op_36038_cast_fp16")]; + tensor var_36042_begin_0 = const()[name = string("op_36042_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_36042_end_0 = const()[name = string("op_36042_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_36042_end_mask_0 = const()[name = string("op_36042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36042_cast_fp16 = slice_by_index(begin = var_36042_begin_0, end = var_36042_end_0, end_mask = var_36042_end_mask_0, x = k_47_cast_fp16)[name = string("op_36042_cast_fp16")]; + tensor var_36046_begin_0 = const()[name = string("op_36046_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_36046_end_0 = const()[name = string("op_36046_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_36046_end_mask_0 = const()[name = string("op_36046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36046_cast_fp16 = slice_by_index(begin = var_36046_begin_0, end = var_36046_end_0, end_mask = var_36046_end_mask_0, x = k_47_cast_fp16)[name = string("op_36046_cast_fp16")]; + tensor var_36050_begin_0 = const()[name = string("op_36050_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_36050_end_0 = const()[name = string("op_36050_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_36050_end_mask_0 = const()[name = string("op_36050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36050_cast_fp16 = slice_by_index(begin = var_36050_begin_0, end = var_36050_end_0, end_mask = var_36050_end_mask_0, x = k_47_cast_fp16)[name = string("op_36050_cast_fp16")]; + tensor var_36054_begin_0 = const()[name = string("op_36054_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_36054_end_0 = const()[name = string("op_36054_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_36054_end_mask_0 = const()[name = string("op_36054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36054_cast_fp16 = slice_by_index(begin = var_36054_begin_0, end = var_36054_end_0, end_mask = var_36054_end_mask_0, x = k_47_cast_fp16)[name = string("op_36054_cast_fp16")]; + tensor var_36058_begin_0 = const()[name = string("op_36058_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_36058_end_0 = const()[name = string("op_36058_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_36058_end_mask_0 = const()[name = string("op_36058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36058_cast_fp16 = slice_by_index(begin = var_36058_begin_0, end = var_36058_end_0, end_mask = var_36058_end_mask_0, x = k_47_cast_fp16)[name = string("op_36058_cast_fp16")]; + tensor var_36062_begin_0 = const()[name = string("op_36062_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_36062_end_0 = const()[name = string("op_36062_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_36062_end_mask_0 = const()[name = string("op_36062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36062_cast_fp16 = slice_by_index(begin = var_36062_begin_0, end = var_36062_end_0, end_mask = var_36062_end_mask_0, x = k_47_cast_fp16)[name = string("op_36062_cast_fp16")]; + tensor var_36066_begin_0 = const()[name = string("op_36066_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_36066_end_0 = const()[name = string("op_36066_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_36066_end_mask_0 = const()[name = string("op_36066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36066_cast_fp16 = slice_by_index(begin = var_36066_begin_0, end = var_36066_end_0, end_mask = var_36066_end_mask_0, x = k_47_cast_fp16)[name = string("op_36066_cast_fp16")]; + tensor var_36070_begin_0 = const()[name = string("op_36070_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_36070_end_0 = const()[name = string("op_36070_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_36070_end_mask_0 = const()[name = string("op_36070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36070_cast_fp16 = slice_by_index(begin = var_36070_begin_0, end = var_36070_end_0, end_mask = var_36070_end_mask_0, x = k_47_cast_fp16)[name = string("op_36070_cast_fp16")]; + tensor var_36074_begin_0 = const()[name = string("op_36074_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_36074_end_0 = const()[name = string("op_36074_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_36074_end_mask_0 = const()[name = string("op_36074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36074_cast_fp16 = slice_by_index(begin = var_36074_begin_0, end = var_36074_end_0, end_mask = var_36074_end_mask_0, x = k_47_cast_fp16)[name = string("op_36074_cast_fp16")]; + tensor var_36078_begin_0 = const()[name = string("op_36078_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_36078_end_0 = const()[name = string("op_36078_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_36078_end_mask_0 = const()[name = string("op_36078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36078_cast_fp16 = slice_by_index(begin = var_36078_begin_0, end = var_36078_end_0, end_mask = var_36078_end_mask_0, x = k_47_cast_fp16)[name = string("op_36078_cast_fp16")]; + tensor var_36082_begin_0 = const()[name = string("op_36082_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_36082_end_0 = const()[name = string("op_36082_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_36082_end_mask_0 = const()[name = string("op_36082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36082_cast_fp16 = slice_by_index(begin = var_36082_begin_0, end = var_36082_end_0, end_mask = var_36082_end_mask_0, x = k_47_cast_fp16)[name = string("op_36082_cast_fp16")]; + tensor var_36086_begin_0 = const()[name = string("op_36086_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_36086_end_0 = const()[name = string("op_36086_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_36086_end_mask_0 = const()[name = string("op_36086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36086_cast_fp16 = slice_by_index(begin = var_36086_begin_0, end = var_36086_end_0, end_mask = var_36086_end_mask_0, x = k_47_cast_fp16)[name = string("op_36086_cast_fp16")]; + tensor var_36090_begin_0 = const()[name = string("op_36090_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_36090_end_0 = const()[name = string("op_36090_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_36090_end_mask_0 = const()[name = string("op_36090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36090_cast_fp16 = slice_by_index(begin = var_36090_begin_0, end = var_36090_end_0, end_mask = var_36090_end_mask_0, x = k_47_cast_fp16)[name = string("op_36090_cast_fp16")]; + tensor var_36094_begin_0 = const()[name = string("op_36094_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_36094_end_0 = const()[name = string("op_36094_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_36094_end_mask_0 = const()[name = string("op_36094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36094_cast_fp16 = slice_by_index(begin = var_36094_begin_0, end = var_36094_end_0, end_mask = var_36094_end_mask_0, x = k_47_cast_fp16)[name = string("op_36094_cast_fp16")]; + tensor var_36098_begin_0 = const()[name = string("op_36098_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_36098_end_0 = const()[name = string("op_36098_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_36098_end_mask_0 = const()[name = string("op_36098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36098_cast_fp16 = slice_by_index(begin = var_36098_begin_0, end = var_36098_end_0, end_mask = var_36098_end_mask_0, x = k_47_cast_fp16)[name = string("op_36098_cast_fp16")]; + tensor var_36102_begin_0 = const()[name = string("op_36102_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_36102_end_0 = const()[name = string("op_36102_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_36102_end_mask_0 = const()[name = string("op_36102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36102_cast_fp16 = slice_by_index(begin = var_36102_begin_0, end = var_36102_end_0, end_mask = var_36102_end_mask_0, x = k_47_cast_fp16)[name = string("op_36102_cast_fp16")]; + tensor var_36104_begin_0 = const()[name = string("op_36104_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36104_end_0 = const()[name = string("op_36104_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36104_end_mask_0 = const()[name = string("op_36104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36104_cast_fp16 = slice_by_index(begin = var_36104_begin_0, end = var_36104_end_0, end_mask = var_36104_end_mask_0, x = value_47_cast_fp16)[name = string("op_36104_cast_fp16")]; + tensor var_36108_begin_0 = const()[name = string("op_36108_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36108_end_0 = const()[name = string("op_36108_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_36108_end_mask_0 = const()[name = string("op_36108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36108_cast_fp16 = slice_by_index(begin = var_36108_begin_0, end = var_36108_end_0, end_mask = var_36108_end_mask_0, x = value_47_cast_fp16)[name = string("op_36108_cast_fp16")]; + tensor var_36112_begin_0 = const()[name = string("op_36112_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36112_end_0 = const()[name = string("op_36112_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_36112_end_mask_0 = const()[name = string("op_36112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36112_cast_fp16 = slice_by_index(begin = var_36112_begin_0, end = var_36112_end_0, end_mask = var_36112_end_mask_0, x = value_47_cast_fp16)[name = string("op_36112_cast_fp16")]; + tensor var_36116_begin_0 = const()[name = string("op_36116_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36116_end_0 = const()[name = string("op_36116_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_36116_end_mask_0 = const()[name = string("op_36116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36116_cast_fp16 = slice_by_index(begin = var_36116_begin_0, end = var_36116_end_0, end_mask = var_36116_end_mask_0, x = value_47_cast_fp16)[name = string("op_36116_cast_fp16")]; + tensor var_36120_begin_0 = const()[name = string("op_36120_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36120_end_0 = const()[name = string("op_36120_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_36120_end_mask_0 = const()[name = string("op_36120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36120_cast_fp16 = slice_by_index(begin = var_36120_begin_0, end = var_36120_end_0, end_mask = var_36120_end_mask_0, x = value_47_cast_fp16)[name = string("op_36120_cast_fp16")]; + tensor var_36124_begin_0 = const()[name = string("op_36124_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36124_end_0 = const()[name = string("op_36124_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_36124_end_mask_0 = const()[name = string("op_36124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36124_cast_fp16 = slice_by_index(begin = var_36124_begin_0, end = var_36124_end_0, end_mask = var_36124_end_mask_0, x = value_47_cast_fp16)[name = string("op_36124_cast_fp16")]; + tensor var_36128_begin_0 = const()[name = string("op_36128_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36128_end_0 = const()[name = string("op_36128_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_36128_end_mask_0 = const()[name = string("op_36128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36128_cast_fp16 = slice_by_index(begin = var_36128_begin_0, end = var_36128_end_0, end_mask = var_36128_end_mask_0, x = value_47_cast_fp16)[name = string("op_36128_cast_fp16")]; + tensor var_36132_begin_0 = const()[name = string("op_36132_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36132_end_0 = const()[name = string("op_36132_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_36132_end_mask_0 = const()[name = string("op_36132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36132_cast_fp16 = slice_by_index(begin = var_36132_begin_0, end = var_36132_end_0, end_mask = var_36132_end_mask_0, x = value_47_cast_fp16)[name = string("op_36132_cast_fp16")]; + tensor var_36136_begin_0 = const()[name = string("op_36136_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36136_end_0 = const()[name = string("op_36136_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_36136_end_mask_0 = const()[name = string("op_36136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36136_cast_fp16 = slice_by_index(begin = var_36136_begin_0, end = var_36136_end_0, end_mask = var_36136_end_mask_0, x = value_47_cast_fp16)[name = string("op_36136_cast_fp16")]; + tensor var_36140_begin_0 = const()[name = string("op_36140_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36140_end_0 = const()[name = string("op_36140_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_36140_end_mask_0 = const()[name = string("op_36140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36140_cast_fp16 = slice_by_index(begin = var_36140_begin_0, end = var_36140_end_0, end_mask = var_36140_end_mask_0, x = value_47_cast_fp16)[name = string("op_36140_cast_fp16")]; + tensor var_36144_begin_0 = const()[name = string("op_36144_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36144_end_0 = const()[name = string("op_36144_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_36144_end_mask_0 = const()[name = string("op_36144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36144_cast_fp16 = slice_by_index(begin = var_36144_begin_0, end = var_36144_end_0, end_mask = var_36144_end_mask_0, x = value_47_cast_fp16)[name = string("op_36144_cast_fp16")]; + tensor var_36148_begin_0 = const()[name = string("op_36148_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36148_end_0 = const()[name = string("op_36148_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_36148_end_mask_0 = const()[name = string("op_36148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36148_cast_fp16 = slice_by_index(begin = var_36148_begin_0, end = var_36148_end_0, end_mask = var_36148_end_mask_0, x = value_47_cast_fp16)[name = string("op_36148_cast_fp16")]; + tensor var_36152_begin_0 = const()[name = string("op_36152_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36152_end_0 = const()[name = string("op_36152_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_36152_end_mask_0 = const()[name = string("op_36152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36152_cast_fp16 = slice_by_index(begin = var_36152_begin_0, end = var_36152_end_0, end_mask = var_36152_end_mask_0, x = value_47_cast_fp16)[name = string("op_36152_cast_fp16")]; + tensor var_36156_begin_0 = const()[name = string("op_36156_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36156_end_0 = const()[name = string("op_36156_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_36156_end_mask_0 = const()[name = string("op_36156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36156_cast_fp16 = slice_by_index(begin = var_36156_begin_0, end = var_36156_end_0, end_mask = var_36156_end_mask_0, x = value_47_cast_fp16)[name = string("op_36156_cast_fp16")]; + tensor var_36160_begin_0 = const()[name = string("op_36160_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36160_end_0 = const()[name = string("op_36160_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_36160_end_mask_0 = const()[name = string("op_36160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36160_cast_fp16 = slice_by_index(begin = var_36160_begin_0, end = var_36160_end_0, end_mask = var_36160_end_mask_0, x = value_47_cast_fp16)[name = string("op_36160_cast_fp16")]; + tensor var_36164_begin_0 = const()[name = string("op_36164_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36164_end_0 = const()[name = string("op_36164_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_36164_end_mask_0 = const()[name = string("op_36164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36164_cast_fp16 = slice_by_index(begin = var_36164_begin_0, end = var_36164_end_0, end_mask = var_36164_end_mask_0, x = value_47_cast_fp16)[name = string("op_36164_cast_fp16")]; + tensor var_36168_begin_0 = const()[name = string("op_36168_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36168_end_0 = const()[name = string("op_36168_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_36168_end_mask_0 = const()[name = string("op_36168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36168_cast_fp16 = slice_by_index(begin = var_36168_begin_0, end = var_36168_end_0, end_mask = var_36168_end_mask_0, x = value_47_cast_fp16)[name = string("op_36168_cast_fp16")]; + tensor var_36172_begin_0 = const()[name = string("op_36172_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36172_end_0 = const()[name = string("op_36172_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_36172_end_mask_0 = const()[name = string("op_36172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36172_cast_fp16 = slice_by_index(begin = var_36172_begin_0, end = var_36172_end_0, end_mask = var_36172_end_mask_0, x = value_47_cast_fp16)[name = string("op_36172_cast_fp16")]; + tensor var_36176_begin_0 = const()[name = string("op_36176_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36176_end_0 = const()[name = string("op_36176_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_36176_end_mask_0 = const()[name = string("op_36176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36176_cast_fp16 = slice_by_index(begin = var_36176_begin_0, end = var_36176_end_0, end_mask = var_36176_end_mask_0, x = value_47_cast_fp16)[name = string("op_36176_cast_fp16")]; + tensor var_36180_begin_0 = const()[name = string("op_36180_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36180_end_0 = const()[name = string("op_36180_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_36180_end_mask_0 = const()[name = string("op_36180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36180_cast_fp16 = slice_by_index(begin = var_36180_begin_0, end = var_36180_end_0, end_mask = var_36180_end_mask_0, x = value_47_cast_fp16)[name = string("op_36180_cast_fp16")]; + string _SplitHeadsQ__mh_w_3681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3681_equation_0, values = (var_36026_cast_fp16, var_35468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3681_cast_fp16")]; + string _SplitHeadsQ__mh_w_3683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3683_equation_0, values = (var_36026_cast_fp16, var_35475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3683_cast_fp16")]; + string _SplitHeadsQ__mh_w_3685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3685_equation_0, values = (var_36026_cast_fp16, var_35482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3685_cast_fp16")]; + string _SplitHeadsQ__mh_w_3687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3687_equation_0, values = (var_36026_cast_fp16, var_35489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3687_cast_fp16")]; + string _SplitHeadsQ__mh_w_3689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3689_equation_0, values = (var_36030_cast_fp16, var_35496_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3689_cast_fp16")]; + string _SplitHeadsQ__mh_w_3691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3691_equation_0, values = (var_36030_cast_fp16, var_35503_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3691_cast_fp16")]; + string _SplitHeadsQ__mh_w_3693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3693_equation_0, values = (var_36030_cast_fp16, var_35510_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3693_cast_fp16")]; + string _SplitHeadsQ__mh_w_3695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3695_equation_0, values = (var_36030_cast_fp16, var_35517_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3695_cast_fp16")]; + string _SplitHeadsQ__mh_w_3697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3697_equation_0, values = (var_36034_cast_fp16, var_35524_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3697_cast_fp16")]; + string _SplitHeadsQ__mh_w_3699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3699_equation_0, values = (var_36034_cast_fp16, var_35531_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3699_cast_fp16")]; + string _SplitHeadsQ__mh_w_3701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3701_equation_0, values = (var_36034_cast_fp16, var_35538_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3701_cast_fp16")]; + string _SplitHeadsQ__mh_w_3703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3703_equation_0, values = (var_36034_cast_fp16, var_35545_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3703_cast_fp16")]; + string _SplitHeadsQ__mh_w_3705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3705_equation_0, values = (var_36038_cast_fp16, var_35552_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3705_cast_fp16")]; + string _SplitHeadsQ__mh_w_3707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3707_equation_0, values = (var_36038_cast_fp16, var_35559_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3707_cast_fp16")]; + string _SplitHeadsQ__mh_w_3709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3709_equation_0, values = (var_36038_cast_fp16, var_35566_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3709_cast_fp16")]; + string _SplitHeadsQ__mh_w_3711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3711_equation_0, values = (var_36038_cast_fp16, var_35573_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3711_cast_fp16")]; + string _SplitHeadsQ__mh_w_3713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3713_equation_0, values = (var_36042_cast_fp16, var_35580_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3713_cast_fp16")]; + string _SplitHeadsQ__mh_w_3715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3715_equation_0, values = (var_36042_cast_fp16, var_35587_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3715_cast_fp16")]; + string _SplitHeadsQ__mh_w_3717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3717_equation_0, values = (var_36042_cast_fp16, var_35594_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3717_cast_fp16")]; + string _SplitHeadsQ__mh_w_3719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3719_equation_0, values = (var_36042_cast_fp16, var_35601_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3719_cast_fp16")]; + string _SplitHeadsQ__mh_w_3721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3721_equation_0, values = (var_36046_cast_fp16, var_35608_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3721_cast_fp16")]; + string _SplitHeadsQ__mh_w_3723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3723_equation_0, values = (var_36046_cast_fp16, var_35615_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3723_cast_fp16")]; + string _SplitHeadsQ__mh_w_3725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3725_equation_0, values = (var_36046_cast_fp16, var_35622_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3725_cast_fp16")]; + string _SplitHeadsQ__mh_w_3727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3727_equation_0, values = (var_36046_cast_fp16, var_35629_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3727_cast_fp16")]; + string _SplitHeadsQ__mh_w_3729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3729_equation_0, values = (var_36050_cast_fp16, var_35636_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3729_cast_fp16")]; + string _SplitHeadsQ__mh_w_3731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3731_equation_0, values = (var_36050_cast_fp16, var_35643_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3731_cast_fp16")]; + string _SplitHeadsQ__mh_w_3733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3733_equation_0, values = (var_36050_cast_fp16, var_35650_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3733_cast_fp16")]; + string _SplitHeadsQ__mh_w_3735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3735_equation_0, values = (var_36050_cast_fp16, var_35657_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3735_cast_fp16")]; + string _SplitHeadsQ__mh_w_3737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3737_equation_0, values = (var_36054_cast_fp16, var_35664_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3737_cast_fp16")]; + string _SplitHeadsQ__mh_w_3739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3739_equation_0, values = (var_36054_cast_fp16, var_35671_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3739_cast_fp16")]; + string _SplitHeadsQ__mh_w_3741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3741_equation_0, values = (var_36054_cast_fp16, var_35678_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3741_cast_fp16")]; + string _SplitHeadsQ__mh_w_3743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3743_equation_0, values = (var_36054_cast_fp16, var_35685_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3743_cast_fp16")]; + string _SplitHeadsQ__mh_w_3745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3745_equation_0, values = (var_36058_cast_fp16, var_35692_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3745_cast_fp16")]; + string _SplitHeadsQ__mh_w_3747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3747_equation_0, values = (var_36058_cast_fp16, var_35699_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3747_cast_fp16")]; + string _SplitHeadsQ__mh_w_3749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3749_equation_0, values = (var_36058_cast_fp16, var_35706_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3749_cast_fp16")]; + string _SplitHeadsQ__mh_w_3751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3751_equation_0, values = (var_36058_cast_fp16, var_35713_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3751_cast_fp16")]; + string _SplitHeadsQ__mh_w_3753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3753_equation_0, values = (var_36062_cast_fp16, var_35720_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3753_cast_fp16")]; + string _SplitHeadsQ__mh_w_3755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3755_equation_0, values = (var_36062_cast_fp16, var_35727_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3755_cast_fp16")]; + string _SplitHeadsQ__mh_w_3757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3757_equation_0, values = (var_36062_cast_fp16, var_35734_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3757_cast_fp16")]; + string _SplitHeadsQ__mh_w_3759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3759_equation_0, values = (var_36062_cast_fp16, var_35741_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3759_cast_fp16")]; + string _SplitHeadsQ__mh_w_3761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3761_equation_0, values = (var_36066_cast_fp16, var_35748_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3761_cast_fp16")]; + string _SplitHeadsQ__mh_w_3763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3763_equation_0, values = (var_36066_cast_fp16, var_35755_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3763_cast_fp16")]; + string _SplitHeadsQ__mh_w_3765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3765_equation_0, values = (var_36066_cast_fp16, var_35762_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3765_cast_fp16")]; + string _SplitHeadsQ__mh_w_3767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3767_equation_0, values = (var_36066_cast_fp16, var_35769_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3767_cast_fp16")]; + string _SplitHeadsQ__mh_w_3769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3769_equation_0, values = (var_36070_cast_fp16, var_35776_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3769_cast_fp16")]; + string _SplitHeadsQ__mh_w_3771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3771_equation_0, values = (var_36070_cast_fp16, var_35783_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3771_cast_fp16")]; + string _SplitHeadsQ__mh_w_3773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3773_equation_0, values = (var_36070_cast_fp16, var_35790_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3773_cast_fp16")]; + string _SplitHeadsQ__mh_w_3775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3775_equation_0, values = (var_36070_cast_fp16, var_35797_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3775_cast_fp16")]; + string _SplitHeadsQ__mh_w_3777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3777_equation_0, values = (var_36074_cast_fp16, var_35804_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3777_cast_fp16")]; + string _SplitHeadsQ__mh_w_3779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3779_equation_0, values = (var_36074_cast_fp16, var_35811_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3779_cast_fp16")]; + string _SplitHeadsQ__mh_w_3781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3781_equation_0, values = (var_36074_cast_fp16, var_35818_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3781_cast_fp16")]; + string _SplitHeadsQ__mh_w_3783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3783_equation_0, values = (var_36074_cast_fp16, var_35825_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3783_cast_fp16")]; + string _SplitHeadsQ__mh_w_3785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3785_equation_0, values = (var_36078_cast_fp16, var_35832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3785_cast_fp16")]; + string _SplitHeadsQ__mh_w_3787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3787_equation_0, values = (var_36078_cast_fp16, var_35839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3787_cast_fp16")]; + string _SplitHeadsQ__mh_w_3789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3789_equation_0, values = (var_36078_cast_fp16, var_35846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3789_cast_fp16")]; + string _SplitHeadsQ__mh_w_3791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3791_equation_0, values = (var_36078_cast_fp16, var_35853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3791_cast_fp16")]; + string _SplitHeadsQ__mh_w_3793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3793_equation_0, values = (var_36082_cast_fp16, var_35860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3793_cast_fp16")]; + string _SplitHeadsQ__mh_w_3795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3795_equation_0, values = (var_36082_cast_fp16, var_35867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3795_cast_fp16")]; + string _SplitHeadsQ__mh_w_3797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3797_equation_0, values = (var_36082_cast_fp16, var_35874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3797_cast_fp16")]; + string _SplitHeadsQ__mh_w_3799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3799_equation_0, values = (var_36082_cast_fp16, var_35881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3799_cast_fp16")]; + string _SplitHeadsQ__mh_w_3801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3801_equation_0, values = (var_36086_cast_fp16, var_35888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3801_cast_fp16")]; + string _SplitHeadsQ__mh_w_3803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3803_equation_0, values = (var_36086_cast_fp16, var_35895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3803_cast_fp16")]; + string _SplitHeadsQ__mh_w_3805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3805_equation_0, values = (var_36086_cast_fp16, var_35902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3805_cast_fp16")]; + string _SplitHeadsQ__mh_w_3807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3807_equation_0, values = (var_36086_cast_fp16, var_35909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3807_cast_fp16")]; + string _SplitHeadsQ__mh_w_3809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3809_equation_0, values = (var_36090_cast_fp16, var_35916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3809_cast_fp16")]; + string _SplitHeadsQ__mh_w_3811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3811_equation_0, values = (var_36090_cast_fp16, var_35923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3811_cast_fp16")]; + string _SplitHeadsQ__mh_w_3813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3813_equation_0, values = (var_36090_cast_fp16, var_35930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3813_cast_fp16")]; + string _SplitHeadsQ__mh_w_3815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3815_equation_0, values = (var_36090_cast_fp16, var_35937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3815_cast_fp16")]; + string _SplitHeadsQ__mh_w_3817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3817_equation_0, values = (var_36094_cast_fp16, var_35944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3817_cast_fp16")]; + string _SplitHeadsQ__mh_w_3819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3819_equation_0, values = (var_36094_cast_fp16, var_35951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3819_cast_fp16")]; + string _SplitHeadsQ__mh_w_3821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3821_equation_0, values = (var_36094_cast_fp16, var_35958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3821_cast_fp16")]; + string _SplitHeadsQ__mh_w_3823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3823_equation_0, values = (var_36094_cast_fp16, var_35965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3823_cast_fp16")]; + string _SplitHeadsQ__mh_w_3825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3825_equation_0, values = (var_36098_cast_fp16, var_35972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3825_cast_fp16")]; + string _SplitHeadsQ__mh_w_3827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3827_equation_0, values = (var_36098_cast_fp16, var_35979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3827_cast_fp16")]; + string _SplitHeadsQ__mh_w_3829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3829_equation_0, values = (var_36098_cast_fp16, var_35986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3829_cast_fp16")]; + string _SplitHeadsQ__mh_w_3831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3831_equation_0, values = (var_36098_cast_fp16, var_35993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3831_cast_fp16")]; + string _SplitHeadsQ__mh_w_3833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3833_equation_0, values = (var_36102_cast_fp16, var_36000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3833_cast_fp16")]; + string _SplitHeadsQ__mh_w_3835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3835_equation_0, values = (var_36102_cast_fp16, var_36007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3835_cast_fp16")]; + string _SplitHeadsQ__mh_w_3837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3837_equation_0, values = (var_36102_cast_fp16, var_36014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3837_cast_fp16")]; + string _SplitHeadsQ__mh_w_3839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3839_equation_0, values = (var_36102_cast_fp16, var_36021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3839_cast_fp16")]; + fp16 var_36343_to_fp16 = const()[name = string("op_36343_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3681_cast_fp16, y = var_36343_to_fp16)[name = string("aw_chunk_3681_cast_fp16")]; + fp16 var_36345_to_fp16 = const()[name = string("op_36345_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3683_cast_fp16, y = var_36345_to_fp16)[name = string("aw_chunk_3683_cast_fp16")]; + fp16 var_36347_to_fp16 = const()[name = string("op_36347_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3685_cast_fp16, y = var_36347_to_fp16)[name = string("aw_chunk_3685_cast_fp16")]; + fp16 var_36349_to_fp16 = const()[name = string("op_36349_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3687_cast_fp16, y = var_36349_to_fp16)[name = string("aw_chunk_3687_cast_fp16")]; + fp16 var_36351_to_fp16 = const()[name = string("op_36351_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3689_cast_fp16, y = var_36351_to_fp16)[name = string("aw_chunk_3689_cast_fp16")]; + fp16 var_36353_to_fp16 = const()[name = string("op_36353_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3691_cast_fp16, y = var_36353_to_fp16)[name = string("aw_chunk_3691_cast_fp16")]; + fp16 var_36355_to_fp16 = const()[name = string("op_36355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3693_cast_fp16, y = var_36355_to_fp16)[name = string("aw_chunk_3693_cast_fp16")]; + fp16 var_36357_to_fp16 = const()[name = string("op_36357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3695_cast_fp16, y = var_36357_to_fp16)[name = string("aw_chunk_3695_cast_fp16")]; + fp16 var_36359_to_fp16 = const()[name = string("op_36359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3697_cast_fp16, y = var_36359_to_fp16)[name = string("aw_chunk_3697_cast_fp16")]; + fp16 var_36361_to_fp16 = const()[name = string("op_36361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3699_cast_fp16, y = var_36361_to_fp16)[name = string("aw_chunk_3699_cast_fp16")]; + fp16 var_36363_to_fp16 = const()[name = string("op_36363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3701_cast_fp16, y = var_36363_to_fp16)[name = string("aw_chunk_3701_cast_fp16")]; + fp16 var_36365_to_fp16 = const()[name = string("op_36365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3703_cast_fp16, y = var_36365_to_fp16)[name = string("aw_chunk_3703_cast_fp16")]; + fp16 var_36367_to_fp16 = const()[name = string("op_36367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3705_cast_fp16, y = var_36367_to_fp16)[name = string("aw_chunk_3705_cast_fp16")]; + fp16 var_36369_to_fp16 = const()[name = string("op_36369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3707_cast_fp16, y = var_36369_to_fp16)[name = string("aw_chunk_3707_cast_fp16")]; + fp16 var_36371_to_fp16 = const()[name = string("op_36371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3709_cast_fp16, y = var_36371_to_fp16)[name = string("aw_chunk_3709_cast_fp16")]; + fp16 var_36373_to_fp16 = const()[name = string("op_36373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3711_cast_fp16, y = var_36373_to_fp16)[name = string("aw_chunk_3711_cast_fp16")]; + fp16 var_36375_to_fp16 = const()[name = string("op_36375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3713_cast_fp16, y = var_36375_to_fp16)[name = string("aw_chunk_3713_cast_fp16")]; + fp16 var_36377_to_fp16 = const()[name = string("op_36377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3715_cast_fp16, y = var_36377_to_fp16)[name = string("aw_chunk_3715_cast_fp16")]; + fp16 var_36379_to_fp16 = const()[name = string("op_36379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3717_cast_fp16, y = var_36379_to_fp16)[name = string("aw_chunk_3717_cast_fp16")]; + fp16 var_36381_to_fp16 = const()[name = string("op_36381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3719_cast_fp16, y = var_36381_to_fp16)[name = string("aw_chunk_3719_cast_fp16")]; + fp16 var_36383_to_fp16 = const()[name = string("op_36383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3721_cast_fp16, y = var_36383_to_fp16)[name = string("aw_chunk_3721_cast_fp16")]; + fp16 var_36385_to_fp16 = const()[name = string("op_36385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3723_cast_fp16, y = var_36385_to_fp16)[name = string("aw_chunk_3723_cast_fp16")]; + fp16 var_36387_to_fp16 = const()[name = string("op_36387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3725_cast_fp16, y = var_36387_to_fp16)[name = string("aw_chunk_3725_cast_fp16")]; + fp16 var_36389_to_fp16 = const()[name = string("op_36389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3727_cast_fp16, y = var_36389_to_fp16)[name = string("aw_chunk_3727_cast_fp16")]; + fp16 var_36391_to_fp16 = const()[name = string("op_36391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3729_cast_fp16, y = var_36391_to_fp16)[name = string("aw_chunk_3729_cast_fp16")]; + fp16 var_36393_to_fp16 = const()[name = string("op_36393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3731_cast_fp16, y = var_36393_to_fp16)[name = string("aw_chunk_3731_cast_fp16")]; + fp16 var_36395_to_fp16 = const()[name = string("op_36395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3733_cast_fp16, y = var_36395_to_fp16)[name = string("aw_chunk_3733_cast_fp16")]; + fp16 var_36397_to_fp16 = const()[name = string("op_36397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3735_cast_fp16, y = var_36397_to_fp16)[name = string("aw_chunk_3735_cast_fp16")]; + fp16 var_36399_to_fp16 = const()[name = string("op_36399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3737_cast_fp16, y = var_36399_to_fp16)[name = string("aw_chunk_3737_cast_fp16")]; + fp16 var_36401_to_fp16 = const()[name = string("op_36401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3739_cast_fp16, y = var_36401_to_fp16)[name = string("aw_chunk_3739_cast_fp16")]; + fp16 var_36403_to_fp16 = const()[name = string("op_36403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3741_cast_fp16, y = var_36403_to_fp16)[name = string("aw_chunk_3741_cast_fp16")]; + fp16 var_36405_to_fp16 = const()[name = string("op_36405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3743_cast_fp16, y = var_36405_to_fp16)[name = string("aw_chunk_3743_cast_fp16")]; + fp16 var_36407_to_fp16 = const()[name = string("op_36407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3745_cast_fp16, y = var_36407_to_fp16)[name = string("aw_chunk_3745_cast_fp16")]; + fp16 var_36409_to_fp16 = const()[name = string("op_36409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3747_cast_fp16, y = var_36409_to_fp16)[name = string("aw_chunk_3747_cast_fp16")]; + fp16 var_36411_to_fp16 = const()[name = string("op_36411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3749_cast_fp16, y = var_36411_to_fp16)[name = string("aw_chunk_3749_cast_fp16")]; + fp16 var_36413_to_fp16 = const()[name = string("op_36413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3751_cast_fp16, y = var_36413_to_fp16)[name = string("aw_chunk_3751_cast_fp16")]; + fp16 var_36415_to_fp16 = const()[name = string("op_36415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3753_cast_fp16, y = var_36415_to_fp16)[name = string("aw_chunk_3753_cast_fp16")]; + fp16 var_36417_to_fp16 = const()[name = string("op_36417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3755_cast_fp16, y = var_36417_to_fp16)[name = string("aw_chunk_3755_cast_fp16")]; + fp16 var_36419_to_fp16 = const()[name = string("op_36419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3757_cast_fp16, y = var_36419_to_fp16)[name = string("aw_chunk_3757_cast_fp16")]; + fp16 var_36421_to_fp16 = const()[name = string("op_36421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3759_cast_fp16, y = var_36421_to_fp16)[name = string("aw_chunk_3759_cast_fp16")]; + fp16 var_36423_to_fp16 = const()[name = string("op_36423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3761_cast_fp16, y = var_36423_to_fp16)[name = string("aw_chunk_3761_cast_fp16")]; + fp16 var_36425_to_fp16 = const()[name = string("op_36425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3763_cast_fp16, y = var_36425_to_fp16)[name = string("aw_chunk_3763_cast_fp16")]; + fp16 var_36427_to_fp16 = const()[name = string("op_36427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3765_cast_fp16, y = var_36427_to_fp16)[name = string("aw_chunk_3765_cast_fp16")]; + fp16 var_36429_to_fp16 = const()[name = string("op_36429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3767_cast_fp16, y = var_36429_to_fp16)[name = string("aw_chunk_3767_cast_fp16")]; + fp16 var_36431_to_fp16 = const()[name = string("op_36431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3769_cast_fp16, y = var_36431_to_fp16)[name = string("aw_chunk_3769_cast_fp16")]; + fp16 var_36433_to_fp16 = const()[name = string("op_36433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3771_cast_fp16, y = var_36433_to_fp16)[name = string("aw_chunk_3771_cast_fp16")]; + fp16 var_36435_to_fp16 = const()[name = string("op_36435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3773_cast_fp16, y = var_36435_to_fp16)[name = string("aw_chunk_3773_cast_fp16")]; + fp16 var_36437_to_fp16 = const()[name = string("op_36437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3775_cast_fp16, y = var_36437_to_fp16)[name = string("aw_chunk_3775_cast_fp16")]; + fp16 var_36439_to_fp16 = const()[name = string("op_36439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3777_cast_fp16, y = var_36439_to_fp16)[name = string("aw_chunk_3777_cast_fp16")]; + fp16 var_36441_to_fp16 = const()[name = string("op_36441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3779_cast_fp16, y = var_36441_to_fp16)[name = string("aw_chunk_3779_cast_fp16")]; + fp16 var_36443_to_fp16 = const()[name = string("op_36443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3781_cast_fp16, y = var_36443_to_fp16)[name = string("aw_chunk_3781_cast_fp16")]; + fp16 var_36445_to_fp16 = const()[name = string("op_36445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3783_cast_fp16, y = var_36445_to_fp16)[name = string("aw_chunk_3783_cast_fp16")]; + fp16 var_36447_to_fp16 = const()[name = string("op_36447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3785_cast_fp16, y = var_36447_to_fp16)[name = string("aw_chunk_3785_cast_fp16")]; + fp16 var_36449_to_fp16 = const()[name = string("op_36449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3787_cast_fp16, y = var_36449_to_fp16)[name = string("aw_chunk_3787_cast_fp16")]; + fp16 var_36451_to_fp16 = const()[name = string("op_36451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3789_cast_fp16, y = var_36451_to_fp16)[name = string("aw_chunk_3789_cast_fp16")]; + fp16 var_36453_to_fp16 = const()[name = string("op_36453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3791_cast_fp16, y = var_36453_to_fp16)[name = string("aw_chunk_3791_cast_fp16")]; + fp16 var_36455_to_fp16 = const()[name = string("op_36455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3793_cast_fp16, y = var_36455_to_fp16)[name = string("aw_chunk_3793_cast_fp16")]; + fp16 var_36457_to_fp16 = const()[name = string("op_36457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3795_cast_fp16, y = var_36457_to_fp16)[name = string("aw_chunk_3795_cast_fp16")]; + fp16 var_36459_to_fp16 = const()[name = string("op_36459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3797_cast_fp16, y = var_36459_to_fp16)[name = string("aw_chunk_3797_cast_fp16")]; + fp16 var_36461_to_fp16 = const()[name = string("op_36461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3799_cast_fp16, y = var_36461_to_fp16)[name = string("aw_chunk_3799_cast_fp16")]; + fp16 var_36463_to_fp16 = const()[name = string("op_36463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3801_cast_fp16, y = var_36463_to_fp16)[name = string("aw_chunk_3801_cast_fp16")]; + fp16 var_36465_to_fp16 = const()[name = string("op_36465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3803_cast_fp16, y = var_36465_to_fp16)[name = string("aw_chunk_3803_cast_fp16")]; + fp16 var_36467_to_fp16 = const()[name = string("op_36467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3805_cast_fp16, y = var_36467_to_fp16)[name = string("aw_chunk_3805_cast_fp16")]; + fp16 var_36469_to_fp16 = const()[name = string("op_36469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3807_cast_fp16, y = var_36469_to_fp16)[name = string("aw_chunk_3807_cast_fp16")]; + fp16 var_36471_to_fp16 = const()[name = string("op_36471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3809_cast_fp16, y = var_36471_to_fp16)[name = string("aw_chunk_3809_cast_fp16")]; + fp16 var_36473_to_fp16 = const()[name = string("op_36473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3811_cast_fp16, y = var_36473_to_fp16)[name = string("aw_chunk_3811_cast_fp16")]; + fp16 var_36475_to_fp16 = const()[name = string("op_36475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3813_cast_fp16, y = var_36475_to_fp16)[name = string("aw_chunk_3813_cast_fp16")]; + fp16 var_36477_to_fp16 = const()[name = string("op_36477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3815_cast_fp16, y = var_36477_to_fp16)[name = string("aw_chunk_3815_cast_fp16")]; + fp16 var_36479_to_fp16 = const()[name = string("op_36479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3817_cast_fp16, y = var_36479_to_fp16)[name = string("aw_chunk_3817_cast_fp16")]; + fp16 var_36481_to_fp16 = const()[name = string("op_36481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3819_cast_fp16, y = var_36481_to_fp16)[name = string("aw_chunk_3819_cast_fp16")]; + fp16 var_36483_to_fp16 = const()[name = string("op_36483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3821_cast_fp16, y = var_36483_to_fp16)[name = string("aw_chunk_3821_cast_fp16")]; + fp16 var_36485_to_fp16 = const()[name = string("op_36485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3823_cast_fp16, y = var_36485_to_fp16)[name = string("aw_chunk_3823_cast_fp16")]; + fp16 var_36487_to_fp16 = const()[name = string("op_36487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3825_cast_fp16, y = var_36487_to_fp16)[name = string("aw_chunk_3825_cast_fp16")]; + fp16 var_36489_to_fp16 = const()[name = string("op_36489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3827_cast_fp16, y = var_36489_to_fp16)[name = string("aw_chunk_3827_cast_fp16")]; + fp16 var_36491_to_fp16 = const()[name = string("op_36491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3829_cast_fp16, y = var_36491_to_fp16)[name = string("aw_chunk_3829_cast_fp16")]; + fp16 var_36493_to_fp16 = const()[name = string("op_36493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3831_cast_fp16, y = var_36493_to_fp16)[name = string("aw_chunk_3831_cast_fp16")]; + fp16 var_36495_to_fp16 = const()[name = string("op_36495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3833_cast_fp16, y = var_36495_to_fp16)[name = string("aw_chunk_3833_cast_fp16")]; + fp16 var_36497_to_fp16 = const()[name = string("op_36497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3835_cast_fp16, y = var_36497_to_fp16)[name = string("aw_chunk_3835_cast_fp16")]; + fp16 var_36499_to_fp16 = const()[name = string("op_36499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3837_cast_fp16, y = var_36499_to_fp16)[name = string("aw_chunk_3837_cast_fp16")]; + fp16 var_36501_to_fp16 = const()[name = string("op_36501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3839_cast_fp16, y = var_36501_to_fp16)[name = string("aw_chunk_3839_cast_fp16")]; + tensor var_36503_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3681_cast_fp16)[name = string("op_36503_cast_fp16")]; + tensor var_36504_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3683_cast_fp16)[name = string("op_36504_cast_fp16")]; + tensor var_36505_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3685_cast_fp16)[name = string("op_36505_cast_fp16")]; + tensor var_36506_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3687_cast_fp16)[name = string("op_36506_cast_fp16")]; + tensor var_36507_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3689_cast_fp16)[name = string("op_36507_cast_fp16")]; + tensor var_36508_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3691_cast_fp16)[name = string("op_36508_cast_fp16")]; + tensor var_36509_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3693_cast_fp16)[name = string("op_36509_cast_fp16")]; + tensor var_36510_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3695_cast_fp16)[name = string("op_36510_cast_fp16")]; + tensor var_36511_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3697_cast_fp16)[name = string("op_36511_cast_fp16")]; + tensor var_36512_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3699_cast_fp16)[name = string("op_36512_cast_fp16")]; + tensor var_36513_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3701_cast_fp16)[name = string("op_36513_cast_fp16")]; + tensor var_36514_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3703_cast_fp16)[name = string("op_36514_cast_fp16")]; + tensor var_36515_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3705_cast_fp16)[name = string("op_36515_cast_fp16")]; + tensor var_36516_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3707_cast_fp16)[name = string("op_36516_cast_fp16")]; + tensor var_36517_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3709_cast_fp16)[name = string("op_36517_cast_fp16")]; + tensor var_36518_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3711_cast_fp16)[name = string("op_36518_cast_fp16")]; + tensor var_36519_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3713_cast_fp16)[name = string("op_36519_cast_fp16")]; + tensor var_36520_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3715_cast_fp16)[name = string("op_36520_cast_fp16")]; + tensor var_36521_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3717_cast_fp16)[name = string("op_36521_cast_fp16")]; + tensor var_36522_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3719_cast_fp16)[name = string("op_36522_cast_fp16")]; + tensor var_36523_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3721_cast_fp16)[name = string("op_36523_cast_fp16")]; + tensor var_36524_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3723_cast_fp16)[name = string("op_36524_cast_fp16")]; + tensor var_36525_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3725_cast_fp16)[name = string("op_36525_cast_fp16")]; + tensor var_36526_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3727_cast_fp16)[name = string("op_36526_cast_fp16")]; + tensor var_36527_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3729_cast_fp16)[name = string("op_36527_cast_fp16")]; + tensor var_36528_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3731_cast_fp16)[name = string("op_36528_cast_fp16")]; + tensor var_36529_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3733_cast_fp16)[name = string("op_36529_cast_fp16")]; + tensor var_36530_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3735_cast_fp16)[name = string("op_36530_cast_fp16")]; + tensor var_36531_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3737_cast_fp16)[name = string("op_36531_cast_fp16")]; + tensor var_36532_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3739_cast_fp16)[name = string("op_36532_cast_fp16")]; + tensor var_36533_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3741_cast_fp16)[name = string("op_36533_cast_fp16")]; + tensor var_36534_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3743_cast_fp16)[name = string("op_36534_cast_fp16")]; + tensor var_36535_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3745_cast_fp16)[name = string("op_36535_cast_fp16")]; + tensor var_36536_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3747_cast_fp16)[name = string("op_36536_cast_fp16")]; + tensor var_36537_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3749_cast_fp16)[name = string("op_36537_cast_fp16")]; + tensor var_36538_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3751_cast_fp16)[name = string("op_36538_cast_fp16")]; + tensor var_36539_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3753_cast_fp16)[name = string("op_36539_cast_fp16")]; + tensor var_36540_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3755_cast_fp16)[name = string("op_36540_cast_fp16")]; + tensor var_36541_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3757_cast_fp16)[name = string("op_36541_cast_fp16")]; + tensor var_36542_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3759_cast_fp16)[name = string("op_36542_cast_fp16")]; + tensor var_36543_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3761_cast_fp16)[name = string("op_36543_cast_fp16")]; + tensor var_36544_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3763_cast_fp16)[name = string("op_36544_cast_fp16")]; + tensor var_36545_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3765_cast_fp16)[name = string("op_36545_cast_fp16")]; + tensor var_36546_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3767_cast_fp16)[name = string("op_36546_cast_fp16")]; + tensor var_36547_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3769_cast_fp16)[name = string("op_36547_cast_fp16")]; + tensor var_36548_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3771_cast_fp16)[name = string("op_36548_cast_fp16")]; + tensor var_36549_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3773_cast_fp16)[name = string("op_36549_cast_fp16")]; + tensor var_36550_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3775_cast_fp16)[name = string("op_36550_cast_fp16")]; + tensor var_36551_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3777_cast_fp16)[name = string("op_36551_cast_fp16")]; + tensor var_36552_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3779_cast_fp16)[name = string("op_36552_cast_fp16")]; + tensor var_36553_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3781_cast_fp16)[name = string("op_36553_cast_fp16")]; + tensor var_36554_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3783_cast_fp16)[name = string("op_36554_cast_fp16")]; + tensor var_36555_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3785_cast_fp16)[name = string("op_36555_cast_fp16")]; + tensor var_36556_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3787_cast_fp16)[name = string("op_36556_cast_fp16")]; + tensor var_36557_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3789_cast_fp16)[name = string("op_36557_cast_fp16")]; + tensor var_36558_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3791_cast_fp16)[name = string("op_36558_cast_fp16")]; + tensor var_36559_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3793_cast_fp16)[name = string("op_36559_cast_fp16")]; + tensor var_36560_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3795_cast_fp16)[name = string("op_36560_cast_fp16")]; + tensor var_36561_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3797_cast_fp16)[name = string("op_36561_cast_fp16")]; + tensor var_36562_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3799_cast_fp16)[name = string("op_36562_cast_fp16")]; + tensor var_36563_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3801_cast_fp16)[name = string("op_36563_cast_fp16")]; + tensor var_36564_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3803_cast_fp16)[name = string("op_36564_cast_fp16")]; + tensor var_36565_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3805_cast_fp16)[name = string("op_36565_cast_fp16")]; + tensor var_36566_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3807_cast_fp16)[name = string("op_36566_cast_fp16")]; + tensor var_36567_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3809_cast_fp16)[name = string("op_36567_cast_fp16")]; + tensor var_36568_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3811_cast_fp16)[name = string("op_36568_cast_fp16")]; + tensor var_36569_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3813_cast_fp16)[name = string("op_36569_cast_fp16")]; + tensor var_36570_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3815_cast_fp16)[name = string("op_36570_cast_fp16")]; + tensor var_36571_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3817_cast_fp16)[name = string("op_36571_cast_fp16")]; + tensor var_36572_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3819_cast_fp16)[name = string("op_36572_cast_fp16")]; + tensor var_36573_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3821_cast_fp16)[name = string("op_36573_cast_fp16")]; + tensor var_36574_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3823_cast_fp16)[name = string("op_36574_cast_fp16")]; + tensor var_36575_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3825_cast_fp16)[name = string("op_36575_cast_fp16")]; + tensor var_36576_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3827_cast_fp16)[name = string("op_36576_cast_fp16")]; + tensor var_36577_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3829_cast_fp16)[name = string("op_36577_cast_fp16")]; + tensor var_36578_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3831_cast_fp16)[name = string("op_36578_cast_fp16")]; + tensor var_36579_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3833_cast_fp16)[name = string("op_36579_cast_fp16")]; + tensor var_36580_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3835_cast_fp16)[name = string("op_36580_cast_fp16")]; + tensor var_36581_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3837_cast_fp16)[name = string("op_36581_cast_fp16")]; + tensor var_36582_cast_fp16 = softmax(axis = var_35328, x = aw_chunk_3839_cast_fp16)[name = string("op_36582_cast_fp16")]; + string var_36584_equation_0 = const()[name = string("op_36584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36584_cast_fp16 = einsum(equation = var_36584_equation_0, values = (var_36104_cast_fp16, var_36503_cast_fp16))[name = string("op_36584_cast_fp16")]; + string var_36586_equation_0 = const()[name = string("op_36586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36586_cast_fp16 = einsum(equation = var_36586_equation_0, values = (var_36104_cast_fp16, var_36504_cast_fp16))[name = string("op_36586_cast_fp16")]; + string var_36588_equation_0 = const()[name = string("op_36588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36588_cast_fp16 = einsum(equation = var_36588_equation_0, values = (var_36104_cast_fp16, var_36505_cast_fp16))[name = string("op_36588_cast_fp16")]; + string var_36590_equation_0 = const()[name = string("op_36590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36590_cast_fp16 = einsum(equation = var_36590_equation_0, values = (var_36104_cast_fp16, var_36506_cast_fp16))[name = string("op_36590_cast_fp16")]; + string var_36592_equation_0 = const()[name = string("op_36592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36592_cast_fp16 = einsum(equation = var_36592_equation_0, values = (var_36108_cast_fp16, var_36507_cast_fp16))[name = string("op_36592_cast_fp16")]; + string var_36594_equation_0 = const()[name = string("op_36594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36594_cast_fp16 = einsum(equation = var_36594_equation_0, values = (var_36108_cast_fp16, var_36508_cast_fp16))[name = string("op_36594_cast_fp16")]; + string var_36596_equation_0 = const()[name = string("op_36596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36596_cast_fp16 = einsum(equation = var_36596_equation_0, values = (var_36108_cast_fp16, var_36509_cast_fp16))[name = string("op_36596_cast_fp16")]; + string var_36598_equation_0 = const()[name = string("op_36598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36598_cast_fp16 = einsum(equation = var_36598_equation_0, values = (var_36108_cast_fp16, var_36510_cast_fp16))[name = string("op_36598_cast_fp16")]; + string var_36600_equation_0 = const()[name = string("op_36600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36600_cast_fp16 = einsum(equation = var_36600_equation_0, values = (var_36112_cast_fp16, var_36511_cast_fp16))[name = string("op_36600_cast_fp16")]; + string var_36602_equation_0 = const()[name = string("op_36602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36602_cast_fp16 = einsum(equation = var_36602_equation_0, values = (var_36112_cast_fp16, var_36512_cast_fp16))[name = string("op_36602_cast_fp16")]; + string var_36604_equation_0 = const()[name = string("op_36604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36604_cast_fp16 = einsum(equation = var_36604_equation_0, values = (var_36112_cast_fp16, var_36513_cast_fp16))[name = string("op_36604_cast_fp16")]; + string var_36606_equation_0 = const()[name = string("op_36606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36606_cast_fp16 = einsum(equation = var_36606_equation_0, values = (var_36112_cast_fp16, var_36514_cast_fp16))[name = string("op_36606_cast_fp16")]; + string var_36608_equation_0 = const()[name = string("op_36608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36608_cast_fp16 = einsum(equation = var_36608_equation_0, values = (var_36116_cast_fp16, var_36515_cast_fp16))[name = string("op_36608_cast_fp16")]; + string var_36610_equation_0 = const()[name = string("op_36610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36610_cast_fp16 = einsum(equation = var_36610_equation_0, values = (var_36116_cast_fp16, var_36516_cast_fp16))[name = string("op_36610_cast_fp16")]; + string var_36612_equation_0 = const()[name = string("op_36612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36612_cast_fp16 = einsum(equation = var_36612_equation_0, values = (var_36116_cast_fp16, var_36517_cast_fp16))[name = string("op_36612_cast_fp16")]; + string var_36614_equation_0 = const()[name = string("op_36614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36614_cast_fp16 = einsum(equation = var_36614_equation_0, values = (var_36116_cast_fp16, var_36518_cast_fp16))[name = string("op_36614_cast_fp16")]; + string var_36616_equation_0 = const()[name = string("op_36616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36616_cast_fp16 = einsum(equation = var_36616_equation_0, values = (var_36120_cast_fp16, var_36519_cast_fp16))[name = string("op_36616_cast_fp16")]; + string var_36618_equation_0 = const()[name = string("op_36618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36618_cast_fp16 = einsum(equation = var_36618_equation_0, values = (var_36120_cast_fp16, var_36520_cast_fp16))[name = string("op_36618_cast_fp16")]; + string var_36620_equation_0 = const()[name = string("op_36620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36620_cast_fp16 = einsum(equation = var_36620_equation_0, values = (var_36120_cast_fp16, var_36521_cast_fp16))[name = string("op_36620_cast_fp16")]; + string var_36622_equation_0 = const()[name = string("op_36622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36622_cast_fp16 = einsum(equation = var_36622_equation_0, values = (var_36120_cast_fp16, var_36522_cast_fp16))[name = string("op_36622_cast_fp16")]; + string var_36624_equation_0 = const()[name = string("op_36624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36624_cast_fp16 = einsum(equation = var_36624_equation_0, values = (var_36124_cast_fp16, var_36523_cast_fp16))[name = string("op_36624_cast_fp16")]; + string var_36626_equation_0 = const()[name = string("op_36626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36626_cast_fp16 = einsum(equation = var_36626_equation_0, values = (var_36124_cast_fp16, var_36524_cast_fp16))[name = string("op_36626_cast_fp16")]; + string var_36628_equation_0 = const()[name = string("op_36628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36628_cast_fp16 = einsum(equation = var_36628_equation_0, values = (var_36124_cast_fp16, var_36525_cast_fp16))[name = string("op_36628_cast_fp16")]; + string var_36630_equation_0 = const()[name = string("op_36630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36630_cast_fp16 = einsum(equation = var_36630_equation_0, values = (var_36124_cast_fp16, var_36526_cast_fp16))[name = string("op_36630_cast_fp16")]; + string var_36632_equation_0 = const()[name = string("op_36632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36632_cast_fp16 = einsum(equation = var_36632_equation_0, values = (var_36128_cast_fp16, var_36527_cast_fp16))[name = string("op_36632_cast_fp16")]; + string var_36634_equation_0 = const()[name = string("op_36634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36634_cast_fp16 = einsum(equation = var_36634_equation_0, values = (var_36128_cast_fp16, var_36528_cast_fp16))[name = string("op_36634_cast_fp16")]; + string var_36636_equation_0 = const()[name = string("op_36636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36636_cast_fp16 = einsum(equation = var_36636_equation_0, values = (var_36128_cast_fp16, var_36529_cast_fp16))[name = string("op_36636_cast_fp16")]; + string var_36638_equation_0 = const()[name = string("op_36638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36638_cast_fp16 = einsum(equation = var_36638_equation_0, values = (var_36128_cast_fp16, var_36530_cast_fp16))[name = string("op_36638_cast_fp16")]; + string var_36640_equation_0 = const()[name = string("op_36640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36640_cast_fp16 = einsum(equation = var_36640_equation_0, values = (var_36132_cast_fp16, var_36531_cast_fp16))[name = string("op_36640_cast_fp16")]; + string var_36642_equation_0 = const()[name = string("op_36642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36642_cast_fp16 = einsum(equation = var_36642_equation_0, values = (var_36132_cast_fp16, var_36532_cast_fp16))[name = string("op_36642_cast_fp16")]; + string var_36644_equation_0 = const()[name = string("op_36644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36644_cast_fp16 = einsum(equation = var_36644_equation_0, values = (var_36132_cast_fp16, var_36533_cast_fp16))[name = string("op_36644_cast_fp16")]; + string var_36646_equation_0 = const()[name = string("op_36646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36646_cast_fp16 = einsum(equation = var_36646_equation_0, values = (var_36132_cast_fp16, var_36534_cast_fp16))[name = string("op_36646_cast_fp16")]; + string var_36648_equation_0 = const()[name = string("op_36648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36648_cast_fp16 = einsum(equation = var_36648_equation_0, values = (var_36136_cast_fp16, var_36535_cast_fp16))[name = string("op_36648_cast_fp16")]; + string var_36650_equation_0 = const()[name = string("op_36650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36650_cast_fp16 = einsum(equation = var_36650_equation_0, values = (var_36136_cast_fp16, var_36536_cast_fp16))[name = string("op_36650_cast_fp16")]; + string var_36652_equation_0 = const()[name = string("op_36652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36652_cast_fp16 = einsum(equation = var_36652_equation_0, values = (var_36136_cast_fp16, var_36537_cast_fp16))[name = string("op_36652_cast_fp16")]; + string var_36654_equation_0 = const()[name = string("op_36654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36654_cast_fp16 = einsum(equation = var_36654_equation_0, values = (var_36136_cast_fp16, var_36538_cast_fp16))[name = string("op_36654_cast_fp16")]; + string var_36656_equation_0 = const()[name = string("op_36656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36656_cast_fp16 = einsum(equation = var_36656_equation_0, values = (var_36140_cast_fp16, var_36539_cast_fp16))[name = string("op_36656_cast_fp16")]; + string var_36658_equation_0 = const()[name = string("op_36658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36658_cast_fp16 = einsum(equation = var_36658_equation_0, values = (var_36140_cast_fp16, var_36540_cast_fp16))[name = string("op_36658_cast_fp16")]; + string var_36660_equation_0 = const()[name = string("op_36660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36660_cast_fp16 = einsum(equation = var_36660_equation_0, values = (var_36140_cast_fp16, var_36541_cast_fp16))[name = string("op_36660_cast_fp16")]; + string var_36662_equation_0 = const()[name = string("op_36662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36662_cast_fp16 = einsum(equation = var_36662_equation_0, values = (var_36140_cast_fp16, var_36542_cast_fp16))[name = string("op_36662_cast_fp16")]; + string var_36664_equation_0 = const()[name = string("op_36664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36664_cast_fp16 = einsum(equation = var_36664_equation_0, values = (var_36144_cast_fp16, var_36543_cast_fp16))[name = string("op_36664_cast_fp16")]; + string var_36666_equation_0 = const()[name = string("op_36666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36666_cast_fp16 = einsum(equation = var_36666_equation_0, values = (var_36144_cast_fp16, var_36544_cast_fp16))[name = string("op_36666_cast_fp16")]; + string var_36668_equation_0 = const()[name = string("op_36668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36668_cast_fp16 = einsum(equation = var_36668_equation_0, values = (var_36144_cast_fp16, var_36545_cast_fp16))[name = string("op_36668_cast_fp16")]; + string var_36670_equation_0 = const()[name = string("op_36670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36670_cast_fp16 = einsum(equation = var_36670_equation_0, values = (var_36144_cast_fp16, var_36546_cast_fp16))[name = string("op_36670_cast_fp16")]; + string var_36672_equation_0 = const()[name = string("op_36672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36672_cast_fp16 = einsum(equation = var_36672_equation_0, values = (var_36148_cast_fp16, var_36547_cast_fp16))[name = string("op_36672_cast_fp16")]; + string var_36674_equation_0 = const()[name = string("op_36674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36674_cast_fp16 = einsum(equation = var_36674_equation_0, values = (var_36148_cast_fp16, var_36548_cast_fp16))[name = string("op_36674_cast_fp16")]; + string var_36676_equation_0 = const()[name = string("op_36676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36676_cast_fp16 = einsum(equation = var_36676_equation_0, values = (var_36148_cast_fp16, var_36549_cast_fp16))[name = string("op_36676_cast_fp16")]; + string var_36678_equation_0 = const()[name = string("op_36678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36678_cast_fp16 = einsum(equation = var_36678_equation_0, values = (var_36148_cast_fp16, var_36550_cast_fp16))[name = string("op_36678_cast_fp16")]; + string var_36680_equation_0 = const()[name = string("op_36680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36680_cast_fp16 = einsum(equation = var_36680_equation_0, values = (var_36152_cast_fp16, var_36551_cast_fp16))[name = string("op_36680_cast_fp16")]; + string var_36682_equation_0 = const()[name = string("op_36682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36682_cast_fp16 = einsum(equation = var_36682_equation_0, values = (var_36152_cast_fp16, var_36552_cast_fp16))[name = string("op_36682_cast_fp16")]; + string var_36684_equation_0 = const()[name = string("op_36684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36684_cast_fp16 = einsum(equation = var_36684_equation_0, values = (var_36152_cast_fp16, var_36553_cast_fp16))[name = string("op_36684_cast_fp16")]; + string var_36686_equation_0 = const()[name = string("op_36686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36686_cast_fp16 = einsum(equation = var_36686_equation_0, values = (var_36152_cast_fp16, var_36554_cast_fp16))[name = string("op_36686_cast_fp16")]; + string var_36688_equation_0 = const()[name = string("op_36688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36688_cast_fp16 = einsum(equation = var_36688_equation_0, values = (var_36156_cast_fp16, var_36555_cast_fp16))[name = string("op_36688_cast_fp16")]; + string var_36690_equation_0 = const()[name = string("op_36690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36690_cast_fp16 = einsum(equation = var_36690_equation_0, values = (var_36156_cast_fp16, var_36556_cast_fp16))[name = string("op_36690_cast_fp16")]; + string var_36692_equation_0 = const()[name = string("op_36692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36692_cast_fp16 = einsum(equation = var_36692_equation_0, values = (var_36156_cast_fp16, var_36557_cast_fp16))[name = string("op_36692_cast_fp16")]; + string var_36694_equation_0 = const()[name = string("op_36694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36694_cast_fp16 = einsum(equation = var_36694_equation_0, values = (var_36156_cast_fp16, var_36558_cast_fp16))[name = string("op_36694_cast_fp16")]; + string var_36696_equation_0 = const()[name = string("op_36696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36696_cast_fp16 = einsum(equation = var_36696_equation_0, values = (var_36160_cast_fp16, var_36559_cast_fp16))[name = string("op_36696_cast_fp16")]; + string var_36698_equation_0 = const()[name = string("op_36698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36698_cast_fp16 = einsum(equation = var_36698_equation_0, values = (var_36160_cast_fp16, var_36560_cast_fp16))[name = string("op_36698_cast_fp16")]; + string var_36700_equation_0 = const()[name = string("op_36700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36700_cast_fp16 = einsum(equation = var_36700_equation_0, values = (var_36160_cast_fp16, var_36561_cast_fp16))[name = string("op_36700_cast_fp16")]; + string var_36702_equation_0 = const()[name = string("op_36702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36702_cast_fp16 = einsum(equation = var_36702_equation_0, values = (var_36160_cast_fp16, var_36562_cast_fp16))[name = string("op_36702_cast_fp16")]; + string var_36704_equation_0 = const()[name = string("op_36704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36704_cast_fp16 = einsum(equation = var_36704_equation_0, values = (var_36164_cast_fp16, var_36563_cast_fp16))[name = string("op_36704_cast_fp16")]; + string var_36706_equation_0 = const()[name = string("op_36706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36706_cast_fp16 = einsum(equation = var_36706_equation_0, values = (var_36164_cast_fp16, var_36564_cast_fp16))[name = string("op_36706_cast_fp16")]; + string var_36708_equation_0 = const()[name = string("op_36708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36708_cast_fp16 = einsum(equation = var_36708_equation_0, values = (var_36164_cast_fp16, var_36565_cast_fp16))[name = string("op_36708_cast_fp16")]; + string var_36710_equation_0 = const()[name = string("op_36710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36710_cast_fp16 = einsum(equation = var_36710_equation_0, values = (var_36164_cast_fp16, var_36566_cast_fp16))[name = string("op_36710_cast_fp16")]; + string var_36712_equation_0 = const()[name = string("op_36712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36712_cast_fp16 = einsum(equation = var_36712_equation_0, values = (var_36168_cast_fp16, var_36567_cast_fp16))[name = string("op_36712_cast_fp16")]; + string var_36714_equation_0 = const()[name = string("op_36714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36714_cast_fp16 = einsum(equation = var_36714_equation_0, values = (var_36168_cast_fp16, var_36568_cast_fp16))[name = string("op_36714_cast_fp16")]; + string var_36716_equation_0 = const()[name = string("op_36716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36716_cast_fp16 = einsum(equation = var_36716_equation_0, values = (var_36168_cast_fp16, var_36569_cast_fp16))[name = string("op_36716_cast_fp16")]; + string var_36718_equation_0 = const()[name = string("op_36718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36718_cast_fp16 = einsum(equation = var_36718_equation_0, values = (var_36168_cast_fp16, var_36570_cast_fp16))[name = string("op_36718_cast_fp16")]; + string var_36720_equation_0 = const()[name = string("op_36720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36720_cast_fp16 = einsum(equation = var_36720_equation_0, values = (var_36172_cast_fp16, var_36571_cast_fp16))[name = string("op_36720_cast_fp16")]; + string var_36722_equation_0 = const()[name = string("op_36722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36722_cast_fp16 = einsum(equation = var_36722_equation_0, values = (var_36172_cast_fp16, var_36572_cast_fp16))[name = string("op_36722_cast_fp16")]; + string var_36724_equation_0 = const()[name = string("op_36724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36724_cast_fp16 = einsum(equation = var_36724_equation_0, values = (var_36172_cast_fp16, var_36573_cast_fp16))[name = string("op_36724_cast_fp16")]; + string var_36726_equation_0 = const()[name = string("op_36726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36726_cast_fp16 = einsum(equation = var_36726_equation_0, values = (var_36172_cast_fp16, var_36574_cast_fp16))[name = string("op_36726_cast_fp16")]; + string var_36728_equation_0 = const()[name = string("op_36728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36728_cast_fp16 = einsum(equation = var_36728_equation_0, values = (var_36176_cast_fp16, var_36575_cast_fp16))[name = string("op_36728_cast_fp16")]; + string var_36730_equation_0 = const()[name = string("op_36730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36730_cast_fp16 = einsum(equation = var_36730_equation_0, values = (var_36176_cast_fp16, var_36576_cast_fp16))[name = string("op_36730_cast_fp16")]; + string var_36732_equation_0 = const()[name = string("op_36732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36732_cast_fp16 = einsum(equation = var_36732_equation_0, values = (var_36176_cast_fp16, var_36577_cast_fp16))[name = string("op_36732_cast_fp16")]; + string var_36734_equation_0 = const()[name = string("op_36734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36734_cast_fp16 = einsum(equation = var_36734_equation_0, values = (var_36176_cast_fp16, var_36578_cast_fp16))[name = string("op_36734_cast_fp16")]; + string var_36736_equation_0 = const()[name = string("op_36736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36736_cast_fp16 = einsum(equation = var_36736_equation_0, values = (var_36180_cast_fp16, var_36579_cast_fp16))[name = string("op_36736_cast_fp16")]; + string var_36738_equation_0 = const()[name = string("op_36738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36738_cast_fp16 = einsum(equation = var_36738_equation_0, values = (var_36180_cast_fp16, var_36580_cast_fp16))[name = string("op_36738_cast_fp16")]; + string var_36740_equation_0 = const()[name = string("op_36740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36740_cast_fp16 = einsum(equation = var_36740_equation_0, values = (var_36180_cast_fp16, var_36581_cast_fp16))[name = string("op_36740_cast_fp16")]; + string var_36742_equation_0 = const()[name = string("op_36742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_36742_cast_fp16 = einsum(equation = var_36742_equation_0, values = (var_36180_cast_fp16, var_36582_cast_fp16))[name = string("op_36742_cast_fp16")]; + bool var_36744_interleave_0 = const()[name = string("op_36744_interleave_0"), val = bool(false)]; + tensor var_36744_cast_fp16 = concat(axis = var_35303, interleave = var_36744_interleave_0, values = (var_36584_cast_fp16, var_36586_cast_fp16, var_36588_cast_fp16, var_36590_cast_fp16))[name = string("op_36744_cast_fp16")]; + bool var_36746_interleave_0 = const()[name = string("op_36746_interleave_0"), val = bool(false)]; + tensor var_36746_cast_fp16 = concat(axis = var_35303, interleave = var_36746_interleave_0, values = (var_36592_cast_fp16, var_36594_cast_fp16, var_36596_cast_fp16, var_36598_cast_fp16))[name = string("op_36746_cast_fp16")]; + bool var_36748_interleave_0 = const()[name = string("op_36748_interleave_0"), val = bool(false)]; + tensor var_36748_cast_fp16 = concat(axis = var_35303, interleave = var_36748_interleave_0, values = (var_36600_cast_fp16, var_36602_cast_fp16, var_36604_cast_fp16, var_36606_cast_fp16))[name = string("op_36748_cast_fp16")]; + bool var_36750_interleave_0 = const()[name = string("op_36750_interleave_0"), val = bool(false)]; + tensor var_36750_cast_fp16 = concat(axis = var_35303, interleave = var_36750_interleave_0, values = (var_36608_cast_fp16, var_36610_cast_fp16, var_36612_cast_fp16, var_36614_cast_fp16))[name = string("op_36750_cast_fp16")]; + bool var_36752_interleave_0 = const()[name = string("op_36752_interleave_0"), val = bool(false)]; + tensor var_36752_cast_fp16 = concat(axis = var_35303, interleave = var_36752_interleave_0, values = (var_36616_cast_fp16, var_36618_cast_fp16, var_36620_cast_fp16, var_36622_cast_fp16))[name = string("op_36752_cast_fp16")]; + bool var_36754_interleave_0 = const()[name = string("op_36754_interleave_0"), val = bool(false)]; + tensor var_36754_cast_fp16 = concat(axis = var_35303, interleave = var_36754_interleave_0, values = (var_36624_cast_fp16, var_36626_cast_fp16, var_36628_cast_fp16, var_36630_cast_fp16))[name = string("op_36754_cast_fp16")]; + bool var_36756_interleave_0 = const()[name = string("op_36756_interleave_0"), val = bool(false)]; + tensor var_36756_cast_fp16 = concat(axis = var_35303, interleave = var_36756_interleave_0, values = (var_36632_cast_fp16, var_36634_cast_fp16, var_36636_cast_fp16, var_36638_cast_fp16))[name = string("op_36756_cast_fp16")]; + bool var_36758_interleave_0 = const()[name = string("op_36758_interleave_0"), val = bool(false)]; + tensor var_36758_cast_fp16 = concat(axis = var_35303, interleave = var_36758_interleave_0, values = (var_36640_cast_fp16, var_36642_cast_fp16, var_36644_cast_fp16, var_36646_cast_fp16))[name = string("op_36758_cast_fp16")]; + bool var_36760_interleave_0 = const()[name = string("op_36760_interleave_0"), val = bool(false)]; + tensor var_36760_cast_fp16 = concat(axis = var_35303, interleave = var_36760_interleave_0, values = (var_36648_cast_fp16, var_36650_cast_fp16, var_36652_cast_fp16, var_36654_cast_fp16))[name = string("op_36760_cast_fp16")]; + bool var_36762_interleave_0 = const()[name = string("op_36762_interleave_0"), val = bool(false)]; + tensor var_36762_cast_fp16 = concat(axis = var_35303, interleave = var_36762_interleave_0, values = (var_36656_cast_fp16, var_36658_cast_fp16, var_36660_cast_fp16, var_36662_cast_fp16))[name = string("op_36762_cast_fp16")]; + bool var_36764_interleave_0 = const()[name = string("op_36764_interleave_0"), val = bool(false)]; + tensor var_36764_cast_fp16 = concat(axis = var_35303, interleave = var_36764_interleave_0, values = (var_36664_cast_fp16, var_36666_cast_fp16, var_36668_cast_fp16, var_36670_cast_fp16))[name = string("op_36764_cast_fp16")]; + bool var_36766_interleave_0 = const()[name = string("op_36766_interleave_0"), val = bool(false)]; + tensor var_36766_cast_fp16 = concat(axis = var_35303, interleave = var_36766_interleave_0, values = (var_36672_cast_fp16, var_36674_cast_fp16, var_36676_cast_fp16, var_36678_cast_fp16))[name = string("op_36766_cast_fp16")]; + bool var_36768_interleave_0 = const()[name = string("op_36768_interleave_0"), val = bool(false)]; + tensor var_36768_cast_fp16 = concat(axis = var_35303, interleave = var_36768_interleave_0, values = (var_36680_cast_fp16, var_36682_cast_fp16, var_36684_cast_fp16, var_36686_cast_fp16))[name = string("op_36768_cast_fp16")]; + bool var_36770_interleave_0 = const()[name = string("op_36770_interleave_0"), val = bool(false)]; + tensor var_36770_cast_fp16 = concat(axis = var_35303, interleave = var_36770_interleave_0, values = (var_36688_cast_fp16, var_36690_cast_fp16, var_36692_cast_fp16, var_36694_cast_fp16))[name = string("op_36770_cast_fp16")]; + bool var_36772_interleave_0 = const()[name = string("op_36772_interleave_0"), val = bool(false)]; + tensor var_36772_cast_fp16 = concat(axis = var_35303, interleave = var_36772_interleave_0, values = (var_36696_cast_fp16, var_36698_cast_fp16, var_36700_cast_fp16, var_36702_cast_fp16))[name = string("op_36772_cast_fp16")]; + bool var_36774_interleave_0 = const()[name = string("op_36774_interleave_0"), val = bool(false)]; + tensor var_36774_cast_fp16 = concat(axis = var_35303, interleave = var_36774_interleave_0, values = (var_36704_cast_fp16, var_36706_cast_fp16, var_36708_cast_fp16, var_36710_cast_fp16))[name = string("op_36774_cast_fp16")]; + bool var_36776_interleave_0 = const()[name = string("op_36776_interleave_0"), val = bool(false)]; + tensor var_36776_cast_fp16 = concat(axis = var_35303, interleave = var_36776_interleave_0, values = (var_36712_cast_fp16, var_36714_cast_fp16, var_36716_cast_fp16, var_36718_cast_fp16))[name = string("op_36776_cast_fp16")]; + bool var_36778_interleave_0 = const()[name = string("op_36778_interleave_0"), val = bool(false)]; + tensor var_36778_cast_fp16 = concat(axis = var_35303, interleave = var_36778_interleave_0, values = (var_36720_cast_fp16, var_36722_cast_fp16, var_36724_cast_fp16, var_36726_cast_fp16))[name = string("op_36778_cast_fp16")]; + bool var_36780_interleave_0 = const()[name = string("op_36780_interleave_0"), val = bool(false)]; + tensor var_36780_cast_fp16 = concat(axis = var_35303, interleave = var_36780_interleave_0, values = (var_36728_cast_fp16, var_36730_cast_fp16, var_36732_cast_fp16, var_36734_cast_fp16))[name = string("op_36780_cast_fp16")]; + bool var_36782_interleave_0 = const()[name = string("op_36782_interleave_0"), val = bool(false)]; + tensor var_36782_cast_fp16 = concat(axis = var_35303, interleave = var_36782_interleave_0, values = (var_36736_cast_fp16, var_36738_cast_fp16, var_36740_cast_fp16, var_36742_cast_fp16))[name = string("op_36782_cast_fp16")]; + bool input_185_interleave_0 = const()[name = string("input_185_interleave_0"), val = bool(false)]; + tensor input_185_cast_fp16 = concat(axis = var_35328, interleave = input_185_interleave_0, values = (var_36744_cast_fp16, var_36746_cast_fp16, var_36748_cast_fp16, var_36750_cast_fp16, var_36752_cast_fp16, var_36754_cast_fp16, var_36756_cast_fp16, var_36758_cast_fp16, var_36760_cast_fp16, var_36762_cast_fp16, var_36764_cast_fp16, var_36766_cast_fp16, var_36768_cast_fp16, var_36770_cast_fp16, var_36772_cast_fp16, var_36774_cast_fp16, var_36776_cast_fp16, var_36778_cast_fp16, var_36780_cast_fp16, var_36782_cast_fp16))[name = string("input_185_cast_fp16")]; + string obj_95_pad_type_0 = const()[name = string("obj_95_pad_type_0"), val = string("valid")]; + tensor obj_95_strides_0 = const()[name = string("obj_95_strides_0"), val = tensor([1, 1])]; + tensor obj_95_pad_0 = const()[name = string("obj_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_95_dilations_0 = const()[name = string("obj_95_dilations_0"), val = tensor([1, 1])]; + int32 obj_95_groups_0 = const()[name = string("obj_95_groups_0"), val = int32(1)]; + tensor layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(929630720)))]; + tensor layers_23_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932907584)))]; + tensor obj_95_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_95_cast_fp16)[name = string("inputs_95_cast_fp16")]; + tensor out_95_axes_0 = const()[name = string("out_95_axes_0"), val = tensor([1])]; + fp16 var_36801_to_fp16 = const()[name = string("op_36801_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_36801_to_fp16, x = inputs_95_cast_fp16)[name = string("out_95_cast_fp16")]; + tensor input_187_gamma_0_to_fp16 = const()[name = string("input_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932910208)))]; + tensor input_187_beta_0_to_fp16 = const()[name = string("input_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932912832)))]; + fp16 input_187_epsilon_0_to_fp16 = const()[name = string("input_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_187_cast_fp16 = batch_norm(beta = input_187_beta_0_to_fp16, epsilon = input_187_epsilon_0_to_fp16, gamma = input_187_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = string("input_187_cast_fp16")]; + string input_189_pad_type_0 = const()[name = string("input_189_pad_type_0"), val = string("valid")]; + tensor input_189_strides_0 = const()[name = string("input_189_strides_0"), val = tensor([1, 1])]; + tensor input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_189_dilations_0 = const()[name = string("input_189_dilations_0"), val = tensor([1, 1])]; + int32 input_189_groups_0 = const()[name = string("input_189_groups_0"), val = int32(1)]; + tensor layers_23_fc1_weight_to_fp16 = const()[name = string("layers_23_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(932915456)))]; + tensor layers_23_fc1_bias_to_fp16 = const()[name = string("layers_23_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946022720)))]; + tensor input_189_cast_fp16 = conv(bias = layers_23_fc1_bias_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = layers_23_fc1_weight_to_fp16, x = input_187_cast_fp16)[name = string("input_189_cast_fp16")]; + string input_191_mode_0 = const()[name = string("input_191_mode_0"), val = string("EXACT")]; + tensor input_191_cast_fp16 = gelu(mode = input_191_mode_0, x = input_189_cast_fp16)[name = string("input_191_cast_fp16")]; + string hidden_states_51_pad_type_0 = const()[name = string("hidden_states_51_pad_type_0"), val = string("valid")]; + tensor hidden_states_51_strides_0 = const()[name = string("hidden_states_51_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_51_pad_0 = const()[name = string("hidden_states_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_51_dilations_0 = const()[name = string("hidden_states_51_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_51_groups_0 = const()[name = string("hidden_states_51_groups_0"), val = int32(1)]; + tensor layers_23_fc2_weight_to_fp16 = const()[name = string("layers_23_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946033024)))]; + tensor layers_23_fc2_bias_to_fp16 = const()[name = string("layers_23_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959140288)))]; + tensor hidden_states_51_cast_fp16 = conv(bias = layers_23_fc2_bias_to_fp16, dilations = hidden_states_51_dilations_0, groups = hidden_states_51_groups_0, pad = hidden_states_51_pad_0, pad_type = hidden_states_51_pad_type_0, strides = hidden_states_51_strides_0, weight = layers_23_fc2_weight_to_fp16, x = input_191_cast_fp16)[name = string("hidden_states_51_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_51_cast_fp16)[name = string("inputs_97_cast_fp16")]; + int32 var_36830 = const()[name = string("op_36830"), val = int32(3)]; + int32 var_36855 = const()[name = string("op_36855"), val = int32(1)]; + tensor out_97_axes_0 = const()[name = string("out_97_axes_0"), val = tensor([1])]; + fp16 var_36872_to_fp16 = const()[name = string("op_36872_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_36872_to_fp16, x = inputs_97_cast_fp16)[name = string("out_97_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959142912)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959145536)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_49_pad_type_0 = const()[name = string("query_49_pad_type_0"), val = string("valid")]; + tensor query_49_strides_0 = const()[name = string("query_49_strides_0"), val = tensor([1, 1])]; + tensor query_49_pad_0 = const()[name = string("query_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_49_dilations_0 = const()[name = string("query_49_dilations_0"), val = tensor([1, 1])]; + int32 query_49_groups_0 = const()[name = string("query_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(959148160)))]; + tensor layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962425024)))]; + tensor query_49_cast_fp16 = conv(bias = layers_24_self_attn_q_proj_bias_to_fp16, dilations = query_49_dilations_0, groups = query_49_groups_0, pad = query_49_pad_0, pad_type = query_49_pad_type_0, strides = query_49_strides_0, weight = layers_24_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_49_cast_fp16")]; + string key_49_pad_type_0 = const()[name = string("key_49_pad_type_0"), val = string("valid")]; + tensor key_49_strides_0 = const()[name = string("key_49_strides_0"), val = tensor([1, 1])]; + tensor key_49_pad_0 = const()[name = string("key_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_49_dilations_0 = const()[name = string("key_49_dilations_0"), val = tensor([1, 1])]; + int32 key_49_groups_0 = const()[name = string("key_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(962427648)))]; + tensor key_49_cast_fp16 = conv(dilations = key_49_dilations_0, groups = key_49_groups_0, pad = key_49_pad_0, pad_type = key_49_pad_type_0, strides = key_49_strides_0, weight = layers_24_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("key_49_cast_fp16")]; + string value_49_pad_type_0 = const()[name = string("value_49_pad_type_0"), val = string("valid")]; + tensor value_49_strides_0 = const()[name = string("value_49_strides_0"), val = tensor([1, 1])]; + tensor value_49_pad_0 = const()[name = string("value_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_49_dilations_0 = const()[name = string("value_49_dilations_0"), val = tensor([1, 1])]; + int32 value_49_groups_0 = const()[name = string("value_49_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(965704512)))]; + tensor layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968981376)))]; + tensor value_49_cast_fp16 = conv(bias = layers_24_self_attn_v_proj_bias_to_fp16, dilations = value_49_dilations_0, groups = value_49_groups_0, pad = value_49_pad_0, pad_type = value_49_pad_type_0, strides = value_49_strides_0, weight = layers_24_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("value_49_cast_fp16")]; + tensor var_36910_begin_0 = const()[name = string("op_36910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36910_end_0 = const()[name = string("op_36910_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36910_end_mask_0 = const()[name = string("op_36910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36910_cast_fp16 = slice_by_index(begin = var_36910_begin_0, end = var_36910_end_0, end_mask = var_36910_end_mask_0, x = query_49_cast_fp16)[name = string("op_36910_cast_fp16")]; + tensor var_36914_begin_0 = const()[name = string("op_36914_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36914_end_0 = const()[name = string("op_36914_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_36914_end_mask_0 = const()[name = string("op_36914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36914_cast_fp16 = slice_by_index(begin = var_36914_begin_0, end = var_36914_end_0, end_mask = var_36914_end_mask_0, x = query_49_cast_fp16)[name = string("op_36914_cast_fp16")]; + tensor var_36918_begin_0 = const()[name = string("op_36918_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36918_end_0 = const()[name = string("op_36918_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_36918_end_mask_0 = const()[name = string("op_36918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36918_cast_fp16 = slice_by_index(begin = var_36918_begin_0, end = var_36918_end_0, end_mask = var_36918_end_mask_0, x = query_49_cast_fp16)[name = string("op_36918_cast_fp16")]; + tensor var_36922_begin_0 = const()[name = string("op_36922_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36922_end_0 = const()[name = string("op_36922_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_36922_end_mask_0 = const()[name = string("op_36922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36922_cast_fp16 = slice_by_index(begin = var_36922_begin_0, end = var_36922_end_0, end_mask = var_36922_end_mask_0, x = query_49_cast_fp16)[name = string("op_36922_cast_fp16")]; + tensor var_36926_begin_0 = const()[name = string("op_36926_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36926_end_0 = const()[name = string("op_36926_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_36926_end_mask_0 = const()[name = string("op_36926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36926_cast_fp16 = slice_by_index(begin = var_36926_begin_0, end = var_36926_end_0, end_mask = var_36926_end_mask_0, x = query_49_cast_fp16)[name = string("op_36926_cast_fp16")]; + tensor var_36930_begin_0 = const()[name = string("op_36930_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36930_end_0 = const()[name = string("op_36930_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_36930_end_mask_0 = const()[name = string("op_36930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36930_cast_fp16 = slice_by_index(begin = var_36930_begin_0, end = var_36930_end_0, end_mask = var_36930_end_mask_0, x = query_49_cast_fp16)[name = string("op_36930_cast_fp16")]; + tensor var_36934_begin_0 = const()[name = string("op_36934_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36934_end_0 = const()[name = string("op_36934_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_36934_end_mask_0 = const()[name = string("op_36934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36934_cast_fp16 = slice_by_index(begin = var_36934_begin_0, end = var_36934_end_0, end_mask = var_36934_end_mask_0, x = query_49_cast_fp16)[name = string("op_36934_cast_fp16")]; + tensor var_36938_begin_0 = const()[name = string("op_36938_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36938_end_0 = const()[name = string("op_36938_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_36938_end_mask_0 = const()[name = string("op_36938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36938_cast_fp16 = slice_by_index(begin = var_36938_begin_0, end = var_36938_end_0, end_mask = var_36938_end_mask_0, x = query_49_cast_fp16)[name = string("op_36938_cast_fp16")]; + tensor var_36942_begin_0 = const()[name = string("op_36942_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36942_end_0 = const()[name = string("op_36942_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_36942_end_mask_0 = const()[name = string("op_36942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36942_cast_fp16 = slice_by_index(begin = var_36942_begin_0, end = var_36942_end_0, end_mask = var_36942_end_mask_0, x = query_49_cast_fp16)[name = string("op_36942_cast_fp16")]; + tensor var_36946_begin_0 = const()[name = string("op_36946_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36946_end_0 = const()[name = string("op_36946_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_36946_end_mask_0 = const()[name = string("op_36946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36946_cast_fp16 = slice_by_index(begin = var_36946_begin_0, end = var_36946_end_0, end_mask = var_36946_end_mask_0, x = query_49_cast_fp16)[name = string("op_36946_cast_fp16")]; + tensor var_36950_begin_0 = const()[name = string("op_36950_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36950_end_0 = const()[name = string("op_36950_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_36950_end_mask_0 = const()[name = string("op_36950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36950_cast_fp16 = slice_by_index(begin = var_36950_begin_0, end = var_36950_end_0, end_mask = var_36950_end_mask_0, x = query_49_cast_fp16)[name = string("op_36950_cast_fp16")]; + tensor var_36954_begin_0 = const()[name = string("op_36954_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36954_end_0 = const()[name = string("op_36954_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_36954_end_mask_0 = const()[name = string("op_36954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36954_cast_fp16 = slice_by_index(begin = var_36954_begin_0, end = var_36954_end_0, end_mask = var_36954_end_mask_0, x = query_49_cast_fp16)[name = string("op_36954_cast_fp16")]; + tensor var_36958_begin_0 = const()[name = string("op_36958_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36958_end_0 = const()[name = string("op_36958_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_36958_end_mask_0 = const()[name = string("op_36958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36958_cast_fp16 = slice_by_index(begin = var_36958_begin_0, end = var_36958_end_0, end_mask = var_36958_end_mask_0, x = query_49_cast_fp16)[name = string("op_36958_cast_fp16")]; + tensor var_36962_begin_0 = const()[name = string("op_36962_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36962_end_0 = const()[name = string("op_36962_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_36962_end_mask_0 = const()[name = string("op_36962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36962_cast_fp16 = slice_by_index(begin = var_36962_begin_0, end = var_36962_end_0, end_mask = var_36962_end_mask_0, x = query_49_cast_fp16)[name = string("op_36962_cast_fp16")]; + tensor var_36966_begin_0 = const()[name = string("op_36966_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36966_end_0 = const()[name = string("op_36966_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_36966_end_mask_0 = const()[name = string("op_36966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36966_cast_fp16 = slice_by_index(begin = var_36966_begin_0, end = var_36966_end_0, end_mask = var_36966_end_mask_0, x = query_49_cast_fp16)[name = string("op_36966_cast_fp16")]; + tensor var_36970_begin_0 = const()[name = string("op_36970_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36970_end_0 = const()[name = string("op_36970_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_36970_end_mask_0 = const()[name = string("op_36970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36970_cast_fp16 = slice_by_index(begin = var_36970_begin_0, end = var_36970_end_0, end_mask = var_36970_end_mask_0, x = query_49_cast_fp16)[name = string("op_36970_cast_fp16")]; + tensor var_36974_begin_0 = const()[name = string("op_36974_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36974_end_0 = const()[name = string("op_36974_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_36974_end_mask_0 = const()[name = string("op_36974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36974_cast_fp16 = slice_by_index(begin = var_36974_begin_0, end = var_36974_end_0, end_mask = var_36974_end_mask_0, x = query_49_cast_fp16)[name = string("op_36974_cast_fp16")]; + tensor var_36978_begin_0 = const()[name = string("op_36978_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36978_end_0 = const()[name = string("op_36978_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_36978_end_mask_0 = const()[name = string("op_36978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36978_cast_fp16 = slice_by_index(begin = var_36978_begin_0, end = var_36978_end_0, end_mask = var_36978_end_mask_0, x = query_49_cast_fp16)[name = string("op_36978_cast_fp16")]; + tensor var_36982_begin_0 = const()[name = string("op_36982_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36982_end_0 = const()[name = string("op_36982_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_36982_end_mask_0 = const()[name = string("op_36982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36982_cast_fp16 = slice_by_index(begin = var_36982_begin_0, end = var_36982_end_0, end_mask = var_36982_end_mask_0, x = query_49_cast_fp16)[name = string("op_36982_cast_fp16")]; + tensor var_36986_begin_0 = const()[name = string("op_36986_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36986_end_0 = const()[name = string("op_36986_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_36986_end_mask_0 = const()[name = string("op_36986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36986_cast_fp16 = slice_by_index(begin = var_36986_begin_0, end = var_36986_end_0, end_mask = var_36986_end_mask_0, x = query_49_cast_fp16)[name = string("op_36986_cast_fp16")]; + tensor var_36995_begin_0 = const()[name = string("op_36995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36995_end_0 = const()[name = string("op_36995_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36995_end_mask_0 = const()[name = string("op_36995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36995_cast_fp16 = slice_by_index(begin = var_36995_begin_0, end = var_36995_end_0, end_mask = var_36995_end_mask_0, x = var_36910_cast_fp16)[name = string("op_36995_cast_fp16")]; + tensor var_37002_begin_0 = const()[name = string("op_37002_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37002_end_0 = const()[name = string("op_37002_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37002_end_mask_0 = const()[name = string("op_37002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37002_cast_fp16 = slice_by_index(begin = var_37002_begin_0, end = var_37002_end_0, end_mask = var_37002_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37002_cast_fp16")]; + tensor var_37009_begin_0 = const()[name = string("op_37009_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37009_end_0 = const()[name = string("op_37009_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37009_end_mask_0 = const()[name = string("op_37009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37009_cast_fp16 = slice_by_index(begin = var_37009_begin_0, end = var_37009_end_0, end_mask = var_37009_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37009_cast_fp16")]; + tensor var_37016_begin_0 = const()[name = string("op_37016_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37016_end_0 = const()[name = string("op_37016_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37016_end_mask_0 = const()[name = string("op_37016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37016_cast_fp16 = slice_by_index(begin = var_37016_begin_0, end = var_37016_end_0, end_mask = var_37016_end_mask_0, x = var_36910_cast_fp16)[name = string("op_37016_cast_fp16")]; + tensor var_37023_begin_0 = const()[name = string("op_37023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37023_end_0 = const()[name = string("op_37023_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37023_end_mask_0 = const()[name = string("op_37023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37023_cast_fp16 = slice_by_index(begin = var_37023_begin_0, end = var_37023_end_0, end_mask = var_37023_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37023_cast_fp16")]; + tensor var_37030_begin_0 = const()[name = string("op_37030_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37030_end_0 = const()[name = string("op_37030_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37030_end_mask_0 = const()[name = string("op_37030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37030_cast_fp16 = slice_by_index(begin = var_37030_begin_0, end = var_37030_end_0, end_mask = var_37030_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37030_cast_fp16")]; + tensor var_37037_begin_0 = const()[name = string("op_37037_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37037_end_0 = const()[name = string("op_37037_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37037_end_mask_0 = const()[name = string("op_37037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37037_cast_fp16 = slice_by_index(begin = var_37037_begin_0, end = var_37037_end_0, end_mask = var_37037_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37037_cast_fp16")]; + tensor var_37044_begin_0 = const()[name = string("op_37044_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37044_end_0 = const()[name = string("op_37044_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37044_end_mask_0 = const()[name = string("op_37044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37044_cast_fp16 = slice_by_index(begin = var_37044_begin_0, end = var_37044_end_0, end_mask = var_37044_end_mask_0, x = var_36914_cast_fp16)[name = string("op_37044_cast_fp16")]; + tensor var_37051_begin_0 = const()[name = string("op_37051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37051_end_0 = const()[name = string("op_37051_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37051_end_mask_0 = const()[name = string("op_37051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37051_cast_fp16 = slice_by_index(begin = var_37051_begin_0, end = var_37051_end_0, end_mask = var_37051_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37051_cast_fp16")]; + tensor var_37058_begin_0 = const()[name = string("op_37058_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37058_end_0 = const()[name = string("op_37058_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37058_end_mask_0 = const()[name = string("op_37058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37058_cast_fp16 = slice_by_index(begin = var_37058_begin_0, end = var_37058_end_0, end_mask = var_37058_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37058_cast_fp16")]; + tensor var_37065_begin_0 = const()[name = string("op_37065_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37065_end_0 = const()[name = string("op_37065_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37065_end_mask_0 = const()[name = string("op_37065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37065_cast_fp16 = slice_by_index(begin = var_37065_begin_0, end = var_37065_end_0, end_mask = var_37065_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37065_cast_fp16")]; + tensor var_37072_begin_0 = const()[name = string("op_37072_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37072_end_0 = const()[name = string("op_37072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37072_end_mask_0 = const()[name = string("op_37072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37072_cast_fp16 = slice_by_index(begin = var_37072_begin_0, end = var_37072_end_0, end_mask = var_37072_end_mask_0, x = var_36918_cast_fp16)[name = string("op_37072_cast_fp16")]; + tensor var_37079_begin_0 = const()[name = string("op_37079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37079_end_0 = const()[name = string("op_37079_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37079_end_mask_0 = const()[name = string("op_37079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37079_cast_fp16 = slice_by_index(begin = var_37079_begin_0, end = var_37079_end_0, end_mask = var_37079_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37079_cast_fp16")]; + tensor var_37086_begin_0 = const()[name = string("op_37086_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37086_end_0 = const()[name = string("op_37086_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37086_end_mask_0 = const()[name = string("op_37086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37086_cast_fp16 = slice_by_index(begin = var_37086_begin_0, end = var_37086_end_0, end_mask = var_37086_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37086_cast_fp16")]; + tensor var_37093_begin_0 = const()[name = string("op_37093_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37093_end_0 = const()[name = string("op_37093_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37093_end_mask_0 = const()[name = string("op_37093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37093_cast_fp16 = slice_by_index(begin = var_37093_begin_0, end = var_37093_end_0, end_mask = var_37093_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37093_cast_fp16")]; + tensor var_37100_begin_0 = const()[name = string("op_37100_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37100_end_0 = const()[name = string("op_37100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37100_end_mask_0 = const()[name = string("op_37100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37100_cast_fp16 = slice_by_index(begin = var_37100_begin_0, end = var_37100_end_0, end_mask = var_37100_end_mask_0, x = var_36922_cast_fp16)[name = string("op_37100_cast_fp16")]; + tensor var_37107_begin_0 = const()[name = string("op_37107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37107_end_0 = const()[name = string("op_37107_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37107_end_mask_0 = const()[name = string("op_37107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37107_cast_fp16 = slice_by_index(begin = var_37107_begin_0, end = var_37107_end_0, end_mask = var_37107_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37107_cast_fp16")]; + tensor var_37114_begin_0 = const()[name = string("op_37114_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37114_end_0 = const()[name = string("op_37114_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37114_end_mask_0 = const()[name = string("op_37114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37114_cast_fp16 = slice_by_index(begin = var_37114_begin_0, end = var_37114_end_0, end_mask = var_37114_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37114_cast_fp16")]; + tensor var_37121_begin_0 = const()[name = string("op_37121_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37121_end_0 = const()[name = string("op_37121_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37121_end_mask_0 = const()[name = string("op_37121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37121_cast_fp16 = slice_by_index(begin = var_37121_begin_0, end = var_37121_end_0, end_mask = var_37121_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37121_cast_fp16")]; + tensor var_37128_begin_0 = const()[name = string("op_37128_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37128_end_0 = const()[name = string("op_37128_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37128_end_mask_0 = const()[name = string("op_37128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37128_cast_fp16 = slice_by_index(begin = var_37128_begin_0, end = var_37128_end_0, end_mask = var_37128_end_mask_0, x = var_36926_cast_fp16)[name = string("op_37128_cast_fp16")]; + tensor var_37135_begin_0 = const()[name = string("op_37135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37135_end_0 = const()[name = string("op_37135_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37135_end_mask_0 = const()[name = string("op_37135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37135_cast_fp16 = slice_by_index(begin = var_37135_begin_0, end = var_37135_end_0, end_mask = var_37135_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37135_cast_fp16")]; + tensor var_37142_begin_0 = const()[name = string("op_37142_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37142_end_0 = const()[name = string("op_37142_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37142_end_mask_0 = const()[name = string("op_37142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37142_cast_fp16 = slice_by_index(begin = var_37142_begin_0, end = var_37142_end_0, end_mask = var_37142_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37142_cast_fp16")]; + tensor var_37149_begin_0 = const()[name = string("op_37149_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37149_end_0 = const()[name = string("op_37149_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37149_end_mask_0 = const()[name = string("op_37149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37149_cast_fp16 = slice_by_index(begin = var_37149_begin_0, end = var_37149_end_0, end_mask = var_37149_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37149_cast_fp16")]; + tensor var_37156_begin_0 = const()[name = string("op_37156_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37156_end_0 = const()[name = string("op_37156_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37156_end_mask_0 = const()[name = string("op_37156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37156_cast_fp16 = slice_by_index(begin = var_37156_begin_0, end = var_37156_end_0, end_mask = var_37156_end_mask_0, x = var_36930_cast_fp16)[name = string("op_37156_cast_fp16")]; + tensor var_37163_begin_0 = const()[name = string("op_37163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37163_end_0 = const()[name = string("op_37163_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37163_end_mask_0 = const()[name = string("op_37163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37163_cast_fp16 = slice_by_index(begin = var_37163_begin_0, end = var_37163_end_0, end_mask = var_37163_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37163_cast_fp16")]; + tensor var_37170_begin_0 = const()[name = string("op_37170_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37170_end_0 = const()[name = string("op_37170_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37170_end_mask_0 = const()[name = string("op_37170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37170_cast_fp16 = slice_by_index(begin = var_37170_begin_0, end = var_37170_end_0, end_mask = var_37170_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37170_cast_fp16")]; + tensor var_37177_begin_0 = const()[name = string("op_37177_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37177_end_0 = const()[name = string("op_37177_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37177_end_mask_0 = const()[name = string("op_37177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37177_cast_fp16 = slice_by_index(begin = var_37177_begin_0, end = var_37177_end_0, end_mask = var_37177_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37177_cast_fp16")]; + tensor var_37184_begin_0 = const()[name = string("op_37184_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37184_end_0 = const()[name = string("op_37184_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37184_end_mask_0 = const()[name = string("op_37184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37184_cast_fp16 = slice_by_index(begin = var_37184_begin_0, end = var_37184_end_0, end_mask = var_37184_end_mask_0, x = var_36934_cast_fp16)[name = string("op_37184_cast_fp16")]; + tensor var_37191_begin_0 = const()[name = string("op_37191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37191_end_0 = const()[name = string("op_37191_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37191_end_mask_0 = const()[name = string("op_37191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37191_cast_fp16 = slice_by_index(begin = var_37191_begin_0, end = var_37191_end_0, end_mask = var_37191_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37191_cast_fp16")]; + tensor var_37198_begin_0 = const()[name = string("op_37198_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37198_end_0 = const()[name = string("op_37198_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37198_end_mask_0 = const()[name = string("op_37198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37198_cast_fp16 = slice_by_index(begin = var_37198_begin_0, end = var_37198_end_0, end_mask = var_37198_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37198_cast_fp16")]; + tensor var_37205_begin_0 = const()[name = string("op_37205_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37205_end_0 = const()[name = string("op_37205_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37205_end_mask_0 = const()[name = string("op_37205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37205_cast_fp16 = slice_by_index(begin = var_37205_begin_0, end = var_37205_end_0, end_mask = var_37205_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37205_cast_fp16")]; + tensor var_37212_begin_0 = const()[name = string("op_37212_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37212_end_0 = const()[name = string("op_37212_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37212_end_mask_0 = const()[name = string("op_37212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37212_cast_fp16 = slice_by_index(begin = var_37212_begin_0, end = var_37212_end_0, end_mask = var_37212_end_mask_0, x = var_36938_cast_fp16)[name = string("op_37212_cast_fp16")]; + tensor var_37219_begin_0 = const()[name = string("op_37219_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37219_end_0 = const()[name = string("op_37219_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37219_end_mask_0 = const()[name = string("op_37219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37219_cast_fp16 = slice_by_index(begin = var_37219_begin_0, end = var_37219_end_0, end_mask = var_37219_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37219_cast_fp16")]; + tensor var_37226_begin_0 = const()[name = string("op_37226_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37226_end_0 = const()[name = string("op_37226_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37226_end_mask_0 = const()[name = string("op_37226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37226_cast_fp16 = slice_by_index(begin = var_37226_begin_0, end = var_37226_end_0, end_mask = var_37226_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37226_cast_fp16")]; + tensor var_37233_begin_0 = const()[name = string("op_37233_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37233_end_0 = const()[name = string("op_37233_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37233_end_mask_0 = const()[name = string("op_37233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37233_cast_fp16 = slice_by_index(begin = var_37233_begin_0, end = var_37233_end_0, end_mask = var_37233_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37233_cast_fp16")]; + tensor var_37240_begin_0 = const()[name = string("op_37240_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37240_end_0 = const()[name = string("op_37240_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37240_end_mask_0 = const()[name = string("op_37240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37240_cast_fp16 = slice_by_index(begin = var_37240_begin_0, end = var_37240_end_0, end_mask = var_37240_end_mask_0, x = var_36942_cast_fp16)[name = string("op_37240_cast_fp16")]; + tensor var_37247_begin_0 = const()[name = string("op_37247_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37247_end_0 = const()[name = string("op_37247_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37247_end_mask_0 = const()[name = string("op_37247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37247_cast_fp16 = slice_by_index(begin = var_37247_begin_0, end = var_37247_end_0, end_mask = var_37247_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37247_cast_fp16")]; + tensor var_37254_begin_0 = const()[name = string("op_37254_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37254_end_0 = const()[name = string("op_37254_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37254_end_mask_0 = const()[name = string("op_37254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37254_cast_fp16 = slice_by_index(begin = var_37254_begin_0, end = var_37254_end_0, end_mask = var_37254_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37254_cast_fp16")]; + tensor var_37261_begin_0 = const()[name = string("op_37261_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37261_end_0 = const()[name = string("op_37261_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37261_end_mask_0 = const()[name = string("op_37261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37261_cast_fp16 = slice_by_index(begin = var_37261_begin_0, end = var_37261_end_0, end_mask = var_37261_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37261_cast_fp16")]; + tensor var_37268_begin_0 = const()[name = string("op_37268_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37268_end_0 = const()[name = string("op_37268_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37268_end_mask_0 = const()[name = string("op_37268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37268_cast_fp16 = slice_by_index(begin = var_37268_begin_0, end = var_37268_end_0, end_mask = var_37268_end_mask_0, x = var_36946_cast_fp16)[name = string("op_37268_cast_fp16")]; + tensor var_37275_begin_0 = const()[name = string("op_37275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37275_end_0 = const()[name = string("op_37275_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37275_end_mask_0 = const()[name = string("op_37275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37275_cast_fp16 = slice_by_index(begin = var_37275_begin_0, end = var_37275_end_0, end_mask = var_37275_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37275_cast_fp16")]; + tensor var_37282_begin_0 = const()[name = string("op_37282_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37282_end_0 = const()[name = string("op_37282_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37282_end_mask_0 = const()[name = string("op_37282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37282_cast_fp16 = slice_by_index(begin = var_37282_begin_0, end = var_37282_end_0, end_mask = var_37282_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37282_cast_fp16")]; + tensor var_37289_begin_0 = const()[name = string("op_37289_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37289_end_0 = const()[name = string("op_37289_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37289_end_mask_0 = const()[name = string("op_37289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37289_cast_fp16 = slice_by_index(begin = var_37289_begin_0, end = var_37289_end_0, end_mask = var_37289_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37289_cast_fp16")]; + tensor var_37296_begin_0 = const()[name = string("op_37296_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37296_end_0 = const()[name = string("op_37296_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37296_end_mask_0 = const()[name = string("op_37296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37296_cast_fp16 = slice_by_index(begin = var_37296_begin_0, end = var_37296_end_0, end_mask = var_37296_end_mask_0, x = var_36950_cast_fp16)[name = string("op_37296_cast_fp16")]; + tensor var_37303_begin_0 = const()[name = string("op_37303_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37303_end_0 = const()[name = string("op_37303_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37303_end_mask_0 = const()[name = string("op_37303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37303_cast_fp16 = slice_by_index(begin = var_37303_begin_0, end = var_37303_end_0, end_mask = var_37303_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37303_cast_fp16")]; + tensor var_37310_begin_0 = const()[name = string("op_37310_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37310_end_0 = const()[name = string("op_37310_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37310_end_mask_0 = const()[name = string("op_37310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37310_cast_fp16 = slice_by_index(begin = var_37310_begin_0, end = var_37310_end_0, end_mask = var_37310_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37310_cast_fp16")]; + tensor var_37317_begin_0 = const()[name = string("op_37317_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37317_end_0 = const()[name = string("op_37317_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37317_end_mask_0 = const()[name = string("op_37317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37317_cast_fp16 = slice_by_index(begin = var_37317_begin_0, end = var_37317_end_0, end_mask = var_37317_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37317_cast_fp16")]; + tensor var_37324_begin_0 = const()[name = string("op_37324_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37324_end_0 = const()[name = string("op_37324_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37324_end_mask_0 = const()[name = string("op_37324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37324_cast_fp16 = slice_by_index(begin = var_37324_begin_0, end = var_37324_end_0, end_mask = var_37324_end_mask_0, x = var_36954_cast_fp16)[name = string("op_37324_cast_fp16")]; + tensor var_37331_begin_0 = const()[name = string("op_37331_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37331_end_0 = const()[name = string("op_37331_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37331_end_mask_0 = const()[name = string("op_37331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37331_cast_fp16 = slice_by_index(begin = var_37331_begin_0, end = var_37331_end_0, end_mask = var_37331_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37331_cast_fp16")]; + tensor var_37338_begin_0 = const()[name = string("op_37338_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37338_end_0 = const()[name = string("op_37338_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37338_end_mask_0 = const()[name = string("op_37338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37338_cast_fp16 = slice_by_index(begin = var_37338_begin_0, end = var_37338_end_0, end_mask = var_37338_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37338_cast_fp16")]; + tensor var_37345_begin_0 = const()[name = string("op_37345_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37345_end_0 = const()[name = string("op_37345_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37345_end_mask_0 = const()[name = string("op_37345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37345_cast_fp16 = slice_by_index(begin = var_37345_begin_0, end = var_37345_end_0, end_mask = var_37345_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37345_cast_fp16")]; + tensor var_37352_begin_0 = const()[name = string("op_37352_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37352_end_0 = const()[name = string("op_37352_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37352_end_mask_0 = const()[name = string("op_37352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37352_cast_fp16 = slice_by_index(begin = var_37352_begin_0, end = var_37352_end_0, end_mask = var_37352_end_mask_0, x = var_36958_cast_fp16)[name = string("op_37352_cast_fp16")]; + tensor var_37359_begin_0 = const()[name = string("op_37359_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37359_end_0 = const()[name = string("op_37359_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37359_end_mask_0 = const()[name = string("op_37359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37359_cast_fp16 = slice_by_index(begin = var_37359_begin_0, end = var_37359_end_0, end_mask = var_37359_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37359_cast_fp16")]; + tensor var_37366_begin_0 = const()[name = string("op_37366_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37366_end_0 = const()[name = string("op_37366_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37366_end_mask_0 = const()[name = string("op_37366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37366_cast_fp16 = slice_by_index(begin = var_37366_begin_0, end = var_37366_end_0, end_mask = var_37366_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37366_cast_fp16")]; + tensor var_37373_begin_0 = const()[name = string("op_37373_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37373_end_0 = const()[name = string("op_37373_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37373_end_mask_0 = const()[name = string("op_37373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37373_cast_fp16 = slice_by_index(begin = var_37373_begin_0, end = var_37373_end_0, end_mask = var_37373_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37373_cast_fp16")]; + tensor var_37380_begin_0 = const()[name = string("op_37380_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37380_end_0 = const()[name = string("op_37380_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37380_end_mask_0 = const()[name = string("op_37380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37380_cast_fp16 = slice_by_index(begin = var_37380_begin_0, end = var_37380_end_0, end_mask = var_37380_end_mask_0, x = var_36962_cast_fp16)[name = string("op_37380_cast_fp16")]; + tensor var_37387_begin_0 = const()[name = string("op_37387_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37387_end_0 = const()[name = string("op_37387_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37387_end_mask_0 = const()[name = string("op_37387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37387_cast_fp16 = slice_by_index(begin = var_37387_begin_0, end = var_37387_end_0, end_mask = var_37387_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37387_cast_fp16")]; + tensor var_37394_begin_0 = const()[name = string("op_37394_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37394_end_0 = const()[name = string("op_37394_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37394_end_mask_0 = const()[name = string("op_37394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37394_cast_fp16 = slice_by_index(begin = var_37394_begin_0, end = var_37394_end_0, end_mask = var_37394_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37394_cast_fp16")]; + tensor var_37401_begin_0 = const()[name = string("op_37401_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37401_end_0 = const()[name = string("op_37401_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37401_end_mask_0 = const()[name = string("op_37401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37401_cast_fp16 = slice_by_index(begin = var_37401_begin_0, end = var_37401_end_0, end_mask = var_37401_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37401_cast_fp16")]; + tensor var_37408_begin_0 = const()[name = string("op_37408_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37408_end_0 = const()[name = string("op_37408_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37408_end_mask_0 = const()[name = string("op_37408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37408_cast_fp16 = slice_by_index(begin = var_37408_begin_0, end = var_37408_end_0, end_mask = var_37408_end_mask_0, x = var_36966_cast_fp16)[name = string("op_37408_cast_fp16")]; + tensor var_37415_begin_0 = const()[name = string("op_37415_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37415_end_0 = const()[name = string("op_37415_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37415_end_mask_0 = const()[name = string("op_37415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37415_cast_fp16 = slice_by_index(begin = var_37415_begin_0, end = var_37415_end_0, end_mask = var_37415_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37415_cast_fp16")]; + tensor var_37422_begin_0 = const()[name = string("op_37422_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37422_end_0 = const()[name = string("op_37422_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37422_end_mask_0 = const()[name = string("op_37422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37422_cast_fp16 = slice_by_index(begin = var_37422_begin_0, end = var_37422_end_0, end_mask = var_37422_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37422_cast_fp16")]; + tensor var_37429_begin_0 = const()[name = string("op_37429_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37429_end_0 = const()[name = string("op_37429_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37429_end_mask_0 = const()[name = string("op_37429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37429_cast_fp16 = slice_by_index(begin = var_37429_begin_0, end = var_37429_end_0, end_mask = var_37429_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37429_cast_fp16")]; + tensor var_37436_begin_0 = const()[name = string("op_37436_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37436_end_0 = const()[name = string("op_37436_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37436_end_mask_0 = const()[name = string("op_37436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37436_cast_fp16 = slice_by_index(begin = var_37436_begin_0, end = var_37436_end_0, end_mask = var_37436_end_mask_0, x = var_36970_cast_fp16)[name = string("op_37436_cast_fp16")]; + tensor var_37443_begin_0 = const()[name = string("op_37443_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37443_end_0 = const()[name = string("op_37443_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37443_end_mask_0 = const()[name = string("op_37443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37443_cast_fp16 = slice_by_index(begin = var_37443_begin_0, end = var_37443_end_0, end_mask = var_37443_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37443_cast_fp16")]; + tensor var_37450_begin_0 = const()[name = string("op_37450_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37450_end_0 = const()[name = string("op_37450_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37450_end_mask_0 = const()[name = string("op_37450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37450_cast_fp16 = slice_by_index(begin = var_37450_begin_0, end = var_37450_end_0, end_mask = var_37450_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37450_cast_fp16")]; + tensor var_37457_begin_0 = const()[name = string("op_37457_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37457_end_0 = const()[name = string("op_37457_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37457_end_mask_0 = const()[name = string("op_37457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37457_cast_fp16 = slice_by_index(begin = var_37457_begin_0, end = var_37457_end_0, end_mask = var_37457_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37457_cast_fp16")]; + tensor var_37464_begin_0 = const()[name = string("op_37464_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37464_end_0 = const()[name = string("op_37464_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37464_end_mask_0 = const()[name = string("op_37464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37464_cast_fp16 = slice_by_index(begin = var_37464_begin_0, end = var_37464_end_0, end_mask = var_37464_end_mask_0, x = var_36974_cast_fp16)[name = string("op_37464_cast_fp16")]; + tensor var_37471_begin_0 = const()[name = string("op_37471_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37471_end_0 = const()[name = string("op_37471_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37471_end_mask_0 = const()[name = string("op_37471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37471_cast_fp16 = slice_by_index(begin = var_37471_begin_0, end = var_37471_end_0, end_mask = var_37471_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37471_cast_fp16")]; + tensor var_37478_begin_0 = const()[name = string("op_37478_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37478_end_0 = const()[name = string("op_37478_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37478_end_mask_0 = const()[name = string("op_37478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37478_cast_fp16 = slice_by_index(begin = var_37478_begin_0, end = var_37478_end_0, end_mask = var_37478_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37478_cast_fp16")]; + tensor var_37485_begin_0 = const()[name = string("op_37485_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37485_end_0 = const()[name = string("op_37485_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37485_end_mask_0 = const()[name = string("op_37485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37485_cast_fp16 = slice_by_index(begin = var_37485_begin_0, end = var_37485_end_0, end_mask = var_37485_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37485_cast_fp16")]; + tensor var_37492_begin_0 = const()[name = string("op_37492_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37492_end_0 = const()[name = string("op_37492_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37492_end_mask_0 = const()[name = string("op_37492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37492_cast_fp16 = slice_by_index(begin = var_37492_begin_0, end = var_37492_end_0, end_mask = var_37492_end_mask_0, x = var_36978_cast_fp16)[name = string("op_37492_cast_fp16")]; + tensor var_37499_begin_0 = const()[name = string("op_37499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37499_end_0 = const()[name = string("op_37499_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37499_end_mask_0 = const()[name = string("op_37499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37499_cast_fp16 = slice_by_index(begin = var_37499_begin_0, end = var_37499_end_0, end_mask = var_37499_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37499_cast_fp16")]; + tensor var_37506_begin_0 = const()[name = string("op_37506_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37506_end_0 = const()[name = string("op_37506_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37506_end_mask_0 = const()[name = string("op_37506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37506_cast_fp16 = slice_by_index(begin = var_37506_begin_0, end = var_37506_end_0, end_mask = var_37506_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37506_cast_fp16")]; + tensor var_37513_begin_0 = const()[name = string("op_37513_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37513_end_0 = const()[name = string("op_37513_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37513_end_mask_0 = const()[name = string("op_37513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37513_cast_fp16 = slice_by_index(begin = var_37513_begin_0, end = var_37513_end_0, end_mask = var_37513_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37513_cast_fp16")]; + tensor var_37520_begin_0 = const()[name = string("op_37520_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37520_end_0 = const()[name = string("op_37520_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37520_end_mask_0 = const()[name = string("op_37520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37520_cast_fp16 = slice_by_index(begin = var_37520_begin_0, end = var_37520_end_0, end_mask = var_37520_end_mask_0, x = var_36982_cast_fp16)[name = string("op_37520_cast_fp16")]; + tensor var_37527_begin_0 = const()[name = string("op_37527_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37527_end_0 = const()[name = string("op_37527_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37527_end_mask_0 = const()[name = string("op_37527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37527_cast_fp16 = slice_by_index(begin = var_37527_begin_0, end = var_37527_end_0, end_mask = var_37527_end_mask_0, x = var_36986_cast_fp16)[name = string("op_37527_cast_fp16")]; + tensor var_37534_begin_0 = const()[name = string("op_37534_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_37534_end_0 = const()[name = string("op_37534_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_37534_end_mask_0 = const()[name = string("op_37534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37534_cast_fp16 = slice_by_index(begin = var_37534_begin_0, end = var_37534_end_0, end_mask = var_37534_end_mask_0, x = var_36986_cast_fp16)[name = string("op_37534_cast_fp16")]; + tensor var_37541_begin_0 = const()[name = string("op_37541_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_37541_end_0 = const()[name = string("op_37541_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_37541_end_mask_0 = const()[name = string("op_37541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37541_cast_fp16 = slice_by_index(begin = var_37541_begin_0, end = var_37541_end_0, end_mask = var_37541_end_mask_0, x = var_36986_cast_fp16)[name = string("op_37541_cast_fp16")]; + tensor var_37548_begin_0 = const()[name = string("op_37548_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_37548_end_0 = const()[name = string("op_37548_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37548_end_mask_0 = const()[name = string("op_37548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37548_cast_fp16 = slice_by_index(begin = var_37548_begin_0, end = var_37548_end_0, end_mask = var_37548_end_mask_0, x = var_36986_cast_fp16)[name = string("op_37548_cast_fp16")]; + tensor k_49_perm_0 = const()[name = string("k_49_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_37553_begin_0 = const()[name = string("op_37553_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37553_end_0 = const()[name = string("op_37553_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_37553_end_mask_0 = const()[name = string("op_37553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_49_cast_fp16 = transpose(perm = k_49_perm_0, x = key_49_cast_fp16)[name = string("transpose_7")]; + tensor var_37553_cast_fp16 = slice_by_index(begin = var_37553_begin_0, end = var_37553_end_0, end_mask = var_37553_end_mask_0, x = k_49_cast_fp16)[name = string("op_37553_cast_fp16")]; + tensor var_37557_begin_0 = const()[name = string("op_37557_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_37557_end_0 = const()[name = string("op_37557_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_37557_end_mask_0 = const()[name = string("op_37557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37557_cast_fp16 = slice_by_index(begin = var_37557_begin_0, end = var_37557_end_0, end_mask = var_37557_end_mask_0, x = k_49_cast_fp16)[name = string("op_37557_cast_fp16")]; + tensor var_37561_begin_0 = const()[name = string("op_37561_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_37561_end_0 = const()[name = string("op_37561_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_37561_end_mask_0 = const()[name = string("op_37561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37561_cast_fp16 = slice_by_index(begin = var_37561_begin_0, end = var_37561_end_0, end_mask = var_37561_end_mask_0, x = k_49_cast_fp16)[name = string("op_37561_cast_fp16")]; + tensor var_37565_begin_0 = const()[name = string("op_37565_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_37565_end_0 = const()[name = string("op_37565_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_37565_end_mask_0 = const()[name = string("op_37565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37565_cast_fp16 = slice_by_index(begin = var_37565_begin_0, end = var_37565_end_0, end_mask = var_37565_end_mask_0, x = k_49_cast_fp16)[name = string("op_37565_cast_fp16")]; + tensor var_37569_begin_0 = const()[name = string("op_37569_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_37569_end_0 = const()[name = string("op_37569_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_37569_end_mask_0 = const()[name = string("op_37569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37569_cast_fp16 = slice_by_index(begin = var_37569_begin_0, end = var_37569_end_0, end_mask = var_37569_end_mask_0, x = k_49_cast_fp16)[name = string("op_37569_cast_fp16")]; + tensor var_37573_begin_0 = const()[name = string("op_37573_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_37573_end_0 = const()[name = string("op_37573_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_37573_end_mask_0 = const()[name = string("op_37573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37573_cast_fp16 = slice_by_index(begin = var_37573_begin_0, end = var_37573_end_0, end_mask = var_37573_end_mask_0, x = k_49_cast_fp16)[name = string("op_37573_cast_fp16")]; + tensor var_37577_begin_0 = const()[name = string("op_37577_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_37577_end_0 = const()[name = string("op_37577_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_37577_end_mask_0 = const()[name = string("op_37577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37577_cast_fp16 = slice_by_index(begin = var_37577_begin_0, end = var_37577_end_0, end_mask = var_37577_end_mask_0, x = k_49_cast_fp16)[name = string("op_37577_cast_fp16")]; + tensor var_37581_begin_0 = const()[name = string("op_37581_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_37581_end_0 = const()[name = string("op_37581_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_37581_end_mask_0 = const()[name = string("op_37581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37581_cast_fp16 = slice_by_index(begin = var_37581_begin_0, end = var_37581_end_0, end_mask = var_37581_end_mask_0, x = k_49_cast_fp16)[name = string("op_37581_cast_fp16")]; + tensor var_37585_begin_0 = const()[name = string("op_37585_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_37585_end_0 = const()[name = string("op_37585_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_37585_end_mask_0 = const()[name = string("op_37585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37585_cast_fp16 = slice_by_index(begin = var_37585_begin_0, end = var_37585_end_0, end_mask = var_37585_end_mask_0, x = k_49_cast_fp16)[name = string("op_37585_cast_fp16")]; + tensor var_37589_begin_0 = const()[name = string("op_37589_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_37589_end_0 = const()[name = string("op_37589_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_37589_end_mask_0 = const()[name = string("op_37589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37589_cast_fp16 = slice_by_index(begin = var_37589_begin_0, end = var_37589_end_0, end_mask = var_37589_end_mask_0, x = k_49_cast_fp16)[name = string("op_37589_cast_fp16")]; + tensor var_37593_begin_0 = const()[name = string("op_37593_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_37593_end_0 = const()[name = string("op_37593_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_37593_end_mask_0 = const()[name = string("op_37593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37593_cast_fp16 = slice_by_index(begin = var_37593_begin_0, end = var_37593_end_0, end_mask = var_37593_end_mask_0, x = k_49_cast_fp16)[name = string("op_37593_cast_fp16")]; + tensor var_37597_begin_0 = const()[name = string("op_37597_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_37597_end_0 = const()[name = string("op_37597_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_37597_end_mask_0 = const()[name = string("op_37597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37597_cast_fp16 = slice_by_index(begin = var_37597_begin_0, end = var_37597_end_0, end_mask = var_37597_end_mask_0, x = k_49_cast_fp16)[name = string("op_37597_cast_fp16")]; + tensor var_37601_begin_0 = const()[name = string("op_37601_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_37601_end_0 = const()[name = string("op_37601_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_37601_end_mask_0 = const()[name = string("op_37601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37601_cast_fp16 = slice_by_index(begin = var_37601_begin_0, end = var_37601_end_0, end_mask = var_37601_end_mask_0, x = k_49_cast_fp16)[name = string("op_37601_cast_fp16")]; + tensor var_37605_begin_0 = const()[name = string("op_37605_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_37605_end_0 = const()[name = string("op_37605_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_37605_end_mask_0 = const()[name = string("op_37605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37605_cast_fp16 = slice_by_index(begin = var_37605_begin_0, end = var_37605_end_0, end_mask = var_37605_end_mask_0, x = k_49_cast_fp16)[name = string("op_37605_cast_fp16")]; + tensor var_37609_begin_0 = const()[name = string("op_37609_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_37609_end_0 = const()[name = string("op_37609_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_37609_end_mask_0 = const()[name = string("op_37609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37609_cast_fp16 = slice_by_index(begin = var_37609_begin_0, end = var_37609_end_0, end_mask = var_37609_end_mask_0, x = k_49_cast_fp16)[name = string("op_37609_cast_fp16")]; + tensor var_37613_begin_0 = const()[name = string("op_37613_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_37613_end_0 = const()[name = string("op_37613_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_37613_end_mask_0 = const()[name = string("op_37613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37613_cast_fp16 = slice_by_index(begin = var_37613_begin_0, end = var_37613_end_0, end_mask = var_37613_end_mask_0, x = k_49_cast_fp16)[name = string("op_37613_cast_fp16")]; + tensor var_37617_begin_0 = const()[name = string("op_37617_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_37617_end_0 = const()[name = string("op_37617_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_37617_end_mask_0 = const()[name = string("op_37617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37617_cast_fp16 = slice_by_index(begin = var_37617_begin_0, end = var_37617_end_0, end_mask = var_37617_end_mask_0, x = k_49_cast_fp16)[name = string("op_37617_cast_fp16")]; + tensor var_37621_begin_0 = const()[name = string("op_37621_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_37621_end_0 = const()[name = string("op_37621_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_37621_end_mask_0 = const()[name = string("op_37621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37621_cast_fp16 = slice_by_index(begin = var_37621_begin_0, end = var_37621_end_0, end_mask = var_37621_end_mask_0, x = k_49_cast_fp16)[name = string("op_37621_cast_fp16")]; + tensor var_37625_begin_0 = const()[name = string("op_37625_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_37625_end_0 = const()[name = string("op_37625_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_37625_end_mask_0 = const()[name = string("op_37625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37625_cast_fp16 = slice_by_index(begin = var_37625_begin_0, end = var_37625_end_0, end_mask = var_37625_end_mask_0, x = k_49_cast_fp16)[name = string("op_37625_cast_fp16")]; + tensor var_37629_begin_0 = const()[name = string("op_37629_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_37629_end_0 = const()[name = string("op_37629_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_37629_end_mask_0 = const()[name = string("op_37629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37629_cast_fp16 = slice_by_index(begin = var_37629_begin_0, end = var_37629_end_0, end_mask = var_37629_end_mask_0, x = k_49_cast_fp16)[name = string("op_37629_cast_fp16")]; + tensor var_37631_begin_0 = const()[name = string("op_37631_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37631_end_0 = const()[name = string("op_37631_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37631_end_mask_0 = const()[name = string("op_37631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37631_cast_fp16 = slice_by_index(begin = var_37631_begin_0, end = var_37631_end_0, end_mask = var_37631_end_mask_0, x = value_49_cast_fp16)[name = string("op_37631_cast_fp16")]; + tensor var_37635_begin_0 = const()[name = string("op_37635_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37635_end_0 = const()[name = string("op_37635_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_37635_end_mask_0 = const()[name = string("op_37635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37635_cast_fp16 = slice_by_index(begin = var_37635_begin_0, end = var_37635_end_0, end_mask = var_37635_end_mask_0, x = value_49_cast_fp16)[name = string("op_37635_cast_fp16")]; + tensor var_37639_begin_0 = const()[name = string("op_37639_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37639_end_0 = const()[name = string("op_37639_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_37639_end_mask_0 = const()[name = string("op_37639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37639_cast_fp16 = slice_by_index(begin = var_37639_begin_0, end = var_37639_end_0, end_mask = var_37639_end_mask_0, x = value_49_cast_fp16)[name = string("op_37639_cast_fp16")]; + tensor var_37643_begin_0 = const()[name = string("op_37643_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37643_end_0 = const()[name = string("op_37643_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_37643_end_mask_0 = const()[name = string("op_37643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37643_cast_fp16 = slice_by_index(begin = var_37643_begin_0, end = var_37643_end_0, end_mask = var_37643_end_mask_0, x = value_49_cast_fp16)[name = string("op_37643_cast_fp16")]; + tensor var_37647_begin_0 = const()[name = string("op_37647_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37647_end_0 = const()[name = string("op_37647_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_37647_end_mask_0 = const()[name = string("op_37647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37647_cast_fp16 = slice_by_index(begin = var_37647_begin_0, end = var_37647_end_0, end_mask = var_37647_end_mask_0, x = value_49_cast_fp16)[name = string("op_37647_cast_fp16")]; + tensor var_37651_begin_0 = const()[name = string("op_37651_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37651_end_0 = const()[name = string("op_37651_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_37651_end_mask_0 = const()[name = string("op_37651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37651_cast_fp16 = slice_by_index(begin = var_37651_begin_0, end = var_37651_end_0, end_mask = var_37651_end_mask_0, x = value_49_cast_fp16)[name = string("op_37651_cast_fp16")]; + tensor var_37655_begin_0 = const()[name = string("op_37655_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37655_end_0 = const()[name = string("op_37655_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_37655_end_mask_0 = const()[name = string("op_37655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37655_cast_fp16 = slice_by_index(begin = var_37655_begin_0, end = var_37655_end_0, end_mask = var_37655_end_mask_0, x = value_49_cast_fp16)[name = string("op_37655_cast_fp16")]; + tensor var_37659_begin_0 = const()[name = string("op_37659_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37659_end_0 = const()[name = string("op_37659_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_37659_end_mask_0 = const()[name = string("op_37659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37659_cast_fp16 = slice_by_index(begin = var_37659_begin_0, end = var_37659_end_0, end_mask = var_37659_end_mask_0, x = value_49_cast_fp16)[name = string("op_37659_cast_fp16")]; + tensor var_37663_begin_0 = const()[name = string("op_37663_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37663_end_0 = const()[name = string("op_37663_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_37663_end_mask_0 = const()[name = string("op_37663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37663_cast_fp16 = slice_by_index(begin = var_37663_begin_0, end = var_37663_end_0, end_mask = var_37663_end_mask_0, x = value_49_cast_fp16)[name = string("op_37663_cast_fp16")]; + tensor var_37667_begin_0 = const()[name = string("op_37667_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37667_end_0 = const()[name = string("op_37667_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_37667_end_mask_0 = const()[name = string("op_37667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37667_cast_fp16 = slice_by_index(begin = var_37667_begin_0, end = var_37667_end_0, end_mask = var_37667_end_mask_0, x = value_49_cast_fp16)[name = string("op_37667_cast_fp16")]; + tensor var_37671_begin_0 = const()[name = string("op_37671_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37671_end_0 = const()[name = string("op_37671_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_37671_end_mask_0 = const()[name = string("op_37671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37671_cast_fp16 = slice_by_index(begin = var_37671_begin_0, end = var_37671_end_0, end_mask = var_37671_end_mask_0, x = value_49_cast_fp16)[name = string("op_37671_cast_fp16")]; + tensor var_37675_begin_0 = const()[name = string("op_37675_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37675_end_0 = const()[name = string("op_37675_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_37675_end_mask_0 = const()[name = string("op_37675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37675_cast_fp16 = slice_by_index(begin = var_37675_begin_0, end = var_37675_end_0, end_mask = var_37675_end_mask_0, x = value_49_cast_fp16)[name = string("op_37675_cast_fp16")]; + tensor var_37679_begin_0 = const()[name = string("op_37679_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37679_end_0 = const()[name = string("op_37679_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_37679_end_mask_0 = const()[name = string("op_37679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37679_cast_fp16 = slice_by_index(begin = var_37679_begin_0, end = var_37679_end_0, end_mask = var_37679_end_mask_0, x = value_49_cast_fp16)[name = string("op_37679_cast_fp16")]; + tensor var_37683_begin_0 = const()[name = string("op_37683_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37683_end_0 = const()[name = string("op_37683_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_37683_end_mask_0 = const()[name = string("op_37683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37683_cast_fp16 = slice_by_index(begin = var_37683_begin_0, end = var_37683_end_0, end_mask = var_37683_end_mask_0, x = value_49_cast_fp16)[name = string("op_37683_cast_fp16")]; + tensor var_37687_begin_0 = const()[name = string("op_37687_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37687_end_0 = const()[name = string("op_37687_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_37687_end_mask_0 = const()[name = string("op_37687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37687_cast_fp16 = slice_by_index(begin = var_37687_begin_0, end = var_37687_end_0, end_mask = var_37687_end_mask_0, x = value_49_cast_fp16)[name = string("op_37687_cast_fp16")]; + tensor var_37691_begin_0 = const()[name = string("op_37691_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37691_end_0 = const()[name = string("op_37691_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_37691_end_mask_0 = const()[name = string("op_37691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37691_cast_fp16 = slice_by_index(begin = var_37691_begin_0, end = var_37691_end_0, end_mask = var_37691_end_mask_0, x = value_49_cast_fp16)[name = string("op_37691_cast_fp16")]; + tensor var_37695_begin_0 = const()[name = string("op_37695_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37695_end_0 = const()[name = string("op_37695_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_37695_end_mask_0 = const()[name = string("op_37695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37695_cast_fp16 = slice_by_index(begin = var_37695_begin_0, end = var_37695_end_0, end_mask = var_37695_end_mask_0, x = value_49_cast_fp16)[name = string("op_37695_cast_fp16")]; + tensor var_37699_begin_0 = const()[name = string("op_37699_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37699_end_0 = const()[name = string("op_37699_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_37699_end_mask_0 = const()[name = string("op_37699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37699_cast_fp16 = slice_by_index(begin = var_37699_begin_0, end = var_37699_end_0, end_mask = var_37699_end_mask_0, x = value_49_cast_fp16)[name = string("op_37699_cast_fp16")]; + tensor var_37703_begin_0 = const()[name = string("op_37703_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37703_end_0 = const()[name = string("op_37703_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_37703_end_mask_0 = const()[name = string("op_37703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37703_cast_fp16 = slice_by_index(begin = var_37703_begin_0, end = var_37703_end_0, end_mask = var_37703_end_mask_0, x = value_49_cast_fp16)[name = string("op_37703_cast_fp16")]; + tensor var_37707_begin_0 = const()[name = string("op_37707_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37707_end_0 = const()[name = string("op_37707_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_37707_end_mask_0 = const()[name = string("op_37707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37707_cast_fp16 = slice_by_index(begin = var_37707_begin_0, end = var_37707_end_0, end_mask = var_37707_end_mask_0, x = value_49_cast_fp16)[name = string("op_37707_cast_fp16")]; + string _SplitHeadsQ__mh_w_3841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3841_equation_0, values = (var_37553_cast_fp16, var_36995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3841_cast_fp16")]; + string _SplitHeadsQ__mh_w_3843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3843_equation_0, values = (var_37553_cast_fp16, var_37002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3843_cast_fp16")]; + string _SplitHeadsQ__mh_w_3845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3845_equation_0, values = (var_37553_cast_fp16, var_37009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3845_cast_fp16")]; + string _SplitHeadsQ__mh_w_3847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3847_equation_0, values = (var_37553_cast_fp16, var_37016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3847_cast_fp16")]; + string _SplitHeadsQ__mh_w_3849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3849_equation_0, values = (var_37557_cast_fp16, var_37023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3849_cast_fp16")]; + string _SplitHeadsQ__mh_w_3851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3851_equation_0, values = (var_37557_cast_fp16, var_37030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3851_cast_fp16")]; + string _SplitHeadsQ__mh_w_3853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3853_equation_0, values = (var_37557_cast_fp16, var_37037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3853_cast_fp16")]; + string _SplitHeadsQ__mh_w_3855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3855_equation_0, values = (var_37557_cast_fp16, var_37044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3855_cast_fp16")]; + string _SplitHeadsQ__mh_w_3857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3857_equation_0, values = (var_37561_cast_fp16, var_37051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3857_cast_fp16")]; + string _SplitHeadsQ__mh_w_3859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3859_equation_0, values = (var_37561_cast_fp16, var_37058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3859_cast_fp16")]; + string _SplitHeadsQ__mh_w_3861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3861_equation_0, values = (var_37561_cast_fp16, var_37065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3861_cast_fp16")]; + string _SplitHeadsQ__mh_w_3863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3863_equation_0, values = (var_37561_cast_fp16, var_37072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3863_cast_fp16")]; + string _SplitHeadsQ__mh_w_3865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3865_equation_0, values = (var_37565_cast_fp16, var_37079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3865_cast_fp16")]; + string _SplitHeadsQ__mh_w_3867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3867_equation_0, values = (var_37565_cast_fp16, var_37086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3867_cast_fp16")]; + string _SplitHeadsQ__mh_w_3869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3869_equation_0, values = (var_37565_cast_fp16, var_37093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3869_cast_fp16")]; + string _SplitHeadsQ__mh_w_3871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3871_equation_0, values = (var_37565_cast_fp16, var_37100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3871_cast_fp16")]; + string _SplitHeadsQ__mh_w_3873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3873_equation_0, values = (var_37569_cast_fp16, var_37107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3873_cast_fp16")]; + string _SplitHeadsQ__mh_w_3875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3875_equation_0, values = (var_37569_cast_fp16, var_37114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3875_cast_fp16")]; + string _SplitHeadsQ__mh_w_3877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3877_equation_0, values = (var_37569_cast_fp16, var_37121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3877_cast_fp16")]; + string _SplitHeadsQ__mh_w_3879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3879_equation_0, values = (var_37569_cast_fp16, var_37128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3879_cast_fp16")]; + string _SplitHeadsQ__mh_w_3881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3881_equation_0, values = (var_37573_cast_fp16, var_37135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3881_cast_fp16")]; + string _SplitHeadsQ__mh_w_3883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3883_equation_0, values = (var_37573_cast_fp16, var_37142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3883_cast_fp16")]; + string _SplitHeadsQ__mh_w_3885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3885_equation_0, values = (var_37573_cast_fp16, var_37149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3885_cast_fp16")]; + string _SplitHeadsQ__mh_w_3887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3887_equation_0, values = (var_37573_cast_fp16, var_37156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3887_cast_fp16")]; + string _SplitHeadsQ__mh_w_3889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3889_equation_0, values = (var_37577_cast_fp16, var_37163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3889_cast_fp16")]; + string _SplitHeadsQ__mh_w_3891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3891_equation_0, values = (var_37577_cast_fp16, var_37170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3891_cast_fp16")]; + string _SplitHeadsQ__mh_w_3893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3893_equation_0, values = (var_37577_cast_fp16, var_37177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3893_cast_fp16")]; + string _SplitHeadsQ__mh_w_3895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3895_equation_0, values = (var_37577_cast_fp16, var_37184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3895_cast_fp16")]; + string _SplitHeadsQ__mh_w_3897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3897_equation_0, values = (var_37581_cast_fp16, var_37191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3897_cast_fp16")]; + string _SplitHeadsQ__mh_w_3899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3899_equation_0, values = (var_37581_cast_fp16, var_37198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3899_cast_fp16")]; + string _SplitHeadsQ__mh_w_3901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3901_equation_0, values = (var_37581_cast_fp16, var_37205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3901_cast_fp16")]; + string _SplitHeadsQ__mh_w_3903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3903_equation_0, values = (var_37581_cast_fp16, var_37212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3903_cast_fp16")]; + string _SplitHeadsQ__mh_w_3905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3905_equation_0, values = (var_37585_cast_fp16, var_37219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3905_cast_fp16")]; + string _SplitHeadsQ__mh_w_3907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3907_equation_0, values = (var_37585_cast_fp16, var_37226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3907_cast_fp16")]; + string _SplitHeadsQ__mh_w_3909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3909_equation_0, values = (var_37585_cast_fp16, var_37233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3909_cast_fp16")]; + string _SplitHeadsQ__mh_w_3911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3911_equation_0, values = (var_37585_cast_fp16, var_37240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3911_cast_fp16")]; + string _SplitHeadsQ__mh_w_3913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3913_equation_0, values = (var_37589_cast_fp16, var_37247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3913_cast_fp16")]; + string _SplitHeadsQ__mh_w_3915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3915_equation_0, values = (var_37589_cast_fp16, var_37254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3915_cast_fp16")]; + string _SplitHeadsQ__mh_w_3917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3917_equation_0, values = (var_37589_cast_fp16, var_37261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3917_cast_fp16")]; + string _SplitHeadsQ__mh_w_3919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3919_equation_0, values = (var_37589_cast_fp16, var_37268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3919_cast_fp16")]; + string _SplitHeadsQ__mh_w_3921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3921_equation_0, values = (var_37593_cast_fp16, var_37275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3921_cast_fp16")]; + string _SplitHeadsQ__mh_w_3923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3923_equation_0, values = (var_37593_cast_fp16, var_37282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3923_cast_fp16")]; + string _SplitHeadsQ__mh_w_3925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3925_equation_0, values = (var_37593_cast_fp16, var_37289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3925_cast_fp16")]; + string _SplitHeadsQ__mh_w_3927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3927_equation_0, values = (var_37593_cast_fp16, var_37296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3927_cast_fp16")]; + string _SplitHeadsQ__mh_w_3929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3929_equation_0, values = (var_37597_cast_fp16, var_37303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3929_cast_fp16")]; + string _SplitHeadsQ__mh_w_3931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3931_equation_0, values = (var_37597_cast_fp16, var_37310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3931_cast_fp16")]; + string _SplitHeadsQ__mh_w_3933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3933_equation_0, values = (var_37597_cast_fp16, var_37317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3933_cast_fp16")]; + string _SplitHeadsQ__mh_w_3935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3935_equation_0, values = (var_37597_cast_fp16, var_37324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3935_cast_fp16")]; + string _SplitHeadsQ__mh_w_3937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3937_equation_0, values = (var_37601_cast_fp16, var_37331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3937_cast_fp16")]; + string _SplitHeadsQ__mh_w_3939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3939_equation_0, values = (var_37601_cast_fp16, var_37338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3939_cast_fp16")]; + string _SplitHeadsQ__mh_w_3941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3941_equation_0, values = (var_37601_cast_fp16, var_37345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3941_cast_fp16")]; + string _SplitHeadsQ__mh_w_3943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3943_equation_0, values = (var_37601_cast_fp16, var_37352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3943_cast_fp16")]; + string _SplitHeadsQ__mh_w_3945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3945_equation_0, values = (var_37605_cast_fp16, var_37359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3945_cast_fp16")]; + string _SplitHeadsQ__mh_w_3947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3947_equation_0, values = (var_37605_cast_fp16, var_37366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3947_cast_fp16")]; + string _SplitHeadsQ__mh_w_3949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3949_equation_0, values = (var_37605_cast_fp16, var_37373_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3949_cast_fp16")]; + string _SplitHeadsQ__mh_w_3951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3951_equation_0, values = (var_37605_cast_fp16, var_37380_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3951_cast_fp16")]; + string _SplitHeadsQ__mh_w_3953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3953_equation_0, values = (var_37609_cast_fp16, var_37387_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3953_cast_fp16")]; + string _SplitHeadsQ__mh_w_3955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3955_equation_0, values = (var_37609_cast_fp16, var_37394_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3955_cast_fp16")]; + string _SplitHeadsQ__mh_w_3957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3957_equation_0, values = (var_37609_cast_fp16, var_37401_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3957_cast_fp16")]; + string _SplitHeadsQ__mh_w_3959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3959_equation_0, values = (var_37609_cast_fp16, var_37408_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3959_cast_fp16")]; + string _SplitHeadsQ__mh_w_3961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3961_equation_0, values = (var_37613_cast_fp16, var_37415_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3961_cast_fp16")]; + string _SplitHeadsQ__mh_w_3963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3963_equation_0, values = (var_37613_cast_fp16, var_37422_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3963_cast_fp16")]; + string _SplitHeadsQ__mh_w_3965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3965_equation_0, values = (var_37613_cast_fp16, var_37429_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3965_cast_fp16")]; + string _SplitHeadsQ__mh_w_3967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3967_equation_0, values = (var_37613_cast_fp16, var_37436_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3967_cast_fp16")]; + string _SplitHeadsQ__mh_w_3969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3969_equation_0, values = (var_37617_cast_fp16, var_37443_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3969_cast_fp16")]; + string _SplitHeadsQ__mh_w_3971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3971_equation_0, values = (var_37617_cast_fp16, var_37450_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3971_cast_fp16")]; + string _SplitHeadsQ__mh_w_3973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3973_equation_0, values = (var_37617_cast_fp16, var_37457_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3973_cast_fp16")]; + string _SplitHeadsQ__mh_w_3975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3975_equation_0, values = (var_37617_cast_fp16, var_37464_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3975_cast_fp16")]; + string _SplitHeadsQ__mh_w_3977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3977_equation_0, values = (var_37621_cast_fp16, var_37471_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3977_cast_fp16")]; + string _SplitHeadsQ__mh_w_3979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3979_equation_0, values = (var_37621_cast_fp16, var_37478_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3979_cast_fp16")]; + string _SplitHeadsQ__mh_w_3981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3981_equation_0, values = (var_37621_cast_fp16, var_37485_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3981_cast_fp16")]; + string _SplitHeadsQ__mh_w_3983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3983_equation_0, values = (var_37621_cast_fp16, var_37492_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3983_cast_fp16")]; + string _SplitHeadsQ__mh_w_3985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3985_equation_0, values = (var_37625_cast_fp16, var_37499_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3985_cast_fp16")]; + string _SplitHeadsQ__mh_w_3987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3987_equation_0, values = (var_37625_cast_fp16, var_37506_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3987_cast_fp16")]; + string _SplitHeadsQ__mh_w_3989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3989_equation_0, values = (var_37625_cast_fp16, var_37513_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3989_cast_fp16")]; + string _SplitHeadsQ__mh_w_3991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3991_equation_0, values = (var_37625_cast_fp16, var_37520_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3991_cast_fp16")]; + string _SplitHeadsQ__mh_w_3993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3993_equation_0, values = (var_37629_cast_fp16, var_37527_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3993_cast_fp16")]; + string _SplitHeadsQ__mh_w_3995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3995_equation_0, values = (var_37629_cast_fp16, var_37534_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3995_cast_fp16")]; + string _SplitHeadsQ__mh_w_3997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3997_equation_0, values = (var_37629_cast_fp16, var_37541_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3997_cast_fp16")]; + string _SplitHeadsQ__mh_w_3999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3999_equation_0, values = (var_37629_cast_fp16, var_37548_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3999_cast_fp16")]; + fp16 var_37870_to_fp16 = const()[name = string("op_37870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3841_cast_fp16, y = var_37870_to_fp16)[name = string("aw_chunk_3841_cast_fp16")]; + fp16 var_37872_to_fp16 = const()[name = string("op_37872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3843_cast_fp16, y = var_37872_to_fp16)[name = string("aw_chunk_3843_cast_fp16")]; + fp16 var_37874_to_fp16 = const()[name = string("op_37874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3845_cast_fp16, y = var_37874_to_fp16)[name = string("aw_chunk_3845_cast_fp16")]; + fp16 var_37876_to_fp16 = const()[name = string("op_37876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3847_cast_fp16, y = var_37876_to_fp16)[name = string("aw_chunk_3847_cast_fp16")]; + fp16 var_37878_to_fp16 = const()[name = string("op_37878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3849_cast_fp16, y = var_37878_to_fp16)[name = string("aw_chunk_3849_cast_fp16")]; + fp16 var_37880_to_fp16 = const()[name = string("op_37880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3851_cast_fp16, y = var_37880_to_fp16)[name = string("aw_chunk_3851_cast_fp16")]; + fp16 var_37882_to_fp16 = const()[name = string("op_37882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3853_cast_fp16, y = var_37882_to_fp16)[name = string("aw_chunk_3853_cast_fp16")]; + fp16 var_37884_to_fp16 = const()[name = string("op_37884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3855_cast_fp16, y = var_37884_to_fp16)[name = string("aw_chunk_3855_cast_fp16")]; + fp16 var_37886_to_fp16 = const()[name = string("op_37886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3857_cast_fp16, y = var_37886_to_fp16)[name = string("aw_chunk_3857_cast_fp16")]; + fp16 var_37888_to_fp16 = const()[name = string("op_37888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3859_cast_fp16, y = var_37888_to_fp16)[name = string("aw_chunk_3859_cast_fp16")]; + fp16 var_37890_to_fp16 = const()[name = string("op_37890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3861_cast_fp16, y = var_37890_to_fp16)[name = string("aw_chunk_3861_cast_fp16")]; + fp16 var_37892_to_fp16 = const()[name = string("op_37892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3863_cast_fp16, y = var_37892_to_fp16)[name = string("aw_chunk_3863_cast_fp16")]; + fp16 var_37894_to_fp16 = const()[name = string("op_37894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3865_cast_fp16, y = var_37894_to_fp16)[name = string("aw_chunk_3865_cast_fp16")]; + fp16 var_37896_to_fp16 = const()[name = string("op_37896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3867_cast_fp16, y = var_37896_to_fp16)[name = string("aw_chunk_3867_cast_fp16")]; + fp16 var_37898_to_fp16 = const()[name = string("op_37898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3869_cast_fp16, y = var_37898_to_fp16)[name = string("aw_chunk_3869_cast_fp16")]; + fp16 var_37900_to_fp16 = const()[name = string("op_37900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3871_cast_fp16, y = var_37900_to_fp16)[name = string("aw_chunk_3871_cast_fp16")]; + fp16 var_37902_to_fp16 = const()[name = string("op_37902_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3873_cast_fp16, y = var_37902_to_fp16)[name = string("aw_chunk_3873_cast_fp16")]; + fp16 var_37904_to_fp16 = const()[name = string("op_37904_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3875_cast_fp16, y = var_37904_to_fp16)[name = string("aw_chunk_3875_cast_fp16")]; + fp16 var_37906_to_fp16 = const()[name = string("op_37906_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3877_cast_fp16, y = var_37906_to_fp16)[name = string("aw_chunk_3877_cast_fp16")]; + fp16 var_37908_to_fp16 = const()[name = string("op_37908_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3879_cast_fp16, y = var_37908_to_fp16)[name = string("aw_chunk_3879_cast_fp16")]; + fp16 var_37910_to_fp16 = const()[name = string("op_37910_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3881_cast_fp16, y = var_37910_to_fp16)[name = string("aw_chunk_3881_cast_fp16")]; + fp16 var_37912_to_fp16 = const()[name = string("op_37912_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3883_cast_fp16, y = var_37912_to_fp16)[name = string("aw_chunk_3883_cast_fp16")]; + fp16 var_37914_to_fp16 = const()[name = string("op_37914_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3885_cast_fp16, y = var_37914_to_fp16)[name = string("aw_chunk_3885_cast_fp16")]; + fp16 var_37916_to_fp16 = const()[name = string("op_37916_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3887_cast_fp16, y = var_37916_to_fp16)[name = string("aw_chunk_3887_cast_fp16")]; + fp16 var_37918_to_fp16 = const()[name = string("op_37918_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3889_cast_fp16, y = var_37918_to_fp16)[name = string("aw_chunk_3889_cast_fp16")]; + fp16 var_37920_to_fp16 = const()[name = string("op_37920_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3891_cast_fp16, y = var_37920_to_fp16)[name = string("aw_chunk_3891_cast_fp16")]; + fp16 var_37922_to_fp16 = const()[name = string("op_37922_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3893_cast_fp16, y = var_37922_to_fp16)[name = string("aw_chunk_3893_cast_fp16")]; + fp16 var_37924_to_fp16 = const()[name = string("op_37924_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3895_cast_fp16, y = var_37924_to_fp16)[name = string("aw_chunk_3895_cast_fp16")]; + fp16 var_37926_to_fp16 = const()[name = string("op_37926_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3897_cast_fp16, y = var_37926_to_fp16)[name = string("aw_chunk_3897_cast_fp16")]; + fp16 var_37928_to_fp16 = const()[name = string("op_37928_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3899_cast_fp16, y = var_37928_to_fp16)[name = string("aw_chunk_3899_cast_fp16")]; + fp16 var_37930_to_fp16 = const()[name = string("op_37930_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3901_cast_fp16, y = var_37930_to_fp16)[name = string("aw_chunk_3901_cast_fp16")]; + fp16 var_37932_to_fp16 = const()[name = string("op_37932_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3903_cast_fp16, y = var_37932_to_fp16)[name = string("aw_chunk_3903_cast_fp16")]; + fp16 var_37934_to_fp16 = const()[name = string("op_37934_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3905_cast_fp16, y = var_37934_to_fp16)[name = string("aw_chunk_3905_cast_fp16")]; + fp16 var_37936_to_fp16 = const()[name = string("op_37936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3907_cast_fp16, y = var_37936_to_fp16)[name = string("aw_chunk_3907_cast_fp16")]; + fp16 var_37938_to_fp16 = const()[name = string("op_37938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3909_cast_fp16, y = var_37938_to_fp16)[name = string("aw_chunk_3909_cast_fp16")]; + fp16 var_37940_to_fp16 = const()[name = string("op_37940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3911_cast_fp16, y = var_37940_to_fp16)[name = string("aw_chunk_3911_cast_fp16")]; + fp16 var_37942_to_fp16 = const()[name = string("op_37942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3913_cast_fp16, y = var_37942_to_fp16)[name = string("aw_chunk_3913_cast_fp16")]; + fp16 var_37944_to_fp16 = const()[name = string("op_37944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3915_cast_fp16, y = var_37944_to_fp16)[name = string("aw_chunk_3915_cast_fp16")]; + fp16 var_37946_to_fp16 = const()[name = string("op_37946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3917_cast_fp16, y = var_37946_to_fp16)[name = string("aw_chunk_3917_cast_fp16")]; + fp16 var_37948_to_fp16 = const()[name = string("op_37948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3919_cast_fp16, y = var_37948_to_fp16)[name = string("aw_chunk_3919_cast_fp16")]; + fp16 var_37950_to_fp16 = const()[name = string("op_37950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3921_cast_fp16, y = var_37950_to_fp16)[name = string("aw_chunk_3921_cast_fp16")]; + fp16 var_37952_to_fp16 = const()[name = string("op_37952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3923_cast_fp16, y = var_37952_to_fp16)[name = string("aw_chunk_3923_cast_fp16")]; + fp16 var_37954_to_fp16 = const()[name = string("op_37954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3925_cast_fp16, y = var_37954_to_fp16)[name = string("aw_chunk_3925_cast_fp16")]; + fp16 var_37956_to_fp16 = const()[name = string("op_37956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3927_cast_fp16, y = var_37956_to_fp16)[name = string("aw_chunk_3927_cast_fp16")]; + fp16 var_37958_to_fp16 = const()[name = string("op_37958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3929_cast_fp16, y = var_37958_to_fp16)[name = string("aw_chunk_3929_cast_fp16")]; + fp16 var_37960_to_fp16 = const()[name = string("op_37960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3931_cast_fp16, y = var_37960_to_fp16)[name = string("aw_chunk_3931_cast_fp16")]; + fp16 var_37962_to_fp16 = const()[name = string("op_37962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3933_cast_fp16, y = var_37962_to_fp16)[name = string("aw_chunk_3933_cast_fp16")]; + fp16 var_37964_to_fp16 = const()[name = string("op_37964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3935_cast_fp16, y = var_37964_to_fp16)[name = string("aw_chunk_3935_cast_fp16")]; + fp16 var_37966_to_fp16 = const()[name = string("op_37966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3937_cast_fp16, y = var_37966_to_fp16)[name = string("aw_chunk_3937_cast_fp16")]; + fp16 var_37968_to_fp16 = const()[name = string("op_37968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3939_cast_fp16, y = var_37968_to_fp16)[name = string("aw_chunk_3939_cast_fp16")]; + fp16 var_37970_to_fp16 = const()[name = string("op_37970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3941_cast_fp16, y = var_37970_to_fp16)[name = string("aw_chunk_3941_cast_fp16")]; + fp16 var_37972_to_fp16 = const()[name = string("op_37972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3943_cast_fp16, y = var_37972_to_fp16)[name = string("aw_chunk_3943_cast_fp16")]; + fp16 var_37974_to_fp16 = const()[name = string("op_37974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3945_cast_fp16, y = var_37974_to_fp16)[name = string("aw_chunk_3945_cast_fp16")]; + fp16 var_37976_to_fp16 = const()[name = string("op_37976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3947_cast_fp16, y = var_37976_to_fp16)[name = string("aw_chunk_3947_cast_fp16")]; + fp16 var_37978_to_fp16 = const()[name = string("op_37978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3949_cast_fp16, y = var_37978_to_fp16)[name = string("aw_chunk_3949_cast_fp16")]; + fp16 var_37980_to_fp16 = const()[name = string("op_37980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3951_cast_fp16, y = var_37980_to_fp16)[name = string("aw_chunk_3951_cast_fp16")]; + fp16 var_37982_to_fp16 = const()[name = string("op_37982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3953_cast_fp16, y = var_37982_to_fp16)[name = string("aw_chunk_3953_cast_fp16")]; + fp16 var_37984_to_fp16 = const()[name = string("op_37984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3955_cast_fp16, y = var_37984_to_fp16)[name = string("aw_chunk_3955_cast_fp16")]; + fp16 var_37986_to_fp16 = const()[name = string("op_37986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3957_cast_fp16, y = var_37986_to_fp16)[name = string("aw_chunk_3957_cast_fp16")]; + fp16 var_37988_to_fp16 = const()[name = string("op_37988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3959_cast_fp16, y = var_37988_to_fp16)[name = string("aw_chunk_3959_cast_fp16")]; + fp16 var_37990_to_fp16 = const()[name = string("op_37990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3961_cast_fp16, y = var_37990_to_fp16)[name = string("aw_chunk_3961_cast_fp16")]; + fp16 var_37992_to_fp16 = const()[name = string("op_37992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3963_cast_fp16, y = var_37992_to_fp16)[name = string("aw_chunk_3963_cast_fp16")]; + fp16 var_37994_to_fp16 = const()[name = string("op_37994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3965_cast_fp16, y = var_37994_to_fp16)[name = string("aw_chunk_3965_cast_fp16")]; + fp16 var_37996_to_fp16 = const()[name = string("op_37996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3967_cast_fp16, y = var_37996_to_fp16)[name = string("aw_chunk_3967_cast_fp16")]; + fp16 var_37998_to_fp16 = const()[name = string("op_37998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3969_cast_fp16, y = var_37998_to_fp16)[name = string("aw_chunk_3969_cast_fp16")]; + fp16 var_38000_to_fp16 = const()[name = string("op_38000_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3971_cast_fp16, y = var_38000_to_fp16)[name = string("aw_chunk_3971_cast_fp16")]; + fp16 var_38002_to_fp16 = const()[name = string("op_38002_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3973_cast_fp16, y = var_38002_to_fp16)[name = string("aw_chunk_3973_cast_fp16")]; + fp16 var_38004_to_fp16 = const()[name = string("op_38004_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3975_cast_fp16, y = var_38004_to_fp16)[name = string("aw_chunk_3975_cast_fp16")]; + fp16 var_38006_to_fp16 = const()[name = string("op_38006_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3977_cast_fp16, y = var_38006_to_fp16)[name = string("aw_chunk_3977_cast_fp16")]; + fp16 var_38008_to_fp16 = const()[name = string("op_38008_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3979_cast_fp16, y = var_38008_to_fp16)[name = string("aw_chunk_3979_cast_fp16")]; + fp16 var_38010_to_fp16 = const()[name = string("op_38010_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3981_cast_fp16, y = var_38010_to_fp16)[name = string("aw_chunk_3981_cast_fp16")]; + fp16 var_38012_to_fp16 = const()[name = string("op_38012_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3983_cast_fp16, y = var_38012_to_fp16)[name = string("aw_chunk_3983_cast_fp16")]; + fp16 var_38014_to_fp16 = const()[name = string("op_38014_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3985_cast_fp16, y = var_38014_to_fp16)[name = string("aw_chunk_3985_cast_fp16")]; + fp16 var_38016_to_fp16 = const()[name = string("op_38016_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3987_cast_fp16, y = var_38016_to_fp16)[name = string("aw_chunk_3987_cast_fp16")]; + fp16 var_38018_to_fp16 = const()[name = string("op_38018_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3989_cast_fp16, y = var_38018_to_fp16)[name = string("aw_chunk_3989_cast_fp16")]; + fp16 var_38020_to_fp16 = const()[name = string("op_38020_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3991_cast_fp16, y = var_38020_to_fp16)[name = string("aw_chunk_3991_cast_fp16")]; + fp16 var_38022_to_fp16 = const()[name = string("op_38022_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3993_cast_fp16, y = var_38022_to_fp16)[name = string("aw_chunk_3993_cast_fp16")]; + fp16 var_38024_to_fp16 = const()[name = string("op_38024_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3995_cast_fp16, y = var_38024_to_fp16)[name = string("aw_chunk_3995_cast_fp16")]; + fp16 var_38026_to_fp16 = const()[name = string("op_38026_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3997_cast_fp16, y = var_38026_to_fp16)[name = string("aw_chunk_3997_cast_fp16")]; + fp16 var_38028_to_fp16 = const()[name = string("op_38028_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3999_cast_fp16, y = var_38028_to_fp16)[name = string("aw_chunk_3999_cast_fp16")]; + tensor var_38030_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3841_cast_fp16)[name = string("op_38030_cast_fp16")]; + tensor var_38031_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3843_cast_fp16)[name = string("op_38031_cast_fp16")]; + tensor var_38032_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3845_cast_fp16)[name = string("op_38032_cast_fp16")]; + tensor var_38033_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3847_cast_fp16)[name = string("op_38033_cast_fp16")]; + tensor var_38034_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3849_cast_fp16)[name = string("op_38034_cast_fp16")]; + tensor var_38035_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3851_cast_fp16)[name = string("op_38035_cast_fp16")]; + tensor var_38036_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3853_cast_fp16)[name = string("op_38036_cast_fp16")]; + tensor var_38037_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3855_cast_fp16)[name = string("op_38037_cast_fp16")]; + tensor var_38038_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3857_cast_fp16)[name = string("op_38038_cast_fp16")]; + tensor var_38039_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3859_cast_fp16)[name = string("op_38039_cast_fp16")]; + tensor var_38040_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3861_cast_fp16)[name = string("op_38040_cast_fp16")]; + tensor var_38041_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3863_cast_fp16)[name = string("op_38041_cast_fp16")]; + tensor var_38042_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3865_cast_fp16)[name = string("op_38042_cast_fp16")]; + tensor var_38043_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3867_cast_fp16)[name = string("op_38043_cast_fp16")]; + tensor var_38044_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3869_cast_fp16)[name = string("op_38044_cast_fp16")]; + tensor var_38045_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3871_cast_fp16)[name = string("op_38045_cast_fp16")]; + tensor var_38046_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3873_cast_fp16)[name = string("op_38046_cast_fp16")]; + tensor var_38047_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3875_cast_fp16)[name = string("op_38047_cast_fp16")]; + tensor var_38048_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3877_cast_fp16)[name = string("op_38048_cast_fp16")]; + tensor var_38049_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3879_cast_fp16)[name = string("op_38049_cast_fp16")]; + tensor var_38050_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3881_cast_fp16)[name = string("op_38050_cast_fp16")]; + tensor var_38051_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3883_cast_fp16)[name = string("op_38051_cast_fp16")]; + tensor var_38052_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3885_cast_fp16)[name = string("op_38052_cast_fp16")]; + tensor var_38053_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3887_cast_fp16)[name = string("op_38053_cast_fp16")]; + tensor var_38054_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3889_cast_fp16)[name = string("op_38054_cast_fp16")]; + tensor var_38055_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3891_cast_fp16)[name = string("op_38055_cast_fp16")]; + tensor var_38056_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3893_cast_fp16)[name = string("op_38056_cast_fp16")]; + tensor var_38057_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3895_cast_fp16)[name = string("op_38057_cast_fp16")]; + tensor var_38058_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3897_cast_fp16)[name = string("op_38058_cast_fp16")]; + tensor var_38059_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3899_cast_fp16)[name = string("op_38059_cast_fp16")]; + tensor var_38060_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3901_cast_fp16)[name = string("op_38060_cast_fp16")]; + tensor var_38061_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3903_cast_fp16)[name = string("op_38061_cast_fp16")]; + tensor var_38062_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3905_cast_fp16)[name = string("op_38062_cast_fp16")]; + tensor var_38063_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3907_cast_fp16)[name = string("op_38063_cast_fp16")]; + tensor var_38064_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3909_cast_fp16)[name = string("op_38064_cast_fp16")]; + tensor var_38065_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3911_cast_fp16)[name = string("op_38065_cast_fp16")]; + tensor var_38066_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3913_cast_fp16)[name = string("op_38066_cast_fp16")]; + tensor var_38067_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3915_cast_fp16)[name = string("op_38067_cast_fp16")]; + tensor var_38068_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3917_cast_fp16)[name = string("op_38068_cast_fp16")]; + tensor var_38069_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3919_cast_fp16)[name = string("op_38069_cast_fp16")]; + tensor var_38070_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3921_cast_fp16)[name = string("op_38070_cast_fp16")]; + tensor var_38071_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3923_cast_fp16)[name = string("op_38071_cast_fp16")]; + tensor var_38072_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3925_cast_fp16)[name = string("op_38072_cast_fp16")]; + tensor var_38073_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3927_cast_fp16)[name = string("op_38073_cast_fp16")]; + tensor var_38074_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3929_cast_fp16)[name = string("op_38074_cast_fp16")]; + tensor var_38075_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3931_cast_fp16)[name = string("op_38075_cast_fp16")]; + tensor var_38076_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3933_cast_fp16)[name = string("op_38076_cast_fp16")]; + tensor var_38077_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3935_cast_fp16)[name = string("op_38077_cast_fp16")]; + tensor var_38078_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3937_cast_fp16)[name = string("op_38078_cast_fp16")]; + tensor var_38079_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3939_cast_fp16)[name = string("op_38079_cast_fp16")]; + tensor var_38080_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3941_cast_fp16)[name = string("op_38080_cast_fp16")]; + tensor var_38081_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3943_cast_fp16)[name = string("op_38081_cast_fp16")]; + tensor var_38082_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3945_cast_fp16)[name = string("op_38082_cast_fp16")]; + tensor var_38083_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3947_cast_fp16)[name = string("op_38083_cast_fp16")]; + tensor var_38084_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3949_cast_fp16)[name = string("op_38084_cast_fp16")]; + tensor var_38085_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3951_cast_fp16)[name = string("op_38085_cast_fp16")]; + tensor var_38086_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3953_cast_fp16)[name = string("op_38086_cast_fp16")]; + tensor var_38087_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3955_cast_fp16)[name = string("op_38087_cast_fp16")]; + tensor var_38088_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3957_cast_fp16)[name = string("op_38088_cast_fp16")]; + tensor var_38089_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3959_cast_fp16)[name = string("op_38089_cast_fp16")]; + tensor var_38090_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3961_cast_fp16)[name = string("op_38090_cast_fp16")]; + tensor var_38091_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3963_cast_fp16)[name = string("op_38091_cast_fp16")]; + tensor var_38092_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3965_cast_fp16)[name = string("op_38092_cast_fp16")]; + tensor var_38093_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3967_cast_fp16)[name = string("op_38093_cast_fp16")]; + tensor var_38094_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3969_cast_fp16)[name = string("op_38094_cast_fp16")]; + tensor var_38095_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3971_cast_fp16)[name = string("op_38095_cast_fp16")]; + tensor var_38096_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3973_cast_fp16)[name = string("op_38096_cast_fp16")]; + tensor var_38097_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3975_cast_fp16)[name = string("op_38097_cast_fp16")]; + tensor var_38098_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3977_cast_fp16)[name = string("op_38098_cast_fp16")]; + tensor var_38099_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3979_cast_fp16)[name = string("op_38099_cast_fp16")]; + tensor var_38100_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3981_cast_fp16)[name = string("op_38100_cast_fp16")]; + tensor var_38101_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3983_cast_fp16)[name = string("op_38101_cast_fp16")]; + tensor var_38102_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3985_cast_fp16)[name = string("op_38102_cast_fp16")]; + tensor var_38103_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3987_cast_fp16)[name = string("op_38103_cast_fp16")]; + tensor var_38104_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3989_cast_fp16)[name = string("op_38104_cast_fp16")]; + tensor var_38105_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3991_cast_fp16)[name = string("op_38105_cast_fp16")]; + tensor var_38106_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3993_cast_fp16)[name = string("op_38106_cast_fp16")]; + tensor var_38107_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3995_cast_fp16)[name = string("op_38107_cast_fp16")]; + tensor var_38108_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3997_cast_fp16)[name = string("op_38108_cast_fp16")]; + tensor var_38109_cast_fp16 = softmax(axis = var_36855, x = aw_chunk_3999_cast_fp16)[name = string("op_38109_cast_fp16")]; + string var_38111_equation_0 = const()[name = string("op_38111_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38111_cast_fp16 = einsum(equation = var_38111_equation_0, values = (var_37631_cast_fp16, var_38030_cast_fp16))[name = string("op_38111_cast_fp16")]; + string var_38113_equation_0 = const()[name = string("op_38113_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38113_cast_fp16 = einsum(equation = var_38113_equation_0, values = (var_37631_cast_fp16, var_38031_cast_fp16))[name = string("op_38113_cast_fp16")]; + string var_38115_equation_0 = const()[name = string("op_38115_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38115_cast_fp16 = einsum(equation = var_38115_equation_0, values = (var_37631_cast_fp16, var_38032_cast_fp16))[name = string("op_38115_cast_fp16")]; + string var_38117_equation_0 = const()[name = string("op_38117_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38117_cast_fp16 = einsum(equation = var_38117_equation_0, values = (var_37631_cast_fp16, var_38033_cast_fp16))[name = string("op_38117_cast_fp16")]; + string var_38119_equation_0 = const()[name = string("op_38119_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38119_cast_fp16 = einsum(equation = var_38119_equation_0, values = (var_37635_cast_fp16, var_38034_cast_fp16))[name = string("op_38119_cast_fp16")]; + string var_38121_equation_0 = const()[name = string("op_38121_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38121_cast_fp16 = einsum(equation = var_38121_equation_0, values = (var_37635_cast_fp16, var_38035_cast_fp16))[name = string("op_38121_cast_fp16")]; + string var_38123_equation_0 = const()[name = string("op_38123_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38123_cast_fp16 = einsum(equation = var_38123_equation_0, values = (var_37635_cast_fp16, var_38036_cast_fp16))[name = string("op_38123_cast_fp16")]; + string var_38125_equation_0 = const()[name = string("op_38125_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38125_cast_fp16 = einsum(equation = var_38125_equation_0, values = (var_37635_cast_fp16, var_38037_cast_fp16))[name = string("op_38125_cast_fp16")]; + string var_38127_equation_0 = const()[name = string("op_38127_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38127_cast_fp16 = einsum(equation = var_38127_equation_0, values = (var_37639_cast_fp16, var_38038_cast_fp16))[name = string("op_38127_cast_fp16")]; + string var_38129_equation_0 = const()[name = string("op_38129_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38129_cast_fp16 = einsum(equation = var_38129_equation_0, values = (var_37639_cast_fp16, var_38039_cast_fp16))[name = string("op_38129_cast_fp16")]; + string var_38131_equation_0 = const()[name = string("op_38131_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38131_cast_fp16 = einsum(equation = var_38131_equation_0, values = (var_37639_cast_fp16, var_38040_cast_fp16))[name = string("op_38131_cast_fp16")]; + string var_38133_equation_0 = const()[name = string("op_38133_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38133_cast_fp16 = einsum(equation = var_38133_equation_0, values = (var_37639_cast_fp16, var_38041_cast_fp16))[name = string("op_38133_cast_fp16")]; + string var_38135_equation_0 = const()[name = string("op_38135_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38135_cast_fp16 = einsum(equation = var_38135_equation_0, values = (var_37643_cast_fp16, var_38042_cast_fp16))[name = string("op_38135_cast_fp16")]; + string var_38137_equation_0 = const()[name = string("op_38137_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38137_cast_fp16 = einsum(equation = var_38137_equation_0, values = (var_37643_cast_fp16, var_38043_cast_fp16))[name = string("op_38137_cast_fp16")]; + string var_38139_equation_0 = const()[name = string("op_38139_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38139_cast_fp16 = einsum(equation = var_38139_equation_0, values = (var_37643_cast_fp16, var_38044_cast_fp16))[name = string("op_38139_cast_fp16")]; + string var_38141_equation_0 = const()[name = string("op_38141_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38141_cast_fp16 = einsum(equation = var_38141_equation_0, values = (var_37643_cast_fp16, var_38045_cast_fp16))[name = string("op_38141_cast_fp16")]; + string var_38143_equation_0 = const()[name = string("op_38143_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38143_cast_fp16 = einsum(equation = var_38143_equation_0, values = (var_37647_cast_fp16, var_38046_cast_fp16))[name = string("op_38143_cast_fp16")]; + string var_38145_equation_0 = const()[name = string("op_38145_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38145_cast_fp16 = einsum(equation = var_38145_equation_0, values = (var_37647_cast_fp16, var_38047_cast_fp16))[name = string("op_38145_cast_fp16")]; + string var_38147_equation_0 = const()[name = string("op_38147_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38147_cast_fp16 = einsum(equation = var_38147_equation_0, values = (var_37647_cast_fp16, var_38048_cast_fp16))[name = string("op_38147_cast_fp16")]; + string var_38149_equation_0 = const()[name = string("op_38149_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38149_cast_fp16 = einsum(equation = var_38149_equation_0, values = (var_37647_cast_fp16, var_38049_cast_fp16))[name = string("op_38149_cast_fp16")]; + string var_38151_equation_0 = const()[name = string("op_38151_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38151_cast_fp16 = einsum(equation = var_38151_equation_0, values = (var_37651_cast_fp16, var_38050_cast_fp16))[name = string("op_38151_cast_fp16")]; + string var_38153_equation_0 = const()[name = string("op_38153_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38153_cast_fp16 = einsum(equation = var_38153_equation_0, values = (var_37651_cast_fp16, var_38051_cast_fp16))[name = string("op_38153_cast_fp16")]; + string var_38155_equation_0 = const()[name = string("op_38155_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38155_cast_fp16 = einsum(equation = var_38155_equation_0, values = (var_37651_cast_fp16, var_38052_cast_fp16))[name = string("op_38155_cast_fp16")]; + string var_38157_equation_0 = const()[name = string("op_38157_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38157_cast_fp16 = einsum(equation = var_38157_equation_0, values = (var_37651_cast_fp16, var_38053_cast_fp16))[name = string("op_38157_cast_fp16")]; + string var_38159_equation_0 = const()[name = string("op_38159_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38159_cast_fp16 = einsum(equation = var_38159_equation_0, values = (var_37655_cast_fp16, var_38054_cast_fp16))[name = string("op_38159_cast_fp16")]; + string var_38161_equation_0 = const()[name = string("op_38161_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38161_cast_fp16 = einsum(equation = var_38161_equation_0, values = (var_37655_cast_fp16, var_38055_cast_fp16))[name = string("op_38161_cast_fp16")]; + string var_38163_equation_0 = const()[name = string("op_38163_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38163_cast_fp16 = einsum(equation = var_38163_equation_0, values = (var_37655_cast_fp16, var_38056_cast_fp16))[name = string("op_38163_cast_fp16")]; + string var_38165_equation_0 = const()[name = string("op_38165_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38165_cast_fp16 = einsum(equation = var_38165_equation_0, values = (var_37655_cast_fp16, var_38057_cast_fp16))[name = string("op_38165_cast_fp16")]; + string var_38167_equation_0 = const()[name = string("op_38167_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38167_cast_fp16 = einsum(equation = var_38167_equation_0, values = (var_37659_cast_fp16, var_38058_cast_fp16))[name = string("op_38167_cast_fp16")]; + string var_38169_equation_0 = const()[name = string("op_38169_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38169_cast_fp16 = einsum(equation = var_38169_equation_0, values = (var_37659_cast_fp16, var_38059_cast_fp16))[name = string("op_38169_cast_fp16")]; + string var_38171_equation_0 = const()[name = string("op_38171_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38171_cast_fp16 = einsum(equation = var_38171_equation_0, values = (var_37659_cast_fp16, var_38060_cast_fp16))[name = string("op_38171_cast_fp16")]; + string var_38173_equation_0 = const()[name = string("op_38173_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38173_cast_fp16 = einsum(equation = var_38173_equation_0, values = (var_37659_cast_fp16, var_38061_cast_fp16))[name = string("op_38173_cast_fp16")]; + string var_38175_equation_0 = const()[name = string("op_38175_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38175_cast_fp16 = einsum(equation = var_38175_equation_0, values = (var_37663_cast_fp16, var_38062_cast_fp16))[name = string("op_38175_cast_fp16")]; + string var_38177_equation_0 = const()[name = string("op_38177_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38177_cast_fp16 = einsum(equation = var_38177_equation_0, values = (var_37663_cast_fp16, var_38063_cast_fp16))[name = string("op_38177_cast_fp16")]; + string var_38179_equation_0 = const()[name = string("op_38179_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38179_cast_fp16 = einsum(equation = var_38179_equation_0, values = (var_37663_cast_fp16, var_38064_cast_fp16))[name = string("op_38179_cast_fp16")]; + string var_38181_equation_0 = const()[name = string("op_38181_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38181_cast_fp16 = einsum(equation = var_38181_equation_0, values = (var_37663_cast_fp16, var_38065_cast_fp16))[name = string("op_38181_cast_fp16")]; + string var_38183_equation_0 = const()[name = string("op_38183_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38183_cast_fp16 = einsum(equation = var_38183_equation_0, values = (var_37667_cast_fp16, var_38066_cast_fp16))[name = string("op_38183_cast_fp16")]; + string var_38185_equation_0 = const()[name = string("op_38185_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38185_cast_fp16 = einsum(equation = var_38185_equation_0, values = (var_37667_cast_fp16, var_38067_cast_fp16))[name = string("op_38185_cast_fp16")]; + string var_38187_equation_0 = const()[name = string("op_38187_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38187_cast_fp16 = einsum(equation = var_38187_equation_0, values = (var_37667_cast_fp16, var_38068_cast_fp16))[name = string("op_38187_cast_fp16")]; + string var_38189_equation_0 = const()[name = string("op_38189_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38189_cast_fp16 = einsum(equation = var_38189_equation_0, values = (var_37667_cast_fp16, var_38069_cast_fp16))[name = string("op_38189_cast_fp16")]; + string var_38191_equation_0 = const()[name = string("op_38191_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38191_cast_fp16 = einsum(equation = var_38191_equation_0, values = (var_37671_cast_fp16, var_38070_cast_fp16))[name = string("op_38191_cast_fp16")]; + string var_38193_equation_0 = const()[name = string("op_38193_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38193_cast_fp16 = einsum(equation = var_38193_equation_0, values = (var_37671_cast_fp16, var_38071_cast_fp16))[name = string("op_38193_cast_fp16")]; + string var_38195_equation_0 = const()[name = string("op_38195_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38195_cast_fp16 = einsum(equation = var_38195_equation_0, values = (var_37671_cast_fp16, var_38072_cast_fp16))[name = string("op_38195_cast_fp16")]; + string var_38197_equation_0 = const()[name = string("op_38197_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38197_cast_fp16 = einsum(equation = var_38197_equation_0, values = (var_37671_cast_fp16, var_38073_cast_fp16))[name = string("op_38197_cast_fp16")]; + string var_38199_equation_0 = const()[name = string("op_38199_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38199_cast_fp16 = einsum(equation = var_38199_equation_0, values = (var_37675_cast_fp16, var_38074_cast_fp16))[name = string("op_38199_cast_fp16")]; + string var_38201_equation_0 = const()[name = string("op_38201_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38201_cast_fp16 = einsum(equation = var_38201_equation_0, values = (var_37675_cast_fp16, var_38075_cast_fp16))[name = string("op_38201_cast_fp16")]; + string var_38203_equation_0 = const()[name = string("op_38203_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38203_cast_fp16 = einsum(equation = var_38203_equation_0, values = (var_37675_cast_fp16, var_38076_cast_fp16))[name = string("op_38203_cast_fp16")]; + string var_38205_equation_0 = const()[name = string("op_38205_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38205_cast_fp16 = einsum(equation = var_38205_equation_0, values = (var_37675_cast_fp16, var_38077_cast_fp16))[name = string("op_38205_cast_fp16")]; + string var_38207_equation_0 = const()[name = string("op_38207_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38207_cast_fp16 = einsum(equation = var_38207_equation_0, values = (var_37679_cast_fp16, var_38078_cast_fp16))[name = string("op_38207_cast_fp16")]; + string var_38209_equation_0 = const()[name = string("op_38209_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38209_cast_fp16 = einsum(equation = var_38209_equation_0, values = (var_37679_cast_fp16, var_38079_cast_fp16))[name = string("op_38209_cast_fp16")]; + string var_38211_equation_0 = const()[name = string("op_38211_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38211_cast_fp16 = einsum(equation = var_38211_equation_0, values = (var_37679_cast_fp16, var_38080_cast_fp16))[name = string("op_38211_cast_fp16")]; + string var_38213_equation_0 = const()[name = string("op_38213_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38213_cast_fp16 = einsum(equation = var_38213_equation_0, values = (var_37679_cast_fp16, var_38081_cast_fp16))[name = string("op_38213_cast_fp16")]; + string var_38215_equation_0 = const()[name = string("op_38215_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38215_cast_fp16 = einsum(equation = var_38215_equation_0, values = (var_37683_cast_fp16, var_38082_cast_fp16))[name = string("op_38215_cast_fp16")]; + string var_38217_equation_0 = const()[name = string("op_38217_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38217_cast_fp16 = einsum(equation = var_38217_equation_0, values = (var_37683_cast_fp16, var_38083_cast_fp16))[name = string("op_38217_cast_fp16")]; + string var_38219_equation_0 = const()[name = string("op_38219_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38219_cast_fp16 = einsum(equation = var_38219_equation_0, values = (var_37683_cast_fp16, var_38084_cast_fp16))[name = string("op_38219_cast_fp16")]; + string var_38221_equation_0 = const()[name = string("op_38221_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38221_cast_fp16 = einsum(equation = var_38221_equation_0, values = (var_37683_cast_fp16, var_38085_cast_fp16))[name = string("op_38221_cast_fp16")]; + string var_38223_equation_0 = const()[name = string("op_38223_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38223_cast_fp16 = einsum(equation = var_38223_equation_0, values = (var_37687_cast_fp16, var_38086_cast_fp16))[name = string("op_38223_cast_fp16")]; + string var_38225_equation_0 = const()[name = string("op_38225_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38225_cast_fp16 = einsum(equation = var_38225_equation_0, values = (var_37687_cast_fp16, var_38087_cast_fp16))[name = string("op_38225_cast_fp16")]; + string var_38227_equation_0 = const()[name = string("op_38227_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38227_cast_fp16 = einsum(equation = var_38227_equation_0, values = (var_37687_cast_fp16, var_38088_cast_fp16))[name = string("op_38227_cast_fp16")]; + string var_38229_equation_0 = const()[name = string("op_38229_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38229_cast_fp16 = einsum(equation = var_38229_equation_0, values = (var_37687_cast_fp16, var_38089_cast_fp16))[name = string("op_38229_cast_fp16")]; + string var_38231_equation_0 = const()[name = string("op_38231_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38231_cast_fp16 = einsum(equation = var_38231_equation_0, values = (var_37691_cast_fp16, var_38090_cast_fp16))[name = string("op_38231_cast_fp16")]; + string var_38233_equation_0 = const()[name = string("op_38233_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38233_cast_fp16 = einsum(equation = var_38233_equation_0, values = (var_37691_cast_fp16, var_38091_cast_fp16))[name = string("op_38233_cast_fp16")]; + string var_38235_equation_0 = const()[name = string("op_38235_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38235_cast_fp16 = einsum(equation = var_38235_equation_0, values = (var_37691_cast_fp16, var_38092_cast_fp16))[name = string("op_38235_cast_fp16")]; + string var_38237_equation_0 = const()[name = string("op_38237_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38237_cast_fp16 = einsum(equation = var_38237_equation_0, values = (var_37691_cast_fp16, var_38093_cast_fp16))[name = string("op_38237_cast_fp16")]; + string var_38239_equation_0 = const()[name = string("op_38239_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38239_cast_fp16 = einsum(equation = var_38239_equation_0, values = (var_37695_cast_fp16, var_38094_cast_fp16))[name = string("op_38239_cast_fp16")]; + string var_38241_equation_0 = const()[name = string("op_38241_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38241_cast_fp16 = einsum(equation = var_38241_equation_0, values = (var_37695_cast_fp16, var_38095_cast_fp16))[name = string("op_38241_cast_fp16")]; + string var_38243_equation_0 = const()[name = string("op_38243_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38243_cast_fp16 = einsum(equation = var_38243_equation_0, values = (var_37695_cast_fp16, var_38096_cast_fp16))[name = string("op_38243_cast_fp16")]; + string var_38245_equation_0 = const()[name = string("op_38245_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38245_cast_fp16 = einsum(equation = var_38245_equation_0, values = (var_37695_cast_fp16, var_38097_cast_fp16))[name = string("op_38245_cast_fp16")]; + string var_38247_equation_0 = const()[name = string("op_38247_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38247_cast_fp16 = einsum(equation = var_38247_equation_0, values = (var_37699_cast_fp16, var_38098_cast_fp16))[name = string("op_38247_cast_fp16")]; + string var_38249_equation_0 = const()[name = string("op_38249_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38249_cast_fp16 = einsum(equation = var_38249_equation_0, values = (var_37699_cast_fp16, var_38099_cast_fp16))[name = string("op_38249_cast_fp16")]; + string var_38251_equation_0 = const()[name = string("op_38251_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38251_cast_fp16 = einsum(equation = var_38251_equation_0, values = (var_37699_cast_fp16, var_38100_cast_fp16))[name = string("op_38251_cast_fp16")]; + string var_38253_equation_0 = const()[name = string("op_38253_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38253_cast_fp16 = einsum(equation = var_38253_equation_0, values = (var_37699_cast_fp16, var_38101_cast_fp16))[name = string("op_38253_cast_fp16")]; + string var_38255_equation_0 = const()[name = string("op_38255_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38255_cast_fp16 = einsum(equation = var_38255_equation_0, values = (var_37703_cast_fp16, var_38102_cast_fp16))[name = string("op_38255_cast_fp16")]; + string var_38257_equation_0 = const()[name = string("op_38257_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38257_cast_fp16 = einsum(equation = var_38257_equation_0, values = (var_37703_cast_fp16, var_38103_cast_fp16))[name = string("op_38257_cast_fp16")]; + string var_38259_equation_0 = const()[name = string("op_38259_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38259_cast_fp16 = einsum(equation = var_38259_equation_0, values = (var_37703_cast_fp16, var_38104_cast_fp16))[name = string("op_38259_cast_fp16")]; + string var_38261_equation_0 = const()[name = string("op_38261_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38261_cast_fp16 = einsum(equation = var_38261_equation_0, values = (var_37703_cast_fp16, var_38105_cast_fp16))[name = string("op_38261_cast_fp16")]; + string var_38263_equation_0 = const()[name = string("op_38263_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38263_cast_fp16 = einsum(equation = var_38263_equation_0, values = (var_37707_cast_fp16, var_38106_cast_fp16))[name = string("op_38263_cast_fp16")]; + string var_38265_equation_0 = const()[name = string("op_38265_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38265_cast_fp16 = einsum(equation = var_38265_equation_0, values = (var_37707_cast_fp16, var_38107_cast_fp16))[name = string("op_38265_cast_fp16")]; + string var_38267_equation_0 = const()[name = string("op_38267_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38267_cast_fp16 = einsum(equation = var_38267_equation_0, values = (var_37707_cast_fp16, var_38108_cast_fp16))[name = string("op_38267_cast_fp16")]; + string var_38269_equation_0 = const()[name = string("op_38269_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_38269_cast_fp16 = einsum(equation = var_38269_equation_0, values = (var_37707_cast_fp16, var_38109_cast_fp16))[name = string("op_38269_cast_fp16")]; + bool var_38271_interleave_0 = const()[name = string("op_38271_interleave_0"), val = bool(false)]; + tensor var_38271_cast_fp16 = concat(axis = var_36830, interleave = var_38271_interleave_0, values = (var_38111_cast_fp16, var_38113_cast_fp16, var_38115_cast_fp16, var_38117_cast_fp16))[name = string("op_38271_cast_fp16")]; + bool var_38273_interleave_0 = const()[name = string("op_38273_interleave_0"), val = bool(false)]; + tensor var_38273_cast_fp16 = concat(axis = var_36830, interleave = var_38273_interleave_0, values = (var_38119_cast_fp16, var_38121_cast_fp16, var_38123_cast_fp16, var_38125_cast_fp16))[name = string("op_38273_cast_fp16")]; + bool var_38275_interleave_0 = const()[name = string("op_38275_interleave_0"), val = bool(false)]; + tensor var_38275_cast_fp16 = concat(axis = var_36830, interleave = var_38275_interleave_0, values = (var_38127_cast_fp16, var_38129_cast_fp16, var_38131_cast_fp16, var_38133_cast_fp16))[name = string("op_38275_cast_fp16")]; + bool var_38277_interleave_0 = const()[name = string("op_38277_interleave_0"), val = bool(false)]; + tensor var_38277_cast_fp16 = concat(axis = var_36830, interleave = var_38277_interleave_0, values = (var_38135_cast_fp16, var_38137_cast_fp16, var_38139_cast_fp16, var_38141_cast_fp16))[name = string("op_38277_cast_fp16")]; + bool var_38279_interleave_0 = const()[name = string("op_38279_interleave_0"), val = bool(false)]; + tensor var_38279_cast_fp16 = concat(axis = var_36830, interleave = var_38279_interleave_0, values = (var_38143_cast_fp16, var_38145_cast_fp16, var_38147_cast_fp16, var_38149_cast_fp16))[name = string("op_38279_cast_fp16")]; + bool var_38281_interleave_0 = const()[name = string("op_38281_interleave_0"), val = bool(false)]; + tensor var_38281_cast_fp16 = concat(axis = var_36830, interleave = var_38281_interleave_0, values = (var_38151_cast_fp16, var_38153_cast_fp16, var_38155_cast_fp16, var_38157_cast_fp16))[name = string("op_38281_cast_fp16")]; + bool var_38283_interleave_0 = const()[name = string("op_38283_interleave_0"), val = bool(false)]; + tensor var_38283_cast_fp16 = concat(axis = var_36830, interleave = var_38283_interleave_0, values = (var_38159_cast_fp16, var_38161_cast_fp16, var_38163_cast_fp16, var_38165_cast_fp16))[name = string("op_38283_cast_fp16")]; + bool var_38285_interleave_0 = const()[name = string("op_38285_interleave_0"), val = bool(false)]; + tensor var_38285_cast_fp16 = concat(axis = var_36830, interleave = var_38285_interleave_0, values = (var_38167_cast_fp16, var_38169_cast_fp16, var_38171_cast_fp16, var_38173_cast_fp16))[name = string("op_38285_cast_fp16")]; + bool var_38287_interleave_0 = const()[name = string("op_38287_interleave_0"), val = bool(false)]; + tensor var_38287_cast_fp16 = concat(axis = var_36830, interleave = var_38287_interleave_0, values = (var_38175_cast_fp16, var_38177_cast_fp16, var_38179_cast_fp16, var_38181_cast_fp16))[name = string("op_38287_cast_fp16")]; + bool var_38289_interleave_0 = const()[name = string("op_38289_interleave_0"), val = bool(false)]; + tensor var_38289_cast_fp16 = concat(axis = var_36830, interleave = var_38289_interleave_0, values = (var_38183_cast_fp16, var_38185_cast_fp16, var_38187_cast_fp16, var_38189_cast_fp16))[name = string("op_38289_cast_fp16")]; + bool var_38291_interleave_0 = const()[name = string("op_38291_interleave_0"), val = bool(false)]; + tensor var_38291_cast_fp16 = concat(axis = var_36830, interleave = var_38291_interleave_0, values = (var_38191_cast_fp16, var_38193_cast_fp16, var_38195_cast_fp16, var_38197_cast_fp16))[name = string("op_38291_cast_fp16")]; + bool var_38293_interleave_0 = const()[name = string("op_38293_interleave_0"), val = bool(false)]; + tensor var_38293_cast_fp16 = concat(axis = var_36830, interleave = var_38293_interleave_0, values = (var_38199_cast_fp16, var_38201_cast_fp16, var_38203_cast_fp16, var_38205_cast_fp16))[name = string("op_38293_cast_fp16")]; + bool var_38295_interleave_0 = const()[name = string("op_38295_interleave_0"), val = bool(false)]; + tensor var_38295_cast_fp16 = concat(axis = var_36830, interleave = var_38295_interleave_0, values = (var_38207_cast_fp16, var_38209_cast_fp16, var_38211_cast_fp16, var_38213_cast_fp16))[name = string("op_38295_cast_fp16")]; + bool var_38297_interleave_0 = const()[name = string("op_38297_interleave_0"), val = bool(false)]; + tensor var_38297_cast_fp16 = concat(axis = var_36830, interleave = var_38297_interleave_0, values = (var_38215_cast_fp16, var_38217_cast_fp16, var_38219_cast_fp16, var_38221_cast_fp16))[name = string("op_38297_cast_fp16")]; + bool var_38299_interleave_0 = const()[name = string("op_38299_interleave_0"), val = bool(false)]; + tensor var_38299_cast_fp16 = concat(axis = var_36830, interleave = var_38299_interleave_0, values = (var_38223_cast_fp16, var_38225_cast_fp16, var_38227_cast_fp16, var_38229_cast_fp16))[name = string("op_38299_cast_fp16")]; + bool var_38301_interleave_0 = const()[name = string("op_38301_interleave_0"), val = bool(false)]; + tensor var_38301_cast_fp16 = concat(axis = var_36830, interleave = var_38301_interleave_0, values = (var_38231_cast_fp16, var_38233_cast_fp16, var_38235_cast_fp16, var_38237_cast_fp16))[name = string("op_38301_cast_fp16")]; + bool var_38303_interleave_0 = const()[name = string("op_38303_interleave_0"), val = bool(false)]; + tensor var_38303_cast_fp16 = concat(axis = var_36830, interleave = var_38303_interleave_0, values = (var_38239_cast_fp16, var_38241_cast_fp16, var_38243_cast_fp16, var_38245_cast_fp16))[name = string("op_38303_cast_fp16")]; + bool var_38305_interleave_0 = const()[name = string("op_38305_interleave_0"), val = bool(false)]; + tensor var_38305_cast_fp16 = concat(axis = var_36830, interleave = var_38305_interleave_0, values = (var_38247_cast_fp16, var_38249_cast_fp16, var_38251_cast_fp16, var_38253_cast_fp16))[name = string("op_38305_cast_fp16")]; + bool var_38307_interleave_0 = const()[name = string("op_38307_interleave_0"), val = bool(false)]; + tensor var_38307_cast_fp16 = concat(axis = var_36830, interleave = var_38307_interleave_0, values = (var_38255_cast_fp16, var_38257_cast_fp16, var_38259_cast_fp16, var_38261_cast_fp16))[name = string("op_38307_cast_fp16")]; + bool var_38309_interleave_0 = const()[name = string("op_38309_interleave_0"), val = bool(false)]; + tensor var_38309_cast_fp16 = concat(axis = var_36830, interleave = var_38309_interleave_0, values = (var_38263_cast_fp16, var_38265_cast_fp16, var_38267_cast_fp16, var_38269_cast_fp16))[name = string("op_38309_cast_fp16")]; + bool input_193_interleave_0 = const()[name = string("input_193_interleave_0"), val = bool(false)]; + tensor input_193_cast_fp16 = concat(axis = var_36855, interleave = input_193_interleave_0, values = (var_38271_cast_fp16, var_38273_cast_fp16, var_38275_cast_fp16, var_38277_cast_fp16, var_38279_cast_fp16, var_38281_cast_fp16, var_38283_cast_fp16, var_38285_cast_fp16, var_38287_cast_fp16, var_38289_cast_fp16, var_38291_cast_fp16, var_38293_cast_fp16, var_38295_cast_fp16, var_38297_cast_fp16, var_38299_cast_fp16, var_38301_cast_fp16, var_38303_cast_fp16, var_38305_cast_fp16, var_38307_cast_fp16, var_38309_cast_fp16))[name = string("input_193_cast_fp16")]; + string obj_99_pad_type_0 = const()[name = string("obj_99_pad_type_0"), val = string("valid")]; + tensor obj_99_strides_0 = const()[name = string("obj_99_strides_0"), val = tensor([1, 1])]; + tensor obj_99_pad_0 = const()[name = string("obj_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_99_dilations_0 = const()[name = string("obj_99_dilations_0"), val = tensor([1, 1])]; + int32 obj_99_groups_0 = const()[name = string("obj_99_groups_0"), val = int32(1)]; + tensor layers_24_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(968984000)))]; + tensor layers_24_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972260864)))]; + tensor obj_99_cast_fp16 = conv(bias = layers_24_self_attn_o_proj_bias_to_fp16, dilations = obj_99_dilations_0, groups = obj_99_groups_0, pad = obj_99_pad_0, pad_type = obj_99_pad_type_0, strides = obj_99_strides_0, weight = layers_24_self_attn_o_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("obj_99_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_99_cast_fp16)[name = string("inputs_99_cast_fp16")]; + tensor out_99_axes_0 = const()[name = string("out_99_axes_0"), val = tensor([1])]; + fp16 var_38328_to_fp16 = const()[name = string("op_38328_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_38328_to_fp16, x = inputs_99_cast_fp16)[name = string("out_99_cast_fp16")]; + tensor input_195_gamma_0_to_fp16 = const()[name = string("input_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972263488)))]; + tensor input_195_beta_0_to_fp16 = const()[name = string("input_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972266112)))]; + fp16 input_195_epsilon_0_to_fp16 = const()[name = string("input_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = string("input_195_cast_fp16")]; + string input_197_pad_type_0 = const()[name = string("input_197_pad_type_0"), val = string("valid")]; + tensor input_197_strides_0 = const()[name = string("input_197_strides_0"), val = tensor([1, 1])]; + tensor input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_197_dilations_0 = const()[name = string("input_197_dilations_0"), val = tensor([1, 1])]; + int32 input_197_groups_0 = const()[name = string("input_197_groups_0"), val = int32(1)]; + tensor layers_24_fc1_weight_to_fp16 = const()[name = string("layers_24_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(972268736)))]; + tensor layers_24_fc1_bias_to_fp16 = const()[name = string("layers_24_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985376000)))]; + tensor input_197_cast_fp16 = conv(bias = layers_24_fc1_bias_to_fp16, dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_24_fc1_weight_to_fp16, x = input_195_cast_fp16)[name = string("input_197_cast_fp16")]; + string input_199_mode_0 = const()[name = string("input_199_mode_0"), val = string("EXACT")]; + tensor input_199_cast_fp16 = gelu(mode = input_199_mode_0, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")]; + string hidden_states_53_pad_type_0 = const()[name = string("hidden_states_53_pad_type_0"), val = string("valid")]; + tensor hidden_states_53_strides_0 = const()[name = string("hidden_states_53_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_53_pad_0 = const()[name = string("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_53_dilations_0 = const()[name = string("hidden_states_53_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_53_groups_0 = const()[name = string("hidden_states_53_groups_0"), val = int32(1)]; + tensor layers_24_fc2_weight_to_fp16 = const()[name = string("layers_24_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(985386304)))]; + tensor layers_24_fc2_bias_to_fp16 = const()[name = string("layers_24_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998493568)))]; + tensor hidden_states_53_cast_fp16 = conv(bias = layers_24_fc2_bias_to_fp16, dilations = hidden_states_53_dilations_0, groups = hidden_states_53_groups_0, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = hidden_states_53_strides_0, weight = layers_24_fc2_weight_to_fp16, x = input_199_cast_fp16)[name = string("hidden_states_53_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = hidden_states_53_cast_fp16)[name = string("inputs_101_cast_fp16")]; + int32 var_38357 = const()[name = string("op_38357"), val = int32(3)]; + int32 var_38382 = const()[name = string("op_38382"), val = int32(1)]; + tensor out_101_axes_0 = const()[name = string("out_101_axes_0"), val = tensor([1])]; + fp16 var_38399_to_fp16 = const()[name = string("op_38399_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_38399_to_fp16, x = inputs_101_cast_fp16)[name = string("out_101_cast_fp16")]; + tensor obj_101_gamma_0_to_fp16 = const()[name = string("obj_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998496192)))]; + tensor obj_101_beta_0_to_fp16 = const()[name = string("obj_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998498816)))]; + fp16 obj_101_epsilon_0_to_fp16 = const()[name = string("obj_101_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_101_cast_fp16 = batch_norm(beta = obj_101_beta_0_to_fp16, epsilon = obj_101_epsilon_0_to_fp16, gamma = obj_101_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = string("obj_101_cast_fp16")]; + string query_51_pad_type_0 = const()[name = string("query_51_pad_type_0"), val = string("valid")]; + tensor query_51_strides_0 = const()[name = string("query_51_strides_0"), val = tensor([1, 1])]; + tensor query_51_pad_0 = const()[name = string("query_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_51_dilations_0 = const()[name = string("query_51_dilations_0"), val = tensor([1, 1])]; + int32 query_51_groups_0 = const()[name = string("query_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(998501440)))]; + tensor layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001778304)))]; + tensor query_51_cast_fp16 = conv(bias = layers_25_self_attn_q_proj_bias_to_fp16, dilations = query_51_dilations_0, groups = query_51_groups_0, pad = query_51_pad_0, pad_type = query_51_pad_type_0, strides = query_51_strides_0, weight = layers_25_self_attn_q_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("query_51_cast_fp16")]; + string key_51_pad_type_0 = const()[name = string("key_51_pad_type_0"), val = string("valid")]; + tensor key_51_strides_0 = const()[name = string("key_51_strides_0"), val = tensor([1, 1])]; + tensor key_51_pad_0 = const()[name = string("key_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_51_dilations_0 = const()[name = string("key_51_dilations_0"), val = tensor([1, 1])]; + int32 key_51_groups_0 = const()[name = string("key_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1001780928)))]; + tensor key_51_cast_fp16 = conv(dilations = key_51_dilations_0, groups = key_51_groups_0, pad = key_51_pad_0, pad_type = key_51_pad_type_0, strides = key_51_strides_0, weight = layers_25_self_attn_k_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("key_51_cast_fp16")]; + string value_51_pad_type_0 = const()[name = string("value_51_pad_type_0"), val = string("valid")]; + tensor value_51_strides_0 = const()[name = string("value_51_strides_0"), val = tensor([1, 1])]; + tensor value_51_pad_0 = const()[name = string("value_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_51_dilations_0 = const()[name = string("value_51_dilations_0"), val = tensor([1, 1])]; + int32 value_51_groups_0 = const()[name = string("value_51_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1005057792)))]; + tensor layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008334656)))]; + tensor value_51_cast_fp16 = conv(bias = layers_25_self_attn_v_proj_bias_to_fp16, dilations = value_51_dilations_0, groups = value_51_groups_0, pad = value_51_pad_0, pad_type = value_51_pad_type_0, strides = value_51_strides_0, weight = layers_25_self_attn_v_proj_weight_to_fp16, x = obj_101_cast_fp16)[name = string("value_51_cast_fp16")]; + tensor var_38437_begin_0 = const()[name = string("op_38437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38437_end_0 = const()[name = string("op_38437_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38437_end_mask_0 = const()[name = string("op_38437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38437_cast_fp16 = slice_by_index(begin = var_38437_begin_0, end = var_38437_end_0, end_mask = var_38437_end_mask_0, x = query_51_cast_fp16)[name = string("op_38437_cast_fp16")]; + tensor var_38441_begin_0 = const()[name = string("op_38441_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38441_end_0 = const()[name = string("op_38441_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_38441_end_mask_0 = const()[name = string("op_38441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38441_cast_fp16 = slice_by_index(begin = var_38441_begin_0, end = var_38441_end_0, end_mask = var_38441_end_mask_0, x = query_51_cast_fp16)[name = string("op_38441_cast_fp16")]; + tensor var_38445_begin_0 = const()[name = string("op_38445_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38445_end_0 = const()[name = string("op_38445_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_38445_end_mask_0 = const()[name = string("op_38445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38445_cast_fp16 = slice_by_index(begin = var_38445_begin_0, end = var_38445_end_0, end_mask = var_38445_end_mask_0, x = query_51_cast_fp16)[name = string("op_38445_cast_fp16")]; + tensor var_38449_begin_0 = const()[name = string("op_38449_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38449_end_0 = const()[name = string("op_38449_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_38449_end_mask_0 = const()[name = string("op_38449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38449_cast_fp16 = slice_by_index(begin = var_38449_begin_0, end = var_38449_end_0, end_mask = var_38449_end_mask_0, x = query_51_cast_fp16)[name = string("op_38449_cast_fp16")]; + tensor var_38453_begin_0 = const()[name = string("op_38453_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38453_end_0 = const()[name = string("op_38453_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_38453_end_mask_0 = const()[name = string("op_38453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38453_cast_fp16 = slice_by_index(begin = var_38453_begin_0, end = var_38453_end_0, end_mask = var_38453_end_mask_0, x = query_51_cast_fp16)[name = string("op_38453_cast_fp16")]; + tensor var_38457_begin_0 = const()[name = string("op_38457_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38457_end_0 = const()[name = string("op_38457_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_38457_end_mask_0 = const()[name = string("op_38457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38457_cast_fp16 = slice_by_index(begin = var_38457_begin_0, end = var_38457_end_0, end_mask = var_38457_end_mask_0, x = query_51_cast_fp16)[name = string("op_38457_cast_fp16")]; + tensor var_38461_begin_0 = const()[name = string("op_38461_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38461_end_0 = const()[name = string("op_38461_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_38461_end_mask_0 = const()[name = string("op_38461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38461_cast_fp16 = slice_by_index(begin = var_38461_begin_0, end = var_38461_end_0, end_mask = var_38461_end_mask_0, x = query_51_cast_fp16)[name = string("op_38461_cast_fp16")]; + tensor var_38465_begin_0 = const()[name = string("op_38465_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38465_end_0 = const()[name = string("op_38465_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_38465_end_mask_0 = const()[name = string("op_38465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38465_cast_fp16 = slice_by_index(begin = var_38465_begin_0, end = var_38465_end_0, end_mask = var_38465_end_mask_0, x = query_51_cast_fp16)[name = string("op_38465_cast_fp16")]; + tensor var_38469_begin_0 = const()[name = string("op_38469_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38469_end_0 = const()[name = string("op_38469_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_38469_end_mask_0 = const()[name = string("op_38469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38469_cast_fp16 = slice_by_index(begin = var_38469_begin_0, end = var_38469_end_0, end_mask = var_38469_end_mask_0, x = query_51_cast_fp16)[name = string("op_38469_cast_fp16")]; + tensor var_38473_begin_0 = const()[name = string("op_38473_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38473_end_0 = const()[name = string("op_38473_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_38473_end_mask_0 = const()[name = string("op_38473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38473_cast_fp16 = slice_by_index(begin = var_38473_begin_0, end = var_38473_end_0, end_mask = var_38473_end_mask_0, x = query_51_cast_fp16)[name = string("op_38473_cast_fp16")]; + tensor var_38477_begin_0 = const()[name = string("op_38477_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38477_end_0 = const()[name = string("op_38477_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_38477_end_mask_0 = const()[name = string("op_38477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38477_cast_fp16 = slice_by_index(begin = var_38477_begin_0, end = var_38477_end_0, end_mask = var_38477_end_mask_0, x = query_51_cast_fp16)[name = string("op_38477_cast_fp16")]; + tensor var_38481_begin_0 = const()[name = string("op_38481_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38481_end_0 = const()[name = string("op_38481_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_38481_end_mask_0 = const()[name = string("op_38481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38481_cast_fp16 = slice_by_index(begin = var_38481_begin_0, end = var_38481_end_0, end_mask = var_38481_end_mask_0, x = query_51_cast_fp16)[name = string("op_38481_cast_fp16")]; + tensor var_38485_begin_0 = const()[name = string("op_38485_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38485_end_0 = const()[name = string("op_38485_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_38485_end_mask_0 = const()[name = string("op_38485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38485_cast_fp16 = slice_by_index(begin = var_38485_begin_0, end = var_38485_end_0, end_mask = var_38485_end_mask_0, x = query_51_cast_fp16)[name = string("op_38485_cast_fp16")]; + tensor var_38489_begin_0 = const()[name = string("op_38489_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38489_end_0 = const()[name = string("op_38489_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_38489_end_mask_0 = const()[name = string("op_38489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38489_cast_fp16 = slice_by_index(begin = var_38489_begin_0, end = var_38489_end_0, end_mask = var_38489_end_mask_0, x = query_51_cast_fp16)[name = string("op_38489_cast_fp16")]; + tensor var_38493_begin_0 = const()[name = string("op_38493_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38493_end_0 = const()[name = string("op_38493_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_38493_end_mask_0 = const()[name = string("op_38493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38493_cast_fp16 = slice_by_index(begin = var_38493_begin_0, end = var_38493_end_0, end_mask = var_38493_end_mask_0, x = query_51_cast_fp16)[name = string("op_38493_cast_fp16")]; + tensor var_38497_begin_0 = const()[name = string("op_38497_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38497_end_0 = const()[name = string("op_38497_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_38497_end_mask_0 = const()[name = string("op_38497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38497_cast_fp16 = slice_by_index(begin = var_38497_begin_0, end = var_38497_end_0, end_mask = var_38497_end_mask_0, x = query_51_cast_fp16)[name = string("op_38497_cast_fp16")]; + tensor var_38501_begin_0 = const()[name = string("op_38501_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38501_end_0 = const()[name = string("op_38501_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_38501_end_mask_0 = const()[name = string("op_38501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38501_cast_fp16 = slice_by_index(begin = var_38501_begin_0, end = var_38501_end_0, end_mask = var_38501_end_mask_0, x = query_51_cast_fp16)[name = string("op_38501_cast_fp16")]; + tensor var_38505_begin_0 = const()[name = string("op_38505_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38505_end_0 = const()[name = string("op_38505_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_38505_end_mask_0 = const()[name = string("op_38505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38505_cast_fp16 = slice_by_index(begin = var_38505_begin_0, end = var_38505_end_0, end_mask = var_38505_end_mask_0, x = query_51_cast_fp16)[name = string("op_38505_cast_fp16")]; + tensor var_38509_begin_0 = const()[name = string("op_38509_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38509_end_0 = const()[name = string("op_38509_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_38509_end_mask_0 = const()[name = string("op_38509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38509_cast_fp16 = slice_by_index(begin = var_38509_begin_0, end = var_38509_end_0, end_mask = var_38509_end_mask_0, x = query_51_cast_fp16)[name = string("op_38509_cast_fp16")]; + tensor var_38513_begin_0 = const()[name = string("op_38513_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38513_end_0 = const()[name = string("op_38513_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_38513_end_mask_0 = const()[name = string("op_38513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38513_cast_fp16 = slice_by_index(begin = var_38513_begin_0, end = var_38513_end_0, end_mask = var_38513_end_mask_0, x = query_51_cast_fp16)[name = string("op_38513_cast_fp16")]; + tensor var_38522_begin_0 = const()[name = string("op_38522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38522_end_0 = const()[name = string("op_38522_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38522_end_mask_0 = const()[name = string("op_38522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38522_cast_fp16 = slice_by_index(begin = var_38522_begin_0, end = var_38522_end_0, end_mask = var_38522_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38522_cast_fp16")]; + tensor var_38529_begin_0 = const()[name = string("op_38529_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38529_end_0 = const()[name = string("op_38529_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38529_end_mask_0 = const()[name = string("op_38529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38529_cast_fp16 = slice_by_index(begin = var_38529_begin_0, end = var_38529_end_0, end_mask = var_38529_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38529_cast_fp16")]; + tensor var_38536_begin_0 = const()[name = string("op_38536_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38536_end_0 = const()[name = string("op_38536_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38536_end_mask_0 = const()[name = string("op_38536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38536_cast_fp16 = slice_by_index(begin = var_38536_begin_0, end = var_38536_end_0, end_mask = var_38536_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38536_cast_fp16")]; + tensor var_38543_begin_0 = const()[name = string("op_38543_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38543_end_0 = const()[name = string("op_38543_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38543_end_mask_0 = const()[name = string("op_38543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38543_cast_fp16 = slice_by_index(begin = var_38543_begin_0, end = var_38543_end_0, end_mask = var_38543_end_mask_0, x = var_38437_cast_fp16)[name = string("op_38543_cast_fp16")]; + tensor var_38550_begin_0 = const()[name = string("op_38550_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38550_end_0 = const()[name = string("op_38550_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38550_end_mask_0 = const()[name = string("op_38550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38550_cast_fp16 = slice_by_index(begin = var_38550_begin_0, end = var_38550_end_0, end_mask = var_38550_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38550_cast_fp16")]; + tensor var_38557_begin_0 = const()[name = string("op_38557_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38557_end_0 = const()[name = string("op_38557_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38557_end_mask_0 = const()[name = string("op_38557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38557_cast_fp16 = slice_by_index(begin = var_38557_begin_0, end = var_38557_end_0, end_mask = var_38557_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38557_cast_fp16")]; + tensor var_38564_begin_0 = const()[name = string("op_38564_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38564_end_0 = const()[name = string("op_38564_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38564_end_mask_0 = const()[name = string("op_38564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38564_cast_fp16 = slice_by_index(begin = var_38564_begin_0, end = var_38564_end_0, end_mask = var_38564_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38564_cast_fp16")]; + tensor var_38571_begin_0 = const()[name = string("op_38571_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38571_end_0 = const()[name = string("op_38571_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38571_end_mask_0 = const()[name = string("op_38571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38571_cast_fp16 = slice_by_index(begin = var_38571_begin_0, end = var_38571_end_0, end_mask = var_38571_end_mask_0, x = var_38441_cast_fp16)[name = string("op_38571_cast_fp16")]; + tensor var_38578_begin_0 = const()[name = string("op_38578_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38578_end_0 = const()[name = string("op_38578_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38578_end_mask_0 = const()[name = string("op_38578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38578_cast_fp16 = slice_by_index(begin = var_38578_begin_0, end = var_38578_end_0, end_mask = var_38578_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38578_cast_fp16")]; + tensor var_38585_begin_0 = const()[name = string("op_38585_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38585_end_0 = const()[name = string("op_38585_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38585_end_mask_0 = const()[name = string("op_38585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38585_cast_fp16 = slice_by_index(begin = var_38585_begin_0, end = var_38585_end_0, end_mask = var_38585_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38585_cast_fp16")]; + tensor var_38592_begin_0 = const()[name = string("op_38592_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38592_end_0 = const()[name = string("op_38592_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38592_end_mask_0 = const()[name = string("op_38592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38592_cast_fp16 = slice_by_index(begin = var_38592_begin_0, end = var_38592_end_0, end_mask = var_38592_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38592_cast_fp16")]; + tensor var_38599_begin_0 = const()[name = string("op_38599_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38599_end_0 = const()[name = string("op_38599_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38599_end_mask_0 = const()[name = string("op_38599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38599_cast_fp16 = slice_by_index(begin = var_38599_begin_0, end = var_38599_end_0, end_mask = var_38599_end_mask_0, x = var_38445_cast_fp16)[name = string("op_38599_cast_fp16")]; + tensor var_38606_begin_0 = const()[name = string("op_38606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38606_end_0 = const()[name = string("op_38606_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38606_end_mask_0 = const()[name = string("op_38606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38606_cast_fp16 = slice_by_index(begin = var_38606_begin_0, end = var_38606_end_0, end_mask = var_38606_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38606_cast_fp16")]; + tensor var_38613_begin_0 = const()[name = string("op_38613_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38613_end_0 = const()[name = string("op_38613_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38613_end_mask_0 = const()[name = string("op_38613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38613_cast_fp16 = slice_by_index(begin = var_38613_begin_0, end = var_38613_end_0, end_mask = var_38613_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38613_cast_fp16")]; + tensor var_38620_begin_0 = const()[name = string("op_38620_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38620_end_0 = const()[name = string("op_38620_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38620_end_mask_0 = const()[name = string("op_38620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38620_cast_fp16 = slice_by_index(begin = var_38620_begin_0, end = var_38620_end_0, end_mask = var_38620_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38620_cast_fp16")]; + tensor var_38627_begin_0 = const()[name = string("op_38627_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38627_end_0 = const()[name = string("op_38627_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38627_end_mask_0 = const()[name = string("op_38627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38627_cast_fp16 = slice_by_index(begin = var_38627_begin_0, end = var_38627_end_0, end_mask = var_38627_end_mask_0, x = var_38449_cast_fp16)[name = string("op_38627_cast_fp16")]; + tensor var_38634_begin_0 = const()[name = string("op_38634_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38634_end_0 = const()[name = string("op_38634_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38634_end_mask_0 = const()[name = string("op_38634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38634_cast_fp16 = slice_by_index(begin = var_38634_begin_0, end = var_38634_end_0, end_mask = var_38634_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38634_cast_fp16")]; + tensor var_38641_begin_0 = const()[name = string("op_38641_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38641_end_0 = const()[name = string("op_38641_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38641_end_mask_0 = const()[name = string("op_38641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38641_cast_fp16 = slice_by_index(begin = var_38641_begin_0, end = var_38641_end_0, end_mask = var_38641_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38641_cast_fp16")]; + tensor var_38648_begin_0 = const()[name = string("op_38648_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38648_end_0 = const()[name = string("op_38648_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38648_end_mask_0 = const()[name = string("op_38648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38648_cast_fp16 = slice_by_index(begin = var_38648_begin_0, end = var_38648_end_0, end_mask = var_38648_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38648_cast_fp16")]; + tensor var_38655_begin_0 = const()[name = string("op_38655_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38655_end_0 = const()[name = string("op_38655_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38655_end_mask_0 = const()[name = string("op_38655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38655_cast_fp16 = slice_by_index(begin = var_38655_begin_0, end = var_38655_end_0, end_mask = var_38655_end_mask_0, x = var_38453_cast_fp16)[name = string("op_38655_cast_fp16")]; + tensor var_38662_begin_0 = const()[name = string("op_38662_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38662_end_0 = const()[name = string("op_38662_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38662_end_mask_0 = const()[name = string("op_38662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38662_cast_fp16 = slice_by_index(begin = var_38662_begin_0, end = var_38662_end_0, end_mask = var_38662_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38662_cast_fp16")]; + tensor var_38669_begin_0 = const()[name = string("op_38669_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38669_end_0 = const()[name = string("op_38669_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38669_end_mask_0 = const()[name = string("op_38669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38669_cast_fp16 = slice_by_index(begin = var_38669_begin_0, end = var_38669_end_0, end_mask = var_38669_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38669_cast_fp16")]; + tensor var_38676_begin_0 = const()[name = string("op_38676_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38676_end_0 = const()[name = string("op_38676_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38676_end_mask_0 = const()[name = string("op_38676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38676_cast_fp16 = slice_by_index(begin = var_38676_begin_0, end = var_38676_end_0, end_mask = var_38676_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38676_cast_fp16")]; + tensor var_38683_begin_0 = const()[name = string("op_38683_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38683_end_0 = const()[name = string("op_38683_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38683_end_mask_0 = const()[name = string("op_38683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38683_cast_fp16 = slice_by_index(begin = var_38683_begin_0, end = var_38683_end_0, end_mask = var_38683_end_mask_0, x = var_38457_cast_fp16)[name = string("op_38683_cast_fp16")]; + tensor var_38690_begin_0 = const()[name = string("op_38690_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38690_end_0 = const()[name = string("op_38690_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38690_end_mask_0 = const()[name = string("op_38690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38690_cast_fp16 = slice_by_index(begin = var_38690_begin_0, end = var_38690_end_0, end_mask = var_38690_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38690_cast_fp16")]; + tensor var_38697_begin_0 = const()[name = string("op_38697_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38697_end_0 = const()[name = string("op_38697_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38697_end_mask_0 = const()[name = string("op_38697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38697_cast_fp16 = slice_by_index(begin = var_38697_begin_0, end = var_38697_end_0, end_mask = var_38697_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38697_cast_fp16")]; + tensor var_38704_begin_0 = const()[name = string("op_38704_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38704_end_0 = const()[name = string("op_38704_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38704_end_mask_0 = const()[name = string("op_38704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38704_cast_fp16 = slice_by_index(begin = var_38704_begin_0, end = var_38704_end_0, end_mask = var_38704_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38704_cast_fp16")]; + tensor var_38711_begin_0 = const()[name = string("op_38711_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38711_end_0 = const()[name = string("op_38711_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38711_end_mask_0 = const()[name = string("op_38711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38711_cast_fp16 = slice_by_index(begin = var_38711_begin_0, end = var_38711_end_0, end_mask = var_38711_end_mask_0, x = var_38461_cast_fp16)[name = string("op_38711_cast_fp16")]; + tensor var_38718_begin_0 = const()[name = string("op_38718_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38718_end_0 = const()[name = string("op_38718_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38718_end_mask_0 = const()[name = string("op_38718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38718_cast_fp16 = slice_by_index(begin = var_38718_begin_0, end = var_38718_end_0, end_mask = var_38718_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38718_cast_fp16")]; + tensor var_38725_begin_0 = const()[name = string("op_38725_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38725_end_0 = const()[name = string("op_38725_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38725_end_mask_0 = const()[name = string("op_38725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38725_cast_fp16 = slice_by_index(begin = var_38725_begin_0, end = var_38725_end_0, end_mask = var_38725_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38725_cast_fp16")]; + tensor var_38732_begin_0 = const()[name = string("op_38732_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38732_end_0 = const()[name = string("op_38732_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38732_end_mask_0 = const()[name = string("op_38732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38732_cast_fp16 = slice_by_index(begin = var_38732_begin_0, end = var_38732_end_0, end_mask = var_38732_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38732_cast_fp16")]; + tensor var_38739_begin_0 = const()[name = string("op_38739_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38739_end_0 = const()[name = string("op_38739_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38739_end_mask_0 = const()[name = string("op_38739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38739_cast_fp16 = slice_by_index(begin = var_38739_begin_0, end = var_38739_end_0, end_mask = var_38739_end_mask_0, x = var_38465_cast_fp16)[name = string("op_38739_cast_fp16")]; + tensor var_38746_begin_0 = const()[name = string("op_38746_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38746_end_0 = const()[name = string("op_38746_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38746_end_mask_0 = const()[name = string("op_38746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38746_cast_fp16 = slice_by_index(begin = var_38746_begin_0, end = var_38746_end_0, end_mask = var_38746_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38746_cast_fp16")]; + tensor var_38753_begin_0 = const()[name = string("op_38753_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38753_end_0 = const()[name = string("op_38753_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38753_end_mask_0 = const()[name = string("op_38753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38753_cast_fp16 = slice_by_index(begin = var_38753_begin_0, end = var_38753_end_0, end_mask = var_38753_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38753_cast_fp16")]; + tensor var_38760_begin_0 = const()[name = string("op_38760_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38760_end_0 = const()[name = string("op_38760_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38760_end_mask_0 = const()[name = string("op_38760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38760_cast_fp16 = slice_by_index(begin = var_38760_begin_0, end = var_38760_end_0, end_mask = var_38760_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38760_cast_fp16")]; + tensor var_38767_begin_0 = const()[name = string("op_38767_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38767_end_0 = const()[name = string("op_38767_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38767_end_mask_0 = const()[name = string("op_38767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38767_cast_fp16 = slice_by_index(begin = var_38767_begin_0, end = var_38767_end_0, end_mask = var_38767_end_mask_0, x = var_38469_cast_fp16)[name = string("op_38767_cast_fp16")]; + tensor var_38774_begin_0 = const()[name = string("op_38774_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38774_end_0 = const()[name = string("op_38774_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38774_end_mask_0 = const()[name = string("op_38774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38774_cast_fp16 = slice_by_index(begin = var_38774_begin_0, end = var_38774_end_0, end_mask = var_38774_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38774_cast_fp16")]; + tensor var_38781_begin_0 = const()[name = string("op_38781_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38781_end_0 = const()[name = string("op_38781_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38781_end_mask_0 = const()[name = string("op_38781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38781_cast_fp16 = slice_by_index(begin = var_38781_begin_0, end = var_38781_end_0, end_mask = var_38781_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38781_cast_fp16")]; + tensor var_38788_begin_0 = const()[name = string("op_38788_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38788_end_0 = const()[name = string("op_38788_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38788_end_mask_0 = const()[name = string("op_38788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38788_cast_fp16 = slice_by_index(begin = var_38788_begin_0, end = var_38788_end_0, end_mask = var_38788_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38788_cast_fp16")]; + tensor var_38795_begin_0 = const()[name = string("op_38795_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38795_end_0 = const()[name = string("op_38795_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38795_end_mask_0 = const()[name = string("op_38795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38795_cast_fp16 = slice_by_index(begin = var_38795_begin_0, end = var_38795_end_0, end_mask = var_38795_end_mask_0, x = var_38473_cast_fp16)[name = string("op_38795_cast_fp16")]; + tensor var_38802_begin_0 = const()[name = string("op_38802_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38802_end_0 = const()[name = string("op_38802_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38802_end_mask_0 = const()[name = string("op_38802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38802_cast_fp16 = slice_by_index(begin = var_38802_begin_0, end = var_38802_end_0, end_mask = var_38802_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38802_cast_fp16")]; + tensor var_38809_begin_0 = const()[name = string("op_38809_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38809_end_0 = const()[name = string("op_38809_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38809_end_mask_0 = const()[name = string("op_38809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38809_cast_fp16 = slice_by_index(begin = var_38809_begin_0, end = var_38809_end_0, end_mask = var_38809_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38809_cast_fp16")]; + tensor var_38816_begin_0 = const()[name = string("op_38816_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38816_end_0 = const()[name = string("op_38816_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38816_end_mask_0 = const()[name = string("op_38816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38816_cast_fp16 = slice_by_index(begin = var_38816_begin_0, end = var_38816_end_0, end_mask = var_38816_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38816_cast_fp16")]; + tensor var_38823_begin_0 = const()[name = string("op_38823_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38823_end_0 = const()[name = string("op_38823_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38823_end_mask_0 = const()[name = string("op_38823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38823_cast_fp16 = slice_by_index(begin = var_38823_begin_0, end = var_38823_end_0, end_mask = var_38823_end_mask_0, x = var_38477_cast_fp16)[name = string("op_38823_cast_fp16")]; + tensor var_38830_begin_0 = const()[name = string("op_38830_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38830_end_0 = const()[name = string("op_38830_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38830_end_mask_0 = const()[name = string("op_38830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38830_cast_fp16 = slice_by_index(begin = var_38830_begin_0, end = var_38830_end_0, end_mask = var_38830_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38830_cast_fp16")]; + tensor var_38837_begin_0 = const()[name = string("op_38837_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38837_end_0 = const()[name = string("op_38837_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38837_end_mask_0 = const()[name = string("op_38837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38837_cast_fp16 = slice_by_index(begin = var_38837_begin_0, end = var_38837_end_0, end_mask = var_38837_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38837_cast_fp16")]; + tensor var_38844_begin_0 = const()[name = string("op_38844_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38844_end_0 = const()[name = string("op_38844_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38844_end_mask_0 = const()[name = string("op_38844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38844_cast_fp16 = slice_by_index(begin = var_38844_begin_0, end = var_38844_end_0, end_mask = var_38844_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38844_cast_fp16")]; + tensor var_38851_begin_0 = const()[name = string("op_38851_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38851_end_0 = const()[name = string("op_38851_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38851_end_mask_0 = const()[name = string("op_38851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38851_cast_fp16 = slice_by_index(begin = var_38851_begin_0, end = var_38851_end_0, end_mask = var_38851_end_mask_0, x = var_38481_cast_fp16)[name = string("op_38851_cast_fp16")]; + tensor var_38858_begin_0 = const()[name = string("op_38858_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38858_end_0 = const()[name = string("op_38858_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38858_end_mask_0 = const()[name = string("op_38858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38858_cast_fp16 = slice_by_index(begin = var_38858_begin_0, end = var_38858_end_0, end_mask = var_38858_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38858_cast_fp16")]; + tensor var_38865_begin_0 = const()[name = string("op_38865_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38865_end_0 = const()[name = string("op_38865_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38865_end_mask_0 = const()[name = string("op_38865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38865_cast_fp16 = slice_by_index(begin = var_38865_begin_0, end = var_38865_end_0, end_mask = var_38865_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38865_cast_fp16")]; + tensor var_38872_begin_0 = const()[name = string("op_38872_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38872_end_0 = const()[name = string("op_38872_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38872_end_mask_0 = const()[name = string("op_38872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38872_cast_fp16 = slice_by_index(begin = var_38872_begin_0, end = var_38872_end_0, end_mask = var_38872_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38872_cast_fp16")]; + tensor var_38879_begin_0 = const()[name = string("op_38879_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38879_end_0 = const()[name = string("op_38879_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38879_end_mask_0 = const()[name = string("op_38879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38879_cast_fp16 = slice_by_index(begin = var_38879_begin_0, end = var_38879_end_0, end_mask = var_38879_end_mask_0, x = var_38485_cast_fp16)[name = string("op_38879_cast_fp16")]; + tensor var_38886_begin_0 = const()[name = string("op_38886_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38886_end_0 = const()[name = string("op_38886_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38886_end_mask_0 = const()[name = string("op_38886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38886_cast_fp16 = slice_by_index(begin = var_38886_begin_0, end = var_38886_end_0, end_mask = var_38886_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38886_cast_fp16")]; + tensor var_38893_begin_0 = const()[name = string("op_38893_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38893_end_0 = const()[name = string("op_38893_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38893_end_mask_0 = const()[name = string("op_38893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38893_cast_fp16 = slice_by_index(begin = var_38893_begin_0, end = var_38893_end_0, end_mask = var_38893_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38893_cast_fp16")]; + tensor var_38900_begin_0 = const()[name = string("op_38900_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38900_end_0 = const()[name = string("op_38900_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38900_end_mask_0 = const()[name = string("op_38900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38900_cast_fp16 = slice_by_index(begin = var_38900_begin_0, end = var_38900_end_0, end_mask = var_38900_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38900_cast_fp16")]; + tensor var_38907_begin_0 = const()[name = string("op_38907_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38907_end_0 = const()[name = string("op_38907_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38907_end_mask_0 = const()[name = string("op_38907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38907_cast_fp16 = slice_by_index(begin = var_38907_begin_0, end = var_38907_end_0, end_mask = var_38907_end_mask_0, x = var_38489_cast_fp16)[name = string("op_38907_cast_fp16")]; + tensor var_38914_begin_0 = const()[name = string("op_38914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38914_end_0 = const()[name = string("op_38914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38914_end_mask_0 = const()[name = string("op_38914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38914_cast_fp16 = slice_by_index(begin = var_38914_begin_0, end = var_38914_end_0, end_mask = var_38914_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38914_cast_fp16")]; + tensor var_38921_begin_0 = const()[name = string("op_38921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38921_end_0 = const()[name = string("op_38921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38921_end_mask_0 = const()[name = string("op_38921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38921_cast_fp16 = slice_by_index(begin = var_38921_begin_0, end = var_38921_end_0, end_mask = var_38921_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38921_cast_fp16")]; + tensor var_38928_begin_0 = const()[name = string("op_38928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38928_end_0 = const()[name = string("op_38928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38928_end_mask_0 = const()[name = string("op_38928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38928_cast_fp16 = slice_by_index(begin = var_38928_begin_0, end = var_38928_end_0, end_mask = var_38928_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38928_cast_fp16")]; + tensor var_38935_begin_0 = const()[name = string("op_38935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38935_end_0 = const()[name = string("op_38935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38935_end_mask_0 = const()[name = string("op_38935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38935_cast_fp16 = slice_by_index(begin = var_38935_begin_0, end = var_38935_end_0, end_mask = var_38935_end_mask_0, x = var_38493_cast_fp16)[name = string("op_38935_cast_fp16")]; + tensor var_38942_begin_0 = const()[name = string("op_38942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38942_end_0 = const()[name = string("op_38942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38942_end_mask_0 = const()[name = string("op_38942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38942_cast_fp16 = slice_by_index(begin = var_38942_begin_0, end = var_38942_end_0, end_mask = var_38942_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38942_cast_fp16")]; + tensor var_38949_begin_0 = const()[name = string("op_38949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38949_end_0 = const()[name = string("op_38949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38949_end_mask_0 = const()[name = string("op_38949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38949_cast_fp16 = slice_by_index(begin = var_38949_begin_0, end = var_38949_end_0, end_mask = var_38949_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38949_cast_fp16")]; + tensor var_38956_begin_0 = const()[name = string("op_38956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38956_end_0 = const()[name = string("op_38956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38956_end_mask_0 = const()[name = string("op_38956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38956_cast_fp16 = slice_by_index(begin = var_38956_begin_0, end = var_38956_end_0, end_mask = var_38956_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38956_cast_fp16")]; + tensor var_38963_begin_0 = const()[name = string("op_38963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38963_end_0 = const()[name = string("op_38963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38963_end_mask_0 = const()[name = string("op_38963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38963_cast_fp16 = slice_by_index(begin = var_38963_begin_0, end = var_38963_end_0, end_mask = var_38963_end_mask_0, x = var_38497_cast_fp16)[name = string("op_38963_cast_fp16")]; + tensor var_38970_begin_0 = const()[name = string("op_38970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38970_end_0 = const()[name = string("op_38970_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38970_end_mask_0 = const()[name = string("op_38970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38970_cast_fp16 = slice_by_index(begin = var_38970_begin_0, end = var_38970_end_0, end_mask = var_38970_end_mask_0, x = var_38501_cast_fp16)[name = string("op_38970_cast_fp16")]; + tensor var_38977_begin_0 = const()[name = string("op_38977_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38977_end_0 = const()[name = string("op_38977_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38977_end_mask_0 = const()[name = string("op_38977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38977_cast_fp16 = slice_by_index(begin = var_38977_begin_0, end = var_38977_end_0, end_mask = var_38977_end_mask_0, x = var_38501_cast_fp16)[name = string("op_38977_cast_fp16")]; + tensor var_38984_begin_0 = const()[name = string("op_38984_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38984_end_0 = const()[name = string("op_38984_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38984_end_mask_0 = const()[name = string("op_38984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38984_cast_fp16 = slice_by_index(begin = var_38984_begin_0, end = var_38984_end_0, end_mask = var_38984_end_mask_0, x = var_38501_cast_fp16)[name = string("op_38984_cast_fp16")]; + tensor var_38991_begin_0 = const()[name = string("op_38991_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38991_end_0 = const()[name = string("op_38991_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38991_end_mask_0 = const()[name = string("op_38991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38991_cast_fp16 = slice_by_index(begin = var_38991_begin_0, end = var_38991_end_0, end_mask = var_38991_end_mask_0, x = var_38501_cast_fp16)[name = string("op_38991_cast_fp16")]; + tensor var_38998_begin_0 = const()[name = string("op_38998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38998_end_0 = const()[name = string("op_38998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38998_end_mask_0 = const()[name = string("op_38998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38998_cast_fp16 = slice_by_index(begin = var_38998_begin_0, end = var_38998_end_0, end_mask = var_38998_end_mask_0, x = var_38505_cast_fp16)[name = string("op_38998_cast_fp16")]; + tensor var_39005_begin_0 = const()[name = string("op_39005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39005_end_0 = const()[name = string("op_39005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39005_end_mask_0 = const()[name = string("op_39005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39005_cast_fp16 = slice_by_index(begin = var_39005_begin_0, end = var_39005_end_0, end_mask = var_39005_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39005_cast_fp16")]; + tensor var_39012_begin_0 = const()[name = string("op_39012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39012_end_0 = const()[name = string("op_39012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39012_end_mask_0 = const()[name = string("op_39012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39012_cast_fp16 = slice_by_index(begin = var_39012_begin_0, end = var_39012_end_0, end_mask = var_39012_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39012_cast_fp16")]; + tensor var_39019_begin_0 = const()[name = string("op_39019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39019_end_0 = const()[name = string("op_39019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39019_end_mask_0 = const()[name = string("op_39019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39019_cast_fp16 = slice_by_index(begin = var_39019_begin_0, end = var_39019_end_0, end_mask = var_39019_end_mask_0, x = var_38505_cast_fp16)[name = string("op_39019_cast_fp16")]; + tensor var_39026_begin_0 = const()[name = string("op_39026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39026_end_0 = const()[name = string("op_39026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39026_end_mask_0 = const()[name = string("op_39026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39026_cast_fp16 = slice_by_index(begin = var_39026_begin_0, end = var_39026_end_0, end_mask = var_39026_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39026_cast_fp16")]; + tensor var_39033_begin_0 = const()[name = string("op_39033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39033_end_0 = const()[name = string("op_39033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39033_end_mask_0 = const()[name = string("op_39033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39033_cast_fp16 = slice_by_index(begin = var_39033_begin_0, end = var_39033_end_0, end_mask = var_39033_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39033_cast_fp16")]; + tensor var_39040_begin_0 = const()[name = string("op_39040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39040_end_0 = const()[name = string("op_39040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39040_end_mask_0 = const()[name = string("op_39040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39040_cast_fp16 = slice_by_index(begin = var_39040_begin_0, end = var_39040_end_0, end_mask = var_39040_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39040_cast_fp16")]; + tensor var_39047_begin_0 = const()[name = string("op_39047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39047_end_0 = const()[name = string("op_39047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39047_end_mask_0 = const()[name = string("op_39047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39047_cast_fp16 = slice_by_index(begin = var_39047_begin_0, end = var_39047_end_0, end_mask = var_39047_end_mask_0, x = var_38509_cast_fp16)[name = string("op_39047_cast_fp16")]; + tensor var_39054_begin_0 = const()[name = string("op_39054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39054_end_0 = const()[name = string("op_39054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39054_end_mask_0 = const()[name = string("op_39054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39054_cast_fp16 = slice_by_index(begin = var_39054_begin_0, end = var_39054_end_0, end_mask = var_39054_end_mask_0, x = var_38513_cast_fp16)[name = string("op_39054_cast_fp16")]; + tensor var_39061_begin_0 = const()[name = string("op_39061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39061_end_0 = const()[name = string("op_39061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39061_end_mask_0 = const()[name = string("op_39061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39061_cast_fp16 = slice_by_index(begin = var_39061_begin_0, end = var_39061_end_0, end_mask = var_39061_end_mask_0, x = var_38513_cast_fp16)[name = string("op_39061_cast_fp16")]; + tensor var_39068_begin_0 = const()[name = string("op_39068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39068_end_0 = const()[name = string("op_39068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39068_end_mask_0 = const()[name = string("op_39068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39068_cast_fp16 = slice_by_index(begin = var_39068_begin_0, end = var_39068_end_0, end_mask = var_39068_end_mask_0, x = var_38513_cast_fp16)[name = string("op_39068_cast_fp16")]; + tensor var_39075_begin_0 = const()[name = string("op_39075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39075_end_0 = const()[name = string("op_39075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39075_end_mask_0 = const()[name = string("op_39075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39075_cast_fp16 = slice_by_index(begin = var_39075_begin_0, end = var_39075_end_0, end_mask = var_39075_end_mask_0, x = var_38513_cast_fp16)[name = string("op_39075_cast_fp16")]; + tensor k_51_perm_0 = const()[name = string("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_39080_begin_0 = const()[name = string("op_39080_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39080_end_0 = const()[name = string("op_39080_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_39080_end_mask_0 = const()[name = string("op_39080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = key_51_cast_fp16)[name = string("transpose_6")]; + tensor var_39080_cast_fp16 = slice_by_index(begin = var_39080_begin_0, end = var_39080_end_0, end_mask = var_39080_end_mask_0, x = k_51_cast_fp16)[name = string("op_39080_cast_fp16")]; + tensor var_39084_begin_0 = const()[name = string("op_39084_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_39084_end_0 = const()[name = string("op_39084_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_39084_end_mask_0 = const()[name = string("op_39084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39084_cast_fp16 = slice_by_index(begin = var_39084_begin_0, end = var_39084_end_0, end_mask = var_39084_end_mask_0, x = k_51_cast_fp16)[name = string("op_39084_cast_fp16")]; + tensor var_39088_begin_0 = const()[name = string("op_39088_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_39088_end_0 = const()[name = string("op_39088_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_39088_end_mask_0 = const()[name = string("op_39088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39088_cast_fp16 = slice_by_index(begin = var_39088_begin_0, end = var_39088_end_0, end_mask = var_39088_end_mask_0, x = k_51_cast_fp16)[name = string("op_39088_cast_fp16")]; + tensor var_39092_begin_0 = const()[name = string("op_39092_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_39092_end_0 = const()[name = string("op_39092_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_39092_end_mask_0 = const()[name = string("op_39092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39092_cast_fp16 = slice_by_index(begin = var_39092_begin_0, end = var_39092_end_0, end_mask = var_39092_end_mask_0, x = k_51_cast_fp16)[name = string("op_39092_cast_fp16")]; + tensor var_39096_begin_0 = const()[name = string("op_39096_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_39096_end_0 = const()[name = string("op_39096_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_39096_end_mask_0 = const()[name = string("op_39096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39096_cast_fp16 = slice_by_index(begin = var_39096_begin_0, end = var_39096_end_0, end_mask = var_39096_end_mask_0, x = k_51_cast_fp16)[name = string("op_39096_cast_fp16")]; + tensor var_39100_begin_0 = const()[name = string("op_39100_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_39100_end_0 = const()[name = string("op_39100_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_39100_end_mask_0 = const()[name = string("op_39100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39100_cast_fp16 = slice_by_index(begin = var_39100_begin_0, end = var_39100_end_0, end_mask = var_39100_end_mask_0, x = k_51_cast_fp16)[name = string("op_39100_cast_fp16")]; + tensor var_39104_begin_0 = const()[name = string("op_39104_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_39104_end_0 = const()[name = string("op_39104_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_39104_end_mask_0 = const()[name = string("op_39104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39104_cast_fp16 = slice_by_index(begin = var_39104_begin_0, end = var_39104_end_0, end_mask = var_39104_end_mask_0, x = k_51_cast_fp16)[name = string("op_39104_cast_fp16")]; + tensor var_39108_begin_0 = const()[name = string("op_39108_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_39108_end_0 = const()[name = string("op_39108_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_39108_end_mask_0 = const()[name = string("op_39108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39108_cast_fp16 = slice_by_index(begin = var_39108_begin_0, end = var_39108_end_0, end_mask = var_39108_end_mask_0, x = k_51_cast_fp16)[name = string("op_39108_cast_fp16")]; + tensor var_39112_begin_0 = const()[name = string("op_39112_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_39112_end_0 = const()[name = string("op_39112_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_39112_end_mask_0 = const()[name = string("op_39112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39112_cast_fp16 = slice_by_index(begin = var_39112_begin_0, end = var_39112_end_0, end_mask = var_39112_end_mask_0, x = k_51_cast_fp16)[name = string("op_39112_cast_fp16")]; + tensor var_39116_begin_0 = const()[name = string("op_39116_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_39116_end_0 = const()[name = string("op_39116_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_39116_end_mask_0 = const()[name = string("op_39116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39116_cast_fp16 = slice_by_index(begin = var_39116_begin_0, end = var_39116_end_0, end_mask = var_39116_end_mask_0, x = k_51_cast_fp16)[name = string("op_39116_cast_fp16")]; + tensor var_39120_begin_0 = const()[name = string("op_39120_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_39120_end_0 = const()[name = string("op_39120_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_39120_end_mask_0 = const()[name = string("op_39120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39120_cast_fp16 = slice_by_index(begin = var_39120_begin_0, end = var_39120_end_0, end_mask = var_39120_end_mask_0, x = k_51_cast_fp16)[name = string("op_39120_cast_fp16")]; + tensor var_39124_begin_0 = const()[name = string("op_39124_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_39124_end_0 = const()[name = string("op_39124_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_39124_end_mask_0 = const()[name = string("op_39124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39124_cast_fp16 = slice_by_index(begin = var_39124_begin_0, end = var_39124_end_0, end_mask = var_39124_end_mask_0, x = k_51_cast_fp16)[name = string("op_39124_cast_fp16")]; + tensor var_39128_begin_0 = const()[name = string("op_39128_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_39128_end_0 = const()[name = string("op_39128_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_39128_end_mask_0 = const()[name = string("op_39128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39128_cast_fp16 = slice_by_index(begin = var_39128_begin_0, end = var_39128_end_0, end_mask = var_39128_end_mask_0, x = k_51_cast_fp16)[name = string("op_39128_cast_fp16")]; + tensor var_39132_begin_0 = const()[name = string("op_39132_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_39132_end_0 = const()[name = string("op_39132_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_39132_end_mask_0 = const()[name = string("op_39132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39132_cast_fp16 = slice_by_index(begin = var_39132_begin_0, end = var_39132_end_0, end_mask = var_39132_end_mask_0, x = k_51_cast_fp16)[name = string("op_39132_cast_fp16")]; + tensor var_39136_begin_0 = const()[name = string("op_39136_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_39136_end_0 = const()[name = string("op_39136_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_39136_end_mask_0 = const()[name = string("op_39136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39136_cast_fp16 = slice_by_index(begin = var_39136_begin_0, end = var_39136_end_0, end_mask = var_39136_end_mask_0, x = k_51_cast_fp16)[name = string("op_39136_cast_fp16")]; + tensor var_39140_begin_0 = const()[name = string("op_39140_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_39140_end_0 = const()[name = string("op_39140_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_39140_end_mask_0 = const()[name = string("op_39140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39140_cast_fp16 = slice_by_index(begin = var_39140_begin_0, end = var_39140_end_0, end_mask = var_39140_end_mask_0, x = k_51_cast_fp16)[name = string("op_39140_cast_fp16")]; + tensor var_39144_begin_0 = const()[name = string("op_39144_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_39144_end_0 = const()[name = string("op_39144_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_39144_end_mask_0 = const()[name = string("op_39144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39144_cast_fp16 = slice_by_index(begin = var_39144_begin_0, end = var_39144_end_0, end_mask = var_39144_end_mask_0, x = k_51_cast_fp16)[name = string("op_39144_cast_fp16")]; + tensor var_39148_begin_0 = const()[name = string("op_39148_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_39148_end_0 = const()[name = string("op_39148_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_39148_end_mask_0 = const()[name = string("op_39148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39148_cast_fp16 = slice_by_index(begin = var_39148_begin_0, end = var_39148_end_0, end_mask = var_39148_end_mask_0, x = k_51_cast_fp16)[name = string("op_39148_cast_fp16")]; + tensor var_39152_begin_0 = const()[name = string("op_39152_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_39152_end_0 = const()[name = string("op_39152_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_39152_end_mask_0 = const()[name = string("op_39152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39152_cast_fp16 = slice_by_index(begin = var_39152_begin_0, end = var_39152_end_0, end_mask = var_39152_end_mask_0, x = k_51_cast_fp16)[name = string("op_39152_cast_fp16")]; + tensor var_39156_begin_0 = const()[name = string("op_39156_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_39156_end_0 = const()[name = string("op_39156_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_39156_end_mask_0 = const()[name = string("op_39156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39156_cast_fp16 = slice_by_index(begin = var_39156_begin_0, end = var_39156_end_0, end_mask = var_39156_end_mask_0, x = k_51_cast_fp16)[name = string("op_39156_cast_fp16")]; + tensor var_39158_begin_0 = const()[name = string("op_39158_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39158_end_0 = const()[name = string("op_39158_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39158_end_mask_0 = const()[name = string("op_39158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39158_cast_fp16 = slice_by_index(begin = var_39158_begin_0, end = var_39158_end_0, end_mask = var_39158_end_mask_0, x = value_51_cast_fp16)[name = string("op_39158_cast_fp16")]; + tensor var_39162_begin_0 = const()[name = string("op_39162_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39162_end_0 = const()[name = string("op_39162_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_39162_end_mask_0 = const()[name = string("op_39162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39162_cast_fp16 = slice_by_index(begin = var_39162_begin_0, end = var_39162_end_0, end_mask = var_39162_end_mask_0, x = value_51_cast_fp16)[name = string("op_39162_cast_fp16")]; + tensor var_39166_begin_0 = const()[name = string("op_39166_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39166_end_0 = const()[name = string("op_39166_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_39166_end_mask_0 = const()[name = string("op_39166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39166_cast_fp16 = slice_by_index(begin = var_39166_begin_0, end = var_39166_end_0, end_mask = var_39166_end_mask_0, x = value_51_cast_fp16)[name = string("op_39166_cast_fp16")]; + tensor var_39170_begin_0 = const()[name = string("op_39170_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39170_end_0 = const()[name = string("op_39170_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_39170_end_mask_0 = const()[name = string("op_39170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39170_cast_fp16 = slice_by_index(begin = var_39170_begin_0, end = var_39170_end_0, end_mask = var_39170_end_mask_0, x = value_51_cast_fp16)[name = string("op_39170_cast_fp16")]; + tensor var_39174_begin_0 = const()[name = string("op_39174_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39174_end_0 = const()[name = string("op_39174_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_39174_end_mask_0 = const()[name = string("op_39174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39174_cast_fp16 = slice_by_index(begin = var_39174_begin_0, end = var_39174_end_0, end_mask = var_39174_end_mask_0, x = value_51_cast_fp16)[name = string("op_39174_cast_fp16")]; + tensor var_39178_begin_0 = const()[name = string("op_39178_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39178_end_0 = const()[name = string("op_39178_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_39178_end_mask_0 = const()[name = string("op_39178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39178_cast_fp16 = slice_by_index(begin = var_39178_begin_0, end = var_39178_end_0, end_mask = var_39178_end_mask_0, x = value_51_cast_fp16)[name = string("op_39178_cast_fp16")]; + tensor var_39182_begin_0 = const()[name = string("op_39182_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39182_end_0 = const()[name = string("op_39182_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_39182_end_mask_0 = const()[name = string("op_39182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39182_cast_fp16 = slice_by_index(begin = var_39182_begin_0, end = var_39182_end_0, end_mask = var_39182_end_mask_0, x = value_51_cast_fp16)[name = string("op_39182_cast_fp16")]; + tensor var_39186_begin_0 = const()[name = string("op_39186_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39186_end_0 = const()[name = string("op_39186_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_39186_end_mask_0 = const()[name = string("op_39186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39186_cast_fp16 = slice_by_index(begin = var_39186_begin_0, end = var_39186_end_0, end_mask = var_39186_end_mask_0, x = value_51_cast_fp16)[name = string("op_39186_cast_fp16")]; + tensor var_39190_begin_0 = const()[name = string("op_39190_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39190_end_0 = const()[name = string("op_39190_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_39190_end_mask_0 = const()[name = string("op_39190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39190_cast_fp16 = slice_by_index(begin = var_39190_begin_0, end = var_39190_end_0, end_mask = var_39190_end_mask_0, x = value_51_cast_fp16)[name = string("op_39190_cast_fp16")]; + tensor var_39194_begin_0 = const()[name = string("op_39194_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39194_end_0 = const()[name = string("op_39194_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_39194_end_mask_0 = const()[name = string("op_39194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39194_cast_fp16 = slice_by_index(begin = var_39194_begin_0, end = var_39194_end_0, end_mask = var_39194_end_mask_0, x = value_51_cast_fp16)[name = string("op_39194_cast_fp16")]; + tensor var_39198_begin_0 = const()[name = string("op_39198_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39198_end_0 = const()[name = string("op_39198_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_39198_end_mask_0 = const()[name = string("op_39198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39198_cast_fp16 = slice_by_index(begin = var_39198_begin_0, end = var_39198_end_0, end_mask = var_39198_end_mask_0, x = value_51_cast_fp16)[name = string("op_39198_cast_fp16")]; + tensor var_39202_begin_0 = const()[name = string("op_39202_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39202_end_0 = const()[name = string("op_39202_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_39202_end_mask_0 = const()[name = string("op_39202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39202_cast_fp16 = slice_by_index(begin = var_39202_begin_0, end = var_39202_end_0, end_mask = var_39202_end_mask_0, x = value_51_cast_fp16)[name = string("op_39202_cast_fp16")]; + tensor var_39206_begin_0 = const()[name = string("op_39206_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39206_end_0 = const()[name = string("op_39206_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_39206_end_mask_0 = const()[name = string("op_39206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39206_cast_fp16 = slice_by_index(begin = var_39206_begin_0, end = var_39206_end_0, end_mask = var_39206_end_mask_0, x = value_51_cast_fp16)[name = string("op_39206_cast_fp16")]; + tensor var_39210_begin_0 = const()[name = string("op_39210_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39210_end_0 = const()[name = string("op_39210_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_39210_end_mask_0 = const()[name = string("op_39210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39210_cast_fp16 = slice_by_index(begin = var_39210_begin_0, end = var_39210_end_0, end_mask = var_39210_end_mask_0, x = value_51_cast_fp16)[name = string("op_39210_cast_fp16")]; + tensor var_39214_begin_0 = const()[name = string("op_39214_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39214_end_0 = const()[name = string("op_39214_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_39214_end_mask_0 = const()[name = string("op_39214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39214_cast_fp16 = slice_by_index(begin = var_39214_begin_0, end = var_39214_end_0, end_mask = var_39214_end_mask_0, x = value_51_cast_fp16)[name = string("op_39214_cast_fp16")]; + tensor var_39218_begin_0 = const()[name = string("op_39218_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39218_end_0 = const()[name = string("op_39218_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_39218_end_mask_0 = const()[name = string("op_39218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39218_cast_fp16 = slice_by_index(begin = var_39218_begin_0, end = var_39218_end_0, end_mask = var_39218_end_mask_0, x = value_51_cast_fp16)[name = string("op_39218_cast_fp16")]; + tensor var_39222_begin_0 = const()[name = string("op_39222_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39222_end_0 = const()[name = string("op_39222_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_39222_end_mask_0 = const()[name = string("op_39222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39222_cast_fp16 = slice_by_index(begin = var_39222_begin_0, end = var_39222_end_0, end_mask = var_39222_end_mask_0, x = value_51_cast_fp16)[name = string("op_39222_cast_fp16")]; + tensor var_39226_begin_0 = const()[name = string("op_39226_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39226_end_0 = const()[name = string("op_39226_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_39226_end_mask_0 = const()[name = string("op_39226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39226_cast_fp16 = slice_by_index(begin = var_39226_begin_0, end = var_39226_end_0, end_mask = var_39226_end_mask_0, x = value_51_cast_fp16)[name = string("op_39226_cast_fp16")]; + tensor var_39230_begin_0 = const()[name = string("op_39230_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39230_end_0 = const()[name = string("op_39230_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_39230_end_mask_0 = const()[name = string("op_39230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39230_cast_fp16 = slice_by_index(begin = var_39230_begin_0, end = var_39230_end_0, end_mask = var_39230_end_mask_0, x = value_51_cast_fp16)[name = string("op_39230_cast_fp16")]; + tensor var_39234_begin_0 = const()[name = string("op_39234_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39234_end_0 = const()[name = string("op_39234_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_39234_end_mask_0 = const()[name = string("op_39234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39234_cast_fp16 = slice_by_index(begin = var_39234_begin_0, end = var_39234_end_0, end_mask = var_39234_end_mask_0, x = value_51_cast_fp16)[name = string("op_39234_cast_fp16")]; + string _SplitHeadsQ__mh_w_4001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4001_equation_0, values = (var_39080_cast_fp16, var_38522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4001_cast_fp16")]; + string _SplitHeadsQ__mh_w_4003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4003_equation_0, values = (var_39080_cast_fp16, var_38529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4003_cast_fp16")]; + string _SplitHeadsQ__mh_w_4005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4005_equation_0, values = (var_39080_cast_fp16, var_38536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4005_cast_fp16")]; + string _SplitHeadsQ__mh_w_4007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4007_equation_0, values = (var_39080_cast_fp16, var_38543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4007_cast_fp16")]; + string _SplitHeadsQ__mh_w_4009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4009_equation_0, values = (var_39084_cast_fp16, var_38550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4009_cast_fp16")]; + string _SplitHeadsQ__mh_w_4011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4011_equation_0, values = (var_39084_cast_fp16, var_38557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4011_cast_fp16")]; + string _SplitHeadsQ__mh_w_4013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4013_equation_0, values = (var_39084_cast_fp16, var_38564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4013_cast_fp16")]; + string _SplitHeadsQ__mh_w_4015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4015_equation_0, values = (var_39084_cast_fp16, var_38571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4015_cast_fp16")]; + string _SplitHeadsQ__mh_w_4017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4017_equation_0, values = (var_39088_cast_fp16, var_38578_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4017_cast_fp16")]; + string _SplitHeadsQ__mh_w_4019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4019_equation_0, values = (var_39088_cast_fp16, var_38585_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4019_cast_fp16")]; + string _SplitHeadsQ__mh_w_4021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4021_equation_0, values = (var_39088_cast_fp16, var_38592_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4021_cast_fp16")]; + string _SplitHeadsQ__mh_w_4023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4023_equation_0, values = (var_39088_cast_fp16, var_38599_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4023_cast_fp16")]; + string _SplitHeadsQ__mh_w_4025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4025_equation_0, values = (var_39092_cast_fp16, var_38606_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4025_cast_fp16")]; + string _SplitHeadsQ__mh_w_4027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4027_equation_0, values = (var_39092_cast_fp16, var_38613_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4027_cast_fp16")]; + string _SplitHeadsQ__mh_w_4029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4029_equation_0, values = (var_39092_cast_fp16, var_38620_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4029_cast_fp16")]; + string _SplitHeadsQ__mh_w_4031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4031_equation_0, values = (var_39092_cast_fp16, var_38627_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4031_cast_fp16")]; + string _SplitHeadsQ__mh_w_4033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4033_equation_0, values = (var_39096_cast_fp16, var_38634_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4033_cast_fp16")]; + string _SplitHeadsQ__mh_w_4035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4035_equation_0, values = (var_39096_cast_fp16, var_38641_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4035_cast_fp16")]; + string _SplitHeadsQ__mh_w_4037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4037_equation_0, values = (var_39096_cast_fp16, var_38648_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4037_cast_fp16")]; + string _SplitHeadsQ__mh_w_4039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4039_equation_0, values = (var_39096_cast_fp16, var_38655_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4039_cast_fp16")]; + string _SplitHeadsQ__mh_w_4041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4041_equation_0, values = (var_39100_cast_fp16, var_38662_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4041_cast_fp16")]; + string _SplitHeadsQ__mh_w_4043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4043_equation_0, values = (var_39100_cast_fp16, var_38669_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4043_cast_fp16")]; + string _SplitHeadsQ__mh_w_4045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4045_equation_0, values = (var_39100_cast_fp16, var_38676_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4045_cast_fp16")]; + string _SplitHeadsQ__mh_w_4047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4047_equation_0, values = (var_39100_cast_fp16, var_38683_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4047_cast_fp16")]; + string _SplitHeadsQ__mh_w_4049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4049_equation_0, values = (var_39104_cast_fp16, var_38690_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4049_cast_fp16")]; + string _SplitHeadsQ__mh_w_4051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4051_equation_0, values = (var_39104_cast_fp16, var_38697_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4051_cast_fp16")]; + string _SplitHeadsQ__mh_w_4053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4053_equation_0, values = (var_39104_cast_fp16, var_38704_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4053_cast_fp16")]; + string _SplitHeadsQ__mh_w_4055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4055_equation_0, values = (var_39104_cast_fp16, var_38711_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4055_cast_fp16")]; + string _SplitHeadsQ__mh_w_4057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4057_equation_0, values = (var_39108_cast_fp16, var_38718_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4057_cast_fp16")]; + string _SplitHeadsQ__mh_w_4059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4059_equation_0, values = (var_39108_cast_fp16, var_38725_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4059_cast_fp16")]; + string _SplitHeadsQ__mh_w_4061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4061_equation_0, values = (var_39108_cast_fp16, var_38732_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4061_cast_fp16")]; + string _SplitHeadsQ__mh_w_4063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4063_equation_0, values = (var_39108_cast_fp16, var_38739_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4063_cast_fp16")]; + string _SplitHeadsQ__mh_w_4065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4065_equation_0, values = (var_39112_cast_fp16, var_38746_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4065_cast_fp16")]; + string _SplitHeadsQ__mh_w_4067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4067_equation_0, values = (var_39112_cast_fp16, var_38753_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4067_cast_fp16")]; + string _SplitHeadsQ__mh_w_4069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4069_equation_0, values = (var_39112_cast_fp16, var_38760_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4069_cast_fp16")]; + string _SplitHeadsQ__mh_w_4071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4071_equation_0, values = (var_39112_cast_fp16, var_38767_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4071_cast_fp16")]; + string _SplitHeadsQ__mh_w_4073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4073_equation_0, values = (var_39116_cast_fp16, var_38774_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4073_cast_fp16")]; + string _SplitHeadsQ__mh_w_4075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4075_equation_0, values = (var_39116_cast_fp16, var_38781_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4075_cast_fp16")]; + string _SplitHeadsQ__mh_w_4077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4077_equation_0, values = (var_39116_cast_fp16, var_38788_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4077_cast_fp16")]; + string _SplitHeadsQ__mh_w_4079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4079_equation_0, values = (var_39116_cast_fp16, var_38795_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4079_cast_fp16")]; + string _SplitHeadsQ__mh_w_4081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4081_equation_0, values = (var_39120_cast_fp16, var_38802_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4081_cast_fp16")]; + string _SplitHeadsQ__mh_w_4083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4083_equation_0, values = (var_39120_cast_fp16, var_38809_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4083_cast_fp16")]; + string _SplitHeadsQ__mh_w_4085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4085_equation_0, values = (var_39120_cast_fp16, var_38816_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4085_cast_fp16")]; + string _SplitHeadsQ__mh_w_4087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4087_equation_0, values = (var_39120_cast_fp16, var_38823_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4087_cast_fp16")]; + string _SplitHeadsQ__mh_w_4089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4089_equation_0, values = (var_39124_cast_fp16, var_38830_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4089_cast_fp16")]; + string _SplitHeadsQ__mh_w_4091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4091_equation_0, values = (var_39124_cast_fp16, var_38837_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4091_cast_fp16")]; + string _SplitHeadsQ__mh_w_4093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4093_equation_0, values = (var_39124_cast_fp16, var_38844_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4093_cast_fp16")]; + string _SplitHeadsQ__mh_w_4095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4095_equation_0, values = (var_39124_cast_fp16, var_38851_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4095_cast_fp16")]; + string _SplitHeadsQ__mh_w_4097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4097_equation_0, values = (var_39128_cast_fp16, var_38858_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4097_cast_fp16")]; + string _SplitHeadsQ__mh_w_4099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4099_equation_0, values = (var_39128_cast_fp16, var_38865_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4099_cast_fp16")]; + string _SplitHeadsQ__mh_w_4101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4101_equation_0, values = (var_39128_cast_fp16, var_38872_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4101_cast_fp16")]; + string _SplitHeadsQ__mh_w_4103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4103_equation_0, values = (var_39128_cast_fp16, var_38879_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4103_cast_fp16")]; + string _SplitHeadsQ__mh_w_4105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4105_equation_0, values = (var_39132_cast_fp16, var_38886_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4105_cast_fp16")]; + string _SplitHeadsQ__mh_w_4107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4107_equation_0, values = (var_39132_cast_fp16, var_38893_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4107_cast_fp16")]; + string _SplitHeadsQ__mh_w_4109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4109_equation_0, values = (var_39132_cast_fp16, var_38900_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4109_cast_fp16")]; + string _SplitHeadsQ__mh_w_4111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4111_equation_0, values = (var_39132_cast_fp16, var_38907_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4111_cast_fp16")]; + string _SplitHeadsQ__mh_w_4113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4113_equation_0, values = (var_39136_cast_fp16, var_38914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4113_cast_fp16")]; + string _SplitHeadsQ__mh_w_4115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4115_equation_0, values = (var_39136_cast_fp16, var_38921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4115_cast_fp16")]; + string _SplitHeadsQ__mh_w_4117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4117_equation_0, values = (var_39136_cast_fp16, var_38928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4117_cast_fp16")]; + string _SplitHeadsQ__mh_w_4119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4119_equation_0, values = (var_39136_cast_fp16, var_38935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4119_cast_fp16")]; + string _SplitHeadsQ__mh_w_4121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4121_equation_0, values = (var_39140_cast_fp16, var_38942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4121_cast_fp16")]; + string _SplitHeadsQ__mh_w_4123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4123_equation_0, values = (var_39140_cast_fp16, var_38949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4123_cast_fp16")]; + string _SplitHeadsQ__mh_w_4125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4125_equation_0, values = (var_39140_cast_fp16, var_38956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4125_cast_fp16")]; + string _SplitHeadsQ__mh_w_4127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4127_equation_0, values = (var_39140_cast_fp16, var_38963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4127_cast_fp16")]; + string _SplitHeadsQ__mh_w_4129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4129_equation_0, values = (var_39144_cast_fp16, var_38970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4129_cast_fp16")]; + string _SplitHeadsQ__mh_w_4131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4131_equation_0, values = (var_39144_cast_fp16, var_38977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4131_cast_fp16")]; + string _SplitHeadsQ__mh_w_4133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4133_equation_0, values = (var_39144_cast_fp16, var_38984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4133_cast_fp16")]; + string _SplitHeadsQ__mh_w_4135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4135_equation_0, values = (var_39144_cast_fp16, var_38991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4135_cast_fp16")]; + string _SplitHeadsQ__mh_w_4137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4137_equation_0, values = (var_39148_cast_fp16, var_38998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4137_cast_fp16")]; + string _SplitHeadsQ__mh_w_4139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4139_equation_0, values = (var_39148_cast_fp16, var_39005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4139_cast_fp16")]; + string _SplitHeadsQ__mh_w_4141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4141_equation_0, values = (var_39148_cast_fp16, var_39012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4141_cast_fp16")]; + string _SplitHeadsQ__mh_w_4143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4143_equation_0, values = (var_39148_cast_fp16, var_39019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4143_cast_fp16")]; + string _SplitHeadsQ__mh_w_4145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4145_equation_0, values = (var_39152_cast_fp16, var_39026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4145_cast_fp16")]; + string _SplitHeadsQ__mh_w_4147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4147_equation_0, values = (var_39152_cast_fp16, var_39033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4147_cast_fp16")]; + string _SplitHeadsQ__mh_w_4149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4149_equation_0, values = (var_39152_cast_fp16, var_39040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4149_cast_fp16")]; + string _SplitHeadsQ__mh_w_4151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4151_equation_0, values = (var_39152_cast_fp16, var_39047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4151_cast_fp16")]; + string _SplitHeadsQ__mh_w_4153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4153_equation_0, values = (var_39156_cast_fp16, var_39054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4153_cast_fp16")]; + string _SplitHeadsQ__mh_w_4155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4155_equation_0, values = (var_39156_cast_fp16, var_39061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4155_cast_fp16")]; + string _SplitHeadsQ__mh_w_4157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4157_equation_0, values = (var_39156_cast_fp16, var_39068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4157_cast_fp16")]; + string _SplitHeadsQ__mh_w_4159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4159_equation_0, values = (var_39156_cast_fp16, var_39075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4159_cast_fp16")]; + fp16 var_39397_to_fp16 = const()[name = string("op_39397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4001_cast_fp16, y = var_39397_to_fp16)[name = string("aw_chunk_4001_cast_fp16")]; + fp16 var_39399_to_fp16 = const()[name = string("op_39399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4003_cast_fp16, y = var_39399_to_fp16)[name = string("aw_chunk_4003_cast_fp16")]; + fp16 var_39401_to_fp16 = const()[name = string("op_39401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4005_cast_fp16, y = var_39401_to_fp16)[name = string("aw_chunk_4005_cast_fp16")]; + fp16 var_39403_to_fp16 = const()[name = string("op_39403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4007_cast_fp16, y = var_39403_to_fp16)[name = string("aw_chunk_4007_cast_fp16")]; + fp16 var_39405_to_fp16 = const()[name = string("op_39405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4009_cast_fp16, y = var_39405_to_fp16)[name = string("aw_chunk_4009_cast_fp16")]; + fp16 var_39407_to_fp16 = const()[name = string("op_39407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4011_cast_fp16, y = var_39407_to_fp16)[name = string("aw_chunk_4011_cast_fp16")]; + fp16 var_39409_to_fp16 = const()[name = string("op_39409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4013_cast_fp16, y = var_39409_to_fp16)[name = string("aw_chunk_4013_cast_fp16")]; + fp16 var_39411_to_fp16 = const()[name = string("op_39411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4015_cast_fp16, y = var_39411_to_fp16)[name = string("aw_chunk_4015_cast_fp16")]; + fp16 var_39413_to_fp16 = const()[name = string("op_39413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4017_cast_fp16, y = var_39413_to_fp16)[name = string("aw_chunk_4017_cast_fp16")]; + fp16 var_39415_to_fp16 = const()[name = string("op_39415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4019_cast_fp16, y = var_39415_to_fp16)[name = string("aw_chunk_4019_cast_fp16")]; + fp16 var_39417_to_fp16 = const()[name = string("op_39417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4021_cast_fp16, y = var_39417_to_fp16)[name = string("aw_chunk_4021_cast_fp16")]; + fp16 var_39419_to_fp16 = const()[name = string("op_39419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4023_cast_fp16, y = var_39419_to_fp16)[name = string("aw_chunk_4023_cast_fp16")]; + fp16 var_39421_to_fp16 = const()[name = string("op_39421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4025_cast_fp16, y = var_39421_to_fp16)[name = string("aw_chunk_4025_cast_fp16")]; + fp16 var_39423_to_fp16 = const()[name = string("op_39423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4027_cast_fp16, y = var_39423_to_fp16)[name = string("aw_chunk_4027_cast_fp16")]; + fp16 var_39425_to_fp16 = const()[name = string("op_39425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4029_cast_fp16, y = var_39425_to_fp16)[name = string("aw_chunk_4029_cast_fp16")]; + fp16 var_39427_to_fp16 = const()[name = string("op_39427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4031_cast_fp16, y = var_39427_to_fp16)[name = string("aw_chunk_4031_cast_fp16")]; + fp16 var_39429_to_fp16 = const()[name = string("op_39429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4033_cast_fp16, y = var_39429_to_fp16)[name = string("aw_chunk_4033_cast_fp16")]; + fp16 var_39431_to_fp16 = const()[name = string("op_39431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4035_cast_fp16, y = var_39431_to_fp16)[name = string("aw_chunk_4035_cast_fp16")]; + fp16 var_39433_to_fp16 = const()[name = string("op_39433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4037_cast_fp16, y = var_39433_to_fp16)[name = string("aw_chunk_4037_cast_fp16")]; + fp16 var_39435_to_fp16 = const()[name = string("op_39435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4039_cast_fp16, y = var_39435_to_fp16)[name = string("aw_chunk_4039_cast_fp16")]; + fp16 var_39437_to_fp16 = const()[name = string("op_39437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4041_cast_fp16, y = var_39437_to_fp16)[name = string("aw_chunk_4041_cast_fp16")]; + fp16 var_39439_to_fp16 = const()[name = string("op_39439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4043_cast_fp16, y = var_39439_to_fp16)[name = string("aw_chunk_4043_cast_fp16")]; + fp16 var_39441_to_fp16 = const()[name = string("op_39441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4045_cast_fp16, y = var_39441_to_fp16)[name = string("aw_chunk_4045_cast_fp16")]; + fp16 var_39443_to_fp16 = const()[name = string("op_39443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4047_cast_fp16, y = var_39443_to_fp16)[name = string("aw_chunk_4047_cast_fp16")]; + fp16 var_39445_to_fp16 = const()[name = string("op_39445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4049_cast_fp16, y = var_39445_to_fp16)[name = string("aw_chunk_4049_cast_fp16")]; + fp16 var_39447_to_fp16 = const()[name = string("op_39447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4051_cast_fp16, y = var_39447_to_fp16)[name = string("aw_chunk_4051_cast_fp16")]; + fp16 var_39449_to_fp16 = const()[name = string("op_39449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4053_cast_fp16, y = var_39449_to_fp16)[name = string("aw_chunk_4053_cast_fp16")]; + fp16 var_39451_to_fp16 = const()[name = string("op_39451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4055_cast_fp16, y = var_39451_to_fp16)[name = string("aw_chunk_4055_cast_fp16")]; + fp16 var_39453_to_fp16 = const()[name = string("op_39453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4057_cast_fp16, y = var_39453_to_fp16)[name = string("aw_chunk_4057_cast_fp16")]; + fp16 var_39455_to_fp16 = const()[name = string("op_39455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4059_cast_fp16, y = var_39455_to_fp16)[name = string("aw_chunk_4059_cast_fp16")]; + fp16 var_39457_to_fp16 = const()[name = string("op_39457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4061_cast_fp16, y = var_39457_to_fp16)[name = string("aw_chunk_4061_cast_fp16")]; + fp16 var_39459_to_fp16 = const()[name = string("op_39459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4063_cast_fp16, y = var_39459_to_fp16)[name = string("aw_chunk_4063_cast_fp16")]; + fp16 var_39461_to_fp16 = const()[name = string("op_39461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4065_cast_fp16, y = var_39461_to_fp16)[name = string("aw_chunk_4065_cast_fp16")]; + fp16 var_39463_to_fp16 = const()[name = string("op_39463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4067_cast_fp16, y = var_39463_to_fp16)[name = string("aw_chunk_4067_cast_fp16")]; + fp16 var_39465_to_fp16 = const()[name = string("op_39465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4069_cast_fp16, y = var_39465_to_fp16)[name = string("aw_chunk_4069_cast_fp16")]; + fp16 var_39467_to_fp16 = const()[name = string("op_39467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4071_cast_fp16, y = var_39467_to_fp16)[name = string("aw_chunk_4071_cast_fp16")]; + fp16 var_39469_to_fp16 = const()[name = string("op_39469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4073_cast_fp16, y = var_39469_to_fp16)[name = string("aw_chunk_4073_cast_fp16")]; + fp16 var_39471_to_fp16 = const()[name = string("op_39471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4075_cast_fp16, y = var_39471_to_fp16)[name = string("aw_chunk_4075_cast_fp16")]; + fp16 var_39473_to_fp16 = const()[name = string("op_39473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4077_cast_fp16, y = var_39473_to_fp16)[name = string("aw_chunk_4077_cast_fp16")]; + fp16 var_39475_to_fp16 = const()[name = string("op_39475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4079_cast_fp16, y = var_39475_to_fp16)[name = string("aw_chunk_4079_cast_fp16")]; + fp16 var_39477_to_fp16 = const()[name = string("op_39477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4081_cast_fp16, y = var_39477_to_fp16)[name = string("aw_chunk_4081_cast_fp16")]; + fp16 var_39479_to_fp16 = const()[name = string("op_39479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4083_cast_fp16, y = var_39479_to_fp16)[name = string("aw_chunk_4083_cast_fp16")]; + fp16 var_39481_to_fp16 = const()[name = string("op_39481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4085_cast_fp16, y = var_39481_to_fp16)[name = string("aw_chunk_4085_cast_fp16")]; + fp16 var_39483_to_fp16 = const()[name = string("op_39483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4087_cast_fp16, y = var_39483_to_fp16)[name = string("aw_chunk_4087_cast_fp16")]; + fp16 var_39485_to_fp16 = const()[name = string("op_39485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4089_cast_fp16, y = var_39485_to_fp16)[name = string("aw_chunk_4089_cast_fp16")]; + fp16 var_39487_to_fp16 = const()[name = string("op_39487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4091_cast_fp16, y = var_39487_to_fp16)[name = string("aw_chunk_4091_cast_fp16")]; + fp16 var_39489_to_fp16 = const()[name = string("op_39489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4093_cast_fp16, y = var_39489_to_fp16)[name = string("aw_chunk_4093_cast_fp16")]; + fp16 var_39491_to_fp16 = const()[name = string("op_39491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4095_cast_fp16, y = var_39491_to_fp16)[name = string("aw_chunk_4095_cast_fp16")]; + fp16 var_39493_to_fp16 = const()[name = string("op_39493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4097_cast_fp16, y = var_39493_to_fp16)[name = string("aw_chunk_4097_cast_fp16")]; + fp16 var_39495_to_fp16 = const()[name = string("op_39495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4099_cast_fp16, y = var_39495_to_fp16)[name = string("aw_chunk_4099_cast_fp16")]; + fp16 var_39497_to_fp16 = const()[name = string("op_39497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4101_cast_fp16, y = var_39497_to_fp16)[name = string("aw_chunk_4101_cast_fp16")]; + fp16 var_39499_to_fp16 = const()[name = string("op_39499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4103_cast_fp16, y = var_39499_to_fp16)[name = string("aw_chunk_4103_cast_fp16")]; + fp16 var_39501_to_fp16 = const()[name = string("op_39501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4105_cast_fp16, y = var_39501_to_fp16)[name = string("aw_chunk_4105_cast_fp16")]; + fp16 var_39503_to_fp16 = const()[name = string("op_39503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4107_cast_fp16, y = var_39503_to_fp16)[name = string("aw_chunk_4107_cast_fp16")]; + fp16 var_39505_to_fp16 = const()[name = string("op_39505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4109_cast_fp16, y = var_39505_to_fp16)[name = string("aw_chunk_4109_cast_fp16")]; + fp16 var_39507_to_fp16 = const()[name = string("op_39507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4111_cast_fp16, y = var_39507_to_fp16)[name = string("aw_chunk_4111_cast_fp16")]; + fp16 var_39509_to_fp16 = const()[name = string("op_39509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4113_cast_fp16, y = var_39509_to_fp16)[name = string("aw_chunk_4113_cast_fp16")]; + fp16 var_39511_to_fp16 = const()[name = string("op_39511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4115_cast_fp16, y = var_39511_to_fp16)[name = string("aw_chunk_4115_cast_fp16")]; + fp16 var_39513_to_fp16 = const()[name = string("op_39513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4117_cast_fp16, y = var_39513_to_fp16)[name = string("aw_chunk_4117_cast_fp16")]; + fp16 var_39515_to_fp16 = const()[name = string("op_39515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4119_cast_fp16, y = var_39515_to_fp16)[name = string("aw_chunk_4119_cast_fp16")]; + fp16 var_39517_to_fp16 = const()[name = string("op_39517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4121_cast_fp16, y = var_39517_to_fp16)[name = string("aw_chunk_4121_cast_fp16")]; + fp16 var_39519_to_fp16 = const()[name = string("op_39519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4123_cast_fp16, y = var_39519_to_fp16)[name = string("aw_chunk_4123_cast_fp16")]; + fp16 var_39521_to_fp16 = const()[name = string("op_39521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4125_cast_fp16, y = var_39521_to_fp16)[name = string("aw_chunk_4125_cast_fp16")]; + fp16 var_39523_to_fp16 = const()[name = string("op_39523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4127_cast_fp16, y = var_39523_to_fp16)[name = string("aw_chunk_4127_cast_fp16")]; + fp16 var_39525_to_fp16 = const()[name = string("op_39525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4129_cast_fp16, y = var_39525_to_fp16)[name = string("aw_chunk_4129_cast_fp16")]; + fp16 var_39527_to_fp16 = const()[name = string("op_39527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4131_cast_fp16, y = var_39527_to_fp16)[name = string("aw_chunk_4131_cast_fp16")]; + fp16 var_39529_to_fp16 = const()[name = string("op_39529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4133_cast_fp16, y = var_39529_to_fp16)[name = string("aw_chunk_4133_cast_fp16")]; + fp16 var_39531_to_fp16 = const()[name = string("op_39531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4135_cast_fp16, y = var_39531_to_fp16)[name = string("aw_chunk_4135_cast_fp16")]; + fp16 var_39533_to_fp16 = const()[name = string("op_39533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4137_cast_fp16, y = var_39533_to_fp16)[name = string("aw_chunk_4137_cast_fp16")]; + fp16 var_39535_to_fp16 = const()[name = string("op_39535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4139_cast_fp16, y = var_39535_to_fp16)[name = string("aw_chunk_4139_cast_fp16")]; + fp16 var_39537_to_fp16 = const()[name = string("op_39537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4141_cast_fp16, y = var_39537_to_fp16)[name = string("aw_chunk_4141_cast_fp16")]; + fp16 var_39539_to_fp16 = const()[name = string("op_39539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4143_cast_fp16, y = var_39539_to_fp16)[name = string("aw_chunk_4143_cast_fp16")]; + fp16 var_39541_to_fp16 = const()[name = string("op_39541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4145_cast_fp16, y = var_39541_to_fp16)[name = string("aw_chunk_4145_cast_fp16")]; + fp16 var_39543_to_fp16 = const()[name = string("op_39543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4147_cast_fp16, y = var_39543_to_fp16)[name = string("aw_chunk_4147_cast_fp16")]; + fp16 var_39545_to_fp16 = const()[name = string("op_39545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4149_cast_fp16, y = var_39545_to_fp16)[name = string("aw_chunk_4149_cast_fp16")]; + fp16 var_39547_to_fp16 = const()[name = string("op_39547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4151_cast_fp16, y = var_39547_to_fp16)[name = string("aw_chunk_4151_cast_fp16")]; + fp16 var_39549_to_fp16 = const()[name = string("op_39549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4153_cast_fp16, y = var_39549_to_fp16)[name = string("aw_chunk_4153_cast_fp16")]; + fp16 var_39551_to_fp16 = const()[name = string("op_39551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4155_cast_fp16, y = var_39551_to_fp16)[name = string("aw_chunk_4155_cast_fp16")]; + fp16 var_39553_to_fp16 = const()[name = string("op_39553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4157_cast_fp16, y = var_39553_to_fp16)[name = string("aw_chunk_4157_cast_fp16")]; + fp16 var_39555_to_fp16 = const()[name = string("op_39555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4159_cast_fp16, y = var_39555_to_fp16)[name = string("aw_chunk_4159_cast_fp16")]; + tensor var_39557_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4001_cast_fp16)[name = string("op_39557_cast_fp16")]; + tensor var_39558_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4003_cast_fp16)[name = string("op_39558_cast_fp16")]; + tensor var_39559_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4005_cast_fp16)[name = string("op_39559_cast_fp16")]; + tensor var_39560_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4007_cast_fp16)[name = string("op_39560_cast_fp16")]; + tensor var_39561_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4009_cast_fp16)[name = string("op_39561_cast_fp16")]; + tensor var_39562_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4011_cast_fp16)[name = string("op_39562_cast_fp16")]; + tensor var_39563_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4013_cast_fp16)[name = string("op_39563_cast_fp16")]; + tensor var_39564_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4015_cast_fp16)[name = string("op_39564_cast_fp16")]; + tensor var_39565_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4017_cast_fp16)[name = string("op_39565_cast_fp16")]; + tensor var_39566_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4019_cast_fp16)[name = string("op_39566_cast_fp16")]; + tensor var_39567_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4021_cast_fp16)[name = string("op_39567_cast_fp16")]; + tensor var_39568_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4023_cast_fp16)[name = string("op_39568_cast_fp16")]; + tensor var_39569_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4025_cast_fp16)[name = string("op_39569_cast_fp16")]; + tensor var_39570_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4027_cast_fp16)[name = string("op_39570_cast_fp16")]; + tensor var_39571_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4029_cast_fp16)[name = string("op_39571_cast_fp16")]; + tensor var_39572_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4031_cast_fp16)[name = string("op_39572_cast_fp16")]; + tensor var_39573_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4033_cast_fp16)[name = string("op_39573_cast_fp16")]; + tensor var_39574_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4035_cast_fp16)[name = string("op_39574_cast_fp16")]; + tensor var_39575_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4037_cast_fp16)[name = string("op_39575_cast_fp16")]; + tensor var_39576_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4039_cast_fp16)[name = string("op_39576_cast_fp16")]; + tensor var_39577_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4041_cast_fp16)[name = string("op_39577_cast_fp16")]; + tensor var_39578_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4043_cast_fp16)[name = string("op_39578_cast_fp16")]; + tensor var_39579_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4045_cast_fp16)[name = string("op_39579_cast_fp16")]; + tensor var_39580_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4047_cast_fp16)[name = string("op_39580_cast_fp16")]; + tensor var_39581_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4049_cast_fp16)[name = string("op_39581_cast_fp16")]; + tensor var_39582_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4051_cast_fp16)[name = string("op_39582_cast_fp16")]; + tensor var_39583_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4053_cast_fp16)[name = string("op_39583_cast_fp16")]; + tensor var_39584_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4055_cast_fp16)[name = string("op_39584_cast_fp16")]; + tensor var_39585_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4057_cast_fp16)[name = string("op_39585_cast_fp16")]; + tensor var_39586_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4059_cast_fp16)[name = string("op_39586_cast_fp16")]; + tensor var_39587_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4061_cast_fp16)[name = string("op_39587_cast_fp16")]; + tensor var_39588_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4063_cast_fp16)[name = string("op_39588_cast_fp16")]; + tensor var_39589_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4065_cast_fp16)[name = string("op_39589_cast_fp16")]; + tensor var_39590_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4067_cast_fp16)[name = string("op_39590_cast_fp16")]; + tensor var_39591_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4069_cast_fp16)[name = string("op_39591_cast_fp16")]; + tensor var_39592_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4071_cast_fp16)[name = string("op_39592_cast_fp16")]; + tensor var_39593_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4073_cast_fp16)[name = string("op_39593_cast_fp16")]; + tensor var_39594_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4075_cast_fp16)[name = string("op_39594_cast_fp16")]; + tensor var_39595_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4077_cast_fp16)[name = string("op_39595_cast_fp16")]; + tensor var_39596_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4079_cast_fp16)[name = string("op_39596_cast_fp16")]; + tensor var_39597_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4081_cast_fp16)[name = string("op_39597_cast_fp16")]; + tensor var_39598_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4083_cast_fp16)[name = string("op_39598_cast_fp16")]; + tensor var_39599_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4085_cast_fp16)[name = string("op_39599_cast_fp16")]; + tensor var_39600_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4087_cast_fp16)[name = string("op_39600_cast_fp16")]; + tensor var_39601_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4089_cast_fp16)[name = string("op_39601_cast_fp16")]; + tensor var_39602_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4091_cast_fp16)[name = string("op_39602_cast_fp16")]; + tensor var_39603_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4093_cast_fp16)[name = string("op_39603_cast_fp16")]; + tensor var_39604_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4095_cast_fp16)[name = string("op_39604_cast_fp16")]; + tensor var_39605_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4097_cast_fp16)[name = string("op_39605_cast_fp16")]; + tensor var_39606_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4099_cast_fp16)[name = string("op_39606_cast_fp16")]; + tensor var_39607_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4101_cast_fp16)[name = string("op_39607_cast_fp16")]; + tensor var_39608_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4103_cast_fp16)[name = string("op_39608_cast_fp16")]; + tensor var_39609_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4105_cast_fp16)[name = string("op_39609_cast_fp16")]; + tensor var_39610_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4107_cast_fp16)[name = string("op_39610_cast_fp16")]; + tensor var_39611_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4109_cast_fp16)[name = string("op_39611_cast_fp16")]; + tensor var_39612_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4111_cast_fp16)[name = string("op_39612_cast_fp16")]; + tensor var_39613_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4113_cast_fp16)[name = string("op_39613_cast_fp16")]; + tensor var_39614_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4115_cast_fp16)[name = string("op_39614_cast_fp16")]; + tensor var_39615_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4117_cast_fp16)[name = string("op_39615_cast_fp16")]; + tensor var_39616_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4119_cast_fp16)[name = string("op_39616_cast_fp16")]; + tensor var_39617_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4121_cast_fp16)[name = string("op_39617_cast_fp16")]; + tensor var_39618_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4123_cast_fp16)[name = string("op_39618_cast_fp16")]; + tensor var_39619_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4125_cast_fp16)[name = string("op_39619_cast_fp16")]; + tensor var_39620_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4127_cast_fp16)[name = string("op_39620_cast_fp16")]; + tensor var_39621_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4129_cast_fp16)[name = string("op_39621_cast_fp16")]; + tensor var_39622_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4131_cast_fp16)[name = string("op_39622_cast_fp16")]; + tensor var_39623_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4133_cast_fp16)[name = string("op_39623_cast_fp16")]; + tensor var_39624_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4135_cast_fp16)[name = string("op_39624_cast_fp16")]; + tensor var_39625_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4137_cast_fp16)[name = string("op_39625_cast_fp16")]; + tensor var_39626_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4139_cast_fp16)[name = string("op_39626_cast_fp16")]; + tensor var_39627_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4141_cast_fp16)[name = string("op_39627_cast_fp16")]; + tensor var_39628_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4143_cast_fp16)[name = string("op_39628_cast_fp16")]; + tensor var_39629_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4145_cast_fp16)[name = string("op_39629_cast_fp16")]; + tensor var_39630_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4147_cast_fp16)[name = string("op_39630_cast_fp16")]; + tensor var_39631_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4149_cast_fp16)[name = string("op_39631_cast_fp16")]; + tensor var_39632_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4151_cast_fp16)[name = string("op_39632_cast_fp16")]; + tensor var_39633_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4153_cast_fp16)[name = string("op_39633_cast_fp16")]; + tensor var_39634_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4155_cast_fp16)[name = string("op_39634_cast_fp16")]; + tensor var_39635_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4157_cast_fp16)[name = string("op_39635_cast_fp16")]; + tensor var_39636_cast_fp16 = softmax(axis = var_38382, x = aw_chunk_4159_cast_fp16)[name = string("op_39636_cast_fp16")]; + string var_39638_equation_0 = const()[name = string("op_39638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39638_cast_fp16 = einsum(equation = var_39638_equation_0, values = (var_39158_cast_fp16, var_39557_cast_fp16))[name = string("op_39638_cast_fp16")]; + string var_39640_equation_0 = const()[name = string("op_39640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39640_cast_fp16 = einsum(equation = var_39640_equation_0, values = (var_39158_cast_fp16, var_39558_cast_fp16))[name = string("op_39640_cast_fp16")]; + string var_39642_equation_0 = const()[name = string("op_39642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39642_cast_fp16 = einsum(equation = var_39642_equation_0, values = (var_39158_cast_fp16, var_39559_cast_fp16))[name = string("op_39642_cast_fp16")]; + string var_39644_equation_0 = const()[name = string("op_39644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39644_cast_fp16 = einsum(equation = var_39644_equation_0, values = (var_39158_cast_fp16, var_39560_cast_fp16))[name = string("op_39644_cast_fp16")]; + string var_39646_equation_0 = const()[name = string("op_39646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39646_cast_fp16 = einsum(equation = var_39646_equation_0, values = (var_39162_cast_fp16, var_39561_cast_fp16))[name = string("op_39646_cast_fp16")]; + string var_39648_equation_0 = const()[name = string("op_39648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39648_cast_fp16 = einsum(equation = var_39648_equation_0, values = (var_39162_cast_fp16, var_39562_cast_fp16))[name = string("op_39648_cast_fp16")]; + string var_39650_equation_0 = const()[name = string("op_39650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39650_cast_fp16 = einsum(equation = var_39650_equation_0, values = (var_39162_cast_fp16, var_39563_cast_fp16))[name = string("op_39650_cast_fp16")]; + string var_39652_equation_0 = const()[name = string("op_39652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39652_cast_fp16 = einsum(equation = var_39652_equation_0, values = (var_39162_cast_fp16, var_39564_cast_fp16))[name = string("op_39652_cast_fp16")]; + string var_39654_equation_0 = const()[name = string("op_39654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39654_cast_fp16 = einsum(equation = var_39654_equation_0, values = (var_39166_cast_fp16, var_39565_cast_fp16))[name = string("op_39654_cast_fp16")]; + string var_39656_equation_0 = const()[name = string("op_39656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39656_cast_fp16 = einsum(equation = var_39656_equation_0, values = (var_39166_cast_fp16, var_39566_cast_fp16))[name = string("op_39656_cast_fp16")]; + string var_39658_equation_0 = const()[name = string("op_39658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39658_cast_fp16 = einsum(equation = var_39658_equation_0, values = (var_39166_cast_fp16, var_39567_cast_fp16))[name = string("op_39658_cast_fp16")]; + string var_39660_equation_0 = const()[name = string("op_39660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39660_cast_fp16 = einsum(equation = var_39660_equation_0, values = (var_39166_cast_fp16, var_39568_cast_fp16))[name = string("op_39660_cast_fp16")]; + string var_39662_equation_0 = const()[name = string("op_39662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39662_cast_fp16 = einsum(equation = var_39662_equation_0, values = (var_39170_cast_fp16, var_39569_cast_fp16))[name = string("op_39662_cast_fp16")]; + string var_39664_equation_0 = const()[name = string("op_39664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39664_cast_fp16 = einsum(equation = var_39664_equation_0, values = (var_39170_cast_fp16, var_39570_cast_fp16))[name = string("op_39664_cast_fp16")]; + string var_39666_equation_0 = const()[name = string("op_39666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39666_cast_fp16 = einsum(equation = var_39666_equation_0, values = (var_39170_cast_fp16, var_39571_cast_fp16))[name = string("op_39666_cast_fp16")]; + string var_39668_equation_0 = const()[name = string("op_39668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39668_cast_fp16 = einsum(equation = var_39668_equation_0, values = (var_39170_cast_fp16, var_39572_cast_fp16))[name = string("op_39668_cast_fp16")]; + string var_39670_equation_0 = const()[name = string("op_39670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39670_cast_fp16 = einsum(equation = var_39670_equation_0, values = (var_39174_cast_fp16, var_39573_cast_fp16))[name = string("op_39670_cast_fp16")]; + string var_39672_equation_0 = const()[name = string("op_39672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39672_cast_fp16 = einsum(equation = var_39672_equation_0, values = (var_39174_cast_fp16, var_39574_cast_fp16))[name = string("op_39672_cast_fp16")]; + string var_39674_equation_0 = const()[name = string("op_39674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39674_cast_fp16 = einsum(equation = var_39674_equation_0, values = (var_39174_cast_fp16, var_39575_cast_fp16))[name = string("op_39674_cast_fp16")]; + string var_39676_equation_0 = const()[name = string("op_39676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39676_cast_fp16 = einsum(equation = var_39676_equation_0, values = (var_39174_cast_fp16, var_39576_cast_fp16))[name = string("op_39676_cast_fp16")]; + string var_39678_equation_0 = const()[name = string("op_39678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39678_cast_fp16 = einsum(equation = var_39678_equation_0, values = (var_39178_cast_fp16, var_39577_cast_fp16))[name = string("op_39678_cast_fp16")]; + string var_39680_equation_0 = const()[name = string("op_39680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39680_cast_fp16 = einsum(equation = var_39680_equation_0, values = (var_39178_cast_fp16, var_39578_cast_fp16))[name = string("op_39680_cast_fp16")]; + string var_39682_equation_0 = const()[name = string("op_39682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39682_cast_fp16 = einsum(equation = var_39682_equation_0, values = (var_39178_cast_fp16, var_39579_cast_fp16))[name = string("op_39682_cast_fp16")]; + string var_39684_equation_0 = const()[name = string("op_39684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39684_cast_fp16 = einsum(equation = var_39684_equation_0, values = (var_39178_cast_fp16, var_39580_cast_fp16))[name = string("op_39684_cast_fp16")]; + string var_39686_equation_0 = const()[name = string("op_39686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39686_cast_fp16 = einsum(equation = var_39686_equation_0, values = (var_39182_cast_fp16, var_39581_cast_fp16))[name = string("op_39686_cast_fp16")]; + string var_39688_equation_0 = const()[name = string("op_39688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39688_cast_fp16 = einsum(equation = var_39688_equation_0, values = (var_39182_cast_fp16, var_39582_cast_fp16))[name = string("op_39688_cast_fp16")]; + string var_39690_equation_0 = const()[name = string("op_39690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39690_cast_fp16 = einsum(equation = var_39690_equation_0, values = (var_39182_cast_fp16, var_39583_cast_fp16))[name = string("op_39690_cast_fp16")]; + string var_39692_equation_0 = const()[name = string("op_39692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39692_cast_fp16 = einsum(equation = var_39692_equation_0, values = (var_39182_cast_fp16, var_39584_cast_fp16))[name = string("op_39692_cast_fp16")]; + string var_39694_equation_0 = const()[name = string("op_39694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39694_cast_fp16 = einsum(equation = var_39694_equation_0, values = (var_39186_cast_fp16, var_39585_cast_fp16))[name = string("op_39694_cast_fp16")]; + string var_39696_equation_0 = const()[name = string("op_39696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39696_cast_fp16 = einsum(equation = var_39696_equation_0, values = (var_39186_cast_fp16, var_39586_cast_fp16))[name = string("op_39696_cast_fp16")]; + string var_39698_equation_0 = const()[name = string("op_39698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39698_cast_fp16 = einsum(equation = var_39698_equation_0, values = (var_39186_cast_fp16, var_39587_cast_fp16))[name = string("op_39698_cast_fp16")]; + string var_39700_equation_0 = const()[name = string("op_39700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39700_cast_fp16 = einsum(equation = var_39700_equation_0, values = (var_39186_cast_fp16, var_39588_cast_fp16))[name = string("op_39700_cast_fp16")]; + string var_39702_equation_0 = const()[name = string("op_39702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39702_cast_fp16 = einsum(equation = var_39702_equation_0, values = (var_39190_cast_fp16, var_39589_cast_fp16))[name = string("op_39702_cast_fp16")]; + string var_39704_equation_0 = const()[name = string("op_39704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39704_cast_fp16 = einsum(equation = var_39704_equation_0, values = (var_39190_cast_fp16, var_39590_cast_fp16))[name = string("op_39704_cast_fp16")]; + string var_39706_equation_0 = const()[name = string("op_39706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39706_cast_fp16 = einsum(equation = var_39706_equation_0, values = (var_39190_cast_fp16, var_39591_cast_fp16))[name = string("op_39706_cast_fp16")]; + string var_39708_equation_0 = const()[name = string("op_39708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39708_cast_fp16 = einsum(equation = var_39708_equation_0, values = (var_39190_cast_fp16, var_39592_cast_fp16))[name = string("op_39708_cast_fp16")]; + string var_39710_equation_0 = const()[name = string("op_39710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39710_cast_fp16 = einsum(equation = var_39710_equation_0, values = (var_39194_cast_fp16, var_39593_cast_fp16))[name = string("op_39710_cast_fp16")]; + string var_39712_equation_0 = const()[name = string("op_39712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39712_cast_fp16 = einsum(equation = var_39712_equation_0, values = (var_39194_cast_fp16, var_39594_cast_fp16))[name = string("op_39712_cast_fp16")]; + string var_39714_equation_0 = const()[name = string("op_39714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39714_cast_fp16 = einsum(equation = var_39714_equation_0, values = (var_39194_cast_fp16, var_39595_cast_fp16))[name = string("op_39714_cast_fp16")]; + string var_39716_equation_0 = const()[name = string("op_39716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39716_cast_fp16 = einsum(equation = var_39716_equation_0, values = (var_39194_cast_fp16, var_39596_cast_fp16))[name = string("op_39716_cast_fp16")]; + string var_39718_equation_0 = const()[name = string("op_39718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39718_cast_fp16 = einsum(equation = var_39718_equation_0, values = (var_39198_cast_fp16, var_39597_cast_fp16))[name = string("op_39718_cast_fp16")]; + string var_39720_equation_0 = const()[name = string("op_39720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39720_cast_fp16 = einsum(equation = var_39720_equation_0, values = (var_39198_cast_fp16, var_39598_cast_fp16))[name = string("op_39720_cast_fp16")]; + string var_39722_equation_0 = const()[name = string("op_39722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39722_cast_fp16 = einsum(equation = var_39722_equation_0, values = (var_39198_cast_fp16, var_39599_cast_fp16))[name = string("op_39722_cast_fp16")]; + string var_39724_equation_0 = const()[name = string("op_39724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39724_cast_fp16 = einsum(equation = var_39724_equation_0, values = (var_39198_cast_fp16, var_39600_cast_fp16))[name = string("op_39724_cast_fp16")]; + string var_39726_equation_0 = const()[name = string("op_39726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39726_cast_fp16 = einsum(equation = var_39726_equation_0, values = (var_39202_cast_fp16, var_39601_cast_fp16))[name = string("op_39726_cast_fp16")]; + string var_39728_equation_0 = const()[name = string("op_39728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39728_cast_fp16 = einsum(equation = var_39728_equation_0, values = (var_39202_cast_fp16, var_39602_cast_fp16))[name = string("op_39728_cast_fp16")]; + string var_39730_equation_0 = const()[name = string("op_39730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39730_cast_fp16 = einsum(equation = var_39730_equation_0, values = (var_39202_cast_fp16, var_39603_cast_fp16))[name = string("op_39730_cast_fp16")]; + string var_39732_equation_0 = const()[name = string("op_39732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39732_cast_fp16 = einsum(equation = var_39732_equation_0, values = (var_39202_cast_fp16, var_39604_cast_fp16))[name = string("op_39732_cast_fp16")]; + string var_39734_equation_0 = const()[name = string("op_39734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39734_cast_fp16 = einsum(equation = var_39734_equation_0, values = (var_39206_cast_fp16, var_39605_cast_fp16))[name = string("op_39734_cast_fp16")]; + string var_39736_equation_0 = const()[name = string("op_39736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39736_cast_fp16 = einsum(equation = var_39736_equation_0, values = (var_39206_cast_fp16, var_39606_cast_fp16))[name = string("op_39736_cast_fp16")]; + string var_39738_equation_0 = const()[name = string("op_39738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39738_cast_fp16 = einsum(equation = var_39738_equation_0, values = (var_39206_cast_fp16, var_39607_cast_fp16))[name = string("op_39738_cast_fp16")]; + string var_39740_equation_0 = const()[name = string("op_39740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39740_cast_fp16 = einsum(equation = var_39740_equation_0, values = (var_39206_cast_fp16, var_39608_cast_fp16))[name = string("op_39740_cast_fp16")]; + string var_39742_equation_0 = const()[name = string("op_39742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39742_cast_fp16 = einsum(equation = var_39742_equation_0, values = (var_39210_cast_fp16, var_39609_cast_fp16))[name = string("op_39742_cast_fp16")]; + string var_39744_equation_0 = const()[name = string("op_39744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39744_cast_fp16 = einsum(equation = var_39744_equation_0, values = (var_39210_cast_fp16, var_39610_cast_fp16))[name = string("op_39744_cast_fp16")]; + string var_39746_equation_0 = const()[name = string("op_39746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39746_cast_fp16 = einsum(equation = var_39746_equation_0, values = (var_39210_cast_fp16, var_39611_cast_fp16))[name = string("op_39746_cast_fp16")]; + string var_39748_equation_0 = const()[name = string("op_39748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39748_cast_fp16 = einsum(equation = var_39748_equation_0, values = (var_39210_cast_fp16, var_39612_cast_fp16))[name = string("op_39748_cast_fp16")]; + string var_39750_equation_0 = const()[name = string("op_39750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39750_cast_fp16 = einsum(equation = var_39750_equation_0, values = (var_39214_cast_fp16, var_39613_cast_fp16))[name = string("op_39750_cast_fp16")]; + string var_39752_equation_0 = const()[name = string("op_39752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39752_cast_fp16 = einsum(equation = var_39752_equation_0, values = (var_39214_cast_fp16, var_39614_cast_fp16))[name = string("op_39752_cast_fp16")]; + string var_39754_equation_0 = const()[name = string("op_39754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39754_cast_fp16 = einsum(equation = var_39754_equation_0, values = (var_39214_cast_fp16, var_39615_cast_fp16))[name = string("op_39754_cast_fp16")]; + string var_39756_equation_0 = const()[name = string("op_39756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39756_cast_fp16 = einsum(equation = var_39756_equation_0, values = (var_39214_cast_fp16, var_39616_cast_fp16))[name = string("op_39756_cast_fp16")]; + string var_39758_equation_0 = const()[name = string("op_39758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39758_cast_fp16 = einsum(equation = var_39758_equation_0, values = (var_39218_cast_fp16, var_39617_cast_fp16))[name = string("op_39758_cast_fp16")]; + string var_39760_equation_0 = const()[name = string("op_39760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39760_cast_fp16 = einsum(equation = var_39760_equation_0, values = (var_39218_cast_fp16, var_39618_cast_fp16))[name = string("op_39760_cast_fp16")]; + string var_39762_equation_0 = const()[name = string("op_39762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39762_cast_fp16 = einsum(equation = var_39762_equation_0, values = (var_39218_cast_fp16, var_39619_cast_fp16))[name = string("op_39762_cast_fp16")]; + string var_39764_equation_0 = const()[name = string("op_39764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39764_cast_fp16 = einsum(equation = var_39764_equation_0, values = (var_39218_cast_fp16, var_39620_cast_fp16))[name = string("op_39764_cast_fp16")]; + string var_39766_equation_0 = const()[name = string("op_39766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39766_cast_fp16 = einsum(equation = var_39766_equation_0, values = (var_39222_cast_fp16, var_39621_cast_fp16))[name = string("op_39766_cast_fp16")]; + string var_39768_equation_0 = const()[name = string("op_39768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39768_cast_fp16 = einsum(equation = var_39768_equation_0, values = (var_39222_cast_fp16, var_39622_cast_fp16))[name = string("op_39768_cast_fp16")]; + string var_39770_equation_0 = const()[name = string("op_39770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39770_cast_fp16 = einsum(equation = var_39770_equation_0, values = (var_39222_cast_fp16, var_39623_cast_fp16))[name = string("op_39770_cast_fp16")]; + string var_39772_equation_0 = const()[name = string("op_39772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39772_cast_fp16 = einsum(equation = var_39772_equation_0, values = (var_39222_cast_fp16, var_39624_cast_fp16))[name = string("op_39772_cast_fp16")]; + string var_39774_equation_0 = const()[name = string("op_39774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39774_cast_fp16 = einsum(equation = var_39774_equation_0, values = (var_39226_cast_fp16, var_39625_cast_fp16))[name = string("op_39774_cast_fp16")]; + string var_39776_equation_0 = const()[name = string("op_39776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39776_cast_fp16 = einsum(equation = var_39776_equation_0, values = (var_39226_cast_fp16, var_39626_cast_fp16))[name = string("op_39776_cast_fp16")]; + string var_39778_equation_0 = const()[name = string("op_39778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39778_cast_fp16 = einsum(equation = var_39778_equation_0, values = (var_39226_cast_fp16, var_39627_cast_fp16))[name = string("op_39778_cast_fp16")]; + string var_39780_equation_0 = const()[name = string("op_39780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39780_cast_fp16 = einsum(equation = var_39780_equation_0, values = (var_39226_cast_fp16, var_39628_cast_fp16))[name = string("op_39780_cast_fp16")]; + string var_39782_equation_0 = const()[name = string("op_39782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39782_cast_fp16 = einsum(equation = var_39782_equation_0, values = (var_39230_cast_fp16, var_39629_cast_fp16))[name = string("op_39782_cast_fp16")]; + string var_39784_equation_0 = const()[name = string("op_39784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39784_cast_fp16 = einsum(equation = var_39784_equation_0, values = (var_39230_cast_fp16, var_39630_cast_fp16))[name = string("op_39784_cast_fp16")]; + string var_39786_equation_0 = const()[name = string("op_39786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39786_cast_fp16 = einsum(equation = var_39786_equation_0, values = (var_39230_cast_fp16, var_39631_cast_fp16))[name = string("op_39786_cast_fp16")]; + string var_39788_equation_0 = const()[name = string("op_39788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39788_cast_fp16 = einsum(equation = var_39788_equation_0, values = (var_39230_cast_fp16, var_39632_cast_fp16))[name = string("op_39788_cast_fp16")]; + string var_39790_equation_0 = const()[name = string("op_39790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39790_cast_fp16 = einsum(equation = var_39790_equation_0, values = (var_39234_cast_fp16, var_39633_cast_fp16))[name = string("op_39790_cast_fp16")]; + string var_39792_equation_0 = const()[name = string("op_39792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39792_cast_fp16 = einsum(equation = var_39792_equation_0, values = (var_39234_cast_fp16, var_39634_cast_fp16))[name = string("op_39792_cast_fp16")]; + string var_39794_equation_0 = const()[name = string("op_39794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39794_cast_fp16 = einsum(equation = var_39794_equation_0, values = (var_39234_cast_fp16, var_39635_cast_fp16))[name = string("op_39794_cast_fp16")]; + string var_39796_equation_0 = const()[name = string("op_39796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_39796_cast_fp16 = einsum(equation = var_39796_equation_0, values = (var_39234_cast_fp16, var_39636_cast_fp16))[name = string("op_39796_cast_fp16")]; + bool var_39798_interleave_0 = const()[name = string("op_39798_interleave_0"), val = bool(false)]; + tensor var_39798_cast_fp16 = concat(axis = var_38357, interleave = var_39798_interleave_0, values = (var_39638_cast_fp16, var_39640_cast_fp16, var_39642_cast_fp16, var_39644_cast_fp16))[name = string("op_39798_cast_fp16")]; + bool var_39800_interleave_0 = const()[name = string("op_39800_interleave_0"), val = bool(false)]; + tensor var_39800_cast_fp16 = concat(axis = var_38357, interleave = var_39800_interleave_0, values = (var_39646_cast_fp16, var_39648_cast_fp16, var_39650_cast_fp16, var_39652_cast_fp16))[name = string("op_39800_cast_fp16")]; + bool var_39802_interleave_0 = const()[name = string("op_39802_interleave_0"), val = bool(false)]; + tensor var_39802_cast_fp16 = concat(axis = var_38357, interleave = var_39802_interleave_0, values = (var_39654_cast_fp16, var_39656_cast_fp16, var_39658_cast_fp16, var_39660_cast_fp16))[name = string("op_39802_cast_fp16")]; + bool var_39804_interleave_0 = const()[name = string("op_39804_interleave_0"), val = bool(false)]; + tensor var_39804_cast_fp16 = concat(axis = var_38357, interleave = var_39804_interleave_0, values = (var_39662_cast_fp16, var_39664_cast_fp16, var_39666_cast_fp16, var_39668_cast_fp16))[name = string("op_39804_cast_fp16")]; + bool var_39806_interleave_0 = const()[name = string("op_39806_interleave_0"), val = bool(false)]; + tensor var_39806_cast_fp16 = concat(axis = var_38357, interleave = var_39806_interleave_0, values = (var_39670_cast_fp16, var_39672_cast_fp16, var_39674_cast_fp16, var_39676_cast_fp16))[name = string("op_39806_cast_fp16")]; + bool var_39808_interleave_0 = const()[name = string("op_39808_interleave_0"), val = bool(false)]; + tensor var_39808_cast_fp16 = concat(axis = var_38357, interleave = var_39808_interleave_0, values = (var_39678_cast_fp16, var_39680_cast_fp16, var_39682_cast_fp16, var_39684_cast_fp16))[name = string("op_39808_cast_fp16")]; + bool var_39810_interleave_0 = const()[name = string("op_39810_interleave_0"), val = bool(false)]; + tensor var_39810_cast_fp16 = concat(axis = var_38357, interleave = var_39810_interleave_0, values = (var_39686_cast_fp16, var_39688_cast_fp16, var_39690_cast_fp16, var_39692_cast_fp16))[name = string("op_39810_cast_fp16")]; + bool var_39812_interleave_0 = const()[name = string("op_39812_interleave_0"), val = bool(false)]; + tensor var_39812_cast_fp16 = concat(axis = var_38357, interleave = var_39812_interleave_0, values = (var_39694_cast_fp16, var_39696_cast_fp16, var_39698_cast_fp16, var_39700_cast_fp16))[name = string("op_39812_cast_fp16")]; + bool var_39814_interleave_0 = const()[name = string("op_39814_interleave_0"), val = bool(false)]; + tensor var_39814_cast_fp16 = concat(axis = var_38357, interleave = var_39814_interleave_0, values = (var_39702_cast_fp16, var_39704_cast_fp16, var_39706_cast_fp16, var_39708_cast_fp16))[name = string("op_39814_cast_fp16")]; + bool var_39816_interleave_0 = const()[name = string("op_39816_interleave_0"), val = bool(false)]; + tensor var_39816_cast_fp16 = concat(axis = var_38357, interleave = var_39816_interleave_0, values = (var_39710_cast_fp16, var_39712_cast_fp16, var_39714_cast_fp16, var_39716_cast_fp16))[name = string("op_39816_cast_fp16")]; + bool var_39818_interleave_0 = const()[name = string("op_39818_interleave_0"), val = bool(false)]; + tensor var_39818_cast_fp16 = concat(axis = var_38357, interleave = var_39818_interleave_0, values = (var_39718_cast_fp16, var_39720_cast_fp16, var_39722_cast_fp16, var_39724_cast_fp16))[name = string("op_39818_cast_fp16")]; + bool var_39820_interleave_0 = const()[name = string("op_39820_interleave_0"), val = bool(false)]; + tensor var_39820_cast_fp16 = concat(axis = var_38357, interleave = var_39820_interleave_0, values = (var_39726_cast_fp16, var_39728_cast_fp16, var_39730_cast_fp16, var_39732_cast_fp16))[name = string("op_39820_cast_fp16")]; + bool var_39822_interleave_0 = const()[name = string("op_39822_interleave_0"), val = bool(false)]; + tensor var_39822_cast_fp16 = concat(axis = var_38357, interleave = var_39822_interleave_0, values = (var_39734_cast_fp16, var_39736_cast_fp16, var_39738_cast_fp16, var_39740_cast_fp16))[name = string("op_39822_cast_fp16")]; + bool var_39824_interleave_0 = const()[name = string("op_39824_interleave_0"), val = bool(false)]; + tensor var_39824_cast_fp16 = concat(axis = var_38357, interleave = var_39824_interleave_0, values = (var_39742_cast_fp16, var_39744_cast_fp16, var_39746_cast_fp16, var_39748_cast_fp16))[name = string("op_39824_cast_fp16")]; + bool var_39826_interleave_0 = const()[name = string("op_39826_interleave_0"), val = bool(false)]; + tensor var_39826_cast_fp16 = concat(axis = var_38357, interleave = var_39826_interleave_0, values = (var_39750_cast_fp16, var_39752_cast_fp16, var_39754_cast_fp16, var_39756_cast_fp16))[name = string("op_39826_cast_fp16")]; + bool var_39828_interleave_0 = const()[name = string("op_39828_interleave_0"), val = bool(false)]; + tensor var_39828_cast_fp16 = concat(axis = var_38357, interleave = var_39828_interleave_0, values = (var_39758_cast_fp16, var_39760_cast_fp16, var_39762_cast_fp16, var_39764_cast_fp16))[name = string("op_39828_cast_fp16")]; + bool var_39830_interleave_0 = const()[name = string("op_39830_interleave_0"), val = bool(false)]; + tensor var_39830_cast_fp16 = concat(axis = var_38357, interleave = var_39830_interleave_0, values = (var_39766_cast_fp16, var_39768_cast_fp16, var_39770_cast_fp16, var_39772_cast_fp16))[name = string("op_39830_cast_fp16")]; + bool var_39832_interleave_0 = const()[name = string("op_39832_interleave_0"), val = bool(false)]; + tensor var_39832_cast_fp16 = concat(axis = var_38357, interleave = var_39832_interleave_0, values = (var_39774_cast_fp16, var_39776_cast_fp16, var_39778_cast_fp16, var_39780_cast_fp16))[name = string("op_39832_cast_fp16")]; + bool var_39834_interleave_0 = const()[name = string("op_39834_interleave_0"), val = bool(false)]; + tensor var_39834_cast_fp16 = concat(axis = var_38357, interleave = var_39834_interleave_0, values = (var_39782_cast_fp16, var_39784_cast_fp16, var_39786_cast_fp16, var_39788_cast_fp16))[name = string("op_39834_cast_fp16")]; + bool var_39836_interleave_0 = const()[name = string("op_39836_interleave_0"), val = bool(false)]; + tensor var_39836_cast_fp16 = concat(axis = var_38357, interleave = var_39836_interleave_0, values = (var_39790_cast_fp16, var_39792_cast_fp16, var_39794_cast_fp16, var_39796_cast_fp16))[name = string("op_39836_cast_fp16")]; + bool input_201_interleave_0 = const()[name = string("input_201_interleave_0"), val = bool(false)]; + tensor input_201_cast_fp16 = concat(axis = var_38382, interleave = input_201_interleave_0, values = (var_39798_cast_fp16, var_39800_cast_fp16, var_39802_cast_fp16, var_39804_cast_fp16, var_39806_cast_fp16, var_39808_cast_fp16, var_39810_cast_fp16, var_39812_cast_fp16, var_39814_cast_fp16, var_39816_cast_fp16, var_39818_cast_fp16, var_39820_cast_fp16, var_39822_cast_fp16, var_39824_cast_fp16, var_39826_cast_fp16, var_39828_cast_fp16, var_39830_cast_fp16, var_39832_cast_fp16, var_39834_cast_fp16, var_39836_cast_fp16))[name = string("input_201_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_25_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008337280)))]; + tensor layers_25_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011614144)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_25_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_25_self_attn_o_proj_weight_to_fp16, x = input_201_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_103_cast_fp16")]; + tensor out_103_axes_0 = const()[name = string("out_103_axes_0"), val = tensor([1])]; + fp16 var_39855_to_fp16 = const()[name = string("op_39855_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_39855_to_fp16, x = inputs_103_cast_fp16)[name = string("out_103_cast_fp16")]; + tensor input_203_gamma_0_to_fp16 = const()[name = string("input_203_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011616768)))]; + tensor input_203_beta_0_to_fp16 = const()[name = string("input_203_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011619392)))]; + fp16 input_203_epsilon_0_to_fp16 = const()[name = string("input_203_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_203_cast_fp16 = batch_norm(beta = input_203_beta_0_to_fp16, epsilon = input_203_epsilon_0_to_fp16, gamma = input_203_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = string("input_203_cast_fp16")]; + string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")]; + tensor input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor([1, 1])]; + tensor input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor([1, 1])]; + int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1)]; + tensor layers_25_fc1_weight_to_fp16 = const()[name = string("layers_25_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1011622016)))]; + tensor layers_25_fc1_bias_to_fp16 = const()[name = string("layers_25_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024729280)))]; + tensor input_205_cast_fp16 = conv(bias = layers_25_fc1_bias_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = layers_25_fc1_weight_to_fp16, x = input_203_cast_fp16)[name = string("input_205_cast_fp16")]; + string input_207_mode_0 = const()[name = string("input_207_mode_0"), val = string("EXACT")]; + tensor input_207_cast_fp16 = gelu(mode = input_207_mode_0, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")]; + string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")]; + tensor hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)]; + tensor layers_25_fc2_weight_to_fp16 = const()[name = string("layers_25_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1024739584)))]; + tensor layers_25_fc2_bias_to_fp16 = const()[name = string("layers_25_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037846848)))]; + tensor hidden_states_55_cast_fp16 = conv(bias = layers_25_fc2_bias_to_fp16, dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = layers_25_fc2_weight_to_fp16, x = input_207_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = hidden_states_55_cast_fp16)[name = string("inputs_105_cast_fp16")]; + int32 var_39884 = const()[name = string("op_39884"), val = int32(3)]; + int32 var_39909 = const()[name = string("op_39909"), val = int32(1)]; + tensor out_105_axes_0 = const()[name = string("out_105_axes_0"), val = tensor([1])]; + fp16 var_39926_to_fp16 = const()[name = string("op_39926_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_39926_to_fp16, x = inputs_105_cast_fp16)[name = string("out_105_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037849472)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037852096)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_53_pad_type_0 = const()[name = string("query_53_pad_type_0"), val = string("valid")]; + tensor query_53_strides_0 = const()[name = string("query_53_strides_0"), val = tensor([1, 1])]; + tensor query_53_pad_0 = const()[name = string("query_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_53_dilations_0 = const()[name = string("query_53_dilations_0"), val = tensor([1, 1])]; + int32 query_53_groups_0 = const()[name = string("query_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1037854720)))]; + tensor layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041131584)))]; + tensor query_53_cast_fp16 = conv(bias = layers_26_self_attn_q_proj_bias_to_fp16, dilations = query_53_dilations_0, groups = query_53_groups_0, pad = query_53_pad_0, pad_type = query_53_pad_type_0, strides = query_53_strides_0, weight = layers_26_self_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_53_cast_fp16")]; + string key_53_pad_type_0 = const()[name = string("key_53_pad_type_0"), val = string("valid")]; + tensor key_53_strides_0 = const()[name = string("key_53_strides_0"), val = tensor([1, 1])]; + tensor key_53_pad_0 = const()[name = string("key_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_53_dilations_0 = const()[name = string("key_53_dilations_0"), val = tensor([1, 1])]; + int32 key_53_groups_0 = const()[name = string("key_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1041134208)))]; + tensor key_53_cast_fp16 = conv(dilations = key_53_dilations_0, groups = key_53_groups_0, pad = key_53_pad_0, pad_type = key_53_pad_type_0, strides = key_53_strides_0, weight = layers_26_self_attn_k_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("key_53_cast_fp16")]; + string value_53_pad_type_0 = const()[name = string("value_53_pad_type_0"), val = string("valid")]; + tensor value_53_strides_0 = const()[name = string("value_53_strides_0"), val = tensor([1, 1])]; + tensor value_53_pad_0 = const()[name = string("value_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_53_dilations_0 = const()[name = string("value_53_dilations_0"), val = tensor([1, 1])]; + int32 value_53_groups_0 = const()[name = string("value_53_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1044411072)))]; + tensor layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047687936)))]; + tensor value_53_cast_fp16 = conv(bias = layers_26_self_attn_v_proj_bias_to_fp16, dilations = value_53_dilations_0, groups = value_53_groups_0, pad = value_53_pad_0, pad_type = value_53_pad_type_0, strides = value_53_strides_0, weight = layers_26_self_attn_v_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("value_53_cast_fp16")]; + tensor var_39964_begin_0 = const()[name = string("op_39964_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39964_end_0 = const()[name = string("op_39964_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39964_end_mask_0 = const()[name = string("op_39964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39964_cast_fp16 = slice_by_index(begin = var_39964_begin_0, end = var_39964_end_0, end_mask = var_39964_end_mask_0, x = query_53_cast_fp16)[name = string("op_39964_cast_fp16")]; + tensor var_39968_begin_0 = const()[name = string("op_39968_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39968_end_0 = const()[name = string("op_39968_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_39968_end_mask_0 = const()[name = string("op_39968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39968_cast_fp16 = slice_by_index(begin = var_39968_begin_0, end = var_39968_end_0, end_mask = var_39968_end_mask_0, x = query_53_cast_fp16)[name = string("op_39968_cast_fp16")]; + tensor var_39972_begin_0 = const()[name = string("op_39972_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39972_end_0 = const()[name = string("op_39972_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_39972_end_mask_0 = const()[name = string("op_39972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39972_cast_fp16 = slice_by_index(begin = var_39972_begin_0, end = var_39972_end_0, end_mask = var_39972_end_mask_0, x = query_53_cast_fp16)[name = string("op_39972_cast_fp16")]; + tensor var_39976_begin_0 = const()[name = string("op_39976_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39976_end_0 = const()[name = string("op_39976_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_39976_end_mask_0 = const()[name = string("op_39976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39976_cast_fp16 = slice_by_index(begin = var_39976_begin_0, end = var_39976_end_0, end_mask = var_39976_end_mask_0, x = query_53_cast_fp16)[name = string("op_39976_cast_fp16")]; + tensor var_39980_begin_0 = const()[name = string("op_39980_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39980_end_0 = const()[name = string("op_39980_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_39980_end_mask_0 = const()[name = string("op_39980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39980_cast_fp16 = slice_by_index(begin = var_39980_begin_0, end = var_39980_end_0, end_mask = var_39980_end_mask_0, x = query_53_cast_fp16)[name = string("op_39980_cast_fp16")]; + tensor var_39984_begin_0 = const()[name = string("op_39984_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39984_end_0 = const()[name = string("op_39984_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_39984_end_mask_0 = const()[name = string("op_39984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39984_cast_fp16 = slice_by_index(begin = var_39984_begin_0, end = var_39984_end_0, end_mask = var_39984_end_mask_0, x = query_53_cast_fp16)[name = string("op_39984_cast_fp16")]; + tensor var_39988_begin_0 = const()[name = string("op_39988_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39988_end_0 = const()[name = string("op_39988_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_39988_end_mask_0 = const()[name = string("op_39988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39988_cast_fp16 = slice_by_index(begin = var_39988_begin_0, end = var_39988_end_0, end_mask = var_39988_end_mask_0, x = query_53_cast_fp16)[name = string("op_39988_cast_fp16")]; + tensor var_39992_begin_0 = const()[name = string("op_39992_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39992_end_0 = const()[name = string("op_39992_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_39992_end_mask_0 = const()[name = string("op_39992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39992_cast_fp16 = slice_by_index(begin = var_39992_begin_0, end = var_39992_end_0, end_mask = var_39992_end_mask_0, x = query_53_cast_fp16)[name = string("op_39992_cast_fp16")]; + tensor var_39996_begin_0 = const()[name = string("op_39996_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39996_end_0 = const()[name = string("op_39996_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_39996_end_mask_0 = const()[name = string("op_39996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39996_cast_fp16 = slice_by_index(begin = var_39996_begin_0, end = var_39996_end_0, end_mask = var_39996_end_mask_0, x = query_53_cast_fp16)[name = string("op_39996_cast_fp16")]; + tensor var_40000_begin_0 = const()[name = string("op_40000_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40000_end_0 = const()[name = string("op_40000_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_40000_end_mask_0 = const()[name = string("op_40000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40000_cast_fp16 = slice_by_index(begin = var_40000_begin_0, end = var_40000_end_0, end_mask = var_40000_end_mask_0, x = query_53_cast_fp16)[name = string("op_40000_cast_fp16")]; + tensor var_40004_begin_0 = const()[name = string("op_40004_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40004_end_0 = const()[name = string("op_40004_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_40004_end_mask_0 = const()[name = string("op_40004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40004_cast_fp16 = slice_by_index(begin = var_40004_begin_0, end = var_40004_end_0, end_mask = var_40004_end_mask_0, x = query_53_cast_fp16)[name = string("op_40004_cast_fp16")]; + tensor var_40008_begin_0 = const()[name = string("op_40008_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40008_end_0 = const()[name = string("op_40008_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_40008_end_mask_0 = const()[name = string("op_40008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40008_cast_fp16 = slice_by_index(begin = var_40008_begin_0, end = var_40008_end_0, end_mask = var_40008_end_mask_0, x = query_53_cast_fp16)[name = string("op_40008_cast_fp16")]; + tensor var_40012_begin_0 = const()[name = string("op_40012_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40012_end_0 = const()[name = string("op_40012_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_40012_end_mask_0 = const()[name = string("op_40012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40012_cast_fp16 = slice_by_index(begin = var_40012_begin_0, end = var_40012_end_0, end_mask = var_40012_end_mask_0, x = query_53_cast_fp16)[name = string("op_40012_cast_fp16")]; + tensor var_40016_begin_0 = const()[name = string("op_40016_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40016_end_0 = const()[name = string("op_40016_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_40016_end_mask_0 = const()[name = string("op_40016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40016_cast_fp16 = slice_by_index(begin = var_40016_begin_0, end = var_40016_end_0, end_mask = var_40016_end_mask_0, x = query_53_cast_fp16)[name = string("op_40016_cast_fp16")]; + tensor var_40020_begin_0 = const()[name = string("op_40020_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40020_end_0 = const()[name = string("op_40020_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_40020_end_mask_0 = const()[name = string("op_40020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40020_cast_fp16 = slice_by_index(begin = var_40020_begin_0, end = var_40020_end_0, end_mask = var_40020_end_mask_0, x = query_53_cast_fp16)[name = string("op_40020_cast_fp16")]; + tensor var_40024_begin_0 = const()[name = string("op_40024_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40024_end_0 = const()[name = string("op_40024_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_40024_end_mask_0 = const()[name = string("op_40024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40024_cast_fp16 = slice_by_index(begin = var_40024_begin_0, end = var_40024_end_0, end_mask = var_40024_end_mask_0, x = query_53_cast_fp16)[name = string("op_40024_cast_fp16")]; + tensor var_40028_begin_0 = const()[name = string("op_40028_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40028_end_0 = const()[name = string("op_40028_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_40028_end_mask_0 = const()[name = string("op_40028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40028_cast_fp16 = slice_by_index(begin = var_40028_begin_0, end = var_40028_end_0, end_mask = var_40028_end_mask_0, x = query_53_cast_fp16)[name = string("op_40028_cast_fp16")]; + tensor var_40032_begin_0 = const()[name = string("op_40032_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40032_end_0 = const()[name = string("op_40032_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_40032_end_mask_0 = const()[name = string("op_40032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40032_cast_fp16 = slice_by_index(begin = var_40032_begin_0, end = var_40032_end_0, end_mask = var_40032_end_mask_0, x = query_53_cast_fp16)[name = string("op_40032_cast_fp16")]; + tensor var_40036_begin_0 = const()[name = string("op_40036_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40036_end_0 = const()[name = string("op_40036_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_40036_end_mask_0 = const()[name = string("op_40036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40036_cast_fp16 = slice_by_index(begin = var_40036_begin_0, end = var_40036_end_0, end_mask = var_40036_end_mask_0, x = query_53_cast_fp16)[name = string("op_40036_cast_fp16")]; + tensor var_40040_begin_0 = const()[name = string("op_40040_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40040_end_0 = const()[name = string("op_40040_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_40040_end_mask_0 = const()[name = string("op_40040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40040_cast_fp16 = slice_by_index(begin = var_40040_begin_0, end = var_40040_end_0, end_mask = var_40040_end_mask_0, x = query_53_cast_fp16)[name = string("op_40040_cast_fp16")]; + tensor var_40049_begin_0 = const()[name = string("op_40049_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40049_end_0 = const()[name = string("op_40049_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40049_end_mask_0 = const()[name = string("op_40049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40049_cast_fp16 = slice_by_index(begin = var_40049_begin_0, end = var_40049_end_0, end_mask = var_40049_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40049_cast_fp16")]; + tensor var_40056_begin_0 = const()[name = string("op_40056_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40056_end_0 = const()[name = string("op_40056_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40056_end_mask_0 = const()[name = string("op_40056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40056_cast_fp16 = slice_by_index(begin = var_40056_begin_0, end = var_40056_end_0, end_mask = var_40056_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40056_cast_fp16")]; + tensor var_40063_begin_0 = const()[name = string("op_40063_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40063_end_0 = const()[name = string("op_40063_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40063_end_mask_0 = const()[name = string("op_40063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40063_cast_fp16 = slice_by_index(begin = var_40063_begin_0, end = var_40063_end_0, end_mask = var_40063_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40063_cast_fp16")]; + tensor var_40070_begin_0 = const()[name = string("op_40070_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40070_end_0 = const()[name = string("op_40070_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40070_end_mask_0 = const()[name = string("op_40070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40070_cast_fp16 = slice_by_index(begin = var_40070_begin_0, end = var_40070_end_0, end_mask = var_40070_end_mask_0, x = var_39964_cast_fp16)[name = string("op_40070_cast_fp16")]; + tensor var_40077_begin_0 = const()[name = string("op_40077_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40077_end_0 = const()[name = string("op_40077_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40077_end_mask_0 = const()[name = string("op_40077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40077_cast_fp16 = slice_by_index(begin = var_40077_begin_0, end = var_40077_end_0, end_mask = var_40077_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40077_cast_fp16")]; + tensor var_40084_begin_0 = const()[name = string("op_40084_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40084_end_0 = const()[name = string("op_40084_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40084_end_mask_0 = const()[name = string("op_40084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40084_cast_fp16 = slice_by_index(begin = var_40084_begin_0, end = var_40084_end_0, end_mask = var_40084_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40084_cast_fp16")]; + tensor var_40091_begin_0 = const()[name = string("op_40091_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40091_end_0 = const()[name = string("op_40091_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40091_end_mask_0 = const()[name = string("op_40091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40091_cast_fp16 = slice_by_index(begin = var_40091_begin_0, end = var_40091_end_0, end_mask = var_40091_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40091_cast_fp16")]; + tensor var_40098_begin_0 = const()[name = string("op_40098_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40098_end_0 = const()[name = string("op_40098_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40098_end_mask_0 = const()[name = string("op_40098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40098_cast_fp16 = slice_by_index(begin = var_40098_begin_0, end = var_40098_end_0, end_mask = var_40098_end_mask_0, x = var_39968_cast_fp16)[name = string("op_40098_cast_fp16")]; + tensor var_40105_begin_0 = const()[name = string("op_40105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40105_end_0 = const()[name = string("op_40105_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40105_end_mask_0 = const()[name = string("op_40105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40105_cast_fp16 = slice_by_index(begin = var_40105_begin_0, end = var_40105_end_0, end_mask = var_40105_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40105_cast_fp16")]; + tensor var_40112_begin_0 = const()[name = string("op_40112_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40112_end_0 = const()[name = string("op_40112_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40112_end_mask_0 = const()[name = string("op_40112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40112_cast_fp16 = slice_by_index(begin = var_40112_begin_0, end = var_40112_end_0, end_mask = var_40112_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40112_cast_fp16")]; + tensor var_40119_begin_0 = const()[name = string("op_40119_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40119_end_0 = const()[name = string("op_40119_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40119_end_mask_0 = const()[name = string("op_40119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40119_cast_fp16 = slice_by_index(begin = var_40119_begin_0, end = var_40119_end_0, end_mask = var_40119_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40119_cast_fp16")]; + tensor var_40126_begin_0 = const()[name = string("op_40126_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40126_end_0 = const()[name = string("op_40126_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40126_end_mask_0 = const()[name = string("op_40126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40126_cast_fp16 = slice_by_index(begin = var_40126_begin_0, end = var_40126_end_0, end_mask = var_40126_end_mask_0, x = var_39972_cast_fp16)[name = string("op_40126_cast_fp16")]; + tensor var_40133_begin_0 = const()[name = string("op_40133_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40133_end_0 = const()[name = string("op_40133_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40133_end_mask_0 = const()[name = string("op_40133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40133_cast_fp16 = slice_by_index(begin = var_40133_begin_0, end = var_40133_end_0, end_mask = var_40133_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40133_cast_fp16")]; + tensor var_40140_begin_0 = const()[name = string("op_40140_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40140_end_0 = const()[name = string("op_40140_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40140_end_mask_0 = const()[name = string("op_40140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40140_cast_fp16 = slice_by_index(begin = var_40140_begin_0, end = var_40140_end_0, end_mask = var_40140_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40140_cast_fp16")]; + tensor var_40147_begin_0 = const()[name = string("op_40147_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40147_end_0 = const()[name = string("op_40147_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40147_end_mask_0 = const()[name = string("op_40147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40147_cast_fp16 = slice_by_index(begin = var_40147_begin_0, end = var_40147_end_0, end_mask = var_40147_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40147_cast_fp16")]; + tensor var_40154_begin_0 = const()[name = string("op_40154_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40154_end_0 = const()[name = string("op_40154_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40154_end_mask_0 = const()[name = string("op_40154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40154_cast_fp16 = slice_by_index(begin = var_40154_begin_0, end = var_40154_end_0, end_mask = var_40154_end_mask_0, x = var_39976_cast_fp16)[name = string("op_40154_cast_fp16")]; + tensor var_40161_begin_0 = const()[name = string("op_40161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40161_end_0 = const()[name = string("op_40161_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40161_end_mask_0 = const()[name = string("op_40161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40161_cast_fp16 = slice_by_index(begin = var_40161_begin_0, end = var_40161_end_0, end_mask = var_40161_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40161_cast_fp16")]; + tensor var_40168_begin_0 = const()[name = string("op_40168_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40168_end_0 = const()[name = string("op_40168_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40168_end_mask_0 = const()[name = string("op_40168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40168_cast_fp16 = slice_by_index(begin = var_40168_begin_0, end = var_40168_end_0, end_mask = var_40168_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40168_cast_fp16")]; + tensor var_40175_begin_0 = const()[name = string("op_40175_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40175_end_0 = const()[name = string("op_40175_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40175_end_mask_0 = const()[name = string("op_40175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40175_cast_fp16 = slice_by_index(begin = var_40175_begin_0, end = var_40175_end_0, end_mask = var_40175_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40175_cast_fp16")]; + tensor var_40182_begin_0 = const()[name = string("op_40182_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40182_end_0 = const()[name = string("op_40182_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40182_end_mask_0 = const()[name = string("op_40182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40182_cast_fp16 = slice_by_index(begin = var_40182_begin_0, end = var_40182_end_0, end_mask = var_40182_end_mask_0, x = var_39980_cast_fp16)[name = string("op_40182_cast_fp16")]; + tensor var_40189_begin_0 = const()[name = string("op_40189_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40189_end_0 = const()[name = string("op_40189_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40189_end_mask_0 = const()[name = string("op_40189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40189_cast_fp16 = slice_by_index(begin = var_40189_begin_0, end = var_40189_end_0, end_mask = var_40189_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40189_cast_fp16")]; + tensor var_40196_begin_0 = const()[name = string("op_40196_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40196_end_0 = const()[name = string("op_40196_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40196_end_mask_0 = const()[name = string("op_40196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40196_cast_fp16 = slice_by_index(begin = var_40196_begin_0, end = var_40196_end_0, end_mask = var_40196_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40196_cast_fp16")]; + tensor var_40203_begin_0 = const()[name = string("op_40203_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40203_end_0 = const()[name = string("op_40203_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40203_end_mask_0 = const()[name = string("op_40203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40203_cast_fp16 = slice_by_index(begin = var_40203_begin_0, end = var_40203_end_0, end_mask = var_40203_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40203_cast_fp16")]; + tensor var_40210_begin_0 = const()[name = string("op_40210_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40210_end_0 = const()[name = string("op_40210_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40210_end_mask_0 = const()[name = string("op_40210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40210_cast_fp16 = slice_by_index(begin = var_40210_begin_0, end = var_40210_end_0, end_mask = var_40210_end_mask_0, x = var_39984_cast_fp16)[name = string("op_40210_cast_fp16")]; + tensor var_40217_begin_0 = const()[name = string("op_40217_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40217_end_0 = const()[name = string("op_40217_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40217_end_mask_0 = const()[name = string("op_40217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40217_cast_fp16 = slice_by_index(begin = var_40217_begin_0, end = var_40217_end_0, end_mask = var_40217_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40217_cast_fp16")]; + tensor var_40224_begin_0 = const()[name = string("op_40224_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40224_end_0 = const()[name = string("op_40224_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40224_end_mask_0 = const()[name = string("op_40224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40224_cast_fp16 = slice_by_index(begin = var_40224_begin_0, end = var_40224_end_0, end_mask = var_40224_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40224_cast_fp16")]; + tensor var_40231_begin_0 = const()[name = string("op_40231_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40231_end_0 = const()[name = string("op_40231_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40231_end_mask_0 = const()[name = string("op_40231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40231_cast_fp16 = slice_by_index(begin = var_40231_begin_0, end = var_40231_end_0, end_mask = var_40231_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40231_cast_fp16")]; + tensor var_40238_begin_0 = const()[name = string("op_40238_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40238_end_0 = const()[name = string("op_40238_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40238_end_mask_0 = const()[name = string("op_40238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40238_cast_fp16 = slice_by_index(begin = var_40238_begin_0, end = var_40238_end_0, end_mask = var_40238_end_mask_0, x = var_39988_cast_fp16)[name = string("op_40238_cast_fp16")]; + tensor var_40245_begin_0 = const()[name = string("op_40245_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40245_end_0 = const()[name = string("op_40245_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40245_end_mask_0 = const()[name = string("op_40245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40245_cast_fp16 = slice_by_index(begin = var_40245_begin_0, end = var_40245_end_0, end_mask = var_40245_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40245_cast_fp16")]; + tensor var_40252_begin_0 = const()[name = string("op_40252_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40252_end_0 = const()[name = string("op_40252_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40252_end_mask_0 = const()[name = string("op_40252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40252_cast_fp16 = slice_by_index(begin = var_40252_begin_0, end = var_40252_end_0, end_mask = var_40252_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40252_cast_fp16")]; + tensor var_40259_begin_0 = const()[name = string("op_40259_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40259_end_0 = const()[name = string("op_40259_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40259_end_mask_0 = const()[name = string("op_40259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40259_cast_fp16 = slice_by_index(begin = var_40259_begin_0, end = var_40259_end_0, end_mask = var_40259_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40259_cast_fp16")]; + tensor var_40266_begin_0 = const()[name = string("op_40266_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40266_end_0 = const()[name = string("op_40266_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40266_end_mask_0 = const()[name = string("op_40266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40266_cast_fp16 = slice_by_index(begin = var_40266_begin_0, end = var_40266_end_0, end_mask = var_40266_end_mask_0, x = var_39992_cast_fp16)[name = string("op_40266_cast_fp16")]; + tensor var_40273_begin_0 = const()[name = string("op_40273_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40273_end_0 = const()[name = string("op_40273_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40273_end_mask_0 = const()[name = string("op_40273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40273_cast_fp16 = slice_by_index(begin = var_40273_begin_0, end = var_40273_end_0, end_mask = var_40273_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40273_cast_fp16")]; + tensor var_40280_begin_0 = const()[name = string("op_40280_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40280_end_0 = const()[name = string("op_40280_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40280_end_mask_0 = const()[name = string("op_40280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40280_cast_fp16 = slice_by_index(begin = var_40280_begin_0, end = var_40280_end_0, end_mask = var_40280_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40280_cast_fp16")]; + tensor var_40287_begin_0 = const()[name = string("op_40287_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40287_end_0 = const()[name = string("op_40287_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40287_end_mask_0 = const()[name = string("op_40287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40287_cast_fp16 = slice_by_index(begin = var_40287_begin_0, end = var_40287_end_0, end_mask = var_40287_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40287_cast_fp16")]; + tensor var_40294_begin_0 = const()[name = string("op_40294_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40294_end_0 = const()[name = string("op_40294_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40294_end_mask_0 = const()[name = string("op_40294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40294_cast_fp16 = slice_by_index(begin = var_40294_begin_0, end = var_40294_end_0, end_mask = var_40294_end_mask_0, x = var_39996_cast_fp16)[name = string("op_40294_cast_fp16")]; + tensor var_40301_begin_0 = const()[name = string("op_40301_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40301_end_0 = const()[name = string("op_40301_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40301_end_mask_0 = const()[name = string("op_40301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40301_cast_fp16 = slice_by_index(begin = var_40301_begin_0, end = var_40301_end_0, end_mask = var_40301_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40301_cast_fp16")]; + tensor var_40308_begin_0 = const()[name = string("op_40308_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40308_end_0 = const()[name = string("op_40308_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40308_end_mask_0 = const()[name = string("op_40308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40308_cast_fp16 = slice_by_index(begin = var_40308_begin_0, end = var_40308_end_0, end_mask = var_40308_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40308_cast_fp16")]; + tensor var_40315_begin_0 = const()[name = string("op_40315_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40315_end_0 = const()[name = string("op_40315_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40315_end_mask_0 = const()[name = string("op_40315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40315_cast_fp16 = slice_by_index(begin = var_40315_begin_0, end = var_40315_end_0, end_mask = var_40315_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40315_cast_fp16")]; + tensor var_40322_begin_0 = const()[name = string("op_40322_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40322_end_0 = const()[name = string("op_40322_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40322_end_mask_0 = const()[name = string("op_40322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40322_cast_fp16 = slice_by_index(begin = var_40322_begin_0, end = var_40322_end_0, end_mask = var_40322_end_mask_0, x = var_40000_cast_fp16)[name = string("op_40322_cast_fp16")]; + tensor var_40329_begin_0 = const()[name = string("op_40329_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40329_end_0 = const()[name = string("op_40329_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40329_end_mask_0 = const()[name = string("op_40329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40329_cast_fp16 = slice_by_index(begin = var_40329_begin_0, end = var_40329_end_0, end_mask = var_40329_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40329_cast_fp16")]; + tensor var_40336_begin_0 = const()[name = string("op_40336_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40336_end_0 = const()[name = string("op_40336_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40336_end_mask_0 = const()[name = string("op_40336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40336_cast_fp16 = slice_by_index(begin = var_40336_begin_0, end = var_40336_end_0, end_mask = var_40336_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40336_cast_fp16")]; + tensor var_40343_begin_0 = const()[name = string("op_40343_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40343_end_0 = const()[name = string("op_40343_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40343_end_mask_0 = const()[name = string("op_40343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40343_cast_fp16 = slice_by_index(begin = var_40343_begin_0, end = var_40343_end_0, end_mask = var_40343_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40343_cast_fp16")]; + tensor var_40350_begin_0 = const()[name = string("op_40350_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40350_end_0 = const()[name = string("op_40350_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40350_end_mask_0 = const()[name = string("op_40350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40350_cast_fp16 = slice_by_index(begin = var_40350_begin_0, end = var_40350_end_0, end_mask = var_40350_end_mask_0, x = var_40004_cast_fp16)[name = string("op_40350_cast_fp16")]; + tensor var_40357_begin_0 = const()[name = string("op_40357_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40357_end_0 = const()[name = string("op_40357_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40357_end_mask_0 = const()[name = string("op_40357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40357_cast_fp16 = slice_by_index(begin = var_40357_begin_0, end = var_40357_end_0, end_mask = var_40357_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40357_cast_fp16")]; + tensor var_40364_begin_0 = const()[name = string("op_40364_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40364_end_0 = const()[name = string("op_40364_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40364_end_mask_0 = const()[name = string("op_40364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40364_cast_fp16 = slice_by_index(begin = var_40364_begin_0, end = var_40364_end_0, end_mask = var_40364_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40364_cast_fp16")]; + tensor var_40371_begin_0 = const()[name = string("op_40371_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40371_end_0 = const()[name = string("op_40371_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40371_end_mask_0 = const()[name = string("op_40371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40371_cast_fp16 = slice_by_index(begin = var_40371_begin_0, end = var_40371_end_0, end_mask = var_40371_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40371_cast_fp16")]; + tensor var_40378_begin_0 = const()[name = string("op_40378_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40378_end_0 = const()[name = string("op_40378_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40378_end_mask_0 = const()[name = string("op_40378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40378_cast_fp16 = slice_by_index(begin = var_40378_begin_0, end = var_40378_end_0, end_mask = var_40378_end_mask_0, x = var_40008_cast_fp16)[name = string("op_40378_cast_fp16")]; + tensor var_40385_begin_0 = const()[name = string("op_40385_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40385_end_0 = const()[name = string("op_40385_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40385_end_mask_0 = const()[name = string("op_40385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40385_cast_fp16 = slice_by_index(begin = var_40385_begin_0, end = var_40385_end_0, end_mask = var_40385_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40385_cast_fp16")]; + tensor var_40392_begin_0 = const()[name = string("op_40392_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40392_end_0 = const()[name = string("op_40392_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40392_end_mask_0 = const()[name = string("op_40392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40392_cast_fp16 = slice_by_index(begin = var_40392_begin_0, end = var_40392_end_0, end_mask = var_40392_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40392_cast_fp16")]; + tensor var_40399_begin_0 = const()[name = string("op_40399_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40399_end_0 = const()[name = string("op_40399_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40399_end_mask_0 = const()[name = string("op_40399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40399_cast_fp16 = slice_by_index(begin = var_40399_begin_0, end = var_40399_end_0, end_mask = var_40399_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40399_cast_fp16")]; + tensor var_40406_begin_0 = const()[name = string("op_40406_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40406_end_0 = const()[name = string("op_40406_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40406_end_mask_0 = const()[name = string("op_40406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40406_cast_fp16 = slice_by_index(begin = var_40406_begin_0, end = var_40406_end_0, end_mask = var_40406_end_mask_0, x = var_40012_cast_fp16)[name = string("op_40406_cast_fp16")]; + tensor var_40413_begin_0 = const()[name = string("op_40413_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40413_end_0 = const()[name = string("op_40413_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40413_end_mask_0 = const()[name = string("op_40413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40413_cast_fp16 = slice_by_index(begin = var_40413_begin_0, end = var_40413_end_0, end_mask = var_40413_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40413_cast_fp16")]; + tensor var_40420_begin_0 = const()[name = string("op_40420_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40420_end_0 = const()[name = string("op_40420_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40420_end_mask_0 = const()[name = string("op_40420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40420_cast_fp16 = slice_by_index(begin = var_40420_begin_0, end = var_40420_end_0, end_mask = var_40420_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40420_cast_fp16")]; + tensor var_40427_begin_0 = const()[name = string("op_40427_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40427_end_0 = const()[name = string("op_40427_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40427_end_mask_0 = const()[name = string("op_40427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40427_cast_fp16 = slice_by_index(begin = var_40427_begin_0, end = var_40427_end_0, end_mask = var_40427_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40427_cast_fp16")]; + tensor var_40434_begin_0 = const()[name = string("op_40434_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40434_end_0 = const()[name = string("op_40434_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40434_end_mask_0 = const()[name = string("op_40434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40434_cast_fp16 = slice_by_index(begin = var_40434_begin_0, end = var_40434_end_0, end_mask = var_40434_end_mask_0, x = var_40016_cast_fp16)[name = string("op_40434_cast_fp16")]; + tensor var_40441_begin_0 = const()[name = string("op_40441_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40441_end_0 = const()[name = string("op_40441_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40441_end_mask_0 = const()[name = string("op_40441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40441_cast_fp16 = slice_by_index(begin = var_40441_begin_0, end = var_40441_end_0, end_mask = var_40441_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40441_cast_fp16")]; + tensor var_40448_begin_0 = const()[name = string("op_40448_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40448_end_0 = const()[name = string("op_40448_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40448_end_mask_0 = const()[name = string("op_40448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40448_cast_fp16 = slice_by_index(begin = var_40448_begin_0, end = var_40448_end_0, end_mask = var_40448_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40448_cast_fp16")]; + tensor var_40455_begin_0 = const()[name = string("op_40455_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40455_end_0 = const()[name = string("op_40455_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40455_end_mask_0 = const()[name = string("op_40455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40455_cast_fp16 = slice_by_index(begin = var_40455_begin_0, end = var_40455_end_0, end_mask = var_40455_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40455_cast_fp16")]; + tensor var_40462_begin_0 = const()[name = string("op_40462_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40462_end_0 = const()[name = string("op_40462_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40462_end_mask_0 = const()[name = string("op_40462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40462_cast_fp16 = slice_by_index(begin = var_40462_begin_0, end = var_40462_end_0, end_mask = var_40462_end_mask_0, x = var_40020_cast_fp16)[name = string("op_40462_cast_fp16")]; + tensor var_40469_begin_0 = const()[name = string("op_40469_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40469_end_0 = const()[name = string("op_40469_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40469_end_mask_0 = const()[name = string("op_40469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40469_cast_fp16 = slice_by_index(begin = var_40469_begin_0, end = var_40469_end_0, end_mask = var_40469_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40469_cast_fp16")]; + tensor var_40476_begin_0 = const()[name = string("op_40476_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40476_end_0 = const()[name = string("op_40476_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40476_end_mask_0 = const()[name = string("op_40476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40476_cast_fp16 = slice_by_index(begin = var_40476_begin_0, end = var_40476_end_0, end_mask = var_40476_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40476_cast_fp16")]; + tensor var_40483_begin_0 = const()[name = string("op_40483_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40483_end_0 = const()[name = string("op_40483_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40483_end_mask_0 = const()[name = string("op_40483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40483_cast_fp16 = slice_by_index(begin = var_40483_begin_0, end = var_40483_end_0, end_mask = var_40483_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40483_cast_fp16")]; + tensor var_40490_begin_0 = const()[name = string("op_40490_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40490_end_0 = const()[name = string("op_40490_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40490_end_mask_0 = const()[name = string("op_40490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40490_cast_fp16 = slice_by_index(begin = var_40490_begin_0, end = var_40490_end_0, end_mask = var_40490_end_mask_0, x = var_40024_cast_fp16)[name = string("op_40490_cast_fp16")]; + tensor var_40497_begin_0 = const()[name = string("op_40497_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40497_end_0 = const()[name = string("op_40497_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40497_end_mask_0 = const()[name = string("op_40497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40497_cast_fp16 = slice_by_index(begin = var_40497_begin_0, end = var_40497_end_0, end_mask = var_40497_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40497_cast_fp16")]; + tensor var_40504_begin_0 = const()[name = string("op_40504_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40504_end_0 = const()[name = string("op_40504_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40504_end_mask_0 = const()[name = string("op_40504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40504_cast_fp16 = slice_by_index(begin = var_40504_begin_0, end = var_40504_end_0, end_mask = var_40504_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40504_cast_fp16")]; + tensor var_40511_begin_0 = const()[name = string("op_40511_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40511_end_0 = const()[name = string("op_40511_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40511_end_mask_0 = const()[name = string("op_40511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40511_cast_fp16 = slice_by_index(begin = var_40511_begin_0, end = var_40511_end_0, end_mask = var_40511_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40511_cast_fp16")]; + tensor var_40518_begin_0 = const()[name = string("op_40518_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40518_end_0 = const()[name = string("op_40518_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40518_end_mask_0 = const()[name = string("op_40518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40518_cast_fp16 = slice_by_index(begin = var_40518_begin_0, end = var_40518_end_0, end_mask = var_40518_end_mask_0, x = var_40028_cast_fp16)[name = string("op_40518_cast_fp16")]; + tensor var_40525_begin_0 = const()[name = string("op_40525_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40525_end_0 = const()[name = string("op_40525_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40525_end_mask_0 = const()[name = string("op_40525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40525_cast_fp16 = slice_by_index(begin = var_40525_begin_0, end = var_40525_end_0, end_mask = var_40525_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40525_cast_fp16")]; + tensor var_40532_begin_0 = const()[name = string("op_40532_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40532_end_0 = const()[name = string("op_40532_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40532_end_mask_0 = const()[name = string("op_40532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40532_cast_fp16 = slice_by_index(begin = var_40532_begin_0, end = var_40532_end_0, end_mask = var_40532_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40532_cast_fp16")]; + tensor var_40539_begin_0 = const()[name = string("op_40539_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40539_end_0 = const()[name = string("op_40539_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40539_end_mask_0 = const()[name = string("op_40539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40539_cast_fp16 = slice_by_index(begin = var_40539_begin_0, end = var_40539_end_0, end_mask = var_40539_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40539_cast_fp16")]; + tensor var_40546_begin_0 = const()[name = string("op_40546_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40546_end_0 = const()[name = string("op_40546_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40546_end_mask_0 = const()[name = string("op_40546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40546_cast_fp16 = slice_by_index(begin = var_40546_begin_0, end = var_40546_end_0, end_mask = var_40546_end_mask_0, x = var_40032_cast_fp16)[name = string("op_40546_cast_fp16")]; + tensor var_40553_begin_0 = const()[name = string("op_40553_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40553_end_0 = const()[name = string("op_40553_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40553_end_mask_0 = const()[name = string("op_40553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40553_cast_fp16 = slice_by_index(begin = var_40553_begin_0, end = var_40553_end_0, end_mask = var_40553_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40553_cast_fp16")]; + tensor var_40560_begin_0 = const()[name = string("op_40560_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40560_end_0 = const()[name = string("op_40560_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40560_end_mask_0 = const()[name = string("op_40560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40560_cast_fp16 = slice_by_index(begin = var_40560_begin_0, end = var_40560_end_0, end_mask = var_40560_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40560_cast_fp16")]; + tensor var_40567_begin_0 = const()[name = string("op_40567_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40567_end_0 = const()[name = string("op_40567_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40567_end_mask_0 = const()[name = string("op_40567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40567_cast_fp16 = slice_by_index(begin = var_40567_begin_0, end = var_40567_end_0, end_mask = var_40567_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40567_cast_fp16")]; + tensor var_40574_begin_0 = const()[name = string("op_40574_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40574_end_0 = const()[name = string("op_40574_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40574_end_mask_0 = const()[name = string("op_40574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40574_cast_fp16 = slice_by_index(begin = var_40574_begin_0, end = var_40574_end_0, end_mask = var_40574_end_mask_0, x = var_40036_cast_fp16)[name = string("op_40574_cast_fp16")]; + tensor var_40581_begin_0 = const()[name = string("op_40581_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40581_end_0 = const()[name = string("op_40581_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40581_end_mask_0 = const()[name = string("op_40581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40581_cast_fp16 = slice_by_index(begin = var_40581_begin_0, end = var_40581_end_0, end_mask = var_40581_end_mask_0, x = var_40040_cast_fp16)[name = string("op_40581_cast_fp16")]; + tensor var_40588_begin_0 = const()[name = string("op_40588_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40588_end_0 = const()[name = string("op_40588_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40588_end_mask_0 = const()[name = string("op_40588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40588_cast_fp16 = slice_by_index(begin = var_40588_begin_0, end = var_40588_end_0, end_mask = var_40588_end_mask_0, x = var_40040_cast_fp16)[name = string("op_40588_cast_fp16")]; + tensor var_40595_begin_0 = const()[name = string("op_40595_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40595_end_0 = const()[name = string("op_40595_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40595_end_mask_0 = const()[name = string("op_40595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40595_cast_fp16 = slice_by_index(begin = var_40595_begin_0, end = var_40595_end_0, end_mask = var_40595_end_mask_0, x = var_40040_cast_fp16)[name = string("op_40595_cast_fp16")]; + tensor var_40602_begin_0 = const()[name = string("op_40602_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40602_end_0 = const()[name = string("op_40602_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40602_end_mask_0 = const()[name = string("op_40602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40602_cast_fp16 = slice_by_index(begin = var_40602_begin_0, end = var_40602_end_0, end_mask = var_40602_end_mask_0, x = var_40040_cast_fp16)[name = string("op_40602_cast_fp16")]; + tensor k_53_perm_0 = const()[name = string("k_53_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_40607_begin_0 = const()[name = string("op_40607_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40607_end_0 = const()[name = string("op_40607_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_40607_end_mask_0 = const()[name = string("op_40607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_53_cast_fp16 = transpose(perm = k_53_perm_0, x = key_53_cast_fp16)[name = string("transpose_5")]; + tensor var_40607_cast_fp16 = slice_by_index(begin = var_40607_begin_0, end = var_40607_end_0, end_mask = var_40607_end_mask_0, x = k_53_cast_fp16)[name = string("op_40607_cast_fp16")]; + tensor var_40611_begin_0 = const()[name = string("op_40611_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_40611_end_0 = const()[name = string("op_40611_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_40611_end_mask_0 = const()[name = string("op_40611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40611_cast_fp16 = slice_by_index(begin = var_40611_begin_0, end = var_40611_end_0, end_mask = var_40611_end_mask_0, x = k_53_cast_fp16)[name = string("op_40611_cast_fp16")]; + tensor var_40615_begin_0 = const()[name = string("op_40615_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_40615_end_0 = const()[name = string("op_40615_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_40615_end_mask_0 = const()[name = string("op_40615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40615_cast_fp16 = slice_by_index(begin = var_40615_begin_0, end = var_40615_end_0, end_mask = var_40615_end_mask_0, x = k_53_cast_fp16)[name = string("op_40615_cast_fp16")]; + tensor var_40619_begin_0 = const()[name = string("op_40619_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_40619_end_0 = const()[name = string("op_40619_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_40619_end_mask_0 = const()[name = string("op_40619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40619_cast_fp16 = slice_by_index(begin = var_40619_begin_0, end = var_40619_end_0, end_mask = var_40619_end_mask_0, x = k_53_cast_fp16)[name = string("op_40619_cast_fp16")]; + tensor var_40623_begin_0 = const()[name = string("op_40623_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_40623_end_0 = const()[name = string("op_40623_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_40623_end_mask_0 = const()[name = string("op_40623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40623_cast_fp16 = slice_by_index(begin = var_40623_begin_0, end = var_40623_end_0, end_mask = var_40623_end_mask_0, x = k_53_cast_fp16)[name = string("op_40623_cast_fp16")]; + tensor var_40627_begin_0 = const()[name = string("op_40627_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_40627_end_0 = const()[name = string("op_40627_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_40627_end_mask_0 = const()[name = string("op_40627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40627_cast_fp16 = slice_by_index(begin = var_40627_begin_0, end = var_40627_end_0, end_mask = var_40627_end_mask_0, x = k_53_cast_fp16)[name = string("op_40627_cast_fp16")]; + tensor var_40631_begin_0 = const()[name = string("op_40631_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_40631_end_0 = const()[name = string("op_40631_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_40631_end_mask_0 = const()[name = string("op_40631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40631_cast_fp16 = slice_by_index(begin = var_40631_begin_0, end = var_40631_end_0, end_mask = var_40631_end_mask_0, x = k_53_cast_fp16)[name = string("op_40631_cast_fp16")]; + tensor var_40635_begin_0 = const()[name = string("op_40635_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_40635_end_0 = const()[name = string("op_40635_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_40635_end_mask_0 = const()[name = string("op_40635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40635_cast_fp16 = slice_by_index(begin = var_40635_begin_0, end = var_40635_end_0, end_mask = var_40635_end_mask_0, x = k_53_cast_fp16)[name = string("op_40635_cast_fp16")]; + tensor var_40639_begin_0 = const()[name = string("op_40639_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_40639_end_0 = const()[name = string("op_40639_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_40639_end_mask_0 = const()[name = string("op_40639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40639_cast_fp16 = slice_by_index(begin = var_40639_begin_0, end = var_40639_end_0, end_mask = var_40639_end_mask_0, x = k_53_cast_fp16)[name = string("op_40639_cast_fp16")]; + tensor var_40643_begin_0 = const()[name = string("op_40643_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_40643_end_0 = const()[name = string("op_40643_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_40643_end_mask_0 = const()[name = string("op_40643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40643_cast_fp16 = slice_by_index(begin = var_40643_begin_0, end = var_40643_end_0, end_mask = var_40643_end_mask_0, x = k_53_cast_fp16)[name = string("op_40643_cast_fp16")]; + tensor var_40647_begin_0 = const()[name = string("op_40647_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_40647_end_0 = const()[name = string("op_40647_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_40647_end_mask_0 = const()[name = string("op_40647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40647_cast_fp16 = slice_by_index(begin = var_40647_begin_0, end = var_40647_end_0, end_mask = var_40647_end_mask_0, x = k_53_cast_fp16)[name = string("op_40647_cast_fp16")]; + tensor var_40651_begin_0 = const()[name = string("op_40651_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_40651_end_0 = const()[name = string("op_40651_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_40651_end_mask_0 = const()[name = string("op_40651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40651_cast_fp16 = slice_by_index(begin = var_40651_begin_0, end = var_40651_end_0, end_mask = var_40651_end_mask_0, x = k_53_cast_fp16)[name = string("op_40651_cast_fp16")]; + tensor var_40655_begin_0 = const()[name = string("op_40655_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_40655_end_0 = const()[name = string("op_40655_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_40655_end_mask_0 = const()[name = string("op_40655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40655_cast_fp16 = slice_by_index(begin = var_40655_begin_0, end = var_40655_end_0, end_mask = var_40655_end_mask_0, x = k_53_cast_fp16)[name = string("op_40655_cast_fp16")]; + tensor var_40659_begin_0 = const()[name = string("op_40659_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_40659_end_0 = const()[name = string("op_40659_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_40659_end_mask_0 = const()[name = string("op_40659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40659_cast_fp16 = slice_by_index(begin = var_40659_begin_0, end = var_40659_end_0, end_mask = var_40659_end_mask_0, x = k_53_cast_fp16)[name = string("op_40659_cast_fp16")]; + tensor var_40663_begin_0 = const()[name = string("op_40663_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_40663_end_0 = const()[name = string("op_40663_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_40663_end_mask_0 = const()[name = string("op_40663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40663_cast_fp16 = slice_by_index(begin = var_40663_begin_0, end = var_40663_end_0, end_mask = var_40663_end_mask_0, x = k_53_cast_fp16)[name = string("op_40663_cast_fp16")]; + tensor var_40667_begin_0 = const()[name = string("op_40667_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_40667_end_0 = const()[name = string("op_40667_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_40667_end_mask_0 = const()[name = string("op_40667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40667_cast_fp16 = slice_by_index(begin = var_40667_begin_0, end = var_40667_end_0, end_mask = var_40667_end_mask_0, x = k_53_cast_fp16)[name = string("op_40667_cast_fp16")]; + tensor var_40671_begin_0 = const()[name = string("op_40671_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_40671_end_0 = const()[name = string("op_40671_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_40671_end_mask_0 = const()[name = string("op_40671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40671_cast_fp16 = slice_by_index(begin = var_40671_begin_0, end = var_40671_end_0, end_mask = var_40671_end_mask_0, x = k_53_cast_fp16)[name = string("op_40671_cast_fp16")]; + tensor var_40675_begin_0 = const()[name = string("op_40675_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_40675_end_0 = const()[name = string("op_40675_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_40675_end_mask_0 = const()[name = string("op_40675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40675_cast_fp16 = slice_by_index(begin = var_40675_begin_0, end = var_40675_end_0, end_mask = var_40675_end_mask_0, x = k_53_cast_fp16)[name = string("op_40675_cast_fp16")]; + tensor var_40679_begin_0 = const()[name = string("op_40679_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_40679_end_0 = const()[name = string("op_40679_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_40679_end_mask_0 = const()[name = string("op_40679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40679_cast_fp16 = slice_by_index(begin = var_40679_begin_0, end = var_40679_end_0, end_mask = var_40679_end_mask_0, x = k_53_cast_fp16)[name = string("op_40679_cast_fp16")]; + tensor var_40683_begin_0 = const()[name = string("op_40683_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_40683_end_0 = const()[name = string("op_40683_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_40683_end_mask_0 = const()[name = string("op_40683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40683_cast_fp16 = slice_by_index(begin = var_40683_begin_0, end = var_40683_end_0, end_mask = var_40683_end_mask_0, x = k_53_cast_fp16)[name = string("op_40683_cast_fp16")]; + tensor var_40685_begin_0 = const()[name = string("op_40685_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40685_end_0 = const()[name = string("op_40685_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40685_end_mask_0 = const()[name = string("op_40685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40685_cast_fp16 = slice_by_index(begin = var_40685_begin_0, end = var_40685_end_0, end_mask = var_40685_end_mask_0, x = value_53_cast_fp16)[name = string("op_40685_cast_fp16")]; + tensor var_40689_begin_0 = const()[name = string("op_40689_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40689_end_0 = const()[name = string("op_40689_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_40689_end_mask_0 = const()[name = string("op_40689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40689_cast_fp16 = slice_by_index(begin = var_40689_begin_0, end = var_40689_end_0, end_mask = var_40689_end_mask_0, x = value_53_cast_fp16)[name = string("op_40689_cast_fp16")]; + tensor var_40693_begin_0 = const()[name = string("op_40693_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40693_end_0 = const()[name = string("op_40693_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_40693_end_mask_0 = const()[name = string("op_40693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40693_cast_fp16 = slice_by_index(begin = var_40693_begin_0, end = var_40693_end_0, end_mask = var_40693_end_mask_0, x = value_53_cast_fp16)[name = string("op_40693_cast_fp16")]; + tensor var_40697_begin_0 = const()[name = string("op_40697_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40697_end_0 = const()[name = string("op_40697_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_40697_end_mask_0 = const()[name = string("op_40697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40697_cast_fp16 = slice_by_index(begin = var_40697_begin_0, end = var_40697_end_0, end_mask = var_40697_end_mask_0, x = value_53_cast_fp16)[name = string("op_40697_cast_fp16")]; + tensor var_40701_begin_0 = const()[name = string("op_40701_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40701_end_0 = const()[name = string("op_40701_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_40701_end_mask_0 = const()[name = string("op_40701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40701_cast_fp16 = slice_by_index(begin = var_40701_begin_0, end = var_40701_end_0, end_mask = var_40701_end_mask_0, x = value_53_cast_fp16)[name = string("op_40701_cast_fp16")]; + tensor var_40705_begin_0 = const()[name = string("op_40705_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40705_end_0 = const()[name = string("op_40705_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_40705_end_mask_0 = const()[name = string("op_40705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40705_cast_fp16 = slice_by_index(begin = var_40705_begin_0, end = var_40705_end_0, end_mask = var_40705_end_mask_0, x = value_53_cast_fp16)[name = string("op_40705_cast_fp16")]; + tensor var_40709_begin_0 = const()[name = string("op_40709_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40709_end_0 = const()[name = string("op_40709_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_40709_end_mask_0 = const()[name = string("op_40709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40709_cast_fp16 = slice_by_index(begin = var_40709_begin_0, end = var_40709_end_0, end_mask = var_40709_end_mask_0, x = value_53_cast_fp16)[name = string("op_40709_cast_fp16")]; + tensor var_40713_begin_0 = const()[name = string("op_40713_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40713_end_0 = const()[name = string("op_40713_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_40713_end_mask_0 = const()[name = string("op_40713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40713_cast_fp16 = slice_by_index(begin = var_40713_begin_0, end = var_40713_end_0, end_mask = var_40713_end_mask_0, x = value_53_cast_fp16)[name = string("op_40713_cast_fp16")]; + tensor var_40717_begin_0 = const()[name = string("op_40717_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40717_end_0 = const()[name = string("op_40717_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_40717_end_mask_0 = const()[name = string("op_40717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40717_cast_fp16 = slice_by_index(begin = var_40717_begin_0, end = var_40717_end_0, end_mask = var_40717_end_mask_0, x = value_53_cast_fp16)[name = string("op_40717_cast_fp16")]; + tensor var_40721_begin_0 = const()[name = string("op_40721_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40721_end_0 = const()[name = string("op_40721_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_40721_end_mask_0 = const()[name = string("op_40721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40721_cast_fp16 = slice_by_index(begin = var_40721_begin_0, end = var_40721_end_0, end_mask = var_40721_end_mask_0, x = value_53_cast_fp16)[name = string("op_40721_cast_fp16")]; + tensor var_40725_begin_0 = const()[name = string("op_40725_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40725_end_0 = const()[name = string("op_40725_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_40725_end_mask_0 = const()[name = string("op_40725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40725_cast_fp16 = slice_by_index(begin = var_40725_begin_0, end = var_40725_end_0, end_mask = var_40725_end_mask_0, x = value_53_cast_fp16)[name = string("op_40725_cast_fp16")]; + tensor var_40729_begin_0 = const()[name = string("op_40729_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40729_end_0 = const()[name = string("op_40729_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_40729_end_mask_0 = const()[name = string("op_40729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40729_cast_fp16 = slice_by_index(begin = var_40729_begin_0, end = var_40729_end_0, end_mask = var_40729_end_mask_0, x = value_53_cast_fp16)[name = string("op_40729_cast_fp16")]; + tensor var_40733_begin_0 = const()[name = string("op_40733_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40733_end_0 = const()[name = string("op_40733_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_40733_end_mask_0 = const()[name = string("op_40733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40733_cast_fp16 = slice_by_index(begin = var_40733_begin_0, end = var_40733_end_0, end_mask = var_40733_end_mask_0, x = value_53_cast_fp16)[name = string("op_40733_cast_fp16")]; + tensor var_40737_begin_0 = const()[name = string("op_40737_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40737_end_0 = const()[name = string("op_40737_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_40737_end_mask_0 = const()[name = string("op_40737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40737_cast_fp16 = slice_by_index(begin = var_40737_begin_0, end = var_40737_end_0, end_mask = var_40737_end_mask_0, x = value_53_cast_fp16)[name = string("op_40737_cast_fp16")]; + tensor var_40741_begin_0 = const()[name = string("op_40741_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40741_end_0 = const()[name = string("op_40741_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_40741_end_mask_0 = const()[name = string("op_40741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40741_cast_fp16 = slice_by_index(begin = var_40741_begin_0, end = var_40741_end_0, end_mask = var_40741_end_mask_0, x = value_53_cast_fp16)[name = string("op_40741_cast_fp16")]; + tensor var_40745_begin_0 = const()[name = string("op_40745_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40745_end_0 = const()[name = string("op_40745_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_40745_end_mask_0 = const()[name = string("op_40745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40745_cast_fp16 = slice_by_index(begin = var_40745_begin_0, end = var_40745_end_0, end_mask = var_40745_end_mask_0, x = value_53_cast_fp16)[name = string("op_40745_cast_fp16")]; + tensor var_40749_begin_0 = const()[name = string("op_40749_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40749_end_0 = const()[name = string("op_40749_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_40749_end_mask_0 = const()[name = string("op_40749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40749_cast_fp16 = slice_by_index(begin = var_40749_begin_0, end = var_40749_end_0, end_mask = var_40749_end_mask_0, x = value_53_cast_fp16)[name = string("op_40749_cast_fp16")]; + tensor var_40753_begin_0 = const()[name = string("op_40753_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40753_end_0 = const()[name = string("op_40753_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_40753_end_mask_0 = const()[name = string("op_40753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40753_cast_fp16 = slice_by_index(begin = var_40753_begin_0, end = var_40753_end_0, end_mask = var_40753_end_mask_0, x = value_53_cast_fp16)[name = string("op_40753_cast_fp16")]; + tensor var_40757_begin_0 = const()[name = string("op_40757_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40757_end_0 = const()[name = string("op_40757_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_40757_end_mask_0 = const()[name = string("op_40757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40757_cast_fp16 = slice_by_index(begin = var_40757_begin_0, end = var_40757_end_0, end_mask = var_40757_end_mask_0, x = value_53_cast_fp16)[name = string("op_40757_cast_fp16")]; + tensor var_40761_begin_0 = const()[name = string("op_40761_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40761_end_0 = const()[name = string("op_40761_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_40761_end_mask_0 = const()[name = string("op_40761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40761_cast_fp16 = slice_by_index(begin = var_40761_begin_0, end = var_40761_end_0, end_mask = var_40761_end_mask_0, x = value_53_cast_fp16)[name = string("op_40761_cast_fp16")]; + string _SplitHeadsQ__mh_w_4161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4161_equation_0, values = (var_40607_cast_fp16, var_40049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4161_cast_fp16")]; + string _SplitHeadsQ__mh_w_4163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4163_equation_0, values = (var_40607_cast_fp16, var_40056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4163_cast_fp16")]; + string _SplitHeadsQ__mh_w_4165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4165_equation_0, values = (var_40607_cast_fp16, var_40063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4165_cast_fp16")]; + string _SplitHeadsQ__mh_w_4167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4167_equation_0, values = (var_40607_cast_fp16, var_40070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4167_cast_fp16")]; + string _SplitHeadsQ__mh_w_4169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4169_equation_0, values = (var_40611_cast_fp16, var_40077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4169_cast_fp16")]; + string _SplitHeadsQ__mh_w_4171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4171_equation_0, values = (var_40611_cast_fp16, var_40084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4171_cast_fp16")]; + string _SplitHeadsQ__mh_w_4173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4173_equation_0, values = (var_40611_cast_fp16, var_40091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4173_cast_fp16")]; + string _SplitHeadsQ__mh_w_4175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4175_equation_0, values = (var_40611_cast_fp16, var_40098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4175_cast_fp16")]; + string _SplitHeadsQ__mh_w_4177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4177_equation_0, values = (var_40615_cast_fp16, var_40105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4177_cast_fp16")]; + string _SplitHeadsQ__mh_w_4179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4179_equation_0, values = (var_40615_cast_fp16, var_40112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4179_cast_fp16")]; + string _SplitHeadsQ__mh_w_4181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4181_equation_0, values = (var_40615_cast_fp16, var_40119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4181_cast_fp16")]; + string _SplitHeadsQ__mh_w_4183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4183_equation_0, values = (var_40615_cast_fp16, var_40126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4183_cast_fp16")]; + string _SplitHeadsQ__mh_w_4185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4185_equation_0, values = (var_40619_cast_fp16, var_40133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4185_cast_fp16")]; + string _SplitHeadsQ__mh_w_4187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4187_equation_0, values = (var_40619_cast_fp16, var_40140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4187_cast_fp16")]; + string _SplitHeadsQ__mh_w_4189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4189_equation_0, values = (var_40619_cast_fp16, var_40147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4189_cast_fp16")]; + string _SplitHeadsQ__mh_w_4191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4191_equation_0, values = (var_40619_cast_fp16, var_40154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4191_cast_fp16")]; + string _SplitHeadsQ__mh_w_4193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4193_equation_0, values = (var_40623_cast_fp16, var_40161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4193_cast_fp16")]; + string _SplitHeadsQ__mh_w_4195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4195_equation_0, values = (var_40623_cast_fp16, var_40168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4195_cast_fp16")]; + string _SplitHeadsQ__mh_w_4197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4197_equation_0, values = (var_40623_cast_fp16, var_40175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4197_cast_fp16")]; + string _SplitHeadsQ__mh_w_4199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4199_equation_0, values = (var_40623_cast_fp16, var_40182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4199_cast_fp16")]; + string _SplitHeadsQ__mh_w_4201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4201_equation_0, values = (var_40627_cast_fp16, var_40189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4201_cast_fp16")]; + string _SplitHeadsQ__mh_w_4203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4203_equation_0, values = (var_40627_cast_fp16, var_40196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4203_cast_fp16")]; + string _SplitHeadsQ__mh_w_4205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4205_equation_0, values = (var_40627_cast_fp16, var_40203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4205_cast_fp16")]; + string _SplitHeadsQ__mh_w_4207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4207_equation_0, values = (var_40627_cast_fp16, var_40210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4207_cast_fp16")]; + string _SplitHeadsQ__mh_w_4209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4209_equation_0, values = (var_40631_cast_fp16, var_40217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4209_cast_fp16")]; + string _SplitHeadsQ__mh_w_4211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4211_equation_0, values = (var_40631_cast_fp16, var_40224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4211_cast_fp16")]; + string _SplitHeadsQ__mh_w_4213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4213_equation_0, values = (var_40631_cast_fp16, var_40231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4213_cast_fp16")]; + string _SplitHeadsQ__mh_w_4215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4215_equation_0, values = (var_40631_cast_fp16, var_40238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4215_cast_fp16")]; + string _SplitHeadsQ__mh_w_4217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4217_equation_0, values = (var_40635_cast_fp16, var_40245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4217_cast_fp16")]; + string _SplitHeadsQ__mh_w_4219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4219_equation_0, values = (var_40635_cast_fp16, var_40252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4219_cast_fp16")]; + string _SplitHeadsQ__mh_w_4221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4221_equation_0, values = (var_40635_cast_fp16, var_40259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4221_cast_fp16")]; + string _SplitHeadsQ__mh_w_4223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4223_equation_0, values = (var_40635_cast_fp16, var_40266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4223_cast_fp16")]; + string _SplitHeadsQ__mh_w_4225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4225_equation_0, values = (var_40639_cast_fp16, var_40273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4225_cast_fp16")]; + string _SplitHeadsQ__mh_w_4227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4227_equation_0, values = (var_40639_cast_fp16, var_40280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4227_cast_fp16")]; + string _SplitHeadsQ__mh_w_4229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4229_equation_0, values = (var_40639_cast_fp16, var_40287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4229_cast_fp16")]; + string _SplitHeadsQ__mh_w_4231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4231_equation_0, values = (var_40639_cast_fp16, var_40294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4231_cast_fp16")]; + string _SplitHeadsQ__mh_w_4233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4233_equation_0, values = (var_40643_cast_fp16, var_40301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4233_cast_fp16")]; + string _SplitHeadsQ__mh_w_4235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4235_equation_0, values = (var_40643_cast_fp16, var_40308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4235_cast_fp16")]; + string _SplitHeadsQ__mh_w_4237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4237_equation_0, values = (var_40643_cast_fp16, var_40315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4237_cast_fp16")]; + string _SplitHeadsQ__mh_w_4239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4239_equation_0, values = (var_40643_cast_fp16, var_40322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4239_cast_fp16")]; + string _SplitHeadsQ__mh_w_4241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4241_equation_0, values = (var_40647_cast_fp16, var_40329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4241_cast_fp16")]; + string _SplitHeadsQ__mh_w_4243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4243_equation_0, values = (var_40647_cast_fp16, var_40336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4243_cast_fp16")]; + string _SplitHeadsQ__mh_w_4245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4245_equation_0, values = (var_40647_cast_fp16, var_40343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4245_cast_fp16")]; + string _SplitHeadsQ__mh_w_4247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4247_equation_0, values = (var_40647_cast_fp16, var_40350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4247_cast_fp16")]; + string _SplitHeadsQ__mh_w_4249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4249_equation_0, values = (var_40651_cast_fp16, var_40357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4249_cast_fp16")]; + string _SplitHeadsQ__mh_w_4251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4251_equation_0, values = (var_40651_cast_fp16, var_40364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4251_cast_fp16")]; + string _SplitHeadsQ__mh_w_4253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4253_equation_0, values = (var_40651_cast_fp16, var_40371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4253_cast_fp16")]; + string _SplitHeadsQ__mh_w_4255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4255_equation_0, values = (var_40651_cast_fp16, var_40378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4255_cast_fp16")]; + string _SplitHeadsQ__mh_w_4257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4257_equation_0, values = (var_40655_cast_fp16, var_40385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4257_cast_fp16")]; + string _SplitHeadsQ__mh_w_4259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4259_equation_0, values = (var_40655_cast_fp16, var_40392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4259_cast_fp16")]; + string _SplitHeadsQ__mh_w_4261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4261_equation_0, values = (var_40655_cast_fp16, var_40399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4261_cast_fp16")]; + string _SplitHeadsQ__mh_w_4263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4263_equation_0, values = (var_40655_cast_fp16, var_40406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4263_cast_fp16")]; + string _SplitHeadsQ__mh_w_4265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4265_equation_0, values = (var_40659_cast_fp16, var_40413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4265_cast_fp16")]; + string _SplitHeadsQ__mh_w_4267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4267_equation_0, values = (var_40659_cast_fp16, var_40420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4267_cast_fp16")]; + string _SplitHeadsQ__mh_w_4269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4269_equation_0, values = (var_40659_cast_fp16, var_40427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4269_cast_fp16")]; + string _SplitHeadsQ__mh_w_4271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4271_equation_0, values = (var_40659_cast_fp16, var_40434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4271_cast_fp16")]; + string _SplitHeadsQ__mh_w_4273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4273_equation_0, values = (var_40663_cast_fp16, var_40441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4273_cast_fp16")]; + string _SplitHeadsQ__mh_w_4275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4275_equation_0, values = (var_40663_cast_fp16, var_40448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4275_cast_fp16")]; + string _SplitHeadsQ__mh_w_4277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4277_equation_0, values = (var_40663_cast_fp16, var_40455_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4277_cast_fp16")]; + string _SplitHeadsQ__mh_w_4279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4279_equation_0, values = (var_40663_cast_fp16, var_40462_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4279_cast_fp16")]; + string _SplitHeadsQ__mh_w_4281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4281_equation_0, values = (var_40667_cast_fp16, var_40469_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4281_cast_fp16")]; + string _SplitHeadsQ__mh_w_4283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4283_equation_0, values = (var_40667_cast_fp16, var_40476_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4283_cast_fp16")]; + string _SplitHeadsQ__mh_w_4285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4285_equation_0, values = (var_40667_cast_fp16, var_40483_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4285_cast_fp16")]; + string _SplitHeadsQ__mh_w_4287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4287_equation_0, values = (var_40667_cast_fp16, var_40490_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4287_cast_fp16")]; + string _SplitHeadsQ__mh_w_4289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4289_equation_0, values = (var_40671_cast_fp16, var_40497_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4289_cast_fp16")]; + string _SplitHeadsQ__mh_w_4291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4291_equation_0, values = (var_40671_cast_fp16, var_40504_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4291_cast_fp16")]; + string _SplitHeadsQ__mh_w_4293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4293_equation_0, values = (var_40671_cast_fp16, var_40511_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4293_cast_fp16")]; + string _SplitHeadsQ__mh_w_4295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4295_equation_0, values = (var_40671_cast_fp16, var_40518_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4295_cast_fp16")]; + string _SplitHeadsQ__mh_w_4297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4297_equation_0, values = (var_40675_cast_fp16, var_40525_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4297_cast_fp16")]; + string _SplitHeadsQ__mh_w_4299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4299_equation_0, values = (var_40675_cast_fp16, var_40532_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4299_cast_fp16")]; + string _SplitHeadsQ__mh_w_4301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4301_equation_0, values = (var_40675_cast_fp16, var_40539_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4301_cast_fp16")]; + string _SplitHeadsQ__mh_w_4303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4303_equation_0, values = (var_40675_cast_fp16, var_40546_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4303_cast_fp16")]; + string _SplitHeadsQ__mh_w_4305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4305_equation_0, values = (var_40679_cast_fp16, var_40553_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4305_cast_fp16")]; + string _SplitHeadsQ__mh_w_4307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4307_equation_0, values = (var_40679_cast_fp16, var_40560_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4307_cast_fp16")]; + string _SplitHeadsQ__mh_w_4309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4309_equation_0, values = (var_40679_cast_fp16, var_40567_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4309_cast_fp16")]; + string _SplitHeadsQ__mh_w_4311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4311_equation_0, values = (var_40679_cast_fp16, var_40574_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4311_cast_fp16")]; + string _SplitHeadsQ__mh_w_4313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4313_equation_0, values = (var_40683_cast_fp16, var_40581_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4313_cast_fp16")]; + string _SplitHeadsQ__mh_w_4315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4315_equation_0, values = (var_40683_cast_fp16, var_40588_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4315_cast_fp16")]; + string _SplitHeadsQ__mh_w_4317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4317_equation_0, values = (var_40683_cast_fp16, var_40595_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4317_cast_fp16")]; + string _SplitHeadsQ__mh_w_4319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4319_equation_0, values = (var_40683_cast_fp16, var_40602_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4319_cast_fp16")]; + fp16 var_40924_to_fp16 = const()[name = string("op_40924_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4161_cast_fp16, y = var_40924_to_fp16)[name = string("aw_chunk_4161_cast_fp16")]; + fp16 var_40926_to_fp16 = const()[name = string("op_40926_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4163_cast_fp16, y = var_40926_to_fp16)[name = string("aw_chunk_4163_cast_fp16")]; + fp16 var_40928_to_fp16 = const()[name = string("op_40928_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4165_cast_fp16, y = var_40928_to_fp16)[name = string("aw_chunk_4165_cast_fp16")]; + fp16 var_40930_to_fp16 = const()[name = string("op_40930_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4167_cast_fp16, y = var_40930_to_fp16)[name = string("aw_chunk_4167_cast_fp16")]; + fp16 var_40932_to_fp16 = const()[name = string("op_40932_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4169_cast_fp16, y = var_40932_to_fp16)[name = string("aw_chunk_4169_cast_fp16")]; + fp16 var_40934_to_fp16 = const()[name = string("op_40934_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4171_cast_fp16, y = var_40934_to_fp16)[name = string("aw_chunk_4171_cast_fp16")]; + fp16 var_40936_to_fp16 = const()[name = string("op_40936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4173_cast_fp16, y = var_40936_to_fp16)[name = string("aw_chunk_4173_cast_fp16")]; + fp16 var_40938_to_fp16 = const()[name = string("op_40938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4175_cast_fp16, y = var_40938_to_fp16)[name = string("aw_chunk_4175_cast_fp16")]; + fp16 var_40940_to_fp16 = const()[name = string("op_40940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4177_cast_fp16, y = var_40940_to_fp16)[name = string("aw_chunk_4177_cast_fp16")]; + fp16 var_40942_to_fp16 = const()[name = string("op_40942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4179_cast_fp16, y = var_40942_to_fp16)[name = string("aw_chunk_4179_cast_fp16")]; + fp16 var_40944_to_fp16 = const()[name = string("op_40944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4181_cast_fp16, y = var_40944_to_fp16)[name = string("aw_chunk_4181_cast_fp16")]; + fp16 var_40946_to_fp16 = const()[name = string("op_40946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4183_cast_fp16, y = var_40946_to_fp16)[name = string("aw_chunk_4183_cast_fp16")]; + fp16 var_40948_to_fp16 = const()[name = string("op_40948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4185_cast_fp16, y = var_40948_to_fp16)[name = string("aw_chunk_4185_cast_fp16")]; + fp16 var_40950_to_fp16 = const()[name = string("op_40950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4187_cast_fp16, y = var_40950_to_fp16)[name = string("aw_chunk_4187_cast_fp16")]; + fp16 var_40952_to_fp16 = const()[name = string("op_40952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4189_cast_fp16, y = var_40952_to_fp16)[name = string("aw_chunk_4189_cast_fp16")]; + fp16 var_40954_to_fp16 = const()[name = string("op_40954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4191_cast_fp16, y = var_40954_to_fp16)[name = string("aw_chunk_4191_cast_fp16")]; + fp16 var_40956_to_fp16 = const()[name = string("op_40956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4193_cast_fp16, y = var_40956_to_fp16)[name = string("aw_chunk_4193_cast_fp16")]; + fp16 var_40958_to_fp16 = const()[name = string("op_40958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4195_cast_fp16, y = var_40958_to_fp16)[name = string("aw_chunk_4195_cast_fp16")]; + fp16 var_40960_to_fp16 = const()[name = string("op_40960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4197_cast_fp16, y = var_40960_to_fp16)[name = string("aw_chunk_4197_cast_fp16")]; + fp16 var_40962_to_fp16 = const()[name = string("op_40962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4199_cast_fp16, y = var_40962_to_fp16)[name = string("aw_chunk_4199_cast_fp16")]; + fp16 var_40964_to_fp16 = const()[name = string("op_40964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4201_cast_fp16, y = var_40964_to_fp16)[name = string("aw_chunk_4201_cast_fp16")]; + fp16 var_40966_to_fp16 = const()[name = string("op_40966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4203_cast_fp16, y = var_40966_to_fp16)[name = string("aw_chunk_4203_cast_fp16")]; + fp16 var_40968_to_fp16 = const()[name = string("op_40968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4205_cast_fp16, y = var_40968_to_fp16)[name = string("aw_chunk_4205_cast_fp16")]; + fp16 var_40970_to_fp16 = const()[name = string("op_40970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4207_cast_fp16, y = var_40970_to_fp16)[name = string("aw_chunk_4207_cast_fp16")]; + fp16 var_40972_to_fp16 = const()[name = string("op_40972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4209_cast_fp16, y = var_40972_to_fp16)[name = string("aw_chunk_4209_cast_fp16")]; + fp16 var_40974_to_fp16 = const()[name = string("op_40974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4211_cast_fp16, y = var_40974_to_fp16)[name = string("aw_chunk_4211_cast_fp16")]; + fp16 var_40976_to_fp16 = const()[name = string("op_40976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4213_cast_fp16, y = var_40976_to_fp16)[name = string("aw_chunk_4213_cast_fp16")]; + fp16 var_40978_to_fp16 = const()[name = string("op_40978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4215_cast_fp16, y = var_40978_to_fp16)[name = string("aw_chunk_4215_cast_fp16")]; + fp16 var_40980_to_fp16 = const()[name = string("op_40980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4217_cast_fp16, y = var_40980_to_fp16)[name = string("aw_chunk_4217_cast_fp16")]; + fp16 var_40982_to_fp16 = const()[name = string("op_40982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4219_cast_fp16, y = var_40982_to_fp16)[name = string("aw_chunk_4219_cast_fp16")]; + fp16 var_40984_to_fp16 = const()[name = string("op_40984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4221_cast_fp16, y = var_40984_to_fp16)[name = string("aw_chunk_4221_cast_fp16")]; + fp16 var_40986_to_fp16 = const()[name = string("op_40986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4223_cast_fp16, y = var_40986_to_fp16)[name = string("aw_chunk_4223_cast_fp16")]; + fp16 var_40988_to_fp16 = const()[name = string("op_40988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4225_cast_fp16, y = var_40988_to_fp16)[name = string("aw_chunk_4225_cast_fp16")]; + fp16 var_40990_to_fp16 = const()[name = string("op_40990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4227_cast_fp16, y = var_40990_to_fp16)[name = string("aw_chunk_4227_cast_fp16")]; + fp16 var_40992_to_fp16 = const()[name = string("op_40992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4229_cast_fp16, y = var_40992_to_fp16)[name = string("aw_chunk_4229_cast_fp16")]; + fp16 var_40994_to_fp16 = const()[name = string("op_40994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4231_cast_fp16, y = var_40994_to_fp16)[name = string("aw_chunk_4231_cast_fp16")]; + fp16 var_40996_to_fp16 = const()[name = string("op_40996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4233_cast_fp16, y = var_40996_to_fp16)[name = string("aw_chunk_4233_cast_fp16")]; + fp16 var_40998_to_fp16 = const()[name = string("op_40998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4235_cast_fp16, y = var_40998_to_fp16)[name = string("aw_chunk_4235_cast_fp16")]; + fp16 var_41000_to_fp16 = const()[name = string("op_41000_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4237_cast_fp16, y = var_41000_to_fp16)[name = string("aw_chunk_4237_cast_fp16")]; + fp16 var_41002_to_fp16 = const()[name = string("op_41002_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4239_cast_fp16, y = var_41002_to_fp16)[name = string("aw_chunk_4239_cast_fp16")]; + fp16 var_41004_to_fp16 = const()[name = string("op_41004_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4241_cast_fp16, y = var_41004_to_fp16)[name = string("aw_chunk_4241_cast_fp16")]; + fp16 var_41006_to_fp16 = const()[name = string("op_41006_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4243_cast_fp16, y = var_41006_to_fp16)[name = string("aw_chunk_4243_cast_fp16")]; + fp16 var_41008_to_fp16 = const()[name = string("op_41008_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4245_cast_fp16, y = var_41008_to_fp16)[name = string("aw_chunk_4245_cast_fp16")]; + fp16 var_41010_to_fp16 = const()[name = string("op_41010_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4247_cast_fp16, y = var_41010_to_fp16)[name = string("aw_chunk_4247_cast_fp16")]; + fp16 var_41012_to_fp16 = const()[name = string("op_41012_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4249_cast_fp16, y = var_41012_to_fp16)[name = string("aw_chunk_4249_cast_fp16")]; + fp16 var_41014_to_fp16 = const()[name = string("op_41014_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4251_cast_fp16, y = var_41014_to_fp16)[name = string("aw_chunk_4251_cast_fp16")]; + fp16 var_41016_to_fp16 = const()[name = string("op_41016_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4253_cast_fp16, y = var_41016_to_fp16)[name = string("aw_chunk_4253_cast_fp16")]; + fp16 var_41018_to_fp16 = const()[name = string("op_41018_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4255_cast_fp16, y = var_41018_to_fp16)[name = string("aw_chunk_4255_cast_fp16")]; + fp16 var_41020_to_fp16 = const()[name = string("op_41020_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4257_cast_fp16, y = var_41020_to_fp16)[name = string("aw_chunk_4257_cast_fp16")]; + fp16 var_41022_to_fp16 = const()[name = string("op_41022_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4259_cast_fp16, y = var_41022_to_fp16)[name = string("aw_chunk_4259_cast_fp16")]; + fp16 var_41024_to_fp16 = const()[name = string("op_41024_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4261_cast_fp16, y = var_41024_to_fp16)[name = string("aw_chunk_4261_cast_fp16")]; + fp16 var_41026_to_fp16 = const()[name = string("op_41026_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4263_cast_fp16, y = var_41026_to_fp16)[name = string("aw_chunk_4263_cast_fp16")]; + fp16 var_41028_to_fp16 = const()[name = string("op_41028_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4265_cast_fp16, y = var_41028_to_fp16)[name = string("aw_chunk_4265_cast_fp16")]; + fp16 var_41030_to_fp16 = const()[name = string("op_41030_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4267_cast_fp16, y = var_41030_to_fp16)[name = string("aw_chunk_4267_cast_fp16")]; + fp16 var_41032_to_fp16 = const()[name = string("op_41032_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4269_cast_fp16, y = var_41032_to_fp16)[name = string("aw_chunk_4269_cast_fp16")]; + fp16 var_41034_to_fp16 = const()[name = string("op_41034_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4271_cast_fp16, y = var_41034_to_fp16)[name = string("aw_chunk_4271_cast_fp16")]; + fp16 var_41036_to_fp16 = const()[name = string("op_41036_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4273_cast_fp16, y = var_41036_to_fp16)[name = string("aw_chunk_4273_cast_fp16")]; + fp16 var_41038_to_fp16 = const()[name = string("op_41038_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4275_cast_fp16, y = var_41038_to_fp16)[name = string("aw_chunk_4275_cast_fp16")]; + fp16 var_41040_to_fp16 = const()[name = string("op_41040_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4277_cast_fp16, y = var_41040_to_fp16)[name = string("aw_chunk_4277_cast_fp16")]; + fp16 var_41042_to_fp16 = const()[name = string("op_41042_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4279_cast_fp16, y = var_41042_to_fp16)[name = string("aw_chunk_4279_cast_fp16")]; + fp16 var_41044_to_fp16 = const()[name = string("op_41044_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4281_cast_fp16, y = var_41044_to_fp16)[name = string("aw_chunk_4281_cast_fp16")]; + fp16 var_41046_to_fp16 = const()[name = string("op_41046_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4283_cast_fp16, y = var_41046_to_fp16)[name = string("aw_chunk_4283_cast_fp16")]; + fp16 var_41048_to_fp16 = const()[name = string("op_41048_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4285_cast_fp16, y = var_41048_to_fp16)[name = string("aw_chunk_4285_cast_fp16")]; + fp16 var_41050_to_fp16 = const()[name = string("op_41050_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4287_cast_fp16, y = var_41050_to_fp16)[name = string("aw_chunk_4287_cast_fp16")]; + fp16 var_41052_to_fp16 = const()[name = string("op_41052_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4289_cast_fp16, y = var_41052_to_fp16)[name = string("aw_chunk_4289_cast_fp16")]; + fp16 var_41054_to_fp16 = const()[name = string("op_41054_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4291_cast_fp16, y = var_41054_to_fp16)[name = string("aw_chunk_4291_cast_fp16")]; + fp16 var_41056_to_fp16 = const()[name = string("op_41056_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4293_cast_fp16, y = var_41056_to_fp16)[name = string("aw_chunk_4293_cast_fp16")]; + fp16 var_41058_to_fp16 = const()[name = string("op_41058_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4295_cast_fp16, y = var_41058_to_fp16)[name = string("aw_chunk_4295_cast_fp16")]; + fp16 var_41060_to_fp16 = const()[name = string("op_41060_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4297_cast_fp16, y = var_41060_to_fp16)[name = string("aw_chunk_4297_cast_fp16")]; + fp16 var_41062_to_fp16 = const()[name = string("op_41062_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4299_cast_fp16, y = var_41062_to_fp16)[name = string("aw_chunk_4299_cast_fp16")]; + fp16 var_41064_to_fp16 = const()[name = string("op_41064_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4301_cast_fp16, y = var_41064_to_fp16)[name = string("aw_chunk_4301_cast_fp16")]; + fp16 var_41066_to_fp16 = const()[name = string("op_41066_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4303_cast_fp16, y = var_41066_to_fp16)[name = string("aw_chunk_4303_cast_fp16")]; + fp16 var_41068_to_fp16 = const()[name = string("op_41068_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4305_cast_fp16, y = var_41068_to_fp16)[name = string("aw_chunk_4305_cast_fp16")]; + fp16 var_41070_to_fp16 = const()[name = string("op_41070_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4307_cast_fp16, y = var_41070_to_fp16)[name = string("aw_chunk_4307_cast_fp16")]; + fp16 var_41072_to_fp16 = const()[name = string("op_41072_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4309_cast_fp16, y = var_41072_to_fp16)[name = string("aw_chunk_4309_cast_fp16")]; + fp16 var_41074_to_fp16 = const()[name = string("op_41074_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4311_cast_fp16, y = var_41074_to_fp16)[name = string("aw_chunk_4311_cast_fp16")]; + fp16 var_41076_to_fp16 = const()[name = string("op_41076_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4313_cast_fp16, y = var_41076_to_fp16)[name = string("aw_chunk_4313_cast_fp16")]; + fp16 var_41078_to_fp16 = const()[name = string("op_41078_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4315_cast_fp16, y = var_41078_to_fp16)[name = string("aw_chunk_4315_cast_fp16")]; + fp16 var_41080_to_fp16 = const()[name = string("op_41080_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4317_cast_fp16, y = var_41080_to_fp16)[name = string("aw_chunk_4317_cast_fp16")]; + fp16 var_41082_to_fp16 = const()[name = string("op_41082_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4319_cast_fp16, y = var_41082_to_fp16)[name = string("aw_chunk_4319_cast_fp16")]; + tensor var_41084_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4161_cast_fp16)[name = string("op_41084_cast_fp16")]; + tensor var_41085_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4163_cast_fp16)[name = string("op_41085_cast_fp16")]; + tensor var_41086_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4165_cast_fp16)[name = string("op_41086_cast_fp16")]; + tensor var_41087_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4167_cast_fp16)[name = string("op_41087_cast_fp16")]; + tensor var_41088_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4169_cast_fp16)[name = string("op_41088_cast_fp16")]; + tensor var_41089_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4171_cast_fp16)[name = string("op_41089_cast_fp16")]; + tensor var_41090_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4173_cast_fp16)[name = string("op_41090_cast_fp16")]; + tensor var_41091_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4175_cast_fp16)[name = string("op_41091_cast_fp16")]; + tensor var_41092_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4177_cast_fp16)[name = string("op_41092_cast_fp16")]; + tensor var_41093_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4179_cast_fp16)[name = string("op_41093_cast_fp16")]; + tensor var_41094_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4181_cast_fp16)[name = string("op_41094_cast_fp16")]; + tensor var_41095_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4183_cast_fp16)[name = string("op_41095_cast_fp16")]; + tensor var_41096_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4185_cast_fp16)[name = string("op_41096_cast_fp16")]; + tensor var_41097_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4187_cast_fp16)[name = string("op_41097_cast_fp16")]; + tensor var_41098_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4189_cast_fp16)[name = string("op_41098_cast_fp16")]; + tensor var_41099_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4191_cast_fp16)[name = string("op_41099_cast_fp16")]; + tensor var_41100_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4193_cast_fp16)[name = string("op_41100_cast_fp16")]; + tensor var_41101_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4195_cast_fp16)[name = string("op_41101_cast_fp16")]; + tensor var_41102_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4197_cast_fp16)[name = string("op_41102_cast_fp16")]; + tensor var_41103_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4199_cast_fp16)[name = string("op_41103_cast_fp16")]; + tensor var_41104_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4201_cast_fp16)[name = string("op_41104_cast_fp16")]; + tensor var_41105_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4203_cast_fp16)[name = string("op_41105_cast_fp16")]; + tensor var_41106_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4205_cast_fp16)[name = string("op_41106_cast_fp16")]; + tensor var_41107_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4207_cast_fp16)[name = string("op_41107_cast_fp16")]; + tensor var_41108_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4209_cast_fp16)[name = string("op_41108_cast_fp16")]; + tensor var_41109_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4211_cast_fp16)[name = string("op_41109_cast_fp16")]; + tensor var_41110_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4213_cast_fp16)[name = string("op_41110_cast_fp16")]; + tensor var_41111_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4215_cast_fp16)[name = string("op_41111_cast_fp16")]; + tensor var_41112_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4217_cast_fp16)[name = string("op_41112_cast_fp16")]; + tensor var_41113_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4219_cast_fp16)[name = string("op_41113_cast_fp16")]; + tensor var_41114_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4221_cast_fp16)[name = string("op_41114_cast_fp16")]; + tensor var_41115_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4223_cast_fp16)[name = string("op_41115_cast_fp16")]; + tensor var_41116_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4225_cast_fp16)[name = string("op_41116_cast_fp16")]; + tensor var_41117_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4227_cast_fp16)[name = string("op_41117_cast_fp16")]; + tensor var_41118_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4229_cast_fp16)[name = string("op_41118_cast_fp16")]; + tensor var_41119_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4231_cast_fp16)[name = string("op_41119_cast_fp16")]; + tensor var_41120_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4233_cast_fp16)[name = string("op_41120_cast_fp16")]; + tensor var_41121_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4235_cast_fp16)[name = string("op_41121_cast_fp16")]; + tensor var_41122_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4237_cast_fp16)[name = string("op_41122_cast_fp16")]; + tensor var_41123_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4239_cast_fp16)[name = string("op_41123_cast_fp16")]; + tensor var_41124_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4241_cast_fp16)[name = string("op_41124_cast_fp16")]; + tensor var_41125_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4243_cast_fp16)[name = string("op_41125_cast_fp16")]; + tensor var_41126_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4245_cast_fp16)[name = string("op_41126_cast_fp16")]; + tensor var_41127_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4247_cast_fp16)[name = string("op_41127_cast_fp16")]; + tensor var_41128_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4249_cast_fp16)[name = string("op_41128_cast_fp16")]; + tensor var_41129_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4251_cast_fp16)[name = string("op_41129_cast_fp16")]; + tensor var_41130_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4253_cast_fp16)[name = string("op_41130_cast_fp16")]; + tensor var_41131_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4255_cast_fp16)[name = string("op_41131_cast_fp16")]; + tensor var_41132_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4257_cast_fp16)[name = string("op_41132_cast_fp16")]; + tensor var_41133_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4259_cast_fp16)[name = string("op_41133_cast_fp16")]; + tensor var_41134_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4261_cast_fp16)[name = string("op_41134_cast_fp16")]; + tensor var_41135_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4263_cast_fp16)[name = string("op_41135_cast_fp16")]; + tensor var_41136_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4265_cast_fp16)[name = string("op_41136_cast_fp16")]; + tensor var_41137_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4267_cast_fp16)[name = string("op_41137_cast_fp16")]; + tensor var_41138_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4269_cast_fp16)[name = string("op_41138_cast_fp16")]; + tensor var_41139_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4271_cast_fp16)[name = string("op_41139_cast_fp16")]; + tensor var_41140_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4273_cast_fp16)[name = string("op_41140_cast_fp16")]; + tensor var_41141_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4275_cast_fp16)[name = string("op_41141_cast_fp16")]; + tensor var_41142_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4277_cast_fp16)[name = string("op_41142_cast_fp16")]; + tensor var_41143_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4279_cast_fp16)[name = string("op_41143_cast_fp16")]; + tensor var_41144_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4281_cast_fp16)[name = string("op_41144_cast_fp16")]; + tensor var_41145_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4283_cast_fp16)[name = string("op_41145_cast_fp16")]; + tensor var_41146_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4285_cast_fp16)[name = string("op_41146_cast_fp16")]; + tensor var_41147_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4287_cast_fp16)[name = string("op_41147_cast_fp16")]; + tensor var_41148_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4289_cast_fp16)[name = string("op_41148_cast_fp16")]; + tensor var_41149_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4291_cast_fp16)[name = string("op_41149_cast_fp16")]; + tensor var_41150_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4293_cast_fp16)[name = string("op_41150_cast_fp16")]; + tensor var_41151_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4295_cast_fp16)[name = string("op_41151_cast_fp16")]; + tensor var_41152_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4297_cast_fp16)[name = string("op_41152_cast_fp16")]; + tensor var_41153_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4299_cast_fp16)[name = string("op_41153_cast_fp16")]; + tensor var_41154_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4301_cast_fp16)[name = string("op_41154_cast_fp16")]; + tensor var_41155_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4303_cast_fp16)[name = string("op_41155_cast_fp16")]; + tensor var_41156_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4305_cast_fp16)[name = string("op_41156_cast_fp16")]; + tensor var_41157_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4307_cast_fp16)[name = string("op_41157_cast_fp16")]; + tensor var_41158_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4309_cast_fp16)[name = string("op_41158_cast_fp16")]; + tensor var_41159_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4311_cast_fp16)[name = string("op_41159_cast_fp16")]; + tensor var_41160_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4313_cast_fp16)[name = string("op_41160_cast_fp16")]; + tensor var_41161_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4315_cast_fp16)[name = string("op_41161_cast_fp16")]; + tensor var_41162_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4317_cast_fp16)[name = string("op_41162_cast_fp16")]; + tensor var_41163_cast_fp16 = softmax(axis = var_39909, x = aw_chunk_4319_cast_fp16)[name = string("op_41163_cast_fp16")]; + string var_41165_equation_0 = const()[name = string("op_41165_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41165_cast_fp16 = einsum(equation = var_41165_equation_0, values = (var_40685_cast_fp16, var_41084_cast_fp16))[name = string("op_41165_cast_fp16")]; + string var_41167_equation_0 = const()[name = string("op_41167_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41167_cast_fp16 = einsum(equation = var_41167_equation_0, values = (var_40685_cast_fp16, var_41085_cast_fp16))[name = string("op_41167_cast_fp16")]; + string var_41169_equation_0 = const()[name = string("op_41169_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41169_cast_fp16 = einsum(equation = var_41169_equation_0, values = (var_40685_cast_fp16, var_41086_cast_fp16))[name = string("op_41169_cast_fp16")]; + string var_41171_equation_0 = const()[name = string("op_41171_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41171_cast_fp16 = einsum(equation = var_41171_equation_0, values = (var_40685_cast_fp16, var_41087_cast_fp16))[name = string("op_41171_cast_fp16")]; + string var_41173_equation_0 = const()[name = string("op_41173_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41173_cast_fp16 = einsum(equation = var_41173_equation_0, values = (var_40689_cast_fp16, var_41088_cast_fp16))[name = string("op_41173_cast_fp16")]; + string var_41175_equation_0 = const()[name = string("op_41175_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41175_cast_fp16 = einsum(equation = var_41175_equation_0, values = (var_40689_cast_fp16, var_41089_cast_fp16))[name = string("op_41175_cast_fp16")]; + string var_41177_equation_0 = const()[name = string("op_41177_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41177_cast_fp16 = einsum(equation = var_41177_equation_0, values = (var_40689_cast_fp16, var_41090_cast_fp16))[name = string("op_41177_cast_fp16")]; + string var_41179_equation_0 = const()[name = string("op_41179_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41179_cast_fp16 = einsum(equation = var_41179_equation_0, values = (var_40689_cast_fp16, var_41091_cast_fp16))[name = string("op_41179_cast_fp16")]; + string var_41181_equation_0 = const()[name = string("op_41181_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41181_cast_fp16 = einsum(equation = var_41181_equation_0, values = (var_40693_cast_fp16, var_41092_cast_fp16))[name = string("op_41181_cast_fp16")]; + string var_41183_equation_0 = const()[name = string("op_41183_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41183_cast_fp16 = einsum(equation = var_41183_equation_0, values = (var_40693_cast_fp16, var_41093_cast_fp16))[name = string("op_41183_cast_fp16")]; + string var_41185_equation_0 = const()[name = string("op_41185_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41185_cast_fp16 = einsum(equation = var_41185_equation_0, values = (var_40693_cast_fp16, var_41094_cast_fp16))[name = string("op_41185_cast_fp16")]; + string var_41187_equation_0 = const()[name = string("op_41187_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41187_cast_fp16 = einsum(equation = var_41187_equation_0, values = (var_40693_cast_fp16, var_41095_cast_fp16))[name = string("op_41187_cast_fp16")]; + string var_41189_equation_0 = const()[name = string("op_41189_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41189_cast_fp16 = einsum(equation = var_41189_equation_0, values = (var_40697_cast_fp16, var_41096_cast_fp16))[name = string("op_41189_cast_fp16")]; + string var_41191_equation_0 = const()[name = string("op_41191_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41191_cast_fp16 = einsum(equation = var_41191_equation_0, values = (var_40697_cast_fp16, var_41097_cast_fp16))[name = string("op_41191_cast_fp16")]; + string var_41193_equation_0 = const()[name = string("op_41193_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41193_cast_fp16 = einsum(equation = var_41193_equation_0, values = (var_40697_cast_fp16, var_41098_cast_fp16))[name = string("op_41193_cast_fp16")]; + string var_41195_equation_0 = const()[name = string("op_41195_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41195_cast_fp16 = einsum(equation = var_41195_equation_0, values = (var_40697_cast_fp16, var_41099_cast_fp16))[name = string("op_41195_cast_fp16")]; + string var_41197_equation_0 = const()[name = string("op_41197_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41197_cast_fp16 = einsum(equation = var_41197_equation_0, values = (var_40701_cast_fp16, var_41100_cast_fp16))[name = string("op_41197_cast_fp16")]; + string var_41199_equation_0 = const()[name = string("op_41199_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41199_cast_fp16 = einsum(equation = var_41199_equation_0, values = (var_40701_cast_fp16, var_41101_cast_fp16))[name = string("op_41199_cast_fp16")]; + string var_41201_equation_0 = const()[name = string("op_41201_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41201_cast_fp16 = einsum(equation = var_41201_equation_0, values = (var_40701_cast_fp16, var_41102_cast_fp16))[name = string("op_41201_cast_fp16")]; + string var_41203_equation_0 = const()[name = string("op_41203_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41203_cast_fp16 = einsum(equation = var_41203_equation_0, values = (var_40701_cast_fp16, var_41103_cast_fp16))[name = string("op_41203_cast_fp16")]; + string var_41205_equation_0 = const()[name = string("op_41205_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41205_cast_fp16 = einsum(equation = var_41205_equation_0, values = (var_40705_cast_fp16, var_41104_cast_fp16))[name = string("op_41205_cast_fp16")]; + string var_41207_equation_0 = const()[name = string("op_41207_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41207_cast_fp16 = einsum(equation = var_41207_equation_0, values = (var_40705_cast_fp16, var_41105_cast_fp16))[name = string("op_41207_cast_fp16")]; + string var_41209_equation_0 = const()[name = string("op_41209_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41209_cast_fp16 = einsum(equation = var_41209_equation_0, values = (var_40705_cast_fp16, var_41106_cast_fp16))[name = string("op_41209_cast_fp16")]; + string var_41211_equation_0 = const()[name = string("op_41211_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41211_cast_fp16 = einsum(equation = var_41211_equation_0, values = (var_40705_cast_fp16, var_41107_cast_fp16))[name = string("op_41211_cast_fp16")]; + string var_41213_equation_0 = const()[name = string("op_41213_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41213_cast_fp16 = einsum(equation = var_41213_equation_0, values = (var_40709_cast_fp16, var_41108_cast_fp16))[name = string("op_41213_cast_fp16")]; + string var_41215_equation_0 = const()[name = string("op_41215_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41215_cast_fp16 = einsum(equation = var_41215_equation_0, values = (var_40709_cast_fp16, var_41109_cast_fp16))[name = string("op_41215_cast_fp16")]; + string var_41217_equation_0 = const()[name = string("op_41217_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41217_cast_fp16 = einsum(equation = var_41217_equation_0, values = (var_40709_cast_fp16, var_41110_cast_fp16))[name = string("op_41217_cast_fp16")]; + string var_41219_equation_0 = const()[name = string("op_41219_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41219_cast_fp16 = einsum(equation = var_41219_equation_0, values = (var_40709_cast_fp16, var_41111_cast_fp16))[name = string("op_41219_cast_fp16")]; + string var_41221_equation_0 = const()[name = string("op_41221_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41221_cast_fp16 = einsum(equation = var_41221_equation_0, values = (var_40713_cast_fp16, var_41112_cast_fp16))[name = string("op_41221_cast_fp16")]; + string var_41223_equation_0 = const()[name = string("op_41223_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41223_cast_fp16 = einsum(equation = var_41223_equation_0, values = (var_40713_cast_fp16, var_41113_cast_fp16))[name = string("op_41223_cast_fp16")]; + string var_41225_equation_0 = const()[name = string("op_41225_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41225_cast_fp16 = einsum(equation = var_41225_equation_0, values = (var_40713_cast_fp16, var_41114_cast_fp16))[name = string("op_41225_cast_fp16")]; + string var_41227_equation_0 = const()[name = string("op_41227_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41227_cast_fp16 = einsum(equation = var_41227_equation_0, values = (var_40713_cast_fp16, var_41115_cast_fp16))[name = string("op_41227_cast_fp16")]; + string var_41229_equation_0 = const()[name = string("op_41229_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41229_cast_fp16 = einsum(equation = var_41229_equation_0, values = (var_40717_cast_fp16, var_41116_cast_fp16))[name = string("op_41229_cast_fp16")]; + string var_41231_equation_0 = const()[name = string("op_41231_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41231_cast_fp16 = einsum(equation = var_41231_equation_0, values = (var_40717_cast_fp16, var_41117_cast_fp16))[name = string("op_41231_cast_fp16")]; + string var_41233_equation_0 = const()[name = string("op_41233_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41233_cast_fp16 = einsum(equation = var_41233_equation_0, values = (var_40717_cast_fp16, var_41118_cast_fp16))[name = string("op_41233_cast_fp16")]; + string var_41235_equation_0 = const()[name = string("op_41235_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41235_cast_fp16 = einsum(equation = var_41235_equation_0, values = (var_40717_cast_fp16, var_41119_cast_fp16))[name = string("op_41235_cast_fp16")]; + string var_41237_equation_0 = const()[name = string("op_41237_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41237_cast_fp16 = einsum(equation = var_41237_equation_0, values = (var_40721_cast_fp16, var_41120_cast_fp16))[name = string("op_41237_cast_fp16")]; + string var_41239_equation_0 = const()[name = string("op_41239_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41239_cast_fp16 = einsum(equation = var_41239_equation_0, values = (var_40721_cast_fp16, var_41121_cast_fp16))[name = string("op_41239_cast_fp16")]; + string var_41241_equation_0 = const()[name = string("op_41241_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41241_cast_fp16 = einsum(equation = var_41241_equation_0, values = (var_40721_cast_fp16, var_41122_cast_fp16))[name = string("op_41241_cast_fp16")]; + string var_41243_equation_0 = const()[name = string("op_41243_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41243_cast_fp16 = einsum(equation = var_41243_equation_0, values = (var_40721_cast_fp16, var_41123_cast_fp16))[name = string("op_41243_cast_fp16")]; + string var_41245_equation_0 = const()[name = string("op_41245_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41245_cast_fp16 = einsum(equation = var_41245_equation_0, values = (var_40725_cast_fp16, var_41124_cast_fp16))[name = string("op_41245_cast_fp16")]; + string var_41247_equation_0 = const()[name = string("op_41247_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41247_cast_fp16 = einsum(equation = var_41247_equation_0, values = (var_40725_cast_fp16, var_41125_cast_fp16))[name = string("op_41247_cast_fp16")]; + string var_41249_equation_0 = const()[name = string("op_41249_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41249_cast_fp16 = einsum(equation = var_41249_equation_0, values = (var_40725_cast_fp16, var_41126_cast_fp16))[name = string("op_41249_cast_fp16")]; + string var_41251_equation_0 = const()[name = string("op_41251_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41251_cast_fp16 = einsum(equation = var_41251_equation_0, values = (var_40725_cast_fp16, var_41127_cast_fp16))[name = string("op_41251_cast_fp16")]; + string var_41253_equation_0 = const()[name = string("op_41253_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41253_cast_fp16 = einsum(equation = var_41253_equation_0, values = (var_40729_cast_fp16, var_41128_cast_fp16))[name = string("op_41253_cast_fp16")]; + string var_41255_equation_0 = const()[name = string("op_41255_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41255_cast_fp16 = einsum(equation = var_41255_equation_0, values = (var_40729_cast_fp16, var_41129_cast_fp16))[name = string("op_41255_cast_fp16")]; + string var_41257_equation_0 = const()[name = string("op_41257_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41257_cast_fp16 = einsum(equation = var_41257_equation_0, values = (var_40729_cast_fp16, var_41130_cast_fp16))[name = string("op_41257_cast_fp16")]; + string var_41259_equation_0 = const()[name = string("op_41259_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41259_cast_fp16 = einsum(equation = var_41259_equation_0, values = (var_40729_cast_fp16, var_41131_cast_fp16))[name = string("op_41259_cast_fp16")]; + string var_41261_equation_0 = const()[name = string("op_41261_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41261_cast_fp16 = einsum(equation = var_41261_equation_0, values = (var_40733_cast_fp16, var_41132_cast_fp16))[name = string("op_41261_cast_fp16")]; + string var_41263_equation_0 = const()[name = string("op_41263_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41263_cast_fp16 = einsum(equation = var_41263_equation_0, values = (var_40733_cast_fp16, var_41133_cast_fp16))[name = string("op_41263_cast_fp16")]; + string var_41265_equation_0 = const()[name = string("op_41265_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41265_cast_fp16 = einsum(equation = var_41265_equation_0, values = (var_40733_cast_fp16, var_41134_cast_fp16))[name = string("op_41265_cast_fp16")]; + string var_41267_equation_0 = const()[name = string("op_41267_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41267_cast_fp16 = einsum(equation = var_41267_equation_0, values = (var_40733_cast_fp16, var_41135_cast_fp16))[name = string("op_41267_cast_fp16")]; + string var_41269_equation_0 = const()[name = string("op_41269_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41269_cast_fp16 = einsum(equation = var_41269_equation_0, values = (var_40737_cast_fp16, var_41136_cast_fp16))[name = string("op_41269_cast_fp16")]; + string var_41271_equation_0 = const()[name = string("op_41271_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41271_cast_fp16 = einsum(equation = var_41271_equation_0, values = (var_40737_cast_fp16, var_41137_cast_fp16))[name = string("op_41271_cast_fp16")]; + string var_41273_equation_0 = const()[name = string("op_41273_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41273_cast_fp16 = einsum(equation = var_41273_equation_0, values = (var_40737_cast_fp16, var_41138_cast_fp16))[name = string("op_41273_cast_fp16")]; + string var_41275_equation_0 = const()[name = string("op_41275_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41275_cast_fp16 = einsum(equation = var_41275_equation_0, values = (var_40737_cast_fp16, var_41139_cast_fp16))[name = string("op_41275_cast_fp16")]; + string var_41277_equation_0 = const()[name = string("op_41277_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41277_cast_fp16 = einsum(equation = var_41277_equation_0, values = (var_40741_cast_fp16, var_41140_cast_fp16))[name = string("op_41277_cast_fp16")]; + string var_41279_equation_0 = const()[name = string("op_41279_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41279_cast_fp16 = einsum(equation = var_41279_equation_0, values = (var_40741_cast_fp16, var_41141_cast_fp16))[name = string("op_41279_cast_fp16")]; + string var_41281_equation_0 = const()[name = string("op_41281_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41281_cast_fp16 = einsum(equation = var_41281_equation_0, values = (var_40741_cast_fp16, var_41142_cast_fp16))[name = string("op_41281_cast_fp16")]; + string var_41283_equation_0 = const()[name = string("op_41283_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41283_cast_fp16 = einsum(equation = var_41283_equation_0, values = (var_40741_cast_fp16, var_41143_cast_fp16))[name = string("op_41283_cast_fp16")]; + string var_41285_equation_0 = const()[name = string("op_41285_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41285_cast_fp16 = einsum(equation = var_41285_equation_0, values = (var_40745_cast_fp16, var_41144_cast_fp16))[name = string("op_41285_cast_fp16")]; + string var_41287_equation_0 = const()[name = string("op_41287_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41287_cast_fp16 = einsum(equation = var_41287_equation_0, values = (var_40745_cast_fp16, var_41145_cast_fp16))[name = string("op_41287_cast_fp16")]; + string var_41289_equation_0 = const()[name = string("op_41289_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41289_cast_fp16 = einsum(equation = var_41289_equation_0, values = (var_40745_cast_fp16, var_41146_cast_fp16))[name = string("op_41289_cast_fp16")]; + string var_41291_equation_0 = const()[name = string("op_41291_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41291_cast_fp16 = einsum(equation = var_41291_equation_0, values = (var_40745_cast_fp16, var_41147_cast_fp16))[name = string("op_41291_cast_fp16")]; + string var_41293_equation_0 = const()[name = string("op_41293_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41293_cast_fp16 = einsum(equation = var_41293_equation_0, values = (var_40749_cast_fp16, var_41148_cast_fp16))[name = string("op_41293_cast_fp16")]; + string var_41295_equation_0 = const()[name = string("op_41295_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41295_cast_fp16 = einsum(equation = var_41295_equation_0, values = (var_40749_cast_fp16, var_41149_cast_fp16))[name = string("op_41295_cast_fp16")]; + string var_41297_equation_0 = const()[name = string("op_41297_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41297_cast_fp16 = einsum(equation = var_41297_equation_0, values = (var_40749_cast_fp16, var_41150_cast_fp16))[name = string("op_41297_cast_fp16")]; + string var_41299_equation_0 = const()[name = string("op_41299_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41299_cast_fp16 = einsum(equation = var_41299_equation_0, values = (var_40749_cast_fp16, var_41151_cast_fp16))[name = string("op_41299_cast_fp16")]; + string var_41301_equation_0 = const()[name = string("op_41301_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41301_cast_fp16 = einsum(equation = var_41301_equation_0, values = (var_40753_cast_fp16, var_41152_cast_fp16))[name = string("op_41301_cast_fp16")]; + string var_41303_equation_0 = const()[name = string("op_41303_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41303_cast_fp16 = einsum(equation = var_41303_equation_0, values = (var_40753_cast_fp16, var_41153_cast_fp16))[name = string("op_41303_cast_fp16")]; + string var_41305_equation_0 = const()[name = string("op_41305_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41305_cast_fp16 = einsum(equation = var_41305_equation_0, values = (var_40753_cast_fp16, var_41154_cast_fp16))[name = string("op_41305_cast_fp16")]; + string var_41307_equation_0 = const()[name = string("op_41307_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41307_cast_fp16 = einsum(equation = var_41307_equation_0, values = (var_40753_cast_fp16, var_41155_cast_fp16))[name = string("op_41307_cast_fp16")]; + string var_41309_equation_0 = const()[name = string("op_41309_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41309_cast_fp16 = einsum(equation = var_41309_equation_0, values = (var_40757_cast_fp16, var_41156_cast_fp16))[name = string("op_41309_cast_fp16")]; + string var_41311_equation_0 = const()[name = string("op_41311_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41311_cast_fp16 = einsum(equation = var_41311_equation_0, values = (var_40757_cast_fp16, var_41157_cast_fp16))[name = string("op_41311_cast_fp16")]; + string var_41313_equation_0 = const()[name = string("op_41313_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41313_cast_fp16 = einsum(equation = var_41313_equation_0, values = (var_40757_cast_fp16, var_41158_cast_fp16))[name = string("op_41313_cast_fp16")]; + string var_41315_equation_0 = const()[name = string("op_41315_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41315_cast_fp16 = einsum(equation = var_41315_equation_0, values = (var_40757_cast_fp16, var_41159_cast_fp16))[name = string("op_41315_cast_fp16")]; + string var_41317_equation_0 = const()[name = string("op_41317_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41317_cast_fp16 = einsum(equation = var_41317_equation_0, values = (var_40761_cast_fp16, var_41160_cast_fp16))[name = string("op_41317_cast_fp16")]; + string var_41319_equation_0 = const()[name = string("op_41319_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41319_cast_fp16 = einsum(equation = var_41319_equation_0, values = (var_40761_cast_fp16, var_41161_cast_fp16))[name = string("op_41319_cast_fp16")]; + string var_41321_equation_0 = const()[name = string("op_41321_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41321_cast_fp16 = einsum(equation = var_41321_equation_0, values = (var_40761_cast_fp16, var_41162_cast_fp16))[name = string("op_41321_cast_fp16")]; + string var_41323_equation_0 = const()[name = string("op_41323_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_41323_cast_fp16 = einsum(equation = var_41323_equation_0, values = (var_40761_cast_fp16, var_41163_cast_fp16))[name = string("op_41323_cast_fp16")]; + bool var_41325_interleave_0 = const()[name = string("op_41325_interleave_0"), val = bool(false)]; + tensor var_41325_cast_fp16 = concat(axis = var_39884, interleave = var_41325_interleave_0, values = (var_41165_cast_fp16, var_41167_cast_fp16, var_41169_cast_fp16, var_41171_cast_fp16))[name = string("op_41325_cast_fp16")]; + bool var_41327_interleave_0 = const()[name = string("op_41327_interleave_0"), val = bool(false)]; + tensor var_41327_cast_fp16 = concat(axis = var_39884, interleave = var_41327_interleave_0, values = (var_41173_cast_fp16, var_41175_cast_fp16, var_41177_cast_fp16, var_41179_cast_fp16))[name = string("op_41327_cast_fp16")]; + bool var_41329_interleave_0 = const()[name = string("op_41329_interleave_0"), val = bool(false)]; + tensor var_41329_cast_fp16 = concat(axis = var_39884, interleave = var_41329_interleave_0, values = (var_41181_cast_fp16, var_41183_cast_fp16, var_41185_cast_fp16, var_41187_cast_fp16))[name = string("op_41329_cast_fp16")]; + bool var_41331_interleave_0 = const()[name = string("op_41331_interleave_0"), val = bool(false)]; + tensor var_41331_cast_fp16 = concat(axis = var_39884, interleave = var_41331_interleave_0, values = (var_41189_cast_fp16, var_41191_cast_fp16, var_41193_cast_fp16, var_41195_cast_fp16))[name = string("op_41331_cast_fp16")]; + bool var_41333_interleave_0 = const()[name = string("op_41333_interleave_0"), val = bool(false)]; + tensor var_41333_cast_fp16 = concat(axis = var_39884, interleave = var_41333_interleave_0, values = (var_41197_cast_fp16, var_41199_cast_fp16, var_41201_cast_fp16, var_41203_cast_fp16))[name = string("op_41333_cast_fp16")]; + bool var_41335_interleave_0 = const()[name = string("op_41335_interleave_0"), val = bool(false)]; + tensor var_41335_cast_fp16 = concat(axis = var_39884, interleave = var_41335_interleave_0, values = (var_41205_cast_fp16, var_41207_cast_fp16, var_41209_cast_fp16, var_41211_cast_fp16))[name = string("op_41335_cast_fp16")]; + bool var_41337_interleave_0 = const()[name = string("op_41337_interleave_0"), val = bool(false)]; + tensor var_41337_cast_fp16 = concat(axis = var_39884, interleave = var_41337_interleave_0, values = (var_41213_cast_fp16, var_41215_cast_fp16, var_41217_cast_fp16, var_41219_cast_fp16))[name = string("op_41337_cast_fp16")]; + bool var_41339_interleave_0 = const()[name = string("op_41339_interleave_0"), val = bool(false)]; + tensor var_41339_cast_fp16 = concat(axis = var_39884, interleave = var_41339_interleave_0, values = (var_41221_cast_fp16, var_41223_cast_fp16, var_41225_cast_fp16, var_41227_cast_fp16))[name = string("op_41339_cast_fp16")]; + bool var_41341_interleave_0 = const()[name = string("op_41341_interleave_0"), val = bool(false)]; + tensor var_41341_cast_fp16 = concat(axis = var_39884, interleave = var_41341_interleave_0, values = (var_41229_cast_fp16, var_41231_cast_fp16, var_41233_cast_fp16, var_41235_cast_fp16))[name = string("op_41341_cast_fp16")]; + bool var_41343_interleave_0 = const()[name = string("op_41343_interleave_0"), val = bool(false)]; + tensor var_41343_cast_fp16 = concat(axis = var_39884, interleave = var_41343_interleave_0, values = (var_41237_cast_fp16, var_41239_cast_fp16, var_41241_cast_fp16, var_41243_cast_fp16))[name = string("op_41343_cast_fp16")]; + bool var_41345_interleave_0 = const()[name = string("op_41345_interleave_0"), val = bool(false)]; + tensor var_41345_cast_fp16 = concat(axis = var_39884, interleave = var_41345_interleave_0, values = (var_41245_cast_fp16, var_41247_cast_fp16, var_41249_cast_fp16, var_41251_cast_fp16))[name = string("op_41345_cast_fp16")]; + bool var_41347_interleave_0 = const()[name = string("op_41347_interleave_0"), val = bool(false)]; + tensor var_41347_cast_fp16 = concat(axis = var_39884, interleave = var_41347_interleave_0, values = (var_41253_cast_fp16, var_41255_cast_fp16, var_41257_cast_fp16, var_41259_cast_fp16))[name = string("op_41347_cast_fp16")]; + bool var_41349_interleave_0 = const()[name = string("op_41349_interleave_0"), val = bool(false)]; + tensor var_41349_cast_fp16 = concat(axis = var_39884, interleave = var_41349_interleave_0, values = (var_41261_cast_fp16, var_41263_cast_fp16, var_41265_cast_fp16, var_41267_cast_fp16))[name = string("op_41349_cast_fp16")]; + bool var_41351_interleave_0 = const()[name = string("op_41351_interleave_0"), val = bool(false)]; + tensor var_41351_cast_fp16 = concat(axis = var_39884, interleave = var_41351_interleave_0, values = (var_41269_cast_fp16, var_41271_cast_fp16, var_41273_cast_fp16, var_41275_cast_fp16))[name = string("op_41351_cast_fp16")]; + bool var_41353_interleave_0 = const()[name = string("op_41353_interleave_0"), val = bool(false)]; + tensor var_41353_cast_fp16 = concat(axis = var_39884, interleave = var_41353_interleave_0, values = (var_41277_cast_fp16, var_41279_cast_fp16, var_41281_cast_fp16, var_41283_cast_fp16))[name = string("op_41353_cast_fp16")]; + bool var_41355_interleave_0 = const()[name = string("op_41355_interleave_0"), val = bool(false)]; + tensor var_41355_cast_fp16 = concat(axis = var_39884, interleave = var_41355_interleave_0, values = (var_41285_cast_fp16, var_41287_cast_fp16, var_41289_cast_fp16, var_41291_cast_fp16))[name = string("op_41355_cast_fp16")]; + bool var_41357_interleave_0 = const()[name = string("op_41357_interleave_0"), val = bool(false)]; + tensor var_41357_cast_fp16 = concat(axis = var_39884, interleave = var_41357_interleave_0, values = (var_41293_cast_fp16, var_41295_cast_fp16, var_41297_cast_fp16, var_41299_cast_fp16))[name = string("op_41357_cast_fp16")]; + bool var_41359_interleave_0 = const()[name = string("op_41359_interleave_0"), val = bool(false)]; + tensor var_41359_cast_fp16 = concat(axis = var_39884, interleave = var_41359_interleave_0, values = (var_41301_cast_fp16, var_41303_cast_fp16, var_41305_cast_fp16, var_41307_cast_fp16))[name = string("op_41359_cast_fp16")]; + bool var_41361_interleave_0 = const()[name = string("op_41361_interleave_0"), val = bool(false)]; + tensor var_41361_cast_fp16 = concat(axis = var_39884, interleave = var_41361_interleave_0, values = (var_41309_cast_fp16, var_41311_cast_fp16, var_41313_cast_fp16, var_41315_cast_fp16))[name = string("op_41361_cast_fp16")]; + bool var_41363_interleave_0 = const()[name = string("op_41363_interleave_0"), val = bool(false)]; + tensor var_41363_cast_fp16 = concat(axis = var_39884, interleave = var_41363_interleave_0, values = (var_41317_cast_fp16, var_41319_cast_fp16, var_41321_cast_fp16, var_41323_cast_fp16))[name = string("op_41363_cast_fp16")]; + bool input_209_interleave_0 = const()[name = string("input_209_interleave_0"), val = bool(false)]; + tensor input_209_cast_fp16 = concat(axis = var_39909, interleave = input_209_interleave_0, values = (var_41325_cast_fp16, var_41327_cast_fp16, var_41329_cast_fp16, var_41331_cast_fp16, var_41333_cast_fp16, var_41335_cast_fp16, var_41337_cast_fp16, var_41339_cast_fp16, var_41341_cast_fp16, var_41343_cast_fp16, var_41345_cast_fp16, var_41347_cast_fp16, var_41349_cast_fp16, var_41351_cast_fp16, var_41353_cast_fp16, var_41355_cast_fp16, var_41357_cast_fp16, var_41359_cast_fp16, var_41361_cast_fp16, var_41363_cast_fp16))[name = string("input_209_cast_fp16")]; + string obj_107_pad_type_0 = const()[name = string("obj_107_pad_type_0"), val = string("valid")]; + tensor obj_107_strides_0 = const()[name = string("obj_107_strides_0"), val = tensor([1, 1])]; + tensor obj_107_pad_0 = const()[name = string("obj_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_107_dilations_0 = const()[name = string("obj_107_dilations_0"), val = tensor([1, 1])]; + int32 obj_107_groups_0 = const()[name = string("obj_107_groups_0"), val = int32(1)]; + tensor layers_26_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1047690560)))]; + tensor layers_26_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050967424)))]; + tensor obj_107_cast_fp16 = conv(bias = layers_26_self_attn_o_proj_bias_to_fp16, dilations = obj_107_dilations_0, groups = obj_107_groups_0, pad = obj_107_pad_0, pad_type = obj_107_pad_type_0, strides = obj_107_strides_0, weight = layers_26_self_attn_o_proj_weight_to_fp16, x = input_209_cast_fp16)[name = string("obj_107_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_107_cast_fp16)[name = string("inputs_107_cast_fp16")]; + tensor out_107_axes_0 = const()[name = string("out_107_axes_0"), val = tensor([1])]; + fp16 var_41382_to_fp16 = const()[name = string("op_41382_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_41382_to_fp16, x = inputs_107_cast_fp16)[name = string("out_107_cast_fp16")]; + tensor input_211_gamma_0_to_fp16 = const()[name = string("input_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050970048)))]; + tensor input_211_beta_0_to_fp16 = const()[name = string("input_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050972672)))]; + fp16 input_211_epsilon_0_to_fp16 = const()[name = string("input_211_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_211_cast_fp16 = batch_norm(beta = input_211_beta_0_to_fp16, epsilon = input_211_epsilon_0_to_fp16, gamma = input_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = string("input_211_cast_fp16")]; + string input_213_pad_type_0 = const()[name = string("input_213_pad_type_0"), val = string("valid")]; + tensor input_213_strides_0 = const()[name = string("input_213_strides_0"), val = tensor([1, 1])]; + tensor input_213_pad_0 = const()[name = string("input_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_213_dilations_0 = const()[name = string("input_213_dilations_0"), val = tensor([1, 1])]; + int32 input_213_groups_0 = const()[name = string("input_213_groups_0"), val = int32(1)]; + tensor layers_26_fc1_weight_to_fp16 = const()[name = string("layers_26_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1050975296)))]; + tensor layers_26_fc1_bias_to_fp16 = const()[name = string("layers_26_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064082560)))]; + tensor input_213_cast_fp16 = conv(bias = layers_26_fc1_bias_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = layers_26_fc1_weight_to_fp16, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")]; + string input_215_mode_0 = const()[name = string("input_215_mode_0"), val = string("EXACT")]; + tensor input_215_cast_fp16 = gelu(mode = input_215_mode_0, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")]; + string hidden_states_57_pad_type_0 = const()[name = string("hidden_states_57_pad_type_0"), val = string("valid")]; + tensor hidden_states_57_strides_0 = const()[name = string("hidden_states_57_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_57_pad_0 = const()[name = string("hidden_states_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_57_dilations_0 = const()[name = string("hidden_states_57_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_57_groups_0 = const()[name = string("hidden_states_57_groups_0"), val = int32(1)]; + tensor layers_26_fc2_weight_to_fp16 = const()[name = string("layers_26_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1064092864)))]; + tensor layers_26_fc2_bias_to_fp16 = const()[name = string("layers_26_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077200128)))]; + tensor hidden_states_57_cast_fp16 = conv(bias = layers_26_fc2_bias_to_fp16, dilations = hidden_states_57_dilations_0, groups = hidden_states_57_groups_0, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = hidden_states_57_strides_0, weight = layers_26_fc2_weight_to_fp16, x = input_215_cast_fp16)[name = string("hidden_states_57_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_57_cast_fp16)[name = string("inputs_109_cast_fp16")]; + int32 var_41411 = const()[name = string("op_41411"), val = int32(3)]; + int32 var_41436 = const()[name = string("op_41436"), val = int32(1)]; + tensor out_109_axes_0 = const()[name = string("out_109_axes_0"), val = tensor([1])]; + fp16 var_41453_to_fp16 = const()[name = string("op_41453_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_41453_to_fp16, x = inputs_109_cast_fp16)[name = string("out_109_cast_fp16")]; + tensor obj_109_gamma_0_to_fp16 = const()[name = string("obj_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077202752)))]; + tensor obj_109_beta_0_to_fp16 = const()[name = string("obj_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077205376)))]; + fp16 obj_109_epsilon_0_to_fp16 = const()[name = string("obj_109_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_109_cast_fp16 = batch_norm(beta = obj_109_beta_0_to_fp16, epsilon = obj_109_epsilon_0_to_fp16, gamma = obj_109_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = string("obj_109_cast_fp16")]; + string query_55_pad_type_0 = const()[name = string("query_55_pad_type_0"), val = string("valid")]; + tensor query_55_strides_0 = const()[name = string("query_55_strides_0"), val = tensor([1, 1])]; + tensor query_55_pad_0 = const()[name = string("query_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_55_dilations_0 = const()[name = string("query_55_dilations_0"), val = tensor([1, 1])]; + int32 query_55_groups_0 = const()[name = string("query_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077208000)))]; + tensor layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080484864)))]; + tensor query_55_cast_fp16 = conv(bias = layers_27_self_attn_q_proj_bias_to_fp16, dilations = query_55_dilations_0, groups = query_55_groups_0, pad = query_55_pad_0, pad_type = query_55_pad_type_0, strides = query_55_strides_0, weight = layers_27_self_attn_q_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("query_55_cast_fp16")]; + string key_55_pad_type_0 = const()[name = string("key_55_pad_type_0"), val = string("valid")]; + tensor key_55_strides_0 = const()[name = string("key_55_strides_0"), val = tensor([1, 1])]; + tensor key_55_pad_0 = const()[name = string("key_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_55_dilations_0 = const()[name = string("key_55_dilations_0"), val = tensor([1, 1])]; + int32 key_55_groups_0 = const()[name = string("key_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1080487488)))]; + tensor key_55_cast_fp16 = conv(dilations = key_55_dilations_0, groups = key_55_groups_0, pad = key_55_pad_0, pad_type = key_55_pad_type_0, strides = key_55_strides_0, weight = layers_27_self_attn_k_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("key_55_cast_fp16")]; + string value_55_pad_type_0 = const()[name = string("value_55_pad_type_0"), val = string("valid")]; + tensor value_55_strides_0 = const()[name = string("value_55_strides_0"), val = tensor([1, 1])]; + tensor value_55_pad_0 = const()[name = string("value_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_55_dilations_0 = const()[name = string("value_55_dilations_0"), val = tensor([1, 1])]; + int32 value_55_groups_0 = const()[name = string("value_55_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1083764352)))]; + tensor layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087041216)))]; + tensor value_55_cast_fp16 = conv(bias = layers_27_self_attn_v_proj_bias_to_fp16, dilations = value_55_dilations_0, groups = value_55_groups_0, pad = value_55_pad_0, pad_type = value_55_pad_type_0, strides = value_55_strides_0, weight = layers_27_self_attn_v_proj_weight_to_fp16, x = obj_109_cast_fp16)[name = string("value_55_cast_fp16")]; + tensor var_41491_begin_0 = const()[name = string("op_41491_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41491_end_0 = const()[name = string("op_41491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41491_end_mask_0 = const()[name = string("op_41491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41491_cast_fp16 = slice_by_index(begin = var_41491_begin_0, end = var_41491_end_0, end_mask = var_41491_end_mask_0, x = query_55_cast_fp16)[name = string("op_41491_cast_fp16")]; + tensor var_41495_begin_0 = const()[name = string("op_41495_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41495_end_0 = const()[name = string("op_41495_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_41495_end_mask_0 = const()[name = string("op_41495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41495_cast_fp16 = slice_by_index(begin = var_41495_begin_0, end = var_41495_end_0, end_mask = var_41495_end_mask_0, x = query_55_cast_fp16)[name = string("op_41495_cast_fp16")]; + tensor var_41499_begin_0 = const()[name = string("op_41499_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41499_end_0 = const()[name = string("op_41499_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_41499_end_mask_0 = const()[name = string("op_41499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41499_cast_fp16 = slice_by_index(begin = var_41499_begin_0, end = var_41499_end_0, end_mask = var_41499_end_mask_0, x = query_55_cast_fp16)[name = string("op_41499_cast_fp16")]; + tensor var_41503_begin_0 = const()[name = string("op_41503_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41503_end_0 = const()[name = string("op_41503_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_41503_end_mask_0 = const()[name = string("op_41503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41503_cast_fp16 = slice_by_index(begin = var_41503_begin_0, end = var_41503_end_0, end_mask = var_41503_end_mask_0, x = query_55_cast_fp16)[name = string("op_41503_cast_fp16")]; + tensor var_41507_begin_0 = const()[name = string("op_41507_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41507_end_0 = const()[name = string("op_41507_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_41507_end_mask_0 = const()[name = string("op_41507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41507_cast_fp16 = slice_by_index(begin = var_41507_begin_0, end = var_41507_end_0, end_mask = var_41507_end_mask_0, x = query_55_cast_fp16)[name = string("op_41507_cast_fp16")]; + tensor var_41511_begin_0 = const()[name = string("op_41511_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41511_end_0 = const()[name = string("op_41511_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_41511_end_mask_0 = const()[name = string("op_41511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41511_cast_fp16 = slice_by_index(begin = var_41511_begin_0, end = var_41511_end_0, end_mask = var_41511_end_mask_0, x = query_55_cast_fp16)[name = string("op_41511_cast_fp16")]; + tensor var_41515_begin_0 = const()[name = string("op_41515_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41515_end_0 = const()[name = string("op_41515_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_41515_end_mask_0 = const()[name = string("op_41515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41515_cast_fp16 = slice_by_index(begin = var_41515_begin_0, end = var_41515_end_0, end_mask = var_41515_end_mask_0, x = query_55_cast_fp16)[name = string("op_41515_cast_fp16")]; + tensor var_41519_begin_0 = const()[name = string("op_41519_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41519_end_0 = const()[name = string("op_41519_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_41519_end_mask_0 = const()[name = string("op_41519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41519_cast_fp16 = slice_by_index(begin = var_41519_begin_0, end = var_41519_end_0, end_mask = var_41519_end_mask_0, x = query_55_cast_fp16)[name = string("op_41519_cast_fp16")]; + tensor var_41523_begin_0 = const()[name = string("op_41523_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41523_end_0 = const()[name = string("op_41523_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_41523_end_mask_0 = const()[name = string("op_41523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41523_cast_fp16 = slice_by_index(begin = var_41523_begin_0, end = var_41523_end_0, end_mask = var_41523_end_mask_0, x = query_55_cast_fp16)[name = string("op_41523_cast_fp16")]; + tensor var_41527_begin_0 = const()[name = string("op_41527_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41527_end_0 = const()[name = string("op_41527_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_41527_end_mask_0 = const()[name = string("op_41527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41527_cast_fp16 = slice_by_index(begin = var_41527_begin_0, end = var_41527_end_0, end_mask = var_41527_end_mask_0, x = query_55_cast_fp16)[name = string("op_41527_cast_fp16")]; + tensor var_41531_begin_0 = const()[name = string("op_41531_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41531_end_0 = const()[name = string("op_41531_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_41531_end_mask_0 = const()[name = string("op_41531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41531_cast_fp16 = slice_by_index(begin = var_41531_begin_0, end = var_41531_end_0, end_mask = var_41531_end_mask_0, x = query_55_cast_fp16)[name = string("op_41531_cast_fp16")]; + tensor var_41535_begin_0 = const()[name = string("op_41535_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41535_end_0 = const()[name = string("op_41535_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_41535_end_mask_0 = const()[name = string("op_41535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41535_cast_fp16 = slice_by_index(begin = var_41535_begin_0, end = var_41535_end_0, end_mask = var_41535_end_mask_0, x = query_55_cast_fp16)[name = string("op_41535_cast_fp16")]; + tensor var_41539_begin_0 = const()[name = string("op_41539_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41539_end_0 = const()[name = string("op_41539_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_41539_end_mask_0 = const()[name = string("op_41539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41539_cast_fp16 = slice_by_index(begin = var_41539_begin_0, end = var_41539_end_0, end_mask = var_41539_end_mask_0, x = query_55_cast_fp16)[name = string("op_41539_cast_fp16")]; + tensor var_41543_begin_0 = const()[name = string("op_41543_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41543_end_0 = const()[name = string("op_41543_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_41543_end_mask_0 = const()[name = string("op_41543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41543_cast_fp16 = slice_by_index(begin = var_41543_begin_0, end = var_41543_end_0, end_mask = var_41543_end_mask_0, x = query_55_cast_fp16)[name = string("op_41543_cast_fp16")]; + tensor var_41547_begin_0 = const()[name = string("op_41547_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41547_end_0 = const()[name = string("op_41547_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_41547_end_mask_0 = const()[name = string("op_41547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41547_cast_fp16 = slice_by_index(begin = var_41547_begin_0, end = var_41547_end_0, end_mask = var_41547_end_mask_0, x = query_55_cast_fp16)[name = string("op_41547_cast_fp16")]; + tensor var_41551_begin_0 = const()[name = string("op_41551_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41551_end_0 = const()[name = string("op_41551_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_41551_end_mask_0 = const()[name = string("op_41551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41551_cast_fp16 = slice_by_index(begin = var_41551_begin_0, end = var_41551_end_0, end_mask = var_41551_end_mask_0, x = query_55_cast_fp16)[name = string("op_41551_cast_fp16")]; + tensor var_41555_begin_0 = const()[name = string("op_41555_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41555_end_0 = const()[name = string("op_41555_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_41555_end_mask_0 = const()[name = string("op_41555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41555_cast_fp16 = slice_by_index(begin = var_41555_begin_0, end = var_41555_end_0, end_mask = var_41555_end_mask_0, x = query_55_cast_fp16)[name = string("op_41555_cast_fp16")]; + tensor var_41559_begin_0 = const()[name = string("op_41559_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41559_end_0 = const()[name = string("op_41559_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_41559_end_mask_0 = const()[name = string("op_41559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41559_cast_fp16 = slice_by_index(begin = var_41559_begin_0, end = var_41559_end_0, end_mask = var_41559_end_mask_0, x = query_55_cast_fp16)[name = string("op_41559_cast_fp16")]; + tensor var_41563_begin_0 = const()[name = string("op_41563_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41563_end_0 = const()[name = string("op_41563_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_41563_end_mask_0 = const()[name = string("op_41563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41563_cast_fp16 = slice_by_index(begin = var_41563_begin_0, end = var_41563_end_0, end_mask = var_41563_end_mask_0, x = query_55_cast_fp16)[name = string("op_41563_cast_fp16")]; + tensor var_41567_begin_0 = const()[name = string("op_41567_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41567_end_0 = const()[name = string("op_41567_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_41567_end_mask_0 = const()[name = string("op_41567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41567_cast_fp16 = slice_by_index(begin = var_41567_begin_0, end = var_41567_end_0, end_mask = var_41567_end_mask_0, x = query_55_cast_fp16)[name = string("op_41567_cast_fp16")]; + tensor var_41576_begin_0 = const()[name = string("op_41576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41576_end_0 = const()[name = string("op_41576_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41576_end_mask_0 = const()[name = string("op_41576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41576_cast_fp16 = slice_by_index(begin = var_41576_begin_0, end = var_41576_end_0, end_mask = var_41576_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41576_cast_fp16")]; + tensor var_41583_begin_0 = const()[name = string("op_41583_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41583_end_0 = const()[name = string("op_41583_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41583_end_mask_0 = const()[name = string("op_41583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41583_cast_fp16 = slice_by_index(begin = var_41583_begin_0, end = var_41583_end_0, end_mask = var_41583_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41583_cast_fp16")]; + tensor var_41590_begin_0 = const()[name = string("op_41590_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41590_end_0 = const()[name = string("op_41590_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41590_end_mask_0 = const()[name = string("op_41590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41590_cast_fp16 = slice_by_index(begin = var_41590_begin_0, end = var_41590_end_0, end_mask = var_41590_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41590_cast_fp16")]; + tensor var_41597_begin_0 = const()[name = string("op_41597_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41597_end_0 = const()[name = string("op_41597_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41597_end_mask_0 = const()[name = string("op_41597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41597_cast_fp16 = slice_by_index(begin = var_41597_begin_0, end = var_41597_end_0, end_mask = var_41597_end_mask_0, x = var_41491_cast_fp16)[name = string("op_41597_cast_fp16")]; + tensor var_41604_begin_0 = const()[name = string("op_41604_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41604_end_0 = const()[name = string("op_41604_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41604_end_mask_0 = const()[name = string("op_41604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41604_cast_fp16 = slice_by_index(begin = var_41604_begin_0, end = var_41604_end_0, end_mask = var_41604_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41604_cast_fp16")]; + tensor var_41611_begin_0 = const()[name = string("op_41611_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41611_end_0 = const()[name = string("op_41611_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41611_end_mask_0 = const()[name = string("op_41611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41611_cast_fp16 = slice_by_index(begin = var_41611_begin_0, end = var_41611_end_0, end_mask = var_41611_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41611_cast_fp16")]; + tensor var_41618_begin_0 = const()[name = string("op_41618_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41618_end_0 = const()[name = string("op_41618_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41618_end_mask_0 = const()[name = string("op_41618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41618_cast_fp16 = slice_by_index(begin = var_41618_begin_0, end = var_41618_end_0, end_mask = var_41618_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41618_cast_fp16")]; + tensor var_41625_begin_0 = const()[name = string("op_41625_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41625_end_0 = const()[name = string("op_41625_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41625_end_mask_0 = const()[name = string("op_41625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41625_cast_fp16 = slice_by_index(begin = var_41625_begin_0, end = var_41625_end_0, end_mask = var_41625_end_mask_0, x = var_41495_cast_fp16)[name = string("op_41625_cast_fp16")]; + tensor var_41632_begin_0 = const()[name = string("op_41632_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41632_end_0 = const()[name = string("op_41632_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41632_end_mask_0 = const()[name = string("op_41632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41632_cast_fp16 = slice_by_index(begin = var_41632_begin_0, end = var_41632_end_0, end_mask = var_41632_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41632_cast_fp16")]; + tensor var_41639_begin_0 = const()[name = string("op_41639_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41639_end_0 = const()[name = string("op_41639_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41639_end_mask_0 = const()[name = string("op_41639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41639_cast_fp16 = slice_by_index(begin = var_41639_begin_0, end = var_41639_end_0, end_mask = var_41639_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41639_cast_fp16")]; + tensor var_41646_begin_0 = const()[name = string("op_41646_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41646_end_0 = const()[name = string("op_41646_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41646_end_mask_0 = const()[name = string("op_41646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41646_cast_fp16 = slice_by_index(begin = var_41646_begin_0, end = var_41646_end_0, end_mask = var_41646_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41646_cast_fp16")]; + tensor var_41653_begin_0 = const()[name = string("op_41653_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41653_end_0 = const()[name = string("op_41653_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41653_end_mask_0 = const()[name = string("op_41653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41653_cast_fp16 = slice_by_index(begin = var_41653_begin_0, end = var_41653_end_0, end_mask = var_41653_end_mask_0, x = var_41499_cast_fp16)[name = string("op_41653_cast_fp16")]; + tensor var_41660_begin_0 = const()[name = string("op_41660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41660_end_0 = const()[name = string("op_41660_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41660_end_mask_0 = const()[name = string("op_41660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41660_cast_fp16 = slice_by_index(begin = var_41660_begin_0, end = var_41660_end_0, end_mask = var_41660_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41660_cast_fp16")]; + tensor var_41667_begin_0 = const()[name = string("op_41667_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41667_end_0 = const()[name = string("op_41667_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41667_end_mask_0 = const()[name = string("op_41667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41667_cast_fp16 = slice_by_index(begin = var_41667_begin_0, end = var_41667_end_0, end_mask = var_41667_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41667_cast_fp16")]; + tensor var_41674_begin_0 = const()[name = string("op_41674_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41674_end_0 = const()[name = string("op_41674_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41674_end_mask_0 = const()[name = string("op_41674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41674_cast_fp16 = slice_by_index(begin = var_41674_begin_0, end = var_41674_end_0, end_mask = var_41674_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41674_cast_fp16")]; + tensor var_41681_begin_0 = const()[name = string("op_41681_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41681_end_0 = const()[name = string("op_41681_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41681_end_mask_0 = const()[name = string("op_41681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41681_cast_fp16 = slice_by_index(begin = var_41681_begin_0, end = var_41681_end_0, end_mask = var_41681_end_mask_0, x = var_41503_cast_fp16)[name = string("op_41681_cast_fp16")]; + tensor var_41688_begin_0 = const()[name = string("op_41688_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41688_end_0 = const()[name = string("op_41688_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41688_end_mask_0 = const()[name = string("op_41688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41688_cast_fp16 = slice_by_index(begin = var_41688_begin_0, end = var_41688_end_0, end_mask = var_41688_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41688_cast_fp16")]; + tensor var_41695_begin_0 = const()[name = string("op_41695_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41695_end_0 = const()[name = string("op_41695_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41695_end_mask_0 = const()[name = string("op_41695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41695_cast_fp16 = slice_by_index(begin = var_41695_begin_0, end = var_41695_end_0, end_mask = var_41695_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41695_cast_fp16")]; + tensor var_41702_begin_0 = const()[name = string("op_41702_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41702_end_0 = const()[name = string("op_41702_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41702_end_mask_0 = const()[name = string("op_41702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41702_cast_fp16 = slice_by_index(begin = var_41702_begin_0, end = var_41702_end_0, end_mask = var_41702_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41702_cast_fp16")]; + tensor var_41709_begin_0 = const()[name = string("op_41709_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41709_end_0 = const()[name = string("op_41709_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41709_end_mask_0 = const()[name = string("op_41709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41709_cast_fp16 = slice_by_index(begin = var_41709_begin_0, end = var_41709_end_0, end_mask = var_41709_end_mask_0, x = var_41507_cast_fp16)[name = string("op_41709_cast_fp16")]; + tensor var_41716_begin_0 = const()[name = string("op_41716_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41716_end_0 = const()[name = string("op_41716_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41716_end_mask_0 = const()[name = string("op_41716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41716_cast_fp16 = slice_by_index(begin = var_41716_begin_0, end = var_41716_end_0, end_mask = var_41716_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41716_cast_fp16")]; + tensor var_41723_begin_0 = const()[name = string("op_41723_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41723_end_0 = const()[name = string("op_41723_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41723_end_mask_0 = const()[name = string("op_41723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41723_cast_fp16 = slice_by_index(begin = var_41723_begin_0, end = var_41723_end_0, end_mask = var_41723_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41723_cast_fp16")]; + tensor var_41730_begin_0 = const()[name = string("op_41730_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41730_end_0 = const()[name = string("op_41730_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41730_end_mask_0 = const()[name = string("op_41730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41730_cast_fp16 = slice_by_index(begin = var_41730_begin_0, end = var_41730_end_0, end_mask = var_41730_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41730_cast_fp16")]; + tensor var_41737_begin_0 = const()[name = string("op_41737_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41737_end_0 = const()[name = string("op_41737_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41737_end_mask_0 = const()[name = string("op_41737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41737_cast_fp16 = slice_by_index(begin = var_41737_begin_0, end = var_41737_end_0, end_mask = var_41737_end_mask_0, x = var_41511_cast_fp16)[name = string("op_41737_cast_fp16")]; + tensor var_41744_begin_0 = const()[name = string("op_41744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41744_end_0 = const()[name = string("op_41744_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41744_end_mask_0 = const()[name = string("op_41744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41744_cast_fp16 = slice_by_index(begin = var_41744_begin_0, end = var_41744_end_0, end_mask = var_41744_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41744_cast_fp16")]; + tensor var_41751_begin_0 = const()[name = string("op_41751_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41751_end_0 = const()[name = string("op_41751_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41751_end_mask_0 = const()[name = string("op_41751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41751_cast_fp16 = slice_by_index(begin = var_41751_begin_0, end = var_41751_end_0, end_mask = var_41751_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41751_cast_fp16")]; + tensor var_41758_begin_0 = const()[name = string("op_41758_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41758_end_0 = const()[name = string("op_41758_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41758_end_mask_0 = const()[name = string("op_41758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41758_cast_fp16 = slice_by_index(begin = var_41758_begin_0, end = var_41758_end_0, end_mask = var_41758_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41758_cast_fp16")]; + tensor var_41765_begin_0 = const()[name = string("op_41765_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41765_end_0 = const()[name = string("op_41765_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41765_end_mask_0 = const()[name = string("op_41765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41765_cast_fp16 = slice_by_index(begin = var_41765_begin_0, end = var_41765_end_0, end_mask = var_41765_end_mask_0, x = var_41515_cast_fp16)[name = string("op_41765_cast_fp16")]; + tensor var_41772_begin_0 = const()[name = string("op_41772_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41772_end_0 = const()[name = string("op_41772_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41772_end_mask_0 = const()[name = string("op_41772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41772_cast_fp16 = slice_by_index(begin = var_41772_begin_0, end = var_41772_end_0, end_mask = var_41772_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41772_cast_fp16")]; + tensor var_41779_begin_0 = const()[name = string("op_41779_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41779_end_0 = const()[name = string("op_41779_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41779_end_mask_0 = const()[name = string("op_41779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41779_cast_fp16 = slice_by_index(begin = var_41779_begin_0, end = var_41779_end_0, end_mask = var_41779_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41779_cast_fp16")]; + tensor var_41786_begin_0 = const()[name = string("op_41786_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41786_end_0 = const()[name = string("op_41786_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41786_end_mask_0 = const()[name = string("op_41786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41786_cast_fp16 = slice_by_index(begin = var_41786_begin_0, end = var_41786_end_0, end_mask = var_41786_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41786_cast_fp16")]; + tensor var_41793_begin_0 = const()[name = string("op_41793_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41793_end_0 = const()[name = string("op_41793_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41793_end_mask_0 = const()[name = string("op_41793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41793_cast_fp16 = slice_by_index(begin = var_41793_begin_0, end = var_41793_end_0, end_mask = var_41793_end_mask_0, x = var_41519_cast_fp16)[name = string("op_41793_cast_fp16")]; + tensor var_41800_begin_0 = const()[name = string("op_41800_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41800_end_0 = const()[name = string("op_41800_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41800_end_mask_0 = const()[name = string("op_41800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41800_cast_fp16 = slice_by_index(begin = var_41800_begin_0, end = var_41800_end_0, end_mask = var_41800_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41800_cast_fp16")]; + tensor var_41807_begin_0 = const()[name = string("op_41807_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41807_end_0 = const()[name = string("op_41807_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41807_end_mask_0 = const()[name = string("op_41807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41807_cast_fp16 = slice_by_index(begin = var_41807_begin_0, end = var_41807_end_0, end_mask = var_41807_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41807_cast_fp16")]; + tensor var_41814_begin_0 = const()[name = string("op_41814_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41814_end_0 = const()[name = string("op_41814_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41814_end_mask_0 = const()[name = string("op_41814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41814_cast_fp16 = slice_by_index(begin = var_41814_begin_0, end = var_41814_end_0, end_mask = var_41814_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41814_cast_fp16")]; + tensor var_41821_begin_0 = const()[name = string("op_41821_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41821_end_0 = const()[name = string("op_41821_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41821_end_mask_0 = const()[name = string("op_41821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41821_cast_fp16 = slice_by_index(begin = var_41821_begin_0, end = var_41821_end_0, end_mask = var_41821_end_mask_0, x = var_41523_cast_fp16)[name = string("op_41821_cast_fp16")]; + tensor var_41828_begin_0 = const()[name = string("op_41828_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41828_end_0 = const()[name = string("op_41828_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41828_end_mask_0 = const()[name = string("op_41828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41828_cast_fp16 = slice_by_index(begin = var_41828_begin_0, end = var_41828_end_0, end_mask = var_41828_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41828_cast_fp16")]; + tensor var_41835_begin_0 = const()[name = string("op_41835_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41835_end_0 = const()[name = string("op_41835_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41835_end_mask_0 = const()[name = string("op_41835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41835_cast_fp16 = slice_by_index(begin = var_41835_begin_0, end = var_41835_end_0, end_mask = var_41835_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41835_cast_fp16")]; + tensor var_41842_begin_0 = const()[name = string("op_41842_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41842_end_0 = const()[name = string("op_41842_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41842_end_mask_0 = const()[name = string("op_41842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41842_cast_fp16 = slice_by_index(begin = var_41842_begin_0, end = var_41842_end_0, end_mask = var_41842_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41842_cast_fp16")]; + tensor var_41849_begin_0 = const()[name = string("op_41849_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41849_end_0 = const()[name = string("op_41849_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41849_end_mask_0 = const()[name = string("op_41849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41849_cast_fp16 = slice_by_index(begin = var_41849_begin_0, end = var_41849_end_0, end_mask = var_41849_end_mask_0, x = var_41527_cast_fp16)[name = string("op_41849_cast_fp16")]; + tensor var_41856_begin_0 = const()[name = string("op_41856_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41856_end_0 = const()[name = string("op_41856_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41856_end_mask_0 = const()[name = string("op_41856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41856_cast_fp16 = slice_by_index(begin = var_41856_begin_0, end = var_41856_end_0, end_mask = var_41856_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41856_cast_fp16")]; + tensor var_41863_begin_0 = const()[name = string("op_41863_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41863_end_0 = const()[name = string("op_41863_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41863_end_mask_0 = const()[name = string("op_41863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41863_cast_fp16 = slice_by_index(begin = var_41863_begin_0, end = var_41863_end_0, end_mask = var_41863_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41863_cast_fp16")]; + tensor var_41870_begin_0 = const()[name = string("op_41870_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41870_end_0 = const()[name = string("op_41870_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41870_end_mask_0 = const()[name = string("op_41870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41870_cast_fp16 = slice_by_index(begin = var_41870_begin_0, end = var_41870_end_0, end_mask = var_41870_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41870_cast_fp16")]; + tensor var_41877_begin_0 = const()[name = string("op_41877_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41877_end_0 = const()[name = string("op_41877_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41877_end_mask_0 = const()[name = string("op_41877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41877_cast_fp16 = slice_by_index(begin = var_41877_begin_0, end = var_41877_end_0, end_mask = var_41877_end_mask_0, x = var_41531_cast_fp16)[name = string("op_41877_cast_fp16")]; + tensor var_41884_begin_0 = const()[name = string("op_41884_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41884_end_0 = const()[name = string("op_41884_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41884_end_mask_0 = const()[name = string("op_41884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41884_cast_fp16 = slice_by_index(begin = var_41884_begin_0, end = var_41884_end_0, end_mask = var_41884_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41884_cast_fp16")]; + tensor var_41891_begin_0 = const()[name = string("op_41891_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41891_end_0 = const()[name = string("op_41891_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41891_end_mask_0 = const()[name = string("op_41891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41891_cast_fp16 = slice_by_index(begin = var_41891_begin_0, end = var_41891_end_0, end_mask = var_41891_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41891_cast_fp16")]; + tensor var_41898_begin_0 = const()[name = string("op_41898_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41898_end_0 = const()[name = string("op_41898_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41898_end_mask_0 = const()[name = string("op_41898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41898_cast_fp16 = slice_by_index(begin = var_41898_begin_0, end = var_41898_end_0, end_mask = var_41898_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41898_cast_fp16")]; + tensor var_41905_begin_0 = const()[name = string("op_41905_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41905_end_0 = const()[name = string("op_41905_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41905_end_mask_0 = const()[name = string("op_41905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41905_cast_fp16 = slice_by_index(begin = var_41905_begin_0, end = var_41905_end_0, end_mask = var_41905_end_mask_0, x = var_41535_cast_fp16)[name = string("op_41905_cast_fp16")]; + tensor var_41912_begin_0 = const()[name = string("op_41912_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41912_end_0 = const()[name = string("op_41912_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41912_end_mask_0 = const()[name = string("op_41912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41912_cast_fp16 = slice_by_index(begin = var_41912_begin_0, end = var_41912_end_0, end_mask = var_41912_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41912_cast_fp16")]; + tensor var_41919_begin_0 = const()[name = string("op_41919_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41919_end_0 = const()[name = string("op_41919_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41919_end_mask_0 = const()[name = string("op_41919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41919_cast_fp16 = slice_by_index(begin = var_41919_begin_0, end = var_41919_end_0, end_mask = var_41919_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41919_cast_fp16")]; + tensor var_41926_begin_0 = const()[name = string("op_41926_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41926_end_0 = const()[name = string("op_41926_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41926_end_mask_0 = const()[name = string("op_41926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41926_cast_fp16 = slice_by_index(begin = var_41926_begin_0, end = var_41926_end_0, end_mask = var_41926_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41926_cast_fp16")]; + tensor var_41933_begin_0 = const()[name = string("op_41933_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41933_end_0 = const()[name = string("op_41933_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41933_end_mask_0 = const()[name = string("op_41933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41933_cast_fp16 = slice_by_index(begin = var_41933_begin_0, end = var_41933_end_0, end_mask = var_41933_end_mask_0, x = var_41539_cast_fp16)[name = string("op_41933_cast_fp16")]; + tensor var_41940_begin_0 = const()[name = string("op_41940_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41940_end_0 = const()[name = string("op_41940_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41940_end_mask_0 = const()[name = string("op_41940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41940_cast_fp16 = slice_by_index(begin = var_41940_begin_0, end = var_41940_end_0, end_mask = var_41940_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41940_cast_fp16")]; + tensor var_41947_begin_0 = const()[name = string("op_41947_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41947_end_0 = const()[name = string("op_41947_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41947_end_mask_0 = const()[name = string("op_41947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41947_cast_fp16 = slice_by_index(begin = var_41947_begin_0, end = var_41947_end_0, end_mask = var_41947_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41947_cast_fp16")]; + tensor var_41954_begin_0 = const()[name = string("op_41954_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41954_end_0 = const()[name = string("op_41954_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41954_end_mask_0 = const()[name = string("op_41954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41954_cast_fp16 = slice_by_index(begin = var_41954_begin_0, end = var_41954_end_0, end_mask = var_41954_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41954_cast_fp16")]; + tensor var_41961_begin_0 = const()[name = string("op_41961_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41961_end_0 = const()[name = string("op_41961_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41961_end_mask_0 = const()[name = string("op_41961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41961_cast_fp16 = slice_by_index(begin = var_41961_begin_0, end = var_41961_end_0, end_mask = var_41961_end_mask_0, x = var_41543_cast_fp16)[name = string("op_41961_cast_fp16")]; + tensor var_41968_begin_0 = const()[name = string("op_41968_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41968_end_0 = const()[name = string("op_41968_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41968_end_mask_0 = const()[name = string("op_41968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41968_cast_fp16 = slice_by_index(begin = var_41968_begin_0, end = var_41968_end_0, end_mask = var_41968_end_mask_0, x = var_41547_cast_fp16)[name = string("op_41968_cast_fp16")]; + tensor var_41975_begin_0 = const()[name = string("op_41975_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41975_end_0 = const()[name = string("op_41975_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41975_end_mask_0 = const()[name = string("op_41975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41975_cast_fp16 = slice_by_index(begin = var_41975_begin_0, end = var_41975_end_0, end_mask = var_41975_end_mask_0, x = var_41547_cast_fp16)[name = string("op_41975_cast_fp16")]; + tensor var_41982_begin_0 = const()[name = string("op_41982_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41982_end_0 = const()[name = string("op_41982_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41982_end_mask_0 = const()[name = string("op_41982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41982_cast_fp16 = slice_by_index(begin = var_41982_begin_0, end = var_41982_end_0, end_mask = var_41982_end_mask_0, x = var_41547_cast_fp16)[name = string("op_41982_cast_fp16")]; + tensor var_41989_begin_0 = const()[name = string("op_41989_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41989_end_0 = const()[name = string("op_41989_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41989_end_mask_0 = const()[name = string("op_41989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41989_cast_fp16 = slice_by_index(begin = var_41989_begin_0, end = var_41989_end_0, end_mask = var_41989_end_mask_0, x = var_41547_cast_fp16)[name = string("op_41989_cast_fp16")]; + tensor var_41996_begin_0 = const()[name = string("op_41996_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41996_end_0 = const()[name = string("op_41996_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41996_end_mask_0 = const()[name = string("op_41996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41996_cast_fp16 = slice_by_index(begin = var_41996_begin_0, end = var_41996_end_0, end_mask = var_41996_end_mask_0, x = var_41551_cast_fp16)[name = string("op_41996_cast_fp16")]; + tensor var_42003_begin_0 = const()[name = string("op_42003_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42003_end_0 = const()[name = string("op_42003_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42003_end_mask_0 = const()[name = string("op_42003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42003_cast_fp16 = slice_by_index(begin = var_42003_begin_0, end = var_42003_end_0, end_mask = var_42003_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42003_cast_fp16")]; + tensor var_42010_begin_0 = const()[name = string("op_42010_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42010_end_0 = const()[name = string("op_42010_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42010_end_mask_0 = const()[name = string("op_42010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42010_cast_fp16 = slice_by_index(begin = var_42010_begin_0, end = var_42010_end_0, end_mask = var_42010_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42010_cast_fp16")]; + tensor var_42017_begin_0 = const()[name = string("op_42017_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42017_end_0 = const()[name = string("op_42017_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42017_end_mask_0 = const()[name = string("op_42017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42017_cast_fp16 = slice_by_index(begin = var_42017_begin_0, end = var_42017_end_0, end_mask = var_42017_end_mask_0, x = var_41551_cast_fp16)[name = string("op_42017_cast_fp16")]; + tensor var_42024_begin_0 = const()[name = string("op_42024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42024_end_0 = const()[name = string("op_42024_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42024_end_mask_0 = const()[name = string("op_42024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42024_cast_fp16 = slice_by_index(begin = var_42024_begin_0, end = var_42024_end_0, end_mask = var_42024_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42024_cast_fp16")]; + tensor var_42031_begin_0 = const()[name = string("op_42031_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42031_end_0 = const()[name = string("op_42031_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42031_end_mask_0 = const()[name = string("op_42031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42031_cast_fp16 = slice_by_index(begin = var_42031_begin_0, end = var_42031_end_0, end_mask = var_42031_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42031_cast_fp16")]; + tensor var_42038_begin_0 = const()[name = string("op_42038_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42038_end_0 = const()[name = string("op_42038_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42038_end_mask_0 = const()[name = string("op_42038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42038_cast_fp16 = slice_by_index(begin = var_42038_begin_0, end = var_42038_end_0, end_mask = var_42038_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42038_cast_fp16")]; + tensor var_42045_begin_0 = const()[name = string("op_42045_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42045_end_0 = const()[name = string("op_42045_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42045_end_mask_0 = const()[name = string("op_42045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42045_cast_fp16 = slice_by_index(begin = var_42045_begin_0, end = var_42045_end_0, end_mask = var_42045_end_mask_0, x = var_41555_cast_fp16)[name = string("op_42045_cast_fp16")]; + tensor var_42052_begin_0 = const()[name = string("op_42052_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42052_end_0 = const()[name = string("op_42052_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42052_end_mask_0 = const()[name = string("op_42052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42052_cast_fp16 = slice_by_index(begin = var_42052_begin_0, end = var_42052_end_0, end_mask = var_42052_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42052_cast_fp16")]; + tensor var_42059_begin_0 = const()[name = string("op_42059_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42059_end_0 = const()[name = string("op_42059_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42059_end_mask_0 = const()[name = string("op_42059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42059_cast_fp16 = slice_by_index(begin = var_42059_begin_0, end = var_42059_end_0, end_mask = var_42059_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42059_cast_fp16")]; + tensor var_42066_begin_0 = const()[name = string("op_42066_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42066_end_0 = const()[name = string("op_42066_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42066_end_mask_0 = const()[name = string("op_42066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42066_cast_fp16 = slice_by_index(begin = var_42066_begin_0, end = var_42066_end_0, end_mask = var_42066_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42066_cast_fp16")]; + tensor var_42073_begin_0 = const()[name = string("op_42073_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42073_end_0 = const()[name = string("op_42073_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42073_end_mask_0 = const()[name = string("op_42073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42073_cast_fp16 = slice_by_index(begin = var_42073_begin_0, end = var_42073_end_0, end_mask = var_42073_end_mask_0, x = var_41559_cast_fp16)[name = string("op_42073_cast_fp16")]; + tensor var_42080_begin_0 = const()[name = string("op_42080_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42080_end_0 = const()[name = string("op_42080_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42080_end_mask_0 = const()[name = string("op_42080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42080_cast_fp16 = slice_by_index(begin = var_42080_begin_0, end = var_42080_end_0, end_mask = var_42080_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42080_cast_fp16")]; + tensor var_42087_begin_0 = const()[name = string("op_42087_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42087_end_0 = const()[name = string("op_42087_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42087_end_mask_0 = const()[name = string("op_42087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42087_cast_fp16 = slice_by_index(begin = var_42087_begin_0, end = var_42087_end_0, end_mask = var_42087_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42087_cast_fp16")]; + tensor var_42094_begin_0 = const()[name = string("op_42094_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42094_end_0 = const()[name = string("op_42094_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42094_end_mask_0 = const()[name = string("op_42094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42094_cast_fp16 = slice_by_index(begin = var_42094_begin_0, end = var_42094_end_0, end_mask = var_42094_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42094_cast_fp16")]; + tensor var_42101_begin_0 = const()[name = string("op_42101_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42101_end_0 = const()[name = string("op_42101_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42101_end_mask_0 = const()[name = string("op_42101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42101_cast_fp16 = slice_by_index(begin = var_42101_begin_0, end = var_42101_end_0, end_mask = var_42101_end_mask_0, x = var_41563_cast_fp16)[name = string("op_42101_cast_fp16")]; + tensor var_42108_begin_0 = const()[name = string("op_42108_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42108_end_0 = const()[name = string("op_42108_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42108_end_mask_0 = const()[name = string("op_42108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42108_cast_fp16 = slice_by_index(begin = var_42108_begin_0, end = var_42108_end_0, end_mask = var_42108_end_mask_0, x = var_41567_cast_fp16)[name = string("op_42108_cast_fp16")]; + tensor var_42115_begin_0 = const()[name = string("op_42115_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42115_end_0 = const()[name = string("op_42115_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42115_end_mask_0 = const()[name = string("op_42115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42115_cast_fp16 = slice_by_index(begin = var_42115_begin_0, end = var_42115_end_0, end_mask = var_42115_end_mask_0, x = var_41567_cast_fp16)[name = string("op_42115_cast_fp16")]; + tensor var_42122_begin_0 = const()[name = string("op_42122_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42122_end_0 = const()[name = string("op_42122_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42122_end_mask_0 = const()[name = string("op_42122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42122_cast_fp16 = slice_by_index(begin = var_42122_begin_0, end = var_42122_end_0, end_mask = var_42122_end_mask_0, x = var_41567_cast_fp16)[name = string("op_42122_cast_fp16")]; + tensor var_42129_begin_0 = const()[name = string("op_42129_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42129_end_0 = const()[name = string("op_42129_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42129_end_mask_0 = const()[name = string("op_42129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42129_cast_fp16 = slice_by_index(begin = var_42129_begin_0, end = var_42129_end_0, end_mask = var_42129_end_mask_0, x = var_41567_cast_fp16)[name = string("op_42129_cast_fp16")]; + tensor k_55_perm_0 = const()[name = string("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_42134_begin_0 = const()[name = string("op_42134_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42134_end_0 = const()[name = string("op_42134_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_42134_end_mask_0 = const()[name = string("op_42134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_55_cast_fp16 = transpose(perm = k_55_perm_0, x = key_55_cast_fp16)[name = string("transpose_4")]; + tensor var_42134_cast_fp16 = slice_by_index(begin = var_42134_begin_0, end = var_42134_end_0, end_mask = var_42134_end_mask_0, x = k_55_cast_fp16)[name = string("op_42134_cast_fp16")]; + tensor var_42138_begin_0 = const()[name = string("op_42138_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_42138_end_0 = const()[name = string("op_42138_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_42138_end_mask_0 = const()[name = string("op_42138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42138_cast_fp16 = slice_by_index(begin = var_42138_begin_0, end = var_42138_end_0, end_mask = var_42138_end_mask_0, x = k_55_cast_fp16)[name = string("op_42138_cast_fp16")]; + tensor var_42142_begin_0 = const()[name = string("op_42142_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_42142_end_0 = const()[name = string("op_42142_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_42142_end_mask_0 = const()[name = string("op_42142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42142_cast_fp16 = slice_by_index(begin = var_42142_begin_0, end = var_42142_end_0, end_mask = var_42142_end_mask_0, x = k_55_cast_fp16)[name = string("op_42142_cast_fp16")]; + tensor var_42146_begin_0 = const()[name = string("op_42146_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_42146_end_0 = const()[name = string("op_42146_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_42146_end_mask_0 = const()[name = string("op_42146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42146_cast_fp16 = slice_by_index(begin = var_42146_begin_0, end = var_42146_end_0, end_mask = var_42146_end_mask_0, x = k_55_cast_fp16)[name = string("op_42146_cast_fp16")]; + tensor var_42150_begin_0 = const()[name = string("op_42150_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_42150_end_0 = const()[name = string("op_42150_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_42150_end_mask_0 = const()[name = string("op_42150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42150_cast_fp16 = slice_by_index(begin = var_42150_begin_0, end = var_42150_end_0, end_mask = var_42150_end_mask_0, x = k_55_cast_fp16)[name = string("op_42150_cast_fp16")]; + tensor var_42154_begin_0 = const()[name = string("op_42154_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_42154_end_0 = const()[name = string("op_42154_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_42154_end_mask_0 = const()[name = string("op_42154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42154_cast_fp16 = slice_by_index(begin = var_42154_begin_0, end = var_42154_end_0, end_mask = var_42154_end_mask_0, x = k_55_cast_fp16)[name = string("op_42154_cast_fp16")]; + tensor var_42158_begin_0 = const()[name = string("op_42158_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_42158_end_0 = const()[name = string("op_42158_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_42158_end_mask_0 = const()[name = string("op_42158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42158_cast_fp16 = slice_by_index(begin = var_42158_begin_0, end = var_42158_end_0, end_mask = var_42158_end_mask_0, x = k_55_cast_fp16)[name = string("op_42158_cast_fp16")]; + tensor var_42162_begin_0 = const()[name = string("op_42162_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_42162_end_0 = const()[name = string("op_42162_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_42162_end_mask_0 = const()[name = string("op_42162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42162_cast_fp16 = slice_by_index(begin = var_42162_begin_0, end = var_42162_end_0, end_mask = var_42162_end_mask_0, x = k_55_cast_fp16)[name = string("op_42162_cast_fp16")]; + tensor var_42166_begin_0 = const()[name = string("op_42166_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_42166_end_0 = const()[name = string("op_42166_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_42166_end_mask_0 = const()[name = string("op_42166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42166_cast_fp16 = slice_by_index(begin = var_42166_begin_0, end = var_42166_end_0, end_mask = var_42166_end_mask_0, x = k_55_cast_fp16)[name = string("op_42166_cast_fp16")]; + tensor var_42170_begin_0 = const()[name = string("op_42170_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_42170_end_0 = const()[name = string("op_42170_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_42170_end_mask_0 = const()[name = string("op_42170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42170_cast_fp16 = slice_by_index(begin = var_42170_begin_0, end = var_42170_end_0, end_mask = var_42170_end_mask_0, x = k_55_cast_fp16)[name = string("op_42170_cast_fp16")]; + tensor var_42174_begin_0 = const()[name = string("op_42174_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_42174_end_0 = const()[name = string("op_42174_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_42174_end_mask_0 = const()[name = string("op_42174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42174_cast_fp16 = slice_by_index(begin = var_42174_begin_0, end = var_42174_end_0, end_mask = var_42174_end_mask_0, x = k_55_cast_fp16)[name = string("op_42174_cast_fp16")]; + tensor var_42178_begin_0 = const()[name = string("op_42178_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_42178_end_0 = const()[name = string("op_42178_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_42178_end_mask_0 = const()[name = string("op_42178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42178_cast_fp16 = slice_by_index(begin = var_42178_begin_0, end = var_42178_end_0, end_mask = var_42178_end_mask_0, x = k_55_cast_fp16)[name = string("op_42178_cast_fp16")]; + tensor var_42182_begin_0 = const()[name = string("op_42182_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_42182_end_0 = const()[name = string("op_42182_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_42182_end_mask_0 = const()[name = string("op_42182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42182_cast_fp16 = slice_by_index(begin = var_42182_begin_0, end = var_42182_end_0, end_mask = var_42182_end_mask_0, x = k_55_cast_fp16)[name = string("op_42182_cast_fp16")]; + tensor var_42186_begin_0 = const()[name = string("op_42186_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_42186_end_0 = const()[name = string("op_42186_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_42186_end_mask_0 = const()[name = string("op_42186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42186_cast_fp16 = slice_by_index(begin = var_42186_begin_0, end = var_42186_end_0, end_mask = var_42186_end_mask_0, x = k_55_cast_fp16)[name = string("op_42186_cast_fp16")]; + tensor var_42190_begin_0 = const()[name = string("op_42190_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_42190_end_0 = const()[name = string("op_42190_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_42190_end_mask_0 = const()[name = string("op_42190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42190_cast_fp16 = slice_by_index(begin = var_42190_begin_0, end = var_42190_end_0, end_mask = var_42190_end_mask_0, x = k_55_cast_fp16)[name = string("op_42190_cast_fp16")]; + tensor var_42194_begin_0 = const()[name = string("op_42194_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_42194_end_0 = const()[name = string("op_42194_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_42194_end_mask_0 = const()[name = string("op_42194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42194_cast_fp16 = slice_by_index(begin = var_42194_begin_0, end = var_42194_end_0, end_mask = var_42194_end_mask_0, x = k_55_cast_fp16)[name = string("op_42194_cast_fp16")]; + tensor var_42198_begin_0 = const()[name = string("op_42198_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_42198_end_0 = const()[name = string("op_42198_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_42198_end_mask_0 = const()[name = string("op_42198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42198_cast_fp16 = slice_by_index(begin = var_42198_begin_0, end = var_42198_end_0, end_mask = var_42198_end_mask_0, x = k_55_cast_fp16)[name = string("op_42198_cast_fp16")]; + tensor var_42202_begin_0 = const()[name = string("op_42202_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_42202_end_0 = const()[name = string("op_42202_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_42202_end_mask_0 = const()[name = string("op_42202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42202_cast_fp16 = slice_by_index(begin = var_42202_begin_0, end = var_42202_end_0, end_mask = var_42202_end_mask_0, x = k_55_cast_fp16)[name = string("op_42202_cast_fp16")]; + tensor var_42206_begin_0 = const()[name = string("op_42206_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_42206_end_0 = const()[name = string("op_42206_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_42206_end_mask_0 = const()[name = string("op_42206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42206_cast_fp16 = slice_by_index(begin = var_42206_begin_0, end = var_42206_end_0, end_mask = var_42206_end_mask_0, x = k_55_cast_fp16)[name = string("op_42206_cast_fp16")]; + tensor var_42210_begin_0 = const()[name = string("op_42210_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_42210_end_0 = const()[name = string("op_42210_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_42210_end_mask_0 = const()[name = string("op_42210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42210_cast_fp16 = slice_by_index(begin = var_42210_begin_0, end = var_42210_end_0, end_mask = var_42210_end_mask_0, x = k_55_cast_fp16)[name = string("op_42210_cast_fp16")]; + tensor var_42212_begin_0 = const()[name = string("op_42212_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42212_end_0 = const()[name = string("op_42212_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42212_end_mask_0 = const()[name = string("op_42212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42212_cast_fp16 = slice_by_index(begin = var_42212_begin_0, end = var_42212_end_0, end_mask = var_42212_end_mask_0, x = value_55_cast_fp16)[name = string("op_42212_cast_fp16")]; + tensor var_42216_begin_0 = const()[name = string("op_42216_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42216_end_0 = const()[name = string("op_42216_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_42216_end_mask_0 = const()[name = string("op_42216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42216_cast_fp16 = slice_by_index(begin = var_42216_begin_0, end = var_42216_end_0, end_mask = var_42216_end_mask_0, x = value_55_cast_fp16)[name = string("op_42216_cast_fp16")]; + tensor var_42220_begin_0 = const()[name = string("op_42220_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42220_end_0 = const()[name = string("op_42220_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_42220_end_mask_0 = const()[name = string("op_42220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42220_cast_fp16 = slice_by_index(begin = var_42220_begin_0, end = var_42220_end_0, end_mask = var_42220_end_mask_0, x = value_55_cast_fp16)[name = string("op_42220_cast_fp16")]; + tensor var_42224_begin_0 = const()[name = string("op_42224_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42224_end_0 = const()[name = string("op_42224_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_42224_end_mask_0 = const()[name = string("op_42224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42224_cast_fp16 = slice_by_index(begin = var_42224_begin_0, end = var_42224_end_0, end_mask = var_42224_end_mask_0, x = value_55_cast_fp16)[name = string("op_42224_cast_fp16")]; + tensor var_42228_begin_0 = const()[name = string("op_42228_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42228_end_0 = const()[name = string("op_42228_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_42228_end_mask_0 = const()[name = string("op_42228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42228_cast_fp16 = slice_by_index(begin = var_42228_begin_0, end = var_42228_end_0, end_mask = var_42228_end_mask_0, x = value_55_cast_fp16)[name = string("op_42228_cast_fp16")]; + tensor var_42232_begin_0 = const()[name = string("op_42232_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42232_end_0 = const()[name = string("op_42232_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_42232_end_mask_0 = const()[name = string("op_42232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42232_cast_fp16 = slice_by_index(begin = var_42232_begin_0, end = var_42232_end_0, end_mask = var_42232_end_mask_0, x = value_55_cast_fp16)[name = string("op_42232_cast_fp16")]; + tensor var_42236_begin_0 = const()[name = string("op_42236_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42236_end_0 = const()[name = string("op_42236_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_42236_end_mask_0 = const()[name = string("op_42236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42236_cast_fp16 = slice_by_index(begin = var_42236_begin_0, end = var_42236_end_0, end_mask = var_42236_end_mask_0, x = value_55_cast_fp16)[name = string("op_42236_cast_fp16")]; + tensor var_42240_begin_0 = const()[name = string("op_42240_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42240_end_0 = const()[name = string("op_42240_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_42240_end_mask_0 = const()[name = string("op_42240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42240_cast_fp16 = slice_by_index(begin = var_42240_begin_0, end = var_42240_end_0, end_mask = var_42240_end_mask_0, x = value_55_cast_fp16)[name = string("op_42240_cast_fp16")]; + tensor var_42244_begin_0 = const()[name = string("op_42244_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42244_end_0 = const()[name = string("op_42244_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_42244_end_mask_0 = const()[name = string("op_42244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42244_cast_fp16 = slice_by_index(begin = var_42244_begin_0, end = var_42244_end_0, end_mask = var_42244_end_mask_0, x = value_55_cast_fp16)[name = string("op_42244_cast_fp16")]; + tensor var_42248_begin_0 = const()[name = string("op_42248_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42248_end_0 = const()[name = string("op_42248_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_42248_end_mask_0 = const()[name = string("op_42248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42248_cast_fp16 = slice_by_index(begin = var_42248_begin_0, end = var_42248_end_0, end_mask = var_42248_end_mask_0, x = value_55_cast_fp16)[name = string("op_42248_cast_fp16")]; + tensor var_42252_begin_0 = const()[name = string("op_42252_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42252_end_0 = const()[name = string("op_42252_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_42252_end_mask_0 = const()[name = string("op_42252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42252_cast_fp16 = slice_by_index(begin = var_42252_begin_0, end = var_42252_end_0, end_mask = var_42252_end_mask_0, x = value_55_cast_fp16)[name = string("op_42252_cast_fp16")]; + tensor var_42256_begin_0 = const()[name = string("op_42256_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42256_end_0 = const()[name = string("op_42256_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_42256_end_mask_0 = const()[name = string("op_42256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42256_cast_fp16 = slice_by_index(begin = var_42256_begin_0, end = var_42256_end_0, end_mask = var_42256_end_mask_0, x = value_55_cast_fp16)[name = string("op_42256_cast_fp16")]; + tensor var_42260_begin_0 = const()[name = string("op_42260_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42260_end_0 = const()[name = string("op_42260_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_42260_end_mask_0 = const()[name = string("op_42260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42260_cast_fp16 = slice_by_index(begin = var_42260_begin_0, end = var_42260_end_0, end_mask = var_42260_end_mask_0, x = value_55_cast_fp16)[name = string("op_42260_cast_fp16")]; + tensor var_42264_begin_0 = const()[name = string("op_42264_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42264_end_0 = const()[name = string("op_42264_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_42264_end_mask_0 = const()[name = string("op_42264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42264_cast_fp16 = slice_by_index(begin = var_42264_begin_0, end = var_42264_end_0, end_mask = var_42264_end_mask_0, x = value_55_cast_fp16)[name = string("op_42264_cast_fp16")]; + tensor var_42268_begin_0 = const()[name = string("op_42268_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42268_end_0 = const()[name = string("op_42268_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_42268_end_mask_0 = const()[name = string("op_42268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42268_cast_fp16 = slice_by_index(begin = var_42268_begin_0, end = var_42268_end_0, end_mask = var_42268_end_mask_0, x = value_55_cast_fp16)[name = string("op_42268_cast_fp16")]; + tensor var_42272_begin_0 = const()[name = string("op_42272_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42272_end_0 = const()[name = string("op_42272_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_42272_end_mask_0 = const()[name = string("op_42272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42272_cast_fp16 = slice_by_index(begin = var_42272_begin_0, end = var_42272_end_0, end_mask = var_42272_end_mask_0, x = value_55_cast_fp16)[name = string("op_42272_cast_fp16")]; + tensor var_42276_begin_0 = const()[name = string("op_42276_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42276_end_0 = const()[name = string("op_42276_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_42276_end_mask_0 = const()[name = string("op_42276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42276_cast_fp16 = slice_by_index(begin = var_42276_begin_0, end = var_42276_end_0, end_mask = var_42276_end_mask_0, x = value_55_cast_fp16)[name = string("op_42276_cast_fp16")]; + tensor var_42280_begin_0 = const()[name = string("op_42280_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42280_end_0 = const()[name = string("op_42280_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_42280_end_mask_0 = const()[name = string("op_42280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42280_cast_fp16 = slice_by_index(begin = var_42280_begin_0, end = var_42280_end_0, end_mask = var_42280_end_mask_0, x = value_55_cast_fp16)[name = string("op_42280_cast_fp16")]; + tensor var_42284_begin_0 = const()[name = string("op_42284_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42284_end_0 = const()[name = string("op_42284_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_42284_end_mask_0 = const()[name = string("op_42284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42284_cast_fp16 = slice_by_index(begin = var_42284_begin_0, end = var_42284_end_0, end_mask = var_42284_end_mask_0, x = value_55_cast_fp16)[name = string("op_42284_cast_fp16")]; + tensor var_42288_begin_0 = const()[name = string("op_42288_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42288_end_0 = const()[name = string("op_42288_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_42288_end_mask_0 = const()[name = string("op_42288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42288_cast_fp16 = slice_by_index(begin = var_42288_begin_0, end = var_42288_end_0, end_mask = var_42288_end_mask_0, x = value_55_cast_fp16)[name = string("op_42288_cast_fp16")]; + string _SplitHeadsQ__mh_w_4321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4321_equation_0, values = (var_42134_cast_fp16, var_41576_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4321_cast_fp16")]; + string _SplitHeadsQ__mh_w_4323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4323_equation_0, values = (var_42134_cast_fp16, var_41583_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4323_cast_fp16")]; + string _SplitHeadsQ__mh_w_4325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4325_equation_0, values = (var_42134_cast_fp16, var_41590_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4325_cast_fp16")]; + string _SplitHeadsQ__mh_w_4327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4327_equation_0, values = (var_42134_cast_fp16, var_41597_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4327_cast_fp16")]; + string _SplitHeadsQ__mh_w_4329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4329_equation_0, values = (var_42138_cast_fp16, var_41604_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4329_cast_fp16")]; + string _SplitHeadsQ__mh_w_4331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4331_equation_0, values = (var_42138_cast_fp16, var_41611_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4331_cast_fp16")]; + string _SplitHeadsQ__mh_w_4333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4333_equation_0, values = (var_42138_cast_fp16, var_41618_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4333_cast_fp16")]; + string _SplitHeadsQ__mh_w_4335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4335_equation_0, values = (var_42138_cast_fp16, var_41625_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4335_cast_fp16")]; + string _SplitHeadsQ__mh_w_4337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4337_equation_0, values = (var_42142_cast_fp16, var_41632_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4337_cast_fp16")]; + string _SplitHeadsQ__mh_w_4339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4339_equation_0, values = (var_42142_cast_fp16, var_41639_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4339_cast_fp16")]; + string _SplitHeadsQ__mh_w_4341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4341_equation_0, values = (var_42142_cast_fp16, var_41646_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4341_cast_fp16")]; + string _SplitHeadsQ__mh_w_4343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4343_equation_0, values = (var_42142_cast_fp16, var_41653_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4343_cast_fp16")]; + string _SplitHeadsQ__mh_w_4345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4345_equation_0, values = (var_42146_cast_fp16, var_41660_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4345_cast_fp16")]; + string _SplitHeadsQ__mh_w_4347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4347_equation_0, values = (var_42146_cast_fp16, var_41667_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4347_cast_fp16")]; + string _SplitHeadsQ__mh_w_4349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4349_equation_0, values = (var_42146_cast_fp16, var_41674_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4349_cast_fp16")]; + string _SplitHeadsQ__mh_w_4351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4351_equation_0, values = (var_42146_cast_fp16, var_41681_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4351_cast_fp16")]; + string _SplitHeadsQ__mh_w_4353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4353_equation_0, values = (var_42150_cast_fp16, var_41688_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4353_cast_fp16")]; + string _SplitHeadsQ__mh_w_4355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4355_equation_0, values = (var_42150_cast_fp16, var_41695_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4355_cast_fp16")]; + string _SplitHeadsQ__mh_w_4357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4357_equation_0, values = (var_42150_cast_fp16, var_41702_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4357_cast_fp16")]; + string _SplitHeadsQ__mh_w_4359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4359_equation_0, values = (var_42150_cast_fp16, var_41709_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4359_cast_fp16")]; + string _SplitHeadsQ__mh_w_4361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4361_equation_0, values = (var_42154_cast_fp16, var_41716_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4361_cast_fp16")]; + string _SplitHeadsQ__mh_w_4363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4363_equation_0, values = (var_42154_cast_fp16, var_41723_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4363_cast_fp16")]; + string _SplitHeadsQ__mh_w_4365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4365_equation_0, values = (var_42154_cast_fp16, var_41730_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4365_cast_fp16")]; + string _SplitHeadsQ__mh_w_4367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4367_equation_0, values = (var_42154_cast_fp16, var_41737_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4367_cast_fp16")]; + string _SplitHeadsQ__mh_w_4369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4369_equation_0, values = (var_42158_cast_fp16, var_41744_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4369_cast_fp16")]; + string _SplitHeadsQ__mh_w_4371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4371_equation_0, values = (var_42158_cast_fp16, var_41751_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4371_cast_fp16")]; + string _SplitHeadsQ__mh_w_4373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4373_equation_0, values = (var_42158_cast_fp16, var_41758_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4373_cast_fp16")]; + string _SplitHeadsQ__mh_w_4375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4375_equation_0, values = (var_42158_cast_fp16, var_41765_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4375_cast_fp16")]; + string _SplitHeadsQ__mh_w_4377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4377_equation_0, values = (var_42162_cast_fp16, var_41772_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4377_cast_fp16")]; + string _SplitHeadsQ__mh_w_4379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4379_equation_0, values = (var_42162_cast_fp16, var_41779_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4379_cast_fp16")]; + string _SplitHeadsQ__mh_w_4381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4381_equation_0, values = (var_42162_cast_fp16, var_41786_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4381_cast_fp16")]; + string _SplitHeadsQ__mh_w_4383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4383_equation_0, values = (var_42162_cast_fp16, var_41793_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4383_cast_fp16")]; + string _SplitHeadsQ__mh_w_4385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4385_equation_0, values = (var_42166_cast_fp16, var_41800_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4385_cast_fp16")]; + string _SplitHeadsQ__mh_w_4387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4387_equation_0, values = (var_42166_cast_fp16, var_41807_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4387_cast_fp16")]; + string _SplitHeadsQ__mh_w_4389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4389_equation_0, values = (var_42166_cast_fp16, var_41814_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4389_cast_fp16")]; + string _SplitHeadsQ__mh_w_4391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4391_equation_0, values = (var_42166_cast_fp16, var_41821_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4391_cast_fp16")]; + string _SplitHeadsQ__mh_w_4393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4393_equation_0, values = (var_42170_cast_fp16, var_41828_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4393_cast_fp16")]; + string _SplitHeadsQ__mh_w_4395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4395_equation_0, values = (var_42170_cast_fp16, var_41835_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4395_cast_fp16")]; + string _SplitHeadsQ__mh_w_4397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4397_equation_0, values = (var_42170_cast_fp16, var_41842_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4397_cast_fp16")]; + string _SplitHeadsQ__mh_w_4399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4399_equation_0, values = (var_42170_cast_fp16, var_41849_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4399_cast_fp16")]; + string _SplitHeadsQ__mh_w_4401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4401_equation_0, values = (var_42174_cast_fp16, var_41856_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4401_cast_fp16")]; + string _SplitHeadsQ__mh_w_4403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4403_equation_0, values = (var_42174_cast_fp16, var_41863_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4403_cast_fp16")]; + string _SplitHeadsQ__mh_w_4405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4405_equation_0, values = (var_42174_cast_fp16, var_41870_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4405_cast_fp16")]; + string _SplitHeadsQ__mh_w_4407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4407_equation_0, values = (var_42174_cast_fp16, var_41877_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4407_cast_fp16")]; + string _SplitHeadsQ__mh_w_4409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4409_equation_0, values = (var_42178_cast_fp16, var_41884_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4409_cast_fp16")]; + string _SplitHeadsQ__mh_w_4411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4411_equation_0, values = (var_42178_cast_fp16, var_41891_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4411_cast_fp16")]; + string _SplitHeadsQ__mh_w_4413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4413_equation_0, values = (var_42178_cast_fp16, var_41898_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4413_cast_fp16")]; + string _SplitHeadsQ__mh_w_4415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4415_equation_0, values = (var_42178_cast_fp16, var_41905_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4415_cast_fp16")]; + string _SplitHeadsQ__mh_w_4417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4417_equation_0, values = (var_42182_cast_fp16, var_41912_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4417_cast_fp16")]; + string _SplitHeadsQ__mh_w_4419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4419_equation_0, values = (var_42182_cast_fp16, var_41919_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4419_cast_fp16")]; + string _SplitHeadsQ__mh_w_4421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4421_equation_0, values = (var_42182_cast_fp16, var_41926_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4421_cast_fp16")]; + string _SplitHeadsQ__mh_w_4423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4423_equation_0, values = (var_42182_cast_fp16, var_41933_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4423_cast_fp16")]; + string _SplitHeadsQ__mh_w_4425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4425_equation_0, values = (var_42186_cast_fp16, var_41940_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4425_cast_fp16")]; + string _SplitHeadsQ__mh_w_4427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4427_equation_0, values = (var_42186_cast_fp16, var_41947_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4427_cast_fp16")]; + string _SplitHeadsQ__mh_w_4429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4429_equation_0, values = (var_42186_cast_fp16, var_41954_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4429_cast_fp16")]; + string _SplitHeadsQ__mh_w_4431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4431_equation_0, values = (var_42186_cast_fp16, var_41961_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4431_cast_fp16")]; + string _SplitHeadsQ__mh_w_4433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4433_equation_0, values = (var_42190_cast_fp16, var_41968_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4433_cast_fp16")]; + string _SplitHeadsQ__mh_w_4435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4435_equation_0, values = (var_42190_cast_fp16, var_41975_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4435_cast_fp16")]; + string _SplitHeadsQ__mh_w_4437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4437_equation_0, values = (var_42190_cast_fp16, var_41982_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4437_cast_fp16")]; + string _SplitHeadsQ__mh_w_4439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4439_equation_0, values = (var_42190_cast_fp16, var_41989_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4439_cast_fp16")]; + string _SplitHeadsQ__mh_w_4441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4441_equation_0, values = (var_42194_cast_fp16, var_41996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4441_cast_fp16")]; + string _SplitHeadsQ__mh_w_4443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4443_equation_0, values = (var_42194_cast_fp16, var_42003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4443_cast_fp16")]; + string _SplitHeadsQ__mh_w_4445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4445_equation_0, values = (var_42194_cast_fp16, var_42010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4445_cast_fp16")]; + string _SplitHeadsQ__mh_w_4447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4447_equation_0, values = (var_42194_cast_fp16, var_42017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4447_cast_fp16")]; + string _SplitHeadsQ__mh_w_4449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4449_equation_0, values = (var_42198_cast_fp16, var_42024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4449_cast_fp16")]; + string _SplitHeadsQ__mh_w_4451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4451_equation_0, values = (var_42198_cast_fp16, var_42031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4451_cast_fp16")]; + string _SplitHeadsQ__mh_w_4453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4453_equation_0, values = (var_42198_cast_fp16, var_42038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4453_cast_fp16")]; + string _SplitHeadsQ__mh_w_4455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4455_equation_0, values = (var_42198_cast_fp16, var_42045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4455_cast_fp16")]; + string _SplitHeadsQ__mh_w_4457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4457_equation_0, values = (var_42202_cast_fp16, var_42052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4457_cast_fp16")]; + string _SplitHeadsQ__mh_w_4459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4459_equation_0, values = (var_42202_cast_fp16, var_42059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4459_cast_fp16")]; + string _SplitHeadsQ__mh_w_4461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4461_equation_0, values = (var_42202_cast_fp16, var_42066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4461_cast_fp16")]; + string _SplitHeadsQ__mh_w_4463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4463_equation_0, values = (var_42202_cast_fp16, var_42073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4463_cast_fp16")]; + string _SplitHeadsQ__mh_w_4465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4465_equation_0, values = (var_42206_cast_fp16, var_42080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4465_cast_fp16")]; + string _SplitHeadsQ__mh_w_4467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4467_equation_0, values = (var_42206_cast_fp16, var_42087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4467_cast_fp16")]; + string _SplitHeadsQ__mh_w_4469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4469_equation_0, values = (var_42206_cast_fp16, var_42094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4469_cast_fp16")]; + string _SplitHeadsQ__mh_w_4471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4471_equation_0, values = (var_42206_cast_fp16, var_42101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4471_cast_fp16")]; + string _SplitHeadsQ__mh_w_4473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4473_equation_0, values = (var_42210_cast_fp16, var_42108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4473_cast_fp16")]; + string _SplitHeadsQ__mh_w_4475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4475_equation_0, values = (var_42210_cast_fp16, var_42115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4475_cast_fp16")]; + string _SplitHeadsQ__mh_w_4477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4477_equation_0, values = (var_42210_cast_fp16, var_42122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4477_cast_fp16")]; + string _SplitHeadsQ__mh_w_4479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4479_equation_0, values = (var_42210_cast_fp16, var_42129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4479_cast_fp16")]; + fp16 var_42451_to_fp16 = const()[name = string("op_42451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4321_cast_fp16, y = var_42451_to_fp16)[name = string("aw_chunk_4321_cast_fp16")]; + fp16 var_42453_to_fp16 = const()[name = string("op_42453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4323_cast_fp16, y = var_42453_to_fp16)[name = string("aw_chunk_4323_cast_fp16")]; + fp16 var_42455_to_fp16 = const()[name = string("op_42455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4325_cast_fp16, y = var_42455_to_fp16)[name = string("aw_chunk_4325_cast_fp16")]; + fp16 var_42457_to_fp16 = const()[name = string("op_42457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4327_cast_fp16, y = var_42457_to_fp16)[name = string("aw_chunk_4327_cast_fp16")]; + fp16 var_42459_to_fp16 = const()[name = string("op_42459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4329_cast_fp16, y = var_42459_to_fp16)[name = string("aw_chunk_4329_cast_fp16")]; + fp16 var_42461_to_fp16 = const()[name = string("op_42461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4331_cast_fp16, y = var_42461_to_fp16)[name = string("aw_chunk_4331_cast_fp16")]; + fp16 var_42463_to_fp16 = const()[name = string("op_42463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4333_cast_fp16, y = var_42463_to_fp16)[name = string("aw_chunk_4333_cast_fp16")]; + fp16 var_42465_to_fp16 = const()[name = string("op_42465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4335_cast_fp16, y = var_42465_to_fp16)[name = string("aw_chunk_4335_cast_fp16")]; + fp16 var_42467_to_fp16 = const()[name = string("op_42467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4337_cast_fp16, y = var_42467_to_fp16)[name = string("aw_chunk_4337_cast_fp16")]; + fp16 var_42469_to_fp16 = const()[name = string("op_42469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4339_cast_fp16, y = var_42469_to_fp16)[name = string("aw_chunk_4339_cast_fp16")]; + fp16 var_42471_to_fp16 = const()[name = string("op_42471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4341_cast_fp16, y = var_42471_to_fp16)[name = string("aw_chunk_4341_cast_fp16")]; + fp16 var_42473_to_fp16 = const()[name = string("op_42473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4343_cast_fp16, y = var_42473_to_fp16)[name = string("aw_chunk_4343_cast_fp16")]; + fp16 var_42475_to_fp16 = const()[name = string("op_42475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4345_cast_fp16, y = var_42475_to_fp16)[name = string("aw_chunk_4345_cast_fp16")]; + fp16 var_42477_to_fp16 = const()[name = string("op_42477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4347_cast_fp16, y = var_42477_to_fp16)[name = string("aw_chunk_4347_cast_fp16")]; + fp16 var_42479_to_fp16 = const()[name = string("op_42479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4349_cast_fp16, y = var_42479_to_fp16)[name = string("aw_chunk_4349_cast_fp16")]; + fp16 var_42481_to_fp16 = const()[name = string("op_42481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4351_cast_fp16, y = var_42481_to_fp16)[name = string("aw_chunk_4351_cast_fp16")]; + fp16 var_42483_to_fp16 = const()[name = string("op_42483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4353_cast_fp16, y = var_42483_to_fp16)[name = string("aw_chunk_4353_cast_fp16")]; + fp16 var_42485_to_fp16 = const()[name = string("op_42485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4355_cast_fp16, y = var_42485_to_fp16)[name = string("aw_chunk_4355_cast_fp16")]; + fp16 var_42487_to_fp16 = const()[name = string("op_42487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4357_cast_fp16, y = var_42487_to_fp16)[name = string("aw_chunk_4357_cast_fp16")]; + fp16 var_42489_to_fp16 = const()[name = string("op_42489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4359_cast_fp16, y = var_42489_to_fp16)[name = string("aw_chunk_4359_cast_fp16")]; + fp16 var_42491_to_fp16 = const()[name = string("op_42491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4361_cast_fp16, y = var_42491_to_fp16)[name = string("aw_chunk_4361_cast_fp16")]; + fp16 var_42493_to_fp16 = const()[name = string("op_42493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4363_cast_fp16, y = var_42493_to_fp16)[name = string("aw_chunk_4363_cast_fp16")]; + fp16 var_42495_to_fp16 = const()[name = string("op_42495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4365_cast_fp16, y = var_42495_to_fp16)[name = string("aw_chunk_4365_cast_fp16")]; + fp16 var_42497_to_fp16 = const()[name = string("op_42497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4367_cast_fp16, y = var_42497_to_fp16)[name = string("aw_chunk_4367_cast_fp16")]; + fp16 var_42499_to_fp16 = const()[name = string("op_42499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4369_cast_fp16, y = var_42499_to_fp16)[name = string("aw_chunk_4369_cast_fp16")]; + fp16 var_42501_to_fp16 = const()[name = string("op_42501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4371_cast_fp16, y = var_42501_to_fp16)[name = string("aw_chunk_4371_cast_fp16")]; + fp16 var_42503_to_fp16 = const()[name = string("op_42503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4373_cast_fp16, y = var_42503_to_fp16)[name = string("aw_chunk_4373_cast_fp16")]; + fp16 var_42505_to_fp16 = const()[name = string("op_42505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4375_cast_fp16, y = var_42505_to_fp16)[name = string("aw_chunk_4375_cast_fp16")]; + fp16 var_42507_to_fp16 = const()[name = string("op_42507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4377_cast_fp16, y = var_42507_to_fp16)[name = string("aw_chunk_4377_cast_fp16")]; + fp16 var_42509_to_fp16 = const()[name = string("op_42509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4379_cast_fp16, y = var_42509_to_fp16)[name = string("aw_chunk_4379_cast_fp16")]; + fp16 var_42511_to_fp16 = const()[name = string("op_42511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4381_cast_fp16, y = var_42511_to_fp16)[name = string("aw_chunk_4381_cast_fp16")]; + fp16 var_42513_to_fp16 = const()[name = string("op_42513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4383_cast_fp16, y = var_42513_to_fp16)[name = string("aw_chunk_4383_cast_fp16")]; + fp16 var_42515_to_fp16 = const()[name = string("op_42515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4385_cast_fp16, y = var_42515_to_fp16)[name = string("aw_chunk_4385_cast_fp16")]; + fp16 var_42517_to_fp16 = const()[name = string("op_42517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4387_cast_fp16, y = var_42517_to_fp16)[name = string("aw_chunk_4387_cast_fp16")]; + fp16 var_42519_to_fp16 = const()[name = string("op_42519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4389_cast_fp16, y = var_42519_to_fp16)[name = string("aw_chunk_4389_cast_fp16")]; + fp16 var_42521_to_fp16 = const()[name = string("op_42521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4391_cast_fp16, y = var_42521_to_fp16)[name = string("aw_chunk_4391_cast_fp16")]; + fp16 var_42523_to_fp16 = const()[name = string("op_42523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4393_cast_fp16, y = var_42523_to_fp16)[name = string("aw_chunk_4393_cast_fp16")]; + fp16 var_42525_to_fp16 = const()[name = string("op_42525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4395_cast_fp16, y = var_42525_to_fp16)[name = string("aw_chunk_4395_cast_fp16")]; + fp16 var_42527_to_fp16 = const()[name = string("op_42527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4397_cast_fp16, y = var_42527_to_fp16)[name = string("aw_chunk_4397_cast_fp16")]; + fp16 var_42529_to_fp16 = const()[name = string("op_42529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4399_cast_fp16, y = var_42529_to_fp16)[name = string("aw_chunk_4399_cast_fp16")]; + fp16 var_42531_to_fp16 = const()[name = string("op_42531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4401_cast_fp16, y = var_42531_to_fp16)[name = string("aw_chunk_4401_cast_fp16")]; + fp16 var_42533_to_fp16 = const()[name = string("op_42533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4403_cast_fp16, y = var_42533_to_fp16)[name = string("aw_chunk_4403_cast_fp16")]; + fp16 var_42535_to_fp16 = const()[name = string("op_42535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4405_cast_fp16, y = var_42535_to_fp16)[name = string("aw_chunk_4405_cast_fp16")]; + fp16 var_42537_to_fp16 = const()[name = string("op_42537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4407_cast_fp16, y = var_42537_to_fp16)[name = string("aw_chunk_4407_cast_fp16")]; + fp16 var_42539_to_fp16 = const()[name = string("op_42539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4409_cast_fp16, y = var_42539_to_fp16)[name = string("aw_chunk_4409_cast_fp16")]; + fp16 var_42541_to_fp16 = const()[name = string("op_42541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4411_cast_fp16, y = var_42541_to_fp16)[name = string("aw_chunk_4411_cast_fp16")]; + fp16 var_42543_to_fp16 = const()[name = string("op_42543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4413_cast_fp16, y = var_42543_to_fp16)[name = string("aw_chunk_4413_cast_fp16")]; + fp16 var_42545_to_fp16 = const()[name = string("op_42545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4415_cast_fp16, y = var_42545_to_fp16)[name = string("aw_chunk_4415_cast_fp16")]; + fp16 var_42547_to_fp16 = const()[name = string("op_42547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4417_cast_fp16, y = var_42547_to_fp16)[name = string("aw_chunk_4417_cast_fp16")]; + fp16 var_42549_to_fp16 = const()[name = string("op_42549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4419_cast_fp16, y = var_42549_to_fp16)[name = string("aw_chunk_4419_cast_fp16")]; + fp16 var_42551_to_fp16 = const()[name = string("op_42551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4421_cast_fp16, y = var_42551_to_fp16)[name = string("aw_chunk_4421_cast_fp16")]; + fp16 var_42553_to_fp16 = const()[name = string("op_42553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4423_cast_fp16, y = var_42553_to_fp16)[name = string("aw_chunk_4423_cast_fp16")]; + fp16 var_42555_to_fp16 = const()[name = string("op_42555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4425_cast_fp16, y = var_42555_to_fp16)[name = string("aw_chunk_4425_cast_fp16")]; + fp16 var_42557_to_fp16 = const()[name = string("op_42557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4427_cast_fp16, y = var_42557_to_fp16)[name = string("aw_chunk_4427_cast_fp16")]; + fp16 var_42559_to_fp16 = const()[name = string("op_42559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4429_cast_fp16, y = var_42559_to_fp16)[name = string("aw_chunk_4429_cast_fp16")]; + fp16 var_42561_to_fp16 = const()[name = string("op_42561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4431_cast_fp16, y = var_42561_to_fp16)[name = string("aw_chunk_4431_cast_fp16")]; + fp16 var_42563_to_fp16 = const()[name = string("op_42563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4433_cast_fp16, y = var_42563_to_fp16)[name = string("aw_chunk_4433_cast_fp16")]; + fp16 var_42565_to_fp16 = const()[name = string("op_42565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4435_cast_fp16, y = var_42565_to_fp16)[name = string("aw_chunk_4435_cast_fp16")]; + fp16 var_42567_to_fp16 = const()[name = string("op_42567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4437_cast_fp16, y = var_42567_to_fp16)[name = string("aw_chunk_4437_cast_fp16")]; + fp16 var_42569_to_fp16 = const()[name = string("op_42569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4439_cast_fp16, y = var_42569_to_fp16)[name = string("aw_chunk_4439_cast_fp16")]; + fp16 var_42571_to_fp16 = const()[name = string("op_42571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4441_cast_fp16, y = var_42571_to_fp16)[name = string("aw_chunk_4441_cast_fp16")]; + fp16 var_42573_to_fp16 = const()[name = string("op_42573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4443_cast_fp16, y = var_42573_to_fp16)[name = string("aw_chunk_4443_cast_fp16")]; + fp16 var_42575_to_fp16 = const()[name = string("op_42575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4445_cast_fp16, y = var_42575_to_fp16)[name = string("aw_chunk_4445_cast_fp16")]; + fp16 var_42577_to_fp16 = const()[name = string("op_42577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4447_cast_fp16, y = var_42577_to_fp16)[name = string("aw_chunk_4447_cast_fp16")]; + fp16 var_42579_to_fp16 = const()[name = string("op_42579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4449_cast_fp16, y = var_42579_to_fp16)[name = string("aw_chunk_4449_cast_fp16")]; + fp16 var_42581_to_fp16 = const()[name = string("op_42581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4451_cast_fp16, y = var_42581_to_fp16)[name = string("aw_chunk_4451_cast_fp16")]; + fp16 var_42583_to_fp16 = const()[name = string("op_42583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4453_cast_fp16, y = var_42583_to_fp16)[name = string("aw_chunk_4453_cast_fp16")]; + fp16 var_42585_to_fp16 = const()[name = string("op_42585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4455_cast_fp16, y = var_42585_to_fp16)[name = string("aw_chunk_4455_cast_fp16")]; + fp16 var_42587_to_fp16 = const()[name = string("op_42587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4457_cast_fp16, y = var_42587_to_fp16)[name = string("aw_chunk_4457_cast_fp16")]; + fp16 var_42589_to_fp16 = const()[name = string("op_42589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4459_cast_fp16, y = var_42589_to_fp16)[name = string("aw_chunk_4459_cast_fp16")]; + fp16 var_42591_to_fp16 = const()[name = string("op_42591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4461_cast_fp16, y = var_42591_to_fp16)[name = string("aw_chunk_4461_cast_fp16")]; + fp16 var_42593_to_fp16 = const()[name = string("op_42593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4463_cast_fp16, y = var_42593_to_fp16)[name = string("aw_chunk_4463_cast_fp16")]; + fp16 var_42595_to_fp16 = const()[name = string("op_42595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4465_cast_fp16, y = var_42595_to_fp16)[name = string("aw_chunk_4465_cast_fp16")]; + fp16 var_42597_to_fp16 = const()[name = string("op_42597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4467_cast_fp16, y = var_42597_to_fp16)[name = string("aw_chunk_4467_cast_fp16")]; + fp16 var_42599_to_fp16 = const()[name = string("op_42599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4469_cast_fp16, y = var_42599_to_fp16)[name = string("aw_chunk_4469_cast_fp16")]; + fp16 var_42601_to_fp16 = const()[name = string("op_42601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4471_cast_fp16, y = var_42601_to_fp16)[name = string("aw_chunk_4471_cast_fp16")]; + fp16 var_42603_to_fp16 = const()[name = string("op_42603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4473_cast_fp16, y = var_42603_to_fp16)[name = string("aw_chunk_4473_cast_fp16")]; + fp16 var_42605_to_fp16 = const()[name = string("op_42605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4475_cast_fp16, y = var_42605_to_fp16)[name = string("aw_chunk_4475_cast_fp16")]; + fp16 var_42607_to_fp16 = const()[name = string("op_42607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4477_cast_fp16, y = var_42607_to_fp16)[name = string("aw_chunk_4477_cast_fp16")]; + fp16 var_42609_to_fp16 = const()[name = string("op_42609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4479_cast_fp16, y = var_42609_to_fp16)[name = string("aw_chunk_4479_cast_fp16")]; + tensor var_42611_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4321_cast_fp16)[name = string("op_42611_cast_fp16")]; + tensor var_42612_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4323_cast_fp16)[name = string("op_42612_cast_fp16")]; + tensor var_42613_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4325_cast_fp16)[name = string("op_42613_cast_fp16")]; + tensor var_42614_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4327_cast_fp16)[name = string("op_42614_cast_fp16")]; + tensor var_42615_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4329_cast_fp16)[name = string("op_42615_cast_fp16")]; + tensor var_42616_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4331_cast_fp16)[name = string("op_42616_cast_fp16")]; + tensor var_42617_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4333_cast_fp16)[name = string("op_42617_cast_fp16")]; + tensor var_42618_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4335_cast_fp16)[name = string("op_42618_cast_fp16")]; + tensor var_42619_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4337_cast_fp16)[name = string("op_42619_cast_fp16")]; + tensor var_42620_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4339_cast_fp16)[name = string("op_42620_cast_fp16")]; + tensor var_42621_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4341_cast_fp16)[name = string("op_42621_cast_fp16")]; + tensor var_42622_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4343_cast_fp16)[name = string("op_42622_cast_fp16")]; + tensor var_42623_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4345_cast_fp16)[name = string("op_42623_cast_fp16")]; + tensor var_42624_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4347_cast_fp16)[name = string("op_42624_cast_fp16")]; + tensor var_42625_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4349_cast_fp16)[name = string("op_42625_cast_fp16")]; + tensor var_42626_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4351_cast_fp16)[name = string("op_42626_cast_fp16")]; + tensor var_42627_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4353_cast_fp16)[name = string("op_42627_cast_fp16")]; + tensor var_42628_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4355_cast_fp16)[name = string("op_42628_cast_fp16")]; + tensor var_42629_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4357_cast_fp16)[name = string("op_42629_cast_fp16")]; + tensor var_42630_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4359_cast_fp16)[name = string("op_42630_cast_fp16")]; + tensor var_42631_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4361_cast_fp16)[name = string("op_42631_cast_fp16")]; + tensor var_42632_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4363_cast_fp16)[name = string("op_42632_cast_fp16")]; + tensor var_42633_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4365_cast_fp16)[name = string("op_42633_cast_fp16")]; + tensor var_42634_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4367_cast_fp16)[name = string("op_42634_cast_fp16")]; + tensor var_42635_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4369_cast_fp16)[name = string("op_42635_cast_fp16")]; + tensor var_42636_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4371_cast_fp16)[name = string("op_42636_cast_fp16")]; + tensor var_42637_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4373_cast_fp16)[name = string("op_42637_cast_fp16")]; + tensor var_42638_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4375_cast_fp16)[name = string("op_42638_cast_fp16")]; + tensor var_42639_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4377_cast_fp16)[name = string("op_42639_cast_fp16")]; + tensor var_42640_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4379_cast_fp16)[name = string("op_42640_cast_fp16")]; + tensor var_42641_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4381_cast_fp16)[name = string("op_42641_cast_fp16")]; + tensor var_42642_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4383_cast_fp16)[name = string("op_42642_cast_fp16")]; + tensor var_42643_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4385_cast_fp16)[name = string("op_42643_cast_fp16")]; + tensor var_42644_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4387_cast_fp16)[name = string("op_42644_cast_fp16")]; + tensor var_42645_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4389_cast_fp16)[name = string("op_42645_cast_fp16")]; + tensor var_42646_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4391_cast_fp16)[name = string("op_42646_cast_fp16")]; + tensor var_42647_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4393_cast_fp16)[name = string("op_42647_cast_fp16")]; + tensor var_42648_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4395_cast_fp16)[name = string("op_42648_cast_fp16")]; + tensor var_42649_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4397_cast_fp16)[name = string("op_42649_cast_fp16")]; + tensor var_42650_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4399_cast_fp16)[name = string("op_42650_cast_fp16")]; + tensor var_42651_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4401_cast_fp16)[name = string("op_42651_cast_fp16")]; + tensor var_42652_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4403_cast_fp16)[name = string("op_42652_cast_fp16")]; + tensor var_42653_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4405_cast_fp16)[name = string("op_42653_cast_fp16")]; + tensor var_42654_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4407_cast_fp16)[name = string("op_42654_cast_fp16")]; + tensor var_42655_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4409_cast_fp16)[name = string("op_42655_cast_fp16")]; + tensor var_42656_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4411_cast_fp16)[name = string("op_42656_cast_fp16")]; + tensor var_42657_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4413_cast_fp16)[name = string("op_42657_cast_fp16")]; + tensor var_42658_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4415_cast_fp16)[name = string("op_42658_cast_fp16")]; + tensor var_42659_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4417_cast_fp16)[name = string("op_42659_cast_fp16")]; + tensor var_42660_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4419_cast_fp16)[name = string("op_42660_cast_fp16")]; + tensor var_42661_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4421_cast_fp16)[name = string("op_42661_cast_fp16")]; + tensor var_42662_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4423_cast_fp16)[name = string("op_42662_cast_fp16")]; + tensor var_42663_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4425_cast_fp16)[name = string("op_42663_cast_fp16")]; + tensor var_42664_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4427_cast_fp16)[name = string("op_42664_cast_fp16")]; + tensor var_42665_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4429_cast_fp16)[name = string("op_42665_cast_fp16")]; + tensor var_42666_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4431_cast_fp16)[name = string("op_42666_cast_fp16")]; + tensor var_42667_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4433_cast_fp16)[name = string("op_42667_cast_fp16")]; + tensor var_42668_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4435_cast_fp16)[name = string("op_42668_cast_fp16")]; + tensor var_42669_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4437_cast_fp16)[name = string("op_42669_cast_fp16")]; + tensor var_42670_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4439_cast_fp16)[name = string("op_42670_cast_fp16")]; + tensor var_42671_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4441_cast_fp16)[name = string("op_42671_cast_fp16")]; + tensor var_42672_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4443_cast_fp16)[name = string("op_42672_cast_fp16")]; + tensor var_42673_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4445_cast_fp16)[name = string("op_42673_cast_fp16")]; + tensor var_42674_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4447_cast_fp16)[name = string("op_42674_cast_fp16")]; + tensor var_42675_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4449_cast_fp16)[name = string("op_42675_cast_fp16")]; + tensor var_42676_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4451_cast_fp16)[name = string("op_42676_cast_fp16")]; + tensor var_42677_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4453_cast_fp16)[name = string("op_42677_cast_fp16")]; + tensor var_42678_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4455_cast_fp16)[name = string("op_42678_cast_fp16")]; + tensor var_42679_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4457_cast_fp16)[name = string("op_42679_cast_fp16")]; + tensor var_42680_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4459_cast_fp16)[name = string("op_42680_cast_fp16")]; + tensor var_42681_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4461_cast_fp16)[name = string("op_42681_cast_fp16")]; + tensor var_42682_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4463_cast_fp16)[name = string("op_42682_cast_fp16")]; + tensor var_42683_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4465_cast_fp16)[name = string("op_42683_cast_fp16")]; + tensor var_42684_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4467_cast_fp16)[name = string("op_42684_cast_fp16")]; + tensor var_42685_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4469_cast_fp16)[name = string("op_42685_cast_fp16")]; + tensor var_42686_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4471_cast_fp16)[name = string("op_42686_cast_fp16")]; + tensor var_42687_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4473_cast_fp16)[name = string("op_42687_cast_fp16")]; + tensor var_42688_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4475_cast_fp16)[name = string("op_42688_cast_fp16")]; + tensor var_42689_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4477_cast_fp16)[name = string("op_42689_cast_fp16")]; + tensor var_42690_cast_fp16 = softmax(axis = var_41436, x = aw_chunk_4479_cast_fp16)[name = string("op_42690_cast_fp16")]; + string var_42692_equation_0 = const()[name = string("op_42692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42692_cast_fp16 = einsum(equation = var_42692_equation_0, values = (var_42212_cast_fp16, var_42611_cast_fp16))[name = string("op_42692_cast_fp16")]; + string var_42694_equation_0 = const()[name = string("op_42694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42694_cast_fp16 = einsum(equation = var_42694_equation_0, values = (var_42212_cast_fp16, var_42612_cast_fp16))[name = string("op_42694_cast_fp16")]; + string var_42696_equation_0 = const()[name = string("op_42696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42696_cast_fp16 = einsum(equation = var_42696_equation_0, values = (var_42212_cast_fp16, var_42613_cast_fp16))[name = string("op_42696_cast_fp16")]; + string var_42698_equation_0 = const()[name = string("op_42698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42698_cast_fp16 = einsum(equation = var_42698_equation_0, values = (var_42212_cast_fp16, var_42614_cast_fp16))[name = string("op_42698_cast_fp16")]; + string var_42700_equation_0 = const()[name = string("op_42700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42700_cast_fp16 = einsum(equation = var_42700_equation_0, values = (var_42216_cast_fp16, var_42615_cast_fp16))[name = string("op_42700_cast_fp16")]; + string var_42702_equation_0 = const()[name = string("op_42702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42702_cast_fp16 = einsum(equation = var_42702_equation_0, values = (var_42216_cast_fp16, var_42616_cast_fp16))[name = string("op_42702_cast_fp16")]; + string var_42704_equation_0 = const()[name = string("op_42704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42704_cast_fp16 = einsum(equation = var_42704_equation_0, values = (var_42216_cast_fp16, var_42617_cast_fp16))[name = string("op_42704_cast_fp16")]; + string var_42706_equation_0 = const()[name = string("op_42706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42706_cast_fp16 = einsum(equation = var_42706_equation_0, values = (var_42216_cast_fp16, var_42618_cast_fp16))[name = string("op_42706_cast_fp16")]; + string var_42708_equation_0 = const()[name = string("op_42708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42708_cast_fp16 = einsum(equation = var_42708_equation_0, values = (var_42220_cast_fp16, var_42619_cast_fp16))[name = string("op_42708_cast_fp16")]; + string var_42710_equation_0 = const()[name = string("op_42710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42710_cast_fp16 = einsum(equation = var_42710_equation_0, values = (var_42220_cast_fp16, var_42620_cast_fp16))[name = string("op_42710_cast_fp16")]; + string var_42712_equation_0 = const()[name = string("op_42712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42712_cast_fp16 = einsum(equation = var_42712_equation_0, values = (var_42220_cast_fp16, var_42621_cast_fp16))[name = string("op_42712_cast_fp16")]; + string var_42714_equation_0 = const()[name = string("op_42714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42714_cast_fp16 = einsum(equation = var_42714_equation_0, values = (var_42220_cast_fp16, var_42622_cast_fp16))[name = string("op_42714_cast_fp16")]; + string var_42716_equation_0 = const()[name = string("op_42716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42716_cast_fp16 = einsum(equation = var_42716_equation_0, values = (var_42224_cast_fp16, var_42623_cast_fp16))[name = string("op_42716_cast_fp16")]; + string var_42718_equation_0 = const()[name = string("op_42718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42718_cast_fp16 = einsum(equation = var_42718_equation_0, values = (var_42224_cast_fp16, var_42624_cast_fp16))[name = string("op_42718_cast_fp16")]; + string var_42720_equation_0 = const()[name = string("op_42720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42720_cast_fp16 = einsum(equation = var_42720_equation_0, values = (var_42224_cast_fp16, var_42625_cast_fp16))[name = string("op_42720_cast_fp16")]; + string var_42722_equation_0 = const()[name = string("op_42722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42722_cast_fp16 = einsum(equation = var_42722_equation_0, values = (var_42224_cast_fp16, var_42626_cast_fp16))[name = string("op_42722_cast_fp16")]; + string var_42724_equation_0 = const()[name = string("op_42724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42724_cast_fp16 = einsum(equation = var_42724_equation_0, values = (var_42228_cast_fp16, var_42627_cast_fp16))[name = string("op_42724_cast_fp16")]; + string var_42726_equation_0 = const()[name = string("op_42726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42726_cast_fp16 = einsum(equation = var_42726_equation_0, values = (var_42228_cast_fp16, var_42628_cast_fp16))[name = string("op_42726_cast_fp16")]; + string var_42728_equation_0 = const()[name = string("op_42728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42728_cast_fp16 = einsum(equation = var_42728_equation_0, values = (var_42228_cast_fp16, var_42629_cast_fp16))[name = string("op_42728_cast_fp16")]; + string var_42730_equation_0 = const()[name = string("op_42730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42730_cast_fp16 = einsum(equation = var_42730_equation_0, values = (var_42228_cast_fp16, var_42630_cast_fp16))[name = string("op_42730_cast_fp16")]; + string var_42732_equation_0 = const()[name = string("op_42732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42732_cast_fp16 = einsum(equation = var_42732_equation_0, values = (var_42232_cast_fp16, var_42631_cast_fp16))[name = string("op_42732_cast_fp16")]; + string var_42734_equation_0 = const()[name = string("op_42734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42734_cast_fp16 = einsum(equation = var_42734_equation_0, values = (var_42232_cast_fp16, var_42632_cast_fp16))[name = string("op_42734_cast_fp16")]; + string var_42736_equation_0 = const()[name = string("op_42736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42736_cast_fp16 = einsum(equation = var_42736_equation_0, values = (var_42232_cast_fp16, var_42633_cast_fp16))[name = string("op_42736_cast_fp16")]; + string var_42738_equation_0 = const()[name = string("op_42738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42738_cast_fp16 = einsum(equation = var_42738_equation_0, values = (var_42232_cast_fp16, var_42634_cast_fp16))[name = string("op_42738_cast_fp16")]; + string var_42740_equation_0 = const()[name = string("op_42740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42740_cast_fp16 = einsum(equation = var_42740_equation_0, values = (var_42236_cast_fp16, var_42635_cast_fp16))[name = string("op_42740_cast_fp16")]; + string var_42742_equation_0 = const()[name = string("op_42742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42742_cast_fp16 = einsum(equation = var_42742_equation_0, values = (var_42236_cast_fp16, var_42636_cast_fp16))[name = string("op_42742_cast_fp16")]; + string var_42744_equation_0 = const()[name = string("op_42744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42744_cast_fp16 = einsum(equation = var_42744_equation_0, values = (var_42236_cast_fp16, var_42637_cast_fp16))[name = string("op_42744_cast_fp16")]; + string var_42746_equation_0 = const()[name = string("op_42746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42746_cast_fp16 = einsum(equation = var_42746_equation_0, values = (var_42236_cast_fp16, var_42638_cast_fp16))[name = string("op_42746_cast_fp16")]; + string var_42748_equation_0 = const()[name = string("op_42748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42748_cast_fp16 = einsum(equation = var_42748_equation_0, values = (var_42240_cast_fp16, var_42639_cast_fp16))[name = string("op_42748_cast_fp16")]; + string var_42750_equation_0 = const()[name = string("op_42750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42750_cast_fp16 = einsum(equation = var_42750_equation_0, values = (var_42240_cast_fp16, var_42640_cast_fp16))[name = string("op_42750_cast_fp16")]; + string var_42752_equation_0 = const()[name = string("op_42752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42752_cast_fp16 = einsum(equation = var_42752_equation_0, values = (var_42240_cast_fp16, var_42641_cast_fp16))[name = string("op_42752_cast_fp16")]; + string var_42754_equation_0 = const()[name = string("op_42754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42754_cast_fp16 = einsum(equation = var_42754_equation_0, values = (var_42240_cast_fp16, var_42642_cast_fp16))[name = string("op_42754_cast_fp16")]; + string var_42756_equation_0 = const()[name = string("op_42756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42756_cast_fp16 = einsum(equation = var_42756_equation_0, values = (var_42244_cast_fp16, var_42643_cast_fp16))[name = string("op_42756_cast_fp16")]; + string var_42758_equation_0 = const()[name = string("op_42758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42758_cast_fp16 = einsum(equation = var_42758_equation_0, values = (var_42244_cast_fp16, var_42644_cast_fp16))[name = string("op_42758_cast_fp16")]; + string var_42760_equation_0 = const()[name = string("op_42760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42760_cast_fp16 = einsum(equation = var_42760_equation_0, values = (var_42244_cast_fp16, var_42645_cast_fp16))[name = string("op_42760_cast_fp16")]; + string var_42762_equation_0 = const()[name = string("op_42762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42762_cast_fp16 = einsum(equation = var_42762_equation_0, values = (var_42244_cast_fp16, var_42646_cast_fp16))[name = string("op_42762_cast_fp16")]; + string var_42764_equation_0 = const()[name = string("op_42764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42764_cast_fp16 = einsum(equation = var_42764_equation_0, values = (var_42248_cast_fp16, var_42647_cast_fp16))[name = string("op_42764_cast_fp16")]; + string var_42766_equation_0 = const()[name = string("op_42766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42766_cast_fp16 = einsum(equation = var_42766_equation_0, values = (var_42248_cast_fp16, var_42648_cast_fp16))[name = string("op_42766_cast_fp16")]; + string var_42768_equation_0 = const()[name = string("op_42768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42768_cast_fp16 = einsum(equation = var_42768_equation_0, values = (var_42248_cast_fp16, var_42649_cast_fp16))[name = string("op_42768_cast_fp16")]; + string var_42770_equation_0 = const()[name = string("op_42770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42770_cast_fp16 = einsum(equation = var_42770_equation_0, values = (var_42248_cast_fp16, var_42650_cast_fp16))[name = string("op_42770_cast_fp16")]; + string var_42772_equation_0 = const()[name = string("op_42772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42772_cast_fp16 = einsum(equation = var_42772_equation_0, values = (var_42252_cast_fp16, var_42651_cast_fp16))[name = string("op_42772_cast_fp16")]; + string var_42774_equation_0 = const()[name = string("op_42774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42774_cast_fp16 = einsum(equation = var_42774_equation_0, values = (var_42252_cast_fp16, var_42652_cast_fp16))[name = string("op_42774_cast_fp16")]; + string var_42776_equation_0 = const()[name = string("op_42776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42776_cast_fp16 = einsum(equation = var_42776_equation_0, values = (var_42252_cast_fp16, var_42653_cast_fp16))[name = string("op_42776_cast_fp16")]; + string var_42778_equation_0 = const()[name = string("op_42778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42778_cast_fp16 = einsum(equation = var_42778_equation_0, values = (var_42252_cast_fp16, var_42654_cast_fp16))[name = string("op_42778_cast_fp16")]; + string var_42780_equation_0 = const()[name = string("op_42780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42780_cast_fp16 = einsum(equation = var_42780_equation_0, values = (var_42256_cast_fp16, var_42655_cast_fp16))[name = string("op_42780_cast_fp16")]; + string var_42782_equation_0 = const()[name = string("op_42782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42782_cast_fp16 = einsum(equation = var_42782_equation_0, values = (var_42256_cast_fp16, var_42656_cast_fp16))[name = string("op_42782_cast_fp16")]; + string var_42784_equation_0 = const()[name = string("op_42784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42784_cast_fp16 = einsum(equation = var_42784_equation_0, values = (var_42256_cast_fp16, var_42657_cast_fp16))[name = string("op_42784_cast_fp16")]; + string var_42786_equation_0 = const()[name = string("op_42786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42786_cast_fp16 = einsum(equation = var_42786_equation_0, values = (var_42256_cast_fp16, var_42658_cast_fp16))[name = string("op_42786_cast_fp16")]; + string var_42788_equation_0 = const()[name = string("op_42788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42788_cast_fp16 = einsum(equation = var_42788_equation_0, values = (var_42260_cast_fp16, var_42659_cast_fp16))[name = string("op_42788_cast_fp16")]; + string var_42790_equation_0 = const()[name = string("op_42790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42790_cast_fp16 = einsum(equation = var_42790_equation_0, values = (var_42260_cast_fp16, var_42660_cast_fp16))[name = string("op_42790_cast_fp16")]; + string var_42792_equation_0 = const()[name = string("op_42792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42792_cast_fp16 = einsum(equation = var_42792_equation_0, values = (var_42260_cast_fp16, var_42661_cast_fp16))[name = string("op_42792_cast_fp16")]; + string var_42794_equation_0 = const()[name = string("op_42794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42794_cast_fp16 = einsum(equation = var_42794_equation_0, values = (var_42260_cast_fp16, var_42662_cast_fp16))[name = string("op_42794_cast_fp16")]; + string var_42796_equation_0 = const()[name = string("op_42796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42796_cast_fp16 = einsum(equation = var_42796_equation_0, values = (var_42264_cast_fp16, var_42663_cast_fp16))[name = string("op_42796_cast_fp16")]; + string var_42798_equation_0 = const()[name = string("op_42798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42798_cast_fp16 = einsum(equation = var_42798_equation_0, values = (var_42264_cast_fp16, var_42664_cast_fp16))[name = string("op_42798_cast_fp16")]; + string var_42800_equation_0 = const()[name = string("op_42800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42800_cast_fp16 = einsum(equation = var_42800_equation_0, values = (var_42264_cast_fp16, var_42665_cast_fp16))[name = string("op_42800_cast_fp16")]; + string var_42802_equation_0 = const()[name = string("op_42802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42802_cast_fp16 = einsum(equation = var_42802_equation_0, values = (var_42264_cast_fp16, var_42666_cast_fp16))[name = string("op_42802_cast_fp16")]; + string var_42804_equation_0 = const()[name = string("op_42804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42804_cast_fp16 = einsum(equation = var_42804_equation_0, values = (var_42268_cast_fp16, var_42667_cast_fp16))[name = string("op_42804_cast_fp16")]; + string var_42806_equation_0 = const()[name = string("op_42806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42806_cast_fp16 = einsum(equation = var_42806_equation_0, values = (var_42268_cast_fp16, var_42668_cast_fp16))[name = string("op_42806_cast_fp16")]; + string var_42808_equation_0 = const()[name = string("op_42808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42808_cast_fp16 = einsum(equation = var_42808_equation_0, values = (var_42268_cast_fp16, var_42669_cast_fp16))[name = string("op_42808_cast_fp16")]; + string var_42810_equation_0 = const()[name = string("op_42810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42810_cast_fp16 = einsum(equation = var_42810_equation_0, values = (var_42268_cast_fp16, var_42670_cast_fp16))[name = string("op_42810_cast_fp16")]; + string var_42812_equation_0 = const()[name = string("op_42812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42812_cast_fp16 = einsum(equation = var_42812_equation_0, values = (var_42272_cast_fp16, var_42671_cast_fp16))[name = string("op_42812_cast_fp16")]; + string var_42814_equation_0 = const()[name = string("op_42814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42814_cast_fp16 = einsum(equation = var_42814_equation_0, values = (var_42272_cast_fp16, var_42672_cast_fp16))[name = string("op_42814_cast_fp16")]; + string var_42816_equation_0 = const()[name = string("op_42816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42816_cast_fp16 = einsum(equation = var_42816_equation_0, values = (var_42272_cast_fp16, var_42673_cast_fp16))[name = string("op_42816_cast_fp16")]; + string var_42818_equation_0 = const()[name = string("op_42818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42818_cast_fp16 = einsum(equation = var_42818_equation_0, values = (var_42272_cast_fp16, var_42674_cast_fp16))[name = string("op_42818_cast_fp16")]; + string var_42820_equation_0 = const()[name = string("op_42820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42820_cast_fp16 = einsum(equation = var_42820_equation_0, values = (var_42276_cast_fp16, var_42675_cast_fp16))[name = string("op_42820_cast_fp16")]; + string var_42822_equation_0 = const()[name = string("op_42822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42822_cast_fp16 = einsum(equation = var_42822_equation_0, values = (var_42276_cast_fp16, var_42676_cast_fp16))[name = string("op_42822_cast_fp16")]; + string var_42824_equation_0 = const()[name = string("op_42824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42824_cast_fp16 = einsum(equation = var_42824_equation_0, values = (var_42276_cast_fp16, var_42677_cast_fp16))[name = string("op_42824_cast_fp16")]; + string var_42826_equation_0 = const()[name = string("op_42826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42826_cast_fp16 = einsum(equation = var_42826_equation_0, values = (var_42276_cast_fp16, var_42678_cast_fp16))[name = string("op_42826_cast_fp16")]; + string var_42828_equation_0 = const()[name = string("op_42828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42828_cast_fp16 = einsum(equation = var_42828_equation_0, values = (var_42280_cast_fp16, var_42679_cast_fp16))[name = string("op_42828_cast_fp16")]; + string var_42830_equation_0 = const()[name = string("op_42830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42830_cast_fp16 = einsum(equation = var_42830_equation_0, values = (var_42280_cast_fp16, var_42680_cast_fp16))[name = string("op_42830_cast_fp16")]; + string var_42832_equation_0 = const()[name = string("op_42832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42832_cast_fp16 = einsum(equation = var_42832_equation_0, values = (var_42280_cast_fp16, var_42681_cast_fp16))[name = string("op_42832_cast_fp16")]; + string var_42834_equation_0 = const()[name = string("op_42834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42834_cast_fp16 = einsum(equation = var_42834_equation_0, values = (var_42280_cast_fp16, var_42682_cast_fp16))[name = string("op_42834_cast_fp16")]; + string var_42836_equation_0 = const()[name = string("op_42836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42836_cast_fp16 = einsum(equation = var_42836_equation_0, values = (var_42284_cast_fp16, var_42683_cast_fp16))[name = string("op_42836_cast_fp16")]; + string var_42838_equation_0 = const()[name = string("op_42838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42838_cast_fp16 = einsum(equation = var_42838_equation_0, values = (var_42284_cast_fp16, var_42684_cast_fp16))[name = string("op_42838_cast_fp16")]; + string var_42840_equation_0 = const()[name = string("op_42840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42840_cast_fp16 = einsum(equation = var_42840_equation_0, values = (var_42284_cast_fp16, var_42685_cast_fp16))[name = string("op_42840_cast_fp16")]; + string var_42842_equation_0 = const()[name = string("op_42842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42842_cast_fp16 = einsum(equation = var_42842_equation_0, values = (var_42284_cast_fp16, var_42686_cast_fp16))[name = string("op_42842_cast_fp16")]; + string var_42844_equation_0 = const()[name = string("op_42844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42844_cast_fp16 = einsum(equation = var_42844_equation_0, values = (var_42288_cast_fp16, var_42687_cast_fp16))[name = string("op_42844_cast_fp16")]; + string var_42846_equation_0 = const()[name = string("op_42846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42846_cast_fp16 = einsum(equation = var_42846_equation_0, values = (var_42288_cast_fp16, var_42688_cast_fp16))[name = string("op_42846_cast_fp16")]; + string var_42848_equation_0 = const()[name = string("op_42848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42848_cast_fp16 = einsum(equation = var_42848_equation_0, values = (var_42288_cast_fp16, var_42689_cast_fp16))[name = string("op_42848_cast_fp16")]; + string var_42850_equation_0 = const()[name = string("op_42850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_42850_cast_fp16 = einsum(equation = var_42850_equation_0, values = (var_42288_cast_fp16, var_42690_cast_fp16))[name = string("op_42850_cast_fp16")]; + bool var_42852_interleave_0 = const()[name = string("op_42852_interleave_0"), val = bool(false)]; + tensor var_42852_cast_fp16 = concat(axis = var_41411, interleave = var_42852_interleave_0, values = (var_42692_cast_fp16, var_42694_cast_fp16, var_42696_cast_fp16, var_42698_cast_fp16))[name = string("op_42852_cast_fp16")]; + bool var_42854_interleave_0 = const()[name = string("op_42854_interleave_0"), val = bool(false)]; + tensor var_42854_cast_fp16 = concat(axis = var_41411, interleave = var_42854_interleave_0, values = (var_42700_cast_fp16, var_42702_cast_fp16, var_42704_cast_fp16, var_42706_cast_fp16))[name = string("op_42854_cast_fp16")]; + bool var_42856_interleave_0 = const()[name = string("op_42856_interleave_0"), val = bool(false)]; + tensor var_42856_cast_fp16 = concat(axis = var_41411, interleave = var_42856_interleave_0, values = (var_42708_cast_fp16, var_42710_cast_fp16, var_42712_cast_fp16, var_42714_cast_fp16))[name = string("op_42856_cast_fp16")]; + bool var_42858_interleave_0 = const()[name = string("op_42858_interleave_0"), val = bool(false)]; + tensor var_42858_cast_fp16 = concat(axis = var_41411, interleave = var_42858_interleave_0, values = (var_42716_cast_fp16, var_42718_cast_fp16, var_42720_cast_fp16, var_42722_cast_fp16))[name = string("op_42858_cast_fp16")]; + bool var_42860_interleave_0 = const()[name = string("op_42860_interleave_0"), val = bool(false)]; + tensor var_42860_cast_fp16 = concat(axis = var_41411, interleave = var_42860_interleave_0, values = (var_42724_cast_fp16, var_42726_cast_fp16, var_42728_cast_fp16, var_42730_cast_fp16))[name = string("op_42860_cast_fp16")]; + bool var_42862_interleave_0 = const()[name = string("op_42862_interleave_0"), val = bool(false)]; + tensor var_42862_cast_fp16 = concat(axis = var_41411, interleave = var_42862_interleave_0, values = (var_42732_cast_fp16, var_42734_cast_fp16, var_42736_cast_fp16, var_42738_cast_fp16))[name = string("op_42862_cast_fp16")]; + bool var_42864_interleave_0 = const()[name = string("op_42864_interleave_0"), val = bool(false)]; + tensor var_42864_cast_fp16 = concat(axis = var_41411, interleave = var_42864_interleave_0, values = (var_42740_cast_fp16, var_42742_cast_fp16, var_42744_cast_fp16, var_42746_cast_fp16))[name = string("op_42864_cast_fp16")]; + bool var_42866_interleave_0 = const()[name = string("op_42866_interleave_0"), val = bool(false)]; + tensor var_42866_cast_fp16 = concat(axis = var_41411, interleave = var_42866_interleave_0, values = (var_42748_cast_fp16, var_42750_cast_fp16, var_42752_cast_fp16, var_42754_cast_fp16))[name = string("op_42866_cast_fp16")]; + bool var_42868_interleave_0 = const()[name = string("op_42868_interleave_0"), val = bool(false)]; + tensor var_42868_cast_fp16 = concat(axis = var_41411, interleave = var_42868_interleave_0, values = (var_42756_cast_fp16, var_42758_cast_fp16, var_42760_cast_fp16, var_42762_cast_fp16))[name = string("op_42868_cast_fp16")]; + bool var_42870_interleave_0 = const()[name = string("op_42870_interleave_0"), val = bool(false)]; + tensor var_42870_cast_fp16 = concat(axis = var_41411, interleave = var_42870_interleave_0, values = (var_42764_cast_fp16, var_42766_cast_fp16, var_42768_cast_fp16, var_42770_cast_fp16))[name = string("op_42870_cast_fp16")]; + bool var_42872_interleave_0 = const()[name = string("op_42872_interleave_0"), val = bool(false)]; + tensor var_42872_cast_fp16 = concat(axis = var_41411, interleave = var_42872_interleave_0, values = (var_42772_cast_fp16, var_42774_cast_fp16, var_42776_cast_fp16, var_42778_cast_fp16))[name = string("op_42872_cast_fp16")]; + bool var_42874_interleave_0 = const()[name = string("op_42874_interleave_0"), val = bool(false)]; + tensor var_42874_cast_fp16 = concat(axis = var_41411, interleave = var_42874_interleave_0, values = (var_42780_cast_fp16, var_42782_cast_fp16, var_42784_cast_fp16, var_42786_cast_fp16))[name = string("op_42874_cast_fp16")]; + bool var_42876_interleave_0 = const()[name = string("op_42876_interleave_0"), val = bool(false)]; + tensor var_42876_cast_fp16 = concat(axis = var_41411, interleave = var_42876_interleave_0, values = (var_42788_cast_fp16, var_42790_cast_fp16, var_42792_cast_fp16, var_42794_cast_fp16))[name = string("op_42876_cast_fp16")]; + bool var_42878_interleave_0 = const()[name = string("op_42878_interleave_0"), val = bool(false)]; + tensor var_42878_cast_fp16 = concat(axis = var_41411, interleave = var_42878_interleave_0, values = (var_42796_cast_fp16, var_42798_cast_fp16, var_42800_cast_fp16, var_42802_cast_fp16))[name = string("op_42878_cast_fp16")]; + bool var_42880_interleave_0 = const()[name = string("op_42880_interleave_0"), val = bool(false)]; + tensor var_42880_cast_fp16 = concat(axis = var_41411, interleave = var_42880_interleave_0, values = (var_42804_cast_fp16, var_42806_cast_fp16, var_42808_cast_fp16, var_42810_cast_fp16))[name = string("op_42880_cast_fp16")]; + bool var_42882_interleave_0 = const()[name = string("op_42882_interleave_0"), val = bool(false)]; + tensor var_42882_cast_fp16 = concat(axis = var_41411, interleave = var_42882_interleave_0, values = (var_42812_cast_fp16, var_42814_cast_fp16, var_42816_cast_fp16, var_42818_cast_fp16))[name = string("op_42882_cast_fp16")]; + bool var_42884_interleave_0 = const()[name = string("op_42884_interleave_0"), val = bool(false)]; + tensor var_42884_cast_fp16 = concat(axis = var_41411, interleave = var_42884_interleave_0, values = (var_42820_cast_fp16, var_42822_cast_fp16, var_42824_cast_fp16, var_42826_cast_fp16))[name = string("op_42884_cast_fp16")]; + bool var_42886_interleave_0 = const()[name = string("op_42886_interleave_0"), val = bool(false)]; + tensor var_42886_cast_fp16 = concat(axis = var_41411, interleave = var_42886_interleave_0, values = (var_42828_cast_fp16, var_42830_cast_fp16, var_42832_cast_fp16, var_42834_cast_fp16))[name = string("op_42886_cast_fp16")]; + bool var_42888_interleave_0 = const()[name = string("op_42888_interleave_0"), val = bool(false)]; + tensor var_42888_cast_fp16 = concat(axis = var_41411, interleave = var_42888_interleave_0, values = (var_42836_cast_fp16, var_42838_cast_fp16, var_42840_cast_fp16, var_42842_cast_fp16))[name = string("op_42888_cast_fp16")]; + bool var_42890_interleave_0 = const()[name = string("op_42890_interleave_0"), val = bool(false)]; + tensor var_42890_cast_fp16 = concat(axis = var_41411, interleave = var_42890_interleave_0, values = (var_42844_cast_fp16, var_42846_cast_fp16, var_42848_cast_fp16, var_42850_cast_fp16))[name = string("op_42890_cast_fp16")]; + bool input_217_interleave_0 = const()[name = string("input_217_interleave_0"), val = bool(false)]; + tensor input_217_cast_fp16 = concat(axis = var_41436, interleave = input_217_interleave_0, values = (var_42852_cast_fp16, var_42854_cast_fp16, var_42856_cast_fp16, var_42858_cast_fp16, var_42860_cast_fp16, var_42862_cast_fp16, var_42864_cast_fp16, var_42866_cast_fp16, var_42868_cast_fp16, var_42870_cast_fp16, var_42872_cast_fp16, var_42874_cast_fp16, var_42876_cast_fp16, var_42878_cast_fp16, var_42880_cast_fp16, var_42882_cast_fp16, var_42884_cast_fp16, var_42886_cast_fp16, var_42888_cast_fp16, var_42890_cast_fp16))[name = string("input_217_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_27_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1087043840)))]; + tensor layers_27_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090320704)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_27_self_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_27_self_attn_o_proj_weight_to_fp16, x = input_217_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_111_cast_fp16")]; + tensor out_111_axes_0 = const()[name = string("out_111_axes_0"), val = tensor([1])]; + fp16 var_42909_to_fp16 = const()[name = string("op_42909_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_42909_to_fp16, x = inputs_111_cast_fp16)[name = string("out_111_cast_fp16")]; + tensor input_219_gamma_0_to_fp16 = const()[name = string("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090323328)))]; + tensor input_219_beta_0_to_fp16 = const()[name = string("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090325952)))]; + fp16 input_219_epsilon_0_to_fp16 = const()[name = string("input_219_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = string("input_219_cast_fp16")]; + string input_221_pad_type_0 = const()[name = string("input_221_pad_type_0"), val = string("valid")]; + tensor input_221_strides_0 = const()[name = string("input_221_strides_0"), val = tensor([1, 1])]; + tensor input_221_pad_0 = const()[name = string("input_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_221_dilations_0 = const()[name = string("input_221_dilations_0"), val = tensor([1, 1])]; + int32 input_221_groups_0 = const()[name = string("input_221_groups_0"), val = int32(1)]; + tensor layers_27_fc1_weight_to_fp16 = const()[name = string("layers_27_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1090328576)))]; + tensor layers_27_fc1_bias_to_fp16 = const()[name = string("layers_27_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103435840)))]; + tensor input_221_cast_fp16 = conv(bias = layers_27_fc1_bias_to_fp16, dilations = input_221_dilations_0, groups = input_221_groups_0, pad = input_221_pad_0, pad_type = input_221_pad_type_0, strides = input_221_strides_0, weight = layers_27_fc1_weight_to_fp16, x = input_219_cast_fp16)[name = string("input_221_cast_fp16")]; + string input_223_mode_0 = const()[name = string("input_223_mode_0"), val = string("EXACT")]; + tensor input_223_cast_fp16 = gelu(mode = input_223_mode_0, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")]; + string hidden_states_59_pad_type_0 = const()[name = string("hidden_states_59_pad_type_0"), val = string("valid")]; + tensor hidden_states_59_strides_0 = const()[name = string("hidden_states_59_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_59_pad_0 = const()[name = string("hidden_states_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_59_dilations_0 = const()[name = string("hidden_states_59_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_59_groups_0 = const()[name = string("hidden_states_59_groups_0"), val = int32(1)]; + tensor layers_27_fc2_weight_to_fp16 = const()[name = string("layers_27_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1103446144)))]; + tensor layers_27_fc2_bias_to_fp16 = const()[name = string("layers_27_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116553408)))]; + tensor hidden_states_59_cast_fp16 = conv(bias = layers_27_fc2_bias_to_fp16, dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = layers_27_fc2_weight_to_fp16, x = input_223_cast_fp16)[name = string("hidden_states_59_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = hidden_states_59_cast_fp16)[name = string("inputs_113_cast_fp16")]; + int32 var_42938 = const()[name = string("op_42938"), val = int32(3)]; + int32 var_42963 = const()[name = string("op_42963"), val = int32(1)]; + tensor out_113_axes_0 = const()[name = string("out_113_axes_0"), val = tensor([1])]; + fp16 var_42980_to_fp16 = const()[name = string("op_42980_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_42980_to_fp16, x = inputs_113_cast_fp16)[name = string("out_113_cast_fp16")]; + tensor obj_113_gamma_0_to_fp16 = const()[name = string("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116556032)))]; + tensor obj_113_beta_0_to_fp16 = const()[name = string("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116558656)))]; + fp16 obj_113_epsilon_0_to_fp16 = const()[name = string("obj_113_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = string("obj_113_cast_fp16")]; + string query_57_pad_type_0 = const()[name = string("query_57_pad_type_0"), val = string("valid")]; + tensor query_57_strides_0 = const()[name = string("query_57_strides_0"), val = tensor([1, 1])]; + tensor query_57_pad_0 = const()[name = string("query_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_57_dilations_0 = const()[name = string("query_57_dilations_0"), val = tensor([1, 1])]; + int32 query_57_groups_0 = const()[name = string("query_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1116561280)))]; + tensor layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119838144)))]; + tensor query_57_cast_fp16 = conv(bias = layers_28_self_attn_q_proj_bias_to_fp16, dilations = query_57_dilations_0, groups = query_57_groups_0, pad = query_57_pad_0, pad_type = query_57_pad_type_0, strides = query_57_strides_0, weight = layers_28_self_attn_q_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("query_57_cast_fp16")]; + string key_57_pad_type_0 = const()[name = string("key_57_pad_type_0"), val = string("valid")]; + tensor key_57_strides_0 = const()[name = string("key_57_strides_0"), val = tensor([1, 1])]; + tensor key_57_pad_0 = const()[name = string("key_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_57_dilations_0 = const()[name = string("key_57_dilations_0"), val = tensor([1, 1])]; + int32 key_57_groups_0 = const()[name = string("key_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1119840768)))]; + tensor key_57_cast_fp16 = conv(dilations = key_57_dilations_0, groups = key_57_groups_0, pad = key_57_pad_0, pad_type = key_57_pad_type_0, strides = key_57_strides_0, weight = layers_28_self_attn_k_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("key_57_cast_fp16")]; + string value_57_pad_type_0 = const()[name = string("value_57_pad_type_0"), val = string("valid")]; + tensor value_57_strides_0 = const()[name = string("value_57_strides_0"), val = tensor([1, 1])]; + tensor value_57_pad_0 = const()[name = string("value_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_57_dilations_0 = const()[name = string("value_57_dilations_0"), val = tensor([1, 1])]; + int32 value_57_groups_0 = const()[name = string("value_57_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1123117632)))]; + tensor layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126394496)))]; + tensor value_57_cast_fp16 = conv(bias = layers_28_self_attn_v_proj_bias_to_fp16, dilations = value_57_dilations_0, groups = value_57_groups_0, pad = value_57_pad_0, pad_type = value_57_pad_type_0, strides = value_57_strides_0, weight = layers_28_self_attn_v_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = string("value_57_cast_fp16")]; + tensor var_43018_begin_0 = const()[name = string("op_43018_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43018_end_0 = const()[name = string("op_43018_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43018_end_mask_0 = const()[name = string("op_43018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43018_cast_fp16 = slice_by_index(begin = var_43018_begin_0, end = var_43018_end_0, end_mask = var_43018_end_mask_0, x = query_57_cast_fp16)[name = string("op_43018_cast_fp16")]; + tensor var_43022_begin_0 = const()[name = string("op_43022_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43022_end_0 = const()[name = string("op_43022_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_43022_end_mask_0 = const()[name = string("op_43022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43022_cast_fp16 = slice_by_index(begin = var_43022_begin_0, end = var_43022_end_0, end_mask = var_43022_end_mask_0, x = query_57_cast_fp16)[name = string("op_43022_cast_fp16")]; + tensor var_43026_begin_0 = const()[name = string("op_43026_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43026_end_0 = const()[name = string("op_43026_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_43026_end_mask_0 = const()[name = string("op_43026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43026_cast_fp16 = slice_by_index(begin = var_43026_begin_0, end = var_43026_end_0, end_mask = var_43026_end_mask_0, x = query_57_cast_fp16)[name = string("op_43026_cast_fp16")]; + tensor var_43030_begin_0 = const()[name = string("op_43030_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43030_end_0 = const()[name = string("op_43030_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_43030_end_mask_0 = const()[name = string("op_43030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43030_cast_fp16 = slice_by_index(begin = var_43030_begin_0, end = var_43030_end_0, end_mask = var_43030_end_mask_0, x = query_57_cast_fp16)[name = string("op_43030_cast_fp16")]; + tensor var_43034_begin_0 = const()[name = string("op_43034_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43034_end_0 = const()[name = string("op_43034_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_43034_end_mask_0 = const()[name = string("op_43034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43034_cast_fp16 = slice_by_index(begin = var_43034_begin_0, end = var_43034_end_0, end_mask = var_43034_end_mask_0, x = query_57_cast_fp16)[name = string("op_43034_cast_fp16")]; + tensor var_43038_begin_0 = const()[name = string("op_43038_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43038_end_0 = const()[name = string("op_43038_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_43038_end_mask_0 = const()[name = string("op_43038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43038_cast_fp16 = slice_by_index(begin = var_43038_begin_0, end = var_43038_end_0, end_mask = var_43038_end_mask_0, x = query_57_cast_fp16)[name = string("op_43038_cast_fp16")]; + tensor var_43042_begin_0 = const()[name = string("op_43042_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43042_end_0 = const()[name = string("op_43042_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_43042_end_mask_0 = const()[name = string("op_43042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43042_cast_fp16 = slice_by_index(begin = var_43042_begin_0, end = var_43042_end_0, end_mask = var_43042_end_mask_0, x = query_57_cast_fp16)[name = string("op_43042_cast_fp16")]; + tensor var_43046_begin_0 = const()[name = string("op_43046_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43046_end_0 = const()[name = string("op_43046_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_43046_end_mask_0 = const()[name = string("op_43046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43046_cast_fp16 = slice_by_index(begin = var_43046_begin_0, end = var_43046_end_0, end_mask = var_43046_end_mask_0, x = query_57_cast_fp16)[name = string("op_43046_cast_fp16")]; + tensor var_43050_begin_0 = const()[name = string("op_43050_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43050_end_0 = const()[name = string("op_43050_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_43050_end_mask_0 = const()[name = string("op_43050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43050_cast_fp16 = slice_by_index(begin = var_43050_begin_0, end = var_43050_end_0, end_mask = var_43050_end_mask_0, x = query_57_cast_fp16)[name = string("op_43050_cast_fp16")]; + tensor var_43054_begin_0 = const()[name = string("op_43054_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43054_end_0 = const()[name = string("op_43054_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_43054_end_mask_0 = const()[name = string("op_43054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43054_cast_fp16 = slice_by_index(begin = var_43054_begin_0, end = var_43054_end_0, end_mask = var_43054_end_mask_0, x = query_57_cast_fp16)[name = string("op_43054_cast_fp16")]; + tensor var_43058_begin_0 = const()[name = string("op_43058_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43058_end_0 = const()[name = string("op_43058_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_43058_end_mask_0 = const()[name = string("op_43058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43058_cast_fp16 = slice_by_index(begin = var_43058_begin_0, end = var_43058_end_0, end_mask = var_43058_end_mask_0, x = query_57_cast_fp16)[name = string("op_43058_cast_fp16")]; + tensor var_43062_begin_0 = const()[name = string("op_43062_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43062_end_0 = const()[name = string("op_43062_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_43062_end_mask_0 = const()[name = string("op_43062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43062_cast_fp16 = slice_by_index(begin = var_43062_begin_0, end = var_43062_end_0, end_mask = var_43062_end_mask_0, x = query_57_cast_fp16)[name = string("op_43062_cast_fp16")]; + tensor var_43066_begin_0 = const()[name = string("op_43066_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43066_end_0 = const()[name = string("op_43066_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_43066_end_mask_0 = const()[name = string("op_43066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43066_cast_fp16 = slice_by_index(begin = var_43066_begin_0, end = var_43066_end_0, end_mask = var_43066_end_mask_0, x = query_57_cast_fp16)[name = string("op_43066_cast_fp16")]; + tensor var_43070_begin_0 = const()[name = string("op_43070_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43070_end_0 = const()[name = string("op_43070_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_43070_end_mask_0 = const()[name = string("op_43070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43070_cast_fp16 = slice_by_index(begin = var_43070_begin_0, end = var_43070_end_0, end_mask = var_43070_end_mask_0, x = query_57_cast_fp16)[name = string("op_43070_cast_fp16")]; + tensor var_43074_begin_0 = const()[name = string("op_43074_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43074_end_0 = const()[name = string("op_43074_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_43074_end_mask_0 = const()[name = string("op_43074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43074_cast_fp16 = slice_by_index(begin = var_43074_begin_0, end = var_43074_end_0, end_mask = var_43074_end_mask_0, x = query_57_cast_fp16)[name = string("op_43074_cast_fp16")]; + tensor var_43078_begin_0 = const()[name = string("op_43078_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43078_end_0 = const()[name = string("op_43078_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_43078_end_mask_0 = const()[name = string("op_43078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43078_cast_fp16 = slice_by_index(begin = var_43078_begin_0, end = var_43078_end_0, end_mask = var_43078_end_mask_0, x = query_57_cast_fp16)[name = string("op_43078_cast_fp16")]; + tensor var_43082_begin_0 = const()[name = string("op_43082_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43082_end_0 = const()[name = string("op_43082_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_43082_end_mask_0 = const()[name = string("op_43082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43082_cast_fp16 = slice_by_index(begin = var_43082_begin_0, end = var_43082_end_0, end_mask = var_43082_end_mask_0, x = query_57_cast_fp16)[name = string("op_43082_cast_fp16")]; + tensor var_43086_begin_0 = const()[name = string("op_43086_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43086_end_0 = const()[name = string("op_43086_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_43086_end_mask_0 = const()[name = string("op_43086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43086_cast_fp16 = slice_by_index(begin = var_43086_begin_0, end = var_43086_end_0, end_mask = var_43086_end_mask_0, x = query_57_cast_fp16)[name = string("op_43086_cast_fp16")]; + tensor var_43090_begin_0 = const()[name = string("op_43090_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43090_end_0 = const()[name = string("op_43090_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_43090_end_mask_0 = const()[name = string("op_43090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43090_cast_fp16 = slice_by_index(begin = var_43090_begin_0, end = var_43090_end_0, end_mask = var_43090_end_mask_0, x = query_57_cast_fp16)[name = string("op_43090_cast_fp16")]; + tensor var_43094_begin_0 = const()[name = string("op_43094_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43094_end_0 = const()[name = string("op_43094_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_43094_end_mask_0 = const()[name = string("op_43094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43094_cast_fp16 = slice_by_index(begin = var_43094_begin_0, end = var_43094_end_0, end_mask = var_43094_end_mask_0, x = query_57_cast_fp16)[name = string("op_43094_cast_fp16")]; + tensor var_43103_begin_0 = const()[name = string("op_43103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43103_end_0 = const()[name = string("op_43103_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43103_end_mask_0 = const()[name = string("op_43103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43103_cast_fp16 = slice_by_index(begin = var_43103_begin_0, end = var_43103_end_0, end_mask = var_43103_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43103_cast_fp16")]; + tensor var_43110_begin_0 = const()[name = string("op_43110_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43110_end_0 = const()[name = string("op_43110_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43110_end_mask_0 = const()[name = string("op_43110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43110_cast_fp16 = slice_by_index(begin = var_43110_begin_0, end = var_43110_end_0, end_mask = var_43110_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43110_cast_fp16")]; + tensor var_43117_begin_0 = const()[name = string("op_43117_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43117_end_0 = const()[name = string("op_43117_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43117_end_mask_0 = const()[name = string("op_43117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43117_cast_fp16 = slice_by_index(begin = var_43117_begin_0, end = var_43117_end_0, end_mask = var_43117_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43117_cast_fp16")]; + tensor var_43124_begin_0 = const()[name = string("op_43124_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43124_end_0 = const()[name = string("op_43124_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43124_end_mask_0 = const()[name = string("op_43124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43124_cast_fp16 = slice_by_index(begin = var_43124_begin_0, end = var_43124_end_0, end_mask = var_43124_end_mask_0, x = var_43018_cast_fp16)[name = string("op_43124_cast_fp16")]; + tensor var_43131_begin_0 = const()[name = string("op_43131_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43131_end_0 = const()[name = string("op_43131_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43131_end_mask_0 = const()[name = string("op_43131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43131_cast_fp16 = slice_by_index(begin = var_43131_begin_0, end = var_43131_end_0, end_mask = var_43131_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43131_cast_fp16")]; + tensor var_43138_begin_0 = const()[name = string("op_43138_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43138_end_0 = const()[name = string("op_43138_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43138_end_mask_0 = const()[name = string("op_43138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43138_cast_fp16 = slice_by_index(begin = var_43138_begin_0, end = var_43138_end_0, end_mask = var_43138_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43138_cast_fp16")]; + tensor var_43145_begin_0 = const()[name = string("op_43145_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43145_end_0 = const()[name = string("op_43145_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43145_end_mask_0 = const()[name = string("op_43145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43145_cast_fp16 = slice_by_index(begin = var_43145_begin_0, end = var_43145_end_0, end_mask = var_43145_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43145_cast_fp16")]; + tensor var_43152_begin_0 = const()[name = string("op_43152_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43152_end_0 = const()[name = string("op_43152_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43152_end_mask_0 = const()[name = string("op_43152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43152_cast_fp16 = slice_by_index(begin = var_43152_begin_0, end = var_43152_end_0, end_mask = var_43152_end_mask_0, x = var_43022_cast_fp16)[name = string("op_43152_cast_fp16")]; + tensor var_43159_begin_0 = const()[name = string("op_43159_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43159_end_0 = const()[name = string("op_43159_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43159_end_mask_0 = const()[name = string("op_43159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43159_cast_fp16 = slice_by_index(begin = var_43159_begin_0, end = var_43159_end_0, end_mask = var_43159_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43159_cast_fp16")]; + tensor var_43166_begin_0 = const()[name = string("op_43166_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43166_end_0 = const()[name = string("op_43166_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43166_end_mask_0 = const()[name = string("op_43166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43166_cast_fp16 = slice_by_index(begin = var_43166_begin_0, end = var_43166_end_0, end_mask = var_43166_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43166_cast_fp16")]; + tensor var_43173_begin_0 = const()[name = string("op_43173_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43173_end_0 = const()[name = string("op_43173_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43173_end_mask_0 = const()[name = string("op_43173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43173_cast_fp16 = slice_by_index(begin = var_43173_begin_0, end = var_43173_end_0, end_mask = var_43173_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43173_cast_fp16")]; + tensor var_43180_begin_0 = const()[name = string("op_43180_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43180_end_0 = const()[name = string("op_43180_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43180_end_mask_0 = const()[name = string("op_43180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43180_cast_fp16 = slice_by_index(begin = var_43180_begin_0, end = var_43180_end_0, end_mask = var_43180_end_mask_0, x = var_43026_cast_fp16)[name = string("op_43180_cast_fp16")]; + tensor var_43187_begin_0 = const()[name = string("op_43187_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43187_end_0 = const()[name = string("op_43187_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43187_end_mask_0 = const()[name = string("op_43187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43187_cast_fp16 = slice_by_index(begin = var_43187_begin_0, end = var_43187_end_0, end_mask = var_43187_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43187_cast_fp16")]; + tensor var_43194_begin_0 = const()[name = string("op_43194_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43194_end_0 = const()[name = string("op_43194_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43194_end_mask_0 = const()[name = string("op_43194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43194_cast_fp16 = slice_by_index(begin = var_43194_begin_0, end = var_43194_end_0, end_mask = var_43194_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43194_cast_fp16")]; + tensor var_43201_begin_0 = const()[name = string("op_43201_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43201_end_0 = const()[name = string("op_43201_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43201_end_mask_0 = const()[name = string("op_43201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43201_cast_fp16 = slice_by_index(begin = var_43201_begin_0, end = var_43201_end_0, end_mask = var_43201_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43201_cast_fp16")]; + tensor var_43208_begin_0 = const()[name = string("op_43208_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43208_end_0 = const()[name = string("op_43208_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43208_end_mask_0 = const()[name = string("op_43208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43208_cast_fp16 = slice_by_index(begin = var_43208_begin_0, end = var_43208_end_0, end_mask = var_43208_end_mask_0, x = var_43030_cast_fp16)[name = string("op_43208_cast_fp16")]; + tensor var_43215_begin_0 = const()[name = string("op_43215_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43215_end_0 = const()[name = string("op_43215_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43215_end_mask_0 = const()[name = string("op_43215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43215_cast_fp16 = slice_by_index(begin = var_43215_begin_0, end = var_43215_end_0, end_mask = var_43215_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43215_cast_fp16")]; + tensor var_43222_begin_0 = const()[name = string("op_43222_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43222_end_0 = const()[name = string("op_43222_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43222_end_mask_0 = const()[name = string("op_43222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43222_cast_fp16 = slice_by_index(begin = var_43222_begin_0, end = var_43222_end_0, end_mask = var_43222_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43222_cast_fp16")]; + tensor var_43229_begin_0 = const()[name = string("op_43229_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43229_end_0 = const()[name = string("op_43229_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43229_end_mask_0 = const()[name = string("op_43229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43229_cast_fp16 = slice_by_index(begin = var_43229_begin_0, end = var_43229_end_0, end_mask = var_43229_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43229_cast_fp16")]; + tensor var_43236_begin_0 = const()[name = string("op_43236_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43236_end_0 = const()[name = string("op_43236_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43236_end_mask_0 = const()[name = string("op_43236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43236_cast_fp16 = slice_by_index(begin = var_43236_begin_0, end = var_43236_end_0, end_mask = var_43236_end_mask_0, x = var_43034_cast_fp16)[name = string("op_43236_cast_fp16")]; + tensor var_43243_begin_0 = const()[name = string("op_43243_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43243_end_0 = const()[name = string("op_43243_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43243_end_mask_0 = const()[name = string("op_43243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43243_cast_fp16 = slice_by_index(begin = var_43243_begin_0, end = var_43243_end_0, end_mask = var_43243_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43243_cast_fp16")]; + tensor var_43250_begin_0 = const()[name = string("op_43250_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43250_end_0 = const()[name = string("op_43250_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43250_end_mask_0 = const()[name = string("op_43250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43250_cast_fp16 = slice_by_index(begin = var_43250_begin_0, end = var_43250_end_0, end_mask = var_43250_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43250_cast_fp16")]; + tensor var_43257_begin_0 = const()[name = string("op_43257_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43257_end_0 = const()[name = string("op_43257_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43257_end_mask_0 = const()[name = string("op_43257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43257_cast_fp16 = slice_by_index(begin = var_43257_begin_0, end = var_43257_end_0, end_mask = var_43257_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43257_cast_fp16")]; + tensor var_43264_begin_0 = const()[name = string("op_43264_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43264_end_0 = const()[name = string("op_43264_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43264_end_mask_0 = const()[name = string("op_43264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43264_cast_fp16 = slice_by_index(begin = var_43264_begin_0, end = var_43264_end_0, end_mask = var_43264_end_mask_0, x = var_43038_cast_fp16)[name = string("op_43264_cast_fp16")]; + tensor var_43271_begin_0 = const()[name = string("op_43271_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43271_end_0 = const()[name = string("op_43271_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43271_end_mask_0 = const()[name = string("op_43271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43271_cast_fp16 = slice_by_index(begin = var_43271_begin_0, end = var_43271_end_0, end_mask = var_43271_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43271_cast_fp16")]; + tensor var_43278_begin_0 = const()[name = string("op_43278_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43278_end_0 = const()[name = string("op_43278_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43278_end_mask_0 = const()[name = string("op_43278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43278_cast_fp16 = slice_by_index(begin = var_43278_begin_0, end = var_43278_end_0, end_mask = var_43278_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43278_cast_fp16")]; + tensor var_43285_begin_0 = const()[name = string("op_43285_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43285_end_0 = const()[name = string("op_43285_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43285_end_mask_0 = const()[name = string("op_43285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43285_cast_fp16 = slice_by_index(begin = var_43285_begin_0, end = var_43285_end_0, end_mask = var_43285_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43285_cast_fp16")]; + tensor var_43292_begin_0 = const()[name = string("op_43292_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43292_end_0 = const()[name = string("op_43292_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43292_end_mask_0 = const()[name = string("op_43292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43292_cast_fp16 = slice_by_index(begin = var_43292_begin_0, end = var_43292_end_0, end_mask = var_43292_end_mask_0, x = var_43042_cast_fp16)[name = string("op_43292_cast_fp16")]; + tensor var_43299_begin_0 = const()[name = string("op_43299_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43299_end_0 = const()[name = string("op_43299_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43299_end_mask_0 = const()[name = string("op_43299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43299_cast_fp16 = slice_by_index(begin = var_43299_begin_0, end = var_43299_end_0, end_mask = var_43299_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43299_cast_fp16")]; + tensor var_43306_begin_0 = const()[name = string("op_43306_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43306_end_0 = const()[name = string("op_43306_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43306_end_mask_0 = const()[name = string("op_43306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43306_cast_fp16 = slice_by_index(begin = var_43306_begin_0, end = var_43306_end_0, end_mask = var_43306_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43306_cast_fp16")]; + tensor var_43313_begin_0 = const()[name = string("op_43313_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43313_end_0 = const()[name = string("op_43313_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43313_end_mask_0 = const()[name = string("op_43313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43313_cast_fp16 = slice_by_index(begin = var_43313_begin_0, end = var_43313_end_0, end_mask = var_43313_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43313_cast_fp16")]; + tensor var_43320_begin_0 = const()[name = string("op_43320_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43320_end_0 = const()[name = string("op_43320_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43320_end_mask_0 = const()[name = string("op_43320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43320_cast_fp16 = slice_by_index(begin = var_43320_begin_0, end = var_43320_end_0, end_mask = var_43320_end_mask_0, x = var_43046_cast_fp16)[name = string("op_43320_cast_fp16")]; + tensor var_43327_begin_0 = const()[name = string("op_43327_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43327_end_0 = const()[name = string("op_43327_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43327_end_mask_0 = const()[name = string("op_43327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43327_cast_fp16 = slice_by_index(begin = var_43327_begin_0, end = var_43327_end_0, end_mask = var_43327_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43327_cast_fp16")]; + tensor var_43334_begin_0 = const()[name = string("op_43334_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43334_end_0 = const()[name = string("op_43334_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43334_end_mask_0 = const()[name = string("op_43334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43334_cast_fp16 = slice_by_index(begin = var_43334_begin_0, end = var_43334_end_0, end_mask = var_43334_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43334_cast_fp16")]; + tensor var_43341_begin_0 = const()[name = string("op_43341_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43341_end_0 = const()[name = string("op_43341_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43341_end_mask_0 = const()[name = string("op_43341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43341_cast_fp16 = slice_by_index(begin = var_43341_begin_0, end = var_43341_end_0, end_mask = var_43341_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43341_cast_fp16")]; + tensor var_43348_begin_0 = const()[name = string("op_43348_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43348_end_0 = const()[name = string("op_43348_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43348_end_mask_0 = const()[name = string("op_43348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43348_cast_fp16 = slice_by_index(begin = var_43348_begin_0, end = var_43348_end_0, end_mask = var_43348_end_mask_0, x = var_43050_cast_fp16)[name = string("op_43348_cast_fp16")]; + tensor var_43355_begin_0 = const()[name = string("op_43355_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43355_end_0 = const()[name = string("op_43355_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43355_end_mask_0 = const()[name = string("op_43355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43355_cast_fp16 = slice_by_index(begin = var_43355_begin_0, end = var_43355_end_0, end_mask = var_43355_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43355_cast_fp16")]; + tensor var_43362_begin_0 = const()[name = string("op_43362_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43362_end_0 = const()[name = string("op_43362_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43362_end_mask_0 = const()[name = string("op_43362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43362_cast_fp16 = slice_by_index(begin = var_43362_begin_0, end = var_43362_end_0, end_mask = var_43362_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43362_cast_fp16")]; + tensor var_43369_begin_0 = const()[name = string("op_43369_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43369_end_0 = const()[name = string("op_43369_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43369_end_mask_0 = const()[name = string("op_43369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43369_cast_fp16 = slice_by_index(begin = var_43369_begin_0, end = var_43369_end_0, end_mask = var_43369_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43369_cast_fp16")]; + tensor var_43376_begin_0 = const()[name = string("op_43376_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43376_end_0 = const()[name = string("op_43376_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43376_end_mask_0 = const()[name = string("op_43376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43376_cast_fp16 = slice_by_index(begin = var_43376_begin_0, end = var_43376_end_0, end_mask = var_43376_end_mask_0, x = var_43054_cast_fp16)[name = string("op_43376_cast_fp16")]; + tensor var_43383_begin_0 = const()[name = string("op_43383_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43383_end_0 = const()[name = string("op_43383_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43383_end_mask_0 = const()[name = string("op_43383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43383_cast_fp16 = slice_by_index(begin = var_43383_begin_0, end = var_43383_end_0, end_mask = var_43383_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43383_cast_fp16")]; + tensor var_43390_begin_0 = const()[name = string("op_43390_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43390_end_0 = const()[name = string("op_43390_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43390_end_mask_0 = const()[name = string("op_43390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43390_cast_fp16 = slice_by_index(begin = var_43390_begin_0, end = var_43390_end_0, end_mask = var_43390_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43390_cast_fp16")]; + tensor var_43397_begin_0 = const()[name = string("op_43397_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43397_end_0 = const()[name = string("op_43397_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43397_end_mask_0 = const()[name = string("op_43397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43397_cast_fp16 = slice_by_index(begin = var_43397_begin_0, end = var_43397_end_0, end_mask = var_43397_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43397_cast_fp16")]; + tensor var_43404_begin_0 = const()[name = string("op_43404_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43404_end_0 = const()[name = string("op_43404_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43404_end_mask_0 = const()[name = string("op_43404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43404_cast_fp16 = slice_by_index(begin = var_43404_begin_0, end = var_43404_end_0, end_mask = var_43404_end_mask_0, x = var_43058_cast_fp16)[name = string("op_43404_cast_fp16")]; + tensor var_43411_begin_0 = const()[name = string("op_43411_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43411_end_0 = const()[name = string("op_43411_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43411_end_mask_0 = const()[name = string("op_43411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43411_cast_fp16 = slice_by_index(begin = var_43411_begin_0, end = var_43411_end_0, end_mask = var_43411_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43411_cast_fp16")]; + tensor var_43418_begin_0 = const()[name = string("op_43418_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43418_end_0 = const()[name = string("op_43418_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43418_end_mask_0 = const()[name = string("op_43418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43418_cast_fp16 = slice_by_index(begin = var_43418_begin_0, end = var_43418_end_0, end_mask = var_43418_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43418_cast_fp16")]; + tensor var_43425_begin_0 = const()[name = string("op_43425_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43425_end_0 = const()[name = string("op_43425_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43425_end_mask_0 = const()[name = string("op_43425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43425_cast_fp16 = slice_by_index(begin = var_43425_begin_0, end = var_43425_end_0, end_mask = var_43425_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43425_cast_fp16")]; + tensor var_43432_begin_0 = const()[name = string("op_43432_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43432_end_0 = const()[name = string("op_43432_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43432_end_mask_0 = const()[name = string("op_43432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43432_cast_fp16 = slice_by_index(begin = var_43432_begin_0, end = var_43432_end_0, end_mask = var_43432_end_mask_0, x = var_43062_cast_fp16)[name = string("op_43432_cast_fp16")]; + tensor var_43439_begin_0 = const()[name = string("op_43439_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43439_end_0 = const()[name = string("op_43439_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43439_end_mask_0 = const()[name = string("op_43439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43439_cast_fp16 = slice_by_index(begin = var_43439_begin_0, end = var_43439_end_0, end_mask = var_43439_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43439_cast_fp16")]; + tensor var_43446_begin_0 = const()[name = string("op_43446_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43446_end_0 = const()[name = string("op_43446_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43446_end_mask_0 = const()[name = string("op_43446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43446_cast_fp16 = slice_by_index(begin = var_43446_begin_0, end = var_43446_end_0, end_mask = var_43446_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43446_cast_fp16")]; + tensor var_43453_begin_0 = const()[name = string("op_43453_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43453_end_0 = const()[name = string("op_43453_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43453_end_mask_0 = const()[name = string("op_43453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43453_cast_fp16 = slice_by_index(begin = var_43453_begin_0, end = var_43453_end_0, end_mask = var_43453_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43453_cast_fp16")]; + tensor var_43460_begin_0 = const()[name = string("op_43460_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43460_end_0 = const()[name = string("op_43460_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43460_end_mask_0 = const()[name = string("op_43460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43460_cast_fp16 = slice_by_index(begin = var_43460_begin_0, end = var_43460_end_0, end_mask = var_43460_end_mask_0, x = var_43066_cast_fp16)[name = string("op_43460_cast_fp16")]; + tensor var_43467_begin_0 = const()[name = string("op_43467_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43467_end_0 = const()[name = string("op_43467_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43467_end_mask_0 = const()[name = string("op_43467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43467_cast_fp16 = slice_by_index(begin = var_43467_begin_0, end = var_43467_end_0, end_mask = var_43467_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43467_cast_fp16")]; + tensor var_43474_begin_0 = const()[name = string("op_43474_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43474_end_0 = const()[name = string("op_43474_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43474_end_mask_0 = const()[name = string("op_43474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43474_cast_fp16 = slice_by_index(begin = var_43474_begin_0, end = var_43474_end_0, end_mask = var_43474_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43474_cast_fp16")]; + tensor var_43481_begin_0 = const()[name = string("op_43481_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43481_end_0 = const()[name = string("op_43481_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43481_end_mask_0 = const()[name = string("op_43481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43481_cast_fp16 = slice_by_index(begin = var_43481_begin_0, end = var_43481_end_0, end_mask = var_43481_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43481_cast_fp16")]; + tensor var_43488_begin_0 = const()[name = string("op_43488_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43488_end_0 = const()[name = string("op_43488_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43488_end_mask_0 = const()[name = string("op_43488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43488_cast_fp16 = slice_by_index(begin = var_43488_begin_0, end = var_43488_end_0, end_mask = var_43488_end_mask_0, x = var_43070_cast_fp16)[name = string("op_43488_cast_fp16")]; + tensor var_43495_begin_0 = const()[name = string("op_43495_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43495_end_0 = const()[name = string("op_43495_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43495_end_mask_0 = const()[name = string("op_43495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43495_cast_fp16 = slice_by_index(begin = var_43495_begin_0, end = var_43495_end_0, end_mask = var_43495_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43495_cast_fp16")]; + tensor var_43502_begin_0 = const()[name = string("op_43502_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43502_end_0 = const()[name = string("op_43502_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43502_end_mask_0 = const()[name = string("op_43502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43502_cast_fp16 = slice_by_index(begin = var_43502_begin_0, end = var_43502_end_0, end_mask = var_43502_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43502_cast_fp16")]; + tensor var_43509_begin_0 = const()[name = string("op_43509_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43509_end_0 = const()[name = string("op_43509_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43509_end_mask_0 = const()[name = string("op_43509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43509_cast_fp16 = slice_by_index(begin = var_43509_begin_0, end = var_43509_end_0, end_mask = var_43509_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43509_cast_fp16")]; + tensor var_43516_begin_0 = const()[name = string("op_43516_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43516_end_0 = const()[name = string("op_43516_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43516_end_mask_0 = const()[name = string("op_43516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43516_cast_fp16 = slice_by_index(begin = var_43516_begin_0, end = var_43516_end_0, end_mask = var_43516_end_mask_0, x = var_43074_cast_fp16)[name = string("op_43516_cast_fp16")]; + tensor var_43523_begin_0 = const()[name = string("op_43523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43523_end_0 = const()[name = string("op_43523_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43523_end_mask_0 = const()[name = string("op_43523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43523_cast_fp16 = slice_by_index(begin = var_43523_begin_0, end = var_43523_end_0, end_mask = var_43523_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43523_cast_fp16")]; + tensor var_43530_begin_0 = const()[name = string("op_43530_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43530_end_0 = const()[name = string("op_43530_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43530_end_mask_0 = const()[name = string("op_43530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43530_cast_fp16 = slice_by_index(begin = var_43530_begin_0, end = var_43530_end_0, end_mask = var_43530_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43530_cast_fp16")]; + tensor var_43537_begin_0 = const()[name = string("op_43537_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43537_end_0 = const()[name = string("op_43537_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43537_end_mask_0 = const()[name = string("op_43537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43537_cast_fp16 = slice_by_index(begin = var_43537_begin_0, end = var_43537_end_0, end_mask = var_43537_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43537_cast_fp16")]; + tensor var_43544_begin_0 = const()[name = string("op_43544_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43544_end_0 = const()[name = string("op_43544_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43544_end_mask_0 = const()[name = string("op_43544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43544_cast_fp16 = slice_by_index(begin = var_43544_begin_0, end = var_43544_end_0, end_mask = var_43544_end_mask_0, x = var_43078_cast_fp16)[name = string("op_43544_cast_fp16")]; + tensor var_43551_begin_0 = const()[name = string("op_43551_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43551_end_0 = const()[name = string("op_43551_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43551_end_mask_0 = const()[name = string("op_43551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43551_cast_fp16 = slice_by_index(begin = var_43551_begin_0, end = var_43551_end_0, end_mask = var_43551_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43551_cast_fp16")]; + tensor var_43558_begin_0 = const()[name = string("op_43558_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43558_end_0 = const()[name = string("op_43558_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43558_end_mask_0 = const()[name = string("op_43558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43558_cast_fp16 = slice_by_index(begin = var_43558_begin_0, end = var_43558_end_0, end_mask = var_43558_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43558_cast_fp16")]; + tensor var_43565_begin_0 = const()[name = string("op_43565_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43565_end_0 = const()[name = string("op_43565_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43565_end_mask_0 = const()[name = string("op_43565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43565_cast_fp16 = slice_by_index(begin = var_43565_begin_0, end = var_43565_end_0, end_mask = var_43565_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43565_cast_fp16")]; + tensor var_43572_begin_0 = const()[name = string("op_43572_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43572_end_0 = const()[name = string("op_43572_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43572_end_mask_0 = const()[name = string("op_43572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43572_cast_fp16 = slice_by_index(begin = var_43572_begin_0, end = var_43572_end_0, end_mask = var_43572_end_mask_0, x = var_43082_cast_fp16)[name = string("op_43572_cast_fp16")]; + tensor var_43579_begin_0 = const()[name = string("op_43579_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43579_end_0 = const()[name = string("op_43579_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43579_end_mask_0 = const()[name = string("op_43579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43579_cast_fp16 = slice_by_index(begin = var_43579_begin_0, end = var_43579_end_0, end_mask = var_43579_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43579_cast_fp16")]; + tensor var_43586_begin_0 = const()[name = string("op_43586_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43586_end_0 = const()[name = string("op_43586_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43586_end_mask_0 = const()[name = string("op_43586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43586_cast_fp16 = slice_by_index(begin = var_43586_begin_0, end = var_43586_end_0, end_mask = var_43586_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43586_cast_fp16")]; + tensor var_43593_begin_0 = const()[name = string("op_43593_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43593_end_0 = const()[name = string("op_43593_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43593_end_mask_0 = const()[name = string("op_43593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43593_cast_fp16 = slice_by_index(begin = var_43593_begin_0, end = var_43593_end_0, end_mask = var_43593_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43593_cast_fp16")]; + tensor var_43600_begin_0 = const()[name = string("op_43600_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43600_end_0 = const()[name = string("op_43600_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43600_end_mask_0 = const()[name = string("op_43600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43600_cast_fp16 = slice_by_index(begin = var_43600_begin_0, end = var_43600_end_0, end_mask = var_43600_end_mask_0, x = var_43086_cast_fp16)[name = string("op_43600_cast_fp16")]; + tensor var_43607_begin_0 = const()[name = string("op_43607_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43607_end_0 = const()[name = string("op_43607_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43607_end_mask_0 = const()[name = string("op_43607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43607_cast_fp16 = slice_by_index(begin = var_43607_begin_0, end = var_43607_end_0, end_mask = var_43607_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43607_cast_fp16")]; + tensor var_43614_begin_0 = const()[name = string("op_43614_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43614_end_0 = const()[name = string("op_43614_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43614_end_mask_0 = const()[name = string("op_43614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43614_cast_fp16 = slice_by_index(begin = var_43614_begin_0, end = var_43614_end_0, end_mask = var_43614_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43614_cast_fp16")]; + tensor var_43621_begin_0 = const()[name = string("op_43621_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43621_end_0 = const()[name = string("op_43621_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43621_end_mask_0 = const()[name = string("op_43621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43621_cast_fp16 = slice_by_index(begin = var_43621_begin_0, end = var_43621_end_0, end_mask = var_43621_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43621_cast_fp16")]; + tensor var_43628_begin_0 = const()[name = string("op_43628_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43628_end_0 = const()[name = string("op_43628_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43628_end_mask_0 = const()[name = string("op_43628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43628_cast_fp16 = slice_by_index(begin = var_43628_begin_0, end = var_43628_end_0, end_mask = var_43628_end_mask_0, x = var_43090_cast_fp16)[name = string("op_43628_cast_fp16")]; + tensor var_43635_begin_0 = const()[name = string("op_43635_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43635_end_0 = const()[name = string("op_43635_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43635_end_mask_0 = const()[name = string("op_43635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43635_cast_fp16 = slice_by_index(begin = var_43635_begin_0, end = var_43635_end_0, end_mask = var_43635_end_mask_0, x = var_43094_cast_fp16)[name = string("op_43635_cast_fp16")]; + tensor var_43642_begin_0 = const()[name = string("op_43642_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43642_end_0 = const()[name = string("op_43642_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43642_end_mask_0 = const()[name = string("op_43642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43642_cast_fp16 = slice_by_index(begin = var_43642_begin_0, end = var_43642_end_0, end_mask = var_43642_end_mask_0, x = var_43094_cast_fp16)[name = string("op_43642_cast_fp16")]; + tensor var_43649_begin_0 = const()[name = string("op_43649_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43649_end_0 = const()[name = string("op_43649_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43649_end_mask_0 = const()[name = string("op_43649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43649_cast_fp16 = slice_by_index(begin = var_43649_begin_0, end = var_43649_end_0, end_mask = var_43649_end_mask_0, x = var_43094_cast_fp16)[name = string("op_43649_cast_fp16")]; + tensor var_43656_begin_0 = const()[name = string("op_43656_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43656_end_0 = const()[name = string("op_43656_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43656_end_mask_0 = const()[name = string("op_43656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43656_cast_fp16 = slice_by_index(begin = var_43656_begin_0, end = var_43656_end_0, end_mask = var_43656_end_mask_0, x = var_43094_cast_fp16)[name = string("op_43656_cast_fp16")]; + tensor k_57_perm_0 = const()[name = string("k_57_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_43661_begin_0 = const()[name = string("op_43661_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43661_end_0 = const()[name = string("op_43661_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_43661_end_mask_0 = const()[name = string("op_43661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_57_cast_fp16 = transpose(perm = k_57_perm_0, x = key_57_cast_fp16)[name = string("transpose_3")]; + tensor var_43661_cast_fp16 = slice_by_index(begin = var_43661_begin_0, end = var_43661_end_0, end_mask = var_43661_end_mask_0, x = k_57_cast_fp16)[name = string("op_43661_cast_fp16")]; + tensor var_43665_begin_0 = const()[name = string("op_43665_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_43665_end_0 = const()[name = string("op_43665_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_43665_end_mask_0 = const()[name = string("op_43665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43665_cast_fp16 = slice_by_index(begin = var_43665_begin_0, end = var_43665_end_0, end_mask = var_43665_end_mask_0, x = k_57_cast_fp16)[name = string("op_43665_cast_fp16")]; + tensor var_43669_begin_0 = const()[name = string("op_43669_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_43669_end_0 = const()[name = string("op_43669_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_43669_end_mask_0 = const()[name = string("op_43669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43669_cast_fp16 = slice_by_index(begin = var_43669_begin_0, end = var_43669_end_0, end_mask = var_43669_end_mask_0, x = k_57_cast_fp16)[name = string("op_43669_cast_fp16")]; + tensor var_43673_begin_0 = const()[name = string("op_43673_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_43673_end_0 = const()[name = string("op_43673_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_43673_end_mask_0 = const()[name = string("op_43673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43673_cast_fp16 = slice_by_index(begin = var_43673_begin_0, end = var_43673_end_0, end_mask = var_43673_end_mask_0, x = k_57_cast_fp16)[name = string("op_43673_cast_fp16")]; + tensor var_43677_begin_0 = const()[name = string("op_43677_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_43677_end_0 = const()[name = string("op_43677_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_43677_end_mask_0 = const()[name = string("op_43677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43677_cast_fp16 = slice_by_index(begin = var_43677_begin_0, end = var_43677_end_0, end_mask = var_43677_end_mask_0, x = k_57_cast_fp16)[name = string("op_43677_cast_fp16")]; + tensor var_43681_begin_0 = const()[name = string("op_43681_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_43681_end_0 = const()[name = string("op_43681_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_43681_end_mask_0 = const()[name = string("op_43681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43681_cast_fp16 = slice_by_index(begin = var_43681_begin_0, end = var_43681_end_0, end_mask = var_43681_end_mask_0, x = k_57_cast_fp16)[name = string("op_43681_cast_fp16")]; + tensor var_43685_begin_0 = const()[name = string("op_43685_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_43685_end_0 = const()[name = string("op_43685_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_43685_end_mask_0 = const()[name = string("op_43685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43685_cast_fp16 = slice_by_index(begin = var_43685_begin_0, end = var_43685_end_0, end_mask = var_43685_end_mask_0, x = k_57_cast_fp16)[name = string("op_43685_cast_fp16")]; + tensor var_43689_begin_0 = const()[name = string("op_43689_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_43689_end_0 = const()[name = string("op_43689_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_43689_end_mask_0 = const()[name = string("op_43689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43689_cast_fp16 = slice_by_index(begin = var_43689_begin_0, end = var_43689_end_0, end_mask = var_43689_end_mask_0, x = k_57_cast_fp16)[name = string("op_43689_cast_fp16")]; + tensor var_43693_begin_0 = const()[name = string("op_43693_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_43693_end_0 = const()[name = string("op_43693_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_43693_end_mask_0 = const()[name = string("op_43693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43693_cast_fp16 = slice_by_index(begin = var_43693_begin_0, end = var_43693_end_0, end_mask = var_43693_end_mask_0, x = k_57_cast_fp16)[name = string("op_43693_cast_fp16")]; + tensor var_43697_begin_0 = const()[name = string("op_43697_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_43697_end_0 = const()[name = string("op_43697_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_43697_end_mask_0 = const()[name = string("op_43697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43697_cast_fp16 = slice_by_index(begin = var_43697_begin_0, end = var_43697_end_0, end_mask = var_43697_end_mask_0, x = k_57_cast_fp16)[name = string("op_43697_cast_fp16")]; + tensor var_43701_begin_0 = const()[name = string("op_43701_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_43701_end_0 = const()[name = string("op_43701_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_43701_end_mask_0 = const()[name = string("op_43701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43701_cast_fp16 = slice_by_index(begin = var_43701_begin_0, end = var_43701_end_0, end_mask = var_43701_end_mask_0, x = k_57_cast_fp16)[name = string("op_43701_cast_fp16")]; + tensor var_43705_begin_0 = const()[name = string("op_43705_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_43705_end_0 = const()[name = string("op_43705_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_43705_end_mask_0 = const()[name = string("op_43705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43705_cast_fp16 = slice_by_index(begin = var_43705_begin_0, end = var_43705_end_0, end_mask = var_43705_end_mask_0, x = k_57_cast_fp16)[name = string("op_43705_cast_fp16")]; + tensor var_43709_begin_0 = const()[name = string("op_43709_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_43709_end_0 = const()[name = string("op_43709_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_43709_end_mask_0 = const()[name = string("op_43709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43709_cast_fp16 = slice_by_index(begin = var_43709_begin_0, end = var_43709_end_0, end_mask = var_43709_end_mask_0, x = k_57_cast_fp16)[name = string("op_43709_cast_fp16")]; + tensor var_43713_begin_0 = const()[name = string("op_43713_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_43713_end_0 = const()[name = string("op_43713_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_43713_end_mask_0 = const()[name = string("op_43713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43713_cast_fp16 = slice_by_index(begin = var_43713_begin_0, end = var_43713_end_0, end_mask = var_43713_end_mask_0, x = k_57_cast_fp16)[name = string("op_43713_cast_fp16")]; + tensor var_43717_begin_0 = const()[name = string("op_43717_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_43717_end_0 = const()[name = string("op_43717_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_43717_end_mask_0 = const()[name = string("op_43717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43717_cast_fp16 = slice_by_index(begin = var_43717_begin_0, end = var_43717_end_0, end_mask = var_43717_end_mask_0, x = k_57_cast_fp16)[name = string("op_43717_cast_fp16")]; + tensor var_43721_begin_0 = const()[name = string("op_43721_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_43721_end_0 = const()[name = string("op_43721_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_43721_end_mask_0 = const()[name = string("op_43721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43721_cast_fp16 = slice_by_index(begin = var_43721_begin_0, end = var_43721_end_0, end_mask = var_43721_end_mask_0, x = k_57_cast_fp16)[name = string("op_43721_cast_fp16")]; + tensor var_43725_begin_0 = const()[name = string("op_43725_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_43725_end_0 = const()[name = string("op_43725_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_43725_end_mask_0 = const()[name = string("op_43725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43725_cast_fp16 = slice_by_index(begin = var_43725_begin_0, end = var_43725_end_0, end_mask = var_43725_end_mask_0, x = k_57_cast_fp16)[name = string("op_43725_cast_fp16")]; + tensor var_43729_begin_0 = const()[name = string("op_43729_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_43729_end_0 = const()[name = string("op_43729_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_43729_end_mask_0 = const()[name = string("op_43729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43729_cast_fp16 = slice_by_index(begin = var_43729_begin_0, end = var_43729_end_0, end_mask = var_43729_end_mask_0, x = k_57_cast_fp16)[name = string("op_43729_cast_fp16")]; + tensor var_43733_begin_0 = const()[name = string("op_43733_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_43733_end_0 = const()[name = string("op_43733_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_43733_end_mask_0 = const()[name = string("op_43733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43733_cast_fp16 = slice_by_index(begin = var_43733_begin_0, end = var_43733_end_0, end_mask = var_43733_end_mask_0, x = k_57_cast_fp16)[name = string("op_43733_cast_fp16")]; + tensor var_43737_begin_0 = const()[name = string("op_43737_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_43737_end_0 = const()[name = string("op_43737_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_43737_end_mask_0 = const()[name = string("op_43737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43737_cast_fp16 = slice_by_index(begin = var_43737_begin_0, end = var_43737_end_0, end_mask = var_43737_end_mask_0, x = k_57_cast_fp16)[name = string("op_43737_cast_fp16")]; + tensor var_43739_begin_0 = const()[name = string("op_43739_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43739_end_0 = const()[name = string("op_43739_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43739_end_mask_0 = const()[name = string("op_43739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43739_cast_fp16 = slice_by_index(begin = var_43739_begin_0, end = var_43739_end_0, end_mask = var_43739_end_mask_0, x = value_57_cast_fp16)[name = string("op_43739_cast_fp16")]; + tensor var_43743_begin_0 = const()[name = string("op_43743_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43743_end_0 = const()[name = string("op_43743_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_43743_end_mask_0 = const()[name = string("op_43743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43743_cast_fp16 = slice_by_index(begin = var_43743_begin_0, end = var_43743_end_0, end_mask = var_43743_end_mask_0, x = value_57_cast_fp16)[name = string("op_43743_cast_fp16")]; + tensor var_43747_begin_0 = const()[name = string("op_43747_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43747_end_0 = const()[name = string("op_43747_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_43747_end_mask_0 = const()[name = string("op_43747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43747_cast_fp16 = slice_by_index(begin = var_43747_begin_0, end = var_43747_end_0, end_mask = var_43747_end_mask_0, x = value_57_cast_fp16)[name = string("op_43747_cast_fp16")]; + tensor var_43751_begin_0 = const()[name = string("op_43751_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43751_end_0 = const()[name = string("op_43751_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_43751_end_mask_0 = const()[name = string("op_43751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43751_cast_fp16 = slice_by_index(begin = var_43751_begin_0, end = var_43751_end_0, end_mask = var_43751_end_mask_0, x = value_57_cast_fp16)[name = string("op_43751_cast_fp16")]; + tensor var_43755_begin_0 = const()[name = string("op_43755_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43755_end_0 = const()[name = string("op_43755_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_43755_end_mask_0 = const()[name = string("op_43755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43755_cast_fp16 = slice_by_index(begin = var_43755_begin_0, end = var_43755_end_0, end_mask = var_43755_end_mask_0, x = value_57_cast_fp16)[name = string("op_43755_cast_fp16")]; + tensor var_43759_begin_0 = const()[name = string("op_43759_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43759_end_0 = const()[name = string("op_43759_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_43759_end_mask_0 = const()[name = string("op_43759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43759_cast_fp16 = slice_by_index(begin = var_43759_begin_0, end = var_43759_end_0, end_mask = var_43759_end_mask_0, x = value_57_cast_fp16)[name = string("op_43759_cast_fp16")]; + tensor var_43763_begin_0 = const()[name = string("op_43763_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43763_end_0 = const()[name = string("op_43763_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_43763_end_mask_0 = const()[name = string("op_43763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43763_cast_fp16 = slice_by_index(begin = var_43763_begin_0, end = var_43763_end_0, end_mask = var_43763_end_mask_0, x = value_57_cast_fp16)[name = string("op_43763_cast_fp16")]; + tensor var_43767_begin_0 = const()[name = string("op_43767_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43767_end_0 = const()[name = string("op_43767_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_43767_end_mask_0 = const()[name = string("op_43767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43767_cast_fp16 = slice_by_index(begin = var_43767_begin_0, end = var_43767_end_0, end_mask = var_43767_end_mask_0, x = value_57_cast_fp16)[name = string("op_43767_cast_fp16")]; + tensor var_43771_begin_0 = const()[name = string("op_43771_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43771_end_0 = const()[name = string("op_43771_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_43771_end_mask_0 = const()[name = string("op_43771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43771_cast_fp16 = slice_by_index(begin = var_43771_begin_0, end = var_43771_end_0, end_mask = var_43771_end_mask_0, x = value_57_cast_fp16)[name = string("op_43771_cast_fp16")]; + tensor var_43775_begin_0 = const()[name = string("op_43775_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43775_end_0 = const()[name = string("op_43775_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_43775_end_mask_0 = const()[name = string("op_43775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43775_cast_fp16 = slice_by_index(begin = var_43775_begin_0, end = var_43775_end_0, end_mask = var_43775_end_mask_0, x = value_57_cast_fp16)[name = string("op_43775_cast_fp16")]; + tensor var_43779_begin_0 = const()[name = string("op_43779_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43779_end_0 = const()[name = string("op_43779_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_43779_end_mask_0 = const()[name = string("op_43779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43779_cast_fp16 = slice_by_index(begin = var_43779_begin_0, end = var_43779_end_0, end_mask = var_43779_end_mask_0, x = value_57_cast_fp16)[name = string("op_43779_cast_fp16")]; + tensor var_43783_begin_0 = const()[name = string("op_43783_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43783_end_0 = const()[name = string("op_43783_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_43783_end_mask_0 = const()[name = string("op_43783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43783_cast_fp16 = slice_by_index(begin = var_43783_begin_0, end = var_43783_end_0, end_mask = var_43783_end_mask_0, x = value_57_cast_fp16)[name = string("op_43783_cast_fp16")]; + tensor var_43787_begin_0 = const()[name = string("op_43787_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43787_end_0 = const()[name = string("op_43787_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_43787_end_mask_0 = const()[name = string("op_43787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43787_cast_fp16 = slice_by_index(begin = var_43787_begin_0, end = var_43787_end_0, end_mask = var_43787_end_mask_0, x = value_57_cast_fp16)[name = string("op_43787_cast_fp16")]; + tensor var_43791_begin_0 = const()[name = string("op_43791_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43791_end_0 = const()[name = string("op_43791_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_43791_end_mask_0 = const()[name = string("op_43791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43791_cast_fp16 = slice_by_index(begin = var_43791_begin_0, end = var_43791_end_0, end_mask = var_43791_end_mask_0, x = value_57_cast_fp16)[name = string("op_43791_cast_fp16")]; + tensor var_43795_begin_0 = const()[name = string("op_43795_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43795_end_0 = const()[name = string("op_43795_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_43795_end_mask_0 = const()[name = string("op_43795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43795_cast_fp16 = slice_by_index(begin = var_43795_begin_0, end = var_43795_end_0, end_mask = var_43795_end_mask_0, x = value_57_cast_fp16)[name = string("op_43795_cast_fp16")]; + tensor var_43799_begin_0 = const()[name = string("op_43799_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43799_end_0 = const()[name = string("op_43799_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_43799_end_mask_0 = const()[name = string("op_43799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43799_cast_fp16 = slice_by_index(begin = var_43799_begin_0, end = var_43799_end_0, end_mask = var_43799_end_mask_0, x = value_57_cast_fp16)[name = string("op_43799_cast_fp16")]; + tensor var_43803_begin_0 = const()[name = string("op_43803_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43803_end_0 = const()[name = string("op_43803_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_43803_end_mask_0 = const()[name = string("op_43803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43803_cast_fp16 = slice_by_index(begin = var_43803_begin_0, end = var_43803_end_0, end_mask = var_43803_end_mask_0, x = value_57_cast_fp16)[name = string("op_43803_cast_fp16")]; + tensor var_43807_begin_0 = const()[name = string("op_43807_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43807_end_0 = const()[name = string("op_43807_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_43807_end_mask_0 = const()[name = string("op_43807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43807_cast_fp16 = slice_by_index(begin = var_43807_begin_0, end = var_43807_end_0, end_mask = var_43807_end_mask_0, x = value_57_cast_fp16)[name = string("op_43807_cast_fp16")]; + tensor var_43811_begin_0 = const()[name = string("op_43811_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43811_end_0 = const()[name = string("op_43811_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_43811_end_mask_0 = const()[name = string("op_43811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43811_cast_fp16 = slice_by_index(begin = var_43811_begin_0, end = var_43811_end_0, end_mask = var_43811_end_mask_0, x = value_57_cast_fp16)[name = string("op_43811_cast_fp16")]; + tensor var_43815_begin_0 = const()[name = string("op_43815_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43815_end_0 = const()[name = string("op_43815_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_43815_end_mask_0 = const()[name = string("op_43815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43815_cast_fp16 = slice_by_index(begin = var_43815_begin_0, end = var_43815_end_0, end_mask = var_43815_end_mask_0, x = value_57_cast_fp16)[name = string("op_43815_cast_fp16")]; + string _SplitHeadsQ__mh_w_4481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4481_equation_0, values = (var_43661_cast_fp16, var_43103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4481_cast_fp16")]; + string _SplitHeadsQ__mh_w_4483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4483_equation_0, values = (var_43661_cast_fp16, var_43110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4483_cast_fp16")]; + string _SplitHeadsQ__mh_w_4485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4485_equation_0, values = (var_43661_cast_fp16, var_43117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4485_cast_fp16")]; + string _SplitHeadsQ__mh_w_4487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4487_equation_0, values = (var_43661_cast_fp16, var_43124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4487_cast_fp16")]; + string _SplitHeadsQ__mh_w_4489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4489_equation_0, values = (var_43665_cast_fp16, var_43131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4489_cast_fp16")]; + string _SplitHeadsQ__mh_w_4491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4491_equation_0, values = (var_43665_cast_fp16, var_43138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4491_cast_fp16")]; + string _SplitHeadsQ__mh_w_4493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4493_equation_0, values = (var_43665_cast_fp16, var_43145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4493_cast_fp16")]; + string _SplitHeadsQ__mh_w_4495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4495_equation_0, values = (var_43665_cast_fp16, var_43152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4495_cast_fp16")]; + string _SplitHeadsQ__mh_w_4497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4497_equation_0, values = (var_43669_cast_fp16, var_43159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4497_cast_fp16")]; + string _SplitHeadsQ__mh_w_4499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4499_equation_0, values = (var_43669_cast_fp16, var_43166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4499_cast_fp16")]; + string _SplitHeadsQ__mh_w_4501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4501_equation_0, values = (var_43669_cast_fp16, var_43173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4501_cast_fp16")]; + string _SplitHeadsQ__mh_w_4503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4503_equation_0, values = (var_43669_cast_fp16, var_43180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4503_cast_fp16")]; + string _SplitHeadsQ__mh_w_4505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4505_equation_0, values = (var_43673_cast_fp16, var_43187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4505_cast_fp16")]; + string _SplitHeadsQ__mh_w_4507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4507_equation_0, values = (var_43673_cast_fp16, var_43194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4507_cast_fp16")]; + string _SplitHeadsQ__mh_w_4509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4509_equation_0, values = (var_43673_cast_fp16, var_43201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4509_cast_fp16")]; + string _SplitHeadsQ__mh_w_4511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4511_equation_0, values = (var_43673_cast_fp16, var_43208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4511_cast_fp16")]; + string _SplitHeadsQ__mh_w_4513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4513_equation_0, values = (var_43677_cast_fp16, var_43215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4513_cast_fp16")]; + string _SplitHeadsQ__mh_w_4515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4515_equation_0, values = (var_43677_cast_fp16, var_43222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4515_cast_fp16")]; + string _SplitHeadsQ__mh_w_4517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4517_equation_0, values = (var_43677_cast_fp16, var_43229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4517_cast_fp16")]; + string _SplitHeadsQ__mh_w_4519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4519_equation_0, values = (var_43677_cast_fp16, var_43236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4519_cast_fp16")]; + string _SplitHeadsQ__mh_w_4521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4521_equation_0, values = (var_43681_cast_fp16, var_43243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4521_cast_fp16")]; + string _SplitHeadsQ__mh_w_4523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4523_equation_0, values = (var_43681_cast_fp16, var_43250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4523_cast_fp16")]; + string _SplitHeadsQ__mh_w_4525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4525_equation_0, values = (var_43681_cast_fp16, var_43257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4525_cast_fp16")]; + string _SplitHeadsQ__mh_w_4527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4527_equation_0, values = (var_43681_cast_fp16, var_43264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4527_cast_fp16")]; + string _SplitHeadsQ__mh_w_4529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4529_equation_0, values = (var_43685_cast_fp16, var_43271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4529_cast_fp16")]; + string _SplitHeadsQ__mh_w_4531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4531_equation_0, values = (var_43685_cast_fp16, var_43278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4531_cast_fp16")]; + string _SplitHeadsQ__mh_w_4533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4533_equation_0, values = (var_43685_cast_fp16, var_43285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4533_cast_fp16")]; + string _SplitHeadsQ__mh_w_4535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4535_equation_0, values = (var_43685_cast_fp16, var_43292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4535_cast_fp16")]; + string _SplitHeadsQ__mh_w_4537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4537_equation_0, values = (var_43689_cast_fp16, var_43299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4537_cast_fp16")]; + string _SplitHeadsQ__mh_w_4539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4539_equation_0, values = (var_43689_cast_fp16, var_43306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4539_cast_fp16")]; + string _SplitHeadsQ__mh_w_4541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4541_equation_0, values = (var_43689_cast_fp16, var_43313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4541_cast_fp16")]; + string _SplitHeadsQ__mh_w_4543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4543_equation_0, values = (var_43689_cast_fp16, var_43320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4543_cast_fp16")]; + string _SplitHeadsQ__mh_w_4545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4545_equation_0, values = (var_43693_cast_fp16, var_43327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4545_cast_fp16")]; + string _SplitHeadsQ__mh_w_4547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4547_equation_0, values = (var_43693_cast_fp16, var_43334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4547_cast_fp16")]; + string _SplitHeadsQ__mh_w_4549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4549_equation_0, values = (var_43693_cast_fp16, var_43341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4549_cast_fp16")]; + string _SplitHeadsQ__mh_w_4551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4551_equation_0, values = (var_43693_cast_fp16, var_43348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4551_cast_fp16")]; + string _SplitHeadsQ__mh_w_4553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4553_equation_0, values = (var_43697_cast_fp16, var_43355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4553_cast_fp16")]; + string _SplitHeadsQ__mh_w_4555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4555_equation_0, values = (var_43697_cast_fp16, var_43362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4555_cast_fp16")]; + string _SplitHeadsQ__mh_w_4557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4557_equation_0, values = (var_43697_cast_fp16, var_43369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4557_cast_fp16")]; + string _SplitHeadsQ__mh_w_4559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4559_equation_0, values = (var_43697_cast_fp16, var_43376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4559_cast_fp16")]; + string _SplitHeadsQ__mh_w_4561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4561_equation_0, values = (var_43701_cast_fp16, var_43383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4561_cast_fp16")]; + string _SplitHeadsQ__mh_w_4563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4563_equation_0, values = (var_43701_cast_fp16, var_43390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4563_cast_fp16")]; + string _SplitHeadsQ__mh_w_4565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4565_equation_0, values = (var_43701_cast_fp16, var_43397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4565_cast_fp16")]; + string _SplitHeadsQ__mh_w_4567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4567_equation_0, values = (var_43701_cast_fp16, var_43404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4567_cast_fp16")]; + string _SplitHeadsQ__mh_w_4569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4569_equation_0, values = (var_43705_cast_fp16, var_43411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4569_cast_fp16")]; + string _SplitHeadsQ__mh_w_4571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4571_equation_0, values = (var_43705_cast_fp16, var_43418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4571_cast_fp16")]; + string _SplitHeadsQ__mh_w_4573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4573_equation_0, values = (var_43705_cast_fp16, var_43425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4573_cast_fp16")]; + string _SplitHeadsQ__mh_w_4575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4575_equation_0, values = (var_43705_cast_fp16, var_43432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4575_cast_fp16")]; + string _SplitHeadsQ__mh_w_4577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4577_equation_0, values = (var_43709_cast_fp16, var_43439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4577_cast_fp16")]; + string _SplitHeadsQ__mh_w_4579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4579_equation_0, values = (var_43709_cast_fp16, var_43446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4579_cast_fp16")]; + string _SplitHeadsQ__mh_w_4581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4581_equation_0, values = (var_43709_cast_fp16, var_43453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4581_cast_fp16")]; + string _SplitHeadsQ__mh_w_4583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4583_equation_0, values = (var_43709_cast_fp16, var_43460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4583_cast_fp16")]; + string _SplitHeadsQ__mh_w_4585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4585_equation_0, values = (var_43713_cast_fp16, var_43467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4585_cast_fp16")]; + string _SplitHeadsQ__mh_w_4587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4587_equation_0, values = (var_43713_cast_fp16, var_43474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4587_cast_fp16")]; + string _SplitHeadsQ__mh_w_4589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4589_equation_0, values = (var_43713_cast_fp16, var_43481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4589_cast_fp16")]; + string _SplitHeadsQ__mh_w_4591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4591_equation_0, values = (var_43713_cast_fp16, var_43488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4591_cast_fp16")]; + string _SplitHeadsQ__mh_w_4593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4593_equation_0, values = (var_43717_cast_fp16, var_43495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4593_cast_fp16")]; + string _SplitHeadsQ__mh_w_4595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4595_equation_0, values = (var_43717_cast_fp16, var_43502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4595_cast_fp16")]; + string _SplitHeadsQ__mh_w_4597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4597_equation_0, values = (var_43717_cast_fp16, var_43509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4597_cast_fp16")]; + string _SplitHeadsQ__mh_w_4599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4599_equation_0, values = (var_43717_cast_fp16, var_43516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4599_cast_fp16")]; + string _SplitHeadsQ__mh_w_4601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4601_equation_0, values = (var_43721_cast_fp16, var_43523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4601_cast_fp16")]; + string _SplitHeadsQ__mh_w_4603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4603_equation_0, values = (var_43721_cast_fp16, var_43530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4603_cast_fp16")]; + string _SplitHeadsQ__mh_w_4605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4605_equation_0, values = (var_43721_cast_fp16, var_43537_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4605_cast_fp16")]; + string _SplitHeadsQ__mh_w_4607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4607_equation_0, values = (var_43721_cast_fp16, var_43544_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4607_cast_fp16")]; + string _SplitHeadsQ__mh_w_4609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4609_equation_0, values = (var_43725_cast_fp16, var_43551_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4609_cast_fp16")]; + string _SplitHeadsQ__mh_w_4611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4611_equation_0, values = (var_43725_cast_fp16, var_43558_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4611_cast_fp16")]; + string _SplitHeadsQ__mh_w_4613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4613_equation_0, values = (var_43725_cast_fp16, var_43565_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4613_cast_fp16")]; + string _SplitHeadsQ__mh_w_4615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4615_equation_0, values = (var_43725_cast_fp16, var_43572_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4615_cast_fp16")]; + string _SplitHeadsQ__mh_w_4617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4617_equation_0, values = (var_43729_cast_fp16, var_43579_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4617_cast_fp16")]; + string _SplitHeadsQ__mh_w_4619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4619_equation_0, values = (var_43729_cast_fp16, var_43586_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4619_cast_fp16")]; + string _SplitHeadsQ__mh_w_4621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4621_equation_0, values = (var_43729_cast_fp16, var_43593_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4621_cast_fp16")]; + string _SplitHeadsQ__mh_w_4623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4623_equation_0, values = (var_43729_cast_fp16, var_43600_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4623_cast_fp16")]; + string _SplitHeadsQ__mh_w_4625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4625_equation_0, values = (var_43733_cast_fp16, var_43607_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4625_cast_fp16")]; + string _SplitHeadsQ__mh_w_4627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4627_equation_0, values = (var_43733_cast_fp16, var_43614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4627_cast_fp16")]; + string _SplitHeadsQ__mh_w_4629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4629_equation_0, values = (var_43733_cast_fp16, var_43621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4629_cast_fp16")]; + string _SplitHeadsQ__mh_w_4631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4631_equation_0, values = (var_43733_cast_fp16, var_43628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4631_cast_fp16")]; + string _SplitHeadsQ__mh_w_4633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4633_equation_0, values = (var_43737_cast_fp16, var_43635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4633_cast_fp16")]; + string _SplitHeadsQ__mh_w_4635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4635_equation_0, values = (var_43737_cast_fp16, var_43642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4635_cast_fp16")]; + string _SplitHeadsQ__mh_w_4637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4637_equation_0, values = (var_43737_cast_fp16, var_43649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4637_cast_fp16")]; + string _SplitHeadsQ__mh_w_4639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4639_equation_0, values = (var_43737_cast_fp16, var_43656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4639_cast_fp16")]; + fp16 var_43978_to_fp16 = const()[name = string("op_43978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4481_cast_fp16, y = var_43978_to_fp16)[name = string("aw_chunk_4481_cast_fp16")]; + fp16 var_43980_to_fp16 = const()[name = string("op_43980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4483_cast_fp16, y = var_43980_to_fp16)[name = string("aw_chunk_4483_cast_fp16")]; + fp16 var_43982_to_fp16 = const()[name = string("op_43982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4485_cast_fp16, y = var_43982_to_fp16)[name = string("aw_chunk_4485_cast_fp16")]; + fp16 var_43984_to_fp16 = const()[name = string("op_43984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4487_cast_fp16, y = var_43984_to_fp16)[name = string("aw_chunk_4487_cast_fp16")]; + fp16 var_43986_to_fp16 = const()[name = string("op_43986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4489_cast_fp16, y = var_43986_to_fp16)[name = string("aw_chunk_4489_cast_fp16")]; + fp16 var_43988_to_fp16 = const()[name = string("op_43988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4491_cast_fp16, y = var_43988_to_fp16)[name = string("aw_chunk_4491_cast_fp16")]; + fp16 var_43990_to_fp16 = const()[name = string("op_43990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4493_cast_fp16, y = var_43990_to_fp16)[name = string("aw_chunk_4493_cast_fp16")]; + fp16 var_43992_to_fp16 = const()[name = string("op_43992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4495_cast_fp16, y = var_43992_to_fp16)[name = string("aw_chunk_4495_cast_fp16")]; + fp16 var_43994_to_fp16 = const()[name = string("op_43994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4497_cast_fp16, y = var_43994_to_fp16)[name = string("aw_chunk_4497_cast_fp16")]; + fp16 var_43996_to_fp16 = const()[name = string("op_43996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4499_cast_fp16, y = var_43996_to_fp16)[name = string("aw_chunk_4499_cast_fp16")]; + fp16 var_43998_to_fp16 = const()[name = string("op_43998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4501_cast_fp16, y = var_43998_to_fp16)[name = string("aw_chunk_4501_cast_fp16")]; + fp16 var_44000_to_fp16 = const()[name = string("op_44000_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4503_cast_fp16, y = var_44000_to_fp16)[name = string("aw_chunk_4503_cast_fp16")]; + fp16 var_44002_to_fp16 = const()[name = string("op_44002_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4505_cast_fp16, y = var_44002_to_fp16)[name = string("aw_chunk_4505_cast_fp16")]; + fp16 var_44004_to_fp16 = const()[name = string("op_44004_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4507_cast_fp16, y = var_44004_to_fp16)[name = string("aw_chunk_4507_cast_fp16")]; + fp16 var_44006_to_fp16 = const()[name = string("op_44006_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4509_cast_fp16, y = var_44006_to_fp16)[name = string("aw_chunk_4509_cast_fp16")]; + fp16 var_44008_to_fp16 = const()[name = string("op_44008_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4511_cast_fp16, y = var_44008_to_fp16)[name = string("aw_chunk_4511_cast_fp16")]; + fp16 var_44010_to_fp16 = const()[name = string("op_44010_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4513_cast_fp16, y = var_44010_to_fp16)[name = string("aw_chunk_4513_cast_fp16")]; + fp16 var_44012_to_fp16 = const()[name = string("op_44012_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4515_cast_fp16, y = var_44012_to_fp16)[name = string("aw_chunk_4515_cast_fp16")]; + fp16 var_44014_to_fp16 = const()[name = string("op_44014_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4517_cast_fp16, y = var_44014_to_fp16)[name = string("aw_chunk_4517_cast_fp16")]; + fp16 var_44016_to_fp16 = const()[name = string("op_44016_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4519_cast_fp16, y = var_44016_to_fp16)[name = string("aw_chunk_4519_cast_fp16")]; + fp16 var_44018_to_fp16 = const()[name = string("op_44018_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4521_cast_fp16, y = var_44018_to_fp16)[name = string("aw_chunk_4521_cast_fp16")]; + fp16 var_44020_to_fp16 = const()[name = string("op_44020_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4523_cast_fp16, y = var_44020_to_fp16)[name = string("aw_chunk_4523_cast_fp16")]; + fp16 var_44022_to_fp16 = const()[name = string("op_44022_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4525_cast_fp16, y = var_44022_to_fp16)[name = string("aw_chunk_4525_cast_fp16")]; + fp16 var_44024_to_fp16 = const()[name = string("op_44024_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4527_cast_fp16, y = var_44024_to_fp16)[name = string("aw_chunk_4527_cast_fp16")]; + fp16 var_44026_to_fp16 = const()[name = string("op_44026_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4529_cast_fp16, y = var_44026_to_fp16)[name = string("aw_chunk_4529_cast_fp16")]; + fp16 var_44028_to_fp16 = const()[name = string("op_44028_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4531_cast_fp16, y = var_44028_to_fp16)[name = string("aw_chunk_4531_cast_fp16")]; + fp16 var_44030_to_fp16 = const()[name = string("op_44030_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4533_cast_fp16, y = var_44030_to_fp16)[name = string("aw_chunk_4533_cast_fp16")]; + fp16 var_44032_to_fp16 = const()[name = string("op_44032_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4535_cast_fp16, y = var_44032_to_fp16)[name = string("aw_chunk_4535_cast_fp16")]; + fp16 var_44034_to_fp16 = const()[name = string("op_44034_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4537_cast_fp16, y = var_44034_to_fp16)[name = string("aw_chunk_4537_cast_fp16")]; + fp16 var_44036_to_fp16 = const()[name = string("op_44036_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4539_cast_fp16, y = var_44036_to_fp16)[name = string("aw_chunk_4539_cast_fp16")]; + fp16 var_44038_to_fp16 = const()[name = string("op_44038_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4541_cast_fp16, y = var_44038_to_fp16)[name = string("aw_chunk_4541_cast_fp16")]; + fp16 var_44040_to_fp16 = const()[name = string("op_44040_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4543_cast_fp16, y = var_44040_to_fp16)[name = string("aw_chunk_4543_cast_fp16")]; + fp16 var_44042_to_fp16 = const()[name = string("op_44042_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4545_cast_fp16, y = var_44042_to_fp16)[name = string("aw_chunk_4545_cast_fp16")]; + fp16 var_44044_to_fp16 = const()[name = string("op_44044_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4547_cast_fp16, y = var_44044_to_fp16)[name = string("aw_chunk_4547_cast_fp16")]; + fp16 var_44046_to_fp16 = const()[name = string("op_44046_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4549_cast_fp16, y = var_44046_to_fp16)[name = string("aw_chunk_4549_cast_fp16")]; + fp16 var_44048_to_fp16 = const()[name = string("op_44048_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4551_cast_fp16, y = var_44048_to_fp16)[name = string("aw_chunk_4551_cast_fp16")]; + fp16 var_44050_to_fp16 = const()[name = string("op_44050_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4553_cast_fp16, y = var_44050_to_fp16)[name = string("aw_chunk_4553_cast_fp16")]; + fp16 var_44052_to_fp16 = const()[name = string("op_44052_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4555_cast_fp16, y = var_44052_to_fp16)[name = string("aw_chunk_4555_cast_fp16")]; + fp16 var_44054_to_fp16 = const()[name = string("op_44054_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4557_cast_fp16, y = var_44054_to_fp16)[name = string("aw_chunk_4557_cast_fp16")]; + fp16 var_44056_to_fp16 = const()[name = string("op_44056_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4559_cast_fp16, y = var_44056_to_fp16)[name = string("aw_chunk_4559_cast_fp16")]; + fp16 var_44058_to_fp16 = const()[name = string("op_44058_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4561_cast_fp16, y = var_44058_to_fp16)[name = string("aw_chunk_4561_cast_fp16")]; + fp16 var_44060_to_fp16 = const()[name = string("op_44060_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4563_cast_fp16, y = var_44060_to_fp16)[name = string("aw_chunk_4563_cast_fp16")]; + fp16 var_44062_to_fp16 = const()[name = string("op_44062_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4565_cast_fp16, y = var_44062_to_fp16)[name = string("aw_chunk_4565_cast_fp16")]; + fp16 var_44064_to_fp16 = const()[name = string("op_44064_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4567_cast_fp16, y = var_44064_to_fp16)[name = string("aw_chunk_4567_cast_fp16")]; + fp16 var_44066_to_fp16 = const()[name = string("op_44066_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4569_cast_fp16, y = var_44066_to_fp16)[name = string("aw_chunk_4569_cast_fp16")]; + fp16 var_44068_to_fp16 = const()[name = string("op_44068_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4571_cast_fp16, y = var_44068_to_fp16)[name = string("aw_chunk_4571_cast_fp16")]; + fp16 var_44070_to_fp16 = const()[name = string("op_44070_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4573_cast_fp16, y = var_44070_to_fp16)[name = string("aw_chunk_4573_cast_fp16")]; + fp16 var_44072_to_fp16 = const()[name = string("op_44072_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4575_cast_fp16, y = var_44072_to_fp16)[name = string("aw_chunk_4575_cast_fp16")]; + fp16 var_44074_to_fp16 = const()[name = string("op_44074_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4577_cast_fp16, y = var_44074_to_fp16)[name = string("aw_chunk_4577_cast_fp16")]; + fp16 var_44076_to_fp16 = const()[name = string("op_44076_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4579_cast_fp16, y = var_44076_to_fp16)[name = string("aw_chunk_4579_cast_fp16")]; + fp16 var_44078_to_fp16 = const()[name = string("op_44078_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4581_cast_fp16, y = var_44078_to_fp16)[name = string("aw_chunk_4581_cast_fp16")]; + fp16 var_44080_to_fp16 = const()[name = string("op_44080_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4583_cast_fp16, y = var_44080_to_fp16)[name = string("aw_chunk_4583_cast_fp16")]; + fp16 var_44082_to_fp16 = const()[name = string("op_44082_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4585_cast_fp16, y = var_44082_to_fp16)[name = string("aw_chunk_4585_cast_fp16")]; + fp16 var_44084_to_fp16 = const()[name = string("op_44084_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4587_cast_fp16, y = var_44084_to_fp16)[name = string("aw_chunk_4587_cast_fp16")]; + fp16 var_44086_to_fp16 = const()[name = string("op_44086_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4589_cast_fp16, y = var_44086_to_fp16)[name = string("aw_chunk_4589_cast_fp16")]; + fp16 var_44088_to_fp16 = const()[name = string("op_44088_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4591_cast_fp16, y = var_44088_to_fp16)[name = string("aw_chunk_4591_cast_fp16")]; + fp16 var_44090_to_fp16 = const()[name = string("op_44090_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4593_cast_fp16, y = var_44090_to_fp16)[name = string("aw_chunk_4593_cast_fp16")]; + fp16 var_44092_to_fp16 = const()[name = string("op_44092_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4595_cast_fp16, y = var_44092_to_fp16)[name = string("aw_chunk_4595_cast_fp16")]; + fp16 var_44094_to_fp16 = const()[name = string("op_44094_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4597_cast_fp16, y = var_44094_to_fp16)[name = string("aw_chunk_4597_cast_fp16")]; + fp16 var_44096_to_fp16 = const()[name = string("op_44096_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4599_cast_fp16, y = var_44096_to_fp16)[name = string("aw_chunk_4599_cast_fp16")]; + fp16 var_44098_to_fp16 = const()[name = string("op_44098_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4601_cast_fp16, y = var_44098_to_fp16)[name = string("aw_chunk_4601_cast_fp16")]; + fp16 var_44100_to_fp16 = const()[name = string("op_44100_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4603_cast_fp16, y = var_44100_to_fp16)[name = string("aw_chunk_4603_cast_fp16")]; + fp16 var_44102_to_fp16 = const()[name = string("op_44102_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4605_cast_fp16, y = var_44102_to_fp16)[name = string("aw_chunk_4605_cast_fp16")]; + fp16 var_44104_to_fp16 = const()[name = string("op_44104_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4607_cast_fp16, y = var_44104_to_fp16)[name = string("aw_chunk_4607_cast_fp16")]; + fp16 var_44106_to_fp16 = const()[name = string("op_44106_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4609_cast_fp16, y = var_44106_to_fp16)[name = string("aw_chunk_4609_cast_fp16")]; + fp16 var_44108_to_fp16 = const()[name = string("op_44108_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4611_cast_fp16, y = var_44108_to_fp16)[name = string("aw_chunk_4611_cast_fp16")]; + fp16 var_44110_to_fp16 = const()[name = string("op_44110_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4613_cast_fp16, y = var_44110_to_fp16)[name = string("aw_chunk_4613_cast_fp16")]; + fp16 var_44112_to_fp16 = const()[name = string("op_44112_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4615_cast_fp16, y = var_44112_to_fp16)[name = string("aw_chunk_4615_cast_fp16")]; + fp16 var_44114_to_fp16 = const()[name = string("op_44114_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4617_cast_fp16, y = var_44114_to_fp16)[name = string("aw_chunk_4617_cast_fp16")]; + fp16 var_44116_to_fp16 = const()[name = string("op_44116_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4619_cast_fp16, y = var_44116_to_fp16)[name = string("aw_chunk_4619_cast_fp16")]; + fp16 var_44118_to_fp16 = const()[name = string("op_44118_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4621_cast_fp16, y = var_44118_to_fp16)[name = string("aw_chunk_4621_cast_fp16")]; + fp16 var_44120_to_fp16 = const()[name = string("op_44120_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4623_cast_fp16, y = var_44120_to_fp16)[name = string("aw_chunk_4623_cast_fp16")]; + fp16 var_44122_to_fp16 = const()[name = string("op_44122_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4625_cast_fp16, y = var_44122_to_fp16)[name = string("aw_chunk_4625_cast_fp16")]; + fp16 var_44124_to_fp16 = const()[name = string("op_44124_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4627_cast_fp16, y = var_44124_to_fp16)[name = string("aw_chunk_4627_cast_fp16")]; + fp16 var_44126_to_fp16 = const()[name = string("op_44126_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4629_cast_fp16, y = var_44126_to_fp16)[name = string("aw_chunk_4629_cast_fp16")]; + fp16 var_44128_to_fp16 = const()[name = string("op_44128_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4631_cast_fp16, y = var_44128_to_fp16)[name = string("aw_chunk_4631_cast_fp16")]; + fp16 var_44130_to_fp16 = const()[name = string("op_44130_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4633_cast_fp16, y = var_44130_to_fp16)[name = string("aw_chunk_4633_cast_fp16")]; + fp16 var_44132_to_fp16 = const()[name = string("op_44132_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4635_cast_fp16, y = var_44132_to_fp16)[name = string("aw_chunk_4635_cast_fp16")]; + fp16 var_44134_to_fp16 = const()[name = string("op_44134_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4637_cast_fp16, y = var_44134_to_fp16)[name = string("aw_chunk_4637_cast_fp16")]; + fp16 var_44136_to_fp16 = const()[name = string("op_44136_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4639_cast_fp16, y = var_44136_to_fp16)[name = string("aw_chunk_4639_cast_fp16")]; + tensor var_44138_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4481_cast_fp16)[name = string("op_44138_cast_fp16")]; + tensor var_44139_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4483_cast_fp16)[name = string("op_44139_cast_fp16")]; + tensor var_44140_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4485_cast_fp16)[name = string("op_44140_cast_fp16")]; + tensor var_44141_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4487_cast_fp16)[name = string("op_44141_cast_fp16")]; + tensor var_44142_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4489_cast_fp16)[name = string("op_44142_cast_fp16")]; + tensor var_44143_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4491_cast_fp16)[name = string("op_44143_cast_fp16")]; + tensor var_44144_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4493_cast_fp16)[name = string("op_44144_cast_fp16")]; + tensor var_44145_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4495_cast_fp16)[name = string("op_44145_cast_fp16")]; + tensor var_44146_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4497_cast_fp16)[name = string("op_44146_cast_fp16")]; + tensor var_44147_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4499_cast_fp16)[name = string("op_44147_cast_fp16")]; + tensor var_44148_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4501_cast_fp16)[name = string("op_44148_cast_fp16")]; + tensor var_44149_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4503_cast_fp16)[name = string("op_44149_cast_fp16")]; + tensor var_44150_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4505_cast_fp16)[name = string("op_44150_cast_fp16")]; + tensor var_44151_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4507_cast_fp16)[name = string("op_44151_cast_fp16")]; + tensor var_44152_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4509_cast_fp16)[name = string("op_44152_cast_fp16")]; + tensor var_44153_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4511_cast_fp16)[name = string("op_44153_cast_fp16")]; + tensor var_44154_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4513_cast_fp16)[name = string("op_44154_cast_fp16")]; + tensor var_44155_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4515_cast_fp16)[name = string("op_44155_cast_fp16")]; + tensor var_44156_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4517_cast_fp16)[name = string("op_44156_cast_fp16")]; + tensor var_44157_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4519_cast_fp16)[name = string("op_44157_cast_fp16")]; + tensor var_44158_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4521_cast_fp16)[name = string("op_44158_cast_fp16")]; + tensor var_44159_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4523_cast_fp16)[name = string("op_44159_cast_fp16")]; + tensor var_44160_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4525_cast_fp16)[name = string("op_44160_cast_fp16")]; + tensor var_44161_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4527_cast_fp16)[name = string("op_44161_cast_fp16")]; + tensor var_44162_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4529_cast_fp16)[name = string("op_44162_cast_fp16")]; + tensor var_44163_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4531_cast_fp16)[name = string("op_44163_cast_fp16")]; + tensor var_44164_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4533_cast_fp16)[name = string("op_44164_cast_fp16")]; + tensor var_44165_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4535_cast_fp16)[name = string("op_44165_cast_fp16")]; + tensor var_44166_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4537_cast_fp16)[name = string("op_44166_cast_fp16")]; + tensor var_44167_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4539_cast_fp16)[name = string("op_44167_cast_fp16")]; + tensor var_44168_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4541_cast_fp16)[name = string("op_44168_cast_fp16")]; + tensor var_44169_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4543_cast_fp16)[name = string("op_44169_cast_fp16")]; + tensor var_44170_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4545_cast_fp16)[name = string("op_44170_cast_fp16")]; + tensor var_44171_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4547_cast_fp16)[name = string("op_44171_cast_fp16")]; + tensor var_44172_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4549_cast_fp16)[name = string("op_44172_cast_fp16")]; + tensor var_44173_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4551_cast_fp16)[name = string("op_44173_cast_fp16")]; + tensor var_44174_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4553_cast_fp16)[name = string("op_44174_cast_fp16")]; + tensor var_44175_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4555_cast_fp16)[name = string("op_44175_cast_fp16")]; + tensor var_44176_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4557_cast_fp16)[name = string("op_44176_cast_fp16")]; + tensor var_44177_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4559_cast_fp16)[name = string("op_44177_cast_fp16")]; + tensor var_44178_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4561_cast_fp16)[name = string("op_44178_cast_fp16")]; + tensor var_44179_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4563_cast_fp16)[name = string("op_44179_cast_fp16")]; + tensor var_44180_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4565_cast_fp16)[name = string("op_44180_cast_fp16")]; + tensor var_44181_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4567_cast_fp16)[name = string("op_44181_cast_fp16")]; + tensor var_44182_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4569_cast_fp16)[name = string("op_44182_cast_fp16")]; + tensor var_44183_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4571_cast_fp16)[name = string("op_44183_cast_fp16")]; + tensor var_44184_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4573_cast_fp16)[name = string("op_44184_cast_fp16")]; + tensor var_44185_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4575_cast_fp16)[name = string("op_44185_cast_fp16")]; + tensor var_44186_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4577_cast_fp16)[name = string("op_44186_cast_fp16")]; + tensor var_44187_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4579_cast_fp16)[name = string("op_44187_cast_fp16")]; + tensor var_44188_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4581_cast_fp16)[name = string("op_44188_cast_fp16")]; + tensor var_44189_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4583_cast_fp16)[name = string("op_44189_cast_fp16")]; + tensor var_44190_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4585_cast_fp16)[name = string("op_44190_cast_fp16")]; + tensor var_44191_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4587_cast_fp16)[name = string("op_44191_cast_fp16")]; + tensor var_44192_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4589_cast_fp16)[name = string("op_44192_cast_fp16")]; + tensor var_44193_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4591_cast_fp16)[name = string("op_44193_cast_fp16")]; + tensor var_44194_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4593_cast_fp16)[name = string("op_44194_cast_fp16")]; + tensor var_44195_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4595_cast_fp16)[name = string("op_44195_cast_fp16")]; + tensor var_44196_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4597_cast_fp16)[name = string("op_44196_cast_fp16")]; + tensor var_44197_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4599_cast_fp16)[name = string("op_44197_cast_fp16")]; + tensor var_44198_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4601_cast_fp16)[name = string("op_44198_cast_fp16")]; + tensor var_44199_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4603_cast_fp16)[name = string("op_44199_cast_fp16")]; + tensor var_44200_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4605_cast_fp16)[name = string("op_44200_cast_fp16")]; + tensor var_44201_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4607_cast_fp16)[name = string("op_44201_cast_fp16")]; + tensor var_44202_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4609_cast_fp16)[name = string("op_44202_cast_fp16")]; + tensor var_44203_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4611_cast_fp16)[name = string("op_44203_cast_fp16")]; + tensor var_44204_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4613_cast_fp16)[name = string("op_44204_cast_fp16")]; + tensor var_44205_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4615_cast_fp16)[name = string("op_44205_cast_fp16")]; + tensor var_44206_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4617_cast_fp16)[name = string("op_44206_cast_fp16")]; + tensor var_44207_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4619_cast_fp16)[name = string("op_44207_cast_fp16")]; + tensor var_44208_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4621_cast_fp16)[name = string("op_44208_cast_fp16")]; + tensor var_44209_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4623_cast_fp16)[name = string("op_44209_cast_fp16")]; + tensor var_44210_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4625_cast_fp16)[name = string("op_44210_cast_fp16")]; + tensor var_44211_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4627_cast_fp16)[name = string("op_44211_cast_fp16")]; + tensor var_44212_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4629_cast_fp16)[name = string("op_44212_cast_fp16")]; + tensor var_44213_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4631_cast_fp16)[name = string("op_44213_cast_fp16")]; + tensor var_44214_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4633_cast_fp16)[name = string("op_44214_cast_fp16")]; + tensor var_44215_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4635_cast_fp16)[name = string("op_44215_cast_fp16")]; + tensor var_44216_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4637_cast_fp16)[name = string("op_44216_cast_fp16")]; + tensor var_44217_cast_fp16 = softmax(axis = var_42963, x = aw_chunk_4639_cast_fp16)[name = string("op_44217_cast_fp16")]; + string var_44219_equation_0 = const()[name = string("op_44219_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44219_cast_fp16 = einsum(equation = var_44219_equation_0, values = (var_43739_cast_fp16, var_44138_cast_fp16))[name = string("op_44219_cast_fp16")]; + string var_44221_equation_0 = const()[name = string("op_44221_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44221_cast_fp16 = einsum(equation = var_44221_equation_0, values = (var_43739_cast_fp16, var_44139_cast_fp16))[name = string("op_44221_cast_fp16")]; + string var_44223_equation_0 = const()[name = string("op_44223_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44223_cast_fp16 = einsum(equation = var_44223_equation_0, values = (var_43739_cast_fp16, var_44140_cast_fp16))[name = string("op_44223_cast_fp16")]; + string var_44225_equation_0 = const()[name = string("op_44225_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44225_cast_fp16 = einsum(equation = var_44225_equation_0, values = (var_43739_cast_fp16, var_44141_cast_fp16))[name = string("op_44225_cast_fp16")]; + string var_44227_equation_0 = const()[name = string("op_44227_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44227_cast_fp16 = einsum(equation = var_44227_equation_0, values = (var_43743_cast_fp16, var_44142_cast_fp16))[name = string("op_44227_cast_fp16")]; + string var_44229_equation_0 = const()[name = string("op_44229_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44229_cast_fp16 = einsum(equation = var_44229_equation_0, values = (var_43743_cast_fp16, var_44143_cast_fp16))[name = string("op_44229_cast_fp16")]; + string var_44231_equation_0 = const()[name = string("op_44231_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44231_cast_fp16 = einsum(equation = var_44231_equation_0, values = (var_43743_cast_fp16, var_44144_cast_fp16))[name = string("op_44231_cast_fp16")]; + string var_44233_equation_0 = const()[name = string("op_44233_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44233_cast_fp16 = einsum(equation = var_44233_equation_0, values = (var_43743_cast_fp16, var_44145_cast_fp16))[name = string("op_44233_cast_fp16")]; + string var_44235_equation_0 = const()[name = string("op_44235_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44235_cast_fp16 = einsum(equation = var_44235_equation_0, values = (var_43747_cast_fp16, var_44146_cast_fp16))[name = string("op_44235_cast_fp16")]; + string var_44237_equation_0 = const()[name = string("op_44237_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44237_cast_fp16 = einsum(equation = var_44237_equation_0, values = (var_43747_cast_fp16, var_44147_cast_fp16))[name = string("op_44237_cast_fp16")]; + string var_44239_equation_0 = const()[name = string("op_44239_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44239_cast_fp16 = einsum(equation = var_44239_equation_0, values = (var_43747_cast_fp16, var_44148_cast_fp16))[name = string("op_44239_cast_fp16")]; + string var_44241_equation_0 = const()[name = string("op_44241_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44241_cast_fp16 = einsum(equation = var_44241_equation_0, values = (var_43747_cast_fp16, var_44149_cast_fp16))[name = string("op_44241_cast_fp16")]; + string var_44243_equation_0 = const()[name = string("op_44243_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44243_cast_fp16 = einsum(equation = var_44243_equation_0, values = (var_43751_cast_fp16, var_44150_cast_fp16))[name = string("op_44243_cast_fp16")]; + string var_44245_equation_0 = const()[name = string("op_44245_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44245_cast_fp16 = einsum(equation = var_44245_equation_0, values = (var_43751_cast_fp16, var_44151_cast_fp16))[name = string("op_44245_cast_fp16")]; + string var_44247_equation_0 = const()[name = string("op_44247_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44247_cast_fp16 = einsum(equation = var_44247_equation_0, values = (var_43751_cast_fp16, var_44152_cast_fp16))[name = string("op_44247_cast_fp16")]; + string var_44249_equation_0 = const()[name = string("op_44249_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44249_cast_fp16 = einsum(equation = var_44249_equation_0, values = (var_43751_cast_fp16, var_44153_cast_fp16))[name = string("op_44249_cast_fp16")]; + string var_44251_equation_0 = const()[name = string("op_44251_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44251_cast_fp16 = einsum(equation = var_44251_equation_0, values = (var_43755_cast_fp16, var_44154_cast_fp16))[name = string("op_44251_cast_fp16")]; + string var_44253_equation_0 = const()[name = string("op_44253_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44253_cast_fp16 = einsum(equation = var_44253_equation_0, values = (var_43755_cast_fp16, var_44155_cast_fp16))[name = string("op_44253_cast_fp16")]; + string var_44255_equation_0 = const()[name = string("op_44255_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44255_cast_fp16 = einsum(equation = var_44255_equation_0, values = (var_43755_cast_fp16, var_44156_cast_fp16))[name = string("op_44255_cast_fp16")]; + string var_44257_equation_0 = const()[name = string("op_44257_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44257_cast_fp16 = einsum(equation = var_44257_equation_0, values = (var_43755_cast_fp16, var_44157_cast_fp16))[name = string("op_44257_cast_fp16")]; + string var_44259_equation_0 = const()[name = string("op_44259_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44259_cast_fp16 = einsum(equation = var_44259_equation_0, values = (var_43759_cast_fp16, var_44158_cast_fp16))[name = string("op_44259_cast_fp16")]; + string var_44261_equation_0 = const()[name = string("op_44261_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44261_cast_fp16 = einsum(equation = var_44261_equation_0, values = (var_43759_cast_fp16, var_44159_cast_fp16))[name = string("op_44261_cast_fp16")]; + string var_44263_equation_0 = const()[name = string("op_44263_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44263_cast_fp16 = einsum(equation = var_44263_equation_0, values = (var_43759_cast_fp16, var_44160_cast_fp16))[name = string("op_44263_cast_fp16")]; + string var_44265_equation_0 = const()[name = string("op_44265_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44265_cast_fp16 = einsum(equation = var_44265_equation_0, values = (var_43759_cast_fp16, var_44161_cast_fp16))[name = string("op_44265_cast_fp16")]; + string var_44267_equation_0 = const()[name = string("op_44267_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44267_cast_fp16 = einsum(equation = var_44267_equation_0, values = (var_43763_cast_fp16, var_44162_cast_fp16))[name = string("op_44267_cast_fp16")]; + string var_44269_equation_0 = const()[name = string("op_44269_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44269_cast_fp16 = einsum(equation = var_44269_equation_0, values = (var_43763_cast_fp16, var_44163_cast_fp16))[name = string("op_44269_cast_fp16")]; + string var_44271_equation_0 = const()[name = string("op_44271_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44271_cast_fp16 = einsum(equation = var_44271_equation_0, values = (var_43763_cast_fp16, var_44164_cast_fp16))[name = string("op_44271_cast_fp16")]; + string var_44273_equation_0 = const()[name = string("op_44273_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44273_cast_fp16 = einsum(equation = var_44273_equation_0, values = (var_43763_cast_fp16, var_44165_cast_fp16))[name = string("op_44273_cast_fp16")]; + string var_44275_equation_0 = const()[name = string("op_44275_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44275_cast_fp16 = einsum(equation = var_44275_equation_0, values = (var_43767_cast_fp16, var_44166_cast_fp16))[name = string("op_44275_cast_fp16")]; + string var_44277_equation_0 = const()[name = string("op_44277_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44277_cast_fp16 = einsum(equation = var_44277_equation_0, values = (var_43767_cast_fp16, var_44167_cast_fp16))[name = string("op_44277_cast_fp16")]; + string var_44279_equation_0 = const()[name = string("op_44279_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44279_cast_fp16 = einsum(equation = var_44279_equation_0, values = (var_43767_cast_fp16, var_44168_cast_fp16))[name = string("op_44279_cast_fp16")]; + string var_44281_equation_0 = const()[name = string("op_44281_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44281_cast_fp16 = einsum(equation = var_44281_equation_0, values = (var_43767_cast_fp16, var_44169_cast_fp16))[name = string("op_44281_cast_fp16")]; + string var_44283_equation_0 = const()[name = string("op_44283_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44283_cast_fp16 = einsum(equation = var_44283_equation_0, values = (var_43771_cast_fp16, var_44170_cast_fp16))[name = string("op_44283_cast_fp16")]; + string var_44285_equation_0 = const()[name = string("op_44285_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44285_cast_fp16 = einsum(equation = var_44285_equation_0, values = (var_43771_cast_fp16, var_44171_cast_fp16))[name = string("op_44285_cast_fp16")]; + string var_44287_equation_0 = const()[name = string("op_44287_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44287_cast_fp16 = einsum(equation = var_44287_equation_0, values = (var_43771_cast_fp16, var_44172_cast_fp16))[name = string("op_44287_cast_fp16")]; + string var_44289_equation_0 = const()[name = string("op_44289_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44289_cast_fp16 = einsum(equation = var_44289_equation_0, values = (var_43771_cast_fp16, var_44173_cast_fp16))[name = string("op_44289_cast_fp16")]; + string var_44291_equation_0 = const()[name = string("op_44291_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44291_cast_fp16 = einsum(equation = var_44291_equation_0, values = (var_43775_cast_fp16, var_44174_cast_fp16))[name = string("op_44291_cast_fp16")]; + string var_44293_equation_0 = const()[name = string("op_44293_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44293_cast_fp16 = einsum(equation = var_44293_equation_0, values = (var_43775_cast_fp16, var_44175_cast_fp16))[name = string("op_44293_cast_fp16")]; + string var_44295_equation_0 = const()[name = string("op_44295_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44295_cast_fp16 = einsum(equation = var_44295_equation_0, values = (var_43775_cast_fp16, var_44176_cast_fp16))[name = string("op_44295_cast_fp16")]; + string var_44297_equation_0 = const()[name = string("op_44297_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44297_cast_fp16 = einsum(equation = var_44297_equation_0, values = (var_43775_cast_fp16, var_44177_cast_fp16))[name = string("op_44297_cast_fp16")]; + string var_44299_equation_0 = const()[name = string("op_44299_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44299_cast_fp16 = einsum(equation = var_44299_equation_0, values = (var_43779_cast_fp16, var_44178_cast_fp16))[name = string("op_44299_cast_fp16")]; + string var_44301_equation_0 = const()[name = string("op_44301_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44301_cast_fp16 = einsum(equation = var_44301_equation_0, values = (var_43779_cast_fp16, var_44179_cast_fp16))[name = string("op_44301_cast_fp16")]; + string var_44303_equation_0 = const()[name = string("op_44303_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44303_cast_fp16 = einsum(equation = var_44303_equation_0, values = (var_43779_cast_fp16, var_44180_cast_fp16))[name = string("op_44303_cast_fp16")]; + string var_44305_equation_0 = const()[name = string("op_44305_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44305_cast_fp16 = einsum(equation = var_44305_equation_0, values = (var_43779_cast_fp16, var_44181_cast_fp16))[name = string("op_44305_cast_fp16")]; + string var_44307_equation_0 = const()[name = string("op_44307_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44307_cast_fp16 = einsum(equation = var_44307_equation_0, values = (var_43783_cast_fp16, var_44182_cast_fp16))[name = string("op_44307_cast_fp16")]; + string var_44309_equation_0 = const()[name = string("op_44309_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44309_cast_fp16 = einsum(equation = var_44309_equation_0, values = (var_43783_cast_fp16, var_44183_cast_fp16))[name = string("op_44309_cast_fp16")]; + string var_44311_equation_0 = const()[name = string("op_44311_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44311_cast_fp16 = einsum(equation = var_44311_equation_0, values = (var_43783_cast_fp16, var_44184_cast_fp16))[name = string("op_44311_cast_fp16")]; + string var_44313_equation_0 = const()[name = string("op_44313_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44313_cast_fp16 = einsum(equation = var_44313_equation_0, values = (var_43783_cast_fp16, var_44185_cast_fp16))[name = string("op_44313_cast_fp16")]; + string var_44315_equation_0 = const()[name = string("op_44315_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44315_cast_fp16 = einsum(equation = var_44315_equation_0, values = (var_43787_cast_fp16, var_44186_cast_fp16))[name = string("op_44315_cast_fp16")]; + string var_44317_equation_0 = const()[name = string("op_44317_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44317_cast_fp16 = einsum(equation = var_44317_equation_0, values = (var_43787_cast_fp16, var_44187_cast_fp16))[name = string("op_44317_cast_fp16")]; + string var_44319_equation_0 = const()[name = string("op_44319_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44319_cast_fp16 = einsum(equation = var_44319_equation_0, values = (var_43787_cast_fp16, var_44188_cast_fp16))[name = string("op_44319_cast_fp16")]; + string var_44321_equation_0 = const()[name = string("op_44321_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44321_cast_fp16 = einsum(equation = var_44321_equation_0, values = (var_43787_cast_fp16, var_44189_cast_fp16))[name = string("op_44321_cast_fp16")]; + string var_44323_equation_0 = const()[name = string("op_44323_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44323_cast_fp16 = einsum(equation = var_44323_equation_0, values = (var_43791_cast_fp16, var_44190_cast_fp16))[name = string("op_44323_cast_fp16")]; + string var_44325_equation_0 = const()[name = string("op_44325_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44325_cast_fp16 = einsum(equation = var_44325_equation_0, values = (var_43791_cast_fp16, var_44191_cast_fp16))[name = string("op_44325_cast_fp16")]; + string var_44327_equation_0 = const()[name = string("op_44327_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44327_cast_fp16 = einsum(equation = var_44327_equation_0, values = (var_43791_cast_fp16, var_44192_cast_fp16))[name = string("op_44327_cast_fp16")]; + string var_44329_equation_0 = const()[name = string("op_44329_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44329_cast_fp16 = einsum(equation = var_44329_equation_0, values = (var_43791_cast_fp16, var_44193_cast_fp16))[name = string("op_44329_cast_fp16")]; + string var_44331_equation_0 = const()[name = string("op_44331_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44331_cast_fp16 = einsum(equation = var_44331_equation_0, values = (var_43795_cast_fp16, var_44194_cast_fp16))[name = string("op_44331_cast_fp16")]; + string var_44333_equation_0 = const()[name = string("op_44333_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44333_cast_fp16 = einsum(equation = var_44333_equation_0, values = (var_43795_cast_fp16, var_44195_cast_fp16))[name = string("op_44333_cast_fp16")]; + string var_44335_equation_0 = const()[name = string("op_44335_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44335_cast_fp16 = einsum(equation = var_44335_equation_0, values = (var_43795_cast_fp16, var_44196_cast_fp16))[name = string("op_44335_cast_fp16")]; + string var_44337_equation_0 = const()[name = string("op_44337_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44337_cast_fp16 = einsum(equation = var_44337_equation_0, values = (var_43795_cast_fp16, var_44197_cast_fp16))[name = string("op_44337_cast_fp16")]; + string var_44339_equation_0 = const()[name = string("op_44339_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44339_cast_fp16 = einsum(equation = var_44339_equation_0, values = (var_43799_cast_fp16, var_44198_cast_fp16))[name = string("op_44339_cast_fp16")]; + string var_44341_equation_0 = const()[name = string("op_44341_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44341_cast_fp16 = einsum(equation = var_44341_equation_0, values = (var_43799_cast_fp16, var_44199_cast_fp16))[name = string("op_44341_cast_fp16")]; + string var_44343_equation_0 = const()[name = string("op_44343_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44343_cast_fp16 = einsum(equation = var_44343_equation_0, values = (var_43799_cast_fp16, var_44200_cast_fp16))[name = string("op_44343_cast_fp16")]; + string var_44345_equation_0 = const()[name = string("op_44345_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44345_cast_fp16 = einsum(equation = var_44345_equation_0, values = (var_43799_cast_fp16, var_44201_cast_fp16))[name = string("op_44345_cast_fp16")]; + string var_44347_equation_0 = const()[name = string("op_44347_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44347_cast_fp16 = einsum(equation = var_44347_equation_0, values = (var_43803_cast_fp16, var_44202_cast_fp16))[name = string("op_44347_cast_fp16")]; + string var_44349_equation_0 = const()[name = string("op_44349_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44349_cast_fp16 = einsum(equation = var_44349_equation_0, values = (var_43803_cast_fp16, var_44203_cast_fp16))[name = string("op_44349_cast_fp16")]; + string var_44351_equation_0 = const()[name = string("op_44351_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44351_cast_fp16 = einsum(equation = var_44351_equation_0, values = (var_43803_cast_fp16, var_44204_cast_fp16))[name = string("op_44351_cast_fp16")]; + string var_44353_equation_0 = const()[name = string("op_44353_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44353_cast_fp16 = einsum(equation = var_44353_equation_0, values = (var_43803_cast_fp16, var_44205_cast_fp16))[name = string("op_44353_cast_fp16")]; + string var_44355_equation_0 = const()[name = string("op_44355_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44355_cast_fp16 = einsum(equation = var_44355_equation_0, values = (var_43807_cast_fp16, var_44206_cast_fp16))[name = string("op_44355_cast_fp16")]; + string var_44357_equation_0 = const()[name = string("op_44357_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44357_cast_fp16 = einsum(equation = var_44357_equation_0, values = (var_43807_cast_fp16, var_44207_cast_fp16))[name = string("op_44357_cast_fp16")]; + string var_44359_equation_0 = const()[name = string("op_44359_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44359_cast_fp16 = einsum(equation = var_44359_equation_0, values = (var_43807_cast_fp16, var_44208_cast_fp16))[name = string("op_44359_cast_fp16")]; + string var_44361_equation_0 = const()[name = string("op_44361_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44361_cast_fp16 = einsum(equation = var_44361_equation_0, values = (var_43807_cast_fp16, var_44209_cast_fp16))[name = string("op_44361_cast_fp16")]; + string var_44363_equation_0 = const()[name = string("op_44363_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44363_cast_fp16 = einsum(equation = var_44363_equation_0, values = (var_43811_cast_fp16, var_44210_cast_fp16))[name = string("op_44363_cast_fp16")]; + string var_44365_equation_0 = const()[name = string("op_44365_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44365_cast_fp16 = einsum(equation = var_44365_equation_0, values = (var_43811_cast_fp16, var_44211_cast_fp16))[name = string("op_44365_cast_fp16")]; + string var_44367_equation_0 = const()[name = string("op_44367_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44367_cast_fp16 = einsum(equation = var_44367_equation_0, values = (var_43811_cast_fp16, var_44212_cast_fp16))[name = string("op_44367_cast_fp16")]; + string var_44369_equation_0 = const()[name = string("op_44369_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44369_cast_fp16 = einsum(equation = var_44369_equation_0, values = (var_43811_cast_fp16, var_44213_cast_fp16))[name = string("op_44369_cast_fp16")]; + string var_44371_equation_0 = const()[name = string("op_44371_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44371_cast_fp16 = einsum(equation = var_44371_equation_0, values = (var_43815_cast_fp16, var_44214_cast_fp16))[name = string("op_44371_cast_fp16")]; + string var_44373_equation_0 = const()[name = string("op_44373_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44373_cast_fp16 = einsum(equation = var_44373_equation_0, values = (var_43815_cast_fp16, var_44215_cast_fp16))[name = string("op_44373_cast_fp16")]; + string var_44375_equation_0 = const()[name = string("op_44375_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44375_cast_fp16 = einsum(equation = var_44375_equation_0, values = (var_43815_cast_fp16, var_44216_cast_fp16))[name = string("op_44375_cast_fp16")]; + string var_44377_equation_0 = const()[name = string("op_44377_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_44377_cast_fp16 = einsum(equation = var_44377_equation_0, values = (var_43815_cast_fp16, var_44217_cast_fp16))[name = string("op_44377_cast_fp16")]; + bool var_44379_interleave_0 = const()[name = string("op_44379_interleave_0"), val = bool(false)]; + tensor var_44379_cast_fp16 = concat(axis = var_42938, interleave = var_44379_interleave_0, values = (var_44219_cast_fp16, var_44221_cast_fp16, var_44223_cast_fp16, var_44225_cast_fp16))[name = string("op_44379_cast_fp16")]; + bool var_44381_interleave_0 = const()[name = string("op_44381_interleave_0"), val = bool(false)]; + tensor var_44381_cast_fp16 = concat(axis = var_42938, interleave = var_44381_interleave_0, values = (var_44227_cast_fp16, var_44229_cast_fp16, var_44231_cast_fp16, var_44233_cast_fp16))[name = string("op_44381_cast_fp16")]; + bool var_44383_interleave_0 = const()[name = string("op_44383_interleave_0"), val = bool(false)]; + tensor var_44383_cast_fp16 = concat(axis = var_42938, interleave = var_44383_interleave_0, values = (var_44235_cast_fp16, var_44237_cast_fp16, var_44239_cast_fp16, var_44241_cast_fp16))[name = string("op_44383_cast_fp16")]; + bool var_44385_interleave_0 = const()[name = string("op_44385_interleave_0"), val = bool(false)]; + tensor var_44385_cast_fp16 = concat(axis = var_42938, interleave = var_44385_interleave_0, values = (var_44243_cast_fp16, var_44245_cast_fp16, var_44247_cast_fp16, var_44249_cast_fp16))[name = string("op_44385_cast_fp16")]; + bool var_44387_interleave_0 = const()[name = string("op_44387_interleave_0"), val = bool(false)]; + tensor var_44387_cast_fp16 = concat(axis = var_42938, interleave = var_44387_interleave_0, values = (var_44251_cast_fp16, var_44253_cast_fp16, var_44255_cast_fp16, var_44257_cast_fp16))[name = string("op_44387_cast_fp16")]; + bool var_44389_interleave_0 = const()[name = string("op_44389_interleave_0"), val = bool(false)]; + tensor var_44389_cast_fp16 = concat(axis = var_42938, interleave = var_44389_interleave_0, values = (var_44259_cast_fp16, var_44261_cast_fp16, var_44263_cast_fp16, var_44265_cast_fp16))[name = string("op_44389_cast_fp16")]; + bool var_44391_interleave_0 = const()[name = string("op_44391_interleave_0"), val = bool(false)]; + tensor var_44391_cast_fp16 = concat(axis = var_42938, interleave = var_44391_interleave_0, values = (var_44267_cast_fp16, var_44269_cast_fp16, var_44271_cast_fp16, var_44273_cast_fp16))[name = string("op_44391_cast_fp16")]; + bool var_44393_interleave_0 = const()[name = string("op_44393_interleave_0"), val = bool(false)]; + tensor var_44393_cast_fp16 = concat(axis = var_42938, interleave = var_44393_interleave_0, values = (var_44275_cast_fp16, var_44277_cast_fp16, var_44279_cast_fp16, var_44281_cast_fp16))[name = string("op_44393_cast_fp16")]; + bool var_44395_interleave_0 = const()[name = string("op_44395_interleave_0"), val = bool(false)]; + tensor var_44395_cast_fp16 = concat(axis = var_42938, interleave = var_44395_interleave_0, values = (var_44283_cast_fp16, var_44285_cast_fp16, var_44287_cast_fp16, var_44289_cast_fp16))[name = string("op_44395_cast_fp16")]; + bool var_44397_interleave_0 = const()[name = string("op_44397_interleave_0"), val = bool(false)]; + tensor var_44397_cast_fp16 = concat(axis = var_42938, interleave = var_44397_interleave_0, values = (var_44291_cast_fp16, var_44293_cast_fp16, var_44295_cast_fp16, var_44297_cast_fp16))[name = string("op_44397_cast_fp16")]; + bool var_44399_interleave_0 = const()[name = string("op_44399_interleave_0"), val = bool(false)]; + tensor var_44399_cast_fp16 = concat(axis = var_42938, interleave = var_44399_interleave_0, values = (var_44299_cast_fp16, var_44301_cast_fp16, var_44303_cast_fp16, var_44305_cast_fp16))[name = string("op_44399_cast_fp16")]; + bool var_44401_interleave_0 = const()[name = string("op_44401_interleave_0"), val = bool(false)]; + tensor var_44401_cast_fp16 = concat(axis = var_42938, interleave = var_44401_interleave_0, values = (var_44307_cast_fp16, var_44309_cast_fp16, var_44311_cast_fp16, var_44313_cast_fp16))[name = string("op_44401_cast_fp16")]; + bool var_44403_interleave_0 = const()[name = string("op_44403_interleave_0"), val = bool(false)]; + tensor var_44403_cast_fp16 = concat(axis = var_42938, interleave = var_44403_interleave_0, values = (var_44315_cast_fp16, var_44317_cast_fp16, var_44319_cast_fp16, var_44321_cast_fp16))[name = string("op_44403_cast_fp16")]; + bool var_44405_interleave_0 = const()[name = string("op_44405_interleave_0"), val = bool(false)]; + tensor var_44405_cast_fp16 = concat(axis = var_42938, interleave = var_44405_interleave_0, values = (var_44323_cast_fp16, var_44325_cast_fp16, var_44327_cast_fp16, var_44329_cast_fp16))[name = string("op_44405_cast_fp16")]; + bool var_44407_interleave_0 = const()[name = string("op_44407_interleave_0"), val = bool(false)]; + tensor var_44407_cast_fp16 = concat(axis = var_42938, interleave = var_44407_interleave_0, values = (var_44331_cast_fp16, var_44333_cast_fp16, var_44335_cast_fp16, var_44337_cast_fp16))[name = string("op_44407_cast_fp16")]; + bool var_44409_interleave_0 = const()[name = string("op_44409_interleave_0"), val = bool(false)]; + tensor var_44409_cast_fp16 = concat(axis = var_42938, interleave = var_44409_interleave_0, values = (var_44339_cast_fp16, var_44341_cast_fp16, var_44343_cast_fp16, var_44345_cast_fp16))[name = string("op_44409_cast_fp16")]; + bool var_44411_interleave_0 = const()[name = string("op_44411_interleave_0"), val = bool(false)]; + tensor var_44411_cast_fp16 = concat(axis = var_42938, interleave = var_44411_interleave_0, values = (var_44347_cast_fp16, var_44349_cast_fp16, var_44351_cast_fp16, var_44353_cast_fp16))[name = string("op_44411_cast_fp16")]; + bool var_44413_interleave_0 = const()[name = string("op_44413_interleave_0"), val = bool(false)]; + tensor var_44413_cast_fp16 = concat(axis = var_42938, interleave = var_44413_interleave_0, values = (var_44355_cast_fp16, var_44357_cast_fp16, var_44359_cast_fp16, var_44361_cast_fp16))[name = string("op_44413_cast_fp16")]; + bool var_44415_interleave_0 = const()[name = string("op_44415_interleave_0"), val = bool(false)]; + tensor var_44415_cast_fp16 = concat(axis = var_42938, interleave = var_44415_interleave_0, values = (var_44363_cast_fp16, var_44365_cast_fp16, var_44367_cast_fp16, var_44369_cast_fp16))[name = string("op_44415_cast_fp16")]; + bool var_44417_interleave_0 = const()[name = string("op_44417_interleave_0"), val = bool(false)]; + tensor var_44417_cast_fp16 = concat(axis = var_42938, interleave = var_44417_interleave_0, values = (var_44371_cast_fp16, var_44373_cast_fp16, var_44375_cast_fp16, var_44377_cast_fp16))[name = string("op_44417_cast_fp16")]; + bool input_225_interleave_0 = const()[name = string("input_225_interleave_0"), val = bool(false)]; + tensor input_225_cast_fp16 = concat(axis = var_42963, interleave = input_225_interleave_0, values = (var_44379_cast_fp16, var_44381_cast_fp16, var_44383_cast_fp16, var_44385_cast_fp16, var_44387_cast_fp16, var_44389_cast_fp16, var_44391_cast_fp16, var_44393_cast_fp16, var_44395_cast_fp16, var_44397_cast_fp16, var_44399_cast_fp16, var_44401_cast_fp16, var_44403_cast_fp16, var_44405_cast_fp16, var_44407_cast_fp16, var_44409_cast_fp16, var_44411_cast_fp16, var_44413_cast_fp16, var_44415_cast_fp16, var_44417_cast_fp16))[name = string("input_225_cast_fp16")]; + string obj_115_pad_type_0 = const()[name = string("obj_115_pad_type_0"), val = string("valid")]; + tensor obj_115_strides_0 = const()[name = string("obj_115_strides_0"), val = tensor([1, 1])]; + tensor obj_115_pad_0 = const()[name = string("obj_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_115_dilations_0 = const()[name = string("obj_115_dilations_0"), val = tensor([1, 1])]; + int32 obj_115_groups_0 = const()[name = string("obj_115_groups_0"), val = int32(1)]; + tensor layers_28_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1126397120)))]; + tensor layers_28_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129673984)))]; + tensor obj_115_cast_fp16 = conv(bias = layers_28_self_attn_o_proj_bias_to_fp16, dilations = obj_115_dilations_0, groups = obj_115_groups_0, pad = obj_115_pad_0, pad_type = obj_115_pad_type_0, strides = obj_115_strides_0, weight = layers_28_self_attn_o_proj_weight_to_fp16, x = input_225_cast_fp16)[name = string("obj_115_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_115_cast_fp16)[name = string("inputs_115_cast_fp16")]; + tensor out_115_axes_0 = const()[name = string("out_115_axes_0"), val = tensor([1])]; + fp16 var_44436_to_fp16 = const()[name = string("op_44436_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_44436_to_fp16, x = inputs_115_cast_fp16)[name = string("out_115_cast_fp16")]; + tensor input_227_gamma_0_to_fp16 = const()[name = string("input_227_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129676608)))]; + tensor input_227_beta_0_to_fp16 = const()[name = string("input_227_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129679232)))]; + fp16 input_227_epsilon_0_to_fp16 = const()[name = string("input_227_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_227_cast_fp16 = batch_norm(beta = input_227_beta_0_to_fp16, epsilon = input_227_epsilon_0_to_fp16, gamma = input_227_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = string("input_227_cast_fp16")]; + string input_229_pad_type_0 = const()[name = string("input_229_pad_type_0"), val = string("valid")]; + tensor input_229_strides_0 = const()[name = string("input_229_strides_0"), val = tensor([1, 1])]; + tensor input_229_pad_0 = const()[name = string("input_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_229_dilations_0 = const()[name = string("input_229_dilations_0"), val = tensor([1, 1])]; + int32 input_229_groups_0 = const()[name = string("input_229_groups_0"), val = int32(1)]; + tensor layers_28_fc1_weight_to_fp16 = const()[name = string("layers_28_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1129681856)))]; + tensor layers_28_fc1_bias_to_fp16 = const()[name = string("layers_28_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142789120)))]; + tensor input_229_cast_fp16 = conv(bias = layers_28_fc1_bias_to_fp16, dilations = input_229_dilations_0, groups = input_229_groups_0, pad = input_229_pad_0, pad_type = input_229_pad_type_0, strides = input_229_strides_0, weight = layers_28_fc1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")]; + string input_231_mode_0 = const()[name = string("input_231_mode_0"), val = string("EXACT")]; + tensor input_231_cast_fp16 = gelu(mode = input_231_mode_0, x = input_229_cast_fp16)[name = string("input_231_cast_fp16")]; + string hidden_states_61_pad_type_0 = const()[name = string("hidden_states_61_pad_type_0"), val = string("valid")]; + tensor hidden_states_61_strides_0 = const()[name = string("hidden_states_61_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_61_pad_0 = const()[name = string("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_61_dilations_0 = const()[name = string("hidden_states_61_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_61_groups_0 = const()[name = string("hidden_states_61_groups_0"), val = int32(1)]; + tensor layers_28_fc2_weight_to_fp16 = const()[name = string("layers_28_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1142799424)))]; + tensor layers_28_fc2_bias_to_fp16 = const()[name = string("layers_28_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155906688)))]; + tensor hidden_states_61_cast_fp16 = conv(bias = layers_28_fc2_bias_to_fp16, dilations = hidden_states_61_dilations_0, groups = hidden_states_61_groups_0, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = hidden_states_61_strides_0, weight = layers_28_fc2_weight_to_fp16, x = input_231_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = hidden_states_61_cast_fp16)[name = string("inputs_117_cast_fp16")]; + int32 var_44465 = const()[name = string("op_44465"), val = int32(3)]; + int32 var_44490 = const()[name = string("op_44490"), val = int32(1)]; + tensor out_117_axes_0 = const()[name = string("out_117_axes_0"), val = tensor([1])]; + fp16 var_44507_to_fp16 = const()[name = string("op_44507_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_44507_to_fp16, x = inputs_117_cast_fp16)[name = string("out_117_cast_fp16")]; + tensor obj_117_gamma_0_to_fp16 = const()[name = string("obj_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155909312)))]; + tensor obj_117_beta_0_to_fp16 = const()[name = string("obj_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155911936)))]; + fp16 obj_117_epsilon_0_to_fp16 = const()[name = string("obj_117_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_117_cast_fp16 = batch_norm(beta = obj_117_beta_0_to_fp16, epsilon = obj_117_epsilon_0_to_fp16, gamma = obj_117_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = string("obj_117_cast_fp16")]; + string query_59_pad_type_0 = const()[name = string("query_59_pad_type_0"), val = string("valid")]; + tensor query_59_strides_0 = const()[name = string("query_59_strides_0"), val = tensor([1, 1])]; + tensor query_59_pad_0 = const()[name = string("query_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_59_dilations_0 = const()[name = string("query_59_dilations_0"), val = tensor([1, 1])]; + int32 query_59_groups_0 = const()[name = string("query_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1155914560)))]; + tensor layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159191424)))]; + tensor query_59_cast_fp16 = conv(bias = layers_29_self_attn_q_proj_bias_to_fp16, dilations = query_59_dilations_0, groups = query_59_groups_0, pad = query_59_pad_0, pad_type = query_59_pad_type_0, strides = query_59_strides_0, weight = layers_29_self_attn_q_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("query_59_cast_fp16")]; + string key_59_pad_type_0 = const()[name = string("key_59_pad_type_0"), val = string("valid")]; + tensor key_59_strides_0 = const()[name = string("key_59_strides_0"), val = tensor([1, 1])]; + tensor key_59_pad_0 = const()[name = string("key_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_59_dilations_0 = const()[name = string("key_59_dilations_0"), val = tensor([1, 1])]; + int32 key_59_groups_0 = const()[name = string("key_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1159194048)))]; + tensor key_59_cast_fp16 = conv(dilations = key_59_dilations_0, groups = key_59_groups_0, pad = key_59_pad_0, pad_type = key_59_pad_type_0, strides = key_59_strides_0, weight = layers_29_self_attn_k_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("key_59_cast_fp16")]; + string value_59_pad_type_0 = const()[name = string("value_59_pad_type_0"), val = string("valid")]; + tensor value_59_strides_0 = const()[name = string("value_59_strides_0"), val = tensor([1, 1])]; + tensor value_59_pad_0 = const()[name = string("value_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_59_dilations_0 = const()[name = string("value_59_dilations_0"), val = tensor([1, 1])]; + int32 value_59_groups_0 = const()[name = string("value_59_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1162470912)))]; + tensor layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165747776)))]; + tensor value_59_cast_fp16 = conv(bias = layers_29_self_attn_v_proj_bias_to_fp16, dilations = value_59_dilations_0, groups = value_59_groups_0, pad = value_59_pad_0, pad_type = value_59_pad_type_0, strides = value_59_strides_0, weight = layers_29_self_attn_v_proj_weight_to_fp16, x = obj_117_cast_fp16)[name = string("value_59_cast_fp16")]; + tensor var_44545_begin_0 = const()[name = string("op_44545_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44545_end_0 = const()[name = string("op_44545_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44545_end_mask_0 = const()[name = string("op_44545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44545_cast_fp16 = slice_by_index(begin = var_44545_begin_0, end = var_44545_end_0, end_mask = var_44545_end_mask_0, x = query_59_cast_fp16)[name = string("op_44545_cast_fp16")]; + tensor var_44549_begin_0 = const()[name = string("op_44549_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44549_end_0 = const()[name = string("op_44549_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_44549_end_mask_0 = const()[name = string("op_44549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44549_cast_fp16 = slice_by_index(begin = var_44549_begin_0, end = var_44549_end_0, end_mask = var_44549_end_mask_0, x = query_59_cast_fp16)[name = string("op_44549_cast_fp16")]; + tensor var_44553_begin_0 = const()[name = string("op_44553_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44553_end_0 = const()[name = string("op_44553_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_44553_end_mask_0 = const()[name = string("op_44553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44553_cast_fp16 = slice_by_index(begin = var_44553_begin_0, end = var_44553_end_0, end_mask = var_44553_end_mask_0, x = query_59_cast_fp16)[name = string("op_44553_cast_fp16")]; + tensor var_44557_begin_0 = const()[name = string("op_44557_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44557_end_0 = const()[name = string("op_44557_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_44557_end_mask_0 = const()[name = string("op_44557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44557_cast_fp16 = slice_by_index(begin = var_44557_begin_0, end = var_44557_end_0, end_mask = var_44557_end_mask_0, x = query_59_cast_fp16)[name = string("op_44557_cast_fp16")]; + tensor var_44561_begin_0 = const()[name = string("op_44561_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44561_end_0 = const()[name = string("op_44561_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_44561_end_mask_0 = const()[name = string("op_44561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44561_cast_fp16 = slice_by_index(begin = var_44561_begin_0, end = var_44561_end_0, end_mask = var_44561_end_mask_0, x = query_59_cast_fp16)[name = string("op_44561_cast_fp16")]; + tensor var_44565_begin_0 = const()[name = string("op_44565_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44565_end_0 = const()[name = string("op_44565_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_44565_end_mask_0 = const()[name = string("op_44565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44565_cast_fp16 = slice_by_index(begin = var_44565_begin_0, end = var_44565_end_0, end_mask = var_44565_end_mask_0, x = query_59_cast_fp16)[name = string("op_44565_cast_fp16")]; + tensor var_44569_begin_0 = const()[name = string("op_44569_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44569_end_0 = const()[name = string("op_44569_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_44569_end_mask_0 = const()[name = string("op_44569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44569_cast_fp16 = slice_by_index(begin = var_44569_begin_0, end = var_44569_end_0, end_mask = var_44569_end_mask_0, x = query_59_cast_fp16)[name = string("op_44569_cast_fp16")]; + tensor var_44573_begin_0 = const()[name = string("op_44573_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44573_end_0 = const()[name = string("op_44573_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_44573_end_mask_0 = const()[name = string("op_44573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44573_cast_fp16 = slice_by_index(begin = var_44573_begin_0, end = var_44573_end_0, end_mask = var_44573_end_mask_0, x = query_59_cast_fp16)[name = string("op_44573_cast_fp16")]; + tensor var_44577_begin_0 = const()[name = string("op_44577_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44577_end_0 = const()[name = string("op_44577_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_44577_end_mask_0 = const()[name = string("op_44577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44577_cast_fp16 = slice_by_index(begin = var_44577_begin_0, end = var_44577_end_0, end_mask = var_44577_end_mask_0, x = query_59_cast_fp16)[name = string("op_44577_cast_fp16")]; + tensor var_44581_begin_0 = const()[name = string("op_44581_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44581_end_0 = const()[name = string("op_44581_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_44581_end_mask_0 = const()[name = string("op_44581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44581_cast_fp16 = slice_by_index(begin = var_44581_begin_0, end = var_44581_end_0, end_mask = var_44581_end_mask_0, x = query_59_cast_fp16)[name = string("op_44581_cast_fp16")]; + tensor var_44585_begin_0 = const()[name = string("op_44585_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44585_end_0 = const()[name = string("op_44585_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_44585_end_mask_0 = const()[name = string("op_44585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44585_cast_fp16 = slice_by_index(begin = var_44585_begin_0, end = var_44585_end_0, end_mask = var_44585_end_mask_0, x = query_59_cast_fp16)[name = string("op_44585_cast_fp16")]; + tensor var_44589_begin_0 = const()[name = string("op_44589_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44589_end_0 = const()[name = string("op_44589_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_44589_end_mask_0 = const()[name = string("op_44589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44589_cast_fp16 = slice_by_index(begin = var_44589_begin_0, end = var_44589_end_0, end_mask = var_44589_end_mask_0, x = query_59_cast_fp16)[name = string("op_44589_cast_fp16")]; + tensor var_44593_begin_0 = const()[name = string("op_44593_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44593_end_0 = const()[name = string("op_44593_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_44593_end_mask_0 = const()[name = string("op_44593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44593_cast_fp16 = slice_by_index(begin = var_44593_begin_0, end = var_44593_end_0, end_mask = var_44593_end_mask_0, x = query_59_cast_fp16)[name = string("op_44593_cast_fp16")]; + tensor var_44597_begin_0 = const()[name = string("op_44597_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44597_end_0 = const()[name = string("op_44597_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_44597_end_mask_0 = const()[name = string("op_44597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44597_cast_fp16 = slice_by_index(begin = var_44597_begin_0, end = var_44597_end_0, end_mask = var_44597_end_mask_0, x = query_59_cast_fp16)[name = string("op_44597_cast_fp16")]; + tensor var_44601_begin_0 = const()[name = string("op_44601_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44601_end_0 = const()[name = string("op_44601_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_44601_end_mask_0 = const()[name = string("op_44601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44601_cast_fp16 = slice_by_index(begin = var_44601_begin_0, end = var_44601_end_0, end_mask = var_44601_end_mask_0, x = query_59_cast_fp16)[name = string("op_44601_cast_fp16")]; + tensor var_44605_begin_0 = const()[name = string("op_44605_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44605_end_0 = const()[name = string("op_44605_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_44605_end_mask_0 = const()[name = string("op_44605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44605_cast_fp16 = slice_by_index(begin = var_44605_begin_0, end = var_44605_end_0, end_mask = var_44605_end_mask_0, x = query_59_cast_fp16)[name = string("op_44605_cast_fp16")]; + tensor var_44609_begin_0 = const()[name = string("op_44609_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44609_end_0 = const()[name = string("op_44609_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_44609_end_mask_0 = const()[name = string("op_44609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44609_cast_fp16 = slice_by_index(begin = var_44609_begin_0, end = var_44609_end_0, end_mask = var_44609_end_mask_0, x = query_59_cast_fp16)[name = string("op_44609_cast_fp16")]; + tensor var_44613_begin_0 = const()[name = string("op_44613_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44613_end_0 = const()[name = string("op_44613_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_44613_end_mask_0 = const()[name = string("op_44613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44613_cast_fp16 = slice_by_index(begin = var_44613_begin_0, end = var_44613_end_0, end_mask = var_44613_end_mask_0, x = query_59_cast_fp16)[name = string("op_44613_cast_fp16")]; + tensor var_44617_begin_0 = const()[name = string("op_44617_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44617_end_0 = const()[name = string("op_44617_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_44617_end_mask_0 = const()[name = string("op_44617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44617_cast_fp16 = slice_by_index(begin = var_44617_begin_0, end = var_44617_end_0, end_mask = var_44617_end_mask_0, x = query_59_cast_fp16)[name = string("op_44617_cast_fp16")]; + tensor var_44621_begin_0 = const()[name = string("op_44621_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44621_end_0 = const()[name = string("op_44621_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_44621_end_mask_0 = const()[name = string("op_44621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44621_cast_fp16 = slice_by_index(begin = var_44621_begin_0, end = var_44621_end_0, end_mask = var_44621_end_mask_0, x = query_59_cast_fp16)[name = string("op_44621_cast_fp16")]; + tensor var_44630_begin_0 = const()[name = string("op_44630_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44630_end_0 = const()[name = string("op_44630_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44630_end_mask_0 = const()[name = string("op_44630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44630_cast_fp16 = slice_by_index(begin = var_44630_begin_0, end = var_44630_end_0, end_mask = var_44630_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44630_cast_fp16")]; + tensor var_44637_begin_0 = const()[name = string("op_44637_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44637_end_0 = const()[name = string("op_44637_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44637_end_mask_0 = const()[name = string("op_44637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44637_cast_fp16 = slice_by_index(begin = var_44637_begin_0, end = var_44637_end_0, end_mask = var_44637_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44637_cast_fp16")]; + tensor var_44644_begin_0 = const()[name = string("op_44644_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44644_end_0 = const()[name = string("op_44644_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44644_end_mask_0 = const()[name = string("op_44644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44644_cast_fp16 = slice_by_index(begin = var_44644_begin_0, end = var_44644_end_0, end_mask = var_44644_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44644_cast_fp16")]; + tensor var_44651_begin_0 = const()[name = string("op_44651_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44651_end_0 = const()[name = string("op_44651_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44651_end_mask_0 = const()[name = string("op_44651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44651_cast_fp16 = slice_by_index(begin = var_44651_begin_0, end = var_44651_end_0, end_mask = var_44651_end_mask_0, x = var_44545_cast_fp16)[name = string("op_44651_cast_fp16")]; + tensor var_44658_begin_0 = const()[name = string("op_44658_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44658_end_0 = const()[name = string("op_44658_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44658_end_mask_0 = const()[name = string("op_44658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44658_cast_fp16 = slice_by_index(begin = var_44658_begin_0, end = var_44658_end_0, end_mask = var_44658_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44658_cast_fp16")]; + tensor var_44665_begin_0 = const()[name = string("op_44665_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44665_end_0 = const()[name = string("op_44665_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44665_end_mask_0 = const()[name = string("op_44665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44665_cast_fp16 = slice_by_index(begin = var_44665_begin_0, end = var_44665_end_0, end_mask = var_44665_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44665_cast_fp16")]; + tensor var_44672_begin_0 = const()[name = string("op_44672_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44672_end_0 = const()[name = string("op_44672_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44672_end_mask_0 = const()[name = string("op_44672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44672_cast_fp16 = slice_by_index(begin = var_44672_begin_0, end = var_44672_end_0, end_mask = var_44672_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44672_cast_fp16")]; + tensor var_44679_begin_0 = const()[name = string("op_44679_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44679_end_0 = const()[name = string("op_44679_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44679_end_mask_0 = const()[name = string("op_44679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44679_cast_fp16 = slice_by_index(begin = var_44679_begin_0, end = var_44679_end_0, end_mask = var_44679_end_mask_0, x = var_44549_cast_fp16)[name = string("op_44679_cast_fp16")]; + tensor var_44686_begin_0 = const()[name = string("op_44686_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44686_end_0 = const()[name = string("op_44686_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44686_end_mask_0 = const()[name = string("op_44686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44686_cast_fp16 = slice_by_index(begin = var_44686_begin_0, end = var_44686_end_0, end_mask = var_44686_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44686_cast_fp16")]; + tensor var_44693_begin_0 = const()[name = string("op_44693_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44693_end_0 = const()[name = string("op_44693_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44693_end_mask_0 = const()[name = string("op_44693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44693_cast_fp16 = slice_by_index(begin = var_44693_begin_0, end = var_44693_end_0, end_mask = var_44693_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44693_cast_fp16")]; + tensor var_44700_begin_0 = const()[name = string("op_44700_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44700_end_0 = const()[name = string("op_44700_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44700_end_mask_0 = const()[name = string("op_44700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44700_cast_fp16 = slice_by_index(begin = var_44700_begin_0, end = var_44700_end_0, end_mask = var_44700_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44700_cast_fp16")]; + tensor var_44707_begin_0 = const()[name = string("op_44707_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44707_end_0 = const()[name = string("op_44707_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44707_end_mask_0 = const()[name = string("op_44707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44707_cast_fp16 = slice_by_index(begin = var_44707_begin_0, end = var_44707_end_0, end_mask = var_44707_end_mask_0, x = var_44553_cast_fp16)[name = string("op_44707_cast_fp16")]; + tensor var_44714_begin_0 = const()[name = string("op_44714_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44714_end_0 = const()[name = string("op_44714_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44714_end_mask_0 = const()[name = string("op_44714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44714_cast_fp16 = slice_by_index(begin = var_44714_begin_0, end = var_44714_end_0, end_mask = var_44714_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44714_cast_fp16")]; + tensor var_44721_begin_0 = const()[name = string("op_44721_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44721_end_0 = const()[name = string("op_44721_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44721_end_mask_0 = const()[name = string("op_44721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44721_cast_fp16 = slice_by_index(begin = var_44721_begin_0, end = var_44721_end_0, end_mask = var_44721_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44721_cast_fp16")]; + tensor var_44728_begin_0 = const()[name = string("op_44728_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44728_end_0 = const()[name = string("op_44728_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44728_end_mask_0 = const()[name = string("op_44728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44728_cast_fp16 = slice_by_index(begin = var_44728_begin_0, end = var_44728_end_0, end_mask = var_44728_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44728_cast_fp16")]; + tensor var_44735_begin_0 = const()[name = string("op_44735_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44735_end_0 = const()[name = string("op_44735_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44735_end_mask_0 = const()[name = string("op_44735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44735_cast_fp16 = slice_by_index(begin = var_44735_begin_0, end = var_44735_end_0, end_mask = var_44735_end_mask_0, x = var_44557_cast_fp16)[name = string("op_44735_cast_fp16")]; + tensor var_44742_begin_0 = const()[name = string("op_44742_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44742_end_0 = const()[name = string("op_44742_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44742_end_mask_0 = const()[name = string("op_44742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44742_cast_fp16 = slice_by_index(begin = var_44742_begin_0, end = var_44742_end_0, end_mask = var_44742_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44742_cast_fp16")]; + tensor var_44749_begin_0 = const()[name = string("op_44749_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44749_end_0 = const()[name = string("op_44749_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44749_end_mask_0 = const()[name = string("op_44749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44749_cast_fp16 = slice_by_index(begin = var_44749_begin_0, end = var_44749_end_0, end_mask = var_44749_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44749_cast_fp16")]; + tensor var_44756_begin_0 = const()[name = string("op_44756_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44756_end_0 = const()[name = string("op_44756_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44756_end_mask_0 = const()[name = string("op_44756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44756_cast_fp16 = slice_by_index(begin = var_44756_begin_0, end = var_44756_end_0, end_mask = var_44756_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44756_cast_fp16")]; + tensor var_44763_begin_0 = const()[name = string("op_44763_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44763_end_0 = const()[name = string("op_44763_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44763_end_mask_0 = const()[name = string("op_44763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44763_cast_fp16 = slice_by_index(begin = var_44763_begin_0, end = var_44763_end_0, end_mask = var_44763_end_mask_0, x = var_44561_cast_fp16)[name = string("op_44763_cast_fp16")]; + tensor var_44770_begin_0 = const()[name = string("op_44770_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44770_end_0 = const()[name = string("op_44770_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44770_end_mask_0 = const()[name = string("op_44770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44770_cast_fp16 = slice_by_index(begin = var_44770_begin_0, end = var_44770_end_0, end_mask = var_44770_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44770_cast_fp16")]; + tensor var_44777_begin_0 = const()[name = string("op_44777_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44777_end_0 = const()[name = string("op_44777_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44777_end_mask_0 = const()[name = string("op_44777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44777_cast_fp16 = slice_by_index(begin = var_44777_begin_0, end = var_44777_end_0, end_mask = var_44777_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44777_cast_fp16")]; + tensor var_44784_begin_0 = const()[name = string("op_44784_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44784_end_0 = const()[name = string("op_44784_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44784_end_mask_0 = const()[name = string("op_44784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44784_cast_fp16 = slice_by_index(begin = var_44784_begin_0, end = var_44784_end_0, end_mask = var_44784_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44784_cast_fp16")]; + tensor var_44791_begin_0 = const()[name = string("op_44791_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44791_end_0 = const()[name = string("op_44791_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44791_end_mask_0 = const()[name = string("op_44791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44791_cast_fp16 = slice_by_index(begin = var_44791_begin_0, end = var_44791_end_0, end_mask = var_44791_end_mask_0, x = var_44565_cast_fp16)[name = string("op_44791_cast_fp16")]; + tensor var_44798_begin_0 = const()[name = string("op_44798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44798_end_0 = const()[name = string("op_44798_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44798_end_mask_0 = const()[name = string("op_44798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44798_cast_fp16 = slice_by_index(begin = var_44798_begin_0, end = var_44798_end_0, end_mask = var_44798_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44798_cast_fp16")]; + tensor var_44805_begin_0 = const()[name = string("op_44805_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44805_end_0 = const()[name = string("op_44805_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44805_end_mask_0 = const()[name = string("op_44805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44805_cast_fp16 = slice_by_index(begin = var_44805_begin_0, end = var_44805_end_0, end_mask = var_44805_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44805_cast_fp16")]; + tensor var_44812_begin_0 = const()[name = string("op_44812_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44812_end_0 = const()[name = string("op_44812_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44812_end_mask_0 = const()[name = string("op_44812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44812_cast_fp16 = slice_by_index(begin = var_44812_begin_0, end = var_44812_end_0, end_mask = var_44812_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44812_cast_fp16")]; + tensor var_44819_begin_0 = const()[name = string("op_44819_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44819_end_0 = const()[name = string("op_44819_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44819_end_mask_0 = const()[name = string("op_44819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44819_cast_fp16 = slice_by_index(begin = var_44819_begin_0, end = var_44819_end_0, end_mask = var_44819_end_mask_0, x = var_44569_cast_fp16)[name = string("op_44819_cast_fp16")]; + tensor var_44826_begin_0 = const()[name = string("op_44826_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44826_end_0 = const()[name = string("op_44826_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44826_end_mask_0 = const()[name = string("op_44826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44826_cast_fp16 = slice_by_index(begin = var_44826_begin_0, end = var_44826_end_0, end_mask = var_44826_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44826_cast_fp16")]; + tensor var_44833_begin_0 = const()[name = string("op_44833_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44833_end_0 = const()[name = string("op_44833_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44833_end_mask_0 = const()[name = string("op_44833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44833_cast_fp16 = slice_by_index(begin = var_44833_begin_0, end = var_44833_end_0, end_mask = var_44833_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44833_cast_fp16")]; + tensor var_44840_begin_0 = const()[name = string("op_44840_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44840_end_0 = const()[name = string("op_44840_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44840_end_mask_0 = const()[name = string("op_44840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44840_cast_fp16 = slice_by_index(begin = var_44840_begin_0, end = var_44840_end_0, end_mask = var_44840_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44840_cast_fp16")]; + tensor var_44847_begin_0 = const()[name = string("op_44847_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44847_end_0 = const()[name = string("op_44847_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44847_end_mask_0 = const()[name = string("op_44847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44847_cast_fp16 = slice_by_index(begin = var_44847_begin_0, end = var_44847_end_0, end_mask = var_44847_end_mask_0, x = var_44573_cast_fp16)[name = string("op_44847_cast_fp16")]; + tensor var_44854_begin_0 = const()[name = string("op_44854_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44854_end_0 = const()[name = string("op_44854_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44854_end_mask_0 = const()[name = string("op_44854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44854_cast_fp16 = slice_by_index(begin = var_44854_begin_0, end = var_44854_end_0, end_mask = var_44854_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44854_cast_fp16")]; + tensor var_44861_begin_0 = const()[name = string("op_44861_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44861_end_0 = const()[name = string("op_44861_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44861_end_mask_0 = const()[name = string("op_44861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44861_cast_fp16 = slice_by_index(begin = var_44861_begin_0, end = var_44861_end_0, end_mask = var_44861_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44861_cast_fp16")]; + tensor var_44868_begin_0 = const()[name = string("op_44868_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44868_end_0 = const()[name = string("op_44868_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44868_end_mask_0 = const()[name = string("op_44868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44868_cast_fp16 = slice_by_index(begin = var_44868_begin_0, end = var_44868_end_0, end_mask = var_44868_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44868_cast_fp16")]; + tensor var_44875_begin_0 = const()[name = string("op_44875_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44875_end_0 = const()[name = string("op_44875_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44875_end_mask_0 = const()[name = string("op_44875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44875_cast_fp16 = slice_by_index(begin = var_44875_begin_0, end = var_44875_end_0, end_mask = var_44875_end_mask_0, x = var_44577_cast_fp16)[name = string("op_44875_cast_fp16")]; + tensor var_44882_begin_0 = const()[name = string("op_44882_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44882_end_0 = const()[name = string("op_44882_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44882_end_mask_0 = const()[name = string("op_44882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44882_cast_fp16 = slice_by_index(begin = var_44882_begin_0, end = var_44882_end_0, end_mask = var_44882_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44882_cast_fp16")]; + tensor var_44889_begin_0 = const()[name = string("op_44889_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44889_end_0 = const()[name = string("op_44889_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44889_end_mask_0 = const()[name = string("op_44889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44889_cast_fp16 = slice_by_index(begin = var_44889_begin_0, end = var_44889_end_0, end_mask = var_44889_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44889_cast_fp16")]; + tensor var_44896_begin_0 = const()[name = string("op_44896_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44896_end_0 = const()[name = string("op_44896_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44896_end_mask_0 = const()[name = string("op_44896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44896_cast_fp16 = slice_by_index(begin = var_44896_begin_0, end = var_44896_end_0, end_mask = var_44896_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44896_cast_fp16")]; + tensor var_44903_begin_0 = const()[name = string("op_44903_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44903_end_0 = const()[name = string("op_44903_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44903_end_mask_0 = const()[name = string("op_44903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44903_cast_fp16 = slice_by_index(begin = var_44903_begin_0, end = var_44903_end_0, end_mask = var_44903_end_mask_0, x = var_44581_cast_fp16)[name = string("op_44903_cast_fp16")]; + tensor var_44910_begin_0 = const()[name = string("op_44910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44910_end_0 = const()[name = string("op_44910_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44910_end_mask_0 = const()[name = string("op_44910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44910_cast_fp16 = slice_by_index(begin = var_44910_begin_0, end = var_44910_end_0, end_mask = var_44910_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44910_cast_fp16")]; + tensor var_44917_begin_0 = const()[name = string("op_44917_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44917_end_0 = const()[name = string("op_44917_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44917_end_mask_0 = const()[name = string("op_44917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44917_cast_fp16 = slice_by_index(begin = var_44917_begin_0, end = var_44917_end_0, end_mask = var_44917_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44917_cast_fp16")]; + tensor var_44924_begin_0 = const()[name = string("op_44924_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44924_end_0 = const()[name = string("op_44924_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44924_end_mask_0 = const()[name = string("op_44924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44924_cast_fp16 = slice_by_index(begin = var_44924_begin_0, end = var_44924_end_0, end_mask = var_44924_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44924_cast_fp16")]; + tensor var_44931_begin_0 = const()[name = string("op_44931_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44931_end_0 = const()[name = string("op_44931_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44931_end_mask_0 = const()[name = string("op_44931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44931_cast_fp16 = slice_by_index(begin = var_44931_begin_0, end = var_44931_end_0, end_mask = var_44931_end_mask_0, x = var_44585_cast_fp16)[name = string("op_44931_cast_fp16")]; + tensor var_44938_begin_0 = const()[name = string("op_44938_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44938_end_0 = const()[name = string("op_44938_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44938_end_mask_0 = const()[name = string("op_44938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44938_cast_fp16 = slice_by_index(begin = var_44938_begin_0, end = var_44938_end_0, end_mask = var_44938_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44938_cast_fp16")]; + tensor var_44945_begin_0 = const()[name = string("op_44945_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44945_end_0 = const()[name = string("op_44945_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44945_end_mask_0 = const()[name = string("op_44945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44945_cast_fp16 = slice_by_index(begin = var_44945_begin_0, end = var_44945_end_0, end_mask = var_44945_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44945_cast_fp16")]; + tensor var_44952_begin_0 = const()[name = string("op_44952_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44952_end_0 = const()[name = string("op_44952_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44952_end_mask_0 = const()[name = string("op_44952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44952_cast_fp16 = slice_by_index(begin = var_44952_begin_0, end = var_44952_end_0, end_mask = var_44952_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44952_cast_fp16")]; + tensor var_44959_begin_0 = const()[name = string("op_44959_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44959_end_0 = const()[name = string("op_44959_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44959_end_mask_0 = const()[name = string("op_44959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44959_cast_fp16 = slice_by_index(begin = var_44959_begin_0, end = var_44959_end_0, end_mask = var_44959_end_mask_0, x = var_44589_cast_fp16)[name = string("op_44959_cast_fp16")]; + tensor var_44966_begin_0 = const()[name = string("op_44966_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44966_end_0 = const()[name = string("op_44966_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44966_end_mask_0 = const()[name = string("op_44966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44966_cast_fp16 = slice_by_index(begin = var_44966_begin_0, end = var_44966_end_0, end_mask = var_44966_end_mask_0, x = var_44593_cast_fp16)[name = string("op_44966_cast_fp16")]; + tensor var_44973_begin_0 = const()[name = string("op_44973_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44973_end_0 = const()[name = string("op_44973_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44973_end_mask_0 = const()[name = string("op_44973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44973_cast_fp16 = slice_by_index(begin = var_44973_begin_0, end = var_44973_end_0, end_mask = var_44973_end_mask_0, x = var_44593_cast_fp16)[name = string("op_44973_cast_fp16")]; + tensor var_44980_begin_0 = const()[name = string("op_44980_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44980_end_0 = const()[name = string("op_44980_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44980_end_mask_0 = const()[name = string("op_44980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44980_cast_fp16 = slice_by_index(begin = var_44980_begin_0, end = var_44980_end_0, end_mask = var_44980_end_mask_0, x = var_44593_cast_fp16)[name = string("op_44980_cast_fp16")]; + tensor var_44987_begin_0 = const()[name = string("op_44987_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44987_end_0 = const()[name = string("op_44987_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44987_end_mask_0 = const()[name = string("op_44987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44987_cast_fp16 = slice_by_index(begin = var_44987_begin_0, end = var_44987_end_0, end_mask = var_44987_end_mask_0, x = var_44593_cast_fp16)[name = string("op_44987_cast_fp16")]; + tensor var_44994_begin_0 = const()[name = string("op_44994_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44994_end_0 = const()[name = string("op_44994_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44994_end_mask_0 = const()[name = string("op_44994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44994_cast_fp16 = slice_by_index(begin = var_44994_begin_0, end = var_44994_end_0, end_mask = var_44994_end_mask_0, x = var_44597_cast_fp16)[name = string("op_44994_cast_fp16")]; + tensor var_45001_begin_0 = const()[name = string("op_45001_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45001_end_0 = const()[name = string("op_45001_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45001_end_mask_0 = const()[name = string("op_45001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45001_cast_fp16 = slice_by_index(begin = var_45001_begin_0, end = var_45001_end_0, end_mask = var_45001_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45001_cast_fp16")]; + tensor var_45008_begin_0 = const()[name = string("op_45008_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45008_end_0 = const()[name = string("op_45008_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45008_end_mask_0 = const()[name = string("op_45008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45008_cast_fp16 = slice_by_index(begin = var_45008_begin_0, end = var_45008_end_0, end_mask = var_45008_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45008_cast_fp16")]; + tensor var_45015_begin_0 = const()[name = string("op_45015_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45015_end_0 = const()[name = string("op_45015_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45015_end_mask_0 = const()[name = string("op_45015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45015_cast_fp16 = slice_by_index(begin = var_45015_begin_0, end = var_45015_end_0, end_mask = var_45015_end_mask_0, x = var_44597_cast_fp16)[name = string("op_45015_cast_fp16")]; + tensor var_45022_begin_0 = const()[name = string("op_45022_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45022_end_0 = const()[name = string("op_45022_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45022_end_mask_0 = const()[name = string("op_45022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45022_cast_fp16 = slice_by_index(begin = var_45022_begin_0, end = var_45022_end_0, end_mask = var_45022_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45022_cast_fp16")]; + tensor var_45029_begin_0 = const()[name = string("op_45029_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45029_end_0 = const()[name = string("op_45029_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45029_end_mask_0 = const()[name = string("op_45029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45029_cast_fp16 = slice_by_index(begin = var_45029_begin_0, end = var_45029_end_0, end_mask = var_45029_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45029_cast_fp16")]; + tensor var_45036_begin_0 = const()[name = string("op_45036_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45036_end_0 = const()[name = string("op_45036_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45036_end_mask_0 = const()[name = string("op_45036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45036_cast_fp16 = slice_by_index(begin = var_45036_begin_0, end = var_45036_end_0, end_mask = var_45036_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45036_cast_fp16")]; + tensor var_45043_begin_0 = const()[name = string("op_45043_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45043_end_0 = const()[name = string("op_45043_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45043_end_mask_0 = const()[name = string("op_45043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45043_cast_fp16 = slice_by_index(begin = var_45043_begin_0, end = var_45043_end_0, end_mask = var_45043_end_mask_0, x = var_44601_cast_fp16)[name = string("op_45043_cast_fp16")]; + tensor var_45050_begin_0 = const()[name = string("op_45050_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45050_end_0 = const()[name = string("op_45050_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45050_end_mask_0 = const()[name = string("op_45050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45050_cast_fp16 = slice_by_index(begin = var_45050_begin_0, end = var_45050_end_0, end_mask = var_45050_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45050_cast_fp16")]; + tensor var_45057_begin_0 = const()[name = string("op_45057_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45057_end_0 = const()[name = string("op_45057_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45057_end_mask_0 = const()[name = string("op_45057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45057_cast_fp16 = slice_by_index(begin = var_45057_begin_0, end = var_45057_end_0, end_mask = var_45057_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45057_cast_fp16")]; + tensor var_45064_begin_0 = const()[name = string("op_45064_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45064_end_0 = const()[name = string("op_45064_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45064_end_mask_0 = const()[name = string("op_45064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45064_cast_fp16 = slice_by_index(begin = var_45064_begin_0, end = var_45064_end_0, end_mask = var_45064_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45064_cast_fp16")]; + tensor var_45071_begin_0 = const()[name = string("op_45071_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45071_end_0 = const()[name = string("op_45071_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45071_end_mask_0 = const()[name = string("op_45071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45071_cast_fp16 = slice_by_index(begin = var_45071_begin_0, end = var_45071_end_0, end_mask = var_45071_end_mask_0, x = var_44605_cast_fp16)[name = string("op_45071_cast_fp16")]; + tensor var_45078_begin_0 = const()[name = string("op_45078_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45078_end_0 = const()[name = string("op_45078_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45078_end_mask_0 = const()[name = string("op_45078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45078_cast_fp16 = slice_by_index(begin = var_45078_begin_0, end = var_45078_end_0, end_mask = var_45078_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45078_cast_fp16")]; + tensor var_45085_begin_0 = const()[name = string("op_45085_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45085_end_0 = const()[name = string("op_45085_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45085_end_mask_0 = const()[name = string("op_45085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45085_cast_fp16 = slice_by_index(begin = var_45085_begin_0, end = var_45085_end_0, end_mask = var_45085_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45085_cast_fp16")]; + tensor var_45092_begin_0 = const()[name = string("op_45092_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45092_end_0 = const()[name = string("op_45092_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45092_end_mask_0 = const()[name = string("op_45092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45092_cast_fp16 = slice_by_index(begin = var_45092_begin_0, end = var_45092_end_0, end_mask = var_45092_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45092_cast_fp16")]; + tensor var_45099_begin_0 = const()[name = string("op_45099_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45099_end_0 = const()[name = string("op_45099_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45099_end_mask_0 = const()[name = string("op_45099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45099_cast_fp16 = slice_by_index(begin = var_45099_begin_0, end = var_45099_end_0, end_mask = var_45099_end_mask_0, x = var_44609_cast_fp16)[name = string("op_45099_cast_fp16")]; + tensor var_45106_begin_0 = const()[name = string("op_45106_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45106_end_0 = const()[name = string("op_45106_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45106_end_mask_0 = const()[name = string("op_45106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45106_cast_fp16 = slice_by_index(begin = var_45106_begin_0, end = var_45106_end_0, end_mask = var_45106_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45106_cast_fp16")]; + tensor var_45113_begin_0 = const()[name = string("op_45113_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45113_end_0 = const()[name = string("op_45113_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45113_end_mask_0 = const()[name = string("op_45113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45113_cast_fp16 = slice_by_index(begin = var_45113_begin_0, end = var_45113_end_0, end_mask = var_45113_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45113_cast_fp16")]; + tensor var_45120_begin_0 = const()[name = string("op_45120_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45120_end_0 = const()[name = string("op_45120_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45120_end_mask_0 = const()[name = string("op_45120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45120_cast_fp16 = slice_by_index(begin = var_45120_begin_0, end = var_45120_end_0, end_mask = var_45120_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45120_cast_fp16")]; + tensor var_45127_begin_0 = const()[name = string("op_45127_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45127_end_0 = const()[name = string("op_45127_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45127_end_mask_0 = const()[name = string("op_45127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45127_cast_fp16 = slice_by_index(begin = var_45127_begin_0, end = var_45127_end_0, end_mask = var_45127_end_mask_0, x = var_44613_cast_fp16)[name = string("op_45127_cast_fp16")]; + tensor var_45134_begin_0 = const()[name = string("op_45134_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45134_end_0 = const()[name = string("op_45134_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45134_end_mask_0 = const()[name = string("op_45134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45134_cast_fp16 = slice_by_index(begin = var_45134_begin_0, end = var_45134_end_0, end_mask = var_45134_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45134_cast_fp16")]; + tensor var_45141_begin_0 = const()[name = string("op_45141_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45141_end_0 = const()[name = string("op_45141_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45141_end_mask_0 = const()[name = string("op_45141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45141_cast_fp16 = slice_by_index(begin = var_45141_begin_0, end = var_45141_end_0, end_mask = var_45141_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45141_cast_fp16")]; + tensor var_45148_begin_0 = const()[name = string("op_45148_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45148_end_0 = const()[name = string("op_45148_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45148_end_mask_0 = const()[name = string("op_45148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45148_cast_fp16 = slice_by_index(begin = var_45148_begin_0, end = var_45148_end_0, end_mask = var_45148_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45148_cast_fp16")]; + tensor var_45155_begin_0 = const()[name = string("op_45155_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45155_end_0 = const()[name = string("op_45155_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45155_end_mask_0 = const()[name = string("op_45155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45155_cast_fp16 = slice_by_index(begin = var_45155_begin_0, end = var_45155_end_0, end_mask = var_45155_end_mask_0, x = var_44617_cast_fp16)[name = string("op_45155_cast_fp16")]; + tensor var_45162_begin_0 = const()[name = string("op_45162_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45162_end_0 = const()[name = string("op_45162_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45162_end_mask_0 = const()[name = string("op_45162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45162_cast_fp16 = slice_by_index(begin = var_45162_begin_0, end = var_45162_end_0, end_mask = var_45162_end_mask_0, x = var_44621_cast_fp16)[name = string("op_45162_cast_fp16")]; + tensor var_45169_begin_0 = const()[name = string("op_45169_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45169_end_0 = const()[name = string("op_45169_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45169_end_mask_0 = const()[name = string("op_45169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45169_cast_fp16 = slice_by_index(begin = var_45169_begin_0, end = var_45169_end_0, end_mask = var_45169_end_mask_0, x = var_44621_cast_fp16)[name = string("op_45169_cast_fp16")]; + tensor var_45176_begin_0 = const()[name = string("op_45176_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45176_end_0 = const()[name = string("op_45176_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45176_end_mask_0 = const()[name = string("op_45176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45176_cast_fp16 = slice_by_index(begin = var_45176_begin_0, end = var_45176_end_0, end_mask = var_45176_end_mask_0, x = var_44621_cast_fp16)[name = string("op_45176_cast_fp16")]; + tensor var_45183_begin_0 = const()[name = string("op_45183_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45183_end_0 = const()[name = string("op_45183_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45183_end_mask_0 = const()[name = string("op_45183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45183_cast_fp16 = slice_by_index(begin = var_45183_begin_0, end = var_45183_end_0, end_mask = var_45183_end_mask_0, x = var_44621_cast_fp16)[name = string("op_45183_cast_fp16")]; + tensor k_59_perm_0 = const()[name = string("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_45188_begin_0 = const()[name = string("op_45188_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45188_end_0 = const()[name = string("op_45188_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_45188_end_mask_0 = const()[name = string("op_45188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_59_cast_fp16 = transpose(perm = k_59_perm_0, x = key_59_cast_fp16)[name = string("transpose_2")]; + tensor var_45188_cast_fp16 = slice_by_index(begin = var_45188_begin_0, end = var_45188_end_0, end_mask = var_45188_end_mask_0, x = k_59_cast_fp16)[name = string("op_45188_cast_fp16")]; + tensor var_45192_begin_0 = const()[name = string("op_45192_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_45192_end_0 = const()[name = string("op_45192_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_45192_end_mask_0 = const()[name = string("op_45192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45192_cast_fp16 = slice_by_index(begin = var_45192_begin_0, end = var_45192_end_0, end_mask = var_45192_end_mask_0, x = k_59_cast_fp16)[name = string("op_45192_cast_fp16")]; + tensor var_45196_begin_0 = const()[name = string("op_45196_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_45196_end_0 = const()[name = string("op_45196_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_45196_end_mask_0 = const()[name = string("op_45196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45196_cast_fp16 = slice_by_index(begin = var_45196_begin_0, end = var_45196_end_0, end_mask = var_45196_end_mask_0, x = k_59_cast_fp16)[name = string("op_45196_cast_fp16")]; + tensor var_45200_begin_0 = const()[name = string("op_45200_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_45200_end_0 = const()[name = string("op_45200_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_45200_end_mask_0 = const()[name = string("op_45200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45200_cast_fp16 = slice_by_index(begin = var_45200_begin_0, end = var_45200_end_0, end_mask = var_45200_end_mask_0, x = k_59_cast_fp16)[name = string("op_45200_cast_fp16")]; + tensor var_45204_begin_0 = const()[name = string("op_45204_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_45204_end_0 = const()[name = string("op_45204_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_45204_end_mask_0 = const()[name = string("op_45204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45204_cast_fp16 = slice_by_index(begin = var_45204_begin_0, end = var_45204_end_0, end_mask = var_45204_end_mask_0, x = k_59_cast_fp16)[name = string("op_45204_cast_fp16")]; + tensor var_45208_begin_0 = const()[name = string("op_45208_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_45208_end_0 = const()[name = string("op_45208_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_45208_end_mask_0 = const()[name = string("op_45208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45208_cast_fp16 = slice_by_index(begin = var_45208_begin_0, end = var_45208_end_0, end_mask = var_45208_end_mask_0, x = k_59_cast_fp16)[name = string("op_45208_cast_fp16")]; + tensor var_45212_begin_0 = const()[name = string("op_45212_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_45212_end_0 = const()[name = string("op_45212_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_45212_end_mask_0 = const()[name = string("op_45212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45212_cast_fp16 = slice_by_index(begin = var_45212_begin_0, end = var_45212_end_0, end_mask = var_45212_end_mask_0, x = k_59_cast_fp16)[name = string("op_45212_cast_fp16")]; + tensor var_45216_begin_0 = const()[name = string("op_45216_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_45216_end_0 = const()[name = string("op_45216_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_45216_end_mask_0 = const()[name = string("op_45216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45216_cast_fp16 = slice_by_index(begin = var_45216_begin_0, end = var_45216_end_0, end_mask = var_45216_end_mask_0, x = k_59_cast_fp16)[name = string("op_45216_cast_fp16")]; + tensor var_45220_begin_0 = const()[name = string("op_45220_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_45220_end_0 = const()[name = string("op_45220_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_45220_end_mask_0 = const()[name = string("op_45220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45220_cast_fp16 = slice_by_index(begin = var_45220_begin_0, end = var_45220_end_0, end_mask = var_45220_end_mask_0, x = k_59_cast_fp16)[name = string("op_45220_cast_fp16")]; + tensor var_45224_begin_0 = const()[name = string("op_45224_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_45224_end_0 = const()[name = string("op_45224_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_45224_end_mask_0 = const()[name = string("op_45224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45224_cast_fp16 = slice_by_index(begin = var_45224_begin_0, end = var_45224_end_0, end_mask = var_45224_end_mask_0, x = k_59_cast_fp16)[name = string("op_45224_cast_fp16")]; + tensor var_45228_begin_0 = const()[name = string("op_45228_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_45228_end_0 = const()[name = string("op_45228_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_45228_end_mask_0 = const()[name = string("op_45228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45228_cast_fp16 = slice_by_index(begin = var_45228_begin_0, end = var_45228_end_0, end_mask = var_45228_end_mask_0, x = k_59_cast_fp16)[name = string("op_45228_cast_fp16")]; + tensor var_45232_begin_0 = const()[name = string("op_45232_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_45232_end_0 = const()[name = string("op_45232_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_45232_end_mask_0 = const()[name = string("op_45232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45232_cast_fp16 = slice_by_index(begin = var_45232_begin_0, end = var_45232_end_0, end_mask = var_45232_end_mask_0, x = k_59_cast_fp16)[name = string("op_45232_cast_fp16")]; + tensor var_45236_begin_0 = const()[name = string("op_45236_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_45236_end_0 = const()[name = string("op_45236_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_45236_end_mask_0 = const()[name = string("op_45236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45236_cast_fp16 = slice_by_index(begin = var_45236_begin_0, end = var_45236_end_0, end_mask = var_45236_end_mask_0, x = k_59_cast_fp16)[name = string("op_45236_cast_fp16")]; + tensor var_45240_begin_0 = const()[name = string("op_45240_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_45240_end_0 = const()[name = string("op_45240_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_45240_end_mask_0 = const()[name = string("op_45240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45240_cast_fp16 = slice_by_index(begin = var_45240_begin_0, end = var_45240_end_0, end_mask = var_45240_end_mask_0, x = k_59_cast_fp16)[name = string("op_45240_cast_fp16")]; + tensor var_45244_begin_0 = const()[name = string("op_45244_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_45244_end_0 = const()[name = string("op_45244_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_45244_end_mask_0 = const()[name = string("op_45244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45244_cast_fp16 = slice_by_index(begin = var_45244_begin_0, end = var_45244_end_0, end_mask = var_45244_end_mask_0, x = k_59_cast_fp16)[name = string("op_45244_cast_fp16")]; + tensor var_45248_begin_0 = const()[name = string("op_45248_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_45248_end_0 = const()[name = string("op_45248_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_45248_end_mask_0 = const()[name = string("op_45248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45248_cast_fp16 = slice_by_index(begin = var_45248_begin_0, end = var_45248_end_0, end_mask = var_45248_end_mask_0, x = k_59_cast_fp16)[name = string("op_45248_cast_fp16")]; + tensor var_45252_begin_0 = const()[name = string("op_45252_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_45252_end_0 = const()[name = string("op_45252_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_45252_end_mask_0 = const()[name = string("op_45252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45252_cast_fp16 = slice_by_index(begin = var_45252_begin_0, end = var_45252_end_0, end_mask = var_45252_end_mask_0, x = k_59_cast_fp16)[name = string("op_45252_cast_fp16")]; + tensor var_45256_begin_0 = const()[name = string("op_45256_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_45256_end_0 = const()[name = string("op_45256_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_45256_end_mask_0 = const()[name = string("op_45256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45256_cast_fp16 = slice_by_index(begin = var_45256_begin_0, end = var_45256_end_0, end_mask = var_45256_end_mask_0, x = k_59_cast_fp16)[name = string("op_45256_cast_fp16")]; + tensor var_45260_begin_0 = const()[name = string("op_45260_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_45260_end_0 = const()[name = string("op_45260_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_45260_end_mask_0 = const()[name = string("op_45260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45260_cast_fp16 = slice_by_index(begin = var_45260_begin_0, end = var_45260_end_0, end_mask = var_45260_end_mask_0, x = k_59_cast_fp16)[name = string("op_45260_cast_fp16")]; + tensor var_45264_begin_0 = const()[name = string("op_45264_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_45264_end_0 = const()[name = string("op_45264_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_45264_end_mask_0 = const()[name = string("op_45264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45264_cast_fp16 = slice_by_index(begin = var_45264_begin_0, end = var_45264_end_0, end_mask = var_45264_end_mask_0, x = k_59_cast_fp16)[name = string("op_45264_cast_fp16")]; + tensor var_45266_begin_0 = const()[name = string("op_45266_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45266_end_0 = const()[name = string("op_45266_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45266_end_mask_0 = const()[name = string("op_45266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45266_cast_fp16 = slice_by_index(begin = var_45266_begin_0, end = var_45266_end_0, end_mask = var_45266_end_mask_0, x = value_59_cast_fp16)[name = string("op_45266_cast_fp16")]; + tensor var_45270_begin_0 = const()[name = string("op_45270_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45270_end_0 = const()[name = string("op_45270_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_45270_end_mask_0 = const()[name = string("op_45270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45270_cast_fp16 = slice_by_index(begin = var_45270_begin_0, end = var_45270_end_0, end_mask = var_45270_end_mask_0, x = value_59_cast_fp16)[name = string("op_45270_cast_fp16")]; + tensor var_45274_begin_0 = const()[name = string("op_45274_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45274_end_0 = const()[name = string("op_45274_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_45274_end_mask_0 = const()[name = string("op_45274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45274_cast_fp16 = slice_by_index(begin = var_45274_begin_0, end = var_45274_end_0, end_mask = var_45274_end_mask_0, x = value_59_cast_fp16)[name = string("op_45274_cast_fp16")]; + tensor var_45278_begin_0 = const()[name = string("op_45278_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45278_end_0 = const()[name = string("op_45278_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_45278_end_mask_0 = const()[name = string("op_45278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45278_cast_fp16 = slice_by_index(begin = var_45278_begin_0, end = var_45278_end_0, end_mask = var_45278_end_mask_0, x = value_59_cast_fp16)[name = string("op_45278_cast_fp16")]; + tensor var_45282_begin_0 = const()[name = string("op_45282_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45282_end_0 = const()[name = string("op_45282_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_45282_end_mask_0 = const()[name = string("op_45282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45282_cast_fp16 = slice_by_index(begin = var_45282_begin_0, end = var_45282_end_0, end_mask = var_45282_end_mask_0, x = value_59_cast_fp16)[name = string("op_45282_cast_fp16")]; + tensor var_45286_begin_0 = const()[name = string("op_45286_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45286_end_0 = const()[name = string("op_45286_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_45286_end_mask_0 = const()[name = string("op_45286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45286_cast_fp16 = slice_by_index(begin = var_45286_begin_0, end = var_45286_end_0, end_mask = var_45286_end_mask_0, x = value_59_cast_fp16)[name = string("op_45286_cast_fp16")]; + tensor var_45290_begin_0 = const()[name = string("op_45290_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45290_end_0 = const()[name = string("op_45290_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_45290_end_mask_0 = const()[name = string("op_45290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45290_cast_fp16 = slice_by_index(begin = var_45290_begin_0, end = var_45290_end_0, end_mask = var_45290_end_mask_0, x = value_59_cast_fp16)[name = string("op_45290_cast_fp16")]; + tensor var_45294_begin_0 = const()[name = string("op_45294_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45294_end_0 = const()[name = string("op_45294_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_45294_end_mask_0 = const()[name = string("op_45294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45294_cast_fp16 = slice_by_index(begin = var_45294_begin_0, end = var_45294_end_0, end_mask = var_45294_end_mask_0, x = value_59_cast_fp16)[name = string("op_45294_cast_fp16")]; + tensor var_45298_begin_0 = const()[name = string("op_45298_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45298_end_0 = const()[name = string("op_45298_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_45298_end_mask_0 = const()[name = string("op_45298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45298_cast_fp16 = slice_by_index(begin = var_45298_begin_0, end = var_45298_end_0, end_mask = var_45298_end_mask_0, x = value_59_cast_fp16)[name = string("op_45298_cast_fp16")]; + tensor var_45302_begin_0 = const()[name = string("op_45302_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45302_end_0 = const()[name = string("op_45302_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_45302_end_mask_0 = const()[name = string("op_45302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45302_cast_fp16 = slice_by_index(begin = var_45302_begin_0, end = var_45302_end_0, end_mask = var_45302_end_mask_0, x = value_59_cast_fp16)[name = string("op_45302_cast_fp16")]; + tensor var_45306_begin_0 = const()[name = string("op_45306_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45306_end_0 = const()[name = string("op_45306_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_45306_end_mask_0 = const()[name = string("op_45306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45306_cast_fp16 = slice_by_index(begin = var_45306_begin_0, end = var_45306_end_0, end_mask = var_45306_end_mask_0, x = value_59_cast_fp16)[name = string("op_45306_cast_fp16")]; + tensor var_45310_begin_0 = const()[name = string("op_45310_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45310_end_0 = const()[name = string("op_45310_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_45310_end_mask_0 = const()[name = string("op_45310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45310_cast_fp16 = slice_by_index(begin = var_45310_begin_0, end = var_45310_end_0, end_mask = var_45310_end_mask_0, x = value_59_cast_fp16)[name = string("op_45310_cast_fp16")]; + tensor var_45314_begin_0 = const()[name = string("op_45314_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45314_end_0 = const()[name = string("op_45314_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_45314_end_mask_0 = const()[name = string("op_45314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45314_cast_fp16 = slice_by_index(begin = var_45314_begin_0, end = var_45314_end_0, end_mask = var_45314_end_mask_0, x = value_59_cast_fp16)[name = string("op_45314_cast_fp16")]; + tensor var_45318_begin_0 = const()[name = string("op_45318_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45318_end_0 = const()[name = string("op_45318_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_45318_end_mask_0 = const()[name = string("op_45318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45318_cast_fp16 = slice_by_index(begin = var_45318_begin_0, end = var_45318_end_0, end_mask = var_45318_end_mask_0, x = value_59_cast_fp16)[name = string("op_45318_cast_fp16")]; + tensor var_45322_begin_0 = const()[name = string("op_45322_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45322_end_0 = const()[name = string("op_45322_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_45322_end_mask_0 = const()[name = string("op_45322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45322_cast_fp16 = slice_by_index(begin = var_45322_begin_0, end = var_45322_end_0, end_mask = var_45322_end_mask_0, x = value_59_cast_fp16)[name = string("op_45322_cast_fp16")]; + tensor var_45326_begin_0 = const()[name = string("op_45326_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45326_end_0 = const()[name = string("op_45326_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_45326_end_mask_0 = const()[name = string("op_45326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45326_cast_fp16 = slice_by_index(begin = var_45326_begin_0, end = var_45326_end_0, end_mask = var_45326_end_mask_0, x = value_59_cast_fp16)[name = string("op_45326_cast_fp16")]; + tensor var_45330_begin_0 = const()[name = string("op_45330_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45330_end_0 = const()[name = string("op_45330_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_45330_end_mask_0 = const()[name = string("op_45330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45330_cast_fp16 = slice_by_index(begin = var_45330_begin_0, end = var_45330_end_0, end_mask = var_45330_end_mask_0, x = value_59_cast_fp16)[name = string("op_45330_cast_fp16")]; + tensor var_45334_begin_0 = const()[name = string("op_45334_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45334_end_0 = const()[name = string("op_45334_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_45334_end_mask_0 = const()[name = string("op_45334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45334_cast_fp16 = slice_by_index(begin = var_45334_begin_0, end = var_45334_end_0, end_mask = var_45334_end_mask_0, x = value_59_cast_fp16)[name = string("op_45334_cast_fp16")]; + tensor var_45338_begin_0 = const()[name = string("op_45338_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45338_end_0 = const()[name = string("op_45338_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_45338_end_mask_0 = const()[name = string("op_45338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45338_cast_fp16 = slice_by_index(begin = var_45338_begin_0, end = var_45338_end_0, end_mask = var_45338_end_mask_0, x = value_59_cast_fp16)[name = string("op_45338_cast_fp16")]; + tensor var_45342_begin_0 = const()[name = string("op_45342_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45342_end_0 = const()[name = string("op_45342_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_45342_end_mask_0 = const()[name = string("op_45342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45342_cast_fp16 = slice_by_index(begin = var_45342_begin_0, end = var_45342_end_0, end_mask = var_45342_end_mask_0, x = value_59_cast_fp16)[name = string("op_45342_cast_fp16")]; + string _SplitHeadsQ__mh_w_4641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4641_equation_0, values = (var_45188_cast_fp16, var_44630_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4641_cast_fp16")]; + string _SplitHeadsQ__mh_w_4643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4643_equation_0, values = (var_45188_cast_fp16, var_44637_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4643_cast_fp16")]; + string _SplitHeadsQ__mh_w_4645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4645_equation_0, values = (var_45188_cast_fp16, var_44644_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4645_cast_fp16")]; + string _SplitHeadsQ__mh_w_4647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4647_equation_0, values = (var_45188_cast_fp16, var_44651_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4647_cast_fp16")]; + string _SplitHeadsQ__mh_w_4649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4649_equation_0, values = (var_45192_cast_fp16, var_44658_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4649_cast_fp16")]; + string _SplitHeadsQ__mh_w_4651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4651_equation_0, values = (var_45192_cast_fp16, var_44665_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4651_cast_fp16")]; + string _SplitHeadsQ__mh_w_4653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4653_equation_0, values = (var_45192_cast_fp16, var_44672_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4653_cast_fp16")]; + string _SplitHeadsQ__mh_w_4655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4655_equation_0, values = (var_45192_cast_fp16, var_44679_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4655_cast_fp16")]; + string _SplitHeadsQ__mh_w_4657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4657_equation_0, values = (var_45196_cast_fp16, var_44686_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4657_cast_fp16")]; + string _SplitHeadsQ__mh_w_4659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4659_equation_0, values = (var_45196_cast_fp16, var_44693_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4659_cast_fp16")]; + string _SplitHeadsQ__mh_w_4661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4661_equation_0, values = (var_45196_cast_fp16, var_44700_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4661_cast_fp16")]; + string _SplitHeadsQ__mh_w_4663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4663_equation_0, values = (var_45196_cast_fp16, var_44707_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4663_cast_fp16")]; + string _SplitHeadsQ__mh_w_4665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4665_equation_0, values = (var_45200_cast_fp16, var_44714_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4665_cast_fp16")]; + string _SplitHeadsQ__mh_w_4667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4667_equation_0, values = (var_45200_cast_fp16, var_44721_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4667_cast_fp16")]; + string _SplitHeadsQ__mh_w_4669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4669_equation_0, values = (var_45200_cast_fp16, var_44728_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4669_cast_fp16")]; + string _SplitHeadsQ__mh_w_4671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4671_equation_0, values = (var_45200_cast_fp16, var_44735_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4671_cast_fp16")]; + string _SplitHeadsQ__mh_w_4673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4673_equation_0, values = (var_45204_cast_fp16, var_44742_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4673_cast_fp16")]; + string _SplitHeadsQ__mh_w_4675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4675_equation_0, values = (var_45204_cast_fp16, var_44749_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4675_cast_fp16")]; + string _SplitHeadsQ__mh_w_4677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4677_equation_0, values = (var_45204_cast_fp16, var_44756_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4677_cast_fp16")]; + string _SplitHeadsQ__mh_w_4679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4679_equation_0, values = (var_45204_cast_fp16, var_44763_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4679_cast_fp16")]; + string _SplitHeadsQ__mh_w_4681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4681_equation_0, values = (var_45208_cast_fp16, var_44770_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4681_cast_fp16")]; + string _SplitHeadsQ__mh_w_4683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4683_equation_0, values = (var_45208_cast_fp16, var_44777_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4683_cast_fp16")]; + string _SplitHeadsQ__mh_w_4685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4685_equation_0, values = (var_45208_cast_fp16, var_44784_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4685_cast_fp16")]; + string _SplitHeadsQ__mh_w_4687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4687_equation_0, values = (var_45208_cast_fp16, var_44791_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4687_cast_fp16")]; + string _SplitHeadsQ__mh_w_4689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4689_equation_0, values = (var_45212_cast_fp16, var_44798_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4689_cast_fp16")]; + string _SplitHeadsQ__mh_w_4691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4691_equation_0, values = (var_45212_cast_fp16, var_44805_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4691_cast_fp16")]; + string _SplitHeadsQ__mh_w_4693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4693_equation_0, values = (var_45212_cast_fp16, var_44812_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4693_cast_fp16")]; + string _SplitHeadsQ__mh_w_4695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4695_equation_0, values = (var_45212_cast_fp16, var_44819_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4695_cast_fp16")]; + string _SplitHeadsQ__mh_w_4697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4697_equation_0, values = (var_45216_cast_fp16, var_44826_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4697_cast_fp16")]; + string _SplitHeadsQ__mh_w_4699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4699_equation_0, values = (var_45216_cast_fp16, var_44833_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4699_cast_fp16")]; + string _SplitHeadsQ__mh_w_4701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4701_equation_0, values = (var_45216_cast_fp16, var_44840_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4701_cast_fp16")]; + string _SplitHeadsQ__mh_w_4703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4703_equation_0, values = (var_45216_cast_fp16, var_44847_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4703_cast_fp16")]; + string _SplitHeadsQ__mh_w_4705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4705_equation_0, values = (var_45220_cast_fp16, var_44854_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4705_cast_fp16")]; + string _SplitHeadsQ__mh_w_4707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4707_equation_0, values = (var_45220_cast_fp16, var_44861_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4707_cast_fp16")]; + string _SplitHeadsQ__mh_w_4709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4709_equation_0, values = (var_45220_cast_fp16, var_44868_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4709_cast_fp16")]; + string _SplitHeadsQ__mh_w_4711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4711_equation_0, values = (var_45220_cast_fp16, var_44875_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4711_cast_fp16")]; + string _SplitHeadsQ__mh_w_4713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4713_equation_0, values = (var_45224_cast_fp16, var_44882_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4713_cast_fp16")]; + string _SplitHeadsQ__mh_w_4715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4715_equation_0, values = (var_45224_cast_fp16, var_44889_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4715_cast_fp16")]; + string _SplitHeadsQ__mh_w_4717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4717_equation_0, values = (var_45224_cast_fp16, var_44896_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4717_cast_fp16")]; + string _SplitHeadsQ__mh_w_4719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4719_equation_0, values = (var_45224_cast_fp16, var_44903_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4719_cast_fp16")]; + string _SplitHeadsQ__mh_w_4721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4721_equation_0, values = (var_45228_cast_fp16, var_44910_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4721_cast_fp16")]; + string _SplitHeadsQ__mh_w_4723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4723_equation_0, values = (var_45228_cast_fp16, var_44917_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4723_cast_fp16")]; + string _SplitHeadsQ__mh_w_4725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4725_equation_0, values = (var_45228_cast_fp16, var_44924_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4725_cast_fp16")]; + string _SplitHeadsQ__mh_w_4727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4727_equation_0, values = (var_45228_cast_fp16, var_44931_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4727_cast_fp16")]; + string _SplitHeadsQ__mh_w_4729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4729_equation_0, values = (var_45232_cast_fp16, var_44938_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4729_cast_fp16")]; + string _SplitHeadsQ__mh_w_4731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4731_equation_0, values = (var_45232_cast_fp16, var_44945_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4731_cast_fp16")]; + string _SplitHeadsQ__mh_w_4733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4733_equation_0, values = (var_45232_cast_fp16, var_44952_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4733_cast_fp16")]; + string _SplitHeadsQ__mh_w_4735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4735_equation_0, values = (var_45232_cast_fp16, var_44959_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4735_cast_fp16")]; + string _SplitHeadsQ__mh_w_4737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4737_equation_0, values = (var_45236_cast_fp16, var_44966_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4737_cast_fp16")]; + string _SplitHeadsQ__mh_w_4739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4739_equation_0, values = (var_45236_cast_fp16, var_44973_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4739_cast_fp16")]; + string _SplitHeadsQ__mh_w_4741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4741_equation_0, values = (var_45236_cast_fp16, var_44980_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4741_cast_fp16")]; + string _SplitHeadsQ__mh_w_4743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4743_equation_0, values = (var_45236_cast_fp16, var_44987_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4743_cast_fp16")]; + string _SplitHeadsQ__mh_w_4745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4745_equation_0, values = (var_45240_cast_fp16, var_44994_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4745_cast_fp16")]; + string _SplitHeadsQ__mh_w_4747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4747_equation_0, values = (var_45240_cast_fp16, var_45001_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4747_cast_fp16")]; + string _SplitHeadsQ__mh_w_4749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4749_equation_0, values = (var_45240_cast_fp16, var_45008_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4749_cast_fp16")]; + string _SplitHeadsQ__mh_w_4751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4751_equation_0, values = (var_45240_cast_fp16, var_45015_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4751_cast_fp16")]; + string _SplitHeadsQ__mh_w_4753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4753_equation_0, values = (var_45244_cast_fp16, var_45022_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4753_cast_fp16")]; + string _SplitHeadsQ__mh_w_4755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4755_equation_0, values = (var_45244_cast_fp16, var_45029_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4755_cast_fp16")]; + string _SplitHeadsQ__mh_w_4757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4757_equation_0, values = (var_45244_cast_fp16, var_45036_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4757_cast_fp16")]; + string _SplitHeadsQ__mh_w_4759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4759_equation_0, values = (var_45244_cast_fp16, var_45043_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4759_cast_fp16")]; + string _SplitHeadsQ__mh_w_4761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4761_equation_0, values = (var_45248_cast_fp16, var_45050_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4761_cast_fp16")]; + string _SplitHeadsQ__mh_w_4763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4763_equation_0, values = (var_45248_cast_fp16, var_45057_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4763_cast_fp16")]; + string _SplitHeadsQ__mh_w_4765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4765_equation_0, values = (var_45248_cast_fp16, var_45064_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4765_cast_fp16")]; + string _SplitHeadsQ__mh_w_4767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4767_equation_0, values = (var_45248_cast_fp16, var_45071_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4767_cast_fp16")]; + string _SplitHeadsQ__mh_w_4769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4769_equation_0, values = (var_45252_cast_fp16, var_45078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4769_cast_fp16")]; + string _SplitHeadsQ__mh_w_4771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4771_equation_0, values = (var_45252_cast_fp16, var_45085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4771_cast_fp16")]; + string _SplitHeadsQ__mh_w_4773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4773_equation_0, values = (var_45252_cast_fp16, var_45092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4773_cast_fp16")]; + string _SplitHeadsQ__mh_w_4775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4775_equation_0, values = (var_45252_cast_fp16, var_45099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4775_cast_fp16")]; + string _SplitHeadsQ__mh_w_4777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4777_equation_0, values = (var_45256_cast_fp16, var_45106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4777_cast_fp16")]; + string _SplitHeadsQ__mh_w_4779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4779_equation_0, values = (var_45256_cast_fp16, var_45113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4779_cast_fp16")]; + string _SplitHeadsQ__mh_w_4781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4781_equation_0, values = (var_45256_cast_fp16, var_45120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4781_cast_fp16")]; + string _SplitHeadsQ__mh_w_4783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4783_equation_0, values = (var_45256_cast_fp16, var_45127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4783_cast_fp16")]; + string _SplitHeadsQ__mh_w_4785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4785_equation_0, values = (var_45260_cast_fp16, var_45134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4785_cast_fp16")]; + string _SplitHeadsQ__mh_w_4787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4787_equation_0, values = (var_45260_cast_fp16, var_45141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4787_cast_fp16")]; + string _SplitHeadsQ__mh_w_4789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4789_equation_0, values = (var_45260_cast_fp16, var_45148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4789_cast_fp16")]; + string _SplitHeadsQ__mh_w_4791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4791_equation_0, values = (var_45260_cast_fp16, var_45155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4791_cast_fp16")]; + string _SplitHeadsQ__mh_w_4793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4793_equation_0, values = (var_45264_cast_fp16, var_45162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4793_cast_fp16")]; + string _SplitHeadsQ__mh_w_4795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4795_equation_0, values = (var_45264_cast_fp16, var_45169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4795_cast_fp16")]; + string _SplitHeadsQ__mh_w_4797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4797_equation_0, values = (var_45264_cast_fp16, var_45176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4797_cast_fp16")]; + string _SplitHeadsQ__mh_w_4799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4799_equation_0, values = (var_45264_cast_fp16, var_45183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4799_cast_fp16")]; + fp16 var_45505_to_fp16 = const()[name = string("op_45505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4641_cast_fp16, y = var_45505_to_fp16)[name = string("aw_chunk_4641_cast_fp16")]; + fp16 var_45507_to_fp16 = const()[name = string("op_45507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4643_cast_fp16, y = var_45507_to_fp16)[name = string("aw_chunk_4643_cast_fp16")]; + fp16 var_45509_to_fp16 = const()[name = string("op_45509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4645_cast_fp16, y = var_45509_to_fp16)[name = string("aw_chunk_4645_cast_fp16")]; + fp16 var_45511_to_fp16 = const()[name = string("op_45511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4647_cast_fp16, y = var_45511_to_fp16)[name = string("aw_chunk_4647_cast_fp16")]; + fp16 var_45513_to_fp16 = const()[name = string("op_45513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4649_cast_fp16, y = var_45513_to_fp16)[name = string("aw_chunk_4649_cast_fp16")]; + fp16 var_45515_to_fp16 = const()[name = string("op_45515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4651_cast_fp16, y = var_45515_to_fp16)[name = string("aw_chunk_4651_cast_fp16")]; + fp16 var_45517_to_fp16 = const()[name = string("op_45517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4653_cast_fp16, y = var_45517_to_fp16)[name = string("aw_chunk_4653_cast_fp16")]; + fp16 var_45519_to_fp16 = const()[name = string("op_45519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4655_cast_fp16, y = var_45519_to_fp16)[name = string("aw_chunk_4655_cast_fp16")]; + fp16 var_45521_to_fp16 = const()[name = string("op_45521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4657_cast_fp16, y = var_45521_to_fp16)[name = string("aw_chunk_4657_cast_fp16")]; + fp16 var_45523_to_fp16 = const()[name = string("op_45523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4659_cast_fp16, y = var_45523_to_fp16)[name = string("aw_chunk_4659_cast_fp16")]; + fp16 var_45525_to_fp16 = const()[name = string("op_45525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4661_cast_fp16, y = var_45525_to_fp16)[name = string("aw_chunk_4661_cast_fp16")]; + fp16 var_45527_to_fp16 = const()[name = string("op_45527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4663_cast_fp16, y = var_45527_to_fp16)[name = string("aw_chunk_4663_cast_fp16")]; + fp16 var_45529_to_fp16 = const()[name = string("op_45529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4665_cast_fp16, y = var_45529_to_fp16)[name = string("aw_chunk_4665_cast_fp16")]; + fp16 var_45531_to_fp16 = const()[name = string("op_45531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4667_cast_fp16, y = var_45531_to_fp16)[name = string("aw_chunk_4667_cast_fp16")]; + fp16 var_45533_to_fp16 = const()[name = string("op_45533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4669_cast_fp16, y = var_45533_to_fp16)[name = string("aw_chunk_4669_cast_fp16")]; + fp16 var_45535_to_fp16 = const()[name = string("op_45535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4671_cast_fp16, y = var_45535_to_fp16)[name = string("aw_chunk_4671_cast_fp16")]; + fp16 var_45537_to_fp16 = const()[name = string("op_45537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4673_cast_fp16, y = var_45537_to_fp16)[name = string("aw_chunk_4673_cast_fp16")]; + fp16 var_45539_to_fp16 = const()[name = string("op_45539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4675_cast_fp16, y = var_45539_to_fp16)[name = string("aw_chunk_4675_cast_fp16")]; + fp16 var_45541_to_fp16 = const()[name = string("op_45541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4677_cast_fp16, y = var_45541_to_fp16)[name = string("aw_chunk_4677_cast_fp16")]; + fp16 var_45543_to_fp16 = const()[name = string("op_45543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4679_cast_fp16, y = var_45543_to_fp16)[name = string("aw_chunk_4679_cast_fp16")]; + fp16 var_45545_to_fp16 = const()[name = string("op_45545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4681_cast_fp16, y = var_45545_to_fp16)[name = string("aw_chunk_4681_cast_fp16")]; + fp16 var_45547_to_fp16 = const()[name = string("op_45547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4683_cast_fp16, y = var_45547_to_fp16)[name = string("aw_chunk_4683_cast_fp16")]; + fp16 var_45549_to_fp16 = const()[name = string("op_45549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4685_cast_fp16, y = var_45549_to_fp16)[name = string("aw_chunk_4685_cast_fp16")]; + fp16 var_45551_to_fp16 = const()[name = string("op_45551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4687_cast_fp16, y = var_45551_to_fp16)[name = string("aw_chunk_4687_cast_fp16")]; + fp16 var_45553_to_fp16 = const()[name = string("op_45553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4689_cast_fp16, y = var_45553_to_fp16)[name = string("aw_chunk_4689_cast_fp16")]; + fp16 var_45555_to_fp16 = const()[name = string("op_45555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4691_cast_fp16, y = var_45555_to_fp16)[name = string("aw_chunk_4691_cast_fp16")]; + fp16 var_45557_to_fp16 = const()[name = string("op_45557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4693_cast_fp16, y = var_45557_to_fp16)[name = string("aw_chunk_4693_cast_fp16")]; + fp16 var_45559_to_fp16 = const()[name = string("op_45559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4695_cast_fp16, y = var_45559_to_fp16)[name = string("aw_chunk_4695_cast_fp16")]; + fp16 var_45561_to_fp16 = const()[name = string("op_45561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4697_cast_fp16, y = var_45561_to_fp16)[name = string("aw_chunk_4697_cast_fp16")]; + fp16 var_45563_to_fp16 = const()[name = string("op_45563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4699_cast_fp16, y = var_45563_to_fp16)[name = string("aw_chunk_4699_cast_fp16")]; + fp16 var_45565_to_fp16 = const()[name = string("op_45565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4701_cast_fp16, y = var_45565_to_fp16)[name = string("aw_chunk_4701_cast_fp16")]; + fp16 var_45567_to_fp16 = const()[name = string("op_45567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4703_cast_fp16, y = var_45567_to_fp16)[name = string("aw_chunk_4703_cast_fp16")]; + fp16 var_45569_to_fp16 = const()[name = string("op_45569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4705_cast_fp16, y = var_45569_to_fp16)[name = string("aw_chunk_4705_cast_fp16")]; + fp16 var_45571_to_fp16 = const()[name = string("op_45571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4707_cast_fp16, y = var_45571_to_fp16)[name = string("aw_chunk_4707_cast_fp16")]; + fp16 var_45573_to_fp16 = const()[name = string("op_45573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4709_cast_fp16, y = var_45573_to_fp16)[name = string("aw_chunk_4709_cast_fp16")]; + fp16 var_45575_to_fp16 = const()[name = string("op_45575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4711_cast_fp16, y = var_45575_to_fp16)[name = string("aw_chunk_4711_cast_fp16")]; + fp16 var_45577_to_fp16 = const()[name = string("op_45577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4713_cast_fp16, y = var_45577_to_fp16)[name = string("aw_chunk_4713_cast_fp16")]; + fp16 var_45579_to_fp16 = const()[name = string("op_45579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4715_cast_fp16, y = var_45579_to_fp16)[name = string("aw_chunk_4715_cast_fp16")]; + fp16 var_45581_to_fp16 = const()[name = string("op_45581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4717_cast_fp16, y = var_45581_to_fp16)[name = string("aw_chunk_4717_cast_fp16")]; + fp16 var_45583_to_fp16 = const()[name = string("op_45583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4719_cast_fp16, y = var_45583_to_fp16)[name = string("aw_chunk_4719_cast_fp16")]; + fp16 var_45585_to_fp16 = const()[name = string("op_45585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4721_cast_fp16, y = var_45585_to_fp16)[name = string("aw_chunk_4721_cast_fp16")]; + fp16 var_45587_to_fp16 = const()[name = string("op_45587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4723_cast_fp16, y = var_45587_to_fp16)[name = string("aw_chunk_4723_cast_fp16")]; + fp16 var_45589_to_fp16 = const()[name = string("op_45589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4725_cast_fp16, y = var_45589_to_fp16)[name = string("aw_chunk_4725_cast_fp16")]; + fp16 var_45591_to_fp16 = const()[name = string("op_45591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4727_cast_fp16, y = var_45591_to_fp16)[name = string("aw_chunk_4727_cast_fp16")]; + fp16 var_45593_to_fp16 = const()[name = string("op_45593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4729_cast_fp16, y = var_45593_to_fp16)[name = string("aw_chunk_4729_cast_fp16")]; + fp16 var_45595_to_fp16 = const()[name = string("op_45595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4731_cast_fp16, y = var_45595_to_fp16)[name = string("aw_chunk_4731_cast_fp16")]; + fp16 var_45597_to_fp16 = const()[name = string("op_45597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4733_cast_fp16, y = var_45597_to_fp16)[name = string("aw_chunk_4733_cast_fp16")]; + fp16 var_45599_to_fp16 = const()[name = string("op_45599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4735_cast_fp16, y = var_45599_to_fp16)[name = string("aw_chunk_4735_cast_fp16")]; + fp16 var_45601_to_fp16 = const()[name = string("op_45601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4737_cast_fp16, y = var_45601_to_fp16)[name = string("aw_chunk_4737_cast_fp16")]; + fp16 var_45603_to_fp16 = const()[name = string("op_45603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4739_cast_fp16, y = var_45603_to_fp16)[name = string("aw_chunk_4739_cast_fp16")]; + fp16 var_45605_to_fp16 = const()[name = string("op_45605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4741_cast_fp16, y = var_45605_to_fp16)[name = string("aw_chunk_4741_cast_fp16")]; + fp16 var_45607_to_fp16 = const()[name = string("op_45607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4743_cast_fp16, y = var_45607_to_fp16)[name = string("aw_chunk_4743_cast_fp16")]; + fp16 var_45609_to_fp16 = const()[name = string("op_45609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4745_cast_fp16, y = var_45609_to_fp16)[name = string("aw_chunk_4745_cast_fp16")]; + fp16 var_45611_to_fp16 = const()[name = string("op_45611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4747_cast_fp16, y = var_45611_to_fp16)[name = string("aw_chunk_4747_cast_fp16")]; + fp16 var_45613_to_fp16 = const()[name = string("op_45613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4749_cast_fp16, y = var_45613_to_fp16)[name = string("aw_chunk_4749_cast_fp16")]; + fp16 var_45615_to_fp16 = const()[name = string("op_45615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4751_cast_fp16, y = var_45615_to_fp16)[name = string("aw_chunk_4751_cast_fp16")]; + fp16 var_45617_to_fp16 = const()[name = string("op_45617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4753_cast_fp16, y = var_45617_to_fp16)[name = string("aw_chunk_4753_cast_fp16")]; + fp16 var_45619_to_fp16 = const()[name = string("op_45619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4755_cast_fp16, y = var_45619_to_fp16)[name = string("aw_chunk_4755_cast_fp16")]; + fp16 var_45621_to_fp16 = const()[name = string("op_45621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4757_cast_fp16, y = var_45621_to_fp16)[name = string("aw_chunk_4757_cast_fp16")]; + fp16 var_45623_to_fp16 = const()[name = string("op_45623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4759_cast_fp16, y = var_45623_to_fp16)[name = string("aw_chunk_4759_cast_fp16")]; + fp16 var_45625_to_fp16 = const()[name = string("op_45625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4761_cast_fp16, y = var_45625_to_fp16)[name = string("aw_chunk_4761_cast_fp16")]; + fp16 var_45627_to_fp16 = const()[name = string("op_45627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4763_cast_fp16, y = var_45627_to_fp16)[name = string("aw_chunk_4763_cast_fp16")]; + fp16 var_45629_to_fp16 = const()[name = string("op_45629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4765_cast_fp16, y = var_45629_to_fp16)[name = string("aw_chunk_4765_cast_fp16")]; + fp16 var_45631_to_fp16 = const()[name = string("op_45631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4767_cast_fp16, y = var_45631_to_fp16)[name = string("aw_chunk_4767_cast_fp16")]; + fp16 var_45633_to_fp16 = const()[name = string("op_45633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4769_cast_fp16, y = var_45633_to_fp16)[name = string("aw_chunk_4769_cast_fp16")]; + fp16 var_45635_to_fp16 = const()[name = string("op_45635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4771_cast_fp16, y = var_45635_to_fp16)[name = string("aw_chunk_4771_cast_fp16")]; + fp16 var_45637_to_fp16 = const()[name = string("op_45637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4773_cast_fp16, y = var_45637_to_fp16)[name = string("aw_chunk_4773_cast_fp16")]; + fp16 var_45639_to_fp16 = const()[name = string("op_45639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4775_cast_fp16, y = var_45639_to_fp16)[name = string("aw_chunk_4775_cast_fp16")]; + fp16 var_45641_to_fp16 = const()[name = string("op_45641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4777_cast_fp16, y = var_45641_to_fp16)[name = string("aw_chunk_4777_cast_fp16")]; + fp16 var_45643_to_fp16 = const()[name = string("op_45643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4779_cast_fp16, y = var_45643_to_fp16)[name = string("aw_chunk_4779_cast_fp16")]; + fp16 var_45645_to_fp16 = const()[name = string("op_45645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4781_cast_fp16, y = var_45645_to_fp16)[name = string("aw_chunk_4781_cast_fp16")]; + fp16 var_45647_to_fp16 = const()[name = string("op_45647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4783_cast_fp16, y = var_45647_to_fp16)[name = string("aw_chunk_4783_cast_fp16")]; + fp16 var_45649_to_fp16 = const()[name = string("op_45649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4785_cast_fp16, y = var_45649_to_fp16)[name = string("aw_chunk_4785_cast_fp16")]; + fp16 var_45651_to_fp16 = const()[name = string("op_45651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4787_cast_fp16, y = var_45651_to_fp16)[name = string("aw_chunk_4787_cast_fp16")]; + fp16 var_45653_to_fp16 = const()[name = string("op_45653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4789_cast_fp16, y = var_45653_to_fp16)[name = string("aw_chunk_4789_cast_fp16")]; + fp16 var_45655_to_fp16 = const()[name = string("op_45655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4791_cast_fp16, y = var_45655_to_fp16)[name = string("aw_chunk_4791_cast_fp16")]; + fp16 var_45657_to_fp16 = const()[name = string("op_45657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4793_cast_fp16, y = var_45657_to_fp16)[name = string("aw_chunk_4793_cast_fp16")]; + fp16 var_45659_to_fp16 = const()[name = string("op_45659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4795_cast_fp16, y = var_45659_to_fp16)[name = string("aw_chunk_4795_cast_fp16")]; + fp16 var_45661_to_fp16 = const()[name = string("op_45661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4797_cast_fp16, y = var_45661_to_fp16)[name = string("aw_chunk_4797_cast_fp16")]; + fp16 var_45663_to_fp16 = const()[name = string("op_45663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4799_cast_fp16, y = var_45663_to_fp16)[name = string("aw_chunk_4799_cast_fp16")]; + tensor var_45665_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4641_cast_fp16)[name = string("op_45665_cast_fp16")]; + tensor var_45666_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4643_cast_fp16)[name = string("op_45666_cast_fp16")]; + tensor var_45667_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4645_cast_fp16)[name = string("op_45667_cast_fp16")]; + tensor var_45668_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4647_cast_fp16)[name = string("op_45668_cast_fp16")]; + tensor var_45669_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4649_cast_fp16)[name = string("op_45669_cast_fp16")]; + tensor var_45670_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4651_cast_fp16)[name = string("op_45670_cast_fp16")]; + tensor var_45671_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4653_cast_fp16)[name = string("op_45671_cast_fp16")]; + tensor var_45672_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4655_cast_fp16)[name = string("op_45672_cast_fp16")]; + tensor var_45673_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4657_cast_fp16)[name = string("op_45673_cast_fp16")]; + tensor var_45674_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4659_cast_fp16)[name = string("op_45674_cast_fp16")]; + tensor var_45675_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4661_cast_fp16)[name = string("op_45675_cast_fp16")]; + tensor var_45676_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4663_cast_fp16)[name = string("op_45676_cast_fp16")]; + tensor var_45677_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4665_cast_fp16)[name = string("op_45677_cast_fp16")]; + tensor var_45678_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4667_cast_fp16)[name = string("op_45678_cast_fp16")]; + tensor var_45679_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4669_cast_fp16)[name = string("op_45679_cast_fp16")]; + tensor var_45680_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4671_cast_fp16)[name = string("op_45680_cast_fp16")]; + tensor var_45681_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4673_cast_fp16)[name = string("op_45681_cast_fp16")]; + tensor var_45682_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4675_cast_fp16)[name = string("op_45682_cast_fp16")]; + tensor var_45683_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4677_cast_fp16)[name = string("op_45683_cast_fp16")]; + tensor var_45684_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4679_cast_fp16)[name = string("op_45684_cast_fp16")]; + tensor var_45685_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4681_cast_fp16)[name = string("op_45685_cast_fp16")]; + tensor var_45686_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4683_cast_fp16)[name = string("op_45686_cast_fp16")]; + tensor var_45687_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4685_cast_fp16)[name = string("op_45687_cast_fp16")]; + tensor var_45688_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4687_cast_fp16)[name = string("op_45688_cast_fp16")]; + tensor var_45689_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4689_cast_fp16)[name = string("op_45689_cast_fp16")]; + tensor var_45690_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4691_cast_fp16)[name = string("op_45690_cast_fp16")]; + tensor var_45691_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4693_cast_fp16)[name = string("op_45691_cast_fp16")]; + tensor var_45692_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4695_cast_fp16)[name = string("op_45692_cast_fp16")]; + tensor var_45693_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4697_cast_fp16)[name = string("op_45693_cast_fp16")]; + tensor var_45694_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4699_cast_fp16)[name = string("op_45694_cast_fp16")]; + tensor var_45695_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4701_cast_fp16)[name = string("op_45695_cast_fp16")]; + tensor var_45696_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4703_cast_fp16)[name = string("op_45696_cast_fp16")]; + tensor var_45697_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4705_cast_fp16)[name = string("op_45697_cast_fp16")]; + tensor var_45698_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4707_cast_fp16)[name = string("op_45698_cast_fp16")]; + tensor var_45699_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4709_cast_fp16)[name = string("op_45699_cast_fp16")]; + tensor var_45700_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4711_cast_fp16)[name = string("op_45700_cast_fp16")]; + tensor var_45701_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4713_cast_fp16)[name = string("op_45701_cast_fp16")]; + tensor var_45702_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4715_cast_fp16)[name = string("op_45702_cast_fp16")]; + tensor var_45703_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4717_cast_fp16)[name = string("op_45703_cast_fp16")]; + tensor var_45704_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4719_cast_fp16)[name = string("op_45704_cast_fp16")]; + tensor var_45705_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4721_cast_fp16)[name = string("op_45705_cast_fp16")]; + tensor var_45706_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4723_cast_fp16)[name = string("op_45706_cast_fp16")]; + tensor var_45707_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4725_cast_fp16)[name = string("op_45707_cast_fp16")]; + tensor var_45708_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4727_cast_fp16)[name = string("op_45708_cast_fp16")]; + tensor var_45709_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4729_cast_fp16)[name = string("op_45709_cast_fp16")]; + tensor var_45710_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4731_cast_fp16)[name = string("op_45710_cast_fp16")]; + tensor var_45711_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4733_cast_fp16)[name = string("op_45711_cast_fp16")]; + tensor var_45712_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4735_cast_fp16)[name = string("op_45712_cast_fp16")]; + tensor var_45713_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4737_cast_fp16)[name = string("op_45713_cast_fp16")]; + tensor var_45714_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4739_cast_fp16)[name = string("op_45714_cast_fp16")]; + tensor var_45715_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4741_cast_fp16)[name = string("op_45715_cast_fp16")]; + tensor var_45716_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4743_cast_fp16)[name = string("op_45716_cast_fp16")]; + tensor var_45717_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4745_cast_fp16)[name = string("op_45717_cast_fp16")]; + tensor var_45718_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4747_cast_fp16)[name = string("op_45718_cast_fp16")]; + tensor var_45719_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4749_cast_fp16)[name = string("op_45719_cast_fp16")]; + tensor var_45720_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4751_cast_fp16)[name = string("op_45720_cast_fp16")]; + tensor var_45721_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4753_cast_fp16)[name = string("op_45721_cast_fp16")]; + tensor var_45722_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4755_cast_fp16)[name = string("op_45722_cast_fp16")]; + tensor var_45723_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4757_cast_fp16)[name = string("op_45723_cast_fp16")]; + tensor var_45724_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4759_cast_fp16)[name = string("op_45724_cast_fp16")]; + tensor var_45725_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4761_cast_fp16)[name = string("op_45725_cast_fp16")]; + tensor var_45726_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4763_cast_fp16)[name = string("op_45726_cast_fp16")]; + tensor var_45727_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4765_cast_fp16)[name = string("op_45727_cast_fp16")]; + tensor var_45728_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4767_cast_fp16)[name = string("op_45728_cast_fp16")]; + tensor var_45729_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4769_cast_fp16)[name = string("op_45729_cast_fp16")]; + tensor var_45730_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4771_cast_fp16)[name = string("op_45730_cast_fp16")]; + tensor var_45731_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4773_cast_fp16)[name = string("op_45731_cast_fp16")]; + tensor var_45732_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4775_cast_fp16)[name = string("op_45732_cast_fp16")]; + tensor var_45733_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4777_cast_fp16)[name = string("op_45733_cast_fp16")]; + tensor var_45734_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4779_cast_fp16)[name = string("op_45734_cast_fp16")]; + tensor var_45735_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4781_cast_fp16)[name = string("op_45735_cast_fp16")]; + tensor var_45736_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4783_cast_fp16)[name = string("op_45736_cast_fp16")]; + tensor var_45737_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4785_cast_fp16)[name = string("op_45737_cast_fp16")]; + tensor var_45738_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4787_cast_fp16)[name = string("op_45738_cast_fp16")]; + tensor var_45739_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4789_cast_fp16)[name = string("op_45739_cast_fp16")]; + tensor var_45740_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4791_cast_fp16)[name = string("op_45740_cast_fp16")]; + tensor var_45741_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4793_cast_fp16)[name = string("op_45741_cast_fp16")]; + tensor var_45742_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4795_cast_fp16)[name = string("op_45742_cast_fp16")]; + tensor var_45743_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4797_cast_fp16)[name = string("op_45743_cast_fp16")]; + tensor var_45744_cast_fp16 = softmax(axis = var_44490, x = aw_chunk_4799_cast_fp16)[name = string("op_45744_cast_fp16")]; + string var_45746_equation_0 = const()[name = string("op_45746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45746_cast_fp16 = einsum(equation = var_45746_equation_0, values = (var_45266_cast_fp16, var_45665_cast_fp16))[name = string("op_45746_cast_fp16")]; + string var_45748_equation_0 = const()[name = string("op_45748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45748_cast_fp16 = einsum(equation = var_45748_equation_0, values = (var_45266_cast_fp16, var_45666_cast_fp16))[name = string("op_45748_cast_fp16")]; + string var_45750_equation_0 = const()[name = string("op_45750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45750_cast_fp16 = einsum(equation = var_45750_equation_0, values = (var_45266_cast_fp16, var_45667_cast_fp16))[name = string("op_45750_cast_fp16")]; + string var_45752_equation_0 = const()[name = string("op_45752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45752_cast_fp16 = einsum(equation = var_45752_equation_0, values = (var_45266_cast_fp16, var_45668_cast_fp16))[name = string("op_45752_cast_fp16")]; + string var_45754_equation_0 = const()[name = string("op_45754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45754_cast_fp16 = einsum(equation = var_45754_equation_0, values = (var_45270_cast_fp16, var_45669_cast_fp16))[name = string("op_45754_cast_fp16")]; + string var_45756_equation_0 = const()[name = string("op_45756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45756_cast_fp16 = einsum(equation = var_45756_equation_0, values = (var_45270_cast_fp16, var_45670_cast_fp16))[name = string("op_45756_cast_fp16")]; + string var_45758_equation_0 = const()[name = string("op_45758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45758_cast_fp16 = einsum(equation = var_45758_equation_0, values = (var_45270_cast_fp16, var_45671_cast_fp16))[name = string("op_45758_cast_fp16")]; + string var_45760_equation_0 = const()[name = string("op_45760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45760_cast_fp16 = einsum(equation = var_45760_equation_0, values = (var_45270_cast_fp16, var_45672_cast_fp16))[name = string("op_45760_cast_fp16")]; + string var_45762_equation_0 = const()[name = string("op_45762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45762_cast_fp16 = einsum(equation = var_45762_equation_0, values = (var_45274_cast_fp16, var_45673_cast_fp16))[name = string("op_45762_cast_fp16")]; + string var_45764_equation_0 = const()[name = string("op_45764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45764_cast_fp16 = einsum(equation = var_45764_equation_0, values = (var_45274_cast_fp16, var_45674_cast_fp16))[name = string("op_45764_cast_fp16")]; + string var_45766_equation_0 = const()[name = string("op_45766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45766_cast_fp16 = einsum(equation = var_45766_equation_0, values = (var_45274_cast_fp16, var_45675_cast_fp16))[name = string("op_45766_cast_fp16")]; + string var_45768_equation_0 = const()[name = string("op_45768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45768_cast_fp16 = einsum(equation = var_45768_equation_0, values = (var_45274_cast_fp16, var_45676_cast_fp16))[name = string("op_45768_cast_fp16")]; + string var_45770_equation_0 = const()[name = string("op_45770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45770_cast_fp16 = einsum(equation = var_45770_equation_0, values = (var_45278_cast_fp16, var_45677_cast_fp16))[name = string("op_45770_cast_fp16")]; + string var_45772_equation_0 = const()[name = string("op_45772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45772_cast_fp16 = einsum(equation = var_45772_equation_0, values = (var_45278_cast_fp16, var_45678_cast_fp16))[name = string("op_45772_cast_fp16")]; + string var_45774_equation_0 = const()[name = string("op_45774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45774_cast_fp16 = einsum(equation = var_45774_equation_0, values = (var_45278_cast_fp16, var_45679_cast_fp16))[name = string("op_45774_cast_fp16")]; + string var_45776_equation_0 = const()[name = string("op_45776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45776_cast_fp16 = einsum(equation = var_45776_equation_0, values = (var_45278_cast_fp16, var_45680_cast_fp16))[name = string("op_45776_cast_fp16")]; + string var_45778_equation_0 = const()[name = string("op_45778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45778_cast_fp16 = einsum(equation = var_45778_equation_0, values = (var_45282_cast_fp16, var_45681_cast_fp16))[name = string("op_45778_cast_fp16")]; + string var_45780_equation_0 = const()[name = string("op_45780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45780_cast_fp16 = einsum(equation = var_45780_equation_0, values = (var_45282_cast_fp16, var_45682_cast_fp16))[name = string("op_45780_cast_fp16")]; + string var_45782_equation_0 = const()[name = string("op_45782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45782_cast_fp16 = einsum(equation = var_45782_equation_0, values = (var_45282_cast_fp16, var_45683_cast_fp16))[name = string("op_45782_cast_fp16")]; + string var_45784_equation_0 = const()[name = string("op_45784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45784_cast_fp16 = einsum(equation = var_45784_equation_0, values = (var_45282_cast_fp16, var_45684_cast_fp16))[name = string("op_45784_cast_fp16")]; + string var_45786_equation_0 = const()[name = string("op_45786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45786_cast_fp16 = einsum(equation = var_45786_equation_0, values = (var_45286_cast_fp16, var_45685_cast_fp16))[name = string("op_45786_cast_fp16")]; + string var_45788_equation_0 = const()[name = string("op_45788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45788_cast_fp16 = einsum(equation = var_45788_equation_0, values = (var_45286_cast_fp16, var_45686_cast_fp16))[name = string("op_45788_cast_fp16")]; + string var_45790_equation_0 = const()[name = string("op_45790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45790_cast_fp16 = einsum(equation = var_45790_equation_0, values = (var_45286_cast_fp16, var_45687_cast_fp16))[name = string("op_45790_cast_fp16")]; + string var_45792_equation_0 = const()[name = string("op_45792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45792_cast_fp16 = einsum(equation = var_45792_equation_0, values = (var_45286_cast_fp16, var_45688_cast_fp16))[name = string("op_45792_cast_fp16")]; + string var_45794_equation_0 = const()[name = string("op_45794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45794_cast_fp16 = einsum(equation = var_45794_equation_0, values = (var_45290_cast_fp16, var_45689_cast_fp16))[name = string("op_45794_cast_fp16")]; + string var_45796_equation_0 = const()[name = string("op_45796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45796_cast_fp16 = einsum(equation = var_45796_equation_0, values = (var_45290_cast_fp16, var_45690_cast_fp16))[name = string("op_45796_cast_fp16")]; + string var_45798_equation_0 = const()[name = string("op_45798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45798_cast_fp16 = einsum(equation = var_45798_equation_0, values = (var_45290_cast_fp16, var_45691_cast_fp16))[name = string("op_45798_cast_fp16")]; + string var_45800_equation_0 = const()[name = string("op_45800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45800_cast_fp16 = einsum(equation = var_45800_equation_0, values = (var_45290_cast_fp16, var_45692_cast_fp16))[name = string("op_45800_cast_fp16")]; + string var_45802_equation_0 = const()[name = string("op_45802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45802_cast_fp16 = einsum(equation = var_45802_equation_0, values = (var_45294_cast_fp16, var_45693_cast_fp16))[name = string("op_45802_cast_fp16")]; + string var_45804_equation_0 = const()[name = string("op_45804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45804_cast_fp16 = einsum(equation = var_45804_equation_0, values = (var_45294_cast_fp16, var_45694_cast_fp16))[name = string("op_45804_cast_fp16")]; + string var_45806_equation_0 = const()[name = string("op_45806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45806_cast_fp16 = einsum(equation = var_45806_equation_0, values = (var_45294_cast_fp16, var_45695_cast_fp16))[name = string("op_45806_cast_fp16")]; + string var_45808_equation_0 = const()[name = string("op_45808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45808_cast_fp16 = einsum(equation = var_45808_equation_0, values = (var_45294_cast_fp16, var_45696_cast_fp16))[name = string("op_45808_cast_fp16")]; + string var_45810_equation_0 = const()[name = string("op_45810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45810_cast_fp16 = einsum(equation = var_45810_equation_0, values = (var_45298_cast_fp16, var_45697_cast_fp16))[name = string("op_45810_cast_fp16")]; + string var_45812_equation_0 = const()[name = string("op_45812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45812_cast_fp16 = einsum(equation = var_45812_equation_0, values = (var_45298_cast_fp16, var_45698_cast_fp16))[name = string("op_45812_cast_fp16")]; + string var_45814_equation_0 = const()[name = string("op_45814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45814_cast_fp16 = einsum(equation = var_45814_equation_0, values = (var_45298_cast_fp16, var_45699_cast_fp16))[name = string("op_45814_cast_fp16")]; + string var_45816_equation_0 = const()[name = string("op_45816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45816_cast_fp16 = einsum(equation = var_45816_equation_0, values = (var_45298_cast_fp16, var_45700_cast_fp16))[name = string("op_45816_cast_fp16")]; + string var_45818_equation_0 = const()[name = string("op_45818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45818_cast_fp16 = einsum(equation = var_45818_equation_0, values = (var_45302_cast_fp16, var_45701_cast_fp16))[name = string("op_45818_cast_fp16")]; + string var_45820_equation_0 = const()[name = string("op_45820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45820_cast_fp16 = einsum(equation = var_45820_equation_0, values = (var_45302_cast_fp16, var_45702_cast_fp16))[name = string("op_45820_cast_fp16")]; + string var_45822_equation_0 = const()[name = string("op_45822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45822_cast_fp16 = einsum(equation = var_45822_equation_0, values = (var_45302_cast_fp16, var_45703_cast_fp16))[name = string("op_45822_cast_fp16")]; + string var_45824_equation_0 = const()[name = string("op_45824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45824_cast_fp16 = einsum(equation = var_45824_equation_0, values = (var_45302_cast_fp16, var_45704_cast_fp16))[name = string("op_45824_cast_fp16")]; + string var_45826_equation_0 = const()[name = string("op_45826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45826_cast_fp16 = einsum(equation = var_45826_equation_0, values = (var_45306_cast_fp16, var_45705_cast_fp16))[name = string("op_45826_cast_fp16")]; + string var_45828_equation_0 = const()[name = string("op_45828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45828_cast_fp16 = einsum(equation = var_45828_equation_0, values = (var_45306_cast_fp16, var_45706_cast_fp16))[name = string("op_45828_cast_fp16")]; + string var_45830_equation_0 = const()[name = string("op_45830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45830_cast_fp16 = einsum(equation = var_45830_equation_0, values = (var_45306_cast_fp16, var_45707_cast_fp16))[name = string("op_45830_cast_fp16")]; + string var_45832_equation_0 = const()[name = string("op_45832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45832_cast_fp16 = einsum(equation = var_45832_equation_0, values = (var_45306_cast_fp16, var_45708_cast_fp16))[name = string("op_45832_cast_fp16")]; + string var_45834_equation_0 = const()[name = string("op_45834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45834_cast_fp16 = einsum(equation = var_45834_equation_0, values = (var_45310_cast_fp16, var_45709_cast_fp16))[name = string("op_45834_cast_fp16")]; + string var_45836_equation_0 = const()[name = string("op_45836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45836_cast_fp16 = einsum(equation = var_45836_equation_0, values = (var_45310_cast_fp16, var_45710_cast_fp16))[name = string("op_45836_cast_fp16")]; + string var_45838_equation_0 = const()[name = string("op_45838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45838_cast_fp16 = einsum(equation = var_45838_equation_0, values = (var_45310_cast_fp16, var_45711_cast_fp16))[name = string("op_45838_cast_fp16")]; + string var_45840_equation_0 = const()[name = string("op_45840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45840_cast_fp16 = einsum(equation = var_45840_equation_0, values = (var_45310_cast_fp16, var_45712_cast_fp16))[name = string("op_45840_cast_fp16")]; + string var_45842_equation_0 = const()[name = string("op_45842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45842_cast_fp16 = einsum(equation = var_45842_equation_0, values = (var_45314_cast_fp16, var_45713_cast_fp16))[name = string("op_45842_cast_fp16")]; + string var_45844_equation_0 = const()[name = string("op_45844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45844_cast_fp16 = einsum(equation = var_45844_equation_0, values = (var_45314_cast_fp16, var_45714_cast_fp16))[name = string("op_45844_cast_fp16")]; + string var_45846_equation_0 = const()[name = string("op_45846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45846_cast_fp16 = einsum(equation = var_45846_equation_0, values = (var_45314_cast_fp16, var_45715_cast_fp16))[name = string("op_45846_cast_fp16")]; + string var_45848_equation_0 = const()[name = string("op_45848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45848_cast_fp16 = einsum(equation = var_45848_equation_0, values = (var_45314_cast_fp16, var_45716_cast_fp16))[name = string("op_45848_cast_fp16")]; + string var_45850_equation_0 = const()[name = string("op_45850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45850_cast_fp16 = einsum(equation = var_45850_equation_0, values = (var_45318_cast_fp16, var_45717_cast_fp16))[name = string("op_45850_cast_fp16")]; + string var_45852_equation_0 = const()[name = string("op_45852_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45852_cast_fp16 = einsum(equation = var_45852_equation_0, values = (var_45318_cast_fp16, var_45718_cast_fp16))[name = string("op_45852_cast_fp16")]; + string var_45854_equation_0 = const()[name = string("op_45854_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45854_cast_fp16 = einsum(equation = var_45854_equation_0, values = (var_45318_cast_fp16, var_45719_cast_fp16))[name = string("op_45854_cast_fp16")]; + string var_45856_equation_0 = const()[name = string("op_45856_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45856_cast_fp16 = einsum(equation = var_45856_equation_0, values = (var_45318_cast_fp16, var_45720_cast_fp16))[name = string("op_45856_cast_fp16")]; + string var_45858_equation_0 = const()[name = string("op_45858_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45858_cast_fp16 = einsum(equation = var_45858_equation_0, values = (var_45322_cast_fp16, var_45721_cast_fp16))[name = string("op_45858_cast_fp16")]; + string var_45860_equation_0 = const()[name = string("op_45860_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45860_cast_fp16 = einsum(equation = var_45860_equation_0, values = (var_45322_cast_fp16, var_45722_cast_fp16))[name = string("op_45860_cast_fp16")]; + string var_45862_equation_0 = const()[name = string("op_45862_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45862_cast_fp16 = einsum(equation = var_45862_equation_0, values = (var_45322_cast_fp16, var_45723_cast_fp16))[name = string("op_45862_cast_fp16")]; + string var_45864_equation_0 = const()[name = string("op_45864_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45864_cast_fp16 = einsum(equation = var_45864_equation_0, values = (var_45322_cast_fp16, var_45724_cast_fp16))[name = string("op_45864_cast_fp16")]; + string var_45866_equation_0 = const()[name = string("op_45866_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45866_cast_fp16 = einsum(equation = var_45866_equation_0, values = (var_45326_cast_fp16, var_45725_cast_fp16))[name = string("op_45866_cast_fp16")]; + string var_45868_equation_0 = const()[name = string("op_45868_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45868_cast_fp16 = einsum(equation = var_45868_equation_0, values = (var_45326_cast_fp16, var_45726_cast_fp16))[name = string("op_45868_cast_fp16")]; + string var_45870_equation_0 = const()[name = string("op_45870_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45870_cast_fp16 = einsum(equation = var_45870_equation_0, values = (var_45326_cast_fp16, var_45727_cast_fp16))[name = string("op_45870_cast_fp16")]; + string var_45872_equation_0 = const()[name = string("op_45872_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45872_cast_fp16 = einsum(equation = var_45872_equation_0, values = (var_45326_cast_fp16, var_45728_cast_fp16))[name = string("op_45872_cast_fp16")]; + string var_45874_equation_0 = const()[name = string("op_45874_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45874_cast_fp16 = einsum(equation = var_45874_equation_0, values = (var_45330_cast_fp16, var_45729_cast_fp16))[name = string("op_45874_cast_fp16")]; + string var_45876_equation_0 = const()[name = string("op_45876_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45876_cast_fp16 = einsum(equation = var_45876_equation_0, values = (var_45330_cast_fp16, var_45730_cast_fp16))[name = string("op_45876_cast_fp16")]; + string var_45878_equation_0 = const()[name = string("op_45878_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45878_cast_fp16 = einsum(equation = var_45878_equation_0, values = (var_45330_cast_fp16, var_45731_cast_fp16))[name = string("op_45878_cast_fp16")]; + string var_45880_equation_0 = const()[name = string("op_45880_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45880_cast_fp16 = einsum(equation = var_45880_equation_0, values = (var_45330_cast_fp16, var_45732_cast_fp16))[name = string("op_45880_cast_fp16")]; + string var_45882_equation_0 = const()[name = string("op_45882_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45882_cast_fp16 = einsum(equation = var_45882_equation_0, values = (var_45334_cast_fp16, var_45733_cast_fp16))[name = string("op_45882_cast_fp16")]; + string var_45884_equation_0 = const()[name = string("op_45884_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45884_cast_fp16 = einsum(equation = var_45884_equation_0, values = (var_45334_cast_fp16, var_45734_cast_fp16))[name = string("op_45884_cast_fp16")]; + string var_45886_equation_0 = const()[name = string("op_45886_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45886_cast_fp16 = einsum(equation = var_45886_equation_0, values = (var_45334_cast_fp16, var_45735_cast_fp16))[name = string("op_45886_cast_fp16")]; + string var_45888_equation_0 = const()[name = string("op_45888_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45888_cast_fp16 = einsum(equation = var_45888_equation_0, values = (var_45334_cast_fp16, var_45736_cast_fp16))[name = string("op_45888_cast_fp16")]; + string var_45890_equation_0 = const()[name = string("op_45890_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45890_cast_fp16 = einsum(equation = var_45890_equation_0, values = (var_45338_cast_fp16, var_45737_cast_fp16))[name = string("op_45890_cast_fp16")]; + string var_45892_equation_0 = const()[name = string("op_45892_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45892_cast_fp16 = einsum(equation = var_45892_equation_0, values = (var_45338_cast_fp16, var_45738_cast_fp16))[name = string("op_45892_cast_fp16")]; + string var_45894_equation_0 = const()[name = string("op_45894_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45894_cast_fp16 = einsum(equation = var_45894_equation_0, values = (var_45338_cast_fp16, var_45739_cast_fp16))[name = string("op_45894_cast_fp16")]; + string var_45896_equation_0 = const()[name = string("op_45896_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45896_cast_fp16 = einsum(equation = var_45896_equation_0, values = (var_45338_cast_fp16, var_45740_cast_fp16))[name = string("op_45896_cast_fp16")]; + string var_45898_equation_0 = const()[name = string("op_45898_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45898_cast_fp16 = einsum(equation = var_45898_equation_0, values = (var_45342_cast_fp16, var_45741_cast_fp16))[name = string("op_45898_cast_fp16")]; + string var_45900_equation_0 = const()[name = string("op_45900_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45900_cast_fp16 = einsum(equation = var_45900_equation_0, values = (var_45342_cast_fp16, var_45742_cast_fp16))[name = string("op_45900_cast_fp16")]; + string var_45902_equation_0 = const()[name = string("op_45902_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45902_cast_fp16 = einsum(equation = var_45902_equation_0, values = (var_45342_cast_fp16, var_45743_cast_fp16))[name = string("op_45902_cast_fp16")]; + string var_45904_equation_0 = const()[name = string("op_45904_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_45904_cast_fp16 = einsum(equation = var_45904_equation_0, values = (var_45342_cast_fp16, var_45744_cast_fp16))[name = string("op_45904_cast_fp16")]; + bool var_45906_interleave_0 = const()[name = string("op_45906_interleave_0"), val = bool(false)]; + tensor var_45906_cast_fp16 = concat(axis = var_44465, interleave = var_45906_interleave_0, values = (var_45746_cast_fp16, var_45748_cast_fp16, var_45750_cast_fp16, var_45752_cast_fp16))[name = string("op_45906_cast_fp16")]; + bool var_45908_interleave_0 = const()[name = string("op_45908_interleave_0"), val = bool(false)]; + tensor var_45908_cast_fp16 = concat(axis = var_44465, interleave = var_45908_interleave_0, values = (var_45754_cast_fp16, var_45756_cast_fp16, var_45758_cast_fp16, var_45760_cast_fp16))[name = string("op_45908_cast_fp16")]; + bool var_45910_interleave_0 = const()[name = string("op_45910_interleave_0"), val = bool(false)]; + tensor var_45910_cast_fp16 = concat(axis = var_44465, interleave = var_45910_interleave_0, values = (var_45762_cast_fp16, var_45764_cast_fp16, var_45766_cast_fp16, var_45768_cast_fp16))[name = string("op_45910_cast_fp16")]; + bool var_45912_interleave_0 = const()[name = string("op_45912_interleave_0"), val = bool(false)]; + tensor var_45912_cast_fp16 = concat(axis = var_44465, interleave = var_45912_interleave_0, values = (var_45770_cast_fp16, var_45772_cast_fp16, var_45774_cast_fp16, var_45776_cast_fp16))[name = string("op_45912_cast_fp16")]; + bool var_45914_interleave_0 = const()[name = string("op_45914_interleave_0"), val = bool(false)]; + tensor var_45914_cast_fp16 = concat(axis = var_44465, interleave = var_45914_interleave_0, values = (var_45778_cast_fp16, var_45780_cast_fp16, var_45782_cast_fp16, var_45784_cast_fp16))[name = string("op_45914_cast_fp16")]; + bool var_45916_interleave_0 = const()[name = string("op_45916_interleave_0"), val = bool(false)]; + tensor var_45916_cast_fp16 = concat(axis = var_44465, interleave = var_45916_interleave_0, values = (var_45786_cast_fp16, var_45788_cast_fp16, var_45790_cast_fp16, var_45792_cast_fp16))[name = string("op_45916_cast_fp16")]; + bool var_45918_interleave_0 = const()[name = string("op_45918_interleave_0"), val = bool(false)]; + tensor var_45918_cast_fp16 = concat(axis = var_44465, interleave = var_45918_interleave_0, values = (var_45794_cast_fp16, var_45796_cast_fp16, var_45798_cast_fp16, var_45800_cast_fp16))[name = string("op_45918_cast_fp16")]; + bool var_45920_interleave_0 = const()[name = string("op_45920_interleave_0"), val = bool(false)]; + tensor var_45920_cast_fp16 = concat(axis = var_44465, interleave = var_45920_interleave_0, values = (var_45802_cast_fp16, var_45804_cast_fp16, var_45806_cast_fp16, var_45808_cast_fp16))[name = string("op_45920_cast_fp16")]; + bool var_45922_interleave_0 = const()[name = string("op_45922_interleave_0"), val = bool(false)]; + tensor var_45922_cast_fp16 = concat(axis = var_44465, interleave = var_45922_interleave_0, values = (var_45810_cast_fp16, var_45812_cast_fp16, var_45814_cast_fp16, var_45816_cast_fp16))[name = string("op_45922_cast_fp16")]; + bool var_45924_interleave_0 = const()[name = string("op_45924_interleave_0"), val = bool(false)]; + tensor var_45924_cast_fp16 = concat(axis = var_44465, interleave = var_45924_interleave_0, values = (var_45818_cast_fp16, var_45820_cast_fp16, var_45822_cast_fp16, var_45824_cast_fp16))[name = string("op_45924_cast_fp16")]; + bool var_45926_interleave_0 = const()[name = string("op_45926_interleave_0"), val = bool(false)]; + tensor var_45926_cast_fp16 = concat(axis = var_44465, interleave = var_45926_interleave_0, values = (var_45826_cast_fp16, var_45828_cast_fp16, var_45830_cast_fp16, var_45832_cast_fp16))[name = string("op_45926_cast_fp16")]; + bool var_45928_interleave_0 = const()[name = string("op_45928_interleave_0"), val = bool(false)]; + tensor var_45928_cast_fp16 = concat(axis = var_44465, interleave = var_45928_interleave_0, values = (var_45834_cast_fp16, var_45836_cast_fp16, var_45838_cast_fp16, var_45840_cast_fp16))[name = string("op_45928_cast_fp16")]; + bool var_45930_interleave_0 = const()[name = string("op_45930_interleave_0"), val = bool(false)]; + tensor var_45930_cast_fp16 = concat(axis = var_44465, interleave = var_45930_interleave_0, values = (var_45842_cast_fp16, var_45844_cast_fp16, var_45846_cast_fp16, var_45848_cast_fp16))[name = string("op_45930_cast_fp16")]; + bool var_45932_interleave_0 = const()[name = string("op_45932_interleave_0"), val = bool(false)]; + tensor var_45932_cast_fp16 = concat(axis = var_44465, interleave = var_45932_interleave_0, values = (var_45850_cast_fp16, var_45852_cast_fp16, var_45854_cast_fp16, var_45856_cast_fp16))[name = string("op_45932_cast_fp16")]; + bool var_45934_interleave_0 = const()[name = string("op_45934_interleave_0"), val = bool(false)]; + tensor var_45934_cast_fp16 = concat(axis = var_44465, interleave = var_45934_interleave_0, values = (var_45858_cast_fp16, var_45860_cast_fp16, var_45862_cast_fp16, var_45864_cast_fp16))[name = string("op_45934_cast_fp16")]; + bool var_45936_interleave_0 = const()[name = string("op_45936_interleave_0"), val = bool(false)]; + tensor var_45936_cast_fp16 = concat(axis = var_44465, interleave = var_45936_interleave_0, values = (var_45866_cast_fp16, var_45868_cast_fp16, var_45870_cast_fp16, var_45872_cast_fp16))[name = string("op_45936_cast_fp16")]; + bool var_45938_interleave_0 = const()[name = string("op_45938_interleave_0"), val = bool(false)]; + tensor var_45938_cast_fp16 = concat(axis = var_44465, interleave = var_45938_interleave_0, values = (var_45874_cast_fp16, var_45876_cast_fp16, var_45878_cast_fp16, var_45880_cast_fp16))[name = string("op_45938_cast_fp16")]; + bool var_45940_interleave_0 = const()[name = string("op_45940_interleave_0"), val = bool(false)]; + tensor var_45940_cast_fp16 = concat(axis = var_44465, interleave = var_45940_interleave_0, values = (var_45882_cast_fp16, var_45884_cast_fp16, var_45886_cast_fp16, var_45888_cast_fp16))[name = string("op_45940_cast_fp16")]; + bool var_45942_interleave_0 = const()[name = string("op_45942_interleave_0"), val = bool(false)]; + tensor var_45942_cast_fp16 = concat(axis = var_44465, interleave = var_45942_interleave_0, values = (var_45890_cast_fp16, var_45892_cast_fp16, var_45894_cast_fp16, var_45896_cast_fp16))[name = string("op_45942_cast_fp16")]; + bool var_45944_interleave_0 = const()[name = string("op_45944_interleave_0"), val = bool(false)]; + tensor var_45944_cast_fp16 = concat(axis = var_44465, interleave = var_45944_interleave_0, values = (var_45898_cast_fp16, var_45900_cast_fp16, var_45902_cast_fp16, var_45904_cast_fp16))[name = string("op_45944_cast_fp16")]; + bool input_233_interleave_0 = const()[name = string("input_233_interleave_0"), val = bool(false)]; + tensor input_233_cast_fp16 = concat(axis = var_44490, interleave = input_233_interleave_0, values = (var_45906_cast_fp16, var_45908_cast_fp16, var_45910_cast_fp16, var_45912_cast_fp16, var_45914_cast_fp16, var_45916_cast_fp16, var_45918_cast_fp16, var_45920_cast_fp16, var_45922_cast_fp16, var_45924_cast_fp16, var_45926_cast_fp16, var_45928_cast_fp16, var_45930_cast_fp16, var_45932_cast_fp16, var_45934_cast_fp16, var_45936_cast_fp16, var_45938_cast_fp16, var_45940_cast_fp16, var_45942_cast_fp16, var_45944_cast_fp16))[name = string("input_233_cast_fp16")]; + string obj_119_pad_type_0 = const()[name = string("obj_119_pad_type_0"), val = string("valid")]; + tensor obj_119_strides_0 = const()[name = string("obj_119_strides_0"), val = tensor([1, 1])]; + tensor obj_119_pad_0 = const()[name = string("obj_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_119_dilations_0 = const()[name = string("obj_119_dilations_0"), val = tensor([1, 1])]; + int32 obj_119_groups_0 = const()[name = string("obj_119_groups_0"), val = int32(1)]; + tensor layers_29_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1165750400)))]; + tensor layers_29_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169027264)))]; + tensor obj_119_cast_fp16 = conv(bias = layers_29_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_29_self_attn_o_proj_weight_to_fp16, x = input_233_cast_fp16)[name = string("obj_119_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_119_cast_fp16)[name = string("inputs_119_cast_fp16")]; + tensor out_119_axes_0 = const()[name = string("out_119_axes_0"), val = tensor([1])]; + fp16 var_45963_to_fp16 = const()[name = string("op_45963_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_45963_to_fp16, x = inputs_119_cast_fp16)[name = string("out_119_cast_fp16")]; + tensor input_235_gamma_0_to_fp16 = const()[name = string("input_235_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169029888)))]; + tensor input_235_beta_0_to_fp16 = const()[name = string("input_235_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169032512)))]; + fp16 input_235_epsilon_0_to_fp16 = const()[name = string("input_235_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = string("input_235_cast_fp16")]; + string input_237_pad_type_0 = const()[name = string("input_237_pad_type_0"), val = string("valid")]; + tensor input_237_strides_0 = const()[name = string("input_237_strides_0"), val = tensor([1, 1])]; + tensor input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_237_dilations_0 = const()[name = string("input_237_dilations_0"), val = tensor([1, 1])]; + int32 input_237_groups_0 = const()[name = string("input_237_groups_0"), val = int32(1)]; + tensor layers_29_fc1_weight_to_fp16 = const()[name = string("layers_29_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1169035136)))]; + tensor layers_29_fc1_bias_to_fp16 = const()[name = string("layers_29_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182142400)))]; + tensor input_237_cast_fp16 = conv(bias = layers_29_fc1_bias_to_fp16, dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_29_fc1_weight_to_fp16, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")]; + string input_239_mode_0 = const()[name = string("input_239_mode_0"), val = string("EXACT")]; + tensor input_239_cast_fp16 = gelu(mode = input_239_mode_0, x = input_237_cast_fp16)[name = string("input_239_cast_fp16")]; + string hidden_states_63_pad_type_0 = const()[name = string("hidden_states_63_pad_type_0"), val = string("valid")]; + tensor hidden_states_63_strides_0 = const()[name = string("hidden_states_63_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_63_pad_0 = const()[name = string("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_63_dilations_0 = const()[name = string("hidden_states_63_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_63_groups_0 = const()[name = string("hidden_states_63_groups_0"), val = int32(1)]; + tensor layers_29_fc2_weight_to_fp16 = const()[name = string("layers_29_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1182152704)))]; + tensor layers_29_fc2_bias_to_fp16 = const()[name = string("layers_29_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195259968)))]; + tensor hidden_states_63_cast_fp16 = conv(bias = layers_29_fc2_bias_to_fp16, dilations = hidden_states_63_dilations_0, groups = hidden_states_63_groups_0, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = hidden_states_63_strides_0, weight = layers_29_fc2_weight_to_fp16, x = input_239_cast_fp16)[name = string("hidden_states_63_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_63_cast_fp16)[name = string("inputs_121_cast_fp16")]; + int32 var_45992 = const()[name = string("op_45992"), val = int32(3)]; + int32 var_46017 = const()[name = string("op_46017"), val = int32(1)]; + tensor out_121_axes_0 = const()[name = string("out_121_axes_0"), val = tensor([1])]; + fp16 var_46034_to_fp16 = const()[name = string("op_46034_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_46034_to_fp16, x = inputs_121_cast_fp16)[name = string("out_121_cast_fp16")]; + tensor obj_121_gamma_0_to_fp16 = const()[name = string("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195262592)))]; + tensor obj_121_beta_0_to_fp16 = const()[name = string("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195265216)))]; + fp16 obj_121_epsilon_0_to_fp16 = const()[name = string("obj_121_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = string("obj_121_cast_fp16")]; + string query_61_pad_type_0 = const()[name = string("query_61_pad_type_0"), val = string("valid")]; + tensor query_61_strides_0 = const()[name = string("query_61_strides_0"), val = tensor([1, 1])]; + tensor query_61_pad_0 = const()[name = string("query_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_61_dilations_0 = const()[name = string("query_61_dilations_0"), val = tensor([1, 1])]; + int32 query_61_groups_0 = const()[name = string("query_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1195267840)))]; + tensor layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198544704)))]; + tensor query_61_cast_fp16 = conv(bias = layers_30_self_attn_q_proj_bias_to_fp16, dilations = query_61_dilations_0, groups = query_61_groups_0, pad = query_61_pad_0, pad_type = query_61_pad_type_0, strides = query_61_strides_0, weight = layers_30_self_attn_q_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("query_61_cast_fp16")]; + string key_61_pad_type_0 = const()[name = string("key_61_pad_type_0"), val = string("valid")]; + tensor key_61_strides_0 = const()[name = string("key_61_strides_0"), val = tensor([1, 1])]; + tensor key_61_pad_0 = const()[name = string("key_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_61_dilations_0 = const()[name = string("key_61_dilations_0"), val = tensor([1, 1])]; + int32 key_61_groups_0 = const()[name = string("key_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1198547328)))]; + tensor key_61_cast_fp16 = conv(dilations = key_61_dilations_0, groups = key_61_groups_0, pad = key_61_pad_0, pad_type = key_61_pad_type_0, strides = key_61_strides_0, weight = layers_30_self_attn_k_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("key_61_cast_fp16")]; + string value_61_pad_type_0 = const()[name = string("value_61_pad_type_0"), val = string("valid")]; + tensor value_61_strides_0 = const()[name = string("value_61_strides_0"), val = tensor([1, 1])]; + tensor value_61_pad_0 = const()[name = string("value_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_61_dilations_0 = const()[name = string("value_61_dilations_0"), val = tensor([1, 1])]; + int32 value_61_groups_0 = const()[name = string("value_61_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1201824192)))]; + tensor layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205101056)))]; + tensor value_61_cast_fp16 = conv(bias = layers_30_self_attn_v_proj_bias_to_fp16, dilations = value_61_dilations_0, groups = value_61_groups_0, pad = value_61_pad_0, pad_type = value_61_pad_type_0, strides = value_61_strides_0, weight = layers_30_self_attn_v_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = string("value_61_cast_fp16")]; + tensor var_46072_begin_0 = const()[name = string("op_46072_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46072_end_0 = const()[name = string("op_46072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46072_end_mask_0 = const()[name = string("op_46072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46072_cast_fp16 = slice_by_index(begin = var_46072_begin_0, end = var_46072_end_0, end_mask = var_46072_end_mask_0, x = query_61_cast_fp16)[name = string("op_46072_cast_fp16")]; + tensor var_46076_begin_0 = const()[name = string("op_46076_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46076_end_0 = const()[name = string("op_46076_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_46076_end_mask_0 = const()[name = string("op_46076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46076_cast_fp16 = slice_by_index(begin = var_46076_begin_0, end = var_46076_end_0, end_mask = var_46076_end_mask_0, x = query_61_cast_fp16)[name = string("op_46076_cast_fp16")]; + tensor var_46080_begin_0 = const()[name = string("op_46080_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46080_end_0 = const()[name = string("op_46080_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_46080_end_mask_0 = const()[name = string("op_46080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46080_cast_fp16 = slice_by_index(begin = var_46080_begin_0, end = var_46080_end_0, end_mask = var_46080_end_mask_0, x = query_61_cast_fp16)[name = string("op_46080_cast_fp16")]; + tensor var_46084_begin_0 = const()[name = string("op_46084_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46084_end_0 = const()[name = string("op_46084_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_46084_end_mask_0 = const()[name = string("op_46084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46084_cast_fp16 = slice_by_index(begin = var_46084_begin_0, end = var_46084_end_0, end_mask = var_46084_end_mask_0, x = query_61_cast_fp16)[name = string("op_46084_cast_fp16")]; + tensor var_46088_begin_0 = const()[name = string("op_46088_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46088_end_0 = const()[name = string("op_46088_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_46088_end_mask_0 = const()[name = string("op_46088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46088_cast_fp16 = slice_by_index(begin = var_46088_begin_0, end = var_46088_end_0, end_mask = var_46088_end_mask_0, x = query_61_cast_fp16)[name = string("op_46088_cast_fp16")]; + tensor var_46092_begin_0 = const()[name = string("op_46092_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46092_end_0 = const()[name = string("op_46092_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_46092_end_mask_0 = const()[name = string("op_46092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46092_cast_fp16 = slice_by_index(begin = var_46092_begin_0, end = var_46092_end_0, end_mask = var_46092_end_mask_0, x = query_61_cast_fp16)[name = string("op_46092_cast_fp16")]; + tensor var_46096_begin_0 = const()[name = string("op_46096_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46096_end_0 = const()[name = string("op_46096_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_46096_end_mask_0 = const()[name = string("op_46096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46096_cast_fp16 = slice_by_index(begin = var_46096_begin_0, end = var_46096_end_0, end_mask = var_46096_end_mask_0, x = query_61_cast_fp16)[name = string("op_46096_cast_fp16")]; + tensor var_46100_begin_0 = const()[name = string("op_46100_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46100_end_0 = const()[name = string("op_46100_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_46100_end_mask_0 = const()[name = string("op_46100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46100_cast_fp16 = slice_by_index(begin = var_46100_begin_0, end = var_46100_end_0, end_mask = var_46100_end_mask_0, x = query_61_cast_fp16)[name = string("op_46100_cast_fp16")]; + tensor var_46104_begin_0 = const()[name = string("op_46104_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46104_end_0 = const()[name = string("op_46104_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_46104_end_mask_0 = const()[name = string("op_46104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46104_cast_fp16 = slice_by_index(begin = var_46104_begin_0, end = var_46104_end_0, end_mask = var_46104_end_mask_0, x = query_61_cast_fp16)[name = string("op_46104_cast_fp16")]; + tensor var_46108_begin_0 = const()[name = string("op_46108_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46108_end_0 = const()[name = string("op_46108_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_46108_end_mask_0 = const()[name = string("op_46108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46108_cast_fp16 = slice_by_index(begin = var_46108_begin_0, end = var_46108_end_0, end_mask = var_46108_end_mask_0, x = query_61_cast_fp16)[name = string("op_46108_cast_fp16")]; + tensor var_46112_begin_0 = const()[name = string("op_46112_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46112_end_0 = const()[name = string("op_46112_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_46112_end_mask_0 = const()[name = string("op_46112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46112_cast_fp16 = slice_by_index(begin = var_46112_begin_0, end = var_46112_end_0, end_mask = var_46112_end_mask_0, x = query_61_cast_fp16)[name = string("op_46112_cast_fp16")]; + tensor var_46116_begin_0 = const()[name = string("op_46116_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46116_end_0 = const()[name = string("op_46116_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_46116_end_mask_0 = const()[name = string("op_46116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46116_cast_fp16 = slice_by_index(begin = var_46116_begin_0, end = var_46116_end_0, end_mask = var_46116_end_mask_0, x = query_61_cast_fp16)[name = string("op_46116_cast_fp16")]; + tensor var_46120_begin_0 = const()[name = string("op_46120_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46120_end_0 = const()[name = string("op_46120_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_46120_end_mask_0 = const()[name = string("op_46120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46120_cast_fp16 = slice_by_index(begin = var_46120_begin_0, end = var_46120_end_0, end_mask = var_46120_end_mask_0, x = query_61_cast_fp16)[name = string("op_46120_cast_fp16")]; + tensor var_46124_begin_0 = const()[name = string("op_46124_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46124_end_0 = const()[name = string("op_46124_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_46124_end_mask_0 = const()[name = string("op_46124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46124_cast_fp16 = slice_by_index(begin = var_46124_begin_0, end = var_46124_end_0, end_mask = var_46124_end_mask_0, x = query_61_cast_fp16)[name = string("op_46124_cast_fp16")]; + tensor var_46128_begin_0 = const()[name = string("op_46128_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46128_end_0 = const()[name = string("op_46128_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_46128_end_mask_0 = const()[name = string("op_46128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46128_cast_fp16 = slice_by_index(begin = var_46128_begin_0, end = var_46128_end_0, end_mask = var_46128_end_mask_0, x = query_61_cast_fp16)[name = string("op_46128_cast_fp16")]; + tensor var_46132_begin_0 = const()[name = string("op_46132_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46132_end_0 = const()[name = string("op_46132_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_46132_end_mask_0 = const()[name = string("op_46132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46132_cast_fp16 = slice_by_index(begin = var_46132_begin_0, end = var_46132_end_0, end_mask = var_46132_end_mask_0, x = query_61_cast_fp16)[name = string("op_46132_cast_fp16")]; + tensor var_46136_begin_0 = const()[name = string("op_46136_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46136_end_0 = const()[name = string("op_46136_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_46136_end_mask_0 = const()[name = string("op_46136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46136_cast_fp16 = slice_by_index(begin = var_46136_begin_0, end = var_46136_end_0, end_mask = var_46136_end_mask_0, x = query_61_cast_fp16)[name = string("op_46136_cast_fp16")]; + tensor var_46140_begin_0 = const()[name = string("op_46140_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46140_end_0 = const()[name = string("op_46140_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_46140_end_mask_0 = const()[name = string("op_46140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46140_cast_fp16 = slice_by_index(begin = var_46140_begin_0, end = var_46140_end_0, end_mask = var_46140_end_mask_0, x = query_61_cast_fp16)[name = string("op_46140_cast_fp16")]; + tensor var_46144_begin_0 = const()[name = string("op_46144_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46144_end_0 = const()[name = string("op_46144_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_46144_end_mask_0 = const()[name = string("op_46144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46144_cast_fp16 = slice_by_index(begin = var_46144_begin_0, end = var_46144_end_0, end_mask = var_46144_end_mask_0, x = query_61_cast_fp16)[name = string("op_46144_cast_fp16")]; + tensor var_46148_begin_0 = const()[name = string("op_46148_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46148_end_0 = const()[name = string("op_46148_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_46148_end_mask_0 = const()[name = string("op_46148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46148_cast_fp16 = slice_by_index(begin = var_46148_begin_0, end = var_46148_end_0, end_mask = var_46148_end_mask_0, x = query_61_cast_fp16)[name = string("op_46148_cast_fp16")]; + tensor var_46157_begin_0 = const()[name = string("op_46157_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46157_end_0 = const()[name = string("op_46157_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46157_end_mask_0 = const()[name = string("op_46157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46157_cast_fp16 = slice_by_index(begin = var_46157_begin_0, end = var_46157_end_0, end_mask = var_46157_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46157_cast_fp16")]; + tensor var_46164_begin_0 = const()[name = string("op_46164_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46164_end_0 = const()[name = string("op_46164_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46164_end_mask_0 = const()[name = string("op_46164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46164_cast_fp16 = slice_by_index(begin = var_46164_begin_0, end = var_46164_end_0, end_mask = var_46164_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46164_cast_fp16")]; + tensor var_46171_begin_0 = const()[name = string("op_46171_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46171_end_0 = const()[name = string("op_46171_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46171_end_mask_0 = const()[name = string("op_46171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46171_cast_fp16 = slice_by_index(begin = var_46171_begin_0, end = var_46171_end_0, end_mask = var_46171_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46171_cast_fp16")]; + tensor var_46178_begin_0 = const()[name = string("op_46178_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46178_end_0 = const()[name = string("op_46178_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46178_end_mask_0 = const()[name = string("op_46178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46178_cast_fp16 = slice_by_index(begin = var_46178_begin_0, end = var_46178_end_0, end_mask = var_46178_end_mask_0, x = var_46072_cast_fp16)[name = string("op_46178_cast_fp16")]; + tensor var_46185_begin_0 = const()[name = string("op_46185_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46185_end_0 = const()[name = string("op_46185_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46185_end_mask_0 = const()[name = string("op_46185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46185_cast_fp16 = slice_by_index(begin = var_46185_begin_0, end = var_46185_end_0, end_mask = var_46185_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46185_cast_fp16")]; + tensor var_46192_begin_0 = const()[name = string("op_46192_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46192_end_0 = const()[name = string("op_46192_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46192_end_mask_0 = const()[name = string("op_46192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46192_cast_fp16 = slice_by_index(begin = var_46192_begin_0, end = var_46192_end_0, end_mask = var_46192_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46192_cast_fp16")]; + tensor var_46199_begin_0 = const()[name = string("op_46199_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46199_end_0 = const()[name = string("op_46199_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46199_end_mask_0 = const()[name = string("op_46199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46199_cast_fp16 = slice_by_index(begin = var_46199_begin_0, end = var_46199_end_0, end_mask = var_46199_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46199_cast_fp16")]; + tensor var_46206_begin_0 = const()[name = string("op_46206_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46206_end_0 = const()[name = string("op_46206_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46206_end_mask_0 = const()[name = string("op_46206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46206_cast_fp16 = slice_by_index(begin = var_46206_begin_0, end = var_46206_end_0, end_mask = var_46206_end_mask_0, x = var_46076_cast_fp16)[name = string("op_46206_cast_fp16")]; + tensor var_46213_begin_0 = const()[name = string("op_46213_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46213_end_0 = const()[name = string("op_46213_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46213_end_mask_0 = const()[name = string("op_46213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46213_cast_fp16 = slice_by_index(begin = var_46213_begin_0, end = var_46213_end_0, end_mask = var_46213_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46213_cast_fp16")]; + tensor var_46220_begin_0 = const()[name = string("op_46220_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46220_end_0 = const()[name = string("op_46220_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46220_end_mask_0 = const()[name = string("op_46220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46220_cast_fp16 = slice_by_index(begin = var_46220_begin_0, end = var_46220_end_0, end_mask = var_46220_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46220_cast_fp16")]; + tensor var_46227_begin_0 = const()[name = string("op_46227_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46227_end_0 = const()[name = string("op_46227_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46227_end_mask_0 = const()[name = string("op_46227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46227_cast_fp16 = slice_by_index(begin = var_46227_begin_0, end = var_46227_end_0, end_mask = var_46227_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46227_cast_fp16")]; + tensor var_46234_begin_0 = const()[name = string("op_46234_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46234_end_0 = const()[name = string("op_46234_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46234_end_mask_0 = const()[name = string("op_46234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46234_cast_fp16 = slice_by_index(begin = var_46234_begin_0, end = var_46234_end_0, end_mask = var_46234_end_mask_0, x = var_46080_cast_fp16)[name = string("op_46234_cast_fp16")]; + tensor var_46241_begin_0 = const()[name = string("op_46241_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46241_end_0 = const()[name = string("op_46241_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46241_end_mask_0 = const()[name = string("op_46241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46241_cast_fp16 = slice_by_index(begin = var_46241_begin_0, end = var_46241_end_0, end_mask = var_46241_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46241_cast_fp16")]; + tensor var_46248_begin_0 = const()[name = string("op_46248_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46248_end_0 = const()[name = string("op_46248_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46248_end_mask_0 = const()[name = string("op_46248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46248_cast_fp16 = slice_by_index(begin = var_46248_begin_0, end = var_46248_end_0, end_mask = var_46248_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46248_cast_fp16")]; + tensor var_46255_begin_0 = const()[name = string("op_46255_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46255_end_0 = const()[name = string("op_46255_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46255_end_mask_0 = const()[name = string("op_46255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46255_cast_fp16 = slice_by_index(begin = var_46255_begin_0, end = var_46255_end_0, end_mask = var_46255_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46255_cast_fp16")]; + tensor var_46262_begin_0 = const()[name = string("op_46262_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46262_end_0 = const()[name = string("op_46262_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46262_end_mask_0 = const()[name = string("op_46262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46262_cast_fp16 = slice_by_index(begin = var_46262_begin_0, end = var_46262_end_0, end_mask = var_46262_end_mask_0, x = var_46084_cast_fp16)[name = string("op_46262_cast_fp16")]; + tensor var_46269_begin_0 = const()[name = string("op_46269_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46269_end_0 = const()[name = string("op_46269_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46269_end_mask_0 = const()[name = string("op_46269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46269_cast_fp16 = slice_by_index(begin = var_46269_begin_0, end = var_46269_end_0, end_mask = var_46269_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46269_cast_fp16")]; + tensor var_46276_begin_0 = const()[name = string("op_46276_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46276_end_0 = const()[name = string("op_46276_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46276_end_mask_0 = const()[name = string("op_46276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46276_cast_fp16 = slice_by_index(begin = var_46276_begin_0, end = var_46276_end_0, end_mask = var_46276_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46276_cast_fp16")]; + tensor var_46283_begin_0 = const()[name = string("op_46283_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46283_end_0 = const()[name = string("op_46283_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46283_end_mask_0 = const()[name = string("op_46283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46283_cast_fp16 = slice_by_index(begin = var_46283_begin_0, end = var_46283_end_0, end_mask = var_46283_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46283_cast_fp16")]; + tensor var_46290_begin_0 = const()[name = string("op_46290_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46290_end_0 = const()[name = string("op_46290_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46290_end_mask_0 = const()[name = string("op_46290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46290_cast_fp16 = slice_by_index(begin = var_46290_begin_0, end = var_46290_end_0, end_mask = var_46290_end_mask_0, x = var_46088_cast_fp16)[name = string("op_46290_cast_fp16")]; + tensor var_46297_begin_0 = const()[name = string("op_46297_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46297_end_0 = const()[name = string("op_46297_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46297_end_mask_0 = const()[name = string("op_46297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46297_cast_fp16 = slice_by_index(begin = var_46297_begin_0, end = var_46297_end_0, end_mask = var_46297_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46297_cast_fp16")]; + tensor var_46304_begin_0 = const()[name = string("op_46304_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46304_end_0 = const()[name = string("op_46304_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46304_end_mask_0 = const()[name = string("op_46304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46304_cast_fp16 = slice_by_index(begin = var_46304_begin_0, end = var_46304_end_0, end_mask = var_46304_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46304_cast_fp16")]; + tensor var_46311_begin_0 = const()[name = string("op_46311_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46311_end_0 = const()[name = string("op_46311_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46311_end_mask_0 = const()[name = string("op_46311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46311_cast_fp16 = slice_by_index(begin = var_46311_begin_0, end = var_46311_end_0, end_mask = var_46311_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46311_cast_fp16")]; + tensor var_46318_begin_0 = const()[name = string("op_46318_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46318_end_0 = const()[name = string("op_46318_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46318_end_mask_0 = const()[name = string("op_46318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46318_cast_fp16 = slice_by_index(begin = var_46318_begin_0, end = var_46318_end_0, end_mask = var_46318_end_mask_0, x = var_46092_cast_fp16)[name = string("op_46318_cast_fp16")]; + tensor var_46325_begin_0 = const()[name = string("op_46325_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46325_end_0 = const()[name = string("op_46325_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46325_end_mask_0 = const()[name = string("op_46325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46325_cast_fp16 = slice_by_index(begin = var_46325_begin_0, end = var_46325_end_0, end_mask = var_46325_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46325_cast_fp16")]; + tensor var_46332_begin_0 = const()[name = string("op_46332_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46332_end_0 = const()[name = string("op_46332_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46332_end_mask_0 = const()[name = string("op_46332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46332_cast_fp16 = slice_by_index(begin = var_46332_begin_0, end = var_46332_end_0, end_mask = var_46332_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46332_cast_fp16")]; + tensor var_46339_begin_0 = const()[name = string("op_46339_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46339_end_0 = const()[name = string("op_46339_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46339_end_mask_0 = const()[name = string("op_46339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46339_cast_fp16 = slice_by_index(begin = var_46339_begin_0, end = var_46339_end_0, end_mask = var_46339_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46339_cast_fp16")]; + tensor var_46346_begin_0 = const()[name = string("op_46346_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46346_end_0 = const()[name = string("op_46346_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46346_end_mask_0 = const()[name = string("op_46346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46346_cast_fp16 = slice_by_index(begin = var_46346_begin_0, end = var_46346_end_0, end_mask = var_46346_end_mask_0, x = var_46096_cast_fp16)[name = string("op_46346_cast_fp16")]; + tensor var_46353_begin_0 = const()[name = string("op_46353_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46353_end_0 = const()[name = string("op_46353_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46353_end_mask_0 = const()[name = string("op_46353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46353_cast_fp16 = slice_by_index(begin = var_46353_begin_0, end = var_46353_end_0, end_mask = var_46353_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46353_cast_fp16")]; + tensor var_46360_begin_0 = const()[name = string("op_46360_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46360_end_0 = const()[name = string("op_46360_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46360_end_mask_0 = const()[name = string("op_46360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46360_cast_fp16 = slice_by_index(begin = var_46360_begin_0, end = var_46360_end_0, end_mask = var_46360_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46360_cast_fp16")]; + tensor var_46367_begin_0 = const()[name = string("op_46367_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46367_end_0 = const()[name = string("op_46367_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46367_end_mask_0 = const()[name = string("op_46367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46367_cast_fp16 = slice_by_index(begin = var_46367_begin_0, end = var_46367_end_0, end_mask = var_46367_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46367_cast_fp16")]; + tensor var_46374_begin_0 = const()[name = string("op_46374_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46374_end_0 = const()[name = string("op_46374_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46374_end_mask_0 = const()[name = string("op_46374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46374_cast_fp16 = slice_by_index(begin = var_46374_begin_0, end = var_46374_end_0, end_mask = var_46374_end_mask_0, x = var_46100_cast_fp16)[name = string("op_46374_cast_fp16")]; + tensor var_46381_begin_0 = const()[name = string("op_46381_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46381_end_0 = const()[name = string("op_46381_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46381_end_mask_0 = const()[name = string("op_46381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46381_cast_fp16 = slice_by_index(begin = var_46381_begin_0, end = var_46381_end_0, end_mask = var_46381_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46381_cast_fp16")]; + tensor var_46388_begin_0 = const()[name = string("op_46388_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46388_end_0 = const()[name = string("op_46388_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46388_end_mask_0 = const()[name = string("op_46388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46388_cast_fp16 = slice_by_index(begin = var_46388_begin_0, end = var_46388_end_0, end_mask = var_46388_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46388_cast_fp16")]; + tensor var_46395_begin_0 = const()[name = string("op_46395_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46395_end_0 = const()[name = string("op_46395_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46395_end_mask_0 = const()[name = string("op_46395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46395_cast_fp16 = slice_by_index(begin = var_46395_begin_0, end = var_46395_end_0, end_mask = var_46395_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46395_cast_fp16")]; + tensor var_46402_begin_0 = const()[name = string("op_46402_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46402_end_0 = const()[name = string("op_46402_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46402_end_mask_0 = const()[name = string("op_46402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46402_cast_fp16 = slice_by_index(begin = var_46402_begin_0, end = var_46402_end_0, end_mask = var_46402_end_mask_0, x = var_46104_cast_fp16)[name = string("op_46402_cast_fp16")]; + tensor var_46409_begin_0 = const()[name = string("op_46409_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46409_end_0 = const()[name = string("op_46409_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46409_end_mask_0 = const()[name = string("op_46409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46409_cast_fp16 = slice_by_index(begin = var_46409_begin_0, end = var_46409_end_0, end_mask = var_46409_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46409_cast_fp16")]; + tensor var_46416_begin_0 = const()[name = string("op_46416_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46416_end_0 = const()[name = string("op_46416_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46416_end_mask_0 = const()[name = string("op_46416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46416_cast_fp16 = slice_by_index(begin = var_46416_begin_0, end = var_46416_end_0, end_mask = var_46416_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46416_cast_fp16")]; + tensor var_46423_begin_0 = const()[name = string("op_46423_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46423_end_0 = const()[name = string("op_46423_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46423_end_mask_0 = const()[name = string("op_46423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46423_cast_fp16 = slice_by_index(begin = var_46423_begin_0, end = var_46423_end_0, end_mask = var_46423_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46423_cast_fp16")]; + tensor var_46430_begin_0 = const()[name = string("op_46430_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46430_end_0 = const()[name = string("op_46430_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46430_end_mask_0 = const()[name = string("op_46430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46430_cast_fp16 = slice_by_index(begin = var_46430_begin_0, end = var_46430_end_0, end_mask = var_46430_end_mask_0, x = var_46108_cast_fp16)[name = string("op_46430_cast_fp16")]; + tensor var_46437_begin_0 = const()[name = string("op_46437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46437_end_0 = const()[name = string("op_46437_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46437_end_mask_0 = const()[name = string("op_46437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46437_cast_fp16 = slice_by_index(begin = var_46437_begin_0, end = var_46437_end_0, end_mask = var_46437_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46437_cast_fp16")]; + tensor var_46444_begin_0 = const()[name = string("op_46444_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46444_end_0 = const()[name = string("op_46444_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46444_end_mask_0 = const()[name = string("op_46444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46444_cast_fp16 = slice_by_index(begin = var_46444_begin_0, end = var_46444_end_0, end_mask = var_46444_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46444_cast_fp16")]; + tensor var_46451_begin_0 = const()[name = string("op_46451_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46451_end_0 = const()[name = string("op_46451_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46451_end_mask_0 = const()[name = string("op_46451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46451_cast_fp16 = slice_by_index(begin = var_46451_begin_0, end = var_46451_end_0, end_mask = var_46451_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46451_cast_fp16")]; + tensor var_46458_begin_0 = const()[name = string("op_46458_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46458_end_0 = const()[name = string("op_46458_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46458_end_mask_0 = const()[name = string("op_46458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46458_cast_fp16 = slice_by_index(begin = var_46458_begin_0, end = var_46458_end_0, end_mask = var_46458_end_mask_0, x = var_46112_cast_fp16)[name = string("op_46458_cast_fp16")]; + tensor var_46465_begin_0 = const()[name = string("op_46465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46465_end_0 = const()[name = string("op_46465_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46465_end_mask_0 = const()[name = string("op_46465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46465_cast_fp16 = slice_by_index(begin = var_46465_begin_0, end = var_46465_end_0, end_mask = var_46465_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46465_cast_fp16")]; + tensor var_46472_begin_0 = const()[name = string("op_46472_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46472_end_0 = const()[name = string("op_46472_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46472_end_mask_0 = const()[name = string("op_46472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46472_cast_fp16 = slice_by_index(begin = var_46472_begin_0, end = var_46472_end_0, end_mask = var_46472_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46472_cast_fp16")]; + tensor var_46479_begin_0 = const()[name = string("op_46479_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46479_end_0 = const()[name = string("op_46479_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46479_end_mask_0 = const()[name = string("op_46479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46479_cast_fp16 = slice_by_index(begin = var_46479_begin_0, end = var_46479_end_0, end_mask = var_46479_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46479_cast_fp16")]; + tensor var_46486_begin_0 = const()[name = string("op_46486_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46486_end_0 = const()[name = string("op_46486_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46486_end_mask_0 = const()[name = string("op_46486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46486_cast_fp16 = slice_by_index(begin = var_46486_begin_0, end = var_46486_end_0, end_mask = var_46486_end_mask_0, x = var_46116_cast_fp16)[name = string("op_46486_cast_fp16")]; + tensor var_46493_begin_0 = const()[name = string("op_46493_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46493_end_0 = const()[name = string("op_46493_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46493_end_mask_0 = const()[name = string("op_46493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46493_cast_fp16 = slice_by_index(begin = var_46493_begin_0, end = var_46493_end_0, end_mask = var_46493_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46493_cast_fp16")]; + tensor var_46500_begin_0 = const()[name = string("op_46500_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46500_end_0 = const()[name = string("op_46500_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46500_end_mask_0 = const()[name = string("op_46500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46500_cast_fp16 = slice_by_index(begin = var_46500_begin_0, end = var_46500_end_0, end_mask = var_46500_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46500_cast_fp16")]; + tensor var_46507_begin_0 = const()[name = string("op_46507_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46507_end_0 = const()[name = string("op_46507_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46507_end_mask_0 = const()[name = string("op_46507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46507_cast_fp16 = slice_by_index(begin = var_46507_begin_0, end = var_46507_end_0, end_mask = var_46507_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46507_cast_fp16")]; + tensor var_46514_begin_0 = const()[name = string("op_46514_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46514_end_0 = const()[name = string("op_46514_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46514_end_mask_0 = const()[name = string("op_46514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46514_cast_fp16 = slice_by_index(begin = var_46514_begin_0, end = var_46514_end_0, end_mask = var_46514_end_mask_0, x = var_46120_cast_fp16)[name = string("op_46514_cast_fp16")]; + tensor var_46521_begin_0 = const()[name = string("op_46521_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46521_end_0 = const()[name = string("op_46521_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46521_end_mask_0 = const()[name = string("op_46521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46521_cast_fp16 = slice_by_index(begin = var_46521_begin_0, end = var_46521_end_0, end_mask = var_46521_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46521_cast_fp16")]; + tensor var_46528_begin_0 = const()[name = string("op_46528_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46528_end_0 = const()[name = string("op_46528_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46528_end_mask_0 = const()[name = string("op_46528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46528_cast_fp16 = slice_by_index(begin = var_46528_begin_0, end = var_46528_end_0, end_mask = var_46528_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46528_cast_fp16")]; + tensor var_46535_begin_0 = const()[name = string("op_46535_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46535_end_0 = const()[name = string("op_46535_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46535_end_mask_0 = const()[name = string("op_46535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46535_cast_fp16 = slice_by_index(begin = var_46535_begin_0, end = var_46535_end_0, end_mask = var_46535_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46535_cast_fp16")]; + tensor var_46542_begin_0 = const()[name = string("op_46542_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46542_end_0 = const()[name = string("op_46542_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46542_end_mask_0 = const()[name = string("op_46542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46542_cast_fp16 = slice_by_index(begin = var_46542_begin_0, end = var_46542_end_0, end_mask = var_46542_end_mask_0, x = var_46124_cast_fp16)[name = string("op_46542_cast_fp16")]; + tensor var_46549_begin_0 = const()[name = string("op_46549_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46549_end_0 = const()[name = string("op_46549_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46549_end_mask_0 = const()[name = string("op_46549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46549_cast_fp16 = slice_by_index(begin = var_46549_begin_0, end = var_46549_end_0, end_mask = var_46549_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46549_cast_fp16")]; + tensor var_46556_begin_0 = const()[name = string("op_46556_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46556_end_0 = const()[name = string("op_46556_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46556_end_mask_0 = const()[name = string("op_46556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46556_cast_fp16 = slice_by_index(begin = var_46556_begin_0, end = var_46556_end_0, end_mask = var_46556_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46556_cast_fp16")]; + tensor var_46563_begin_0 = const()[name = string("op_46563_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46563_end_0 = const()[name = string("op_46563_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46563_end_mask_0 = const()[name = string("op_46563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46563_cast_fp16 = slice_by_index(begin = var_46563_begin_0, end = var_46563_end_0, end_mask = var_46563_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46563_cast_fp16")]; + tensor var_46570_begin_0 = const()[name = string("op_46570_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46570_end_0 = const()[name = string("op_46570_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46570_end_mask_0 = const()[name = string("op_46570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46570_cast_fp16 = slice_by_index(begin = var_46570_begin_0, end = var_46570_end_0, end_mask = var_46570_end_mask_0, x = var_46128_cast_fp16)[name = string("op_46570_cast_fp16")]; + tensor var_46577_begin_0 = const()[name = string("op_46577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46577_end_0 = const()[name = string("op_46577_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46577_end_mask_0 = const()[name = string("op_46577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46577_cast_fp16 = slice_by_index(begin = var_46577_begin_0, end = var_46577_end_0, end_mask = var_46577_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46577_cast_fp16")]; + tensor var_46584_begin_0 = const()[name = string("op_46584_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46584_end_0 = const()[name = string("op_46584_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46584_end_mask_0 = const()[name = string("op_46584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46584_cast_fp16 = slice_by_index(begin = var_46584_begin_0, end = var_46584_end_0, end_mask = var_46584_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46584_cast_fp16")]; + tensor var_46591_begin_0 = const()[name = string("op_46591_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46591_end_0 = const()[name = string("op_46591_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46591_end_mask_0 = const()[name = string("op_46591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46591_cast_fp16 = slice_by_index(begin = var_46591_begin_0, end = var_46591_end_0, end_mask = var_46591_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46591_cast_fp16")]; + tensor var_46598_begin_0 = const()[name = string("op_46598_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46598_end_0 = const()[name = string("op_46598_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46598_end_mask_0 = const()[name = string("op_46598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46598_cast_fp16 = slice_by_index(begin = var_46598_begin_0, end = var_46598_end_0, end_mask = var_46598_end_mask_0, x = var_46132_cast_fp16)[name = string("op_46598_cast_fp16")]; + tensor var_46605_begin_0 = const()[name = string("op_46605_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46605_end_0 = const()[name = string("op_46605_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46605_end_mask_0 = const()[name = string("op_46605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46605_cast_fp16 = slice_by_index(begin = var_46605_begin_0, end = var_46605_end_0, end_mask = var_46605_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46605_cast_fp16")]; + tensor var_46612_begin_0 = const()[name = string("op_46612_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46612_end_0 = const()[name = string("op_46612_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46612_end_mask_0 = const()[name = string("op_46612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46612_cast_fp16 = slice_by_index(begin = var_46612_begin_0, end = var_46612_end_0, end_mask = var_46612_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46612_cast_fp16")]; + tensor var_46619_begin_0 = const()[name = string("op_46619_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46619_end_0 = const()[name = string("op_46619_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46619_end_mask_0 = const()[name = string("op_46619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46619_cast_fp16 = slice_by_index(begin = var_46619_begin_0, end = var_46619_end_0, end_mask = var_46619_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46619_cast_fp16")]; + tensor var_46626_begin_0 = const()[name = string("op_46626_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46626_end_0 = const()[name = string("op_46626_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46626_end_mask_0 = const()[name = string("op_46626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46626_cast_fp16 = slice_by_index(begin = var_46626_begin_0, end = var_46626_end_0, end_mask = var_46626_end_mask_0, x = var_46136_cast_fp16)[name = string("op_46626_cast_fp16")]; + tensor var_46633_begin_0 = const()[name = string("op_46633_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46633_end_0 = const()[name = string("op_46633_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46633_end_mask_0 = const()[name = string("op_46633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46633_cast_fp16 = slice_by_index(begin = var_46633_begin_0, end = var_46633_end_0, end_mask = var_46633_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46633_cast_fp16")]; + tensor var_46640_begin_0 = const()[name = string("op_46640_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46640_end_0 = const()[name = string("op_46640_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46640_end_mask_0 = const()[name = string("op_46640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46640_cast_fp16 = slice_by_index(begin = var_46640_begin_0, end = var_46640_end_0, end_mask = var_46640_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46640_cast_fp16")]; + tensor var_46647_begin_0 = const()[name = string("op_46647_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46647_end_0 = const()[name = string("op_46647_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46647_end_mask_0 = const()[name = string("op_46647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46647_cast_fp16 = slice_by_index(begin = var_46647_begin_0, end = var_46647_end_0, end_mask = var_46647_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46647_cast_fp16")]; + tensor var_46654_begin_0 = const()[name = string("op_46654_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46654_end_0 = const()[name = string("op_46654_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46654_end_mask_0 = const()[name = string("op_46654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46654_cast_fp16 = slice_by_index(begin = var_46654_begin_0, end = var_46654_end_0, end_mask = var_46654_end_mask_0, x = var_46140_cast_fp16)[name = string("op_46654_cast_fp16")]; + tensor var_46661_begin_0 = const()[name = string("op_46661_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46661_end_0 = const()[name = string("op_46661_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46661_end_mask_0 = const()[name = string("op_46661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46661_cast_fp16 = slice_by_index(begin = var_46661_begin_0, end = var_46661_end_0, end_mask = var_46661_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46661_cast_fp16")]; + tensor var_46668_begin_0 = const()[name = string("op_46668_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46668_end_0 = const()[name = string("op_46668_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46668_end_mask_0 = const()[name = string("op_46668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46668_cast_fp16 = slice_by_index(begin = var_46668_begin_0, end = var_46668_end_0, end_mask = var_46668_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46668_cast_fp16")]; + tensor var_46675_begin_0 = const()[name = string("op_46675_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46675_end_0 = const()[name = string("op_46675_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46675_end_mask_0 = const()[name = string("op_46675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46675_cast_fp16 = slice_by_index(begin = var_46675_begin_0, end = var_46675_end_0, end_mask = var_46675_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46675_cast_fp16")]; + tensor var_46682_begin_0 = const()[name = string("op_46682_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46682_end_0 = const()[name = string("op_46682_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46682_end_mask_0 = const()[name = string("op_46682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46682_cast_fp16 = slice_by_index(begin = var_46682_begin_0, end = var_46682_end_0, end_mask = var_46682_end_mask_0, x = var_46144_cast_fp16)[name = string("op_46682_cast_fp16")]; + tensor var_46689_begin_0 = const()[name = string("op_46689_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46689_end_0 = const()[name = string("op_46689_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46689_end_mask_0 = const()[name = string("op_46689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46689_cast_fp16 = slice_by_index(begin = var_46689_begin_0, end = var_46689_end_0, end_mask = var_46689_end_mask_0, x = var_46148_cast_fp16)[name = string("op_46689_cast_fp16")]; + tensor var_46696_begin_0 = const()[name = string("op_46696_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46696_end_0 = const()[name = string("op_46696_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46696_end_mask_0 = const()[name = string("op_46696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46696_cast_fp16 = slice_by_index(begin = var_46696_begin_0, end = var_46696_end_0, end_mask = var_46696_end_mask_0, x = var_46148_cast_fp16)[name = string("op_46696_cast_fp16")]; + tensor var_46703_begin_0 = const()[name = string("op_46703_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46703_end_0 = const()[name = string("op_46703_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46703_end_mask_0 = const()[name = string("op_46703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46703_cast_fp16 = slice_by_index(begin = var_46703_begin_0, end = var_46703_end_0, end_mask = var_46703_end_mask_0, x = var_46148_cast_fp16)[name = string("op_46703_cast_fp16")]; + tensor var_46710_begin_0 = const()[name = string("op_46710_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46710_end_0 = const()[name = string("op_46710_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46710_end_mask_0 = const()[name = string("op_46710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46710_cast_fp16 = slice_by_index(begin = var_46710_begin_0, end = var_46710_end_0, end_mask = var_46710_end_mask_0, x = var_46148_cast_fp16)[name = string("op_46710_cast_fp16")]; + tensor k_61_perm_0 = const()[name = string("k_61_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_46715_begin_0 = const()[name = string("op_46715_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46715_end_0 = const()[name = string("op_46715_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_46715_end_mask_0 = const()[name = string("op_46715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_61_cast_fp16 = transpose(perm = k_61_perm_0, x = key_61_cast_fp16)[name = string("transpose_1")]; + tensor var_46715_cast_fp16 = slice_by_index(begin = var_46715_begin_0, end = var_46715_end_0, end_mask = var_46715_end_mask_0, x = k_61_cast_fp16)[name = string("op_46715_cast_fp16")]; + tensor var_46719_begin_0 = const()[name = string("op_46719_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_46719_end_0 = const()[name = string("op_46719_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_46719_end_mask_0 = const()[name = string("op_46719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46719_cast_fp16 = slice_by_index(begin = var_46719_begin_0, end = var_46719_end_0, end_mask = var_46719_end_mask_0, x = k_61_cast_fp16)[name = string("op_46719_cast_fp16")]; + tensor var_46723_begin_0 = const()[name = string("op_46723_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_46723_end_0 = const()[name = string("op_46723_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_46723_end_mask_0 = const()[name = string("op_46723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46723_cast_fp16 = slice_by_index(begin = var_46723_begin_0, end = var_46723_end_0, end_mask = var_46723_end_mask_0, x = k_61_cast_fp16)[name = string("op_46723_cast_fp16")]; + tensor var_46727_begin_0 = const()[name = string("op_46727_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_46727_end_0 = const()[name = string("op_46727_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_46727_end_mask_0 = const()[name = string("op_46727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46727_cast_fp16 = slice_by_index(begin = var_46727_begin_0, end = var_46727_end_0, end_mask = var_46727_end_mask_0, x = k_61_cast_fp16)[name = string("op_46727_cast_fp16")]; + tensor var_46731_begin_0 = const()[name = string("op_46731_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_46731_end_0 = const()[name = string("op_46731_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_46731_end_mask_0 = const()[name = string("op_46731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46731_cast_fp16 = slice_by_index(begin = var_46731_begin_0, end = var_46731_end_0, end_mask = var_46731_end_mask_0, x = k_61_cast_fp16)[name = string("op_46731_cast_fp16")]; + tensor var_46735_begin_0 = const()[name = string("op_46735_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_46735_end_0 = const()[name = string("op_46735_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_46735_end_mask_0 = const()[name = string("op_46735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46735_cast_fp16 = slice_by_index(begin = var_46735_begin_0, end = var_46735_end_0, end_mask = var_46735_end_mask_0, x = k_61_cast_fp16)[name = string("op_46735_cast_fp16")]; + tensor var_46739_begin_0 = const()[name = string("op_46739_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_46739_end_0 = const()[name = string("op_46739_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_46739_end_mask_0 = const()[name = string("op_46739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46739_cast_fp16 = slice_by_index(begin = var_46739_begin_0, end = var_46739_end_0, end_mask = var_46739_end_mask_0, x = k_61_cast_fp16)[name = string("op_46739_cast_fp16")]; + tensor var_46743_begin_0 = const()[name = string("op_46743_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_46743_end_0 = const()[name = string("op_46743_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_46743_end_mask_0 = const()[name = string("op_46743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46743_cast_fp16 = slice_by_index(begin = var_46743_begin_0, end = var_46743_end_0, end_mask = var_46743_end_mask_0, x = k_61_cast_fp16)[name = string("op_46743_cast_fp16")]; + tensor var_46747_begin_0 = const()[name = string("op_46747_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_46747_end_0 = const()[name = string("op_46747_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_46747_end_mask_0 = const()[name = string("op_46747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46747_cast_fp16 = slice_by_index(begin = var_46747_begin_0, end = var_46747_end_0, end_mask = var_46747_end_mask_0, x = k_61_cast_fp16)[name = string("op_46747_cast_fp16")]; + tensor var_46751_begin_0 = const()[name = string("op_46751_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_46751_end_0 = const()[name = string("op_46751_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_46751_end_mask_0 = const()[name = string("op_46751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46751_cast_fp16 = slice_by_index(begin = var_46751_begin_0, end = var_46751_end_0, end_mask = var_46751_end_mask_0, x = k_61_cast_fp16)[name = string("op_46751_cast_fp16")]; + tensor var_46755_begin_0 = const()[name = string("op_46755_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_46755_end_0 = const()[name = string("op_46755_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_46755_end_mask_0 = const()[name = string("op_46755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46755_cast_fp16 = slice_by_index(begin = var_46755_begin_0, end = var_46755_end_0, end_mask = var_46755_end_mask_0, x = k_61_cast_fp16)[name = string("op_46755_cast_fp16")]; + tensor var_46759_begin_0 = const()[name = string("op_46759_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_46759_end_0 = const()[name = string("op_46759_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_46759_end_mask_0 = const()[name = string("op_46759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46759_cast_fp16 = slice_by_index(begin = var_46759_begin_0, end = var_46759_end_0, end_mask = var_46759_end_mask_0, x = k_61_cast_fp16)[name = string("op_46759_cast_fp16")]; + tensor var_46763_begin_0 = const()[name = string("op_46763_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_46763_end_0 = const()[name = string("op_46763_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_46763_end_mask_0 = const()[name = string("op_46763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46763_cast_fp16 = slice_by_index(begin = var_46763_begin_0, end = var_46763_end_0, end_mask = var_46763_end_mask_0, x = k_61_cast_fp16)[name = string("op_46763_cast_fp16")]; + tensor var_46767_begin_0 = const()[name = string("op_46767_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_46767_end_0 = const()[name = string("op_46767_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_46767_end_mask_0 = const()[name = string("op_46767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46767_cast_fp16 = slice_by_index(begin = var_46767_begin_0, end = var_46767_end_0, end_mask = var_46767_end_mask_0, x = k_61_cast_fp16)[name = string("op_46767_cast_fp16")]; + tensor var_46771_begin_0 = const()[name = string("op_46771_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_46771_end_0 = const()[name = string("op_46771_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_46771_end_mask_0 = const()[name = string("op_46771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46771_cast_fp16 = slice_by_index(begin = var_46771_begin_0, end = var_46771_end_0, end_mask = var_46771_end_mask_0, x = k_61_cast_fp16)[name = string("op_46771_cast_fp16")]; + tensor var_46775_begin_0 = const()[name = string("op_46775_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_46775_end_0 = const()[name = string("op_46775_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_46775_end_mask_0 = const()[name = string("op_46775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46775_cast_fp16 = slice_by_index(begin = var_46775_begin_0, end = var_46775_end_0, end_mask = var_46775_end_mask_0, x = k_61_cast_fp16)[name = string("op_46775_cast_fp16")]; + tensor var_46779_begin_0 = const()[name = string("op_46779_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_46779_end_0 = const()[name = string("op_46779_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_46779_end_mask_0 = const()[name = string("op_46779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46779_cast_fp16 = slice_by_index(begin = var_46779_begin_0, end = var_46779_end_0, end_mask = var_46779_end_mask_0, x = k_61_cast_fp16)[name = string("op_46779_cast_fp16")]; + tensor var_46783_begin_0 = const()[name = string("op_46783_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_46783_end_0 = const()[name = string("op_46783_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_46783_end_mask_0 = const()[name = string("op_46783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46783_cast_fp16 = slice_by_index(begin = var_46783_begin_0, end = var_46783_end_0, end_mask = var_46783_end_mask_0, x = k_61_cast_fp16)[name = string("op_46783_cast_fp16")]; + tensor var_46787_begin_0 = const()[name = string("op_46787_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_46787_end_0 = const()[name = string("op_46787_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_46787_end_mask_0 = const()[name = string("op_46787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46787_cast_fp16 = slice_by_index(begin = var_46787_begin_0, end = var_46787_end_0, end_mask = var_46787_end_mask_0, x = k_61_cast_fp16)[name = string("op_46787_cast_fp16")]; + tensor var_46791_begin_0 = const()[name = string("op_46791_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_46791_end_0 = const()[name = string("op_46791_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_46791_end_mask_0 = const()[name = string("op_46791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46791_cast_fp16 = slice_by_index(begin = var_46791_begin_0, end = var_46791_end_0, end_mask = var_46791_end_mask_0, x = k_61_cast_fp16)[name = string("op_46791_cast_fp16")]; + tensor var_46793_begin_0 = const()[name = string("op_46793_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46793_end_0 = const()[name = string("op_46793_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46793_end_mask_0 = const()[name = string("op_46793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46793_cast_fp16 = slice_by_index(begin = var_46793_begin_0, end = var_46793_end_0, end_mask = var_46793_end_mask_0, x = value_61_cast_fp16)[name = string("op_46793_cast_fp16")]; + tensor var_46797_begin_0 = const()[name = string("op_46797_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46797_end_0 = const()[name = string("op_46797_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_46797_end_mask_0 = const()[name = string("op_46797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46797_cast_fp16 = slice_by_index(begin = var_46797_begin_0, end = var_46797_end_0, end_mask = var_46797_end_mask_0, x = value_61_cast_fp16)[name = string("op_46797_cast_fp16")]; + tensor var_46801_begin_0 = const()[name = string("op_46801_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46801_end_0 = const()[name = string("op_46801_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_46801_end_mask_0 = const()[name = string("op_46801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46801_cast_fp16 = slice_by_index(begin = var_46801_begin_0, end = var_46801_end_0, end_mask = var_46801_end_mask_0, x = value_61_cast_fp16)[name = string("op_46801_cast_fp16")]; + tensor var_46805_begin_0 = const()[name = string("op_46805_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46805_end_0 = const()[name = string("op_46805_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_46805_end_mask_0 = const()[name = string("op_46805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46805_cast_fp16 = slice_by_index(begin = var_46805_begin_0, end = var_46805_end_0, end_mask = var_46805_end_mask_0, x = value_61_cast_fp16)[name = string("op_46805_cast_fp16")]; + tensor var_46809_begin_0 = const()[name = string("op_46809_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46809_end_0 = const()[name = string("op_46809_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_46809_end_mask_0 = const()[name = string("op_46809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46809_cast_fp16 = slice_by_index(begin = var_46809_begin_0, end = var_46809_end_0, end_mask = var_46809_end_mask_0, x = value_61_cast_fp16)[name = string("op_46809_cast_fp16")]; + tensor var_46813_begin_0 = const()[name = string("op_46813_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46813_end_0 = const()[name = string("op_46813_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_46813_end_mask_0 = const()[name = string("op_46813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46813_cast_fp16 = slice_by_index(begin = var_46813_begin_0, end = var_46813_end_0, end_mask = var_46813_end_mask_0, x = value_61_cast_fp16)[name = string("op_46813_cast_fp16")]; + tensor var_46817_begin_0 = const()[name = string("op_46817_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46817_end_0 = const()[name = string("op_46817_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_46817_end_mask_0 = const()[name = string("op_46817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46817_cast_fp16 = slice_by_index(begin = var_46817_begin_0, end = var_46817_end_0, end_mask = var_46817_end_mask_0, x = value_61_cast_fp16)[name = string("op_46817_cast_fp16")]; + tensor var_46821_begin_0 = const()[name = string("op_46821_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46821_end_0 = const()[name = string("op_46821_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_46821_end_mask_0 = const()[name = string("op_46821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46821_cast_fp16 = slice_by_index(begin = var_46821_begin_0, end = var_46821_end_0, end_mask = var_46821_end_mask_0, x = value_61_cast_fp16)[name = string("op_46821_cast_fp16")]; + tensor var_46825_begin_0 = const()[name = string("op_46825_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46825_end_0 = const()[name = string("op_46825_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_46825_end_mask_0 = const()[name = string("op_46825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46825_cast_fp16 = slice_by_index(begin = var_46825_begin_0, end = var_46825_end_0, end_mask = var_46825_end_mask_0, x = value_61_cast_fp16)[name = string("op_46825_cast_fp16")]; + tensor var_46829_begin_0 = const()[name = string("op_46829_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46829_end_0 = const()[name = string("op_46829_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_46829_end_mask_0 = const()[name = string("op_46829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46829_cast_fp16 = slice_by_index(begin = var_46829_begin_0, end = var_46829_end_0, end_mask = var_46829_end_mask_0, x = value_61_cast_fp16)[name = string("op_46829_cast_fp16")]; + tensor var_46833_begin_0 = const()[name = string("op_46833_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46833_end_0 = const()[name = string("op_46833_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_46833_end_mask_0 = const()[name = string("op_46833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46833_cast_fp16 = slice_by_index(begin = var_46833_begin_0, end = var_46833_end_0, end_mask = var_46833_end_mask_0, x = value_61_cast_fp16)[name = string("op_46833_cast_fp16")]; + tensor var_46837_begin_0 = const()[name = string("op_46837_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46837_end_0 = const()[name = string("op_46837_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_46837_end_mask_0 = const()[name = string("op_46837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46837_cast_fp16 = slice_by_index(begin = var_46837_begin_0, end = var_46837_end_0, end_mask = var_46837_end_mask_0, x = value_61_cast_fp16)[name = string("op_46837_cast_fp16")]; + tensor var_46841_begin_0 = const()[name = string("op_46841_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46841_end_0 = const()[name = string("op_46841_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_46841_end_mask_0 = const()[name = string("op_46841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46841_cast_fp16 = slice_by_index(begin = var_46841_begin_0, end = var_46841_end_0, end_mask = var_46841_end_mask_0, x = value_61_cast_fp16)[name = string("op_46841_cast_fp16")]; + tensor var_46845_begin_0 = const()[name = string("op_46845_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46845_end_0 = const()[name = string("op_46845_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_46845_end_mask_0 = const()[name = string("op_46845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46845_cast_fp16 = slice_by_index(begin = var_46845_begin_0, end = var_46845_end_0, end_mask = var_46845_end_mask_0, x = value_61_cast_fp16)[name = string("op_46845_cast_fp16")]; + tensor var_46849_begin_0 = const()[name = string("op_46849_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46849_end_0 = const()[name = string("op_46849_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_46849_end_mask_0 = const()[name = string("op_46849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46849_cast_fp16 = slice_by_index(begin = var_46849_begin_0, end = var_46849_end_0, end_mask = var_46849_end_mask_0, x = value_61_cast_fp16)[name = string("op_46849_cast_fp16")]; + tensor var_46853_begin_0 = const()[name = string("op_46853_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46853_end_0 = const()[name = string("op_46853_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_46853_end_mask_0 = const()[name = string("op_46853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46853_cast_fp16 = slice_by_index(begin = var_46853_begin_0, end = var_46853_end_0, end_mask = var_46853_end_mask_0, x = value_61_cast_fp16)[name = string("op_46853_cast_fp16")]; + tensor var_46857_begin_0 = const()[name = string("op_46857_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46857_end_0 = const()[name = string("op_46857_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_46857_end_mask_0 = const()[name = string("op_46857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46857_cast_fp16 = slice_by_index(begin = var_46857_begin_0, end = var_46857_end_0, end_mask = var_46857_end_mask_0, x = value_61_cast_fp16)[name = string("op_46857_cast_fp16")]; + tensor var_46861_begin_0 = const()[name = string("op_46861_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46861_end_0 = const()[name = string("op_46861_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_46861_end_mask_0 = const()[name = string("op_46861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46861_cast_fp16 = slice_by_index(begin = var_46861_begin_0, end = var_46861_end_0, end_mask = var_46861_end_mask_0, x = value_61_cast_fp16)[name = string("op_46861_cast_fp16")]; + tensor var_46865_begin_0 = const()[name = string("op_46865_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46865_end_0 = const()[name = string("op_46865_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_46865_end_mask_0 = const()[name = string("op_46865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46865_cast_fp16 = slice_by_index(begin = var_46865_begin_0, end = var_46865_end_0, end_mask = var_46865_end_mask_0, x = value_61_cast_fp16)[name = string("op_46865_cast_fp16")]; + tensor var_46869_begin_0 = const()[name = string("op_46869_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46869_end_0 = const()[name = string("op_46869_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_46869_end_mask_0 = const()[name = string("op_46869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46869_cast_fp16 = slice_by_index(begin = var_46869_begin_0, end = var_46869_end_0, end_mask = var_46869_end_mask_0, x = value_61_cast_fp16)[name = string("op_46869_cast_fp16")]; + string _SplitHeadsQ__mh_w_4801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4801_equation_0, values = (var_46715_cast_fp16, var_46157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4801_cast_fp16")]; + string _SplitHeadsQ__mh_w_4803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4803_equation_0, values = (var_46715_cast_fp16, var_46164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4803_cast_fp16")]; + string _SplitHeadsQ__mh_w_4805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4805_equation_0, values = (var_46715_cast_fp16, var_46171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4805_cast_fp16")]; + string _SplitHeadsQ__mh_w_4807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4807_equation_0, values = (var_46715_cast_fp16, var_46178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4807_cast_fp16")]; + string _SplitHeadsQ__mh_w_4809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4809_equation_0, values = (var_46719_cast_fp16, var_46185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4809_cast_fp16")]; + string _SplitHeadsQ__mh_w_4811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4811_equation_0, values = (var_46719_cast_fp16, var_46192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4811_cast_fp16")]; + string _SplitHeadsQ__mh_w_4813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4813_equation_0, values = (var_46719_cast_fp16, var_46199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4813_cast_fp16")]; + string _SplitHeadsQ__mh_w_4815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4815_equation_0, values = (var_46719_cast_fp16, var_46206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4815_cast_fp16")]; + string _SplitHeadsQ__mh_w_4817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4817_equation_0, values = (var_46723_cast_fp16, var_46213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4817_cast_fp16")]; + string _SplitHeadsQ__mh_w_4819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4819_equation_0, values = (var_46723_cast_fp16, var_46220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4819_cast_fp16")]; + string _SplitHeadsQ__mh_w_4821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4821_equation_0, values = (var_46723_cast_fp16, var_46227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4821_cast_fp16")]; + string _SplitHeadsQ__mh_w_4823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4823_equation_0, values = (var_46723_cast_fp16, var_46234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4823_cast_fp16")]; + string _SplitHeadsQ__mh_w_4825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4825_equation_0, values = (var_46727_cast_fp16, var_46241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4825_cast_fp16")]; + string _SplitHeadsQ__mh_w_4827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4827_equation_0, values = (var_46727_cast_fp16, var_46248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4827_cast_fp16")]; + string _SplitHeadsQ__mh_w_4829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4829_equation_0, values = (var_46727_cast_fp16, var_46255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4829_cast_fp16")]; + string _SplitHeadsQ__mh_w_4831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4831_equation_0, values = (var_46727_cast_fp16, var_46262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4831_cast_fp16")]; + string _SplitHeadsQ__mh_w_4833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4833_equation_0, values = (var_46731_cast_fp16, var_46269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4833_cast_fp16")]; + string _SplitHeadsQ__mh_w_4835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4835_equation_0, values = (var_46731_cast_fp16, var_46276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4835_cast_fp16")]; + string _SplitHeadsQ__mh_w_4837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4837_equation_0, values = (var_46731_cast_fp16, var_46283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4837_cast_fp16")]; + string _SplitHeadsQ__mh_w_4839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4839_equation_0, values = (var_46731_cast_fp16, var_46290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4839_cast_fp16")]; + string _SplitHeadsQ__mh_w_4841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4841_equation_0, values = (var_46735_cast_fp16, var_46297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4841_cast_fp16")]; + string _SplitHeadsQ__mh_w_4843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4843_equation_0, values = (var_46735_cast_fp16, var_46304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4843_cast_fp16")]; + string _SplitHeadsQ__mh_w_4845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4845_equation_0, values = (var_46735_cast_fp16, var_46311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4845_cast_fp16")]; + string _SplitHeadsQ__mh_w_4847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4847_equation_0, values = (var_46735_cast_fp16, var_46318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4847_cast_fp16")]; + string _SplitHeadsQ__mh_w_4849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4849_equation_0, values = (var_46739_cast_fp16, var_46325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4849_cast_fp16")]; + string _SplitHeadsQ__mh_w_4851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4851_equation_0, values = (var_46739_cast_fp16, var_46332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4851_cast_fp16")]; + string _SplitHeadsQ__mh_w_4853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4853_equation_0, values = (var_46739_cast_fp16, var_46339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4853_cast_fp16")]; + string _SplitHeadsQ__mh_w_4855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4855_equation_0, values = (var_46739_cast_fp16, var_46346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4855_cast_fp16")]; + string _SplitHeadsQ__mh_w_4857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4857_equation_0, values = (var_46743_cast_fp16, var_46353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4857_cast_fp16")]; + string _SplitHeadsQ__mh_w_4859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4859_equation_0, values = (var_46743_cast_fp16, var_46360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4859_cast_fp16")]; + string _SplitHeadsQ__mh_w_4861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4861_equation_0, values = (var_46743_cast_fp16, var_46367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4861_cast_fp16")]; + string _SplitHeadsQ__mh_w_4863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4863_equation_0, values = (var_46743_cast_fp16, var_46374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4863_cast_fp16")]; + string _SplitHeadsQ__mh_w_4865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4865_equation_0, values = (var_46747_cast_fp16, var_46381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4865_cast_fp16")]; + string _SplitHeadsQ__mh_w_4867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4867_equation_0, values = (var_46747_cast_fp16, var_46388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4867_cast_fp16")]; + string _SplitHeadsQ__mh_w_4869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4869_equation_0, values = (var_46747_cast_fp16, var_46395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4869_cast_fp16")]; + string _SplitHeadsQ__mh_w_4871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4871_equation_0, values = (var_46747_cast_fp16, var_46402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4871_cast_fp16")]; + string _SplitHeadsQ__mh_w_4873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4873_equation_0, values = (var_46751_cast_fp16, var_46409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4873_cast_fp16")]; + string _SplitHeadsQ__mh_w_4875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4875_equation_0, values = (var_46751_cast_fp16, var_46416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4875_cast_fp16")]; + string _SplitHeadsQ__mh_w_4877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4877_equation_0, values = (var_46751_cast_fp16, var_46423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4877_cast_fp16")]; + string _SplitHeadsQ__mh_w_4879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4879_equation_0, values = (var_46751_cast_fp16, var_46430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4879_cast_fp16")]; + string _SplitHeadsQ__mh_w_4881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4881_equation_0, values = (var_46755_cast_fp16, var_46437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4881_cast_fp16")]; + string _SplitHeadsQ__mh_w_4883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4883_equation_0, values = (var_46755_cast_fp16, var_46444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4883_cast_fp16")]; + string _SplitHeadsQ__mh_w_4885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4885_equation_0, values = (var_46755_cast_fp16, var_46451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4885_cast_fp16")]; + string _SplitHeadsQ__mh_w_4887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4887_equation_0, values = (var_46755_cast_fp16, var_46458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4887_cast_fp16")]; + string _SplitHeadsQ__mh_w_4889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4889_equation_0, values = (var_46759_cast_fp16, var_46465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4889_cast_fp16")]; + string _SplitHeadsQ__mh_w_4891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4891_equation_0, values = (var_46759_cast_fp16, var_46472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4891_cast_fp16")]; + string _SplitHeadsQ__mh_w_4893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4893_equation_0, values = (var_46759_cast_fp16, var_46479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4893_cast_fp16")]; + string _SplitHeadsQ__mh_w_4895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4895_equation_0, values = (var_46759_cast_fp16, var_46486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4895_cast_fp16")]; + string _SplitHeadsQ__mh_w_4897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4897_equation_0, values = (var_46763_cast_fp16, var_46493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4897_cast_fp16")]; + string _SplitHeadsQ__mh_w_4899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4899_equation_0, values = (var_46763_cast_fp16, var_46500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4899_cast_fp16")]; + string _SplitHeadsQ__mh_w_4901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4901_equation_0, values = (var_46763_cast_fp16, var_46507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4901_cast_fp16")]; + string _SplitHeadsQ__mh_w_4903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4903_equation_0, values = (var_46763_cast_fp16, var_46514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4903_cast_fp16")]; + string _SplitHeadsQ__mh_w_4905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4905_equation_0, values = (var_46767_cast_fp16, var_46521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4905_cast_fp16")]; + string _SplitHeadsQ__mh_w_4907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4907_equation_0, values = (var_46767_cast_fp16, var_46528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4907_cast_fp16")]; + string _SplitHeadsQ__mh_w_4909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4909_equation_0, values = (var_46767_cast_fp16, var_46535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4909_cast_fp16")]; + string _SplitHeadsQ__mh_w_4911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4911_equation_0, values = (var_46767_cast_fp16, var_46542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4911_cast_fp16")]; + string _SplitHeadsQ__mh_w_4913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4913_equation_0, values = (var_46771_cast_fp16, var_46549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4913_cast_fp16")]; + string _SplitHeadsQ__mh_w_4915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4915_equation_0, values = (var_46771_cast_fp16, var_46556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4915_cast_fp16")]; + string _SplitHeadsQ__mh_w_4917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4917_equation_0, values = (var_46771_cast_fp16, var_46563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4917_cast_fp16")]; + string _SplitHeadsQ__mh_w_4919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4919_equation_0, values = (var_46771_cast_fp16, var_46570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4919_cast_fp16")]; + string _SplitHeadsQ__mh_w_4921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4921_equation_0, values = (var_46775_cast_fp16, var_46577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4921_cast_fp16")]; + string _SplitHeadsQ__mh_w_4923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4923_equation_0, values = (var_46775_cast_fp16, var_46584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4923_cast_fp16")]; + string _SplitHeadsQ__mh_w_4925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4925_equation_0, values = (var_46775_cast_fp16, var_46591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4925_cast_fp16")]; + string _SplitHeadsQ__mh_w_4927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4927_equation_0, values = (var_46775_cast_fp16, var_46598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4927_cast_fp16")]; + string _SplitHeadsQ__mh_w_4929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4929_equation_0, values = (var_46779_cast_fp16, var_46605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4929_cast_fp16")]; + string _SplitHeadsQ__mh_w_4931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4931_equation_0, values = (var_46779_cast_fp16, var_46612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4931_cast_fp16")]; + string _SplitHeadsQ__mh_w_4933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4933_equation_0, values = (var_46779_cast_fp16, var_46619_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4933_cast_fp16")]; + string _SplitHeadsQ__mh_w_4935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4935_equation_0, values = (var_46779_cast_fp16, var_46626_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4935_cast_fp16")]; + string _SplitHeadsQ__mh_w_4937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4937_equation_0, values = (var_46783_cast_fp16, var_46633_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4937_cast_fp16")]; + string _SplitHeadsQ__mh_w_4939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4939_equation_0, values = (var_46783_cast_fp16, var_46640_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4939_cast_fp16")]; + string _SplitHeadsQ__mh_w_4941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4941_equation_0, values = (var_46783_cast_fp16, var_46647_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4941_cast_fp16")]; + string _SplitHeadsQ__mh_w_4943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4943_equation_0, values = (var_46783_cast_fp16, var_46654_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4943_cast_fp16")]; + string _SplitHeadsQ__mh_w_4945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4945_equation_0, values = (var_46787_cast_fp16, var_46661_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4945_cast_fp16")]; + string _SplitHeadsQ__mh_w_4947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4947_equation_0, values = (var_46787_cast_fp16, var_46668_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4947_cast_fp16")]; + string _SplitHeadsQ__mh_w_4949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4949_equation_0, values = (var_46787_cast_fp16, var_46675_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4949_cast_fp16")]; + string _SplitHeadsQ__mh_w_4951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4951_equation_0, values = (var_46787_cast_fp16, var_46682_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4951_cast_fp16")]; + string _SplitHeadsQ__mh_w_4953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4953_equation_0, values = (var_46791_cast_fp16, var_46689_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4953_cast_fp16")]; + string _SplitHeadsQ__mh_w_4955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4955_equation_0, values = (var_46791_cast_fp16, var_46696_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4955_cast_fp16")]; + string _SplitHeadsQ__mh_w_4957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4957_equation_0, values = (var_46791_cast_fp16, var_46703_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4957_cast_fp16")]; + string _SplitHeadsQ__mh_w_4959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4959_equation_0, values = (var_46791_cast_fp16, var_46710_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4959_cast_fp16")]; + fp16 var_47032_to_fp16 = const()[name = string("op_47032_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4801_cast_fp16, y = var_47032_to_fp16)[name = string("aw_chunk_4801_cast_fp16")]; + fp16 var_47034_to_fp16 = const()[name = string("op_47034_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4803_cast_fp16, y = var_47034_to_fp16)[name = string("aw_chunk_4803_cast_fp16")]; + fp16 var_47036_to_fp16 = const()[name = string("op_47036_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4805_cast_fp16, y = var_47036_to_fp16)[name = string("aw_chunk_4805_cast_fp16")]; + fp16 var_47038_to_fp16 = const()[name = string("op_47038_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4807_cast_fp16, y = var_47038_to_fp16)[name = string("aw_chunk_4807_cast_fp16")]; + fp16 var_47040_to_fp16 = const()[name = string("op_47040_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4809_cast_fp16, y = var_47040_to_fp16)[name = string("aw_chunk_4809_cast_fp16")]; + fp16 var_47042_to_fp16 = const()[name = string("op_47042_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4811_cast_fp16, y = var_47042_to_fp16)[name = string("aw_chunk_4811_cast_fp16")]; + fp16 var_47044_to_fp16 = const()[name = string("op_47044_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4813_cast_fp16, y = var_47044_to_fp16)[name = string("aw_chunk_4813_cast_fp16")]; + fp16 var_47046_to_fp16 = const()[name = string("op_47046_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4815_cast_fp16, y = var_47046_to_fp16)[name = string("aw_chunk_4815_cast_fp16")]; + fp16 var_47048_to_fp16 = const()[name = string("op_47048_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4817_cast_fp16, y = var_47048_to_fp16)[name = string("aw_chunk_4817_cast_fp16")]; + fp16 var_47050_to_fp16 = const()[name = string("op_47050_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4819_cast_fp16, y = var_47050_to_fp16)[name = string("aw_chunk_4819_cast_fp16")]; + fp16 var_47052_to_fp16 = const()[name = string("op_47052_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4821_cast_fp16, y = var_47052_to_fp16)[name = string("aw_chunk_4821_cast_fp16")]; + fp16 var_47054_to_fp16 = const()[name = string("op_47054_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4823_cast_fp16, y = var_47054_to_fp16)[name = string("aw_chunk_4823_cast_fp16")]; + fp16 var_47056_to_fp16 = const()[name = string("op_47056_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4825_cast_fp16, y = var_47056_to_fp16)[name = string("aw_chunk_4825_cast_fp16")]; + fp16 var_47058_to_fp16 = const()[name = string("op_47058_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4827_cast_fp16, y = var_47058_to_fp16)[name = string("aw_chunk_4827_cast_fp16")]; + fp16 var_47060_to_fp16 = const()[name = string("op_47060_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4829_cast_fp16, y = var_47060_to_fp16)[name = string("aw_chunk_4829_cast_fp16")]; + fp16 var_47062_to_fp16 = const()[name = string("op_47062_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4831_cast_fp16, y = var_47062_to_fp16)[name = string("aw_chunk_4831_cast_fp16")]; + fp16 var_47064_to_fp16 = const()[name = string("op_47064_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4833_cast_fp16, y = var_47064_to_fp16)[name = string("aw_chunk_4833_cast_fp16")]; + fp16 var_47066_to_fp16 = const()[name = string("op_47066_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4835_cast_fp16, y = var_47066_to_fp16)[name = string("aw_chunk_4835_cast_fp16")]; + fp16 var_47068_to_fp16 = const()[name = string("op_47068_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4837_cast_fp16, y = var_47068_to_fp16)[name = string("aw_chunk_4837_cast_fp16")]; + fp16 var_47070_to_fp16 = const()[name = string("op_47070_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4839_cast_fp16, y = var_47070_to_fp16)[name = string("aw_chunk_4839_cast_fp16")]; + fp16 var_47072_to_fp16 = const()[name = string("op_47072_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4841_cast_fp16, y = var_47072_to_fp16)[name = string("aw_chunk_4841_cast_fp16")]; + fp16 var_47074_to_fp16 = const()[name = string("op_47074_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4843_cast_fp16, y = var_47074_to_fp16)[name = string("aw_chunk_4843_cast_fp16")]; + fp16 var_47076_to_fp16 = const()[name = string("op_47076_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4845_cast_fp16, y = var_47076_to_fp16)[name = string("aw_chunk_4845_cast_fp16")]; + fp16 var_47078_to_fp16 = const()[name = string("op_47078_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4847_cast_fp16, y = var_47078_to_fp16)[name = string("aw_chunk_4847_cast_fp16")]; + fp16 var_47080_to_fp16 = const()[name = string("op_47080_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4849_cast_fp16, y = var_47080_to_fp16)[name = string("aw_chunk_4849_cast_fp16")]; + fp16 var_47082_to_fp16 = const()[name = string("op_47082_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4851_cast_fp16, y = var_47082_to_fp16)[name = string("aw_chunk_4851_cast_fp16")]; + fp16 var_47084_to_fp16 = const()[name = string("op_47084_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4853_cast_fp16, y = var_47084_to_fp16)[name = string("aw_chunk_4853_cast_fp16")]; + fp16 var_47086_to_fp16 = const()[name = string("op_47086_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4855_cast_fp16, y = var_47086_to_fp16)[name = string("aw_chunk_4855_cast_fp16")]; + fp16 var_47088_to_fp16 = const()[name = string("op_47088_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4857_cast_fp16, y = var_47088_to_fp16)[name = string("aw_chunk_4857_cast_fp16")]; + fp16 var_47090_to_fp16 = const()[name = string("op_47090_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4859_cast_fp16, y = var_47090_to_fp16)[name = string("aw_chunk_4859_cast_fp16")]; + fp16 var_47092_to_fp16 = const()[name = string("op_47092_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4861_cast_fp16, y = var_47092_to_fp16)[name = string("aw_chunk_4861_cast_fp16")]; + fp16 var_47094_to_fp16 = const()[name = string("op_47094_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4863_cast_fp16, y = var_47094_to_fp16)[name = string("aw_chunk_4863_cast_fp16")]; + fp16 var_47096_to_fp16 = const()[name = string("op_47096_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4865_cast_fp16, y = var_47096_to_fp16)[name = string("aw_chunk_4865_cast_fp16")]; + fp16 var_47098_to_fp16 = const()[name = string("op_47098_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4867_cast_fp16, y = var_47098_to_fp16)[name = string("aw_chunk_4867_cast_fp16")]; + fp16 var_47100_to_fp16 = const()[name = string("op_47100_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4869_cast_fp16, y = var_47100_to_fp16)[name = string("aw_chunk_4869_cast_fp16")]; + fp16 var_47102_to_fp16 = const()[name = string("op_47102_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4871_cast_fp16, y = var_47102_to_fp16)[name = string("aw_chunk_4871_cast_fp16")]; + fp16 var_47104_to_fp16 = const()[name = string("op_47104_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4873_cast_fp16, y = var_47104_to_fp16)[name = string("aw_chunk_4873_cast_fp16")]; + fp16 var_47106_to_fp16 = const()[name = string("op_47106_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4875_cast_fp16, y = var_47106_to_fp16)[name = string("aw_chunk_4875_cast_fp16")]; + fp16 var_47108_to_fp16 = const()[name = string("op_47108_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4877_cast_fp16, y = var_47108_to_fp16)[name = string("aw_chunk_4877_cast_fp16")]; + fp16 var_47110_to_fp16 = const()[name = string("op_47110_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4879_cast_fp16, y = var_47110_to_fp16)[name = string("aw_chunk_4879_cast_fp16")]; + fp16 var_47112_to_fp16 = const()[name = string("op_47112_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4881_cast_fp16, y = var_47112_to_fp16)[name = string("aw_chunk_4881_cast_fp16")]; + fp16 var_47114_to_fp16 = const()[name = string("op_47114_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4883_cast_fp16, y = var_47114_to_fp16)[name = string("aw_chunk_4883_cast_fp16")]; + fp16 var_47116_to_fp16 = const()[name = string("op_47116_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4885_cast_fp16, y = var_47116_to_fp16)[name = string("aw_chunk_4885_cast_fp16")]; + fp16 var_47118_to_fp16 = const()[name = string("op_47118_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4887_cast_fp16, y = var_47118_to_fp16)[name = string("aw_chunk_4887_cast_fp16")]; + fp16 var_47120_to_fp16 = const()[name = string("op_47120_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4889_cast_fp16, y = var_47120_to_fp16)[name = string("aw_chunk_4889_cast_fp16")]; + fp16 var_47122_to_fp16 = const()[name = string("op_47122_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4891_cast_fp16, y = var_47122_to_fp16)[name = string("aw_chunk_4891_cast_fp16")]; + fp16 var_47124_to_fp16 = const()[name = string("op_47124_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4893_cast_fp16, y = var_47124_to_fp16)[name = string("aw_chunk_4893_cast_fp16")]; + fp16 var_47126_to_fp16 = const()[name = string("op_47126_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4895_cast_fp16, y = var_47126_to_fp16)[name = string("aw_chunk_4895_cast_fp16")]; + fp16 var_47128_to_fp16 = const()[name = string("op_47128_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4897_cast_fp16, y = var_47128_to_fp16)[name = string("aw_chunk_4897_cast_fp16")]; + fp16 var_47130_to_fp16 = const()[name = string("op_47130_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4899_cast_fp16, y = var_47130_to_fp16)[name = string("aw_chunk_4899_cast_fp16")]; + fp16 var_47132_to_fp16 = const()[name = string("op_47132_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4901_cast_fp16, y = var_47132_to_fp16)[name = string("aw_chunk_4901_cast_fp16")]; + fp16 var_47134_to_fp16 = const()[name = string("op_47134_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4903_cast_fp16, y = var_47134_to_fp16)[name = string("aw_chunk_4903_cast_fp16")]; + fp16 var_47136_to_fp16 = const()[name = string("op_47136_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4905_cast_fp16, y = var_47136_to_fp16)[name = string("aw_chunk_4905_cast_fp16")]; + fp16 var_47138_to_fp16 = const()[name = string("op_47138_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4907_cast_fp16, y = var_47138_to_fp16)[name = string("aw_chunk_4907_cast_fp16")]; + fp16 var_47140_to_fp16 = const()[name = string("op_47140_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4909_cast_fp16, y = var_47140_to_fp16)[name = string("aw_chunk_4909_cast_fp16")]; + fp16 var_47142_to_fp16 = const()[name = string("op_47142_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4911_cast_fp16, y = var_47142_to_fp16)[name = string("aw_chunk_4911_cast_fp16")]; + fp16 var_47144_to_fp16 = const()[name = string("op_47144_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4913_cast_fp16, y = var_47144_to_fp16)[name = string("aw_chunk_4913_cast_fp16")]; + fp16 var_47146_to_fp16 = const()[name = string("op_47146_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4915_cast_fp16, y = var_47146_to_fp16)[name = string("aw_chunk_4915_cast_fp16")]; + fp16 var_47148_to_fp16 = const()[name = string("op_47148_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4917_cast_fp16, y = var_47148_to_fp16)[name = string("aw_chunk_4917_cast_fp16")]; + fp16 var_47150_to_fp16 = const()[name = string("op_47150_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4919_cast_fp16, y = var_47150_to_fp16)[name = string("aw_chunk_4919_cast_fp16")]; + fp16 var_47152_to_fp16 = const()[name = string("op_47152_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4921_cast_fp16, y = var_47152_to_fp16)[name = string("aw_chunk_4921_cast_fp16")]; + fp16 var_47154_to_fp16 = const()[name = string("op_47154_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4923_cast_fp16, y = var_47154_to_fp16)[name = string("aw_chunk_4923_cast_fp16")]; + fp16 var_47156_to_fp16 = const()[name = string("op_47156_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4925_cast_fp16, y = var_47156_to_fp16)[name = string("aw_chunk_4925_cast_fp16")]; + fp16 var_47158_to_fp16 = const()[name = string("op_47158_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4927_cast_fp16, y = var_47158_to_fp16)[name = string("aw_chunk_4927_cast_fp16")]; + fp16 var_47160_to_fp16 = const()[name = string("op_47160_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4929_cast_fp16, y = var_47160_to_fp16)[name = string("aw_chunk_4929_cast_fp16")]; + fp16 var_47162_to_fp16 = const()[name = string("op_47162_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4931_cast_fp16, y = var_47162_to_fp16)[name = string("aw_chunk_4931_cast_fp16")]; + fp16 var_47164_to_fp16 = const()[name = string("op_47164_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4933_cast_fp16, y = var_47164_to_fp16)[name = string("aw_chunk_4933_cast_fp16")]; + fp16 var_47166_to_fp16 = const()[name = string("op_47166_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4935_cast_fp16, y = var_47166_to_fp16)[name = string("aw_chunk_4935_cast_fp16")]; + fp16 var_47168_to_fp16 = const()[name = string("op_47168_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4937_cast_fp16, y = var_47168_to_fp16)[name = string("aw_chunk_4937_cast_fp16")]; + fp16 var_47170_to_fp16 = const()[name = string("op_47170_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4939_cast_fp16, y = var_47170_to_fp16)[name = string("aw_chunk_4939_cast_fp16")]; + fp16 var_47172_to_fp16 = const()[name = string("op_47172_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4941_cast_fp16, y = var_47172_to_fp16)[name = string("aw_chunk_4941_cast_fp16")]; + fp16 var_47174_to_fp16 = const()[name = string("op_47174_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4943_cast_fp16, y = var_47174_to_fp16)[name = string("aw_chunk_4943_cast_fp16")]; + fp16 var_47176_to_fp16 = const()[name = string("op_47176_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4945_cast_fp16, y = var_47176_to_fp16)[name = string("aw_chunk_4945_cast_fp16")]; + fp16 var_47178_to_fp16 = const()[name = string("op_47178_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4947_cast_fp16, y = var_47178_to_fp16)[name = string("aw_chunk_4947_cast_fp16")]; + fp16 var_47180_to_fp16 = const()[name = string("op_47180_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4949_cast_fp16, y = var_47180_to_fp16)[name = string("aw_chunk_4949_cast_fp16")]; + fp16 var_47182_to_fp16 = const()[name = string("op_47182_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4951_cast_fp16, y = var_47182_to_fp16)[name = string("aw_chunk_4951_cast_fp16")]; + fp16 var_47184_to_fp16 = const()[name = string("op_47184_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4953_cast_fp16, y = var_47184_to_fp16)[name = string("aw_chunk_4953_cast_fp16")]; + fp16 var_47186_to_fp16 = const()[name = string("op_47186_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4955_cast_fp16, y = var_47186_to_fp16)[name = string("aw_chunk_4955_cast_fp16")]; + fp16 var_47188_to_fp16 = const()[name = string("op_47188_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4957_cast_fp16, y = var_47188_to_fp16)[name = string("aw_chunk_4957_cast_fp16")]; + fp16 var_47190_to_fp16 = const()[name = string("op_47190_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4959_cast_fp16, y = var_47190_to_fp16)[name = string("aw_chunk_4959_cast_fp16")]; + tensor var_47192_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4801_cast_fp16)[name = string("op_47192_cast_fp16")]; + tensor var_47193_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4803_cast_fp16)[name = string("op_47193_cast_fp16")]; + tensor var_47194_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4805_cast_fp16)[name = string("op_47194_cast_fp16")]; + tensor var_47195_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4807_cast_fp16)[name = string("op_47195_cast_fp16")]; + tensor var_47196_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4809_cast_fp16)[name = string("op_47196_cast_fp16")]; + tensor var_47197_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4811_cast_fp16)[name = string("op_47197_cast_fp16")]; + tensor var_47198_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4813_cast_fp16)[name = string("op_47198_cast_fp16")]; + tensor var_47199_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4815_cast_fp16)[name = string("op_47199_cast_fp16")]; + tensor var_47200_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4817_cast_fp16)[name = string("op_47200_cast_fp16")]; + tensor var_47201_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4819_cast_fp16)[name = string("op_47201_cast_fp16")]; + tensor var_47202_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4821_cast_fp16)[name = string("op_47202_cast_fp16")]; + tensor var_47203_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4823_cast_fp16)[name = string("op_47203_cast_fp16")]; + tensor var_47204_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4825_cast_fp16)[name = string("op_47204_cast_fp16")]; + tensor var_47205_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4827_cast_fp16)[name = string("op_47205_cast_fp16")]; + tensor var_47206_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4829_cast_fp16)[name = string("op_47206_cast_fp16")]; + tensor var_47207_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4831_cast_fp16)[name = string("op_47207_cast_fp16")]; + tensor var_47208_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4833_cast_fp16)[name = string("op_47208_cast_fp16")]; + tensor var_47209_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4835_cast_fp16)[name = string("op_47209_cast_fp16")]; + tensor var_47210_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4837_cast_fp16)[name = string("op_47210_cast_fp16")]; + tensor var_47211_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4839_cast_fp16)[name = string("op_47211_cast_fp16")]; + tensor var_47212_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4841_cast_fp16)[name = string("op_47212_cast_fp16")]; + tensor var_47213_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4843_cast_fp16)[name = string("op_47213_cast_fp16")]; + tensor var_47214_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4845_cast_fp16)[name = string("op_47214_cast_fp16")]; + tensor var_47215_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4847_cast_fp16)[name = string("op_47215_cast_fp16")]; + tensor var_47216_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4849_cast_fp16)[name = string("op_47216_cast_fp16")]; + tensor var_47217_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4851_cast_fp16)[name = string("op_47217_cast_fp16")]; + tensor var_47218_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4853_cast_fp16)[name = string("op_47218_cast_fp16")]; + tensor var_47219_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4855_cast_fp16)[name = string("op_47219_cast_fp16")]; + tensor var_47220_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4857_cast_fp16)[name = string("op_47220_cast_fp16")]; + tensor var_47221_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4859_cast_fp16)[name = string("op_47221_cast_fp16")]; + tensor var_47222_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4861_cast_fp16)[name = string("op_47222_cast_fp16")]; + tensor var_47223_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4863_cast_fp16)[name = string("op_47223_cast_fp16")]; + tensor var_47224_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4865_cast_fp16)[name = string("op_47224_cast_fp16")]; + tensor var_47225_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4867_cast_fp16)[name = string("op_47225_cast_fp16")]; + tensor var_47226_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4869_cast_fp16)[name = string("op_47226_cast_fp16")]; + tensor var_47227_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4871_cast_fp16)[name = string("op_47227_cast_fp16")]; + tensor var_47228_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4873_cast_fp16)[name = string("op_47228_cast_fp16")]; + tensor var_47229_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4875_cast_fp16)[name = string("op_47229_cast_fp16")]; + tensor var_47230_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4877_cast_fp16)[name = string("op_47230_cast_fp16")]; + tensor var_47231_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4879_cast_fp16)[name = string("op_47231_cast_fp16")]; + tensor var_47232_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4881_cast_fp16)[name = string("op_47232_cast_fp16")]; + tensor var_47233_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4883_cast_fp16)[name = string("op_47233_cast_fp16")]; + tensor var_47234_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4885_cast_fp16)[name = string("op_47234_cast_fp16")]; + tensor var_47235_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4887_cast_fp16)[name = string("op_47235_cast_fp16")]; + tensor var_47236_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4889_cast_fp16)[name = string("op_47236_cast_fp16")]; + tensor var_47237_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4891_cast_fp16)[name = string("op_47237_cast_fp16")]; + tensor var_47238_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4893_cast_fp16)[name = string("op_47238_cast_fp16")]; + tensor var_47239_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4895_cast_fp16)[name = string("op_47239_cast_fp16")]; + tensor var_47240_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4897_cast_fp16)[name = string("op_47240_cast_fp16")]; + tensor var_47241_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4899_cast_fp16)[name = string("op_47241_cast_fp16")]; + tensor var_47242_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4901_cast_fp16)[name = string("op_47242_cast_fp16")]; + tensor var_47243_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4903_cast_fp16)[name = string("op_47243_cast_fp16")]; + tensor var_47244_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4905_cast_fp16)[name = string("op_47244_cast_fp16")]; + tensor var_47245_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4907_cast_fp16)[name = string("op_47245_cast_fp16")]; + tensor var_47246_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4909_cast_fp16)[name = string("op_47246_cast_fp16")]; + tensor var_47247_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4911_cast_fp16)[name = string("op_47247_cast_fp16")]; + tensor var_47248_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4913_cast_fp16)[name = string("op_47248_cast_fp16")]; + tensor var_47249_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4915_cast_fp16)[name = string("op_47249_cast_fp16")]; + tensor var_47250_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4917_cast_fp16)[name = string("op_47250_cast_fp16")]; + tensor var_47251_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4919_cast_fp16)[name = string("op_47251_cast_fp16")]; + tensor var_47252_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4921_cast_fp16)[name = string("op_47252_cast_fp16")]; + tensor var_47253_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4923_cast_fp16)[name = string("op_47253_cast_fp16")]; + tensor var_47254_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4925_cast_fp16)[name = string("op_47254_cast_fp16")]; + tensor var_47255_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4927_cast_fp16)[name = string("op_47255_cast_fp16")]; + tensor var_47256_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4929_cast_fp16)[name = string("op_47256_cast_fp16")]; + tensor var_47257_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4931_cast_fp16)[name = string("op_47257_cast_fp16")]; + tensor var_47258_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4933_cast_fp16)[name = string("op_47258_cast_fp16")]; + tensor var_47259_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4935_cast_fp16)[name = string("op_47259_cast_fp16")]; + tensor var_47260_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4937_cast_fp16)[name = string("op_47260_cast_fp16")]; + tensor var_47261_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4939_cast_fp16)[name = string("op_47261_cast_fp16")]; + tensor var_47262_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4941_cast_fp16)[name = string("op_47262_cast_fp16")]; + tensor var_47263_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4943_cast_fp16)[name = string("op_47263_cast_fp16")]; + tensor var_47264_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4945_cast_fp16)[name = string("op_47264_cast_fp16")]; + tensor var_47265_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4947_cast_fp16)[name = string("op_47265_cast_fp16")]; + tensor var_47266_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4949_cast_fp16)[name = string("op_47266_cast_fp16")]; + tensor var_47267_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4951_cast_fp16)[name = string("op_47267_cast_fp16")]; + tensor var_47268_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4953_cast_fp16)[name = string("op_47268_cast_fp16")]; + tensor var_47269_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4955_cast_fp16)[name = string("op_47269_cast_fp16")]; + tensor var_47270_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4957_cast_fp16)[name = string("op_47270_cast_fp16")]; + tensor var_47271_cast_fp16 = softmax(axis = var_46017, x = aw_chunk_4959_cast_fp16)[name = string("op_47271_cast_fp16")]; + string var_47273_equation_0 = const()[name = string("op_47273_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47273_cast_fp16 = einsum(equation = var_47273_equation_0, values = (var_46793_cast_fp16, var_47192_cast_fp16))[name = string("op_47273_cast_fp16")]; + string var_47275_equation_0 = const()[name = string("op_47275_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47275_cast_fp16 = einsum(equation = var_47275_equation_0, values = (var_46793_cast_fp16, var_47193_cast_fp16))[name = string("op_47275_cast_fp16")]; + string var_47277_equation_0 = const()[name = string("op_47277_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47277_cast_fp16 = einsum(equation = var_47277_equation_0, values = (var_46793_cast_fp16, var_47194_cast_fp16))[name = string("op_47277_cast_fp16")]; + string var_47279_equation_0 = const()[name = string("op_47279_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47279_cast_fp16 = einsum(equation = var_47279_equation_0, values = (var_46793_cast_fp16, var_47195_cast_fp16))[name = string("op_47279_cast_fp16")]; + string var_47281_equation_0 = const()[name = string("op_47281_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47281_cast_fp16 = einsum(equation = var_47281_equation_0, values = (var_46797_cast_fp16, var_47196_cast_fp16))[name = string("op_47281_cast_fp16")]; + string var_47283_equation_0 = const()[name = string("op_47283_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47283_cast_fp16 = einsum(equation = var_47283_equation_0, values = (var_46797_cast_fp16, var_47197_cast_fp16))[name = string("op_47283_cast_fp16")]; + string var_47285_equation_0 = const()[name = string("op_47285_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47285_cast_fp16 = einsum(equation = var_47285_equation_0, values = (var_46797_cast_fp16, var_47198_cast_fp16))[name = string("op_47285_cast_fp16")]; + string var_47287_equation_0 = const()[name = string("op_47287_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47287_cast_fp16 = einsum(equation = var_47287_equation_0, values = (var_46797_cast_fp16, var_47199_cast_fp16))[name = string("op_47287_cast_fp16")]; + string var_47289_equation_0 = const()[name = string("op_47289_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47289_cast_fp16 = einsum(equation = var_47289_equation_0, values = (var_46801_cast_fp16, var_47200_cast_fp16))[name = string("op_47289_cast_fp16")]; + string var_47291_equation_0 = const()[name = string("op_47291_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47291_cast_fp16 = einsum(equation = var_47291_equation_0, values = (var_46801_cast_fp16, var_47201_cast_fp16))[name = string("op_47291_cast_fp16")]; + string var_47293_equation_0 = const()[name = string("op_47293_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47293_cast_fp16 = einsum(equation = var_47293_equation_0, values = (var_46801_cast_fp16, var_47202_cast_fp16))[name = string("op_47293_cast_fp16")]; + string var_47295_equation_0 = const()[name = string("op_47295_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47295_cast_fp16 = einsum(equation = var_47295_equation_0, values = (var_46801_cast_fp16, var_47203_cast_fp16))[name = string("op_47295_cast_fp16")]; + string var_47297_equation_0 = const()[name = string("op_47297_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47297_cast_fp16 = einsum(equation = var_47297_equation_0, values = (var_46805_cast_fp16, var_47204_cast_fp16))[name = string("op_47297_cast_fp16")]; + string var_47299_equation_0 = const()[name = string("op_47299_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47299_cast_fp16 = einsum(equation = var_47299_equation_0, values = (var_46805_cast_fp16, var_47205_cast_fp16))[name = string("op_47299_cast_fp16")]; + string var_47301_equation_0 = const()[name = string("op_47301_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47301_cast_fp16 = einsum(equation = var_47301_equation_0, values = (var_46805_cast_fp16, var_47206_cast_fp16))[name = string("op_47301_cast_fp16")]; + string var_47303_equation_0 = const()[name = string("op_47303_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47303_cast_fp16 = einsum(equation = var_47303_equation_0, values = (var_46805_cast_fp16, var_47207_cast_fp16))[name = string("op_47303_cast_fp16")]; + string var_47305_equation_0 = const()[name = string("op_47305_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47305_cast_fp16 = einsum(equation = var_47305_equation_0, values = (var_46809_cast_fp16, var_47208_cast_fp16))[name = string("op_47305_cast_fp16")]; + string var_47307_equation_0 = const()[name = string("op_47307_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47307_cast_fp16 = einsum(equation = var_47307_equation_0, values = (var_46809_cast_fp16, var_47209_cast_fp16))[name = string("op_47307_cast_fp16")]; + string var_47309_equation_0 = const()[name = string("op_47309_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47309_cast_fp16 = einsum(equation = var_47309_equation_0, values = (var_46809_cast_fp16, var_47210_cast_fp16))[name = string("op_47309_cast_fp16")]; + string var_47311_equation_0 = const()[name = string("op_47311_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47311_cast_fp16 = einsum(equation = var_47311_equation_0, values = (var_46809_cast_fp16, var_47211_cast_fp16))[name = string("op_47311_cast_fp16")]; + string var_47313_equation_0 = const()[name = string("op_47313_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47313_cast_fp16 = einsum(equation = var_47313_equation_0, values = (var_46813_cast_fp16, var_47212_cast_fp16))[name = string("op_47313_cast_fp16")]; + string var_47315_equation_0 = const()[name = string("op_47315_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47315_cast_fp16 = einsum(equation = var_47315_equation_0, values = (var_46813_cast_fp16, var_47213_cast_fp16))[name = string("op_47315_cast_fp16")]; + string var_47317_equation_0 = const()[name = string("op_47317_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47317_cast_fp16 = einsum(equation = var_47317_equation_0, values = (var_46813_cast_fp16, var_47214_cast_fp16))[name = string("op_47317_cast_fp16")]; + string var_47319_equation_0 = const()[name = string("op_47319_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47319_cast_fp16 = einsum(equation = var_47319_equation_0, values = (var_46813_cast_fp16, var_47215_cast_fp16))[name = string("op_47319_cast_fp16")]; + string var_47321_equation_0 = const()[name = string("op_47321_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47321_cast_fp16 = einsum(equation = var_47321_equation_0, values = (var_46817_cast_fp16, var_47216_cast_fp16))[name = string("op_47321_cast_fp16")]; + string var_47323_equation_0 = const()[name = string("op_47323_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47323_cast_fp16 = einsum(equation = var_47323_equation_0, values = (var_46817_cast_fp16, var_47217_cast_fp16))[name = string("op_47323_cast_fp16")]; + string var_47325_equation_0 = const()[name = string("op_47325_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47325_cast_fp16 = einsum(equation = var_47325_equation_0, values = (var_46817_cast_fp16, var_47218_cast_fp16))[name = string("op_47325_cast_fp16")]; + string var_47327_equation_0 = const()[name = string("op_47327_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47327_cast_fp16 = einsum(equation = var_47327_equation_0, values = (var_46817_cast_fp16, var_47219_cast_fp16))[name = string("op_47327_cast_fp16")]; + string var_47329_equation_0 = const()[name = string("op_47329_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47329_cast_fp16 = einsum(equation = var_47329_equation_0, values = (var_46821_cast_fp16, var_47220_cast_fp16))[name = string("op_47329_cast_fp16")]; + string var_47331_equation_0 = const()[name = string("op_47331_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47331_cast_fp16 = einsum(equation = var_47331_equation_0, values = (var_46821_cast_fp16, var_47221_cast_fp16))[name = string("op_47331_cast_fp16")]; + string var_47333_equation_0 = const()[name = string("op_47333_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47333_cast_fp16 = einsum(equation = var_47333_equation_0, values = (var_46821_cast_fp16, var_47222_cast_fp16))[name = string("op_47333_cast_fp16")]; + string var_47335_equation_0 = const()[name = string("op_47335_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47335_cast_fp16 = einsum(equation = var_47335_equation_0, values = (var_46821_cast_fp16, var_47223_cast_fp16))[name = string("op_47335_cast_fp16")]; + string var_47337_equation_0 = const()[name = string("op_47337_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47337_cast_fp16 = einsum(equation = var_47337_equation_0, values = (var_46825_cast_fp16, var_47224_cast_fp16))[name = string("op_47337_cast_fp16")]; + string var_47339_equation_0 = const()[name = string("op_47339_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47339_cast_fp16 = einsum(equation = var_47339_equation_0, values = (var_46825_cast_fp16, var_47225_cast_fp16))[name = string("op_47339_cast_fp16")]; + string var_47341_equation_0 = const()[name = string("op_47341_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47341_cast_fp16 = einsum(equation = var_47341_equation_0, values = (var_46825_cast_fp16, var_47226_cast_fp16))[name = string("op_47341_cast_fp16")]; + string var_47343_equation_0 = const()[name = string("op_47343_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47343_cast_fp16 = einsum(equation = var_47343_equation_0, values = (var_46825_cast_fp16, var_47227_cast_fp16))[name = string("op_47343_cast_fp16")]; + string var_47345_equation_0 = const()[name = string("op_47345_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47345_cast_fp16 = einsum(equation = var_47345_equation_0, values = (var_46829_cast_fp16, var_47228_cast_fp16))[name = string("op_47345_cast_fp16")]; + string var_47347_equation_0 = const()[name = string("op_47347_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47347_cast_fp16 = einsum(equation = var_47347_equation_0, values = (var_46829_cast_fp16, var_47229_cast_fp16))[name = string("op_47347_cast_fp16")]; + string var_47349_equation_0 = const()[name = string("op_47349_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47349_cast_fp16 = einsum(equation = var_47349_equation_0, values = (var_46829_cast_fp16, var_47230_cast_fp16))[name = string("op_47349_cast_fp16")]; + string var_47351_equation_0 = const()[name = string("op_47351_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47351_cast_fp16 = einsum(equation = var_47351_equation_0, values = (var_46829_cast_fp16, var_47231_cast_fp16))[name = string("op_47351_cast_fp16")]; + string var_47353_equation_0 = const()[name = string("op_47353_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47353_cast_fp16 = einsum(equation = var_47353_equation_0, values = (var_46833_cast_fp16, var_47232_cast_fp16))[name = string("op_47353_cast_fp16")]; + string var_47355_equation_0 = const()[name = string("op_47355_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47355_cast_fp16 = einsum(equation = var_47355_equation_0, values = (var_46833_cast_fp16, var_47233_cast_fp16))[name = string("op_47355_cast_fp16")]; + string var_47357_equation_0 = const()[name = string("op_47357_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47357_cast_fp16 = einsum(equation = var_47357_equation_0, values = (var_46833_cast_fp16, var_47234_cast_fp16))[name = string("op_47357_cast_fp16")]; + string var_47359_equation_0 = const()[name = string("op_47359_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47359_cast_fp16 = einsum(equation = var_47359_equation_0, values = (var_46833_cast_fp16, var_47235_cast_fp16))[name = string("op_47359_cast_fp16")]; + string var_47361_equation_0 = const()[name = string("op_47361_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47361_cast_fp16 = einsum(equation = var_47361_equation_0, values = (var_46837_cast_fp16, var_47236_cast_fp16))[name = string("op_47361_cast_fp16")]; + string var_47363_equation_0 = const()[name = string("op_47363_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47363_cast_fp16 = einsum(equation = var_47363_equation_0, values = (var_46837_cast_fp16, var_47237_cast_fp16))[name = string("op_47363_cast_fp16")]; + string var_47365_equation_0 = const()[name = string("op_47365_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47365_cast_fp16 = einsum(equation = var_47365_equation_0, values = (var_46837_cast_fp16, var_47238_cast_fp16))[name = string("op_47365_cast_fp16")]; + string var_47367_equation_0 = const()[name = string("op_47367_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47367_cast_fp16 = einsum(equation = var_47367_equation_0, values = (var_46837_cast_fp16, var_47239_cast_fp16))[name = string("op_47367_cast_fp16")]; + string var_47369_equation_0 = const()[name = string("op_47369_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47369_cast_fp16 = einsum(equation = var_47369_equation_0, values = (var_46841_cast_fp16, var_47240_cast_fp16))[name = string("op_47369_cast_fp16")]; + string var_47371_equation_0 = const()[name = string("op_47371_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47371_cast_fp16 = einsum(equation = var_47371_equation_0, values = (var_46841_cast_fp16, var_47241_cast_fp16))[name = string("op_47371_cast_fp16")]; + string var_47373_equation_0 = const()[name = string("op_47373_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47373_cast_fp16 = einsum(equation = var_47373_equation_0, values = (var_46841_cast_fp16, var_47242_cast_fp16))[name = string("op_47373_cast_fp16")]; + string var_47375_equation_0 = const()[name = string("op_47375_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47375_cast_fp16 = einsum(equation = var_47375_equation_0, values = (var_46841_cast_fp16, var_47243_cast_fp16))[name = string("op_47375_cast_fp16")]; + string var_47377_equation_0 = const()[name = string("op_47377_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47377_cast_fp16 = einsum(equation = var_47377_equation_0, values = (var_46845_cast_fp16, var_47244_cast_fp16))[name = string("op_47377_cast_fp16")]; + string var_47379_equation_0 = const()[name = string("op_47379_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47379_cast_fp16 = einsum(equation = var_47379_equation_0, values = (var_46845_cast_fp16, var_47245_cast_fp16))[name = string("op_47379_cast_fp16")]; + string var_47381_equation_0 = const()[name = string("op_47381_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47381_cast_fp16 = einsum(equation = var_47381_equation_0, values = (var_46845_cast_fp16, var_47246_cast_fp16))[name = string("op_47381_cast_fp16")]; + string var_47383_equation_0 = const()[name = string("op_47383_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47383_cast_fp16 = einsum(equation = var_47383_equation_0, values = (var_46845_cast_fp16, var_47247_cast_fp16))[name = string("op_47383_cast_fp16")]; + string var_47385_equation_0 = const()[name = string("op_47385_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47385_cast_fp16 = einsum(equation = var_47385_equation_0, values = (var_46849_cast_fp16, var_47248_cast_fp16))[name = string("op_47385_cast_fp16")]; + string var_47387_equation_0 = const()[name = string("op_47387_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47387_cast_fp16 = einsum(equation = var_47387_equation_0, values = (var_46849_cast_fp16, var_47249_cast_fp16))[name = string("op_47387_cast_fp16")]; + string var_47389_equation_0 = const()[name = string("op_47389_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47389_cast_fp16 = einsum(equation = var_47389_equation_0, values = (var_46849_cast_fp16, var_47250_cast_fp16))[name = string("op_47389_cast_fp16")]; + string var_47391_equation_0 = const()[name = string("op_47391_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47391_cast_fp16 = einsum(equation = var_47391_equation_0, values = (var_46849_cast_fp16, var_47251_cast_fp16))[name = string("op_47391_cast_fp16")]; + string var_47393_equation_0 = const()[name = string("op_47393_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47393_cast_fp16 = einsum(equation = var_47393_equation_0, values = (var_46853_cast_fp16, var_47252_cast_fp16))[name = string("op_47393_cast_fp16")]; + string var_47395_equation_0 = const()[name = string("op_47395_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47395_cast_fp16 = einsum(equation = var_47395_equation_0, values = (var_46853_cast_fp16, var_47253_cast_fp16))[name = string("op_47395_cast_fp16")]; + string var_47397_equation_0 = const()[name = string("op_47397_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47397_cast_fp16 = einsum(equation = var_47397_equation_0, values = (var_46853_cast_fp16, var_47254_cast_fp16))[name = string("op_47397_cast_fp16")]; + string var_47399_equation_0 = const()[name = string("op_47399_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47399_cast_fp16 = einsum(equation = var_47399_equation_0, values = (var_46853_cast_fp16, var_47255_cast_fp16))[name = string("op_47399_cast_fp16")]; + string var_47401_equation_0 = const()[name = string("op_47401_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47401_cast_fp16 = einsum(equation = var_47401_equation_0, values = (var_46857_cast_fp16, var_47256_cast_fp16))[name = string("op_47401_cast_fp16")]; + string var_47403_equation_0 = const()[name = string("op_47403_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47403_cast_fp16 = einsum(equation = var_47403_equation_0, values = (var_46857_cast_fp16, var_47257_cast_fp16))[name = string("op_47403_cast_fp16")]; + string var_47405_equation_0 = const()[name = string("op_47405_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47405_cast_fp16 = einsum(equation = var_47405_equation_0, values = (var_46857_cast_fp16, var_47258_cast_fp16))[name = string("op_47405_cast_fp16")]; + string var_47407_equation_0 = const()[name = string("op_47407_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47407_cast_fp16 = einsum(equation = var_47407_equation_0, values = (var_46857_cast_fp16, var_47259_cast_fp16))[name = string("op_47407_cast_fp16")]; + string var_47409_equation_0 = const()[name = string("op_47409_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47409_cast_fp16 = einsum(equation = var_47409_equation_0, values = (var_46861_cast_fp16, var_47260_cast_fp16))[name = string("op_47409_cast_fp16")]; + string var_47411_equation_0 = const()[name = string("op_47411_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47411_cast_fp16 = einsum(equation = var_47411_equation_0, values = (var_46861_cast_fp16, var_47261_cast_fp16))[name = string("op_47411_cast_fp16")]; + string var_47413_equation_0 = const()[name = string("op_47413_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47413_cast_fp16 = einsum(equation = var_47413_equation_0, values = (var_46861_cast_fp16, var_47262_cast_fp16))[name = string("op_47413_cast_fp16")]; + string var_47415_equation_0 = const()[name = string("op_47415_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47415_cast_fp16 = einsum(equation = var_47415_equation_0, values = (var_46861_cast_fp16, var_47263_cast_fp16))[name = string("op_47415_cast_fp16")]; + string var_47417_equation_0 = const()[name = string("op_47417_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47417_cast_fp16 = einsum(equation = var_47417_equation_0, values = (var_46865_cast_fp16, var_47264_cast_fp16))[name = string("op_47417_cast_fp16")]; + string var_47419_equation_0 = const()[name = string("op_47419_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47419_cast_fp16 = einsum(equation = var_47419_equation_0, values = (var_46865_cast_fp16, var_47265_cast_fp16))[name = string("op_47419_cast_fp16")]; + string var_47421_equation_0 = const()[name = string("op_47421_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47421_cast_fp16 = einsum(equation = var_47421_equation_0, values = (var_46865_cast_fp16, var_47266_cast_fp16))[name = string("op_47421_cast_fp16")]; + string var_47423_equation_0 = const()[name = string("op_47423_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47423_cast_fp16 = einsum(equation = var_47423_equation_0, values = (var_46865_cast_fp16, var_47267_cast_fp16))[name = string("op_47423_cast_fp16")]; + string var_47425_equation_0 = const()[name = string("op_47425_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47425_cast_fp16 = einsum(equation = var_47425_equation_0, values = (var_46869_cast_fp16, var_47268_cast_fp16))[name = string("op_47425_cast_fp16")]; + string var_47427_equation_0 = const()[name = string("op_47427_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47427_cast_fp16 = einsum(equation = var_47427_equation_0, values = (var_46869_cast_fp16, var_47269_cast_fp16))[name = string("op_47427_cast_fp16")]; + string var_47429_equation_0 = const()[name = string("op_47429_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47429_cast_fp16 = einsum(equation = var_47429_equation_0, values = (var_46869_cast_fp16, var_47270_cast_fp16))[name = string("op_47429_cast_fp16")]; + string var_47431_equation_0 = const()[name = string("op_47431_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_47431_cast_fp16 = einsum(equation = var_47431_equation_0, values = (var_46869_cast_fp16, var_47271_cast_fp16))[name = string("op_47431_cast_fp16")]; + bool var_47433_interleave_0 = const()[name = string("op_47433_interleave_0"), val = bool(false)]; + tensor var_47433_cast_fp16 = concat(axis = var_45992, interleave = var_47433_interleave_0, values = (var_47273_cast_fp16, var_47275_cast_fp16, var_47277_cast_fp16, var_47279_cast_fp16))[name = string("op_47433_cast_fp16")]; + bool var_47435_interleave_0 = const()[name = string("op_47435_interleave_0"), val = bool(false)]; + tensor var_47435_cast_fp16 = concat(axis = var_45992, interleave = var_47435_interleave_0, values = (var_47281_cast_fp16, var_47283_cast_fp16, var_47285_cast_fp16, var_47287_cast_fp16))[name = string("op_47435_cast_fp16")]; + bool var_47437_interleave_0 = const()[name = string("op_47437_interleave_0"), val = bool(false)]; + tensor var_47437_cast_fp16 = concat(axis = var_45992, interleave = var_47437_interleave_0, values = (var_47289_cast_fp16, var_47291_cast_fp16, var_47293_cast_fp16, var_47295_cast_fp16))[name = string("op_47437_cast_fp16")]; + bool var_47439_interleave_0 = const()[name = string("op_47439_interleave_0"), val = bool(false)]; + tensor var_47439_cast_fp16 = concat(axis = var_45992, interleave = var_47439_interleave_0, values = (var_47297_cast_fp16, var_47299_cast_fp16, var_47301_cast_fp16, var_47303_cast_fp16))[name = string("op_47439_cast_fp16")]; + bool var_47441_interleave_0 = const()[name = string("op_47441_interleave_0"), val = bool(false)]; + tensor var_47441_cast_fp16 = concat(axis = var_45992, interleave = var_47441_interleave_0, values = (var_47305_cast_fp16, var_47307_cast_fp16, var_47309_cast_fp16, var_47311_cast_fp16))[name = string("op_47441_cast_fp16")]; + bool var_47443_interleave_0 = const()[name = string("op_47443_interleave_0"), val = bool(false)]; + tensor var_47443_cast_fp16 = concat(axis = var_45992, interleave = var_47443_interleave_0, values = (var_47313_cast_fp16, var_47315_cast_fp16, var_47317_cast_fp16, var_47319_cast_fp16))[name = string("op_47443_cast_fp16")]; + bool var_47445_interleave_0 = const()[name = string("op_47445_interleave_0"), val = bool(false)]; + tensor var_47445_cast_fp16 = concat(axis = var_45992, interleave = var_47445_interleave_0, values = (var_47321_cast_fp16, var_47323_cast_fp16, var_47325_cast_fp16, var_47327_cast_fp16))[name = string("op_47445_cast_fp16")]; + bool var_47447_interleave_0 = const()[name = string("op_47447_interleave_0"), val = bool(false)]; + tensor var_47447_cast_fp16 = concat(axis = var_45992, interleave = var_47447_interleave_0, values = (var_47329_cast_fp16, var_47331_cast_fp16, var_47333_cast_fp16, var_47335_cast_fp16))[name = string("op_47447_cast_fp16")]; + bool var_47449_interleave_0 = const()[name = string("op_47449_interleave_0"), val = bool(false)]; + tensor var_47449_cast_fp16 = concat(axis = var_45992, interleave = var_47449_interleave_0, values = (var_47337_cast_fp16, var_47339_cast_fp16, var_47341_cast_fp16, var_47343_cast_fp16))[name = string("op_47449_cast_fp16")]; + bool var_47451_interleave_0 = const()[name = string("op_47451_interleave_0"), val = bool(false)]; + tensor var_47451_cast_fp16 = concat(axis = var_45992, interleave = var_47451_interleave_0, values = (var_47345_cast_fp16, var_47347_cast_fp16, var_47349_cast_fp16, var_47351_cast_fp16))[name = string("op_47451_cast_fp16")]; + bool var_47453_interleave_0 = const()[name = string("op_47453_interleave_0"), val = bool(false)]; + tensor var_47453_cast_fp16 = concat(axis = var_45992, interleave = var_47453_interleave_0, values = (var_47353_cast_fp16, var_47355_cast_fp16, var_47357_cast_fp16, var_47359_cast_fp16))[name = string("op_47453_cast_fp16")]; + bool var_47455_interleave_0 = const()[name = string("op_47455_interleave_0"), val = bool(false)]; + tensor var_47455_cast_fp16 = concat(axis = var_45992, interleave = var_47455_interleave_0, values = (var_47361_cast_fp16, var_47363_cast_fp16, var_47365_cast_fp16, var_47367_cast_fp16))[name = string("op_47455_cast_fp16")]; + bool var_47457_interleave_0 = const()[name = string("op_47457_interleave_0"), val = bool(false)]; + tensor var_47457_cast_fp16 = concat(axis = var_45992, interleave = var_47457_interleave_0, values = (var_47369_cast_fp16, var_47371_cast_fp16, var_47373_cast_fp16, var_47375_cast_fp16))[name = string("op_47457_cast_fp16")]; + bool var_47459_interleave_0 = const()[name = string("op_47459_interleave_0"), val = bool(false)]; + tensor var_47459_cast_fp16 = concat(axis = var_45992, interleave = var_47459_interleave_0, values = (var_47377_cast_fp16, var_47379_cast_fp16, var_47381_cast_fp16, var_47383_cast_fp16))[name = string("op_47459_cast_fp16")]; + bool var_47461_interleave_0 = const()[name = string("op_47461_interleave_0"), val = bool(false)]; + tensor var_47461_cast_fp16 = concat(axis = var_45992, interleave = var_47461_interleave_0, values = (var_47385_cast_fp16, var_47387_cast_fp16, var_47389_cast_fp16, var_47391_cast_fp16))[name = string("op_47461_cast_fp16")]; + bool var_47463_interleave_0 = const()[name = string("op_47463_interleave_0"), val = bool(false)]; + tensor var_47463_cast_fp16 = concat(axis = var_45992, interleave = var_47463_interleave_0, values = (var_47393_cast_fp16, var_47395_cast_fp16, var_47397_cast_fp16, var_47399_cast_fp16))[name = string("op_47463_cast_fp16")]; + bool var_47465_interleave_0 = const()[name = string("op_47465_interleave_0"), val = bool(false)]; + tensor var_47465_cast_fp16 = concat(axis = var_45992, interleave = var_47465_interleave_0, values = (var_47401_cast_fp16, var_47403_cast_fp16, var_47405_cast_fp16, var_47407_cast_fp16))[name = string("op_47465_cast_fp16")]; + bool var_47467_interleave_0 = const()[name = string("op_47467_interleave_0"), val = bool(false)]; + tensor var_47467_cast_fp16 = concat(axis = var_45992, interleave = var_47467_interleave_0, values = (var_47409_cast_fp16, var_47411_cast_fp16, var_47413_cast_fp16, var_47415_cast_fp16))[name = string("op_47467_cast_fp16")]; + bool var_47469_interleave_0 = const()[name = string("op_47469_interleave_0"), val = bool(false)]; + tensor var_47469_cast_fp16 = concat(axis = var_45992, interleave = var_47469_interleave_0, values = (var_47417_cast_fp16, var_47419_cast_fp16, var_47421_cast_fp16, var_47423_cast_fp16))[name = string("op_47469_cast_fp16")]; + bool var_47471_interleave_0 = const()[name = string("op_47471_interleave_0"), val = bool(false)]; + tensor var_47471_cast_fp16 = concat(axis = var_45992, interleave = var_47471_interleave_0, values = (var_47425_cast_fp16, var_47427_cast_fp16, var_47429_cast_fp16, var_47431_cast_fp16))[name = string("op_47471_cast_fp16")]; + bool input_241_interleave_0 = const()[name = string("input_241_interleave_0"), val = bool(false)]; + tensor input_241_cast_fp16 = concat(axis = var_46017, interleave = input_241_interleave_0, values = (var_47433_cast_fp16, var_47435_cast_fp16, var_47437_cast_fp16, var_47439_cast_fp16, var_47441_cast_fp16, var_47443_cast_fp16, var_47445_cast_fp16, var_47447_cast_fp16, var_47449_cast_fp16, var_47451_cast_fp16, var_47453_cast_fp16, var_47455_cast_fp16, var_47457_cast_fp16, var_47459_cast_fp16, var_47461_cast_fp16, var_47463_cast_fp16, var_47465_cast_fp16, var_47467_cast_fp16, var_47469_cast_fp16, var_47471_cast_fp16))[name = string("input_241_cast_fp16")]; + string obj_123_pad_type_0 = const()[name = string("obj_123_pad_type_0"), val = string("valid")]; + tensor obj_123_strides_0 = const()[name = string("obj_123_strides_0"), val = tensor([1, 1])]; + tensor obj_123_pad_0 = const()[name = string("obj_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_123_dilations_0 = const()[name = string("obj_123_dilations_0"), val = tensor([1, 1])]; + int32 obj_123_groups_0 = const()[name = string("obj_123_groups_0"), val = int32(1)]; + tensor layers_30_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1205103680)))]; + tensor layers_30_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_30_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208380544)))]; + tensor obj_123_cast_fp16 = conv(bias = layers_30_self_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_30_self_attn_o_proj_weight_to_fp16, x = input_241_cast_fp16)[name = string("obj_123_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_123_cast_fp16)[name = string("inputs_123_cast_fp16")]; + tensor out_123_axes_0 = const()[name = string("out_123_axes_0"), val = tensor([1])]; + fp16 var_47490_to_fp16 = const()[name = string("op_47490_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_47490_to_fp16, x = inputs_123_cast_fp16)[name = string("out_123_cast_fp16")]; + tensor input_243_gamma_0_to_fp16 = const()[name = string("input_243_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208383168)))]; + tensor input_243_beta_0_to_fp16 = const()[name = string("input_243_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208385792)))]; + fp16 input_243_epsilon_0_to_fp16 = const()[name = string("input_243_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_243_cast_fp16 = batch_norm(beta = input_243_beta_0_to_fp16, epsilon = input_243_epsilon_0_to_fp16, gamma = input_243_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = string("input_243_cast_fp16")]; + string input_245_pad_type_0 = const()[name = string("input_245_pad_type_0"), val = string("valid")]; + tensor input_245_strides_0 = const()[name = string("input_245_strides_0"), val = tensor([1, 1])]; + tensor input_245_pad_0 = const()[name = string("input_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_245_dilations_0 = const()[name = string("input_245_dilations_0"), val = tensor([1, 1])]; + int32 input_245_groups_0 = const()[name = string("input_245_groups_0"), val = int32(1)]; + tensor layers_30_fc1_weight_to_fp16 = const()[name = string("layers_30_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208388416)))]; + tensor layers_30_fc1_bias_to_fp16 = const()[name = string("layers_30_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221495680)))]; + tensor input_245_cast_fp16 = conv(bias = layers_30_fc1_bias_to_fp16, dilations = input_245_dilations_0, groups = input_245_groups_0, pad = input_245_pad_0, pad_type = input_245_pad_type_0, strides = input_245_strides_0, weight = layers_30_fc1_weight_to_fp16, x = input_243_cast_fp16)[name = string("input_245_cast_fp16")]; + string input_247_mode_0 = const()[name = string("input_247_mode_0"), val = string("EXACT")]; + tensor input_247_cast_fp16 = gelu(mode = input_247_mode_0, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")]; + string hidden_states_65_pad_type_0 = const()[name = string("hidden_states_65_pad_type_0"), val = string("valid")]; + tensor hidden_states_65_strides_0 = const()[name = string("hidden_states_65_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_65_pad_0 = const()[name = string("hidden_states_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_65_dilations_0 = const()[name = string("hidden_states_65_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_65_groups_0 = const()[name = string("hidden_states_65_groups_0"), val = int32(1)]; + tensor layers_30_fc2_weight_to_fp16 = const()[name = string("layers_30_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1221505984)))]; + tensor layers_30_fc2_bias_to_fp16 = const()[name = string("layers_30_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234613248)))]; + tensor hidden_states_65_cast_fp16 = conv(bias = layers_30_fc2_bias_to_fp16, dilations = hidden_states_65_dilations_0, groups = hidden_states_65_groups_0, pad = hidden_states_65_pad_0, pad_type = hidden_states_65_pad_type_0, strides = hidden_states_65_strides_0, weight = layers_30_fc2_weight_to_fp16, x = input_247_cast_fp16)[name = string("hidden_states_65_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = hidden_states_65_cast_fp16)[name = string("inputs_125_cast_fp16")]; + int32 var_47519 = const()[name = string("op_47519"), val = int32(3)]; + int32 var_47544 = const()[name = string("op_47544"), val = int32(1)]; + tensor out_125_axes_0 = const()[name = string("out_125_axes_0"), val = tensor([1])]; + fp16 var_47561_to_fp16 = const()[name = string("op_47561_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_47561_to_fp16, x = inputs_125_cast_fp16)[name = string("out_125_cast_fp16")]; + tensor obj_125_gamma_0_to_fp16 = const()[name = string("obj_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234615872)))]; + tensor obj_125_beta_0_to_fp16 = const()[name = string("obj_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234618496)))]; + fp16 obj_125_epsilon_0_to_fp16 = const()[name = string("obj_125_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_125_cast_fp16 = batch_norm(beta = obj_125_beta_0_to_fp16, epsilon = obj_125_epsilon_0_to_fp16, gamma = obj_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = string("obj_125_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1234621120)))]; + tensor layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237897984)))]; + tensor query_cast_fp16 = conv(bias = layers_31_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_31_self_attn_q_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1237900608)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_31_self_attn_k_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1241177472)))]; + tensor layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244454336)))]; + tensor value_cast_fp16 = conv(bias = layers_31_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_31_self_attn_v_proj_weight_to_fp16, x = obj_125_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_47599_begin_0 = const()[name = string("op_47599_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47599_end_0 = const()[name = string("op_47599_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47599_end_mask_0 = const()[name = string("op_47599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47599_cast_fp16 = slice_by_index(begin = var_47599_begin_0, end = var_47599_end_0, end_mask = var_47599_end_mask_0, x = query_cast_fp16)[name = string("op_47599_cast_fp16")]; + tensor var_47603_begin_0 = const()[name = string("op_47603_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47603_end_0 = const()[name = string("op_47603_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_47603_end_mask_0 = const()[name = string("op_47603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47603_cast_fp16 = slice_by_index(begin = var_47603_begin_0, end = var_47603_end_0, end_mask = var_47603_end_mask_0, x = query_cast_fp16)[name = string("op_47603_cast_fp16")]; + tensor var_47607_begin_0 = const()[name = string("op_47607_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47607_end_0 = const()[name = string("op_47607_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_47607_end_mask_0 = const()[name = string("op_47607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47607_cast_fp16 = slice_by_index(begin = var_47607_begin_0, end = var_47607_end_0, end_mask = var_47607_end_mask_0, x = query_cast_fp16)[name = string("op_47607_cast_fp16")]; + tensor var_47611_begin_0 = const()[name = string("op_47611_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47611_end_0 = const()[name = string("op_47611_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_47611_end_mask_0 = const()[name = string("op_47611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47611_cast_fp16 = slice_by_index(begin = var_47611_begin_0, end = var_47611_end_0, end_mask = var_47611_end_mask_0, x = query_cast_fp16)[name = string("op_47611_cast_fp16")]; + tensor var_47615_begin_0 = const()[name = string("op_47615_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47615_end_0 = const()[name = string("op_47615_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_47615_end_mask_0 = const()[name = string("op_47615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47615_cast_fp16 = slice_by_index(begin = var_47615_begin_0, end = var_47615_end_0, end_mask = var_47615_end_mask_0, x = query_cast_fp16)[name = string("op_47615_cast_fp16")]; + tensor var_47619_begin_0 = const()[name = string("op_47619_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47619_end_0 = const()[name = string("op_47619_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_47619_end_mask_0 = const()[name = string("op_47619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47619_cast_fp16 = slice_by_index(begin = var_47619_begin_0, end = var_47619_end_0, end_mask = var_47619_end_mask_0, x = query_cast_fp16)[name = string("op_47619_cast_fp16")]; + tensor var_47623_begin_0 = const()[name = string("op_47623_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47623_end_0 = const()[name = string("op_47623_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_47623_end_mask_0 = const()[name = string("op_47623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47623_cast_fp16 = slice_by_index(begin = var_47623_begin_0, end = var_47623_end_0, end_mask = var_47623_end_mask_0, x = query_cast_fp16)[name = string("op_47623_cast_fp16")]; + tensor var_47627_begin_0 = const()[name = string("op_47627_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47627_end_0 = const()[name = string("op_47627_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_47627_end_mask_0 = const()[name = string("op_47627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47627_cast_fp16 = slice_by_index(begin = var_47627_begin_0, end = var_47627_end_0, end_mask = var_47627_end_mask_0, x = query_cast_fp16)[name = string("op_47627_cast_fp16")]; + tensor var_47631_begin_0 = const()[name = string("op_47631_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47631_end_0 = const()[name = string("op_47631_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_47631_end_mask_0 = const()[name = string("op_47631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47631_cast_fp16 = slice_by_index(begin = var_47631_begin_0, end = var_47631_end_0, end_mask = var_47631_end_mask_0, x = query_cast_fp16)[name = string("op_47631_cast_fp16")]; + tensor var_47635_begin_0 = const()[name = string("op_47635_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47635_end_0 = const()[name = string("op_47635_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_47635_end_mask_0 = const()[name = string("op_47635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47635_cast_fp16 = slice_by_index(begin = var_47635_begin_0, end = var_47635_end_0, end_mask = var_47635_end_mask_0, x = query_cast_fp16)[name = string("op_47635_cast_fp16")]; + tensor var_47639_begin_0 = const()[name = string("op_47639_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47639_end_0 = const()[name = string("op_47639_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_47639_end_mask_0 = const()[name = string("op_47639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47639_cast_fp16 = slice_by_index(begin = var_47639_begin_0, end = var_47639_end_0, end_mask = var_47639_end_mask_0, x = query_cast_fp16)[name = string("op_47639_cast_fp16")]; + tensor var_47643_begin_0 = const()[name = string("op_47643_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47643_end_0 = const()[name = string("op_47643_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_47643_end_mask_0 = const()[name = string("op_47643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47643_cast_fp16 = slice_by_index(begin = var_47643_begin_0, end = var_47643_end_0, end_mask = var_47643_end_mask_0, x = query_cast_fp16)[name = string("op_47643_cast_fp16")]; + tensor var_47647_begin_0 = const()[name = string("op_47647_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47647_end_0 = const()[name = string("op_47647_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_47647_end_mask_0 = const()[name = string("op_47647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47647_cast_fp16 = slice_by_index(begin = var_47647_begin_0, end = var_47647_end_0, end_mask = var_47647_end_mask_0, x = query_cast_fp16)[name = string("op_47647_cast_fp16")]; + tensor var_47651_begin_0 = const()[name = string("op_47651_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47651_end_0 = const()[name = string("op_47651_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_47651_end_mask_0 = const()[name = string("op_47651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47651_cast_fp16 = slice_by_index(begin = var_47651_begin_0, end = var_47651_end_0, end_mask = var_47651_end_mask_0, x = query_cast_fp16)[name = string("op_47651_cast_fp16")]; + tensor var_47655_begin_0 = const()[name = string("op_47655_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47655_end_0 = const()[name = string("op_47655_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_47655_end_mask_0 = const()[name = string("op_47655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47655_cast_fp16 = slice_by_index(begin = var_47655_begin_0, end = var_47655_end_0, end_mask = var_47655_end_mask_0, x = query_cast_fp16)[name = string("op_47655_cast_fp16")]; + tensor var_47659_begin_0 = const()[name = string("op_47659_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47659_end_0 = const()[name = string("op_47659_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_47659_end_mask_0 = const()[name = string("op_47659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47659_cast_fp16 = slice_by_index(begin = var_47659_begin_0, end = var_47659_end_0, end_mask = var_47659_end_mask_0, x = query_cast_fp16)[name = string("op_47659_cast_fp16")]; + tensor var_47663_begin_0 = const()[name = string("op_47663_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47663_end_0 = const()[name = string("op_47663_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_47663_end_mask_0 = const()[name = string("op_47663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47663_cast_fp16 = slice_by_index(begin = var_47663_begin_0, end = var_47663_end_0, end_mask = var_47663_end_mask_0, x = query_cast_fp16)[name = string("op_47663_cast_fp16")]; + tensor var_47667_begin_0 = const()[name = string("op_47667_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47667_end_0 = const()[name = string("op_47667_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_47667_end_mask_0 = const()[name = string("op_47667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47667_cast_fp16 = slice_by_index(begin = var_47667_begin_0, end = var_47667_end_0, end_mask = var_47667_end_mask_0, x = query_cast_fp16)[name = string("op_47667_cast_fp16")]; + tensor var_47671_begin_0 = const()[name = string("op_47671_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47671_end_0 = const()[name = string("op_47671_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_47671_end_mask_0 = const()[name = string("op_47671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47671_cast_fp16 = slice_by_index(begin = var_47671_begin_0, end = var_47671_end_0, end_mask = var_47671_end_mask_0, x = query_cast_fp16)[name = string("op_47671_cast_fp16")]; + tensor var_47675_begin_0 = const()[name = string("op_47675_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47675_end_0 = const()[name = string("op_47675_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_47675_end_mask_0 = const()[name = string("op_47675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47675_cast_fp16 = slice_by_index(begin = var_47675_begin_0, end = var_47675_end_0, end_mask = var_47675_end_mask_0, x = query_cast_fp16)[name = string("op_47675_cast_fp16")]; + tensor var_47684_begin_0 = const()[name = string("op_47684_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47684_end_0 = const()[name = string("op_47684_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47684_end_mask_0 = const()[name = string("op_47684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47684_cast_fp16 = slice_by_index(begin = var_47684_begin_0, end = var_47684_end_0, end_mask = var_47684_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47684_cast_fp16")]; + tensor var_47691_begin_0 = const()[name = string("op_47691_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47691_end_0 = const()[name = string("op_47691_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47691_end_mask_0 = const()[name = string("op_47691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47691_cast_fp16 = slice_by_index(begin = var_47691_begin_0, end = var_47691_end_0, end_mask = var_47691_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47691_cast_fp16")]; + tensor var_47698_begin_0 = const()[name = string("op_47698_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47698_end_0 = const()[name = string("op_47698_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47698_end_mask_0 = const()[name = string("op_47698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47698_cast_fp16 = slice_by_index(begin = var_47698_begin_0, end = var_47698_end_0, end_mask = var_47698_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47698_cast_fp16")]; + tensor var_47705_begin_0 = const()[name = string("op_47705_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47705_end_0 = const()[name = string("op_47705_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47705_end_mask_0 = const()[name = string("op_47705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47705_cast_fp16 = slice_by_index(begin = var_47705_begin_0, end = var_47705_end_0, end_mask = var_47705_end_mask_0, x = var_47599_cast_fp16)[name = string("op_47705_cast_fp16")]; + tensor var_47712_begin_0 = const()[name = string("op_47712_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47712_end_0 = const()[name = string("op_47712_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47712_end_mask_0 = const()[name = string("op_47712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47712_cast_fp16 = slice_by_index(begin = var_47712_begin_0, end = var_47712_end_0, end_mask = var_47712_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47712_cast_fp16")]; + tensor var_47719_begin_0 = const()[name = string("op_47719_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47719_end_0 = const()[name = string("op_47719_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47719_end_mask_0 = const()[name = string("op_47719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47719_cast_fp16 = slice_by_index(begin = var_47719_begin_0, end = var_47719_end_0, end_mask = var_47719_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47719_cast_fp16")]; + tensor var_47726_begin_0 = const()[name = string("op_47726_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47726_end_0 = const()[name = string("op_47726_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47726_end_mask_0 = const()[name = string("op_47726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47726_cast_fp16 = slice_by_index(begin = var_47726_begin_0, end = var_47726_end_0, end_mask = var_47726_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47726_cast_fp16")]; + tensor var_47733_begin_0 = const()[name = string("op_47733_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47733_end_0 = const()[name = string("op_47733_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47733_end_mask_0 = const()[name = string("op_47733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47733_cast_fp16 = slice_by_index(begin = var_47733_begin_0, end = var_47733_end_0, end_mask = var_47733_end_mask_0, x = var_47603_cast_fp16)[name = string("op_47733_cast_fp16")]; + tensor var_47740_begin_0 = const()[name = string("op_47740_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47740_end_0 = const()[name = string("op_47740_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47740_end_mask_0 = const()[name = string("op_47740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47740_cast_fp16 = slice_by_index(begin = var_47740_begin_0, end = var_47740_end_0, end_mask = var_47740_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47740_cast_fp16")]; + tensor var_47747_begin_0 = const()[name = string("op_47747_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47747_end_0 = const()[name = string("op_47747_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47747_end_mask_0 = const()[name = string("op_47747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47747_cast_fp16 = slice_by_index(begin = var_47747_begin_0, end = var_47747_end_0, end_mask = var_47747_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47747_cast_fp16")]; + tensor var_47754_begin_0 = const()[name = string("op_47754_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47754_end_0 = const()[name = string("op_47754_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47754_end_mask_0 = const()[name = string("op_47754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47754_cast_fp16 = slice_by_index(begin = var_47754_begin_0, end = var_47754_end_0, end_mask = var_47754_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47754_cast_fp16")]; + tensor var_47761_begin_0 = const()[name = string("op_47761_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47761_end_0 = const()[name = string("op_47761_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47761_end_mask_0 = const()[name = string("op_47761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47761_cast_fp16 = slice_by_index(begin = var_47761_begin_0, end = var_47761_end_0, end_mask = var_47761_end_mask_0, x = var_47607_cast_fp16)[name = string("op_47761_cast_fp16")]; + tensor var_47768_begin_0 = const()[name = string("op_47768_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47768_end_0 = const()[name = string("op_47768_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47768_end_mask_0 = const()[name = string("op_47768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47768_cast_fp16 = slice_by_index(begin = var_47768_begin_0, end = var_47768_end_0, end_mask = var_47768_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47768_cast_fp16")]; + tensor var_47775_begin_0 = const()[name = string("op_47775_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47775_end_0 = const()[name = string("op_47775_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47775_end_mask_0 = const()[name = string("op_47775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47775_cast_fp16 = slice_by_index(begin = var_47775_begin_0, end = var_47775_end_0, end_mask = var_47775_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47775_cast_fp16")]; + tensor var_47782_begin_0 = const()[name = string("op_47782_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47782_end_0 = const()[name = string("op_47782_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47782_end_mask_0 = const()[name = string("op_47782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47782_cast_fp16 = slice_by_index(begin = var_47782_begin_0, end = var_47782_end_0, end_mask = var_47782_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47782_cast_fp16")]; + tensor var_47789_begin_0 = const()[name = string("op_47789_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47789_end_0 = const()[name = string("op_47789_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47789_end_mask_0 = const()[name = string("op_47789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47789_cast_fp16 = slice_by_index(begin = var_47789_begin_0, end = var_47789_end_0, end_mask = var_47789_end_mask_0, x = var_47611_cast_fp16)[name = string("op_47789_cast_fp16")]; + tensor var_47796_begin_0 = const()[name = string("op_47796_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47796_end_0 = const()[name = string("op_47796_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47796_end_mask_0 = const()[name = string("op_47796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47796_cast_fp16 = slice_by_index(begin = var_47796_begin_0, end = var_47796_end_0, end_mask = var_47796_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47796_cast_fp16")]; + tensor var_47803_begin_0 = const()[name = string("op_47803_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47803_end_0 = const()[name = string("op_47803_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47803_end_mask_0 = const()[name = string("op_47803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47803_cast_fp16 = slice_by_index(begin = var_47803_begin_0, end = var_47803_end_0, end_mask = var_47803_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47803_cast_fp16")]; + tensor var_47810_begin_0 = const()[name = string("op_47810_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47810_end_0 = const()[name = string("op_47810_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47810_end_mask_0 = const()[name = string("op_47810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47810_cast_fp16 = slice_by_index(begin = var_47810_begin_0, end = var_47810_end_0, end_mask = var_47810_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47810_cast_fp16")]; + tensor var_47817_begin_0 = const()[name = string("op_47817_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47817_end_0 = const()[name = string("op_47817_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47817_end_mask_0 = const()[name = string("op_47817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47817_cast_fp16 = slice_by_index(begin = var_47817_begin_0, end = var_47817_end_0, end_mask = var_47817_end_mask_0, x = var_47615_cast_fp16)[name = string("op_47817_cast_fp16")]; + tensor var_47824_begin_0 = const()[name = string("op_47824_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47824_end_0 = const()[name = string("op_47824_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47824_end_mask_0 = const()[name = string("op_47824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47824_cast_fp16 = slice_by_index(begin = var_47824_begin_0, end = var_47824_end_0, end_mask = var_47824_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47824_cast_fp16")]; + tensor var_47831_begin_0 = const()[name = string("op_47831_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47831_end_0 = const()[name = string("op_47831_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47831_end_mask_0 = const()[name = string("op_47831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47831_cast_fp16 = slice_by_index(begin = var_47831_begin_0, end = var_47831_end_0, end_mask = var_47831_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47831_cast_fp16")]; + tensor var_47838_begin_0 = const()[name = string("op_47838_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47838_end_0 = const()[name = string("op_47838_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47838_end_mask_0 = const()[name = string("op_47838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47838_cast_fp16 = slice_by_index(begin = var_47838_begin_0, end = var_47838_end_0, end_mask = var_47838_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47838_cast_fp16")]; + tensor var_47845_begin_0 = const()[name = string("op_47845_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47845_end_0 = const()[name = string("op_47845_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47845_end_mask_0 = const()[name = string("op_47845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47845_cast_fp16 = slice_by_index(begin = var_47845_begin_0, end = var_47845_end_0, end_mask = var_47845_end_mask_0, x = var_47619_cast_fp16)[name = string("op_47845_cast_fp16")]; + tensor var_47852_begin_0 = const()[name = string("op_47852_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47852_end_0 = const()[name = string("op_47852_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47852_end_mask_0 = const()[name = string("op_47852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47852_cast_fp16 = slice_by_index(begin = var_47852_begin_0, end = var_47852_end_0, end_mask = var_47852_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47852_cast_fp16")]; + tensor var_47859_begin_0 = const()[name = string("op_47859_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47859_end_0 = const()[name = string("op_47859_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47859_end_mask_0 = const()[name = string("op_47859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47859_cast_fp16 = slice_by_index(begin = var_47859_begin_0, end = var_47859_end_0, end_mask = var_47859_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47859_cast_fp16")]; + tensor var_47866_begin_0 = const()[name = string("op_47866_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47866_end_0 = const()[name = string("op_47866_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47866_end_mask_0 = const()[name = string("op_47866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47866_cast_fp16 = slice_by_index(begin = var_47866_begin_0, end = var_47866_end_0, end_mask = var_47866_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47866_cast_fp16")]; + tensor var_47873_begin_0 = const()[name = string("op_47873_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47873_end_0 = const()[name = string("op_47873_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47873_end_mask_0 = const()[name = string("op_47873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47873_cast_fp16 = slice_by_index(begin = var_47873_begin_0, end = var_47873_end_0, end_mask = var_47873_end_mask_0, x = var_47623_cast_fp16)[name = string("op_47873_cast_fp16")]; + tensor var_47880_begin_0 = const()[name = string("op_47880_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47880_end_0 = const()[name = string("op_47880_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47880_end_mask_0 = const()[name = string("op_47880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47880_cast_fp16 = slice_by_index(begin = var_47880_begin_0, end = var_47880_end_0, end_mask = var_47880_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47880_cast_fp16")]; + tensor var_47887_begin_0 = const()[name = string("op_47887_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47887_end_0 = const()[name = string("op_47887_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47887_end_mask_0 = const()[name = string("op_47887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47887_cast_fp16 = slice_by_index(begin = var_47887_begin_0, end = var_47887_end_0, end_mask = var_47887_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47887_cast_fp16")]; + tensor var_47894_begin_0 = const()[name = string("op_47894_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47894_end_0 = const()[name = string("op_47894_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47894_end_mask_0 = const()[name = string("op_47894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47894_cast_fp16 = slice_by_index(begin = var_47894_begin_0, end = var_47894_end_0, end_mask = var_47894_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47894_cast_fp16")]; + tensor var_47901_begin_0 = const()[name = string("op_47901_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47901_end_0 = const()[name = string("op_47901_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47901_end_mask_0 = const()[name = string("op_47901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47901_cast_fp16 = slice_by_index(begin = var_47901_begin_0, end = var_47901_end_0, end_mask = var_47901_end_mask_0, x = var_47627_cast_fp16)[name = string("op_47901_cast_fp16")]; + tensor var_47908_begin_0 = const()[name = string("op_47908_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47908_end_0 = const()[name = string("op_47908_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47908_end_mask_0 = const()[name = string("op_47908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47908_cast_fp16 = slice_by_index(begin = var_47908_begin_0, end = var_47908_end_0, end_mask = var_47908_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47908_cast_fp16")]; + tensor var_47915_begin_0 = const()[name = string("op_47915_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47915_end_0 = const()[name = string("op_47915_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47915_end_mask_0 = const()[name = string("op_47915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47915_cast_fp16 = slice_by_index(begin = var_47915_begin_0, end = var_47915_end_0, end_mask = var_47915_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47915_cast_fp16")]; + tensor var_47922_begin_0 = const()[name = string("op_47922_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47922_end_0 = const()[name = string("op_47922_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47922_end_mask_0 = const()[name = string("op_47922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47922_cast_fp16 = slice_by_index(begin = var_47922_begin_0, end = var_47922_end_0, end_mask = var_47922_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47922_cast_fp16")]; + tensor var_47929_begin_0 = const()[name = string("op_47929_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47929_end_0 = const()[name = string("op_47929_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47929_end_mask_0 = const()[name = string("op_47929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47929_cast_fp16 = slice_by_index(begin = var_47929_begin_0, end = var_47929_end_0, end_mask = var_47929_end_mask_0, x = var_47631_cast_fp16)[name = string("op_47929_cast_fp16")]; + tensor var_47936_begin_0 = const()[name = string("op_47936_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47936_end_0 = const()[name = string("op_47936_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47936_end_mask_0 = const()[name = string("op_47936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47936_cast_fp16 = slice_by_index(begin = var_47936_begin_0, end = var_47936_end_0, end_mask = var_47936_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47936_cast_fp16")]; + tensor var_47943_begin_0 = const()[name = string("op_47943_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47943_end_0 = const()[name = string("op_47943_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47943_end_mask_0 = const()[name = string("op_47943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47943_cast_fp16 = slice_by_index(begin = var_47943_begin_0, end = var_47943_end_0, end_mask = var_47943_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47943_cast_fp16")]; + tensor var_47950_begin_0 = const()[name = string("op_47950_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47950_end_0 = const()[name = string("op_47950_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47950_end_mask_0 = const()[name = string("op_47950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47950_cast_fp16 = slice_by_index(begin = var_47950_begin_0, end = var_47950_end_0, end_mask = var_47950_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47950_cast_fp16")]; + tensor var_47957_begin_0 = const()[name = string("op_47957_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47957_end_0 = const()[name = string("op_47957_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47957_end_mask_0 = const()[name = string("op_47957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47957_cast_fp16 = slice_by_index(begin = var_47957_begin_0, end = var_47957_end_0, end_mask = var_47957_end_mask_0, x = var_47635_cast_fp16)[name = string("op_47957_cast_fp16")]; + tensor var_47964_begin_0 = const()[name = string("op_47964_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47964_end_0 = const()[name = string("op_47964_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47964_end_mask_0 = const()[name = string("op_47964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47964_cast_fp16 = slice_by_index(begin = var_47964_begin_0, end = var_47964_end_0, end_mask = var_47964_end_mask_0, x = var_47639_cast_fp16)[name = string("op_47964_cast_fp16")]; + tensor var_47971_begin_0 = const()[name = string("op_47971_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47971_end_0 = const()[name = string("op_47971_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47971_end_mask_0 = const()[name = string("op_47971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47971_cast_fp16 = slice_by_index(begin = var_47971_begin_0, end = var_47971_end_0, end_mask = var_47971_end_mask_0, x = var_47639_cast_fp16)[name = string("op_47971_cast_fp16")]; + tensor var_47978_begin_0 = const()[name = string("op_47978_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47978_end_0 = const()[name = string("op_47978_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47978_end_mask_0 = const()[name = string("op_47978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47978_cast_fp16 = slice_by_index(begin = var_47978_begin_0, end = var_47978_end_0, end_mask = var_47978_end_mask_0, x = var_47639_cast_fp16)[name = string("op_47978_cast_fp16")]; + tensor var_47985_begin_0 = const()[name = string("op_47985_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47985_end_0 = const()[name = string("op_47985_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47985_end_mask_0 = const()[name = string("op_47985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47985_cast_fp16 = slice_by_index(begin = var_47985_begin_0, end = var_47985_end_0, end_mask = var_47985_end_mask_0, x = var_47639_cast_fp16)[name = string("op_47985_cast_fp16")]; + tensor var_47992_begin_0 = const()[name = string("op_47992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47992_end_0 = const()[name = string("op_47992_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47992_end_mask_0 = const()[name = string("op_47992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47992_cast_fp16 = slice_by_index(begin = var_47992_begin_0, end = var_47992_end_0, end_mask = var_47992_end_mask_0, x = var_47643_cast_fp16)[name = string("op_47992_cast_fp16")]; + tensor var_47999_begin_0 = const()[name = string("op_47999_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47999_end_0 = const()[name = string("op_47999_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47999_end_mask_0 = const()[name = string("op_47999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47999_cast_fp16 = slice_by_index(begin = var_47999_begin_0, end = var_47999_end_0, end_mask = var_47999_end_mask_0, x = var_47643_cast_fp16)[name = string("op_47999_cast_fp16")]; + tensor var_48006_begin_0 = const()[name = string("op_48006_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48006_end_0 = const()[name = string("op_48006_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48006_end_mask_0 = const()[name = string("op_48006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48006_cast_fp16 = slice_by_index(begin = var_48006_begin_0, end = var_48006_end_0, end_mask = var_48006_end_mask_0, x = var_47643_cast_fp16)[name = string("op_48006_cast_fp16")]; + tensor var_48013_begin_0 = const()[name = string("op_48013_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48013_end_0 = const()[name = string("op_48013_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48013_end_mask_0 = const()[name = string("op_48013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48013_cast_fp16 = slice_by_index(begin = var_48013_begin_0, end = var_48013_end_0, end_mask = var_48013_end_mask_0, x = var_47643_cast_fp16)[name = string("op_48013_cast_fp16")]; + tensor var_48020_begin_0 = const()[name = string("op_48020_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48020_end_0 = const()[name = string("op_48020_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48020_end_mask_0 = const()[name = string("op_48020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48020_cast_fp16 = slice_by_index(begin = var_48020_begin_0, end = var_48020_end_0, end_mask = var_48020_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48020_cast_fp16")]; + tensor var_48027_begin_0 = const()[name = string("op_48027_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48027_end_0 = const()[name = string("op_48027_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48027_end_mask_0 = const()[name = string("op_48027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48027_cast_fp16 = slice_by_index(begin = var_48027_begin_0, end = var_48027_end_0, end_mask = var_48027_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48027_cast_fp16")]; + tensor var_48034_begin_0 = const()[name = string("op_48034_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48034_end_0 = const()[name = string("op_48034_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48034_end_mask_0 = const()[name = string("op_48034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48034_cast_fp16 = slice_by_index(begin = var_48034_begin_0, end = var_48034_end_0, end_mask = var_48034_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48034_cast_fp16")]; + tensor var_48041_begin_0 = const()[name = string("op_48041_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48041_end_0 = const()[name = string("op_48041_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48041_end_mask_0 = const()[name = string("op_48041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48041_cast_fp16 = slice_by_index(begin = var_48041_begin_0, end = var_48041_end_0, end_mask = var_48041_end_mask_0, x = var_47647_cast_fp16)[name = string("op_48041_cast_fp16")]; + tensor var_48048_begin_0 = const()[name = string("op_48048_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48048_end_0 = const()[name = string("op_48048_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48048_end_mask_0 = const()[name = string("op_48048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48048_cast_fp16 = slice_by_index(begin = var_48048_begin_0, end = var_48048_end_0, end_mask = var_48048_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48048_cast_fp16")]; + tensor var_48055_begin_0 = const()[name = string("op_48055_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48055_end_0 = const()[name = string("op_48055_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48055_end_mask_0 = const()[name = string("op_48055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48055_cast_fp16 = slice_by_index(begin = var_48055_begin_0, end = var_48055_end_0, end_mask = var_48055_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48055_cast_fp16")]; + tensor var_48062_begin_0 = const()[name = string("op_48062_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48062_end_0 = const()[name = string("op_48062_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48062_end_mask_0 = const()[name = string("op_48062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48062_cast_fp16 = slice_by_index(begin = var_48062_begin_0, end = var_48062_end_0, end_mask = var_48062_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48062_cast_fp16")]; + tensor var_48069_begin_0 = const()[name = string("op_48069_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48069_end_0 = const()[name = string("op_48069_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48069_end_mask_0 = const()[name = string("op_48069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48069_cast_fp16 = slice_by_index(begin = var_48069_begin_0, end = var_48069_end_0, end_mask = var_48069_end_mask_0, x = var_47651_cast_fp16)[name = string("op_48069_cast_fp16")]; + tensor var_48076_begin_0 = const()[name = string("op_48076_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48076_end_0 = const()[name = string("op_48076_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48076_end_mask_0 = const()[name = string("op_48076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48076_cast_fp16 = slice_by_index(begin = var_48076_begin_0, end = var_48076_end_0, end_mask = var_48076_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48076_cast_fp16")]; + tensor var_48083_begin_0 = const()[name = string("op_48083_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48083_end_0 = const()[name = string("op_48083_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48083_end_mask_0 = const()[name = string("op_48083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48083_cast_fp16 = slice_by_index(begin = var_48083_begin_0, end = var_48083_end_0, end_mask = var_48083_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48083_cast_fp16")]; + tensor var_48090_begin_0 = const()[name = string("op_48090_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48090_end_0 = const()[name = string("op_48090_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48090_end_mask_0 = const()[name = string("op_48090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48090_cast_fp16 = slice_by_index(begin = var_48090_begin_0, end = var_48090_end_0, end_mask = var_48090_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48090_cast_fp16")]; + tensor var_48097_begin_0 = const()[name = string("op_48097_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48097_end_0 = const()[name = string("op_48097_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48097_end_mask_0 = const()[name = string("op_48097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48097_cast_fp16 = slice_by_index(begin = var_48097_begin_0, end = var_48097_end_0, end_mask = var_48097_end_mask_0, x = var_47655_cast_fp16)[name = string("op_48097_cast_fp16")]; + tensor var_48104_begin_0 = const()[name = string("op_48104_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48104_end_0 = const()[name = string("op_48104_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48104_end_mask_0 = const()[name = string("op_48104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48104_cast_fp16 = slice_by_index(begin = var_48104_begin_0, end = var_48104_end_0, end_mask = var_48104_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48104_cast_fp16")]; + tensor var_48111_begin_0 = const()[name = string("op_48111_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48111_end_0 = const()[name = string("op_48111_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48111_end_mask_0 = const()[name = string("op_48111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48111_cast_fp16 = slice_by_index(begin = var_48111_begin_0, end = var_48111_end_0, end_mask = var_48111_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48111_cast_fp16")]; + tensor var_48118_begin_0 = const()[name = string("op_48118_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48118_end_0 = const()[name = string("op_48118_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48118_end_mask_0 = const()[name = string("op_48118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48118_cast_fp16 = slice_by_index(begin = var_48118_begin_0, end = var_48118_end_0, end_mask = var_48118_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48118_cast_fp16")]; + tensor var_48125_begin_0 = const()[name = string("op_48125_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48125_end_0 = const()[name = string("op_48125_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48125_end_mask_0 = const()[name = string("op_48125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48125_cast_fp16 = slice_by_index(begin = var_48125_begin_0, end = var_48125_end_0, end_mask = var_48125_end_mask_0, x = var_47659_cast_fp16)[name = string("op_48125_cast_fp16")]; + tensor var_48132_begin_0 = const()[name = string("op_48132_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48132_end_0 = const()[name = string("op_48132_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48132_end_mask_0 = const()[name = string("op_48132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48132_cast_fp16 = slice_by_index(begin = var_48132_begin_0, end = var_48132_end_0, end_mask = var_48132_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48132_cast_fp16")]; + tensor var_48139_begin_0 = const()[name = string("op_48139_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48139_end_0 = const()[name = string("op_48139_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48139_end_mask_0 = const()[name = string("op_48139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48139_cast_fp16 = slice_by_index(begin = var_48139_begin_0, end = var_48139_end_0, end_mask = var_48139_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48139_cast_fp16")]; + tensor var_48146_begin_0 = const()[name = string("op_48146_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48146_end_0 = const()[name = string("op_48146_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48146_end_mask_0 = const()[name = string("op_48146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48146_cast_fp16 = slice_by_index(begin = var_48146_begin_0, end = var_48146_end_0, end_mask = var_48146_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48146_cast_fp16")]; + tensor var_48153_begin_0 = const()[name = string("op_48153_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48153_end_0 = const()[name = string("op_48153_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48153_end_mask_0 = const()[name = string("op_48153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48153_cast_fp16 = slice_by_index(begin = var_48153_begin_0, end = var_48153_end_0, end_mask = var_48153_end_mask_0, x = var_47663_cast_fp16)[name = string("op_48153_cast_fp16")]; + tensor var_48160_begin_0 = const()[name = string("op_48160_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48160_end_0 = const()[name = string("op_48160_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48160_end_mask_0 = const()[name = string("op_48160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48160_cast_fp16 = slice_by_index(begin = var_48160_begin_0, end = var_48160_end_0, end_mask = var_48160_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48160_cast_fp16")]; + tensor var_48167_begin_0 = const()[name = string("op_48167_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48167_end_0 = const()[name = string("op_48167_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48167_end_mask_0 = const()[name = string("op_48167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48167_cast_fp16 = slice_by_index(begin = var_48167_begin_0, end = var_48167_end_0, end_mask = var_48167_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48167_cast_fp16")]; + tensor var_48174_begin_0 = const()[name = string("op_48174_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48174_end_0 = const()[name = string("op_48174_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48174_end_mask_0 = const()[name = string("op_48174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48174_cast_fp16 = slice_by_index(begin = var_48174_begin_0, end = var_48174_end_0, end_mask = var_48174_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48174_cast_fp16")]; + tensor var_48181_begin_0 = const()[name = string("op_48181_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48181_end_0 = const()[name = string("op_48181_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48181_end_mask_0 = const()[name = string("op_48181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48181_cast_fp16 = slice_by_index(begin = var_48181_begin_0, end = var_48181_end_0, end_mask = var_48181_end_mask_0, x = var_47667_cast_fp16)[name = string("op_48181_cast_fp16")]; + tensor var_48188_begin_0 = const()[name = string("op_48188_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48188_end_0 = const()[name = string("op_48188_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48188_end_mask_0 = const()[name = string("op_48188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48188_cast_fp16 = slice_by_index(begin = var_48188_begin_0, end = var_48188_end_0, end_mask = var_48188_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48188_cast_fp16")]; + tensor var_48195_begin_0 = const()[name = string("op_48195_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48195_end_0 = const()[name = string("op_48195_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48195_end_mask_0 = const()[name = string("op_48195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48195_cast_fp16 = slice_by_index(begin = var_48195_begin_0, end = var_48195_end_0, end_mask = var_48195_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48195_cast_fp16")]; + tensor var_48202_begin_0 = const()[name = string("op_48202_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48202_end_0 = const()[name = string("op_48202_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48202_end_mask_0 = const()[name = string("op_48202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48202_cast_fp16 = slice_by_index(begin = var_48202_begin_0, end = var_48202_end_0, end_mask = var_48202_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48202_cast_fp16")]; + tensor var_48209_begin_0 = const()[name = string("op_48209_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48209_end_0 = const()[name = string("op_48209_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48209_end_mask_0 = const()[name = string("op_48209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48209_cast_fp16 = slice_by_index(begin = var_48209_begin_0, end = var_48209_end_0, end_mask = var_48209_end_mask_0, x = var_47671_cast_fp16)[name = string("op_48209_cast_fp16")]; + tensor var_48216_begin_0 = const()[name = string("op_48216_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48216_end_0 = const()[name = string("op_48216_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48216_end_mask_0 = const()[name = string("op_48216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48216_cast_fp16 = slice_by_index(begin = var_48216_begin_0, end = var_48216_end_0, end_mask = var_48216_end_mask_0, x = var_47675_cast_fp16)[name = string("op_48216_cast_fp16")]; + tensor var_48223_begin_0 = const()[name = string("op_48223_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48223_end_0 = const()[name = string("op_48223_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48223_end_mask_0 = const()[name = string("op_48223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48223_cast_fp16 = slice_by_index(begin = var_48223_begin_0, end = var_48223_end_0, end_mask = var_48223_end_mask_0, x = var_47675_cast_fp16)[name = string("op_48223_cast_fp16")]; + tensor var_48230_begin_0 = const()[name = string("op_48230_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48230_end_0 = const()[name = string("op_48230_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48230_end_mask_0 = const()[name = string("op_48230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48230_cast_fp16 = slice_by_index(begin = var_48230_begin_0, end = var_48230_end_0, end_mask = var_48230_end_mask_0, x = var_47675_cast_fp16)[name = string("op_48230_cast_fp16")]; + tensor var_48237_begin_0 = const()[name = string("op_48237_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48237_end_0 = const()[name = string("op_48237_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48237_end_mask_0 = const()[name = string("op_48237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48237_cast_fp16 = slice_by_index(begin = var_48237_begin_0, end = var_48237_end_0, end_mask = var_48237_end_mask_0, x = var_47675_cast_fp16)[name = string("op_48237_cast_fp16")]; + tensor k_63_perm_0 = const()[name = string("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_48242_begin_0 = const()[name = string("op_48242_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48242_end_0 = const()[name = string("op_48242_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_48242_end_mask_0 = const()[name = string("op_48242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_48242_cast_fp16 = slice_by_index(begin = var_48242_begin_0, end = var_48242_end_0, end_mask = var_48242_end_mask_0, x = k_63_cast_fp16)[name = string("op_48242_cast_fp16")]; + tensor var_48246_begin_0 = const()[name = string("op_48246_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_48246_end_0 = const()[name = string("op_48246_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_48246_end_mask_0 = const()[name = string("op_48246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48246_cast_fp16 = slice_by_index(begin = var_48246_begin_0, end = var_48246_end_0, end_mask = var_48246_end_mask_0, x = k_63_cast_fp16)[name = string("op_48246_cast_fp16")]; + tensor var_48250_begin_0 = const()[name = string("op_48250_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_48250_end_0 = const()[name = string("op_48250_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_48250_end_mask_0 = const()[name = string("op_48250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48250_cast_fp16 = slice_by_index(begin = var_48250_begin_0, end = var_48250_end_0, end_mask = var_48250_end_mask_0, x = k_63_cast_fp16)[name = string("op_48250_cast_fp16")]; + tensor var_48254_begin_0 = const()[name = string("op_48254_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_48254_end_0 = const()[name = string("op_48254_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_48254_end_mask_0 = const()[name = string("op_48254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48254_cast_fp16 = slice_by_index(begin = var_48254_begin_0, end = var_48254_end_0, end_mask = var_48254_end_mask_0, x = k_63_cast_fp16)[name = string("op_48254_cast_fp16")]; + tensor var_48258_begin_0 = const()[name = string("op_48258_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_48258_end_0 = const()[name = string("op_48258_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_48258_end_mask_0 = const()[name = string("op_48258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48258_cast_fp16 = slice_by_index(begin = var_48258_begin_0, end = var_48258_end_0, end_mask = var_48258_end_mask_0, x = k_63_cast_fp16)[name = string("op_48258_cast_fp16")]; + tensor var_48262_begin_0 = const()[name = string("op_48262_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_48262_end_0 = const()[name = string("op_48262_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_48262_end_mask_0 = const()[name = string("op_48262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48262_cast_fp16 = slice_by_index(begin = var_48262_begin_0, end = var_48262_end_0, end_mask = var_48262_end_mask_0, x = k_63_cast_fp16)[name = string("op_48262_cast_fp16")]; + tensor var_48266_begin_0 = const()[name = string("op_48266_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_48266_end_0 = const()[name = string("op_48266_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_48266_end_mask_0 = const()[name = string("op_48266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48266_cast_fp16 = slice_by_index(begin = var_48266_begin_0, end = var_48266_end_0, end_mask = var_48266_end_mask_0, x = k_63_cast_fp16)[name = string("op_48266_cast_fp16")]; + tensor var_48270_begin_0 = const()[name = string("op_48270_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_48270_end_0 = const()[name = string("op_48270_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_48270_end_mask_0 = const()[name = string("op_48270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48270_cast_fp16 = slice_by_index(begin = var_48270_begin_0, end = var_48270_end_0, end_mask = var_48270_end_mask_0, x = k_63_cast_fp16)[name = string("op_48270_cast_fp16")]; + tensor var_48274_begin_0 = const()[name = string("op_48274_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_48274_end_0 = const()[name = string("op_48274_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_48274_end_mask_0 = const()[name = string("op_48274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48274_cast_fp16 = slice_by_index(begin = var_48274_begin_0, end = var_48274_end_0, end_mask = var_48274_end_mask_0, x = k_63_cast_fp16)[name = string("op_48274_cast_fp16")]; + tensor var_48278_begin_0 = const()[name = string("op_48278_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_48278_end_0 = const()[name = string("op_48278_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_48278_end_mask_0 = const()[name = string("op_48278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48278_cast_fp16 = slice_by_index(begin = var_48278_begin_0, end = var_48278_end_0, end_mask = var_48278_end_mask_0, x = k_63_cast_fp16)[name = string("op_48278_cast_fp16")]; + tensor var_48282_begin_0 = const()[name = string("op_48282_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_48282_end_0 = const()[name = string("op_48282_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_48282_end_mask_0 = const()[name = string("op_48282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48282_cast_fp16 = slice_by_index(begin = var_48282_begin_0, end = var_48282_end_0, end_mask = var_48282_end_mask_0, x = k_63_cast_fp16)[name = string("op_48282_cast_fp16")]; + tensor var_48286_begin_0 = const()[name = string("op_48286_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_48286_end_0 = const()[name = string("op_48286_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_48286_end_mask_0 = const()[name = string("op_48286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48286_cast_fp16 = slice_by_index(begin = var_48286_begin_0, end = var_48286_end_0, end_mask = var_48286_end_mask_0, x = k_63_cast_fp16)[name = string("op_48286_cast_fp16")]; + tensor var_48290_begin_0 = const()[name = string("op_48290_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_48290_end_0 = const()[name = string("op_48290_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_48290_end_mask_0 = const()[name = string("op_48290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48290_cast_fp16 = slice_by_index(begin = var_48290_begin_0, end = var_48290_end_0, end_mask = var_48290_end_mask_0, x = k_63_cast_fp16)[name = string("op_48290_cast_fp16")]; + tensor var_48294_begin_0 = const()[name = string("op_48294_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_48294_end_0 = const()[name = string("op_48294_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_48294_end_mask_0 = const()[name = string("op_48294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48294_cast_fp16 = slice_by_index(begin = var_48294_begin_0, end = var_48294_end_0, end_mask = var_48294_end_mask_0, x = k_63_cast_fp16)[name = string("op_48294_cast_fp16")]; + tensor var_48298_begin_0 = const()[name = string("op_48298_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_48298_end_0 = const()[name = string("op_48298_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_48298_end_mask_0 = const()[name = string("op_48298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48298_cast_fp16 = slice_by_index(begin = var_48298_begin_0, end = var_48298_end_0, end_mask = var_48298_end_mask_0, x = k_63_cast_fp16)[name = string("op_48298_cast_fp16")]; + tensor var_48302_begin_0 = const()[name = string("op_48302_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_48302_end_0 = const()[name = string("op_48302_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_48302_end_mask_0 = const()[name = string("op_48302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48302_cast_fp16 = slice_by_index(begin = var_48302_begin_0, end = var_48302_end_0, end_mask = var_48302_end_mask_0, x = k_63_cast_fp16)[name = string("op_48302_cast_fp16")]; + tensor var_48306_begin_0 = const()[name = string("op_48306_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_48306_end_0 = const()[name = string("op_48306_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_48306_end_mask_0 = const()[name = string("op_48306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48306_cast_fp16 = slice_by_index(begin = var_48306_begin_0, end = var_48306_end_0, end_mask = var_48306_end_mask_0, x = k_63_cast_fp16)[name = string("op_48306_cast_fp16")]; + tensor var_48310_begin_0 = const()[name = string("op_48310_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_48310_end_0 = const()[name = string("op_48310_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_48310_end_mask_0 = const()[name = string("op_48310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48310_cast_fp16 = slice_by_index(begin = var_48310_begin_0, end = var_48310_end_0, end_mask = var_48310_end_mask_0, x = k_63_cast_fp16)[name = string("op_48310_cast_fp16")]; + tensor var_48314_begin_0 = const()[name = string("op_48314_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_48314_end_0 = const()[name = string("op_48314_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_48314_end_mask_0 = const()[name = string("op_48314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48314_cast_fp16 = slice_by_index(begin = var_48314_begin_0, end = var_48314_end_0, end_mask = var_48314_end_mask_0, x = k_63_cast_fp16)[name = string("op_48314_cast_fp16")]; + tensor var_48318_begin_0 = const()[name = string("op_48318_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_48318_end_0 = const()[name = string("op_48318_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_48318_end_mask_0 = const()[name = string("op_48318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48318_cast_fp16 = slice_by_index(begin = var_48318_begin_0, end = var_48318_end_0, end_mask = var_48318_end_mask_0, x = k_63_cast_fp16)[name = string("op_48318_cast_fp16")]; + tensor var_48320_begin_0 = const()[name = string("op_48320_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48320_end_0 = const()[name = string("op_48320_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48320_end_mask_0 = const()[name = string("op_48320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48320_cast_fp16 = slice_by_index(begin = var_48320_begin_0, end = var_48320_end_0, end_mask = var_48320_end_mask_0, x = value_cast_fp16)[name = string("op_48320_cast_fp16")]; + tensor var_48324_begin_0 = const()[name = string("op_48324_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48324_end_0 = const()[name = string("op_48324_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_48324_end_mask_0 = const()[name = string("op_48324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48324_cast_fp16 = slice_by_index(begin = var_48324_begin_0, end = var_48324_end_0, end_mask = var_48324_end_mask_0, x = value_cast_fp16)[name = string("op_48324_cast_fp16")]; + tensor var_48328_begin_0 = const()[name = string("op_48328_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48328_end_0 = const()[name = string("op_48328_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_48328_end_mask_0 = const()[name = string("op_48328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48328_cast_fp16 = slice_by_index(begin = var_48328_begin_0, end = var_48328_end_0, end_mask = var_48328_end_mask_0, x = value_cast_fp16)[name = string("op_48328_cast_fp16")]; + tensor var_48332_begin_0 = const()[name = string("op_48332_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48332_end_0 = const()[name = string("op_48332_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_48332_end_mask_0 = const()[name = string("op_48332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48332_cast_fp16 = slice_by_index(begin = var_48332_begin_0, end = var_48332_end_0, end_mask = var_48332_end_mask_0, x = value_cast_fp16)[name = string("op_48332_cast_fp16")]; + tensor var_48336_begin_0 = const()[name = string("op_48336_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48336_end_0 = const()[name = string("op_48336_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_48336_end_mask_0 = const()[name = string("op_48336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48336_cast_fp16 = slice_by_index(begin = var_48336_begin_0, end = var_48336_end_0, end_mask = var_48336_end_mask_0, x = value_cast_fp16)[name = string("op_48336_cast_fp16")]; + tensor var_48340_begin_0 = const()[name = string("op_48340_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48340_end_0 = const()[name = string("op_48340_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_48340_end_mask_0 = const()[name = string("op_48340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48340_cast_fp16 = slice_by_index(begin = var_48340_begin_0, end = var_48340_end_0, end_mask = var_48340_end_mask_0, x = value_cast_fp16)[name = string("op_48340_cast_fp16")]; + tensor var_48344_begin_0 = const()[name = string("op_48344_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48344_end_0 = const()[name = string("op_48344_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_48344_end_mask_0 = const()[name = string("op_48344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48344_cast_fp16 = slice_by_index(begin = var_48344_begin_0, end = var_48344_end_0, end_mask = var_48344_end_mask_0, x = value_cast_fp16)[name = string("op_48344_cast_fp16")]; + tensor var_48348_begin_0 = const()[name = string("op_48348_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48348_end_0 = const()[name = string("op_48348_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_48348_end_mask_0 = const()[name = string("op_48348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48348_cast_fp16 = slice_by_index(begin = var_48348_begin_0, end = var_48348_end_0, end_mask = var_48348_end_mask_0, x = value_cast_fp16)[name = string("op_48348_cast_fp16")]; + tensor var_48352_begin_0 = const()[name = string("op_48352_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48352_end_0 = const()[name = string("op_48352_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_48352_end_mask_0 = const()[name = string("op_48352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48352_cast_fp16 = slice_by_index(begin = var_48352_begin_0, end = var_48352_end_0, end_mask = var_48352_end_mask_0, x = value_cast_fp16)[name = string("op_48352_cast_fp16")]; + tensor var_48356_begin_0 = const()[name = string("op_48356_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48356_end_0 = const()[name = string("op_48356_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_48356_end_mask_0 = const()[name = string("op_48356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48356_cast_fp16 = slice_by_index(begin = var_48356_begin_0, end = var_48356_end_0, end_mask = var_48356_end_mask_0, x = value_cast_fp16)[name = string("op_48356_cast_fp16")]; + tensor var_48360_begin_0 = const()[name = string("op_48360_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48360_end_0 = const()[name = string("op_48360_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_48360_end_mask_0 = const()[name = string("op_48360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48360_cast_fp16 = slice_by_index(begin = var_48360_begin_0, end = var_48360_end_0, end_mask = var_48360_end_mask_0, x = value_cast_fp16)[name = string("op_48360_cast_fp16")]; + tensor var_48364_begin_0 = const()[name = string("op_48364_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48364_end_0 = const()[name = string("op_48364_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_48364_end_mask_0 = const()[name = string("op_48364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48364_cast_fp16 = slice_by_index(begin = var_48364_begin_0, end = var_48364_end_0, end_mask = var_48364_end_mask_0, x = value_cast_fp16)[name = string("op_48364_cast_fp16")]; + tensor var_48368_begin_0 = const()[name = string("op_48368_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48368_end_0 = const()[name = string("op_48368_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_48368_end_mask_0 = const()[name = string("op_48368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48368_cast_fp16 = slice_by_index(begin = var_48368_begin_0, end = var_48368_end_0, end_mask = var_48368_end_mask_0, x = value_cast_fp16)[name = string("op_48368_cast_fp16")]; + tensor var_48372_begin_0 = const()[name = string("op_48372_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48372_end_0 = const()[name = string("op_48372_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_48372_end_mask_0 = const()[name = string("op_48372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48372_cast_fp16 = slice_by_index(begin = var_48372_begin_0, end = var_48372_end_0, end_mask = var_48372_end_mask_0, x = value_cast_fp16)[name = string("op_48372_cast_fp16")]; + tensor var_48376_begin_0 = const()[name = string("op_48376_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48376_end_0 = const()[name = string("op_48376_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_48376_end_mask_0 = const()[name = string("op_48376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48376_cast_fp16 = slice_by_index(begin = var_48376_begin_0, end = var_48376_end_0, end_mask = var_48376_end_mask_0, x = value_cast_fp16)[name = string("op_48376_cast_fp16")]; + tensor var_48380_begin_0 = const()[name = string("op_48380_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48380_end_0 = const()[name = string("op_48380_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_48380_end_mask_0 = const()[name = string("op_48380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48380_cast_fp16 = slice_by_index(begin = var_48380_begin_0, end = var_48380_end_0, end_mask = var_48380_end_mask_0, x = value_cast_fp16)[name = string("op_48380_cast_fp16")]; + tensor var_48384_begin_0 = const()[name = string("op_48384_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48384_end_0 = const()[name = string("op_48384_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_48384_end_mask_0 = const()[name = string("op_48384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48384_cast_fp16 = slice_by_index(begin = var_48384_begin_0, end = var_48384_end_0, end_mask = var_48384_end_mask_0, x = value_cast_fp16)[name = string("op_48384_cast_fp16")]; + tensor var_48388_begin_0 = const()[name = string("op_48388_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48388_end_0 = const()[name = string("op_48388_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_48388_end_mask_0 = const()[name = string("op_48388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48388_cast_fp16 = slice_by_index(begin = var_48388_begin_0, end = var_48388_end_0, end_mask = var_48388_end_mask_0, x = value_cast_fp16)[name = string("op_48388_cast_fp16")]; + tensor var_48392_begin_0 = const()[name = string("op_48392_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48392_end_0 = const()[name = string("op_48392_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_48392_end_mask_0 = const()[name = string("op_48392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48392_cast_fp16 = slice_by_index(begin = var_48392_begin_0, end = var_48392_end_0, end_mask = var_48392_end_mask_0, x = value_cast_fp16)[name = string("op_48392_cast_fp16")]; + tensor var_48396_begin_0 = const()[name = string("op_48396_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_48396_end_0 = const()[name = string("op_48396_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_48396_end_mask_0 = const()[name = string("op_48396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48396_cast_fp16 = slice_by_index(begin = var_48396_begin_0, end = var_48396_end_0, end_mask = var_48396_end_mask_0, x = value_cast_fp16)[name = string("op_48396_cast_fp16")]; + string _SplitHeadsQ__mh_w_4961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4961_equation_0, values = (var_48242_cast_fp16, var_47684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4961_cast_fp16")]; + string _SplitHeadsQ__mh_w_4963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4963_equation_0, values = (var_48242_cast_fp16, var_47691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4963_cast_fp16")]; + string _SplitHeadsQ__mh_w_4965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4965_equation_0, values = (var_48242_cast_fp16, var_47698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4965_cast_fp16")]; + string _SplitHeadsQ__mh_w_4967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4967_equation_0, values = (var_48242_cast_fp16, var_47705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4967_cast_fp16")]; + string _SplitHeadsQ__mh_w_4969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4969_equation_0, values = (var_48246_cast_fp16, var_47712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4969_cast_fp16")]; + string _SplitHeadsQ__mh_w_4971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4971_equation_0, values = (var_48246_cast_fp16, var_47719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4971_cast_fp16")]; + string _SplitHeadsQ__mh_w_4973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4973_equation_0, values = (var_48246_cast_fp16, var_47726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4973_cast_fp16")]; + string _SplitHeadsQ__mh_w_4975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4975_equation_0, values = (var_48246_cast_fp16, var_47733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4975_cast_fp16")]; + string _SplitHeadsQ__mh_w_4977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4977_equation_0, values = (var_48250_cast_fp16, var_47740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4977_cast_fp16")]; + string _SplitHeadsQ__mh_w_4979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4979_equation_0, values = (var_48250_cast_fp16, var_47747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4979_cast_fp16")]; + string _SplitHeadsQ__mh_w_4981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4981_equation_0, values = (var_48250_cast_fp16, var_47754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4981_cast_fp16")]; + string _SplitHeadsQ__mh_w_4983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4983_equation_0, values = (var_48250_cast_fp16, var_47761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4983_cast_fp16")]; + string _SplitHeadsQ__mh_w_4985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4985_equation_0, values = (var_48254_cast_fp16, var_47768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4985_cast_fp16")]; + string _SplitHeadsQ__mh_w_4987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4987_equation_0, values = (var_48254_cast_fp16, var_47775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4987_cast_fp16")]; + string _SplitHeadsQ__mh_w_4989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4989_equation_0, values = (var_48254_cast_fp16, var_47782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4989_cast_fp16")]; + string _SplitHeadsQ__mh_w_4991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4991_equation_0, values = (var_48254_cast_fp16, var_47789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4991_cast_fp16")]; + string _SplitHeadsQ__mh_w_4993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4993_equation_0, values = (var_48258_cast_fp16, var_47796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4993_cast_fp16")]; + string _SplitHeadsQ__mh_w_4995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4995_equation_0, values = (var_48258_cast_fp16, var_47803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4995_cast_fp16")]; + string _SplitHeadsQ__mh_w_4997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4997_equation_0, values = (var_48258_cast_fp16, var_47810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4997_cast_fp16")]; + string _SplitHeadsQ__mh_w_4999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_4999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_4999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_4999_equation_0, values = (var_48258_cast_fp16, var_47817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_4999_cast_fp16")]; + string _SplitHeadsQ__mh_w_5001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5001_equation_0, values = (var_48262_cast_fp16, var_47824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5001_cast_fp16")]; + string _SplitHeadsQ__mh_w_5003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5003_equation_0, values = (var_48262_cast_fp16, var_47831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5003_cast_fp16")]; + string _SplitHeadsQ__mh_w_5005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5005_equation_0, values = (var_48262_cast_fp16, var_47838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5005_cast_fp16")]; + string _SplitHeadsQ__mh_w_5007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5007_equation_0, values = (var_48262_cast_fp16, var_47845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5007_cast_fp16")]; + string _SplitHeadsQ__mh_w_5009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5009_equation_0, values = (var_48266_cast_fp16, var_47852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5009_cast_fp16")]; + string _SplitHeadsQ__mh_w_5011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5011_equation_0, values = (var_48266_cast_fp16, var_47859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5011_cast_fp16")]; + string _SplitHeadsQ__mh_w_5013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5013_equation_0, values = (var_48266_cast_fp16, var_47866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5013_cast_fp16")]; + string _SplitHeadsQ__mh_w_5015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5015_equation_0, values = (var_48266_cast_fp16, var_47873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5015_cast_fp16")]; + string _SplitHeadsQ__mh_w_5017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5017_equation_0, values = (var_48270_cast_fp16, var_47880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5017_cast_fp16")]; + string _SplitHeadsQ__mh_w_5019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5019_equation_0, values = (var_48270_cast_fp16, var_47887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5019_cast_fp16")]; + string _SplitHeadsQ__mh_w_5021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5021_equation_0, values = (var_48270_cast_fp16, var_47894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5021_cast_fp16")]; + string _SplitHeadsQ__mh_w_5023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5023_equation_0, values = (var_48270_cast_fp16, var_47901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5023_cast_fp16")]; + string _SplitHeadsQ__mh_w_5025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5025_equation_0, values = (var_48274_cast_fp16, var_47908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5025_cast_fp16")]; + string _SplitHeadsQ__mh_w_5027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5027_equation_0, values = (var_48274_cast_fp16, var_47915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5027_cast_fp16")]; + string _SplitHeadsQ__mh_w_5029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5029_equation_0, values = (var_48274_cast_fp16, var_47922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5029_cast_fp16")]; + string _SplitHeadsQ__mh_w_5031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5031_equation_0, values = (var_48274_cast_fp16, var_47929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5031_cast_fp16")]; + string _SplitHeadsQ__mh_w_5033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5033_equation_0, values = (var_48278_cast_fp16, var_47936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5033_cast_fp16")]; + string _SplitHeadsQ__mh_w_5035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5035_equation_0, values = (var_48278_cast_fp16, var_47943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5035_cast_fp16")]; + string _SplitHeadsQ__mh_w_5037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5037_equation_0, values = (var_48278_cast_fp16, var_47950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5037_cast_fp16")]; + string _SplitHeadsQ__mh_w_5039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5039_equation_0, values = (var_48278_cast_fp16, var_47957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5039_cast_fp16")]; + string _SplitHeadsQ__mh_w_5041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5041_equation_0, values = (var_48282_cast_fp16, var_47964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5041_cast_fp16")]; + string _SplitHeadsQ__mh_w_5043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5043_equation_0, values = (var_48282_cast_fp16, var_47971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5043_cast_fp16")]; + string _SplitHeadsQ__mh_w_5045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5045_equation_0, values = (var_48282_cast_fp16, var_47978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5045_cast_fp16")]; + string _SplitHeadsQ__mh_w_5047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5047_equation_0, values = (var_48282_cast_fp16, var_47985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5047_cast_fp16")]; + string _SplitHeadsQ__mh_w_5049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5049_equation_0, values = (var_48286_cast_fp16, var_47992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5049_cast_fp16")]; + string _SplitHeadsQ__mh_w_5051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5051_equation_0, values = (var_48286_cast_fp16, var_47999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5051_cast_fp16")]; + string _SplitHeadsQ__mh_w_5053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5053_equation_0, values = (var_48286_cast_fp16, var_48006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5053_cast_fp16")]; + string _SplitHeadsQ__mh_w_5055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5055_equation_0, values = (var_48286_cast_fp16, var_48013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5055_cast_fp16")]; + string _SplitHeadsQ__mh_w_5057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5057_equation_0, values = (var_48290_cast_fp16, var_48020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5057_cast_fp16")]; + string _SplitHeadsQ__mh_w_5059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5059_equation_0, values = (var_48290_cast_fp16, var_48027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5059_cast_fp16")]; + string _SplitHeadsQ__mh_w_5061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5061_equation_0, values = (var_48290_cast_fp16, var_48034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5061_cast_fp16")]; + string _SplitHeadsQ__mh_w_5063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5063_equation_0, values = (var_48290_cast_fp16, var_48041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5063_cast_fp16")]; + string _SplitHeadsQ__mh_w_5065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5065_equation_0, values = (var_48294_cast_fp16, var_48048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5065_cast_fp16")]; + string _SplitHeadsQ__mh_w_5067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5067_equation_0, values = (var_48294_cast_fp16, var_48055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5067_cast_fp16")]; + string _SplitHeadsQ__mh_w_5069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5069_equation_0, values = (var_48294_cast_fp16, var_48062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5069_cast_fp16")]; + string _SplitHeadsQ__mh_w_5071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5071_equation_0, values = (var_48294_cast_fp16, var_48069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5071_cast_fp16")]; + string _SplitHeadsQ__mh_w_5073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5073_equation_0, values = (var_48298_cast_fp16, var_48076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5073_cast_fp16")]; + string _SplitHeadsQ__mh_w_5075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5075_equation_0, values = (var_48298_cast_fp16, var_48083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5075_cast_fp16")]; + string _SplitHeadsQ__mh_w_5077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5077_equation_0, values = (var_48298_cast_fp16, var_48090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5077_cast_fp16")]; + string _SplitHeadsQ__mh_w_5079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5079_equation_0, values = (var_48298_cast_fp16, var_48097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5079_cast_fp16")]; + string _SplitHeadsQ__mh_w_5081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5081_equation_0, values = (var_48302_cast_fp16, var_48104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5081_cast_fp16")]; + string _SplitHeadsQ__mh_w_5083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5083_equation_0, values = (var_48302_cast_fp16, var_48111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5083_cast_fp16")]; + string _SplitHeadsQ__mh_w_5085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5085_equation_0, values = (var_48302_cast_fp16, var_48118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5085_cast_fp16")]; + string _SplitHeadsQ__mh_w_5087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5087_equation_0, values = (var_48302_cast_fp16, var_48125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5087_cast_fp16")]; + string _SplitHeadsQ__mh_w_5089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5089_equation_0, values = (var_48306_cast_fp16, var_48132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5089_cast_fp16")]; + string _SplitHeadsQ__mh_w_5091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5091_equation_0, values = (var_48306_cast_fp16, var_48139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5091_cast_fp16")]; + string _SplitHeadsQ__mh_w_5093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5093_equation_0, values = (var_48306_cast_fp16, var_48146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5093_cast_fp16")]; + string _SplitHeadsQ__mh_w_5095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5095_equation_0, values = (var_48306_cast_fp16, var_48153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5095_cast_fp16")]; + string _SplitHeadsQ__mh_w_5097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5097_equation_0, values = (var_48310_cast_fp16, var_48160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5097_cast_fp16")]; + string _SplitHeadsQ__mh_w_5099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5099_equation_0, values = (var_48310_cast_fp16, var_48167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5099_cast_fp16")]; + string _SplitHeadsQ__mh_w_5101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5101_equation_0, values = (var_48310_cast_fp16, var_48174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5101_cast_fp16")]; + string _SplitHeadsQ__mh_w_5103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5103_equation_0, values = (var_48310_cast_fp16, var_48181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5103_cast_fp16")]; + string _SplitHeadsQ__mh_w_5105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5105_equation_0, values = (var_48314_cast_fp16, var_48188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5105_cast_fp16")]; + string _SplitHeadsQ__mh_w_5107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5107_equation_0, values = (var_48314_cast_fp16, var_48195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5107_cast_fp16")]; + string _SplitHeadsQ__mh_w_5109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5109_equation_0, values = (var_48314_cast_fp16, var_48202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5109_cast_fp16")]; + string _SplitHeadsQ__mh_w_5111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5111_equation_0, values = (var_48314_cast_fp16, var_48209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5111_cast_fp16")]; + string _SplitHeadsQ__mh_w_5113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5113_equation_0, values = (var_48318_cast_fp16, var_48216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5113_cast_fp16")]; + string _SplitHeadsQ__mh_w_5115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5115_equation_0, values = (var_48318_cast_fp16, var_48223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5115_cast_fp16")]; + string _SplitHeadsQ__mh_w_5117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5117_equation_0, values = (var_48318_cast_fp16, var_48230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5117_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_48318_cast_fp16, var_48237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_48559_to_fp16 = const()[name = string("op_48559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4961_cast_fp16, y = var_48559_to_fp16)[name = string("aw_chunk_4961_cast_fp16")]; + fp16 var_48561_to_fp16 = const()[name = string("op_48561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4963_cast_fp16, y = var_48561_to_fp16)[name = string("aw_chunk_4963_cast_fp16")]; + fp16 var_48563_to_fp16 = const()[name = string("op_48563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4965_cast_fp16, y = var_48563_to_fp16)[name = string("aw_chunk_4965_cast_fp16")]; + fp16 var_48565_to_fp16 = const()[name = string("op_48565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4967_cast_fp16, y = var_48565_to_fp16)[name = string("aw_chunk_4967_cast_fp16")]; + fp16 var_48567_to_fp16 = const()[name = string("op_48567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4969_cast_fp16, y = var_48567_to_fp16)[name = string("aw_chunk_4969_cast_fp16")]; + fp16 var_48569_to_fp16 = const()[name = string("op_48569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4971_cast_fp16, y = var_48569_to_fp16)[name = string("aw_chunk_4971_cast_fp16")]; + fp16 var_48571_to_fp16 = const()[name = string("op_48571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4973_cast_fp16, y = var_48571_to_fp16)[name = string("aw_chunk_4973_cast_fp16")]; + fp16 var_48573_to_fp16 = const()[name = string("op_48573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4975_cast_fp16, y = var_48573_to_fp16)[name = string("aw_chunk_4975_cast_fp16")]; + fp16 var_48575_to_fp16 = const()[name = string("op_48575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4977_cast_fp16, y = var_48575_to_fp16)[name = string("aw_chunk_4977_cast_fp16")]; + fp16 var_48577_to_fp16 = const()[name = string("op_48577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4979_cast_fp16, y = var_48577_to_fp16)[name = string("aw_chunk_4979_cast_fp16")]; + fp16 var_48579_to_fp16 = const()[name = string("op_48579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4981_cast_fp16, y = var_48579_to_fp16)[name = string("aw_chunk_4981_cast_fp16")]; + fp16 var_48581_to_fp16 = const()[name = string("op_48581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4983_cast_fp16, y = var_48581_to_fp16)[name = string("aw_chunk_4983_cast_fp16")]; + fp16 var_48583_to_fp16 = const()[name = string("op_48583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4985_cast_fp16, y = var_48583_to_fp16)[name = string("aw_chunk_4985_cast_fp16")]; + fp16 var_48585_to_fp16 = const()[name = string("op_48585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4987_cast_fp16, y = var_48585_to_fp16)[name = string("aw_chunk_4987_cast_fp16")]; + fp16 var_48587_to_fp16 = const()[name = string("op_48587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4989_cast_fp16, y = var_48587_to_fp16)[name = string("aw_chunk_4989_cast_fp16")]; + fp16 var_48589_to_fp16 = const()[name = string("op_48589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4991_cast_fp16, y = var_48589_to_fp16)[name = string("aw_chunk_4991_cast_fp16")]; + fp16 var_48591_to_fp16 = const()[name = string("op_48591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4993_cast_fp16, y = var_48591_to_fp16)[name = string("aw_chunk_4993_cast_fp16")]; + fp16 var_48593_to_fp16 = const()[name = string("op_48593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4995_cast_fp16, y = var_48593_to_fp16)[name = string("aw_chunk_4995_cast_fp16")]; + fp16 var_48595_to_fp16 = const()[name = string("op_48595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4997_cast_fp16, y = var_48595_to_fp16)[name = string("aw_chunk_4997_cast_fp16")]; + fp16 var_48597_to_fp16 = const()[name = string("op_48597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_4999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_4999_cast_fp16, y = var_48597_to_fp16)[name = string("aw_chunk_4999_cast_fp16")]; + fp16 var_48599_to_fp16 = const()[name = string("op_48599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5001_cast_fp16, y = var_48599_to_fp16)[name = string("aw_chunk_5001_cast_fp16")]; + fp16 var_48601_to_fp16 = const()[name = string("op_48601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5003_cast_fp16, y = var_48601_to_fp16)[name = string("aw_chunk_5003_cast_fp16")]; + fp16 var_48603_to_fp16 = const()[name = string("op_48603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5005_cast_fp16, y = var_48603_to_fp16)[name = string("aw_chunk_5005_cast_fp16")]; + fp16 var_48605_to_fp16 = const()[name = string("op_48605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5007_cast_fp16, y = var_48605_to_fp16)[name = string("aw_chunk_5007_cast_fp16")]; + fp16 var_48607_to_fp16 = const()[name = string("op_48607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5009_cast_fp16, y = var_48607_to_fp16)[name = string("aw_chunk_5009_cast_fp16")]; + fp16 var_48609_to_fp16 = const()[name = string("op_48609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5011_cast_fp16, y = var_48609_to_fp16)[name = string("aw_chunk_5011_cast_fp16")]; + fp16 var_48611_to_fp16 = const()[name = string("op_48611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5013_cast_fp16, y = var_48611_to_fp16)[name = string("aw_chunk_5013_cast_fp16")]; + fp16 var_48613_to_fp16 = const()[name = string("op_48613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5015_cast_fp16, y = var_48613_to_fp16)[name = string("aw_chunk_5015_cast_fp16")]; + fp16 var_48615_to_fp16 = const()[name = string("op_48615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5017_cast_fp16, y = var_48615_to_fp16)[name = string("aw_chunk_5017_cast_fp16")]; + fp16 var_48617_to_fp16 = const()[name = string("op_48617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5019_cast_fp16, y = var_48617_to_fp16)[name = string("aw_chunk_5019_cast_fp16")]; + fp16 var_48619_to_fp16 = const()[name = string("op_48619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5021_cast_fp16, y = var_48619_to_fp16)[name = string("aw_chunk_5021_cast_fp16")]; + fp16 var_48621_to_fp16 = const()[name = string("op_48621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5023_cast_fp16, y = var_48621_to_fp16)[name = string("aw_chunk_5023_cast_fp16")]; + fp16 var_48623_to_fp16 = const()[name = string("op_48623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5025_cast_fp16, y = var_48623_to_fp16)[name = string("aw_chunk_5025_cast_fp16")]; + fp16 var_48625_to_fp16 = const()[name = string("op_48625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5027_cast_fp16, y = var_48625_to_fp16)[name = string("aw_chunk_5027_cast_fp16")]; + fp16 var_48627_to_fp16 = const()[name = string("op_48627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5029_cast_fp16, y = var_48627_to_fp16)[name = string("aw_chunk_5029_cast_fp16")]; + fp16 var_48629_to_fp16 = const()[name = string("op_48629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5031_cast_fp16, y = var_48629_to_fp16)[name = string("aw_chunk_5031_cast_fp16")]; + fp16 var_48631_to_fp16 = const()[name = string("op_48631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5033_cast_fp16, y = var_48631_to_fp16)[name = string("aw_chunk_5033_cast_fp16")]; + fp16 var_48633_to_fp16 = const()[name = string("op_48633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5035_cast_fp16, y = var_48633_to_fp16)[name = string("aw_chunk_5035_cast_fp16")]; + fp16 var_48635_to_fp16 = const()[name = string("op_48635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5037_cast_fp16, y = var_48635_to_fp16)[name = string("aw_chunk_5037_cast_fp16")]; + fp16 var_48637_to_fp16 = const()[name = string("op_48637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5039_cast_fp16, y = var_48637_to_fp16)[name = string("aw_chunk_5039_cast_fp16")]; + fp16 var_48639_to_fp16 = const()[name = string("op_48639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5041_cast_fp16, y = var_48639_to_fp16)[name = string("aw_chunk_5041_cast_fp16")]; + fp16 var_48641_to_fp16 = const()[name = string("op_48641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5043_cast_fp16, y = var_48641_to_fp16)[name = string("aw_chunk_5043_cast_fp16")]; + fp16 var_48643_to_fp16 = const()[name = string("op_48643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5045_cast_fp16, y = var_48643_to_fp16)[name = string("aw_chunk_5045_cast_fp16")]; + fp16 var_48645_to_fp16 = const()[name = string("op_48645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5047_cast_fp16, y = var_48645_to_fp16)[name = string("aw_chunk_5047_cast_fp16")]; + fp16 var_48647_to_fp16 = const()[name = string("op_48647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5049_cast_fp16, y = var_48647_to_fp16)[name = string("aw_chunk_5049_cast_fp16")]; + fp16 var_48649_to_fp16 = const()[name = string("op_48649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5051_cast_fp16, y = var_48649_to_fp16)[name = string("aw_chunk_5051_cast_fp16")]; + fp16 var_48651_to_fp16 = const()[name = string("op_48651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5053_cast_fp16, y = var_48651_to_fp16)[name = string("aw_chunk_5053_cast_fp16")]; + fp16 var_48653_to_fp16 = const()[name = string("op_48653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5055_cast_fp16, y = var_48653_to_fp16)[name = string("aw_chunk_5055_cast_fp16")]; + fp16 var_48655_to_fp16 = const()[name = string("op_48655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5057_cast_fp16, y = var_48655_to_fp16)[name = string("aw_chunk_5057_cast_fp16")]; + fp16 var_48657_to_fp16 = const()[name = string("op_48657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5059_cast_fp16, y = var_48657_to_fp16)[name = string("aw_chunk_5059_cast_fp16")]; + fp16 var_48659_to_fp16 = const()[name = string("op_48659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5061_cast_fp16, y = var_48659_to_fp16)[name = string("aw_chunk_5061_cast_fp16")]; + fp16 var_48661_to_fp16 = const()[name = string("op_48661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5063_cast_fp16, y = var_48661_to_fp16)[name = string("aw_chunk_5063_cast_fp16")]; + fp16 var_48663_to_fp16 = const()[name = string("op_48663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5065_cast_fp16, y = var_48663_to_fp16)[name = string("aw_chunk_5065_cast_fp16")]; + fp16 var_48665_to_fp16 = const()[name = string("op_48665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5067_cast_fp16, y = var_48665_to_fp16)[name = string("aw_chunk_5067_cast_fp16")]; + fp16 var_48667_to_fp16 = const()[name = string("op_48667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5069_cast_fp16, y = var_48667_to_fp16)[name = string("aw_chunk_5069_cast_fp16")]; + fp16 var_48669_to_fp16 = const()[name = string("op_48669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5071_cast_fp16, y = var_48669_to_fp16)[name = string("aw_chunk_5071_cast_fp16")]; + fp16 var_48671_to_fp16 = const()[name = string("op_48671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5073_cast_fp16, y = var_48671_to_fp16)[name = string("aw_chunk_5073_cast_fp16")]; + fp16 var_48673_to_fp16 = const()[name = string("op_48673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5075_cast_fp16, y = var_48673_to_fp16)[name = string("aw_chunk_5075_cast_fp16")]; + fp16 var_48675_to_fp16 = const()[name = string("op_48675_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5077_cast_fp16, y = var_48675_to_fp16)[name = string("aw_chunk_5077_cast_fp16")]; + fp16 var_48677_to_fp16 = const()[name = string("op_48677_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5079_cast_fp16, y = var_48677_to_fp16)[name = string("aw_chunk_5079_cast_fp16")]; + fp16 var_48679_to_fp16 = const()[name = string("op_48679_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5081_cast_fp16, y = var_48679_to_fp16)[name = string("aw_chunk_5081_cast_fp16")]; + fp16 var_48681_to_fp16 = const()[name = string("op_48681_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5083_cast_fp16, y = var_48681_to_fp16)[name = string("aw_chunk_5083_cast_fp16")]; + fp16 var_48683_to_fp16 = const()[name = string("op_48683_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5085_cast_fp16, y = var_48683_to_fp16)[name = string("aw_chunk_5085_cast_fp16")]; + fp16 var_48685_to_fp16 = const()[name = string("op_48685_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5087_cast_fp16, y = var_48685_to_fp16)[name = string("aw_chunk_5087_cast_fp16")]; + fp16 var_48687_to_fp16 = const()[name = string("op_48687_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5089_cast_fp16, y = var_48687_to_fp16)[name = string("aw_chunk_5089_cast_fp16")]; + fp16 var_48689_to_fp16 = const()[name = string("op_48689_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5091_cast_fp16, y = var_48689_to_fp16)[name = string("aw_chunk_5091_cast_fp16")]; + fp16 var_48691_to_fp16 = const()[name = string("op_48691_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5093_cast_fp16, y = var_48691_to_fp16)[name = string("aw_chunk_5093_cast_fp16")]; + fp16 var_48693_to_fp16 = const()[name = string("op_48693_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5095_cast_fp16, y = var_48693_to_fp16)[name = string("aw_chunk_5095_cast_fp16")]; + fp16 var_48695_to_fp16 = const()[name = string("op_48695_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5097_cast_fp16, y = var_48695_to_fp16)[name = string("aw_chunk_5097_cast_fp16")]; + fp16 var_48697_to_fp16 = const()[name = string("op_48697_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5099_cast_fp16, y = var_48697_to_fp16)[name = string("aw_chunk_5099_cast_fp16")]; + fp16 var_48699_to_fp16 = const()[name = string("op_48699_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5101_cast_fp16, y = var_48699_to_fp16)[name = string("aw_chunk_5101_cast_fp16")]; + fp16 var_48701_to_fp16 = const()[name = string("op_48701_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5103_cast_fp16, y = var_48701_to_fp16)[name = string("aw_chunk_5103_cast_fp16")]; + fp16 var_48703_to_fp16 = const()[name = string("op_48703_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5105_cast_fp16, y = var_48703_to_fp16)[name = string("aw_chunk_5105_cast_fp16")]; + fp16 var_48705_to_fp16 = const()[name = string("op_48705_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5107_cast_fp16, y = var_48705_to_fp16)[name = string("aw_chunk_5107_cast_fp16")]; + fp16 var_48707_to_fp16 = const()[name = string("op_48707_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5109_cast_fp16, y = var_48707_to_fp16)[name = string("aw_chunk_5109_cast_fp16")]; + fp16 var_48709_to_fp16 = const()[name = string("op_48709_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5111_cast_fp16, y = var_48709_to_fp16)[name = string("aw_chunk_5111_cast_fp16")]; + fp16 var_48711_to_fp16 = const()[name = string("op_48711_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5113_cast_fp16, y = var_48711_to_fp16)[name = string("aw_chunk_5113_cast_fp16")]; + fp16 var_48713_to_fp16 = const()[name = string("op_48713_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5115_cast_fp16, y = var_48713_to_fp16)[name = string("aw_chunk_5115_cast_fp16")]; + fp16 var_48715_to_fp16 = const()[name = string("op_48715_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5117_cast_fp16, y = var_48715_to_fp16)[name = string("aw_chunk_5117_cast_fp16")]; + fp16 var_48717_to_fp16 = const()[name = string("op_48717_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_48717_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_48719_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4961_cast_fp16)[name = string("op_48719_cast_fp16")]; + tensor var_48720_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4963_cast_fp16)[name = string("op_48720_cast_fp16")]; + tensor var_48721_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4965_cast_fp16)[name = string("op_48721_cast_fp16")]; + tensor var_48722_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4967_cast_fp16)[name = string("op_48722_cast_fp16")]; + tensor var_48723_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4969_cast_fp16)[name = string("op_48723_cast_fp16")]; + tensor var_48724_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4971_cast_fp16)[name = string("op_48724_cast_fp16")]; + tensor var_48725_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4973_cast_fp16)[name = string("op_48725_cast_fp16")]; + tensor var_48726_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4975_cast_fp16)[name = string("op_48726_cast_fp16")]; + tensor var_48727_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4977_cast_fp16)[name = string("op_48727_cast_fp16")]; + tensor var_48728_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4979_cast_fp16)[name = string("op_48728_cast_fp16")]; + tensor var_48729_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4981_cast_fp16)[name = string("op_48729_cast_fp16")]; + tensor var_48730_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4983_cast_fp16)[name = string("op_48730_cast_fp16")]; + tensor var_48731_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4985_cast_fp16)[name = string("op_48731_cast_fp16")]; + tensor var_48732_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4987_cast_fp16)[name = string("op_48732_cast_fp16")]; + tensor var_48733_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4989_cast_fp16)[name = string("op_48733_cast_fp16")]; + tensor var_48734_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4991_cast_fp16)[name = string("op_48734_cast_fp16")]; + tensor var_48735_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4993_cast_fp16)[name = string("op_48735_cast_fp16")]; + tensor var_48736_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4995_cast_fp16)[name = string("op_48736_cast_fp16")]; + tensor var_48737_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4997_cast_fp16)[name = string("op_48737_cast_fp16")]; + tensor var_48738_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_4999_cast_fp16)[name = string("op_48738_cast_fp16")]; + tensor var_48739_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5001_cast_fp16)[name = string("op_48739_cast_fp16")]; + tensor var_48740_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5003_cast_fp16)[name = string("op_48740_cast_fp16")]; + tensor var_48741_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5005_cast_fp16)[name = string("op_48741_cast_fp16")]; + tensor var_48742_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5007_cast_fp16)[name = string("op_48742_cast_fp16")]; + tensor var_48743_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5009_cast_fp16)[name = string("op_48743_cast_fp16")]; + tensor var_48744_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5011_cast_fp16)[name = string("op_48744_cast_fp16")]; + tensor var_48745_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5013_cast_fp16)[name = string("op_48745_cast_fp16")]; + tensor var_48746_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5015_cast_fp16)[name = string("op_48746_cast_fp16")]; + tensor var_48747_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5017_cast_fp16)[name = string("op_48747_cast_fp16")]; + tensor var_48748_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5019_cast_fp16)[name = string("op_48748_cast_fp16")]; + tensor var_48749_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5021_cast_fp16)[name = string("op_48749_cast_fp16")]; + tensor var_48750_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5023_cast_fp16)[name = string("op_48750_cast_fp16")]; + tensor var_48751_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5025_cast_fp16)[name = string("op_48751_cast_fp16")]; + tensor var_48752_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5027_cast_fp16)[name = string("op_48752_cast_fp16")]; + tensor var_48753_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5029_cast_fp16)[name = string("op_48753_cast_fp16")]; + tensor var_48754_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5031_cast_fp16)[name = string("op_48754_cast_fp16")]; + tensor var_48755_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5033_cast_fp16)[name = string("op_48755_cast_fp16")]; + tensor var_48756_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5035_cast_fp16)[name = string("op_48756_cast_fp16")]; + tensor var_48757_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5037_cast_fp16)[name = string("op_48757_cast_fp16")]; + tensor var_48758_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5039_cast_fp16)[name = string("op_48758_cast_fp16")]; + tensor var_48759_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5041_cast_fp16)[name = string("op_48759_cast_fp16")]; + tensor var_48760_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5043_cast_fp16)[name = string("op_48760_cast_fp16")]; + tensor var_48761_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5045_cast_fp16)[name = string("op_48761_cast_fp16")]; + tensor var_48762_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5047_cast_fp16)[name = string("op_48762_cast_fp16")]; + tensor var_48763_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5049_cast_fp16)[name = string("op_48763_cast_fp16")]; + tensor var_48764_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5051_cast_fp16)[name = string("op_48764_cast_fp16")]; + tensor var_48765_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5053_cast_fp16)[name = string("op_48765_cast_fp16")]; + tensor var_48766_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5055_cast_fp16)[name = string("op_48766_cast_fp16")]; + tensor var_48767_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5057_cast_fp16)[name = string("op_48767_cast_fp16")]; + tensor var_48768_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5059_cast_fp16)[name = string("op_48768_cast_fp16")]; + tensor var_48769_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5061_cast_fp16)[name = string("op_48769_cast_fp16")]; + tensor var_48770_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5063_cast_fp16)[name = string("op_48770_cast_fp16")]; + tensor var_48771_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5065_cast_fp16)[name = string("op_48771_cast_fp16")]; + tensor var_48772_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5067_cast_fp16)[name = string("op_48772_cast_fp16")]; + tensor var_48773_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5069_cast_fp16)[name = string("op_48773_cast_fp16")]; + tensor var_48774_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5071_cast_fp16)[name = string("op_48774_cast_fp16")]; + tensor var_48775_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5073_cast_fp16)[name = string("op_48775_cast_fp16")]; + tensor var_48776_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5075_cast_fp16)[name = string("op_48776_cast_fp16")]; + tensor var_48777_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5077_cast_fp16)[name = string("op_48777_cast_fp16")]; + tensor var_48778_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5079_cast_fp16)[name = string("op_48778_cast_fp16")]; + tensor var_48779_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5081_cast_fp16)[name = string("op_48779_cast_fp16")]; + tensor var_48780_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5083_cast_fp16)[name = string("op_48780_cast_fp16")]; + tensor var_48781_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5085_cast_fp16)[name = string("op_48781_cast_fp16")]; + tensor var_48782_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5087_cast_fp16)[name = string("op_48782_cast_fp16")]; + tensor var_48783_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5089_cast_fp16)[name = string("op_48783_cast_fp16")]; + tensor var_48784_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5091_cast_fp16)[name = string("op_48784_cast_fp16")]; + tensor var_48785_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5093_cast_fp16)[name = string("op_48785_cast_fp16")]; + tensor var_48786_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5095_cast_fp16)[name = string("op_48786_cast_fp16")]; + tensor var_48787_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5097_cast_fp16)[name = string("op_48787_cast_fp16")]; + tensor var_48788_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5099_cast_fp16)[name = string("op_48788_cast_fp16")]; + tensor var_48789_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5101_cast_fp16)[name = string("op_48789_cast_fp16")]; + tensor var_48790_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5103_cast_fp16)[name = string("op_48790_cast_fp16")]; + tensor var_48791_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5105_cast_fp16)[name = string("op_48791_cast_fp16")]; + tensor var_48792_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5107_cast_fp16)[name = string("op_48792_cast_fp16")]; + tensor var_48793_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5109_cast_fp16)[name = string("op_48793_cast_fp16")]; + tensor var_48794_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5111_cast_fp16)[name = string("op_48794_cast_fp16")]; + tensor var_48795_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5113_cast_fp16)[name = string("op_48795_cast_fp16")]; + tensor var_48796_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5115_cast_fp16)[name = string("op_48796_cast_fp16")]; + tensor var_48797_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_5117_cast_fp16)[name = string("op_48797_cast_fp16")]; + tensor var_48798_cast_fp16 = softmax(axis = var_47544, x = aw_chunk_cast_fp16)[name = string("op_48798_cast_fp16")]; + string var_48800_equation_0 = const()[name = string("op_48800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48800_cast_fp16 = einsum(equation = var_48800_equation_0, values = (var_48320_cast_fp16, var_48719_cast_fp16))[name = string("op_48800_cast_fp16")]; + string var_48802_equation_0 = const()[name = string("op_48802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48802_cast_fp16 = einsum(equation = var_48802_equation_0, values = (var_48320_cast_fp16, var_48720_cast_fp16))[name = string("op_48802_cast_fp16")]; + string var_48804_equation_0 = const()[name = string("op_48804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48804_cast_fp16 = einsum(equation = var_48804_equation_0, values = (var_48320_cast_fp16, var_48721_cast_fp16))[name = string("op_48804_cast_fp16")]; + string var_48806_equation_0 = const()[name = string("op_48806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48806_cast_fp16 = einsum(equation = var_48806_equation_0, values = (var_48320_cast_fp16, var_48722_cast_fp16))[name = string("op_48806_cast_fp16")]; + string var_48808_equation_0 = const()[name = string("op_48808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48808_cast_fp16 = einsum(equation = var_48808_equation_0, values = (var_48324_cast_fp16, var_48723_cast_fp16))[name = string("op_48808_cast_fp16")]; + string var_48810_equation_0 = const()[name = string("op_48810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48810_cast_fp16 = einsum(equation = var_48810_equation_0, values = (var_48324_cast_fp16, var_48724_cast_fp16))[name = string("op_48810_cast_fp16")]; + string var_48812_equation_0 = const()[name = string("op_48812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48812_cast_fp16 = einsum(equation = var_48812_equation_0, values = (var_48324_cast_fp16, var_48725_cast_fp16))[name = string("op_48812_cast_fp16")]; + string var_48814_equation_0 = const()[name = string("op_48814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48814_cast_fp16 = einsum(equation = var_48814_equation_0, values = (var_48324_cast_fp16, var_48726_cast_fp16))[name = string("op_48814_cast_fp16")]; + string var_48816_equation_0 = const()[name = string("op_48816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48816_cast_fp16 = einsum(equation = var_48816_equation_0, values = (var_48328_cast_fp16, var_48727_cast_fp16))[name = string("op_48816_cast_fp16")]; + string var_48818_equation_0 = const()[name = string("op_48818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48818_cast_fp16 = einsum(equation = var_48818_equation_0, values = (var_48328_cast_fp16, var_48728_cast_fp16))[name = string("op_48818_cast_fp16")]; + string var_48820_equation_0 = const()[name = string("op_48820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48820_cast_fp16 = einsum(equation = var_48820_equation_0, values = (var_48328_cast_fp16, var_48729_cast_fp16))[name = string("op_48820_cast_fp16")]; + string var_48822_equation_0 = const()[name = string("op_48822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48822_cast_fp16 = einsum(equation = var_48822_equation_0, values = (var_48328_cast_fp16, var_48730_cast_fp16))[name = string("op_48822_cast_fp16")]; + string var_48824_equation_0 = const()[name = string("op_48824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48824_cast_fp16 = einsum(equation = var_48824_equation_0, values = (var_48332_cast_fp16, var_48731_cast_fp16))[name = string("op_48824_cast_fp16")]; + string var_48826_equation_0 = const()[name = string("op_48826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48826_cast_fp16 = einsum(equation = var_48826_equation_0, values = (var_48332_cast_fp16, var_48732_cast_fp16))[name = string("op_48826_cast_fp16")]; + string var_48828_equation_0 = const()[name = string("op_48828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48828_cast_fp16 = einsum(equation = var_48828_equation_0, values = (var_48332_cast_fp16, var_48733_cast_fp16))[name = string("op_48828_cast_fp16")]; + string var_48830_equation_0 = const()[name = string("op_48830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48830_cast_fp16 = einsum(equation = var_48830_equation_0, values = (var_48332_cast_fp16, var_48734_cast_fp16))[name = string("op_48830_cast_fp16")]; + string var_48832_equation_0 = const()[name = string("op_48832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48832_cast_fp16 = einsum(equation = var_48832_equation_0, values = (var_48336_cast_fp16, var_48735_cast_fp16))[name = string("op_48832_cast_fp16")]; + string var_48834_equation_0 = const()[name = string("op_48834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48834_cast_fp16 = einsum(equation = var_48834_equation_0, values = (var_48336_cast_fp16, var_48736_cast_fp16))[name = string("op_48834_cast_fp16")]; + string var_48836_equation_0 = const()[name = string("op_48836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48836_cast_fp16 = einsum(equation = var_48836_equation_0, values = (var_48336_cast_fp16, var_48737_cast_fp16))[name = string("op_48836_cast_fp16")]; + string var_48838_equation_0 = const()[name = string("op_48838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48838_cast_fp16 = einsum(equation = var_48838_equation_0, values = (var_48336_cast_fp16, var_48738_cast_fp16))[name = string("op_48838_cast_fp16")]; + string var_48840_equation_0 = const()[name = string("op_48840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48840_cast_fp16 = einsum(equation = var_48840_equation_0, values = (var_48340_cast_fp16, var_48739_cast_fp16))[name = string("op_48840_cast_fp16")]; + string var_48842_equation_0 = const()[name = string("op_48842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48842_cast_fp16 = einsum(equation = var_48842_equation_0, values = (var_48340_cast_fp16, var_48740_cast_fp16))[name = string("op_48842_cast_fp16")]; + string var_48844_equation_0 = const()[name = string("op_48844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48844_cast_fp16 = einsum(equation = var_48844_equation_0, values = (var_48340_cast_fp16, var_48741_cast_fp16))[name = string("op_48844_cast_fp16")]; + string var_48846_equation_0 = const()[name = string("op_48846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48846_cast_fp16 = einsum(equation = var_48846_equation_0, values = (var_48340_cast_fp16, var_48742_cast_fp16))[name = string("op_48846_cast_fp16")]; + string var_48848_equation_0 = const()[name = string("op_48848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48848_cast_fp16 = einsum(equation = var_48848_equation_0, values = (var_48344_cast_fp16, var_48743_cast_fp16))[name = string("op_48848_cast_fp16")]; + string var_48850_equation_0 = const()[name = string("op_48850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48850_cast_fp16 = einsum(equation = var_48850_equation_0, values = (var_48344_cast_fp16, var_48744_cast_fp16))[name = string("op_48850_cast_fp16")]; + string var_48852_equation_0 = const()[name = string("op_48852_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48852_cast_fp16 = einsum(equation = var_48852_equation_0, values = (var_48344_cast_fp16, var_48745_cast_fp16))[name = string("op_48852_cast_fp16")]; + string var_48854_equation_0 = const()[name = string("op_48854_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48854_cast_fp16 = einsum(equation = var_48854_equation_0, values = (var_48344_cast_fp16, var_48746_cast_fp16))[name = string("op_48854_cast_fp16")]; + string var_48856_equation_0 = const()[name = string("op_48856_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48856_cast_fp16 = einsum(equation = var_48856_equation_0, values = (var_48348_cast_fp16, var_48747_cast_fp16))[name = string("op_48856_cast_fp16")]; + string var_48858_equation_0 = const()[name = string("op_48858_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48858_cast_fp16 = einsum(equation = var_48858_equation_0, values = (var_48348_cast_fp16, var_48748_cast_fp16))[name = string("op_48858_cast_fp16")]; + string var_48860_equation_0 = const()[name = string("op_48860_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48860_cast_fp16 = einsum(equation = var_48860_equation_0, values = (var_48348_cast_fp16, var_48749_cast_fp16))[name = string("op_48860_cast_fp16")]; + string var_48862_equation_0 = const()[name = string("op_48862_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48862_cast_fp16 = einsum(equation = var_48862_equation_0, values = (var_48348_cast_fp16, var_48750_cast_fp16))[name = string("op_48862_cast_fp16")]; + string var_48864_equation_0 = const()[name = string("op_48864_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48864_cast_fp16 = einsum(equation = var_48864_equation_0, values = (var_48352_cast_fp16, var_48751_cast_fp16))[name = string("op_48864_cast_fp16")]; + string var_48866_equation_0 = const()[name = string("op_48866_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48866_cast_fp16 = einsum(equation = var_48866_equation_0, values = (var_48352_cast_fp16, var_48752_cast_fp16))[name = string("op_48866_cast_fp16")]; + string var_48868_equation_0 = const()[name = string("op_48868_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48868_cast_fp16 = einsum(equation = var_48868_equation_0, values = (var_48352_cast_fp16, var_48753_cast_fp16))[name = string("op_48868_cast_fp16")]; + string var_48870_equation_0 = const()[name = string("op_48870_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48870_cast_fp16 = einsum(equation = var_48870_equation_0, values = (var_48352_cast_fp16, var_48754_cast_fp16))[name = string("op_48870_cast_fp16")]; + string var_48872_equation_0 = const()[name = string("op_48872_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48872_cast_fp16 = einsum(equation = var_48872_equation_0, values = (var_48356_cast_fp16, var_48755_cast_fp16))[name = string("op_48872_cast_fp16")]; + string var_48874_equation_0 = const()[name = string("op_48874_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48874_cast_fp16 = einsum(equation = var_48874_equation_0, values = (var_48356_cast_fp16, var_48756_cast_fp16))[name = string("op_48874_cast_fp16")]; + string var_48876_equation_0 = const()[name = string("op_48876_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48876_cast_fp16 = einsum(equation = var_48876_equation_0, values = (var_48356_cast_fp16, var_48757_cast_fp16))[name = string("op_48876_cast_fp16")]; + string var_48878_equation_0 = const()[name = string("op_48878_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48878_cast_fp16 = einsum(equation = var_48878_equation_0, values = (var_48356_cast_fp16, var_48758_cast_fp16))[name = string("op_48878_cast_fp16")]; + string var_48880_equation_0 = const()[name = string("op_48880_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48880_cast_fp16 = einsum(equation = var_48880_equation_0, values = (var_48360_cast_fp16, var_48759_cast_fp16))[name = string("op_48880_cast_fp16")]; + string var_48882_equation_0 = const()[name = string("op_48882_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48882_cast_fp16 = einsum(equation = var_48882_equation_0, values = (var_48360_cast_fp16, var_48760_cast_fp16))[name = string("op_48882_cast_fp16")]; + string var_48884_equation_0 = const()[name = string("op_48884_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48884_cast_fp16 = einsum(equation = var_48884_equation_0, values = (var_48360_cast_fp16, var_48761_cast_fp16))[name = string("op_48884_cast_fp16")]; + string var_48886_equation_0 = const()[name = string("op_48886_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48886_cast_fp16 = einsum(equation = var_48886_equation_0, values = (var_48360_cast_fp16, var_48762_cast_fp16))[name = string("op_48886_cast_fp16")]; + string var_48888_equation_0 = const()[name = string("op_48888_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48888_cast_fp16 = einsum(equation = var_48888_equation_0, values = (var_48364_cast_fp16, var_48763_cast_fp16))[name = string("op_48888_cast_fp16")]; + string var_48890_equation_0 = const()[name = string("op_48890_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48890_cast_fp16 = einsum(equation = var_48890_equation_0, values = (var_48364_cast_fp16, var_48764_cast_fp16))[name = string("op_48890_cast_fp16")]; + string var_48892_equation_0 = const()[name = string("op_48892_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48892_cast_fp16 = einsum(equation = var_48892_equation_0, values = (var_48364_cast_fp16, var_48765_cast_fp16))[name = string("op_48892_cast_fp16")]; + string var_48894_equation_0 = const()[name = string("op_48894_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48894_cast_fp16 = einsum(equation = var_48894_equation_0, values = (var_48364_cast_fp16, var_48766_cast_fp16))[name = string("op_48894_cast_fp16")]; + string var_48896_equation_0 = const()[name = string("op_48896_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48896_cast_fp16 = einsum(equation = var_48896_equation_0, values = (var_48368_cast_fp16, var_48767_cast_fp16))[name = string("op_48896_cast_fp16")]; + string var_48898_equation_0 = const()[name = string("op_48898_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48898_cast_fp16 = einsum(equation = var_48898_equation_0, values = (var_48368_cast_fp16, var_48768_cast_fp16))[name = string("op_48898_cast_fp16")]; + string var_48900_equation_0 = const()[name = string("op_48900_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48900_cast_fp16 = einsum(equation = var_48900_equation_0, values = (var_48368_cast_fp16, var_48769_cast_fp16))[name = string("op_48900_cast_fp16")]; + string var_48902_equation_0 = const()[name = string("op_48902_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48902_cast_fp16 = einsum(equation = var_48902_equation_0, values = (var_48368_cast_fp16, var_48770_cast_fp16))[name = string("op_48902_cast_fp16")]; + string var_48904_equation_0 = const()[name = string("op_48904_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48904_cast_fp16 = einsum(equation = var_48904_equation_0, values = (var_48372_cast_fp16, var_48771_cast_fp16))[name = string("op_48904_cast_fp16")]; + string var_48906_equation_0 = const()[name = string("op_48906_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48906_cast_fp16 = einsum(equation = var_48906_equation_0, values = (var_48372_cast_fp16, var_48772_cast_fp16))[name = string("op_48906_cast_fp16")]; + string var_48908_equation_0 = const()[name = string("op_48908_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48908_cast_fp16 = einsum(equation = var_48908_equation_0, values = (var_48372_cast_fp16, var_48773_cast_fp16))[name = string("op_48908_cast_fp16")]; + string var_48910_equation_0 = const()[name = string("op_48910_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48910_cast_fp16 = einsum(equation = var_48910_equation_0, values = (var_48372_cast_fp16, var_48774_cast_fp16))[name = string("op_48910_cast_fp16")]; + string var_48912_equation_0 = const()[name = string("op_48912_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48912_cast_fp16 = einsum(equation = var_48912_equation_0, values = (var_48376_cast_fp16, var_48775_cast_fp16))[name = string("op_48912_cast_fp16")]; + string var_48914_equation_0 = const()[name = string("op_48914_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48914_cast_fp16 = einsum(equation = var_48914_equation_0, values = (var_48376_cast_fp16, var_48776_cast_fp16))[name = string("op_48914_cast_fp16")]; + string var_48916_equation_0 = const()[name = string("op_48916_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48916_cast_fp16 = einsum(equation = var_48916_equation_0, values = (var_48376_cast_fp16, var_48777_cast_fp16))[name = string("op_48916_cast_fp16")]; + string var_48918_equation_0 = const()[name = string("op_48918_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48918_cast_fp16 = einsum(equation = var_48918_equation_0, values = (var_48376_cast_fp16, var_48778_cast_fp16))[name = string("op_48918_cast_fp16")]; + string var_48920_equation_0 = const()[name = string("op_48920_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48920_cast_fp16 = einsum(equation = var_48920_equation_0, values = (var_48380_cast_fp16, var_48779_cast_fp16))[name = string("op_48920_cast_fp16")]; + string var_48922_equation_0 = const()[name = string("op_48922_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48922_cast_fp16 = einsum(equation = var_48922_equation_0, values = (var_48380_cast_fp16, var_48780_cast_fp16))[name = string("op_48922_cast_fp16")]; + string var_48924_equation_0 = const()[name = string("op_48924_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48924_cast_fp16 = einsum(equation = var_48924_equation_0, values = (var_48380_cast_fp16, var_48781_cast_fp16))[name = string("op_48924_cast_fp16")]; + string var_48926_equation_0 = const()[name = string("op_48926_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48926_cast_fp16 = einsum(equation = var_48926_equation_0, values = (var_48380_cast_fp16, var_48782_cast_fp16))[name = string("op_48926_cast_fp16")]; + string var_48928_equation_0 = const()[name = string("op_48928_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48928_cast_fp16 = einsum(equation = var_48928_equation_0, values = (var_48384_cast_fp16, var_48783_cast_fp16))[name = string("op_48928_cast_fp16")]; + string var_48930_equation_0 = const()[name = string("op_48930_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48930_cast_fp16 = einsum(equation = var_48930_equation_0, values = (var_48384_cast_fp16, var_48784_cast_fp16))[name = string("op_48930_cast_fp16")]; + string var_48932_equation_0 = const()[name = string("op_48932_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48932_cast_fp16 = einsum(equation = var_48932_equation_0, values = (var_48384_cast_fp16, var_48785_cast_fp16))[name = string("op_48932_cast_fp16")]; + string var_48934_equation_0 = const()[name = string("op_48934_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48934_cast_fp16 = einsum(equation = var_48934_equation_0, values = (var_48384_cast_fp16, var_48786_cast_fp16))[name = string("op_48934_cast_fp16")]; + string var_48936_equation_0 = const()[name = string("op_48936_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48936_cast_fp16 = einsum(equation = var_48936_equation_0, values = (var_48388_cast_fp16, var_48787_cast_fp16))[name = string("op_48936_cast_fp16")]; + string var_48938_equation_0 = const()[name = string("op_48938_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48938_cast_fp16 = einsum(equation = var_48938_equation_0, values = (var_48388_cast_fp16, var_48788_cast_fp16))[name = string("op_48938_cast_fp16")]; + string var_48940_equation_0 = const()[name = string("op_48940_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48940_cast_fp16 = einsum(equation = var_48940_equation_0, values = (var_48388_cast_fp16, var_48789_cast_fp16))[name = string("op_48940_cast_fp16")]; + string var_48942_equation_0 = const()[name = string("op_48942_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48942_cast_fp16 = einsum(equation = var_48942_equation_0, values = (var_48388_cast_fp16, var_48790_cast_fp16))[name = string("op_48942_cast_fp16")]; + string var_48944_equation_0 = const()[name = string("op_48944_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48944_cast_fp16 = einsum(equation = var_48944_equation_0, values = (var_48392_cast_fp16, var_48791_cast_fp16))[name = string("op_48944_cast_fp16")]; + string var_48946_equation_0 = const()[name = string("op_48946_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48946_cast_fp16 = einsum(equation = var_48946_equation_0, values = (var_48392_cast_fp16, var_48792_cast_fp16))[name = string("op_48946_cast_fp16")]; + string var_48948_equation_0 = const()[name = string("op_48948_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48948_cast_fp16 = einsum(equation = var_48948_equation_0, values = (var_48392_cast_fp16, var_48793_cast_fp16))[name = string("op_48948_cast_fp16")]; + string var_48950_equation_0 = const()[name = string("op_48950_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48950_cast_fp16 = einsum(equation = var_48950_equation_0, values = (var_48392_cast_fp16, var_48794_cast_fp16))[name = string("op_48950_cast_fp16")]; + string var_48952_equation_0 = const()[name = string("op_48952_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48952_cast_fp16 = einsum(equation = var_48952_equation_0, values = (var_48396_cast_fp16, var_48795_cast_fp16))[name = string("op_48952_cast_fp16")]; + string var_48954_equation_0 = const()[name = string("op_48954_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48954_cast_fp16 = einsum(equation = var_48954_equation_0, values = (var_48396_cast_fp16, var_48796_cast_fp16))[name = string("op_48954_cast_fp16")]; + string var_48956_equation_0 = const()[name = string("op_48956_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48956_cast_fp16 = einsum(equation = var_48956_equation_0, values = (var_48396_cast_fp16, var_48797_cast_fp16))[name = string("op_48956_cast_fp16")]; + string var_48958_equation_0 = const()[name = string("op_48958_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_48958_cast_fp16 = einsum(equation = var_48958_equation_0, values = (var_48396_cast_fp16, var_48798_cast_fp16))[name = string("op_48958_cast_fp16")]; + bool var_48960_interleave_0 = const()[name = string("op_48960_interleave_0"), val = bool(false)]; + tensor var_48960_cast_fp16 = concat(axis = var_47519, interleave = var_48960_interleave_0, values = (var_48800_cast_fp16, var_48802_cast_fp16, var_48804_cast_fp16, var_48806_cast_fp16))[name = string("op_48960_cast_fp16")]; + bool var_48962_interleave_0 = const()[name = string("op_48962_interleave_0"), val = bool(false)]; + tensor var_48962_cast_fp16 = concat(axis = var_47519, interleave = var_48962_interleave_0, values = (var_48808_cast_fp16, var_48810_cast_fp16, var_48812_cast_fp16, var_48814_cast_fp16))[name = string("op_48962_cast_fp16")]; + bool var_48964_interleave_0 = const()[name = string("op_48964_interleave_0"), val = bool(false)]; + tensor var_48964_cast_fp16 = concat(axis = var_47519, interleave = var_48964_interleave_0, values = (var_48816_cast_fp16, var_48818_cast_fp16, var_48820_cast_fp16, var_48822_cast_fp16))[name = string("op_48964_cast_fp16")]; + bool var_48966_interleave_0 = const()[name = string("op_48966_interleave_0"), val = bool(false)]; + tensor var_48966_cast_fp16 = concat(axis = var_47519, interleave = var_48966_interleave_0, values = (var_48824_cast_fp16, var_48826_cast_fp16, var_48828_cast_fp16, var_48830_cast_fp16))[name = string("op_48966_cast_fp16")]; + bool var_48968_interleave_0 = const()[name = string("op_48968_interleave_0"), val = bool(false)]; + tensor var_48968_cast_fp16 = concat(axis = var_47519, interleave = var_48968_interleave_0, values = (var_48832_cast_fp16, var_48834_cast_fp16, var_48836_cast_fp16, var_48838_cast_fp16))[name = string("op_48968_cast_fp16")]; + bool var_48970_interleave_0 = const()[name = string("op_48970_interleave_0"), val = bool(false)]; + tensor var_48970_cast_fp16 = concat(axis = var_47519, interleave = var_48970_interleave_0, values = (var_48840_cast_fp16, var_48842_cast_fp16, var_48844_cast_fp16, var_48846_cast_fp16))[name = string("op_48970_cast_fp16")]; + bool var_48972_interleave_0 = const()[name = string("op_48972_interleave_0"), val = bool(false)]; + tensor var_48972_cast_fp16 = concat(axis = var_47519, interleave = var_48972_interleave_0, values = (var_48848_cast_fp16, var_48850_cast_fp16, var_48852_cast_fp16, var_48854_cast_fp16))[name = string("op_48972_cast_fp16")]; + bool var_48974_interleave_0 = const()[name = string("op_48974_interleave_0"), val = bool(false)]; + tensor var_48974_cast_fp16 = concat(axis = var_47519, interleave = var_48974_interleave_0, values = (var_48856_cast_fp16, var_48858_cast_fp16, var_48860_cast_fp16, var_48862_cast_fp16))[name = string("op_48974_cast_fp16")]; + bool var_48976_interleave_0 = const()[name = string("op_48976_interleave_0"), val = bool(false)]; + tensor var_48976_cast_fp16 = concat(axis = var_47519, interleave = var_48976_interleave_0, values = (var_48864_cast_fp16, var_48866_cast_fp16, var_48868_cast_fp16, var_48870_cast_fp16))[name = string("op_48976_cast_fp16")]; + bool var_48978_interleave_0 = const()[name = string("op_48978_interleave_0"), val = bool(false)]; + tensor var_48978_cast_fp16 = concat(axis = var_47519, interleave = var_48978_interleave_0, values = (var_48872_cast_fp16, var_48874_cast_fp16, var_48876_cast_fp16, var_48878_cast_fp16))[name = string("op_48978_cast_fp16")]; + bool var_48980_interleave_0 = const()[name = string("op_48980_interleave_0"), val = bool(false)]; + tensor var_48980_cast_fp16 = concat(axis = var_47519, interleave = var_48980_interleave_0, values = (var_48880_cast_fp16, var_48882_cast_fp16, var_48884_cast_fp16, var_48886_cast_fp16))[name = string("op_48980_cast_fp16")]; + bool var_48982_interleave_0 = const()[name = string("op_48982_interleave_0"), val = bool(false)]; + tensor var_48982_cast_fp16 = concat(axis = var_47519, interleave = var_48982_interleave_0, values = (var_48888_cast_fp16, var_48890_cast_fp16, var_48892_cast_fp16, var_48894_cast_fp16))[name = string("op_48982_cast_fp16")]; + bool var_48984_interleave_0 = const()[name = string("op_48984_interleave_0"), val = bool(false)]; + tensor var_48984_cast_fp16 = concat(axis = var_47519, interleave = var_48984_interleave_0, values = (var_48896_cast_fp16, var_48898_cast_fp16, var_48900_cast_fp16, var_48902_cast_fp16))[name = string("op_48984_cast_fp16")]; + bool var_48986_interleave_0 = const()[name = string("op_48986_interleave_0"), val = bool(false)]; + tensor var_48986_cast_fp16 = concat(axis = var_47519, interleave = var_48986_interleave_0, values = (var_48904_cast_fp16, var_48906_cast_fp16, var_48908_cast_fp16, var_48910_cast_fp16))[name = string("op_48986_cast_fp16")]; + bool var_48988_interleave_0 = const()[name = string("op_48988_interleave_0"), val = bool(false)]; + tensor var_48988_cast_fp16 = concat(axis = var_47519, interleave = var_48988_interleave_0, values = (var_48912_cast_fp16, var_48914_cast_fp16, var_48916_cast_fp16, var_48918_cast_fp16))[name = string("op_48988_cast_fp16")]; + bool var_48990_interleave_0 = const()[name = string("op_48990_interleave_0"), val = bool(false)]; + tensor var_48990_cast_fp16 = concat(axis = var_47519, interleave = var_48990_interleave_0, values = (var_48920_cast_fp16, var_48922_cast_fp16, var_48924_cast_fp16, var_48926_cast_fp16))[name = string("op_48990_cast_fp16")]; + bool var_48992_interleave_0 = const()[name = string("op_48992_interleave_0"), val = bool(false)]; + tensor var_48992_cast_fp16 = concat(axis = var_47519, interleave = var_48992_interleave_0, values = (var_48928_cast_fp16, var_48930_cast_fp16, var_48932_cast_fp16, var_48934_cast_fp16))[name = string("op_48992_cast_fp16")]; + bool var_48994_interleave_0 = const()[name = string("op_48994_interleave_0"), val = bool(false)]; + tensor var_48994_cast_fp16 = concat(axis = var_47519, interleave = var_48994_interleave_0, values = (var_48936_cast_fp16, var_48938_cast_fp16, var_48940_cast_fp16, var_48942_cast_fp16))[name = string("op_48994_cast_fp16")]; + bool var_48996_interleave_0 = const()[name = string("op_48996_interleave_0"), val = bool(false)]; + tensor var_48996_cast_fp16 = concat(axis = var_47519, interleave = var_48996_interleave_0, values = (var_48944_cast_fp16, var_48946_cast_fp16, var_48948_cast_fp16, var_48950_cast_fp16))[name = string("op_48996_cast_fp16")]; + bool var_48998_interleave_0 = const()[name = string("op_48998_interleave_0"), val = bool(false)]; + tensor var_48998_cast_fp16 = concat(axis = var_47519, interleave = var_48998_interleave_0, values = (var_48952_cast_fp16, var_48954_cast_fp16, var_48956_cast_fp16, var_48958_cast_fp16))[name = string("op_48998_cast_fp16")]; + bool input_249_interleave_0 = const()[name = string("input_249_interleave_0"), val = bool(false)]; + tensor input_249_cast_fp16 = concat(axis = var_47544, interleave = input_249_interleave_0, values = (var_48960_cast_fp16, var_48962_cast_fp16, var_48964_cast_fp16, var_48966_cast_fp16, var_48968_cast_fp16, var_48970_cast_fp16, var_48972_cast_fp16, var_48974_cast_fp16, var_48976_cast_fp16, var_48978_cast_fp16, var_48980_cast_fp16, var_48982_cast_fp16, var_48984_cast_fp16, var_48986_cast_fp16, var_48988_cast_fp16, var_48990_cast_fp16, var_48992_cast_fp16, var_48994_cast_fp16, var_48996_cast_fp16, var_48998_cast_fp16))[name = string("input_249_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_31_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1244456960)))]; + tensor layers_31_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_31_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247733824)))]; + tensor obj_cast_fp16 = conv(bias = layers_31_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_31_self_attn_o_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = obj_cast_fp16)[name = string("inputs_127_cast_fp16")]; + tensor out_127_axes_0 = const()[name = string("out_127_axes_0"), val = tensor([1])]; + fp16 var_49017_to_fp16 = const()[name = string("op_49017_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_49017_to_fp16, x = inputs_127_cast_fp16)[name = string("out_127_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = string("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247736448)))]; + tensor input_251_beta_0_to_fp16 = const()[name = string("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247739072)))]; + fp16 input_251_epsilon_0_to_fp16 = const()[name = string("input_251_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = string("input_251_cast_fp16")]; + string input_253_pad_type_0 = const()[name = string("input_253_pad_type_0"), val = string("valid")]; + tensor input_253_strides_0 = const()[name = string("input_253_strides_0"), val = tensor([1, 1])]; + tensor input_253_pad_0 = const()[name = string("input_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_253_dilations_0 = const()[name = string("input_253_dilations_0"), val = tensor([1, 1])]; + int32 input_253_groups_0 = const()[name = string("input_253_groups_0"), val = int32(1)]; + tensor layers_31_fc1_weight_to_fp16 = const()[name = string("layers_31_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1247741696)))]; + tensor layers_31_fc1_bias_to_fp16 = const()[name = string("layers_31_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260848960)))]; + tensor input_253_cast_fp16 = conv(bias = layers_31_fc1_bias_to_fp16, dilations = input_253_dilations_0, groups = input_253_groups_0, pad = input_253_pad_0, pad_type = input_253_pad_type_0, strides = input_253_strides_0, weight = layers_31_fc1_weight_to_fp16, x = input_251_cast_fp16)[name = string("input_253_cast_fp16")]; + string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("EXACT")]; + tensor input_255_cast_fp16 = gelu(mode = input_255_mode_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_31_fc2_weight_to_fp16 = const()[name = string("layers_31_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1260859264)))]; + tensor layers_31_fc2_bias_to_fp16 = const()[name = string("layers_31_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273966528)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_31_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_31_fc2_weight_to_fp16, x = input_255_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_127_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_49055_to_fp16 = const()[name = string("op_49055_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_49055_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273969152)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273971776)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_49079_pad_type_0 = const()[name = string("op_49079_pad_type_0"), val = string("valid")]; + tensor var_49079_strides_0 = const()[name = string("op_49079_strides_0"), val = tensor([1, 1])]; + tensor var_49079_pad_0 = const()[name = string("op_49079_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49079_dilations_0 = const()[name = string("op_49079_dilations_0"), val = tensor([1, 1])]; + int32 var_49079_groups_0 = const()[name = string("op_49079_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1273974400)))]; + tensor var_49079_cast_fp16 = conv(dilations = var_49079_dilations_0, groups = var_49079_groups_0, pad = var_49079_pad_0, pad_type = var_49079_pad_type_0, strides = var_49079_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49079_cast_fp16")]; + string var_49086_pad_type_0 = const()[name = string("op_49086_pad_type_0"), val = string("valid")]; + tensor var_49086_strides_0 = const()[name = string("op_49086_strides_0"), val = tensor([1, 1])]; + tensor var_49086_pad_0 = const()[name = string("op_49086_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49086_dilations_0 = const()[name = string("op_49086_dilations_0"), val = tensor([1, 1])]; + int32 var_49086_groups_0 = const()[name = string("op_49086_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1277251264)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280528128)))]; + tensor var_49086_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_49086_dilations_0, groups = var_49086_groups_0, pad = var_49086_pad_0, pad_type = var_49086_pad_type_0, strides = var_49086_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49086_cast_fp16")]; + string var_49104_pad_type_0 = const()[name = string("op_49104_pad_type_0"), val = string("valid")]; + tensor var_49104_strides_0 = const()[name = string("op_49104_strides_0"), val = tensor([1, 1])]; + tensor var_49104_pad_0 = const()[name = string("op_49104_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49104_dilations_0 = const()[name = string("op_49104_dilations_0"), val = tensor([1, 1])]; + int32 var_49104_groups_0 = const()[name = string("op_49104_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1280530752)))]; + tensor var_49104_cast_fp16 = conv(dilations = var_49104_dilations_0, groups = var_49104_groups_0, pad = var_49104_pad_0, pad_type = var_49104_pad_type_0, strides = var_49104_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49104_cast_fp16")]; + string var_49111_pad_type_0 = const()[name = string("op_49111_pad_type_0"), val = string("valid")]; + tensor var_49111_strides_0 = const()[name = string("op_49111_strides_0"), val = tensor([1, 1])]; + tensor var_49111_pad_0 = const()[name = string("op_49111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49111_dilations_0 = const()[name = string("op_49111_dilations_0"), val = tensor([1, 1])]; + int32 var_49111_groups_0 = const()[name = string("op_49111_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1283807616)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1287084480)))]; + tensor var_49111_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_49111_dilations_0, groups = var_49111_groups_0, pad = var_49111_pad_0, pad_type = var_49111_pad_type_0, strides = var_49111_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49111_cast_fp16")]; + string var_49129_pad_type_0 = const()[name = string("op_49129_pad_type_0"), val = string("valid")]; + tensor var_49129_strides_0 = const()[name = string("op_49129_strides_0"), val = tensor([1, 1])]; + tensor var_49129_pad_0 = const()[name = string("op_49129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49129_dilations_0 = const()[name = string("op_49129_dilations_0"), val = tensor([1, 1])]; + int32 var_49129_groups_0 = const()[name = string("op_49129_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1287087104)))]; + tensor var_49129_cast_fp16 = conv(dilations = var_49129_dilations_0, groups = var_49129_groups_0, pad = var_49129_pad_0, pad_type = var_49129_pad_type_0, strides = var_49129_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49129_cast_fp16")]; + string var_49136_pad_type_0 = const()[name = string("op_49136_pad_type_0"), val = string("valid")]; + tensor var_49136_strides_0 = const()[name = string("op_49136_strides_0"), val = tensor([1, 1])]; + tensor var_49136_pad_0 = const()[name = string("op_49136_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49136_dilations_0 = const()[name = string("op_49136_dilations_0"), val = tensor([1, 1])]; + int32 var_49136_groups_0 = const()[name = string("op_49136_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1290363968)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1293640832)))]; + tensor var_49136_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_49136_dilations_0, groups = var_49136_groups_0, pad = var_49136_pad_0, pad_type = var_49136_pad_type_0, strides = var_49136_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_49136_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1293643456)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1296920320)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1300197184)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_49166 = const()[name = string("op_49166"), val = int32(0)]; + bool input_259_interleave_0 = const()[name = string("input_259_interleave_0"), val = bool(false)]; + tensor input_259_cast_fp16 = concat(axis = var_49166, interleave = input_259_interleave_0, values = (var_49079_cast_fp16, var_49104_cast_fp16, var_49129_cast_fp16, k_cast_fp16))[name = string("input_259_cast_fp16")]; + int32 var_49169 = const()[name = string("op_49169"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_49169, interleave = input_interleave_0, values = (var_49086_cast_fp16, var_49111_cast_fp16, var_49136_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_49176_pad_0 = const()[name = string("op_49176_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_49176_mode_0 = const()[name = string("op_49176_mode_0"), val = string("constant")]; + fp16 const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_33_to_fp16, mode = var_49176_mode_0, pad = var_49176_pad_0, x = input_259_cast_fp16)[name = string("op_49176_cast_fp16")]; + tensor var_49182_pad_0 = const()[name = string("op_49182_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_49182_mode_0 = const()[name = string("op_49182_mode_0"), val = string("constant")]; + fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_34_to_fp16, mode = var_49182_mode_0, pad = var_49182_pad_0, x = input_cast_fp16)[name = string("op_49182_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..b7782cab9f13e7f9bfb526d1e60e6b169f9cf6df --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:151f7b8578b144f50f29cd3542a1247bf6f2ba707d6ac0a480b630d82fe3d236 +size 1300199808 diff --git a/openai_whisper-large-v3-v20240930_turbo/LICENSE_NOTICE.txt b/openai_whisper-large-v3-v20240930_turbo/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..3ba3246801c85f92f79ac029f59b94e7fb646f85 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0980462db89a546e1e90888ea38e0a5ddf1f1fec84608802cdbb12f8a5cc7215 +size 243 diff --git a/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..780171e73cd57a772ec0457470f0c8b86f4c73cd --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6475c6649047ce609e3fe84b2525843c03342820662404540baf28146c174014 +size 329 diff --git a/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..65be90aad1d0e5f73a1f50b19705ccad3c0da822 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 128 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 128, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..6cf57d7dbf15af35e56636caf15aff60353296f0 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..2ae170c9000db89326cc2600450001654bb10f7f --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:009d9fb8f6b589accfa08cebf1c712ef07c3405229ce3cfb3a57ee033c9d8a49 +size 373376 diff --git a/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..e663d681b043c4befaf670427a5aa56c5e6f35f7 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:554d32ced57d792d052df3888d8de65b1bd12c8ad56f876979de0beff2a07abc +size 243 diff --git a/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..89928605833c60a71a2f6f55f19c9a96d133f403 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4901c5e4249e42e8f37325412eb6fcf9ca9c5e22660271613675afed77cff8f +size 754 diff --git a/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/metadata.json b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..06d7d6050d5ca1219bfe92c01608ff3f67bb7365 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51866)", + "shortDescription" : "", + "shape" : "[1, 1, 51866]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 5120 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 5120, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 5120 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 5120, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 8, + "Ios18.mul" : 16, + "Ios18.matmul" : 16, + "Ios18.batchNorm" : 13, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 29, + "Ios18.layerNorm" : 13, + "Ios18.reshape" : 32, + "Ios18.linear" : 1, + "Ios18.conv" : 32, + "Ios18.gelu" : 4, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 20, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 1280 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 1280, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.version" : "8.0", + "com.github.apple.coremltools.source" : "torch==2.5.1" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/model.mil b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..9f060563f9c7e6cab241feb43aab826c7414c868 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,679 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; + int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; + bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_26_cast_fp16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = input_ids, validate_indices = var_26_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_26_cast_fp16")]; + int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; + int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; + bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132777088)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_71")]; + tensor var_30_cast_fp16_cast_uint16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_30_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_30_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_26_cast_fp16, y = var_30_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_44_axes_0 = const()[name = string("op_44_axes_0"), val = tensor([2])]; + tensor var_44_cast_fp16 = expand_dims(axes = var_44_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_44_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_44_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1])]; + int32 var_49_axis_0 = const()[name = string("op_49_axis_0"), val = int32(0)]; + tensor var_49_cast_fp16_0, tensor var_49_cast_fp16_1, tensor var_49_cast_fp16_2, tensor var_49_cast_fp16_3 = split(axis = var_49_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_49_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1])]; + int32 var_56_axis_0 = const()[name = string("op_56_axis_0"), val = int32(0)]; + tensor var_56_cast_fp16_0, tensor var_56_cast_fp16_1, tensor var_56_cast_fp16_2, tensor var_56_cast_fp16_3 = split(axis = var_56_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_56_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 1280, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_76 = const()[name = string("op_76"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_101_to_fp16 = const()[name = string("op_101_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_101_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133924032)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133926656)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133929280)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133931904)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133934528)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137211392)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137214016)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140490880)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143767744)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor([1])]; + tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = kv_cache_update_mask)[name = string("op_136_cast_fp16")]; + tensor var_137_axes_0 = const()[name = string("op_137_axes_0"), val = tensor([2])]; + tensor var_137_cast_fp16 = expand_dims(axes = var_137_axes_0, x = var_136_cast_fp16)[name = string("op_137_cast_fp16")]; + tensor var_139_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_139_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_49_cast_fp16_0, y = var_139_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_141_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_141_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_56_cast_fp16_0, y = var_141_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_144 = const()[name = string("op_144"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_144, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_146_to_fp16 = const()[name = string("op_146_to_fp16"), val = fp16(0x1p-3)]; + tensor var_147_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_146_to_fp16)[name = string("op_147_cast_fp16")]; + tensor var_148 = const()[name = string("op_148"), val = tensor([1, 20, 64, -1])]; + tensor var_149_cast_fp16 = reshape(shape = var_148, x = key_1_cast_fp16)[name = string("op_149_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_147_cast_fp16, y = var_149_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_153_axes_0 = const()[name = string("op_153_axes_0"), val = tensor([1])]; + tensor var_153_cast_fp16 = expand_dims(axes = var_153_axes_0, x = decoder_key_padding_mask)[name = string("op_153_cast_fp16")]; + tensor var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor([2])]; + tensor var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = var_153_cast_fp16)[name = string("op_154_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_157_cast_fp16 = softmax(axis = var_76, x = mh_w_3_cast_fp16)[name = string("op_157_cast_fp16")]; + tensor var_158 = const()[name = string("op_158"), val = tensor([1, 20, 64, -1])]; + tensor var_159_cast_fp16 = reshape(shape = var_158, x = value_1_cast_fp16)[name = string("op_159_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_159_cast_fp16, y = var_157_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_162 = const()[name = string("op_162"), val = tensor([1, 1280, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_162, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143770368)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147047232)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_184_to_fp16 = const()[name = string("op_184_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_184_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147049856)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147052480)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147055104)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150331968)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_204 = const()[name = string("op_204"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_204, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_206_to_fp16 = const()[name = string("op_206_to_fp16"), val = fp16(0x1p-3)]; + tensor var_207_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_206_to_fp16)[name = string("op_207_cast_fp16")]; + tensor var_208 = const()[name = string("op_208"), val = tensor([1, 20, 64, -1])]; + tensor var_209_cast_fp16 = reshape(shape = var_208, x = obj_17_cast_fp16)[name = string("op_209_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_207_cast_fp16, y = var_209_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_213_axes_0 = const()[name = string("op_213_axes_0"), val = tensor([1])]; + tensor var_213_cast_fp16 = expand_dims(axes = var_213_axes_0, x = read_state_4)[name = string("op_213_cast_fp16")]; + tensor var_214_axes_0 = const()[name = string("op_214_axes_0"), val = tensor([2])]; + tensor var_214_cast_fp16 = expand_dims(axes = var_214_axes_0, x = var_213_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_76, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_218 = const()[name = string("op_218"), val = tensor([1, 20, 64, -1])]; + tensor var_219_cast_fp16 = reshape(shape = var_218, x = obj_19_cast_fp16)[name = string("op_219_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_219_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_222 = const()[name = string("op_222"), val = tensor([1, 1280, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_222, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334592)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_240_to_fp16 = const()[name = string("op_240_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_240_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614080)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616704)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153619328)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166726592)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166736896)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179844160)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 1280, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_285 = const()[name = string("op_285"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_310_to_fp16 = const()[name = string("op_310_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_310_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179846784)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179849408)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179852032)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128896)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183131520)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186408384)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189685248)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_348_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_348_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_49_cast_fp16_1, y = var_348_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_350_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_350_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_56_cast_fp16_1, y = var_350_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_353 = const()[name = string("op_353"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_353, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_355_to_fp16 = const()[name = string("op_355_to_fp16"), val = fp16(0x1p-3)]; + tensor var_356_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_355_to_fp16)[name = string("op_356_cast_fp16")]; + tensor var_357 = const()[name = string("op_357"), val = tensor([1, 20, 64, -1])]; + tensor var_358_cast_fp16 = reshape(shape = var_357, x = key_3_cast_fp16)[name = string("op_358_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_356_cast_fp16, y = var_358_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_366_cast_fp16 = softmax(axis = var_285, x = mh_w_11_cast_fp16)[name = string("op_366_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([1, 20, 64, -1])]; + tensor var_368_cast_fp16 = reshape(shape = var_367, x = value_3_cast_fp16)[name = string("op_368_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_368_cast_fp16, y = var_366_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_371 = const()[name = string("op_371"), val = tensor([1, 1280, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_371, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189687872)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192964736)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_393_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192967360)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192969984)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192972608)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196249472)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_413 = const()[name = string("op_413"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_413, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_415_to_fp16 = const()[name = string("op_415_to_fp16"), val = fp16(0x1p-3)]; + tensor var_416_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_415_to_fp16)[name = string("op_416_cast_fp16")]; + tensor var_417 = const()[name = string("op_417"), val = tensor([1, 20, 64, -1])]; + tensor var_418_cast_fp16 = reshape(shape = var_417, x = obj_35_cast_fp16)[name = string("op_418_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_416_cast_fp16, y = var_418_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_285, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_427 = const()[name = string("op_427"), val = tensor([1, 20, 64, -1])]; + tensor var_428_cast_fp16 = reshape(shape = var_427, x = obj_37_cast_fp16)[name = string("op_428_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_428_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_431 = const()[name = string("op_431"), val = tensor([1, 1280, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_431, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196252096)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199528960)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_449_to_fp16 = const()[name = string("op_449_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_449_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199531584)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199534208)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199536832)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212644096)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212654400)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225761664)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 1280, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 1280, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_494 = const()[name = string("op_494"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_519_to_fp16 = const()[name = string("op_519_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_519_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225764288)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225766912)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225769536)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229046400)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229049024)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232325888)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235602752)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_557_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_557_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_49_cast_fp16_2, y = var_557_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_559_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_559_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_56_cast_fp16_2, y = var_559_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_562 = const()[name = string("op_562"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_562, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_564_to_fp16 = const()[name = string("op_564_to_fp16"), val = fp16(0x1p-3)]; + tensor var_565_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_564_to_fp16)[name = string("op_565_cast_fp16")]; + tensor var_566 = const()[name = string("op_566"), val = tensor([1, 20, 64, -1])]; + tensor var_567_cast_fp16 = reshape(shape = var_566, x = key_5_cast_fp16)[name = string("op_567_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_565_cast_fp16, y = var_567_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_575_cast_fp16 = softmax(axis = var_494, x = mh_w_19_cast_fp16)[name = string("op_575_cast_fp16")]; + tensor var_576 = const()[name = string("op_576"), val = tensor([1, 20, 64, -1])]; + tensor var_577_cast_fp16 = reshape(shape = var_576, x = value_5_cast_fp16)[name = string("op_577_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_577_cast_fp16, y = var_575_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_580 = const()[name = string("op_580"), val = tensor([1, 1280, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_580, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235605376)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238882240)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_602_to_fp16 = const()[name = string("op_602_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_602_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238884864)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238887488)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238890112)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242166976)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_622 = const()[name = string("op_622"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_622, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_624_to_fp16 = const()[name = string("op_624_to_fp16"), val = fp16(0x1p-3)]; + tensor var_625_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_624_to_fp16)[name = string("op_625_cast_fp16")]; + tensor var_626 = const()[name = string("op_626"), val = tensor([1, 20, 64, -1])]; + tensor var_627_cast_fp16 = reshape(shape = var_626, x = obj_53_cast_fp16)[name = string("op_627_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_625_cast_fp16, y = var_627_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_494, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_636 = const()[name = string("op_636"), val = tensor([1, 20, 64, -1])]; + tensor var_637_cast_fp16 = reshape(shape = var_636, x = obj_55_cast_fp16)[name = string("op_637_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_637_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_640 = const()[name = string("op_640"), val = tensor([1, 1280, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_640, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242169600)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245446464)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_661_to_fp16 = const()[name = string("op_661_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_661_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245449088)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245451712)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245454336)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258561600)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258571904)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271679168)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 1280, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 1280, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_707 = const()[name = string("op_707"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_732_to_fp16 = const()[name = string("op_732_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_732_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271681792)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271684416)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271687040)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274963904)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274966528)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278243392)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281520256)))]; + tensor current_value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_770_cast_fp16 = mul(x = current_key_cast_fp16, y = var_137_cast_fp16)[name = string("op_770_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_49_cast_fp16_3, y = var_770_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_772_cast_fp16 = mul(x = current_value_cast_fp16, y = var_137_cast_fp16)[name = string("op_772_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_56_cast_fp16_3, y = var_772_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_775 = const()[name = string("op_775"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_775, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_777_to_fp16 = const()[name = string("op_777_to_fp16"), val = fp16(0x1p-3)]; + tensor var_778_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_777_to_fp16)[name = string("op_778_cast_fp16")]; + tensor var_779 = const()[name = string("op_779"), val = tensor([1, 20, 64, -1])]; + tensor var_780_cast_fp16 = reshape(shape = var_779, x = key_cast_fp16)[name = string("op_780_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_778_cast_fp16, y = var_780_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_788_cast_fp16 = softmax(axis = var_707, x = mh_w_27_cast_fp16)[name = string("op_788_cast_fp16")]; + tensor var_789 = const()[name = string("op_789"), val = tensor([1, 20, 64, -1])]; + tensor var_790_cast_fp16 = reshape(shape = var_789, x = value_cast_fp16)[name = string("op_790_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_790_cast_fp16, y = var_788_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_793 = const()[name = string("op_793"), val = tensor([1, 1280, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_793, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281522880)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284799744)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_815_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284802368)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284804992)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284807616)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288084480)))]; + tensor query_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_835 = const()[name = string("op_835"), val = tensor([1, 20, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_835, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_837_to_fp16 = const()[name = string("op_837_to_fp16"), val = fp16(0x1p-3)]; + tensor var_838_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_837_to_fp16)[name = string("op_838_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 20, 64, -1])]; + tensor var_840_cast_fp16 = reshape(shape = var_839, x = obj_71_cast_fp16)[name = string("op_840_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_838_cast_fp16, y = var_840_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_707, x = mh_w_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_849 = const()[name = string("op_849"), val = tensor([1, 20, 64, -1])]; + tensor var_850_cast_fp16 = reshape(shape = var_849, x = obj_73_cast_fp16)[name = string("op_850_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_850_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_853 = const()[name = string("op_853"), val = tensor([1, 1280, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_853, x = attn_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288087104)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291363968)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_874_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291366592)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291369216)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291371840)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304479104)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_37_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304489408)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317596672)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_917_to_fp16 = const()[name = string("op_917_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_917_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317599296)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317601920)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_928_axes_0 = const()[name = string("op_928_axes_0"), val = tensor([2])]; + tensor var_928_cast_fp16 = squeeze(axes = var_928_axes_0, x = hidden_states_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_931_perm_0 = const()[name = string("op_931_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317604544)))]; + tensor var_931_cast_fp16 = transpose(perm = var_931_perm_0, x = var_928_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_931_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_935 = const()[name = string("op_935"), val = int32(1)]; + bool obj_81_interleave_0 = const()[name = string("obj_81_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_935, interleave = obj_81_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_cast_fp16))[name = string("obj_81_cast_fp16")]; + int32 var_938 = const()[name = string("op_938"), val = int32(1)]; + bool obj_83_interleave_0 = const()[name = string("obj_83_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_938, interleave = obj_83_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_cast_fp16))[name = string("obj_83_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = obj_59_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_952_begin_0 = const()[name = string("op_952_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_952_end_0 = const()[name = string("op_952_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_952_end_mask_0 = const()[name = string("op_952_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_952_squeeze_mask_0 = const()[name = string("op_952_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_952_cast_fp16 = slice_by_index(begin = var_952_begin_0, end = var_952_end_0, end_mask = var_952_end_mask_0, squeeze_mask = var_952_squeeze_mask_0, x = var_949_cast_fp16)[name = string("op_952_cast_fp16")]; + tensor var_967_begin_0 = const()[name = string("op_967_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_967_end_0 = const()[name = string("op_967_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_967_end_mask_0 = const()[name = string("op_967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_967_cast_fp16 = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = obj_59_cast_fp16)[name = string("op_967_cast_fp16")]; + tensor var_970_begin_0 = const()[name = string("op_970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_end_0 = const()[name = string("op_970_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_970_end_mask_0 = const()[name = string("op_970_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_970_squeeze_mask_0 = const()[name = string("op_970_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, squeeze_mask = var_970_squeeze_mask_0, x = var_967_cast_fp16)[name = string("op_970_cast_fp16")]; + tensor var_985_begin_0 = const()[name = string("op_985_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_985_end_0 = const()[name = string("op_985_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_985_end_mask_0 = const()[name = string("op_985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = obj_77_cast_fp16)[name = string("op_985_cast_fp16")]; + tensor var_988_begin_0 = const()[name = string("op_988_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_988_end_0 = const()[name = string("op_988_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_988_end_mask_0 = const()[name = string("op_988_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_988_squeeze_mask_0 = const()[name = string("op_988_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_988_cast_fp16 = slice_by_index(begin = var_988_begin_0, end = var_988_end_0, end_mask = var_988_end_mask_0, squeeze_mask = var_988_squeeze_mask_0, x = var_985_cast_fp16)[name = string("op_988_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = string("op_1003_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1003_end_0 = const()[name = string("op_1003_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_1003_end_mask_0 = const()[name = string("op_1003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1003_cast_fp16")]; + tensor var_1006_begin_0 = const()[name = string("op_1006_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1006_end_0 = const()[name = string("op_1006_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1006_end_mask_0 = const()[name = string("op_1006_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1006_squeeze_mask_0 = const()[name = string("op_1006_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1006_cast_fp16 = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, squeeze_mask = var_1006_squeeze_mask_0, x = var_1003_cast_fp16)[name = string("op_1006_cast_fp16")]; + tensor var_1021_begin_0 = const()[name = string("op_1021_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1021_end_0 = const()[name = string("op_1021_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_1021_end_mask_0 = const()[name = string("op_1021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1021_cast_fp16 = slice_by_index(begin = var_1021_begin_0, end = var_1021_end_0, end_mask = var_1021_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1021_cast_fp16")]; + tensor var_1024_begin_0 = const()[name = string("op_1024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1024_end_0 = const()[name = string("op_1024_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1024_end_mask_0 = const()[name = string("op_1024_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1024_squeeze_mask_0 = const()[name = string("op_1024_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1024_cast_fp16 = slice_by_index(begin = var_1024_begin_0, end = var_1024_end_0, end_mask = var_1024_end_mask_0, squeeze_mask = var_1024_squeeze_mask_0, x = var_1021_cast_fp16)[name = string("op_1024_cast_fp16")]; + tensor var_1039_begin_0 = const()[name = string("op_1039_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1039_end_0 = const()[name = string("op_1039_end_0"), val = tensor([1, 15, 1, 1536])]; + tensor var_1039_end_mask_0 = const()[name = string("op_1039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1039_cast_fp16")]; + tensor var_1042_begin_0 = const()[name = string("op_1042_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1042_end_0 = const()[name = string("op_1042_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1042_end_mask_0 = const()[name = string("op_1042_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1042_squeeze_mask_0 = const()[name = string("op_1042_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1042_cast_fp16 = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, squeeze_mask = var_1042_squeeze_mask_0, x = var_1039_cast_fp16)[name = string("op_1042_cast_fp16")]; + int32 var_1049 = const()[name = string("op_1049"), val = int32(1)]; + bool var_1050_interleave_0 = const()[name = string("op_1050_interleave_0"), val = bool(false)]; + tensor var_1050_cast_fp16 = concat(axis = var_1049, interleave = var_1050_interleave_0, values = (var_952_cast_fp16, var_970_cast_fp16, var_988_cast_fp16, var_1006_cast_fp16, var_1024_cast_fp16, var_1042_cast_fp16))[name = string("op_1050_cast_fp16")]; + bool var_1053 = const()[name = string("op_1053"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1053, x = var_1050_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..ce17ffd9db34a87639997c3cd17fa5a72513d3eb --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9be9fd71781e6e440994496f915a5b21f02745788dfb0e6e84978c8f8c89bfc8 +size 317708340 diff --git a/openai_whisper-large-v3-v20240930_turbo/config.json b/openai_whisper-large-v3-v20240930_turbo/config.json new file mode 100644 index 0000000000000000000000000000000000000000..f6c3db01a3124f90b3bfbf3d58e47ba5de1f5ccd --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/config.json @@ -0,0 +1 @@ +{"_name_or_path": "/raid/yoach/tmp_whisper_turbo", "activation_dropout": 0.0, "activation_function": "gelu", "apply_spec_augment": false, "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50256], "bos_token_id": 50257, "classifier_proj_size": 256, "d_model": 1280, "decoder_attention_heads": 20, "decoder_ffn_dim": 5120, "decoder_layerdrop": 0.0, "decoder_layers": 4, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 20, "encoder_ffn_dim": 5120, "encoder_layerdrop": 0.0, "encoder_layers": 32, "eos_token_id": 50257, "init_std": 0.02, "is_encoder_decoder": true, "mask_feature_length": 10, "mask_feature_min_masks": 0, "mask_feature_prob": 0.0, "mask_time_length": 10, "mask_time_min_masks": 2, "mask_time_prob": 0.05, "max_source_positions": 1500, "max_target_positions": 448, "median_filter_width": 7, "model_type": "whisper", "num_hidden_layers": 32, "num_mel_bins": 128, "pad_token_id": 50257, "scale_embedding": false, "torch_dtype": "float16", "transformers_version": "4.46.0.dev0", "use_cache": true, "use_weighted_layer_sum": false, "vocab_size": 51866} \ No newline at end of file diff --git a/openai_whisper-large-v3-v20240930_turbo/generation_config.json b/openai_whisper-large-v3-v20240930_turbo/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..a023f6f947b0fc4ab7a2986a1c23b014836ec720 --- /dev/null +++ b/openai_whisper-large-v3-v20240930_turbo/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[2, 4], [2, 11], [3, 3], [3, 6], [3, 11], [3, 14]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50360]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|yue|>": 50358, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50364, "pad_token_id": 50257, "prev_sot_token_id": 50362, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50359, 50360, 50361, 50362, 50363], "task_to_id": {"transcribe": 50360, "translate": 50359}, "transformers_version": "4.46.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-small.en/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-small.en/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..f6ea167c2b8bc62d8606159c65bc2cd55c892f4f --- /dev/null +++ b/openai_whisper-small.en/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2178a11e8833a7745057ce38dce4588cb4bca380e3398b952761f12aee0a93cb +size 243 diff --git a/openai_whisper-small.en/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-small.en/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..4e7a2bb9e7ef3870c50986867ffd5a6788e5ed84 --- /dev/null +++ b/openai_whisper-small.en/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a078e65c9369ce8a4a687a2bbb0a8befbd4ed459250c0442176824906fa95ee1 +size 433 diff --git a/openai_whisper-small.en/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-small.en/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1587204f11c01ae4618a054b766c946375a44b18 --- /dev/null +++ b/openai_whisper-small.en/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 768 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 768, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 25, + "Ios18.conv" : 98, + "Ios18.gelu" : 14, + "Ios18.concat" : 158, + "Ios16.einsum" : 1152, + "Ios18.add" : 25, + "Ios18.softmax" : 576, + "Ios18.sliceByIndex" : 1008, + "Ios18.layerNorm" : 25, + "Ios18.transpose" : 12, + "Ios18.mul" : 576 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-small.en/AudioEncoder.mlmodelc/model.mil b/openai_whisper-small.en/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..9caa00c57cd12baeb3f79c05eac5d3c9f247585c --- /dev/null +++ b/openai_whisper-small.en/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,9435 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_90_pad_type_0 = const()[name = string("op_90_pad_type_0"), val = string("custom")]; + tensor var_90_pad_0 = const()[name = string("op_90_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_90_strides_0 = const()[name = string("op_90_strides_0"), val = tensor([1, 1])]; + tensor var_90_dilations_0 = const()[name = string("op_90_dilations_0"), val = tensor([1, 1])]; + int32 var_90_groups_0 = const()[name = string("op_90_groups_0"), val = int32(1)]; + tensor var_65_to_fp16 = const()[name = string("op_65_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_71_to_fp16 = const()[name = string("op_71_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368768)))]; + tensor var_90_cast_fp16 = conv(bias = var_71_to_fp16, dilations = var_90_dilations_0, groups = var_90_groups_0, pad = var_90_pad_0, pad_type = var_90_pad_type_0, strides = var_90_strides_0, weight = var_65_to_fp16, x = melspectrogram_features)[name = string("op_90_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_90_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_130_pad_type_0 = const()[name = string("op_130_pad_type_0"), val = string("custom")]; + tensor var_130_pad_0 = const()[name = string("op_130_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_130_strides_0 = const()[name = string("op_130_strides_0"), val = tensor([2, 2])]; + tensor var_130_dilations_0 = const()[name = string("op_130_dilations_0"), val = tensor([1, 1])]; + int32 var_130_groups_0 = const()[name = string("op_130_groups_0"), val = int32(1)]; + tensor var_105_to_fp16 = const()[name = string("op_105_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370368)))]; + tensor var_111_to_fp16 = const()[name = string("op_111_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3909376)))]; + tensor var_130_cast_fp16 = conv(bias = var_111_to_fp16, dilations = var_130_dilations_0, groups = var_130_groups_0, pad = var_130_pad_0, pad_type = var_130_pad_type_0, strides = var_130_strides_0, weight = var_105_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_130_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_130_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_148_to_fp16 = const()[name = string("op_148_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3910976)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_148_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_158 = const()[name = string("op_158"), val = int32(3)]; + int32 var_175 = const()[name = string("op_175"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_192_to_fp16 = const()[name = string("op_192_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_192_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6215040)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6216640)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6218240)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6219840)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6221440)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7401152)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7402752)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8582464)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9762176)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = string("op_230_cast_fp16")]; + tensor var_234_begin_0 = const()[name = string("op_234_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_234_end_0 = const()[name = string("op_234_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_234_end_mask_0 = const()[name = string("op_234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_234_cast_fp16 = slice_by_index(begin = var_234_begin_0, end = var_234_end_0, end_mask = var_234_end_mask_0, x = query_1_cast_fp16)[name = string("op_234_cast_fp16")]; + tensor var_238_begin_0 = const()[name = string("op_238_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_238_end_0 = const()[name = string("op_238_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_238_end_mask_0 = const()[name = string("op_238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_238_cast_fp16 = slice_by_index(begin = var_238_begin_0, end = var_238_end_0, end_mask = var_238_end_mask_0, x = query_1_cast_fp16)[name = string("op_238_cast_fp16")]; + tensor var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = query_1_cast_fp16)[name = string("op_242_cast_fp16")]; + tensor var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = query_1_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor var_250_begin_0 = const()[name = string("op_250_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_250_end_0 = const()[name = string("op_250_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_250_end_mask_0 = const()[name = string("op_250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_250_cast_fp16 = slice_by_index(begin = var_250_begin_0, end = var_250_end_0, end_mask = var_250_end_mask_0, x = query_1_cast_fp16)[name = string("op_250_cast_fp16")]; + tensor var_254_begin_0 = const()[name = string("op_254_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_254_end_0 = const()[name = string("op_254_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_254_end_mask_0 = const()[name = string("op_254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_254_cast_fp16 = slice_by_index(begin = var_254_begin_0, end = var_254_end_0, end_mask = var_254_end_mask_0, x = query_1_cast_fp16)[name = string("op_254_cast_fp16")]; + tensor var_258_begin_0 = const()[name = string("op_258_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_258_end_0 = const()[name = string("op_258_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_258_end_mask_0 = const()[name = string("op_258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_258_cast_fp16 = slice_by_index(begin = var_258_begin_0, end = var_258_end_0, end_mask = var_258_end_mask_0, x = query_1_cast_fp16)[name = string("op_258_cast_fp16")]; + tensor var_262_begin_0 = const()[name = string("op_262_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_262_end_0 = const()[name = string("op_262_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_262_end_mask_0 = const()[name = string("op_262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_262_cast_fp16 = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = query_1_cast_fp16)[name = string("op_262_cast_fp16")]; + tensor var_266_begin_0 = const()[name = string("op_266_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_266_end_0 = const()[name = string("op_266_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_266_end_mask_0 = const()[name = string("op_266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = query_1_cast_fp16)[name = string("op_266_cast_fp16")]; + tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = query_1_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = query_1_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_283_begin_0 = const()[name = string("op_283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_283_end_0 = const()[name = string("op_283_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_283_end_mask_0 = const()[name = string("op_283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_283_cast_fp16 = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_230_cast_fp16)[name = string("op_283_cast_fp16")]; + tensor var_290_begin_0 = const()[name = string("op_290_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_290_end_0 = const()[name = string("op_290_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_290_end_mask_0 = const()[name = string("op_290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_230_cast_fp16)[name = string("op_290_cast_fp16")]; + tensor var_297_begin_0 = const()[name = string("op_297_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_297_end_0 = const()[name = string("op_297_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_297_end_mask_0 = const()[name = string("op_297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_297_cast_fp16 = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_230_cast_fp16)[name = string("op_297_cast_fp16")]; + tensor var_304_begin_0 = const()[name = string("op_304_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_304_end_0 = const()[name = string("op_304_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_304_end_mask_0 = const()[name = string("op_304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_304_cast_fp16 = slice_by_index(begin = var_304_begin_0, end = var_304_end_0, end_mask = var_304_end_mask_0, x = var_230_cast_fp16)[name = string("op_304_cast_fp16")]; + tensor var_311_begin_0 = const()[name = string("op_311_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_311_end_0 = const()[name = string("op_311_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_311_end_mask_0 = const()[name = string("op_311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = var_234_cast_fp16)[name = string("op_311_cast_fp16")]; + tensor var_318_begin_0 = const()[name = string("op_318_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_318_end_0 = const()[name = string("op_318_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_318_end_mask_0 = const()[name = string("op_318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = var_234_cast_fp16)[name = string("op_318_cast_fp16")]; + tensor var_325_begin_0 = const()[name = string("op_325_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_325_end_0 = const()[name = string("op_325_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_325_end_mask_0 = const()[name = string("op_325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_325_cast_fp16 = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = var_234_cast_fp16)[name = string("op_325_cast_fp16")]; + tensor var_332_begin_0 = const()[name = string("op_332_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_332_end_0 = const()[name = string("op_332_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_332_end_mask_0 = const()[name = string("op_332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = var_234_cast_fp16)[name = string("op_332_cast_fp16")]; + tensor var_339_begin_0 = const()[name = string("op_339_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_339_end_0 = const()[name = string("op_339_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_339_end_mask_0 = const()[name = string("op_339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_339_cast_fp16 = slice_by_index(begin = var_339_begin_0, end = var_339_end_0, end_mask = var_339_end_mask_0, x = var_238_cast_fp16)[name = string("op_339_cast_fp16")]; + tensor var_346_begin_0 = const()[name = string("op_346_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_346_end_0 = const()[name = string("op_346_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_346_end_mask_0 = const()[name = string("op_346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_346_cast_fp16 = slice_by_index(begin = var_346_begin_0, end = var_346_end_0, end_mask = var_346_end_mask_0, x = var_238_cast_fp16)[name = string("op_346_cast_fp16")]; + tensor var_353_begin_0 = const()[name = string("op_353_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_353_end_0 = const()[name = string("op_353_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_353_end_mask_0 = const()[name = string("op_353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_353_cast_fp16 = slice_by_index(begin = var_353_begin_0, end = var_353_end_0, end_mask = var_353_end_mask_0, x = var_238_cast_fp16)[name = string("op_353_cast_fp16")]; + tensor var_360_begin_0 = const()[name = string("op_360_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_360_end_0 = const()[name = string("op_360_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_360_end_mask_0 = const()[name = string("op_360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_360_cast_fp16 = slice_by_index(begin = var_360_begin_0, end = var_360_end_0, end_mask = var_360_end_mask_0, x = var_238_cast_fp16)[name = string("op_360_cast_fp16")]; + tensor var_367_begin_0 = const()[name = string("op_367_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_367_end_0 = const()[name = string("op_367_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_367_end_mask_0 = const()[name = string("op_367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_367_cast_fp16 = slice_by_index(begin = var_367_begin_0, end = var_367_end_0, end_mask = var_367_end_mask_0, x = var_242_cast_fp16)[name = string("op_367_cast_fp16")]; + tensor var_374_begin_0 = const()[name = string("op_374_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_374_end_0 = const()[name = string("op_374_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_374_end_mask_0 = const()[name = string("op_374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_374_cast_fp16 = slice_by_index(begin = var_374_begin_0, end = var_374_end_0, end_mask = var_374_end_mask_0, x = var_242_cast_fp16)[name = string("op_374_cast_fp16")]; + tensor var_381_begin_0 = const()[name = string("op_381_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_381_end_0 = const()[name = string("op_381_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_381_end_mask_0 = const()[name = string("op_381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_381_cast_fp16 = slice_by_index(begin = var_381_begin_0, end = var_381_end_0, end_mask = var_381_end_mask_0, x = var_242_cast_fp16)[name = string("op_381_cast_fp16")]; + tensor var_388_begin_0 = const()[name = string("op_388_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_388_end_0 = const()[name = string("op_388_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_388_end_mask_0 = const()[name = string("op_388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_388_cast_fp16 = slice_by_index(begin = var_388_begin_0, end = var_388_end_0, end_mask = var_388_end_mask_0, x = var_242_cast_fp16)[name = string("op_388_cast_fp16")]; + tensor var_395_begin_0 = const()[name = string("op_395_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_395_end_0 = const()[name = string("op_395_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_395_end_mask_0 = const()[name = string("op_395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_395_cast_fp16 = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = var_246_cast_fp16)[name = string("op_395_cast_fp16")]; + tensor var_402_begin_0 = const()[name = string("op_402_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_402_end_0 = const()[name = string("op_402_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_402_end_mask_0 = const()[name = string("op_402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_402_cast_fp16 = slice_by_index(begin = var_402_begin_0, end = var_402_end_0, end_mask = var_402_end_mask_0, x = var_246_cast_fp16)[name = string("op_402_cast_fp16")]; + tensor var_409_begin_0 = const()[name = string("op_409_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_409_end_0 = const()[name = string("op_409_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_409_end_mask_0 = const()[name = string("op_409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_409_cast_fp16 = slice_by_index(begin = var_409_begin_0, end = var_409_end_0, end_mask = var_409_end_mask_0, x = var_246_cast_fp16)[name = string("op_409_cast_fp16")]; + tensor var_416_begin_0 = const()[name = string("op_416_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_416_end_0 = const()[name = string("op_416_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_416_end_mask_0 = const()[name = string("op_416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_416_cast_fp16 = slice_by_index(begin = var_416_begin_0, end = var_416_end_0, end_mask = var_416_end_mask_0, x = var_246_cast_fp16)[name = string("op_416_cast_fp16")]; + tensor var_423_begin_0 = const()[name = string("op_423_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_423_end_0 = const()[name = string("op_423_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_423_end_mask_0 = const()[name = string("op_423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_423_cast_fp16 = slice_by_index(begin = var_423_begin_0, end = var_423_end_0, end_mask = var_423_end_mask_0, x = var_250_cast_fp16)[name = string("op_423_cast_fp16")]; + tensor var_430_begin_0 = const()[name = string("op_430_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_430_end_0 = const()[name = string("op_430_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_430_end_mask_0 = const()[name = string("op_430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_430_cast_fp16 = slice_by_index(begin = var_430_begin_0, end = var_430_end_0, end_mask = var_430_end_mask_0, x = var_250_cast_fp16)[name = string("op_430_cast_fp16")]; + tensor var_437_begin_0 = const()[name = string("op_437_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_437_end_0 = const()[name = string("op_437_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_437_end_mask_0 = const()[name = string("op_437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_437_cast_fp16 = slice_by_index(begin = var_437_begin_0, end = var_437_end_0, end_mask = var_437_end_mask_0, x = var_250_cast_fp16)[name = string("op_437_cast_fp16")]; + tensor var_444_begin_0 = const()[name = string("op_444_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_444_end_0 = const()[name = string("op_444_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_444_end_mask_0 = const()[name = string("op_444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_444_cast_fp16 = slice_by_index(begin = var_444_begin_0, end = var_444_end_0, end_mask = var_444_end_mask_0, x = var_250_cast_fp16)[name = string("op_444_cast_fp16")]; + tensor var_451_begin_0 = const()[name = string("op_451_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_451_end_0 = const()[name = string("op_451_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_451_end_mask_0 = const()[name = string("op_451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_451_cast_fp16 = slice_by_index(begin = var_451_begin_0, end = var_451_end_0, end_mask = var_451_end_mask_0, x = var_254_cast_fp16)[name = string("op_451_cast_fp16")]; + tensor var_458_begin_0 = const()[name = string("op_458_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_458_end_0 = const()[name = string("op_458_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_458_end_mask_0 = const()[name = string("op_458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_458_cast_fp16 = slice_by_index(begin = var_458_begin_0, end = var_458_end_0, end_mask = var_458_end_mask_0, x = var_254_cast_fp16)[name = string("op_458_cast_fp16")]; + tensor var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = var_254_cast_fp16)[name = string("op_465_cast_fp16")]; + tensor var_472_begin_0 = const()[name = string("op_472_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_472_end_0 = const()[name = string("op_472_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_472_end_mask_0 = const()[name = string("op_472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_472_cast_fp16 = slice_by_index(begin = var_472_begin_0, end = var_472_end_0, end_mask = var_472_end_mask_0, x = var_254_cast_fp16)[name = string("op_472_cast_fp16")]; + tensor var_479_begin_0 = const()[name = string("op_479_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_479_end_0 = const()[name = string("op_479_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_479_end_mask_0 = const()[name = string("op_479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_479_cast_fp16 = slice_by_index(begin = var_479_begin_0, end = var_479_end_0, end_mask = var_479_end_mask_0, x = var_258_cast_fp16)[name = string("op_479_cast_fp16")]; + tensor var_486_begin_0 = const()[name = string("op_486_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_486_end_0 = const()[name = string("op_486_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_486_end_mask_0 = const()[name = string("op_486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_486_cast_fp16 = slice_by_index(begin = var_486_begin_0, end = var_486_end_0, end_mask = var_486_end_mask_0, x = var_258_cast_fp16)[name = string("op_486_cast_fp16")]; + tensor var_493_begin_0 = const()[name = string("op_493_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_493_end_0 = const()[name = string("op_493_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_493_end_mask_0 = const()[name = string("op_493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_493_cast_fp16 = slice_by_index(begin = var_493_begin_0, end = var_493_end_0, end_mask = var_493_end_mask_0, x = var_258_cast_fp16)[name = string("op_493_cast_fp16")]; + tensor var_500_begin_0 = const()[name = string("op_500_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_500_end_0 = const()[name = string("op_500_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_500_end_mask_0 = const()[name = string("op_500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_500_cast_fp16 = slice_by_index(begin = var_500_begin_0, end = var_500_end_0, end_mask = var_500_end_mask_0, x = var_258_cast_fp16)[name = string("op_500_cast_fp16")]; + tensor var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = var_262_cast_fp16)[name = string("op_507_cast_fp16")]; + tensor var_514_begin_0 = const()[name = string("op_514_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_514_end_0 = const()[name = string("op_514_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_514_end_mask_0 = const()[name = string("op_514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_514_cast_fp16 = slice_by_index(begin = var_514_begin_0, end = var_514_end_0, end_mask = var_514_end_mask_0, x = var_262_cast_fp16)[name = string("op_514_cast_fp16")]; + tensor var_521_begin_0 = const()[name = string("op_521_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_521_end_0 = const()[name = string("op_521_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_521_end_mask_0 = const()[name = string("op_521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_521_cast_fp16 = slice_by_index(begin = var_521_begin_0, end = var_521_end_0, end_mask = var_521_end_mask_0, x = var_262_cast_fp16)[name = string("op_521_cast_fp16")]; + tensor var_528_begin_0 = const()[name = string("op_528_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_528_end_0 = const()[name = string("op_528_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_528_end_mask_0 = const()[name = string("op_528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_528_cast_fp16 = slice_by_index(begin = var_528_begin_0, end = var_528_end_0, end_mask = var_528_end_mask_0, x = var_262_cast_fp16)[name = string("op_528_cast_fp16")]; + tensor var_535_begin_0 = const()[name = string("op_535_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_535_end_0 = const()[name = string("op_535_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_535_end_mask_0 = const()[name = string("op_535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_535_cast_fp16 = slice_by_index(begin = var_535_begin_0, end = var_535_end_0, end_mask = var_535_end_mask_0, x = var_266_cast_fp16)[name = string("op_535_cast_fp16")]; + tensor var_542_begin_0 = const()[name = string("op_542_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_542_end_0 = const()[name = string("op_542_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_542_end_mask_0 = const()[name = string("op_542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_542_cast_fp16 = slice_by_index(begin = var_542_begin_0, end = var_542_end_0, end_mask = var_542_end_mask_0, x = var_266_cast_fp16)[name = string("op_542_cast_fp16")]; + tensor var_549_begin_0 = const()[name = string("op_549_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_549_end_0 = const()[name = string("op_549_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_549_end_mask_0 = const()[name = string("op_549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_549_cast_fp16 = slice_by_index(begin = var_549_begin_0, end = var_549_end_0, end_mask = var_549_end_mask_0, x = var_266_cast_fp16)[name = string("op_549_cast_fp16")]; + tensor var_556_begin_0 = const()[name = string("op_556_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_556_end_0 = const()[name = string("op_556_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_556_end_mask_0 = const()[name = string("op_556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_556_cast_fp16 = slice_by_index(begin = var_556_begin_0, end = var_556_end_0, end_mask = var_556_end_mask_0, x = var_266_cast_fp16)[name = string("op_556_cast_fp16")]; + tensor var_563_begin_0 = const()[name = string("op_563_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_563_end_0 = const()[name = string("op_563_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_563_end_mask_0 = const()[name = string("op_563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_563_cast_fp16 = slice_by_index(begin = var_563_begin_0, end = var_563_end_0, end_mask = var_563_end_mask_0, x = var_270_cast_fp16)[name = string("op_563_cast_fp16")]; + tensor var_570_begin_0 = const()[name = string("op_570_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_570_end_0 = const()[name = string("op_570_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_570_end_mask_0 = const()[name = string("op_570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_570_cast_fp16 = slice_by_index(begin = var_570_begin_0, end = var_570_end_0, end_mask = var_570_end_mask_0, x = var_270_cast_fp16)[name = string("op_570_cast_fp16")]; + tensor var_577_begin_0 = const()[name = string("op_577_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_577_end_0 = const()[name = string("op_577_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_577_end_mask_0 = const()[name = string("op_577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_577_cast_fp16 = slice_by_index(begin = var_577_begin_0, end = var_577_end_0, end_mask = var_577_end_mask_0, x = var_270_cast_fp16)[name = string("op_577_cast_fp16")]; + tensor var_584_begin_0 = const()[name = string("op_584_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_584_end_0 = const()[name = string("op_584_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_584_end_mask_0 = const()[name = string("op_584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_584_cast_fp16 = slice_by_index(begin = var_584_begin_0, end = var_584_end_0, end_mask = var_584_end_mask_0, x = var_270_cast_fp16)[name = string("op_584_cast_fp16")]; + tensor var_591_begin_0 = const()[name = string("op_591_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_591_end_0 = const()[name = string("op_591_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_591_end_mask_0 = const()[name = string("op_591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_591_cast_fp16 = slice_by_index(begin = var_591_begin_0, end = var_591_end_0, end_mask = var_591_end_mask_0, x = var_274_cast_fp16)[name = string("op_591_cast_fp16")]; + tensor var_598_begin_0 = const()[name = string("op_598_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_598_end_0 = const()[name = string("op_598_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_598_end_mask_0 = const()[name = string("op_598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_598_cast_fp16 = slice_by_index(begin = var_598_begin_0, end = var_598_end_0, end_mask = var_598_end_mask_0, x = var_274_cast_fp16)[name = string("op_598_cast_fp16")]; + tensor var_605_begin_0 = const()[name = string("op_605_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_605_end_0 = const()[name = string("op_605_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_605_end_mask_0 = const()[name = string("op_605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_605_cast_fp16 = slice_by_index(begin = var_605_begin_0, end = var_605_end_0, end_mask = var_605_end_mask_0, x = var_274_cast_fp16)[name = string("op_605_cast_fp16")]; + tensor var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, x = var_274_cast_fp16)[name = string("op_612_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_617_begin_0 = const()[name = string("op_617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_617_end_0 = const()[name = string("op_617_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_617_end_mask_0 = const()[name = string("op_617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_11")]; + tensor var_617_cast_fp16 = slice_by_index(begin = var_617_begin_0, end = var_617_end_0, end_mask = var_617_end_mask_0, x = k_1_cast_fp16)[name = string("op_617_cast_fp16")]; + tensor var_621_begin_0 = const()[name = string("op_621_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_621_end_0 = const()[name = string("op_621_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_621_end_mask_0 = const()[name = string("op_621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_621_cast_fp16 = slice_by_index(begin = var_621_begin_0, end = var_621_end_0, end_mask = var_621_end_mask_0, x = k_1_cast_fp16)[name = string("op_621_cast_fp16")]; + tensor var_625_begin_0 = const()[name = string("op_625_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_625_end_0 = const()[name = string("op_625_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_625_end_mask_0 = const()[name = string("op_625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_625_cast_fp16 = slice_by_index(begin = var_625_begin_0, end = var_625_end_0, end_mask = var_625_end_mask_0, x = k_1_cast_fp16)[name = string("op_625_cast_fp16")]; + tensor var_629_begin_0 = const()[name = string("op_629_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_629_end_0 = const()[name = string("op_629_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_629_end_mask_0 = const()[name = string("op_629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_629_cast_fp16 = slice_by_index(begin = var_629_begin_0, end = var_629_end_0, end_mask = var_629_end_mask_0, x = k_1_cast_fp16)[name = string("op_629_cast_fp16")]; + tensor var_633_begin_0 = const()[name = string("op_633_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_633_end_0 = const()[name = string("op_633_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_633_end_mask_0 = const()[name = string("op_633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_633_cast_fp16 = slice_by_index(begin = var_633_begin_0, end = var_633_end_0, end_mask = var_633_end_mask_0, x = k_1_cast_fp16)[name = string("op_633_cast_fp16")]; + tensor var_637_begin_0 = const()[name = string("op_637_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_637_end_0 = const()[name = string("op_637_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_637_end_mask_0 = const()[name = string("op_637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_637_cast_fp16 = slice_by_index(begin = var_637_begin_0, end = var_637_end_0, end_mask = var_637_end_mask_0, x = k_1_cast_fp16)[name = string("op_637_cast_fp16")]; + tensor var_641_begin_0 = const()[name = string("op_641_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_641_end_0 = const()[name = string("op_641_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_641_end_mask_0 = const()[name = string("op_641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_641_cast_fp16 = slice_by_index(begin = var_641_begin_0, end = var_641_end_0, end_mask = var_641_end_mask_0, x = k_1_cast_fp16)[name = string("op_641_cast_fp16")]; + tensor var_645_begin_0 = const()[name = string("op_645_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_645_end_0 = const()[name = string("op_645_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_645_end_mask_0 = const()[name = string("op_645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = k_1_cast_fp16)[name = string("op_645_cast_fp16")]; + tensor var_649_begin_0 = const()[name = string("op_649_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_649_end_0 = const()[name = string("op_649_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_649_end_mask_0 = const()[name = string("op_649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_649_cast_fp16 = slice_by_index(begin = var_649_begin_0, end = var_649_end_0, end_mask = var_649_end_mask_0, x = k_1_cast_fp16)[name = string("op_649_cast_fp16")]; + tensor var_653_begin_0 = const()[name = string("op_653_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_653_end_0 = const()[name = string("op_653_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_653_end_mask_0 = const()[name = string("op_653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_653_cast_fp16 = slice_by_index(begin = var_653_begin_0, end = var_653_end_0, end_mask = var_653_end_mask_0, x = k_1_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor var_657_begin_0 = const()[name = string("op_657_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_657_end_0 = const()[name = string("op_657_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_657_end_mask_0 = const()[name = string("op_657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_657_cast_fp16 = slice_by_index(begin = var_657_begin_0, end = var_657_end_0, end_mask = var_657_end_mask_0, x = k_1_cast_fp16)[name = string("op_657_cast_fp16")]; + tensor var_661_begin_0 = const()[name = string("op_661_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_661_end_0 = const()[name = string("op_661_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_661_end_mask_0 = const()[name = string("op_661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_661_cast_fp16 = slice_by_index(begin = var_661_begin_0, end = var_661_end_0, end_mask = var_661_end_mask_0, x = k_1_cast_fp16)[name = string("op_661_cast_fp16")]; + tensor var_663_begin_0 = const()[name = string("op_663_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_663_end_0 = const()[name = string("op_663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_663_end_mask_0 = const()[name = string("op_663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = value_1_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_667_begin_0 = const()[name = string("op_667_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_667_end_0 = const()[name = string("op_667_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_667_end_mask_0 = const()[name = string("op_667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_667_cast_fp16 = slice_by_index(begin = var_667_begin_0, end = var_667_end_0, end_mask = var_667_end_mask_0, x = value_1_cast_fp16)[name = string("op_667_cast_fp16")]; + tensor var_671_begin_0 = const()[name = string("op_671_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_671_end_0 = const()[name = string("op_671_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_671_end_mask_0 = const()[name = string("op_671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_671_cast_fp16 = slice_by_index(begin = var_671_begin_0, end = var_671_end_0, end_mask = var_671_end_mask_0, x = value_1_cast_fp16)[name = string("op_671_cast_fp16")]; + tensor var_675_begin_0 = const()[name = string("op_675_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_675_end_0 = const()[name = string("op_675_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_675_end_mask_0 = const()[name = string("op_675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_675_cast_fp16 = slice_by_index(begin = var_675_begin_0, end = var_675_end_0, end_mask = var_675_end_mask_0, x = value_1_cast_fp16)[name = string("op_675_cast_fp16")]; + tensor var_679_begin_0 = const()[name = string("op_679_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_679_end_0 = const()[name = string("op_679_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_679_end_mask_0 = const()[name = string("op_679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_679_cast_fp16 = slice_by_index(begin = var_679_begin_0, end = var_679_end_0, end_mask = var_679_end_mask_0, x = value_1_cast_fp16)[name = string("op_679_cast_fp16")]; + tensor var_683_begin_0 = const()[name = string("op_683_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_683_end_0 = const()[name = string("op_683_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_683_end_mask_0 = const()[name = string("op_683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_683_cast_fp16 = slice_by_index(begin = var_683_begin_0, end = var_683_end_0, end_mask = var_683_end_mask_0, x = value_1_cast_fp16)[name = string("op_683_cast_fp16")]; + tensor var_687_begin_0 = const()[name = string("op_687_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_687_end_0 = const()[name = string("op_687_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_687_end_mask_0 = const()[name = string("op_687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_687_cast_fp16 = slice_by_index(begin = var_687_begin_0, end = var_687_end_0, end_mask = var_687_end_mask_0, x = value_1_cast_fp16)[name = string("op_687_cast_fp16")]; + tensor var_691_begin_0 = const()[name = string("op_691_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_691_end_0 = const()[name = string("op_691_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_691_end_mask_0 = const()[name = string("op_691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_691_cast_fp16 = slice_by_index(begin = var_691_begin_0, end = var_691_end_0, end_mask = var_691_end_mask_0, x = value_1_cast_fp16)[name = string("op_691_cast_fp16")]; + tensor var_695_begin_0 = const()[name = string("op_695_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_695_end_0 = const()[name = string("op_695_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_695_end_mask_0 = const()[name = string("op_695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_695_cast_fp16 = slice_by_index(begin = var_695_begin_0, end = var_695_end_0, end_mask = var_695_end_mask_0, x = value_1_cast_fp16)[name = string("op_695_cast_fp16")]; + tensor var_699_begin_0 = const()[name = string("op_699_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_699_end_0 = const()[name = string("op_699_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_699_end_mask_0 = const()[name = string("op_699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_699_cast_fp16 = slice_by_index(begin = var_699_begin_0, end = var_699_end_0, end_mask = var_699_end_mask_0, x = value_1_cast_fp16)[name = string("op_699_cast_fp16")]; + tensor var_703_begin_0 = const()[name = string("op_703_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_703_end_0 = const()[name = string("op_703_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_703_end_mask_0 = const()[name = string("op_703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_703_cast_fp16 = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = value_1_cast_fp16)[name = string("op_703_cast_fp16")]; + tensor var_707_begin_0 = const()[name = string("op_707_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_707_end_0 = const()[name = string("op_707_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_707_end_mask_0 = const()[name = string("op_707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = value_1_cast_fp16)[name = string("op_707_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_617_cast_fp16, var_283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_617_cast_fp16, var_290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_617_cast_fp16, var_297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_617_cast_fp16, var_304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_621_cast_fp16, var_311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_621_cast_fp16, var_318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_621_cast_fp16, var_325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_621_cast_fp16, var_332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_625_cast_fp16, var_339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_625_cast_fp16, var_346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_625_cast_fp16, var_353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_625_cast_fp16, var_360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_629_cast_fp16, var_367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_629_cast_fp16, var_374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_629_cast_fp16, var_381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_629_cast_fp16, var_388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_633_cast_fp16, var_395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_633_cast_fp16, var_402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_633_cast_fp16, var_409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_633_cast_fp16, var_416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_637_cast_fp16, var_423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_637_cast_fp16, var_430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_637_cast_fp16, var_437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_637_cast_fp16, var_444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_641_cast_fp16, var_451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_641_cast_fp16, var_458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_641_cast_fp16, var_465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_641_cast_fp16, var_472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_645_cast_fp16, var_479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_645_cast_fp16, var_486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_645_cast_fp16, var_493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_645_cast_fp16, var_500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_649_cast_fp16, var_507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_649_cast_fp16, var_514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_649_cast_fp16, var_521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_649_cast_fp16, var_528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_653_cast_fp16, var_535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_653_cast_fp16, var_542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_653_cast_fp16, var_549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_653_cast_fp16, var_556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_657_cast_fp16, var_563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_657_cast_fp16, var_570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_657_cast_fp16, var_577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_657_cast_fp16, var_584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_661_cast_fp16, var_591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_661_cast_fp16, var_598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_661_cast_fp16, var_605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_661_cast_fp16, var_612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + fp16 var_806_to_fp16 = const()[name = string("op_806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_806_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_808_to_fp16 = const()[name = string("op_808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_808_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_810_to_fp16 = const()[name = string("op_810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_810_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_812_to_fp16 = const()[name = string("op_812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_812_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_814_to_fp16 = const()[name = string("op_814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_814_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_816_to_fp16 = const()[name = string("op_816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_816_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_818_to_fp16 = const()[name = string("op_818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_818_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_820_to_fp16 = const()[name = string("op_820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_820_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_822_to_fp16 = const()[name = string("op_822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_822_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_824_to_fp16 = const()[name = string("op_824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_824_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_826_to_fp16 = const()[name = string("op_826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_826_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_828_to_fp16 = const()[name = string("op_828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_828_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_830_to_fp16 = const()[name = string("op_830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_830_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_832_to_fp16 = const()[name = string("op_832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_832_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_834_to_fp16 = const()[name = string("op_834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_834_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_836_to_fp16 = const()[name = string("op_836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_836_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_838_to_fp16 = const()[name = string("op_838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_838_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_840_to_fp16 = const()[name = string("op_840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_840_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_842_to_fp16 = const()[name = string("op_842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_842_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_844_to_fp16 = const()[name = string("op_844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_844_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_846_to_fp16 = const()[name = string("op_846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_846_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_848_to_fp16 = const()[name = string("op_848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_848_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_850_to_fp16 = const()[name = string("op_850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_850_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_852_to_fp16 = const()[name = string("op_852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_852_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_854_to_fp16 = const()[name = string("op_854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_854_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_856_to_fp16 = const()[name = string("op_856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_856_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_858_to_fp16 = const()[name = string("op_858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_858_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_860_to_fp16 = const()[name = string("op_860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_860_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_862_to_fp16 = const()[name = string("op_862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_862_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_864_to_fp16 = const()[name = string("op_864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_864_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_866_to_fp16 = const()[name = string("op_866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_866_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_868_to_fp16 = const()[name = string("op_868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_868_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + fp16 var_870_to_fp16 = const()[name = string("op_870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_870_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_872_to_fp16 = const()[name = string("op_872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_872_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_874_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_876_to_fp16 = const()[name = string("op_876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_876_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_878_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_880_to_fp16 = const()[name = string("op_880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_880_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_882_to_fp16 = const()[name = string("op_882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_882_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_884_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_886_to_fp16 = const()[name = string("op_886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_886_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_888_to_fp16 = const()[name = string("op_888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_888_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_890_to_fp16 = const()[name = string("op_890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_890_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_892_to_fp16 = const()[name = string("op_892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_892_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_894_to_fp16 = const()[name = string("op_894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_894_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_896_to_fp16 = const()[name = string("op_896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_896_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_898_to_fp16 = const()[name = string("op_898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_898_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_900_to_fp16 = const()[name = string("op_900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_900_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + tensor var_902_cast_fp16 = softmax(axis = var_175, x = aw_chunk_1_cast_fp16)[name = string("op_902_cast_fp16")]; + tensor var_903_cast_fp16 = softmax(axis = var_175, x = aw_chunk_3_cast_fp16)[name = string("op_903_cast_fp16")]; + tensor var_904_cast_fp16 = softmax(axis = var_175, x = aw_chunk_5_cast_fp16)[name = string("op_904_cast_fp16")]; + tensor var_905_cast_fp16 = softmax(axis = var_175, x = aw_chunk_7_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_906_cast_fp16 = softmax(axis = var_175, x = aw_chunk_9_cast_fp16)[name = string("op_906_cast_fp16")]; + tensor var_907_cast_fp16 = softmax(axis = var_175, x = aw_chunk_11_cast_fp16)[name = string("op_907_cast_fp16")]; + tensor var_908_cast_fp16 = softmax(axis = var_175, x = aw_chunk_13_cast_fp16)[name = string("op_908_cast_fp16")]; + tensor var_909_cast_fp16 = softmax(axis = var_175, x = aw_chunk_15_cast_fp16)[name = string("op_909_cast_fp16")]; + tensor var_910_cast_fp16 = softmax(axis = var_175, x = aw_chunk_17_cast_fp16)[name = string("op_910_cast_fp16")]; + tensor var_911_cast_fp16 = softmax(axis = var_175, x = aw_chunk_19_cast_fp16)[name = string("op_911_cast_fp16")]; + tensor var_912_cast_fp16 = softmax(axis = var_175, x = aw_chunk_21_cast_fp16)[name = string("op_912_cast_fp16")]; + tensor var_913_cast_fp16 = softmax(axis = var_175, x = aw_chunk_23_cast_fp16)[name = string("op_913_cast_fp16")]; + tensor var_914_cast_fp16 = softmax(axis = var_175, x = aw_chunk_25_cast_fp16)[name = string("op_914_cast_fp16")]; + tensor var_915_cast_fp16 = softmax(axis = var_175, x = aw_chunk_27_cast_fp16)[name = string("op_915_cast_fp16")]; + tensor var_916_cast_fp16 = softmax(axis = var_175, x = aw_chunk_29_cast_fp16)[name = string("op_916_cast_fp16")]; + tensor var_917_cast_fp16 = softmax(axis = var_175, x = aw_chunk_31_cast_fp16)[name = string("op_917_cast_fp16")]; + tensor var_918_cast_fp16 = softmax(axis = var_175, x = aw_chunk_33_cast_fp16)[name = string("op_918_cast_fp16")]; + tensor var_919_cast_fp16 = softmax(axis = var_175, x = aw_chunk_35_cast_fp16)[name = string("op_919_cast_fp16")]; + tensor var_920_cast_fp16 = softmax(axis = var_175, x = aw_chunk_37_cast_fp16)[name = string("op_920_cast_fp16")]; + tensor var_921_cast_fp16 = softmax(axis = var_175, x = aw_chunk_39_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_922_cast_fp16 = softmax(axis = var_175, x = aw_chunk_41_cast_fp16)[name = string("op_922_cast_fp16")]; + tensor var_923_cast_fp16 = softmax(axis = var_175, x = aw_chunk_43_cast_fp16)[name = string("op_923_cast_fp16")]; + tensor var_924_cast_fp16 = softmax(axis = var_175, x = aw_chunk_45_cast_fp16)[name = string("op_924_cast_fp16")]; + tensor var_925_cast_fp16 = softmax(axis = var_175, x = aw_chunk_47_cast_fp16)[name = string("op_925_cast_fp16")]; + tensor var_926_cast_fp16 = softmax(axis = var_175, x = aw_chunk_49_cast_fp16)[name = string("op_926_cast_fp16")]; + tensor var_927_cast_fp16 = softmax(axis = var_175, x = aw_chunk_51_cast_fp16)[name = string("op_927_cast_fp16")]; + tensor var_928_cast_fp16 = softmax(axis = var_175, x = aw_chunk_53_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_929_cast_fp16 = softmax(axis = var_175, x = aw_chunk_55_cast_fp16)[name = string("op_929_cast_fp16")]; + tensor var_930_cast_fp16 = softmax(axis = var_175, x = aw_chunk_57_cast_fp16)[name = string("op_930_cast_fp16")]; + tensor var_931_cast_fp16 = softmax(axis = var_175, x = aw_chunk_59_cast_fp16)[name = string("op_931_cast_fp16")]; + tensor var_932_cast_fp16 = softmax(axis = var_175, x = aw_chunk_61_cast_fp16)[name = string("op_932_cast_fp16")]; + tensor var_933_cast_fp16 = softmax(axis = var_175, x = aw_chunk_63_cast_fp16)[name = string("op_933_cast_fp16")]; + tensor var_934_cast_fp16 = softmax(axis = var_175, x = aw_chunk_65_cast_fp16)[name = string("op_934_cast_fp16")]; + tensor var_935_cast_fp16 = softmax(axis = var_175, x = aw_chunk_67_cast_fp16)[name = string("op_935_cast_fp16")]; + tensor var_936_cast_fp16 = softmax(axis = var_175, x = aw_chunk_69_cast_fp16)[name = string("op_936_cast_fp16")]; + tensor var_937_cast_fp16 = softmax(axis = var_175, x = aw_chunk_71_cast_fp16)[name = string("op_937_cast_fp16")]; + tensor var_938_cast_fp16 = softmax(axis = var_175, x = aw_chunk_73_cast_fp16)[name = string("op_938_cast_fp16")]; + tensor var_939_cast_fp16 = softmax(axis = var_175, x = aw_chunk_75_cast_fp16)[name = string("op_939_cast_fp16")]; + tensor var_940_cast_fp16 = softmax(axis = var_175, x = aw_chunk_77_cast_fp16)[name = string("op_940_cast_fp16")]; + tensor var_941_cast_fp16 = softmax(axis = var_175, x = aw_chunk_79_cast_fp16)[name = string("op_941_cast_fp16")]; + tensor var_942_cast_fp16 = softmax(axis = var_175, x = aw_chunk_81_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_943_cast_fp16 = softmax(axis = var_175, x = aw_chunk_83_cast_fp16)[name = string("op_943_cast_fp16")]; + tensor var_944_cast_fp16 = softmax(axis = var_175, x = aw_chunk_85_cast_fp16)[name = string("op_944_cast_fp16")]; + tensor var_945_cast_fp16 = softmax(axis = var_175, x = aw_chunk_87_cast_fp16)[name = string("op_945_cast_fp16")]; + tensor var_946_cast_fp16 = softmax(axis = var_175, x = aw_chunk_89_cast_fp16)[name = string("op_946_cast_fp16")]; + tensor var_947_cast_fp16 = softmax(axis = var_175, x = aw_chunk_91_cast_fp16)[name = string("op_947_cast_fp16")]; + tensor var_948_cast_fp16 = softmax(axis = var_175, x = aw_chunk_93_cast_fp16)[name = string("op_948_cast_fp16")]; + tensor var_949_cast_fp16 = softmax(axis = var_175, x = aw_chunk_95_cast_fp16)[name = string("op_949_cast_fp16")]; + string var_951_equation_0 = const()[name = string("op_951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_951_cast_fp16 = einsum(equation = var_951_equation_0, values = (var_663_cast_fp16, var_902_cast_fp16))[name = string("op_951_cast_fp16")]; + string var_953_equation_0 = const()[name = string("op_953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_953_cast_fp16 = einsum(equation = var_953_equation_0, values = (var_663_cast_fp16, var_903_cast_fp16))[name = string("op_953_cast_fp16")]; + string var_955_equation_0 = const()[name = string("op_955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_955_cast_fp16 = einsum(equation = var_955_equation_0, values = (var_663_cast_fp16, var_904_cast_fp16))[name = string("op_955_cast_fp16")]; + string var_957_equation_0 = const()[name = string("op_957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_957_cast_fp16 = einsum(equation = var_957_equation_0, values = (var_663_cast_fp16, var_905_cast_fp16))[name = string("op_957_cast_fp16")]; + string var_959_equation_0 = const()[name = string("op_959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_959_cast_fp16 = einsum(equation = var_959_equation_0, values = (var_667_cast_fp16, var_906_cast_fp16))[name = string("op_959_cast_fp16")]; + string var_961_equation_0 = const()[name = string("op_961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_961_cast_fp16 = einsum(equation = var_961_equation_0, values = (var_667_cast_fp16, var_907_cast_fp16))[name = string("op_961_cast_fp16")]; + string var_963_equation_0 = const()[name = string("op_963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_963_cast_fp16 = einsum(equation = var_963_equation_0, values = (var_667_cast_fp16, var_908_cast_fp16))[name = string("op_963_cast_fp16")]; + string var_965_equation_0 = const()[name = string("op_965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_965_cast_fp16 = einsum(equation = var_965_equation_0, values = (var_667_cast_fp16, var_909_cast_fp16))[name = string("op_965_cast_fp16")]; + string var_967_equation_0 = const()[name = string("op_967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_967_cast_fp16 = einsum(equation = var_967_equation_0, values = (var_671_cast_fp16, var_910_cast_fp16))[name = string("op_967_cast_fp16")]; + string var_969_equation_0 = const()[name = string("op_969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_969_cast_fp16 = einsum(equation = var_969_equation_0, values = (var_671_cast_fp16, var_911_cast_fp16))[name = string("op_969_cast_fp16")]; + string var_971_equation_0 = const()[name = string("op_971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_971_cast_fp16 = einsum(equation = var_971_equation_0, values = (var_671_cast_fp16, var_912_cast_fp16))[name = string("op_971_cast_fp16")]; + string var_973_equation_0 = const()[name = string("op_973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_973_cast_fp16 = einsum(equation = var_973_equation_0, values = (var_671_cast_fp16, var_913_cast_fp16))[name = string("op_973_cast_fp16")]; + string var_975_equation_0 = const()[name = string("op_975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_975_cast_fp16 = einsum(equation = var_975_equation_0, values = (var_675_cast_fp16, var_914_cast_fp16))[name = string("op_975_cast_fp16")]; + string var_977_equation_0 = const()[name = string("op_977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_977_cast_fp16 = einsum(equation = var_977_equation_0, values = (var_675_cast_fp16, var_915_cast_fp16))[name = string("op_977_cast_fp16")]; + string var_979_equation_0 = const()[name = string("op_979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_979_cast_fp16 = einsum(equation = var_979_equation_0, values = (var_675_cast_fp16, var_916_cast_fp16))[name = string("op_979_cast_fp16")]; + string var_981_equation_0 = const()[name = string("op_981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_981_cast_fp16 = einsum(equation = var_981_equation_0, values = (var_675_cast_fp16, var_917_cast_fp16))[name = string("op_981_cast_fp16")]; + string var_983_equation_0 = const()[name = string("op_983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_983_cast_fp16 = einsum(equation = var_983_equation_0, values = (var_679_cast_fp16, var_918_cast_fp16))[name = string("op_983_cast_fp16")]; + string var_985_equation_0 = const()[name = string("op_985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_985_cast_fp16 = einsum(equation = var_985_equation_0, values = (var_679_cast_fp16, var_919_cast_fp16))[name = string("op_985_cast_fp16")]; + string var_987_equation_0 = const()[name = string("op_987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_987_cast_fp16 = einsum(equation = var_987_equation_0, values = (var_679_cast_fp16, var_920_cast_fp16))[name = string("op_987_cast_fp16")]; + string var_989_equation_0 = const()[name = string("op_989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_989_cast_fp16 = einsum(equation = var_989_equation_0, values = (var_679_cast_fp16, var_921_cast_fp16))[name = string("op_989_cast_fp16")]; + string var_991_equation_0 = const()[name = string("op_991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_991_cast_fp16 = einsum(equation = var_991_equation_0, values = (var_683_cast_fp16, var_922_cast_fp16))[name = string("op_991_cast_fp16")]; + string var_993_equation_0 = const()[name = string("op_993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_993_cast_fp16 = einsum(equation = var_993_equation_0, values = (var_683_cast_fp16, var_923_cast_fp16))[name = string("op_993_cast_fp16")]; + string var_995_equation_0 = const()[name = string("op_995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_995_cast_fp16 = einsum(equation = var_995_equation_0, values = (var_683_cast_fp16, var_924_cast_fp16))[name = string("op_995_cast_fp16")]; + string var_997_equation_0 = const()[name = string("op_997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_997_cast_fp16 = einsum(equation = var_997_equation_0, values = (var_683_cast_fp16, var_925_cast_fp16))[name = string("op_997_cast_fp16")]; + string var_999_equation_0 = const()[name = string("op_999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_999_cast_fp16 = einsum(equation = var_999_equation_0, values = (var_687_cast_fp16, var_926_cast_fp16))[name = string("op_999_cast_fp16")]; + string var_1001_equation_0 = const()[name = string("op_1001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1001_cast_fp16 = einsum(equation = var_1001_equation_0, values = (var_687_cast_fp16, var_927_cast_fp16))[name = string("op_1001_cast_fp16")]; + string var_1003_equation_0 = const()[name = string("op_1003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1003_cast_fp16 = einsum(equation = var_1003_equation_0, values = (var_687_cast_fp16, var_928_cast_fp16))[name = string("op_1003_cast_fp16")]; + string var_1005_equation_0 = const()[name = string("op_1005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1005_cast_fp16 = einsum(equation = var_1005_equation_0, values = (var_687_cast_fp16, var_929_cast_fp16))[name = string("op_1005_cast_fp16")]; + string var_1007_equation_0 = const()[name = string("op_1007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1007_cast_fp16 = einsum(equation = var_1007_equation_0, values = (var_691_cast_fp16, var_930_cast_fp16))[name = string("op_1007_cast_fp16")]; + string var_1009_equation_0 = const()[name = string("op_1009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1009_cast_fp16 = einsum(equation = var_1009_equation_0, values = (var_691_cast_fp16, var_931_cast_fp16))[name = string("op_1009_cast_fp16")]; + string var_1011_equation_0 = const()[name = string("op_1011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1011_cast_fp16 = einsum(equation = var_1011_equation_0, values = (var_691_cast_fp16, var_932_cast_fp16))[name = string("op_1011_cast_fp16")]; + string var_1013_equation_0 = const()[name = string("op_1013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1013_cast_fp16 = einsum(equation = var_1013_equation_0, values = (var_691_cast_fp16, var_933_cast_fp16))[name = string("op_1013_cast_fp16")]; + string var_1015_equation_0 = const()[name = string("op_1015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1015_cast_fp16 = einsum(equation = var_1015_equation_0, values = (var_695_cast_fp16, var_934_cast_fp16))[name = string("op_1015_cast_fp16")]; + string var_1017_equation_0 = const()[name = string("op_1017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1017_cast_fp16 = einsum(equation = var_1017_equation_0, values = (var_695_cast_fp16, var_935_cast_fp16))[name = string("op_1017_cast_fp16")]; + string var_1019_equation_0 = const()[name = string("op_1019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1019_cast_fp16 = einsum(equation = var_1019_equation_0, values = (var_695_cast_fp16, var_936_cast_fp16))[name = string("op_1019_cast_fp16")]; + string var_1021_equation_0 = const()[name = string("op_1021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1021_cast_fp16 = einsum(equation = var_1021_equation_0, values = (var_695_cast_fp16, var_937_cast_fp16))[name = string("op_1021_cast_fp16")]; + string var_1023_equation_0 = const()[name = string("op_1023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1023_cast_fp16 = einsum(equation = var_1023_equation_0, values = (var_699_cast_fp16, var_938_cast_fp16))[name = string("op_1023_cast_fp16")]; + string var_1025_equation_0 = const()[name = string("op_1025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1025_cast_fp16 = einsum(equation = var_1025_equation_0, values = (var_699_cast_fp16, var_939_cast_fp16))[name = string("op_1025_cast_fp16")]; + string var_1027_equation_0 = const()[name = string("op_1027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1027_cast_fp16 = einsum(equation = var_1027_equation_0, values = (var_699_cast_fp16, var_940_cast_fp16))[name = string("op_1027_cast_fp16")]; + string var_1029_equation_0 = const()[name = string("op_1029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1029_cast_fp16 = einsum(equation = var_1029_equation_0, values = (var_699_cast_fp16, var_941_cast_fp16))[name = string("op_1029_cast_fp16")]; + string var_1031_equation_0 = const()[name = string("op_1031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1031_cast_fp16 = einsum(equation = var_1031_equation_0, values = (var_703_cast_fp16, var_942_cast_fp16))[name = string("op_1031_cast_fp16")]; + string var_1033_equation_0 = const()[name = string("op_1033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1033_cast_fp16 = einsum(equation = var_1033_equation_0, values = (var_703_cast_fp16, var_943_cast_fp16))[name = string("op_1033_cast_fp16")]; + string var_1035_equation_0 = const()[name = string("op_1035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1035_cast_fp16 = einsum(equation = var_1035_equation_0, values = (var_703_cast_fp16, var_944_cast_fp16))[name = string("op_1035_cast_fp16")]; + string var_1037_equation_0 = const()[name = string("op_1037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1037_cast_fp16 = einsum(equation = var_1037_equation_0, values = (var_703_cast_fp16, var_945_cast_fp16))[name = string("op_1037_cast_fp16")]; + string var_1039_equation_0 = const()[name = string("op_1039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1039_cast_fp16 = einsum(equation = var_1039_equation_0, values = (var_707_cast_fp16, var_946_cast_fp16))[name = string("op_1039_cast_fp16")]; + string var_1041_equation_0 = const()[name = string("op_1041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1041_cast_fp16 = einsum(equation = var_1041_equation_0, values = (var_707_cast_fp16, var_947_cast_fp16))[name = string("op_1041_cast_fp16")]; + string var_1043_equation_0 = const()[name = string("op_1043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1043_cast_fp16 = einsum(equation = var_1043_equation_0, values = (var_707_cast_fp16, var_948_cast_fp16))[name = string("op_1043_cast_fp16")]; + string var_1045_equation_0 = const()[name = string("op_1045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1045_cast_fp16 = einsum(equation = var_1045_equation_0, values = (var_707_cast_fp16, var_949_cast_fp16))[name = string("op_1045_cast_fp16")]; + bool var_1047_interleave_0 = const()[name = string("op_1047_interleave_0"), val = bool(false)]; + tensor var_1047_cast_fp16 = concat(axis = var_158, interleave = var_1047_interleave_0, values = (var_951_cast_fp16, var_953_cast_fp16, var_955_cast_fp16, var_957_cast_fp16))[name = string("op_1047_cast_fp16")]; + bool var_1049_interleave_0 = const()[name = string("op_1049_interleave_0"), val = bool(false)]; + tensor var_1049_cast_fp16 = concat(axis = var_158, interleave = var_1049_interleave_0, values = (var_959_cast_fp16, var_961_cast_fp16, var_963_cast_fp16, var_965_cast_fp16))[name = string("op_1049_cast_fp16")]; + bool var_1051_interleave_0 = const()[name = string("op_1051_interleave_0"), val = bool(false)]; + tensor var_1051_cast_fp16 = concat(axis = var_158, interleave = var_1051_interleave_0, values = (var_967_cast_fp16, var_969_cast_fp16, var_971_cast_fp16, var_973_cast_fp16))[name = string("op_1051_cast_fp16")]; + bool var_1053_interleave_0 = const()[name = string("op_1053_interleave_0"), val = bool(false)]; + tensor var_1053_cast_fp16 = concat(axis = var_158, interleave = var_1053_interleave_0, values = (var_975_cast_fp16, var_977_cast_fp16, var_979_cast_fp16, var_981_cast_fp16))[name = string("op_1053_cast_fp16")]; + bool var_1055_interleave_0 = const()[name = string("op_1055_interleave_0"), val = bool(false)]; + tensor var_1055_cast_fp16 = concat(axis = var_158, interleave = var_1055_interleave_0, values = (var_983_cast_fp16, var_985_cast_fp16, var_987_cast_fp16, var_989_cast_fp16))[name = string("op_1055_cast_fp16")]; + bool var_1057_interleave_0 = const()[name = string("op_1057_interleave_0"), val = bool(false)]; + tensor var_1057_cast_fp16 = concat(axis = var_158, interleave = var_1057_interleave_0, values = (var_991_cast_fp16, var_993_cast_fp16, var_995_cast_fp16, var_997_cast_fp16))[name = string("op_1057_cast_fp16")]; + bool var_1059_interleave_0 = const()[name = string("op_1059_interleave_0"), val = bool(false)]; + tensor var_1059_cast_fp16 = concat(axis = var_158, interleave = var_1059_interleave_0, values = (var_999_cast_fp16, var_1001_cast_fp16, var_1003_cast_fp16, var_1005_cast_fp16))[name = string("op_1059_cast_fp16")]; + bool var_1061_interleave_0 = const()[name = string("op_1061_interleave_0"), val = bool(false)]; + tensor var_1061_cast_fp16 = concat(axis = var_158, interleave = var_1061_interleave_0, values = (var_1007_cast_fp16, var_1009_cast_fp16, var_1011_cast_fp16, var_1013_cast_fp16))[name = string("op_1061_cast_fp16")]; + bool var_1063_interleave_0 = const()[name = string("op_1063_interleave_0"), val = bool(false)]; + tensor var_1063_cast_fp16 = concat(axis = var_158, interleave = var_1063_interleave_0, values = (var_1015_cast_fp16, var_1017_cast_fp16, var_1019_cast_fp16, var_1021_cast_fp16))[name = string("op_1063_cast_fp16")]; + bool var_1065_interleave_0 = const()[name = string("op_1065_interleave_0"), val = bool(false)]; + tensor var_1065_cast_fp16 = concat(axis = var_158, interleave = var_1065_interleave_0, values = (var_1023_cast_fp16, var_1025_cast_fp16, var_1027_cast_fp16, var_1029_cast_fp16))[name = string("op_1065_cast_fp16")]; + bool var_1067_interleave_0 = const()[name = string("op_1067_interleave_0"), val = bool(false)]; + tensor var_1067_cast_fp16 = concat(axis = var_158, interleave = var_1067_interleave_0, values = (var_1031_cast_fp16, var_1033_cast_fp16, var_1035_cast_fp16, var_1037_cast_fp16))[name = string("op_1067_cast_fp16")]; + bool var_1069_interleave_0 = const()[name = string("op_1069_interleave_0"), val = bool(false)]; + tensor var_1069_cast_fp16 = concat(axis = var_158, interleave = var_1069_interleave_0, values = (var_1039_cast_fp16, var_1041_cast_fp16, var_1043_cast_fp16, var_1045_cast_fp16))[name = string("op_1069_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_175, interleave = input_1_interleave_0, values = (var_1047_cast_fp16, var_1049_cast_fp16, var_1051_cast_fp16, var_1053_cast_fp16, var_1055_cast_fp16, var_1057_cast_fp16, var_1059_cast_fp16, var_1061_cast_fp16, var_1063_cast_fp16, var_1065_cast_fp16, var_1067_cast_fp16, var_1069_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9763776)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10943488)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_1088_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10945088)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10946688)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10948288)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15666944)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15673152)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20391808)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_1117 = const()[name = string("op_1117"), val = int32(3)]; + int32 var_1134 = const()[name = string("op_1134"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_1151_to_fp16 = const()[name = string("op_1151_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_1151_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20393408)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20395008)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20396608)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21576320)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21577920)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22757632)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937344)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_1189_begin_0 = const()[name = string("op_1189_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1189_end_0 = const()[name = string("op_1189_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1189_end_mask_0 = const()[name = string("op_1189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1189_cast_fp16 = slice_by_index(begin = var_1189_begin_0, end = var_1189_end_0, end_mask = var_1189_end_mask_0, x = query_3_cast_fp16)[name = string("op_1189_cast_fp16")]; + tensor var_1193_begin_0 = const()[name = string("op_1193_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1193_end_0 = const()[name = string("op_1193_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1193_end_mask_0 = const()[name = string("op_1193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1193_cast_fp16 = slice_by_index(begin = var_1193_begin_0, end = var_1193_end_0, end_mask = var_1193_end_mask_0, x = query_3_cast_fp16)[name = string("op_1193_cast_fp16")]; + tensor var_1197_begin_0 = const()[name = string("op_1197_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1197_end_0 = const()[name = string("op_1197_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1197_end_mask_0 = const()[name = string("op_1197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1197_cast_fp16 = slice_by_index(begin = var_1197_begin_0, end = var_1197_end_0, end_mask = var_1197_end_mask_0, x = query_3_cast_fp16)[name = string("op_1197_cast_fp16")]; + tensor var_1201_begin_0 = const()[name = string("op_1201_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1201_end_0 = const()[name = string("op_1201_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1201_end_mask_0 = const()[name = string("op_1201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1201_cast_fp16 = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = query_3_cast_fp16)[name = string("op_1201_cast_fp16")]; + tensor var_1205_begin_0 = const()[name = string("op_1205_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1205_end_0 = const()[name = string("op_1205_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1205_end_mask_0 = const()[name = string("op_1205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1205_cast_fp16 = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, x = query_3_cast_fp16)[name = string("op_1205_cast_fp16")]; + tensor var_1209_begin_0 = const()[name = string("op_1209_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1209_end_0 = const()[name = string("op_1209_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1209_end_mask_0 = const()[name = string("op_1209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1209_cast_fp16 = slice_by_index(begin = var_1209_begin_0, end = var_1209_end_0, end_mask = var_1209_end_mask_0, x = query_3_cast_fp16)[name = string("op_1209_cast_fp16")]; + tensor var_1213_begin_0 = const()[name = string("op_1213_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1213_end_0 = const()[name = string("op_1213_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1213_end_mask_0 = const()[name = string("op_1213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1213_cast_fp16 = slice_by_index(begin = var_1213_begin_0, end = var_1213_end_0, end_mask = var_1213_end_mask_0, x = query_3_cast_fp16)[name = string("op_1213_cast_fp16")]; + tensor var_1217_begin_0 = const()[name = string("op_1217_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1217_end_0 = const()[name = string("op_1217_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1217_end_mask_0 = const()[name = string("op_1217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1217_cast_fp16 = slice_by_index(begin = var_1217_begin_0, end = var_1217_end_0, end_mask = var_1217_end_mask_0, x = query_3_cast_fp16)[name = string("op_1217_cast_fp16")]; + tensor var_1221_begin_0 = const()[name = string("op_1221_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1221_end_0 = const()[name = string("op_1221_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1221_end_mask_0 = const()[name = string("op_1221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1221_cast_fp16 = slice_by_index(begin = var_1221_begin_0, end = var_1221_end_0, end_mask = var_1221_end_mask_0, x = query_3_cast_fp16)[name = string("op_1221_cast_fp16")]; + tensor var_1225_begin_0 = const()[name = string("op_1225_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1225_end_0 = const()[name = string("op_1225_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1225_end_mask_0 = const()[name = string("op_1225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1225_cast_fp16 = slice_by_index(begin = var_1225_begin_0, end = var_1225_end_0, end_mask = var_1225_end_mask_0, x = query_3_cast_fp16)[name = string("op_1225_cast_fp16")]; + tensor var_1229_begin_0 = const()[name = string("op_1229_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1229_end_0 = const()[name = string("op_1229_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1229_end_mask_0 = const()[name = string("op_1229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1229_cast_fp16 = slice_by_index(begin = var_1229_begin_0, end = var_1229_end_0, end_mask = var_1229_end_mask_0, x = query_3_cast_fp16)[name = string("op_1229_cast_fp16")]; + tensor var_1233_begin_0 = const()[name = string("op_1233_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1233_end_0 = const()[name = string("op_1233_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1233_end_mask_0 = const()[name = string("op_1233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1233_cast_fp16 = slice_by_index(begin = var_1233_begin_0, end = var_1233_end_0, end_mask = var_1233_end_mask_0, x = query_3_cast_fp16)[name = string("op_1233_cast_fp16")]; + tensor var_1242_begin_0 = const()[name = string("op_1242_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1242_end_0 = const()[name = string("op_1242_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1242_end_mask_0 = const()[name = string("op_1242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1242_cast_fp16 = slice_by_index(begin = var_1242_begin_0, end = var_1242_end_0, end_mask = var_1242_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1242_cast_fp16")]; + tensor var_1249_begin_0 = const()[name = string("op_1249_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1249_end_0 = const()[name = string("op_1249_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1249_end_mask_0 = const()[name = string("op_1249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1249_cast_fp16 = slice_by_index(begin = var_1249_begin_0, end = var_1249_end_0, end_mask = var_1249_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1249_cast_fp16")]; + tensor var_1256_begin_0 = const()[name = string("op_1256_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1256_end_0 = const()[name = string("op_1256_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1256_end_mask_0 = const()[name = string("op_1256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1256_cast_fp16 = slice_by_index(begin = var_1256_begin_0, end = var_1256_end_0, end_mask = var_1256_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1256_cast_fp16")]; + tensor var_1263_begin_0 = const()[name = string("op_1263_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1263_end_mask_0 = const()[name = string("op_1263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1263_cast_fp16")]; + tensor var_1270_begin_0 = const()[name = string("op_1270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1270_end_0 = const()[name = string("op_1270_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1270_end_mask_0 = const()[name = string("op_1270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1270_cast_fp16")]; + tensor var_1277_begin_0 = const()[name = string("op_1277_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1277_end_0 = const()[name = string("op_1277_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1277_end_mask_0 = const()[name = string("op_1277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1277_cast_fp16 = slice_by_index(begin = var_1277_begin_0, end = var_1277_end_0, end_mask = var_1277_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1277_cast_fp16")]; + tensor var_1284_begin_0 = const()[name = string("op_1284_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1284_end_0 = const()[name = string("op_1284_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1284_end_mask_0 = const()[name = string("op_1284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1284_cast_fp16 = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1284_cast_fp16")]; + tensor var_1291_begin_0 = const()[name = string("op_1291_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1291_end_0 = const()[name = string("op_1291_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1291_end_mask_0 = const()[name = string("op_1291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1291_cast_fp16 = slice_by_index(begin = var_1291_begin_0, end = var_1291_end_0, end_mask = var_1291_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1291_cast_fp16")]; + tensor var_1298_begin_0 = const()[name = string("op_1298_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1298_end_0 = const()[name = string("op_1298_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1298_end_mask_0 = const()[name = string("op_1298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1298_cast_fp16 = slice_by_index(begin = var_1298_begin_0, end = var_1298_end_0, end_mask = var_1298_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1298_cast_fp16")]; + tensor var_1305_begin_0 = const()[name = string("op_1305_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1305_end_0 = const()[name = string("op_1305_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1305_end_mask_0 = const()[name = string("op_1305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1305_cast_fp16 = slice_by_index(begin = var_1305_begin_0, end = var_1305_end_0, end_mask = var_1305_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1305_cast_fp16")]; + tensor var_1312_begin_0 = const()[name = string("op_1312_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1312_end_0 = const()[name = string("op_1312_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1312_end_mask_0 = const()[name = string("op_1312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1312_cast_fp16 = slice_by_index(begin = var_1312_begin_0, end = var_1312_end_0, end_mask = var_1312_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1312_cast_fp16")]; + tensor var_1319_begin_0 = const()[name = string("op_1319_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1319_end_0 = const()[name = string("op_1319_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1319_end_mask_0 = const()[name = string("op_1319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1319_cast_fp16 = slice_by_index(begin = var_1319_begin_0, end = var_1319_end_0, end_mask = var_1319_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1319_cast_fp16")]; + tensor var_1326_begin_0 = const()[name = string("op_1326_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1326_end_0 = const()[name = string("op_1326_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1326_end_mask_0 = const()[name = string("op_1326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1326_cast_fp16 = slice_by_index(begin = var_1326_begin_0, end = var_1326_end_0, end_mask = var_1326_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1326_cast_fp16")]; + tensor var_1333_begin_0 = const()[name = string("op_1333_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1333_end_0 = const()[name = string("op_1333_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1333_end_mask_0 = const()[name = string("op_1333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1333_cast_fp16 = slice_by_index(begin = var_1333_begin_0, end = var_1333_end_0, end_mask = var_1333_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1333_cast_fp16")]; + tensor var_1340_begin_0 = const()[name = string("op_1340_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1340_end_0 = const()[name = string("op_1340_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1340_end_mask_0 = const()[name = string("op_1340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1340_cast_fp16 = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1340_cast_fp16")]; + tensor var_1347_begin_0 = const()[name = string("op_1347_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1347_end_0 = const()[name = string("op_1347_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1347_end_mask_0 = const()[name = string("op_1347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1347_cast_fp16 = slice_by_index(begin = var_1347_begin_0, end = var_1347_end_0, end_mask = var_1347_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1347_cast_fp16")]; + tensor var_1354_begin_0 = const()[name = string("op_1354_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1354_end_0 = const()[name = string("op_1354_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1354_end_mask_0 = const()[name = string("op_1354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1354_cast_fp16 = slice_by_index(begin = var_1354_begin_0, end = var_1354_end_0, end_mask = var_1354_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1354_cast_fp16")]; + tensor var_1361_begin_0 = const()[name = string("op_1361_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1361_end_0 = const()[name = string("op_1361_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1361_end_mask_0 = const()[name = string("op_1361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1361_cast_fp16 = slice_by_index(begin = var_1361_begin_0, end = var_1361_end_0, end_mask = var_1361_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1361_cast_fp16")]; + tensor var_1368_begin_0 = const()[name = string("op_1368_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1368_end_0 = const()[name = string("op_1368_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1368_end_mask_0 = const()[name = string("op_1368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1368_cast_fp16 = slice_by_index(begin = var_1368_begin_0, end = var_1368_end_0, end_mask = var_1368_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1368_cast_fp16")]; + tensor var_1375_begin_0 = const()[name = string("op_1375_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1375_end_0 = const()[name = string("op_1375_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1375_end_mask_0 = const()[name = string("op_1375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1375_cast_fp16 = slice_by_index(begin = var_1375_begin_0, end = var_1375_end_0, end_mask = var_1375_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1375_cast_fp16")]; + tensor var_1382_begin_0 = const()[name = string("op_1382_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1382_end_0 = const()[name = string("op_1382_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1382_end_mask_0 = const()[name = string("op_1382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1382_cast_fp16 = slice_by_index(begin = var_1382_begin_0, end = var_1382_end_0, end_mask = var_1382_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1382_cast_fp16")]; + tensor var_1389_begin_0 = const()[name = string("op_1389_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1389_end_0 = const()[name = string("op_1389_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1389_end_mask_0 = const()[name = string("op_1389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1389_cast_fp16 = slice_by_index(begin = var_1389_begin_0, end = var_1389_end_0, end_mask = var_1389_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1389_cast_fp16")]; + tensor var_1396_begin_0 = const()[name = string("op_1396_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1396_end_0 = const()[name = string("op_1396_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1396_end_mask_0 = const()[name = string("op_1396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1396_cast_fp16 = slice_by_index(begin = var_1396_begin_0, end = var_1396_end_0, end_mask = var_1396_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1396_cast_fp16")]; + tensor var_1403_begin_0 = const()[name = string("op_1403_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1403_end_0 = const()[name = string("op_1403_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1403_end_mask_0 = const()[name = string("op_1403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1403_cast_fp16 = slice_by_index(begin = var_1403_begin_0, end = var_1403_end_0, end_mask = var_1403_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1403_cast_fp16")]; + tensor var_1410_begin_0 = const()[name = string("op_1410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1410_end_0 = const()[name = string("op_1410_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1410_end_mask_0 = const()[name = string("op_1410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1410_cast_fp16 = slice_by_index(begin = var_1410_begin_0, end = var_1410_end_0, end_mask = var_1410_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1410_cast_fp16")]; + tensor var_1417_begin_0 = const()[name = string("op_1417_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1417_end_0 = const()[name = string("op_1417_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1417_end_mask_0 = const()[name = string("op_1417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1417_cast_fp16 = slice_by_index(begin = var_1417_begin_0, end = var_1417_end_0, end_mask = var_1417_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1417_cast_fp16")]; + tensor var_1424_begin_0 = const()[name = string("op_1424_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1424_end_0 = const()[name = string("op_1424_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1424_end_mask_0 = const()[name = string("op_1424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1424_cast_fp16 = slice_by_index(begin = var_1424_begin_0, end = var_1424_end_0, end_mask = var_1424_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1424_cast_fp16")]; + tensor var_1431_begin_0 = const()[name = string("op_1431_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1431_end_0 = const()[name = string("op_1431_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1431_end_mask_0 = const()[name = string("op_1431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1431_cast_fp16 = slice_by_index(begin = var_1431_begin_0, end = var_1431_end_0, end_mask = var_1431_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1431_cast_fp16")]; + tensor var_1438_begin_0 = const()[name = string("op_1438_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1438_end_0 = const()[name = string("op_1438_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1438_end_mask_0 = const()[name = string("op_1438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1438_cast_fp16 = slice_by_index(begin = var_1438_begin_0, end = var_1438_end_0, end_mask = var_1438_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1438_cast_fp16")]; + tensor var_1445_begin_0 = const()[name = string("op_1445_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1445_end_0 = const()[name = string("op_1445_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1445_end_mask_0 = const()[name = string("op_1445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1445_cast_fp16 = slice_by_index(begin = var_1445_begin_0, end = var_1445_end_0, end_mask = var_1445_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1445_cast_fp16")]; + tensor var_1452_begin_0 = const()[name = string("op_1452_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1452_end_0 = const()[name = string("op_1452_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1452_end_mask_0 = const()[name = string("op_1452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1452_cast_fp16 = slice_by_index(begin = var_1452_begin_0, end = var_1452_end_0, end_mask = var_1452_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1452_cast_fp16")]; + tensor var_1459_begin_0 = const()[name = string("op_1459_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1459_end_0 = const()[name = string("op_1459_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1459_end_mask_0 = const()[name = string("op_1459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1459_cast_fp16 = slice_by_index(begin = var_1459_begin_0, end = var_1459_end_0, end_mask = var_1459_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1459_cast_fp16")]; + tensor var_1466_begin_0 = const()[name = string("op_1466_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1466_end_0 = const()[name = string("op_1466_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1466_end_mask_0 = const()[name = string("op_1466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1466_cast_fp16 = slice_by_index(begin = var_1466_begin_0, end = var_1466_end_0, end_mask = var_1466_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1466_cast_fp16")]; + tensor var_1473_begin_0 = const()[name = string("op_1473_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1473_end_0 = const()[name = string("op_1473_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1473_end_mask_0 = const()[name = string("op_1473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1473_cast_fp16")]; + tensor var_1480_begin_0 = const()[name = string("op_1480_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1480_end_0 = const()[name = string("op_1480_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1480_end_mask_0 = const()[name = string("op_1480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1480_cast_fp16 = slice_by_index(begin = var_1480_begin_0, end = var_1480_end_0, end_mask = var_1480_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1480_cast_fp16")]; + tensor var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1487_cast_fp16")]; + tensor var_1494_begin_0 = const()[name = string("op_1494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1494_end_0 = const()[name = string("op_1494_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1494_end_mask_0 = const()[name = string("op_1494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1494_cast_fp16 = slice_by_index(begin = var_1494_begin_0, end = var_1494_end_0, end_mask = var_1494_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1494_cast_fp16")]; + tensor var_1501_begin_0 = const()[name = string("op_1501_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1501_end_0 = const()[name = string("op_1501_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1501_end_mask_0 = const()[name = string("op_1501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1501_cast_fp16 = slice_by_index(begin = var_1501_begin_0, end = var_1501_end_0, end_mask = var_1501_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1501_cast_fp16")]; + tensor var_1508_begin_0 = const()[name = string("op_1508_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1508_end_0 = const()[name = string("op_1508_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1508_end_mask_0 = const()[name = string("op_1508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1508_cast_fp16 = slice_by_index(begin = var_1508_begin_0, end = var_1508_end_0, end_mask = var_1508_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1508_cast_fp16")]; + tensor var_1515_begin_0 = const()[name = string("op_1515_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1515_end_0 = const()[name = string("op_1515_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1515_end_mask_0 = const()[name = string("op_1515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1515_cast_fp16 = slice_by_index(begin = var_1515_begin_0, end = var_1515_end_0, end_mask = var_1515_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1515_cast_fp16")]; + tensor var_1522_begin_0 = const()[name = string("op_1522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1522_end_0 = const()[name = string("op_1522_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1522_end_mask_0 = const()[name = string("op_1522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1522_cast_fp16 = slice_by_index(begin = var_1522_begin_0, end = var_1522_end_0, end_mask = var_1522_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1522_cast_fp16")]; + tensor var_1529_begin_0 = const()[name = string("op_1529_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1529_end_0 = const()[name = string("op_1529_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1529_end_mask_0 = const()[name = string("op_1529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1529_cast_fp16 = slice_by_index(begin = var_1529_begin_0, end = var_1529_end_0, end_mask = var_1529_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1529_cast_fp16")]; + tensor var_1536_begin_0 = const()[name = string("op_1536_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1536_end_0 = const()[name = string("op_1536_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1536_end_mask_0 = const()[name = string("op_1536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1536_cast_fp16 = slice_by_index(begin = var_1536_begin_0, end = var_1536_end_0, end_mask = var_1536_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1536_cast_fp16")]; + tensor var_1543_begin_0 = const()[name = string("op_1543_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1543_end_0 = const()[name = string("op_1543_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1543_end_mask_0 = const()[name = string("op_1543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1543_cast_fp16 = slice_by_index(begin = var_1543_begin_0, end = var_1543_end_0, end_mask = var_1543_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1543_cast_fp16")]; + tensor var_1550_begin_0 = const()[name = string("op_1550_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1550_end_0 = const()[name = string("op_1550_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1550_end_mask_0 = const()[name = string("op_1550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1550_cast_fp16 = slice_by_index(begin = var_1550_begin_0, end = var_1550_end_0, end_mask = var_1550_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1550_cast_fp16")]; + tensor var_1557_begin_0 = const()[name = string("op_1557_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1557_end_0 = const()[name = string("op_1557_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1557_end_mask_0 = const()[name = string("op_1557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1557_cast_fp16 = slice_by_index(begin = var_1557_begin_0, end = var_1557_end_0, end_mask = var_1557_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1557_cast_fp16")]; + tensor var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1564_cast_fp16")]; + tensor var_1571_begin_0 = const()[name = string("op_1571_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1571_end_0 = const()[name = string("op_1571_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1571_end_mask_0 = const()[name = string("op_1571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1571_cast_fp16 = slice_by_index(begin = var_1571_begin_0, end = var_1571_end_0, end_mask = var_1571_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1571_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_10")]; + tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = k_3_cast_fp16)[name = string("op_1576_cast_fp16")]; + tensor var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = k_3_cast_fp16)[name = string("op_1580_cast_fp16")]; + tensor var_1584_begin_0 = const()[name = string("op_1584_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1584_end_0 = const()[name = string("op_1584_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1584_end_mask_0 = const()[name = string("op_1584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = k_3_cast_fp16)[name = string("op_1584_cast_fp16")]; + tensor var_1588_begin_0 = const()[name = string("op_1588_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1588_end_0 = const()[name = string("op_1588_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1588_end_mask_0 = const()[name = string("op_1588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1588_cast_fp16 = slice_by_index(begin = var_1588_begin_0, end = var_1588_end_0, end_mask = var_1588_end_mask_0, x = k_3_cast_fp16)[name = string("op_1588_cast_fp16")]; + tensor var_1592_begin_0 = const()[name = string("op_1592_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1592_end_0 = const()[name = string("op_1592_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1592_end_mask_0 = const()[name = string("op_1592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1592_cast_fp16 = slice_by_index(begin = var_1592_begin_0, end = var_1592_end_0, end_mask = var_1592_end_mask_0, x = k_3_cast_fp16)[name = string("op_1592_cast_fp16")]; + tensor var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = k_3_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1600_begin_0 = const()[name = string("op_1600_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1600_end_0 = const()[name = string("op_1600_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1600_end_mask_0 = const()[name = string("op_1600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1600_cast_fp16 = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, x = k_3_cast_fp16)[name = string("op_1600_cast_fp16")]; + tensor var_1604_begin_0 = const()[name = string("op_1604_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1604_end_0 = const()[name = string("op_1604_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1604_end_mask_0 = const()[name = string("op_1604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1604_cast_fp16 = slice_by_index(begin = var_1604_begin_0, end = var_1604_end_0, end_mask = var_1604_end_mask_0, x = k_3_cast_fp16)[name = string("op_1604_cast_fp16")]; + tensor var_1608_begin_0 = const()[name = string("op_1608_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1608_end_0 = const()[name = string("op_1608_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_1608_end_mask_0 = const()[name = string("op_1608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1608_cast_fp16 = slice_by_index(begin = var_1608_begin_0, end = var_1608_end_0, end_mask = var_1608_end_mask_0, x = k_3_cast_fp16)[name = string("op_1608_cast_fp16")]; + tensor var_1612_begin_0 = const()[name = string("op_1612_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1612_end_0 = const()[name = string("op_1612_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_1612_end_mask_0 = const()[name = string("op_1612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1612_cast_fp16 = slice_by_index(begin = var_1612_begin_0, end = var_1612_end_0, end_mask = var_1612_end_mask_0, x = k_3_cast_fp16)[name = string("op_1612_cast_fp16")]; + tensor var_1616_begin_0 = const()[name = string("op_1616_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_1616_end_0 = const()[name = string("op_1616_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_1616_end_mask_0 = const()[name = string("op_1616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1616_cast_fp16 = slice_by_index(begin = var_1616_begin_0, end = var_1616_end_0, end_mask = var_1616_end_mask_0, x = k_3_cast_fp16)[name = string("op_1616_cast_fp16")]; + tensor var_1620_begin_0 = const()[name = string("op_1620_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_1620_end_0 = const()[name = string("op_1620_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_1620_end_mask_0 = const()[name = string("op_1620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1620_cast_fp16 = slice_by_index(begin = var_1620_begin_0, end = var_1620_end_0, end_mask = var_1620_end_mask_0, x = k_3_cast_fp16)[name = string("op_1620_cast_fp16")]; + tensor var_1622_begin_0 = const()[name = string("op_1622_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1622_end_0 = const()[name = string("op_1622_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1622_end_mask_0 = const()[name = string("op_1622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1622_cast_fp16 = slice_by_index(begin = var_1622_begin_0, end = var_1622_end_0, end_mask = var_1622_end_mask_0, x = value_3_cast_fp16)[name = string("op_1622_cast_fp16")]; + tensor var_1626_begin_0 = const()[name = string("op_1626_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1626_end_0 = const()[name = string("op_1626_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1626_end_mask_0 = const()[name = string("op_1626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1626_cast_fp16 = slice_by_index(begin = var_1626_begin_0, end = var_1626_end_0, end_mask = var_1626_end_mask_0, x = value_3_cast_fp16)[name = string("op_1626_cast_fp16")]; + tensor var_1630_begin_0 = const()[name = string("op_1630_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1630_end_0 = const()[name = string("op_1630_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1630_end_mask_0 = const()[name = string("op_1630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1630_cast_fp16 = slice_by_index(begin = var_1630_begin_0, end = var_1630_end_0, end_mask = var_1630_end_mask_0, x = value_3_cast_fp16)[name = string("op_1630_cast_fp16")]; + tensor var_1634_begin_0 = const()[name = string("op_1634_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1634_end_0 = const()[name = string("op_1634_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1634_end_mask_0 = const()[name = string("op_1634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1634_cast_fp16 = slice_by_index(begin = var_1634_begin_0, end = var_1634_end_0, end_mask = var_1634_end_mask_0, x = value_3_cast_fp16)[name = string("op_1634_cast_fp16")]; + tensor var_1638_begin_0 = const()[name = string("op_1638_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1638_end_0 = const()[name = string("op_1638_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1638_end_mask_0 = const()[name = string("op_1638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = value_3_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor var_1642_begin_0 = const()[name = string("op_1642_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1642_end_0 = const()[name = string("op_1642_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1642_end_mask_0 = const()[name = string("op_1642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1642_cast_fp16 = slice_by_index(begin = var_1642_begin_0, end = var_1642_end_0, end_mask = var_1642_end_mask_0, x = value_3_cast_fp16)[name = string("op_1642_cast_fp16")]; + tensor var_1646_begin_0 = const()[name = string("op_1646_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1646_end_0 = const()[name = string("op_1646_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1646_end_mask_0 = const()[name = string("op_1646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1646_cast_fp16 = slice_by_index(begin = var_1646_begin_0, end = var_1646_end_0, end_mask = var_1646_end_mask_0, x = value_3_cast_fp16)[name = string("op_1646_cast_fp16")]; + tensor var_1650_begin_0 = const()[name = string("op_1650_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1650_end_0 = const()[name = string("op_1650_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1650_end_mask_0 = const()[name = string("op_1650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1650_cast_fp16 = slice_by_index(begin = var_1650_begin_0, end = var_1650_end_0, end_mask = var_1650_end_mask_0, x = value_3_cast_fp16)[name = string("op_1650_cast_fp16")]; + tensor var_1654_begin_0 = const()[name = string("op_1654_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1654_end_0 = const()[name = string("op_1654_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1654_end_mask_0 = const()[name = string("op_1654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1654_cast_fp16 = slice_by_index(begin = var_1654_begin_0, end = var_1654_end_0, end_mask = var_1654_end_mask_0, x = value_3_cast_fp16)[name = string("op_1654_cast_fp16")]; + tensor var_1658_begin_0 = const()[name = string("op_1658_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1658_end_0 = const()[name = string("op_1658_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1658_end_mask_0 = const()[name = string("op_1658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1658_cast_fp16 = slice_by_index(begin = var_1658_begin_0, end = var_1658_end_0, end_mask = var_1658_end_mask_0, x = value_3_cast_fp16)[name = string("op_1658_cast_fp16")]; + tensor var_1662_begin_0 = const()[name = string("op_1662_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1662_end_0 = const()[name = string("op_1662_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1662_end_mask_0 = const()[name = string("op_1662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1662_cast_fp16 = slice_by_index(begin = var_1662_begin_0, end = var_1662_end_0, end_mask = var_1662_end_mask_0, x = value_3_cast_fp16)[name = string("op_1662_cast_fp16")]; + tensor var_1666_begin_0 = const()[name = string("op_1666_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1666_end_0 = const()[name = string("op_1666_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1666_end_mask_0 = const()[name = string("op_1666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1666_cast_fp16 = slice_by_index(begin = var_1666_begin_0, end = var_1666_end_0, end_mask = var_1666_end_mask_0, x = value_3_cast_fp16)[name = string("op_1666_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1576_cast_fp16, var_1242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1576_cast_fp16, var_1249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1576_cast_fp16, var_1256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1576_cast_fp16, var_1263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1580_cast_fp16, var_1270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1580_cast_fp16, var_1277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1580_cast_fp16, var_1284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1580_cast_fp16, var_1291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1584_cast_fp16, var_1298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1584_cast_fp16, var_1305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1584_cast_fp16, var_1312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1584_cast_fp16, var_1319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1588_cast_fp16, var_1326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1588_cast_fp16, var_1333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1588_cast_fp16, var_1340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1588_cast_fp16, var_1347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1592_cast_fp16, var_1354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1592_cast_fp16, var_1361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1592_cast_fp16, var_1368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1592_cast_fp16, var_1375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1596_cast_fp16, var_1382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1596_cast_fp16, var_1389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1596_cast_fp16, var_1396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1596_cast_fp16, var_1403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1600_cast_fp16, var_1410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1600_cast_fp16, var_1417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1600_cast_fp16, var_1424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1600_cast_fp16, var_1431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1604_cast_fp16, var_1438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1604_cast_fp16, var_1445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1604_cast_fp16, var_1452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1604_cast_fp16, var_1459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1608_cast_fp16, var_1466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1608_cast_fp16, var_1473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1608_cast_fp16, var_1480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1608_cast_fp16, var_1487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1612_cast_fp16, var_1494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1612_cast_fp16, var_1501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1612_cast_fp16, var_1508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1612_cast_fp16, var_1515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_1616_cast_fp16, var_1522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_1616_cast_fp16, var_1529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_1616_cast_fp16, var_1536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_1616_cast_fp16, var_1543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_1620_cast_fp16, var_1550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_1620_cast_fp16, var_1557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_1620_cast_fp16, var_1564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_1620_cast_fp16, var_1571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + fp16 var_1765_to_fp16 = const()[name = string("op_1765_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1765_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1767_to_fp16 = const()[name = string("op_1767_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1767_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1769_to_fp16 = const()[name = string("op_1769_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1769_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1771_to_fp16 = const()[name = string("op_1771_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1771_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1773_to_fp16 = const()[name = string("op_1773_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1773_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1775_to_fp16 = const()[name = string("op_1775_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1775_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1777_to_fp16 = const()[name = string("op_1777_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1777_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1779_to_fp16 = const()[name = string("op_1779_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1779_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1781_to_fp16 = const()[name = string("op_1781_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1781_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1783_to_fp16 = const()[name = string("op_1783_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1783_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1785_to_fp16 = const()[name = string("op_1785_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1785_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1787_to_fp16 = const()[name = string("op_1787_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1787_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1789_to_fp16 = const()[name = string("op_1789_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1789_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1791_to_fp16 = const()[name = string("op_1791_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1791_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1793_to_fp16 = const()[name = string("op_1793_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1793_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1795_to_fp16 = const()[name = string("op_1795_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1795_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + fp16 var_1797_to_fp16 = const()[name = string("op_1797_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1797_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1799_to_fp16 = const()[name = string("op_1799_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1799_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1801_to_fp16 = const()[name = string("op_1801_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1801_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1803_to_fp16 = const()[name = string("op_1803_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1803_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1805_to_fp16 = const()[name = string("op_1805_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1805_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1807_to_fp16 = const()[name = string("op_1807_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1807_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1809_to_fp16 = const()[name = string("op_1809_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1809_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1811_to_fp16 = const()[name = string("op_1811_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1811_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1813_to_fp16 = const()[name = string("op_1813_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1813_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1815_to_fp16 = const()[name = string("op_1815_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1815_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1817_to_fp16 = const()[name = string("op_1817_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1817_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1819_to_fp16 = const()[name = string("op_1819_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1819_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1821_to_fp16 = const()[name = string("op_1821_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1821_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1823_to_fp16 = const()[name = string("op_1823_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1823_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1825_to_fp16 = const()[name = string("op_1825_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1825_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1827_to_fp16 = const()[name = string("op_1827_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1827_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_1829_to_fp16 = const()[name = string("op_1829_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_1829_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_1831_to_fp16 = const()[name = string("op_1831_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_1831_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_1833_to_fp16 = const()[name = string("op_1833_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_1833_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_1835_to_fp16 = const()[name = string("op_1835_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_1835_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_1837_to_fp16 = const()[name = string("op_1837_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_1837_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_1839_to_fp16 = const()[name = string("op_1839_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_1839_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_1841_to_fp16 = const()[name = string("op_1841_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_1841_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_1843_to_fp16 = const()[name = string("op_1843_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_1843_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_1845_to_fp16 = const()[name = string("op_1845_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_1845_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_1847_to_fp16 = const()[name = string("op_1847_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_1847_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_1849_to_fp16 = const()[name = string("op_1849_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_1849_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_1851_to_fp16 = const()[name = string("op_1851_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_1851_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_1853_to_fp16 = const()[name = string("op_1853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_1853_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_1855_to_fp16 = const()[name = string("op_1855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_1855_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_1857_to_fp16 = const()[name = string("op_1857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_1857_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_1859_to_fp16 = const()[name = string("op_1859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_1859_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + tensor var_1861_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_97_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1862_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_99_cast_fp16)[name = string("op_1862_cast_fp16")]; + tensor var_1863_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_101_cast_fp16)[name = string("op_1863_cast_fp16")]; + tensor var_1864_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_103_cast_fp16)[name = string("op_1864_cast_fp16")]; + tensor var_1865_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_105_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1866_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_107_cast_fp16)[name = string("op_1866_cast_fp16")]; + tensor var_1867_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_109_cast_fp16)[name = string("op_1867_cast_fp16")]; + tensor var_1868_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_111_cast_fp16)[name = string("op_1868_cast_fp16")]; + tensor var_1869_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_113_cast_fp16)[name = string("op_1869_cast_fp16")]; + tensor var_1870_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_115_cast_fp16)[name = string("op_1870_cast_fp16")]; + tensor var_1871_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_117_cast_fp16)[name = string("op_1871_cast_fp16")]; + tensor var_1872_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_119_cast_fp16)[name = string("op_1872_cast_fp16")]; + tensor var_1873_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_121_cast_fp16)[name = string("op_1873_cast_fp16")]; + tensor var_1874_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_123_cast_fp16)[name = string("op_1874_cast_fp16")]; + tensor var_1875_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_125_cast_fp16)[name = string("op_1875_cast_fp16")]; + tensor var_1876_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_127_cast_fp16)[name = string("op_1876_cast_fp16")]; + tensor var_1877_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_129_cast_fp16)[name = string("op_1877_cast_fp16")]; + tensor var_1878_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_131_cast_fp16)[name = string("op_1878_cast_fp16")]; + tensor var_1879_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_133_cast_fp16)[name = string("op_1879_cast_fp16")]; + tensor var_1880_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_135_cast_fp16)[name = string("op_1880_cast_fp16")]; + tensor var_1881_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_137_cast_fp16)[name = string("op_1881_cast_fp16")]; + tensor var_1882_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_139_cast_fp16)[name = string("op_1882_cast_fp16")]; + tensor var_1883_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_141_cast_fp16)[name = string("op_1883_cast_fp16")]; + tensor var_1884_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_143_cast_fp16)[name = string("op_1884_cast_fp16")]; + tensor var_1885_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_145_cast_fp16)[name = string("op_1885_cast_fp16")]; + tensor var_1886_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_147_cast_fp16)[name = string("op_1886_cast_fp16")]; + tensor var_1887_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_149_cast_fp16)[name = string("op_1887_cast_fp16")]; + tensor var_1888_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_151_cast_fp16)[name = string("op_1888_cast_fp16")]; + tensor var_1889_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_153_cast_fp16)[name = string("op_1889_cast_fp16")]; + tensor var_1890_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_155_cast_fp16)[name = string("op_1890_cast_fp16")]; + tensor var_1891_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_157_cast_fp16)[name = string("op_1891_cast_fp16")]; + tensor var_1892_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_159_cast_fp16)[name = string("op_1892_cast_fp16")]; + tensor var_1893_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_161_cast_fp16)[name = string("op_1893_cast_fp16")]; + tensor var_1894_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_163_cast_fp16)[name = string("op_1894_cast_fp16")]; + tensor var_1895_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_165_cast_fp16)[name = string("op_1895_cast_fp16")]; + tensor var_1896_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_167_cast_fp16)[name = string("op_1896_cast_fp16")]; + tensor var_1897_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_169_cast_fp16)[name = string("op_1897_cast_fp16")]; + tensor var_1898_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_171_cast_fp16)[name = string("op_1898_cast_fp16")]; + tensor var_1899_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_173_cast_fp16)[name = string("op_1899_cast_fp16")]; + tensor var_1900_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_175_cast_fp16)[name = string("op_1900_cast_fp16")]; + tensor var_1901_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_177_cast_fp16)[name = string("op_1901_cast_fp16")]; + tensor var_1902_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_179_cast_fp16)[name = string("op_1902_cast_fp16")]; + tensor var_1903_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_181_cast_fp16)[name = string("op_1903_cast_fp16")]; + tensor var_1904_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_183_cast_fp16)[name = string("op_1904_cast_fp16")]; + tensor var_1905_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_185_cast_fp16)[name = string("op_1905_cast_fp16")]; + tensor var_1906_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_187_cast_fp16)[name = string("op_1906_cast_fp16")]; + tensor var_1907_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_189_cast_fp16)[name = string("op_1907_cast_fp16")]; + tensor var_1908_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_191_cast_fp16)[name = string("op_1908_cast_fp16")]; + string var_1910_equation_0 = const()[name = string("op_1910_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1910_cast_fp16 = einsum(equation = var_1910_equation_0, values = (var_1622_cast_fp16, var_1861_cast_fp16))[name = string("op_1910_cast_fp16")]; + string var_1912_equation_0 = const()[name = string("op_1912_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1912_cast_fp16 = einsum(equation = var_1912_equation_0, values = (var_1622_cast_fp16, var_1862_cast_fp16))[name = string("op_1912_cast_fp16")]; + string var_1914_equation_0 = const()[name = string("op_1914_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1914_cast_fp16 = einsum(equation = var_1914_equation_0, values = (var_1622_cast_fp16, var_1863_cast_fp16))[name = string("op_1914_cast_fp16")]; + string var_1916_equation_0 = const()[name = string("op_1916_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1916_cast_fp16 = einsum(equation = var_1916_equation_0, values = (var_1622_cast_fp16, var_1864_cast_fp16))[name = string("op_1916_cast_fp16")]; + string var_1918_equation_0 = const()[name = string("op_1918_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1918_cast_fp16 = einsum(equation = var_1918_equation_0, values = (var_1626_cast_fp16, var_1865_cast_fp16))[name = string("op_1918_cast_fp16")]; + string var_1920_equation_0 = const()[name = string("op_1920_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1920_cast_fp16 = einsum(equation = var_1920_equation_0, values = (var_1626_cast_fp16, var_1866_cast_fp16))[name = string("op_1920_cast_fp16")]; + string var_1922_equation_0 = const()[name = string("op_1922_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1922_cast_fp16 = einsum(equation = var_1922_equation_0, values = (var_1626_cast_fp16, var_1867_cast_fp16))[name = string("op_1922_cast_fp16")]; + string var_1924_equation_0 = const()[name = string("op_1924_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1924_cast_fp16 = einsum(equation = var_1924_equation_0, values = (var_1626_cast_fp16, var_1868_cast_fp16))[name = string("op_1924_cast_fp16")]; + string var_1926_equation_0 = const()[name = string("op_1926_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1926_cast_fp16 = einsum(equation = var_1926_equation_0, values = (var_1630_cast_fp16, var_1869_cast_fp16))[name = string("op_1926_cast_fp16")]; + string var_1928_equation_0 = const()[name = string("op_1928_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1928_cast_fp16 = einsum(equation = var_1928_equation_0, values = (var_1630_cast_fp16, var_1870_cast_fp16))[name = string("op_1928_cast_fp16")]; + string var_1930_equation_0 = const()[name = string("op_1930_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1930_cast_fp16 = einsum(equation = var_1930_equation_0, values = (var_1630_cast_fp16, var_1871_cast_fp16))[name = string("op_1930_cast_fp16")]; + string var_1932_equation_0 = const()[name = string("op_1932_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1932_cast_fp16 = einsum(equation = var_1932_equation_0, values = (var_1630_cast_fp16, var_1872_cast_fp16))[name = string("op_1932_cast_fp16")]; + string var_1934_equation_0 = const()[name = string("op_1934_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1934_cast_fp16 = einsum(equation = var_1934_equation_0, values = (var_1634_cast_fp16, var_1873_cast_fp16))[name = string("op_1934_cast_fp16")]; + string var_1936_equation_0 = const()[name = string("op_1936_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1936_cast_fp16 = einsum(equation = var_1936_equation_0, values = (var_1634_cast_fp16, var_1874_cast_fp16))[name = string("op_1936_cast_fp16")]; + string var_1938_equation_0 = const()[name = string("op_1938_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1938_cast_fp16 = einsum(equation = var_1938_equation_0, values = (var_1634_cast_fp16, var_1875_cast_fp16))[name = string("op_1938_cast_fp16")]; + string var_1940_equation_0 = const()[name = string("op_1940_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1940_cast_fp16 = einsum(equation = var_1940_equation_0, values = (var_1634_cast_fp16, var_1876_cast_fp16))[name = string("op_1940_cast_fp16")]; + string var_1942_equation_0 = const()[name = string("op_1942_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1942_cast_fp16 = einsum(equation = var_1942_equation_0, values = (var_1638_cast_fp16, var_1877_cast_fp16))[name = string("op_1942_cast_fp16")]; + string var_1944_equation_0 = const()[name = string("op_1944_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1944_cast_fp16 = einsum(equation = var_1944_equation_0, values = (var_1638_cast_fp16, var_1878_cast_fp16))[name = string("op_1944_cast_fp16")]; + string var_1946_equation_0 = const()[name = string("op_1946_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1946_cast_fp16 = einsum(equation = var_1946_equation_0, values = (var_1638_cast_fp16, var_1879_cast_fp16))[name = string("op_1946_cast_fp16")]; + string var_1948_equation_0 = const()[name = string("op_1948_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1948_cast_fp16 = einsum(equation = var_1948_equation_0, values = (var_1638_cast_fp16, var_1880_cast_fp16))[name = string("op_1948_cast_fp16")]; + string var_1950_equation_0 = const()[name = string("op_1950_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1950_cast_fp16 = einsum(equation = var_1950_equation_0, values = (var_1642_cast_fp16, var_1881_cast_fp16))[name = string("op_1950_cast_fp16")]; + string var_1952_equation_0 = const()[name = string("op_1952_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1952_cast_fp16 = einsum(equation = var_1952_equation_0, values = (var_1642_cast_fp16, var_1882_cast_fp16))[name = string("op_1952_cast_fp16")]; + string var_1954_equation_0 = const()[name = string("op_1954_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1954_cast_fp16 = einsum(equation = var_1954_equation_0, values = (var_1642_cast_fp16, var_1883_cast_fp16))[name = string("op_1954_cast_fp16")]; + string var_1956_equation_0 = const()[name = string("op_1956_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1956_cast_fp16 = einsum(equation = var_1956_equation_0, values = (var_1642_cast_fp16, var_1884_cast_fp16))[name = string("op_1956_cast_fp16")]; + string var_1958_equation_0 = const()[name = string("op_1958_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1958_cast_fp16 = einsum(equation = var_1958_equation_0, values = (var_1646_cast_fp16, var_1885_cast_fp16))[name = string("op_1958_cast_fp16")]; + string var_1960_equation_0 = const()[name = string("op_1960_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1960_cast_fp16 = einsum(equation = var_1960_equation_0, values = (var_1646_cast_fp16, var_1886_cast_fp16))[name = string("op_1960_cast_fp16")]; + string var_1962_equation_0 = const()[name = string("op_1962_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1962_cast_fp16 = einsum(equation = var_1962_equation_0, values = (var_1646_cast_fp16, var_1887_cast_fp16))[name = string("op_1962_cast_fp16")]; + string var_1964_equation_0 = const()[name = string("op_1964_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1964_cast_fp16 = einsum(equation = var_1964_equation_0, values = (var_1646_cast_fp16, var_1888_cast_fp16))[name = string("op_1964_cast_fp16")]; + string var_1966_equation_0 = const()[name = string("op_1966_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1966_cast_fp16 = einsum(equation = var_1966_equation_0, values = (var_1650_cast_fp16, var_1889_cast_fp16))[name = string("op_1966_cast_fp16")]; + string var_1968_equation_0 = const()[name = string("op_1968_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1968_cast_fp16 = einsum(equation = var_1968_equation_0, values = (var_1650_cast_fp16, var_1890_cast_fp16))[name = string("op_1968_cast_fp16")]; + string var_1970_equation_0 = const()[name = string("op_1970_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1970_cast_fp16 = einsum(equation = var_1970_equation_0, values = (var_1650_cast_fp16, var_1891_cast_fp16))[name = string("op_1970_cast_fp16")]; + string var_1972_equation_0 = const()[name = string("op_1972_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1972_cast_fp16 = einsum(equation = var_1972_equation_0, values = (var_1650_cast_fp16, var_1892_cast_fp16))[name = string("op_1972_cast_fp16")]; + string var_1974_equation_0 = const()[name = string("op_1974_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1974_cast_fp16 = einsum(equation = var_1974_equation_0, values = (var_1654_cast_fp16, var_1893_cast_fp16))[name = string("op_1974_cast_fp16")]; + string var_1976_equation_0 = const()[name = string("op_1976_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1976_cast_fp16 = einsum(equation = var_1976_equation_0, values = (var_1654_cast_fp16, var_1894_cast_fp16))[name = string("op_1976_cast_fp16")]; + string var_1978_equation_0 = const()[name = string("op_1978_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1978_cast_fp16 = einsum(equation = var_1978_equation_0, values = (var_1654_cast_fp16, var_1895_cast_fp16))[name = string("op_1978_cast_fp16")]; + string var_1980_equation_0 = const()[name = string("op_1980_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1980_cast_fp16 = einsum(equation = var_1980_equation_0, values = (var_1654_cast_fp16, var_1896_cast_fp16))[name = string("op_1980_cast_fp16")]; + string var_1982_equation_0 = const()[name = string("op_1982_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1982_cast_fp16 = einsum(equation = var_1982_equation_0, values = (var_1658_cast_fp16, var_1897_cast_fp16))[name = string("op_1982_cast_fp16")]; + string var_1984_equation_0 = const()[name = string("op_1984_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1984_cast_fp16 = einsum(equation = var_1984_equation_0, values = (var_1658_cast_fp16, var_1898_cast_fp16))[name = string("op_1984_cast_fp16")]; + string var_1986_equation_0 = const()[name = string("op_1986_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1986_cast_fp16 = einsum(equation = var_1986_equation_0, values = (var_1658_cast_fp16, var_1899_cast_fp16))[name = string("op_1986_cast_fp16")]; + string var_1988_equation_0 = const()[name = string("op_1988_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1988_cast_fp16 = einsum(equation = var_1988_equation_0, values = (var_1658_cast_fp16, var_1900_cast_fp16))[name = string("op_1988_cast_fp16")]; + string var_1990_equation_0 = const()[name = string("op_1990_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1990_cast_fp16 = einsum(equation = var_1990_equation_0, values = (var_1662_cast_fp16, var_1901_cast_fp16))[name = string("op_1990_cast_fp16")]; + string var_1992_equation_0 = const()[name = string("op_1992_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1992_cast_fp16 = einsum(equation = var_1992_equation_0, values = (var_1662_cast_fp16, var_1902_cast_fp16))[name = string("op_1992_cast_fp16")]; + string var_1994_equation_0 = const()[name = string("op_1994_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1994_cast_fp16 = einsum(equation = var_1994_equation_0, values = (var_1662_cast_fp16, var_1903_cast_fp16))[name = string("op_1994_cast_fp16")]; + string var_1996_equation_0 = const()[name = string("op_1996_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1996_cast_fp16 = einsum(equation = var_1996_equation_0, values = (var_1662_cast_fp16, var_1904_cast_fp16))[name = string("op_1996_cast_fp16")]; + string var_1998_equation_0 = const()[name = string("op_1998_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1998_cast_fp16 = einsum(equation = var_1998_equation_0, values = (var_1666_cast_fp16, var_1905_cast_fp16))[name = string("op_1998_cast_fp16")]; + string var_2000_equation_0 = const()[name = string("op_2000_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2000_cast_fp16 = einsum(equation = var_2000_equation_0, values = (var_1666_cast_fp16, var_1906_cast_fp16))[name = string("op_2000_cast_fp16")]; + string var_2002_equation_0 = const()[name = string("op_2002_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2002_cast_fp16 = einsum(equation = var_2002_equation_0, values = (var_1666_cast_fp16, var_1907_cast_fp16))[name = string("op_2002_cast_fp16")]; + string var_2004_equation_0 = const()[name = string("op_2004_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2004_cast_fp16 = einsum(equation = var_2004_equation_0, values = (var_1666_cast_fp16, var_1908_cast_fp16))[name = string("op_2004_cast_fp16")]; + bool var_2006_interleave_0 = const()[name = string("op_2006_interleave_0"), val = bool(false)]; + tensor var_2006_cast_fp16 = concat(axis = var_1117, interleave = var_2006_interleave_0, values = (var_1910_cast_fp16, var_1912_cast_fp16, var_1914_cast_fp16, var_1916_cast_fp16))[name = string("op_2006_cast_fp16")]; + bool var_2008_interleave_0 = const()[name = string("op_2008_interleave_0"), val = bool(false)]; + tensor var_2008_cast_fp16 = concat(axis = var_1117, interleave = var_2008_interleave_0, values = (var_1918_cast_fp16, var_1920_cast_fp16, var_1922_cast_fp16, var_1924_cast_fp16))[name = string("op_2008_cast_fp16")]; + bool var_2010_interleave_0 = const()[name = string("op_2010_interleave_0"), val = bool(false)]; + tensor var_2010_cast_fp16 = concat(axis = var_1117, interleave = var_2010_interleave_0, values = (var_1926_cast_fp16, var_1928_cast_fp16, var_1930_cast_fp16, var_1932_cast_fp16))[name = string("op_2010_cast_fp16")]; + bool var_2012_interleave_0 = const()[name = string("op_2012_interleave_0"), val = bool(false)]; + tensor var_2012_cast_fp16 = concat(axis = var_1117, interleave = var_2012_interleave_0, values = (var_1934_cast_fp16, var_1936_cast_fp16, var_1938_cast_fp16, var_1940_cast_fp16))[name = string("op_2012_cast_fp16")]; + bool var_2014_interleave_0 = const()[name = string("op_2014_interleave_0"), val = bool(false)]; + tensor var_2014_cast_fp16 = concat(axis = var_1117, interleave = var_2014_interleave_0, values = (var_1942_cast_fp16, var_1944_cast_fp16, var_1946_cast_fp16, var_1948_cast_fp16))[name = string("op_2014_cast_fp16")]; + bool var_2016_interleave_0 = const()[name = string("op_2016_interleave_0"), val = bool(false)]; + tensor var_2016_cast_fp16 = concat(axis = var_1117, interleave = var_2016_interleave_0, values = (var_1950_cast_fp16, var_1952_cast_fp16, var_1954_cast_fp16, var_1956_cast_fp16))[name = string("op_2016_cast_fp16")]; + bool var_2018_interleave_0 = const()[name = string("op_2018_interleave_0"), val = bool(false)]; + tensor var_2018_cast_fp16 = concat(axis = var_1117, interleave = var_2018_interleave_0, values = (var_1958_cast_fp16, var_1960_cast_fp16, var_1962_cast_fp16, var_1964_cast_fp16))[name = string("op_2018_cast_fp16")]; + bool var_2020_interleave_0 = const()[name = string("op_2020_interleave_0"), val = bool(false)]; + tensor var_2020_cast_fp16 = concat(axis = var_1117, interleave = var_2020_interleave_0, values = (var_1966_cast_fp16, var_1968_cast_fp16, var_1970_cast_fp16, var_1972_cast_fp16))[name = string("op_2020_cast_fp16")]; + bool var_2022_interleave_0 = const()[name = string("op_2022_interleave_0"), val = bool(false)]; + tensor var_2022_cast_fp16 = concat(axis = var_1117, interleave = var_2022_interleave_0, values = (var_1974_cast_fp16, var_1976_cast_fp16, var_1978_cast_fp16, var_1980_cast_fp16))[name = string("op_2022_cast_fp16")]; + bool var_2024_interleave_0 = const()[name = string("op_2024_interleave_0"), val = bool(false)]; + tensor var_2024_cast_fp16 = concat(axis = var_1117, interleave = var_2024_interleave_0, values = (var_1982_cast_fp16, var_1984_cast_fp16, var_1986_cast_fp16, var_1988_cast_fp16))[name = string("op_2024_cast_fp16")]; + bool var_2026_interleave_0 = const()[name = string("op_2026_interleave_0"), val = bool(false)]; + tensor var_2026_cast_fp16 = concat(axis = var_1117, interleave = var_2026_interleave_0, values = (var_1990_cast_fp16, var_1992_cast_fp16, var_1994_cast_fp16, var_1996_cast_fp16))[name = string("op_2026_cast_fp16")]; + bool var_2028_interleave_0 = const()[name = string("op_2028_interleave_0"), val = bool(false)]; + tensor var_2028_cast_fp16 = concat(axis = var_1117, interleave = var_2028_interleave_0, values = (var_1998_cast_fp16, var_2000_cast_fp16, var_2002_cast_fp16, var_2004_cast_fp16))[name = string("op_2028_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_1134, interleave = input_9_interleave_0, values = (var_2006_cast_fp16, var_2008_cast_fp16, var_2010_cast_fp16, var_2012_cast_fp16, var_2014_cast_fp16, var_2016_cast_fp16, var_2018_cast_fp16, var_2020_cast_fp16, var_2022_cast_fp16, var_2024_cast_fp16, var_2026_cast_fp16, var_2028_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23938944)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25118656)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_2047_to_fp16 = const()[name = string("op_2047_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_2047_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25120256)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25121856)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25123456)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29842112)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29848320)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34566976)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_2076 = const()[name = string("op_2076"), val = int32(3)]; + int32 var_2093 = const()[name = string("op_2093"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_2110_to_fp16 = const()[name = string("op_2110_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_2110_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34568576)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34570176)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34571776)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35751488)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35753088)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36932800)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38112512)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_2148_begin_0 = const()[name = string("op_2148_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2148_end_0 = const()[name = string("op_2148_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2148_end_mask_0 = const()[name = string("op_2148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2148_cast_fp16 = slice_by_index(begin = var_2148_begin_0, end = var_2148_end_0, end_mask = var_2148_end_mask_0, x = query_5_cast_fp16)[name = string("op_2148_cast_fp16")]; + tensor var_2152_begin_0 = const()[name = string("op_2152_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2152_end_0 = const()[name = string("op_2152_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2152_end_mask_0 = const()[name = string("op_2152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2152_cast_fp16 = slice_by_index(begin = var_2152_begin_0, end = var_2152_end_0, end_mask = var_2152_end_mask_0, x = query_5_cast_fp16)[name = string("op_2152_cast_fp16")]; + tensor var_2156_begin_0 = const()[name = string("op_2156_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2156_end_0 = const()[name = string("op_2156_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2156_end_mask_0 = const()[name = string("op_2156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2156_cast_fp16 = slice_by_index(begin = var_2156_begin_0, end = var_2156_end_0, end_mask = var_2156_end_mask_0, x = query_5_cast_fp16)[name = string("op_2156_cast_fp16")]; + tensor var_2160_begin_0 = const()[name = string("op_2160_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2160_end_0 = const()[name = string("op_2160_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2160_end_mask_0 = const()[name = string("op_2160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = query_5_cast_fp16)[name = string("op_2160_cast_fp16")]; + tensor var_2164_begin_0 = const()[name = string("op_2164_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2164_end_0 = const()[name = string("op_2164_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2164_end_mask_0 = const()[name = string("op_2164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = query_5_cast_fp16)[name = string("op_2164_cast_fp16")]; + tensor var_2168_begin_0 = const()[name = string("op_2168_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2168_end_0 = const()[name = string("op_2168_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2168_end_mask_0 = const()[name = string("op_2168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2168_cast_fp16 = slice_by_index(begin = var_2168_begin_0, end = var_2168_end_0, end_mask = var_2168_end_mask_0, x = query_5_cast_fp16)[name = string("op_2168_cast_fp16")]; + tensor var_2172_begin_0 = const()[name = string("op_2172_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2172_end_0 = const()[name = string("op_2172_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2172_end_mask_0 = const()[name = string("op_2172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2172_cast_fp16 = slice_by_index(begin = var_2172_begin_0, end = var_2172_end_0, end_mask = var_2172_end_mask_0, x = query_5_cast_fp16)[name = string("op_2172_cast_fp16")]; + tensor var_2176_begin_0 = const()[name = string("op_2176_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2176_end_0 = const()[name = string("op_2176_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2176_end_mask_0 = const()[name = string("op_2176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2176_cast_fp16 = slice_by_index(begin = var_2176_begin_0, end = var_2176_end_0, end_mask = var_2176_end_mask_0, x = query_5_cast_fp16)[name = string("op_2176_cast_fp16")]; + tensor var_2180_begin_0 = const()[name = string("op_2180_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2180_end_0 = const()[name = string("op_2180_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2180_end_mask_0 = const()[name = string("op_2180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2180_cast_fp16 = slice_by_index(begin = var_2180_begin_0, end = var_2180_end_0, end_mask = var_2180_end_mask_0, x = query_5_cast_fp16)[name = string("op_2180_cast_fp16")]; + tensor var_2184_begin_0 = const()[name = string("op_2184_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2184_end_0 = const()[name = string("op_2184_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2184_end_mask_0 = const()[name = string("op_2184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2184_cast_fp16 = slice_by_index(begin = var_2184_begin_0, end = var_2184_end_0, end_mask = var_2184_end_mask_0, x = query_5_cast_fp16)[name = string("op_2184_cast_fp16")]; + tensor var_2188_begin_0 = const()[name = string("op_2188_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2188_end_0 = const()[name = string("op_2188_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2188_end_mask_0 = const()[name = string("op_2188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2188_cast_fp16 = slice_by_index(begin = var_2188_begin_0, end = var_2188_end_0, end_mask = var_2188_end_mask_0, x = query_5_cast_fp16)[name = string("op_2188_cast_fp16")]; + tensor var_2192_begin_0 = const()[name = string("op_2192_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2192_end_0 = const()[name = string("op_2192_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2192_end_mask_0 = const()[name = string("op_2192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2192_cast_fp16 = slice_by_index(begin = var_2192_begin_0, end = var_2192_end_0, end_mask = var_2192_end_mask_0, x = query_5_cast_fp16)[name = string("op_2192_cast_fp16")]; + tensor var_2201_begin_0 = const()[name = string("op_2201_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2201_end_0 = const()[name = string("op_2201_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2201_end_mask_0 = const()[name = string("op_2201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2201_cast_fp16 = slice_by_index(begin = var_2201_begin_0, end = var_2201_end_0, end_mask = var_2201_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2201_cast_fp16")]; + tensor var_2208_begin_0 = const()[name = string("op_2208_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2208_end_0 = const()[name = string("op_2208_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2208_end_mask_0 = const()[name = string("op_2208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2208_cast_fp16 = slice_by_index(begin = var_2208_begin_0, end = var_2208_end_0, end_mask = var_2208_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2208_cast_fp16")]; + tensor var_2215_begin_0 = const()[name = string("op_2215_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2215_end_0 = const()[name = string("op_2215_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2215_end_mask_0 = const()[name = string("op_2215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2215_cast_fp16 = slice_by_index(begin = var_2215_begin_0, end = var_2215_end_0, end_mask = var_2215_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2215_cast_fp16")]; + tensor var_2222_begin_0 = const()[name = string("op_2222_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2222_end_0 = const()[name = string("op_2222_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2222_end_mask_0 = const()[name = string("op_2222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2222_cast_fp16 = slice_by_index(begin = var_2222_begin_0, end = var_2222_end_0, end_mask = var_2222_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2222_cast_fp16")]; + tensor var_2229_begin_0 = const()[name = string("op_2229_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2229_end_0 = const()[name = string("op_2229_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2229_end_mask_0 = const()[name = string("op_2229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2229_cast_fp16 = slice_by_index(begin = var_2229_begin_0, end = var_2229_end_0, end_mask = var_2229_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2229_cast_fp16")]; + tensor var_2236_begin_0 = const()[name = string("op_2236_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2236_end_0 = const()[name = string("op_2236_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2236_end_mask_0 = const()[name = string("op_2236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2236_cast_fp16 = slice_by_index(begin = var_2236_begin_0, end = var_2236_end_0, end_mask = var_2236_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2236_cast_fp16")]; + tensor var_2243_begin_0 = const()[name = string("op_2243_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2243_end_0 = const()[name = string("op_2243_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2243_end_mask_0 = const()[name = string("op_2243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2243_cast_fp16")]; + tensor var_2250_begin_0 = const()[name = string("op_2250_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2250_end_0 = const()[name = string("op_2250_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2250_end_mask_0 = const()[name = string("op_2250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2250_cast_fp16 = slice_by_index(begin = var_2250_begin_0, end = var_2250_end_0, end_mask = var_2250_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2250_cast_fp16")]; + tensor var_2257_begin_0 = const()[name = string("op_2257_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2257_end_0 = const()[name = string("op_2257_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2257_end_mask_0 = const()[name = string("op_2257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2257_cast_fp16 = slice_by_index(begin = var_2257_begin_0, end = var_2257_end_0, end_mask = var_2257_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2257_cast_fp16")]; + tensor var_2264_begin_0 = const()[name = string("op_2264_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2264_end_0 = const()[name = string("op_2264_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2264_end_mask_0 = const()[name = string("op_2264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2264_cast_fp16 = slice_by_index(begin = var_2264_begin_0, end = var_2264_end_0, end_mask = var_2264_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2264_cast_fp16")]; + tensor var_2271_begin_0 = const()[name = string("op_2271_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2271_end_0 = const()[name = string("op_2271_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2271_end_mask_0 = const()[name = string("op_2271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2271_cast_fp16 = slice_by_index(begin = var_2271_begin_0, end = var_2271_end_0, end_mask = var_2271_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2271_cast_fp16")]; + tensor var_2278_begin_0 = const()[name = string("op_2278_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2278_end_0 = const()[name = string("op_2278_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2278_end_mask_0 = const()[name = string("op_2278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2278_cast_fp16 = slice_by_index(begin = var_2278_begin_0, end = var_2278_end_0, end_mask = var_2278_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2278_cast_fp16")]; + tensor var_2285_begin_0 = const()[name = string("op_2285_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2285_end_0 = const()[name = string("op_2285_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2285_end_mask_0 = const()[name = string("op_2285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2285_cast_fp16 = slice_by_index(begin = var_2285_begin_0, end = var_2285_end_0, end_mask = var_2285_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2285_cast_fp16")]; + tensor var_2292_begin_0 = const()[name = string("op_2292_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2292_end_0 = const()[name = string("op_2292_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2292_end_mask_0 = const()[name = string("op_2292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2292_cast_fp16 = slice_by_index(begin = var_2292_begin_0, end = var_2292_end_0, end_mask = var_2292_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2292_cast_fp16")]; + tensor var_2299_begin_0 = const()[name = string("op_2299_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2299_end_0 = const()[name = string("op_2299_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2299_end_mask_0 = const()[name = string("op_2299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2299_cast_fp16 = slice_by_index(begin = var_2299_begin_0, end = var_2299_end_0, end_mask = var_2299_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2299_cast_fp16")]; + tensor var_2306_begin_0 = const()[name = string("op_2306_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2306_end_0 = const()[name = string("op_2306_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2306_end_mask_0 = const()[name = string("op_2306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2306_cast_fp16 = slice_by_index(begin = var_2306_begin_0, end = var_2306_end_0, end_mask = var_2306_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2306_cast_fp16")]; + tensor var_2313_begin_0 = const()[name = string("op_2313_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2313_end_0 = const()[name = string("op_2313_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2313_end_mask_0 = const()[name = string("op_2313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2313_cast_fp16 = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2313_cast_fp16")]; + tensor var_2320_begin_0 = const()[name = string("op_2320_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2320_end_0 = const()[name = string("op_2320_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2320_end_mask_0 = const()[name = string("op_2320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2320_cast_fp16 = slice_by_index(begin = var_2320_begin_0, end = var_2320_end_0, end_mask = var_2320_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2320_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = string("op_2327_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2327_end_0 = const()[name = string("op_2327_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2327_end_mask_0 = const()[name = string("op_2327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2327_cast_fp16")]; + tensor var_2334_begin_0 = const()[name = string("op_2334_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2334_end_0 = const()[name = string("op_2334_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2334_end_mask_0 = const()[name = string("op_2334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2334_cast_fp16 = slice_by_index(begin = var_2334_begin_0, end = var_2334_end_0, end_mask = var_2334_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2334_cast_fp16")]; + tensor var_2341_begin_0 = const()[name = string("op_2341_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2341_end_0 = const()[name = string("op_2341_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2341_end_mask_0 = const()[name = string("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2341_cast_fp16 = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2341_cast_fp16")]; + tensor var_2348_begin_0 = const()[name = string("op_2348_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2348_end_0 = const()[name = string("op_2348_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2348_end_mask_0 = const()[name = string("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2348_cast_fp16 = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2348_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = string("op_2355_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2355_end_0 = const()[name = string("op_2355_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2355_end_mask_0 = const()[name = string("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2355_cast_fp16")]; + tensor var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2362_cast_fp16")]; + tensor var_2369_begin_0 = const()[name = string("op_2369_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2369_end_0 = const()[name = string("op_2369_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2369_end_mask_0 = const()[name = string("op_2369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2369_cast_fp16 = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2369_cast_fp16")]; + tensor var_2376_begin_0 = const()[name = string("op_2376_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2376_end_0 = const()[name = string("op_2376_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2376_end_mask_0 = const()[name = string("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2376_cast_fp16 = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2376_cast_fp16")]; + tensor var_2383_begin_0 = const()[name = string("op_2383_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2383_end_0 = const()[name = string("op_2383_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2383_end_mask_0 = const()[name = string("op_2383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2383_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = string("op_2390_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2390_end_0 = const()[name = string("op_2390_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2390_end_mask_0 = const()[name = string("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2390_cast_fp16")]; + tensor var_2397_begin_0 = const()[name = string("op_2397_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2397_end_0 = const()[name = string("op_2397_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2397_end_mask_0 = const()[name = string("op_2397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2397_cast_fp16 = slice_by_index(begin = var_2397_begin_0, end = var_2397_end_0, end_mask = var_2397_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2397_cast_fp16")]; + tensor var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2404_cast_fp16")]; + tensor var_2411_begin_0 = const()[name = string("op_2411_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2411_end_0 = const()[name = string("op_2411_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2411_end_mask_0 = const()[name = string("op_2411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2411_cast_fp16 = slice_by_index(begin = var_2411_begin_0, end = var_2411_end_0, end_mask = var_2411_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2411_cast_fp16")]; + tensor var_2418_begin_0 = const()[name = string("op_2418_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2418_end_0 = const()[name = string("op_2418_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2418_end_mask_0 = const()[name = string("op_2418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2418_cast_fp16")]; + tensor var_2425_begin_0 = const()[name = string("op_2425_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2425_end_0 = const()[name = string("op_2425_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2425_end_mask_0 = const()[name = string("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2425_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2439_begin_0 = const()[name = string("op_2439_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2439_end_0 = const()[name = string("op_2439_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2439_end_mask_0 = const()[name = string("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2439_cast_fp16")]; + tensor var_2446_begin_0 = const()[name = string("op_2446_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2446_end_0 = const()[name = string("op_2446_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2446_end_mask_0 = const()[name = string("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2446_cast_fp16")]; + tensor var_2453_begin_0 = const()[name = string("op_2453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2453_end_0 = const()[name = string("op_2453_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2453_end_mask_0 = const()[name = string("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2453_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2467_begin_0 = const()[name = string("op_2467_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2467_end_0 = const()[name = string("op_2467_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2467_end_mask_0 = const()[name = string("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2467_cast_fp16")]; + tensor var_2474_begin_0 = const()[name = string("op_2474_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2474_end_0 = const()[name = string("op_2474_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2474_end_mask_0 = const()[name = string("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2474_cast_fp16")]; + tensor var_2481_begin_0 = const()[name = string("op_2481_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2481_end_0 = const()[name = string("op_2481_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2481_end_mask_0 = const()[name = string("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2481_cast_fp16")]; + tensor var_2488_begin_0 = const()[name = string("op_2488_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2488_end_0 = const()[name = string("op_2488_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2488_end_mask_0 = const()[name = string("op_2488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2488_cast_fp16 = slice_by_index(begin = var_2488_begin_0, end = var_2488_end_0, end_mask = var_2488_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2488_cast_fp16")]; + tensor var_2495_begin_0 = const()[name = string("op_2495_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2495_end_0 = const()[name = string("op_2495_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2495_end_mask_0 = const()[name = string("op_2495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2495_cast_fp16 = slice_by_index(begin = var_2495_begin_0, end = var_2495_end_0, end_mask = var_2495_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2495_cast_fp16")]; + tensor var_2502_begin_0 = const()[name = string("op_2502_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2502_end_0 = const()[name = string("op_2502_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2502_end_mask_0 = const()[name = string("op_2502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2502_cast_fp16")]; + tensor var_2509_begin_0 = const()[name = string("op_2509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2509_end_0 = const()[name = string("op_2509_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2509_end_mask_0 = const()[name = string("op_2509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2509_cast_fp16 = slice_by_index(begin = var_2509_begin_0, end = var_2509_end_0, end_mask = var_2509_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2509_cast_fp16")]; + tensor var_2516_begin_0 = const()[name = string("op_2516_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2516_end_0 = const()[name = string("op_2516_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2516_end_mask_0 = const()[name = string("op_2516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2516_cast_fp16 = slice_by_index(begin = var_2516_begin_0, end = var_2516_end_0, end_mask = var_2516_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2516_cast_fp16")]; + tensor var_2523_begin_0 = const()[name = string("op_2523_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2523_end_0 = const()[name = string("op_2523_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2523_end_mask_0 = const()[name = string("op_2523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2523_cast_fp16 = slice_by_index(begin = var_2523_begin_0, end = var_2523_end_0, end_mask = var_2523_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2523_cast_fp16")]; + tensor var_2530_begin_0 = const()[name = string("op_2530_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2530_end_0 = const()[name = string("op_2530_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2530_end_mask_0 = const()[name = string("op_2530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2530_cast_fp16 = slice_by_index(begin = var_2530_begin_0, end = var_2530_end_0, end_mask = var_2530_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2530_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2535_begin_0 = const()[name = string("op_2535_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2535_end_0 = const()[name = string("op_2535_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2535_end_mask_0 = const()[name = string("op_2535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_9")]; + tensor var_2535_cast_fp16 = slice_by_index(begin = var_2535_begin_0, end = var_2535_end_0, end_mask = var_2535_end_mask_0, x = k_5_cast_fp16)[name = string("op_2535_cast_fp16")]; + tensor var_2539_begin_0 = const()[name = string("op_2539_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2539_end_0 = const()[name = string("op_2539_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2539_end_mask_0 = const()[name = string("op_2539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2539_cast_fp16 = slice_by_index(begin = var_2539_begin_0, end = var_2539_end_0, end_mask = var_2539_end_mask_0, x = k_5_cast_fp16)[name = string("op_2539_cast_fp16")]; + tensor var_2543_begin_0 = const()[name = string("op_2543_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2543_end_0 = const()[name = string("op_2543_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2543_end_mask_0 = const()[name = string("op_2543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2543_cast_fp16 = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = k_5_cast_fp16)[name = string("op_2543_cast_fp16")]; + tensor var_2547_begin_0 = const()[name = string("op_2547_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2547_end_0 = const()[name = string("op_2547_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2547_end_mask_0 = const()[name = string("op_2547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2547_cast_fp16 = slice_by_index(begin = var_2547_begin_0, end = var_2547_end_0, end_mask = var_2547_end_mask_0, x = k_5_cast_fp16)[name = string("op_2547_cast_fp16")]; + tensor var_2551_begin_0 = const()[name = string("op_2551_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2551_end_0 = const()[name = string("op_2551_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2551_end_mask_0 = const()[name = string("op_2551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2551_cast_fp16 = slice_by_index(begin = var_2551_begin_0, end = var_2551_end_0, end_mask = var_2551_end_mask_0, x = k_5_cast_fp16)[name = string("op_2551_cast_fp16")]; + tensor var_2555_begin_0 = const()[name = string("op_2555_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2555_end_0 = const()[name = string("op_2555_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2555_end_mask_0 = const()[name = string("op_2555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2555_cast_fp16 = slice_by_index(begin = var_2555_begin_0, end = var_2555_end_0, end_mask = var_2555_end_mask_0, x = k_5_cast_fp16)[name = string("op_2555_cast_fp16")]; + tensor var_2559_begin_0 = const()[name = string("op_2559_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2559_end_0 = const()[name = string("op_2559_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2559_end_mask_0 = const()[name = string("op_2559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2559_cast_fp16 = slice_by_index(begin = var_2559_begin_0, end = var_2559_end_0, end_mask = var_2559_end_mask_0, x = k_5_cast_fp16)[name = string("op_2559_cast_fp16")]; + tensor var_2563_begin_0 = const()[name = string("op_2563_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2563_end_0 = const()[name = string("op_2563_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2563_end_mask_0 = const()[name = string("op_2563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2563_cast_fp16 = slice_by_index(begin = var_2563_begin_0, end = var_2563_end_0, end_mask = var_2563_end_mask_0, x = k_5_cast_fp16)[name = string("op_2563_cast_fp16")]; + tensor var_2567_begin_0 = const()[name = string("op_2567_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_2567_end_mask_0 = const()[name = string("op_2567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = k_5_cast_fp16)[name = string("op_2567_cast_fp16")]; + tensor var_2571_begin_0 = const()[name = string("op_2571_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2571_end_0 = const()[name = string("op_2571_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_2571_end_mask_0 = const()[name = string("op_2571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2571_cast_fp16 = slice_by_index(begin = var_2571_begin_0, end = var_2571_end_0, end_mask = var_2571_end_mask_0, x = k_5_cast_fp16)[name = string("op_2571_cast_fp16")]; + tensor var_2575_begin_0 = const()[name = string("op_2575_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2575_end_0 = const()[name = string("op_2575_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_2575_end_mask_0 = const()[name = string("op_2575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2575_cast_fp16 = slice_by_index(begin = var_2575_begin_0, end = var_2575_end_0, end_mask = var_2575_end_mask_0, x = k_5_cast_fp16)[name = string("op_2575_cast_fp16")]; + tensor var_2579_begin_0 = const()[name = string("op_2579_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2579_end_0 = const()[name = string("op_2579_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_2579_end_mask_0 = const()[name = string("op_2579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2579_cast_fp16 = slice_by_index(begin = var_2579_begin_0, end = var_2579_end_0, end_mask = var_2579_end_mask_0, x = k_5_cast_fp16)[name = string("op_2579_cast_fp16")]; + tensor var_2581_begin_0 = const()[name = string("op_2581_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2581_end_0 = const()[name = string("op_2581_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2581_end_mask_0 = const()[name = string("op_2581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2581_cast_fp16 = slice_by_index(begin = var_2581_begin_0, end = var_2581_end_0, end_mask = var_2581_end_mask_0, x = value_5_cast_fp16)[name = string("op_2581_cast_fp16")]; + tensor var_2585_begin_0 = const()[name = string("op_2585_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2585_end_0 = const()[name = string("op_2585_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2585_end_mask_0 = const()[name = string("op_2585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2585_cast_fp16 = slice_by_index(begin = var_2585_begin_0, end = var_2585_end_0, end_mask = var_2585_end_mask_0, x = value_5_cast_fp16)[name = string("op_2585_cast_fp16")]; + tensor var_2589_begin_0 = const()[name = string("op_2589_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2589_end_0 = const()[name = string("op_2589_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2589_end_mask_0 = const()[name = string("op_2589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2589_cast_fp16 = slice_by_index(begin = var_2589_begin_0, end = var_2589_end_0, end_mask = var_2589_end_mask_0, x = value_5_cast_fp16)[name = string("op_2589_cast_fp16")]; + tensor var_2593_begin_0 = const()[name = string("op_2593_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2593_end_0 = const()[name = string("op_2593_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2593_end_mask_0 = const()[name = string("op_2593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2593_cast_fp16 = slice_by_index(begin = var_2593_begin_0, end = var_2593_end_0, end_mask = var_2593_end_mask_0, x = value_5_cast_fp16)[name = string("op_2593_cast_fp16")]; + tensor var_2597_begin_0 = const()[name = string("op_2597_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2597_end_0 = const()[name = string("op_2597_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2597_end_mask_0 = const()[name = string("op_2597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2597_cast_fp16 = slice_by_index(begin = var_2597_begin_0, end = var_2597_end_0, end_mask = var_2597_end_mask_0, x = value_5_cast_fp16)[name = string("op_2597_cast_fp16")]; + tensor var_2601_begin_0 = const()[name = string("op_2601_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2601_end_0 = const()[name = string("op_2601_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2601_end_mask_0 = const()[name = string("op_2601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2601_cast_fp16 = slice_by_index(begin = var_2601_begin_0, end = var_2601_end_0, end_mask = var_2601_end_mask_0, x = value_5_cast_fp16)[name = string("op_2601_cast_fp16")]; + tensor var_2605_begin_0 = const()[name = string("op_2605_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2605_end_0 = const()[name = string("op_2605_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2605_end_mask_0 = const()[name = string("op_2605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2605_cast_fp16 = slice_by_index(begin = var_2605_begin_0, end = var_2605_end_0, end_mask = var_2605_end_mask_0, x = value_5_cast_fp16)[name = string("op_2605_cast_fp16")]; + tensor var_2609_begin_0 = const()[name = string("op_2609_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2609_end_0 = const()[name = string("op_2609_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2609_end_mask_0 = const()[name = string("op_2609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2609_cast_fp16 = slice_by_index(begin = var_2609_begin_0, end = var_2609_end_0, end_mask = var_2609_end_mask_0, x = value_5_cast_fp16)[name = string("op_2609_cast_fp16")]; + tensor var_2613_begin_0 = const()[name = string("op_2613_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2613_end_0 = const()[name = string("op_2613_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2613_end_mask_0 = const()[name = string("op_2613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2613_cast_fp16 = slice_by_index(begin = var_2613_begin_0, end = var_2613_end_0, end_mask = var_2613_end_mask_0, x = value_5_cast_fp16)[name = string("op_2613_cast_fp16")]; + tensor var_2617_begin_0 = const()[name = string("op_2617_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2617_end_0 = const()[name = string("op_2617_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2617_end_mask_0 = const()[name = string("op_2617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2617_cast_fp16 = slice_by_index(begin = var_2617_begin_0, end = var_2617_end_0, end_mask = var_2617_end_mask_0, x = value_5_cast_fp16)[name = string("op_2617_cast_fp16")]; + tensor var_2621_begin_0 = const()[name = string("op_2621_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2621_end_0 = const()[name = string("op_2621_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2621_end_mask_0 = const()[name = string("op_2621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2621_cast_fp16 = slice_by_index(begin = var_2621_begin_0, end = var_2621_end_0, end_mask = var_2621_end_mask_0, x = value_5_cast_fp16)[name = string("op_2621_cast_fp16")]; + tensor var_2625_begin_0 = const()[name = string("op_2625_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2625_end_0 = const()[name = string("op_2625_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2625_end_mask_0 = const()[name = string("op_2625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2625_cast_fp16 = slice_by_index(begin = var_2625_begin_0, end = var_2625_end_0, end_mask = var_2625_end_mask_0, x = value_5_cast_fp16)[name = string("op_2625_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2535_cast_fp16, var_2201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2535_cast_fp16, var_2208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2535_cast_fp16, var_2215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2535_cast_fp16, var_2222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2539_cast_fp16, var_2229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2539_cast_fp16, var_2236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2539_cast_fp16, var_2243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2539_cast_fp16, var_2250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2543_cast_fp16, var_2257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2543_cast_fp16, var_2264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2543_cast_fp16, var_2271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2543_cast_fp16, var_2278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2547_cast_fp16, var_2285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2547_cast_fp16, var_2292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2547_cast_fp16, var_2299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2547_cast_fp16, var_2306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2551_cast_fp16, var_2313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2551_cast_fp16, var_2320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2551_cast_fp16, var_2327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2551_cast_fp16, var_2334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2555_cast_fp16, var_2341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2555_cast_fp16, var_2348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2555_cast_fp16, var_2355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2555_cast_fp16, var_2362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2559_cast_fp16, var_2369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2559_cast_fp16, var_2376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2559_cast_fp16, var_2383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2559_cast_fp16, var_2390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2563_cast_fp16, var_2397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2563_cast_fp16, var_2404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2563_cast_fp16, var_2411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2563_cast_fp16, var_2418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_2567_cast_fp16, var_2425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_2567_cast_fp16, var_2432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_2567_cast_fp16, var_2439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_2567_cast_fp16, var_2446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_2571_cast_fp16, var_2453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_2571_cast_fp16, var_2460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_2571_cast_fp16, var_2467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_2571_cast_fp16, var_2474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_2575_cast_fp16, var_2481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_2575_cast_fp16, var_2488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_2575_cast_fp16, var_2495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_2575_cast_fp16, var_2502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_2579_cast_fp16, var_2509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_2579_cast_fp16, var_2516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_2579_cast_fp16, var_2523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_2579_cast_fp16, var_2530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + fp16 var_2724_to_fp16 = const()[name = string("op_2724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2724_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2726_to_fp16 = const()[name = string("op_2726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2726_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2728_to_fp16 = const()[name = string("op_2728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2728_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2730_to_fp16 = const()[name = string("op_2730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2730_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2732_to_fp16 = const()[name = string("op_2732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2732_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2734_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2736_to_fp16 = const()[name = string("op_2736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2736_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2738_to_fp16 = const()[name = string("op_2738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2738_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2740_to_fp16 = const()[name = string("op_2740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2740_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2742_to_fp16 = const()[name = string("op_2742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2742_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2744_to_fp16 = const()[name = string("op_2744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2744_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2746_to_fp16 = const()[name = string("op_2746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2746_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2748_to_fp16 = const()[name = string("op_2748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2748_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2750_to_fp16 = const()[name = string("op_2750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2750_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2752_to_fp16 = const()[name = string("op_2752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2752_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2754_to_fp16 = const()[name = string("op_2754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2754_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2756_to_fp16 = const()[name = string("op_2756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2756_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2758_to_fp16 = const()[name = string("op_2758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2758_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2760_to_fp16 = const()[name = string("op_2760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2760_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2762_to_fp16 = const()[name = string("op_2762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2762_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2764_to_fp16 = const()[name = string("op_2764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2764_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2766_to_fp16 = const()[name = string("op_2766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2766_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2768_to_fp16 = const()[name = string("op_2768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2768_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2770_to_fp16 = const()[name = string("op_2770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2770_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2772_to_fp16 = const()[name = string("op_2772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2772_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2774_to_fp16 = const()[name = string("op_2774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2774_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2776_to_fp16 = const()[name = string("op_2776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2776_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2778_to_fp16 = const()[name = string("op_2778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2778_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2780_to_fp16 = const()[name = string("op_2780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2780_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2782_to_fp16 = const()[name = string("op_2782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2782_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2784_to_fp16 = const()[name = string("op_2784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2784_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2786_to_fp16 = const()[name = string("op_2786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2786_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + fp16 var_2788_to_fp16 = const()[name = string("op_2788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_2788_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_2790_to_fp16 = const()[name = string("op_2790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_2790_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_2792_to_fp16 = const()[name = string("op_2792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_2792_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_2794_to_fp16 = const()[name = string("op_2794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_2794_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_2796_to_fp16 = const()[name = string("op_2796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_2796_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_2798_to_fp16 = const()[name = string("op_2798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_2798_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_2800_to_fp16 = const()[name = string("op_2800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_2800_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_2802_to_fp16 = const()[name = string("op_2802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_2802_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_2804_to_fp16 = const()[name = string("op_2804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_2804_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_2806_to_fp16 = const()[name = string("op_2806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_2806_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_2808_to_fp16 = const()[name = string("op_2808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_2808_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_2810_to_fp16 = const()[name = string("op_2810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_2810_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_2812_to_fp16 = const()[name = string("op_2812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_2812_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_2814_to_fp16 = const()[name = string("op_2814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_2814_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_2816_to_fp16 = const()[name = string("op_2816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_2816_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_2818_to_fp16 = const()[name = string("op_2818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_2818_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + tensor var_2820_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_193_cast_fp16)[name = string("op_2820_cast_fp16")]; + tensor var_2821_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_195_cast_fp16)[name = string("op_2821_cast_fp16")]; + tensor var_2822_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_197_cast_fp16)[name = string("op_2822_cast_fp16")]; + tensor var_2823_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_199_cast_fp16)[name = string("op_2823_cast_fp16")]; + tensor var_2824_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_201_cast_fp16)[name = string("op_2824_cast_fp16")]; + tensor var_2825_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_203_cast_fp16)[name = string("op_2825_cast_fp16")]; + tensor var_2826_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_205_cast_fp16)[name = string("op_2826_cast_fp16")]; + tensor var_2827_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_207_cast_fp16)[name = string("op_2827_cast_fp16")]; + tensor var_2828_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_209_cast_fp16)[name = string("op_2828_cast_fp16")]; + tensor var_2829_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_211_cast_fp16)[name = string("op_2829_cast_fp16")]; + tensor var_2830_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_213_cast_fp16)[name = string("op_2830_cast_fp16")]; + tensor var_2831_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_215_cast_fp16)[name = string("op_2831_cast_fp16")]; + tensor var_2832_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_217_cast_fp16)[name = string("op_2832_cast_fp16")]; + tensor var_2833_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_219_cast_fp16)[name = string("op_2833_cast_fp16")]; + tensor var_2834_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_221_cast_fp16)[name = string("op_2834_cast_fp16")]; + tensor var_2835_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_223_cast_fp16)[name = string("op_2835_cast_fp16")]; + tensor var_2836_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_225_cast_fp16)[name = string("op_2836_cast_fp16")]; + tensor var_2837_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_227_cast_fp16)[name = string("op_2837_cast_fp16")]; + tensor var_2838_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_229_cast_fp16)[name = string("op_2838_cast_fp16")]; + tensor var_2839_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_231_cast_fp16)[name = string("op_2839_cast_fp16")]; + tensor var_2840_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_233_cast_fp16)[name = string("op_2840_cast_fp16")]; + tensor var_2841_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_235_cast_fp16)[name = string("op_2841_cast_fp16")]; + tensor var_2842_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_237_cast_fp16)[name = string("op_2842_cast_fp16")]; + tensor var_2843_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_239_cast_fp16)[name = string("op_2843_cast_fp16")]; + tensor var_2844_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_241_cast_fp16)[name = string("op_2844_cast_fp16")]; + tensor var_2845_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_243_cast_fp16)[name = string("op_2845_cast_fp16")]; + tensor var_2846_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_245_cast_fp16)[name = string("op_2846_cast_fp16")]; + tensor var_2847_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_247_cast_fp16)[name = string("op_2847_cast_fp16")]; + tensor var_2848_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_249_cast_fp16)[name = string("op_2848_cast_fp16")]; + tensor var_2849_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_251_cast_fp16)[name = string("op_2849_cast_fp16")]; + tensor var_2850_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_253_cast_fp16)[name = string("op_2850_cast_fp16")]; + tensor var_2851_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_255_cast_fp16)[name = string("op_2851_cast_fp16")]; + tensor var_2852_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_257_cast_fp16)[name = string("op_2852_cast_fp16")]; + tensor var_2853_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_259_cast_fp16)[name = string("op_2853_cast_fp16")]; + tensor var_2854_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_261_cast_fp16)[name = string("op_2854_cast_fp16")]; + tensor var_2855_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_263_cast_fp16)[name = string("op_2855_cast_fp16")]; + tensor var_2856_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_265_cast_fp16)[name = string("op_2856_cast_fp16")]; + tensor var_2857_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_267_cast_fp16)[name = string("op_2857_cast_fp16")]; + tensor var_2858_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_269_cast_fp16)[name = string("op_2858_cast_fp16")]; + tensor var_2859_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_271_cast_fp16)[name = string("op_2859_cast_fp16")]; + tensor var_2860_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_273_cast_fp16)[name = string("op_2860_cast_fp16")]; + tensor var_2861_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_275_cast_fp16)[name = string("op_2861_cast_fp16")]; + tensor var_2862_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_277_cast_fp16)[name = string("op_2862_cast_fp16")]; + tensor var_2863_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_279_cast_fp16)[name = string("op_2863_cast_fp16")]; + tensor var_2864_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_281_cast_fp16)[name = string("op_2864_cast_fp16")]; + tensor var_2865_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_283_cast_fp16)[name = string("op_2865_cast_fp16")]; + tensor var_2866_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_285_cast_fp16)[name = string("op_2866_cast_fp16")]; + tensor var_2867_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_287_cast_fp16)[name = string("op_2867_cast_fp16")]; + string var_2869_equation_0 = const()[name = string("op_2869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2869_cast_fp16 = einsum(equation = var_2869_equation_0, values = (var_2581_cast_fp16, var_2820_cast_fp16))[name = string("op_2869_cast_fp16")]; + string var_2871_equation_0 = const()[name = string("op_2871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2871_cast_fp16 = einsum(equation = var_2871_equation_0, values = (var_2581_cast_fp16, var_2821_cast_fp16))[name = string("op_2871_cast_fp16")]; + string var_2873_equation_0 = const()[name = string("op_2873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2873_cast_fp16 = einsum(equation = var_2873_equation_0, values = (var_2581_cast_fp16, var_2822_cast_fp16))[name = string("op_2873_cast_fp16")]; + string var_2875_equation_0 = const()[name = string("op_2875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2875_cast_fp16 = einsum(equation = var_2875_equation_0, values = (var_2581_cast_fp16, var_2823_cast_fp16))[name = string("op_2875_cast_fp16")]; + string var_2877_equation_0 = const()[name = string("op_2877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2877_cast_fp16 = einsum(equation = var_2877_equation_0, values = (var_2585_cast_fp16, var_2824_cast_fp16))[name = string("op_2877_cast_fp16")]; + string var_2879_equation_0 = const()[name = string("op_2879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2879_cast_fp16 = einsum(equation = var_2879_equation_0, values = (var_2585_cast_fp16, var_2825_cast_fp16))[name = string("op_2879_cast_fp16")]; + string var_2881_equation_0 = const()[name = string("op_2881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2881_cast_fp16 = einsum(equation = var_2881_equation_0, values = (var_2585_cast_fp16, var_2826_cast_fp16))[name = string("op_2881_cast_fp16")]; + string var_2883_equation_0 = const()[name = string("op_2883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2883_cast_fp16 = einsum(equation = var_2883_equation_0, values = (var_2585_cast_fp16, var_2827_cast_fp16))[name = string("op_2883_cast_fp16")]; + string var_2885_equation_0 = const()[name = string("op_2885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2885_cast_fp16 = einsum(equation = var_2885_equation_0, values = (var_2589_cast_fp16, var_2828_cast_fp16))[name = string("op_2885_cast_fp16")]; + string var_2887_equation_0 = const()[name = string("op_2887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2887_cast_fp16 = einsum(equation = var_2887_equation_0, values = (var_2589_cast_fp16, var_2829_cast_fp16))[name = string("op_2887_cast_fp16")]; + string var_2889_equation_0 = const()[name = string("op_2889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2889_cast_fp16 = einsum(equation = var_2889_equation_0, values = (var_2589_cast_fp16, var_2830_cast_fp16))[name = string("op_2889_cast_fp16")]; + string var_2891_equation_0 = const()[name = string("op_2891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2891_cast_fp16 = einsum(equation = var_2891_equation_0, values = (var_2589_cast_fp16, var_2831_cast_fp16))[name = string("op_2891_cast_fp16")]; + string var_2893_equation_0 = const()[name = string("op_2893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2893_cast_fp16 = einsum(equation = var_2893_equation_0, values = (var_2593_cast_fp16, var_2832_cast_fp16))[name = string("op_2893_cast_fp16")]; + string var_2895_equation_0 = const()[name = string("op_2895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2895_cast_fp16 = einsum(equation = var_2895_equation_0, values = (var_2593_cast_fp16, var_2833_cast_fp16))[name = string("op_2895_cast_fp16")]; + string var_2897_equation_0 = const()[name = string("op_2897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2897_cast_fp16 = einsum(equation = var_2897_equation_0, values = (var_2593_cast_fp16, var_2834_cast_fp16))[name = string("op_2897_cast_fp16")]; + string var_2899_equation_0 = const()[name = string("op_2899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2899_cast_fp16 = einsum(equation = var_2899_equation_0, values = (var_2593_cast_fp16, var_2835_cast_fp16))[name = string("op_2899_cast_fp16")]; + string var_2901_equation_0 = const()[name = string("op_2901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2901_cast_fp16 = einsum(equation = var_2901_equation_0, values = (var_2597_cast_fp16, var_2836_cast_fp16))[name = string("op_2901_cast_fp16")]; + string var_2903_equation_0 = const()[name = string("op_2903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2903_cast_fp16 = einsum(equation = var_2903_equation_0, values = (var_2597_cast_fp16, var_2837_cast_fp16))[name = string("op_2903_cast_fp16")]; + string var_2905_equation_0 = const()[name = string("op_2905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2905_cast_fp16 = einsum(equation = var_2905_equation_0, values = (var_2597_cast_fp16, var_2838_cast_fp16))[name = string("op_2905_cast_fp16")]; + string var_2907_equation_0 = const()[name = string("op_2907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2907_cast_fp16 = einsum(equation = var_2907_equation_0, values = (var_2597_cast_fp16, var_2839_cast_fp16))[name = string("op_2907_cast_fp16")]; + string var_2909_equation_0 = const()[name = string("op_2909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2909_cast_fp16 = einsum(equation = var_2909_equation_0, values = (var_2601_cast_fp16, var_2840_cast_fp16))[name = string("op_2909_cast_fp16")]; + string var_2911_equation_0 = const()[name = string("op_2911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2911_cast_fp16 = einsum(equation = var_2911_equation_0, values = (var_2601_cast_fp16, var_2841_cast_fp16))[name = string("op_2911_cast_fp16")]; + string var_2913_equation_0 = const()[name = string("op_2913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2913_cast_fp16 = einsum(equation = var_2913_equation_0, values = (var_2601_cast_fp16, var_2842_cast_fp16))[name = string("op_2913_cast_fp16")]; + string var_2915_equation_0 = const()[name = string("op_2915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2915_cast_fp16 = einsum(equation = var_2915_equation_0, values = (var_2601_cast_fp16, var_2843_cast_fp16))[name = string("op_2915_cast_fp16")]; + string var_2917_equation_0 = const()[name = string("op_2917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2917_cast_fp16 = einsum(equation = var_2917_equation_0, values = (var_2605_cast_fp16, var_2844_cast_fp16))[name = string("op_2917_cast_fp16")]; + string var_2919_equation_0 = const()[name = string("op_2919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2919_cast_fp16 = einsum(equation = var_2919_equation_0, values = (var_2605_cast_fp16, var_2845_cast_fp16))[name = string("op_2919_cast_fp16")]; + string var_2921_equation_0 = const()[name = string("op_2921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2921_cast_fp16 = einsum(equation = var_2921_equation_0, values = (var_2605_cast_fp16, var_2846_cast_fp16))[name = string("op_2921_cast_fp16")]; + string var_2923_equation_0 = const()[name = string("op_2923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2923_cast_fp16 = einsum(equation = var_2923_equation_0, values = (var_2605_cast_fp16, var_2847_cast_fp16))[name = string("op_2923_cast_fp16")]; + string var_2925_equation_0 = const()[name = string("op_2925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2925_cast_fp16 = einsum(equation = var_2925_equation_0, values = (var_2609_cast_fp16, var_2848_cast_fp16))[name = string("op_2925_cast_fp16")]; + string var_2927_equation_0 = const()[name = string("op_2927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2927_cast_fp16 = einsum(equation = var_2927_equation_0, values = (var_2609_cast_fp16, var_2849_cast_fp16))[name = string("op_2927_cast_fp16")]; + string var_2929_equation_0 = const()[name = string("op_2929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2929_cast_fp16 = einsum(equation = var_2929_equation_0, values = (var_2609_cast_fp16, var_2850_cast_fp16))[name = string("op_2929_cast_fp16")]; + string var_2931_equation_0 = const()[name = string("op_2931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2931_cast_fp16 = einsum(equation = var_2931_equation_0, values = (var_2609_cast_fp16, var_2851_cast_fp16))[name = string("op_2931_cast_fp16")]; + string var_2933_equation_0 = const()[name = string("op_2933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2933_cast_fp16 = einsum(equation = var_2933_equation_0, values = (var_2613_cast_fp16, var_2852_cast_fp16))[name = string("op_2933_cast_fp16")]; + string var_2935_equation_0 = const()[name = string("op_2935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2935_cast_fp16 = einsum(equation = var_2935_equation_0, values = (var_2613_cast_fp16, var_2853_cast_fp16))[name = string("op_2935_cast_fp16")]; + string var_2937_equation_0 = const()[name = string("op_2937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2937_cast_fp16 = einsum(equation = var_2937_equation_0, values = (var_2613_cast_fp16, var_2854_cast_fp16))[name = string("op_2937_cast_fp16")]; + string var_2939_equation_0 = const()[name = string("op_2939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2939_cast_fp16 = einsum(equation = var_2939_equation_0, values = (var_2613_cast_fp16, var_2855_cast_fp16))[name = string("op_2939_cast_fp16")]; + string var_2941_equation_0 = const()[name = string("op_2941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2941_cast_fp16 = einsum(equation = var_2941_equation_0, values = (var_2617_cast_fp16, var_2856_cast_fp16))[name = string("op_2941_cast_fp16")]; + string var_2943_equation_0 = const()[name = string("op_2943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2943_cast_fp16 = einsum(equation = var_2943_equation_0, values = (var_2617_cast_fp16, var_2857_cast_fp16))[name = string("op_2943_cast_fp16")]; + string var_2945_equation_0 = const()[name = string("op_2945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2945_cast_fp16 = einsum(equation = var_2945_equation_0, values = (var_2617_cast_fp16, var_2858_cast_fp16))[name = string("op_2945_cast_fp16")]; + string var_2947_equation_0 = const()[name = string("op_2947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2947_cast_fp16 = einsum(equation = var_2947_equation_0, values = (var_2617_cast_fp16, var_2859_cast_fp16))[name = string("op_2947_cast_fp16")]; + string var_2949_equation_0 = const()[name = string("op_2949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2949_cast_fp16 = einsum(equation = var_2949_equation_0, values = (var_2621_cast_fp16, var_2860_cast_fp16))[name = string("op_2949_cast_fp16")]; + string var_2951_equation_0 = const()[name = string("op_2951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2951_cast_fp16 = einsum(equation = var_2951_equation_0, values = (var_2621_cast_fp16, var_2861_cast_fp16))[name = string("op_2951_cast_fp16")]; + string var_2953_equation_0 = const()[name = string("op_2953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2953_cast_fp16 = einsum(equation = var_2953_equation_0, values = (var_2621_cast_fp16, var_2862_cast_fp16))[name = string("op_2953_cast_fp16")]; + string var_2955_equation_0 = const()[name = string("op_2955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2955_cast_fp16 = einsum(equation = var_2955_equation_0, values = (var_2621_cast_fp16, var_2863_cast_fp16))[name = string("op_2955_cast_fp16")]; + string var_2957_equation_0 = const()[name = string("op_2957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2957_cast_fp16 = einsum(equation = var_2957_equation_0, values = (var_2625_cast_fp16, var_2864_cast_fp16))[name = string("op_2957_cast_fp16")]; + string var_2959_equation_0 = const()[name = string("op_2959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2959_cast_fp16 = einsum(equation = var_2959_equation_0, values = (var_2625_cast_fp16, var_2865_cast_fp16))[name = string("op_2959_cast_fp16")]; + string var_2961_equation_0 = const()[name = string("op_2961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2961_cast_fp16 = einsum(equation = var_2961_equation_0, values = (var_2625_cast_fp16, var_2866_cast_fp16))[name = string("op_2961_cast_fp16")]; + string var_2963_equation_0 = const()[name = string("op_2963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2963_cast_fp16 = einsum(equation = var_2963_equation_0, values = (var_2625_cast_fp16, var_2867_cast_fp16))[name = string("op_2963_cast_fp16")]; + bool var_2965_interleave_0 = const()[name = string("op_2965_interleave_0"), val = bool(false)]; + tensor var_2965_cast_fp16 = concat(axis = var_2076, interleave = var_2965_interleave_0, values = (var_2869_cast_fp16, var_2871_cast_fp16, var_2873_cast_fp16, var_2875_cast_fp16))[name = string("op_2965_cast_fp16")]; + bool var_2967_interleave_0 = const()[name = string("op_2967_interleave_0"), val = bool(false)]; + tensor var_2967_cast_fp16 = concat(axis = var_2076, interleave = var_2967_interleave_0, values = (var_2877_cast_fp16, var_2879_cast_fp16, var_2881_cast_fp16, var_2883_cast_fp16))[name = string("op_2967_cast_fp16")]; + bool var_2969_interleave_0 = const()[name = string("op_2969_interleave_0"), val = bool(false)]; + tensor var_2969_cast_fp16 = concat(axis = var_2076, interleave = var_2969_interleave_0, values = (var_2885_cast_fp16, var_2887_cast_fp16, var_2889_cast_fp16, var_2891_cast_fp16))[name = string("op_2969_cast_fp16")]; + bool var_2971_interleave_0 = const()[name = string("op_2971_interleave_0"), val = bool(false)]; + tensor var_2971_cast_fp16 = concat(axis = var_2076, interleave = var_2971_interleave_0, values = (var_2893_cast_fp16, var_2895_cast_fp16, var_2897_cast_fp16, var_2899_cast_fp16))[name = string("op_2971_cast_fp16")]; + bool var_2973_interleave_0 = const()[name = string("op_2973_interleave_0"), val = bool(false)]; + tensor var_2973_cast_fp16 = concat(axis = var_2076, interleave = var_2973_interleave_0, values = (var_2901_cast_fp16, var_2903_cast_fp16, var_2905_cast_fp16, var_2907_cast_fp16))[name = string("op_2973_cast_fp16")]; + bool var_2975_interleave_0 = const()[name = string("op_2975_interleave_0"), val = bool(false)]; + tensor var_2975_cast_fp16 = concat(axis = var_2076, interleave = var_2975_interleave_0, values = (var_2909_cast_fp16, var_2911_cast_fp16, var_2913_cast_fp16, var_2915_cast_fp16))[name = string("op_2975_cast_fp16")]; + bool var_2977_interleave_0 = const()[name = string("op_2977_interleave_0"), val = bool(false)]; + tensor var_2977_cast_fp16 = concat(axis = var_2076, interleave = var_2977_interleave_0, values = (var_2917_cast_fp16, var_2919_cast_fp16, var_2921_cast_fp16, var_2923_cast_fp16))[name = string("op_2977_cast_fp16")]; + bool var_2979_interleave_0 = const()[name = string("op_2979_interleave_0"), val = bool(false)]; + tensor var_2979_cast_fp16 = concat(axis = var_2076, interleave = var_2979_interleave_0, values = (var_2925_cast_fp16, var_2927_cast_fp16, var_2929_cast_fp16, var_2931_cast_fp16))[name = string("op_2979_cast_fp16")]; + bool var_2981_interleave_0 = const()[name = string("op_2981_interleave_0"), val = bool(false)]; + tensor var_2981_cast_fp16 = concat(axis = var_2076, interleave = var_2981_interleave_0, values = (var_2933_cast_fp16, var_2935_cast_fp16, var_2937_cast_fp16, var_2939_cast_fp16))[name = string("op_2981_cast_fp16")]; + bool var_2983_interleave_0 = const()[name = string("op_2983_interleave_0"), val = bool(false)]; + tensor var_2983_cast_fp16 = concat(axis = var_2076, interleave = var_2983_interleave_0, values = (var_2941_cast_fp16, var_2943_cast_fp16, var_2945_cast_fp16, var_2947_cast_fp16))[name = string("op_2983_cast_fp16")]; + bool var_2985_interleave_0 = const()[name = string("op_2985_interleave_0"), val = bool(false)]; + tensor var_2985_cast_fp16 = concat(axis = var_2076, interleave = var_2985_interleave_0, values = (var_2949_cast_fp16, var_2951_cast_fp16, var_2953_cast_fp16, var_2955_cast_fp16))[name = string("op_2985_cast_fp16")]; + bool var_2987_interleave_0 = const()[name = string("op_2987_interleave_0"), val = bool(false)]; + tensor var_2987_cast_fp16 = concat(axis = var_2076, interleave = var_2987_interleave_0, values = (var_2957_cast_fp16, var_2959_cast_fp16, var_2961_cast_fp16, var_2963_cast_fp16))[name = string("op_2987_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_2093, interleave = input_17_interleave_0, values = (var_2965_cast_fp16, var_2967_cast_fp16, var_2969_cast_fp16, var_2971_cast_fp16, var_2973_cast_fp16, var_2975_cast_fp16, var_2977_cast_fp16, var_2979_cast_fp16, var_2981_cast_fp16, var_2983_cast_fp16, var_2985_cast_fp16, var_2987_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38114112)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39293824)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_3006_to_fp16 = const()[name = string("op_3006_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_3006_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39295424)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39297024)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39298624)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44017280)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44023488)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48742144)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_3035 = const()[name = string("op_3035"), val = int32(3)]; + int32 var_3052 = const()[name = string("op_3052"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_3069_to_fp16 = const()[name = string("op_3069_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_3069_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48743744)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48745344)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48746944)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49926656)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49928256)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51107968)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52287680)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_3107_begin_0 = const()[name = string("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3107_end_0 = const()[name = string("op_3107_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3107_end_mask_0 = const()[name = string("op_3107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = query_7_cast_fp16)[name = string("op_3107_cast_fp16")]; + tensor var_3111_begin_0 = const()[name = string("op_3111_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3111_end_0 = const()[name = string("op_3111_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3111_end_mask_0 = const()[name = string("op_3111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3111_cast_fp16 = slice_by_index(begin = var_3111_begin_0, end = var_3111_end_0, end_mask = var_3111_end_mask_0, x = query_7_cast_fp16)[name = string("op_3111_cast_fp16")]; + tensor var_3115_begin_0 = const()[name = string("op_3115_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3115_end_0 = const()[name = string("op_3115_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3115_end_mask_0 = const()[name = string("op_3115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3115_cast_fp16 = slice_by_index(begin = var_3115_begin_0, end = var_3115_end_0, end_mask = var_3115_end_mask_0, x = query_7_cast_fp16)[name = string("op_3115_cast_fp16")]; + tensor var_3119_begin_0 = const()[name = string("op_3119_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3119_end_0 = const()[name = string("op_3119_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3119_end_mask_0 = const()[name = string("op_3119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3119_cast_fp16 = slice_by_index(begin = var_3119_begin_0, end = var_3119_end_0, end_mask = var_3119_end_mask_0, x = query_7_cast_fp16)[name = string("op_3119_cast_fp16")]; + tensor var_3123_begin_0 = const()[name = string("op_3123_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3123_end_0 = const()[name = string("op_3123_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3123_end_mask_0 = const()[name = string("op_3123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3123_cast_fp16 = slice_by_index(begin = var_3123_begin_0, end = var_3123_end_0, end_mask = var_3123_end_mask_0, x = query_7_cast_fp16)[name = string("op_3123_cast_fp16")]; + tensor var_3127_begin_0 = const()[name = string("op_3127_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3127_end_0 = const()[name = string("op_3127_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3127_end_mask_0 = const()[name = string("op_3127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3127_cast_fp16 = slice_by_index(begin = var_3127_begin_0, end = var_3127_end_0, end_mask = var_3127_end_mask_0, x = query_7_cast_fp16)[name = string("op_3127_cast_fp16")]; + tensor var_3131_begin_0 = const()[name = string("op_3131_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3131_end_0 = const()[name = string("op_3131_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3131_end_mask_0 = const()[name = string("op_3131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3131_cast_fp16 = slice_by_index(begin = var_3131_begin_0, end = var_3131_end_0, end_mask = var_3131_end_mask_0, x = query_7_cast_fp16)[name = string("op_3131_cast_fp16")]; + tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = query_7_cast_fp16)[name = string("op_3135_cast_fp16")]; + tensor var_3139_begin_0 = const()[name = string("op_3139_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3139_end_0 = const()[name = string("op_3139_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3139_end_mask_0 = const()[name = string("op_3139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3139_cast_fp16 = slice_by_index(begin = var_3139_begin_0, end = var_3139_end_0, end_mask = var_3139_end_mask_0, x = query_7_cast_fp16)[name = string("op_3139_cast_fp16")]; + tensor var_3143_begin_0 = const()[name = string("op_3143_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3143_end_0 = const()[name = string("op_3143_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3143_end_mask_0 = const()[name = string("op_3143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3143_cast_fp16 = slice_by_index(begin = var_3143_begin_0, end = var_3143_end_0, end_mask = var_3143_end_mask_0, x = query_7_cast_fp16)[name = string("op_3143_cast_fp16")]; + tensor var_3147_begin_0 = const()[name = string("op_3147_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3147_end_0 = const()[name = string("op_3147_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3147_end_mask_0 = const()[name = string("op_3147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3147_cast_fp16 = slice_by_index(begin = var_3147_begin_0, end = var_3147_end_0, end_mask = var_3147_end_mask_0, x = query_7_cast_fp16)[name = string("op_3147_cast_fp16")]; + tensor var_3151_begin_0 = const()[name = string("op_3151_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3151_end_0 = const()[name = string("op_3151_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3151_end_mask_0 = const()[name = string("op_3151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3151_cast_fp16 = slice_by_index(begin = var_3151_begin_0, end = var_3151_end_0, end_mask = var_3151_end_mask_0, x = query_7_cast_fp16)[name = string("op_3151_cast_fp16")]; + tensor var_3160_begin_0 = const()[name = string("op_3160_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3160_end_0 = const()[name = string("op_3160_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3160_end_mask_0 = const()[name = string("op_3160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3160_cast_fp16 = slice_by_index(begin = var_3160_begin_0, end = var_3160_end_0, end_mask = var_3160_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3160_cast_fp16")]; + tensor var_3167_begin_0 = const()[name = string("op_3167_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3167_end_0 = const()[name = string("op_3167_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3167_end_mask_0 = const()[name = string("op_3167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3167_cast_fp16 = slice_by_index(begin = var_3167_begin_0, end = var_3167_end_0, end_mask = var_3167_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3167_cast_fp16")]; + tensor var_3174_begin_0 = const()[name = string("op_3174_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3174_end_0 = const()[name = string("op_3174_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3174_end_mask_0 = const()[name = string("op_3174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3174_cast_fp16 = slice_by_index(begin = var_3174_begin_0, end = var_3174_end_0, end_mask = var_3174_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3174_cast_fp16")]; + tensor var_3181_begin_0 = const()[name = string("op_3181_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3181_end_0 = const()[name = string("op_3181_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3181_end_mask_0 = const()[name = string("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3181_cast_fp16 = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3181_cast_fp16")]; + tensor var_3188_begin_0 = const()[name = string("op_3188_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3188_end_0 = const()[name = string("op_3188_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3188_end_mask_0 = const()[name = string("op_3188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3188_cast_fp16 = slice_by_index(begin = var_3188_begin_0, end = var_3188_end_0, end_mask = var_3188_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3188_cast_fp16")]; + tensor var_3195_begin_0 = const()[name = string("op_3195_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3195_end_0 = const()[name = string("op_3195_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3195_end_mask_0 = const()[name = string("op_3195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3195_cast_fp16 = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3195_cast_fp16")]; + tensor var_3202_begin_0 = const()[name = string("op_3202_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3202_end_0 = const()[name = string("op_3202_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3202_end_mask_0 = const()[name = string("op_3202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3202_cast_fp16 = slice_by_index(begin = var_3202_begin_0, end = var_3202_end_0, end_mask = var_3202_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3202_cast_fp16")]; + tensor var_3209_begin_0 = const()[name = string("op_3209_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3209_end_0 = const()[name = string("op_3209_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3209_end_mask_0 = const()[name = string("op_3209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3209_cast_fp16 = slice_by_index(begin = var_3209_begin_0, end = var_3209_end_0, end_mask = var_3209_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3209_cast_fp16")]; + tensor var_3216_begin_0 = const()[name = string("op_3216_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3216_end_0 = const()[name = string("op_3216_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3216_end_mask_0 = const()[name = string("op_3216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3216_cast_fp16 = slice_by_index(begin = var_3216_begin_0, end = var_3216_end_0, end_mask = var_3216_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3216_cast_fp16")]; + tensor var_3223_begin_0 = const()[name = string("op_3223_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3223_end_0 = const()[name = string("op_3223_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3223_end_mask_0 = const()[name = string("op_3223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3223_cast_fp16 = slice_by_index(begin = var_3223_begin_0, end = var_3223_end_0, end_mask = var_3223_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3223_cast_fp16")]; + tensor var_3230_begin_0 = const()[name = string("op_3230_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3230_end_0 = const()[name = string("op_3230_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3230_end_mask_0 = const()[name = string("op_3230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3230_cast_fp16 = slice_by_index(begin = var_3230_begin_0, end = var_3230_end_0, end_mask = var_3230_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3230_cast_fp16")]; + tensor var_3237_begin_0 = const()[name = string("op_3237_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3237_end_0 = const()[name = string("op_3237_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3237_end_mask_0 = const()[name = string("op_3237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3237_cast_fp16 = slice_by_index(begin = var_3237_begin_0, end = var_3237_end_0, end_mask = var_3237_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3237_cast_fp16")]; + tensor var_3244_begin_0 = const()[name = string("op_3244_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3244_end_0 = const()[name = string("op_3244_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3244_end_mask_0 = const()[name = string("op_3244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3244_cast_fp16 = slice_by_index(begin = var_3244_begin_0, end = var_3244_end_0, end_mask = var_3244_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3244_cast_fp16")]; + tensor var_3251_begin_0 = const()[name = string("op_3251_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3251_end_0 = const()[name = string("op_3251_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3251_end_mask_0 = const()[name = string("op_3251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3251_cast_fp16 = slice_by_index(begin = var_3251_begin_0, end = var_3251_end_0, end_mask = var_3251_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3251_cast_fp16")]; + tensor var_3258_begin_0 = const()[name = string("op_3258_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3258_end_0 = const()[name = string("op_3258_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3258_end_mask_0 = const()[name = string("op_3258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3258_cast_fp16 = slice_by_index(begin = var_3258_begin_0, end = var_3258_end_0, end_mask = var_3258_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3258_cast_fp16")]; + tensor var_3265_begin_0 = const()[name = string("op_3265_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3265_end_0 = const()[name = string("op_3265_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3265_end_mask_0 = const()[name = string("op_3265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3265_cast_fp16 = slice_by_index(begin = var_3265_begin_0, end = var_3265_end_0, end_mask = var_3265_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3265_cast_fp16")]; + tensor var_3272_begin_0 = const()[name = string("op_3272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3272_end_0 = const()[name = string("op_3272_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3272_end_mask_0 = const()[name = string("op_3272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3272_cast_fp16 = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3272_cast_fp16")]; + tensor var_3279_begin_0 = const()[name = string("op_3279_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3279_end_0 = const()[name = string("op_3279_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3279_end_mask_0 = const()[name = string("op_3279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3279_cast_fp16 = slice_by_index(begin = var_3279_begin_0, end = var_3279_end_0, end_mask = var_3279_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3279_cast_fp16")]; + tensor var_3286_begin_0 = const()[name = string("op_3286_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3286_end_0 = const()[name = string("op_3286_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3286_end_mask_0 = const()[name = string("op_3286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3286_cast_fp16 = slice_by_index(begin = var_3286_begin_0, end = var_3286_end_0, end_mask = var_3286_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3286_cast_fp16")]; + tensor var_3293_begin_0 = const()[name = string("op_3293_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3293_end_0 = const()[name = string("op_3293_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3293_end_mask_0 = const()[name = string("op_3293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3293_cast_fp16 = slice_by_index(begin = var_3293_begin_0, end = var_3293_end_0, end_mask = var_3293_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3293_cast_fp16")]; + tensor var_3300_begin_0 = const()[name = string("op_3300_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3300_end_0 = const()[name = string("op_3300_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3300_end_mask_0 = const()[name = string("op_3300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3300_cast_fp16 = slice_by_index(begin = var_3300_begin_0, end = var_3300_end_0, end_mask = var_3300_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3300_cast_fp16")]; + tensor var_3307_begin_0 = const()[name = string("op_3307_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3307_end_0 = const()[name = string("op_3307_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3307_end_mask_0 = const()[name = string("op_3307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3307_cast_fp16 = slice_by_index(begin = var_3307_begin_0, end = var_3307_end_0, end_mask = var_3307_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3307_cast_fp16")]; + tensor var_3314_begin_0 = const()[name = string("op_3314_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3314_end_0 = const()[name = string("op_3314_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3314_end_mask_0 = const()[name = string("op_3314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3314_cast_fp16 = slice_by_index(begin = var_3314_begin_0, end = var_3314_end_0, end_mask = var_3314_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3314_cast_fp16")]; + tensor var_3321_begin_0 = const()[name = string("op_3321_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3321_end_0 = const()[name = string("op_3321_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3321_end_mask_0 = const()[name = string("op_3321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3321_cast_fp16 = slice_by_index(begin = var_3321_begin_0, end = var_3321_end_0, end_mask = var_3321_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3321_cast_fp16")]; + tensor var_3328_begin_0 = const()[name = string("op_3328_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3328_end_0 = const()[name = string("op_3328_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3328_end_mask_0 = const()[name = string("op_3328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3328_cast_fp16 = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3328_cast_fp16")]; + tensor var_3335_begin_0 = const()[name = string("op_3335_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3335_end_0 = const()[name = string("op_3335_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3335_end_mask_0 = const()[name = string("op_3335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3335_cast_fp16 = slice_by_index(begin = var_3335_begin_0, end = var_3335_end_0, end_mask = var_3335_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3335_cast_fp16")]; + tensor var_3342_begin_0 = const()[name = string("op_3342_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3342_end_0 = const()[name = string("op_3342_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3342_end_mask_0 = const()[name = string("op_3342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3342_cast_fp16 = slice_by_index(begin = var_3342_begin_0, end = var_3342_end_0, end_mask = var_3342_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3342_cast_fp16")]; + tensor var_3349_begin_0 = const()[name = string("op_3349_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3349_end_0 = const()[name = string("op_3349_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3349_end_mask_0 = const()[name = string("op_3349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3349_cast_fp16 = slice_by_index(begin = var_3349_begin_0, end = var_3349_end_0, end_mask = var_3349_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3349_cast_fp16")]; + tensor var_3356_begin_0 = const()[name = string("op_3356_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3356_end_0 = const()[name = string("op_3356_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3356_end_mask_0 = const()[name = string("op_3356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3356_cast_fp16 = slice_by_index(begin = var_3356_begin_0, end = var_3356_end_0, end_mask = var_3356_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3363_begin_0 = const()[name = string("op_3363_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3363_end_0 = const()[name = string("op_3363_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3363_end_mask_0 = const()[name = string("op_3363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3363_cast_fp16 = slice_by_index(begin = var_3363_begin_0, end = var_3363_end_0, end_mask = var_3363_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3363_cast_fp16")]; + tensor var_3370_begin_0 = const()[name = string("op_3370_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3370_end_0 = const()[name = string("op_3370_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3370_end_mask_0 = const()[name = string("op_3370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3370_cast_fp16 = slice_by_index(begin = var_3370_begin_0, end = var_3370_end_0, end_mask = var_3370_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3370_cast_fp16")]; + tensor var_3377_begin_0 = const()[name = string("op_3377_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3377_end_0 = const()[name = string("op_3377_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3377_end_mask_0 = const()[name = string("op_3377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3377_cast_fp16")]; + tensor var_3384_begin_0 = const()[name = string("op_3384_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3384_end_0 = const()[name = string("op_3384_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3384_end_mask_0 = const()[name = string("op_3384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3384_cast_fp16 = slice_by_index(begin = var_3384_begin_0, end = var_3384_end_0, end_mask = var_3384_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3384_cast_fp16")]; + tensor var_3391_begin_0 = const()[name = string("op_3391_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3391_end_0 = const()[name = string("op_3391_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3391_end_mask_0 = const()[name = string("op_3391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3391_cast_fp16 = slice_by_index(begin = var_3391_begin_0, end = var_3391_end_0, end_mask = var_3391_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3391_cast_fp16")]; + tensor var_3398_begin_0 = const()[name = string("op_3398_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3398_end_0 = const()[name = string("op_3398_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3398_end_mask_0 = const()[name = string("op_3398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3398_cast_fp16 = slice_by_index(begin = var_3398_begin_0, end = var_3398_end_0, end_mask = var_3398_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3398_cast_fp16")]; + tensor var_3405_begin_0 = const()[name = string("op_3405_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3405_end_0 = const()[name = string("op_3405_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3405_end_mask_0 = const()[name = string("op_3405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3405_cast_fp16 = slice_by_index(begin = var_3405_begin_0, end = var_3405_end_0, end_mask = var_3405_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3405_cast_fp16")]; + tensor var_3412_begin_0 = const()[name = string("op_3412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3412_end_0 = const()[name = string("op_3412_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3412_end_mask_0 = const()[name = string("op_3412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3412_cast_fp16 = slice_by_index(begin = var_3412_begin_0, end = var_3412_end_0, end_mask = var_3412_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3412_cast_fp16")]; + tensor var_3419_begin_0 = const()[name = string("op_3419_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3419_end_0 = const()[name = string("op_3419_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3419_end_mask_0 = const()[name = string("op_3419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3419_cast_fp16 = slice_by_index(begin = var_3419_begin_0, end = var_3419_end_0, end_mask = var_3419_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3419_cast_fp16")]; + tensor var_3426_begin_0 = const()[name = string("op_3426_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3426_end_0 = const()[name = string("op_3426_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3426_end_mask_0 = const()[name = string("op_3426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3426_cast_fp16 = slice_by_index(begin = var_3426_begin_0, end = var_3426_end_0, end_mask = var_3426_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3426_cast_fp16")]; + tensor var_3433_begin_0 = const()[name = string("op_3433_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3433_end_0 = const()[name = string("op_3433_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3433_end_mask_0 = const()[name = string("op_3433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3433_cast_fp16 = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3433_cast_fp16")]; + tensor var_3440_begin_0 = const()[name = string("op_3440_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3440_end_0 = const()[name = string("op_3440_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3440_end_mask_0 = const()[name = string("op_3440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3440_cast_fp16 = slice_by_index(begin = var_3440_begin_0, end = var_3440_end_0, end_mask = var_3440_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3440_cast_fp16")]; + tensor var_3447_begin_0 = const()[name = string("op_3447_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3447_end_0 = const()[name = string("op_3447_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3447_end_mask_0 = const()[name = string("op_3447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3447_cast_fp16 = slice_by_index(begin = var_3447_begin_0, end = var_3447_end_0, end_mask = var_3447_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3447_cast_fp16")]; + tensor var_3454_begin_0 = const()[name = string("op_3454_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3454_end_0 = const()[name = string("op_3454_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3454_end_mask_0 = const()[name = string("op_3454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3454_cast_fp16 = slice_by_index(begin = var_3454_begin_0, end = var_3454_end_0, end_mask = var_3454_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3454_cast_fp16")]; + tensor var_3461_begin_0 = const()[name = string("op_3461_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3461_end_0 = const()[name = string("op_3461_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3461_end_mask_0 = const()[name = string("op_3461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3461_cast_fp16 = slice_by_index(begin = var_3461_begin_0, end = var_3461_end_0, end_mask = var_3461_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3461_cast_fp16")]; + tensor var_3468_begin_0 = const()[name = string("op_3468_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3468_end_0 = const()[name = string("op_3468_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3468_end_mask_0 = const()[name = string("op_3468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3468_cast_fp16 = slice_by_index(begin = var_3468_begin_0, end = var_3468_end_0, end_mask = var_3468_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3468_cast_fp16")]; + tensor var_3475_begin_0 = const()[name = string("op_3475_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3475_end_0 = const()[name = string("op_3475_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3475_end_mask_0 = const()[name = string("op_3475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3475_cast_fp16 = slice_by_index(begin = var_3475_begin_0, end = var_3475_end_0, end_mask = var_3475_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3475_cast_fp16")]; + tensor var_3482_begin_0 = const()[name = string("op_3482_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3482_end_0 = const()[name = string("op_3482_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3482_end_mask_0 = const()[name = string("op_3482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3482_cast_fp16 = slice_by_index(begin = var_3482_begin_0, end = var_3482_end_0, end_mask = var_3482_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3482_cast_fp16")]; + tensor var_3489_begin_0 = const()[name = string("op_3489_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3489_end_0 = const()[name = string("op_3489_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3489_end_mask_0 = const()[name = string("op_3489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3489_cast_fp16 = slice_by_index(begin = var_3489_begin_0, end = var_3489_end_0, end_mask = var_3489_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3489_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3494_begin_0 = const()[name = string("op_3494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3494_end_0 = const()[name = string("op_3494_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3494_end_mask_0 = const()[name = string("op_3494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_8")]; + tensor var_3494_cast_fp16 = slice_by_index(begin = var_3494_begin_0, end = var_3494_end_0, end_mask = var_3494_end_mask_0, x = k_7_cast_fp16)[name = string("op_3494_cast_fp16")]; + tensor var_3498_begin_0 = const()[name = string("op_3498_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3498_end_0 = const()[name = string("op_3498_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3498_end_mask_0 = const()[name = string("op_3498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3498_cast_fp16 = slice_by_index(begin = var_3498_begin_0, end = var_3498_end_0, end_mask = var_3498_end_mask_0, x = k_7_cast_fp16)[name = string("op_3498_cast_fp16")]; + tensor var_3502_begin_0 = const()[name = string("op_3502_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3502_end_0 = const()[name = string("op_3502_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3502_end_mask_0 = const()[name = string("op_3502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3502_cast_fp16 = slice_by_index(begin = var_3502_begin_0, end = var_3502_end_0, end_mask = var_3502_end_mask_0, x = k_7_cast_fp16)[name = string("op_3502_cast_fp16")]; + tensor var_3506_begin_0 = const()[name = string("op_3506_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3506_end_0 = const()[name = string("op_3506_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3506_end_mask_0 = const()[name = string("op_3506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3506_cast_fp16 = slice_by_index(begin = var_3506_begin_0, end = var_3506_end_0, end_mask = var_3506_end_mask_0, x = k_7_cast_fp16)[name = string("op_3506_cast_fp16")]; + tensor var_3510_begin_0 = const()[name = string("op_3510_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3510_end_0 = const()[name = string("op_3510_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3510_end_mask_0 = const()[name = string("op_3510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3510_cast_fp16 = slice_by_index(begin = var_3510_begin_0, end = var_3510_end_0, end_mask = var_3510_end_mask_0, x = k_7_cast_fp16)[name = string("op_3510_cast_fp16")]; + tensor var_3514_begin_0 = const()[name = string("op_3514_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3514_end_0 = const()[name = string("op_3514_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3514_end_mask_0 = const()[name = string("op_3514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3514_cast_fp16 = slice_by_index(begin = var_3514_begin_0, end = var_3514_end_0, end_mask = var_3514_end_mask_0, x = k_7_cast_fp16)[name = string("op_3514_cast_fp16")]; + tensor var_3518_begin_0 = const()[name = string("op_3518_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3518_end_0 = const()[name = string("op_3518_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3518_end_mask_0 = const()[name = string("op_3518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3518_cast_fp16 = slice_by_index(begin = var_3518_begin_0, end = var_3518_end_0, end_mask = var_3518_end_mask_0, x = k_7_cast_fp16)[name = string("op_3518_cast_fp16")]; + tensor var_3522_begin_0 = const()[name = string("op_3522_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3522_end_0 = const()[name = string("op_3522_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3522_end_mask_0 = const()[name = string("op_3522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3522_cast_fp16 = slice_by_index(begin = var_3522_begin_0, end = var_3522_end_0, end_mask = var_3522_end_mask_0, x = k_7_cast_fp16)[name = string("op_3522_cast_fp16")]; + tensor var_3526_begin_0 = const()[name = string("op_3526_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3526_end_0 = const()[name = string("op_3526_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_3526_end_mask_0 = const()[name = string("op_3526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3526_cast_fp16 = slice_by_index(begin = var_3526_begin_0, end = var_3526_end_0, end_mask = var_3526_end_mask_0, x = k_7_cast_fp16)[name = string("op_3526_cast_fp16")]; + tensor var_3530_begin_0 = const()[name = string("op_3530_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3530_end_0 = const()[name = string("op_3530_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_3530_end_mask_0 = const()[name = string("op_3530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3530_cast_fp16 = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = k_7_cast_fp16)[name = string("op_3530_cast_fp16")]; + tensor var_3534_begin_0 = const()[name = string("op_3534_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3534_end_0 = const()[name = string("op_3534_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_3534_end_mask_0 = const()[name = string("op_3534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3534_cast_fp16 = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = k_7_cast_fp16)[name = string("op_3534_cast_fp16")]; + tensor var_3538_begin_0 = const()[name = string("op_3538_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3538_end_0 = const()[name = string("op_3538_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_3538_end_mask_0 = const()[name = string("op_3538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3538_cast_fp16 = slice_by_index(begin = var_3538_begin_0, end = var_3538_end_0, end_mask = var_3538_end_mask_0, x = k_7_cast_fp16)[name = string("op_3538_cast_fp16")]; + tensor var_3540_begin_0 = const()[name = string("op_3540_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3540_end_0 = const()[name = string("op_3540_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3540_end_mask_0 = const()[name = string("op_3540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3540_cast_fp16 = slice_by_index(begin = var_3540_begin_0, end = var_3540_end_0, end_mask = var_3540_end_mask_0, x = value_7_cast_fp16)[name = string("op_3540_cast_fp16")]; + tensor var_3544_begin_0 = const()[name = string("op_3544_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3544_end_0 = const()[name = string("op_3544_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3544_end_mask_0 = const()[name = string("op_3544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3544_cast_fp16 = slice_by_index(begin = var_3544_begin_0, end = var_3544_end_0, end_mask = var_3544_end_mask_0, x = value_7_cast_fp16)[name = string("op_3544_cast_fp16")]; + tensor var_3548_begin_0 = const()[name = string("op_3548_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3548_end_0 = const()[name = string("op_3548_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3548_end_mask_0 = const()[name = string("op_3548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3548_cast_fp16 = slice_by_index(begin = var_3548_begin_0, end = var_3548_end_0, end_mask = var_3548_end_mask_0, x = value_7_cast_fp16)[name = string("op_3548_cast_fp16")]; + tensor var_3552_begin_0 = const()[name = string("op_3552_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3552_end_0 = const()[name = string("op_3552_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3552_end_mask_0 = const()[name = string("op_3552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3552_cast_fp16 = slice_by_index(begin = var_3552_begin_0, end = var_3552_end_0, end_mask = var_3552_end_mask_0, x = value_7_cast_fp16)[name = string("op_3552_cast_fp16")]; + tensor var_3556_begin_0 = const()[name = string("op_3556_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3556_end_0 = const()[name = string("op_3556_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3556_end_mask_0 = const()[name = string("op_3556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3556_cast_fp16 = slice_by_index(begin = var_3556_begin_0, end = var_3556_end_0, end_mask = var_3556_end_mask_0, x = value_7_cast_fp16)[name = string("op_3556_cast_fp16")]; + tensor var_3560_begin_0 = const()[name = string("op_3560_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3560_end_0 = const()[name = string("op_3560_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3560_end_mask_0 = const()[name = string("op_3560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3560_cast_fp16 = slice_by_index(begin = var_3560_begin_0, end = var_3560_end_0, end_mask = var_3560_end_mask_0, x = value_7_cast_fp16)[name = string("op_3560_cast_fp16")]; + tensor var_3564_begin_0 = const()[name = string("op_3564_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3564_end_0 = const()[name = string("op_3564_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3564_end_mask_0 = const()[name = string("op_3564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3564_cast_fp16 = slice_by_index(begin = var_3564_begin_0, end = var_3564_end_0, end_mask = var_3564_end_mask_0, x = value_7_cast_fp16)[name = string("op_3564_cast_fp16")]; + tensor var_3568_begin_0 = const()[name = string("op_3568_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3568_end_0 = const()[name = string("op_3568_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3568_end_mask_0 = const()[name = string("op_3568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3568_cast_fp16 = slice_by_index(begin = var_3568_begin_0, end = var_3568_end_0, end_mask = var_3568_end_mask_0, x = value_7_cast_fp16)[name = string("op_3568_cast_fp16")]; + tensor var_3572_begin_0 = const()[name = string("op_3572_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3572_end_0 = const()[name = string("op_3572_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3572_end_mask_0 = const()[name = string("op_3572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3572_cast_fp16 = slice_by_index(begin = var_3572_begin_0, end = var_3572_end_0, end_mask = var_3572_end_mask_0, x = value_7_cast_fp16)[name = string("op_3572_cast_fp16")]; + tensor var_3576_begin_0 = const()[name = string("op_3576_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3576_end_0 = const()[name = string("op_3576_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3576_end_mask_0 = const()[name = string("op_3576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3576_cast_fp16 = slice_by_index(begin = var_3576_begin_0, end = var_3576_end_0, end_mask = var_3576_end_mask_0, x = value_7_cast_fp16)[name = string("op_3576_cast_fp16")]; + tensor var_3580_begin_0 = const()[name = string("op_3580_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3580_end_0 = const()[name = string("op_3580_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3580_end_mask_0 = const()[name = string("op_3580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3580_cast_fp16 = slice_by_index(begin = var_3580_begin_0, end = var_3580_end_0, end_mask = var_3580_end_mask_0, x = value_7_cast_fp16)[name = string("op_3580_cast_fp16")]; + tensor var_3584_begin_0 = const()[name = string("op_3584_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3584_end_0 = const()[name = string("op_3584_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3584_end_mask_0 = const()[name = string("op_3584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3584_cast_fp16 = slice_by_index(begin = var_3584_begin_0, end = var_3584_end_0, end_mask = var_3584_end_mask_0, x = value_7_cast_fp16)[name = string("op_3584_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_3494_cast_fp16, var_3160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_3494_cast_fp16, var_3167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_3494_cast_fp16, var_3174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_3494_cast_fp16, var_3181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_3498_cast_fp16, var_3188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_3498_cast_fp16, var_3195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_3498_cast_fp16, var_3202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_3498_cast_fp16, var_3209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_3502_cast_fp16, var_3216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_3502_cast_fp16, var_3223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_3502_cast_fp16, var_3230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_3502_cast_fp16, var_3237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_3506_cast_fp16, var_3244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_3506_cast_fp16, var_3251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_3506_cast_fp16, var_3258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_3506_cast_fp16, var_3265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3510_cast_fp16, var_3272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3510_cast_fp16, var_3279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3510_cast_fp16, var_3286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3510_cast_fp16, var_3293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3514_cast_fp16, var_3300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3514_cast_fp16, var_3307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3514_cast_fp16, var_3314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3514_cast_fp16, var_3321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3518_cast_fp16, var_3328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3518_cast_fp16, var_3335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3518_cast_fp16, var_3342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3518_cast_fp16, var_3349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3522_cast_fp16, var_3356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3522_cast_fp16, var_3363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3522_cast_fp16, var_3370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3522_cast_fp16, var_3377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3526_cast_fp16, var_3384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3526_cast_fp16, var_3391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3526_cast_fp16, var_3398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3526_cast_fp16, var_3405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3530_cast_fp16, var_3412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3530_cast_fp16, var_3419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3530_cast_fp16, var_3426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3530_cast_fp16, var_3433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3534_cast_fp16, var_3440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3534_cast_fp16, var_3447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3534_cast_fp16, var_3454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3534_cast_fp16, var_3461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3538_cast_fp16, var_3468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3538_cast_fp16, var_3475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3538_cast_fp16, var_3482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_383_equation_0, values = (var_3538_cast_fp16, var_3489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_383_cast_fp16")]; + fp16 var_3683_to_fp16 = const()[name = string("op_3683_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_3683_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_3685_to_fp16 = const()[name = string("op_3685_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_3685_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_3687_to_fp16 = const()[name = string("op_3687_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_3687_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_3689_to_fp16 = const()[name = string("op_3689_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_3689_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_3691_to_fp16 = const()[name = string("op_3691_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_3691_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_3693_to_fp16 = const()[name = string("op_3693_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_3693_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_3695_to_fp16 = const()[name = string("op_3695_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_3695_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_3697_to_fp16 = const()[name = string("op_3697_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_3697_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_3699_to_fp16 = const()[name = string("op_3699_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_3699_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_3701_to_fp16 = const()[name = string("op_3701_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_3701_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_3703_to_fp16 = const()[name = string("op_3703_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_3703_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_3705_to_fp16 = const()[name = string("op_3705_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_3705_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_3707_to_fp16 = const()[name = string("op_3707_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_3707_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_3709_to_fp16 = const()[name = string("op_3709_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_3709_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_3711_to_fp16 = const()[name = string("op_3711_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_3711_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_3713_to_fp16 = const()[name = string("op_3713_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_3713_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + fp16 var_3715_to_fp16 = const()[name = string("op_3715_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_3715_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_3717_to_fp16 = const()[name = string("op_3717_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_3717_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_3719_to_fp16 = const()[name = string("op_3719_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_3719_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_3721_to_fp16 = const()[name = string("op_3721_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_3721_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_3723_to_fp16 = const()[name = string("op_3723_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_3723_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_3725_to_fp16 = const()[name = string("op_3725_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_3725_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_3727_to_fp16 = const()[name = string("op_3727_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_3727_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_3729_to_fp16 = const()[name = string("op_3729_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_3729_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_3731_to_fp16 = const()[name = string("op_3731_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_3731_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_3733_to_fp16 = const()[name = string("op_3733_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_3733_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_3735_to_fp16 = const()[name = string("op_3735_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_3735_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_3737_to_fp16 = const()[name = string("op_3737_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_3737_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_3739_to_fp16 = const()[name = string("op_3739_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_3739_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_3741_to_fp16 = const()[name = string("op_3741_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_3741_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_3743_to_fp16 = const()[name = string("op_3743_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_3743_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_3745_to_fp16 = const()[name = string("op_3745_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_3745_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_3747_to_fp16 = const()[name = string("op_3747_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_3747_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_3749_to_fp16 = const()[name = string("op_3749_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_3749_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_3751_to_fp16 = const()[name = string("op_3751_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_3751_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_3753_to_fp16 = const()[name = string("op_3753_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_3753_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_3755_to_fp16 = const()[name = string("op_3755_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_3755_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_3757_to_fp16 = const()[name = string("op_3757_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_3757_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_3759_to_fp16 = const()[name = string("op_3759_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_3759_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_3761_to_fp16 = const()[name = string("op_3761_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_3761_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_3763_to_fp16 = const()[name = string("op_3763_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_3763_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_3765_to_fp16 = const()[name = string("op_3765_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_3765_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_3767_to_fp16 = const()[name = string("op_3767_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_3767_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_3769_to_fp16 = const()[name = string("op_3769_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_3769_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_3771_to_fp16 = const()[name = string("op_3771_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_3771_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_3773_to_fp16 = const()[name = string("op_3773_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_3773_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_3775_to_fp16 = const()[name = string("op_3775_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_3775_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_3777_to_fp16 = const()[name = string("op_3777_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_383_cast_fp16, y = var_3777_to_fp16)[name = string("aw_chunk_383_cast_fp16")]; + tensor var_3779_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_289_cast_fp16)[name = string("op_3779_cast_fp16")]; + tensor var_3780_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_291_cast_fp16)[name = string("op_3780_cast_fp16")]; + tensor var_3781_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_293_cast_fp16)[name = string("op_3781_cast_fp16")]; + tensor var_3782_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_295_cast_fp16)[name = string("op_3782_cast_fp16")]; + tensor var_3783_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_297_cast_fp16)[name = string("op_3783_cast_fp16")]; + tensor var_3784_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_299_cast_fp16)[name = string("op_3784_cast_fp16")]; + tensor var_3785_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_301_cast_fp16)[name = string("op_3785_cast_fp16")]; + tensor var_3786_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_303_cast_fp16)[name = string("op_3786_cast_fp16")]; + tensor var_3787_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_305_cast_fp16)[name = string("op_3787_cast_fp16")]; + tensor var_3788_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_307_cast_fp16)[name = string("op_3788_cast_fp16")]; + tensor var_3789_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_309_cast_fp16)[name = string("op_3789_cast_fp16")]; + tensor var_3790_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_311_cast_fp16)[name = string("op_3790_cast_fp16")]; + tensor var_3791_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_313_cast_fp16)[name = string("op_3791_cast_fp16")]; + tensor var_3792_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_315_cast_fp16)[name = string("op_3792_cast_fp16")]; + tensor var_3793_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_317_cast_fp16)[name = string("op_3793_cast_fp16")]; + tensor var_3794_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_319_cast_fp16)[name = string("op_3794_cast_fp16")]; + tensor var_3795_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_321_cast_fp16)[name = string("op_3795_cast_fp16")]; + tensor var_3796_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_323_cast_fp16)[name = string("op_3796_cast_fp16")]; + tensor var_3797_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_325_cast_fp16)[name = string("op_3797_cast_fp16")]; + tensor var_3798_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_327_cast_fp16)[name = string("op_3798_cast_fp16")]; + tensor var_3799_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_329_cast_fp16)[name = string("op_3799_cast_fp16")]; + tensor var_3800_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_331_cast_fp16)[name = string("op_3800_cast_fp16")]; + tensor var_3801_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_333_cast_fp16)[name = string("op_3801_cast_fp16")]; + tensor var_3802_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_335_cast_fp16)[name = string("op_3802_cast_fp16")]; + tensor var_3803_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_337_cast_fp16)[name = string("op_3803_cast_fp16")]; + tensor var_3804_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_339_cast_fp16)[name = string("op_3804_cast_fp16")]; + tensor var_3805_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_341_cast_fp16)[name = string("op_3805_cast_fp16")]; + tensor var_3806_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_343_cast_fp16)[name = string("op_3806_cast_fp16")]; + tensor var_3807_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_345_cast_fp16)[name = string("op_3807_cast_fp16")]; + tensor var_3808_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_347_cast_fp16)[name = string("op_3808_cast_fp16")]; + tensor var_3809_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_349_cast_fp16)[name = string("op_3809_cast_fp16")]; + tensor var_3810_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_351_cast_fp16)[name = string("op_3810_cast_fp16")]; + tensor var_3811_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_353_cast_fp16)[name = string("op_3811_cast_fp16")]; + tensor var_3812_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_355_cast_fp16)[name = string("op_3812_cast_fp16")]; + tensor var_3813_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_357_cast_fp16)[name = string("op_3813_cast_fp16")]; + tensor var_3814_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_359_cast_fp16)[name = string("op_3814_cast_fp16")]; + tensor var_3815_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_361_cast_fp16)[name = string("op_3815_cast_fp16")]; + tensor var_3816_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_363_cast_fp16)[name = string("op_3816_cast_fp16")]; + tensor var_3817_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_365_cast_fp16)[name = string("op_3817_cast_fp16")]; + tensor var_3818_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_367_cast_fp16)[name = string("op_3818_cast_fp16")]; + tensor var_3819_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_369_cast_fp16)[name = string("op_3819_cast_fp16")]; + tensor var_3820_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_371_cast_fp16)[name = string("op_3820_cast_fp16")]; + tensor var_3821_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_373_cast_fp16)[name = string("op_3821_cast_fp16")]; + tensor var_3822_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_375_cast_fp16)[name = string("op_3822_cast_fp16")]; + tensor var_3823_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_377_cast_fp16)[name = string("op_3823_cast_fp16")]; + tensor var_3824_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_379_cast_fp16)[name = string("op_3824_cast_fp16")]; + tensor var_3825_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_381_cast_fp16)[name = string("op_3825_cast_fp16")]; + tensor var_3826_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_383_cast_fp16)[name = string("op_3826_cast_fp16")]; + string var_3828_equation_0 = const()[name = string("op_3828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3828_cast_fp16 = einsum(equation = var_3828_equation_0, values = (var_3540_cast_fp16, var_3779_cast_fp16))[name = string("op_3828_cast_fp16")]; + string var_3830_equation_0 = const()[name = string("op_3830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3830_cast_fp16 = einsum(equation = var_3830_equation_0, values = (var_3540_cast_fp16, var_3780_cast_fp16))[name = string("op_3830_cast_fp16")]; + string var_3832_equation_0 = const()[name = string("op_3832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3832_cast_fp16 = einsum(equation = var_3832_equation_0, values = (var_3540_cast_fp16, var_3781_cast_fp16))[name = string("op_3832_cast_fp16")]; + string var_3834_equation_0 = const()[name = string("op_3834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3834_cast_fp16 = einsum(equation = var_3834_equation_0, values = (var_3540_cast_fp16, var_3782_cast_fp16))[name = string("op_3834_cast_fp16")]; + string var_3836_equation_0 = const()[name = string("op_3836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3836_cast_fp16 = einsum(equation = var_3836_equation_0, values = (var_3544_cast_fp16, var_3783_cast_fp16))[name = string("op_3836_cast_fp16")]; + string var_3838_equation_0 = const()[name = string("op_3838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3838_cast_fp16 = einsum(equation = var_3838_equation_0, values = (var_3544_cast_fp16, var_3784_cast_fp16))[name = string("op_3838_cast_fp16")]; + string var_3840_equation_0 = const()[name = string("op_3840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3840_cast_fp16 = einsum(equation = var_3840_equation_0, values = (var_3544_cast_fp16, var_3785_cast_fp16))[name = string("op_3840_cast_fp16")]; + string var_3842_equation_0 = const()[name = string("op_3842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3842_cast_fp16 = einsum(equation = var_3842_equation_0, values = (var_3544_cast_fp16, var_3786_cast_fp16))[name = string("op_3842_cast_fp16")]; + string var_3844_equation_0 = const()[name = string("op_3844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3844_cast_fp16 = einsum(equation = var_3844_equation_0, values = (var_3548_cast_fp16, var_3787_cast_fp16))[name = string("op_3844_cast_fp16")]; + string var_3846_equation_0 = const()[name = string("op_3846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3846_cast_fp16 = einsum(equation = var_3846_equation_0, values = (var_3548_cast_fp16, var_3788_cast_fp16))[name = string("op_3846_cast_fp16")]; + string var_3848_equation_0 = const()[name = string("op_3848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3848_cast_fp16 = einsum(equation = var_3848_equation_0, values = (var_3548_cast_fp16, var_3789_cast_fp16))[name = string("op_3848_cast_fp16")]; + string var_3850_equation_0 = const()[name = string("op_3850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3850_cast_fp16 = einsum(equation = var_3850_equation_0, values = (var_3548_cast_fp16, var_3790_cast_fp16))[name = string("op_3850_cast_fp16")]; + string var_3852_equation_0 = const()[name = string("op_3852_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3852_cast_fp16 = einsum(equation = var_3852_equation_0, values = (var_3552_cast_fp16, var_3791_cast_fp16))[name = string("op_3852_cast_fp16")]; + string var_3854_equation_0 = const()[name = string("op_3854_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3854_cast_fp16 = einsum(equation = var_3854_equation_0, values = (var_3552_cast_fp16, var_3792_cast_fp16))[name = string("op_3854_cast_fp16")]; + string var_3856_equation_0 = const()[name = string("op_3856_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3856_cast_fp16 = einsum(equation = var_3856_equation_0, values = (var_3552_cast_fp16, var_3793_cast_fp16))[name = string("op_3856_cast_fp16")]; + string var_3858_equation_0 = const()[name = string("op_3858_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3858_cast_fp16 = einsum(equation = var_3858_equation_0, values = (var_3552_cast_fp16, var_3794_cast_fp16))[name = string("op_3858_cast_fp16")]; + string var_3860_equation_0 = const()[name = string("op_3860_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3860_cast_fp16 = einsum(equation = var_3860_equation_0, values = (var_3556_cast_fp16, var_3795_cast_fp16))[name = string("op_3860_cast_fp16")]; + string var_3862_equation_0 = const()[name = string("op_3862_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3862_cast_fp16 = einsum(equation = var_3862_equation_0, values = (var_3556_cast_fp16, var_3796_cast_fp16))[name = string("op_3862_cast_fp16")]; + string var_3864_equation_0 = const()[name = string("op_3864_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3864_cast_fp16 = einsum(equation = var_3864_equation_0, values = (var_3556_cast_fp16, var_3797_cast_fp16))[name = string("op_3864_cast_fp16")]; + string var_3866_equation_0 = const()[name = string("op_3866_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3866_cast_fp16 = einsum(equation = var_3866_equation_0, values = (var_3556_cast_fp16, var_3798_cast_fp16))[name = string("op_3866_cast_fp16")]; + string var_3868_equation_0 = const()[name = string("op_3868_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3868_cast_fp16 = einsum(equation = var_3868_equation_0, values = (var_3560_cast_fp16, var_3799_cast_fp16))[name = string("op_3868_cast_fp16")]; + string var_3870_equation_0 = const()[name = string("op_3870_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3870_cast_fp16 = einsum(equation = var_3870_equation_0, values = (var_3560_cast_fp16, var_3800_cast_fp16))[name = string("op_3870_cast_fp16")]; + string var_3872_equation_0 = const()[name = string("op_3872_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3872_cast_fp16 = einsum(equation = var_3872_equation_0, values = (var_3560_cast_fp16, var_3801_cast_fp16))[name = string("op_3872_cast_fp16")]; + string var_3874_equation_0 = const()[name = string("op_3874_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3874_cast_fp16 = einsum(equation = var_3874_equation_0, values = (var_3560_cast_fp16, var_3802_cast_fp16))[name = string("op_3874_cast_fp16")]; + string var_3876_equation_0 = const()[name = string("op_3876_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3876_cast_fp16 = einsum(equation = var_3876_equation_0, values = (var_3564_cast_fp16, var_3803_cast_fp16))[name = string("op_3876_cast_fp16")]; + string var_3878_equation_0 = const()[name = string("op_3878_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3878_cast_fp16 = einsum(equation = var_3878_equation_0, values = (var_3564_cast_fp16, var_3804_cast_fp16))[name = string("op_3878_cast_fp16")]; + string var_3880_equation_0 = const()[name = string("op_3880_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3880_cast_fp16 = einsum(equation = var_3880_equation_0, values = (var_3564_cast_fp16, var_3805_cast_fp16))[name = string("op_3880_cast_fp16")]; + string var_3882_equation_0 = const()[name = string("op_3882_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3882_cast_fp16 = einsum(equation = var_3882_equation_0, values = (var_3564_cast_fp16, var_3806_cast_fp16))[name = string("op_3882_cast_fp16")]; + string var_3884_equation_0 = const()[name = string("op_3884_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3884_cast_fp16 = einsum(equation = var_3884_equation_0, values = (var_3568_cast_fp16, var_3807_cast_fp16))[name = string("op_3884_cast_fp16")]; + string var_3886_equation_0 = const()[name = string("op_3886_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3886_cast_fp16 = einsum(equation = var_3886_equation_0, values = (var_3568_cast_fp16, var_3808_cast_fp16))[name = string("op_3886_cast_fp16")]; + string var_3888_equation_0 = const()[name = string("op_3888_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3888_cast_fp16 = einsum(equation = var_3888_equation_0, values = (var_3568_cast_fp16, var_3809_cast_fp16))[name = string("op_3888_cast_fp16")]; + string var_3890_equation_0 = const()[name = string("op_3890_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3890_cast_fp16 = einsum(equation = var_3890_equation_0, values = (var_3568_cast_fp16, var_3810_cast_fp16))[name = string("op_3890_cast_fp16")]; + string var_3892_equation_0 = const()[name = string("op_3892_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3892_cast_fp16 = einsum(equation = var_3892_equation_0, values = (var_3572_cast_fp16, var_3811_cast_fp16))[name = string("op_3892_cast_fp16")]; + string var_3894_equation_0 = const()[name = string("op_3894_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3894_cast_fp16 = einsum(equation = var_3894_equation_0, values = (var_3572_cast_fp16, var_3812_cast_fp16))[name = string("op_3894_cast_fp16")]; + string var_3896_equation_0 = const()[name = string("op_3896_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3896_cast_fp16 = einsum(equation = var_3896_equation_0, values = (var_3572_cast_fp16, var_3813_cast_fp16))[name = string("op_3896_cast_fp16")]; + string var_3898_equation_0 = const()[name = string("op_3898_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3898_cast_fp16 = einsum(equation = var_3898_equation_0, values = (var_3572_cast_fp16, var_3814_cast_fp16))[name = string("op_3898_cast_fp16")]; + string var_3900_equation_0 = const()[name = string("op_3900_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3900_cast_fp16 = einsum(equation = var_3900_equation_0, values = (var_3576_cast_fp16, var_3815_cast_fp16))[name = string("op_3900_cast_fp16")]; + string var_3902_equation_0 = const()[name = string("op_3902_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3902_cast_fp16 = einsum(equation = var_3902_equation_0, values = (var_3576_cast_fp16, var_3816_cast_fp16))[name = string("op_3902_cast_fp16")]; + string var_3904_equation_0 = const()[name = string("op_3904_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3904_cast_fp16 = einsum(equation = var_3904_equation_0, values = (var_3576_cast_fp16, var_3817_cast_fp16))[name = string("op_3904_cast_fp16")]; + string var_3906_equation_0 = const()[name = string("op_3906_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3906_cast_fp16 = einsum(equation = var_3906_equation_0, values = (var_3576_cast_fp16, var_3818_cast_fp16))[name = string("op_3906_cast_fp16")]; + string var_3908_equation_0 = const()[name = string("op_3908_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3908_cast_fp16 = einsum(equation = var_3908_equation_0, values = (var_3580_cast_fp16, var_3819_cast_fp16))[name = string("op_3908_cast_fp16")]; + string var_3910_equation_0 = const()[name = string("op_3910_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3910_cast_fp16 = einsum(equation = var_3910_equation_0, values = (var_3580_cast_fp16, var_3820_cast_fp16))[name = string("op_3910_cast_fp16")]; + string var_3912_equation_0 = const()[name = string("op_3912_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3912_cast_fp16 = einsum(equation = var_3912_equation_0, values = (var_3580_cast_fp16, var_3821_cast_fp16))[name = string("op_3912_cast_fp16")]; + string var_3914_equation_0 = const()[name = string("op_3914_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3914_cast_fp16 = einsum(equation = var_3914_equation_0, values = (var_3580_cast_fp16, var_3822_cast_fp16))[name = string("op_3914_cast_fp16")]; + string var_3916_equation_0 = const()[name = string("op_3916_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3916_cast_fp16 = einsum(equation = var_3916_equation_0, values = (var_3584_cast_fp16, var_3823_cast_fp16))[name = string("op_3916_cast_fp16")]; + string var_3918_equation_0 = const()[name = string("op_3918_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3918_cast_fp16 = einsum(equation = var_3918_equation_0, values = (var_3584_cast_fp16, var_3824_cast_fp16))[name = string("op_3918_cast_fp16")]; + string var_3920_equation_0 = const()[name = string("op_3920_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3920_cast_fp16 = einsum(equation = var_3920_equation_0, values = (var_3584_cast_fp16, var_3825_cast_fp16))[name = string("op_3920_cast_fp16")]; + string var_3922_equation_0 = const()[name = string("op_3922_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3922_cast_fp16 = einsum(equation = var_3922_equation_0, values = (var_3584_cast_fp16, var_3826_cast_fp16))[name = string("op_3922_cast_fp16")]; + bool var_3924_interleave_0 = const()[name = string("op_3924_interleave_0"), val = bool(false)]; + tensor var_3924_cast_fp16 = concat(axis = var_3035, interleave = var_3924_interleave_0, values = (var_3828_cast_fp16, var_3830_cast_fp16, var_3832_cast_fp16, var_3834_cast_fp16))[name = string("op_3924_cast_fp16")]; + bool var_3926_interleave_0 = const()[name = string("op_3926_interleave_0"), val = bool(false)]; + tensor var_3926_cast_fp16 = concat(axis = var_3035, interleave = var_3926_interleave_0, values = (var_3836_cast_fp16, var_3838_cast_fp16, var_3840_cast_fp16, var_3842_cast_fp16))[name = string("op_3926_cast_fp16")]; + bool var_3928_interleave_0 = const()[name = string("op_3928_interleave_0"), val = bool(false)]; + tensor var_3928_cast_fp16 = concat(axis = var_3035, interleave = var_3928_interleave_0, values = (var_3844_cast_fp16, var_3846_cast_fp16, var_3848_cast_fp16, var_3850_cast_fp16))[name = string("op_3928_cast_fp16")]; + bool var_3930_interleave_0 = const()[name = string("op_3930_interleave_0"), val = bool(false)]; + tensor var_3930_cast_fp16 = concat(axis = var_3035, interleave = var_3930_interleave_0, values = (var_3852_cast_fp16, var_3854_cast_fp16, var_3856_cast_fp16, var_3858_cast_fp16))[name = string("op_3930_cast_fp16")]; + bool var_3932_interleave_0 = const()[name = string("op_3932_interleave_0"), val = bool(false)]; + tensor var_3932_cast_fp16 = concat(axis = var_3035, interleave = var_3932_interleave_0, values = (var_3860_cast_fp16, var_3862_cast_fp16, var_3864_cast_fp16, var_3866_cast_fp16))[name = string("op_3932_cast_fp16")]; + bool var_3934_interleave_0 = const()[name = string("op_3934_interleave_0"), val = bool(false)]; + tensor var_3934_cast_fp16 = concat(axis = var_3035, interleave = var_3934_interleave_0, values = (var_3868_cast_fp16, var_3870_cast_fp16, var_3872_cast_fp16, var_3874_cast_fp16))[name = string("op_3934_cast_fp16")]; + bool var_3936_interleave_0 = const()[name = string("op_3936_interleave_0"), val = bool(false)]; + tensor var_3936_cast_fp16 = concat(axis = var_3035, interleave = var_3936_interleave_0, values = (var_3876_cast_fp16, var_3878_cast_fp16, var_3880_cast_fp16, var_3882_cast_fp16))[name = string("op_3936_cast_fp16")]; + bool var_3938_interleave_0 = const()[name = string("op_3938_interleave_0"), val = bool(false)]; + tensor var_3938_cast_fp16 = concat(axis = var_3035, interleave = var_3938_interleave_0, values = (var_3884_cast_fp16, var_3886_cast_fp16, var_3888_cast_fp16, var_3890_cast_fp16))[name = string("op_3938_cast_fp16")]; + bool var_3940_interleave_0 = const()[name = string("op_3940_interleave_0"), val = bool(false)]; + tensor var_3940_cast_fp16 = concat(axis = var_3035, interleave = var_3940_interleave_0, values = (var_3892_cast_fp16, var_3894_cast_fp16, var_3896_cast_fp16, var_3898_cast_fp16))[name = string("op_3940_cast_fp16")]; + bool var_3942_interleave_0 = const()[name = string("op_3942_interleave_0"), val = bool(false)]; + tensor var_3942_cast_fp16 = concat(axis = var_3035, interleave = var_3942_interleave_0, values = (var_3900_cast_fp16, var_3902_cast_fp16, var_3904_cast_fp16, var_3906_cast_fp16))[name = string("op_3942_cast_fp16")]; + bool var_3944_interleave_0 = const()[name = string("op_3944_interleave_0"), val = bool(false)]; + tensor var_3944_cast_fp16 = concat(axis = var_3035, interleave = var_3944_interleave_0, values = (var_3908_cast_fp16, var_3910_cast_fp16, var_3912_cast_fp16, var_3914_cast_fp16))[name = string("op_3944_cast_fp16")]; + bool var_3946_interleave_0 = const()[name = string("op_3946_interleave_0"), val = bool(false)]; + tensor var_3946_cast_fp16 = concat(axis = var_3035, interleave = var_3946_interleave_0, values = (var_3916_cast_fp16, var_3918_cast_fp16, var_3920_cast_fp16, var_3922_cast_fp16))[name = string("op_3946_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_3052, interleave = input_25_interleave_0, values = (var_3924_cast_fp16, var_3926_cast_fp16, var_3928_cast_fp16, var_3930_cast_fp16, var_3932_cast_fp16, var_3934_cast_fp16, var_3936_cast_fp16, var_3938_cast_fp16, var_3940_cast_fp16, var_3942_cast_fp16, var_3944_cast_fp16, var_3946_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52289280)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53468992)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_3965_to_fp16 = const()[name = string("op_3965_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_3965_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53470592)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53472192)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53473792)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58192448)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58198656)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62917312)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_3994 = const()[name = string("op_3994"), val = int32(3)]; + int32 var_4011 = const()[name = string("op_4011"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_4028_to_fp16 = const()[name = string("op_4028_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_4028_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62918912)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62920512)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62922112)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64101824)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64103424)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65283136)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66462848)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_4066_begin_0 = const()[name = string("op_4066_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4066_end_0 = const()[name = string("op_4066_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4066_end_mask_0 = const()[name = string("op_4066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4066_cast_fp16 = slice_by_index(begin = var_4066_begin_0, end = var_4066_end_0, end_mask = var_4066_end_mask_0, x = query_9_cast_fp16)[name = string("op_4066_cast_fp16")]; + tensor var_4070_begin_0 = const()[name = string("op_4070_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4070_end_0 = const()[name = string("op_4070_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4070_end_mask_0 = const()[name = string("op_4070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4070_cast_fp16 = slice_by_index(begin = var_4070_begin_0, end = var_4070_end_0, end_mask = var_4070_end_mask_0, x = query_9_cast_fp16)[name = string("op_4070_cast_fp16")]; + tensor var_4074_begin_0 = const()[name = string("op_4074_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4074_end_0 = const()[name = string("op_4074_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4074_end_mask_0 = const()[name = string("op_4074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4074_cast_fp16 = slice_by_index(begin = var_4074_begin_0, end = var_4074_end_0, end_mask = var_4074_end_mask_0, x = query_9_cast_fp16)[name = string("op_4074_cast_fp16")]; + tensor var_4078_begin_0 = const()[name = string("op_4078_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4078_end_0 = const()[name = string("op_4078_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4078_end_mask_0 = const()[name = string("op_4078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4078_cast_fp16 = slice_by_index(begin = var_4078_begin_0, end = var_4078_end_0, end_mask = var_4078_end_mask_0, x = query_9_cast_fp16)[name = string("op_4078_cast_fp16")]; + tensor var_4082_begin_0 = const()[name = string("op_4082_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4082_end_0 = const()[name = string("op_4082_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4082_end_mask_0 = const()[name = string("op_4082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4082_cast_fp16 = slice_by_index(begin = var_4082_begin_0, end = var_4082_end_0, end_mask = var_4082_end_mask_0, x = query_9_cast_fp16)[name = string("op_4082_cast_fp16")]; + tensor var_4086_begin_0 = const()[name = string("op_4086_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4086_end_0 = const()[name = string("op_4086_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4086_end_mask_0 = const()[name = string("op_4086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4086_cast_fp16 = slice_by_index(begin = var_4086_begin_0, end = var_4086_end_0, end_mask = var_4086_end_mask_0, x = query_9_cast_fp16)[name = string("op_4086_cast_fp16")]; + tensor var_4090_begin_0 = const()[name = string("op_4090_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4090_end_0 = const()[name = string("op_4090_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4090_end_mask_0 = const()[name = string("op_4090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4090_cast_fp16 = slice_by_index(begin = var_4090_begin_0, end = var_4090_end_0, end_mask = var_4090_end_mask_0, x = query_9_cast_fp16)[name = string("op_4090_cast_fp16")]; + tensor var_4094_begin_0 = const()[name = string("op_4094_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4094_end_0 = const()[name = string("op_4094_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4094_end_mask_0 = const()[name = string("op_4094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4094_cast_fp16 = slice_by_index(begin = var_4094_begin_0, end = var_4094_end_0, end_mask = var_4094_end_mask_0, x = query_9_cast_fp16)[name = string("op_4094_cast_fp16")]; + tensor var_4098_begin_0 = const()[name = string("op_4098_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4098_end_0 = const()[name = string("op_4098_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4098_end_mask_0 = const()[name = string("op_4098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4098_cast_fp16 = slice_by_index(begin = var_4098_begin_0, end = var_4098_end_0, end_mask = var_4098_end_mask_0, x = query_9_cast_fp16)[name = string("op_4098_cast_fp16")]; + tensor var_4102_begin_0 = const()[name = string("op_4102_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4102_end_0 = const()[name = string("op_4102_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4102_end_mask_0 = const()[name = string("op_4102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4102_cast_fp16 = slice_by_index(begin = var_4102_begin_0, end = var_4102_end_0, end_mask = var_4102_end_mask_0, x = query_9_cast_fp16)[name = string("op_4102_cast_fp16")]; + tensor var_4106_begin_0 = const()[name = string("op_4106_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4106_end_0 = const()[name = string("op_4106_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4106_end_mask_0 = const()[name = string("op_4106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4106_cast_fp16 = slice_by_index(begin = var_4106_begin_0, end = var_4106_end_0, end_mask = var_4106_end_mask_0, x = query_9_cast_fp16)[name = string("op_4106_cast_fp16")]; + tensor var_4110_begin_0 = const()[name = string("op_4110_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4110_end_0 = const()[name = string("op_4110_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4110_end_mask_0 = const()[name = string("op_4110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4110_cast_fp16 = slice_by_index(begin = var_4110_begin_0, end = var_4110_end_0, end_mask = var_4110_end_mask_0, x = query_9_cast_fp16)[name = string("op_4110_cast_fp16")]; + tensor var_4119_begin_0 = const()[name = string("op_4119_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4119_end_0 = const()[name = string("op_4119_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4119_end_mask_0 = const()[name = string("op_4119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4119_cast_fp16 = slice_by_index(begin = var_4119_begin_0, end = var_4119_end_0, end_mask = var_4119_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4119_cast_fp16")]; + tensor var_4126_begin_0 = const()[name = string("op_4126_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4126_end_0 = const()[name = string("op_4126_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4126_end_mask_0 = const()[name = string("op_4126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4126_cast_fp16 = slice_by_index(begin = var_4126_begin_0, end = var_4126_end_0, end_mask = var_4126_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4126_cast_fp16")]; + tensor var_4133_begin_0 = const()[name = string("op_4133_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4133_end_0 = const()[name = string("op_4133_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4133_end_mask_0 = const()[name = string("op_4133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4133_cast_fp16 = slice_by_index(begin = var_4133_begin_0, end = var_4133_end_0, end_mask = var_4133_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4133_cast_fp16")]; + tensor var_4140_begin_0 = const()[name = string("op_4140_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4140_end_0 = const()[name = string("op_4140_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4140_end_mask_0 = const()[name = string("op_4140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4140_cast_fp16 = slice_by_index(begin = var_4140_begin_0, end = var_4140_end_0, end_mask = var_4140_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4140_cast_fp16")]; + tensor var_4147_begin_0 = const()[name = string("op_4147_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4147_end_0 = const()[name = string("op_4147_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4147_end_mask_0 = const()[name = string("op_4147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4147_cast_fp16 = slice_by_index(begin = var_4147_begin_0, end = var_4147_end_0, end_mask = var_4147_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4147_cast_fp16")]; + tensor var_4154_begin_0 = const()[name = string("op_4154_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4154_end_0 = const()[name = string("op_4154_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4154_end_mask_0 = const()[name = string("op_4154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4154_cast_fp16 = slice_by_index(begin = var_4154_begin_0, end = var_4154_end_0, end_mask = var_4154_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4154_cast_fp16")]; + tensor var_4161_begin_0 = const()[name = string("op_4161_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4161_end_0 = const()[name = string("op_4161_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4161_end_mask_0 = const()[name = string("op_4161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4161_cast_fp16 = slice_by_index(begin = var_4161_begin_0, end = var_4161_end_0, end_mask = var_4161_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4161_cast_fp16")]; + tensor var_4168_begin_0 = const()[name = string("op_4168_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4168_end_0 = const()[name = string("op_4168_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4168_end_mask_0 = const()[name = string("op_4168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4168_cast_fp16 = slice_by_index(begin = var_4168_begin_0, end = var_4168_end_0, end_mask = var_4168_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4168_cast_fp16")]; + tensor var_4175_begin_0 = const()[name = string("op_4175_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4175_end_0 = const()[name = string("op_4175_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4175_end_mask_0 = const()[name = string("op_4175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4175_cast_fp16 = slice_by_index(begin = var_4175_begin_0, end = var_4175_end_0, end_mask = var_4175_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4175_cast_fp16")]; + tensor var_4182_begin_0 = const()[name = string("op_4182_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4182_end_0 = const()[name = string("op_4182_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4182_end_mask_0 = const()[name = string("op_4182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4182_cast_fp16 = slice_by_index(begin = var_4182_begin_0, end = var_4182_end_0, end_mask = var_4182_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4182_cast_fp16")]; + tensor var_4189_begin_0 = const()[name = string("op_4189_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4189_end_0 = const()[name = string("op_4189_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4189_end_mask_0 = const()[name = string("op_4189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4189_cast_fp16 = slice_by_index(begin = var_4189_begin_0, end = var_4189_end_0, end_mask = var_4189_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4189_cast_fp16")]; + tensor var_4196_begin_0 = const()[name = string("op_4196_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4196_end_0 = const()[name = string("op_4196_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4196_end_mask_0 = const()[name = string("op_4196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4196_cast_fp16 = slice_by_index(begin = var_4196_begin_0, end = var_4196_end_0, end_mask = var_4196_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4196_cast_fp16")]; + tensor var_4203_begin_0 = const()[name = string("op_4203_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4203_end_0 = const()[name = string("op_4203_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4203_end_mask_0 = const()[name = string("op_4203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4203_cast_fp16 = slice_by_index(begin = var_4203_begin_0, end = var_4203_end_0, end_mask = var_4203_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4203_cast_fp16")]; + tensor var_4210_begin_0 = const()[name = string("op_4210_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4210_end_0 = const()[name = string("op_4210_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4210_end_mask_0 = const()[name = string("op_4210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4210_cast_fp16 = slice_by_index(begin = var_4210_begin_0, end = var_4210_end_0, end_mask = var_4210_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4210_cast_fp16")]; + tensor var_4217_begin_0 = const()[name = string("op_4217_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4217_end_0 = const()[name = string("op_4217_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4217_end_mask_0 = const()[name = string("op_4217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4217_cast_fp16 = slice_by_index(begin = var_4217_begin_0, end = var_4217_end_0, end_mask = var_4217_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4217_cast_fp16")]; + tensor var_4224_begin_0 = const()[name = string("op_4224_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4224_end_0 = const()[name = string("op_4224_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4224_end_mask_0 = const()[name = string("op_4224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4224_cast_fp16 = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4224_cast_fp16")]; + tensor var_4231_begin_0 = const()[name = string("op_4231_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4231_end_0 = const()[name = string("op_4231_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4231_end_mask_0 = const()[name = string("op_4231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4231_cast_fp16 = slice_by_index(begin = var_4231_begin_0, end = var_4231_end_0, end_mask = var_4231_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4231_cast_fp16")]; + tensor var_4238_begin_0 = const()[name = string("op_4238_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4238_end_0 = const()[name = string("op_4238_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4238_end_mask_0 = const()[name = string("op_4238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4238_cast_fp16 = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4238_cast_fp16")]; + tensor var_4245_begin_0 = const()[name = string("op_4245_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4245_end_0 = const()[name = string("op_4245_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4245_end_mask_0 = const()[name = string("op_4245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4245_cast_fp16 = slice_by_index(begin = var_4245_begin_0, end = var_4245_end_0, end_mask = var_4245_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4245_cast_fp16")]; + tensor var_4252_begin_0 = const()[name = string("op_4252_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4252_end_0 = const()[name = string("op_4252_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4252_end_mask_0 = const()[name = string("op_4252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4252_cast_fp16 = slice_by_index(begin = var_4252_begin_0, end = var_4252_end_0, end_mask = var_4252_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4252_cast_fp16")]; + tensor var_4259_begin_0 = const()[name = string("op_4259_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4259_end_0 = const()[name = string("op_4259_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4259_end_mask_0 = const()[name = string("op_4259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4259_cast_fp16 = slice_by_index(begin = var_4259_begin_0, end = var_4259_end_0, end_mask = var_4259_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4259_cast_fp16")]; + tensor var_4266_begin_0 = const()[name = string("op_4266_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4266_end_0 = const()[name = string("op_4266_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4266_end_mask_0 = const()[name = string("op_4266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4266_cast_fp16")]; + tensor var_4273_begin_0 = const()[name = string("op_4273_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4273_end_0 = const()[name = string("op_4273_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4273_end_mask_0 = const()[name = string("op_4273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4273_cast_fp16 = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4273_cast_fp16")]; + tensor var_4280_begin_0 = const()[name = string("op_4280_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4280_end_0 = const()[name = string("op_4280_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4280_end_mask_0 = const()[name = string("op_4280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4280_cast_fp16")]; + tensor var_4287_begin_0 = const()[name = string("op_4287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4287_end_0 = const()[name = string("op_4287_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4287_end_mask_0 = const()[name = string("op_4287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4287_cast_fp16 = slice_by_index(begin = var_4287_begin_0, end = var_4287_end_0, end_mask = var_4287_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4287_cast_fp16")]; + tensor var_4294_begin_0 = const()[name = string("op_4294_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4294_end_0 = const()[name = string("op_4294_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4294_end_mask_0 = const()[name = string("op_4294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4294_cast_fp16 = slice_by_index(begin = var_4294_begin_0, end = var_4294_end_0, end_mask = var_4294_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4294_cast_fp16")]; + tensor var_4301_begin_0 = const()[name = string("op_4301_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4301_end_0 = const()[name = string("op_4301_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4301_end_mask_0 = const()[name = string("op_4301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4301_cast_fp16 = slice_by_index(begin = var_4301_begin_0, end = var_4301_end_0, end_mask = var_4301_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4301_cast_fp16")]; + tensor var_4308_begin_0 = const()[name = string("op_4308_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4308_end_0 = const()[name = string("op_4308_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4308_end_mask_0 = const()[name = string("op_4308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4308_cast_fp16 = slice_by_index(begin = var_4308_begin_0, end = var_4308_end_0, end_mask = var_4308_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4308_cast_fp16")]; + tensor var_4315_begin_0 = const()[name = string("op_4315_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4315_end_0 = const()[name = string("op_4315_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4315_end_mask_0 = const()[name = string("op_4315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4315_cast_fp16 = slice_by_index(begin = var_4315_begin_0, end = var_4315_end_0, end_mask = var_4315_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4315_cast_fp16")]; + tensor var_4322_begin_0 = const()[name = string("op_4322_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4322_end_0 = const()[name = string("op_4322_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4322_end_mask_0 = const()[name = string("op_4322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4322_cast_fp16 = slice_by_index(begin = var_4322_begin_0, end = var_4322_end_0, end_mask = var_4322_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4322_cast_fp16")]; + tensor var_4329_begin_0 = const()[name = string("op_4329_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4329_end_0 = const()[name = string("op_4329_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4329_end_mask_0 = const()[name = string("op_4329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4329_cast_fp16 = slice_by_index(begin = var_4329_begin_0, end = var_4329_end_0, end_mask = var_4329_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4329_cast_fp16")]; + tensor var_4336_begin_0 = const()[name = string("op_4336_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4336_end_0 = const()[name = string("op_4336_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4336_end_mask_0 = const()[name = string("op_4336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4336_cast_fp16 = slice_by_index(begin = var_4336_begin_0, end = var_4336_end_0, end_mask = var_4336_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4336_cast_fp16")]; + tensor var_4343_begin_0 = const()[name = string("op_4343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4343_end_0 = const()[name = string("op_4343_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4343_end_mask_0 = const()[name = string("op_4343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4343_cast_fp16 = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4343_cast_fp16")]; + tensor var_4350_begin_0 = const()[name = string("op_4350_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4350_end_0 = const()[name = string("op_4350_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4350_end_mask_0 = const()[name = string("op_4350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4350_cast_fp16 = slice_by_index(begin = var_4350_begin_0, end = var_4350_end_0, end_mask = var_4350_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4350_cast_fp16")]; + tensor var_4357_begin_0 = const()[name = string("op_4357_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4357_end_0 = const()[name = string("op_4357_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4357_end_mask_0 = const()[name = string("op_4357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4357_cast_fp16 = slice_by_index(begin = var_4357_begin_0, end = var_4357_end_0, end_mask = var_4357_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4357_cast_fp16")]; + tensor var_4364_begin_0 = const()[name = string("op_4364_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4364_end_0 = const()[name = string("op_4364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4364_end_mask_0 = const()[name = string("op_4364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4364_cast_fp16 = slice_by_index(begin = var_4364_begin_0, end = var_4364_end_0, end_mask = var_4364_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4364_cast_fp16")]; + tensor var_4371_begin_0 = const()[name = string("op_4371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4371_end_0 = const()[name = string("op_4371_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4371_end_mask_0 = const()[name = string("op_4371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4371_cast_fp16 = slice_by_index(begin = var_4371_begin_0, end = var_4371_end_0, end_mask = var_4371_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4371_cast_fp16")]; + tensor var_4378_begin_0 = const()[name = string("op_4378_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4378_end_0 = const()[name = string("op_4378_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4378_end_mask_0 = const()[name = string("op_4378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4378_cast_fp16 = slice_by_index(begin = var_4378_begin_0, end = var_4378_end_0, end_mask = var_4378_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4378_cast_fp16")]; + tensor var_4385_begin_0 = const()[name = string("op_4385_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4385_end_0 = const()[name = string("op_4385_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4385_end_mask_0 = const()[name = string("op_4385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4385_cast_fp16 = slice_by_index(begin = var_4385_begin_0, end = var_4385_end_0, end_mask = var_4385_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4385_cast_fp16")]; + tensor var_4392_begin_0 = const()[name = string("op_4392_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4392_end_0 = const()[name = string("op_4392_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4392_end_mask_0 = const()[name = string("op_4392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4392_cast_fp16 = slice_by_index(begin = var_4392_begin_0, end = var_4392_end_0, end_mask = var_4392_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4392_cast_fp16")]; + tensor var_4399_begin_0 = const()[name = string("op_4399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4399_end_0 = const()[name = string("op_4399_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4399_end_mask_0 = const()[name = string("op_4399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4399_cast_fp16 = slice_by_index(begin = var_4399_begin_0, end = var_4399_end_0, end_mask = var_4399_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4399_cast_fp16")]; + tensor var_4406_begin_0 = const()[name = string("op_4406_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4406_end_0 = const()[name = string("op_4406_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4406_end_mask_0 = const()[name = string("op_4406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4406_cast_fp16 = slice_by_index(begin = var_4406_begin_0, end = var_4406_end_0, end_mask = var_4406_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4406_cast_fp16")]; + tensor var_4413_begin_0 = const()[name = string("op_4413_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4413_end_0 = const()[name = string("op_4413_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4413_end_mask_0 = const()[name = string("op_4413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4413_cast_fp16 = slice_by_index(begin = var_4413_begin_0, end = var_4413_end_0, end_mask = var_4413_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4413_cast_fp16")]; + tensor var_4420_begin_0 = const()[name = string("op_4420_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4420_end_0 = const()[name = string("op_4420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4420_end_mask_0 = const()[name = string("op_4420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4420_cast_fp16 = slice_by_index(begin = var_4420_begin_0, end = var_4420_end_0, end_mask = var_4420_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4420_cast_fp16")]; + tensor var_4427_begin_0 = const()[name = string("op_4427_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4427_end_0 = const()[name = string("op_4427_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4427_end_mask_0 = const()[name = string("op_4427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4427_cast_fp16 = slice_by_index(begin = var_4427_begin_0, end = var_4427_end_0, end_mask = var_4427_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4427_cast_fp16")]; + tensor var_4434_begin_0 = const()[name = string("op_4434_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4434_end_0 = const()[name = string("op_4434_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4434_end_mask_0 = const()[name = string("op_4434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4434_cast_fp16 = slice_by_index(begin = var_4434_begin_0, end = var_4434_end_0, end_mask = var_4434_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4434_cast_fp16")]; + tensor var_4441_begin_0 = const()[name = string("op_4441_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4441_end_0 = const()[name = string("op_4441_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4441_end_mask_0 = const()[name = string("op_4441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4441_cast_fp16 = slice_by_index(begin = var_4441_begin_0, end = var_4441_end_0, end_mask = var_4441_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4441_cast_fp16")]; + tensor var_4448_begin_0 = const()[name = string("op_4448_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4448_end_0 = const()[name = string("op_4448_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4448_end_mask_0 = const()[name = string("op_4448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4448_cast_fp16 = slice_by_index(begin = var_4448_begin_0, end = var_4448_end_0, end_mask = var_4448_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4448_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4453_begin_0 = const()[name = string("op_4453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4453_end_0 = const()[name = string("op_4453_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_4453_end_mask_0 = const()[name = string("op_4453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_7")]; + tensor var_4453_cast_fp16 = slice_by_index(begin = var_4453_begin_0, end = var_4453_end_0, end_mask = var_4453_end_mask_0, x = k_9_cast_fp16)[name = string("op_4453_cast_fp16")]; + tensor var_4457_begin_0 = const()[name = string("op_4457_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4457_end_0 = const()[name = string("op_4457_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_4457_end_mask_0 = const()[name = string("op_4457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4457_cast_fp16 = slice_by_index(begin = var_4457_begin_0, end = var_4457_end_0, end_mask = var_4457_end_mask_0, x = k_9_cast_fp16)[name = string("op_4457_cast_fp16")]; + tensor var_4461_begin_0 = const()[name = string("op_4461_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4461_end_0 = const()[name = string("op_4461_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_4461_end_mask_0 = const()[name = string("op_4461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4461_cast_fp16 = slice_by_index(begin = var_4461_begin_0, end = var_4461_end_0, end_mask = var_4461_end_mask_0, x = k_9_cast_fp16)[name = string("op_4461_cast_fp16")]; + tensor var_4465_begin_0 = const()[name = string("op_4465_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4465_end_0 = const()[name = string("op_4465_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_4465_end_mask_0 = const()[name = string("op_4465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4465_cast_fp16 = slice_by_index(begin = var_4465_begin_0, end = var_4465_end_0, end_mask = var_4465_end_mask_0, x = k_9_cast_fp16)[name = string("op_4465_cast_fp16")]; + tensor var_4469_begin_0 = const()[name = string("op_4469_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4469_end_0 = const()[name = string("op_4469_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_4469_end_mask_0 = const()[name = string("op_4469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4469_cast_fp16 = slice_by_index(begin = var_4469_begin_0, end = var_4469_end_0, end_mask = var_4469_end_mask_0, x = k_9_cast_fp16)[name = string("op_4469_cast_fp16")]; + tensor var_4473_begin_0 = const()[name = string("op_4473_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4473_end_0 = const()[name = string("op_4473_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_4473_end_mask_0 = const()[name = string("op_4473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4473_cast_fp16 = slice_by_index(begin = var_4473_begin_0, end = var_4473_end_0, end_mask = var_4473_end_mask_0, x = k_9_cast_fp16)[name = string("op_4473_cast_fp16")]; + tensor var_4477_begin_0 = const()[name = string("op_4477_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4477_end_0 = const()[name = string("op_4477_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_4477_end_mask_0 = const()[name = string("op_4477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4477_cast_fp16 = slice_by_index(begin = var_4477_begin_0, end = var_4477_end_0, end_mask = var_4477_end_mask_0, x = k_9_cast_fp16)[name = string("op_4477_cast_fp16")]; + tensor var_4481_begin_0 = const()[name = string("op_4481_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4481_end_0 = const()[name = string("op_4481_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_4481_end_mask_0 = const()[name = string("op_4481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4481_cast_fp16 = slice_by_index(begin = var_4481_begin_0, end = var_4481_end_0, end_mask = var_4481_end_mask_0, x = k_9_cast_fp16)[name = string("op_4481_cast_fp16")]; + tensor var_4485_begin_0 = const()[name = string("op_4485_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4485_end_0 = const()[name = string("op_4485_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_4485_end_mask_0 = const()[name = string("op_4485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4485_cast_fp16 = slice_by_index(begin = var_4485_begin_0, end = var_4485_end_0, end_mask = var_4485_end_mask_0, x = k_9_cast_fp16)[name = string("op_4485_cast_fp16")]; + tensor var_4489_begin_0 = const()[name = string("op_4489_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4489_end_0 = const()[name = string("op_4489_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_4489_end_mask_0 = const()[name = string("op_4489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4489_cast_fp16 = slice_by_index(begin = var_4489_begin_0, end = var_4489_end_0, end_mask = var_4489_end_mask_0, x = k_9_cast_fp16)[name = string("op_4489_cast_fp16")]; + tensor var_4493_begin_0 = const()[name = string("op_4493_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4493_end_0 = const()[name = string("op_4493_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_4493_end_mask_0 = const()[name = string("op_4493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4493_cast_fp16 = slice_by_index(begin = var_4493_begin_0, end = var_4493_end_0, end_mask = var_4493_end_mask_0, x = k_9_cast_fp16)[name = string("op_4493_cast_fp16")]; + tensor var_4497_begin_0 = const()[name = string("op_4497_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4497_end_0 = const()[name = string("op_4497_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_4497_end_mask_0 = const()[name = string("op_4497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4497_cast_fp16 = slice_by_index(begin = var_4497_begin_0, end = var_4497_end_0, end_mask = var_4497_end_mask_0, x = k_9_cast_fp16)[name = string("op_4497_cast_fp16")]; + tensor var_4499_begin_0 = const()[name = string("op_4499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4499_end_0 = const()[name = string("op_4499_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4499_end_mask_0 = const()[name = string("op_4499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4499_cast_fp16 = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = value_9_cast_fp16)[name = string("op_4499_cast_fp16")]; + tensor var_4503_begin_0 = const()[name = string("op_4503_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4503_end_0 = const()[name = string("op_4503_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4503_end_mask_0 = const()[name = string("op_4503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4503_cast_fp16 = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = value_9_cast_fp16)[name = string("op_4503_cast_fp16")]; + tensor var_4507_begin_0 = const()[name = string("op_4507_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4507_end_0 = const()[name = string("op_4507_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4507_end_mask_0 = const()[name = string("op_4507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4507_cast_fp16 = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = value_9_cast_fp16)[name = string("op_4507_cast_fp16")]; + tensor var_4511_begin_0 = const()[name = string("op_4511_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4511_end_0 = const()[name = string("op_4511_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4511_end_mask_0 = const()[name = string("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4511_cast_fp16 = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = value_9_cast_fp16)[name = string("op_4511_cast_fp16")]; + tensor var_4515_begin_0 = const()[name = string("op_4515_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4515_end_0 = const()[name = string("op_4515_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4515_end_mask_0 = const()[name = string("op_4515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4515_cast_fp16 = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = value_9_cast_fp16)[name = string("op_4515_cast_fp16")]; + tensor var_4519_begin_0 = const()[name = string("op_4519_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4519_end_0 = const()[name = string("op_4519_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4519_end_mask_0 = const()[name = string("op_4519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4519_cast_fp16 = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = value_9_cast_fp16)[name = string("op_4519_cast_fp16")]; + tensor var_4523_begin_0 = const()[name = string("op_4523_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4523_end_0 = const()[name = string("op_4523_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4523_end_mask_0 = const()[name = string("op_4523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4523_cast_fp16 = slice_by_index(begin = var_4523_begin_0, end = var_4523_end_0, end_mask = var_4523_end_mask_0, x = value_9_cast_fp16)[name = string("op_4523_cast_fp16")]; + tensor var_4527_begin_0 = const()[name = string("op_4527_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4527_end_0 = const()[name = string("op_4527_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4527_end_mask_0 = const()[name = string("op_4527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4527_cast_fp16 = slice_by_index(begin = var_4527_begin_0, end = var_4527_end_0, end_mask = var_4527_end_mask_0, x = value_9_cast_fp16)[name = string("op_4527_cast_fp16")]; + tensor var_4531_begin_0 = const()[name = string("op_4531_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4531_end_0 = const()[name = string("op_4531_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4531_end_mask_0 = const()[name = string("op_4531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4531_cast_fp16 = slice_by_index(begin = var_4531_begin_0, end = var_4531_end_0, end_mask = var_4531_end_mask_0, x = value_9_cast_fp16)[name = string("op_4531_cast_fp16")]; + tensor var_4535_begin_0 = const()[name = string("op_4535_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4535_end_0 = const()[name = string("op_4535_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4535_end_mask_0 = const()[name = string("op_4535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4535_cast_fp16 = slice_by_index(begin = var_4535_begin_0, end = var_4535_end_0, end_mask = var_4535_end_mask_0, x = value_9_cast_fp16)[name = string("op_4535_cast_fp16")]; + tensor var_4539_begin_0 = const()[name = string("op_4539_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4539_end_0 = const()[name = string("op_4539_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4539_end_mask_0 = const()[name = string("op_4539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4539_cast_fp16 = slice_by_index(begin = var_4539_begin_0, end = var_4539_end_0, end_mask = var_4539_end_mask_0, x = value_9_cast_fp16)[name = string("op_4539_cast_fp16")]; + tensor var_4543_begin_0 = const()[name = string("op_4543_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4543_end_0 = const()[name = string("op_4543_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4543_end_mask_0 = const()[name = string("op_4543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4543_cast_fp16 = slice_by_index(begin = var_4543_begin_0, end = var_4543_end_0, end_mask = var_4543_end_mask_0, x = value_9_cast_fp16)[name = string("op_4543_cast_fp16")]; + string _SplitHeadsQ__mh_w_385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_385_equation_0, values = (var_4453_cast_fp16, var_4119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_385_cast_fp16")]; + string _SplitHeadsQ__mh_w_387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_387_equation_0, values = (var_4453_cast_fp16, var_4126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_387_cast_fp16")]; + string _SplitHeadsQ__mh_w_389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_389_equation_0, values = (var_4453_cast_fp16, var_4133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_389_cast_fp16")]; + string _SplitHeadsQ__mh_w_391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_391_equation_0, values = (var_4453_cast_fp16, var_4140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_391_cast_fp16")]; + string _SplitHeadsQ__mh_w_393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_393_equation_0, values = (var_4457_cast_fp16, var_4147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_393_cast_fp16")]; + string _SplitHeadsQ__mh_w_395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_395_equation_0, values = (var_4457_cast_fp16, var_4154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_395_cast_fp16")]; + string _SplitHeadsQ__mh_w_397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_397_equation_0, values = (var_4457_cast_fp16, var_4161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_397_cast_fp16")]; + string _SplitHeadsQ__mh_w_399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_399_equation_0, values = (var_4457_cast_fp16, var_4168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_399_cast_fp16")]; + string _SplitHeadsQ__mh_w_401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_401_equation_0, values = (var_4461_cast_fp16, var_4175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_401_cast_fp16")]; + string _SplitHeadsQ__mh_w_403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_403_equation_0, values = (var_4461_cast_fp16, var_4182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_403_cast_fp16")]; + string _SplitHeadsQ__mh_w_405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_405_equation_0, values = (var_4461_cast_fp16, var_4189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_405_cast_fp16")]; + string _SplitHeadsQ__mh_w_407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_407_equation_0, values = (var_4461_cast_fp16, var_4196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_407_cast_fp16")]; + string _SplitHeadsQ__mh_w_409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_409_equation_0, values = (var_4465_cast_fp16, var_4203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_409_cast_fp16")]; + string _SplitHeadsQ__mh_w_411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_411_equation_0, values = (var_4465_cast_fp16, var_4210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_411_cast_fp16")]; + string _SplitHeadsQ__mh_w_413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_413_equation_0, values = (var_4465_cast_fp16, var_4217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_413_cast_fp16")]; + string _SplitHeadsQ__mh_w_415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_415_equation_0, values = (var_4465_cast_fp16, var_4224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_415_cast_fp16")]; + string _SplitHeadsQ__mh_w_417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_417_equation_0, values = (var_4469_cast_fp16, var_4231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_417_cast_fp16")]; + string _SplitHeadsQ__mh_w_419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_419_equation_0, values = (var_4469_cast_fp16, var_4238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_419_cast_fp16")]; + string _SplitHeadsQ__mh_w_421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_421_equation_0, values = (var_4469_cast_fp16, var_4245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_421_cast_fp16")]; + string _SplitHeadsQ__mh_w_423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_423_equation_0, values = (var_4469_cast_fp16, var_4252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_423_cast_fp16")]; + string _SplitHeadsQ__mh_w_425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_425_equation_0, values = (var_4473_cast_fp16, var_4259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_425_cast_fp16")]; + string _SplitHeadsQ__mh_w_427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_427_equation_0, values = (var_4473_cast_fp16, var_4266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_427_cast_fp16")]; + string _SplitHeadsQ__mh_w_429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_429_equation_0, values = (var_4473_cast_fp16, var_4273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_429_cast_fp16")]; + string _SplitHeadsQ__mh_w_431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_431_equation_0, values = (var_4473_cast_fp16, var_4280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_431_cast_fp16")]; + string _SplitHeadsQ__mh_w_433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_433_equation_0, values = (var_4477_cast_fp16, var_4287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_433_cast_fp16")]; + string _SplitHeadsQ__mh_w_435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_435_equation_0, values = (var_4477_cast_fp16, var_4294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_435_cast_fp16")]; + string _SplitHeadsQ__mh_w_437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_437_equation_0, values = (var_4477_cast_fp16, var_4301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_437_cast_fp16")]; + string _SplitHeadsQ__mh_w_439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_439_equation_0, values = (var_4477_cast_fp16, var_4308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_439_cast_fp16")]; + string _SplitHeadsQ__mh_w_441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_441_equation_0, values = (var_4481_cast_fp16, var_4315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_441_cast_fp16")]; + string _SplitHeadsQ__mh_w_443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_443_equation_0, values = (var_4481_cast_fp16, var_4322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_443_cast_fp16")]; + string _SplitHeadsQ__mh_w_445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_445_equation_0, values = (var_4481_cast_fp16, var_4329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_445_cast_fp16")]; + string _SplitHeadsQ__mh_w_447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_447_equation_0, values = (var_4481_cast_fp16, var_4336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_447_cast_fp16")]; + string _SplitHeadsQ__mh_w_449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_449_equation_0, values = (var_4485_cast_fp16, var_4343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_449_cast_fp16")]; + string _SplitHeadsQ__mh_w_451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_451_equation_0, values = (var_4485_cast_fp16, var_4350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_451_cast_fp16")]; + string _SplitHeadsQ__mh_w_453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_453_equation_0, values = (var_4485_cast_fp16, var_4357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_453_cast_fp16")]; + string _SplitHeadsQ__mh_w_455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_455_equation_0, values = (var_4485_cast_fp16, var_4364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_455_cast_fp16")]; + string _SplitHeadsQ__mh_w_457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_457_equation_0, values = (var_4489_cast_fp16, var_4371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_457_cast_fp16")]; + string _SplitHeadsQ__mh_w_459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_459_equation_0, values = (var_4489_cast_fp16, var_4378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_459_cast_fp16")]; + string _SplitHeadsQ__mh_w_461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_461_equation_0, values = (var_4489_cast_fp16, var_4385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_461_cast_fp16")]; + string _SplitHeadsQ__mh_w_463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_463_equation_0, values = (var_4489_cast_fp16, var_4392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_463_cast_fp16")]; + string _SplitHeadsQ__mh_w_465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_465_equation_0, values = (var_4493_cast_fp16, var_4399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_465_cast_fp16")]; + string _SplitHeadsQ__mh_w_467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_467_equation_0, values = (var_4493_cast_fp16, var_4406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_467_cast_fp16")]; + string _SplitHeadsQ__mh_w_469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_469_equation_0, values = (var_4493_cast_fp16, var_4413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_469_cast_fp16")]; + string _SplitHeadsQ__mh_w_471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_471_equation_0, values = (var_4493_cast_fp16, var_4420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_471_cast_fp16")]; + string _SplitHeadsQ__mh_w_473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_473_equation_0, values = (var_4497_cast_fp16, var_4427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_473_cast_fp16")]; + string _SplitHeadsQ__mh_w_475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_475_equation_0, values = (var_4497_cast_fp16, var_4434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_475_cast_fp16")]; + string _SplitHeadsQ__mh_w_477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_477_equation_0, values = (var_4497_cast_fp16, var_4441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_477_cast_fp16")]; + string _SplitHeadsQ__mh_w_479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_479_equation_0, values = (var_4497_cast_fp16, var_4448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_479_cast_fp16")]; + fp16 var_4642_to_fp16 = const()[name = string("op_4642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_385_cast_fp16, y = var_4642_to_fp16)[name = string("aw_chunk_385_cast_fp16")]; + fp16 var_4644_to_fp16 = const()[name = string("op_4644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_387_cast_fp16, y = var_4644_to_fp16)[name = string("aw_chunk_387_cast_fp16")]; + fp16 var_4646_to_fp16 = const()[name = string("op_4646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_389_cast_fp16, y = var_4646_to_fp16)[name = string("aw_chunk_389_cast_fp16")]; + fp16 var_4648_to_fp16 = const()[name = string("op_4648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_391_cast_fp16, y = var_4648_to_fp16)[name = string("aw_chunk_391_cast_fp16")]; + fp16 var_4650_to_fp16 = const()[name = string("op_4650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_393_cast_fp16, y = var_4650_to_fp16)[name = string("aw_chunk_393_cast_fp16")]; + fp16 var_4652_to_fp16 = const()[name = string("op_4652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_395_cast_fp16, y = var_4652_to_fp16)[name = string("aw_chunk_395_cast_fp16")]; + fp16 var_4654_to_fp16 = const()[name = string("op_4654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_397_cast_fp16, y = var_4654_to_fp16)[name = string("aw_chunk_397_cast_fp16")]; + fp16 var_4656_to_fp16 = const()[name = string("op_4656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_399_cast_fp16, y = var_4656_to_fp16)[name = string("aw_chunk_399_cast_fp16")]; + fp16 var_4658_to_fp16 = const()[name = string("op_4658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_401_cast_fp16, y = var_4658_to_fp16)[name = string("aw_chunk_401_cast_fp16")]; + fp16 var_4660_to_fp16 = const()[name = string("op_4660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_403_cast_fp16, y = var_4660_to_fp16)[name = string("aw_chunk_403_cast_fp16")]; + fp16 var_4662_to_fp16 = const()[name = string("op_4662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_405_cast_fp16, y = var_4662_to_fp16)[name = string("aw_chunk_405_cast_fp16")]; + fp16 var_4664_to_fp16 = const()[name = string("op_4664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_407_cast_fp16, y = var_4664_to_fp16)[name = string("aw_chunk_407_cast_fp16")]; + fp16 var_4666_to_fp16 = const()[name = string("op_4666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_409_cast_fp16, y = var_4666_to_fp16)[name = string("aw_chunk_409_cast_fp16")]; + fp16 var_4668_to_fp16 = const()[name = string("op_4668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_411_cast_fp16, y = var_4668_to_fp16)[name = string("aw_chunk_411_cast_fp16")]; + fp16 var_4670_to_fp16 = const()[name = string("op_4670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_413_cast_fp16, y = var_4670_to_fp16)[name = string("aw_chunk_413_cast_fp16")]; + fp16 var_4672_to_fp16 = const()[name = string("op_4672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_415_cast_fp16, y = var_4672_to_fp16)[name = string("aw_chunk_415_cast_fp16")]; + fp16 var_4674_to_fp16 = const()[name = string("op_4674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_417_cast_fp16, y = var_4674_to_fp16)[name = string("aw_chunk_417_cast_fp16")]; + fp16 var_4676_to_fp16 = const()[name = string("op_4676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_419_cast_fp16, y = var_4676_to_fp16)[name = string("aw_chunk_419_cast_fp16")]; + fp16 var_4678_to_fp16 = const()[name = string("op_4678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_421_cast_fp16, y = var_4678_to_fp16)[name = string("aw_chunk_421_cast_fp16")]; + fp16 var_4680_to_fp16 = const()[name = string("op_4680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_423_cast_fp16, y = var_4680_to_fp16)[name = string("aw_chunk_423_cast_fp16")]; + fp16 var_4682_to_fp16 = const()[name = string("op_4682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_425_cast_fp16, y = var_4682_to_fp16)[name = string("aw_chunk_425_cast_fp16")]; + fp16 var_4684_to_fp16 = const()[name = string("op_4684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_427_cast_fp16, y = var_4684_to_fp16)[name = string("aw_chunk_427_cast_fp16")]; + fp16 var_4686_to_fp16 = const()[name = string("op_4686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_429_cast_fp16, y = var_4686_to_fp16)[name = string("aw_chunk_429_cast_fp16")]; + fp16 var_4688_to_fp16 = const()[name = string("op_4688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_431_cast_fp16, y = var_4688_to_fp16)[name = string("aw_chunk_431_cast_fp16")]; + fp16 var_4690_to_fp16 = const()[name = string("op_4690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_433_cast_fp16, y = var_4690_to_fp16)[name = string("aw_chunk_433_cast_fp16")]; + fp16 var_4692_to_fp16 = const()[name = string("op_4692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_435_cast_fp16, y = var_4692_to_fp16)[name = string("aw_chunk_435_cast_fp16")]; + fp16 var_4694_to_fp16 = const()[name = string("op_4694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_437_cast_fp16, y = var_4694_to_fp16)[name = string("aw_chunk_437_cast_fp16")]; + fp16 var_4696_to_fp16 = const()[name = string("op_4696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_439_cast_fp16, y = var_4696_to_fp16)[name = string("aw_chunk_439_cast_fp16")]; + fp16 var_4698_to_fp16 = const()[name = string("op_4698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_441_cast_fp16, y = var_4698_to_fp16)[name = string("aw_chunk_441_cast_fp16")]; + fp16 var_4700_to_fp16 = const()[name = string("op_4700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_443_cast_fp16, y = var_4700_to_fp16)[name = string("aw_chunk_443_cast_fp16")]; + fp16 var_4702_to_fp16 = const()[name = string("op_4702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_445_cast_fp16, y = var_4702_to_fp16)[name = string("aw_chunk_445_cast_fp16")]; + fp16 var_4704_to_fp16 = const()[name = string("op_4704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_447_cast_fp16, y = var_4704_to_fp16)[name = string("aw_chunk_447_cast_fp16")]; + fp16 var_4706_to_fp16 = const()[name = string("op_4706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_449_cast_fp16, y = var_4706_to_fp16)[name = string("aw_chunk_449_cast_fp16")]; + fp16 var_4708_to_fp16 = const()[name = string("op_4708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_451_cast_fp16, y = var_4708_to_fp16)[name = string("aw_chunk_451_cast_fp16")]; + fp16 var_4710_to_fp16 = const()[name = string("op_4710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_453_cast_fp16, y = var_4710_to_fp16)[name = string("aw_chunk_453_cast_fp16")]; + fp16 var_4712_to_fp16 = const()[name = string("op_4712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_455_cast_fp16, y = var_4712_to_fp16)[name = string("aw_chunk_455_cast_fp16")]; + fp16 var_4714_to_fp16 = const()[name = string("op_4714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_457_cast_fp16, y = var_4714_to_fp16)[name = string("aw_chunk_457_cast_fp16")]; + fp16 var_4716_to_fp16 = const()[name = string("op_4716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_459_cast_fp16, y = var_4716_to_fp16)[name = string("aw_chunk_459_cast_fp16")]; + fp16 var_4718_to_fp16 = const()[name = string("op_4718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_461_cast_fp16, y = var_4718_to_fp16)[name = string("aw_chunk_461_cast_fp16")]; + fp16 var_4720_to_fp16 = const()[name = string("op_4720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_463_cast_fp16, y = var_4720_to_fp16)[name = string("aw_chunk_463_cast_fp16")]; + fp16 var_4722_to_fp16 = const()[name = string("op_4722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_465_cast_fp16, y = var_4722_to_fp16)[name = string("aw_chunk_465_cast_fp16")]; + fp16 var_4724_to_fp16 = const()[name = string("op_4724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_467_cast_fp16, y = var_4724_to_fp16)[name = string("aw_chunk_467_cast_fp16")]; + fp16 var_4726_to_fp16 = const()[name = string("op_4726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_469_cast_fp16, y = var_4726_to_fp16)[name = string("aw_chunk_469_cast_fp16")]; + fp16 var_4728_to_fp16 = const()[name = string("op_4728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_471_cast_fp16, y = var_4728_to_fp16)[name = string("aw_chunk_471_cast_fp16")]; + fp16 var_4730_to_fp16 = const()[name = string("op_4730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_473_cast_fp16, y = var_4730_to_fp16)[name = string("aw_chunk_473_cast_fp16")]; + fp16 var_4732_to_fp16 = const()[name = string("op_4732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_475_cast_fp16, y = var_4732_to_fp16)[name = string("aw_chunk_475_cast_fp16")]; + fp16 var_4734_to_fp16 = const()[name = string("op_4734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_477_cast_fp16, y = var_4734_to_fp16)[name = string("aw_chunk_477_cast_fp16")]; + fp16 var_4736_to_fp16 = const()[name = string("op_4736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_479_cast_fp16, y = var_4736_to_fp16)[name = string("aw_chunk_479_cast_fp16")]; + tensor var_4738_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_385_cast_fp16)[name = string("op_4738_cast_fp16")]; + tensor var_4739_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_387_cast_fp16)[name = string("op_4739_cast_fp16")]; + tensor var_4740_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_389_cast_fp16)[name = string("op_4740_cast_fp16")]; + tensor var_4741_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_391_cast_fp16)[name = string("op_4741_cast_fp16")]; + tensor var_4742_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_393_cast_fp16)[name = string("op_4742_cast_fp16")]; + tensor var_4743_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_395_cast_fp16)[name = string("op_4743_cast_fp16")]; + tensor var_4744_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_397_cast_fp16)[name = string("op_4744_cast_fp16")]; + tensor var_4745_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_399_cast_fp16)[name = string("op_4745_cast_fp16")]; + tensor var_4746_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_401_cast_fp16)[name = string("op_4746_cast_fp16")]; + tensor var_4747_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_403_cast_fp16)[name = string("op_4747_cast_fp16")]; + tensor var_4748_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_405_cast_fp16)[name = string("op_4748_cast_fp16")]; + tensor var_4749_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_407_cast_fp16)[name = string("op_4749_cast_fp16")]; + tensor var_4750_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_409_cast_fp16)[name = string("op_4750_cast_fp16")]; + tensor var_4751_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_411_cast_fp16)[name = string("op_4751_cast_fp16")]; + tensor var_4752_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_413_cast_fp16)[name = string("op_4752_cast_fp16")]; + tensor var_4753_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_415_cast_fp16)[name = string("op_4753_cast_fp16")]; + tensor var_4754_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_417_cast_fp16)[name = string("op_4754_cast_fp16")]; + tensor var_4755_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_419_cast_fp16)[name = string("op_4755_cast_fp16")]; + tensor var_4756_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_421_cast_fp16)[name = string("op_4756_cast_fp16")]; + tensor var_4757_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_423_cast_fp16)[name = string("op_4757_cast_fp16")]; + tensor var_4758_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_425_cast_fp16)[name = string("op_4758_cast_fp16")]; + tensor var_4759_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_427_cast_fp16)[name = string("op_4759_cast_fp16")]; + tensor var_4760_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_429_cast_fp16)[name = string("op_4760_cast_fp16")]; + tensor var_4761_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_431_cast_fp16)[name = string("op_4761_cast_fp16")]; + tensor var_4762_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_433_cast_fp16)[name = string("op_4762_cast_fp16")]; + tensor var_4763_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_435_cast_fp16)[name = string("op_4763_cast_fp16")]; + tensor var_4764_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_437_cast_fp16)[name = string("op_4764_cast_fp16")]; + tensor var_4765_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_439_cast_fp16)[name = string("op_4765_cast_fp16")]; + tensor var_4766_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_441_cast_fp16)[name = string("op_4766_cast_fp16")]; + tensor var_4767_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_443_cast_fp16)[name = string("op_4767_cast_fp16")]; + tensor var_4768_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_445_cast_fp16)[name = string("op_4768_cast_fp16")]; + tensor var_4769_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_447_cast_fp16)[name = string("op_4769_cast_fp16")]; + tensor var_4770_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_449_cast_fp16)[name = string("op_4770_cast_fp16")]; + tensor var_4771_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_451_cast_fp16)[name = string("op_4771_cast_fp16")]; + tensor var_4772_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_453_cast_fp16)[name = string("op_4772_cast_fp16")]; + tensor var_4773_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_455_cast_fp16)[name = string("op_4773_cast_fp16")]; + tensor var_4774_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_457_cast_fp16)[name = string("op_4774_cast_fp16")]; + tensor var_4775_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_459_cast_fp16)[name = string("op_4775_cast_fp16")]; + tensor var_4776_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_461_cast_fp16)[name = string("op_4776_cast_fp16")]; + tensor var_4777_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_463_cast_fp16)[name = string("op_4777_cast_fp16")]; + tensor var_4778_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_465_cast_fp16)[name = string("op_4778_cast_fp16")]; + tensor var_4779_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_467_cast_fp16)[name = string("op_4779_cast_fp16")]; + tensor var_4780_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_469_cast_fp16)[name = string("op_4780_cast_fp16")]; + tensor var_4781_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_471_cast_fp16)[name = string("op_4781_cast_fp16")]; + tensor var_4782_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_473_cast_fp16)[name = string("op_4782_cast_fp16")]; + tensor var_4783_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_475_cast_fp16)[name = string("op_4783_cast_fp16")]; + tensor var_4784_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_477_cast_fp16)[name = string("op_4784_cast_fp16")]; + tensor var_4785_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_479_cast_fp16)[name = string("op_4785_cast_fp16")]; + string var_4787_equation_0 = const()[name = string("op_4787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4787_cast_fp16 = einsum(equation = var_4787_equation_0, values = (var_4499_cast_fp16, var_4738_cast_fp16))[name = string("op_4787_cast_fp16")]; + string var_4789_equation_0 = const()[name = string("op_4789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4789_cast_fp16 = einsum(equation = var_4789_equation_0, values = (var_4499_cast_fp16, var_4739_cast_fp16))[name = string("op_4789_cast_fp16")]; + string var_4791_equation_0 = const()[name = string("op_4791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4791_cast_fp16 = einsum(equation = var_4791_equation_0, values = (var_4499_cast_fp16, var_4740_cast_fp16))[name = string("op_4791_cast_fp16")]; + string var_4793_equation_0 = const()[name = string("op_4793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4793_cast_fp16 = einsum(equation = var_4793_equation_0, values = (var_4499_cast_fp16, var_4741_cast_fp16))[name = string("op_4793_cast_fp16")]; + string var_4795_equation_0 = const()[name = string("op_4795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4795_cast_fp16 = einsum(equation = var_4795_equation_0, values = (var_4503_cast_fp16, var_4742_cast_fp16))[name = string("op_4795_cast_fp16")]; + string var_4797_equation_0 = const()[name = string("op_4797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4797_cast_fp16 = einsum(equation = var_4797_equation_0, values = (var_4503_cast_fp16, var_4743_cast_fp16))[name = string("op_4797_cast_fp16")]; + string var_4799_equation_0 = const()[name = string("op_4799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4799_cast_fp16 = einsum(equation = var_4799_equation_0, values = (var_4503_cast_fp16, var_4744_cast_fp16))[name = string("op_4799_cast_fp16")]; + string var_4801_equation_0 = const()[name = string("op_4801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4801_cast_fp16 = einsum(equation = var_4801_equation_0, values = (var_4503_cast_fp16, var_4745_cast_fp16))[name = string("op_4801_cast_fp16")]; + string var_4803_equation_0 = const()[name = string("op_4803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4803_cast_fp16 = einsum(equation = var_4803_equation_0, values = (var_4507_cast_fp16, var_4746_cast_fp16))[name = string("op_4803_cast_fp16")]; + string var_4805_equation_0 = const()[name = string("op_4805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4805_cast_fp16 = einsum(equation = var_4805_equation_0, values = (var_4507_cast_fp16, var_4747_cast_fp16))[name = string("op_4805_cast_fp16")]; + string var_4807_equation_0 = const()[name = string("op_4807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4807_cast_fp16 = einsum(equation = var_4807_equation_0, values = (var_4507_cast_fp16, var_4748_cast_fp16))[name = string("op_4807_cast_fp16")]; + string var_4809_equation_0 = const()[name = string("op_4809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4809_cast_fp16 = einsum(equation = var_4809_equation_0, values = (var_4507_cast_fp16, var_4749_cast_fp16))[name = string("op_4809_cast_fp16")]; + string var_4811_equation_0 = const()[name = string("op_4811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4811_cast_fp16 = einsum(equation = var_4811_equation_0, values = (var_4511_cast_fp16, var_4750_cast_fp16))[name = string("op_4811_cast_fp16")]; + string var_4813_equation_0 = const()[name = string("op_4813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4813_cast_fp16 = einsum(equation = var_4813_equation_0, values = (var_4511_cast_fp16, var_4751_cast_fp16))[name = string("op_4813_cast_fp16")]; + string var_4815_equation_0 = const()[name = string("op_4815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4815_cast_fp16 = einsum(equation = var_4815_equation_0, values = (var_4511_cast_fp16, var_4752_cast_fp16))[name = string("op_4815_cast_fp16")]; + string var_4817_equation_0 = const()[name = string("op_4817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4817_cast_fp16 = einsum(equation = var_4817_equation_0, values = (var_4511_cast_fp16, var_4753_cast_fp16))[name = string("op_4817_cast_fp16")]; + string var_4819_equation_0 = const()[name = string("op_4819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4819_cast_fp16 = einsum(equation = var_4819_equation_0, values = (var_4515_cast_fp16, var_4754_cast_fp16))[name = string("op_4819_cast_fp16")]; + string var_4821_equation_0 = const()[name = string("op_4821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4821_cast_fp16 = einsum(equation = var_4821_equation_0, values = (var_4515_cast_fp16, var_4755_cast_fp16))[name = string("op_4821_cast_fp16")]; + string var_4823_equation_0 = const()[name = string("op_4823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4823_cast_fp16 = einsum(equation = var_4823_equation_0, values = (var_4515_cast_fp16, var_4756_cast_fp16))[name = string("op_4823_cast_fp16")]; + string var_4825_equation_0 = const()[name = string("op_4825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4825_cast_fp16 = einsum(equation = var_4825_equation_0, values = (var_4515_cast_fp16, var_4757_cast_fp16))[name = string("op_4825_cast_fp16")]; + string var_4827_equation_0 = const()[name = string("op_4827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4827_cast_fp16 = einsum(equation = var_4827_equation_0, values = (var_4519_cast_fp16, var_4758_cast_fp16))[name = string("op_4827_cast_fp16")]; + string var_4829_equation_0 = const()[name = string("op_4829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4829_cast_fp16 = einsum(equation = var_4829_equation_0, values = (var_4519_cast_fp16, var_4759_cast_fp16))[name = string("op_4829_cast_fp16")]; + string var_4831_equation_0 = const()[name = string("op_4831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4519_cast_fp16, var_4760_cast_fp16))[name = string("op_4831_cast_fp16")]; + string var_4833_equation_0 = const()[name = string("op_4833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4833_cast_fp16 = einsum(equation = var_4833_equation_0, values = (var_4519_cast_fp16, var_4761_cast_fp16))[name = string("op_4833_cast_fp16")]; + string var_4835_equation_0 = const()[name = string("op_4835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4835_cast_fp16 = einsum(equation = var_4835_equation_0, values = (var_4523_cast_fp16, var_4762_cast_fp16))[name = string("op_4835_cast_fp16")]; + string var_4837_equation_0 = const()[name = string("op_4837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4837_cast_fp16 = einsum(equation = var_4837_equation_0, values = (var_4523_cast_fp16, var_4763_cast_fp16))[name = string("op_4837_cast_fp16")]; + string var_4839_equation_0 = const()[name = string("op_4839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4839_cast_fp16 = einsum(equation = var_4839_equation_0, values = (var_4523_cast_fp16, var_4764_cast_fp16))[name = string("op_4839_cast_fp16")]; + string var_4841_equation_0 = const()[name = string("op_4841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4841_cast_fp16 = einsum(equation = var_4841_equation_0, values = (var_4523_cast_fp16, var_4765_cast_fp16))[name = string("op_4841_cast_fp16")]; + string var_4843_equation_0 = const()[name = string("op_4843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4843_cast_fp16 = einsum(equation = var_4843_equation_0, values = (var_4527_cast_fp16, var_4766_cast_fp16))[name = string("op_4843_cast_fp16")]; + string var_4845_equation_0 = const()[name = string("op_4845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4845_cast_fp16 = einsum(equation = var_4845_equation_0, values = (var_4527_cast_fp16, var_4767_cast_fp16))[name = string("op_4845_cast_fp16")]; + string var_4847_equation_0 = const()[name = string("op_4847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4847_cast_fp16 = einsum(equation = var_4847_equation_0, values = (var_4527_cast_fp16, var_4768_cast_fp16))[name = string("op_4847_cast_fp16")]; + string var_4849_equation_0 = const()[name = string("op_4849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4849_cast_fp16 = einsum(equation = var_4849_equation_0, values = (var_4527_cast_fp16, var_4769_cast_fp16))[name = string("op_4849_cast_fp16")]; + string var_4851_equation_0 = const()[name = string("op_4851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4851_cast_fp16 = einsum(equation = var_4851_equation_0, values = (var_4531_cast_fp16, var_4770_cast_fp16))[name = string("op_4851_cast_fp16")]; + string var_4853_equation_0 = const()[name = string("op_4853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4853_cast_fp16 = einsum(equation = var_4853_equation_0, values = (var_4531_cast_fp16, var_4771_cast_fp16))[name = string("op_4853_cast_fp16")]; + string var_4855_equation_0 = const()[name = string("op_4855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4855_cast_fp16 = einsum(equation = var_4855_equation_0, values = (var_4531_cast_fp16, var_4772_cast_fp16))[name = string("op_4855_cast_fp16")]; + string var_4857_equation_0 = const()[name = string("op_4857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4857_cast_fp16 = einsum(equation = var_4857_equation_0, values = (var_4531_cast_fp16, var_4773_cast_fp16))[name = string("op_4857_cast_fp16")]; + string var_4859_equation_0 = const()[name = string("op_4859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4859_cast_fp16 = einsum(equation = var_4859_equation_0, values = (var_4535_cast_fp16, var_4774_cast_fp16))[name = string("op_4859_cast_fp16")]; + string var_4861_equation_0 = const()[name = string("op_4861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4861_cast_fp16 = einsum(equation = var_4861_equation_0, values = (var_4535_cast_fp16, var_4775_cast_fp16))[name = string("op_4861_cast_fp16")]; + string var_4863_equation_0 = const()[name = string("op_4863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4863_cast_fp16 = einsum(equation = var_4863_equation_0, values = (var_4535_cast_fp16, var_4776_cast_fp16))[name = string("op_4863_cast_fp16")]; + string var_4865_equation_0 = const()[name = string("op_4865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4865_cast_fp16 = einsum(equation = var_4865_equation_0, values = (var_4535_cast_fp16, var_4777_cast_fp16))[name = string("op_4865_cast_fp16")]; + string var_4867_equation_0 = const()[name = string("op_4867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4867_cast_fp16 = einsum(equation = var_4867_equation_0, values = (var_4539_cast_fp16, var_4778_cast_fp16))[name = string("op_4867_cast_fp16")]; + string var_4869_equation_0 = const()[name = string("op_4869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4869_cast_fp16 = einsum(equation = var_4869_equation_0, values = (var_4539_cast_fp16, var_4779_cast_fp16))[name = string("op_4869_cast_fp16")]; + string var_4871_equation_0 = const()[name = string("op_4871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4871_cast_fp16 = einsum(equation = var_4871_equation_0, values = (var_4539_cast_fp16, var_4780_cast_fp16))[name = string("op_4871_cast_fp16")]; + string var_4873_equation_0 = const()[name = string("op_4873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4873_cast_fp16 = einsum(equation = var_4873_equation_0, values = (var_4539_cast_fp16, var_4781_cast_fp16))[name = string("op_4873_cast_fp16")]; + string var_4875_equation_0 = const()[name = string("op_4875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4875_cast_fp16 = einsum(equation = var_4875_equation_0, values = (var_4543_cast_fp16, var_4782_cast_fp16))[name = string("op_4875_cast_fp16")]; + string var_4877_equation_0 = const()[name = string("op_4877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4877_cast_fp16 = einsum(equation = var_4877_equation_0, values = (var_4543_cast_fp16, var_4783_cast_fp16))[name = string("op_4877_cast_fp16")]; + string var_4879_equation_0 = const()[name = string("op_4879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4879_cast_fp16 = einsum(equation = var_4879_equation_0, values = (var_4543_cast_fp16, var_4784_cast_fp16))[name = string("op_4879_cast_fp16")]; + string var_4881_equation_0 = const()[name = string("op_4881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4881_cast_fp16 = einsum(equation = var_4881_equation_0, values = (var_4543_cast_fp16, var_4785_cast_fp16))[name = string("op_4881_cast_fp16")]; + bool var_4883_interleave_0 = const()[name = string("op_4883_interleave_0"), val = bool(false)]; + tensor var_4883_cast_fp16 = concat(axis = var_3994, interleave = var_4883_interleave_0, values = (var_4787_cast_fp16, var_4789_cast_fp16, var_4791_cast_fp16, var_4793_cast_fp16))[name = string("op_4883_cast_fp16")]; + bool var_4885_interleave_0 = const()[name = string("op_4885_interleave_0"), val = bool(false)]; + tensor var_4885_cast_fp16 = concat(axis = var_3994, interleave = var_4885_interleave_0, values = (var_4795_cast_fp16, var_4797_cast_fp16, var_4799_cast_fp16, var_4801_cast_fp16))[name = string("op_4885_cast_fp16")]; + bool var_4887_interleave_0 = const()[name = string("op_4887_interleave_0"), val = bool(false)]; + tensor var_4887_cast_fp16 = concat(axis = var_3994, interleave = var_4887_interleave_0, values = (var_4803_cast_fp16, var_4805_cast_fp16, var_4807_cast_fp16, var_4809_cast_fp16))[name = string("op_4887_cast_fp16")]; + bool var_4889_interleave_0 = const()[name = string("op_4889_interleave_0"), val = bool(false)]; + tensor var_4889_cast_fp16 = concat(axis = var_3994, interleave = var_4889_interleave_0, values = (var_4811_cast_fp16, var_4813_cast_fp16, var_4815_cast_fp16, var_4817_cast_fp16))[name = string("op_4889_cast_fp16")]; + bool var_4891_interleave_0 = const()[name = string("op_4891_interleave_0"), val = bool(false)]; + tensor var_4891_cast_fp16 = concat(axis = var_3994, interleave = var_4891_interleave_0, values = (var_4819_cast_fp16, var_4821_cast_fp16, var_4823_cast_fp16, var_4825_cast_fp16))[name = string("op_4891_cast_fp16")]; + bool var_4893_interleave_0 = const()[name = string("op_4893_interleave_0"), val = bool(false)]; + tensor var_4893_cast_fp16 = concat(axis = var_3994, interleave = var_4893_interleave_0, values = (var_4827_cast_fp16, var_4829_cast_fp16, var_4831_cast_fp16, var_4833_cast_fp16))[name = string("op_4893_cast_fp16")]; + bool var_4895_interleave_0 = const()[name = string("op_4895_interleave_0"), val = bool(false)]; + tensor var_4895_cast_fp16 = concat(axis = var_3994, interleave = var_4895_interleave_0, values = (var_4835_cast_fp16, var_4837_cast_fp16, var_4839_cast_fp16, var_4841_cast_fp16))[name = string("op_4895_cast_fp16")]; + bool var_4897_interleave_0 = const()[name = string("op_4897_interleave_0"), val = bool(false)]; + tensor var_4897_cast_fp16 = concat(axis = var_3994, interleave = var_4897_interleave_0, values = (var_4843_cast_fp16, var_4845_cast_fp16, var_4847_cast_fp16, var_4849_cast_fp16))[name = string("op_4897_cast_fp16")]; + bool var_4899_interleave_0 = const()[name = string("op_4899_interleave_0"), val = bool(false)]; + tensor var_4899_cast_fp16 = concat(axis = var_3994, interleave = var_4899_interleave_0, values = (var_4851_cast_fp16, var_4853_cast_fp16, var_4855_cast_fp16, var_4857_cast_fp16))[name = string("op_4899_cast_fp16")]; + bool var_4901_interleave_0 = const()[name = string("op_4901_interleave_0"), val = bool(false)]; + tensor var_4901_cast_fp16 = concat(axis = var_3994, interleave = var_4901_interleave_0, values = (var_4859_cast_fp16, var_4861_cast_fp16, var_4863_cast_fp16, var_4865_cast_fp16))[name = string("op_4901_cast_fp16")]; + bool var_4903_interleave_0 = const()[name = string("op_4903_interleave_0"), val = bool(false)]; + tensor var_4903_cast_fp16 = concat(axis = var_3994, interleave = var_4903_interleave_0, values = (var_4867_cast_fp16, var_4869_cast_fp16, var_4871_cast_fp16, var_4873_cast_fp16))[name = string("op_4903_cast_fp16")]; + bool var_4905_interleave_0 = const()[name = string("op_4905_interleave_0"), val = bool(false)]; + tensor var_4905_cast_fp16 = concat(axis = var_3994, interleave = var_4905_interleave_0, values = (var_4875_cast_fp16, var_4877_cast_fp16, var_4879_cast_fp16, var_4881_cast_fp16))[name = string("op_4905_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_4011, interleave = input_33_interleave_0, values = (var_4883_cast_fp16, var_4885_cast_fp16, var_4887_cast_fp16, var_4889_cast_fp16, var_4891_cast_fp16, var_4893_cast_fp16, var_4895_cast_fp16, var_4897_cast_fp16, var_4899_cast_fp16, var_4901_cast_fp16, var_4903_cast_fp16, var_4905_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66464448)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67644160)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_4924_to_fp16 = const()[name = string("op_4924_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_4924_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67645760)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67647360)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67648960)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72367616)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72373824)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77092480)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_4953 = const()[name = string("op_4953"), val = int32(3)]; + int32 var_4970 = const()[name = string("op_4970"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_4987_to_fp16 = const()[name = string("op_4987_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_4987_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77094080)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77095680)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77097280)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78276992)))]; + tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; + string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; + tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; + tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; + int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78278592)))]; + tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; + string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; + tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; + tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; + int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79458304)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80638016)))]; + tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_5025_begin_0 = const()[name = string("op_5025_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5025_end_0 = const()[name = string("op_5025_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5025_end_mask_0 = const()[name = string("op_5025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5025_cast_fp16 = slice_by_index(begin = var_5025_begin_0, end = var_5025_end_0, end_mask = var_5025_end_mask_0, x = query_11_cast_fp16)[name = string("op_5025_cast_fp16")]; + tensor var_5029_begin_0 = const()[name = string("op_5029_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5029_end_0 = const()[name = string("op_5029_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5029_end_mask_0 = const()[name = string("op_5029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5029_cast_fp16 = slice_by_index(begin = var_5029_begin_0, end = var_5029_end_0, end_mask = var_5029_end_mask_0, x = query_11_cast_fp16)[name = string("op_5029_cast_fp16")]; + tensor var_5033_begin_0 = const()[name = string("op_5033_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5033_end_0 = const()[name = string("op_5033_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5033_end_mask_0 = const()[name = string("op_5033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5033_cast_fp16 = slice_by_index(begin = var_5033_begin_0, end = var_5033_end_0, end_mask = var_5033_end_mask_0, x = query_11_cast_fp16)[name = string("op_5033_cast_fp16")]; + tensor var_5037_begin_0 = const()[name = string("op_5037_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5037_end_0 = const()[name = string("op_5037_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5037_end_mask_0 = const()[name = string("op_5037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5037_cast_fp16 = slice_by_index(begin = var_5037_begin_0, end = var_5037_end_0, end_mask = var_5037_end_mask_0, x = query_11_cast_fp16)[name = string("op_5037_cast_fp16")]; + tensor var_5041_begin_0 = const()[name = string("op_5041_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5041_end_0 = const()[name = string("op_5041_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5041_end_mask_0 = const()[name = string("op_5041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5041_cast_fp16 = slice_by_index(begin = var_5041_begin_0, end = var_5041_end_0, end_mask = var_5041_end_mask_0, x = query_11_cast_fp16)[name = string("op_5041_cast_fp16")]; + tensor var_5045_begin_0 = const()[name = string("op_5045_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5045_end_0 = const()[name = string("op_5045_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5045_end_mask_0 = const()[name = string("op_5045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5045_cast_fp16 = slice_by_index(begin = var_5045_begin_0, end = var_5045_end_0, end_mask = var_5045_end_mask_0, x = query_11_cast_fp16)[name = string("op_5045_cast_fp16")]; + tensor var_5049_begin_0 = const()[name = string("op_5049_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5049_end_0 = const()[name = string("op_5049_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5049_end_mask_0 = const()[name = string("op_5049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5049_cast_fp16 = slice_by_index(begin = var_5049_begin_0, end = var_5049_end_0, end_mask = var_5049_end_mask_0, x = query_11_cast_fp16)[name = string("op_5049_cast_fp16")]; + tensor var_5053_begin_0 = const()[name = string("op_5053_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5053_end_0 = const()[name = string("op_5053_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5053_end_mask_0 = const()[name = string("op_5053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5053_cast_fp16 = slice_by_index(begin = var_5053_begin_0, end = var_5053_end_0, end_mask = var_5053_end_mask_0, x = query_11_cast_fp16)[name = string("op_5053_cast_fp16")]; + tensor var_5057_begin_0 = const()[name = string("op_5057_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5057_end_0 = const()[name = string("op_5057_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5057_end_mask_0 = const()[name = string("op_5057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5057_cast_fp16 = slice_by_index(begin = var_5057_begin_0, end = var_5057_end_0, end_mask = var_5057_end_mask_0, x = query_11_cast_fp16)[name = string("op_5057_cast_fp16")]; + tensor var_5061_begin_0 = const()[name = string("op_5061_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5061_end_0 = const()[name = string("op_5061_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5061_end_mask_0 = const()[name = string("op_5061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5061_cast_fp16 = slice_by_index(begin = var_5061_begin_0, end = var_5061_end_0, end_mask = var_5061_end_mask_0, x = query_11_cast_fp16)[name = string("op_5061_cast_fp16")]; + tensor var_5065_begin_0 = const()[name = string("op_5065_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5065_end_0 = const()[name = string("op_5065_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5065_end_mask_0 = const()[name = string("op_5065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5065_cast_fp16 = slice_by_index(begin = var_5065_begin_0, end = var_5065_end_0, end_mask = var_5065_end_mask_0, x = query_11_cast_fp16)[name = string("op_5065_cast_fp16")]; + tensor var_5069_begin_0 = const()[name = string("op_5069_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5069_end_0 = const()[name = string("op_5069_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5069_end_mask_0 = const()[name = string("op_5069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5069_cast_fp16 = slice_by_index(begin = var_5069_begin_0, end = var_5069_end_0, end_mask = var_5069_end_mask_0, x = query_11_cast_fp16)[name = string("op_5069_cast_fp16")]; + tensor var_5078_begin_0 = const()[name = string("op_5078_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5078_end_0 = const()[name = string("op_5078_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5078_end_mask_0 = const()[name = string("op_5078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5078_cast_fp16 = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5078_cast_fp16")]; + tensor var_5085_begin_0 = const()[name = string("op_5085_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5085_end_0 = const()[name = string("op_5085_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5085_end_mask_0 = const()[name = string("op_5085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5085_cast_fp16 = slice_by_index(begin = var_5085_begin_0, end = var_5085_end_0, end_mask = var_5085_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5085_cast_fp16")]; + tensor var_5092_begin_0 = const()[name = string("op_5092_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5092_end_0 = const()[name = string("op_5092_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5092_end_mask_0 = const()[name = string("op_5092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5092_cast_fp16 = slice_by_index(begin = var_5092_begin_0, end = var_5092_end_0, end_mask = var_5092_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5092_cast_fp16")]; + tensor var_5099_begin_0 = const()[name = string("op_5099_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5099_end_0 = const()[name = string("op_5099_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5099_end_mask_0 = const()[name = string("op_5099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5099_cast_fp16 = slice_by_index(begin = var_5099_begin_0, end = var_5099_end_0, end_mask = var_5099_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5099_cast_fp16")]; + tensor var_5106_begin_0 = const()[name = string("op_5106_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5106_end_0 = const()[name = string("op_5106_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5106_end_mask_0 = const()[name = string("op_5106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5106_cast_fp16 = slice_by_index(begin = var_5106_begin_0, end = var_5106_end_0, end_mask = var_5106_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5106_cast_fp16")]; + tensor var_5113_begin_0 = const()[name = string("op_5113_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5113_end_0 = const()[name = string("op_5113_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5113_end_mask_0 = const()[name = string("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5113_cast_fp16 = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5113_cast_fp16")]; + tensor var_5120_begin_0 = const()[name = string("op_5120_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5120_end_0 = const()[name = string("op_5120_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5120_end_mask_0 = const()[name = string("op_5120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5120_cast_fp16 = slice_by_index(begin = var_5120_begin_0, end = var_5120_end_0, end_mask = var_5120_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5120_cast_fp16")]; + tensor var_5127_begin_0 = const()[name = string("op_5127_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5127_end_0 = const()[name = string("op_5127_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5127_end_mask_0 = const()[name = string("op_5127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5127_cast_fp16 = slice_by_index(begin = var_5127_begin_0, end = var_5127_end_0, end_mask = var_5127_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5127_cast_fp16")]; + tensor var_5134_begin_0 = const()[name = string("op_5134_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5134_end_0 = const()[name = string("op_5134_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5134_end_mask_0 = const()[name = string("op_5134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5134_cast_fp16 = slice_by_index(begin = var_5134_begin_0, end = var_5134_end_0, end_mask = var_5134_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5134_cast_fp16")]; + tensor var_5141_begin_0 = const()[name = string("op_5141_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5141_end_0 = const()[name = string("op_5141_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5141_end_mask_0 = const()[name = string("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5141_cast_fp16 = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5141_cast_fp16")]; + tensor var_5148_begin_0 = const()[name = string("op_5148_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5148_end_0 = const()[name = string("op_5148_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5148_end_mask_0 = const()[name = string("op_5148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5148_cast_fp16 = slice_by_index(begin = var_5148_begin_0, end = var_5148_end_0, end_mask = var_5148_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5148_cast_fp16")]; + tensor var_5155_begin_0 = const()[name = string("op_5155_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5155_end_0 = const()[name = string("op_5155_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5155_end_mask_0 = const()[name = string("op_5155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5155_cast_fp16 = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5155_cast_fp16")]; + tensor var_5162_begin_0 = const()[name = string("op_5162_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5162_end_0 = const()[name = string("op_5162_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5162_end_mask_0 = const()[name = string("op_5162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5162_cast_fp16 = slice_by_index(begin = var_5162_begin_0, end = var_5162_end_0, end_mask = var_5162_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5162_cast_fp16")]; + tensor var_5169_begin_0 = const()[name = string("op_5169_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5169_end_0 = const()[name = string("op_5169_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5169_end_mask_0 = const()[name = string("op_5169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5169_cast_fp16 = slice_by_index(begin = var_5169_begin_0, end = var_5169_end_0, end_mask = var_5169_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5169_cast_fp16")]; + tensor var_5176_begin_0 = const()[name = string("op_5176_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5176_end_0 = const()[name = string("op_5176_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5176_end_mask_0 = const()[name = string("op_5176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5176_cast_fp16 = slice_by_index(begin = var_5176_begin_0, end = var_5176_end_0, end_mask = var_5176_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5176_cast_fp16")]; + tensor var_5183_begin_0 = const()[name = string("op_5183_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5183_end_0 = const()[name = string("op_5183_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5183_end_mask_0 = const()[name = string("op_5183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5183_cast_fp16 = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5183_cast_fp16")]; + tensor var_5190_begin_0 = const()[name = string("op_5190_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5190_end_0 = const()[name = string("op_5190_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5190_end_mask_0 = const()[name = string("op_5190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5190_cast_fp16 = slice_by_index(begin = var_5190_begin_0, end = var_5190_end_0, end_mask = var_5190_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5190_cast_fp16")]; + tensor var_5197_begin_0 = const()[name = string("op_5197_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5197_end_0 = const()[name = string("op_5197_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5197_end_mask_0 = const()[name = string("op_5197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5197_cast_fp16 = slice_by_index(begin = var_5197_begin_0, end = var_5197_end_0, end_mask = var_5197_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5197_cast_fp16")]; + tensor var_5204_begin_0 = const()[name = string("op_5204_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5204_end_0 = const()[name = string("op_5204_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5204_end_mask_0 = const()[name = string("op_5204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5204_cast_fp16 = slice_by_index(begin = var_5204_begin_0, end = var_5204_end_0, end_mask = var_5204_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5204_cast_fp16")]; + tensor var_5211_begin_0 = const()[name = string("op_5211_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5211_end_0 = const()[name = string("op_5211_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5211_end_mask_0 = const()[name = string("op_5211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5211_cast_fp16 = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5211_cast_fp16")]; + tensor var_5218_begin_0 = const()[name = string("op_5218_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5218_end_0 = const()[name = string("op_5218_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5218_end_mask_0 = const()[name = string("op_5218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5218_cast_fp16 = slice_by_index(begin = var_5218_begin_0, end = var_5218_end_0, end_mask = var_5218_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5218_cast_fp16")]; + tensor var_5225_begin_0 = const()[name = string("op_5225_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5225_end_0 = const()[name = string("op_5225_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5225_end_mask_0 = const()[name = string("op_5225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5225_cast_fp16 = slice_by_index(begin = var_5225_begin_0, end = var_5225_end_0, end_mask = var_5225_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5225_cast_fp16")]; + tensor var_5232_begin_0 = const()[name = string("op_5232_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5232_end_0 = const()[name = string("op_5232_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5232_end_mask_0 = const()[name = string("op_5232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5232_cast_fp16 = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5232_cast_fp16")]; + tensor var_5239_begin_0 = const()[name = string("op_5239_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5239_end_0 = const()[name = string("op_5239_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5239_end_mask_0 = const()[name = string("op_5239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5239_cast_fp16")]; + tensor var_5246_begin_0 = const()[name = string("op_5246_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5246_end_0 = const()[name = string("op_5246_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5246_end_mask_0 = const()[name = string("op_5246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5246_cast_fp16 = slice_by_index(begin = var_5246_begin_0, end = var_5246_end_0, end_mask = var_5246_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5246_cast_fp16")]; + tensor var_5253_begin_0 = const()[name = string("op_5253_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5253_end_0 = const()[name = string("op_5253_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5253_end_mask_0 = const()[name = string("op_5253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5253_cast_fp16 = slice_by_index(begin = var_5253_begin_0, end = var_5253_end_0, end_mask = var_5253_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5253_cast_fp16")]; + tensor var_5260_begin_0 = const()[name = string("op_5260_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5260_end_0 = const()[name = string("op_5260_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5260_end_mask_0 = const()[name = string("op_5260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5260_cast_fp16 = slice_by_index(begin = var_5260_begin_0, end = var_5260_end_0, end_mask = var_5260_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5260_cast_fp16")]; + tensor var_5267_begin_0 = const()[name = string("op_5267_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5267_end_0 = const()[name = string("op_5267_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5267_end_mask_0 = const()[name = string("op_5267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5267_cast_fp16 = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5267_cast_fp16")]; + tensor var_5274_begin_0 = const()[name = string("op_5274_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5274_end_0 = const()[name = string("op_5274_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5274_end_mask_0 = const()[name = string("op_5274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5274_cast_fp16 = slice_by_index(begin = var_5274_begin_0, end = var_5274_end_0, end_mask = var_5274_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5274_cast_fp16")]; + tensor var_5281_begin_0 = const()[name = string("op_5281_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5281_end_0 = const()[name = string("op_5281_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5281_end_mask_0 = const()[name = string("op_5281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5281_cast_fp16")]; + tensor var_5288_begin_0 = const()[name = string("op_5288_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5288_end_0 = const()[name = string("op_5288_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5288_end_mask_0 = const()[name = string("op_5288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5288_cast_fp16 = slice_by_index(begin = var_5288_begin_0, end = var_5288_end_0, end_mask = var_5288_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5288_cast_fp16")]; + tensor var_5295_begin_0 = const()[name = string("op_5295_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5295_end_0 = const()[name = string("op_5295_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5295_end_mask_0 = const()[name = string("op_5295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5295_cast_fp16 = slice_by_index(begin = var_5295_begin_0, end = var_5295_end_0, end_mask = var_5295_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5295_cast_fp16")]; + tensor var_5302_begin_0 = const()[name = string("op_5302_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5302_end_0 = const()[name = string("op_5302_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5302_end_mask_0 = const()[name = string("op_5302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5302_cast_fp16 = slice_by_index(begin = var_5302_begin_0, end = var_5302_end_0, end_mask = var_5302_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5302_cast_fp16")]; + tensor var_5309_begin_0 = const()[name = string("op_5309_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5309_end_0 = const()[name = string("op_5309_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5309_end_mask_0 = const()[name = string("op_5309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5309_cast_fp16 = slice_by_index(begin = var_5309_begin_0, end = var_5309_end_0, end_mask = var_5309_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5309_cast_fp16")]; + tensor var_5316_begin_0 = const()[name = string("op_5316_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5316_end_0 = const()[name = string("op_5316_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5316_end_mask_0 = const()[name = string("op_5316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5316_cast_fp16 = slice_by_index(begin = var_5316_begin_0, end = var_5316_end_0, end_mask = var_5316_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5316_cast_fp16")]; + tensor var_5323_begin_0 = const()[name = string("op_5323_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5323_end_0 = const()[name = string("op_5323_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5323_end_mask_0 = const()[name = string("op_5323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5323_cast_fp16 = slice_by_index(begin = var_5323_begin_0, end = var_5323_end_0, end_mask = var_5323_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5323_cast_fp16")]; + tensor var_5330_begin_0 = const()[name = string("op_5330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5330_end_0 = const()[name = string("op_5330_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5330_end_mask_0 = const()[name = string("op_5330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5330_cast_fp16 = slice_by_index(begin = var_5330_begin_0, end = var_5330_end_0, end_mask = var_5330_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5330_cast_fp16")]; + tensor var_5337_begin_0 = const()[name = string("op_5337_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5337_end_0 = const()[name = string("op_5337_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5337_end_mask_0 = const()[name = string("op_5337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5337_cast_fp16 = slice_by_index(begin = var_5337_begin_0, end = var_5337_end_0, end_mask = var_5337_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5337_cast_fp16")]; + tensor var_5344_begin_0 = const()[name = string("op_5344_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5344_end_0 = const()[name = string("op_5344_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5344_end_mask_0 = const()[name = string("op_5344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5344_cast_fp16 = slice_by_index(begin = var_5344_begin_0, end = var_5344_end_0, end_mask = var_5344_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5344_cast_fp16")]; + tensor var_5351_begin_0 = const()[name = string("op_5351_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5351_end_0 = const()[name = string("op_5351_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5351_end_mask_0 = const()[name = string("op_5351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5351_cast_fp16 = slice_by_index(begin = var_5351_begin_0, end = var_5351_end_0, end_mask = var_5351_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5351_cast_fp16")]; + tensor var_5358_begin_0 = const()[name = string("op_5358_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5358_end_0 = const()[name = string("op_5358_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5358_end_mask_0 = const()[name = string("op_5358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5358_cast_fp16 = slice_by_index(begin = var_5358_begin_0, end = var_5358_end_0, end_mask = var_5358_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5358_cast_fp16")]; + tensor var_5365_begin_0 = const()[name = string("op_5365_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5365_end_0 = const()[name = string("op_5365_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5365_end_mask_0 = const()[name = string("op_5365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5365_cast_fp16 = slice_by_index(begin = var_5365_begin_0, end = var_5365_end_0, end_mask = var_5365_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5365_cast_fp16")]; + tensor var_5372_begin_0 = const()[name = string("op_5372_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5372_end_0 = const()[name = string("op_5372_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5372_end_mask_0 = const()[name = string("op_5372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5372_cast_fp16 = slice_by_index(begin = var_5372_begin_0, end = var_5372_end_0, end_mask = var_5372_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5372_cast_fp16")]; + tensor var_5379_begin_0 = const()[name = string("op_5379_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5379_end_0 = const()[name = string("op_5379_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5379_end_mask_0 = const()[name = string("op_5379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5379_cast_fp16 = slice_by_index(begin = var_5379_begin_0, end = var_5379_end_0, end_mask = var_5379_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5379_cast_fp16")]; + tensor var_5386_begin_0 = const()[name = string("op_5386_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5386_end_0 = const()[name = string("op_5386_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5386_end_mask_0 = const()[name = string("op_5386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5386_cast_fp16 = slice_by_index(begin = var_5386_begin_0, end = var_5386_end_0, end_mask = var_5386_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5386_cast_fp16")]; + tensor var_5393_begin_0 = const()[name = string("op_5393_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5393_end_0 = const()[name = string("op_5393_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5393_end_mask_0 = const()[name = string("op_5393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5393_cast_fp16 = slice_by_index(begin = var_5393_begin_0, end = var_5393_end_0, end_mask = var_5393_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5393_cast_fp16")]; + tensor var_5400_begin_0 = const()[name = string("op_5400_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5400_end_0 = const()[name = string("op_5400_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5400_end_mask_0 = const()[name = string("op_5400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5400_cast_fp16 = slice_by_index(begin = var_5400_begin_0, end = var_5400_end_0, end_mask = var_5400_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5400_cast_fp16")]; + tensor var_5407_begin_0 = const()[name = string("op_5407_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5407_end_0 = const()[name = string("op_5407_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5407_end_mask_0 = const()[name = string("op_5407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5407_cast_fp16 = slice_by_index(begin = var_5407_begin_0, end = var_5407_end_0, end_mask = var_5407_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5407_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5412_begin_0 = const()[name = string("op_5412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5412_end_0 = const()[name = string("op_5412_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_5412_end_mask_0 = const()[name = string("op_5412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = string("transpose_6")]; + tensor var_5412_cast_fp16 = slice_by_index(begin = var_5412_begin_0, end = var_5412_end_0, end_mask = var_5412_end_mask_0, x = k_11_cast_fp16)[name = string("op_5412_cast_fp16")]; + tensor var_5416_begin_0 = const()[name = string("op_5416_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5416_end_0 = const()[name = string("op_5416_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_5416_end_mask_0 = const()[name = string("op_5416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5416_cast_fp16 = slice_by_index(begin = var_5416_begin_0, end = var_5416_end_0, end_mask = var_5416_end_mask_0, x = k_11_cast_fp16)[name = string("op_5416_cast_fp16")]; + tensor var_5420_begin_0 = const()[name = string("op_5420_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5420_end_0 = const()[name = string("op_5420_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_5420_end_mask_0 = const()[name = string("op_5420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5420_cast_fp16 = slice_by_index(begin = var_5420_begin_0, end = var_5420_end_0, end_mask = var_5420_end_mask_0, x = k_11_cast_fp16)[name = string("op_5420_cast_fp16")]; + tensor var_5424_begin_0 = const()[name = string("op_5424_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5424_end_0 = const()[name = string("op_5424_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_5424_end_mask_0 = const()[name = string("op_5424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5424_cast_fp16 = slice_by_index(begin = var_5424_begin_0, end = var_5424_end_0, end_mask = var_5424_end_mask_0, x = k_11_cast_fp16)[name = string("op_5424_cast_fp16")]; + tensor var_5428_begin_0 = const()[name = string("op_5428_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5428_end_0 = const()[name = string("op_5428_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_5428_end_mask_0 = const()[name = string("op_5428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5428_cast_fp16 = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = k_11_cast_fp16)[name = string("op_5428_cast_fp16")]; + tensor var_5432_begin_0 = const()[name = string("op_5432_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5432_end_0 = const()[name = string("op_5432_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_5432_end_mask_0 = const()[name = string("op_5432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5432_cast_fp16 = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = k_11_cast_fp16)[name = string("op_5432_cast_fp16")]; + tensor var_5436_begin_0 = const()[name = string("op_5436_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5436_end_0 = const()[name = string("op_5436_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_5436_end_mask_0 = const()[name = string("op_5436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5436_cast_fp16 = slice_by_index(begin = var_5436_begin_0, end = var_5436_end_0, end_mask = var_5436_end_mask_0, x = k_11_cast_fp16)[name = string("op_5436_cast_fp16")]; + tensor var_5440_begin_0 = const()[name = string("op_5440_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5440_end_0 = const()[name = string("op_5440_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_5440_end_mask_0 = const()[name = string("op_5440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5440_cast_fp16 = slice_by_index(begin = var_5440_begin_0, end = var_5440_end_0, end_mask = var_5440_end_mask_0, x = k_11_cast_fp16)[name = string("op_5440_cast_fp16")]; + tensor var_5444_begin_0 = const()[name = string("op_5444_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5444_end_0 = const()[name = string("op_5444_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_5444_end_mask_0 = const()[name = string("op_5444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5444_cast_fp16 = slice_by_index(begin = var_5444_begin_0, end = var_5444_end_0, end_mask = var_5444_end_mask_0, x = k_11_cast_fp16)[name = string("op_5444_cast_fp16")]; + tensor var_5448_begin_0 = const()[name = string("op_5448_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5448_end_0 = const()[name = string("op_5448_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_5448_end_mask_0 = const()[name = string("op_5448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5448_cast_fp16 = slice_by_index(begin = var_5448_begin_0, end = var_5448_end_0, end_mask = var_5448_end_mask_0, x = k_11_cast_fp16)[name = string("op_5448_cast_fp16")]; + tensor var_5452_begin_0 = const()[name = string("op_5452_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5452_end_0 = const()[name = string("op_5452_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_5452_end_mask_0 = const()[name = string("op_5452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5452_cast_fp16 = slice_by_index(begin = var_5452_begin_0, end = var_5452_end_0, end_mask = var_5452_end_mask_0, x = k_11_cast_fp16)[name = string("op_5452_cast_fp16")]; + tensor var_5456_begin_0 = const()[name = string("op_5456_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5456_end_0 = const()[name = string("op_5456_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_5456_end_mask_0 = const()[name = string("op_5456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5456_cast_fp16 = slice_by_index(begin = var_5456_begin_0, end = var_5456_end_0, end_mask = var_5456_end_mask_0, x = k_11_cast_fp16)[name = string("op_5456_cast_fp16")]; + tensor var_5458_begin_0 = const()[name = string("op_5458_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5458_end_0 = const()[name = string("op_5458_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5458_end_mask_0 = const()[name = string("op_5458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5458_cast_fp16 = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = value_11_cast_fp16)[name = string("op_5458_cast_fp16")]; + tensor var_5462_begin_0 = const()[name = string("op_5462_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5462_end_0 = const()[name = string("op_5462_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5462_end_mask_0 = const()[name = string("op_5462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5462_cast_fp16 = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, x = value_11_cast_fp16)[name = string("op_5462_cast_fp16")]; + tensor var_5466_begin_0 = const()[name = string("op_5466_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5466_end_0 = const()[name = string("op_5466_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5466_end_mask_0 = const()[name = string("op_5466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5466_cast_fp16 = slice_by_index(begin = var_5466_begin_0, end = var_5466_end_0, end_mask = var_5466_end_mask_0, x = value_11_cast_fp16)[name = string("op_5466_cast_fp16")]; + tensor var_5470_begin_0 = const()[name = string("op_5470_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5470_end_0 = const()[name = string("op_5470_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5470_end_mask_0 = const()[name = string("op_5470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5470_cast_fp16 = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = value_11_cast_fp16)[name = string("op_5470_cast_fp16")]; + tensor var_5474_begin_0 = const()[name = string("op_5474_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5474_end_0 = const()[name = string("op_5474_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5474_end_mask_0 = const()[name = string("op_5474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5474_cast_fp16 = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = value_11_cast_fp16)[name = string("op_5474_cast_fp16")]; + tensor var_5478_begin_0 = const()[name = string("op_5478_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5478_end_0 = const()[name = string("op_5478_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5478_end_mask_0 = const()[name = string("op_5478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5478_cast_fp16 = slice_by_index(begin = var_5478_begin_0, end = var_5478_end_0, end_mask = var_5478_end_mask_0, x = value_11_cast_fp16)[name = string("op_5478_cast_fp16")]; + tensor var_5482_begin_0 = const()[name = string("op_5482_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5482_end_0 = const()[name = string("op_5482_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5482_end_mask_0 = const()[name = string("op_5482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5482_cast_fp16 = slice_by_index(begin = var_5482_begin_0, end = var_5482_end_0, end_mask = var_5482_end_mask_0, x = value_11_cast_fp16)[name = string("op_5482_cast_fp16")]; + tensor var_5486_begin_0 = const()[name = string("op_5486_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5486_end_0 = const()[name = string("op_5486_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5486_end_mask_0 = const()[name = string("op_5486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5486_cast_fp16 = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = value_11_cast_fp16)[name = string("op_5486_cast_fp16")]; + tensor var_5490_begin_0 = const()[name = string("op_5490_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5490_end_0 = const()[name = string("op_5490_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5490_end_mask_0 = const()[name = string("op_5490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5490_cast_fp16 = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = value_11_cast_fp16)[name = string("op_5490_cast_fp16")]; + tensor var_5494_begin_0 = const()[name = string("op_5494_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5494_end_0 = const()[name = string("op_5494_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5494_end_mask_0 = const()[name = string("op_5494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5494_cast_fp16 = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = value_11_cast_fp16)[name = string("op_5494_cast_fp16")]; + tensor var_5498_begin_0 = const()[name = string("op_5498_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5498_end_0 = const()[name = string("op_5498_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5498_end_mask_0 = const()[name = string("op_5498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = value_11_cast_fp16)[name = string("op_5498_cast_fp16")]; + tensor var_5502_begin_0 = const()[name = string("op_5502_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5502_end_0 = const()[name = string("op_5502_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5502_end_mask_0 = const()[name = string("op_5502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5502_cast_fp16 = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = value_11_cast_fp16)[name = string("op_5502_cast_fp16")]; + string _SplitHeadsQ__mh_w_481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_481_equation_0, values = (var_5412_cast_fp16, var_5078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_481_cast_fp16")]; + string _SplitHeadsQ__mh_w_483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_483_equation_0, values = (var_5412_cast_fp16, var_5085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_483_cast_fp16")]; + string _SplitHeadsQ__mh_w_485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_485_equation_0, values = (var_5412_cast_fp16, var_5092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_485_cast_fp16")]; + string _SplitHeadsQ__mh_w_487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_487_equation_0, values = (var_5412_cast_fp16, var_5099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_487_cast_fp16")]; + string _SplitHeadsQ__mh_w_489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_489_equation_0, values = (var_5416_cast_fp16, var_5106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_489_cast_fp16")]; + string _SplitHeadsQ__mh_w_491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_491_equation_0, values = (var_5416_cast_fp16, var_5113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_491_cast_fp16")]; + string _SplitHeadsQ__mh_w_493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_493_equation_0, values = (var_5416_cast_fp16, var_5120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_493_cast_fp16")]; + string _SplitHeadsQ__mh_w_495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_495_equation_0, values = (var_5416_cast_fp16, var_5127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_495_cast_fp16")]; + string _SplitHeadsQ__mh_w_497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_497_equation_0, values = (var_5420_cast_fp16, var_5134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_497_cast_fp16")]; + string _SplitHeadsQ__mh_w_499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_499_equation_0, values = (var_5420_cast_fp16, var_5141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_499_cast_fp16")]; + string _SplitHeadsQ__mh_w_501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_501_equation_0, values = (var_5420_cast_fp16, var_5148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_501_cast_fp16")]; + string _SplitHeadsQ__mh_w_503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_503_equation_0, values = (var_5420_cast_fp16, var_5155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_503_cast_fp16")]; + string _SplitHeadsQ__mh_w_505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_505_equation_0, values = (var_5424_cast_fp16, var_5162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_505_cast_fp16")]; + string _SplitHeadsQ__mh_w_507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_507_equation_0, values = (var_5424_cast_fp16, var_5169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_507_cast_fp16")]; + string _SplitHeadsQ__mh_w_509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_509_equation_0, values = (var_5424_cast_fp16, var_5176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_509_cast_fp16")]; + string _SplitHeadsQ__mh_w_511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_511_equation_0, values = (var_5424_cast_fp16, var_5183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_511_cast_fp16")]; + string _SplitHeadsQ__mh_w_513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_513_equation_0, values = (var_5428_cast_fp16, var_5190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_513_cast_fp16")]; + string _SplitHeadsQ__mh_w_515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_515_equation_0, values = (var_5428_cast_fp16, var_5197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_515_cast_fp16")]; + string _SplitHeadsQ__mh_w_517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_517_equation_0, values = (var_5428_cast_fp16, var_5204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_517_cast_fp16")]; + string _SplitHeadsQ__mh_w_519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_519_equation_0, values = (var_5428_cast_fp16, var_5211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_521_equation_0, values = (var_5432_cast_fp16, var_5218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_521_cast_fp16")]; + string _SplitHeadsQ__mh_w_523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_523_equation_0, values = (var_5432_cast_fp16, var_5225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_523_cast_fp16")]; + string _SplitHeadsQ__mh_w_525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_525_equation_0, values = (var_5432_cast_fp16, var_5232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_525_cast_fp16")]; + string _SplitHeadsQ__mh_w_527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_527_equation_0, values = (var_5432_cast_fp16, var_5239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_527_cast_fp16")]; + string _SplitHeadsQ__mh_w_529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_529_equation_0, values = (var_5436_cast_fp16, var_5246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_529_cast_fp16")]; + string _SplitHeadsQ__mh_w_531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_531_equation_0, values = (var_5436_cast_fp16, var_5253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_531_cast_fp16")]; + string _SplitHeadsQ__mh_w_533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_533_equation_0, values = (var_5436_cast_fp16, var_5260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_533_cast_fp16")]; + string _SplitHeadsQ__mh_w_535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_535_equation_0, values = (var_5436_cast_fp16, var_5267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_535_cast_fp16")]; + string _SplitHeadsQ__mh_w_537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_537_equation_0, values = (var_5440_cast_fp16, var_5274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_537_cast_fp16")]; + string _SplitHeadsQ__mh_w_539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_539_equation_0, values = (var_5440_cast_fp16, var_5281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_539_cast_fp16")]; + string _SplitHeadsQ__mh_w_541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_541_equation_0, values = (var_5440_cast_fp16, var_5288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_541_cast_fp16")]; + string _SplitHeadsQ__mh_w_543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_543_equation_0, values = (var_5440_cast_fp16, var_5295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_543_cast_fp16")]; + string _SplitHeadsQ__mh_w_545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_545_equation_0, values = (var_5444_cast_fp16, var_5302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_545_cast_fp16")]; + string _SplitHeadsQ__mh_w_547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_547_equation_0, values = (var_5444_cast_fp16, var_5309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_547_cast_fp16")]; + string _SplitHeadsQ__mh_w_549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_549_equation_0, values = (var_5444_cast_fp16, var_5316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_549_cast_fp16")]; + string _SplitHeadsQ__mh_w_551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_551_equation_0, values = (var_5444_cast_fp16, var_5323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_551_cast_fp16")]; + string _SplitHeadsQ__mh_w_553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_553_equation_0, values = (var_5448_cast_fp16, var_5330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_553_cast_fp16")]; + string _SplitHeadsQ__mh_w_555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_555_equation_0, values = (var_5448_cast_fp16, var_5337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_555_cast_fp16")]; + string _SplitHeadsQ__mh_w_557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_557_equation_0, values = (var_5448_cast_fp16, var_5344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_557_cast_fp16")]; + string _SplitHeadsQ__mh_w_559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_559_equation_0, values = (var_5448_cast_fp16, var_5351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_559_cast_fp16")]; + string _SplitHeadsQ__mh_w_561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_561_equation_0, values = (var_5452_cast_fp16, var_5358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_561_cast_fp16")]; + string _SplitHeadsQ__mh_w_563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_563_equation_0, values = (var_5452_cast_fp16, var_5365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_563_cast_fp16")]; + string _SplitHeadsQ__mh_w_565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_565_equation_0, values = (var_5452_cast_fp16, var_5372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_565_cast_fp16")]; + string _SplitHeadsQ__mh_w_567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_567_equation_0, values = (var_5452_cast_fp16, var_5379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_567_cast_fp16")]; + string _SplitHeadsQ__mh_w_569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_569_equation_0, values = (var_5456_cast_fp16, var_5386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_569_cast_fp16")]; + string _SplitHeadsQ__mh_w_571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_571_equation_0, values = (var_5456_cast_fp16, var_5393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_571_cast_fp16")]; + string _SplitHeadsQ__mh_w_573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_573_equation_0, values = (var_5456_cast_fp16, var_5400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_573_cast_fp16")]; + string _SplitHeadsQ__mh_w_575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_575_equation_0, values = (var_5456_cast_fp16, var_5407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_575_cast_fp16")]; + fp16 var_5601_to_fp16 = const()[name = string("op_5601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_481_cast_fp16, y = var_5601_to_fp16)[name = string("aw_chunk_481_cast_fp16")]; + fp16 var_5603_to_fp16 = const()[name = string("op_5603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_483_cast_fp16, y = var_5603_to_fp16)[name = string("aw_chunk_483_cast_fp16")]; + fp16 var_5605_to_fp16 = const()[name = string("op_5605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_485_cast_fp16, y = var_5605_to_fp16)[name = string("aw_chunk_485_cast_fp16")]; + fp16 var_5607_to_fp16 = const()[name = string("op_5607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_487_cast_fp16, y = var_5607_to_fp16)[name = string("aw_chunk_487_cast_fp16")]; + fp16 var_5609_to_fp16 = const()[name = string("op_5609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_489_cast_fp16, y = var_5609_to_fp16)[name = string("aw_chunk_489_cast_fp16")]; + fp16 var_5611_to_fp16 = const()[name = string("op_5611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_491_cast_fp16, y = var_5611_to_fp16)[name = string("aw_chunk_491_cast_fp16")]; + fp16 var_5613_to_fp16 = const()[name = string("op_5613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_493_cast_fp16, y = var_5613_to_fp16)[name = string("aw_chunk_493_cast_fp16")]; + fp16 var_5615_to_fp16 = const()[name = string("op_5615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_495_cast_fp16, y = var_5615_to_fp16)[name = string("aw_chunk_495_cast_fp16")]; + fp16 var_5617_to_fp16 = const()[name = string("op_5617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_497_cast_fp16, y = var_5617_to_fp16)[name = string("aw_chunk_497_cast_fp16")]; + fp16 var_5619_to_fp16 = const()[name = string("op_5619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_499_cast_fp16, y = var_5619_to_fp16)[name = string("aw_chunk_499_cast_fp16")]; + fp16 var_5621_to_fp16 = const()[name = string("op_5621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_501_cast_fp16, y = var_5621_to_fp16)[name = string("aw_chunk_501_cast_fp16")]; + fp16 var_5623_to_fp16 = const()[name = string("op_5623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_503_cast_fp16, y = var_5623_to_fp16)[name = string("aw_chunk_503_cast_fp16")]; + fp16 var_5625_to_fp16 = const()[name = string("op_5625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_505_cast_fp16, y = var_5625_to_fp16)[name = string("aw_chunk_505_cast_fp16")]; + fp16 var_5627_to_fp16 = const()[name = string("op_5627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_507_cast_fp16, y = var_5627_to_fp16)[name = string("aw_chunk_507_cast_fp16")]; + fp16 var_5629_to_fp16 = const()[name = string("op_5629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_509_cast_fp16, y = var_5629_to_fp16)[name = string("aw_chunk_509_cast_fp16")]; + fp16 var_5631_to_fp16 = const()[name = string("op_5631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_511_cast_fp16, y = var_5631_to_fp16)[name = string("aw_chunk_511_cast_fp16")]; + fp16 var_5633_to_fp16 = const()[name = string("op_5633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_513_cast_fp16, y = var_5633_to_fp16)[name = string("aw_chunk_513_cast_fp16")]; + fp16 var_5635_to_fp16 = const()[name = string("op_5635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_515_cast_fp16, y = var_5635_to_fp16)[name = string("aw_chunk_515_cast_fp16")]; + fp16 var_5637_to_fp16 = const()[name = string("op_5637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_517_cast_fp16, y = var_5637_to_fp16)[name = string("aw_chunk_517_cast_fp16")]; + fp16 var_5639_to_fp16 = const()[name = string("op_5639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_519_cast_fp16, y = var_5639_to_fp16)[name = string("aw_chunk_519_cast_fp16")]; + fp16 var_5641_to_fp16 = const()[name = string("op_5641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_521_cast_fp16, y = var_5641_to_fp16)[name = string("aw_chunk_521_cast_fp16")]; + fp16 var_5643_to_fp16 = const()[name = string("op_5643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_523_cast_fp16, y = var_5643_to_fp16)[name = string("aw_chunk_523_cast_fp16")]; + fp16 var_5645_to_fp16 = const()[name = string("op_5645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_525_cast_fp16, y = var_5645_to_fp16)[name = string("aw_chunk_525_cast_fp16")]; + fp16 var_5647_to_fp16 = const()[name = string("op_5647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_527_cast_fp16, y = var_5647_to_fp16)[name = string("aw_chunk_527_cast_fp16")]; + fp16 var_5649_to_fp16 = const()[name = string("op_5649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_529_cast_fp16, y = var_5649_to_fp16)[name = string("aw_chunk_529_cast_fp16")]; + fp16 var_5651_to_fp16 = const()[name = string("op_5651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_531_cast_fp16, y = var_5651_to_fp16)[name = string("aw_chunk_531_cast_fp16")]; + fp16 var_5653_to_fp16 = const()[name = string("op_5653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_533_cast_fp16, y = var_5653_to_fp16)[name = string("aw_chunk_533_cast_fp16")]; + fp16 var_5655_to_fp16 = const()[name = string("op_5655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_535_cast_fp16, y = var_5655_to_fp16)[name = string("aw_chunk_535_cast_fp16")]; + fp16 var_5657_to_fp16 = const()[name = string("op_5657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_537_cast_fp16, y = var_5657_to_fp16)[name = string("aw_chunk_537_cast_fp16")]; + fp16 var_5659_to_fp16 = const()[name = string("op_5659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_539_cast_fp16, y = var_5659_to_fp16)[name = string("aw_chunk_539_cast_fp16")]; + fp16 var_5661_to_fp16 = const()[name = string("op_5661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_541_cast_fp16, y = var_5661_to_fp16)[name = string("aw_chunk_541_cast_fp16")]; + fp16 var_5663_to_fp16 = const()[name = string("op_5663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_543_cast_fp16, y = var_5663_to_fp16)[name = string("aw_chunk_543_cast_fp16")]; + fp16 var_5665_to_fp16 = const()[name = string("op_5665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_545_cast_fp16, y = var_5665_to_fp16)[name = string("aw_chunk_545_cast_fp16")]; + fp16 var_5667_to_fp16 = const()[name = string("op_5667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_547_cast_fp16, y = var_5667_to_fp16)[name = string("aw_chunk_547_cast_fp16")]; + fp16 var_5669_to_fp16 = const()[name = string("op_5669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_549_cast_fp16, y = var_5669_to_fp16)[name = string("aw_chunk_549_cast_fp16")]; + fp16 var_5671_to_fp16 = const()[name = string("op_5671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_551_cast_fp16, y = var_5671_to_fp16)[name = string("aw_chunk_551_cast_fp16")]; + fp16 var_5673_to_fp16 = const()[name = string("op_5673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_553_cast_fp16, y = var_5673_to_fp16)[name = string("aw_chunk_553_cast_fp16")]; + fp16 var_5675_to_fp16 = const()[name = string("op_5675_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_555_cast_fp16, y = var_5675_to_fp16)[name = string("aw_chunk_555_cast_fp16")]; + fp16 var_5677_to_fp16 = const()[name = string("op_5677_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_557_cast_fp16, y = var_5677_to_fp16)[name = string("aw_chunk_557_cast_fp16")]; + fp16 var_5679_to_fp16 = const()[name = string("op_5679_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_559_cast_fp16, y = var_5679_to_fp16)[name = string("aw_chunk_559_cast_fp16")]; + fp16 var_5681_to_fp16 = const()[name = string("op_5681_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_561_cast_fp16, y = var_5681_to_fp16)[name = string("aw_chunk_561_cast_fp16")]; + fp16 var_5683_to_fp16 = const()[name = string("op_5683_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_563_cast_fp16, y = var_5683_to_fp16)[name = string("aw_chunk_563_cast_fp16")]; + fp16 var_5685_to_fp16 = const()[name = string("op_5685_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_565_cast_fp16, y = var_5685_to_fp16)[name = string("aw_chunk_565_cast_fp16")]; + fp16 var_5687_to_fp16 = const()[name = string("op_5687_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_567_cast_fp16, y = var_5687_to_fp16)[name = string("aw_chunk_567_cast_fp16")]; + fp16 var_5689_to_fp16 = const()[name = string("op_5689_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_569_cast_fp16, y = var_5689_to_fp16)[name = string("aw_chunk_569_cast_fp16")]; + fp16 var_5691_to_fp16 = const()[name = string("op_5691_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_571_cast_fp16, y = var_5691_to_fp16)[name = string("aw_chunk_571_cast_fp16")]; + fp16 var_5693_to_fp16 = const()[name = string("op_5693_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_573_cast_fp16, y = var_5693_to_fp16)[name = string("aw_chunk_573_cast_fp16")]; + fp16 var_5695_to_fp16 = const()[name = string("op_5695_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_575_cast_fp16, y = var_5695_to_fp16)[name = string("aw_chunk_575_cast_fp16")]; + tensor var_5697_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_481_cast_fp16)[name = string("op_5697_cast_fp16")]; + tensor var_5698_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_483_cast_fp16)[name = string("op_5698_cast_fp16")]; + tensor var_5699_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_485_cast_fp16)[name = string("op_5699_cast_fp16")]; + tensor var_5700_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_487_cast_fp16)[name = string("op_5700_cast_fp16")]; + tensor var_5701_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_489_cast_fp16)[name = string("op_5701_cast_fp16")]; + tensor var_5702_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_491_cast_fp16)[name = string("op_5702_cast_fp16")]; + tensor var_5703_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_493_cast_fp16)[name = string("op_5703_cast_fp16")]; + tensor var_5704_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_495_cast_fp16)[name = string("op_5704_cast_fp16")]; + tensor var_5705_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_497_cast_fp16)[name = string("op_5705_cast_fp16")]; + tensor var_5706_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_499_cast_fp16)[name = string("op_5706_cast_fp16")]; + tensor var_5707_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_501_cast_fp16)[name = string("op_5707_cast_fp16")]; + tensor var_5708_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_503_cast_fp16)[name = string("op_5708_cast_fp16")]; + tensor var_5709_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_505_cast_fp16)[name = string("op_5709_cast_fp16")]; + tensor var_5710_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_507_cast_fp16)[name = string("op_5710_cast_fp16")]; + tensor var_5711_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_509_cast_fp16)[name = string("op_5711_cast_fp16")]; + tensor var_5712_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_511_cast_fp16)[name = string("op_5712_cast_fp16")]; + tensor var_5713_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_513_cast_fp16)[name = string("op_5713_cast_fp16")]; + tensor var_5714_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_515_cast_fp16)[name = string("op_5714_cast_fp16")]; + tensor var_5715_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_517_cast_fp16)[name = string("op_5715_cast_fp16")]; + tensor var_5716_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_519_cast_fp16)[name = string("op_5716_cast_fp16")]; + tensor var_5717_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_521_cast_fp16)[name = string("op_5717_cast_fp16")]; + tensor var_5718_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_523_cast_fp16)[name = string("op_5718_cast_fp16")]; + tensor var_5719_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_525_cast_fp16)[name = string("op_5719_cast_fp16")]; + tensor var_5720_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_527_cast_fp16)[name = string("op_5720_cast_fp16")]; + tensor var_5721_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_529_cast_fp16)[name = string("op_5721_cast_fp16")]; + tensor var_5722_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_531_cast_fp16)[name = string("op_5722_cast_fp16")]; + tensor var_5723_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_533_cast_fp16)[name = string("op_5723_cast_fp16")]; + tensor var_5724_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_535_cast_fp16)[name = string("op_5724_cast_fp16")]; + tensor var_5725_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_537_cast_fp16)[name = string("op_5725_cast_fp16")]; + tensor var_5726_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_539_cast_fp16)[name = string("op_5726_cast_fp16")]; + tensor var_5727_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_541_cast_fp16)[name = string("op_5727_cast_fp16")]; + tensor var_5728_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_543_cast_fp16)[name = string("op_5728_cast_fp16")]; + tensor var_5729_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_545_cast_fp16)[name = string("op_5729_cast_fp16")]; + tensor var_5730_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_547_cast_fp16)[name = string("op_5730_cast_fp16")]; + tensor var_5731_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_549_cast_fp16)[name = string("op_5731_cast_fp16")]; + tensor var_5732_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_551_cast_fp16)[name = string("op_5732_cast_fp16")]; + tensor var_5733_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_553_cast_fp16)[name = string("op_5733_cast_fp16")]; + tensor var_5734_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_555_cast_fp16)[name = string("op_5734_cast_fp16")]; + tensor var_5735_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_557_cast_fp16)[name = string("op_5735_cast_fp16")]; + tensor var_5736_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_559_cast_fp16)[name = string("op_5736_cast_fp16")]; + tensor var_5737_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_561_cast_fp16)[name = string("op_5737_cast_fp16")]; + tensor var_5738_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_563_cast_fp16)[name = string("op_5738_cast_fp16")]; + tensor var_5739_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_565_cast_fp16)[name = string("op_5739_cast_fp16")]; + tensor var_5740_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_567_cast_fp16)[name = string("op_5740_cast_fp16")]; + tensor var_5741_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_569_cast_fp16)[name = string("op_5741_cast_fp16")]; + tensor var_5742_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_571_cast_fp16)[name = string("op_5742_cast_fp16")]; + tensor var_5743_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_573_cast_fp16)[name = string("op_5743_cast_fp16")]; + tensor var_5744_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_575_cast_fp16)[name = string("op_5744_cast_fp16")]; + string var_5746_equation_0 = const()[name = string("op_5746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5746_cast_fp16 = einsum(equation = var_5746_equation_0, values = (var_5458_cast_fp16, var_5697_cast_fp16))[name = string("op_5746_cast_fp16")]; + string var_5748_equation_0 = const()[name = string("op_5748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5748_cast_fp16 = einsum(equation = var_5748_equation_0, values = (var_5458_cast_fp16, var_5698_cast_fp16))[name = string("op_5748_cast_fp16")]; + string var_5750_equation_0 = const()[name = string("op_5750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5750_cast_fp16 = einsum(equation = var_5750_equation_0, values = (var_5458_cast_fp16, var_5699_cast_fp16))[name = string("op_5750_cast_fp16")]; + string var_5752_equation_0 = const()[name = string("op_5752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5752_cast_fp16 = einsum(equation = var_5752_equation_0, values = (var_5458_cast_fp16, var_5700_cast_fp16))[name = string("op_5752_cast_fp16")]; + string var_5754_equation_0 = const()[name = string("op_5754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5754_cast_fp16 = einsum(equation = var_5754_equation_0, values = (var_5462_cast_fp16, var_5701_cast_fp16))[name = string("op_5754_cast_fp16")]; + string var_5756_equation_0 = const()[name = string("op_5756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5756_cast_fp16 = einsum(equation = var_5756_equation_0, values = (var_5462_cast_fp16, var_5702_cast_fp16))[name = string("op_5756_cast_fp16")]; + string var_5758_equation_0 = const()[name = string("op_5758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5758_cast_fp16 = einsum(equation = var_5758_equation_0, values = (var_5462_cast_fp16, var_5703_cast_fp16))[name = string("op_5758_cast_fp16")]; + string var_5760_equation_0 = const()[name = string("op_5760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5760_cast_fp16 = einsum(equation = var_5760_equation_0, values = (var_5462_cast_fp16, var_5704_cast_fp16))[name = string("op_5760_cast_fp16")]; + string var_5762_equation_0 = const()[name = string("op_5762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5762_cast_fp16 = einsum(equation = var_5762_equation_0, values = (var_5466_cast_fp16, var_5705_cast_fp16))[name = string("op_5762_cast_fp16")]; + string var_5764_equation_0 = const()[name = string("op_5764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5764_cast_fp16 = einsum(equation = var_5764_equation_0, values = (var_5466_cast_fp16, var_5706_cast_fp16))[name = string("op_5764_cast_fp16")]; + string var_5766_equation_0 = const()[name = string("op_5766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5766_cast_fp16 = einsum(equation = var_5766_equation_0, values = (var_5466_cast_fp16, var_5707_cast_fp16))[name = string("op_5766_cast_fp16")]; + string var_5768_equation_0 = const()[name = string("op_5768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5768_cast_fp16 = einsum(equation = var_5768_equation_0, values = (var_5466_cast_fp16, var_5708_cast_fp16))[name = string("op_5768_cast_fp16")]; + string var_5770_equation_0 = const()[name = string("op_5770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5770_cast_fp16 = einsum(equation = var_5770_equation_0, values = (var_5470_cast_fp16, var_5709_cast_fp16))[name = string("op_5770_cast_fp16")]; + string var_5772_equation_0 = const()[name = string("op_5772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5772_cast_fp16 = einsum(equation = var_5772_equation_0, values = (var_5470_cast_fp16, var_5710_cast_fp16))[name = string("op_5772_cast_fp16")]; + string var_5774_equation_0 = const()[name = string("op_5774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5774_cast_fp16 = einsum(equation = var_5774_equation_0, values = (var_5470_cast_fp16, var_5711_cast_fp16))[name = string("op_5774_cast_fp16")]; + string var_5776_equation_0 = const()[name = string("op_5776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5776_cast_fp16 = einsum(equation = var_5776_equation_0, values = (var_5470_cast_fp16, var_5712_cast_fp16))[name = string("op_5776_cast_fp16")]; + string var_5778_equation_0 = const()[name = string("op_5778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5778_cast_fp16 = einsum(equation = var_5778_equation_0, values = (var_5474_cast_fp16, var_5713_cast_fp16))[name = string("op_5778_cast_fp16")]; + string var_5780_equation_0 = const()[name = string("op_5780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5780_cast_fp16 = einsum(equation = var_5780_equation_0, values = (var_5474_cast_fp16, var_5714_cast_fp16))[name = string("op_5780_cast_fp16")]; + string var_5782_equation_0 = const()[name = string("op_5782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5782_cast_fp16 = einsum(equation = var_5782_equation_0, values = (var_5474_cast_fp16, var_5715_cast_fp16))[name = string("op_5782_cast_fp16")]; + string var_5784_equation_0 = const()[name = string("op_5784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5784_cast_fp16 = einsum(equation = var_5784_equation_0, values = (var_5474_cast_fp16, var_5716_cast_fp16))[name = string("op_5784_cast_fp16")]; + string var_5786_equation_0 = const()[name = string("op_5786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5786_cast_fp16 = einsum(equation = var_5786_equation_0, values = (var_5478_cast_fp16, var_5717_cast_fp16))[name = string("op_5786_cast_fp16")]; + string var_5788_equation_0 = const()[name = string("op_5788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5788_cast_fp16 = einsum(equation = var_5788_equation_0, values = (var_5478_cast_fp16, var_5718_cast_fp16))[name = string("op_5788_cast_fp16")]; + string var_5790_equation_0 = const()[name = string("op_5790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5790_cast_fp16 = einsum(equation = var_5790_equation_0, values = (var_5478_cast_fp16, var_5719_cast_fp16))[name = string("op_5790_cast_fp16")]; + string var_5792_equation_0 = const()[name = string("op_5792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5792_cast_fp16 = einsum(equation = var_5792_equation_0, values = (var_5478_cast_fp16, var_5720_cast_fp16))[name = string("op_5792_cast_fp16")]; + string var_5794_equation_0 = const()[name = string("op_5794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5794_cast_fp16 = einsum(equation = var_5794_equation_0, values = (var_5482_cast_fp16, var_5721_cast_fp16))[name = string("op_5794_cast_fp16")]; + string var_5796_equation_0 = const()[name = string("op_5796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5796_cast_fp16 = einsum(equation = var_5796_equation_0, values = (var_5482_cast_fp16, var_5722_cast_fp16))[name = string("op_5796_cast_fp16")]; + string var_5798_equation_0 = const()[name = string("op_5798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5798_cast_fp16 = einsum(equation = var_5798_equation_0, values = (var_5482_cast_fp16, var_5723_cast_fp16))[name = string("op_5798_cast_fp16")]; + string var_5800_equation_0 = const()[name = string("op_5800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5800_cast_fp16 = einsum(equation = var_5800_equation_0, values = (var_5482_cast_fp16, var_5724_cast_fp16))[name = string("op_5800_cast_fp16")]; + string var_5802_equation_0 = const()[name = string("op_5802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5802_cast_fp16 = einsum(equation = var_5802_equation_0, values = (var_5486_cast_fp16, var_5725_cast_fp16))[name = string("op_5802_cast_fp16")]; + string var_5804_equation_0 = const()[name = string("op_5804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5804_cast_fp16 = einsum(equation = var_5804_equation_0, values = (var_5486_cast_fp16, var_5726_cast_fp16))[name = string("op_5804_cast_fp16")]; + string var_5806_equation_0 = const()[name = string("op_5806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5806_cast_fp16 = einsum(equation = var_5806_equation_0, values = (var_5486_cast_fp16, var_5727_cast_fp16))[name = string("op_5806_cast_fp16")]; + string var_5808_equation_0 = const()[name = string("op_5808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5808_cast_fp16 = einsum(equation = var_5808_equation_0, values = (var_5486_cast_fp16, var_5728_cast_fp16))[name = string("op_5808_cast_fp16")]; + string var_5810_equation_0 = const()[name = string("op_5810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5810_cast_fp16 = einsum(equation = var_5810_equation_0, values = (var_5490_cast_fp16, var_5729_cast_fp16))[name = string("op_5810_cast_fp16")]; + string var_5812_equation_0 = const()[name = string("op_5812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5812_cast_fp16 = einsum(equation = var_5812_equation_0, values = (var_5490_cast_fp16, var_5730_cast_fp16))[name = string("op_5812_cast_fp16")]; + string var_5814_equation_0 = const()[name = string("op_5814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5814_cast_fp16 = einsum(equation = var_5814_equation_0, values = (var_5490_cast_fp16, var_5731_cast_fp16))[name = string("op_5814_cast_fp16")]; + string var_5816_equation_0 = const()[name = string("op_5816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5816_cast_fp16 = einsum(equation = var_5816_equation_0, values = (var_5490_cast_fp16, var_5732_cast_fp16))[name = string("op_5816_cast_fp16")]; + string var_5818_equation_0 = const()[name = string("op_5818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5818_cast_fp16 = einsum(equation = var_5818_equation_0, values = (var_5494_cast_fp16, var_5733_cast_fp16))[name = string("op_5818_cast_fp16")]; + string var_5820_equation_0 = const()[name = string("op_5820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5820_cast_fp16 = einsum(equation = var_5820_equation_0, values = (var_5494_cast_fp16, var_5734_cast_fp16))[name = string("op_5820_cast_fp16")]; + string var_5822_equation_0 = const()[name = string("op_5822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5822_cast_fp16 = einsum(equation = var_5822_equation_0, values = (var_5494_cast_fp16, var_5735_cast_fp16))[name = string("op_5822_cast_fp16")]; + string var_5824_equation_0 = const()[name = string("op_5824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5824_cast_fp16 = einsum(equation = var_5824_equation_0, values = (var_5494_cast_fp16, var_5736_cast_fp16))[name = string("op_5824_cast_fp16")]; + string var_5826_equation_0 = const()[name = string("op_5826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5826_cast_fp16 = einsum(equation = var_5826_equation_0, values = (var_5498_cast_fp16, var_5737_cast_fp16))[name = string("op_5826_cast_fp16")]; + string var_5828_equation_0 = const()[name = string("op_5828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5828_cast_fp16 = einsum(equation = var_5828_equation_0, values = (var_5498_cast_fp16, var_5738_cast_fp16))[name = string("op_5828_cast_fp16")]; + string var_5830_equation_0 = const()[name = string("op_5830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5830_cast_fp16 = einsum(equation = var_5830_equation_0, values = (var_5498_cast_fp16, var_5739_cast_fp16))[name = string("op_5830_cast_fp16")]; + string var_5832_equation_0 = const()[name = string("op_5832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5832_cast_fp16 = einsum(equation = var_5832_equation_0, values = (var_5498_cast_fp16, var_5740_cast_fp16))[name = string("op_5832_cast_fp16")]; + string var_5834_equation_0 = const()[name = string("op_5834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5834_cast_fp16 = einsum(equation = var_5834_equation_0, values = (var_5502_cast_fp16, var_5741_cast_fp16))[name = string("op_5834_cast_fp16")]; + string var_5836_equation_0 = const()[name = string("op_5836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5836_cast_fp16 = einsum(equation = var_5836_equation_0, values = (var_5502_cast_fp16, var_5742_cast_fp16))[name = string("op_5836_cast_fp16")]; + string var_5838_equation_0 = const()[name = string("op_5838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5838_cast_fp16 = einsum(equation = var_5838_equation_0, values = (var_5502_cast_fp16, var_5743_cast_fp16))[name = string("op_5838_cast_fp16")]; + string var_5840_equation_0 = const()[name = string("op_5840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5840_cast_fp16 = einsum(equation = var_5840_equation_0, values = (var_5502_cast_fp16, var_5744_cast_fp16))[name = string("op_5840_cast_fp16")]; + bool var_5842_interleave_0 = const()[name = string("op_5842_interleave_0"), val = bool(false)]; + tensor var_5842_cast_fp16 = concat(axis = var_4953, interleave = var_5842_interleave_0, values = (var_5746_cast_fp16, var_5748_cast_fp16, var_5750_cast_fp16, var_5752_cast_fp16))[name = string("op_5842_cast_fp16")]; + bool var_5844_interleave_0 = const()[name = string("op_5844_interleave_0"), val = bool(false)]; + tensor var_5844_cast_fp16 = concat(axis = var_4953, interleave = var_5844_interleave_0, values = (var_5754_cast_fp16, var_5756_cast_fp16, var_5758_cast_fp16, var_5760_cast_fp16))[name = string("op_5844_cast_fp16")]; + bool var_5846_interleave_0 = const()[name = string("op_5846_interleave_0"), val = bool(false)]; + tensor var_5846_cast_fp16 = concat(axis = var_4953, interleave = var_5846_interleave_0, values = (var_5762_cast_fp16, var_5764_cast_fp16, var_5766_cast_fp16, var_5768_cast_fp16))[name = string("op_5846_cast_fp16")]; + bool var_5848_interleave_0 = const()[name = string("op_5848_interleave_0"), val = bool(false)]; + tensor var_5848_cast_fp16 = concat(axis = var_4953, interleave = var_5848_interleave_0, values = (var_5770_cast_fp16, var_5772_cast_fp16, var_5774_cast_fp16, var_5776_cast_fp16))[name = string("op_5848_cast_fp16")]; + bool var_5850_interleave_0 = const()[name = string("op_5850_interleave_0"), val = bool(false)]; + tensor var_5850_cast_fp16 = concat(axis = var_4953, interleave = var_5850_interleave_0, values = (var_5778_cast_fp16, var_5780_cast_fp16, var_5782_cast_fp16, var_5784_cast_fp16))[name = string("op_5850_cast_fp16")]; + bool var_5852_interleave_0 = const()[name = string("op_5852_interleave_0"), val = bool(false)]; + tensor var_5852_cast_fp16 = concat(axis = var_4953, interleave = var_5852_interleave_0, values = (var_5786_cast_fp16, var_5788_cast_fp16, var_5790_cast_fp16, var_5792_cast_fp16))[name = string("op_5852_cast_fp16")]; + bool var_5854_interleave_0 = const()[name = string("op_5854_interleave_0"), val = bool(false)]; + tensor var_5854_cast_fp16 = concat(axis = var_4953, interleave = var_5854_interleave_0, values = (var_5794_cast_fp16, var_5796_cast_fp16, var_5798_cast_fp16, var_5800_cast_fp16))[name = string("op_5854_cast_fp16")]; + bool var_5856_interleave_0 = const()[name = string("op_5856_interleave_0"), val = bool(false)]; + tensor var_5856_cast_fp16 = concat(axis = var_4953, interleave = var_5856_interleave_0, values = (var_5802_cast_fp16, var_5804_cast_fp16, var_5806_cast_fp16, var_5808_cast_fp16))[name = string("op_5856_cast_fp16")]; + bool var_5858_interleave_0 = const()[name = string("op_5858_interleave_0"), val = bool(false)]; + tensor var_5858_cast_fp16 = concat(axis = var_4953, interleave = var_5858_interleave_0, values = (var_5810_cast_fp16, var_5812_cast_fp16, var_5814_cast_fp16, var_5816_cast_fp16))[name = string("op_5858_cast_fp16")]; + bool var_5860_interleave_0 = const()[name = string("op_5860_interleave_0"), val = bool(false)]; + tensor var_5860_cast_fp16 = concat(axis = var_4953, interleave = var_5860_interleave_0, values = (var_5818_cast_fp16, var_5820_cast_fp16, var_5822_cast_fp16, var_5824_cast_fp16))[name = string("op_5860_cast_fp16")]; + bool var_5862_interleave_0 = const()[name = string("op_5862_interleave_0"), val = bool(false)]; + tensor var_5862_cast_fp16 = concat(axis = var_4953, interleave = var_5862_interleave_0, values = (var_5826_cast_fp16, var_5828_cast_fp16, var_5830_cast_fp16, var_5832_cast_fp16))[name = string("op_5862_cast_fp16")]; + bool var_5864_interleave_0 = const()[name = string("op_5864_interleave_0"), val = bool(false)]; + tensor var_5864_cast_fp16 = concat(axis = var_4953, interleave = var_5864_interleave_0, values = (var_5834_cast_fp16, var_5836_cast_fp16, var_5838_cast_fp16, var_5840_cast_fp16))[name = string("op_5864_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_4970, interleave = input_41_interleave_0, values = (var_5842_cast_fp16, var_5844_cast_fp16, var_5846_cast_fp16, var_5848_cast_fp16, var_5850_cast_fp16, var_5852_cast_fp16, var_5854_cast_fp16, var_5856_cast_fp16, var_5858_cast_fp16, var_5860_cast_fp16, var_5862_cast_fp16, var_5864_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; + tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; + tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; + int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80639616)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81819328)))]; + tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_5883_to_fp16 = const()[name = string("op_5883_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_5883_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81820928)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81822528)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81824128)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86542784)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86548992)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91267648)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; + int32 var_5912 = const()[name = string("op_5912"), val = int32(3)]; + int32 var_5929 = const()[name = string("op_5929"), val = int32(1)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_5946_to_fp16 = const()[name = string("op_5946_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_5946_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91269248)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91270848)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91272448)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92452160)))]; + tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; + string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; + tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; + tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; + int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92453760)))]; + tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; + string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; + tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; + tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; + int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93633472)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94813184)))]; + tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_5984_begin_0 = const()[name = string("op_5984_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5984_end_0 = const()[name = string("op_5984_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5984_end_mask_0 = const()[name = string("op_5984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5984_cast_fp16 = slice_by_index(begin = var_5984_begin_0, end = var_5984_end_0, end_mask = var_5984_end_mask_0, x = query_13_cast_fp16)[name = string("op_5984_cast_fp16")]; + tensor var_5988_begin_0 = const()[name = string("op_5988_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5988_end_0 = const()[name = string("op_5988_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5988_end_mask_0 = const()[name = string("op_5988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5988_cast_fp16 = slice_by_index(begin = var_5988_begin_0, end = var_5988_end_0, end_mask = var_5988_end_mask_0, x = query_13_cast_fp16)[name = string("op_5988_cast_fp16")]; + tensor var_5992_begin_0 = const()[name = string("op_5992_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5992_end_0 = const()[name = string("op_5992_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5992_end_mask_0 = const()[name = string("op_5992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5992_cast_fp16 = slice_by_index(begin = var_5992_begin_0, end = var_5992_end_0, end_mask = var_5992_end_mask_0, x = query_13_cast_fp16)[name = string("op_5992_cast_fp16")]; + tensor var_5996_begin_0 = const()[name = string("op_5996_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5996_end_0 = const()[name = string("op_5996_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5996_end_mask_0 = const()[name = string("op_5996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5996_cast_fp16 = slice_by_index(begin = var_5996_begin_0, end = var_5996_end_0, end_mask = var_5996_end_mask_0, x = query_13_cast_fp16)[name = string("op_5996_cast_fp16")]; + tensor var_6000_begin_0 = const()[name = string("op_6000_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6000_end_0 = const()[name = string("op_6000_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6000_end_mask_0 = const()[name = string("op_6000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6000_cast_fp16 = slice_by_index(begin = var_6000_begin_0, end = var_6000_end_0, end_mask = var_6000_end_mask_0, x = query_13_cast_fp16)[name = string("op_6000_cast_fp16")]; + tensor var_6004_begin_0 = const()[name = string("op_6004_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6004_end_0 = const()[name = string("op_6004_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6004_end_mask_0 = const()[name = string("op_6004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6004_cast_fp16 = slice_by_index(begin = var_6004_begin_0, end = var_6004_end_0, end_mask = var_6004_end_mask_0, x = query_13_cast_fp16)[name = string("op_6004_cast_fp16")]; + tensor var_6008_begin_0 = const()[name = string("op_6008_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6008_end_0 = const()[name = string("op_6008_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6008_end_mask_0 = const()[name = string("op_6008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6008_cast_fp16 = slice_by_index(begin = var_6008_begin_0, end = var_6008_end_0, end_mask = var_6008_end_mask_0, x = query_13_cast_fp16)[name = string("op_6008_cast_fp16")]; + tensor var_6012_begin_0 = const()[name = string("op_6012_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6012_end_0 = const()[name = string("op_6012_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6012_end_mask_0 = const()[name = string("op_6012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6012_cast_fp16 = slice_by_index(begin = var_6012_begin_0, end = var_6012_end_0, end_mask = var_6012_end_mask_0, x = query_13_cast_fp16)[name = string("op_6012_cast_fp16")]; + tensor var_6016_begin_0 = const()[name = string("op_6016_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6016_end_0 = const()[name = string("op_6016_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6016_end_mask_0 = const()[name = string("op_6016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6016_cast_fp16 = slice_by_index(begin = var_6016_begin_0, end = var_6016_end_0, end_mask = var_6016_end_mask_0, x = query_13_cast_fp16)[name = string("op_6016_cast_fp16")]; + tensor var_6020_begin_0 = const()[name = string("op_6020_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6020_end_0 = const()[name = string("op_6020_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6020_end_mask_0 = const()[name = string("op_6020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6020_cast_fp16 = slice_by_index(begin = var_6020_begin_0, end = var_6020_end_0, end_mask = var_6020_end_mask_0, x = query_13_cast_fp16)[name = string("op_6020_cast_fp16")]; + tensor var_6024_begin_0 = const()[name = string("op_6024_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6024_end_0 = const()[name = string("op_6024_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6024_end_mask_0 = const()[name = string("op_6024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6024_cast_fp16 = slice_by_index(begin = var_6024_begin_0, end = var_6024_end_0, end_mask = var_6024_end_mask_0, x = query_13_cast_fp16)[name = string("op_6024_cast_fp16")]; + tensor var_6028_begin_0 = const()[name = string("op_6028_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6028_end_0 = const()[name = string("op_6028_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6028_end_mask_0 = const()[name = string("op_6028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6028_cast_fp16 = slice_by_index(begin = var_6028_begin_0, end = var_6028_end_0, end_mask = var_6028_end_mask_0, x = query_13_cast_fp16)[name = string("op_6028_cast_fp16")]; + tensor var_6037_begin_0 = const()[name = string("op_6037_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6037_end_0 = const()[name = string("op_6037_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6037_end_mask_0 = const()[name = string("op_6037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6037_cast_fp16 = slice_by_index(begin = var_6037_begin_0, end = var_6037_end_0, end_mask = var_6037_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6037_cast_fp16")]; + tensor var_6044_begin_0 = const()[name = string("op_6044_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6044_end_0 = const()[name = string("op_6044_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6044_end_mask_0 = const()[name = string("op_6044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6044_cast_fp16 = slice_by_index(begin = var_6044_begin_0, end = var_6044_end_0, end_mask = var_6044_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6044_cast_fp16")]; + tensor var_6051_begin_0 = const()[name = string("op_6051_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6051_end_0 = const()[name = string("op_6051_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6051_end_mask_0 = const()[name = string("op_6051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6051_cast_fp16 = slice_by_index(begin = var_6051_begin_0, end = var_6051_end_0, end_mask = var_6051_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6051_cast_fp16")]; + tensor var_6058_begin_0 = const()[name = string("op_6058_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6058_end_0 = const()[name = string("op_6058_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6058_end_mask_0 = const()[name = string("op_6058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6058_cast_fp16 = slice_by_index(begin = var_6058_begin_0, end = var_6058_end_0, end_mask = var_6058_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6058_cast_fp16")]; + tensor var_6065_begin_0 = const()[name = string("op_6065_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6065_end_0 = const()[name = string("op_6065_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6065_end_mask_0 = const()[name = string("op_6065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6065_cast_fp16 = slice_by_index(begin = var_6065_begin_0, end = var_6065_end_0, end_mask = var_6065_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6065_cast_fp16")]; + tensor var_6072_begin_0 = const()[name = string("op_6072_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6072_end_0 = const()[name = string("op_6072_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6072_end_mask_0 = const()[name = string("op_6072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6072_cast_fp16 = slice_by_index(begin = var_6072_begin_0, end = var_6072_end_0, end_mask = var_6072_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6072_cast_fp16")]; + tensor var_6079_begin_0 = const()[name = string("op_6079_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6079_end_0 = const()[name = string("op_6079_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6079_end_mask_0 = const()[name = string("op_6079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6079_cast_fp16 = slice_by_index(begin = var_6079_begin_0, end = var_6079_end_0, end_mask = var_6079_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6079_cast_fp16")]; + tensor var_6086_begin_0 = const()[name = string("op_6086_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6086_end_0 = const()[name = string("op_6086_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6086_end_mask_0 = const()[name = string("op_6086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6086_cast_fp16 = slice_by_index(begin = var_6086_begin_0, end = var_6086_end_0, end_mask = var_6086_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6086_cast_fp16")]; + tensor var_6093_begin_0 = const()[name = string("op_6093_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6093_end_0 = const()[name = string("op_6093_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6093_end_mask_0 = const()[name = string("op_6093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6093_cast_fp16 = slice_by_index(begin = var_6093_begin_0, end = var_6093_end_0, end_mask = var_6093_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6093_cast_fp16")]; + tensor var_6100_begin_0 = const()[name = string("op_6100_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6100_end_0 = const()[name = string("op_6100_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6100_end_mask_0 = const()[name = string("op_6100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6100_cast_fp16 = slice_by_index(begin = var_6100_begin_0, end = var_6100_end_0, end_mask = var_6100_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6100_cast_fp16")]; + tensor var_6107_begin_0 = const()[name = string("op_6107_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6107_end_0 = const()[name = string("op_6107_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6107_end_mask_0 = const()[name = string("op_6107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6107_cast_fp16 = slice_by_index(begin = var_6107_begin_0, end = var_6107_end_0, end_mask = var_6107_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6107_cast_fp16")]; + tensor var_6114_begin_0 = const()[name = string("op_6114_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6114_end_0 = const()[name = string("op_6114_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6114_end_mask_0 = const()[name = string("op_6114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6114_cast_fp16 = slice_by_index(begin = var_6114_begin_0, end = var_6114_end_0, end_mask = var_6114_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6114_cast_fp16")]; + tensor var_6121_begin_0 = const()[name = string("op_6121_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6121_end_0 = const()[name = string("op_6121_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6121_end_mask_0 = const()[name = string("op_6121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6121_cast_fp16 = slice_by_index(begin = var_6121_begin_0, end = var_6121_end_0, end_mask = var_6121_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6121_cast_fp16")]; + tensor var_6128_begin_0 = const()[name = string("op_6128_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6128_end_0 = const()[name = string("op_6128_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6128_end_mask_0 = const()[name = string("op_6128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6128_cast_fp16 = slice_by_index(begin = var_6128_begin_0, end = var_6128_end_0, end_mask = var_6128_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6128_cast_fp16")]; + tensor var_6135_begin_0 = const()[name = string("op_6135_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6135_end_0 = const()[name = string("op_6135_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6135_end_mask_0 = const()[name = string("op_6135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6135_cast_fp16 = slice_by_index(begin = var_6135_begin_0, end = var_6135_end_0, end_mask = var_6135_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6135_cast_fp16")]; + tensor var_6142_begin_0 = const()[name = string("op_6142_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6142_end_0 = const()[name = string("op_6142_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6142_end_mask_0 = const()[name = string("op_6142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6142_cast_fp16 = slice_by_index(begin = var_6142_begin_0, end = var_6142_end_0, end_mask = var_6142_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6142_cast_fp16")]; + tensor var_6149_begin_0 = const()[name = string("op_6149_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6149_end_0 = const()[name = string("op_6149_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6149_end_mask_0 = const()[name = string("op_6149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6149_cast_fp16 = slice_by_index(begin = var_6149_begin_0, end = var_6149_end_0, end_mask = var_6149_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6149_cast_fp16")]; + tensor var_6156_begin_0 = const()[name = string("op_6156_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6156_end_0 = const()[name = string("op_6156_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6156_end_mask_0 = const()[name = string("op_6156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6156_cast_fp16 = slice_by_index(begin = var_6156_begin_0, end = var_6156_end_0, end_mask = var_6156_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6156_cast_fp16")]; + tensor var_6163_begin_0 = const()[name = string("op_6163_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6163_end_0 = const()[name = string("op_6163_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6163_end_mask_0 = const()[name = string("op_6163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6163_cast_fp16 = slice_by_index(begin = var_6163_begin_0, end = var_6163_end_0, end_mask = var_6163_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6163_cast_fp16")]; + tensor var_6170_begin_0 = const()[name = string("op_6170_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6170_end_0 = const()[name = string("op_6170_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6170_end_mask_0 = const()[name = string("op_6170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6170_cast_fp16 = slice_by_index(begin = var_6170_begin_0, end = var_6170_end_0, end_mask = var_6170_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6170_cast_fp16")]; + tensor var_6177_begin_0 = const()[name = string("op_6177_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6177_end_0 = const()[name = string("op_6177_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6177_end_mask_0 = const()[name = string("op_6177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6177_cast_fp16 = slice_by_index(begin = var_6177_begin_0, end = var_6177_end_0, end_mask = var_6177_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6177_cast_fp16")]; + tensor var_6184_begin_0 = const()[name = string("op_6184_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6184_end_0 = const()[name = string("op_6184_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6184_end_mask_0 = const()[name = string("op_6184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6184_cast_fp16 = slice_by_index(begin = var_6184_begin_0, end = var_6184_end_0, end_mask = var_6184_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6184_cast_fp16")]; + tensor var_6191_begin_0 = const()[name = string("op_6191_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6191_end_0 = const()[name = string("op_6191_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6191_end_mask_0 = const()[name = string("op_6191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6191_cast_fp16 = slice_by_index(begin = var_6191_begin_0, end = var_6191_end_0, end_mask = var_6191_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6191_cast_fp16")]; + tensor var_6198_begin_0 = const()[name = string("op_6198_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6198_end_0 = const()[name = string("op_6198_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6198_end_mask_0 = const()[name = string("op_6198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6198_cast_fp16 = slice_by_index(begin = var_6198_begin_0, end = var_6198_end_0, end_mask = var_6198_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6198_cast_fp16")]; + tensor var_6205_begin_0 = const()[name = string("op_6205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6205_end_0 = const()[name = string("op_6205_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6205_end_mask_0 = const()[name = string("op_6205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6205_cast_fp16 = slice_by_index(begin = var_6205_begin_0, end = var_6205_end_0, end_mask = var_6205_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6205_cast_fp16")]; + tensor var_6212_begin_0 = const()[name = string("op_6212_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6212_end_0 = const()[name = string("op_6212_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6212_end_mask_0 = const()[name = string("op_6212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6212_cast_fp16 = slice_by_index(begin = var_6212_begin_0, end = var_6212_end_0, end_mask = var_6212_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6212_cast_fp16")]; + tensor var_6219_begin_0 = const()[name = string("op_6219_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6219_end_0 = const()[name = string("op_6219_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6219_end_mask_0 = const()[name = string("op_6219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6219_cast_fp16 = slice_by_index(begin = var_6219_begin_0, end = var_6219_end_0, end_mask = var_6219_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6219_cast_fp16")]; + tensor var_6226_begin_0 = const()[name = string("op_6226_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6226_end_0 = const()[name = string("op_6226_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6226_end_mask_0 = const()[name = string("op_6226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6226_cast_fp16 = slice_by_index(begin = var_6226_begin_0, end = var_6226_end_0, end_mask = var_6226_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6226_cast_fp16")]; + tensor var_6233_begin_0 = const()[name = string("op_6233_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6233_end_0 = const()[name = string("op_6233_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6233_end_mask_0 = const()[name = string("op_6233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6233_cast_fp16 = slice_by_index(begin = var_6233_begin_0, end = var_6233_end_0, end_mask = var_6233_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6233_cast_fp16")]; + tensor var_6240_begin_0 = const()[name = string("op_6240_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6240_end_0 = const()[name = string("op_6240_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6240_end_mask_0 = const()[name = string("op_6240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6240_cast_fp16 = slice_by_index(begin = var_6240_begin_0, end = var_6240_end_0, end_mask = var_6240_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6240_cast_fp16")]; + tensor var_6247_begin_0 = const()[name = string("op_6247_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6247_end_0 = const()[name = string("op_6247_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6247_end_mask_0 = const()[name = string("op_6247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6247_cast_fp16 = slice_by_index(begin = var_6247_begin_0, end = var_6247_end_0, end_mask = var_6247_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6247_cast_fp16")]; + tensor var_6254_begin_0 = const()[name = string("op_6254_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6254_end_0 = const()[name = string("op_6254_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6254_end_mask_0 = const()[name = string("op_6254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6254_cast_fp16 = slice_by_index(begin = var_6254_begin_0, end = var_6254_end_0, end_mask = var_6254_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6254_cast_fp16")]; + tensor var_6261_begin_0 = const()[name = string("op_6261_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6261_end_0 = const()[name = string("op_6261_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6261_end_mask_0 = const()[name = string("op_6261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6261_cast_fp16 = slice_by_index(begin = var_6261_begin_0, end = var_6261_end_0, end_mask = var_6261_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6261_cast_fp16")]; + tensor var_6268_begin_0 = const()[name = string("op_6268_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6268_end_0 = const()[name = string("op_6268_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6268_end_mask_0 = const()[name = string("op_6268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6268_cast_fp16 = slice_by_index(begin = var_6268_begin_0, end = var_6268_end_0, end_mask = var_6268_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6268_cast_fp16")]; + tensor var_6275_begin_0 = const()[name = string("op_6275_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6275_end_0 = const()[name = string("op_6275_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6275_end_mask_0 = const()[name = string("op_6275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6275_cast_fp16 = slice_by_index(begin = var_6275_begin_0, end = var_6275_end_0, end_mask = var_6275_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6275_cast_fp16")]; + tensor var_6282_begin_0 = const()[name = string("op_6282_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6282_end_0 = const()[name = string("op_6282_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6282_end_mask_0 = const()[name = string("op_6282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6282_cast_fp16 = slice_by_index(begin = var_6282_begin_0, end = var_6282_end_0, end_mask = var_6282_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6282_cast_fp16")]; + tensor var_6289_begin_0 = const()[name = string("op_6289_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6289_end_0 = const()[name = string("op_6289_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6289_end_mask_0 = const()[name = string("op_6289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6289_cast_fp16 = slice_by_index(begin = var_6289_begin_0, end = var_6289_end_0, end_mask = var_6289_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6289_cast_fp16")]; + tensor var_6296_begin_0 = const()[name = string("op_6296_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6296_end_0 = const()[name = string("op_6296_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6296_end_mask_0 = const()[name = string("op_6296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6296_cast_fp16 = slice_by_index(begin = var_6296_begin_0, end = var_6296_end_0, end_mask = var_6296_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6296_cast_fp16")]; + tensor var_6303_begin_0 = const()[name = string("op_6303_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6303_end_0 = const()[name = string("op_6303_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6303_end_mask_0 = const()[name = string("op_6303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6303_cast_fp16 = slice_by_index(begin = var_6303_begin_0, end = var_6303_end_0, end_mask = var_6303_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6303_cast_fp16")]; + tensor var_6310_begin_0 = const()[name = string("op_6310_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6310_end_0 = const()[name = string("op_6310_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6310_end_mask_0 = const()[name = string("op_6310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6310_cast_fp16 = slice_by_index(begin = var_6310_begin_0, end = var_6310_end_0, end_mask = var_6310_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6310_cast_fp16")]; + tensor var_6317_begin_0 = const()[name = string("op_6317_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6317_end_0 = const()[name = string("op_6317_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6317_end_mask_0 = const()[name = string("op_6317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6317_cast_fp16 = slice_by_index(begin = var_6317_begin_0, end = var_6317_end_0, end_mask = var_6317_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6317_cast_fp16")]; + tensor var_6324_begin_0 = const()[name = string("op_6324_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6324_end_0 = const()[name = string("op_6324_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6324_end_mask_0 = const()[name = string("op_6324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6324_cast_fp16 = slice_by_index(begin = var_6324_begin_0, end = var_6324_end_0, end_mask = var_6324_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6324_cast_fp16")]; + tensor var_6331_begin_0 = const()[name = string("op_6331_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6331_end_0 = const()[name = string("op_6331_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6331_end_mask_0 = const()[name = string("op_6331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6331_cast_fp16 = slice_by_index(begin = var_6331_begin_0, end = var_6331_end_0, end_mask = var_6331_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6331_cast_fp16")]; + tensor var_6338_begin_0 = const()[name = string("op_6338_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6338_end_0 = const()[name = string("op_6338_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6338_end_mask_0 = const()[name = string("op_6338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6338_cast_fp16 = slice_by_index(begin = var_6338_begin_0, end = var_6338_end_0, end_mask = var_6338_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6338_cast_fp16")]; + tensor var_6345_begin_0 = const()[name = string("op_6345_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6345_end_0 = const()[name = string("op_6345_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6345_end_mask_0 = const()[name = string("op_6345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6345_cast_fp16 = slice_by_index(begin = var_6345_begin_0, end = var_6345_end_0, end_mask = var_6345_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6345_cast_fp16")]; + tensor var_6352_begin_0 = const()[name = string("op_6352_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6352_end_0 = const()[name = string("op_6352_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6352_end_mask_0 = const()[name = string("op_6352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6352_cast_fp16 = slice_by_index(begin = var_6352_begin_0, end = var_6352_end_0, end_mask = var_6352_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6352_cast_fp16")]; + tensor var_6359_begin_0 = const()[name = string("op_6359_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6359_end_0 = const()[name = string("op_6359_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6359_end_mask_0 = const()[name = string("op_6359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6359_cast_fp16 = slice_by_index(begin = var_6359_begin_0, end = var_6359_end_0, end_mask = var_6359_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6359_cast_fp16")]; + tensor var_6366_begin_0 = const()[name = string("op_6366_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6366_end_0 = const()[name = string("op_6366_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6366_end_mask_0 = const()[name = string("op_6366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6366_cast_fp16 = slice_by_index(begin = var_6366_begin_0, end = var_6366_end_0, end_mask = var_6366_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6366_cast_fp16")]; + tensor k_13_perm_0 = const()[name = string("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6371_begin_0 = const()[name = string("op_6371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6371_end_0 = const()[name = string("op_6371_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_6371_end_mask_0 = const()[name = string("op_6371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_13_cast_fp16 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = string("transpose_5")]; + tensor var_6371_cast_fp16 = slice_by_index(begin = var_6371_begin_0, end = var_6371_end_0, end_mask = var_6371_end_mask_0, x = k_13_cast_fp16)[name = string("op_6371_cast_fp16")]; + tensor var_6375_begin_0 = const()[name = string("op_6375_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6375_end_0 = const()[name = string("op_6375_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_6375_end_mask_0 = const()[name = string("op_6375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6375_cast_fp16 = slice_by_index(begin = var_6375_begin_0, end = var_6375_end_0, end_mask = var_6375_end_mask_0, x = k_13_cast_fp16)[name = string("op_6375_cast_fp16")]; + tensor var_6379_begin_0 = const()[name = string("op_6379_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6379_end_0 = const()[name = string("op_6379_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_6379_end_mask_0 = const()[name = string("op_6379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6379_cast_fp16 = slice_by_index(begin = var_6379_begin_0, end = var_6379_end_0, end_mask = var_6379_end_mask_0, x = k_13_cast_fp16)[name = string("op_6379_cast_fp16")]; + tensor var_6383_begin_0 = const()[name = string("op_6383_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6383_end_0 = const()[name = string("op_6383_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_6383_end_mask_0 = const()[name = string("op_6383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6383_cast_fp16 = slice_by_index(begin = var_6383_begin_0, end = var_6383_end_0, end_mask = var_6383_end_mask_0, x = k_13_cast_fp16)[name = string("op_6383_cast_fp16")]; + tensor var_6387_begin_0 = const()[name = string("op_6387_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6387_end_0 = const()[name = string("op_6387_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_6387_end_mask_0 = const()[name = string("op_6387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6387_cast_fp16 = slice_by_index(begin = var_6387_begin_0, end = var_6387_end_0, end_mask = var_6387_end_mask_0, x = k_13_cast_fp16)[name = string("op_6387_cast_fp16")]; + tensor var_6391_begin_0 = const()[name = string("op_6391_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6391_end_0 = const()[name = string("op_6391_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_6391_end_mask_0 = const()[name = string("op_6391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6391_cast_fp16 = slice_by_index(begin = var_6391_begin_0, end = var_6391_end_0, end_mask = var_6391_end_mask_0, x = k_13_cast_fp16)[name = string("op_6391_cast_fp16")]; + tensor var_6395_begin_0 = const()[name = string("op_6395_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6395_end_0 = const()[name = string("op_6395_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_6395_end_mask_0 = const()[name = string("op_6395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6395_cast_fp16 = slice_by_index(begin = var_6395_begin_0, end = var_6395_end_0, end_mask = var_6395_end_mask_0, x = k_13_cast_fp16)[name = string("op_6395_cast_fp16")]; + tensor var_6399_begin_0 = const()[name = string("op_6399_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6399_end_0 = const()[name = string("op_6399_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_6399_end_mask_0 = const()[name = string("op_6399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6399_cast_fp16 = slice_by_index(begin = var_6399_begin_0, end = var_6399_end_0, end_mask = var_6399_end_mask_0, x = k_13_cast_fp16)[name = string("op_6399_cast_fp16")]; + tensor var_6403_begin_0 = const()[name = string("op_6403_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6403_end_0 = const()[name = string("op_6403_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_6403_end_mask_0 = const()[name = string("op_6403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6403_cast_fp16 = slice_by_index(begin = var_6403_begin_0, end = var_6403_end_0, end_mask = var_6403_end_mask_0, x = k_13_cast_fp16)[name = string("op_6403_cast_fp16")]; + tensor var_6407_begin_0 = const()[name = string("op_6407_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6407_end_0 = const()[name = string("op_6407_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_6407_end_mask_0 = const()[name = string("op_6407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6407_cast_fp16 = slice_by_index(begin = var_6407_begin_0, end = var_6407_end_0, end_mask = var_6407_end_mask_0, x = k_13_cast_fp16)[name = string("op_6407_cast_fp16")]; + tensor var_6411_begin_0 = const()[name = string("op_6411_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6411_end_0 = const()[name = string("op_6411_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_6411_end_mask_0 = const()[name = string("op_6411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6411_cast_fp16 = slice_by_index(begin = var_6411_begin_0, end = var_6411_end_0, end_mask = var_6411_end_mask_0, x = k_13_cast_fp16)[name = string("op_6411_cast_fp16")]; + tensor var_6415_begin_0 = const()[name = string("op_6415_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6415_end_0 = const()[name = string("op_6415_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_6415_end_mask_0 = const()[name = string("op_6415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6415_cast_fp16 = slice_by_index(begin = var_6415_begin_0, end = var_6415_end_0, end_mask = var_6415_end_mask_0, x = k_13_cast_fp16)[name = string("op_6415_cast_fp16")]; + tensor var_6417_begin_0 = const()[name = string("op_6417_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6417_end_0 = const()[name = string("op_6417_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6417_end_mask_0 = const()[name = string("op_6417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6417_cast_fp16 = slice_by_index(begin = var_6417_begin_0, end = var_6417_end_0, end_mask = var_6417_end_mask_0, x = value_13_cast_fp16)[name = string("op_6417_cast_fp16")]; + tensor var_6421_begin_0 = const()[name = string("op_6421_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6421_end_0 = const()[name = string("op_6421_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6421_end_mask_0 = const()[name = string("op_6421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6421_cast_fp16 = slice_by_index(begin = var_6421_begin_0, end = var_6421_end_0, end_mask = var_6421_end_mask_0, x = value_13_cast_fp16)[name = string("op_6421_cast_fp16")]; + tensor var_6425_begin_0 = const()[name = string("op_6425_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6425_end_0 = const()[name = string("op_6425_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6425_end_mask_0 = const()[name = string("op_6425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6425_cast_fp16 = slice_by_index(begin = var_6425_begin_0, end = var_6425_end_0, end_mask = var_6425_end_mask_0, x = value_13_cast_fp16)[name = string("op_6425_cast_fp16")]; + tensor var_6429_begin_0 = const()[name = string("op_6429_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6429_end_0 = const()[name = string("op_6429_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6429_end_mask_0 = const()[name = string("op_6429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6429_cast_fp16 = slice_by_index(begin = var_6429_begin_0, end = var_6429_end_0, end_mask = var_6429_end_mask_0, x = value_13_cast_fp16)[name = string("op_6429_cast_fp16")]; + tensor var_6433_begin_0 = const()[name = string("op_6433_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6433_end_0 = const()[name = string("op_6433_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6433_end_mask_0 = const()[name = string("op_6433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6433_cast_fp16 = slice_by_index(begin = var_6433_begin_0, end = var_6433_end_0, end_mask = var_6433_end_mask_0, x = value_13_cast_fp16)[name = string("op_6433_cast_fp16")]; + tensor var_6437_begin_0 = const()[name = string("op_6437_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6437_end_0 = const()[name = string("op_6437_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6437_end_mask_0 = const()[name = string("op_6437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6437_cast_fp16 = slice_by_index(begin = var_6437_begin_0, end = var_6437_end_0, end_mask = var_6437_end_mask_0, x = value_13_cast_fp16)[name = string("op_6437_cast_fp16")]; + tensor var_6441_begin_0 = const()[name = string("op_6441_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6441_end_0 = const()[name = string("op_6441_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6441_end_mask_0 = const()[name = string("op_6441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6441_cast_fp16 = slice_by_index(begin = var_6441_begin_0, end = var_6441_end_0, end_mask = var_6441_end_mask_0, x = value_13_cast_fp16)[name = string("op_6441_cast_fp16")]; + tensor var_6445_begin_0 = const()[name = string("op_6445_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6445_end_0 = const()[name = string("op_6445_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6445_end_mask_0 = const()[name = string("op_6445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6445_cast_fp16 = slice_by_index(begin = var_6445_begin_0, end = var_6445_end_0, end_mask = var_6445_end_mask_0, x = value_13_cast_fp16)[name = string("op_6445_cast_fp16")]; + tensor var_6449_begin_0 = const()[name = string("op_6449_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6449_end_0 = const()[name = string("op_6449_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6449_end_mask_0 = const()[name = string("op_6449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6449_cast_fp16 = slice_by_index(begin = var_6449_begin_0, end = var_6449_end_0, end_mask = var_6449_end_mask_0, x = value_13_cast_fp16)[name = string("op_6449_cast_fp16")]; + tensor var_6453_begin_0 = const()[name = string("op_6453_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6453_end_0 = const()[name = string("op_6453_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6453_end_mask_0 = const()[name = string("op_6453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6453_cast_fp16 = slice_by_index(begin = var_6453_begin_0, end = var_6453_end_0, end_mask = var_6453_end_mask_0, x = value_13_cast_fp16)[name = string("op_6453_cast_fp16")]; + tensor var_6457_begin_0 = const()[name = string("op_6457_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6457_end_0 = const()[name = string("op_6457_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6457_end_mask_0 = const()[name = string("op_6457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6457_cast_fp16 = slice_by_index(begin = var_6457_begin_0, end = var_6457_end_0, end_mask = var_6457_end_mask_0, x = value_13_cast_fp16)[name = string("op_6457_cast_fp16")]; + tensor var_6461_begin_0 = const()[name = string("op_6461_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6461_end_0 = const()[name = string("op_6461_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6461_end_mask_0 = const()[name = string("op_6461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6461_cast_fp16 = slice_by_index(begin = var_6461_begin_0, end = var_6461_end_0, end_mask = var_6461_end_mask_0, x = value_13_cast_fp16)[name = string("op_6461_cast_fp16")]; + string _SplitHeadsQ__mh_w_577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_577_equation_0, values = (var_6371_cast_fp16, var_6037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_577_cast_fp16")]; + string _SplitHeadsQ__mh_w_579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_579_equation_0, values = (var_6371_cast_fp16, var_6044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_579_cast_fp16")]; + string _SplitHeadsQ__mh_w_581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_581_equation_0, values = (var_6371_cast_fp16, var_6051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_581_cast_fp16")]; + string _SplitHeadsQ__mh_w_583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_583_equation_0, values = (var_6371_cast_fp16, var_6058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_583_cast_fp16")]; + string _SplitHeadsQ__mh_w_585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_585_equation_0, values = (var_6375_cast_fp16, var_6065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_585_cast_fp16")]; + string _SplitHeadsQ__mh_w_587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_587_equation_0, values = (var_6375_cast_fp16, var_6072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_587_cast_fp16")]; + string _SplitHeadsQ__mh_w_589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_589_equation_0, values = (var_6375_cast_fp16, var_6079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_589_cast_fp16")]; + string _SplitHeadsQ__mh_w_591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_591_equation_0, values = (var_6375_cast_fp16, var_6086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_591_cast_fp16")]; + string _SplitHeadsQ__mh_w_593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_593_equation_0, values = (var_6379_cast_fp16, var_6093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_593_cast_fp16")]; + string _SplitHeadsQ__mh_w_595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_595_equation_0, values = (var_6379_cast_fp16, var_6100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_595_cast_fp16")]; + string _SplitHeadsQ__mh_w_597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_597_equation_0, values = (var_6379_cast_fp16, var_6107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_597_cast_fp16")]; + string _SplitHeadsQ__mh_w_599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_599_equation_0, values = (var_6379_cast_fp16, var_6114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_599_cast_fp16")]; + string _SplitHeadsQ__mh_w_601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_601_equation_0, values = (var_6383_cast_fp16, var_6121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_601_cast_fp16")]; + string _SplitHeadsQ__mh_w_603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_603_equation_0, values = (var_6383_cast_fp16, var_6128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_603_cast_fp16")]; + string _SplitHeadsQ__mh_w_605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_605_equation_0, values = (var_6383_cast_fp16, var_6135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_605_cast_fp16")]; + string _SplitHeadsQ__mh_w_607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_607_equation_0, values = (var_6383_cast_fp16, var_6142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_607_cast_fp16")]; + string _SplitHeadsQ__mh_w_609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_609_equation_0, values = (var_6387_cast_fp16, var_6149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_609_cast_fp16")]; + string _SplitHeadsQ__mh_w_611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_611_equation_0, values = (var_6387_cast_fp16, var_6156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_611_cast_fp16")]; + string _SplitHeadsQ__mh_w_613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_613_equation_0, values = (var_6387_cast_fp16, var_6163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_613_cast_fp16")]; + string _SplitHeadsQ__mh_w_615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_615_equation_0, values = (var_6387_cast_fp16, var_6170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_615_cast_fp16")]; + string _SplitHeadsQ__mh_w_617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_617_equation_0, values = (var_6391_cast_fp16, var_6177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_617_cast_fp16")]; + string _SplitHeadsQ__mh_w_619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_619_equation_0, values = (var_6391_cast_fp16, var_6184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_619_cast_fp16")]; + string _SplitHeadsQ__mh_w_621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_621_equation_0, values = (var_6391_cast_fp16, var_6191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_621_cast_fp16")]; + string _SplitHeadsQ__mh_w_623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_623_equation_0, values = (var_6391_cast_fp16, var_6198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_623_cast_fp16")]; + string _SplitHeadsQ__mh_w_625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_625_equation_0, values = (var_6395_cast_fp16, var_6205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_625_cast_fp16")]; + string _SplitHeadsQ__mh_w_627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_627_equation_0, values = (var_6395_cast_fp16, var_6212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_627_cast_fp16")]; + string _SplitHeadsQ__mh_w_629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_629_equation_0, values = (var_6395_cast_fp16, var_6219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_629_cast_fp16")]; + string _SplitHeadsQ__mh_w_631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_631_equation_0, values = (var_6395_cast_fp16, var_6226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_631_cast_fp16")]; + string _SplitHeadsQ__mh_w_633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_633_equation_0, values = (var_6399_cast_fp16, var_6233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_633_cast_fp16")]; + string _SplitHeadsQ__mh_w_635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_635_equation_0, values = (var_6399_cast_fp16, var_6240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_635_cast_fp16")]; + string _SplitHeadsQ__mh_w_637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_637_equation_0, values = (var_6399_cast_fp16, var_6247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_637_cast_fp16")]; + string _SplitHeadsQ__mh_w_639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_639_equation_0, values = (var_6399_cast_fp16, var_6254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_639_cast_fp16")]; + string _SplitHeadsQ__mh_w_641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_641_equation_0, values = (var_6403_cast_fp16, var_6261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_641_cast_fp16")]; + string _SplitHeadsQ__mh_w_643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_643_equation_0, values = (var_6403_cast_fp16, var_6268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_643_cast_fp16")]; + string _SplitHeadsQ__mh_w_645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_645_equation_0, values = (var_6403_cast_fp16, var_6275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_645_cast_fp16")]; + string _SplitHeadsQ__mh_w_647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_647_equation_0, values = (var_6403_cast_fp16, var_6282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_647_cast_fp16")]; + string _SplitHeadsQ__mh_w_649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_649_equation_0, values = (var_6407_cast_fp16, var_6289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_649_cast_fp16")]; + string _SplitHeadsQ__mh_w_651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_651_equation_0, values = (var_6407_cast_fp16, var_6296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_651_cast_fp16")]; + string _SplitHeadsQ__mh_w_653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_653_equation_0, values = (var_6407_cast_fp16, var_6303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_653_cast_fp16")]; + string _SplitHeadsQ__mh_w_655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_655_equation_0, values = (var_6407_cast_fp16, var_6310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_655_cast_fp16")]; + string _SplitHeadsQ__mh_w_657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_657_equation_0, values = (var_6411_cast_fp16, var_6317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_657_cast_fp16")]; + string _SplitHeadsQ__mh_w_659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_659_equation_0, values = (var_6411_cast_fp16, var_6324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_659_cast_fp16")]; + string _SplitHeadsQ__mh_w_661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_661_equation_0, values = (var_6411_cast_fp16, var_6331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_661_cast_fp16")]; + string _SplitHeadsQ__mh_w_663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_663_equation_0, values = (var_6411_cast_fp16, var_6338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_663_cast_fp16")]; + string _SplitHeadsQ__mh_w_665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_665_equation_0, values = (var_6415_cast_fp16, var_6345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_665_cast_fp16")]; + string _SplitHeadsQ__mh_w_667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_667_equation_0, values = (var_6415_cast_fp16, var_6352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_667_cast_fp16")]; + string _SplitHeadsQ__mh_w_669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_669_equation_0, values = (var_6415_cast_fp16, var_6359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_669_cast_fp16")]; + string _SplitHeadsQ__mh_w_671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_671_equation_0, values = (var_6415_cast_fp16, var_6366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_671_cast_fp16")]; + fp16 var_6560_to_fp16 = const()[name = string("op_6560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_577_cast_fp16, y = var_6560_to_fp16)[name = string("aw_chunk_577_cast_fp16")]; + fp16 var_6562_to_fp16 = const()[name = string("op_6562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_579_cast_fp16, y = var_6562_to_fp16)[name = string("aw_chunk_579_cast_fp16")]; + fp16 var_6564_to_fp16 = const()[name = string("op_6564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_581_cast_fp16, y = var_6564_to_fp16)[name = string("aw_chunk_581_cast_fp16")]; + fp16 var_6566_to_fp16 = const()[name = string("op_6566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_583_cast_fp16, y = var_6566_to_fp16)[name = string("aw_chunk_583_cast_fp16")]; + fp16 var_6568_to_fp16 = const()[name = string("op_6568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_585_cast_fp16, y = var_6568_to_fp16)[name = string("aw_chunk_585_cast_fp16")]; + fp16 var_6570_to_fp16 = const()[name = string("op_6570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_587_cast_fp16, y = var_6570_to_fp16)[name = string("aw_chunk_587_cast_fp16")]; + fp16 var_6572_to_fp16 = const()[name = string("op_6572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_589_cast_fp16, y = var_6572_to_fp16)[name = string("aw_chunk_589_cast_fp16")]; + fp16 var_6574_to_fp16 = const()[name = string("op_6574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_591_cast_fp16, y = var_6574_to_fp16)[name = string("aw_chunk_591_cast_fp16")]; + fp16 var_6576_to_fp16 = const()[name = string("op_6576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_593_cast_fp16, y = var_6576_to_fp16)[name = string("aw_chunk_593_cast_fp16")]; + fp16 var_6578_to_fp16 = const()[name = string("op_6578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_595_cast_fp16, y = var_6578_to_fp16)[name = string("aw_chunk_595_cast_fp16")]; + fp16 var_6580_to_fp16 = const()[name = string("op_6580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_597_cast_fp16, y = var_6580_to_fp16)[name = string("aw_chunk_597_cast_fp16")]; + fp16 var_6582_to_fp16 = const()[name = string("op_6582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_599_cast_fp16, y = var_6582_to_fp16)[name = string("aw_chunk_599_cast_fp16")]; + fp16 var_6584_to_fp16 = const()[name = string("op_6584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_601_cast_fp16, y = var_6584_to_fp16)[name = string("aw_chunk_601_cast_fp16")]; + fp16 var_6586_to_fp16 = const()[name = string("op_6586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_603_cast_fp16, y = var_6586_to_fp16)[name = string("aw_chunk_603_cast_fp16")]; + fp16 var_6588_to_fp16 = const()[name = string("op_6588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_605_cast_fp16, y = var_6588_to_fp16)[name = string("aw_chunk_605_cast_fp16")]; + fp16 var_6590_to_fp16 = const()[name = string("op_6590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_607_cast_fp16, y = var_6590_to_fp16)[name = string("aw_chunk_607_cast_fp16")]; + fp16 var_6592_to_fp16 = const()[name = string("op_6592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_609_cast_fp16, y = var_6592_to_fp16)[name = string("aw_chunk_609_cast_fp16")]; + fp16 var_6594_to_fp16 = const()[name = string("op_6594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_611_cast_fp16, y = var_6594_to_fp16)[name = string("aw_chunk_611_cast_fp16")]; + fp16 var_6596_to_fp16 = const()[name = string("op_6596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_613_cast_fp16, y = var_6596_to_fp16)[name = string("aw_chunk_613_cast_fp16")]; + fp16 var_6598_to_fp16 = const()[name = string("op_6598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_615_cast_fp16, y = var_6598_to_fp16)[name = string("aw_chunk_615_cast_fp16")]; + fp16 var_6600_to_fp16 = const()[name = string("op_6600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_617_cast_fp16, y = var_6600_to_fp16)[name = string("aw_chunk_617_cast_fp16")]; + fp16 var_6602_to_fp16 = const()[name = string("op_6602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_619_cast_fp16, y = var_6602_to_fp16)[name = string("aw_chunk_619_cast_fp16")]; + fp16 var_6604_to_fp16 = const()[name = string("op_6604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_621_cast_fp16, y = var_6604_to_fp16)[name = string("aw_chunk_621_cast_fp16")]; + fp16 var_6606_to_fp16 = const()[name = string("op_6606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_623_cast_fp16, y = var_6606_to_fp16)[name = string("aw_chunk_623_cast_fp16")]; + fp16 var_6608_to_fp16 = const()[name = string("op_6608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_625_cast_fp16, y = var_6608_to_fp16)[name = string("aw_chunk_625_cast_fp16")]; + fp16 var_6610_to_fp16 = const()[name = string("op_6610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_627_cast_fp16, y = var_6610_to_fp16)[name = string("aw_chunk_627_cast_fp16")]; + fp16 var_6612_to_fp16 = const()[name = string("op_6612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_629_cast_fp16, y = var_6612_to_fp16)[name = string("aw_chunk_629_cast_fp16")]; + fp16 var_6614_to_fp16 = const()[name = string("op_6614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_631_cast_fp16, y = var_6614_to_fp16)[name = string("aw_chunk_631_cast_fp16")]; + fp16 var_6616_to_fp16 = const()[name = string("op_6616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_633_cast_fp16, y = var_6616_to_fp16)[name = string("aw_chunk_633_cast_fp16")]; + fp16 var_6618_to_fp16 = const()[name = string("op_6618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_635_cast_fp16, y = var_6618_to_fp16)[name = string("aw_chunk_635_cast_fp16")]; + fp16 var_6620_to_fp16 = const()[name = string("op_6620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_637_cast_fp16, y = var_6620_to_fp16)[name = string("aw_chunk_637_cast_fp16")]; + fp16 var_6622_to_fp16 = const()[name = string("op_6622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_639_cast_fp16, y = var_6622_to_fp16)[name = string("aw_chunk_639_cast_fp16")]; + fp16 var_6624_to_fp16 = const()[name = string("op_6624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_641_cast_fp16, y = var_6624_to_fp16)[name = string("aw_chunk_641_cast_fp16")]; + fp16 var_6626_to_fp16 = const()[name = string("op_6626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_643_cast_fp16, y = var_6626_to_fp16)[name = string("aw_chunk_643_cast_fp16")]; + fp16 var_6628_to_fp16 = const()[name = string("op_6628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_645_cast_fp16, y = var_6628_to_fp16)[name = string("aw_chunk_645_cast_fp16")]; + fp16 var_6630_to_fp16 = const()[name = string("op_6630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_647_cast_fp16, y = var_6630_to_fp16)[name = string("aw_chunk_647_cast_fp16")]; + fp16 var_6632_to_fp16 = const()[name = string("op_6632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_649_cast_fp16, y = var_6632_to_fp16)[name = string("aw_chunk_649_cast_fp16")]; + fp16 var_6634_to_fp16 = const()[name = string("op_6634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_651_cast_fp16, y = var_6634_to_fp16)[name = string("aw_chunk_651_cast_fp16")]; + fp16 var_6636_to_fp16 = const()[name = string("op_6636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_653_cast_fp16, y = var_6636_to_fp16)[name = string("aw_chunk_653_cast_fp16")]; + fp16 var_6638_to_fp16 = const()[name = string("op_6638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_655_cast_fp16, y = var_6638_to_fp16)[name = string("aw_chunk_655_cast_fp16")]; + fp16 var_6640_to_fp16 = const()[name = string("op_6640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_657_cast_fp16, y = var_6640_to_fp16)[name = string("aw_chunk_657_cast_fp16")]; + fp16 var_6642_to_fp16 = const()[name = string("op_6642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_659_cast_fp16, y = var_6642_to_fp16)[name = string("aw_chunk_659_cast_fp16")]; + fp16 var_6644_to_fp16 = const()[name = string("op_6644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_661_cast_fp16, y = var_6644_to_fp16)[name = string("aw_chunk_661_cast_fp16")]; + fp16 var_6646_to_fp16 = const()[name = string("op_6646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_663_cast_fp16, y = var_6646_to_fp16)[name = string("aw_chunk_663_cast_fp16")]; + fp16 var_6648_to_fp16 = const()[name = string("op_6648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_665_cast_fp16, y = var_6648_to_fp16)[name = string("aw_chunk_665_cast_fp16")]; + fp16 var_6650_to_fp16 = const()[name = string("op_6650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_667_cast_fp16, y = var_6650_to_fp16)[name = string("aw_chunk_667_cast_fp16")]; + fp16 var_6652_to_fp16 = const()[name = string("op_6652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_669_cast_fp16, y = var_6652_to_fp16)[name = string("aw_chunk_669_cast_fp16")]; + fp16 var_6654_to_fp16 = const()[name = string("op_6654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_671_cast_fp16, y = var_6654_to_fp16)[name = string("aw_chunk_671_cast_fp16")]; + tensor var_6656_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_577_cast_fp16)[name = string("op_6656_cast_fp16")]; + tensor var_6657_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_579_cast_fp16)[name = string("op_6657_cast_fp16")]; + tensor var_6658_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_581_cast_fp16)[name = string("op_6658_cast_fp16")]; + tensor var_6659_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_583_cast_fp16)[name = string("op_6659_cast_fp16")]; + tensor var_6660_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_585_cast_fp16)[name = string("op_6660_cast_fp16")]; + tensor var_6661_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_587_cast_fp16)[name = string("op_6661_cast_fp16")]; + tensor var_6662_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_589_cast_fp16)[name = string("op_6662_cast_fp16")]; + tensor var_6663_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_591_cast_fp16)[name = string("op_6663_cast_fp16")]; + tensor var_6664_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_593_cast_fp16)[name = string("op_6664_cast_fp16")]; + tensor var_6665_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_595_cast_fp16)[name = string("op_6665_cast_fp16")]; + tensor var_6666_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_597_cast_fp16)[name = string("op_6666_cast_fp16")]; + tensor var_6667_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_599_cast_fp16)[name = string("op_6667_cast_fp16")]; + tensor var_6668_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_601_cast_fp16)[name = string("op_6668_cast_fp16")]; + tensor var_6669_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_603_cast_fp16)[name = string("op_6669_cast_fp16")]; + tensor var_6670_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_605_cast_fp16)[name = string("op_6670_cast_fp16")]; + tensor var_6671_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_607_cast_fp16)[name = string("op_6671_cast_fp16")]; + tensor var_6672_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_609_cast_fp16)[name = string("op_6672_cast_fp16")]; + tensor var_6673_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_611_cast_fp16)[name = string("op_6673_cast_fp16")]; + tensor var_6674_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_613_cast_fp16)[name = string("op_6674_cast_fp16")]; + tensor var_6675_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_615_cast_fp16)[name = string("op_6675_cast_fp16")]; + tensor var_6676_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_617_cast_fp16)[name = string("op_6676_cast_fp16")]; + tensor var_6677_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_619_cast_fp16)[name = string("op_6677_cast_fp16")]; + tensor var_6678_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_621_cast_fp16)[name = string("op_6678_cast_fp16")]; + tensor var_6679_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_623_cast_fp16)[name = string("op_6679_cast_fp16")]; + tensor var_6680_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_625_cast_fp16)[name = string("op_6680_cast_fp16")]; + tensor var_6681_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_627_cast_fp16)[name = string("op_6681_cast_fp16")]; + tensor var_6682_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_629_cast_fp16)[name = string("op_6682_cast_fp16")]; + tensor var_6683_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_631_cast_fp16)[name = string("op_6683_cast_fp16")]; + tensor var_6684_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_633_cast_fp16)[name = string("op_6684_cast_fp16")]; + tensor var_6685_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_635_cast_fp16)[name = string("op_6685_cast_fp16")]; + tensor var_6686_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_637_cast_fp16)[name = string("op_6686_cast_fp16")]; + tensor var_6687_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_639_cast_fp16)[name = string("op_6687_cast_fp16")]; + tensor var_6688_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_641_cast_fp16)[name = string("op_6688_cast_fp16")]; + tensor var_6689_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_643_cast_fp16)[name = string("op_6689_cast_fp16")]; + tensor var_6690_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_645_cast_fp16)[name = string("op_6690_cast_fp16")]; + tensor var_6691_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_647_cast_fp16)[name = string("op_6691_cast_fp16")]; + tensor var_6692_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_649_cast_fp16)[name = string("op_6692_cast_fp16")]; + tensor var_6693_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_651_cast_fp16)[name = string("op_6693_cast_fp16")]; + tensor var_6694_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_653_cast_fp16)[name = string("op_6694_cast_fp16")]; + tensor var_6695_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_655_cast_fp16)[name = string("op_6695_cast_fp16")]; + tensor var_6696_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_657_cast_fp16)[name = string("op_6696_cast_fp16")]; + tensor var_6697_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_659_cast_fp16)[name = string("op_6697_cast_fp16")]; + tensor var_6698_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_661_cast_fp16)[name = string("op_6698_cast_fp16")]; + tensor var_6699_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_663_cast_fp16)[name = string("op_6699_cast_fp16")]; + tensor var_6700_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_665_cast_fp16)[name = string("op_6700_cast_fp16")]; + tensor var_6701_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_667_cast_fp16)[name = string("op_6701_cast_fp16")]; + tensor var_6702_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_669_cast_fp16)[name = string("op_6702_cast_fp16")]; + tensor var_6703_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_671_cast_fp16)[name = string("op_6703_cast_fp16")]; + string var_6705_equation_0 = const()[name = string("op_6705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6705_cast_fp16 = einsum(equation = var_6705_equation_0, values = (var_6417_cast_fp16, var_6656_cast_fp16))[name = string("op_6705_cast_fp16")]; + string var_6707_equation_0 = const()[name = string("op_6707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6707_cast_fp16 = einsum(equation = var_6707_equation_0, values = (var_6417_cast_fp16, var_6657_cast_fp16))[name = string("op_6707_cast_fp16")]; + string var_6709_equation_0 = const()[name = string("op_6709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6709_cast_fp16 = einsum(equation = var_6709_equation_0, values = (var_6417_cast_fp16, var_6658_cast_fp16))[name = string("op_6709_cast_fp16")]; + string var_6711_equation_0 = const()[name = string("op_6711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6711_cast_fp16 = einsum(equation = var_6711_equation_0, values = (var_6417_cast_fp16, var_6659_cast_fp16))[name = string("op_6711_cast_fp16")]; + string var_6713_equation_0 = const()[name = string("op_6713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6713_cast_fp16 = einsum(equation = var_6713_equation_0, values = (var_6421_cast_fp16, var_6660_cast_fp16))[name = string("op_6713_cast_fp16")]; + string var_6715_equation_0 = const()[name = string("op_6715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6715_cast_fp16 = einsum(equation = var_6715_equation_0, values = (var_6421_cast_fp16, var_6661_cast_fp16))[name = string("op_6715_cast_fp16")]; + string var_6717_equation_0 = const()[name = string("op_6717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6717_cast_fp16 = einsum(equation = var_6717_equation_0, values = (var_6421_cast_fp16, var_6662_cast_fp16))[name = string("op_6717_cast_fp16")]; + string var_6719_equation_0 = const()[name = string("op_6719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6719_cast_fp16 = einsum(equation = var_6719_equation_0, values = (var_6421_cast_fp16, var_6663_cast_fp16))[name = string("op_6719_cast_fp16")]; + string var_6721_equation_0 = const()[name = string("op_6721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6721_cast_fp16 = einsum(equation = var_6721_equation_0, values = (var_6425_cast_fp16, var_6664_cast_fp16))[name = string("op_6721_cast_fp16")]; + string var_6723_equation_0 = const()[name = string("op_6723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6723_cast_fp16 = einsum(equation = var_6723_equation_0, values = (var_6425_cast_fp16, var_6665_cast_fp16))[name = string("op_6723_cast_fp16")]; + string var_6725_equation_0 = const()[name = string("op_6725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6725_cast_fp16 = einsum(equation = var_6725_equation_0, values = (var_6425_cast_fp16, var_6666_cast_fp16))[name = string("op_6725_cast_fp16")]; + string var_6727_equation_0 = const()[name = string("op_6727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6727_cast_fp16 = einsum(equation = var_6727_equation_0, values = (var_6425_cast_fp16, var_6667_cast_fp16))[name = string("op_6727_cast_fp16")]; + string var_6729_equation_0 = const()[name = string("op_6729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6729_cast_fp16 = einsum(equation = var_6729_equation_0, values = (var_6429_cast_fp16, var_6668_cast_fp16))[name = string("op_6729_cast_fp16")]; + string var_6731_equation_0 = const()[name = string("op_6731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6731_cast_fp16 = einsum(equation = var_6731_equation_0, values = (var_6429_cast_fp16, var_6669_cast_fp16))[name = string("op_6731_cast_fp16")]; + string var_6733_equation_0 = const()[name = string("op_6733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6733_cast_fp16 = einsum(equation = var_6733_equation_0, values = (var_6429_cast_fp16, var_6670_cast_fp16))[name = string("op_6733_cast_fp16")]; + string var_6735_equation_0 = const()[name = string("op_6735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6735_cast_fp16 = einsum(equation = var_6735_equation_0, values = (var_6429_cast_fp16, var_6671_cast_fp16))[name = string("op_6735_cast_fp16")]; + string var_6737_equation_0 = const()[name = string("op_6737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6737_cast_fp16 = einsum(equation = var_6737_equation_0, values = (var_6433_cast_fp16, var_6672_cast_fp16))[name = string("op_6737_cast_fp16")]; + string var_6739_equation_0 = const()[name = string("op_6739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6739_cast_fp16 = einsum(equation = var_6739_equation_0, values = (var_6433_cast_fp16, var_6673_cast_fp16))[name = string("op_6739_cast_fp16")]; + string var_6741_equation_0 = const()[name = string("op_6741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6741_cast_fp16 = einsum(equation = var_6741_equation_0, values = (var_6433_cast_fp16, var_6674_cast_fp16))[name = string("op_6741_cast_fp16")]; + string var_6743_equation_0 = const()[name = string("op_6743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6743_cast_fp16 = einsum(equation = var_6743_equation_0, values = (var_6433_cast_fp16, var_6675_cast_fp16))[name = string("op_6743_cast_fp16")]; + string var_6745_equation_0 = const()[name = string("op_6745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6745_cast_fp16 = einsum(equation = var_6745_equation_0, values = (var_6437_cast_fp16, var_6676_cast_fp16))[name = string("op_6745_cast_fp16")]; + string var_6747_equation_0 = const()[name = string("op_6747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6747_cast_fp16 = einsum(equation = var_6747_equation_0, values = (var_6437_cast_fp16, var_6677_cast_fp16))[name = string("op_6747_cast_fp16")]; + string var_6749_equation_0 = const()[name = string("op_6749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6749_cast_fp16 = einsum(equation = var_6749_equation_0, values = (var_6437_cast_fp16, var_6678_cast_fp16))[name = string("op_6749_cast_fp16")]; + string var_6751_equation_0 = const()[name = string("op_6751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6751_cast_fp16 = einsum(equation = var_6751_equation_0, values = (var_6437_cast_fp16, var_6679_cast_fp16))[name = string("op_6751_cast_fp16")]; + string var_6753_equation_0 = const()[name = string("op_6753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6441_cast_fp16, var_6680_cast_fp16))[name = string("op_6753_cast_fp16")]; + string var_6755_equation_0 = const()[name = string("op_6755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6755_cast_fp16 = einsum(equation = var_6755_equation_0, values = (var_6441_cast_fp16, var_6681_cast_fp16))[name = string("op_6755_cast_fp16")]; + string var_6757_equation_0 = const()[name = string("op_6757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6441_cast_fp16, var_6682_cast_fp16))[name = string("op_6757_cast_fp16")]; + string var_6759_equation_0 = const()[name = string("op_6759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6759_cast_fp16 = einsum(equation = var_6759_equation_0, values = (var_6441_cast_fp16, var_6683_cast_fp16))[name = string("op_6759_cast_fp16")]; + string var_6761_equation_0 = const()[name = string("op_6761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6445_cast_fp16, var_6684_cast_fp16))[name = string("op_6761_cast_fp16")]; + string var_6763_equation_0 = const()[name = string("op_6763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6763_cast_fp16 = einsum(equation = var_6763_equation_0, values = (var_6445_cast_fp16, var_6685_cast_fp16))[name = string("op_6763_cast_fp16")]; + string var_6765_equation_0 = const()[name = string("op_6765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6445_cast_fp16, var_6686_cast_fp16))[name = string("op_6765_cast_fp16")]; + string var_6767_equation_0 = const()[name = string("op_6767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6767_cast_fp16 = einsum(equation = var_6767_equation_0, values = (var_6445_cast_fp16, var_6687_cast_fp16))[name = string("op_6767_cast_fp16")]; + string var_6769_equation_0 = const()[name = string("op_6769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6449_cast_fp16, var_6688_cast_fp16))[name = string("op_6769_cast_fp16")]; + string var_6771_equation_0 = const()[name = string("op_6771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6771_cast_fp16 = einsum(equation = var_6771_equation_0, values = (var_6449_cast_fp16, var_6689_cast_fp16))[name = string("op_6771_cast_fp16")]; + string var_6773_equation_0 = const()[name = string("op_6773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6449_cast_fp16, var_6690_cast_fp16))[name = string("op_6773_cast_fp16")]; + string var_6775_equation_0 = const()[name = string("op_6775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6775_cast_fp16 = einsum(equation = var_6775_equation_0, values = (var_6449_cast_fp16, var_6691_cast_fp16))[name = string("op_6775_cast_fp16")]; + string var_6777_equation_0 = const()[name = string("op_6777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6777_cast_fp16 = einsum(equation = var_6777_equation_0, values = (var_6453_cast_fp16, var_6692_cast_fp16))[name = string("op_6777_cast_fp16")]; + string var_6779_equation_0 = const()[name = string("op_6779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6779_cast_fp16 = einsum(equation = var_6779_equation_0, values = (var_6453_cast_fp16, var_6693_cast_fp16))[name = string("op_6779_cast_fp16")]; + string var_6781_equation_0 = const()[name = string("op_6781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6781_cast_fp16 = einsum(equation = var_6781_equation_0, values = (var_6453_cast_fp16, var_6694_cast_fp16))[name = string("op_6781_cast_fp16")]; + string var_6783_equation_0 = const()[name = string("op_6783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6783_cast_fp16 = einsum(equation = var_6783_equation_0, values = (var_6453_cast_fp16, var_6695_cast_fp16))[name = string("op_6783_cast_fp16")]; + string var_6785_equation_0 = const()[name = string("op_6785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6785_cast_fp16 = einsum(equation = var_6785_equation_0, values = (var_6457_cast_fp16, var_6696_cast_fp16))[name = string("op_6785_cast_fp16")]; + string var_6787_equation_0 = const()[name = string("op_6787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6787_cast_fp16 = einsum(equation = var_6787_equation_0, values = (var_6457_cast_fp16, var_6697_cast_fp16))[name = string("op_6787_cast_fp16")]; + string var_6789_equation_0 = const()[name = string("op_6789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6789_cast_fp16 = einsum(equation = var_6789_equation_0, values = (var_6457_cast_fp16, var_6698_cast_fp16))[name = string("op_6789_cast_fp16")]; + string var_6791_equation_0 = const()[name = string("op_6791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6791_cast_fp16 = einsum(equation = var_6791_equation_0, values = (var_6457_cast_fp16, var_6699_cast_fp16))[name = string("op_6791_cast_fp16")]; + string var_6793_equation_0 = const()[name = string("op_6793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6793_cast_fp16 = einsum(equation = var_6793_equation_0, values = (var_6461_cast_fp16, var_6700_cast_fp16))[name = string("op_6793_cast_fp16")]; + string var_6795_equation_0 = const()[name = string("op_6795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6795_cast_fp16 = einsum(equation = var_6795_equation_0, values = (var_6461_cast_fp16, var_6701_cast_fp16))[name = string("op_6795_cast_fp16")]; + string var_6797_equation_0 = const()[name = string("op_6797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6797_cast_fp16 = einsum(equation = var_6797_equation_0, values = (var_6461_cast_fp16, var_6702_cast_fp16))[name = string("op_6797_cast_fp16")]; + string var_6799_equation_0 = const()[name = string("op_6799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6799_cast_fp16 = einsum(equation = var_6799_equation_0, values = (var_6461_cast_fp16, var_6703_cast_fp16))[name = string("op_6799_cast_fp16")]; + bool var_6801_interleave_0 = const()[name = string("op_6801_interleave_0"), val = bool(false)]; + tensor var_6801_cast_fp16 = concat(axis = var_5912, interleave = var_6801_interleave_0, values = (var_6705_cast_fp16, var_6707_cast_fp16, var_6709_cast_fp16, var_6711_cast_fp16))[name = string("op_6801_cast_fp16")]; + bool var_6803_interleave_0 = const()[name = string("op_6803_interleave_0"), val = bool(false)]; + tensor var_6803_cast_fp16 = concat(axis = var_5912, interleave = var_6803_interleave_0, values = (var_6713_cast_fp16, var_6715_cast_fp16, var_6717_cast_fp16, var_6719_cast_fp16))[name = string("op_6803_cast_fp16")]; + bool var_6805_interleave_0 = const()[name = string("op_6805_interleave_0"), val = bool(false)]; + tensor var_6805_cast_fp16 = concat(axis = var_5912, interleave = var_6805_interleave_0, values = (var_6721_cast_fp16, var_6723_cast_fp16, var_6725_cast_fp16, var_6727_cast_fp16))[name = string("op_6805_cast_fp16")]; + bool var_6807_interleave_0 = const()[name = string("op_6807_interleave_0"), val = bool(false)]; + tensor var_6807_cast_fp16 = concat(axis = var_5912, interleave = var_6807_interleave_0, values = (var_6729_cast_fp16, var_6731_cast_fp16, var_6733_cast_fp16, var_6735_cast_fp16))[name = string("op_6807_cast_fp16")]; + bool var_6809_interleave_0 = const()[name = string("op_6809_interleave_0"), val = bool(false)]; + tensor var_6809_cast_fp16 = concat(axis = var_5912, interleave = var_6809_interleave_0, values = (var_6737_cast_fp16, var_6739_cast_fp16, var_6741_cast_fp16, var_6743_cast_fp16))[name = string("op_6809_cast_fp16")]; + bool var_6811_interleave_0 = const()[name = string("op_6811_interleave_0"), val = bool(false)]; + tensor var_6811_cast_fp16 = concat(axis = var_5912, interleave = var_6811_interleave_0, values = (var_6745_cast_fp16, var_6747_cast_fp16, var_6749_cast_fp16, var_6751_cast_fp16))[name = string("op_6811_cast_fp16")]; + bool var_6813_interleave_0 = const()[name = string("op_6813_interleave_0"), val = bool(false)]; + tensor var_6813_cast_fp16 = concat(axis = var_5912, interleave = var_6813_interleave_0, values = (var_6753_cast_fp16, var_6755_cast_fp16, var_6757_cast_fp16, var_6759_cast_fp16))[name = string("op_6813_cast_fp16")]; + bool var_6815_interleave_0 = const()[name = string("op_6815_interleave_0"), val = bool(false)]; + tensor var_6815_cast_fp16 = concat(axis = var_5912, interleave = var_6815_interleave_0, values = (var_6761_cast_fp16, var_6763_cast_fp16, var_6765_cast_fp16, var_6767_cast_fp16))[name = string("op_6815_cast_fp16")]; + bool var_6817_interleave_0 = const()[name = string("op_6817_interleave_0"), val = bool(false)]; + tensor var_6817_cast_fp16 = concat(axis = var_5912, interleave = var_6817_interleave_0, values = (var_6769_cast_fp16, var_6771_cast_fp16, var_6773_cast_fp16, var_6775_cast_fp16))[name = string("op_6817_cast_fp16")]; + bool var_6819_interleave_0 = const()[name = string("op_6819_interleave_0"), val = bool(false)]; + tensor var_6819_cast_fp16 = concat(axis = var_5912, interleave = var_6819_interleave_0, values = (var_6777_cast_fp16, var_6779_cast_fp16, var_6781_cast_fp16, var_6783_cast_fp16))[name = string("op_6819_cast_fp16")]; + bool var_6821_interleave_0 = const()[name = string("op_6821_interleave_0"), val = bool(false)]; + tensor var_6821_cast_fp16 = concat(axis = var_5912, interleave = var_6821_interleave_0, values = (var_6785_cast_fp16, var_6787_cast_fp16, var_6789_cast_fp16, var_6791_cast_fp16))[name = string("op_6821_cast_fp16")]; + bool var_6823_interleave_0 = const()[name = string("op_6823_interleave_0"), val = bool(false)]; + tensor var_6823_cast_fp16 = concat(axis = var_5912, interleave = var_6823_interleave_0, values = (var_6793_cast_fp16, var_6795_cast_fp16, var_6797_cast_fp16, var_6799_cast_fp16))[name = string("op_6823_cast_fp16")]; + bool input_49_interleave_0 = const()[name = string("input_49_interleave_0"), val = bool(false)]; + tensor input_49_cast_fp16 = concat(axis = var_5929, interleave = input_49_interleave_0, values = (var_6801_cast_fp16, var_6803_cast_fp16, var_6805_cast_fp16, var_6807_cast_fp16, var_6809_cast_fp16, var_6811_cast_fp16, var_6813_cast_fp16, var_6815_cast_fp16, var_6817_cast_fp16, var_6819_cast_fp16, var_6821_cast_fp16, var_6823_cast_fp16))[name = string("input_49_cast_fp16")]; + string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; + tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; + tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; + int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94814784)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95994496)))]; + tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_6842_to_fp16 = const()[name = string("op_6842_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_6842_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95996096)))]; + tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95997696)))]; + fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; + string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; + tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; + tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; + int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95999296)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100717952)))]; + tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; + string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; + tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100724160)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105442816)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; + int32 var_6871 = const()[name = string("op_6871"), val = int32(3)]; + int32 var_6888 = const()[name = string("op_6888"), val = int32(1)]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_6905_to_fp16 = const()[name = string("op_6905_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_6905_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105444416)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105446016)))]; + fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105447616)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106627328)))]; + tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; + string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; + tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; + tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; + int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106628928)))]; + tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; + string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; + tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; + tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; + int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107808640)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108988352)))]; + tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_6943_begin_0 = const()[name = string("op_6943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6943_end_0 = const()[name = string("op_6943_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6943_end_mask_0 = const()[name = string("op_6943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = query_15_cast_fp16)[name = string("op_6943_cast_fp16")]; + tensor var_6947_begin_0 = const()[name = string("op_6947_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6947_end_0 = const()[name = string("op_6947_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6947_end_mask_0 = const()[name = string("op_6947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = query_15_cast_fp16)[name = string("op_6947_cast_fp16")]; + tensor var_6951_begin_0 = const()[name = string("op_6951_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6951_end_0 = const()[name = string("op_6951_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6951_end_mask_0 = const()[name = string("op_6951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = query_15_cast_fp16)[name = string("op_6951_cast_fp16")]; + tensor var_6955_begin_0 = const()[name = string("op_6955_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6955_end_0 = const()[name = string("op_6955_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6955_end_mask_0 = const()[name = string("op_6955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = query_15_cast_fp16)[name = string("op_6955_cast_fp16")]; + tensor var_6959_begin_0 = const()[name = string("op_6959_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6959_end_0 = const()[name = string("op_6959_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6959_end_mask_0 = const()[name = string("op_6959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = query_15_cast_fp16)[name = string("op_6959_cast_fp16")]; + tensor var_6963_begin_0 = const()[name = string("op_6963_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6963_end_0 = const()[name = string("op_6963_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6963_end_mask_0 = const()[name = string("op_6963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = query_15_cast_fp16)[name = string("op_6963_cast_fp16")]; + tensor var_6967_begin_0 = const()[name = string("op_6967_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6967_end_0 = const()[name = string("op_6967_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6967_end_mask_0 = const()[name = string("op_6967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6967_cast_fp16 = slice_by_index(begin = var_6967_begin_0, end = var_6967_end_0, end_mask = var_6967_end_mask_0, x = query_15_cast_fp16)[name = string("op_6967_cast_fp16")]; + tensor var_6971_begin_0 = const()[name = string("op_6971_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6971_end_0 = const()[name = string("op_6971_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6971_end_mask_0 = const()[name = string("op_6971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6971_cast_fp16 = slice_by_index(begin = var_6971_begin_0, end = var_6971_end_0, end_mask = var_6971_end_mask_0, x = query_15_cast_fp16)[name = string("op_6971_cast_fp16")]; + tensor var_6975_begin_0 = const()[name = string("op_6975_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6975_end_0 = const()[name = string("op_6975_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6975_end_mask_0 = const()[name = string("op_6975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6975_cast_fp16 = slice_by_index(begin = var_6975_begin_0, end = var_6975_end_0, end_mask = var_6975_end_mask_0, x = query_15_cast_fp16)[name = string("op_6975_cast_fp16")]; + tensor var_6979_begin_0 = const()[name = string("op_6979_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6979_end_0 = const()[name = string("op_6979_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6979_end_mask_0 = const()[name = string("op_6979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6979_cast_fp16 = slice_by_index(begin = var_6979_begin_0, end = var_6979_end_0, end_mask = var_6979_end_mask_0, x = query_15_cast_fp16)[name = string("op_6979_cast_fp16")]; + tensor var_6983_begin_0 = const()[name = string("op_6983_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6983_end_0 = const()[name = string("op_6983_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6983_end_mask_0 = const()[name = string("op_6983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6983_cast_fp16 = slice_by_index(begin = var_6983_begin_0, end = var_6983_end_0, end_mask = var_6983_end_mask_0, x = query_15_cast_fp16)[name = string("op_6983_cast_fp16")]; + tensor var_6987_begin_0 = const()[name = string("op_6987_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6987_end_0 = const()[name = string("op_6987_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6987_end_mask_0 = const()[name = string("op_6987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6987_cast_fp16 = slice_by_index(begin = var_6987_begin_0, end = var_6987_end_0, end_mask = var_6987_end_mask_0, x = query_15_cast_fp16)[name = string("op_6987_cast_fp16")]; + tensor var_6996_begin_0 = const()[name = string("op_6996_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6996_end_0 = const()[name = string("op_6996_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6996_end_mask_0 = const()[name = string("op_6996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6996_cast_fp16 = slice_by_index(begin = var_6996_begin_0, end = var_6996_end_0, end_mask = var_6996_end_mask_0, x = var_6943_cast_fp16)[name = string("op_6996_cast_fp16")]; + tensor var_7003_begin_0 = const()[name = string("op_7003_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7003_end_0 = const()[name = string("op_7003_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7003_end_mask_0 = const()[name = string("op_7003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7003_cast_fp16 = slice_by_index(begin = var_7003_begin_0, end = var_7003_end_0, end_mask = var_7003_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7003_cast_fp16")]; + tensor var_7010_begin_0 = const()[name = string("op_7010_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7010_end_0 = const()[name = string("op_7010_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7010_end_mask_0 = const()[name = string("op_7010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7010_cast_fp16 = slice_by_index(begin = var_7010_begin_0, end = var_7010_end_0, end_mask = var_7010_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7010_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = string("op_7017_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7017_end_0 = const()[name = string("op_7017_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7017_end_mask_0 = const()[name = string("op_7017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7017_cast_fp16")]; + tensor var_7024_begin_0 = const()[name = string("op_7024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7024_end_0 = const()[name = string("op_7024_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7024_end_mask_0 = const()[name = string("op_7024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7024_cast_fp16 = slice_by_index(begin = var_7024_begin_0, end = var_7024_end_0, end_mask = var_7024_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7024_cast_fp16")]; + tensor var_7031_begin_0 = const()[name = string("op_7031_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7031_end_0 = const()[name = string("op_7031_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7031_end_mask_0 = const()[name = string("op_7031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7031_cast_fp16 = slice_by_index(begin = var_7031_begin_0, end = var_7031_end_0, end_mask = var_7031_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7031_cast_fp16")]; + tensor var_7038_begin_0 = const()[name = string("op_7038_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7038_end_0 = const()[name = string("op_7038_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7038_end_mask_0 = const()[name = string("op_7038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7038_cast_fp16 = slice_by_index(begin = var_7038_begin_0, end = var_7038_end_0, end_mask = var_7038_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7038_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = string("op_7045_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7045_end_0 = const()[name = string("op_7045_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7045_end_mask_0 = const()[name = string("op_7045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7045_cast_fp16")]; + tensor var_7052_begin_0 = const()[name = string("op_7052_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7052_end_0 = const()[name = string("op_7052_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7052_end_mask_0 = const()[name = string("op_7052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7052_cast_fp16 = slice_by_index(begin = var_7052_begin_0, end = var_7052_end_0, end_mask = var_7052_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7052_cast_fp16")]; + tensor var_7059_begin_0 = const()[name = string("op_7059_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7059_end_0 = const()[name = string("op_7059_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7059_end_mask_0 = const()[name = string("op_7059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7059_cast_fp16 = slice_by_index(begin = var_7059_begin_0, end = var_7059_end_0, end_mask = var_7059_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7059_cast_fp16")]; + tensor var_7066_begin_0 = const()[name = string("op_7066_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7066_end_0 = const()[name = string("op_7066_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7066_end_mask_0 = const()[name = string("op_7066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7066_cast_fp16 = slice_by_index(begin = var_7066_begin_0, end = var_7066_end_0, end_mask = var_7066_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7066_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = string("op_7073_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7073_end_0 = const()[name = string("op_7073_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7073_end_mask_0 = const()[name = string("op_7073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7073_cast_fp16")]; + tensor var_7080_begin_0 = const()[name = string("op_7080_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7080_end_0 = const()[name = string("op_7080_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7080_end_mask_0 = const()[name = string("op_7080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7080_cast_fp16 = slice_by_index(begin = var_7080_begin_0, end = var_7080_end_0, end_mask = var_7080_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7080_cast_fp16")]; + tensor var_7087_begin_0 = const()[name = string("op_7087_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7087_end_0 = const()[name = string("op_7087_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7087_end_mask_0 = const()[name = string("op_7087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7087_cast_fp16 = slice_by_index(begin = var_7087_begin_0, end = var_7087_end_0, end_mask = var_7087_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7087_cast_fp16")]; + tensor var_7094_begin_0 = const()[name = string("op_7094_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7094_end_0 = const()[name = string("op_7094_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7094_end_mask_0 = const()[name = string("op_7094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7094_cast_fp16 = slice_by_index(begin = var_7094_begin_0, end = var_7094_end_0, end_mask = var_7094_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7094_cast_fp16")]; + tensor var_7101_begin_0 = const()[name = string("op_7101_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7101_end_0 = const()[name = string("op_7101_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7101_end_mask_0 = const()[name = string("op_7101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7101_cast_fp16 = slice_by_index(begin = var_7101_begin_0, end = var_7101_end_0, end_mask = var_7101_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7101_cast_fp16")]; + tensor var_7108_begin_0 = const()[name = string("op_7108_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7108_end_0 = const()[name = string("op_7108_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7108_end_mask_0 = const()[name = string("op_7108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7108_cast_fp16 = slice_by_index(begin = var_7108_begin_0, end = var_7108_end_0, end_mask = var_7108_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7108_cast_fp16")]; + tensor var_7115_begin_0 = const()[name = string("op_7115_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7115_end_0 = const()[name = string("op_7115_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7115_end_mask_0 = const()[name = string("op_7115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7115_cast_fp16 = slice_by_index(begin = var_7115_begin_0, end = var_7115_end_0, end_mask = var_7115_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7115_cast_fp16")]; + tensor var_7122_begin_0 = const()[name = string("op_7122_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7122_end_0 = const()[name = string("op_7122_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7122_end_mask_0 = const()[name = string("op_7122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7122_cast_fp16 = slice_by_index(begin = var_7122_begin_0, end = var_7122_end_0, end_mask = var_7122_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7122_cast_fp16")]; + tensor var_7129_begin_0 = const()[name = string("op_7129_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7129_end_0 = const()[name = string("op_7129_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7129_end_mask_0 = const()[name = string("op_7129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7129_cast_fp16 = slice_by_index(begin = var_7129_begin_0, end = var_7129_end_0, end_mask = var_7129_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7129_cast_fp16")]; + tensor var_7136_begin_0 = const()[name = string("op_7136_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7136_end_0 = const()[name = string("op_7136_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7136_end_mask_0 = const()[name = string("op_7136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7136_cast_fp16 = slice_by_index(begin = var_7136_begin_0, end = var_7136_end_0, end_mask = var_7136_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7136_cast_fp16")]; + tensor var_7143_begin_0 = const()[name = string("op_7143_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7143_end_0 = const()[name = string("op_7143_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7143_end_mask_0 = const()[name = string("op_7143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7143_cast_fp16 = slice_by_index(begin = var_7143_begin_0, end = var_7143_end_0, end_mask = var_7143_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7143_cast_fp16")]; + tensor var_7150_begin_0 = const()[name = string("op_7150_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7150_end_0 = const()[name = string("op_7150_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7150_end_mask_0 = const()[name = string("op_7150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7150_cast_fp16 = slice_by_index(begin = var_7150_begin_0, end = var_7150_end_0, end_mask = var_7150_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7150_cast_fp16")]; + tensor var_7157_begin_0 = const()[name = string("op_7157_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7157_end_0 = const()[name = string("op_7157_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7157_end_mask_0 = const()[name = string("op_7157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7157_cast_fp16 = slice_by_index(begin = var_7157_begin_0, end = var_7157_end_0, end_mask = var_7157_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7157_cast_fp16")]; + tensor var_7164_begin_0 = const()[name = string("op_7164_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7164_end_0 = const()[name = string("op_7164_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7164_end_mask_0 = const()[name = string("op_7164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7164_cast_fp16 = slice_by_index(begin = var_7164_begin_0, end = var_7164_end_0, end_mask = var_7164_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7164_cast_fp16")]; + tensor var_7171_begin_0 = const()[name = string("op_7171_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7171_end_0 = const()[name = string("op_7171_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7171_end_mask_0 = const()[name = string("op_7171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7171_cast_fp16 = slice_by_index(begin = var_7171_begin_0, end = var_7171_end_0, end_mask = var_7171_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7171_cast_fp16")]; + tensor var_7178_begin_0 = const()[name = string("op_7178_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7178_end_0 = const()[name = string("op_7178_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7178_end_mask_0 = const()[name = string("op_7178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7178_cast_fp16 = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7178_cast_fp16")]; + tensor var_7185_begin_0 = const()[name = string("op_7185_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7185_end_0 = const()[name = string("op_7185_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7185_end_mask_0 = const()[name = string("op_7185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7185_cast_fp16 = slice_by_index(begin = var_7185_begin_0, end = var_7185_end_0, end_mask = var_7185_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7185_cast_fp16")]; + tensor var_7192_begin_0 = const()[name = string("op_7192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7192_end_0 = const()[name = string("op_7192_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7192_end_mask_0 = const()[name = string("op_7192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7192_cast_fp16 = slice_by_index(begin = var_7192_begin_0, end = var_7192_end_0, end_mask = var_7192_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7192_cast_fp16")]; + tensor var_7199_begin_0 = const()[name = string("op_7199_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7199_end_0 = const()[name = string("op_7199_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7199_end_mask_0 = const()[name = string("op_7199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7199_cast_fp16 = slice_by_index(begin = var_7199_begin_0, end = var_7199_end_0, end_mask = var_7199_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7199_cast_fp16")]; + tensor var_7206_begin_0 = const()[name = string("op_7206_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7206_end_0 = const()[name = string("op_7206_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7206_end_mask_0 = const()[name = string("op_7206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7206_cast_fp16 = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7206_cast_fp16")]; + tensor var_7213_begin_0 = const()[name = string("op_7213_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7213_end_0 = const()[name = string("op_7213_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7213_end_mask_0 = const()[name = string("op_7213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7213_cast_fp16 = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7213_cast_fp16")]; + tensor var_7220_begin_0 = const()[name = string("op_7220_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7220_end_0 = const()[name = string("op_7220_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7220_end_mask_0 = const()[name = string("op_7220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7220_cast_fp16 = slice_by_index(begin = var_7220_begin_0, end = var_7220_end_0, end_mask = var_7220_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7220_cast_fp16")]; + tensor var_7227_begin_0 = const()[name = string("op_7227_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7227_end_0 = const()[name = string("op_7227_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7227_end_mask_0 = const()[name = string("op_7227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7227_cast_fp16 = slice_by_index(begin = var_7227_begin_0, end = var_7227_end_0, end_mask = var_7227_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7227_cast_fp16")]; + tensor var_7234_begin_0 = const()[name = string("op_7234_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7234_end_0 = const()[name = string("op_7234_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7234_end_mask_0 = const()[name = string("op_7234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7234_cast_fp16 = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7234_cast_fp16")]; + tensor var_7241_begin_0 = const()[name = string("op_7241_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7241_end_0 = const()[name = string("op_7241_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7241_end_mask_0 = const()[name = string("op_7241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7241_cast_fp16 = slice_by_index(begin = var_7241_begin_0, end = var_7241_end_0, end_mask = var_7241_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7241_cast_fp16")]; + tensor var_7248_begin_0 = const()[name = string("op_7248_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7248_end_0 = const()[name = string("op_7248_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7248_end_mask_0 = const()[name = string("op_7248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7248_cast_fp16 = slice_by_index(begin = var_7248_begin_0, end = var_7248_end_0, end_mask = var_7248_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7248_cast_fp16")]; + tensor var_7255_begin_0 = const()[name = string("op_7255_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7255_end_0 = const()[name = string("op_7255_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7255_end_mask_0 = const()[name = string("op_7255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7255_cast_fp16 = slice_by_index(begin = var_7255_begin_0, end = var_7255_end_0, end_mask = var_7255_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7255_cast_fp16")]; + tensor var_7262_begin_0 = const()[name = string("op_7262_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7262_end_0 = const()[name = string("op_7262_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7262_end_mask_0 = const()[name = string("op_7262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7262_cast_fp16 = slice_by_index(begin = var_7262_begin_0, end = var_7262_end_0, end_mask = var_7262_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7262_cast_fp16")]; + tensor var_7269_begin_0 = const()[name = string("op_7269_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7269_end_0 = const()[name = string("op_7269_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7269_end_mask_0 = const()[name = string("op_7269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7269_cast_fp16 = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7269_cast_fp16")]; + tensor var_7276_begin_0 = const()[name = string("op_7276_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7276_end_0 = const()[name = string("op_7276_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7276_end_mask_0 = const()[name = string("op_7276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7276_cast_fp16 = slice_by_index(begin = var_7276_begin_0, end = var_7276_end_0, end_mask = var_7276_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7276_cast_fp16")]; + tensor var_7283_begin_0 = const()[name = string("op_7283_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7283_end_0 = const()[name = string("op_7283_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7283_end_mask_0 = const()[name = string("op_7283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7283_cast_fp16 = slice_by_index(begin = var_7283_begin_0, end = var_7283_end_0, end_mask = var_7283_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7283_cast_fp16")]; + tensor var_7290_begin_0 = const()[name = string("op_7290_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7290_end_0 = const()[name = string("op_7290_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7290_end_mask_0 = const()[name = string("op_7290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7290_cast_fp16 = slice_by_index(begin = var_7290_begin_0, end = var_7290_end_0, end_mask = var_7290_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7290_cast_fp16")]; + tensor var_7297_begin_0 = const()[name = string("op_7297_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7297_end_0 = const()[name = string("op_7297_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7297_end_mask_0 = const()[name = string("op_7297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7297_cast_fp16 = slice_by_index(begin = var_7297_begin_0, end = var_7297_end_0, end_mask = var_7297_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7297_cast_fp16")]; + tensor var_7304_begin_0 = const()[name = string("op_7304_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7304_end_0 = const()[name = string("op_7304_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7304_end_mask_0 = const()[name = string("op_7304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7304_cast_fp16 = slice_by_index(begin = var_7304_begin_0, end = var_7304_end_0, end_mask = var_7304_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7304_cast_fp16")]; + tensor var_7311_begin_0 = const()[name = string("op_7311_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7311_end_0 = const()[name = string("op_7311_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7311_end_mask_0 = const()[name = string("op_7311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7311_cast_fp16 = slice_by_index(begin = var_7311_begin_0, end = var_7311_end_0, end_mask = var_7311_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7311_cast_fp16")]; + tensor var_7318_begin_0 = const()[name = string("op_7318_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7318_end_0 = const()[name = string("op_7318_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7318_end_mask_0 = const()[name = string("op_7318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7318_cast_fp16 = slice_by_index(begin = var_7318_begin_0, end = var_7318_end_0, end_mask = var_7318_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7318_cast_fp16")]; + tensor var_7325_begin_0 = const()[name = string("op_7325_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7325_end_0 = const()[name = string("op_7325_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7325_end_mask_0 = const()[name = string("op_7325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7325_cast_fp16 = slice_by_index(begin = var_7325_begin_0, end = var_7325_end_0, end_mask = var_7325_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7325_cast_fp16")]; + tensor k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7330_begin_0 = const()[name = string("op_7330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7330_end_0 = const()[name = string("op_7330_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_7330_end_mask_0 = const()[name = string("op_7330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = string("transpose_4")]; + tensor var_7330_cast_fp16 = slice_by_index(begin = var_7330_begin_0, end = var_7330_end_0, end_mask = var_7330_end_mask_0, x = k_15_cast_fp16)[name = string("op_7330_cast_fp16")]; + tensor var_7334_begin_0 = const()[name = string("op_7334_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7334_end_0 = const()[name = string("op_7334_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_7334_end_mask_0 = const()[name = string("op_7334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7334_cast_fp16 = slice_by_index(begin = var_7334_begin_0, end = var_7334_end_0, end_mask = var_7334_end_mask_0, x = k_15_cast_fp16)[name = string("op_7334_cast_fp16")]; + tensor var_7338_begin_0 = const()[name = string("op_7338_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7338_end_0 = const()[name = string("op_7338_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_7338_end_mask_0 = const()[name = string("op_7338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7338_cast_fp16 = slice_by_index(begin = var_7338_begin_0, end = var_7338_end_0, end_mask = var_7338_end_mask_0, x = k_15_cast_fp16)[name = string("op_7338_cast_fp16")]; + tensor var_7342_begin_0 = const()[name = string("op_7342_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7342_end_0 = const()[name = string("op_7342_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_7342_end_mask_0 = const()[name = string("op_7342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7342_cast_fp16 = slice_by_index(begin = var_7342_begin_0, end = var_7342_end_0, end_mask = var_7342_end_mask_0, x = k_15_cast_fp16)[name = string("op_7342_cast_fp16")]; + tensor var_7346_begin_0 = const()[name = string("op_7346_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7346_end_0 = const()[name = string("op_7346_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_7346_end_mask_0 = const()[name = string("op_7346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7346_cast_fp16 = slice_by_index(begin = var_7346_begin_0, end = var_7346_end_0, end_mask = var_7346_end_mask_0, x = k_15_cast_fp16)[name = string("op_7346_cast_fp16")]; + tensor var_7350_begin_0 = const()[name = string("op_7350_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7350_end_0 = const()[name = string("op_7350_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_7350_end_mask_0 = const()[name = string("op_7350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7350_cast_fp16 = slice_by_index(begin = var_7350_begin_0, end = var_7350_end_0, end_mask = var_7350_end_mask_0, x = k_15_cast_fp16)[name = string("op_7350_cast_fp16")]; + tensor var_7354_begin_0 = const()[name = string("op_7354_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7354_end_0 = const()[name = string("op_7354_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_7354_end_mask_0 = const()[name = string("op_7354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7354_cast_fp16 = slice_by_index(begin = var_7354_begin_0, end = var_7354_end_0, end_mask = var_7354_end_mask_0, x = k_15_cast_fp16)[name = string("op_7354_cast_fp16")]; + tensor var_7358_begin_0 = const()[name = string("op_7358_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7358_end_0 = const()[name = string("op_7358_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_7358_end_mask_0 = const()[name = string("op_7358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7358_cast_fp16 = slice_by_index(begin = var_7358_begin_0, end = var_7358_end_0, end_mask = var_7358_end_mask_0, x = k_15_cast_fp16)[name = string("op_7358_cast_fp16")]; + tensor var_7362_begin_0 = const()[name = string("op_7362_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7362_end_0 = const()[name = string("op_7362_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_7362_end_mask_0 = const()[name = string("op_7362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7362_cast_fp16 = slice_by_index(begin = var_7362_begin_0, end = var_7362_end_0, end_mask = var_7362_end_mask_0, x = k_15_cast_fp16)[name = string("op_7362_cast_fp16")]; + tensor var_7366_begin_0 = const()[name = string("op_7366_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7366_end_0 = const()[name = string("op_7366_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_7366_end_mask_0 = const()[name = string("op_7366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7366_cast_fp16 = slice_by_index(begin = var_7366_begin_0, end = var_7366_end_0, end_mask = var_7366_end_mask_0, x = k_15_cast_fp16)[name = string("op_7366_cast_fp16")]; + tensor var_7370_begin_0 = const()[name = string("op_7370_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7370_end_0 = const()[name = string("op_7370_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_7370_end_mask_0 = const()[name = string("op_7370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7370_cast_fp16 = slice_by_index(begin = var_7370_begin_0, end = var_7370_end_0, end_mask = var_7370_end_mask_0, x = k_15_cast_fp16)[name = string("op_7370_cast_fp16")]; + tensor var_7374_begin_0 = const()[name = string("op_7374_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7374_end_0 = const()[name = string("op_7374_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_7374_end_mask_0 = const()[name = string("op_7374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7374_cast_fp16 = slice_by_index(begin = var_7374_begin_0, end = var_7374_end_0, end_mask = var_7374_end_mask_0, x = k_15_cast_fp16)[name = string("op_7374_cast_fp16")]; + tensor var_7376_begin_0 = const()[name = string("op_7376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7376_end_0 = const()[name = string("op_7376_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7376_end_mask_0 = const()[name = string("op_7376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7376_cast_fp16 = slice_by_index(begin = var_7376_begin_0, end = var_7376_end_0, end_mask = var_7376_end_mask_0, x = value_15_cast_fp16)[name = string("op_7376_cast_fp16")]; + tensor var_7380_begin_0 = const()[name = string("op_7380_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7380_end_0 = const()[name = string("op_7380_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7380_end_mask_0 = const()[name = string("op_7380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7380_cast_fp16 = slice_by_index(begin = var_7380_begin_0, end = var_7380_end_0, end_mask = var_7380_end_mask_0, x = value_15_cast_fp16)[name = string("op_7380_cast_fp16")]; + tensor var_7384_begin_0 = const()[name = string("op_7384_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7384_end_0 = const()[name = string("op_7384_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7384_end_mask_0 = const()[name = string("op_7384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7384_cast_fp16 = slice_by_index(begin = var_7384_begin_0, end = var_7384_end_0, end_mask = var_7384_end_mask_0, x = value_15_cast_fp16)[name = string("op_7384_cast_fp16")]; + tensor var_7388_begin_0 = const()[name = string("op_7388_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7388_end_0 = const()[name = string("op_7388_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7388_end_mask_0 = const()[name = string("op_7388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7388_cast_fp16 = slice_by_index(begin = var_7388_begin_0, end = var_7388_end_0, end_mask = var_7388_end_mask_0, x = value_15_cast_fp16)[name = string("op_7388_cast_fp16")]; + tensor var_7392_begin_0 = const()[name = string("op_7392_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7392_end_0 = const()[name = string("op_7392_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7392_end_mask_0 = const()[name = string("op_7392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7392_cast_fp16 = slice_by_index(begin = var_7392_begin_0, end = var_7392_end_0, end_mask = var_7392_end_mask_0, x = value_15_cast_fp16)[name = string("op_7392_cast_fp16")]; + tensor var_7396_begin_0 = const()[name = string("op_7396_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7396_end_0 = const()[name = string("op_7396_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7396_end_mask_0 = const()[name = string("op_7396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7396_cast_fp16 = slice_by_index(begin = var_7396_begin_0, end = var_7396_end_0, end_mask = var_7396_end_mask_0, x = value_15_cast_fp16)[name = string("op_7396_cast_fp16")]; + tensor var_7400_begin_0 = const()[name = string("op_7400_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7400_end_0 = const()[name = string("op_7400_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7400_end_mask_0 = const()[name = string("op_7400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7400_cast_fp16 = slice_by_index(begin = var_7400_begin_0, end = var_7400_end_0, end_mask = var_7400_end_mask_0, x = value_15_cast_fp16)[name = string("op_7400_cast_fp16")]; + tensor var_7404_begin_0 = const()[name = string("op_7404_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7404_end_0 = const()[name = string("op_7404_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7404_end_mask_0 = const()[name = string("op_7404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7404_cast_fp16 = slice_by_index(begin = var_7404_begin_0, end = var_7404_end_0, end_mask = var_7404_end_mask_0, x = value_15_cast_fp16)[name = string("op_7404_cast_fp16")]; + tensor var_7408_begin_0 = const()[name = string("op_7408_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7408_end_0 = const()[name = string("op_7408_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7408_end_mask_0 = const()[name = string("op_7408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7408_cast_fp16 = slice_by_index(begin = var_7408_begin_0, end = var_7408_end_0, end_mask = var_7408_end_mask_0, x = value_15_cast_fp16)[name = string("op_7408_cast_fp16")]; + tensor var_7412_begin_0 = const()[name = string("op_7412_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7412_end_0 = const()[name = string("op_7412_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7412_end_mask_0 = const()[name = string("op_7412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7412_cast_fp16 = slice_by_index(begin = var_7412_begin_0, end = var_7412_end_0, end_mask = var_7412_end_mask_0, x = value_15_cast_fp16)[name = string("op_7412_cast_fp16")]; + tensor var_7416_begin_0 = const()[name = string("op_7416_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7416_end_0 = const()[name = string("op_7416_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7416_end_mask_0 = const()[name = string("op_7416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7416_cast_fp16 = slice_by_index(begin = var_7416_begin_0, end = var_7416_end_0, end_mask = var_7416_end_mask_0, x = value_15_cast_fp16)[name = string("op_7416_cast_fp16")]; + tensor var_7420_begin_0 = const()[name = string("op_7420_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7420_end_0 = const()[name = string("op_7420_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7420_end_mask_0 = const()[name = string("op_7420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7420_cast_fp16 = slice_by_index(begin = var_7420_begin_0, end = var_7420_end_0, end_mask = var_7420_end_mask_0, x = value_15_cast_fp16)[name = string("op_7420_cast_fp16")]; + string _SplitHeadsQ__mh_w_673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_673_equation_0, values = (var_7330_cast_fp16, var_6996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_673_cast_fp16")]; + string _SplitHeadsQ__mh_w_675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_675_equation_0, values = (var_7330_cast_fp16, var_7003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_675_cast_fp16")]; + string _SplitHeadsQ__mh_w_677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_677_equation_0, values = (var_7330_cast_fp16, var_7010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_677_cast_fp16")]; + string _SplitHeadsQ__mh_w_679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_679_equation_0, values = (var_7330_cast_fp16, var_7017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_679_cast_fp16")]; + string _SplitHeadsQ__mh_w_681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_681_equation_0, values = (var_7334_cast_fp16, var_7024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_681_cast_fp16")]; + string _SplitHeadsQ__mh_w_683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_683_equation_0, values = (var_7334_cast_fp16, var_7031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_683_cast_fp16")]; + string _SplitHeadsQ__mh_w_685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_685_equation_0, values = (var_7334_cast_fp16, var_7038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_685_cast_fp16")]; + string _SplitHeadsQ__mh_w_687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_687_equation_0, values = (var_7334_cast_fp16, var_7045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_687_cast_fp16")]; + string _SplitHeadsQ__mh_w_689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_689_equation_0, values = (var_7338_cast_fp16, var_7052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_689_cast_fp16")]; + string _SplitHeadsQ__mh_w_691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_691_equation_0, values = (var_7338_cast_fp16, var_7059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_691_cast_fp16")]; + string _SplitHeadsQ__mh_w_693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_693_equation_0, values = (var_7338_cast_fp16, var_7066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_693_cast_fp16")]; + string _SplitHeadsQ__mh_w_695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_695_equation_0, values = (var_7338_cast_fp16, var_7073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_695_cast_fp16")]; + string _SplitHeadsQ__mh_w_697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_697_equation_0, values = (var_7342_cast_fp16, var_7080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_697_cast_fp16")]; + string _SplitHeadsQ__mh_w_699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_699_equation_0, values = (var_7342_cast_fp16, var_7087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_699_cast_fp16")]; + string _SplitHeadsQ__mh_w_701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_701_equation_0, values = (var_7342_cast_fp16, var_7094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_701_cast_fp16")]; + string _SplitHeadsQ__mh_w_703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_703_equation_0, values = (var_7342_cast_fp16, var_7101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_703_cast_fp16")]; + string _SplitHeadsQ__mh_w_705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_705_equation_0, values = (var_7346_cast_fp16, var_7108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_705_cast_fp16")]; + string _SplitHeadsQ__mh_w_707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_707_equation_0, values = (var_7346_cast_fp16, var_7115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_707_cast_fp16")]; + string _SplitHeadsQ__mh_w_709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_709_equation_0, values = (var_7346_cast_fp16, var_7122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_709_cast_fp16")]; + string _SplitHeadsQ__mh_w_711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_711_equation_0, values = (var_7346_cast_fp16, var_7129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_711_cast_fp16")]; + string _SplitHeadsQ__mh_w_713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_713_equation_0, values = (var_7350_cast_fp16, var_7136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_713_cast_fp16")]; + string _SplitHeadsQ__mh_w_715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_715_equation_0, values = (var_7350_cast_fp16, var_7143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_715_cast_fp16")]; + string _SplitHeadsQ__mh_w_717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_717_equation_0, values = (var_7350_cast_fp16, var_7150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_717_cast_fp16")]; + string _SplitHeadsQ__mh_w_719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_719_equation_0, values = (var_7350_cast_fp16, var_7157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_719_cast_fp16")]; + string _SplitHeadsQ__mh_w_721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_721_equation_0, values = (var_7354_cast_fp16, var_7164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_721_cast_fp16")]; + string _SplitHeadsQ__mh_w_723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_723_equation_0, values = (var_7354_cast_fp16, var_7171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_723_cast_fp16")]; + string _SplitHeadsQ__mh_w_725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_725_equation_0, values = (var_7354_cast_fp16, var_7178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_725_cast_fp16")]; + string _SplitHeadsQ__mh_w_727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_727_equation_0, values = (var_7354_cast_fp16, var_7185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_727_cast_fp16")]; + string _SplitHeadsQ__mh_w_729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_729_equation_0, values = (var_7358_cast_fp16, var_7192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_729_cast_fp16")]; + string _SplitHeadsQ__mh_w_731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_731_equation_0, values = (var_7358_cast_fp16, var_7199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_731_cast_fp16")]; + string _SplitHeadsQ__mh_w_733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_733_equation_0, values = (var_7358_cast_fp16, var_7206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_733_cast_fp16")]; + string _SplitHeadsQ__mh_w_735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_735_equation_0, values = (var_7358_cast_fp16, var_7213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_735_cast_fp16")]; + string _SplitHeadsQ__mh_w_737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_737_equation_0, values = (var_7362_cast_fp16, var_7220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_737_cast_fp16")]; + string _SplitHeadsQ__mh_w_739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_739_equation_0, values = (var_7362_cast_fp16, var_7227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_739_cast_fp16")]; + string _SplitHeadsQ__mh_w_741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_741_equation_0, values = (var_7362_cast_fp16, var_7234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_741_cast_fp16")]; + string _SplitHeadsQ__mh_w_743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_743_equation_0, values = (var_7362_cast_fp16, var_7241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_743_cast_fp16")]; + string _SplitHeadsQ__mh_w_745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_745_equation_0, values = (var_7366_cast_fp16, var_7248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_745_cast_fp16")]; + string _SplitHeadsQ__mh_w_747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_747_equation_0, values = (var_7366_cast_fp16, var_7255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_747_cast_fp16")]; + string _SplitHeadsQ__mh_w_749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_749_equation_0, values = (var_7366_cast_fp16, var_7262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_749_cast_fp16")]; + string _SplitHeadsQ__mh_w_751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_751_equation_0, values = (var_7366_cast_fp16, var_7269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_751_cast_fp16")]; + string _SplitHeadsQ__mh_w_753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_753_equation_0, values = (var_7370_cast_fp16, var_7276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_753_cast_fp16")]; + string _SplitHeadsQ__mh_w_755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_755_equation_0, values = (var_7370_cast_fp16, var_7283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_755_cast_fp16")]; + string _SplitHeadsQ__mh_w_757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_757_equation_0, values = (var_7370_cast_fp16, var_7290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_757_cast_fp16")]; + string _SplitHeadsQ__mh_w_759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_759_equation_0, values = (var_7370_cast_fp16, var_7297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_759_cast_fp16")]; + string _SplitHeadsQ__mh_w_761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_761_equation_0, values = (var_7374_cast_fp16, var_7304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_761_cast_fp16")]; + string _SplitHeadsQ__mh_w_763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_763_equation_0, values = (var_7374_cast_fp16, var_7311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_763_cast_fp16")]; + string _SplitHeadsQ__mh_w_765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_765_equation_0, values = (var_7374_cast_fp16, var_7318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_765_cast_fp16")]; + string _SplitHeadsQ__mh_w_767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_767_equation_0, values = (var_7374_cast_fp16, var_7325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_767_cast_fp16")]; + fp16 var_7519_to_fp16 = const()[name = string("op_7519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_673_cast_fp16, y = var_7519_to_fp16)[name = string("aw_chunk_673_cast_fp16")]; + fp16 var_7521_to_fp16 = const()[name = string("op_7521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_675_cast_fp16, y = var_7521_to_fp16)[name = string("aw_chunk_675_cast_fp16")]; + fp16 var_7523_to_fp16 = const()[name = string("op_7523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_677_cast_fp16, y = var_7523_to_fp16)[name = string("aw_chunk_677_cast_fp16")]; + fp16 var_7525_to_fp16 = const()[name = string("op_7525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_679_cast_fp16, y = var_7525_to_fp16)[name = string("aw_chunk_679_cast_fp16")]; + fp16 var_7527_to_fp16 = const()[name = string("op_7527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_681_cast_fp16, y = var_7527_to_fp16)[name = string("aw_chunk_681_cast_fp16")]; + fp16 var_7529_to_fp16 = const()[name = string("op_7529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_683_cast_fp16, y = var_7529_to_fp16)[name = string("aw_chunk_683_cast_fp16")]; + fp16 var_7531_to_fp16 = const()[name = string("op_7531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_685_cast_fp16, y = var_7531_to_fp16)[name = string("aw_chunk_685_cast_fp16")]; + fp16 var_7533_to_fp16 = const()[name = string("op_7533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_687_cast_fp16, y = var_7533_to_fp16)[name = string("aw_chunk_687_cast_fp16")]; + fp16 var_7535_to_fp16 = const()[name = string("op_7535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_689_cast_fp16, y = var_7535_to_fp16)[name = string("aw_chunk_689_cast_fp16")]; + fp16 var_7537_to_fp16 = const()[name = string("op_7537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_691_cast_fp16, y = var_7537_to_fp16)[name = string("aw_chunk_691_cast_fp16")]; + fp16 var_7539_to_fp16 = const()[name = string("op_7539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_693_cast_fp16, y = var_7539_to_fp16)[name = string("aw_chunk_693_cast_fp16")]; + fp16 var_7541_to_fp16 = const()[name = string("op_7541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_695_cast_fp16, y = var_7541_to_fp16)[name = string("aw_chunk_695_cast_fp16")]; + fp16 var_7543_to_fp16 = const()[name = string("op_7543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_697_cast_fp16, y = var_7543_to_fp16)[name = string("aw_chunk_697_cast_fp16")]; + fp16 var_7545_to_fp16 = const()[name = string("op_7545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_699_cast_fp16, y = var_7545_to_fp16)[name = string("aw_chunk_699_cast_fp16")]; + fp16 var_7547_to_fp16 = const()[name = string("op_7547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_701_cast_fp16, y = var_7547_to_fp16)[name = string("aw_chunk_701_cast_fp16")]; + fp16 var_7549_to_fp16 = const()[name = string("op_7549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_703_cast_fp16, y = var_7549_to_fp16)[name = string("aw_chunk_703_cast_fp16")]; + fp16 var_7551_to_fp16 = const()[name = string("op_7551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_705_cast_fp16, y = var_7551_to_fp16)[name = string("aw_chunk_705_cast_fp16")]; + fp16 var_7553_to_fp16 = const()[name = string("op_7553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_707_cast_fp16, y = var_7553_to_fp16)[name = string("aw_chunk_707_cast_fp16")]; + fp16 var_7555_to_fp16 = const()[name = string("op_7555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_709_cast_fp16, y = var_7555_to_fp16)[name = string("aw_chunk_709_cast_fp16")]; + fp16 var_7557_to_fp16 = const()[name = string("op_7557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_711_cast_fp16, y = var_7557_to_fp16)[name = string("aw_chunk_711_cast_fp16")]; + fp16 var_7559_to_fp16 = const()[name = string("op_7559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_713_cast_fp16, y = var_7559_to_fp16)[name = string("aw_chunk_713_cast_fp16")]; + fp16 var_7561_to_fp16 = const()[name = string("op_7561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_715_cast_fp16, y = var_7561_to_fp16)[name = string("aw_chunk_715_cast_fp16")]; + fp16 var_7563_to_fp16 = const()[name = string("op_7563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_717_cast_fp16, y = var_7563_to_fp16)[name = string("aw_chunk_717_cast_fp16")]; + fp16 var_7565_to_fp16 = const()[name = string("op_7565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_719_cast_fp16, y = var_7565_to_fp16)[name = string("aw_chunk_719_cast_fp16")]; + fp16 var_7567_to_fp16 = const()[name = string("op_7567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_721_cast_fp16, y = var_7567_to_fp16)[name = string("aw_chunk_721_cast_fp16")]; + fp16 var_7569_to_fp16 = const()[name = string("op_7569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_723_cast_fp16, y = var_7569_to_fp16)[name = string("aw_chunk_723_cast_fp16")]; + fp16 var_7571_to_fp16 = const()[name = string("op_7571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_725_cast_fp16, y = var_7571_to_fp16)[name = string("aw_chunk_725_cast_fp16")]; + fp16 var_7573_to_fp16 = const()[name = string("op_7573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_727_cast_fp16, y = var_7573_to_fp16)[name = string("aw_chunk_727_cast_fp16")]; + fp16 var_7575_to_fp16 = const()[name = string("op_7575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_729_cast_fp16, y = var_7575_to_fp16)[name = string("aw_chunk_729_cast_fp16")]; + fp16 var_7577_to_fp16 = const()[name = string("op_7577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_731_cast_fp16, y = var_7577_to_fp16)[name = string("aw_chunk_731_cast_fp16")]; + fp16 var_7579_to_fp16 = const()[name = string("op_7579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_733_cast_fp16, y = var_7579_to_fp16)[name = string("aw_chunk_733_cast_fp16")]; + fp16 var_7581_to_fp16 = const()[name = string("op_7581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_735_cast_fp16, y = var_7581_to_fp16)[name = string("aw_chunk_735_cast_fp16")]; + fp16 var_7583_to_fp16 = const()[name = string("op_7583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_737_cast_fp16, y = var_7583_to_fp16)[name = string("aw_chunk_737_cast_fp16")]; + fp16 var_7585_to_fp16 = const()[name = string("op_7585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_739_cast_fp16, y = var_7585_to_fp16)[name = string("aw_chunk_739_cast_fp16")]; + fp16 var_7587_to_fp16 = const()[name = string("op_7587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_741_cast_fp16, y = var_7587_to_fp16)[name = string("aw_chunk_741_cast_fp16")]; + fp16 var_7589_to_fp16 = const()[name = string("op_7589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_743_cast_fp16, y = var_7589_to_fp16)[name = string("aw_chunk_743_cast_fp16")]; + fp16 var_7591_to_fp16 = const()[name = string("op_7591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_745_cast_fp16, y = var_7591_to_fp16)[name = string("aw_chunk_745_cast_fp16")]; + fp16 var_7593_to_fp16 = const()[name = string("op_7593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_747_cast_fp16, y = var_7593_to_fp16)[name = string("aw_chunk_747_cast_fp16")]; + fp16 var_7595_to_fp16 = const()[name = string("op_7595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_749_cast_fp16, y = var_7595_to_fp16)[name = string("aw_chunk_749_cast_fp16")]; + fp16 var_7597_to_fp16 = const()[name = string("op_7597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_751_cast_fp16, y = var_7597_to_fp16)[name = string("aw_chunk_751_cast_fp16")]; + fp16 var_7599_to_fp16 = const()[name = string("op_7599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_753_cast_fp16, y = var_7599_to_fp16)[name = string("aw_chunk_753_cast_fp16")]; + fp16 var_7601_to_fp16 = const()[name = string("op_7601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_755_cast_fp16, y = var_7601_to_fp16)[name = string("aw_chunk_755_cast_fp16")]; + fp16 var_7603_to_fp16 = const()[name = string("op_7603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_757_cast_fp16, y = var_7603_to_fp16)[name = string("aw_chunk_757_cast_fp16")]; + fp16 var_7605_to_fp16 = const()[name = string("op_7605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_759_cast_fp16, y = var_7605_to_fp16)[name = string("aw_chunk_759_cast_fp16")]; + fp16 var_7607_to_fp16 = const()[name = string("op_7607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_761_cast_fp16, y = var_7607_to_fp16)[name = string("aw_chunk_761_cast_fp16")]; + fp16 var_7609_to_fp16 = const()[name = string("op_7609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_763_cast_fp16, y = var_7609_to_fp16)[name = string("aw_chunk_763_cast_fp16")]; + fp16 var_7611_to_fp16 = const()[name = string("op_7611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_765_cast_fp16, y = var_7611_to_fp16)[name = string("aw_chunk_765_cast_fp16")]; + fp16 var_7613_to_fp16 = const()[name = string("op_7613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_767_cast_fp16, y = var_7613_to_fp16)[name = string("aw_chunk_767_cast_fp16")]; + tensor var_7615_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_673_cast_fp16)[name = string("op_7615_cast_fp16")]; + tensor var_7616_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_675_cast_fp16)[name = string("op_7616_cast_fp16")]; + tensor var_7617_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_677_cast_fp16)[name = string("op_7617_cast_fp16")]; + tensor var_7618_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_679_cast_fp16)[name = string("op_7618_cast_fp16")]; + tensor var_7619_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_681_cast_fp16)[name = string("op_7619_cast_fp16")]; + tensor var_7620_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_683_cast_fp16)[name = string("op_7620_cast_fp16")]; + tensor var_7621_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_685_cast_fp16)[name = string("op_7621_cast_fp16")]; + tensor var_7622_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_687_cast_fp16)[name = string("op_7622_cast_fp16")]; + tensor var_7623_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_689_cast_fp16)[name = string("op_7623_cast_fp16")]; + tensor var_7624_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_691_cast_fp16)[name = string("op_7624_cast_fp16")]; + tensor var_7625_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_693_cast_fp16)[name = string("op_7625_cast_fp16")]; + tensor var_7626_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_695_cast_fp16)[name = string("op_7626_cast_fp16")]; + tensor var_7627_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_697_cast_fp16)[name = string("op_7627_cast_fp16")]; + tensor var_7628_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_699_cast_fp16)[name = string("op_7628_cast_fp16")]; + tensor var_7629_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_701_cast_fp16)[name = string("op_7629_cast_fp16")]; + tensor var_7630_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_703_cast_fp16)[name = string("op_7630_cast_fp16")]; + tensor var_7631_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_705_cast_fp16)[name = string("op_7631_cast_fp16")]; + tensor var_7632_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_707_cast_fp16)[name = string("op_7632_cast_fp16")]; + tensor var_7633_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_709_cast_fp16)[name = string("op_7633_cast_fp16")]; + tensor var_7634_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_711_cast_fp16)[name = string("op_7634_cast_fp16")]; + tensor var_7635_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_713_cast_fp16)[name = string("op_7635_cast_fp16")]; + tensor var_7636_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_715_cast_fp16)[name = string("op_7636_cast_fp16")]; + tensor var_7637_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_717_cast_fp16)[name = string("op_7637_cast_fp16")]; + tensor var_7638_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_719_cast_fp16)[name = string("op_7638_cast_fp16")]; + tensor var_7639_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_721_cast_fp16)[name = string("op_7639_cast_fp16")]; + tensor var_7640_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_723_cast_fp16)[name = string("op_7640_cast_fp16")]; + tensor var_7641_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_725_cast_fp16)[name = string("op_7641_cast_fp16")]; + tensor var_7642_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_727_cast_fp16)[name = string("op_7642_cast_fp16")]; + tensor var_7643_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_729_cast_fp16)[name = string("op_7643_cast_fp16")]; + tensor var_7644_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_731_cast_fp16)[name = string("op_7644_cast_fp16")]; + tensor var_7645_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_733_cast_fp16)[name = string("op_7645_cast_fp16")]; + tensor var_7646_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_735_cast_fp16)[name = string("op_7646_cast_fp16")]; + tensor var_7647_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_737_cast_fp16)[name = string("op_7647_cast_fp16")]; + tensor var_7648_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_739_cast_fp16)[name = string("op_7648_cast_fp16")]; + tensor var_7649_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_741_cast_fp16)[name = string("op_7649_cast_fp16")]; + tensor var_7650_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_743_cast_fp16)[name = string("op_7650_cast_fp16")]; + tensor var_7651_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_745_cast_fp16)[name = string("op_7651_cast_fp16")]; + tensor var_7652_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_747_cast_fp16)[name = string("op_7652_cast_fp16")]; + tensor var_7653_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_749_cast_fp16)[name = string("op_7653_cast_fp16")]; + tensor var_7654_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_751_cast_fp16)[name = string("op_7654_cast_fp16")]; + tensor var_7655_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_753_cast_fp16)[name = string("op_7655_cast_fp16")]; + tensor var_7656_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_755_cast_fp16)[name = string("op_7656_cast_fp16")]; + tensor var_7657_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_757_cast_fp16)[name = string("op_7657_cast_fp16")]; + tensor var_7658_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_759_cast_fp16)[name = string("op_7658_cast_fp16")]; + tensor var_7659_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_761_cast_fp16)[name = string("op_7659_cast_fp16")]; + tensor var_7660_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_763_cast_fp16)[name = string("op_7660_cast_fp16")]; + tensor var_7661_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_765_cast_fp16)[name = string("op_7661_cast_fp16")]; + tensor var_7662_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_767_cast_fp16)[name = string("op_7662_cast_fp16")]; + string var_7664_equation_0 = const()[name = string("op_7664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7664_cast_fp16 = einsum(equation = var_7664_equation_0, values = (var_7376_cast_fp16, var_7615_cast_fp16))[name = string("op_7664_cast_fp16")]; + string var_7666_equation_0 = const()[name = string("op_7666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7666_cast_fp16 = einsum(equation = var_7666_equation_0, values = (var_7376_cast_fp16, var_7616_cast_fp16))[name = string("op_7666_cast_fp16")]; + string var_7668_equation_0 = const()[name = string("op_7668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7668_cast_fp16 = einsum(equation = var_7668_equation_0, values = (var_7376_cast_fp16, var_7617_cast_fp16))[name = string("op_7668_cast_fp16")]; + string var_7670_equation_0 = const()[name = string("op_7670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7670_cast_fp16 = einsum(equation = var_7670_equation_0, values = (var_7376_cast_fp16, var_7618_cast_fp16))[name = string("op_7670_cast_fp16")]; + string var_7672_equation_0 = const()[name = string("op_7672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7672_cast_fp16 = einsum(equation = var_7672_equation_0, values = (var_7380_cast_fp16, var_7619_cast_fp16))[name = string("op_7672_cast_fp16")]; + string var_7674_equation_0 = const()[name = string("op_7674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7674_cast_fp16 = einsum(equation = var_7674_equation_0, values = (var_7380_cast_fp16, var_7620_cast_fp16))[name = string("op_7674_cast_fp16")]; + string var_7676_equation_0 = const()[name = string("op_7676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7676_cast_fp16 = einsum(equation = var_7676_equation_0, values = (var_7380_cast_fp16, var_7621_cast_fp16))[name = string("op_7676_cast_fp16")]; + string var_7678_equation_0 = const()[name = string("op_7678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7678_cast_fp16 = einsum(equation = var_7678_equation_0, values = (var_7380_cast_fp16, var_7622_cast_fp16))[name = string("op_7678_cast_fp16")]; + string var_7680_equation_0 = const()[name = string("op_7680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7680_cast_fp16 = einsum(equation = var_7680_equation_0, values = (var_7384_cast_fp16, var_7623_cast_fp16))[name = string("op_7680_cast_fp16")]; + string var_7682_equation_0 = const()[name = string("op_7682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7682_cast_fp16 = einsum(equation = var_7682_equation_0, values = (var_7384_cast_fp16, var_7624_cast_fp16))[name = string("op_7682_cast_fp16")]; + string var_7684_equation_0 = const()[name = string("op_7684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7684_cast_fp16 = einsum(equation = var_7684_equation_0, values = (var_7384_cast_fp16, var_7625_cast_fp16))[name = string("op_7684_cast_fp16")]; + string var_7686_equation_0 = const()[name = string("op_7686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7686_cast_fp16 = einsum(equation = var_7686_equation_0, values = (var_7384_cast_fp16, var_7626_cast_fp16))[name = string("op_7686_cast_fp16")]; + string var_7688_equation_0 = const()[name = string("op_7688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7688_cast_fp16 = einsum(equation = var_7688_equation_0, values = (var_7388_cast_fp16, var_7627_cast_fp16))[name = string("op_7688_cast_fp16")]; + string var_7690_equation_0 = const()[name = string("op_7690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7690_cast_fp16 = einsum(equation = var_7690_equation_0, values = (var_7388_cast_fp16, var_7628_cast_fp16))[name = string("op_7690_cast_fp16")]; + string var_7692_equation_0 = const()[name = string("op_7692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7692_cast_fp16 = einsum(equation = var_7692_equation_0, values = (var_7388_cast_fp16, var_7629_cast_fp16))[name = string("op_7692_cast_fp16")]; + string var_7694_equation_0 = const()[name = string("op_7694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7694_cast_fp16 = einsum(equation = var_7694_equation_0, values = (var_7388_cast_fp16, var_7630_cast_fp16))[name = string("op_7694_cast_fp16")]; + string var_7696_equation_0 = const()[name = string("op_7696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7696_cast_fp16 = einsum(equation = var_7696_equation_0, values = (var_7392_cast_fp16, var_7631_cast_fp16))[name = string("op_7696_cast_fp16")]; + string var_7698_equation_0 = const()[name = string("op_7698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7698_cast_fp16 = einsum(equation = var_7698_equation_0, values = (var_7392_cast_fp16, var_7632_cast_fp16))[name = string("op_7698_cast_fp16")]; + string var_7700_equation_0 = const()[name = string("op_7700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7700_cast_fp16 = einsum(equation = var_7700_equation_0, values = (var_7392_cast_fp16, var_7633_cast_fp16))[name = string("op_7700_cast_fp16")]; + string var_7702_equation_0 = const()[name = string("op_7702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7702_cast_fp16 = einsum(equation = var_7702_equation_0, values = (var_7392_cast_fp16, var_7634_cast_fp16))[name = string("op_7702_cast_fp16")]; + string var_7704_equation_0 = const()[name = string("op_7704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7704_cast_fp16 = einsum(equation = var_7704_equation_0, values = (var_7396_cast_fp16, var_7635_cast_fp16))[name = string("op_7704_cast_fp16")]; + string var_7706_equation_0 = const()[name = string("op_7706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7706_cast_fp16 = einsum(equation = var_7706_equation_0, values = (var_7396_cast_fp16, var_7636_cast_fp16))[name = string("op_7706_cast_fp16")]; + string var_7708_equation_0 = const()[name = string("op_7708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7708_cast_fp16 = einsum(equation = var_7708_equation_0, values = (var_7396_cast_fp16, var_7637_cast_fp16))[name = string("op_7708_cast_fp16")]; + string var_7710_equation_0 = const()[name = string("op_7710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7710_cast_fp16 = einsum(equation = var_7710_equation_0, values = (var_7396_cast_fp16, var_7638_cast_fp16))[name = string("op_7710_cast_fp16")]; + string var_7712_equation_0 = const()[name = string("op_7712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7712_cast_fp16 = einsum(equation = var_7712_equation_0, values = (var_7400_cast_fp16, var_7639_cast_fp16))[name = string("op_7712_cast_fp16")]; + string var_7714_equation_0 = const()[name = string("op_7714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7714_cast_fp16 = einsum(equation = var_7714_equation_0, values = (var_7400_cast_fp16, var_7640_cast_fp16))[name = string("op_7714_cast_fp16")]; + string var_7716_equation_0 = const()[name = string("op_7716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7716_cast_fp16 = einsum(equation = var_7716_equation_0, values = (var_7400_cast_fp16, var_7641_cast_fp16))[name = string("op_7716_cast_fp16")]; + string var_7718_equation_0 = const()[name = string("op_7718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7718_cast_fp16 = einsum(equation = var_7718_equation_0, values = (var_7400_cast_fp16, var_7642_cast_fp16))[name = string("op_7718_cast_fp16")]; + string var_7720_equation_0 = const()[name = string("op_7720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7720_cast_fp16 = einsum(equation = var_7720_equation_0, values = (var_7404_cast_fp16, var_7643_cast_fp16))[name = string("op_7720_cast_fp16")]; + string var_7722_equation_0 = const()[name = string("op_7722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7722_cast_fp16 = einsum(equation = var_7722_equation_0, values = (var_7404_cast_fp16, var_7644_cast_fp16))[name = string("op_7722_cast_fp16")]; + string var_7724_equation_0 = const()[name = string("op_7724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7724_cast_fp16 = einsum(equation = var_7724_equation_0, values = (var_7404_cast_fp16, var_7645_cast_fp16))[name = string("op_7724_cast_fp16")]; + string var_7726_equation_0 = const()[name = string("op_7726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7726_cast_fp16 = einsum(equation = var_7726_equation_0, values = (var_7404_cast_fp16, var_7646_cast_fp16))[name = string("op_7726_cast_fp16")]; + string var_7728_equation_0 = const()[name = string("op_7728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7728_cast_fp16 = einsum(equation = var_7728_equation_0, values = (var_7408_cast_fp16, var_7647_cast_fp16))[name = string("op_7728_cast_fp16")]; + string var_7730_equation_0 = const()[name = string("op_7730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7730_cast_fp16 = einsum(equation = var_7730_equation_0, values = (var_7408_cast_fp16, var_7648_cast_fp16))[name = string("op_7730_cast_fp16")]; + string var_7732_equation_0 = const()[name = string("op_7732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7732_cast_fp16 = einsum(equation = var_7732_equation_0, values = (var_7408_cast_fp16, var_7649_cast_fp16))[name = string("op_7732_cast_fp16")]; + string var_7734_equation_0 = const()[name = string("op_7734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7734_cast_fp16 = einsum(equation = var_7734_equation_0, values = (var_7408_cast_fp16, var_7650_cast_fp16))[name = string("op_7734_cast_fp16")]; + string var_7736_equation_0 = const()[name = string("op_7736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7736_cast_fp16 = einsum(equation = var_7736_equation_0, values = (var_7412_cast_fp16, var_7651_cast_fp16))[name = string("op_7736_cast_fp16")]; + string var_7738_equation_0 = const()[name = string("op_7738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7738_cast_fp16 = einsum(equation = var_7738_equation_0, values = (var_7412_cast_fp16, var_7652_cast_fp16))[name = string("op_7738_cast_fp16")]; + string var_7740_equation_0 = const()[name = string("op_7740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7740_cast_fp16 = einsum(equation = var_7740_equation_0, values = (var_7412_cast_fp16, var_7653_cast_fp16))[name = string("op_7740_cast_fp16")]; + string var_7742_equation_0 = const()[name = string("op_7742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7742_cast_fp16 = einsum(equation = var_7742_equation_0, values = (var_7412_cast_fp16, var_7654_cast_fp16))[name = string("op_7742_cast_fp16")]; + string var_7744_equation_0 = const()[name = string("op_7744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7744_cast_fp16 = einsum(equation = var_7744_equation_0, values = (var_7416_cast_fp16, var_7655_cast_fp16))[name = string("op_7744_cast_fp16")]; + string var_7746_equation_0 = const()[name = string("op_7746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7746_cast_fp16 = einsum(equation = var_7746_equation_0, values = (var_7416_cast_fp16, var_7656_cast_fp16))[name = string("op_7746_cast_fp16")]; + string var_7748_equation_0 = const()[name = string("op_7748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7748_cast_fp16 = einsum(equation = var_7748_equation_0, values = (var_7416_cast_fp16, var_7657_cast_fp16))[name = string("op_7748_cast_fp16")]; + string var_7750_equation_0 = const()[name = string("op_7750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7750_cast_fp16 = einsum(equation = var_7750_equation_0, values = (var_7416_cast_fp16, var_7658_cast_fp16))[name = string("op_7750_cast_fp16")]; + string var_7752_equation_0 = const()[name = string("op_7752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7752_cast_fp16 = einsum(equation = var_7752_equation_0, values = (var_7420_cast_fp16, var_7659_cast_fp16))[name = string("op_7752_cast_fp16")]; + string var_7754_equation_0 = const()[name = string("op_7754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7754_cast_fp16 = einsum(equation = var_7754_equation_0, values = (var_7420_cast_fp16, var_7660_cast_fp16))[name = string("op_7754_cast_fp16")]; + string var_7756_equation_0 = const()[name = string("op_7756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7756_cast_fp16 = einsum(equation = var_7756_equation_0, values = (var_7420_cast_fp16, var_7661_cast_fp16))[name = string("op_7756_cast_fp16")]; + string var_7758_equation_0 = const()[name = string("op_7758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7758_cast_fp16 = einsum(equation = var_7758_equation_0, values = (var_7420_cast_fp16, var_7662_cast_fp16))[name = string("op_7758_cast_fp16")]; + bool var_7760_interleave_0 = const()[name = string("op_7760_interleave_0"), val = bool(false)]; + tensor var_7760_cast_fp16 = concat(axis = var_6871, interleave = var_7760_interleave_0, values = (var_7664_cast_fp16, var_7666_cast_fp16, var_7668_cast_fp16, var_7670_cast_fp16))[name = string("op_7760_cast_fp16")]; + bool var_7762_interleave_0 = const()[name = string("op_7762_interleave_0"), val = bool(false)]; + tensor var_7762_cast_fp16 = concat(axis = var_6871, interleave = var_7762_interleave_0, values = (var_7672_cast_fp16, var_7674_cast_fp16, var_7676_cast_fp16, var_7678_cast_fp16))[name = string("op_7762_cast_fp16")]; + bool var_7764_interleave_0 = const()[name = string("op_7764_interleave_0"), val = bool(false)]; + tensor var_7764_cast_fp16 = concat(axis = var_6871, interleave = var_7764_interleave_0, values = (var_7680_cast_fp16, var_7682_cast_fp16, var_7684_cast_fp16, var_7686_cast_fp16))[name = string("op_7764_cast_fp16")]; + bool var_7766_interleave_0 = const()[name = string("op_7766_interleave_0"), val = bool(false)]; + tensor var_7766_cast_fp16 = concat(axis = var_6871, interleave = var_7766_interleave_0, values = (var_7688_cast_fp16, var_7690_cast_fp16, var_7692_cast_fp16, var_7694_cast_fp16))[name = string("op_7766_cast_fp16")]; + bool var_7768_interleave_0 = const()[name = string("op_7768_interleave_0"), val = bool(false)]; + tensor var_7768_cast_fp16 = concat(axis = var_6871, interleave = var_7768_interleave_0, values = (var_7696_cast_fp16, var_7698_cast_fp16, var_7700_cast_fp16, var_7702_cast_fp16))[name = string("op_7768_cast_fp16")]; + bool var_7770_interleave_0 = const()[name = string("op_7770_interleave_0"), val = bool(false)]; + tensor var_7770_cast_fp16 = concat(axis = var_6871, interleave = var_7770_interleave_0, values = (var_7704_cast_fp16, var_7706_cast_fp16, var_7708_cast_fp16, var_7710_cast_fp16))[name = string("op_7770_cast_fp16")]; + bool var_7772_interleave_0 = const()[name = string("op_7772_interleave_0"), val = bool(false)]; + tensor var_7772_cast_fp16 = concat(axis = var_6871, interleave = var_7772_interleave_0, values = (var_7712_cast_fp16, var_7714_cast_fp16, var_7716_cast_fp16, var_7718_cast_fp16))[name = string("op_7772_cast_fp16")]; + bool var_7774_interleave_0 = const()[name = string("op_7774_interleave_0"), val = bool(false)]; + tensor var_7774_cast_fp16 = concat(axis = var_6871, interleave = var_7774_interleave_0, values = (var_7720_cast_fp16, var_7722_cast_fp16, var_7724_cast_fp16, var_7726_cast_fp16))[name = string("op_7774_cast_fp16")]; + bool var_7776_interleave_0 = const()[name = string("op_7776_interleave_0"), val = bool(false)]; + tensor var_7776_cast_fp16 = concat(axis = var_6871, interleave = var_7776_interleave_0, values = (var_7728_cast_fp16, var_7730_cast_fp16, var_7732_cast_fp16, var_7734_cast_fp16))[name = string("op_7776_cast_fp16")]; + bool var_7778_interleave_0 = const()[name = string("op_7778_interleave_0"), val = bool(false)]; + tensor var_7778_cast_fp16 = concat(axis = var_6871, interleave = var_7778_interleave_0, values = (var_7736_cast_fp16, var_7738_cast_fp16, var_7740_cast_fp16, var_7742_cast_fp16))[name = string("op_7778_cast_fp16")]; + bool var_7780_interleave_0 = const()[name = string("op_7780_interleave_0"), val = bool(false)]; + tensor var_7780_cast_fp16 = concat(axis = var_6871, interleave = var_7780_interleave_0, values = (var_7744_cast_fp16, var_7746_cast_fp16, var_7748_cast_fp16, var_7750_cast_fp16))[name = string("op_7780_cast_fp16")]; + bool var_7782_interleave_0 = const()[name = string("op_7782_interleave_0"), val = bool(false)]; + tensor var_7782_cast_fp16 = concat(axis = var_6871, interleave = var_7782_interleave_0, values = (var_7752_cast_fp16, var_7754_cast_fp16, var_7756_cast_fp16, var_7758_cast_fp16))[name = string("op_7782_cast_fp16")]; + bool input_57_interleave_0 = const()[name = string("input_57_interleave_0"), val = bool(false)]; + tensor input_57_cast_fp16 = concat(axis = var_6888, interleave = input_57_interleave_0, values = (var_7760_cast_fp16, var_7762_cast_fp16, var_7764_cast_fp16, var_7766_cast_fp16, var_7768_cast_fp16, var_7770_cast_fp16, var_7772_cast_fp16, var_7774_cast_fp16, var_7776_cast_fp16, var_7778_cast_fp16, var_7780_cast_fp16, var_7782_cast_fp16))[name = string("input_57_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108989952)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110169664)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_7801_to_fp16 = const()[name = string("op_7801_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_7801_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110171264)))]; + tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110172864)))]; + fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110174464)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114893120)))]; + tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; + tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114899328)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119617984)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; + int32 var_7830 = const()[name = string("op_7830"), val = int32(3)]; + int32 var_7847 = const()[name = string("op_7847"), val = int32(1)]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_7864_to_fp16 = const()[name = string("op_7864_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_7864_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119619584)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119621184)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119622784)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120802496)))]; + tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; + string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; + tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; + tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; + int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120804096)))]; + tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; + string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; + tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; + tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; + int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121983808)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123163520)))]; + tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_7902_begin_0 = const()[name = string("op_7902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7902_end_0 = const()[name = string("op_7902_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7902_end_mask_0 = const()[name = string("op_7902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = query_17_cast_fp16)[name = string("op_7902_cast_fp16")]; + tensor var_7906_begin_0 = const()[name = string("op_7906_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7906_end_0 = const()[name = string("op_7906_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7906_end_mask_0 = const()[name = string("op_7906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = query_17_cast_fp16)[name = string("op_7906_cast_fp16")]; + tensor var_7910_begin_0 = const()[name = string("op_7910_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7910_end_0 = const()[name = string("op_7910_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7910_end_mask_0 = const()[name = string("op_7910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = query_17_cast_fp16)[name = string("op_7910_cast_fp16")]; + tensor var_7914_begin_0 = const()[name = string("op_7914_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7914_end_0 = const()[name = string("op_7914_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7914_end_mask_0 = const()[name = string("op_7914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = query_17_cast_fp16)[name = string("op_7914_cast_fp16")]; + tensor var_7918_begin_0 = const()[name = string("op_7918_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7918_end_0 = const()[name = string("op_7918_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7918_end_mask_0 = const()[name = string("op_7918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7918_cast_fp16 = slice_by_index(begin = var_7918_begin_0, end = var_7918_end_0, end_mask = var_7918_end_mask_0, x = query_17_cast_fp16)[name = string("op_7918_cast_fp16")]; + tensor var_7922_begin_0 = const()[name = string("op_7922_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7922_end_0 = const()[name = string("op_7922_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7922_end_mask_0 = const()[name = string("op_7922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7922_cast_fp16 = slice_by_index(begin = var_7922_begin_0, end = var_7922_end_0, end_mask = var_7922_end_mask_0, x = query_17_cast_fp16)[name = string("op_7922_cast_fp16")]; + tensor var_7926_begin_0 = const()[name = string("op_7926_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7926_end_0 = const()[name = string("op_7926_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7926_end_mask_0 = const()[name = string("op_7926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7926_cast_fp16 = slice_by_index(begin = var_7926_begin_0, end = var_7926_end_0, end_mask = var_7926_end_mask_0, x = query_17_cast_fp16)[name = string("op_7926_cast_fp16")]; + tensor var_7930_begin_0 = const()[name = string("op_7930_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7930_end_0 = const()[name = string("op_7930_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7930_end_mask_0 = const()[name = string("op_7930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7930_cast_fp16 = slice_by_index(begin = var_7930_begin_0, end = var_7930_end_0, end_mask = var_7930_end_mask_0, x = query_17_cast_fp16)[name = string("op_7930_cast_fp16")]; + tensor var_7934_begin_0 = const()[name = string("op_7934_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7934_end_0 = const()[name = string("op_7934_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7934_end_mask_0 = const()[name = string("op_7934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7934_cast_fp16 = slice_by_index(begin = var_7934_begin_0, end = var_7934_end_0, end_mask = var_7934_end_mask_0, x = query_17_cast_fp16)[name = string("op_7934_cast_fp16")]; + tensor var_7938_begin_0 = const()[name = string("op_7938_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7938_end_0 = const()[name = string("op_7938_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7938_end_mask_0 = const()[name = string("op_7938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7938_cast_fp16 = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = query_17_cast_fp16)[name = string("op_7938_cast_fp16")]; + tensor var_7942_begin_0 = const()[name = string("op_7942_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7942_end_0 = const()[name = string("op_7942_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7942_end_mask_0 = const()[name = string("op_7942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7942_cast_fp16 = slice_by_index(begin = var_7942_begin_0, end = var_7942_end_0, end_mask = var_7942_end_mask_0, x = query_17_cast_fp16)[name = string("op_7942_cast_fp16")]; + tensor var_7946_begin_0 = const()[name = string("op_7946_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7946_end_0 = const()[name = string("op_7946_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7946_end_mask_0 = const()[name = string("op_7946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7946_cast_fp16 = slice_by_index(begin = var_7946_begin_0, end = var_7946_end_0, end_mask = var_7946_end_mask_0, x = query_17_cast_fp16)[name = string("op_7946_cast_fp16")]; + tensor var_7955_begin_0 = const()[name = string("op_7955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7955_end_0 = const()[name = string("op_7955_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7955_end_mask_0 = const()[name = string("op_7955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7955_cast_fp16 = slice_by_index(begin = var_7955_begin_0, end = var_7955_end_0, end_mask = var_7955_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7955_cast_fp16")]; + tensor var_7962_begin_0 = const()[name = string("op_7962_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7962_end_0 = const()[name = string("op_7962_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7962_end_mask_0 = const()[name = string("op_7962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7962_cast_fp16 = slice_by_index(begin = var_7962_begin_0, end = var_7962_end_0, end_mask = var_7962_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7962_cast_fp16")]; + tensor var_7969_begin_0 = const()[name = string("op_7969_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7969_end_0 = const()[name = string("op_7969_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7969_end_mask_0 = const()[name = string("op_7969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7969_cast_fp16 = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7969_cast_fp16")]; + tensor var_7976_begin_0 = const()[name = string("op_7976_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7976_end_0 = const()[name = string("op_7976_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7976_end_mask_0 = const()[name = string("op_7976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7976_cast_fp16 = slice_by_index(begin = var_7976_begin_0, end = var_7976_end_0, end_mask = var_7976_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7976_cast_fp16")]; + tensor var_7983_begin_0 = const()[name = string("op_7983_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7983_end_0 = const()[name = string("op_7983_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7983_end_mask_0 = const()[name = string("op_7983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7983_cast_fp16 = slice_by_index(begin = var_7983_begin_0, end = var_7983_end_0, end_mask = var_7983_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7983_cast_fp16")]; + tensor var_7990_begin_0 = const()[name = string("op_7990_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7990_end_0 = const()[name = string("op_7990_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7990_end_mask_0 = const()[name = string("op_7990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7990_cast_fp16 = slice_by_index(begin = var_7990_begin_0, end = var_7990_end_0, end_mask = var_7990_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7990_cast_fp16")]; + tensor var_7997_begin_0 = const()[name = string("op_7997_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7997_end_0 = const()[name = string("op_7997_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7997_end_mask_0 = const()[name = string("op_7997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7997_cast_fp16 = slice_by_index(begin = var_7997_begin_0, end = var_7997_end_0, end_mask = var_7997_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7997_cast_fp16")]; + tensor var_8004_begin_0 = const()[name = string("op_8004_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8004_end_0 = const()[name = string("op_8004_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8004_end_mask_0 = const()[name = string("op_8004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8004_cast_fp16 = slice_by_index(begin = var_8004_begin_0, end = var_8004_end_0, end_mask = var_8004_end_mask_0, x = var_7906_cast_fp16)[name = string("op_8004_cast_fp16")]; + tensor var_8011_begin_0 = const()[name = string("op_8011_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8011_end_0 = const()[name = string("op_8011_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8011_end_mask_0 = const()[name = string("op_8011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8011_cast_fp16 = slice_by_index(begin = var_8011_begin_0, end = var_8011_end_0, end_mask = var_8011_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8011_cast_fp16")]; + tensor var_8018_begin_0 = const()[name = string("op_8018_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8018_end_0 = const()[name = string("op_8018_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8018_end_mask_0 = const()[name = string("op_8018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8018_cast_fp16 = slice_by_index(begin = var_8018_begin_0, end = var_8018_end_0, end_mask = var_8018_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8018_cast_fp16")]; + tensor var_8025_begin_0 = const()[name = string("op_8025_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8025_end_0 = const()[name = string("op_8025_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8025_end_mask_0 = const()[name = string("op_8025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8025_cast_fp16 = slice_by_index(begin = var_8025_begin_0, end = var_8025_end_0, end_mask = var_8025_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8025_cast_fp16")]; + tensor var_8032_begin_0 = const()[name = string("op_8032_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8032_end_0 = const()[name = string("op_8032_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8032_end_mask_0 = const()[name = string("op_8032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8032_cast_fp16 = slice_by_index(begin = var_8032_begin_0, end = var_8032_end_0, end_mask = var_8032_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8032_cast_fp16")]; + tensor var_8039_begin_0 = const()[name = string("op_8039_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8039_end_0 = const()[name = string("op_8039_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8039_end_mask_0 = const()[name = string("op_8039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8039_cast_fp16 = slice_by_index(begin = var_8039_begin_0, end = var_8039_end_0, end_mask = var_8039_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8039_cast_fp16")]; + tensor var_8046_begin_0 = const()[name = string("op_8046_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8046_end_0 = const()[name = string("op_8046_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8046_end_mask_0 = const()[name = string("op_8046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8046_cast_fp16 = slice_by_index(begin = var_8046_begin_0, end = var_8046_end_0, end_mask = var_8046_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8046_cast_fp16")]; + tensor var_8053_begin_0 = const()[name = string("op_8053_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8053_end_0 = const()[name = string("op_8053_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8053_end_mask_0 = const()[name = string("op_8053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8053_cast_fp16 = slice_by_index(begin = var_8053_begin_0, end = var_8053_end_0, end_mask = var_8053_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8053_cast_fp16")]; + tensor var_8060_begin_0 = const()[name = string("op_8060_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8060_end_0 = const()[name = string("op_8060_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8060_end_mask_0 = const()[name = string("op_8060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8060_cast_fp16 = slice_by_index(begin = var_8060_begin_0, end = var_8060_end_0, end_mask = var_8060_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8060_cast_fp16")]; + tensor var_8067_begin_0 = const()[name = string("op_8067_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8067_end_0 = const()[name = string("op_8067_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8067_end_mask_0 = const()[name = string("op_8067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8067_cast_fp16 = slice_by_index(begin = var_8067_begin_0, end = var_8067_end_0, end_mask = var_8067_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8067_cast_fp16")]; + tensor var_8074_begin_0 = const()[name = string("op_8074_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8074_end_0 = const()[name = string("op_8074_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8074_end_mask_0 = const()[name = string("op_8074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8074_cast_fp16 = slice_by_index(begin = var_8074_begin_0, end = var_8074_end_0, end_mask = var_8074_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8074_cast_fp16")]; + tensor var_8081_begin_0 = const()[name = string("op_8081_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8081_end_0 = const()[name = string("op_8081_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8081_end_mask_0 = const()[name = string("op_8081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8081_cast_fp16 = slice_by_index(begin = var_8081_begin_0, end = var_8081_end_0, end_mask = var_8081_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8081_cast_fp16")]; + tensor var_8088_begin_0 = const()[name = string("op_8088_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8088_end_0 = const()[name = string("op_8088_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8088_end_mask_0 = const()[name = string("op_8088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8088_cast_fp16 = slice_by_index(begin = var_8088_begin_0, end = var_8088_end_0, end_mask = var_8088_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8088_cast_fp16")]; + tensor var_8095_begin_0 = const()[name = string("op_8095_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8095_end_0 = const()[name = string("op_8095_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8095_end_mask_0 = const()[name = string("op_8095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8095_cast_fp16 = slice_by_index(begin = var_8095_begin_0, end = var_8095_end_0, end_mask = var_8095_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8095_cast_fp16")]; + tensor var_8102_begin_0 = const()[name = string("op_8102_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8102_end_0 = const()[name = string("op_8102_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8102_end_mask_0 = const()[name = string("op_8102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8102_cast_fp16 = slice_by_index(begin = var_8102_begin_0, end = var_8102_end_0, end_mask = var_8102_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8102_cast_fp16")]; + tensor var_8109_begin_0 = const()[name = string("op_8109_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8109_end_0 = const()[name = string("op_8109_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8109_end_mask_0 = const()[name = string("op_8109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8109_cast_fp16 = slice_by_index(begin = var_8109_begin_0, end = var_8109_end_0, end_mask = var_8109_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8109_cast_fp16")]; + tensor var_8116_begin_0 = const()[name = string("op_8116_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8116_end_0 = const()[name = string("op_8116_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8116_end_mask_0 = const()[name = string("op_8116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8116_cast_fp16 = slice_by_index(begin = var_8116_begin_0, end = var_8116_end_0, end_mask = var_8116_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8116_cast_fp16")]; + tensor var_8123_begin_0 = const()[name = string("op_8123_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8123_end_0 = const()[name = string("op_8123_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8123_end_mask_0 = const()[name = string("op_8123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8123_cast_fp16 = slice_by_index(begin = var_8123_begin_0, end = var_8123_end_0, end_mask = var_8123_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8123_cast_fp16")]; + tensor var_8130_begin_0 = const()[name = string("op_8130_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8130_end_0 = const()[name = string("op_8130_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8130_end_mask_0 = const()[name = string("op_8130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8130_cast_fp16 = slice_by_index(begin = var_8130_begin_0, end = var_8130_end_0, end_mask = var_8130_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8130_cast_fp16")]; + tensor var_8137_begin_0 = const()[name = string("op_8137_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8137_end_0 = const()[name = string("op_8137_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8137_end_mask_0 = const()[name = string("op_8137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8137_cast_fp16 = slice_by_index(begin = var_8137_begin_0, end = var_8137_end_0, end_mask = var_8137_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8137_cast_fp16")]; + tensor var_8144_begin_0 = const()[name = string("op_8144_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8144_end_0 = const()[name = string("op_8144_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8144_end_mask_0 = const()[name = string("op_8144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8144_cast_fp16 = slice_by_index(begin = var_8144_begin_0, end = var_8144_end_0, end_mask = var_8144_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8144_cast_fp16")]; + tensor var_8151_begin_0 = const()[name = string("op_8151_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8151_end_0 = const()[name = string("op_8151_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8151_end_mask_0 = const()[name = string("op_8151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8151_cast_fp16 = slice_by_index(begin = var_8151_begin_0, end = var_8151_end_0, end_mask = var_8151_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8151_cast_fp16")]; + tensor var_8158_begin_0 = const()[name = string("op_8158_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8158_end_0 = const()[name = string("op_8158_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8158_end_mask_0 = const()[name = string("op_8158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8158_cast_fp16 = slice_by_index(begin = var_8158_begin_0, end = var_8158_end_0, end_mask = var_8158_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8158_cast_fp16")]; + tensor var_8165_begin_0 = const()[name = string("op_8165_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8165_end_0 = const()[name = string("op_8165_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8165_end_mask_0 = const()[name = string("op_8165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8165_cast_fp16 = slice_by_index(begin = var_8165_begin_0, end = var_8165_end_0, end_mask = var_8165_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8165_cast_fp16")]; + tensor var_8172_begin_0 = const()[name = string("op_8172_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8172_end_0 = const()[name = string("op_8172_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8172_end_mask_0 = const()[name = string("op_8172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8172_cast_fp16 = slice_by_index(begin = var_8172_begin_0, end = var_8172_end_0, end_mask = var_8172_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8172_cast_fp16")]; + tensor var_8179_begin_0 = const()[name = string("op_8179_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8179_end_0 = const()[name = string("op_8179_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8179_end_mask_0 = const()[name = string("op_8179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8179_cast_fp16 = slice_by_index(begin = var_8179_begin_0, end = var_8179_end_0, end_mask = var_8179_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8179_cast_fp16")]; + tensor var_8186_begin_0 = const()[name = string("op_8186_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8186_end_0 = const()[name = string("op_8186_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8186_end_mask_0 = const()[name = string("op_8186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8186_cast_fp16 = slice_by_index(begin = var_8186_begin_0, end = var_8186_end_0, end_mask = var_8186_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8186_cast_fp16")]; + tensor var_8193_begin_0 = const()[name = string("op_8193_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8193_end_0 = const()[name = string("op_8193_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8193_end_mask_0 = const()[name = string("op_8193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8193_cast_fp16 = slice_by_index(begin = var_8193_begin_0, end = var_8193_end_0, end_mask = var_8193_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8193_cast_fp16")]; + tensor var_8200_begin_0 = const()[name = string("op_8200_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8200_end_0 = const()[name = string("op_8200_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8200_end_mask_0 = const()[name = string("op_8200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8200_cast_fp16 = slice_by_index(begin = var_8200_begin_0, end = var_8200_end_0, end_mask = var_8200_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8200_cast_fp16")]; + tensor var_8207_begin_0 = const()[name = string("op_8207_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8207_end_0 = const()[name = string("op_8207_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8207_end_mask_0 = const()[name = string("op_8207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8207_cast_fp16 = slice_by_index(begin = var_8207_begin_0, end = var_8207_end_0, end_mask = var_8207_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8207_cast_fp16")]; + tensor var_8214_begin_0 = const()[name = string("op_8214_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8214_end_0 = const()[name = string("op_8214_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8214_end_mask_0 = const()[name = string("op_8214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8214_cast_fp16 = slice_by_index(begin = var_8214_begin_0, end = var_8214_end_0, end_mask = var_8214_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8214_cast_fp16")]; + tensor var_8221_begin_0 = const()[name = string("op_8221_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8221_end_0 = const()[name = string("op_8221_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8221_end_mask_0 = const()[name = string("op_8221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8221_cast_fp16 = slice_by_index(begin = var_8221_begin_0, end = var_8221_end_0, end_mask = var_8221_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8221_cast_fp16")]; + tensor var_8228_begin_0 = const()[name = string("op_8228_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8228_end_0 = const()[name = string("op_8228_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8228_end_mask_0 = const()[name = string("op_8228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8228_cast_fp16 = slice_by_index(begin = var_8228_begin_0, end = var_8228_end_0, end_mask = var_8228_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8228_cast_fp16")]; + tensor var_8235_begin_0 = const()[name = string("op_8235_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8235_end_0 = const()[name = string("op_8235_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8235_end_mask_0 = const()[name = string("op_8235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8235_cast_fp16 = slice_by_index(begin = var_8235_begin_0, end = var_8235_end_0, end_mask = var_8235_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8235_cast_fp16")]; + tensor var_8242_begin_0 = const()[name = string("op_8242_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8242_end_0 = const()[name = string("op_8242_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8242_end_mask_0 = const()[name = string("op_8242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8242_cast_fp16 = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8242_cast_fp16")]; + tensor var_8249_begin_0 = const()[name = string("op_8249_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8249_end_0 = const()[name = string("op_8249_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8249_end_mask_0 = const()[name = string("op_8249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8249_cast_fp16 = slice_by_index(begin = var_8249_begin_0, end = var_8249_end_0, end_mask = var_8249_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8249_cast_fp16")]; + tensor var_8256_begin_0 = const()[name = string("op_8256_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8256_end_0 = const()[name = string("op_8256_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8256_end_mask_0 = const()[name = string("op_8256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8256_cast_fp16 = slice_by_index(begin = var_8256_begin_0, end = var_8256_end_0, end_mask = var_8256_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8256_cast_fp16")]; + tensor var_8263_begin_0 = const()[name = string("op_8263_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8263_end_0 = const()[name = string("op_8263_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8263_end_mask_0 = const()[name = string("op_8263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8263_cast_fp16 = slice_by_index(begin = var_8263_begin_0, end = var_8263_end_0, end_mask = var_8263_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8263_cast_fp16")]; + tensor var_8270_begin_0 = const()[name = string("op_8270_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8270_end_0 = const()[name = string("op_8270_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8270_end_mask_0 = const()[name = string("op_8270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8270_cast_fp16 = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8270_cast_fp16")]; + tensor var_8277_begin_0 = const()[name = string("op_8277_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8277_end_0 = const()[name = string("op_8277_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8277_end_mask_0 = const()[name = string("op_8277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8277_cast_fp16 = slice_by_index(begin = var_8277_begin_0, end = var_8277_end_0, end_mask = var_8277_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8277_cast_fp16")]; + tensor var_8284_begin_0 = const()[name = string("op_8284_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8284_end_0 = const()[name = string("op_8284_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8284_end_mask_0 = const()[name = string("op_8284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8284_cast_fp16 = slice_by_index(begin = var_8284_begin_0, end = var_8284_end_0, end_mask = var_8284_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8284_cast_fp16")]; + tensor k_17_perm_0 = const()[name = string("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8289_begin_0 = const()[name = string("op_8289_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8289_end_0 = const()[name = string("op_8289_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_8289_end_mask_0 = const()[name = string("op_8289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_17_cast_fp16 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = string("transpose_3")]; + tensor var_8289_cast_fp16 = slice_by_index(begin = var_8289_begin_0, end = var_8289_end_0, end_mask = var_8289_end_mask_0, x = k_17_cast_fp16)[name = string("op_8289_cast_fp16")]; + tensor var_8293_begin_0 = const()[name = string("op_8293_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8293_end_0 = const()[name = string("op_8293_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_8293_end_mask_0 = const()[name = string("op_8293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8293_cast_fp16 = slice_by_index(begin = var_8293_begin_0, end = var_8293_end_0, end_mask = var_8293_end_mask_0, x = k_17_cast_fp16)[name = string("op_8293_cast_fp16")]; + tensor var_8297_begin_0 = const()[name = string("op_8297_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8297_end_0 = const()[name = string("op_8297_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_8297_end_mask_0 = const()[name = string("op_8297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8297_cast_fp16 = slice_by_index(begin = var_8297_begin_0, end = var_8297_end_0, end_mask = var_8297_end_mask_0, x = k_17_cast_fp16)[name = string("op_8297_cast_fp16")]; + tensor var_8301_begin_0 = const()[name = string("op_8301_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8301_end_0 = const()[name = string("op_8301_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_8301_end_mask_0 = const()[name = string("op_8301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8301_cast_fp16 = slice_by_index(begin = var_8301_begin_0, end = var_8301_end_0, end_mask = var_8301_end_mask_0, x = k_17_cast_fp16)[name = string("op_8301_cast_fp16")]; + tensor var_8305_begin_0 = const()[name = string("op_8305_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8305_end_0 = const()[name = string("op_8305_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_8305_end_mask_0 = const()[name = string("op_8305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8305_cast_fp16 = slice_by_index(begin = var_8305_begin_0, end = var_8305_end_0, end_mask = var_8305_end_mask_0, x = k_17_cast_fp16)[name = string("op_8305_cast_fp16")]; + tensor var_8309_begin_0 = const()[name = string("op_8309_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8309_end_0 = const()[name = string("op_8309_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_8309_end_mask_0 = const()[name = string("op_8309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8309_cast_fp16 = slice_by_index(begin = var_8309_begin_0, end = var_8309_end_0, end_mask = var_8309_end_mask_0, x = k_17_cast_fp16)[name = string("op_8309_cast_fp16")]; + tensor var_8313_begin_0 = const()[name = string("op_8313_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8313_end_0 = const()[name = string("op_8313_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_8313_end_mask_0 = const()[name = string("op_8313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8313_cast_fp16 = slice_by_index(begin = var_8313_begin_0, end = var_8313_end_0, end_mask = var_8313_end_mask_0, x = k_17_cast_fp16)[name = string("op_8313_cast_fp16")]; + tensor var_8317_begin_0 = const()[name = string("op_8317_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8317_end_0 = const()[name = string("op_8317_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_8317_end_mask_0 = const()[name = string("op_8317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8317_cast_fp16 = slice_by_index(begin = var_8317_begin_0, end = var_8317_end_0, end_mask = var_8317_end_mask_0, x = k_17_cast_fp16)[name = string("op_8317_cast_fp16")]; + tensor var_8321_begin_0 = const()[name = string("op_8321_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8321_end_0 = const()[name = string("op_8321_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_8321_end_mask_0 = const()[name = string("op_8321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8321_cast_fp16 = slice_by_index(begin = var_8321_begin_0, end = var_8321_end_0, end_mask = var_8321_end_mask_0, x = k_17_cast_fp16)[name = string("op_8321_cast_fp16")]; + tensor var_8325_begin_0 = const()[name = string("op_8325_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8325_end_0 = const()[name = string("op_8325_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_8325_end_mask_0 = const()[name = string("op_8325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8325_cast_fp16 = slice_by_index(begin = var_8325_begin_0, end = var_8325_end_0, end_mask = var_8325_end_mask_0, x = k_17_cast_fp16)[name = string("op_8325_cast_fp16")]; + tensor var_8329_begin_0 = const()[name = string("op_8329_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8329_end_0 = const()[name = string("op_8329_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_8329_end_mask_0 = const()[name = string("op_8329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8329_cast_fp16 = slice_by_index(begin = var_8329_begin_0, end = var_8329_end_0, end_mask = var_8329_end_mask_0, x = k_17_cast_fp16)[name = string("op_8329_cast_fp16")]; + tensor var_8333_begin_0 = const()[name = string("op_8333_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8333_end_0 = const()[name = string("op_8333_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_8333_end_mask_0 = const()[name = string("op_8333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8333_cast_fp16 = slice_by_index(begin = var_8333_begin_0, end = var_8333_end_0, end_mask = var_8333_end_mask_0, x = k_17_cast_fp16)[name = string("op_8333_cast_fp16")]; + tensor var_8335_begin_0 = const()[name = string("op_8335_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8335_end_0 = const()[name = string("op_8335_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8335_end_mask_0 = const()[name = string("op_8335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8335_cast_fp16 = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = value_17_cast_fp16)[name = string("op_8335_cast_fp16")]; + tensor var_8339_begin_0 = const()[name = string("op_8339_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8339_end_0 = const()[name = string("op_8339_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8339_end_mask_0 = const()[name = string("op_8339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8339_cast_fp16 = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = value_17_cast_fp16)[name = string("op_8339_cast_fp16")]; + tensor var_8343_begin_0 = const()[name = string("op_8343_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8343_end_0 = const()[name = string("op_8343_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8343_end_mask_0 = const()[name = string("op_8343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8343_cast_fp16 = slice_by_index(begin = var_8343_begin_0, end = var_8343_end_0, end_mask = var_8343_end_mask_0, x = value_17_cast_fp16)[name = string("op_8343_cast_fp16")]; + tensor var_8347_begin_0 = const()[name = string("op_8347_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8347_end_0 = const()[name = string("op_8347_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8347_end_mask_0 = const()[name = string("op_8347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8347_cast_fp16 = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = value_17_cast_fp16)[name = string("op_8347_cast_fp16")]; + tensor var_8351_begin_0 = const()[name = string("op_8351_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8351_end_0 = const()[name = string("op_8351_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8351_end_mask_0 = const()[name = string("op_8351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8351_cast_fp16 = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = value_17_cast_fp16)[name = string("op_8351_cast_fp16")]; + tensor var_8355_begin_0 = const()[name = string("op_8355_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8355_end_0 = const()[name = string("op_8355_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8355_end_mask_0 = const()[name = string("op_8355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8355_cast_fp16 = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = value_17_cast_fp16)[name = string("op_8355_cast_fp16")]; + tensor var_8359_begin_0 = const()[name = string("op_8359_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8359_end_0 = const()[name = string("op_8359_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8359_end_mask_0 = const()[name = string("op_8359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8359_cast_fp16 = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = value_17_cast_fp16)[name = string("op_8359_cast_fp16")]; + tensor var_8363_begin_0 = const()[name = string("op_8363_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8363_end_0 = const()[name = string("op_8363_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8363_end_mask_0 = const()[name = string("op_8363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8363_cast_fp16 = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = value_17_cast_fp16)[name = string("op_8363_cast_fp16")]; + tensor var_8367_begin_0 = const()[name = string("op_8367_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8367_end_0 = const()[name = string("op_8367_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8367_end_mask_0 = const()[name = string("op_8367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8367_cast_fp16 = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = value_17_cast_fp16)[name = string("op_8367_cast_fp16")]; + tensor var_8371_begin_0 = const()[name = string("op_8371_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8371_end_0 = const()[name = string("op_8371_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8371_end_mask_0 = const()[name = string("op_8371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8371_cast_fp16 = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = value_17_cast_fp16)[name = string("op_8371_cast_fp16")]; + tensor var_8375_begin_0 = const()[name = string("op_8375_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8375_end_0 = const()[name = string("op_8375_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8375_end_mask_0 = const()[name = string("op_8375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8375_cast_fp16 = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = value_17_cast_fp16)[name = string("op_8375_cast_fp16")]; + tensor var_8379_begin_0 = const()[name = string("op_8379_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8379_end_0 = const()[name = string("op_8379_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8379_end_mask_0 = const()[name = string("op_8379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8379_cast_fp16 = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = value_17_cast_fp16)[name = string("op_8379_cast_fp16")]; + string _SplitHeadsQ__mh_w_769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_769_equation_0, values = (var_8289_cast_fp16, var_7955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_769_cast_fp16")]; + string _SplitHeadsQ__mh_w_771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_771_equation_0, values = (var_8289_cast_fp16, var_7962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_771_cast_fp16")]; + string _SplitHeadsQ__mh_w_773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_773_equation_0, values = (var_8289_cast_fp16, var_7969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_773_cast_fp16")]; + string _SplitHeadsQ__mh_w_775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_775_equation_0, values = (var_8289_cast_fp16, var_7976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_775_cast_fp16")]; + string _SplitHeadsQ__mh_w_777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_777_equation_0, values = (var_8293_cast_fp16, var_7983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_777_cast_fp16")]; + string _SplitHeadsQ__mh_w_779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_779_equation_0, values = (var_8293_cast_fp16, var_7990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_779_cast_fp16")]; + string _SplitHeadsQ__mh_w_781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_781_equation_0, values = (var_8293_cast_fp16, var_7997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_781_cast_fp16")]; + string _SplitHeadsQ__mh_w_783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_783_equation_0, values = (var_8293_cast_fp16, var_8004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_783_cast_fp16")]; + string _SplitHeadsQ__mh_w_785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_785_equation_0, values = (var_8297_cast_fp16, var_8011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_785_cast_fp16")]; + string _SplitHeadsQ__mh_w_787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_787_equation_0, values = (var_8297_cast_fp16, var_8018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_787_cast_fp16")]; + string _SplitHeadsQ__mh_w_789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_789_equation_0, values = (var_8297_cast_fp16, var_8025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_789_cast_fp16")]; + string _SplitHeadsQ__mh_w_791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_791_equation_0, values = (var_8297_cast_fp16, var_8032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_791_cast_fp16")]; + string _SplitHeadsQ__mh_w_793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_793_equation_0, values = (var_8301_cast_fp16, var_8039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_793_cast_fp16")]; + string _SplitHeadsQ__mh_w_795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_795_equation_0, values = (var_8301_cast_fp16, var_8046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_795_cast_fp16")]; + string _SplitHeadsQ__mh_w_797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_797_equation_0, values = (var_8301_cast_fp16, var_8053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_797_cast_fp16")]; + string _SplitHeadsQ__mh_w_799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_799_equation_0, values = (var_8301_cast_fp16, var_8060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_799_cast_fp16")]; + string _SplitHeadsQ__mh_w_801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_801_equation_0, values = (var_8305_cast_fp16, var_8067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_801_cast_fp16")]; + string _SplitHeadsQ__mh_w_803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_803_equation_0, values = (var_8305_cast_fp16, var_8074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_803_cast_fp16")]; + string _SplitHeadsQ__mh_w_805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_805_equation_0, values = (var_8305_cast_fp16, var_8081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_805_cast_fp16")]; + string _SplitHeadsQ__mh_w_807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_807_equation_0, values = (var_8305_cast_fp16, var_8088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_807_cast_fp16")]; + string _SplitHeadsQ__mh_w_809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_809_equation_0, values = (var_8309_cast_fp16, var_8095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_809_cast_fp16")]; + string _SplitHeadsQ__mh_w_811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_811_equation_0, values = (var_8309_cast_fp16, var_8102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_811_cast_fp16")]; + string _SplitHeadsQ__mh_w_813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_813_equation_0, values = (var_8309_cast_fp16, var_8109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_813_cast_fp16")]; + string _SplitHeadsQ__mh_w_815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_815_equation_0, values = (var_8309_cast_fp16, var_8116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_815_cast_fp16")]; + string _SplitHeadsQ__mh_w_817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_817_equation_0, values = (var_8313_cast_fp16, var_8123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_817_cast_fp16")]; + string _SplitHeadsQ__mh_w_819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_819_equation_0, values = (var_8313_cast_fp16, var_8130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_819_cast_fp16")]; + string _SplitHeadsQ__mh_w_821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_821_equation_0, values = (var_8313_cast_fp16, var_8137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_821_cast_fp16")]; + string _SplitHeadsQ__mh_w_823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_823_equation_0, values = (var_8313_cast_fp16, var_8144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_823_cast_fp16")]; + string _SplitHeadsQ__mh_w_825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_825_equation_0, values = (var_8317_cast_fp16, var_8151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_825_cast_fp16")]; + string _SplitHeadsQ__mh_w_827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_827_equation_0, values = (var_8317_cast_fp16, var_8158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_827_cast_fp16")]; + string _SplitHeadsQ__mh_w_829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_829_equation_0, values = (var_8317_cast_fp16, var_8165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_829_cast_fp16")]; + string _SplitHeadsQ__mh_w_831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_831_equation_0, values = (var_8317_cast_fp16, var_8172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_831_cast_fp16")]; + string _SplitHeadsQ__mh_w_833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_833_equation_0, values = (var_8321_cast_fp16, var_8179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_833_cast_fp16")]; + string _SplitHeadsQ__mh_w_835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_835_equation_0, values = (var_8321_cast_fp16, var_8186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_835_cast_fp16")]; + string _SplitHeadsQ__mh_w_837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_837_equation_0, values = (var_8321_cast_fp16, var_8193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_837_cast_fp16")]; + string _SplitHeadsQ__mh_w_839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_839_equation_0, values = (var_8321_cast_fp16, var_8200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_839_cast_fp16")]; + string _SplitHeadsQ__mh_w_841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_841_equation_0, values = (var_8325_cast_fp16, var_8207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_841_cast_fp16")]; + string _SplitHeadsQ__mh_w_843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_843_equation_0, values = (var_8325_cast_fp16, var_8214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_843_cast_fp16")]; + string _SplitHeadsQ__mh_w_845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_845_equation_0, values = (var_8325_cast_fp16, var_8221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_845_cast_fp16")]; + string _SplitHeadsQ__mh_w_847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_847_equation_0, values = (var_8325_cast_fp16, var_8228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_847_cast_fp16")]; + string _SplitHeadsQ__mh_w_849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_849_equation_0, values = (var_8329_cast_fp16, var_8235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_849_cast_fp16")]; + string _SplitHeadsQ__mh_w_851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_851_equation_0, values = (var_8329_cast_fp16, var_8242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_851_cast_fp16")]; + string _SplitHeadsQ__mh_w_853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_853_equation_0, values = (var_8329_cast_fp16, var_8249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_853_cast_fp16")]; + string _SplitHeadsQ__mh_w_855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_855_equation_0, values = (var_8329_cast_fp16, var_8256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_855_cast_fp16")]; + string _SplitHeadsQ__mh_w_857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_857_equation_0, values = (var_8333_cast_fp16, var_8263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_857_cast_fp16")]; + string _SplitHeadsQ__mh_w_859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_859_equation_0, values = (var_8333_cast_fp16, var_8270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_859_cast_fp16")]; + string _SplitHeadsQ__mh_w_861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_861_equation_0, values = (var_8333_cast_fp16, var_8277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_861_cast_fp16")]; + string _SplitHeadsQ__mh_w_863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_863_equation_0, values = (var_8333_cast_fp16, var_8284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_863_cast_fp16")]; + fp16 var_8478_to_fp16 = const()[name = string("op_8478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_769_cast_fp16, y = var_8478_to_fp16)[name = string("aw_chunk_769_cast_fp16")]; + fp16 var_8480_to_fp16 = const()[name = string("op_8480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_771_cast_fp16, y = var_8480_to_fp16)[name = string("aw_chunk_771_cast_fp16")]; + fp16 var_8482_to_fp16 = const()[name = string("op_8482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_773_cast_fp16, y = var_8482_to_fp16)[name = string("aw_chunk_773_cast_fp16")]; + fp16 var_8484_to_fp16 = const()[name = string("op_8484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_775_cast_fp16, y = var_8484_to_fp16)[name = string("aw_chunk_775_cast_fp16")]; + fp16 var_8486_to_fp16 = const()[name = string("op_8486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_777_cast_fp16, y = var_8486_to_fp16)[name = string("aw_chunk_777_cast_fp16")]; + fp16 var_8488_to_fp16 = const()[name = string("op_8488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_779_cast_fp16, y = var_8488_to_fp16)[name = string("aw_chunk_779_cast_fp16")]; + fp16 var_8490_to_fp16 = const()[name = string("op_8490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_781_cast_fp16, y = var_8490_to_fp16)[name = string("aw_chunk_781_cast_fp16")]; + fp16 var_8492_to_fp16 = const()[name = string("op_8492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_783_cast_fp16, y = var_8492_to_fp16)[name = string("aw_chunk_783_cast_fp16")]; + fp16 var_8494_to_fp16 = const()[name = string("op_8494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_785_cast_fp16, y = var_8494_to_fp16)[name = string("aw_chunk_785_cast_fp16")]; + fp16 var_8496_to_fp16 = const()[name = string("op_8496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_787_cast_fp16, y = var_8496_to_fp16)[name = string("aw_chunk_787_cast_fp16")]; + fp16 var_8498_to_fp16 = const()[name = string("op_8498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_789_cast_fp16, y = var_8498_to_fp16)[name = string("aw_chunk_789_cast_fp16")]; + fp16 var_8500_to_fp16 = const()[name = string("op_8500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_791_cast_fp16, y = var_8500_to_fp16)[name = string("aw_chunk_791_cast_fp16")]; + fp16 var_8502_to_fp16 = const()[name = string("op_8502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_793_cast_fp16, y = var_8502_to_fp16)[name = string("aw_chunk_793_cast_fp16")]; + fp16 var_8504_to_fp16 = const()[name = string("op_8504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_795_cast_fp16, y = var_8504_to_fp16)[name = string("aw_chunk_795_cast_fp16")]; + fp16 var_8506_to_fp16 = const()[name = string("op_8506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_797_cast_fp16, y = var_8506_to_fp16)[name = string("aw_chunk_797_cast_fp16")]; + fp16 var_8508_to_fp16 = const()[name = string("op_8508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_799_cast_fp16, y = var_8508_to_fp16)[name = string("aw_chunk_799_cast_fp16")]; + fp16 var_8510_to_fp16 = const()[name = string("op_8510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_801_cast_fp16, y = var_8510_to_fp16)[name = string("aw_chunk_801_cast_fp16")]; + fp16 var_8512_to_fp16 = const()[name = string("op_8512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_803_cast_fp16, y = var_8512_to_fp16)[name = string("aw_chunk_803_cast_fp16")]; + fp16 var_8514_to_fp16 = const()[name = string("op_8514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_805_cast_fp16, y = var_8514_to_fp16)[name = string("aw_chunk_805_cast_fp16")]; + fp16 var_8516_to_fp16 = const()[name = string("op_8516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_807_cast_fp16, y = var_8516_to_fp16)[name = string("aw_chunk_807_cast_fp16")]; + fp16 var_8518_to_fp16 = const()[name = string("op_8518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_809_cast_fp16, y = var_8518_to_fp16)[name = string("aw_chunk_809_cast_fp16")]; + fp16 var_8520_to_fp16 = const()[name = string("op_8520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_811_cast_fp16, y = var_8520_to_fp16)[name = string("aw_chunk_811_cast_fp16")]; + fp16 var_8522_to_fp16 = const()[name = string("op_8522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_813_cast_fp16, y = var_8522_to_fp16)[name = string("aw_chunk_813_cast_fp16")]; + fp16 var_8524_to_fp16 = const()[name = string("op_8524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_815_cast_fp16, y = var_8524_to_fp16)[name = string("aw_chunk_815_cast_fp16")]; + fp16 var_8526_to_fp16 = const()[name = string("op_8526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_817_cast_fp16, y = var_8526_to_fp16)[name = string("aw_chunk_817_cast_fp16")]; + fp16 var_8528_to_fp16 = const()[name = string("op_8528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_819_cast_fp16, y = var_8528_to_fp16)[name = string("aw_chunk_819_cast_fp16")]; + fp16 var_8530_to_fp16 = const()[name = string("op_8530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_821_cast_fp16, y = var_8530_to_fp16)[name = string("aw_chunk_821_cast_fp16")]; + fp16 var_8532_to_fp16 = const()[name = string("op_8532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_823_cast_fp16, y = var_8532_to_fp16)[name = string("aw_chunk_823_cast_fp16")]; + fp16 var_8534_to_fp16 = const()[name = string("op_8534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_825_cast_fp16, y = var_8534_to_fp16)[name = string("aw_chunk_825_cast_fp16")]; + fp16 var_8536_to_fp16 = const()[name = string("op_8536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_827_cast_fp16, y = var_8536_to_fp16)[name = string("aw_chunk_827_cast_fp16")]; + fp16 var_8538_to_fp16 = const()[name = string("op_8538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_829_cast_fp16, y = var_8538_to_fp16)[name = string("aw_chunk_829_cast_fp16")]; + fp16 var_8540_to_fp16 = const()[name = string("op_8540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_831_cast_fp16, y = var_8540_to_fp16)[name = string("aw_chunk_831_cast_fp16")]; + fp16 var_8542_to_fp16 = const()[name = string("op_8542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_833_cast_fp16, y = var_8542_to_fp16)[name = string("aw_chunk_833_cast_fp16")]; + fp16 var_8544_to_fp16 = const()[name = string("op_8544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_835_cast_fp16, y = var_8544_to_fp16)[name = string("aw_chunk_835_cast_fp16")]; + fp16 var_8546_to_fp16 = const()[name = string("op_8546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_837_cast_fp16, y = var_8546_to_fp16)[name = string("aw_chunk_837_cast_fp16")]; + fp16 var_8548_to_fp16 = const()[name = string("op_8548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_839_cast_fp16, y = var_8548_to_fp16)[name = string("aw_chunk_839_cast_fp16")]; + fp16 var_8550_to_fp16 = const()[name = string("op_8550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_841_cast_fp16, y = var_8550_to_fp16)[name = string("aw_chunk_841_cast_fp16")]; + fp16 var_8552_to_fp16 = const()[name = string("op_8552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_843_cast_fp16, y = var_8552_to_fp16)[name = string("aw_chunk_843_cast_fp16")]; + fp16 var_8554_to_fp16 = const()[name = string("op_8554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_845_cast_fp16, y = var_8554_to_fp16)[name = string("aw_chunk_845_cast_fp16")]; + fp16 var_8556_to_fp16 = const()[name = string("op_8556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_847_cast_fp16, y = var_8556_to_fp16)[name = string("aw_chunk_847_cast_fp16")]; + fp16 var_8558_to_fp16 = const()[name = string("op_8558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_849_cast_fp16, y = var_8558_to_fp16)[name = string("aw_chunk_849_cast_fp16")]; + fp16 var_8560_to_fp16 = const()[name = string("op_8560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_851_cast_fp16, y = var_8560_to_fp16)[name = string("aw_chunk_851_cast_fp16")]; + fp16 var_8562_to_fp16 = const()[name = string("op_8562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_853_cast_fp16, y = var_8562_to_fp16)[name = string("aw_chunk_853_cast_fp16")]; + fp16 var_8564_to_fp16 = const()[name = string("op_8564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_855_cast_fp16, y = var_8564_to_fp16)[name = string("aw_chunk_855_cast_fp16")]; + fp16 var_8566_to_fp16 = const()[name = string("op_8566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_857_cast_fp16, y = var_8566_to_fp16)[name = string("aw_chunk_857_cast_fp16")]; + fp16 var_8568_to_fp16 = const()[name = string("op_8568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_859_cast_fp16, y = var_8568_to_fp16)[name = string("aw_chunk_859_cast_fp16")]; + fp16 var_8570_to_fp16 = const()[name = string("op_8570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_861_cast_fp16, y = var_8570_to_fp16)[name = string("aw_chunk_861_cast_fp16")]; + fp16 var_8572_to_fp16 = const()[name = string("op_8572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_863_cast_fp16, y = var_8572_to_fp16)[name = string("aw_chunk_863_cast_fp16")]; + tensor var_8574_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_769_cast_fp16)[name = string("op_8574_cast_fp16")]; + tensor var_8575_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_771_cast_fp16)[name = string("op_8575_cast_fp16")]; + tensor var_8576_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_773_cast_fp16)[name = string("op_8576_cast_fp16")]; + tensor var_8577_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_775_cast_fp16)[name = string("op_8577_cast_fp16")]; + tensor var_8578_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_777_cast_fp16)[name = string("op_8578_cast_fp16")]; + tensor var_8579_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_779_cast_fp16)[name = string("op_8579_cast_fp16")]; + tensor var_8580_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_781_cast_fp16)[name = string("op_8580_cast_fp16")]; + tensor var_8581_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_783_cast_fp16)[name = string("op_8581_cast_fp16")]; + tensor var_8582_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_785_cast_fp16)[name = string("op_8582_cast_fp16")]; + tensor var_8583_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_787_cast_fp16)[name = string("op_8583_cast_fp16")]; + tensor var_8584_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_789_cast_fp16)[name = string("op_8584_cast_fp16")]; + tensor var_8585_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_791_cast_fp16)[name = string("op_8585_cast_fp16")]; + tensor var_8586_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_793_cast_fp16)[name = string("op_8586_cast_fp16")]; + tensor var_8587_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_795_cast_fp16)[name = string("op_8587_cast_fp16")]; + tensor var_8588_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_797_cast_fp16)[name = string("op_8588_cast_fp16")]; + tensor var_8589_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_799_cast_fp16)[name = string("op_8589_cast_fp16")]; + tensor var_8590_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_801_cast_fp16)[name = string("op_8590_cast_fp16")]; + tensor var_8591_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_803_cast_fp16)[name = string("op_8591_cast_fp16")]; + tensor var_8592_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_805_cast_fp16)[name = string("op_8592_cast_fp16")]; + tensor var_8593_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_807_cast_fp16)[name = string("op_8593_cast_fp16")]; + tensor var_8594_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_809_cast_fp16)[name = string("op_8594_cast_fp16")]; + tensor var_8595_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_811_cast_fp16)[name = string("op_8595_cast_fp16")]; + tensor var_8596_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_813_cast_fp16)[name = string("op_8596_cast_fp16")]; + tensor var_8597_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_815_cast_fp16)[name = string("op_8597_cast_fp16")]; + tensor var_8598_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_817_cast_fp16)[name = string("op_8598_cast_fp16")]; + tensor var_8599_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_819_cast_fp16)[name = string("op_8599_cast_fp16")]; + tensor var_8600_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_821_cast_fp16)[name = string("op_8600_cast_fp16")]; + tensor var_8601_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_823_cast_fp16)[name = string("op_8601_cast_fp16")]; + tensor var_8602_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_825_cast_fp16)[name = string("op_8602_cast_fp16")]; + tensor var_8603_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_827_cast_fp16)[name = string("op_8603_cast_fp16")]; + tensor var_8604_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_829_cast_fp16)[name = string("op_8604_cast_fp16")]; + tensor var_8605_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_831_cast_fp16)[name = string("op_8605_cast_fp16")]; + tensor var_8606_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_833_cast_fp16)[name = string("op_8606_cast_fp16")]; + tensor var_8607_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_835_cast_fp16)[name = string("op_8607_cast_fp16")]; + tensor var_8608_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_837_cast_fp16)[name = string("op_8608_cast_fp16")]; + tensor var_8609_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_839_cast_fp16)[name = string("op_8609_cast_fp16")]; + tensor var_8610_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_841_cast_fp16)[name = string("op_8610_cast_fp16")]; + tensor var_8611_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_843_cast_fp16)[name = string("op_8611_cast_fp16")]; + tensor var_8612_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_845_cast_fp16)[name = string("op_8612_cast_fp16")]; + tensor var_8613_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_847_cast_fp16)[name = string("op_8613_cast_fp16")]; + tensor var_8614_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_849_cast_fp16)[name = string("op_8614_cast_fp16")]; + tensor var_8615_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_851_cast_fp16)[name = string("op_8615_cast_fp16")]; + tensor var_8616_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_853_cast_fp16)[name = string("op_8616_cast_fp16")]; + tensor var_8617_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_855_cast_fp16)[name = string("op_8617_cast_fp16")]; + tensor var_8618_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_857_cast_fp16)[name = string("op_8618_cast_fp16")]; + tensor var_8619_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_859_cast_fp16)[name = string("op_8619_cast_fp16")]; + tensor var_8620_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_861_cast_fp16)[name = string("op_8620_cast_fp16")]; + tensor var_8621_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_863_cast_fp16)[name = string("op_8621_cast_fp16")]; + string var_8623_equation_0 = const()[name = string("op_8623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8623_cast_fp16 = einsum(equation = var_8623_equation_0, values = (var_8335_cast_fp16, var_8574_cast_fp16))[name = string("op_8623_cast_fp16")]; + string var_8625_equation_0 = const()[name = string("op_8625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8625_cast_fp16 = einsum(equation = var_8625_equation_0, values = (var_8335_cast_fp16, var_8575_cast_fp16))[name = string("op_8625_cast_fp16")]; + string var_8627_equation_0 = const()[name = string("op_8627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8627_cast_fp16 = einsum(equation = var_8627_equation_0, values = (var_8335_cast_fp16, var_8576_cast_fp16))[name = string("op_8627_cast_fp16")]; + string var_8629_equation_0 = const()[name = string("op_8629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8629_cast_fp16 = einsum(equation = var_8629_equation_0, values = (var_8335_cast_fp16, var_8577_cast_fp16))[name = string("op_8629_cast_fp16")]; + string var_8631_equation_0 = const()[name = string("op_8631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8631_cast_fp16 = einsum(equation = var_8631_equation_0, values = (var_8339_cast_fp16, var_8578_cast_fp16))[name = string("op_8631_cast_fp16")]; + string var_8633_equation_0 = const()[name = string("op_8633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8633_cast_fp16 = einsum(equation = var_8633_equation_0, values = (var_8339_cast_fp16, var_8579_cast_fp16))[name = string("op_8633_cast_fp16")]; + string var_8635_equation_0 = const()[name = string("op_8635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8635_cast_fp16 = einsum(equation = var_8635_equation_0, values = (var_8339_cast_fp16, var_8580_cast_fp16))[name = string("op_8635_cast_fp16")]; + string var_8637_equation_0 = const()[name = string("op_8637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8637_cast_fp16 = einsum(equation = var_8637_equation_0, values = (var_8339_cast_fp16, var_8581_cast_fp16))[name = string("op_8637_cast_fp16")]; + string var_8639_equation_0 = const()[name = string("op_8639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8639_cast_fp16 = einsum(equation = var_8639_equation_0, values = (var_8343_cast_fp16, var_8582_cast_fp16))[name = string("op_8639_cast_fp16")]; + string var_8641_equation_0 = const()[name = string("op_8641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8641_cast_fp16 = einsum(equation = var_8641_equation_0, values = (var_8343_cast_fp16, var_8583_cast_fp16))[name = string("op_8641_cast_fp16")]; + string var_8643_equation_0 = const()[name = string("op_8643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8643_cast_fp16 = einsum(equation = var_8643_equation_0, values = (var_8343_cast_fp16, var_8584_cast_fp16))[name = string("op_8643_cast_fp16")]; + string var_8645_equation_0 = const()[name = string("op_8645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8645_cast_fp16 = einsum(equation = var_8645_equation_0, values = (var_8343_cast_fp16, var_8585_cast_fp16))[name = string("op_8645_cast_fp16")]; + string var_8647_equation_0 = const()[name = string("op_8647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8647_cast_fp16 = einsum(equation = var_8647_equation_0, values = (var_8347_cast_fp16, var_8586_cast_fp16))[name = string("op_8647_cast_fp16")]; + string var_8649_equation_0 = const()[name = string("op_8649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8649_cast_fp16 = einsum(equation = var_8649_equation_0, values = (var_8347_cast_fp16, var_8587_cast_fp16))[name = string("op_8649_cast_fp16")]; + string var_8651_equation_0 = const()[name = string("op_8651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8651_cast_fp16 = einsum(equation = var_8651_equation_0, values = (var_8347_cast_fp16, var_8588_cast_fp16))[name = string("op_8651_cast_fp16")]; + string var_8653_equation_0 = const()[name = string("op_8653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8653_cast_fp16 = einsum(equation = var_8653_equation_0, values = (var_8347_cast_fp16, var_8589_cast_fp16))[name = string("op_8653_cast_fp16")]; + string var_8655_equation_0 = const()[name = string("op_8655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8655_cast_fp16 = einsum(equation = var_8655_equation_0, values = (var_8351_cast_fp16, var_8590_cast_fp16))[name = string("op_8655_cast_fp16")]; + string var_8657_equation_0 = const()[name = string("op_8657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8657_cast_fp16 = einsum(equation = var_8657_equation_0, values = (var_8351_cast_fp16, var_8591_cast_fp16))[name = string("op_8657_cast_fp16")]; + string var_8659_equation_0 = const()[name = string("op_8659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8659_cast_fp16 = einsum(equation = var_8659_equation_0, values = (var_8351_cast_fp16, var_8592_cast_fp16))[name = string("op_8659_cast_fp16")]; + string var_8661_equation_0 = const()[name = string("op_8661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8661_cast_fp16 = einsum(equation = var_8661_equation_0, values = (var_8351_cast_fp16, var_8593_cast_fp16))[name = string("op_8661_cast_fp16")]; + string var_8663_equation_0 = const()[name = string("op_8663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8663_cast_fp16 = einsum(equation = var_8663_equation_0, values = (var_8355_cast_fp16, var_8594_cast_fp16))[name = string("op_8663_cast_fp16")]; + string var_8665_equation_0 = const()[name = string("op_8665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8665_cast_fp16 = einsum(equation = var_8665_equation_0, values = (var_8355_cast_fp16, var_8595_cast_fp16))[name = string("op_8665_cast_fp16")]; + string var_8667_equation_0 = const()[name = string("op_8667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8667_cast_fp16 = einsum(equation = var_8667_equation_0, values = (var_8355_cast_fp16, var_8596_cast_fp16))[name = string("op_8667_cast_fp16")]; + string var_8669_equation_0 = const()[name = string("op_8669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8669_cast_fp16 = einsum(equation = var_8669_equation_0, values = (var_8355_cast_fp16, var_8597_cast_fp16))[name = string("op_8669_cast_fp16")]; + string var_8671_equation_0 = const()[name = string("op_8671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8671_cast_fp16 = einsum(equation = var_8671_equation_0, values = (var_8359_cast_fp16, var_8598_cast_fp16))[name = string("op_8671_cast_fp16")]; + string var_8673_equation_0 = const()[name = string("op_8673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8673_cast_fp16 = einsum(equation = var_8673_equation_0, values = (var_8359_cast_fp16, var_8599_cast_fp16))[name = string("op_8673_cast_fp16")]; + string var_8675_equation_0 = const()[name = string("op_8675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8675_cast_fp16 = einsum(equation = var_8675_equation_0, values = (var_8359_cast_fp16, var_8600_cast_fp16))[name = string("op_8675_cast_fp16")]; + string var_8677_equation_0 = const()[name = string("op_8677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8677_cast_fp16 = einsum(equation = var_8677_equation_0, values = (var_8359_cast_fp16, var_8601_cast_fp16))[name = string("op_8677_cast_fp16")]; + string var_8679_equation_0 = const()[name = string("op_8679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8679_cast_fp16 = einsum(equation = var_8679_equation_0, values = (var_8363_cast_fp16, var_8602_cast_fp16))[name = string("op_8679_cast_fp16")]; + string var_8681_equation_0 = const()[name = string("op_8681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8681_cast_fp16 = einsum(equation = var_8681_equation_0, values = (var_8363_cast_fp16, var_8603_cast_fp16))[name = string("op_8681_cast_fp16")]; + string var_8683_equation_0 = const()[name = string("op_8683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8683_cast_fp16 = einsum(equation = var_8683_equation_0, values = (var_8363_cast_fp16, var_8604_cast_fp16))[name = string("op_8683_cast_fp16")]; + string var_8685_equation_0 = const()[name = string("op_8685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8685_cast_fp16 = einsum(equation = var_8685_equation_0, values = (var_8363_cast_fp16, var_8605_cast_fp16))[name = string("op_8685_cast_fp16")]; + string var_8687_equation_0 = const()[name = string("op_8687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8687_cast_fp16 = einsum(equation = var_8687_equation_0, values = (var_8367_cast_fp16, var_8606_cast_fp16))[name = string("op_8687_cast_fp16")]; + string var_8689_equation_0 = const()[name = string("op_8689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8689_cast_fp16 = einsum(equation = var_8689_equation_0, values = (var_8367_cast_fp16, var_8607_cast_fp16))[name = string("op_8689_cast_fp16")]; + string var_8691_equation_0 = const()[name = string("op_8691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8691_cast_fp16 = einsum(equation = var_8691_equation_0, values = (var_8367_cast_fp16, var_8608_cast_fp16))[name = string("op_8691_cast_fp16")]; + string var_8693_equation_0 = const()[name = string("op_8693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8693_cast_fp16 = einsum(equation = var_8693_equation_0, values = (var_8367_cast_fp16, var_8609_cast_fp16))[name = string("op_8693_cast_fp16")]; + string var_8695_equation_0 = const()[name = string("op_8695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8695_cast_fp16 = einsum(equation = var_8695_equation_0, values = (var_8371_cast_fp16, var_8610_cast_fp16))[name = string("op_8695_cast_fp16")]; + string var_8697_equation_0 = const()[name = string("op_8697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8697_cast_fp16 = einsum(equation = var_8697_equation_0, values = (var_8371_cast_fp16, var_8611_cast_fp16))[name = string("op_8697_cast_fp16")]; + string var_8699_equation_0 = const()[name = string("op_8699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8699_cast_fp16 = einsum(equation = var_8699_equation_0, values = (var_8371_cast_fp16, var_8612_cast_fp16))[name = string("op_8699_cast_fp16")]; + string var_8701_equation_0 = const()[name = string("op_8701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8701_cast_fp16 = einsum(equation = var_8701_equation_0, values = (var_8371_cast_fp16, var_8613_cast_fp16))[name = string("op_8701_cast_fp16")]; + string var_8703_equation_0 = const()[name = string("op_8703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8703_cast_fp16 = einsum(equation = var_8703_equation_0, values = (var_8375_cast_fp16, var_8614_cast_fp16))[name = string("op_8703_cast_fp16")]; + string var_8705_equation_0 = const()[name = string("op_8705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8705_cast_fp16 = einsum(equation = var_8705_equation_0, values = (var_8375_cast_fp16, var_8615_cast_fp16))[name = string("op_8705_cast_fp16")]; + string var_8707_equation_0 = const()[name = string("op_8707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8707_cast_fp16 = einsum(equation = var_8707_equation_0, values = (var_8375_cast_fp16, var_8616_cast_fp16))[name = string("op_8707_cast_fp16")]; + string var_8709_equation_0 = const()[name = string("op_8709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8709_cast_fp16 = einsum(equation = var_8709_equation_0, values = (var_8375_cast_fp16, var_8617_cast_fp16))[name = string("op_8709_cast_fp16")]; + string var_8711_equation_0 = const()[name = string("op_8711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8711_cast_fp16 = einsum(equation = var_8711_equation_0, values = (var_8379_cast_fp16, var_8618_cast_fp16))[name = string("op_8711_cast_fp16")]; + string var_8713_equation_0 = const()[name = string("op_8713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8713_cast_fp16 = einsum(equation = var_8713_equation_0, values = (var_8379_cast_fp16, var_8619_cast_fp16))[name = string("op_8713_cast_fp16")]; + string var_8715_equation_0 = const()[name = string("op_8715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8715_cast_fp16 = einsum(equation = var_8715_equation_0, values = (var_8379_cast_fp16, var_8620_cast_fp16))[name = string("op_8715_cast_fp16")]; + string var_8717_equation_0 = const()[name = string("op_8717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8717_cast_fp16 = einsum(equation = var_8717_equation_0, values = (var_8379_cast_fp16, var_8621_cast_fp16))[name = string("op_8717_cast_fp16")]; + bool var_8719_interleave_0 = const()[name = string("op_8719_interleave_0"), val = bool(false)]; + tensor var_8719_cast_fp16 = concat(axis = var_7830, interleave = var_8719_interleave_0, values = (var_8623_cast_fp16, var_8625_cast_fp16, var_8627_cast_fp16, var_8629_cast_fp16))[name = string("op_8719_cast_fp16")]; + bool var_8721_interleave_0 = const()[name = string("op_8721_interleave_0"), val = bool(false)]; + tensor var_8721_cast_fp16 = concat(axis = var_7830, interleave = var_8721_interleave_0, values = (var_8631_cast_fp16, var_8633_cast_fp16, var_8635_cast_fp16, var_8637_cast_fp16))[name = string("op_8721_cast_fp16")]; + bool var_8723_interleave_0 = const()[name = string("op_8723_interleave_0"), val = bool(false)]; + tensor var_8723_cast_fp16 = concat(axis = var_7830, interleave = var_8723_interleave_0, values = (var_8639_cast_fp16, var_8641_cast_fp16, var_8643_cast_fp16, var_8645_cast_fp16))[name = string("op_8723_cast_fp16")]; + bool var_8725_interleave_0 = const()[name = string("op_8725_interleave_0"), val = bool(false)]; + tensor var_8725_cast_fp16 = concat(axis = var_7830, interleave = var_8725_interleave_0, values = (var_8647_cast_fp16, var_8649_cast_fp16, var_8651_cast_fp16, var_8653_cast_fp16))[name = string("op_8725_cast_fp16")]; + bool var_8727_interleave_0 = const()[name = string("op_8727_interleave_0"), val = bool(false)]; + tensor var_8727_cast_fp16 = concat(axis = var_7830, interleave = var_8727_interleave_0, values = (var_8655_cast_fp16, var_8657_cast_fp16, var_8659_cast_fp16, var_8661_cast_fp16))[name = string("op_8727_cast_fp16")]; + bool var_8729_interleave_0 = const()[name = string("op_8729_interleave_0"), val = bool(false)]; + tensor var_8729_cast_fp16 = concat(axis = var_7830, interleave = var_8729_interleave_0, values = (var_8663_cast_fp16, var_8665_cast_fp16, var_8667_cast_fp16, var_8669_cast_fp16))[name = string("op_8729_cast_fp16")]; + bool var_8731_interleave_0 = const()[name = string("op_8731_interleave_0"), val = bool(false)]; + tensor var_8731_cast_fp16 = concat(axis = var_7830, interleave = var_8731_interleave_0, values = (var_8671_cast_fp16, var_8673_cast_fp16, var_8675_cast_fp16, var_8677_cast_fp16))[name = string("op_8731_cast_fp16")]; + bool var_8733_interleave_0 = const()[name = string("op_8733_interleave_0"), val = bool(false)]; + tensor var_8733_cast_fp16 = concat(axis = var_7830, interleave = var_8733_interleave_0, values = (var_8679_cast_fp16, var_8681_cast_fp16, var_8683_cast_fp16, var_8685_cast_fp16))[name = string("op_8733_cast_fp16")]; + bool var_8735_interleave_0 = const()[name = string("op_8735_interleave_0"), val = bool(false)]; + tensor var_8735_cast_fp16 = concat(axis = var_7830, interleave = var_8735_interleave_0, values = (var_8687_cast_fp16, var_8689_cast_fp16, var_8691_cast_fp16, var_8693_cast_fp16))[name = string("op_8735_cast_fp16")]; + bool var_8737_interleave_0 = const()[name = string("op_8737_interleave_0"), val = bool(false)]; + tensor var_8737_cast_fp16 = concat(axis = var_7830, interleave = var_8737_interleave_0, values = (var_8695_cast_fp16, var_8697_cast_fp16, var_8699_cast_fp16, var_8701_cast_fp16))[name = string("op_8737_cast_fp16")]; + bool var_8739_interleave_0 = const()[name = string("op_8739_interleave_0"), val = bool(false)]; + tensor var_8739_cast_fp16 = concat(axis = var_7830, interleave = var_8739_interleave_0, values = (var_8703_cast_fp16, var_8705_cast_fp16, var_8707_cast_fp16, var_8709_cast_fp16))[name = string("op_8739_cast_fp16")]; + bool var_8741_interleave_0 = const()[name = string("op_8741_interleave_0"), val = bool(false)]; + tensor var_8741_cast_fp16 = concat(axis = var_7830, interleave = var_8741_interleave_0, values = (var_8711_cast_fp16, var_8713_cast_fp16, var_8715_cast_fp16, var_8717_cast_fp16))[name = string("op_8741_cast_fp16")]; + bool input_65_interleave_0 = const()[name = string("input_65_interleave_0"), val = bool(false)]; + tensor input_65_cast_fp16 = concat(axis = var_7847, interleave = input_65_interleave_0, values = (var_8719_cast_fp16, var_8721_cast_fp16, var_8723_cast_fp16, var_8725_cast_fp16, var_8727_cast_fp16, var_8729_cast_fp16, var_8731_cast_fp16, var_8733_cast_fp16, var_8735_cast_fp16, var_8737_cast_fp16, var_8739_cast_fp16, var_8741_cast_fp16))[name = string("input_65_cast_fp16")]; + string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; + tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; + tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; + int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123165120)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124344832)))]; + tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_8760_to_fp16 = const()[name = string("op_8760_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_8760_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124346432)))]; + tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124348032)))]; + fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; + tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; + tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; + int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124349632)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129068288)))]; + tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; + tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129074496)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133793152)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; + int32 var_8789 = const()[name = string("op_8789"), val = int32(3)]; + int32 var_8806 = const()[name = string("op_8806"), val = int32(1)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_8823_to_fp16 = const()[name = string("op_8823_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_8823_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133794752)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133796352)))]; + fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133797952)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134977664)))]; + tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; + string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; + tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; + tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; + int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134979264)))]; + tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; + string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; + tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; + tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; + int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136158976)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137338688)))]; + tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_8861_begin_0 = const()[name = string("op_8861_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8861_end_0 = const()[name = string("op_8861_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8861_end_mask_0 = const()[name = string("op_8861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8861_cast_fp16 = slice_by_index(begin = var_8861_begin_0, end = var_8861_end_0, end_mask = var_8861_end_mask_0, x = query_19_cast_fp16)[name = string("op_8861_cast_fp16")]; + tensor var_8865_begin_0 = const()[name = string("op_8865_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8865_end_0 = const()[name = string("op_8865_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8865_end_mask_0 = const()[name = string("op_8865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8865_cast_fp16 = slice_by_index(begin = var_8865_begin_0, end = var_8865_end_0, end_mask = var_8865_end_mask_0, x = query_19_cast_fp16)[name = string("op_8865_cast_fp16")]; + tensor var_8869_begin_0 = const()[name = string("op_8869_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8869_end_0 = const()[name = string("op_8869_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8869_end_mask_0 = const()[name = string("op_8869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8869_cast_fp16 = slice_by_index(begin = var_8869_begin_0, end = var_8869_end_0, end_mask = var_8869_end_mask_0, x = query_19_cast_fp16)[name = string("op_8869_cast_fp16")]; + tensor var_8873_begin_0 = const()[name = string("op_8873_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8873_end_0 = const()[name = string("op_8873_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8873_end_mask_0 = const()[name = string("op_8873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8873_cast_fp16 = slice_by_index(begin = var_8873_begin_0, end = var_8873_end_0, end_mask = var_8873_end_mask_0, x = query_19_cast_fp16)[name = string("op_8873_cast_fp16")]; + tensor var_8877_begin_0 = const()[name = string("op_8877_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8877_end_0 = const()[name = string("op_8877_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8877_end_mask_0 = const()[name = string("op_8877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8877_cast_fp16 = slice_by_index(begin = var_8877_begin_0, end = var_8877_end_0, end_mask = var_8877_end_mask_0, x = query_19_cast_fp16)[name = string("op_8877_cast_fp16")]; + tensor var_8881_begin_0 = const()[name = string("op_8881_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8881_end_0 = const()[name = string("op_8881_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8881_end_mask_0 = const()[name = string("op_8881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8881_cast_fp16 = slice_by_index(begin = var_8881_begin_0, end = var_8881_end_0, end_mask = var_8881_end_mask_0, x = query_19_cast_fp16)[name = string("op_8881_cast_fp16")]; + tensor var_8885_begin_0 = const()[name = string("op_8885_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8885_end_0 = const()[name = string("op_8885_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8885_end_mask_0 = const()[name = string("op_8885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8885_cast_fp16 = slice_by_index(begin = var_8885_begin_0, end = var_8885_end_0, end_mask = var_8885_end_mask_0, x = query_19_cast_fp16)[name = string("op_8885_cast_fp16")]; + tensor var_8889_begin_0 = const()[name = string("op_8889_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8889_end_0 = const()[name = string("op_8889_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8889_end_mask_0 = const()[name = string("op_8889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8889_cast_fp16 = slice_by_index(begin = var_8889_begin_0, end = var_8889_end_0, end_mask = var_8889_end_mask_0, x = query_19_cast_fp16)[name = string("op_8889_cast_fp16")]; + tensor var_8893_begin_0 = const()[name = string("op_8893_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8893_end_0 = const()[name = string("op_8893_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8893_end_mask_0 = const()[name = string("op_8893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8893_cast_fp16 = slice_by_index(begin = var_8893_begin_0, end = var_8893_end_0, end_mask = var_8893_end_mask_0, x = query_19_cast_fp16)[name = string("op_8893_cast_fp16")]; + tensor var_8897_begin_0 = const()[name = string("op_8897_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8897_end_0 = const()[name = string("op_8897_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8897_end_mask_0 = const()[name = string("op_8897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8897_cast_fp16 = slice_by_index(begin = var_8897_begin_0, end = var_8897_end_0, end_mask = var_8897_end_mask_0, x = query_19_cast_fp16)[name = string("op_8897_cast_fp16")]; + tensor var_8901_begin_0 = const()[name = string("op_8901_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8901_end_0 = const()[name = string("op_8901_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8901_end_mask_0 = const()[name = string("op_8901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8901_cast_fp16 = slice_by_index(begin = var_8901_begin_0, end = var_8901_end_0, end_mask = var_8901_end_mask_0, x = query_19_cast_fp16)[name = string("op_8901_cast_fp16")]; + tensor var_8905_begin_0 = const()[name = string("op_8905_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8905_end_0 = const()[name = string("op_8905_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8905_end_mask_0 = const()[name = string("op_8905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8905_cast_fp16 = slice_by_index(begin = var_8905_begin_0, end = var_8905_end_0, end_mask = var_8905_end_mask_0, x = query_19_cast_fp16)[name = string("op_8905_cast_fp16")]; + tensor var_8914_begin_0 = const()[name = string("op_8914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8914_end_0 = const()[name = string("op_8914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8914_end_mask_0 = const()[name = string("op_8914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8914_cast_fp16 = slice_by_index(begin = var_8914_begin_0, end = var_8914_end_0, end_mask = var_8914_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8914_cast_fp16")]; + tensor var_8921_begin_0 = const()[name = string("op_8921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8921_end_0 = const()[name = string("op_8921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8921_end_mask_0 = const()[name = string("op_8921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8921_cast_fp16 = slice_by_index(begin = var_8921_begin_0, end = var_8921_end_0, end_mask = var_8921_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8921_cast_fp16")]; + tensor var_8928_begin_0 = const()[name = string("op_8928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8928_end_0 = const()[name = string("op_8928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8928_end_mask_0 = const()[name = string("op_8928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8928_cast_fp16 = slice_by_index(begin = var_8928_begin_0, end = var_8928_end_0, end_mask = var_8928_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8928_cast_fp16")]; + tensor var_8935_begin_0 = const()[name = string("op_8935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8935_end_0 = const()[name = string("op_8935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8935_end_mask_0 = const()[name = string("op_8935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8935_cast_fp16 = slice_by_index(begin = var_8935_begin_0, end = var_8935_end_0, end_mask = var_8935_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8935_cast_fp16")]; + tensor var_8942_begin_0 = const()[name = string("op_8942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8942_end_0 = const()[name = string("op_8942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8942_end_mask_0 = const()[name = string("op_8942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8942_cast_fp16 = slice_by_index(begin = var_8942_begin_0, end = var_8942_end_0, end_mask = var_8942_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8942_cast_fp16")]; + tensor var_8949_begin_0 = const()[name = string("op_8949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8949_end_0 = const()[name = string("op_8949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8949_end_mask_0 = const()[name = string("op_8949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8949_cast_fp16 = slice_by_index(begin = var_8949_begin_0, end = var_8949_end_0, end_mask = var_8949_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8949_cast_fp16")]; + tensor var_8956_begin_0 = const()[name = string("op_8956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8956_end_0 = const()[name = string("op_8956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8956_end_mask_0 = const()[name = string("op_8956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8956_cast_fp16 = slice_by_index(begin = var_8956_begin_0, end = var_8956_end_0, end_mask = var_8956_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8956_cast_fp16")]; + tensor var_8963_begin_0 = const()[name = string("op_8963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8963_end_0 = const()[name = string("op_8963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8963_end_mask_0 = const()[name = string("op_8963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8963_cast_fp16 = slice_by_index(begin = var_8963_begin_0, end = var_8963_end_0, end_mask = var_8963_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8963_cast_fp16")]; + tensor var_8970_begin_0 = const()[name = string("op_8970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8970_end_0 = const()[name = string("op_8970_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8970_end_mask_0 = const()[name = string("op_8970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8970_cast_fp16 = slice_by_index(begin = var_8970_begin_0, end = var_8970_end_0, end_mask = var_8970_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8970_cast_fp16")]; + tensor var_8977_begin_0 = const()[name = string("op_8977_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8977_end_0 = const()[name = string("op_8977_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8977_end_mask_0 = const()[name = string("op_8977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8977_cast_fp16 = slice_by_index(begin = var_8977_begin_0, end = var_8977_end_0, end_mask = var_8977_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8977_cast_fp16")]; + tensor var_8984_begin_0 = const()[name = string("op_8984_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8984_end_0 = const()[name = string("op_8984_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8984_end_mask_0 = const()[name = string("op_8984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8984_cast_fp16 = slice_by_index(begin = var_8984_begin_0, end = var_8984_end_0, end_mask = var_8984_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8984_cast_fp16")]; + tensor var_8991_begin_0 = const()[name = string("op_8991_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8991_end_0 = const()[name = string("op_8991_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8991_end_mask_0 = const()[name = string("op_8991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8991_cast_fp16 = slice_by_index(begin = var_8991_begin_0, end = var_8991_end_0, end_mask = var_8991_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8991_cast_fp16")]; + tensor var_8998_begin_0 = const()[name = string("op_8998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8998_end_0 = const()[name = string("op_8998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8998_end_mask_0 = const()[name = string("op_8998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8998_cast_fp16 = slice_by_index(begin = var_8998_begin_0, end = var_8998_end_0, end_mask = var_8998_end_mask_0, x = var_8873_cast_fp16)[name = string("op_8998_cast_fp16")]; + tensor var_9005_begin_0 = const()[name = string("op_9005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9005_end_0 = const()[name = string("op_9005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9005_end_mask_0 = const()[name = string("op_9005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9005_cast_fp16 = slice_by_index(begin = var_9005_begin_0, end = var_9005_end_0, end_mask = var_9005_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9005_cast_fp16")]; + tensor var_9012_begin_0 = const()[name = string("op_9012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9012_end_0 = const()[name = string("op_9012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9012_end_mask_0 = const()[name = string("op_9012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9012_cast_fp16 = slice_by_index(begin = var_9012_begin_0, end = var_9012_end_0, end_mask = var_9012_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9012_cast_fp16")]; + tensor var_9019_begin_0 = const()[name = string("op_9019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9019_end_0 = const()[name = string("op_9019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9019_end_mask_0 = const()[name = string("op_9019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9019_cast_fp16 = slice_by_index(begin = var_9019_begin_0, end = var_9019_end_0, end_mask = var_9019_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9019_cast_fp16")]; + tensor var_9026_begin_0 = const()[name = string("op_9026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9026_end_0 = const()[name = string("op_9026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9026_end_mask_0 = const()[name = string("op_9026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9026_cast_fp16 = slice_by_index(begin = var_9026_begin_0, end = var_9026_end_0, end_mask = var_9026_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9026_cast_fp16")]; + tensor var_9033_begin_0 = const()[name = string("op_9033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9033_end_0 = const()[name = string("op_9033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9033_end_mask_0 = const()[name = string("op_9033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9033_cast_fp16 = slice_by_index(begin = var_9033_begin_0, end = var_9033_end_0, end_mask = var_9033_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9033_cast_fp16")]; + tensor var_9040_begin_0 = const()[name = string("op_9040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9040_end_0 = const()[name = string("op_9040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9040_end_mask_0 = const()[name = string("op_9040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9040_cast_fp16 = slice_by_index(begin = var_9040_begin_0, end = var_9040_end_0, end_mask = var_9040_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9040_cast_fp16")]; + tensor var_9047_begin_0 = const()[name = string("op_9047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9047_end_0 = const()[name = string("op_9047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9047_end_mask_0 = const()[name = string("op_9047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9047_cast_fp16 = slice_by_index(begin = var_9047_begin_0, end = var_9047_end_0, end_mask = var_9047_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9047_cast_fp16")]; + tensor var_9054_begin_0 = const()[name = string("op_9054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9054_end_0 = const()[name = string("op_9054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9054_end_mask_0 = const()[name = string("op_9054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9054_cast_fp16 = slice_by_index(begin = var_9054_begin_0, end = var_9054_end_0, end_mask = var_9054_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9054_cast_fp16")]; + tensor var_9061_begin_0 = const()[name = string("op_9061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9061_end_0 = const()[name = string("op_9061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9061_end_mask_0 = const()[name = string("op_9061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9061_cast_fp16 = slice_by_index(begin = var_9061_begin_0, end = var_9061_end_0, end_mask = var_9061_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9061_cast_fp16")]; + tensor var_9068_begin_0 = const()[name = string("op_9068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9068_end_0 = const()[name = string("op_9068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9068_end_mask_0 = const()[name = string("op_9068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9068_cast_fp16 = slice_by_index(begin = var_9068_begin_0, end = var_9068_end_0, end_mask = var_9068_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9068_cast_fp16")]; + tensor var_9075_begin_0 = const()[name = string("op_9075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9075_end_0 = const()[name = string("op_9075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9075_end_mask_0 = const()[name = string("op_9075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9075_cast_fp16 = slice_by_index(begin = var_9075_begin_0, end = var_9075_end_0, end_mask = var_9075_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9075_cast_fp16")]; + tensor var_9082_begin_0 = const()[name = string("op_9082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9082_end_0 = const()[name = string("op_9082_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9082_end_mask_0 = const()[name = string("op_9082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9082_cast_fp16 = slice_by_index(begin = var_9082_begin_0, end = var_9082_end_0, end_mask = var_9082_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9082_cast_fp16")]; + tensor var_9089_begin_0 = const()[name = string("op_9089_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9089_end_0 = const()[name = string("op_9089_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9089_end_mask_0 = const()[name = string("op_9089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9089_cast_fp16 = slice_by_index(begin = var_9089_begin_0, end = var_9089_end_0, end_mask = var_9089_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9089_cast_fp16")]; + tensor var_9096_begin_0 = const()[name = string("op_9096_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9096_end_0 = const()[name = string("op_9096_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9096_end_mask_0 = const()[name = string("op_9096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9096_cast_fp16 = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9096_cast_fp16")]; + tensor var_9103_begin_0 = const()[name = string("op_9103_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9103_end_0 = const()[name = string("op_9103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9103_end_mask_0 = const()[name = string("op_9103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9103_cast_fp16 = slice_by_index(begin = var_9103_begin_0, end = var_9103_end_0, end_mask = var_9103_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9103_cast_fp16")]; + tensor var_9110_begin_0 = const()[name = string("op_9110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9110_end_0 = const()[name = string("op_9110_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9110_end_mask_0 = const()[name = string("op_9110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9110_cast_fp16 = slice_by_index(begin = var_9110_begin_0, end = var_9110_end_0, end_mask = var_9110_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9110_cast_fp16")]; + tensor var_9117_begin_0 = const()[name = string("op_9117_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9117_end_0 = const()[name = string("op_9117_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9117_end_mask_0 = const()[name = string("op_9117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9117_cast_fp16 = slice_by_index(begin = var_9117_begin_0, end = var_9117_end_0, end_mask = var_9117_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9117_cast_fp16")]; + tensor var_9124_begin_0 = const()[name = string("op_9124_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9124_end_0 = const()[name = string("op_9124_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9124_end_mask_0 = const()[name = string("op_9124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9124_cast_fp16 = slice_by_index(begin = var_9124_begin_0, end = var_9124_end_0, end_mask = var_9124_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9124_cast_fp16")]; + tensor var_9131_begin_0 = const()[name = string("op_9131_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9131_end_0 = const()[name = string("op_9131_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9131_end_mask_0 = const()[name = string("op_9131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9131_cast_fp16 = slice_by_index(begin = var_9131_begin_0, end = var_9131_end_0, end_mask = var_9131_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9131_cast_fp16")]; + tensor var_9138_begin_0 = const()[name = string("op_9138_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9138_end_0 = const()[name = string("op_9138_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9138_end_mask_0 = const()[name = string("op_9138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9138_cast_fp16 = slice_by_index(begin = var_9138_begin_0, end = var_9138_end_0, end_mask = var_9138_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9138_cast_fp16")]; + tensor var_9145_begin_0 = const()[name = string("op_9145_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9145_end_0 = const()[name = string("op_9145_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9145_end_mask_0 = const()[name = string("op_9145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9145_cast_fp16 = slice_by_index(begin = var_9145_begin_0, end = var_9145_end_0, end_mask = var_9145_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9145_cast_fp16")]; + tensor var_9152_begin_0 = const()[name = string("op_9152_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9152_end_0 = const()[name = string("op_9152_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9152_end_mask_0 = const()[name = string("op_9152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9152_cast_fp16 = slice_by_index(begin = var_9152_begin_0, end = var_9152_end_0, end_mask = var_9152_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9152_cast_fp16")]; + tensor var_9159_begin_0 = const()[name = string("op_9159_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9159_end_0 = const()[name = string("op_9159_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9159_end_mask_0 = const()[name = string("op_9159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9159_cast_fp16 = slice_by_index(begin = var_9159_begin_0, end = var_9159_end_0, end_mask = var_9159_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9159_cast_fp16")]; + tensor var_9166_begin_0 = const()[name = string("op_9166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9166_end_0 = const()[name = string("op_9166_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9166_end_mask_0 = const()[name = string("op_9166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9166_cast_fp16 = slice_by_index(begin = var_9166_begin_0, end = var_9166_end_0, end_mask = var_9166_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9166_cast_fp16")]; + tensor var_9173_begin_0 = const()[name = string("op_9173_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9173_end_0 = const()[name = string("op_9173_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9173_end_mask_0 = const()[name = string("op_9173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9173_cast_fp16 = slice_by_index(begin = var_9173_begin_0, end = var_9173_end_0, end_mask = var_9173_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9173_cast_fp16")]; + tensor var_9180_begin_0 = const()[name = string("op_9180_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9180_end_0 = const()[name = string("op_9180_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9180_end_mask_0 = const()[name = string("op_9180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9180_cast_fp16 = slice_by_index(begin = var_9180_begin_0, end = var_9180_end_0, end_mask = var_9180_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9180_cast_fp16")]; + tensor var_9187_begin_0 = const()[name = string("op_9187_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9187_end_0 = const()[name = string("op_9187_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9187_end_mask_0 = const()[name = string("op_9187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9187_cast_fp16 = slice_by_index(begin = var_9187_begin_0, end = var_9187_end_0, end_mask = var_9187_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9187_cast_fp16")]; + tensor var_9194_begin_0 = const()[name = string("op_9194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9194_end_0 = const()[name = string("op_9194_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9194_end_mask_0 = const()[name = string("op_9194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9194_cast_fp16 = slice_by_index(begin = var_9194_begin_0, end = var_9194_end_0, end_mask = var_9194_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9194_cast_fp16")]; + tensor var_9201_begin_0 = const()[name = string("op_9201_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9201_end_0 = const()[name = string("op_9201_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9201_end_mask_0 = const()[name = string("op_9201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9201_cast_fp16 = slice_by_index(begin = var_9201_begin_0, end = var_9201_end_0, end_mask = var_9201_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9201_cast_fp16")]; + tensor var_9208_begin_0 = const()[name = string("op_9208_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9208_end_0 = const()[name = string("op_9208_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9208_end_mask_0 = const()[name = string("op_9208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9208_cast_fp16 = slice_by_index(begin = var_9208_begin_0, end = var_9208_end_0, end_mask = var_9208_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9208_cast_fp16")]; + tensor var_9215_begin_0 = const()[name = string("op_9215_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9215_end_0 = const()[name = string("op_9215_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9215_end_mask_0 = const()[name = string("op_9215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9215_cast_fp16 = slice_by_index(begin = var_9215_begin_0, end = var_9215_end_0, end_mask = var_9215_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9215_cast_fp16")]; + tensor var_9222_begin_0 = const()[name = string("op_9222_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9222_end_0 = const()[name = string("op_9222_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9222_end_mask_0 = const()[name = string("op_9222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9222_cast_fp16 = slice_by_index(begin = var_9222_begin_0, end = var_9222_end_0, end_mask = var_9222_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9222_cast_fp16")]; + tensor var_9229_begin_0 = const()[name = string("op_9229_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9229_end_0 = const()[name = string("op_9229_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9229_end_mask_0 = const()[name = string("op_9229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9229_cast_fp16 = slice_by_index(begin = var_9229_begin_0, end = var_9229_end_0, end_mask = var_9229_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9229_cast_fp16")]; + tensor var_9236_begin_0 = const()[name = string("op_9236_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9236_end_0 = const()[name = string("op_9236_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9236_end_mask_0 = const()[name = string("op_9236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9236_cast_fp16 = slice_by_index(begin = var_9236_begin_0, end = var_9236_end_0, end_mask = var_9236_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9236_cast_fp16")]; + tensor var_9243_begin_0 = const()[name = string("op_9243_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9243_end_0 = const()[name = string("op_9243_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9243_end_mask_0 = const()[name = string("op_9243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9243_cast_fp16 = slice_by_index(begin = var_9243_begin_0, end = var_9243_end_0, end_mask = var_9243_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9243_cast_fp16")]; + tensor k_19_perm_0 = const()[name = string("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9248_begin_0 = const()[name = string("op_9248_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9248_end_0 = const()[name = string("op_9248_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_9248_end_mask_0 = const()[name = string("op_9248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = string("transpose_2")]; + tensor var_9248_cast_fp16 = slice_by_index(begin = var_9248_begin_0, end = var_9248_end_0, end_mask = var_9248_end_mask_0, x = k_19_cast_fp16)[name = string("op_9248_cast_fp16")]; + tensor var_9252_begin_0 = const()[name = string("op_9252_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9252_end_0 = const()[name = string("op_9252_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_9252_end_mask_0 = const()[name = string("op_9252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9252_cast_fp16 = slice_by_index(begin = var_9252_begin_0, end = var_9252_end_0, end_mask = var_9252_end_mask_0, x = k_19_cast_fp16)[name = string("op_9252_cast_fp16")]; + tensor var_9256_begin_0 = const()[name = string("op_9256_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9256_end_0 = const()[name = string("op_9256_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_9256_end_mask_0 = const()[name = string("op_9256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9256_cast_fp16 = slice_by_index(begin = var_9256_begin_0, end = var_9256_end_0, end_mask = var_9256_end_mask_0, x = k_19_cast_fp16)[name = string("op_9256_cast_fp16")]; + tensor var_9260_begin_0 = const()[name = string("op_9260_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9260_end_0 = const()[name = string("op_9260_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_9260_end_mask_0 = const()[name = string("op_9260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9260_cast_fp16 = slice_by_index(begin = var_9260_begin_0, end = var_9260_end_0, end_mask = var_9260_end_mask_0, x = k_19_cast_fp16)[name = string("op_9260_cast_fp16")]; + tensor var_9264_begin_0 = const()[name = string("op_9264_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9264_end_0 = const()[name = string("op_9264_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_9264_end_mask_0 = const()[name = string("op_9264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9264_cast_fp16 = slice_by_index(begin = var_9264_begin_0, end = var_9264_end_0, end_mask = var_9264_end_mask_0, x = k_19_cast_fp16)[name = string("op_9264_cast_fp16")]; + tensor var_9268_begin_0 = const()[name = string("op_9268_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9268_end_0 = const()[name = string("op_9268_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_9268_end_mask_0 = const()[name = string("op_9268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9268_cast_fp16 = slice_by_index(begin = var_9268_begin_0, end = var_9268_end_0, end_mask = var_9268_end_mask_0, x = k_19_cast_fp16)[name = string("op_9268_cast_fp16")]; + tensor var_9272_begin_0 = const()[name = string("op_9272_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9272_end_0 = const()[name = string("op_9272_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_9272_end_mask_0 = const()[name = string("op_9272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9272_cast_fp16 = slice_by_index(begin = var_9272_begin_0, end = var_9272_end_0, end_mask = var_9272_end_mask_0, x = k_19_cast_fp16)[name = string("op_9272_cast_fp16")]; + tensor var_9276_begin_0 = const()[name = string("op_9276_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9276_end_0 = const()[name = string("op_9276_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_9276_end_mask_0 = const()[name = string("op_9276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9276_cast_fp16 = slice_by_index(begin = var_9276_begin_0, end = var_9276_end_0, end_mask = var_9276_end_mask_0, x = k_19_cast_fp16)[name = string("op_9276_cast_fp16")]; + tensor var_9280_begin_0 = const()[name = string("op_9280_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9280_end_0 = const()[name = string("op_9280_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_9280_end_mask_0 = const()[name = string("op_9280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9280_cast_fp16 = slice_by_index(begin = var_9280_begin_0, end = var_9280_end_0, end_mask = var_9280_end_mask_0, x = k_19_cast_fp16)[name = string("op_9280_cast_fp16")]; + tensor var_9284_begin_0 = const()[name = string("op_9284_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9284_end_0 = const()[name = string("op_9284_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_9284_end_mask_0 = const()[name = string("op_9284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9284_cast_fp16 = slice_by_index(begin = var_9284_begin_0, end = var_9284_end_0, end_mask = var_9284_end_mask_0, x = k_19_cast_fp16)[name = string("op_9284_cast_fp16")]; + tensor var_9288_begin_0 = const()[name = string("op_9288_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9288_end_0 = const()[name = string("op_9288_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_9288_end_mask_0 = const()[name = string("op_9288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9288_cast_fp16 = slice_by_index(begin = var_9288_begin_0, end = var_9288_end_0, end_mask = var_9288_end_mask_0, x = k_19_cast_fp16)[name = string("op_9288_cast_fp16")]; + tensor var_9292_begin_0 = const()[name = string("op_9292_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9292_end_0 = const()[name = string("op_9292_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_9292_end_mask_0 = const()[name = string("op_9292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9292_cast_fp16 = slice_by_index(begin = var_9292_begin_0, end = var_9292_end_0, end_mask = var_9292_end_mask_0, x = k_19_cast_fp16)[name = string("op_9292_cast_fp16")]; + tensor var_9294_begin_0 = const()[name = string("op_9294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9294_end_0 = const()[name = string("op_9294_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9294_end_mask_0 = const()[name = string("op_9294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9294_cast_fp16 = slice_by_index(begin = var_9294_begin_0, end = var_9294_end_0, end_mask = var_9294_end_mask_0, x = value_19_cast_fp16)[name = string("op_9294_cast_fp16")]; + tensor var_9298_begin_0 = const()[name = string("op_9298_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9298_end_0 = const()[name = string("op_9298_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9298_end_mask_0 = const()[name = string("op_9298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9298_cast_fp16 = slice_by_index(begin = var_9298_begin_0, end = var_9298_end_0, end_mask = var_9298_end_mask_0, x = value_19_cast_fp16)[name = string("op_9298_cast_fp16")]; + tensor var_9302_begin_0 = const()[name = string("op_9302_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9302_end_0 = const()[name = string("op_9302_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9302_end_mask_0 = const()[name = string("op_9302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9302_cast_fp16 = slice_by_index(begin = var_9302_begin_0, end = var_9302_end_0, end_mask = var_9302_end_mask_0, x = value_19_cast_fp16)[name = string("op_9302_cast_fp16")]; + tensor var_9306_begin_0 = const()[name = string("op_9306_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9306_end_0 = const()[name = string("op_9306_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9306_end_mask_0 = const()[name = string("op_9306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9306_cast_fp16 = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = value_19_cast_fp16)[name = string("op_9306_cast_fp16")]; + tensor var_9310_begin_0 = const()[name = string("op_9310_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9310_end_0 = const()[name = string("op_9310_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9310_end_mask_0 = const()[name = string("op_9310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9310_cast_fp16 = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = value_19_cast_fp16)[name = string("op_9310_cast_fp16")]; + tensor var_9314_begin_0 = const()[name = string("op_9314_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9314_end_0 = const()[name = string("op_9314_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9314_end_mask_0 = const()[name = string("op_9314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9314_cast_fp16 = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = value_19_cast_fp16)[name = string("op_9314_cast_fp16")]; + tensor var_9318_begin_0 = const()[name = string("op_9318_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9318_end_0 = const()[name = string("op_9318_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9318_end_mask_0 = const()[name = string("op_9318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9318_cast_fp16 = slice_by_index(begin = var_9318_begin_0, end = var_9318_end_0, end_mask = var_9318_end_mask_0, x = value_19_cast_fp16)[name = string("op_9318_cast_fp16")]; + tensor var_9322_begin_0 = const()[name = string("op_9322_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9322_end_0 = const()[name = string("op_9322_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9322_end_mask_0 = const()[name = string("op_9322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9322_cast_fp16 = slice_by_index(begin = var_9322_begin_0, end = var_9322_end_0, end_mask = var_9322_end_mask_0, x = value_19_cast_fp16)[name = string("op_9322_cast_fp16")]; + tensor var_9326_begin_0 = const()[name = string("op_9326_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9326_end_0 = const()[name = string("op_9326_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9326_end_mask_0 = const()[name = string("op_9326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9326_cast_fp16 = slice_by_index(begin = var_9326_begin_0, end = var_9326_end_0, end_mask = var_9326_end_mask_0, x = value_19_cast_fp16)[name = string("op_9326_cast_fp16")]; + tensor var_9330_begin_0 = const()[name = string("op_9330_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9330_end_0 = const()[name = string("op_9330_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9330_end_mask_0 = const()[name = string("op_9330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9330_cast_fp16 = slice_by_index(begin = var_9330_begin_0, end = var_9330_end_0, end_mask = var_9330_end_mask_0, x = value_19_cast_fp16)[name = string("op_9330_cast_fp16")]; + tensor var_9334_begin_0 = const()[name = string("op_9334_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9334_end_0 = const()[name = string("op_9334_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9334_end_mask_0 = const()[name = string("op_9334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9334_cast_fp16 = slice_by_index(begin = var_9334_begin_0, end = var_9334_end_0, end_mask = var_9334_end_mask_0, x = value_19_cast_fp16)[name = string("op_9334_cast_fp16")]; + tensor var_9338_begin_0 = const()[name = string("op_9338_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9338_end_0 = const()[name = string("op_9338_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9338_end_mask_0 = const()[name = string("op_9338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9338_cast_fp16 = slice_by_index(begin = var_9338_begin_0, end = var_9338_end_0, end_mask = var_9338_end_mask_0, x = value_19_cast_fp16)[name = string("op_9338_cast_fp16")]; + string _SplitHeadsQ__mh_w_865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_865_equation_0, values = (var_9248_cast_fp16, var_8914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_865_cast_fp16")]; + string _SplitHeadsQ__mh_w_867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_867_equation_0, values = (var_9248_cast_fp16, var_8921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_867_cast_fp16")]; + string _SplitHeadsQ__mh_w_869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_869_equation_0, values = (var_9248_cast_fp16, var_8928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_869_cast_fp16")]; + string _SplitHeadsQ__mh_w_871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_871_equation_0, values = (var_9248_cast_fp16, var_8935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_871_cast_fp16")]; + string _SplitHeadsQ__mh_w_873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_873_equation_0, values = (var_9252_cast_fp16, var_8942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_873_cast_fp16")]; + string _SplitHeadsQ__mh_w_875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_875_equation_0, values = (var_9252_cast_fp16, var_8949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_875_cast_fp16")]; + string _SplitHeadsQ__mh_w_877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_877_equation_0, values = (var_9252_cast_fp16, var_8956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_877_cast_fp16")]; + string _SplitHeadsQ__mh_w_879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_879_equation_0, values = (var_9252_cast_fp16, var_8963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_879_cast_fp16")]; + string _SplitHeadsQ__mh_w_881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_881_equation_0, values = (var_9256_cast_fp16, var_8970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_881_cast_fp16")]; + string _SplitHeadsQ__mh_w_883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_883_equation_0, values = (var_9256_cast_fp16, var_8977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_883_cast_fp16")]; + string _SplitHeadsQ__mh_w_885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_885_equation_0, values = (var_9256_cast_fp16, var_8984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_885_cast_fp16")]; + string _SplitHeadsQ__mh_w_887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_887_equation_0, values = (var_9256_cast_fp16, var_8991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_887_cast_fp16")]; + string _SplitHeadsQ__mh_w_889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_889_equation_0, values = (var_9260_cast_fp16, var_8998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_889_cast_fp16")]; + string _SplitHeadsQ__mh_w_891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_891_equation_0, values = (var_9260_cast_fp16, var_9005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_891_cast_fp16")]; + string _SplitHeadsQ__mh_w_893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_893_equation_0, values = (var_9260_cast_fp16, var_9012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_893_cast_fp16")]; + string _SplitHeadsQ__mh_w_895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_895_equation_0, values = (var_9260_cast_fp16, var_9019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_895_cast_fp16")]; + string _SplitHeadsQ__mh_w_897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_897_equation_0, values = (var_9264_cast_fp16, var_9026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_897_cast_fp16")]; + string _SplitHeadsQ__mh_w_899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_899_equation_0, values = (var_9264_cast_fp16, var_9033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_899_cast_fp16")]; + string _SplitHeadsQ__mh_w_901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_901_equation_0, values = (var_9264_cast_fp16, var_9040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_901_cast_fp16")]; + string _SplitHeadsQ__mh_w_903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_903_equation_0, values = (var_9264_cast_fp16, var_9047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_903_cast_fp16")]; + string _SplitHeadsQ__mh_w_905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_905_equation_0, values = (var_9268_cast_fp16, var_9054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_905_cast_fp16")]; + string _SplitHeadsQ__mh_w_907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_907_equation_0, values = (var_9268_cast_fp16, var_9061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_907_cast_fp16")]; + string _SplitHeadsQ__mh_w_909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_909_equation_0, values = (var_9268_cast_fp16, var_9068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_909_cast_fp16")]; + string _SplitHeadsQ__mh_w_911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_911_equation_0, values = (var_9268_cast_fp16, var_9075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_911_cast_fp16")]; + string _SplitHeadsQ__mh_w_913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_913_equation_0, values = (var_9272_cast_fp16, var_9082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_913_cast_fp16")]; + string _SplitHeadsQ__mh_w_915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_915_equation_0, values = (var_9272_cast_fp16, var_9089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_915_cast_fp16")]; + string _SplitHeadsQ__mh_w_917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_917_equation_0, values = (var_9272_cast_fp16, var_9096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_917_cast_fp16")]; + string _SplitHeadsQ__mh_w_919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_919_equation_0, values = (var_9272_cast_fp16, var_9103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_919_cast_fp16")]; + string _SplitHeadsQ__mh_w_921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_921_equation_0, values = (var_9276_cast_fp16, var_9110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_921_cast_fp16")]; + string _SplitHeadsQ__mh_w_923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_923_equation_0, values = (var_9276_cast_fp16, var_9117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_923_cast_fp16")]; + string _SplitHeadsQ__mh_w_925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_925_equation_0, values = (var_9276_cast_fp16, var_9124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_925_cast_fp16")]; + string _SplitHeadsQ__mh_w_927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_927_equation_0, values = (var_9276_cast_fp16, var_9131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_927_cast_fp16")]; + string _SplitHeadsQ__mh_w_929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_929_equation_0, values = (var_9280_cast_fp16, var_9138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_929_cast_fp16")]; + string _SplitHeadsQ__mh_w_931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_931_equation_0, values = (var_9280_cast_fp16, var_9145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_931_cast_fp16")]; + string _SplitHeadsQ__mh_w_933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_933_equation_0, values = (var_9280_cast_fp16, var_9152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_933_cast_fp16")]; + string _SplitHeadsQ__mh_w_935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_935_equation_0, values = (var_9280_cast_fp16, var_9159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_935_cast_fp16")]; + string _SplitHeadsQ__mh_w_937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_937_equation_0, values = (var_9284_cast_fp16, var_9166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_937_cast_fp16")]; + string _SplitHeadsQ__mh_w_939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_939_equation_0, values = (var_9284_cast_fp16, var_9173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_939_cast_fp16")]; + string _SplitHeadsQ__mh_w_941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_941_equation_0, values = (var_9284_cast_fp16, var_9180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_941_cast_fp16")]; + string _SplitHeadsQ__mh_w_943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_943_equation_0, values = (var_9284_cast_fp16, var_9187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_943_cast_fp16")]; + string _SplitHeadsQ__mh_w_945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_945_equation_0, values = (var_9288_cast_fp16, var_9194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_945_cast_fp16")]; + string _SplitHeadsQ__mh_w_947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_947_equation_0, values = (var_9288_cast_fp16, var_9201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_947_cast_fp16")]; + string _SplitHeadsQ__mh_w_949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_949_equation_0, values = (var_9288_cast_fp16, var_9208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_949_cast_fp16")]; + string _SplitHeadsQ__mh_w_951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_951_equation_0, values = (var_9288_cast_fp16, var_9215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_951_cast_fp16")]; + string _SplitHeadsQ__mh_w_953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_953_equation_0, values = (var_9292_cast_fp16, var_9222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_953_cast_fp16")]; + string _SplitHeadsQ__mh_w_955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_955_equation_0, values = (var_9292_cast_fp16, var_9229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_955_cast_fp16")]; + string _SplitHeadsQ__mh_w_957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_957_equation_0, values = (var_9292_cast_fp16, var_9236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_957_cast_fp16")]; + string _SplitHeadsQ__mh_w_959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_959_equation_0, values = (var_9292_cast_fp16, var_9243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_959_cast_fp16")]; + fp16 var_9437_to_fp16 = const()[name = string("op_9437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_865_cast_fp16, y = var_9437_to_fp16)[name = string("aw_chunk_865_cast_fp16")]; + fp16 var_9439_to_fp16 = const()[name = string("op_9439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_867_cast_fp16, y = var_9439_to_fp16)[name = string("aw_chunk_867_cast_fp16")]; + fp16 var_9441_to_fp16 = const()[name = string("op_9441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_869_cast_fp16, y = var_9441_to_fp16)[name = string("aw_chunk_869_cast_fp16")]; + fp16 var_9443_to_fp16 = const()[name = string("op_9443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_871_cast_fp16, y = var_9443_to_fp16)[name = string("aw_chunk_871_cast_fp16")]; + fp16 var_9445_to_fp16 = const()[name = string("op_9445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_873_cast_fp16, y = var_9445_to_fp16)[name = string("aw_chunk_873_cast_fp16")]; + fp16 var_9447_to_fp16 = const()[name = string("op_9447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_875_cast_fp16, y = var_9447_to_fp16)[name = string("aw_chunk_875_cast_fp16")]; + fp16 var_9449_to_fp16 = const()[name = string("op_9449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_877_cast_fp16, y = var_9449_to_fp16)[name = string("aw_chunk_877_cast_fp16")]; + fp16 var_9451_to_fp16 = const()[name = string("op_9451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_879_cast_fp16, y = var_9451_to_fp16)[name = string("aw_chunk_879_cast_fp16")]; + fp16 var_9453_to_fp16 = const()[name = string("op_9453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_881_cast_fp16, y = var_9453_to_fp16)[name = string("aw_chunk_881_cast_fp16")]; + fp16 var_9455_to_fp16 = const()[name = string("op_9455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_883_cast_fp16, y = var_9455_to_fp16)[name = string("aw_chunk_883_cast_fp16")]; + fp16 var_9457_to_fp16 = const()[name = string("op_9457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_885_cast_fp16, y = var_9457_to_fp16)[name = string("aw_chunk_885_cast_fp16")]; + fp16 var_9459_to_fp16 = const()[name = string("op_9459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_887_cast_fp16, y = var_9459_to_fp16)[name = string("aw_chunk_887_cast_fp16")]; + fp16 var_9461_to_fp16 = const()[name = string("op_9461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_889_cast_fp16, y = var_9461_to_fp16)[name = string("aw_chunk_889_cast_fp16")]; + fp16 var_9463_to_fp16 = const()[name = string("op_9463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_891_cast_fp16, y = var_9463_to_fp16)[name = string("aw_chunk_891_cast_fp16")]; + fp16 var_9465_to_fp16 = const()[name = string("op_9465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_893_cast_fp16, y = var_9465_to_fp16)[name = string("aw_chunk_893_cast_fp16")]; + fp16 var_9467_to_fp16 = const()[name = string("op_9467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_895_cast_fp16, y = var_9467_to_fp16)[name = string("aw_chunk_895_cast_fp16")]; + fp16 var_9469_to_fp16 = const()[name = string("op_9469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_897_cast_fp16, y = var_9469_to_fp16)[name = string("aw_chunk_897_cast_fp16")]; + fp16 var_9471_to_fp16 = const()[name = string("op_9471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_899_cast_fp16, y = var_9471_to_fp16)[name = string("aw_chunk_899_cast_fp16")]; + fp16 var_9473_to_fp16 = const()[name = string("op_9473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_901_cast_fp16, y = var_9473_to_fp16)[name = string("aw_chunk_901_cast_fp16")]; + fp16 var_9475_to_fp16 = const()[name = string("op_9475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_903_cast_fp16, y = var_9475_to_fp16)[name = string("aw_chunk_903_cast_fp16")]; + fp16 var_9477_to_fp16 = const()[name = string("op_9477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_905_cast_fp16, y = var_9477_to_fp16)[name = string("aw_chunk_905_cast_fp16")]; + fp16 var_9479_to_fp16 = const()[name = string("op_9479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_907_cast_fp16, y = var_9479_to_fp16)[name = string("aw_chunk_907_cast_fp16")]; + fp16 var_9481_to_fp16 = const()[name = string("op_9481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_909_cast_fp16, y = var_9481_to_fp16)[name = string("aw_chunk_909_cast_fp16")]; + fp16 var_9483_to_fp16 = const()[name = string("op_9483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_911_cast_fp16, y = var_9483_to_fp16)[name = string("aw_chunk_911_cast_fp16")]; + fp16 var_9485_to_fp16 = const()[name = string("op_9485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_913_cast_fp16, y = var_9485_to_fp16)[name = string("aw_chunk_913_cast_fp16")]; + fp16 var_9487_to_fp16 = const()[name = string("op_9487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_915_cast_fp16, y = var_9487_to_fp16)[name = string("aw_chunk_915_cast_fp16")]; + fp16 var_9489_to_fp16 = const()[name = string("op_9489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_917_cast_fp16, y = var_9489_to_fp16)[name = string("aw_chunk_917_cast_fp16")]; + fp16 var_9491_to_fp16 = const()[name = string("op_9491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_919_cast_fp16, y = var_9491_to_fp16)[name = string("aw_chunk_919_cast_fp16")]; + fp16 var_9493_to_fp16 = const()[name = string("op_9493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_921_cast_fp16, y = var_9493_to_fp16)[name = string("aw_chunk_921_cast_fp16")]; + fp16 var_9495_to_fp16 = const()[name = string("op_9495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_923_cast_fp16, y = var_9495_to_fp16)[name = string("aw_chunk_923_cast_fp16")]; + fp16 var_9497_to_fp16 = const()[name = string("op_9497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_925_cast_fp16, y = var_9497_to_fp16)[name = string("aw_chunk_925_cast_fp16")]; + fp16 var_9499_to_fp16 = const()[name = string("op_9499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_927_cast_fp16, y = var_9499_to_fp16)[name = string("aw_chunk_927_cast_fp16")]; + fp16 var_9501_to_fp16 = const()[name = string("op_9501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_929_cast_fp16, y = var_9501_to_fp16)[name = string("aw_chunk_929_cast_fp16")]; + fp16 var_9503_to_fp16 = const()[name = string("op_9503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_931_cast_fp16, y = var_9503_to_fp16)[name = string("aw_chunk_931_cast_fp16")]; + fp16 var_9505_to_fp16 = const()[name = string("op_9505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_933_cast_fp16, y = var_9505_to_fp16)[name = string("aw_chunk_933_cast_fp16")]; + fp16 var_9507_to_fp16 = const()[name = string("op_9507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_935_cast_fp16, y = var_9507_to_fp16)[name = string("aw_chunk_935_cast_fp16")]; + fp16 var_9509_to_fp16 = const()[name = string("op_9509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_937_cast_fp16, y = var_9509_to_fp16)[name = string("aw_chunk_937_cast_fp16")]; + fp16 var_9511_to_fp16 = const()[name = string("op_9511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_939_cast_fp16, y = var_9511_to_fp16)[name = string("aw_chunk_939_cast_fp16")]; + fp16 var_9513_to_fp16 = const()[name = string("op_9513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_941_cast_fp16, y = var_9513_to_fp16)[name = string("aw_chunk_941_cast_fp16")]; + fp16 var_9515_to_fp16 = const()[name = string("op_9515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_943_cast_fp16, y = var_9515_to_fp16)[name = string("aw_chunk_943_cast_fp16")]; + fp16 var_9517_to_fp16 = const()[name = string("op_9517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_945_cast_fp16, y = var_9517_to_fp16)[name = string("aw_chunk_945_cast_fp16")]; + fp16 var_9519_to_fp16 = const()[name = string("op_9519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_947_cast_fp16, y = var_9519_to_fp16)[name = string("aw_chunk_947_cast_fp16")]; + fp16 var_9521_to_fp16 = const()[name = string("op_9521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_949_cast_fp16, y = var_9521_to_fp16)[name = string("aw_chunk_949_cast_fp16")]; + fp16 var_9523_to_fp16 = const()[name = string("op_9523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_951_cast_fp16, y = var_9523_to_fp16)[name = string("aw_chunk_951_cast_fp16")]; + fp16 var_9525_to_fp16 = const()[name = string("op_9525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_953_cast_fp16, y = var_9525_to_fp16)[name = string("aw_chunk_953_cast_fp16")]; + fp16 var_9527_to_fp16 = const()[name = string("op_9527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_955_cast_fp16, y = var_9527_to_fp16)[name = string("aw_chunk_955_cast_fp16")]; + fp16 var_9529_to_fp16 = const()[name = string("op_9529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_957_cast_fp16, y = var_9529_to_fp16)[name = string("aw_chunk_957_cast_fp16")]; + fp16 var_9531_to_fp16 = const()[name = string("op_9531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_959_cast_fp16, y = var_9531_to_fp16)[name = string("aw_chunk_959_cast_fp16")]; + tensor var_9533_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_865_cast_fp16)[name = string("op_9533_cast_fp16")]; + tensor var_9534_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_867_cast_fp16)[name = string("op_9534_cast_fp16")]; + tensor var_9535_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_869_cast_fp16)[name = string("op_9535_cast_fp16")]; + tensor var_9536_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_871_cast_fp16)[name = string("op_9536_cast_fp16")]; + tensor var_9537_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_873_cast_fp16)[name = string("op_9537_cast_fp16")]; + tensor var_9538_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_875_cast_fp16)[name = string("op_9538_cast_fp16")]; + tensor var_9539_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_877_cast_fp16)[name = string("op_9539_cast_fp16")]; + tensor var_9540_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_879_cast_fp16)[name = string("op_9540_cast_fp16")]; + tensor var_9541_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_881_cast_fp16)[name = string("op_9541_cast_fp16")]; + tensor var_9542_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_883_cast_fp16)[name = string("op_9542_cast_fp16")]; + tensor var_9543_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_885_cast_fp16)[name = string("op_9543_cast_fp16")]; + tensor var_9544_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_887_cast_fp16)[name = string("op_9544_cast_fp16")]; + tensor var_9545_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_889_cast_fp16)[name = string("op_9545_cast_fp16")]; + tensor var_9546_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_891_cast_fp16)[name = string("op_9546_cast_fp16")]; + tensor var_9547_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_893_cast_fp16)[name = string("op_9547_cast_fp16")]; + tensor var_9548_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_895_cast_fp16)[name = string("op_9548_cast_fp16")]; + tensor var_9549_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_897_cast_fp16)[name = string("op_9549_cast_fp16")]; + tensor var_9550_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_899_cast_fp16)[name = string("op_9550_cast_fp16")]; + tensor var_9551_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_901_cast_fp16)[name = string("op_9551_cast_fp16")]; + tensor var_9552_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_903_cast_fp16)[name = string("op_9552_cast_fp16")]; + tensor var_9553_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_905_cast_fp16)[name = string("op_9553_cast_fp16")]; + tensor var_9554_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_907_cast_fp16)[name = string("op_9554_cast_fp16")]; + tensor var_9555_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_909_cast_fp16)[name = string("op_9555_cast_fp16")]; + tensor var_9556_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_911_cast_fp16)[name = string("op_9556_cast_fp16")]; + tensor var_9557_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_913_cast_fp16)[name = string("op_9557_cast_fp16")]; + tensor var_9558_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_915_cast_fp16)[name = string("op_9558_cast_fp16")]; + tensor var_9559_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_917_cast_fp16)[name = string("op_9559_cast_fp16")]; + tensor var_9560_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_919_cast_fp16)[name = string("op_9560_cast_fp16")]; + tensor var_9561_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_921_cast_fp16)[name = string("op_9561_cast_fp16")]; + tensor var_9562_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_923_cast_fp16)[name = string("op_9562_cast_fp16")]; + tensor var_9563_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_925_cast_fp16)[name = string("op_9563_cast_fp16")]; + tensor var_9564_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_927_cast_fp16)[name = string("op_9564_cast_fp16")]; + tensor var_9565_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_929_cast_fp16)[name = string("op_9565_cast_fp16")]; + tensor var_9566_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_931_cast_fp16)[name = string("op_9566_cast_fp16")]; + tensor var_9567_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_933_cast_fp16)[name = string("op_9567_cast_fp16")]; + tensor var_9568_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_935_cast_fp16)[name = string("op_9568_cast_fp16")]; + tensor var_9569_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_937_cast_fp16)[name = string("op_9569_cast_fp16")]; + tensor var_9570_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_939_cast_fp16)[name = string("op_9570_cast_fp16")]; + tensor var_9571_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_941_cast_fp16)[name = string("op_9571_cast_fp16")]; + tensor var_9572_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_943_cast_fp16)[name = string("op_9572_cast_fp16")]; + tensor var_9573_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_945_cast_fp16)[name = string("op_9573_cast_fp16")]; + tensor var_9574_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_947_cast_fp16)[name = string("op_9574_cast_fp16")]; + tensor var_9575_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_949_cast_fp16)[name = string("op_9575_cast_fp16")]; + tensor var_9576_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_951_cast_fp16)[name = string("op_9576_cast_fp16")]; + tensor var_9577_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_953_cast_fp16)[name = string("op_9577_cast_fp16")]; + tensor var_9578_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_955_cast_fp16)[name = string("op_9578_cast_fp16")]; + tensor var_9579_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_957_cast_fp16)[name = string("op_9579_cast_fp16")]; + tensor var_9580_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_959_cast_fp16)[name = string("op_9580_cast_fp16")]; + string var_9582_equation_0 = const()[name = string("op_9582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9582_cast_fp16 = einsum(equation = var_9582_equation_0, values = (var_9294_cast_fp16, var_9533_cast_fp16))[name = string("op_9582_cast_fp16")]; + string var_9584_equation_0 = const()[name = string("op_9584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9584_cast_fp16 = einsum(equation = var_9584_equation_0, values = (var_9294_cast_fp16, var_9534_cast_fp16))[name = string("op_9584_cast_fp16")]; + string var_9586_equation_0 = const()[name = string("op_9586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9586_cast_fp16 = einsum(equation = var_9586_equation_0, values = (var_9294_cast_fp16, var_9535_cast_fp16))[name = string("op_9586_cast_fp16")]; + string var_9588_equation_0 = const()[name = string("op_9588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9588_cast_fp16 = einsum(equation = var_9588_equation_0, values = (var_9294_cast_fp16, var_9536_cast_fp16))[name = string("op_9588_cast_fp16")]; + string var_9590_equation_0 = const()[name = string("op_9590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9590_cast_fp16 = einsum(equation = var_9590_equation_0, values = (var_9298_cast_fp16, var_9537_cast_fp16))[name = string("op_9590_cast_fp16")]; + string var_9592_equation_0 = const()[name = string("op_9592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9592_cast_fp16 = einsum(equation = var_9592_equation_0, values = (var_9298_cast_fp16, var_9538_cast_fp16))[name = string("op_9592_cast_fp16")]; + string var_9594_equation_0 = const()[name = string("op_9594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9594_cast_fp16 = einsum(equation = var_9594_equation_0, values = (var_9298_cast_fp16, var_9539_cast_fp16))[name = string("op_9594_cast_fp16")]; + string var_9596_equation_0 = const()[name = string("op_9596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9596_cast_fp16 = einsum(equation = var_9596_equation_0, values = (var_9298_cast_fp16, var_9540_cast_fp16))[name = string("op_9596_cast_fp16")]; + string var_9598_equation_0 = const()[name = string("op_9598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9598_cast_fp16 = einsum(equation = var_9598_equation_0, values = (var_9302_cast_fp16, var_9541_cast_fp16))[name = string("op_9598_cast_fp16")]; + string var_9600_equation_0 = const()[name = string("op_9600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9600_cast_fp16 = einsum(equation = var_9600_equation_0, values = (var_9302_cast_fp16, var_9542_cast_fp16))[name = string("op_9600_cast_fp16")]; + string var_9602_equation_0 = const()[name = string("op_9602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9602_cast_fp16 = einsum(equation = var_9602_equation_0, values = (var_9302_cast_fp16, var_9543_cast_fp16))[name = string("op_9602_cast_fp16")]; + string var_9604_equation_0 = const()[name = string("op_9604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9604_cast_fp16 = einsum(equation = var_9604_equation_0, values = (var_9302_cast_fp16, var_9544_cast_fp16))[name = string("op_9604_cast_fp16")]; + string var_9606_equation_0 = const()[name = string("op_9606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9606_cast_fp16 = einsum(equation = var_9606_equation_0, values = (var_9306_cast_fp16, var_9545_cast_fp16))[name = string("op_9606_cast_fp16")]; + string var_9608_equation_0 = const()[name = string("op_9608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9608_cast_fp16 = einsum(equation = var_9608_equation_0, values = (var_9306_cast_fp16, var_9546_cast_fp16))[name = string("op_9608_cast_fp16")]; + string var_9610_equation_0 = const()[name = string("op_9610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9610_cast_fp16 = einsum(equation = var_9610_equation_0, values = (var_9306_cast_fp16, var_9547_cast_fp16))[name = string("op_9610_cast_fp16")]; + string var_9612_equation_0 = const()[name = string("op_9612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9612_cast_fp16 = einsum(equation = var_9612_equation_0, values = (var_9306_cast_fp16, var_9548_cast_fp16))[name = string("op_9612_cast_fp16")]; + string var_9614_equation_0 = const()[name = string("op_9614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9614_cast_fp16 = einsum(equation = var_9614_equation_0, values = (var_9310_cast_fp16, var_9549_cast_fp16))[name = string("op_9614_cast_fp16")]; + string var_9616_equation_0 = const()[name = string("op_9616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9616_cast_fp16 = einsum(equation = var_9616_equation_0, values = (var_9310_cast_fp16, var_9550_cast_fp16))[name = string("op_9616_cast_fp16")]; + string var_9618_equation_0 = const()[name = string("op_9618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9618_cast_fp16 = einsum(equation = var_9618_equation_0, values = (var_9310_cast_fp16, var_9551_cast_fp16))[name = string("op_9618_cast_fp16")]; + string var_9620_equation_0 = const()[name = string("op_9620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9620_cast_fp16 = einsum(equation = var_9620_equation_0, values = (var_9310_cast_fp16, var_9552_cast_fp16))[name = string("op_9620_cast_fp16")]; + string var_9622_equation_0 = const()[name = string("op_9622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9622_cast_fp16 = einsum(equation = var_9622_equation_0, values = (var_9314_cast_fp16, var_9553_cast_fp16))[name = string("op_9622_cast_fp16")]; + string var_9624_equation_0 = const()[name = string("op_9624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9624_cast_fp16 = einsum(equation = var_9624_equation_0, values = (var_9314_cast_fp16, var_9554_cast_fp16))[name = string("op_9624_cast_fp16")]; + string var_9626_equation_0 = const()[name = string("op_9626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9626_cast_fp16 = einsum(equation = var_9626_equation_0, values = (var_9314_cast_fp16, var_9555_cast_fp16))[name = string("op_9626_cast_fp16")]; + string var_9628_equation_0 = const()[name = string("op_9628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9628_cast_fp16 = einsum(equation = var_9628_equation_0, values = (var_9314_cast_fp16, var_9556_cast_fp16))[name = string("op_9628_cast_fp16")]; + string var_9630_equation_0 = const()[name = string("op_9630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9630_cast_fp16 = einsum(equation = var_9630_equation_0, values = (var_9318_cast_fp16, var_9557_cast_fp16))[name = string("op_9630_cast_fp16")]; + string var_9632_equation_0 = const()[name = string("op_9632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9632_cast_fp16 = einsum(equation = var_9632_equation_0, values = (var_9318_cast_fp16, var_9558_cast_fp16))[name = string("op_9632_cast_fp16")]; + string var_9634_equation_0 = const()[name = string("op_9634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9634_cast_fp16 = einsum(equation = var_9634_equation_0, values = (var_9318_cast_fp16, var_9559_cast_fp16))[name = string("op_9634_cast_fp16")]; + string var_9636_equation_0 = const()[name = string("op_9636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9636_cast_fp16 = einsum(equation = var_9636_equation_0, values = (var_9318_cast_fp16, var_9560_cast_fp16))[name = string("op_9636_cast_fp16")]; + string var_9638_equation_0 = const()[name = string("op_9638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9638_cast_fp16 = einsum(equation = var_9638_equation_0, values = (var_9322_cast_fp16, var_9561_cast_fp16))[name = string("op_9638_cast_fp16")]; + string var_9640_equation_0 = const()[name = string("op_9640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9640_cast_fp16 = einsum(equation = var_9640_equation_0, values = (var_9322_cast_fp16, var_9562_cast_fp16))[name = string("op_9640_cast_fp16")]; + string var_9642_equation_0 = const()[name = string("op_9642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9642_cast_fp16 = einsum(equation = var_9642_equation_0, values = (var_9322_cast_fp16, var_9563_cast_fp16))[name = string("op_9642_cast_fp16")]; + string var_9644_equation_0 = const()[name = string("op_9644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9644_cast_fp16 = einsum(equation = var_9644_equation_0, values = (var_9322_cast_fp16, var_9564_cast_fp16))[name = string("op_9644_cast_fp16")]; + string var_9646_equation_0 = const()[name = string("op_9646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9646_cast_fp16 = einsum(equation = var_9646_equation_0, values = (var_9326_cast_fp16, var_9565_cast_fp16))[name = string("op_9646_cast_fp16")]; + string var_9648_equation_0 = const()[name = string("op_9648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9648_cast_fp16 = einsum(equation = var_9648_equation_0, values = (var_9326_cast_fp16, var_9566_cast_fp16))[name = string("op_9648_cast_fp16")]; + string var_9650_equation_0 = const()[name = string("op_9650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9650_cast_fp16 = einsum(equation = var_9650_equation_0, values = (var_9326_cast_fp16, var_9567_cast_fp16))[name = string("op_9650_cast_fp16")]; + string var_9652_equation_0 = const()[name = string("op_9652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9652_cast_fp16 = einsum(equation = var_9652_equation_0, values = (var_9326_cast_fp16, var_9568_cast_fp16))[name = string("op_9652_cast_fp16")]; + string var_9654_equation_0 = const()[name = string("op_9654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9654_cast_fp16 = einsum(equation = var_9654_equation_0, values = (var_9330_cast_fp16, var_9569_cast_fp16))[name = string("op_9654_cast_fp16")]; + string var_9656_equation_0 = const()[name = string("op_9656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9656_cast_fp16 = einsum(equation = var_9656_equation_0, values = (var_9330_cast_fp16, var_9570_cast_fp16))[name = string("op_9656_cast_fp16")]; + string var_9658_equation_0 = const()[name = string("op_9658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9658_cast_fp16 = einsum(equation = var_9658_equation_0, values = (var_9330_cast_fp16, var_9571_cast_fp16))[name = string("op_9658_cast_fp16")]; + string var_9660_equation_0 = const()[name = string("op_9660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9660_cast_fp16 = einsum(equation = var_9660_equation_0, values = (var_9330_cast_fp16, var_9572_cast_fp16))[name = string("op_9660_cast_fp16")]; + string var_9662_equation_0 = const()[name = string("op_9662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9662_cast_fp16 = einsum(equation = var_9662_equation_0, values = (var_9334_cast_fp16, var_9573_cast_fp16))[name = string("op_9662_cast_fp16")]; + string var_9664_equation_0 = const()[name = string("op_9664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9664_cast_fp16 = einsum(equation = var_9664_equation_0, values = (var_9334_cast_fp16, var_9574_cast_fp16))[name = string("op_9664_cast_fp16")]; + string var_9666_equation_0 = const()[name = string("op_9666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9666_cast_fp16 = einsum(equation = var_9666_equation_0, values = (var_9334_cast_fp16, var_9575_cast_fp16))[name = string("op_9666_cast_fp16")]; + string var_9668_equation_0 = const()[name = string("op_9668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9668_cast_fp16 = einsum(equation = var_9668_equation_0, values = (var_9334_cast_fp16, var_9576_cast_fp16))[name = string("op_9668_cast_fp16")]; + string var_9670_equation_0 = const()[name = string("op_9670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9670_cast_fp16 = einsum(equation = var_9670_equation_0, values = (var_9338_cast_fp16, var_9577_cast_fp16))[name = string("op_9670_cast_fp16")]; + string var_9672_equation_0 = const()[name = string("op_9672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9672_cast_fp16 = einsum(equation = var_9672_equation_0, values = (var_9338_cast_fp16, var_9578_cast_fp16))[name = string("op_9672_cast_fp16")]; + string var_9674_equation_0 = const()[name = string("op_9674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9674_cast_fp16 = einsum(equation = var_9674_equation_0, values = (var_9338_cast_fp16, var_9579_cast_fp16))[name = string("op_9674_cast_fp16")]; + string var_9676_equation_0 = const()[name = string("op_9676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9676_cast_fp16 = einsum(equation = var_9676_equation_0, values = (var_9338_cast_fp16, var_9580_cast_fp16))[name = string("op_9676_cast_fp16")]; + bool var_9678_interleave_0 = const()[name = string("op_9678_interleave_0"), val = bool(false)]; + tensor var_9678_cast_fp16 = concat(axis = var_8789, interleave = var_9678_interleave_0, values = (var_9582_cast_fp16, var_9584_cast_fp16, var_9586_cast_fp16, var_9588_cast_fp16))[name = string("op_9678_cast_fp16")]; + bool var_9680_interleave_0 = const()[name = string("op_9680_interleave_0"), val = bool(false)]; + tensor var_9680_cast_fp16 = concat(axis = var_8789, interleave = var_9680_interleave_0, values = (var_9590_cast_fp16, var_9592_cast_fp16, var_9594_cast_fp16, var_9596_cast_fp16))[name = string("op_9680_cast_fp16")]; + bool var_9682_interleave_0 = const()[name = string("op_9682_interleave_0"), val = bool(false)]; + tensor var_9682_cast_fp16 = concat(axis = var_8789, interleave = var_9682_interleave_0, values = (var_9598_cast_fp16, var_9600_cast_fp16, var_9602_cast_fp16, var_9604_cast_fp16))[name = string("op_9682_cast_fp16")]; + bool var_9684_interleave_0 = const()[name = string("op_9684_interleave_0"), val = bool(false)]; + tensor var_9684_cast_fp16 = concat(axis = var_8789, interleave = var_9684_interleave_0, values = (var_9606_cast_fp16, var_9608_cast_fp16, var_9610_cast_fp16, var_9612_cast_fp16))[name = string("op_9684_cast_fp16")]; + bool var_9686_interleave_0 = const()[name = string("op_9686_interleave_0"), val = bool(false)]; + tensor var_9686_cast_fp16 = concat(axis = var_8789, interleave = var_9686_interleave_0, values = (var_9614_cast_fp16, var_9616_cast_fp16, var_9618_cast_fp16, var_9620_cast_fp16))[name = string("op_9686_cast_fp16")]; + bool var_9688_interleave_0 = const()[name = string("op_9688_interleave_0"), val = bool(false)]; + tensor var_9688_cast_fp16 = concat(axis = var_8789, interleave = var_9688_interleave_0, values = (var_9622_cast_fp16, var_9624_cast_fp16, var_9626_cast_fp16, var_9628_cast_fp16))[name = string("op_9688_cast_fp16")]; + bool var_9690_interleave_0 = const()[name = string("op_9690_interleave_0"), val = bool(false)]; + tensor var_9690_cast_fp16 = concat(axis = var_8789, interleave = var_9690_interleave_0, values = (var_9630_cast_fp16, var_9632_cast_fp16, var_9634_cast_fp16, var_9636_cast_fp16))[name = string("op_9690_cast_fp16")]; + bool var_9692_interleave_0 = const()[name = string("op_9692_interleave_0"), val = bool(false)]; + tensor var_9692_cast_fp16 = concat(axis = var_8789, interleave = var_9692_interleave_0, values = (var_9638_cast_fp16, var_9640_cast_fp16, var_9642_cast_fp16, var_9644_cast_fp16))[name = string("op_9692_cast_fp16")]; + bool var_9694_interleave_0 = const()[name = string("op_9694_interleave_0"), val = bool(false)]; + tensor var_9694_cast_fp16 = concat(axis = var_8789, interleave = var_9694_interleave_0, values = (var_9646_cast_fp16, var_9648_cast_fp16, var_9650_cast_fp16, var_9652_cast_fp16))[name = string("op_9694_cast_fp16")]; + bool var_9696_interleave_0 = const()[name = string("op_9696_interleave_0"), val = bool(false)]; + tensor var_9696_cast_fp16 = concat(axis = var_8789, interleave = var_9696_interleave_0, values = (var_9654_cast_fp16, var_9656_cast_fp16, var_9658_cast_fp16, var_9660_cast_fp16))[name = string("op_9696_cast_fp16")]; + bool var_9698_interleave_0 = const()[name = string("op_9698_interleave_0"), val = bool(false)]; + tensor var_9698_cast_fp16 = concat(axis = var_8789, interleave = var_9698_interleave_0, values = (var_9662_cast_fp16, var_9664_cast_fp16, var_9666_cast_fp16, var_9668_cast_fp16))[name = string("op_9698_cast_fp16")]; + bool var_9700_interleave_0 = const()[name = string("op_9700_interleave_0"), val = bool(false)]; + tensor var_9700_cast_fp16 = concat(axis = var_8789, interleave = var_9700_interleave_0, values = (var_9670_cast_fp16, var_9672_cast_fp16, var_9674_cast_fp16, var_9676_cast_fp16))[name = string("op_9700_cast_fp16")]; + bool input_73_interleave_0 = const()[name = string("input_73_interleave_0"), val = bool(false)]; + tensor input_73_cast_fp16 = concat(axis = var_8806, interleave = input_73_interleave_0, values = (var_9678_cast_fp16, var_9680_cast_fp16, var_9682_cast_fp16, var_9684_cast_fp16, var_9686_cast_fp16, var_9688_cast_fp16, var_9690_cast_fp16, var_9692_cast_fp16, var_9694_cast_fp16, var_9696_cast_fp16, var_9698_cast_fp16, var_9700_cast_fp16))[name = string("input_73_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137340288)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138520000)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_9719_to_fp16 = const()[name = string("op_9719_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_9719_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138521600)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138523200)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138524800)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143243456)))]; + tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143249664)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147968320)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; + int32 var_9748 = const()[name = string("op_9748"), val = int32(3)]; + int32 var_9765 = const()[name = string("op_9765"), val = int32(1)]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_9782_to_fp16 = const()[name = string("op_9782_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_9782_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147969920)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147971520)))]; + fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147973120)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149152832)))]; + tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; + string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; + tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; + tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; + int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149154432)))]; + tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; + string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; + tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; + tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; + int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334144)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151513856)))]; + tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_9820_begin_0 = const()[name = string("op_9820_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9820_end_0 = const()[name = string("op_9820_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9820_end_mask_0 = const()[name = string("op_9820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9820_cast_fp16 = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = query_21_cast_fp16)[name = string("op_9820_cast_fp16")]; + tensor var_9824_begin_0 = const()[name = string("op_9824_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9824_end_0 = const()[name = string("op_9824_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9824_end_mask_0 = const()[name = string("op_9824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9824_cast_fp16 = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = query_21_cast_fp16)[name = string("op_9824_cast_fp16")]; + tensor var_9828_begin_0 = const()[name = string("op_9828_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9828_end_0 = const()[name = string("op_9828_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9828_end_mask_0 = const()[name = string("op_9828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9828_cast_fp16 = slice_by_index(begin = var_9828_begin_0, end = var_9828_end_0, end_mask = var_9828_end_mask_0, x = query_21_cast_fp16)[name = string("op_9828_cast_fp16")]; + tensor var_9832_begin_0 = const()[name = string("op_9832_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9832_end_0 = const()[name = string("op_9832_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9832_end_mask_0 = const()[name = string("op_9832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9832_cast_fp16 = slice_by_index(begin = var_9832_begin_0, end = var_9832_end_0, end_mask = var_9832_end_mask_0, x = query_21_cast_fp16)[name = string("op_9832_cast_fp16")]; + tensor var_9836_begin_0 = const()[name = string("op_9836_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9836_end_0 = const()[name = string("op_9836_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9836_end_mask_0 = const()[name = string("op_9836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9836_cast_fp16 = slice_by_index(begin = var_9836_begin_0, end = var_9836_end_0, end_mask = var_9836_end_mask_0, x = query_21_cast_fp16)[name = string("op_9836_cast_fp16")]; + tensor var_9840_begin_0 = const()[name = string("op_9840_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9840_end_0 = const()[name = string("op_9840_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9840_end_mask_0 = const()[name = string("op_9840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9840_cast_fp16 = slice_by_index(begin = var_9840_begin_0, end = var_9840_end_0, end_mask = var_9840_end_mask_0, x = query_21_cast_fp16)[name = string("op_9840_cast_fp16")]; + tensor var_9844_begin_0 = const()[name = string("op_9844_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9844_end_0 = const()[name = string("op_9844_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9844_end_mask_0 = const()[name = string("op_9844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9844_cast_fp16 = slice_by_index(begin = var_9844_begin_0, end = var_9844_end_0, end_mask = var_9844_end_mask_0, x = query_21_cast_fp16)[name = string("op_9844_cast_fp16")]; + tensor var_9848_begin_0 = const()[name = string("op_9848_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9848_end_0 = const()[name = string("op_9848_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9848_end_mask_0 = const()[name = string("op_9848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9848_cast_fp16 = slice_by_index(begin = var_9848_begin_0, end = var_9848_end_0, end_mask = var_9848_end_mask_0, x = query_21_cast_fp16)[name = string("op_9848_cast_fp16")]; + tensor var_9852_begin_0 = const()[name = string("op_9852_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9852_end_0 = const()[name = string("op_9852_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9852_end_mask_0 = const()[name = string("op_9852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9852_cast_fp16 = slice_by_index(begin = var_9852_begin_0, end = var_9852_end_0, end_mask = var_9852_end_mask_0, x = query_21_cast_fp16)[name = string("op_9852_cast_fp16")]; + tensor var_9856_begin_0 = const()[name = string("op_9856_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9856_end_0 = const()[name = string("op_9856_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9856_end_mask_0 = const()[name = string("op_9856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9856_cast_fp16 = slice_by_index(begin = var_9856_begin_0, end = var_9856_end_0, end_mask = var_9856_end_mask_0, x = query_21_cast_fp16)[name = string("op_9856_cast_fp16")]; + tensor var_9860_begin_0 = const()[name = string("op_9860_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9860_end_0 = const()[name = string("op_9860_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9860_end_mask_0 = const()[name = string("op_9860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9860_cast_fp16 = slice_by_index(begin = var_9860_begin_0, end = var_9860_end_0, end_mask = var_9860_end_mask_0, x = query_21_cast_fp16)[name = string("op_9860_cast_fp16")]; + tensor var_9864_begin_0 = const()[name = string("op_9864_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9864_end_0 = const()[name = string("op_9864_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9864_end_mask_0 = const()[name = string("op_9864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9864_cast_fp16 = slice_by_index(begin = var_9864_begin_0, end = var_9864_end_0, end_mask = var_9864_end_mask_0, x = query_21_cast_fp16)[name = string("op_9864_cast_fp16")]; + tensor var_9873_begin_0 = const()[name = string("op_9873_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9873_end_0 = const()[name = string("op_9873_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9873_end_mask_0 = const()[name = string("op_9873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9873_cast_fp16 = slice_by_index(begin = var_9873_begin_0, end = var_9873_end_0, end_mask = var_9873_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9873_cast_fp16")]; + tensor var_9880_begin_0 = const()[name = string("op_9880_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9880_end_0 = const()[name = string("op_9880_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9880_end_mask_0 = const()[name = string("op_9880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9880_cast_fp16 = slice_by_index(begin = var_9880_begin_0, end = var_9880_end_0, end_mask = var_9880_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9880_cast_fp16")]; + tensor var_9887_begin_0 = const()[name = string("op_9887_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9887_end_0 = const()[name = string("op_9887_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9887_end_mask_0 = const()[name = string("op_9887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9887_cast_fp16 = slice_by_index(begin = var_9887_begin_0, end = var_9887_end_0, end_mask = var_9887_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9887_cast_fp16")]; + tensor var_9894_begin_0 = const()[name = string("op_9894_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9894_end_0 = const()[name = string("op_9894_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9894_end_mask_0 = const()[name = string("op_9894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9894_cast_fp16 = slice_by_index(begin = var_9894_begin_0, end = var_9894_end_0, end_mask = var_9894_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9894_cast_fp16")]; + tensor var_9901_begin_0 = const()[name = string("op_9901_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9901_end_0 = const()[name = string("op_9901_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9901_end_mask_0 = const()[name = string("op_9901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9901_cast_fp16 = slice_by_index(begin = var_9901_begin_0, end = var_9901_end_0, end_mask = var_9901_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9901_cast_fp16")]; + tensor var_9908_begin_0 = const()[name = string("op_9908_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9908_end_0 = const()[name = string("op_9908_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9908_end_mask_0 = const()[name = string("op_9908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9908_cast_fp16 = slice_by_index(begin = var_9908_begin_0, end = var_9908_end_0, end_mask = var_9908_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9908_cast_fp16")]; + tensor var_9915_begin_0 = const()[name = string("op_9915_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9915_end_0 = const()[name = string("op_9915_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9915_end_mask_0 = const()[name = string("op_9915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9915_cast_fp16 = slice_by_index(begin = var_9915_begin_0, end = var_9915_end_0, end_mask = var_9915_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9915_cast_fp16")]; + tensor var_9922_begin_0 = const()[name = string("op_9922_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9922_end_0 = const()[name = string("op_9922_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9922_end_mask_0 = const()[name = string("op_9922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9922_cast_fp16 = slice_by_index(begin = var_9922_begin_0, end = var_9922_end_0, end_mask = var_9922_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9922_cast_fp16")]; + tensor var_9929_begin_0 = const()[name = string("op_9929_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9929_end_0 = const()[name = string("op_9929_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9929_end_mask_0 = const()[name = string("op_9929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9929_cast_fp16 = slice_by_index(begin = var_9929_begin_0, end = var_9929_end_0, end_mask = var_9929_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9929_cast_fp16")]; + tensor var_9936_begin_0 = const()[name = string("op_9936_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9936_end_0 = const()[name = string("op_9936_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9936_end_mask_0 = const()[name = string("op_9936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9936_cast_fp16 = slice_by_index(begin = var_9936_begin_0, end = var_9936_end_0, end_mask = var_9936_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9936_cast_fp16")]; + tensor var_9943_begin_0 = const()[name = string("op_9943_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9943_end_0 = const()[name = string("op_9943_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9943_end_mask_0 = const()[name = string("op_9943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9943_cast_fp16 = slice_by_index(begin = var_9943_begin_0, end = var_9943_end_0, end_mask = var_9943_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9943_cast_fp16")]; + tensor var_9950_begin_0 = const()[name = string("op_9950_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9950_end_0 = const()[name = string("op_9950_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9950_end_mask_0 = const()[name = string("op_9950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9950_cast_fp16 = slice_by_index(begin = var_9950_begin_0, end = var_9950_end_0, end_mask = var_9950_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9950_cast_fp16")]; + tensor var_9957_begin_0 = const()[name = string("op_9957_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9957_end_0 = const()[name = string("op_9957_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9957_end_mask_0 = const()[name = string("op_9957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9957_cast_fp16 = slice_by_index(begin = var_9957_begin_0, end = var_9957_end_0, end_mask = var_9957_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9957_cast_fp16")]; + tensor var_9964_begin_0 = const()[name = string("op_9964_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9964_end_0 = const()[name = string("op_9964_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9964_end_mask_0 = const()[name = string("op_9964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9964_cast_fp16 = slice_by_index(begin = var_9964_begin_0, end = var_9964_end_0, end_mask = var_9964_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9964_cast_fp16")]; + tensor var_9971_begin_0 = const()[name = string("op_9971_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9971_end_0 = const()[name = string("op_9971_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9971_end_mask_0 = const()[name = string("op_9971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9971_cast_fp16 = slice_by_index(begin = var_9971_begin_0, end = var_9971_end_0, end_mask = var_9971_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9971_cast_fp16")]; + tensor var_9978_begin_0 = const()[name = string("op_9978_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9978_end_0 = const()[name = string("op_9978_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9978_end_mask_0 = const()[name = string("op_9978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9978_cast_fp16 = slice_by_index(begin = var_9978_begin_0, end = var_9978_end_0, end_mask = var_9978_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9978_cast_fp16")]; + tensor var_9985_begin_0 = const()[name = string("op_9985_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9985_end_0 = const()[name = string("op_9985_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9985_end_mask_0 = const()[name = string("op_9985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9985_cast_fp16 = slice_by_index(begin = var_9985_begin_0, end = var_9985_end_0, end_mask = var_9985_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9985_cast_fp16")]; + tensor var_9992_begin_0 = const()[name = string("op_9992_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9992_end_0 = const()[name = string("op_9992_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9992_end_mask_0 = const()[name = string("op_9992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9992_cast_fp16 = slice_by_index(begin = var_9992_begin_0, end = var_9992_end_0, end_mask = var_9992_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9992_cast_fp16")]; + tensor var_9999_begin_0 = const()[name = string("op_9999_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9999_end_0 = const()[name = string("op_9999_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9999_end_mask_0 = const()[name = string("op_9999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9999_cast_fp16 = slice_by_index(begin = var_9999_begin_0, end = var_9999_end_0, end_mask = var_9999_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9999_cast_fp16")]; + tensor var_10006_begin_0 = const()[name = string("op_10006_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10006_end_0 = const()[name = string("op_10006_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10006_end_mask_0 = const()[name = string("op_10006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10006_cast_fp16 = slice_by_index(begin = var_10006_begin_0, end = var_10006_end_0, end_mask = var_10006_end_mask_0, x = var_9836_cast_fp16)[name = string("op_10006_cast_fp16")]; + tensor var_10013_begin_0 = const()[name = string("op_10013_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10013_end_0 = const()[name = string("op_10013_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10013_end_mask_0 = const()[name = string("op_10013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10013_cast_fp16 = slice_by_index(begin = var_10013_begin_0, end = var_10013_end_0, end_mask = var_10013_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10013_cast_fp16")]; + tensor var_10020_begin_0 = const()[name = string("op_10020_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10020_end_0 = const()[name = string("op_10020_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10020_end_mask_0 = const()[name = string("op_10020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10020_cast_fp16 = slice_by_index(begin = var_10020_begin_0, end = var_10020_end_0, end_mask = var_10020_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10020_cast_fp16")]; + tensor var_10027_begin_0 = const()[name = string("op_10027_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10027_end_0 = const()[name = string("op_10027_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10027_end_mask_0 = const()[name = string("op_10027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10027_cast_fp16 = slice_by_index(begin = var_10027_begin_0, end = var_10027_end_0, end_mask = var_10027_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10027_cast_fp16")]; + tensor var_10034_begin_0 = const()[name = string("op_10034_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10034_end_0 = const()[name = string("op_10034_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10034_end_mask_0 = const()[name = string("op_10034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10034_cast_fp16 = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10034_cast_fp16")]; + tensor var_10041_begin_0 = const()[name = string("op_10041_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10041_end_0 = const()[name = string("op_10041_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10041_end_mask_0 = const()[name = string("op_10041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10041_cast_fp16 = slice_by_index(begin = var_10041_begin_0, end = var_10041_end_0, end_mask = var_10041_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10041_cast_fp16")]; + tensor var_10048_begin_0 = const()[name = string("op_10048_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10048_end_0 = const()[name = string("op_10048_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10048_end_mask_0 = const()[name = string("op_10048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10048_cast_fp16 = slice_by_index(begin = var_10048_begin_0, end = var_10048_end_0, end_mask = var_10048_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10048_cast_fp16")]; + tensor var_10055_begin_0 = const()[name = string("op_10055_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10055_end_0 = const()[name = string("op_10055_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10055_end_mask_0 = const()[name = string("op_10055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10055_cast_fp16 = slice_by_index(begin = var_10055_begin_0, end = var_10055_end_0, end_mask = var_10055_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10055_cast_fp16")]; + tensor var_10062_begin_0 = const()[name = string("op_10062_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10062_end_0 = const()[name = string("op_10062_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10062_end_mask_0 = const()[name = string("op_10062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10062_cast_fp16 = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10062_cast_fp16")]; + tensor var_10069_begin_0 = const()[name = string("op_10069_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10069_end_0 = const()[name = string("op_10069_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10069_end_mask_0 = const()[name = string("op_10069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10069_cast_fp16 = slice_by_index(begin = var_10069_begin_0, end = var_10069_end_0, end_mask = var_10069_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10069_cast_fp16")]; + tensor var_10076_begin_0 = const()[name = string("op_10076_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10076_end_0 = const()[name = string("op_10076_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10076_end_mask_0 = const()[name = string("op_10076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10076_cast_fp16")]; + tensor var_10083_begin_0 = const()[name = string("op_10083_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10083_end_0 = const()[name = string("op_10083_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10083_end_mask_0 = const()[name = string("op_10083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10083_cast_fp16 = slice_by_index(begin = var_10083_begin_0, end = var_10083_end_0, end_mask = var_10083_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10083_cast_fp16")]; + tensor var_10090_begin_0 = const()[name = string("op_10090_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10090_end_0 = const()[name = string("op_10090_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10090_end_mask_0 = const()[name = string("op_10090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10090_cast_fp16 = slice_by_index(begin = var_10090_begin_0, end = var_10090_end_0, end_mask = var_10090_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10090_cast_fp16")]; + tensor var_10097_begin_0 = const()[name = string("op_10097_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10097_end_0 = const()[name = string("op_10097_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10097_end_mask_0 = const()[name = string("op_10097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10097_cast_fp16 = slice_by_index(begin = var_10097_begin_0, end = var_10097_end_0, end_mask = var_10097_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10097_cast_fp16")]; + tensor var_10104_begin_0 = const()[name = string("op_10104_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10104_end_0 = const()[name = string("op_10104_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10104_end_mask_0 = const()[name = string("op_10104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10104_cast_fp16 = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10104_cast_fp16")]; + tensor var_10111_begin_0 = const()[name = string("op_10111_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10111_end_0 = const()[name = string("op_10111_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10111_end_mask_0 = const()[name = string("op_10111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10111_cast_fp16 = slice_by_index(begin = var_10111_begin_0, end = var_10111_end_0, end_mask = var_10111_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10111_cast_fp16")]; + tensor var_10118_begin_0 = const()[name = string("op_10118_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10118_end_0 = const()[name = string("op_10118_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10118_end_mask_0 = const()[name = string("op_10118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10118_cast_fp16 = slice_by_index(begin = var_10118_begin_0, end = var_10118_end_0, end_mask = var_10118_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10118_cast_fp16")]; + tensor var_10125_begin_0 = const()[name = string("op_10125_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10125_end_0 = const()[name = string("op_10125_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10125_end_mask_0 = const()[name = string("op_10125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10125_cast_fp16 = slice_by_index(begin = var_10125_begin_0, end = var_10125_end_0, end_mask = var_10125_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10125_cast_fp16")]; + tensor var_10132_begin_0 = const()[name = string("op_10132_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10132_end_0 = const()[name = string("op_10132_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10132_end_mask_0 = const()[name = string("op_10132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10132_cast_fp16 = slice_by_index(begin = var_10132_begin_0, end = var_10132_end_0, end_mask = var_10132_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10132_cast_fp16")]; + tensor var_10139_begin_0 = const()[name = string("op_10139_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10139_end_0 = const()[name = string("op_10139_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10139_end_mask_0 = const()[name = string("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10139_cast_fp16")]; + tensor var_10146_begin_0 = const()[name = string("op_10146_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10146_end_0 = const()[name = string("op_10146_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10146_end_mask_0 = const()[name = string("op_10146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10146_cast_fp16 = slice_by_index(begin = var_10146_begin_0, end = var_10146_end_0, end_mask = var_10146_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10146_cast_fp16")]; + tensor var_10153_begin_0 = const()[name = string("op_10153_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10153_end_0 = const()[name = string("op_10153_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10153_end_mask_0 = const()[name = string("op_10153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10153_cast_fp16 = slice_by_index(begin = var_10153_begin_0, end = var_10153_end_0, end_mask = var_10153_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10153_cast_fp16")]; + tensor var_10160_begin_0 = const()[name = string("op_10160_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10160_end_0 = const()[name = string("op_10160_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10160_end_mask_0 = const()[name = string("op_10160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10160_cast_fp16 = slice_by_index(begin = var_10160_begin_0, end = var_10160_end_0, end_mask = var_10160_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10160_cast_fp16")]; + tensor var_10167_begin_0 = const()[name = string("op_10167_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10167_end_0 = const()[name = string("op_10167_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10167_end_mask_0 = const()[name = string("op_10167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10167_cast_fp16 = slice_by_index(begin = var_10167_begin_0, end = var_10167_end_0, end_mask = var_10167_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10167_cast_fp16")]; + tensor var_10174_begin_0 = const()[name = string("op_10174_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10174_end_0 = const()[name = string("op_10174_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10174_end_mask_0 = const()[name = string("op_10174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10174_cast_fp16 = slice_by_index(begin = var_10174_begin_0, end = var_10174_end_0, end_mask = var_10174_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10174_cast_fp16")]; + tensor var_10181_begin_0 = const()[name = string("op_10181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10181_end_0 = const()[name = string("op_10181_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10181_end_mask_0 = const()[name = string("op_10181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10181_cast_fp16 = slice_by_index(begin = var_10181_begin_0, end = var_10181_end_0, end_mask = var_10181_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10181_cast_fp16")]; + tensor var_10188_begin_0 = const()[name = string("op_10188_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10188_end_0 = const()[name = string("op_10188_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10188_end_mask_0 = const()[name = string("op_10188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10188_cast_fp16 = slice_by_index(begin = var_10188_begin_0, end = var_10188_end_0, end_mask = var_10188_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10188_cast_fp16")]; + tensor var_10195_begin_0 = const()[name = string("op_10195_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10195_end_0 = const()[name = string("op_10195_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10195_end_mask_0 = const()[name = string("op_10195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10195_cast_fp16 = slice_by_index(begin = var_10195_begin_0, end = var_10195_end_0, end_mask = var_10195_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10195_cast_fp16")]; + tensor var_10202_begin_0 = const()[name = string("op_10202_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10202_end_0 = const()[name = string("op_10202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10202_end_mask_0 = const()[name = string("op_10202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10202_cast_fp16 = slice_by_index(begin = var_10202_begin_0, end = var_10202_end_0, end_mask = var_10202_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10202_cast_fp16")]; + tensor k_21_perm_0 = const()[name = string("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10207_begin_0 = const()[name = string("op_10207_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10207_end_0 = const()[name = string("op_10207_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_10207_end_mask_0 = const()[name = string("op_10207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = string("transpose_1")]; + tensor var_10207_cast_fp16 = slice_by_index(begin = var_10207_begin_0, end = var_10207_end_0, end_mask = var_10207_end_mask_0, x = k_21_cast_fp16)[name = string("op_10207_cast_fp16")]; + tensor var_10211_begin_0 = const()[name = string("op_10211_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10211_end_0 = const()[name = string("op_10211_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_10211_end_mask_0 = const()[name = string("op_10211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10211_cast_fp16 = slice_by_index(begin = var_10211_begin_0, end = var_10211_end_0, end_mask = var_10211_end_mask_0, x = k_21_cast_fp16)[name = string("op_10211_cast_fp16")]; + tensor var_10215_begin_0 = const()[name = string("op_10215_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10215_end_0 = const()[name = string("op_10215_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_10215_end_mask_0 = const()[name = string("op_10215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10215_cast_fp16 = slice_by_index(begin = var_10215_begin_0, end = var_10215_end_0, end_mask = var_10215_end_mask_0, x = k_21_cast_fp16)[name = string("op_10215_cast_fp16")]; + tensor var_10219_begin_0 = const()[name = string("op_10219_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10219_end_0 = const()[name = string("op_10219_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_10219_end_mask_0 = const()[name = string("op_10219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10219_cast_fp16 = slice_by_index(begin = var_10219_begin_0, end = var_10219_end_0, end_mask = var_10219_end_mask_0, x = k_21_cast_fp16)[name = string("op_10219_cast_fp16")]; + tensor var_10223_begin_0 = const()[name = string("op_10223_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10223_end_0 = const()[name = string("op_10223_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_10223_end_mask_0 = const()[name = string("op_10223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10223_cast_fp16 = slice_by_index(begin = var_10223_begin_0, end = var_10223_end_0, end_mask = var_10223_end_mask_0, x = k_21_cast_fp16)[name = string("op_10223_cast_fp16")]; + tensor var_10227_begin_0 = const()[name = string("op_10227_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10227_end_0 = const()[name = string("op_10227_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_10227_end_mask_0 = const()[name = string("op_10227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10227_cast_fp16 = slice_by_index(begin = var_10227_begin_0, end = var_10227_end_0, end_mask = var_10227_end_mask_0, x = k_21_cast_fp16)[name = string("op_10227_cast_fp16")]; + tensor var_10231_begin_0 = const()[name = string("op_10231_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10231_end_0 = const()[name = string("op_10231_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_10231_end_mask_0 = const()[name = string("op_10231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10231_cast_fp16 = slice_by_index(begin = var_10231_begin_0, end = var_10231_end_0, end_mask = var_10231_end_mask_0, x = k_21_cast_fp16)[name = string("op_10231_cast_fp16")]; + tensor var_10235_begin_0 = const()[name = string("op_10235_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10235_end_0 = const()[name = string("op_10235_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_10235_end_mask_0 = const()[name = string("op_10235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10235_cast_fp16 = slice_by_index(begin = var_10235_begin_0, end = var_10235_end_0, end_mask = var_10235_end_mask_0, x = k_21_cast_fp16)[name = string("op_10235_cast_fp16")]; + tensor var_10239_begin_0 = const()[name = string("op_10239_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10239_end_0 = const()[name = string("op_10239_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_10239_end_mask_0 = const()[name = string("op_10239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10239_cast_fp16 = slice_by_index(begin = var_10239_begin_0, end = var_10239_end_0, end_mask = var_10239_end_mask_0, x = k_21_cast_fp16)[name = string("op_10239_cast_fp16")]; + tensor var_10243_begin_0 = const()[name = string("op_10243_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10243_end_0 = const()[name = string("op_10243_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_10243_end_mask_0 = const()[name = string("op_10243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10243_cast_fp16 = slice_by_index(begin = var_10243_begin_0, end = var_10243_end_0, end_mask = var_10243_end_mask_0, x = k_21_cast_fp16)[name = string("op_10243_cast_fp16")]; + tensor var_10247_begin_0 = const()[name = string("op_10247_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10247_end_0 = const()[name = string("op_10247_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_10247_end_mask_0 = const()[name = string("op_10247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10247_cast_fp16 = slice_by_index(begin = var_10247_begin_0, end = var_10247_end_0, end_mask = var_10247_end_mask_0, x = k_21_cast_fp16)[name = string("op_10247_cast_fp16")]; + tensor var_10251_begin_0 = const()[name = string("op_10251_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10251_end_0 = const()[name = string("op_10251_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_10251_end_mask_0 = const()[name = string("op_10251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10251_cast_fp16 = slice_by_index(begin = var_10251_begin_0, end = var_10251_end_0, end_mask = var_10251_end_mask_0, x = k_21_cast_fp16)[name = string("op_10251_cast_fp16")]; + tensor var_10253_begin_0 = const()[name = string("op_10253_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10253_end_0 = const()[name = string("op_10253_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10253_end_mask_0 = const()[name = string("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = value_21_cast_fp16)[name = string("op_10253_cast_fp16")]; + tensor var_10257_begin_0 = const()[name = string("op_10257_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10257_end_0 = const()[name = string("op_10257_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10257_end_mask_0 = const()[name = string("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = value_21_cast_fp16)[name = string("op_10257_cast_fp16")]; + tensor var_10261_begin_0 = const()[name = string("op_10261_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10261_end_0 = const()[name = string("op_10261_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10261_end_mask_0 = const()[name = string("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = value_21_cast_fp16)[name = string("op_10261_cast_fp16")]; + tensor var_10265_begin_0 = const()[name = string("op_10265_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10265_end_0 = const()[name = string("op_10265_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10265_end_mask_0 = const()[name = string("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = value_21_cast_fp16)[name = string("op_10265_cast_fp16")]; + tensor var_10269_begin_0 = const()[name = string("op_10269_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10269_end_0 = const()[name = string("op_10269_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10269_end_mask_0 = const()[name = string("op_10269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10269_cast_fp16 = slice_by_index(begin = var_10269_begin_0, end = var_10269_end_0, end_mask = var_10269_end_mask_0, x = value_21_cast_fp16)[name = string("op_10269_cast_fp16")]; + tensor var_10273_begin_0 = const()[name = string("op_10273_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10273_end_0 = const()[name = string("op_10273_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10273_end_mask_0 = const()[name = string("op_10273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10273_cast_fp16 = slice_by_index(begin = var_10273_begin_0, end = var_10273_end_0, end_mask = var_10273_end_mask_0, x = value_21_cast_fp16)[name = string("op_10273_cast_fp16")]; + tensor var_10277_begin_0 = const()[name = string("op_10277_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10277_end_0 = const()[name = string("op_10277_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10277_end_mask_0 = const()[name = string("op_10277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10277_cast_fp16 = slice_by_index(begin = var_10277_begin_0, end = var_10277_end_0, end_mask = var_10277_end_mask_0, x = value_21_cast_fp16)[name = string("op_10277_cast_fp16")]; + tensor var_10281_begin_0 = const()[name = string("op_10281_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10281_end_0 = const()[name = string("op_10281_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10281_end_mask_0 = const()[name = string("op_10281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10281_cast_fp16 = slice_by_index(begin = var_10281_begin_0, end = var_10281_end_0, end_mask = var_10281_end_mask_0, x = value_21_cast_fp16)[name = string("op_10281_cast_fp16")]; + tensor var_10285_begin_0 = const()[name = string("op_10285_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10285_end_0 = const()[name = string("op_10285_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10285_end_mask_0 = const()[name = string("op_10285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10285_cast_fp16 = slice_by_index(begin = var_10285_begin_0, end = var_10285_end_0, end_mask = var_10285_end_mask_0, x = value_21_cast_fp16)[name = string("op_10285_cast_fp16")]; + tensor var_10289_begin_0 = const()[name = string("op_10289_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10289_end_0 = const()[name = string("op_10289_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10289_end_mask_0 = const()[name = string("op_10289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10289_cast_fp16 = slice_by_index(begin = var_10289_begin_0, end = var_10289_end_0, end_mask = var_10289_end_mask_0, x = value_21_cast_fp16)[name = string("op_10289_cast_fp16")]; + tensor var_10293_begin_0 = const()[name = string("op_10293_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10293_end_0 = const()[name = string("op_10293_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10293_end_mask_0 = const()[name = string("op_10293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10293_cast_fp16 = slice_by_index(begin = var_10293_begin_0, end = var_10293_end_0, end_mask = var_10293_end_mask_0, x = value_21_cast_fp16)[name = string("op_10293_cast_fp16")]; + tensor var_10297_begin_0 = const()[name = string("op_10297_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10297_end_0 = const()[name = string("op_10297_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10297_end_mask_0 = const()[name = string("op_10297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10297_cast_fp16 = slice_by_index(begin = var_10297_begin_0, end = var_10297_end_0, end_mask = var_10297_end_mask_0, x = value_21_cast_fp16)[name = string("op_10297_cast_fp16")]; + string _SplitHeadsQ__mh_w_961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_961_equation_0, values = (var_10207_cast_fp16, var_9873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_961_cast_fp16")]; + string _SplitHeadsQ__mh_w_963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_963_equation_0, values = (var_10207_cast_fp16, var_9880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_963_cast_fp16")]; + string _SplitHeadsQ__mh_w_965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_965_equation_0, values = (var_10207_cast_fp16, var_9887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_965_cast_fp16")]; + string _SplitHeadsQ__mh_w_967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_967_equation_0, values = (var_10207_cast_fp16, var_9894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_967_cast_fp16")]; + string _SplitHeadsQ__mh_w_969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_969_equation_0, values = (var_10211_cast_fp16, var_9901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_969_cast_fp16")]; + string _SplitHeadsQ__mh_w_971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_971_equation_0, values = (var_10211_cast_fp16, var_9908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_971_cast_fp16")]; + string _SplitHeadsQ__mh_w_973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_973_equation_0, values = (var_10211_cast_fp16, var_9915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_973_cast_fp16")]; + string _SplitHeadsQ__mh_w_975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_975_equation_0, values = (var_10211_cast_fp16, var_9922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_975_cast_fp16")]; + string _SplitHeadsQ__mh_w_977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_977_equation_0, values = (var_10215_cast_fp16, var_9929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_977_cast_fp16")]; + string _SplitHeadsQ__mh_w_979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_979_equation_0, values = (var_10215_cast_fp16, var_9936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_979_cast_fp16")]; + string _SplitHeadsQ__mh_w_981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_981_equation_0, values = (var_10215_cast_fp16, var_9943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_981_cast_fp16")]; + string _SplitHeadsQ__mh_w_983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_983_equation_0, values = (var_10215_cast_fp16, var_9950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_983_cast_fp16")]; + string _SplitHeadsQ__mh_w_985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_985_equation_0, values = (var_10219_cast_fp16, var_9957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_985_cast_fp16")]; + string _SplitHeadsQ__mh_w_987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_987_equation_0, values = (var_10219_cast_fp16, var_9964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_987_cast_fp16")]; + string _SplitHeadsQ__mh_w_989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_989_equation_0, values = (var_10219_cast_fp16, var_9971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_989_cast_fp16")]; + string _SplitHeadsQ__mh_w_991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_991_equation_0, values = (var_10219_cast_fp16, var_9978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_991_cast_fp16")]; + string _SplitHeadsQ__mh_w_993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_993_equation_0, values = (var_10223_cast_fp16, var_9985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_993_cast_fp16")]; + string _SplitHeadsQ__mh_w_995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_995_equation_0, values = (var_10223_cast_fp16, var_9992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_995_cast_fp16")]; + string _SplitHeadsQ__mh_w_997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_997_equation_0, values = (var_10223_cast_fp16, var_9999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_997_cast_fp16")]; + string _SplitHeadsQ__mh_w_999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_999_equation_0, values = (var_10223_cast_fp16, var_10006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_999_cast_fp16")]; + string _SplitHeadsQ__mh_w_1001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1001_equation_0, values = (var_10227_cast_fp16, var_10013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1001_cast_fp16")]; + string _SplitHeadsQ__mh_w_1003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1003_equation_0, values = (var_10227_cast_fp16, var_10020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1003_cast_fp16")]; + string _SplitHeadsQ__mh_w_1005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1005_equation_0, values = (var_10227_cast_fp16, var_10027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1005_cast_fp16")]; + string _SplitHeadsQ__mh_w_1007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1007_equation_0, values = (var_10227_cast_fp16, var_10034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1007_cast_fp16")]; + string _SplitHeadsQ__mh_w_1009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1009_equation_0, values = (var_10231_cast_fp16, var_10041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1009_cast_fp16")]; + string _SplitHeadsQ__mh_w_1011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1011_equation_0, values = (var_10231_cast_fp16, var_10048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1011_cast_fp16")]; + string _SplitHeadsQ__mh_w_1013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1013_equation_0, values = (var_10231_cast_fp16, var_10055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1013_cast_fp16")]; + string _SplitHeadsQ__mh_w_1015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1015_equation_0, values = (var_10231_cast_fp16, var_10062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1015_cast_fp16")]; + string _SplitHeadsQ__mh_w_1017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1017_equation_0, values = (var_10235_cast_fp16, var_10069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1017_cast_fp16")]; + string _SplitHeadsQ__mh_w_1019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1019_equation_0, values = (var_10235_cast_fp16, var_10076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1019_cast_fp16")]; + string _SplitHeadsQ__mh_w_1021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1021_equation_0, values = (var_10235_cast_fp16, var_10083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1021_cast_fp16")]; + string _SplitHeadsQ__mh_w_1023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1023_equation_0, values = (var_10235_cast_fp16, var_10090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1023_cast_fp16")]; + string _SplitHeadsQ__mh_w_1025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1025_equation_0, values = (var_10239_cast_fp16, var_10097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1025_cast_fp16")]; + string _SplitHeadsQ__mh_w_1027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1027_equation_0, values = (var_10239_cast_fp16, var_10104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1027_cast_fp16")]; + string _SplitHeadsQ__mh_w_1029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1029_equation_0, values = (var_10239_cast_fp16, var_10111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1029_cast_fp16")]; + string _SplitHeadsQ__mh_w_1031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1031_equation_0, values = (var_10239_cast_fp16, var_10118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1031_cast_fp16")]; + string _SplitHeadsQ__mh_w_1033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1033_equation_0, values = (var_10243_cast_fp16, var_10125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1033_cast_fp16")]; + string _SplitHeadsQ__mh_w_1035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1035_equation_0, values = (var_10243_cast_fp16, var_10132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1035_cast_fp16")]; + string _SplitHeadsQ__mh_w_1037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1037_equation_0, values = (var_10243_cast_fp16, var_10139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1037_cast_fp16")]; + string _SplitHeadsQ__mh_w_1039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1039_equation_0, values = (var_10243_cast_fp16, var_10146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1039_cast_fp16")]; + string _SplitHeadsQ__mh_w_1041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1041_equation_0, values = (var_10247_cast_fp16, var_10153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1041_cast_fp16")]; + string _SplitHeadsQ__mh_w_1043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1043_equation_0, values = (var_10247_cast_fp16, var_10160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1043_cast_fp16")]; + string _SplitHeadsQ__mh_w_1045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1045_equation_0, values = (var_10247_cast_fp16, var_10167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1045_cast_fp16")]; + string _SplitHeadsQ__mh_w_1047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1047_equation_0, values = (var_10247_cast_fp16, var_10174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1047_cast_fp16")]; + string _SplitHeadsQ__mh_w_1049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1049_equation_0, values = (var_10251_cast_fp16, var_10181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1049_cast_fp16")]; + string _SplitHeadsQ__mh_w_1051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1051_equation_0, values = (var_10251_cast_fp16, var_10188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1051_cast_fp16")]; + string _SplitHeadsQ__mh_w_1053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1053_equation_0, values = (var_10251_cast_fp16, var_10195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1053_cast_fp16")]; + string _SplitHeadsQ__mh_w_1055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1055_equation_0, values = (var_10251_cast_fp16, var_10202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1055_cast_fp16")]; + fp16 var_10396_to_fp16 = const()[name = string("op_10396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_961_cast_fp16, y = var_10396_to_fp16)[name = string("aw_chunk_961_cast_fp16")]; + fp16 var_10398_to_fp16 = const()[name = string("op_10398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_963_cast_fp16, y = var_10398_to_fp16)[name = string("aw_chunk_963_cast_fp16")]; + fp16 var_10400_to_fp16 = const()[name = string("op_10400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_965_cast_fp16, y = var_10400_to_fp16)[name = string("aw_chunk_965_cast_fp16")]; + fp16 var_10402_to_fp16 = const()[name = string("op_10402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_967_cast_fp16, y = var_10402_to_fp16)[name = string("aw_chunk_967_cast_fp16")]; + fp16 var_10404_to_fp16 = const()[name = string("op_10404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_969_cast_fp16, y = var_10404_to_fp16)[name = string("aw_chunk_969_cast_fp16")]; + fp16 var_10406_to_fp16 = const()[name = string("op_10406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_971_cast_fp16, y = var_10406_to_fp16)[name = string("aw_chunk_971_cast_fp16")]; + fp16 var_10408_to_fp16 = const()[name = string("op_10408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_973_cast_fp16, y = var_10408_to_fp16)[name = string("aw_chunk_973_cast_fp16")]; + fp16 var_10410_to_fp16 = const()[name = string("op_10410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_975_cast_fp16, y = var_10410_to_fp16)[name = string("aw_chunk_975_cast_fp16")]; + fp16 var_10412_to_fp16 = const()[name = string("op_10412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_977_cast_fp16, y = var_10412_to_fp16)[name = string("aw_chunk_977_cast_fp16")]; + fp16 var_10414_to_fp16 = const()[name = string("op_10414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_979_cast_fp16, y = var_10414_to_fp16)[name = string("aw_chunk_979_cast_fp16")]; + fp16 var_10416_to_fp16 = const()[name = string("op_10416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_981_cast_fp16, y = var_10416_to_fp16)[name = string("aw_chunk_981_cast_fp16")]; + fp16 var_10418_to_fp16 = const()[name = string("op_10418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_983_cast_fp16, y = var_10418_to_fp16)[name = string("aw_chunk_983_cast_fp16")]; + fp16 var_10420_to_fp16 = const()[name = string("op_10420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_985_cast_fp16, y = var_10420_to_fp16)[name = string("aw_chunk_985_cast_fp16")]; + fp16 var_10422_to_fp16 = const()[name = string("op_10422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_987_cast_fp16, y = var_10422_to_fp16)[name = string("aw_chunk_987_cast_fp16")]; + fp16 var_10424_to_fp16 = const()[name = string("op_10424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_989_cast_fp16, y = var_10424_to_fp16)[name = string("aw_chunk_989_cast_fp16")]; + fp16 var_10426_to_fp16 = const()[name = string("op_10426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_991_cast_fp16, y = var_10426_to_fp16)[name = string("aw_chunk_991_cast_fp16")]; + fp16 var_10428_to_fp16 = const()[name = string("op_10428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_993_cast_fp16, y = var_10428_to_fp16)[name = string("aw_chunk_993_cast_fp16")]; + fp16 var_10430_to_fp16 = const()[name = string("op_10430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_995_cast_fp16, y = var_10430_to_fp16)[name = string("aw_chunk_995_cast_fp16")]; + fp16 var_10432_to_fp16 = const()[name = string("op_10432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_997_cast_fp16, y = var_10432_to_fp16)[name = string("aw_chunk_997_cast_fp16")]; + fp16 var_10434_to_fp16 = const()[name = string("op_10434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_999_cast_fp16, y = var_10434_to_fp16)[name = string("aw_chunk_999_cast_fp16")]; + fp16 var_10436_to_fp16 = const()[name = string("op_10436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1001_cast_fp16, y = var_10436_to_fp16)[name = string("aw_chunk_1001_cast_fp16")]; + fp16 var_10438_to_fp16 = const()[name = string("op_10438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1003_cast_fp16, y = var_10438_to_fp16)[name = string("aw_chunk_1003_cast_fp16")]; + fp16 var_10440_to_fp16 = const()[name = string("op_10440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1005_cast_fp16, y = var_10440_to_fp16)[name = string("aw_chunk_1005_cast_fp16")]; + fp16 var_10442_to_fp16 = const()[name = string("op_10442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1007_cast_fp16, y = var_10442_to_fp16)[name = string("aw_chunk_1007_cast_fp16")]; + fp16 var_10444_to_fp16 = const()[name = string("op_10444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1009_cast_fp16, y = var_10444_to_fp16)[name = string("aw_chunk_1009_cast_fp16")]; + fp16 var_10446_to_fp16 = const()[name = string("op_10446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1011_cast_fp16, y = var_10446_to_fp16)[name = string("aw_chunk_1011_cast_fp16")]; + fp16 var_10448_to_fp16 = const()[name = string("op_10448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1013_cast_fp16, y = var_10448_to_fp16)[name = string("aw_chunk_1013_cast_fp16")]; + fp16 var_10450_to_fp16 = const()[name = string("op_10450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1015_cast_fp16, y = var_10450_to_fp16)[name = string("aw_chunk_1015_cast_fp16")]; + fp16 var_10452_to_fp16 = const()[name = string("op_10452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1017_cast_fp16, y = var_10452_to_fp16)[name = string("aw_chunk_1017_cast_fp16")]; + fp16 var_10454_to_fp16 = const()[name = string("op_10454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1019_cast_fp16, y = var_10454_to_fp16)[name = string("aw_chunk_1019_cast_fp16")]; + fp16 var_10456_to_fp16 = const()[name = string("op_10456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1021_cast_fp16, y = var_10456_to_fp16)[name = string("aw_chunk_1021_cast_fp16")]; + fp16 var_10458_to_fp16 = const()[name = string("op_10458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1023_cast_fp16, y = var_10458_to_fp16)[name = string("aw_chunk_1023_cast_fp16")]; + fp16 var_10460_to_fp16 = const()[name = string("op_10460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1025_cast_fp16, y = var_10460_to_fp16)[name = string("aw_chunk_1025_cast_fp16")]; + fp16 var_10462_to_fp16 = const()[name = string("op_10462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1027_cast_fp16, y = var_10462_to_fp16)[name = string("aw_chunk_1027_cast_fp16")]; + fp16 var_10464_to_fp16 = const()[name = string("op_10464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1029_cast_fp16, y = var_10464_to_fp16)[name = string("aw_chunk_1029_cast_fp16")]; + fp16 var_10466_to_fp16 = const()[name = string("op_10466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1031_cast_fp16, y = var_10466_to_fp16)[name = string("aw_chunk_1031_cast_fp16")]; + fp16 var_10468_to_fp16 = const()[name = string("op_10468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1033_cast_fp16, y = var_10468_to_fp16)[name = string("aw_chunk_1033_cast_fp16")]; + fp16 var_10470_to_fp16 = const()[name = string("op_10470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1035_cast_fp16, y = var_10470_to_fp16)[name = string("aw_chunk_1035_cast_fp16")]; + fp16 var_10472_to_fp16 = const()[name = string("op_10472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1037_cast_fp16, y = var_10472_to_fp16)[name = string("aw_chunk_1037_cast_fp16")]; + fp16 var_10474_to_fp16 = const()[name = string("op_10474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1039_cast_fp16, y = var_10474_to_fp16)[name = string("aw_chunk_1039_cast_fp16")]; + fp16 var_10476_to_fp16 = const()[name = string("op_10476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1041_cast_fp16, y = var_10476_to_fp16)[name = string("aw_chunk_1041_cast_fp16")]; + fp16 var_10478_to_fp16 = const()[name = string("op_10478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1043_cast_fp16, y = var_10478_to_fp16)[name = string("aw_chunk_1043_cast_fp16")]; + fp16 var_10480_to_fp16 = const()[name = string("op_10480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1045_cast_fp16, y = var_10480_to_fp16)[name = string("aw_chunk_1045_cast_fp16")]; + fp16 var_10482_to_fp16 = const()[name = string("op_10482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1047_cast_fp16, y = var_10482_to_fp16)[name = string("aw_chunk_1047_cast_fp16")]; + fp16 var_10484_to_fp16 = const()[name = string("op_10484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1049_cast_fp16, y = var_10484_to_fp16)[name = string("aw_chunk_1049_cast_fp16")]; + fp16 var_10486_to_fp16 = const()[name = string("op_10486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1051_cast_fp16, y = var_10486_to_fp16)[name = string("aw_chunk_1051_cast_fp16")]; + fp16 var_10488_to_fp16 = const()[name = string("op_10488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1053_cast_fp16, y = var_10488_to_fp16)[name = string("aw_chunk_1053_cast_fp16")]; + fp16 var_10490_to_fp16 = const()[name = string("op_10490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1055_cast_fp16, y = var_10490_to_fp16)[name = string("aw_chunk_1055_cast_fp16")]; + tensor var_10492_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_961_cast_fp16)[name = string("op_10492_cast_fp16")]; + tensor var_10493_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_963_cast_fp16)[name = string("op_10493_cast_fp16")]; + tensor var_10494_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_965_cast_fp16)[name = string("op_10494_cast_fp16")]; + tensor var_10495_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_967_cast_fp16)[name = string("op_10495_cast_fp16")]; + tensor var_10496_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_969_cast_fp16)[name = string("op_10496_cast_fp16")]; + tensor var_10497_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_971_cast_fp16)[name = string("op_10497_cast_fp16")]; + tensor var_10498_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_973_cast_fp16)[name = string("op_10498_cast_fp16")]; + tensor var_10499_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_975_cast_fp16)[name = string("op_10499_cast_fp16")]; + tensor var_10500_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_977_cast_fp16)[name = string("op_10500_cast_fp16")]; + tensor var_10501_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_979_cast_fp16)[name = string("op_10501_cast_fp16")]; + tensor var_10502_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_981_cast_fp16)[name = string("op_10502_cast_fp16")]; + tensor var_10503_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_983_cast_fp16)[name = string("op_10503_cast_fp16")]; + tensor var_10504_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_985_cast_fp16)[name = string("op_10504_cast_fp16")]; + tensor var_10505_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_987_cast_fp16)[name = string("op_10505_cast_fp16")]; + tensor var_10506_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_989_cast_fp16)[name = string("op_10506_cast_fp16")]; + tensor var_10507_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_991_cast_fp16)[name = string("op_10507_cast_fp16")]; + tensor var_10508_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_993_cast_fp16)[name = string("op_10508_cast_fp16")]; + tensor var_10509_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_995_cast_fp16)[name = string("op_10509_cast_fp16")]; + tensor var_10510_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_997_cast_fp16)[name = string("op_10510_cast_fp16")]; + tensor var_10511_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_999_cast_fp16)[name = string("op_10511_cast_fp16")]; + tensor var_10512_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1001_cast_fp16)[name = string("op_10512_cast_fp16")]; + tensor var_10513_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1003_cast_fp16)[name = string("op_10513_cast_fp16")]; + tensor var_10514_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1005_cast_fp16)[name = string("op_10514_cast_fp16")]; + tensor var_10515_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1007_cast_fp16)[name = string("op_10515_cast_fp16")]; + tensor var_10516_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1009_cast_fp16)[name = string("op_10516_cast_fp16")]; + tensor var_10517_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1011_cast_fp16)[name = string("op_10517_cast_fp16")]; + tensor var_10518_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1013_cast_fp16)[name = string("op_10518_cast_fp16")]; + tensor var_10519_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1015_cast_fp16)[name = string("op_10519_cast_fp16")]; + tensor var_10520_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1017_cast_fp16)[name = string("op_10520_cast_fp16")]; + tensor var_10521_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1019_cast_fp16)[name = string("op_10521_cast_fp16")]; + tensor var_10522_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1021_cast_fp16)[name = string("op_10522_cast_fp16")]; + tensor var_10523_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1023_cast_fp16)[name = string("op_10523_cast_fp16")]; + tensor var_10524_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1025_cast_fp16)[name = string("op_10524_cast_fp16")]; + tensor var_10525_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1027_cast_fp16)[name = string("op_10525_cast_fp16")]; + tensor var_10526_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1029_cast_fp16)[name = string("op_10526_cast_fp16")]; + tensor var_10527_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1031_cast_fp16)[name = string("op_10527_cast_fp16")]; + tensor var_10528_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1033_cast_fp16)[name = string("op_10528_cast_fp16")]; + tensor var_10529_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1035_cast_fp16)[name = string("op_10529_cast_fp16")]; + tensor var_10530_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1037_cast_fp16)[name = string("op_10530_cast_fp16")]; + tensor var_10531_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1039_cast_fp16)[name = string("op_10531_cast_fp16")]; + tensor var_10532_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1041_cast_fp16)[name = string("op_10532_cast_fp16")]; + tensor var_10533_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1043_cast_fp16)[name = string("op_10533_cast_fp16")]; + tensor var_10534_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1045_cast_fp16)[name = string("op_10534_cast_fp16")]; + tensor var_10535_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1047_cast_fp16)[name = string("op_10535_cast_fp16")]; + tensor var_10536_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1049_cast_fp16)[name = string("op_10536_cast_fp16")]; + tensor var_10537_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1051_cast_fp16)[name = string("op_10537_cast_fp16")]; + tensor var_10538_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1053_cast_fp16)[name = string("op_10538_cast_fp16")]; + tensor var_10539_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1055_cast_fp16)[name = string("op_10539_cast_fp16")]; + string var_10541_equation_0 = const()[name = string("op_10541_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10541_cast_fp16 = einsum(equation = var_10541_equation_0, values = (var_10253_cast_fp16, var_10492_cast_fp16))[name = string("op_10541_cast_fp16")]; + string var_10543_equation_0 = const()[name = string("op_10543_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10543_cast_fp16 = einsum(equation = var_10543_equation_0, values = (var_10253_cast_fp16, var_10493_cast_fp16))[name = string("op_10543_cast_fp16")]; + string var_10545_equation_0 = const()[name = string("op_10545_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10545_cast_fp16 = einsum(equation = var_10545_equation_0, values = (var_10253_cast_fp16, var_10494_cast_fp16))[name = string("op_10545_cast_fp16")]; + string var_10547_equation_0 = const()[name = string("op_10547_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10547_cast_fp16 = einsum(equation = var_10547_equation_0, values = (var_10253_cast_fp16, var_10495_cast_fp16))[name = string("op_10547_cast_fp16")]; + string var_10549_equation_0 = const()[name = string("op_10549_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10549_cast_fp16 = einsum(equation = var_10549_equation_0, values = (var_10257_cast_fp16, var_10496_cast_fp16))[name = string("op_10549_cast_fp16")]; + string var_10551_equation_0 = const()[name = string("op_10551_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10551_cast_fp16 = einsum(equation = var_10551_equation_0, values = (var_10257_cast_fp16, var_10497_cast_fp16))[name = string("op_10551_cast_fp16")]; + string var_10553_equation_0 = const()[name = string("op_10553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10553_cast_fp16 = einsum(equation = var_10553_equation_0, values = (var_10257_cast_fp16, var_10498_cast_fp16))[name = string("op_10553_cast_fp16")]; + string var_10555_equation_0 = const()[name = string("op_10555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10555_cast_fp16 = einsum(equation = var_10555_equation_0, values = (var_10257_cast_fp16, var_10499_cast_fp16))[name = string("op_10555_cast_fp16")]; + string var_10557_equation_0 = const()[name = string("op_10557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10557_cast_fp16 = einsum(equation = var_10557_equation_0, values = (var_10261_cast_fp16, var_10500_cast_fp16))[name = string("op_10557_cast_fp16")]; + string var_10559_equation_0 = const()[name = string("op_10559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10559_cast_fp16 = einsum(equation = var_10559_equation_0, values = (var_10261_cast_fp16, var_10501_cast_fp16))[name = string("op_10559_cast_fp16")]; + string var_10561_equation_0 = const()[name = string("op_10561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10561_cast_fp16 = einsum(equation = var_10561_equation_0, values = (var_10261_cast_fp16, var_10502_cast_fp16))[name = string("op_10561_cast_fp16")]; + string var_10563_equation_0 = const()[name = string("op_10563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10563_cast_fp16 = einsum(equation = var_10563_equation_0, values = (var_10261_cast_fp16, var_10503_cast_fp16))[name = string("op_10563_cast_fp16")]; + string var_10565_equation_0 = const()[name = string("op_10565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10565_cast_fp16 = einsum(equation = var_10565_equation_0, values = (var_10265_cast_fp16, var_10504_cast_fp16))[name = string("op_10565_cast_fp16")]; + string var_10567_equation_0 = const()[name = string("op_10567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10567_cast_fp16 = einsum(equation = var_10567_equation_0, values = (var_10265_cast_fp16, var_10505_cast_fp16))[name = string("op_10567_cast_fp16")]; + string var_10569_equation_0 = const()[name = string("op_10569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10569_cast_fp16 = einsum(equation = var_10569_equation_0, values = (var_10265_cast_fp16, var_10506_cast_fp16))[name = string("op_10569_cast_fp16")]; + string var_10571_equation_0 = const()[name = string("op_10571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10571_cast_fp16 = einsum(equation = var_10571_equation_0, values = (var_10265_cast_fp16, var_10507_cast_fp16))[name = string("op_10571_cast_fp16")]; + string var_10573_equation_0 = const()[name = string("op_10573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10573_cast_fp16 = einsum(equation = var_10573_equation_0, values = (var_10269_cast_fp16, var_10508_cast_fp16))[name = string("op_10573_cast_fp16")]; + string var_10575_equation_0 = const()[name = string("op_10575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10575_cast_fp16 = einsum(equation = var_10575_equation_0, values = (var_10269_cast_fp16, var_10509_cast_fp16))[name = string("op_10575_cast_fp16")]; + string var_10577_equation_0 = const()[name = string("op_10577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10577_cast_fp16 = einsum(equation = var_10577_equation_0, values = (var_10269_cast_fp16, var_10510_cast_fp16))[name = string("op_10577_cast_fp16")]; + string var_10579_equation_0 = const()[name = string("op_10579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10579_cast_fp16 = einsum(equation = var_10579_equation_0, values = (var_10269_cast_fp16, var_10511_cast_fp16))[name = string("op_10579_cast_fp16")]; + string var_10581_equation_0 = const()[name = string("op_10581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10581_cast_fp16 = einsum(equation = var_10581_equation_0, values = (var_10273_cast_fp16, var_10512_cast_fp16))[name = string("op_10581_cast_fp16")]; + string var_10583_equation_0 = const()[name = string("op_10583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10583_cast_fp16 = einsum(equation = var_10583_equation_0, values = (var_10273_cast_fp16, var_10513_cast_fp16))[name = string("op_10583_cast_fp16")]; + string var_10585_equation_0 = const()[name = string("op_10585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10585_cast_fp16 = einsum(equation = var_10585_equation_0, values = (var_10273_cast_fp16, var_10514_cast_fp16))[name = string("op_10585_cast_fp16")]; + string var_10587_equation_0 = const()[name = string("op_10587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10587_cast_fp16 = einsum(equation = var_10587_equation_0, values = (var_10273_cast_fp16, var_10515_cast_fp16))[name = string("op_10587_cast_fp16")]; + string var_10589_equation_0 = const()[name = string("op_10589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10589_cast_fp16 = einsum(equation = var_10589_equation_0, values = (var_10277_cast_fp16, var_10516_cast_fp16))[name = string("op_10589_cast_fp16")]; + string var_10591_equation_0 = const()[name = string("op_10591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10591_cast_fp16 = einsum(equation = var_10591_equation_0, values = (var_10277_cast_fp16, var_10517_cast_fp16))[name = string("op_10591_cast_fp16")]; + string var_10593_equation_0 = const()[name = string("op_10593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10593_cast_fp16 = einsum(equation = var_10593_equation_0, values = (var_10277_cast_fp16, var_10518_cast_fp16))[name = string("op_10593_cast_fp16")]; + string var_10595_equation_0 = const()[name = string("op_10595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10595_cast_fp16 = einsum(equation = var_10595_equation_0, values = (var_10277_cast_fp16, var_10519_cast_fp16))[name = string("op_10595_cast_fp16")]; + string var_10597_equation_0 = const()[name = string("op_10597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10597_cast_fp16 = einsum(equation = var_10597_equation_0, values = (var_10281_cast_fp16, var_10520_cast_fp16))[name = string("op_10597_cast_fp16")]; + string var_10599_equation_0 = const()[name = string("op_10599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10599_cast_fp16 = einsum(equation = var_10599_equation_0, values = (var_10281_cast_fp16, var_10521_cast_fp16))[name = string("op_10599_cast_fp16")]; + string var_10601_equation_0 = const()[name = string("op_10601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10601_cast_fp16 = einsum(equation = var_10601_equation_0, values = (var_10281_cast_fp16, var_10522_cast_fp16))[name = string("op_10601_cast_fp16")]; + string var_10603_equation_0 = const()[name = string("op_10603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10603_cast_fp16 = einsum(equation = var_10603_equation_0, values = (var_10281_cast_fp16, var_10523_cast_fp16))[name = string("op_10603_cast_fp16")]; + string var_10605_equation_0 = const()[name = string("op_10605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10605_cast_fp16 = einsum(equation = var_10605_equation_0, values = (var_10285_cast_fp16, var_10524_cast_fp16))[name = string("op_10605_cast_fp16")]; + string var_10607_equation_0 = const()[name = string("op_10607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10607_cast_fp16 = einsum(equation = var_10607_equation_0, values = (var_10285_cast_fp16, var_10525_cast_fp16))[name = string("op_10607_cast_fp16")]; + string var_10609_equation_0 = const()[name = string("op_10609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10609_cast_fp16 = einsum(equation = var_10609_equation_0, values = (var_10285_cast_fp16, var_10526_cast_fp16))[name = string("op_10609_cast_fp16")]; + string var_10611_equation_0 = const()[name = string("op_10611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10611_cast_fp16 = einsum(equation = var_10611_equation_0, values = (var_10285_cast_fp16, var_10527_cast_fp16))[name = string("op_10611_cast_fp16")]; + string var_10613_equation_0 = const()[name = string("op_10613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10613_cast_fp16 = einsum(equation = var_10613_equation_0, values = (var_10289_cast_fp16, var_10528_cast_fp16))[name = string("op_10613_cast_fp16")]; + string var_10615_equation_0 = const()[name = string("op_10615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10615_cast_fp16 = einsum(equation = var_10615_equation_0, values = (var_10289_cast_fp16, var_10529_cast_fp16))[name = string("op_10615_cast_fp16")]; + string var_10617_equation_0 = const()[name = string("op_10617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10617_cast_fp16 = einsum(equation = var_10617_equation_0, values = (var_10289_cast_fp16, var_10530_cast_fp16))[name = string("op_10617_cast_fp16")]; + string var_10619_equation_0 = const()[name = string("op_10619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10619_cast_fp16 = einsum(equation = var_10619_equation_0, values = (var_10289_cast_fp16, var_10531_cast_fp16))[name = string("op_10619_cast_fp16")]; + string var_10621_equation_0 = const()[name = string("op_10621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10621_cast_fp16 = einsum(equation = var_10621_equation_0, values = (var_10293_cast_fp16, var_10532_cast_fp16))[name = string("op_10621_cast_fp16")]; + string var_10623_equation_0 = const()[name = string("op_10623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10623_cast_fp16 = einsum(equation = var_10623_equation_0, values = (var_10293_cast_fp16, var_10533_cast_fp16))[name = string("op_10623_cast_fp16")]; + string var_10625_equation_0 = const()[name = string("op_10625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10625_cast_fp16 = einsum(equation = var_10625_equation_0, values = (var_10293_cast_fp16, var_10534_cast_fp16))[name = string("op_10625_cast_fp16")]; + string var_10627_equation_0 = const()[name = string("op_10627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10627_cast_fp16 = einsum(equation = var_10627_equation_0, values = (var_10293_cast_fp16, var_10535_cast_fp16))[name = string("op_10627_cast_fp16")]; + string var_10629_equation_0 = const()[name = string("op_10629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10629_cast_fp16 = einsum(equation = var_10629_equation_0, values = (var_10297_cast_fp16, var_10536_cast_fp16))[name = string("op_10629_cast_fp16")]; + string var_10631_equation_0 = const()[name = string("op_10631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10631_cast_fp16 = einsum(equation = var_10631_equation_0, values = (var_10297_cast_fp16, var_10537_cast_fp16))[name = string("op_10631_cast_fp16")]; + string var_10633_equation_0 = const()[name = string("op_10633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10633_cast_fp16 = einsum(equation = var_10633_equation_0, values = (var_10297_cast_fp16, var_10538_cast_fp16))[name = string("op_10633_cast_fp16")]; + string var_10635_equation_0 = const()[name = string("op_10635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10635_cast_fp16 = einsum(equation = var_10635_equation_0, values = (var_10297_cast_fp16, var_10539_cast_fp16))[name = string("op_10635_cast_fp16")]; + bool var_10637_interleave_0 = const()[name = string("op_10637_interleave_0"), val = bool(false)]; + tensor var_10637_cast_fp16 = concat(axis = var_9748, interleave = var_10637_interleave_0, values = (var_10541_cast_fp16, var_10543_cast_fp16, var_10545_cast_fp16, var_10547_cast_fp16))[name = string("op_10637_cast_fp16")]; + bool var_10639_interleave_0 = const()[name = string("op_10639_interleave_0"), val = bool(false)]; + tensor var_10639_cast_fp16 = concat(axis = var_9748, interleave = var_10639_interleave_0, values = (var_10549_cast_fp16, var_10551_cast_fp16, var_10553_cast_fp16, var_10555_cast_fp16))[name = string("op_10639_cast_fp16")]; + bool var_10641_interleave_0 = const()[name = string("op_10641_interleave_0"), val = bool(false)]; + tensor var_10641_cast_fp16 = concat(axis = var_9748, interleave = var_10641_interleave_0, values = (var_10557_cast_fp16, var_10559_cast_fp16, var_10561_cast_fp16, var_10563_cast_fp16))[name = string("op_10641_cast_fp16")]; + bool var_10643_interleave_0 = const()[name = string("op_10643_interleave_0"), val = bool(false)]; + tensor var_10643_cast_fp16 = concat(axis = var_9748, interleave = var_10643_interleave_0, values = (var_10565_cast_fp16, var_10567_cast_fp16, var_10569_cast_fp16, var_10571_cast_fp16))[name = string("op_10643_cast_fp16")]; + bool var_10645_interleave_0 = const()[name = string("op_10645_interleave_0"), val = bool(false)]; + tensor var_10645_cast_fp16 = concat(axis = var_9748, interleave = var_10645_interleave_0, values = (var_10573_cast_fp16, var_10575_cast_fp16, var_10577_cast_fp16, var_10579_cast_fp16))[name = string("op_10645_cast_fp16")]; + bool var_10647_interleave_0 = const()[name = string("op_10647_interleave_0"), val = bool(false)]; + tensor var_10647_cast_fp16 = concat(axis = var_9748, interleave = var_10647_interleave_0, values = (var_10581_cast_fp16, var_10583_cast_fp16, var_10585_cast_fp16, var_10587_cast_fp16))[name = string("op_10647_cast_fp16")]; + bool var_10649_interleave_0 = const()[name = string("op_10649_interleave_0"), val = bool(false)]; + tensor var_10649_cast_fp16 = concat(axis = var_9748, interleave = var_10649_interleave_0, values = (var_10589_cast_fp16, var_10591_cast_fp16, var_10593_cast_fp16, var_10595_cast_fp16))[name = string("op_10649_cast_fp16")]; + bool var_10651_interleave_0 = const()[name = string("op_10651_interleave_0"), val = bool(false)]; + tensor var_10651_cast_fp16 = concat(axis = var_9748, interleave = var_10651_interleave_0, values = (var_10597_cast_fp16, var_10599_cast_fp16, var_10601_cast_fp16, var_10603_cast_fp16))[name = string("op_10651_cast_fp16")]; + bool var_10653_interleave_0 = const()[name = string("op_10653_interleave_0"), val = bool(false)]; + tensor var_10653_cast_fp16 = concat(axis = var_9748, interleave = var_10653_interleave_0, values = (var_10605_cast_fp16, var_10607_cast_fp16, var_10609_cast_fp16, var_10611_cast_fp16))[name = string("op_10653_cast_fp16")]; + bool var_10655_interleave_0 = const()[name = string("op_10655_interleave_0"), val = bool(false)]; + tensor var_10655_cast_fp16 = concat(axis = var_9748, interleave = var_10655_interleave_0, values = (var_10613_cast_fp16, var_10615_cast_fp16, var_10617_cast_fp16, var_10619_cast_fp16))[name = string("op_10655_cast_fp16")]; + bool var_10657_interleave_0 = const()[name = string("op_10657_interleave_0"), val = bool(false)]; + tensor var_10657_cast_fp16 = concat(axis = var_9748, interleave = var_10657_interleave_0, values = (var_10621_cast_fp16, var_10623_cast_fp16, var_10625_cast_fp16, var_10627_cast_fp16))[name = string("op_10657_cast_fp16")]; + bool var_10659_interleave_0 = const()[name = string("op_10659_interleave_0"), val = bool(false)]; + tensor var_10659_cast_fp16 = concat(axis = var_9748, interleave = var_10659_interleave_0, values = (var_10629_cast_fp16, var_10631_cast_fp16, var_10633_cast_fp16, var_10635_cast_fp16))[name = string("op_10659_cast_fp16")]; + bool input_81_interleave_0 = const()[name = string("input_81_interleave_0"), val = bool(false)]; + tensor input_81_cast_fp16 = concat(axis = var_9765, interleave = input_81_interleave_0, values = (var_10637_cast_fp16, var_10639_cast_fp16, var_10641_cast_fp16, var_10643_cast_fp16, var_10645_cast_fp16, var_10647_cast_fp16, var_10649_cast_fp16, var_10651_cast_fp16, var_10653_cast_fp16, var_10655_cast_fp16, var_10657_cast_fp16, var_10659_cast_fp16))[name = string("input_81_cast_fp16")]; + string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; + tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; + tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; + int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151515456)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152695168)))]; + tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_10678_to_fp16 = const()[name = string("op_10678_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_10678_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152696768)))]; + tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152698368)))]; + fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; + string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; + tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; + tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; + int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152699968)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157418624)))]; + tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; + tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157424832)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162143488)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; + int32 var_10707 = const()[name = string("op_10707"), val = int32(3)]; + int32 var_10724 = const()[name = string("op_10724"), val = int32(1)]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_10741_to_fp16 = const()[name = string("op_10741_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_10741_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162145088)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162146688)))]; + fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162148288)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163328000)))]; + tensor query_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163329600)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164509312)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165689024)))]; + tensor value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_10779_begin_0 = const()[name = string("op_10779_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10779_end_0 = const()[name = string("op_10779_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10779_end_mask_0 = const()[name = string("op_10779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10779_cast_fp16 = slice_by_index(begin = var_10779_begin_0, end = var_10779_end_0, end_mask = var_10779_end_mask_0, x = query_cast_fp16)[name = string("op_10779_cast_fp16")]; + tensor var_10783_begin_0 = const()[name = string("op_10783_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10783_end_0 = const()[name = string("op_10783_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10783_end_mask_0 = const()[name = string("op_10783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10783_cast_fp16 = slice_by_index(begin = var_10783_begin_0, end = var_10783_end_0, end_mask = var_10783_end_mask_0, x = query_cast_fp16)[name = string("op_10783_cast_fp16")]; + tensor var_10787_begin_0 = const()[name = string("op_10787_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10787_end_0 = const()[name = string("op_10787_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10787_end_mask_0 = const()[name = string("op_10787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10787_cast_fp16 = slice_by_index(begin = var_10787_begin_0, end = var_10787_end_0, end_mask = var_10787_end_mask_0, x = query_cast_fp16)[name = string("op_10787_cast_fp16")]; + tensor var_10791_begin_0 = const()[name = string("op_10791_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10791_end_0 = const()[name = string("op_10791_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10791_end_mask_0 = const()[name = string("op_10791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10791_cast_fp16 = slice_by_index(begin = var_10791_begin_0, end = var_10791_end_0, end_mask = var_10791_end_mask_0, x = query_cast_fp16)[name = string("op_10791_cast_fp16")]; + tensor var_10795_begin_0 = const()[name = string("op_10795_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10795_end_0 = const()[name = string("op_10795_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10795_end_mask_0 = const()[name = string("op_10795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10795_cast_fp16 = slice_by_index(begin = var_10795_begin_0, end = var_10795_end_0, end_mask = var_10795_end_mask_0, x = query_cast_fp16)[name = string("op_10795_cast_fp16")]; + tensor var_10799_begin_0 = const()[name = string("op_10799_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10799_end_0 = const()[name = string("op_10799_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10799_end_mask_0 = const()[name = string("op_10799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10799_cast_fp16 = slice_by_index(begin = var_10799_begin_0, end = var_10799_end_0, end_mask = var_10799_end_mask_0, x = query_cast_fp16)[name = string("op_10799_cast_fp16")]; + tensor var_10803_begin_0 = const()[name = string("op_10803_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10803_end_0 = const()[name = string("op_10803_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10803_end_mask_0 = const()[name = string("op_10803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10803_cast_fp16 = slice_by_index(begin = var_10803_begin_0, end = var_10803_end_0, end_mask = var_10803_end_mask_0, x = query_cast_fp16)[name = string("op_10803_cast_fp16")]; + tensor var_10807_begin_0 = const()[name = string("op_10807_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10807_end_0 = const()[name = string("op_10807_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10807_end_mask_0 = const()[name = string("op_10807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10807_cast_fp16 = slice_by_index(begin = var_10807_begin_0, end = var_10807_end_0, end_mask = var_10807_end_mask_0, x = query_cast_fp16)[name = string("op_10807_cast_fp16")]; + tensor var_10811_begin_0 = const()[name = string("op_10811_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10811_end_0 = const()[name = string("op_10811_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10811_end_mask_0 = const()[name = string("op_10811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10811_cast_fp16 = slice_by_index(begin = var_10811_begin_0, end = var_10811_end_0, end_mask = var_10811_end_mask_0, x = query_cast_fp16)[name = string("op_10811_cast_fp16")]; + tensor var_10815_begin_0 = const()[name = string("op_10815_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10815_end_0 = const()[name = string("op_10815_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10815_end_mask_0 = const()[name = string("op_10815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10815_cast_fp16 = slice_by_index(begin = var_10815_begin_0, end = var_10815_end_0, end_mask = var_10815_end_mask_0, x = query_cast_fp16)[name = string("op_10815_cast_fp16")]; + tensor var_10819_begin_0 = const()[name = string("op_10819_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10819_end_0 = const()[name = string("op_10819_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10819_end_mask_0 = const()[name = string("op_10819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10819_cast_fp16 = slice_by_index(begin = var_10819_begin_0, end = var_10819_end_0, end_mask = var_10819_end_mask_0, x = query_cast_fp16)[name = string("op_10819_cast_fp16")]; + tensor var_10823_begin_0 = const()[name = string("op_10823_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10823_end_0 = const()[name = string("op_10823_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10823_end_mask_0 = const()[name = string("op_10823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10823_cast_fp16 = slice_by_index(begin = var_10823_begin_0, end = var_10823_end_0, end_mask = var_10823_end_mask_0, x = query_cast_fp16)[name = string("op_10823_cast_fp16")]; + tensor var_10832_begin_0 = const()[name = string("op_10832_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10832_end_0 = const()[name = string("op_10832_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10832_end_mask_0 = const()[name = string("op_10832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10832_cast_fp16 = slice_by_index(begin = var_10832_begin_0, end = var_10832_end_0, end_mask = var_10832_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10832_cast_fp16")]; + tensor var_10839_begin_0 = const()[name = string("op_10839_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10839_end_0 = const()[name = string("op_10839_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10839_end_mask_0 = const()[name = string("op_10839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10839_cast_fp16 = slice_by_index(begin = var_10839_begin_0, end = var_10839_end_0, end_mask = var_10839_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10839_cast_fp16")]; + tensor var_10846_begin_0 = const()[name = string("op_10846_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10846_end_0 = const()[name = string("op_10846_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10846_end_mask_0 = const()[name = string("op_10846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10846_cast_fp16 = slice_by_index(begin = var_10846_begin_0, end = var_10846_end_0, end_mask = var_10846_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10846_cast_fp16")]; + tensor var_10853_begin_0 = const()[name = string("op_10853_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10853_end_0 = const()[name = string("op_10853_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10853_end_mask_0 = const()[name = string("op_10853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10853_cast_fp16 = slice_by_index(begin = var_10853_begin_0, end = var_10853_end_0, end_mask = var_10853_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10853_cast_fp16")]; + tensor var_10860_begin_0 = const()[name = string("op_10860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10860_end_0 = const()[name = string("op_10860_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10860_end_mask_0 = const()[name = string("op_10860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10860_cast_fp16 = slice_by_index(begin = var_10860_begin_0, end = var_10860_end_0, end_mask = var_10860_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10860_cast_fp16")]; + tensor var_10867_begin_0 = const()[name = string("op_10867_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10867_end_0 = const()[name = string("op_10867_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10867_end_mask_0 = const()[name = string("op_10867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10867_cast_fp16 = slice_by_index(begin = var_10867_begin_0, end = var_10867_end_0, end_mask = var_10867_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10867_cast_fp16")]; + tensor var_10874_begin_0 = const()[name = string("op_10874_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10874_end_0 = const()[name = string("op_10874_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10874_end_mask_0 = const()[name = string("op_10874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10874_cast_fp16 = slice_by_index(begin = var_10874_begin_0, end = var_10874_end_0, end_mask = var_10874_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10874_cast_fp16")]; + tensor var_10881_begin_0 = const()[name = string("op_10881_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10881_end_0 = const()[name = string("op_10881_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10881_end_mask_0 = const()[name = string("op_10881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10881_cast_fp16 = slice_by_index(begin = var_10881_begin_0, end = var_10881_end_0, end_mask = var_10881_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10881_cast_fp16")]; + tensor var_10888_begin_0 = const()[name = string("op_10888_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10888_end_0 = const()[name = string("op_10888_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10888_end_mask_0 = const()[name = string("op_10888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10888_cast_fp16 = slice_by_index(begin = var_10888_begin_0, end = var_10888_end_0, end_mask = var_10888_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10888_cast_fp16")]; + tensor var_10895_begin_0 = const()[name = string("op_10895_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10895_end_0 = const()[name = string("op_10895_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10895_end_mask_0 = const()[name = string("op_10895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10895_cast_fp16 = slice_by_index(begin = var_10895_begin_0, end = var_10895_end_0, end_mask = var_10895_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10895_cast_fp16")]; + tensor var_10902_begin_0 = const()[name = string("op_10902_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10902_end_0 = const()[name = string("op_10902_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10902_end_mask_0 = const()[name = string("op_10902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10902_cast_fp16 = slice_by_index(begin = var_10902_begin_0, end = var_10902_end_0, end_mask = var_10902_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10902_cast_fp16")]; + tensor var_10909_begin_0 = const()[name = string("op_10909_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10909_end_0 = const()[name = string("op_10909_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10909_end_mask_0 = const()[name = string("op_10909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10909_cast_fp16 = slice_by_index(begin = var_10909_begin_0, end = var_10909_end_0, end_mask = var_10909_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10909_cast_fp16")]; + tensor var_10916_begin_0 = const()[name = string("op_10916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10916_end_0 = const()[name = string("op_10916_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10916_end_mask_0 = const()[name = string("op_10916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10916_cast_fp16 = slice_by_index(begin = var_10916_begin_0, end = var_10916_end_0, end_mask = var_10916_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10916_cast_fp16")]; + tensor var_10923_begin_0 = const()[name = string("op_10923_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10923_end_0 = const()[name = string("op_10923_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10923_end_mask_0 = const()[name = string("op_10923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10923_cast_fp16 = slice_by_index(begin = var_10923_begin_0, end = var_10923_end_0, end_mask = var_10923_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10923_cast_fp16")]; + tensor var_10930_begin_0 = const()[name = string("op_10930_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10930_end_0 = const()[name = string("op_10930_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10930_end_mask_0 = const()[name = string("op_10930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10930_cast_fp16 = slice_by_index(begin = var_10930_begin_0, end = var_10930_end_0, end_mask = var_10930_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10930_cast_fp16")]; + tensor var_10937_begin_0 = const()[name = string("op_10937_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10937_end_0 = const()[name = string("op_10937_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10937_end_mask_0 = const()[name = string("op_10937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10937_cast_fp16 = slice_by_index(begin = var_10937_begin_0, end = var_10937_end_0, end_mask = var_10937_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10937_cast_fp16")]; + tensor var_10944_begin_0 = const()[name = string("op_10944_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10944_end_0 = const()[name = string("op_10944_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10944_end_mask_0 = const()[name = string("op_10944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10944_cast_fp16 = slice_by_index(begin = var_10944_begin_0, end = var_10944_end_0, end_mask = var_10944_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10944_cast_fp16")]; + tensor var_10951_begin_0 = const()[name = string("op_10951_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10951_end_0 = const()[name = string("op_10951_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10951_end_mask_0 = const()[name = string("op_10951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10951_cast_fp16 = slice_by_index(begin = var_10951_begin_0, end = var_10951_end_0, end_mask = var_10951_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10951_cast_fp16")]; + tensor var_10958_begin_0 = const()[name = string("op_10958_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10958_end_0 = const()[name = string("op_10958_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10958_end_mask_0 = const()[name = string("op_10958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10958_cast_fp16 = slice_by_index(begin = var_10958_begin_0, end = var_10958_end_0, end_mask = var_10958_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10958_cast_fp16")]; + tensor var_10965_begin_0 = const()[name = string("op_10965_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10965_end_0 = const()[name = string("op_10965_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10965_end_mask_0 = const()[name = string("op_10965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10965_cast_fp16 = slice_by_index(begin = var_10965_begin_0, end = var_10965_end_0, end_mask = var_10965_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10965_cast_fp16")]; + tensor var_10972_begin_0 = const()[name = string("op_10972_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10972_end_0 = const()[name = string("op_10972_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10972_end_mask_0 = const()[name = string("op_10972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10972_cast_fp16 = slice_by_index(begin = var_10972_begin_0, end = var_10972_end_0, end_mask = var_10972_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10972_cast_fp16")]; + tensor var_10979_begin_0 = const()[name = string("op_10979_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10979_end_0 = const()[name = string("op_10979_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10979_end_mask_0 = const()[name = string("op_10979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10979_cast_fp16")]; + tensor var_10986_begin_0 = const()[name = string("op_10986_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10986_end_0 = const()[name = string("op_10986_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10986_end_mask_0 = const()[name = string("op_10986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10986_cast_fp16 = slice_by_index(begin = var_10986_begin_0, end = var_10986_end_0, end_mask = var_10986_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10986_cast_fp16")]; + tensor var_10993_begin_0 = const()[name = string("op_10993_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10993_end_0 = const()[name = string("op_10993_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10993_end_mask_0 = const()[name = string("op_10993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10993_cast_fp16 = slice_by_index(begin = var_10993_begin_0, end = var_10993_end_0, end_mask = var_10993_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10993_cast_fp16")]; + tensor var_11000_begin_0 = const()[name = string("op_11000_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11000_end_0 = const()[name = string("op_11000_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11000_end_mask_0 = const()[name = string("op_11000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11000_cast_fp16 = slice_by_index(begin = var_11000_begin_0, end = var_11000_end_0, end_mask = var_11000_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11000_cast_fp16")]; + tensor var_11007_begin_0 = const()[name = string("op_11007_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11007_end_0 = const()[name = string("op_11007_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11007_end_mask_0 = const()[name = string("op_11007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11007_cast_fp16")]; + tensor var_11014_begin_0 = const()[name = string("op_11014_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11014_end_0 = const()[name = string("op_11014_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11014_end_mask_0 = const()[name = string("op_11014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11014_cast_fp16 = slice_by_index(begin = var_11014_begin_0, end = var_11014_end_0, end_mask = var_11014_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11014_cast_fp16")]; + tensor var_11021_begin_0 = const()[name = string("op_11021_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11021_end_0 = const()[name = string("op_11021_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11021_end_mask_0 = const()[name = string("op_11021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11021_cast_fp16 = slice_by_index(begin = var_11021_begin_0, end = var_11021_end_0, end_mask = var_11021_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11021_cast_fp16")]; + tensor var_11028_begin_0 = const()[name = string("op_11028_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11028_end_0 = const()[name = string("op_11028_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11028_end_mask_0 = const()[name = string("op_11028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11028_cast_fp16 = slice_by_index(begin = var_11028_begin_0, end = var_11028_end_0, end_mask = var_11028_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11028_cast_fp16")]; + tensor var_11035_begin_0 = const()[name = string("op_11035_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11035_end_0 = const()[name = string("op_11035_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11035_end_mask_0 = const()[name = string("op_11035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11035_cast_fp16 = slice_by_index(begin = var_11035_begin_0, end = var_11035_end_0, end_mask = var_11035_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11035_cast_fp16")]; + tensor var_11042_begin_0 = const()[name = string("op_11042_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11042_end_0 = const()[name = string("op_11042_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11042_end_mask_0 = const()[name = string("op_11042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11042_cast_fp16 = slice_by_index(begin = var_11042_begin_0, end = var_11042_end_0, end_mask = var_11042_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11042_cast_fp16")]; + tensor var_11049_begin_0 = const()[name = string("op_11049_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11049_end_0 = const()[name = string("op_11049_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11049_end_mask_0 = const()[name = string("op_11049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11049_cast_fp16 = slice_by_index(begin = var_11049_begin_0, end = var_11049_end_0, end_mask = var_11049_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11049_cast_fp16")]; + tensor var_11056_begin_0 = const()[name = string("op_11056_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11056_end_0 = const()[name = string("op_11056_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11056_end_mask_0 = const()[name = string("op_11056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11056_cast_fp16 = slice_by_index(begin = var_11056_begin_0, end = var_11056_end_0, end_mask = var_11056_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11056_cast_fp16")]; + tensor var_11063_begin_0 = const()[name = string("op_11063_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11063_end_0 = const()[name = string("op_11063_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11063_end_mask_0 = const()[name = string("op_11063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11063_cast_fp16 = slice_by_index(begin = var_11063_begin_0, end = var_11063_end_0, end_mask = var_11063_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11063_cast_fp16")]; + tensor var_11070_begin_0 = const()[name = string("op_11070_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11070_end_0 = const()[name = string("op_11070_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11070_end_mask_0 = const()[name = string("op_11070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11070_cast_fp16 = slice_by_index(begin = var_11070_begin_0, end = var_11070_end_0, end_mask = var_11070_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11070_cast_fp16")]; + tensor var_11077_begin_0 = const()[name = string("op_11077_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11077_end_0 = const()[name = string("op_11077_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11077_end_mask_0 = const()[name = string("op_11077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11077_cast_fp16 = slice_by_index(begin = var_11077_begin_0, end = var_11077_end_0, end_mask = var_11077_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11077_cast_fp16")]; + tensor var_11084_begin_0 = const()[name = string("op_11084_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11084_end_0 = const()[name = string("op_11084_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11084_end_mask_0 = const()[name = string("op_11084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11084_cast_fp16 = slice_by_index(begin = var_11084_begin_0, end = var_11084_end_0, end_mask = var_11084_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11084_cast_fp16")]; + tensor var_11091_begin_0 = const()[name = string("op_11091_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11091_end_0 = const()[name = string("op_11091_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11091_end_mask_0 = const()[name = string("op_11091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11091_cast_fp16 = slice_by_index(begin = var_11091_begin_0, end = var_11091_end_0, end_mask = var_11091_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11091_cast_fp16")]; + tensor var_11098_begin_0 = const()[name = string("op_11098_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11098_end_0 = const()[name = string("op_11098_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11098_end_mask_0 = const()[name = string("op_11098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11098_cast_fp16 = slice_by_index(begin = var_11098_begin_0, end = var_11098_end_0, end_mask = var_11098_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11098_cast_fp16")]; + tensor var_11105_begin_0 = const()[name = string("op_11105_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11105_end_0 = const()[name = string("op_11105_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11105_end_mask_0 = const()[name = string("op_11105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11105_cast_fp16 = slice_by_index(begin = var_11105_begin_0, end = var_11105_end_0, end_mask = var_11105_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11105_cast_fp16")]; + tensor var_11112_begin_0 = const()[name = string("op_11112_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11112_end_0 = const()[name = string("op_11112_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11112_end_mask_0 = const()[name = string("op_11112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11112_cast_fp16 = slice_by_index(begin = var_11112_begin_0, end = var_11112_end_0, end_mask = var_11112_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11112_cast_fp16")]; + tensor var_11119_begin_0 = const()[name = string("op_11119_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11119_end_0 = const()[name = string("op_11119_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11119_end_mask_0 = const()[name = string("op_11119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11119_cast_fp16 = slice_by_index(begin = var_11119_begin_0, end = var_11119_end_0, end_mask = var_11119_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11119_cast_fp16")]; + tensor var_11126_begin_0 = const()[name = string("op_11126_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11126_end_0 = const()[name = string("op_11126_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11126_end_mask_0 = const()[name = string("op_11126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11126_cast_fp16 = slice_by_index(begin = var_11126_begin_0, end = var_11126_end_0, end_mask = var_11126_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11126_cast_fp16")]; + tensor var_11133_begin_0 = const()[name = string("op_11133_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11133_end_0 = const()[name = string("op_11133_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11133_end_mask_0 = const()[name = string("op_11133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11133_cast_fp16 = slice_by_index(begin = var_11133_begin_0, end = var_11133_end_0, end_mask = var_11133_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11133_cast_fp16")]; + tensor var_11140_begin_0 = const()[name = string("op_11140_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11140_end_0 = const()[name = string("op_11140_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11140_end_mask_0 = const()[name = string("op_11140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11140_cast_fp16 = slice_by_index(begin = var_11140_begin_0, end = var_11140_end_0, end_mask = var_11140_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11140_cast_fp16")]; + tensor var_11147_begin_0 = const()[name = string("op_11147_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11147_end_0 = const()[name = string("op_11147_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11147_end_mask_0 = const()[name = string("op_11147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11147_cast_fp16 = slice_by_index(begin = var_11147_begin_0, end = var_11147_end_0, end_mask = var_11147_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11147_cast_fp16")]; + tensor var_11154_begin_0 = const()[name = string("op_11154_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11154_end_0 = const()[name = string("op_11154_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11154_end_mask_0 = const()[name = string("op_11154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11154_cast_fp16 = slice_by_index(begin = var_11154_begin_0, end = var_11154_end_0, end_mask = var_11154_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11154_cast_fp16")]; + tensor var_11161_begin_0 = const()[name = string("op_11161_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11161_end_0 = const()[name = string("op_11161_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11161_end_mask_0 = const()[name = string("op_11161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11161_cast_fp16 = slice_by_index(begin = var_11161_begin_0, end = var_11161_end_0, end_mask = var_11161_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11161_cast_fp16")]; + tensor k_23_perm_0 = const()[name = string("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11166_begin_0 = const()[name = string("op_11166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11166_end_0 = const()[name = string("op_11166_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_11166_end_mask_0 = const()[name = string("op_11166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_11166_cast_fp16 = slice_by_index(begin = var_11166_begin_0, end = var_11166_end_0, end_mask = var_11166_end_mask_0, x = k_23_cast_fp16)[name = string("op_11166_cast_fp16")]; + tensor var_11170_begin_0 = const()[name = string("op_11170_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11170_end_0 = const()[name = string("op_11170_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_11170_end_mask_0 = const()[name = string("op_11170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11170_cast_fp16 = slice_by_index(begin = var_11170_begin_0, end = var_11170_end_0, end_mask = var_11170_end_mask_0, x = k_23_cast_fp16)[name = string("op_11170_cast_fp16")]; + tensor var_11174_begin_0 = const()[name = string("op_11174_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11174_end_0 = const()[name = string("op_11174_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_11174_end_mask_0 = const()[name = string("op_11174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11174_cast_fp16 = slice_by_index(begin = var_11174_begin_0, end = var_11174_end_0, end_mask = var_11174_end_mask_0, x = k_23_cast_fp16)[name = string("op_11174_cast_fp16")]; + tensor var_11178_begin_0 = const()[name = string("op_11178_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11178_end_0 = const()[name = string("op_11178_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_11178_end_mask_0 = const()[name = string("op_11178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11178_cast_fp16 = slice_by_index(begin = var_11178_begin_0, end = var_11178_end_0, end_mask = var_11178_end_mask_0, x = k_23_cast_fp16)[name = string("op_11178_cast_fp16")]; + tensor var_11182_begin_0 = const()[name = string("op_11182_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11182_end_0 = const()[name = string("op_11182_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_11182_end_mask_0 = const()[name = string("op_11182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11182_cast_fp16 = slice_by_index(begin = var_11182_begin_0, end = var_11182_end_0, end_mask = var_11182_end_mask_0, x = k_23_cast_fp16)[name = string("op_11182_cast_fp16")]; + tensor var_11186_begin_0 = const()[name = string("op_11186_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11186_end_0 = const()[name = string("op_11186_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_11186_end_mask_0 = const()[name = string("op_11186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11186_cast_fp16 = slice_by_index(begin = var_11186_begin_0, end = var_11186_end_0, end_mask = var_11186_end_mask_0, x = k_23_cast_fp16)[name = string("op_11186_cast_fp16")]; + tensor var_11190_begin_0 = const()[name = string("op_11190_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11190_end_0 = const()[name = string("op_11190_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_11190_end_mask_0 = const()[name = string("op_11190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11190_cast_fp16 = slice_by_index(begin = var_11190_begin_0, end = var_11190_end_0, end_mask = var_11190_end_mask_0, x = k_23_cast_fp16)[name = string("op_11190_cast_fp16")]; + tensor var_11194_begin_0 = const()[name = string("op_11194_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11194_end_0 = const()[name = string("op_11194_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_11194_end_mask_0 = const()[name = string("op_11194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11194_cast_fp16 = slice_by_index(begin = var_11194_begin_0, end = var_11194_end_0, end_mask = var_11194_end_mask_0, x = k_23_cast_fp16)[name = string("op_11194_cast_fp16")]; + tensor var_11198_begin_0 = const()[name = string("op_11198_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11198_end_0 = const()[name = string("op_11198_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_11198_end_mask_0 = const()[name = string("op_11198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11198_cast_fp16 = slice_by_index(begin = var_11198_begin_0, end = var_11198_end_0, end_mask = var_11198_end_mask_0, x = k_23_cast_fp16)[name = string("op_11198_cast_fp16")]; + tensor var_11202_begin_0 = const()[name = string("op_11202_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11202_end_0 = const()[name = string("op_11202_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_11202_end_mask_0 = const()[name = string("op_11202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11202_cast_fp16 = slice_by_index(begin = var_11202_begin_0, end = var_11202_end_0, end_mask = var_11202_end_mask_0, x = k_23_cast_fp16)[name = string("op_11202_cast_fp16")]; + tensor var_11206_begin_0 = const()[name = string("op_11206_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11206_end_0 = const()[name = string("op_11206_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_11206_end_mask_0 = const()[name = string("op_11206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11206_cast_fp16 = slice_by_index(begin = var_11206_begin_0, end = var_11206_end_0, end_mask = var_11206_end_mask_0, x = k_23_cast_fp16)[name = string("op_11206_cast_fp16")]; + tensor var_11210_begin_0 = const()[name = string("op_11210_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11210_end_0 = const()[name = string("op_11210_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_11210_end_mask_0 = const()[name = string("op_11210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11210_cast_fp16 = slice_by_index(begin = var_11210_begin_0, end = var_11210_end_0, end_mask = var_11210_end_mask_0, x = k_23_cast_fp16)[name = string("op_11210_cast_fp16")]; + tensor var_11212_begin_0 = const()[name = string("op_11212_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11212_end_0 = const()[name = string("op_11212_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11212_end_mask_0 = const()[name = string("op_11212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11212_cast_fp16 = slice_by_index(begin = var_11212_begin_0, end = var_11212_end_0, end_mask = var_11212_end_mask_0, x = value_cast_fp16)[name = string("op_11212_cast_fp16")]; + tensor var_11216_begin_0 = const()[name = string("op_11216_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11216_end_0 = const()[name = string("op_11216_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_11216_end_mask_0 = const()[name = string("op_11216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11216_cast_fp16 = slice_by_index(begin = var_11216_begin_0, end = var_11216_end_0, end_mask = var_11216_end_mask_0, x = value_cast_fp16)[name = string("op_11216_cast_fp16")]; + tensor var_11220_begin_0 = const()[name = string("op_11220_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11220_end_0 = const()[name = string("op_11220_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_11220_end_mask_0 = const()[name = string("op_11220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11220_cast_fp16 = slice_by_index(begin = var_11220_begin_0, end = var_11220_end_0, end_mask = var_11220_end_mask_0, x = value_cast_fp16)[name = string("op_11220_cast_fp16")]; + tensor var_11224_begin_0 = const()[name = string("op_11224_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11224_end_0 = const()[name = string("op_11224_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_11224_end_mask_0 = const()[name = string("op_11224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11224_cast_fp16 = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = value_cast_fp16)[name = string("op_11224_cast_fp16")]; + tensor var_11228_begin_0 = const()[name = string("op_11228_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11228_end_0 = const()[name = string("op_11228_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_11228_end_mask_0 = const()[name = string("op_11228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11228_cast_fp16 = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = value_cast_fp16)[name = string("op_11228_cast_fp16")]; + tensor var_11232_begin_0 = const()[name = string("op_11232_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11232_end_0 = const()[name = string("op_11232_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_11232_end_mask_0 = const()[name = string("op_11232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11232_cast_fp16 = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = value_cast_fp16)[name = string("op_11232_cast_fp16")]; + tensor var_11236_begin_0 = const()[name = string("op_11236_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11236_end_0 = const()[name = string("op_11236_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_11236_end_mask_0 = const()[name = string("op_11236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11236_cast_fp16 = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = value_cast_fp16)[name = string("op_11236_cast_fp16")]; + tensor var_11240_begin_0 = const()[name = string("op_11240_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11240_end_0 = const()[name = string("op_11240_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_11240_end_mask_0 = const()[name = string("op_11240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11240_cast_fp16 = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = value_cast_fp16)[name = string("op_11240_cast_fp16")]; + tensor var_11244_begin_0 = const()[name = string("op_11244_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11244_end_0 = const()[name = string("op_11244_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_11244_end_mask_0 = const()[name = string("op_11244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11244_cast_fp16 = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = value_cast_fp16)[name = string("op_11244_cast_fp16")]; + tensor var_11248_begin_0 = const()[name = string("op_11248_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11248_end_0 = const()[name = string("op_11248_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_11248_end_mask_0 = const()[name = string("op_11248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11248_cast_fp16 = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = value_cast_fp16)[name = string("op_11248_cast_fp16")]; + tensor var_11252_begin_0 = const()[name = string("op_11252_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11252_end_0 = const()[name = string("op_11252_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_11252_end_mask_0 = const()[name = string("op_11252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11252_cast_fp16 = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = value_cast_fp16)[name = string("op_11252_cast_fp16")]; + tensor var_11256_begin_0 = const()[name = string("op_11256_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11256_end_0 = const()[name = string("op_11256_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_11256_end_mask_0 = const()[name = string("op_11256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11256_cast_fp16 = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = value_cast_fp16)[name = string("op_11256_cast_fp16")]; + string _SplitHeadsQ__mh_w_1057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1057_equation_0, values = (var_11166_cast_fp16, var_10832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1057_cast_fp16")]; + string _SplitHeadsQ__mh_w_1059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1059_equation_0, values = (var_11166_cast_fp16, var_10839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1059_cast_fp16")]; + string _SplitHeadsQ__mh_w_1061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1061_equation_0, values = (var_11166_cast_fp16, var_10846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1061_cast_fp16")]; + string _SplitHeadsQ__mh_w_1063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1063_equation_0, values = (var_11166_cast_fp16, var_10853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1063_cast_fp16")]; + string _SplitHeadsQ__mh_w_1065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1065_equation_0, values = (var_11170_cast_fp16, var_10860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1065_cast_fp16")]; + string _SplitHeadsQ__mh_w_1067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1067_equation_0, values = (var_11170_cast_fp16, var_10867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1067_cast_fp16")]; + string _SplitHeadsQ__mh_w_1069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1069_equation_0, values = (var_11170_cast_fp16, var_10874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1069_cast_fp16")]; + string _SplitHeadsQ__mh_w_1071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1071_equation_0, values = (var_11170_cast_fp16, var_10881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1071_cast_fp16")]; + string _SplitHeadsQ__mh_w_1073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1073_equation_0, values = (var_11174_cast_fp16, var_10888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1073_cast_fp16")]; + string _SplitHeadsQ__mh_w_1075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1075_equation_0, values = (var_11174_cast_fp16, var_10895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1075_cast_fp16")]; + string _SplitHeadsQ__mh_w_1077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1077_equation_0, values = (var_11174_cast_fp16, var_10902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1077_cast_fp16")]; + string _SplitHeadsQ__mh_w_1079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1079_equation_0, values = (var_11174_cast_fp16, var_10909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1079_cast_fp16")]; + string _SplitHeadsQ__mh_w_1081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1081_equation_0, values = (var_11178_cast_fp16, var_10916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1081_cast_fp16")]; + string _SplitHeadsQ__mh_w_1083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1083_equation_0, values = (var_11178_cast_fp16, var_10923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1083_cast_fp16")]; + string _SplitHeadsQ__mh_w_1085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1085_equation_0, values = (var_11178_cast_fp16, var_10930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1085_cast_fp16")]; + string _SplitHeadsQ__mh_w_1087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1087_equation_0, values = (var_11178_cast_fp16, var_10937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1087_cast_fp16")]; + string _SplitHeadsQ__mh_w_1089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1089_equation_0, values = (var_11182_cast_fp16, var_10944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1089_cast_fp16")]; + string _SplitHeadsQ__mh_w_1091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1091_equation_0, values = (var_11182_cast_fp16, var_10951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1091_cast_fp16")]; + string _SplitHeadsQ__mh_w_1093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1093_equation_0, values = (var_11182_cast_fp16, var_10958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1093_cast_fp16")]; + string _SplitHeadsQ__mh_w_1095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1095_equation_0, values = (var_11182_cast_fp16, var_10965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1095_cast_fp16")]; + string _SplitHeadsQ__mh_w_1097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1097_equation_0, values = (var_11186_cast_fp16, var_10972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1097_cast_fp16")]; + string _SplitHeadsQ__mh_w_1099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1099_equation_0, values = (var_11186_cast_fp16, var_10979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1099_cast_fp16")]; + string _SplitHeadsQ__mh_w_1101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1101_equation_0, values = (var_11186_cast_fp16, var_10986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1101_cast_fp16")]; + string _SplitHeadsQ__mh_w_1103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1103_equation_0, values = (var_11186_cast_fp16, var_10993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1103_cast_fp16")]; + string _SplitHeadsQ__mh_w_1105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1105_equation_0, values = (var_11190_cast_fp16, var_11000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1105_cast_fp16")]; + string _SplitHeadsQ__mh_w_1107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1107_equation_0, values = (var_11190_cast_fp16, var_11007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1107_cast_fp16")]; + string _SplitHeadsQ__mh_w_1109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1109_equation_0, values = (var_11190_cast_fp16, var_11014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1109_cast_fp16")]; + string _SplitHeadsQ__mh_w_1111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1111_equation_0, values = (var_11190_cast_fp16, var_11021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1111_cast_fp16")]; + string _SplitHeadsQ__mh_w_1113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1113_equation_0, values = (var_11194_cast_fp16, var_11028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1113_cast_fp16")]; + string _SplitHeadsQ__mh_w_1115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1115_equation_0, values = (var_11194_cast_fp16, var_11035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1115_cast_fp16")]; + string _SplitHeadsQ__mh_w_1117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1117_equation_0, values = (var_11194_cast_fp16, var_11042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1117_cast_fp16")]; + string _SplitHeadsQ__mh_w_1119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1119_equation_0, values = (var_11194_cast_fp16, var_11049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1119_cast_fp16")]; + string _SplitHeadsQ__mh_w_1121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1121_equation_0, values = (var_11198_cast_fp16, var_11056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1121_cast_fp16")]; + string _SplitHeadsQ__mh_w_1123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1123_equation_0, values = (var_11198_cast_fp16, var_11063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1123_cast_fp16")]; + string _SplitHeadsQ__mh_w_1125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1125_equation_0, values = (var_11198_cast_fp16, var_11070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1125_cast_fp16")]; + string _SplitHeadsQ__mh_w_1127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1127_equation_0, values = (var_11198_cast_fp16, var_11077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1127_cast_fp16")]; + string _SplitHeadsQ__mh_w_1129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1129_equation_0, values = (var_11202_cast_fp16, var_11084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1129_cast_fp16")]; + string _SplitHeadsQ__mh_w_1131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1131_equation_0, values = (var_11202_cast_fp16, var_11091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1131_cast_fp16")]; + string _SplitHeadsQ__mh_w_1133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1133_equation_0, values = (var_11202_cast_fp16, var_11098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1133_cast_fp16")]; + string _SplitHeadsQ__mh_w_1135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1135_equation_0, values = (var_11202_cast_fp16, var_11105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1135_cast_fp16")]; + string _SplitHeadsQ__mh_w_1137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1137_equation_0, values = (var_11206_cast_fp16, var_11112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1137_cast_fp16")]; + string _SplitHeadsQ__mh_w_1139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1139_equation_0, values = (var_11206_cast_fp16, var_11119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1139_cast_fp16")]; + string _SplitHeadsQ__mh_w_1141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1141_equation_0, values = (var_11206_cast_fp16, var_11126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1141_cast_fp16")]; + string _SplitHeadsQ__mh_w_1143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1143_equation_0, values = (var_11206_cast_fp16, var_11133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1143_cast_fp16")]; + string _SplitHeadsQ__mh_w_1145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1145_equation_0, values = (var_11210_cast_fp16, var_11140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1145_cast_fp16")]; + string _SplitHeadsQ__mh_w_1147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1147_equation_0, values = (var_11210_cast_fp16, var_11147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1147_cast_fp16")]; + string _SplitHeadsQ__mh_w_1149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1149_equation_0, values = (var_11210_cast_fp16, var_11154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1149_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_11210_cast_fp16, var_11161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_11355_to_fp16 = const()[name = string("op_11355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1057_cast_fp16, y = var_11355_to_fp16)[name = string("aw_chunk_1057_cast_fp16")]; + fp16 var_11357_to_fp16 = const()[name = string("op_11357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1059_cast_fp16, y = var_11357_to_fp16)[name = string("aw_chunk_1059_cast_fp16")]; + fp16 var_11359_to_fp16 = const()[name = string("op_11359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1061_cast_fp16, y = var_11359_to_fp16)[name = string("aw_chunk_1061_cast_fp16")]; + fp16 var_11361_to_fp16 = const()[name = string("op_11361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1063_cast_fp16, y = var_11361_to_fp16)[name = string("aw_chunk_1063_cast_fp16")]; + fp16 var_11363_to_fp16 = const()[name = string("op_11363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1065_cast_fp16, y = var_11363_to_fp16)[name = string("aw_chunk_1065_cast_fp16")]; + fp16 var_11365_to_fp16 = const()[name = string("op_11365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1067_cast_fp16, y = var_11365_to_fp16)[name = string("aw_chunk_1067_cast_fp16")]; + fp16 var_11367_to_fp16 = const()[name = string("op_11367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1069_cast_fp16, y = var_11367_to_fp16)[name = string("aw_chunk_1069_cast_fp16")]; + fp16 var_11369_to_fp16 = const()[name = string("op_11369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1071_cast_fp16, y = var_11369_to_fp16)[name = string("aw_chunk_1071_cast_fp16")]; + fp16 var_11371_to_fp16 = const()[name = string("op_11371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1073_cast_fp16, y = var_11371_to_fp16)[name = string("aw_chunk_1073_cast_fp16")]; + fp16 var_11373_to_fp16 = const()[name = string("op_11373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1075_cast_fp16, y = var_11373_to_fp16)[name = string("aw_chunk_1075_cast_fp16")]; + fp16 var_11375_to_fp16 = const()[name = string("op_11375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1077_cast_fp16, y = var_11375_to_fp16)[name = string("aw_chunk_1077_cast_fp16")]; + fp16 var_11377_to_fp16 = const()[name = string("op_11377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1079_cast_fp16, y = var_11377_to_fp16)[name = string("aw_chunk_1079_cast_fp16")]; + fp16 var_11379_to_fp16 = const()[name = string("op_11379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1081_cast_fp16, y = var_11379_to_fp16)[name = string("aw_chunk_1081_cast_fp16")]; + fp16 var_11381_to_fp16 = const()[name = string("op_11381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1083_cast_fp16, y = var_11381_to_fp16)[name = string("aw_chunk_1083_cast_fp16")]; + fp16 var_11383_to_fp16 = const()[name = string("op_11383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1085_cast_fp16, y = var_11383_to_fp16)[name = string("aw_chunk_1085_cast_fp16")]; + fp16 var_11385_to_fp16 = const()[name = string("op_11385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1087_cast_fp16, y = var_11385_to_fp16)[name = string("aw_chunk_1087_cast_fp16")]; + fp16 var_11387_to_fp16 = const()[name = string("op_11387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1089_cast_fp16, y = var_11387_to_fp16)[name = string("aw_chunk_1089_cast_fp16")]; + fp16 var_11389_to_fp16 = const()[name = string("op_11389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1091_cast_fp16, y = var_11389_to_fp16)[name = string("aw_chunk_1091_cast_fp16")]; + fp16 var_11391_to_fp16 = const()[name = string("op_11391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1093_cast_fp16, y = var_11391_to_fp16)[name = string("aw_chunk_1093_cast_fp16")]; + fp16 var_11393_to_fp16 = const()[name = string("op_11393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1095_cast_fp16, y = var_11393_to_fp16)[name = string("aw_chunk_1095_cast_fp16")]; + fp16 var_11395_to_fp16 = const()[name = string("op_11395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1097_cast_fp16, y = var_11395_to_fp16)[name = string("aw_chunk_1097_cast_fp16")]; + fp16 var_11397_to_fp16 = const()[name = string("op_11397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1099_cast_fp16, y = var_11397_to_fp16)[name = string("aw_chunk_1099_cast_fp16")]; + fp16 var_11399_to_fp16 = const()[name = string("op_11399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1101_cast_fp16, y = var_11399_to_fp16)[name = string("aw_chunk_1101_cast_fp16")]; + fp16 var_11401_to_fp16 = const()[name = string("op_11401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1103_cast_fp16, y = var_11401_to_fp16)[name = string("aw_chunk_1103_cast_fp16")]; + fp16 var_11403_to_fp16 = const()[name = string("op_11403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1105_cast_fp16, y = var_11403_to_fp16)[name = string("aw_chunk_1105_cast_fp16")]; + fp16 var_11405_to_fp16 = const()[name = string("op_11405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1107_cast_fp16, y = var_11405_to_fp16)[name = string("aw_chunk_1107_cast_fp16")]; + fp16 var_11407_to_fp16 = const()[name = string("op_11407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1109_cast_fp16, y = var_11407_to_fp16)[name = string("aw_chunk_1109_cast_fp16")]; + fp16 var_11409_to_fp16 = const()[name = string("op_11409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1111_cast_fp16, y = var_11409_to_fp16)[name = string("aw_chunk_1111_cast_fp16")]; + fp16 var_11411_to_fp16 = const()[name = string("op_11411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1113_cast_fp16, y = var_11411_to_fp16)[name = string("aw_chunk_1113_cast_fp16")]; + fp16 var_11413_to_fp16 = const()[name = string("op_11413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1115_cast_fp16, y = var_11413_to_fp16)[name = string("aw_chunk_1115_cast_fp16")]; + fp16 var_11415_to_fp16 = const()[name = string("op_11415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1117_cast_fp16, y = var_11415_to_fp16)[name = string("aw_chunk_1117_cast_fp16")]; + fp16 var_11417_to_fp16 = const()[name = string("op_11417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1119_cast_fp16, y = var_11417_to_fp16)[name = string("aw_chunk_1119_cast_fp16")]; + fp16 var_11419_to_fp16 = const()[name = string("op_11419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1121_cast_fp16, y = var_11419_to_fp16)[name = string("aw_chunk_1121_cast_fp16")]; + fp16 var_11421_to_fp16 = const()[name = string("op_11421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1123_cast_fp16, y = var_11421_to_fp16)[name = string("aw_chunk_1123_cast_fp16")]; + fp16 var_11423_to_fp16 = const()[name = string("op_11423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1125_cast_fp16, y = var_11423_to_fp16)[name = string("aw_chunk_1125_cast_fp16")]; + fp16 var_11425_to_fp16 = const()[name = string("op_11425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1127_cast_fp16, y = var_11425_to_fp16)[name = string("aw_chunk_1127_cast_fp16")]; + fp16 var_11427_to_fp16 = const()[name = string("op_11427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1129_cast_fp16, y = var_11427_to_fp16)[name = string("aw_chunk_1129_cast_fp16")]; + fp16 var_11429_to_fp16 = const()[name = string("op_11429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1131_cast_fp16, y = var_11429_to_fp16)[name = string("aw_chunk_1131_cast_fp16")]; + fp16 var_11431_to_fp16 = const()[name = string("op_11431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1133_cast_fp16, y = var_11431_to_fp16)[name = string("aw_chunk_1133_cast_fp16")]; + fp16 var_11433_to_fp16 = const()[name = string("op_11433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1135_cast_fp16, y = var_11433_to_fp16)[name = string("aw_chunk_1135_cast_fp16")]; + fp16 var_11435_to_fp16 = const()[name = string("op_11435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1137_cast_fp16, y = var_11435_to_fp16)[name = string("aw_chunk_1137_cast_fp16")]; + fp16 var_11437_to_fp16 = const()[name = string("op_11437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1139_cast_fp16, y = var_11437_to_fp16)[name = string("aw_chunk_1139_cast_fp16")]; + fp16 var_11439_to_fp16 = const()[name = string("op_11439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1141_cast_fp16, y = var_11439_to_fp16)[name = string("aw_chunk_1141_cast_fp16")]; + fp16 var_11441_to_fp16 = const()[name = string("op_11441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1143_cast_fp16, y = var_11441_to_fp16)[name = string("aw_chunk_1143_cast_fp16")]; + fp16 var_11443_to_fp16 = const()[name = string("op_11443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1145_cast_fp16, y = var_11443_to_fp16)[name = string("aw_chunk_1145_cast_fp16")]; + fp16 var_11445_to_fp16 = const()[name = string("op_11445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1147_cast_fp16, y = var_11445_to_fp16)[name = string("aw_chunk_1147_cast_fp16")]; + fp16 var_11447_to_fp16 = const()[name = string("op_11447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1149_cast_fp16, y = var_11447_to_fp16)[name = string("aw_chunk_1149_cast_fp16")]; + fp16 var_11449_to_fp16 = const()[name = string("op_11449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_11449_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_11451_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1057_cast_fp16)[name = string("op_11451_cast_fp16")]; + tensor var_11452_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1059_cast_fp16)[name = string("op_11452_cast_fp16")]; + tensor var_11453_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1061_cast_fp16)[name = string("op_11453_cast_fp16")]; + tensor var_11454_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1063_cast_fp16)[name = string("op_11454_cast_fp16")]; + tensor var_11455_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1065_cast_fp16)[name = string("op_11455_cast_fp16")]; + tensor var_11456_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1067_cast_fp16)[name = string("op_11456_cast_fp16")]; + tensor var_11457_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1069_cast_fp16)[name = string("op_11457_cast_fp16")]; + tensor var_11458_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1071_cast_fp16)[name = string("op_11458_cast_fp16")]; + tensor var_11459_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1073_cast_fp16)[name = string("op_11459_cast_fp16")]; + tensor var_11460_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1075_cast_fp16)[name = string("op_11460_cast_fp16")]; + tensor var_11461_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1077_cast_fp16)[name = string("op_11461_cast_fp16")]; + tensor var_11462_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1079_cast_fp16)[name = string("op_11462_cast_fp16")]; + tensor var_11463_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1081_cast_fp16)[name = string("op_11463_cast_fp16")]; + tensor var_11464_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1083_cast_fp16)[name = string("op_11464_cast_fp16")]; + tensor var_11465_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1085_cast_fp16)[name = string("op_11465_cast_fp16")]; + tensor var_11466_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1087_cast_fp16)[name = string("op_11466_cast_fp16")]; + tensor var_11467_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1089_cast_fp16)[name = string("op_11467_cast_fp16")]; + tensor var_11468_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1091_cast_fp16)[name = string("op_11468_cast_fp16")]; + tensor var_11469_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1093_cast_fp16)[name = string("op_11469_cast_fp16")]; + tensor var_11470_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1095_cast_fp16)[name = string("op_11470_cast_fp16")]; + tensor var_11471_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1097_cast_fp16)[name = string("op_11471_cast_fp16")]; + tensor var_11472_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1099_cast_fp16)[name = string("op_11472_cast_fp16")]; + tensor var_11473_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1101_cast_fp16)[name = string("op_11473_cast_fp16")]; + tensor var_11474_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1103_cast_fp16)[name = string("op_11474_cast_fp16")]; + tensor var_11475_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1105_cast_fp16)[name = string("op_11475_cast_fp16")]; + tensor var_11476_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1107_cast_fp16)[name = string("op_11476_cast_fp16")]; + tensor var_11477_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1109_cast_fp16)[name = string("op_11477_cast_fp16")]; + tensor var_11478_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1111_cast_fp16)[name = string("op_11478_cast_fp16")]; + tensor var_11479_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1113_cast_fp16)[name = string("op_11479_cast_fp16")]; + tensor var_11480_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1115_cast_fp16)[name = string("op_11480_cast_fp16")]; + tensor var_11481_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1117_cast_fp16)[name = string("op_11481_cast_fp16")]; + tensor var_11482_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1119_cast_fp16)[name = string("op_11482_cast_fp16")]; + tensor var_11483_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1121_cast_fp16)[name = string("op_11483_cast_fp16")]; + tensor var_11484_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1123_cast_fp16)[name = string("op_11484_cast_fp16")]; + tensor var_11485_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1125_cast_fp16)[name = string("op_11485_cast_fp16")]; + tensor var_11486_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1127_cast_fp16)[name = string("op_11486_cast_fp16")]; + tensor var_11487_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1129_cast_fp16)[name = string("op_11487_cast_fp16")]; + tensor var_11488_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1131_cast_fp16)[name = string("op_11488_cast_fp16")]; + tensor var_11489_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1133_cast_fp16)[name = string("op_11489_cast_fp16")]; + tensor var_11490_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1135_cast_fp16)[name = string("op_11490_cast_fp16")]; + tensor var_11491_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1137_cast_fp16)[name = string("op_11491_cast_fp16")]; + tensor var_11492_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1139_cast_fp16)[name = string("op_11492_cast_fp16")]; + tensor var_11493_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1141_cast_fp16)[name = string("op_11493_cast_fp16")]; + tensor var_11494_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1143_cast_fp16)[name = string("op_11494_cast_fp16")]; + tensor var_11495_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1145_cast_fp16)[name = string("op_11495_cast_fp16")]; + tensor var_11496_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1147_cast_fp16)[name = string("op_11496_cast_fp16")]; + tensor var_11497_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1149_cast_fp16)[name = string("op_11497_cast_fp16")]; + tensor var_11498_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_cast_fp16)[name = string("op_11498_cast_fp16")]; + string var_11500_equation_0 = const()[name = string("op_11500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11500_cast_fp16 = einsum(equation = var_11500_equation_0, values = (var_11212_cast_fp16, var_11451_cast_fp16))[name = string("op_11500_cast_fp16")]; + string var_11502_equation_0 = const()[name = string("op_11502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11502_cast_fp16 = einsum(equation = var_11502_equation_0, values = (var_11212_cast_fp16, var_11452_cast_fp16))[name = string("op_11502_cast_fp16")]; + string var_11504_equation_0 = const()[name = string("op_11504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11504_cast_fp16 = einsum(equation = var_11504_equation_0, values = (var_11212_cast_fp16, var_11453_cast_fp16))[name = string("op_11504_cast_fp16")]; + string var_11506_equation_0 = const()[name = string("op_11506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11506_cast_fp16 = einsum(equation = var_11506_equation_0, values = (var_11212_cast_fp16, var_11454_cast_fp16))[name = string("op_11506_cast_fp16")]; + string var_11508_equation_0 = const()[name = string("op_11508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11508_cast_fp16 = einsum(equation = var_11508_equation_0, values = (var_11216_cast_fp16, var_11455_cast_fp16))[name = string("op_11508_cast_fp16")]; + string var_11510_equation_0 = const()[name = string("op_11510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11510_cast_fp16 = einsum(equation = var_11510_equation_0, values = (var_11216_cast_fp16, var_11456_cast_fp16))[name = string("op_11510_cast_fp16")]; + string var_11512_equation_0 = const()[name = string("op_11512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11512_cast_fp16 = einsum(equation = var_11512_equation_0, values = (var_11216_cast_fp16, var_11457_cast_fp16))[name = string("op_11512_cast_fp16")]; + string var_11514_equation_0 = const()[name = string("op_11514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11514_cast_fp16 = einsum(equation = var_11514_equation_0, values = (var_11216_cast_fp16, var_11458_cast_fp16))[name = string("op_11514_cast_fp16")]; + string var_11516_equation_0 = const()[name = string("op_11516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11516_cast_fp16 = einsum(equation = var_11516_equation_0, values = (var_11220_cast_fp16, var_11459_cast_fp16))[name = string("op_11516_cast_fp16")]; + string var_11518_equation_0 = const()[name = string("op_11518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11518_cast_fp16 = einsum(equation = var_11518_equation_0, values = (var_11220_cast_fp16, var_11460_cast_fp16))[name = string("op_11518_cast_fp16")]; + string var_11520_equation_0 = const()[name = string("op_11520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11520_cast_fp16 = einsum(equation = var_11520_equation_0, values = (var_11220_cast_fp16, var_11461_cast_fp16))[name = string("op_11520_cast_fp16")]; + string var_11522_equation_0 = const()[name = string("op_11522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11522_cast_fp16 = einsum(equation = var_11522_equation_0, values = (var_11220_cast_fp16, var_11462_cast_fp16))[name = string("op_11522_cast_fp16")]; + string var_11524_equation_0 = const()[name = string("op_11524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11524_cast_fp16 = einsum(equation = var_11524_equation_0, values = (var_11224_cast_fp16, var_11463_cast_fp16))[name = string("op_11524_cast_fp16")]; + string var_11526_equation_0 = const()[name = string("op_11526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11526_cast_fp16 = einsum(equation = var_11526_equation_0, values = (var_11224_cast_fp16, var_11464_cast_fp16))[name = string("op_11526_cast_fp16")]; + string var_11528_equation_0 = const()[name = string("op_11528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11528_cast_fp16 = einsum(equation = var_11528_equation_0, values = (var_11224_cast_fp16, var_11465_cast_fp16))[name = string("op_11528_cast_fp16")]; + string var_11530_equation_0 = const()[name = string("op_11530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11530_cast_fp16 = einsum(equation = var_11530_equation_0, values = (var_11224_cast_fp16, var_11466_cast_fp16))[name = string("op_11530_cast_fp16")]; + string var_11532_equation_0 = const()[name = string("op_11532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11532_cast_fp16 = einsum(equation = var_11532_equation_0, values = (var_11228_cast_fp16, var_11467_cast_fp16))[name = string("op_11532_cast_fp16")]; + string var_11534_equation_0 = const()[name = string("op_11534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11534_cast_fp16 = einsum(equation = var_11534_equation_0, values = (var_11228_cast_fp16, var_11468_cast_fp16))[name = string("op_11534_cast_fp16")]; + string var_11536_equation_0 = const()[name = string("op_11536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11536_cast_fp16 = einsum(equation = var_11536_equation_0, values = (var_11228_cast_fp16, var_11469_cast_fp16))[name = string("op_11536_cast_fp16")]; + string var_11538_equation_0 = const()[name = string("op_11538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11538_cast_fp16 = einsum(equation = var_11538_equation_0, values = (var_11228_cast_fp16, var_11470_cast_fp16))[name = string("op_11538_cast_fp16")]; + string var_11540_equation_0 = const()[name = string("op_11540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11540_cast_fp16 = einsum(equation = var_11540_equation_0, values = (var_11232_cast_fp16, var_11471_cast_fp16))[name = string("op_11540_cast_fp16")]; + string var_11542_equation_0 = const()[name = string("op_11542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11542_cast_fp16 = einsum(equation = var_11542_equation_0, values = (var_11232_cast_fp16, var_11472_cast_fp16))[name = string("op_11542_cast_fp16")]; + string var_11544_equation_0 = const()[name = string("op_11544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11544_cast_fp16 = einsum(equation = var_11544_equation_0, values = (var_11232_cast_fp16, var_11473_cast_fp16))[name = string("op_11544_cast_fp16")]; + string var_11546_equation_0 = const()[name = string("op_11546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11546_cast_fp16 = einsum(equation = var_11546_equation_0, values = (var_11232_cast_fp16, var_11474_cast_fp16))[name = string("op_11546_cast_fp16")]; + string var_11548_equation_0 = const()[name = string("op_11548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11548_cast_fp16 = einsum(equation = var_11548_equation_0, values = (var_11236_cast_fp16, var_11475_cast_fp16))[name = string("op_11548_cast_fp16")]; + string var_11550_equation_0 = const()[name = string("op_11550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11550_cast_fp16 = einsum(equation = var_11550_equation_0, values = (var_11236_cast_fp16, var_11476_cast_fp16))[name = string("op_11550_cast_fp16")]; + string var_11552_equation_0 = const()[name = string("op_11552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11552_cast_fp16 = einsum(equation = var_11552_equation_0, values = (var_11236_cast_fp16, var_11477_cast_fp16))[name = string("op_11552_cast_fp16")]; + string var_11554_equation_0 = const()[name = string("op_11554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11554_cast_fp16 = einsum(equation = var_11554_equation_0, values = (var_11236_cast_fp16, var_11478_cast_fp16))[name = string("op_11554_cast_fp16")]; + string var_11556_equation_0 = const()[name = string("op_11556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11556_cast_fp16 = einsum(equation = var_11556_equation_0, values = (var_11240_cast_fp16, var_11479_cast_fp16))[name = string("op_11556_cast_fp16")]; + string var_11558_equation_0 = const()[name = string("op_11558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11558_cast_fp16 = einsum(equation = var_11558_equation_0, values = (var_11240_cast_fp16, var_11480_cast_fp16))[name = string("op_11558_cast_fp16")]; + string var_11560_equation_0 = const()[name = string("op_11560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11560_cast_fp16 = einsum(equation = var_11560_equation_0, values = (var_11240_cast_fp16, var_11481_cast_fp16))[name = string("op_11560_cast_fp16")]; + string var_11562_equation_0 = const()[name = string("op_11562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11562_cast_fp16 = einsum(equation = var_11562_equation_0, values = (var_11240_cast_fp16, var_11482_cast_fp16))[name = string("op_11562_cast_fp16")]; + string var_11564_equation_0 = const()[name = string("op_11564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11564_cast_fp16 = einsum(equation = var_11564_equation_0, values = (var_11244_cast_fp16, var_11483_cast_fp16))[name = string("op_11564_cast_fp16")]; + string var_11566_equation_0 = const()[name = string("op_11566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11566_cast_fp16 = einsum(equation = var_11566_equation_0, values = (var_11244_cast_fp16, var_11484_cast_fp16))[name = string("op_11566_cast_fp16")]; + string var_11568_equation_0 = const()[name = string("op_11568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11568_cast_fp16 = einsum(equation = var_11568_equation_0, values = (var_11244_cast_fp16, var_11485_cast_fp16))[name = string("op_11568_cast_fp16")]; + string var_11570_equation_0 = const()[name = string("op_11570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11570_cast_fp16 = einsum(equation = var_11570_equation_0, values = (var_11244_cast_fp16, var_11486_cast_fp16))[name = string("op_11570_cast_fp16")]; + string var_11572_equation_0 = const()[name = string("op_11572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11572_cast_fp16 = einsum(equation = var_11572_equation_0, values = (var_11248_cast_fp16, var_11487_cast_fp16))[name = string("op_11572_cast_fp16")]; + string var_11574_equation_0 = const()[name = string("op_11574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11574_cast_fp16 = einsum(equation = var_11574_equation_0, values = (var_11248_cast_fp16, var_11488_cast_fp16))[name = string("op_11574_cast_fp16")]; + string var_11576_equation_0 = const()[name = string("op_11576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11576_cast_fp16 = einsum(equation = var_11576_equation_0, values = (var_11248_cast_fp16, var_11489_cast_fp16))[name = string("op_11576_cast_fp16")]; + string var_11578_equation_0 = const()[name = string("op_11578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11578_cast_fp16 = einsum(equation = var_11578_equation_0, values = (var_11248_cast_fp16, var_11490_cast_fp16))[name = string("op_11578_cast_fp16")]; + string var_11580_equation_0 = const()[name = string("op_11580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11580_cast_fp16 = einsum(equation = var_11580_equation_0, values = (var_11252_cast_fp16, var_11491_cast_fp16))[name = string("op_11580_cast_fp16")]; + string var_11582_equation_0 = const()[name = string("op_11582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11582_cast_fp16 = einsum(equation = var_11582_equation_0, values = (var_11252_cast_fp16, var_11492_cast_fp16))[name = string("op_11582_cast_fp16")]; + string var_11584_equation_0 = const()[name = string("op_11584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11584_cast_fp16 = einsum(equation = var_11584_equation_0, values = (var_11252_cast_fp16, var_11493_cast_fp16))[name = string("op_11584_cast_fp16")]; + string var_11586_equation_0 = const()[name = string("op_11586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11586_cast_fp16 = einsum(equation = var_11586_equation_0, values = (var_11252_cast_fp16, var_11494_cast_fp16))[name = string("op_11586_cast_fp16")]; + string var_11588_equation_0 = const()[name = string("op_11588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11588_cast_fp16 = einsum(equation = var_11588_equation_0, values = (var_11256_cast_fp16, var_11495_cast_fp16))[name = string("op_11588_cast_fp16")]; + string var_11590_equation_0 = const()[name = string("op_11590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11590_cast_fp16 = einsum(equation = var_11590_equation_0, values = (var_11256_cast_fp16, var_11496_cast_fp16))[name = string("op_11590_cast_fp16")]; + string var_11592_equation_0 = const()[name = string("op_11592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11592_cast_fp16 = einsum(equation = var_11592_equation_0, values = (var_11256_cast_fp16, var_11497_cast_fp16))[name = string("op_11592_cast_fp16")]; + string var_11594_equation_0 = const()[name = string("op_11594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11594_cast_fp16 = einsum(equation = var_11594_equation_0, values = (var_11256_cast_fp16, var_11498_cast_fp16))[name = string("op_11594_cast_fp16")]; + bool var_11596_interleave_0 = const()[name = string("op_11596_interleave_0"), val = bool(false)]; + tensor var_11596_cast_fp16 = concat(axis = var_10707, interleave = var_11596_interleave_0, values = (var_11500_cast_fp16, var_11502_cast_fp16, var_11504_cast_fp16, var_11506_cast_fp16))[name = string("op_11596_cast_fp16")]; + bool var_11598_interleave_0 = const()[name = string("op_11598_interleave_0"), val = bool(false)]; + tensor var_11598_cast_fp16 = concat(axis = var_10707, interleave = var_11598_interleave_0, values = (var_11508_cast_fp16, var_11510_cast_fp16, var_11512_cast_fp16, var_11514_cast_fp16))[name = string("op_11598_cast_fp16")]; + bool var_11600_interleave_0 = const()[name = string("op_11600_interleave_0"), val = bool(false)]; + tensor var_11600_cast_fp16 = concat(axis = var_10707, interleave = var_11600_interleave_0, values = (var_11516_cast_fp16, var_11518_cast_fp16, var_11520_cast_fp16, var_11522_cast_fp16))[name = string("op_11600_cast_fp16")]; + bool var_11602_interleave_0 = const()[name = string("op_11602_interleave_0"), val = bool(false)]; + tensor var_11602_cast_fp16 = concat(axis = var_10707, interleave = var_11602_interleave_0, values = (var_11524_cast_fp16, var_11526_cast_fp16, var_11528_cast_fp16, var_11530_cast_fp16))[name = string("op_11602_cast_fp16")]; + bool var_11604_interleave_0 = const()[name = string("op_11604_interleave_0"), val = bool(false)]; + tensor var_11604_cast_fp16 = concat(axis = var_10707, interleave = var_11604_interleave_0, values = (var_11532_cast_fp16, var_11534_cast_fp16, var_11536_cast_fp16, var_11538_cast_fp16))[name = string("op_11604_cast_fp16")]; + bool var_11606_interleave_0 = const()[name = string("op_11606_interleave_0"), val = bool(false)]; + tensor var_11606_cast_fp16 = concat(axis = var_10707, interleave = var_11606_interleave_0, values = (var_11540_cast_fp16, var_11542_cast_fp16, var_11544_cast_fp16, var_11546_cast_fp16))[name = string("op_11606_cast_fp16")]; + bool var_11608_interleave_0 = const()[name = string("op_11608_interleave_0"), val = bool(false)]; + tensor var_11608_cast_fp16 = concat(axis = var_10707, interleave = var_11608_interleave_0, values = (var_11548_cast_fp16, var_11550_cast_fp16, var_11552_cast_fp16, var_11554_cast_fp16))[name = string("op_11608_cast_fp16")]; + bool var_11610_interleave_0 = const()[name = string("op_11610_interleave_0"), val = bool(false)]; + tensor var_11610_cast_fp16 = concat(axis = var_10707, interleave = var_11610_interleave_0, values = (var_11556_cast_fp16, var_11558_cast_fp16, var_11560_cast_fp16, var_11562_cast_fp16))[name = string("op_11610_cast_fp16")]; + bool var_11612_interleave_0 = const()[name = string("op_11612_interleave_0"), val = bool(false)]; + tensor var_11612_cast_fp16 = concat(axis = var_10707, interleave = var_11612_interleave_0, values = (var_11564_cast_fp16, var_11566_cast_fp16, var_11568_cast_fp16, var_11570_cast_fp16))[name = string("op_11612_cast_fp16")]; + bool var_11614_interleave_0 = const()[name = string("op_11614_interleave_0"), val = bool(false)]; + tensor var_11614_cast_fp16 = concat(axis = var_10707, interleave = var_11614_interleave_0, values = (var_11572_cast_fp16, var_11574_cast_fp16, var_11576_cast_fp16, var_11578_cast_fp16))[name = string("op_11614_cast_fp16")]; + bool var_11616_interleave_0 = const()[name = string("op_11616_interleave_0"), val = bool(false)]; + tensor var_11616_cast_fp16 = concat(axis = var_10707, interleave = var_11616_interleave_0, values = (var_11580_cast_fp16, var_11582_cast_fp16, var_11584_cast_fp16, var_11586_cast_fp16))[name = string("op_11616_cast_fp16")]; + bool var_11618_interleave_0 = const()[name = string("op_11618_interleave_0"), val = bool(false)]; + tensor var_11618_cast_fp16 = concat(axis = var_10707, interleave = var_11618_interleave_0, values = (var_11588_cast_fp16, var_11590_cast_fp16, var_11592_cast_fp16, var_11594_cast_fp16))[name = string("op_11618_cast_fp16")]; + bool input_89_interleave_0 = const()[name = string("input_89_interleave_0"), val = bool(false)]; + tensor input_89_cast_fp16 = concat(axis = var_10724, interleave = input_89_interleave_0, values = (var_11596_cast_fp16, var_11598_cast_fp16, var_11600_cast_fp16, var_11602_cast_fp16, var_11604_cast_fp16, var_11606_cast_fp16, var_11608_cast_fp16, var_11610_cast_fp16, var_11612_cast_fp16, var_11614_cast_fp16, var_11616_cast_fp16, var_11618_cast_fp16))[name = string("input_89_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165690624)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166870336)))]; + tensor obj_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_11637_to_fp16 = const()[name = string("op_11637_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_11637_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166871936)))]; + tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166873536)))]; + fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166875136)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171593792)))]; + tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; + tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171600000)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176318656)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_11675_to_fp16 = const()[name = string("op_11675_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_11675_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176320256)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176321856)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_11699_pad_type_0 = const()[name = string("op_11699_pad_type_0"), val = string("valid")]; + tensor var_11699_strides_0 = const()[name = string("op_11699_strides_0"), val = tensor([1, 1])]; + tensor var_11699_pad_0 = const()[name = string("op_11699_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11699_dilations_0 = const()[name = string("op_11699_dilations_0"), val = tensor([1, 1])]; + int32 var_11699_groups_0 = const()[name = string("op_11699_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176323456)))]; + tensor var_11699_cast_fp16 = conv(dilations = var_11699_dilations_0, groups = var_11699_groups_0, pad = var_11699_pad_0, pad_type = var_11699_pad_type_0, strides = var_11699_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11699_cast_fp16")]; + string var_11706_pad_type_0 = const()[name = string("op_11706_pad_type_0"), val = string("valid")]; + tensor var_11706_strides_0 = const()[name = string("op_11706_strides_0"), val = tensor([1, 1])]; + tensor var_11706_pad_0 = const()[name = string("op_11706_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11706_dilations_0 = const()[name = string("op_11706_dilations_0"), val = tensor([1, 1])]; + int32 var_11706_groups_0 = const()[name = string("op_11706_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177503168)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178682880)))]; + tensor var_11706_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_11706_dilations_0, groups = var_11706_groups_0, pad = var_11706_pad_0, pad_type = var_11706_pad_type_0, strides = var_11706_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11706_cast_fp16")]; + string var_11724_pad_type_0 = const()[name = string("op_11724_pad_type_0"), val = string("valid")]; + tensor var_11724_strides_0 = const()[name = string("op_11724_strides_0"), val = tensor([1, 1])]; + tensor var_11724_pad_0 = const()[name = string("op_11724_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11724_dilations_0 = const()[name = string("op_11724_dilations_0"), val = tensor([1, 1])]; + int32 var_11724_groups_0 = const()[name = string("op_11724_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178684480)))]; + tensor var_11724_cast_fp16 = conv(dilations = var_11724_dilations_0, groups = var_11724_groups_0, pad = var_11724_pad_0, pad_type = var_11724_pad_type_0, strides = var_11724_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11724_cast_fp16")]; + string var_11731_pad_type_0 = const()[name = string("op_11731_pad_type_0"), val = string("valid")]; + tensor var_11731_strides_0 = const()[name = string("op_11731_strides_0"), val = tensor([1, 1])]; + tensor var_11731_pad_0 = const()[name = string("op_11731_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11731_dilations_0 = const()[name = string("op_11731_dilations_0"), val = tensor([1, 1])]; + int32 var_11731_groups_0 = const()[name = string("op_11731_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179864192)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181043904)))]; + tensor var_11731_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_11731_dilations_0, groups = var_11731_groups_0, pad = var_11731_pad_0, pad_type = var_11731_pad_type_0, strides = var_11731_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11731_cast_fp16")]; + string var_11749_pad_type_0 = const()[name = string("op_11749_pad_type_0"), val = string("valid")]; + tensor var_11749_strides_0 = const()[name = string("op_11749_strides_0"), val = tensor([1, 1])]; + tensor var_11749_pad_0 = const()[name = string("op_11749_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11749_dilations_0 = const()[name = string("op_11749_dilations_0"), val = tensor([1, 1])]; + int32 var_11749_groups_0 = const()[name = string("op_11749_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181045504)))]; + tensor var_11749_cast_fp16 = conv(dilations = var_11749_dilations_0, groups = var_11749_groups_0, pad = var_11749_pad_0, pad_type = var_11749_pad_type_0, strides = var_11749_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11749_cast_fp16")]; + string var_11756_pad_type_0 = const()[name = string("op_11756_pad_type_0"), val = string("valid")]; + tensor var_11756_strides_0 = const()[name = string("op_11756_strides_0"), val = tensor([1, 1])]; + tensor var_11756_pad_0 = const()[name = string("op_11756_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11756_dilations_0 = const()[name = string("op_11756_dilations_0"), val = tensor([1, 1])]; + int32 var_11756_groups_0 = const()[name = string("op_11756_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182225216)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183404928)))]; + tensor var_11756_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_11756_dilations_0, groups = var_11756_groups_0, pad = var_11756_pad_0, pad_type = var_11756_pad_type_0, strides = var_11756_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11756_cast_fp16")]; + string var_11774_pad_type_0 = const()[name = string("op_11774_pad_type_0"), val = string("valid")]; + tensor var_11774_strides_0 = const()[name = string("op_11774_strides_0"), val = tensor([1, 1])]; + tensor var_11774_pad_0 = const()[name = string("op_11774_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11774_dilations_0 = const()[name = string("op_11774_dilations_0"), val = tensor([1, 1])]; + int32 var_11774_groups_0 = const()[name = string("op_11774_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183406528)))]; + tensor var_11774_cast_fp16 = conv(dilations = var_11774_dilations_0, groups = var_11774_groups_0, pad = var_11774_pad_0, pad_type = var_11774_pad_type_0, strides = var_11774_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11774_cast_fp16")]; + string var_11781_pad_type_0 = const()[name = string("op_11781_pad_type_0"), val = string("valid")]; + tensor var_11781_strides_0 = const()[name = string("op_11781_strides_0"), val = tensor([1, 1])]; + tensor var_11781_pad_0 = const()[name = string("op_11781_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11781_dilations_0 = const()[name = string("op_11781_dilations_0"), val = tensor([1, 1])]; + int32 var_11781_groups_0 = const()[name = string("op_11781_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184586240)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185765952)))]; + tensor var_11781_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = var_11781_dilations_0, groups = var_11781_groups_0, pad = var_11781_pad_0, pad_type = var_11781_pad_type_0, strides = var_11781_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11781_cast_fp16")]; + string var_11799_pad_type_0 = const()[name = string("op_11799_pad_type_0"), val = string("valid")]; + tensor var_11799_strides_0 = const()[name = string("op_11799_strides_0"), val = tensor([1, 1])]; + tensor var_11799_pad_0 = const()[name = string("op_11799_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11799_dilations_0 = const()[name = string("op_11799_dilations_0"), val = tensor([1, 1])]; + int32 var_11799_groups_0 = const()[name = string("op_11799_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185767552)))]; + tensor var_11799_cast_fp16 = conv(dilations = var_11799_dilations_0, groups = var_11799_groups_0, pad = var_11799_pad_0, pad_type = var_11799_pad_type_0, strides = var_11799_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11799_cast_fp16")]; + string var_11806_pad_type_0 = const()[name = string("op_11806_pad_type_0"), val = string("valid")]; + tensor var_11806_strides_0 = const()[name = string("op_11806_strides_0"), val = tensor([1, 1])]; + tensor var_11806_pad_0 = const()[name = string("op_11806_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11806_dilations_0 = const()[name = string("op_11806_dilations_0"), val = tensor([1, 1])]; + int32 var_11806_groups_0 = const()[name = string("op_11806_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186947264)))]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188126976)))]; + tensor var_11806_cast_fp16 = conv(bias = decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16, dilations = var_11806_dilations_0, groups = var_11806_groups_0, pad = var_11806_pad_0, pad_type = var_11806_pad_type_0, strides = var_11806_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11806_cast_fp16")]; + string var_11824_pad_type_0 = const()[name = string("op_11824_pad_type_0"), val = string("valid")]; + tensor var_11824_strides_0 = const()[name = string("op_11824_strides_0"), val = tensor([1, 1])]; + tensor var_11824_pad_0 = const()[name = string("op_11824_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11824_dilations_0 = const()[name = string("op_11824_dilations_0"), val = tensor([1, 1])]; + int32 var_11824_groups_0 = const()[name = string("op_11824_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188128576)))]; + tensor var_11824_cast_fp16 = conv(dilations = var_11824_dilations_0, groups = var_11824_groups_0, pad = var_11824_pad_0, pad_type = var_11824_pad_type_0, strides = var_11824_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11824_cast_fp16")]; + string var_11831_pad_type_0 = const()[name = string("op_11831_pad_type_0"), val = string("valid")]; + tensor var_11831_strides_0 = const()[name = string("op_11831_strides_0"), val = tensor([1, 1])]; + tensor var_11831_pad_0 = const()[name = string("op_11831_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11831_dilations_0 = const()[name = string("op_11831_dilations_0"), val = tensor([1, 1])]; + int32 var_11831_groups_0 = const()[name = string("op_11831_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189308288)))]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190488000)))]; + tensor var_11831_cast_fp16 = conv(bias = decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16, dilations = var_11831_dilations_0, groups = var_11831_groups_0, pad = var_11831_pad_0, pad_type = var_11831_pad_type_0, strides = var_11831_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11831_cast_fp16")]; + string var_11849_pad_type_0 = const()[name = string("op_11849_pad_type_0"), val = string("valid")]; + tensor var_11849_strides_0 = const()[name = string("op_11849_strides_0"), val = tensor([1, 1])]; + tensor var_11849_pad_0 = const()[name = string("op_11849_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11849_dilations_0 = const()[name = string("op_11849_dilations_0"), val = tensor([1, 1])]; + int32 var_11849_groups_0 = const()[name = string("op_11849_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190489600)))]; + tensor var_11849_cast_fp16 = conv(dilations = var_11849_dilations_0, groups = var_11849_groups_0, pad = var_11849_pad_0, pad_type = var_11849_pad_type_0, strides = var_11849_strides_0, weight = decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11849_cast_fp16")]; + string var_11856_pad_type_0 = const()[name = string("op_11856_pad_type_0"), val = string("valid")]; + tensor var_11856_strides_0 = const()[name = string("op_11856_strides_0"), val = tensor([1, 1])]; + tensor var_11856_pad_0 = const()[name = string("op_11856_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11856_dilations_0 = const()[name = string("op_11856_dilations_0"), val = tensor([1, 1])]; + int32 var_11856_groups_0 = const()[name = string("op_11856_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191669312)))]; + tensor decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192849024)))]; + tensor var_11856_cast_fp16 = conv(bias = decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16, dilations = var_11856_dilations_0, groups = var_11856_groups_0, pad = var_11856_pad_0, pad_type = var_11856_pad_type_0, strides = var_11856_strides_0, weight = decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11856_cast_fp16")]; + string var_11874_pad_type_0 = const()[name = string("op_11874_pad_type_0"), val = string("valid")]; + tensor var_11874_strides_0 = const()[name = string("op_11874_strides_0"), val = tensor([1, 1])]; + tensor var_11874_pad_0 = const()[name = string("op_11874_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11874_dilations_0 = const()[name = string("op_11874_dilations_0"), val = tensor([1, 1])]; + int32 var_11874_groups_0 = const()[name = string("op_11874_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192850624)))]; + tensor var_11874_cast_fp16 = conv(dilations = var_11874_dilations_0, groups = var_11874_groups_0, pad = var_11874_pad_0, pad_type = var_11874_pad_type_0, strides = var_11874_strides_0, weight = decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11874_cast_fp16")]; + string var_11881_pad_type_0 = const()[name = string("op_11881_pad_type_0"), val = string("valid")]; + tensor var_11881_strides_0 = const()[name = string("op_11881_strides_0"), val = tensor([1, 1])]; + tensor var_11881_pad_0 = const()[name = string("op_11881_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11881_dilations_0 = const()[name = string("op_11881_dilations_0"), val = tensor([1, 1])]; + int32 var_11881_groups_0 = const()[name = string("op_11881_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194030336)))]; + tensor decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195210048)))]; + tensor var_11881_cast_fp16 = conv(bias = decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16, dilations = var_11881_dilations_0, groups = var_11881_groups_0, pad = var_11881_pad_0, pad_type = var_11881_pad_type_0, strides = var_11881_strides_0, weight = decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11881_cast_fp16")]; + string var_11899_pad_type_0 = const()[name = string("op_11899_pad_type_0"), val = string("valid")]; + tensor var_11899_strides_0 = const()[name = string("op_11899_strides_0"), val = tensor([1, 1])]; + tensor var_11899_pad_0 = const()[name = string("op_11899_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11899_dilations_0 = const()[name = string("op_11899_dilations_0"), val = tensor([1, 1])]; + int32 var_11899_groups_0 = const()[name = string("op_11899_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195211648)))]; + tensor var_11899_cast_fp16 = conv(dilations = var_11899_dilations_0, groups = var_11899_groups_0, pad = var_11899_pad_0, pad_type = var_11899_pad_type_0, strides = var_11899_strides_0, weight = decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11899_cast_fp16")]; + string var_11906_pad_type_0 = const()[name = string("op_11906_pad_type_0"), val = string("valid")]; + tensor var_11906_strides_0 = const()[name = string("op_11906_strides_0"), val = tensor([1, 1])]; + tensor var_11906_pad_0 = const()[name = string("op_11906_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11906_dilations_0 = const()[name = string("op_11906_dilations_0"), val = tensor([1, 1])]; + int32 var_11906_groups_0 = const()[name = string("op_11906_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196391360)))]; + tensor decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197571072)))]; + tensor var_11906_cast_fp16 = conv(bias = decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16, dilations = var_11906_dilations_0, groups = var_11906_groups_0, pad = var_11906_pad_0, pad_type = var_11906_pad_type_0, strides = var_11906_strides_0, weight = decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11906_cast_fp16")]; + string var_11924_pad_type_0 = const()[name = string("op_11924_pad_type_0"), val = string("valid")]; + tensor var_11924_strides_0 = const()[name = string("op_11924_strides_0"), val = tensor([1, 1])]; + tensor var_11924_pad_0 = const()[name = string("op_11924_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11924_dilations_0 = const()[name = string("op_11924_dilations_0"), val = tensor([1, 1])]; + int32 var_11924_groups_0 = const()[name = string("op_11924_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197572672)))]; + tensor var_11924_cast_fp16 = conv(dilations = var_11924_dilations_0, groups = var_11924_groups_0, pad = var_11924_pad_0, pad_type = var_11924_pad_type_0, strides = var_11924_strides_0, weight = decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11924_cast_fp16")]; + string var_11931_pad_type_0 = const()[name = string("op_11931_pad_type_0"), val = string("valid")]; + tensor var_11931_strides_0 = const()[name = string("op_11931_strides_0"), val = tensor([1, 1])]; + tensor var_11931_pad_0 = const()[name = string("op_11931_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11931_dilations_0 = const()[name = string("op_11931_dilations_0"), val = tensor([1, 1])]; + int32 var_11931_groups_0 = const()[name = string("op_11931_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198752384)))]; + tensor decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199932096)))]; + tensor var_11931_cast_fp16 = conv(bias = decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16, dilations = var_11931_dilations_0, groups = var_11931_groups_0, pad = var_11931_pad_0, pad_type = var_11931_pad_type_0, strides = var_11931_strides_0, weight = decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11931_cast_fp16")]; + string var_11949_pad_type_0 = const()[name = string("op_11949_pad_type_0"), val = string("valid")]; + tensor var_11949_strides_0 = const()[name = string("op_11949_strides_0"), val = tensor([1, 1])]; + tensor var_11949_pad_0 = const()[name = string("op_11949_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11949_dilations_0 = const()[name = string("op_11949_dilations_0"), val = tensor([1, 1])]; + int32 var_11949_groups_0 = const()[name = string("op_11949_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199933696)))]; + tensor var_11949_cast_fp16 = conv(dilations = var_11949_dilations_0, groups = var_11949_groups_0, pad = var_11949_pad_0, pad_type = var_11949_pad_type_0, strides = var_11949_strides_0, weight = decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11949_cast_fp16")]; + string var_11956_pad_type_0 = const()[name = string("op_11956_pad_type_0"), val = string("valid")]; + tensor var_11956_strides_0 = const()[name = string("op_11956_strides_0"), val = tensor([1, 1])]; + tensor var_11956_pad_0 = const()[name = string("op_11956_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11956_dilations_0 = const()[name = string("op_11956_dilations_0"), val = tensor([1, 1])]; + int32 var_11956_groups_0 = const()[name = string("op_11956_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201113408)))]; + tensor decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202293120)))]; + tensor var_11956_cast_fp16 = conv(bias = decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16, dilations = var_11956_dilations_0, groups = var_11956_groups_0, pad = var_11956_pad_0, pad_type = var_11956_pad_type_0, strides = var_11956_strides_0, weight = decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11956_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202294720)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203474432)))]; + tensor decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204654144)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_11986 = const()[name = string("op_11986"), val = int32(0)]; + bool input_99_interleave_0 = const()[name = string("input_99_interleave_0"), val = bool(false)]; + tensor input_99_cast_fp16 = concat(axis = var_11986, interleave = input_99_interleave_0, values = (var_11699_cast_fp16, var_11724_cast_fp16, var_11749_cast_fp16, var_11774_cast_fp16, var_11799_cast_fp16, var_11824_cast_fp16, var_11849_cast_fp16, var_11874_cast_fp16, var_11899_cast_fp16, var_11924_cast_fp16, var_11949_cast_fp16, k_cast_fp16))[name = string("input_99_cast_fp16")]; + int32 var_11989 = const()[name = string("op_11989"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_11989, interleave = input_interleave_0, values = (var_11706_cast_fp16, var_11731_cast_fp16, var_11756_cast_fp16, var_11781_cast_fp16, var_11806_cast_fp16, var_11831_cast_fp16, var_11856_cast_fp16, var_11881_cast_fp16, var_11906_cast_fp16, var_11931_cast_fp16, var_11956_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_11996_pad_0 = const()[name = string("op_11996_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_11996_mode_0 = const()[name = string("op_11996_mode_0"), val = string("constant")]; + fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_13_to_fp16, mode = var_11996_mode_0, pad = var_11996_pad_0, x = input_99_cast_fp16)[name = string("op_11996_cast_fp16")]; + tensor var_12002_pad_0 = const()[name = string("op_12002_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_12002_mode_0 = const()[name = string("op_12002_mode_0"), val = string("constant")]; + fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_14_to_fp16, mode = var_12002_mode_0, pad = var_12002_pad_0, x = input_cast_fp16)[name = string("op_12002_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-small.en/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-small.en/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..6a1518cfa297431939c886918dfe996c06b25a76 --- /dev/null +++ b/openai_whisper-small.en/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:506a026b04016ef63909aff039c6945ee5216be3bc57be18dd454c92c3b890ad +size 204655744 diff --git a/openai_whisper-small.en/LICENSE_NOTICE.txt b/openai_whisper-small.en/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-small.en/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-small.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-small.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a11fbb2cd75b96eb2120a672afefa298c2ef857b --- /dev/null +++ b/openai_whisper-small.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc05e563ee0c556e3f578e04be5fb67b4e7520124403f2561f39102f0f2b33d +size 243 diff --git a/openai_whisper-small.en/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-small.en/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a3544b6644c1af93ca6bdabb67a1c51e80eaa552 --- /dev/null +++ b/openai_whisper-small.en/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ef11ea703011eab03287ec661f999e19c2c78cf67d531b5e6afa02e18f913d +size 328 diff --git a/openai_whisper-small.en/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-small.en/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1a60dd494a857817b67d87cd920baa6824e74b61 --- /dev/null +++ b/openai_whisper-small.en/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-small.en/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-small.en/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cf4cd446f68b88655d00a7df7063aa46937a9bdd --- /dev/null +++ b/openai_whisper-small.en/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-small.en/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-small.en/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f7d28ffac464e9e7086a526930f0059187de8d01 --- /dev/null +++ b/openai_whisper-small.en/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83 +size 354080 diff --git a/openai_whisper-small.en/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-small.en/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c0ffdbc591c30fcbb87b76e96c5540f4ca9752e6 --- /dev/null +++ b/openai_whisper-small.en/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49de74e378937d7086068566ffd13793233f52566ad301ac0882611fc36e4bdd +size 243 diff --git a/openai_whisper-small.en/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-small.en/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..2a559681d08f2ad78842877bb0279ba2e76a65ae --- /dev/null +++ b/openai_whisper-small.en/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7af00de6eebff972d49c192df21fdcc49dc037f8043e50e89fdd8e3831e1a8e8 +size 754 diff --git a/openai_whisper-small.en/TextDecoder.mlmodelc/metadata.json b/openai_whisper-small.en/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..a576977650e28cfe81ec1023166e97d6aa1dc78b --- /dev/null +++ b/openai_whisper-small.en/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51864)", + "shortDescription" : "", + "shape" : "[1, 1, 51864]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 9216 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 9216, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 9216 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 9216, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 24, + "Ios18.mul" : 48, + "Ios18.matmul" : 48, + "Ios18.batchNorm" : 37, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 85, + "Ios18.layerNorm" : 37, + "Ios18.reshape" : 96, + "Ios18.linear" : 1, + "Ios18.conv" : 96, + "Ios18.gelu" : 12, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 62, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 448)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 448)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-small.en/TextDecoder.mlmodelc/model.mil b/openai_whisper-small.en/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..c76e0075f2a33d958f18590ac2eced29f81a369b --- /dev/null +++ b/openai_whisper-small.en/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,1900 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_42_axis_0 = const()[name = string("op_42_axis_0"), val = int32(0)]; + int32 var_42_batch_dims_0 = const()[name = string("op_42_batch_dims_0"), val = int32(0)]; + bool var_42_validate_indices_0 = const()[name = string("op_42_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_42_cast_fp16 = gather(axis = var_42_axis_0, batch_dims = var_42_batch_dims_0, indices = input_ids, validate_indices = var_42_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_42_cast_fp16")]; + int32 var_46_axis_0 = const()[name = string("op_46_axis_0"), val = int32(0)]; + int32 var_46_batch_dims_0 = const()[name = string("op_46_batch_dims_0"), val = int32(0)]; + bool var_46_validate_indices_0 = const()[name = string("op_46_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79663232)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_183")]; + tensor var_46_cast_fp16_cast_uint16 = gather(axis = var_46_axis_0, batch_dims = var_46_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_46_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_46_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_42_cast_fp16, y = var_46_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_60_axes_0 = const()[name = string("op_60_axes_0"), val = tensor([2])]; + tensor var_60_cast_fp16 = expand_dims(axes = var_60_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_60_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_60_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80351424)))]; + int32 var_65_axis_0 = const()[name = string("op_65_axis_0"), val = int32(0)]; + tensor var_65_cast_fp16_0, tensor var_65_cast_fp16_1, tensor var_65_cast_fp16_2, tensor var_65_cast_fp16_3, tensor var_65_cast_fp16_4, tensor var_65_cast_fp16_5, tensor var_65_cast_fp16_6, tensor var_65_cast_fp16_7, tensor var_65_cast_fp16_8, tensor var_65_cast_fp16_9, tensor var_65_cast_fp16_10, tensor var_65_cast_fp16_11 = split(axis = var_65_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_65_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80351552)))]; + int32 var_80_axis_0 = const()[name = string("op_80_axis_0"), val = int32(0)]; + tensor var_80_cast_fp16_0, tensor var_80_cast_fp16_1, tensor var_80_cast_fp16_2, tensor var_80_cast_fp16_3, tensor var_80_cast_fp16_4, tensor var_80_cast_fp16_5, tensor var_80_cast_fp16_6, tensor var_80_cast_fp16_7, tensor var_80_cast_fp16_8, tensor var_80_cast_fp16_9, tensor var_80_cast_fp16_10, tensor var_80_cast_fp16_11 = split(axis = var_80_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_80_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 768, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 768, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_108 = const()[name = string("op_108"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_133_to_fp16 = const()[name = string("op_133_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_133_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80351680)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80353280)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80354880)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80356480)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80358080)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81537792)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81539392)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82719104)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83898816)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_168_axes_0 = const()[name = string("op_168_axes_0"), val = tensor([1])]; + tensor var_168_cast_fp16 = expand_dims(axes = var_168_axes_0, x = kv_cache_update_mask)[name = string("op_168_cast_fp16")]; + tensor var_169_axes_0 = const()[name = string("op_169_axes_0"), val = tensor([2])]; + tensor var_169_cast_fp16 = expand_dims(axes = var_169_axes_0, x = var_168_cast_fp16)[name = string("op_169_cast_fp16")]; + tensor var_171_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_169_cast_fp16)[name = string("op_171_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_65_cast_fp16_0, y = var_171_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_173_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_169_cast_fp16)[name = string("op_173_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_80_cast_fp16_0, y = var_173_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_176 = const()[name = string("op_176"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_176, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_178_to_fp16 = const()[name = string("op_178_to_fp16"), val = fp16(0x1p-3)]; + tensor var_179_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_178_to_fp16)[name = string("op_179_cast_fp16")]; + tensor var_180 = const()[name = string("op_180"), val = tensor([1, 12, 64, -1])]; + tensor var_181_cast_fp16 = reshape(shape = var_180, x = key_1_cast_fp16)[name = string("op_181_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_179_cast_fp16, y = var_181_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_185_axes_0 = const()[name = string("op_185_axes_0"), val = tensor([1])]; + tensor var_185_cast_fp16 = expand_dims(axes = var_185_axes_0, x = decoder_key_padding_mask)[name = string("op_185_cast_fp16")]; + tensor var_186_axes_0 = const()[name = string("op_186_axes_0"), val = tensor([2])]; + tensor var_186_cast_fp16 = expand_dims(axes = var_186_axes_0, x = var_185_cast_fp16)[name = string("op_186_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_189_cast_fp16 = softmax(axis = var_108, x = mh_w_3_cast_fp16)[name = string("op_189_cast_fp16")]; + tensor var_190 = const()[name = string("op_190"), val = tensor([1, 12, 64, -1])]; + tensor var_191_cast_fp16 = reshape(shape = var_190, x = value_1_cast_fp16)[name = string("op_191_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_191_cast_fp16, y = var_189_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_194 = const()[name = string("op_194"), val = tensor([1, 768, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_194, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83900416)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85080128)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_216_to_fp16 = const()[name = string("op_216_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_216_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85081728)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85083328)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85084928)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86264640)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_236 = const()[name = string("op_236"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_236, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_238_to_fp16 = const()[name = string("op_238_to_fp16"), val = fp16(0x1p-3)]; + tensor var_239_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_238_to_fp16)[name = string("op_239_cast_fp16")]; + tensor var_240 = const()[name = string("op_240"), val = tensor([1, 12, 64, -1])]; + tensor var_241_cast_fp16 = reshape(shape = var_240, x = obj_17_cast_fp16)[name = string("op_241_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_239_cast_fp16, y = var_241_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_245_axes_0 = const()[name = string("op_245_axes_0"), val = tensor([1])]; + tensor var_245_cast_fp16 = expand_dims(axes = var_245_axes_0, x = read_state_4)[name = string("op_245_cast_fp16")]; + tensor var_246_axes_0 = const()[name = string("op_246_axes_0"), val = tensor([2])]; + tensor var_246_cast_fp16 = expand_dims(axes = var_246_axes_0, x = var_245_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_108, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_250 = const()[name = string("op_250"), val = tensor([1, 12, 64, -1])]; + tensor var_251_cast_fp16 = reshape(shape = var_250, x = obj_19_cast_fp16)[name = string("op_251_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_251_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_254 = const()[name = string("op_254"), val = tensor([1, 768, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_254, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86266240)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87445952)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_272_to_fp16 = const()[name = string("op_272_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_272_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87447552)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87449152)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87450752)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92169408)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92175616)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96894272)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 768, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 768, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_317 = const()[name = string("op_317"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_342_to_fp16 = const()[name = string("op_342_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_342_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96895872)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96897472)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96899072)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98078784)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98080384)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99260096)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100439808)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_380_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_169_cast_fp16)[name = string("op_380_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_65_cast_fp16_1, y = var_380_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_382_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_169_cast_fp16)[name = string("op_382_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_80_cast_fp16_1, y = var_382_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_385 = const()[name = string("op_385"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_385, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_387_to_fp16 = const()[name = string("op_387_to_fp16"), val = fp16(0x1p-3)]; + tensor var_388_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_387_to_fp16)[name = string("op_388_cast_fp16")]; + tensor var_389 = const()[name = string("op_389"), val = tensor([1, 12, 64, -1])]; + tensor var_390_cast_fp16 = reshape(shape = var_389, x = key_3_cast_fp16)[name = string("op_390_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_388_cast_fp16, y = var_390_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_398_cast_fp16 = softmax(axis = var_317, x = mh_w_11_cast_fp16)[name = string("op_398_cast_fp16")]; + tensor var_399 = const()[name = string("op_399"), val = tensor([1, 12, 64, -1])]; + tensor var_400_cast_fp16 = reshape(shape = var_399, x = value_3_cast_fp16)[name = string("op_400_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_400_cast_fp16, y = var_398_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_403 = const()[name = string("op_403"), val = tensor([1, 768, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_403, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100441408)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101621120)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_425_to_fp16 = const()[name = string("op_425_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_425_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101622720)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101624320)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101625920)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102805632)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_445 = const()[name = string("op_445"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_445, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_447_to_fp16 = const()[name = string("op_447_to_fp16"), val = fp16(0x1p-3)]; + tensor var_448_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_447_to_fp16)[name = string("op_448_cast_fp16")]; + tensor var_449 = const()[name = string("op_449"), val = tensor([1, 12, 64, -1])]; + tensor var_450_cast_fp16 = reshape(shape = var_449, x = obj_35_cast_fp16)[name = string("op_450_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_448_cast_fp16, y = var_450_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_317, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_459 = const()[name = string("op_459"), val = tensor([1, 12, 64, -1])]; + tensor var_460_cast_fp16 = reshape(shape = var_459, x = obj_37_cast_fp16)[name = string("op_460_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_460_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_463 = const()[name = string("op_463"), val = tensor([1, 768, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_463, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102807232)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103986944)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_481_to_fp16 = const()[name = string("op_481_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_481_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103988544)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103990144)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103991744)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108710400)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108716608)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113435264)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 768, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 768, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_526 = const()[name = string("op_526"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_551_to_fp16 = const()[name = string("op_551_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_551_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113436864)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113438464)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113440064)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114619776)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114621376)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115801088)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116980800)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_589_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_169_cast_fp16)[name = string("op_589_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_65_cast_fp16_2, y = var_589_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_591_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_169_cast_fp16)[name = string("op_591_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_80_cast_fp16_2, y = var_591_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_594 = const()[name = string("op_594"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_594, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; + tensor var_597_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_596_to_fp16)[name = string("op_597_cast_fp16")]; + tensor var_598 = const()[name = string("op_598"), val = tensor([1, 12, 64, -1])]; + tensor var_599_cast_fp16 = reshape(shape = var_598, x = key_5_cast_fp16)[name = string("op_599_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_597_cast_fp16, y = var_599_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_607_cast_fp16 = softmax(axis = var_526, x = mh_w_19_cast_fp16)[name = string("op_607_cast_fp16")]; + tensor var_608 = const()[name = string("op_608"), val = tensor([1, 12, 64, -1])]; + tensor var_609_cast_fp16 = reshape(shape = var_608, x = value_5_cast_fp16)[name = string("op_609_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_609_cast_fp16, y = var_607_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_612 = const()[name = string("op_612"), val = tensor([1, 768, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_612, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116982400)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118162112)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_634_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118163712)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118165312)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118166912)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119346624)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_654 = const()[name = string("op_654"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_654, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_656_to_fp16 = const()[name = string("op_656_to_fp16"), val = fp16(0x1p-3)]; + tensor var_657_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_656_to_fp16)[name = string("op_657_cast_fp16")]; + tensor var_658 = const()[name = string("op_658"), val = tensor([1, 12, 64, -1])]; + tensor var_659_cast_fp16 = reshape(shape = var_658, x = obj_53_cast_fp16)[name = string("op_659_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_657_cast_fp16, y = var_659_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_526, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_668 = const()[name = string("op_668"), val = tensor([1, 12, 64, -1])]; + tensor var_669_cast_fp16 = reshape(shape = var_668, x = obj_55_cast_fp16)[name = string("op_669_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_669_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_672 = const()[name = string("op_672"), val = tensor([1, 768, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_672, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119348224)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120527936)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_690_to_fp16 = const()[name = string("op_690_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_690_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120529536)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120531136)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120532736)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125251392)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125257600)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129976256)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 768, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 768, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_735 = const()[name = string("op_735"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_760_to_fp16 = const()[name = string("op_760_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_760_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129977856)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129979456)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129981056)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131160768)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_7_pad_type_0 = const()[name = string("current_key_7_pad_type_0"), val = string("valid")]; + tensor current_key_7_strides_0 = const()[name = string("current_key_7_strides_0"), val = tensor([1, 1])]; + tensor current_key_7_pad_0 = const()[name = string("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_7_dilations_0 = const()[name = string("current_key_7_dilations_0"), val = tensor([1, 1])]; + int32 current_key_7_groups_0 = const()[name = string("current_key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131162368)))]; + tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_7_cast_fp16")]; + string current_value_7_pad_type_0 = const()[name = string("current_value_7_pad_type_0"), val = string("valid")]; + tensor current_value_7_strides_0 = const()[name = string("current_value_7_strides_0"), val = tensor([1, 1])]; + tensor current_value_7_pad_0 = const()[name = string("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_7_dilations_0 = const()[name = string("current_value_7_dilations_0"), val = tensor([1, 1])]; + int32 current_value_7_groups_0 = const()[name = string("current_value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132342080)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133521792)))]; + tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_7_cast_fp16")]; + tensor var_798_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_169_cast_fp16)[name = string("op_798_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = var_65_cast_fp16_3, y = var_798_cast_fp16)[name = string("key_7_cast_fp16")]; + tensor var_800_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_169_cast_fp16)[name = string("op_800_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = var_80_cast_fp16_3, y = var_800_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_803 = const()[name = string("op_803"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_803, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_805_to_fp16 = const()[name = string("op_805_to_fp16"), val = fp16(0x1p-3)]; + tensor var_806_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_805_to_fp16)[name = string("op_806_cast_fp16")]; + tensor var_807 = const()[name = string("op_807"), val = tensor([1, 12, 64, -1])]; + tensor var_808_cast_fp16 = reshape(shape = var_807, x = key_7_cast_fp16)[name = string("op_808_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_806_cast_fp16, y = var_808_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_816_cast_fp16 = softmax(axis = var_735, x = mh_w_27_cast_fp16)[name = string("op_816_cast_fp16")]; + tensor var_817 = const()[name = string("op_817"), val = tensor([1, 12, 64, -1])]; + tensor var_818_cast_fp16 = reshape(shape = var_817, x = value_7_cast_fp16)[name = string("op_818_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_818_cast_fp16, y = var_816_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_821 = const()[name = string("op_821"), val = tensor([1, 768, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_821, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133523392)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134703104)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_843_to_fp16 = const()[name = string("op_843_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_843_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134704704)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134706304)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134707904)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135887616)))]; + tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_15_cast_fp16")]; + tensor var_863 = const()[name = string("op_863"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_863, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; + fp16 var_865_to_fp16 = const()[name = string("op_865_to_fp16"), val = fp16(0x1p-3)]; + tensor var_866_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_865_to_fp16)[name = string("op_866_cast_fp16")]; + tensor var_867 = const()[name = string("op_867"), val = tensor([1, 12, 64, -1])]; + tensor var_868_cast_fp16 = reshape(shape = var_867, x = obj_71_cast_fp16)[name = string("op_868_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_866_cast_fp16, y = var_868_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_31_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_735, x = mh_w_31_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_877 = const()[name = string("op_877"), val = tensor([1, 12, 64, -1])]; + tensor var_878_cast_fp16 = reshape(shape = var_877, x = obj_73_cast_fp16)[name = string("op_878_cast_fp16")]; + bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; + bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_878_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_15_cast_fp16")]; + tensor var_881 = const()[name = string("op_881"), val = tensor([1, 768, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_881, x = attn_15_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135889216)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137068928)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_899_to_fp16 = const()[name = string("op_899_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_899_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137070528)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137072128)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137073728)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141792384)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141798592)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146517248)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_25_cast_fp16")]; + tensor obj_89_begin_0 = const()[name = string("obj_89_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_89_end_0 = const()[name = string("obj_89_end_0"), val = tensor([5, 768, 1, 1536])]; + tensor obj_89_end_mask_0 = const()[name = string("obj_89_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_89_cast_fp16 = slice_by_index(begin = obj_89_begin_0, end = obj_89_end_0, end_mask = obj_89_end_mask_0, x = read_state_2)[name = string("obj_89_cast_fp16")]; + tensor obj_91_begin_0 = const()[name = string("obj_91_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_91_end_0 = const()[name = string("obj_91_end_0"), val = tensor([5, 768, 1, 1536])]; + tensor obj_91_end_mask_0 = const()[name = string("obj_91_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_91_cast_fp16 = slice_by_index(begin = obj_91_begin_0, end = obj_91_end_0, end_mask = obj_91_end_mask_0, x = read_state_3)[name = string("obj_91_cast_fp16")]; + int32 var_944 = const()[name = string("op_944"), val = int32(3)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_969_to_fp16 = const()[name = string("op_969_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_969_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146518848)))]; + tensor obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146520448)))]; + fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_79_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146522048)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147701760)))]; + tensor query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("query_17_cast_fp16")]; + string current_key_9_pad_type_0 = const()[name = string("current_key_9_pad_type_0"), val = string("valid")]; + tensor current_key_9_strides_0 = const()[name = string("current_key_9_strides_0"), val = tensor([1, 1])]; + tensor current_key_9_pad_0 = const()[name = string("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_9_dilations_0 = const()[name = string("current_key_9_dilations_0"), val = tensor([1, 1])]; + int32 current_key_9_groups_0 = const()[name = string("current_key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147703360)))]; + tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_key_9_cast_fp16")]; + string current_value_9_pad_type_0 = const()[name = string("current_value_9_pad_type_0"), val = string("valid")]; + tensor current_value_9_strides_0 = const()[name = string("current_value_9_strides_0"), val = tensor([1, 1])]; + tensor current_value_9_pad_0 = const()[name = string("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_9_dilations_0 = const()[name = string("current_value_9_dilations_0"), val = tensor([1, 1])]; + int32 current_value_9_groups_0 = const()[name = string("current_value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148883072)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150062784)))]; + tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_value_9_cast_fp16")]; + tensor var_1007_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_169_cast_fp16)[name = string("op_1007_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = var_65_cast_fp16_4, y = var_1007_cast_fp16)[name = string("key_9_cast_fp16")]; + tensor var_1009_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_169_cast_fp16)[name = string("op_1009_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = var_80_cast_fp16_4, y = var_1009_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_1012 = const()[name = string("op_1012"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_1012, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; + fp16 var_1014_to_fp16 = const()[name = string("op_1014_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1015_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1014_to_fp16)[name = string("op_1015_cast_fp16")]; + tensor var_1016 = const()[name = string("op_1016"), val = tensor([1, 12, 64, -1])]; + tensor var_1017_cast_fp16 = reshape(shape = var_1016, x = key_9_cast_fp16)[name = string("op_1017_cast_fp16")]; + bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; + bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_1015_cast_fp16, y = var_1017_cast_fp16)[name = string("mh_w_33_cast_fp16")]; + tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_35_cast_fp16")]; + tensor var_1025_cast_fp16 = softmax(axis = var_944, x = mh_w_35_cast_fp16)[name = string("op_1025_cast_fp16")]; + tensor var_1026 = const()[name = string("op_1026"), val = tensor([1, 12, 64, -1])]; + tensor var_1027_cast_fp16 = reshape(shape = var_1026, x = value_9_cast_fp16)[name = string("op_1027_cast_fp16")]; + bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; + bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1027_cast_fp16, y = var_1025_cast_fp16)[name = string("attn_17_cast_fp16")]; + tensor var_1030 = const()[name = string("op_1030"), val = tensor([1, 768, 1, -1])]; + tensor input_41_cast_fp16 = reshape(shape = var_1030, x = attn_17_cast_fp16)[name = string("input_41_cast_fp16")]; + string obj_85_pad_type_0 = const()[name = string("obj_85_pad_type_0"), val = string("valid")]; + tensor obj_85_strides_0 = const()[name = string("obj_85_strides_0"), val = tensor([1, 1])]; + tensor obj_85_pad_0 = const()[name = string("obj_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_85_dilations_0 = const()[name = string("obj_85_dilations_0"), val = tensor([1, 1])]; + int32 obj_85_groups_0 = const()[name = string("obj_85_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150064384)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151244096)))]; + tensor obj_85_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_85_dilations_0, groups = obj_85_groups_0, pad = obj_85_pad_0, pad_type = obj_85_pad_type_0, strides = obj_85_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_85_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_85_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_1052_to_fp16 = const()[name = string("op_1052_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1052_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor obj_87_gamma_0_to_fp16 = const()[name = string("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151245696)))]; + tensor obj_87_beta_0_to_fp16 = const()[name = string("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151247296)))]; + fp16 obj_87_epsilon_0_to_fp16 = const()[name = string("obj_87_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("obj_87_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151248896)))]; + tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152428608)))]; + tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16, x = obj_87_cast_fp16)[name = string("query_19_cast_fp16")]; + tensor var_1072 = const()[name = string("op_1072"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_1072, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; + fp16 var_1074_to_fp16 = const()[name = string("op_1074_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1075_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1074_to_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1076 = const()[name = string("op_1076"), val = tensor([1, 12, 64, -1])]; + tensor var_1077_cast_fp16 = reshape(shape = var_1076, x = obj_89_cast_fp16)[name = string("op_1077_cast_fp16")]; + bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; + bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1075_cast_fp16, y = var_1077_cast_fp16)[name = string("mh_w_37_cast_fp16")]; + tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_39_cast_fp16")]; + tensor obj_95_cast_fp16 = softmax(axis = var_944, x = mh_w_39_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor var_1086 = const()[name = string("op_1086"), val = tensor([1, 12, 64, -1])]; + tensor var_1087_cast_fp16 = reshape(shape = var_1086, x = obj_91_cast_fp16)[name = string("op_1087_cast_fp16")]; + bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; + bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1087_cast_fp16, y = obj_95_cast_fp16)[name = string("attn_19_cast_fp16")]; + tensor var_1090 = const()[name = string("op_1090"), val = tensor([1, 768, 1, -1])]; + tensor input_43_cast_fp16 = reshape(shape = var_1090, x = attn_19_cast_fp16)[name = string("input_43_cast_fp16")]; + string obj_93_pad_type_0 = const()[name = string("obj_93_pad_type_0"), val = string("valid")]; + tensor obj_93_strides_0 = const()[name = string("obj_93_strides_0"), val = tensor([1, 1])]; + tensor obj_93_pad_0 = const()[name = string("obj_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_93_dilations_0 = const()[name = string("obj_93_dilations_0"), val = tensor([1, 1])]; + int32 obj_93_groups_0 = const()[name = string("obj_93_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152430208)))]; + tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153609920)))]; + tensor obj_93_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_93_dilations_0, groups = obj_93_groups_0, pad = obj_93_pad_0, pad_type = obj_93_pad_type_0, strides = obj_93_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("obj_93_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_93_cast_fp16)[name = string("inputs_29_cast_fp16")]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_1108_to_fp16 = const()[name = string("op_1108_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1108_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611520)))]; + tensor input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153613120)))]; + fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")]; + tensor input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor([1, 1])]; + tensor input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor([1, 1])]; + int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614720)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158333376)))]; + tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")]; + tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = string("input_49_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158339584)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163058240)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor obj_107_begin_0 = const()[name = string("obj_107_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_107_end_0 = const()[name = string("obj_107_end_0"), val = tensor([6, 768, 1, 1536])]; + tensor obj_107_end_mask_0 = const()[name = string("obj_107_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_107_cast_fp16 = slice_by_index(begin = obj_107_begin_0, end = obj_107_end_0, end_mask = obj_107_end_mask_0, x = read_state_2)[name = string("obj_107_cast_fp16")]; + tensor obj_109_begin_0 = const()[name = string("obj_109_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_109_end_0 = const()[name = string("obj_109_end_0"), val = tensor([6, 768, 1, 1536])]; + tensor obj_109_end_mask_0 = const()[name = string("obj_109_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_109_cast_fp16 = slice_by_index(begin = obj_109_begin_0, end = obj_109_end_0, end_mask = obj_109_end_mask_0, x = read_state_3)[name = string("obj_109_cast_fp16")]; + int32 var_1153 = const()[name = string("op_1153"), val = int32(3)]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_1178_to_fp16 = const()[name = string("op_1178_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1178_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163059840)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163061440)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163063040)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164242752)))]; + tensor query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_21_cast_fp16")]; + string current_key_11_pad_type_0 = const()[name = string("current_key_11_pad_type_0"), val = string("valid")]; + tensor current_key_11_strides_0 = const()[name = string("current_key_11_strides_0"), val = tensor([1, 1])]; + tensor current_key_11_pad_0 = const()[name = string("current_key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_11_dilations_0 = const()[name = string("current_key_11_dilations_0"), val = tensor([1, 1])]; + int32 current_key_11_groups_0 = const()[name = string("current_key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164244352)))]; + tensor current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_key_11_cast_fp16")]; + string current_value_11_pad_type_0 = const()[name = string("current_value_11_pad_type_0"), val = string("valid")]; + tensor current_value_11_strides_0 = const()[name = string("current_value_11_strides_0"), val = tensor([1, 1])]; + tensor current_value_11_pad_0 = const()[name = string("current_value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_11_dilations_0 = const()[name = string("current_value_11_dilations_0"), val = tensor([1, 1])]; + int32 current_value_11_groups_0 = const()[name = string("current_value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165424064)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166603776)))]; + tensor current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_value_11_cast_fp16")]; + tensor var_1216_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_169_cast_fp16)[name = string("op_1216_cast_fp16")]; + tensor key_11_cast_fp16 = add(x = var_65_cast_fp16_5, y = var_1216_cast_fp16)[name = string("key_11_cast_fp16")]; + tensor var_1218_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_169_cast_fp16)[name = string("op_1218_cast_fp16")]; + tensor value_11_cast_fp16 = add(x = var_80_cast_fp16_5, y = var_1218_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_1221 = const()[name = string("op_1221"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_1221, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; + fp16 var_1223_to_fp16 = const()[name = string("op_1223_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1224_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1223_to_fp16)[name = string("op_1224_cast_fp16")]; + tensor var_1225 = const()[name = string("op_1225"), val = tensor([1, 12, 64, -1])]; + tensor var_1226_cast_fp16 = reshape(shape = var_1225, x = key_11_cast_fp16)[name = string("op_1226_cast_fp16")]; + bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; + bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1224_cast_fp16, y = var_1226_cast_fp16)[name = string("mh_w_41_cast_fp16")]; + tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_43_cast_fp16")]; + tensor var_1234_cast_fp16 = softmax(axis = var_1153, x = mh_w_43_cast_fp16)[name = string("op_1234_cast_fp16")]; + tensor var_1235 = const()[name = string("op_1235"), val = tensor([1, 12, 64, -1])]; + tensor var_1236_cast_fp16 = reshape(shape = var_1235, x = value_11_cast_fp16)[name = string("op_1236_cast_fp16")]; + bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; + bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1236_cast_fp16, y = var_1234_cast_fp16)[name = string("attn_21_cast_fp16")]; + tensor var_1239 = const()[name = string("op_1239"), val = tensor([1, 768, 1, -1])]; + tensor input_51_cast_fp16 = reshape(shape = var_1239, x = attn_21_cast_fp16)[name = string("input_51_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166605376)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167785088)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_33_cast_fp16")]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_1261_to_fp16 = const()[name = string("op_1261_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1261_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167786688)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167788288)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; + tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; + tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; + int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167789888)))]; + tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168969600)))]; + tensor query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_23_cast_fp16")]; + tensor var_1281 = const()[name = string("op_1281"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_23_cast_fp16 = reshape(shape = var_1281, x = query_23_cast_fp16)[name = string("mh_q_23_cast_fp16")]; + fp16 var_1283_to_fp16 = const()[name = string("op_1283_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1284_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1283_to_fp16)[name = string("op_1284_cast_fp16")]; + tensor var_1285 = const()[name = string("op_1285"), val = tensor([1, 12, 64, -1])]; + tensor var_1286_cast_fp16 = reshape(shape = var_1285, x = obj_107_cast_fp16)[name = string("op_1286_cast_fp16")]; + bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; + bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_1284_cast_fp16, y = var_1286_cast_fp16)[name = string("mh_w_45_cast_fp16")]; + tensor mh_w_47_cast_fp16 = add(x = mh_w_45_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_47_cast_fp16")]; + tensor obj_113_cast_fp16 = softmax(axis = var_1153, x = mh_w_47_cast_fp16)[name = string("obj_113_cast_fp16")]; + tensor var_1295 = const()[name = string("op_1295"), val = tensor([1, 12, 64, -1])]; + tensor var_1296_cast_fp16 = reshape(shape = var_1295, x = obj_109_cast_fp16)[name = string("op_1296_cast_fp16")]; + bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)]; + bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)]; + tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1296_cast_fp16, y = obj_113_cast_fp16)[name = string("attn_23_cast_fp16")]; + tensor var_1299 = const()[name = string("op_1299"), val = tensor([1, 768, 1, -1])]; + tensor input_53_cast_fp16 = reshape(shape = var_1299, x = attn_23_cast_fp16)[name = string("input_53_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168971200)))]; + tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170150912)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_1317_to_fp16 = const()[name = string("op_1317_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1317_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_55_gamma_0_to_fp16 = const()[name = string("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170152512)))]; + tensor input_55_beta_0_to_fp16 = const()[name = string("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170154112)))]; + fp16 input_55_epsilon_0_to_fp16 = const()[name = string("input_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_55_cast_fp16")]; + string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")]; + tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; + int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170155712)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174874368)))]; + tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; + string input_59_mode_0 = const()[name = string("input_59_mode_0"), val = string("EXACT")]; + tensor input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174880576)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179599232)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_59_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_37_cast_fp16")]; + tensor obj_125_begin_0 = const()[name = string("obj_125_begin_0"), val = tensor([6, 0, 0, 0])]; + tensor obj_125_end_0 = const()[name = string("obj_125_end_0"), val = tensor([7, 768, 1, 1536])]; + tensor obj_125_end_mask_0 = const()[name = string("obj_125_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_125_cast_fp16 = slice_by_index(begin = obj_125_begin_0, end = obj_125_end_0, end_mask = obj_125_end_mask_0, x = read_state_2)[name = string("obj_125_cast_fp16")]; + tensor obj_127_begin_0 = const()[name = string("obj_127_begin_0"), val = tensor([6, 0, 0, 0])]; + tensor obj_127_end_0 = const()[name = string("obj_127_end_0"), val = tensor([7, 768, 1, 1536])]; + tensor obj_127_end_mask_0 = const()[name = string("obj_127_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_127_cast_fp16 = slice_by_index(begin = obj_127_begin_0, end = obj_127_end_0, end_mask = obj_127_end_mask_0, x = read_state_3)[name = string("obj_127_cast_fp16")]; + int32 var_1362 = const()[name = string("op_1362"), val = int32(3)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_1387_to_fp16 = const()[name = string("op_1387_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1387_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_115_gamma_0_to_fp16 = const()[name = string("obj_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179600832)))]; + tensor obj_115_beta_0_to_fp16 = const()[name = string("obj_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179602432)))]; + fp16 obj_115_epsilon_0_to_fp16 = const()[name = string("obj_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_115_cast_fp16 = batch_norm(beta = obj_115_beta_0_to_fp16, epsilon = obj_115_epsilon_0_to_fp16, gamma = obj_115_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_115_cast_fp16")]; + string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; + tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; + tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; + int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179604032)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180783744)))]; + tensor query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("query_25_cast_fp16")]; + string current_key_13_pad_type_0 = const()[name = string("current_key_13_pad_type_0"), val = string("valid")]; + tensor current_key_13_strides_0 = const()[name = string("current_key_13_strides_0"), val = tensor([1, 1])]; + tensor current_key_13_pad_0 = const()[name = string("current_key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_13_dilations_0 = const()[name = string("current_key_13_dilations_0"), val = tensor([1, 1])]; + int32 current_key_13_groups_0 = const()[name = string("current_key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180785344)))]; + tensor current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("current_key_13_cast_fp16")]; + string current_value_13_pad_type_0 = const()[name = string("current_value_13_pad_type_0"), val = string("valid")]; + tensor current_value_13_strides_0 = const()[name = string("current_value_13_strides_0"), val = tensor([1, 1])]; + tensor current_value_13_pad_0 = const()[name = string("current_value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_13_dilations_0 = const()[name = string("current_value_13_dilations_0"), val = tensor([1, 1])]; + int32 current_value_13_groups_0 = const()[name = string("current_value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181965056)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183144768)))]; + tensor current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("current_value_13_cast_fp16")]; + tensor var_1425_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_169_cast_fp16)[name = string("op_1425_cast_fp16")]; + tensor key_13_cast_fp16 = add(x = var_65_cast_fp16_6, y = var_1425_cast_fp16)[name = string("key_13_cast_fp16")]; + tensor var_1427_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_169_cast_fp16)[name = string("op_1427_cast_fp16")]; + tensor value_13_cast_fp16 = add(x = var_80_cast_fp16_6, y = var_1427_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_1430 = const()[name = string("op_1430"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_25_cast_fp16 = reshape(shape = var_1430, x = query_25_cast_fp16)[name = string("mh_q_25_cast_fp16")]; + fp16 var_1432_to_fp16 = const()[name = string("op_1432_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1433_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1432_to_fp16)[name = string("op_1433_cast_fp16")]; + tensor var_1434 = const()[name = string("op_1434"), val = tensor([1, 12, 64, -1])]; + tensor var_1435_cast_fp16 = reshape(shape = var_1434, x = key_13_cast_fp16)[name = string("op_1435_cast_fp16")]; + bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)]; + bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)]; + tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_1433_cast_fp16, y = var_1435_cast_fp16)[name = string("mh_w_49_cast_fp16")]; + tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_51_cast_fp16")]; + tensor var_1443_cast_fp16 = softmax(axis = var_1362, x = mh_w_51_cast_fp16)[name = string("op_1443_cast_fp16")]; + tensor var_1444 = const()[name = string("op_1444"), val = tensor([1, 12, 64, -1])]; + tensor var_1445_cast_fp16 = reshape(shape = var_1444, x = value_13_cast_fp16)[name = string("op_1445_cast_fp16")]; + bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)]; + bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)]; + tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1445_cast_fp16, y = var_1443_cast_fp16)[name = string("attn_25_cast_fp16")]; + tensor var_1448 = const()[name = string("op_1448"), val = tensor([1, 768, 1, -1])]; + tensor input_61_cast_fp16 = reshape(shape = var_1448, x = attn_25_cast_fp16)[name = string("input_61_cast_fp16")]; + string obj_121_pad_type_0 = const()[name = string("obj_121_pad_type_0"), val = string("valid")]; + tensor obj_121_strides_0 = const()[name = string("obj_121_strides_0"), val = tensor([1, 1])]; + tensor obj_121_pad_0 = const()[name = string("obj_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_121_dilations_0 = const()[name = string("obj_121_dilations_0"), val = tensor([1, 1])]; + int32 obj_121_groups_0 = const()[name = string("obj_121_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183146368)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184326080)))]; + tensor obj_121_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_121_dilations_0, groups = obj_121_groups_0, pad = obj_121_pad_0, pad_type = obj_121_pad_type_0, strides = obj_121_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_61_cast_fp16)[name = string("obj_121_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_121_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_1470_to_fp16 = const()[name = string("op_1470_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1470_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor obj_123_gamma_0_to_fp16 = const()[name = string("obj_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184327680)))]; + tensor obj_123_beta_0_to_fp16 = const()[name = string("obj_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184329280)))]; + fp16 obj_123_epsilon_0_to_fp16 = const()[name = string("obj_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_123_cast_fp16 = batch_norm(beta = obj_123_beta_0_to_fp16, epsilon = obj_123_epsilon_0_to_fp16, gamma = obj_123_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("obj_123_cast_fp16")]; + string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; + tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; + tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; + int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; + tensor layers_6_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184330880)))]; + tensor layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185510592)))]; + tensor query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16, x = obj_123_cast_fp16)[name = string("query_27_cast_fp16")]; + tensor var_1490 = const()[name = string("op_1490"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_27_cast_fp16 = reshape(shape = var_1490, x = query_27_cast_fp16)[name = string("mh_q_27_cast_fp16")]; + fp16 var_1492_to_fp16 = const()[name = string("op_1492_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1493_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1492_to_fp16)[name = string("op_1493_cast_fp16")]; + tensor var_1494 = const()[name = string("op_1494"), val = tensor([1, 12, 64, -1])]; + tensor var_1495_cast_fp16 = reshape(shape = var_1494, x = obj_125_cast_fp16)[name = string("op_1495_cast_fp16")]; + bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)]; + bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)]; + tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_1493_cast_fp16, y = var_1495_cast_fp16)[name = string("mh_w_53_cast_fp16")]; + tensor mh_w_55_cast_fp16 = add(x = mh_w_53_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_55_cast_fp16")]; + tensor obj_131_cast_fp16 = softmax(axis = var_1362, x = mh_w_55_cast_fp16)[name = string("obj_131_cast_fp16")]; + tensor var_1504 = const()[name = string("op_1504"), val = tensor([1, 12, 64, -1])]; + tensor var_1505_cast_fp16 = reshape(shape = var_1504, x = obj_127_cast_fp16)[name = string("op_1505_cast_fp16")]; + bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)]; + bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)]; + tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1505_cast_fp16, y = obj_131_cast_fp16)[name = string("attn_27_cast_fp16")]; + tensor var_1508 = const()[name = string("op_1508"), val = tensor([1, 768, 1, -1])]; + tensor input_63_cast_fp16 = reshape(shape = var_1508, x = attn_27_cast_fp16)[name = string("input_63_cast_fp16")]; + string obj_129_pad_type_0 = const()[name = string("obj_129_pad_type_0"), val = string("valid")]; + tensor obj_129_strides_0 = const()[name = string("obj_129_strides_0"), val = tensor([1, 1])]; + tensor obj_129_pad_0 = const()[name = string("obj_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_129_dilations_0 = const()[name = string("obj_129_dilations_0"), val = tensor([1, 1])]; + int32 obj_129_groups_0 = const()[name = string("obj_129_groups_0"), val = int32(1)]; + tensor layers_6_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185512192)))]; + tensor layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186691904)))]; + tensor obj_129_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_129_dilations_0, groups = obj_129_groups_0, pad = obj_129_pad_0, pad_type = obj_129_pad_type_0, strides = obj_129_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("obj_129_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_129_cast_fp16)[name = string("inputs_41_cast_fp16")]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_1529_to_fp16 = const()[name = string("op_1529_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1529_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor input_65_gamma_0_to_fp16 = const()[name = string("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186693504)))]; + tensor input_65_beta_0_to_fp16 = const()[name = string("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186695104)))]; + fp16 input_65_epsilon_0_to_fp16 = const()[name = string("input_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("input_65_cast_fp16")]; + string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")]; + tensor input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor([1, 1])]; + tensor input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor([1, 1])]; + int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186696704)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191415360)))]; + tensor input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_mode_0 = const()[name = string("input_69_mode_0"), val = string("EXACT")]; + tensor input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191421568)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196140224)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_69_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor obj_143_begin_0 = const()[name = string("obj_143_begin_0"), val = tensor([7, 0, 0, 0])]; + tensor obj_143_end_0 = const()[name = string("obj_143_end_0"), val = tensor([8, 768, 1, 1536])]; + tensor obj_143_end_mask_0 = const()[name = string("obj_143_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_143_cast_fp16 = slice_by_index(begin = obj_143_begin_0, end = obj_143_end_0, end_mask = obj_143_end_mask_0, x = read_state_2)[name = string("obj_143_cast_fp16")]; + tensor obj_145_begin_0 = const()[name = string("obj_145_begin_0"), val = tensor([7, 0, 0, 0])]; + tensor obj_145_end_0 = const()[name = string("obj_145_end_0"), val = tensor([8, 768, 1, 1536])]; + tensor obj_145_end_mask_0 = const()[name = string("obj_145_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_145_cast_fp16 = slice_by_index(begin = obj_145_begin_0, end = obj_145_end_0, end_mask = obj_145_end_mask_0, x = read_state_3)[name = string("obj_145_cast_fp16")]; + int32 var_1575 = const()[name = string("op_1575"), val = int32(3)]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_1600_to_fp16 = const()[name = string("op_1600_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1600_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor obj_133_gamma_0_to_fp16 = const()[name = string("obj_133_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196141824)))]; + tensor obj_133_beta_0_to_fp16 = const()[name = string("obj_133_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196143424)))]; + fp16 obj_133_epsilon_0_to_fp16 = const()[name = string("obj_133_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_133_cast_fp16 = batch_norm(beta = obj_133_beta_0_to_fp16, epsilon = obj_133_epsilon_0_to_fp16, gamma = obj_133_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("obj_133_cast_fp16")]; + string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; + tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; + tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; + int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196145024)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197324736)))]; + tensor query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("query_29_cast_fp16")]; + string current_key_15_pad_type_0 = const()[name = string("current_key_15_pad_type_0"), val = string("valid")]; + tensor current_key_15_strides_0 = const()[name = string("current_key_15_strides_0"), val = tensor([1, 1])]; + tensor current_key_15_pad_0 = const()[name = string("current_key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_15_dilations_0 = const()[name = string("current_key_15_dilations_0"), val = tensor([1, 1])]; + int32 current_key_15_groups_0 = const()[name = string("current_key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197326336)))]; + tensor current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("current_key_15_cast_fp16")]; + string current_value_15_pad_type_0 = const()[name = string("current_value_15_pad_type_0"), val = string("valid")]; + tensor current_value_15_strides_0 = const()[name = string("current_value_15_strides_0"), val = tensor([1, 1])]; + tensor current_value_15_pad_0 = const()[name = string("current_value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_15_dilations_0 = const()[name = string("current_value_15_dilations_0"), val = tensor([1, 1])]; + int32 current_value_15_groups_0 = const()[name = string("current_value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198506048)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199685760)))]; + tensor current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("current_value_15_cast_fp16")]; + tensor var_1638_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_169_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor key_15_cast_fp16 = add(x = var_65_cast_fp16_7, y = var_1638_cast_fp16)[name = string("key_15_cast_fp16")]; + tensor var_1640_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_169_cast_fp16)[name = string("op_1640_cast_fp16")]; + tensor value_15_cast_fp16 = add(x = var_80_cast_fp16_7, y = var_1640_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_1643 = const()[name = string("op_1643"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_29_cast_fp16 = reshape(shape = var_1643, x = query_29_cast_fp16)[name = string("mh_q_29_cast_fp16")]; + fp16 var_1645_to_fp16 = const()[name = string("op_1645_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1646_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1645_to_fp16)[name = string("op_1646_cast_fp16")]; + tensor var_1647 = const()[name = string("op_1647"), val = tensor([1, 12, 64, -1])]; + tensor var_1648_cast_fp16 = reshape(shape = var_1647, x = key_15_cast_fp16)[name = string("op_1648_cast_fp16")]; + bool mh_w_57_transpose_x_0 = const()[name = string("mh_w_57_transpose_x_0"), val = bool(true)]; + bool mh_w_57_transpose_y_0 = const()[name = string("mh_w_57_transpose_y_0"), val = bool(false)]; + tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_1646_cast_fp16, y = var_1648_cast_fp16)[name = string("mh_w_57_cast_fp16")]; + tensor mh_w_59_cast_fp16 = add(x = mh_w_57_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_59_cast_fp16")]; + tensor var_1656_cast_fp16 = softmax(axis = var_1575, x = mh_w_59_cast_fp16)[name = string("op_1656_cast_fp16")]; + tensor var_1657 = const()[name = string("op_1657"), val = tensor([1, 12, 64, -1])]; + tensor var_1658_cast_fp16 = reshape(shape = var_1657, x = value_15_cast_fp16)[name = string("op_1658_cast_fp16")]; + bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)]; + bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)]; + tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1658_cast_fp16, y = var_1656_cast_fp16)[name = string("attn_29_cast_fp16")]; + tensor var_1661 = const()[name = string("op_1661"), val = tensor([1, 768, 1, -1])]; + tensor input_71_cast_fp16 = reshape(shape = var_1661, x = attn_29_cast_fp16)[name = string("input_71_cast_fp16")]; + string obj_139_pad_type_0 = const()[name = string("obj_139_pad_type_0"), val = string("valid")]; + tensor obj_139_strides_0 = const()[name = string("obj_139_strides_0"), val = tensor([1, 1])]; + tensor obj_139_pad_0 = const()[name = string("obj_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_139_dilations_0 = const()[name = string("obj_139_dilations_0"), val = tensor([1, 1])]; + int32 obj_139_groups_0 = const()[name = string("obj_139_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199687360)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200867072)))]; + tensor obj_139_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_139_dilations_0, groups = obj_139_groups_0, pad = obj_139_pad_0, pad_type = obj_139_pad_type_0, strides = obj_139_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("obj_139_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_139_cast_fp16)[name = string("inputs_45_cast_fp16")]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_1683_to_fp16 = const()[name = string("op_1683_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1683_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_141_gamma_0_to_fp16 = const()[name = string("obj_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200868672)))]; + tensor obj_141_beta_0_to_fp16 = const()[name = string("obj_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200870272)))]; + fp16 obj_141_epsilon_0_to_fp16 = const()[name = string("obj_141_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_141_cast_fp16")]; + string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; + tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; + tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; + int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; + tensor layers_7_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200871872)))]; + tensor layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202051584)))]; + tensor query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = string("query_31_cast_fp16")]; + tensor var_1703 = const()[name = string("op_1703"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_31_cast_fp16 = reshape(shape = var_1703, x = query_31_cast_fp16)[name = string("mh_q_31_cast_fp16")]; + fp16 var_1705_to_fp16 = const()[name = string("op_1705_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1706_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1705_to_fp16)[name = string("op_1706_cast_fp16")]; + tensor var_1707 = const()[name = string("op_1707"), val = tensor([1, 12, 64, -1])]; + tensor var_1708_cast_fp16 = reshape(shape = var_1707, x = obj_143_cast_fp16)[name = string("op_1708_cast_fp16")]; + bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)]; + bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)]; + tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_1706_cast_fp16, y = var_1708_cast_fp16)[name = string("mh_w_61_cast_fp16")]; + tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_63_cast_fp16")]; + tensor obj_149_cast_fp16 = softmax(axis = var_1575, x = mh_w_63_cast_fp16)[name = string("obj_149_cast_fp16")]; + tensor var_1717 = const()[name = string("op_1717"), val = tensor([1, 12, 64, -1])]; + tensor var_1718_cast_fp16 = reshape(shape = var_1717, x = obj_145_cast_fp16)[name = string("op_1718_cast_fp16")]; + bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)]; + bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)]; + tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1718_cast_fp16, y = obj_149_cast_fp16)[name = string("attn_31_cast_fp16")]; + tensor var_1721 = const()[name = string("op_1721"), val = tensor([1, 768, 1, -1])]; + tensor input_73_cast_fp16 = reshape(shape = var_1721, x = attn_31_cast_fp16)[name = string("input_73_cast_fp16")]; + string obj_147_pad_type_0 = const()[name = string("obj_147_pad_type_0"), val = string("valid")]; + tensor obj_147_strides_0 = const()[name = string("obj_147_strides_0"), val = tensor([1, 1])]; + tensor obj_147_pad_0 = const()[name = string("obj_147_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_147_dilations_0 = const()[name = string("obj_147_dilations_0"), val = tensor([1, 1])]; + int32 obj_147_groups_0 = const()[name = string("obj_147_groups_0"), val = int32(1)]; + tensor layers_7_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202053184)))]; + tensor layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203232896)))]; + tensor obj_147_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_147_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_147_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_1742_to_fp16 = const()[name = string("op_1742_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1742_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203234496)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203236096)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203237696)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207956352)))]; + tensor input_77_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207962560)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212681216)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_49_cast_fp16")]; + tensor obj_161_begin_0 = const()[name = string("obj_161_begin_0"), val = tensor([8, 0, 0, 0])]; + tensor obj_161_end_0 = const()[name = string("obj_161_end_0"), val = tensor([9, 768, 1, 1536])]; + tensor obj_161_end_mask_0 = const()[name = string("obj_161_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_161_cast_fp16 = slice_by_index(begin = obj_161_begin_0, end = obj_161_end_0, end_mask = obj_161_end_mask_0, x = read_state_2)[name = string("obj_161_cast_fp16")]; + tensor obj_163_begin_0 = const()[name = string("obj_163_begin_0"), val = tensor([8, 0, 0, 0])]; + tensor obj_163_end_0 = const()[name = string("obj_163_end_0"), val = tensor([9, 768, 1, 1536])]; + tensor obj_163_end_mask_0 = const()[name = string("obj_163_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_163_cast_fp16 = slice_by_index(begin = obj_163_begin_0, end = obj_163_end_0, end_mask = obj_163_end_mask_0, x = read_state_3)[name = string("obj_163_cast_fp16")]; + int32 var_1788 = const()[name = string("op_1788"), val = int32(3)]; + tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; + fp16 var_1813_to_fp16 = const()[name = string("op_1813_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1813_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; + tensor obj_151_gamma_0_to_fp16 = const()[name = string("obj_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212682816)))]; + tensor obj_151_beta_0_to_fp16 = const()[name = string("obj_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212684416)))]; + fp16 obj_151_epsilon_0_to_fp16 = const()[name = string("obj_151_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_151_cast_fp16 = batch_norm(beta = obj_151_beta_0_to_fp16, epsilon = obj_151_epsilon_0_to_fp16, gamma = obj_151_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_151_cast_fp16")]; + string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; + tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; + tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; + int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212686016)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213865728)))]; + tensor query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("query_33_cast_fp16")]; + string current_key_17_pad_type_0 = const()[name = string("current_key_17_pad_type_0"), val = string("valid")]; + tensor current_key_17_strides_0 = const()[name = string("current_key_17_strides_0"), val = tensor([1, 1])]; + tensor current_key_17_pad_0 = const()[name = string("current_key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_17_dilations_0 = const()[name = string("current_key_17_dilations_0"), val = tensor([1, 1])]; + int32 current_key_17_groups_0 = const()[name = string("current_key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213867328)))]; + tensor current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("current_key_17_cast_fp16")]; + string current_value_17_pad_type_0 = const()[name = string("current_value_17_pad_type_0"), val = string("valid")]; + tensor current_value_17_strides_0 = const()[name = string("current_value_17_strides_0"), val = tensor([1, 1])]; + tensor current_value_17_pad_0 = const()[name = string("current_value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_17_dilations_0 = const()[name = string("current_value_17_dilations_0"), val = tensor([1, 1])]; + int32 current_value_17_groups_0 = const()[name = string("current_value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215047040)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216226752)))]; + tensor current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("current_value_17_cast_fp16")]; + tensor var_1851_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_169_cast_fp16)[name = string("op_1851_cast_fp16")]; + tensor key_17_cast_fp16 = add(x = var_65_cast_fp16_8, y = var_1851_cast_fp16)[name = string("key_17_cast_fp16")]; + tensor var_1853_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_169_cast_fp16)[name = string("op_1853_cast_fp16")]; + tensor value_17_cast_fp16 = add(x = var_80_cast_fp16_8, y = var_1853_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_1856 = const()[name = string("op_1856"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_33_cast_fp16 = reshape(shape = var_1856, x = query_33_cast_fp16)[name = string("mh_q_33_cast_fp16")]; + fp16 var_1858_to_fp16 = const()[name = string("op_1858_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1859_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_1858_to_fp16)[name = string("op_1859_cast_fp16")]; + tensor var_1860 = const()[name = string("op_1860"), val = tensor([1, 12, 64, -1])]; + tensor var_1861_cast_fp16 = reshape(shape = var_1860, x = key_17_cast_fp16)[name = string("op_1861_cast_fp16")]; + bool mh_w_65_transpose_x_0 = const()[name = string("mh_w_65_transpose_x_0"), val = bool(true)]; + bool mh_w_65_transpose_y_0 = const()[name = string("mh_w_65_transpose_y_0"), val = bool(false)]; + tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_1859_cast_fp16, y = var_1861_cast_fp16)[name = string("mh_w_65_cast_fp16")]; + tensor mh_w_67_cast_fp16 = add(x = mh_w_65_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_67_cast_fp16")]; + tensor var_1869_cast_fp16 = softmax(axis = var_1788, x = mh_w_67_cast_fp16)[name = string("op_1869_cast_fp16")]; + tensor var_1870 = const()[name = string("op_1870"), val = tensor([1, 12, 64, -1])]; + tensor var_1871_cast_fp16 = reshape(shape = var_1870, x = value_17_cast_fp16)[name = string("op_1871_cast_fp16")]; + bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)]; + bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)]; + tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_1871_cast_fp16, y = var_1869_cast_fp16)[name = string("attn_33_cast_fp16")]; + tensor var_1874 = const()[name = string("op_1874"), val = tensor([1, 768, 1, -1])]; + tensor input_81_cast_fp16 = reshape(shape = var_1874, x = attn_33_cast_fp16)[name = string("input_81_cast_fp16")]; + string obj_157_pad_type_0 = const()[name = string("obj_157_pad_type_0"), val = string("valid")]; + tensor obj_157_strides_0 = const()[name = string("obj_157_strides_0"), val = tensor([1, 1])]; + tensor obj_157_pad_0 = const()[name = string("obj_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_157_dilations_0 = const()[name = string("obj_157_dilations_0"), val = tensor([1, 1])]; + int32 obj_157_groups_0 = const()[name = string("obj_157_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216228352)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217408064)))]; + tensor obj_157_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_157_dilations_0, groups = obj_157_groups_0, pad = obj_157_pad_0, pad_type = obj_157_pad_type_0, strides = obj_157_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_157_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_157_cast_fp16)[name = string("inputs_51_cast_fp16")]; + tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; + fp16 var_1896_to_fp16 = const()[name = string("op_1896_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1896_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; + tensor obj_159_gamma_0_to_fp16 = const()[name = string("obj_159_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217409664)))]; + tensor obj_159_beta_0_to_fp16 = const()[name = string("obj_159_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217411264)))]; + fp16 obj_159_epsilon_0_to_fp16 = const()[name = string("obj_159_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_159_cast_fp16 = batch_norm(beta = obj_159_beta_0_to_fp16, epsilon = obj_159_epsilon_0_to_fp16, gamma = obj_159_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("obj_159_cast_fp16")]; + string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; + tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; + tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; + int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; + tensor layers_8_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217412864)))]; + tensor layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218592576)))]; + tensor query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16, x = obj_159_cast_fp16)[name = string("query_35_cast_fp16")]; + tensor var_1916 = const()[name = string("op_1916"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_35_cast_fp16 = reshape(shape = var_1916, x = query_35_cast_fp16)[name = string("mh_q_35_cast_fp16")]; + fp16 var_1918_to_fp16 = const()[name = string("op_1918_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1919_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_1918_to_fp16)[name = string("op_1919_cast_fp16")]; + tensor var_1920 = const()[name = string("op_1920"), val = tensor([1, 12, 64, -1])]; + tensor var_1921_cast_fp16 = reshape(shape = var_1920, x = obj_161_cast_fp16)[name = string("op_1921_cast_fp16")]; + bool mh_w_69_transpose_x_0 = const()[name = string("mh_w_69_transpose_x_0"), val = bool(true)]; + bool mh_w_69_transpose_y_0 = const()[name = string("mh_w_69_transpose_y_0"), val = bool(false)]; + tensor mh_w_69_cast_fp16 = matmul(transpose_x = mh_w_69_transpose_x_0, transpose_y = mh_w_69_transpose_y_0, x = var_1919_cast_fp16, y = var_1921_cast_fp16)[name = string("mh_w_69_cast_fp16")]; + tensor mh_w_71_cast_fp16 = add(x = mh_w_69_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_71_cast_fp16")]; + tensor obj_167_cast_fp16 = softmax(axis = var_1788, x = mh_w_71_cast_fp16)[name = string("obj_167_cast_fp16")]; + tensor var_1930 = const()[name = string("op_1930"), val = tensor([1, 12, 64, -1])]; + tensor var_1931_cast_fp16 = reshape(shape = var_1930, x = obj_163_cast_fp16)[name = string("op_1931_cast_fp16")]; + bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)]; + bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)]; + tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_1931_cast_fp16, y = obj_167_cast_fp16)[name = string("attn_35_cast_fp16")]; + tensor var_1934 = const()[name = string("op_1934"), val = tensor([1, 768, 1, -1])]; + tensor input_83_cast_fp16 = reshape(shape = var_1934, x = attn_35_cast_fp16)[name = string("input_83_cast_fp16")]; + string obj_165_pad_type_0 = const()[name = string("obj_165_pad_type_0"), val = string("valid")]; + tensor obj_165_strides_0 = const()[name = string("obj_165_strides_0"), val = tensor([1, 1])]; + tensor obj_165_pad_0 = const()[name = string("obj_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_165_dilations_0 = const()[name = string("obj_165_dilations_0"), val = tensor([1, 1])]; + int32 obj_165_groups_0 = const()[name = string("obj_165_groups_0"), val = int32(1)]; + tensor layers_8_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218594176)))]; + tensor layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219773888)))]; + tensor obj_165_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("obj_165_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_165_cast_fp16)[name = string("inputs_53_cast_fp16")]; + tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; + fp16 var_1955_to_fp16 = const()[name = string("op_1955_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_1955_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; + tensor input_85_gamma_0_to_fp16 = const()[name = string("input_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219775488)))]; + tensor input_85_beta_0_to_fp16 = const()[name = string("input_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219777088)))]; + fp16 input_85_epsilon_0_to_fp16 = const()[name = string("input_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_pad_type_0 = const()[name = string("input_87_pad_type_0"), val = string("valid")]; + tensor input_87_strides_0 = const()[name = string("input_87_strides_0"), val = tensor([1, 1])]; + tensor input_87_pad_0 = const()[name = string("input_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_87_dilations_0 = const()[name = string("input_87_dilations_0"), val = tensor([1, 1])]; + int32 input_87_groups_0 = const()[name = string("input_87_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219778688)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224497344)))]; + tensor input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string input_89_mode_0 = const()[name = string("input_89_mode_0"), val = string("EXACT")]; + tensor input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = string("input_89_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224503552)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229222208)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_89_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_55_cast_fp16")]; + tensor obj_179_begin_0 = const()[name = string("obj_179_begin_0"), val = tensor([9, 0, 0, 0])]; + tensor obj_179_end_0 = const()[name = string("obj_179_end_0"), val = tensor([10, 768, 1, 1536])]; + tensor obj_179_end_mask_0 = const()[name = string("obj_179_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_179_cast_fp16 = slice_by_index(begin = obj_179_begin_0, end = obj_179_end_0, end_mask = obj_179_end_mask_0, x = read_state_2)[name = string("obj_179_cast_fp16")]; + tensor obj_181_begin_0 = const()[name = string("obj_181_begin_0"), val = tensor([9, 0, 0, 0])]; + tensor obj_181_end_0 = const()[name = string("obj_181_end_0"), val = tensor([10, 768, 1, 1536])]; + tensor obj_181_end_mask_0 = const()[name = string("obj_181_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_181_cast_fp16 = slice_by_index(begin = obj_181_begin_0, end = obj_181_end_0, end_mask = obj_181_end_mask_0, x = read_state_3)[name = string("obj_181_cast_fp16")]; + int32 var_2001 = const()[name = string("op_2001"), val = int32(3)]; + tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; + fp16 var_2026_to_fp16 = const()[name = string("op_2026_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2026_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; + tensor obj_169_gamma_0_to_fp16 = const()[name = string("obj_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229223808)))]; + tensor obj_169_beta_0_to_fp16 = const()[name = string("obj_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229225408)))]; + fp16 obj_169_epsilon_0_to_fp16 = const()[name = string("obj_169_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_169_cast_fp16 = batch_norm(beta = obj_169_beta_0_to_fp16, epsilon = obj_169_epsilon_0_to_fp16, gamma = obj_169_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("obj_169_cast_fp16")]; + string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; + tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; + tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; + int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229227008)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230406720)))]; + tensor query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("query_37_cast_fp16")]; + string current_key_19_pad_type_0 = const()[name = string("current_key_19_pad_type_0"), val = string("valid")]; + tensor current_key_19_strides_0 = const()[name = string("current_key_19_strides_0"), val = tensor([1, 1])]; + tensor current_key_19_pad_0 = const()[name = string("current_key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_19_dilations_0 = const()[name = string("current_key_19_dilations_0"), val = tensor([1, 1])]; + int32 current_key_19_groups_0 = const()[name = string("current_key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230408320)))]; + tensor current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("current_key_19_cast_fp16")]; + string current_value_19_pad_type_0 = const()[name = string("current_value_19_pad_type_0"), val = string("valid")]; + tensor current_value_19_strides_0 = const()[name = string("current_value_19_strides_0"), val = tensor([1, 1])]; + tensor current_value_19_pad_0 = const()[name = string("current_value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_19_dilations_0 = const()[name = string("current_value_19_dilations_0"), val = tensor([1, 1])]; + int32 current_value_19_groups_0 = const()[name = string("current_value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231588032)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232767744)))]; + tensor current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("current_value_19_cast_fp16")]; + tensor var_2064_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_169_cast_fp16)[name = string("op_2064_cast_fp16")]; + tensor key_19_cast_fp16 = add(x = var_65_cast_fp16_9, y = var_2064_cast_fp16)[name = string("key_19_cast_fp16")]; + tensor var_2066_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_169_cast_fp16)[name = string("op_2066_cast_fp16")]; + tensor value_19_cast_fp16 = add(x = var_80_cast_fp16_9, y = var_2066_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_2069 = const()[name = string("op_2069"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_37_cast_fp16 = reshape(shape = var_2069, x = query_37_cast_fp16)[name = string("mh_q_37_cast_fp16")]; + fp16 var_2071_to_fp16 = const()[name = string("op_2071_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2072_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2071_to_fp16)[name = string("op_2072_cast_fp16")]; + tensor var_2073 = const()[name = string("op_2073"), val = tensor([1, 12, 64, -1])]; + tensor var_2074_cast_fp16 = reshape(shape = var_2073, x = key_19_cast_fp16)[name = string("op_2074_cast_fp16")]; + bool mh_w_73_transpose_x_0 = const()[name = string("mh_w_73_transpose_x_0"), val = bool(true)]; + bool mh_w_73_transpose_y_0 = const()[name = string("mh_w_73_transpose_y_0"), val = bool(false)]; + tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_2072_cast_fp16, y = var_2074_cast_fp16)[name = string("mh_w_73_cast_fp16")]; + tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_75_cast_fp16")]; + tensor var_2082_cast_fp16 = softmax(axis = var_2001, x = mh_w_75_cast_fp16)[name = string("op_2082_cast_fp16")]; + tensor var_2083 = const()[name = string("op_2083"), val = tensor([1, 12, 64, -1])]; + tensor var_2084_cast_fp16 = reshape(shape = var_2083, x = value_19_cast_fp16)[name = string("op_2084_cast_fp16")]; + bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)]; + bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)]; + tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2084_cast_fp16, y = var_2082_cast_fp16)[name = string("attn_37_cast_fp16")]; + tensor var_2087 = const()[name = string("op_2087"), val = tensor([1, 768, 1, -1])]; + tensor input_91_cast_fp16 = reshape(shape = var_2087, x = attn_37_cast_fp16)[name = string("input_91_cast_fp16")]; + string obj_175_pad_type_0 = const()[name = string("obj_175_pad_type_0"), val = string("valid")]; + tensor obj_175_strides_0 = const()[name = string("obj_175_strides_0"), val = tensor([1, 1])]; + tensor obj_175_pad_0 = const()[name = string("obj_175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_175_dilations_0 = const()[name = string("obj_175_dilations_0"), val = tensor([1, 1])]; + int32 obj_175_groups_0 = const()[name = string("obj_175_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232769344)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233949056)))]; + tensor obj_175_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_175_dilations_0, groups = obj_175_groups_0, pad = obj_175_pad_0, pad_type = obj_175_pad_type_0, strides = obj_175_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = string("obj_175_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_175_cast_fp16)[name = string("inputs_57_cast_fp16")]; + tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; + fp16 var_2109_to_fp16 = const()[name = string("op_2109_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2109_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; + tensor obj_177_gamma_0_to_fp16 = const()[name = string("obj_177_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233950656)))]; + tensor obj_177_beta_0_to_fp16 = const()[name = string("obj_177_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233952256)))]; + fp16 obj_177_epsilon_0_to_fp16 = const()[name = string("obj_177_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_177_cast_fp16 = batch_norm(beta = obj_177_beta_0_to_fp16, epsilon = obj_177_epsilon_0_to_fp16, gamma = obj_177_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_177_cast_fp16")]; + string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; + tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; + tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; + int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; + tensor layers_9_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233953856)))]; + tensor layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235133568)))]; + tensor query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16, x = obj_177_cast_fp16)[name = string("query_39_cast_fp16")]; + tensor var_2129 = const()[name = string("op_2129"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_39_cast_fp16 = reshape(shape = var_2129, x = query_39_cast_fp16)[name = string("mh_q_39_cast_fp16")]; + fp16 var_2131_to_fp16 = const()[name = string("op_2131_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2132_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2131_to_fp16)[name = string("op_2132_cast_fp16")]; + tensor var_2133 = const()[name = string("op_2133"), val = tensor([1, 12, 64, -1])]; + tensor var_2134_cast_fp16 = reshape(shape = var_2133, x = obj_179_cast_fp16)[name = string("op_2134_cast_fp16")]; + bool mh_w_77_transpose_x_0 = const()[name = string("mh_w_77_transpose_x_0"), val = bool(true)]; + bool mh_w_77_transpose_y_0 = const()[name = string("mh_w_77_transpose_y_0"), val = bool(false)]; + tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_2132_cast_fp16, y = var_2134_cast_fp16)[name = string("mh_w_77_cast_fp16")]; + tensor mh_w_79_cast_fp16 = add(x = mh_w_77_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_79_cast_fp16")]; + tensor obj_185_cast_fp16 = softmax(axis = var_2001, x = mh_w_79_cast_fp16)[name = string("obj_185_cast_fp16")]; + tensor var_2143 = const()[name = string("op_2143"), val = tensor([1, 12, 64, -1])]; + tensor var_2144_cast_fp16 = reshape(shape = var_2143, x = obj_181_cast_fp16)[name = string("op_2144_cast_fp16")]; + bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)]; + bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)]; + tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2144_cast_fp16, y = obj_185_cast_fp16)[name = string("attn_39_cast_fp16")]; + tensor var_2147 = const()[name = string("op_2147"), val = tensor([1, 768, 1, -1])]; + tensor input_93_cast_fp16 = reshape(shape = var_2147, x = attn_39_cast_fp16)[name = string("input_93_cast_fp16")]; + string obj_183_pad_type_0 = const()[name = string("obj_183_pad_type_0"), val = string("valid")]; + tensor obj_183_strides_0 = const()[name = string("obj_183_strides_0"), val = tensor([1, 1])]; + tensor obj_183_pad_0 = const()[name = string("obj_183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_183_dilations_0 = const()[name = string("obj_183_dilations_0"), val = tensor([1, 1])]; + int32 obj_183_groups_0 = const()[name = string("obj_183_groups_0"), val = int32(1)]; + tensor layers_9_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235135168)))]; + tensor layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236314880)))]; + tensor obj_183_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_183_dilations_0, groups = obj_183_groups_0, pad = obj_183_pad_0, pad_type = obj_183_pad_type_0, strides = obj_183_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("obj_183_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_183_cast_fp16)[name = string("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; + fp16 var_2168_to_fp16 = const()[name = string("op_2168_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2168_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; + tensor input_95_gamma_0_to_fp16 = const()[name = string("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236316480)))]; + tensor input_95_beta_0_to_fp16 = const()[name = string("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236318080)))]; + fp16 input_95_epsilon_0_to_fp16 = const()[name = string("input_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_95_cast_fp16")]; + string input_97_pad_type_0 = const()[name = string("input_97_pad_type_0"), val = string("valid")]; + tensor input_97_strides_0 = const()[name = string("input_97_strides_0"), val = tensor([1, 1])]; + tensor input_97_pad_0 = const()[name = string("input_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_97_dilations_0 = const()[name = string("input_97_dilations_0"), val = tensor([1, 1])]; + int32 input_97_groups_0 = const()[name = string("input_97_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236319680)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241038336)))]; + tensor input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_95_cast_fp16)[name = string("input_97_cast_fp16")]; + string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("EXACT")]; + tensor input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241044544)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245763200)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_99_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_61_cast_fp16")]; + tensor obj_197_begin_0 = const()[name = string("obj_197_begin_0"), val = tensor([10, 0, 0, 0])]; + tensor obj_197_end_0 = const()[name = string("obj_197_end_0"), val = tensor([11, 768, 1, 1536])]; + tensor obj_197_end_mask_0 = const()[name = string("obj_197_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_197_cast_fp16 = slice_by_index(begin = obj_197_begin_0, end = obj_197_end_0, end_mask = obj_197_end_mask_0, x = read_state_2)[name = string("obj_197_cast_fp16")]; + tensor obj_199_begin_0 = const()[name = string("obj_199_begin_0"), val = tensor([10, 0, 0, 0])]; + tensor obj_199_end_0 = const()[name = string("obj_199_end_0"), val = tensor([11, 768, 1, 1536])]; + tensor obj_199_end_mask_0 = const()[name = string("obj_199_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_199_cast_fp16 = slice_by_index(begin = obj_199_begin_0, end = obj_199_end_0, end_mask = obj_199_end_mask_0, x = read_state_3)[name = string("obj_199_cast_fp16")]; + int32 var_2214 = const()[name = string("op_2214"), val = int32(3)]; + tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; + fp16 var_2239_to_fp16 = const()[name = string("op_2239_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2239_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; + tensor obj_187_gamma_0_to_fp16 = const()[name = string("obj_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245764800)))]; + tensor obj_187_beta_0_to_fp16 = const()[name = string("obj_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245766400)))]; + fp16 obj_187_epsilon_0_to_fp16 = const()[name = string("obj_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_187_cast_fp16 = batch_norm(beta = obj_187_beta_0_to_fp16, epsilon = obj_187_epsilon_0_to_fp16, gamma = obj_187_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_187_cast_fp16")]; + string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; + tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; + tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; + int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245768000)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246947712)))]; + tensor query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("query_41_cast_fp16")]; + string current_key_21_pad_type_0 = const()[name = string("current_key_21_pad_type_0"), val = string("valid")]; + tensor current_key_21_strides_0 = const()[name = string("current_key_21_strides_0"), val = tensor([1, 1])]; + tensor current_key_21_pad_0 = const()[name = string("current_key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_21_dilations_0 = const()[name = string("current_key_21_dilations_0"), val = tensor([1, 1])]; + int32 current_key_21_groups_0 = const()[name = string("current_key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246949312)))]; + tensor current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("current_key_21_cast_fp16")]; + string current_value_21_pad_type_0 = const()[name = string("current_value_21_pad_type_0"), val = string("valid")]; + tensor current_value_21_strides_0 = const()[name = string("current_value_21_strides_0"), val = tensor([1, 1])]; + tensor current_value_21_pad_0 = const()[name = string("current_value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_21_dilations_0 = const()[name = string("current_value_21_dilations_0"), val = tensor([1, 1])]; + int32 current_value_21_groups_0 = const()[name = string("current_value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248129024)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249308736)))]; + tensor current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("current_value_21_cast_fp16")]; + tensor var_2277_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_169_cast_fp16)[name = string("op_2277_cast_fp16")]; + tensor key_21_cast_fp16 = add(x = var_65_cast_fp16_10, y = var_2277_cast_fp16)[name = string("key_21_cast_fp16")]; + tensor var_2279_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_169_cast_fp16)[name = string("op_2279_cast_fp16")]; + tensor value_21_cast_fp16 = add(x = var_80_cast_fp16_10, y = var_2279_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_2282 = const()[name = string("op_2282"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_41_cast_fp16 = reshape(shape = var_2282, x = query_41_cast_fp16)[name = string("mh_q_41_cast_fp16")]; + fp16 var_2284_to_fp16 = const()[name = string("op_2284_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2285_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2284_to_fp16)[name = string("op_2285_cast_fp16")]; + tensor var_2286 = const()[name = string("op_2286"), val = tensor([1, 12, 64, -1])]; + tensor var_2287_cast_fp16 = reshape(shape = var_2286, x = key_21_cast_fp16)[name = string("op_2287_cast_fp16")]; + bool mh_w_81_transpose_x_0 = const()[name = string("mh_w_81_transpose_x_0"), val = bool(true)]; + bool mh_w_81_transpose_y_0 = const()[name = string("mh_w_81_transpose_y_0"), val = bool(false)]; + tensor mh_w_81_cast_fp16 = matmul(transpose_x = mh_w_81_transpose_x_0, transpose_y = mh_w_81_transpose_y_0, x = var_2285_cast_fp16, y = var_2287_cast_fp16)[name = string("mh_w_81_cast_fp16")]; + tensor mh_w_83_cast_fp16 = add(x = mh_w_81_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_83_cast_fp16")]; + tensor var_2295_cast_fp16 = softmax(axis = var_2214, x = mh_w_83_cast_fp16)[name = string("op_2295_cast_fp16")]; + tensor var_2296 = const()[name = string("op_2296"), val = tensor([1, 12, 64, -1])]; + tensor var_2297_cast_fp16 = reshape(shape = var_2296, x = value_21_cast_fp16)[name = string("op_2297_cast_fp16")]; + bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)]; + bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)]; + tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2297_cast_fp16, y = var_2295_cast_fp16)[name = string("attn_41_cast_fp16")]; + tensor var_2300 = const()[name = string("op_2300"), val = tensor([1, 768, 1, -1])]; + tensor input_101_cast_fp16 = reshape(shape = var_2300, x = attn_41_cast_fp16)[name = string("input_101_cast_fp16")]; + string obj_193_pad_type_0 = const()[name = string("obj_193_pad_type_0"), val = string("valid")]; + tensor obj_193_strides_0 = const()[name = string("obj_193_strides_0"), val = tensor([1, 1])]; + tensor obj_193_pad_0 = const()[name = string("obj_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_193_dilations_0 = const()[name = string("obj_193_dilations_0"), val = tensor([1, 1])]; + int32 obj_193_groups_0 = const()[name = string("obj_193_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249310336)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250490048)))]; + tensor obj_193_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_193_dilations_0, groups = obj_193_groups_0, pad = obj_193_pad_0, pad_type = obj_193_pad_type_0, strides = obj_193_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_101_cast_fp16)[name = string("obj_193_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_193_cast_fp16)[name = string("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; + fp16 var_2322_to_fp16 = const()[name = string("op_2322_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2322_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; + tensor obj_195_gamma_0_to_fp16 = const()[name = string("obj_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250491648)))]; + tensor obj_195_beta_0_to_fp16 = const()[name = string("obj_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250493248)))]; + fp16 obj_195_epsilon_0_to_fp16 = const()[name = string("obj_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_195_cast_fp16 = batch_norm(beta = obj_195_beta_0_to_fp16, epsilon = obj_195_epsilon_0_to_fp16, gamma = obj_195_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("obj_195_cast_fp16")]; + string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; + tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; + tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; + int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; + tensor layers_10_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250494848)))]; + tensor layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251674560)))]; + tensor query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16, x = obj_195_cast_fp16)[name = string("query_43_cast_fp16")]; + tensor var_2342 = const()[name = string("op_2342"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_43_cast_fp16 = reshape(shape = var_2342, x = query_43_cast_fp16)[name = string("mh_q_43_cast_fp16")]; + fp16 var_2344_to_fp16 = const()[name = string("op_2344_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2345_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2344_to_fp16)[name = string("op_2345_cast_fp16")]; + tensor var_2346 = const()[name = string("op_2346"), val = tensor([1, 12, 64, -1])]; + tensor var_2347_cast_fp16 = reshape(shape = var_2346, x = obj_197_cast_fp16)[name = string("op_2347_cast_fp16")]; + bool mh_w_85_transpose_x_0 = const()[name = string("mh_w_85_transpose_x_0"), val = bool(true)]; + bool mh_w_85_transpose_y_0 = const()[name = string("mh_w_85_transpose_y_0"), val = bool(false)]; + tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_2345_cast_fp16, y = var_2347_cast_fp16)[name = string("mh_w_85_cast_fp16")]; + tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_87_cast_fp16")]; + tensor obj_203_cast_fp16 = softmax(axis = var_2214, x = mh_w_87_cast_fp16)[name = string("obj_203_cast_fp16")]; + tensor var_2356 = const()[name = string("op_2356"), val = tensor([1, 12, 64, -1])]; + tensor var_2357_cast_fp16 = reshape(shape = var_2356, x = obj_199_cast_fp16)[name = string("op_2357_cast_fp16")]; + bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)]; + bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)]; + tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2357_cast_fp16, y = obj_203_cast_fp16)[name = string("attn_43_cast_fp16")]; + tensor var_2360 = const()[name = string("op_2360"), val = tensor([1, 768, 1, -1])]; + tensor input_103_cast_fp16 = reshape(shape = var_2360, x = attn_43_cast_fp16)[name = string("input_103_cast_fp16")]; + string obj_201_pad_type_0 = const()[name = string("obj_201_pad_type_0"), val = string("valid")]; + tensor obj_201_strides_0 = const()[name = string("obj_201_strides_0"), val = tensor([1, 1])]; + tensor obj_201_pad_0 = const()[name = string("obj_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_201_dilations_0 = const()[name = string("obj_201_dilations_0"), val = tensor([1, 1])]; + int32 obj_201_groups_0 = const()[name = string("obj_201_groups_0"), val = int32(1)]; + tensor layers_10_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251676160)))]; + tensor layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252855872)))]; + tensor obj_201_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_201_dilations_0, groups = obj_201_groups_0, pad = obj_201_pad_0, pad_type = obj_201_pad_type_0, strides = obj_201_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("obj_201_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_201_cast_fp16)[name = string("inputs_65_cast_fp16")]; + tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; + fp16 var_2381_to_fp16 = const()[name = string("op_2381_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2381_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; + tensor input_105_gamma_0_to_fp16 = const()[name = string("input_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252857472)))]; + tensor input_105_beta_0_to_fp16 = const()[name = string("input_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252859072)))]; + fp16 input_105_epsilon_0_to_fp16 = const()[name = string("input_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("input_105_cast_fp16")]; + string input_107_pad_type_0 = const()[name = string("input_107_pad_type_0"), val = string("valid")]; + tensor input_107_strides_0 = const()[name = string("input_107_strides_0"), val = tensor([1, 1])]; + tensor input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_107_dilations_0 = const()[name = string("input_107_dilations_0"), val = tensor([1, 1])]; + int32 input_107_groups_0 = const()[name = string("input_107_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252860672)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257579328)))]; + tensor input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")]; + string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("EXACT")]; + tensor input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257585536)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262304192)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_67_cast_fp16")]; + tensor obj_215_begin_0 = const()[name = string("obj_215_begin_0"), val = tensor([11, 0, 0, 0])]; + tensor obj_215_end_0 = const()[name = string("obj_215_end_0"), val = tensor([12, 768, 1, 1536])]; + tensor obj_215_end_mask_0 = const()[name = string("obj_215_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_215_cast_fp16 = slice_by_index(begin = obj_215_begin_0, end = obj_215_end_0, end_mask = obj_215_end_mask_0, x = read_state_2)[name = string("obj_215_cast_fp16")]; + tensor obj_217_begin_0 = const()[name = string("obj_217_begin_0"), val = tensor([11, 0, 0, 0])]; + tensor obj_217_end_0 = const()[name = string("obj_217_end_0"), val = tensor([12, 768, 1, 1536])]; + tensor obj_217_end_mask_0 = const()[name = string("obj_217_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_217_cast_fp16 = slice_by_index(begin = obj_217_begin_0, end = obj_217_end_0, end_mask = obj_217_end_mask_0, x = read_state_3)[name = string("obj_217_cast_fp16")]; + int32 var_2427 = const()[name = string("op_2427"), val = int32(3)]; + tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; + fp16 var_2452_to_fp16 = const()[name = string("op_2452_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2452_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; + tensor obj_205_gamma_0_to_fp16 = const()[name = string("obj_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262305792)))]; + tensor obj_205_beta_0_to_fp16 = const()[name = string("obj_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262307392)))]; + fp16 obj_205_epsilon_0_to_fp16 = const()[name = string("obj_205_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_205_cast_fp16 = batch_norm(beta = obj_205_beta_0_to_fp16, epsilon = obj_205_epsilon_0_to_fp16, gamma = obj_205_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("obj_205_cast_fp16")]; + string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; + tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; + tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; + int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262308992)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263488704)))]; + tensor query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("query_45_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263490304)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264670016)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265849728)))]; + tensor current_value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_2490_cast_fp16 = mul(x = current_key_cast_fp16, y = var_169_cast_fp16)[name = string("op_2490_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_65_cast_fp16_11, y = var_2490_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_2492_cast_fp16 = mul(x = current_value_cast_fp16, y = var_169_cast_fp16)[name = string("op_2492_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_80_cast_fp16_11, y = var_2492_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_2495 = const()[name = string("op_2495"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_45_cast_fp16 = reshape(shape = var_2495, x = query_45_cast_fp16)[name = string("mh_q_45_cast_fp16")]; + fp16 var_2497_to_fp16 = const()[name = string("op_2497_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2498_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2497_to_fp16)[name = string("op_2498_cast_fp16")]; + tensor var_2499 = const()[name = string("op_2499"), val = tensor([1, 12, 64, -1])]; + tensor var_2500_cast_fp16 = reshape(shape = var_2499, x = key_cast_fp16)[name = string("op_2500_cast_fp16")]; + bool mh_w_89_transpose_x_0 = const()[name = string("mh_w_89_transpose_x_0"), val = bool(true)]; + bool mh_w_89_transpose_y_0 = const()[name = string("mh_w_89_transpose_y_0"), val = bool(false)]; + tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_2498_cast_fp16, y = var_2500_cast_fp16)[name = string("mh_w_89_cast_fp16")]; + tensor mh_w_91_cast_fp16 = add(x = mh_w_89_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_91_cast_fp16")]; + tensor var_2508_cast_fp16 = softmax(axis = var_2427, x = mh_w_91_cast_fp16)[name = string("op_2508_cast_fp16")]; + tensor var_2509 = const()[name = string("op_2509"), val = tensor([1, 12, 64, -1])]; + tensor var_2510_cast_fp16 = reshape(shape = var_2509, x = value_cast_fp16)[name = string("op_2510_cast_fp16")]; + bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)]; + bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)]; + tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2510_cast_fp16, y = var_2508_cast_fp16)[name = string("attn_45_cast_fp16")]; + tensor var_2513 = const()[name = string("op_2513"), val = tensor([1, 768, 1, -1])]; + tensor input_111_cast_fp16 = reshape(shape = var_2513, x = attn_45_cast_fp16)[name = string("input_111_cast_fp16")]; + string obj_211_pad_type_0 = const()[name = string("obj_211_pad_type_0"), val = string("valid")]; + tensor obj_211_strides_0 = const()[name = string("obj_211_strides_0"), val = tensor([1, 1])]; + tensor obj_211_pad_0 = const()[name = string("obj_211_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_211_dilations_0 = const()[name = string("obj_211_dilations_0"), val = tensor([1, 1])]; + int32 obj_211_groups_0 = const()[name = string("obj_211_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265851328)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267031040)))]; + tensor obj_211_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_211_dilations_0, groups = obj_211_groups_0, pad = obj_211_pad_0, pad_type = obj_211_pad_type_0, strides = obj_211_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("obj_211_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_211_cast_fp16)[name = string("inputs_69_cast_fp16")]; + tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; + fp16 var_2535_to_fp16 = const()[name = string("op_2535_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2535_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; + tensor obj_213_gamma_0_to_fp16 = const()[name = string("obj_213_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267032640)))]; + tensor obj_213_beta_0_to_fp16 = const()[name = string("obj_213_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267034240)))]; + fp16 obj_213_epsilon_0_to_fp16 = const()[name = string("obj_213_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_213_cast_fp16 = batch_norm(beta = obj_213_beta_0_to_fp16, epsilon = obj_213_epsilon_0_to_fp16, gamma = obj_213_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_213_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_11_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267035840)))]; + tensor layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268215552)))]; + tensor query_cast_fp16 = conv(bias = layers_11_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_11_encoder_attn_q_proj_weight_to_fp16, x = obj_213_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_2555 = const()[name = string("op_2555"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_2555, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_2557_to_fp16 = const()[name = string("op_2557_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2558_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_2557_to_fp16)[name = string("op_2558_cast_fp16")]; + tensor var_2559 = const()[name = string("op_2559"), val = tensor([1, 12, 64, -1])]; + tensor var_2560_cast_fp16 = reshape(shape = var_2559, x = obj_215_cast_fp16)[name = string("op_2560_cast_fp16")]; + bool mh_w_93_transpose_x_0 = const()[name = string("mh_w_93_transpose_x_0"), val = bool(true)]; + bool mh_w_93_transpose_y_0 = const()[name = string("mh_w_93_transpose_y_0"), val = bool(false)]; + tensor mh_w_93_cast_fp16 = matmul(transpose_x = mh_w_93_transpose_x_0, transpose_y = mh_w_93_transpose_y_0, x = var_2558_cast_fp16, y = var_2560_cast_fp16)[name = string("mh_w_93_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_93_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_221_cast_fp16 = softmax(axis = var_2427, x = mh_w_cast_fp16)[name = string("obj_221_cast_fp16")]; + tensor var_2569 = const()[name = string("op_2569"), val = tensor([1, 12, 64, -1])]; + tensor var_2570_cast_fp16 = reshape(shape = var_2569, x = obj_217_cast_fp16)[name = string("op_2570_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_2570_cast_fp16, y = obj_221_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_2573 = const()[name = string("op_2573"), val = tensor([1, 768, 1, -1])]; + tensor input_113_cast_fp16 = reshape(shape = var_2573, x = attn_cast_fp16)[name = string("input_113_cast_fp16")]; + string obj_219_pad_type_0 = const()[name = string("obj_219_pad_type_0"), val = string("valid")]; + tensor obj_219_strides_0 = const()[name = string("obj_219_strides_0"), val = tensor([1, 1])]; + tensor obj_219_pad_0 = const()[name = string("obj_219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_219_dilations_0 = const()[name = string("obj_219_dilations_0"), val = tensor([1, 1])]; + int32 obj_219_groups_0 = const()[name = string("obj_219_groups_0"), val = int32(1)]; + tensor layers_11_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268217152)))]; + tensor layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269396864)))]; + tensor obj_219_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_219_dilations_0, groups = obj_219_groups_0, pad = obj_219_pad_0, pad_type = obj_219_pad_type_0, strides = obj_219_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_219_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_219_cast_fp16)[name = string("inputs_71_cast_fp16")]; + tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; + fp16 var_2594_to_fp16 = const()[name = string("op_2594_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2594_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269398464)))]; + tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269400064)))]; + fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269401664)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274120320)))]; + tensor input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_117_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274126528)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278845184)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_2637_to_fp16 = const()[name = string("op_2637_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2637_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278846784)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278848384)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_2648_axes_0 = const()[name = string("op_2648_axes_0"), val = tensor([2])]; + tensor var_2648_cast_fp16 = squeeze(axes = var_2648_axes_0, x = hidden_states_cast_fp16)[name = string("op_2648_cast_fp16")]; + tensor var_2651_perm_0 = const()[name = string("op_2651_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278849984)))]; + tensor var_2651_cast_fp16 = transpose(perm = var_2651_perm_0, x = var_2648_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_2651_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_2655 = const()[name = string("op_2655"), val = int32(1)]; + bool obj_225_interleave_0 = const()[name = string("obj_225_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_2655, interleave = obj_225_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_cast_fp16))[name = string("obj_225_cast_fp16")]; + int32 var_2658 = const()[name = string("op_2658"), val = int32(1)]; + bool obj_227_interleave_0 = const()[name = string("obj_227_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_2658, interleave = obj_227_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_cast_fp16))[name = string("obj_227_cast_fp16")]; + tensor var_2669_begin_0 = const()[name = string("op_2669_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2669_end_0 = const()[name = string("op_2669_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_2669_end_mask_0 = const()[name = string("op_2669_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2669_cast_fp16 = slice_by_index(begin = var_2669_begin_0, end = var_2669_end_0, end_mask = var_2669_end_mask_0, x = obj_131_cast_fp16)[name = string("op_2669_cast_fp16")]; + tensor var_2672_begin_0 = const()[name = string("op_2672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2672_end_0 = const()[name = string("op_2672_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2672_end_mask_0 = const()[name = string("op_2672_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2672_squeeze_mask_0 = const()[name = string("op_2672_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2672_cast_fp16 = slice_by_index(begin = var_2672_begin_0, end = var_2672_end_0, end_mask = var_2672_end_mask_0, squeeze_mask = var_2672_squeeze_mask_0, x = var_2669_cast_fp16)[name = string("op_2672_cast_fp16")]; + tensor var_2687_begin_0 = const()[name = string("op_2687_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2687_end_0 = const()[name = string("op_2687_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2687_end_mask_0 = const()[name = string("op_2687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2687_cast_fp16 = slice_by_index(begin = var_2687_begin_0, end = var_2687_end_0, end_mask = var_2687_end_mask_0, x = obj_149_cast_fp16)[name = string("op_2687_cast_fp16")]; + tensor var_2690_begin_0 = const()[name = string("op_2690_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2690_end_0 = const()[name = string("op_2690_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2690_end_mask_0 = const()[name = string("op_2690_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2690_squeeze_mask_0 = const()[name = string("op_2690_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2690_cast_fp16 = slice_by_index(begin = var_2690_begin_0, end = var_2690_end_0, end_mask = var_2690_end_mask_0, squeeze_mask = var_2690_squeeze_mask_0, x = var_2687_cast_fp16)[name = string("op_2690_cast_fp16")]; + tensor var_2705_begin_0 = const()[name = string("op_2705_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2705_end_0 = const()[name = string("op_2705_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_2705_end_mask_0 = const()[name = string("op_2705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2705_cast_fp16 = slice_by_index(begin = var_2705_begin_0, end = var_2705_end_0, end_mask = var_2705_end_mask_0, x = obj_149_cast_fp16)[name = string("op_2705_cast_fp16")]; + tensor var_2708_begin_0 = const()[name = string("op_2708_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2708_end_0 = const()[name = string("op_2708_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2708_end_mask_0 = const()[name = string("op_2708_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2708_squeeze_mask_0 = const()[name = string("op_2708_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2708_cast_fp16 = slice_by_index(begin = var_2708_begin_0, end = var_2708_end_0, end_mask = var_2708_end_mask_0, squeeze_mask = var_2708_squeeze_mask_0, x = var_2705_cast_fp16)[name = string("op_2708_cast_fp16")]; + tensor var_2723_begin_0 = const()[name = string("op_2723_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2723_end_0 = const()[name = string("op_2723_end_0"), val = tensor([1, 9, 1, 1536])]; + tensor var_2723_end_mask_0 = const()[name = string("op_2723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2723_cast_fp16 = slice_by_index(begin = var_2723_begin_0, end = var_2723_end_0, end_mask = var_2723_end_mask_0, x = obj_149_cast_fp16)[name = string("op_2723_cast_fp16")]; + tensor var_2726_begin_0 = const()[name = string("op_2726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2726_end_0 = const()[name = string("op_2726_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2726_end_mask_0 = const()[name = string("op_2726_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2726_squeeze_mask_0 = const()[name = string("op_2726_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2726_cast_fp16 = slice_by_index(begin = var_2726_begin_0, end = var_2726_end_0, end_mask = var_2726_end_mask_0, squeeze_mask = var_2726_squeeze_mask_0, x = var_2723_cast_fp16)[name = string("op_2726_cast_fp16")]; + tensor var_2741_begin_0 = const()[name = string("op_2741_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2741_end_0 = const()[name = string("op_2741_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_2741_end_mask_0 = const()[name = string("op_2741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2741_cast_fp16 = slice_by_index(begin = var_2741_begin_0, end = var_2741_end_0, end_mask = var_2741_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2741_cast_fp16")]; + tensor var_2744_begin_0 = const()[name = string("op_2744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2744_end_0 = const()[name = string("op_2744_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2744_end_mask_0 = const()[name = string("op_2744_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2744_squeeze_mask_0 = const()[name = string("op_2744_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2744_cast_fp16 = slice_by_index(begin = var_2744_begin_0, end = var_2744_end_0, end_mask = var_2744_end_mask_0, squeeze_mask = var_2744_squeeze_mask_0, x = var_2741_cast_fp16)[name = string("op_2744_cast_fp16")]; + tensor var_2759_begin_0 = const()[name = string("op_2759_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2759_end_0 = const()[name = string("op_2759_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_2759_end_mask_0 = const()[name = string("op_2759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2759_cast_fp16 = slice_by_index(begin = var_2759_begin_0, end = var_2759_end_0, end_mask = var_2759_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2759_cast_fp16")]; + tensor var_2762_begin_0 = const()[name = string("op_2762_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2762_end_0 = const()[name = string("op_2762_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2762_end_mask_0 = const()[name = string("op_2762_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2762_squeeze_mask_0 = const()[name = string("op_2762_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2762_cast_fp16 = slice_by_index(begin = var_2762_begin_0, end = var_2762_end_0, end_mask = var_2762_end_mask_0, squeeze_mask = var_2762_squeeze_mask_0, x = var_2759_cast_fp16)[name = string("op_2762_cast_fp16")]; + tensor var_2777_begin_0 = const()[name = string("op_2777_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2777_end_0 = const()[name = string("op_2777_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_2777_end_mask_0 = const()[name = string("op_2777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2777_cast_fp16 = slice_by_index(begin = var_2777_begin_0, end = var_2777_end_0, end_mask = var_2777_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2777_cast_fp16")]; + tensor var_2780_begin_0 = const()[name = string("op_2780_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2780_end_0 = const()[name = string("op_2780_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2780_end_mask_0 = const()[name = string("op_2780_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2780_squeeze_mask_0 = const()[name = string("op_2780_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2780_cast_fp16 = slice_by_index(begin = var_2780_begin_0, end = var_2780_end_0, end_mask = var_2780_end_mask_0, squeeze_mask = var_2780_squeeze_mask_0, x = var_2777_cast_fp16)[name = string("op_2780_cast_fp16")]; + tensor var_2795_begin_0 = const()[name = string("op_2795_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2795_end_0 = const()[name = string("op_2795_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2795_end_mask_0 = const()[name = string("op_2795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2795_cast_fp16 = slice_by_index(begin = var_2795_begin_0, end = var_2795_end_0, end_mask = var_2795_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2795_cast_fp16")]; + tensor var_2798_begin_0 = const()[name = string("op_2798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2798_end_0 = const()[name = string("op_2798_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2798_end_mask_0 = const()[name = string("op_2798_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2798_squeeze_mask_0 = const()[name = string("op_2798_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2798_cast_fp16 = slice_by_index(begin = var_2798_begin_0, end = var_2798_end_0, end_mask = var_2798_end_mask_0, squeeze_mask = var_2798_squeeze_mask_0, x = var_2795_cast_fp16)[name = string("op_2798_cast_fp16")]; + tensor var_2813_begin_0 = const()[name = string("op_2813_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2813_end_0 = const()[name = string("op_2813_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_2813_end_mask_0 = const()[name = string("op_2813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2813_cast_fp16 = slice_by_index(begin = var_2813_begin_0, end = var_2813_end_0, end_mask = var_2813_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2813_cast_fp16")]; + tensor var_2816_begin_0 = const()[name = string("op_2816_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2816_end_0 = const()[name = string("op_2816_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2816_end_mask_0 = const()[name = string("op_2816_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2816_squeeze_mask_0 = const()[name = string("op_2816_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2816_cast_fp16 = slice_by_index(begin = var_2816_begin_0, end = var_2816_end_0, end_mask = var_2816_end_mask_0, squeeze_mask = var_2816_squeeze_mask_0, x = var_2813_cast_fp16)[name = string("op_2816_cast_fp16")]; + tensor var_2831_begin_0 = const()[name = string("op_2831_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2831_end_0 = const()[name = string("op_2831_end_0"), val = tensor([1, 9, 1, 1536])]; + tensor var_2831_end_mask_0 = const()[name = string("op_2831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2831_cast_fp16 = slice_by_index(begin = var_2831_begin_0, end = var_2831_end_0, end_mask = var_2831_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2831_cast_fp16")]; + tensor var_2834_begin_0 = const()[name = string("op_2834_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2834_end_0 = const()[name = string("op_2834_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2834_end_mask_0 = const()[name = string("op_2834_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2834_squeeze_mask_0 = const()[name = string("op_2834_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2834_cast_fp16 = slice_by_index(begin = var_2834_begin_0, end = var_2834_end_0, end_mask = var_2834_end_mask_0, squeeze_mask = var_2834_squeeze_mask_0, x = var_2831_cast_fp16)[name = string("op_2834_cast_fp16")]; + tensor var_2849_begin_0 = const()[name = string("op_2849_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2849_end_0 = const()[name = string("op_2849_end_0"), val = tensor([1, 11, 1, 1536])]; + tensor var_2849_end_mask_0 = const()[name = string("op_2849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2849_cast_fp16 = slice_by_index(begin = var_2849_begin_0, end = var_2849_end_0, end_mask = var_2849_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2849_cast_fp16")]; + tensor var_2852_begin_0 = const()[name = string("op_2852_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2852_end_0 = const()[name = string("op_2852_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2852_end_mask_0 = const()[name = string("op_2852_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2852_squeeze_mask_0 = const()[name = string("op_2852_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2852_cast_fp16 = slice_by_index(begin = var_2852_begin_0, end = var_2852_end_0, end_mask = var_2852_end_mask_0, squeeze_mask = var_2852_squeeze_mask_0, x = var_2849_cast_fp16)[name = string("op_2852_cast_fp16")]; + tensor var_2867_begin_0 = const()[name = string("op_2867_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2867_end_0 = const()[name = string("op_2867_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2867_end_mask_0 = const()[name = string("op_2867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2867_cast_fp16 = slice_by_index(begin = var_2867_begin_0, end = var_2867_end_0, end_mask = var_2867_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2867_cast_fp16")]; + tensor var_2870_begin_0 = const()[name = string("op_2870_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2870_end_0 = const()[name = string("op_2870_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2870_end_mask_0 = const()[name = string("op_2870_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2870_squeeze_mask_0 = const()[name = string("op_2870_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2870_cast_fp16 = slice_by_index(begin = var_2870_begin_0, end = var_2870_end_0, end_mask = var_2870_end_mask_0, squeeze_mask = var_2870_squeeze_mask_0, x = var_2867_cast_fp16)[name = string("op_2870_cast_fp16")]; + tensor var_2885_begin_0 = const()[name = string("op_2885_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2885_end_0 = const()[name = string("op_2885_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_2885_end_mask_0 = const()[name = string("op_2885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2885_cast_fp16 = slice_by_index(begin = var_2885_begin_0, end = var_2885_end_0, end_mask = var_2885_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2885_cast_fp16")]; + tensor var_2888_begin_0 = const()[name = string("op_2888_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2888_end_0 = const()[name = string("op_2888_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2888_end_mask_0 = const()[name = string("op_2888_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2888_squeeze_mask_0 = const()[name = string("op_2888_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2888_cast_fp16 = slice_by_index(begin = var_2888_begin_0, end = var_2888_end_0, end_mask = var_2888_end_mask_0, squeeze_mask = var_2888_squeeze_mask_0, x = var_2885_cast_fp16)[name = string("op_2888_cast_fp16")]; + tensor var_2903_begin_0 = const()[name = string("op_2903_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2903_end_0 = const()[name = string("op_2903_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_2903_end_mask_0 = const()[name = string("op_2903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2903_cast_fp16 = slice_by_index(begin = var_2903_begin_0, end = var_2903_end_0, end_mask = var_2903_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2903_cast_fp16")]; + tensor var_2906_begin_0 = const()[name = string("op_2906_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2906_end_0 = const()[name = string("op_2906_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2906_end_mask_0 = const()[name = string("op_2906_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2906_squeeze_mask_0 = const()[name = string("op_2906_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2906_cast_fp16 = slice_by_index(begin = var_2906_begin_0, end = var_2906_end_0, end_mask = var_2906_end_mask_0, squeeze_mask = var_2906_squeeze_mask_0, x = var_2903_cast_fp16)[name = string("op_2906_cast_fp16")]; + tensor var_2921_begin_0 = const()[name = string("op_2921_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2921_end_0 = const()[name = string("op_2921_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_2921_end_mask_0 = const()[name = string("op_2921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2921_cast_fp16 = slice_by_index(begin = var_2921_begin_0, end = var_2921_end_0, end_mask = var_2921_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2921_cast_fp16")]; + tensor var_2924_begin_0 = const()[name = string("op_2924_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2924_end_0 = const()[name = string("op_2924_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2924_end_mask_0 = const()[name = string("op_2924_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2924_squeeze_mask_0 = const()[name = string("op_2924_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2924_cast_fp16 = slice_by_index(begin = var_2924_begin_0, end = var_2924_end_0, end_mask = var_2924_end_mask_0, squeeze_mask = var_2924_squeeze_mask_0, x = var_2921_cast_fp16)[name = string("op_2924_cast_fp16")]; + tensor var_2939_begin_0 = const()[name = string("op_2939_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2939_end_0 = const()[name = string("op_2939_end_0"), val = tensor([1, 7, 1, 1536])]; + tensor var_2939_end_mask_0 = const()[name = string("op_2939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2939_cast_fp16 = slice_by_index(begin = var_2939_begin_0, end = var_2939_end_0, end_mask = var_2939_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2939_cast_fp16")]; + tensor var_2942_begin_0 = const()[name = string("op_2942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2942_end_0 = const()[name = string("op_2942_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2942_end_mask_0 = const()[name = string("op_2942_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2942_squeeze_mask_0 = const()[name = string("op_2942_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2942_cast_fp16 = slice_by_index(begin = var_2942_begin_0, end = var_2942_end_0, end_mask = var_2942_end_mask_0, squeeze_mask = var_2942_squeeze_mask_0, x = var_2939_cast_fp16)[name = string("op_2942_cast_fp16")]; + tensor var_2957_begin_0 = const()[name = string("op_2957_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2957_end_0 = const()[name = string("op_2957_end_0"), val = tensor([1, 12, 1, 1536])]; + tensor var_2957_end_mask_0 = const()[name = string("op_2957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2957_cast_fp16 = slice_by_index(begin = var_2957_begin_0, end = var_2957_end_0, end_mask = var_2957_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2957_cast_fp16")]; + tensor var_2960_begin_0 = const()[name = string("op_2960_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2960_end_0 = const()[name = string("op_2960_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2960_end_mask_0 = const()[name = string("op_2960_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2960_squeeze_mask_0 = const()[name = string("op_2960_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2960_cast_fp16 = slice_by_index(begin = var_2960_begin_0, end = var_2960_end_0, end_mask = var_2960_end_mask_0, squeeze_mask = var_2960_squeeze_mask_0, x = var_2957_cast_fp16)[name = string("op_2960_cast_fp16")]; + tensor var_2975_begin_0 = const()[name = string("op_2975_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2975_end_0 = const()[name = string("op_2975_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_2975_end_mask_0 = const()[name = string("op_2975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2975_cast_fp16 = slice_by_index(begin = var_2975_begin_0, end = var_2975_end_0, end_mask = var_2975_end_mask_0, x = obj_221_cast_fp16)[name = string("op_2975_cast_fp16")]; + tensor var_2978_begin_0 = const()[name = string("op_2978_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2978_end_0 = const()[name = string("op_2978_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2978_end_mask_0 = const()[name = string("op_2978_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2978_squeeze_mask_0 = const()[name = string("op_2978_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2978_cast_fp16 = slice_by_index(begin = var_2978_begin_0, end = var_2978_end_0, end_mask = var_2978_end_mask_0, squeeze_mask = var_2978_squeeze_mask_0, x = var_2975_cast_fp16)[name = string("op_2978_cast_fp16")]; + tensor var_2993_begin_0 = const()[name = string("op_2993_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2993_end_0 = const()[name = string("op_2993_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_2993_end_mask_0 = const()[name = string("op_2993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2993_cast_fp16 = slice_by_index(begin = var_2993_begin_0, end = var_2993_end_0, end_mask = var_2993_end_mask_0, x = obj_221_cast_fp16)[name = string("op_2993_cast_fp16")]; + tensor var_2996_begin_0 = const()[name = string("op_2996_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2996_end_0 = const()[name = string("op_2996_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2996_end_mask_0 = const()[name = string("op_2996_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2996_squeeze_mask_0 = const()[name = string("op_2996_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2996_cast_fp16 = slice_by_index(begin = var_2996_begin_0, end = var_2996_end_0, end_mask = var_2996_end_mask_0, squeeze_mask = var_2996_squeeze_mask_0, x = var_2993_cast_fp16)[name = string("op_2996_cast_fp16")]; + int32 var_3003 = const()[name = string("op_3003"), val = int32(1)]; + bool var_3004_interleave_0 = const()[name = string("op_3004_interleave_0"), val = bool(false)]; + tensor var_3004_cast_fp16 = concat(axis = var_3003, interleave = var_3004_interleave_0, values = (var_2672_cast_fp16, var_2690_cast_fp16, var_2708_cast_fp16, var_2726_cast_fp16, var_2744_cast_fp16, var_2762_cast_fp16, var_2780_cast_fp16, var_2798_cast_fp16, var_2816_cast_fp16, var_2834_cast_fp16, var_2852_cast_fp16, var_2870_cast_fp16, var_2888_cast_fp16, var_2906_cast_fp16, var_2924_cast_fp16, var_2942_cast_fp16, var_2960_cast_fp16, var_2978_cast_fp16, var_2996_cast_fp16))[name = string("op_3004_cast_fp16")]; + bool var_3007 = const()[name = string("op_3007"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_3007, x = var_3004_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-small.en/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-small.en/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..9a7a6e86683e4afb379eff5d1bac7865d9d6a586 --- /dev/null +++ b/openai_whisper-small.en/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6be0784d200cceb2ade0deddfdd7acdf78f9475c4a63f31b248fe62e65c6985 +size 278953776 diff --git a/openai_whisper-small.en/config.json b/openai_whisper-small.en/config.json new file mode 100644 index 0000000000000000000000000000000000000000..9dee569cf0c20925208ec84fecbb95e873f8bf24 --- /dev/null +++ b/openai_whisper-small.en/config.json @@ -0,0 +1 @@ +{"_name_or_path": "openai/whisper-small", "activation_dropout": 0.0, "activation_function": "gelu", "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "d_model": 768, "decoder_attention_heads": 12, "decoder_ffn_dim": 3072, "decoder_layerdrop": 0.0, "decoder_layers": 12, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 12, "encoder_ffn_dim": 3072, "encoder_layerdrop": 0.0, "encoder_layers": 12, "eos_token_id": 50257, "forced_decoder_ids": [[1, 50259], [2, 50359], [3, 50363]], "init_std": 0.02, "is_encoder_decoder": true, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "model_type": "whisper", "num_hidden_layers": 12, "num_mel_bins": 80, "pad_token_id": 50257, "scale_embedding": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50360, 50361, 50362], "torch_dtype": "float32", "transformers_version": "4.27.0.dev0", "use_cache": true, "vocab_size": 51865} \ No newline at end of file diff --git a/openai_whisper-small.en/generation_config.json b/openai_whisper-small.en/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..cdd26273f9cd1ab8ecda49f5b8c033134c61cb4a --- /dev/null +++ b/openai_whisper-small.en/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[5, 3], [5, 9], [8, 0], [8, 4], [8, 7], [8, 8], [9, 0], [9, 7], [9, 9], [10, 5]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50359]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50363, "pad_token_id": 50257, "prev_sot_token_id": 50361, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "task_to_id": {"transcribe": 50359, "translate": 50358}, "transformers_version": "4.31.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-small/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-small/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..f6ea167c2b8bc62d8606159c65bc2cd55c892f4f --- /dev/null +++ b/openai_whisper-small/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2178a11e8833a7745057ce38dce4588cb4bca380e3398b952761f12aee0a93cb +size 243 diff --git a/openai_whisper-small/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-small/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..4e7a2bb9e7ef3870c50986867ffd5a6788e5ed84 --- /dev/null +++ b/openai_whisper-small/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a078e65c9369ce8a4a687a2bbb0a8befbd4ed459250c0442176824906fa95ee1 +size 433 diff --git a/openai_whisper-small/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-small/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1587204f11c01ae4618a054b766c946375a44b18 --- /dev/null +++ b/openai_whisper-small/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 768 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 768, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 25, + "Ios18.conv" : 98, + "Ios18.gelu" : 14, + "Ios18.concat" : 158, + "Ios16.einsum" : 1152, + "Ios18.add" : 25, + "Ios18.softmax" : 576, + "Ios18.sliceByIndex" : 1008, + "Ios18.layerNorm" : 25, + "Ios18.transpose" : 12, + "Ios18.mul" : 576 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-small/AudioEncoder.mlmodelc/model.mil b/openai_whisper-small/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..9caa00c57cd12baeb3f79c05eac5d3c9f247585c --- /dev/null +++ b/openai_whisper-small/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,9435 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_90_pad_type_0 = const()[name = string("op_90_pad_type_0"), val = string("custom")]; + tensor var_90_pad_0 = const()[name = string("op_90_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_90_strides_0 = const()[name = string("op_90_strides_0"), val = tensor([1, 1])]; + tensor var_90_dilations_0 = const()[name = string("op_90_dilations_0"), val = tensor([1, 1])]; + int32 var_90_groups_0 = const()[name = string("op_90_groups_0"), val = int32(1)]; + tensor var_65_to_fp16 = const()[name = string("op_65_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_71_to_fp16 = const()[name = string("op_71_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368768)))]; + tensor var_90_cast_fp16 = conv(bias = var_71_to_fp16, dilations = var_90_dilations_0, groups = var_90_groups_0, pad = var_90_pad_0, pad_type = var_90_pad_type_0, strides = var_90_strides_0, weight = var_65_to_fp16, x = melspectrogram_features)[name = string("op_90_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_90_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_130_pad_type_0 = const()[name = string("op_130_pad_type_0"), val = string("custom")]; + tensor var_130_pad_0 = const()[name = string("op_130_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_130_strides_0 = const()[name = string("op_130_strides_0"), val = tensor([2, 2])]; + tensor var_130_dilations_0 = const()[name = string("op_130_dilations_0"), val = tensor([1, 1])]; + int32 var_130_groups_0 = const()[name = string("op_130_groups_0"), val = int32(1)]; + tensor var_105_to_fp16 = const()[name = string("op_105_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370368)))]; + tensor var_111_to_fp16 = const()[name = string("op_111_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3909376)))]; + tensor var_130_cast_fp16 = conv(bias = var_111_to_fp16, dilations = var_130_dilations_0, groups = var_130_groups_0, pad = var_130_pad_0, pad_type = var_130_pad_type_0, strides = var_130_strides_0, weight = var_105_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_130_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_130_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_148_to_fp16 = const()[name = string("op_148_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3910976)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_148_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_158 = const()[name = string("op_158"), val = int32(3)]; + int32 var_175 = const()[name = string("op_175"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_192_to_fp16 = const()[name = string("op_192_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_192_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6215040)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6216640)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6218240)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6219840)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6221440)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7401152)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7402752)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8582464)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9762176)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = string("op_230_cast_fp16")]; + tensor var_234_begin_0 = const()[name = string("op_234_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_234_end_0 = const()[name = string("op_234_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_234_end_mask_0 = const()[name = string("op_234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_234_cast_fp16 = slice_by_index(begin = var_234_begin_0, end = var_234_end_0, end_mask = var_234_end_mask_0, x = query_1_cast_fp16)[name = string("op_234_cast_fp16")]; + tensor var_238_begin_0 = const()[name = string("op_238_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_238_end_0 = const()[name = string("op_238_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_238_end_mask_0 = const()[name = string("op_238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_238_cast_fp16 = slice_by_index(begin = var_238_begin_0, end = var_238_end_0, end_mask = var_238_end_mask_0, x = query_1_cast_fp16)[name = string("op_238_cast_fp16")]; + tensor var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = query_1_cast_fp16)[name = string("op_242_cast_fp16")]; + tensor var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = query_1_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor var_250_begin_0 = const()[name = string("op_250_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_250_end_0 = const()[name = string("op_250_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_250_end_mask_0 = const()[name = string("op_250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_250_cast_fp16 = slice_by_index(begin = var_250_begin_0, end = var_250_end_0, end_mask = var_250_end_mask_0, x = query_1_cast_fp16)[name = string("op_250_cast_fp16")]; + tensor var_254_begin_0 = const()[name = string("op_254_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_254_end_0 = const()[name = string("op_254_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_254_end_mask_0 = const()[name = string("op_254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_254_cast_fp16 = slice_by_index(begin = var_254_begin_0, end = var_254_end_0, end_mask = var_254_end_mask_0, x = query_1_cast_fp16)[name = string("op_254_cast_fp16")]; + tensor var_258_begin_0 = const()[name = string("op_258_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_258_end_0 = const()[name = string("op_258_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_258_end_mask_0 = const()[name = string("op_258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_258_cast_fp16 = slice_by_index(begin = var_258_begin_0, end = var_258_end_0, end_mask = var_258_end_mask_0, x = query_1_cast_fp16)[name = string("op_258_cast_fp16")]; + tensor var_262_begin_0 = const()[name = string("op_262_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_262_end_0 = const()[name = string("op_262_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_262_end_mask_0 = const()[name = string("op_262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_262_cast_fp16 = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = query_1_cast_fp16)[name = string("op_262_cast_fp16")]; + tensor var_266_begin_0 = const()[name = string("op_266_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_266_end_0 = const()[name = string("op_266_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_266_end_mask_0 = const()[name = string("op_266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = query_1_cast_fp16)[name = string("op_266_cast_fp16")]; + tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = query_1_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = query_1_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_283_begin_0 = const()[name = string("op_283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_283_end_0 = const()[name = string("op_283_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_283_end_mask_0 = const()[name = string("op_283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_283_cast_fp16 = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_230_cast_fp16)[name = string("op_283_cast_fp16")]; + tensor var_290_begin_0 = const()[name = string("op_290_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_290_end_0 = const()[name = string("op_290_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_290_end_mask_0 = const()[name = string("op_290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_230_cast_fp16)[name = string("op_290_cast_fp16")]; + tensor var_297_begin_0 = const()[name = string("op_297_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_297_end_0 = const()[name = string("op_297_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_297_end_mask_0 = const()[name = string("op_297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_297_cast_fp16 = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_230_cast_fp16)[name = string("op_297_cast_fp16")]; + tensor var_304_begin_0 = const()[name = string("op_304_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_304_end_0 = const()[name = string("op_304_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_304_end_mask_0 = const()[name = string("op_304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_304_cast_fp16 = slice_by_index(begin = var_304_begin_0, end = var_304_end_0, end_mask = var_304_end_mask_0, x = var_230_cast_fp16)[name = string("op_304_cast_fp16")]; + tensor var_311_begin_0 = const()[name = string("op_311_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_311_end_0 = const()[name = string("op_311_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_311_end_mask_0 = const()[name = string("op_311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = var_234_cast_fp16)[name = string("op_311_cast_fp16")]; + tensor var_318_begin_0 = const()[name = string("op_318_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_318_end_0 = const()[name = string("op_318_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_318_end_mask_0 = const()[name = string("op_318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = var_234_cast_fp16)[name = string("op_318_cast_fp16")]; + tensor var_325_begin_0 = const()[name = string("op_325_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_325_end_0 = const()[name = string("op_325_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_325_end_mask_0 = const()[name = string("op_325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_325_cast_fp16 = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = var_234_cast_fp16)[name = string("op_325_cast_fp16")]; + tensor var_332_begin_0 = const()[name = string("op_332_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_332_end_0 = const()[name = string("op_332_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_332_end_mask_0 = const()[name = string("op_332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = var_234_cast_fp16)[name = string("op_332_cast_fp16")]; + tensor var_339_begin_0 = const()[name = string("op_339_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_339_end_0 = const()[name = string("op_339_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_339_end_mask_0 = const()[name = string("op_339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_339_cast_fp16 = slice_by_index(begin = var_339_begin_0, end = var_339_end_0, end_mask = var_339_end_mask_0, x = var_238_cast_fp16)[name = string("op_339_cast_fp16")]; + tensor var_346_begin_0 = const()[name = string("op_346_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_346_end_0 = const()[name = string("op_346_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_346_end_mask_0 = const()[name = string("op_346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_346_cast_fp16 = slice_by_index(begin = var_346_begin_0, end = var_346_end_0, end_mask = var_346_end_mask_0, x = var_238_cast_fp16)[name = string("op_346_cast_fp16")]; + tensor var_353_begin_0 = const()[name = string("op_353_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_353_end_0 = const()[name = string("op_353_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_353_end_mask_0 = const()[name = string("op_353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_353_cast_fp16 = slice_by_index(begin = var_353_begin_0, end = var_353_end_0, end_mask = var_353_end_mask_0, x = var_238_cast_fp16)[name = string("op_353_cast_fp16")]; + tensor var_360_begin_0 = const()[name = string("op_360_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_360_end_0 = const()[name = string("op_360_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_360_end_mask_0 = const()[name = string("op_360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_360_cast_fp16 = slice_by_index(begin = var_360_begin_0, end = var_360_end_0, end_mask = var_360_end_mask_0, x = var_238_cast_fp16)[name = string("op_360_cast_fp16")]; + tensor var_367_begin_0 = const()[name = string("op_367_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_367_end_0 = const()[name = string("op_367_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_367_end_mask_0 = const()[name = string("op_367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_367_cast_fp16 = slice_by_index(begin = var_367_begin_0, end = var_367_end_0, end_mask = var_367_end_mask_0, x = var_242_cast_fp16)[name = string("op_367_cast_fp16")]; + tensor var_374_begin_0 = const()[name = string("op_374_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_374_end_0 = const()[name = string("op_374_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_374_end_mask_0 = const()[name = string("op_374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_374_cast_fp16 = slice_by_index(begin = var_374_begin_0, end = var_374_end_0, end_mask = var_374_end_mask_0, x = var_242_cast_fp16)[name = string("op_374_cast_fp16")]; + tensor var_381_begin_0 = const()[name = string("op_381_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_381_end_0 = const()[name = string("op_381_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_381_end_mask_0 = const()[name = string("op_381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_381_cast_fp16 = slice_by_index(begin = var_381_begin_0, end = var_381_end_0, end_mask = var_381_end_mask_0, x = var_242_cast_fp16)[name = string("op_381_cast_fp16")]; + tensor var_388_begin_0 = const()[name = string("op_388_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_388_end_0 = const()[name = string("op_388_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_388_end_mask_0 = const()[name = string("op_388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_388_cast_fp16 = slice_by_index(begin = var_388_begin_0, end = var_388_end_0, end_mask = var_388_end_mask_0, x = var_242_cast_fp16)[name = string("op_388_cast_fp16")]; + tensor var_395_begin_0 = const()[name = string("op_395_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_395_end_0 = const()[name = string("op_395_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_395_end_mask_0 = const()[name = string("op_395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_395_cast_fp16 = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = var_246_cast_fp16)[name = string("op_395_cast_fp16")]; + tensor var_402_begin_0 = const()[name = string("op_402_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_402_end_0 = const()[name = string("op_402_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_402_end_mask_0 = const()[name = string("op_402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_402_cast_fp16 = slice_by_index(begin = var_402_begin_0, end = var_402_end_0, end_mask = var_402_end_mask_0, x = var_246_cast_fp16)[name = string("op_402_cast_fp16")]; + tensor var_409_begin_0 = const()[name = string("op_409_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_409_end_0 = const()[name = string("op_409_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_409_end_mask_0 = const()[name = string("op_409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_409_cast_fp16 = slice_by_index(begin = var_409_begin_0, end = var_409_end_0, end_mask = var_409_end_mask_0, x = var_246_cast_fp16)[name = string("op_409_cast_fp16")]; + tensor var_416_begin_0 = const()[name = string("op_416_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_416_end_0 = const()[name = string("op_416_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_416_end_mask_0 = const()[name = string("op_416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_416_cast_fp16 = slice_by_index(begin = var_416_begin_0, end = var_416_end_0, end_mask = var_416_end_mask_0, x = var_246_cast_fp16)[name = string("op_416_cast_fp16")]; + tensor var_423_begin_0 = const()[name = string("op_423_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_423_end_0 = const()[name = string("op_423_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_423_end_mask_0 = const()[name = string("op_423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_423_cast_fp16 = slice_by_index(begin = var_423_begin_0, end = var_423_end_0, end_mask = var_423_end_mask_0, x = var_250_cast_fp16)[name = string("op_423_cast_fp16")]; + tensor var_430_begin_0 = const()[name = string("op_430_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_430_end_0 = const()[name = string("op_430_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_430_end_mask_0 = const()[name = string("op_430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_430_cast_fp16 = slice_by_index(begin = var_430_begin_0, end = var_430_end_0, end_mask = var_430_end_mask_0, x = var_250_cast_fp16)[name = string("op_430_cast_fp16")]; + tensor var_437_begin_0 = const()[name = string("op_437_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_437_end_0 = const()[name = string("op_437_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_437_end_mask_0 = const()[name = string("op_437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_437_cast_fp16 = slice_by_index(begin = var_437_begin_0, end = var_437_end_0, end_mask = var_437_end_mask_0, x = var_250_cast_fp16)[name = string("op_437_cast_fp16")]; + tensor var_444_begin_0 = const()[name = string("op_444_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_444_end_0 = const()[name = string("op_444_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_444_end_mask_0 = const()[name = string("op_444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_444_cast_fp16 = slice_by_index(begin = var_444_begin_0, end = var_444_end_0, end_mask = var_444_end_mask_0, x = var_250_cast_fp16)[name = string("op_444_cast_fp16")]; + tensor var_451_begin_0 = const()[name = string("op_451_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_451_end_0 = const()[name = string("op_451_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_451_end_mask_0 = const()[name = string("op_451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_451_cast_fp16 = slice_by_index(begin = var_451_begin_0, end = var_451_end_0, end_mask = var_451_end_mask_0, x = var_254_cast_fp16)[name = string("op_451_cast_fp16")]; + tensor var_458_begin_0 = const()[name = string("op_458_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_458_end_0 = const()[name = string("op_458_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_458_end_mask_0 = const()[name = string("op_458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_458_cast_fp16 = slice_by_index(begin = var_458_begin_0, end = var_458_end_0, end_mask = var_458_end_mask_0, x = var_254_cast_fp16)[name = string("op_458_cast_fp16")]; + tensor var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = var_254_cast_fp16)[name = string("op_465_cast_fp16")]; + tensor var_472_begin_0 = const()[name = string("op_472_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_472_end_0 = const()[name = string("op_472_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_472_end_mask_0 = const()[name = string("op_472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_472_cast_fp16 = slice_by_index(begin = var_472_begin_0, end = var_472_end_0, end_mask = var_472_end_mask_0, x = var_254_cast_fp16)[name = string("op_472_cast_fp16")]; + tensor var_479_begin_0 = const()[name = string("op_479_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_479_end_0 = const()[name = string("op_479_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_479_end_mask_0 = const()[name = string("op_479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_479_cast_fp16 = slice_by_index(begin = var_479_begin_0, end = var_479_end_0, end_mask = var_479_end_mask_0, x = var_258_cast_fp16)[name = string("op_479_cast_fp16")]; + tensor var_486_begin_0 = const()[name = string("op_486_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_486_end_0 = const()[name = string("op_486_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_486_end_mask_0 = const()[name = string("op_486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_486_cast_fp16 = slice_by_index(begin = var_486_begin_0, end = var_486_end_0, end_mask = var_486_end_mask_0, x = var_258_cast_fp16)[name = string("op_486_cast_fp16")]; + tensor var_493_begin_0 = const()[name = string("op_493_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_493_end_0 = const()[name = string("op_493_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_493_end_mask_0 = const()[name = string("op_493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_493_cast_fp16 = slice_by_index(begin = var_493_begin_0, end = var_493_end_0, end_mask = var_493_end_mask_0, x = var_258_cast_fp16)[name = string("op_493_cast_fp16")]; + tensor var_500_begin_0 = const()[name = string("op_500_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_500_end_0 = const()[name = string("op_500_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_500_end_mask_0 = const()[name = string("op_500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_500_cast_fp16 = slice_by_index(begin = var_500_begin_0, end = var_500_end_0, end_mask = var_500_end_mask_0, x = var_258_cast_fp16)[name = string("op_500_cast_fp16")]; + tensor var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = var_262_cast_fp16)[name = string("op_507_cast_fp16")]; + tensor var_514_begin_0 = const()[name = string("op_514_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_514_end_0 = const()[name = string("op_514_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_514_end_mask_0 = const()[name = string("op_514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_514_cast_fp16 = slice_by_index(begin = var_514_begin_0, end = var_514_end_0, end_mask = var_514_end_mask_0, x = var_262_cast_fp16)[name = string("op_514_cast_fp16")]; + tensor var_521_begin_0 = const()[name = string("op_521_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_521_end_0 = const()[name = string("op_521_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_521_end_mask_0 = const()[name = string("op_521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_521_cast_fp16 = slice_by_index(begin = var_521_begin_0, end = var_521_end_0, end_mask = var_521_end_mask_0, x = var_262_cast_fp16)[name = string("op_521_cast_fp16")]; + tensor var_528_begin_0 = const()[name = string("op_528_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_528_end_0 = const()[name = string("op_528_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_528_end_mask_0 = const()[name = string("op_528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_528_cast_fp16 = slice_by_index(begin = var_528_begin_0, end = var_528_end_0, end_mask = var_528_end_mask_0, x = var_262_cast_fp16)[name = string("op_528_cast_fp16")]; + tensor var_535_begin_0 = const()[name = string("op_535_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_535_end_0 = const()[name = string("op_535_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_535_end_mask_0 = const()[name = string("op_535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_535_cast_fp16 = slice_by_index(begin = var_535_begin_0, end = var_535_end_0, end_mask = var_535_end_mask_0, x = var_266_cast_fp16)[name = string("op_535_cast_fp16")]; + tensor var_542_begin_0 = const()[name = string("op_542_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_542_end_0 = const()[name = string("op_542_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_542_end_mask_0 = const()[name = string("op_542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_542_cast_fp16 = slice_by_index(begin = var_542_begin_0, end = var_542_end_0, end_mask = var_542_end_mask_0, x = var_266_cast_fp16)[name = string("op_542_cast_fp16")]; + tensor var_549_begin_0 = const()[name = string("op_549_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_549_end_0 = const()[name = string("op_549_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_549_end_mask_0 = const()[name = string("op_549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_549_cast_fp16 = slice_by_index(begin = var_549_begin_0, end = var_549_end_0, end_mask = var_549_end_mask_0, x = var_266_cast_fp16)[name = string("op_549_cast_fp16")]; + tensor var_556_begin_0 = const()[name = string("op_556_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_556_end_0 = const()[name = string("op_556_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_556_end_mask_0 = const()[name = string("op_556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_556_cast_fp16 = slice_by_index(begin = var_556_begin_0, end = var_556_end_0, end_mask = var_556_end_mask_0, x = var_266_cast_fp16)[name = string("op_556_cast_fp16")]; + tensor var_563_begin_0 = const()[name = string("op_563_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_563_end_0 = const()[name = string("op_563_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_563_end_mask_0 = const()[name = string("op_563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_563_cast_fp16 = slice_by_index(begin = var_563_begin_0, end = var_563_end_0, end_mask = var_563_end_mask_0, x = var_270_cast_fp16)[name = string("op_563_cast_fp16")]; + tensor var_570_begin_0 = const()[name = string("op_570_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_570_end_0 = const()[name = string("op_570_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_570_end_mask_0 = const()[name = string("op_570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_570_cast_fp16 = slice_by_index(begin = var_570_begin_0, end = var_570_end_0, end_mask = var_570_end_mask_0, x = var_270_cast_fp16)[name = string("op_570_cast_fp16")]; + tensor var_577_begin_0 = const()[name = string("op_577_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_577_end_0 = const()[name = string("op_577_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_577_end_mask_0 = const()[name = string("op_577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_577_cast_fp16 = slice_by_index(begin = var_577_begin_0, end = var_577_end_0, end_mask = var_577_end_mask_0, x = var_270_cast_fp16)[name = string("op_577_cast_fp16")]; + tensor var_584_begin_0 = const()[name = string("op_584_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_584_end_0 = const()[name = string("op_584_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_584_end_mask_0 = const()[name = string("op_584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_584_cast_fp16 = slice_by_index(begin = var_584_begin_0, end = var_584_end_0, end_mask = var_584_end_mask_0, x = var_270_cast_fp16)[name = string("op_584_cast_fp16")]; + tensor var_591_begin_0 = const()[name = string("op_591_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_591_end_0 = const()[name = string("op_591_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_591_end_mask_0 = const()[name = string("op_591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_591_cast_fp16 = slice_by_index(begin = var_591_begin_0, end = var_591_end_0, end_mask = var_591_end_mask_0, x = var_274_cast_fp16)[name = string("op_591_cast_fp16")]; + tensor var_598_begin_0 = const()[name = string("op_598_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_598_end_0 = const()[name = string("op_598_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_598_end_mask_0 = const()[name = string("op_598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_598_cast_fp16 = slice_by_index(begin = var_598_begin_0, end = var_598_end_0, end_mask = var_598_end_mask_0, x = var_274_cast_fp16)[name = string("op_598_cast_fp16")]; + tensor var_605_begin_0 = const()[name = string("op_605_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_605_end_0 = const()[name = string("op_605_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_605_end_mask_0 = const()[name = string("op_605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_605_cast_fp16 = slice_by_index(begin = var_605_begin_0, end = var_605_end_0, end_mask = var_605_end_mask_0, x = var_274_cast_fp16)[name = string("op_605_cast_fp16")]; + tensor var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, x = var_274_cast_fp16)[name = string("op_612_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_617_begin_0 = const()[name = string("op_617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_617_end_0 = const()[name = string("op_617_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_617_end_mask_0 = const()[name = string("op_617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_11")]; + tensor var_617_cast_fp16 = slice_by_index(begin = var_617_begin_0, end = var_617_end_0, end_mask = var_617_end_mask_0, x = k_1_cast_fp16)[name = string("op_617_cast_fp16")]; + tensor var_621_begin_0 = const()[name = string("op_621_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_621_end_0 = const()[name = string("op_621_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_621_end_mask_0 = const()[name = string("op_621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_621_cast_fp16 = slice_by_index(begin = var_621_begin_0, end = var_621_end_0, end_mask = var_621_end_mask_0, x = k_1_cast_fp16)[name = string("op_621_cast_fp16")]; + tensor var_625_begin_0 = const()[name = string("op_625_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_625_end_0 = const()[name = string("op_625_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_625_end_mask_0 = const()[name = string("op_625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_625_cast_fp16 = slice_by_index(begin = var_625_begin_0, end = var_625_end_0, end_mask = var_625_end_mask_0, x = k_1_cast_fp16)[name = string("op_625_cast_fp16")]; + tensor var_629_begin_0 = const()[name = string("op_629_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_629_end_0 = const()[name = string("op_629_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_629_end_mask_0 = const()[name = string("op_629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_629_cast_fp16 = slice_by_index(begin = var_629_begin_0, end = var_629_end_0, end_mask = var_629_end_mask_0, x = k_1_cast_fp16)[name = string("op_629_cast_fp16")]; + tensor var_633_begin_0 = const()[name = string("op_633_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_633_end_0 = const()[name = string("op_633_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_633_end_mask_0 = const()[name = string("op_633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_633_cast_fp16 = slice_by_index(begin = var_633_begin_0, end = var_633_end_0, end_mask = var_633_end_mask_0, x = k_1_cast_fp16)[name = string("op_633_cast_fp16")]; + tensor var_637_begin_0 = const()[name = string("op_637_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_637_end_0 = const()[name = string("op_637_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_637_end_mask_0 = const()[name = string("op_637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_637_cast_fp16 = slice_by_index(begin = var_637_begin_0, end = var_637_end_0, end_mask = var_637_end_mask_0, x = k_1_cast_fp16)[name = string("op_637_cast_fp16")]; + tensor var_641_begin_0 = const()[name = string("op_641_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_641_end_0 = const()[name = string("op_641_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_641_end_mask_0 = const()[name = string("op_641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_641_cast_fp16 = slice_by_index(begin = var_641_begin_0, end = var_641_end_0, end_mask = var_641_end_mask_0, x = k_1_cast_fp16)[name = string("op_641_cast_fp16")]; + tensor var_645_begin_0 = const()[name = string("op_645_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_645_end_0 = const()[name = string("op_645_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_645_end_mask_0 = const()[name = string("op_645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = k_1_cast_fp16)[name = string("op_645_cast_fp16")]; + tensor var_649_begin_0 = const()[name = string("op_649_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_649_end_0 = const()[name = string("op_649_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_649_end_mask_0 = const()[name = string("op_649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_649_cast_fp16 = slice_by_index(begin = var_649_begin_0, end = var_649_end_0, end_mask = var_649_end_mask_0, x = k_1_cast_fp16)[name = string("op_649_cast_fp16")]; + tensor var_653_begin_0 = const()[name = string("op_653_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_653_end_0 = const()[name = string("op_653_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_653_end_mask_0 = const()[name = string("op_653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_653_cast_fp16 = slice_by_index(begin = var_653_begin_0, end = var_653_end_0, end_mask = var_653_end_mask_0, x = k_1_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor var_657_begin_0 = const()[name = string("op_657_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_657_end_0 = const()[name = string("op_657_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_657_end_mask_0 = const()[name = string("op_657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_657_cast_fp16 = slice_by_index(begin = var_657_begin_0, end = var_657_end_0, end_mask = var_657_end_mask_0, x = k_1_cast_fp16)[name = string("op_657_cast_fp16")]; + tensor var_661_begin_0 = const()[name = string("op_661_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_661_end_0 = const()[name = string("op_661_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_661_end_mask_0 = const()[name = string("op_661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_661_cast_fp16 = slice_by_index(begin = var_661_begin_0, end = var_661_end_0, end_mask = var_661_end_mask_0, x = k_1_cast_fp16)[name = string("op_661_cast_fp16")]; + tensor var_663_begin_0 = const()[name = string("op_663_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_663_end_0 = const()[name = string("op_663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_663_end_mask_0 = const()[name = string("op_663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = value_1_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_667_begin_0 = const()[name = string("op_667_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_667_end_0 = const()[name = string("op_667_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_667_end_mask_0 = const()[name = string("op_667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_667_cast_fp16 = slice_by_index(begin = var_667_begin_0, end = var_667_end_0, end_mask = var_667_end_mask_0, x = value_1_cast_fp16)[name = string("op_667_cast_fp16")]; + tensor var_671_begin_0 = const()[name = string("op_671_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_671_end_0 = const()[name = string("op_671_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_671_end_mask_0 = const()[name = string("op_671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_671_cast_fp16 = slice_by_index(begin = var_671_begin_0, end = var_671_end_0, end_mask = var_671_end_mask_0, x = value_1_cast_fp16)[name = string("op_671_cast_fp16")]; + tensor var_675_begin_0 = const()[name = string("op_675_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_675_end_0 = const()[name = string("op_675_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_675_end_mask_0 = const()[name = string("op_675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_675_cast_fp16 = slice_by_index(begin = var_675_begin_0, end = var_675_end_0, end_mask = var_675_end_mask_0, x = value_1_cast_fp16)[name = string("op_675_cast_fp16")]; + tensor var_679_begin_0 = const()[name = string("op_679_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_679_end_0 = const()[name = string("op_679_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_679_end_mask_0 = const()[name = string("op_679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_679_cast_fp16 = slice_by_index(begin = var_679_begin_0, end = var_679_end_0, end_mask = var_679_end_mask_0, x = value_1_cast_fp16)[name = string("op_679_cast_fp16")]; + tensor var_683_begin_0 = const()[name = string("op_683_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_683_end_0 = const()[name = string("op_683_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_683_end_mask_0 = const()[name = string("op_683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_683_cast_fp16 = slice_by_index(begin = var_683_begin_0, end = var_683_end_0, end_mask = var_683_end_mask_0, x = value_1_cast_fp16)[name = string("op_683_cast_fp16")]; + tensor var_687_begin_0 = const()[name = string("op_687_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_687_end_0 = const()[name = string("op_687_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_687_end_mask_0 = const()[name = string("op_687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_687_cast_fp16 = slice_by_index(begin = var_687_begin_0, end = var_687_end_0, end_mask = var_687_end_mask_0, x = value_1_cast_fp16)[name = string("op_687_cast_fp16")]; + tensor var_691_begin_0 = const()[name = string("op_691_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_691_end_0 = const()[name = string("op_691_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_691_end_mask_0 = const()[name = string("op_691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_691_cast_fp16 = slice_by_index(begin = var_691_begin_0, end = var_691_end_0, end_mask = var_691_end_mask_0, x = value_1_cast_fp16)[name = string("op_691_cast_fp16")]; + tensor var_695_begin_0 = const()[name = string("op_695_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_695_end_0 = const()[name = string("op_695_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_695_end_mask_0 = const()[name = string("op_695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_695_cast_fp16 = slice_by_index(begin = var_695_begin_0, end = var_695_end_0, end_mask = var_695_end_mask_0, x = value_1_cast_fp16)[name = string("op_695_cast_fp16")]; + tensor var_699_begin_0 = const()[name = string("op_699_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_699_end_0 = const()[name = string("op_699_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_699_end_mask_0 = const()[name = string("op_699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_699_cast_fp16 = slice_by_index(begin = var_699_begin_0, end = var_699_end_0, end_mask = var_699_end_mask_0, x = value_1_cast_fp16)[name = string("op_699_cast_fp16")]; + tensor var_703_begin_0 = const()[name = string("op_703_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_703_end_0 = const()[name = string("op_703_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_703_end_mask_0 = const()[name = string("op_703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_703_cast_fp16 = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = value_1_cast_fp16)[name = string("op_703_cast_fp16")]; + tensor var_707_begin_0 = const()[name = string("op_707_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_707_end_0 = const()[name = string("op_707_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_707_end_mask_0 = const()[name = string("op_707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = value_1_cast_fp16)[name = string("op_707_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_617_cast_fp16, var_283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_617_cast_fp16, var_290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_617_cast_fp16, var_297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_617_cast_fp16, var_304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_621_cast_fp16, var_311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_621_cast_fp16, var_318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_621_cast_fp16, var_325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_621_cast_fp16, var_332_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_625_cast_fp16, var_339_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_625_cast_fp16, var_346_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_625_cast_fp16, var_353_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_625_cast_fp16, var_360_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_629_cast_fp16, var_367_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_629_cast_fp16, var_374_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_629_cast_fp16, var_381_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_629_cast_fp16, var_388_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_633_cast_fp16, var_395_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_633_cast_fp16, var_402_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_633_cast_fp16, var_409_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_633_cast_fp16, var_416_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_637_cast_fp16, var_423_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_637_cast_fp16, var_430_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_637_cast_fp16, var_437_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_637_cast_fp16, var_444_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_641_cast_fp16, var_451_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_641_cast_fp16, var_458_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_641_cast_fp16, var_465_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_641_cast_fp16, var_472_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_645_cast_fp16, var_479_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_645_cast_fp16, var_486_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_645_cast_fp16, var_493_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_645_cast_fp16, var_500_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_649_cast_fp16, var_507_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_649_cast_fp16, var_514_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_649_cast_fp16, var_521_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_649_cast_fp16, var_528_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_653_cast_fp16, var_535_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_653_cast_fp16, var_542_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_653_cast_fp16, var_549_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_653_cast_fp16, var_556_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_657_cast_fp16, var_563_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_657_cast_fp16, var_570_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_657_cast_fp16, var_577_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_657_cast_fp16, var_584_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_661_cast_fp16, var_591_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_661_cast_fp16, var_598_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_661_cast_fp16, var_605_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_661_cast_fp16, var_612_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + fp16 var_806_to_fp16 = const()[name = string("op_806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_806_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_808_to_fp16 = const()[name = string("op_808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_808_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_810_to_fp16 = const()[name = string("op_810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_810_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_812_to_fp16 = const()[name = string("op_812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_812_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_814_to_fp16 = const()[name = string("op_814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_814_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_816_to_fp16 = const()[name = string("op_816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_816_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_818_to_fp16 = const()[name = string("op_818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_818_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_820_to_fp16 = const()[name = string("op_820_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_820_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_822_to_fp16 = const()[name = string("op_822_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_822_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_824_to_fp16 = const()[name = string("op_824_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_824_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_826_to_fp16 = const()[name = string("op_826_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_826_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_828_to_fp16 = const()[name = string("op_828_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_828_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_830_to_fp16 = const()[name = string("op_830_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_830_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_832_to_fp16 = const()[name = string("op_832_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_832_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_834_to_fp16 = const()[name = string("op_834_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_834_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_836_to_fp16 = const()[name = string("op_836_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_836_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_838_to_fp16 = const()[name = string("op_838_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_838_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_840_to_fp16 = const()[name = string("op_840_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_840_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_842_to_fp16 = const()[name = string("op_842_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_842_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_844_to_fp16 = const()[name = string("op_844_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_844_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_846_to_fp16 = const()[name = string("op_846_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_846_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_848_to_fp16 = const()[name = string("op_848_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_848_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_850_to_fp16 = const()[name = string("op_850_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_850_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_852_to_fp16 = const()[name = string("op_852_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_852_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_854_to_fp16 = const()[name = string("op_854_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_854_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_856_to_fp16 = const()[name = string("op_856_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_856_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_858_to_fp16 = const()[name = string("op_858_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_858_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_860_to_fp16 = const()[name = string("op_860_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_860_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_862_to_fp16 = const()[name = string("op_862_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_862_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_864_to_fp16 = const()[name = string("op_864_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_864_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_866_to_fp16 = const()[name = string("op_866_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_866_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_868_to_fp16 = const()[name = string("op_868_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_868_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + fp16 var_870_to_fp16 = const()[name = string("op_870_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_870_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_872_to_fp16 = const()[name = string("op_872_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_872_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_874_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_876_to_fp16 = const()[name = string("op_876_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_876_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_878_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_880_to_fp16 = const()[name = string("op_880_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_880_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_882_to_fp16 = const()[name = string("op_882_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_882_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_884_to_fp16 = const()[name = string("op_884_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_884_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_886_to_fp16 = const()[name = string("op_886_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_886_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_888_to_fp16 = const()[name = string("op_888_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_888_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_890_to_fp16 = const()[name = string("op_890_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_890_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_892_to_fp16 = const()[name = string("op_892_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_892_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_894_to_fp16 = const()[name = string("op_894_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_894_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_896_to_fp16 = const()[name = string("op_896_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_896_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_898_to_fp16 = const()[name = string("op_898_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_898_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_900_to_fp16 = const()[name = string("op_900_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_900_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + tensor var_902_cast_fp16 = softmax(axis = var_175, x = aw_chunk_1_cast_fp16)[name = string("op_902_cast_fp16")]; + tensor var_903_cast_fp16 = softmax(axis = var_175, x = aw_chunk_3_cast_fp16)[name = string("op_903_cast_fp16")]; + tensor var_904_cast_fp16 = softmax(axis = var_175, x = aw_chunk_5_cast_fp16)[name = string("op_904_cast_fp16")]; + tensor var_905_cast_fp16 = softmax(axis = var_175, x = aw_chunk_7_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_906_cast_fp16 = softmax(axis = var_175, x = aw_chunk_9_cast_fp16)[name = string("op_906_cast_fp16")]; + tensor var_907_cast_fp16 = softmax(axis = var_175, x = aw_chunk_11_cast_fp16)[name = string("op_907_cast_fp16")]; + tensor var_908_cast_fp16 = softmax(axis = var_175, x = aw_chunk_13_cast_fp16)[name = string("op_908_cast_fp16")]; + tensor var_909_cast_fp16 = softmax(axis = var_175, x = aw_chunk_15_cast_fp16)[name = string("op_909_cast_fp16")]; + tensor var_910_cast_fp16 = softmax(axis = var_175, x = aw_chunk_17_cast_fp16)[name = string("op_910_cast_fp16")]; + tensor var_911_cast_fp16 = softmax(axis = var_175, x = aw_chunk_19_cast_fp16)[name = string("op_911_cast_fp16")]; + tensor var_912_cast_fp16 = softmax(axis = var_175, x = aw_chunk_21_cast_fp16)[name = string("op_912_cast_fp16")]; + tensor var_913_cast_fp16 = softmax(axis = var_175, x = aw_chunk_23_cast_fp16)[name = string("op_913_cast_fp16")]; + tensor var_914_cast_fp16 = softmax(axis = var_175, x = aw_chunk_25_cast_fp16)[name = string("op_914_cast_fp16")]; + tensor var_915_cast_fp16 = softmax(axis = var_175, x = aw_chunk_27_cast_fp16)[name = string("op_915_cast_fp16")]; + tensor var_916_cast_fp16 = softmax(axis = var_175, x = aw_chunk_29_cast_fp16)[name = string("op_916_cast_fp16")]; + tensor var_917_cast_fp16 = softmax(axis = var_175, x = aw_chunk_31_cast_fp16)[name = string("op_917_cast_fp16")]; + tensor var_918_cast_fp16 = softmax(axis = var_175, x = aw_chunk_33_cast_fp16)[name = string("op_918_cast_fp16")]; + tensor var_919_cast_fp16 = softmax(axis = var_175, x = aw_chunk_35_cast_fp16)[name = string("op_919_cast_fp16")]; + tensor var_920_cast_fp16 = softmax(axis = var_175, x = aw_chunk_37_cast_fp16)[name = string("op_920_cast_fp16")]; + tensor var_921_cast_fp16 = softmax(axis = var_175, x = aw_chunk_39_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_922_cast_fp16 = softmax(axis = var_175, x = aw_chunk_41_cast_fp16)[name = string("op_922_cast_fp16")]; + tensor var_923_cast_fp16 = softmax(axis = var_175, x = aw_chunk_43_cast_fp16)[name = string("op_923_cast_fp16")]; + tensor var_924_cast_fp16 = softmax(axis = var_175, x = aw_chunk_45_cast_fp16)[name = string("op_924_cast_fp16")]; + tensor var_925_cast_fp16 = softmax(axis = var_175, x = aw_chunk_47_cast_fp16)[name = string("op_925_cast_fp16")]; + tensor var_926_cast_fp16 = softmax(axis = var_175, x = aw_chunk_49_cast_fp16)[name = string("op_926_cast_fp16")]; + tensor var_927_cast_fp16 = softmax(axis = var_175, x = aw_chunk_51_cast_fp16)[name = string("op_927_cast_fp16")]; + tensor var_928_cast_fp16 = softmax(axis = var_175, x = aw_chunk_53_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_929_cast_fp16 = softmax(axis = var_175, x = aw_chunk_55_cast_fp16)[name = string("op_929_cast_fp16")]; + tensor var_930_cast_fp16 = softmax(axis = var_175, x = aw_chunk_57_cast_fp16)[name = string("op_930_cast_fp16")]; + tensor var_931_cast_fp16 = softmax(axis = var_175, x = aw_chunk_59_cast_fp16)[name = string("op_931_cast_fp16")]; + tensor var_932_cast_fp16 = softmax(axis = var_175, x = aw_chunk_61_cast_fp16)[name = string("op_932_cast_fp16")]; + tensor var_933_cast_fp16 = softmax(axis = var_175, x = aw_chunk_63_cast_fp16)[name = string("op_933_cast_fp16")]; + tensor var_934_cast_fp16 = softmax(axis = var_175, x = aw_chunk_65_cast_fp16)[name = string("op_934_cast_fp16")]; + tensor var_935_cast_fp16 = softmax(axis = var_175, x = aw_chunk_67_cast_fp16)[name = string("op_935_cast_fp16")]; + tensor var_936_cast_fp16 = softmax(axis = var_175, x = aw_chunk_69_cast_fp16)[name = string("op_936_cast_fp16")]; + tensor var_937_cast_fp16 = softmax(axis = var_175, x = aw_chunk_71_cast_fp16)[name = string("op_937_cast_fp16")]; + tensor var_938_cast_fp16 = softmax(axis = var_175, x = aw_chunk_73_cast_fp16)[name = string("op_938_cast_fp16")]; + tensor var_939_cast_fp16 = softmax(axis = var_175, x = aw_chunk_75_cast_fp16)[name = string("op_939_cast_fp16")]; + tensor var_940_cast_fp16 = softmax(axis = var_175, x = aw_chunk_77_cast_fp16)[name = string("op_940_cast_fp16")]; + tensor var_941_cast_fp16 = softmax(axis = var_175, x = aw_chunk_79_cast_fp16)[name = string("op_941_cast_fp16")]; + tensor var_942_cast_fp16 = softmax(axis = var_175, x = aw_chunk_81_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_943_cast_fp16 = softmax(axis = var_175, x = aw_chunk_83_cast_fp16)[name = string("op_943_cast_fp16")]; + tensor var_944_cast_fp16 = softmax(axis = var_175, x = aw_chunk_85_cast_fp16)[name = string("op_944_cast_fp16")]; + tensor var_945_cast_fp16 = softmax(axis = var_175, x = aw_chunk_87_cast_fp16)[name = string("op_945_cast_fp16")]; + tensor var_946_cast_fp16 = softmax(axis = var_175, x = aw_chunk_89_cast_fp16)[name = string("op_946_cast_fp16")]; + tensor var_947_cast_fp16 = softmax(axis = var_175, x = aw_chunk_91_cast_fp16)[name = string("op_947_cast_fp16")]; + tensor var_948_cast_fp16 = softmax(axis = var_175, x = aw_chunk_93_cast_fp16)[name = string("op_948_cast_fp16")]; + tensor var_949_cast_fp16 = softmax(axis = var_175, x = aw_chunk_95_cast_fp16)[name = string("op_949_cast_fp16")]; + string var_951_equation_0 = const()[name = string("op_951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_951_cast_fp16 = einsum(equation = var_951_equation_0, values = (var_663_cast_fp16, var_902_cast_fp16))[name = string("op_951_cast_fp16")]; + string var_953_equation_0 = const()[name = string("op_953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_953_cast_fp16 = einsum(equation = var_953_equation_0, values = (var_663_cast_fp16, var_903_cast_fp16))[name = string("op_953_cast_fp16")]; + string var_955_equation_0 = const()[name = string("op_955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_955_cast_fp16 = einsum(equation = var_955_equation_0, values = (var_663_cast_fp16, var_904_cast_fp16))[name = string("op_955_cast_fp16")]; + string var_957_equation_0 = const()[name = string("op_957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_957_cast_fp16 = einsum(equation = var_957_equation_0, values = (var_663_cast_fp16, var_905_cast_fp16))[name = string("op_957_cast_fp16")]; + string var_959_equation_0 = const()[name = string("op_959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_959_cast_fp16 = einsum(equation = var_959_equation_0, values = (var_667_cast_fp16, var_906_cast_fp16))[name = string("op_959_cast_fp16")]; + string var_961_equation_0 = const()[name = string("op_961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_961_cast_fp16 = einsum(equation = var_961_equation_0, values = (var_667_cast_fp16, var_907_cast_fp16))[name = string("op_961_cast_fp16")]; + string var_963_equation_0 = const()[name = string("op_963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_963_cast_fp16 = einsum(equation = var_963_equation_0, values = (var_667_cast_fp16, var_908_cast_fp16))[name = string("op_963_cast_fp16")]; + string var_965_equation_0 = const()[name = string("op_965_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_965_cast_fp16 = einsum(equation = var_965_equation_0, values = (var_667_cast_fp16, var_909_cast_fp16))[name = string("op_965_cast_fp16")]; + string var_967_equation_0 = const()[name = string("op_967_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_967_cast_fp16 = einsum(equation = var_967_equation_0, values = (var_671_cast_fp16, var_910_cast_fp16))[name = string("op_967_cast_fp16")]; + string var_969_equation_0 = const()[name = string("op_969_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_969_cast_fp16 = einsum(equation = var_969_equation_0, values = (var_671_cast_fp16, var_911_cast_fp16))[name = string("op_969_cast_fp16")]; + string var_971_equation_0 = const()[name = string("op_971_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_971_cast_fp16 = einsum(equation = var_971_equation_0, values = (var_671_cast_fp16, var_912_cast_fp16))[name = string("op_971_cast_fp16")]; + string var_973_equation_0 = const()[name = string("op_973_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_973_cast_fp16 = einsum(equation = var_973_equation_0, values = (var_671_cast_fp16, var_913_cast_fp16))[name = string("op_973_cast_fp16")]; + string var_975_equation_0 = const()[name = string("op_975_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_975_cast_fp16 = einsum(equation = var_975_equation_0, values = (var_675_cast_fp16, var_914_cast_fp16))[name = string("op_975_cast_fp16")]; + string var_977_equation_0 = const()[name = string("op_977_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_977_cast_fp16 = einsum(equation = var_977_equation_0, values = (var_675_cast_fp16, var_915_cast_fp16))[name = string("op_977_cast_fp16")]; + string var_979_equation_0 = const()[name = string("op_979_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_979_cast_fp16 = einsum(equation = var_979_equation_0, values = (var_675_cast_fp16, var_916_cast_fp16))[name = string("op_979_cast_fp16")]; + string var_981_equation_0 = const()[name = string("op_981_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_981_cast_fp16 = einsum(equation = var_981_equation_0, values = (var_675_cast_fp16, var_917_cast_fp16))[name = string("op_981_cast_fp16")]; + string var_983_equation_0 = const()[name = string("op_983_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_983_cast_fp16 = einsum(equation = var_983_equation_0, values = (var_679_cast_fp16, var_918_cast_fp16))[name = string("op_983_cast_fp16")]; + string var_985_equation_0 = const()[name = string("op_985_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_985_cast_fp16 = einsum(equation = var_985_equation_0, values = (var_679_cast_fp16, var_919_cast_fp16))[name = string("op_985_cast_fp16")]; + string var_987_equation_0 = const()[name = string("op_987_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_987_cast_fp16 = einsum(equation = var_987_equation_0, values = (var_679_cast_fp16, var_920_cast_fp16))[name = string("op_987_cast_fp16")]; + string var_989_equation_0 = const()[name = string("op_989_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_989_cast_fp16 = einsum(equation = var_989_equation_0, values = (var_679_cast_fp16, var_921_cast_fp16))[name = string("op_989_cast_fp16")]; + string var_991_equation_0 = const()[name = string("op_991_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_991_cast_fp16 = einsum(equation = var_991_equation_0, values = (var_683_cast_fp16, var_922_cast_fp16))[name = string("op_991_cast_fp16")]; + string var_993_equation_0 = const()[name = string("op_993_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_993_cast_fp16 = einsum(equation = var_993_equation_0, values = (var_683_cast_fp16, var_923_cast_fp16))[name = string("op_993_cast_fp16")]; + string var_995_equation_0 = const()[name = string("op_995_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_995_cast_fp16 = einsum(equation = var_995_equation_0, values = (var_683_cast_fp16, var_924_cast_fp16))[name = string("op_995_cast_fp16")]; + string var_997_equation_0 = const()[name = string("op_997_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_997_cast_fp16 = einsum(equation = var_997_equation_0, values = (var_683_cast_fp16, var_925_cast_fp16))[name = string("op_997_cast_fp16")]; + string var_999_equation_0 = const()[name = string("op_999_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_999_cast_fp16 = einsum(equation = var_999_equation_0, values = (var_687_cast_fp16, var_926_cast_fp16))[name = string("op_999_cast_fp16")]; + string var_1001_equation_0 = const()[name = string("op_1001_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1001_cast_fp16 = einsum(equation = var_1001_equation_0, values = (var_687_cast_fp16, var_927_cast_fp16))[name = string("op_1001_cast_fp16")]; + string var_1003_equation_0 = const()[name = string("op_1003_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1003_cast_fp16 = einsum(equation = var_1003_equation_0, values = (var_687_cast_fp16, var_928_cast_fp16))[name = string("op_1003_cast_fp16")]; + string var_1005_equation_0 = const()[name = string("op_1005_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1005_cast_fp16 = einsum(equation = var_1005_equation_0, values = (var_687_cast_fp16, var_929_cast_fp16))[name = string("op_1005_cast_fp16")]; + string var_1007_equation_0 = const()[name = string("op_1007_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1007_cast_fp16 = einsum(equation = var_1007_equation_0, values = (var_691_cast_fp16, var_930_cast_fp16))[name = string("op_1007_cast_fp16")]; + string var_1009_equation_0 = const()[name = string("op_1009_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1009_cast_fp16 = einsum(equation = var_1009_equation_0, values = (var_691_cast_fp16, var_931_cast_fp16))[name = string("op_1009_cast_fp16")]; + string var_1011_equation_0 = const()[name = string("op_1011_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1011_cast_fp16 = einsum(equation = var_1011_equation_0, values = (var_691_cast_fp16, var_932_cast_fp16))[name = string("op_1011_cast_fp16")]; + string var_1013_equation_0 = const()[name = string("op_1013_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1013_cast_fp16 = einsum(equation = var_1013_equation_0, values = (var_691_cast_fp16, var_933_cast_fp16))[name = string("op_1013_cast_fp16")]; + string var_1015_equation_0 = const()[name = string("op_1015_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1015_cast_fp16 = einsum(equation = var_1015_equation_0, values = (var_695_cast_fp16, var_934_cast_fp16))[name = string("op_1015_cast_fp16")]; + string var_1017_equation_0 = const()[name = string("op_1017_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1017_cast_fp16 = einsum(equation = var_1017_equation_0, values = (var_695_cast_fp16, var_935_cast_fp16))[name = string("op_1017_cast_fp16")]; + string var_1019_equation_0 = const()[name = string("op_1019_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1019_cast_fp16 = einsum(equation = var_1019_equation_0, values = (var_695_cast_fp16, var_936_cast_fp16))[name = string("op_1019_cast_fp16")]; + string var_1021_equation_0 = const()[name = string("op_1021_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1021_cast_fp16 = einsum(equation = var_1021_equation_0, values = (var_695_cast_fp16, var_937_cast_fp16))[name = string("op_1021_cast_fp16")]; + string var_1023_equation_0 = const()[name = string("op_1023_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1023_cast_fp16 = einsum(equation = var_1023_equation_0, values = (var_699_cast_fp16, var_938_cast_fp16))[name = string("op_1023_cast_fp16")]; + string var_1025_equation_0 = const()[name = string("op_1025_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1025_cast_fp16 = einsum(equation = var_1025_equation_0, values = (var_699_cast_fp16, var_939_cast_fp16))[name = string("op_1025_cast_fp16")]; + string var_1027_equation_0 = const()[name = string("op_1027_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1027_cast_fp16 = einsum(equation = var_1027_equation_0, values = (var_699_cast_fp16, var_940_cast_fp16))[name = string("op_1027_cast_fp16")]; + string var_1029_equation_0 = const()[name = string("op_1029_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1029_cast_fp16 = einsum(equation = var_1029_equation_0, values = (var_699_cast_fp16, var_941_cast_fp16))[name = string("op_1029_cast_fp16")]; + string var_1031_equation_0 = const()[name = string("op_1031_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1031_cast_fp16 = einsum(equation = var_1031_equation_0, values = (var_703_cast_fp16, var_942_cast_fp16))[name = string("op_1031_cast_fp16")]; + string var_1033_equation_0 = const()[name = string("op_1033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1033_cast_fp16 = einsum(equation = var_1033_equation_0, values = (var_703_cast_fp16, var_943_cast_fp16))[name = string("op_1033_cast_fp16")]; + string var_1035_equation_0 = const()[name = string("op_1035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1035_cast_fp16 = einsum(equation = var_1035_equation_0, values = (var_703_cast_fp16, var_944_cast_fp16))[name = string("op_1035_cast_fp16")]; + string var_1037_equation_0 = const()[name = string("op_1037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1037_cast_fp16 = einsum(equation = var_1037_equation_0, values = (var_703_cast_fp16, var_945_cast_fp16))[name = string("op_1037_cast_fp16")]; + string var_1039_equation_0 = const()[name = string("op_1039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1039_cast_fp16 = einsum(equation = var_1039_equation_0, values = (var_707_cast_fp16, var_946_cast_fp16))[name = string("op_1039_cast_fp16")]; + string var_1041_equation_0 = const()[name = string("op_1041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1041_cast_fp16 = einsum(equation = var_1041_equation_0, values = (var_707_cast_fp16, var_947_cast_fp16))[name = string("op_1041_cast_fp16")]; + string var_1043_equation_0 = const()[name = string("op_1043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1043_cast_fp16 = einsum(equation = var_1043_equation_0, values = (var_707_cast_fp16, var_948_cast_fp16))[name = string("op_1043_cast_fp16")]; + string var_1045_equation_0 = const()[name = string("op_1045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1045_cast_fp16 = einsum(equation = var_1045_equation_0, values = (var_707_cast_fp16, var_949_cast_fp16))[name = string("op_1045_cast_fp16")]; + bool var_1047_interleave_0 = const()[name = string("op_1047_interleave_0"), val = bool(false)]; + tensor var_1047_cast_fp16 = concat(axis = var_158, interleave = var_1047_interleave_0, values = (var_951_cast_fp16, var_953_cast_fp16, var_955_cast_fp16, var_957_cast_fp16))[name = string("op_1047_cast_fp16")]; + bool var_1049_interleave_0 = const()[name = string("op_1049_interleave_0"), val = bool(false)]; + tensor var_1049_cast_fp16 = concat(axis = var_158, interleave = var_1049_interleave_0, values = (var_959_cast_fp16, var_961_cast_fp16, var_963_cast_fp16, var_965_cast_fp16))[name = string("op_1049_cast_fp16")]; + bool var_1051_interleave_0 = const()[name = string("op_1051_interleave_0"), val = bool(false)]; + tensor var_1051_cast_fp16 = concat(axis = var_158, interleave = var_1051_interleave_0, values = (var_967_cast_fp16, var_969_cast_fp16, var_971_cast_fp16, var_973_cast_fp16))[name = string("op_1051_cast_fp16")]; + bool var_1053_interleave_0 = const()[name = string("op_1053_interleave_0"), val = bool(false)]; + tensor var_1053_cast_fp16 = concat(axis = var_158, interleave = var_1053_interleave_0, values = (var_975_cast_fp16, var_977_cast_fp16, var_979_cast_fp16, var_981_cast_fp16))[name = string("op_1053_cast_fp16")]; + bool var_1055_interleave_0 = const()[name = string("op_1055_interleave_0"), val = bool(false)]; + tensor var_1055_cast_fp16 = concat(axis = var_158, interleave = var_1055_interleave_0, values = (var_983_cast_fp16, var_985_cast_fp16, var_987_cast_fp16, var_989_cast_fp16))[name = string("op_1055_cast_fp16")]; + bool var_1057_interleave_0 = const()[name = string("op_1057_interleave_0"), val = bool(false)]; + tensor var_1057_cast_fp16 = concat(axis = var_158, interleave = var_1057_interleave_0, values = (var_991_cast_fp16, var_993_cast_fp16, var_995_cast_fp16, var_997_cast_fp16))[name = string("op_1057_cast_fp16")]; + bool var_1059_interleave_0 = const()[name = string("op_1059_interleave_0"), val = bool(false)]; + tensor var_1059_cast_fp16 = concat(axis = var_158, interleave = var_1059_interleave_0, values = (var_999_cast_fp16, var_1001_cast_fp16, var_1003_cast_fp16, var_1005_cast_fp16))[name = string("op_1059_cast_fp16")]; + bool var_1061_interleave_0 = const()[name = string("op_1061_interleave_0"), val = bool(false)]; + tensor var_1061_cast_fp16 = concat(axis = var_158, interleave = var_1061_interleave_0, values = (var_1007_cast_fp16, var_1009_cast_fp16, var_1011_cast_fp16, var_1013_cast_fp16))[name = string("op_1061_cast_fp16")]; + bool var_1063_interleave_0 = const()[name = string("op_1063_interleave_0"), val = bool(false)]; + tensor var_1063_cast_fp16 = concat(axis = var_158, interleave = var_1063_interleave_0, values = (var_1015_cast_fp16, var_1017_cast_fp16, var_1019_cast_fp16, var_1021_cast_fp16))[name = string("op_1063_cast_fp16")]; + bool var_1065_interleave_0 = const()[name = string("op_1065_interleave_0"), val = bool(false)]; + tensor var_1065_cast_fp16 = concat(axis = var_158, interleave = var_1065_interleave_0, values = (var_1023_cast_fp16, var_1025_cast_fp16, var_1027_cast_fp16, var_1029_cast_fp16))[name = string("op_1065_cast_fp16")]; + bool var_1067_interleave_0 = const()[name = string("op_1067_interleave_0"), val = bool(false)]; + tensor var_1067_cast_fp16 = concat(axis = var_158, interleave = var_1067_interleave_0, values = (var_1031_cast_fp16, var_1033_cast_fp16, var_1035_cast_fp16, var_1037_cast_fp16))[name = string("op_1067_cast_fp16")]; + bool var_1069_interleave_0 = const()[name = string("op_1069_interleave_0"), val = bool(false)]; + tensor var_1069_cast_fp16 = concat(axis = var_158, interleave = var_1069_interleave_0, values = (var_1039_cast_fp16, var_1041_cast_fp16, var_1043_cast_fp16, var_1045_cast_fp16))[name = string("op_1069_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_175, interleave = input_1_interleave_0, values = (var_1047_cast_fp16, var_1049_cast_fp16, var_1051_cast_fp16, var_1053_cast_fp16, var_1055_cast_fp16, var_1057_cast_fp16, var_1059_cast_fp16, var_1061_cast_fp16, var_1063_cast_fp16, var_1065_cast_fp16, var_1067_cast_fp16, var_1069_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9763776)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10943488)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_1088_to_fp16 = const()[name = string("op_1088_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_1088_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10945088)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10946688)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10948288)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15666944)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15673152)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20391808)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_1117 = const()[name = string("op_1117"), val = int32(3)]; + int32 var_1134 = const()[name = string("op_1134"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_1151_to_fp16 = const()[name = string("op_1151_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_1151_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20393408)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20395008)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20396608)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21576320)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21577920)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22757632)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937344)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_1189_begin_0 = const()[name = string("op_1189_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1189_end_0 = const()[name = string("op_1189_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1189_end_mask_0 = const()[name = string("op_1189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1189_cast_fp16 = slice_by_index(begin = var_1189_begin_0, end = var_1189_end_0, end_mask = var_1189_end_mask_0, x = query_3_cast_fp16)[name = string("op_1189_cast_fp16")]; + tensor var_1193_begin_0 = const()[name = string("op_1193_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1193_end_0 = const()[name = string("op_1193_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1193_end_mask_0 = const()[name = string("op_1193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1193_cast_fp16 = slice_by_index(begin = var_1193_begin_0, end = var_1193_end_0, end_mask = var_1193_end_mask_0, x = query_3_cast_fp16)[name = string("op_1193_cast_fp16")]; + tensor var_1197_begin_0 = const()[name = string("op_1197_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1197_end_0 = const()[name = string("op_1197_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1197_end_mask_0 = const()[name = string("op_1197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1197_cast_fp16 = slice_by_index(begin = var_1197_begin_0, end = var_1197_end_0, end_mask = var_1197_end_mask_0, x = query_3_cast_fp16)[name = string("op_1197_cast_fp16")]; + tensor var_1201_begin_0 = const()[name = string("op_1201_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1201_end_0 = const()[name = string("op_1201_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1201_end_mask_0 = const()[name = string("op_1201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1201_cast_fp16 = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = query_3_cast_fp16)[name = string("op_1201_cast_fp16")]; + tensor var_1205_begin_0 = const()[name = string("op_1205_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1205_end_0 = const()[name = string("op_1205_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1205_end_mask_0 = const()[name = string("op_1205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1205_cast_fp16 = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, x = query_3_cast_fp16)[name = string("op_1205_cast_fp16")]; + tensor var_1209_begin_0 = const()[name = string("op_1209_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1209_end_0 = const()[name = string("op_1209_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1209_end_mask_0 = const()[name = string("op_1209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1209_cast_fp16 = slice_by_index(begin = var_1209_begin_0, end = var_1209_end_0, end_mask = var_1209_end_mask_0, x = query_3_cast_fp16)[name = string("op_1209_cast_fp16")]; + tensor var_1213_begin_0 = const()[name = string("op_1213_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1213_end_0 = const()[name = string("op_1213_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1213_end_mask_0 = const()[name = string("op_1213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1213_cast_fp16 = slice_by_index(begin = var_1213_begin_0, end = var_1213_end_0, end_mask = var_1213_end_mask_0, x = query_3_cast_fp16)[name = string("op_1213_cast_fp16")]; + tensor var_1217_begin_0 = const()[name = string("op_1217_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1217_end_0 = const()[name = string("op_1217_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1217_end_mask_0 = const()[name = string("op_1217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1217_cast_fp16 = slice_by_index(begin = var_1217_begin_0, end = var_1217_end_0, end_mask = var_1217_end_mask_0, x = query_3_cast_fp16)[name = string("op_1217_cast_fp16")]; + tensor var_1221_begin_0 = const()[name = string("op_1221_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1221_end_0 = const()[name = string("op_1221_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1221_end_mask_0 = const()[name = string("op_1221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1221_cast_fp16 = slice_by_index(begin = var_1221_begin_0, end = var_1221_end_0, end_mask = var_1221_end_mask_0, x = query_3_cast_fp16)[name = string("op_1221_cast_fp16")]; + tensor var_1225_begin_0 = const()[name = string("op_1225_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1225_end_0 = const()[name = string("op_1225_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1225_end_mask_0 = const()[name = string("op_1225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1225_cast_fp16 = slice_by_index(begin = var_1225_begin_0, end = var_1225_end_0, end_mask = var_1225_end_mask_0, x = query_3_cast_fp16)[name = string("op_1225_cast_fp16")]; + tensor var_1229_begin_0 = const()[name = string("op_1229_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1229_end_0 = const()[name = string("op_1229_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1229_end_mask_0 = const()[name = string("op_1229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1229_cast_fp16 = slice_by_index(begin = var_1229_begin_0, end = var_1229_end_0, end_mask = var_1229_end_mask_0, x = query_3_cast_fp16)[name = string("op_1229_cast_fp16")]; + tensor var_1233_begin_0 = const()[name = string("op_1233_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1233_end_0 = const()[name = string("op_1233_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1233_end_mask_0 = const()[name = string("op_1233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1233_cast_fp16 = slice_by_index(begin = var_1233_begin_0, end = var_1233_end_0, end_mask = var_1233_end_mask_0, x = query_3_cast_fp16)[name = string("op_1233_cast_fp16")]; + tensor var_1242_begin_0 = const()[name = string("op_1242_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1242_end_0 = const()[name = string("op_1242_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1242_end_mask_0 = const()[name = string("op_1242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1242_cast_fp16 = slice_by_index(begin = var_1242_begin_0, end = var_1242_end_0, end_mask = var_1242_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1242_cast_fp16")]; + tensor var_1249_begin_0 = const()[name = string("op_1249_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1249_end_0 = const()[name = string("op_1249_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1249_end_mask_0 = const()[name = string("op_1249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1249_cast_fp16 = slice_by_index(begin = var_1249_begin_0, end = var_1249_end_0, end_mask = var_1249_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1249_cast_fp16")]; + tensor var_1256_begin_0 = const()[name = string("op_1256_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1256_end_0 = const()[name = string("op_1256_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1256_end_mask_0 = const()[name = string("op_1256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1256_cast_fp16 = slice_by_index(begin = var_1256_begin_0, end = var_1256_end_0, end_mask = var_1256_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1256_cast_fp16")]; + tensor var_1263_begin_0 = const()[name = string("op_1263_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1263_end_0 = const()[name = string("op_1263_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1263_end_mask_0 = const()[name = string("op_1263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1263_cast_fp16 = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = var_1189_cast_fp16)[name = string("op_1263_cast_fp16")]; + tensor var_1270_begin_0 = const()[name = string("op_1270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1270_end_0 = const()[name = string("op_1270_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1270_end_mask_0 = const()[name = string("op_1270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1270_cast_fp16")]; + tensor var_1277_begin_0 = const()[name = string("op_1277_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1277_end_0 = const()[name = string("op_1277_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1277_end_mask_0 = const()[name = string("op_1277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1277_cast_fp16 = slice_by_index(begin = var_1277_begin_0, end = var_1277_end_0, end_mask = var_1277_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1277_cast_fp16")]; + tensor var_1284_begin_0 = const()[name = string("op_1284_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1284_end_0 = const()[name = string("op_1284_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1284_end_mask_0 = const()[name = string("op_1284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1284_cast_fp16 = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1284_cast_fp16")]; + tensor var_1291_begin_0 = const()[name = string("op_1291_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1291_end_0 = const()[name = string("op_1291_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1291_end_mask_0 = const()[name = string("op_1291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1291_cast_fp16 = slice_by_index(begin = var_1291_begin_0, end = var_1291_end_0, end_mask = var_1291_end_mask_0, x = var_1193_cast_fp16)[name = string("op_1291_cast_fp16")]; + tensor var_1298_begin_0 = const()[name = string("op_1298_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1298_end_0 = const()[name = string("op_1298_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1298_end_mask_0 = const()[name = string("op_1298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1298_cast_fp16 = slice_by_index(begin = var_1298_begin_0, end = var_1298_end_0, end_mask = var_1298_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1298_cast_fp16")]; + tensor var_1305_begin_0 = const()[name = string("op_1305_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1305_end_0 = const()[name = string("op_1305_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1305_end_mask_0 = const()[name = string("op_1305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1305_cast_fp16 = slice_by_index(begin = var_1305_begin_0, end = var_1305_end_0, end_mask = var_1305_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1305_cast_fp16")]; + tensor var_1312_begin_0 = const()[name = string("op_1312_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1312_end_0 = const()[name = string("op_1312_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1312_end_mask_0 = const()[name = string("op_1312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1312_cast_fp16 = slice_by_index(begin = var_1312_begin_0, end = var_1312_end_0, end_mask = var_1312_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1312_cast_fp16")]; + tensor var_1319_begin_0 = const()[name = string("op_1319_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1319_end_0 = const()[name = string("op_1319_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1319_end_mask_0 = const()[name = string("op_1319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1319_cast_fp16 = slice_by_index(begin = var_1319_begin_0, end = var_1319_end_0, end_mask = var_1319_end_mask_0, x = var_1197_cast_fp16)[name = string("op_1319_cast_fp16")]; + tensor var_1326_begin_0 = const()[name = string("op_1326_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1326_end_0 = const()[name = string("op_1326_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1326_end_mask_0 = const()[name = string("op_1326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1326_cast_fp16 = slice_by_index(begin = var_1326_begin_0, end = var_1326_end_0, end_mask = var_1326_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1326_cast_fp16")]; + tensor var_1333_begin_0 = const()[name = string("op_1333_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1333_end_0 = const()[name = string("op_1333_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1333_end_mask_0 = const()[name = string("op_1333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1333_cast_fp16 = slice_by_index(begin = var_1333_begin_0, end = var_1333_end_0, end_mask = var_1333_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1333_cast_fp16")]; + tensor var_1340_begin_0 = const()[name = string("op_1340_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1340_end_0 = const()[name = string("op_1340_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1340_end_mask_0 = const()[name = string("op_1340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1340_cast_fp16 = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1340_cast_fp16")]; + tensor var_1347_begin_0 = const()[name = string("op_1347_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1347_end_0 = const()[name = string("op_1347_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1347_end_mask_0 = const()[name = string("op_1347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1347_cast_fp16 = slice_by_index(begin = var_1347_begin_0, end = var_1347_end_0, end_mask = var_1347_end_mask_0, x = var_1201_cast_fp16)[name = string("op_1347_cast_fp16")]; + tensor var_1354_begin_0 = const()[name = string("op_1354_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1354_end_0 = const()[name = string("op_1354_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1354_end_mask_0 = const()[name = string("op_1354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1354_cast_fp16 = slice_by_index(begin = var_1354_begin_0, end = var_1354_end_0, end_mask = var_1354_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1354_cast_fp16")]; + tensor var_1361_begin_0 = const()[name = string("op_1361_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1361_end_0 = const()[name = string("op_1361_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1361_end_mask_0 = const()[name = string("op_1361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1361_cast_fp16 = slice_by_index(begin = var_1361_begin_0, end = var_1361_end_0, end_mask = var_1361_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1361_cast_fp16")]; + tensor var_1368_begin_0 = const()[name = string("op_1368_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1368_end_0 = const()[name = string("op_1368_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1368_end_mask_0 = const()[name = string("op_1368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1368_cast_fp16 = slice_by_index(begin = var_1368_begin_0, end = var_1368_end_0, end_mask = var_1368_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1368_cast_fp16")]; + tensor var_1375_begin_0 = const()[name = string("op_1375_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1375_end_0 = const()[name = string("op_1375_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1375_end_mask_0 = const()[name = string("op_1375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1375_cast_fp16 = slice_by_index(begin = var_1375_begin_0, end = var_1375_end_0, end_mask = var_1375_end_mask_0, x = var_1205_cast_fp16)[name = string("op_1375_cast_fp16")]; + tensor var_1382_begin_0 = const()[name = string("op_1382_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1382_end_0 = const()[name = string("op_1382_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1382_end_mask_0 = const()[name = string("op_1382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1382_cast_fp16 = slice_by_index(begin = var_1382_begin_0, end = var_1382_end_0, end_mask = var_1382_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1382_cast_fp16")]; + tensor var_1389_begin_0 = const()[name = string("op_1389_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1389_end_0 = const()[name = string("op_1389_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1389_end_mask_0 = const()[name = string("op_1389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1389_cast_fp16 = slice_by_index(begin = var_1389_begin_0, end = var_1389_end_0, end_mask = var_1389_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1389_cast_fp16")]; + tensor var_1396_begin_0 = const()[name = string("op_1396_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1396_end_0 = const()[name = string("op_1396_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1396_end_mask_0 = const()[name = string("op_1396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1396_cast_fp16 = slice_by_index(begin = var_1396_begin_0, end = var_1396_end_0, end_mask = var_1396_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1396_cast_fp16")]; + tensor var_1403_begin_0 = const()[name = string("op_1403_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1403_end_0 = const()[name = string("op_1403_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1403_end_mask_0 = const()[name = string("op_1403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1403_cast_fp16 = slice_by_index(begin = var_1403_begin_0, end = var_1403_end_0, end_mask = var_1403_end_mask_0, x = var_1209_cast_fp16)[name = string("op_1403_cast_fp16")]; + tensor var_1410_begin_0 = const()[name = string("op_1410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1410_end_0 = const()[name = string("op_1410_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1410_end_mask_0 = const()[name = string("op_1410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1410_cast_fp16 = slice_by_index(begin = var_1410_begin_0, end = var_1410_end_0, end_mask = var_1410_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1410_cast_fp16")]; + tensor var_1417_begin_0 = const()[name = string("op_1417_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1417_end_0 = const()[name = string("op_1417_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1417_end_mask_0 = const()[name = string("op_1417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1417_cast_fp16 = slice_by_index(begin = var_1417_begin_0, end = var_1417_end_0, end_mask = var_1417_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1417_cast_fp16")]; + tensor var_1424_begin_0 = const()[name = string("op_1424_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1424_end_0 = const()[name = string("op_1424_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1424_end_mask_0 = const()[name = string("op_1424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1424_cast_fp16 = slice_by_index(begin = var_1424_begin_0, end = var_1424_end_0, end_mask = var_1424_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1424_cast_fp16")]; + tensor var_1431_begin_0 = const()[name = string("op_1431_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1431_end_0 = const()[name = string("op_1431_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1431_end_mask_0 = const()[name = string("op_1431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1431_cast_fp16 = slice_by_index(begin = var_1431_begin_0, end = var_1431_end_0, end_mask = var_1431_end_mask_0, x = var_1213_cast_fp16)[name = string("op_1431_cast_fp16")]; + tensor var_1438_begin_0 = const()[name = string("op_1438_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1438_end_0 = const()[name = string("op_1438_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1438_end_mask_0 = const()[name = string("op_1438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1438_cast_fp16 = slice_by_index(begin = var_1438_begin_0, end = var_1438_end_0, end_mask = var_1438_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1438_cast_fp16")]; + tensor var_1445_begin_0 = const()[name = string("op_1445_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1445_end_0 = const()[name = string("op_1445_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1445_end_mask_0 = const()[name = string("op_1445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1445_cast_fp16 = slice_by_index(begin = var_1445_begin_0, end = var_1445_end_0, end_mask = var_1445_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1445_cast_fp16")]; + tensor var_1452_begin_0 = const()[name = string("op_1452_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1452_end_0 = const()[name = string("op_1452_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1452_end_mask_0 = const()[name = string("op_1452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1452_cast_fp16 = slice_by_index(begin = var_1452_begin_0, end = var_1452_end_0, end_mask = var_1452_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1452_cast_fp16")]; + tensor var_1459_begin_0 = const()[name = string("op_1459_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1459_end_0 = const()[name = string("op_1459_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1459_end_mask_0 = const()[name = string("op_1459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1459_cast_fp16 = slice_by_index(begin = var_1459_begin_0, end = var_1459_end_0, end_mask = var_1459_end_mask_0, x = var_1217_cast_fp16)[name = string("op_1459_cast_fp16")]; + tensor var_1466_begin_0 = const()[name = string("op_1466_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1466_end_0 = const()[name = string("op_1466_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1466_end_mask_0 = const()[name = string("op_1466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1466_cast_fp16 = slice_by_index(begin = var_1466_begin_0, end = var_1466_end_0, end_mask = var_1466_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1466_cast_fp16")]; + tensor var_1473_begin_0 = const()[name = string("op_1473_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1473_end_0 = const()[name = string("op_1473_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1473_end_mask_0 = const()[name = string("op_1473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1473_cast_fp16")]; + tensor var_1480_begin_0 = const()[name = string("op_1480_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1480_end_0 = const()[name = string("op_1480_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1480_end_mask_0 = const()[name = string("op_1480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1480_cast_fp16 = slice_by_index(begin = var_1480_begin_0, end = var_1480_end_0, end_mask = var_1480_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1480_cast_fp16")]; + tensor var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = var_1221_cast_fp16)[name = string("op_1487_cast_fp16")]; + tensor var_1494_begin_0 = const()[name = string("op_1494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1494_end_0 = const()[name = string("op_1494_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1494_end_mask_0 = const()[name = string("op_1494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1494_cast_fp16 = slice_by_index(begin = var_1494_begin_0, end = var_1494_end_0, end_mask = var_1494_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1494_cast_fp16")]; + tensor var_1501_begin_0 = const()[name = string("op_1501_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1501_end_0 = const()[name = string("op_1501_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1501_end_mask_0 = const()[name = string("op_1501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1501_cast_fp16 = slice_by_index(begin = var_1501_begin_0, end = var_1501_end_0, end_mask = var_1501_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1501_cast_fp16")]; + tensor var_1508_begin_0 = const()[name = string("op_1508_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1508_end_0 = const()[name = string("op_1508_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1508_end_mask_0 = const()[name = string("op_1508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1508_cast_fp16 = slice_by_index(begin = var_1508_begin_0, end = var_1508_end_0, end_mask = var_1508_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1508_cast_fp16")]; + tensor var_1515_begin_0 = const()[name = string("op_1515_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1515_end_0 = const()[name = string("op_1515_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1515_end_mask_0 = const()[name = string("op_1515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1515_cast_fp16 = slice_by_index(begin = var_1515_begin_0, end = var_1515_end_0, end_mask = var_1515_end_mask_0, x = var_1225_cast_fp16)[name = string("op_1515_cast_fp16")]; + tensor var_1522_begin_0 = const()[name = string("op_1522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1522_end_0 = const()[name = string("op_1522_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1522_end_mask_0 = const()[name = string("op_1522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1522_cast_fp16 = slice_by_index(begin = var_1522_begin_0, end = var_1522_end_0, end_mask = var_1522_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1522_cast_fp16")]; + tensor var_1529_begin_0 = const()[name = string("op_1529_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1529_end_0 = const()[name = string("op_1529_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1529_end_mask_0 = const()[name = string("op_1529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1529_cast_fp16 = slice_by_index(begin = var_1529_begin_0, end = var_1529_end_0, end_mask = var_1529_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1529_cast_fp16")]; + tensor var_1536_begin_0 = const()[name = string("op_1536_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1536_end_0 = const()[name = string("op_1536_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1536_end_mask_0 = const()[name = string("op_1536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1536_cast_fp16 = slice_by_index(begin = var_1536_begin_0, end = var_1536_end_0, end_mask = var_1536_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1536_cast_fp16")]; + tensor var_1543_begin_0 = const()[name = string("op_1543_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1543_end_0 = const()[name = string("op_1543_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1543_end_mask_0 = const()[name = string("op_1543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1543_cast_fp16 = slice_by_index(begin = var_1543_begin_0, end = var_1543_end_0, end_mask = var_1543_end_mask_0, x = var_1229_cast_fp16)[name = string("op_1543_cast_fp16")]; + tensor var_1550_begin_0 = const()[name = string("op_1550_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1550_end_0 = const()[name = string("op_1550_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1550_end_mask_0 = const()[name = string("op_1550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1550_cast_fp16 = slice_by_index(begin = var_1550_begin_0, end = var_1550_end_0, end_mask = var_1550_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1550_cast_fp16")]; + tensor var_1557_begin_0 = const()[name = string("op_1557_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1557_end_0 = const()[name = string("op_1557_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1557_end_mask_0 = const()[name = string("op_1557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1557_cast_fp16 = slice_by_index(begin = var_1557_begin_0, end = var_1557_end_0, end_mask = var_1557_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1557_cast_fp16")]; + tensor var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1564_cast_fp16")]; + tensor var_1571_begin_0 = const()[name = string("op_1571_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1571_end_0 = const()[name = string("op_1571_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1571_end_mask_0 = const()[name = string("op_1571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1571_cast_fp16 = slice_by_index(begin = var_1571_begin_0, end = var_1571_end_0, end_mask = var_1571_end_mask_0, x = var_1233_cast_fp16)[name = string("op_1571_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_10")]; + tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = k_3_cast_fp16)[name = string("op_1576_cast_fp16")]; + tensor var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = k_3_cast_fp16)[name = string("op_1580_cast_fp16")]; + tensor var_1584_begin_0 = const()[name = string("op_1584_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1584_end_0 = const()[name = string("op_1584_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1584_end_mask_0 = const()[name = string("op_1584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = k_3_cast_fp16)[name = string("op_1584_cast_fp16")]; + tensor var_1588_begin_0 = const()[name = string("op_1588_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1588_end_0 = const()[name = string("op_1588_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1588_end_mask_0 = const()[name = string("op_1588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1588_cast_fp16 = slice_by_index(begin = var_1588_begin_0, end = var_1588_end_0, end_mask = var_1588_end_mask_0, x = k_3_cast_fp16)[name = string("op_1588_cast_fp16")]; + tensor var_1592_begin_0 = const()[name = string("op_1592_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1592_end_0 = const()[name = string("op_1592_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1592_end_mask_0 = const()[name = string("op_1592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1592_cast_fp16 = slice_by_index(begin = var_1592_begin_0, end = var_1592_end_0, end_mask = var_1592_end_mask_0, x = k_3_cast_fp16)[name = string("op_1592_cast_fp16")]; + tensor var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = k_3_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1600_begin_0 = const()[name = string("op_1600_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1600_end_0 = const()[name = string("op_1600_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1600_end_mask_0 = const()[name = string("op_1600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1600_cast_fp16 = slice_by_index(begin = var_1600_begin_0, end = var_1600_end_0, end_mask = var_1600_end_mask_0, x = k_3_cast_fp16)[name = string("op_1600_cast_fp16")]; + tensor var_1604_begin_0 = const()[name = string("op_1604_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1604_end_0 = const()[name = string("op_1604_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1604_end_mask_0 = const()[name = string("op_1604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1604_cast_fp16 = slice_by_index(begin = var_1604_begin_0, end = var_1604_end_0, end_mask = var_1604_end_mask_0, x = k_3_cast_fp16)[name = string("op_1604_cast_fp16")]; + tensor var_1608_begin_0 = const()[name = string("op_1608_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1608_end_0 = const()[name = string("op_1608_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_1608_end_mask_0 = const()[name = string("op_1608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1608_cast_fp16 = slice_by_index(begin = var_1608_begin_0, end = var_1608_end_0, end_mask = var_1608_end_mask_0, x = k_3_cast_fp16)[name = string("op_1608_cast_fp16")]; + tensor var_1612_begin_0 = const()[name = string("op_1612_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_1612_end_0 = const()[name = string("op_1612_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_1612_end_mask_0 = const()[name = string("op_1612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1612_cast_fp16 = slice_by_index(begin = var_1612_begin_0, end = var_1612_end_0, end_mask = var_1612_end_mask_0, x = k_3_cast_fp16)[name = string("op_1612_cast_fp16")]; + tensor var_1616_begin_0 = const()[name = string("op_1616_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_1616_end_0 = const()[name = string("op_1616_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_1616_end_mask_0 = const()[name = string("op_1616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1616_cast_fp16 = slice_by_index(begin = var_1616_begin_0, end = var_1616_end_0, end_mask = var_1616_end_mask_0, x = k_3_cast_fp16)[name = string("op_1616_cast_fp16")]; + tensor var_1620_begin_0 = const()[name = string("op_1620_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_1620_end_0 = const()[name = string("op_1620_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_1620_end_mask_0 = const()[name = string("op_1620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1620_cast_fp16 = slice_by_index(begin = var_1620_begin_0, end = var_1620_end_0, end_mask = var_1620_end_mask_0, x = k_3_cast_fp16)[name = string("op_1620_cast_fp16")]; + tensor var_1622_begin_0 = const()[name = string("op_1622_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1622_end_0 = const()[name = string("op_1622_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1622_end_mask_0 = const()[name = string("op_1622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1622_cast_fp16 = slice_by_index(begin = var_1622_begin_0, end = var_1622_end_0, end_mask = var_1622_end_mask_0, x = value_3_cast_fp16)[name = string("op_1622_cast_fp16")]; + tensor var_1626_begin_0 = const()[name = string("op_1626_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1626_end_0 = const()[name = string("op_1626_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1626_end_mask_0 = const()[name = string("op_1626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1626_cast_fp16 = slice_by_index(begin = var_1626_begin_0, end = var_1626_end_0, end_mask = var_1626_end_mask_0, x = value_3_cast_fp16)[name = string("op_1626_cast_fp16")]; + tensor var_1630_begin_0 = const()[name = string("op_1630_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1630_end_0 = const()[name = string("op_1630_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1630_end_mask_0 = const()[name = string("op_1630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1630_cast_fp16 = slice_by_index(begin = var_1630_begin_0, end = var_1630_end_0, end_mask = var_1630_end_mask_0, x = value_3_cast_fp16)[name = string("op_1630_cast_fp16")]; + tensor var_1634_begin_0 = const()[name = string("op_1634_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1634_end_0 = const()[name = string("op_1634_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1634_end_mask_0 = const()[name = string("op_1634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1634_cast_fp16 = slice_by_index(begin = var_1634_begin_0, end = var_1634_end_0, end_mask = var_1634_end_mask_0, x = value_3_cast_fp16)[name = string("op_1634_cast_fp16")]; + tensor var_1638_begin_0 = const()[name = string("op_1638_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1638_end_0 = const()[name = string("op_1638_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1638_end_mask_0 = const()[name = string("op_1638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = value_3_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor var_1642_begin_0 = const()[name = string("op_1642_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1642_end_0 = const()[name = string("op_1642_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1642_end_mask_0 = const()[name = string("op_1642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1642_cast_fp16 = slice_by_index(begin = var_1642_begin_0, end = var_1642_end_0, end_mask = var_1642_end_mask_0, x = value_3_cast_fp16)[name = string("op_1642_cast_fp16")]; + tensor var_1646_begin_0 = const()[name = string("op_1646_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1646_end_0 = const()[name = string("op_1646_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1646_end_mask_0 = const()[name = string("op_1646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1646_cast_fp16 = slice_by_index(begin = var_1646_begin_0, end = var_1646_end_0, end_mask = var_1646_end_mask_0, x = value_3_cast_fp16)[name = string("op_1646_cast_fp16")]; + tensor var_1650_begin_0 = const()[name = string("op_1650_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1650_end_0 = const()[name = string("op_1650_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1650_end_mask_0 = const()[name = string("op_1650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1650_cast_fp16 = slice_by_index(begin = var_1650_begin_0, end = var_1650_end_0, end_mask = var_1650_end_mask_0, x = value_3_cast_fp16)[name = string("op_1650_cast_fp16")]; + tensor var_1654_begin_0 = const()[name = string("op_1654_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1654_end_0 = const()[name = string("op_1654_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1654_end_mask_0 = const()[name = string("op_1654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1654_cast_fp16 = slice_by_index(begin = var_1654_begin_0, end = var_1654_end_0, end_mask = var_1654_end_mask_0, x = value_3_cast_fp16)[name = string("op_1654_cast_fp16")]; + tensor var_1658_begin_0 = const()[name = string("op_1658_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1658_end_0 = const()[name = string("op_1658_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1658_end_mask_0 = const()[name = string("op_1658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1658_cast_fp16 = slice_by_index(begin = var_1658_begin_0, end = var_1658_end_0, end_mask = var_1658_end_mask_0, x = value_3_cast_fp16)[name = string("op_1658_cast_fp16")]; + tensor var_1662_begin_0 = const()[name = string("op_1662_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1662_end_0 = const()[name = string("op_1662_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1662_end_mask_0 = const()[name = string("op_1662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1662_cast_fp16 = slice_by_index(begin = var_1662_begin_0, end = var_1662_end_0, end_mask = var_1662_end_mask_0, x = value_3_cast_fp16)[name = string("op_1662_cast_fp16")]; + tensor var_1666_begin_0 = const()[name = string("op_1666_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1666_end_0 = const()[name = string("op_1666_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1666_end_mask_0 = const()[name = string("op_1666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1666_cast_fp16 = slice_by_index(begin = var_1666_begin_0, end = var_1666_end_0, end_mask = var_1666_end_mask_0, x = value_3_cast_fp16)[name = string("op_1666_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1576_cast_fp16, var_1242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1576_cast_fp16, var_1249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1576_cast_fp16, var_1256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1576_cast_fp16, var_1263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1580_cast_fp16, var_1270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1580_cast_fp16, var_1277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1580_cast_fp16, var_1284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1580_cast_fp16, var_1291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1584_cast_fp16, var_1298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1584_cast_fp16, var_1305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1584_cast_fp16, var_1312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1584_cast_fp16, var_1319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1588_cast_fp16, var_1326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1588_cast_fp16, var_1333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1588_cast_fp16, var_1340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1588_cast_fp16, var_1347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1592_cast_fp16, var_1354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1592_cast_fp16, var_1361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1592_cast_fp16, var_1368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1592_cast_fp16, var_1375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1596_cast_fp16, var_1382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1596_cast_fp16, var_1389_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1596_cast_fp16, var_1396_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1596_cast_fp16, var_1403_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1600_cast_fp16, var_1410_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1600_cast_fp16, var_1417_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1600_cast_fp16, var_1424_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1600_cast_fp16, var_1431_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1604_cast_fp16, var_1438_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1604_cast_fp16, var_1445_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1604_cast_fp16, var_1452_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1604_cast_fp16, var_1459_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1608_cast_fp16, var_1466_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1608_cast_fp16, var_1473_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1608_cast_fp16, var_1480_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1608_cast_fp16, var_1487_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1612_cast_fp16, var_1494_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1612_cast_fp16, var_1501_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1612_cast_fp16, var_1508_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1612_cast_fp16, var_1515_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_1616_cast_fp16, var_1522_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_1616_cast_fp16, var_1529_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_1616_cast_fp16, var_1536_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_1616_cast_fp16, var_1543_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_1620_cast_fp16, var_1550_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_1620_cast_fp16, var_1557_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_1620_cast_fp16, var_1564_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_1620_cast_fp16, var_1571_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + fp16 var_1765_to_fp16 = const()[name = string("op_1765_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1765_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1767_to_fp16 = const()[name = string("op_1767_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1767_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1769_to_fp16 = const()[name = string("op_1769_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1769_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1771_to_fp16 = const()[name = string("op_1771_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1771_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1773_to_fp16 = const()[name = string("op_1773_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1773_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1775_to_fp16 = const()[name = string("op_1775_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1775_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1777_to_fp16 = const()[name = string("op_1777_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1777_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1779_to_fp16 = const()[name = string("op_1779_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1779_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1781_to_fp16 = const()[name = string("op_1781_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1781_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1783_to_fp16 = const()[name = string("op_1783_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1783_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1785_to_fp16 = const()[name = string("op_1785_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1785_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1787_to_fp16 = const()[name = string("op_1787_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1787_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1789_to_fp16 = const()[name = string("op_1789_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1789_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1791_to_fp16 = const()[name = string("op_1791_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1791_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1793_to_fp16 = const()[name = string("op_1793_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1793_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1795_to_fp16 = const()[name = string("op_1795_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1795_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + fp16 var_1797_to_fp16 = const()[name = string("op_1797_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1797_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1799_to_fp16 = const()[name = string("op_1799_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1799_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1801_to_fp16 = const()[name = string("op_1801_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1801_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1803_to_fp16 = const()[name = string("op_1803_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1803_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1805_to_fp16 = const()[name = string("op_1805_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1805_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1807_to_fp16 = const()[name = string("op_1807_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1807_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1809_to_fp16 = const()[name = string("op_1809_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1809_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1811_to_fp16 = const()[name = string("op_1811_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1811_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1813_to_fp16 = const()[name = string("op_1813_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1813_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1815_to_fp16 = const()[name = string("op_1815_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1815_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1817_to_fp16 = const()[name = string("op_1817_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1817_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1819_to_fp16 = const()[name = string("op_1819_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1819_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1821_to_fp16 = const()[name = string("op_1821_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1821_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1823_to_fp16 = const()[name = string("op_1823_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1823_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1825_to_fp16 = const()[name = string("op_1825_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1825_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1827_to_fp16 = const()[name = string("op_1827_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1827_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_1829_to_fp16 = const()[name = string("op_1829_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_1829_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_1831_to_fp16 = const()[name = string("op_1831_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_1831_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_1833_to_fp16 = const()[name = string("op_1833_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_1833_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_1835_to_fp16 = const()[name = string("op_1835_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_1835_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_1837_to_fp16 = const()[name = string("op_1837_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_1837_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_1839_to_fp16 = const()[name = string("op_1839_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_1839_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_1841_to_fp16 = const()[name = string("op_1841_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_1841_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_1843_to_fp16 = const()[name = string("op_1843_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_1843_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_1845_to_fp16 = const()[name = string("op_1845_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_1845_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_1847_to_fp16 = const()[name = string("op_1847_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_1847_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_1849_to_fp16 = const()[name = string("op_1849_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_1849_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_1851_to_fp16 = const()[name = string("op_1851_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_1851_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_1853_to_fp16 = const()[name = string("op_1853_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_1853_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_1855_to_fp16 = const()[name = string("op_1855_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_1855_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_1857_to_fp16 = const()[name = string("op_1857_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_1857_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_1859_to_fp16 = const()[name = string("op_1859_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_1859_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + tensor var_1861_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_97_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1862_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_99_cast_fp16)[name = string("op_1862_cast_fp16")]; + tensor var_1863_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_101_cast_fp16)[name = string("op_1863_cast_fp16")]; + tensor var_1864_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_103_cast_fp16)[name = string("op_1864_cast_fp16")]; + tensor var_1865_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_105_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1866_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_107_cast_fp16)[name = string("op_1866_cast_fp16")]; + tensor var_1867_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_109_cast_fp16)[name = string("op_1867_cast_fp16")]; + tensor var_1868_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_111_cast_fp16)[name = string("op_1868_cast_fp16")]; + tensor var_1869_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_113_cast_fp16)[name = string("op_1869_cast_fp16")]; + tensor var_1870_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_115_cast_fp16)[name = string("op_1870_cast_fp16")]; + tensor var_1871_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_117_cast_fp16)[name = string("op_1871_cast_fp16")]; + tensor var_1872_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_119_cast_fp16)[name = string("op_1872_cast_fp16")]; + tensor var_1873_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_121_cast_fp16)[name = string("op_1873_cast_fp16")]; + tensor var_1874_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_123_cast_fp16)[name = string("op_1874_cast_fp16")]; + tensor var_1875_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_125_cast_fp16)[name = string("op_1875_cast_fp16")]; + tensor var_1876_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_127_cast_fp16)[name = string("op_1876_cast_fp16")]; + tensor var_1877_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_129_cast_fp16)[name = string("op_1877_cast_fp16")]; + tensor var_1878_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_131_cast_fp16)[name = string("op_1878_cast_fp16")]; + tensor var_1879_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_133_cast_fp16)[name = string("op_1879_cast_fp16")]; + tensor var_1880_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_135_cast_fp16)[name = string("op_1880_cast_fp16")]; + tensor var_1881_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_137_cast_fp16)[name = string("op_1881_cast_fp16")]; + tensor var_1882_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_139_cast_fp16)[name = string("op_1882_cast_fp16")]; + tensor var_1883_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_141_cast_fp16)[name = string("op_1883_cast_fp16")]; + tensor var_1884_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_143_cast_fp16)[name = string("op_1884_cast_fp16")]; + tensor var_1885_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_145_cast_fp16)[name = string("op_1885_cast_fp16")]; + tensor var_1886_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_147_cast_fp16)[name = string("op_1886_cast_fp16")]; + tensor var_1887_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_149_cast_fp16)[name = string("op_1887_cast_fp16")]; + tensor var_1888_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_151_cast_fp16)[name = string("op_1888_cast_fp16")]; + tensor var_1889_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_153_cast_fp16)[name = string("op_1889_cast_fp16")]; + tensor var_1890_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_155_cast_fp16)[name = string("op_1890_cast_fp16")]; + tensor var_1891_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_157_cast_fp16)[name = string("op_1891_cast_fp16")]; + tensor var_1892_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_159_cast_fp16)[name = string("op_1892_cast_fp16")]; + tensor var_1893_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_161_cast_fp16)[name = string("op_1893_cast_fp16")]; + tensor var_1894_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_163_cast_fp16)[name = string("op_1894_cast_fp16")]; + tensor var_1895_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_165_cast_fp16)[name = string("op_1895_cast_fp16")]; + tensor var_1896_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_167_cast_fp16)[name = string("op_1896_cast_fp16")]; + tensor var_1897_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_169_cast_fp16)[name = string("op_1897_cast_fp16")]; + tensor var_1898_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_171_cast_fp16)[name = string("op_1898_cast_fp16")]; + tensor var_1899_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_173_cast_fp16)[name = string("op_1899_cast_fp16")]; + tensor var_1900_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_175_cast_fp16)[name = string("op_1900_cast_fp16")]; + tensor var_1901_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_177_cast_fp16)[name = string("op_1901_cast_fp16")]; + tensor var_1902_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_179_cast_fp16)[name = string("op_1902_cast_fp16")]; + tensor var_1903_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_181_cast_fp16)[name = string("op_1903_cast_fp16")]; + tensor var_1904_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_183_cast_fp16)[name = string("op_1904_cast_fp16")]; + tensor var_1905_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_185_cast_fp16)[name = string("op_1905_cast_fp16")]; + tensor var_1906_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_187_cast_fp16)[name = string("op_1906_cast_fp16")]; + tensor var_1907_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_189_cast_fp16)[name = string("op_1907_cast_fp16")]; + tensor var_1908_cast_fp16 = softmax(axis = var_1134, x = aw_chunk_191_cast_fp16)[name = string("op_1908_cast_fp16")]; + string var_1910_equation_0 = const()[name = string("op_1910_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1910_cast_fp16 = einsum(equation = var_1910_equation_0, values = (var_1622_cast_fp16, var_1861_cast_fp16))[name = string("op_1910_cast_fp16")]; + string var_1912_equation_0 = const()[name = string("op_1912_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1912_cast_fp16 = einsum(equation = var_1912_equation_0, values = (var_1622_cast_fp16, var_1862_cast_fp16))[name = string("op_1912_cast_fp16")]; + string var_1914_equation_0 = const()[name = string("op_1914_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1914_cast_fp16 = einsum(equation = var_1914_equation_0, values = (var_1622_cast_fp16, var_1863_cast_fp16))[name = string("op_1914_cast_fp16")]; + string var_1916_equation_0 = const()[name = string("op_1916_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1916_cast_fp16 = einsum(equation = var_1916_equation_0, values = (var_1622_cast_fp16, var_1864_cast_fp16))[name = string("op_1916_cast_fp16")]; + string var_1918_equation_0 = const()[name = string("op_1918_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1918_cast_fp16 = einsum(equation = var_1918_equation_0, values = (var_1626_cast_fp16, var_1865_cast_fp16))[name = string("op_1918_cast_fp16")]; + string var_1920_equation_0 = const()[name = string("op_1920_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1920_cast_fp16 = einsum(equation = var_1920_equation_0, values = (var_1626_cast_fp16, var_1866_cast_fp16))[name = string("op_1920_cast_fp16")]; + string var_1922_equation_0 = const()[name = string("op_1922_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1922_cast_fp16 = einsum(equation = var_1922_equation_0, values = (var_1626_cast_fp16, var_1867_cast_fp16))[name = string("op_1922_cast_fp16")]; + string var_1924_equation_0 = const()[name = string("op_1924_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1924_cast_fp16 = einsum(equation = var_1924_equation_0, values = (var_1626_cast_fp16, var_1868_cast_fp16))[name = string("op_1924_cast_fp16")]; + string var_1926_equation_0 = const()[name = string("op_1926_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1926_cast_fp16 = einsum(equation = var_1926_equation_0, values = (var_1630_cast_fp16, var_1869_cast_fp16))[name = string("op_1926_cast_fp16")]; + string var_1928_equation_0 = const()[name = string("op_1928_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1928_cast_fp16 = einsum(equation = var_1928_equation_0, values = (var_1630_cast_fp16, var_1870_cast_fp16))[name = string("op_1928_cast_fp16")]; + string var_1930_equation_0 = const()[name = string("op_1930_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1930_cast_fp16 = einsum(equation = var_1930_equation_0, values = (var_1630_cast_fp16, var_1871_cast_fp16))[name = string("op_1930_cast_fp16")]; + string var_1932_equation_0 = const()[name = string("op_1932_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1932_cast_fp16 = einsum(equation = var_1932_equation_0, values = (var_1630_cast_fp16, var_1872_cast_fp16))[name = string("op_1932_cast_fp16")]; + string var_1934_equation_0 = const()[name = string("op_1934_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1934_cast_fp16 = einsum(equation = var_1934_equation_0, values = (var_1634_cast_fp16, var_1873_cast_fp16))[name = string("op_1934_cast_fp16")]; + string var_1936_equation_0 = const()[name = string("op_1936_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1936_cast_fp16 = einsum(equation = var_1936_equation_0, values = (var_1634_cast_fp16, var_1874_cast_fp16))[name = string("op_1936_cast_fp16")]; + string var_1938_equation_0 = const()[name = string("op_1938_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1938_cast_fp16 = einsum(equation = var_1938_equation_0, values = (var_1634_cast_fp16, var_1875_cast_fp16))[name = string("op_1938_cast_fp16")]; + string var_1940_equation_0 = const()[name = string("op_1940_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1940_cast_fp16 = einsum(equation = var_1940_equation_0, values = (var_1634_cast_fp16, var_1876_cast_fp16))[name = string("op_1940_cast_fp16")]; + string var_1942_equation_0 = const()[name = string("op_1942_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1942_cast_fp16 = einsum(equation = var_1942_equation_0, values = (var_1638_cast_fp16, var_1877_cast_fp16))[name = string("op_1942_cast_fp16")]; + string var_1944_equation_0 = const()[name = string("op_1944_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1944_cast_fp16 = einsum(equation = var_1944_equation_0, values = (var_1638_cast_fp16, var_1878_cast_fp16))[name = string("op_1944_cast_fp16")]; + string var_1946_equation_0 = const()[name = string("op_1946_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1946_cast_fp16 = einsum(equation = var_1946_equation_0, values = (var_1638_cast_fp16, var_1879_cast_fp16))[name = string("op_1946_cast_fp16")]; + string var_1948_equation_0 = const()[name = string("op_1948_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1948_cast_fp16 = einsum(equation = var_1948_equation_0, values = (var_1638_cast_fp16, var_1880_cast_fp16))[name = string("op_1948_cast_fp16")]; + string var_1950_equation_0 = const()[name = string("op_1950_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1950_cast_fp16 = einsum(equation = var_1950_equation_0, values = (var_1642_cast_fp16, var_1881_cast_fp16))[name = string("op_1950_cast_fp16")]; + string var_1952_equation_0 = const()[name = string("op_1952_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1952_cast_fp16 = einsum(equation = var_1952_equation_0, values = (var_1642_cast_fp16, var_1882_cast_fp16))[name = string("op_1952_cast_fp16")]; + string var_1954_equation_0 = const()[name = string("op_1954_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1954_cast_fp16 = einsum(equation = var_1954_equation_0, values = (var_1642_cast_fp16, var_1883_cast_fp16))[name = string("op_1954_cast_fp16")]; + string var_1956_equation_0 = const()[name = string("op_1956_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1956_cast_fp16 = einsum(equation = var_1956_equation_0, values = (var_1642_cast_fp16, var_1884_cast_fp16))[name = string("op_1956_cast_fp16")]; + string var_1958_equation_0 = const()[name = string("op_1958_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1958_cast_fp16 = einsum(equation = var_1958_equation_0, values = (var_1646_cast_fp16, var_1885_cast_fp16))[name = string("op_1958_cast_fp16")]; + string var_1960_equation_0 = const()[name = string("op_1960_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1960_cast_fp16 = einsum(equation = var_1960_equation_0, values = (var_1646_cast_fp16, var_1886_cast_fp16))[name = string("op_1960_cast_fp16")]; + string var_1962_equation_0 = const()[name = string("op_1962_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1962_cast_fp16 = einsum(equation = var_1962_equation_0, values = (var_1646_cast_fp16, var_1887_cast_fp16))[name = string("op_1962_cast_fp16")]; + string var_1964_equation_0 = const()[name = string("op_1964_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1964_cast_fp16 = einsum(equation = var_1964_equation_0, values = (var_1646_cast_fp16, var_1888_cast_fp16))[name = string("op_1964_cast_fp16")]; + string var_1966_equation_0 = const()[name = string("op_1966_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1966_cast_fp16 = einsum(equation = var_1966_equation_0, values = (var_1650_cast_fp16, var_1889_cast_fp16))[name = string("op_1966_cast_fp16")]; + string var_1968_equation_0 = const()[name = string("op_1968_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1968_cast_fp16 = einsum(equation = var_1968_equation_0, values = (var_1650_cast_fp16, var_1890_cast_fp16))[name = string("op_1968_cast_fp16")]; + string var_1970_equation_0 = const()[name = string("op_1970_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1970_cast_fp16 = einsum(equation = var_1970_equation_0, values = (var_1650_cast_fp16, var_1891_cast_fp16))[name = string("op_1970_cast_fp16")]; + string var_1972_equation_0 = const()[name = string("op_1972_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1972_cast_fp16 = einsum(equation = var_1972_equation_0, values = (var_1650_cast_fp16, var_1892_cast_fp16))[name = string("op_1972_cast_fp16")]; + string var_1974_equation_0 = const()[name = string("op_1974_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1974_cast_fp16 = einsum(equation = var_1974_equation_0, values = (var_1654_cast_fp16, var_1893_cast_fp16))[name = string("op_1974_cast_fp16")]; + string var_1976_equation_0 = const()[name = string("op_1976_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1976_cast_fp16 = einsum(equation = var_1976_equation_0, values = (var_1654_cast_fp16, var_1894_cast_fp16))[name = string("op_1976_cast_fp16")]; + string var_1978_equation_0 = const()[name = string("op_1978_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1978_cast_fp16 = einsum(equation = var_1978_equation_0, values = (var_1654_cast_fp16, var_1895_cast_fp16))[name = string("op_1978_cast_fp16")]; + string var_1980_equation_0 = const()[name = string("op_1980_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1980_cast_fp16 = einsum(equation = var_1980_equation_0, values = (var_1654_cast_fp16, var_1896_cast_fp16))[name = string("op_1980_cast_fp16")]; + string var_1982_equation_0 = const()[name = string("op_1982_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1982_cast_fp16 = einsum(equation = var_1982_equation_0, values = (var_1658_cast_fp16, var_1897_cast_fp16))[name = string("op_1982_cast_fp16")]; + string var_1984_equation_0 = const()[name = string("op_1984_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1984_cast_fp16 = einsum(equation = var_1984_equation_0, values = (var_1658_cast_fp16, var_1898_cast_fp16))[name = string("op_1984_cast_fp16")]; + string var_1986_equation_0 = const()[name = string("op_1986_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1986_cast_fp16 = einsum(equation = var_1986_equation_0, values = (var_1658_cast_fp16, var_1899_cast_fp16))[name = string("op_1986_cast_fp16")]; + string var_1988_equation_0 = const()[name = string("op_1988_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1988_cast_fp16 = einsum(equation = var_1988_equation_0, values = (var_1658_cast_fp16, var_1900_cast_fp16))[name = string("op_1988_cast_fp16")]; + string var_1990_equation_0 = const()[name = string("op_1990_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1990_cast_fp16 = einsum(equation = var_1990_equation_0, values = (var_1662_cast_fp16, var_1901_cast_fp16))[name = string("op_1990_cast_fp16")]; + string var_1992_equation_0 = const()[name = string("op_1992_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1992_cast_fp16 = einsum(equation = var_1992_equation_0, values = (var_1662_cast_fp16, var_1902_cast_fp16))[name = string("op_1992_cast_fp16")]; + string var_1994_equation_0 = const()[name = string("op_1994_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1994_cast_fp16 = einsum(equation = var_1994_equation_0, values = (var_1662_cast_fp16, var_1903_cast_fp16))[name = string("op_1994_cast_fp16")]; + string var_1996_equation_0 = const()[name = string("op_1996_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1996_cast_fp16 = einsum(equation = var_1996_equation_0, values = (var_1662_cast_fp16, var_1904_cast_fp16))[name = string("op_1996_cast_fp16")]; + string var_1998_equation_0 = const()[name = string("op_1998_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1998_cast_fp16 = einsum(equation = var_1998_equation_0, values = (var_1666_cast_fp16, var_1905_cast_fp16))[name = string("op_1998_cast_fp16")]; + string var_2000_equation_0 = const()[name = string("op_2000_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2000_cast_fp16 = einsum(equation = var_2000_equation_0, values = (var_1666_cast_fp16, var_1906_cast_fp16))[name = string("op_2000_cast_fp16")]; + string var_2002_equation_0 = const()[name = string("op_2002_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2002_cast_fp16 = einsum(equation = var_2002_equation_0, values = (var_1666_cast_fp16, var_1907_cast_fp16))[name = string("op_2002_cast_fp16")]; + string var_2004_equation_0 = const()[name = string("op_2004_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2004_cast_fp16 = einsum(equation = var_2004_equation_0, values = (var_1666_cast_fp16, var_1908_cast_fp16))[name = string("op_2004_cast_fp16")]; + bool var_2006_interleave_0 = const()[name = string("op_2006_interleave_0"), val = bool(false)]; + tensor var_2006_cast_fp16 = concat(axis = var_1117, interleave = var_2006_interleave_0, values = (var_1910_cast_fp16, var_1912_cast_fp16, var_1914_cast_fp16, var_1916_cast_fp16))[name = string("op_2006_cast_fp16")]; + bool var_2008_interleave_0 = const()[name = string("op_2008_interleave_0"), val = bool(false)]; + tensor var_2008_cast_fp16 = concat(axis = var_1117, interleave = var_2008_interleave_0, values = (var_1918_cast_fp16, var_1920_cast_fp16, var_1922_cast_fp16, var_1924_cast_fp16))[name = string("op_2008_cast_fp16")]; + bool var_2010_interleave_0 = const()[name = string("op_2010_interleave_0"), val = bool(false)]; + tensor var_2010_cast_fp16 = concat(axis = var_1117, interleave = var_2010_interleave_0, values = (var_1926_cast_fp16, var_1928_cast_fp16, var_1930_cast_fp16, var_1932_cast_fp16))[name = string("op_2010_cast_fp16")]; + bool var_2012_interleave_0 = const()[name = string("op_2012_interleave_0"), val = bool(false)]; + tensor var_2012_cast_fp16 = concat(axis = var_1117, interleave = var_2012_interleave_0, values = (var_1934_cast_fp16, var_1936_cast_fp16, var_1938_cast_fp16, var_1940_cast_fp16))[name = string("op_2012_cast_fp16")]; + bool var_2014_interleave_0 = const()[name = string("op_2014_interleave_0"), val = bool(false)]; + tensor var_2014_cast_fp16 = concat(axis = var_1117, interleave = var_2014_interleave_0, values = (var_1942_cast_fp16, var_1944_cast_fp16, var_1946_cast_fp16, var_1948_cast_fp16))[name = string("op_2014_cast_fp16")]; + bool var_2016_interleave_0 = const()[name = string("op_2016_interleave_0"), val = bool(false)]; + tensor var_2016_cast_fp16 = concat(axis = var_1117, interleave = var_2016_interleave_0, values = (var_1950_cast_fp16, var_1952_cast_fp16, var_1954_cast_fp16, var_1956_cast_fp16))[name = string("op_2016_cast_fp16")]; + bool var_2018_interleave_0 = const()[name = string("op_2018_interleave_0"), val = bool(false)]; + tensor var_2018_cast_fp16 = concat(axis = var_1117, interleave = var_2018_interleave_0, values = (var_1958_cast_fp16, var_1960_cast_fp16, var_1962_cast_fp16, var_1964_cast_fp16))[name = string("op_2018_cast_fp16")]; + bool var_2020_interleave_0 = const()[name = string("op_2020_interleave_0"), val = bool(false)]; + tensor var_2020_cast_fp16 = concat(axis = var_1117, interleave = var_2020_interleave_0, values = (var_1966_cast_fp16, var_1968_cast_fp16, var_1970_cast_fp16, var_1972_cast_fp16))[name = string("op_2020_cast_fp16")]; + bool var_2022_interleave_0 = const()[name = string("op_2022_interleave_0"), val = bool(false)]; + tensor var_2022_cast_fp16 = concat(axis = var_1117, interleave = var_2022_interleave_0, values = (var_1974_cast_fp16, var_1976_cast_fp16, var_1978_cast_fp16, var_1980_cast_fp16))[name = string("op_2022_cast_fp16")]; + bool var_2024_interleave_0 = const()[name = string("op_2024_interleave_0"), val = bool(false)]; + tensor var_2024_cast_fp16 = concat(axis = var_1117, interleave = var_2024_interleave_0, values = (var_1982_cast_fp16, var_1984_cast_fp16, var_1986_cast_fp16, var_1988_cast_fp16))[name = string("op_2024_cast_fp16")]; + bool var_2026_interleave_0 = const()[name = string("op_2026_interleave_0"), val = bool(false)]; + tensor var_2026_cast_fp16 = concat(axis = var_1117, interleave = var_2026_interleave_0, values = (var_1990_cast_fp16, var_1992_cast_fp16, var_1994_cast_fp16, var_1996_cast_fp16))[name = string("op_2026_cast_fp16")]; + bool var_2028_interleave_0 = const()[name = string("op_2028_interleave_0"), val = bool(false)]; + tensor var_2028_cast_fp16 = concat(axis = var_1117, interleave = var_2028_interleave_0, values = (var_1998_cast_fp16, var_2000_cast_fp16, var_2002_cast_fp16, var_2004_cast_fp16))[name = string("op_2028_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_1134, interleave = input_9_interleave_0, values = (var_2006_cast_fp16, var_2008_cast_fp16, var_2010_cast_fp16, var_2012_cast_fp16, var_2014_cast_fp16, var_2016_cast_fp16, var_2018_cast_fp16, var_2020_cast_fp16, var_2022_cast_fp16, var_2024_cast_fp16, var_2026_cast_fp16, var_2028_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23938944)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25118656)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_2047_to_fp16 = const()[name = string("op_2047_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_2047_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25120256)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25121856)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25123456)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29842112)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29848320)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34566976)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_2076 = const()[name = string("op_2076"), val = int32(3)]; + int32 var_2093 = const()[name = string("op_2093"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_2110_to_fp16 = const()[name = string("op_2110_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_2110_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34568576)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34570176)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34571776)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35751488)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35753088)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36932800)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38112512)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_2148_begin_0 = const()[name = string("op_2148_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2148_end_0 = const()[name = string("op_2148_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2148_end_mask_0 = const()[name = string("op_2148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2148_cast_fp16 = slice_by_index(begin = var_2148_begin_0, end = var_2148_end_0, end_mask = var_2148_end_mask_0, x = query_5_cast_fp16)[name = string("op_2148_cast_fp16")]; + tensor var_2152_begin_0 = const()[name = string("op_2152_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2152_end_0 = const()[name = string("op_2152_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2152_end_mask_0 = const()[name = string("op_2152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2152_cast_fp16 = slice_by_index(begin = var_2152_begin_0, end = var_2152_end_0, end_mask = var_2152_end_mask_0, x = query_5_cast_fp16)[name = string("op_2152_cast_fp16")]; + tensor var_2156_begin_0 = const()[name = string("op_2156_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2156_end_0 = const()[name = string("op_2156_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2156_end_mask_0 = const()[name = string("op_2156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2156_cast_fp16 = slice_by_index(begin = var_2156_begin_0, end = var_2156_end_0, end_mask = var_2156_end_mask_0, x = query_5_cast_fp16)[name = string("op_2156_cast_fp16")]; + tensor var_2160_begin_0 = const()[name = string("op_2160_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2160_end_0 = const()[name = string("op_2160_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2160_end_mask_0 = const()[name = string("op_2160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = query_5_cast_fp16)[name = string("op_2160_cast_fp16")]; + tensor var_2164_begin_0 = const()[name = string("op_2164_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2164_end_0 = const()[name = string("op_2164_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2164_end_mask_0 = const()[name = string("op_2164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = query_5_cast_fp16)[name = string("op_2164_cast_fp16")]; + tensor var_2168_begin_0 = const()[name = string("op_2168_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2168_end_0 = const()[name = string("op_2168_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2168_end_mask_0 = const()[name = string("op_2168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2168_cast_fp16 = slice_by_index(begin = var_2168_begin_0, end = var_2168_end_0, end_mask = var_2168_end_mask_0, x = query_5_cast_fp16)[name = string("op_2168_cast_fp16")]; + tensor var_2172_begin_0 = const()[name = string("op_2172_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2172_end_0 = const()[name = string("op_2172_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2172_end_mask_0 = const()[name = string("op_2172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2172_cast_fp16 = slice_by_index(begin = var_2172_begin_0, end = var_2172_end_0, end_mask = var_2172_end_mask_0, x = query_5_cast_fp16)[name = string("op_2172_cast_fp16")]; + tensor var_2176_begin_0 = const()[name = string("op_2176_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2176_end_0 = const()[name = string("op_2176_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2176_end_mask_0 = const()[name = string("op_2176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2176_cast_fp16 = slice_by_index(begin = var_2176_begin_0, end = var_2176_end_0, end_mask = var_2176_end_mask_0, x = query_5_cast_fp16)[name = string("op_2176_cast_fp16")]; + tensor var_2180_begin_0 = const()[name = string("op_2180_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2180_end_0 = const()[name = string("op_2180_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2180_end_mask_0 = const()[name = string("op_2180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2180_cast_fp16 = slice_by_index(begin = var_2180_begin_0, end = var_2180_end_0, end_mask = var_2180_end_mask_0, x = query_5_cast_fp16)[name = string("op_2180_cast_fp16")]; + tensor var_2184_begin_0 = const()[name = string("op_2184_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2184_end_0 = const()[name = string("op_2184_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2184_end_mask_0 = const()[name = string("op_2184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2184_cast_fp16 = slice_by_index(begin = var_2184_begin_0, end = var_2184_end_0, end_mask = var_2184_end_mask_0, x = query_5_cast_fp16)[name = string("op_2184_cast_fp16")]; + tensor var_2188_begin_0 = const()[name = string("op_2188_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2188_end_0 = const()[name = string("op_2188_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2188_end_mask_0 = const()[name = string("op_2188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2188_cast_fp16 = slice_by_index(begin = var_2188_begin_0, end = var_2188_end_0, end_mask = var_2188_end_mask_0, x = query_5_cast_fp16)[name = string("op_2188_cast_fp16")]; + tensor var_2192_begin_0 = const()[name = string("op_2192_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2192_end_0 = const()[name = string("op_2192_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2192_end_mask_0 = const()[name = string("op_2192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2192_cast_fp16 = slice_by_index(begin = var_2192_begin_0, end = var_2192_end_0, end_mask = var_2192_end_mask_0, x = query_5_cast_fp16)[name = string("op_2192_cast_fp16")]; + tensor var_2201_begin_0 = const()[name = string("op_2201_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2201_end_0 = const()[name = string("op_2201_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2201_end_mask_0 = const()[name = string("op_2201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2201_cast_fp16 = slice_by_index(begin = var_2201_begin_0, end = var_2201_end_0, end_mask = var_2201_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2201_cast_fp16")]; + tensor var_2208_begin_0 = const()[name = string("op_2208_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2208_end_0 = const()[name = string("op_2208_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2208_end_mask_0 = const()[name = string("op_2208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2208_cast_fp16 = slice_by_index(begin = var_2208_begin_0, end = var_2208_end_0, end_mask = var_2208_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2208_cast_fp16")]; + tensor var_2215_begin_0 = const()[name = string("op_2215_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2215_end_0 = const()[name = string("op_2215_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2215_end_mask_0 = const()[name = string("op_2215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2215_cast_fp16 = slice_by_index(begin = var_2215_begin_0, end = var_2215_end_0, end_mask = var_2215_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2215_cast_fp16")]; + tensor var_2222_begin_0 = const()[name = string("op_2222_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2222_end_0 = const()[name = string("op_2222_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2222_end_mask_0 = const()[name = string("op_2222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2222_cast_fp16 = slice_by_index(begin = var_2222_begin_0, end = var_2222_end_0, end_mask = var_2222_end_mask_0, x = var_2148_cast_fp16)[name = string("op_2222_cast_fp16")]; + tensor var_2229_begin_0 = const()[name = string("op_2229_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2229_end_0 = const()[name = string("op_2229_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2229_end_mask_0 = const()[name = string("op_2229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2229_cast_fp16 = slice_by_index(begin = var_2229_begin_0, end = var_2229_end_0, end_mask = var_2229_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2229_cast_fp16")]; + tensor var_2236_begin_0 = const()[name = string("op_2236_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2236_end_0 = const()[name = string("op_2236_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2236_end_mask_0 = const()[name = string("op_2236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2236_cast_fp16 = slice_by_index(begin = var_2236_begin_0, end = var_2236_end_0, end_mask = var_2236_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2236_cast_fp16")]; + tensor var_2243_begin_0 = const()[name = string("op_2243_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2243_end_0 = const()[name = string("op_2243_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2243_end_mask_0 = const()[name = string("op_2243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2243_cast_fp16")]; + tensor var_2250_begin_0 = const()[name = string("op_2250_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2250_end_0 = const()[name = string("op_2250_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2250_end_mask_0 = const()[name = string("op_2250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2250_cast_fp16 = slice_by_index(begin = var_2250_begin_0, end = var_2250_end_0, end_mask = var_2250_end_mask_0, x = var_2152_cast_fp16)[name = string("op_2250_cast_fp16")]; + tensor var_2257_begin_0 = const()[name = string("op_2257_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2257_end_0 = const()[name = string("op_2257_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2257_end_mask_0 = const()[name = string("op_2257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2257_cast_fp16 = slice_by_index(begin = var_2257_begin_0, end = var_2257_end_0, end_mask = var_2257_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2257_cast_fp16")]; + tensor var_2264_begin_0 = const()[name = string("op_2264_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2264_end_0 = const()[name = string("op_2264_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2264_end_mask_0 = const()[name = string("op_2264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2264_cast_fp16 = slice_by_index(begin = var_2264_begin_0, end = var_2264_end_0, end_mask = var_2264_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2264_cast_fp16")]; + tensor var_2271_begin_0 = const()[name = string("op_2271_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2271_end_0 = const()[name = string("op_2271_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2271_end_mask_0 = const()[name = string("op_2271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2271_cast_fp16 = slice_by_index(begin = var_2271_begin_0, end = var_2271_end_0, end_mask = var_2271_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2271_cast_fp16")]; + tensor var_2278_begin_0 = const()[name = string("op_2278_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2278_end_0 = const()[name = string("op_2278_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2278_end_mask_0 = const()[name = string("op_2278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2278_cast_fp16 = slice_by_index(begin = var_2278_begin_0, end = var_2278_end_0, end_mask = var_2278_end_mask_0, x = var_2156_cast_fp16)[name = string("op_2278_cast_fp16")]; + tensor var_2285_begin_0 = const()[name = string("op_2285_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2285_end_0 = const()[name = string("op_2285_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2285_end_mask_0 = const()[name = string("op_2285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2285_cast_fp16 = slice_by_index(begin = var_2285_begin_0, end = var_2285_end_0, end_mask = var_2285_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2285_cast_fp16")]; + tensor var_2292_begin_0 = const()[name = string("op_2292_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2292_end_0 = const()[name = string("op_2292_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2292_end_mask_0 = const()[name = string("op_2292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2292_cast_fp16 = slice_by_index(begin = var_2292_begin_0, end = var_2292_end_0, end_mask = var_2292_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2292_cast_fp16")]; + tensor var_2299_begin_0 = const()[name = string("op_2299_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2299_end_0 = const()[name = string("op_2299_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2299_end_mask_0 = const()[name = string("op_2299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2299_cast_fp16 = slice_by_index(begin = var_2299_begin_0, end = var_2299_end_0, end_mask = var_2299_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2299_cast_fp16")]; + tensor var_2306_begin_0 = const()[name = string("op_2306_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2306_end_0 = const()[name = string("op_2306_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2306_end_mask_0 = const()[name = string("op_2306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2306_cast_fp16 = slice_by_index(begin = var_2306_begin_0, end = var_2306_end_0, end_mask = var_2306_end_mask_0, x = var_2160_cast_fp16)[name = string("op_2306_cast_fp16")]; + tensor var_2313_begin_0 = const()[name = string("op_2313_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2313_end_0 = const()[name = string("op_2313_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2313_end_mask_0 = const()[name = string("op_2313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2313_cast_fp16 = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2313_cast_fp16")]; + tensor var_2320_begin_0 = const()[name = string("op_2320_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2320_end_0 = const()[name = string("op_2320_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2320_end_mask_0 = const()[name = string("op_2320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2320_cast_fp16 = slice_by_index(begin = var_2320_begin_0, end = var_2320_end_0, end_mask = var_2320_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2320_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = string("op_2327_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2327_end_0 = const()[name = string("op_2327_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2327_end_mask_0 = const()[name = string("op_2327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2327_cast_fp16")]; + tensor var_2334_begin_0 = const()[name = string("op_2334_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2334_end_0 = const()[name = string("op_2334_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2334_end_mask_0 = const()[name = string("op_2334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2334_cast_fp16 = slice_by_index(begin = var_2334_begin_0, end = var_2334_end_0, end_mask = var_2334_end_mask_0, x = var_2164_cast_fp16)[name = string("op_2334_cast_fp16")]; + tensor var_2341_begin_0 = const()[name = string("op_2341_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2341_end_0 = const()[name = string("op_2341_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2341_end_mask_0 = const()[name = string("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2341_cast_fp16 = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2341_cast_fp16")]; + tensor var_2348_begin_0 = const()[name = string("op_2348_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2348_end_0 = const()[name = string("op_2348_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2348_end_mask_0 = const()[name = string("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2348_cast_fp16 = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2348_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = string("op_2355_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2355_end_0 = const()[name = string("op_2355_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2355_end_mask_0 = const()[name = string("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2355_cast_fp16")]; + tensor var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2168_cast_fp16)[name = string("op_2362_cast_fp16")]; + tensor var_2369_begin_0 = const()[name = string("op_2369_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2369_end_0 = const()[name = string("op_2369_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2369_end_mask_0 = const()[name = string("op_2369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2369_cast_fp16 = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2369_cast_fp16")]; + tensor var_2376_begin_0 = const()[name = string("op_2376_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2376_end_0 = const()[name = string("op_2376_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2376_end_mask_0 = const()[name = string("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2376_cast_fp16 = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2376_cast_fp16")]; + tensor var_2383_begin_0 = const()[name = string("op_2383_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2383_end_0 = const()[name = string("op_2383_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2383_end_mask_0 = const()[name = string("op_2383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2383_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = string("op_2390_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2390_end_0 = const()[name = string("op_2390_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2390_end_mask_0 = const()[name = string("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = var_2172_cast_fp16)[name = string("op_2390_cast_fp16")]; + tensor var_2397_begin_0 = const()[name = string("op_2397_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2397_end_0 = const()[name = string("op_2397_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2397_end_mask_0 = const()[name = string("op_2397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2397_cast_fp16 = slice_by_index(begin = var_2397_begin_0, end = var_2397_end_0, end_mask = var_2397_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2397_cast_fp16")]; + tensor var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2404_cast_fp16")]; + tensor var_2411_begin_0 = const()[name = string("op_2411_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2411_end_0 = const()[name = string("op_2411_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2411_end_mask_0 = const()[name = string("op_2411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2411_cast_fp16 = slice_by_index(begin = var_2411_begin_0, end = var_2411_end_0, end_mask = var_2411_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2411_cast_fp16")]; + tensor var_2418_begin_0 = const()[name = string("op_2418_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2418_end_0 = const()[name = string("op_2418_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2418_end_mask_0 = const()[name = string("op_2418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = var_2176_cast_fp16)[name = string("op_2418_cast_fp16")]; + tensor var_2425_begin_0 = const()[name = string("op_2425_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2425_end_0 = const()[name = string("op_2425_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2425_end_mask_0 = const()[name = string("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2425_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2439_begin_0 = const()[name = string("op_2439_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2439_end_0 = const()[name = string("op_2439_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2439_end_mask_0 = const()[name = string("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2439_cast_fp16")]; + tensor var_2446_begin_0 = const()[name = string("op_2446_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2446_end_0 = const()[name = string("op_2446_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2446_end_mask_0 = const()[name = string("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2180_cast_fp16)[name = string("op_2446_cast_fp16")]; + tensor var_2453_begin_0 = const()[name = string("op_2453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2453_end_0 = const()[name = string("op_2453_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2453_end_mask_0 = const()[name = string("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2453_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2467_begin_0 = const()[name = string("op_2467_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2467_end_0 = const()[name = string("op_2467_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2467_end_mask_0 = const()[name = string("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2467_cast_fp16")]; + tensor var_2474_begin_0 = const()[name = string("op_2474_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2474_end_0 = const()[name = string("op_2474_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2474_end_mask_0 = const()[name = string("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2184_cast_fp16)[name = string("op_2474_cast_fp16")]; + tensor var_2481_begin_0 = const()[name = string("op_2481_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2481_end_0 = const()[name = string("op_2481_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2481_end_mask_0 = const()[name = string("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2481_cast_fp16")]; + tensor var_2488_begin_0 = const()[name = string("op_2488_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2488_end_0 = const()[name = string("op_2488_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2488_end_mask_0 = const()[name = string("op_2488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2488_cast_fp16 = slice_by_index(begin = var_2488_begin_0, end = var_2488_end_0, end_mask = var_2488_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2488_cast_fp16")]; + tensor var_2495_begin_0 = const()[name = string("op_2495_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2495_end_0 = const()[name = string("op_2495_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2495_end_mask_0 = const()[name = string("op_2495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2495_cast_fp16 = slice_by_index(begin = var_2495_begin_0, end = var_2495_end_0, end_mask = var_2495_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2495_cast_fp16")]; + tensor var_2502_begin_0 = const()[name = string("op_2502_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2502_end_0 = const()[name = string("op_2502_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2502_end_mask_0 = const()[name = string("op_2502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = var_2188_cast_fp16)[name = string("op_2502_cast_fp16")]; + tensor var_2509_begin_0 = const()[name = string("op_2509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2509_end_0 = const()[name = string("op_2509_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2509_end_mask_0 = const()[name = string("op_2509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2509_cast_fp16 = slice_by_index(begin = var_2509_begin_0, end = var_2509_end_0, end_mask = var_2509_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2509_cast_fp16")]; + tensor var_2516_begin_0 = const()[name = string("op_2516_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2516_end_0 = const()[name = string("op_2516_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2516_end_mask_0 = const()[name = string("op_2516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2516_cast_fp16 = slice_by_index(begin = var_2516_begin_0, end = var_2516_end_0, end_mask = var_2516_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2516_cast_fp16")]; + tensor var_2523_begin_0 = const()[name = string("op_2523_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2523_end_0 = const()[name = string("op_2523_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2523_end_mask_0 = const()[name = string("op_2523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2523_cast_fp16 = slice_by_index(begin = var_2523_begin_0, end = var_2523_end_0, end_mask = var_2523_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2523_cast_fp16")]; + tensor var_2530_begin_0 = const()[name = string("op_2530_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2530_end_0 = const()[name = string("op_2530_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2530_end_mask_0 = const()[name = string("op_2530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2530_cast_fp16 = slice_by_index(begin = var_2530_begin_0, end = var_2530_end_0, end_mask = var_2530_end_mask_0, x = var_2192_cast_fp16)[name = string("op_2530_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2535_begin_0 = const()[name = string("op_2535_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2535_end_0 = const()[name = string("op_2535_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2535_end_mask_0 = const()[name = string("op_2535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_9")]; + tensor var_2535_cast_fp16 = slice_by_index(begin = var_2535_begin_0, end = var_2535_end_0, end_mask = var_2535_end_mask_0, x = k_5_cast_fp16)[name = string("op_2535_cast_fp16")]; + tensor var_2539_begin_0 = const()[name = string("op_2539_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2539_end_0 = const()[name = string("op_2539_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2539_end_mask_0 = const()[name = string("op_2539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2539_cast_fp16 = slice_by_index(begin = var_2539_begin_0, end = var_2539_end_0, end_mask = var_2539_end_mask_0, x = k_5_cast_fp16)[name = string("op_2539_cast_fp16")]; + tensor var_2543_begin_0 = const()[name = string("op_2543_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2543_end_0 = const()[name = string("op_2543_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2543_end_mask_0 = const()[name = string("op_2543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2543_cast_fp16 = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = k_5_cast_fp16)[name = string("op_2543_cast_fp16")]; + tensor var_2547_begin_0 = const()[name = string("op_2547_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2547_end_0 = const()[name = string("op_2547_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2547_end_mask_0 = const()[name = string("op_2547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2547_cast_fp16 = slice_by_index(begin = var_2547_begin_0, end = var_2547_end_0, end_mask = var_2547_end_mask_0, x = k_5_cast_fp16)[name = string("op_2547_cast_fp16")]; + tensor var_2551_begin_0 = const()[name = string("op_2551_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2551_end_0 = const()[name = string("op_2551_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2551_end_mask_0 = const()[name = string("op_2551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2551_cast_fp16 = slice_by_index(begin = var_2551_begin_0, end = var_2551_end_0, end_mask = var_2551_end_mask_0, x = k_5_cast_fp16)[name = string("op_2551_cast_fp16")]; + tensor var_2555_begin_0 = const()[name = string("op_2555_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2555_end_0 = const()[name = string("op_2555_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2555_end_mask_0 = const()[name = string("op_2555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2555_cast_fp16 = slice_by_index(begin = var_2555_begin_0, end = var_2555_end_0, end_mask = var_2555_end_mask_0, x = k_5_cast_fp16)[name = string("op_2555_cast_fp16")]; + tensor var_2559_begin_0 = const()[name = string("op_2559_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2559_end_0 = const()[name = string("op_2559_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2559_end_mask_0 = const()[name = string("op_2559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2559_cast_fp16 = slice_by_index(begin = var_2559_begin_0, end = var_2559_end_0, end_mask = var_2559_end_mask_0, x = k_5_cast_fp16)[name = string("op_2559_cast_fp16")]; + tensor var_2563_begin_0 = const()[name = string("op_2563_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2563_end_0 = const()[name = string("op_2563_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2563_end_mask_0 = const()[name = string("op_2563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2563_cast_fp16 = slice_by_index(begin = var_2563_begin_0, end = var_2563_end_0, end_mask = var_2563_end_mask_0, x = k_5_cast_fp16)[name = string("op_2563_cast_fp16")]; + tensor var_2567_begin_0 = const()[name = string("op_2567_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2567_end_0 = const()[name = string("op_2567_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_2567_end_mask_0 = const()[name = string("op_2567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = k_5_cast_fp16)[name = string("op_2567_cast_fp16")]; + tensor var_2571_begin_0 = const()[name = string("op_2571_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2571_end_0 = const()[name = string("op_2571_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_2571_end_mask_0 = const()[name = string("op_2571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2571_cast_fp16 = slice_by_index(begin = var_2571_begin_0, end = var_2571_end_0, end_mask = var_2571_end_mask_0, x = k_5_cast_fp16)[name = string("op_2571_cast_fp16")]; + tensor var_2575_begin_0 = const()[name = string("op_2575_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2575_end_0 = const()[name = string("op_2575_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_2575_end_mask_0 = const()[name = string("op_2575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2575_cast_fp16 = slice_by_index(begin = var_2575_begin_0, end = var_2575_end_0, end_mask = var_2575_end_mask_0, x = k_5_cast_fp16)[name = string("op_2575_cast_fp16")]; + tensor var_2579_begin_0 = const()[name = string("op_2579_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2579_end_0 = const()[name = string("op_2579_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_2579_end_mask_0 = const()[name = string("op_2579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2579_cast_fp16 = slice_by_index(begin = var_2579_begin_0, end = var_2579_end_0, end_mask = var_2579_end_mask_0, x = k_5_cast_fp16)[name = string("op_2579_cast_fp16")]; + tensor var_2581_begin_0 = const()[name = string("op_2581_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2581_end_0 = const()[name = string("op_2581_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2581_end_mask_0 = const()[name = string("op_2581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2581_cast_fp16 = slice_by_index(begin = var_2581_begin_0, end = var_2581_end_0, end_mask = var_2581_end_mask_0, x = value_5_cast_fp16)[name = string("op_2581_cast_fp16")]; + tensor var_2585_begin_0 = const()[name = string("op_2585_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2585_end_0 = const()[name = string("op_2585_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2585_end_mask_0 = const()[name = string("op_2585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2585_cast_fp16 = slice_by_index(begin = var_2585_begin_0, end = var_2585_end_0, end_mask = var_2585_end_mask_0, x = value_5_cast_fp16)[name = string("op_2585_cast_fp16")]; + tensor var_2589_begin_0 = const()[name = string("op_2589_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2589_end_0 = const()[name = string("op_2589_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2589_end_mask_0 = const()[name = string("op_2589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2589_cast_fp16 = slice_by_index(begin = var_2589_begin_0, end = var_2589_end_0, end_mask = var_2589_end_mask_0, x = value_5_cast_fp16)[name = string("op_2589_cast_fp16")]; + tensor var_2593_begin_0 = const()[name = string("op_2593_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2593_end_0 = const()[name = string("op_2593_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2593_end_mask_0 = const()[name = string("op_2593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2593_cast_fp16 = slice_by_index(begin = var_2593_begin_0, end = var_2593_end_0, end_mask = var_2593_end_mask_0, x = value_5_cast_fp16)[name = string("op_2593_cast_fp16")]; + tensor var_2597_begin_0 = const()[name = string("op_2597_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2597_end_0 = const()[name = string("op_2597_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2597_end_mask_0 = const()[name = string("op_2597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2597_cast_fp16 = slice_by_index(begin = var_2597_begin_0, end = var_2597_end_0, end_mask = var_2597_end_mask_0, x = value_5_cast_fp16)[name = string("op_2597_cast_fp16")]; + tensor var_2601_begin_0 = const()[name = string("op_2601_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2601_end_0 = const()[name = string("op_2601_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2601_end_mask_0 = const()[name = string("op_2601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2601_cast_fp16 = slice_by_index(begin = var_2601_begin_0, end = var_2601_end_0, end_mask = var_2601_end_mask_0, x = value_5_cast_fp16)[name = string("op_2601_cast_fp16")]; + tensor var_2605_begin_0 = const()[name = string("op_2605_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2605_end_0 = const()[name = string("op_2605_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2605_end_mask_0 = const()[name = string("op_2605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2605_cast_fp16 = slice_by_index(begin = var_2605_begin_0, end = var_2605_end_0, end_mask = var_2605_end_mask_0, x = value_5_cast_fp16)[name = string("op_2605_cast_fp16")]; + tensor var_2609_begin_0 = const()[name = string("op_2609_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2609_end_0 = const()[name = string("op_2609_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2609_end_mask_0 = const()[name = string("op_2609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2609_cast_fp16 = slice_by_index(begin = var_2609_begin_0, end = var_2609_end_0, end_mask = var_2609_end_mask_0, x = value_5_cast_fp16)[name = string("op_2609_cast_fp16")]; + tensor var_2613_begin_0 = const()[name = string("op_2613_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2613_end_0 = const()[name = string("op_2613_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2613_end_mask_0 = const()[name = string("op_2613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2613_cast_fp16 = slice_by_index(begin = var_2613_begin_0, end = var_2613_end_0, end_mask = var_2613_end_mask_0, x = value_5_cast_fp16)[name = string("op_2613_cast_fp16")]; + tensor var_2617_begin_0 = const()[name = string("op_2617_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2617_end_0 = const()[name = string("op_2617_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2617_end_mask_0 = const()[name = string("op_2617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2617_cast_fp16 = slice_by_index(begin = var_2617_begin_0, end = var_2617_end_0, end_mask = var_2617_end_mask_0, x = value_5_cast_fp16)[name = string("op_2617_cast_fp16")]; + tensor var_2621_begin_0 = const()[name = string("op_2621_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2621_end_0 = const()[name = string("op_2621_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2621_end_mask_0 = const()[name = string("op_2621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2621_cast_fp16 = slice_by_index(begin = var_2621_begin_0, end = var_2621_end_0, end_mask = var_2621_end_mask_0, x = value_5_cast_fp16)[name = string("op_2621_cast_fp16")]; + tensor var_2625_begin_0 = const()[name = string("op_2625_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2625_end_0 = const()[name = string("op_2625_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2625_end_mask_0 = const()[name = string("op_2625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2625_cast_fp16 = slice_by_index(begin = var_2625_begin_0, end = var_2625_end_0, end_mask = var_2625_end_mask_0, x = value_5_cast_fp16)[name = string("op_2625_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2535_cast_fp16, var_2201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2535_cast_fp16, var_2208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2535_cast_fp16, var_2215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2535_cast_fp16, var_2222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2539_cast_fp16, var_2229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2539_cast_fp16, var_2236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2539_cast_fp16, var_2243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2539_cast_fp16, var_2250_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2543_cast_fp16, var_2257_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2543_cast_fp16, var_2264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2543_cast_fp16, var_2271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2543_cast_fp16, var_2278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2547_cast_fp16, var_2285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2547_cast_fp16, var_2292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2547_cast_fp16, var_2299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2547_cast_fp16, var_2306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2551_cast_fp16, var_2313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2551_cast_fp16, var_2320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2551_cast_fp16, var_2327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2551_cast_fp16, var_2334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2555_cast_fp16, var_2341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2555_cast_fp16, var_2348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2555_cast_fp16, var_2355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2555_cast_fp16, var_2362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2559_cast_fp16, var_2369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2559_cast_fp16, var_2376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2559_cast_fp16, var_2383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2559_cast_fp16, var_2390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2563_cast_fp16, var_2397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2563_cast_fp16, var_2404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2563_cast_fp16, var_2411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2563_cast_fp16, var_2418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_2567_cast_fp16, var_2425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_2567_cast_fp16, var_2432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_2567_cast_fp16, var_2439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_2567_cast_fp16, var_2446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_2571_cast_fp16, var_2453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_2571_cast_fp16, var_2460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_2571_cast_fp16, var_2467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_2571_cast_fp16, var_2474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_2575_cast_fp16, var_2481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_2575_cast_fp16, var_2488_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_2575_cast_fp16, var_2495_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_2575_cast_fp16, var_2502_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_2579_cast_fp16, var_2509_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_2579_cast_fp16, var_2516_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_2579_cast_fp16, var_2523_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_2579_cast_fp16, var_2530_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + fp16 var_2724_to_fp16 = const()[name = string("op_2724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2724_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2726_to_fp16 = const()[name = string("op_2726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2726_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2728_to_fp16 = const()[name = string("op_2728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2728_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2730_to_fp16 = const()[name = string("op_2730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2730_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2732_to_fp16 = const()[name = string("op_2732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2732_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2734_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2736_to_fp16 = const()[name = string("op_2736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2736_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2738_to_fp16 = const()[name = string("op_2738_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2738_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2740_to_fp16 = const()[name = string("op_2740_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2740_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2742_to_fp16 = const()[name = string("op_2742_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2742_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2744_to_fp16 = const()[name = string("op_2744_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2744_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2746_to_fp16 = const()[name = string("op_2746_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2746_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2748_to_fp16 = const()[name = string("op_2748_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2748_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2750_to_fp16 = const()[name = string("op_2750_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2750_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2752_to_fp16 = const()[name = string("op_2752_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2752_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2754_to_fp16 = const()[name = string("op_2754_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2754_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2756_to_fp16 = const()[name = string("op_2756_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2756_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2758_to_fp16 = const()[name = string("op_2758_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2758_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2760_to_fp16 = const()[name = string("op_2760_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2760_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2762_to_fp16 = const()[name = string("op_2762_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2762_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2764_to_fp16 = const()[name = string("op_2764_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2764_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2766_to_fp16 = const()[name = string("op_2766_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2766_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2768_to_fp16 = const()[name = string("op_2768_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2768_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2770_to_fp16 = const()[name = string("op_2770_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2770_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2772_to_fp16 = const()[name = string("op_2772_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2772_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2774_to_fp16 = const()[name = string("op_2774_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2774_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2776_to_fp16 = const()[name = string("op_2776_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2776_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2778_to_fp16 = const()[name = string("op_2778_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2778_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2780_to_fp16 = const()[name = string("op_2780_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2780_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2782_to_fp16 = const()[name = string("op_2782_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2782_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2784_to_fp16 = const()[name = string("op_2784_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2784_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2786_to_fp16 = const()[name = string("op_2786_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2786_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + fp16 var_2788_to_fp16 = const()[name = string("op_2788_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_2788_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_2790_to_fp16 = const()[name = string("op_2790_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_2790_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_2792_to_fp16 = const()[name = string("op_2792_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_2792_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_2794_to_fp16 = const()[name = string("op_2794_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_2794_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_2796_to_fp16 = const()[name = string("op_2796_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_2796_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_2798_to_fp16 = const()[name = string("op_2798_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_2798_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_2800_to_fp16 = const()[name = string("op_2800_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_2800_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_2802_to_fp16 = const()[name = string("op_2802_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_2802_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_2804_to_fp16 = const()[name = string("op_2804_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_2804_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_2806_to_fp16 = const()[name = string("op_2806_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_2806_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_2808_to_fp16 = const()[name = string("op_2808_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_2808_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_2810_to_fp16 = const()[name = string("op_2810_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_2810_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_2812_to_fp16 = const()[name = string("op_2812_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_2812_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_2814_to_fp16 = const()[name = string("op_2814_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_2814_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_2816_to_fp16 = const()[name = string("op_2816_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_2816_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_2818_to_fp16 = const()[name = string("op_2818_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_2818_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + tensor var_2820_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_193_cast_fp16)[name = string("op_2820_cast_fp16")]; + tensor var_2821_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_195_cast_fp16)[name = string("op_2821_cast_fp16")]; + tensor var_2822_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_197_cast_fp16)[name = string("op_2822_cast_fp16")]; + tensor var_2823_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_199_cast_fp16)[name = string("op_2823_cast_fp16")]; + tensor var_2824_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_201_cast_fp16)[name = string("op_2824_cast_fp16")]; + tensor var_2825_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_203_cast_fp16)[name = string("op_2825_cast_fp16")]; + tensor var_2826_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_205_cast_fp16)[name = string("op_2826_cast_fp16")]; + tensor var_2827_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_207_cast_fp16)[name = string("op_2827_cast_fp16")]; + tensor var_2828_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_209_cast_fp16)[name = string("op_2828_cast_fp16")]; + tensor var_2829_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_211_cast_fp16)[name = string("op_2829_cast_fp16")]; + tensor var_2830_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_213_cast_fp16)[name = string("op_2830_cast_fp16")]; + tensor var_2831_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_215_cast_fp16)[name = string("op_2831_cast_fp16")]; + tensor var_2832_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_217_cast_fp16)[name = string("op_2832_cast_fp16")]; + tensor var_2833_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_219_cast_fp16)[name = string("op_2833_cast_fp16")]; + tensor var_2834_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_221_cast_fp16)[name = string("op_2834_cast_fp16")]; + tensor var_2835_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_223_cast_fp16)[name = string("op_2835_cast_fp16")]; + tensor var_2836_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_225_cast_fp16)[name = string("op_2836_cast_fp16")]; + tensor var_2837_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_227_cast_fp16)[name = string("op_2837_cast_fp16")]; + tensor var_2838_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_229_cast_fp16)[name = string("op_2838_cast_fp16")]; + tensor var_2839_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_231_cast_fp16)[name = string("op_2839_cast_fp16")]; + tensor var_2840_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_233_cast_fp16)[name = string("op_2840_cast_fp16")]; + tensor var_2841_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_235_cast_fp16)[name = string("op_2841_cast_fp16")]; + tensor var_2842_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_237_cast_fp16)[name = string("op_2842_cast_fp16")]; + tensor var_2843_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_239_cast_fp16)[name = string("op_2843_cast_fp16")]; + tensor var_2844_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_241_cast_fp16)[name = string("op_2844_cast_fp16")]; + tensor var_2845_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_243_cast_fp16)[name = string("op_2845_cast_fp16")]; + tensor var_2846_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_245_cast_fp16)[name = string("op_2846_cast_fp16")]; + tensor var_2847_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_247_cast_fp16)[name = string("op_2847_cast_fp16")]; + tensor var_2848_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_249_cast_fp16)[name = string("op_2848_cast_fp16")]; + tensor var_2849_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_251_cast_fp16)[name = string("op_2849_cast_fp16")]; + tensor var_2850_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_253_cast_fp16)[name = string("op_2850_cast_fp16")]; + tensor var_2851_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_255_cast_fp16)[name = string("op_2851_cast_fp16")]; + tensor var_2852_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_257_cast_fp16)[name = string("op_2852_cast_fp16")]; + tensor var_2853_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_259_cast_fp16)[name = string("op_2853_cast_fp16")]; + tensor var_2854_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_261_cast_fp16)[name = string("op_2854_cast_fp16")]; + tensor var_2855_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_263_cast_fp16)[name = string("op_2855_cast_fp16")]; + tensor var_2856_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_265_cast_fp16)[name = string("op_2856_cast_fp16")]; + tensor var_2857_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_267_cast_fp16)[name = string("op_2857_cast_fp16")]; + tensor var_2858_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_269_cast_fp16)[name = string("op_2858_cast_fp16")]; + tensor var_2859_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_271_cast_fp16)[name = string("op_2859_cast_fp16")]; + tensor var_2860_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_273_cast_fp16)[name = string("op_2860_cast_fp16")]; + tensor var_2861_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_275_cast_fp16)[name = string("op_2861_cast_fp16")]; + tensor var_2862_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_277_cast_fp16)[name = string("op_2862_cast_fp16")]; + tensor var_2863_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_279_cast_fp16)[name = string("op_2863_cast_fp16")]; + tensor var_2864_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_281_cast_fp16)[name = string("op_2864_cast_fp16")]; + tensor var_2865_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_283_cast_fp16)[name = string("op_2865_cast_fp16")]; + tensor var_2866_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_285_cast_fp16)[name = string("op_2866_cast_fp16")]; + tensor var_2867_cast_fp16 = softmax(axis = var_2093, x = aw_chunk_287_cast_fp16)[name = string("op_2867_cast_fp16")]; + string var_2869_equation_0 = const()[name = string("op_2869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2869_cast_fp16 = einsum(equation = var_2869_equation_0, values = (var_2581_cast_fp16, var_2820_cast_fp16))[name = string("op_2869_cast_fp16")]; + string var_2871_equation_0 = const()[name = string("op_2871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2871_cast_fp16 = einsum(equation = var_2871_equation_0, values = (var_2581_cast_fp16, var_2821_cast_fp16))[name = string("op_2871_cast_fp16")]; + string var_2873_equation_0 = const()[name = string("op_2873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2873_cast_fp16 = einsum(equation = var_2873_equation_0, values = (var_2581_cast_fp16, var_2822_cast_fp16))[name = string("op_2873_cast_fp16")]; + string var_2875_equation_0 = const()[name = string("op_2875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2875_cast_fp16 = einsum(equation = var_2875_equation_0, values = (var_2581_cast_fp16, var_2823_cast_fp16))[name = string("op_2875_cast_fp16")]; + string var_2877_equation_0 = const()[name = string("op_2877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2877_cast_fp16 = einsum(equation = var_2877_equation_0, values = (var_2585_cast_fp16, var_2824_cast_fp16))[name = string("op_2877_cast_fp16")]; + string var_2879_equation_0 = const()[name = string("op_2879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2879_cast_fp16 = einsum(equation = var_2879_equation_0, values = (var_2585_cast_fp16, var_2825_cast_fp16))[name = string("op_2879_cast_fp16")]; + string var_2881_equation_0 = const()[name = string("op_2881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2881_cast_fp16 = einsum(equation = var_2881_equation_0, values = (var_2585_cast_fp16, var_2826_cast_fp16))[name = string("op_2881_cast_fp16")]; + string var_2883_equation_0 = const()[name = string("op_2883_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2883_cast_fp16 = einsum(equation = var_2883_equation_0, values = (var_2585_cast_fp16, var_2827_cast_fp16))[name = string("op_2883_cast_fp16")]; + string var_2885_equation_0 = const()[name = string("op_2885_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2885_cast_fp16 = einsum(equation = var_2885_equation_0, values = (var_2589_cast_fp16, var_2828_cast_fp16))[name = string("op_2885_cast_fp16")]; + string var_2887_equation_0 = const()[name = string("op_2887_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2887_cast_fp16 = einsum(equation = var_2887_equation_0, values = (var_2589_cast_fp16, var_2829_cast_fp16))[name = string("op_2887_cast_fp16")]; + string var_2889_equation_0 = const()[name = string("op_2889_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2889_cast_fp16 = einsum(equation = var_2889_equation_0, values = (var_2589_cast_fp16, var_2830_cast_fp16))[name = string("op_2889_cast_fp16")]; + string var_2891_equation_0 = const()[name = string("op_2891_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2891_cast_fp16 = einsum(equation = var_2891_equation_0, values = (var_2589_cast_fp16, var_2831_cast_fp16))[name = string("op_2891_cast_fp16")]; + string var_2893_equation_0 = const()[name = string("op_2893_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2893_cast_fp16 = einsum(equation = var_2893_equation_0, values = (var_2593_cast_fp16, var_2832_cast_fp16))[name = string("op_2893_cast_fp16")]; + string var_2895_equation_0 = const()[name = string("op_2895_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2895_cast_fp16 = einsum(equation = var_2895_equation_0, values = (var_2593_cast_fp16, var_2833_cast_fp16))[name = string("op_2895_cast_fp16")]; + string var_2897_equation_0 = const()[name = string("op_2897_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2897_cast_fp16 = einsum(equation = var_2897_equation_0, values = (var_2593_cast_fp16, var_2834_cast_fp16))[name = string("op_2897_cast_fp16")]; + string var_2899_equation_0 = const()[name = string("op_2899_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2899_cast_fp16 = einsum(equation = var_2899_equation_0, values = (var_2593_cast_fp16, var_2835_cast_fp16))[name = string("op_2899_cast_fp16")]; + string var_2901_equation_0 = const()[name = string("op_2901_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2901_cast_fp16 = einsum(equation = var_2901_equation_0, values = (var_2597_cast_fp16, var_2836_cast_fp16))[name = string("op_2901_cast_fp16")]; + string var_2903_equation_0 = const()[name = string("op_2903_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2903_cast_fp16 = einsum(equation = var_2903_equation_0, values = (var_2597_cast_fp16, var_2837_cast_fp16))[name = string("op_2903_cast_fp16")]; + string var_2905_equation_0 = const()[name = string("op_2905_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2905_cast_fp16 = einsum(equation = var_2905_equation_0, values = (var_2597_cast_fp16, var_2838_cast_fp16))[name = string("op_2905_cast_fp16")]; + string var_2907_equation_0 = const()[name = string("op_2907_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2907_cast_fp16 = einsum(equation = var_2907_equation_0, values = (var_2597_cast_fp16, var_2839_cast_fp16))[name = string("op_2907_cast_fp16")]; + string var_2909_equation_0 = const()[name = string("op_2909_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2909_cast_fp16 = einsum(equation = var_2909_equation_0, values = (var_2601_cast_fp16, var_2840_cast_fp16))[name = string("op_2909_cast_fp16")]; + string var_2911_equation_0 = const()[name = string("op_2911_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2911_cast_fp16 = einsum(equation = var_2911_equation_0, values = (var_2601_cast_fp16, var_2841_cast_fp16))[name = string("op_2911_cast_fp16")]; + string var_2913_equation_0 = const()[name = string("op_2913_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2913_cast_fp16 = einsum(equation = var_2913_equation_0, values = (var_2601_cast_fp16, var_2842_cast_fp16))[name = string("op_2913_cast_fp16")]; + string var_2915_equation_0 = const()[name = string("op_2915_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2915_cast_fp16 = einsum(equation = var_2915_equation_0, values = (var_2601_cast_fp16, var_2843_cast_fp16))[name = string("op_2915_cast_fp16")]; + string var_2917_equation_0 = const()[name = string("op_2917_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2917_cast_fp16 = einsum(equation = var_2917_equation_0, values = (var_2605_cast_fp16, var_2844_cast_fp16))[name = string("op_2917_cast_fp16")]; + string var_2919_equation_0 = const()[name = string("op_2919_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2919_cast_fp16 = einsum(equation = var_2919_equation_0, values = (var_2605_cast_fp16, var_2845_cast_fp16))[name = string("op_2919_cast_fp16")]; + string var_2921_equation_0 = const()[name = string("op_2921_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2921_cast_fp16 = einsum(equation = var_2921_equation_0, values = (var_2605_cast_fp16, var_2846_cast_fp16))[name = string("op_2921_cast_fp16")]; + string var_2923_equation_0 = const()[name = string("op_2923_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2923_cast_fp16 = einsum(equation = var_2923_equation_0, values = (var_2605_cast_fp16, var_2847_cast_fp16))[name = string("op_2923_cast_fp16")]; + string var_2925_equation_0 = const()[name = string("op_2925_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2925_cast_fp16 = einsum(equation = var_2925_equation_0, values = (var_2609_cast_fp16, var_2848_cast_fp16))[name = string("op_2925_cast_fp16")]; + string var_2927_equation_0 = const()[name = string("op_2927_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2927_cast_fp16 = einsum(equation = var_2927_equation_0, values = (var_2609_cast_fp16, var_2849_cast_fp16))[name = string("op_2927_cast_fp16")]; + string var_2929_equation_0 = const()[name = string("op_2929_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2929_cast_fp16 = einsum(equation = var_2929_equation_0, values = (var_2609_cast_fp16, var_2850_cast_fp16))[name = string("op_2929_cast_fp16")]; + string var_2931_equation_0 = const()[name = string("op_2931_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2931_cast_fp16 = einsum(equation = var_2931_equation_0, values = (var_2609_cast_fp16, var_2851_cast_fp16))[name = string("op_2931_cast_fp16")]; + string var_2933_equation_0 = const()[name = string("op_2933_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2933_cast_fp16 = einsum(equation = var_2933_equation_0, values = (var_2613_cast_fp16, var_2852_cast_fp16))[name = string("op_2933_cast_fp16")]; + string var_2935_equation_0 = const()[name = string("op_2935_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2935_cast_fp16 = einsum(equation = var_2935_equation_0, values = (var_2613_cast_fp16, var_2853_cast_fp16))[name = string("op_2935_cast_fp16")]; + string var_2937_equation_0 = const()[name = string("op_2937_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2937_cast_fp16 = einsum(equation = var_2937_equation_0, values = (var_2613_cast_fp16, var_2854_cast_fp16))[name = string("op_2937_cast_fp16")]; + string var_2939_equation_0 = const()[name = string("op_2939_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2939_cast_fp16 = einsum(equation = var_2939_equation_0, values = (var_2613_cast_fp16, var_2855_cast_fp16))[name = string("op_2939_cast_fp16")]; + string var_2941_equation_0 = const()[name = string("op_2941_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2941_cast_fp16 = einsum(equation = var_2941_equation_0, values = (var_2617_cast_fp16, var_2856_cast_fp16))[name = string("op_2941_cast_fp16")]; + string var_2943_equation_0 = const()[name = string("op_2943_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2943_cast_fp16 = einsum(equation = var_2943_equation_0, values = (var_2617_cast_fp16, var_2857_cast_fp16))[name = string("op_2943_cast_fp16")]; + string var_2945_equation_0 = const()[name = string("op_2945_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2945_cast_fp16 = einsum(equation = var_2945_equation_0, values = (var_2617_cast_fp16, var_2858_cast_fp16))[name = string("op_2945_cast_fp16")]; + string var_2947_equation_0 = const()[name = string("op_2947_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2947_cast_fp16 = einsum(equation = var_2947_equation_0, values = (var_2617_cast_fp16, var_2859_cast_fp16))[name = string("op_2947_cast_fp16")]; + string var_2949_equation_0 = const()[name = string("op_2949_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2949_cast_fp16 = einsum(equation = var_2949_equation_0, values = (var_2621_cast_fp16, var_2860_cast_fp16))[name = string("op_2949_cast_fp16")]; + string var_2951_equation_0 = const()[name = string("op_2951_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2951_cast_fp16 = einsum(equation = var_2951_equation_0, values = (var_2621_cast_fp16, var_2861_cast_fp16))[name = string("op_2951_cast_fp16")]; + string var_2953_equation_0 = const()[name = string("op_2953_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2953_cast_fp16 = einsum(equation = var_2953_equation_0, values = (var_2621_cast_fp16, var_2862_cast_fp16))[name = string("op_2953_cast_fp16")]; + string var_2955_equation_0 = const()[name = string("op_2955_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2955_cast_fp16 = einsum(equation = var_2955_equation_0, values = (var_2621_cast_fp16, var_2863_cast_fp16))[name = string("op_2955_cast_fp16")]; + string var_2957_equation_0 = const()[name = string("op_2957_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2957_cast_fp16 = einsum(equation = var_2957_equation_0, values = (var_2625_cast_fp16, var_2864_cast_fp16))[name = string("op_2957_cast_fp16")]; + string var_2959_equation_0 = const()[name = string("op_2959_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2959_cast_fp16 = einsum(equation = var_2959_equation_0, values = (var_2625_cast_fp16, var_2865_cast_fp16))[name = string("op_2959_cast_fp16")]; + string var_2961_equation_0 = const()[name = string("op_2961_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2961_cast_fp16 = einsum(equation = var_2961_equation_0, values = (var_2625_cast_fp16, var_2866_cast_fp16))[name = string("op_2961_cast_fp16")]; + string var_2963_equation_0 = const()[name = string("op_2963_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2963_cast_fp16 = einsum(equation = var_2963_equation_0, values = (var_2625_cast_fp16, var_2867_cast_fp16))[name = string("op_2963_cast_fp16")]; + bool var_2965_interleave_0 = const()[name = string("op_2965_interleave_0"), val = bool(false)]; + tensor var_2965_cast_fp16 = concat(axis = var_2076, interleave = var_2965_interleave_0, values = (var_2869_cast_fp16, var_2871_cast_fp16, var_2873_cast_fp16, var_2875_cast_fp16))[name = string("op_2965_cast_fp16")]; + bool var_2967_interleave_0 = const()[name = string("op_2967_interleave_0"), val = bool(false)]; + tensor var_2967_cast_fp16 = concat(axis = var_2076, interleave = var_2967_interleave_0, values = (var_2877_cast_fp16, var_2879_cast_fp16, var_2881_cast_fp16, var_2883_cast_fp16))[name = string("op_2967_cast_fp16")]; + bool var_2969_interleave_0 = const()[name = string("op_2969_interleave_0"), val = bool(false)]; + tensor var_2969_cast_fp16 = concat(axis = var_2076, interleave = var_2969_interleave_0, values = (var_2885_cast_fp16, var_2887_cast_fp16, var_2889_cast_fp16, var_2891_cast_fp16))[name = string("op_2969_cast_fp16")]; + bool var_2971_interleave_0 = const()[name = string("op_2971_interleave_0"), val = bool(false)]; + tensor var_2971_cast_fp16 = concat(axis = var_2076, interleave = var_2971_interleave_0, values = (var_2893_cast_fp16, var_2895_cast_fp16, var_2897_cast_fp16, var_2899_cast_fp16))[name = string("op_2971_cast_fp16")]; + bool var_2973_interleave_0 = const()[name = string("op_2973_interleave_0"), val = bool(false)]; + tensor var_2973_cast_fp16 = concat(axis = var_2076, interleave = var_2973_interleave_0, values = (var_2901_cast_fp16, var_2903_cast_fp16, var_2905_cast_fp16, var_2907_cast_fp16))[name = string("op_2973_cast_fp16")]; + bool var_2975_interleave_0 = const()[name = string("op_2975_interleave_0"), val = bool(false)]; + tensor var_2975_cast_fp16 = concat(axis = var_2076, interleave = var_2975_interleave_0, values = (var_2909_cast_fp16, var_2911_cast_fp16, var_2913_cast_fp16, var_2915_cast_fp16))[name = string("op_2975_cast_fp16")]; + bool var_2977_interleave_0 = const()[name = string("op_2977_interleave_0"), val = bool(false)]; + tensor var_2977_cast_fp16 = concat(axis = var_2076, interleave = var_2977_interleave_0, values = (var_2917_cast_fp16, var_2919_cast_fp16, var_2921_cast_fp16, var_2923_cast_fp16))[name = string("op_2977_cast_fp16")]; + bool var_2979_interleave_0 = const()[name = string("op_2979_interleave_0"), val = bool(false)]; + tensor var_2979_cast_fp16 = concat(axis = var_2076, interleave = var_2979_interleave_0, values = (var_2925_cast_fp16, var_2927_cast_fp16, var_2929_cast_fp16, var_2931_cast_fp16))[name = string("op_2979_cast_fp16")]; + bool var_2981_interleave_0 = const()[name = string("op_2981_interleave_0"), val = bool(false)]; + tensor var_2981_cast_fp16 = concat(axis = var_2076, interleave = var_2981_interleave_0, values = (var_2933_cast_fp16, var_2935_cast_fp16, var_2937_cast_fp16, var_2939_cast_fp16))[name = string("op_2981_cast_fp16")]; + bool var_2983_interleave_0 = const()[name = string("op_2983_interleave_0"), val = bool(false)]; + tensor var_2983_cast_fp16 = concat(axis = var_2076, interleave = var_2983_interleave_0, values = (var_2941_cast_fp16, var_2943_cast_fp16, var_2945_cast_fp16, var_2947_cast_fp16))[name = string("op_2983_cast_fp16")]; + bool var_2985_interleave_0 = const()[name = string("op_2985_interleave_0"), val = bool(false)]; + tensor var_2985_cast_fp16 = concat(axis = var_2076, interleave = var_2985_interleave_0, values = (var_2949_cast_fp16, var_2951_cast_fp16, var_2953_cast_fp16, var_2955_cast_fp16))[name = string("op_2985_cast_fp16")]; + bool var_2987_interleave_0 = const()[name = string("op_2987_interleave_0"), val = bool(false)]; + tensor var_2987_cast_fp16 = concat(axis = var_2076, interleave = var_2987_interleave_0, values = (var_2957_cast_fp16, var_2959_cast_fp16, var_2961_cast_fp16, var_2963_cast_fp16))[name = string("op_2987_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_2093, interleave = input_17_interleave_0, values = (var_2965_cast_fp16, var_2967_cast_fp16, var_2969_cast_fp16, var_2971_cast_fp16, var_2973_cast_fp16, var_2975_cast_fp16, var_2977_cast_fp16, var_2979_cast_fp16, var_2981_cast_fp16, var_2983_cast_fp16, var_2985_cast_fp16, var_2987_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38114112)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39293824)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_3006_to_fp16 = const()[name = string("op_3006_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_3006_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39295424)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39297024)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39298624)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44017280)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44023488)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48742144)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_3035 = const()[name = string("op_3035"), val = int32(3)]; + int32 var_3052 = const()[name = string("op_3052"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_3069_to_fp16 = const()[name = string("op_3069_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_3069_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48743744)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48745344)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48746944)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49926656)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49928256)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51107968)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52287680)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_3107_begin_0 = const()[name = string("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3107_end_0 = const()[name = string("op_3107_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3107_end_mask_0 = const()[name = string("op_3107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = query_7_cast_fp16)[name = string("op_3107_cast_fp16")]; + tensor var_3111_begin_0 = const()[name = string("op_3111_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3111_end_0 = const()[name = string("op_3111_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3111_end_mask_0 = const()[name = string("op_3111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3111_cast_fp16 = slice_by_index(begin = var_3111_begin_0, end = var_3111_end_0, end_mask = var_3111_end_mask_0, x = query_7_cast_fp16)[name = string("op_3111_cast_fp16")]; + tensor var_3115_begin_0 = const()[name = string("op_3115_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3115_end_0 = const()[name = string("op_3115_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3115_end_mask_0 = const()[name = string("op_3115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3115_cast_fp16 = slice_by_index(begin = var_3115_begin_0, end = var_3115_end_0, end_mask = var_3115_end_mask_0, x = query_7_cast_fp16)[name = string("op_3115_cast_fp16")]; + tensor var_3119_begin_0 = const()[name = string("op_3119_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3119_end_0 = const()[name = string("op_3119_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3119_end_mask_0 = const()[name = string("op_3119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3119_cast_fp16 = slice_by_index(begin = var_3119_begin_0, end = var_3119_end_0, end_mask = var_3119_end_mask_0, x = query_7_cast_fp16)[name = string("op_3119_cast_fp16")]; + tensor var_3123_begin_0 = const()[name = string("op_3123_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3123_end_0 = const()[name = string("op_3123_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3123_end_mask_0 = const()[name = string("op_3123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3123_cast_fp16 = slice_by_index(begin = var_3123_begin_0, end = var_3123_end_0, end_mask = var_3123_end_mask_0, x = query_7_cast_fp16)[name = string("op_3123_cast_fp16")]; + tensor var_3127_begin_0 = const()[name = string("op_3127_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3127_end_0 = const()[name = string("op_3127_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3127_end_mask_0 = const()[name = string("op_3127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3127_cast_fp16 = slice_by_index(begin = var_3127_begin_0, end = var_3127_end_0, end_mask = var_3127_end_mask_0, x = query_7_cast_fp16)[name = string("op_3127_cast_fp16")]; + tensor var_3131_begin_0 = const()[name = string("op_3131_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3131_end_0 = const()[name = string("op_3131_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3131_end_mask_0 = const()[name = string("op_3131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3131_cast_fp16 = slice_by_index(begin = var_3131_begin_0, end = var_3131_end_0, end_mask = var_3131_end_mask_0, x = query_7_cast_fp16)[name = string("op_3131_cast_fp16")]; + tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = query_7_cast_fp16)[name = string("op_3135_cast_fp16")]; + tensor var_3139_begin_0 = const()[name = string("op_3139_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3139_end_0 = const()[name = string("op_3139_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3139_end_mask_0 = const()[name = string("op_3139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3139_cast_fp16 = slice_by_index(begin = var_3139_begin_0, end = var_3139_end_0, end_mask = var_3139_end_mask_0, x = query_7_cast_fp16)[name = string("op_3139_cast_fp16")]; + tensor var_3143_begin_0 = const()[name = string("op_3143_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3143_end_0 = const()[name = string("op_3143_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3143_end_mask_0 = const()[name = string("op_3143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3143_cast_fp16 = slice_by_index(begin = var_3143_begin_0, end = var_3143_end_0, end_mask = var_3143_end_mask_0, x = query_7_cast_fp16)[name = string("op_3143_cast_fp16")]; + tensor var_3147_begin_0 = const()[name = string("op_3147_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3147_end_0 = const()[name = string("op_3147_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3147_end_mask_0 = const()[name = string("op_3147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3147_cast_fp16 = slice_by_index(begin = var_3147_begin_0, end = var_3147_end_0, end_mask = var_3147_end_mask_0, x = query_7_cast_fp16)[name = string("op_3147_cast_fp16")]; + tensor var_3151_begin_0 = const()[name = string("op_3151_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3151_end_0 = const()[name = string("op_3151_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3151_end_mask_0 = const()[name = string("op_3151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3151_cast_fp16 = slice_by_index(begin = var_3151_begin_0, end = var_3151_end_0, end_mask = var_3151_end_mask_0, x = query_7_cast_fp16)[name = string("op_3151_cast_fp16")]; + tensor var_3160_begin_0 = const()[name = string("op_3160_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3160_end_0 = const()[name = string("op_3160_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3160_end_mask_0 = const()[name = string("op_3160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3160_cast_fp16 = slice_by_index(begin = var_3160_begin_0, end = var_3160_end_0, end_mask = var_3160_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3160_cast_fp16")]; + tensor var_3167_begin_0 = const()[name = string("op_3167_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3167_end_0 = const()[name = string("op_3167_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3167_end_mask_0 = const()[name = string("op_3167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3167_cast_fp16 = slice_by_index(begin = var_3167_begin_0, end = var_3167_end_0, end_mask = var_3167_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3167_cast_fp16")]; + tensor var_3174_begin_0 = const()[name = string("op_3174_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3174_end_0 = const()[name = string("op_3174_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3174_end_mask_0 = const()[name = string("op_3174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3174_cast_fp16 = slice_by_index(begin = var_3174_begin_0, end = var_3174_end_0, end_mask = var_3174_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3174_cast_fp16")]; + tensor var_3181_begin_0 = const()[name = string("op_3181_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3181_end_0 = const()[name = string("op_3181_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3181_end_mask_0 = const()[name = string("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3181_cast_fp16 = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = var_3107_cast_fp16)[name = string("op_3181_cast_fp16")]; + tensor var_3188_begin_0 = const()[name = string("op_3188_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3188_end_0 = const()[name = string("op_3188_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3188_end_mask_0 = const()[name = string("op_3188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3188_cast_fp16 = slice_by_index(begin = var_3188_begin_0, end = var_3188_end_0, end_mask = var_3188_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3188_cast_fp16")]; + tensor var_3195_begin_0 = const()[name = string("op_3195_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3195_end_0 = const()[name = string("op_3195_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3195_end_mask_0 = const()[name = string("op_3195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3195_cast_fp16 = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3195_cast_fp16")]; + tensor var_3202_begin_0 = const()[name = string("op_3202_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3202_end_0 = const()[name = string("op_3202_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3202_end_mask_0 = const()[name = string("op_3202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3202_cast_fp16 = slice_by_index(begin = var_3202_begin_0, end = var_3202_end_0, end_mask = var_3202_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3202_cast_fp16")]; + tensor var_3209_begin_0 = const()[name = string("op_3209_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3209_end_0 = const()[name = string("op_3209_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3209_end_mask_0 = const()[name = string("op_3209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3209_cast_fp16 = slice_by_index(begin = var_3209_begin_0, end = var_3209_end_0, end_mask = var_3209_end_mask_0, x = var_3111_cast_fp16)[name = string("op_3209_cast_fp16")]; + tensor var_3216_begin_0 = const()[name = string("op_3216_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3216_end_0 = const()[name = string("op_3216_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3216_end_mask_0 = const()[name = string("op_3216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3216_cast_fp16 = slice_by_index(begin = var_3216_begin_0, end = var_3216_end_0, end_mask = var_3216_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3216_cast_fp16")]; + tensor var_3223_begin_0 = const()[name = string("op_3223_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3223_end_0 = const()[name = string("op_3223_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3223_end_mask_0 = const()[name = string("op_3223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3223_cast_fp16 = slice_by_index(begin = var_3223_begin_0, end = var_3223_end_0, end_mask = var_3223_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3223_cast_fp16")]; + tensor var_3230_begin_0 = const()[name = string("op_3230_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3230_end_0 = const()[name = string("op_3230_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3230_end_mask_0 = const()[name = string("op_3230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3230_cast_fp16 = slice_by_index(begin = var_3230_begin_0, end = var_3230_end_0, end_mask = var_3230_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3230_cast_fp16")]; + tensor var_3237_begin_0 = const()[name = string("op_3237_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3237_end_0 = const()[name = string("op_3237_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3237_end_mask_0 = const()[name = string("op_3237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3237_cast_fp16 = slice_by_index(begin = var_3237_begin_0, end = var_3237_end_0, end_mask = var_3237_end_mask_0, x = var_3115_cast_fp16)[name = string("op_3237_cast_fp16")]; + tensor var_3244_begin_0 = const()[name = string("op_3244_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3244_end_0 = const()[name = string("op_3244_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3244_end_mask_0 = const()[name = string("op_3244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3244_cast_fp16 = slice_by_index(begin = var_3244_begin_0, end = var_3244_end_0, end_mask = var_3244_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3244_cast_fp16")]; + tensor var_3251_begin_0 = const()[name = string("op_3251_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3251_end_0 = const()[name = string("op_3251_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3251_end_mask_0 = const()[name = string("op_3251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3251_cast_fp16 = slice_by_index(begin = var_3251_begin_0, end = var_3251_end_0, end_mask = var_3251_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3251_cast_fp16")]; + tensor var_3258_begin_0 = const()[name = string("op_3258_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3258_end_0 = const()[name = string("op_3258_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3258_end_mask_0 = const()[name = string("op_3258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3258_cast_fp16 = slice_by_index(begin = var_3258_begin_0, end = var_3258_end_0, end_mask = var_3258_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3258_cast_fp16")]; + tensor var_3265_begin_0 = const()[name = string("op_3265_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3265_end_0 = const()[name = string("op_3265_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3265_end_mask_0 = const()[name = string("op_3265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3265_cast_fp16 = slice_by_index(begin = var_3265_begin_0, end = var_3265_end_0, end_mask = var_3265_end_mask_0, x = var_3119_cast_fp16)[name = string("op_3265_cast_fp16")]; + tensor var_3272_begin_0 = const()[name = string("op_3272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3272_end_0 = const()[name = string("op_3272_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3272_end_mask_0 = const()[name = string("op_3272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3272_cast_fp16 = slice_by_index(begin = var_3272_begin_0, end = var_3272_end_0, end_mask = var_3272_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3272_cast_fp16")]; + tensor var_3279_begin_0 = const()[name = string("op_3279_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3279_end_0 = const()[name = string("op_3279_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3279_end_mask_0 = const()[name = string("op_3279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3279_cast_fp16 = slice_by_index(begin = var_3279_begin_0, end = var_3279_end_0, end_mask = var_3279_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3279_cast_fp16")]; + tensor var_3286_begin_0 = const()[name = string("op_3286_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3286_end_0 = const()[name = string("op_3286_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3286_end_mask_0 = const()[name = string("op_3286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3286_cast_fp16 = slice_by_index(begin = var_3286_begin_0, end = var_3286_end_0, end_mask = var_3286_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3286_cast_fp16")]; + tensor var_3293_begin_0 = const()[name = string("op_3293_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3293_end_0 = const()[name = string("op_3293_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3293_end_mask_0 = const()[name = string("op_3293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3293_cast_fp16 = slice_by_index(begin = var_3293_begin_0, end = var_3293_end_0, end_mask = var_3293_end_mask_0, x = var_3123_cast_fp16)[name = string("op_3293_cast_fp16")]; + tensor var_3300_begin_0 = const()[name = string("op_3300_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3300_end_0 = const()[name = string("op_3300_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3300_end_mask_0 = const()[name = string("op_3300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3300_cast_fp16 = slice_by_index(begin = var_3300_begin_0, end = var_3300_end_0, end_mask = var_3300_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3300_cast_fp16")]; + tensor var_3307_begin_0 = const()[name = string("op_3307_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3307_end_0 = const()[name = string("op_3307_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3307_end_mask_0 = const()[name = string("op_3307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3307_cast_fp16 = slice_by_index(begin = var_3307_begin_0, end = var_3307_end_0, end_mask = var_3307_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3307_cast_fp16")]; + tensor var_3314_begin_0 = const()[name = string("op_3314_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3314_end_0 = const()[name = string("op_3314_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3314_end_mask_0 = const()[name = string("op_3314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3314_cast_fp16 = slice_by_index(begin = var_3314_begin_0, end = var_3314_end_0, end_mask = var_3314_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3314_cast_fp16")]; + tensor var_3321_begin_0 = const()[name = string("op_3321_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3321_end_0 = const()[name = string("op_3321_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3321_end_mask_0 = const()[name = string("op_3321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3321_cast_fp16 = slice_by_index(begin = var_3321_begin_0, end = var_3321_end_0, end_mask = var_3321_end_mask_0, x = var_3127_cast_fp16)[name = string("op_3321_cast_fp16")]; + tensor var_3328_begin_0 = const()[name = string("op_3328_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3328_end_0 = const()[name = string("op_3328_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3328_end_mask_0 = const()[name = string("op_3328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3328_cast_fp16 = slice_by_index(begin = var_3328_begin_0, end = var_3328_end_0, end_mask = var_3328_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3328_cast_fp16")]; + tensor var_3335_begin_0 = const()[name = string("op_3335_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3335_end_0 = const()[name = string("op_3335_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3335_end_mask_0 = const()[name = string("op_3335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3335_cast_fp16 = slice_by_index(begin = var_3335_begin_0, end = var_3335_end_0, end_mask = var_3335_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3335_cast_fp16")]; + tensor var_3342_begin_0 = const()[name = string("op_3342_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3342_end_0 = const()[name = string("op_3342_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3342_end_mask_0 = const()[name = string("op_3342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3342_cast_fp16 = slice_by_index(begin = var_3342_begin_0, end = var_3342_end_0, end_mask = var_3342_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3342_cast_fp16")]; + tensor var_3349_begin_0 = const()[name = string("op_3349_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3349_end_0 = const()[name = string("op_3349_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3349_end_mask_0 = const()[name = string("op_3349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3349_cast_fp16 = slice_by_index(begin = var_3349_begin_0, end = var_3349_end_0, end_mask = var_3349_end_mask_0, x = var_3131_cast_fp16)[name = string("op_3349_cast_fp16")]; + tensor var_3356_begin_0 = const()[name = string("op_3356_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3356_end_0 = const()[name = string("op_3356_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3356_end_mask_0 = const()[name = string("op_3356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3356_cast_fp16 = slice_by_index(begin = var_3356_begin_0, end = var_3356_end_0, end_mask = var_3356_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3363_begin_0 = const()[name = string("op_3363_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3363_end_0 = const()[name = string("op_3363_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3363_end_mask_0 = const()[name = string("op_3363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3363_cast_fp16 = slice_by_index(begin = var_3363_begin_0, end = var_3363_end_0, end_mask = var_3363_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3363_cast_fp16")]; + tensor var_3370_begin_0 = const()[name = string("op_3370_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3370_end_0 = const()[name = string("op_3370_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3370_end_mask_0 = const()[name = string("op_3370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3370_cast_fp16 = slice_by_index(begin = var_3370_begin_0, end = var_3370_end_0, end_mask = var_3370_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3370_cast_fp16")]; + tensor var_3377_begin_0 = const()[name = string("op_3377_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3377_end_0 = const()[name = string("op_3377_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3377_end_mask_0 = const()[name = string("op_3377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = var_3135_cast_fp16)[name = string("op_3377_cast_fp16")]; + tensor var_3384_begin_0 = const()[name = string("op_3384_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3384_end_0 = const()[name = string("op_3384_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3384_end_mask_0 = const()[name = string("op_3384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3384_cast_fp16 = slice_by_index(begin = var_3384_begin_0, end = var_3384_end_0, end_mask = var_3384_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3384_cast_fp16")]; + tensor var_3391_begin_0 = const()[name = string("op_3391_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3391_end_0 = const()[name = string("op_3391_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3391_end_mask_0 = const()[name = string("op_3391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3391_cast_fp16 = slice_by_index(begin = var_3391_begin_0, end = var_3391_end_0, end_mask = var_3391_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3391_cast_fp16")]; + tensor var_3398_begin_0 = const()[name = string("op_3398_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3398_end_0 = const()[name = string("op_3398_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3398_end_mask_0 = const()[name = string("op_3398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3398_cast_fp16 = slice_by_index(begin = var_3398_begin_0, end = var_3398_end_0, end_mask = var_3398_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3398_cast_fp16")]; + tensor var_3405_begin_0 = const()[name = string("op_3405_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3405_end_0 = const()[name = string("op_3405_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3405_end_mask_0 = const()[name = string("op_3405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3405_cast_fp16 = slice_by_index(begin = var_3405_begin_0, end = var_3405_end_0, end_mask = var_3405_end_mask_0, x = var_3139_cast_fp16)[name = string("op_3405_cast_fp16")]; + tensor var_3412_begin_0 = const()[name = string("op_3412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3412_end_0 = const()[name = string("op_3412_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3412_end_mask_0 = const()[name = string("op_3412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3412_cast_fp16 = slice_by_index(begin = var_3412_begin_0, end = var_3412_end_0, end_mask = var_3412_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3412_cast_fp16")]; + tensor var_3419_begin_0 = const()[name = string("op_3419_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3419_end_0 = const()[name = string("op_3419_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3419_end_mask_0 = const()[name = string("op_3419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3419_cast_fp16 = slice_by_index(begin = var_3419_begin_0, end = var_3419_end_0, end_mask = var_3419_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3419_cast_fp16")]; + tensor var_3426_begin_0 = const()[name = string("op_3426_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3426_end_0 = const()[name = string("op_3426_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3426_end_mask_0 = const()[name = string("op_3426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3426_cast_fp16 = slice_by_index(begin = var_3426_begin_0, end = var_3426_end_0, end_mask = var_3426_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3426_cast_fp16")]; + tensor var_3433_begin_0 = const()[name = string("op_3433_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3433_end_0 = const()[name = string("op_3433_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3433_end_mask_0 = const()[name = string("op_3433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3433_cast_fp16 = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = var_3143_cast_fp16)[name = string("op_3433_cast_fp16")]; + tensor var_3440_begin_0 = const()[name = string("op_3440_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3440_end_0 = const()[name = string("op_3440_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3440_end_mask_0 = const()[name = string("op_3440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3440_cast_fp16 = slice_by_index(begin = var_3440_begin_0, end = var_3440_end_0, end_mask = var_3440_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3440_cast_fp16")]; + tensor var_3447_begin_0 = const()[name = string("op_3447_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3447_end_0 = const()[name = string("op_3447_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3447_end_mask_0 = const()[name = string("op_3447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3447_cast_fp16 = slice_by_index(begin = var_3447_begin_0, end = var_3447_end_0, end_mask = var_3447_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3447_cast_fp16")]; + tensor var_3454_begin_0 = const()[name = string("op_3454_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3454_end_0 = const()[name = string("op_3454_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3454_end_mask_0 = const()[name = string("op_3454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3454_cast_fp16 = slice_by_index(begin = var_3454_begin_0, end = var_3454_end_0, end_mask = var_3454_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3454_cast_fp16")]; + tensor var_3461_begin_0 = const()[name = string("op_3461_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3461_end_0 = const()[name = string("op_3461_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3461_end_mask_0 = const()[name = string("op_3461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3461_cast_fp16 = slice_by_index(begin = var_3461_begin_0, end = var_3461_end_0, end_mask = var_3461_end_mask_0, x = var_3147_cast_fp16)[name = string("op_3461_cast_fp16")]; + tensor var_3468_begin_0 = const()[name = string("op_3468_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3468_end_0 = const()[name = string("op_3468_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3468_end_mask_0 = const()[name = string("op_3468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3468_cast_fp16 = slice_by_index(begin = var_3468_begin_0, end = var_3468_end_0, end_mask = var_3468_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3468_cast_fp16")]; + tensor var_3475_begin_0 = const()[name = string("op_3475_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3475_end_0 = const()[name = string("op_3475_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3475_end_mask_0 = const()[name = string("op_3475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3475_cast_fp16 = slice_by_index(begin = var_3475_begin_0, end = var_3475_end_0, end_mask = var_3475_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3475_cast_fp16")]; + tensor var_3482_begin_0 = const()[name = string("op_3482_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3482_end_0 = const()[name = string("op_3482_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3482_end_mask_0 = const()[name = string("op_3482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3482_cast_fp16 = slice_by_index(begin = var_3482_begin_0, end = var_3482_end_0, end_mask = var_3482_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3482_cast_fp16")]; + tensor var_3489_begin_0 = const()[name = string("op_3489_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3489_end_0 = const()[name = string("op_3489_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3489_end_mask_0 = const()[name = string("op_3489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3489_cast_fp16 = slice_by_index(begin = var_3489_begin_0, end = var_3489_end_0, end_mask = var_3489_end_mask_0, x = var_3151_cast_fp16)[name = string("op_3489_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3494_begin_0 = const()[name = string("op_3494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3494_end_0 = const()[name = string("op_3494_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3494_end_mask_0 = const()[name = string("op_3494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_8")]; + tensor var_3494_cast_fp16 = slice_by_index(begin = var_3494_begin_0, end = var_3494_end_0, end_mask = var_3494_end_mask_0, x = k_7_cast_fp16)[name = string("op_3494_cast_fp16")]; + tensor var_3498_begin_0 = const()[name = string("op_3498_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3498_end_0 = const()[name = string("op_3498_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3498_end_mask_0 = const()[name = string("op_3498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3498_cast_fp16 = slice_by_index(begin = var_3498_begin_0, end = var_3498_end_0, end_mask = var_3498_end_mask_0, x = k_7_cast_fp16)[name = string("op_3498_cast_fp16")]; + tensor var_3502_begin_0 = const()[name = string("op_3502_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3502_end_0 = const()[name = string("op_3502_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3502_end_mask_0 = const()[name = string("op_3502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3502_cast_fp16 = slice_by_index(begin = var_3502_begin_0, end = var_3502_end_0, end_mask = var_3502_end_mask_0, x = k_7_cast_fp16)[name = string("op_3502_cast_fp16")]; + tensor var_3506_begin_0 = const()[name = string("op_3506_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3506_end_0 = const()[name = string("op_3506_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3506_end_mask_0 = const()[name = string("op_3506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3506_cast_fp16 = slice_by_index(begin = var_3506_begin_0, end = var_3506_end_0, end_mask = var_3506_end_mask_0, x = k_7_cast_fp16)[name = string("op_3506_cast_fp16")]; + tensor var_3510_begin_0 = const()[name = string("op_3510_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3510_end_0 = const()[name = string("op_3510_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3510_end_mask_0 = const()[name = string("op_3510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3510_cast_fp16 = slice_by_index(begin = var_3510_begin_0, end = var_3510_end_0, end_mask = var_3510_end_mask_0, x = k_7_cast_fp16)[name = string("op_3510_cast_fp16")]; + tensor var_3514_begin_0 = const()[name = string("op_3514_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3514_end_0 = const()[name = string("op_3514_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3514_end_mask_0 = const()[name = string("op_3514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3514_cast_fp16 = slice_by_index(begin = var_3514_begin_0, end = var_3514_end_0, end_mask = var_3514_end_mask_0, x = k_7_cast_fp16)[name = string("op_3514_cast_fp16")]; + tensor var_3518_begin_0 = const()[name = string("op_3518_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3518_end_0 = const()[name = string("op_3518_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3518_end_mask_0 = const()[name = string("op_3518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3518_cast_fp16 = slice_by_index(begin = var_3518_begin_0, end = var_3518_end_0, end_mask = var_3518_end_mask_0, x = k_7_cast_fp16)[name = string("op_3518_cast_fp16")]; + tensor var_3522_begin_0 = const()[name = string("op_3522_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3522_end_0 = const()[name = string("op_3522_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3522_end_mask_0 = const()[name = string("op_3522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3522_cast_fp16 = slice_by_index(begin = var_3522_begin_0, end = var_3522_end_0, end_mask = var_3522_end_mask_0, x = k_7_cast_fp16)[name = string("op_3522_cast_fp16")]; + tensor var_3526_begin_0 = const()[name = string("op_3526_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3526_end_0 = const()[name = string("op_3526_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_3526_end_mask_0 = const()[name = string("op_3526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3526_cast_fp16 = slice_by_index(begin = var_3526_begin_0, end = var_3526_end_0, end_mask = var_3526_end_mask_0, x = k_7_cast_fp16)[name = string("op_3526_cast_fp16")]; + tensor var_3530_begin_0 = const()[name = string("op_3530_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3530_end_0 = const()[name = string("op_3530_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_3530_end_mask_0 = const()[name = string("op_3530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3530_cast_fp16 = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = k_7_cast_fp16)[name = string("op_3530_cast_fp16")]; + tensor var_3534_begin_0 = const()[name = string("op_3534_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_3534_end_0 = const()[name = string("op_3534_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_3534_end_mask_0 = const()[name = string("op_3534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3534_cast_fp16 = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = k_7_cast_fp16)[name = string("op_3534_cast_fp16")]; + tensor var_3538_begin_0 = const()[name = string("op_3538_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_3538_end_0 = const()[name = string("op_3538_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_3538_end_mask_0 = const()[name = string("op_3538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3538_cast_fp16 = slice_by_index(begin = var_3538_begin_0, end = var_3538_end_0, end_mask = var_3538_end_mask_0, x = k_7_cast_fp16)[name = string("op_3538_cast_fp16")]; + tensor var_3540_begin_0 = const()[name = string("op_3540_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3540_end_0 = const()[name = string("op_3540_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3540_end_mask_0 = const()[name = string("op_3540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3540_cast_fp16 = slice_by_index(begin = var_3540_begin_0, end = var_3540_end_0, end_mask = var_3540_end_mask_0, x = value_7_cast_fp16)[name = string("op_3540_cast_fp16")]; + tensor var_3544_begin_0 = const()[name = string("op_3544_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3544_end_0 = const()[name = string("op_3544_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3544_end_mask_0 = const()[name = string("op_3544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3544_cast_fp16 = slice_by_index(begin = var_3544_begin_0, end = var_3544_end_0, end_mask = var_3544_end_mask_0, x = value_7_cast_fp16)[name = string("op_3544_cast_fp16")]; + tensor var_3548_begin_0 = const()[name = string("op_3548_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3548_end_0 = const()[name = string("op_3548_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3548_end_mask_0 = const()[name = string("op_3548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3548_cast_fp16 = slice_by_index(begin = var_3548_begin_0, end = var_3548_end_0, end_mask = var_3548_end_mask_0, x = value_7_cast_fp16)[name = string("op_3548_cast_fp16")]; + tensor var_3552_begin_0 = const()[name = string("op_3552_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3552_end_0 = const()[name = string("op_3552_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3552_end_mask_0 = const()[name = string("op_3552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3552_cast_fp16 = slice_by_index(begin = var_3552_begin_0, end = var_3552_end_0, end_mask = var_3552_end_mask_0, x = value_7_cast_fp16)[name = string("op_3552_cast_fp16")]; + tensor var_3556_begin_0 = const()[name = string("op_3556_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3556_end_0 = const()[name = string("op_3556_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3556_end_mask_0 = const()[name = string("op_3556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3556_cast_fp16 = slice_by_index(begin = var_3556_begin_0, end = var_3556_end_0, end_mask = var_3556_end_mask_0, x = value_7_cast_fp16)[name = string("op_3556_cast_fp16")]; + tensor var_3560_begin_0 = const()[name = string("op_3560_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3560_end_0 = const()[name = string("op_3560_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3560_end_mask_0 = const()[name = string("op_3560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3560_cast_fp16 = slice_by_index(begin = var_3560_begin_0, end = var_3560_end_0, end_mask = var_3560_end_mask_0, x = value_7_cast_fp16)[name = string("op_3560_cast_fp16")]; + tensor var_3564_begin_0 = const()[name = string("op_3564_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3564_end_0 = const()[name = string("op_3564_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3564_end_mask_0 = const()[name = string("op_3564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3564_cast_fp16 = slice_by_index(begin = var_3564_begin_0, end = var_3564_end_0, end_mask = var_3564_end_mask_0, x = value_7_cast_fp16)[name = string("op_3564_cast_fp16")]; + tensor var_3568_begin_0 = const()[name = string("op_3568_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3568_end_0 = const()[name = string("op_3568_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3568_end_mask_0 = const()[name = string("op_3568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3568_cast_fp16 = slice_by_index(begin = var_3568_begin_0, end = var_3568_end_0, end_mask = var_3568_end_mask_0, x = value_7_cast_fp16)[name = string("op_3568_cast_fp16")]; + tensor var_3572_begin_0 = const()[name = string("op_3572_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3572_end_0 = const()[name = string("op_3572_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3572_end_mask_0 = const()[name = string("op_3572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3572_cast_fp16 = slice_by_index(begin = var_3572_begin_0, end = var_3572_end_0, end_mask = var_3572_end_mask_0, x = value_7_cast_fp16)[name = string("op_3572_cast_fp16")]; + tensor var_3576_begin_0 = const()[name = string("op_3576_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3576_end_0 = const()[name = string("op_3576_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3576_end_mask_0 = const()[name = string("op_3576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3576_cast_fp16 = slice_by_index(begin = var_3576_begin_0, end = var_3576_end_0, end_mask = var_3576_end_mask_0, x = value_7_cast_fp16)[name = string("op_3576_cast_fp16")]; + tensor var_3580_begin_0 = const()[name = string("op_3580_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3580_end_0 = const()[name = string("op_3580_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3580_end_mask_0 = const()[name = string("op_3580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3580_cast_fp16 = slice_by_index(begin = var_3580_begin_0, end = var_3580_end_0, end_mask = var_3580_end_mask_0, x = value_7_cast_fp16)[name = string("op_3580_cast_fp16")]; + tensor var_3584_begin_0 = const()[name = string("op_3584_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3584_end_0 = const()[name = string("op_3584_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3584_end_mask_0 = const()[name = string("op_3584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3584_cast_fp16 = slice_by_index(begin = var_3584_begin_0, end = var_3584_end_0, end_mask = var_3584_end_mask_0, x = value_7_cast_fp16)[name = string("op_3584_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_3494_cast_fp16, var_3160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_3494_cast_fp16, var_3167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_3494_cast_fp16, var_3174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_3494_cast_fp16, var_3181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_3498_cast_fp16, var_3188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_3498_cast_fp16, var_3195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_3498_cast_fp16, var_3202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_3498_cast_fp16, var_3209_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_3502_cast_fp16, var_3216_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_3502_cast_fp16, var_3223_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_3502_cast_fp16, var_3230_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_3502_cast_fp16, var_3237_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_3506_cast_fp16, var_3244_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_3506_cast_fp16, var_3251_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_3506_cast_fp16, var_3258_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_3506_cast_fp16, var_3265_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3510_cast_fp16, var_3272_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3510_cast_fp16, var_3279_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3510_cast_fp16, var_3286_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3510_cast_fp16, var_3293_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3514_cast_fp16, var_3300_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3514_cast_fp16, var_3307_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3514_cast_fp16, var_3314_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3514_cast_fp16, var_3321_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3518_cast_fp16, var_3328_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3518_cast_fp16, var_3335_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3518_cast_fp16, var_3342_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3518_cast_fp16, var_3349_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3522_cast_fp16, var_3356_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3522_cast_fp16, var_3363_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3522_cast_fp16, var_3370_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3522_cast_fp16, var_3377_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3526_cast_fp16, var_3384_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3526_cast_fp16, var_3391_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3526_cast_fp16, var_3398_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3526_cast_fp16, var_3405_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3530_cast_fp16, var_3412_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3530_cast_fp16, var_3419_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3530_cast_fp16, var_3426_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3530_cast_fp16, var_3433_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3534_cast_fp16, var_3440_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3534_cast_fp16, var_3447_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3534_cast_fp16, var_3454_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3534_cast_fp16, var_3461_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3538_cast_fp16, var_3468_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3538_cast_fp16, var_3475_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3538_cast_fp16, var_3482_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_383_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_383_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_383_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_383_equation_0, values = (var_3538_cast_fp16, var_3489_cast_fp16))[name = string("_SplitHeadsQ__mh_w_383_cast_fp16")]; + fp16 var_3683_to_fp16 = const()[name = string("op_3683_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_3683_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_3685_to_fp16 = const()[name = string("op_3685_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_3685_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_3687_to_fp16 = const()[name = string("op_3687_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_3687_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_3689_to_fp16 = const()[name = string("op_3689_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_3689_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_3691_to_fp16 = const()[name = string("op_3691_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_3691_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_3693_to_fp16 = const()[name = string("op_3693_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_3693_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_3695_to_fp16 = const()[name = string("op_3695_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_3695_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_3697_to_fp16 = const()[name = string("op_3697_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_3697_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_3699_to_fp16 = const()[name = string("op_3699_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_3699_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_3701_to_fp16 = const()[name = string("op_3701_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_3701_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_3703_to_fp16 = const()[name = string("op_3703_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_3703_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_3705_to_fp16 = const()[name = string("op_3705_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_3705_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_3707_to_fp16 = const()[name = string("op_3707_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_3707_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_3709_to_fp16 = const()[name = string("op_3709_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_3709_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_3711_to_fp16 = const()[name = string("op_3711_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_3711_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_3713_to_fp16 = const()[name = string("op_3713_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_3713_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + fp16 var_3715_to_fp16 = const()[name = string("op_3715_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_3715_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_3717_to_fp16 = const()[name = string("op_3717_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_3717_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_3719_to_fp16 = const()[name = string("op_3719_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_3719_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_3721_to_fp16 = const()[name = string("op_3721_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_3721_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_3723_to_fp16 = const()[name = string("op_3723_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_3723_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_3725_to_fp16 = const()[name = string("op_3725_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_3725_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_3727_to_fp16 = const()[name = string("op_3727_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_3727_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_3729_to_fp16 = const()[name = string("op_3729_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_3729_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_3731_to_fp16 = const()[name = string("op_3731_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_3731_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_3733_to_fp16 = const()[name = string("op_3733_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_3733_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_3735_to_fp16 = const()[name = string("op_3735_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_3735_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_3737_to_fp16 = const()[name = string("op_3737_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_3737_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_3739_to_fp16 = const()[name = string("op_3739_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_3739_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_3741_to_fp16 = const()[name = string("op_3741_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_3741_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_3743_to_fp16 = const()[name = string("op_3743_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_3743_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_3745_to_fp16 = const()[name = string("op_3745_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_3745_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_3747_to_fp16 = const()[name = string("op_3747_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_3747_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_3749_to_fp16 = const()[name = string("op_3749_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_3749_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_3751_to_fp16 = const()[name = string("op_3751_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_3751_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_3753_to_fp16 = const()[name = string("op_3753_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_3753_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_3755_to_fp16 = const()[name = string("op_3755_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_3755_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_3757_to_fp16 = const()[name = string("op_3757_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_3757_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_3759_to_fp16 = const()[name = string("op_3759_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_3759_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_3761_to_fp16 = const()[name = string("op_3761_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_3761_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_3763_to_fp16 = const()[name = string("op_3763_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_3763_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_3765_to_fp16 = const()[name = string("op_3765_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_3765_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_3767_to_fp16 = const()[name = string("op_3767_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_3767_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_3769_to_fp16 = const()[name = string("op_3769_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_3769_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_3771_to_fp16 = const()[name = string("op_3771_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_3771_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_3773_to_fp16 = const()[name = string("op_3773_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_3773_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_3775_to_fp16 = const()[name = string("op_3775_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_3775_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_3777_to_fp16 = const()[name = string("op_3777_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_383_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_383_cast_fp16, y = var_3777_to_fp16)[name = string("aw_chunk_383_cast_fp16")]; + tensor var_3779_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_289_cast_fp16)[name = string("op_3779_cast_fp16")]; + tensor var_3780_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_291_cast_fp16)[name = string("op_3780_cast_fp16")]; + tensor var_3781_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_293_cast_fp16)[name = string("op_3781_cast_fp16")]; + tensor var_3782_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_295_cast_fp16)[name = string("op_3782_cast_fp16")]; + tensor var_3783_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_297_cast_fp16)[name = string("op_3783_cast_fp16")]; + tensor var_3784_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_299_cast_fp16)[name = string("op_3784_cast_fp16")]; + tensor var_3785_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_301_cast_fp16)[name = string("op_3785_cast_fp16")]; + tensor var_3786_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_303_cast_fp16)[name = string("op_3786_cast_fp16")]; + tensor var_3787_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_305_cast_fp16)[name = string("op_3787_cast_fp16")]; + tensor var_3788_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_307_cast_fp16)[name = string("op_3788_cast_fp16")]; + tensor var_3789_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_309_cast_fp16)[name = string("op_3789_cast_fp16")]; + tensor var_3790_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_311_cast_fp16)[name = string("op_3790_cast_fp16")]; + tensor var_3791_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_313_cast_fp16)[name = string("op_3791_cast_fp16")]; + tensor var_3792_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_315_cast_fp16)[name = string("op_3792_cast_fp16")]; + tensor var_3793_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_317_cast_fp16)[name = string("op_3793_cast_fp16")]; + tensor var_3794_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_319_cast_fp16)[name = string("op_3794_cast_fp16")]; + tensor var_3795_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_321_cast_fp16)[name = string("op_3795_cast_fp16")]; + tensor var_3796_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_323_cast_fp16)[name = string("op_3796_cast_fp16")]; + tensor var_3797_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_325_cast_fp16)[name = string("op_3797_cast_fp16")]; + tensor var_3798_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_327_cast_fp16)[name = string("op_3798_cast_fp16")]; + tensor var_3799_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_329_cast_fp16)[name = string("op_3799_cast_fp16")]; + tensor var_3800_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_331_cast_fp16)[name = string("op_3800_cast_fp16")]; + tensor var_3801_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_333_cast_fp16)[name = string("op_3801_cast_fp16")]; + tensor var_3802_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_335_cast_fp16)[name = string("op_3802_cast_fp16")]; + tensor var_3803_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_337_cast_fp16)[name = string("op_3803_cast_fp16")]; + tensor var_3804_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_339_cast_fp16)[name = string("op_3804_cast_fp16")]; + tensor var_3805_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_341_cast_fp16)[name = string("op_3805_cast_fp16")]; + tensor var_3806_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_343_cast_fp16)[name = string("op_3806_cast_fp16")]; + tensor var_3807_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_345_cast_fp16)[name = string("op_3807_cast_fp16")]; + tensor var_3808_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_347_cast_fp16)[name = string("op_3808_cast_fp16")]; + tensor var_3809_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_349_cast_fp16)[name = string("op_3809_cast_fp16")]; + tensor var_3810_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_351_cast_fp16)[name = string("op_3810_cast_fp16")]; + tensor var_3811_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_353_cast_fp16)[name = string("op_3811_cast_fp16")]; + tensor var_3812_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_355_cast_fp16)[name = string("op_3812_cast_fp16")]; + tensor var_3813_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_357_cast_fp16)[name = string("op_3813_cast_fp16")]; + tensor var_3814_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_359_cast_fp16)[name = string("op_3814_cast_fp16")]; + tensor var_3815_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_361_cast_fp16)[name = string("op_3815_cast_fp16")]; + tensor var_3816_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_363_cast_fp16)[name = string("op_3816_cast_fp16")]; + tensor var_3817_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_365_cast_fp16)[name = string("op_3817_cast_fp16")]; + tensor var_3818_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_367_cast_fp16)[name = string("op_3818_cast_fp16")]; + tensor var_3819_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_369_cast_fp16)[name = string("op_3819_cast_fp16")]; + tensor var_3820_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_371_cast_fp16)[name = string("op_3820_cast_fp16")]; + tensor var_3821_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_373_cast_fp16)[name = string("op_3821_cast_fp16")]; + tensor var_3822_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_375_cast_fp16)[name = string("op_3822_cast_fp16")]; + tensor var_3823_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_377_cast_fp16)[name = string("op_3823_cast_fp16")]; + tensor var_3824_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_379_cast_fp16)[name = string("op_3824_cast_fp16")]; + tensor var_3825_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_381_cast_fp16)[name = string("op_3825_cast_fp16")]; + tensor var_3826_cast_fp16 = softmax(axis = var_3052, x = aw_chunk_383_cast_fp16)[name = string("op_3826_cast_fp16")]; + string var_3828_equation_0 = const()[name = string("op_3828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3828_cast_fp16 = einsum(equation = var_3828_equation_0, values = (var_3540_cast_fp16, var_3779_cast_fp16))[name = string("op_3828_cast_fp16")]; + string var_3830_equation_0 = const()[name = string("op_3830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3830_cast_fp16 = einsum(equation = var_3830_equation_0, values = (var_3540_cast_fp16, var_3780_cast_fp16))[name = string("op_3830_cast_fp16")]; + string var_3832_equation_0 = const()[name = string("op_3832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3832_cast_fp16 = einsum(equation = var_3832_equation_0, values = (var_3540_cast_fp16, var_3781_cast_fp16))[name = string("op_3832_cast_fp16")]; + string var_3834_equation_0 = const()[name = string("op_3834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3834_cast_fp16 = einsum(equation = var_3834_equation_0, values = (var_3540_cast_fp16, var_3782_cast_fp16))[name = string("op_3834_cast_fp16")]; + string var_3836_equation_0 = const()[name = string("op_3836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3836_cast_fp16 = einsum(equation = var_3836_equation_0, values = (var_3544_cast_fp16, var_3783_cast_fp16))[name = string("op_3836_cast_fp16")]; + string var_3838_equation_0 = const()[name = string("op_3838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3838_cast_fp16 = einsum(equation = var_3838_equation_0, values = (var_3544_cast_fp16, var_3784_cast_fp16))[name = string("op_3838_cast_fp16")]; + string var_3840_equation_0 = const()[name = string("op_3840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3840_cast_fp16 = einsum(equation = var_3840_equation_0, values = (var_3544_cast_fp16, var_3785_cast_fp16))[name = string("op_3840_cast_fp16")]; + string var_3842_equation_0 = const()[name = string("op_3842_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3842_cast_fp16 = einsum(equation = var_3842_equation_0, values = (var_3544_cast_fp16, var_3786_cast_fp16))[name = string("op_3842_cast_fp16")]; + string var_3844_equation_0 = const()[name = string("op_3844_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3844_cast_fp16 = einsum(equation = var_3844_equation_0, values = (var_3548_cast_fp16, var_3787_cast_fp16))[name = string("op_3844_cast_fp16")]; + string var_3846_equation_0 = const()[name = string("op_3846_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3846_cast_fp16 = einsum(equation = var_3846_equation_0, values = (var_3548_cast_fp16, var_3788_cast_fp16))[name = string("op_3846_cast_fp16")]; + string var_3848_equation_0 = const()[name = string("op_3848_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3848_cast_fp16 = einsum(equation = var_3848_equation_0, values = (var_3548_cast_fp16, var_3789_cast_fp16))[name = string("op_3848_cast_fp16")]; + string var_3850_equation_0 = const()[name = string("op_3850_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3850_cast_fp16 = einsum(equation = var_3850_equation_0, values = (var_3548_cast_fp16, var_3790_cast_fp16))[name = string("op_3850_cast_fp16")]; + string var_3852_equation_0 = const()[name = string("op_3852_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3852_cast_fp16 = einsum(equation = var_3852_equation_0, values = (var_3552_cast_fp16, var_3791_cast_fp16))[name = string("op_3852_cast_fp16")]; + string var_3854_equation_0 = const()[name = string("op_3854_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3854_cast_fp16 = einsum(equation = var_3854_equation_0, values = (var_3552_cast_fp16, var_3792_cast_fp16))[name = string("op_3854_cast_fp16")]; + string var_3856_equation_0 = const()[name = string("op_3856_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3856_cast_fp16 = einsum(equation = var_3856_equation_0, values = (var_3552_cast_fp16, var_3793_cast_fp16))[name = string("op_3856_cast_fp16")]; + string var_3858_equation_0 = const()[name = string("op_3858_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3858_cast_fp16 = einsum(equation = var_3858_equation_0, values = (var_3552_cast_fp16, var_3794_cast_fp16))[name = string("op_3858_cast_fp16")]; + string var_3860_equation_0 = const()[name = string("op_3860_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3860_cast_fp16 = einsum(equation = var_3860_equation_0, values = (var_3556_cast_fp16, var_3795_cast_fp16))[name = string("op_3860_cast_fp16")]; + string var_3862_equation_0 = const()[name = string("op_3862_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3862_cast_fp16 = einsum(equation = var_3862_equation_0, values = (var_3556_cast_fp16, var_3796_cast_fp16))[name = string("op_3862_cast_fp16")]; + string var_3864_equation_0 = const()[name = string("op_3864_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3864_cast_fp16 = einsum(equation = var_3864_equation_0, values = (var_3556_cast_fp16, var_3797_cast_fp16))[name = string("op_3864_cast_fp16")]; + string var_3866_equation_0 = const()[name = string("op_3866_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3866_cast_fp16 = einsum(equation = var_3866_equation_0, values = (var_3556_cast_fp16, var_3798_cast_fp16))[name = string("op_3866_cast_fp16")]; + string var_3868_equation_0 = const()[name = string("op_3868_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3868_cast_fp16 = einsum(equation = var_3868_equation_0, values = (var_3560_cast_fp16, var_3799_cast_fp16))[name = string("op_3868_cast_fp16")]; + string var_3870_equation_0 = const()[name = string("op_3870_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3870_cast_fp16 = einsum(equation = var_3870_equation_0, values = (var_3560_cast_fp16, var_3800_cast_fp16))[name = string("op_3870_cast_fp16")]; + string var_3872_equation_0 = const()[name = string("op_3872_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3872_cast_fp16 = einsum(equation = var_3872_equation_0, values = (var_3560_cast_fp16, var_3801_cast_fp16))[name = string("op_3872_cast_fp16")]; + string var_3874_equation_0 = const()[name = string("op_3874_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3874_cast_fp16 = einsum(equation = var_3874_equation_0, values = (var_3560_cast_fp16, var_3802_cast_fp16))[name = string("op_3874_cast_fp16")]; + string var_3876_equation_0 = const()[name = string("op_3876_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3876_cast_fp16 = einsum(equation = var_3876_equation_0, values = (var_3564_cast_fp16, var_3803_cast_fp16))[name = string("op_3876_cast_fp16")]; + string var_3878_equation_0 = const()[name = string("op_3878_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3878_cast_fp16 = einsum(equation = var_3878_equation_0, values = (var_3564_cast_fp16, var_3804_cast_fp16))[name = string("op_3878_cast_fp16")]; + string var_3880_equation_0 = const()[name = string("op_3880_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3880_cast_fp16 = einsum(equation = var_3880_equation_0, values = (var_3564_cast_fp16, var_3805_cast_fp16))[name = string("op_3880_cast_fp16")]; + string var_3882_equation_0 = const()[name = string("op_3882_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3882_cast_fp16 = einsum(equation = var_3882_equation_0, values = (var_3564_cast_fp16, var_3806_cast_fp16))[name = string("op_3882_cast_fp16")]; + string var_3884_equation_0 = const()[name = string("op_3884_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3884_cast_fp16 = einsum(equation = var_3884_equation_0, values = (var_3568_cast_fp16, var_3807_cast_fp16))[name = string("op_3884_cast_fp16")]; + string var_3886_equation_0 = const()[name = string("op_3886_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3886_cast_fp16 = einsum(equation = var_3886_equation_0, values = (var_3568_cast_fp16, var_3808_cast_fp16))[name = string("op_3886_cast_fp16")]; + string var_3888_equation_0 = const()[name = string("op_3888_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3888_cast_fp16 = einsum(equation = var_3888_equation_0, values = (var_3568_cast_fp16, var_3809_cast_fp16))[name = string("op_3888_cast_fp16")]; + string var_3890_equation_0 = const()[name = string("op_3890_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3890_cast_fp16 = einsum(equation = var_3890_equation_0, values = (var_3568_cast_fp16, var_3810_cast_fp16))[name = string("op_3890_cast_fp16")]; + string var_3892_equation_0 = const()[name = string("op_3892_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3892_cast_fp16 = einsum(equation = var_3892_equation_0, values = (var_3572_cast_fp16, var_3811_cast_fp16))[name = string("op_3892_cast_fp16")]; + string var_3894_equation_0 = const()[name = string("op_3894_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3894_cast_fp16 = einsum(equation = var_3894_equation_0, values = (var_3572_cast_fp16, var_3812_cast_fp16))[name = string("op_3894_cast_fp16")]; + string var_3896_equation_0 = const()[name = string("op_3896_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3896_cast_fp16 = einsum(equation = var_3896_equation_0, values = (var_3572_cast_fp16, var_3813_cast_fp16))[name = string("op_3896_cast_fp16")]; + string var_3898_equation_0 = const()[name = string("op_3898_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3898_cast_fp16 = einsum(equation = var_3898_equation_0, values = (var_3572_cast_fp16, var_3814_cast_fp16))[name = string("op_3898_cast_fp16")]; + string var_3900_equation_0 = const()[name = string("op_3900_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3900_cast_fp16 = einsum(equation = var_3900_equation_0, values = (var_3576_cast_fp16, var_3815_cast_fp16))[name = string("op_3900_cast_fp16")]; + string var_3902_equation_0 = const()[name = string("op_3902_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3902_cast_fp16 = einsum(equation = var_3902_equation_0, values = (var_3576_cast_fp16, var_3816_cast_fp16))[name = string("op_3902_cast_fp16")]; + string var_3904_equation_0 = const()[name = string("op_3904_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3904_cast_fp16 = einsum(equation = var_3904_equation_0, values = (var_3576_cast_fp16, var_3817_cast_fp16))[name = string("op_3904_cast_fp16")]; + string var_3906_equation_0 = const()[name = string("op_3906_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3906_cast_fp16 = einsum(equation = var_3906_equation_0, values = (var_3576_cast_fp16, var_3818_cast_fp16))[name = string("op_3906_cast_fp16")]; + string var_3908_equation_0 = const()[name = string("op_3908_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3908_cast_fp16 = einsum(equation = var_3908_equation_0, values = (var_3580_cast_fp16, var_3819_cast_fp16))[name = string("op_3908_cast_fp16")]; + string var_3910_equation_0 = const()[name = string("op_3910_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3910_cast_fp16 = einsum(equation = var_3910_equation_0, values = (var_3580_cast_fp16, var_3820_cast_fp16))[name = string("op_3910_cast_fp16")]; + string var_3912_equation_0 = const()[name = string("op_3912_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3912_cast_fp16 = einsum(equation = var_3912_equation_0, values = (var_3580_cast_fp16, var_3821_cast_fp16))[name = string("op_3912_cast_fp16")]; + string var_3914_equation_0 = const()[name = string("op_3914_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3914_cast_fp16 = einsum(equation = var_3914_equation_0, values = (var_3580_cast_fp16, var_3822_cast_fp16))[name = string("op_3914_cast_fp16")]; + string var_3916_equation_0 = const()[name = string("op_3916_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3916_cast_fp16 = einsum(equation = var_3916_equation_0, values = (var_3584_cast_fp16, var_3823_cast_fp16))[name = string("op_3916_cast_fp16")]; + string var_3918_equation_0 = const()[name = string("op_3918_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3918_cast_fp16 = einsum(equation = var_3918_equation_0, values = (var_3584_cast_fp16, var_3824_cast_fp16))[name = string("op_3918_cast_fp16")]; + string var_3920_equation_0 = const()[name = string("op_3920_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3920_cast_fp16 = einsum(equation = var_3920_equation_0, values = (var_3584_cast_fp16, var_3825_cast_fp16))[name = string("op_3920_cast_fp16")]; + string var_3922_equation_0 = const()[name = string("op_3922_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3922_cast_fp16 = einsum(equation = var_3922_equation_0, values = (var_3584_cast_fp16, var_3826_cast_fp16))[name = string("op_3922_cast_fp16")]; + bool var_3924_interleave_0 = const()[name = string("op_3924_interleave_0"), val = bool(false)]; + tensor var_3924_cast_fp16 = concat(axis = var_3035, interleave = var_3924_interleave_0, values = (var_3828_cast_fp16, var_3830_cast_fp16, var_3832_cast_fp16, var_3834_cast_fp16))[name = string("op_3924_cast_fp16")]; + bool var_3926_interleave_0 = const()[name = string("op_3926_interleave_0"), val = bool(false)]; + tensor var_3926_cast_fp16 = concat(axis = var_3035, interleave = var_3926_interleave_0, values = (var_3836_cast_fp16, var_3838_cast_fp16, var_3840_cast_fp16, var_3842_cast_fp16))[name = string("op_3926_cast_fp16")]; + bool var_3928_interleave_0 = const()[name = string("op_3928_interleave_0"), val = bool(false)]; + tensor var_3928_cast_fp16 = concat(axis = var_3035, interleave = var_3928_interleave_0, values = (var_3844_cast_fp16, var_3846_cast_fp16, var_3848_cast_fp16, var_3850_cast_fp16))[name = string("op_3928_cast_fp16")]; + bool var_3930_interleave_0 = const()[name = string("op_3930_interleave_0"), val = bool(false)]; + tensor var_3930_cast_fp16 = concat(axis = var_3035, interleave = var_3930_interleave_0, values = (var_3852_cast_fp16, var_3854_cast_fp16, var_3856_cast_fp16, var_3858_cast_fp16))[name = string("op_3930_cast_fp16")]; + bool var_3932_interleave_0 = const()[name = string("op_3932_interleave_0"), val = bool(false)]; + tensor var_3932_cast_fp16 = concat(axis = var_3035, interleave = var_3932_interleave_0, values = (var_3860_cast_fp16, var_3862_cast_fp16, var_3864_cast_fp16, var_3866_cast_fp16))[name = string("op_3932_cast_fp16")]; + bool var_3934_interleave_0 = const()[name = string("op_3934_interleave_0"), val = bool(false)]; + tensor var_3934_cast_fp16 = concat(axis = var_3035, interleave = var_3934_interleave_0, values = (var_3868_cast_fp16, var_3870_cast_fp16, var_3872_cast_fp16, var_3874_cast_fp16))[name = string("op_3934_cast_fp16")]; + bool var_3936_interleave_0 = const()[name = string("op_3936_interleave_0"), val = bool(false)]; + tensor var_3936_cast_fp16 = concat(axis = var_3035, interleave = var_3936_interleave_0, values = (var_3876_cast_fp16, var_3878_cast_fp16, var_3880_cast_fp16, var_3882_cast_fp16))[name = string("op_3936_cast_fp16")]; + bool var_3938_interleave_0 = const()[name = string("op_3938_interleave_0"), val = bool(false)]; + tensor var_3938_cast_fp16 = concat(axis = var_3035, interleave = var_3938_interleave_0, values = (var_3884_cast_fp16, var_3886_cast_fp16, var_3888_cast_fp16, var_3890_cast_fp16))[name = string("op_3938_cast_fp16")]; + bool var_3940_interleave_0 = const()[name = string("op_3940_interleave_0"), val = bool(false)]; + tensor var_3940_cast_fp16 = concat(axis = var_3035, interleave = var_3940_interleave_0, values = (var_3892_cast_fp16, var_3894_cast_fp16, var_3896_cast_fp16, var_3898_cast_fp16))[name = string("op_3940_cast_fp16")]; + bool var_3942_interleave_0 = const()[name = string("op_3942_interleave_0"), val = bool(false)]; + tensor var_3942_cast_fp16 = concat(axis = var_3035, interleave = var_3942_interleave_0, values = (var_3900_cast_fp16, var_3902_cast_fp16, var_3904_cast_fp16, var_3906_cast_fp16))[name = string("op_3942_cast_fp16")]; + bool var_3944_interleave_0 = const()[name = string("op_3944_interleave_0"), val = bool(false)]; + tensor var_3944_cast_fp16 = concat(axis = var_3035, interleave = var_3944_interleave_0, values = (var_3908_cast_fp16, var_3910_cast_fp16, var_3912_cast_fp16, var_3914_cast_fp16))[name = string("op_3944_cast_fp16")]; + bool var_3946_interleave_0 = const()[name = string("op_3946_interleave_0"), val = bool(false)]; + tensor var_3946_cast_fp16 = concat(axis = var_3035, interleave = var_3946_interleave_0, values = (var_3916_cast_fp16, var_3918_cast_fp16, var_3920_cast_fp16, var_3922_cast_fp16))[name = string("op_3946_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_3052, interleave = input_25_interleave_0, values = (var_3924_cast_fp16, var_3926_cast_fp16, var_3928_cast_fp16, var_3930_cast_fp16, var_3932_cast_fp16, var_3934_cast_fp16, var_3936_cast_fp16, var_3938_cast_fp16, var_3940_cast_fp16, var_3942_cast_fp16, var_3944_cast_fp16, var_3946_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52289280)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53468992)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_3965_to_fp16 = const()[name = string("op_3965_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_3965_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53470592)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53472192)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53473792)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58192448)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58198656)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62917312)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_3994 = const()[name = string("op_3994"), val = int32(3)]; + int32 var_4011 = const()[name = string("op_4011"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_4028_to_fp16 = const()[name = string("op_4028_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_4028_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62918912)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62920512)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62922112)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64101824)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64103424)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65283136)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66462848)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_4066_begin_0 = const()[name = string("op_4066_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4066_end_0 = const()[name = string("op_4066_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4066_end_mask_0 = const()[name = string("op_4066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4066_cast_fp16 = slice_by_index(begin = var_4066_begin_0, end = var_4066_end_0, end_mask = var_4066_end_mask_0, x = query_9_cast_fp16)[name = string("op_4066_cast_fp16")]; + tensor var_4070_begin_0 = const()[name = string("op_4070_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4070_end_0 = const()[name = string("op_4070_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4070_end_mask_0 = const()[name = string("op_4070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4070_cast_fp16 = slice_by_index(begin = var_4070_begin_0, end = var_4070_end_0, end_mask = var_4070_end_mask_0, x = query_9_cast_fp16)[name = string("op_4070_cast_fp16")]; + tensor var_4074_begin_0 = const()[name = string("op_4074_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4074_end_0 = const()[name = string("op_4074_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4074_end_mask_0 = const()[name = string("op_4074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4074_cast_fp16 = slice_by_index(begin = var_4074_begin_0, end = var_4074_end_0, end_mask = var_4074_end_mask_0, x = query_9_cast_fp16)[name = string("op_4074_cast_fp16")]; + tensor var_4078_begin_0 = const()[name = string("op_4078_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4078_end_0 = const()[name = string("op_4078_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4078_end_mask_0 = const()[name = string("op_4078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4078_cast_fp16 = slice_by_index(begin = var_4078_begin_0, end = var_4078_end_0, end_mask = var_4078_end_mask_0, x = query_9_cast_fp16)[name = string("op_4078_cast_fp16")]; + tensor var_4082_begin_0 = const()[name = string("op_4082_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4082_end_0 = const()[name = string("op_4082_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4082_end_mask_0 = const()[name = string("op_4082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4082_cast_fp16 = slice_by_index(begin = var_4082_begin_0, end = var_4082_end_0, end_mask = var_4082_end_mask_0, x = query_9_cast_fp16)[name = string("op_4082_cast_fp16")]; + tensor var_4086_begin_0 = const()[name = string("op_4086_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4086_end_0 = const()[name = string("op_4086_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4086_end_mask_0 = const()[name = string("op_4086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4086_cast_fp16 = slice_by_index(begin = var_4086_begin_0, end = var_4086_end_0, end_mask = var_4086_end_mask_0, x = query_9_cast_fp16)[name = string("op_4086_cast_fp16")]; + tensor var_4090_begin_0 = const()[name = string("op_4090_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4090_end_0 = const()[name = string("op_4090_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4090_end_mask_0 = const()[name = string("op_4090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4090_cast_fp16 = slice_by_index(begin = var_4090_begin_0, end = var_4090_end_0, end_mask = var_4090_end_mask_0, x = query_9_cast_fp16)[name = string("op_4090_cast_fp16")]; + tensor var_4094_begin_0 = const()[name = string("op_4094_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4094_end_0 = const()[name = string("op_4094_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4094_end_mask_0 = const()[name = string("op_4094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4094_cast_fp16 = slice_by_index(begin = var_4094_begin_0, end = var_4094_end_0, end_mask = var_4094_end_mask_0, x = query_9_cast_fp16)[name = string("op_4094_cast_fp16")]; + tensor var_4098_begin_0 = const()[name = string("op_4098_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4098_end_0 = const()[name = string("op_4098_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4098_end_mask_0 = const()[name = string("op_4098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4098_cast_fp16 = slice_by_index(begin = var_4098_begin_0, end = var_4098_end_0, end_mask = var_4098_end_mask_0, x = query_9_cast_fp16)[name = string("op_4098_cast_fp16")]; + tensor var_4102_begin_0 = const()[name = string("op_4102_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4102_end_0 = const()[name = string("op_4102_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4102_end_mask_0 = const()[name = string("op_4102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4102_cast_fp16 = slice_by_index(begin = var_4102_begin_0, end = var_4102_end_0, end_mask = var_4102_end_mask_0, x = query_9_cast_fp16)[name = string("op_4102_cast_fp16")]; + tensor var_4106_begin_0 = const()[name = string("op_4106_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4106_end_0 = const()[name = string("op_4106_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4106_end_mask_0 = const()[name = string("op_4106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4106_cast_fp16 = slice_by_index(begin = var_4106_begin_0, end = var_4106_end_0, end_mask = var_4106_end_mask_0, x = query_9_cast_fp16)[name = string("op_4106_cast_fp16")]; + tensor var_4110_begin_0 = const()[name = string("op_4110_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4110_end_0 = const()[name = string("op_4110_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4110_end_mask_0 = const()[name = string("op_4110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4110_cast_fp16 = slice_by_index(begin = var_4110_begin_0, end = var_4110_end_0, end_mask = var_4110_end_mask_0, x = query_9_cast_fp16)[name = string("op_4110_cast_fp16")]; + tensor var_4119_begin_0 = const()[name = string("op_4119_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4119_end_0 = const()[name = string("op_4119_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4119_end_mask_0 = const()[name = string("op_4119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4119_cast_fp16 = slice_by_index(begin = var_4119_begin_0, end = var_4119_end_0, end_mask = var_4119_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4119_cast_fp16")]; + tensor var_4126_begin_0 = const()[name = string("op_4126_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4126_end_0 = const()[name = string("op_4126_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4126_end_mask_0 = const()[name = string("op_4126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4126_cast_fp16 = slice_by_index(begin = var_4126_begin_0, end = var_4126_end_0, end_mask = var_4126_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4126_cast_fp16")]; + tensor var_4133_begin_0 = const()[name = string("op_4133_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4133_end_0 = const()[name = string("op_4133_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4133_end_mask_0 = const()[name = string("op_4133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4133_cast_fp16 = slice_by_index(begin = var_4133_begin_0, end = var_4133_end_0, end_mask = var_4133_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4133_cast_fp16")]; + tensor var_4140_begin_0 = const()[name = string("op_4140_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4140_end_0 = const()[name = string("op_4140_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4140_end_mask_0 = const()[name = string("op_4140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4140_cast_fp16 = slice_by_index(begin = var_4140_begin_0, end = var_4140_end_0, end_mask = var_4140_end_mask_0, x = var_4066_cast_fp16)[name = string("op_4140_cast_fp16")]; + tensor var_4147_begin_0 = const()[name = string("op_4147_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4147_end_0 = const()[name = string("op_4147_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4147_end_mask_0 = const()[name = string("op_4147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4147_cast_fp16 = slice_by_index(begin = var_4147_begin_0, end = var_4147_end_0, end_mask = var_4147_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4147_cast_fp16")]; + tensor var_4154_begin_0 = const()[name = string("op_4154_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4154_end_0 = const()[name = string("op_4154_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4154_end_mask_0 = const()[name = string("op_4154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4154_cast_fp16 = slice_by_index(begin = var_4154_begin_0, end = var_4154_end_0, end_mask = var_4154_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4154_cast_fp16")]; + tensor var_4161_begin_0 = const()[name = string("op_4161_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4161_end_0 = const()[name = string("op_4161_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4161_end_mask_0 = const()[name = string("op_4161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4161_cast_fp16 = slice_by_index(begin = var_4161_begin_0, end = var_4161_end_0, end_mask = var_4161_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4161_cast_fp16")]; + tensor var_4168_begin_0 = const()[name = string("op_4168_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4168_end_0 = const()[name = string("op_4168_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4168_end_mask_0 = const()[name = string("op_4168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4168_cast_fp16 = slice_by_index(begin = var_4168_begin_0, end = var_4168_end_0, end_mask = var_4168_end_mask_0, x = var_4070_cast_fp16)[name = string("op_4168_cast_fp16")]; + tensor var_4175_begin_0 = const()[name = string("op_4175_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4175_end_0 = const()[name = string("op_4175_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4175_end_mask_0 = const()[name = string("op_4175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4175_cast_fp16 = slice_by_index(begin = var_4175_begin_0, end = var_4175_end_0, end_mask = var_4175_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4175_cast_fp16")]; + tensor var_4182_begin_0 = const()[name = string("op_4182_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4182_end_0 = const()[name = string("op_4182_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4182_end_mask_0 = const()[name = string("op_4182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4182_cast_fp16 = slice_by_index(begin = var_4182_begin_0, end = var_4182_end_0, end_mask = var_4182_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4182_cast_fp16")]; + tensor var_4189_begin_0 = const()[name = string("op_4189_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4189_end_0 = const()[name = string("op_4189_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4189_end_mask_0 = const()[name = string("op_4189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4189_cast_fp16 = slice_by_index(begin = var_4189_begin_0, end = var_4189_end_0, end_mask = var_4189_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4189_cast_fp16")]; + tensor var_4196_begin_0 = const()[name = string("op_4196_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4196_end_0 = const()[name = string("op_4196_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4196_end_mask_0 = const()[name = string("op_4196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4196_cast_fp16 = slice_by_index(begin = var_4196_begin_0, end = var_4196_end_0, end_mask = var_4196_end_mask_0, x = var_4074_cast_fp16)[name = string("op_4196_cast_fp16")]; + tensor var_4203_begin_0 = const()[name = string("op_4203_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4203_end_0 = const()[name = string("op_4203_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4203_end_mask_0 = const()[name = string("op_4203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4203_cast_fp16 = slice_by_index(begin = var_4203_begin_0, end = var_4203_end_0, end_mask = var_4203_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4203_cast_fp16")]; + tensor var_4210_begin_0 = const()[name = string("op_4210_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4210_end_0 = const()[name = string("op_4210_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4210_end_mask_0 = const()[name = string("op_4210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4210_cast_fp16 = slice_by_index(begin = var_4210_begin_0, end = var_4210_end_0, end_mask = var_4210_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4210_cast_fp16")]; + tensor var_4217_begin_0 = const()[name = string("op_4217_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4217_end_0 = const()[name = string("op_4217_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4217_end_mask_0 = const()[name = string("op_4217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4217_cast_fp16 = slice_by_index(begin = var_4217_begin_0, end = var_4217_end_0, end_mask = var_4217_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4217_cast_fp16")]; + tensor var_4224_begin_0 = const()[name = string("op_4224_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4224_end_0 = const()[name = string("op_4224_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4224_end_mask_0 = const()[name = string("op_4224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4224_cast_fp16 = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = var_4078_cast_fp16)[name = string("op_4224_cast_fp16")]; + tensor var_4231_begin_0 = const()[name = string("op_4231_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4231_end_0 = const()[name = string("op_4231_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4231_end_mask_0 = const()[name = string("op_4231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4231_cast_fp16 = slice_by_index(begin = var_4231_begin_0, end = var_4231_end_0, end_mask = var_4231_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4231_cast_fp16")]; + tensor var_4238_begin_0 = const()[name = string("op_4238_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4238_end_0 = const()[name = string("op_4238_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4238_end_mask_0 = const()[name = string("op_4238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4238_cast_fp16 = slice_by_index(begin = var_4238_begin_0, end = var_4238_end_0, end_mask = var_4238_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4238_cast_fp16")]; + tensor var_4245_begin_0 = const()[name = string("op_4245_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4245_end_0 = const()[name = string("op_4245_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4245_end_mask_0 = const()[name = string("op_4245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4245_cast_fp16 = slice_by_index(begin = var_4245_begin_0, end = var_4245_end_0, end_mask = var_4245_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4245_cast_fp16")]; + tensor var_4252_begin_0 = const()[name = string("op_4252_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4252_end_0 = const()[name = string("op_4252_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4252_end_mask_0 = const()[name = string("op_4252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4252_cast_fp16 = slice_by_index(begin = var_4252_begin_0, end = var_4252_end_0, end_mask = var_4252_end_mask_0, x = var_4082_cast_fp16)[name = string("op_4252_cast_fp16")]; + tensor var_4259_begin_0 = const()[name = string("op_4259_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4259_end_0 = const()[name = string("op_4259_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4259_end_mask_0 = const()[name = string("op_4259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4259_cast_fp16 = slice_by_index(begin = var_4259_begin_0, end = var_4259_end_0, end_mask = var_4259_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4259_cast_fp16")]; + tensor var_4266_begin_0 = const()[name = string("op_4266_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4266_end_0 = const()[name = string("op_4266_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4266_end_mask_0 = const()[name = string("op_4266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4266_cast_fp16")]; + tensor var_4273_begin_0 = const()[name = string("op_4273_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4273_end_0 = const()[name = string("op_4273_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4273_end_mask_0 = const()[name = string("op_4273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4273_cast_fp16 = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4273_cast_fp16")]; + tensor var_4280_begin_0 = const()[name = string("op_4280_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4280_end_0 = const()[name = string("op_4280_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4280_end_mask_0 = const()[name = string("op_4280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = var_4086_cast_fp16)[name = string("op_4280_cast_fp16")]; + tensor var_4287_begin_0 = const()[name = string("op_4287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4287_end_0 = const()[name = string("op_4287_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4287_end_mask_0 = const()[name = string("op_4287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4287_cast_fp16 = slice_by_index(begin = var_4287_begin_0, end = var_4287_end_0, end_mask = var_4287_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4287_cast_fp16")]; + tensor var_4294_begin_0 = const()[name = string("op_4294_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4294_end_0 = const()[name = string("op_4294_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4294_end_mask_0 = const()[name = string("op_4294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4294_cast_fp16 = slice_by_index(begin = var_4294_begin_0, end = var_4294_end_0, end_mask = var_4294_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4294_cast_fp16")]; + tensor var_4301_begin_0 = const()[name = string("op_4301_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4301_end_0 = const()[name = string("op_4301_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4301_end_mask_0 = const()[name = string("op_4301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4301_cast_fp16 = slice_by_index(begin = var_4301_begin_0, end = var_4301_end_0, end_mask = var_4301_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4301_cast_fp16")]; + tensor var_4308_begin_0 = const()[name = string("op_4308_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4308_end_0 = const()[name = string("op_4308_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4308_end_mask_0 = const()[name = string("op_4308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4308_cast_fp16 = slice_by_index(begin = var_4308_begin_0, end = var_4308_end_0, end_mask = var_4308_end_mask_0, x = var_4090_cast_fp16)[name = string("op_4308_cast_fp16")]; + tensor var_4315_begin_0 = const()[name = string("op_4315_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4315_end_0 = const()[name = string("op_4315_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4315_end_mask_0 = const()[name = string("op_4315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4315_cast_fp16 = slice_by_index(begin = var_4315_begin_0, end = var_4315_end_0, end_mask = var_4315_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4315_cast_fp16")]; + tensor var_4322_begin_0 = const()[name = string("op_4322_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4322_end_0 = const()[name = string("op_4322_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4322_end_mask_0 = const()[name = string("op_4322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4322_cast_fp16 = slice_by_index(begin = var_4322_begin_0, end = var_4322_end_0, end_mask = var_4322_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4322_cast_fp16")]; + tensor var_4329_begin_0 = const()[name = string("op_4329_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4329_end_0 = const()[name = string("op_4329_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4329_end_mask_0 = const()[name = string("op_4329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4329_cast_fp16 = slice_by_index(begin = var_4329_begin_0, end = var_4329_end_0, end_mask = var_4329_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4329_cast_fp16")]; + tensor var_4336_begin_0 = const()[name = string("op_4336_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4336_end_0 = const()[name = string("op_4336_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4336_end_mask_0 = const()[name = string("op_4336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4336_cast_fp16 = slice_by_index(begin = var_4336_begin_0, end = var_4336_end_0, end_mask = var_4336_end_mask_0, x = var_4094_cast_fp16)[name = string("op_4336_cast_fp16")]; + tensor var_4343_begin_0 = const()[name = string("op_4343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4343_end_0 = const()[name = string("op_4343_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4343_end_mask_0 = const()[name = string("op_4343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4343_cast_fp16 = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4343_cast_fp16")]; + tensor var_4350_begin_0 = const()[name = string("op_4350_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4350_end_0 = const()[name = string("op_4350_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4350_end_mask_0 = const()[name = string("op_4350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4350_cast_fp16 = slice_by_index(begin = var_4350_begin_0, end = var_4350_end_0, end_mask = var_4350_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4350_cast_fp16")]; + tensor var_4357_begin_0 = const()[name = string("op_4357_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4357_end_0 = const()[name = string("op_4357_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4357_end_mask_0 = const()[name = string("op_4357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4357_cast_fp16 = slice_by_index(begin = var_4357_begin_0, end = var_4357_end_0, end_mask = var_4357_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4357_cast_fp16")]; + tensor var_4364_begin_0 = const()[name = string("op_4364_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4364_end_0 = const()[name = string("op_4364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4364_end_mask_0 = const()[name = string("op_4364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4364_cast_fp16 = slice_by_index(begin = var_4364_begin_0, end = var_4364_end_0, end_mask = var_4364_end_mask_0, x = var_4098_cast_fp16)[name = string("op_4364_cast_fp16")]; + tensor var_4371_begin_0 = const()[name = string("op_4371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4371_end_0 = const()[name = string("op_4371_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4371_end_mask_0 = const()[name = string("op_4371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4371_cast_fp16 = slice_by_index(begin = var_4371_begin_0, end = var_4371_end_0, end_mask = var_4371_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4371_cast_fp16")]; + tensor var_4378_begin_0 = const()[name = string("op_4378_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4378_end_0 = const()[name = string("op_4378_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4378_end_mask_0 = const()[name = string("op_4378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4378_cast_fp16 = slice_by_index(begin = var_4378_begin_0, end = var_4378_end_0, end_mask = var_4378_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4378_cast_fp16")]; + tensor var_4385_begin_0 = const()[name = string("op_4385_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4385_end_0 = const()[name = string("op_4385_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4385_end_mask_0 = const()[name = string("op_4385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4385_cast_fp16 = slice_by_index(begin = var_4385_begin_0, end = var_4385_end_0, end_mask = var_4385_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4385_cast_fp16")]; + tensor var_4392_begin_0 = const()[name = string("op_4392_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4392_end_0 = const()[name = string("op_4392_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4392_end_mask_0 = const()[name = string("op_4392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4392_cast_fp16 = slice_by_index(begin = var_4392_begin_0, end = var_4392_end_0, end_mask = var_4392_end_mask_0, x = var_4102_cast_fp16)[name = string("op_4392_cast_fp16")]; + tensor var_4399_begin_0 = const()[name = string("op_4399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4399_end_0 = const()[name = string("op_4399_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4399_end_mask_0 = const()[name = string("op_4399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4399_cast_fp16 = slice_by_index(begin = var_4399_begin_0, end = var_4399_end_0, end_mask = var_4399_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4399_cast_fp16")]; + tensor var_4406_begin_0 = const()[name = string("op_4406_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4406_end_0 = const()[name = string("op_4406_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4406_end_mask_0 = const()[name = string("op_4406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4406_cast_fp16 = slice_by_index(begin = var_4406_begin_0, end = var_4406_end_0, end_mask = var_4406_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4406_cast_fp16")]; + tensor var_4413_begin_0 = const()[name = string("op_4413_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4413_end_0 = const()[name = string("op_4413_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4413_end_mask_0 = const()[name = string("op_4413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4413_cast_fp16 = slice_by_index(begin = var_4413_begin_0, end = var_4413_end_0, end_mask = var_4413_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4413_cast_fp16")]; + tensor var_4420_begin_0 = const()[name = string("op_4420_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4420_end_0 = const()[name = string("op_4420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4420_end_mask_0 = const()[name = string("op_4420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4420_cast_fp16 = slice_by_index(begin = var_4420_begin_0, end = var_4420_end_0, end_mask = var_4420_end_mask_0, x = var_4106_cast_fp16)[name = string("op_4420_cast_fp16")]; + tensor var_4427_begin_0 = const()[name = string("op_4427_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4427_end_0 = const()[name = string("op_4427_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4427_end_mask_0 = const()[name = string("op_4427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4427_cast_fp16 = slice_by_index(begin = var_4427_begin_0, end = var_4427_end_0, end_mask = var_4427_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4427_cast_fp16")]; + tensor var_4434_begin_0 = const()[name = string("op_4434_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4434_end_0 = const()[name = string("op_4434_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4434_end_mask_0 = const()[name = string("op_4434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4434_cast_fp16 = slice_by_index(begin = var_4434_begin_0, end = var_4434_end_0, end_mask = var_4434_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4434_cast_fp16")]; + tensor var_4441_begin_0 = const()[name = string("op_4441_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4441_end_0 = const()[name = string("op_4441_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4441_end_mask_0 = const()[name = string("op_4441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4441_cast_fp16 = slice_by_index(begin = var_4441_begin_0, end = var_4441_end_0, end_mask = var_4441_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4441_cast_fp16")]; + tensor var_4448_begin_0 = const()[name = string("op_4448_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4448_end_0 = const()[name = string("op_4448_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4448_end_mask_0 = const()[name = string("op_4448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4448_cast_fp16 = slice_by_index(begin = var_4448_begin_0, end = var_4448_end_0, end_mask = var_4448_end_mask_0, x = var_4110_cast_fp16)[name = string("op_4448_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4453_begin_0 = const()[name = string("op_4453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4453_end_0 = const()[name = string("op_4453_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_4453_end_mask_0 = const()[name = string("op_4453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_7")]; + tensor var_4453_cast_fp16 = slice_by_index(begin = var_4453_begin_0, end = var_4453_end_0, end_mask = var_4453_end_mask_0, x = k_9_cast_fp16)[name = string("op_4453_cast_fp16")]; + tensor var_4457_begin_0 = const()[name = string("op_4457_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4457_end_0 = const()[name = string("op_4457_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_4457_end_mask_0 = const()[name = string("op_4457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4457_cast_fp16 = slice_by_index(begin = var_4457_begin_0, end = var_4457_end_0, end_mask = var_4457_end_mask_0, x = k_9_cast_fp16)[name = string("op_4457_cast_fp16")]; + tensor var_4461_begin_0 = const()[name = string("op_4461_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4461_end_0 = const()[name = string("op_4461_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_4461_end_mask_0 = const()[name = string("op_4461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4461_cast_fp16 = slice_by_index(begin = var_4461_begin_0, end = var_4461_end_0, end_mask = var_4461_end_mask_0, x = k_9_cast_fp16)[name = string("op_4461_cast_fp16")]; + tensor var_4465_begin_0 = const()[name = string("op_4465_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4465_end_0 = const()[name = string("op_4465_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_4465_end_mask_0 = const()[name = string("op_4465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4465_cast_fp16 = slice_by_index(begin = var_4465_begin_0, end = var_4465_end_0, end_mask = var_4465_end_mask_0, x = k_9_cast_fp16)[name = string("op_4465_cast_fp16")]; + tensor var_4469_begin_0 = const()[name = string("op_4469_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4469_end_0 = const()[name = string("op_4469_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_4469_end_mask_0 = const()[name = string("op_4469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4469_cast_fp16 = slice_by_index(begin = var_4469_begin_0, end = var_4469_end_0, end_mask = var_4469_end_mask_0, x = k_9_cast_fp16)[name = string("op_4469_cast_fp16")]; + tensor var_4473_begin_0 = const()[name = string("op_4473_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4473_end_0 = const()[name = string("op_4473_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_4473_end_mask_0 = const()[name = string("op_4473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4473_cast_fp16 = slice_by_index(begin = var_4473_begin_0, end = var_4473_end_0, end_mask = var_4473_end_mask_0, x = k_9_cast_fp16)[name = string("op_4473_cast_fp16")]; + tensor var_4477_begin_0 = const()[name = string("op_4477_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4477_end_0 = const()[name = string("op_4477_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_4477_end_mask_0 = const()[name = string("op_4477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4477_cast_fp16 = slice_by_index(begin = var_4477_begin_0, end = var_4477_end_0, end_mask = var_4477_end_mask_0, x = k_9_cast_fp16)[name = string("op_4477_cast_fp16")]; + tensor var_4481_begin_0 = const()[name = string("op_4481_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4481_end_0 = const()[name = string("op_4481_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_4481_end_mask_0 = const()[name = string("op_4481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4481_cast_fp16 = slice_by_index(begin = var_4481_begin_0, end = var_4481_end_0, end_mask = var_4481_end_mask_0, x = k_9_cast_fp16)[name = string("op_4481_cast_fp16")]; + tensor var_4485_begin_0 = const()[name = string("op_4485_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4485_end_0 = const()[name = string("op_4485_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_4485_end_mask_0 = const()[name = string("op_4485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4485_cast_fp16 = slice_by_index(begin = var_4485_begin_0, end = var_4485_end_0, end_mask = var_4485_end_mask_0, x = k_9_cast_fp16)[name = string("op_4485_cast_fp16")]; + tensor var_4489_begin_0 = const()[name = string("op_4489_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4489_end_0 = const()[name = string("op_4489_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_4489_end_mask_0 = const()[name = string("op_4489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4489_cast_fp16 = slice_by_index(begin = var_4489_begin_0, end = var_4489_end_0, end_mask = var_4489_end_mask_0, x = k_9_cast_fp16)[name = string("op_4489_cast_fp16")]; + tensor var_4493_begin_0 = const()[name = string("op_4493_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4493_end_0 = const()[name = string("op_4493_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_4493_end_mask_0 = const()[name = string("op_4493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4493_cast_fp16 = slice_by_index(begin = var_4493_begin_0, end = var_4493_end_0, end_mask = var_4493_end_mask_0, x = k_9_cast_fp16)[name = string("op_4493_cast_fp16")]; + tensor var_4497_begin_0 = const()[name = string("op_4497_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4497_end_0 = const()[name = string("op_4497_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_4497_end_mask_0 = const()[name = string("op_4497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4497_cast_fp16 = slice_by_index(begin = var_4497_begin_0, end = var_4497_end_0, end_mask = var_4497_end_mask_0, x = k_9_cast_fp16)[name = string("op_4497_cast_fp16")]; + tensor var_4499_begin_0 = const()[name = string("op_4499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4499_end_0 = const()[name = string("op_4499_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4499_end_mask_0 = const()[name = string("op_4499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4499_cast_fp16 = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = value_9_cast_fp16)[name = string("op_4499_cast_fp16")]; + tensor var_4503_begin_0 = const()[name = string("op_4503_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4503_end_0 = const()[name = string("op_4503_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4503_end_mask_0 = const()[name = string("op_4503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4503_cast_fp16 = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = value_9_cast_fp16)[name = string("op_4503_cast_fp16")]; + tensor var_4507_begin_0 = const()[name = string("op_4507_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4507_end_0 = const()[name = string("op_4507_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4507_end_mask_0 = const()[name = string("op_4507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4507_cast_fp16 = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = value_9_cast_fp16)[name = string("op_4507_cast_fp16")]; + tensor var_4511_begin_0 = const()[name = string("op_4511_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4511_end_0 = const()[name = string("op_4511_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4511_end_mask_0 = const()[name = string("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4511_cast_fp16 = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = value_9_cast_fp16)[name = string("op_4511_cast_fp16")]; + tensor var_4515_begin_0 = const()[name = string("op_4515_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4515_end_0 = const()[name = string("op_4515_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4515_end_mask_0 = const()[name = string("op_4515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4515_cast_fp16 = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = value_9_cast_fp16)[name = string("op_4515_cast_fp16")]; + tensor var_4519_begin_0 = const()[name = string("op_4519_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4519_end_0 = const()[name = string("op_4519_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4519_end_mask_0 = const()[name = string("op_4519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4519_cast_fp16 = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = value_9_cast_fp16)[name = string("op_4519_cast_fp16")]; + tensor var_4523_begin_0 = const()[name = string("op_4523_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4523_end_0 = const()[name = string("op_4523_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4523_end_mask_0 = const()[name = string("op_4523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4523_cast_fp16 = slice_by_index(begin = var_4523_begin_0, end = var_4523_end_0, end_mask = var_4523_end_mask_0, x = value_9_cast_fp16)[name = string("op_4523_cast_fp16")]; + tensor var_4527_begin_0 = const()[name = string("op_4527_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4527_end_0 = const()[name = string("op_4527_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4527_end_mask_0 = const()[name = string("op_4527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4527_cast_fp16 = slice_by_index(begin = var_4527_begin_0, end = var_4527_end_0, end_mask = var_4527_end_mask_0, x = value_9_cast_fp16)[name = string("op_4527_cast_fp16")]; + tensor var_4531_begin_0 = const()[name = string("op_4531_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4531_end_0 = const()[name = string("op_4531_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4531_end_mask_0 = const()[name = string("op_4531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4531_cast_fp16 = slice_by_index(begin = var_4531_begin_0, end = var_4531_end_0, end_mask = var_4531_end_mask_0, x = value_9_cast_fp16)[name = string("op_4531_cast_fp16")]; + tensor var_4535_begin_0 = const()[name = string("op_4535_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4535_end_0 = const()[name = string("op_4535_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4535_end_mask_0 = const()[name = string("op_4535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4535_cast_fp16 = slice_by_index(begin = var_4535_begin_0, end = var_4535_end_0, end_mask = var_4535_end_mask_0, x = value_9_cast_fp16)[name = string("op_4535_cast_fp16")]; + tensor var_4539_begin_0 = const()[name = string("op_4539_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4539_end_0 = const()[name = string("op_4539_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4539_end_mask_0 = const()[name = string("op_4539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4539_cast_fp16 = slice_by_index(begin = var_4539_begin_0, end = var_4539_end_0, end_mask = var_4539_end_mask_0, x = value_9_cast_fp16)[name = string("op_4539_cast_fp16")]; + tensor var_4543_begin_0 = const()[name = string("op_4543_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4543_end_0 = const()[name = string("op_4543_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4543_end_mask_0 = const()[name = string("op_4543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4543_cast_fp16 = slice_by_index(begin = var_4543_begin_0, end = var_4543_end_0, end_mask = var_4543_end_mask_0, x = value_9_cast_fp16)[name = string("op_4543_cast_fp16")]; + string _SplitHeadsQ__mh_w_385_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_385_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_385_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_385_equation_0, values = (var_4453_cast_fp16, var_4119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_385_cast_fp16")]; + string _SplitHeadsQ__mh_w_387_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_387_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_387_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_387_equation_0, values = (var_4453_cast_fp16, var_4126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_387_cast_fp16")]; + string _SplitHeadsQ__mh_w_389_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_389_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_389_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_389_equation_0, values = (var_4453_cast_fp16, var_4133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_389_cast_fp16")]; + string _SplitHeadsQ__mh_w_391_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_391_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_391_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_391_equation_0, values = (var_4453_cast_fp16, var_4140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_391_cast_fp16")]; + string _SplitHeadsQ__mh_w_393_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_393_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_393_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_393_equation_0, values = (var_4457_cast_fp16, var_4147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_393_cast_fp16")]; + string _SplitHeadsQ__mh_w_395_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_395_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_395_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_395_equation_0, values = (var_4457_cast_fp16, var_4154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_395_cast_fp16")]; + string _SplitHeadsQ__mh_w_397_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_397_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_397_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_397_equation_0, values = (var_4457_cast_fp16, var_4161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_397_cast_fp16")]; + string _SplitHeadsQ__mh_w_399_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_399_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_399_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_399_equation_0, values = (var_4457_cast_fp16, var_4168_cast_fp16))[name = string("_SplitHeadsQ__mh_w_399_cast_fp16")]; + string _SplitHeadsQ__mh_w_401_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_401_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_401_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_401_equation_0, values = (var_4461_cast_fp16, var_4175_cast_fp16))[name = string("_SplitHeadsQ__mh_w_401_cast_fp16")]; + string _SplitHeadsQ__mh_w_403_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_403_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_403_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_403_equation_0, values = (var_4461_cast_fp16, var_4182_cast_fp16))[name = string("_SplitHeadsQ__mh_w_403_cast_fp16")]; + string _SplitHeadsQ__mh_w_405_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_405_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_405_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_405_equation_0, values = (var_4461_cast_fp16, var_4189_cast_fp16))[name = string("_SplitHeadsQ__mh_w_405_cast_fp16")]; + string _SplitHeadsQ__mh_w_407_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_407_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_407_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_407_equation_0, values = (var_4461_cast_fp16, var_4196_cast_fp16))[name = string("_SplitHeadsQ__mh_w_407_cast_fp16")]; + string _SplitHeadsQ__mh_w_409_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_409_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_409_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_409_equation_0, values = (var_4465_cast_fp16, var_4203_cast_fp16))[name = string("_SplitHeadsQ__mh_w_409_cast_fp16")]; + string _SplitHeadsQ__mh_w_411_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_411_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_411_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_411_equation_0, values = (var_4465_cast_fp16, var_4210_cast_fp16))[name = string("_SplitHeadsQ__mh_w_411_cast_fp16")]; + string _SplitHeadsQ__mh_w_413_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_413_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_413_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_413_equation_0, values = (var_4465_cast_fp16, var_4217_cast_fp16))[name = string("_SplitHeadsQ__mh_w_413_cast_fp16")]; + string _SplitHeadsQ__mh_w_415_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_415_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_415_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_415_equation_0, values = (var_4465_cast_fp16, var_4224_cast_fp16))[name = string("_SplitHeadsQ__mh_w_415_cast_fp16")]; + string _SplitHeadsQ__mh_w_417_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_417_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_417_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_417_equation_0, values = (var_4469_cast_fp16, var_4231_cast_fp16))[name = string("_SplitHeadsQ__mh_w_417_cast_fp16")]; + string _SplitHeadsQ__mh_w_419_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_419_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_419_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_419_equation_0, values = (var_4469_cast_fp16, var_4238_cast_fp16))[name = string("_SplitHeadsQ__mh_w_419_cast_fp16")]; + string _SplitHeadsQ__mh_w_421_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_421_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_421_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_421_equation_0, values = (var_4469_cast_fp16, var_4245_cast_fp16))[name = string("_SplitHeadsQ__mh_w_421_cast_fp16")]; + string _SplitHeadsQ__mh_w_423_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_423_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_423_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_423_equation_0, values = (var_4469_cast_fp16, var_4252_cast_fp16))[name = string("_SplitHeadsQ__mh_w_423_cast_fp16")]; + string _SplitHeadsQ__mh_w_425_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_425_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_425_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_425_equation_0, values = (var_4473_cast_fp16, var_4259_cast_fp16))[name = string("_SplitHeadsQ__mh_w_425_cast_fp16")]; + string _SplitHeadsQ__mh_w_427_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_427_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_427_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_427_equation_0, values = (var_4473_cast_fp16, var_4266_cast_fp16))[name = string("_SplitHeadsQ__mh_w_427_cast_fp16")]; + string _SplitHeadsQ__mh_w_429_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_429_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_429_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_429_equation_0, values = (var_4473_cast_fp16, var_4273_cast_fp16))[name = string("_SplitHeadsQ__mh_w_429_cast_fp16")]; + string _SplitHeadsQ__mh_w_431_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_431_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_431_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_431_equation_0, values = (var_4473_cast_fp16, var_4280_cast_fp16))[name = string("_SplitHeadsQ__mh_w_431_cast_fp16")]; + string _SplitHeadsQ__mh_w_433_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_433_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_433_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_433_equation_0, values = (var_4477_cast_fp16, var_4287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_433_cast_fp16")]; + string _SplitHeadsQ__mh_w_435_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_435_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_435_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_435_equation_0, values = (var_4477_cast_fp16, var_4294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_435_cast_fp16")]; + string _SplitHeadsQ__mh_w_437_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_437_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_437_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_437_equation_0, values = (var_4477_cast_fp16, var_4301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_437_cast_fp16")]; + string _SplitHeadsQ__mh_w_439_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_439_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_439_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_439_equation_0, values = (var_4477_cast_fp16, var_4308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_439_cast_fp16")]; + string _SplitHeadsQ__mh_w_441_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_441_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_441_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_441_equation_0, values = (var_4481_cast_fp16, var_4315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_441_cast_fp16")]; + string _SplitHeadsQ__mh_w_443_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_443_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_443_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_443_equation_0, values = (var_4481_cast_fp16, var_4322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_443_cast_fp16")]; + string _SplitHeadsQ__mh_w_445_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_445_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_445_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_445_equation_0, values = (var_4481_cast_fp16, var_4329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_445_cast_fp16")]; + string _SplitHeadsQ__mh_w_447_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_447_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_447_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_447_equation_0, values = (var_4481_cast_fp16, var_4336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_447_cast_fp16")]; + string _SplitHeadsQ__mh_w_449_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_449_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_449_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_449_equation_0, values = (var_4485_cast_fp16, var_4343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_449_cast_fp16")]; + string _SplitHeadsQ__mh_w_451_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_451_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_451_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_451_equation_0, values = (var_4485_cast_fp16, var_4350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_451_cast_fp16")]; + string _SplitHeadsQ__mh_w_453_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_453_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_453_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_453_equation_0, values = (var_4485_cast_fp16, var_4357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_453_cast_fp16")]; + string _SplitHeadsQ__mh_w_455_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_455_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_455_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_455_equation_0, values = (var_4485_cast_fp16, var_4364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_455_cast_fp16")]; + string _SplitHeadsQ__mh_w_457_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_457_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_457_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_457_equation_0, values = (var_4489_cast_fp16, var_4371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_457_cast_fp16")]; + string _SplitHeadsQ__mh_w_459_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_459_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_459_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_459_equation_0, values = (var_4489_cast_fp16, var_4378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_459_cast_fp16")]; + string _SplitHeadsQ__mh_w_461_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_461_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_461_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_461_equation_0, values = (var_4489_cast_fp16, var_4385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_461_cast_fp16")]; + string _SplitHeadsQ__mh_w_463_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_463_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_463_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_463_equation_0, values = (var_4489_cast_fp16, var_4392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_463_cast_fp16")]; + string _SplitHeadsQ__mh_w_465_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_465_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_465_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_465_equation_0, values = (var_4493_cast_fp16, var_4399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_465_cast_fp16")]; + string _SplitHeadsQ__mh_w_467_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_467_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_467_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_467_equation_0, values = (var_4493_cast_fp16, var_4406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_467_cast_fp16")]; + string _SplitHeadsQ__mh_w_469_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_469_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_469_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_469_equation_0, values = (var_4493_cast_fp16, var_4413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_469_cast_fp16")]; + string _SplitHeadsQ__mh_w_471_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_471_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_471_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_471_equation_0, values = (var_4493_cast_fp16, var_4420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_471_cast_fp16")]; + string _SplitHeadsQ__mh_w_473_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_473_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_473_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_473_equation_0, values = (var_4497_cast_fp16, var_4427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_473_cast_fp16")]; + string _SplitHeadsQ__mh_w_475_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_475_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_475_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_475_equation_0, values = (var_4497_cast_fp16, var_4434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_475_cast_fp16")]; + string _SplitHeadsQ__mh_w_477_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_477_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_477_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_477_equation_0, values = (var_4497_cast_fp16, var_4441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_477_cast_fp16")]; + string _SplitHeadsQ__mh_w_479_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_479_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_479_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_479_equation_0, values = (var_4497_cast_fp16, var_4448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_479_cast_fp16")]; + fp16 var_4642_to_fp16 = const()[name = string("op_4642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_385_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_385_cast_fp16, y = var_4642_to_fp16)[name = string("aw_chunk_385_cast_fp16")]; + fp16 var_4644_to_fp16 = const()[name = string("op_4644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_387_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_387_cast_fp16, y = var_4644_to_fp16)[name = string("aw_chunk_387_cast_fp16")]; + fp16 var_4646_to_fp16 = const()[name = string("op_4646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_389_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_389_cast_fp16, y = var_4646_to_fp16)[name = string("aw_chunk_389_cast_fp16")]; + fp16 var_4648_to_fp16 = const()[name = string("op_4648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_391_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_391_cast_fp16, y = var_4648_to_fp16)[name = string("aw_chunk_391_cast_fp16")]; + fp16 var_4650_to_fp16 = const()[name = string("op_4650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_393_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_393_cast_fp16, y = var_4650_to_fp16)[name = string("aw_chunk_393_cast_fp16")]; + fp16 var_4652_to_fp16 = const()[name = string("op_4652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_395_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_395_cast_fp16, y = var_4652_to_fp16)[name = string("aw_chunk_395_cast_fp16")]; + fp16 var_4654_to_fp16 = const()[name = string("op_4654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_397_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_397_cast_fp16, y = var_4654_to_fp16)[name = string("aw_chunk_397_cast_fp16")]; + fp16 var_4656_to_fp16 = const()[name = string("op_4656_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_399_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_399_cast_fp16, y = var_4656_to_fp16)[name = string("aw_chunk_399_cast_fp16")]; + fp16 var_4658_to_fp16 = const()[name = string("op_4658_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_401_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_401_cast_fp16, y = var_4658_to_fp16)[name = string("aw_chunk_401_cast_fp16")]; + fp16 var_4660_to_fp16 = const()[name = string("op_4660_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_403_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_403_cast_fp16, y = var_4660_to_fp16)[name = string("aw_chunk_403_cast_fp16")]; + fp16 var_4662_to_fp16 = const()[name = string("op_4662_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_405_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_405_cast_fp16, y = var_4662_to_fp16)[name = string("aw_chunk_405_cast_fp16")]; + fp16 var_4664_to_fp16 = const()[name = string("op_4664_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_407_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_407_cast_fp16, y = var_4664_to_fp16)[name = string("aw_chunk_407_cast_fp16")]; + fp16 var_4666_to_fp16 = const()[name = string("op_4666_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_409_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_409_cast_fp16, y = var_4666_to_fp16)[name = string("aw_chunk_409_cast_fp16")]; + fp16 var_4668_to_fp16 = const()[name = string("op_4668_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_411_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_411_cast_fp16, y = var_4668_to_fp16)[name = string("aw_chunk_411_cast_fp16")]; + fp16 var_4670_to_fp16 = const()[name = string("op_4670_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_413_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_413_cast_fp16, y = var_4670_to_fp16)[name = string("aw_chunk_413_cast_fp16")]; + fp16 var_4672_to_fp16 = const()[name = string("op_4672_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_415_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_415_cast_fp16, y = var_4672_to_fp16)[name = string("aw_chunk_415_cast_fp16")]; + fp16 var_4674_to_fp16 = const()[name = string("op_4674_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_417_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_417_cast_fp16, y = var_4674_to_fp16)[name = string("aw_chunk_417_cast_fp16")]; + fp16 var_4676_to_fp16 = const()[name = string("op_4676_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_419_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_419_cast_fp16, y = var_4676_to_fp16)[name = string("aw_chunk_419_cast_fp16")]; + fp16 var_4678_to_fp16 = const()[name = string("op_4678_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_421_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_421_cast_fp16, y = var_4678_to_fp16)[name = string("aw_chunk_421_cast_fp16")]; + fp16 var_4680_to_fp16 = const()[name = string("op_4680_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_423_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_423_cast_fp16, y = var_4680_to_fp16)[name = string("aw_chunk_423_cast_fp16")]; + fp16 var_4682_to_fp16 = const()[name = string("op_4682_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_425_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_425_cast_fp16, y = var_4682_to_fp16)[name = string("aw_chunk_425_cast_fp16")]; + fp16 var_4684_to_fp16 = const()[name = string("op_4684_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_427_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_427_cast_fp16, y = var_4684_to_fp16)[name = string("aw_chunk_427_cast_fp16")]; + fp16 var_4686_to_fp16 = const()[name = string("op_4686_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_429_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_429_cast_fp16, y = var_4686_to_fp16)[name = string("aw_chunk_429_cast_fp16")]; + fp16 var_4688_to_fp16 = const()[name = string("op_4688_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_431_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_431_cast_fp16, y = var_4688_to_fp16)[name = string("aw_chunk_431_cast_fp16")]; + fp16 var_4690_to_fp16 = const()[name = string("op_4690_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_433_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_433_cast_fp16, y = var_4690_to_fp16)[name = string("aw_chunk_433_cast_fp16")]; + fp16 var_4692_to_fp16 = const()[name = string("op_4692_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_435_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_435_cast_fp16, y = var_4692_to_fp16)[name = string("aw_chunk_435_cast_fp16")]; + fp16 var_4694_to_fp16 = const()[name = string("op_4694_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_437_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_437_cast_fp16, y = var_4694_to_fp16)[name = string("aw_chunk_437_cast_fp16")]; + fp16 var_4696_to_fp16 = const()[name = string("op_4696_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_439_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_439_cast_fp16, y = var_4696_to_fp16)[name = string("aw_chunk_439_cast_fp16")]; + fp16 var_4698_to_fp16 = const()[name = string("op_4698_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_441_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_441_cast_fp16, y = var_4698_to_fp16)[name = string("aw_chunk_441_cast_fp16")]; + fp16 var_4700_to_fp16 = const()[name = string("op_4700_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_443_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_443_cast_fp16, y = var_4700_to_fp16)[name = string("aw_chunk_443_cast_fp16")]; + fp16 var_4702_to_fp16 = const()[name = string("op_4702_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_445_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_445_cast_fp16, y = var_4702_to_fp16)[name = string("aw_chunk_445_cast_fp16")]; + fp16 var_4704_to_fp16 = const()[name = string("op_4704_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_447_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_447_cast_fp16, y = var_4704_to_fp16)[name = string("aw_chunk_447_cast_fp16")]; + fp16 var_4706_to_fp16 = const()[name = string("op_4706_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_449_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_449_cast_fp16, y = var_4706_to_fp16)[name = string("aw_chunk_449_cast_fp16")]; + fp16 var_4708_to_fp16 = const()[name = string("op_4708_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_451_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_451_cast_fp16, y = var_4708_to_fp16)[name = string("aw_chunk_451_cast_fp16")]; + fp16 var_4710_to_fp16 = const()[name = string("op_4710_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_453_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_453_cast_fp16, y = var_4710_to_fp16)[name = string("aw_chunk_453_cast_fp16")]; + fp16 var_4712_to_fp16 = const()[name = string("op_4712_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_455_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_455_cast_fp16, y = var_4712_to_fp16)[name = string("aw_chunk_455_cast_fp16")]; + fp16 var_4714_to_fp16 = const()[name = string("op_4714_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_457_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_457_cast_fp16, y = var_4714_to_fp16)[name = string("aw_chunk_457_cast_fp16")]; + fp16 var_4716_to_fp16 = const()[name = string("op_4716_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_459_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_459_cast_fp16, y = var_4716_to_fp16)[name = string("aw_chunk_459_cast_fp16")]; + fp16 var_4718_to_fp16 = const()[name = string("op_4718_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_461_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_461_cast_fp16, y = var_4718_to_fp16)[name = string("aw_chunk_461_cast_fp16")]; + fp16 var_4720_to_fp16 = const()[name = string("op_4720_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_463_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_463_cast_fp16, y = var_4720_to_fp16)[name = string("aw_chunk_463_cast_fp16")]; + fp16 var_4722_to_fp16 = const()[name = string("op_4722_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_465_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_465_cast_fp16, y = var_4722_to_fp16)[name = string("aw_chunk_465_cast_fp16")]; + fp16 var_4724_to_fp16 = const()[name = string("op_4724_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_467_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_467_cast_fp16, y = var_4724_to_fp16)[name = string("aw_chunk_467_cast_fp16")]; + fp16 var_4726_to_fp16 = const()[name = string("op_4726_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_469_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_469_cast_fp16, y = var_4726_to_fp16)[name = string("aw_chunk_469_cast_fp16")]; + fp16 var_4728_to_fp16 = const()[name = string("op_4728_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_471_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_471_cast_fp16, y = var_4728_to_fp16)[name = string("aw_chunk_471_cast_fp16")]; + fp16 var_4730_to_fp16 = const()[name = string("op_4730_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_473_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_473_cast_fp16, y = var_4730_to_fp16)[name = string("aw_chunk_473_cast_fp16")]; + fp16 var_4732_to_fp16 = const()[name = string("op_4732_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_475_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_475_cast_fp16, y = var_4732_to_fp16)[name = string("aw_chunk_475_cast_fp16")]; + fp16 var_4734_to_fp16 = const()[name = string("op_4734_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_477_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_477_cast_fp16, y = var_4734_to_fp16)[name = string("aw_chunk_477_cast_fp16")]; + fp16 var_4736_to_fp16 = const()[name = string("op_4736_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_479_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_479_cast_fp16, y = var_4736_to_fp16)[name = string("aw_chunk_479_cast_fp16")]; + tensor var_4738_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_385_cast_fp16)[name = string("op_4738_cast_fp16")]; + tensor var_4739_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_387_cast_fp16)[name = string("op_4739_cast_fp16")]; + tensor var_4740_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_389_cast_fp16)[name = string("op_4740_cast_fp16")]; + tensor var_4741_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_391_cast_fp16)[name = string("op_4741_cast_fp16")]; + tensor var_4742_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_393_cast_fp16)[name = string("op_4742_cast_fp16")]; + tensor var_4743_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_395_cast_fp16)[name = string("op_4743_cast_fp16")]; + tensor var_4744_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_397_cast_fp16)[name = string("op_4744_cast_fp16")]; + tensor var_4745_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_399_cast_fp16)[name = string("op_4745_cast_fp16")]; + tensor var_4746_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_401_cast_fp16)[name = string("op_4746_cast_fp16")]; + tensor var_4747_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_403_cast_fp16)[name = string("op_4747_cast_fp16")]; + tensor var_4748_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_405_cast_fp16)[name = string("op_4748_cast_fp16")]; + tensor var_4749_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_407_cast_fp16)[name = string("op_4749_cast_fp16")]; + tensor var_4750_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_409_cast_fp16)[name = string("op_4750_cast_fp16")]; + tensor var_4751_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_411_cast_fp16)[name = string("op_4751_cast_fp16")]; + tensor var_4752_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_413_cast_fp16)[name = string("op_4752_cast_fp16")]; + tensor var_4753_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_415_cast_fp16)[name = string("op_4753_cast_fp16")]; + tensor var_4754_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_417_cast_fp16)[name = string("op_4754_cast_fp16")]; + tensor var_4755_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_419_cast_fp16)[name = string("op_4755_cast_fp16")]; + tensor var_4756_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_421_cast_fp16)[name = string("op_4756_cast_fp16")]; + tensor var_4757_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_423_cast_fp16)[name = string("op_4757_cast_fp16")]; + tensor var_4758_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_425_cast_fp16)[name = string("op_4758_cast_fp16")]; + tensor var_4759_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_427_cast_fp16)[name = string("op_4759_cast_fp16")]; + tensor var_4760_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_429_cast_fp16)[name = string("op_4760_cast_fp16")]; + tensor var_4761_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_431_cast_fp16)[name = string("op_4761_cast_fp16")]; + tensor var_4762_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_433_cast_fp16)[name = string("op_4762_cast_fp16")]; + tensor var_4763_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_435_cast_fp16)[name = string("op_4763_cast_fp16")]; + tensor var_4764_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_437_cast_fp16)[name = string("op_4764_cast_fp16")]; + tensor var_4765_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_439_cast_fp16)[name = string("op_4765_cast_fp16")]; + tensor var_4766_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_441_cast_fp16)[name = string("op_4766_cast_fp16")]; + tensor var_4767_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_443_cast_fp16)[name = string("op_4767_cast_fp16")]; + tensor var_4768_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_445_cast_fp16)[name = string("op_4768_cast_fp16")]; + tensor var_4769_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_447_cast_fp16)[name = string("op_4769_cast_fp16")]; + tensor var_4770_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_449_cast_fp16)[name = string("op_4770_cast_fp16")]; + tensor var_4771_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_451_cast_fp16)[name = string("op_4771_cast_fp16")]; + tensor var_4772_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_453_cast_fp16)[name = string("op_4772_cast_fp16")]; + tensor var_4773_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_455_cast_fp16)[name = string("op_4773_cast_fp16")]; + tensor var_4774_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_457_cast_fp16)[name = string("op_4774_cast_fp16")]; + tensor var_4775_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_459_cast_fp16)[name = string("op_4775_cast_fp16")]; + tensor var_4776_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_461_cast_fp16)[name = string("op_4776_cast_fp16")]; + tensor var_4777_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_463_cast_fp16)[name = string("op_4777_cast_fp16")]; + tensor var_4778_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_465_cast_fp16)[name = string("op_4778_cast_fp16")]; + tensor var_4779_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_467_cast_fp16)[name = string("op_4779_cast_fp16")]; + tensor var_4780_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_469_cast_fp16)[name = string("op_4780_cast_fp16")]; + tensor var_4781_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_471_cast_fp16)[name = string("op_4781_cast_fp16")]; + tensor var_4782_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_473_cast_fp16)[name = string("op_4782_cast_fp16")]; + tensor var_4783_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_475_cast_fp16)[name = string("op_4783_cast_fp16")]; + tensor var_4784_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_477_cast_fp16)[name = string("op_4784_cast_fp16")]; + tensor var_4785_cast_fp16 = softmax(axis = var_4011, x = aw_chunk_479_cast_fp16)[name = string("op_4785_cast_fp16")]; + string var_4787_equation_0 = const()[name = string("op_4787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4787_cast_fp16 = einsum(equation = var_4787_equation_0, values = (var_4499_cast_fp16, var_4738_cast_fp16))[name = string("op_4787_cast_fp16")]; + string var_4789_equation_0 = const()[name = string("op_4789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4789_cast_fp16 = einsum(equation = var_4789_equation_0, values = (var_4499_cast_fp16, var_4739_cast_fp16))[name = string("op_4789_cast_fp16")]; + string var_4791_equation_0 = const()[name = string("op_4791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4791_cast_fp16 = einsum(equation = var_4791_equation_0, values = (var_4499_cast_fp16, var_4740_cast_fp16))[name = string("op_4791_cast_fp16")]; + string var_4793_equation_0 = const()[name = string("op_4793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4793_cast_fp16 = einsum(equation = var_4793_equation_0, values = (var_4499_cast_fp16, var_4741_cast_fp16))[name = string("op_4793_cast_fp16")]; + string var_4795_equation_0 = const()[name = string("op_4795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4795_cast_fp16 = einsum(equation = var_4795_equation_0, values = (var_4503_cast_fp16, var_4742_cast_fp16))[name = string("op_4795_cast_fp16")]; + string var_4797_equation_0 = const()[name = string("op_4797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4797_cast_fp16 = einsum(equation = var_4797_equation_0, values = (var_4503_cast_fp16, var_4743_cast_fp16))[name = string("op_4797_cast_fp16")]; + string var_4799_equation_0 = const()[name = string("op_4799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4799_cast_fp16 = einsum(equation = var_4799_equation_0, values = (var_4503_cast_fp16, var_4744_cast_fp16))[name = string("op_4799_cast_fp16")]; + string var_4801_equation_0 = const()[name = string("op_4801_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4801_cast_fp16 = einsum(equation = var_4801_equation_0, values = (var_4503_cast_fp16, var_4745_cast_fp16))[name = string("op_4801_cast_fp16")]; + string var_4803_equation_0 = const()[name = string("op_4803_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4803_cast_fp16 = einsum(equation = var_4803_equation_0, values = (var_4507_cast_fp16, var_4746_cast_fp16))[name = string("op_4803_cast_fp16")]; + string var_4805_equation_0 = const()[name = string("op_4805_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4805_cast_fp16 = einsum(equation = var_4805_equation_0, values = (var_4507_cast_fp16, var_4747_cast_fp16))[name = string("op_4805_cast_fp16")]; + string var_4807_equation_0 = const()[name = string("op_4807_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4807_cast_fp16 = einsum(equation = var_4807_equation_0, values = (var_4507_cast_fp16, var_4748_cast_fp16))[name = string("op_4807_cast_fp16")]; + string var_4809_equation_0 = const()[name = string("op_4809_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4809_cast_fp16 = einsum(equation = var_4809_equation_0, values = (var_4507_cast_fp16, var_4749_cast_fp16))[name = string("op_4809_cast_fp16")]; + string var_4811_equation_0 = const()[name = string("op_4811_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4811_cast_fp16 = einsum(equation = var_4811_equation_0, values = (var_4511_cast_fp16, var_4750_cast_fp16))[name = string("op_4811_cast_fp16")]; + string var_4813_equation_0 = const()[name = string("op_4813_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4813_cast_fp16 = einsum(equation = var_4813_equation_0, values = (var_4511_cast_fp16, var_4751_cast_fp16))[name = string("op_4813_cast_fp16")]; + string var_4815_equation_0 = const()[name = string("op_4815_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4815_cast_fp16 = einsum(equation = var_4815_equation_0, values = (var_4511_cast_fp16, var_4752_cast_fp16))[name = string("op_4815_cast_fp16")]; + string var_4817_equation_0 = const()[name = string("op_4817_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4817_cast_fp16 = einsum(equation = var_4817_equation_0, values = (var_4511_cast_fp16, var_4753_cast_fp16))[name = string("op_4817_cast_fp16")]; + string var_4819_equation_0 = const()[name = string("op_4819_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4819_cast_fp16 = einsum(equation = var_4819_equation_0, values = (var_4515_cast_fp16, var_4754_cast_fp16))[name = string("op_4819_cast_fp16")]; + string var_4821_equation_0 = const()[name = string("op_4821_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4821_cast_fp16 = einsum(equation = var_4821_equation_0, values = (var_4515_cast_fp16, var_4755_cast_fp16))[name = string("op_4821_cast_fp16")]; + string var_4823_equation_0 = const()[name = string("op_4823_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4823_cast_fp16 = einsum(equation = var_4823_equation_0, values = (var_4515_cast_fp16, var_4756_cast_fp16))[name = string("op_4823_cast_fp16")]; + string var_4825_equation_0 = const()[name = string("op_4825_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4825_cast_fp16 = einsum(equation = var_4825_equation_0, values = (var_4515_cast_fp16, var_4757_cast_fp16))[name = string("op_4825_cast_fp16")]; + string var_4827_equation_0 = const()[name = string("op_4827_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4827_cast_fp16 = einsum(equation = var_4827_equation_0, values = (var_4519_cast_fp16, var_4758_cast_fp16))[name = string("op_4827_cast_fp16")]; + string var_4829_equation_0 = const()[name = string("op_4829_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4829_cast_fp16 = einsum(equation = var_4829_equation_0, values = (var_4519_cast_fp16, var_4759_cast_fp16))[name = string("op_4829_cast_fp16")]; + string var_4831_equation_0 = const()[name = string("op_4831_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4519_cast_fp16, var_4760_cast_fp16))[name = string("op_4831_cast_fp16")]; + string var_4833_equation_0 = const()[name = string("op_4833_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4833_cast_fp16 = einsum(equation = var_4833_equation_0, values = (var_4519_cast_fp16, var_4761_cast_fp16))[name = string("op_4833_cast_fp16")]; + string var_4835_equation_0 = const()[name = string("op_4835_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4835_cast_fp16 = einsum(equation = var_4835_equation_0, values = (var_4523_cast_fp16, var_4762_cast_fp16))[name = string("op_4835_cast_fp16")]; + string var_4837_equation_0 = const()[name = string("op_4837_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4837_cast_fp16 = einsum(equation = var_4837_equation_0, values = (var_4523_cast_fp16, var_4763_cast_fp16))[name = string("op_4837_cast_fp16")]; + string var_4839_equation_0 = const()[name = string("op_4839_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4839_cast_fp16 = einsum(equation = var_4839_equation_0, values = (var_4523_cast_fp16, var_4764_cast_fp16))[name = string("op_4839_cast_fp16")]; + string var_4841_equation_0 = const()[name = string("op_4841_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4841_cast_fp16 = einsum(equation = var_4841_equation_0, values = (var_4523_cast_fp16, var_4765_cast_fp16))[name = string("op_4841_cast_fp16")]; + string var_4843_equation_0 = const()[name = string("op_4843_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4843_cast_fp16 = einsum(equation = var_4843_equation_0, values = (var_4527_cast_fp16, var_4766_cast_fp16))[name = string("op_4843_cast_fp16")]; + string var_4845_equation_0 = const()[name = string("op_4845_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4845_cast_fp16 = einsum(equation = var_4845_equation_0, values = (var_4527_cast_fp16, var_4767_cast_fp16))[name = string("op_4845_cast_fp16")]; + string var_4847_equation_0 = const()[name = string("op_4847_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4847_cast_fp16 = einsum(equation = var_4847_equation_0, values = (var_4527_cast_fp16, var_4768_cast_fp16))[name = string("op_4847_cast_fp16")]; + string var_4849_equation_0 = const()[name = string("op_4849_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4849_cast_fp16 = einsum(equation = var_4849_equation_0, values = (var_4527_cast_fp16, var_4769_cast_fp16))[name = string("op_4849_cast_fp16")]; + string var_4851_equation_0 = const()[name = string("op_4851_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4851_cast_fp16 = einsum(equation = var_4851_equation_0, values = (var_4531_cast_fp16, var_4770_cast_fp16))[name = string("op_4851_cast_fp16")]; + string var_4853_equation_0 = const()[name = string("op_4853_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4853_cast_fp16 = einsum(equation = var_4853_equation_0, values = (var_4531_cast_fp16, var_4771_cast_fp16))[name = string("op_4853_cast_fp16")]; + string var_4855_equation_0 = const()[name = string("op_4855_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4855_cast_fp16 = einsum(equation = var_4855_equation_0, values = (var_4531_cast_fp16, var_4772_cast_fp16))[name = string("op_4855_cast_fp16")]; + string var_4857_equation_0 = const()[name = string("op_4857_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4857_cast_fp16 = einsum(equation = var_4857_equation_0, values = (var_4531_cast_fp16, var_4773_cast_fp16))[name = string("op_4857_cast_fp16")]; + string var_4859_equation_0 = const()[name = string("op_4859_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4859_cast_fp16 = einsum(equation = var_4859_equation_0, values = (var_4535_cast_fp16, var_4774_cast_fp16))[name = string("op_4859_cast_fp16")]; + string var_4861_equation_0 = const()[name = string("op_4861_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4861_cast_fp16 = einsum(equation = var_4861_equation_0, values = (var_4535_cast_fp16, var_4775_cast_fp16))[name = string("op_4861_cast_fp16")]; + string var_4863_equation_0 = const()[name = string("op_4863_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4863_cast_fp16 = einsum(equation = var_4863_equation_0, values = (var_4535_cast_fp16, var_4776_cast_fp16))[name = string("op_4863_cast_fp16")]; + string var_4865_equation_0 = const()[name = string("op_4865_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4865_cast_fp16 = einsum(equation = var_4865_equation_0, values = (var_4535_cast_fp16, var_4777_cast_fp16))[name = string("op_4865_cast_fp16")]; + string var_4867_equation_0 = const()[name = string("op_4867_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4867_cast_fp16 = einsum(equation = var_4867_equation_0, values = (var_4539_cast_fp16, var_4778_cast_fp16))[name = string("op_4867_cast_fp16")]; + string var_4869_equation_0 = const()[name = string("op_4869_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4869_cast_fp16 = einsum(equation = var_4869_equation_0, values = (var_4539_cast_fp16, var_4779_cast_fp16))[name = string("op_4869_cast_fp16")]; + string var_4871_equation_0 = const()[name = string("op_4871_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4871_cast_fp16 = einsum(equation = var_4871_equation_0, values = (var_4539_cast_fp16, var_4780_cast_fp16))[name = string("op_4871_cast_fp16")]; + string var_4873_equation_0 = const()[name = string("op_4873_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4873_cast_fp16 = einsum(equation = var_4873_equation_0, values = (var_4539_cast_fp16, var_4781_cast_fp16))[name = string("op_4873_cast_fp16")]; + string var_4875_equation_0 = const()[name = string("op_4875_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4875_cast_fp16 = einsum(equation = var_4875_equation_0, values = (var_4543_cast_fp16, var_4782_cast_fp16))[name = string("op_4875_cast_fp16")]; + string var_4877_equation_0 = const()[name = string("op_4877_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4877_cast_fp16 = einsum(equation = var_4877_equation_0, values = (var_4543_cast_fp16, var_4783_cast_fp16))[name = string("op_4877_cast_fp16")]; + string var_4879_equation_0 = const()[name = string("op_4879_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4879_cast_fp16 = einsum(equation = var_4879_equation_0, values = (var_4543_cast_fp16, var_4784_cast_fp16))[name = string("op_4879_cast_fp16")]; + string var_4881_equation_0 = const()[name = string("op_4881_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4881_cast_fp16 = einsum(equation = var_4881_equation_0, values = (var_4543_cast_fp16, var_4785_cast_fp16))[name = string("op_4881_cast_fp16")]; + bool var_4883_interleave_0 = const()[name = string("op_4883_interleave_0"), val = bool(false)]; + tensor var_4883_cast_fp16 = concat(axis = var_3994, interleave = var_4883_interleave_0, values = (var_4787_cast_fp16, var_4789_cast_fp16, var_4791_cast_fp16, var_4793_cast_fp16))[name = string("op_4883_cast_fp16")]; + bool var_4885_interleave_0 = const()[name = string("op_4885_interleave_0"), val = bool(false)]; + tensor var_4885_cast_fp16 = concat(axis = var_3994, interleave = var_4885_interleave_0, values = (var_4795_cast_fp16, var_4797_cast_fp16, var_4799_cast_fp16, var_4801_cast_fp16))[name = string("op_4885_cast_fp16")]; + bool var_4887_interleave_0 = const()[name = string("op_4887_interleave_0"), val = bool(false)]; + tensor var_4887_cast_fp16 = concat(axis = var_3994, interleave = var_4887_interleave_0, values = (var_4803_cast_fp16, var_4805_cast_fp16, var_4807_cast_fp16, var_4809_cast_fp16))[name = string("op_4887_cast_fp16")]; + bool var_4889_interleave_0 = const()[name = string("op_4889_interleave_0"), val = bool(false)]; + tensor var_4889_cast_fp16 = concat(axis = var_3994, interleave = var_4889_interleave_0, values = (var_4811_cast_fp16, var_4813_cast_fp16, var_4815_cast_fp16, var_4817_cast_fp16))[name = string("op_4889_cast_fp16")]; + bool var_4891_interleave_0 = const()[name = string("op_4891_interleave_0"), val = bool(false)]; + tensor var_4891_cast_fp16 = concat(axis = var_3994, interleave = var_4891_interleave_0, values = (var_4819_cast_fp16, var_4821_cast_fp16, var_4823_cast_fp16, var_4825_cast_fp16))[name = string("op_4891_cast_fp16")]; + bool var_4893_interleave_0 = const()[name = string("op_4893_interleave_0"), val = bool(false)]; + tensor var_4893_cast_fp16 = concat(axis = var_3994, interleave = var_4893_interleave_0, values = (var_4827_cast_fp16, var_4829_cast_fp16, var_4831_cast_fp16, var_4833_cast_fp16))[name = string("op_4893_cast_fp16")]; + bool var_4895_interleave_0 = const()[name = string("op_4895_interleave_0"), val = bool(false)]; + tensor var_4895_cast_fp16 = concat(axis = var_3994, interleave = var_4895_interleave_0, values = (var_4835_cast_fp16, var_4837_cast_fp16, var_4839_cast_fp16, var_4841_cast_fp16))[name = string("op_4895_cast_fp16")]; + bool var_4897_interleave_0 = const()[name = string("op_4897_interleave_0"), val = bool(false)]; + tensor var_4897_cast_fp16 = concat(axis = var_3994, interleave = var_4897_interleave_0, values = (var_4843_cast_fp16, var_4845_cast_fp16, var_4847_cast_fp16, var_4849_cast_fp16))[name = string("op_4897_cast_fp16")]; + bool var_4899_interleave_0 = const()[name = string("op_4899_interleave_0"), val = bool(false)]; + tensor var_4899_cast_fp16 = concat(axis = var_3994, interleave = var_4899_interleave_0, values = (var_4851_cast_fp16, var_4853_cast_fp16, var_4855_cast_fp16, var_4857_cast_fp16))[name = string("op_4899_cast_fp16")]; + bool var_4901_interleave_0 = const()[name = string("op_4901_interleave_0"), val = bool(false)]; + tensor var_4901_cast_fp16 = concat(axis = var_3994, interleave = var_4901_interleave_0, values = (var_4859_cast_fp16, var_4861_cast_fp16, var_4863_cast_fp16, var_4865_cast_fp16))[name = string("op_4901_cast_fp16")]; + bool var_4903_interleave_0 = const()[name = string("op_4903_interleave_0"), val = bool(false)]; + tensor var_4903_cast_fp16 = concat(axis = var_3994, interleave = var_4903_interleave_0, values = (var_4867_cast_fp16, var_4869_cast_fp16, var_4871_cast_fp16, var_4873_cast_fp16))[name = string("op_4903_cast_fp16")]; + bool var_4905_interleave_0 = const()[name = string("op_4905_interleave_0"), val = bool(false)]; + tensor var_4905_cast_fp16 = concat(axis = var_3994, interleave = var_4905_interleave_0, values = (var_4875_cast_fp16, var_4877_cast_fp16, var_4879_cast_fp16, var_4881_cast_fp16))[name = string("op_4905_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_4011, interleave = input_33_interleave_0, values = (var_4883_cast_fp16, var_4885_cast_fp16, var_4887_cast_fp16, var_4889_cast_fp16, var_4891_cast_fp16, var_4893_cast_fp16, var_4895_cast_fp16, var_4897_cast_fp16, var_4899_cast_fp16, var_4901_cast_fp16, var_4903_cast_fp16, var_4905_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66464448)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67644160)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_4924_to_fp16 = const()[name = string("op_4924_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_4924_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67645760)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67647360)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67648960)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72367616)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72373824)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77092480)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_4953 = const()[name = string("op_4953"), val = int32(3)]; + int32 var_4970 = const()[name = string("op_4970"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_4987_to_fp16 = const()[name = string("op_4987_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_4987_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77094080)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77095680)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77097280)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78276992)))]; + tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_11_cast_fp16")]; + string key_11_pad_type_0 = const()[name = string("key_11_pad_type_0"), val = string("valid")]; + tensor key_11_strides_0 = const()[name = string("key_11_strides_0"), val = tensor([1, 1])]; + tensor key_11_pad_0 = const()[name = string("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_11_dilations_0 = const()[name = string("key_11_dilations_0"), val = tensor([1, 1])]; + int32 key_11_groups_0 = const()[name = string("key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78278592)))]; + tensor key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_11_cast_fp16")]; + string value_11_pad_type_0 = const()[name = string("value_11_pad_type_0"), val = string("valid")]; + tensor value_11_strides_0 = const()[name = string("value_11_strides_0"), val = tensor([1, 1])]; + tensor value_11_pad_0 = const()[name = string("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_11_dilations_0 = const()[name = string("value_11_dilations_0"), val = tensor([1, 1])]; + int32 value_11_groups_0 = const()[name = string("value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79458304)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80638016)))]; + tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_5025_begin_0 = const()[name = string("op_5025_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5025_end_0 = const()[name = string("op_5025_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5025_end_mask_0 = const()[name = string("op_5025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5025_cast_fp16 = slice_by_index(begin = var_5025_begin_0, end = var_5025_end_0, end_mask = var_5025_end_mask_0, x = query_11_cast_fp16)[name = string("op_5025_cast_fp16")]; + tensor var_5029_begin_0 = const()[name = string("op_5029_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5029_end_0 = const()[name = string("op_5029_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5029_end_mask_0 = const()[name = string("op_5029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5029_cast_fp16 = slice_by_index(begin = var_5029_begin_0, end = var_5029_end_0, end_mask = var_5029_end_mask_0, x = query_11_cast_fp16)[name = string("op_5029_cast_fp16")]; + tensor var_5033_begin_0 = const()[name = string("op_5033_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5033_end_0 = const()[name = string("op_5033_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5033_end_mask_0 = const()[name = string("op_5033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5033_cast_fp16 = slice_by_index(begin = var_5033_begin_0, end = var_5033_end_0, end_mask = var_5033_end_mask_0, x = query_11_cast_fp16)[name = string("op_5033_cast_fp16")]; + tensor var_5037_begin_0 = const()[name = string("op_5037_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5037_end_0 = const()[name = string("op_5037_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5037_end_mask_0 = const()[name = string("op_5037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5037_cast_fp16 = slice_by_index(begin = var_5037_begin_0, end = var_5037_end_0, end_mask = var_5037_end_mask_0, x = query_11_cast_fp16)[name = string("op_5037_cast_fp16")]; + tensor var_5041_begin_0 = const()[name = string("op_5041_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5041_end_0 = const()[name = string("op_5041_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5041_end_mask_0 = const()[name = string("op_5041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5041_cast_fp16 = slice_by_index(begin = var_5041_begin_0, end = var_5041_end_0, end_mask = var_5041_end_mask_0, x = query_11_cast_fp16)[name = string("op_5041_cast_fp16")]; + tensor var_5045_begin_0 = const()[name = string("op_5045_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5045_end_0 = const()[name = string("op_5045_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5045_end_mask_0 = const()[name = string("op_5045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5045_cast_fp16 = slice_by_index(begin = var_5045_begin_0, end = var_5045_end_0, end_mask = var_5045_end_mask_0, x = query_11_cast_fp16)[name = string("op_5045_cast_fp16")]; + tensor var_5049_begin_0 = const()[name = string("op_5049_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5049_end_0 = const()[name = string("op_5049_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5049_end_mask_0 = const()[name = string("op_5049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5049_cast_fp16 = slice_by_index(begin = var_5049_begin_0, end = var_5049_end_0, end_mask = var_5049_end_mask_0, x = query_11_cast_fp16)[name = string("op_5049_cast_fp16")]; + tensor var_5053_begin_0 = const()[name = string("op_5053_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5053_end_0 = const()[name = string("op_5053_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5053_end_mask_0 = const()[name = string("op_5053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5053_cast_fp16 = slice_by_index(begin = var_5053_begin_0, end = var_5053_end_0, end_mask = var_5053_end_mask_0, x = query_11_cast_fp16)[name = string("op_5053_cast_fp16")]; + tensor var_5057_begin_0 = const()[name = string("op_5057_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5057_end_0 = const()[name = string("op_5057_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5057_end_mask_0 = const()[name = string("op_5057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5057_cast_fp16 = slice_by_index(begin = var_5057_begin_0, end = var_5057_end_0, end_mask = var_5057_end_mask_0, x = query_11_cast_fp16)[name = string("op_5057_cast_fp16")]; + tensor var_5061_begin_0 = const()[name = string("op_5061_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5061_end_0 = const()[name = string("op_5061_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5061_end_mask_0 = const()[name = string("op_5061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5061_cast_fp16 = slice_by_index(begin = var_5061_begin_0, end = var_5061_end_0, end_mask = var_5061_end_mask_0, x = query_11_cast_fp16)[name = string("op_5061_cast_fp16")]; + tensor var_5065_begin_0 = const()[name = string("op_5065_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5065_end_0 = const()[name = string("op_5065_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5065_end_mask_0 = const()[name = string("op_5065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5065_cast_fp16 = slice_by_index(begin = var_5065_begin_0, end = var_5065_end_0, end_mask = var_5065_end_mask_0, x = query_11_cast_fp16)[name = string("op_5065_cast_fp16")]; + tensor var_5069_begin_0 = const()[name = string("op_5069_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5069_end_0 = const()[name = string("op_5069_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5069_end_mask_0 = const()[name = string("op_5069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5069_cast_fp16 = slice_by_index(begin = var_5069_begin_0, end = var_5069_end_0, end_mask = var_5069_end_mask_0, x = query_11_cast_fp16)[name = string("op_5069_cast_fp16")]; + tensor var_5078_begin_0 = const()[name = string("op_5078_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5078_end_0 = const()[name = string("op_5078_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5078_end_mask_0 = const()[name = string("op_5078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5078_cast_fp16 = slice_by_index(begin = var_5078_begin_0, end = var_5078_end_0, end_mask = var_5078_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5078_cast_fp16")]; + tensor var_5085_begin_0 = const()[name = string("op_5085_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5085_end_0 = const()[name = string("op_5085_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5085_end_mask_0 = const()[name = string("op_5085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5085_cast_fp16 = slice_by_index(begin = var_5085_begin_0, end = var_5085_end_0, end_mask = var_5085_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5085_cast_fp16")]; + tensor var_5092_begin_0 = const()[name = string("op_5092_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5092_end_0 = const()[name = string("op_5092_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5092_end_mask_0 = const()[name = string("op_5092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5092_cast_fp16 = slice_by_index(begin = var_5092_begin_0, end = var_5092_end_0, end_mask = var_5092_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5092_cast_fp16")]; + tensor var_5099_begin_0 = const()[name = string("op_5099_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5099_end_0 = const()[name = string("op_5099_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5099_end_mask_0 = const()[name = string("op_5099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5099_cast_fp16 = slice_by_index(begin = var_5099_begin_0, end = var_5099_end_0, end_mask = var_5099_end_mask_0, x = var_5025_cast_fp16)[name = string("op_5099_cast_fp16")]; + tensor var_5106_begin_0 = const()[name = string("op_5106_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5106_end_0 = const()[name = string("op_5106_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5106_end_mask_0 = const()[name = string("op_5106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5106_cast_fp16 = slice_by_index(begin = var_5106_begin_0, end = var_5106_end_0, end_mask = var_5106_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5106_cast_fp16")]; + tensor var_5113_begin_0 = const()[name = string("op_5113_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5113_end_0 = const()[name = string("op_5113_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5113_end_mask_0 = const()[name = string("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5113_cast_fp16 = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5113_cast_fp16")]; + tensor var_5120_begin_0 = const()[name = string("op_5120_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5120_end_0 = const()[name = string("op_5120_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5120_end_mask_0 = const()[name = string("op_5120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5120_cast_fp16 = slice_by_index(begin = var_5120_begin_0, end = var_5120_end_0, end_mask = var_5120_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5120_cast_fp16")]; + tensor var_5127_begin_0 = const()[name = string("op_5127_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5127_end_0 = const()[name = string("op_5127_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5127_end_mask_0 = const()[name = string("op_5127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5127_cast_fp16 = slice_by_index(begin = var_5127_begin_0, end = var_5127_end_0, end_mask = var_5127_end_mask_0, x = var_5029_cast_fp16)[name = string("op_5127_cast_fp16")]; + tensor var_5134_begin_0 = const()[name = string("op_5134_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5134_end_0 = const()[name = string("op_5134_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5134_end_mask_0 = const()[name = string("op_5134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5134_cast_fp16 = slice_by_index(begin = var_5134_begin_0, end = var_5134_end_0, end_mask = var_5134_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5134_cast_fp16")]; + tensor var_5141_begin_0 = const()[name = string("op_5141_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5141_end_0 = const()[name = string("op_5141_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5141_end_mask_0 = const()[name = string("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5141_cast_fp16 = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5141_cast_fp16")]; + tensor var_5148_begin_0 = const()[name = string("op_5148_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5148_end_0 = const()[name = string("op_5148_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5148_end_mask_0 = const()[name = string("op_5148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5148_cast_fp16 = slice_by_index(begin = var_5148_begin_0, end = var_5148_end_0, end_mask = var_5148_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5148_cast_fp16")]; + tensor var_5155_begin_0 = const()[name = string("op_5155_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5155_end_0 = const()[name = string("op_5155_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5155_end_mask_0 = const()[name = string("op_5155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5155_cast_fp16 = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = var_5033_cast_fp16)[name = string("op_5155_cast_fp16")]; + tensor var_5162_begin_0 = const()[name = string("op_5162_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5162_end_0 = const()[name = string("op_5162_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5162_end_mask_0 = const()[name = string("op_5162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5162_cast_fp16 = slice_by_index(begin = var_5162_begin_0, end = var_5162_end_0, end_mask = var_5162_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5162_cast_fp16")]; + tensor var_5169_begin_0 = const()[name = string("op_5169_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5169_end_0 = const()[name = string("op_5169_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5169_end_mask_0 = const()[name = string("op_5169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5169_cast_fp16 = slice_by_index(begin = var_5169_begin_0, end = var_5169_end_0, end_mask = var_5169_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5169_cast_fp16")]; + tensor var_5176_begin_0 = const()[name = string("op_5176_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5176_end_0 = const()[name = string("op_5176_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5176_end_mask_0 = const()[name = string("op_5176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5176_cast_fp16 = slice_by_index(begin = var_5176_begin_0, end = var_5176_end_0, end_mask = var_5176_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5176_cast_fp16")]; + tensor var_5183_begin_0 = const()[name = string("op_5183_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5183_end_0 = const()[name = string("op_5183_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5183_end_mask_0 = const()[name = string("op_5183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5183_cast_fp16 = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = var_5037_cast_fp16)[name = string("op_5183_cast_fp16")]; + tensor var_5190_begin_0 = const()[name = string("op_5190_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5190_end_0 = const()[name = string("op_5190_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5190_end_mask_0 = const()[name = string("op_5190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5190_cast_fp16 = slice_by_index(begin = var_5190_begin_0, end = var_5190_end_0, end_mask = var_5190_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5190_cast_fp16")]; + tensor var_5197_begin_0 = const()[name = string("op_5197_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5197_end_0 = const()[name = string("op_5197_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5197_end_mask_0 = const()[name = string("op_5197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5197_cast_fp16 = slice_by_index(begin = var_5197_begin_0, end = var_5197_end_0, end_mask = var_5197_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5197_cast_fp16")]; + tensor var_5204_begin_0 = const()[name = string("op_5204_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5204_end_0 = const()[name = string("op_5204_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5204_end_mask_0 = const()[name = string("op_5204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5204_cast_fp16 = slice_by_index(begin = var_5204_begin_0, end = var_5204_end_0, end_mask = var_5204_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5204_cast_fp16")]; + tensor var_5211_begin_0 = const()[name = string("op_5211_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5211_end_0 = const()[name = string("op_5211_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5211_end_mask_0 = const()[name = string("op_5211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5211_cast_fp16 = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = var_5041_cast_fp16)[name = string("op_5211_cast_fp16")]; + tensor var_5218_begin_0 = const()[name = string("op_5218_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5218_end_0 = const()[name = string("op_5218_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5218_end_mask_0 = const()[name = string("op_5218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5218_cast_fp16 = slice_by_index(begin = var_5218_begin_0, end = var_5218_end_0, end_mask = var_5218_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5218_cast_fp16")]; + tensor var_5225_begin_0 = const()[name = string("op_5225_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5225_end_0 = const()[name = string("op_5225_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5225_end_mask_0 = const()[name = string("op_5225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5225_cast_fp16 = slice_by_index(begin = var_5225_begin_0, end = var_5225_end_0, end_mask = var_5225_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5225_cast_fp16")]; + tensor var_5232_begin_0 = const()[name = string("op_5232_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5232_end_0 = const()[name = string("op_5232_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5232_end_mask_0 = const()[name = string("op_5232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5232_cast_fp16 = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5232_cast_fp16")]; + tensor var_5239_begin_0 = const()[name = string("op_5239_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5239_end_0 = const()[name = string("op_5239_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5239_end_mask_0 = const()[name = string("op_5239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = var_5045_cast_fp16)[name = string("op_5239_cast_fp16")]; + tensor var_5246_begin_0 = const()[name = string("op_5246_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5246_end_0 = const()[name = string("op_5246_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5246_end_mask_0 = const()[name = string("op_5246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5246_cast_fp16 = slice_by_index(begin = var_5246_begin_0, end = var_5246_end_0, end_mask = var_5246_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5246_cast_fp16")]; + tensor var_5253_begin_0 = const()[name = string("op_5253_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5253_end_0 = const()[name = string("op_5253_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5253_end_mask_0 = const()[name = string("op_5253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5253_cast_fp16 = slice_by_index(begin = var_5253_begin_0, end = var_5253_end_0, end_mask = var_5253_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5253_cast_fp16")]; + tensor var_5260_begin_0 = const()[name = string("op_5260_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5260_end_0 = const()[name = string("op_5260_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5260_end_mask_0 = const()[name = string("op_5260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5260_cast_fp16 = slice_by_index(begin = var_5260_begin_0, end = var_5260_end_0, end_mask = var_5260_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5260_cast_fp16")]; + tensor var_5267_begin_0 = const()[name = string("op_5267_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5267_end_0 = const()[name = string("op_5267_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5267_end_mask_0 = const()[name = string("op_5267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5267_cast_fp16 = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = var_5049_cast_fp16)[name = string("op_5267_cast_fp16")]; + tensor var_5274_begin_0 = const()[name = string("op_5274_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5274_end_0 = const()[name = string("op_5274_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5274_end_mask_0 = const()[name = string("op_5274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5274_cast_fp16 = slice_by_index(begin = var_5274_begin_0, end = var_5274_end_0, end_mask = var_5274_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5274_cast_fp16")]; + tensor var_5281_begin_0 = const()[name = string("op_5281_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5281_end_0 = const()[name = string("op_5281_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5281_end_mask_0 = const()[name = string("op_5281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5281_cast_fp16")]; + tensor var_5288_begin_0 = const()[name = string("op_5288_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5288_end_0 = const()[name = string("op_5288_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5288_end_mask_0 = const()[name = string("op_5288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5288_cast_fp16 = slice_by_index(begin = var_5288_begin_0, end = var_5288_end_0, end_mask = var_5288_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5288_cast_fp16")]; + tensor var_5295_begin_0 = const()[name = string("op_5295_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5295_end_0 = const()[name = string("op_5295_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5295_end_mask_0 = const()[name = string("op_5295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5295_cast_fp16 = slice_by_index(begin = var_5295_begin_0, end = var_5295_end_0, end_mask = var_5295_end_mask_0, x = var_5053_cast_fp16)[name = string("op_5295_cast_fp16")]; + tensor var_5302_begin_0 = const()[name = string("op_5302_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5302_end_0 = const()[name = string("op_5302_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5302_end_mask_0 = const()[name = string("op_5302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5302_cast_fp16 = slice_by_index(begin = var_5302_begin_0, end = var_5302_end_0, end_mask = var_5302_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5302_cast_fp16")]; + tensor var_5309_begin_0 = const()[name = string("op_5309_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5309_end_0 = const()[name = string("op_5309_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5309_end_mask_0 = const()[name = string("op_5309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5309_cast_fp16 = slice_by_index(begin = var_5309_begin_0, end = var_5309_end_0, end_mask = var_5309_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5309_cast_fp16")]; + tensor var_5316_begin_0 = const()[name = string("op_5316_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5316_end_0 = const()[name = string("op_5316_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5316_end_mask_0 = const()[name = string("op_5316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5316_cast_fp16 = slice_by_index(begin = var_5316_begin_0, end = var_5316_end_0, end_mask = var_5316_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5316_cast_fp16")]; + tensor var_5323_begin_0 = const()[name = string("op_5323_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5323_end_0 = const()[name = string("op_5323_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5323_end_mask_0 = const()[name = string("op_5323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5323_cast_fp16 = slice_by_index(begin = var_5323_begin_0, end = var_5323_end_0, end_mask = var_5323_end_mask_0, x = var_5057_cast_fp16)[name = string("op_5323_cast_fp16")]; + tensor var_5330_begin_0 = const()[name = string("op_5330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5330_end_0 = const()[name = string("op_5330_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5330_end_mask_0 = const()[name = string("op_5330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5330_cast_fp16 = slice_by_index(begin = var_5330_begin_0, end = var_5330_end_0, end_mask = var_5330_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5330_cast_fp16")]; + tensor var_5337_begin_0 = const()[name = string("op_5337_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5337_end_0 = const()[name = string("op_5337_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5337_end_mask_0 = const()[name = string("op_5337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5337_cast_fp16 = slice_by_index(begin = var_5337_begin_0, end = var_5337_end_0, end_mask = var_5337_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5337_cast_fp16")]; + tensor var_5344_begin_0 = const()[name = string("op_5344_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5344_end_0 = const()[name = string("op_5344_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5344_end_mask_0 = const()[name = string("op_5344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5344_cast_fp16 = slice_by_index(begin = var_5344_begin_0, end = var_5344_end_0, end_mask = var_5344_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5344_cast_fp16")]; + tensor var_5351_begin_0 = const()[name = string("op_5351_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5351_end_0 = const()[name = string("op_5351_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5351_end_mask_0 = const()[name = string("op_5351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5351_cast_fp16 = slice_by_index(begin = var_5351_begin_0, end = var_5351_end_0, end_mask = var_5351_end_mask_0, x = var_5061_cast_fp16)[name = string("op_5351_cast_fp16")]; + tensor var_5358_begin_0 = const()[name = string("op_5358_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5358_end_0 = const()[name = string("op_5358_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5358_end_mask_0 = const()[name = string("op_5358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5358_cast_fp16 = slice_by_index(begin = var_5358_begin_0, end = var_5358_end_0, end_mask = var_5358_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5358_cast_fp16")]; + tensor var_5365_begin_0 = const()[name = string("op_5365_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5365_end_0 = const()[name = string("op_5365_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5365_end_mask_0 = const()[name = string("op_5365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5365_cast_fp16 = slice_by_index(begin = var_5365_begin_0, end = var_5365_end_0, end_mask = var_5365_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5365_cast_fp16")]; + tensor var_5372_begin_0 = const()[name = string("op_5372_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5372_end_0 = const()[name = string("op_5372_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5372_end_mask_0 = const()[name = string("op_5372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5372_cast_fp16 = slice_by_index(begin = var_5372_begin_0, end = var_5372_end_0, end_mask = var_5372_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5372_cast_fp16")]; + tensor var_5379_begin_0 = const()[name = string("op_5379_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5379_end_0 = const()[name = string("op_5379_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5379_end_mask_0 = const()[name = string("op_5379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5379_cast_fp16 = slice_by_index(begin = var_5379_begin_0, end = var_5379_end_0, end_mask = var_5379_end_mask_0, x = var_5065_cast_fp16)[name = string("op_5379_cast_fp16")]; + tensor var_5386_begin_0 = const()[name = string("op_5386_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5386_end_0 = const()[name = string("op_5386_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5386_end_mask_0 = const()[name = string("op_5386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5386_cast_fp16 = slice_by_index(begin = var_5386_begin_0, end = var_5386_end_0, end_mask = var_5386_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5386_cast_fp16")]; + tensor var_5393_begin_0 = const()[name = string("op_5393_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5393_end_0 = const()[name = string("op_5393_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5393_end_mask_0 = const()[name = string("op_5393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5393_cast_fp16 = slice_by_index(begin = var_5393_begin_0, end = var_5393_end_0, end_mask = var_5393_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5393_cast_fp16")]; + tensor var_5400_begin_0 = const()[name = string("op_5400_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5400_end_0 = const()[name = string("op_5400_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5400_end_mask_0 = const()[name = string("op_5400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5400_cast_fp16 = slice_by_index(begin = var_5400_begin_0, end = var_5400_end_0, end_mask = var_5400_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5400_cast_fp16")]; + tensor var_5407_begin_0 = const()[name = string("op_5407_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5407_end_0 = const()[name = string("op_5407_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5407_end_mask_0 = const()[name = string("op_5407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5407_cast_fp16 = slice_by_index(begin = var_5407_begin_0, end = var_5407_end_0, end_mask = var_5407_end_mask_0, x = var_5069_cast_fp16)[name = string("op_5407_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5412_begin_0 = const()[name = string("op_5412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5412_end_0 = const()[name = string("op_5412_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_5412_end_mask_0 = const()[name = string("op_5412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = string("transpose_6")]; + tensor var_5412_cast_fp16 = slice_by_index(begin = var_5412_begin_0, end = var_5412_end_0, end_mask = var_5412_end_mask_0, x = k_11_cast_fp16)[name = string("op_5412_cast_fp16")]; + tensor var_5416_begin_0 = const()[name = string("op_5416_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5416_end_0 = const()[name = string("op_5416_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_5416_end_mask_0 = const()[name = string("op_5416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5416_cast_fp16 = slice_by_index(begin = var_5416_begin_0, end = var_5416_end_0, end_mask = var_5416_end_mask_0, x = k_11_cast_fp16)[name = string("op_5416_cast_fp16")]; + tensor var_5420_begin_0 = const()[name = string("op_5420_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5420_end_0 = const()[name = string("op_5420_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_5420_end_mask_0 = const()[name = string("op_5420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5420_cast_fp16 = slice_by_index(begin = var_5420_begin_0, end = var_5420_end_0, end_mask = var_5420_end_mask_0, x = k_11_cast_fp16)[name = string("op_5420_cast_fp16")]; + tensor var_5424_begin_0 = const()[name = string("op_5424_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5424_end_0 = const()[name = string("op_5424_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_5424_end_mask_0 = const()[name = string("op_5424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5424_cast_fp16 = slice_by_index(begin = var_5424_begin_0, end = var_5424_end_0, end_mask = var_5424_end_mask_0, x = k_11_cast_fp16)[name = string("op_5424_cast_fp16")]; + tensor var_5428_begin_0 = const()[name = string("op_5428_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5428_end_0 = const()[name = string("op_5428_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_5428_end_mask_0 = const()[name = string("op_5428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5428_cast_fp16 = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = k_11_cast_fp16)[name = string("op_5428_cast_fp16")]; + tensor var_5432_begin_0 = const()[name = string("op_5432_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5432_end_0 = const()[name = string("op_5432_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_5432_end_mask_0 = const()[name = string("op_5432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5432_cast_fp16 = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = k_11_cast_fp16)[name = string("op_5432_cast_fp16")]; + tensor var_5436_begin_0 = const()[name = string("op_5436_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5436_end_0 = const()[name = string("op_5436_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_5436_end_mask_0 = const()[name = string("op_5436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5436_cast_fp16 = slice_by_index(begin = var_5436_begin_0, end = var_5436_end_0, end_mask = var_5436_end_mask_0, x = k_11_cast_fp16)[name = string("op_5436_cast_fp16")]; + tensor var_5440_begin_0 = const()[name = string("op_5440_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5440_end_0 = const()[name = string("op_5440_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_5440_end_mask_0 = const()[name = string("op_5440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5440_cast_fp16 = slice_by_index(begin = var_5440_begin_0, end = var_5440_end_0, end_mask = var_5440_end_mask_0, x = k_11_cast_fp16)[name = string("op_5440_cast_fp16")]; + tensor var_5444_begin_0 = const()[name = string("op_5444_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5444_end_0 = const()[name = string("op_5444_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_5444_end_mask_0 = const()[name = string("op_5444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5444_cast_fp16 = slice_by_index(begin = var_5444_begin_0, end = var_5444_end_0, end_mask = var_5444_end_mask_0, x = k_11_cast_fp16)[name = string("op_5444_cast_fp16")]; + tensor var_5448_begin_0 = const()[name = string("op_5448_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5448_end_0 = const()[name = string("op_5448_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_5448_end_mask_0 = const()[name = string("op_5448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5448_cast_fp16 = slice_by_index(begin = var_5448_begin_0, end = var_5448_end_0, end_mask = var_5448_end_mask_0, x = k_11_cast_fp16)[name = string("op_5448_cast_fp16")]; + tensor var_5452_begin_0 = const()[name = string("op_5452_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5452_end_0 = const()[name = string("op_5452_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_5452_end_mask_0 = const()[name = string("op_5452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5452_cast_fp16 = slice_by_index(begin = var_5452_begin_0, end = var_5452_end_0, end_mask = var_5452_end_mask_0, x = k_11_cast_fp16)[name = string("op_5452_cast_fp16")]; + tensor var_5456_begin_0 = const()[name = string("op_5456_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5456_end_0 = const()[name = string("op_5456_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_5456_end_mask_0 = const()[name = string("op_5456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5456_cast_fp16 = slice_by_index(begin = var_5456_begin_0, end = var_5456_end_0, end_mask = var_5456_end_mask_0, x = k_11_cast_fp16)[name = string("op_5456_cast_fp16")]; + tensor var_5458_begin_0 = const()[name = string("op_5458_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5458_end_0 = const()[name = string("op_5458_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5458_end_mask_0 = const()[name = string("op_5458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5458_cast_fp16 = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = value_11_cast_fp16)[name = string("op_5458_cast_fp16")]; + tensor var_5462_begin_0 = const()[name = string("op_5462_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5462_end_0 = const()[name = string("op_5462_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5462_end_mask_0 = const()[name = string("op_5462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5462_cast_fp16 = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, x = value_11_cast_fp16)[name = string("op_5462_cast_fp16")]; + tensor var_5466_begin_0 = const()[name = string("op_5466_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5466_end_0 = const()[name = string("op_5466_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5466_end_mask_0 = const()[name = string("op_5466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5466_cast_fp16 = slice_by_index(begin = var_5466_begin_0, end = var_5466_end_0, end_mask = var_5466_end_mask_0, x = value_11_cast_fp16)[name = string("op_5466_cast_fp16")]; + tensor var_5470_begin_0 = const()[name = string("op_5470_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5470_end_0 = const()[name = string("op_5470_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5470_end_mask_0 = const()[name = string("op_5470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5470_cast_fp16 = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = value_11_cast_fp16)[name = string("op_5470_cast_fp16")]; + tensor var_5474_begin_0 = const()[name = string("op_5474_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5474_end_0 = const()[name = string("op_5474_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5474_end_mask_0 = const()[name = string("op_5474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5474_cast_fp16 = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = value_11_cast_fp16)[name = string("op_5474_cast_fp16")]; + tensor var_5478_begin_0 = const()[name = string("op_5478_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5478_end_0 = const()[name = string("op_5478_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5478_end_mask_0 = const()[name = string("op_5478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5478_cast_fp16 = slice_by_index(begin = var_5478_begin_0, end = var_5478_end_0, end_mask = var_5478_end_mask_0, x = value_11_cast_fp16)[name = string("op_5478_cast_fp16")]; + tensor var_5482_begin_0 = const()[name = string("op_5482_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5482_end_0 = const()[name = string("op_5482_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5482_end_mask_0 = const()[name = string("op_5482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5482_cast_fp16 = slice_by_index(begin = var_5482_begin_0, end = var_5482_end_0, end_mask = var_5482_end_mask_0, x = value_11_cast_fp16)[name = string("op_5482_cast_fp16")]; + tensor var_5486_begin_0 = const()[name = string("op_5486_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5486_end_0 = const()[name = string("op_5486_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5486_end_mask_0 = const()[name = string("op_5486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5486_cast_fp16 = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = value_11_cast_fp16)[name = string("op_5486_cast_fp16")]; + tensor var_5490_begin_0 = const()[name = string("op_5490_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5490_end_0 = const()[name = string("op_5490_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5490_end_mask_0 = const()[name = string("op_5490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5490_cast_fp16 = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = value_11_cast_fp16)[name = string("op_5490_cast_fp16")]; + tensor var_5494_begin_0 = const()[name = string("op_5494_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5494_end_0 = const()[name = string("op_5494_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5494_end_mask_0 = const()[name = string("op_5494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5494_cast_fp16 = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = value_11_cast_fp16)[name = string("op_5494_cast_fp16")]; + tensor var_5498_begin_0 = const()[name = string("op_5498_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5498_end_0 = const()[name = string("op_5498_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5498_end_mask_0 = const()[name = string("op_5498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5498_cast_fp16 = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = value_11_cast_fp16)[name = string("op_5498_cast_fp16")]; + tensor var_5502_begin_0 = const()[name = string("op_5502_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5502_end_0 = const()[name = string("op_5502_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5502_end_mask_0 = const()[name = string("op_5502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5502_cast_fp16 = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = value_11_cast_fp16)[name = string("op_5502_cast_fp16")]; + string _SplitHeadsQ__mh_w_481_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_481_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_481_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_481_equation_0, values = (var_5412_cast_fp16, var_5078_cast_fp16))[name = string("_SplitHeadsQ__mh_w_481_cast_fp16")]; + string _SplitHeadsQ__mh_w_483_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_483_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_483_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_483_equation_0, values = (var_5412_cast_fp16, var_5085_cast_fp16))[name = string("_SplitHeadsQ__mh_w_483_cast_fp16")]; + string _SplitHeadsQ__mh_w_485_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_485_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_485_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_485_equation_0, values = (var_5412_cast_fp16, var_5092_cast_fp16))[name = string("_SplitHeadsQ__mh_w_485_cast_fp16")]; + string _SplitHeadsQ__mh_w_487_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_487_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_487_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_487_equation_0, values = (var_5412_cast_fp16, var_5099_cast_fp16))[name = string("_SplitHeadsQ__mh_w_487_cast_fp16")]; + string _SplitHeadsQ__mh_w_489_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_489_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_489_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_489_equation_0, values = (var_5416_cast_fp16, var_5106_cast_fp16))[name = string("_SplitHeadsQ__mh_w_489_cast_fp16")]; + string _SplitHeadsQ__mh_w_491_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_491_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_491_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_491_equation_0, values = (var_5416_cast_fp16, var_5113_cast_fp16))[name = string("_SplitHeadsQ__mh_w_491_cast_fp16")]; + string _SplitHeadsQ__mh_w_493_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_493_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_493_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_493_equation_0, values = (var_5416_cast_fp16, var_5120_cast_fp16))[name = string("_SplitHeadsQ__mh_w_493_cast_fp16")]; + string _SplitHeadsQ__mh_w_495_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_495_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_495_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_495_equation_0, values = (var_5416_cast_fp16, var_5127_cast_fp16))[name = string("_SplitHeadsQ__mh_w_495_cast_fp16")]; + string _SplitHeadsQ__mh_w_497_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_497_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_497_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_497_equation_0, values = (var_5420_cast_fp16, var_5134_cast_fp16))[name = string("_SplitHeadsQ__mh_w_497_cast_fp16")]; + string _SplitHeadsQ__mh_w_499_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_499_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_499_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_499_equation_0, values = (var_5420_cast_fp16, var_5141_cast_fp16))[name = string("_SplitHeadsQ__mh_w_499_cast_fp16")]; + string _SplitHeadsQ__mh_w_501_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_501_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_501_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_501_equation_0, values = (var_5420_cast_fp16, var_5148_cast_fp16))[name = string("_SplitHeadsQ__mh_w_501_cast_fp16")]; + string _SplitHeadsQ__mh_w_503_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_503_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_503_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_503_equation_0, values = (var_5420_cast_fp16, var_5155_cast_fp16))[name = string("_SplitHeadsQ__mh_w_503_cast_fp16")]; + string _SplitHeadsQ__mh_w_505_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_505_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_505_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_505_equation_0, values = (var_5424_cast_fp16, var_5162_cast_fp16))[name = string("_SplitHeadsQ__mh_w_505_cast_fp16")]; + string _SplitHeadsQ__mh_w_507_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_507_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_507_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_507_equation_0, values = (var_5424_cast_fp16, var_5169_cast_fp16))[name = string("_SplitHeadsQ__mh_w_507_cast_fp16")]; + string _SplitHeadsQ__mh_w_509_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_509_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_509_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_509_equation_0, values = (var_5424_cast_fp16, var_5176_cast_fp16))[name = string("_SplitHeadsQ__mh_w_509_cast_fp16")]; + string _SplitHeadsQ__mh_w_511_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_511_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_511_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_511_equation_0, values = (var_5424_cast_fp16, var_5183_cast_fp16))[name = string("_SplitHeadsQ__mh_w_511_cast_fp16")]; + string _SplitHeadsQ__mh_w_513_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_513_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_513_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_513_equation_0, values = (var_5428_cast_fp16, var_5190_cast_fp16))[name = string("_SplitHeadsQ__mh_w_513_cast_fp16")]; + string _SplitHeadsQ__mh_w_515_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_515_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_515_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_515_equation_0, values = (var_5428_cast_fp16, var_5197_cast_fp16))[name = string("_SplitHeadsQ__mh_w_515_cast_fp16")]; + string _SplitHeadsQ__mh_w_517_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_517_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_517_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_517_equation_0, values = (var_5428_cast_fp16, var_5204_cast_fp16))[name = string("_SplitHeadsQ__mh_w_517_cast_fp16")]; + string _SplitHeadsQ__mh_w_519_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_519_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_519_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_519_equation_0, values = (var_5428_cast_fp16, var_5211_cast_fp16))[name = string("_SplitHeadsQ__mh_w_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_521_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_521_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_521_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_521_equation_0, values = (var_5432_cast_fp16, var_5218_cast_fp16))[name = string("_SplitHeadsQ__mh_w_521_cast_fp16")]; + string _SplitHeadsQ__mh_w_523_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_523_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_523_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_523_equation_0, values = (var_5432_cast_fp16, var_5225_cast_fp16))[name = string("_SplitHeadsQ__mh_w_523_cast_fp16")]; + string _SplitHeadsQ__mh_w_525_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_525_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_525_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_525_equation_0, values = (var_5432_cast_fp16, var_5232_cast_fp16))[name = string("_SplitHeadsQ__mh_w_525_cast_fp16")]; + string _SplitHeadsQ__mh_w_527_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_527_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_527_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_527_equation_0, values = (var_5432_cast_fp16, var_5239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_527_cast_fp16")]; + string _SplitHeadsQ__mh_w_529_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_529_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_529_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_529_equation_0, values = (var_5436_cast_fp16, var_5246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_529_cast_fp16")]; + string _SplitHeadsQ__mh_w_531_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_531_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_531_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_531_equation_0, values = (var_5436_cast_fp16, var_5253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_531_cast_fp16")]; + string _SplitHeadsQ__mh_w_533_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_533_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_533_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_533_equation_0, values = (var_5436_cast_fp16, var_5260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_533_cast_fp16")]; + string _SplitHeadsQ__mh_w_535_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_535_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_535_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_535_equation_0, values = (var_5436_cast_fp16, var_5267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_535_cast_fp16")]; + string _SplitHeadsQ__mh_w_537_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_537_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_537_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_537_equation_0, values = (var_5440_cast_fp16, var_5274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_537_cast_fp16")]; + string _SplitHeadsQ__mh_w_539_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_539_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_539_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_539_equation_0, values = (var_5440_cast_fp16, var_5281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_539_cast_fp16")]; + string _SplitHeadsQ__mh_w_541_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_541_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_541_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_541_equation_0, values = (var_5440_cast_fp16, var_5288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_541_cast_fp16")]; + string _SplitHeadsQ__mh_w_543_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_543_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_543_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_543_equation_0, values = (var_5440_cast_fp16, var_5295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_543_cast_fp16")]; + string _SplitHeadsQ__mh_w_545_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_545_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_545_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_545_equation_0, values = (var_5444_cast_fp16, var_5302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_545_cast_fp16")]; + string _SplitHeadsQ__mh_w_547_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_547_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_547_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_547_equation_0, values = (var_5444_cast_fp16, var_5309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_547_cast_fp16")]; + string _SplitHeadsQ__mh_w_549_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_549_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_549_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_549_equation_0, values = (var_5444_cast_fp16, var_5316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_549_cast_fp16")]; + string _SplitHeadsQ__mh_w_551_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_551_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_551_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_551_equation_0, values = (var_5444_cast_fp16, var_5323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_551_cast_fp16")]; + string _SplitHeadsQ__mh_w_553_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_553_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_553_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_553_equation_0, values = (var_5448_cast_fp16, var_5330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_553_cast_fp16")]; + string _SplitHeadsQ__mh_w_555_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_555_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_555_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_555_equation_0, values = (var_5448_cast_fp16, var_5337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_555_cast_fp16")]; + string _SplitHeadsQ__mh_w_557_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_557_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_557_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_557_equation_0, values = (var_5448_cast_fp16, var_5344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_557_cast_fp16")]; + string _SplitHeadsQ__mh_w_559_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_559_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_559_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_559_equation_0, values = (var_5448_cast_fp16, var_5351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_559_cast_fp16")]; + string _SplitHeadsQ__mh_w_561_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_561_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_561_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_561_equation_0, values = (var_5452_cast_fp16, var_5358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_561_cast_fp16")]; + string _SplitHeadsQ__mh_w_563_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_563_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_563_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_563_equation_0, values = (var_5452_cast_fp16, var_5365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_563_cast_fp16")]; + string _SplitHeadsQ__mh_w_565_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_565_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_565_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_565_equation_0, values = (var_5452_cast_fp16, var_5372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_565_cast_fp16")]; + string _SplitHeadsQ__mh_w_567_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_567_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_567_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_567_equation_0, values = (var_5452_cast_fp16, var_5379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_567_cast_fp16")]; + string _SplitHeadsQ__mh_w_569_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_569_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_569_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_569_equation_0, values = (var_5456_cast_fp16, var_5386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_569_cast_fp16")]; + string _SplitHeadsQ__mh_w_571_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_571_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_571_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_571_equation_0, values = (var_5456_cast_fp16, var_5393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_571_cast_fp16")]; + string _SplitHeadsQ__mh_w_573_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_573_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_573_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_573_equation_0, values = (var_5456_cast_fp16, var_5400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_573_cast_fp16")]; + string _SplitHeadsQ__mh_w_575_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_575_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_575_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_575_equation_0, values = (var_5456_cast_fp16, var_5407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_575_cast_fp16")]; + fp16 var_5601_to_fp16 = const()[name = string("op_5601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_481_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_481_cast_fp16, y = var_5601_to_fp16)[name = string("aw_chunk_481_cast_fp16")]; + fp16 var_5603_to_fp16 = const()[name = string("op_5603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_483_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_483_cast_fp16, y = var_5603_to_fp16)[name = string("aw_chunk_483_cast_fp16")]; + fp16 var_5605_to_fp16 = const()[name = string("op_5605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_485_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_485_cast_fp16, y = var_5605_to_fp16)[name = string("aw_chunk_485_cast_fp16")]; + fp16 var_5607_to_fp16 = const()[name = string("op_5607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_487_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_487_cast_fp16, y = var_5607_to_fp16)[name = string("aw_chunk_487_cast_fp16")]; + fp16 var_5609_to_fp16 = const()[name = string("op_5609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_489_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_489_cast_fp16, y = var_5609_to_fp16)[name = string("aw_chunk_489_cast_fp16")]; + fp16 var_5611_to_fp16 = const()[name = string("op_5611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_491_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_491_cast_fp16, y = var_5611_to_fp16)[name = string("aw_chunk_491_cast_fp16")]; + fp16 var_5613_to_fp16 = const()[name = string("op_5613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_493_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_493_cast_fp16, y = var_5613_to_fp16)[name = string("aw_chunk_493_cast_fp16")]; + fp16 var_5615_to_fp16 = const()[name = string("op_5615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_495_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_495_cast_fp16, y = var_5615_to_fp16)[name = string("aw_chunk_495_cast_fp16")]; + fp16 var_5617_to_fp16 = const()[name = string("op_5617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_497_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_497_cast_fp16, y = var_5617_to_fp16)[name = string("aw_chunk_497_cast_fp16")]; + fp16 var_5619_to_fp16 = const()[name = string("op_5619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_499_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_499_cast_fp16, y = var_5619_to_fp16)[name = string("aw_chunk_499_cast_fp16")]; + fp16 var_5621_to_fp16 = const()[name = string("op_5621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_501_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_501_cast_fp16, y = var_5621_to_fp16)[name = string("aw_chunk_501_cast_fp16")]; + fp16 var_5623_to_fp16 = const()[name = string("op_5623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_503_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_503_cast_fp16, y = var_5623_to_fp16)[name = string("aw_chunk_503_cast_fp16")]; + fp16 var_5625_to_fp16 = const()[name = string("op_5625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_505_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_505_cast_fp16, y = var_5625_to_fp16)[name = string("aw_chunk_505_cast_fp16")]; + fp16 var_5627_to_fp16 = const()[name = string("op_5627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_507_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_507_cast_fp16, y = var_5627_to_fp16)[name = string("aw_chunk_507_cast_fp16")]; + fp16 var_5629_to_fp16 = const()[name = string("op_5629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_509_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_509_cast_fp16, y = var_5629_to_fp16)[name = string("aw_chunk_509_cast_fp16")]; + fp16 var_5631_to_fp16 = const()[name = string("op_5631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_511_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_511_cast_fp16, y = var_5631_to_fp16)[name = string("aw_chunk_511_cast_fp16")]; + fp16 var_5633_to_fp16 = const()[name = string("op_5633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_513_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_513_cast_fp16, y = var_5633_to_fp16)[name = string("aw_chunk_513_cast_fp16")]; + fp16 var_5635_to_fp16 = const()[name = string("op_5635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_515_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_515_cast_fp16, y = var_5635_to_fp16)[name = string("aw_chunk_515_cast_fp16")]; + fp16 var_5637_to_fp16 = const()[name = string("op_5637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_517_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_517_cast_fp16, y = var_5637_to_fp16)[name = string("aw_chunk_517_cast_fp16")]; + fp16 var_5639_to_fp16 = const()[name = string("op_5639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_519_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_519_cast_fp16, y = var_5639_to_fp16)[name = string("aw_chunk_519_cast_fp16")]; + fp16 var_5641_to_fp16 = const()[name = string("op_5641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_521_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_521_cast_fp16, y = var_5641_to_fp16)[name = string("aw_chunk_521_cast_fp16")]; + fp16 var_5643_to_fp16 = const()[name = string("op_5643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_523_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_523_cast_fp16, y = var_5643_to_fp16)[name = string("aw_chunk_523_cast_fp16")]; + fp16 var_5645_to_fp16 = const()[name = string("op_5645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_525_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_525_cast_fp16, y = var_5645_to_fp16)[name = string("aw_chunk_525_cast_fp16")]; + fp16 var_5647_to_fp16 = const()[name = string("op_5647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_527_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_527_cast_fp16, y = var_5647_to_fp16)[name = string("aw_chunk_527_cast_fp16")]; + fp16 var_5649_to_fp16 = const()[name = string("op_5649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_529_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_529_cast_fp16, y = var_5649_to_fp16)[name = string("aw_chunk_529_cast_fp16")]; + fp16 var_5651_to_fp16 = const()[name = string("op_5651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_531_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_531_cast_fp16, y = var_5651_to_fp16)[name = string("aw_chunk_531_cast_fp16")]; + fp16 var_5653_to_fp16 = const()[name = string("op_5653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_533_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_533_cast_fp16, y = var_5653_to_fp16)[name = string("aw_chunk_533_cast_fp16")]; + fp16 var_5655_to_fp16 = const()[name = string("op_5655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_535_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_535_cast_fp16, y = var_5655_to_fp16)[name = string("aw_chunk_535_cast_fp16")]; + fp16 var_5657_to_fp16 = const()[name = string("op_5657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_537_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_537_cast_fp16, y = var_5657_to_fp16)[name = string("aw_chunk_537_cast_fp16")]; + fp16 var_5659_to_fp16 = const()[name = string("op_5659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_539_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_539_cast_fp16, y = var_5659_to_fp16)[name = string("aw_chunk_539_cast_fp16")]; + fp16 var_5661_to_fp16 = const()[name = string("op_5661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_541_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_541_cast_fp16, y = var_5661_to_fp16)[name = string("aw_chunk_541_cast_fp16")]; + fp16 var_5663_to_fp16 = const()[name = string("op_5663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_543_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_543_cast_fp16, y = var_5663_to_fp16)[name = string("aw_chunk_543_cast_fp16")]; + fp16 var_5665_to_fp16 = const()[name = string("op_5665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_545_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_545_cast_fp16, y = var_5665_to_fp16)[name = string("aw_chunk_545_cast_fp16")]; + fp16 var_5667_to_fp16 = const()[name = string("op_5667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_547_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_547_cast_fp16, y = var_5667_to_fp16)[name = string("aw_chunk_547_cast_fp16")]; + fp16 var_5669_to_fp16 = const()[name = string("op_5669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_549_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_549_cast_fp16, y = var_5669_to_fp16)[name = string("aw_chunk_549_cast_fp16")]; + fp16 var_5671_to_fp16 = const()[name = string("op_5671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_551_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_551_cast_fp16, y = var_5671_to_fp16)[name = string("aw_chunk_551_cast_fp16")]; + fp16 var_5673_to_fp16 = const()[name = string("op_5673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_553_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_553_cast_fp16, y = var_5673_to_fp16)[name = string("aw_chunk_553_cast_fp16")]; + fp16 var_5675_to_fp16 = const()[name = string("op_5675_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_555_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_555_cast_fp16, y = var_5675_to_fp16)[name = string("aw_chunk_555_cast_fp16")]; + fp16 var_5677_to_fp16 = const()[name = string("op_5677_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_557_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_557_cast_fp16, y = var_5677_to_fp16)[name = string("aw_chunk_557_cast_fp16")]; + fp16 var_5679_to_fp16 = const()[name = string("op_5679_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_559_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_559_cast_fp16, y = var_5679_to_fp16)[name = string("aw_chunk_559_cast_fp16")]; + fp16 var_5681_to_fp16 = const()[name = string("op_5681_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_561_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_561_cast_fp16, y = var_5681_to_fp16)[name = string("aw_chunk_561_cast_fp16")]; + fp16 var_5683_to_fp16 = const()[name = string("op_5683_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_563_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_563_cast_fp16, y = var_5683_to_fp16)[name = string("aw_chunk_563_cast_fp16")]; + fp16 var_5685_to_fp16 = const()[name = string("op_5685_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_565_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_565_cast_fp16, y = var_5685_to_fp16)[name = string("aw_chunk_565_cast_fp16")]; + fp16 var_5687_to_fp16 = const()[name = string("op_5687_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_567_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_567_cast_fp16, y = var_5687_to_fp16)[name = string("aw_chunk_567_cast_fp16")]; + fp16 var_5689_to_fp16 = const()[name = string("op_5689_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_569_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_569_cast_fp16, y = var_5689_to_fp16)[name = string("aw_chunk_569_cast_fp16")]; + fp16 var_5691_to_fp16 = const()[name = string("op_5691_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_571_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_571_cast_fp16, y = var_5691_to_fp16)[name = string("aw_chunk_571_cast_fp16")]; + fp16 var_5693_to_fp16 = const()[name = string("op_5693_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_573_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_573_cast_fp16, y = var_5693_to_fp16)[name = string("aw_chunk_573_cast_fp16")]; + fp16 var_5695_to_fp16 = const()[name = string("op_5695_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_575_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_575_cast_fp16, y = var_5695_to_fp16)[name = string("aw_chunk_575_cast_fp16")]; + tensor var_5697_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_481_cast_fp16)[name = string("op_5697_cast_fp16")]; + tensor var_5698_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_483_cast_fp16)[name = string("op_5698_cast_fp16")]; + tensor var_5699_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_485_cast_fp16)[name = string("op_5699_cast_fp16")]; + tensor var_5700_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_487_cast_fp16)[name = string("op_5700_cast_fp16")]; + tensor var_5701_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_489_cast_fp16)[name = string("op_5701_cast_fp16")]; + tensor var_5702_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_491_cast_fp16)[name = string("op_5702_cast_fp16")]; + tensor var_5703_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_493_cast_fp16)[name = string("op_5703_cast_fp16")]; + tensor var_5704_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_495_cast_fp16)[name = string("op_5704_cast_fp16")]; + tensor var_5705_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_497_cast_fp16)[name = string("op_5705_cast_fp16")]; + tensor var_5706_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_499_cast_fp16)[name = string("op_5706_cast_fp16")]; + tensor var_5707_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_501_cast_fp16)[name = string("op_5707_cast_fp16")]; + tensor var_5708_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_503_cast_fp16)[name = string("op_5708_cast_fp16")]; + tensor var_5709_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_505_cast_fp16)[name = string("op_5709_cast_fp16")]; + tensor var_5710_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_507_cast_fp16)[name = string("op_5710_cast_fp16")]; + tensor var_5711_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_509_cast_fp16)[name = string("op_5711_cast_fp16")]; + tensor var_5712_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_511_cast_fp16)[name = string("op_5712_cast_fp16")]; + tensor var_5713_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_513_cast_fp16)[name = string("op_5713_cast_fp16")]; + tensor var_5714_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_515_cast_fp16)[name = string("op_5714_cast_fp16")]; + tensor var_5715_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_517_cast_fp16)[name = string("op_5715_cast_fp16")]; + tensor var_5716_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_519_cast_fp16)[name = string("op_5716_cast_fp16")]; + tensor var_5717_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_521_cast_fp16)[name = string("op_5717_cast_fp16")]; + tensor var_5718_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_523_cast_fp16)[name = string("op_5718_cast_fp16")]; + tensor var_5719_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_525_cast_fp16)[name = string("op_5719_cast_fp16")]; + tensor var_5720_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_527_cast_fp16)[name = string("op_5720_cast_fp16")]; + tensor var_5721_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_529_cast_fp16)[name = string("op_5721_cast_fp16")]; + tensor var_5722_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_531_cast_fp16)[name = string("op_5722_cast_fp16")]; + tensor var_5723_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_533_cast_fp16)[name = string("op_5723_cast_fp16")]; + tensor var_5724_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_535_cast_fp16)[name = string("op_5724_cast_fp16")]; + tensor var_5725_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_537_cast_fp16)[name = string("op_5725_cast_fp16")]; + tensor var_5726_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_539_cast_fp16)[name = string("op_5726_cast_fp16")]; + tensor var_5727_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_541_cast_fp16)[name = string("op_5727_cast_fp16")]; + tensor var_5728_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_543_cast_fp16)[name = string("op_5728_cast_fp16")]; + tensor var_5729_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_545_cast_fp16)[name = string("op_5729_cast_fp16")]; + tensor var_5730_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_547_cast_fp16)[name = string("op_5730_cast_fp16")]; + tensor var_5731_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_549_cast_fp16)[name = string("op_5731_cast_fp16")]; + tensor var_5732_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_551_cast_fp16)[name = string("op_5732_cast_fp16")]; + tensor var_5733_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_553_cast_fp16)[name = string("op_5733_cast_fp16")]; + tensor var_5734_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_555_cast_fp16)[name = string("op_5734_cast_fp16")]; + tensor var_5735_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_557_cast_fp16)[name = string("op_5735_cast_fp16")]; + tensor var_5736_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_559_cast_fp16)[name = string("op_5736_cast_fp16")]; + tensor var_5737_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_561_cast_fp16)[name = string("op_5737_cast_fp16")]; + tensor var_5738_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_563_cast_fp16)[name = string("op_5738_cast_fp16")]; + tensor var_5739_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_565_cast_fp16)[name = string("op_5739_cast_fp16")]; + tensor var_5740_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_567_cast_fp16)[name = string("op_5740_cast_fp16")]; + tensor var_5741_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_569_cast_fp16)[name = string("op_5741_cast_fp16")]; + tensor var_5742_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_571_cast_fp16)[name = string("op_5742_cast_fp16")]; + tensor var_5743_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_573_cast_fp16)[name = string("op_5743_cast_fp16")]; + tensor var_5744_cast_fp16 = softmax(axis = var_4970, x = aw_chunk_575_cast_fp16)[name = string("op_5744_cast_fp16")]; + string var_5746_equation_0 = const()[name = string("op_5746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5746_cast_fp16 = einsum(equation = var_5746_equation_0, values = (var_5458_cast_fp16, var_5697_cast_fp16))[name = string("op_5746_cast_fp16")]; + string var_5748_equation_0 = const()[name = string("op_5748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5748_cast_fp16 = einsum(equation = var_5748_equation_0, values = (var_5458_cast_fp16, var_5698_cast_fp16))[name = string("op_5748_cast_fp16")]; + string var_5750_equation_0 = const()[name = string("op_5750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5750_cast_fp16 = einsum(equation = var_5750_equation_0, values = (var_5458_cast_fp16, var_5699_cast_fp16))[name = string("op_5750_cast_fp16")]; + string var_5752_equation_0 = const()[name = string("op_5752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5752_cast_fp16 = einsum(equation = var_5752_equation_0, values = (var_5458_cast_fp16, var_5700_cast_fp16))[name = string("op_5752_cast_fp16")]; + string var_5754_equation_0 = const()[name = string("op_5754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5754_cast_fp16 = einsum(equation = var_5754_equation_0, values = (var_5462_cast_fp16, var_5701_cast_fp16))[name = string("op_5754_cast_fp16")]; + string var_5756_equation_0 = const()[name = string("op_5756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5756_cast_fp16 = einsum(equation = var_5756_equation_0, values = (var_5462_cast_fp16, var_5702_cast_fp16))[name = string("op_5756_cast_fp16")]; + string var_5758_equation_0 = const()[name = string("op_5758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5758_cast_fp16 = einsum(equation = var_5758_equation_0, values = (var_5462_cast_fp16, var_5703_cast_fp16))[name = string("op_5758_cast_fp16")]; + string var_5760_equation_0 = const()[name = string("op_5760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5760_cast_fp16 = einsum(equation = var_5760_equation_0, values = (var_5462_cast_fp16, var_5704_cast_fp16))[name = string("op_5760_cast_fp16")]; + string var_5762_equation_0 = const()[name = string("op_5762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5762_cast_fp16 = einsum(equation = var_5762_equation_0, values = (var_5466_cast_fp16, var_5705_cast_fp16))[name = string("op_5762_cast_fp16")]; + string var_5764_equation_0 = const()[name = string("op_5764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5764_cast_fp16 = einsum(equation = var_5764_equation_0, values = (var_5466_cast_fp16, var_5706_cast_fp16))[name = string("op_5764_cast_fp16")]; + string var_5766_equation_0 = const()[name = string("op_5766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5766_cast_fp16 = einsum(equation = var_5766_equation_0, values = (var_5466_cast_fp16, var_5707_cast_fp16))[name = string("op_5766_cast_fp16")]; + string var_5768_equation_0 = const()[name = string("op_5768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5768_cast_fp16 = einsum(equation = var_5768_equation_0, values = (var_5466_cast_fp16, var_5708_cast_fp16))[name = string("op_5768_cast_fp16")]; + string var_5770_equation_0 = const()[name = string("op_5770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5770_cast_fp16 = einsum(equation = var_5770_equation_0, values = (var_5470_cast_fp16, var_5709_cast_fp16))[name = string("op_5770_cast_fp16")]; + string var_5772_equation_0 = const()[name = string("op_5772_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5772_cast_fp16 = einsum(equation = var_5772_equation_0, values = (var_5470_cast_fp16, var_5710_cast_fp16))[name = string("op_5772_cast_fp16")]; + string var_5774_equation_0 = const()[name = string("op_5774_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5774_cast_fp16 = einsum(equation = var_5774_equation_0, values = (var_5470_cast_fp16, var_5711_cast_fp16))[name = string("op_5774_cast_fp16")]; + string var_5776_equation_0 = const()[name = string("op_5776_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5776_cast_fp16 = einsum(equation = var_5776_equation_0, values = (var_5470_cast_fp16, var_5712_cast_fp16))[name = string("op_5776_cast_fp16")]; + string var_5778_equation_0 = const()[name = string("op_5778_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5778_cast_fp16 = einsum(equation = var_5778_equation_0, values = (var_5474_cast_fp16, var_5713_cast_fp16))[name = string("op_5778_cast_fp16")]; + string var_5780_equation_0 = const()[name = string("op_5780_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5780_cast_fp16 = einsum(equation = var_5780_equation_0, values = (var_5474_cast_fp16, var_5714_cast_fp16))[name = string("op_5780_cast_fp16")]; + string var_5782_equation_0 = const()[name = string("op_5782_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5782_cast_fp16 = einsum(equation = var_5782_equation_0, values = (var_5474_cast_fp16, var_5715_cast_fp16))[name = string("op_5782_cast_fp16")]; + string var_5784_equation_0 = const()[name = string("op_5784_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5784_cast_fp16 = einsum(equation = var_5784_equation_0, values = (var_5474_cast_fp16, var_5716_cast_fp16))[name = string("op_5784_cast_fp16")]; + string var_5786_equation_0 = const()[name = string("op_5786_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5786_cast_fp16 = einsum(equation = var_5786_equation_0, values = (var_5478_cast_fp16, var_5717_cast_fp16))[name = string("op_5786_cast_fp16")]; + string var_5788_equation_0 = const()[name = string("op_5788_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5788_cast_fp16 = einsum(equation = var_5788_equation_0, values = (var_5478_cast_fp16, var_5718_cast_fp16))[name = string("op_5788_cast_fp16")]; + string var_5790_equation_0 = const()[name = string("op_5790_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5790_cast_fp16 = einsum(equation = var_5790_equation_0, values = (var_5478_cast_fp16, var_5719_cast_fp16))[name = string("op_5790_cast_fp16")]; + string var_5792_equation_0 = const()[name = string("op_5792_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5792_cast_fp16 = einsum(equation = var_5792_equation_0, values = (var_5478_cast_fp16, var_5720_cast_fp16))[name = string("op_5792_cast_fp16")]; + string var_5794_equation_0 = const()[name = string("op_5794_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5794_cast_fp16 = einsum(equation = var_5794_equation_0, values = (var_5482_cast_fp16, var_5721_cast_fp16))[name = string("op_5794_cast_fp16")]; + string var_5796_equation_0 = const()[name = string("op_5796_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5796_cast_fp16 = einsum(equation = var_5796_equation_0, values = (var_5482_cast_fp16, var_5722_cast_fp16))[name = string("op_5796_cast_fp16")]; + string var_5798_equation_0 = const()[name = string("op_5798_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5798_cast_fp16 = einsum(equation = var_5798_equation_0, values = (var_5482_cast_fp16, var_5723_cast_fp16))[name = string("op_5798_cast_fp16")]; + string var_5800_equation_0 = const()[name = string("op_5800_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5800_cast_fp16 = einsum(equation = var_5800_equation_0, values = (var_5482_cast_fp16, var_5724_cast_fp16))[name = string("op_5800_cast_fp16")]; + string var_5802_equation_0 = const()[name = string("op_5802_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5802_cast_fp16 = einsum(equation = var_5802_equation_0, values = (var_5486_cast_fp16, var_5725_cast_fp16))[name = string("op_5802_cast_fp16")]; + string var_5804_equation_0 = const()[name = string("op_5804_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5804_cast_fp16 = einsum(equation = var_5804_equation_0, values = (var_5486_cast_fp16, var_5726_cast_fp16))[name = string("op_5804_cast_fp16")]; + string var_5806_equation_0 = const()[name = string("op_5806_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5806_cast_fp16 = einsum(equation = var_5806_equation_0, values = (var_5486_cast_fp16, var_5727_cast_fp16))[name = string("op_5806_cast_fp16")]; + string var_5808_equation_0 = const()[name = string("op_5808_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5808_cast_fp16 = einsum(equation = var_5808_equation_0, values = (var_5486_cast_fp16, var_5728_cast_fp16))[name = string("op_5808_cast_fp16")]; + string var_5810_equation_0 = const()[name = string("op_5810_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5810_cast_fp16 = einsum(equation = var_5810_equation_0, values = (var_5490_cast_fp16, var_5729_cast_fp16))[name = string("op_5810_cast_fp16")]; + string var_5812_equation_0 = const()[name = string("op_5812_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5812_cast_fp16 = einsum(equation = var_5812_equation_0, values = (var_5490_cast_fp16, var_5730_cast_fp16))[name = string("op_5812_cast_fp16")]; + string var_5814_equation_0 = const()[name = string("op_5814_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5814_cast_fp16 = einsum(equation = var_5814_equation_0, values = (var_5490_cast_fp16, var_5731_cast_fp16))[name = string("op_5814_cast_fp16")]; + string var_5816_equation_0 = const()[name = string("op_5816_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5816_cast_fp16 = einsum(equation = var_5816_equation_0, values = (var_5490_cast_fp16, var_5732_cast_fp16))[name = string("op_5816_cast_fp16")]; + string var_5818_equation_0 = const()[name = string("op_5818_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5818_cast_fp16 = einsum(equation = var_5818_equation_0, values = (var_5494_cast_fp16, var_5733_cast_fp16))[name = string("op_5818_cast_fp16")]; + string var_5820_equation_0 = const()[name = string("op_5820_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5820_cast_fp16 = einsum(equation = var_5820_equation_0, values = (var_5494_cast_fp16, var_5734_cast_fp16))[name = string("op_5820_cast_fp16")]; + string var_5822_equation_0 = const()[name = string("op_5822_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5822_cast_fp16 = einsum(equation = var_5822_equation_0, values = (var_5494_cast_fp16, var_5735_cast_fp16))[name = string("op_5822_cast_fp16")]; + string var_5824_equation_0 = const()[name = string("op_5824_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5824_cast_fp16 = einsum(equation = var_5824_equation_0, values = (var_5494_cast_fp16, var_5736_cast_fp16))[name = string("op_5824_cast_fp16")]; + string var_5826_equation_0 = const()[name = string("op_5826_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5826_cast_fp16 = einsum(equation = var_5826_equation_0, values = (var_5498_cast_fp16, var_5737_cast_fp16))[name = string("op_5826_cast_fp16")]; + string var_5828_equation_0 = const()[name = string("op_5828_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5828_cast_fp16 = einsum(equation = var_5828_equation_0, values = (var_5498_cast_fp16, var_5738_cast_fp16))[name = string("op_5828_cast_fp16")]; + string var_5830_equation_0 = const()[name = string("op_5830_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5830_cast_fp16 = einsum(equation = var_5830_equation_0, values = (var_5498_cast_fp16, var_5739_cast_fp16))[name = string("op_5830_cast_fp16")]; + string var_5832_equation_0 = const()[name = string("op_5832_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5832_cast_fp16 = einsum(equation = var_5832_equation_0, values = (var_5498_cast_fp16, var_5740_cast_fp16))[name = string("op_5832_cast_fp16")]; + string var_5834_equation_0 = const()[name = string("op_5834_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5834_cast_fp16 = einsum(equation = var_5834_equation_0, values = (var_5502_cast_fp16, var_5741_cast_fp16))[name = string("op_5834_cast_fp16")]; + string var_5836_equation_0 = const()[name = string("op_5836_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5836_cast_fp16 = einsum(equation = var_5836_equation_0, values = (var_5502_cast_fp16, var_5742_cast_fp16))[name = string("op_5836_cast_fp16")]; + string var_5838_equation_0 = const()[name = string("op_5838_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5838_cast_fp16 = einsum(equation = var_5838_equation_0, values = (var_5502_cast_fp16, var_5743_cast_fp16))[name = string("op_5838_cast_fp16")]; + string var_5840_equation_0 = const()[name = string("op_5840_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_5840_cast_fp16 = einsum(equation = var_5840_equation_0, values = (var_5502_cast_fp16, var_5744_cast_fp16))[name = string("op_5840_cast_fp16")]; + bool var_5842_interleave_0 = const()[name = string("op_5842_interleave_0"), val = bool(false)]; + tensor var_5842_cast_fp16 = concat(axis = var_4953, interleave = var_5842_interleave_0, values = (var_5746_cast_fp16, var_5748_cast_fp16, var_5750_cast_fp16, var_5752_cast_fp16))[name = string("op_5842_cast_fp16")]; + bool var_5844_interleave_0 = const()[name = string("op_5844_interleave_0"), val = bool(false)]; + tensor var_5844_cast_fp16 = concat(axis = var_4953, interleave = var_5844_interleave_0, values = (var_5754_cast_fp16, var_5756_cast_fp16, var_5758_cast_fp16, var_5760_cast_fp16))[name = string("op_5844_cast_fp16")]; + bool var_5846_interleave_0 = const()[name = string("op_5846_interleave_0"), val = bool(false)]; + tensor var_5846_cast_fp16 = concat(axis = var_4953, interleave = var_5846_interleave_0, values = (var_5762_cast_fp16, var_5764_cast_fp16, var_5766_cast_fp16, var_5768_cast_fp16))[name = string("op_5846_cast_fp16")]; + bool var_5848_interleave_0 = const()[name = string("op_5848_interleave_0"), val = bool(false)]; + tensor var_5848_cast_fp16 = concat(axis = var_4953, interleave = var_5848_interleave_0, values = (var_5770_cast_fp16, var_5772_cast_fp16, var_5774_cast_fp16, var_5776_cast_fp16))[name = string("op_5848_cast_fp16")]; + bool var_5850_interleave_0 = const()[name = string("op_5850_interleave_0"), val = bool(false)]; + tensor var_5850_cast_fp16 = concat(axis = var_4953, interleave = var_5850_interleave_0, values = (var_5778_cast_fp16, var_5780_cast_fp16, var_5782_cast_fp16, var_5784_cast_fp16))[name = string("op_5850_cast_fp16")]; + bool var_5852_interleave_0 = const()[name = string("op_5852_interleave_0"), val = bool(false)]; + tensor var_5852_cast_fp16 = concat(axis = var_4953, interleave = var_5852_interleave_0, values = (var_5786_cast_fp16, var_5788_cast_fp16, var_5790_cast_fp16, var_5792_cast_fp16))[name = string("op_5852_cast_fp16")]; + bool var_5854_interleave_0 = const()[name = string("op_5854_interleave_0"), val = bool(false)]; + tensor var_5854_cast_fp16 = concat(axis = var_4953, interleave = var_5854_interleave_0, values = (var_5794_cast_fp16, var_5796_cast_fp16, var_5798_cast_fp16, var_5800_cast_fp16))[name = string("op_5854_cast_fp16")]; + bool var_5856_interleave_0 = const()[name = string("op_5856_interleave_0"), val = bool(false)]; + tensor var_5856_cast_fp16 = concat(axis = var_4953, interleave = var_5856_interleave_0, values = (var_5802_cast_fp16, var_5804_cast_fp16, var_5806_cast_fp16, var_5808_cast_fp16))[name = string("op_5856_cast_fp16")]; + bool var_5858_interleave_0 = const()[name = string("op_5858_interleave_0"), val = bool(false)]; + tensor var_5858_cast_fp16 = concat(axis = var_4953, interleave = var_5858_interleave_0, values = (var_5810_cast_fp16, var_5812_cast_fp16, var_5814_cast_fp16, var_5816_cast_fp16))[name = string("op_5858_cast_fp16")]; + bool var_5860_interleave_0 = const()[name = string("op_5860_interleave_0"), val = bool(false)]; + tensor var_5860_cast_fp16 = concat(axis = var_4953, interleave = var_5860_interleave_0, values = (var_5818_cast_fp16, var_5820_cast_fp16, var_5822_cast_fp16, var_5824_cast_fp16))[name = string("op_5860_cast_fp16")]; + bool var_5862_interleave_0 = const()[name = string("op_5862_interleave_0"), val = bool(false)]; + tensor var_5862_cast_fp16 = concat(axis = var_4953, interleave = var_5862_interleave_0, values = (var_5826_cast_fp16, var_5828_cast_fp16, var_5830_cast_fp16, var_5832_cast_fp16))[name = string("op_5862_cast_fp16")]; + bool var_5864_interleave_0 = const()[name = string("op_5864_interleave_0"), val = bool(false)]; + tensor var_5864_cast_fp16 = concat(axis = var_4953, interleave = var_5864_interleave_0, values = (var_5834_cast_fp16, var_5836_cast_fp16, var_5838_cast_fp16, var_5840_cast_fp16))[name = string("op_5864_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_4970, interleave = input_41_interleave_0, values = (var_5842_cast_fp16, var_5844_cast_fp16, var_5846_cast_fp16, var_5848_cast_fp16, var_5850_cast_fp16, var_5852_cast_fp16, var_5854_cast_fp16, var_5856_cast_fp16, var_5858_cast_fp16, var_5860_cast_fp16, var_5862_cast_fp16, var_5864_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_23_pad_type_0 = const()[name = string("obj_23_pad_type_0"), val = string("valid")]; + tensor obj_23_strides_0 = const()[name = string("obj_23_strides_0"), val = tensor([1, 1])]; + tensor obj_23_pad_0 = const()[name = string("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_23_dilations_0 = const()[name = string("obj_23_dilations_0"), val = tensor([1, 1])]; + int32 obj_23_groups_0 = const()[name = string("obj_23_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80639616)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81819328)))]; + tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_23_dilations_0, groups = obj_23_groups_0, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = obj_23_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_5883_to_fp16 = const()[name = string("op_5883_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_5883_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81820928)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81822528)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81824128)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86542784)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86548992)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91267648)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_25_cast_fp16")]; + int32 var_5912 = const()[name = string("op_5912"), val = int32(3)]; + int32 var_5929 = const()[name = string("op_5929"), val = int32(1)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_5946_to_fp16 = const()[name = string("op_5946_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_5946_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91269248)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91270848)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91272448)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92452160)))]; + tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_13_cast_fp16")]; + string key_13_pad_type_0 = const()[name = string("key_13_pad_type_0"), val = string("valid")]; + tensor key_13_strides_0 = const()[name = string("key_13_strides_0"), val = tensor([1, 1])]; + tensor key_13_pad_0 = const()[name = string("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_13_dilations_0 = const()[name = string("key_13_dilations_0"), val = tensor([1, 1])]; + int32 key_13_groups_0 = const()[name = string("key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92453760)))]; + tensor key_13_cast_fp16 = conv(dilations = key_13_dilations_0, groups = key_13_groups_0, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("key_13_cast_fp16")]; + string value_13_pad_type_0 = const()[name = string("value_13_pad_type_0"), val = string("valid")]; + tensor value_13_strides_0 = const()[name = string("value_13_strides_0"), val = tensor([1, 1])]; + tensor value_13_pad_0 = const()[name = string("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_13_dilations_0 = const()[name = string("value_13_dilations_0"), val = tensor([1, 1])]; + int32 value_13_groups_0 = const()[name = string("value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93633472)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94813184)))]; + tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = value_13_dilations_0, groups = value_13_groups_0, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_5984_begin_0 = const()[name = string("op_5984_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5984_end_0 = const()[name = string("op_5984_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5984_end_mask_0 = const()[name = string("op_5984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5984_cast_fp16 = slice_by_index(begin = var_5984_begin_0, end = var_5984_end_0, end_mask = var_5984_end_mask_0, x = query_13_cast_fp16)[name = string("op_5984_cast_fp16")]; + tensor var_5988_begin_0 = const()[name = string("op_5988_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5988_end_0 = const()[name = string("op_5988_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5988_end_mask_0 = const()[name = string("op_5988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5988_cast_fp16 = slice_by_index(begin = var_5988_begin_0, end = var_5988_end_0, end_mask = var_5988_end_mask_0, x = query_13_cast_fp16)[name = string("op_5988_cast_fp16")]; + tensor var_5992_begin_0 = const()[name = string("op_5992_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5992_end_0 = const()[name = string("op_5992_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5992_end_mask_0 = const()[name = string("op_5992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5992_cast_fp16 = slice_by_index(begin = var_5992_begin_0, end = var_5992_end_0, end_mask = var_5992_end_mask_0, x = query_13_cast_fp16)[name = string("op_5992_cast_fp16")]; + tensor var_5996_begin_0 = const()[name = string("op_5996_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5996_end_0 = const()[name = string("op_5996_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5996_end_mask_0 = const()[name = string("op_5996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5996_cast_fp16 = slice_by_index(begin = var_5996_begin_0, end = var_5996_end_0, end_mask = var_5996_end_mask_0, x = query_13_cast_fp16)[name = string("op_5996_cast_fp16")]; + tensor var_6000_begin_0 = const()[name = string("op_6000_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6000_end_0 = const()[name = string("op_6000_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6000_end_mask_0 = const()[name = string("op_6000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6000_cast_fp16 = slice_by_index(begin = var_6000_begin_0, end = var_6000_end_0, end_mask = var_6000_end_mask_0, x = query_13_cast_fp16)[name = string("op_6000_cast_fp16")]; + tensor var_6004_begin_0 = const()[name = string("op_6004_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6004_end_0 = const()[name = string("op_6004_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6004_end_mask_0 = const()[name = string("op_6004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6004_cast_fp16 = slice_by_index(begin = var_6004_begin_0, end = var_6004_end_0, end_mask = var_6004_end_mask_0, x = query_13_cast_fp16)[name = string("op_6004_cast_fp16")]; + tensor var_6008_begin_0 = const()[name = string("op_6008_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6008_end_0 = const()[name = string("op_6008_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6008_end_mask_0 = const()[name = string("op_6008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6008_cast_fp16 = slice_by_index(begin = var_6008_begin_0, end = var_6008_end_0, end_mask = var_6008_end_mask_0, x = query_13_cast_fp16)[name = string("op_6008_cast_fp16")]; + tensor var_6012_begin_0 = const()[name = string("op_6012_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6012_end_0 = const()[name = string("op_6012_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6012_end_mask_0 = const()[name = string("op_6012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6012_cast_fp16 = slice_by_index(begin = var_6012_begin_0, end = var_6012_end_0, end_mask = var_6012_end_mask_0, x = query_13_cast_fp16)[name = string("op_6012_cast_fp16")]; + tensor var_6016_begin_0 = const()[name = string("op_6016_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6016_end_0 = const()[name = string("op_6016_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6016_end_mask_0 = const()[name = string("op_6016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6016_cast_fp16 = slice_by_index(begin = var_6016_begin_0, end = var_6016_end_0, end_mask = var_6016_end_mask_0, x = query_13_cast_fp16)[name = string("op_6016_cast_fp16")]; + tensor var_6020_begin_0 = const()[name = string("op_6020_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6020_end_0 = const()[name = string("op_6020_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6020_end_mask_0 = const()[name = string("op_6020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6020_cast_fp16 = slice_by_index(begin = var_6020_begin_0, end = var_6020_end_0, end_mask = var_6020_end_mask_0, x = query_13_cast_fp16)[name = string("op_6020_cast_fp16")]; + tensor var_6024_begin_0 = const()[name = string("op_6024_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6024_end_0 = const()[name = string("op_6024_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6024_end_mask_0 = const()[name = string("op_6024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6024_cast_fp16 = slice_by_index(begin = var_6024_begin_0, end = var_6024_end_0, end_mask = var_6024_end_mask_0, x = query_13_cast_fp16)[name = string("op_6024_cast_fp16")]; + tensor var_6028_begin_0 = const()[name = string("op_6028_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6028_end_0 = const()[name = string("op_6028_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6028_end_mask_0 = const()[name = string("op_6028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6028_cast_fp16 = slice_by_index(begin = var_6028_begin_0, end = var_6028_end_0, end_mask = var_6028_end_mask_0, x = query_13_cast_fp16)[name = string("op_6028_cast_fp16")]; + tensor var_6037_begin_0 = const()[name = string("op_6037_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6037_end_0 = const()[name = string("op_6037_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6037_end_mask_0 = const()[name = string("op_6037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6037_cast_fp16 = slice_by_index(begin = var_6037_begin_0, end = var_6037_end_0, end_mask = var_6037_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6037_cast_fp16")]; + tensor var_6044_begin_0 = const()[name = string("op_6044_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6044_end_0 = const()[name = string("op_6044_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6044_end_mask_0 = const()[name = string("op_6044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6044_cast_fp16 = slice_by_index(begin = var_6044_begin_0, end = var_6044_end_0, end_mask = var_6044_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6044_cast_fp16")]; + tensor var_6051_begin_0 = const()[name = string("op_6051_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6051_end_0 = const()[name = string("op_6051_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6051_end_mask_0 = const()[name = string("op_6051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6051_cast_fp16 = slice_by_index(begin = var_6051_begin_0, end = var_6051_end_0, end_mask = var_6051_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6051_cast_fp16")]; + tensor var_6058_begin_0 = const()[name = string("op_6058_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6058_end_0 = const()[name = string("op_6058_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6058_end_mask_0 = const()[name = string("op_6058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6058_cast_fp16 = slice_by_index(begin = var_6058_begin_0, end = var_6058_end_0, end_mask = var_6058_end_mask_0, x = var_5984_cast_fp16)[name = string("op_6058_cast_fp16")]; + tensor var_6065_begin_0 = const()[name = string("op_6065_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6065_end_0 = const()[name = string("op_6065_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6065_end_mask_0 = const()[name = string("op_6065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6065_cast_fp16 = slice_by_index(begin = var_6065_begin_0, end = var_6065_end_0, end_mask = var_6065_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6065_cast_fp16")]; + tensor var_6072_begin_0 = const()[name = string("op_6072_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6072_end_0 = const()[name = string("op_6072_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6072_end_mask_0 = const()[name = string("op_6072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6072_cast_fp16 = slice_by_index(begin = var_6072_begin_0, end = var_6072_end_0, end_mask = var_6072_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6072_cast_fp16")]; + tensor var_6079_begin_0 = const()[name = string("op_6079_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6079_end_0 = const()[name = string("op_6079_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6079_end_mask_0 = const()[name = string("op_6079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6079_cast_fp16 = slice_by_index(begin = var_6079_begin_0, end = var_6079_end_0, end_mask = var_6079_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6079_cast_fp16")]; + tensor var_6086_begin_0 = const()[name = string("op_6086_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6086_end_0 = const()[name = string("op_6086_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6086_end_mask_0 = const()[name = string("op_6086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6086_cast_fp16 = slice_by_index(begin = var_6086_begin_0, end = var_6086_end_0, end_mask = var_6086_end_mask_0, x = var_5988_cast_fp16)[name = string("op_6086_cast_fp16")]; + tensor var_6093_begin_0 = const()[name = string("op_6093_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6093_end_0 = const()[name = string("op_6093_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6093_end_mask_0 = const()[name = string("op_6093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6093_cast_fp16 = slice_by_index(begin = var_6093_begin_0, end = var_6093_end_0, end_mask = var_6093_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6093_cast_fp16")]; + tensor var_6100_begin_0 = const()[name = string("op_6100_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6100_end_0 = const()[name = string("op_6100_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6100_end_mask_0 = const()[name = string("op_6100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6100_cast_fp16 = slice_by_index(begin = var_6100_begin_0, end = var_6100_end_0, end_mask = var_6100_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6100_cast_fp16")]; + tensor var_6107_begin_0 = const()[name = string("op_6107_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6107_end_0 = const()[name = string("op_6107_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6107_end_mask_0 = const()[name = string("op_6107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6107_cast_fp16 = slice_by_index(begin = var_6107_begin_0, end = var_6107_end_0, end_mask = var_6107_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6107_cast_fp16")]; + tensor var_6114_begin_0 = const()[name = string("op_6114_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6114_end_0 = const()[name = string("op_6114_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6114_end_mask_0 = const()[name = string("op_6114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6114_cast_fp16 = slice_by_index(begin = var_6114_begin_0, end = var_6114_end_0, end_mask = var_6114_end_mask_0, x = var_5992_cast_fp16)[name = string("op_6114_cast_fp16")]; + tensor var_6121_begin_0 = const()[name = string("op_6121_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6121_end_0 = const()[name = string("op_6121_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6121_end_mask_0 = const()[name = string("op_6121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6121_cast_fp16 = slice_by_index(begin = var_6121_begin_0, end = var_6121_end_0, end_mask = var_6121_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6121_cast_fp16")]; + tensor var_6128_begin_0 = const()[name = string("op_6128_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6128_end_0 = const()[name = string("op_6128_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6128_end_mask_0 = const()[name = string("op_6128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6128_cast_fp16 = slice_by_index(begin = var_6128_begin_0, end = var_6128_end_0, end_mask = var_6128_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6128_cast_fp16")]; + tensor var_6135_begin_0 = const()[name = string("op_6135_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6135_end_0 = const()[name = string("op_6135_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6135_end_mask_0 = const()[name = string("op_6135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6135_cast_fp16 = slice_by_index(begin = var_6135_begin_0, end = var_6135_end_0, end_mask = var_6135_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6135_cast_fp16")]; + tensor var_6142_begin_0 = const()[name = string("op_6142_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6142_end_0 = const()[name = string("op_6142_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6142_end_mask_0 = const()[name = string("op_6142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6142_cast_fp16 = slice_by_index(begin = var_6142_begin_0, end = var_6142_end_0, end_mask = var_6142_end_mask_0, x = var_5996_cast_fp16)[name = string("op_6142_cast_fp16")]; + tensor var_6149_begin_0 = const()[name = string("op_6149_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6149_end_0 = const()[name = string("op_6149_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6149_end_mask_0 = const()[name = string("op_6149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6149_cast_fp16 = slice_by_index(begin = var_6149_begin_0, end = var_6149_end_0, end_mask = var_6149_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6149_cast_fp16")]; + tensor var_6156_begin_0 = const()[name = string("op_6156_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6156_end_0 = const()[name = string("op_6156_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6156_end_mask_0 = const()[name = string("op_6156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6156_cast_fp16 = slice_by_index(begin = var_6156_begin_0, end = var_6156_end_0, end_mask = var_6156_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6156_cast_fp16")]; + tensor var_6163_begin_0 = const()[name = string("op_6163_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6163_end_0 = const()[name = string("op_6163_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6163_end_mask_0 = const()[name = string("op_6163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6163_cast_fp16 = slice_by_index(begin = var_6163_begin_0, end = var_6163_end_0, end_mask = var_6163_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6163_cast_fp16")]; + tensor var_6170_begin_0 = const()[name = string("op_6170_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6170_end_0 = const()[name = string("op_6170_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6170_end_mask_0 = const()[name = string("op_6170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6170_cast_fp16 = slice_by_index(begin = var_6170_begin_0, end = var_6170_end_0, end_mask = var_6170_end_mask_0, x = var_6000_cast_fp16)[name = string("op_6170_cast_fp16")]; + tensor var_6177_begin_0 = const()[name = string("op_6177_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6177_end_0 = const()[name = string("op_6177_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6177_end_mask_0 = const()[name = string("op_6177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6177_cast_fp16 = slice_by_index(begin = var_6177_begin_0, end = var_6177_end_0, end_mask = var_6177_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6177_cast_fp16")]; + tensor var_6184_begin_0 = const()[name = string("op_6184_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6184_end_0 = const()[name = string("op_6184_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6184_end_mask_0 = const()[name = string("op_6184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6184_cast_fp16 = slice_by_index(begin = var_6184_begin_0, end = var_6184_end_0, end_mask = var_6184_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6184_cast_fp16")]; + tensor var_6191_begin_0 = const()[name = string("op_6191_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6191_end_0 = const()[name = string("op_6191_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6191_end_mask_0 = const()[name = string("op_6191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6191_cast_fp16 = slice_by_index(begin = var_6191_begin_0, end = var_6191_end_0, end_mask = var_6191_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6191_cast_fp16")]; + tensor var_6198_begin_0 = const()[name = string("op_6198_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6198_end_0 = const()[name = string("op_6198_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6198_end_mask_0 = const()[name = string("op_6198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6198_cast_fp16 = slice_by_index(begin = var_6198_begin_0, end = var_6198_end_0, end_mask = var_6198_end_mask_0, x = var_6004_cast_fp16)[name = string("op_6198_cast_fp16")]; + tensor var_6205_begin_0 = const()[name = string("op_6205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6205_end_0 = const()[name = string("op_6205_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6205_end_mask_0 = const()[name = string("op_6205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6205_cast_fp16 = slice_by_index(begin = var_6205_begin_0, end = var_6205_end_0, end_mask = var_6205_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6205_cast_fp16")]; + tensor var_6212_begin_0 = const()[name = string("op_6212_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6212_end_0 = const()[name = string("op_6212_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6212_end_mask_0 = const()[name = string("op_6212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6212_cast_fp16 = slice_by_index(begin = var_6212_begin_0, end = var_6212_end_0, end_mask = var_6212_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6212_cast_fp16")]; + tensor var_6219_begin_0 = const()[name = string("op_6219_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6219_end_0 = const()[name = string("op_6219_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6219_end_mask_0 = const()[name = string("op_6219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6219_cast_fp16 = slice_by_index(begin = var_6219_begin_0, end = var_6219_end_0, end_mask = var_6219_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6219_cast_fp16")]; + tensor var_6226_begin_0 = const()[name = string("op_6226_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6226_end_0 = const()[name = string("op_6226_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6226_end_mask_0 = const()[name = string("op_6226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6226_cast_fp16 = slice_by_index(begin = var_6226_begin_0, end = var_6226_end_0, end_mask = var_6226_end_mask_0, x = var_6008_cast_fp16)[name = string("op_6226_cast_fp16")]; + tensor var_6233_begin_0 = const()[name = string("op_6233_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6233_end_0 = const()[name = string("op_6233_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6233_end_mask_0 = const()[name = string("op_6233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6233_cast_fp16 = slice_by_index(begin = var_6233_begin_0, end = var_6233_end_0, end_mask = var_6233_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6233_cast_fp16")]; + tensor var_6240_begin_0 = const()[name = string("op_6240_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6240_end_0 = const()[name = string("op_6240_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6240_end_mask_0 = const()[name = string("op_6240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6240_cast_fp16 = slice_by_index(begin = var_6240_begin_0, end = var_6240_end_0, end_mask = var_6240_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6240_cast_fp16")]; + tensor var_6247_begin_0 = const()[name = string("op_6247_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6247_end_0 = const()[name = string("op_6247_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6247_end_mask_0 = const()[name = string("op_6247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6247_cast_fp16 = slice_by_index(begin = var_6247_begin_0, end = var_6247_end_0, end_mask = var_6247_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6247_cast_fp16")]; + tensor var_6254_begin_0 = const()[name = string("op_6254_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6254_end_0 = const()[name = string("op_6254_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6254_end_mask_0 = const()[name = string("op_6254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6254_cast_fp16 = slice_by_index(begin = var_6254_begin_0, end = var_6254_end_0, end_mask = var_6254_end_mask_0, x = var_6012_cast_fp16)[name = string("op_6254_cast_fp16")]; + tensor var_6261_begin_0 = const()[name = string("op_6261_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6261_end_0 = const()[name = string("op_6261_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6261_end_mask_0 = const()[name = string("op_6261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6261_cast_fp16 = slice_by_index(begin = var_6261_begin_0, end = var_6261_end_0, end_mask = var_6261_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6261_cast_fp16")]; + tensor var_6268_begin_0 = const()[name = string("op_6268_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6268_end_0 = const()[name = string("op_6268_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6268_end_mask_0 = const()[name = string("op_6268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6268_cast_fp16 = slice_by_index(begin = var_6268_begin_0, end = var_6268_end_0, end_mask = var_6268_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6268_cast_fp16")]; + tensor var_6275_begin_0 = const()[name = string("op_6275_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6275_end_0 = const()[name = string("op_6275_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6275_end_mask_0 = const()[name = string("op_6275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6275_cast_fp16 = slice_by_index(begin = var_6275_begin_0, end = var_6275_end_0, end_mask = var_6275_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6275_cast_fp16")]; + tensor var_6282_begin_0 = const()[name = string("op_6282_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6282_end_0 = const()[name = string("op_6282_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6282_end_mask_0 = const()[name = string("op_6282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6282_cast_fp16 = slice_by_index(begin = var_6282_begin_0, end = var_6282_end_0, end_mask = var_6282_end_mask_0, x = var_6016_cast_fp16)[name = string("op_6282_cast_fp16")]; + tensor var_6289_begin_0 = const()[name = string("op_6289_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6289_end_0 = const()[name = string("op_6289_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6289_end_mask_0 = const()[name = string("op_6289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6289_cast_fp16 = slice_by_index(begin = var_6289_begin_0, end = var_6289_end_0, end_mask = var_6289_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6289_cast_fp16")]; + tensor var_6296_begin_0 = const()[name = string("op_6296_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6296_end_0 = const()[name = string("op_6296_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6296_end_mask_0 = const()[name = string("op_6296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6296_cast_fp16 = slice_by_index(begin = var_6296_begin_0, end = var_6296_end_0, end_mask = var_6296_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6296_cast_fp16")]; + tensor var_6303_begin_0 = const()[name = string("op_6303_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6303_end_0 = const()[name = string("op_6303_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6303_end_mask_0 = const()[name = string("op_6303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6303_cast_fp16 = slice_by_index(begin = var_6303_begin_0, end = var_6303_end_0, end_mask = var_6303_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6303_cast_fp16")]; + tensor var_6310_begin_0 = const()[name = string("op_6310_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6310_end_0 = const()[name = string("op_6310_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6310_end_mask_0 = const()[name = string("op_6310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6310_cast_fp16 = slice_by_index(begin = var_6310_begin_0, end = var_6310_end_0, end_mask = var_6310_end_mask_0, x = var_6020_cast_fp16)[name = string("op_6310_cast_fp16")]; + tensor var_6317_begin_0 = const()[name = string("op_6317_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6317_end_0 = const()[name = string("op_6317_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6317_end_mask_0 = const()[name = string("op_6317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6317_cast_fp16 = slice_by_index(begin = var_6317_begin_0, end = var_6317_end_0, end_mask = var_6317_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6317_cast_fp16")]; + tensor var_6324_begin_0 = const()[name = string("op_6324_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6324_end_0 = const()[name = string("op_6324_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6324_end_mask_0 = const()[name = string("op_6324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6324_cast_fp16 = slice_by_index(begin = var_6324_begin_0, end = var_6324_end_0, end_mask = var_6324_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6324_cast_fp16")]; + tensor var_6331_begin_0 = const()[name = string("op_6331_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6331_end_0 = const()[name = string("op_6331_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6331_end_mask_0 = const()[name = string("op_6331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6331_cast_fp16 = slice_by_index(begin = var_6331_begin_0, end = var_6331_end_0, end_mask = var_6331_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6331_cast_fp16")]; + tensor var_6338_begin_0 = const()[name = string("op_6338_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6338_end_0 = const()[name = string("op_6338_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6338_end_mask_0 = const()[name = string("op_6338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6338_cast_fp16 = slice_by_index(begin = var_6338_begin_0, end = var_6338_end_0, end_mask = var_6338_end_mask_0, x = var_6024_cast_fp16)[name = string("op_6338_cast_fp16")]; + tensor var_6345_begin_0 = const()[name = string("op_6345_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6345_end_0 = const()[name = string("op_6345_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6345_end_mask_0 = const()[name = string("op_6345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6345_cast_fp16 = slice_by_index(begin = var_6345_begin_0, end = var_6345_end_0, end_mask = var_6345_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6345_cast_fp16")]; + tensor var_6352_begin_0 = const()[name = string("op_6352_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6352_end_0 = const()[name = string("op_6352_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6352_end_mask_0 = const()[name = string("op_6352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6352_cast_fp16 = slice_by_index(begin = var_6352_begin_0, end = var_6352_end_0, end_mask = var_6352_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6352_cast_fp16")]; + tensor var_6359_begin_0 = const()[name = string("op_6359_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6359_end_0 = const()[name = string("op_6359_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6359_end_mask_0 = const()[name = string("op_6359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6359_cast_fp16 = slice_by_index(begin = var_6359_begin_0, end = var_6359_end_0, end_mask = var_6359_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6359_cast_fp16")]; + tensor var_6366_begin_0 = const()[name = string("op_6366_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6366_end_0 = const()[name = string("op_6366_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6366_end_mask_0 = const()[name = string("op_6366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6366_cast_fp16 = slice_by_index(begin = var_6366_begin_0, end = var_6366_end_0, end_mask = var_6366_end_mask_0, x = var_6028_cast_fp16)[name = string("op_6366_cast_fp16")]; + tensor k_13_perm_0 = const()[name = string("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6371_begin_0 = const()[name = string("op_6371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6371_end_0 = const()[name = string("op_6371_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_6371_end_mask_0 = const()[name = string("op_6371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_13_cast_fp16 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = string("transpose_5")]; + tensor var_6371_cast_fp16 = slice_by_index(begin = var_6371_begin_0, end = var_6371_end_0, end_mask = var_6371_end_mask_0, x = k_13_cast_fp16)[name = string("op_6371_cast_fp16")]; + tensor var_6375_begin_0 = const()[name = string("op_6375_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6375_end_0 = const()[name = string("op_6375_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_6375_end_mask_0 = const()[name = string("op_6375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6375_cast_fp16 = slice_by_index(begin = var_6375_begin_0, end = var_6375_end_0, end_mask = var_6375_end_mask_0, x = k_13_cast_fp16)[name = string("op_6375_cast_fp16")]; + tensor var_6379_begin_0 = const()[name = string("op_6379_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6379_end_0 = const()[name = string("op_6379_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_6379_end_mask_0 = const()[name = string("op_6379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6379_cast_fp16 = slice_by_index(begin = var_6379_begin_0, end = var_6379_end_0, end_mask = var_6379_end_mask_0, x = k_13_cast_fp16)[name = string("op_6379_cast_fp16")]; + tensor var_6383_begin_0 = const()[name = string("op_6383_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6383_end_0 = const()[name = string("op_6383_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_6383_end_mask_0 = const()[name = string("op_6383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6383_cast_fp16 = slice_by_index(begin = var_6383_begin_0, end = var_6383_end_0, end_mask = var_6383_end_mask_0, x = k_13_cast_fp16)[name = string("op_6383_cast_fp16")]; + tensor var_6387_begin_0 = const()[name = string("op_6387_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6387_end_0 = const()[name = string("op_6387_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_6387_end_mask_0 = const()[name = string("op_6387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6387_cast_fp16 = slice_by_index(begin = var_6387_begin_0, end = var_6387_end_0, end_mask = var_6387_end_mask_0, x = k_13_cast_fp16)[name = string("op_6387_cast_fp16")]; + tensor var_6391_begin_0 = const()[name = string("op_6391_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6391_end_0 = const()[name = string("op_6391_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_6391_end_mask_0 = const()[name = string("op_6391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6391_cast_fp16 = slice_by_index(begin = var_6391_begin_0, end = var_6391_end_0, end_mask = var_6391_end_mask_0, x = k_13_cast_fp16)[name = string("op_6391_cast_fp16")]; + tensor var_6395_begin_0 = const()[name = string("op_6395_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6395_end_0 = const()[name = string("op_6395_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_6395_end_mask_0 = const()[name = string("op_6395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6395_cast_fp16 = slice_by_index(begin = var_6395_begin_0, end = var_6395_end_0, end_mask = var_6395_end_mask_0, x = k_13_cast_fp16)[name = string("op_6395_cast_fp16")]; + tensor var_6399_begin_0 = const()[name = string("op_6399_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6399_end_0 = const()[name = string("op_6399_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_6399_end_mask_0 = const()[name = string("op_6399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6399_cast_fp16 = slice_by_index(begin = var_6399_begin_0, end = var_6399_end_0, end_mask = var_6399_end_mask_0, x = k_13_cast_fp16)[name = string("op_6399_cast_fp16")]; + tensor var_6403_begin_0 = const()[name = string("op_6403_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6403_end_0 = const()[name = string("op_6403_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_6403_end_mask_0 = const()[name = string("op_6403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6403_cast_fp16 = slice_by_index(begin = var_6403_begin_0, end = var_6403_end_0, end_mask = var_6403_end_mask_0, x = k_13_cast_fp16)[name = string("op_6403_cast_fp16")]; + tensor var_6407_begin_0 = const()[name = string("op_6407_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6407_end_0 = const()[name = string("op_6407_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_6407_end_mask_0 = const()[name = string("op_6407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6407_cast_fp16 = slice_by_index(begin = var_6407_begin_0, end = var_6407_end_0, end_mask = var_6407_end_mask_0, x = k_13_cast_fp16)[name = string("op_6407_cast_fp16")]; + tensor var_6411_begin_0 = const()[name = string("op_6411_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6411_end_0 = const()[name = string("op_6411_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_6411_end_mask_0 = const()[name = string("op_6411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6411_cast_fp16 = slice_by_index(begin = var_6411_begin_0, end = var_6411_end_0, end_mask = var_6411_end_mask_0, x = k_13_cast_fp16)[name = string("op_6411_cast_fp16")]; + tensor var_6415_begin_0 = const()[name = string("op_6415_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6415_end_0 = const()[name = string("op_6415_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_6415_end_mask_0 = const()[name = string("op_6415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6415_cast_fp16 = slice_by_index(begin = var_6415_begin_0, end = var_6415_end_0, end_mask = var_6415_end_mask_0, x = k_13_cast_fp16)[name = string("op_6415_cast_fp16")]; + tensor var_6417_begin_0 = const()[name = string("op_6417_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6417_end_0 = const()[name = string("op_6417_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6417_end_mask_0 = const()[name = string("op_6417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6417_cast_fp16 = slice_by_index(begin = var_6417_begin_0, end = var_6417_end_0, end_mask = var_6417_end_mask_0, x = value_13_cast_fp16)[name = string("op_6417_cast_fp16")]; + tensor var_6421_begin_0 = const()[name = string("op_6421_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6421_end_0 = const()[name = string("op_6421_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6421_end_mask_0 = const()[name = string("op_6421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6421_cast_fp16 = slice_by_index(begin = var_6421_begin_0, end = var_6421_end_0, end_mask = var_6421_end_mask_0, x = value_13_cast_fp16)[name = string("op_6421_cast_fp16")]; + tensor var_6425_begin_0 = const()[name = string("op_6425_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6425_end_0 = const()[name = string("op_6425_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6425_end_mask_0 = const()[name = string("op_6425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6425_cast_fp16 = slice_by_index(begin = var_6425_begin_0, end = var_6425_end_0, end_mask = var_6425_end_mask_0, x = value_13_cast_fp16)[name = string("op_6425_cast_fp16")]; + tensor var_6429_begin_0 = const()[name = string("op_6429_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6429_end_0 = const()[name = string("op_6429_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6429_end_mask_0 = const()[name = string("op_6429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6429_cast_fp16 = slice_by_index(begin = var_6429_begin_0, end = var_6429_end_0, end_mask = var_6429_end_mask_0, x = value_13_cast_fp16)[name = string("op_6429_cast_fp16")]; + tensor var_6433_begin_0 = const()[name = string("op_6433_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6433_end_0 = const()[name = string("op_6433_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6433_end_mask_0 = const()[name = string("op_6433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6433_cast_fp16 = slice_by_index(begin = var_6433_begin_0, end = var_6433_end_0, end_mask = var_6433_end_mask_0, x = value_13_cast_fp16)[name = string("op_6433_cast_fp16")]; + tensor var_6437_begin_0 = const()[name = string("op_6437_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6437_end_0 = const()[name = string("op_6437_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6437_end_mask_0 = const()[name = string("op_6437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6437_cast_fp16 = slice_by_index(begin = var_6437_begin_0, end = var_6437_end_0, end_mask = var_6437_end_mask_0, x = value_13_cast_fp16)[name = string("op_6437_cast_fp16")]; + tensor var_6441_begin_0 = const()[name = string("op_6441_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6441_end_0 = const()[name = string("op_6441_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6441_end_mask_0 = const()[name = string("op_6441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6441_cast_fp16 = slice_by_index(begin = var_6441_begin_0, end = var_6441_end_0, end_mask = var_6441_end_mask_0, x = value_13_cast_fp16)[name = string("op_6441_cast_fp16")]; + tensor var_6445_begin_0 = const()[name = string("op_6445_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6445_end_0 = const()[name = string("op_6445_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6445_end_mask_0 = const()[name = string("op_6445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6445_cast_fp16 = slice_by_index(begin = var_6445_begin_0, end = var_6445_end_0, end_mask = var_6445_end_mask_0, x = value_13_cast_fp16)[name = string("op_6445_cast_fp16")]; + tensor var_6449_begin_0 = const()[name = string("op_6449_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6449_end_0 = const()[name = string("op_6449_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6449_end_mask_0 = const()[name = string("op_6449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6449_cast_fp16 = slice_by_index(begin = var_6449_begin_0, end = var_6449_end_0, end_mask = var_6449_end_mask_0, x = value_13_cast_fp16)[name = string("op_6449_cast_fp16")]; + tensor var_6453_begin_0 = const()[name = string("op_6453_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6453_end_0 = const()[name = string("op_6453_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6453_end_mask_0 = const()[name = string("op_6453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6453_cast_fp16 = slice_by_index(begin = var_6453_begin_0, end = var_6453_end_0, end_mask = var_6453_end_mask_0, x = value_13_cast_fp16)[name = string("op_6453_cast_fp16")]; + tensor var_6457_begin_0 = const()[name = string("op_6457_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6457_end_0 = const()[name = string("op_6457_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6457_end_mask_0 = const()[name = string("op_6457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6457_cast_fp16 = slice_by_index(begin = var_6457_begin_0, end = var_6457_end_0, end_mask = var_6457_end_mask_0, x = value_13_cast_fp16)[name = string("op_6457_cast_fp16")]; + tensor var_6461_begin_0 = const()[name = string("op_6461_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6461_end_0 = const()[name = string("op_6461_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6461_end_mask_0 = const()[name = string("op_6461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6461_cast_fp16 = slice_by_index(begin = var_6461_begin_0, end = var_6461_end_0, end_mask = var_6461_end_mask_0, x = value_13_cast_fp16)[name = string("op_6461_cast_fp16")]; + string _SplitHeadsQ__mh_w_577_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_577_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_577_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_577_equation_0, values = (var_6371_cast_fp16, var_6037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_577_cast_fp16")]; + string _SplitHeadsQ__mh_w_579_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_579_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_579_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_579_equation_0, values = (var_6371_cast_fp16, var_6044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_579_cast_fp16")]; + string _SplitHeadsQ__mh_w_581_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_581_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_581_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_581_equation_0, values = (var_6371_cast_fp16, var_6051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_581_cast_fp16")]; + string _SplitHeadsQ__mh_w_583_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_583_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_583_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_583_equation_0, values = (var_6371_cast_fp16, var_6058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_583_cast_fp16")]; + string _SplitHeadsQ__mh_w_585_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_585_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_585_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_585_equation_0, values = (var_6375_cast_fp16, var_6065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_585_cast_fp16")]; + string _SplitHeadsQ__mh_w_587_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_587_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_587_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_587_equation_0, values = (var_6375_cast_fp16, var_6072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_587_cast_fp16")]; + string _SplitHeadsQ__mh_w_589_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_589_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_589_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_589_equation_0, values = (var_6375_cast_fp16, var_6079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_589_cast_fp16")]; + string _SplitHeadsQ__mh_w_591_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_591_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_591_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_591_equation_0, values = (var_6375_cast_fp16, var_6086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_591_cast_fp16")]; + string _SplitHeadsQ__mh_w_593_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_593_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_593_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_593_equation_0, values = (var_6379_cast_fp16, var_6093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_593_cast_fp16")]; + string _SplitHeadsQ__mh_w_595_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_595_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_595_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_595_equation_0, values = (var_6379_cast_fp16, var_6100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_595_cast_fp16")]; + string _SplitHeadsQ__mh_w_597_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_597_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_597_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_597_equation_0, values = (var_6379_cast_fp16, var_6107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_597_cast_fp16")]; + string _SplitHeadsQ__mh_w_599_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_599_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_599_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_599_equation_0, values = (var_6379_cast_fp16, var_6114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_599_cast_fp16")]; + string _SplitHeadsQ__mh_w_601_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_601_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_601_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_601_equation_0, values = (var_6383_cast_fp16, var_6121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_601_cast_fp16")]; + string _SplitHeadsQ__mh_w_603_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_603_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_603_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_603_equation_0, values = (var_6383_cast_fp16, var_6128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_603_cast_fp16")]; + string _SplitHeadsQ__mh_w_605_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_605_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_605_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_605_equation_0, values = (var_6383_cast_fp16, var_6135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_605_cast_fp16")]; + string _SplitHeadsQ__mh_w_607_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_607_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_607_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_607_equation_0, values = (var_6383_cast_fp16, var_6142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_607_cast_fp16")]; + string _SplitHeadsQ__mh_w_609_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_609_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_609_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_609_equation_0, values = (var_6387_cast_fp16, var_6149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_609_cast_fp16")]; + string _SplitHeadsQ__mh_w_611_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_611_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_611_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_611_equation_0, values = (var_6387_cast_fp16, var_6156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_611_cast_fp16")]; + string _SplitHeadsQ__mh_w_613_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_613_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_613_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_613_equation_0, values = (var_6387_cast_fp16, var_6163_cast_fp16))[name = string("_SplitHeadsQ__mh_w_613_cast_fp16")]; + string _SplitHeadsQ__mh_w_615_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_615_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_615_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_615_equation_0, values = (var_6387_cast_fp16, var_6170_cast_fp16))[name = string("_SplitHeadsQ__mh_w_615_cast_fp16")]; + string _SplitHeadsQ__mh_w_617_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_617_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_617_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_617_equation_0, values = (var_6391_cast_fp16, var_6177_cast_fp16))[name = string("_SplitHeadsQ__mh_w_617_cast_fp16")]; + string _SplitHeadsQ__mh_w_619_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_619_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_619_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_619_equation_0, values = (var_6391_cast_fp16, var_6184_cast_fp16))[name = string("_SplitHeadsQ__mh_w_619_cast_fp16")]; + string _SplitHeadsQ__mh_w_621_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_621_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_621_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_621_equation_0, values = (var_6391_cast_fp16, var_6191_cast_fp16))[name = string("_SplitHeadsQ__mh_w_621_cast_fp16")]; + string _SplitHeadsQ__mh_w_623_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_623_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_623_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_623_equation_0, values = (var_6391_cast_fp16, var_6198_cast_fp16))[name = string("_SplitHeadsQ__mh_w_623_cast_fp16")]; + string _SplitHeadsQ__mh_w_625_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_625_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_625_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_625_equation_0, values = (var_6395_cast_fp16, var_6205_cast_fp16))[name = string("_SplitHeadsQ__mh_w_625_cast_fp16")]; + string _SplitHeadsQ__mh_w_627_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_627_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_627_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_627_equation_0, values = (var_6395_cast_fp16, var_6212_cast_fp16))[name = string("_SplitHeadsQ__mh_w_627_cast_fp16")]; + string _SplitHeadsQ__mh_w_629_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_629_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_629_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_629_equation_0, values = (var_6395_cast_fp16, var_6219_cast_fp16))[name = string("_SplitHeadsQ__mh_w_629_cast_fp16")]; + string _SplitHeadsQ__mh_w_631_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_631_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_631_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_631_equation_0, values = (var_6395_cast_fp16, var_6226_cast_fp16))[name = string("_SplitHeadsQ__mh_w_631_cast_fp16")]; + string _SplitHeadsQ__mh_w_633_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_633_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_633_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_633_equation_0, values = (var_6399_cast_fp16, var_6233_cast_fp16))[name = string("_SplitHeadsQ__mh_w_633_cast_fp16")]; + string _SplitHeadsQ__mh_w_635_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_635_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_635_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_635_equation_0, values = (var_6399_cast_fp16, var_6240_cast_fp16))[name = string("_SplitHeadsQ__mh_w_635_cast_fp16")]; + string _SplitHeadsQ__mh_w_637_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_637_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_637_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_637_equation_0, values = (var_6399_cast_fp16, var_6247_cast_fp16))[name = string("_SplitHeadsQ__mh_w_637_cast_fp16")]; + string _SplitHeadsQ__mh_w_639_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_639_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_639_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_639_equation_0, values = (var_6399_cast_fp16, var_6254_cast_fp16))[name = string("_SplitHeadsQ__mh_w_639_cast_fp16")]; + string _SplitHeadsQ__mh_w_641_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_641_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_641_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_641_equation_0, values = (var_6403_cast_fp16, var_6261_cast_fp16))[name = string("_SplitHeadsQ__mh_w_641_cast_fp16")]; + string _SplitHeadsQ__mh_w_643_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_643_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_643_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_643_equation_0, values = (var_6403_cast_fp16, var_6268_cast_fp16))[name = string("_SplitHeadsQ__mh_w_643_cast_fp16")]; + string _SplitHeadsQ__mh_w_645_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_645_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_645_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_645_equation_0, values = (var_6403_cast_fp16, var_6275_cast_fp16))[name = string("_SplitHeadsQ__mh_w_645_cast_fp16")]; + string _SplitHeadsQ__mh_w_647_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_647_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_647_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_647_equation_0, values = (var_6403_cast_fp16, var_6282_cast_fp16))[name = string("_SplitHeadsQ__mh_w_647_cast_fp16")]; + string _SplitHeadsQ__mh_w_649_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_649_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_649_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_649_equation_0, values = (var_6407_cast_fp16, var_6289_cast_fp16))[name = string("_SplitHeadsQ__mh_w_649_cast_fp16")]; + string _SplitHeadsQ__mh_w_651_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_651_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_651_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_651_equation_0, values = (var_6407_cast_fp16, var_6296_cast_fp16))[name = string("_SplitHeadsQ__mh_w_651_cast_fp16")]; + string _SplitHeadsQ__mh_w_653_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_653_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_653_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_653_equation_0, values = (var_6407_cast_fp16, var_6303_cast_fp16))[name = string("_SplitHeadsQ__mh_w_653_cast_fp16")]; + string _SplitHeadsQ__mh_w_655_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_655_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_655_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_655_equation_0, values = (var_6407_cast_fp16, var_6310_cast_fp16))[name = string("_SplitHeadsQ__mh_w_655_cast_fp16")]; + string _SplitHeadsQ__mh_w_657_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_657_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_657_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_657_equation_0, values = (var_6411_cast_fp16, var_6317_cast_fp16))[name = string("_SplitHeadsQ__mh_w_657_cast_fp16")]; + string _SplitHeadsQ__mh_w_659_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_659_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_659_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_659_equation_0, values = (var_6411_cast_fp16, var_6324_cast_fp16))[name = string("_SplitHeadsQ__mh_w_659_cast_fp16")]; + string _SplitHeadsQ__mh_w_661_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_661_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_661_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_661_equation_0, values = (var_6411_cast_fp16, var_6331_cast_fp16))[name = string("_SplitHeadsQ__mh_w_661_cast_fp16")]; + string _SplitHeadsQ__mh_w_663_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_663_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_663_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_663_equation_0, values = (var_6411_cast_fp16, var_6338_cast_fp16))[name = string("_SplitHeadsQ__mh_w_663_cast_fp16")]; + string _SplitHeadsQ__mh_w_665_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_665_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_665_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_665_equation_0, values = (var_6415_cast_fp16, var_6345_cast_fp16))[name = string("_SplitHeadsQ__mh_w_665_cast_fp16")]; + string _SplitHeadsQ__mh_w_667_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_667_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_667_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_667_equation_0, values = (var_6415_cast_fp16, var_6352_cast_fp16))[name = string("_SplitHeadsQ__mh_w_667_cast_fp16")]; + string _SplitHeadsQ__mh_w_669_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_669_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_669_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_669_equation_0, values = (var_6415_cast_fp16, var_6359_cast_fp16))[name = string("_SplitHeadsQ__mh_w_669_cast_fp16")]; + string _SplitHeadsQ__mh_w_671_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_671_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_671_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_671_equation_0, values = (var_6415_cast_fp16, var_6366_cast_fp16))[name = string("_SplitHeadsQ__mh_w_671_cast_fp16")]; + fp16 var_6560_to_fp16 = const()[name = string("op_6560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_577_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_577_cast_fp16, y = var_6560_to_fp16)[name = string("aw_chunk_577_cast_fp16")]; + fp16 var_6562_to_fp16 = const()[name = string("op_6562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_579_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_579_cast_fp16, y = var_6562_to_fp16)[name = string("aw_chunk_579_cast_fp16")]; + fp16 var_6564_to_fp16 = const()[name = string("op_6564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_581_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_581_cast_fp16, y = var_6564_to_fp16)[name = string("aw_chunk_581_cast_fp16")]; + fp16 var_6566_to_fp16 = const()[name = string("op_6566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_583_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_583_cast_fp16, y = var_6566_to_fp16)[name = string("aw_chunk_583_cast_fp16")]; + fp16 var_6568_to_fp16 = const()[name = string("op_6568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_585_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_585_cast_fp16, y = var_6568_to_fp16)[name = string("aw_chunk_585_cast_fp16")]; + fp16 var_6570_to_fp16 = const()[name = string("op_6570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_587_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_587_cast_fp16, y = var_6570_to_fp16)[name = string("aw_chunk_587_cast_fp16")]; + fp16 var_6572_to_fp16 = const()[name = string("op_6572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_589_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_589_cast_fp16, y = var_6572_to_fp16)[name = string("aw_chunk_589_cast_fp16")]; + fp16 var_6574_to_fp16 = const()[name = string("op_6574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_591_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_591_cast_fp16, y = var_6574_to_fp16)[name = string("aw_chunk_591_cast_fp16")]; + fp16 var_6576_to_fp16 = const()[name = string("op_6576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_593_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_593_cast_fp16, y = var_6576_to_fp16)[name = string("aw_chunk_593_cast_fp16")]; + fp16 var_6578_to_fp16 = const()[name = string("op_6578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_595_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_595_cast_fp16, y = var_6578_to_fp16)[name = string("aw_chunk_595_cast_fp16")]; + fp16 var_6580_to_fp16 = const()[name = string("op_6580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_597_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_597_cast_fp16, y = var_6580_to_fp16)[name = string("aw_chunk_597_cast_fp16")]; + fp16 var_6582_to_fp16 = const()[name = string("op_6582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_599_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_599_cast_fp16, y = var_6582_to_fp16)[name = string("aw_chunk_599_cast_fp16")]; + fp16 var_6584_to_fp16 = const()[name = string("op_6584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_601_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_601_cast_fp16, y = var_6584_to_fp16)[name = string("aw_chunk_601_cast_fp16")]; + fp16 var_6586_to_fp16 = const()[name = string("op_6586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_603_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_603_cast_fp16, y = var_6586_to_fp16)[name = string("aw_chunk_603_cast_fp16")]; + fp16 var_6588_to_fp16 = const()[name = string("op_6588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_605_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_605_cast_fp16, y = var_6588_to_fp16)[name = string("aw_chunk_605_cast_fp16")]; + fp16 var_6590_to_fp16 = const()[name = string("op_6590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_607_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_607_cast_fp16, y = var_6590_to_fp16)[name = string("aw_chunk_607_cast_fp16")]; + fp16 var_6592_to_fp16 = const()[name = string("op_6592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_609_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_609_cast_fp16, y = var_6592_to_fp16)[name = string("aw_chunk_609_cast_fp16")]; + fp16 var_6594_to_fp16 = const()[name = string("op_6594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_611_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_611_cast_fp16, y = var_6594_to_fp16)[name = string("aw_chunk_611_cast_fp16")]; + fp16 var_6596_to_fp16 = const()[name = string("op_6596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_613_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_613_cast_fp16, y = var_6596_to_fp16)[name = string("aw_chunk_613_cast_fp16")]; + fp16 var_6598_to_fp16 = const()[name = string("op_6598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_615_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_615_cast_fp16, y = var_6598_to_fp16)[name = string("aw_chunk_615_cast_fp16")]; + fp16 var_6600_to_fp16 = const()[name = string("op_6600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_617_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_617_cast_fp16, y = var_6600_to_fp16)[name = string("aw_chunk_617_cast_fp16")]; + fp16 var_6602_to_fp16 = const()[name = string("op_6602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_619_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_619_cast_fp16, y = var_6602_to_fp16)[name = string("aw_chunk_619_cast_fp16")]; + fp16 var_6604_to_fp16 = const()[name = string("op_6604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_621_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_621_cast_fp16, y = var_6604_to_fp16)[name = string("aw_chunk_621_cast_fp16")]; + fp16 var_6606_to_fp16 = const()[name = string("op_6606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_623_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_623_cast_fp16, y = var_6606_to_fp16)[name = string("aw_chunk_623_cast_fp16")]; + fp16 var_6608_to_fp16 = const()[name = string("op_6608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_625_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_625_cast_fp16, y = var_6608_to_fp16)[name = string("aw_chunk_625_cast_fp16")]; + fp16 var_6610_to_fp16 = const()[name = string("op_6610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_627_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_627_cast_fp16, y = var_6610_to_fp16)[name = string("aw_chunk_627_cast_fp16")]; + fp16 var_6612_to_fp16 = const()[name = string("op_6612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_629_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_629_cast_fp16, y = var_6612_to_fp16)[name = string("aw_chunk_629_cast_fp16")]; + fp16 var_6614_to_fp16 = const()[name = string("op_6614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_631_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_631_cast_fp16, y = var_6614_to_fp16)[name = string("aw_chunk_631_cast_fp16")]; + fp16 var_6616_to_fp16 = const()[name = string("op_6616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_633_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_633_cast_fp16, y = var_6616_to_fp16)[name = string("aw_chunk_633_cast_fp16")]; + fp16 var_6618_to_fp16 = const()[name = string("op_6618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_635_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_635_cast_fp16, y = var_6618_to_fp16)[name = string("aw_chunk_635_cast_fp16")]; + fp16 var_6620_to_fp16 = const()[name = string("op_6620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_637_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_637_cast_fp16, y = var_6620_to_fp16)[name = string("aw_chunk_637_cast_fp16")]; + fp16 var_6622_to_fp16 = const()[name = string("op_6622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_639_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_639_cast_fp16, y = var_6622_to_fp16)[name = string("aw_chunk_639_cast_fp16")]; + fp16 var_6624_to_fp16 = const()[name = string("op_6624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_641_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_641_cast_fp16, y = var_6624_to_fp16)[name = string("aw_chunk_641_cast_fp16")]; + fp16 var_6626_to_fp16 = const()[name = string("op_6626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_643_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_643_cast_fp16, y = var_6626_to_fp16)[name = string("aw_chunk_643_cast_fp16")]; + fp16 var_6628_to_fp16 = const()[name = string("op_6628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_645_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_645_cast_fp16, y = var_6628_to_fp16)[name = string("aw_chunk_645_cast_fp16")]; + fp16 var_6630_to_fp16 = const()[name = string("op_6630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_647_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_647_cast_fp16, y = var_6630_to_fp16)[name = string("aw_chunk_647_cast_fp16")]; + fp16 var_6632_to_fp16 = const()[name = string("op_6632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_649_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_649_cast_fp16, y = var_6632_to_fp16)[name = string("aw_chunk_649_cast_fp16")]; + fp16 var_6634_to_fp16 = const()[name = string("op_6634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_651_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_651_cast_fp16, y = var_6634_to_fp16)[name = string("aw_chunk_651_cast_fp16")]; + fp16 var_6636_to_fp16 = const()[name = string("op_6636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_653_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_653_cast_fp16, y = var_6636_to_fp16)[name = string("aw_chunk_653_cast_fp16")]; + fp16 var_6638_to_fp16 = const()[name = string("op_6638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_655_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_655_cast_fp16, y = var_6638_to_fp16)[name = string("aw_chunk_655_cast_fp16")]; + fp16 var_6640_to_fp16 = const()[name = string("op_6640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_657_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_657_cast_fp16, y = var_6640_to_fp16)[name = string("aw_chunk_657_cast_fp16")]; + fp16 var_6642_to_fp16 = const()[name = string("op_6642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_659_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_659_cast_fp16, y = var_6642_to_fp16)[name = string("aw_chunk_659_cast_fp16")]; + fp16 var_6644_to_fp16 = const()[name = string("op_6644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_661_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_661_cast_fp16, y = var_6644_to_fp16)[name = string("aw_chunk_661_cast_fp16")]; + fp16 var_6646_to_fp16 = const()[name = string("op_6646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_663_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_663_cast_fp16, y = var_6646_to_fp16)[name = string("aw_chunk_663_cast_fp16")]; + fp16 var_6648_to_fp16 = const()[name = string("op_6648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_665_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_665_cast_fp16, y = var_6648_to_fp16)[name = string("aw_chunk_665_cast_fp16")]; + fp16 var_6650_to_fp16 = const()[name = string("op_6650_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_667_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_667_cast_fp16, y = var_6650_to_fp16)[name = string("aw_chunk_667_cast_fp16")]; + fp16 var_6652_to_fp16 = const()[name = string("op_6652_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_669_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_669_cast_fp16, y = var_6652_to_fp16)[name = string("aw_chunk_669_cast_fp16")]; + fp16 var_6654_to_fp16 = const()[name = string("op_6654_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_671_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_671_cast_fp16, y = var_6654_to_fp16)[name = string("aw_chunk_671_cast_fp16")]; + tensor var_6656_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_577_cast_fp16)[name = string("op_6656_cast_fp16")]; + tensor var_6657_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_579_cast_fp16)[name = string("op_6657_cast_fp16")]; + tensor var_6658_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_581_cast_fp16)[name = string("op_6658_cast_fp16")]; + tensor var_6659_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_583_cast_fp16)[name = string("op_6659_cast_fp16")]; + tensor var_6660_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_585_cast_fp16)[name = string("op_6660_cast_fp16")]; + tensor var_6661_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_587_cast_fp16)[name = string("op_6661_cast_fp16")]; + tensor var_6662_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_589_cast_fp16)[name = string("op_6662_cast_fp16")]; + tensor var_6663_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_591_cast_fp16)[name = string("op_6663_cast_fp16")]; + tensor var_6664_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_593_cast_fp16)[name = string("op_6664_cast_fp16")]; + tensor var_6665_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_595_cast_fp16)[name = string("op_6665_cast_fp16")]; + tensor var_6666_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_597_cast_fp16)[name = string("op_6666_cast_fp16")]; + tensor var_6667_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_599_cast_fp16)[name = string("op_6667_cast_fp16")]; + tensor var_6668_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_601_cast_fp16)[name = string("op_6668_cast_fp16")]; + tensor var_6669_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_603_cast_fp16)[name = string("op_6669_cast_fp16")]; + tensor var_6670_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_605_cast_fp16)[name = string("op_6670_cast_fp16")]; + tensor var_6671_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_607_cast_fp16)[name = string("op_6671_cast_fp16")]; + tensor var_6672_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_609_cast_fp16)[name = string("op_6672_cast_fp16")]; + tensor var_6673_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_611_cast_fp16)[name = string("op_6673_cast_fp16")]; + tensor var_6674_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_613_cast_fp16)[name = string("op_6674_cast_fp16")]; + tensor var_6675_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_615_cast_fp16)[name = string("op_6675_cast_fp16")]; + tensor var_6676_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_617_cast_fp16)[name = string("op_6676_cast_fp16")]; + tensor var_6677_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_619_cast_fp16)[name = string("op_6677_cast_fp16")]; + tensor var_6678_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_621_cast_fp16)[name = string("op_6678_cast_fp16")]; + tensor var_6679_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_623_cast_fp16)[name = string("op_6679_cast_fp16")]; + tensor var_6680_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_625_cast_fp16)[name = string("op_6680_cast_fp16")]; + tensor var_6681_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_627_cast_fp16)[name = string("op_6681_cast_fp16")]; + tensor var_6682_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_629_cast_fp16)[name = string("op_6682_cast_fp16")]; + tensor var_6683_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_631_cast_fp16)[name = string("op_6683_cast_fp16")]; + tensor var_6684_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_633_cast_fp16)[name = string("op_6684_cast_fp16")]; + tensor var_6685_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_635_cast_fp16)[name = string("op_6685_cast_fp16")]; + tensor var_6686_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_637_cast_fp16)[name = string("op_6686_cast_fp16")]; + tensor var_6687_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_639_cast_fp16)[name = string("op_6687_cast_fp16")]; + tensor var_6688_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_641_cast_fp16)[name = string("op_6688_cast_fp16")]; + tensor var_6689_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_643_cast_fp16)[name = string("op_6689_cast_fp16")]; + tensor var_6690_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_645_cast_fp16)[name = string("op_6690_cast_fp16")]; + tensor var_6691_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_647_cast_fp16)[name = string("op_6691_cast_fp16")]; + tensor var_6692_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_649_cast_fp16)[name = string("op_6692_cast_fp16")]; + tensor var_6693_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_651_cast_fp16)[name = string("op_6693_cast_fp16")]; + tensor var_6694_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_653_cast_fp16)[name = string("op_6694_cast_fp16")]; + tensor var_6695_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_655_cast_fp16)[name = string("op_6695_cast_fp16")]; + tensor var_6696_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_657_cast_fp16)[name = string("op_6696_cast_fp16")]; + tensor var_6697_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_659_cast_fp16)[name = string("op_6697_cast_fp16")]; + tensor var_6698_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_661_cast_fp16)[name = string("op_6698_cast_fp16")]; + tensor var_6699_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_663_cast_fp16)[name = string("op_6699_cast_fp16")]; + tensor var_6700_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_665_cast_fp16)[name = string("op_6700_cast_fp16")]; + tensor var_6701_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_667_cast_fp16)[name = string("op_6701_cast_fp16")]; + tensor var_6702_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_669_cast_fp16)[name = string("op_6702_cast_fp16")]; + tensor var_6703_cast_fp16 = softmax(axis = var_5929, x = aw_chunk_671_cast_fp16)[name = string("op_6703_cast_fp16")]; + string var_6705_equation_0 = const()[name = string("op_6705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6705_cast_fp16 = einsum(equation = var_6705_equation_0, values = (var_6417_cast_fp16, var_6656_cast_fp16))[name = string("op_6705_cast_fp16")]; + string var_6707_equation_0 = const()[name = string("op_6707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6707_cast_fp16 = einsum(equation = var_6707_equation_0, values = (var_6417_cast_fp16, var_6657_cast_fp16))[name = string("op_6707_cast_fp16")]; + string var_6709_equation_0 = const()[name = string("op_6709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6709_cast_fp16 = einsum(equation = var_6709_equation_0, values = (var_6417_cast_fp16, var_6658_cast_fp16))[name = string("op_6709_cast_fp16")]; + string var_6711_equation_0 = const()[name = string("op_6711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6711_cast_fp16 = einsum(equation = var_6711_equation_0, values = (var_6417_cast_fp16, var_6659_cast_fp16))[name = string("op_6711_cast_fp16")]; + string var_6713_equation_0 = const()[name = string("op_6713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6713_cast_fp16 = einsum(equation = var_6713_equation_0, values = (var_6421_cast_fp16, var_6660_cast_fp16))[name = string("op_6713_cast_fp16")]; + string var_6715_equation_0 = const()[name = string("op_6715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6715_cast_fp16 = einsum(equation = var_6715_equation_0, values = (var_6421_cast_fp16, var_6661_cast_fp16))[name = string("op_6715_cast_fp16")]; + string var_6717_equation_0 = const()[name = string("op_6717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6717_cast_fp16 = einsum(equation = var_6717_equation_0, values = (var_6421_cast_fp16, var_6662_cast_fp16))[name = string("op_6717_cast_fp16")]; + string var_6719_equation_0 = const()[name = string("op_6719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6719_cast_fp16 = einsum(equation = var_6719_equation_0, values = (var_6421_cast_fp16, var_6663_cast_fp16))[name = string("op_6719_cast_fp16")]; + string var_6721_equation_0 = const()[name = string("op_6721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6721_cast_fp16 = einsum(equation = var_6721_equation_0, values = (var_6425_cast_fp16, var_6664_cast_fp16))[name = string("op_6721_cast_fp16")]; + string var_6723_equation_0 = const()[name = string("op_6723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6723_cast_fp16 = einsum(equation = var_6723_equation_0, values = (var_6425_cast_fp16, var_6665_cast_fp16))[name = string("op_6723_cast_fp16")]; + string var_6725_equation_0 = const()[name = string("op_6725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6725_cast_fp16 = einsum(equation = var_6725_equation_0, values = (var_6425_cast_fp16, var_6666_cast_fp16))[name = string("op_6725_cast_fp16")]; + string var_6727_equation_0 = const()[name = string("op_6727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6727_cast_fp16 = einsum(equation = var_6727_equation_0, values = (var_6425_cast_fp16, var_6667_cast_fp16))[name = string("op_6727_cast_fp16")]; + string var_6729_equation_0 = const()[name = string("op_6729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6729_cast_fp16 = einsum(equation = var_6729_equation_0, values = (var_6429_cast_fp16, var_6668_cast_fp16))[name = string("op_6729_cast_fp16")]; + string var_6731_equation_0 = const()[name = string("op_6731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6731_cast_fp16 = einsum(equation = var_6731_equation_0, values = (var_6429_cast_fp16, var_6669_cast_fp16))[name = string("op_6731_cast_fp16")]; + string var_6733_equation_0 = const()[name = string("op_6733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6733_cast_fp16 = einsum(equation = var_6733_equation_0, values = (var_6429_cast_fp16, var_6670_cast_fp16))[name = string("op_6733_cast_fp16")]; + string var_6735_equation_0 = const()[name = string("op_6735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6735_cast_fp16 = einsum(equation = var_6735_equation_0, values = (var_6429_cast_fp16, var_6671_cast_fp16))[name = string("op_6735_cast_fp16")]; + string var_6737_equation_0 = const()[name = string("op_6737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6737_cast_fp16 = einsum(equation = var_6737_equation_0, values = (var_6433_cast_fp16, var_6672_cast_fp16))[name = string("op_6737_cast_fp16")]; + string var_6739_equation_0 = const()[name = string("op_6739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6739_cast_fp16 = einsum(equation = var_6739_equation_0, values = (var_6433_cast_fp16, var_6673_cast_fp16))[name = string("op_6739_cast_fp16")]; + string var_6741_equation_0 = const()[name = string("op_6741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6741_cast_fp16 = einsum(equation = var_6741_equation_0, values = (var_6433_cast_fp16, var_6674_cast_fp16))[name = string("op_6741_cast_fp16")]; + string var_6743_equation_0 = const()[name = string("op_6743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6743_cast_fp16 = einsum(equation = var_6743_equation_0, values = (var_6433_cast_fp16, var_6675_cast_fp16))[name = string("op_6743_cast_fp16")]; + string var_6745_equation_0 = const()[name = string("op_6745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6745_cast_fp16 = einsum(equation = var_6745_equation_0, values = (var_6437_cast_fp16, var_6676_cast_fp16))[name = string("op_6745_cast_fp16")]; + string var_6747_equation_0 = const()[name = string("op_6747_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6747_cast_fp16 = einsum(equation = var_6747_equation_0, values = (var_6437_cast_fp16, var_6677_cast_fp16))[name = string("op_6747_cast_fp16")]; + string var_6749_equation_0 = const()[name = string("op_6749_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6749_cast_fp16 = einsum(equation = var_6749_equation_0, values = (var_6437_cast_fp16, var_6678_cast_fp16))[name = string("op_6749_cast_fp16")]; + string var_6751_equation_0 = const()[name = string("op_6751_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6751_cast_fp16 = einsum(equation = var_6751_equation_0, values = (var_6437_cast_fp16, var_6679_cast_fp16))[name = string("op_6751_cast_fp16")]; + string var_6753_equation_0 = const()[name = string("op_6753_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6441_cast_fp16, var_6680_cast_fp16))[name = string("op_6753_cast_fp16")]; + string var_6755_equation_0 = const()[name = string("op_6755_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6755_cast_fp16 = einsum(equation = var_6755_equation_0, values = (var_6441_cast_fp16, var_6681_cast_fp16))[name = string("op_6755_cast_fp16")]; + string var_6757_equation_0 = const()[name = string("op_6757_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6441_cast_fp16, var_6682_cast_fp16))[name = string("op_6757_cast_fp16")]; + string var_6759_equation_0 = const()[name = string("op_6759_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6759_cast_fp16 = einsum(equation = var_6759_equation_0, values = (var_6441_cast_fp16, var_6683_cast_fp16))[name = string("op_6759_cast_fp16")]; + string var_6761_equation_0 = const()[name = string("op_6761_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6445_cast_fp16, var_6684_cast_fp16))[name = string("op_6761_cast_fp16")]; + string var_6763_equation_0 = const()[name = string("op_6763_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6763_cast_fp16 = einsum(equation = var_6763_equation_0, values = (var_6445_cast_fp16, var_6685_cast_fp16))[name = string("op_6763_cast_fp16")]; + string var_6765_equation_0 = const()[name = string("op_6765_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6445_cast_fp16, var_6686_cast_fp16))[name = string("op_6765_cast_fp16")]; + string var_6767_equation_0 = const()[name = string("op_6767_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6767_cast_fp16 = einsum(equation = var_6767_equation_0, values = (var_6445_cast_fp16, var_6687_cast_fp16))[name = string("op_6767_cast_fp16")]; + string var_6769_equation_0 = const()[name = string("op_6769_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6449_cast_fp16, var_6688_cast_fp16))[name = string("op_6769_cast_fp16")]; + string var_6771_equation_0 = const()[name = string("op_6771_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6771_cast_fp16 = einsum(equation = var_6771_equation_0, values = (var_6449_cast_fp16, var_6689_cast_fp16))[name = string("op_6771_cast_fp16")]; + string var_6773_equation_0 = const()[name = string("op_6773_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6449_cast_fp16, var_6690_cast_fp16))[name = string("op_6773_cast_fp16")]; + string var_6775_equation_0 = const()[name = string("op_6775_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6775_cast_fp16 = einsum(equation = var_6775_equation_0, values = (var_6449_cast_fp16, var_6691_cast_fp16))[name = string("op_6775_cast_fp16")]; + string var_6777_equation_0 = const()[name = string("op_6777_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6777_cast_fp16 = einsum(equation = var_6777_equation_0, values = (var_6453_cast_fp16, var_6692_cast_fp16))[name = string("op_6777_cast_fp16")]; + string var_6779_equation_0 = const()[name = string("op_6779_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6779_cast_fp16 = einsum(equation = var_6779_equation_0, values = (var_6453_cast_fp16, var_6693_cast_fp16))[name = string("op_6779_cast_fp16")]; + string var_6781_equation_0 = const()[name = string("op_6781_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6781_cast_fp16 = einsum(equation = var_6781_equation_0, values = (var_6453_cast_fp16, var_6694_cast_fp16))[name = string("op_6781_cast_fp16")]; + string var_6783_equation_0 = const()[name = string("op_6783_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6783_cast_fp16 = einsum(equation = var_6783_equation_0, values = (var_6453_cast_fp16, var_6695_cast_fp16))[name = string("op_6783_cast_fp16")]; + string var_6785_equation_0 = const()[name = string("op_6785_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6785_cast_fp16 = einsum(equation = var_6785_equation_0, values = (var_6457_cast_fp16, var_6696_cast_fp16))[name = string("op_6785_cast_fp16")]; + string var_6787_equation_0 = const()[name = string("op_6787_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6787_cast_fp16 = einsum(equation = var_6787_equation_0, values = (var_6457_cast_fp16, var_6697_cast_fp16))[name = string("op_6787_cast_fp16")]; + string var_6789_equation_0 = const()[name = string("op_6789_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6789_cast_fp16 = einsum(equation = var_6789_equation_0, values = (var_6457_cast_fp16, var_6698_cast_fp16))[name = string("op_6789_cast_fp16")]; + string var_6791_equation_0 = const()[name = string("op_6791_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6791_cast_fp16 = einsum(equation = var_6791_equation_0, values = (var_6457_cast_fp16, var_6699_cast_fp16))[name = string("op_6791_cast_fp16")]; + string var_6793_equation_0 = const()[name = string("op_6793_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6793_cast_fp16 = einsum(equation = var_6793_equation_0, values = (var_6461_cast_fp16, var_6700_cast_fp16))[name = string("op_6793_cast_fp16")]; + string var_6795_equation_0 = const()[name = string("op_6795_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6795_cast_fp16 = einsum(equation = var_6795_equation_0, values = (var_6461_cast_fp16, var_6701_cast_fp16))[name = string("op_6795_cast_fp16")]; + string var_6797_equation_0 = const()[name = string("op_6797_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6797_cast_fp16 = einsum(equation = var_6797_equation_0, values = (var_6461_cast_fp16, var_6702_cast_fp16))[name = string("op_6797_cast_fp16")]; + string var_6799_equation_0 = const()[name = string("op_6799_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_6799_cast_fp16 = einsum(equation = var_6799_equation_0, values = (var_6461_cast_fp16, var_6703_cast_fp16))[name = string("op_6799_cast_fp16")]; + bool var_6801_interleave_0 = const()[name = string("op_6801_interleave_0"), val = bool(false)]; + tensor var_6801_cast_fp16 = concat(axis = var_5912, interleave = var_6801_interleave_0, values = (var_6705_cast_fp16, var_6707_cast_fp16, var_6709_cast_fp16, var_6711_cast_fp16))[name = string("op_6801_cast_fp16")]; + bool var_6803_interleave_0 = const()[name = string("op_6803_interleave_0"), val = bool(false)]; + tensor var_6803_cast_fp16 = concat(axis = var_5912, interleave = var_6803_interleave_0, values = (var_6713_cast_fp16, var_6715_cast_fp16, var_6717_cast_fp16, var_6719_cast_fp16))[name = string("op_6803_cast_fp16")]; + bool var_6805_interleave_0 = const()[name = string("op_6805_interleave_0"), val = bool(false)]; + tensor var_6805_cast_fp16 = concat(axis = var_5912, interleave = var_6805_interleave_0, values = (var_6721_cast_fp16, var_6723_cast_fp16, var_6725_cast_fp16, var_6727_cast_fp16))[name = string("op_6805_cast_fp16")]; + bool var_6807_interleave_0 = const()[name = string("op_6807_interleave_0"), val = bool(false)]; + tensor var_6807_cast_fp16 = concat(axis = var_5912, interleave = var_6807_interleave_0, values = (var_6729_cast_fp16, var_6731_cast_fp16, var_6733_cast_fp16, var_6735_cast_fp16))[name = string("op_6807_cast_fp16")]; + bool var_6809_interleave_0 = const()[name = string("op_6809_interleave_0"), val = bool(false)]; + tensor var_6809_cast_fp16 = concat(axis = var_5912, interleave = var_6809_interleave_0, values = (var_6737_cast_fp16, var_6739_cast_fp16, var_6741_cast_fp16, var_6743_cast_fp16))[name = string("op_6809_cast_fp16")]; + bool var_6811_interleave_0 = const()[name = string("op_6811_interleave_0"), val = bool(false)]; + tensor var_6811_cast_fp16 = concat(axis = var_5912, interleave = var_6811_interleave_0, values = (var_6745_cast_fp16, var_6747_cast_fp16, var_6749_cast_fp16, var_6751_cast_fp16))[name = string("op_6811_cast_fp16")]; + bool var_6813_interleave_0 = const()[name = string("op_6813_interleave_0"), val = bool(false)]; + tensor var_6813_cast_fp16 = concat(axis = var_5912, interleave = var_6813_interleave_0, values = (var_6753_cast_fp16, var_6755_cast_fp16, var_6757_cast_fp16, var_6759_cast_fp16))[name = string("op_6813_cast_fp16")]; + bool var_6815_interleave_0 = const()[name = string("op_6815_interleave_0"), val = bool(false)]; + tensor var_6815_cast_fp16 = concat(axis = var_5912, interleave = var_6815_interleave_0, values = (var_6761_cast_fp16, var_6763_cast_fp16, var_6765_cast_fp16, var_6767_cast_fp16))[name = string("op_6815_cast_fp16")]; + bool var_6817_interleave_0 = const()[name = string("op_6817_interleave_0"), val = bool(false)]; + tensor var_6817_cast_fp16 = concat(axis = var_5912, interleave = var_6817_interleave_0, values = (var_6769_cast_fp16, var_6771_cast_fp16, var_6773_cast_fp16, var_6775_cast_fp16))[name = string("op_6817_cast_fp16")]; + bool var_6819_interleave_0 = const()[name = string("op_6819_interleave_0"), val = bool(false)]; + tensor var_6819_cast_fp16 = concat(axis = var_5912, interleave = var_6819_interleave_0, values = (var_6777_cast_fp16, var_6779_cast_fp16, var_6781_cast_fp16, var_6783_cast_fp16))[name = string("op_6819_cast_fp16")]; + bool var_6821_interleave_0 = const()[name = string("op_6821_interleave_0"), val = bool(false)]; + tensor var_6821_cast_fp16 = concat(axis = var_5912, interleave = var_6821_interleave_0, values = (var_6785_cast_fp16, var_6787_cast_fp16, var_6789_cast_fp16, var_6791_cast_fp16))[name = string("op_6821_cast_fp16")]; + bool var_6823_interleave_0 = const()[name = string("op_6823_interleave_0"), val = bool(false)]; + tensor var_6823_cast_fp16 = concat(axis = var_5912, interleave = var_6823_interleave_0, values = (var_6793_cast_fp16, var_6795_cast_fp16, var_6797_cast_fp16, var_6799_cast_fp16))[name = string("op_6823_cast_fp16")]; + bool input_49_interleave_0 = const()[name = string("input_49_interleave_0"), val = bool(false)]; + tensor input_49_cast_fp16 = concat(axis = var_5929, interleave = input_49_interleave_0, values = (var_6801_cast_fp16, var_6803_cast_fp16, var_6805_cast_fp16, var_6807_cast_fp16, var_6809_cast_fp16, var_6811_cast_fp16, var_6813_cast_fp16, var_6815_cast_fp16, var_6817_cast_fp16, var_6819_cast_fp16, var_6821_cast_fp16, var_6823_cast_fp16))[name = string("input_49_cast_fp16")]; + string obj_27_pad_type_0 = const()[name = string("obj_27_pad_type_0"), val = string("valid")]; + tensor obj_27_strides_0 = const()[name = string("obj_27_strides_0"), val = tensor([1, 1])]; + tensor obj_27_pad_0 = const()[name = string("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_27_dilations_0 = const()[name = string("obj_27_dilations_0"), val = tensor([1, 1])]; + int32 obj_27_groups_0 = const()[name = string("obj_27_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94814784)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95994496)))]; + tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_27_dilations_0, groups = obj_27_groups_0, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = obj_27_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_6842_to_fp16 = const()[name = string("op_6842_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_6842_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = string("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95996096)))]; + tensor input_51_beta_0_to_fp16 = const()[name = string("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95997696)))]; + fp16 input_51_epsilon_0_to_fp16 = const()[name = string("input_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("input_51_cast_fp16")]; + string input_53_pad_type_0 = const()[name = string("input_53_pad_type_0"), val = string("valid")]; + tensor input_53_strides_0 = const()[name = string("input_53_strides_0"), val = tensor([1, 1])]; + tensor input_53_pad_0 = const()[name = string("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_53_dilations_0 = const()[name = string("input_53_dilations_0"), val = tensor([1, 1])]; + int32 input_53_groups_0 = const()[name = string("input_53_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95999296)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100717952)))]; + tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_53_dilations_0, groups = input_53_groups_0, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = input_53_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")]; + string input_55_mode_0 = const()[name = string("input_55_mode_0"), val = string("EXACT")]; + tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100724160)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105442816)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_55_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_29_cast_fp16")]; + int32 var_6871 = const()[name = string("op_6871"), val = int32(3)]; + int32 var_6888 = const()[name = string("op_6888"), val = int32(1)]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_6905_to_fp16 = const()[name = string("op_6905_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_6905_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = string("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105444416)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = string("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105446016)))]; + fp16 obj_29_epsilon_0_to_fp16 = const()[name = string("obj_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("obj_29_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105447616)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106627328)))]; + tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("query_15_cast_fp16")]; + string key_15_pad_type_0 = const()[name = string("key_15_pad_type_0"), val = string("valid")]; + tensor key_15_strides_0 = const()[name = string("key_15_strides_0"), val = tensor([1, 1])]; + tensor key_15_pad_0 = const()[name = string("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_15_dilations_0 = const()[name = string("key_15_dilations_0"), val = tensor([1, 1])]; + int32 key_15_groups_0 = const()[name = string("key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106628928)))]; + tensor key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("key_15_cast_fp16")]; + string value_15_pad_type_0 = const()[name = string("value_15_pad_type_0"), val = string("valid")]; + tensor value_15_strides_0 = const()[name = string("value_15_strides_0"), val = tensor([1, 1])]; + tensor value_15_pad_0 = const()[name = string("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_15_dilations_0 = const()[name = string("value_15_dilations_0"), val = tensor([1, 1])]; + int32 value_15_groups_0 = const()[name = string("value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107808640)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108988352)))]; + tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_6943_begin_0 = const()[name = string("op_6943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6943_end_0 = const()[name = string("op_6943_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6943_end_mask_0 = const()[name = string("op_6943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = query_15_cast_fp16)[name = string("op_6943_cast_fp16")]; + tensor var_6947_begin_0 = const()[name = string("op_6947_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6947_end_0 = const()[name = string("op_6947_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6947_end_mask_0 = const()[name = string("op_6947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = query_15_cast_fp16)[name = string("op_6947_cast_fp16")]; + tensor var_6951_begin_0 = const()[name = string("op_6951_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6951_end_0 = const()[name = string("op_6951_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6951_end_mask_0 = const()[name = string("op_6951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = query_15_cast_fp16)[name = string("op_6951_cast_fp16")]; + tensor var_6955_begin_0 = const()[name = string("op_6955_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6955_end_0 = const()[name = string("op_6955_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6955_end_mask_0 = const()[name = string("op_6955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = query_15_cast_fp16)[name = string("op_6955_cast_fp16")]; + tensor var_6959_begin_0 = const()[name = string("op_6959_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6959_end_0 = const()[name = string("op_6959_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6959_end_mask_0 = const()[name = string("op_6959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = query_15_cast_fp16)[name = string("op_6959_cast_fp16")]; + tensor var_6963_begin_0 = const()[name = string("op_6963_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6963_end_0 = const()[name = string("op_6963_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6963_end_mask_0 = const()[name = string("op_6963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = query_15_cast_fp16)[name = string("op_6963_cast_fp16")]; + tensor var_6967_begin_0 = const()[name = string("op_6967_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6967_end_0 = const()[name = string("op_6967_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6967_end_mask_0 = const()[name = string("op_6967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6967_cast_fp16 = slice_by_index(begin = var_6967_begin_0, end = var_6967_end_0, end_mask = var_6967_end_mask_0, x = query_15_cast_fp16)[name = string("op_6967_cast_fp16")]; + tensor var_6971_begin_0 = const()[name = string("op_6971_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6971_end_0 = const()[name = string("op_6971_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6971_end_mask_0 = const()[name = string("op_6971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6971_cast_fp16 = slice_by_index(begin = var_6971_begin_0, end = var_6971_end_0, end_mask = var_6971_end_mask_0, x = query_15_cast_fp16)[name = string("op_6971_cast_fp16")]; + tensor var_6975_begin_0 = const()[name = string("op_6975_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6975_end_0 = const()[name = string("op_6975_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6975_end_mask_0 = const()[name = string("op_6975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6975_cast_fp16 = slice_by_index(begin = var_6975_begin_0, end = var_6975_end_0, end_mask = var_6975_end_mask_0, x = query_15_cast_fp16)[name = string("op_6975_cast_fp16")]; + tensor var_6979_begin_0 = const()[name = string("op_6979_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6979_end_0 = const()[name = string("op_6979_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6979_end_mask_0 = const()[name = string("op_6979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6979_cast_fp16 = slice_by_index(begin = var_6979_begin_0, end = var_6979_end_0, end_mask = var_6979_end_mask_0, x = query_15_cast_fp16)[name = string("op_6979_cast_fp16")]; + tensor var_6983_begin_0 = const()[name = string("op_6983_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6983_end_0 = const()[name = string("op_6983_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6983_end_mask_0 = const()[name = string("op_6983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6983_cast_fp16 = slice_by_index(begin = var_6983_begin_0, end = var_6983_end_0, end_mask = var_6983_end_mask_0, x = query_15_cast_fp16)[name = string("op_6983_cast_fp16")]; + tensor var_6987_begin_0 = const()[name = string("op_6987_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6987_end_0 = const()[name = string("op_6987_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6987_end_mask_0 = const()[name = string("op_6987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6987_cast_fp16 = slice_by_index(begin = var_6987_begin_0, end = var_6987_end_0, end_mask = var_6987_end_mask_0, x = query_15_cast_fp16)[name = string("op_6987_cast_fp16")]; + tensor var_6996_begin_0 = const()[name = string("op_6996_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6996_end_0 = const()[name = string("op_6996_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6996_end_mask_0 = const()[name = string("op_6996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6996_cast_fp16 = slice_by_index(begin = var_6996_begin_0, end = var_6996_end_0, end_mask = var_6996_end_mask_0, x = var_6943_cast_fp16)[name = string("op_6996_cast_fp16")]; + tensor var_7003_begin_0 = const()[name = string("op_7003_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7003_end_0 = const()[name = string("op_7003_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7003_end_mask_0 = const()[name = string("op_7003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7003_cast_fp16 = slice_by_index(begin = var_7003_begin_0, end = var_7003_end_0, end_mask = var_7003_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7003_cast_fp16")]; + tensor var_7010_begin_0 = const()[name = string("op_7010_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7010_end_0 = const()[name = string("op_7010_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7010_end_mask_0 = const()[name = string("op_7010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7010_cast_fp16 = slice_by_index(begin = var_7010_begin_0, end = var_7010_end_0, end_mask = var_7010_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7010_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = string("op_7017_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7017_end_0 = const()[name = string("op_7017_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7017_end_mask_0 = const()[name = string("op_7017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = var_6943_cast_fp16)[name = string("op_7017_cast_fp16")]; + tensor var_7024_begin_0 = const()[name = string("op_7024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7024_end_0 = const()[name = string("op_7024_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7024_end_mask_0 = const()[name = string("op_7024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7024_cast_fp16 = slice_by_index(begin = var_7024_begin_0, end = var_7024_end_0, end_mask = var_7024_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7024_cast_fp16")]; + tensor var_7031_begin_0 = const()[name = string("op_7031_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7031_end_0 = const()[name = string("op_7031_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7031_end_mask_0 = const()[name = string("op_7031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7031_cast_fp16 = slice_by_index(begin = var_7031_begin_0, end = var_7031_end_0, end_mask = var_7031_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7031_cast_fp16")]; + tensor var_7038_begin_0 = const()[name = string("op_7038_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7038_end_0 = const()[name = string("op_7038_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7038_end_mask_0 = const()[name = string("op_7038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7038_cast_fp16 = slice_by_index(begin = var_7038_begin_0, end = var_7038_end_0, end_mask = var_7038_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7038_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = string("op_7045_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7045_end_0 = const()[name = string("op_7045_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7045_end_mask_0 = const()[name = string("op_7045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = var_6947_cast_fp16)[name = string("op_7045_cast_fp16")]; + tensor var_7052_begin_0 = const()[name = string("op_7052_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7052_end_0 = const()[name = string("op_7052_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7052_end_mask_0 = const()[name = string("op_7052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7052_cast_fp16 = slice_by_index(begin = var_7052_begin_0, end = var_7052_end_0, end_mask = var_7052_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7052_cast_fp16")]; + tensor var_7059_begin_0 = const()[name = string("op_7059_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7059_end_0 = const()[name = string("op_7059_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7059_end_mask_0 = const()[name = string("op_7059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7059_cast_fp16 = slice_by_index(begin = var_7059_begin_0, end = var_7059_end_0, end_mask = var_7059_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7059_cast_fp16")]; + tensor var_7066_begin_0 = const()[name = string("op_7066_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7066_end_0 = const()[name = string("op_7066_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7066_end_mask_0 = const()[name = string("op_7066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7066_cast_fp16 = slice_by_index(begin = var_7066_begin_0, end = var_7066_end_0, end_mask = var_7066_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7066_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = string("op_7073_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7073_end_0 = const()[name = string("op_7073_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7073_end_mask_0 = const()[name = string("op_7073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = var_6951_cast_fp16)[name = string("op_7073_cast_fp16")]; + tensor var_7080_begin_0 = const()[name = string("op_7080_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7080_end_0 = const()[name = string("op_7080_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7080_end_mask_0 = const()[name = string("op_7080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7080_cast_fp16 = slice_by_index(begin = var_7080_begin_0, end = var_7080_end_0, end_mask = var_7080_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7080_cast_fp16")]; + tensor var_7087_begin_0 = const()[name = string("op_7087_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7087_end_0 = const()[name = string("op_7087_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7087_end_mask_0 = const()[name = string("op_7087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7087_cast_fp16 = slice_by_index(begin = var_7087_begin_0, end = var_7087_end_0, end_mask = var_7087_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7087_cast_fp16")]; + tensor var_7094_begin_0 = const()[name = string("op_7094_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7094_end_0 = const()[name = string("op_7094_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7094_end_mask_0 = const()[name = string("op_7094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7094_cast_fp16 = slice_by_index(begin = var_7094_begin_0, end = var_7094_end_0, end_mask = var_7094_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7094_cast_fp16")]; + tensor var_7101_begin_0 = const()[name = string("op_7101_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7101_end_0 = const()[name = string("op_7101_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7101_end_mask_0 = const()[name = string("op_7101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7101_cast_fp16 = slice_by_index(begin = var_7101_begin_0, end = var_7101_end_0, end_mask = var_7101_end_mask_0, x = var_6955_cast_fp16)[name = string("op_7101_cast_fp16")]; + tensor var_7108_begin_0 = const()[name = string("op_7108_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7108_end_0 = const()[name = string("op_7108_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7108_end_mask_0 = const()[name = string("op_7108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7108_cast_fp16 = slice_by_index(begin = var_7108_begin_0, end = var_7108_end_0, end_mask = var_7108_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7108_cast_fp16")]; + tensor var_7115_begin_0 = const()[name = string("op_7115_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7115_end_0 = const()[name = string("op_7115_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7115_end_mask_0 = const()[name = string("op_7115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7115_cast_fp16 = slice_by_index(begin = var_7115_begin_0, end = var_7115_end_0, end_mask = var_7115_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7115_cast_fp16")]; + tensor var_7122_begin_0 = const()[name = string("op_7122_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7122_end_0 = const()[name = string("op_7122_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7122_end_mask_0 = const()[name = string("op_7122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7122_cast_fp16 = slice_by_index(begin = var_7122_begin_0, end = var_7122_end_0, end_mask = var_7122_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7122_cast_fp16")]; + tensor var_7129_begin_0 = const()[name = string("op_7129_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7129_end_0 = const()[name = string("op_7129_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7129_end_mask_0 = const()[name = string("op_7129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7129_cast_fp16 = slice_by_index(begin = var_7129_begin_0, end = var_7129_end_0, end_mask = var_7129_end_mask_0, x = var_6959_cast_fp16)[name = string("op_7129_cast_fp16")]; + tensor var_7136_begin_0 = const()[name = string("op_7136_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7136_end_0 = const()[name = string("op_7136_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7136_end_mask_0 = const()[name = string("op_7136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7136_cast_fp16 = slice_by_index(begin = var_7136_begin_0, end = var_7136_end_0, end_mask = var_7136_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7136_cast_fp16")]; + tensor var_7143_begin_0 = const()[name = string("op_7143_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7143_end_0 = const()[name = string("op_7143_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7143_end_mask_0 = const()[name = string("op_7143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7143_cast_fp16 = slice_by_index(begin = var_7143_begin_0, end = var_7143_end_0, end_mask = var_7143_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7143_cast_fp16")]; + tensor var_7150_begin_0 = const()[name = string("op_7150_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7150_end_0 = const()[name = string("op_7150_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7150_end_mask_0 = const()[name = string("op_7150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7150_cast_fp16 = slice_by_index(begin = var_7150_begin_0, end = var_7150_end_0, end_mask = var_7150_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7150_cast_fp16")]; + tensor var_7157_begin_0 = const()[name = string("op_7157_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7157_end_0 = const()[name = string("op_7157_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7157_end_mask_0 = const()[name = string("op_7157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7157_cast_fp16 = slice_by_index(begin = var_7157_begin_0, end = var_7157_end_0, end_mask = var_7157_end_mask_0, x = var_6963_cast_fp16)[name = string("op_7157_cast_fp16")]; + tensor var_7164_begin_0 = const()[name = string("op_7164_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7164_end_0 = const()[name = string("op_7164_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7164_end_mask_0 = const()[name = string("op_7164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7164_cast_fp16 = slice_by_index(begin = var_7164_begin_0, end = var_7164_end_0, end_mask = var_7164_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7164_cast_fp16")]; + tensor var_7171_begin_0 = const()[name = string("op_7171_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7171_end_0 = const()[name = string("op_7171_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7171_end_mask_0 = const()[name = string("op_7171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7171_cast_fp16 = slice_by_index(begin = var_7171_begin_0, end = var_7171_end_0, end_mask = var_7171_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7171_cast_fp16")]; + tensor var_7178_begin_0 = const()[name = string("op_7178_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7178_end_0 = const()[name = string("op_7178_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7178_end_mask_0 = const()[name = string("op_7178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7178_cast_fp16 = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7178_cast_fp16")]; + tensor var_7185_begin_0 = const()[name = string("op_7185_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7185_end_0 = const()[name = string("op_7185_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7185_end_mask_0 = const()[name = string("op_7185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7185_cast_fp16 = slice_by_index(begin = var_7185_begin_0, end = var_7185_end_0, end_mask = var_7185_end_mask_0, x = var_6967_cast_fp16)[name = string("op_7185_cast_fp16")]; + tensor var_7192_begin_0 = const()[name = string("op_7192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7192_end_0 = const()[name = string("op_7192_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7192_end_mask_0 = const()[name = string("op_7192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7192_cast_fp16 = slice_by_index(begin = var_7192_begin_0, end = var_7192_end_0, end_mask = var_7192_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7192_cast_fp16")]; + tensor var_7199_begin_0 = const()[name = string("op_7199_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7199_end_0 = const()[name = string("op_7199_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7199_end_mask_0 = const()[name = string("op_7199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7199_cast_fp16 = slice_by_index(begin = var_7199_begin_0, end = var_7199_end_0, end_mask = var_7199_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7199_cast_fp16")]; + tensor var_7206_begin_0 = const()[name = string("op_7206_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7206_end_0 = const()[name = string("op_7206_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7206_end_mask_0 = const()[name = string("op_7206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7206_cast_fp16 = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7206_cast_fp16")]; + tensor var_7213_begin_0 = const()[name = string("op_7213_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7213_end_0 = const()[name = string("op_7213_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7213_end_mask_0 = const()[name = string("op_7213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7213_cast_fp16 = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = var_6971_cast_fp16)[name = string("op_7213_cast_fp16")]; + tensor var_7220_begin_0 = const()[name = string("op_7220_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7220_end_0 = const()[name = string("op_7220_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7220_end_mask_0 = const()[name = string("op_7220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7220_cast_fp16 = slice_by_index(begin = var_7220_begin_0, end = var_7220_end_0, end_mask = var_7220_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7220_cast_fp16")]; + tensor var_7227_begin_0 = const()[name = string("op_7227_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7227_end_0 = const()[name = string("op_7227_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7227_end_mask_0 = const()[name = string("op_7227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7227_cast_fp16 = slice_by_index(begin = var_7227_begin_0, end = var_7227_end_0, end_mask = var_7227_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7227_cast_fp16")]; + tensor var_7234_begin_0 = const()[name = string("op_7234_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7234_end_0 = const()[name = string("op_7234_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7234_end_mask_0 = const()[name = string("op_7234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7234_cast_fp16 = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7234_cast_fp16")]; + tensor var_7241_begin_0 = const()[name = string("op_7241_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7241_end_0 = const()[name = string("op_7241_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7241_end_mask_0 = const()[name = string("op_7241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7241_cast_fp16 = slice_by_index(begin = var_7241_begin_0, end = var_7241_end_0, end_mask = var_7241_end_mask_0, x = var_6975_cast_fp16)[name = string("op_7241_cast_fp16")]; + tensor var_7248_begin_0 = const()[name = string("op_7248_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7248_end_0 = const()[name = string("op_7248_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7248_end_mask_0 = const()[name = string("op_7248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7248_cast_fp16 = slice_by_index(begin = var_7248_begin_0, end = var_7248_end_0, end_mask = var_7248_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7248_cast_fp16")]; + tensor var_7255_begin_0 = const()[name = string("op_7255_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7255_end_0 = const()[name = string("op_7255_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7255_end_mask_0 = const()[name = string("op_7255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7255_cast_fp16 = slice_by_index(begin = var_7255_begin_0, end = var_7255_end_0, end_mask = var_7255_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7255_cast_fp16")]; + tensor var_7262_begin_0 = const()[name = string("op_7262_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7262_end_0 = const()[name = string("op_7262_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7262_end_mask_0 = const()[name = string("op_7262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7262_cast_fp16 = slice_by_index(begin = var_7262_begin_0, end = var_7262_end_0, end_mask = var_7262_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7262_cast_fp16")]; + tensor var_7269_begin_0 = const()[name = string("op_7269_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7269_end_0 = const()[name = string("op_7269_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7269_end_mask_0 = const()[name = string("op_7269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7269_cast_fp16 = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = var_6979_cast_fp16)[name = string("op_7269_cast_fp16")]; + tensor var_7276_begin_0 = const()[name = string("op_7276_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7276_end_0 = const()[name = string("op_7276_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7276_end_mask_0 = const()[name = string("op_7276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7276_cast_fp16 = slice_by_index(begin = var_7276_begin_0, end = var_7276_end_0, end_mask = var_7276_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7276_cast_fp16")]; + tensor var_7283_begin_0 = const()[name = string("op_7283_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7283_end_0 = const()[name = string("op_7283_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7283_end_mask_0 = const()[name = string("op_7283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7283_cast_fp16 = slice_by_index(begin = var_7283_begin_0, end = var_7283_end_0, end_mask = var_7283_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7283_cast_fp16")]; + tensor var_7290_begin_0 = const()[name = string("op_7290_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7290_end_0 = const()[name = string("op_7290_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7290_end_mask_0 = const()[name = string("op_7290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7290_cast_fp16 = slice_by_index(begin = var_7290_begin_0, end = var_7290_end_0, end_mask = var_7290_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7290_cast_fp16")]; + tensor var_7297_begin_0 = const()[name = string("op_7297_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7297_end_0 = const()[name = string("op_7297_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7297_end_mask_0 = const()[name = string("op_7297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7297_cast_fp16 = slice_by_index(begin = var_7297_begin_0, end = var_7297_end_0, end_mask = var_7297_end_mask_0, x = var_6983_cast_fp16)[name = string("op_7297_cast_fp16")]; + tensor var_7304_begin_0 = const()[name = string("op_7304_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7304_end_0 = const()[name = string("op_7304_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7304_end_mask_0 = const()[name = string("op_7304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7304_cast_fp16 = slice_by_index(begin = var_7304_begin_0, end = var_7304_end_0, end_mask = var_7304_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7304_cast_fp16")]; + tensor var_7311_begin_0 = const()[name = string("op_7311_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7311_end_0 = const()[name = string("op_7311_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7311_end_mask_0 = const()[name = string("op_7311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7311_cast_fp16 = slice_by_index(begin = var_7311_begin_0, end = var_7311_end_0, end_mask = var_7311_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7311_cast_fp16")]; + tensor var_7318_begin_0 = const()[name = string("op_7318_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7318_end_0 = const()[name = string("op_7318_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7318_end_mask_0 = const()[name = string("op_7318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7318_cast_fp16 = slice_by_index(begin = var_7318_begin_0, end = var_7318_end_0, end_mask = var_7318_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7318_cast_fp16")]; + tensor var_7325_begin_0 = const()[name = string("op_7325_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7325_end_0 = const()[name = string("op_7325_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7325_end_mask_0 = const()[name = string("op_7325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7325_cast_fp16 = slice_by_index(begin = var_7325_begin_0, end = var_7325_end_0, end_mask = var_7325_end_mask_0, x = var_6987_cast_fp16)[name = string("op_7325_cast_fp16")]; + tensor k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7330_begin_0 = const()[name = string("op_7330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7330_end_0 = const()[name = string("op_7330_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_7330_end_mask_0 = const()[name = string("op_7330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = string("transpose_4")]; + tensor var_7330_cast_fp16 = slice_by_index(begin = var_7330_begin_0, end = var_7330_end_0, end_mask = var_7330_end_mask_0, x = k_15_cast_fp16)[name = string("op_7330_cast_fp16")]; + tensor var_7334_begin_0 = const()[name = string("op_7334_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7334_end_0 = const()[name = string("op_7334_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_7334_end_mask_0 = const()[name = string("op_7334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7334_cast_fp16 = slice_by_index(begin = var_7334_begin_0, end = var_7334_end_0, end_mask = var_7334_end_mask_0, x = k_15_cast_fp16)[name = string("op_7334_cast_fp16")]; + tensor var_7338_begin_0 = const()[name = string("op_7338_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7338_end_0 = const()[name = string("op_7338_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_7338_end_mask_0 = const()[name = string("op_7338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7338_cast_fp16 = slice_by_index(begin = var_7338_begin_0, end = var_7338_end_0, end_mask = var_7338_end_mask_0, x = k_15_cast_fp16)[name = string("op_7338_cast_fp16")]; + tensor var_7342_begin_0 = const()[name = string("op_7342_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7342_end_0 = const()[name = string("op_7342_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_7342_end_mask_0 = const()[name = string("op_7342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7342_cast_fp16 = slice_by_index(begin = var_7342_begin_0, end = var_7342_end_0, end_mask = var_7342_end_mask_0, x = k_15_cast_fp16)[name = string("op_7342_cast_fp16")]; + tensor var_7346_begin_0 = const()[name = string("op_7346_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7346_end_0 = const()[name = string("op_7346_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_7346_end_mask_0 = const()[name = string("op_7346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7346_cast_fp16 = slice_by_index(begin = var_7346_begin_0, end = var_7346_end_0, end_mask = var_7346_end_mask_0, x = k_15_cast_fp16)[name = string("op_7346_cast_fp16")]; + tensor var_7350_begin_0 = const()[name = string("op_7350_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7350_end_0 = const()[name = string("op_7350_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_7350_end_mask_0 = const()[name = string("op_7350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7350_cast_fp16 = slice_by_index(begin = var_7350_begin_0, end = var_7350_end_0, end_mask = var_7350_end_mask_0, x = k_15_cast_fp16)[name = string("op_7350_cast_fp16")]; + tensor var_7354_begin_0 = const()[name = string("op_7354_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7354_end_0 = const()[name = string("op_7354_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_7354_end_mask_0 = const()[name = string("op_7354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7354_cast_fp16 = slice_by_index(begin = var_7354_begin_0, end = var_7354_end_0, end_mask = var_7354_end_mask_0, x = k_15_cast_fp16)[name = string("op_7354_cast_fp16")]; + tensor var_7358_begin_0 = const()[name = string("op_7358_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7358_end_0 = const()[name = string("op_7358_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_7358_end_mask_0 = const()[name = string("op_7358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7358_cast_fp16 = slice_by_index(begin = var_7358_begin_0, end = var_7358_end_0, end_mask = var_7358_end_mask_0, x = k_15_cast_fp16)[name = string("op_7358_cast_fp16")]; + tensor var_7362_begin_0 = const()[name = string("op_7362_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7362_end_0 = const()[name = string("op_7362_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_7362_end_mask_0 = const()[name = string("op_7362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7362_cast_fp16 = slice_by_index(begin = var_7362_begin_0, end = var_7362_end_0, end_mask = var_7362_end_mask_0, x = k_15_cast_fp16)[name = string("op_7362_cast_fp16")]; + tensor var_7366_begin_0 = const()[name = string("op_7366_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7366_end_0 = const()[name = string("op_7366_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_7366_end_mask_0 = const()[name = string("op_7366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7366_cast_fp16 = slice_by_index(begin = var_7366_begin_0, end = var_7366_end_0, end_mask = var_7366_end_mask_0, x = k_15_cast_fp16)[name = string("op_7366_cast_fp16")]; + tensor var_7370_begin_0 = const()[name = string("op_7370_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7370_end_0 = const()[name = string("op_7370_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_7370_end_mask_0 = const()[name = string("op_7370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7370_cast_fp16 = slice_by_index(begin = var_7370_begin_0, end = var_7370_end_0, end_mask = var_7370_end_mask_0, x = k_15_cast_fp16)[name = string("op_7370_cast_fp16")]; + tensor var_7374_begin_0 = const()[name = string("op_7374_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7374_end_0 = const()[name = string("op_7374_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_7374_end_mask_0 = const()[name = string("op_7374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7374_cast_fp16 = slice_by_index(begin = var_7374_begin_0, end = var_7374_end_0, end_mask = var_7374_end_mask_0, x = k_15_cast_fp16)[name = string("op_7374_cast_fp16")]; + tensor var_7376_begin_0 = const()[name = string("op_7376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7376_end_0 = const()[name = string("op_7376_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7376_end_mask_0 = const()[name = string("op_7376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7376_cast_fp16 = slice_by_index(begin = var_7376_begin_0, end = var_7376_end_0, end_mask = var_7376_end_mask_0, x = value_15_cast_fp16)[name = string("op_7376_cast_fp16")]; + tensor var_7380_begin_0 = const()[name = string("op_7380_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7380_end_0 = const()[name = string("op_7380_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7380_end_mask_0 = const()[name = string("op_7380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7380_cast_fp16 = slice_by_index(begin = var_7380_begin_0, end = var_7380_end_0, end_mask = var_7380_end_mask_0, x = value_15_cast_fp16)[name = string("op_7380_cast_fp16")]; + tensor var_7384_begin_0 = const()[name = string("op_7384_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7384_end_0 = const()[name = string("op_7384_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7384_end_mask_0 = const()[name = string("op_7384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7384_cast_fp16 = slice_by_index(begin = var_7384_begin_0, end = var_7384_end_0, end_mask = var_7384_end_mask_0, x = value_15_cast_fp16)[name = string("op_7384_cast_fp16")]; + tensor var_7388_begin_0 = const()[name = string("op_7388_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7388_end_0 = const()[name = string("op_7388_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7388_end_mask_0 = const()[name = string("op_7388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7388_cast_fp16 = slice_by_index(begin = var_7388_begin_0, end = var_7388_end_0, end_mask = var_7388_end_mask_0, x = value_15_cast_fp16)[name = string("op_7388_cast_fp16")]; + tensor var_7392_begin_0 = const()[name = string("op_7392_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7392_end_0 = const()[name = string("op_7392_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7392_end_mask_0 = const()[name = string("op_7392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7392_cast_fp16 = slice_by_index(begin = var_7392_begin_0, end = var_7392_end_0, end_mask = var_7392_end_mask_0, x = value_15_cast_fp16)[name = string("op_7392_cast_fp16")]; + tensor var_7396_begin_0 = const()[name = string("op_7396_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7396_end_0 = const()[name = string("op_7396_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7396_end_mask_0 = const()[name = string("op_7396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7396_cast_fp16 = slice_by_index(begin = var_7396_begin_0, end = var_7396_end_0, end_mask = var_7396_end_mask_0, x = value_15_cast_fp16)[name = string("op_7396_cast_fp16")]; + tensor var_7400_begin_0 = const()[name = string("op_7400_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7400_end_0 = const()[name = string("op_7400_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7400_end_mask_0 = const()[name = string("op_7400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7400_cast_fp16 = slice_by_index(begin = var_7400_begin_0, end = var_7400_end_0, end_mask = var_7400_end_mask_0, x = value_15_cast_fp16)[name = string("op_7400_cast_fp16")]; + tensor var_7404_begin_0 = const()[name = string("op_7404_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7404_end_0 = const()[name = string("op_7404_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7404_end_mask_0 = const()[name = string("op_7404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7404_cast_fp16 = slice_by_index(begin = var_7404_begin_0, end = var_7404_end_0, end_mask = var_7404_end_mask_0, x = value_15_cast_fp16)[name = string("op_7404_cast_fp16")]; + tensor var_7408_begin_0 = const()[name = string("op_7408_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7408_end_0 = const()[name = string("op_7408_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7408_end_mask_0 = const()[name = string("op_7408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7408_cast_fp16 = slice_by_index(begin = var_7408_begin_0, end = var_7408_end_0, end_mask = var_7408_end_mask_0, x = value_15_cast_fp16)[name = string("op_7408_cast_fp16")]; + tensor var_7412_begin_0 = const()[name = string("op_7412_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7412_end_0 = const()[name = string("op_7412_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7412_end_mask_0 = const()[name = string("op_7412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7412_cast_fp16 = slice_by_index(begin = var_7412_begin_0, end = var_7412_end_0, end_mask = var_7412_end_mask_0, x = value_15_cast_fp16)[name = string("op_7412_cast_fp16")]; + tensor var_7416_begin_0 = const()[name = string("op_7416_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7416_end_0 = const()[name = string("op_7416_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7416_end_mask_0 = const()[name = string("op_7416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7416_cast_fp16 = slice_by_index(begin = var_7416_begin_0, end = var_7416_end_0, end_mask = var_7416_end_mask_0, x = value_15_cast_fp16)[name = string("op_7416_cast_fp16")]; + tensor var_7420_begin_0 = const()[name = string("op_7420_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7420_end_0 = const()[name = string("op_7420_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7420_end_mask_0 = const()[name = string("op_7420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7420_cast_fp16 = slice_by_index(begin = var_7420_begin_0, end = var_7420_end_0, end_mask = var_7420_end_mask_0, x = value_15_cast_fp16)[name = string("op_7420_cast_fp16")]; + string _SplitHeadsQ__mh_w_673_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_673_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_673_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_673_equation_0, values = (var_7330_cast_fp16, var_6996_cast_fp16))[name = string("_SplitHeadsQ__mh_w_673_cast_fp16")]; + string _SplitHeadsQ__mh_w_675_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_675_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_675_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_675_equation_0, values = (var_7330_cast_fp16, var_7003_cast_fp16))[name = string("_SplitHeadsQ__mh_w_675_cast_fp16")]; + string _SplitHeadsQ__mh_w_677_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_677_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_677_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_677_equation_0, values = (var_7330_cast_fp16, var_7010_cast_fp16))[name = string("_SplitHeadsQ__mh_w_677_cast_fp16")]; + string _SplitHeadsQ__mh_w_679_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_679_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_679_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_679_equation_0, values = (var_7330_cast_fp16, var_7017_cast_fp16))[name = string("_SplitHeadsQ__mh_w_679_cast_fp16")]; + string _SplitHeadsQ__mh_w_681_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_681_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_681_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_681_equation_0, values = (var_7334_cast_fp16, var_7024_cast_fp16))[name = string("_SplitHeadsQ__mh_w_681_cast_fp16")]; + string _SplitHeadsQ__mh_w_683_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_683_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_683_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_683_equation_0, values = (var_7334_cast_fp16, var_7031_cast_fp16))[name = string("_SplitHeadsQ__mh_w_683_cast_fp16")]; + string _SplitHeadsQ__mh_w_685_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_685_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_685_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_685_equation_0, values = (var_7334_cast_fp16, var_7038_cast_fp16))[name = string("_SplitHeadsQ__mh_w_685_cast_fp16")]; + string _SplitHeadsQ__mh_w_687_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_687_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_687_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_687_equation_0, values = (var_7334_cast_fp16, var_7045_cast_fp16))[name = string("_SplitHeadsQ__mh_w_687_cast_fp16")]; + string _SplitHeadsQ__mh_w_689_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_689_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_689_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_689_equation_0, values = (var_7338_cast_fp16, var_7052_cast_fp16))[name = string("_SplitHeadsQ__mh_w_689_cast_fp16")]; + string _SplitHeadsQ__mh_w_691_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_691_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_691_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_691_equation_0, values = (var_7338_cast_fp16, var_7059_cast_fp16))[name = string("_SplitHeadsQ__mh_w_691_cast_fp16")]; + string _SplitHeadsQ__mh_w_693_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_693_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_693_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_693_equation_0, values = (var_7338_cast_fp16, var_7066_cast_fp16))[name = string("_SplitHeadsQ__mh_w_693_cast_fp16")]; + string _SplitHeadsQ__mh_w_695_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_695_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_695_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_695_equation_0, values = (var_7338_cast_fp16, var_7073_cast_fp16))[name = string("_SplitHeadsQ__mh_w_695_cast_fp16")]; + string _SplitHeadsQ__mh_w_697_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_697_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_697_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_697_equation_0, values = (var_7342_cast_fp16, var_7080_cast_fp16))[name = string("_SplitHeadsQ__mh_w_697_cast_fp16")]; + string _SplitHeadsQ__mh_w_699_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_699_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_699_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_699_equation_0, values = (var_7342_cast_fp16, var_7087_cast_fp16))[name = string("_SplitHeadsQ__mh_w_699_cast_fp16")]; + string _SplitHeadsQ__mh_w_701_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_701_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_701_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_701_equation_0, values = (var_7342_cast_fp16, var_7094_cast_fp16))[name = string("_SplitHeadsQ__mh_w_701_cast_fp16")]; + string _SplitHeadsQ__mh_w_703_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_703_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_703_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_703_equation_0, values = (var_7342_cast_fp16, var_7101_cast_fp16))[name = string("_SplitHeadsQ__mh_w_703_cast_fp16")]; + string _SplitHeadsQ__mh_w_705_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_705_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_705_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_705_equation_0, values = (var_7346_cast_fp16, var_7108_cast_fp16))[name = string("_SplitHeadsQ__mh_w_705_cast_fp16")]; + string _SplitHeadsQ__mh_w_707_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_707_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_707_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_707_equation_0, values = (var_7346_cast_fp16, var_7115_cast_fp16))[name = string("_SplitHeadsQ__mh_w_707_cast_fp16")]; + string _SplitHeadsQ__mh_w_709_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_709_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_709_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_709_equation_0, values = (var_7346_cast_fp16, var_7122_cast_fp16))[name = string("_SplitHeadsQ__mh_w_709_cast_fp16")]; + string _SplitHeadsQ__mh_w_711_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_711_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_711_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_711_equation_0, values = (var_7346_cast_fp16, var_7129_cast_fp16))[name = string("_SplitHeadsQ__mh_w_711_cast_fp16")]; + string _SplitHeadsQ__mh_w_713_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_713_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_713_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_713_equation_0, values = (var_7350_cast_fp16, var_7136_cast_fp16))[name = string("_SplitHeadsQ__mh_w_713_cast_fp16")]; + string _SplitHeadsQ__mh_w_715_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_715_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_715_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_715_equation_0, values = (var_7350_cast_fp16, var_7143_cast_fp16))[name = string("_SplitHeadsQ__mh_w_715_cast_fp16")]; + string _SplitHeadsQ__mh_w_717_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_717_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_717_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_717_equation_0, values = (var_7350_cast_fp16, var_7150_cast_fp16))[name = string("_SplitHeadsQ__mh_w_717_cast_fp16")]; + string _SplitHeadsQ__mh_w_719_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_719_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_719_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_719_equation_0, values = (var_7350_cast_fp16, var_7157_cast_fp16))[name = string("_SplitHeadsQ__mh_w_719_cast_fp16")]; + string _SplitHeadsQ__mh_w_721_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_721_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_721_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_721_equation_0, values = (var_7354_cast_fp16, var_7164_cast_fp16))[name = string("_SplitHeadsQ__mh_w_721_cast_fp16")]; + string _SplitHeadsQ__mh_w_723_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_723_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_723_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_723_equation_0, values = (var_7354_cast_fp16, var_7171_cast_fp16))[name = string("_SplitHeadsQ__mh_w_723_cast_fp16")]; + string _SplitHeadsQ__mh_w_725_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_725_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_725_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_725_equation_0, values = (var_7354_cast_fp16, var_7178_cast_fp16))[name = string("_SplitHeadsQ__mh_w_725_cast_fp16")]; + string _SplitHeadsQ__mh_w_727_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_727_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_727_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_727_equation_0, values = (var_7354_cast_fp16, var_7185_cast_fp16))[name = string("_SplitHeadsQ__mh_w_727_cast_fp16")]; + string _SplitHeadsQ__mh_w_729_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_729_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_729_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_729_equation_0, values = (var_7358_cast_fp16, var_7192_cast_fp16))[name = string("_SplitHeadsQ__mh_w_729_cast_fp16")]; + string _SplitHeadsQ__mh_w_731_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_731_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_731_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_731_equation_0, values = (var_7358_cast_fp16, var_7199_cast_fp16))[name = string("_SplitHeadsQ__mh_w_731_cast_fp16")]; + string _SplitHeadsQ__mh_w_733_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_733_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_733_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_733_equation_0, values = (var_7358_cast_fp16, var_7206_cast_fp16))[name = string("_SplitHeadsQ__mh_w_733_cast_fp16")]; + string _SplitHeadsQ__mh_w_735_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_735_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_735_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_735_equation_0, values = (var_7358_cast_fp16, var_7213_cast_fp16))[name = string("_SplitHeadsQ__mh_w_735_cast_fp16")]; + string _SplitHeadsQ__mh_w_737_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_737_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_737_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_737_equation_0, values = (var_7362_cast_fp16, var_7220_cast_fp16))[name = string("_SplitHeadsQ__mh_w_737_cast_fp16")]; + string _SplitHeadsQ__mh_w_739_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_739_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_739_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_739_equation_0, values = (var_7362_cast_fp16, var_7227_cast_fp16))[name = string("_SplitHeadsQ__mh_w_739_cast_fp16")]; + string _SplitHeadsQ__mh_w_741_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_741_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_741_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_741_equation_0, values = (var_7362_cast_fp16, var_7234_cast_fp16))[name = string("_SplitHeadsQ__mh_w_741_cast_fp16")]; + string _SplitHeadsQ__mh_w_743_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_743_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_743_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_743_equation_0, values = (var_7362_cast_fp16, var_7241_cast_fp16))[name = string("_SplitHeadsQ__mh_w_743_cast_fp16")]; + string _SplitHeadsQ__mh_w_745_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_745_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_745_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_745_equation_0, values = (var_7366_cast_fp16, var_7248_cast_fp16))[name = string("_SplitHeadsQ__mh_w_745_cast_fp16")]; + string _SplitHeadsQ__mh_w_747_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_747_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_747_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_747_equation_0, values = (var_7366_cast_fp16, var_7255_cast_fp16))[name = string("_SplitHeadsQ__mh_w_747_cast_fp16")]; + string _SplitHeadsQ__mh_w_749_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_749_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_749_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_749_equation_0, values = (var_7366_cast_fp16, var_7262_cast_fp16))[name = string("_SplitHeadsQ__mh_w_749_cast_fp16")]; + string _SplitHeadsQ__mh_w_751_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_751_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_751_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_751_equation_0, values = (var_7366_cast_fp16, var_7269_cast_fp16))[name = string("_SplitHeadsQ__mh_w_751_cast_fp16")]; + string _SplitHeadsQ__mh_w_753_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_753_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_753_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_753_equation_0, values = (var_7370_cast_fp16, var_7276_cast_fp16))[name = string("_SplitHeadsQ__mh_w_753_cast_fp16")]; + string _SplitHeadsQ__mh_w_755_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_755_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_755_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_755_equation_0, values = (var_7370_cast_fp16, var_7283_cast_fp16))[name = string("_SplitHeadsQ__mh_w_755_cast_fp16")]; + string _SplitHeadsQ__mh_w_757_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_757_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_757_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_757_equation_0, values = (var_7370_cast_fp16, var_7290_cast_fp16))[name = string("_SplitHeadsQ__mh_w_757_cast_fp16")]; + string _SplitHeadsQ__mh_w_759_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_759_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_759_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_759_equation_0, values = (var_7370_cast_fp16, var_7297_cast_fp16))[name = string("_SplitHeadsQ__mh_w_759_cast_fp16")]; + string _SplitHeadsQ__mh_w_761_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_761_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_761_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_761_equation_0, values = (var_7374_cast_fp16, var_7304_cast_fp16))[name = string("_SplitHeadsQ__mh_w_761_cast_fp16")]; + string _SplitHeadsQ__mh_w_763_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_763_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_763_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_763_equation_0, values = (var_7374_cast_fp16, var_7311_cast_fp16))[name = string("_SplitHeadsQ__mh_w_763_cast_fp16")]; + string _SplitHeadsQ__mh_w_765_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_765_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_765_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_765_equation_0, values = (var_7374_cast_fp16, var_7318_cast_fp16))[name = string("_SplitHeadsQ__mh_w_765_cast_fp16")]; + string _SplitHeadsQ__mh_w_767_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_767_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_767_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_767_equation_0, values = (var_7374_cast_fp16, var_7325_cast_fp16))[name = string("_SplitHeadsQ__mh_w_767_cast_fp16")]; + fp16 var_7519_to_fp16 = const()[name = string("op_7519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_673_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_673_cast_fp16, y = var_7519_to_fp16)[name = string("aw_chunk_673_cast_fp16")]; + fp16 var_7521_to_fp16 = const()[name = string("op_7521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_675_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_675_cast_fp16, y = var_7521_to_fp16)[name = string("aw_chunk_675_cast_fp16")]; + fp16 var_7523_to_fp16 = const()[name = string("op_7523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_677_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_677_cast_fp16, y = var_7523_to_fp16)[name = string("aw_chunk_677_cast_fp16")]; + fp16 var_7525_to_fp16 = const()[name = string("op_7525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_679_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_679_cast_fp16, y = var_7525_to_fp16)[name = string("aw_chunk_679_cast_fp16")]; + fp16 var_7527_to_fp16 = const()[name = string("op_7527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_681_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_681_cast_fp16, y = var_7527_to_fp16)[name = string("aw_chunk_681_cast_fp16")]; + fp16 var_7529_to_fp16 = const()[name = string("op_7529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_683_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_683_cast_fp16, y = var_7529_to_fp16)[name = string("aw_chunk_683_cast_fp16")]; + fp16 var_7531_to_fp16 = const()[name = string("op_7531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_685_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_685_cast_fp16, y = var_7531_to_fp16)[name = string("aw_chunk_685_cast_fp16")]; + fp16 var_7533_to_fp16 = const()[name = string("op_7533_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_687_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_687_cast_fp16, y = var_7533_to_fp16)[name = string("aw_chunk_687_cast_fp16")]; + fp16 var_7535_to_fp16 = const()[name = string("op_7535_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_689_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_689_cast_fp16, y = var_7535_to_fp16)[name = string("aw_chunk_689_cast_fp16")]; + fp16 var_7537_to_fp16 = const()[name = string("op_7537_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_691_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_691_cast_fp16, y = var_7537_to_fp16)[name = string("aw_chunk_691_cast_fp16")]; + fp16 var_7539_to_fp16 = const()[name = string("op_7539_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_693_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_693_cast_fp16, y = var_7539_to_fp16)[name = string("aw_chunk_693_cast_fp16")]; + fp16 var_7541_to_fp16 = const()[name = string("op_7541_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_695_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_695_cast_fp16, y = var_7541_to_fp16)[name = string("aw_chunk_695_cast_fp16")]; + fp16 var_7543_to_fp16 = const()[name = string("op_7543_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_697_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_697_cast_fp16, y = var_7543_to_fp16)[name = string("aw_chunk_697_cast_fp16")]; + fp16 var_7545_to_fp16 = const()[name = string("op_7545_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_699_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_699_cast_fp16, y = var_7545_to_fp16)[name = string("aw_chunk_699_cast_fp16")]; + fp16 var_7547_to_fp16 = const()[name = string("op_7547_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_701_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_701_cast_fp16, y = var_7547_to_fp16)[name = string("aw_chunk_701_cast_fp16")]; + fp16 var_7549_to_fp16 = const()[name = string("op_7549_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_703_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_703_cast_fp16, y = var_7549_to_fp16)[name = string("aw_chunk_703_cast_fp16")]; + fp16 var_7551_to_fp16 = const()[name = string("op_7551_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_705_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_705_cast_fp16, y = var_7551_to_fp16)[name = string("aw_chunk_705_cast_fp16")]; + fp16 var_7553_to_fp16 = const()[name = string("op_7553_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_707_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_707_cast_fp16, y = var_7553_to_fp16)[name = string("aw_chunk_707_cast_fp16")]; + fp16 var_7555_to_fp16 = const()[name = string("op_7555_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_709_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_709_cast_fp16, y = var_7555_to_fp16)[name = string("aw_chunk_709_cast_fp16")]; + fp16 var_7557_to_fp16 = const()[name = string("op_7557_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_711_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_711_cast_fp16, y = var_7557_to_fp16)[name = string("aw_chunk_711_cast_fp16")]; + fp16 var_7559_to_fp16 = const()[name = string("op_7559_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_713_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_713_cast_fp16, y = var_7559_to_fp16)[name = string("aw_chunk_713_cast_fp16")]; + fp16 var_7561_to_fp16 = const()[name = string("op_7561_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_715_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_715_cast_fp16, y = var_7561_to_fp16)[name = string("aw_chunk_715_cast_fp16")]; + fp16 var_7563_to_fp16 = const()[name = string("op_7563_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_717_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_717_cast_fp16, y = var_7563_to_fp16)[name = string("aw_chunk_717_cast_fp16")]; + fp16 var_7565_to_fp16 = const()[name = string("op_7565_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_719_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_719_cast_fp16, y = var_7565_to_fp16)[name = string("aw_chunk_719_cast_fp16")]; + fp16 var_7567_to_fp16 = const()[name = string("op_7567_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_721_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_721_cast_fp16, y = var_7567_to_fp16)[name = string("aw_chunk_721_cast_fp16")]; + fp16 var_7569_to_fp16 = const()[name = string("op_7569_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_723_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_723_cast_fp16, y = var_7569_to_fp16)[name = string("aw_chunk_723_cast_fp16")]; + fp16 var_7571_to_fp16 = const()[name = string("op_7571_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_725_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_725_cast_fp16, y = var_7571_to_fp16)[name = string("aw_chunk_725_cast_fp16")]; + fp16 var_7573_to_fp16 = const()[name = string("op_7573_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_727_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_727_cast_fp16, y = var_7573_to_fp16)[name = string("aw_chunk_727_cast_fp16")]; + fp16 var_7575_to_fp16 = const()[name = string("op_7575_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_729_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_729_cast_fp16, y = var_7575_to_fp16)[name = string("aw_chunk_729_cast_fp16")]; + fp16 var_7577_to_fp16 = const()[name = string("op_7577_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_731_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_731_cast_fp16, y = var_7577_to_fp16)[name = string("aw_chunk_731_cast_fp16")]; + fp16 var_7579_to_fp16 = const()[name = string("op_7579_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_733_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_733_cast_fp16, y = var_7579_to_fp16)[name = string("aw_chunk_733_cast_fp16")]; + fp16 var_7581_to_fp16 = const()[name = string("op_7581_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_735_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_735_cast_fp16, y = var_7581_to_fp16)[name = string("aw_chunk_735_cast_fp16")]; + fp16 var_7583_to_fp16 = const()[name = string("op_7583_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_737_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_737_cast_fp16, y = var_7583_to_fp16)[name = string("aw_chunk_737_cast_fp16")]; + fp16 var_7585_to_fp16 = const()[name = string("op_7585_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_739_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_739_cast_fp16, y = var_7585_to_fp16)[name = string("aw_chunk_739_cast_fp16")]; + fp16 var_7587_to_fp16 = const()[name = string("op_7587_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_741_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_741_cast_fp16, y = var_7587_to_fp16)[name = string("aw_chunk_741_cast_fp16")]; + fp16 var_7589_to_fp16 = const()[name = string("op_7589_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_743_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_743_cast_fp16, y = var_7589_to_fp16)[name = string("aw_chunk_743_cast_fp16")]; + fp16 var_7591_to_fp16 = const()[name = string("op_7591_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_745_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_745_cast_fp16, y = var_7591_to_fp16)[name = string("aw_chunk_745_cast_fp16")]; + fp16 var_7593_to_fp16 = const()[name = string("op_7593_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_747_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_747_cast_fp16, y = var_7593_to_fp16)[name = string("aw_chunk_747_cast_fp16")]; + fp16 var_7595_to_fp16 = const()[name = string("op_7595_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_749_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_749_cast_fp16, y = var_7595_to_fp16)[name = string("aw_chunk_749_cast_fp16")]; + fp16 var_7597_to_fp16 = const()[name = string("op_7597_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_751_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_751_cast_fp16, y = var_7597_to_fp16)[name = string("aw_chunk_751_cast_fp16")]; + fp16 var_7599_to_fp16 = const()[name = string("op_7599_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_753_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_753_cast_fp16, y = var_7599_to_fp16)[name = string("aw_chunk_753_cast_fp16")]; + fp16 var_7601_to_fp16 = const()[name = string("op_7601_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_755_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_755_cast_fp16, y = var_7601_to_fp16)[name = string("aw_chunk_755_cast_fp16")]; + fp16 var_7603_to_fp16 = const()[name = string("op_7603_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_757_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_757_cast_fp16, y = var_7603_to_fp16)[name = string("aw_chunk_757_cast_fp16")]; + fp16 var_7605_to_fp16 = const()[name = string("op_7605_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_759_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_759_cast_fp16, y = var_7605_to_fp16)[name = string("aw_chunk_759_cast_fp16")]; + fp16 var_7607_to_fp16 = const()[name = string("op_7607_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_761_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_761_cast_fp16, y = var_7607_to_fp16)[name = string("aw_chunk_761_cast_fp16")]; + fp16 var_7609_to_fp16 = const()[name = string("op_7609_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_763_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_763_cast_fp16, y = var_7609_to_fp16)[name = string("aw_chunk_763_cast_fp16")]; + fp16 var_7611_to_fp16 = const()[name = string("op_7611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_765_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_765_cast_fp16, y = var_7611_to_fp16)[name = string("aw_chunk_765_cast_fp16")]; + fp16 var_7613_to_fp16 = const()[name = string("op_7613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_767_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_767_cast_fp16, y = var_7613_to_fp16)[name = string("aw_chunk_767_cast_fp16")]; + tensor var_7615_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_673_cast_fp16)[name = string("op_7615_cast_fp16")]; + tensor var_7616_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_675_cast_fp16)[name = string("op_7616_cast_fp16")]; + tensor var_7617_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_677_cast_fp16)[name = string("op_7617_cast_fp16")]; + tensor var_7618_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_679_cast_fp16)[name = string("op_7618_cast_fp16")]; + tensor var_7619_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_681_cast_fp16)[name = string("op_7619_cast_fp16")]; + tensor var_7620_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_683_cast_fp16)[name = string("op_7620_cast_fp16")]; + tensor var_7621_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_685_cast_fp16)[name = string("op_7621_cast_fp16")]; + tensor var_7622_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_687_cast_fp16)[name = string("op_7622_cast_fp16")]; + tensor var_7623_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_689_cast_fp16)[name = string("op_7623_cast_fp16")]; + tensor var_7624_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_691_cast_fp16)[name = string("op_7624_cast_fp16")]; + tensor var_7625_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_693_cast_fp16)[name = string("op_7625_cast_fp16")]; + tensor var_7626_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_695_cast_fp16)[name = string("op_7626_cast_fp16")]; + tensor var_7627_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_697_cast_fp16)[name = string("op_7627_cast_fp16")]; + tensor var_7628_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_699_cast_fp16)[name = string("op_7628_cast_fp16")]; + tensor var_7629_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_701_cast_fp16)[name = string("op_7629_cast_fp16")]; + tensor var_7630_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_703_cast_fp16)[name = string("op_7630_cast_fp16")]; + tensor var_7631_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_705_cast_fp16)[name = string("op_7631_cast_fp16")]; + tensor var_7632_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_707_cast_fp16)[name = string("op_7632_cast_fp16")]; + tensor var_7633_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_709_cast_fp16)[name = string("op_7633_cast_fp16")]; + tensor var_7634_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_711_cast_fp16)[name = string("op_7634_cast_fp16")]; + tensor var_7635_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_713_cast_fp16)[name = string("op_7635_cast_fp16")]; + tensor var_7636_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_715_cast_fp16)[name = string("op_7636_cast_fp16")]; + tensor var_7637_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_717_cast_fp16)[name = string("op_7637_cast_fp16")]; + tensor var_7638_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_719_cast_fp16)[name = string("op_7638_cast_fp16")]; + tensor var_7639_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_721_cast_fp16)[name = string("op_7639_cast_fp16")]; + tensor var_7640_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_723_cast_fp16)[name = string("op_7640_cast_fp16")]; + tensor var_7641_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_725_cast_fp16)[name = string("op_7641_cast_fp16")]; + tensor var_7642_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_727_cast_fp16)[name = string("op_7642_cast_fp16")]; + tensor var_7643_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_729_cast_fp16)[name = string("op_7643_cast_fp16")]; + tensor var_7644_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_731_cast_fp16)[name = string("op_7644_cast_fp16")]; + tensor var_7645_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_733_cast_fp16)[name = string("op_7645_cast_fp16")]; + tensor var_7646_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_735_cast_fp16)[name = string("op_7646_cast_fp16")]; + tensor var_7647_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_737_cast_fp16)[name = string("op_7647_cast_fp16")]; + tensor var_7648_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_739_cast_fp16)[name = string("op_7648_cast_fp16")]; + tensor var_7649_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_741_cast_fp16)[name = string("op_7649_cast_fp16")]; + tensor var_7650_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_743_cast_fp16)[name = string("op_7650_cast_fp16")]; + tensor var_7651_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_745_cast_fp16)[name = string("op_7651_cast_fp16")]; + tensor var_7652_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_747_cast_fp16)[name = string("op_7652_cast_fp16")]; + tensor var_7653_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_749_cast_fp16)[name = string("op_7653_cast_fp16")]; + tensor var_7654_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_751_cast_fp16)[name = string("op_7654_cast_fp16")]; + tensor var_7655_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_753_cast_fp16)[name = string("op_7655_cast_fp16")]; + tensor var_7656_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_755_cast_fp16)[name = string("op_7656_cast_fp16")]; + tensor var_7657_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_757_cast_fp16)[name = string("op_7657_cast_fp16")]; + tensor var_7658_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_759_cast_fp16)[name = string("op_7658_cast_fp16")]; + tensor var_7659_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_761_cast_fp16)[name = string("op_7659_cast_fp16")]; + tensor var_7660_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_763_cast_fp16)[name = string("op_7660_cast_fp16")]; + tensor var_7661_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_765_cast_fp16)[name = string("op_7661_cast_fp16")]; + tensor var_7662_cast_fp16 = softmax(axis = var_6888, x = aw_chunk_767_cast_fp16)[name = string("op_7662_cast_fp16")]; + string var_7664_equation_0 = const()[name = string("op_7664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7664_cast_fp16 = einsum(equation = var_7664_equation_0, values = (var_7376_cast_fp16, var_7615_cast_fp16))[name = string("op_7664_cast_fp16")]; + string var_7666_equation_0 = const()[name = string("op_7666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7666_cast_fp16 = einsum(equation = var_7666_equation_0, values = (var_7376_cast_fp16, var_7616_cast_fp16))[name = string("op_7666_cast_fp16")]; + string var_7668_equation_0 = const()[name = string("op_7668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7668_cast_fp16 = einsum(equation = var_7668_equation_0, values = (var_7376_cast_fp16, var_7617_cast_fp16))[name = string("op_7668_cast_fp16")]; + string var_7670_equation_0 = const()[name = string("op_7670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7670_cast_fp16 = einsum(equation = var_7670_equation_0, values = (var_7376_cast_fp16, var_7618_cast_fp16))[name = string("op_7670_cast_fp16")]; + string var_7672_equation_0 = const()[name = string("op_7672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7672_cast_fp16 = einsum(equation = var_7672_equation_0, values = (var_7380_cast_fp16, var_7619_cast_fp16))[name = string("op_7672_cast_fp16")]; + string var_7674_equation_0 = const()[name = string("op_7674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7674_cast_fp16 = einsum(equation = var_7674_equation_0, values = (var_7380_cast_fp16, var_7620_cast_fp16))[name = string("op_7674_cast_fp16")]; + string var_7676_equation_0 = const()[name = string("op_7676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7676_cast_fp16 = einsum(equation = var_7676_equation_0, values = (var_7380_cast_fp16, var_7621_cast_fp16))[name = string("op_7676_cast_fp16")]; + string var_7678_equation_0 = const()[name = string("op_7678_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7678_cast_fp16 = einsum(equation = var_7678_equation_0, values = (var_7380_cast_fp16, var_7622_cast_fp16))[name = string("op_7678_cast_fp16")]; + string var_7680_equation_0 = const()[name = string("op_7680_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7680_cast_fp16 = einsum(equation = var_7680_equation_0, values = (var_7384_cast_fp16, var_7623_cast_fp16))[name = string("op_7680_cast_fp16")]; + string var_7682_equation_0 = const()[name = string("op_7682_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7682_cast_fp16 = einsum(equation = var_7682_equation_0, values = (var_7384_cast_fp16, var_7624_cast_fp16))[name = string("op_7682_cast_fp16")]; + string var_7684_equation_0 = const()[name = string("op_7684_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7684_cast_fp16 = einsum(equation = var_7684_equation_0, values = (var_7384_cast_fp16, var_7625_cast_fp16))[name = string("op_7684_cast_fp16")]; + string var_7686_equation_0 = const()[name = string("op_7686_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7686_cast_fp16 = einsum(equation = var_7686_equation_0, values = (var_7384_cast_fp16, var_7626_cast_fp16))[name = string("op_7686_cast_fp16")]; + string var_7688_equation_0 = const()[name = string("op_7688_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7688_cast_fp16 = einsum(equation = var_7688_equation_0, values = (var_7388_cast_fp16, var_7627_cast_fp16))[name = string("op_7688_cast_fp16")]; + string var_7690_equation_0 = const()[name = string("op_7690_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7690_cast_fp16 = einsum(equation = var_7690_equation_0, values = (var_7388_cast_fp16, var_7628_cast_fp16))[name = string("op_7690_cast_fp16")]; + string var_7692_equation_0 = const()[name = string("op_7692_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7692_cast_fp16 = einsum(equation = var_7692_equation_0, values = (var_7388_cast_fp16, var_7629_cast_fp16))[name = string("op_7692_cast_fp16")]; + string var_7694_equation_0 = const()[name = string("op_7694_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7694_cast_fp16 = einsum(equation = var_7694_equation_0, values = (var_7388_cast_fp16, var_7630_cast_fp16))[name = string("op_7694_cast_fp16")]; + string var_7696_equation_0 = const()[name = string("op_7696_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7696_cast_fp16 = einsum(equation = var_7696_equation_0, values = (var_7392_cast_fp16, var_7631_cast_fp16))[name = string("op_7696_cast_fp16")]; + string var_7698_equation_0 = const()[name = string("op_7698_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7698_cast_fp16 = einsum(equation = var_7698_equation_0, values = (var_7392_cast_fp16, var_7632_cast_fp16))[name = string("op_7698_cast_fp16")]; + string var_7700_equation_0 = const()[name = string("op_7700_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7700_cast_fp16 = einsum(equation = var_7700_equation_0, values = (var_7392_cast_fp16, var_7633_cast_fp16))[name = string("op_7700_cast_fp16")]; + string var_7702_equation_0 = const()[name = string("op_7702_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7702_cast_fp16 = einsum(equation = var_7702_equation_0, values = (var_7392_cast_fp16, var_7634_cast_fp16))[name = string("op_7702_cast_fp16")]; + string var_7704_equation_0 = const()[name = string("op_7704_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7704_cast_fp16 = einsum(equation = var_7704_equation_0, values = (var_7396_cast_fp16, var_7635_cast_fp16))[name = string("op_7704_cast_fp16")]; + string var_7706_equation_0 = const()[name = string("op_7706_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7706_cast_fp16 = einsum(equation = var_7706_equation_0, values = (var_7396_cast_fp16, var_7636_cast_fp16))[name = string("op_7706_cast_fp16")]; + string var_7708_equation_0 = const()[name = string("op_7708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7708_cast_fp16 = einsum(equation = var_7708_equation_0, values = (var_7396_cast_fp16, var_7637_cast_fp16))[name = string("op_7708_cast_fp16")]; + string var_7710_equation_0 = const()[name = string("op_7710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7710_cast_fp16 = einsum(equation = var_7710_equation_0, values = (var_7396_cast_fp16, var_7638_cast_fp16))[name = string("op_7710_cast_fp16")]; + string var_7712_equation_0 = const()[name = string("op_7712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7712_cast_fp16 = einsum(equation = var_7712_equation_0, values = (var_7400_cast_fp16, var_7639_cast_fp16))[name = string("op_7712_cast_fp16")]; + string var_7714_equation_0 = const()[name = string("op_7714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7714_cast_fp16 = einsum(equation = var_7714_equation_0, values = (var_7400_cast_fp16, var_7640_cast_fp16))[name = string("op_7714_cast_fp16")]; + string var_7716_equation_0 = const()[name = string("op_7716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7716_cast_fp16 = einsum(equation = var_7716_equation_0, values = (var_7400_cast_fp16, var_7641_cast_fp16))[name = string("op_7716_cast_fp16")]; + string var_7718_equation_0 = const()[name = string("op_7718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7718_cast_fp16 = einsum(equation = var_7718_equation_0, values = (var_7400_cast_fp16, var_7642_cast_fp16))[name = string("op_7718_cast_fp16")]; + string var_7720_equation_0 = const()[name = string("op_7720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7720_cast_fp16 = einsum(equation = var_7720_equation_0, values = (var_7404_cast_fp16, var_7643_cast_fp16))[name = string("op_7720_cast_fp16")]; + string var_7722_equation_0 = const()[name = string("op_7722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7722_cast_fp16 = einsum(equation = var_7722_equation_0, values = (var_7404_cast_fp16, var_7644_cast_fp16))[name = string("op_7722_cast_fp16")]; + string var_7724_equation_0 = const()[name = string("op_7724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7724_cast_fp16 = einsum(equation = var_7724_equation_0, values = (var_7404_cast_fp16, var_7645_cast_fp16))[name = string("op_7724_cast_fp16")]; + string var_7726_equation_0 = const()[name = string("op_7726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7726_cast_fp16 = einsum(equation = var_7726_equation_0, values = (var_7404_cast_fp16, var_7646_cast_fp16))[name = string("op_7726_cast_fp16")]; + string var_7728_equation_0 = const()[name = string("op_7728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7728_cast_fp16 = einsum(equation = var_7728_equation_0, values = (var_7408_cast_fp16, var_7647_cast_fp16))[name = string("op_7728_cast_fp16")]; + string var_7730_equation_0 = const()[name = string("op_7730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7730_cast_fp16 = einsum(equation = var_7730_equation_0, values = (var_7408_cast_fp16, var_7648_cast_fp16))[name = string("op_7730_cast_fp16")]; + string var_7732_equation_0 = const()[name = string("op_7732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7732_cast_fp16 = einsum(equation = var_7732_equation_0, values = (var_7408_cast_fp16, var_7649_cast_fp16))[name = string("op_7732_cast_fp16")]; + string var_7734_equation_0 = const()[name = string("op_7734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7734_cast_fp16 = einsum(equation = var_7734_equation_0, values = (var_7408_cast_fp16, var_7650_cast_fp16))[name = string("op_7734_cast_fp16")]; + string var_7736_equation_0 = const()[name = string("op_7736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7736_cast_fp16 = einsum(equation = var_7736_equation_0, values = (var_7412_cast_fp16, var_7651_cast_fp16))[name = string("op_7736_cast_fp16")]; + string var_7738_equation_0 = const()[name = string("op_7738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7738_cast_fp16 = einsum(equation = var_7738_equation_0, values = (var_7412_cast_fp16, var_7652_cast_fp16))[name = string("op_7738_cast_fp16")]; + string var_7740_equation_0 = const()[name = string("op_7740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7740_cast_fp16 = einsum(equation = var_7740_equation_0, values = (var_7412_cast_fp16, var_7653_cast_fp16))[name = string("op_7740_cast_fp16")]; + string var_7742_equation_0 = const()[name = string("op_7742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7742_cast_fp16 = einsum(equation = var_7742_equation_0, values = (var_7412_cast_fp16, var_7654_cast_fp16))[name = string("op_7742_cast_fp16")]; + string var_7744_equation_0 = const()[name = string("op_7744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7744_cast_fp16 = einsum(equation = var_7744_equation_0, values = (var_7416_cast_fp16, var_7655_cast_fp16))[name = string("op_7744_cast_fp16")]; + string var_7746_equation_0 = const()[name = string("op_7746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7746_cast_fp16 = einsum(equation = var_7746_equation_0, values = (var_7416_cast_fp16, var_7656_cast_fp16))[name = string("op_7746_cast_fp16")]; + string var_7748_equation_0 = const()[name = string("op_7748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7748_cast_fp16 = einsum(equation = var_7748_equation_0, values = (var_7416_cast_fp16, var_7657_cast_fp16))[name = string("op_7748_cast_fp16")]; + string var_7750_equation_0 = const()[name = string("op_7750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7750_cast_fp16 = einsum(equation = var_7750_equation_0, values = (var_7416_cast_fp16, var_7658_cast_fp16))[name = string("op_7750_cast_fp16")]; + string var_7752_equation_0 = const()[name = string("op_7752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7752_cast_fp16 = einsum(equation = var_7752_equation_0, values = (var_7420_cast_fp16, var_7659_cast_fp16))[name = string("op_7752_cast_fp16")]; + string var_7754_equation_0 = const()[name = string("op_7754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7754_cast_fp16 = einsum(equation = var_7754_equation_0, values = (var_7420_cast_fp16, var_7660_cast_fp16))[name = string("op_7754_cast_fp16")]; + string var_7756_equation_0 = const()[name = string("op_7756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7756_cast_fp16 = einsum(equation = var_7756_equation_0, values = (var_7420_cast_fp16, var_7661_cast_fp16))[name = string("op_7756_cast_fp16")]; + string var_7758_equation_0 = const()[name = string("op_7758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_7758_cast_fp16 = einsum(equation = var_7758_equation_0, values = (var_7420_cast_fp16, var_7662_cast_fp16))[name = string("op_7758_cast_fp16")]; + bool var_7760_interleave_0 = const()[name = string("op_7760_interleave_0"), val = bool(false)]; + tensor var_7760_cast_fp16 = concat(axis = var_6871, interleave = var_7760_interleave_0, values = (var_7664_cast_fp16, var_7666_cast_fp16, var_7668_cast_fp16, var_7670_cast_fp16))[name = string("op_7760_cast_fp16")]; + bool var_7762_interleave_0 = const()[name = string("op_7762_interleave_0"), val = bool(false)]; + tensor var_7762_cast_fp16 = concat(axis = var_6871, interleave = var_7762_interleave_0, values = (var_7672_cast_fp16, var_7674_cast_fp16, var_7676_cast_fp16, var_7678_cast_fp16))[name = string("op_7762_cast_fp16")]; + bool var_7764_interleave_0 = const()[name = string("op_7764_interleave_0"), val = bool(false)]; + tensor var_7764_cast_fp16 = concat(axis = var_6871, interleave = var_7764_interleave_0, values = (var_7680_cast_fp16, var_7682_cast_fp16, var_7684_cast_fp16, var_7686_cast_fp16))[name = string("op_7764_cast_fp16")]; + bool var_7766_interleave_0 = const()[name = string("op_7766_interleave_0"), val = bool(false)]; + tensor var_7766_cast_fp16 = concat(axis = var_6871, interleave = var_7766_interleave_0, values = (var_7688_cast_fp16, var_7690_cast_fp16, var_7692_cast_fp16, var_7694_cast_fp16))[name = string("op_7766_cast_fp16")]; + bool var_7768_interleave_0 = const()[name = string("op_7768_interleave_0"), val = bool(false)]; + tensor var_7768_cast_fp16 = concat(axis = var_6871, interleave = var_7768_interleave_0, values = (var_7696_cast_fp16, var_7698_cast_fp16, var_7700_cast_fp16, var_7702_cast_fp16))[name = string("op_7768_cast_fp16")]; + bool var_7770_interleave_0 = const()[name = string("op_7770_interleave_0"), val = bool(false)]; + tensor var_7770_cast_fp16 = concat(axis = var_6871, interleave = var_7770_interleave_0, values = (var_7704_cast_fp16, var_7706_cast_fp16, var_7708_cast_fp16, var_7710_cast_fp16))[name = string("op_7770_cast_fp16")]; + bool var_7772_interleave_0 = const()[name = string("op_7772_interleave_0"), val = bool(false)]; + tensor var_7772_cast_fp16 = concat(axis = var_6871, interleave = var_7772_interleave_0, values = (var_7712_cast_fp16, var_7714_cast_fp16, var_7716_cast_fp16, var_7718_cast_fp16))[name = string("op_7772_cast_fp16")]; + bool var_7774_interleave_0 = const()[name = string("op_7774_interleave_0"), val = bool(false)]; + tensor var_7774_cast_fp16 = concat(axis = var_6871, interleave = var_7774_interleave_0, values = (var_7720_cast_fp16, var_7722_cast_fp16, var_7724_cast_fp16, var_7726_cast_fp16))[name = string("op_7774_cast_fp16")]; + bool var_7776_interleave_0 = const()[name = string("op_7776_interleave_0"), val = bool(false)]; + tensor var_7776_cast_fp16 = concat(axis = var_6871, interleave = var_7776_interleave_0, values = (var_7728_cast_fp16, var_7730_cast_fp16, var_7732_cast_fp16, var_7734_cast_fp16))[name = string("op_7776_cast_fp16")]; + bool var_7778_interleave_0 = const()[name = string("op_7778_interleave_0"), val = bool(false)]; + tensor var_7778_cast_fp16 = concat(axis = var_6871, interleave = var_7778_interleave_0, values = (var_7736_cast_fp16, var_7738_cast_fp16, var_7740_cast_fp16, var_7742_cast_fp16))[name = string("op_7778_cast_fp16")]; + bool var_7780_interleave_0 = const()[name = string("op_7780_interleave_0"), val = bool(false)]; + tensor var_7780_cast_fp16 = concat(axis = var_6871, interleave = var_7780_interleave_0, values = (var_7744_cast_fp16, var_7746_cast_fp16, var_7748_cast_fp16, var_7750_cast_fp16))[name = string("op_7780_cast_fp16")]; + bool var_7782_interleave_0 = const()[name = string("op_7782_interleave_0"), val = bool(false)]; + tensor var_7782_cast_fp16 = concat(axis = var_6871, interleave = var_7782_interleave_0, values = (var_7752_cast_fp16, var_7754_cast_fp16, var_7756_cast_fp16, var_7758_cast_fp16))[name = string("op_7782_cast_fp16")]; + bool input_57_interleave_0 = const()[name = string("input_57_interleave_0"), val = bool(false)]; + tensor input_57_cast_fp16 = concat(axis = var_6888, interleave = input_57_interleave_0, values = (var_7760_cast_fp16, var_7762_cast_fp16, var_7764_cast_fp16, var_7766_cast_fp16, var_7768_cast_fp16, var_7770_cast_fp16, var_7772_cast_fp16, var_7774_cast_fp16, var_7776_cast_fp16, var_7778_cast_fp16, var_7780_cast_fp16, var_7782_cast_fp16))[name = string("input_57_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108989952)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110169664)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_7801_to_fp16 = const()[name = string("op_7801_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_7801_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor input_59_gamma_0_to_fp16 = const()[name = string("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110171264)))]; + tensor input_59_beta_0_to_fp16 = const()[name = string("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110172864)))]; + fp16 input_59_epsilon_0_to_fp16 = const()[name = string("input_59_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1, 1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1, 1])]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110174464)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114893120)))]; + tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")]; + tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = string("input_63_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114899328)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119617984)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_33_cast_fp16")]; + int32 var_7830 = const()[name = string("op_7830"), val = int32(3)]; + int32 var_7847 = const()[name = string("op_7847"), val = int32(1)]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_7864_to_fp16 = const()[name = string("op_7864_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_7864_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119619584)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119621184)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119622784)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120802496)))]; + tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_17_cast_fp16")]; + string key_17_pad_type_0 = const()[name = string("key_17_pad_type_0"), val = string("valid")]; + tensor key_17_strides_0 = const()[name = string("key_17_strides_0"), val = tensor([1, 1])]; + tensor key_17_pad_0 = const()[name = string("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_17_dilations_0 = const()[name = string("key_17_dilations_0"), val = tensor([1, 1])]; + int32 key_17_groups_0 = const()[name = string("key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120804096)))]; + tensor key_17_cast_fp16 = conv(dilations = key_17_dilations_0, groups = key_17_groups_0, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("key_17_cast_fp16")]; + string value_17_pad_type_0 = const()[name = string("value_17_pad_type_0"), val = string("valid")]; + tensor value_17_strides_0 = const()[name = string("value_17_strides_0"), val = tensor([1, 1])]; + tensor value_17_pad_0 = const()[name = string("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_17_dilations_0 = const()[name = string("value_17_dilations_0"), val = tensor([1, 1])]; + int32 value_17_groups_0 = const()[name = string("value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121983808)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123163520)))]; + tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = value_17_dilations_0, groups = value_17_groups_0, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_7902_begin_0 = const()[name = string("op_7902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7902_end_0 = const()[name = string("op_7902_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7902_end_mask_0 = const()[name = string("op_7902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = query_17_cast_fp16)[name = string("op_7902_cast_fp16")]; + tensor var_7906_begin_0 = const()[name = string("op_7906_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7906_end_0 = const()[name = string("op_7906_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7906_end_mask_0 = const()[name = string("op_7906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = query_17_cast_fp16)[name = string("op_7906_cast_fp16")]; + tensor var_7910_begin_0 = const()[name = string("op_7910_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7910_end_0 = const()[name = string("op_7910_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7910_end_mask_0 = const()[name = string("op_7910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = query_17_cast_fp16)[name = string("op_7910_cast_fp16")]; + tensor var_7914_begin_0 = const()[name = string("op_7914_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7914_end_0 = const()[name = string("op_7914_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7914_end_mask_0 = const()[name = string("op_7914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = query_17_cast_fp16)[name = string("op_7914_cast_fp16")]; + tensor var_7918_begin_0 = const()[name = string("op_7918_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7918_end_0 = const()[name = string("op_7918_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7918_end_mask_0 = const()[name = string("op_7918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7918_cast_fp16 = slice_by_index(begin = var_7918_begin_0, end = var_7918_end_0, end_mask = var_7918_end_mask_0, x = query_17_cast_fp16)[name = string("op_7918_cast_fp16")]; + tensor var_7922_begin_0 = const()[name = string("op_7922_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7922_end_0 = const()[name = string("op_7922_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7922_end_mask_0 = const()[name = string("op_7922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7922_cast_fp16 = slice_by_index(begin = var_7922_begin_0, end = var_7922_end_0, end_mask = var_7922_end_mask_0, x = query_17_cast_fp16)[name = string("op_7922_cast_fp16")]; + tensor var_7926_begin_0 = const()[name = string("op_7926_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7926_end_0 = const()[name = string("op_7926_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7926_end_mask_0 = const()[name = string("op_7926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7926_cast_fp16 = slice_by_index(begin = var_7926_begin_0, end = var_7926_end_0, end_mask = var_7926_end_mask_0, x = query_17_cast_fp16)[name = string("op_7926_cast_fp16")]; + tensor var_7930_begin_0 = const()[name = string("op_7930_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7930_end_0 = const()[name = string("op_7930_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7930_end_mask_0 = const()[name = string("op_7930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7930_cast_fp16 = slice_by_index(begin = var_7930_begin_0, end = var_7930_end_0, end_mask = var_7930_end_mask_0, x = query_17_cast_fp16)[name = string("op_7930_cast_fp16")]; + tensor var_7934_begin_0 = const()[name = string("op_7934_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7934_end_0 = const()[name = string("op_7934_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7934_end_mask_0 = const()[name = string("op_7934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7934_cast_fp16 = slice_by_index(begin = var_7934_begin_0, end = var_7934_end_0, end_mask = var_7934_end_mask_0, x = query_17_cast_fp16)[name = string("op_7934_cast_fp16")]; + tensor var_7938_begin_0 = const()[name = string("op_7938_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7938_end_0 = const()[name = string("op_7938_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7938_end_mask_0 = const()[name = string("op_7938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7938_cast_fp16 = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = query_17_cast_fp16)[name = string("op_7938_cast_fp16")]; + tensor var_7942_begin_0 = const()[name = string("op_7942_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7942_end_0 = const()[name = string("op_7942_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7942_end_mask_0 = const()[name = string("op_7942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7942_cast_fp16 = slice_by_index(begin = var_7942_begin_0, end = var_7942_end_0, end_mask = var_7942_end_mask_0, x = query_17_cast_fp16)[name = string("op_7942_cast_fp16")]; + tensor var_7946_begin_0 = const()[name = string("op_7946_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7946_end_0 = const()[name = string("op_7946_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7946_end_mask_0 = const()[name = string("op_7946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7946_cast_fp16 = slice_by_index(begin = var_7946_begin_0, end = var_7946_end_0, end_mask = var_7946_end_mask_0, x = query_17_cast_fp16)[name = string("op_7946_cast_fp16")]; + tensor var_7955_begin_0 = const()[name = string("op_7955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7955_end_0 = const()[name = string("op_7955_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7955_end_mask_0 = const()[name = string("op_7955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7955_cast_fp16 = slice_by_index(begin = var_7955_begin_0, end = var_7955_end_0, end_mask = var_7955_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7955_cast_fp16")]; + tensor var_7962_begin_0 = const()[name = string("op_7962_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7962_end_0 = const()[name = string("op_7962_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7962_end_mask_0 = const()[name = string("op_7962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7962_cast_fp16 = slice_by_index(begin = var_7962_begin_0, end = var_7962_end_0, end_mask = var_7962_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7962_cast_fp16")]; + tensor var_7969_begin_0 = const()[name = string("op_7969_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7969_end_0 = const()[name = string("op_7969_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7969_end_mask_0 = const()[name = string("op_7969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7969_cast_fp16 = slice_by_index(begin = var_7969_begin_0, end = var_7969_end_0, end_mask = var_7969_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7969_cast_fp16")]; + tensor var_7976_begin_0 = const()[name = string("op_7976_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7976_end_0 = const()[name = string("op_7976_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7976_end_mask_0 = const()[name = string("op_7976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7976_cast_fp16 = slice_by_index(begin = var_7976_begin_0, end = var_7976_end_0, end_mask = var_7976_end_mask_0, x = var_7902_cast_fp16)[name = string("op_7976_cast_fp16")]; + tensor var_7983_begin_0 = const()[name = string("op_7983_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7983_end_0 = const()[name = string("op_7983_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7983_end_mask_0 = const()[name = string("op_7983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7983_cast_fp16 = slice_by_index(begin = var_7983_begin_0, end = var_7983_end_0, end_mask = var_7983_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7983_cast_fp16")]; + tensor var_7990_begin_0 = const()[name = string("op_7990_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7990_end_0 = const()[name = string("op_7990_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7990_end_mask_0 = const()[name = string("op_7990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7990_cast_fp16 = slice_by_index(begin = var_7990_begin_0, end = var_7990_end_0, end_mask = var_7990_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7990_cast_fp16")]; + tensor var_7997_begin_0 = const()[name = string("op_7997_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7997_end_0 = const()[name = string("op_7997_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7997_end_mask_0 = const()[name = string("op_7997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7997_cast_fp16 = slice_by_index(begin = var_7997_begin_0, end = var_7997_end_0, end_mask = var_7997_end_mask_0, x = var_7906_cast_fp16)[name = string("op_7997_cast_fp16")]; + tensor var_8004_begin_0 = const()[name = string("op_8004_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8004_end_0 = const()[name = string("op_8004_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8004_end_mask_0 = const()[name = string("op_8004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8004_cast_fp16 = slice_by_index(begin = var_8004_begin_0, end = var_8004_end_0, end_mask = var_8004_end_mask_0, x = var_7906_cast_fp16)[name = string("op_8004_cast_fp16")]; + tensor var_8011_begin_0 = const()[name = string("op_8011_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8011_end_0 = const()[name = string("op_8011_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8011_end_mask_0 = const()[name = string("op_8011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8011_cast_fp16 = slice_by_index(begin = var_8011_begin_0, end = var_8011_end_0, end_mask = var_8011_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8011_cast_fp16")]; + tensor var_8018_begin_0 = const()[name = string("op_8018_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8018_end_0 = const()[name = string("op_8018_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8018_end_mask_0 = const()[name = string("op_8018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8018_cast_fp16 = slice_by_index(begin = var_8018_begin_0, end = var_8018_end_0, end_mask = var_8018_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8018_cast_fp16")]; + tensor var_8025_begin_0 = const()[name = string("op_8025_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8025_end_0 = const()[name = string("op_8025_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8025_end_mask_0 = const()[name = string("op_8025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8025_cast_fp16 = slice_by_index(begin = var_8025_begin_0, end = var_8025_end_0, end_mask = var_8025_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8025_cast_fp16")]; + tensor var_8032_begin_0 = const()[name = string("op_8032_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8032_end_0 = const()[name = string("op_8032_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8032_end_mask_0 = const()[name = string("op_8032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8032_cast_fp16 = slice_by_index(begin = var_8032_begin_0, end = var_8032_end_0, end_mask = var_8032_end_mask_0, x = var_7910_cast_fp16)[name = string("op_8032_cast_fp16")]; + tensor var_8039_begin_0 = const()[name = string("op_8039_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8039_end_0 = const()[name = string("op_8039_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8039_end_mask_0 = const()[name = string("op_8039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8039_cast_fp16 = slice_by_index(begin = var_8039_begin_0, end = var_8039_end_0, end_mask = var_8039_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8039_cast_fp16")]; + tensor var_8046_begin_0 = const()[name = string("op_8046_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8046_end_0 = const()[name = string("op_8046_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8046_end_mask_0 = const()[name = string("op_8046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8046_cast_fp16 = slice_by_index(begin = var_8046_begin_0, end = var_8046_end_0, end_mask = var_8046_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8046_cast_fp16")]; + tensor var_8053_begin_0 = const()[name = string("op_8053_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8053_end_0 = const()[name = string("op_8053_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8053_end_mask_0 = const()[name = string("op_8053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8053_cast_fp16 = slice_by_index(begin = var_8053_begin_0, end = var_8053_end_0, end_mask = var_8053_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8053_cast_fp16")]; + tensor var_8060_begin_0 = const()[name = string("op_8060_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8060_end_0 = const()[name = string("op_8060_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8060_end_mask_0 = const()[name = string("op_8060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8060_cast_fp16 = slice_by_index(begin = var_8060_begin_0, end = var_8060_end_0, end_mask = var_8060_end_mask_0, x = var_7914_cast_fp16)[name = string("op_8060_cast_fp16")]; + tensor var_8067_begin_0 = const()[name = string("op_8067_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8067_end_0 = const()[name = string("op_8067_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8067_end_mask_0 = const()[name = string("op_8067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8067_cast_fp16 = slice_by_index(begin = var_8067_begin_0, end = var_8067_end_0, end_mask = var_8067_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8067_cast_fp16")]; + tensor var_8074_begin_0 = const()[name = string("op_8074_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8074_end_0 = const()[name = string("op_8074_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8074_end_mask_0 = const()[name = string("op_8074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8074_cast_fp16 = slice_by_index(begin = var_8074_begin_0, end = var_8074_end_0, end_mask = var_8074_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8074_cast_fp16")]; + tensor var_8081_begin_0 = const()[name = string("op_8081_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8081_end_0 = const()[name = string("op_8081_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8081_end_mask_0 = const()[name = string("op_8081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8081_cast_fp16 = slice_by_index(begin = var_8081_begin_0, end = var_8081_end_0, end_mask = var_8081_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8081_cast_fp16")]; + tensor var_8088_begin_0 = const()[name = string("op_8088_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8088_end_0 = const()[name = string("op_8088_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8088_end_mask_0 = const()[name = string("op_8088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8088_cast_fp16 = slice_by_index(begin = var_8088_begin_0, end = var_8088_end_0, end_mask = var_8088_end_mask_0, x = var_7918_cast_fp16)[name = string("op_8088_cast_fp16")]; + tensor var_8095_begin_0 = const()[name = string("op_8095_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8095_end_0 = const()[name = string("op_8095_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8095_end_mask_0 = const()[name = string("op_8095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8095_cast_fp16 = slice_by_index(begin = var_8095_begin_0, end = var_8095_end_0, end_mask = var_8095_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8095_cast_fp16")]; + tensor var_8102_begin_0 = const()[name = string("op_8102_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8102_end_0 = const()[name = string("op_8102_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8102_end_mask_0 = const()[name = string("op_8102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8102_cast_fp16 = slice_by_index(begin = var_8102_begin_0, end = var_8102_end_0, end_mask = var_8102_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8102_cast_fp16")]; + tensor var_8109_begin_0 = const()[name = string("op_8109_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8109_end_0 = const()[name = string("op_8109_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8109_end_mask_0 = const()[name = string("op_8109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8109_cast_fp16 = slice_by_index(begin = var_8109_begin_0, end = var_8109_end_0, end_mask = var_8109_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8109_cast_fp16")]; + tensor var_8116_begin_0 = const()[name = string("op_8116_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8116_end_0 = const()[name = string("op_8116_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8116_end_mask_0 = const()[name = string("op_8116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8116_cast_fp16 = slice_by_index(begin = var_8116_begin_0, end = var_8116_end_0, end_mask = var_8116_end_mask_0, x = var_7922_cast_fp16)[name = string("op_8116_cast_fp16")]; + tensor var_8123_begin_0 = const()[name = string("op_8123_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8123_end_0 = const()[name = string("op_8123_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8123_end_mask_0 = const()[name = string("op_8123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8123_cast_fp16 = slice_by_index(begin = var_8123_begin_0, end = var_8123_end_0, end_mask = var_8123_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8123_cast_fp16")]; + tensor var_8130_begin_0 = const()[name = string("op_8130_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8130_end_0 = const()[name = string("op_8130_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8130_end_mask_0 = const()[name = string("op_8130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8130_cast_fp16 = slice_by_index(begin = var_8130_begin_0, end = var_8130_end_0, end_mask = var_8130_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8130_cast_fp16")]; + tensor var_8137_begin_0 = const()[name = string("op_8137_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8137_end_0 = const()[name = string("op_8137_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8137_end_mask_0 = const()[name = string("op_8137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8137_cast_fp16 = slice_by_index(begin = var_8137_begin_0, end = var_8137_end_0, end_mask = var_8137_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8137_cast_fp16")]; + tensor var_8144_begin_0 = const()[name = string("op_8144_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8144_end_0 = const()[name = string("op_8144_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8144_end_mask_0 = const()[name = string("op_8144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8144_cast_fp16 = slice_by_index(begin = var_8144_begin_0, end = var_8144_end_0, end_mask = var_8144_end_mask_0, x = var_7926_cast_fp16)[name = string("op_8144_cast_fp16")]; + tensor var_8151_begin_0 = const()[name = string("op_8151_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8151_end_0 = const()[name = string("op_8151_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8151_end_mask_0 = const()[name = string("op_8151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8151_cast_fp16 = slice_by_index(begin = var_8151_begin_0, end = var_8151_end_0, end_mask = var_8151_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8151_cast_fp16")]; + tensor var_8158_begin_0 = const()[name = string("op_8158_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8158_end_0 = const()[name = string("op_8158_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8158_end_mask_0 = const()[name = string("op_8158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8158_cast_fp16 = slice_by_index(begin = var_8158_begin_0, end = var_8158_end_0, end_mask = var_8158_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8158_cast_fp16")]; + tensor var_8165_begin_0 = const()[name = string("op_8165_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8165_end_0 = const()[name = string("op_8165_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8165_end_mask_0 = const()[name = string("op_8165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8165_cast_fp16 = slice_by_index(begin = var_8165_begin_0, end = var_8165_end_0, end_mask = var_8165_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8165_cast_fp16")]; + tensor var_8172_begin_0 = const()[name = string("op_8172_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8172_end_0 = const()[name = string("op_8172_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8172_end_mask_0 = const()[name = string("op_8172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8172_cast_fp16 = slice_by_index(begin = var_8172_begin_0, end = var_8172_end_0, end_mask = var_8172_end_mask_0, x = var_7930_cast_fp16)[name = string("op_8172_cast_fp16")]; + tensor var_8179_begin_0 = const()[name = string("op_8179_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8179_end_0 = const()[name = string("op_8179_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8179_end_mask_0 = const()[name = string("op_8179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8179_cast_fp16 = slice_by_index(begin = var_8179_begin_0, end = var_8179_end_0, end_mask = var_8179_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8179_cast_fp16")]; + tensor var_8186_begin_0 = const()[name = string("op_8186_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8186_end_0 = const()[name = string("op_8186_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8186_end_mask_0 = const()[name = string("op_8186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8186_cast_fp16 = slice_by_index(begin = var_8186_begin_0, end = var_8186_end_0, end_mask = var_8186_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8186_cast_fp16")]; + tensor var_8193_begin_0 = const()[name = string("op_8193_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8193_end_0 = const()[name = string("op_8193_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8193_end_mask_0 = const()[name = string("op_8193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8193_cast_fp16 = slice_by_index(begin = var_8193_begin_0, end = var_8193_end_0, end_mask = var_8193_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8193_cast_fp16")]; + tensor var_8200_begin_0 = const()[name = string("op_8200_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8200_end_0 = const()[name = string("op_8200_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8200_end_mask_0 = const()[name = string("op_8200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8200_cast_fp16 = slice_by_index(begin = var_8200_begin_0, end = var_8200_end_0, end_mask = var_8200_end_mask_0, x = var_7934_cast_fp16)[name = string("op_8200_cast_fp16")]; + tensor var_8207_begin_0 = const()[name = string("op_8207_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8207_end_0 = const()[name = string("op_8207_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8207_end_mask_0 = const()[name = string("op_8207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8207_cast_fp16 = slice_by_index(begin = var_8207_begin_0, end = var_8207_end_0, end_mask = var_8207_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8207_cast_fp16")]; + tensor var_8214_begin_0 = const()[name = string("op_8214_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8214_end_0 = const()[name = string("op_8214_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8214_end_mask_0 = const()[name = string("op_8214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8214_cast_fp16 = slice_by_index(begin = var_8214_begin_0, end = var_8214_end_0, end_mask = var_8214_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8214_cast_fp16")]; + tensor var_8221_begin_0 = const()[name = string("op_8221_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8221_end_0 = const()[name = string("op_8221_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8221_end_mask_0 = const()[name = string("op_8221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8221_cast_fp16 = slice_by_index(begin = var_8221_begin_0, end = var_8221_end_0, end_mask = var_8221_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8221_cast_fp16")]; + tensor var_8228_begin_0 = const()[name = string("op_8228_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8228_end_0 = const()[name = string("op_8228_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8228_end_mask_0 = const()[name = string("op_8228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8228_cast_fp16 = slice_by_index(begin = var_8228_begin_0, end = var_8228_end_0, end_mask = var_8228_end_mask_0, x = var_7938_cast_fp16)[name = string("op_8228_cast_fp16")]; + tensor var_8235_begin_0 = const()[name = string("op_8235_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8235_end_0 = const()[name = string("op_8235_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8235_end_mask_0 = const()[name = string("op_8235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8235_cast_fp16 = slice_by_index(begin = var_8235_begin_0, end = var_8235_end_0, end_mask = var_8235_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8235_cast_fp16")]; + tensor var_8242_begin_0 = const()[name = string("op_8242_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8242_end_0 = const()[name = string("op_8242_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8242_end_mask_0 = const()[name = string("op_8242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8242_cast_fp16 = slice_by_index(begin = var_8242_begin_0, end = var_8242_end_0, end_mask = var_8242_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8242_cast_fp16")]; + tensor var_8249_begin_0 = const()[name = string("op_8249_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8249_end_0 = const()[name = string("op_8249_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8249_end_mask_0 = const()[name = string("op_8249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8249_cast_fp16 = slice_by_index(begin = var_8249_begin_0, end = var_8249_end_0, end_mask = var_8249_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8249_cast_fp16")]; + tensor var_8256_begin_0 = const()[name = string("op_8256_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8256_end_0 = const()[name = string("op_8256_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8256_end_mask_0 = const()[name = string("op_8256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8256_cast_fp16 = slice_by_index(begin = var_8256_begin_0, end = var_8256_end_0, end_mask = var_8256_end_mask_0, x = var_7942_cast_fp16)[name = string("op_8256_cast_fp16")]; + tensor var_8263_begin_0 = const()[name = string("op_8263_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8263_end_0 = const()[name = string("op_8263_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8263_end_mask_0 = const()[name = string("op_8263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8263_cast_fp16 = slice_by_index(begin = var_8263_begin_0, end = var_8263_end_0, end_mask = var_8263_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8263_cast_fp16")]; + tensor var_8270_begin_0 = const()[name = string("op_8270_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8270_end_0 = const()[name = string("op_8270_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8270_end_mask_0 = const()[name = string("op_8270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8270_cast_fp16 = slice_by_index(begin = var_8270_begin_0, end = var_8270_end_0, end_mask = var_8270_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8270_cast_fp16")]; + tensor var_8277_begin_0 = const()[name = string("op_8277_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8277_end_0 = const()[name = string("op_8277_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8277_end_mask_0 = const()[name = string("op_8277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8277_cast_fp16 = slice_by_index(begin = var_8277_begin_0, end = var_8277_end_0, end_mask = var_8277_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8277_cast_fp16")]; + tensor var_8284_begin_0 = const()[name = string("op_8284_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8284_end_0 = const()[name = string("op_8284_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8284_end_mask_0 = const()[name = string("op_8284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8284_cast_fp16 = slice_by_index(begin = var_8284_begin_0, end = var_8284_end_0, end_mask = var_8284_end_mask_0, x = var_7946_cast_fp16)[name = string("op_8284_cast_fp16")]; + tensor k_17_perm_0 = const()[name = string("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8289_begin_0 = const()[name = string("op_8289_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8289_end_0 = const()[name = string("op_8289_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_8289_end_mask_0 = const()[name = string("op_8289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_17_cast_fp16 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = string("transpose_3")]; + tensor var_8289_cast_fp16 = slice_by_index(begin = var_8289_begin_0, end = var_8289_end_0, end_mask = var_8289_end_mask_0, x = k_17_cast_fp16)[name = string("op_8289_cast_fp16")]; + tensor var_8293_begin_0 = const()[name = string("op_8293_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8293_end_0 = const()[name = string("op_8293_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_8293_end_mask_0 = const()[name = string("op_8293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8293_cast_fp16 = slice_by_index(begin = var_8293_begin_0, end = var_8293_end_0, end_mask = var_8293_end_mask_0, x = k_17_cast_fp16)[name = string("op_8293_cast_fp16")]; + tensor var_8297_begin_0 = const()[name = string("op_8297_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8297_end_0 = const()[name = string("op_8297_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_8297_end_mask_0 = const()[name = string("op_8297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8297_cast_fp16 = slice_by_index(begin = var_8297_begin_0, end = var_8297_end_0, end_mask = var_8297_end_mask_0, x = k_17_cast_fp16)[name = string("op_8297_cast_fp16")]; + tensor var_8301_begin_0 = const()[name = string("op_8301_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8301_end_0 = const()[name = string("op_8301_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_8301_end_mask_0 = const()[name = string("op_8301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8301_cast_fp16 = slice_by_index(begin = var_8301_begin_0, end = var_8301_end_0, end_mask = var_8301_end_mask_0, x = k_17_cast_fp16)[name = string("op_8301_cast_fp16")]; + tensor var_8305_begin_0 = const()[name = string("op_8305_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8305_end_0 = const()[name = string("op_8305_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_8305_end_mask_0 = const()[name = string("op_8305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8305_cast_fp16 = slice_by_index(begin = var_8305_begin_0, end = var_8305_end_0, end_mask = var_8305_end_mask_0, x = k_17_cast_fp16)[name = string("op_8305_cast_fp16")]; + tensor var_8309_begin_0 = const()[name = string("op_8309_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8309_end_0 = const()[name = string("op_8309_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_8309_end_mask_0 = const()[name = string("op_8309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8309_cast_fp16 = slice_by_index(begin = var_8309_begin_0, end = var_8309_end_0, end_mask = var_8309_end_mask_0, x = k_17_cast_fp16)[name = string("op_8309_cast_fp16")]; + tensor var_8313_begin_0 = const()[name = string("op_8313_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8313_end_0 = const()[name = string("op_8313_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_8313_end_mask_0 = const()[name = string("op_8313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8313_cast_fp16 = slice_by_index(begin = var_8313_begin_0, end = var_8313_end_0, end_mask = var_8313_end_mask_0, x = k_17_cast_fp16)[name = string("op_8313_cast_fp16")]; + tensor var_8317_begin_0 = const()[name = string("op_8317_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8317_end_0 = const()[name = string("op_8317_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_8317_end_mask_0 = const()[name = string("op_8317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8317_cast_fp16 = slice_by_index(begin = var_8317_begin_0, end = var_8317_end_0, end_mask = var_8317_end_mask_0, x = k_17_cast_fp16)[name = string("op_8317_cast_fp16")]; + tensor var_8321_begin_0 = const()[name = string("op_8321_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8321_end_0 = const()[name = string("op_8321_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_8321_end_mask_0 = const()[name = string("op_8321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8321_cast_fp16 = slice_by_index(begin = var_8321_begin_0, end = var_8321_end_0, end_mask = var_8321_end_mask_0, x = k_17_cast_fp16)[name = string("op_8321_cast_fp16")]; + tensor var_8325_begin_0 = const()[name = string("op_8325_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8325_end_0 = const()[name = string("op_8325_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_8325_end_mask_0 = const()[name = string("op_8325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8325_cast_fp16 = slice_by_index(begin = var_8325_begin_0, end = var_8325_end_0, end_mask = var_8325_end_mask_0, x = k_17_cast_fp16)[name = string("op_8325_cast_fp16")]; + tensor var_8329_begin_0 = const()[name = string("op_8329_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8329_end_0 = const()[name = string("op_8329_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_8329_end_mask_0 = const()[name = string("op_8329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8329_cast_fp16 = slice_by_index(begin = var_8329_begin_0, end = var_8329_end_0, end_mask = var_8329_end_mask_0, x = k_17_cast_fp16)[name = string("op_8329_cast_fp16")]; + tensor var_8333_begin_0 = const()[name = string("op_8333_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8333_end_0 = const()[name = string("op_8333_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_8333_end_mask_0 = const()[name = string("op_8333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8333_cast_fp16 = slice_by_index(begin = var_8333_begin_0, end = var_8333_end_0, end_mask = var_8333_end_mask_0, x = k_17_cast_fp16)[name = string("op_8333_cast_fp16")]; + tensor var_8335_begin_0 = const()[name = string("op_8335_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8335_end_0 = const()[name = string("op_8335_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8335_end_mask_0 = const()[name = string("op_8335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8335_cast_fp16 = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = value_17_cast_fp16)[name = string("op_8335_cast_fp16")]; + tensor var_8339_begin_0 = const()[name = string("op_8339_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8339_end_0 = const()[name = string("op_8339_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8339_end_mask_0 = const()[name = string("op_8339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8339_cast_fp16 = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = value_17_cast_fp16)[name = string("op_8339_cast_fp16")]; + tensor var_8343_begin_0 = const()[name = string("op_8343_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8343_end_0 = const()[name = string("op_8343_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8343_end_mask_0 = const()[name = string("op_8343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8343_cast_fp16 = slice_by_index(begin = var_8343_begin_0, end = var_8343_end_0, end_mask = var_8343_end_mask_0, x = value_17_cast_fp16)[name = string("op_8343_cast_fp16")]; + tensor var_8347_begin_0 = const()[name = string("op_8347_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8347_end_0 = const()[name = string("op_8347_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8347_end_mask_0 = const()[name = string("op_8347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8347_cast_fp16 = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = value_17_cast_fp16)[name = string("op_8347_cast_fp16")]; + tensor var_8351_begin_0 = const()[name = string("op_8351_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8351_end_0 = const()[name = string("op_8351_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8351_end_mask_0 = const()[name = string("op_8351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8351_cast_fp16 = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = value_17_cast_fp16)[name = string("op_8351_cast_fp16")]; + tensor var_8355_begin_0 = const()[name = string("op_8355_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8355_end_0 = const()[name = string("op_8355_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8355_end_mask_0 = const()[name = string("op_8355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8355_cast_fp16 = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = value_17_cast_fp16)[name = string("op_8355_cast_fp16")]; + tensor var_8359_begin_0 = const()[name = string("op_8359_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8359_end_0 = const()[name = string("op_8359_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8359_end_mask_0 = const()[name = string("op_8359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8359_cast_fp16 = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = value_17_cast_fp16)[name = string("op_8359_cast_fp16")]; + tensor var_8363_begin_0 = const()[name = string("op_8363_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8363_end_0 = const()[name = string("op_8363_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8363_end_mask_0 = const()[name = string("op_8363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8363_cast_fp16 = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = value_17_cast_fp16)[name = string("op_8363_cast_fp16")]; + tensor var_8367_begin_0 = const()[name = string("op_8367_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8367_end_0 = const()[name = string("op_8367_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8367_end_mask_0 = const()[name = string("op_8367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8367_cast_fp16 = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = value_17_cast_fp16)[name = string("op_8367_cast_fp16")]; + tensor var_8371_begin_0 = const()[name = string("op_8371_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8371_end_0 = const()[name = string("op_8371_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8371_end_mask_0 = const()[name = string("op_8371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8371_cast_fp16 = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = value_17_cast_fp16)[name = string("op_8371_cast_fp16")]; + tensor var_8375_begin_0 = const()[name = string("op_8375_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8375_end_0 = const()[name = string("op_8375_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8375_end_mask_0 = const()[name = string("op_8375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8375_cast_fp16 = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = value_17_cast_fp16)[name = string("op_8375_cast_fp16")]; + tensor var_8379_begin_0 = const()[name = string("op_8379_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8379_end_0 = const()[name = string("op_8379_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8379_end_mask_0 = const()[name = string("op_8379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8379_cast_fp16 = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = value_17_cast_fp16)[name = string("op_8379_cast_fp16")]; + string _SplitHeadsQ__mh_w_769_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_769_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_769_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_769_equation_0, values = (var_8289_cast_fp16, var_7955_cast_fp16))[name = string("_SplitHeadsQ__mh_w_769_cast_fp16")]; + string _SplitHeadsQ__mh_w_771_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_771_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_771_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_771_equation_0, values = (var_8289_cast_fp16, var_7962_cast_fp16))[name = string("_SplitHeadsQ__mh_w_771_cast_fp16")]; + string _SplitHeadsQ__mh_w_773_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_773_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_773_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_773_equation_0, values = (var_8289_cast_fp16, var_7969_cast_fp16))[name = string("_SplitHeadsQ__mh_w_773_cast_fp16")]; + string _SplitHeadsQ__mh_w_775_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_775_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_775_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_775_equation_0, values = (var_8289_cast_fp16, var_7976_cast_fp16))[name = string("_SplitHeadsQ__mh_w_775_cast_fp16")]; + string _SplitHeadsQ__mh_w_777_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_777_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_777_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_777_equation_0, values = (var_8293_cast_fp16, var_7983_cast_fp16))[name = string("_SplitHeadsQ__mh_w_777_cast_fp16")]; + string _SplitHeadsQ__mh_w_779_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_779_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_779_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_779_equation_0, values = (var_8293_cast_fp16, var_7990_cast_fp16))[name = string("_SplitHeadsQ__mh_w_779_cast_fp16")]; + string _SplitHeadsQ__mh_w_781_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_781_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_781_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_781_equation_0, values = (var_8293_cast_fp16, var_7997_cast_fp16))[name = string("_SplitHeadsQ__mh_w_781_cast_fp16")]; + string _SplitHeadsQ__mh_w_783_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_783_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_783_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_783_equation_0, values = (var_8293_cast_fp16, var_8004_cast_fp16))[name = string("_SplitHeadsQ__mh_w_783_cast_fp16")]; + string _SplitHeadsQ__mh_w_785_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_785_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_785_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_785_equation_0, values = (var_8297_cast_fp16, var_8011_cast_fp16))[name = string("_SplitHeadsQ__mh_w_785_cast_fp16")]; + string _SplitHeadsQ__mh_w_787_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_787_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_787_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_787_equation_0, values = (var_8297_cast_fp16, var_8018_cast_fp16))[name = string("_SplitHeadsQ__mh_w_787_cast_fp16")]; + string _SplitHeadsQ__mh_w_789_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_789_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_789_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_789_equation_0, values = (var_8297_cast_fp16, var_8025_cast_fp16))[name = string("_SplitHeadsQ__mh_w_789_cast_fp16")]; + string _SplitHeadsQ__mh_w_791_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_791_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_791_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_791_equation_0, values = (var_8297_cast_fp16, var_8032_cast_fp16))[name = string("_SplitHeadsQ__mh_w_791_cast_fp16")]; + string _SplitHeadsQ__mh_w_793_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_793_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_793_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_793_equation_0, values = (var_8301_cast_fp16, var_8039_cast_fp16))[name = string("_SplitHeadsQ__mh_w_793_cast_fp16")]; + string _SplitHeadsQ__mh_w_795_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_795_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_795_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_795_equation_0, values = (var_8301_cast_fp16, var_8046_cast_fp16))[name = string("_SplitHeadsQ__mh_w_795_cast_fp16")]; + string _SplitHeadsQ__mh_w_797_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_797_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_797_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_797_equation_0, values = (var_8301_cast_fp16, var_8053_cast_fp16))[name = string("_SplitHeadsQ__mh_w_797_cast_fp16")]; + string _SplitHeadsQ__mh_w_799_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_799_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_799_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_799_equation_0, values = (var_8301_cast_fp16, var_8060_cast_fp16))[name = string("_SplitHeadsQ__mh_w_799_cast_fp16")]; + string _SplitHeadsQ__mh_w_801_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_801_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_801_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_801_equation_0, values = (var_8305_cast_fp16, var_8067_cast_fp16))[name = string("_SplitHeadsQ__mh_w_801_cast_fp16")]; + string _SplitHeadsQ__mh_w_803_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_803_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_803_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_803_equation_0, values = (var_8305_cast_fp16, var_8074_cast_fp16))[name = string("_SplitHeadsQ__mh_w_803_cast_fp16")]; + string _SplitHeadsQ__mh_w_805_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_805_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_805_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_805_equation_0, values = (var_8305_cast_fp16, var_8081_cast_fp16))[name = string("_SplitHeadsQ__mh_w_805_cast_fp16")]; + string _SplitHeadsQ__mh_w_807_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_807_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_807_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_807_equation_0, values = (var_8305_cast_fp16, var_8088_cast_fp16))[name = string("_SplitHeadsQ__mh_w_807_cast_fp16")]; + string _SplitHeadsQ__mh_w_809_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_809_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_809_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_809_equation_0, values = (var_8309_cast_fp16, var_8095_cast_fp16))[name = string("_SplitHeadsQ__mh_w_809_cast_fp16")]; + string _SplitHeadsQ__mh_w_811_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_811_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_811_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_811_equation_0, values = (var_8309_cast_fp16, var_8102_cast_fp16))[name = string("_SplitHeadsQ__mh_w_811_cast_fp16")]; + string _SplitHeadsQ__mh_w_813_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_813_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_813_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_813_equation_0, values = (var_8309_cast_fp16, var_8109_cast_fp16))[name = string("_SplitHeadsQ__mh_w_813_cast_fp16")]; + string _SplitHeadsQ__mh_w_815_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_815_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_815_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_815_equation_0, values = (var_8309_cast_fp16, var_8116_cast_fp16))[name = string("_SplitHeadsQ__mh_w_815_cast_fp16")]; + string _SplitHeadsQ__mh_w_817_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_817_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_817_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_817_equation_0, values = (var_8313_cast_fp16, var_8123_cast_fp16))[name = string("_SplitHeadsQ__mh_w_817_cast_fp16")]; + string _SplitHeadsQ__mh_w_819_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_819_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_819_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_819_equation_0, values = (var_8313_cast_fp16, var_8130_cast_fp16))[name = string("_SplitHeadsQ__mh_w_819_cast_fp16")]; + string _SplitHeadsQ__mh_w_821_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_821_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_821_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_821_equation_0, values = (var_8313_cast_fp16, var_8137_cast_fp16))[name = string("_SplitHeadsQ__mh_w_821_cast_fp16")]; + string _SplitHeadsQ__mh_w_823_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_823_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_823_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_823_equation_0, values = (var_8313_cast_fp16, var_8144_cast_fp16))[name = string("_SplitHeadsQ__mh_w_823_cast_fp16")]; + string _SplitHeadsQ__mh_w_825_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_825_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_825_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_825_equation_0, values = (var_8317_cast_fp16, var_8151_cast_fp16))[name = string("_SplitHeadsQ__mh_w_825_cast_fp16")]; + string _SplitHeadsQ__mh_w_827_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_827_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_827_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_827_equation_0, values = (var_8317_cast_fp16, var_8158_cast_fp16))[name = string("_SplitHeadsQ__mh_w_827_cast_fp16")]; + string _SplitHeadsQ__mh_w_829_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_829_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_829_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_829_equation_0, values = (var_8317_cast_fp16, var_8165_cast_fp16))[name = string("_SplitHeadsQ__mh_w_829_cast_fp16")]; + string _SplitHeadsQ__mh_w_831_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_831_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_831_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_831_equation_0, values = (var_8317_cast_fp16, var_8172_cast_fp16))[name = string("_SplitHeadsQ__mh_w_831_cast_fp16")]; + string _SplitHeadsQ__mh_w_833_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_833_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_833_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_833_equation_0, values = (var_8321_cast_fp16, var_8179_cast_fp16))[name = string("_SplitHeadsQ__mh_w_833_cast_fp16")]; + string _SplitHeadsQ__mh_w_835_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_835_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_835_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_835_equation_0, values = (var_8321_cast_fp16, var_8186_cast_fp16))[name = string("_SplitHeadsQ__mh_w_835_cast_fp16")]; + string _SplitHeadsQ__mh_w_837_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_837_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_837_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_837_equation_0, values = (var_8321_cast_fp16, var_8193_cast_fp16))[name = string("_SplitHeadsQ__mh_w_837_cast_fp16")]; + string _SplitHeadsQ__mh_w_839_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_839_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_839_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_839_equation_0, values = (var_8321_cast_fp16, var_8200_cast_fp16))[name = string("_SplitHeadsQ__mh_w_839_cast_fp16")]; + string _SplitHeadsQ__mh_w_841_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_841_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_841_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_841_equation_0, values = (var_8325_cast_fp16, var_8207_cast_fp16))[name = string("_SplitHeadsQ__mh_w_841_cast_fp16")]; + string _SplitHeadsQ__mh_w_843_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_843_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_843_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_843_equation_0, values = (var_8325_cast_fp16, var_8214_cast_fp16))[name = string("_SplitHeadsQ__mh_w_843_cast_fp16")]; + string _SplitHeadsQ__mh_w_845_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_845_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_845_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_845_equation_0, values = (var_8325_cast_fp16, var_8221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_845_cast_fp16")]; + string _SplitHeadsQ__mh_w_847_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_847_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_847_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_847_equation_0, values = (var_8325_cast_fp16, var_8228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_847_cast_fp16")]; + string _SplitHeadsQ__mh_w_849_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_849_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_849_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_849_equation_0, values = (var_8329_cast_fp16, var_8235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_849_cast_fp16")]; + string _SplitHeadsQ__mh_w_851_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_851_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_851_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_851_equation_0, values = (var_8329_cast_fp16, var_8242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_851_cast_fp16")]; + string _SplitHeadsQ__mh_w_853_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_853_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_853_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_853_equation_0, values = (var_8329_cast_fp16, var_8249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_853_cast_fp16")]; + string _SplitHeadsQ__mh_w_855_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_855_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_855_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_855_equation_0, values = (var_8329_cast_fp16, var_8256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_855_cast_fp16")]; + string _SplitHeadsQ__mh_w_857_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_857_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_857_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_857_equation_0, values = (var_8333_cast_fp16, var_8263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_857_cast_fp16")]; + string _SplitHeadsQ__mh_w_859_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_859_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_859_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_859_equation_0, values = (var_8333_cast_fp16, var_8270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_859_cast_fp16")]; + string _SplitHeadsQ__mh_w_861_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_861_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_861_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_861_equation_0, values = (var_8333_cast_fp16, var_8277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_861_cast_fp16")]; + string _SplitHeadsQ__mh_w_863_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_863_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_863_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_863_equation_0, values = (var_8333_cast_fp16, var_8284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_863_cast_fp16")]; + fp16 var_8478_to_fp16 = const()[name = string("op_8478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_769_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_769_cast_fp16, y = var_8478_to_fp16)[name = string("aw_chunk_769_cast_fp16")]; + fp16 var_8480_to_fp16 = const()[name = string("op_8480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_771_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_771_cast_fp16, y = var_8480_to_fp16)[name = string("aw_chunk_771_cast_fp16")]; + fp16 var_8482_to_fp16 = const()[name = string("op_8482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_773_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_773_cast_fp16, y = var_8482_to_fp16)[name = string("aw_chunk_773_cast_fp16")]; + fp16 var_8484_to_fp16 = const()[name = string("op_8484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_775_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_775_cast_fp16, y = var_8484_to_fp16)[name = string("aw_chunk_775_cast_fp16")]; + fp16 var_8486_to_fp16 = const()[name = string("op_8486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_777_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_777_cast_fp16, y = var_8486_to_fp16)[name = string("aw_chunk_777_cast_fp16")]; + fp16 var_8488_to_fp16 = const()[name = string("op_8488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_779_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_779_cast_fp16, y = var_8488_to_fp16)[name = string("aw_chunk_779_cast_fp16")]; + fp16 var_8490_to_fp16 = const()[name = string("op_8490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_781_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_781_cast_fp16, y = var_8490_to_fp16)[name = string("aw_chunk_781_cast_fp16")]; + fp16 var_8492_to_fp16 = const()[name = string("op_8492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_783_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_783_cast_fp16, y = var_8492_to_fp16)[name = string("aw_chunk_783_cast_fp16")]; + fp16 var_8494_to_fp16 = const()[name = string("op_8494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_785_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_785_cast_fp16, y = var_8494_to_fp16)[name = string("aw_chunk_785_cast_fp16")]; + fp16 var_8496_to_fp16 = const()[name = string("op_8496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_787_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_787_cast_fp16, y = var_8496_to_fp16)[name = string("aw_chunk_787_cast_fp16")]; + fp16 var_8498_to_fp16 = const()[name = string("op_8498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_789_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_789_cast_fp16, y = var_8498_to_fp16)[name = string("aw_chunk_789_cast_fp16")]; + fp16 var_8500_to_fp16 = const()[name = string("op_8500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_791_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_791_cast_fp16, y = var_8500_to_fp16)[name = string("aw_chunk_791_cast_fp16")]; + fp16 var_8502_to_fp16 = const()[name = string("op_8502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_793_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_793_cast_fp16, y = var_8502_to_fp16)[name = string("aw_chunk_793_cast_fp16")]; + fp16 var_8504_to_fp16 = const()[name = string("op_8504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_795_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_795_cast_fp16, y = var_8504_to_fp16)[name = string("aw_chunk_795_cast_fp16")]; + fp16 var_8506_to_fp16 = const()[name = string("op_8506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_797_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_797_cast_fp16, y = var_8506_to_fp16)[name = string("aw_chunk_797_cast_fp16")]; + fp16 var_8508_to_fp16 = const()[name = string("op_8508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_799_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_799_cast_fp16, y = var_8508_to_fp16)[name = string("aw_chunk_799_cast_fp16")]; + fp16 var_8510_to_fp16 = const()[name = string("op_8510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_801_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_801_cast_fp16, y = var_8510_to_fp16)[name = string("aw_chunk_801_cast_fp16")]; + fp16 var_8512_to_fp16 = const()[name = string("op_8512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_803_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_803_cast_fp16, y = var_8512_to_fp16)[name = string("aw_chunk_803_cast_fp16")]; + fp16 var_8514_to_fp16 = const()[name = string("op_8514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_805_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_805_cast_fp16, y = var_8514_to_fp16)[name = string("aw_chunk_805_cast_fp16")]; + fp16 var_8516_to_fp16 = const()[name = string("op_8516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_807_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_807_cast_fp16, y = var_8516_to_fp16)[name = string("aw_chunk_807_cast_fp16")]; + fp16 var_8518_to_fp16 = const()[name = string("op_8518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_809_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_809_cast_fp16, y = var_8518_to_fp16)[name = string("aw_chunk_809_cast_fp16")]; + fp16 var_8520_to_fp16 = const()[name = string("op_8520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_811_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_811_cast_fp16, y = var_8520_to_fp16)[name = string("aw_chunk_811_cast_fp16")]; + fp16 var_8522_to_fp16 = const()[name = string("op_8522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_813_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_813_cast_fp16, y = var_8522_to_fp16)[name = string("aw_chunk_813_cast_fp16")]; + fp16 var_8524_to_fp16 = const()[name = string("op_8524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_815_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_815_cast_fp16, y = var_8524_to_fp16)[name = string("aw_chunk_815_cast_fp16")]; + fp16 var_8526_to_fp16 = const()[name = string("op_8526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_817_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_817_cast_fp16, y = var_8526_to_fp16)[name = string("aw_chunk_817_cast_fp16")]; + fp16 var_8528_to_fp16 = const()[name = string("op_8528_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_819_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_819_cast_fp16, y = var_8528_to_fp16)[name = string("aw_chunk_819_cast_fp16")]; + fp16 var_8530_to_fp16 = const()[name = string("op_8530_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_821_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_821_cast_fp16, y = var_8530_to_fp16)[name = string("aw_chunk_821_cast_fp16")]; + fp16 var_8532_to_fp16 = const()[name = string("op_8532_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_823_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_823_cast_fp16, y = var_8532_to_fp16)[name = string("aw_chunk_823_cast_fp16")]; + fp16 var_8534_to_fp16 = const()[name = string("op_8534_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_825_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_825_cast_fp16, y = var_8534_to_fp16)[name = string("aw_chunk_825_cast_fp16")]; + fp16 var_8536_to_fp16 = const()[name = string("op_8536_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_827_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_827_cast_fp16, y = var_8536_to_fp16)[name = string("aw_chunk_827_cast_fp16")]; + fp16 var_8538_to_fp16 = const()[name = string("op_8538_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_829_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_829_cast_fp16, y = var_8538_to_fp16)[name = string("aw_chunk_829_cast_fp16")]; + fp16 var_8540_to_fp16 = const()[name = string("op_8540_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_831_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_831_cast_fp16, y = var_8540_to_fp16)[name = string("aw_chunk_831_cast_fp16")]; + fp16 var_8542_to_fp16 = const()[name = string("op_8542_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_833_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_833_cast_fp16, y = var_8542_to_fp16)[name = string("aw_chunk_833_cast_fp16")]; + fp16 var_8544_to_fp16 = const()[name = string("op_8544_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_835_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_835_cast_fp16, y = var_8544_to_fp16)[name = string("aw_chunk_835_cast_fp16")]; + fp16 var_8546_to_fp16 = const()[name = string("op_8546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_837_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_837_cast_fp16, y = var_8546_to_fp16)[name = string("aw_chunk_837_cast_fp16")]; + fp16 var_8548_to_fp16 = const()[name = string("op_8548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_839_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_839_cast_fp16, y = var_8548_to_fp16)[name = string("aw_chunk_839_cast_fp16")]; + fp16 var_8550_to_fp16 = const()[name = string("op_8550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_841_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_841_cast_fp16, y = var_8550_to_fp16)[name = string("aw_chunk_841_cast_fp16")]; + fp16 var_8552_to_fp16 = const()[name = string("op_8552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_843_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_843_cast_fp16, y = var_8552_to_fp16)[name = string("aw_chunk_843_cast_fp16")]; + fp16 var_8554_to_fp16 = const()[name = string("op_8554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_845_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_845_cast_fp16, y = var_8554_to_fp16)[name = string("aw_chunk_845_cast_fp16")]; + fp16 var_8556_to_fp16 = const()[name = string("op_8556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_847_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_847_cast_fp16, y = var_8556_to_fp16)[name = string("aw_chunk_847_cast_fp16")]; + fp16 var_8558_to_fp16 = const()[name = string("op_8558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_849_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_849_cast_fp16, y = var_8558_to_fp16)[name = string("aw_chunk_849_cast_fp16")]; + fp16 var_8560_to_fp16 = const()[name = string("op_8560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_851_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_851_cast_fp16, y = var_8560_to_fp16)[name = string("aw_chunk_851_cast_fp16")]; + fp16 var_8562_to_fp16 = const()[name = string("op_8562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_853_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_853_cast_fp16, y = var_8562_to_fp16)[name = string("aw_chunk_853_cast_fp16")]; + fp16 var_8564_to_fp16 = const()[name = string("op_8564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_855_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_855_cast_fp16, y = var_8564_to_fp16)[name = string("aw_chunk_855_cast_fp16")]; + fp16 var_8566_to_fp16 = const()[name = string("op_8566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_857_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_857_cast_fp16, y = var_8566_to_fp16)[name = string("aw_chunk_857_cast_fp16")]; + fp16 var_8568_to_fp16 = const()[name = string("op_8568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_859_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_859_cast_fp16, y = var_8568_to_fp16)[name = string("aw_chunk_859_cast_fp16")]; + fp16 var_8570_to_fp16 = const()[name = string("op_8570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_861_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_861_cast_fp16, y = var_8570_to_fp16)[name = string("aw_chunk_861_cast_fp16")]; + fp16 var_8572_to_fp16 = const()[name = string("op_8572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_863_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_863_cast_fp16, y = var_8572_to_fp16)[name = string("aw_chunk_863_cast_fp16")]; + tensor var_8574_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_769_cast_fp16)[name = string("op_8574_cast_fp16")]; + tensor var_8575_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_771_cast_fp16)[name = string("op_8575_cast_fp16")]; + tensor var_8576_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_773_cast_fp16)[name = string("op_8576_cast_fp16")]; + tensor var_8577_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_775_cast_fp16)[name = string("op_8577_cast_fp16")]; + tensor var_8578_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_777_cast_fp16)[name = string("op_8578_cast_fp16")]; + tensor var_8579_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_779_cast_fp16)[name = string("op_8579_cast_fp16")]; + tensor var_8580_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_781_cast_fp16)[name = string("op_8580_cast_fp16")]; + tensor var_8581_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_783_cast_fp16)[name = string("op_8581_cast_fp16")]; + tensor var_8582_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_785_cast_fp16)[name = string("op_8582_cast_fp16")]; + tensor var_8583_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_787_cast_fp16)[name = string("op_8583_cast_fp16")]; + tensor var_8584_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_789_cast_fp16)[name = string("op_8584_cast_fp16")]; + tensor var_8585_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_791_cast_fp16)[name = string("op_8585_cast_fp16")]; + tensor var_8586_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_793_cast_fp16)[name = string("op_8586_cast_fp16")]; + tensor var_8587_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_795_cast_fp16)[name = string("op_8587_cast_fp16")]; + tensor var_8588_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_797_cast_fp16)[name = string("op_8588_cast_fp16")]; + tensor var_8589_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_799_cast_fp16)[name = string("op_8589_cast_fp16")]; + tensor var_8590_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_801_cast_fp16)[name = string("op_8590_cast_fp16")]; + tensor var_8591_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_803_cast_fp16)[name = string("op_8591_cast_fp16")]; + tensor var_8592_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_805_cast_fp16)[name = string("op_8592_cast_fp16")]; + tensor var_8593_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_807_cast_fp16)[name = string("op_8593_cast_fp16")]; + tensor var_8594_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_809_cast_fp16)[name = string("op_8594_cast_fp16")]; + tensor var_8595_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_811_cast_fp16)[name = string("op_8595_cast_fp16")]; + tensor var_8596_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_813_cast_fp16)[name = string("op_8596_cast_fp16")]; + tensor var_8597_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_815_cast_fp16)[name = string("op_8597_cast_fp16")]; + tensor var_8598_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_817_cast_fp16)[name = string("op_8598_cast_fp16")]; + tensor var_8599_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_819_cast_fp16)[name = string("op_8599_cast_fp16")]; + tensor var_8600_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_821_cast_fp16)[name = string("op_8600_cast_fp16")]; + tensor var_8601_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_823_cast_fp16)[name = string("op_8601_cast_fp16")]; + tensor var_8602_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_825_cast_fp16)[name = string("op_8602_cast_fp16")]; + tensor var_8603_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_827_cast_fp16)[name = string("op_8603_cast_fp16")]; + tensor var_8604_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_829_cast_fp16)[name = string("op_8604_cast_fp16")]; + tensor var_8605_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_831_cast_fp16)[name = string("op_8605_cast_fp16")]; + tensor var_8606_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_833_cast_fp16)[name = string("op_8606_cast_fp16")]; + tensor var_8607_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_835_cast_fp16)[name = string("op_8607_cast_fp16")]; + tensor var_8608_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_837_cast_fp16)[name = string("op_8608_cast_fp16")]; + tensor var_8609_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_839_cast_fp16)[name = string("op_8609_cast_fp16")]; + tensor var_8610_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_841_cast_fp16)[name = string("op_8610_cast_fp16")]; + tensor var_8611_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_843_cast_fp16)[name = string("op_8611_cast_fp16")]; + tensor var_8612_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_845_cast_fp16)[name = string("op_8612_cast_fp16")]; + tensor var_8613_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_847_cast_fp16)[name = string("op_8613_cast_fp16")]; + tensor var_8614_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_849_cast_fp16)[name = string("op_8614_cast_fp16")]; + tensor var_8615_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_851_cast_fp16)[name = string("op_8615_cast_fp16")]; + tensor var_8616_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_853_cast_fp16)[name = string("op_8616_cast_fp16")]; + tensor var_8617_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_855_cast_fp16)[name = string("op_8617_cast_fp16")]; + tensor var_8618_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_857_cast_fp16)[name = string("op_8618_cast_fp16")]; + tensor var_8619_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_859_cast_fp16)[name = string("op_8619_cast_fp16")]; + tensor var_8620_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_861_cast_fp16)[name = string("op_8620_cast_fp16")]; + tensor var_8621_cast_fp16 = softmax(axis = var_7847, x = aw_chunk_863_cast_fp16)[name = string("op_8621_cast_fp16")]; + string var_8623_equation_0 = const()[name = string("op_8623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8623_cast_fp16 = einsum(equation = var_8623_equation_0, values = (var_8335_cast_fp16, var_8574_cast_fp16))[name = string("op_8623_cast_fp16")]; + string var_8625_equation_0 = const()[name = string("op_8625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8625_cast_fp16 = einsum(equation = var_8625_equation_0, values = (var_8335_cast_fp16, var_8575_cast_fp16))[name = string("op_8625_cast_fp16")]; + string var_8627_equation_0 = const()[name = string("op_8627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8627_cast_fp16 = einsum(equation = var_8627_equation_0, values = (var_8335_cast_fp16, var_8576_cast_fp16))[name = string("op_8627_cast_fp16")]; + string var_8629_equation_0 = const()[name = string("op_8629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8629_cast_fp16 = einsum(equation = var_8629_equation_0, values = (var_8335_cast_fp16, var_8577_cast_fp16))[name = string("op_8629_cast_fp16")]; + string var_8631_equation_0 = const()[name = string("op_8631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8631_cast_fp16 = einsum(equation = var_8631_equation_0, values = (var_8339_cast_fp16, var_8578_cast_fp16))[name = string("op_8631_cast_fp16")]; + string var_8633_equation_0 = const()[name = string("op_8633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8633_cast_fp16 = einsum(equation = var_8633_equation_0, values = (var_8339_cast_fp16, var_8579_cast_fp16))[name = string("op_8633_cast_fp16")]; + string var_8635_equation_0 = const()[name = string("op_8635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8635_cast_fp16 = einsum(equation = var_8635_equation_0, values = (var_8339_cast_fp16, var_8580_cast_fp16))[name = string("op_8635_cast_fp16")]; + string var_8637_equation_0 = const()[name = string("op_8637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8637_cast_fp16 = einsum(equation = var_8637_equation_0, values = (var_8339_cast_fp16, var_8581_cast_fp16))[name = string("op_8637_cast_fp16")]; + string var_8639_equation_0 = const()[name = string("op_8639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8639_cast_fp16 = einsum(equation = var_8639_equation_0, values = (var_8343_cast_fp16, var_8582_cast_fp16))[name = string("op_8639_cast_fp16")]; + string var_8641_equation_0 = const()[name = string("op_8641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8641_cast_fp16 = einsum(equation = var_8641_equation_0, values = (var_8343_cast_fp16, var_8583_cast_fp16))[name = string("op_8641_cast_fp16")]; + string var_8643_equation_0 = const()[name = string("op_8643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8643_cast_fp16 = einsum(equation = var_8643_equation_0, values = (var_8343_cast_fp16, var_8584_cast_fp16))[name = string("op_8643_cast_fp16")]; + string var_8645_equation_0 = const()[name = string("op_8645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8645_cast_fp16 = einsum(equation = var_8645_equation_0, values = (var_8343_cast_fp16, var_8585_cast_fp16))[name = string("op_8645_cast_fp16")]; + string var_8647_equation_0 = const()[name = string("op_8647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8647_cast_fp16 = einsum(equation = var_8647_equation_0, values = (var_8347_cast_fp16, var_8586_cast_fp16))[name = string("op_8647_cast_fp16")]; + string var_8649_equation_0 = const()[name = string("op_8649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8649_cast_fp16 = einsum(equation = var_8649_equation_0, values = (var_8347_cast_fp16, var_8587_cast_fp16))[name = string("op_8649_cast_fp16")]; + string var_8651_equation_0 = const()[name = string("op_8651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8651_cast_fp16 = einsum(equation = var_8651_equation_0, values = (var_8347_cast_fp16, var_8588_cast_fp16))[name = string("op_8651_cast_fp16")]; + string var_8653_equation_0 = const()[name = string("op_8653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8653_cast_fp16 = einsum(equation = var_8653_equation_0, values = (var_8347_cast_fp16, var_8589_cast_fp16))[name = string("op_8653_cast_fp16")]; + string var_8655_equation_0 = const()[name = string("op_8655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8655_cast_fp16 = einsum(equation = var_8655_equation_0, values = (var_8351_cast_fp16, var_8590_cast_fp16))[name = string("op_8655_cast_fp16")]; + string var_8657_equation_0 = const()[name = string("op_8657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8657_cast_fp16 = einsum(equation = var_8657_equation_0, values = (var_8351_cast_fp16, var_8591_cast_fp16))[name = string("op_8657_cast_fp16")]; + string var_8659_equation_0 = const()[name = string("op_8659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8659_cast_fp16 = einsum(equation = var_8659_equation_0, values = (var_8351_cast_fp16, var_8592_cast_fp16))[name = string("op_8659_cast_fp16")]; + string var_8661_equation_0 = const()[name = string("op_8661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8661_cast_fp16 = einsum(equation = var_8661_equation_0, values = (var_8351_cast_fp16, var_8593_cast_fp16))[name = string("op_8661_cast_fp16")]; + string var_8663_equation_0 = const()[name = string("op_8663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8663_cast_fp16 = einsum(equation = var_8663_equation_0, values = (var_8355_cast_fp16, var_8594_cast_fp16))[name = string("op_8663_cast_fp16")]; + string var_8665_equation_0 = const()[name = string("op_8665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8665_cast_fp16 = einsum(equation = var_8665_equation_0, values = (var_8355_cast_fp16, var_8595_cast_fp16))[name = string("op_8665_cast_fp16")]; + string var_8667_equation_0 = const()[name = string("op_8667_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8667_cast_fp16 = einsum(equation = var_8667_equation_0, values = (var_8355_cast_fp16, var_8596_cast_fp16))[name = string("op_8667_cast_fp16")]; + string var_8669_equation_0 = const()[name = string("op_8669_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8669_cast_fp16 = einsum(equation = var_8669_equation_0, values = (var_8355_cast_fp16, var_8597_cast_fp16))[name = string("op_8669_cast_fp16")]; + string var_8671_equation_0 = const()[name = string("op_8671_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8671_cast_fp16 = einsum(equation = var_8671_equation_0, values = (var_8359_cast_fp16, var_8598_cast_fp16))[name = string("op_8671_cast_fp16")]; + string var_8673_equation_0 = const()[name = string("op_8673_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8673_cast_fp16 = einsum(equation = var_8673_equation_0, values = (var_8359_cast_fp16, var_8599_cast_fp16))[name = string("op_8673_cast_fp16")]; + string var_8675_equation_0 = const()[name = string("op_8675_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8675_cast_fp16 = einsum(equation = var_8675_equation_0, values = (var_8359_cast_fp16, var_8600_cast_fp16))[name = string("op_8675_cast_fp16")]; + string var_8677_equation_0 = const()[name = string("op_8677_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8677_cast_fp16 = einsum(equation = var_8677_equation_0, values = (var_8359_cast_fp16, var_8601_cast_fp16))[name = string("op_8677_cast_fp16")]; + string var_8679_equation_0 = const()[name = string("op_8679_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8679_cast_fp16 = einsum(equation = var_8679_equation_0, values = (var_8363_cast_fp16, var_8602_cast_fp16))[name = string("op_8679_cast_fp16")]; + string var_8681_equation_0 = const()[name = string("op_8681_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8681_cast_fp16 = einsum(equation = var_8681_equation_0, values = (var_8363_cast_fp16, var_8603_cast_fp16))[name = string("op_8681_cast_fp16")]; + string var_8683_equation_0 = const()[name = string("op_8683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8683_cast_fp16 = einsum(equation = var_8683_equation_0, values = (var_8363_cast_fp16, var_8604_cast_fp16))[name = string("op_8683_cast_fp16")]; + string var_8685_equation_0 = const()[name = string("op_8685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8685_cast_fp16 = einsum(equation = var_8685_equation_0, values = (var_8363_cast_fp16, var_8605_cast_fp16))[name = string("op_8685_cast_fp16")]; + string var_8687_equation_0 = const()[name = string("op_8687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8687_cast_fp16 = einsum(equation = var_8687_equation_0, values = (var_8367_cast_fp16, var_8606_cast_fp16))[name = string("op_8687_cast_fp16")]; + string var_8689_equation_0 = const()[name = string("op_8689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8689_cast_fp16 = einsum(equation = var_8689_equation_0, values = (var_8367_cast_fp16, var_8607_cast_fp16))[name = string("op_8689_cast_fp16")]; + string var_8691_equation_0 = const()[name = string("op_8691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8691_cast_fp16 = einsum(equation = var_8691_equation_0, values = (var_8367_cast_fp16, var_8608_cast_fp16))[name = string("op_8691_cast_fp16")]; + string var_8693_equation_0 = const()[name = string("op_8693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8693_cast_fp16 = einsum(equation = var_8693_equation_0, values = (var_8367_cast_fp16, var_8609_cast_fp16))[name = string("op_8693_cast_fp16")]; + string var_8695_equation_0 = const()[name = string("op_8695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8695_cast_fp16 = einsum(equation = var_8695_equation_0, values = (var_8371_cast_fp16, var_8610_cast_fp16))[name = string("op_8695_cast_fp16")]; + string var_8697_equation_0 = const()[name = string("op_8697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8697_cast_fp16 = einsum(equation = var_8697_equation_0, values = (var_8371_cast_fp16, var_8611_cast_fp16))[name = string("op_8697_cast_fp16")]; + string var_8699_equation_0 = const()[name = string("op_8699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8699_cast_fp16 = einsum(equation = var_8699_equation_0, values = (var_8371_cast_fp16, var_8612_cast_fp16))[name = string("op_8699_cast_fp16")]; + string var_8701_equation_0 = const()[name = string("op_8701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8701_cast_fp16 = einsum(equation = var_8701_equation_0, values = (var_8371_cast_fp16, var_8613_cast_fp16))[name = string("op_8701_cast_fp16")]; + string var_8703_equation_0 = const()[name = string("op_8703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8703_cast_fp16 = einsum(equation = var_8703_equation_0, values = (var_8375_cast_fp16, var_8614_cast_fp16))[name = string("op_8703_cast_fp16")]; + string var_8705_equation_0 = const()[name = string("op_8705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8705_cast_fp16 = einsum(equation = var_8705_equation_0, values = (var_8375_cast_fp16, var_8615_cast_fp16))[name = string("op_8705_cast_fp16")]; + string var_8707_equation_0 = const()[name = string("op_8707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8707_cast_fp16 = einsum(equation = var_8707_equation_0, values = (var_8375_cast_fp16, var_8616_cast_fp16))[name = string("op_8707_cast_fp16")]; + string var_8709_equation_0 = const()[name = string("op_8709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8709_cast_fp16 = einsum(equation = var_8709_equation_0, values = (var_8375_cast_fp16, var_8617_cast_fp16))[name = string("op_8709_cast_fp16")]; + string var_8711_equation_0 = const()[name = string("op_8711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8711_cast_fp16 = einsum(equation = var_8711_equation_0, values = (var_8379_cast_fp16, var_8618_cast_fp16))[name = string("op_8711_cast_fp16")]; + string var_8713_equation_0 = const()[name = string("op_8713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8713_cast_fp16 = einsum(equation = var_8713_equation_0, values = (var_8379_cast_fp16, var_8619_cast_fp16))[name = string("op_8713_cast_fp16")]; + string var_8715_equation_0 = const()[name = string("op_8715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8715_cast_fp16 = einsum(equation = var_8715_equation_0, values = (var_8379_cast_fp16, var_8620_cast_fp16))[name = string("op_8715_cast_fp16")]; + string var_8717_equation_0 = const()[name = string("op_8717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_8717_cast_fp16 = einsum(equation = var_8717_equation_0, values = (var_8379_cast_fp16, var_8621_cast_fp16))[name = string("op_8717_cast_fp16")]; + bool var_8719_interleave_0 = const()[name = string("op_8719_interleave_0"), val = bool(false)]; + tensor var_8719_cast_fp16 = concat(axis = var_7830, interleave = var_8719_interleave_0, values = (var_8623_cast_fp16, var_8625_cast_fp16, var_8627_cast_fp16, var_8629_cast_fp16))[name = string("op_8719_cast_fp16")]; + bool var_8721_interleave_0 = const()[name = string("op_8721_interleave_0"), val = bool(false)]; + tensor var_8721_cast_fp16 = concat(axis = var_7830, interleave = var_8721_interleave_0, values = (var_8631_cast_fp16, var_8633_cast_fp16, var_8635_cast_fp16, var_8637_cast_fp16))[name = string("op_8721_cast_fp16")]; + bool var_8723_interleave_0 = const()[name = string("op_8723_interleave_0"), val = bool(false)]; + tensor var_8723_cast_fp16 = concat(axis = var_7830, interleave = var_8723_interleave_0, values = (var_8639_cast_fp16, var_8641_cast_fp16, var_8643_cast_fp16, var_8645_cast_fp16))[name = string("op_8723_cast_fp16")]; + bool var_8725_interleave_0 = const()[name = string("op_8725_interleave_0"), val = bool(false)]; + tensor var_8725_cast_fp16 = concat(axis = var_7830, interleave = var_8725_interleave_0, values = (var_8647_cast_fp16, var_8649_cast_fp16, var_8651_cast_fp16, var_8653_cast_fp16))[name = string("op_8725_cast_fp16")]; + bool var_8727_interleave_0 = const()[name = string("op_8727_interleave_0"), val = bool(false)]; + tensor var_8727_cast_fp16 = concat(axis = var_7830, interleave = var_8727_interleave_0, values = (var_8655_cast_fp16, var_8657_cast_fp16, var_8659_cast_fp16, var_8661_cast_fp16))[name = string("op_8727_cast_fp16")]; + bool var_8729_interleave_0 = const()[name = string("op_8729_interleave_0"), val = bool(false)]; + tensor var_8729_cast_fp16 = concat(axis = var_7830, interleave = var_8729_interleave_0, values = (var_8663_cast_fp16, var_8665_cast_fp16, var_8667_cast_fp16, var_8669_cast_fp16))[name = string("op_8729_cast_fp16")]; + bool var_8731_interleave_0 = const()[name = string("op_8731_interleave_0"), val = bool(false)]; + tensor var_8731_cast_fp16 = concat(axis = var_7830, interleave = var_8731_interleave_0, values = (var_8671_cast_fp16, var_8673_cast_fp16, var_8675_cast_fp16, var_8677_cast_fp16))[name = string("op_8731_cast_fp16")]; + bool var_8733_interleave_0 = const()[name = string("op_8733_interleave_0"), val = bool(false)]; + tensor var_8733_cast_fp16 = concat(axis = var_7830, interleave = var_8733_interleave_0, values = (var_8679_cast_fp16, var_8681_cast_fp16, var_8683_cast_fp16, var_8685_cast_fp16))[name = string("op_8733_cast_fp16")]; + bool var_8735_interleave_0 = const()[name = string("op_8735_interleave_0"), val = bool(false)]; + tensor var_8735_cast_fp16 = concat(axis = var_7830, interleave = var_8735_interleave_0, values = (var_8687_cast_fp16, var_8689_cast_fp16, var_8691_cast_fp16, var_8693_cast_fp16))[name = string("op_8735_cast_fp16")]; + bool var_8737_interleave_0 = const()[name = string("op_8737_interleave_0"), val = bool(false)]; + tensor var_8737_cast_fp16 = concat(axis = var_7830, interleave = var_8737_interleave_0, values = (var_8695_cast_fp16, var_8697_cast_fp16, var_8699_cast_fp16, var_8701_cast_fp16))[name = string("op_8737_cast_fp16")]; + bool var_8739_interleave_0 = const()[name = string("op_8739_interleave_0"), val = bool(false)]; + tensor var_8739_cast_fp16 = concat(axis = var_7830, interleave = var_8739_interleave_0, values = (var_8703_cast_fp16, var_8705_cast_fp16, var_8707_cast_fp16, var_8709_cast_fp16))[name = string("op_8739_cast_fp16")]; + bool var_8741_interleave_0 = const()[name = string("op_8741_interleave_0"), val = bool(false)]; + tensor var_8741_cast_fp16 = concat(axis = var_7830, interleave = var_8741_interleave_0, values = (var_8711_cast_fp16, var_8713_cast_fp16, var_8715_cast_fp16, var_8717_cast_fp16))[name = string("op_8741_cast_fp16")]; + bool input_65_interleave_0 = const()[name = string("input_65_interleave_0"), val = bool(false)]; + tensor input_65_cast_fp16 = concat(axis = var_7847, interleave = input_65_interleave_0, values = (var_8719_cast_fp16, var_8721_cast_fp16, var_8723_cast_fp16, var_8725_cast_fp16, var_8727_cast_fp16, var_8729_cast_fp16, var_8731_cast_fp16, var_8733_cast_fp16, var_8735_cast_fp16, var_8737_cast_fp16, var_8739_cast_fp16, var_8741_cast_fp16))[name = string("input_65_cast_fp16")]; + string obj_35_pad_type_0 = const()[name = string("obj_35_pad_type_0"), val = string("valid")]; + tensor obj_35_strides_0 = const()[name = string("obj_35_strides_0"), val = tensor([1, 1])]; + tensor obj_35_pad_0 = const()[name = string("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_35_dilations_0 = const()[name = string("obj_35_dilations_0"), val = tensor([1, 1])]; + int32 obj_35_groups_0 = const()[name = string("obj_35_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123165120)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124344832)))]; + tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_8760_to_fp16 = const()[name = string("op_8760_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_8760_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_67_gamma_0_to_fp16 = const()[name = string("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124346432)))]; + tensor input_67_beta_0_to_fp16 = const()[name = string("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124348032)))]; + fp16 input_67_epsilon_0_to_fp16 = const()[name = string("input_67_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_pad_type_0 = const()[name = string("input_69_pad_type_0"), val = string("valid")]; + tensor input_69_strides_0 = const()[name = string("input_69_strides_0"), val = tensor([1, 1])]; + tensor input_69_pad_0 = const()[name = string("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_69_dilations_0 = const()[name = string("input_69_dilations_0"), val = tensor([1, 1])]; + int32 input_69_groups_0 = const()[name = string("input_69_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124349632)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129068288)))]; + tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string input_71_mode_0 = const()[name = string("input_71_mode_0"), val = string("EXACT")]; + tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129074496)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133793152)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_71_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_37_cast_fp16")]; + int32 var_8789 = const()[name = string("op_8789"), val = int32(3)]; + int32 var_8806 = const()[name = string("op_8806"), val = int32(1)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_8823_to_fp16 = const()[name = string("op_8823_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_8823_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = string("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133794752)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = string("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133796352)))]; + fp16 obj_37_epsilon_0_to_fp16 = const()[name = string("obj_37_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_37_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133797952)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134977664)))]; + tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("query_19_cast_fp16")]; + string key_19_pad_type_0 = const()[name = string("key_19_pad_type_0"), val = string("valid")]; + tensor key_19_strides_0 = const()[name = string("key_19_strides_0"), val = tensor([1, 1])]; + tensor key_19_pad_0 = const()[name = string("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_19_dilations_0 = const()[name = string("key_19_dilations_0"), val = tensor([1, 1])]; + int32 key_19_groups_0 = const()[name = string("key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134979264)))]; + tensor key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("key_19_cast_fp16")]; + string value_19_pad_type_0 = const()[name = string("value_19_pad_type_0"), val = string("valid")]; + tensor value_19_strides_0 = const()[name = string("value_19_strides_0"), val = tensor([1, 1])]; + tensor value_19_pad_0 = const()[name = string("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_19_dilations_0 = const()[name = string("value_19_dilations_0"), val = tensor([1, 1])]; + int32 value_19_groups_0 = const()[name = string("value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136158976)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137338688)))]; + tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_8861_begin_0 = const()[name = string("op_8861_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8861_end_0 = const()[name = string("op_8861_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8861_end_mask_0 = const()[name = string("op_8861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8861_cast_fp16 = slice_by_index(begin = var_8861_begin_0, end = var_8861_end_0, end_mask = var_8861_end_mask_0, x = query_19_cast_fp16)[name = string("op_8861_cast_fp16")]; + tensor var_8865_begin_0 = const()[name = string("op_8865_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8865_end_0 = const()[name = string("op_8865_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8865_end_mask_0 = const()[name = string("op_8865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8865_cast_fp16 = slice_by_index(begin = var_8865_begin_0, end = var_8865_end_0, end_mask = var_8865_end_mask_0, x = query_19_cast_fp16)[name = string("op_8865_cast_fp16")]; + tensor var_8869_begin_0 = const()[name = string("op_8869_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8869_end_0 = const()[name = string("op_8869_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8869_end_mask_0 = const()[name = string("op_8869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8869_cast_fp16 = slice_by_index(begin = var_8869_begin_0, end = var_8869_end_0, end_mask = var_8869_end_mask_0, x = query_19_cast_fp16)[name = string("op_8869_cast_fp16")]; + tensor var_8873_begin_0 = const()[name = string("op_8873_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8873_end_0 = const()[name = string("op_8873_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8873_end_mask_0 = const()[name = string("op_8873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8873_cast_fp16 = slice_by_index(begin = var_8873_begin_0, end = var_8873_end_0, end_mask = var_8873_end_mask_0, x = query_19_cast_fp16)[name = string("op_8873_cast_fp16")]; + tensor var_8877_begin_0 = const()[name = string("op_8877_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8877_end_0 = const()[name = string("op_8877_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8877_end_mask_0 = const()[name = string("op_8877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8877_cast_fp16 = slice_by_index(begin = var_8877_begin_0, end = var_8877_end_0, end_mask = var_8877_end_mask_0, x = query_19_cast_fp16)[name = string("op_8877_cast_fp16")]; + tensor var_8881_begin_0 = const()[name = string("op_8881_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8881_end_0 = const()[name = string("op_8881_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8881_end_mask_0 = const()[name = string("op_8881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8881_cast_fp16 = slice_by_index(begin = var_8881_begin_0, end = var_8881_end_0, end_mask = var_8881_end_mask_0, x = query_19_cast_fp16)[name = string("op_8881_cast_fp16")]; + tensor var_8885_begin_0 = const()[name = string("op_8885_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8885_end_0 = const()[name = string("op_8885_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8885_end_mask_0 = const()[name = string("op_8885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8885_cast_fp16 = slice_by_index(begin = var_8885_begin_0, end = var_8885_end_0, end_mask = var_8885_end_mask_0, x = query_19_cast_fp16)[name = string("op_8885_cast_fp16")]; + tensor var_8889_begin_0 = const()[name = string("op_8889_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8889_end_0 = const()[name = string("op_8889_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8889_end_mask_0 = const()[name = string("op_8889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8889_cast_fp16 = slice_by_index(begin = var_8889_begin_0, end = var_8889_end_0, end_mask = var_8889_end_mask_0, x = query_19_cast_fp16)[name = string("op_8889_cast_fp16")]; + tensor var_8893_begin_0 = const()[name = string("op_8893_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8893_end_0 = const()[name = string("op_8893_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8893_end_mask_0 = const()[name = string("op_8893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8893_cast_fp16 = slice_by_index(begin = var_8893_begin_0, end = var_8893_end_0, end_mask = var_8893_end_mask_0, x = query_19_cast_fp16)[name = string("op_8893_cast_fp16")]; + tensor var_8897_begin_0 = const()[name = string("op_8897_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8897_end_0 = const()[name = string("op_8897_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8897_end_mask_0 = const()[name = string("op_8897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8897_cast_fp16 = slice_by_index(begin = var_8897_begin_0, end = var_8897_end_0, end_mask = var_8897_end_mask_0, x = query_19_cast_fp16)[name = string("op_8897_cast_fp16")]; + tensor var_8901_begin_0 = const()[name = string("op_8901_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8901_end_0 = const()[name = string("op_8901_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8901_end_mask_0 = const()[name = string("op_8901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8901_cast_fp16 = slice_by_index(begin = var_8901_begin_0, end = var_8901_end_0, end_mask = var_8901_end_mask_0, x = query_19_cast_fp16)[name = string("op_8901_cast_fp16")]; + tensor var_8905_begin_0 = const()[name = string("op_8905_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8905_end_0 = const()[name = string("op_8905_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8905_end_mask_0 = const()[name = string("op_8905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8905_cast_fp16 = slice_by_index(begin = var_8905_begin_0, end = var_8905_end_0, end_mask = var_8905_end_mask_0, x = query_19_cast_fp16)[name = string("op_8905_cast_fp16")]; + tensor var_8914_begin_0 = const()[name = string("op_8914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8914_end_0 = const()[name = string("op_8914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8914_end_mask_0 = const()[name = string("op_8914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8914_cast_fp16 = slice_by_index(begin = var_8914_begin_0, end = var_8914_end_0, end_mask = var_8914_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8914_cast_fp16")]; + tensor var_8921_begin_0 = const()[name = string("op_8921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8921_end_0 = const()[name = string("op_8921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8921_end_mask_0 = const()[name = string("op_8921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8921_cast_fp16 = slice_by_index(begin = var_8921_begin_0, end = var_8921_end_0, end_mask = var_8921_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8921_cast_fp16")]; + tensor var_8928_begin_0 = const()[name = string("op_8928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8928_end_0 = const()[name = string("op_8928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8928_end_mask_0 = const()[name = string("op_8928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8928_cast_fp16 = slice_by_index(begin = var_8928_begin_0, end = var_8928_end_0, end_mask = var_8928_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8928_cast_fp16")]; + tensor var_8935_begin_0 = const()[name = string("op_8935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8935_end_0 = const()[name = string("op_8935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8935_end_mask_0 = const()[name = string("op_8935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8935_cast_fp16 = slice_by_index(begin = var_8935_begin_0, end = var_8935_end_0, end_mask = var_8935_end_mask_0, x = var_8861_cast_fp16)[name = string("op_8935_cast_fp16")]; + tensor var_8942_begin_0 = const()[name = string("op_8942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8942_end_0 = const()[name = string("op_8942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8942_end_mask_0 = const()[name = string("op_8942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8942_cast_fp16 = slice_by_index(begin = var_8942_begin_0, end = var_8942_end_0, end_mask = var_8942_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8942_cast_fp16")]; + tensor var_8949_begin_0 = const()[name = string("op_8949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8949_end_0 = const()[name = string("op_8949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8949_end_mask_0 = const()[name = string("op_8949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8949_cast_fp16 = slice_by_index(begin = var_8949_begin_0, end = var_8949_end_0, end_mask = var_8949_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8949_cast_fp16")]; + tensor var_8956_begin_0 = const()[name = string("op_8956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8956_end_0 = const()[name = string("op_8956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8956_end_mask_0 = const()[name = string("op_8956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8956_cast_fp16 = slice_by_index(begin = var_8956_begin_0, end = var_8956_end_0, end_mask = var_8956_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8956_cast_fp16")]; + tensor var_8963_begin_0 = const()[name = string("op_8963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8963_end_0 = const()[name = string("op_8963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8963_end_mask_0 = const()[name = string("op_8963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8963_cast_fp16 = slice_by_index(begin = var_8963_begin_0, end = var_8963_end_0, end_mask = var_8963_end_mask_0, x = var_8865_cast_fp16)[name = string("op_8963_cast_fp16")]; + tensor var_8970_begin_0 = const()[name = string("op_8970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8970_end_0 = const()[name = string("op_8970_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8970_end_mask_0 = const()[name = string("op_8970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8970_cast_fp16 = slice_by_index(begin = var_8970_begin_0, end = var_8970_end_0, end_mask = var_8970_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8970_cast_fp16")]; + tensor var_8977_begin_0 = const()[name = string("op_8977_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8977_end_0 = const()[name = string("op_8977_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8977_end_mask_0 = const()[name = string("op_8977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8977_cast_fp16 = slice_by_index(begin = var_8977_begin_0, end = var_8977_end_0, end_mask = var_8977_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8977_cast_fp16")]; + tensor var_8984_begin_0 = const()[name = string("op_8984_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8984_end_0 = const()[name = string("op_8984_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8984_end_mask_0 = const()[name = string("op_8984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8984_cast_fp16 = slice_by_index(begin = var_8984_begin_0, end = var_8984_end_0, end_mask = var_8984_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8984_cast_fp16")]; + tensor var_8991_begin_0 = const()[name = string("op_8991_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8991_end_0 = const()[name = string("op_8991_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8991_end_mask_0 = const()[name = string("op_8991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8991_cast_fp16 = slice_by_index(begin = var_8991_begin_0, end = var_8991_end_0, end_mask = var_8991_end_mask_0, x = var_8869_cast_fp16)[name = string("op_8991_cast_fp16")]; + tensor var_8998_begin_0 = const()[name = string("op_8998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8998_end_0 = const()[name = string("op_8998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8998_end_mask_0 = const()[name = string("op_8998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8998_cast_fp16 = slice_by_index(begin = var_8998_begin_0, end = var_8998_end_0, end_mask = var_8998_end_mask_0, x = var_8873_cast_fp16)[name = string("op_8998_cast_fp16")]; + tensor var_9005_begin_0 = const()[name = string("op_9005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9005_end_0 = const()[name = string("op_9005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9005_end_mask_0 = const()[name = string("op_9005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9005_cast_fp16 = slice_by_index(begin = var_9005_begin_0, end = var_9005_end_0, end_mask = var_9005_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9005_cast_fp16")]; + tensor var_9012_begin_0 = const()[name = string("op_9012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9012_end_0 = const()[name = string("op_9012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9012_end_mask_0 = const()[name = string("op_9012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9012_cast_fp16 = slice_by_index(begin = var_9012_begin_0, end = var_9012_end_0, end_mask = var_9012_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9012_cast_fp16")]; + tensor var_9019_begin_0 = const()[name = string("op_9019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9019_end_0 = const()[name = string("op_9019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9019_end_mask_0 = const()[name = string("op_9019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9019_cast_fp16 = slice_by_index(begin = var_9019_begin_0, end = var_9019_end_0, end_mask = var_9019_end_mask_0, x = var_8873_cast_fp16)[name = string("op_9019_cast_fp16")]; + tensor var_9026_begin_0 = const()[name = string("op_9026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9026_end_0 = const()[name = string("op_9026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9026_end_mask_0 = const()[name = string("op_9026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9026_cast_fp16 = slice_by_index(begin = var_9026_begin_0, end = var_9026_end_0, end_mask = var_9026_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9026_cast_fp16")]; + tensor var_9033_begin_0 = const()[name = string("op_9033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9033_end_0 = const()[name = string("op_9033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9033_end_mask_0 = const()[name = string("op_9033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9033_cast_fp16 = slice_by_index(begin = var_9033_begin_0, end = var_9033_end_0, end_mask = var_9033_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9033_cast_fp16")]; + tensor var_9040_begin_0 = const()[name = string("op_9040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9040_end_0 = const()[name = string("op_9040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9040_end_mask_0 = const()[name = string("op_9040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9040_cast_fp16 = slice_by_index(begin = var_9040_begin_0, end = var_9040_end_0, end_mask = var_9040_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9040_cast_fp16")]; + tensor var_9047_begin_0 = const()[name = string("op_9047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9047_end_0 = const()[name = string("op_9047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9047_end_mask_0 = const()[name = string("op_9047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9047_cast_fp16 = slice_by_index(begin = var_9047_begin_0, end = var_9047_end_0, end_mask = var_9047_end_mask_0, x = var_8877_cast_fp16)[name = string("op_9047_cast_fp16")]; + tensor var_9054_begin_0 = const()[name = string("op_9054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9054_end_0 = const()[name = string("op_9054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9054_end_mask_0 = const()[name = string("op_9054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9054_cast_fp16 = slice_by_index(begin = var_9054_begin_0, end = var_9054_end_0, end_mask = var_9054_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9054_cast_fp16")]; + tensor var_9061_begin_0 = const()[name = string("op_9061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9061_end_0 = const()[name = string("op_9061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9061_end_mask_0 = const()[name = string("op_9061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9061_cast_fp16 = slice_by_index(begin = var_9061_begin_0, end = var_9061_end_0, end_mask = var_9061_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9061_cast_fp16")]; + tensor var_9068_begin_0 = const()[name = string("op_9068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9068_end_0 = const()[name = string("op_9068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9068_end_mask_0 = const()[name = string("op_9068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9068_cast_fp16 = slice_by_index(begin = var_9068_begin_0, end = var_9068_end_0, end_mask = var_9068_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9068_cast_fp16")]; + tensor var_9075_begin_0 = const()[name = string("op_9075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9075_end_0 = const()[name = string("op_9075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9075_end_mask_0 = const()[name = string("op_9075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9075_cast_fp16 = slice_by_index(begin = var_9075_begin_0, end = var_9075_end_0, end_mask = var_9075_end_mask_0, x = var_8881_cast_fp16)[name = string("op_9075_cast_fp16")]; + tensor var_9082_begin_0 = const()[name = string("op_9082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9082_end_0 = const()[name = string("op_9082_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9082_end_mask_0 = const()[name = string("op_9082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9082_cast_fp16 = slice_by_index(begin = var_9082_begin_0, end = var_9082_end_0, end_mask = var_9082_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9082_cast_fp16")]; + tensor var_9089_begin_0 = const()[name = string("op_9089_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9089_end_0 = const()[name = string("op_9089_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9089_end_mask_0 = const()[name = string("op_9089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9089_cast_fp16 = slice_by_index(begin = var_9089_begin_0, end = var_9089_end_0, end_mask = var_9089_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9089_cast_fp16")]; + tensor var_9096_begin_0 = const()[name = string("op_9096_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9096_end_0 = const()[name = string("op_9096_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9096_end_mask_0 = const()[name = string("op_9096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9096_cast_fp16 = slice_by_index(begin = var_9096_begin_0, end = var_9096_end_0, end_mask = var_9096_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9096_cast_fp16")]; + tensor var_9103_begin_0 = const()[name = string("op_9103_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9103_end_0 = const()[name = string("op_9103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9103_end_mask_0 = const()[name = string("op_9103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9103_cast_fp16 = slice_by_index(begin = var_9103_begin_0, end = var_9103_end_0, end_mask = var_9103_end_mask_0, x = var_8885_cast_fp16)[name = string("op_9103_cast_fp16")]; + tensor var_9110_begin_0 = const()[name = string("op_9110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9110_end_0 = const()[name = string("op_9110_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9110_end_mask_0 = const()[name = string("op_9110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9110_cast_fp16 = slice_by_index(begin = var_9110_begin_0, end = var_9110_end_0, end_mask = var_9110_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9110_cast_fp16")]; + tensor var_9117_begin_0 = const()[name = string("op_9117_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9117_end_0 = const()[name = string("op_9117_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9117_end_mask_0 = const()[name = string("op_9117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9117_cast_fp16 = slice_by_index(begin = var_9117_begin_0, end = var_9117_end_0, end_mask = var_9117_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9117_cast_fp16")]; + tensor var_9124_begin_0 = const()[name = string("op_9124_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9124_end_0 = const()[name = string("op_9124_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9124_end_mask_0 = const()[name = string("op_9124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9124_cast_fp16 = slice_by_index(begin = var_9124_begin_0, end = var_9124_end_0, end_mask = var_9124_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9124_cast_fp16")]; + tensor var_9131_begin_0 = const()[name = string("op_9131_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9131_end_0 = const()[name = string("op_9131_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9131_end_mask_0 = const()[name = string("op_9131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9131_cast_fp16 = slice_by_index(begin = var_9131_begin_0, end = var_9131_end_0, end_mask = var_9131_end_mask_0, x = var_8889_cast_fp16)[name = string("op_9131_cast_fp16")]; + tensor var_9138_begin_0 = const()[name = string("op_9138_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9138_end_0 = const()[name = string("op_9138_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9138_end_mask_0 = const()[name = string("op_9138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9138_cast_fp16 = slice_by_index(begin = var_9138_begin_0, end = var_9138_end_0, end_mask = var_9138_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9138_cast_fp16")]; + tensor var_9145_begin_0 = const()[name = string("op_9145_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9145_end_0 = const()[name = string("op_9145_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9145_end_mask_0 = const()[name = string("op_9145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9145_cast_fp16 = slice_by_index(begin = var_9145_begin_0, end = var_9145_end_0, end_mask = var_9145_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9145_cast_fp16")]; + tensor var_9152_begin_0 = const()[name = string("op_9152_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9152_end_0 = const()[name = string("op_9152_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9152_end_mask_0 = const()[name = string("op_9152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9152_cast_fp16 = slice_by_index(begin = var_9152_begin_0, end = var_9152_end_0, end_mask = var_9152_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9152_cast_fp16")]; + tensor var_9159_begin_0 = const()[name = string("op_9159_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9159_end_0 = const()[name = string("op_9159_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9159_end_mask_0 = const()[name = string("op_9159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9159_cast_fp16 = slice_by_index(begin = var_9159_begin_0, end = var_9159_end_0, end_mask = var_9159_end_mask_0, x = var_8893_cast_fp16)[name = string("op_9159_cast_fp16")]; + tensor var_9166_begin_0 = const()[name = string("op_9166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9166_end_0 = const()[name = string("op_9166_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9166_end_mask_0 = const()[name = string("op_9166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9166_cast_fp16 = slice_by_index(begin = var_9166_begin_0, end = var_9166_end_0, end_mask = var_9166_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9166_cast_fp16")]; + tensor var_9173_begin_0 = const()[name = string("op_9173_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9173_end_0 = const()[name = string("op_9173_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9173_end_mask_0 = const()[name = string("op_9173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9173_cast_fp16 = slice_by_index(begin = var_9173_begin_0, end = var_9173_end_0, end_mask = var_9173_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9173_cast_fp16")]; + tensor var_9180_begin_0 = const()[name = string("op_9180_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9180_end_0 = const()[name = string("op_9180_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9180_end_mask_0 = const()[name = string("op_9180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9180_cast_fp16 = slice_by_index(begin = var_9180_begin_0, end = var_9180_end_0, end_mask = var_9180_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9180_cast_fp16")]; + tensor var_9187_begin_0 = const()[name = string("op_9187_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9187_end_0 = const()[name = string("op_9187_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9187_end_mask_0 = const()[name = string("op_9187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9187_cast_fp16 = slice_by_index(begin = var_9187_begin_0, end = var_9187_end_0, end_mask = var_9187_end_mask_0, x = var_8897_cast_fp16)[name = string("op_9187_cast_fp16")]; + tensor var_9194_begin_0 = const()[name = string("op_9194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9194_end_0 = const()[name = string("op_9194_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9194_end_mask_0 = const()[name = string("op_9194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9194_cast_fp16 = slice_by_index(begin = var_9194_begin_0, end = var_9194_end_0, end_mask = var_9194_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9194_cast_fp16")]; + tensor var_9201_begin_0 = const()[name = string("op_9201_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9201_end_0 = const()[name = string("op_9201_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9201_end_mask_0 = const()[name = string("op_9201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9201_cast_fp16 = slice_by_index(begin = var_9201_begin_0, end = var_9201_end_0, end_mask = var_9201_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9201_cast_fp16")]; + tensor var_9208_begin_0 = const()[name = string("op_9208_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9208_end_0 = const()[name = string("op_9208_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9208_end_mask_0 = const()[name = string("op_9208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9208_cast_fp16 = slice_by_index(begin = var_9208_begin_0, end = var_9208_end_0, end_mask = var_9208_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9208_cast_fp16")]; + tensor var_9215_begin_0 = const()[name = string("op_9215_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9215_end_0 = const()[name = string("op_9215_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9215_end_mask_0 = const()[name = string("op_9215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9215_cast_fp16 = slice_by_index(begin = var_9215_begin_0, end = var_9215_end_0, end_mask = var_9215_end_mask_0, x = var_8901_cast_fp16)[name = string("op_9215_cast_fp16")]; + tensor var_9222_begin_0 = const()[name = string("op_9222_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9222_end_0 = const()[name = string("op_9222_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9222_end_mask_0 = const()[name = string("op_9222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9222_cast_fp16 = slice_by_index(begin = var_9222_begin_0, end = var_9222_end_0, end_mask = var_9222_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9222_cast_fp16")]; + tensor var_9229_begin_0 = const()[name = string("op_9229_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9229_end_0 = const()[name = string("op_9229_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9229_end_mask_0 = const()[name = string("op_9229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9229_cast_fp16 = slice_by_index(begin = var_9229_begin_0, end = var_9229_end_0, end_mask = var_9229_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9229_cast_fp16")]; + tensor var_9236_begin_0 = const()[name = string("op_9236_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9236_end_0 = const()[name = string("op_9236_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9236_end_mask_0 = const()[name = string("op_9236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9236_cast_fp16 = slice_by_index(begin = var_9236_begin_0, end = var_9236_end_0, end_mask = var_9236_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9236_cast_fp16")]; + tensor var_9243_begin_0 = const()[name = string("op_9243_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9243_end_0 = const()[name = string("op_9243_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9243_end_mask_0 = const()[name = string("op_9243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9243_cast_fp16 = slice_by_index(begin = var_9243_begin_0, end = var_9243_end_0, end_mask = var_9243_end_mask_0, x = var_8905_cast_fp16)[name = string("op_9243_cast_fp16")]; + tensor k_19_perm_0 = const()[name = string("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9248_begin_0 = const()[name = string("op_9248_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9248_end_0 = const()[name = string("op_9248_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_9248_end_mask_0 = const()[name = string("op_9248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_19_cast_fp16 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = string("transpose_2")]; + tensor var_9248_cast_fp16 = slice_by_index(begin = var_9248_begin_0, end = var_9248_end_0, end_mask = var_9248_end_mask_0, x = k_19_cast_fp16)[name = string("op_9248_cast_fp16")]; + tensor var_9252_begin_0 = const()[name = string("op_9252_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9252_end_0 = const()[name = string("op_9252_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_9252_end_mask_0 = const()[name = string("op_9252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9252_cast_fp16 = slice_by_index(begin = var_9252_begin_0, end = var_9252_end_0, end_mask = var_9252_end_mask_0, x = k_19_cast_fp16)[name = string("op_9252_cast_fp16")]; + tensor var_9256_begin_0 = const()[name = string("op_9256_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9256_end_0 = const()[name = string("op_9256_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_9256_end_mask_0 = const()[name = string("op_9256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9256_cast_fp16 = slice_by_index(begin = var_9256_begin_0, end = var_9256_end_0, end_mask = var_9256_end_mask_0, x = k_19_cast_fp16)[name = string("op_9256_cast_fp16")]; + tensor var_9260_begin_0 = const()[name = string("op_9260_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9260_end_0 = const()[name = string("op_9260_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_9260_end_mask_0 = const()[name = string("op_9260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9260_cast_fp16 = slice_by_index(begin = var_9260_begin_0, end = var_9260_end_0, end_mask = var_9260_end_mask_0, x = k_19_cast_fp16)[name = string("op_9260_cast_fp16")]; + tensor var_9264_begin_0 = const()[name = string("op_9264_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9264_end_0 = const()[name = string("op_9264_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_9264_end_mask_0 = const()[name = string("op_9264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9264_cast_fp16 = slice_by_index(begin = var_9264_begin_0, end = var_9264_end_0, end_mask = var_9264_end_mask_0, x = k_19_cast_fp16)[name = string("op_9264_cast_fp16")]; + tensor var_9268_begin_0 = const()[name = string("op_9268_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9268_end_0 = const()[name = string("op_9268_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_9268_end_mask_0 = const()[name = string("op_9268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9268_cast_fp16 = slice_by_index(begin = var_9268_begin_0, end = var_9268_end_0, end_mask = var_9268_end_mask_0, x = k_19_cast_fp16)[name = string("op_9268_cast_fp16")]; + tensor var_9272_begin_0 = const()[name = string("op_9272_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9272_end_0 = const()[name = string("op_9272_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_9272_end_mask_0 = const()[name = string("op_9272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9272_cast_fp16 = slice_by_index(begin = var_9272_begin_0, end = var_9272_end_0, end_mask = var_9272_end_mask_0, x = k_19_cast_fp16)[name = string("op_9272_cast_fp16")]; + tensor var_9276_begin_0 = const()[name = string("op_9276_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9276_end_0 = const()[name = string("op_9276_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_9276_end_mask_0 = const()[name = string("op_9276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9276_cast_fp16 = slice_by_index(begin = var_9276_begin_0, end = var_9276_end_0, end_mask = var_9276_end_mask_0, x = k_19_cast_fp16)[name = string("op_9276_cast_fp16")]; + tensor var_9280_begin_0 = const()[name = string("op_9280_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9280_end_0 = const()[name = string("op_9280_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_9280_end_mask_0 = const()[name = string("op_9280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9280_cast_fp16 = slice_by_index(begin = var_9280_begin_0, end = var_9280_end_0, end_mask = var_9280_end_mask_0, x = k_19_cast_fp16)[name = string("op_9280_cast_fp16")]; + tensor var_9284_begin_0 = const()[name = string("op_9284_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9284_end_0 = const()[name = string("op_9284_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_9284_end_mask_0 = const()[name = string("op_9284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9284_cast_fp16 = slice_by_index(begin = var_9284_begin_0, end = var_9284_end_0, end_mask = var_9284_end_mask_0, x = k_19_cast_fp16)[name = string("op_9284_cast_fp16")]; + tensor var_9288_begin_0 = const()[name = string("op_9288_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9288_end_0 = const()[name = string("op_9288_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_9288_end_mask_0 = const()[name = string("op_9288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9288_cast_fp16 = slice_by_index(begin = var_9288_begin_0, end = var_9288_end_0, end_mask = var_9288_end_mask_0, x = k_19_cast_fp16)[name = string("op_9288_cast_fp16")]; + tensor var_9292_begin_0 = const()[name = string("op_9292_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9292_end_0 = const()[name = string("op_9292_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_9292_end_mask_0 = const()[name = string("op_9292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9292_cast_fp16 = slice_by_index(begin = var_9292_begin_0, end = var_9292_end_0, end_mask = var_9292_end_mask_0, x = k_19_cast_fp16)[name = string("op_9292_cast_fp16")]; + tensor var_9294_begin_0 = const()[name = string("op_9294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9294_end_0 = const()[name = string("op_9294_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9294_end_mask_0 = const()[name = string("op_9294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9294_cast_fp16 = slice_by_index(begin = var_9294_begin_0, end = var_9294_end_0, end_mask = var_9294_end_mask_0, x = value_19_cast_fp16)[name = string("op_9294_cast_fp16")]; + tensor var_9298_begin_0 = const()[name = string("op_9298_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9298_end_0 = const()[name = string("op_9298_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9298_end_mask_0 = const()[name = string("op_9298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9298_cast_fp16 = slice_by_index(begin = var_9298_begin_0, end = var_9298_end_0, end_mask = var_9298_end_mask_0, x = value_19_cast_fp16)[name = string("op_9298_cast_fp16")]; + tensor var_9302_begin_0 = const()[name = string("op_9302_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9302_end_0 = const()[name = string("op_9302_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9302_end_mask_0 = const()[name = string("op_9302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9302_cast_fp16 = slice_by_index(begin = var_9302_begin_0, end = var_9302_end_0, end_mask = var_9302_end_mask_0, x = value_19_cast_fp16)[name = string("op_9302_cast_fp16")]; + tensor var_9306_begin_0 = const()[name = string("op_9306_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9306_end_0 = const()[name = string("op_9306_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9306_end_mask_0 = const()[name = string("op_9306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9306_cast_fp16 = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = value_19_cast_fp16)[name = string("op_9306_cast_fp16")]; + tensor var_9310_begin_0 = const()[name = string("op_9310_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9310_end_0 = const()[name = string("op_9310_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9310_end_mask_0 = const()[name = string("op_9310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9310_cast_fp16 = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = value_19_cast_fp16)[name = string("op_9310_cast_fp16")]; + tensor var_9314_begin_0 = const()[name = string("op_9314_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9314_end_0 = const()[name = string("op_9314_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9314_end_mask_0 = const()[name = string("op_9314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9314_cast_fp16 = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = value_19_cast_fp16)[name = string("op_9314_cast_fp16")]; + tensor var_9318_begin_0 = const()[name = string("op_9318_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9318_end_0 = const()[name = string("op_9318_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9318_end_mask_0 = const()[name = string("op_9318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9318_cast_fp16 = slice_by_index(begin = var_9318_begin_0, end = var_9318_end_0, end_mask = var_9318_end_mask_0, x = value_19_cast_fp16)[name = string("op_9318_cast_fp16")]; + tensor var_9322_begin_0 = const()[name = string("op_9322_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9322_end_0 = const()[name = string("op_9322_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9322_end_mask_0 = const()[name = string("op_9322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9322_cast_fp16 = slice_by_index(begin = var_9322_begin_0, end = var_9322_end_0, end_mask = var_9322_end_mask_0, x = value_19_cast_fp16)[name = string("op_9322_cast_fp16")]; + tensor var_9326_begin_0 = const()[name = string("op_9326_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9326_end_0 = const()[name = string("op_9326_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9326_end_mask_0 = const()[name = string("op_9326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9326_cast_fp16 = slice_by_index(begin = var_9326_begin_0, end = var_9326_end_0, end_mask = var_9326_end_mask_0, x = value_19_cast_fp16)[name = string("op_9326_cast_fp16")]; + tensor var_9330_begin_0 = const()[name = string("op_9330_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9330_end_0 = const()[name = string("op_9330_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9330_end_mask_0 = const()[name = string("op_9330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9330_cast_fp16 = slice_by_index(begin = var_9330_begin_0, end = var_9330_end_0, end_mask = var_9330_end_mask_0, x = value_19_cast_fp16)[name = string("op_9330_cast_fp16")]; + tensor var_9334_begin_0 = const()[name = string("op_9334_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9334_end_0 = const()[name = string("op_9334_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9334_end_mask_0 = const()[name = string("op_9334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9334_cast_fp16 = slice_by_index(begin = var_9334_begin_0, end = var_9334_end_0, end_mask = var_9334_end_mask_0, x = value_19_cast_fp16)[name = string("op_9334_cast_fp16")]; + tensor var_9338_begin_0 = const()[name = string("op_9338_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9338_end_0 = const()[name = string("op_9338_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9338_end_mask_0 = const()[name = string("op_9338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9338_cast_fp16 = slice_by_index(begin = var_9338_begin_0, end = var_9338_end_0, end_mask = var_9338_end_mask_0, x = value_19_cast_fp16)[name = string("op_9338_cast_fp16")]; + string _SplitHeadsQ__mh_w_865_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_865_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_865_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_865_equation_0, values = (var_9248_cast_fp16, var_8914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_865_cast_fp16")]; + string _SplitHeadsQ__mh_w_867_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_867_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_867_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_867_equation_0, values = (var_9248_cast_fp16, var_8921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_867_cast_fp16")]; + string _SplitHeadsQ__mh_w_869_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_869_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_869_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_869_equation_0, values = (var_9248_cast_fp16, var_8928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_869_cast_fp16")]; + string _SplitHeadsQ__mh_w_871_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_871_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_871_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_871_equation_0, values = (var_9248_cast_fp16, var_8935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_871_cast_fp16")]; + string _SplitHeadsQ__mh_w_873_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_873_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_873_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_873_equation_0, values = (var_9252_cast_fp16, var_8942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_873_cast_fp16")]; + string _SplitHeadsQ__mh_w_875_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_875_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_875_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_875_equation_0, values = (var_9252_cast_fp16, var_8949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_875_cast_fp16")]; + string _SplitHeadsQ__mh_w_877_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_877_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_877_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_877_equation_0, values = (var_9252_cast_fp16, var_8956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_877_cast_fp16")]; + string _SplitHeadsQ__mh_w_879_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_879_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_879_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_879_equation_0, values = (var_9252_cast_fp16, var_8963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_879_cast_fp16")]; + string _SplitHeadsQ__mh_w_881_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_881_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_881_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_881_equation_0, values = (var_9256_cast_fp16, var_8970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_881_cast_fp16")]; + string _SplitHeadsQ__mh_w_883_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_883_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_883_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_883_equation_0, values = (var_9256_cast_fp16, var_8977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_883_cast_fp16")]; + string _SplitHeadsQ__mh_w_885_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_885_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_885_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_885_equation_0, values = (var_9256_cast_fp16, var_8984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_885_cast_fp16")]; + string _SplitHeadsQ__mh_w_887_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_887_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_887_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_887_equation_0, values = (var_9256_cast_fp16, var_8991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_887_cast_fp16")]; + string _SplitHeadsQ__mh_w_889_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_889_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_889_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_889_equation_0, values = (var_9260_cast_fp16, var_8998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_889_cast_fp16")]; + string _SplitHeadsQ__mh_w_891_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_891_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_891_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_891_equation_0, values = (var_9260_cast_fp16, var_9005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_891_cast_fp16")]; + string _SplitHeadsQ__mh_w_893_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_893_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_893_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_893_equation_0, values = (var_9260_cast_fp16, var_9012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_893_cast_fp16")]; + string _SplitHeadsQ__mh_w_895_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_895_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_895_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_895_equation_0, values = (var_9260_cast_fp16, var_9019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_895_cast_fp16")]; + string _SplitHeadsQ__mh_w_897_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_897_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_897_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_897_equation_0, values = (var_9264_cast_fp16, var_9026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_897_cast_fp16")]; + string _SplitHeadsQ__mh_w_899_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_899_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_899_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_899_equation_0, values = (var_9264_cast_fp16, var_9033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_899_cast_fp16")]; + string _SplitHeadsQ__mh_w_901_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_901_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_901_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_901_equation_0, values = (var_9264_cast_fp16, var_9040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_901_cast_fp16")]; + string _SplitHeadsQ__mh_w_903_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_903_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_903_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_903_equation_0, values = (var_9264_cast_fp16, var_9047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_903_cast_fp16")]; + string _SplitHeadsQ__mh_w_905_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_905_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_905_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_905_equation_0, values = (var_9268_cast_fp16, var_9054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_905_cast_fp16")]; + string _SplitHeadsQ__mh_w_907_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_907_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_907_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_907_equation_0, values = (var_9268_cast_fp16, var_9061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_907_cast_fp16")]; + string _SplitHeadsQ__mh_w_909_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_909_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_909_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_909_equation_0, values = (var_9268_cast_fp16, var_9068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_909_cast_fp16")]; + string _SplitHeadsQ__mh_w_911_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_911_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_911_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_911_equation_0, values = (var_9268_cast_fp16, var_9075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_911_cast_fp16")]; + string _SplitHeadsQ__mh_w_913_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_913_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_913_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_913_equation_0, values = (var_9272_cast_fp16, var_9082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_913_cast_fp16")]; + string _SplitHeadsQ__mh_w_915_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_915_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_915_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_915_equation_0, values = (var_9272_cast_fp16, var_9089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_915_cast_fp16")]; + string _SplitHeadsQ__mh_w_917_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_917_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_917_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_917_equation_0, values = (var_9272_cast_fp16, var_9096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_917_cast_fp16")]; + string _SplitHeadsQ__mh_w_919_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_919_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_919_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_919_equation_0, values = (var_9272_cast_fp16, var_9103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_919_cast_fp16")]; + string _SplitHeadsQ__mh_w_921_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_921_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_921_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_921_equation_0, values = (var_9276_cast_fp16, var_9110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_921_cast_fp16")]; + string _SplitHeadsQ__mh_w_923_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_923_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_923_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_923_equation_0, values = (var_9276_cast_fp16, var_9117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_923_cast_fp16")]; + string _SplitHeadsQ__mh_w_925_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_925_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_925_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_925_equation_0, values = (var_9276_cast_fp16, var_9124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_925_cast_fp16")]; + string _SplitHeadsQ__mh_w_927_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_927_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_927_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_927_equation_0, values = (var_9276_cast_fp16, var_9131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_927_cast_fp16")]; + string _SplitHeadsQ__mh_w_929_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_929_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_929_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_929_equation_0, values = (var_9280_cast_fp16, var_9138_cast_fp16))[name = string("_SplitHeadsQ__mh_w_929_cast_fp16")]; + string _SplitHeadsQ__mh_w_931_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_931_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_931_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_931_equation_0, values = (var_9280_cast_fp16, var_9145_cast_fp16))[name = string("_SplitHeadsQ__mh_w_931_cast_fp16")]; + string _SplitHeadsQ__mh_w_933_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_933_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_933_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_933_equation_0, values = (var_9280_cast_fp16, var_9152_cast_fp16))[name = string("_SplitHeadsQ__mh_w_933_cast_fp16")]; + string _SplitHeadsQ__mh_w_935_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_935_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_935_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_935_equation_0, values = (var_9280_cast_fp16, var_9159_cast_fp16))[name = string("_SplitHeadsQ__mh_w_935_cast_fp16")]; + string _SplitHeadsQ__mh_w_937_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_937_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_937_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_937_equation_0, values = (var_9284_cast_fp16, var_9166_cast_fp16))[name = string("_SplitHeadsQ__mh_w_937_cast_fp16")]; + string _SplitHeadsQ__mh_w_939_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_939_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_939_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_939_equation_0, values = (var_9284_cast_fp16, var_9173_cast_fp16))[name = string("_SplitHeadsQ__mh_w_939_cast_fp16")]; + string _SplitHeadsQ__mh_w_941_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_941_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_941_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_941_equation_0, values = (var_9284_cast_fp16, var_9180_cast_fp16))[name = string("_SplitHeadsQ__mh_w_941_cast_fp16")]; + string _SplitHeadsQ__mh_w_943_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_943_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_943_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_943_equation_0, values = (var_9284_cast_fp16, var_9187_cast_fp16))[name = string("_SplitHeadsQ__mh_w_943_cast_fp16")]; + string _SplitHeadsQ__mh_w_945_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_945_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_945_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_945_equation_0, values = (var_9288_cast_fp16, var_9194_cast_fp16))[name = string("_SplitHeadsQ__mh_w_945_cast_fp16")]; + string _SplitHeadsQ__mh_w_947_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_947_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_947_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_947_equation_0, values = (var_9288_cast_fp16, var_9201_cast_fp16))[name = string("_SplitHeadsQ__mh_w_947_cast_fp16")]; + string _SplitHeadsQ__mh_w_949_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_949_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_949_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_949_equation_0, values = (var_9288_cast_fp16, var_9208_cast_fp16))[name = string("_SplitHeadsQ__mh_w_949_cast_fp16")]; + string _SplitHeadsQ__mh_w_951_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_951_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_951_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_951_equation_0, values = (var_9288_cast_fp16, var_9215_cast_fp16))[name = string("_SplitHeadsQ__mh_w_951_cast_fp16")]; + string _SplitHeadsQ__mh_w_953_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_953_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_953_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_953_equation_0, values = (var_9292_cast_fp16, var_9222_cast_fp16))[name = string("_SplitHeadsQ__mh_w_953_cast_fp16")]; + string _SplitHeadsQ__mh_w_955_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_955_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_955_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_955_equation_0, values = (var_9292_cast_fp16, var_9229_cast_fp16))[name = string("_SplitHeadsQ__mh_w_955_cast_fp16")]; + string _SplitHeadsQ__mh_w_957_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_957_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_957_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_957_equation_0, values = (var_9292_cast_fp16, var_9236_cast_fp16))[name = string("_SplitHeadsQ__mh_w_957_cast_fp16")]; + string _SplitHeadsQ__mh_w_959_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_959_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_959_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_959_equation_0, values = (var_9292_cast_fp16, var_9243_cast_fp16))[name = string("_SplitHeadsQ__mh_w_959_cast_fp16")]; + fp16 var_9437_to_fp16 = const()[name = string("op_9437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_865_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_865_cast_fp16, y = var_9437_to_fp16)[name = string("aw_chunk_865_cast_fp16")]; + fp16 var_9439_to_fp16 = const()[name = string("op_9439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_867_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_867_cast_fp16, y = var_9439_to_fp16)[name = string("aw_chunk_867_cast_fp16")]; + fp16 var_9441_to_fp16 = const()[name = string("op_9441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_869_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_869_cast_fp16, y = var_9441_to_fp16)[name = string("aw_chunk_869_cast_fp16")]; + fp16 var_9443_to_fp16 = const()[name = string("op_9443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_871_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_871_cast_fp16, y = var_9443_to_fp16)[name = string("aw_chunk_871_cast_fp16")]; + fp16 var_9445_to_fp16 = const()[name = string("op_9445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_873_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_873_cast_fp16, y = var_9445_to_fp16)[name = string("aw_chunk_873_cast_fp16")]; + fp16 var_9447_to_fp16 = const()[name = string("op_9447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_875_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_875_cast_fp16, y = var_9447_to_fp16)[name = string("aw_chunk_875_cast_fp16")]; + fp16 var_9449_to_fp16 = const()[name = string("op_9449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_877_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_877_cast_fp16, y = var_9449_to_fp16)[name = string("aw_chunk_877_cast_fp16")]; + fp16 var_9451_to_fp16 = const()[name = string("op_9451_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_879_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_879_cast_fp16, y = var_9451_to_fp16)[name = string("aw_chunk_879_cast_fp16")]; + fp16 var_9453_to_fp16 = const()[name = string("op_9453_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_881_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_881_cast_fp16, y = var_9453_to_fp16)[name = string("aw_chunk_881_cast_fp16")]; + fp16 var_9455_to_fp16 = const()[name = string("op_9455_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_883_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_883_cast_fp16, y = var_9455_to_fp16)[name = string("aw_chunk_883_cast_fp16")]; + fp16 var_9457_to_fp16 = const()[name = string("op_9457_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_885_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_885_cast_fp16, y = var_9457_to_fp16)[name = string("aw_chunk_885_cast_fp16")]; + fp16 var_9459_to_fp16 = const()[name = string("op_9459_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_887_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_887_cast_fp16, y = var_9459_to_fp16)[name = string("aw_chunk_887_cast_fp16")]; + fp16 var_9461_to_fp16 = const()[name = string("op_9461_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_889_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_889_cast_fp16, y = var_9461_to_fp16)[name = string("aw_chunk_889_cast_fp16")]; + fp16 var_9463_to_fp16 = const()[name = string("op_9463_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_891_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_891_cast_fp16, y = var_9463_to_fp16)[name = string("aw_chunk_891_cast_fp16")]; + fp16 var_9465_to_fp16 = const()[name = string("op_9465_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_893_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_893_cast_fp16, y = var_9465_to_fp16)[name = string("aw_chunk_893_cast_fp16")]; + fp16 var_9467_to_fp16 = const()[name = string("op_9467_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_895_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_895_cast_fp16, y = var_9467_to_fp16)[name = string("aw_chunk_895_cast_fp16")]; + fp16 var_9469_to_fp16 = const()[name = string("op_9469_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_897_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_897_cast_fp16, y = var_9469_to_fp16)[name = string("aw_chunk_897_cast_fp16")]; + fp16 var_9471_to_fp16 = const()[name = string("op_9471_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_899_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_899_cast_fp16, y = var_9471_to_fp16)[name = string("aw_chunk_899_cast_fp16")]; + fp16 var_9473_to_fp16 = const()[name = string("op_9473_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_901_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_901_cast_fp16, y = var_9473_to_fp16)[name = string("aw_chunk_901_cast_fp16")]; + fp16 var_9475_to_fp16 = const()[name = string("op_9475_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_903_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_903_cast_fp16, y = var_9475_to_fp16)[name = string("aw_chunk_903_cast_fp16")]; + fp16 var_9477_to_fp16 = const()[name = string("op_9477_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_905_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_905_cast_fp16, y = var_9477_to_fp16)[name = string("aw_chunk_905_cast_fp16")]; + fp16 var_9479_to_fp16 = const()[name = string("op_9479_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_907_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_907_cast_fp16, y = var_9479_to_fp16)[name = string("aw_chunk_907_cast_fp16")]; + fp16 var_9481_to_fp16 = const()[name = string("op_9481_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_909_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_909_cast_fp16, y = var_9481_to_fp16)[name = string("aw_chunk_909_cast_fp16")]; + fp16 var_9483_to_fp16 = const()[name = string("op_9483_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_911_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_911_cast_fp16, y = var_9483_to_fp16)[name = string("aw_chunk_911_cast_fp16")]; + fp16 var_9485_to_fp16 = const()[name = string("op_9485_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_913_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_913_cast_fp16, y = var_9485_to_fp16)[name = string("aw_chunk_913_cast_fp16")]; + fp16 var_9487_to_fp16 = const()[name = string("op_9487_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_915_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_915_cast_fp16, y = var_9487_to_fp16)[name = string("aw_chunk_915_cast_fp16")]; + fp16 var_9489_to_fp16 = const()[name = string("op_9489_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_917_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_917_cast_fp16, y = var_9489_to_fp16)[name = string("aw_chunk_917_cast_fp16")]; + fp16 var_9491_to_fp16 = const()[name = string("op_9491_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_919_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_919_cast_fp16, y = var_9491_to_fp16)[name = string("aw_chunk_919_cast_fp16")]; + fp16 var_9493_to_fp16 = const()[name = string("op_9493_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_921_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_921_cast_fp16, y = var_9493_to_fp16)[name = string("aw_chunk_921_cast_fp16")]; + fp16 var_9495_to_fp16 = const()[name = string("op_9495_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_923_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_923_cast_fp16, y = var_9495_to_fp16)[name = string("aw_chunk_923_cast_fp16")]; + fp16 var_9497_to_fp16 = const()[name = string("op_9497_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_925_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_925_cast_fp16, y = var_9497_to_fp16)[name = string("aw_chunk_925_cast_fp16")]; + fp16 var_9499_to_fp16 = const()[name = string("op_9499_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_927_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_927_cast_fp16, y = var_9499_to_fp16)[name = string("aw_chunk_927_cast_fp16")]; + fp16 var_9501_to_fp16 = const()[name = string("op_9501_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_929_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_929_cast_fp16, y = var_9501_to_fp16)[name = string("aw_chunk_929_cast_fp16")]; + fp16 var_9503_to_fp16 = const()[name = string("op_9503_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_931_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_931_cast_fp16, y = var_9503_to_fp16)[name = string("aw_chunk_931_cast_fp16")]; + fp16 var_9505_to_fp16 = const()[name = string("op_9505_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_933_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_933_cast_fp16, y = var_9505_to_fp16)[name = string("aw_chunk_933_cast_fp16")]; + fp16 var_9507_to_fp16 = const()[name = string("op_9507_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_935_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_935_cast_fp16, y = var_9507_to_fp16)[name = string("aw_chunk_935_cast_fp16")]; + fp16 var_9509_to_fp16 = const()[name = string("op_9509_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_937_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_937_cast_fp16, y = var_9509_to_fp16)[name = string("aw_chunk_937_cast_fp16")]; + fp16 var_9511_to_fp16 = const()[name = string("op_9511_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_939_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_939_cast_fp16, y = var_9511_to_fp16)[name = string("aw_chunk_939_cast_fp16")]; + fp16 var_9513_to_fp16 = const()[name = string("op_9513_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_941_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_941_cast_fp16, y = var_9513_to_fp16)[name = string("aw_chunk_941_cast_fp16")]; + fp16 var_9515_to_fp16 = const()[name = string("op_9515_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_943_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_943_cast_fp16, y = var_9515_to_fp16)[name = string("aw_chunk_943_cast_fp16")]; + fp16 var_9517_to_fp16 = const()[name = string("op_9517_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_945_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_945_cast_fp16, y = var_9517_to_fp16)[name = string("aw_chunk_945_cast_fp16")]; + fp16 var_9519_to_fp16 = const()[name = string("op_9519_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_947_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_947_cast_fp16, y = var_9519_to_fp16)[name = string("aw_chunk_947_cast_fp16")]; + fp16 var_9521_to_fp16 = const()[name = string("op_9521_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_949_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_949_cast_fp16, y = var_9521_to_fp16)[name = string("aw_chunk_949_cast_fp16")]; + fp16 var_9523_to_fp16 = const()[name = string("op_9523_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_951_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_951_cast_fp16, y = var_9523_to_fp16)[name = string("aw_chunk_951_cast_fp16")]; + fp16 var_9525_to_fp16 = const()[name = string("op_9525_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_953_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_953_cast_fp16, y = var_9525_to_fp16)[name = string("aw_chunk_953_cast_fp16")]; + fp16 var_9527_to_fp16 = const()[name = string("op_9527_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_955_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_955_cast_fp16, y = var_9527_to_fp16)[name = string("aw_chunk_955_cast_fp16")]; + fp16 var_9529_to_fp16 = const()[name = string("op_9529_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_957_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_957_cast_fp16, y = var_9529_to_fp16)[name = string("aw_chunk_957_cast_fp16")]; + fp16 var_9531_to_fp16 = const()[name = string("op_9531_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_959_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_959_cast_fp16, y = var_9531_to_fp16)[name = string("aw_chunk_959_cast_fp16")]; + tensor var_9533_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_865_cast_fp16)[name = string("op_9533_cast_fp16")]; + tensor var_9534_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_867_cast_fp16)[name = string("op_9534_cast_fp16")]; + tensor var_9535_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_869_cast_fp16)[name = string("op_9535_cast_fp16")]; + tensor var_9536_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_871_cast_fp16)[name = string("op_9536_cast_fp16")]; + tensor var_9537_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_873_cast_fp16)[name = string("op_9537_cast_fp16")]; + tensor var_9538_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_875_cast_fp16)[name = string("op_9538_cast_fp16")]; + tensor var_9539_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_877_cast_fp16)[name = string("op_9539_cast_fp16")]; + tensor var_9540_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_879_cast_fp16)[name = string("op_9540_cast_fp16")]; + tensor var_9541_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_881_cast_fp16)[name = string("op_9541_cast_fp16")]; + tensor var_9542_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_883_cast_fp16)[name = string("op_9542_cast_fp16")]; + tensor var_9543_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_885_cast_fp16)[name = string("op_9543_cast_fp16")]; + tensor var_9544_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_887_cast_fp16)[name = string("op_9544_cast_fp16")]; + tensor var_9545_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_889_cast_fp16)[name = string("op_9545_cast_fp16")]; + tensor var_9546_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_891_cast_fp16)[name = string("op_9546_cast_fp16")]; + tensor var_9547_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_893_cast_fp16)[name = string("op_9547_cast_fp16")]; + tensor var_9548_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_895_cast_fp16)[name = string("op_9548_cast_fp16")]; + tensor var_9549_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_897_cast_fp16)[name = string("op_9549_cast_fp16")]; + tensor var_9550_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_899_cast_fp16)[name = string("op_9550_cast_fp16")]; + tensor var_9551_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_901_cast_fp16)[name = string("op_9551_cast_fp16")]; + tensor var_9552_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_903_cast_fp16)[name = string("op_9552_cast_fp16")]; + tensor var_9553_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_905_cast_fp16)[name = string("op_9553_cast_fp16")]; + tensor var_9554_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_907_cast_fp16)[name = string("op_9554_cast_fp16")]; + tensor var_9555_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_909_cast_fp16)[name = string("op_9555_cast_fp16")]; + tensor var_9556_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_911_cast_fp16)[name = string("op_9556_cast_fp16")]; + tensor var_9557_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_913_cast_fp16)[name = string("op_9557_cast_fp16")]; + tensor var_9558_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_915_cast_fp16)[name = string("op_9558_cast_fp16")]; + tensor var_9559_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_917_cast_fp16)[name = string("op_9559_cast_fp16")]; + tensor var_9560_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_919_cast_fp16)[name = string("op_9560_cast_fp16")]; + tensor var_9561_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_921_cast_fp16)[name = string("op_9561_cast_fp16")]; + tensor var_9562_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_923_cast_fp16)[name = string("op_9562_cast_fp16")]; + tensor var_9563_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_925_cast_fp16)[name = string("op_9563_cast_fp16")]; + tensor var_9564_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_927_cast_fp16)[name = string("op_9564_cast_fp16")]; + tensor var_9565_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_929_cast_fp16)[name = string("op_9565_cast_fp16")]; + tensor var_9566_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_931_cast_fp16)[name = string("op_9566_cast_fp16")]; + tensor var_9567_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_933_cast_fp16)[name = string("op_9567_cast_fp16")]; + tensor var_9568_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_935_cast_fp16)[name = string("op_9568_cast_fp16")]; + tensor var_9569_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_937_cast_fp16)[name = string("op_9569_cast_fp16")]; + tensor var_9570_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_939_cast_fp16)[name = string("op_9570_cast_fp16")]; + tensor var_9571_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_941_cast_fp16)[name = string("op_9571_cast_fp16")]; + tensor var_9572_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_943_cast_fp16)[name = string("op_9572_cast_fp16")]; + tensor var_9573_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_945_cast_fp16)[name = string("op_9573_cast_fp16")]; + tensor var_9574_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_947_cast_fp16)[name = string("op_9574_cast_fp16")]; + tensor var_9575_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_949_cast_fp16)[name = string("op_9575_cast_fp16")]; + tensor var_9576_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_951_cast_fp16)[name = string("op_9576_cast_fp16")]; + tensor var_9577_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_953_cast_fp16)[name = string("op_9577_cast_fp16")]; + tensor var_9578_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_955_cast_fp16)[name = string("op_9578_cast_fp16")]; + tensor var_9579_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_957_cast_fp16)[name = string("op_9579_cast_fp16")]; + tensor var_9580_cast_fp16 = softmax(axis = var_8806, x = aw_chunk_959_cast_fp16)[name = string("op_9580_cast_fp16")]; + string var_9582_equation_0 = const()[name = string("op_9582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9582_cast_fp16 = einsum(equation = var_9582_equation_0, values = (var_9294_cast_fp16, var_9533_cast_fp16))[name = string("op_9582_cast_fp16")]; + string var_9584_equation_0 = const()[name = string("op_9584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9584_cast_fp16 = einsum(equation = var_9584_equation_0, values = (var_9294_cast_fp16, var_9534_cast_fp16))[name = string("op_9584_cast_fp16")]; + string var_9586_equation_0 = const()[name = string("op_9586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9586_cast_fp16 = einsum(equation = var_9586_equation_0, values = (var_9294_cast_fp16, var_9535_cast_fp16))[name = string("op_9586_cast_fp16")]; + string var_9588_equation_0 = const()[name = string("op_9588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9588_cast_fp16 = einsum(equation = var_9588_equation_0, values = (var_9294_cast_fp16, var_9536_cast_fp16))[name = string("op_9588_cast_fp16")]; + string var_9590_equation_0 = const()[name = string("op_9590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9590_cast_fp16 = einsum(equation = var_9590_equation_0, values = (var_9298_cast_fp16, var_9537_cast_fp16))[name = string("op_9590_cast_fp16")]; + string var_9592_equation_0 = const()[name = string("op_9592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9592_cast_fp16 = einsum(equation = var_9592_equation_0, values = (var_9298_cast_fp16, var_9538_cast_fp16))[name = string("op_9592_cast_fp16")]; + string var_9594_equation_0 = const()[name = string("op_9594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9594_cast_fp16 = einsum(equation = var_9594_equation_0, values = (var_9298_cast_fp16, var_9539_cast_fp16))[name = string("op_9594_cast_fp16")]; + string var_9596_equation_0 = const()[name = string("op_9596_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9596_cast_fp16 = einsum(equation = var_9596_equation_0, values = (var_9298_cast_fp16, var_9540_cast_fp16))[name = string("op_9596_cast_fp16")]; + string var_9598_equation_0 = const()[name = string("op_9598_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9598_cast_fp16 = einsum(equation = var_9598_equation_0, values = (var_9302_cast_fp16, var_9541_cast_fp16))[name = string("op_9598_cast_fp16")]; + string var_9600_equation_0 = const()[name = string("op_9600_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9600_cast_fp16 = einsum(equation = var_9600_equation_0, values = (var_9302_cast_fp16, var_9542_cast_fp16))[name = string("op_9600_cast_fp16")]; + string var_9602_equation_0 = const()[name = string("op_9602_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9602_cast_fp16 = einsum(equation = var_9602_equation_0, values = (var_9302_cast_fp16, var_9543_cast_fp16))[name = string("op_9602_cast_fp16")]; + string var_9604_equation_0 = const()[name = string("op_9604_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9604_cast_fp16 = einsum(equation = var_9604_equation_0, values = (var_9302_cast_fp16, var_9544_cast_fp16))[name = string("op_9604_cast_fp16")]; + string var_9606_equation_0 = const()[name = string("op_9606_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9606_cast_fp16 = einsum(equation = var_9606_equation_0, values = (var_9306_cast_fp16, var_9545_cast_fp16))[name = string("op_9606_cast_fp16")]; + string var_9608_equation_0 = const()[name = string("op_9608_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9608_cast_fp16 = einsum(equation = var_9608_equation_0, values = (var_9306_cast_fp16, var_9546_cast_fp16))[name = string("op_9608_cast_fp16")]; + string var_9610_equation_0 = const()[name = string("op_9610_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9610_cast_fp16 = einsum(equation = var_9610_equation_0, values = (var_9306_cast_fp16, var_9547_cast_fp16))[name = string("op_9610_cast_fp16")]; + string var_9612_equation_0 = const()[name = string("op_9612_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9612_cast_fp16 = einsum(equation = var_9612_equation_0, values = (var_9306_cast_fp16, var_9548_cast_fp16))[name = string("op_9612_cast_fp16")]; + string var_9614_equation_0 = const()[name = string("op_9614_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9614_cast_fp16 = einsum(equation = var_9614_equation_0, values = (var_9310_cast_fp16, var_9549_cast_fp16))[name = string("op_9614_cast_fp16")]; + string var_9616_equation_0 = const()[name = string("op_9616_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9616_cast_fp16 = einsum(equation = var_9616_equation_0, values = (var_9310_cast_fp16, var_9550_cast_fp16))[name = string("op_9616_cast_fp16")]; + string var_9618_equation_0 = const()[name = string("op_9618_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9618_cast_fp16 = einsum(equation = var_9618_equation_0, values = (var_9310_cast_fp16, var_9551_cast_fp16))[name = string("op_9618_cast_fp16")]; + string var_9620_equation_0 = const()[name = string("op_9620_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9620_cast_fp16 = einsum(equation = var_9620_equation_0, values = (var_9310_cast_fp16, var_9552_cast_fp16))[name = string("op_9620_cast_fp16")]; + string var_9622_equation_0 = const()[name = string("op_9622_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9622_cast_fp16 = einsum(equation = var_9622_equation_0, values = (var_9314_cast_fp16, var_9553_cast_fp16))[name = string("op_9622_cast_fp16")]; + string var_9624_equation_0 = const()[name = string("op_9624_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9624_cast_fp16 = einsum(equation = var_9624_equation_0, values = (var_9314_cast_fp16, var_9554_cast_fp16))[name = string("op_9624_cast_fp16")]; + string var_9626_equation_0 = const()[name = string("op_9626_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9626_cast_fp16 = einsum(equation = var_9626_equation_0, values = (var_9314_cast_fp16, var_9555_cast_fp16))[name = string("op_9626_cast_fp16")]; + string var_9628_equation_0 = const()[name = string("op_9628_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9628_cast_fp16 = einsum(equation = var_9628_equation_0, values = (var_9314_cast_fp16, var_9556_cast_fp16))[name = string("op_9628_cast_fp16")]; + string var_9630_equation_0 = const()[name = string("op_9630_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9630_cast_fp16 = einsum(equation = var_9630_equation_0, values = (var_9318_cast_fp16, var_9557_cast_fp16))[name = string("op_9630_cast_fp16")]; + string var_9632_equation_0 = const()[name = string("op_9632_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9632_cast_fp16 = einsum(equation = var_9632_equation_0, values = (var_9318_cast_fp16, var_9558_cast_fp16))[name = string("op_9632_cast_fp16")]; + string var_9634_equation_0 = const()[name = string("op_9634_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9634_cast_fp16 = einsum(equation = var_9634_equation_0, values = (var_9318_cast_fp16, var_9559_cast_fp16))[name = string("op_9634_cast_fp16")]; + string var_9636_equation_0 = const()[name = string("op_9636_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9636_cast_fp16 = einsum(equation = var_9636_equation_0, values = (var_9318_cast_fp16, var_9560_cast_fp16))[name = string("op_9636_cast_fp16")]; + string var_9638_equation_0 = const()[name = string("op_9638_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9638_cast_fp16 = einsum(equation = var_9638_equation_0, values = (var_9322_cast_fp16, var_9561_cast_fp16))[name = string("op_9638_cast_fp16")]; + string var_9640_equation_0 = const()[name = string("op_9640_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9640_cast_fp16 = einsum(equation = var_9640_equation_0, values = (var_9322_cast_fp16, var_9562_cast_fp16))[name = string("op_9640_cast_fp16")]; + string var_9642_equation_0 = const()[name = string("op_9642_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9642_cast_fp16 = einsum(equation = var_9642_equation_0, values = (var_9322_cast_fp16, var_9563_cast_fp16))[name = string("op_9642_cast_fp16")]; + string var_9644_equation_0 = const()[name = string("op_9644_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9644_cast_fp16 = einsum(equation = var_9644_equation_0, values = (var_9322_cast_fp16, var_9564_cast_fp16))[name = string("op_9644_cast_fp16")]; + string var_9646_equation_0 = const()[name = string("op_9646_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9646_cast_fp16 = einsum(equation = var_9646_equation_0, values = (var_9326_cast_fp16, var_9565_cast_fp16))[name = string("op_9646_cast_fp16")]; + string var_9648_equation_0 = const()[name = string("op_9648_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9648_cast_fp16 = einsum(equation = var_9648_equation_0, values = (var_9326_cast_fp16, var_9566_cast_fp16))[name = string("op_9648_cast_fp16")]; + string var_9650_equation_0 = const()[name = string("op_9650_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9650_cast_fp16 = einsum(equation = var_9650_equation_0, values = (var_9326_cast_fp16, var_9567_cast_fp16))[name = string("op_9650_cast_fp16")]; + string var_9652_equation_0 = const()[name = string("op_9652_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9652_cast_fp16 = einsum(equation = var_9652_equation_0, values = (var_9326_cast_fp16, var_9568_cast_fp16))[name = string("op_9652_cast_fp16")]; + string var_9654_equation_0 = const()[name = string("op_9654_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9654_cast_fp16 = einsum(equation = var_9654_equation_0, values = (var_9330_cast_fp16, var_9569_cast_fp16))[name = string("op_9654_cast_fp16")]; + string var_9656_equation_0 = const()[name = string("op_9656_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9656_cast_fp16 = einsum(equation = var_9656_equation_0, values = (var_9330_cast_fp16, var_9570_cast_fp16))[name = string("op_9656_cast_fp16")]; + string var_9658_equation_0 = const()[name = string("op_9658_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9658_cast_fp16 = einsum(equation = var_9658_equation_0, values = (var_9330_cast_fp16, var_9571_cast_fp16))[name = string("op_9658_cast_fp16")]; + string var_9660_equation_0 = const()[name = string("op_9660_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9660_cast_fp16 = einsum(equation = var_9660_equation_0, values = (var_9330_cast_fp16, var_9572_cast_fp16))[name = string("op_9660_cast_fp16")]; + string var_9662_equation_0 = const()[name = string("op_9662_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9662_cast_fp16 = einsum(equation = var_9662_equation_0, values = (var_9334_cast_fp16, var_9573_cast_fp16))[name = string("op_9662_cast_fp16")]; + string var_9664_equation_0 = const()[name = string("op_9664_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9664_cast_fp16 = einsum(equation = var_9664_equation_0, values = (var_9334_cast_fp16, var_9574_cast_fp16))[name = string("op_9664_cast_fp16")]; + string var_9666_equation_0 = const()[name = string("op_9666_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9666_cast_fp16 = einsum(equation = var_9666_equation_0, values = (var_9334_cast_fp16, var_9575_cast_fp16))[name = string("op_9666_cast_fp16")]; + string var_9668_equation_0 = const()[name = string("op_9668_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9668_cast_fp16 = einsum(equation = var_9668_equation_0, values = (var_9334_cast_fp16, var_9576_cast_fp16))[name = string("op_9668_cast_fp16")]; + string var_9670_equation_0 = const()[name = string("op_9670_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9670_cast_fp16 = einsum(equation = var_9670_equation_0, values = (var_9338_cast_fp16, var_9577_cast_fp16))[name = string("op_9670_cast_fp16")]; + string var_9672_equation_0 = const()[name = string("op_9672_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9672_cast_fp16 = einsum(equation = var_9672_equation_0, values = (var_9338_cast_fp16, var_9578_cast_fp16))[name = string("op_9672_cast_fp16")]; + string var_9674_equation_0 = const()[name = string("op_9674_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9674_cast_fp16 = einsum(equation = var_9674_equation_0, values = (var_9338_cast_fp16, var_9579_cast_fp16))[name = string("op_9674_cast_fp16")]; + string var_9676_equation_0 = const()[name = string("op_9676_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_9676_cast_fp16 = einsum(equation = var_9676_equation_0, values = (var_9338_cast_fp16, var_9580_cast_fp16))[name = string("op_9676_cast_fp16")]; + bool var_9678_interleave_0 = const()[name = string("op_9678_interleave_0"), val = bool(false)]; + tensor var_9678_cast_fp16 = concat(axis = var_8789, interleave = var_9678_interleave_0, values = (var_9582_cast_fp16, var_9584_cast_fp16, var_9586_cast_fp16, var_9588_cast_fp16))[name = string("op_9678_cast_fp16")]; + bool var_9680_interleave_0 = const()[name = string("op_9680_interleave_0"), val = bool(false)]; + tensor var_9680_cast_fp16 = concat(axis = var_8789, interleave = var_9680_interleave_0, values = (var_9590_cast_fp16, var_9592_cast_fp16, var_9594_cast_fp16, var_9596_cast_fp16))[name = string("op_9680_cast_fp16")]; + bool var_9682_interleave_0 = const()[name = string("op_9682_interleave_0"), val = bool(false)]; + tensor var_9682_cast_fp16 = concat(axis = var_8789, interleave = var_9682_interleave_0, values = (var_9598_cast_fp16, var_9600_cast_fp16, var_9602_cast_fp16, var_9604_cast_fp16))[name = string("op_9682_cast_fp16")]; + bool var_9684_interleave_0 = const()[name = string("op_9684_interleave_0"), val = bool(false)]; + tensor var_9684_cast_fp16 = concat(axis = var_8789, interleave = var_9684_interleave_0, values = (var_9606_cast_fp16, var_9608_cast_fp16, var_9610_cast_fp16, var_9612_cast_fp16))[name = string("op_9684_cast_fp16")]; + bool var_9686_interleave_0 = const()[name = string("op_9686_interleave_0"), val = bool(false)]; + tensor var_9686_cast_fp16 = concat(axis = var_8789, interleave = var_9686_interleave_0, values = (var_9614_cast_fp16, var_9616_cast_fp16, var_9618_cast_fp16, var_9620_cast_fp16))[name = string("op_9686_cast_fp16")]; + bool var_9688_interleave_0 = const()[name = string("op_9688_interleave_0"), val = bool(false)]; + tensor var_9688_cast_fp16 = concat(axis = var_8789, interleave = var_9688_interleave_0, values = (var_9622_cast_fp16, var_9624_cast_fp16, var_9626_cast_fp16, var_9628_cast_fp16))[name = string("op_9688_cast_fp16")]; + bool var_9690_interleave_0 = const()[name = string("op_9690_interleave_0"), val = bool(false)]; + tensor var_9690_cast_fp16 = concat(axis = var_8789, interleave = var_9690_interleave_0, values = (var_9630_cast_fp16, var_9632_cast_fp16, var_9634_cast_fp16, var_9636_cast_fp16))[name = string("op_9690_cast_fp16")]; + bool var_9692_interleave_0 = const()[name = string("op_9692_interleave_0"), val = bool(false)]; + tensor var_9692_cast_fp16 = concat(axis = var_8789, interleave = var_9692_interleave_0, values = (var_9638_cast_fp16, var_9640_cast_fp16, var_9642_cast_fp16, var_9644_cast_fp16))[name = string("op_9692_cast_fp16")]; + bool var_9694_interleave_0 = const()[name = string("op_9694_interleave_0"), val = bool(false)]; + tensor var_9694_cast_fp16 = concat(axis = var_8789, interleave = var_9694_interleave_0, values = (var_9646_cast_fp16, var_9648_cast_fp16, var_9650_cast_fp16, var_9652_cast_fp16))[name = string("op_9694_cast_fp16")]; + bool var_9696_interleave_0 = const()[name = string("op_9696_interleave_0"), val = bool(false)]; + tensor var_9696_cast_fp16 = concat(axis = var_8789, interleave = var_9696_interleave_0, values = (var_9654_cast_fp16, var_9656_cast_fp16, var_9658_cast_fp16, var_9660_cast_fp16))[name = string("op_9696_cast_fp16")]; + bool var_9698_interleave_0 = const()[name = string("op_9698_interleave_0"), val = bool(false)]; + tensor var_9698_cast_fp16 = concat(axis = var_8789, interleave = var_9698_interleave_0, values = (var_9662_cast_fp16, var_9664_cast_fp16, var_9666_cast_fp16, var_9668_cast_fp16))[name = string("op_9698_cast_fp16")]; + bool var_9700_interleave_0 = const()[name = string("op_9700_interleave_0"), val = bool(false)]; + tensor var_9700_cast_fp16 = concat(axis = var_8789, interleave = var_9700_interleave_0, values = (var_9670_cast_fp16, var_9672_cast_fp16, var_9674_cast_fp16, var_9676_cast_fp16))[name = string("op_9700_cast_fp16")]; + bool input_73_interleave_0 = const()[name = string("input_73_interleave_0"), val = bool(false)]; + tensor input_73_cast_fp16 = concat(axis = var_8806, interleave = input_73_interleave_0, values = (var_9678_cast_fp16, var_9680_cast_fp16, var_9682_cast_fp16, var_9684_cast_fp16, var_9686_cast_fp16, var_9688_cast_fp16, var_9690_cast_fp16, var_9692_cast_fp16, var_9694_cast_fp16, var_9696_cast_fp16, var_9698_cast_fp16, var_9700_cast_fp16))[name = string("input_73_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137340288)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138520000)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_9719_to_fp16 = const()[name = string("op_9719_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_9719_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138521600)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138523200)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138524800)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143243456)))]; + tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143249664)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147968320)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_41_cast_fp16")]; + int32 var_9748 = const()[name = string("op_9748"), val = int32(3)]; + int32 var_9765 = const()[name = string("op_9765"), val = int32(1)]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_9782_to_fp16 = const()[name = string("op_9782_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_9782_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = string("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147969920)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = string("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147971520)))]; + fp16 obj_41_epsilon_0_to_fp16 = const()[name = string("obj_41_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("obj_41_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147973120)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149152832)))]; + tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("query_21_cast_fp16")]; + string key_21_pad_type_0 = const()[name = string("key_21_pad_type_0"), val = string("valid")]; + tensor key_21_strides_0 = const()[name = string("key_21_strides_0"), val = tensor([1, 1])]; + tensor key_21_pad_0 = const()[name = string("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_21_dilations_0 = const()[name = string("key_21_dilations_0"), val = tensor([1, 1])]; + int32 key_21_groups_0 = const()[name = string("key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149154432)))]; + tensor key_21_cast_fp16 = conv(dilations = key_21_dilations_0, groups = key_21_groups_0, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("key_21_cast_fp16")]; + string value_21_pad_type_0 = const()[name = string("value_21_pad_type_0"), val = string("valid")]; + tensor value_21_strides_0 = const()[name = string("value_21_strides_0"), val = tensor([1, 1])]; + tensor value_21_pad_0 = const()[name = string("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_21_dilations_0 = const()[name = string("value_21_dilations_0"), val = tensor([1, 1])]; + int32 value_21_groups_0 = const()[name = string("value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150334144)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151513856)))]; + tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = value_21_dilations_0, groups = value_21_groups_0, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_9820_begin_0 = const()[name = string("op_9820_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9820_end_0 = const()[name = string("op_9820_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9820_end_mask_0 = const()[name = string("op_9820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9820_cast_fp16 = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = query_21_cast_fp16)[name = string("op_9820_cast_fp16")]; + tensor var_9824_begin_0 = const()[name = string("op_9824_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9824_end_0 = const()[name = string("op_9824_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9824_end_mask_0 = const()[name = string("op_9824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9824_cast_fp16 = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = query_21_cast_fp16)[name = string("op_9824_cast_fp16")]; + tensor var_9828_begin_0 = const()[name = string("op_9828_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9828_end_0 = const()[name = string("op_9828_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9828_end_mask_0 = const()[name = string("op_9828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9828_cast_fp16 = slice_by_index(begin = var_9828_begin_0, end = var_9828_end_0, end_mask = var_9828_end_mask_0, x = query_21_cast_fp16)[name = string("op_9828_cast_fp16")]; + tensor var_9832_begin_0 = const()[name = string("op_9832_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9832_end_0 = const()[name = string("op_9832_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9832_end_mask_0 = const()[name = string("op_9832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9832_cast_fp16 = slice_by_index(begin = var_9832_begin_0, end = var_9832_end_0, end_mask = var_9832_end_mask_0, x = query_21_cast_fp16)[name = string("op_9832_cast_fp16")]; + tensor var_9836_begin_0 = const()[name = string("op_9836_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9836_end_0 = const()[name = string("op_9836_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9836_end_mask_0 = const()[name = string("op_9836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9836_cast_fp16 = slice_by_index(begin = var_9836_begin_0, end = var_9836_end_0, end_mask = var_9836_end_mask_0, x = query_21_cast_fp16)[name = string("op_9836_cast_fp16")]; + tensor var_9840_begin_0 = const()[name = string("op_9840_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9840_end_0 = const()[name = string("op_9840_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9840_end_mask_0 = const()[name = string("op_9840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9840_cast_fp16 = slice_by_index(begin = var_9840_begin_0, end = var_9840_end_0, end_mask = var_9840_end_mask_0, x = query_21_cast_fp16)[name = string("op_9840_cast_fp16")]; + tensor var_9844_begin_0 = const()[name = string("op_9844_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9844_end_0 = const()[name = string("op_9844_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9844_end_mask_0 = const()[name = string("op_9844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9844_cast_fp16 = slice_by_index(begin = var_9844_begin_0, end = var_9844_end_0, end_mask = var_9844_end_mask_0, x = query_21_cast_fp16)[name = string("op_9844_cast_fp16")]; + tensor var_9848_begin_0 = const()[name = string("op_9848_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9848_end_0 = const()[name = string("op_9848_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9848_end_mask_0 = const()[name = string("op_9848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9848_cast_fp16 = slice_by_index(begin = var_9848_begin_0, end = var_9848_end_0, end_mask = var_9848_end_mask_0, x = query_21_cast_fp16)[name = string("op_9848_cast_fp16")]; + tensor var_9852_begin_0 = const()[name = string("op_9852_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9852_end_0 = const()[name = string("op_9852_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9852_end_mask_0 = const()[name = string("op_9852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9852_cast_fp16 = slice_by_index(begin = var_9852_begin_0, end = var_9852_end_0, end_mask = var_9852_end_mask_0, x = query_21_cast_fp16)[name = string("op_9852_cast_fp16")]; + tensor var_9856_begin_0 = const()[name = string("op_9856_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9856_end_0 = const()[name = string("op_9856_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9856_end_mask_0 = const()[name = string("op_9856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9856_cast_fp16 = slice_by_index(begin = var_9856_begin_0, end = var_9856_end_0, end_mask = var_9856_end_mask_0, x = query_21_cast_fp16)[name = string("op_9856_cast_fp16")]; + tensor var_9860_begin_0 = const()[name = string("op_9860_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9860_end_0 = const()[name = string("op_9860_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9860_end_mask_0 = const()[name = string("op_9860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9860_cast_fp16 = slice_by_index(begin = var_9860_begin_0, end = var_9860_end_0, end_mask = var_9860_end_mask_0, x = query_21_cast_fp16)[name = string("op_9860_cast_fp16")]; + tensor var_9864_begin_0 = const()[name = string("op_9864_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9864_end_0 = const()[name = string("op_9864_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9864_end_mask_0 = const()[name = string("op_9864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9864_cast_fp16 = slice_by_index(begin = var_9864_begin_0, end = var_9864_end_0, end_mask = var_9864_end_mask_0, x = query_21_cast_fp16)[name = string("op_9864_cast_fp16")]; + tensor var_9873_begin_0 = const()[name = string("op_9873_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9873_end_0 = const()[name = string("op_9873_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9873_end_mask_0 = const()[name = string("op_9873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9873_cast_fp16 = slice_by_index(begin = var_9873_begin_0, end = var_9873_end_0, end_mask = var_9873_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9873_cast_fp16")]; + tensor var_9880_begin_0 = const()[name = string("op_9880_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9880_end_0 = const()[name = string("op_9880_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9880_end_mask_0 = const()[name = string("op_9880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9880_cast_fp16 = slice_by_index(begin = var_9880_begin_0, end = var_9880_end_0, end_mask = var_9880_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9880_cast_fp16")]; + tensor var_9887_begin_0 = const()[name = string("op_9887_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9887_end_0 = const()[name = string("op_9887_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9887_end_mask_0 = const()[name = string("op_9887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9887_cast_fp16 = slice_by_index(begin = var_9887_begin_0, end = var_9887_end_0, end_mask = var_9887_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9887_cast_fp16")]; + tensor var_9894_begin_0 = const()[name = string("op_9894_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9894_end_0 = const()[name = string("op_9894_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9894_end_mask_0 = const()[name = string("op_9894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9894_cast_fp16 = slice_by_index(begin = var_9894_begin_0, end = var_9894_end_0, end_mask = var_9894_end_mask_0, x = var_9820_cast_fp16)[name = string("op_9894_cast_fp16")]; + tensor var_9901_begin_0 = const()[name = string("op_9901_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9901_end_0 = const()[name = string("op_9901_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9901_end_mask_0 = const()[name = string("op_9901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9901_cast_fp16 = slice_by_index(begin = var_9901_begin_0, end = var_9901_end_0, end_mask = var_9901_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9901_cast_fp16")]; + tensor var_9908_begin_0 = const()[name = string("op_9908_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9908_end_0 = const()[name = string("op_9908_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9908_end_mask_0 = const()[name = string("op_9908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9908_cast_fp16 = slice_by_index(begin = var_9908_begin_0, end = var_9908_end_0, end_mask = var_9908_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9908_cast_fp16")]; + tensor var_9915_begin_0 = const()[name = string("op_9915_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9915_end_0 = const()[name = string("op_9915_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9915_end_mask_0 = const()[name = string("op_9915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9915_cast_fp16 = slice_by_index(begin = var_9915_begin_0, end = var_9915_end_0, end_mask = var_9915_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9915_cast_fp16")]; + tensor var_9922_begin_0 = const()[name = string("op_9922_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9922_end_0 = const()[name = string("op_9922_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9922_end_mask_0 = const()[name = string("op_9922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9922_cast_fp16 = slice_by_index(begin = var_9922_begin_0, end = var_9922_end_0, end_mask = var_9922_end_mask_0, x = var_9824_cast_fp16)[name = string("op_9922_cast_fp16")]; + tensor var_9929_begin_0 = const()[name = string("op_9929_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9929_end_0 = const()[name = string("op_9929_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9929_end_mask_0 = const()[name = string("op_9929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9929_cast_fp16 = slice_by_index(begin = var_9929_begin_0, end = var_9929_end_0, end_mask = var_9929_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9929_cast_fp16")]; + tensor var_9936_begin_0 = const()[name = string("op_9936_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9936_end_0 = const()[name = string("op_9936_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9936_end_mask_0 = const()[name = string("op_9936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9936_cast_fp16 = slice_by_index(begin = var_9936_begin_0, end = var_9936_end_0, end_mask = var_9936_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9936_cast_fp16")]; + tensor var_9943_begin_0 = const()[name = string("op_9943_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9943_end_0 = const()[name = string("op_9943_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9943_end_mask_0 = const()[name = string("op_9943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9943_cast_fp16 = slice_by_index(begin = var_9943_begin_0, end = var_9943_end_0, end_mask = var_9943_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9943_cast_fp16")]; + tensor var_9950_begin_0 = const()[name = string("op_9950_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9950_end_0 = const()[name = string("op_9950_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9950_end_mask_0 = const()[name = string("op_9950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9950_cast_fp16 = slice_by_index(begin = var_9950_begin_0, end = var_9950_end_0, end_mask = var_9950_end_mask_0, x = var_9828_cast_fp16)[name = string("op_9950_cast_fp16")]; + tensor var_9957_begin_0 = const()[name = string("op_9957_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9957_end_0 = const()[name = string("op_9957_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9957_end_mask_0 = const()[name = string("op_9957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9957_cast_fp16 = slice_by_index(begin = var_9957_begin_0, end = var_9957_end_0, end_mask = var_9957_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9957_cast_fp16")]; + tensor var_9964_begin_0 = const()[name = string("op_9964_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9964_end_0 = const()[name = string("op_9964_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9964_end_mask_0 = const()[name = string("op_9964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9964_cast_fp16 = slice_by_index(begin = var_9964_begin_0, end = var_9964_end_0, end_mask = var_9964_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9964_cast_fp16")]; + tensor var_9971_begin_0 = const()[name = string("op_9971_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9971_end_0 = const()[name = string("op_9971_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9971_end_mask_0 = const()[name = string("op_9971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9971_cast_fp16 = slice_by_index(begin = var_9971_begin_0, end = var_9971_end_0, end_mask = var_9971_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9971_cast_fp16")]; + tensor var_9978_begin_0 = const()[name = string("op_9978_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9978_end_0 = const()[name = string("op_9978_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9978_end_mask_0 = const()[name = string("op_9978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9978_cast_fp16 = slice_by_index(begin = var_9978_begin_0, end = var_9978_end_0, end_mask = var_9978_end_mask_0, x = var_9832_cast_fp16)[name = string("op_9978_cast_fp16")]; + tensor var_9985_begin_0 = const()[name = string("op_9985_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9985_end_0 = const()[name = string("op_9985_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9985_end_mask_0 = const()[name = string("op_9985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9985_cast_fp16 = slice_by_index(begin = var_9985_begin_0, end = var_9985_end_0, end_mask = var_9985_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9985_cast_fp16")]; + tensor var_9992_begin_0 = const()[name = string("op_9992_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9992_end_0 = const()[name = string("op_9992_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9992_end_mask_0 = const()[name = string("op_9992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9992_cast_fp16 = slice_by_index(begin = var_9992_begin_0, end = var_9992_end_0, end_mask = var_9992_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9992_cast_fp16")]; + tensor var_9999_begin_0 = const()[name = string("op_9999_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9999_end_0 = const()[name = string("op_9999_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9999_end_mask_0 = const()[name = string("op_9999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9999_cast_fp16 = slice_by_index(begin = var_9999_begin_0, end = var_9999_end_0, end_mask = var_9999_end_mask_0, x = var_9836_cast_fp16)[name = string("op_9999_cast_fp16")]; + tensor var_10006_begin_0 = const()[name = string("op_10006_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10006_end_0 = const()[name = string("op_10006_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10006_end_mask_0 = const()[name = string("op_10006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10006_cast_fp16 = slice_by_index(begin = var_10006_begin_0, end = var_10006_end_0, end_mask = var_10006_end_mask_0, x = var_9836_cast_fp16)[name = string("op_10006_cast_fp16")]; + tensor var_10013_begin_0 = const()[name = string("op_10013_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10013_end_0 = const()[name = string("op_10013_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10013_end_mask_0 = const()[name = string("op_10013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10013_cast_fp16 = slice_by_index(begin = var_10013_begin_0, end = var_10013_end_0, end_mask = var_10013_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10013_cast_fp16")]; + tensor var_10020_begin_0 = const()[name = string("op_10020_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10020_end_0 = const()[name = string("op_10020_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10020_end_mask_0 = const()[name = string("op_10020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10020_cast_fp16 = slice_by_index(begin = var_10020_begin_0, end = var_10020_end_0, end_mask = var_10020_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10020_cast_fp16")]; + tensor var_10027_begin_0 = const()[name = string("op_10027_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10027_end_0 = const()[name = string("op_10027_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10027_end_mask_0 = const()[name = string("op_10027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10027_cast_fp16 = slice_by_index(begin = var_10027_begin_0, end = var_10027_end_0, end_mask = var_10027_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10027_cast_fp16")]; + tensor var_10034_begin_0 = const()[name = string("op_10034_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10034_end_0 = const()[name = string("op_10034_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10034_end_mask_0 = const()[name = string("op_10034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10034_cast_fp16 = slice_by_index(begin = var_10034_begin_0, end = var_10034_end_0, end_mask = var_10034_end_mask_0, x = var_9840_cast_fp16)[name = string("op_10034_cast_fp16")]; + tensor var_10041_begin_0 = const()[name = string("op_10041_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10041_end_0 = const()[name = string("op_10041_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10041_end_mask_0 = const()[name = string("op_10041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10041_cast_fp16 = slice_by_index(begin = var_10041_begin_0, end = var_10041_end_0, end_mask = var_10041_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10041_cast_fp16")]; + tensor var_10048_begin_0 = const()[name = string("op_10048_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10048_end_0 = const()[name = string("op_10048_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10048_end_mask_0 = const()[name = string("op_10048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10048_cast_fp16 = slice_by_index(begin = var_10048_begin_0, end = var_10048_end_0, end_mask = var_10048_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10048_cast_fp16")]; + tensor var_10055_begin_0 = const()[name = string("op_10055_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10055_end_0 = const()[name = string("op_10055_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10055_end_mask_0 = const()[name = string("op_10055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10055_cast_fp16 = slice_by_index(begin = var_10055_begin_0, end = var_10055_end_0, end_mask = var_10055_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10055_cast_fp16")]; + tensor var_10062_begin_0 = const()[name = string("op_10062_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10062_end_0 = const()[name = string("op_10062_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10062_end_mask_0 = const()[name = string("op_10062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10062_cast_fp16 = slice_by_index(begin = var_10062_begin_0, end = var_10062_end_0, end_mask = var_10062_end_mask_0, x = var_9844_cast_fp16)[name = string("op_10062_cast_fp16")]; + tensor var_10069_begin_0 = const()[name = string("op_10069_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10069_end_0 = const()[name = string("op_10069_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10069_end_mask_0 = const()[name = string("op_10069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10069_cast_fp16 = slice_by_index(begin = var_10069_begin_0, end = var_10069_end_0, end_mask = var_10069_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10069_cast_fp16")]; + tensor var_10076_begin_0 = const()[name = string("op_10076_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10076_end_0 = const()[name = string("op_10076_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10076_end_mask_0 = const()[name = string("op_10076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10076_cast_fp16 = slice_by_index(begin = var_10076_begin_0, end = var_10076_end_0, end_mask = var_10076_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10076_cast_fp16")]; + tensor var_10083_begin_0 = const()[name = string("op_10083_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10083_end_0 = const()[name = string("op_10083_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10083_end_mask_0 = const()[name = string("op_10083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10083_cast_fp16 = slice_by_index(begin = var_10083_begin_0, end = var_10083_end_0, end_mask = var_10083_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10083_cast_fp16")]; + tensor var_10090_begin_0 = const()[name = string("op_10090_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10090_end_0 = const()[name = string("op_10090_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10090_end_mask_0 = const()[name = string("op_10090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10090_cast_fp16 = slice_by_index(begin = var_10090_begin_0, end = var_10090_end_0, end_mask = var_10090_end_mask_0, x = var_9848_cast_fp16)[name = string("op_10090_cast_fp16")]; + tensor var_10097_begin_0 = const()[name = string("op_10097_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10097_end_0 = const()[name = string("op_10097_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10097_end_mask_0 = const()[name = string("op_10097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10097_cast_fp16 = slice_by_index(begin = var_10097_begin_0, end = var_10097_end_0, end_mask = var_10097_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10097_cast_fp16")]; + tensor var_10104_begin_0 = const()[name = string("op_10104_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10104_end_0 = const()[name = string("op_10104_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10104_end_mask_0 = const()[name = string("op_10104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10104_cast_fp16 = slice_by_index(begin = var_10104_begin_0, end = var_10104_end_0, end_mask = var_10104_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10104_cast_fp16")]; + tensor var_10111_begin_0 = const()[name = string("op_10111_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10111_end_0 = const()[name = string("op_10111_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10111_end_mask_0 = const()[name = string("op_10111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10111_cast_fp16 = slice_by_index(begin = var_10111_begin_0, end = var_10111_end_0, end_mask = var_10111_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10111_cast_fp16")]; + tensor var_10118_begin_0 = const()[name = string("op_10118_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10118_end_0 = const()[name = string("op_10118_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10118_end_mask_0 = const()[name = string("op_10118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10118_cast_fp16 = slice_by_index(begin = var_10118_begin_0, end = var_10118_end_0, end_mask = var_10118_end_mask_0, x = var_9852_cast_fp16)[name = string("op_10118_cast_fp16")]; + tensor var_10125_begin_0 = const()[name = string("op_10125_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10125_end_0 = const()[name = string("op_10125_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10125_end_mask_0 = const()[name = string("op_10125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10125_cast_fp16 = slice_by_index(begin = var_10125_begin_0, end = var_10125_end_0, end_mask = var_10125_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10125_cast_fp16")]; + tensor var_10132_begin_0 = const()[name = string("op_10132_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10132_end_0 = const()[name = string("op_10132_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10132_end_mask_0 = const()[name = string("op_10132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10132_cast_fp16 = slice_by_index(begin = var_10132_begin_0, end = var_10132_end_0, end_mask = var_10132_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10132_cast_fp16")]; + tensor var_10139_begin_0 = const()[name = string("op_10139_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10139_end_0 = const()[name = string("op_10139_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10139_end_mask_0 = const()[name = string("op_10139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10139_cast_fp16 = slice_by_index(begin = var_10139_begin_0, end = var_10139_end_0, end_mask = var_10139_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10139_cast_fp16")]; + tensor var_10146_begin_0 = const()[name = string("op_10146_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10146_end_0 = const()[name = string("op_10146_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10146_end_mask_0 = const()[name = string("op_10146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10146_cast_fp16 = slice_by_index(begin = var_10146_begin_0, end = var_10146_end_0, end_mask = var_10146_end_mask_0, x = var_9856_cast_fp16)[name = string("op_10146_cast_fp16")]; + tensor var_10153_begin_0 = const()[name = string("op_10153_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10153_end_0 = const()[name = string("op_10153_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10153_end_mask_0 = const()[name = string("op_10153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10153_cast_fp16 = slice_by_index(begin = var_10153_begin_0, end = var_10153_end_0, end_mask = var_10153_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10153_cast_fp16")]; + tensor var_10160_begin_0 = const()[name = string("op_10160_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10160_end_0 = const()[name = string("op_10160_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10160_end_mask_0 = const()[name = string("op_10160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10160_cast_fp16 = slice_by_index(begin = var_10160_begin_0, end = var_10160_end_0, end_mask = var_10160_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10160_cast_fp16")]; + tensor var_10167_begin_0 = const()[name = string("op_10167_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10167_end_0 = const()[name = string("op_10167_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10167_end_mask_0 = const()[name = string("op_10167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10167_cast_fp16 = slice_by_index(begin = var_10167_begin_0, end = var_10167_end_0, end_mask = var_10167_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10167_cast_fp16")]; + tensor var_10174_begin_0 = const()[name = string("op_10174_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10174_end_0 = const()[name = string("op_10174_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10174_end_mask_0 = const()[name = string("op_10174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10174_cast_fp16 = slice_by_index(begin = var_10174_begin_0, end = var_10174_end_0, end_mask = var_10174_end_mask_0, x = var_9860_cast_fp16)[name = string("op_10174_cast_fp16")]; + tensor var_10181_begin_0 = const()[name = string("op_10181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10181_end_0 = const()[name = string("op_10181_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10181_end_mask_0 = const()[name = string("op_10181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10181_cast_fp16 = slice_by_index(begin = var_10181_begin_0, end = var_10181_end_0, end_mask = var_10181_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10181_cast_fp16")]; + tensor var_10188_begin_0 = const()[name = string("op_10188_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10188_end_0 = const()[name = string("op_10188_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10188_end_mask_0 = const()[name = string("op_10188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10188_cast_fp16 = slice_by_index(begin = var_10188_begin_0, end = var_10188_end_0, end_mask = var_10188_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10188_cast_fp16")]; + tensor var_10195_begin_0 = const()[name = string("op_10195_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10195_end_0 = const()[name = string("op_10195_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10195_end_mask_0 = const()[name = string("op_10195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10195_cast_fp16 = slice_by_index(begin = var_10195_begin_0, end = var_10195_end_0, end_mask = var_10195_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10195_cast_fp16")]; + tensor var_10202_begin_0 = const()[name = string("op_10202_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10202_end_0 = const()[name = string("op_10202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10202_end_mask_0 = const()[name = string("op_10202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10202_cast_fp16 = slice_by_index(begin = var_10202_begin_0, end = var_10202_end_0, end_mask = var_10202_end_mask_0, x = var_9864_cast_fp16)[name = string("op_10202_cast_fp16")]; + tensor k_21_perm_0 = const()[name = string("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10207_begin_0 = const()[name = string("op_10207_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10207_end_0 = const()[name = string("op_10207_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_10207_end_mask_0 = const()[name = string("op_10207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = string("transpose_1")]; + tensor var_10207_cast_fp16 = slice_by_index(begin = var_10207_begin_0, end = var_10207_end_0, end_mask = var_10207_end_mask_0, x = k_21_cast_fp16)[name = string("op_10207_cast_fp16")]; + tensor var_10211_begin_0 = const()[name = string("op_10211_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10211_end_0 = const()[name = string("op_10211_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_10211_end_mask_0 = const()[name = string("op_10211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10211_cast_fp16 = slice_by_index(begin = var_10211_begin_0, end = var_10211_end_0, end_mask = var_10211_end_mask_0, x = k_21_cast_fp16)[name = string("op_10211_cast_fp16")]; + tensor var_10215_begin_0 = const()[name = string("op_10215_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10215_end_0 = const()[name = string("op_10215_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_10215_end_mask_0 = const()[name = string("op_10215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10215_cast_fp16 = slice_by_index(begin = var_10215_begin_0, end = var_10215_end_0, end_mask = var_10215_end_mask_0, x = k_21_cast_fp16)[name = string("op_10215_cast_fp16")]; + tensor var_10219_begin_0 = const()[name = string("op_10219_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10219_end_0 = const()[name = string("op_10219_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_10219_end_mask_0 = const()[name = string("op_10219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10219_cast_fp16 = slice_by_index(begin = var_10219_begin_0, end = var_10219_end_0, end_mask = var_10219_end_mask_0, x = k_21_cast_fp16)[name = string("op_10219_cast_fp16")]; + tensor var_10223_begin_0 = const()[name = string("op_10223_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10223_end_0 = const()[name = string("op_10223_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_10223_end_mask_0 = const()[name = string("op_10223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10223_cast_fp16 = slice_by_index(begin = var_10223_begin_0, end = var_10223_end_0, end_mask = var_10223_end_mask_0, x = k_21_cast_fp16)[name = string("op_10223_cast_fp16")]; + tensor var_10227_begin_0 = const()[name = string("op_10227_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10227_end_0 = const()[name = string("op_10227_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_10227_end_mask_0 = const()[name = string("op_10227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10227_cast_fp16 = slice_by_index(begin = var_10227_begin_0, end = var_10227_end_0, end_mask = var_10227_end_mask_0, x = k_21_cast_fp16)[name = string("op_10227_cast_fp16")]; + tensor var_10231_begin_0 = const()[name = string("op_10231_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10231_end_0 = const()[name = string("op_10231_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_10231_end_mask_0 = const()[name = string("op_10231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10231_cast_fp16 = slice_by_index(begin = var_10231_begin_0, end = var_10231_end_0, end_mask = var_10231_end_mask_0, x = k_21_cast_fp16)[name = string("op_10231_cast_fp16")]; + tensor var_10235_begin_0 = const()[name = string("op_10235_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10235_end_0 = const()[name = string("op_10235_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_10235_end_mask_0 = const()[name = string("op_10235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10235_cast_fp16 = slice_by_index(begin = var_10235_begin_0, end = var_10235_end_0, end_mask = var_10235_end_mask_0, x = k_21_cast_fp16)[name = string("op_10235_cast_fp16")]; + tensor var_10239_begin_0 = const()[name = string("op_10239_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10239_end_0 = const()[name = string("op_10239_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_10239_end_mask_0 = const()[name = string("op_10239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10239_cast_fp16 = slice_by_index(begin = var_10239_begin_0, end = var_10239_end_0, end_mask = var_10239_end_mask_0, x = k_21_cast_fp16)[name = string("op_10239_cast_fp16")]; + tensor var_10243_begin_0 = const()[name = string("op_10243_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10243_end_0 = const()[name = string("op_10243_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_10243_end_mask_0 = const()[name = string("op_10243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10243_cast_fp16 = slice_by_index(begin = var_10243_begin_0, end = var_10243_end_0, end_mask = var_10243_end_mask_0, x = k_21_cast_fp16)[name = string("op_10243_cast_fp16")]; + tensor var_10247_begin_0 = const()[name = string("op_10247_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10247_end_0 = const()[name = string("op_10247_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_10247_end_mask_0 = const()[name = string("op_10247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10247_cast_fp16 = slice_by_index(begin = var_10247_begin_0, end = var_10247_end_0, end_mask = var_10247_end_mask_0, x = k_21_cast_fp16)[name = string("op_10247_cast_fp16")]; + tensor var_10251_begin_0 = const()[name = string("op_10251_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10251_end_0 = const()[name = string("op_10251_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_10251_end_mask_0 = const()[name = string("op_10251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10251_cast_fp16 = slice_by_index(begin = var_10251_begin_0, end = var_10251_end_0, end_mask = var_10251_end_mask_0, x = k_21_cast_fp16)[name = string("op_10251_cast_fp16")]; + tensor var_10253_begin_0 = const()[name = string("op_10253_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10253_end_0 = const()[name = string("op_10253_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10253_end_mask_0 = const()[name = string("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = value_21_cast_fp16)[name = string("op_10253_cast_fp16")]; + tensor var_10257_begin_0 = const()[name = string("op_10257_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10257_end_0 = const()[name = string("op_10257_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10257_end_mask_0 = const()[name = string("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = value_21_cast_fp16)[name = string("op_10257_cast_fp16")]; + tensor var_10261_begin_0 = const()[name = string("op_10261_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10261_end_0 = const()[name = string("op_10261_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10261_end_mask_0 = const()[name = string("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = value_21_cast_fp16)[name = string("op_10261_cast_fp16")]; + tensor var_10265_begin_0 = const()[name = string("op_10265_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10265_end_0 = const()[name = string("op_10265_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10265_end_mask_0 = const()[name = string("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = value_21_cast_fp16)[name = string("op_10265_cast_fp16")]; + tensor var_10269_begin_0 = const()[name = string("op_10269_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10269_end_0 = const()[name = string("op_10269_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10269_end_mask_0 = const()[name = string("op_10269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10269_cast_fp16 = slice_by_index(begin = var_10269_begin_0, end = var_10269_end_0, end_mask = var_10269_end_mask_0, x = value_21_cast_fp16)[name = string("op_10269_cast_fp16")]; + tensor var_10273_begin_0 = const()[name = string("op_10273_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10273_end_0 = const()[name = string("op_10273_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10273_end_mask_0 = const()[name = string("op_10273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10273_cast_fp16 = slice_by_index(begin = var_10273_begin_0, end = var_10273_end_0, end_mask = var_10273_end_mask_0, x = value_21_cast_fp16)[name = string("op_10273_cast_fp16")]; + tensor var_10277_begin_0 = const()[name = string("op_10277_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10277_end_0 = const()[name = string("op_10277_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10277_end_mask_0 = const()[name = string("op_10277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10277_cast_fp16 = slice_by_index(begin = var_10277_begin_0, end = var_10277_end_0, end_mask = var_10277_end_mask_0, x = value_21_cast_fp16)[name = string("op_10277_cast_fp16")]; + tensor var_10281_begin_0 = const()[name = string("op_10281_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10281_end_0 = const()[name = string("op_10281_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10281_end_mask_0 = const()[name = string("op_10281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10281_cast_fp16 = slice_by_index(begin = var_10281_begin_0, end = var_10281_end_0, end_mask = var_10281_end_mask_0, x = value_21_cast_fp16)[name = string("op_10281_cast_fp16")]; + tensor var_10285_begin_0 = const()[name = string("op_10285_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10285_end_0 = const()[name = string("op_10285_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10285_end_mask_0 = const()[name = string("op_10285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10285_cast_fp16 = slice_by_index(begin = var_10285_begin_0, end = var_10285_end_0, end_mask = var_10285_end_mask_0, x = value_21_cast_fp16)[name = string("op_10285_cast_fp16")]; + tensor var_10289_begin_0 = const()[name = string("op_10289_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10289_end_0 = const()[name = string("op_10289_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10289_end_mask_0 = const()[name = string("op_10289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10289_cast_fp16 = slice_by_index(begin = var_10289_begin_0, end = var_10289_end_0, end_mask = var_10289_end_mask_0, x = value_21_cast_fp16)[name = string("op_10289_cast_fp16")]; + tensor var_10293_begin_0 = const()[name = string("op_10293_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10293_end_0 = const()[name = string("op_10293_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10293_end_mask_0 = const()[name = string("op_10293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10293_cast_fp16 = slice_by_index(begin = var_10293_begin_0, end = var_10293_end_0, end_mask = var_10293_end_mask_0, x = value_21_cast_fp16)[name = string("op_10293_cast_fp16")]; + tensor var_10297_begin_0 = const()[name = string("op_10297_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10297_end_0 = const()[name = string("op_10297_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10297_end_mask_0 = const()[name = string("op_10297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10297_cast_fp16 = slice_by_index(begin = var_10297_begin_0, end = var_10297_end_0, end_mask = var_10297_end_mask_0, x = value_21_cast_fp16)[name = string("op_10297_cast_fp16")]; + string _SplitHeadsQ__mh_w_961_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_961_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_961_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_961_equation_0, values = (var_10207_cast_fp16, var_9873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_961_cast_fp16")]; + string _SplitHeadsQ__mh_w_963_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_963_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_963_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_963_equation_0, values = (var_10207_cast_fp16, var_9880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_963_cast_fp16")]; + string _SplitHeadsQ__mh_w_965_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_965_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_965_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_965_equation_0, values = (var_10207_cast_fp16, var_9887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_965_cast_fp16")]; + string _SplitHeadsQ__mh_w_967_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_967_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_967_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_967_equation_0, values = (var_10207_cast_fp16, var_9894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_967_cast_fp16")]; + string _SplitHeadsQ__mh_w_969_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_969_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_969_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_969_equation_0, values = (var_10211_cast_fp16, var_9901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_969_cast_fp16")]; + string _SplitHeadsQ__mh_w_971_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_971_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_971_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_971_equation_0, values = (var_10211_cast_fp16, var_9908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_971_cast_fp16")]; + string _SplitHeadsQ__mh_w_973_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_973_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_973_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_973_equation_0, values = (var_10211_cast_fp16, var_9915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_973_cast_fp16")]; + string _SplitHeadsQ__mh_w_975_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_975_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_975_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_975_equation_0, values = (var_10211_cast_fp16, var_9922_cast_fp16))[name = string("_SplitHeadsQ__mh_w_975_cast_fp16")]; + string _SplitHeadsQ__mh_w_977_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_977_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_977_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_977_equation_0, values = (var_10215_cast_fp16, var_9929_cast_fp16))[name = string("_SplitHeadsQ__mh_w_977_cast_fp16")]; + string _SplitHeadsQ__mh_w_979_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_979_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_979_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_979_equation_0, values = (var_10215_cast_fp16, var_9936_cast_fp16))[name = string("_SplitHeadsQ__mh_w_979_cast_fp16")]; + string _SplitHeadsQ__mh_w_981_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_981_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_981_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_981_equation_0, values = (var_10215_cast_fp16, var_9943_cast_fp16))[name = string("_SplitHeadsQ__mh_w_981_cast_fp16")]; + string _SplitHeadsQ__mh_w_983_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_983_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_983_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_983_equation_0, values = (var_10215_cast_fp16, var_9950_cast_fp16))[name = string("_SplitHeadsQ__mh_w_983_cast_fp16")]; + string _SplitHeadsQ__mh_w_985_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_985_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_985_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_985_equation_0, values = (var_10219_cast_fp16, var_9957_cast_fp16))[name = string("_SplitHeadsQ__mh_w_985_cast_fp16")]; + string _SplitHeadsQ__mh_w_987_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_987_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_987_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_987_equation_0, values = (var_10219_cast_fp16, var_9964_cast_fp16))[name = string("_SplitHeadsQ__mh_w_987_cast_fp16")]; + string _SplitHeadsQ__mh_w_989_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_989_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_989_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_989_equation_0, values = (var_10219_cast_fp16, var_9971_cast_fp16))[name = string("_SplitHeadsQ__mh_w_989_cast_fp16")]; + string _SplitHeadsQ__mh_w_991_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_991_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_991_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_991_equation_0, values = (var_10219_cast_fp16, var_9978_cast_fp16))[name = string("_SplitHeadsQ__mh_w_991_cast_fp16")]; + string _SplitHeadsQ__mh_w_993_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_993_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_993_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_993_equation_0, values = (var_10223_cast_fp16, var_9985_cast_fp16))[name = string("_SplitHeadsQ__mh_w_993_cast_fp16")]; + string _SplitHeadsQ__mh_w_995_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_995_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_995_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_995_equation_0, values = (var_10223_cast_fp16, var_9992_cast_fp16))[name = string("_SplitHeadsQ__mh_w_995_cast_fp16")]; + string _SplitHeadsQ__mh_w_997_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_997_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_997_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_997_equation_0, values = (var_10223_cast_fp16, var_9999_cast_fp16))[name = string("_SplitHeadsQ__mh_w_997_cast_fp16")]; + string _SplitHeadsQ__mh_w_999_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_999_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_999_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_999_equation_0, values = (var_10223_cast_fp16, var_10006_cast_fp16))[name = string("_SplitHeadsQ__mh_w_999_cast_fp16")]; + string _SplitHeadsQ__mh_w_1001_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1001_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1001_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1001_equation_0, values = (var_10227_cast_fp16, var_10013_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1001_cast_fp16")]; + string _SplitHeadsQ__mh_w_1003_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1003_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1003_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1003_equation_0, values = (var_10227_cast_fp16, var_10020_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1003_cast_fp16")]; + string _SplitHeadsQ__mh_w_1005_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1005_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1005_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1005_equation_0, values = (var_10227_cast_fp16, var_10027_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1005_cast_fp16")]; + string _SplitHeadsQ__mh_w_1007_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1007_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1007_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1007_equation_0, values = (var_10227_cast_fp16, var_10034_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1007_cast_fp16")]; + string _SplitHeadsQ__mh_w_1009_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1009_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1009_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1009_equation_0, values = (var_10231_cast_fp16, var_10041_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1009_cast_fp16")]; + string _SplitHeadsQ__mh_w_1011_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1011_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1011_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1011_equation_0, values = (var_10231_cast_fp16, var_10048_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1011_cast_fp16")]; + string _SplitHeadsQ__mh_w_1013_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1013_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1013_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1013_equation_0, values = (var_10231_cast_fp16, var_10055_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1013_cast_fp16")]; + string _SplitHeadsQ__mh_w_1015_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1015_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1015_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1015_equation_0, values = (var_10231_cast_fp16, var_10062_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1015_cast_fp16")]; + string _SplitHeadsQ__mh_w_1017_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1017_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1017_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1017_equation_0, values = (var_10235_cast_fp16, var_10069_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1017_cast_fp16")]; + string _SplitHeadsQ__mh_w_1019_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1019_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1019_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1019_equation_0, values = (var_10235_cast_fp16, var_10076_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1019_cast_fp16")]; + string _SplitHeadsQ__mh_w_1021_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1021_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1021_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1021_equation_0, values = (var_10235_cast_fp16, var_10083_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1021_cast_fp16")]; + string _SplitHeadsQ__mh_w_1023_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1023_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1023_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1023_equation_0, values = (var_10235_cast_fp16, var_10090_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1023_cast_fp16")]; + string _SplitHeadsQ__mh_w_1025_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1025_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1025_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1025_equation_0, values = (var_10239_cast_fp16, var_10097_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1025_cast_fp16")]; + string _SplitHeadsQ__mh_w_1027_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1027_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1027_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1027_equation_0, values = (var_10239_cast_fp16, var_10104_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1027_cast_fp16")]; + string _SplitHeadsQ__mh_w_1029_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1029_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1029_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1029_equation_0, values = (var_10239_cast_fp16, var_10111_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1029_cast_fp16")]; + string _SplitHeadsQ__mh_w_1031_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1031_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1031_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1031_equation_0, values = (var_10239_cast_fp16, var_10118_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1031_cast_fp16")]; + string _SplitHeadsQ__mh_w_1033_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1033_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1033_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1033_equation_0, values = (var_10243_cast_fp16, var_10125_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1033_cast_fp16")]; + string _SplitHeadsQ__mh_w_1035_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1035_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1035_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1035_equation_0, values = (var_10243_cast_fp16, var_10132_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1035_cast_fp16")]; + string _SplitHeadsQ__mh_w_1037_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1037_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1037_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1037_equation_0, values = (var_10243_cast_fp16, var_10139_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1037_cast_fp16")]; + string _SplitHeadsQ__mh_w_1039_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1039_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1039_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1039_equation_0, values = (var_10243_cast_fp16, var_10146_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1039_cast_fp16")]; + string _SplitHeadsQ__mh_w_1041_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1041_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1041_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1041_equation_0, values = (var_10247_cast_fp16, var_10153_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1041_cast_fp16")]; + string _SplitHeadsQ__mh_w_1043_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1043_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1043_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1043_equation_0, values = (var_10247_cast_fp16, var_10160_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1043_cast_fp16")]; + string _SplitHeadsQ__mh_w_1045_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1045_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1045_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1045_equation_0, values = (var_10247_cast_fp16, var_10167_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1045_cast_fp16")]; + string _SplitHeadsQ__mh_w_1047_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1047_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1047_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1047_equation_0, values = (var_10247_cast_fp16, var_10174_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1047_cast_fp16")]; + string _SplitHeadsQ__mh_w_1049_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1049_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1049_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1049_equation_0, values = (var_10251_cast_fp16, var_10181_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1049_cast_fp16")]; + string _SplitHeadsQ__mh_w_1051_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1051_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1051_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1051_equation_0, values = (var_10251_cast_fp16, var_10188_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1051_cast_fp16")]; + string _SplitHeadsQ__mh_w_1053_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1053_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1053_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1053_equation_0, values = (var_10251_cast_fp16, var_10195_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1053_cast_fp16")]; + string _SplitHeadsQ__mh_w_1055_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1055_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1055_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1055_equation_0, values = (var_10251_cast_fp16, var_10202_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1055_cast_fp16")]; + fp16 var_10396_to_fp16 = const()[name = string("op_10396_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_961_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_961_cast_fp16, y = var_10396_to_fp16)[name = string("aw_chunk_961_cast_fp16")]; + fp16 var_10398_to_fp16 = const()[name = string("op_10398_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_963_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_963_cast_fp16, y = var_10398_to_fp16)[name = string("aw_chunk_963_cast_fp16")]; + fp16 var_10400_to_fp16 = const()[name = string("op_10400_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_965_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_965_cast_fp16, y = var_10400_to_fp16)[name = string("aw_chunk_965_cast_fp16")]; + fp16 var_10402_to_fp16 = const()[name = string("op_10402_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_967_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_967_cast_fp16, y = var_10402_to_fp16)[name = string("aw_chunk_967_cast_fp16")]; + fp16 var_10404_to_fp16 = const()[name = string("op_10404_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_969_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_969_cast_fp16, y = var_10404_to_fp16)[name = string("aw_chunk_969_cast_fp16")]; + fp16 var_10406_to_fp16 = const()[name = string("op_10406_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_971_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_971_cast_fp16, y = var_10406_to_fp16)[name = string("aw_chunk_971_cast_fp16")]; + fp16 var_10408_to_fp16 = const()[name = string("op_10408_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_973_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_973_cast_fp16, y = var_10408_to_fp16)[name = string("aw_chunk_973_cast_fp16")]; + fp16 var_10410_to_fp16 = const()[name = string("op_10410_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_975_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_975_cast_fp16, y = var_10410_to_fp16)[name = string("aw_chunk_975_cast_fp16")]; + fp16 var_10412_to_fp16 = const()[name = string("op_10412_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_977_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_977_cast_fp16, y = var_10412_to_fp16)[name = string("aw_chunk_977_cast_fp16")]; + fp16 var_10414_to_fp16 = const()[name = string("op_10414_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_979_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_979_cast_fp16, y = var_10414_to_fp16)[name = string("aw_chunk_979_cast_fp16")]; + fp16 var_10416_to_fp16 = const()[name = string("op_10416_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_981_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_981_cast_fp16, y = var_10416_to_fp16)[name = string("aw_chunk_981_cast_fp16")]; + fp16 var_10418_to_fp16 = const()[name = string("op_10418_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_983_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_983_cast_fp16, y = var_10418_to_fp16)[name = string("aw_chunk_983_cast_fp16")]; + fp16 var_10420_to_fp16 = const()[name = string("op_10420_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_985_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_985_cast_fp16, y = var_10420_to_fp16)[name = string("aw_chunk_985_cast_fp16")]; + fp16 var_10422_to_fp16 = const()[name = string("op_10422_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_987_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_987_cast_fp16, y = var_10422_to_fp16)[name = string("aw_chunk_987_cast_fp16")]; + fp16 var_10424_to_fp16 = const()[name = string("op_10424_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_989_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_989_cast_fp16, y = var_10424_to_fp16)[name = string("aw_chunk_989_cast_fp16")]; + fp16 var_10426_to_fp16 = const()[name = string("op_10426_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_991_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_991_cast_fp16, y = var_10426_to_fp16)[name = string("aw_chunk_991_cast_fp16")]; + fp16 var_10428_to_fp16 = const()[name = string("op_10428_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_993_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_993_cast_fp16, y = var_10428_to_fp16)[name = string("aw_chunk_993_cast_fp16")]; + fp16 var_10430_to_fp16 = const()[name = string("op_10430_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_995_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_995_cast_fp16, y = var_10430_to_fp16)[name = string("aw_chunk_995_cast_fp16")]; + fp16 var_10432_to_fp16 = const()[name = string("op_10432_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_997_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_997_cast_fp16, y = var_10432_to_fp16)[name = string("aw_chunk_997_cast_fp16")]; + fp16 var_10434_to_fp16 = const()[name = string("op_10434_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_999_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_999_cast_fp16, y = var_10434_to_fp16)[name = string("aw_chunk_999_cast_fp16")]; + fp16 var_10436_to_fp16 = const()[name = string("op_10436_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1001_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1001_cast_fp16, y = var_10436_to_fp16)[name = string("aw_chunk_1001_cast_fp16")]; + fp16 var_10438_to_fp16 = const()[name = string("op_10438_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1003_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1003_cast_fp16, y = var_10438_to_fp16)[name = string("aw_chunk_1003_cast_fp16")]; + fp16 var_10440_to_fp16 = const()[name = string("op_10440_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1005_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1005_cast_fp16, y = var_10440_to_fp16)[name = string("aw_chunk_1005_cast_fp16")]; + fp16 var_10442_to_fp16 = const()[name = string("op_10442_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1007_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1007_cast_fp16, y = var_10442_to_fp16)[name = string("aw_chunk_1007_cast_fp16")]; + fp16 var_10444_to_fp16 = const()[name = string("op_10444_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1009_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1009_cast_fp16, y = var_10444_to_fp16)[name = string("aw_chunk_1009_cast_fp16")]; + fp16 var_10446_to_fp16 = const()[name = string("op_10446_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1011_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1011_cast_fp16, y = var_10446_to_fp16)[name = string("aw_chunk_1011_cast_fp16")]; + fp16 var_10448_to_fp16 = const()[name = string("op_10448_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1013_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1013_cast_fp16, y = var_10448_to_fp16)[name = string("aw_chunk_1013_cast_fp16")]; + fp16 var_10450_to_fp16 = const()[name = string("op_10450_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1015_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1015_cast_fp16, y = var_10450_to_fp16)[name = string("aw_chunk_1015_cast_fp16")]; + fp16 var_10452_to_fp16 = const()[name = string("op_10452_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1017_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1017_cast_fp16, y = var_10452_to_fp16)[name = string("aw_chunk_1017_cast_fp16")]; + fp16 var_10454_to_fp16 = const()[name = string("op_10454_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1019_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1019_cast_fp16, y = var_10454_to_fp16)[name = string("aw_chunk_1019_cast_fp16")]; + fp16 var_10456_to_fp16 = const()[name = string("op_10456_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1021_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1021_cast_fp16, y = var_10456_to_fp16)[name = string("aw_chunk_1021_cast_fp16")]; + fp16 var_10458_to_fp16 = const()[name = string("op_10458_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1023_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1023_cast_fp16, y = var_10458_to_fp16)[name = string("aw_chunk_1023_cast_fp16")]; + fp16 var_10460_to_fp16 = const()[name = string("op_10460_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1025_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1025_cast_fp16, y = var_10460_to_fp16)[name = string("aw_chunk_1025_cast_fp16")]; + fp16 var_10462_to_fp16 = const()[name = string("op_10462_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1027_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1027_cast_fp16, y = var_10462_to_fp16)[name = string("aw_chunk_1027_cast_fp16")]; + fp16 var_10464_to_fp16 = const()[name = string("op_10464_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1029_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1029_cast_fp16, y = var_10464_to_fp16)[name = string("aw_chunk_1029_cast_fp16")]; + fp16 var_10466_to_fp16 = const()[name = string("op_10466_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1031_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1031_cast_fp16, y = var_10466_to_fp16)[name = string("aw_chunk_1031_cast_fp16")]; + fp16 var_10468_to_fp16 = const()[name = string("op_10468_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1033_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1033_cast_fp16, y = var_10468_to_fp16)[name = string("aw_chunk_1033_cast_fp16")]; + fp16 var_10470_to_fp16 = const()[name = string("op_10470_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1035_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1035_cast_fp16, y = var_10470_to_fp16)[name = string("aw_chunk_1035_cast_fp16")]; + fp16 var_10472_to_fp16 = const()[name = string("op_10472_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1037_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1037_cast_fp16, y = var_10472_to_fp16)[name = string("aw_chunk_1037_cast_fp16")]; + fp16 var_10474_to_fp16 = const()[name = string("op_10474_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1039_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1039_cast_fp16, y = var_10474_to_fp16)[name = string("aw_chunk_1039_cast_fp16")]; + fp16 var_10476_to_fp16 = const()[name = string("op_10476_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1041_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1041_cast_fp16, y = var_10476_to_fp16)[name = string("aw_chunk_1041_cast_fp16")]; + fp16 var_10478_to_fp16 = const()[name = string("op_10478_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1043_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1043_cast_fp16, y = var_10478_to_fp16)[name = string("aw_chunk_1043_cast_fp16")]; + fp16 var_10480_to_fp16 = const()[name = string("op_10480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1045_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1045_cast_fp16, y = var_10480_to_fp16)[name = string("aw_chunk_1045_cast_fp16")]; + fp16 var_10482_to_fp16 = const()[name = string("op_10482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1047_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1047_cast_fp16, y = var_10482_to_fp16)[name = string("aw_chunk_1047_cast_fp16")]; + fp16 var_10484_to_fp16 = const()[name = string("op_10484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1049_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1049_cast_fp16, y = var_10484_to_fp16)[name = string("aw_chunk_1049_cast_fp16")]; + fp16 var_10486_to_fp16 = const()[name = string("op_10486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1051_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1051_cast_fp16, y = var_10486_to_fp16)[name = string("aw_chunk_1051_cast_fp16")]; + fp16 var_10488_to_fp16 = const()[name = string("op_10488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1053_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1053_cast_fp16, y = var_10488_to_fp16)[name = string("aw_chunk_1053_cast_fp16")]; + fp16 var_10490_to_fp16 = const()[name = string("op_10490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1055_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1055_cast_fp16, y = var_10490_to_fp16)[name = string("aw_chunk_1055_cast_fp16")]; + tensor var_10492_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_961_cast_fp16)[name = string("op_10492_cast_fp16")]; + tensor var_10493_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_963_cast_fp16)[name = string("op_10493_cast_fp16")]; + tensor var_10494_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_965_cast_fp16)[name = string("op_10494_cast_fp16")]; + tensor var_10495_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_967_cast_fp16)[name = string("op_10495_cast_fp16")]; + tensor var_10496_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_969_cast_fp16)[name = string("op_10496_cast_fp16")]; + tensor var_10497_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_971_cast_fp16)[name = string("op_10497_cast_fp16")]; + tensor var_10498_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_973_cast_fp16)[name = string("op_10498_cast_fp16")]; + tensor var_10499_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_975_cast_fp16)[name = string("op_10499_cast_fp16")]; + tensor var_10500_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_977_cast_fp16)[name = string("op_10500_cast_fp16")]; + tensor var_10501_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_979_cast_fp16)[name = string("op_10501_cast_fp16")]; + tensor var_10502_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_981_cast_fp16)[name = string("op_10502_cast_fp16")]; + tensor var_10503_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_983_cast_fp16)[name = string("op_10503_cast_fp16")]; + tensor var_10504_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_985_cast_fp16)[name = string("op_10504_cast_fp16")]; + tensor var_10505_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_987_cast_fp16)[name = string("op_10505_cast_fp16")]; + tensor var_10506_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_989_cast_fp16)[name = string("op_10506_cast_fp16")]; + tensor var_10507_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_991_cast_fp16)[name = string("op_10507_cast_fp16")]; + tensor var_10508_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_993_cast_fp16)[name = string("op_10508_cast_fp16")]; + tensor var_10509_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_995_cast_fp16)[name = string("op_10509_cast_fp16")]; + tensor var_10510_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_997_cast_fp16)[name = string("op_10510_cast_fp16")]; + tensor var_10511_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_999_cast_fp16)[name = string("op_10511_cast_fp16")]; + tensor var_10512_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1001_cast_fp16)[name = string("op_10512_cast_fp16")]; + tensor var_10513_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1003_cast_fp16)[name = string("op_10513_cast_fp16")]; + tensor var_10514_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1005_cast_fp16)[name = string("op_10514_cast_fp16")]; + tensor var_10515_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1007_cast_fp16)[name = string("op_10515_cast_fp16")]; + tensor var_10516_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1009_cast_fp16)[name = string("op_10516_cast_fp16")]; + tensor var_10517_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1011_cast_fp16)[name = string("op_10517_cast_fp16")]; + tensor var_10518_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1013_cast_fp16)[name = string("op_10518_cast_fp16")]; + tensor var_10519_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1015_cast_fp16)[name = string("op_10519_cast_fp16")]; + tensor var_10520_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1017_cast_fp16)[name = string("op_10520_cast_fp16")]; + tensor var_10521_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1019_cast_fp16)[name = string("op_10521_cast_fp16")]; + tensor var_10522_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1021_cast_fp16)[name = string("op_10522_cast_fp16")]; + tensor var_10523_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1023_cast_fp16)[name = string("op_10523_cast_fp16")]; + tensor var_10524_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1025_cast_fp16)[name = string("op_10524_cast_fp16")]; + tensor var_10525_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1027_cast_fp16)[name = string("op_10525_cast_fp16")]; + tensor var_10526_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1029_cast_fp16)[name = string("op_10526_cast_fp16")]; + tensor var_10527_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1031_cast_fp16)[name = string("op_10527_cast_fp16")]; + tensor var_10528_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1033_cast_fp16)[name = string("op_10528_cast_fp16")]; + tensor var_10529_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1035_cast_fp16)[name = string("op_10529_cast_fp16")]; + tensor var_10530_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1037_cast_fp16)[name = string("op_10530_cast_fp16")]; + tensor var_10531_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1039_cast_fp16)[name = string("op_10531_cast_fp16")]; + tensor var_10532_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1041_cast_fp16)[name = string("op_10532_cast_fp16")]; + tensor var_10533_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1043_cast_fp16)[name = string("op_10533_cast_fp16")]; + tensor var_10534_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1045_cast_fp16)[name = string("op_10534_cast_fp16")]; + tensor var_10535_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1047_cast_fp16)[name = string("op_10535_cast_fp16")]; + tensor var_10536_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1049_cast_fp16)[name = string("op_10536_cast_fp16")]; + tensor var_10537_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1051_cast_fp16)[name = string("op_10537_cast_fp16")]; + tensor var_10538_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1053_cast_fp16)[name = string("op_10538_cast_fp16")]; + tensor var_10539_cast_fp16 = softmax(axis = var_9765, x = aw_chunk_1055_cast_fp16)[name = string("op_10539_cast_fp16")]; + string var_10541_equation_0 = const()[name = string("op_10541_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10541_cast_fp16 = einsum(equation = var_10541_equation_0, values = (var_10253_cast_fp16, var_10492_cast_fp16))[name = string("op_10541_cast_fp16")]; + string var_10543_equation_0 = const()[name = string("op_10543_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10543_cast_fp16 = einsum(equation = var_10543_equation_0, values = (var_10253_cast_fp16, var_10493_cast_fp16))[name = string("op_10543_cast_fp16")]; + string var_10545_equation_0 = const()[name = string("op_10545_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10545_cast_fp16 = einsum(equation = var_10545_equation_0, values = (var_10253_cast_fp16, var_10494_cast_fp16))[name = string("op_10545_cast_fp16")]; + string var_10547_equation_0 = const()[name = string("op_10547_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10547_cast_fp16 = einsum(equation = var_10547_equation_0, values = (var_10253_cast_fp16, var_10495_cast_fp16))[name = string("op_10547_cast_fp16")]; + string var_10549_equation_0 = const()[name = string("op_10549_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10549_cast_fp16 = einsum(equation = var_10549_equation_0, values = (var_10257_cast_fp16, var_10496_cast_fp16))[name = string("op_10549_cast_fp16")]; + string var_10551_equation_0 = const()[name = string("op_10551_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10551_cast_fp16 = einsum(equation = var_10551_equation_0, values = (var_10257_cast_fp16, var_10497_cast_fp16))[name = string("op_10551_cast_fp16")]; + string var_10553_equation_0 = const()[name = string("op_10553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10553_cast_fp16 = einsum(equation = var_10553_equation_0, values = (var_10257_cast_fp16, var_10498_cast_fp16))[name = string("op_10553_cast_fp16")]; + string var_10555_equation_0 = const()[name = string("op_10555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10555_cast_fp16 = einsum(equation = var_10555_equation_0, values = (var_10257_cast_fp16, var_10499_cast_fp16))[name = string("op_10555_cast_fp16")]; + string var_10557_equation_0 = const()[name = string("op_10557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10557_cast_fp16 = einsum(equation = var_10557_equation_0, values = (var_10261_cast_fp16, var_10500_cast_fp16))[name = string("op_10557_cast_fp16")]; + string var_10559_equation_0 = const()[name = string("op_10559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10559_cast_fp16 = einsum(equation = var_10559_equation_0, values = (var_10261_cast_fp16, var_10501_cast_fp16))[name = string("op_10559_cast_fp16")]; + string var_10561_equation_0 = const()[name = string("op_10561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10561_cast_fp16 = einsum(equation = var_10561_equation_0, values = (var_10261_cast_fp16, var_10502_cast_fp16))[name = string("op_10561_cast_fp16")]; + string var_10563_equation_0 = const()[name = string("op_10563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10563_cast_fp16 = einsum(equation = var_10563_equation_0, values = (var_10261_cast_fp16, var_10503_cast_fp16))[name = string("op_10563_cast_fp16")]; + string var_10565_equation_0 = const()[name = string("op_10565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10565_cast_fp16 = einsum(equation = var_10565_equation_0, values = (var_10265_cast_fp16, var_10504_cast_fp16))[name = string("op_10565_cast_fp16")]; + string var_10567_equation_0 = const()[name = string("op_10567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10567_cast_fp16 = einsum(equation = var_10567_equation_0, values = (var_10265_cast_fp16, var_10505_cast_fp16))[name = string("op_10567_cast_fp16")]; + string var_10569_equation_0 = const()[name = string("op_10569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10569_cast_fp16 = einsum(equation = var_10569_equation_0, values = (var_10265_cast_fp16, var_10506_cast_fp16))[name = string("op_10569_cast_fp16")]; + string var_10571_equation_0 = const()[name = string("op_10571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10571_cast_fp16 = einsum(equation = var_10571_equation_0, values = (var_10265_cast_fp16, var_10507_cast_fp16))[name = string("op_10571_cast_fp16")]; + string var_10573_equation_0 = const()[name = string("op_10573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10573_cast_fp16 = einsum(equation = var_10573_equation_0, values = (var_10269_cast_fp16, var_10508_cast_fp16))[name = string("op_10573_cast_fp16")]; + string var_10575_equation_0 = const()[name = string("op_10575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10575_cast_fp16 = einsum(equation = var_10575_equation_0, values = (var_10269_cast_fp16, var_10509_cast_fp16))[name = string("op_10575_cast_fp16")]; + string var_10577_equation_0 = const()[name = string("op_10577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10577_cast_fp16 = einsum(equation = var_10577_equation_0, values = (var_10269_cast_fp16, var_10510_cast_fp16))[name = string("op_10577_cast_fp16")]; + string var_10579_equation_0 = const()[name = string("op_10579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10579_cast_fp16 = einsum(equation = var_10579_equation_0, values = (var_10269_cast_fp16, var_10511_cast_fp16))[name = string("op_10579_cast_fp16")]; + string var_10581_equation_0 = const()[name = string("op_10581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10581_cast_fp16 = einsum(equation = var_10581_equation_0, values = (var_10273_cast_fp16, var_10512_cast_fp16))[name = string("op_10581_cast_fp16")]; + string var_10583_equation_0 = const()[name = string("op_10583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10583_cast_fp16 = einsum(equation = var_10583_equation_0, values = (var_10273_cast_fp16, var_10513_cast_fp16))[name = string("op_10583_cast_fp16")]; + string var_10585_equation_0 = const()[name = string("op_10585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10585_cast_fp16 = einsum(equation = var_10585_equation_0, values = (var_10273_cast_fp16, var_10514_cast_fp16))[name = string("op_10585_cast_fp16")]; + string var_10587_equation_0 = const()[name = string("op_10587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10587_cast_fp16 = einsum(equation = var_10587_equation_0, values = (var_10273_cast_fp16, var_10515_cast_fp16))[name = string("op_10587_cast_fp16")]; + string var_10589_equation_0 = const()[name = string("op_10589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10589_cast_fp16 = einsum(equation = var_10589_equation_0, values = (var_10277_cast_fp16, var_10516_cast_fp16))[name = string("op_10589_cast_fp16")]; + string var_10591_equation_0 = const()[name = string("op_10591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10591_cast_fp16 = einsum(equation = var_10591_equation_0, values = (var_10277_cast_fp16, var_10517_cast_fp16))[name = string("op_10591_cast_fp16")]; + string var_10593_equation_0 = const()[name = string("op_10593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10593_cast_fp16 = einsum(equation = var_10593_equation_0, values = (var_10277_cast_fp16, var_10518_cast_fp16))[name = string("op_10593_cast_fp16")]; + string var_10595_equation_0 = const()[name = string("op_10595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10595_cast_fp16 = einsum(equation = var_10595_equation_0, values = (var_10277_cast_fp16, var_10519_cast_fp16))[name = string("op_10595_cast_fp16")]; + string var_10597_equation_0 = const()[name = string("op_10597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10597_cast_fp16 = einsum(equation = var_10597_equation_0, values = (var_10281_cast_fp16, var_10520_cast_fp16))[name = string("op_10597_cast_fp16")]; + string var_10599_equation_0 = const()[name = string("op_10599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10599_cast_fp16 = einsum(equation = var_10599_equation_0, values = (var_10281_cast_fp16, var_10521_cast_fp16))[name = string("op_10599_cast_fp16")]; + string var_10601_equation_0 = const()[name = string("op_10601_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10601_cast_fp16 = einsum(equation = var_10601_equation_0, values = (var_10281_cast_fp16, var_10522_cast_fp16))[name = string("op_10601_cast_fp16")]; + string var_10603_equation_0 = const()[name = string("op_10603_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10603_cast_fp16 = einsum(equation = var_10603_equation_0, values = (var_10281_cast_fp16, var_10523_cast_fp16))[name = string("op_10603_cast_fp16")]; + string var_10605_equation_0 = const()[name = string("op_10605_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10605_cast_fp16 = einsum(equation = var_10605_equation_0, values = (var_10285_cast_fp16, var_10524_cast_fp16))[name = string("op_10605_cast_fp16")]; + string var_10607_equation_0 = const()[name = string("op_10607_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10607_cast_fp16 = einsum(equation = var_10607_equation_0, values = (var_10285_cast_fp16, var_10525_cast_fp16))[name = string("op_10607_cast_fp16")]; + string var_10609_equation_0 = const()[name = string("op_10609_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10609_cast_fp16 = einsum(equation = var_10609_equation_0, values = (var_10285_cast_fp16, var_10526_cast_fp16))[name = string("op_10609_cast_fp16")]; + string var_10611_equation_0 = const()[name = string("op_10611_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10611_cast_fp16 = einsum(equation = var_10611_equation_0, values = (var_10285_cast_fp16, var_10527_cast_fp16))[name = string("op_10611_cast_fp16")]; + string var_10613_equation_0 = const()[name = string("op_10613_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10613_cast_fp16 = einsum(equation = var_10613_equation_0, values = (var_10289_cast_fp16, var_10528_cast_fp16))[name = string("op_10613_cast_fp16")]; + string var_10615_equation_0 = const()[name = string("op_10615_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10615_cast_fp16 = einsum(equation = var_10615_equation_0, values = (var_10289_cast_fp16, var_10529_cast_fp16))[name = string("op_10615_cast_fp16")]; + string var_10617_equation_0 = const()[name = string("op_10617_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10617_cast_fp16 = einsum(equation = var_10617_equation_0, values = (var_10289_cast_fp16, var_10530_cast_fp16))[name = string("op_10617_cast_fp16")]; + string var_10619_equation_0 = const()[name = string("op_10619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10619_cast_fp16 = einsum(equation = var_10619_equation_0, values = (var_10289_cast_fp16, var_10531_cast_fp16))[name = string("op_10619_cast_fp16")]; + string var_10621_equation_0 = const()[name = string("op_10621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10621_cast_fp16 = einsum(equation = var_10621_equation_0, values = (var_10293_cast_fp16, var_10532_cast_fp16))[name = string("op_10621_cast_fp16")]; + string var_10623_equation_0 = const()[name = string("op_10623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10623_cast_fp16 = einsum(equation = var_10623_equation_0, values = (var_10293_cast_fp16, var_10533_cast_fp16))[name = string("op_10623_cast_fp16")]; + string var_10625_equation_0 = const()[name = string("op_10625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10625_cast_fp16 = einsum(equation = var_10625_equation_0, values = (var_10293_cast_fp16, var_10534_cast_fp16))[name = string("op_10625_cast_fp16")]; + string var_10627_equation_0 = const()[name = string("op_10627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10627_cast_fp16 = einsum(equation = var_10627_equation_0, values = (var_10293_cast_fp16, var_10535_cast_fp16))[name = string("op_10627_cast_fp16")]; + string var_10629_equation_0 = const()[name = string("op_10629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10629_cast_fp16 = einsum(equation = var_10629_equation_0, values = (var_10297_cast_fp16, var_10536_cast_fp16))[name = string("op_10629_cast_fp16")]; + string var_10631_equation_0 = const()[name = string("op_10631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10631_cast_fp16 = einsum(equation = var_10631_equation_0, values = (var_10297_cast_fp16, var_10537_cast_fp16))[name = string("op_10631_cast_fp16")]; + string var_10633_equation_0 = const()[name = string("op_10633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10633_cast_fp16 = einsum(equation = var_10633_equation_0, values = (var_10297_cast_fp16, var_10538_cast_fp16))[name = string("op_10633_cast_fp16")]; + string var_10635_equation_0 = const()[name = string("op_10635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_10635_cast_fp16 = einsum(equation = var_10635_equation_0, values = (var_10297_cast_fp16, var_10539_cast_fp16))[name = string("op_10635_cast_fp16")]; + bool var_10637_interleave_0 = const()[name = string("op_10637_interleave_0"), val = bool(false)]; + tensor var_10637_cast_fp16 = concat(axis = var_9748, interleave = var_10637_interleave_0, values = (var_10541_cast_fp16, var_10543_cast_fp16, var_10545_cast_fp16, var_10547_cast_fp16))[name = string("op_10637_cast_fp16")]; + bool var_10639_interleave_0 = const()[name = string("op_10639_interleave_0"), val = bool(false)]; + tensor var_10639_cast_fp16 = concat(axis = var_9748, interleave = var_10639_interleave_0, values = (var_10549_cast_fp16, var_10551_cast_fp16, var_10553_cast_fp16, var_10555_cast_fp16))[name = string("op_10639_cast_fp16")]; + bool var_10641_interleave_0 = const()[name = string("op_10641_interleave_0"), val = bool(false)]; + tensor var_10641_cast_fp16 = concat(axis = var_9748, interleave = var_10641_interleave_0, values = (var_10557_cast_fp16, var_10559_cast_fp16, var_10561_cast_fp16, var_10563_cast_fp16))[name = string("op_10641_cast_fp16")]; + bool var_10643_interleave_0 = const()[name = string("op_10643_interleave_0"), val = bool(false)]; + tensor var_10643_cast_fp16 = concat(axis = var_9748, interleave = var_10643_interleave_0, values = (var_10565_cast_fp16, var_10567_cast_fp16, var_10569_cast_fp16, var_10571_cast_fp16))[name = string("op_10643_cast_fp16")]; + bool var_10645_interleave_0 = const()[name = string("op_10645_interleave_0"), val = bool(false)]; + tensor var_10645_cast_fp16 = concat(axis = var_9748, interleave = var_10645_interleave_0, values = (var_10573_cast_fp16, var_10575_cast_fp16, var_10577_cast_fp16, var_10579_cast_fp16))[name = string("op_10645_cast_fp16")]; + bool var_10647_interleave_0 = const()[name = string("op_10647_interleave_0"), val = bool(false)]; + tensor var_10647_cast_fp16 = concat(axis = var_9748, interleave = var_10647_interleave_0, values = (var_10581_cast_fp16, var_10583_cast_fp16, var_10585_cast_fp16, var_10587_cast_fp16))[name = string("op_10647_cast_fp16")]; + bool var_10649_interleave_0 = const()[name = string("op_10649_interleave_0"), val = bool(false)]; + tensor var_10649_cast_fp16 = concat(axis = var_9748, interleave = var_10649_interleave_0, values = (var_10589_cast_fp16, var_10591_cast_fp16, var_10593_cast_fp16, var_10595_cast_fp16))[name = string("op_10649_cast_fp16")]; + bool var_10651_interleave_0 = const()[name = string("op_10651_interleave_0"), val = bool(false)]; + tensor var_10651_cast_fp16 = concat(axis = var_9748, interleave = var_10651_interleave_0, values = (var_10597_cast_fp16, var_10599_cast_fp16, var_10601_cast_fp16, var_10603_cast_fp16))[name = string("op_10651_cast_fp16")]; + bool var_10653_interleave_0 = const()[name = string("op_10653_interleave_0"), val = bool(false)]; + tensor var_10653_cast_fp16 = concat(axis = var_9748, interleave = var_10653_interleave_0, values = (var_10605_cast_fp16, var_10607_cast_fp16, var_10609_cast_fp16, var_10611_cast_fp16))[name = string("op_10653_cast_fp16")]; + bool var_10655_interleave_0 = const()[name = string("op_10655_interleave_0"), val = bool(false)]; + tensor var_10655_cast_fp16 = concat(axis = var_9748, interleave = var_10655_interleave_0, values = (var_10613_cast_fp16, var_10615_cast_fp16, var_10617_cast_fp16, var_10619_cast_fp16))[name = string("op_10655_cast_fp16")]; + bool var_10657_interleave_0 = const()[name = string("op_10657_interleave_0"), val = bool(false)]; + tensor var_10657_cast_fp16 = concat(axis = var_9748, interleave = var_10657_interleave_0, values = (var_10621_cast_fp16, var_10623_cast_fp16, var_10625_cast_fp16, var_10627_cast_fp16))[name = string("op_10657_cast_fp16")]; + bool var_10659_interleave_0 = const()[name = string("op_10659_interleave_0"), val = bool(false)]; + tensor var_10659_cast_fp16 = concat(axis = var_9748, interleave = var_10659_interleave_0, values = (var_10629_cast_fp16, var_10631_cast_fp16, var_10633_cast_fp16, var_10635_cast_fp16))[name = string("op_10659_cast_fp16")]; + bool input_81_interleave_0 = const()[name = string("input_81_interleave_0"), val = bool(false)]; + tensor input_81_cast_fp16 = concat(axis = var_9765, interleave = input_81_interleave_0, values = (var_10637_cast_fp16, var_10639_cast_fp16, var_10641_cast_fp16, var_10643_cast_fp16, var_10645_cast_fp16, var_10647_cast_fp16, var_10649_cast_fp16, var_10651_cast_fp16, var_10653_cast_fp16, var_10655_cast_fp16, var_10657_cast_fp16, var_10659_cast_fp16))[name = string("input_81_cast_fp16")]; + string obj_43_pad_type_0 = const()[name = string("obj_43_pad_type_0"), val = string("valid")]; + tensor obj_43_strides_0 = const()[name = string("obj_43_strides_0"), val = tensor([1, 1])]; + tensor obj_43_pad_0 = const()[name = string("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_43_dilations_0 = const()[name = string("obj_43_dilations_0"), val = tensor([1, 1])]; + int32 obj_43_groups_0 = const()[name = string("obj_43_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151515456)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152695168)))]; + tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_43_dilations_0, groups = obj_43_groups_0, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = obj_43_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_10678_to_fp16 = const()[name = string("op_10678_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_10678_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor input_83_gamma_0_to_fp16 = const()[name = string("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152696768)))]; + tensor input_83_beta_0_to_fp16 = const()[name = string("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152698368)))]; + fp16 input_83_epsilon_0_to_fp16 = const()[name = string("input_83_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("input_83_cast_fp16")]; + string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("valid")]; + tensor input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor([1, 1])]; + tensor input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor([1, 1])]; + int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152699968)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157418624)))]; + tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_mode_0 = const()[name = string("input_87_mode_0"), val = string("EXACT")]; + tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157424832)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162143488)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_45_cast_fp16")]; + int32 var_10707 = const()[name = string("op_10707"), val = int32(3)]; + int32 var_10724 = const()[name = string("op_10724"), val = int32(1)]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_10741_to_fp16 = const()[name = string("op_10741_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_10741_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = string("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162145088)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = string("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162146688)))]; + fp16 obj_45_epsilon_0_to_fp16 = const()[name = string("obj_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_45_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162148288)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163328000)))]; + tensor query_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163329600)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164509312)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165689024)))]; + tensor value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_10779_begin_0 = const()[name = string("op_10779_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10779_end_0 = const()[name = string("op_10779_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10779_end_mask_0 = const()[name = string("op_10779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10779_cast_fp16 = slice_by_index(begin = var_10779_begin_0, end = var_10779_end_0, end_mask = var_10779_end_mask_0, x = query_cast_fp16)[name = string("op_10779_cast_fp16")]; + tensor var_10783_begin_0 = const()[name = string("op_10783_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10783_end_0 = const()[name = string("op_10783_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10783_end_mask_0 = const()[name = string("op_10783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10783_cast_fp16 = slice_by_index(begin = var_10783_begin_0, end = var_10783_end_0, end_mask = var_10783_end_mask_0, x = query_cast_fp16)[name = string("op_10783_cast_fp16")]; + tensor var_10787_begin_0 = const()[name = string("op_10787_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10787_end_0 = const()[name = string("op_10787_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10787_end_mask_0 = const()[name = string("op_10787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10787_cast_fp16 = slice_by_index(begin = var_10787_begin_0, end = var_10787_end_0, end_mask = var_10787_end_mask_0, x = query_cast_fp16)[name = string("op_10787_cast_fp16")]; + tensor var_10791_begin_0 = const()[name = string("op_10791_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10791_end_0 = const()[name = string("op_10791_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10791_end_mask_0 = const()[name = string("op_10791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10791_cast_fp16 = slice_by_index(begin = var_10791_begin_0, end = var_10791_end_0, end_mask = var_10791_end_mask_0, x = query_cast_fp16)[name = string("op_10791_cast_fp16")]; + tensor var_10795_begin_0 = const()[name = string("op_10795_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10795_end_0 = const()[name = string("op_10795_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10795_end_mask_0 = const()[name = string("op_10795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10795_cast_fp16 = slice_by_index(begin = var_10795_begin_0, end = var_10795_end_0, end_mask = var_10795_end_mask_0, x = query_cast_fp16)[name = string("op_10795_cast_fp16")]; + tensor var_10799_begin_0 = const()[name = string("op_10799_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10799_end_0 = const()[name = string("op_10799_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10799_end_mask_0 = const()[name = string("op_10799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10799_cast_fp16 = slice_by_index(begin = var_10799_begin_0, end = var_10799_end_0, end_mask = var_10799_end_mask_0, x = query_cast_fp16)[name = string("op_10799_cast_fp16")]; + tensor var_10803_begin_0 = const()[name = string("op_10803_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10803_end_0 = const()[name = string("op_10803_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10803_end_mask_0 = const()[name = string("op_10803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10803_cast_fp16 = slice_by_index(begin = var_10803_begin_0, end = var_10803_end_0, end_mask = var_10803_end_mask_0, x = query_cast_fp16)[name = string("op_10803_cast_fp16")]; + tensor var_10807_begin_0 = const()[name = string("op_10807_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10807_end_0 = const()[name = string("op_10807_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10807_end_mask_0 = const()[name = string("op_10807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10807_cast_fp16 = slice_by_index(begin = var_10807_begin_0, end = var_10807_end_0, end_mask = var_10807_end_mask_0, x = query_cast_fp16)[name = string("op_10807_cast_fp16")]; + tensor var_10811_begin_0 = const()[name = string("op_10811_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10811_end_0 = const()[name = string("op_10811_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10811_end_mask_0 = const()[name = string("op_10811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10811_cast_fp16 = slice_by_index(begin = var_10811_begin_0, end = var_10811_end_0, end_mask = var_10811_end_mask_0, x = query_cast_fp16)[name = string("op_10811_cast_fp16")]; + tensor var_10815_begin_0 = const()[name = string("op_10815_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10815_end_0 = const()[name = string("op_10815_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10815_end_mask_0 = const()[name = string("op_10815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10815_cast_fp16 = slice_by_index(begin = var_10815_begin_0, end = var_10815_end_0, end_mask = var_10815_end_mask_0, x = query_cast_fp16)[name = string("op_10815_cast_fp16")]; + tensor var_10819_begin_0 = const()[name = string("op_10819_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10819_end_0 = const()[name = string("op_10819_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10819_end_mask_0 = const()[name = string("op_10819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10819_cast_fp16 = slice_by_index(begin = var_10819_begin_0, end = var_10819_end_0, end_mask = var_10819_end_mask_0, x = query_cast_fp16)[name = string("op_10819_cast_fp16")]; + tensor var_10823_begin_0 = const()[name = string("op_10823_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10823_end_0 = const()[name = string("op_10823_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10823_end_mask_0 = const()[name = string("op_10823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10823_cast_fp16 = slice_by_index(begin = var_10823_begin_0, end = var_10823_end_0, end_mask = var_10823_end_mask_0, x = query_cast_fp16)[name = string("op_10823_cast_fp16")]; + tensor var_10832_begin_0 = const()[name = string("op_10832_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10832_end_0 = const()[name = string("op_10832_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10832_end_mask_0 = const()[name = string("op_10832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10832_cast_fp16 = slice_by_index(begin = var_10832_begin_0, end = var_10832_end_0, end_mask = var_10832_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10832_cast_fp16")]; + tensor var_10839_begin_0 = const()[name = string("op_10839_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10839_end_0 = const()[name = string("op_10839_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10839_end_mask_0 = const()[name = string("op_10839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10839_cast_fp16 = slice_by_index(begin = var_10839_begin_0, end = var_10839_end_0, end_mask = var_10839_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10839_cast_fp16")]; + tensor var_10846_begin_0 = const()[name = string("op_10846_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10846_end_0 = const()[name = string("op_10846_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10846_end_mask_0 = const()[name = string("op_10846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10846_cast_fp16 = slice_by_index(begin = var_10846_begin_0, end = var_10846_end_0, end_mask = var_10846_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10846_cast_fp16")]; + tensor var_10853_begin_0 = const()[name = string("op_10853_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10853_end_0 = const()[name = string("op_10853_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10853_end_mask_0 = const()[name = string("op_10853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10853_cast_fp16 = slice_by_index(begin = var_10853_begin_0, end = var_10853_end_0, end_mask = var_10853_end_mask_0, x = var_10779_cast_fp16)[name = string("op_10853_cast_fp16")]; + tensor var_10860_begin_0 = const()[name = string("op_10860_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10860_end_0 = const()[name = string("op_10860_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10860_end_mask_0 = const()[name = string("op_10860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10860_cast_fp16 = slice_by_index(begin = var_10860_begin_0, end = var_10860_end_0, end_mask = var_10860_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10860_cast_fp16")]; + tensor var_10867_begin_0 = const()[name = string("op_10867_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10867_end_0 = const()[name = string("op_10867_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10867_end_mask_0 = const()[name = string("op_10867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10867_cast_fp16 = slice_by_index(begin = var_10867_begin_0, end = var_10867_end_0, end_mask = var_10867_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10867_cast_fp16")]; + tensor var_10874_begin_0 = const()[name = string("op_10874_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10874_end_0 = const()[name = string("op_10874_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10874_end_mask_0 = const()[name = string("op_10874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10874_cast_fp16 = slice_by_index(begin = var_10874_begin_0, end = var_10874_end_0, end_mask = var_10874_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10874_cast_fp16")]; + tensor var_10881_begin_0 = const()[name = string("op_10881_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10881_end_0 = const()[name = string("op_10881_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10881_end_mask_0 = const()[name = string("op_10881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10881_cast_fp16 = slice_by_index(begin = var_10881_begin_0, end = var_10881_end_0, end_mask = var_10881_end_mask_0, x = var_10783_cast_fp16)[name = string("op_10881_cast_fp16")]; + tensor var_10888_begin_0 = const()[name = string("op_10888_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10888_end_0 = const()[name = string("op_10888_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10888_end_mask_0 = const()[name = string("op_10888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10888_cast_fp16 = slice_by_index(begin = var_10888_begin_0, end = var_10888_end_0, end_mask = var_10888_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10888_cast_fp16")]; + tensor var_10895_begin_0 = const()[name = string("op_10895_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10895_end_0 = const()[name = string("op_10895_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10895_end_mask_0 = const()[name = string("op_10895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10895_cast_fp16 = slice_by_index(begin = var_10895_begin_0, end = var_10895_end_0, end_mask = var_10895_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10895_cast_fp16")]; + tensor var_10902_begin_0 = const()[name = string("op_10902_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10902_end_0 = const()[name = string("op_10902_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10902_end_mask_0 = const()[name = string("op_10902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10902_cast_fp16 = slice_by_index(begin = var_10902_begin_0, end = var_10902_end_0, end_mask = var_10902_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10902_cast_fp16")]; + tensor var_10909_begin_0 = const()[name = string("op_10909_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10909_end_0 = const()[name = string("op_10909_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10909_end_mask_0 = const()[name = string("op_10909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10909_cast_fp16 = slice_by_index(begin = var_10909_begin_0, end = var_10909_end_0, end_mask = var_10909_end_mask_0, x = var_10787_cast_fp16)[name = string("op_10909_cast_fp16")]; + tensor var_10916_begin_0 = const()[name = string("op_10916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10916_end_0 = const()[name = string("op_10916_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10916_end_mask_0 = const()[name = string("op_10916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10916_cast_fp16 = slice_by_index(begin = var_10916_begin_0, end = var_10916_end_0, end_mask = var_10916_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10916_cast_fp16")]; + tensor var_10923_begin_0 = const()[name = string("op_10923_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10923_end_0 = const()[name = string("op_10923_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10923_end_mask_0 = const()[name = string("op_10923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10923_cast_fp16 = slice_by_index(begin = var_10923_begin_0, end = var_10923_end_0, end_mask = var_10923_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10923_cast_fp16")]; + tensor var_10930_begin_0 = const()[name = string("op_10930_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10930_end_0 = const()[name = string("op_10930_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10930_end_mask_0 = const()[name = string("op_10930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10930_cast_fp16 = slice_by_index(begin = var_10930_begin_0, end = var_10930_end_0, end_mask = var_10930_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10930_cast_fp16")]; + tensor var_10937_begin_0 = const()[name = string("op_10937_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10937_end_0 = const()[name = string("op_10937_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10937_end_mask_0 = const()[name = string("op_10937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10937_cast_fp16 = slice_by_index(begin = var_10937_begin_0, end = var_10937_end_0, end_mask = var_10937_end_mask_0, x = var_10791_cast_fp16)[name = string("op_10937_cast_fp16")]; + tensor var_10944_begin_0 = const()[name = string("op_10944_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10944_end_0 = const()[name = string("op_10944_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10944_end_mask_0 = const()[name = string("op_10944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10944_cast_fp16 = slice_by_index(begin = var_10944_begin_0, end = var_10944_end_0, end_mask = var_10944_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10944_cast_fp16")]; + tensor var_10951_begin_0 = const()[name = string("op_10951_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10951_end_0 = const()[name = string("op_10951_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10951_end_mask_0 = const()[name = string("op_10951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10951_cast_fp16 = slice_by_index(begin = var_10951_begin_0, end = var_10951_end_0, end_mask = var_10951_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10951_cast_fp16")]; + tensor var_10958_begin_0 = const()[name = string("op_10958_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10958_end_0 = const()[name = string("op_10958_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10958_end_mask_0 = const()[name = string("op_10958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10958_cast_fp16 = slice_by_index(begin = var_10958_begin_0, end = var_10958_end_0, end_mask = var_10958_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10958_cast_fp16")]; + tensor var_10965_begin_0 = const()[name = string("op_10965_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10965_end_0 = const()[name = string("op_10965_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10965_end_mask_0 = const()[name = string("op_10965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10965_cast_fp16 = slice_by_index(begin = var_10965_begin_0, end = var_10965_end_0, end_mask = var_10965_end_mask_0, x = var_10795_cast_fp16)[name = string("op_10965_cast_fp16")]; + tensor var_10972_begin_0 = const()[name = string("op_10972_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10972_end_0 = const()[name = string("op_10972_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10972_end_mask_0 = const()[name = string("op_10972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10972_cast_fp16 = slice_by_index(begin = var_10972_begin_0, end = var_10972_end_0, end_mask = var_10972_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10972_cast_fp16")]; + tensor var_10979_begin_0 = const()[name = string("op_10979_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10979_end_0 = const()[name = string("op_10979_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10979_end_mask_0 = const()[name = string("op_10979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10979_cast_fp16 = slice_by_index(begin = var_10979_begin_0, end = var_10979_end_0, end_mask = var_10979_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10979_cast_fp16")]; + tensor var_10986_begin_0 = const()[name = string("op_10986_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10986_end_0 = const()[name = string("op_10986_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10986_end_mask_0 = const()[name = string("op_10986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10986_cast_fp16 = slice_by_index(begin = var_10986_begin_0, end = var_10986_end_0, end_mask = var_10986_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10986_cast_fp16")]; + tensor var_10993_begin_0 = const()[name = string("op_10993_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10993_end_0 = const()[name = string("op_10993_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10993_end_mask_0 = const()[name = string("op_10993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10993_cast_fp16 = slice_by_index(begin = var_10993_begin_0, end = var_10993_end_0, end_mask = var_10993_end_mask_0, x = var_10799_cast_fp16)[name = string("op_10993_cast_fp16")]; + tensor var_11000_begin_0 = const()[name = string("op_11000_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11000_end_0 = const()[name = string("op_11000_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11000_end_mask_0 = const()[name = string("op_11000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11000_cast_fp16 = slice_by_index(begin = var_11000_begin_0, end = var_11000_end_0, end_mask = var_11000_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11000_cast_fp16")]; + tensor var_11007_begin_0 = const()[name = string("op_11007_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11007_end_0 = const()[name = string("op_11007_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11007_end_mask_0 = const()[name = string("op_11007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11007_cast_fp16 = slice_by_index(begin = var_11007_begin_0, end = var_11007_end_0, end_mask = var_11007_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11007_cast_fp16")]; + tensor var_11014_begin_0 = const()[name = string("op_11014_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11014_end_0 = const()[name = string("op_11014_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11014_end_mask_0 = const()[name = string("op_11014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11014_cast_fp16 = slice_by_index(begin = var_11014_begin_0, end = var_11014_end_0, end_mask = var_11014_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11014_cast_fp16")]; + tensor var_11021_begin_0 = const()[name = string("op_11021_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11021_end_0 = const()[name = string("op_11021_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11021_end_mask_0 = const()[name = string("op_11021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11021_cast_fp16 = slice_by_index(begin = var_11021_begin_0, end = var_11021_end_0, end_mask = var_11021_end_mask_0, x = var_10803_cast_fp16)[name = string("op_11021_cast_fp16")]; + tensor var_11028_begin_0 = const()[name = string("op_11028_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11028_end_0 = const()[name = string("op_11028_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11028_end_mask_0 = const()[name = string("op_11028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11028_cast_fp16 = slice_by_index(begin = var_11028_begin_0, end = var_11028_end_0, end_mask = var_11028_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11028_cast_fp16")]; + tensor var_11035_begin_0 = const()[name = string("op_11035_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11035_end_0 = const()[name = string("op_11035_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11035_end_mask_0 = const()[name = string("op_11035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11035_cast_fp16 = slice_by_index(begin = var_11035_begin_0, end = var_11035_end_0, end_mask = var_11035_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11035_cast_fp16")]; + tensor var_11042_begin_0 = const()[name = string("op_11042_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11042_end_0 = const()[name = string("op_11042_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11042_end_mask_0 = const()[name = string("op_11042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11042_cast_fp16 = slice_by_index(begin = var_11042_begin_0, end = var_11042_end_0, end_mask = var_11042_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11042_cast_fp16")]; + tensor var_11049_begin_0 = const()[name = string("op_11049_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11049_end_0 = const()[name = string("op_11049_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11049_end_mask_0 = const()[name = string("op_11049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11049_cast_fp16 = slice_by_index(begin = var_11049_begin_0, end = var_11049_end_0, end_mask = var_11049_end_mask_0, x = var_10807_cast_fp16)[name = string("op_11049_cast_fp16")]; + tensor var_11056_begin_0 = const()[name = string("op_11056_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11056_end_0 = const()[name = string("op_11056_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11056_end_mask_0 = const()[name = string("op_11056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11056_cast_fp16 = slice_by_index(begin = var_11056_begin_0, end = var_11056_end_0, end_mask = var_11056_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11056_cast_fp16")]; + tensor var_11063_begin_0 = const()[name = string("op_11063_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11063_end_0 = const()[name = string("op_11063_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11063_end_mask_0 = const()[name = string("op_11063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11063_cast_fp16 = slice_by_index(begin = var_11063_begin_0, end = var_11063_end_0, end_mask = var_11063_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11063_cast_fp16")]; + tensor var_11070_begin_0 = const()[name = string("op_11070_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11070_end_0 = const()[name = string("op_11070_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11070_end_mask_0 = const()[name = string("op_11070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11070_cast_fp16 = slice_by_index(begin = var_11070_begin_0, end = var_11070_end_0, end_mask = var_11070_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11070_cast_fp16")]; + tensor var_11077_begin_0 = const()[name = string("op_11077_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11077_end_0 = const()[name = string("op_11077_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11077_end_mask_0 = const()[name = string("op_11077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11077_cast_fp16 = slice_by_index(begin = var_11077_begin_0, end = var_11077_end_0, end_mask = var_11077_end_mask_0, x = var_10811_cast_fp16)[name = string("op_11077_cast_fp16")]; + tensor var_11084_begin_0 = const()[name = string("op_11084_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11084_end_0 = const()[name = string("op_11084_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11084_end_mask_0 = const()[name = string("op_11084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11084_cast_fp16 = slice_by_index(begin = var_11084_begin_0, end = var_11084_end_0, end_mask = var_11084_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11084_cast_fp16")]; + tensor var_11091_begin_0 = const()[name = string("op_11091_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11091_end_0 = const()[name = string("op_11091_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11091_end_mask_0 = const()[name = string("op_11091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11091_cast_fp16 = slice_by_index(begin = var_11091_begin_0, end = var_11091_end_0, end_mask = var_11091_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11091_cast_fp16")]; + tensor var_11098_begin_0 = const()[name = string("op_11098_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11098_end_0 = const()[name = string("op_11098_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11098_end_mask_0 = const()[name = string("op_11098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11098_cast_fp16 = slice_by_index(begin = var_11098_begin_0, end = var_11098_end_0, end_mask = var_11098_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11098_cast_fp16")]; + tensor var_11105_begin_0 = const()[name = string("op_11105_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11105_end_0 = const()[name = string("op_11105_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11105_end_mask_0 = const()[name = string("op_11105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11105_cast_fp16 = slice_by_index(begin = var_11105_begin_0, end = var_11105_end_0, end_mask = var_11105_end_mask_0, x = var_10815_cast_fp16)[name = string("op_11105_cast_fp16")]; + tensor var_11112_begin_0 = const()[name = string("op_11112_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11112_end_0 = const()[name = string("op_11112_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11112_end_mask_0 = const()[name = string("op_11112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11112_cast_fp16 = slice_by_index(begin = var_11112_begin_0, end = var_11112_end_0, end_mask = var_11112_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11112_cast_fp16")]; + tensor var_11119_begin_0 = const()[name = string("op_11119_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11119_end_0 = const()[name = string("op_11119_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11119_end_mask_0 = const()[name = string("op_11119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11119_cast_fp16 = slice_by_index(begin = var_11119_begin_0, end = var_11119_end_0, end_mask = var_11119_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11119_cast_fp16")]; + tensor var_11126_begin_0 = const()[name = string("op_11126_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11126_end_0 = const()[name = string("op_11126_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11126_end_mask_0 = const()[name = string("op_11126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11126_cast_fp16 = slice_by_index(begin = var_11126_begin_0, end = var_11126_end_0, end_mask = var_11126_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11126_cast_fp16")]; + tensor var_11133_begin_0 = const()[name = string("op_11133_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11133_end_0 = const()[name = string("op_11133_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11133_end_mask_0 = const()[name = string("op_11133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11133_cast_fp16 = slice_by_index(begin = var_11133_begin_0, end = var_11133_end_0, end_mask = var_11133_end_mask_0, x = var_10819_cast_fp16)[name = string("op_11133_cast_fp16")]; + tensor var_11140_begin_0 = const()[name = string("op_11140_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11140_end_0 = const()[name = string("op_11140_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11140_end_mask_0 = const()[name = string("op_11140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11140_cast_fp16 = slice_by_index(begin = var_11140_begin_0, end = var_11140_end_0, end_mask = var_11140_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11140_cast_fp16")]; + tensor var_11147_begin_0 = const()[name = string("op_11147_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11147_end_0 = const()[name = string("op_11147_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11147_end_mask_0 = const()[name = string("op_11147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11147_cast_fp16 = slice_by_index(begin = var_11147_begin_0, end = var_11147_end_0, end_mask = var_11147_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11147_cast_fp16")]; + tensor var_11154_begin_0 = const()[name = string("op_11154_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11154_end_0 = const()[name = string("op_11154_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11154_end_mask_0 = const()[name = string("op_11154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11154_cast_fp16 = slice_by_index(begin = var_11154_begin_0, end = var_11154_end_0, end_mask = var_11154_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11154_cast_fp16")]; + tensor var_11161_begin_0 = const()[name = string("op_11161_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11161_end_0 = const()[name = string("op_11161_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11161_end_mask_0 = const()[name = string("op_11161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11161_cast_fp16 = slice_by_index(begin = var_11161_begin_0, end = var_11161_end_0, end_mask = var_11161_end_mask_0, x = var_10823_cast_fp16)[name = string("op_11161_cast_fp16")]; + tensor k_23_perm_0 = const()[name = string("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11166_begin_0 = const()[name = string("op_11166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11166_end_0 = const()[name = string("op_11166_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_11166_end_mask_0 = const()[name = string("op_11166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_23_cast_fp16 = transpose(perm = k_23_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_11166_cast_fp16 = slice_by_index(begin = var_11166_begin_0, end = var_11166_end_0, end_mask = var_11166_end_mask_0, x = k_23_cast_fp16)[name = string("op_11166_cast_fp16")]; + tensor var_11170_begin_0 = const()[name = string("op_11170_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11170_end_0 = const()[name = string("op_11170_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_11170_end_mask_0 = const()[name = string("op_11170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11170_cast_fp16 = slice_by_index(begin = var_11170_begin_0, end = var_11170_end_0, end_mask = var_11170_end_mask_0, x = k_23_cast_fp16)[name = string("op_11170_cast_fp16")]; + tensor var_11174_begin_0 = const()[name = string("op_11174_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11174_end_0 = const()[name = string("op_11174_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_11174_end_mask_0 = const()[name = string("op_11174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11174_cast_fp16 = slice_by_index(begin = var_11174_begin_0, end = var_11174_end_0, end_mask = var_11174_end_mask_0, x = k_23_cast_fp16)[name = string("op_11174_cast_fp16")]; + tensor var_11178_begin_0 = const()[name = string("op_11178_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11178_end_0 = const()[name = string("op_11178_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_11178_end_mask_0 = const()[name = string("op_11178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11178_cast_fp16 = slice_by_index(begin = var_11178_begin_0, end = var_11178_end_0, end_mask = var_11178_end_mask_0, x = k_23_cast_fp16)[name = string("op_11178_cast_fp16")]; + tensor var_11182_begin_0 = const()[name = string("op_11182_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11182_end_0 = const()[name = string("op_11182_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_11182_end_mask_0 = const()[name = string("op_11182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11182_cast_fp16 = slice_by_index(begin = var_11182_begin_0, end = var_11182_end_0, end_mask = var_11182_end_mask_0, x = k_23_cast_fp16)[name = string("op_11182_cast_fp16")]; + tensor var_11186_begin_0 = const()[name = string("op_11186_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11186_end_0 = const()[name = string("op_11186_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_11186_end_mask_0 = const()[name = string("op_11186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11186_cast_fp16 = slice_by_index(begin = var_11186_begin_0, end = var_11186_end_0, end_mask = var_11186_end_mask_0, x = k_23_cast_fp16)[name = string("op_11186_cast_fp16")]; + tensor var_11190_begin_0 = const()[name = string("op_11190_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11190_end_0 = const()[name = string("op_11190_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_11190_end_mask_0 = const()[name = string("op_11190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11190_cast_fp16 = slice_by_index(begin = var_11190_begin_0, end = var_11190_end_0, end_mask = var_11190_end_mask_0, x = k_23_cast_fp16)[name = string("op_11190_cast_fp16")]; + tensor var_11194_begin_0 = const()[name = string("op_11194_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11194_end_0 = const()[name = string("op_11194_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_11194_end_mask_0 = const()[name = string("op_11194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11194_cast_fp16 = slice_by_index(begin = var_11194_begin_0, end = var_11194_end_0, end_mask = var_11194_end_mask_0, x = k_23_cast_fp16)[name = string("op_11194_cast_fp16")]; + tensor var_11198_begin_0 = const()[name = string("op_11198_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11198_end_0 = const()[name = string("op_11198_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_11198_end_mask_0 = const()[name = string("op_11198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11198_cast_fp16 = slice_by_index(begin = var_11198_begin_0, end = var_11198_end_0, end_mask = var_11198_end_mask_0, x = k_23_cast_fp16)[name = string("op_11198_cast_fp16")]; + tensor var_11202_begin_0 = const()[name = string("op_11202_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11202_end_0 = const()[name = string("op_11202_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_11202_end_mask_0 = const()[name = string("op_11202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11202_cast_fp16 = slice_by_index(begin = var_11202_begin_0, end = var_11202_end_0, end_mask = var_11202_end_mask_0, x = k_23_cast_fp16)[name = string("op_11202_cast_fp16")]; + tensor var_11206_begin_0 = const()[name = string("op_11206_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_11206_end_0 = const()[name = string("op_11206_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_11206_end_mask_0 = const()[name = string("op_11206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11206_cast_fp16 = slice_by_index(begin = var_11206_begin_0, end = var_11206_end_0, end_mask = var_11206_end_mask_0, x = k_23_cast_fp16)[name = string("op_11206_cast_fp16")]; + tensor var_11210_begin_0 = const()[name = string("op_11210_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_11210_end_0 = const()[name = string("op_11210_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_11210_end_mask_0 = const()[name = string("op_11210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11210_cast_fp16 = slice_by_index(begin = var_11210_begin_0, end = var_11210_end_0, end_mask = var_11210_end_mask_0, x = k_23_cast_fp16)[name = string("op_11210_cast_fp16")]; + tensor var_11212_begin_0 = const()[name = string("op_11212_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11212_end_0 = const()[name = string("op_11212_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11212_end_mask_0 = const()[name = string("op_11212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11212_cast_fp16 = slice_by_index(begin = var_11212_begin_0, end = var_11212_end_0, end_mask = var_11212_end_mask_0, x = value_cast_fp16)[name = string("op_11212_cast_fp16")]; + tensor var_11216_begin_0 = const()[name = string("op_11216_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11216_end_0 = const()[name = string("op_11216_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_11216_end_mask_0 = const()[name = string("op_11216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11216_cast_fp16 = slice_by_index(begin = var_11216_begin_0, end = var_11216_end_0, end_mask = var_11216_end_mask_0, x = value_cast_fp16)[name = string("op_11216_cast_fp16")]; + tensor var_11220_begin_0 = const()[name = string("op_11220_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11220_end_0 = const()[name = string("op_11220_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_11220_end_mask_0 = const()[name = string("op_11220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11220_cast_fp16 = slice_by_index(begin = var_11220_begin_0, end = var_11220_end_0, end_mask = var_11220_end_mask_0, x = value_cast_fp16)[name = string("op_11220_cast_fp16")]; + tensor var_11224_begin_0 = const()[name = string("op_11224_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11224_end_0 = const()[name = string("op_11224_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_11224_end_mask_0 = const()[name = string("op_11224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11224_cast_fp16 = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = value_cast_fp16)[name = string("op_11224_cast_fp16")]; + tensor var_11228_begin_0 = const()[name = string("op_11228_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11228_end_0 = const()[name = string("op_11228_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_11228_end_mask_0 = const()[name = string("op_11228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11228_cast_fp16 = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = value_cast_fp16)[name = string("op_11228_cast_fp16")]; + tensor var_11232_begin_0 = const()[name = string("op_11232_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11232_end_0 = const()[name = string("op_11232_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_11232_end_mask_0 = const()[name = string("op_11232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11232_cast_fp16 = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = value_cast_fp16)[name = string("op_11232_cast_fp16")]; + tensor var_11236_begin_0 = const()[name = string("op_11236_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11236_end_0 = const()[name = string("op_11236_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_11236_end_mask_0 = const()[name = string("op_11236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11236_cast_fp16 = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = value_cast_fp16)[name = string("op_11236_cast_fp16")]; + tensor var_11240_begin_0 = const()[name = string("op_11240_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11240_end_0 = const()[name = string("op_11240_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_11240_end_mask_0 = const()[name = string("op_11240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11240_cast_fp16 = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = value_cast_fp16)[name = string("op_11240_cast_fp16")]; + tensor var_11244_begin_0 = const()[name = string("op_11244_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11244_end_0 = const()[name = string("op_11244_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_11244_end_mask_0 = const()[name = string("op_11244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11244_cast_fp16 = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = value_cast_fp16)[name = string("op_11244_cast_fp16")]; + tensor var_11248_begin_0 = const()[name = string("op_11248_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11248_end_0 = const()[name = string("op_11248_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_11248_end_mask_0 = const()[name = string("op_11248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11248_cast_fp16 = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = value_cast_fp16)[name = string("op_11248_cast_fp16")]; + tensor var_11252_begin_0 = const()[name = string("op_11252_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11252_end_0 = const()[name = string("op_11252_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_11252_end_mask_0 = const()[name = string("op_11252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11252_cast_fp16 = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = value_cast_fp16)[name = string("op_11252_cast_fp16")]; + tensor var_11256_begin_0 = const()[name = string("op_11256_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11256_end_0 = const()[name = string("op_11256_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_11256_end_mask_0 = const()[name = string("op_11256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11256_cast_fp16 = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = value_cast_fp16)[name = string("op_11256_cast_fp16")]; + string _SplitHeadsQ__mh_w_1057_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1057_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1057_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1057_equation_0, values = (var_11166_cast_fp16, var_10832_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1057_cast_fp16")]; + string _SplitHeadsQ__mh_w_1059_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1059_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1059_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1059_equation_0, values = (var_11166_cast_fp16, var_10839_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1059_cast_fp16")]; + string _SplitHeadsQ__mh_w_1061_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1061_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1061_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1061_equation_0, values = (var_11166_cast_fp16, var_10846_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1061_cast_fp16")]; + string _SplitHeadsQ__mh_w_1063_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1063_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1063_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1063_equation_0, values = (var_11166_cast_fp16, var_10853_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1063_cast_fp16")]; + string _SplitHeadsQ__mh_w_1065_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1065_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1065_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1065_equation_0, values = (var_11170_cast_fp16, var_10860_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1065_cast_fp16")]; + string _SplitHeadsQ__mh_w_1067_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1067_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1067_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1067_equation_0, values = (var_11170_cast_fp16, var_10867_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1067_cast_fp16")]; + string _SplitHeadsQ__mh_w_1069_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1069_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1069_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1069_equation_0, values = (var_11170_cast_fp16, var_10874_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1069_cast_fp16")]; + string _SplitHeadsQ__mh_w_1071_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1071_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1071_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1071_equation_0, values = (var_11170_cast_fp16, var_10881_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1071_cast_fp16")]; + string _SplitHeadsQ__mh_w_1073_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1073_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1073_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1073_equation_0, values = (var_11174_cast_fp16, var_10888_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1073_cast_fp16")]; + string _SplitHeadsQ__mh_w_1075_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1075_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1075_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1075_equation_0, values = (var_11174_cast_fp16, var_10895_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1075_cast_fp16")]; + string _SplitHeadsQ__mh_w_1077_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1077_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1077_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1077_equation_0, values = (var_11174_cast_fp16, var_10902_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1077_cast_fp16")]; + string _SplitHeadsQ__mh_w_1079_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1079_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1079_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1079_equation_0, values = (var_11174_cast_fp16, var_10909_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1079_cast_fp16")]; + string _SplitHeadsQ__mh_w_1081_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1081_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1081_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1081_equation_0, values = (var_11178_cast_fp16, var_10916_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1081_cast_fp16")]; + string _SplitHeadsQ__mh_w_1083_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1083_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1083_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1083_equation_0, values = (var_11178_cast_fp16, var_10923_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1083_cast_fp16")]; + string _SplitHeadsQ__mh_w_1085_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1085_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1085_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1085_equation_0, values = (var_11178_cast_fp16, var_10930_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1085_cast_fp16")]; + string _SplitHeadsQ__mh_w_1087_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1087_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1087_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1087_equation_0, values = (var_11178_cast_fp16, var_10937_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1087_cast_fp16")]; + string _SplitHeadsQ__mh_w_1089_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1089_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1089_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1089_equation_0, values = (var_11182_cast_fp16, var_10944_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1089_cast_fp16")]; + string _SplitHeadsQ__mh_w_1091_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1091_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1091_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1091_equation_0, values = (var_11182_cast_fp16, var_10951_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1091_cast_fp16")]; + string _SplitHeadsQ__mh_w_1093_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1093_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1093_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1093_equation_0, values = (var_11182_cast_fp16, var_10958_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1093_cast_fp16")]; + string _SplitHeadsQ__mh_w_1095_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1095_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1095_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1095_equation_0, values = (var_11182_cast_fp16, var_10965_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1095_cast_fp16")]; + string _SplitHeadsQ__mh_w_1097_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1097_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1097_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1097_equation_0, values = (var_11186_cast_fp16, var_10972_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1097_cast_fp16")]; + string _SplitHeadsQ__mh_w_1099_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1099_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1099_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1099_equation_0, values = (var_11186_cast_fp16, var_10979_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1099_cast_fp16")]; + string _SplitHeadsQ__mh_w_1101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1101_equation_0, values = (var_11186_cast_fp16, var_10986_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1101_cast_fp16")]; + string _SplitHeadsQ__mh_w_1103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1103_equation_0, values = (var_11186_cast_fp16, var_10993_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1103_cast_fp16")]; + string _SplitHeadsQ__mh_w_1105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1105_equation_0, values = (var_11190_cast_fp16, var_11000_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1105_cast_fp16")]; + string _SplitHeadsQ__mh_w_1107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1107_equation_0, values = (var_11190_cast_fp16, var_11007_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1107_cast_fp16")]; + string _SplitHeadsQ__mh_w_1109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1109_equation_0, values = (var_11190_cast_fp16, var_11014_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1109_cast_fp16")]; + string _SplitHeadsQ__mh_w_1111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1111_equation_0, values = (var_11190_cast_fp16, var_11021_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1111_cast_fp16")]; + string _SplitHeadsQ__mh_w_1113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1113_equation_0, values = (var_11194_cast_fp16, var_11028_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1113_cast_fp16")]; + string _SplitHeadsQ__mh_w_1115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1115_equation_0, values = (var_11194_cast_fp16, var_11035_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1115_cast_fp16")]; + string _SplitHeadsQ__mh_w_1117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1117_equation_0, values = (var_11194_cast_fp16, var_11042_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1117_cast_fp16")]; + string _SplitHeadsQ__mh_w_1119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1119_equation_0, values = (var_11194_cast_fp16, var_11049_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1119_cast_fp16")]; + string _SplitHeadsQ__mh_w_1121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1121_equation_0, values = (var_11198_cast_fp16, var_11056_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1121_cast_fp16")]; + string _SplitHeadsQ__mh_w_1123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1123_equation_0, values = (var_11198_cast_fp16, var_11063_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1123_cast_fp16")]; + string _SplitHeadsQ__mh_w_1125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1125_equation_0, values = (var_11198_cast_fp16, var_11070_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1125_cast_fp16")]; + string _SplitHeadsQ__mh_w_1127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1127_equation_0, values = (var_11198_cast_fp16, var_11077_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1127_cast_fp16")]; + string _SplitHeadsQ__mh_w_1129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1129_equation_0, values = (var_11202_cast_fp16, var_11084_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1129_cast_fp16")]; + string _SplitHeadsQ__mh_w_1131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1131_equation_0, values = (var_11202_cast_fp16, var_11091_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1131_cast_fp16")]; + string _SplitHeadsQ__mh_w_1133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1133_equation_0, values = (var_11202_cast_fp16, var_11098_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1133_cast_fp16")]; + string _SplitHeadsQ__mh_w_1135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1135_equation_0, values = (var_11202_cast_fp16, var_11105_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1135_cast_fp16")]; + string _SplitHeadsQ__mh_w_1137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1137_equation_0, values = (var_11206_cast_fp16, var_11112_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1137_cast_fp16")]; + string _SplitHeadsQ__mh_w_1139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1139_equation_0, values = (var_11206_cast_fp16, var_11119_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1139_cast_fp16")]; + string _SplitHeadsQ__mh_w_1141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1141_equation_0, values = (var_11206_cast_fp16, var_11126_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1141_cast_fp16")]; + string _SplitHeadsQ__mh_w_1143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1143_equation_0, values = (var_11206_cast_fp16, var_11133_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1143_cast_fp16")]; + string _SplitHeadsQ__mh_w_1145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1145_equation_0, values = (var_11210_cast_fp16, var_11140_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1145_cast_fp16")]; + string _SplitHeadsQ__mh_w_1147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1147_equation_0, values = (var_11210_cast_fp16, var_11147_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1147_cast_fp16")]; + string _SplitHeadsQ__mh_w_1149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1149_equation_0, values = (var_11210_cast_fp16, var_11154_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1149_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_11210_cast_fp16, var_11161_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_11355_to_fp16 = const()[name = string("op_11355_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1057_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1057_cast_fp16, y = var_11355_to_fp16)[name = string("aw_chunk_1057_cast_fp16")]; + fp16 var_11357_to_fp16 = const()[name = string("op_11357_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1059_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1059_cast_fp16, y = var_11357_to_fp16)[name = string("aw_chunk_1059_cast_fp16")]; + fp16 var_11359_to_fp16 = const()[name = string("op_11359_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1061_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1061_cast_fp16, y = var_11359_to_fp16)[name = string("aw_chunk_1061_cast_fp16")]; + fp16 var_11361_to_fp16 = const()[name = string("op_11361_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1063_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1063_cast_fp16, y = var_11361_to_fp16)[name = string("aw_chunk_1063_cast_fp16")]; + fp16 var_11363_to_fp16 = const()[name = string("op_11363_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1065_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1065_cast_fp16, y = var_11363_to_fp16)[name = string("aw_chunk_1065_cast_fp16")]; + fp16 var_11365_to_fp16 = const()[name = string("op_11365_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1067_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1067_cast_fp16, y = var_11365_to_fp16)[name = string("aw_chunk_1067_cast_fp16")]; + fp16 var_11367_to_fp16 = const()[name = string("op_11367_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1069_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1069_cast_fp16, y = var_11367_to_fp16)[name = string("aw_chunk_1069_cast_fp16")]; + fp16 var_11369_to_fp16 = const()[name = string("op_11369_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1071_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1071_cast_fp16, y = var_11369_to_fp16)[name = string("aw_chunk_1071_cast_fp16")]; + fp16 var_11371_to_fp16 = const()[name = string("op_11371_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1073_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1073_cast_fp16, y = var_11371_to_fp16)[name = string("aw_chunk_1073_cast_fp16")]; + fp16 var_11373_to_fp16 = const()[name = string("op_11373_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1075_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1075_cast_fp16, y = var_11373_to_fp16)[name = string("aw_chunk_1075_cast_fp16")]; + fp16 var_11375_to_fp16 = const()[name = string("op_11375_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1077_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1077_cast_fp16, y = var_11375_to_fp16)[name = string("aw_chunk_1077_cast_fp16")]; + fp16 var_11377_to_fp16 = const()[name = string("op_11377_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1079_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1079_cast_fp16, y = var_11377_to_fp16)[name = string("aw_chunk_1079_cast_fp16")]; + fp16 var_11379_to_fp16 = const()[name = string("op_11379_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1081_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1081_cast_fp16, y = var_11379_to_fp16)[name = string("aw_chunk_1081_cast_fp16")]; + fp16 var_11381_to_fp16 = const()[name = string("op_11381_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1083_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1083_cast_fp16, y = var_11381_to_fp16)[name = string("aw_chunk_1083_cast_fp16")]; + fp16 var_11383_to_fp16 = const()[name = string("op_11383_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1085_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1085_cast_fp16, y = var_11383_to_fp16)[name = string("aw_chunk_1085_cast_fp16")]; + fp16 var_11385_to_fp16 = const()[name = string("op_11385_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1087_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1087_cast_fp16, y = var_11385_to_fp16)[name = string("aw_chunk_1087_cast_fp16")]; + fp16 var_11387_to_fp16 = const()[name = string("op_11387_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1089_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1089_cast_fp16, y = var_11387_to_fp16)[name = string("aw_chunk_1089_cast_fp16")]; + fp16 var_11389_to_fp16 = const()[name = string("op_11389_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1091_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1091_cast_fp16, y = var_11389_to_fp16)[name = string("aw_chunk_1091_cast_fp16")]; + fp16 var_11391_to_fp16 = const()[name = string("op_11391_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1093_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1093_cast_fp16, y = var_11391_to_fp16)[name = string("aw_chunk_1093_cast_fp16")]; + fp16 var_11393_to_fp16 = const()[name = string("op_11393_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1095_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1095_cast_fp16, y = var_11393_to_fp16)[name = string("aw_chunk_1095_cast_fp16")]; + fp16 var_11395_to_fp16 = const()[name = string("op_11395_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1097_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1097_cast_fp16, y = var_11395_to_fp16)[name = string("aw_chunk_1097_cast_fp16")]; + fp16 var_11397_to_fp16 = const()[name = string("op_11397_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1099_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1099_cast_fp16, y = var_11397_to_fp16)[name = string("aw_chunk_1099_cast_fp16")]; + fp16 var_11399_to_fp16 = const()[name = string("op_11399_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1101_cast_fp16, y = var_11399_to_fp16)[name = string("aw_chunk_1101_cast_fp16")]; + fp16 var_11401_to_fp16 = const()[name = string("op_11401_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1103_cast_fp16, y = var_11401_to_fp16)[name = string("aw_chunk_1103_cast_fp16")]; + fp16 var_11403_to_fp16 = const()[name = string("op_11403_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1105_cast_fp16, y = var_11403_to_fp16)[name = string("aw_chunk_1105_cast_fp16")]; + fp16 var_11405_to_fp16 = const()[name = string("op_11405_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1107_cast_fp16, y = var_11405_to_fp16)[name = string("aw_chunk_1107_cast_fp16")]; + fp16 var_11407_to_fp16 = const()[name = string("op_11407_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1109_cast_fp16, y = var_11407_to_fp16)[name = string("aw_chunk_1109_cast_fp16")]; + fp16 var_11409_to_fp16 = const()[name = string("op_11409_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1111_cast_fp16, y = var_11409_to_fp16)[name = string("aw_chunk_1111_cast_fp16")]; + fp16 var_11411_to_fp16 = const()[name = string("op_11411_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1113_cast_fp16, y = var_11411_to_fp16)[name = string("aw_chunk_1113_cast_fp16")]; + fp16 var_11413_to_fp16 = const()[name = string("op_11413_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1115_cast_fp16, y = var_11413_to_fp16)[name = string("aw_chunk_1115_cast_fp16")]; + fp16 var_11415_to_fp16 = const()[name = string("op_11415_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1117_cast_fp16, y = var_11415_to_fp16)[name = string("aw_chunk_1117_cast_fp16")]; + fp16 var_11417_to_fp16 = const()[name = string("op_11417_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1119_cast_fp16, y = var_11417_to_fp16)[name = string("aw_chunk_1119_cast_fp16")]; + fp16 var_11419_to_fp16 = const()[name = string("op_11419_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1121_cast_fp16, y = var_11419_to_fp16)[name = string("aw_chunk_1121_cast_fp16")]; + fp16 var_11421_to_fp16 = const()[name = string("op_11421_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1123_cast_fp16, y = var_11421_to_fp16)[name = string("aw_chunk_1123_cast_fp16")]; + fp16 var_11423_to_fp16 = const()[name = string("op_11423_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1125_cast_fp16, y = var_11423_to_fp16)[name = string("aw_chunk_1125_cast_fp16")]; + fp16 var_11425_to_fp16 = const()[name = string("op_11425_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1127_cast_fp16, y = var_11425_to_fp16)[name = string("aw_chunk_1127_cast_fp16")]; + fp16 var_11427_to_fp16 = const()[name = string("op_11427_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1129_cast_fp16, y = var_11427_to_fp16)[name = string("aw_chunk_1129_cast_fp16")]; + fp16 var_11429_to_fp16 = const()[name = string("op_11429_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1131_cast_fp16, y = var_11429_to_fp16)[name = string("aw_chunk_1131_cast_fp16")]; + fp16 var_11431_to_fp16 = const()[name = string("op_11431_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1133_cast_fp16, y = var_11431_to_fp16)[name = string("aw_chunk_1133_cast_fp16")]; + fp16 var_11433_to_fp16 = const()[name = string("op_11433_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1135_cast_fp16, y = var_11433_to_fp16)[name = string("aw_chunk_1135_cast_fp16")]; + fp16 var_11435_to_fp16 = const()[name = string("op_11435_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1137_cast_fp16, y = var_11435_to_fp16)[name = string("aw_chunk_1137_cast_fp16")]; + fp16 var_11437_to_fp16 = const()[name = string("op_11437_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1139_cast_fp16, y = var_11437_to_fp16)[name = string("aw_chunk_1139_cast_fp16")]; + fp16 var_11439_to_fp16 = const()[name = string("op_11439_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1141_cast_fp16, y = var_11439_to_fp16)[name = string("aw_chunk_1141_cast_fp16")]; + fp16 var_11441_to_fp16 = const()[name = string("op_11441_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1143_cast_fp16, y = var_11441_to_fp16)[name = string("aw_chunk_1143_cast_fp16")]; + fp16 var_11443_to_fp16 = const()[name = string("op_11443_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1145_cast_fp16, y = var_11443_to_fp16)[name = string("aw_chunk_1145_cast_fp16")]; + fp16 var_11445_to_fp16 = const()[name = string("op_11445_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1147_cast_fp16, y = var_11445_to_fp16)[name = string("aw_chunk_1147_cast_fp16")]; + fp16 var_11447_to_fp16 = const()[name = string("op_11447_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1149_cast_fp16, y = var_11447_to_fp16)[name = string("aw_chunk_1149_cast_fp16")]; + fp16 var_11449_to_fp16 = const()[name = string("op_11449_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_11449_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_11451_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1057_cast_fp16)[name = string("op_11451_cast_fp16")]; + tensor var_11452_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1059_cast_fp16)[name = string("op_11452_cast_fp16")]; + tensor var_11453_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1061_cast_fp16)[name = string("op_11453_cast_fp16")]; + tensor var_11454_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1063_cast_fp16)[name = string("op_11454_cast_fp16")]; + tensor var_11455_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1065_cast_fp16)[name = string("op_11455_cast_fp16")]; + tensor var_11456_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1067_cast_fp16)[name = string("op_11456_cast_fp16")]; + tensor var_11457_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1069_cast_fp16)[name = string("op_11457_cast_fp16")]; + tensor var_11458_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1071_cast_fp16)[name = string("op_11458_cast_fp16")]; + tensor var_11459_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1073_cast_fp16)[name = string("op_11459_cast_fp16")]; + tensor var_11460_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1075_cast_fp16)[name = string("op_11460_cast_fp16")]; + tensor var_11461_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1077_cast_fp16)[name = string("op_11461_cast_fp16")]; + tensor var_11462_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1079_cast_fp16)[name = string("op_11462_cast_fp16")]; + tensor var_11463_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1081_cast_fp16)[name = string("op_11463_cast_fp16")]; + tensor var_11464_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1083_cast_fp16)[name = string("op_11464_cast_fp16")]; + tensor var_11465_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1085_cast_fp16)[name = string("op_11465_cast_fp16")]; + tensor var_11466_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1087_cast_fp16)[name = string("op_11466_cast_fp16")]; + tensor var_11467_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1089_cast_fp16)[name = string("op_11467_cast_fp16")]; + tensor var_11468_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1091_cast_fp16)[name = string("op_11468_cast_fp16")]; + tensor var_11469_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1093_cast_fp16)[name = string("op_11469_cast_fp16")]; + tensor var_11470_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1095_cast_fp16)[name = string("op_11470_cast_fp16")]; + tensor var_11471_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1097_cast_fp16)[name = string("op_11471_cast_fp16")]; + tensor var_11472_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1099_cast_fp16)[name = string("op_11472_cast_fp16")]; + tensor var_11473_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1101_cast_fp16)[name = string("op_11473_cast_fp16")]; + tensor var_11474_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1103_cast_fp16)[name = string("op_11474_cast_fp16")]; + tensor var_11475_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1105_cast_fp16)[name = string("op_11475_cast_fp16")]; + tensor var_11476_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1107_cast_fp16)[name = string("op_11476_cast_fp16")]; + tensor var_11477_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1109_cast_fp16)[name = string("op_11477_cast_fp16")]; + tensor var_11478_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1111_cast_fp16)[name = string("op_11478_cast_fp16")]; + tensor var_11479_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1113_cast_fp16)[name = string("op_11479_cast_fp16")]; + tensor var_11480_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1115_cast_fp16)[name = string("op_11480_cast_fp16")]; + tensor var_11481_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1117_cast_fp16)[name = string("op_11481_cast_fp16")]; + tensor var_11482_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1119_cast_fp16)[name = string("op_11482_cast_fp16")]; + tensor var_11483_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1121_cast_fp16)[name = string("op_11483_cast_fp16")]; + tensor var_11484_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1123_cast_fp16)[name = string("op_11484_cast_fp16")]; + tensor var_11485_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1125_cast_fp16)[name = string("op_11485_cast_fp16")]; + tensor var_11486_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1127_cast_fp16)[name = string("op_11486_cast_fp16")]; + tensor var_11487_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1129_cast_fp16)[name = string("op_11487_cast_fp16")]; + tensor var_11488_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1131_cast_fp16)[name = string("op_11488_cast_fp16")]; + tensor var_11489_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1133_cast_fp16)[name = string("op_11489_cast_fp16")]; + tensor var_11490_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1135_cast_fp16)[name = string("op_11490_cast_fp16")]; + tensor var_11491_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1137_cast_fp16)[name = string("op_11491_cast_fp16")]; + tensor var_11492_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1139_cast_fp16)[name = string("op_11492_cast_fp16")]; + tensor var_11493_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1141_cast_fp16)[name = string("op_11493_cast_fp16")]; + tensor var_11494_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1143_cast_fp16)[name = string("op_11494_cast_fp16")]; + tensor var_11495_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1145_cast_fp16)[name = string("op_11495_cast_fp16")]; + tensor var_11496_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1147_cast_fp16)[name = string("op_11496_cast_fp16")]; + tensor var_11497_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_1149_cast_fp16)[name = string("op_11497_cast_fp16")]; + tensor var_11498_cast_fp16 = softmax(axis = var_10724, x = aw_chunk_cast_fp16)[name = string("op_11498_cast_fp16")]; + string var_11500_equation_0 = const()[name = string("op_11500_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11500_cast_fp16 = einsum(equation = var_11500_equation_0, values = (var_11212_cast_fp16, var_11451_cast_fp16))[name = string("op_11500_cast_fp16")]; + string var_11502_equation_0 = const()[name = string("op_11502_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11502_cast_fp16 = einsum(equation = var_11502_equation_0, values = (var_11212_cast_fp16, var_11452_cast_fp16))[name = string("op_11502_cast_fp16")]; + string var_11504_equation_0 = const()[name = string("op_11504_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11504_cast_fp16 = einsum(equation = var_11504_equation_0, values = (var_11212_cast_fp16, var_11453_cast_fp16))[name = string("op_11504_cast_fp16")]; + string var_11506_equation_0 = const()[name = string("op_11506_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11506_cast_fp16 = einsum(equation = var_11506_equation_0, values = (var_11212_cast_fp16, var_11454_cast_fp16))[name = string("op_11506_cast_fp16")]; + string var_11508_equation_0 = const()[name = string("op_11508_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11508_cast_fp16 = einsum(equation = var_11508_equation_0, values = (var_11216_cast_fp16, var_11455_cast_fp16))[name = string("op_11508_cast_fp16")]; + string var_11510_equation_0 = const()[name = string("op_11510_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11510_cast_fp16 = einsum(equation = var_11510_equation_0, values = (var_11216_cast_fp16, var_11456_cast_fp16))[name = string("op_11510_cast_fp16")]; + string var_11512_equation_0 = const()[name = string("op_11512_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11512_cast_fp16 = einsum(equation = var_11512_equation_0, values = (var_11216_cast_fp16, var_11457_cast_fp16))[name = string("op_11512_cast_fp16")]; + string var_11514_equation_0 = const()[name = string("op_11514_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11514_cast_fp16 = einsum(equation = var_11514_equation_0, values = (var_11216_cast_fp16, var_11458_cast_fp16))[name = string("op_11514_cast_fp16")]; + string var_11516_equation_0 = const()[name = string("op_11516_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11516_cast_fp16 = einsum(equation = var_11516_equation_0, values = (var_11220_cast_fp16, var_11459_cast_fp16))[name = string("op_11516_cast_fp16")]; + string var_11518_equation_0 = const()[name = string("op_11518_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11518_cast_fp16 = einsum(equation = var_11518_equation_0, values = (var_11220_cast_fp16, var_11460_cast_fp16))[name = string("op_11518_cast_fp16")]; + string var_11520_equation_0 = const()[name = string("op_11520_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11520_cast_fp16 = einsum(equation = var_11520_equation_0, values = (var_11220_cast_fp16, var_11461_cast_fp16))[name = string("op_11520_cast_fp16")]; + string var_11522_equation_0 = const()[name = string("op_11522_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11522_cast_fp16 = einsum(equation = var_11522_equation_0, values = (var_11220_cast_fp16, var_11462_cast_fp16))[name = string("op_11522_cast_fp16")]; + string var_11524_equation_0 = const()[name = string("op_11524_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11524_cast_fp16 = einsum(equation = var_11524_equation_0, values = (var_11224_cast_fp16, var_11463_cast_fp16))[name = string("op_11524_cast_fp16")]; + string var_11526_equation_0 = const()[name = string("op_11526_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11526_cast_fp16 = einsum(equation = var_11526_equation_0, values = (var_11224_cast_fp16, var_11464_cast_fp16))[name = string("op_11526_cast_fp16")]; + string var_11528_equation_0 = const()[name = string("op_11528_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11528_cast_fp16 = einsum(equation = var_11528_equation_0, values = (var_11224_cast_fp16, var_11465_cast_fp16))[name = string("op_11528_cast_fp16")]; + string var_11530_equation_0 = const()[name = string("op_11530_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11530_cast_fp16 = einsum(equation = var_11530_equation_0, values = (var_11224_cast_fp16, var_11466_cast_fp16))[name = string("op_11530_cast_fp16")]; + string var_11532_equation_0 = const()[name = string("op_11532_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11532_cast_fp16 = einsum(equation = var_11532_equation_0, values = (var_11228_cast_fp16, var_11467_cast_fp16))[name = string("op_11532_cast_fp16")]; + string var_11534_equation_0 = const()[name = string("op_11534_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11534_cast_fp16 = einsum(equation = var_11534_equation_0, values = (var_11228_cast_fp16, var_11468_cast_fp16))[name = string("op_11534_cast_fp16")]; + string var_11536_equation_0 = const()[name = string("op_11536_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11536_cast_fp16 = einsum(equation = var_11536_equation_0, values = (var_11228_cast_fp16, var_11469_cast_fp16))[name = string("op_11536_cast_fp16")]; + string var_11538_equation_0 = const()[name = string("op_11538_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11538_cast_fp16 = einsum(equation = var_11538_equation_0, values = (var_11228_cast_fp16, var_11470_cast_fp16))[name = string("op_11538_cast_fp16")]; + string var_11540_equation_0 = const()[name = string("op_11540_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11540_cast_fp16 = einsum(equation = var_11540_equation_0, values = (var_11232_cast_fp16, var_11471_cast_fp16))[name = string("op_11540_cast_fp16")]; + string var_11542_equation_0 = const()[name = string("op_11542_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11542_cast_fp16 = einsum(equation = var_11542_equation_0, values = (var_11232_cast_fp16, var_11472_cast_fp16))[name = string("op_11542_cast_fp16")]; + string var_11544_equation_0 = const()[name = string("op_11544_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11544_cast_fp16 = einsum(equation = var_11544_equation_0, values = (var_11232_cast_fp16, var_11473_cast_fp16))[name = string("op_11544_cast_fp16")]; + string var_11546_equation_0 = const()[name = string("op_11546_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11546_cast_fp16 = einsum(equation = var_11546_equation_0, values = (var_11232_cast_fp16, var_11474_cast_fp16))[name = string("op_11546_cast_fp16")]; + string var_11548_equation_0 = const()[name = string("op_11548_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11548_cast_fp16 = einsum(equation = var_11548_equation_0, values = (var_11236_cast_fp16, var_11475_cast_fp16))[name = string("op_11548_cast_fp16")]; + string var_11550_equation_0 = const()[name = string("op_11550_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11550_cast_fp16 = einsum(equation = var_11550_equation_0, values = (var_11236_cast_fp16, var_11476_cast_fp16))[name = string("op_11550_cast_fp16")]; + string var_11552_equation_0 = const()[name = string("op_11552_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11552_cast_fp16 = einsum(equation = var_11552_equation_0, values = (var_11236_cast_fp16, var_11477_cast_fp16))[name = string("op_11552_cast_fp16")]; + string var_11554_equation_0 = const()[name = string("op_11554_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11554_cast_fp16 = einsum(equation = var_11554_equation_0, values = (var_11236_cast_fp16, var_11478_cast_fp16))[name = string("op_11554_cast_fp16")]; + string var_11556_equation_0 = const()[name = string("op_11556_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11556_cast_fp16 = einsum(equation = var_11556_equation_0, values = (var_11240_cast_fp16, var_11479_cast_fp16))[name = string("op_11556_cast_fp16")]; + string var_11558_equation_0 = const()[name = string("op_11558_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11558_cast_fp16 = einsum(equation = var_11558_equation_0, values = (var_11240_cast_fp16, var_11480_cast_fp16))[name = string("op_11558_cast_fp16")]; + string var_11560_equation_0 = const()[name = string("op_11560_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11560_cast_fp16 = einsum(equation = var_11560_equation_0, values = (var_11240_cast_fp16, var_11481_cast_fp16))[name = string("op_11560_cast_fp16")]; + string var_11562_equation_0 = const()[name = string("op_11562_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11562_cast_fp16 = einsum(equation = var_11562_equation_0, values = (var_11240_cast_fp16, var_11482_cast_fp16))[name = string("op_11562_cast_fp16")]; + string var_11564_equation_0 = const()[name = string("op_11564_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11564_cast_fp16 = einsum(equation = var_11564_equation_0, values = (var_11244_cast_fp16, var_11483_cast_fp16))[name = string("op_11564_cast_fp16")]; + string var_11566_equation_0 = const()[name = string("op_11566_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11566_cast_fp16 = einsum(equation = var_11566_equation_0, values = (var_11244_cast_fp16, var_11484_cast_fp16))[name = string("op_11566_cast_fp16")]; + string var_11568_equation_0 = const()[name = string("op_11568_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11568_cast_fp16 = einsum(equation = var_11568_equation_0, values = (var_11244_cast_fp16, var_11485_cast_fp16))[name = string("op_11568_cast_fp16")]; + string var_11570_equation_0 = const()[name = string("op_11570_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11570_cast_fp16 = einsum(equation = var_11570_equation_0, values = (var_11244_cast_fp16, var_11486_cast_fp16))[name = string("op_11570_cast_fp16")]; + string var_11572_equation_0 = const()[name = string("op_11572_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11572_cast_fp16 = einsum(equation = var_11572_equation_0, values = (var_11248_cast_fp16, var_11487_cast_fp16))[name = string("op_11572_cast_fp16")]; + string var_11574_equation_0 = const()[name = string("op_11574_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11574_cast_fp16 = einsum(equation = var_11574_equation_0, values = (var_11248_cast_fp16, var_11488_cast_fp16))[name = string("op_11574_cast_fp16")]; + string var_11576_equation_0 = const()[name = string("op_11576_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11576_cast_fp16 = einsum(equation = var_11576_equation_0, values = (var_11248_cast_fp16, var_11489_cast_fp16))[name = string("op_11576_cast_fp16")]; + string var_11578_equation_0 = const()[name = string("op_11578_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11578_cast_fp16 = einsum(equation = var_11578_equation_0, values = (var_11248_cast_fp16, var_11490_cast_fp16))[name = string("op_11578_cast_fp16")]; + string var_11580_equation_0 = const()[name = string("op_11580_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11580_cast_fp16 = einsum(equation = var_11580_equation_0, values = (var_11252_cast_fp16, var_11491_cast_fp16))[name = string("op_11580_cast_fp16")]; + string var_11582_equation_0 = const()[name = string("op_11582_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11582_cast_fp16 = einsum(equation = var_11582_equation_0, values = (var_11252_cast_fp16, var_11492_cast_fp16))[name = string("op_11582_cast_fp16")]; + string var_11584_equation_0 = const()[name = string("op_11584_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11584_cast_fp16 = einsum(equation = var_11584_equation_0, values = (var_11252_cast_fp16, var_11493_cast_fp16))[name = string("op_11584_cast_fp16")]; + string var_11586_equation_0 = const()[name = string("op_11586_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11586_cast_fp16 = einsum(equation = var_11586_equation_0, values = (var_11252_cast_fp16, var_11494_cast_fp16))[name = string("op_11586_cast_fp16")]; + string var_11588_equation_0 = const()[name = string("op_11588_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11588_cast_fp16 = einsum(equation = var_11588_equation_0, values = (var_11256_cast_fp16, var_11495_cast_fp16))[name = string("op_11588_cast_fp16")]; + string var_11590_equation_0 = const()[name = string("op_11590_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11590_cast_fp16 = einsum(equation = var_11590_equation_0, values = (var_11256_cast_fp16, var_11496_cast_fp16))[name = string("op_11590_cast_fp16")]; + string var_11592_equation_0 = const()[name = string("op_11592_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11592_cast_fp16 = einsum(equation = var_11592_equation_0, values = (var_11256_cast_fp16, var_11497_cast_fp16))[name = string("op_11592_cast_fp16")]; + string var_11594_equation_0 = const()[name = string("op_11594_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_11594_cast_fp16 = einsum(equation = var_11594_equation_0, values = (var_11256_cast_fp16, var_11498_cast_fp16))[name = string("op_11594_cast_fp16")]; + bool var_11596_interleave_0 = const()[name = string("op_11596_interleave_0"), val = bool(false)]; + tensor var_11596_cast_fp16 = concat(axis = var_10707, interleave = var_11596_interleave_0, values = (var_11500_cast_fp16, var_11502_cast_fp16, var_11504_cast_fp16, var_11506_cast_fp16))[name = string("op_11596_cast_fp16")]; + bool var_11598_interleave_0 = const()[name = string("op_11598_interleave_0"), val = bool(false)]; + tensor var_11598_cast_fp16 = concat(axis = var_10707, interleave = var_11598_interleave_0, values = (var_11508_cast_fp16, var_11510_cast_fp16, var_11512_cast_fp16, var_11514_cast_fp16))[name = string("op_11598_cast_fp16")]; + bool var_11600_interleave_0 = const()[name = string("op_11600_interleave_0"), val = bool(false)]; + tensor var_11600_cast_fp16 = concat(axis = var_10707, interleave = var_11600_interleave_0, values = (var_11516_cast_fp16, var_11518_cast_fp16, var_11520_cast_fp16, var_11522_cast_fp16))[name = string("op_11600_cast_fp16")]; + bool var_11602_interleave_0 = const()[name = string("op_11602_interleave_0"), val = bool(false)]; + tensor var_11602_cast_fp16 = concat(axis = var_10707, interleave = var_11602_interleave_0, values = (var_11524_cast_fp16, var_11526_cast_fp16, var_11528_cast_fp16, var_11530_cast_fp16))[name = string("op_11602_cast_fp16")]; + bool var_11604_interleave_0 = const()[name = string("op_11604_interleave_0"), val = bool(false)]; + tensor var_11604_cast_fp16 = concat(axis = var_10707, interleave = var_11604_interleave_0, values = (var_11532_cast_fp16, var_11534_cast_fp16, var_11536_cast_fp16, var_11538_cast_fp16))[name = string("op_11604_cast_fp16")]; + bool var_11606_interleave_0 = const()[name = string("op_11606_interleave_0"), val = bool(false)]; + tensor var_11606_cast_fp16 = concat(axis = var_10707, interleave = var_11606_interleave_0, values = (var_11540_cast_fp16, var_11542_cast_fp16, var_11544_cast_fp16, var_11546_cast_fp16))[name = string("op_11606_cast_fp16")]; + bool var_11608_interleave_0 = const()[name = string("op_11608_interleave_0"), val = bool(false)]; + tensor var_11608_cast_fp16 = concat(axis = var_10707, interleave = var_11608_interleave_0, values = (var_11548_cast_fp16, var_11550_cast_fp16, var_11552_cast_fp16, var_11554_cast_fp16))[name = string("op_11608_cast_fp16")]; + bool var_11610_interleave_0 = const()[name = string("op_11610_interleave_0"), val = bool(false)]; + tensor var_11610_cast_fp16 = concat(axis = var_10707, interleave = var_11610_interleave_0, values = (var_11556_cast_fp16, var_11558_cast_fp16, var_11560_cast_fp16, var_11562_cast_fp16))[name = string("op_11610_cast_fp16")]; + bool var_11612_interleave_0 = const()[name = string("op_11612_interleave_0"), val = bool(false)]; + tensor var_11612_cast_fp16 = concat(axis = var_10707, interleave = var_11612_interleave_0, values = (var_11564_cast_fp16, var_11566_cast_fp16, var_11568_cast_fp16, var_11570_cast_fp16))[name = string("op_11612_cast_fp16")]; + bool var_11614_interleave_0 = const()[name = string("op_11614_interleave_0"), val = bool(false)]; + tensor var_11614_cast_fp16 = concat(axis = var_10707, interleave = var_11614_interleave_0, values = (var_11572_cast_fp16, var_11574_cast_fp16, var_11576_cast_fp16, var_11578_cast_fp16))[name = string("op_11614_cast_fp16")]; + bool var_11616_interleave_0 = const()[name = string("op_11616_interleave_0"), val = bool(false)]; + tensor var_11616_cast_fp16 = concat(axis = var_10707, interleave = var_11616_interleave_0, values = (var_11580_cast_fp16, var_11582_cast_fp16, var_11584_cast_fp16, var_11586_cast_fp16))[name = string("op_11616_cast_fp16")]; + bool var_11618_interleave_0 = const()[name = string("op_11618_interleave_0"), val = bool(false)]; + tensor var_11618_cast_fp16 = concat(axis = var_10707, interleave = var_11618_interleave_0, values = (var_11588_cast_fp16, var_11590_cast_fp16, var_11592_cast_fp16, var_11594_cast_fp16))[name = string("op_11618_cast_fp16")]; + bool input_89_interleave_0 = const()[name = string("input_89_interleave_0"), val = bool(false)]; + tensor input_89_cast_fp16 = concat(axis = var_10724, interleave = input_89_interleave_0, values = (var_11596_cast_fp16, var_11598_cast_fp16, var_11600_cast_fp16, var_11602_cast_fp16, var_11604_cast_fp16, var_11606_cast_fp16, var_11608_cast_fp16, var_11610_cast_fp16, var_11612_cast_fp16, var_11614_cast_fp16, var_11616_cast_fp16, var_11618_cast_fp16))[name = string("input_89_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165690624)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166870336)))]; + tensor obj_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_11637_to_fp16 = const()[name = string("op_11637_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_11637_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166871936)))]; + tensor input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166873536)))]; + fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1, 1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1, 1])]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166875136)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171593792)))]; + tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + string input_95_mode_0 = const()[name = string("input_95_mode_0"), val = string("EXACT")]; + tensor input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171600000)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176318656)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_11675_to_fp16 = const()[name = string("op_11675_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_11675_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176320256)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176321856)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_11699_pad_type_0 = const()[name = string("op_11699_pad_type_0"), val = string("valid")]; + tensor var_11699_strides_0 = const()[name = string("op_11699_strides_0"), val = tensor([1, 1])]; + tensor var_11699_pad_0 = const()[name = string("op_11699_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11699_dilations_0 = const()[name = string("op_11699_dilations_0"), val = tensor([1, 1])]; + int32 var_11699_groups_0 = const()[name = string("op_11699_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176323456)))]; + tensor var_11699_cast_fp16 = conv(dilations = var_11699_dilations_0, groups = var_11699_groups_0, pad = var_11699_pad_0, pad_type = var_11699_pad_type_0, strides = var_11699_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11699_cast_fp16")]; + string var_11706_pad_type_0 = const()[name = string("op_11706_pad_type_0"), val = string("valid")]; + tensor var_11706_strides_0 = const()[name = string("op_11706_strides_0"), val = tensor([1, 1])]; + tensor var_11706_pad_0 = const()[name = string("op_11706_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11706_dilations_0 = const()[name = string("op_11706_dilations_0"), val = tensor([1, 1])]; + int32 var_11706_groups_0 = const()[name = string("op_11706_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177503168)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178682880)))]; + tensor var_11706_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_11706_dilations_0, groups = var_11706_groups_0, pad = var_11706_pad_0, pad_type = var_11706_pad_type_0, strides = var_11706_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11706_cast_fp16")]; + string var_11724_pad_type_0 = const()[name = string("op_11724_pad_type_0"), val = string("valid")]; + tensor var_11724_strides_0 = const()[name = string("op_11724_strides_0"), val = tensor([1, 1])]; + tensor var_11724_pad_0 = const()[name = string("op_11724_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11724_dilations_0 = const()[name = string("op_11724_dilations_0"), val = tensor([1, 1])]; + int32 var_11724_groups_0 = const()[name = string("op_11724_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178684480)))]; + tensor var_11724_cast_fp16 = conv(dilations = var_11724_dilations_0, groups = var_11724_groups_0, pad = var_11724_pad_0, pad_type = var_11724_pad_type_0, strides = var_11724_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11724_cast_fp16")]; + string var_11731_pad_type_0 = const()[name = string("op_11731_pad_type_0"), val = string("valid")]; + tensor var_11731_strides_0 = const()[name = string("op_11731_strides_0"), val = tensor([1, 1])]; + tensor var_11731_pad_0 = const()[name = string("op_11731_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11731_dilations_0 = const()[name = string("op_11731_dilations_0"), val = tensor([1, 1])]; + int32 var_11731_groups_0 = const()[name = string("op_11731_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179864192)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181043904)))]; + tensor var_11731_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_11731_dilations_0, groups = var_11731_groups_0, pad = var_11731_pad_0, pad_type = var_11731_pad_type_0, strides = var_11731_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11731_cast_fp16")]; + string var_11749_pad_type_0 = const()[name = string("op_11749_pad_type_0"), val = string("valid")]; + tensor var_11749_strides_0 = const()[name = string("op_11749_strides_0"), val = tensor([1, 1])]; + tensor var_11749_pad_0 = const()[name = string("op_11749_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11749_dilations_0 = const()[name = string("op_11749_dilations_0"), val = tensor([1, 1])]; + int32 var_11749_groups_0 = const()[name = string("op_11749_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181045504)))]; + tensor var_11749_cast_fp16 = conv(dilations = var_11749_dilations_0, groups = var_11749_groups_0, pad = var_11749_pad_0, pad_type = var_11749_pad_type_0, strides = var_11749_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11749_cast_fp16")]; + string var_11756_pad_type_0 = const()[name = string("op_11756_pad_type_0"), val = string("valid")]; + tensor var_11756_strides_0 = const()[name = string("op_11756_strides_0"), val = tensor([1, 1])]; + tensor var_11756_pad_0 = const()[name = string("op_11756_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11756_dilations_0 = const()[name = string("op_11756_dilations_0"), val = tensor([1, 1])]; + int32 var_11756_groups_0 = const()[name = string("op_11756_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182225216)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183404928)))]; + tensor var_11756_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_11756_dilations_0, groups = var_11756_groups_0, pad = var_11756_pad_0, pad_type = var_11756_pad_type_0, strides = var_11756_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11756_cast_fp16")]; + string var_11774_pad_type_0 = const()[name = string("op_11774_pad_type_0"), val = string("valid")]; + tensor var_11774_strides_0 = const()[name = string("op_11774_strides_0"), val = tensor([1, 1])]; + tensor var_11774_pad_0 = const()[name = string("op_11774_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11774_dilations_0 = const()[name = string("op_11774_dilations_0"), val = tensor([1, 1])]; + int32 var_11774_groups_0 = const()[name = string("op_11774_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183406528)))]; + tensor var_11774_cast_fp16 = conv(dilations = var_11774_dilations_0, groups = var_11774_groups_0, pad = var_11774_pad_0, pad_type = var_11774_pad_type_0, strides = var_11774_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11774_cast_fp16")]; + string var_11781_pad_type_0 = const()[name = string("op_11781_pad_type_0"), val = string("valid")]; + tensor var_11781_strides_0 = const()[name = string("op_11781_strides_0"), val = tensor([1, 1])]; + tensor var_11781_pad_0 = const()[name = string("op_11781_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11781_dilations_0 = const()[name = string("op_11781_dilations_0"), val = tensor([1, 1])]; + int32 var_11781_groups_0 = const()[name = string("op_11781_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184586240)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185765952)))]; + tensor var_11781_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = var_11781_dilations_0, groups = var_11781_groups_0, pad = var_11781_pad_0, pad_type = var_11781_pad_type_0, strides = var_11781_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11781_cast_fp16")]; + string var_11799_pad_type_0 = const()[name = string("op_11799_pad_type_0"), val = string("valid")]; + tensor var_11799_strides_0 = const()[name = string("op_11799_strides_0"), val = tensor([1, 1])]; + tensor var_11799_pad_0 = const()[name = string("op_11799_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11799_dilations_0 = const()[name = string("op_11799_dilations_0"), val = tensor([1, 1])]; + int32 var_11799_groups_0 = const()[name = string("op_11799_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185767552)))]; + tensor var_11799_cast_fp16 = conv(dilations = var_11799_dilations_0, groups = var_11799_groups_0, pad = var_11799_pad_0, pad_type = var_11799_pad_type_0, strides = var_11799_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11799_cast_fp16")]; + string var_11806_pad_type_0 = const()[name = string("op_11806_pad_type_0"), val = string("valid")]; + tensor var_11806_strides_0 = const()[name = string("op_11806_strides_0"), val = tensor([1, 1])]; + tensor var_11806_pad_0 = const()[name = string("op_11806_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11806_dilations_0 = const()[name = string("op_11806_dilations_0"), val = tensor([1, 1])]; + int32 var_11806_groups_0 = const()[name = string("op_11806_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186947264)))]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188126976)))]; + tensor var_11806_cast_fp16 = conv(bias = decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16, dilations = var_11806_dilations_0, groups = var_11806_groups_0, pad = var_11806_pad_0, pad_type = var_11806_pad_type_0, strides = var_11806_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11806_cast_fp16")]; + string var_11824_pad_type_0 = const()[name = string("op_11824_pad_type_0"), val = string("valid")]; + tensor var_11824_strides_0 = const()[name = string("op_11824_strides_0"), val = tensor([1, 1])]; + tensor var_11824_pad_0 = const()[name = string("op_11824_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11824_dilations_0 = const()[name = string("op_11824_dilations_0"), val = tensor([1, 1])]; + int32 var_11824_groups_0 = const()[name = string("op_11824_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188128576)))]; + tensor var_11824_cast_fp16 = conv(dilations = var_11824_dilations_0, groups = var_11824_groups_0, pad = var_11824_pad_0, pad_type = var_11824_pad_type_0, strides = var_11824_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11824_cast_fp16")]; + string var_11831_pad_type_0 = const()[name = string("op_11831_pad_type_0"), val = string("valid")]; + tensor var_11831_strides_0 = const()[name = string("op_11831_strides_0"), val = tensor([1, 1])]; + tensor var_11831_pad_0 = const()[name = string("op_11831_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11831_dilations_0 = const()[name = string("op_11831_dilations_0"), val = tensor([1, 1])]; + int32 var_11831_groups_0 = const()[name = string("op_11831_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189308288)))]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190488000)))]; + tensor var_11831_cast_fp16 = conv(bias = decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16, dilations = var_11831_dilations_0, groups = var_11831_groups_0, pad = var_11831_pad_0, pad_type = var_11831_pad_type_0, strides = var_11831_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11831_cast_fp16")]; + string var_11849_pad_type_0 = const()[name = string("op_11849_pad_type_0"), val = string("valid")]; + tensor var_11849_strides_0 = const()[name = string("op_11849_strides_0"), val = tensor([1, 1])]; + tensor var_11849_pad_0 = const()[name = string("op_11849_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11849_dilations_0 = const()[name = string("op_11849_dilations_0"), val = tensor([1, 1])]; + int32 var_11849_groups_0 = const()[name = string("op_11849_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190489600)))]; + tensor var_11849_cast_fp16 = conv(dilations = var_11849_dilations_0, groups = var_11849_groups_0, pad = var_11849_pad_0, pad_type = var_11849_pad_type_0, strides = var_11849_strides_0, weight = decoder_kv_cache_prep_6_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11849_cast_fp16")]; + string var_11856_pad_type_0 = const()[name = string("op_11856_pad_type_0"), val = string("valid")]; + tensor var_11856_strides_0 = const()[name = string("op_11856_strides_0"), val = tensor([1, 1])]; + tensor var_11856_pad_0 = const()[name = string("op_11856_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11856_dilations_0 = const()[name = string("op_11856_dilations_0"), val = tensor([1, 1])]; + int32 var_11856_groups_0 = const()[name = string("op_11856_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191669312)))]; + tensor decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192849024)))]; + tensor var_11856_cast_fp16 = conv(bias = decoder_kv_cache_prep_6_encoder_attn_v_proj_bias_to_fp16, dilations = var_11856_dilations_0, groups = var_11856_groups_0, pad = var_11856_pad_0, pad_type = var_11856_pad_type_0, strides = var_11856_strides_0, weight = decoder_kv_cache_prep_6_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11856_cast_fp16")]; + string var_11874_pad_type_0 = const()[name = string("op_11874_pad_type_0"), val = string("valid")]; + tensor var_11874_strides_0 = const()[name = string("op_11874_strides_0"), val = tensor([1, 1])]; + tensor var_11874_pad_0 = const()[name = string("op_11874_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11874_dilations_0 = const()[name = string("op_11874_dilations_0"), val = tensor([1, 1])]; + int32 var_11874_groups_0 = const()[name = string("op_11874_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192850624)))]; + tensor var_11874_cast_fp16 = conv(dilations = var_11874_dilations_0, groups = var_11874_groups_0, pad = var_11874_pad_0, pad_type = var_11874_pad_type_0, strides = var_11874_strides_0, weight = decoder_kv_cache_prep_7_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11874_cast_fp16")]; + string var_11881_pad_type_0 = const()[name = string("op_11881_pad_type_0"), val = string("valid")]; + tensor var_11881_strides_0 = const()[name = string("op_11881_strides_0"), val = tensor([1, 1])]; + tensor var_11881_pad_0 = const()[name = string("op_11881_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11881_dilations_0 = const()[name = string("op_11881_dilations_0"), val = tensor([1, 1])]; + int32 var_11881_groups_0 = const()[name = string("op_11881_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194030336)))]; + tensor decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195210048)))]; + tensor var_11881_cast_fp16 = conv(bias = decoder_kv_cache_prep_7_encoder_attn_v_proj_bias_to_fp16, dilations = var_11881_dilations_0, groups = var_11881_groups_0, pad = var_11881_pad_0, pad_type = var_11881_pad_type_0, strides = var_11881_strides_0, weight = decoder_kv_cache_prep_7_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11881_cast_fp16")]; + string var_11899_pad_type_0 = const()[name = string("op_11899_pad_type_0"), val = string("valid")]; + tensor var_11899_strides_0 = const()[name = string("op_11899_strides_0"), val = tensor([1, 1])]; + tensor var_11899_pad_0 = const()[name = string("op_11899_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11899_dilations_0 = const()[name = string("op_11899_dilations_0"), val = tensor([1, 1])]; + int32 var_11899_groups_0 = const()[name = string("op_11899_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195211648)))]; + tensor var_11899_cast_fp16 = conv(dilations = var_11899_dilations_0, groups = var_11899_groups_0, pad = var_11899_pad_0, pad_type = var_11899_pad_type_0, strides = var_11899_strides_0, weight = decoder_kv_cache_prep_8_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11899_cast_fp16")]; + string var_11906_pad_type_0 = const()[name = string("op_11906_pad_type_0"), val = string("valid")]; + tensor var_11906_strides_0 = const()[name = string("op_11906_strides_0"), val = tensor([1, 1])]; + tensor var_11906_pad_0 = const()[name = string("op_11906_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11906_dilations_0 = const()[name = string("op_11906_dilations_0"), val = tensor([1, 1])]; + int32 var_11906_groups_0 = const()[name = string("op_11906_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196391360)))]; + tensor decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197571072)))]; + tensor var_11906_cast_fp16 = conv(bias = decoder_kv_cache_prep_8_encoder_attn_v_proj_bias_to_fp16, dilations = var_11906_dilations_0, groups = var_11906_groups_0, pad = var_11906_pad_0, pad_type = var_11906_pad_type_0, strides = var_11906_strides_0, weight = decoder_kv_cache_prep_8_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11906_cast_fp16")]; + string var_11924_pad_type_0 = const()[name = string("op_11924_pad_type_0"), val = string("valid")]; + tensor var_11924_strides_0 = const()[name = string("op_11924_strides_0"), val = tensor([1, 1])]; + tensor var_11924_pad_0 = const()[name = string("op_11924_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11924_dilations_0 = const()[name = string("op_11924_dilations_0"), val = tensor([1, 1])]; + int32 var_11924_groups_0 = const()[name = string("op_11924_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197572672)))]; + tensor var_11924_cast_fp16 = conv(dilations = var_11924_dilations_0, groups = var_11924_groups_0, pad = var_11924_pad_0, pad_type = var_11924_pad_type_0, strides = var_11924_strides_0, weight = decoder_kv_cache_prep_9_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11924_cast_fp16")]; + string var_11931_pad_type_0 = const()[name = string("op_11931_pad_type_0"), val = string("valid")]; + tensor var_11931_strides_0 = const()[name = string("op_11931_strides_0"), val = tensor([1, 1])]; + tensor var_11931_pad_0 = const()[name = string("op_11931_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11931_dilations_0 = const()[name = string("op_11931_dilations_0"), val = tensor([1, 1])]; + int32 var_11931_groups_0 = const()[name = string("op_11931_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198752384)))]; + tensor decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199932096)))]; + tensor var_11931_cast_fp16 = conv(bias = decoder_kv_cache_prep_9_encoder_attn_v_proj_bias_to_fp16, dilations = var_11931_dilations_0, groups = var_11931_groups_0, pad = var_11931_pad_0, pad_type = var_11931_pad_type_0, strides = var_11931_strides_0, weight = decoder_kv_cache_prep_9_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11931_cast_fp16")]; + string var_11949_pad_type_0 = const()[name = string("op_11949_pad_type_0"), val = string("valid")]; + tensor var_11949_strides_0 = const()[name = string("op_11949_strides_0"), val = tensor([1, 1])]; + tensor var_11949_pad_0 = const()[name = string("op_11949_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11949_dilations_0 = const()[name = string("op_11949_dilations_0"), val = tensor([1, 1])]; + int32 var_11949_groups_0 = const()[name = string("op_11949_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199933696)))]; + tensor var_11949_cast_fp16 = conv(dilations = var_11949_dilations_0, groups = var_11949_groups_0, pad = var_11949_pad_0, pad_type = var_11949_pad_type_0, strides = var_11949_strides_0, weight = decoder_kv_cache_prep_10_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11949_cast_fp16")]; + string var_11956_pad_type_0 = const()[name = string("op_11956_pad_type_0"), val = string("valid")]; + tensor var_11956_strides_0 = const()[name = string("op_11956_strides_0"), val = tensor([1, 1])]; + tensor var_11956_pad_0 = const()[name = string("op_11956_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11956_dilations_0 = const()[name = string("op_11956_dilations_0"), val = tensor([1, 1])]; + int32 var_11956_groups_0 = const()[name = string("op_11956_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201113408)))]; + tensor decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202293120)))]; + tensor var_11956_cast_fp16 = conv(bias = decoder_kv_cache_prep_10_encoder_attn_v_proj_bias_to_fp16, dilations = var_11956_dilations_0, groups = var_11956_groups_0, pad = var_11956_pad_0, pad_type = var_11956_pad_type_0, strides = var_11956_strides_0, weight = decoder_kv_cache_prep_10_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_11956_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202294720)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_11_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203474432)))]; + tensor decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204654144)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_11_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_11_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_11986 = const()[name = string("op_11986"), val = int32(0)]; + bool input_99_interleave_0 = const()[name = string("input_99_interleave_0"), val = bool(false)]; + tensor input_99_cast_fp16 = concat(axis = var_11986, interleave = input_99_interleave_0, values = (var_11699_cast_fp16, var_11724_cast_fp16, var_11749_cast_fp16, var_11774_cast_fp16, var_11799_cast_fp16, var_11824_cast_fp16, var_11849_cast_fp16, var_11874_cast_fp16, var_11899_cast_fp16, var_11924_cast_fp16, var_11949_cast_fp16, k_cast_fp16))[name = string("input_99_cast_fp16")]; + int32 var_11989 = const()[name = string("op_11989"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_11989, interleave = input_interleave_0, values = (var_11706_cast_fp16, var_11731_cast_fp16, var_11756_cast_fp16, var_11781_cast_fp16, var_11806_cast_fp16, var_11831_cast_fp16, var_11856_cast_fp16, var_11881_cast_fp16, var_11906_cast_fp16, var_11931_cast_fp16, var_11956_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_11996_pad_0 = const()[name = string("op_11996_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_11996_mode_0 = const()[name = string("op_11996_mode_0"), val = string("constant")]; + fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_13_to_fp16, mode = var_11996_mode_0, pad = var_11996_pad_0, x = input_99_cast_fp16)[name = string("op_11996_cast_fp16")]; + tensor var_12002_pad_0 = const()[name = string("op_12002_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_12002_mode_0 = const()[name = string("op_12002_mode_0"), val = string("constant")]; + fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_14_to_fp16, mode = var_12002_mode_0, pad = var_12002_pad_0, x = input_cast_fp16)[name = string("op_12002_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-small/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-small/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f0ec4fe72f4cf66fa31b627ed210614989e648de --- /dev/null +++ b/openai_whisper-small/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75cddf7d8ff2459c67edeae85f8fac629898fac1ff58f90165deaba6662f7294 +size 204655744 diff --git a/openai_whisper-small/LICENSE_NOTICE.txt b/openai_whisper-small/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-small/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-small/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-small/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a11fbb2cd75b96eb2120a672afefa298c2ef857b --- /dev/null +++ b/openai_whisper-small/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc05e563ee0c556e3f578e04be5fb67b4e7520124403f2561f39102f0f2b33d +size 243 diff --git a/openai_whisper-small/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-small/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a3544b6644c1af93ca6bdabb67a1c51e80eaa552 --- /dev/null +++ b/openai_whisper-small/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ef11ea703011eab03287ec661f999e19c2c78cf67d531b5e6afa02e18f913d +size 328 diff --git a/openai_whisper-small/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-small/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1a60dd494a857817b67d87cd920baa6824e74b61 --- /dev/null +++ b/openai_whisper-small/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-small/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-small/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cf4cd446f68b88655d00a7df7063aa46937a9bdd --- /dev/null +++ b/openai_whisper-small/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-small/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-small/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f7d28ffac464e9e7086a526930f0059187de8d01 --- /dev/null +++ b/openai_whisper-small/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83 +size 354080 diff --git a/openai_whisper-small/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-small/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..00a1401bce1115c2d4f2dff72d79b394839d65f6 --- /dev/null +++ b/openai_whisper-small/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:918c3e65821b3232b3aadb00d842d1853743bb677e20c139361103a4511320e2 +size 243 diff --git a/openai_whisper-small/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-small/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..2e6f8ebf3fd61b7d0e989054d1e1d4161ccd8e95 --- /dev/null +++ b/openai_whisper-small/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3015f3429f673bc4a311b034f2dd12abd86a84e3653afa6f166654e6e6478aeb +size 754 diff --git a/openai_whisper-small/TextDecoder.mlmodelc/metadata.json b/openai_whisper-small/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..8cd673f9673ce7a9f34b5d15b9e41ce02efe4669 --- /dev/null +++ b/openai_whisper-small/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51865)", + "shortDescription" : "", + "shape" : "[1, 1, 51865]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 9216 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 9216, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 9216 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 9216, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 24, + "Ios18.mul" : 48, + "Ios18.matmul" : 48, + "Ios18.batchNorm" : 37, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 85, + "Ios18.layerNorm" : 37, + "Ios18.reshape" : 96, + "Ios18.linear" : 1, + "Ios18.conv" : 96, + "Ios18.gelu" : 12, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 44, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 448)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 12 × 768 × 1 × 448)", + "shortDescription" : "", + "shape" : "[12, 768, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-small/TextDecoder.mlmodelc/model.mil b/openai_whisper-small/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..de22e23a82efb3e66bc855bbc0683f62b815a12b --- /dev/null +++ b/openai_whisper-small/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,1819 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_42_axis_0 = const()[name = string("op_42_axis_0"), val = int32(0)]; + int32 var_42_batch_dims_0 = const()[name = string("op_42_batch_dims_0"), val = int32(0)]; + bool var_42_validate_indices_0 = const()[name = string("op_42_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_42_cast_fp16 = gather(axis = var_42_axis_0, batch_dims = var_42_batch_dims_0, indices = input_ids, validate_indices = var_42_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_42_cast_fp16")]; + int32 var_46_axis_0 = const()[name = string("op_46_axis_0"), val = int32(0)]; + int32 var_46_batch_dims_0 = const()[name = string("op_46_batch_dims_0"), val = int32(0)]; + bool var_46_validate_indices_0 = const()[name = string("op_46_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79664768)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_183")]; + tensor var_46_cast_fp16_cast_uint16 = gather(axis = var_46_axis_0, batch_dims = var_46_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_46_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_46_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_42_cast_fp16, y = var_46_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_60_axes_0 = const()[name = string("op_60_axes_0"), val = tensor([2])]; + tensor var_60_cast_fp16 = expand_dims(axes = var_60_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_60_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_60_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80352960)))]; + int32 var_65_axis_0 = const()[name = string("op_65_axis_0"), val = int32(0)]; + tensor var_65_cast_fp16_0, tensor var_65_cast_fp16_1, tensor var_65_cast_fp16_2, tensor var_65_cast_fp16_3, tensor var_65_cast_fp16_4, tensor var_65_cast_fp16_5, tensor var_65_cast_fp16_6, tensor var_65_cast_fp16_7, tensor var_65_cast_fp16_8, tensor var_65_cast_fp16_9, tensor var_65_cast_fp16_10, tensor var_65_cast_fp16_11 = split(axis = var_65_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_65_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80353088)))]; + int32 var_80_axis_0 = const()[name = string("op_80_axis_0"), val = int32(0)]; + tensor var_80_cast_fp16_0, tensor var_80_cast_fp16_1, tensor var_80_cast_fp16_2, tensor var_80_cast_fp16_3, tensor var_80_cast_fp16_4, tensor var_80_cast_fp16_5, tensor var_80_cast_fp16_6, tensor var_80_cast_fp16_7, tensor var_80_cast_fp16_8, tensor var_80_cast_fp16_9, tensor var_80_cast_fp16_10, tensor var_80_cast_fp16_11 = split(axis = var_80_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_80_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 768, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 768, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_108 = const()[name = string("op_108"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_133_to_fp16 = const()[name = string("op_133_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_133_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80353216)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80354816)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80356416)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80358016)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80359616)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81539328)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81540928)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82720640)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83900352)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_168_axes_0 = const()[name = string("op_168_axes_0"), val = tensor([1])]; + tensor var_168_cast_fp16 = expand_dims(axes = var_168_axes_0, x = kv_cache_update_mask)[name = string("op_168_cast_fp16")]; + tensor var_169_axes_0 = const()[name = string("op_169_axes_0"), val = tensor([2])]; + tensor var_169_cast_fp16 = expand_dims(axes = var_169_axes_0, x = var_168_cast_fp16)[name = string("op_169_cast_fp16")]; + tensor var_171_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_169_cast_fp16)[name = string("op_171_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_65_cast_fp16_0, y = var_171_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_173_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_169_cast_fp16)[name = string("op_173_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_80_cast_fp16_0, y = var_173_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_176 = const()[name = string("op_176"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_176, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_178_to_fp16 = const()[name = string("op_178_to_fp16"), val = fp16(0x1p-3)]; + tensor var_179_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_178_to_fp16)[name = string("op_179_cast_fp16")]; + tensor var_180 = const()[name = string("op_180"), val = tensor([1, 12, 64, -1])]; + tensor var_181_cast_fp16 = reshape(shape = var_180, x = key_1_cast_fp16)[name = string("op_181_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_179_cast_fp16, y = var_181_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_185_axes_0 = const()[name = string("op_185_axes_0"), val = tensor([1])]; + tensor var_185_cast_fp16 = expand_dims(axes = var_185_axes_0, x = decoder_key_padding_mask)[name = string("op_185_cast_fp16")]; + tensor var_186_axes_0 = const()[name = string("op_186_axes_0"), val = tensor([2])]; + tensor var_186_cast_fp16 = expand_dims(axes = var_186_axes_0, x = var_185_cast_fp16)[name = string("op_186_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_189_cast_fp16 = softmax(axis = var_108, x = mh_w_3_cast_fp16)[name = string("op_189_cast_fp16")]; + tensor var_190 = const()[name = string("op_190"), val = tensor([1, 12, 64, -1])]; + tensor var_191_cast_fp16 = reshape(shape = var_190, x = value_1_cast_fp16)[name = string("op_191_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_191_cast_fp16, y = var_189_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_194 = const()[name = string("op_194"), val = tensor([1, 768, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_194, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83901952)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85081664)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_216_to_fp16 = const()[name = string("op_216_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_216_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85083264)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85084864)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85086464)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86266176)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_236 = const()[name = string("op_236"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_236, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_238_to_fp16 = const()[name = string("op_238_to_fp16"), val = fp16(0x1p-3)]; + tensor var_239_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_238_to_fp16)[name = string("op_239_cast_fp16")]; + tensor var_240 = const()[name = string("op_240"), val = tensor([1, 12, 64, -1])]; + tensor var_241_cast_fp16 = reshape(shape = var_240, x = obj_17_cast_fp16)[name = string("op_241_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_239_cast_fp16, y = var_241_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_245_axes_0 = const()[name = string("op_245_axes_0"), val = tensor([1])]; + tensor var_245_cast_fp16 = expand_dims(axes = var_245_axes_0, x = read_state_4)[name = string("op_245_cast_fp16")]; + tensor var_246_axes_0 = const()[name = string("op_246_axes_0"), val = tensor([2])]; + tensor var_246_cast_fp16 = expand_dims(axes = var_246_axes_0, x = var_245_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_108, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_250 = const()[name = string("op_250"), val = tensor([1, 12, 64, -1])]; + tensor var_251_cast_fp16 = reshape(shape = var_250, x = obj_19_cast_fp16)[name = string("op_251_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_251_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_254 = const()[name = string("op_254"), val = tensor([1, 768, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_254, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86267776)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87447488)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_272_to_fp16 = const()[name = string("op_272_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_272_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87449088)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87450688)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87452288)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92170944)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92177152)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96895808)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 768, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 768, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_317 = const()[name = string("op_317"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_342_to_fp16 = const()[name = string("op_342_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_342_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96897408)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96899008)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96900608)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98080320)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98081920)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99261632)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100441344)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_380_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_169_cast_fp16)[name = string("op_380_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_65_cast_fp16_1, y = var_380_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_382_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_169_cast_fp16)[name = string("op_382_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_80_cast_fp16_1, y = var_382_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_385 = const()[name = string("op_385"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_385, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_387_to_fp16 = const()[name = string("op_387_to_fp16"), val = fp16(0x1p-3)]; + tensor var_388_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_387_to_fp16)[name = string("op_388_cast_fp16")]; + tensor var_389 = const()[name = string("op_389"), val = tensor([1, 12, 64, -1])]; + tensor var_390_cast_fp16 = reshape(shape = var_389, x = key_3_cast_fp16)[name = string("op_390_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_388_cast_fp16, y = var_390_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_398_cast_fp16 = softmax(axis = var_317, x = mh_w_11_cast_fp16)[name = string("op_398_cast_fp16")]; + tensor var_399 = const()[name = string("op_399"), val = tensor([1, 12, 64, -1])]; + tensor var_400_cast_fp16 = reshape(shape = var_399, x = value_3_cast_fp16)[name = string("op_400_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_400_cast_fp16, y = var_398_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_403 = const()[name = string("op_403"), val = tensor([1, 768, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_403, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100442944)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101622656)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_425_to_fp16 = const()[name = string("op_425_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_425_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101624256)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101625856)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101627456)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102807168)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_445 = const()[name = string("op_445"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_445, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_447_to_fp16 = const()[name = string("op_447_to_fp16"), val = fp16(0x1p-3)]; + tensor var_448_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_447_to_fp16)[name = string("op_448_cast_fp16")]; + tensor var_449 = const()[name = string("op_449"), val = tensor([1, 12, 64, -1])]; + tensor var_450_cast_fp16 = reshape(shape = var_449, x = obj_35_cast_fp16)[name = string("op_450_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_448_cast_fp16, y = var_450_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_317, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_459 = const()[name = string("op_459"), val = tensor([1, 12, 64, -1])]; + tensor var_460_cast_fp16 = reshape(shape = var_459, x = obj_37_cast_fp16)[name = string("op_460_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_460_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_463 = const()[name = string("op_463"), val = tensor([1, 768, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_463, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102808768)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103988480)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_481_to_fp16 = const()[name = string("op_481_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_481_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103990080)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103991680)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103993280)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108711936)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108718144)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113436800)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 768, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 768, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_526 = const()[name = string("op_526"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_551_to_fp16 = const()[name = string("op_551_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_551_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113438400)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113440000)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113441600)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114621312)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114622912)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115802624)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116982336)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_589_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_169_cast_fp16)[name = string("op_589_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_65_cast_fp16_2, y = var_589_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_591_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_169_cast_fp16)[name = string("op_591_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_80_cast_fp16_2, y = var_591_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_594 = const()[name = string("op_594"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_594, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; + tensor var_597_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_596_to_fp16)[name = string("op_597_cast_fp16")]; + tensor var_598 = const()[name = string("op_598"), val = tensor([1, 12, 64, -1])]; + tensor var_599_cast_fp16 = reshape(shape = var_598, x = key_5_cast_fp16)[name = string("op_599_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_597_cast_fp16, y = var_599_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_607_cast_fp16 = softmax(axis = var_526, x = mh_w_19_cast_fp16)[name = string("op_607_cast_fp16")]; + tensor var_608 = const()[name = string("op_608"), val = tensor([1, 12, 64, -1])]; + tensor var_609_cast_fp16 = reshape(shape = var_608, x = value_5_cast_fp16)[name = string("op_609_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_609_cast_fp16, y = var_607_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_612 = const()[name = string("op_612"), val = tensor([1, 768, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_612, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116983936)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118163648)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_634_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118165248)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118166848)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118168448)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119348160)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_654 = const()[name = string("op_654"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_654, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_656_to_fp16 = const()[name = string("op_656_to_fp16"), val = fp16(0x1p-3)]; + tensor var_657_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_656_to_fp16)[name = string("op_657_cast_fp16")]; + tensor var_658 = const()[name = string("op_658"), val = tensor([1, 12, 64, -1])]; + tensor var_659_cast_fp16 = reshape(shape = var_658, x = obj_53_cast_fp16)[name = string("op_659_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_657_cast_fp16, y = var_659_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_526, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_668 = const()[name = string("op_668"), val = tensor([1, 12, 64, -1])]; + tensor var_669_cast_fp16 = reshape(shape = var_668, x = obj_55_cast_fp16)[name = string("op_669_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_669_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_672 = const()[name = string("op_672"), val = tensor([1, 768, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_672, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119349760)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120529472)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_690_to_fp16 = const()[name = string("op_690_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_690_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120531072)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120532672)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120534272)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125252928)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125259136)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129977792)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 768, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 768, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_735 = const()[name = string("op_735"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_760_to_fp16 = const()[name = string("op_760_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_760_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129979392)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129980992)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129982592)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131162304)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_7_pad_type_0 = const()[name = string("current_key_7_pad_type_0"), val = string("valid")]; + tensor current_key_7_strides_0 = const()[name = string("current_key_7_strides_0"), val = tensor([1, 1])]; + tensor current_key_7_pad_0 = const()[name = string("current_key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_7_dilations_0 = const()[name = string("current_key_7_dilations_0"), val = tensor([1, 1])]; + int32 current_key_7_groups_0 = const()[name = string("current_key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131163904)))]; + tensor current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_7_cast_fp16")]; + string current_value_7_pad_type_0 = const()[name = string("current_value_7_pad_type_0"), val = string("valid")]; + tensor current_value_7_strides_0 = const()[name = string("current_value_7_strides_0"), val = tensor([1, 1])]; + tensor current_value_7_pad_0 = const()[name = string("current_value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_7_dilations_0 = const()[name = string("current_value_7_dilations_0"), val = tensor([1, 1])]; + int32 current_value_7_groups_0 = const()[name = string("current_value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132343616)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133523328)))]; + tensor current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_7_cast_fp16")]; + tensor var_798_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_169_cast_fp16)[name = string("op_798_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = var_65_cast_fp16_3, y = var_798_cast_fp16)[name = string("key_7_cast_fp16")]; + tensor var_800_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_169_cast_fp16)[name = string("op_800_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = var_80_cast_fp16_3, y = var_800_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_803 = const()[name = string("op_803"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_803, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_805_to_fp16 = const()[name = string("op_805_to_fp16"), val = fp16(0x1p-3)]; + tensor var_806_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_805_to_fp16)[name = string("op_806_cast_fp16")]; + tensor var_807 = const()[name = string("op_807"), val = tensor([1, 12, 64, -1])]; + tensor var_808_cast_fp16 = reshape(shape = var_807, x = key_7_cast_fp16)[name = string("op_808_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_806_cast_fp16, y = var_808_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_816_cast_fp16 = softmax(axis = var_735, x = mh_w_27_cast_fp16)[name = string("op_816_cast_fp16")]; + tensor var_817 = const()[name = string("op_817"), val = tensor([1, 12, 64, -1])]; + tensor var_818_cast_fp16 = reshape(shape = var_817, x = value_7_cast_fp16)[name = string("op_818_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_818_cast_fp16, y = var_816_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_821 = const()[name = string("op_821"), val = tensor([1, 768, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_821, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133524928)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134704640)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_843_to_fp16 = const()[name = string("op_843_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_843_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134706240)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134707840)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_15_pad_type_0 = const()[name = string("query_15_pad_type_0"), val = string("valid")]; + tensor query_15_strides_0 = const()[name = string("query_15_strides_0"), val = tensor([1, 1])]; + tensor query_15_pad_0 = const()[name = string("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_15_dilations_0 = const()[name = string("query_15_dilations_0"), val = tensor([1, 1])]; + int32 query_15_groups_0 = const()[name = string("query_15_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134709440)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135889152)))]; + tensor query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_15_cast_fp16")]; + tensor var_863 = const()[name = string("op_863"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_863, x = query_15_cast_fp16)[name = string("mh_q_15_cast_fp16")]; + fp16 var_865_to_fp16 = const()[name = string("op_865_to_fp16"), val = fp16(0x1p-3)]; + tensor var_866_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_865_to_fp16)[name = string("op_866_cast_fp16")]; + tensor var_867 = const()[name = string("op_867"), val = tensor([1, 12, 64, -1])]; + tensor var_868_cast_fp16 = reshape(shape = var_867, x = obj_71_cast_fp16)[name = string("op_868_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_866_cast_fp16, y = var_868_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_31_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_735, x = mh_w_31_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_877 = const()[name = string("op_877"), val = tensor([1, 12, 64, -1])]; + tensor var_878_cast_fp16 = reshape(shape = var_877, x = obj_73_cast_fp16)[name = string("op_878_cast_fp16")]; + bool attn_15_transpose_x_0 = const()[name = string("attn_15_transpose_x_0"), val = bool(false)]; + bool attn_15_transpose_y_0 = const()[name = string("attn_15_transpose_y_0"), val = bool(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_878_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_15_cast_fp16")]; + tensor var_881 = const()[name = string("op_881"), val = tensor([1, 768, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_881, x = attn_15_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135890752)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137070464)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_899_to_fp16 = const()[name = string("op_899_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_899_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137072064)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137073664)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137075264)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141793920)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141800128)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146518784)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_25_cast_fp16")]; + tensor obj_89_begin_0 = const()[name = string("obj_89_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_89_end_0 = const()[name = string("obj_89_end_0"), val = tensor([5, 768, 1, 1536])]; + tensor obj_89_end_mask_0 = const()[name = string("obj_89_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_89_cast_fp16 = slice_by_index(begin = obj_89_begin_0, end = obj_89_end_0, end_mask = obj_89_end_mask_0, x = read_state_2)[name = string("obj_89_cast_fp16")]; + tensor obj_91_begin_0 = const()[name = string("obj_91_begin_0"), val = tensor([4, 0, 0, 0])]; + tensor obj_91_end_0 = const()[name = string("obj_91_end_0"), val = tensor([5, 768, 1, 1536])]; + tensor obj_91_end_mask_0 = const()[name = string("obj_91_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_91_cast_fp16 = slice_by_index(begin = obj_91_begin_0, end = obj_91_end_0, end_mask = obj_91_end_mask_0, x = read_state_3)[name = string("obj_91_cast_fp16")]; + int32 var_944 = const()[name = string("op_944"), val = int32(3)]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + fp16 var_969_to_fp16 = const()[name = string("op_969_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_969_to_fp16, x = inputs_25_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor obj_79_gamma_0_to_fp16 = const()[name = string("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146520384)))]; + tensor obj_79_beta_0_to_fp16 = const()[name = string("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146521984)))]; + fp16 obj_79_epsilon_0_to_fp16 = const()[name = string("obj_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_25_cast_fp16)[name = string("obj_79_cast_fp16")]; + string query_17_pad_type_0 = const()[name = string("query_17_pad_type_0"), val = string("valid")]; + tensor query_17_strides_0 = const()[name = string("query_17_strides_0"), val = tensor([1, 1])]; + tensor query_17_pad_0 = const()[name = string("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_17_dilations_0 = const()[name = string("query_17_dilations_0"), val = tensor([1, 1])]; + int32 query_17_groups_0 = const()[name = string("query_17_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146523584)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147703296)))]; + tensor query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("query_17_cast_fp16")]; + string current_key_9_pad_type_0 = const()[name = string("current_key_9_pad_type_0"), val = string("valid")]; + tensor current_key_9_strides_0 = const()[name = string("current_key_9_strides_0"), val = tensor([1, 1])]; + tensor current_key_9_pad_0 = const()[name = string("current_key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_9_dilations_0 = const()[name = string("current_key_9_dilations_0"), val = tensor([1, 1])]; + int32 current_key_9_groups_0 = const()[name = string("current_key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147704896)))]; + tensor current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_key_9_cast_fp16")]; + string current_value_9_pad_type_0 = const()[name = string("current_value_9_pad_type_0"), val = string("valid")]; + tensor current_value_9_strides_0 = const()[name = string("current_value_9_strides_0"), val = tensor([1, 1])]; + tensor current_value_9_pad_0 = const()[name = string("current_value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_9_dilations_0 = const()[name = string("current_value_9_dilations_0"), val = tensor([1, 1])]; + int32 current_value_9_groups_0 = const()[name = string("current_value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148884608)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150064320)))]; + tensor current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = string("current_value_9_cast_fp16")]; + tensor var_1007_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_169_cast_fp16)[name = string("op_1007_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = var_65_cast_fp16_4, y = var_1007_cast_fp16)[name = string("key_9_cast_fp16")]; + tensor var_1009_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_169_cast_fp16)[name = string("op_1009_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = var_80_cast_fp16_4, y = var_1009_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_1012 = const()[name = string("op_1012"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_1012, x = query_17_cast_fp16)[name = string("mh_q_17_cast_fp16")]; + fp16 var_1014_to_fp16 = const()[name = string("op_1014_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1015_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1014_to_fp16)[name = string("op_1015_cast_fp16")]; + tensor var_1016 = const()[name = string("op_1016"), val = tensor([1, 12, 64, -1])]; + tensor var_1017_cast_fp16 = reshape(shape = var_1016, x = key_9_cast_fp16)[name = string("op_1017_cast_fp16")]; + bool mh_w_33_transpose_x_0 = const()[name = string("mh_w_33_transpose_x_0"), val = bool(true)]; + bool mh_w_33_transpose_y_0 = const()[name = string("mh_w_33_transpose_y_0"), val = bool(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_1015_cast_fp16, y = var_1017_cast_fp16)[name = string("mh_w_33_cast_fp16")]; + tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_35_cast_fp16")]; + tensor var_1025_cast_fp16 = softmax(axis = var_944, x = mh_w_35_cast_fp16)[name = string("op_1025_cast_fp16")]; + tensor var_1026 = const()[name = string("op_1026"), val = tensor([1, 12, 64, -1])]; + tensor var_1027_cast_fp16 = reshape(shape = var_1026, x = value_9_cast_fp16)[name = string("op_1027_cast_fp16")]; + bool attn_17_transpose_x_0 = const()[name = string("attn_17_transpose_x_0"), val = bool(false)]; + bool attn_17_transpose_y_0 = const()[name = string("attn_17_transpose_y_0"), val = bool(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1027_cast_fp16, y = var_1025_cast_fp16)[name = string("attn_17_cast_fp16")]; + tensor var_1030 = const()[name = string("op_1030"), val = tensor([1, 768, 1, -1])]; + tensor input_41_cast_fp16 = reshape(shape = var_1030, x = attn_17_cast_fp16)[name = string("input_41_cast_fp16")]; + string obj_85_pad_type_0 = const()[name = string("obj_85_pad_type_0"), val = string("valid")]; + tensor obj_85_strides_0 = const()[name = string("obj_85_strides_0"), val = tensor([1, 1])]; + tensor obj_85_pad_0 = const()[name = string("obj_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_85_dilations_0 = const()[name = string("obj_85_dilations_0"), val = tensor([1, 1])]; + int32 obj_85_groups_0 = const()[name = string("obj_85_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150065920)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151245632)))]; + tensor obj_85_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_85_dilations_0, groups = obj_85_groups_0, pad = obj_85_pad_0, pad_type = obj_85_pad_type_0, strides = obj_85_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_85_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_85_cast_fp16)[name = string("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + fp16 var_1052_to_fp16 = const()[name = string("op_1052_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1052_to_fp16, x = inputs_27_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor obj_87_gamma_0_to_fp16 = const()[name = string("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151247232)))]; + tensor obj_87_beta_0_to_fp16 = const()[name = string("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151248832)))]; + fp16 obj_87_epsilon_0_to_fp16 = const()[name = string("obj_87_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_27_cast_fp16)[name = string("obj_87_cast_fp16")]; + string query_19_pad_type_0 = const()[name = string("query_19_pad_type_0"), val = string("valid")]; + tensor query_19_strides_0 = const()[name = string("query_19_strides_0"), val = tensor([1, 1])]; + tensor query_19_pad_0 = const()[name = string("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_19_dilations_0 = const()[name = string("query_19_dilations_0"), val = tensor([1, 1])]; + int32 query_19_groups_0 = const()[name = string("query_19_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151250432)))]; + tensor layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152430144)))]; + tensor query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16, x = obj_87_cast_fp16)[name = string("query_19_cast_fp16")]; + tensor var_1072 = const()[name = string("op_1072"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_1072, x = query_19_cast_fp16)[name = string("mh_q_19_cast_fp16")]; + fp16 var_1074_to_fp16 = const()[name = string("op_1074_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1075_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1074_to_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1076 = const()[name = string("op_1076"), val = tensor([1, 12, 64, -1])]; + tensor var_1077_cast_fp16 = reshape(shape = var_1076, x = obj_89_cast_fp16)[name = string("op_1077_cast_fp16")]; + bool mh_w_37_transpose_x_0 = const()[name = string("mh_w_37_transpose_x_0"), val = bool(true)]; + bool mh_w_37_transpose_y_0 = const()[name = string("mh_w_37_transpose_y_0"), val = bool(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1075_cast_fp16, y = var_1077_cast_fp16)[name = string("mh_w_37_cast_fp16")]; + tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_39_cast_fp16")]; + tensor obj_95_cast_fp16 = softmax(axis = var_944, x = mh_w_39_cast_fp16)[name = string("obj_95_cast_fp16")]; + tensor var_1086 = const()[name = string("op_1086"), val = tensor([1, 12, 64, -1])]; + tensor var_1087_cast_fp16 = reshape(shape = var_1086, x = obj_91_cast_fp16)[name = string("op_1087_cast_fp16")]; + bool attn_19_transpose_x_0 = const()[name = string("attn_19_transpose_x_0"), val = bool(false)]; + bool attn_19_transpose_y_0 = const()[name = string("attn_19_transpose_y_0"), val = bool(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1087_cast_fp16, y = obj_95_cast_fp16)[name = string("attn_19_cast_fp16")]; + tensor var_1090 = const()[name = string("op_1090"), val = tensor([1, 768, 1, -1])]; + tensor input_43_cast_fp16 = reshape(shape = var_1090, x = attn_19_cast_fp16)[name = string("input_43_cast_fp16")]; + string obj_93_pad_type_0 = const()[name = string("obj_93_pad_type_0"), val = string("valid")]; + tensor obj_93_strides_0 = const()[name = string("obj_93_strides_0"), val = tensor([1, 1])]; + tensor obj_93_pad_0 = const()[name = string("obj_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_93_dilations_0 = const()[name = string("obj_93_dilations_0"), val = tensor([1, 1])]; + int32 obj_93_groups_0 = const()[name = string("obj_93_groups_0"), val = int32(1)]; + tensor layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152431744)))]; + tensor layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153611456)))]; + tensor obj_93_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_93_dilations_0, groups = obj_93_groups_0, pad = obj_93_pad_0, pad_type = obj_93_pad_type_0, strides = obj_93_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("obj_93_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_93_cast_fp16)[name = string("inputs_29_cast_fp16")]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + fp16 var_1108_to_fp16 = const()[name = string("op_1108_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1108_to_fp16, x = inputs_29_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153613056)))]; + tensor input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153614656)))]; + fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_29_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")]; + tensor input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor([1, 1])]; + tensor input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor([1, 1])]; + int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153616256)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158334912)))]; + tensor input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")]; + tensor input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = string("input_49_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158341120)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163059776)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_31_cast_fp16")]; + tensor obj_107_begin_0 = const()[name = string("obj_107_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_107_end_0 = const()[name = string("obj_107_end_0"), val = tensor([6, 768, 1, 1536])]; + tensor obj_107_end_mask_0 = const()[name = string("obj_107_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_107_cast_fp16 = slice_by_index(begin = obj_107_begin_0, end = obj_107_end_0, end_mask = obj_107_end_mask_0, x = read_state_2)[name = string("obj_107_cast_fp16")]; + tensor obj_109_begin_0 = const()[name = string("obj_109_begin_0"), val = tensor([5, 0, 0, 0])]; + tensor obj_109_end_0 = const()[name = string("obj_109_end_0"), val = tensor([6, 768, 1, 1536])]; + tensor obj_109_end_mask_0 = const()[name = string("obj_109_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_109_cast_fp16 = slice_by_index(begin = obj_109_begin_0, end = obj_109_end_0, end_mask = obj_109_end_mask_0, x = read_state_3)[name = string("obj_109_cast_fp16")]; + int32 var_1153 = const()[name = string("op_1153"), val = int32(3)]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + fp16 var_1178_to_fp16 = const()[name = string("op_1178_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1178_to_fp16, x = inputs_31_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = string("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163061376)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = string("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163062976)))]; + fp16 obj_97_epsilon_0_to_fp16 = const()[name = string("obj_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_31_cast_fp16)[name = string("obj_97_cast_fp16")]; + string query_21_pad_type_0 = const()[name = string("query_21_pad_type_0"), val = string("valid")]; + tensor query_21_strides_0 = const()[name = string("query_21_strides_0"), val = tensor([1, 1])]; + tensor query_21_pad_0 = const()[name = string("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_21_dilations_0 = const()[name = string("query_21_dilations_0"), val = tensor([1, 1])]; + int32 query_21_groups_0 = const()[name = string("query_21_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163064576)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164244288)))]; + tensor query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("query_21_cast_fp16")]; + string current_key_11_pad_type_0 = const()[name = string("current_key_11_pad_type_0"), val = string("valid")]; + tensor current_key_11_strides_0 = const()[name = string("current_key_11_strides_0"), val = tensor([1, 1])]; + tensor current_key_11_pad_0 = const()[name = string("current_key_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_11_dilations_0 = const()[name = string("current_key_11_dilations_0"), val = tensor([1, 1])]; + int32 current_key_11_groups_0 = const()[name = string("current_key_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164245888)))]; + tensor current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_key_11_cast_fp16")]; + string current_value_11_pad_type_0 = const()[name = string("current_value_11_pad_type_0"), val = string("valid")]; + tensor current_value_11_strides_0 = const()[name = string("current_value_11_strides_0"), val = tensor([1, 1])]; + tensor current_value_11_pad_0 = const()[name = string("current_value_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_11_dilations_0 = const()[name = string("current_value_11_dilations_0"), val = tensor([1, 1])]; + int32 current_value_11_groups_0 = const()[name = string("current_value_11_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165425600)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166605312)))]; + tensor current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_97_cast_fp16)[name = string("current_value_11_cast_fp16")]; + tensor var_1216_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_169_cast_fp16)[name = string("op_1216_cast_fp16")]; + tensor key_11_cast_fp16 = add(x = var_65_cast_fp16_5, y = var_1216_cast_fp16)[name = string("key_11_cast_fp16")]; + tensor var_1218_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_169_cast_fp16)[name = string("op_1218_cast_fp16")]; + tensor value_11_cast_fp16 = add(x = var_80_cast_fp16_5, y = var_1218_cast_fp16)[name = string("value_11_cast_fp16")]; + tensor var_1221 = const()[name = string("op_1221"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_1221, x = query_21_cast_fp16)[name = string("mh_q_21_cast_fp16")]; + fp16 var_1223_to_fp16 = const()[name = string("op_1223_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1224_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1223_to_fp16)[name = string("op_1224_cast_fp16")]; + tensor var_1225 = const()[name = string("op_1225"), val = tensor([1, 12, 64, -1])]; + tensor var_1226_cast_fp16 = reshape(shape = var_1225, x = key_11_cast_fp16)[name = string("op_1226_cast_fp16")]; + bool mh_w_41_transpose_x_0 = const()[name = string("mh_w_41_transpose_x_0"), val = bool(true)]; + bool mh_w_41_transpose_y_0 = const()[name = string("mh_w_41_transpose_y_0"), val = bool(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1224_cast_fp16, y = var_1226_cast_fp16)[name = string("mh_w_41_cast_fp16")]; + tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_43_cast_fp16")]; + tensor var_1234_cast_fp16 = softmax(axis = var_1153, x = mh_w_43_cast_fp16)[name = string("op_1234_cast_fp16")]; + tensor var_1235 = const()[name = string("op_1235"), val = tensor([1, 12, 64, -1])]; + tensor var_1236_cast_fp16 = reshape(shape = var_1235, x = value_11_cast_fp16)[name = string("op_1236_cast_fp16")]; + bool attn_21_transpose_x_0 = const()[name = string("attn_21_transpose_x_0"), val = bool(false)]; + bool attn_21_transpose_y_0 = const()[name = string("attn_21_transpose_y_0"), val = bool(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1236_cast_fp16, y = var_1234_cast_fp16)[name = string("attn_21_cast_fp16")]; + tensor var_1239 = const()[name = string("op_1239"), val = tensor([1, 768, 1, -1])]; + tensor input_51_cast_fp16 = reshape(shape = var_1239, x = attn_21_cast_fp16)[name = string("input_51_cast_fp16")]; + string obj_103_pad_type_0 = const()[name = string("obj_103_pad_type_0"), val = string("valid")]; + tensor obj_103_strides_0 = const()[name = string("obj_103_strides_0"), val = tensor([1, 1])]; + tensor obj_103_pad_0 = const()[name = string("obj_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_103_dilations_0 = const()[name = string("obj_103_dilations_0"), val = tensor([1, 1])]; + int32 obj_103_groups_0 = const()[name = string("obj_103_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166606912)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167786624)))]; + tensor obj_103_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_103_dilations_0, groups = obj_103_groups_0, pad = obj_103_pad_0, pad_type = obj_103_pad_type_0, strides = obj_103_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("obj_103_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_103_cast_fp16)[name = string("inputs_33_cast_fp16")]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + fp16 var_1261_to_fp16 = const()[name = string("op_1261_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1261_to_fp16, x = inputs_33_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = string("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167788224)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = string("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167789824)))]; + fp16 obj_105_epsilon_0_to_fp16 = const()[name = string("obj_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_33_cast_fp16)[name = string("obj_105_cast_fp16")]; + string query_23_pad_type_0 = const()[name = string("query_23_pad_type_0"), val = string("valid")]; + tensor query_23_strides_0 = const()[name = string("query_23_strides_0"), val = tensor([1, 1])]; + tensor query_23_pad_0 = const()[name = string("query_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_23_dilations_0 = const()[name = string("query_23_dilations_0"), val = tensor([1, 1])]; + int32 query_23_groups_0 = const()[name = string("query_23_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167791424)))]; + tensor layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168971136)))]; + tensor query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_105_cast_fp16)[name = string("query_23_cast_fp16")]; + tensor var_1281 = const()[name = string("op_1281"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_23_cast_fp16 = reshape(shape = var_1281, x = query_23_cast_fp16)[name = string("mh_q_23_cast_fp16")]; + fp16 var_1283_to_fp16 = const()[name = string("op_1283_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1284_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1283_to_fp16)[name = string("op_1284_cast_fp16")]; + tensor var_1285 = const()[name = string("op_1285"), val = tensor([1, 12, 64, -1])]; + tensor var_1286_cast_fp16 = reshape(shape = var_1285, x = obj_107_cast_fp16)[name = string("op_1286_cast_fp16")]; + bool mh_w_45_transpose_x_0 = const()[name = string("mh_w_45_transpose_x_0"), val = bool(true)]; + bool mh_w_45_transpose_y_0 = const()[name = string("mh_w_45_transpose_y_0"), val = bool(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_1284_cast_fp16, y = var_1286_cast_fp16)[name = string("mh_w_45_cast_fp16")]; + tensor mh_w_47_cast_fp16 = add(x = mh_w_45_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_47_cast_fp16")]; + tensor obj_113_cast_fp16 = softmax(axis = var_1153, x = mh_w_47_cast_fp16)[name = string("obj_113_cast_fp16")]; + tensor var_1295 = const()[name = string("op_1295"), val = tensor([1, 12, 64, -1])]; + tensor var_1296_cast_fp16 = reshape(shape = var_1295, x = obj_109_cast_fp16)[name = string("op_1296_cast_fp16")]; + bool attn_23_transpose_x_0 = const()[name = string("attn_23_transpose_x_0"), val = bool(false)]; + bool attn_23_transpose_y_0 = const()[name = string("attn_23_transpose_y_0"), val = bool(true)]; + tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1296_cast_fp16, y = obj_113_cast_fp16)[name = string("attn_23_cast_fp16")]; + tensor var_1299 = const()[name = string("op_1299"), val = tensor([1, 768, 1, -1])]; + tensor input_53_cast_fp16 = reshape(shape = var_1299, x = attn_23_cast_fp16)[name = string("input_53_cast_fp16")]; + string obj_111_pad_type_0 = const()[name = string("obj_111_pad_type_0"), val = string("valid")]; + tensor obj_111_strides_0 = const()[name = string("obj_111_strides_0"), val = tensor([1, 1])]; + tensor obj_111_pad_0 = const()[name = string("obj_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_111_dilations_0 = const()[name = string("obj_111_dilations_0"), val = tensor([1, 1])]; + int32 obj_111_groups_0 = const()[name = string("obj_111_groups_0"), val = int32(1)]; + tensor layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168972736)))]; + tensor layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170152448)))]; + tensor obj_111_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_111_dilations_0, groups = obj_111_groups_0, pad = obj_111_pad_0, pad_type = obj_111_pad_type_0, strides = obj_111_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("obj_111_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_111_cast_fp16)[name = string("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + fp16 var_1320_to_fp16 = const()[name = string("op_1320_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1320_to_fp16, x = inputs_35_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor input_55_gamma_0_to_fp16 = const()[name = string("input_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170154048)))]; + tensor input_55_beta_0_to_fp16 = const()[name = string("input_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170155648)))]; + fp16 input_55_epsilon_0_to_fp16 = const()[name = string("input_55_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_35_cast_fp16)[name = string("input_55_cast_fp16")]; + string input_57_pad_type_0 = const()[name = string("input_57_pad_type_0"), val = string("valid")]; + tensor input_57_strides_0 = const()[name = string("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_pad_0 = const()[name = string("input_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_57_dilations_0 = const()[name = string("input_57_dilations_0"), val = tensor([1, 1])]; + int32 input_57_groups_0 = const()[name = string("input_57_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170157248)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174875904)))]; + tensor input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; + string input_59_mode_0 = const()[name = string("input_59_mode_0"), val = string("EXACT")]; + tensor input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174882112)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179600768)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_59_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_37_cast_fp16")]; + tensor obj_125_begin_0 = const()[name = string("obj_125_begin_0"), val = tensor([6, 0, 0, 0])]; + tensor obj_125_end_0 = const()[name = string("obj_125_end_0"), val = tensor([7, 768, 1, 1536])]; + tensor obj_125_end_mask_0 = const()[name = string("obj_125_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_125_cast_fp16 = slice_by_index(begin = obj_125_begin_0, end = obj_125_end_0, end_mask = obj_125_end_mask_0, x = read_state_2)[name = string("obj_125_cast_fp16")]; + tensor obj_127_begin_0 = const()[name = string("obj_127_begin_0"), val = tensor([6, 0, 0, 0])]; + tensor obj_127_end_0 = const()[name = string("obj_127_end_0"), val = tensor([7, 768, 1, 1536])]; + tensor obj_127_end_mask_0 = const()[name = string("obj_127_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_127_cast_fp16 = slice_by_index(begin = obj_127_begin_0, end = obj_127_end_0, end_mask = obj_127_end_mask_0, x = read_state_3)[name = string("obj_127_cast_fp16")]; + int32 var_1366 = const()[name = string("op_1366"), val = int32(3)]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + fp16 var_1391_to_fp16 = const()[name = string("op_1391_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1391_to_fp16, x = inputs_37_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor obj_115_gamma_0_to_fp16 = const()[name = string("obj_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179602368)))]; + tensor obj_115_beta_0_to_fp16 = const()[name = string("obj_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179603968)))]; + fp16 obj_115_epsilon_0_to_fp16 = const()[name = string("obj_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_115_cast_fp16 = batch_norm(beta = obj_115_beta_0_to_fp16, epsilon = obj_115_epsilon_0_to_fp16, gamma = obj_115_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_37_cast_fp16)[name = string("obj_115_cast_fp16")]; + string query_25_pad_type_0 = const()[name = string("query_25_pad_type_0"), val = string("valid")]; + tensor query_25_strides_0 = const()[name = string("query_25_strides_0"), val = tensor([1, 1])]; + tensor query_25_pad_0 = const()[name = string("query_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_25_dilations_0 = const()[name = string("query_25_dilations_0"), val = tensor([1, 1])]; + int32 query_25_groups_0 = const()[name = string("query_25_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179605568)))]; + tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180785280)))]; + tensor query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("query_25_cast_fp16")]; + string current_key_13_pad_type_0 = const()[name = string("current_key_13_pad_type_0"), val = string("valid")]; + tensor current_key_13_strides_0 = const()[name = string("current_key_13_strides_0"), val = tensor([1, 1])]; + tensor current_key_13_pad_0 = const()[name = string("current_key_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_13_dilations_0 = const()[name = string("current_key_13_dilations_0"), val = tensor([1, 1])]; + int32 current_key_13_groups_0 = const()[name = string("current_key_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180786880)))]; + tensor current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("current_key_13_cast_fp16")]; + string current_value_13_pad_type_0 = const()[name = string("current_value_13_pad_type_0"), val = string("valid")]; + tensor current_value_13_strides_0 = const()[name = string("current_value_13_strides_0"), val = tensor([1, 1])]; + tensor current_value_13_pad_0 = const()[name = string("current_value_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_13_dilations_0 = const()[name = string("current_value_13_dilations_0"), val = tensor([1, 1])]; + int32 current_value_13_groups_0 = const()[name = string("current_value_13_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181966592)))]; + tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183146304)))]; + tensor current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_115_cast_fp16)[name = string("current_value_13_cast_fp16")]; + tensor var_1429_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_169_cast_fp16)[name = string("op_1429_cast_fp16")]; + tensor key_13_cast_fp16 = add(x = var_65_cast_fp16_6, y = var_1429_cast_fp16)[name = string("key_13_cast_fp16")]; + tensor var_1431_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_169_cast_fp16)[name = string("op_1431_cast_fp16")]; + tensor value_13_cast_fp16 = add(x = var_80_cast_fp16_6, y = var_1431_cast_fp16)[name = string("value_13_cast_fp16")]; + tensor var_1434 = const()[name = string("op_1434"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_25_cast_fp16 = reshape(shape = var_1434, x = query_25_cast_fp16)[name = string("mh_q_25_cast_fp16")]; + fp16 var_1436_to_fp16 = const()[name = string("op_1436_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1437_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1436_to_fp16)[name = string("op_1437_cast_fp16")]; + tensor var_1438 = const()[name = string("op_1438"), val = tensor([1, 12, 64, -1])]; + tensor var_1439_cast_fp16 = reshape(shape = var_1438, x = key_13_cast_fp16)[name = string("op_1439_cast_fp16")]; + bool mh_w_49_transpose_x_0 = const()[name = string("mh_w_49_transpose_x_0"), val = bool(true)]; + bool mh_w_49_transpose_y_0 = const()[name = string("mh_w_49_transpose_y_0"), val = bool(false)]; + tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_1437_cast_fp16, y = var_1439_cast_fp16)[name = string("mh_w_49_cast_fp16")]; + tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_51_cast_fp16")]; + tensor var_1447_cast_fp16 = softmax(axis = var_1366, x = mh_w_51_cast_fp16)[name = string("op_1447_cast_fp16")]; + tensor var_1448 = const()[name = string("op_1448"), val = tensor([1, 12, 64, -1])]; + tensor var_1449_cast_fp16 = reshape(shape = var_1448, x = value_13_cast_fp16)[name = string("op_1449_cast_fp16")]; + bool attn_25_transpose_x_0 = const()[name = string("attn_25_transpose_x_0"), val = bool(false)]; + bool attn_25_transpose_y_0 = const()[name = string("attn_25_transpose_y_0"), val = bool(true)]; + tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1449_cast_fp16, y = var_1447_cast_fp16)[name = string("attn_25_cast_fp16")]; + tensor var_1452 = const()[name = string("op_1452"), val = tensor([1, 768, 1, -1])]; + tensor input_61_cast_fp16 = reshape(shape = var_1452, x = attn_25_cast_fp16)[name = string("input_61_cast_fp16")]; + string obj_121_pad_type_0 = const()[name = string("obj_121_pad_type_0"), val = string("valid")]; + tensor obj_121_strides_0 = const()[name = string("obj_121_strides_0"), val = tensor([1, 1])]; + tensor obj_121_pad_0 = const()[name = string("obj_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_121_dilations_0 = const()[name = string("obj_121_dilations_0"), val = tensor([1, 1])]; + int32 obj_121_groups_0 = const()[name = string("obj_121_groups_0"), val = int32(1)]; + tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183147904)))]; + tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184327616)))]; + tensor obj_121_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_121_dilations_0, groups = obj_121_groups_0, pad = obj_121_pad_0, pad_type = obj_121_pad_type_0, strides = obj_121_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_61_cast_fp16)[name = string("obj_121_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_121_cast_fp16)[name = string("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + fp16 var_1474_to_fp16 = const()[name = string("op_1474_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1474_to_fp16, x = inputs_39_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor obj_123_gamma_0_to_fp16 = const()[name = string("obj_123_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184329216)))]; + tensor obj_123_beta_0_to_fp16 = const()[name = string("obj_123_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184330816)))]; + fp16 obj_123_epsilon_0_to_fp16 = const()[name = string("obj_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_123_cast_fp16 = batch_norm(beta = obj_123_beta_0_to_fp16, epsilon = obj_123_epsilon_0_to_fp16, gamma = obj_123_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_39_cast_fp16)[name = string("obj_123_cast_fp16")]; + string query_27_pad_type_0 = const()[name = string("query_27_pad_type_0"), val = string("valid")]; + tensor query_27_strides_0 = const()[name = string("query_27_strides_0"), val = tensor([1, 1])]; + tensor query_27_pad_0 = const()[name = string("query_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_27_dilations_0 = const()[name = string("query_27_dilations_0"), val = tensor([1, 1])]; + int32 query_27_groups_0 = const()[name = string("query_27_groups_0"), val = int32(1)]; + tensor layers_6_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184332416)))]; + tensor layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185512128)))]; + tensor query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16, x = obj_123_cast_fp16)[name = string("query_27_cast_fp16")]; + tensor var_1494 = const()[name = string("op_1494"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_27_cast_fp16 = reshape(shape = var_1494, x = query_27_cast_fp16)[name = string("mh_q_27_cast_fp16")]; + fp16 var_1496_to_fp16 = const()[name = string("op_1496_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1497_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1496_to_fp16)[name = string("op_1497_cast_fp16")]; + tensor var_1498 = const()[name = string("op_1498"), val = tensor([1, 12, 64, -1])]; + tensor var_1499_cast_fp16 = reshape(shape = var_1498, x = obj_125_cast_fp16)[name = string("op_1499_cast_fp16")]; + bool mh_w_53_transpose_x_0 = const()[name = string("mh_w_53_transpose_x_0"), val = bool(true)]; + bool mh_w_53_transpose_y_0 = const()[name = string("mh_w_53_transpose_y_0"), val = bool(false)]; + tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_1497_cast_fp16, y = var_1499_cast_fp16)[name = string("mh_w_53_cast_fp16")]; + tensor mh_w_55_cast_fp16 = add(x = mh_w_53_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_55_cast_fp16")]; + tensor obj_131_cast_fp16 = softmax(axis = var_1366, x = mh_w_55_cast_fp16)[name = string("obj_131_cast_fp16")]; + tensor var_1508 = const()[name = string("op_1508"), val = tensor([1, 12, 64, -1])]; + tensor var_1509_cast_fp16 = reshape(shape = var_1508, x = obj_127_cast_fp16)[name = string("op_1509_cast_fp16")]; + bool attn_27_transpose_x_0 = const()[name = string("attn_27_transpose_x_0"), val = bool(false)]; + bool attn_27_transpose_y_0 = const()[name = string("attn_27_transpose_y_0"), val = bool(true)]; + tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1509_cast_fp16, y = obj_131_cast_fp16)[name = string("attn_27_cast_fp16")]; + tensor var_1512 = const()[name = string("op_1512"), val = tensor([1, 768, 1, -1])]; + tensor input_63_cast_fp16 = reshape(shape = var_1512, x = attn_27_cast_fp16)[name = string("input_63_cast_fp16")]; + string obj_129_pad_type_0 = const()[name = string("obj_129_pad_type_0"), val = string("valid")]; + tensor obj_129_strides_0 = const()[name = string("obj_129_strides_0"), val = tensor([1, 1])]; + tensor obj_129_pad_0 = const()[name = string("obj_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_129_dilations_0 = const()[name = string("obj_129_dilations_0"), val = tensor([1, 1])]; + int32 obj_129_groups_0 = const()[name = string("obj_129_groups_0"), val = int32(1)]; + tensor layers_6_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185513728)))]; + tensor layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186693440)))]; + tensor obj_129_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_129_dilations_0, groups = obj_129_groups_0, pad = obj_129_pad_0, pad_type = obj_129_pad_type_0, strides = obj_129_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("obj_129_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_129_cast_fp16)[name = string("inputs_41_cast_fp16")]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + fp16 var_1530_to_fp16 = const()[name = string("op_1530_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1530_to_fp16, x = inputs_41_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor input_65_gamma_0_to_fp16 = const()[name = string("input_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186695040)))]; + tensor input_65_beta_0_to_fp16 = const()[name = string("input_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186696640)))]; + fp16 input_65_epsilon_0_to_fp16 = const()[name = string("input_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_41_cast_fp16)[name = string("input_65_cast_fp16")]; + string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")]; + tensor input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor([1, 1])]; + tensor input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor([1, 1])]; + int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)]; + tensor layers_6_fc1_weight_to_fp16 = const()[name = string("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186698240)))]; + tensor layers_6_fc1_bias_to_fp16 = const()[name = string("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191416896)))]; + tensor input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")]; + string input_69_mode_0 = const()[name = string("input_69_mode_0"), val = string("EXACT")]; + tensor input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor layers_6_fc2_weight_to_fp16 = const()[name = string("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191423104)))]; + tensor layers_6_fc2_bias_to_fp16 = const()[name = string("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196141760)))]; + tensor hidden_states_15_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_69_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("inputs_43_cast_fp16")]; + tensor obj_143_begin_0 = const()[name = string("obj_143_begin_0"), val = tensor([7, 0, 0, 0])]; + tensor obj_143_end_0 = const()[name = string("obj_143_end_0"), val = tensor([8, 768, 1, 1536])]; + tensor obj_143_end_mask_0 = const()[name = string("obj_143_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_143_cast_fp16 = slice_by_index(begin = obj_143_begin_0, end = obj_143_end_0, end_mask = obj_143_end_mask_0, x = read_state_2)[name = string("obj_143_cast_fp16")]; + tensor obj_145_begin_0 = const()[name = string("obj_145_begin_0"), val = tensor([7, 0, 0, 0])]; + tensor obj_145_end_0 = const()[name = string("obj_145_end_0"), val = tensor([8, 768, 1, 1536])]; + tensor obj_145_end_mask_0 = const()[name = string("obj_145_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_145_cast_fp16 = slice_by_index(begin = obj_145_begin_0, end = obj_145_end_0, end_mask = obj_145_end_mask_0, x = read_state_3)[name = string("obj_145_cast_fp16")]; + int32 var_1575 = const()[name = string("op_1575"), val = int32(3)]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + fp16 var_1600_to_fp16 = const()[name = string("op_1600_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1600_to_fp16, x = inputs_43_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor obj_133_gamma_0_to_fp16 = const()[name = string("obj_133_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196143360)))]; + tensor obj_133_beta_0_to_fp16 = const()[name = string("obj_133_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196144960)))]; + fp16 obj_133_epsilon_0_to_fp16 = const()[name = string("obj_133_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_133_cast_fp16 = batch_norm(beta = obj_133_beta_0_to_fp16, epsilon = obj_133_epsilon_0_to_fp16, gamma = obj_133_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_43_cast_fp16)[name = string("obj_133_cast_fp16")]; + string query_29_pad_type_0 = const()[name = string("query_29_pad_type_0"), val = string("valid")]; + tensor query_29_strides_0 = const()[name = string("query_29_strides_0"), val = tensor([1, 1])]; + tensor query_29_pad_0 = const()[name = string("query_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_29_dilations_0 = const()[name = string("query_29_dilations_0"), val = tensor([1, 1])]; + int32 query_29_groups_0 = const()[name = string("query_29_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196146560)))]; + tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197326272)))]; + tensor query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("query_29_cast_fp16")]; + string current_key_15_pad_type_0 = const()[name = string("current_key_15_pad_type_0"), val = string("valid")]; + tensor current_key_15_strides_0 = const()[name = string("current_key_15_strides_0"), val = tensor([1, 1])]; + tensor current_key_15_pad_0 = const()[name = string("current_key_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_15_dilations_0 = const()[name = string("current_key_15_dilations_0"), val = tensor([1, 1])]; + int32 current_key_15_groups_0 = const()[name = string("current_key_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197327872)))]; + tensor current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("current_key_15_cast_fp16")]; + string current_value_15_pad_type_0 = const()[name = string("current_value_15_pad_type_0"), val = string("valid")]; + tensor current_value_15_strides_0 = const()[name = string("current_value_15_strides_0"), val = tensor([1, 1])]; + tensor current_value_15_pad_0 = const()[name = string("current_value_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_15_dilations_0 = const()[name = string("current_value_15_dilations_0"), val = tensor([1, 1])]; + int32 current_value_15_groups_0 = const()[name = string("current_value_15_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198507584)))]; + tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199687296)))]; + tensor current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_133_cast_fp16)[name = string("current_value_15_cast_fp16")]; + tensor var_1638_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_169_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor key_15_cast_fp16 = add(x = var_65_cast_fp16_7, y = var_1638_cast_fp16)[name = string("key_15_cast_fp16")]; + tensor var_1640_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_169_cast_fp16)[name = string("op_1640_cast_fp16")]; + tensor value_15_cast_fp16 = add(x = var_80_cast_fp16_7, y = var_1640_cast_fp16)[name = string("value_15_cast_fp16")]; + tensor var_1643 = const()[name = string("op_1643"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_29_cast_fp16 = reshape(shape = var_1643, x = query_29_cast_fp16)[name = string("mh_q_29_cast_fp16")]; + fp16 var_1645_to_fp16 = const()[name = string("op_1645_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1646_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1645_to_fp16)[name = string("op_1646_cast_fp16")]; + tensor var_1647 = const()[name = string("op_1647"), val = tensor([1, 12, 64, -1])]; + tensor var_1648_cast_fp16 = reshape(shape = var_1647, x = key_15_cast_fp16)[name = string("op_1648_cast_fp16")]; + bool mh_w_57_transpose_x_0 = const()[name = string("mh_w_57_transpose_x_0"), val = bool(true)]; + bool mh_w_57_transpose_y_0 = const()[name = string("mh_w_57_transpose_y_0"), val = bool(false)]; + tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_1646_cast_fp16, y = var_1648_cast_fp16)[name = string("mh_w_57_cast_fp16")]; + tensor mh_w_59_cast_fp16 = add(x = mh_w_57_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_59_cast_fp16")]; + tensor var_1656_cast_fp16 = softmax(axis = var_1575, x = mh_w_59_cast_fp16)[name = string("op_1656_cast_fp16")]; + tensor var_1657 = const()[name = string("op_1657"), val = tensor([1, 12, 64, -1])]; + tensor var_1658_cast_fp16 = reshape(shape = var_1657, x = value_15_cast_fp16)[name = string("op_1658_cast_fp16")]; + bool attn_29_transpose_x_0 = const()[name = string("attn_29_transpose_x_0"), val = bool(false)]; + bool attn_29_transpose_y_0 = const()[name = string("attn_29_transpose_y_0"), val = bool(true)]; + tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1658_cast_fp16, y = var_1656_cast_fp16)[name = string("attn_29_cast_fp16")]; + tensor var_1661 = const()[name = string("op_1661"), val = tensor([1, 768, 1, -1])]; + tensor input_71_cast_fp16 = reshape(shape = var_1661, x = attn_29_cast_fp16)[name = string("input_71_cast_fp16")]; + string obj_139_pad_type_0 = const()[name = string("obj_139_pad_type_0"), val = string("valid")]; + tensor obj_139_strides_0 = const()[name = string("obj_139_strides_0"), val = tensor([1, 1])]; + tensor obj_139_pad_0 = const()[name = string("obj_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_139_dilations_0 = const()[name = string("obj_139_dilations_0"), val = tensor([1, 1])]; + int32 obj_139_groups_0 = const()[name = string("obj_139_groups_0"), val = int32(1)]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199688896)))]; + tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200868608)))]; + tensor obj_139_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_139_dilations_0, groups = obj_139_groups_0, pad = obj_139_pad_0, pad_type = obj_139_pad_type_0, strides = obj_139_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("obj_139_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_139_cast_fp16)[name = string("inputs_45_cast_fp16")]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + fp16 var_1683_to_fp16 = const()[name = string("op_1683_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1683_to_fp16, x = inputs_45_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor obj_141_gamma_0_to_fp16 = const()[name = string("obj_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200870208)))]; + tensor obj_141_beta_0_to_fp16 = const()[name = string("obj_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200871808)))]; + fp16 obj_141_epsilon_0_to_fp16 = const()[name = string("obj_141_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_45_cast_fp16)[name = string("obj_141_cast_fp16")]; + string query_31_pad_type_0 = const()[name = string("query_31_pad_type_0"), val = string("valid")]; + tensor query_31_strides_0 = const()[name = string("query_31_strides_0"), val = tensor([1, 1])]; + tensor query_31_pad_0 = const()[name = string("query_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_31_dilations_0 = const()[name = string("query_31_dilations_0"), val = tensor([1, 1])]; + int32 query_31_groups_0 = const()[name = string("query_31_groups_0"), val = int32(1)]; + tensor layers_7_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200873408)))]; + tensor layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202053120)))]; + tensor query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = string("query_31_cast_fp16")]; + tensor var_1703 = const()[name = string("op_1703"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_31_cast_fp16 = reshape(shape = var_1703, x = query_31_cast_fp16)[name = string("mh_q_31_cast_fp16")]; + fp16 var_1705_to_fp16 = const()[name = string("op_1705_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1706_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1705_to_fp16)[name = string("op_1706_cast_fp16")]; + tensor var_1707 = const()[name = string("op_1707"), val = tensor([1, 12, 64, -1])]; + tensor var_1708_cast_fp16 = reshape(shape = var_1707, x = obj_143_cast_fp16)[name = string("op_1708_cast_fp16")]; + bool mh_w_61_transpose_x_0 = const()[name = string("mh_w_61_transpose_x_0"), val = bool(true)]; + bool mh_w_61_transpose_y_0 = const()[name = string("mh_w_61_transpose_y_0"), val = bool(false)]; + tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_1706_cast_fp16, y = var_1708_cast_fp16)[name = string("mh_w_61_cast_fp16")]; + tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_63_cast_fp16")]; + tensor obj_149_cast_fp16 = softmax(axis = var_1575, x = mh_w_63_cast_fp16)[name = string("obj_149_cast_fp16")]; + tensor var_1717 = const()[name = string("op_1717"), val = tensor([1, 12, 64, -1])]; + tensor var_1718_cast_fp16 = reshape(shape = var_1717, x = obj_145_cast_fp16)[name = string("op_1718_cast_fp16")]; + bool attn_31_transpose_x_0 = const()[name = string("attn_31_transpose_x_0"), val = bool(false)]; + bool attn_31_transpose_y_0 = const()[name = string("attn_31_transpose_y_0"), val = bool(true)]; + tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1718_cast_fp16, y = obj_149_cast_fp16)[name = string("attn_31_cast_fp16")]; + tensor var_1721 = const()[name = string("op_1721"), val = tensor([1, 768, 1, -1])]; + tensor input_73_cast_fp16 = reshape(shape = var_1721, x = attn_31_cast_fp16)[name = string("input_73_cast_fp16")]; + string obj_147_pad_type_0 = const()[name = string("obj_147_pad_type_0"), val = string("valid")]; + tensor obj_147_strides_0 = const()[name = string("obj_147_strides_0"), val = tensor([1, 1])]; + tensor obj_147_pad_0 = const()[name = string("obj_147_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_147_dilations_0 = const()[name = string("obj_147_dilations_0"), val = tensor([1, 1])]; + int32 obj_147_groups_0 = const()[name = string("obj_147_groups_0"), val = int32(1)]; + tensor layers_7_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202054720)))]; + tensor layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203234432)))]; + tensor obj_147_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = string("obj_147_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_147_cast_fp16)[name = string("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + fp16 var_1739_to_fp16 = const()[name = string("op_1739_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1739_to_fp16, x = inputs_47_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor input_75_gamma_0_to_fp16 = const()[name = string("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203236032)))]; + tensor input_75_beta_0_to_fp16 = const()[name = string("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203237632)))]; + fp16 input_75_epsilon_0_to_fp16 = const()[name = string("input_75_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_47_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1, 1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1, 1])]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(1)]; + tensor layers_7_fc1_weight_to_fp16 = const()[name = string("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203239232)))]; + tensor layers_7_fc1_bias_to_fp16 = const()[name = string("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207957888)))]; + tensor input_77_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + string input_79_mode_0 = const()[name = string("input_79_mode_0"), val = string("EXACT")]; + tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = string("input_79_cast_fp16")]; + string hidden_states_17_pad_type_0 = const()[name = string("hidden_states_17_pad_type_0"), val = string("valid")]; + tensor hidden_states_17_strides_0 = const()[name = string("hidden_states_17_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_0 = const()[name = string("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_17_dilations_0 = const()[name = string("hidden_states_17_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_17_groups_0 = const()[name = string("hidden_states_17_groups_0"), val = int32(1)]; + tensor layers_7_fc2_weight_to_fp16 = const()[name = string("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207964096)))]; + tensor layers_7_fc2_bias_to_fp16 = const()[name = string("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212682752)))]; + tensor hidden_states_17_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = string("inputs_49_cast_fp16")]; + tensor obj_161_begin_0 = const()[name = string("obj_161_begin_0"), val = tensor([8, 0, 0, 0])]; + tensor obj_161_end_0 = const()[name = string("obj_161_end_0"), val = tensor([9, 768, 1, 1536])]; + tensor obj_161_end_mask_0 = const()[name = string("obj_161_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_161_cast_fp16 = slice_by_index(begin = obj_161_begin_0, end = obj_161_end_0, end_mask = obj_161_end_mask_0, x = read_state_2)[name = string("obj_161_cast_fp16")]; + tensor obj_163_begin_0 = const()[name = string("obj_163_begin_0"), val = tensor([8, 0, 0, 0])]; + tensor obj_163_end_0 = const()[name = string("obj_163_end_0"), val = tensor([9, 768, 1, 1536])]; + tensor obj_163_end_mask_0 = const()[name = string("obj_163_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_163_cast_fp16 = slice_by_index(begin = obj_163_begin_0, end = obj_163_end_0, end_mask = obj_163_end_mask_0, x = read_state_3)[name = string("obj_163_cast_fp16")]; + int32 var_1784 = const()[name = string("op_1784"), val = int32(3)]; + tensor out_49_axes_0 = const()[name = string("out_49_axes_0"), val = tensor([1])]; + fp16 var_1809_to_fp16 = const()[name = string("op_1809_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1809_to_fp16, x = inputs_49_cast_fp16)[name = string("out_49_cast_fp16")]; + tensor obj_151_gamma_0_to_fp16 = const()[name = string("obj_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212684352)))]; + tensor obj_151_beta_0_to_fp16 = const()[name = string("obj_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212685952)))]; + fp16 obj_151_epsilon_0_to_fp16 = const()[name = string("obj_151_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_151_cast_fp16 = batch_norm(beta = obj_151_beta_0_to_fp16, epsilon = obj_151_epsilon_0_to_fp16, gamma = obj_151_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_49_cast_fp16)[name = string("obj_151_cast_fp16")]; + string query_33_pad_type_0 = const()[name = string("query_33_pad_type_0"), val = string("valid")]; + tensor query_33_strides_0 = const()[name = string("query_33_strides_0"), val = tensor([1, 1])]; + tensor query_33_pad_0 = const()[name = string("query_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_33_dilations_0 = const()[name = string("query_33_dilations_0"), val = tensor([1, 1])]; + int32 query_33_groups_0 = const()[name = string("query_33_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212687552)))]; + tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213867264)))]; + tensor query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("query_33_cast_fp16")]; + string current_key_17_pad_type_0 = const()[name = string("current_key_17_pad_type_0"), val = string("valid")]; + tensor current_key_17_strides_0 = const()[name = string("current_key_17_strides_0"), val = tensor([1, 1])]; + tensor current_key_17_pad_0 = const()[name = string("current_key_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_17_dilations_0 = const()[name = string("current_key_17_dilations_0"), val = tensor([1, 1])]; + int32 current_key_17_groups_0 = const()[name = string("current_key_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213868864)))]; + tensor current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("current_key_17_cast_fp16")]; + string current_value_17_pad_type_0 = const()[name = string("current_value_17_pad_type_0"), val = string("valid")]; + tensor current_value_17_strides_0 = const()[name = string("current_value_17_strides_0"), val = tensor([1, 1])]; + tensor current_value_17_pad_0 = const()[name = string("current_value_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_17_dilations_0 = const()[name = string("current_value_17_dilations_0"), val = tensor([1, 1])]; + int32 current_value_17_groups_0 = const()[name = string("current_value_17_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215048576)))]; + tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216228288)))]; + tensor current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_151_cast_fp16)[name = string("current_value_17_cast_fp16")]; + tensor var_1847_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_169_cast_fp16)[name = string("op_1847_cast_fp16")]; + tensor key_17_cast_fp16 = add(x = var_65_cast_fp16_8, y = var_1847_cast_fp16)[name = string("key_17_cast_fp16")]; + tensor var_1849_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_169_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor value_17_cast_fp16 = add(x = var_80_cast_fp16_8, y = var_1849_cast_fp16)[name = string("value_17_cast_fp16")]; + tensor var_1852 = const()[name = string("op_1852"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_33_cast_fp16 = reshape(shape = var_1852, x = query_33_cast_fp16)[name = string("mh_q_33_cast_fp16")]; + fp16 var_1854_to_fp16 = const()[name = string("op_1854_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1855_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_1854_to_fp16)[name = string("op_1855_cast_fp16")]; + tensor var_1856 = const()[name = string("op_1856"), val = tensor([1, 12, 64, -1])]; + tensor var_1857_cast_fp16 = reshape(shape = var_1856, x = key_17_cast_fp16)[name = string("op_1857_cast_fp16")]; + bool mh_w_65_transpose_x_0 = const()[name = string("mh_w_65_transpose_x_0"), val = bool(true)]; + bool mh_w_65_transpose_y_0 = const()[name = string("mh_w_65_transpose_y_0"), val = bool(false)]; + tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_1855_cast_fp16, y = var_1857_cast_fp16)[name = string("mh_w_65_cast_fp16")]; + tensor mh_w_67_cast_fp16 = add(x = mh_w_65_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_67_cast_fp16")]; + tensor var_1865_cast_fp16 = softmax(axis = var_1784, x = mh_w_67_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1866 = const()[name = string("op_1866"), val = tensor([1, 12, 64, -1])]; + tensor var_1867_cast_fp16 = reshape(shape = var_1866, x = value_17_cast_fp16)[name = string("op_1867_cast_fp16")]; + bool attn_33_transpose_x_0 = const()[name = string("attn_33_transpose_x_0"), val = bool(false)]; + bool attn_33_transpose_y_0 = const()[name = string("attn_33_transpose_y_0"), val = bool(true)]; + tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_1867_cast_fp16, y = var_1865_cast_fp16)[name = string("attn_33_cast_fp16")]; + tensor var_1870 = const()[name = string("op_1870"), val = tensor([1, 768, 1, -1])]; + tensor input_81_cast_fp16 = reshape(shape = var_1870, x = attn_33_cast_fp16)[name = string("input_81_cast_fp16")]; + string obj_157_pad_type_0 = const()[name = string("obj_157_pad_type_0"), val = string("valid")]; + tensor obj_157_strides_0 = const()[name = string("obj_157_strides_0"), val = tensor([1, 1])]; + tensor obj_157_pad_0 = const()[name = string("obj_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_157_dilations_0 = const()[name = string("obj_157_dilations_0"), val = tensor([1, 1])]; + int32 obj_157_groups_0 = const()[name = string("obj_157_groups_0"), val = int32(1)]; + tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216229888)))]; + tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217409600)))]; + tensor obj_157_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_157_dilations_0, groups = obj_157_groups_0, pad = obj_157_pad_0, pad_type = obj_157_pad_type_0, strides = obj_157_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("obj_157_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_157_cast_fp16)[name = string("inputs_51_cast_fp16")]; + tensor out_51_axes_0 = const()[name = string("out_51_axes_0"), val = tensor([1])]; + fp16 var_1892_to_fp16 = const()[name = string("op_1892_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1892_to_fp16, x = inputs_51_cast_fp16)[name = string("out_51_cast_fp16")]; + tensor obj_159_gamma_0_to_fp16 = const()[name = string("obj_159_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217411200)))]; + tensor obj_159_beta_0_to_fp16 = const()[name = string("obj_159_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217412800)))]; + fp16 obj_159_epsilon_0_to_fp16 = const()[name = string("obj_159_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_159_cast_fp16 = batch_norm(beta = obj_159_beta_0_to_fp16, epsilon = obj_159_epsilon_0_to_fp16, gamma = obj_159_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_51_cast_fp16)[name = string("obj_159_cast_fp16")]; + string query_35_pad_type_0 = const()[name = string("query_35_pad_type_0"), val = string("valid")]; + tensor query_35_strides_0 = const()[name = string("query_35_strides_0"), val = tensor([1, 1])]; + tensor query_35_pad_0 = const()[name = string("query_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_35_dilations_0 = const()[name = string("query_35_dilations_0"), val = tensor([1, 1])]; + int32 query_35_groups_0 = const()[name = string("query_35_groups_0"), val = int32(1)]; + tensor layers_8_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217414400)))]; + tensor layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218594112)))]; + tensor query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16, x = obj_159_cast_fp16)[name = string("query_35_cast_fp16")]; + tensor var_1912 = const()[name = string("op_1912"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_35_cast_fp16 = reshape(shape = var_1912, x = query_35_cast_fp16)[name = string("mh_q_35_cast_fp16")]; + fp16 var_1914_to_fp16 = const()[name = string("op_1914_to_fp16"), val = fp16(0x1p-3)]; + tensor var_1915_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_1914_to_fp16)[name = string("op_1915_cast_fp16")]; + tensor var_1916 = const()[name = string("op_1916"), val = tensor([1, 12, 64, -1])]; + tensor var_1917_cast_fp16 = reshape(shape = var_1916, x = obj_161_cast_fp16)[name = string("op_1917_cast_fp16")]; + bool mh_w_69_transpose_x_0 = const()[name = string("mh_w_69_transpose_x_0"), val = bool(true)]; + bool mh_w_69_transpose_y_0 = const()[name = string("mh_w_69_transpose_y_0"), val = bool(false)]; + tensor mh_w_69_cast_fp16 = matmul(transpose_x = mh_w_69_transpose_x_0, transpose_y = mh_w_69_transpose_y_0, x = var_1915_cast_fp16, y = var_1917_cast_fp16)[name = string("mh_w_69_cast_fp16")]; + tensor mh_w_71_cast_fp16 = add(x = mh_w_69_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_71_cast_fp16")]; + tensor obj_167_cast_fp16 = softmax(axis = var_1784, x = mh_w_71_cast_fp16)[name = string("obj_167_cast_fp16")]; + tensor var_1926 = const()[name = string("op_1926"), val = tensor([1, 12, 64, -1])]; + tensor var_1927_cast_fp16 = reshape(shape = var_1926, x = obj_163_cast_fp16)[name = string("op_1927_cast_fp16")]; + bool attn_35_transpose_x_0 = const()[name = string("attn_35_transpose_x_0"), val = bool(false)]; + bool attn_35_transpose_y_0 = const()[name = string("attn_35_transpose_y_0"), val = bool(true)]; + tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_1927_cast_fp16, y = obj_167_cast_fp16)[name = string("attn_35_cast_fp16")]; + tensor var_1930 = const()[name = string("op_1930"), val = tensor([1, 768, 1, -1])]; + tensor input_83_cast_fp16 = reshape(shape = var_1930, x = attn_35_cast_fp16)[name = string("input_83_cast_fp16")]; + string obj_165_pad_type_0 = const()[name = string("obj_165_pad_type_0"), val = string("valid")]; + tensor obj_165_strides_0 = const()[name = string("obj_165_strides_0"), val = tensor([1, 1])]; + tensor obj_165_pad_0 = const()[name = string("obj_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_165_dilations_0 = const()[name = string("obj_165_dilations_0"), val = tensor([1, 1])]; + int32 obj_165_groups_0 = const()[name = string("obj_165_groups_0"), val = int32(1)]; + tensor layers_8_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218595712)))]; + tensor layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219775424)))]; + tensor obj_165_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("obj_165_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_165_cast_fp16)[name = string("inputs_53_cast_fp16")]; + tensor out_53_axes_0 = const()[name = string("out_53_axes_0"), val = tensor([1])]; + fp16 var_1951_to_fp16 = const()[name = string("op_1951_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_1951_to_fp16, x = inputs_53_cast_fp16)[name = string("out_53_cast_fp16")]; + tensor input_85_gamma_0_to_fp16 = const()[name = string("input_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219777024)))]; + tensor input_85_beta_0_to_fp16 = const()[name = string("input_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219778624)))]; + fp16 input_85_epsilon_0_to_fp16 = const()[name = string("input_85_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_53_cast_fp16)[name = string("input_85_cast_fp16")]; + string input_87_pad_type_0 = const()[name = string("input_87_pad_type_0"), val = string("valid")]; + tensor input_87_strides_0 = const()[name = string("input_87_strides_0"), val = tensor([1, 1])]; + tensor input_87_pad_0 = const()[name = string("input_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_87_dilations_0 = const()[name = string("input_87_dilations_0"), val = tensor([1, 1])]; + int32 input_87_groups_0 = const()[name = string("input_87_groups_0"), val = int32(1)]; + tensor layers_8_fc1_weight_to_fp16 = const()[name = string("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219780224)))]; + tensor layers_8_fc1_bias_to_fp16 = const()[name = string("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224498880)))]; + tensor input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")]; + string input_89_mode_0 = const()[name = string("input_89_mode_0"), val = string("EXACT")]; + tensor input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = string("input_89_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor layers_8_fc2_weight_to_fp16 = const()[name = string("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224505088)))]; + tensor layers_8_fc2_bias_to_fp16 = const()[name = string("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229223744)))]; + tensor hidden_states_19_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_89_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("inputs_55_cast_fp16")]; + tensor obj_179_begin_0 = const()[name = string("obj_179_begin_0"), val = tensor([9, 0, 0, 0])]; + tensor obj_179_end_0 = const()[name = string("obj_179_end_0"), val = tensor([10, 768, 1, 1536])]; + tensor obj_179_end_mask_0 = const()[name = string("obj_179_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_179_cast_fp16 = slice_by_index(begin = obj_179_begin_0, end = obj_179_end_0, end_mask = obj_179_end_mask_0, x = read_state_2)[name = string("obj_179_cast_fp16")]; + tensor obj_181_begin_0 = const()[name = string("obj_181_begin_0"), val = tensor([9, 0, 0, 0])]; + tensor obj_181_end_0 = const()[name = string("obj_181_end_0"), val = tensor([10, 768, 1, 1536])]; + tensor obj_181_end_mask_0 = const()[name = string("obj_181_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_181_cast_fp16 = slice_by_index(begin = obj_181_begin_0, end = obj_181_end_0, end_mask = obj_181_end_mask_0, x = read_state_3)[name = string("obj_181_cast_fp16")]; + int32 var_1997 = const()[name = string("op_1997"), val = int32(3)]; + tensor out_55_axes_0 = const()[name = string("out_55_axes_0"), val = tensor([1])]; + fp16 var_2022_to_fp16 = const()[name = string("op_2022_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2022_to_fp16, x = inputs_55_cast_fp16)[name = string("out_55_cast_fp16")]; + tensor obj_169_gamma_0_to_fp16 = const()[name = string("obj_169_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229225344)))]; + tensor obj_169_beta_0_to_fp16 = const()[name = string("obj_169_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229226944)))]; + fp16 obj_169_epsilon_0_to_fp16 = const()[name = string("obj_169_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_169_cast_fp16 = batch_norm(beta = obj_169_beta_0_to_fp16, epsilon = obj_169_epsilon_0_to_fp16, gamma = obj_169_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_55_cast_fp16)[name = string("obj_169_cast_fp16")]; + string query_37_pad_type_0 = const()[name = string("query_37_pad_type_0"), val = string("valid")]; + tensor query_37_strides_0 = const()[name = string("query_37_strides_0"), val = tensor([1, 1])]; + tensor query_37_pad_0 = const()[name = string("query_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_37_dilations_0 = const()[name = string("query_37_dilations_0"), val = tensor([1, 1])]; + int32 query_37_groups_0 = const()[name = string("query_37_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229228544)))]; + tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230408256)))]; + tensor query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("query_37_cast_fp16")]; + string current_key_19_pad_type_0 = const()[name = string("current_key_19_pad_type_0"), val = string("valid")]; + tensor current_key_19_strides_0 = const()[name = string("current_key_19_strides_0"), val = tensor([1, 1])]; + tensor current_key_19_pad_0 = const()[name = string("current_key_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_19_dilations_0 = const()[name = string("current_key_19_dilations_0"), val = tensor([1, 1])]; + int32 current_key_19_groups_0 = const()[name = string("current_key_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230409856)))]; + tensor current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("current_key_19_cast_fp16")]; + string current_value_19_pad_type_0 = const()[name = string("current_value_19_pad_type_0"), val = string("valid")]; + tensor current_value_19_strides_0 = const()[name = string("current_value_19_strides_0"), val = tensor([1, 1])]; + tensor current_value_19_pad_0 = const()[name = string("current_value_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_19_dilations_0 = const()[name = string("current_value_19_dilations_0"), val = tensor([1, 1])]; + int32 current_value_19_groups_0 = const()[name = string("current_value_19_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231589568)))]; + tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232769280)))]; + tensor current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_169_cast_fp16)[name = string("current_value_19_cast_fp16")]; + tensor var_2060_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_169_cast_fp16)[name = string("op_2060_cast_fp16")]; + tensor key_19_cast_fp16 = add(x = var_65_cast_fp16_9, y = var_2060_cast_fp16)[name = string("key_19_cast_fp16")]; + tensor var_2062_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_169_cast_fp16)[name = string("op_2062_cast_fp16")]; + tensor value_19_cast_fp16 = add(x = var_80_cast_fp16_9, y = var_2062_cast_fp16)[name = string("value_19_cast_fp16")]; + tensor var_2065 = const()[name = string("op_2065"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_37_cast_fp16 = reshape(shape = var_2065, x = query_37_cast_fp16)[name = string("mh_q_37_cast_fp16")]; + fp16 var_2067_to_fp16 = const()[name = string("op_2067_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2068_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2067_to_fp16)[name = string("op_2068_cast_fp16")]; + tensor var_2069 = const()[name = string("op_2069"), val = tensor([1, 12, 64, -1])]; + tensor var_2070_cast_fp16 = reshape(shape = var_2069, x = key_19_cast_fp16)[name = string("op_2070_cast_fp16")]; + bool mh_w_73_transpose_x_0 = const()[name = string("mh_w_73_transpose_x_0"), val = bool(true)]; + bool mh_w_73_transpose_y_0 = const()[name = string("mh_w_73_transpose_y_0"), val = bool(false)]; + tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_2068_cast_fp16, y = var_2070_cast_fp16)[name = string("mh_w_73_cast_fp16")]; + tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_75_cast_fp16")]; + tensor var_2078_cast_fp16 = softmax(axis = var_1997, x = mh_w_75_cast_fp16)[name = string("op_2078_cast_fp16")]; + tensor var_2079 = const()[name = string("op_2079"), val = tensor([1, 12, 64, -1])]; + tensor var_2080_cast_fp16 = reshape(shape = var_2079, x = value_19_cast_fp16)[name = string("op_2080_cast_fp16")]; + bool attn_37_transpose_x_0 = const()[name = string("attn_37_transpose_x_0"), val = bool(false)]; + bool attn_37_transpose_y_0 = const()[name = string("attn_37_transpose_y_0"), val = bool(true)]; + tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2080_cast_fp16, y = var_2078_cast_fp16)[name = string("attn_37_cast_fp16")]; + tensor var_2083 = const()[name = string("op_2083"), val = tensor([1, 768, 1, -1])]; + tensor input_91_cast_fp16 = reshape(shape = var_2083, x = attn_37_cast_fp16)[name = string("input_91_cast_fp16")]; + string obj_175_pad_type_0 = const()[name = string("obj_175_pad_type_0"), val = string("valid")]; + tensor obj_175_strides_0 = const()[name = string("obj_175_strides_0"), val = tensor([1, 1])]; + tensor obj_175_pad_0 = const()[name = string("obj_175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_175_dilations_0 = const()[name = string("obj_175_dilations_0"), val = tensor([1, 1])]; + int32 obj_175_groups_0 = const()[name = string("obj_175_groups_0"), val = int32(1)]; + tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232770880)))]; + tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233950592)))]; + tensor obj_175_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_175_dilations_0, groups = obj_175_groups_0, pad = obj_175_pad_0, pad_type = obj_175_pad_type_0, strides = obj_175_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = string("obj_175_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_175_cast_fp16)[name = string("inputs_57_cast_fp16")]; + tensor out_57_axes_0 = const()[name = string("out_57_axes_0"), val = tensor([1])]; + fp16 var_2105_to_fp16 = const()[name = string("op_2105_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2105_to_fp16, x = inputs_57_cast_fp16)[name = string("out_57_cast_fp16")]; + tensor obj_177_gamma_0_to_fp16 = const()[name = string("obj_177_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233952192)))]; + tensor obj_177_beta_0_to_fp16 = const()[name = string("obj_177_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233953792)))]; + fp16 obj_177_epsilon_0_to_fp16 = const()[name = string("obj_177_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_177_cast_fp16 = batch_norm(beta = obj_177_beta_0_to_fp16, epsilon = obj_177_epsilon_0_to_fp16, gamma = obj_177_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_57_cast_fp16)[name = string("obj_177_cast_fp16")]; + string query_39_pad_type_0 = const()[name = string("query_39_pad_type_0"), val = string("valid")]; + tensor query_39_strides_0 = const()[name = string("query_39_strides_0"), val = tensor([1, 1])]; + tensor query_39_pad_0 = const()[name = string("query_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_39_dilations_0 = const()[name = string("query_39_dilations_0"), val = tensor([1, 1])]; + int32 query_39_groups_0 = const()[name = string("query_39_groups_0"), val = int32(1)]; + tensor layers_9_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233955392)))]; + tensor layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235135104)))]; + tensor query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16, x = obj_177_cast_fp16)[name = string("query_39_cast_fp16")]; + tensor var_2125 = const()[name = string("op_2125"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_39_cast_fp16 = reshape(shape = var_2125, x = query_39_cast_fp16)[name = string("mh_q_39_cast_fp16")]; + fp16 var_2127_to_fp16 = const()[name = string("op_2127_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2128_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2127_to_fp16)[name = string("op_2128_cast_fp16")]; + tensor var_2129 = const()[name = string("op_2129"), val = tensor([1, 12, 64, -1])]; + tensor var_2130_cast_fp16 = reshape(shape = var_2129, x = obj_179_cast_fp16)[name = string("op_2130_cast_fp16")]; + bool mh_w_77_transpose_x_0 = const()[name = string("mh_w_77_transpose_x_0"), val = bool(true)]; + bool mh_w_77_transpose_y_0 = const()[name = string("mh_w_77_transpose_y_0"), val = bool(false)]; + tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_2128_cast_fp16, y = var_2130_cast_fp16)[name = string("mh_w_77_cast_fp16")]; + tensor mh_w_79_cast_fp16 = add(x = mh_w_77_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_79_cast_fp16")]; + tensor obj_185_cast_fp16 = softmax(axis = var_1997, x = mh_w_79_cast_fp16)[name = string("obj_185_cast_fp16")]; + tensor var_2139 = const()[name = string("op_2139"), val = tensor([1, 12, 64, -1])]; + tensor var_2140_cast_fp16 = reshape(shape = var_2139, x = obj_181_cast_fp16)[name = string("op_2140_cast_fp16")]; + bool attn_39_transpose_x_0 = const()[name = string("attn_39_transpose_x_0"), val = bool(false)]; + bool attn_39_transpose_y_0 = const()[name = string("attn_39_transpose_y_0"), val = bool(true)]; + tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2140_cast_fp16, y = obj_185_cast_fp16)[name = string("attn_39_cast_fp16")]; + tensor var_2143 = const()[name = string("op_2143"), val = tensor([1, 768, 1, -1])]; + tensor input_93_cast_fp16 = reshape(shape = var_2143, x = attn_39_cast_fp16)[name = string("input_93_cast_fp16")]; + string obj_183_pad_type_0 = const()[name = string("obj_183_pad_type_0"), val = string("valid")]; + tensor obj_183_strides_0 = const()[name = string("obj_183_strides_0"), val = tensor([1, 1])]; + tensor obj_183_pad_0 = const()[name = string("obj_183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_183_dilations_0 = const()[name = string("obj_183_dilations_0"), val = tensor([1, 1])]; + int32 obj_183_groups_0 = const()[name = string("obj_183_groups_0"), val = int32(1)]; + tensor layers_9_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235136704)))]; + tensor layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236316416)))]; + tensor obj_183_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_183_dilations_0, groups = obj_183_groups_0, pad = obj_183_pad_0, pad_type = obj_183_pad_type_0, strides = obj_183_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("obj_183_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_183_cast_fp16)[name = string("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = string("out_59_axes_0"), val = tensor([1])]; + fp16 var_2164_to_fp16 = const()[name = string("op_2164_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2164_to_fp16, x = inputs_59_cast_fp16)[name = string("out_59_cast_fp16")]; + tensor input_95_gamma_0_to_fp16 = const()[name = string("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236318016)))]; + tensor input_95_beta_0_to_fp16 = const()[name = string("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236319616)))]; + fp16 input_95_epsilon_0_to_fp16 = const()[name = string("input_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_59_cast_fp16)[name = string("input_95_cast_fp16")]; + string input_97_pad_type_0 = const()[name = string("input_97_pad_type_0"), val = string("valid")]; + tensor input_97_strides_0 = const()[name = string("input_97_strides_0"), val = tensor([1, 1])]; + tensor input_97_pad_0 = const()[name = string("input_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_97_dilations_0 = const()[name = string("input_97_dilations_0"), val = tensor([1, 1])]; + int32 input_97_groups_0 = const()[name = string("input_97_groups_0"), val = int32(1)]; + tensor layers_9_fc1_weight_to_fp16 = const()[name = string("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236321216)))]; + tensor layers_9_fc1_bias_to_fp16 = const()[name = string("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241039872)))]; + tensor input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_95_cast_fp16)[name = string("input_97_cast_fp16")]; + string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("EXACT")]; + tensor input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")]; + string hidden_states_21_pad_type_0 = const()[name = string("hidden_states_21_pad_type_0"), val = string("valid")]; + tensor hidden_states_21_strides_0 = const()[name = string("hidden_states_21_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_0 = const()[name = string("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_21_dilations_0 = const()[name = string("hidden_states_21_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_21_groups_0 = const()[name = string("hidden_states_21_groups_0"), val = int32(1)]; + tensor layers_9_fc2_weight_to_fp16 = const()[name = string("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241046080)))]; + tensor layers_9_fc2_bias_to_fp16 = const()[name = string("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245764736)))]; + tensor hidden_states_21_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_99_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = string("inputs_61_cast_fp16")]; + tensor obj_197_begin_0 = const()[name = string("obj_197_begin_0"), val = tensor([10, 0, 0, 0])]; + tensor obj_197_end_0 = const()[name = string("obj_197_end_0"), val = tensor([11, 768, 1, 1536])]; + tensor obj_197_end_mask_0 = const()[name = string("obj_197_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_197_cast_fp16 = slice_by_index(begin = obj_197_begin_0, end = obj_197_end_0, end_mask = obj_197_end_mask_0, x = read_state_2)[name = string("obj_197_cast_fp16")]; + tensor obj_199_begin_0 = const()[name = string("obj_199_begin_0"), val = tensor([10, 0, 0, 0])]; + tensor obj_199_end_0 = const()[name = string("obj_199_end_0"), val = tensor([11, 768, 1, 1536])]; + tensor obj_199_end_mask_0 = const()[name = string("obj_199_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_199_cast_fp16 = slice_by_index(begin = obj_199_begin_0, end = obj_199_end_0, end_mask = obj_199_end_mask_0, x = read_state_3)[name = string("obj_199_cast_fp16")]; + int32 var_2210 = const()[name = string("op_2210"), val = int32(3)]; + tensor out_61_axes_0 = const()[name = string("out_61_axes_0"), val = tensor([1])]; + fp16 var_2235_to_fp16 = const()[name = string("op_2235_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2235_to_fp16, x = inputs_61_cast_fp16)[name = string("out_61_cast_fp16")]; + tensor obj_187_gamma_0_to_fp16 = const()[name = string("obj_187_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245766336)))]; + tensor obj_187_beta_0_to_fp16 = const()[name = string("obj_187_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245767936)))]; + fp16 obj_187_epsilon_0_to_fp16 = const()[name = string("obj_187_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_187_cast_fp16 = batch_norm(beta = obj_187_beta_0_to_fp16, epsilon = obj_187_epsilon_0_to_fp16, gamma = obj_187_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_61_cast_fp16)[name = string("obj_187_cast_fp16")]; + string query_41_pad_type_0 = const()[name = string("query_41_pad_type_0"), val = string("valid")]; + tensor query_41_strides_0 = const()[name = string("query_41_strides_0"), val = tensor([1, 1])]; + tensor query_41_pad_0 = const()[name = string("query_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_41_dilations_0 = const()[name = string("query_41_dilations_0"), val = tensor([1, 1])]; + int32 query_41_groups_0 = const()[name = string("query_41_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245769536)))]; + tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246949248)))]; + tensor query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("query_41_cast_fp16")]; + string current_key_21_pad_type_0 = const()[name = string("current_key_21_pad_type_0"), val = string("valid")]; + tensor current_key_21_strides_0 = const()[name = string("current_key_21_strides_0"), val = tensor([1, 1])]; + tensor current_key_21_pad_0 = const()[name = string("current_key_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_21_dilations_0 = const()[name = string("current_key_21_dilations_0"), val = tensor([1, 1])]; + int32 current_key_21_groups_0 = const()[name = string("current_key_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246950848)))]; + tensor current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("current_key_21_cast_fp16")]; + string current_value_21_pad_type_0 = const()[name = string("current_value_21_pad_type_0"), val = string("valid")]; + tensor current_value_21_strides_0 = const()[name = string("current_value_21_strides_0"), val = tensor([1, 1])]; + tensor current_value_21_pad_0 = const()[name = string("current_value_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_21_dilations_0 = const()[name = string("current_value_21_dilations_0"), val = tensor([1, 1])]; + int32 current_value_21_groups_0 = const()[name = string("current_value_21_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248130560)))]; + tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249310272)))]; + tensor current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_187_cast_fp16)[name = string("current_value_21_cast_fp16")]; + tensor var_2273_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_169_cast_fp16)[name = string("op_2273_cast_fp16")]; + tensor key_21_cast_fp16 = add(x = var_65_cast_fp16_10, y = var_2273_cast_fp16)[name = string("key_21_cast_fp16")]; + tensor var_2275_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_169_cast_fp16)[name = string("op_2275_cast_fp16")]; + tensor value_21_cast_fp16 = add(x = var_80_cast_fp16_10, y = var_2275_cast_fp16)[name = string("value_21_cast_fp16")]; + tensor var_2278 = const()[name = string("op_2278"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_41_cast_fp16 = reshape(shape = var_2278, x = query_41_cast_fp16)[name = string("mh_q_41_cast_fp16")]; + fp16 var_2280_to_fp16 = const()[name = string("op_2280_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2281_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2280_to_fp16)[name = string("op_2281_cast_fp16")]; + tensor var_2282 = const()[name = string("op_2282"), val = tensor([1, 12, 64, -1])]; + tensor var_2283_cast_fp16 = reshape(shape = var_2282, x = key_21_cast_fp16)[name = string("op_2283_cast_fp16")]; + bool mh_w_81_transpose_x_0 = const()[name = string("mh_w_81_transpose_x_0"), val = bool(true)]; + bool mh_w_81_transpose_y_0 = const()[name = string("mh_w_81_transpose_y_0"), val = bool(false)]; + tensor mh_w_81_cast_fp16 = matmul(transpose_x = mh_w_81_transpose_x_0, transpose_y = mh_w_81_transpose_y_0, x = var_2281_cast_fp16, y = var_2283_cast_fp16)[name = string("mh_w_81_cast_fp16")]; + tensor mh_w_83_cast_fp16 = add(x = mh_w_81_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_83_cast_fp16")]; + tensor var_2291_cast_fp16 = softmax(axis = var_2210, x = mh_w_83_cast_fp16)[name = string("op_2291_cast_fp16")]; + tensor var_2292 = const()[name = string("op_2292"), val = tensor([1, 12, 64, -1])]; + tensor var_2293_cast_fp16 = reshape(shape = var_2292, x = value_21_cast_fp16)[name = string("op_2293_cast_fp16")]; + bool attn_41_transpose_x_0 = const()[name = string("attn_41_transpose_x_0"), val = bool(false)]; + bool attn_41_transpose_y_0 = const()[name = string("attn_41_transpose_y_0"), val = bool(true)]; + tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2293_cast_fp16, y = var_2291_cast_fp16)[name = string("attn_41_cast_fp16")]; + tensor var_2296 = const()[name = string("op_2296"), val = tensor([1, 768, 1, -1])]; + tensor input_101_cast_fp16 = reshape(shape = var_2296, x = attn_41_cast_fp16)[name = string("input_101_cast_fp16")]; + string obj_193_pad_type_0 = const()[name = string("obj_193_pad_type_0"), val = string("valid")]; + tensor obj_193_strides_0 = const()[name = string("obj_193_strides_0"), val = tensor([1, 1])]; + tensor obj_193_pad_0 = const()[name = string("obj_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_193_dilations_0 = const()[name = string("obj_193_dilations_0"), val = tensor([1, 1])]; + int32 obj_193_groups_0 = const()[name = string("obj_193_groups_0"), val = int32(1)]; + tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249311872)))]; + tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250491584)))]; + tensor obj_193_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_193_dilations_0, groups = obj_193_groups_0, pad = obj_193_pad_0, pad_type = obj_193_pad_type_0, strides = obj_193_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_101_cast_fp16)[name = string("obj_193_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_193_cast_fp16)[name = string("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = string("out_63_axes_0"), val = tensor([1])]; + fp16 var_2318_to_fp16 = const()[name = string("op_2318_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2318_to_fp16, x = inputs_63_cast_fp16)[name = string("out_63_cast_fp16")]; + tensor obj_195_gamma_0_to_fp16 = const()[name = string("obj_195_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250493184)))]; + tensor obj_195_beta_0_to_fp16 = const()[name = string("obj_195_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250494784)))]; + fp16 obj_195_epsilon_0_to_fp16 = const()[name = string("obj_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_195_cast_fp16 = batch_norm(beta = obj_195_beta_0_to_fp16, epsilon = obj_195_epsilon_0_to_fp16, gamma = obj_195_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_63_cast_fp16)[name = string("obj_195_cast_fp16")]; + string query_43_pad_type_0 = const()[name = string("query_43_pad_type_0"), val = string("valid")]; + tensor query_43_strides_0 = const()[name = string("query_43_strides_0"), val = tensor([1, 1])]; + tensor query_43_pad_0 = const()[name = string("query_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_43_dilations_0 = const()[name = string("query_43_dilations_0"), val = tensor([1, 1])]; + int32 query_43_groups_0 = const()[name = string("query_43_groups_0"), val = int32(1)]; + tensor layers_10_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250496384)))]; + tensor layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251676096)))]; + tensor query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16, x = obj_195_cast_fp16)[name = string("query_43_cast_fp16")]; + tensor var_2338 = const()[name = string("op_2338"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_43_cast_fp16 = reshape(shape = var_2338, x = query_43_cast_fp16)[name = string("mh_q_43_cast_fp16")]; + fp16 var_2340_to_fp16 = const()[name = string("op_2340_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2341_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2340_to_fp16)[name = string("op_2341_cast_fp16")]; + tensor var_2342 = const()[name = string("op_2342"), val = tensor([1, 12, 64, -1])]; + tensor var_2343_cast_fp16 = reshape(shape = var_2342, x = obj_197_cast_fp16)[name = string("op_2343_cast_fp16")]; + bool mh_w_85_transpose_x_0 = const()[name = string("mh_w_85_transpose_x_0"), val = bool(true)]; + bool mh_w_85_transpose_y_0 = const()[name = string("mh_w_85_transpose_y_0"), val = bool(false)]; + tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_2341_cast_fp16, y = var_2343_cast_fp16)[name = string("mh_w_85_cast_fp16")]; + tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_87_cast_fp16")]; + tensor obj_203_cast_fp16 = softmax(axis = var_2210, x = mh_w_87_cast_fp16)[name = string("obj_203_cast_fp16")]; + tensor var_2352 = const()[name = string("op_2352"), val = tensor([1, 12, 64, -1])]; + tensor var_2353_cast_fp16 = reshape(shape = var_2352, x = obj_199_cast_fp16)[name = string("op_2353_cast_fp16")]; + bool attn_43_transpose_x_0 = const()[name = string("attn_43_transpose_x_0"), val = bool(false)]; + bool attn_43_transpose_y_0 = const()[name = string("attn_43_transpose_y_0"), val = bool(true)]; + tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2353_cast_fp16, y = obj_203_cast_fp16)[name = string("attn_43_cast_fp16")]; + tensor var_2356 = const()[name = string("op_2356"), val = tensor([1, 768, 1, -1])]; + tensor input_103_cast_fp16 = reshape(shape = var_2356, x = attn_43_cast_fp16)[name = string("input_103_cast_fp16")]; + string obj_201_pad_type_0 = const()[name = string("obj_201_pad_type_0"), val = string("valid")]; + tensor obj_201_strides_0 = const()[name = string("obj_201_strides_0"), val = tensor([1, 1])]; + tensor obj_201_pad_0 = const()[name = string("obj_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_201_dilations_0 = const()[name = string("obj_201_dilations_0"), val = tensor([1, 1])]; + int32 obj_201_groups_0 = const()[name = string("obj_201_groups_0"), val = int32(1)]; + tensor layers_10_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251677696)))]; + tensor layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252857408)))]; + tensor obj_201_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_201_dilations_0, groups = obj_201_groups_0, pad = obj_201_pad_0, pad_type = obj_201_pad_type_0, strides = obj_201_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("obj_201_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_201_cast_fp16)[name = string("inputs_65_cast_fp16")]; + tensor out_65_axes_0 = const()[name = string("out_65_axes_0"), val = tensor([1])]; + fp16 var_2377_to_fp16 = const()[name = string("op_2377_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2377_to_fp16, x = inputs_65_cast_fp16)[name = string("out_65_cast_fp16")]; + tensor input_105_gamma_0_to_fp16 = const()[name = string("input_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252859008)))]; + tensor input_105_beta_0_to_fp16 = const()[name = string("input_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252860608)))]; + fp16 input_105_epsilon_0_to_fp16 = const()[name = string("input_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_65_cast_fp16)[name = string("input_105_cast_fp16")]; + string input_107_pad_type_0 = const()[name = string("input_107_pad_type_0"), val = string("valid")]; + tensor input_107_strides_0 = const()[name = string("input_107_strides_0"), val = tensor([1, 1])]; + tensor input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_107_dilations_0 = const()[name = string("input_107_dilations_0"), val = tensor([1, 1])]; + int32 input_107_groups_0 = const()[name = string("input_107_groups_0"), val = int32(1)]; + tensor layers_10_fc1_weight_to_fp16 = const()[name = string("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(252862208)))]; + tensor layers_10_fc1_bias_to_fp16 = const()[name = string("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257580864)))]; + tensor input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")]; + string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("EXACT")]; + tensor input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")]; + string hidden_states_23_pad_type_0 = const()[name = string("hidden_states_23_pad_type_0"), val = string("valid")]; + tensor hidden_states_23_strides_0 = const()[name = string("hidden_states_23_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_23_pad_0 = const()[name = string("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_23_dilations_0 = const()[name = string("hidden_states_23_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_23_groups_0 = const()[name = string("hidden_states_23_groups_0"), val = int32(1)]; + tensor layers_10_fc2_weight_to_fp16 = const()[name = string("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257587072)))]; + tensor layers_10_fc2_bias_to_fp16 = const()[name = string("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262305728)))]; + tensor hidden_states_23_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = string("hidden_states_23_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = string("inputs_67_cast_fp16")]; + tensor obj_215_begin_0 = const()[name = string("obj_215_begin_0"), val = tensor([11, 0, 0, 0])]; + tensor obj_215_end_0 = const()[name = string("obj_215_end_0"), val = tensor([12, 768, 1, 1536])]; + tensor obj_215_end_mask_0 = const()[name = string("obj_215_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_215_cast_fp16 = slice_by_index(begin = obj_215_begin_0, end = obj_215_end_0, end_mask = obj_215_end_mask_0, x = read_state_2)[name = string("obj_215_cast_fp16")]; + tensor obj_217_begin_0 = const()[name = string("obj_217_begin_0"), val = tensor([11, 0, 0, 0])]; + tensor obj_217_end_0 = const()[name = string("obj_217_end_0"), val = tensor([12, 768, 1, 1536])]; + tensor obj_217_end_mask_0 = const()[name = string("obj_217_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_217_cast_fp16 = slice_by_index(begin = obj_217_begin_0, end = obj_217_end_0, end_mask = obj_217_end_mask_0, x = read_state_3)[name = string("obj_217_cast_fp16")]; + int32 var_2423 = const()[name = string("op_2423"), val = int32(3)]; + tensor out_67_axes_0 = const()[name = string("out_67_axes_0"), val = tensor([1])]; + fp16 var_2448_to_fp16 = const()[name = string("op_2448_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2448_to_fp16, x = inputs_67_cast_fp16)[name = string("out_67_cast_fp16")]; + tensor obj_205_gamma_0_to_fp16 = const()[name = string("obj_205_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262307328)))]; + tensor obj_205_beta_0_to_fp16 = const()[name = string("obj_205_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262308928)))]; + fp16 obj_205_epsilon_0_to_fp16 = const()[name = string("obj_205_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_205_cast_fp16 = batch_norm(beta = obj_205_beta_0_to_fp16, epsilon = obj_205_epsilon_0_to_fp16, gamma = obj_205_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_67_cast_fp16)[name = string("obj_205_cast_fp16")]; + string query_45_pad_type_0 = const()[name = string("query_45_pad_type_0"), val = string("valid")]; + tensor query_45_strides_0 = const()[name = string("query_45_strides_0"), val = tensor([1, 1])]; + tensor query_45_pad_0 = const()[name = string("query_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_45_dilations_0 = const()[name = string("query_45_dilations_0"), val = tensor([1, 1])]; + int32 query_45_groups_0 = const()[name = string("query_45_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262310528)))]; + tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263490240)))]; + tensor query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("query_45_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263491840)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(264671552)))]; + tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265851264)))]; + tensor current_value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_205_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_2486_cast_fp16 = mul(x = current_key_cast_fp16, y = var_169_cast_fp16)[name = string("op_2486_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_65_cast_fp16_11, y = var_2486_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_2488_cast_fp16 = mul(x = current_value_cast_fp16, y = var_169_cast_fp16)[name = string("op_2488_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_80_cast_fp16_11, y = var_2488_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_2491 = const()[name = string("op_2491"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_45_cast_fp16 = reshape(shape = var_2491, x = query_45_cast_fp16)[name = string("mh_q_45_cast_fp16")]; + fp16 var_2493_to_fp16 = const()[name = string("op_2493_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2494_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2493_to_fp16)[name = string("op_2494_cast_fp16")]; + tensor var_2495 = const()[name = string("op_2495"), val = tensor([1, 12, 64, -1])]; + tensor var_2496_cast_fp16 = reshape(shape = var_2495, x = key_cast_fp16)[name = string("op_2496_cast_fp16")]; + bool mh_w_89_transpose_x_0 = const()[name = string("mh_w_89_transpose_x_0"), val = bool(true)]; + bool mh_w_89_transpose_y_0 = const()[name = string("mh_w_89_transpose_y_0"), val = bool(false)]; + tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_2494_cast_fp16, y = var_2496_cast_fp16)[name = string("mh_w_89_cast_fp16")]; + tensor mh_w_91_cast_fp16 = add(x = mh_w_89_cast_fp16, y = var_186_cast_fp16)[name = string("mh_w_91_cast_fp16")]; + tensor var_2504_cast_fp16 = softmax(axis = var_2423, x = mh_w_91_cast_fp16)[name = string("op_2504_cast_fp16")]; + tensor var_2505 = const()[name = string("op_2505"), val = tensor([1, 12, 64, -1])]; + tensor var_2506_cast_fp16 = reshape(shape = var_2505, x = value_cast_fp16)[name = string("op_2506_cast_fp16")]; + bool attn_45_transpose_x_0 = const()[name = string("attn_45_transpose_x_0"), val = bool(false)]; + bool attn_45_transpose_y_0 = const()[name = string("attn_45_transpose_y_0"), val = bool(true)]; + tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2506_cast_fp16, y = var_2504_cast_fp16)[name = string("attn_45_cast_fp16")]; + tensor var_2509 = const()[name = string("op_2509"), val = tensor([1, 768, 1, -1])]; + tensor input_111_cast_fp16 = reshape(shape = var_2509, x = attn_45_cast_fp16)[name = string("input_111_cast_fp16")]; + string obj_211_pad_type_0 = const()[name = string("obj_211_pad_type_0"), val = string("valid")]; + tensor obj_211_strides_0 = const()[name = string("obj_211_strides_0"), val = tensor([1, 1])]; + tensor obj_211_pad_0 = const()[name = string("obj_211_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_211_dilations_0 = const()[name = string("obj_211_dilations_0"), val = tensor([1, 1])]; + int32 obj_211_groups_0 = const()[name = string("obj_211_groups_0"), val = int32(1)]; + tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265852864)))]; + tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267032576)))]; + tensor obj_211_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_211_dilations_0, groups = obj_211_groups_0, pad = obj_211_pad_0, pad_type = obj_211_pad_type_0, strides = obj_211_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("obj_211_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_211_cast_fp16)[name = string("inputs_69_cast_fp16")]; + tensor out_69_axes_0 = const()[name = string("out_69_axes_0"), val = tensor([1])]; + fp16 var_2531_to_fp16 = const()[name = string("op_2531_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2531_to_fp16, x = inputs_69_cast_fp16)[name = string("out_69_cast_fp16")]; + tensor obj_213_gamma_0_to_fp16 = const()[name = string("obj_213_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267034176)))]; + tensor obj_213_beta_0_to_fp16 = const()[name = string("obj_213_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267035776)))]; + fp16 obj_213_epsilon_0_to_fp16 = const()[name = string("obj_213_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_213_cast_fp16 = batch_norm(beta = obj_213_beta_0_to_fp16, epsilon = obj_213_epsilon_0_to_fp16, gamma = obj_213_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_69_cast_fp16)[name = string("obj_213_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_11_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267037376)))]; + tensor layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268217088)))]; + tensor query_cast_fp16 = conv(bias = layers_11_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_11_encoder_attn_q_proj_weight_to_fp16, x = obj_213_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_2551 = const()[name = string("op_2551"), val = tensor([1, 12, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_2551, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_2553_to_fp16 = const()[name = string("op_2553_to_fp16"), val = fp16(0x1p-3)]; + tensor var_2554_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_2553_to_fp16)[name = string("op_2554_cast_fp16")]; + tensor var_2555 = const()[name = string("op_2555"), val = tensor([1, 12, 64, -1])]; + tensor var_2556_cast_fp16 = reshape(shape = var_2555, x = obj_215_cast_fp16)[name = string("op_2556_cast_fp16")]; + bool mh_w_93_transpose_x_0 = const()[name = string("mh_w_93_transpose_x_0"), val = bool(true)]; + bool mh_w_93_transpose_y_0 = const()[name = string("mh_w_93_transpose_y_0"), val = bool(false)]; + tensor mh_w_93_cast_fp16 = matmul(transpose_x = mh_w_93_transpose_x_0, transpose_y = mh_w_93_transpose_y_0, x = var_2554_cast_fp16, y = var_2556_cast_fp16)[name = string("mh_w_93_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_93_cast_fp16, y = var_246_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_221_cast_fp16 = softmax(axis = var_2423, x = mh_w_cast_fp16)[name = string("obj_221_cast_fp16")]; + tensor var_2565 = const()[name = string("op_2565"), val = tensor([1, 12, 64, -1])]; + tensor var_2566_cast_fp16 = reshape(shape = var_2565, x = obj_217_cast_fp16)[name = string("op_2566_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_2566_cast_fp16, y = obj_221_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_2569 = const()[name = string("op_2569"), val = tensor([1, 768, 1, -1])]; + tensor input_113_cast_fp16 = reshape(shape = var_2569, x = attn_cast_fp16)[name = string("input_113_cast_fp16")]; + string obj_219_pad_type_0 = const()[name = string("obj_219_pad_type_0"), val = string("valid")]; + tensor obj_219_strides_0 = const()[name = string("obj_219_strides_0"), val = tensor([1, 1])]; + tensor obj_219_pad_0 = const()[name = string("obj_219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_219_dilations_0 = const()[name = string("obj_219_dilations_0"), val = tensor([1, 1])]; + int32 obj_219_groups_0 = const()[name = string("obj_219_groups_0"), val = int32(1)]; + tensor layers_11_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268218688)))]; + tensor layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269398400)))]; + tensor obj_219_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_219_dilations_0, groups = obj_219_groups_0, pad = obj_219_pad_0, pad_type = obj_219_pad_type_0, strides = obj_219_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = string("obj_219_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_219_cast_fp16)[name = string("inputs_71_cast_fp16")]; + tensor out_71_axes_0 = const()[name = string("out_71_axes_0"), val = tensor([1])]; + fp16 var_2587_to_fp16 = const()[name = string("op_2587_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2587_to_fp16, x = inputs_71_cast_fp16)[name = string("out_71_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = string("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269400000)))]; + tensor input_115_beta_0_to_fp16 = const()[name = string("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269401600)))]; + fp16 input_115_epsilon_0_to_fp16 = const()[name = string("input_115_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_71_cast_fp16)[name = string("input_115_cast_fp16")]; + string input_117_pad_type_0 = const()[name = string("input_117_pad_type_0"), val = string("valid")]; + tensor input_117_strides_0 = const()[name = string("input_117_strides_0"), val = tensor([1, 1])]; + tensor input_117_pad_0 = const()[name = string("input_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_117_dilations_0 = const()[name = string("input_117_dilations_0"), val = tensor([1, 1])]; + int32 input_117_groups_0 = const()[name = string("input_117_groups_0"), val = int32(1)]; + tensor layers_11_fc1_weight_to_fp16 = const()[name = string("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269403200)))]; + tensor layers_11_fc1_bias_to_fp16 = const()[name = string("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274121856)))]; + tensor input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_117_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor layers_11_fc2_weight_to_fp16 = const()[name = string("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274128064)))]; + tensor layers_11_fc2_bias_to_fp16 = const()[name = string("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278846720)))]; + tensor hidden_states_25_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_2629_to_fp16 = const()[name = string("op_2629_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2629_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278848320)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278849920)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_2640_axes_0 = const()[name = string("op_2640_axes_0"), val = tensor([2])]; + tensor var_2640_cast_fp16 = squeeze(axes = var_2640_axes_0, x = hidden_states_cast_fp16)[name = string("op_2640_cast_fp16")]; + tensor var_2643_perm_0 = const()[name = string("op_2643_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278851520)))]; + tensor var_2643_cast_fp16 = transpose(perm = var_2643_perm_0, x = var_2640_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_2643_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_2647 = const()[name = string("op_2647"), val = int32(1)]; + bool obj_225_interleave_0 = const()[name = string("obj_225_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_2647, interleave = obj_225_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_cast_fp16))[name = string("obj_225_cast_fp16")]; + int32 var_2650 = const()[name = string("op_2650"), val = int32(1)]; + bool obj_227_interleave_0 = const()[name = string("obj_227_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_2650, interleave = obj_227_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_cast_fp16))[name = string("obj_227_cast_fp16")]; + tensor var_2661_begin_0 = const()[name = string("op_2661_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2661_end_0 = const()[name = string("op_2661_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_2661_end_mask_0 = const()[name = string("op_2661_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2661_cast_fp16 = slice_by_index(begin = var_2661_begin_0, end = var_2661_end_0, end_mask = var_2661_end_mask_0, x = obj_113_cast_fp16)[name = string("op_2661_cast_fp16")]; + tensor var_2664_begin_0 = const()[name = string("op_2664_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2664_end_0 = const()[name = string("op_2664_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2664_end_mask_0 = const()[name = string("op_2664_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2664_squeeze_mask_0 = const()[name = string("op_2664_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2664_cast_fp16 = slice_by_index(begin = var_2664_begin_0, end = var_2664_end_0, end_mask = var_2664_end_mask_0, squeeze_mask = var_2664_squeeze_mask_0, x = var_2661_cast_fp16)[name = string("op_2664_cast_fp16")]; + tensor var_2679_begin_0 = const()[name = string("op_2679_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2679_end_0 = const()[name = string("op_2679_end_0"), val = tensor([1, 10, 1, 1536])]; + tensor var_2679_end_mask_0 = const()[name = string("op_2679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2679_cast_fp16 = slice_by_index(begin = var_2679_begin_0, end = var_2679_end_0, end_mask = var_2679_end_mask_0, x = obj_113_cast_fp16)[name = string("op_2679_cast_fp16")]; + tensor var_2682_begin_0 = const()[name = string("op_2682_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2682_end_0 = const()[name = string("op_2682_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2682_end_mask_0 = const()[name = string("op_2682_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2682_squeeze_mask_0 = const()[name = string("op_2682_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2682_cast_fp16 = slice_by_index(begin = var_2682_begin_0, end = var_2682_end_0, end_mask = var_2682_end_mask_0, squeeze_mask = var_2682_squeeze_mask_0, x = var_2679_cast_fp16)[name = string("op_2682_cast_fp16")]; + tensor var_2697_begin_0 = const()[name = string("op_2697_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2697_end_0 = const()[name = string("op_2697_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2697_end_mask_0 = const()[name = string("op_2697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2697_cast_fp16 = slice_by_index(begin = var_2697_begin_0, end = var_2697_end_0, end_mask = var_2697_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2697_cast_fp16")]; + tensor var_2700_begin_0 = const()[name = string("op_2700_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2700_end_0 = const()[name = string("op_2700_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2700_end_mask_0 = const()[name = string("op_2700_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2700_squeeze_mask_0 = const()[name = string("op_2700_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2700_cast_fp16 = slice_by_index(begin = var_2700_begin_0, end = var_2700_end_0, end_mask = var_2700_end_mask_0, squeeze_mask = var_2700_squeeze_mask_0, x = var_2697_cast_fp16)[name = string("op_2700_cast_fp16")]; + tensor var_2715_begin_0 = const()[name = string("op_2715_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2715_end_0 = const()[name = string("op_2715_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_2715_end_mask_0 = const()[name = string("op_2715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2715_cast_fp16 = slice_by_index(begin = var_2715_begin_0, end = var_2715_end_0, end_mask = var_2715_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2715_cast_fp16")]; + tensor var_2718_begin_0 = const()[name = string("op_2718_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2718_end_0 = const()[name = string("op_2718_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2718_end_mask_0 = const()[name = string("op_2718_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2718_squeeze_mask_0 = const()[name = string("op_2718_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2718_cast_fp16 = slice_by_index(begin = var_2718_begin_0, end = var_2718_end_0, end_mask = var_2718_end_mask_0, squeeze_mask = var_2718_squeeze_mask_0, x = var_2715_cast_fp16)[name = string("op_2718_cast_fp16")]; + tensor var_2733_begin_0 = const()[name = string("op_2733_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2733_end_0 = const()[name = string("op_2733_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_2733_end_mask_0 = const()[name = string("op_2733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2733_cast_fp16 = slice_by_index(begin = var_2733_begin_0, end = var_2733_end_0, end_mask = var_2733_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2733_cast_fp16")]; + tensor var_2736_begin_0 = const()[name = string("op_2736_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2736_end_0 = const()[name = string("op_2736_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2736_end_mask_0 = const()[name = string("op_2736_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2736_squeeze_mask_0 = const()[name = string("op_2736_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2736_cast_fp16 = slice_by_index(begin = var_2736_begin_0, end = var_2736_end_0, end_mask = var_2736_end_mask_0, squeeze_mask = var_2736_squeeze_mask_0, x = var_2733_cast_fp16)[name = string("op_2736_cast_fp16")]; + tensor var_2751_begin_0 = const()[name = string("op_2751_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2751_end_0 = const()[name = string("op_2751_end_0"), val = tensor([1, 9, 1, 1536])]; + tensor var_2751_end_mask_0 = const()[name = string("op_2751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2751_cast_fp16 = slice_by_index(begin = var_2751_begin_0, end = var_2751_end_0, end_mask = var_2751_end_mask_0, x = obj_167_cast_fp16)[name = string("op_2751_cast_fp16")]; + tensor var_2754_begin_0 = const()[name = string("op_2754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2754_end_0 = const()[name = string("op_2754_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2754_end_mask_0 = const()[name = string("op_2754_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2754_squeeze_mask_0 = const()[name = string("op_2754_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2754_cast_fp16 = slice_by_index(begin = var_2754_begin_0, end = var_2754_end_0, end_mask = var_2754_end_mask_0, squeeze_mask = var_2754_squeeze_mask_0, x = var_2751_cast_fp16)[name = string("op_2754_cast_fp16")]; + tensor var_2769_begin_0 = const()[name = string("op_2769_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2769_end_0 = const()[name = string("op_2769_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2769_end_mask_0 = const()[name = string("op_2769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2769_cast_fp16 = slice_by_index(begin = var_2769_begin_0, end = var_2769_end_0, end_mask = var_2769_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2769_cast_fp16")]; + tensor var_2772_begin_0 = const()[name = string("op_2772_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2772_end_0 = const()[name = string("op_2772_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2772_end_mask_0 = const()[name = string("op_2772_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2772_squeeze_mask_0 = const()[name = string("op_2772_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2772_cast_fp16 = slice_by_index(begin = var_2772_begin_0, end = var_2772_end_0, end_mask = var_2772_end_mask_0, squeeze_mask = var_2772_squeeze_mask_0, x = var_2769_cast_fp16)[name = string("op_2772_cast_fp16")]; + tensor var_2787_begin_0 = const()[name = string("op_2787_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2787_end_0 = const()[name = string("op_2787_end_0"), val = tensor([1, 8, 1, 1536])]; + tensor var_2787_end_mask_0 = const()[name = string("op_2787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2787_cast_fp16 = slice_by_index(begin = var_2787_begin_0, end = var_2787_end_0, end_mask = var_2787_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2787_cast_fp16")]; + tensor var_2790_begin_0 = const()[name = string("op_2790_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2790_end_0 = const()[name = string("op_2790_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2790_end_mask_0 = const()[name = string("op_2790_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2790_squeeze_mask_0 = const()[name = string("op_2790_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2790_cast_fp16 = slice_by_index(begin = var_2790_begin_0, end = var_2790_end_0, end_mask = var_2790_end_mask_0, squeeze_mask = var_2790_squeeze_mask_0, x = var_2787_cast_fp16)[name = string("op_2790_cast_fp16")]; + tensor var_2805_begin_0 = const()[name = string("op_2805_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2805_end_0 = const()[name = string("op_2805_end_0"), val = tensor([1, 10, 1, 1536])]; + tensor var_2805_end_mask_0 = const()[name = string("op_2805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2805_cast_fp16 = slice_by_index(begin = var_2805_begin_0, end = var_2805_end_0, end_mask = var_2805_end_mask_0, x = obj_185_cast_fp16)[name = string("op_2805_cast_fp16")]; + tensor var_2808_begin_0 = const()[name = string("op_2808_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2808_end_0 = const()[name = string("op_2808_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2808_end_mask_0 = const()[name = string("op_2808_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2808_squeeze_mask_0 = const()[name = string("op_2808_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2808_cast_fp16 = slice_by_index(begin = var_2808_begin_0, end = var_2808_end_0, end_mask = var_2808_end_mask_0, squeeze_mask = var_2808_squeeze_mask_0, x = var_2805_cast_fp16)[name = string("op_2808_cast_fp16")]; + tensor var_2823_begin_0 = const()[name = string("op_2823_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2823_end_0 = const()[name = string("op_2823_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_2823_end_mask_0 = const()[name = string("op_2823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2823_cast_fp16 = slice_by_index(begin = var_2823_begin_0, end = var_2823_end_0, end_mask = var_2823_end_mask_0, x = obj_203_cast_fp16)[name = string("op_2823_cast_fp16")]; + tensor var_2826_begin_0 = const()[name = string("op_2826_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2826_end_0 = const()[name = string("op_2826_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_2826_end_mask_0 = const()[name = string("op_2826_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_2826_squeeze_mask_0 = const()[name = string("op_2826_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_2826_cast_fp16 = slice_by_index(begin = var_2826_begin_0, end = var_2826_end_0, end_mask = var_2826_end_mask_0, squeeze_mask = var_2826_squeeze_mask_0, x = var_2823_cast_fp16)[name = string("op_2826_cast_fp16")]; + int32 var_2833 = const()[name = string("op_2833"), val = int32(1)]; + bool var_2834_interleave_0 = const()[name = string("op_2834_interleave_0"), val = bool(false)]; + tensor var_2834_cast_fp16 = concat(axis = var_2833, interleave = var_2834_interleave_0, values = (var_2664_cast_fp16, var_2682_cast_fp16, var_2700_cast_fp16, var_2718_cast_fp16, var_2736_cast_fp16, var_2754_cast_fp16, var_2772_cast_fp16, var_2790_cast_fp16, var_2808_cast_fp16, var_2826_cast_fp16))[name = string("op_2834_cast_fp16")]; + bool var_2837 = const()[name = string("op_2837"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_2837, x = var_2834_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-small/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-small/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..4d9e72d8e1cd93e1018fa90bcc90cd7013e0cd79 --- /dev/null +++ b/openai_whisper-small/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdb2aeabac321c24a143687f2b25a6ee459abe0d5e029028f59d2d8bb0769c8d +size 278955314 diff --git a/openai_whisper-small/config.json b/openai_whisper-small/config.json new file mode 100644 index 0000000000000000000000000000000000000000..9dee569cf0c20925208ec84fecbb95e873f8bf24 --- /dev/null +++ b/openai_whisper-small/config.json @@ -0,0 +1 @@ +{"_name_or_path": "openai/whisper-small", "activation_dropout": 0.0, "activation_function": "gelu", "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "d_model": 768, "decoder_attention_heads": 12, "decoder_ffn_dim": 3072, "decoder_layerdrop": 0.0, "decoder_layers": 12, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 12, "encoder_ffn_dim": 3072, "encoder_layerdrop": 0.0, "encoder_layers": 12, "eos_token_id": 50257, "forced_decoder_ids": [[1, 50259], [2, 50359], [3, 50363]], "init_std": 0.02, "is_encoder_decoder": true, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "model_type": "whisper", "num_hidden_layers": 12, "num_mel_bins": 80, "pad_token_id": 50257, "scale_embedding": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50360, 50361, 50362], "torch_dtype": "float32", "transformers_version": "4.27.0.dev0", "use_cache": true, "vocab_size": 51865} \ No newline at end of file diff --git a/openai_whisper-small/generation_config.json b/openai_whisper-small/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..cdd26273f9cd1ab8ecda49f5b8c033134c61cb4a --- /dev/null +++ b/openai_whisper-small/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[5, 3], [5, 9], [8, 0], [8, 4], [8, 7], [8, 8], [9, 0], [9, 7], [9, 9], [10, 5]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50359]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50363, "pad_token_id": 50257, "prev_sot_token_id": 50361, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "task_to_id": {"transcribe": 50359, "translate": 50358}, "transformers_version": "4.31.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-tiny.en/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..f115ac22a2e1abcad93ca64fd64ed778a9aeaa60 --- /dev/null +++ b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89a9aaa1f05fb2dbca4368c545605a472819dc4f6b919313ebf4b86a5e90c315 +size 243 diff --git a/openai_whisper-tiny.en/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..8b61d15f0001c9ecd279614e2f8881fbe65a7274 --- /dev/null +++ b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:924bf280f20536a89020e7dd13efdec0a41bb22ce5fb9b2f7062384417accb52 +size 433 diff --git a/openai_whisper-tiny.en/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..45385c1fd8200081ab3d24e79652954d54b9e575 --- /dev/null +++ b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 384 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 384, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 9, + "Ios18.conv" : 34, + "Ios18.gelu" : 6, + "Ios18.concat" : 30, + "Ios16.einsum" : 192, + "Ios18.add" : 9, + "Ios18.softmax" : 96, + "Ios18.sliceByIndex" : 168, + "Ios18.layerNorm" : 9, + "Ios18.transpose" : 4, + "Ios18.mul" : 96 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-tiny.en/AudioEncoder.mlmodelc/model.mil b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..5ad9dee5622f8ff5d216e86322c8cf1cdd4ccba3 --- /dev/null +++ b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,1787 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_58_pad_type_0 = const()[name = string("op_58_pad_type_0"), val = string("custom")]; + tensor var_58_pad_0 = const()[name = string("op_58_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_58_strides_0 = const()[name = string("op_58_strides_0"), val = tensor([1, 1])]; + tensor var_58_dilations_0 = const()[name = string("op_58_dilations_0"), val = tensor([1, 1])]; + int32 var_58_groups_0 = const()[name = string("op_58_groups_0"), val = int32(1)]; + tensor var_33_to_fp16 = const()[name = string("op_33_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_39_to_fp16 = const()[name = string("op_39_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184448)))]; + tensor var_58_cast_fp16 = conv(bias = var_39_to_fp16, dilations = var_58_dilations_0, groups = var_58_groups_0, pad = var_58_pad_0, pad_type = var_58_pad_type_0, strides = var_58_strides_0, weight = var_33_to_fp16, x = melspectrogram_features)[name = string("op_58_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_58_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_98_pad_type_0 = const()[name = string("op_98_pad_type_0"), val = string("custom")]; + tensor var_98_pad_0 = const()[name = string("op_98_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_98_strides_0 = const()[name = string("op_98_strides_0"), val = tensor([2, 2])]; + tensor var_98_dilations_0 = const()[name = string("op_98_dilations_0"), val = tensor([1, 1])]; + int32 var_98_groups_0 = const()[name = string("op_98_groups_0"), val = int32(1)]; + tensor var_73_to_fp16 = const()[name = string("op_73_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185280)))]; + tensor var_79_to_fp16 = const()[name = string("op_79_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070080)))]; + tensor var_98_cast_fp16 = conv(bias = var_79_to_fp16, dilations = var_98_dilations_0, groups = var_98_groups_0, pad = var_98_pad_0, pad_type = var_98_pad_type_0, strides = var_98_strides_0, weight = var_73_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_98_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_98_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_116_to_fp16 = const()[name = string("op_116_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070912)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_116_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_126 = const()[name = string("op_126"), val = int32(3)]; + int32 var_137 = const()[name = string("op_137"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_154_to_fp16 = const()[name = string("op_154_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_154_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2222976)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2223808)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2224640)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2225472)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2226304)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2521280)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2522112)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2817088)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112064)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_192_begin_0 = const()[name = string("op_192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_192_end_0 = const()[name = string("op_192_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_192_end_mask_0 = const()[name = string("op_192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_192_cast_fp16 = slice_by_index(begin = var_192_begin_0, end = var_192_end_0, end_mask = var_192_end_mask_0, x = query_1_cast_fp16)[name = string("op_192_cast_fp16")]; + tensor var_196_begin_0 = const()[name = string("op_196_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_196_end_0 = const()[name = string("op_196_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_196_end_mask_0 = const()[name = string("op_196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_196_cast_fp16 = slice_by_index(begin = var_196_begin_0, end = var_196_end_0, end_mask = var_196_end_mask_0, x = query_1_cast_fp16)[name = string("op_196_cast_fp16")]; + tensor var_200_begin_0 = const()[name = string("op_200_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_200_end_0 = const()[name = string("op_200_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_200_end_mask_0 = const()[name = string("op_200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_200_cast_fp16 = slice_by_index(begin = var_200_begin_0, end = var_200_end_0, end_mask = var_200_end_mask_0, x = query_1_cast_fp16)[name = string("op_200_cast_fp16")]; + tensor var_204_begin_0 = const()[name = string("op_204_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_204_end_0 = const()[name = string("op_204_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_204_end_mask_0 = const()[name = string("op_204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_204_cast_fp16 = slice_by_index(begin = var_204_begin_0, end = var_204_end_0, end_mask = var_204_end_mask_0, x = query_1_cast_fp16)[name = string("op_204_cast_fp16")]; + tensor var_208_begin_0 = const()[name = string("op_208_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_208_end_0 = const()[name = string("op_208_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_208_end_mask_0 = const()[name = string("op_208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_208_cast_fp16 = slice_by_index(begin = var_208_begin_0, end = var_208_end_0, end_mask = var_208_end_mask_0, x = query_1_cast_fp16)[name = string("op_208_cast_fp16")]; + tensor var_212_begin_0 = const()[name = string("op_212_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_212_end_0 = const()[name = string("op_212_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_212_end_mask_0 = const()[name = string("op_212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_212_cast_fp16 = slice_by_index(begin = var_212_begin_0, end = var_212_end_0, end_mask = var_212_end_mask_0, x = query_1_cast_fp16)[name = string("op_212_cast_fp16")]; + tensor var_221_begin_0 = const()[name = string("op_221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_221_end_0 = const()[name = string("op_221_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_221_end_mask_0 = const()[name = string("op_221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_221_cast_fp16 = slice_by_index(begin = var_221_begin_0, end = var_221_end_0, end_mask = var_221_end_mask_0, x = var_192_cast_fp16)[name = string("op_221_cast_fp16")]; + tensor var_228_begin_0 = const()[name = string("op_228_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_228_end_0 = const()[name = string("op_228_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_228_end_mask_0 = const()[name = string("op_228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_228_cast_fp16 = slice_by_index(begin = var_228_begin_0, end = var_228_end_0, end_mask = var_228_end_mask_0, x = var_192_cast_fp16)[name = string("op_228_cast_fp16")]; + tensor var_235_begin_0 = const()[name = string("op_235_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_235_end_0 = const()[name = string("op_235_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_235_end_mask_0 = const()[name = string("op_235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_235_cast_fp16 = slice_by_index(begin = var_235_begin_0, end = var_235_end_0, end_mask = var_235_end_mask_0, x = var_192_cast_fp16)[name = string("op_235_cast_fp16")]; + tensor var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = var_192_cast_fp16)[name = string("op_242_cast_fp16")]; + tensor var_249_begin_0 = const()[name = string("op_249_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_249_end_0 = const()[name = string("op_249_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_249_end_mask_0 = const()[name = string("op_249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_249_cast_fp16 = slice_by_index(begin = var_249_begin_0, end = var_249_end_0, end_mask = var_249_end_mask_0, x = var_196_cast_fp16)[name = string("op_249_cast_fp16")]; + tensor var_256_begin_0 = const()[name = string("op_256_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_256_end_0 = const()[name = string("op_256_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_256_end_mask_0 = const()[name = string("op_256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_256_cast_fp16 = slice_by_index(begin = var_256_begin_0, end = var_256_end_0, end_mask = var_256_end_mask_0, x = var_196_cast_fp16)[name = string("op_256_cast_fp16")]; + tensor var_263_begin_0 = const()[name = string("op_263_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_263_end_0 = const()[name = string("op_263_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_263_end_mask_0 = const()[name = string("op_263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_263_cast_fp16 = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = var_196_cast_fp16)[name = string("op_263_cast_fp16")]; + tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = var_196_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_277_begin_0 = const()[name = string("op_277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_277_end_0 = const()[name = string("op_277_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_277_end_mask_0 = const()[name = string("op_277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_277_cast_fp16 = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_200_cast_fp16)[name = string("op_277_cast_fp16")]; + tensor var_284_begin_0 = const()[name = string("op_284_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_284_end_0 = const()[name = string("op_284_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_284_end_mask_0 = const()[name = string("op_284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_284_cast_fp16 = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_200_cast_fp16)[name = string("op_284_cast_fp16")]; + tensor var_291_begin_0 = const()[name = string("op_291_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_291_end_0 = const()[name = string("op_291_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_291_end_mask_0 = const()[name = string("op_291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_291_cast_fp16 = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_200_cast_fp16)[name = string("op_291_cast_fp16")]; + tensor var_298_begin_0 = const()[name = string("op_298_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_298_end_0 = const()[name = string("op_298_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_298_end_mask_0 = const()[name = string("op_298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_200_cast_fp16)[name = string("op_298_cast_fp16")]; + tensor var_305_begin_0 = const()[name = string("op_305_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_305_end_0 = const()[name = string("op_305_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_305_end_mask_0 = const()[name = string("op_305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_305_cast_fp16 = slice_by_index(begin = var_305_begin_0, end = var_305_end_0, end_mask = var_305_end_mask_0, x = var_204_cast_fp16)[name = string("op_305_cast_fp16")]; + tensor var_312_begin_0 = const()[name = string("op_312_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_312_end_0 = const()[name = string("op_312_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_312_end_mask_0 = const()[name = string("op_312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_312_cast_fp16 = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = var_204_cast_fp16)[name = string("op_312_cast_fp16")]; + tensor var_319_begin_0 = const()[name = string("op_319_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_319_end_0 = const()[name = string("op_319_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_319_end_mask_0 = const()[name = string("op_319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = var_204_cast_fp16)[name = string("op_319_cast_fp16")]; + tensor var_326_begin_0 = const()[name = string("op_326_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_326_end_0 = const()[name = string("op_326_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_326_end_mask_0 = const()[name = string("op_326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = var_204_cast_fp16)[name = string("op_326_cast_fp16")]; + tensor var_333_begin_0 = const()[name = string("op_333_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_333_end_0 = const()[name = string("op_333_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_333_end_mask_0 = const()[name = string("op_333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_333_cast_fp16 = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = var_208_cast_fp16)[name = string("op_333_cast_fp16")]; + tensor var_340_begin_0 = const()[name = string("op_340_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_340_end_0 = const()[name = string("op_340_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_340_end_mask_0 = const()[name = string("op_340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_340_cast_fp16 = slice_by_index(begin = var_340_begin_0, end = var_340_end_0, end_mask = var_340_end_mask_0, x = var_208_cast_fp16)[name = string("op_340_cast_fp16")]; + tensor var_347_begin_0 = const()[name = string("op_347_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_347_end_0 = const()[name = string("op_347_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_347_end_mask_0 = const()[name = string("op_347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_347_cast_fp16 = slice_by_index(begin = var_347_begin_0, end = var_347_end_0, end_mask = var_347_end_mask_0, x = var_208_cast_fp16)[name = string("op_347_cast_fp16")]; + tensor var_354_begin_0 = const()[name = string("op_354_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_354_end_0 = const()[name = string("op_354_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_354_end_mask_0 = const()[name = string("op_354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_354_cast_fp16 = slice_by_index(begin = var_354_begin_0, end = var_354_end_0, end_mask = var_354_end_mask_0, x = var_208_cast_fp16)[name = string("op_354_cast_fp16")]; + tensor var_361_begin_0 = const()[name = string("op_361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_361_end_0 = const()[name = string("op_361_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_361_end_mask_0 = const()[name = string("op_361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_361_cast_fp16 = slice_by_index(begin = var_361_begin_0, end = var_361_end_0, end_mask = var_361_end_mask_0, x = var_212_cast_fp16)[name = string("op_361_cast_fp16")]; + tensor var_368_begin_0 = const()[name = string("op_368_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_368_end_0 = const()[name = string("op_368_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_368_end_mask_0 = const()[name = string("op_368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_368_cast_fp16 = slice_by_index(begin = var_368_begin_0, end = var_368_end_0, end_mask = var_368_end_mask_0, x = var_212_cast_fp16)[name = string("op_368_cast_fp16")]; + tensor var_375_begin_0 = const()[name = string("op_375_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_375_end_0 = const()[name = string("op_375_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_375_end_mask_0 = const()[name = string("op_375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_375_cast_fp16 = slice_by_index(begin = var_375_begin_0, end = var_375_end_0, end_mask = var_375_end_mask_0, x = var_212_cast_fp16)[name = string("op_375_cast_fp16")]; + tensor var_382_begin_0 = const()[name = string("op_382_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_382_end_0 = const()[name = string("op_382_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_382_end_mask_0 = const()[name = string("op_382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_382_cast_fp16 = slice_by_index(begin = var_382_begin_0, end = var_382_end_0, end_mask = var_382_end_mask_0, x = var_212_cast_fp16)[name = string("op_382_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_387_begin_0 = const()[name = string("op_387_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_387_end_0 = const()[name = string("op_387_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_387_end_mask_0 = const()[name = string("op_387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_3")]; + tensor var_387_cast_fp16 = slice_by_index(begin = var_387_begin_0, end = var_387_end_0, end_mask = var_387_end_mask_0, x = k_1_cast_fp16)[name = string("op_387_cast_fp16")]; + tensor var_391_begin_0 = const()[name = string("op_391_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_391_end_0 = const()[name = string("op_391_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_391_end_mask_0 = const()[name = string("op_391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_391_cast_fp16 = slice_by_index(begin = var_391_begin_0, end = var_391_end_0, end_mask = var_391_end_mask_0, x = k_1_cast_fp16)[name = string("op_391_cast_fp16")]; + tensor var_395_begin_0 = const()[name = string("op_395_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_395_end_0 = const()[name = string("op_395_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_395_end_mask_0 = const()[name = string("op_395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_395_cast_fp16 = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = k_1_cast_fp16)[name = string("op_395_cast_fp16")]; + tensor var_399_begin_0 = const()[name = string("op_399_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_399_end_0 = const()[name = string("op_399_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_399_end_mask_0 = const()[name = string("op_399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_399_cast_fp16 = slice_by_index(begin = var_399_begin_0, end = var_399_end_0, end_mask = var_399_end_mask_0, x = k_1_cast_fp16)[name = string("op_399_cast_fp16")]; + tensor var_403_begin_0 = const()[name = string("op_403_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_403_end_0 = const()[name = string("op_403_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_403_end_mask_0 = const()[name = string("op_403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_403_cast_fp16 = slice_by_index(begin = var_403_begin_0, end = var_403_end_0, end_mask = var_403_end_mask_0, x = k_1_cast_fp16)[name = string("op_403_cast_fp16")]; + tensor var_407_begin_0 = const()[name = string("op_407_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_407_end_0 = const()[name = string("op_407_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_407_end_mask_0 = const()[name = string("op_407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = k_1_cast_fp16)[name = string("op_407_cast_fp16")]; + tensor var_409_begin_0 = const()[name = string("op_409_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_409_end_0 = const()[name = string("op_409_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_409_end_mask_0 = const()[name = string("op_409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_409_cast_fp16 = slice_by_index(begin = var_409_begin_0, end = var_409_end_0, end_mask = var_409_end_mask_0, x = value_1_cast_fp16)[name = string("op_409_cast_fp16")]; + tensor var_413_begin_0 = const()[name = string("op_413_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_413_end_0 = const()[name = string("op_413_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_413_end_mask_0 = const()[name = string("op_413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, x = value_1_cast_fp16)[name = string("op_413_cast_fp16")]; + tensor var_417_begin_0 = const()[name = string("op_417_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_417_end_0 = const()[name = string("op_417_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_417_end_mask_0 = const()[name = string("op_417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_417_cast_fp16 = slice_by_index(begin = var_417_begin_0, end = var_417_end_0, end_mask = var_417_end_mask_0, x = value_1_cast_fp16)[name = string("op_417_cast_fp16")]; + tensor var_421_begin_0 = const()[name = string("op_421_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_421_end_0 = const()[name = string("op_421_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_421_end_mask_0 = const()[name = string("op_421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = value_1_cast_fp16)[name = string("op_421_cast_fp16")]; + tensor var_425_begin_0 = const()[name = string("op_425_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_425_end_0 = const()[name = string("op_425_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_425_end_mask_0 = const()[name = string("op_425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_425_cast_fp16 = slice_by_index(begin = var_425_begin_0, end = var_425_end_0, end_mask = var_425_end_mask_0, x = value_1_cast_fp16)[name = string("op_425_cast_fp16")]; + tensor var_429_begin_0 = const()[name = string("op_429_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_429_end_0 = const()[name = string("op_429_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_429_end_mask_0 = const()[name = string("op_429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_429_cast_fp16 = slice_by_index(begin = var_429_begin_0, end = var_429_end_0, end_mask = var_429_end_mask_0, x = value_1_cast_fp16)[name = string("op_429_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_387_cast_fp16, var_221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_387_cast_fp16, var_228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_387_cast_fp16, var_235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_387_cast_fp16, var_242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_391_cast_fp16, var_249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_391_cast_fp16, var_256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_391_cast_fp16, var_263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_391_cast_fp16, var_270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_395_cast_fp16, var_277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_395_cast_fp16, var_284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_395_cast_fp16, var_291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_395_cast_fp16, var_298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_399_cast_fp16, var_305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_399_cast_fp16, var_312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_399_cast_fp16, var_319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_399_cast_fp16, var_326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_403_cast_fp16, var_333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_403_cast_fp16, var_340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_403_cast_fp16, var_347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_403_cast_fp16, var_354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_407_cast_fp16, var_361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_407_cast_fp16, var_368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_407_cast_fp16, var_375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_407_cast_fp16, var_382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + fp16 var_480_to_fp16 = const()[name = string("op_480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_480_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_482_to_fp16 = const()[name = string("op_482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_482_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_484_to_fp16 = const()[name = string("op_484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_484_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_486_to_fp16 = const()[name = string("op_486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_486_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_488_to_fp16 = const()[name = string("op_488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_488_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_490_to_fp16 = const()[name = string("op_490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_490_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_492_to_fp16 = const()[name = string("op_492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_492_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_494_to_fp16 = const()[name = string("op_494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_494_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_496_to_fp16 = const()[name = string("op_496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_496_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_498_to_fp16 = const()[name = string("op_498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_498_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_500_to_fp16 = const()[name = string("op_500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_500_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_502_to_fp16 = const()[name = string("op_502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_502_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_504_to_fp16 = const()[name = string("op_504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_504_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_506_to_fp16 = const()[name = string("op_506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_506_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_508_to_fp16 = const()[name = string("op_508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_508_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_510_to_fp16 = const()[name = string("op_510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_510_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_512_to_fp16 = const()[name = string("op_512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_512_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_514_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_516_to_fp16 = const()[name = string("op_516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_516_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_518_to_fp16 = const()[name = string("op_518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_518_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_520_to_fp16 = const()[name = string("op_520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_520_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_522_to_fp16 = const()[name = string("op_522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_522_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_524_to_fp16 = const()[name = string("op_524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_524_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_526_to_fp16 = const()[name = string("op_526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_526_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + tensor var_528_cast_fp16 = softmax(axis = var_137, x = aw_chunk_1_cast_fp16)[name = string("op_528_cast_fp16")]; + tensor var_529_cast_fp16 = softmax(axis = var_137, x = aw_chunk_3_cast_fp16)[name = string("op_529_cast_fp16")]; + tensor var_530_cast_fp16 = softmax(axis = var_137, x = aw_chunk_5_cast_fp16)[name = string("op_530_cast_fp16")]; + tensor var_531_cast_fp16 = softmax(axis = var_137, x = aw_chunk_7_cast_fp16)[name = string("op_531_cast_fp16")]; + tensor var_532_cast_fp16 = softmax(axis = var_137, x = aw_chunk_9_cast_fp16)[name = string("op_532_cast_fp16")]; + tensor var_533_cast_fp16 = softmax(axis = var_137, x = aw_chunk_11_cast_fp16)[name = string("op_533_cast_fp16")]; + tensor var_534_cast_fp16 = softmax(axis = var_137, x = aw_chunk_13_cast_fp16)[name = string("op_534_cast_fp16")]; + tensor var_535_cast_fp16 = softmax(axis = var_137, x = aw_chunk_15_cast_fp16)[name = string("op_535_cast_fp16")]; + tensor var_536_cast_fp16 = softmax(axis = var_137, x = aw_chunk_17_cast_fp16)[name = string("op_536_cast_fp16")]; + tensor var_537_cast_fp16 = softmax(axis = var_137, x = aw_chunk_19_cast_fp16)[name = string("op_537_cast_fp16")]; + tensor var_538_cast_fp16 = softmax(axis = var_137, x = aw_chunk_21_cast_fp16)[name = string("op_538_cast_fp16")]; + tensor var_539_cast_fp16 = softmax(axis = var_137, x = aw_chunk_23_cast_fp16)[name = string("op_539_cast_fp16")]; + tensor var_540_cast_fp16 = softmax(axis = var_137, x = aw_chunk_25_cast_fp16)[name = string("op_540_cast_fp16")]; + tensor var_541_cast_fp16 = softmax(axis = var_137, x = aw_chunk_27_cast_fp16)[name = string("op_541_cast_fp16")]; + tensor var_542_cast_fp16 = softmax(axis = var_137, x = aw_chunk_29_cast_fp16)[name = string("op_542_cast_fp16")]; + tensor var_543_cast_fp16 = softmax(axis = var_137, x = aw_chunk_31_cast_fp16)[name = string("op_543_cast_fp16")]; + tensor var_544_cast_fp16 = softmax(axis = var_137, x = aw_chunk_33_cast_fp16)[name = string("op_544_cast_fp16")]; + tensor var_545_cast_fp16 = softmax(axis = var_137, x = aw_chunk_35_cast_fp16)[name = string("op_545_cast_fp16")]; + tensor var_546_cast_fp16 = softmax(axis = var_137, x = aw_chunk_37_cast_fp16)[name = string("op_546_cast_fp16")]; + tensor var_547_cast_fp16 = softmax(axis = var_137, x = aw_chunk_39_cast_fp16)[name = string("op_547_cast_fp16")]; + tensor var_548_cast_fp16 = softmax(axis = var_137, x = aw_chunk_41_cast_fp16)[name = string("op_548_cast_fp16")]; + tensor var_549_cast_fp16 = softmax(axis = var_137, x = aw_chunk_43_cast_fp16)[name = string("op_549_cast_fp16")]; + tensor var_550_cast_fp16 = softmax(axis = var_137, x = aw_chunk_45_cast_fp16)[name = string("op_550_cast_fp16")]; + tensor var_551_cast_fp16 = softmax(axis = var_137, x = aw_chunk_47_cast_fp16)[name = string("op_551_cast_fp16")]; + string var_553_equation_0 = const()[name = string("op_553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_553_cast_fp16 = einsum(equation = var_553_equation_0, values = (var_409_cast_fp16, var_528_cast_fp16))[name = string("op_553_cast_fp16")]; + string var_555_equation_0 = const()[name = string("op_555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_409_cast_fp16, var_529_cast_fp16))[name = string("op_555_cast_fp16")]; + string var_557_equation_0 = const()[name = string("op_557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_557_cast_fp16 = einsum(equation = var_557_equation_0, values = (var_409_cast_fp16, var_530_cast_fp16))[name = string("op_557_cast_fp16")]; + string var_559_equation_0 = const()[name = string("op_559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_409_cast_fp16, var_531_cast_fp16))[name = string("op_559_cast_fp16")]; + string var_561_equation_0 = const()[name = string("op_561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_561_cast_fp16 = einsum(equation = var_561_equation_0, values = (var_413_cast_fp16, var_532_cast_fp16))[name = string("op_561_cast_fp16")]; + string var_563_equation_0 = const()[name = string("op_563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_413_cast_fp16, var_533_cast_fp16))[name = string("op_563_cast_fp16")]; + string var_565_equation_0 = const()[name = string("op_565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_565_cast_fp16 = einsum(equation = var_565_equation_0, values = (var_413_cast_fp16, var_534_cast_fp16))[name = string("op_565_cast_fp16")]; + string var_567_equation_0 = const()[name = string("op_567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_413_cast_fp16, var_535_cast_fp16))[name = string("op_567_cast_fp16")]; + string var_569_equation_0 = const()[name = string("op_569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_569_cast_fp16 = einsum(equation = var_569_equation_0, values = (var_417_cast_fp16, var_536_cast_fp16))[name = string("op_569_cast_fp16")]; + string var_571_equation_0 = const()[name = string("op_571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_417_cast_fp16, var_537_cast_fp16))[name = string("op_571_cast_fp16")]; + string var_573_equation_0 = const()[name = string("op_573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_573_cast_fp16 = einsum(equation = var_573_equation_0, values = (var_417_cast_fp16, var_538_cast_fp16))[name = string("op_573_cast_fp16")]; + string var_575_equation_0 = const()[name = string("op_575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_417_cast_fp16, var_539_cast_fp16))[name = string("op_575_cast_fp16")]; + string var_577_equation_0 = const()[name = string("op_577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_577_cast_fp16 = einsum(equation = var_577_equation_0, values = (var_421_cast_fp16, var_540_cast_fp16))[name = string("op_577_cast_fp16")]; + string var_579_equation_0 = const()[name = string("op_579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_579_cast_fp16 = einsum(equation = var_579_equation_0, values = (var_421_cast_fp16, var_541_cast_fp16))[name = string("op_579_cast_fp16")]; + string var_581_equation_0 = const()[name = string("op_581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_581_cast_fp16 = einsum(equation = var_581_equation_0, values = (var_421_cast_fp16, var_542_cast_fp16))[name = string("op_581_cast_fp16")]; + string var_583_equation_0 = const()[name = string("op_583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_583_cast_fp16 = einsum(equation = var_583_equation_0, values = (var_421_cast_fp16, var_543_cast_fp16))[name = string("op_583_cast_fp16")]; + string var_585_equation_0 = const()[name = string("op_585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_585_cast_fp16 = einsum(equation = var_585_equation_0, values = (var_425_cast_fp16, var_544_cast_fp16))[name = string("op_585_cast_fp16")]; + string var_587_equation_0 = const()[name = string("op_587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_587_cast_fp16 = einsum(equation = var_587_equation_0, values = (var_425_cast_fp16, var_545_cast_fp16))[name = string("op_587_cast_fp16")]; + string var_589_equation_0 = const()[name = string("op_589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_589_cast_fp16 = einsum(equation = var_589_equation_0, values = (var_425_cast_fp16, var_546_cast_fp16))[name = string("op_589_cast_fp16")]; + string var_591_equation_0 = const()[name = string("op_591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_591_cast_fp16 = einsum(equation = var_591_equation_0, values = (var_425_cast_fp16, var_547_cast_fp16))[name = string("op_591_cast_fp16")]; + string var_593_equation_0 = const()[name = string("op_593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_593_cast_fp16 = einsum(equation = var_593_equation_0, values = (var_429_cast_fp16, var_548_cast_fp16))[name = string("op_593_cast_fp16")]; + string var_595_equation_0 = const()[name = string("op_595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_595_cast_fp16 = einsum(equation = var_595_equation_0, values = (var_429_cast_fp16, var_549_cast_fp16))[name = string("op_595_cast_fp16")]; + string var_597_equation_0 = const()[name = string("op_597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_597_cast_fp16 = einsum(equation = var_597_equation_0, values = (var_429_cast_fp16, var_550_cast_fp16))[name = string("op_597_cast_fp16")]; + string var_599_equation_0 = const()[name = string("op_599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_599_cast_fp16 = einsum(equation = var_599_equation_0, values = (var_429_cast_fp16, var_551_cast_fp16))[name = string("op_599_cast_fp16")]; + bool var_601_interleave_0 = const()[name = string("op_601_interleave_0"), val = bool(false)]; + tensor var_601_cast_fp16 = concat(axis = var_126, interleave = var_601_interleave_0, values = (var_553_cast_fp16, var_555_cast_fp16, var_557_cast_fp16, var_559_cast_fp16))[name = string("op_601_cast_fp16")]; + bool var_603_interleave_0 = const()[name = string("op_603_interleave_0"), val = bool(false)]; + tensor var_603_cast_fp16 = concat(axis = var_126, interleave = var_603_interleave_0, values = (var_561_cast_fp16, var_563_cast_fp16, var_565_cast_fp16, var_567_cast_fp16))[name = string("op_603_cast_fp16")]; + bool var_605_interleave_0 = const()[name = string("op_605_interleave_0"), val = bool(false)]; + tensor var_605_cast_fp16 = concat(axis = var_126, interleave = var_605_interleave_0, values = (var_569_cast_fp16, var_571_cast_fp16, var_573_cast_fp16, var_575_cast_fp16))[name = string("op_605_cast_fp16")]; + bool var_607_interleave_0 = const()[name = string("op_607_interleave_0"), val = bool(false)]; + tensor var_607_cast_fp16 = concat(axis = var_126, interleave = var_607_interleave_0, values = (var_577_cast_fp16, var_579_cast_fp16, var_581_cast_fp16, var_583_cast_fp16))[name = string("op_607_cast_fp16")]; + bool var_609_interleave_0 = const()[name = string("op_609_interleave_0"), val = bool(false)]; + tensor var_609_cast_fp16 = concat(axis = var_126, interleave = var_609_interleave_0, values = (var_585_cast_fp16, var_587_cast_fp16, var_589_cast_fp16, var_591_cast_fp16))[name = string("op_609_cast_fp16")]; + bool var_611_interleave_0 = const()[name = string("op_611_interleave_0"), val = bool(false)]; + tensor var_611_cast_fp16 = concat(axis = var_126, interleave = var_611_interleave_0, values = (var_593_cast_fp16, var_595_cast_fp16, var_597_cast_fp16, var_599_cast_fp16))[name = string("op_611_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_137, interleave = input_1_interleave_0, values = (var_601_cast_fp16, var_603_cast_fp16, var_605_cast_fp16, var_607_cast_fp16, var_609_cast_fp16, var_611_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112896)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407872)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_630_to_fp16 = const()[name = string("op_630_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_630_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3408704)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3409536)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3410368)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4590080)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4593216)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772928)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_659 = const()[name = string("op_659"), val = int32(3)]; + int32 var_670 = const()[name = string("op_670"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_687_to_fp16 = const()[name = string("op_687_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_687_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5773760)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5774592)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5775424)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6070400)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6071232)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6366208)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6661184)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_725_begin_0 = const()[name = string("op_725_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_725_end_0 = const()[name = string("op_725_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_725_end_mask_0 = const()[name = string("op_725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_725_cast_fp16 = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = query_3_cast_fp16)[name = string("op_725_cast_fp16")]; + tensor var_729_begin_0 = const()[name = string("op_729_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_729_end_0 = const()[name = string("op_729_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_729_end_mask_0 = const()[name = string("op_729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_729_cast_fp16 = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = query_3_cast_fp16)[name = string("op_729_cast_fp16")]; + tensor var_733_begin_0 = const()[name = string("op_733_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_733_end_0 = const()[name = string("op_733_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_733_end_mask_0 = const()[name = string("op_733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_733_cast_fp16 = slice_by_index(begin = var_733_begin_0, end = var_733_end_0, end_mask = var_733_end_mask_0, x = query_3_cast_fp16)[name = string("op_733_cast_fp16")]; + tensor var_737_begin_0 = const()[name = string("op_737_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_737_end_0 = const()[name = string("op_737_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_737_end_mask_0 = const()[name = string("op_737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_737_cast_fp16 = slice_by_index(begin = var_737_begin_0, end = var_737_end_0, end_mask = var_737_end_mask_0, x = query_3_cast_fp16)[name = string("op_737_cast_fp16")]; + tensor var_741_begin_0 = const()[name = string("op_741_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_741_end_0 = const()[name = string("op_741_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_741_end_mask_0 = const()[name = string("op_741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_741_cast_fp16 = slice_by_index(begin = var_741_begin_0, end = var_741_end_0, end_mask = var_741_end_mask_0, x = query_3_cast_fp16)[name = string("op_741_cast_fp16")]; + tensor var_745_begin_0 = const()[name = string("op_745_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_745_end_0 = const()[name = string("op_745_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_745_end_mask_0 = const()[name = string("op_745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_745_cast_fp16 = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = query_3_cast_fp16)[name = string("op_745_cast_fp16")]; + tensor var_754_begin_0 = const()[name = string("op_754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_754_end_0 = const()[name = string("op_754_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_754_end_mask_0 = const()[name = string("op_754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_754_cast_fp16 = slice_by_index(begin = var_754_begin_0, end = var_754_end_0, end_mask = var_754_end_mask_0, x = var_725_cast_fp16)[name = string("op_754_cast_fp16")]; + tensor var_761_begin_0 = const()[name = string("op_761_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_761_end_0 = const()[name = string("op_761_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_761_end_mask_0 = const()[name = string("op_761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_761_cast_fp16 = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = var_725_cast_fp16)[name = string("op_761_cast_fp16")]; + tensor var_768_begin_0 = const()[name = string("op_768_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_768_end_0 = const()[name = string("op_768_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_768_end_mask_0 = const()[name = string("op_768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_768_cast_fp16 = slice_by_index(begin = var_768_begin_0, end = var_768_end_0, end_mask = var_768_end_mask_0, x = var_725_cast_fp16)[name = string("op_768_cast_fp16")]; + tensor var_775_begin_0 = const()[name = string("op_775_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_775_end_0 = const()[name = string("op_775_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_775_end_mask_0 = const()[name = string("op_775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_775_cast_fp16 = slice_by_index(begin = var_775_begin_0, end = var_775_end_0, end_mask = var_775_end_mask_0, x = var_725_cast_fp16)[name = string("op_775_cast_fp16")]; + tensor var_782_begin_0 = const()[name = string("op_782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_782_end_0 = const()[name = string("op_782_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_782_end_mask_0 = const()[name = string("op_782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_782_cast_fp16 = slice_by_index(begin = var_782_begin_0, end = var_782_end_0, end_mask = var_782_end_mask_0, x = var_729_cast_fp16)[name = string("op_782_cast_fp16")]; + tensor var_789_begin_0 = const()[name = string("op_789_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_789_end_0 = const()[name = string("op_789_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_789_end_mask_0 = const()[name = string("op_789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_789_cast_fp16 = slice_by_index(begin = var_789_begin_0, end = var_789_end_0, end_mask = var_789_end_mask_0, x = var_729_cast_fp16)[name = string("op_789_cast_fp16")]; + tensor var_796_begin_0 = const()[name = string("op_796_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_796_end_0 = const()[name = string("op_796_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_796_end_mask_0 = const()[name = string("op_796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_796_cast_fp16 = slice_by_index(begin = var_796_begin_0, end = var_796_end_0, end_mask = var_796_end_mask_0, x = var_729_cast_fp16)[name = string("op_796_cast_fp16")]; + tensor var_803_begin_0 = const()[name = string("op_803_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_803_end_0 = const()[name = string("op_803_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_803_end_mask_0 = const()[name = string("op_803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_803_cast_fp16 = slice_by_index(begin = var_803_begin_0, end = var_803_end_0, end_mask = var_803_end_mask_0, x = var_729_cast_fp16)[name = string("op_803_cast_fp16")]; + tensor var_810_begin_0 = const()[name = string("op_810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_810_end_0 = const()[name = string("op_810_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_810_end_mask_0 = const()[name = string("op_810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_810_cast_fp16 = slice_by_index(begin = var_810_begin_0, end = var_810_end_0, end_mask = var_810_end_mask_0, x = var_733_cast_fp16)[name = string("op_810_cast_fp16")]; + tensor var_817_begin_0 = const()[name = string("op_817_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_817_end_0 = const()[name = string("op_817_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_817_end_mask_0 = const()[name = string("op_817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_817_cast_fp16 = slice_by_index(begin = var_817_begin_0, end = var_817_end_0, end_mask = var_817_end_mask_0, x = var_733_cast_fp16)[name = string("op_817_cast_fp16")]; + tensor var_824_begin_0 = const()[name = string("op_824_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_824_end_0 = const()[name = string("op_824_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_824_end_mask_0 = const()[name = string("op_824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_824_cast_fp16 = slice_by_index(begin = var_824_begin_0, end = var_824_end_0, end_mask = var_824_end_mask_0, x = var_733_cast_fp16)[name = string("op_824_cast_fp16")]; + tensor var_831_begin_0 = const()[name = string("op_831_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_831_end_0 = const()[name = string("op_831_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_831_end_mask_0 = const()[name = string("op_831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_831_cast_fp16 = slice_by_index(begin = var_831_begin_0, end = var_831_end_0, end_mask = var_831_end_mask_0, x = var_733_cast_fp16)[name = string("op_831_cast_fp16")]; + tensor var_838_begin_0 = const()[name = string("op_838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_838_end_0 = const()[name = string("op_838_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_838_end_mask_0 = const()[name = string("op_838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_838_cast_fp16 = slice_by_index(begin = var_838_begin_0, end = var_838_end_0, end_mask = var_838_end_mask_0, x = var_737_cast_fp16)[name = string("op_838_cast_fp16")]; + tensor var_845_begin_0 = const()[name = string("op_845_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_845_end_0 = const()[name = string("op_845_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_845_end_mask_0 = const()[name = string("op_845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_845_cast_fp16 = slice_by_index(begin = var_845_begin_0, end = var_845_end_0, end_mask = var_845_end_mask_0, x = var_737_cast_fp16)[name = string("op_845_cast_fp16")]; + tensor var_852_begin_0 = const()[name = string("op_852_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_852_end_0 = const()[name = string("op_852_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_852_end_mask_0 = const()[name = string("op_852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_852_cast_fp16 = slice_by_index(begin = var_852_begin_0, end = var_852_end_0, end_mask = var_852_end_mask_0, x = var_737_cast_fp16)[name = string("op_852_cast_fp16")]; + tensor var_859_begin_0 = const()[name = string("op_859_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_859_end_0 = const()[name = string("op_859_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_859_end_mask_0 = const()[name = string("op_859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_859_cast_fp16 = slice_by_index(begin = var_859_begin_0, end = var_859_end_0, end_mask = var_859_end_mask_0, x = var_737_cast_fp16)[name = string("op_859_cast_fp16")]; + tensor var_866_begin_0 = const()[name = string("op_866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_866_end_0 = const()[name = string("op_866_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_866_end_mask_0 = const()[name = string("op_866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_866_cast_fp16 = slice_by_index(begin = var_866_begin_0, end = var_866_end_0, end_mask = var_866_end_mask_0, x = var_741_cast_fp16)[name = string("op_866_cast_fp16")]; + tensor var_873_begin_0 = const()[name = string("op_873_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_873_end_0 = const()[name = string("op_873_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_873_end_mask_0 = const()[name = string("op_873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_873_cast_fp16 = slice_by_index(begin = var_873_begin_0, end = var_873_end_0, end_mask = var_873_end_mask_0, x = var_741_cast_fp16)[name = string("op_873_cast_fp16")]; + tensor var_880_begin_0 = const()[name = string("op_880_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_880_end_0 = const()[name = string("op_880_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_880_end_mask_0 = const()[name = string("op_880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_880_cast_fp16 = slice_by_index(begin = var_880_begin_0, end = var_880_end_0, end_mask = var_880_end_mask_0, x = var_741_cast_fp16)[name = string("op_880_cast_fp16")]; + tensor var_887_begin_0 = const()[name = string("op_887_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_887_end_0 = const()[name = string("op_887_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_887_end_mask_0 = const()[name = string("op_887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_887_cast_fp16 = slice_by_index(begin = var_887_begin_0, end = var_887_end_0, end_mask = var_887_end_mask_0, x = var_741_cast_fp16)[name = string("op_887_cast_fp16")]; + tensor var_894_begin_0 = const()[name = string("op_894_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_894_end_0 = const()[name = string("op_894_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_894_end_mask_0 = const()[name = string("op_894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_894_cast_fp16 = slice_by_index(begin = var_894_begin_0, end = var_894_end_0, end_mask = var_894_end_mask_0, x = var_745_cast_fp16)[name = string("op_894_cast_fp16")]; + tensor var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = var_745_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_908_begin_0 = const()[name = string("op_908_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_908_end_0 = const()[name = string("op_908_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_908_end_mask_0 = const()[name = string("op_908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_908_cast_fp16 = slice_by_index(begin = var_908_begin_0, end = var_908_end_0, end_mask = var_908_end_mask_0, x = var_745_cast_fp16)[name = string("op_908_cast_fp16")]; + tensor var_915_begin_0 = const()[name = string("op_915_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_915_end_0 = const()[name = string("op_915_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_915_end_mask_0 = const()[name = string("op_915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_915_cast_fp16 = slice_by_index(begin = var_915_begin_0, end = var_915_end_0, end_mask = var_915_end_mask_0, x = var_745_cast_fp16)[name = string("op_915_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_920_begin_0 = const()[name = string("op_920_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_920_end_0 = const()[name = string("op_920_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_920_end_mask_0 = const()[name = string("op_920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_2")]; + tensor var_920_cast_fp16 = slice_by_index(begin = var_920_begin_0, end = var_920_end_0, end_mask = var_920_end_mask_0, x = k_3_cast_fp16)[name = string("op_920_cast_fp16")]; + tensor var_924_begin_0 = const()[name = string("op_924_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_924_end_0 = const()[name = string("op_924_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_924_end_mask_0 = const()[name = string("op_924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_924_cast_fp16 = slice_by_index(begin = var_924_begin_0, end = var_924_end_0, end_mask = var_924_end_mask_0, x = k_3_cast_fp16)[name = string("op_924_cast_fp16")]; + tensor var_928_begin_0 = const()[name = string("op_928_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_928_end_0 = const()[name = string("op_928_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_928_end_mask_0 = const()[name = string("op_928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_928_cast_fp16 = slice_by_index(begin = var_928_begin_0, end = var_928_end_0, end_mask = var_928_end_mask_0, x = k_3_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_932_begin_0 = const()[name = string("op_932_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_932_end_0 = const()[name = string("op_932_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_932_end_mask_0 = const()[name = string("op_932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_932_cast_fp16 = slice_by_index(begin = var_932_begin_0, end = var_932_end_0, end_mask = var_932_end_mask_0, x = k_3_cast_fp16)[name = string("op_932_cast_fp16")]; + tensor var_936_begin_0 = const()[name = string("op_936_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_936_end_0 = const()[name = string("op_936_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_936_end_mask_0 = const()[name = string("op_936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_936_cast_fp16 = slice_by_index(begin = var_936_begin_0, end = var_936_end_0, end_mask = var_936_end_mask_0, x = k_3_cast_fp16)[name = string("op_936_cast_fp16")]; + tensor var_940_begin_0 = const()[name = string("op_940_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_940_end_0 = const()[name = string("op_940_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_940_end_mask_0 = const()[name = string("op_940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_940_cast_fp16 = slice_by_index(begin = var_940_begin_0, end = var_940_end_0, end_mask = var_940_end_mask_0, x = k_3_cast_fp16)[name = string("op_940_cast_fp16")]; + tensor var_942_begin_0 = const()[name = string("op_942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_942_end_0 = const()[name = string("op_942_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_942_end_mask_0 = const()[name = string("op_942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = value_3_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_946_begin_0 = const()[name = string("op_946_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_946_end_0 = const()[name = string("op_946_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_946_end_mask_0 = const()[name = string("op_946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_946_cast_fp16 = slice_by_index(begin = var_946_begin_0, end = var_946_end_0, end_mask = var_946_end_mask_0, x = value_3_cast_fp16)[name = string("op_946_cast_fp16")]; + tensor var_950_begin_0 = const()[name = string("op_950_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_950_end_0 = const()[name = string("op_950_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_950_end_mask_0 = const()[name = string("op_950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_950_cast_fp16 = slice_by_index(begin = var_950_begin_0, end = var_950_end_0, end_mask = var_950_end_mask_0, x = value_3_cast_fp16)[name = string("op_950_cast_fp16")]; + tensor var_954_begin_0 = const()[name = string("op_954_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_954_end_0 = const()[name = string("op_954_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_954_end_mask_0 = const()[name = string("op_954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_954_cast_fp16 = slice_by_index(begin = var_954_begin_0, end = var_954_end_0, end_mask = var_954_end_mask_0, x = value_3_cast_fp16)[name = string("op_954_cast_fp16")]; + tensor var_958_begin_0 = const()[name = string("op_958_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_958_end_0 = const()[name = string("op_958_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_958_end_mask_0 = const()[name = string("op_958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_958_cast_fp16 = slice_by_index(begin = var_958_begin_0, end = var_958_end_0, end_mask = var_958_end_mask_0, x = value_3_cast_fp16)[name = string("op_958_cast_fp16")]; + tensor var_962_begin_0 = const()[name = string("op_962_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_962_end_0 = const()[name = string("op_962_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_962_end_mask_0 = const()[name = string("op_962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, x = value_3_cast_fp16)[name = string("op_962_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_920_cast_fp16, var_754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_920_cast_fp16, var_761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_920_cast_fp16, var_768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_920_cast_fp16, var_775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_924_cast_fp16, var_782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_924_cast_fp16, var_789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_924_cast_fp16, var_796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_924_cast_fp16, var_803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_928_cast_fp16, var_810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_928_cast_fp16, var_817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_928_cast_fp16, var_824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_928_cast_fp16, var_831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_932_cast_fp16, var_838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_932_cast_fp16, var_845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_932_cast_fp16, var_852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_932_cast_fp16, var_859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_936_cast_fp16, var_866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_936_cast_fp16, var_873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_936_cast_fp16, var_880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_936_cast_fp16, var_887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_940_cast_fp16, var_894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_940_cast_fp16, var_901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_940_cast_fp16, var_908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_940_cast_fp16, var_915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + fp16 var_1013_to_fp16 = const()[name = string("op_1013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_1013_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_1015_to_fp16 = const()[name = string("op_1015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_1015_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_1017_to_fp16 = const()[name = string("op_1017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_1017_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_1019_to_fp16 = const()[name = string("op_1019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_1019_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_1021_to_fp16 = const()[name = string("op_1021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_1021_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_1023_to_fp16 = const()[name = string("op_1023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_1023_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_1025_to_fp16 = const()[name = string("op_1025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_1025_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_1027_to_fp16 = const()[name = string("op_1027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_1027_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + fp16 var_1029_to_fp16 = const()[name = string("op_1029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1029_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1031_to_fp16 = const()[name = string("op_1031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1031_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1033_to_fp16 = const()[name = string("op_1033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1033_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1035_to_fp16 = const()[name = string("op_1035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1035_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1037_to_fp16 = const()[name = string("op_1037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1037_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1039_to_fp16 = const()[name = string("op_1039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1039_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1041_to_fp16 = const()[name = string("op_1041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1041_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1043_to_fp16 = const()[name = string("op_1043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1043_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1045_to_fp16 = const()[name = string("op_1045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1045_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1047_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1049_to_fp16 = const()[name = string("op_1049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1049_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1051_to_fp16 = const()[name = string("op_1051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1051_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1053_to_fp16 = const()[name = string("op_1053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1053_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1055_to_fp16 = const()[name = string("op_1055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1055_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1057_to_fp16 = const()[name = string("op_1057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1057_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1059_to_fp16 = const()[name = string("op_1059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1059_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + tensor var_1061_cast_fp16 = softmax(axis = var_670, x = aw_chunk_49_cast_fp16)[name = string("op_1061_cast_fp16")]; + tensor var_1062_cast_fp16 = softmax(axis = var_670, x = aw_chunk_51_cast_fp16)[name = string("op_1062_cast_fp16")]; + tensor var_1063_cast_fp16 = softmax(axis = var_670, x = aw_chunk_53_cast_fp16)[name = string("op_1063_cast_fp16")]; + tensor var_1064_cast_fp16 = softmax(axis = var_670, x = aw_chunk_55_cast_fp16)[name = string("op_1064_cast_fp16")]; + tensor var_1065_cast_fp16 = softmax(axis = var_670, x = aw_chunk_57_cast_fp16)[name = string("op_1065_cast_fp16")]; + tensor var_1066_cast_fp16 = softmax(axis = var_670, x = aw_chunk_59_cast_fp16)[name = string("op_1066_cast_fp16")]; + tensor var_1067_cast_fp16 = softmax(axis = var_670, x = aw_chunk_61_cast_fp16)[name = string("op_1067_cast_fp16")]; + tensor var_1068_cast_fp16 = softmax(axis = var_670, x = aw_chunk_63_cast_fp16)[name = string("op_1068_cast_fp16")]; + tensor var_1069_cast_fp16 = softmax(axis = var_670, x = aw_chunk_65_cast_fp16)[name = string("op_1069_cast_fp16")]; + tensor var_1070_cast_fp16 = softmax(axis = var_670, x = aw_chunk_67_cast_fp16)[name = string("op_1070_cast_fp16")]; + tensor var_1071_cast_fp16 = softmax(axis = var_670, x = aw_chunk_69_cast_fp16)[name = string("op_1071_cast_fp16")]; + tensor var_1072_cast_fp16 = softmax(axis = var_670, x = aw_chunk_71_cast_fp16)[name = string("op_1072_cast_fp16")]; + tensor var_1073_cast_fp16 = softmax(axis = var_670, x = aw_chunk_73_cast_fp16)[name = string("op_1073_cast_fp16")]; + tensor var_1074_cast_fp16 = softmax(axis = var_670, x = aw_chunk_75_cast_fp16)[name = string("op_1074_cast_fp16")]; + tensor var_1075_cast_fp16 = softmax(axis = var_670, x = aw_chunk_77_cast_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1076_cast_fp16 = softmax(axis = var_670, x = aw_chunk_79_cast_fp16)[name = string("op_1076_cast_fp16")]; + tensor var_1077_cast_fp16 = softmax(axis = var_670, x = aw_chunk_81_cast_fp16)[name = string("op_1077_cast_fp16")]; + tensor var_1078_cast_fp16 = softmax(axis = var_670, x = aw_chunk_83_cast_fp16)[name = string("op_1078_cast_fp16")]; + tensor var_1079_cast_fp16 = softmax(axis = var_670, x = aw_chunk_85_cast_fp16)[name = string("op_1079_cast_fp16")]; + tensor var_1080_cast_fp16 = softmax(axis = var_670, x = aw_chunk_87_cast_fp16)[name = string("op_1080_cast_fp16")]; + tensor var_1081_cast_fp16 = softmax(axis = var_670, x = aw_chunk_89_cast_fp16)[name = string("op_1081_cast_fp16")]; + tensor var_1082_cast_fp16 = softmax(axis = var_670, x = aw_chunk_91_cast_fp16)[name = string("op_1082_cast_fp16")]; + tensor var_1083_cast_fp16 = softmax(axis = var_670, x = aw_chunk_93_cast_fp16)[name = string("op_1083_cast_fp16")]; + tensor var_1084_cast_fp16 = softmax(axis = var_670, x = aw_chunk_95_cast_fp16)[name = string("op_1084_cast_fp16")]; + string var_1086_equation_0 = const()[name = string("op_1086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1086_cast_fp16 = einsum(equation = var_1086_equation_0, values = (var_942_cast_fp16, var_1061_cast_fp16))[name = string("op_1086_cast_fp16")]; + string var_1088_equation_0 = const()[name = string("op_1088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1088_cast_fp16 = einsum(equation = var_1088_equation_0, values = (var_942_cast_fp16, var_1062_cast_fp16))[name = string("op_1088_cast_fp16")]; + string var_1090_equation_0 = const()[name = string("op_1090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1090_cast_fp16 = einsum(equation = var_1090_equation_0, values = (var_942_cast_fp16, var_1063_cast_fp16))[name = string("op_1090_cast_fp16")]; + string var_1092_equation_0 = const()[name = string("op_1092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1092_cast_fp16 = einsum(equation = var_1092_equation_0, values = (var_942_cast_fp16, var_1064_cast_fp16))[name = string("op_1092_cast_fp16")]; + string var_1094_equation_0 = const()[name = string("op_1094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1094_cast_fp16 = einsum(equation = var_1094_equation_0, values = (var_946_cast_fp16, var_1065_cast_fp16))[name = string("op_1094_cast_fp16")]; + string var_1096_equation_0 = const()[name = string("op_1096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1096_cast_fp16 = einsum(equation = var_1096_equation_0, values = (var_946_cast_fp16, var_1066_cast_fp16))[name = string("op_1096_cast_fp16")]; + string var_1098_equation_0 = const()[name = string("op_1098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1098_cast_fp16 = einsum(equation = var_1098_equation_0, values = (var_946_cast_fp16, var_1067_cast_fp16))[name = string("op_1098_cast_fp16")]; + string var_1100_equation_0 = const()[name = string("op_1100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1100_cast_fp16 = einsum(equation = var_1100_equation_0, values = (var_946_cast_fp16, var_1068_cast_fp16))[name = string("op_1100_cast_fp16")]; + string var_1102_equation_0 = const()[name = string("op_1102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1102_cast_fp16 = einsum(equation = var_1102_equation_0, values = (var_950_cast_fp16, var_1069_cast_fp16))[name = string("op_1102_cast_fp16")]; + string var_1104_equation_0 = const()[name = string("op_1104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1104_cast_fp16 = einsum(equation = var_1104_equation_0, values = (var_950_cast_fp16, var_1070_cast_fp16))[name = string("op_1104_cast_fp16")]; + string var_1106_equation_0 = const()[name = string("op_1106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1106_cast_fp16 = einsum(equation = var_1106_equation_0, values = (var_950_cast_fp16, var_1071_cast_fp16))[name = string("op_1106_cast_fp16")]; + string var_1108_equation_0 = const()[name = string("op_1108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1108_cast_fp16 = einsum(equation = var_1108_equation_0, values = (var_950_cast_fp16, var_1072_cast_fp16))[name = string("op_1108_cast_fp16")]; + string var_1110_equation_0 = const()[name = string("op_1110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1110_cast_fp16 = einsum(equation = var_1110_equation_0, values = (var_954_cast_fp16, var_1073_cast_fp16))[name = string("op_1110_cast_fp16")]; + string var_1112_equation_0 = const()[name = string("op_1112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1112_cast_fp16 = einsum(equation = var_1112_equation_0, values = (var_954_cast_fp16, var_1074_cast_fp16))[name = string("op_1112_cast_fp16")]; + string var_1114_equation_0 = const()[name = string("op_1114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1114_cast_fp16 = einsum(equation = var_1114_equation_0, values = (var_954_cast_fp16, var_1075_cast_fp16))[name = string("op_1114_cast_fp16")]; + string var_1116_equation_0 = const()[name = string("op_1116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1116_cast_fp16 = einsum(equation = var_1116_equation_0, values = (var_954_cast_fp16, var_1076_cast_fp16))[name = string("op_1116_cast_fp16")]; + string var_1118_equation_0 = const()[name = string("op_1118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1118_cast_fp16 = einsum(equation = var_1118_equation_0, values = (var_958_cast_fp16, var_1077_cast_fp16))[name = string("op_1118_cast_fp16")]; + string var_1120_equation_0 = const()[name = string("op_1120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1120_cast_fp16 = einsum(equation = var_1120_equation_0, values = (var_958_cast_fp16, var_1078_cast_fp16))[name = string("op_1120_cast_fp16")]; + string var_1122_equation_0 = const()[name = string("op_1122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1122_cast_fp16 = einsum(equation = var_1122_equation_0, values = (var_958_cast_fp16, var_1079_cast_fp16))[name = string("op_1122_cast_fp16")]; + string var_1124_equation_0 = const()[name = string("op_1124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1124_cast_fp16 = einsum(equation = var_1124_equation_0, values = (var_958_cast_fp16, var_1080_cast_fp16))[name = string("op_1124_cast_fp16")]; + string var_1126_equation_0 = const()[name = string("op_1126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1126_cast_fp16 = einsum(equation = var_1126_equation_0, values = (var_962_cast_fp16, var_1081_cast_fp16))[name = string("op_1126_cast_fp16")]; + string var_1128_equation_0 = const()[name = string("op_1128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1128_cast_fp16 = einsum(equation = var_1128_equation_0, values = (var_962_cast_fp16, var_1082_cast_fp16))[name = string("op_1128_cast_fp16")]; + string var_1130_equation_0 = const()[name = string("op_1130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1130_cast_fp16 = einsum(equation = var_1130_equation_0, values = (var_962_cast_fp16, var_1083_cast_fp16))[name = string("op_1130_cast_fp16")]; + string var_1132_equation_0 = const()[name = string("op_1132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1132_cast_fp16 = einsum(equation = var_1132_equation_0, values = (var_962_cast_fp16, var_1084_cast_fp16))[name = string("op_1132_cast_fp16")]; + bool var_1134_interleave_0 = const()[name = string("op_1134_interleave_0"), val = bool(false)]; + tensor var_1134_cast_fp16 = concat(axis = var_659, interleave = var_1134_interleave_0, values = (var_1086_cast_fp16, var_1088_cast_fp16, var_1090_cast_fp16, var_1092_cast_fp16))[name = string("op_1134_cast_fp16")]; + bool var_1136_interleave_0 = const()[name = string("op_1136_interleave_0"), val = bool(false)]; + tensor var_1136_cast_fp16 = concat(axis = var_659, interleave = var_1136_interleave_0, values = (var_1094_cast_fp16, var_1096_cast_fp16, var_1098_cast_fp16, var_1100_cast_fp16))[name = string("op_1136_cast_fp16")]; + bool var_1138_interleave_0 = const()[name = string("op_1138_interleave_0"), val = bool(false)]; + tensor var_1138_cast_fp16 = concat(axis = var_659, interleave = var_1138_interleave_0, values = (var_1102_cast_fp16, var_1104_cast_fp16, var_1106_cast_fp16, var_1108_cast_fp16))[name = string("op_1138_cast_fp16")]; + bool var_1140_interleave_0 = const()[name = string("op_1140_interleave_0"), val = bool(false)]; + tensor var_1140_cast_fp16 = concat(axis = var_659, interleave = var_1140_interleave_0, values = (var_1110_cast_fp16, var_1112_cast_fp16, var_1114_cast_fp16, var_1116_cast_fp16))[name = string("op_1140_cast_fp16")]; + bool var_1142_interleave_0 = const()[name = string("op_1142_interleave_0"), val = bool(false)]; + tensor var_1142_cast_fp16 = concat(axis = var_659, interleave = var_1142_interleave_0, values = (var_1118_cast_fp16, var_1120_cast_fp16, var_1122_cast_fp16, var_1124_cast_fp16))[name = string("op_1142_cast_fp16")]; + bool var_1144_interleave_0 = const()[name = string("op_1144_interleave_0"), val = bool(false)]; + tensor var_1144_cast_fp16 = concat(axis = var_659, interleave = var_1144_interleave_0, values = (var_1126_cast_fp16, var_1128_cast_fp16, var_1130_cast_fp16, var_1132_cast_fp16))[name = string("op_1144_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_670, interleave = input_9_interleave_0, values = (var_1134_cast_fp16, var_1136_cast_fp16, var_1138_cast_fp16, var_1140_cast_fp16, var_1142_cast_fp16, var_1144_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6662016)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956992)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_1163_to_fp16 = const()[name = string("op_1163_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_1163_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6957824)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6958656)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6959488)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8139200)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8142336)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322048)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_1192 = const()[name = string("op_1192"), val = int32(3)]; + int32 var_1203 = const()[name = string("op_1203"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_1220_to_fp16 = const()[name = string("op_1220_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_1220_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322880)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9323712)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9324544)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9619520)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9620352)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9915328)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10210304)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_1258_begin_0 = const()[name = string("op_1258_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1258_end_0 = const()[name = string("op_1258_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1258_end_mask_0 = const()[name = string("op_1258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = query_5_cast_fp16)[name = string("op_1258_cast_fp16")]; + tensor var_1262_begin_0 = const()[name = string("op_1262_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1262_end_0 = const()[name = string("op_1262_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1262_end_mask_0 = const()[name = string("op_1262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1262_cast_fp16 = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = query_5_cast_fp16)[name = string("op_1262_cast_fp16")]; + tensor var_1266_begin_0 = const()[name = string("op_1266_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1266_end_0 = const()[name = string("op_1266_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1266_end_mask_0 = const()[name = string("op_1266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1266_cast_fp16 = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = query_5_cast_fp16)[name = string("op_1266_cast_fp16")]; + tensor var_1270_begin_0 = const()[name = string("op_1270_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1270_end_0 = const()[name = string("op_1270_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1270_end_mask_0 = const()[name = string("op_1270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = query_5_cast_fp16)[name = string("op_1270_cast_fp16")]; + tensor var_1274_begin_0 = const()[name = string("op_1274_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1274_end_0 = const()[name = string("op_1274_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1274_end_mask_0 = const()[name = string("op_1274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1274_cast_fp16 = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = query_5_cast_fp16)[name = string("op_1274_cast_fp16")]; + tensor var_1278_begin_0 = const()[name = string("op_1278_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1278_end_0 = const()[name = string("op_1278_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1278_end_mask_0 = const()[name = string("op_1278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1278_cast_fp16 = slice_by_index(begin = var_1278_begin_0, end = var_1278_end_0, end_mask = var_1278_end_mask_0, x = query_5_cast_fp16)[name = string("op_1278_cast_fp16")]; + tensor var_1287_begin_0 = const()[name = string("op_1287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1287_end_0 = const()[name = string("op_1287_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1287_end_mask_0 = const()[name = string("op_1287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1287_cast_fp16 = slice_by_index(begin = var_1287_begin_0, end = var_1287_end_0, end_mask = var_1287_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1287_cast_fp16")]; + tensor var_1294_begin_0 = const()[name = string("op_1294_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1294_end_0 = const()[name = string("op_1294_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1294_end_mask_0 = const()[name = string("op_1294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1294_cast_fp16 = slice_by_index(begin = var_1294_begin_0, end = var_1294_end_0, end_mask = var_1294_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1294_cast_fp16")]; + tensor var_1301_begin_0 = const()[name = string("op_1301_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1301_end_0 = const()[name = string("op_1301_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1301_end_mask_0 = const()[name = string("op_1301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1301_cast_fp16 = slice_by_index(begin = var_1301_begin_0, end = var_1301_end_0, end_mask = var_1301_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1301_cast_fp16")]; + tensor var_1308_begin_0 = const()[name = string("op_1308_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1308_end_0 = const()[name = string("op_1308_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1308_end_mask_0 = const()[name = string("op_1308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1308_cast_fp16 = slice_by_index(begin = var_1308_begin_0, end = var_1308_end_0, end_mask = var_1308_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1308_cast_fp16")]; + tensor var_1315_begin_0 = const()[name = string("op_1315_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1315_end_0 = const()[name = string("op_1315_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1315_end_mask_0 = const()[name = string("op_1315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1315_cast_fp16 = slice_by_index(begin = var_1315_begin_0, end = var_1315_end_0, end_mask = var_1315_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1315_cast_fp16")]; + tensor var_1322_begin_0 = const()[name = string("op_1322_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1322_end_0 = const()[name = string("op_1322_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1322_end_mask_0 = const()[name = string("op_1322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1322_cast_fp16 = slice_by_index(begin = var_1322_begin_0, end = var_1322_end_0, end_mask = var_1322_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1322_cast_fp16")]; + tensor var_1329_begin_0 = const()[name = string("op_1329_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1329_end_0 = const()[name = string("op_1329_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1329_end_mask_0 = const()[name = string("op_1329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1329_cast_fp16 = slice_by_index(begin = var_1329_begin_0, end = var_1329_end_0, end_mask = var_1329_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1329_cast_fp16")]; + tensor var_1336_begin_0 = const()[name = string("op_1336_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1336_end_0 = const()[name = string("op_1336_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1336_end_mask_0 = const()[name = string("op_1336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1336_cast_fp16 = slice_by_index(begin = var_1336_begin_0, end = var_1336_end_0, end_mask = var_1336_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor var_1343_begin_0 = const()[name = string("op_1343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1343_end_0 = const()[name = string("op_1343_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1343_end_mask_0 = const()[name = string("op_1343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1343_cast_fp16 = slice_by_index(begin = var_1343_begin_0, end = var_1343_end_0, end_mask = var_1343_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1343_cast_fp16")]; + tensor var_1350_begin_0 = const()[name = string("op_1350_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1350_end_0 = const()[name = string("op_1350_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1350_end_mask_0 = const()[name = string("op_1350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1350_cast_fp16 = slice_by_index(begin = var_1350_begin_0, end = var_1350_end_0, end_mask = var_1350_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1350_cast_fp16")]; + tensor var_1357_begin_0 = const()[name = string("op_1357_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1357_end_0 = const()[name = string("op_1357_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1357_end_mask_0 = const()[name = string("op_1357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1357_cast_fp16 = slice_by_index(begin = var_1357_begin_0, end = var_1357_end_0, end_mask = var_1357_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1357_cast_fp16")]; + tensor var_1364_begin_0 = const()[name = string("op_1364_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1364_end_0 = const()[name = string("op_1364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1364_end_mask_0 = const()[name = string("op_1364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1364_cast_fp16 = slice_by_index(begin = var_1364_begin_0, end = var_1364_end_0, end_mask = var_1364_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1364_cast_fp16")]; + tensor var_1371_begin_0 = const()[name = string("op_1371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1371_end_0 = const()[name = string("op_1371_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1371_end_mask_0 = const()[name = string("op_1371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1371_cast_fp16 = slice_by_index(begin = var_1371_begin_0, end = var_1371_end_0, end_mask = var_1371_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1371_cast_fp16")]; + tensor var_1378_begin_0 = const()[name = string("op_1378_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1378_end_0 = const()[name = string("op_1378_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1378_end_mask_0 = const()[name = string("op_1378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1378_cast_fp16 = slice_by_index(begin = var_1378_begin_0, end = var_1378_end_0, end_mask = var_1378_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1378_cast_fp16")]; + tensor var_1385_begin_0 = const()[name = string("op_1385_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1385_end_0 = const()[name = string("op_1385_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1385_end_mask_0 = const()[name = string("op_1385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1385_cast_fp16 = slice_by_index(begin = var_1385_begin_0, end = var_1385_end_0, end_mask = var_1385_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1385_cast_fp16")]; + tensor var_1392_begin_0 = const()[name = string("op_1392_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1392_end_0 = const()[name = string("op_1392_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1392_end_mask_0 = const()[name = string("op_1392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1392_cast_fp16 = slice_by_index(begin = var_1392_begin_0, end = var_1392_end_0, end_mask = var_1392_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1392_cast_fp16")]; + tensor var_1399_begin_0 = const()[name = string("op_1399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1399_end_0 = const()[name = string("op_1399_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1399_end_mask_0 = const()[name = string("op_1399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1399_cast_fp16 = slice_by_index(begin = var_1399_begin_0, end = var_1399_end_0, end_mask = var_1399_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1399_cast_fp16")]; + tensor var_1406_begin_0 = const()[name = string("op_1406_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1406_end_0 = const()[name = string("op_1406_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1406_end_mask_0 = const()[name = string("op_1406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1406_cast_fp16 = slice_by_index(begin = var_1406_begin_0, end = var_1406_end_0, end_mask = var_1406_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1406_cast_fp16")]; + tensor var_1413_begin_0 = const()[name = string("op_1413_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1413_end_0 = const()[name = string("op_1413_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1413_end_mask_0 = const()[name = string("op_1413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1413_cast_fp16 = slice_by_index(begin = var_1413_begin_0, end = var_1413_end_0, end_mask = var_1413_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1413_cast_fp16")]; + tensor var_1420_begin_0 = const()[name = string("op_1420_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1420_end_0 = const()[name = string("op_1420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1420_end_mask_0 = const()[name = string("op_1420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1420_cast_fp16 = slice_by_index(begin = var_1420_begin_0, end = var_1420_end_0, end_mask = var_1420_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1420_cast_fp16")]; + tensor var_1427_begin_0 = const()[name = string("op_1427_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1427_end_0 = const()[name = string("op_1427_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1427_end_mask_0 = const()[name = string("op_1427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1427_cast_fp16 = slice_by_index(begin = var_1427_begin_0, end = var_1427_end_0, end_mask = var_1427_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1427_cast_fp16")]; + tensor var_1434_begin_0 = const()[name = string("op_1434_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1434_end_0 = const()[name = string("op_1434_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1434_end_mask_0 = const()[name = string("op_1434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1434_cast_fp16 = slice_by_index(begin = var_1434_begin_0, end = var_1434_end_0, end_mask = var_1434_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1434_cast_fp16")]; + tensor var_1441_begin_0 = const()[name = string("op_1441_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1441_end_0 = const()[name = string("op_1441_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1441_end_mask_0 = const()[name = string("op_1441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1441_cast_fp16 = slice_by_index(begin = var_1441_begin_0, end = var_1441_end_0, end_mask = var_1441_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1441_cast_fp16")]; + tensor var_1448_begin_0 = const()[name = string("op_1448_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1448_end_0 = const()[name = string("op_1448_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1448_end_mask_0 = const()[name = string("op_1448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1448_cast_fp16 = slice_by_index(begin = var_1448_begin_0, end = var_1448_end_0, end_mask = var_1448_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1448_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1453_begin_0 = const()[name = string("op_1453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1453_end_0 = const()[name = string("op_1453_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1453_end_mask_0 = const()[name = string("op_1453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_1")]; + tensor var_1453_cast_fp16 = slice_by_index(begin = var_1453_begin_0, end = var_1453_end_0, end_mask = var_1453_end_mask_0, x = k_5_cast_fp16)[name = string("op_1453_cast_fp16")]; + tensor var_1457_begin_0 = const()[name = string("op_1457_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1457_end_0 = const()[name = string("op_1457_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1457_end_mask_0 = const()[name = string("op_1457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1457_cast_fp16 = slice_by_index(begin = var_1457_begin_0, end = var_1457_end_0, end_mask = var_1457_end_mask_0, x = k_5_cast_fp16)[name = string("op_1457_cast_fp16")]; + tensor var_1461_begin_0 = const()[name = string("op_1461_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1461_end_0 = const()[name = string("op_1461_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1461_end_mask_0 = const()[name = string("op_1461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1461_cast_fp16 = slice_by_index(begin = var_1461_begin_0, end = var_1461_end_0, end_mask = var_1461_end_mask_0, x = k_5_cast_fp16)[name = string("op_1461_cast_fp16")]; + tensor var_1465_begin_0 = const()[name = string("op_1465_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1465_end_0 = const()[name = string("op_1465_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1465_end_mask_0 = const()[name = string("op_1465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1465_cast_fp16 = slice_by_index(begin = var_1465_begin_0, end = var_1465_end_0, end_mask = var_1465_end_mask_0, x = k_5_cast_fp16)[name = string("op_1465_cast_fp16")]; + tensor var_1469_begin_0 = const()[name = string("op_1469_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1469_end_0 = const()[name = string("op_1469_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1469_end_mask_0 = const()[name = string("op_1469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1469_cast_fp16 = slice_by_index(begin = var_1469_begin_0, end = var_1469_end_0, end_mask = var_1469_end_mask_0, x = k_5_cast_fp16)[name = string("op_1469_cast_fp16")]; + tensor var_1473_begin_0 = const()[name = string("op_1473_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1473_end_0 = const()[name = string("op_1473_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1473_end_mask_0 = const()[name = string("op_1473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = k_5_cast_fp16)[name = string("op_1473_cast_fp16")]; + tensor var_1475_begin_0 = const()[name = string("op_1475_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1475_end_0 = const()[name = string("op_1475_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1475_end_mask_0 = const()[name = string("op_1475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1475_cast_fp16 = slice_by_index(begin = var_1475_begin_0, end = var_1475_end_0, end_mask = var_1475_end_mask_0, x = value_5_cast_fp16)[name = string("op_1475_cast_fp16")]; + tensor var_1479_begin_0 = const()[name = string("op_1479_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1479_end_0 = const()[name = string("op_1479_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1479_end_mask_0 = const()[name = string("op_1479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1479_cast_fp16 = slice_by_index(begin = var_1479_begin_0, end = var_1479_end_0, end_mask = var_1479_end_mask_0, x = value_5_cast_fp16)[name = string("op_1479_cast_fp16")]; + tensor var_1483_begin_0 = const()[name = string("op_1483_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1483_end_0 = const()[name = string("op_1483_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1483_end_mask_0 = const()[name = string("op_1483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1483_cast_fp16 = slice_by_index(begin = var_1483_begin_0, end = var_1483_end_0, end_mask = var_1483_end_mask_0, x = value_5_cast_fp16)[name = string("op_1483_cast_fp16")]; + tensor var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = value_5_cast_fp16)[name = string("op_1487_cast_fp16")]; + tensor var_1491_begin_0 = const()[name = string("op_1491_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1491_end_0 = const()[name = string("op_1491_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1491_end_mask_0 = const()[name = string("op_1491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1491_cast_fp16 = slice_by_index(begin = var_1491_begin_0, end = var_1491_end_0, end_mask = var_1491_end_mask_0, x = value_5_cast_fp16)[name = string("op_1491_cast_fp16")]; + tensor var_1495_begin_0 = const()[name = string("op_1495_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1495_end_0 = const()[name = string("op_1495_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1495_end_mask_0 = const()[name = string("op_1495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1495_cast_fp16 = slice_by_index(begin = var_1495_begin_0, end = var_1495_end_0, end_mask = var_1495_end_mask_0, x = value_5_cast_fp16)[name = string("op_1495_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1453_cast_fp16, var_1287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1453_cast_fp16, var_1294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1453_cast_fp16, var_1301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1453_cast_fp16, var_1308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1457_cast_fp16, var_1315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1457_cast_fp16, var_1322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1457_cast_fp16, var_1329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1457_cast_fp16, var_1336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1461_cast_fp16, var_1343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1461_cast_fp16, var_1350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1461_cast_fp16, var_1357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1461_cast_fp16, var_1364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1465_cast_fp16, var_1371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1465_cast_fp16, var_1378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1465_cast_fp16, var_1385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1465_cast_fp16, var_1392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1469_cast_fp16, var_1399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1469_cast_fp16, var_1406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1469_cast_fp16, var_1413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1469_cast_fp16, var_1420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1473_cast_fp16, var_1427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1473_cast_fp16, var_1434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1473_cast_fp16, var_1441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1473_cast_fp16, var_1448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + fp16 var_1546_to_fp16 = const()[name = string("op_1546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1546_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1548_to_fp16 = const()[name = string("op_1548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1548_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1550_to_fp16 = const()[name = string("op_1550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1550_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1552_to_fp16 = const()[name = string("op_1552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1552_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1554_to_fp16 = const()[name = string("op_1554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1554_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1556_to_fp16 = const()[name = string("op_1556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1556_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1558_to_fp16 = const()[name = string("op_1558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1558_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1560_to_fp16 = const()[name = string("op_1560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1560_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1562_to_fp16 = const()[name = string("op_1562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1562_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1564_to_fp16 = const()[name = string("op_1564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1564_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1566_to_fp16 = const()[name = string("op_1566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1566_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1568_to_fp16 = const()[name = string("op_1568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1568_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1570_to_fp16 = const()[name = string("op_1570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1570_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1572_to_fp16 = const()[name = string("op_1572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1572_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1574_to_fp16 = const()[name = string("op_1574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1574_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1576_to_fp16 = const()[name = string("op_1576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1576_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + fp16 var_1578_to_fp16 = const()[name = string("op_1578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1578_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1580_to_fp16 = const()[name = string("op_1580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1580_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1582_to_fp16 = const()[name = string("op_1582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1582_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1584_to_fp16 = const()[name = string("op_1584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1584_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1586_to_fp16 = const()[name = string("op_1586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1586_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1588_to_fp16 = const()[name = string("op_1588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1588_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1590_to_fp16 = const()[name = string("op_1590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1590_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1592_to_fp16 = const()[name = string("op_1592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1592_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + tensor var_1594_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_97_cast_fp16)[name = string("op_1594_cast_fp16")]; + tensor var_1595_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_99_cast_fp16)[name = string("op_1595_cast_fp16")]; + tensor var_1596_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_101_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1597_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_103_cast_fp16)[name = string("op_1597_cast_fp16")]; + tensor var_1598_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_105_cast_fp16)[name = string("op_1598_cast_fp16")]; + tensor var_1599_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_107_cast_fp16)[name = string("op_1599_cast_fp16")]; + tensor var_1600_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_109_cast_fp16)[name = string("op_1600_cast_fp16")]; + tensor var_1601_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_111_cast_fp16)[name = string("op_1601_cast_fp16")]; + tensor var_1602_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_113_cast_fp16)[name = string("op_1602_cast_fp16")]; + tensor var_1603_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_115_cast_fp16)[name = string("op_1603_cast_fp16")]; + tensor var_1604_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_117_cast_fp16)[name = string("op_1604_cast_fp16")]; + tensor var_1605_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_119_cast_fp16)[name = string("op_1605_cast_fp16")]; + tensor var_1606_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_121_cast_fp16)[name = string("op_1606_cast_fp16")]; + tensor var_1607_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_123_cast_fp16)[name = string("op_1607_cast_fp16")]; + tensor var_1608_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_125_cast_fp16)[name = string("op_1608_cast_fp16")]; + tensor var_1609_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_127_cast_fp16)[name = string("op_1609_cast_fp16")]; + tensor var_1610_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_129_cast_fp16)[name = string("op_1610_cast_fp16")]; + tensor var_1611_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_131_cast_fp16)[name = string("op_1611_cast_fp16")]; + tensor var_1612_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_133_cast_fp16)[name = string("op_1612_cast_fp16")]; + tensor var_1613_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_135_cast_fp16)[name = string("op_1613_cast_fp16")]; + tensor var_1614_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_137_cast_fp16)[name = string("op_1614_cast_fp16")]; + tensor var_1615_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_139_cast_fp16)[name = string("op_1615_cast_fp16")]; + tensor var_1616_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_141_cast_fp16)[name = string("op_1616_cast_fp16")]; + tensor var_1617_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_143_cast_fp16)[name = string("op_1617_cast_fp16")]; + string var_1619_equation_0 = const()[name = string("op_1619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1619_cast_fp16 = einsum(equation = var_1619_equation_0, values = (var_1475_cast_fp16, var_1594_cast_fp16))[name = string("op_1619_cast_fp16")]; + string var_1621_equation_0 = const()[name = string("op_1621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1621_cast_fp16 = einsum(equation = var_1621_equation_0, values = (var_1475_cast_fp16, var_1595_cast_fp16))[name = string("op_1621_cast_fp16")]; + string var_1623_equation_0 = const()[name = string("op_1623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1623_cast_fp16 = einsum(equation = var_1623_equation_0, values = (var_1475_cast_fp16, var_1596_cast_fp16))[name = string("op_1623_cast_fp16")]; + string var_1625_equation_0 = const()[name = string("op_1625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1625_cast_fp16 = einsum(equation = var_1625_equation_0, values = (var_1475_cast_fp16, var_1597_cast_fp16))[name = string("op_1625_cast_fp16")]; + string var_1627_equation_0 = const()[name = string("op_1627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1627_cast_fp16 = einsum(equation = var_1627_equation_0, values = (var_1479_cast_fp16, var_1598_cast_fp16))[name = string("op_1627_cast_fp16")]; + string var_1629_equation_0 = const()[name = string("op_1629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1629_cast_fp16 = einsum(equation = var_1629_equation_0, values = (var_1479_cast_fp16, var_1599_cast_fp16))[name = string("op_1629_cast_fp16")]; + string var_1631_equation_0 = const()[name = string("op_1631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1631_cast_fp16 = einsum(equation = var_1631_equation_0, values = (var_1479_cast_fp16, var_1600_cast_fp16))[name = string("op_1631_cast_fp16")]; + string var_1633_equation_0 = const()[name = string("op_1633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1633_cast_fp16 = einsum(equation = var_1633_equation_0, values = (var_1479_cast_fp16, var_1601_cast_fp16))[name = string("op_1633_cast_fp16")]; + string var_1635_equation_0 = const()[name = string("op_1635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1635_cast_fp16 = einsum(equation = var_1635_equation_0, values = (var_1483_cast_fp16, var_1602_cast_fp16))[name = string("op_1635_cast_fp16")]; + string var_1637_equation_0 = const()[name = string("op_1637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1637_cast_fp16 = einsum(equation = var_1637_equation_0, values = (var_1483_cast_fp16, var_1603_cast_fp16))[name = string("op_1637_cast_fp16")]; + string var_1639_equation_0 = const()[name = string("op_1639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1639_cast_fp16 = einsum(equation = var_1639_equation_0, values = (var_1483_cast_fp16, var_1604_cast_fp16))[name = string("op_1639_cast_fp16")]; + string var_1641_equation_0 = const()[name = string("op_1641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1641_cast_fp16 = einsum(equation = var_1641_equation_0, values = (var_1483_cast_fp16, var_1605_cast_fp16))[name = string("op_1641_cast_fp16")]; + string var_1643_equation_0 = const()[name = string("op_1643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1643_cast_fp16 = einsum(equation = var_1643_equation_0, values = (var_1487_cast_fp16, var_1606_cast_fp16))[name = string("op_1643_cast_fp16")]; + string var_1645_equation_0 = const()[name = string("op_1645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1645_cast_fp16 = einsum(equation = var_1645_equation_0, values = (var_1487_cast_fp16, var_1607_cast_fp16))[name = string("op_1645_cast_fp16")]; + string var_1647_equation_0 = const()[name = string("op_1647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1647_cast_fp16 = einsum(equation = var_1647_equation_0, values = (var_1487_cast_fp16, var_1608_cast_fp16))[name = string("op_1647_cast_fp16")]; + string var_1649_equation_0 = const()[name = string("op_1649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1649_cast_fp16 = einsum(equation = var_1649_equation_0, values = (var_1487_cast_fp16, var_1609_cast_fp16))[name = string("op_1649_cast_fp16")]; + string var_1651_equation_0 = const()[name = string("op_1651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1651_cast_fp16 = einsum(equation = var_1651_equation_0, values = (var_1491_cast_fp16, var_1610_cast_fp16))[name = string("op_1651_cast_fp16")]; + string var_1653_equation_0 = const()[name = string("op_1653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1653_cast_fp16 = einsum(equation = var_1653_equation_0, values = (var_1491_cast_fp16, var_1611_cast_fp16))[name = string("op_1653_cast_fp16")]; + string var_1655_equation_0 = const()[name = string("op_1655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1655_cast_fp16 = einsum(equation = var_1655_equation_0, values = (var_1491_cast_fp16, var_1612_cast_fp16))[name = string("op_1655_cast_fp16")]; + string var_1657_equation_0 = const()[name = string("op_1657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1657_cast_fp16 = einsum(equation = var_1657_equation_0, values = (var_1491_cast_fp16, var_1613_cast_fp16))[name = string("op_1657_cast_fp16")]; + string var_1659_equation_0 = const()[name = string("op_1659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1659_cast_fp16 = einsum(equation = var_1659_equation_0, values = (var_1495_cast_fp16, var_1614_cast_fp16))[name = string("op_1659_cast_fp16")]; + string var_1661_equation_0 = const()[name = string("op_1661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1661_cast_fp16 = einsum(equation = var_1661_equation_0, values = (var_1495_cast_fp16, var_1615_cast_fp16))[name = string("op_1661_cast_fp16")]; + string var_1663_equation_0 = const()[name = string("op_1663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1663_cast_fp16 = einsum(equation = var_1663_equation_0, values = (var_1495_cast_fp16, var_1616_cast_fp16))[name = string("op_1663_cast_fp16")]; + string var_1665_equation_0 = const()[name = string("op_1665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1665_cast_fp16 = einsum(equation = var_1665_equation_0, values = (var_1495_cast_fp16, var_1617_cast_fp16))[name = string("op_1665_cast_fp16")]; + bool var_1667_interleave_0 = const()[name = string("op_1667_interleave_0"), val = bool(false)]; + tensor var_1667_cast_fp16 = concat(axis = var_1192, interleave = var_1667_interleave_0, values = (var_1619_cast_fp16, var_1621_cast_fp16, var_1623_cast_fp16, var_1625_cast_fp16))[name = string("op_1667_cast_fp16")]; + bool var_1669_interleave_0 = const()[name = string("op_1669_interleave_0"), val = bool(false)]; + tensor var_1669_cast_fp16 = concat(axis = var_1192, interleave = var_1669_interleave_0, values = (var_1627_cast_fp16, var_1629_cast_fp16, var_1631_cast_fp16, var_1633_cast_fp16))[name = string("op_1669_cast_fp16")]; + bool var_1671_interleave_0 = const()[name = string("op_1671_interleave_0"), val = bool(false)]; + tensor var_1671_cast_fp16 = concat(axis = var_1192, interleave = var_1671_interleave_0, values = (var_1635_cast_fp16, var_1637_cast_fp16, var_1639_cast_fp16, var_1641_cast_fp16))[name = string("op_1671_cast_fp16")]; + bool var_1673_interleave_0 = const()[name = string("op_1673_interleave_0"), val = bool(false)]; + tensor var_1673_cast_fp16 = concat(axis = var_1192, interleave = var_1673_interleave_0, values = (var_1643_cast_fp16, var_1645_cast_fp16, var_1647_cast_fp16, var_1649_cast_fp16))[name = string("op_1673_cast_fp16")]; + bool var_1675_interleave_0 = const()[name = string("op_1675_interleave_0"), val = bool(false)]; + tensor var_1675_cast_fp16 = concat(axis = var_1192, interleave = var_1675_interleave_0, values = (var_1651_cast_fp16, var_1653_cast_fp16, var_1655_cast_fp16, var_1657_cast_fp16))[name = string("op_1675_cast_fp16")]; + bool var_1677_interleave_0 = const()[name = string("op_1677_interleave_0"), val = bool(false)]; + tensor var_1677_cast_fp16 = concat(axis = var_1192, interleave = var_1677_interleave_0, values = (var_1659_cast_fp16, var_1661_cast_fp16, var_1663_cast_fp16, var_1665_cast_fp16))[name = string("op_1677_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_1203, interleave = input_17_interleave_0, values = (var_1667_cast_fp16, var_1669_cast_fp16, var_1671_cast_fp16, var_1673_cast_fp16, var_1675_cast_fp16, var_1677_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10211136)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506112)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_1696_to_fp16 = const()[name = string("op_1696_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_1696_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506944)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10507776)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10508608)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11688320)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11691456)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12871168)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_1725 = const()[name = string("op_1725"), val = int32(3)]; + int32 var_1736 = const()[name = string("op_1736"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_1753_to_fp16 = const()[name = string("op_1753_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_1753_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872000)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872832)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12873664)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13168640)))]; + tensor query_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13169472)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13464448)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13759424)))]; + tensor value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_1791_begin_0 = const()[name = string("op_1791_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1791_end_0 = const()[name = string("op_1791_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1791_end_mask_0 = const()[name = string("op_1791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1791_cast_fp16 = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, x = query_cast_fp16)[name = string("op_1791_cast_fp16")]; + tensor var_1795_begin_0 = const()[name = string("op_1795_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1795_end_0 = const()[name = string("op_1795_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1795_end_mask_0 = const()[name = string("op_1795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1795_cast_fp16 = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = query_cast_fp16)[name = string("op_1795_cast_fp16")]; + tensor var_1799_begin_0 = const()[name = string("op_1799_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1799_end_0 = const()[name = string("op_1799_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1799_end_mask_0 = const()[name = string("op_1799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = query_cast_fp16)[name = string("op_1799_cast_fp16")]; + tensor var_1803_begin_0 = const()[name = string("op_1803_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1803_end_0 = const()[name = string("op_1803_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1803_end_mask_0 = const()[name = string("op_1803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1803_cast_fp16 = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, x = query_cast_fp16)[name = string("op_1803_cast_fp16")]; + tensor var_1807_begin_0 = const()[name = string("op_1807_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1807_end_0 = const()[name = string("op_1807_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1807_end_mask_0 = const()[name = string("op_1807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1807_cast_fp16 = slice_by_index(begin = var_1807_begin_0, end = var_1807_end_0, end_mask = var_1807_end_mask_0, x = query_cast_fp16)[name = string("op_1807_cast_fp16")]; + tensor var_1811_begin_0 = const()[name = string("op_1811_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1811_end_0 = const()[name = string("op_1811_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1811_end_mask_0 = const()[name = string("op_1811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1811_cast_fp16 = slice_by_index(begin = var_1811_begin_0, end = var_1811_end_0, end_mask = var_1811_end_mask_0, x = query_cast_fp16)[name = string("op_1811_cast_fp16")]; + tensor var_1820_begin_0 = const()[name = string("op_1820_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1820_end_0 = const()[name = string("op_1820_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1820_end_mask_0 = const()[name = string("op_1820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1820_cast_fp16 = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1820_cast_fp16")]; + tensor var_1827_begin_0 = const()[name = string("op_1827_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1827_end_0 = const()[name = string("op_1827_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1827_end_mask_0 = const()[name = string("op_1827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1827_cast_fp16 = slice_by_index(begin = var_1827_begin_0, end = var_1827_end_0, end_mask = var_1827_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1827_cast_fp16")]; + tensor var_1834_begin_0 = const()[name = string("op_1834_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1834_end_0 = const()[name = string("op_1834_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1834_end_mask_0 = const()[name = string("op_1834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1834_cast_fp16 = slice_by_index(begin = var_1834_begin_0, end = var_1834_end_0, end_mask = var_1834_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1834_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1848_begin_0 = const()[name = string("op_1848_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1848_end_0 = const()[name = string("op_1848_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1848_end_mask_0 = const()[name = string("op_1848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1848_cast_fp16 = slice_by_index(begin = var_1848_begin_0, end = var_1848_end_0, end_mask = var_1848_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1848_cast_fp16")]; + tensor var_1855_begin_0 = const()[name = string("op_1855_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1855_end_0 = const()[name = string("op_1855_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1855_end_mask_0 = const()[name = string("op_1855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1855_cast_fp16")]; + tensor var_1862_begin_0 = const()[name = string("op_1862_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1862_end_0 = const()[name = string("op_1862_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1862_end_mask_0 = const()[name = string("op_1862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1862_cast_fp16 = slice_by_index(begin = var_1862_begin_0, end = var_1862_end_0, end_mask = var_1862_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1862_cast_fp16")]; + tensor var_1869_begin_0 = const()[name = string("op_1869_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1869_end_0 = const()[name = string("op_1869_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1869_end_mask_0 = const()[name = string("op_1869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1869_cast_fp16 = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1869_cast_fp16")]; + tensor var_1876_begin_0 = const()[name = string("op_1876_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1876_end_0 = const()[name = string("op_1876_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1876_end_mask_0 = const()[name = string("op_1876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1876_cast_fp16 = slice_by_index(begin = var_1876_begin_0, end = var_1876_end_0, end_mask = var_1876_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1876_cast_fp16")]; + tensor var_1883_begin_0 = const()[name = string("op_1883_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1883_end_0 = const()[name = string("op_1883_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1883_end_mask_0 = const()[name = string("op_1883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1883_cast_fp16 = slice_by_index(begin = var_1883_begin_0, end = var_1883_end_0, end_mask = var_1883_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1883_cast_fp16")]; + tensor var_1890_begin_0 = const()[name = string("op_1890_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1890_end_0 = const()[name = string("op_1890_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1890_end_mask_0 = const()[name = string("op_1890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1890_cast_fp16 = slice_by_index(begin = var_1890_begin_0, end = var_1890_end_0, end_mask = var_1890_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1890_cast_fp16")]; + tensor var_1897_begin_0 = const()[name = string("op_1897_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1897_end_0 = const()[name = string("op_1897_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1897_end_mask_0 = const()[name = string("op_1897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1897_cast_fp16 = slice_by_index(begin = var_1897_begin_0, end = var_1897_end_0, end_mask = var_1897_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1897_cast_fp16")]; + tensor var_1904_begin_0 = const()[name = string("op_1904_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1904_end_0 = const()[name = string("op_1904_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1904_end_mask_0 = const()[name = string("op_1904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1904_cast_fp16 = slice_by_index(begin = var_1904_begin_0, end = var_1904_end_0, end_mask = var_1904_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1904_cast_fp16")]; + tensor var_1911_begin_0 = const()[name = string("op_1911_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1911_end_0 = const()[name = string("op_1911_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1911_end_mask_0 = const()[name = string("op_1911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1911_cast_fp16 = slice_by_index(begin = var_1911_begin_0, end = var_1911_end_0, end_mask = var_1911_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1911_cast_fp16")]; + tensor var_1918_begin_0 = const()[name = string("op_1918_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1918_end_0 = const()[name = string("op_1918_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1918_end_mask_0 = const()[name = string("op_1918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1918_cast_fp16 = slice_by_index(begin = var_1918_begin_0, end = var_1918_end_0, end_mask = var_1918_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1918_cast_fp16")]; + tensor var_1925_begin_0 = const()[name = string("op_1925_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1925_end_0 = const()[name = string("op_1925_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1925_end_mask_0 = const()[name = string("op_1925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1925_cast_fp16 = slice_by_index(begin = var_1925_begin_0, end = var_1925_end_0, end_mask = var_1925_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1925_cast_fp16")]; + tensor var_1932_begin_0 = const()[name = string("op_1932_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1932_end_0 = const()[name = string("op_1932_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1932_end_mask_0 = const()[name = string("op_1932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1932_cast_fp16 = slice_by_index(begin = var_1932_begin_0, end = var_1932_end_0, end_mask = var_1932_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1932_cast_fp16")]; + tensor var_1939_begin_0 = const()[name = string("op_1939_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1939_end_0 = const()[name = string("op_1939_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1939_end_mask_0 = const()[name = string("op_1939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1939_cast_fp16 = slice_by_index(begin = var_1939_begin_0, end = var_1939_end_0, end_mask = var_1939_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1939_cast_fp16")]; + tensor var_1946_begin_0 = const()[name = string("op_1946_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1946_end_0 = const()[name = string("op_1946_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1946_end_mask_0 = const()[name = string("op_1946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1946_cast_fp16 = slice_by_index(begin = var_1946_begin_0, end = var_1946_end_0, end_mask = var_1946_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1946_cast_fp16")]; + tensor var_1953_begin_0 = const()[name = string("op_1953_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1953_end_0 = const()[name = string("op_1953_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1953_end_mask_0 = const()[name = string("op_1953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1953_cast_fp16 = slice_by_index(begin = var_1953_begin_0, end = var_1953_end_0, end_mask = var_1953_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1953_cast_fp16")]; + tensor var_1960_begin_0 = const()[name = string("op_1960_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1960_end_0 = const()[name = string("op_1960_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1960_end_mask_0 = const()[name = string("op_1960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1960_cast_fp16 = slice_by_index(begin = var_1960_begin_0, end = var_1960_end_0, end_mask = var_1960_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1960_cast_fp16")]; + tensor var_1967_begin_0 = const()[name = string("op_1967_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1967_end_0 = const()[name = string("op_1967_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1967_end_mask_0 = const()[name = string("op_1967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1967_cast_fp16 = slice_by_index(begin = var_1967_begin_0, end = var_1967_end_0, end_mask = var_1967_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1967_cast_fp16")]; + tensor var_1974_begin_0 = const()[name = string("op_1974_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1974_end_0 = const()[name = string("op_1974_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1974_end_mask_0 = const()[name = string("op_1974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1974_cast_fp16 = slice_by_index(begin = var_1974_begin_0, end = var_1974_end_0, end_mask = var_1974_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1974_cast_fp16")]; + tensor var_1981_begin_0 = const()[name = string("op_1981_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1981_end_0 = const()[name = string("op_1981_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1981_end_mask_0 = const()[name = string("op_1981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1981_cast_fp16 = slice_by_index(begin = var_1981_begin_0, end = var_1981_end_0, end_mask = var_1981_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1981_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1986_begin_0 = const()[name = string("op_1986_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1986_end_0 = const()[name = string("op_1986_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1986_end_mask_0 = const()[name = string("op_1986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_1986_cast_fp16 = slice_by_index(begin = var_1986_begin_0, end = var_1986_end_0, end_mask = var_1986_end_mask_0, x = k_7_cast_fp16)[name = string("op_1986_cast_fp16")]; + tensor var_1990_begin_0 = const()[name = string("op_1990_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1990_end_0 = const()[name = string("op_1990_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1990_end_mask_0 = const()[name = string("op_1990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1990_cast_fp16 = slice_by_index(begin = var_1990_begin_0, end = var_1990_end_0, end_mask = var_1990_end_mask_0, x = k_7_cast_fp16)[name = string("op_1990_cast_fp16")]; + tensor var_1994_begin_0 = const()[name = string("op_1994_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1994_end_0 = const()[name = string("op_1994_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1994_end_mask_0 = const()[name = string("op_1994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1994_cast_fp16 = slice_by_index(begin = var_1994_begin_0, end = var_1994_end_0, end_mask = var_1994_end_mask_0, x = k_7_cast_fp16)[name = string("op_1994_cast_fp16")]; + tensor var_1998_begin_0 = const()[name = string("op_1998_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1998_end_0 = const()[name = string("op_1998_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1998_end_mask_0 = const()[name = string("op_1998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1998_cast_fp16 = slice_by_index(begin = var_1998_begin_0, end = var_1998_end_0, end_mask = var_1998_end_mask_0, x = k_7_cast_fp16)[name = string("op_1998_cast_fp16")]; + tensor var_2002_begin_0 = const()[name = string("op_2002_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2002_end_0 = const()[name = string("op_2002_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2002_end_mask_0 = const()[name = string("op_2002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2002_cast_fp16 = slice_by_index(begin = var_2002_begin_0, end = var_2002_end_0, end_mask = var_2002_end_mask_0, x = k_7_cast_fp16)[name = string("op_2002_cast_fp16")]; + tensor var_2006_begin_0 = const()[name = string("op_2006_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2006_end_0 = const()[name = string("op_2006_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2006_end_mask_0 = const()[name = string("op_2006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2006_cast_fp16 = slice_by_index(begin = var_2006_begin_0, end = var_2006_end_0, end_mask = var_2006_end_mask_0, x = k_7_cast_fp16)[name = string("op_2006_cast_fp16")]; + tensor var_2008_begin_0 = const()[name = string("op_2008_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2008_end_0 = const()[name = string("op_2008_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2008_end_mask_0 = const()[name = string("op_2008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2008_cast_fp16 = slice_by_index(begin = var_2008_begin_0, end = var_2008_end_0, end_mask = var_2008_end_mask_0, x = value_cast_fp16)[name = string("op_2008_cast_fp16")]; + tensor var_2012_begin_0 = const()[name = string("op_2012_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2012_end_0 = const()[name = string("op_2012_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2012_end_mask_0 = const()[name = string("op_2012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2012_cast_fp16 = slice_by_index(begin = var_2012_begin_0, end = var_2012_end_0, end_mask = var_2012_end_mask_0, x = value_cast_fp16)[name = string("op_2012_cast_fp16")]; + tensor var_2016_begin_0 = const()[name = string("op_2016_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2016_end_0 = const()[name = string("op_2016_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2016_end_mask_0 = const()[name = string("op_2016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2016_cast_fp16 = slice_by_index(begin = var_2016_begin_0, end = var_2016_end_0, end_mask = var_2016_end_mask_0, x = value_cast_fp16)[name = string("op_2016_cast_fp16")]; + tensor var_2020_begin_0 = const()[name = string("op_2020_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2020_end_0 = const()[name = string("op_2020_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2020_end_mask_0 = const()[name = string("op_2020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2020_cast_fp16 = slice_by_index(begin = var_2020_begin_0, end = var_2020_end_0, end_mask = var_2020_end_mask_0, x = value_cast_fp16)[name = string("op_2020_cast_fp16")]; + tensor var_2024_begin_0 = const()[name = string("op_2024_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2024_end_0 = const()[name = string("op_2024_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2024_end_mask_0 = const()[name = string("op_2024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2024_cast_fp16 = slice_by_index(begin = var_2024_begin_0, end = var_2024_end_0, end_mask = var_2024_end_mask_0, x = value_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2028_begin_0 = const()[name = string("op_2028_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2028_end_0 = const()[name = string("op_2028_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2028_end_mask_0 = const()[name = string("op_2028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2028_cast_fp16 = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, x = value_cast_fp16)[name = string("op_2028_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1986_cast_fp16, var_1820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1986_cast_fp16, var_1827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1986_cast_fp16, var_1834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1986_cast_fp16, var_1841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1990_cast_fp16, var_1848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1990_cast_fp16, var_1855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1990_cast_fp16, var_1862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1990_cast_fp16, var_1869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1994_cast_fp16, var_1876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1994_cast_fp16, var_1883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1994_cast_fp16, var_1890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1994_cast_fp16, var_1897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1998_cast_fp16, var_1904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1998_cast_fp16, var_1911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1998_cast_fp16, var_1918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1998_cast_fp16, var_1925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_2002_cast_fp16, var_1932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_2002_cast_fp16, var_1939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_2002_cast_fp16, var_1946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_2002_cast_fp16, var_1953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_2006_cast_fp16, var_1960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_2006_cast_fp16, var_1967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_2006_cast_fp16, var_1974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_2006_cast_fp16, var_1981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_2079_to_fp16 = const()[name = string("op_2079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_2079_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_2081_to_fp16 = const()[name = string("op_2081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_2081_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_2083_to_fp16 = const()[name = string("op_2083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_2083_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_2085_to_fp16 = const()[name = string("op_2085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_2085_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_2087_to_fp16 = const()[name = string("op_2087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_2087_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_2089_to_fp16 = const()[name = string("op_2089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_2089_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_2091_to_fp16 = const()[name = string("op_2091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_2091_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_2093_to_fp16 = const()[name = string("op_2093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_2093_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_2095_to_fp16 = const()[name = string("op_2095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_2095_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_2097_to_fp16 = const()[name = string("op_2097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_2097_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_2099_to_fp16 = const()[name = string("op_2099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_2099_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_2101_to_fp16 = const()[name = string("op_2101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_2101_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_2103_to_fp16 = const()[name = string("op_2103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_2103_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_2105_to_fp16 = const()[name = string("op_2105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_2105_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_2107_to_fp16 = const()[name = string("op_2107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_2107_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_2109_to_fp16 = const()[name = string("op_2109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_2109_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_2111_to_fp16 = const()[name = string("op_2111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_2111_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_2113_to_fp16 = const()[name = string("op_2113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_2113_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_2115_to_fp16 = const()[name = string("op_2115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_2115_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_2117_to_fp16 = const()[name = string("op_2117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_2117_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_2119_to_fp16 = const()[name = string("op_2119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_2119_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_2121_to_fp16 = const()[name = string("op_2121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_2121_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_2123_to_fp16 = const()[name = string("op_2123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_2123_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_2125_to_fp16 = const()[name = string("op_2125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_2125_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_2127_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_145_cast_fp16)[name = string("op_2127_cast_fp16")]; + tensor var_2128_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_147_cast_fp16)[name = string("op_2128_cast_fp16")]; + tensor var_2129_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_149_cast_fp16)[name = string("op_2129_cast_fp16")]; + tensor var_2130_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_151_cast_fp16)[name = string("op_2130_cast_fp16")]; + tensor var_2131_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_153_cast_fp16)[name = string("op_2131_cast_fp16")]; + tensor var_2132_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_155_cast_fp16)[name = string("op_2132_cast_fp16")]; + tensor var_2133_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_157_cast_fp16)[name = string("op_2133_cast_fp16")]; + tensor var_2134_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_159_cast_fp16)[name = string("op_2134_cast_fp16")]; + tensor var_2135_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_161_cast_fp16)[name = string("op_2135_cast_fp16")]; + tensor var_2136_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_163_cast_fp16)[name = string("op_2136_cast_fp16")]; + tensor var_2137_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_165_cast_fp16)[name = string("op_2137_cast_fp16")]; + tensor var_2138_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_167_cast_fp16)[name = string("op_2138_cast_fp16")]; + tensor var_2139_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_169_cast_fp16)[name = string("op_2139_cast_fp16")]; + tensor var_2140_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_171_cast_fp16)[name = string("op_2140_cast_fp16")]; + tensor var_2141_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_173_cast_fp16)[name = string("op_2141_cast_fp16")]; + tensor var_2142_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_175_cast_fp16)[name = string("op_2142_cast_fp16")]; + tensor var_2143_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_177_cast_fp16)[name = string("op_2143_cast_fp16")]; + tensor var_2144_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_179_cast_fp16)[name = string("op_2144_cast_fp16")]; + tensor var_2145_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_181_cast_fp16)[name = string("op_2145_cast_fp16")]; + tensor var_2146_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_183_cast_fp16)[name = string("op_2146_cast_fp16")]; + tensor var_2147_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_185_cast_fp16)[name = string("op_2147_cast_fp16")]; + tensor var_2148_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_187_cast_fp16)[name = string("op_2148_cast_fp16")]; + tensor var_2149_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_189_cast_fp16)[name = string("op_2149_cast_fp16")]; + tensor var_2150_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_cast_fp16)[name = string("op_2150_cast_fp16")]; + string var_2152_equation_0 = const()[name = string("op_2152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2152_cast_fp16 = einsum(equation = var_2152_equation_0, values = (var_2008_cast_fp16, var_2127_cast_fp16))[name = string("op_2152_cast_fp16")]; + string var_2154_equation_0 = const()[name = string("op_2154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2154_cast_fp16 = einsum(equation = var_2154_equation_0, values = (var_2008_cast_fp16, var_2128_cast_fp16))[name = string("op_2154_cast_fp16")]; + string var_2156_equation_0 = const()[name = string("op_2156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2156_cast_fp16 = einsum(equation = var_2156_equation_0, values = (var_2008_cast_fp16, var_2129_cast_fp16))[name = string("op_2156_cast_fp16")]; + string var_2158_equation_0 = const()[name = string("op_2158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2158_cast_fp16 = einsum(equation = var_2158_equation_0, values = (var_2008_cast_fp16, var_2130_cast_fp16))[name = string("op_2158_cast_fp16")]; + string var_2160_equation_0 = const()[name = string("op_2160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2160_cast_fp16 = einsum(equation = var_2160_equation_0, values = (var_2012_cast_fp16, var_2131_cast_fp16))[name = string("op_2160_cast_fp16")]; + string var_2162_equation_0 = const()[name = string("op_2162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2162_cast_fp16 = einsum(equation = var_2162_equation_0, values = (var_2012_cast_fp16, var_2132_cast_fp16))[name = string("op_2162_cast_fp16")]; + string var_2164_equation_0 = const()[name = string("op_2164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2164_cast_fp16 = einsum(equation = var_2164_equation_0, values = (var_2012_cast_fp16, var_2133_cast_fp16))[name = string("op_2164_cast_fp16")]; + string var_2166_equation_0 = const()[name = string("op_2166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2166_cast_fp16 = einsum(equation = var_2166_equation_0, values = (var_2012_cast_fp16, var_2134_cast_fp16))[name = string("op_2166_cast_fp16")]; + string var_2168_equation_0 = const()[name = string("op_2168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2168_cast_fp16 = einsum(equation = var_2168_equation_0, values = (var_2016_cast_fp16, var_2135_cast_fp16))[name = string("op_2168_cast_fp16")]; + string var_2170_equation_0 = const()[name = string("op_2170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2170_cast_fp16 = einsum(equation = var_2170_equation_0, values = (var_2016_cast_fp16, var_2136_cast_fp16))[name = string("op_2170_cast_fp16")]; + string var_2172_equation_0 = const()[name = string("op_2172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2172_cast_fp16 = einsum(equation = var_2172_equation_0, values = (var_2016_cast_fp16, var_2137_cast_fp16))[name = string("op_2172_cast_fp16")]; + string var_2174_equation_0 = const()[name = string("op_2174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2174_cast_fp16 = einsum(equation = var_2174_equation_0, values = (var_2016_cast_fp16, var_2138_cast_fp16))[name = string("op_2174_cast_fp16")]; + string var_2176_equation_0 = const()[name = string("op_2176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2176_cast_fp16 = einsum(equation = var_2176_equation_0, values = (var_2020_cast_fp16, var_2139_cast_fp16))[name = string("op_2176_cast_fp16")]; + string var_2178_equation_0 = const()[name = string("op_2178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2178_cast_fp16 = einsum(equation = var_2178_equation_0, values = (var_2020_cast_fp16, var_2140_cast_fp16))[name = string("op_2178_cast_fp16")]; + string var_2180_equation_0 = const()[name = string("op_2180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2180_cast_fp16 = einsum(equation = var_2180_equation_0, values = (var_2020_cast_fp16, var_2141_cast_fp16))[name = string("op_2180_cast_fp16")]; + string var_2182_equation_0 = const()[name = string("op_2182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2182_cast_fp16 = einsum(equation = var_2182_equation_0, values = (var_2020_cast_fp16, var_2142_cast_fp16))[name = string("op_2182_cast_fp16")]; + string var_2184_equation_0 = const()[name = string("op_2184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2184_cast_fp16 = einsum(equation = var_2184_equation_0, values = (var_2024_cast_fp16, var_2143_cast_fp16))[name = string("op_2184_cast_fp16")]; + string var_2186_equation_0 = const()[name = string("op_2186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2186_cast_fp16 = einsum(equation = var_2186_equation_0, values = (var_2024_cast_fp16, var_2144_cast_fp16))[name = string("op_2186_cast_fp16")]; + string var_2188_equation_0 = const()[name = string("op_2188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2188_cast_fp16 = einsum(equation = var_2188_equation_0, values = (var_2024_cast_fp16, var_2145_cast_fp16))[name = string("op_2188_cast_fp16")]; + string var_2190_equation_0 = const()[name = string("op_2190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2190_cast_fp16 = einsum(equation = var_2190_equation_0, values = (var_2024_cast_fp16, var_2146_cast_fp16))[name = string("op_2190_cast_fp16")]; + string var_2192_equation_0 = const()[name = string("op_2192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2192_cast_fp16 = einsum(equation = var_2192_equation_0, values = (var_2028_cast_fp16, var_2147_cast_fp16))[name = string("op_2192_cast_fp16")]; + string var_2194_equation_0 = const()[name = string("op_2194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2194_cast_fp16 = einsum(equation = var_2194_equation_0, values = (var_2028_cast_fp16, var_2148_cast_fp16))[name = string("op_2194_cast_fp16")]; + string var_2196_equation_0 = const()[name = string("op_2196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2196_cast_fp16 = einsum(equation = var_2196_equation_0, values = (var_2028_cast_fp16, var_2149_cast_fp16))[name = string("op_2196_cast_fp16")]; + string var_2198_equation_0 = const()[name = string("op_2198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2198_cast_fp16 = einsum(equation = var_2198_equation_0, values = (var_2028_cast_fp16, var_2150_cast_fp16))[name = string("op_2198_cast_fp16")]; + bool var_2200_interleave_0 = const()[name = string("op_2200_interleave_0"), val = bool(false)]; + tensor var_2200_cast_fp16 = concat(axis = var_1725, interleave = var_2200_interleave_0, values = (var_2152_cast_fp16, var_2154_cast_fp16, var_2156_cast_fp16, var_2158_cast_fp16))[name = string("op_2200_cast_fp16")]; + bool var_2202_interleave_0 = const()[name = string("op_2202_interleave_0"), val = bool(false)]; + tensor var_2202_cast_fp16 = concat(axis = var_1725, interleave = var_2202_interleave_0, values = (var_2160_cast_fp16, var_2162_cast_fp16, var_2164_cast_fp16, var_2166_cast_fp16))[name = string("op_2202_cast_fp16")]; + bool var_2204_interleave_0 = const()[name = string("op_2204_interleave_0"), val = bool(false)]; + tensor var_2204_cast_fp16 = concat(axis = var_1725, interleave = var_2204_interleave_0, values = (var_2168_cast_fp16, var_2170_cast_fp16, var_2172_cast_fp16, var_2174_cast_fp16))[name = string("op_2204_cast_fp16")]; + bool var_2206_interleave_0 = const()[name = string("op_2206_interleave_0"), val = bool(false)]; + tensor var_2206_cast_fp16 = concat(axis = var_1725, interleave = var_2206_interleave_0, values = (var_2176_cast_fp16, var_2178_cast_fp16, var_2180_cast_fp16, var_2182_cast_fp16))[name = string("op_2206_cast_fp16")]; + bool var_2208_interleave_0 = const()[name = string("op_2208_interleave_0"), val = bool(false)]; + tensor var_2208_cast_fp16 = concat(axis = var_1725, interleave = var_2208_interleave_0, values = (var_2184_cast_fp16, var_2186_cast_fp16, var_2188_cast_fp16, var_2190_cast_fp16))[name = string("op_2208_cast_fp16")]; + bool var_2210_interleave_0 = const()[name = string("op_2210_interleave_0"), val = bool(false)]; + tensor var_2210_cast_fp16 = concat(axis = var_1725, interleave = var_2210_interleave_0, values = (var_2192_cast_fp16, var_2194_cast_fp16, var_2196_cast_fp16, var_2198_cast_fp16))[name = string("op_2210_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_1736, interleave = input_25_interleave_0, values = (var_2200_cast_fp16, var_2202_cast_fp16, var_2204_cast_fp16, var_2206_cast_fp16, var_2208_cast_fp16, var_2210_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13760256)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14055232)))]; + tensor obj_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_2229_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056064)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056896)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14057728)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15237440)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15240576)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16420288)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_2267_to_fp16 = const()[name = string("op_2267_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2267_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421120)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421952)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_2291_pad_type_0 = const()[name = string("op_2291_pad_type_0"), val = string("valid")]; + tensor var_2291_strides_0 = const()[name = string("op_2291_strides_0"), val = tensor([1, 1])]; + tensor var_2291_pad_0 = const()[name = string("op_2291_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2291_dilations_0 = const()[name = string("op_2291_dilations_0"), val = tensor([1, 1])]; + int32 var_2291_groups_0 = const()[name = string("op_2291_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16422784)))]; + tensor var_2291_cast_fp16 = conv(dilations = var_2291_dilations_0, groups = var_2291_groups_0, pad = var_2291_pad_0, pad_type = var_2291_pad_type_0, strides = var_2291_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2291_cast_fp16")]; + string var_2298_pad_type_0 = const()[name = string("op_2298_pad_type_0"), val = string("valid")]; + tensor var_2298_strides_0 = const()[name = string("op_2298_strides_0"), val = tensor([1, 1])]; + tensor var_2298_pad_0 = const()[name = string("op_2298_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2298_dilations_0 = const()[name = string("op_2298_dilations_0"), val = tensor([1, 1])]; + int32 var_2298_groups_0 = const()[name = string("op_2298_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16717760)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17012736)))]; + tensor var_2298_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_2298_dilations_0, groups = var_2298_groups_0, pad = var_2298_pad_0, pad_type = var_2298_pad_type_0, strides = var_2298_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2298_cast_fp16")]; + string var_2316_pad_type_0 = const()[name = string("op_2316_pad_type_0"), val = string("valid")]; + tensor var_2316_strides_0 = const()[name = string("op_2316_strides_0"), val = tensor([1, 1])]; + tensor var_2316_pad_0 = const()[name = string("op_2316_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2316_dilations_0 = const()[name = string("op_2316_dilations_0"), val = tensor([1, 1])]; + int32 var_2316_groups_0 = const()[name = string("op_2316_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17013568)))]; + tensor var_2316_cast_fp16 = conv(dilations = var_2316_dilations_0, groups = var_2316_groups_0, pad = var_2316_pad_0, pad_type = var_2316_pad_type_0, strides = var_2316_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2316_cast_fp16")]; + string var_2323_pad_type_0 = const()[name = string("op_2323_pad_type_0"), val = string("valid")]; + tensor var_2323_strides_0 = const()[name = string("op_2323_strides_0"), val = tensor([1, 1])]; + tensor var_2323_pad_0 = const()[name = string("op_2323_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2323_dilations_0 = const()[name = string("op_2323_dilations_0"), val = tensor([1, 1])]; + int32 var_2323_groups_0 = const()[name = string("op_2323_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17308544)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17603520)))]; + tensor var_2323_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_2323_dilations_0, groups = var_2323_groups_0, pad = var_2323_pad_0, pad_type = var_2323_pad_type_0, strides = var_2323_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2323_cast_fp16")]; + string var_2341_pad_type_0 = const()[name = string("op_2341_pad_type_0"), val = string("valid")]; + tensor var_2341_strides_0 = const()[name = string("op_2341_strides_0"), val = tensor([1, 1])]; + tensor var_2341_pad_0 = const()[name = string("op_2341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2341_dilations_0 = const()[name = string("op_2341_dilations_0"), val = tensor([1, 1])]; + int32 var_2341_groups_0 = const()[name = string("op_2341_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17604352)))]; + tensor var_2341_cast_fp16 = conv(dilations = var_2341_dilations_0, groups = var_2341_groups_0, pad = var_2341_pad_0, pad_type = var_2341_pad_type_0, strides = var_2341_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2341_cast_fp16")]; + string var_2348_pad_type_0 = const()[name = string("op_2348_pad_type_0"), val = string("valid")]; + tensor var_2348_strides_0 = const()[name = string("op_2348_strides_0"), val = tensor([1, 1])]; + tensor var_2348_pad_0 = const()[name = string("op_2348_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2348_dilations_0 = const()[name = string("op_2348_dilations_0"), val = tensor([1, 1])]; + int32 var_2348_groups_0 = const()[name = string("op_2348_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17899328)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18194304)))]; + tensor var_2348_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_2348_dilations_0, groups = var_2348_groups_0, pad = var_2348_pad_0, pad_type = var_2348_pad_type_0, strides = var_2348_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2348_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18195136)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18490112)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18785088)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_2378 = const()[name = string("op_2378"), val = int32(0)]; + bool input_35_interleave_0 = const()[name = string("input_35_interleave_0"), val = bool(false)]; + tensor input_35_cast_fp16 = concat(axis = var_2378, interleave = input_35_interleave_0, values = (var_2291_cast_fp16, var_2316_cast_fp16, var_2341_cast_fp16, k_cast_fp16))[name = string("input_35_cast_fp16")]; + int32 var_2381 = const()[name = string("op_2381"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_2381, interleave = input_interleave_0, values = (var_2298_cast_fp16, var_2323_cast_fp16, var_2348_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_2388_pad_0 = const()[name = string("op_2388_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_2388_mode_0 = const()[name = string("op_2388_mode_0"), val = string("constant")]; + fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_5_to_fp16, mode = var_2388_mode_0, pad = var_2388_pad_0, x = input_35_cast_fp16)[name = string("op_2388_cast_fp16")]; + tensor var_2394_pad_0 = const()[name = string("op_2394_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_2394_mode_0 = const()[name = string("op_2394_mode_0"), val = string("constant")]; + fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_6_to_fp16, mode = var_2394_mode_0, pad = var_2394_pad_0, x = input_cast_fp16)[name = string("op_2394_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-tiny.en/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..fcd7f0881b306a14f71c2ed0aaf74332240c593b --- /dev/null +++ b/openai_whisper-tiny.en/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b2e3e03c62664c5178e50afc7943ed83c2d7e00c21db4e1b953a47b5361ff8c +size 18785920 diff --git a/openai_whisper-tiny.en/LICENSE_NOTICE.txt b/openai_whisper-tiny.en/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-tiny.en/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a11fbb2cd75b96eb2120a672afefa298c2ef857b --- /dev/null +++ b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc05e563ee0c556e3f578e04be5fb67b4e7520124403f2561f39102f0f2b33d +size 243 diff --git a/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a3544b6644c1af93ca6bdabb67a1c51e80eaa552 --- /dev/null +++ b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ef11ea703011eab03287ec661f999e19c2c78cf67d531b5e6afa02e18f913d +size 328 diff --git a/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1a60dd494a857817b67d87cd920baa6824e74b61 --- /dev/null +++ b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cf4cd446f68b88655d00a7df7063aa46937a9bdd --- /dev/null +++ b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f7d28ffac464e9e7086a526930f0059187de8d01 --- /dev/null +++ b/openai_whisper-tiny.en/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83 +size 354080 diff --git a/openai_whisper-tiny.en/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-tiny.en/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..8657e28acfdcaa743cb06ea18661815f95fa0110 --- /dev/null +++ b/openai_whisper-tiny.en/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8709d424db3b08227c7dbee233605d0fee873882af81023d36d859d7ffa206ae +size 243 diff --git a/openai_whisper-tiny.en/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-tiny.en/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..574dc82c7a35a66197af08067eaa5ec9074e6e4b --- /dev/null +++ b/openai_whisper-tiny.en/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ca2c5b34d43b5f84088cd3ca412753bc59e1f132e05e8e3989a3c92d4101371 +size 754 diff --git a/openai_whisper-tiny.en/TextDecoder.mlmodelc/metadata.json b/openai_whisper-tiny.en/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..06f777eef8b02a53fad9e6474c69557d3e8d0364 --- /dev/null +++ b/openai_whisper-tiny.en/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51864)", + "shortDescription" : "", + "shape" : "[1, 1, 51864]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 1536, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 1536, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 8, + "Ios18.mul" : 16, + "Ios18.matmul" : 16, + "Ios18.batchNorm" : 13, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 29, + "Ios18.layerNorm" : 13, + "Ios18.reshape" : 32, + "Ios18.linear" : 1, + "Ios18.conv" : 32, + "Ios18.gelu" : 4, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 24, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-tiny.en/TextDecoder.mlmodelc/model.mil b/openai_whisper-tiny.en/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..a4c8ce213048345a1752bfb32f0692f45b6cc28c --- /dev/null +++ b/openai_whisper-tiny.en/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,697 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; + int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; + bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_26_cast_fp16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = input_ids, validate_indices = var_26_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_26_cast_fp16")]; + int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; + int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; + bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39831680)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_71")]; + tensor var_30_cast_fp16_cast_uint16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_30_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_30_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_26_cast_fp16, y = var_30_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_44_axes_0 = const()[name = string("op_44_axes_0"), val = tensor([2])]; + tensor var_44_cast_fp16 = expand_dims(axes = var_44_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_44_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_44_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1])]; + int32 var_49_axis_0 = const()[name = string("op_49_axis_0"), val = int32(0)]; + tensor var_49_cast_fp16_0, tensor var_49_cast_fp16_1, tensor var_49_cast_fp16_2, tensor var_49_cast_fp16_3 = split(axis = var_49_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_49_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1])]; + int32 var_56_axis_0 = const()[name = string("op_56_axis_0"), val = int32(0)]; + tensor var_56_cast_fp16_0, tensor var_56_cast_fp16_1, tensor var_56_cast_fp16_2, tensor var_56_cast_fp16_3 = split(axis = var_56_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_56_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 384, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 384, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_76 = const()[name = string("op_76"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_101_to_fp16 = const()[name = string("op_101_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_101_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40175808)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40176640)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40177472)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40178304)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40179136)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40474112)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40474944)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40769920)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41064896)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor([1])]; + tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = kv_cache_update_mask)[name = string("op_136_cast_fp16")]; + tensor var_137_axes_0 = const()[name = string("op_137_axes_0"), val = tensor([2])]; + tensor var_137_cast_fp16 = expand_dims(axes = var_137_axes_0, x = var_136_cast_fp16)[name = string("op_137_cast_fp16")]; + tensor var_139_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_139_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_49_cast_fp16_0, y = var_139_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_141_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_141_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_56_cast_fp16_0, y = var_141_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_144 = const()[name = string("op_144"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_144, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_146_to_fp16 = const()[name = string("op_146_to_fp16"), val = fp16(0x1p-3)]; + tensor var_147_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_146_to_fp16)[name = string("op_147_cast_fp16")]; + tensor var_148 = const()[name = string("op_148"), val = tensor([1, 6, 64, -1])]; + tensor var_149_cast_fp16 = reshape(shape = var_148, x = key_1_cast_fp16)[name = string("op_149_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_147_cast_fp16, y = var_149_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_153_axes_0 = const()[name = string("op_153_axes_0"), val = tensor([1])]; + tensor var_153_cast_fp16 = expand_dims(axes = var_153_axes_0, x = decoder_key_padding_mask)[name = string("op_153_cast_fp16")]; + tensor var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor([2])]; + tensor var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = var_153_cast_fp16)[name = string("op_154_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_157_cast_fp16 = softmax(axis = var_76, x = mh_w_3_cast_fp16)[name = string("op_157_cast_fp16")]; + tensor var_158 = const()[name = string("op_158"), val = tensor([1, 6, 64, -1])]; + tensor var_159_cast_fp16 = reshape(shape = var_158, x = value_1_cast_fp16)[name = string("op_159_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_159_cast_fp16, y = var_157_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_162 = const()[name = string("op_162"), val = tensor([1, 384, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_162, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41065728)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41360704)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_184_to_fp16 = const()[name = string("op_184_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_184_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41361536)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41362368)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41363200)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41658176)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_204 = const()[name = string("op_204"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_204, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_206_to_fp16 = const()[name = string("op_206_to_fp16"), val = fp16(0x1p-3)]; + tensor var_207_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_206_to_fp16)[name = string("op_207_cast_fp16")]; + tensor var_208 = const()[name = string("op_208"), val = tensor([1, 6, 64, -1])]; + tensor var_209_cast_fp16 = reshape(shape = var_208, x = obj_17_cast_fp16)[name = string("op_209_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_207_cast_fp16, y = var_209_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_213_axes_0 = const()[name = string("op_213_axes_0"), val = tensor([1])]; + tensor var_213_cast_fp16 = expand_dims(axes = var_213_axes_0, x = read_state_4)[name = string("op_213_cast_fp16")]; + tensor var_214_axes_0 = const()[name = string("op_214_axes_0"), val = tensor([2])]; + tensor var_214_cast_fp16 = expand_dims(axes = var_214_axes_0, x = var_213_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_76, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_218 = const()[name = string("op_218"), val = tensor([1, 6, 64, -1])]; + tensor var_219_cast_fp16 = reshape(shape = var_218, x = obj_19_cast_fp16)[name = string("op_219_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_219_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_222 = const()[name = string("op_222"), val = tensor([1, 384, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_222, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41659008)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41953984)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_240_to_fp16 = const()[name = string("op_240_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_240_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41954816)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41955648)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41956480)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43136192)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43139328)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44319040)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 384, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 384, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_285 = const()[name = string("op_285"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_310_to_fp16 = const()[name = string("op_310_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_310_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44319872)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44320704)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44321536)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44616512)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44617344)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44912320)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45207296)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_348_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_348_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_49_cast_fp16_1, y = var_348_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_350_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_350_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_56_cast_fp16_1, y = var_350_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_353 = const()[name = string("op_353"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_353, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_355_to_fp16 = const()[name = string("op_355_to_fp16"), val = fp16(0x1p-3)]; + tensor var_356_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_355_to_fp16)[name = string("op_356_cast_fp16")]; + tensor var_357 = const()[name = string("op_357"), val = tensor([1, 6, 64, -1])]; + tensor var_358_cast_fp16 = reshape(shape = var_357, x = key_3_cast_fp16)[name = string("op_358_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_356_cast_fp16, y = var_358_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_366_cast_fp16 = softmax(axis = var_285, x = mh_w_11_cast_fp16)[name = string("op_366_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([1, 6, 64, -1])]; + tensor var_368_cast_fp16 = reshape(shape = var_367, x = value_3_cast_fp16)[name = string("op_368_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_368_cast_fp16, y = var_366_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_371 = const()[name = string("op_371"), val = tensor([1, 384, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_371, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45208128)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45503104)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_393_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45503936)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45504768)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45505600)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45800576)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_413 = const()[name = string("op_413"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_413, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_415_to_fp16 = const()[name = string("op_415_to_fp16"), val = fp16(0x1p-3)]; + tensor var_416_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_415_to_fp16)[name = string("op_416_cast_fp16")]; + tensor var_417 = const()[name = string("op_417"), val = tensor([1, 6, 64, -1])]; + tensor var_418_cast_fp16 = reshape(shape = var_417, x = obj_35_cast_fp16)[name = string("op_418_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_416_cast_fp16, y = var_418_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_285, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_427 = const()[name = string("op_427"), val = tensor([1, 6, 64, -1])]; + tensor var_428_cast_fp16 = reshape(shape = var_427, x = obj_37_cast_fp16)[name = string("op_428_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_428_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_431 = const()[name = string("op_431"), val = tensor([1, 384, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_431, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45801408)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46096384)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_452_to_fp16 = const()[name = string("op_452_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_452_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46097216)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46098048)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46098880)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47278592)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47281728)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48461440)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 384, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 384, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_498 = const()[name = string("op_498"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_523_to_fp16 = const()[name = string("op_523_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_523_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48462272)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48463104)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48463936)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48758912)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48759744)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49054720)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49349696)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_561_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_561_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_49_cast_fp16_2, y = var_561_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_563_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_563_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_56_cast_fp16_2, y = var_563_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_566 = const()[name = string("op_566"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_566, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_568_to_fp16 = const()[name = string("op_568_to_fp16"), val = fp16(0x1p-3)]; + tensor var_569_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_568_to_fp16)[name = string("op_569_cast_fp16")]; + tensor var_570 = const()[name = string("op_570"), val = tensor([1, 6, 64, -1])]; + tensor var_571_cast_fp16 = reshape(shape = var_570, x = key_5_cast_fp16)[name = string("op_571_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_569_cast_fp16, y = var_571_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_579_cast_fp16 = softmax(axis = var_498, x = mh_w_19_cast_fp16)[name = string("op_579_cast_fp16")]; + tensor var_580 = const()[name = string("op_580"), val = tensor([1, 6, 64, -1])]; + tensor var_581_cast_fp16 = reshape(shape = var_580, x = value_5_cast_fp16)[name = string("op_581_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_581_cast_fp16, y = var_579_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_584 = const()[name = string("op_584"), val = tensor([1, 384, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_584, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49350528)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49645504)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_606_to_fp16 = const()[name = string("op_606_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_606_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49646336)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49647168)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49648000)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49942976)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_626 = const()[name = string("op_626"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_626, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_628_to_fp16 = const()[name = string("op_628_to_fp16"), val = fp16(0x1p-3)]; + tensor var_629_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_628_to_fp16)[name = string("op_629_cast_fp16")]; + tensor var_630 = const()[name = string("op_630"), val = tensor([1, 6, 64, -1])]; + tensor var_631_cast_fp16 = reshape(shape = var_630, x = obj_53_cast_fp16)[name = string("op_631_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_629_cast_fp16, y = var_631_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_498, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_640 = const()[name = string("op_640"), val = tensor([1, 6, 64, -1])]; + tensor var_641_cast_fp16 = reshape(shape = var_640, x = obj_55_cast_fp16)[name = string("op_641_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_641_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_644 = const()[name = string("op_644"), val = tensor([1, 384, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_644, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49943808)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50238784)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_665_to_fp16 = const()[name = string("op_665_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_665_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50239616)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50240448)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50241280)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51420992)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51424128)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52603840)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 384, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 384, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_711 = const()[name = string("op_711"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_736_to_fp16 = const()[name = string("op_736_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_736_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52604672)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52605504)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52606336)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52901312)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52902144)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53197120)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53492096)))]; + tensor current_value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_774_cast_fp16 = mul(x = current_key_cast_fp16, y = var_137_cast_fp16)[name = string("op_774_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_49_cast_fp16_3, y = var_774_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_776_cast_fp16 = mul(x = current_value_cast_fp16, y = var_137_cast_fp16)[name = string("op_776_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_56_cast_fp16_3, y = var_776_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_779 = const()[name = string("op_779"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_779, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_781_to_fp16 = const()[name = string("op_781_to_fp16"), val = fp16(0x1p-3)]; + tensor var_782_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_781_to_fp16)[name = string("op_782_cast_fp16")]; + tensor var_783 = const()[name = string("op_783"), val = tensor([1, 6, 64, -1])]; + tensor var_784_cast_fp16 = reshape(shape = var_783, x = key_cast_fp16)[name = string("op_784_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_782_cast_fp16, y = var_784_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_792_cast_fp16 = softmax(axis = var_711, x = mh_w_27_cast_fp16)[name = string("op_792_cast_fp16")]; + tensor var_793 = const()[name = string("op_793"), val = tensor([1, 6, 64, -1])]; + tensor var_794_cast_fp16 = reshape(shape = var_793, x = value_cast_fp16)[name = string("op_794_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_794_cast_fp16, y = var_792_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_797 = const()[name = string("op_797"), val = tensor([1, 384, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_797, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53492928)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53787904)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_819_to_fp16 = const()[name = string("op_819_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_819_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53788736)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53789568)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53790400)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54085376)))]; + tensor query_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_839, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_841_to_fp16 = const()[name = string("op_841_to_fp16"), val = fp16(0x1p-3)]; + tensor var_842_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_841_to_fp16)[name = string("op_842_cast_fp16")]; + tensor var_843 = const()[name = string("op_843"), val = tensor([1, 6, 64, -1])]; + tensor var_844_cast_fp16 = reshape(shape = var_843, x = obj_71_cast_fp16)[name = string("op_844_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_842_cast_fp16, y = var_844_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_711, x = mh_w_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_853 = const()[name = string("op_853"), val = tensor([1, 6, 64, -1])]; + tensor var_854_cast_fp16 = reshape(shape = var_853, x = obj_73_cast_fp16)[name = string("op_854_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_854_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_857 = const()[name = string("op_857"), val = tensor([1, 384, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_857, x = attn_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54086208)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54381184)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_878_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54382016)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54382848)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54383680)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55563392)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_37_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55566528)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56746240)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_921_to_fp16 = const()[name = string("op_921_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_921_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56747072)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56747904)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_932_axes_0 = const()[name = string("op_932_axes_0"), val = tensor([2])]; + tensor var_932_cast_fp16 = squeeze(axes = var_932_axes_0, x = hidden_states_cast_fp16)[name = string("op_932_cast_fp16")]; + tensor var_935_perm_0 = const()[name = string("op_935_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56748736)))]; + tensor var_935_cast_fp16 = transpose(perm = var_935_perm_0, x = var_932_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_935_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_939 = const()[name = string("op_939"), val = int32(1)]; + bool obj_81_interleave_0 = const()[name = string("obj_81_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_939, interleave = obj_81_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_cast_fp16))[name = string("obj_81_cast_fp16")]; + int32 var_942 = const()[name = string("op_942"), val = int32(1)]; + bool obj_83_interleave_0 = const()[name = string("obj_83_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_942, interleave = obj_83_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_cast_fp16))[name = string("obj_83_cast_fp16")]; + tensor var_953_begin_0 = const()[name = string("op_953_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_953_end_0 = const()[name = string("op_953_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_953_end_mask_0 = const()[name = string("op_953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = obj_41_cast_fp16)[name = string("op_953_cast_fp16")]; + tensor var_956_begin_0 = const()[name = string("op_956_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_956_end_0 = const()[name = string("op_956_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_956_end_mask_0 = const()[name = string("op_956_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_956_squeeze_mask_0 = const()[name = string("op_956_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_956_cast_fp16 = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, squeeze_mask = var_956_squeeze_mask_0, x = var_953_cast_fp16)[name = string("op_956_cast_fp16")]; + tensor var_971_begin_0 = const()[name = string("op_971_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_971_end_0 = const()[name = string("op_971_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_971_end_mask_0 = const()[name = string("op_971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_971_cast_fp16 = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = obj_59_cast_fp16)[name = string("op_971_cast_fp16")]; + tensor var_974_begin_0 = const()[name = string("op_974_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_974_end_0 = const()[name = string("op_974_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_974_end_mask_0 = const()[name = string("op_974_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_974_squeeze_mask_0 = const()[name = string("op_974_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_974_cast_fp16 = slice_by_index(begin = var_974_begin_0, end = var_974_end_0, end_mask = var_974_end_mask_0, squeeze_mask = var_974_squeeze_mask_0, x = var_971_cast_fp16)[name = string("op_974_cast_fp16")]; + tensor var_989_begin_0 = const()[name = string("op_989_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_989_end_0 = const()[name = string("op_989_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_989_end_mask_0 = const()[name = string("op_989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_989_cast_fp16 = slice_by_index(begin = var_989_begin_0, end = var_989_end_0, end_mask = var_989_end_mask_0, x = obj_59_cast_fp16)[name = string("op_989_cast_fp16")]; + tensor var_992_begin_0 = const()[name = string("op_992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_992_end_0 = const()[name = string("op_992_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_992_end_mask_0 = const()[name = string("op_992_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_992_squeeze_mask_0 = const()[name = string("op_992_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_992_cast_fp16 = slice_by_index(begin = var_992_begin_0, end = var_992_end_0, end_mask = var_992_end_mask_0, squeeze_mask = var_992_squeeze_mask_0, x = var_989_cast_fp16)[name = string("op_992_cast_fp16")]; + tensor var_1007_begin_0 = const()[name = string("op_1007_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1007_end_0 = const()[name = string("op_1007_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1007_end_mask_0 = const()[name = string("op_1007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1007_cast_fp16 = slice_by_index(begin = var_1007_begin_0, end = var_1007_end_0, end_mask = var_1007_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1007_cast_fp16")]; + tensor var_1010_begin_0 = const()[name = string("op_1010_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1010_end_0 = const()[name = string("op_1010_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1010_end_mask_0 = const()[name = string("op_1010_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1010_squeeze_mask_0 = const()[name = string("op_1010_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1010_cast_fp16 = slice_by_index(begin = var_1010_begin_0, end = var_1010_end_0, end_mask = var_1010_end_mask_0, squeeze_mask = var_1010_squeeze_mask_0, x = var_1007_cast_fp16)[name = string("op_1010_cast_fp16")]; + tensor var_1025_begin_0 = const()[name = string("op_1025_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1025_end_0 = const()[name = string("op_1025_end_0"), val = tensor([1, 2, 1, 1536])]; + tensor var_1025_end_mask_0 = const()[name = string("op_1025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1025_cast_fp16 = slice_by_index(begin = var_1025_begin_0, end = var_1025_end_0, end_mask = var_1025_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1025_cast_fp16")]; + tensor var_1028_begin_0 = const()[name = string("op_1028_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1028_end_0 = const()[name = string("op_1028_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1028_end_mask_0 = const()[name = string("op_1028_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1028_squeeze_mask_0 = const()[name = string("op_1028_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1028_cast_fp16 = slice_by_index(begin = var_1028_begin_0, end = var_1028_end_0, end_mask = var_1028_end_mask_0, squeeze_mask = var_1028_squeeze_mask_0, x = var_1025_cast_fp16)[name = string("op_1028_cast_fp16")]; + tensor var_1043_begin_0 = const()[name = string("op_1043_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1043_end_0 = const()[name = string("op_1043_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_1043_end_mask_0 = const()[name = string("op_1043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1043_cast_fp16 = slice_by_index(begin = var_1043_begin_0, end = var_1043_end_0, end_mask = var_1043_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1043_cast_fp16")]; + tensor var_1046_begin_0 = const()[name = string("op_1046_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1046_end_0 = const()[name = string("op_1046_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1046_end_mask_0 = const()[name = string("op_1046_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1046_squeeze_mask_0 = const()[name = string("op_1046_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1046_cast_fp16 = slice_by_index(begin = var_1046_begin_0, end = var_1046_end_0, end_mask = var_1046_end_mask_0, squeeze_mask = var_1046_squeeze_mask_0, x = var_1043_cast_fp16)[name = string("op_1046_cast_fp16")]; + tensor var_1061_begin_0 = const()[name = string("op_1061_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1061_end_0 = const()[name = string("op_1061_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_1061_end_mask_0 = const()[name = string("op_1061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1061_cast_fp16 = slice_by_index(begin = var_1061_begin_0, end = var_1061_end_0, end_mask = var_1061_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1061_cast_fp16")]; + tensor var_1064_begin_0 = const()[name = string("op_1064_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1064_end_0 = const()[name = string("op_1064_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1064_end_mask_0 = const()[name = string("op_1064_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1064_squeeze_mask_0 = const()[name = string("op_1064_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1064_cast_fp16 = slice_by_index(begin = var_1064_begin_0, end = var_1064_end_0, end_mask = var_1064_end_mask_0, squeeze_mask = var_1064_squeeze_mask_0, x = var_1061_cast_fp16)[name = string("op_1064_cast_fp16")]; + tensor var_1079_begin_0 = const()[name = string("op_1079_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1079_end_0 = const()[name = string("op_1079_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_1079_end_mask_0 = const()[name = string("op_1079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1079_cast_fp16 = slice_by_index(begin = var_1079_begin_0, end = var_1079_end_0, end_mask = var_1079_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1079_cast_fp16")]; + tensor var_1082_begin_0 = const()[name = string("op_1082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1082_end_0 = const()[name = string("op_1082_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1082_end_mask_0 = const()[name = string("op_1082_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1082_squeeze_mask_0 = const()[name = string("op_1082_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1082_cast_fp16 = slice_by_index(begin = var_1082_begin_0, end = var_1082_end_0, end_mask = var_1082_end_mask_0, squeeze_mask = var_1082_squeeze_mask_0, x = var_1079_cast_fp16)[name = string("op_1082_cast_fp16")]; + int32 var_1089 = const()[name = string("op_1089"), val = int32(1)]; + bool var_1090_interleave_0 = const()[name = string("op_1090_interleave_0"), val = bool(false)]; + tensor var_1090_cast_fp16 = concat(axis = var_1089, interleave = var_1090_interleave_0, values = (var_956_cast_fp16, var_974_cast_fp16, var_992_cast_fp16, var_1010_cast_fp16, var_1028_cast_fp16, var_1046_cast_fp16, var_1064_cast_fp16, var_1082_cast_fp16))[name = string("op_1090_cast_fp16")]; + bool var_1093 = const()[name = string("op_1093"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1093, x = var_1090_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-tiny.en/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-tiny.en/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f54dcd9f2274d197f36d91c93790021250e6ac8a --- /dev/null +++ b/openai_whisper-tiny.en/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:689c4044e74d94550721f73ffa2257118f66fd8feacc81a69bf45c45ea22f9a7 +size 56852528 diff --git a/openai_whisper-tiny.en/config.json b/openai_whisper-tiny.en/config.json new file mode 100644 index 0000000000000000000000000000000000000000..543fc73de943d608a1370ac9ae7916bdda2e76bc --- /dev/null +++ b/openai_whisper-tiny.en/config.json @@ -0,0 +1 @@ +{"_name_or_path": "openai/whisper-tiny", "activation_dropout": 0.0, "activation_function": "gelu", "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "d_model": 384, "decoder_attention_heads": 6, "decoder_ffn_dim": 1536, "decoder_layerdrop": 0.0, "decoder_layers": 4, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 6, "encoder_ffn_dim": 1536, "encoder_layerdrop": 0.0, "encoder_layers": 4, "eos_token_id": 50257, "forced_decoder_ids": [[1, 50259], [2, 50359], [3, 50363]], "init_std": 0.02, "is_encoder_decoder": true, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "model_type": "whisper", "num_hidden_layers": 4, "num_mel_bins": 80, "pad_token_id": 50257, "scale_embedding": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "torch_dtype": "float32", "transformers_version": "4.27.0.dev0", "use_cache": true, "vocab_size": 51865} \ No newline at end of file diff --git a/openai_whisper-tiny.en/generation_config.json b/openai_whisper-tiny.en/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..1d347402dfedd5e005cfab2688cb1b92fc971c7f --- /dev/null +++ b/openai_whisper-tiny.en/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50359]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50363, "pad_token_id": 50257, "prev_sot_token_id": 50361, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "task_to_id": {"transcribe": 50359, "translate": 50358}, "transformers_version": "4.31.0.dev0"} \ No newline at end of file diff --git a/openai_whisper-tiny/AudioEncoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-tiny/AudioEncoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..f115ac22a2e1abcad93ca64fd64ed778a9aeaa60 --- /dev/null +++ b/openai_whisper-tiny/AudioEncoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89a9aaa1f05fb2dbca4368c545605a472819dc4f6b919313ebf4b86a5e90c315 +size 243 diff --git a/openai_whisper-tiny/AudioEncoder.mlmodelc/coremldata.bin b/openai_whisper-tiny/AudioEncoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..8b61d15f0001c9ecd279614e2f8881fbe65a7274 --- /dev/null +++ b/openai_whisper-tiny/AudioEncoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:924bf280f20536a89020e7dd13efdec0a41bb22ce5fb9b2f7062384417accb52 +size 433 diff --git a/openai_whisper-tiny/AudioEncoder.mlmodelc/metadata.json b/openai_whisper-tiny/AudioEncoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..45385c1fd8200081ab3d24e79652954d54b9e575 --- /dev/null +++ b/openai_whisper-tiny/AudioEncoder.mlmodelc/metadata.json @@ -0,0 +1,91 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 384 × 1 × 1500)", + "shortDescription" : "", + "shape" : "[1, 384, 1, 1500]", + "name" : "encoder_output_embeds", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Pad" : 2, + "Ios18.batchNorm" : 9, + "Ios18.conv" : 34, + "Ios18.gelu" : 6, + "Ios18.concat" : 30, + "Ios16.einsum" : 192, + "Ios18.add" : 9, + "Ios18.softmax" : 96, + "Ios18.sliceByIndex" : 168, + "Ios18.layerNorm" : 9, + "Ios18.transpose" : 4, + "Ios18.mul" : 96 + }, + "computePrecision" : "Mixed (Float16, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "generatedClassName" : "AudioEncoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-tiny/AudioEncoder.mlmodelc/model.mil b/openai_whisper-tiny/AudioEncoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..5ad9dee5622f8ff5d216e86322c8cf1cdd4ccba3 --- /dev/null +++ b/openai_whisper-tiny/AudioEncoder.mlmodelc/model.mil @@ -0,0 +1,1787 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_58_pad_type_0 = const()[name = string("op_58_pad_type_0"), val = string("custom")]; + tensor var_58_pad_0 = const()[name = string("op_58_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_58_strides_0 = const()[name = string("op_58_strides_0"), val = tensor([1, 1])]; + tensor var_58_dilations_0 = const()[name = string("op_58_dilations_0"), val = tensor([1, 1])]; + int32 var_58_groups_0 = const()[name = string("op_58_groups_0"), val = int32(1)]; + tensor var_33_to_fp16 = const()[name = string("op_33_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_39_to_fp16 = const()[name = string("op_39_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184448)))]; + tensor var_58_cast_fp16 = conv(bias = var_39_to_fp16, dilations = var_58_dilations_0, groups = var_58_groups_0, pad = var_58_pad_0, pad_type = var_58_pad_type_0, strides = var_58_strides_0, weight = var_33_to_fp16, x = melspectrogram_features)[name = string("op_58_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_58_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_98_pad_type_0 = const()[name = string("op_98_pad_type_0"), val = string("custom")]; + tensor var_98_pad_0 = const()[name = string("op_98_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_98_strides_0 = const()[name = string("op_98_strides_0"), val = tensor([2, 2])]; + tensor var_98_dilations_0 = const()[name = string("op_98_dilations_0"), val = tensor([1, 1])]; + int32 var_98_groups_0 = const()[name = string("op_98_groups_0"), val = int32(1)]; + tensor var_73_to_fp16 = const()[name = string("op_73_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185280)))]; + tensor var_79_to_fp16 = const()[name = string("op_79_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070080)))]; + tensor var_98_cast_fp16 = conv(bias = var_79_to_fp16, dilations = var_98_dilations_0, groups = var_98_groups_0, pad = var_98_pad_0, pad_type = var_98_pad_type_0, strides = var_98_strides_0, weight = var_73_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_98_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_98_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_116_to_fp16 = const()[name = string("op_116_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070912)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_116_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_126 = const()[name = string("op_126"), val = int32(3)]; + int32 var_137 = const()[name = string("op_137"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_154_to_fp16 = const()[name = string("op_154_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_154_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2222976)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2223808)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2224640)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2225472)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2226304)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2521280)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2522112)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2817088)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112064)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_192_begin_0 = const()[name = string("op_192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_192_end_0 = const()[name = string("op_192_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_192_end_mask_0 = const()[name = string("op_192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_192_cast_fp16 = slice_by_index(begin = var_192_begin_0, end = var_192_end_0, end_mask = var_192_end_mask_0, x = query_1_cast_fp16)[name = string("op_192_cast_fp16")]; + tensor var_196_begin_0 = const()[name = string("op_196_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_196_end_0 = const()[name = string("op_196_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_196_end_mask_0 = const()[name = string("op_196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_196_cast_fp16 = slice_by_index(begin = var_196_begin_0, end = var_196_end_0, end_mask = var_196_end_mask_0, x = query_1_cast_fp16)[name = string("op_196_cast_fp16")]; + tensor var_200_begin_0 = const()[name = string("op_200_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_200_end_0 = const()[name = string("op_200_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_200_end_mask_0 = const()[name = string("op_200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_200_cast_fp16 = slice_by_index(begin = var_200_begin_0, end = var_200_end_0, end_mask = var_200_end_mask_0, x = query_1_cast_fp16)[name = string("op_200_cast_fp16")]; + tensor var_204_begin_0 = const()[name = string("op_204_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_204_end_0 = const()[name = string("op_204_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_204_end_mask_0 = const()[name = string("op_204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_204_cast_fp16 = slice_by_index(begin = var_204_begin_0, end = var_204_end_0, end_mask = var_204_end_mask_0, x = query_1_cast_fp16)[name = string("op_204_cast_fp16")]; + tensor var_208_begin_0 = const()[name = string("op_208_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_208_end_0 = const()[name = string("op_208_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_208_end_mask_0 = const()[name = string("op_208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_208_cast_fp16 = slice_by_index(begin = var_208_begin_0, end = var_208_end_0, end_mask = var_208_end_mask_0, x = query_1_cast_fp16)[name = string("op_208_cast_fp16")]; + tensor var_212_begin_0 = const()[name = string("op_212_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_212_end_0 = const()[name = string("op_212_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_212_end_mask_0 = const()[name = string("op_212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_212_cast_fp16 = slice_by_index(begin = var_212_begin_0, end = var_212_end_0, end_mask = var_212_end_mask_0, x = query_1_cast_fp16)[name = string("op_212_cast_fp16")]; + tensor var_221_begin_0 = const()[name = string("op_221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_221_end_0 = const()[name = string("op_221_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_221_end_mask_0 = const()[name = string("op_221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_221_cast_fp16 = slice_by_index(begin = var_221_begin_0, end = var_221_end_0, end_mask = var_221_end_mask_0, x = var_192_cast_fp16)[name = string("op_221_cast_fp16")]; + tensor var_228_begin_0 = const()[name = string("op_228_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_228_end_0 = const()[name = string("op_228_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_228_end_mask_0 = const()[name = string("op_228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_228_cast_fp16 = slice_by_index(begin = var_228_begin_0, end = var_228_end_0, end_mask = var_228_end_mask_0, x = var_192_cast_fp16)[name = string("op_228_cast_fp16")]; + tensor var_235_begin_0 = const()[name = string("op_235_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_235_end_0 = const()[name = string("op_235_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_235_end_mask_0 = const()[name = string("op_235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_235_cast_fp16 = slice_by_index(begin = var_235_begin_0, end = var_235_end_0, end_mask = var_235_end_mask_0, x = var_192_cast_fp16)[name = string("op_235_cast_fp16")]; + tensor var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = var_192_cast_fp16)[name = string("op_242_cast_fp16")]; + tensor var_249_begin_0 = const()[name = string("op_249_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_249_end_0 = const()[name = string("op_249_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_249_end_mask_0 = const()[name = string("op_249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_249_cast_fp16 = slice_by_index(begin = var_249_begin_0, end = var_249_end_0, end_mask = var_249_end_mask_0, x = var_196_cast_fp16)[name = string("op_249_cast_fp16")]; + tensor var_256_begin_0 = const()[name = string("op_256_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_256_end_0 = const()[name = string("op_256_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_256_end_mask_0 = const()[name = string("op_256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_256_cast_fp16 = slice_by_index(begin = var_256_begin_0, end = var_256_end_0, end_mask = var_256_end_mask_0, x = var_196_cast_fp16)[name = string("op_256_cast_fp16")]; + tensor var_263_begin_0 = const()[name = string("op_263_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_263_end_0 = const()[name = string("op_263_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_263_end_mask_0 = const()[name = string("op_263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_263_cast_fp16 = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = var_196_cast_fp16)[name = string("op_263_cast_fp16")]; + tensor var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = var_196_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_277_begin_0 = const()[name = string("op_277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_277_end_0 = const()[name = string("op_277_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_277_end_mask_0 = const()[name = string("op_277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_277_cast_fp16 = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_200_cast_fp16)[name = string("op_277_cast_fp16")]; + tensor var_284_begin_0 = const()[name = string("op_284_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_284_end_0 = const()[name = string("op_284_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_284_end_mask_0 = const()[name = string("op_284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_284_cast_fp16 = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_200_cast_fp16)[name = string("op_284_cast_fp16")]; + tensor var_291_begin_0 = const()[name = string("op_291_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_291_end_0 = const()[name = string("op_291_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_291_end_mask_0 = const()[name = string("op_291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_291_cast_fp16 = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_200_cast_fp16)[name = string("op_291_cast_fp16")]; + tensor var_298_begin_0 = const()[name = string("op_298_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_298_end_0 = const()[name = string("op_298_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_298_end_mask_0 = const()[name = string("op_298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_200_cast_fp16)[name = string("op_298_cast_fp16")]; + tensor var_305_begin_0 = const()[name = string("op_305_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_305_end_0 = const()[name = string("op_305_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_305_end_mask_0 = const()[name = string("op_305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_305_cast_fp16 = slice_by_index(begin = var_305_begin_0, end = var_305_end_0, end_mask = var_305_end_mask_0, x = var_204_cast_fp16)[name = string("op_305_cast_fp16")]; + tensor var_312_begin_0 = const()[name = string("op_312_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_312_end_0 = const()[name = string("op_312_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_312_end_mask_0 = const()[name = string("op_312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_312_cast_fp16 = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = var_204_cast_fp16)[name = string("op_312_cast_fp16")]; + tensor var_319_begin_0 = const()[name = string("op_319_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_319_end_0 = const()[name = string("op_319_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_319_end_mask_0 = const()[name = string("op_319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = var_204_cast_fp16)[name = string("op_319_cast_fp16")]; + tensor var_326_begin_0 = const()[name = string("op_326_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_326_end_0 = const()[name = string("op_326_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_326_end_mask_0 = const()[name = string("op_326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = var_204_cast_fp16)[name = string("op_326_cast_fp16")]; + tensor var_333_begin_0 = const()[name = string("op_333_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_333_end_0 = const()[name = string("op_333_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_333_end_mask_0 = const()[name = string("op_333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_333_cast_fp16 = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = var_208_cast_fp16)[name = string("op_333_cast_fp16")]; + tensor var_340_begin_0 = const()[name = string("op_340_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_340_end_0 = const()[name = string("op_340_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_340_end_mask_0 = const()[name = string("op_340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_340_cast_fp16 = slice_by_index(begin = var_340_begin_0, end = var_340_end_0, end_mask = var_340_end_mask_0, x = var_208_cast_fp16)[name = string("op_340_cast_fp16")]; + tensor var_347_begin_0 = const()[name = string("op_347_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_347_end_0 = const()[name = string("op_347_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_347_end_mask_0 = const()[name = string("op_347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_347_cast_fp16 = slice_by_index(begin = var_347_begin_0, end = var_347_end_0, end_mask = var_347_end_mask_0, x = var_208_cast_fp16)[name = string("op_347_cast_fp16")]; + tensor var_354_begin_0 = const()[name = string("op_354_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_354_end_0 = const()[name = string("op_354_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_354_end_mask_0 = const()[name = string("op_354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_354_cast_fp16 = slice_by_index(begin = var_354_begin_0, end = var_354_end_0, end_mask = var_354_end_mask_0, x = var_208_cast_fp16)[name = string("op_354_cast_fp16")]; + tensor var_361_begin_0 = const()[name = string("op_361_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_361_end_0 = const()[name = string("op_361_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_361_end_mask_0 = const()[name = string("op_361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_361_cast_fp16 = slice_by_index(begin = var_361_begin_0, end = var_361_end_0, end_mask = var_361_end_mask_0, x = var_212_cast_fp16)[name = string("op_361_cast_fp16")]; + tensor var_368_begin_0 = const()[name = string("op_368_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_368_end_0 = const()[name = string("op_368_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_368_end_mask_0 = const()[name = string("op_368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_368_cast_fp16 = slice_by_index(begin = var_368_begin_0, end = var_368_end_0, end_mask = var_368_end_mask_0, x = var_212_cast_fp16)[name = string("op_368_cast_fp16")]; + tensor var_375_begin_0 = const()[name = string("op_375_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_375_end_0 = const()[name = string("op_375_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_375_end_mask_0 = const()[name = string("op_375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_375_cast_fp16 = slice_by_index(begin = var_375_begin_0, end = var_375_end_0, end_mask = var_375_end_mask_0, x = var_212_cast_fp16)[name = string("op_375_cast_fp16")]; + tensor var_382_begin_0 = const()[name = string("op_382_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_382_end_0 = const()[name = string("op_382_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_382_end_mask_0 = const()[name = string("op_382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_382_cast_fp16 = slice_by_index(begin = var_382_begin_0, end = var_382_end_0, end_mask = var_382_end_mask_0, x = var_212_cast_fp16)[name = string("op_382_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_387_begin_0 = const()[name = string("op_387_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_387_end_0 = const()[name = string("op_387_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_387_end_mask_0 = const()[name = string("op_387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_3")]; + tensor var_387_cast_fp16 = slice_by_index(begin = var_387_begin_0, end = var_387_end_0, end_mask = var_387_end_mask_0, x = k_1_cast_fp16)[name = string("op_387_cast_fp16")]; + tensor var_391_begin_0 = const()[name = string("op_391_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_391_end_0 = const()[name = string("op_391_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_391_end_mask_0 = const()[name = string("op_391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_391_cast_fp16 = slice_by_index(begin = var_391_begin_0, end = var_391_end_0, end_mask = var_391_end_mask_0, x = k_1_cast_fp16)[name = string("op_391_cast_fp16")]; + tensor var_395_begin_0 = const()[name = string("op_395_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_395_end_0 = const()[name = string("op_395_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_395_end_mask_0 = const()[name = string("op_395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_395_cast_fp16 = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = k_1_cast_fp16)[name = string("op_395_cast_fp16")]; + tensor var_399_begin_0 = const()[name = string("op_399_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_399_end_0 = const()[name = string("op_399_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_399_end_mask_0 = const()[name = string("op_399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_399_cast_fp16 = slice_by_index(begin = var_399_begin_0, end = var_399_end_0, end_mask = var_399_end_mask_0, x = k_1_cast_fp16)[name = string("op_399_cast_fp16")]; + tensor var_403_begin_0 = const()[name = string("op_403_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_403_end_0 = const()[name = string("op_403_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_403_end_mask_0 = const()[name = string("op_403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_403_cast_fp16 = slice_by_index(begin = var_403_begin_0, end = var_403_end_0, end_mask = var_403_end_mask_0, x = k_1_cast_fp16)[name = string("op_403_cast_fp16")]; + tensor var_407_begin_0 = const()[name = string("op_407_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_407_end_0 = const()[name = string("op_407_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_407_end_mask_0 = const()[name = string("op_407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = k_1_cast_fp16)[name = string("op_407_cast_fp16")]; + tensor var_409_begin_0 = const()[name = string("op_409_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_409_end_0 = const()[name = string("op_409_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_409_end_mask_0 = const()[name = string("op_409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_409_cast_fp16 = slice_by_index(begin = var_409_begin_0, end = var_409_end_0, end_mask = var_409_end_mask_0, x = value_1_cast_fp16)[name = string("op_409_cast_fp16")]; + tensor var_413_begin_0 = const()[name = string("op_413_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_413_end_0 = const()[name = string("op_413_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_413_end_mask_0 = const()[name = string("op_413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, x = value_1_cast_fp16)[name = string("op_413_cast_fp16")]; + tensor var_417_begin_0 = const()[name = string("op_417_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_417_end_0 = const()[name = string("op_417_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_417_end_mask_0 = const()[name = string("op_417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_417_cast_fp16 = slice_by_index(begin = var_417_begin_0, end = var_417_end_0, end_mask = var_417_end_mask_0, x = value_1_cast_fp16)[name = string("op_417_cast_fp16")]; + tensor var_421_begin_0 = const()[name = string("op_421_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_421_end_0 = const()[name = string("op_421_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_421_end_mask_0 = const()[name = string("op_421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = value_1_cast_fp16)[name = string("op_421_cast_fp16")]; + tensor var_425_begin_0 = const()[name = string("op_425_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_425_end_0 = const()[name = string("op_425_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_425_end_mask_0 = const()[name = string("op_425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_425_cast_fp16 = slice_by_index(begin = var_425_begin_0, end = var_425_end_0, end_mask = var_425_end_mask_0, x = value_1_cast_fp16)[name = string("op_425_cast_fp16")]; + tensor var_429_begin_0 = const()[name = string("op_429_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_429_end_0 = const()[name = string("op_429_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_429_end_mask_0 = const()[name = string("op_429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_429_cast_fp16 = slice_by_index(begin = var_429_begin_0, end = var_429_end_0, end_mask = var_429_end_mask_0, x = value_1_cast_fp16)[name = string("op_429_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_387_cast_fp16, var_221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_387_cast_fp16, var_228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_387_cast_fp16, var_235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_387_cast_fp16, var_242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_391_cast_fp16, var_249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_391_cast_fp16, var_256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_391_cast_fp16, var_263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_391_cast_fp16, var_270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_395_cast_fp16, var_277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_395_cast_fp16, var_284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_395_cast_fp16, var_291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_395_cast_fp16, var_298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_399_cast_fp16, var_305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_399_cast_fp16, var_312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_399_cast_fp16, var_319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_399_cast_fp16, var_326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_403_cast_fp16, var_333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_403_cast_fp16, var_340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_403_cast_fp16, var_347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_403_cast_fp16, var_354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_407_cast_fp16, var_361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_407_cast_fp16, var_368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_407_cast_fp16, var_375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_407_cast_fp16, var_382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + fp16 var_480_to_fp16 = const()[name = string("op_480_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_480_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_482_to_fp16 = const()[name = string("op_482_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_482_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_484_to_fp16 = const()[name = string("op_484_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_484_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_486_to_fp16 = const()[name = string("op_486_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_486_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_488_to_fp16 = const()[name = string("op_488_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_488_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_490_to_fp16 = const()[name = string("op_490_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_490_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_492_to_fp16 = const()[name = string("op_492_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_492_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_494_to_fp16 = const()[name = string("op_494_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_494_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_496_to_fp16 = const()[name = string("op_496_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_496_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_498_to_fp16 = const()[name = string("op_498_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_498_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_500_to_fp16 = const()[name = string("op_500_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_500_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_502_to_fp16 = const()[name = string("op_502_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_502_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_504_to_fp16 = const()[name = string("op_504_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_504_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_506_to_fp16 = const()[name = string("op_506_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_506_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_508_to_fp16 = const()[name = string("op_508_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_508_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_510_to_fp16 = const()[name = string("op_510_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_510_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_512_to_fp16 = const()[name = string("op_512_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_512_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_514_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_516_to_fp16 = const()[name = string("op_516_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_516_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_518_to_fp16 = const()[name = string("op_518_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_518_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_520_to_fp16 = const()[name = string("op_520_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_520_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_522_to_fp16 = const()[name = string("op_522_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_522_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_524_to_fp16 = const()[name = string("op_524_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_524_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_526_to_fp16 = const()[name = string("op_526_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_526_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + tensor var_528_cast_fp16 = softmax(axis = var_137, x = aw_chunk_1_cast_fp16)[name = string("op_528_cast_fp16")]; + tensor var_529_cast_fp16 = softmax(axis = var_137, x = aw_chunk_3_cast_fp16)[name = string("op_529_cast_fp16")]; + tensor var_530_cast_fp16 = softmax(axis = var_137, x = aw_chunk_5_cast_fp16)[name = string("op_530_cast_fp16")]; + tensor var_531_cast_fp16 = softmax(axis = var_137, x = aw_chunk_7_cast_fp16)[name = string("op_531_cast_fp16")]; + tensor var_532_cast_fp16 = softmax(axis = var_137, x = aw_chunk_9_cast_fp16)[name = string("op_532_cast_fp16")]; + tensor var_533_cast_fp16 = softmax(axis = var_137, x = aw_chunk_11_cast_fp16)[name = string("op_533_cast_fp16")]; + tensor var_534_cast_fp16 = softmax(axis = var_137, x = aw_chunk_13_cast_fp16)[name = string("op_534_cast_fp16")]; + tensor var_535_cast_fp16 = softmax(axis = var_137, x = aw_chunk_15_cast_fp16)[name = string("op_535_cast_fp16")]; + tensor var_536_cast_fp16 = softmax(axis = var_137, x = aw_chunk_17_cast_fp16)[name = string("op_536_cast_fp16")]; + tensor var_537_cast_fp16 = softmax(axis = var_137, x = aw_chunk_19_cast_fp16)[name = string("op_537_cast_fp16")]; + tensor var_538_cast_fp16 = softmax(axis = var_137, x = aw_chunk_21_cast_fp16)[name = string("op_538_cast_fp16")]; + tensor var_539_cast_fp16 = softmax(axis = var_137, x = aw_chunk_23_cast_fp16)[name = string("op_539_cast_fp16")]; + tensor var_540_cast_fp16 = softmax(axis = var_137, x = aw_chunk_25_cast_fp16)[name = string("op_540_cast_fp16")]; + tensor var_541_cast_fp16 = softmax(axis = var_137, x = aw_chunk_27_cast_fp16)[name = string("op_541_cast_fp16")]; + tensor var_542_cast_fp16 = softmax(axis = var_137, x = aw_chunk_29_cast_fp16)[name = string("op_542_cast_fp16")]; + tensor var_543_cast_fp16 = softmax(axis = var_137, x = aw_chunk_31_cast_fp16)[name = string("op_543_cast_fp16")]; + tensor var_544_cast_fp16 = softmax(axis = var_137, x = aw_chunk_33_cast_fp16)[name = string("op_544_cast_fp16")]; + tensor var_545_cast_fp16 = softmax(axis = var_137, x = aw_chunk_35_cast_fp16)[name = string("op_545_cast_fp16")]; + tensor var_546_cast_fp16 = softmax(axis = var_137, x = aw_chunk_37_cast_fp16)[name = string("op_546_cast_fp16")]; + tensor var_547_cast_fp16 = softmax(axis = var_137, x = aw_chunk_39_cast_fp16)[name = string("op_547_cast_fp16")]; + tensor var_548_cast_fp16 = softmax(axis = var_137, x = aw_chunk_41_cast_fp16)[name = string("op_548_cast_fp16")]; + tensor var_549_cast_fp16 = softmax(axis = var_137, x = aw_chunk_43_cast_fp16)[name = string("op_549_cast_fp16")]; + tensor var_550_cast_fp16 = softmax(axis = var_137, x = aw_chunk_45_cast_fp16)[name = string("op_550_cast_fp16")]; + tensor var_551_cast_fp16 = softmax(axis = var_137, x = aw_chunk_47_cast_fp16)[name = string("op_551_cast_fp16")]; + string var_553_equation_0 = const()[name = string("op_553_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_553_cast_fp16 = einsum(equation = var_553_equation_0, values = (var_409_cast_fp16, var_528_cast_fp16))[name = string("op_553_cast_fp16")]; + string var_555_equation_0 = const()[name = string("op_555_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_409_cast_fp16, var_529_cast_fp16))[name = string("op_555_cast_fp16")]; + string var_557_equation_0 = const()[name = string("op_557_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_557_cast_fp16 = einsum(equation = var_557_equation_0, values = (var_409_cast_fp16, var_530_cast_fp16))[name = string("op_557_cast_fp16")]; + string var_559_equation_0 = const()[name = string("op_559_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_409_cast_fp16, var_531_cast_fp16))[name = string("op_559_cast_fp16")]; + string var_561_equation_0 = const()[name = string("op_561_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_561_cast_fp16 = einsum(equation = var_561_equation_0, values = (var_413_cast_fp16, var_532_cast_fp16))[name = string("op_561_cast_fp16")]; + string var_563_equation_0 = const()[name = string("op_563_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_413_cast_fp16, var_533_cast_fp16))[name = string("op_563_cast_fp16")]; + string var_565_equation_0 = const()[name = string("op_565_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_565_cast_fp16 = einsum(equation = var_565_equation_0, values = (var_413_cast_fp16, var_534_cast_fp16))[name = string("op_565_cast_fp16")]; + string var_567_equation_0 = const()[name = string("op_567_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_413_cast_fp16, var_535_cast_fp16))[name = string("op_567_cast_fp16")]; + string var_569_equation_0 = const()[name = string("op_569_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_569_cast_fp16 = einsum(equation = var_569_equation_0, values = (var_417_cast_fp16, var_536_cast_fp16))[name = string("op_569_cast_fp16")]; + string var_571_equation_0 = const()[name = string("op_571_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_417_cast_fp16, var_537_cast_fp16))[name = string("op_571_cast_fp16")]; + string var_573_equation_0 = const()[name = string("op_573_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_573_cast_fp16 = einsum(equation = var_573_equation_0, values = (var_417_cast_fp16, var_538_cast_fp16))[name = string("op_573_cast_fp16")]; + string var_575_equation_0 = const()[name = string("op_575_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_417_cast_fp16, var_539_cast_fp16))[name = string("op_575_cast_fp16")]; + string var_577_equation_0 = const()[name = string("op_577_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_577_cast_fp16 = einsum(equation = var_577_equation_0, values = (var_421_cast_fp16, var_540_cast_fp16))[name = string("op_577_cast_fp16")]; + string var_579_equation_0 = const()[name = string("op_579_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_579_cast_fp16 = einsum(equation = var_579_equation_0, values = (var_421_cast_fp16, var_541_cast_fp16))[name = string("op_579_cast_fp16")]; + string var_581_equation_0 = const()[name = string("op_581_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_581_cast_fp16 = einsum(equation = var_581_equation_0, values = (var_421_cast_fp16, var_542_cast_fp16))[name = string("op_581_cast_fp16")]; + string var_583_equation_0 = const()[name = string("op_583_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_583_cast_fp16 = einsum(equation = var_583_equation_0, values = (var_421_cast_fp16, var_543_cast_fp16))[name = string("op_583_cast_fp16")]; + string var_585_equation_0 = const()[name = string("op_585_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_585_cast_fp16 = einsum(equation = var_585_equation_0, values = (var_425_cast_fp16, var_544_cast_fp16))[name = string("op_585_cast_fp16")]; + string var_587_equation_0 = const()[name = string("op_587_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_587_cast_fp16 = einsum(equation = var_587_equation_0, values = (var_425_cast_fp16, var_545_cast_fp16))[name = string("op_587_cast_fp16")]; + string var_589_equation_0 = const()[name = string("op_589_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_589_cast_fp16 = einsum(equation = var_589_equation_0, values = (var_425_cast_fp16, var_546_cast_fp16))[name = string("op_589_cast_fp16")]; + string var_591_equation_0 = const()[name = string("op_591_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_591_cast_fp16 = einsum(equation = var_591_equation_0, values = (var_425_cast_fp16, var_547_cast_fp16))[name = string("op_591_cast_fp16")]; + string var_593_equation_0 = const()[name = string("op_593_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_593_cast_fp16 = einsum(equation = var_593_equation_0, values = (var_429_cast_fp16, var_548_cast_fp16))[name = string("op_593_cast_fp16")]; + string var_595_equation_0 = const()[name = string("op_595_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_595_cast_fp16 = einsum(equation = var_595_equation_0, values = (var_429_cast_fp16, var_549_cast_fp16))[name = string("op_595_cast_fp16")]; + string var_597_equation_0 = const()[name = string("op_597_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_597_cast_fp16 = einsum(equation = var_597_equation_0, values = (var_429_cast_fp16, var_550_cast_fp16))[name = string("op_597_cast_fp16")]; + string var_599_equation_0 = const()[name = string("op_599_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_599_cast_fp16 = einsum(equation = var_599_equation_0, values = (var_429_cast_fp16, var_551_cast_fp16))[name = string("op_599_cast_fp16")]; + bool var_601_interleave_0 = const()[name = string("op_601_interleave_0"), val = bool(false)]; + tensor var_601_cast_fp16 = concat(axis = var_126, interleave = var_601_interleave_0, values = (var_553_cast_fp16, var_555_cast_fp16, var_557_cast_fp16, var_559_cast_fp16))[name = string("op_601_cast_fp16")]; + bool var_603_interleave_0 = const()[name = string("op_603_interleave_0"), val = bool(false)]; + tensor var_603_cast_fp16 = concat(axis = var_126, interleave = var_603_interleave_0, values = (var_561_cast_fp16, var_563_cast_fp16, var_565_cast_fp16, var_567_cast_fp16))[name = string("op_603_cast_fp16")]; + bool var_605_interleave_0 = const()[name = string("op_605_interleave_0"), val = bool(false)]; + tensor var_605_cast_fp16 = concat(axis = var_126, interleave = var_605_interleave_0, values = (var_569_cast_fp16, var_571_cast_fp16, var_573_cast_fp16, var_575_cast_fp16))[name = string("op_605_cast_fp16")]; + bool var_607_interleave_0 = const()[name = string("op_607_interleave_0"), val = bool(false)]; + tensor var_607_cast_fp16 = concat(axis = var_126, interleave = var_607_interleave_0, values = (var_577_cast_fp16, var_579_cast_fp16, var_581_cast_fp16, var_583_cast_fp16))[name = string("op_607_cast_fp16")]; + bool var_609_interleave_0 = const()[name = string("op_609_interleave_0"), val = bool(false)]; + tensor var_609_cast_fp16 = concat(axis = var_126, interleave = var_609_interleave_0, values = (var_585_cast_fp16, var_587_cast_fp16, var_589_cast_fp16, var_591_cast_fp16))[name = string("op_609_cast_fp16")]; + bool var_611_interleave_0 = const()[name = string("op_611_interleave_0"), val = bool(false)]; + tensor var_611_cast_fp16 = concat(axis = var_126, interleave = var_611_interleave_0, values = (var_593_cast_fp16, var_595_cast_fp16, var_597_cast_fp16, var_599_cast_fp16))[name = string("op_611_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_137, interleave = input_1_interleave_0, values = (var_601_cast_fp16, var_603_cast_fp16, var_605_cast_fp16, var_607_cast_fp16, var_609_cast_fp16, var_611_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112896)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407872)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_630_to_fp16 = const()[name = string("op_630_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_630_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3408704)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3409536)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3410368)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4590080)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4593216)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772928)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_659 = const()[name = string("op_659"), val = int32(3)]; + int32 var_670 = const()[name = string("op_670"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_687_to_fp16 = const()[name = string("op_687_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_687_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5773760)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5774592)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5775424)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6070400)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6071232)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6366208)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6661184)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_725_begin_0 = const()[name = string("op_725_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_725_end_0 = const()[name = string("op_725_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_725_end_mask_0 = const()[name = string("op_725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_725_cast_fp16 = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = query_3_cast_fp16)[name = string("op_725_cast_fp16")]; + tensor var_729_begin_0 = const()[name = string("op_729_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_729_end_0 = const()[name = string("op_729_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_729_end_mask_0 = const()[name = string("op_729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_729_cast_fp16 = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = query_3_cast_fp16)[name = string("op_729_cast_fp16")]; + tensor var_733_begin_0 = const()[name = string("op_733_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_733_end_0 = const()[name = string("op_733_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_733_end_mask_0 = const()[name = string("op_733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_733_cast_fp16 = slice_by_index(begin = var_733_begin_0, end = var_733_end_0, end_mask = var_733_end_mask_0, x = query_3_cast_fp16)[name = string("op_733_cast_fp16")]; + tensor var_737_begin_0 = const()[name = string("op_737_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_737_end_0 = const()[name = string("op_737_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_737_end_mask_0 = const()[name = string("op_737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_737_cast_fp16 = slice_by_index(begin = var_737_begin_0, end = var_737_end_0, end_mask = var_737_end_mask_0, x = query_3_cast_fp16)[name = string("op_737_cast_fp16")]; + tensor var_741_begin_0 = const()[name = string("op_741_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_741_end_0 = const()[name = string("op_741_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_741_end_mask_0 = const()[name = string("op_741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_741_cast_fp16 = slice_by_index(begin = var_741_begin_0, end = var_741_end_0, end_mask = var_741_end_mask_0, x = query_3_cast_fp16)[name = string("op_741_cast_fp16")]; + tensor var_745_begin_0 = const()[name = string("op_745_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_745_end_0 = const()[name = string("op_745_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_745_end_mask_0 = const()[name = string("op_745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_745_cast_fp16 = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = query_3_cast_fp16)[name = string("op_745_cast_fp16")]; + tensor var_754_begin_0 = const()[name = string("op_754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_754_end_0 = const()[name = string("op_754_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_754_end_mask_0 = const()[name = string("op_754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_754_cast_fp16 = slice_by_index(begin = var_754_begin_0, end = var_754_end_0, end_mask = var_754_end_mask_0, x = var_725_cast_fp16)[name = string("op_754_cast_fp16")]; + tensor var_761_begin_0 = const()[name = string("op_761_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_761_end_0 = const()[name = string("op_761_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_761_end_mask_0 = const()[name = string("op_761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_761_cast_fp16 = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = var_725_cast_fp16)[name = string("op_761_cast_fp16")]; + tensor var_768_begin_0 = const()[name = string("op_768_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_768_end_0 = const()[name = string("op_768_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_768_end_mask_0 = const()[name = string("op_768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_768_cast_fp16 = slice_by_index(begin = var_768_begin_0, end = var_768_end_0, end_mask = var_768_end_mask_0, x = var_725_cast_fp16)[name = string("op_768_cast_fp16")]; + tensor var_775_begin_0 = const()[name = string("op_775_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_775_end_0 = const()[name = string("op_775_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_775_end_mask_0 = const()[name = string("op_775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_775_cast_fp16 = slice_by_index(begin = var_775_begin_0, end = var_775_end_0, end_mask = var_775_end_mask_0, x = var_725_cast_fp16)[name = string("op_775_cast_fp16")]; + tensor var_782_begin_0 = const()[name = string("op_782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_782_end_0 = const()[name = string("op_782_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_782_end_mask_0 = const()[name = string("op_782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_782_cast_fp16 = slice_by_index(begin = var_782_begin_0, end = var_782_end_0, end_mask = var_782_end_mask_0, x = var_729_cast_fp16)[name = string("op_782_cast_fp16")]; + tensor var_789_begin_0 = const()[name = string("op_789_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_789_end_0 = const()[name = string("op_789_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_789_end_mask_0 = const()[name = string("op_789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_789_cast_fp16 = slice_by_index(begin = var_789_begin_0, end = var_789_end_0, end_mask = var_789_end_mask_0, x = var_729_cast_fp16)[name = string("op_789_cast_fp16")]; + tensor var_796_begin_0 = const()[name = string("op_796_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_796_end_0 = const()[name = string("op_796_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_796_end_mask_0 = const()[name = string("op_796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_796_cast_fp16 = slice_by_index(begin = var_796_begin_0, end = var_796_end_0, end_mask = var_796_end_mask_0, x = var_729_cast_fp16)[name = string("op_796_cast_fp16")]; + tensor var_803_begin_0 = const()[name = string("op_803_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_803_end_0 = const()[name = string("op_803_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_803_end_mask_0 = const()[name = string("op_803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_803_cast_fp16 = slice_by_index(begin = var_803_begin_0, end = var_803_end_0, end_mask = var_803_end_mask_0, x = var_729_cast_fp16)[name = string("op_803_cast_fp16")]; + tensor var_810_begin_0 = const()[name = string("op_810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_810_end_0 = const()[name = string("op_810_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_810_end_mask_0 = const()[name = string("op_810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_810_cast_fp16 = slice_by_index(begin = var_810_begin_0, end = var_810_end_0, end_mask = var_810_end_mask_0, x = var_733_cast_fp16)[name = string("op_810_cast_fp16")]; + tensor var_817_begin_0 = const()[name = string("op_817_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_817_end_0 = const()[name = string("op_817_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_817_end_mask_0 = const()[name = string("op_817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_817_cast_fp16 = slice_by_index(begin = var_817_begin_0, end = var_817_end_0, end_mask = var_817_end_mask_0, x = var_733_cast_fp16)[name = string("op_817_cast_fp16")]; + tensor var_824_begin_0 = const()[name = string("op_824_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_824_end_0 = const()[name = string("op_824_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_824_end_mask_0 = const()[name = string("op_824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_824_cast_fp16 = slice_by_index(begin = var_824_begin_0, end = var_824_end_0, end_mask = var_824_end_mask_0, x = var_733_cast_fp16)[name = string("op_824_cast_fp16")]; + tensor var_831_begin_0 = const()[name = string("op_831_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_831_end_0 = const()[name = string("op_831_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_831_end_mask_0 = const()[name = string("op_831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_831_cast_fp16 = slice_by_index(begin = var_831_begin_0, end = var_831_end_0, end_mask = var_831_end_mask_0, x = var_733_cast_fp16)[name = string("op_831_cast_fp16")]; + tensor var_838_begin_0 = const()[name = string("op_838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_838_end_0 = const()[name = string("op_838_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_838_end_mask_0 = const()[name = string("op_838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_838_cast_fp16 = slice_by_index(begin = var_838_begin_0, end = var_838_end_0, end_mask = var_838_end_mask_0, x = var_737_cast_fp16)[name = string("op_838_cast_fp16")]; + tensor var_845_begin_0 = const()[name = string("op_845_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_845_end_0 = const()[name = string("op_845_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_845_end_mask_0 = const()[name = string("op_845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_845_cast_fp16 = slice_by_index(begin = var_845_begin_0, end = var_845_end_0, end_mask = var_845_end_mask_0, x = var_737_cast_fp16)[name = string("op_845_cast_fp16")]; + tensor var_852_begin_0 = const()[name = string("op_852_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_852_end_0 = const()[name = string("op_852_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_852_end_mask_0 = const()[name = string("op_852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_852_cast_fp16 = slice_by_index(begin = var_852_begin_0, end = var_852_end_0, end_mask = var_852_end_mask_0, x = var_737_cast_fp16)[name = string("op_852_cast_fp16")]; + tensor var_859_begin_0 = const()[name = string("op_859_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_859_end_0 = const()[name = string("op_859_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_859_end_mask_0 = const()[name = string("op_859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_859_cast_fp16 = slice_by_index(begin = var_859_begin_0, end = var_859_end_0, end_mask = var_859_end_mask_0, x = var_737_cast_fp16)[name = string("op_859_cast_fp16")]; + tensor var_866_begin_0 = const()[name = string("op_866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_866_end_0 = const()[name = string("op_866_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_866_end_mask_0 = const()[name = string("op_866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_866_cast_fp16 = slice_by_index(begin = var_866_begin_0, end = var_866_end_0, end_mask = var_866_end_mask_0, x = var_741_cast_fp16)[name = string("op_866_cast_fp16")]; + tensor var_873_begin_0 = const()[name = string("op_873_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_873_end_0 = const()[name = string("op_873_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_873_end_mask_0 = const()[name = string("op_873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_873_cast_fp16 = slice_by_index(begin = var_873_begin_0, end = var_873_end_0, end_mask = var_873_end_mask_0, x = var_741_cast_fp16)[name = string("op_873_cast_fp16")]; + tensor var_880_begin_0 = const()[name = string("op_880_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_880_end_0 = const()[name = string("op_880_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_880_end_mask_0 = const()[name = string("op_880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_880_cast_fp16 = slice_by_index(begin = var_880_begin_0, end = var_880_end_0, end_mask = var_880_end_mask_0, x = var_741_cast_fp16)[name = string("op_880_cast_fp16")]; + tensor var_887_begin_0 = const()[name = string("op_887_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_887_end_0 = const()[name = string("op_887_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_887_end_mask_0 = const()[name = string("op_887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_887_cast_fp16 = slice_by_index(begin = var_887_begin_0, end = var_887_end_0, end_mask = var_887_end_mask_0, x = var_741_cast_fp16)[name = string("op_887_cast_fp16")]; + tensor var_894_begin_0 = const()[name = string("op_894_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_894_end_0 = const()[name = string("op_894_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_894_end_mask_0 = const()[name = string("op_894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_894_cast_fp16 = slice_by_index(begin = var_894_begin_0, end = var_894_end_0, end_mask = var_894_end_mask_0, x = var_745_cast_fp16)[name = string("op_894_cast_fp16")]; + tensor var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = var_745_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_908_begin_0 = const()[name = string("op_908_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_908_end_0 = const()[name = string("op_908_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_908_end_mask_0 = const()[name = string("op_908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_908_cast_fp16 = slice_by_index(begin = var_908_begin_0, end = var_908_end_0, end_mask = var_908_end_mask_0, x = var_745_cast_fp16)[name = string("op_908_cast_fp16")]; + tensor var_915_begin_0 = const()[name = string("op_915_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_915_end_0 = const()[name = string("op_915_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_915_end_mask_0 = const()[name = string("op_915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_915_cast_fp16 = slice_by_index(begin = var_915_begin_0, end = var_915_end_0, end_mask = var_915_end_mask_0, x = var_745_cast_fp16)[name = string("op_915_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_920_begin_0 = const()[name = string("op_920_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_920_end_0 = const()[name = string("op_920_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_920_end_mask_0 = const()[name = string("op_920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_2")]; + tensor var_920_cast_fp16 = slice_by_index(begin = var_920_begin_0, end = var_920_end_0, end_mask = var_920_end_mask_0, x = k_3_cast_fp16)[name = string("op_920_cast_fp16")]; + tensor var_924_begin_0 = const()[name = string("op_924_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_924_end_0 = const()[name = string("op_924_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_924_end_mask_0 = const()[name = string("op_924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_924_cast_fp16 = slice_by_index(begin = var_924_begin_0, end = var_924_end_0, end_mask = var_924_end_mask_0, x = k_3_cast_fp16)[name = string("op_924_cast_fp16")]; + tensor var_928_begin_0 = const()[name = string("op_928_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_928_end_0 = const()[name = string("op_928_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_928_end_mask_0 = const()[name = string("op_928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_928_cast_fp16 = slice_by_index(begin = var_928_begin_0, end = var_928_end_0, end_mask = var_928_end_mask_0, x = k_3_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_932_begin_0 = const()[name = string("op_932_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_932_end_0 = const()[name = string("op_932_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_932_end_mask_0 = const()[name = string("op_932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_932_cast_fp16 = slice_by_index(begin = var_932_begin_0, end = var_932_end_0, end_mask = var_932_end_mask_0, x = k_3_cast_fp16)[name = string("op_932_cast_fp16")]; + tensor var_936_begin_0 = const()[name = string("op_936_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_936_end_0 = const()[name = string("op_936_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_936_end_mask_0 = const()[name = string("op_936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_936_cast_fp16 = slice_by_index(begin = var_936_begin_0, end = var_936_end_0, end_mask = var_936_end_mask_0, x = k_3_cast_fp16)[name = string("op_936_cast_fp16")]; + tensor var_940_begin_0 = const()[name = string("op_940_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_940_end_0 = const()[name = string("op_940_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_940_end_mask_0 = const()[name = string("op_940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_940_cast_fp16 = slice_by_index(begin = var_940_begin_0, end = var_940_end_0, end_mask = var_940_end_mask_0, x = k_3_cast_fp16)[name = string("op_940_cast_fp16")]; + tensor var_942_begin_0 = const()[name = string("op_942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_942_end_0 = const()[name = string("op_942_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_942_end_mask_0 = const()[name = string("op_942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = value_3_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_946_begin_0 = const()[name = string("op_946_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_946_end_0 = const()[name = string("op_946_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_946_end_mask_0 = const()[name = string("op_946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_946_cast_fp16 = slice_by_index(begin = var_946_begin_0, end = var_946_end_0, end_mask = var_946_end_mask_0, x = value_3_cast_fp16)[name = string("op_946_cast_fp16")]; + tensor var_950_begin_0 = const()[name = string("op_950_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_950_end_0 = const()[name = string("op_950_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_950_end_mask_0 = const()[name = string("op_950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_950_cast_fp16 = slice_by_index(begin = var_950_begin_0, end = var_950_end_0, end_mask = var_950_end_mask_0, x = value_3_cast_fp16)[name = string("op_950_cast_fp16")]; + tensor var_954_begin_0 = const()[name = string("op_954_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_954_end_0 = const()[name = string("op_954_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_954_end_mask_0 = const()[name = string("op_954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_954_cast_fp16 = slice_by_index(begin = var_954_begin_0, end = var_954_end_0, end_mask = var_954_end_mask_0, x = value_3_cast_fp16)[name = string("op_954_cast_fp16")]; + tensor var_958_begin_0 = const()[name = string("op_958_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_958_end_0 = const()[name = string("op_958_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_958_end_mask_0 = const()[name = string("op_958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_958_cast_fp16 = slice_by_index(begin = var_958_begin_0, end = var_958_end_0, end_mask = var_958_end_mask_0, x = value_3_cast_fp16)[name = string("op_958_cast_fp16")]; + tensor var_962_begin_0 = const()[name = string("op_962_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_962_end_0 = const()[name = string("op_962_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_962_end_mask_0 = const()[name = string("op_962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, x = value_3_cast_fp16)[name = string("op_962_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_920_cast_fp16, var_754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_920_cast_fp16, var_761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_920_cast_fp16, var_768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_920_cast_fp16, var_775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_924_cast_fp16, var_782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_924_cast_fp16, var_789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_924_cast_fp16, var_796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_924_cast_fp16, var_803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_928_cast_fp16, var_810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_928_cast_fp16, var_817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_928_cast_fp16, var_824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_928_cast_fp16, var_831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_932_cast_fp16, var_838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_932_cast_fp16, var_845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_932_cast_fp16, var_852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_932_cast_fp16, var_859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_936_cast_fp16, var_866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_936_cast_fp16, var_873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_936_cast_fp16, var_880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_936_cast_fp16, var_887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_940_cast_fp16, var_894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_940_cast_fp16, var_901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_940_cast_fp16, var_908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_940_cast_fp16, var_915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + fp16 var_1013_to_fp16 = const()[name = string("op_1013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_1013_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_1015_to_fp16 = const()[name = string("op_1015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_1015_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_1017_to_fp16 = const()[name = string("op_1017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_1017_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_1019_to_fp16 = const()[name = string("op_1019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_1019_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_1021_to_fp16 = const()[name = string("op_1021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_1021_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_1023_to_fp16 = const()[name = string("op_1023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_1023_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_1025_to_fp16 = const()[name = string("op_1025_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_1025_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_1027_to_fp16 = const()[name = string("op_1027_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_1027_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + fp16 var_1029_to_fp16 = const()[name = string("op_1029_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1029_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1031_to_fp16 = const()[name = string("op_1031_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1031_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1033_to_fp16 = const()[name = string("op_1033_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1033_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1035_to_fp16 = const()[name = string("op_1035_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1035_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1037_to_fp16 = const()[name = string("op_1037_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1037_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1039_to_fp16 = const()[name = string("op_1039_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1039_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1041_to_fp16 = const()[name = string("op_1041_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1041_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1043_to_fp16 = const()[name = string("op_1043_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1043_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1045_to_fp16 = const()[name = string("op_1045_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1045_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1047_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1049_to_fp16 = const()[name = string("op_1049_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1049_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1051_to_fp16 = const()[name = string("op_1051_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1051_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1053_to_fp16 = const()[name = string("op_1053_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1053_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1055_to_fp16 = const()[name = string("op_1055_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1055_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1057_to_fp16 = const()[name = string("op_1057_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1057_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1059_to_fp16 = const()[name = string("op_1059_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1059_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + tensor var_1061_cast_fp16 = softmax(axis = var_670, x = aw_chunk_49_cast_fp16)[name = string("op_1061_cast_fp16")]; + tensor var_1062_cast_fp16 = softmax(axis = var_670, x = aw_chunk_51_cast_fp16)[name = string("op_1062_cast_fp16")]; + tensor var_1063_cast_fp16 = softmax(axis = var_670, x = aw_chunk_53_cast_fp16)[name = string("op_1063_cast_fp16")]; + tensor var_1064_cast_fp16 = softmax(axis = var_670, x = aw_chunk_55_cast_fp16)[name = string("op_1064_cast_fp16")]; + tensor var_1065_cast_fp16 = softmax(axis = var_670, x = aw_chunk_57_cast_fp16)[name = string("op_1065_cast_fp16")]; + tensor var_1066_cast_fp16 = softmax(axis = var_670, x = aw_chunk_59_cast_fp16)[name = string("op_1066_cast_fp16")]; + tensor var_1067_cast_fp16 = softmax(axis = var_670, x = aw_chunk_61_cast_fp16)[name = string("op_1067_cast_fp16")]; + tensor var_1068_cast_fp16 = softmax(axis = var_670, x = aw_chunk_63_cast_fp16)[name = string("op_1068_cast_fp16")]; + tensor var_1069_cast_fp16 = softmax(axis = var_670, x = aw_chunk_65_cast_fp16)[name = string("op_1069_cast_fp16")]; + tensor var_1070_cast_fp16 = softmax(axis = var_670, x = aw_chunk_67_cast_fp16)[name = string("op_1070_cast_fp16")]; + tensor var_1071_cast_fp16 = softmax(axis = var_670, x = aw_chunk_69_cast_fp16)[name = string("op_1071_cast_fp16")]; + tensor var_1072_cast_fp16 = softmax(axis = var_670, x = aw_chunk_71_cast_fp16)[name = string("op_1072_cast_fp16")]; + tensor var_1073_cast_fp16 = softmax(axis = var_670, x = aw_chunk_73_cast_fp16)[name = string("op_1073_cast_fp16")]; + tensor var_1074_cast_fp16 = softmax(axis = var_670, x = aw_chunk_75_cast_fp16)[name = string("op_1074_cast_fp16")]; + tensor var_1075_cast_fp16 = softmax(axis = var_670, x = aw_chunk_77_cast_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1076_cast_fp16 = softmax(axis = var_670, x = aw_chunk_79_cast_fp16)[name = string("op_1076_cast_fp16")]; + tensor var_1077_cast_fp16 = softmax(axis = var_670, x = aw_chunk_81_cast_fp16)[name = string("op_1077_cast_fp16")]; + tensor var_1078_cast_fp16 = softmax(axis = var_670, x = aw_chunk_83_cast_fp16)[name = string("op_1078_cast_fp16")]; + tensor var_1079_cast_fp16 = softmax(axis = var_670, x = aw_chunk_85_cast_fp16)[name = string("op_1079_cast_fp16")]; + tensor var_1080_cast_fp16 = softmax(axis = var_670, x = aw_chunk_87_cast_fp16)[name = string("op_1080_cast_fp16")]; + tensor var_1081_cast_fp16 = softmax(axis = var_670, x = aw_chunk_89_cast_fp16)[name = string("op_1081_cast_fp16")]; + tensor var_1082_cast_fp16 = softmax(axis = var_670, x = aw_chunk_91_cast_fp16)[name = string("op_1082_cast_fp16")]; + tensor var_1083_cast_fp16 = softmax(axis = var_670, x = aw_chunk_93_cast_fp16)[name = string("op_1083_cast_fp16")]; + tensor var_1084_cast_fp16 = softmax(axis = var_670, x = aw_chunk_95_cast_fp16)[name = string("op_1084_cast_fp16")]; + string var_1086_equation_0 = const()[name = string("op_1086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1086_cast_fp16 = einsum(equation = var_1086_equation_0, values = (var_942_cast_fp16, var_1061_cast_fp16))[name = string("op_1086_cast_fp16")]; + string var_1088_equation_0 = const()[name = string("op_1088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1088_cast_fp16 = einsum(equation = var_1088_equation_0, values = (var_942_cast_fp16, var_1062_cast_fp16))[name = string("op_1088_cast_fp16")]; + string var_1090_equation_0 = const()[name = string("op_1090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1090_cast_fp16 = einsum(equation = var_1090_equation_0, values = (var_942_cast_fp16, var_1063_cast_fp16))[name = string("op_1090_cast_fp16")]; + string var_1092_equation_0 = const()[name = string("op_1092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1092_cast_fp16 = einsum(equation = var_1092_equation_0, values = (var_942_cast_fp16, var_1064_cast_fp16))[name = string("op_1092_cast_fp16")]; + string var_1094_equation_0 = const()[name = string("op_1094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1094_cast_fp16 = einsum(equation = var_1094_equation_0, values = (var_946_cast_fp16, var_1065_cast_fp16))[name = string("op_1094_cast_fp16")]; + string var_1096_equation_0 = const()[name = string("op_1096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1096_cast_fp16 = einsum(equation = var_1096_equation_0, values = (var_946_cast_fp16, var_1066_cast_fp16))[name = string("op_1096_cast_fp16")]; + string var_1098_equation_0 = const()[name = string("op_1098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1098_cast_fp16 = einsum(equation = var_1098_equation_0, values = (var_946_cast_fp16, var_1067_cast_fp16))[name = string("op_1098_cast_fp16")]; + string var_1100_equation_0 = const()[name = string("op_1100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1100_cast_fp16 = einsum(equation = var_1100_equation_0, values = (var_946_cast_fp16, var_1068_cast_fp16))[name = string("op_1100_cast_fp16")]; + string var_1102_equation_0 = const()[name = string("op_1102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1102_cast_fp16 = einsum(equation = var_1102_equation_0, values = (var_950_cast_fp16, var_1069_cast_fp16))[name = string("op_1102_cast_fp16")]; + string var_1104_equation_0 = const()[name = string("op_1104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1104_cast_fp16 = einsum(equation = var_1104_equation_0, values = (var_950_cast_fp16, var_1070_cast_fp16))[name = string("op_1104_cast_fp16")]; + string var_1106_equation_0 = const()[name = string("op_1106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1106_cast_fp16 = einsum(equation = var_1106_equation_0, values = (var_950_cast_fp16, var_1071_cast_fp16))[name = string("op_1106_cast_fp16")]; + string var_1108_equation_0 = const()[name = string("op_1108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1108_cast_fp16 = einsum(equation = var_1108_equation_0, values = (var_950_cast_fp16, var_1072_cast_fp16))[name = string("op_1108_cast_fp16")]; + string var_1110_equation_0 = const()[name = string("op_1110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1110_cast_fp16 = einsum(equation = var_1110_equation_0, values = (var_954_cast_fp16, var_1073_cast_fp16))[name = string("op_1110_cast_fp16")]; + string var_1112_equation_0 = const()[name = string("op_1112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1112_cast_fp16 = einsum(equation = var_1112_equation_0, values = (var_954_cast_fp16, var_1074_cast_fp16))[name = string("op_1112_cast_fp16")]; + string var_1114_equation_0 = const()[name = string("op_1114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1114_cast_fp16 = einsum(equation = var_1114_equation_0, values = (var_954_cast_fp16, var_1075_cast_fp16))[name = string("op_1114_cast_fp16")]; + string var_1116_equation_0 = const()[name = string("op_1116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1116_cast_fp16 = einsum(equation = var_1116_equation_0, values = (var_954_cast_fp16, var_1076_cast_fp16))[name = string("op_1116_cast_fp16")]; + string var_1118_equation_0 = const()[name = string("op_1118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1118_cast_fp16 = einsum(equation = var_1118_equation_0, values = (var_958_cast_fp16, var_1077_cast_fp16))[name = string("op_1118_cast_fp16")]; + string var_1120_equation_0 = const()[name = string("op_1120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1120_cast_fp16 = einsum(equation = var_1120_equation_0, values = (var_958_cast_fp16, var_1078_cast_fp16))[name = string("op_1120_cast_fp16")]; + string var_1122_equation_0 = const()[name = string("op_1122_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1122_cast_fp16 = einsum(equation = var_1122_equation_0, values = (var_958_cast_fp16, var_1079_cast_fp16))[name = string("op_1122_cast_fp16")]; + string var_1124_equation_0 = const()[name = string("op_1124_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1124_cast_fp16 = einsum(equation = var_1124_equation_0, values = (var_958_cast_fp16, var_1080_cast_fp16))[name = string("op_1124_cast_fp16")]; + string var_1126_equation_0 = const()[name = string("op_1126_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1126_cast_fp16 = einsum(equation = var_1126_equation_0, values = (var_962_cast_fp16, var_1081_cast_fp16))[name = string("op_1126_cast_fp16")]; + string var_1128_equation_0 = const()[name = string("op_1128_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1128_cast_fp16 = einsum(equation = var_1128_equation_0, values = (var_962_cast_fp16, var_1082_cast_fp16))[name = string("op_1128_cast_fp16")]; + string var_1130_equation_0 = const()[name = string("op_1130_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1130_cast_fp16 = einsum(equation = var_1130_equation_0, values = (var_962_cast_fp16, var_1083_cast_fp16))[name = string("op_1130_cast_fp16")]; + string var_1132_equation_0 = const()[name = string("op_1132_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1132_cast_fp16 = einsum(equation = var_1132_equation_0, values = (var_962_cast_fp16, var_1084_cast_fp16))[name = string("op_1132_cast_fp16")]; + bool var_1134_interleave_0 = const()[name = string("op_1134_interleave_0"), val = bool(false)]; + tensor var_1134_cast_fp16 = concat(axis = var_659, interleave = var_1134_interleave_0, values = (var_1086_cast_fp16, var_1088_cast_fp16, var_1090_cast_fp16, var_1092_cast_fp16))[name = string("op_1134_cast_fp16")]; + bool var_1136_interleave_0 = const()[name = string("op_1136_interleave_0"), val = bool(false)]; + tensor var_1136_cast_fp16 = concat(axis = var_659, interleave = var_1136_interleave_0, values = (var_1094_cast_fp16, var_1096_cast_fp16, var_1098_cast_fp16, var_1100_cast_fp16))[name = string("op_1136_cast_fp16")]; + bool var_1138_interleave_0 = const()[name = string("op_1138_interleave_0"), val = bool(false)]; + tensor var_1138_cast_fp16 = concat(axis = var_659, interleave = var_1138_interleave_0, values = (var_1102_cast_fp16, var_1104_cast_fp16, var_1106_cast_fp16, var_1108_cast_fp16))[name = string("op_1138_cast_fp16")]; + bool var_1140_interleave_0 = const()[name = string("op_1140_interleave_0"), val = bool(false)]; + tensor var_1140_cast_fp16 = concat(axis = var_659, interleave = var_1140_interleave_0, values = (var_1110_cast_fp16, var_1112_cast_fp16, var_1114_cast_fp16, var_1116_cast_fp16))[name = string("op_1140_cast_fp16")]; + bool var_1142_interleave_0 = const()[name = string("op_1142_interleave_0"), val = bool(false)]; + tensor var_1142_cast_fp16 = concat(axis = var_659, interleave = var_1142_interleave_0, values = (var_1118_cast_fp16, var_1120_cast_fp16, var_1122_cast_fp16, var_1124_cast_fp16))[name = string("op_1142_cast_fp16")]; + bool var_1144_interleave_0 = const()[name = string("op_1144_interleave_0"), val = bool(false)]; + tensor var_1144_cast_fp16 = concat(axis = var_659, interleave = var_1144_interleave_0, values = (var_1126_cast_fp16, var_1128_cast_fp16, var_1130_cast_fp16, var_1132_cast_fp16))[name = string("op_1144_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_670, interleave = input_9_interleave_0, values = (var_1134_cast_fp16, var_1136_cast_fp16, var_1138_cast_fp16, var_1140_cast_fp16, var_1142_cast_fp16, var_1144_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6662016)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956992)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_1163_to_fp16 = const()[name = string("op_1163_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_1163_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6957824)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6958656)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6959488)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8139200)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8142336)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322048)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_1192 = const()[name = string("op_1192"), val = int32(3)]; + int32 var_1203 = const()[name = string("op_1203"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_1220_to_fp16 = const()[name = string("op_1220_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_1220_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322880)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9323712)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9324544)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9619520)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9620352)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9915328)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10210304)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_1258_begin_0 = const()[name = string("op_1258_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1258_end_0 = const()[name = string("op_1258_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1258_end_mask_0 = const()[name = string("op_1258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = query_5_cast_fp16)[name = string("op_1258_cast_fp16")]; + tensor var_1262_begin_0 = const()[name = string("op_1262_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1262_end_0 = const()[name = string("op_1262_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1262_end_mask_0 = const()[name = string("op_1262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1262_cast_fp16 = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = query_5_cast_fp16)[name = string("op_1262_cast_fp16")]; + tensor var_1266_begin_0 = const()[name = string("op_1266_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1266_end_0 = const()[name = string("op_1266_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1266_end_mask_0 = const()[name = string("op_1266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1266_cast_fp16 = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = query_5_cast_fp16)[name = string("op_1266_cast_fp16")]; + tensor var_1270_begin_0 = const()[name = string("op_1270_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1270_end_0 = const()[name = string("op_1270_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1270_end_mask_0 = const()[name = string("op_1270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = query_5_cast_fp16)[name = string("op_1270_cast_fp16")]; + tensor var_1274_begin_0 = const()[name = string("op_1274_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1274_end_0 = const()[name = string("op_1274_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1274_end_mask_0 = const()[name = string("op_1274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1274_cast_fp16 = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = query_5_cast_fp16)[name = string("op_1274_cast_fp16")]; + tensor var_1278_begin_0 = const()[name = string("op_1278_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1278_end_0 = const()[name = string("op_1278_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1278_end_mask_0 = const()[name = string("op_1278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1278_cast_fp16 = slice_by_index(begin = var_1278_begin_0, end = var_1278_end_0, end_mask = var_1278_end_mask_0, x = query_5_cast_fp16)[name = string("op_1278_cast_fp16")]; + tensor var_1287_begin_0 = const()[name = string("op_1287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1287_end_0 = const()[name = string("op_1287_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1287_end_mask_0 = const()[name = string("op_1287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1287_cast_fp16 = slice_by_index(begin = var_1287_begin_0, end = var_1287_end_0, end_mask = var_1287_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1287_cast_fp16")]; + tensor var_1294_begin_0 = const()[name = string("op_1294_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1294_end_0 = const()[name = string("op_1294_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1294_end_mask_0 = const()[name = string("op_1294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1294_cast_fp16 = slice_by_index(begin = var_1294_begin_0, end = var_1294_end_0, end_mask = var_1294_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1294_cast_fp16")]; + tensor var_1301_begin_0 = const()[name = string("op_1301_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1301_end_0 = const()[name = string("op_1301_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1301_end_mask_0 = const()[name = string("op_1301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1301_cast_fp16 = slice_by_index(begin = var_1301_begin_0, end = var_1301_end_0, end_mask = var_1301_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1301_cast_fp16")]; + tensor var_1308_begin_0 = const()[name = string("op_1308_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1308_end_0 = const()[name = string("op_1308_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1308_end_mask_0 = const()[name = string("op_1308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1308_cast_fp16 = slice_by_index(begin = var_1308_begin_0, end = var_1308_end_0, end_mask = var_1308_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1308_cast_fp16")]; + tensor var_1315_begin_0 = const()[name = string("op_1315_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1315_end_0 = const()[name = string("op_1315_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1315_end_mask_0 = const()[name = string("op_1315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1315_cast_fp16 = slice_by_index(begin = var_1315_begin_0, end = var_1315_end_0, end_mask = var_1315_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1315_cast_fp16")]; + tensor var_1322_begin_0 = const()[name = string("op_1322_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1322_end_0 = const()[name = string("op_1322_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1322_end_mask_0 = const()[name = string("op_1322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1322_cast_fp16 = slice_by_index(begin = var_1322_begin_0, end = var_1322_end_0, end_mask = var_1322_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1322_cast_fp16")]; + tensor var_1329_begin_0 = const()[name = string("op_1329_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1329_end_0 = const()[name = string("op_1329_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1329_end_mask_0 = const()[name = string("op_1329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1329_cast_fp16 = slice_by_index(begin = var_1329_begin_0, end = var_1329_end_0, end_mask = var_1329_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1329_cast_fp16")]; + tensor var_1336_begin_0 = const()[name = string("op_1336_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1336_end_0 = const()[name = string("op_1336_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1336_end_mask_0 = const()[name = string("op_1336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1336_cast_fp16 = slice_by_index(begin = var_1336_begin_0, end = var_1336_end_0, end_mask = var_1336_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor var_1343_begin_0 = const()[name = string("op_1343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1343_end_0 = const()[name = string("op_1343_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1343_end_mask_0 = const()[name = string("op_1343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1343_cast_fp16 = slice_by_index(begin = var_1343_begin_0, end = var_1343_end_0, end_mask = var_1343_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1343_cast_fp16")]; + tensor var_1350_begin_0 = const()[name = string("op_1350_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1350_end_0 = const()[name = string("op_1350_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1350_end_mask_0 = const()[name = string("op_1350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1350_cast_fp16 = slice_by_index(begin = var_1350_begin_0, end = var_1350_end_0, end_mask = var_1350_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1350_cast_fp16")]; + tensor var_1357_begin_0 = const()[name = string("op_1357_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1357_end_0 = const()[name = string("op_1357_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1357_end_mask_0 = const()[name = string("op_1357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1357_cast_fp16 = slice_by_index(begin = var_1357_begin_0, end = var_1357_end_0, end_mask = var_1357_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1357_cast_fp16")]; + tensor var_1364_begin_0 = const()[name = string("op_1364_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1364_end_0 = const()[name = string("op_1364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1364_end_mask_0 = const()[name = string("op_1364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1364_cast_fp16 = slice_by_index(begin = var_1364_begin_0, end = var_1364_end_0, end_mask = var_1364_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1364_cast_fp16")]; + tensor var_1371_begin_0 = const()[name = string("op_1371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1371_end_0 = const()[name = string("op_1371_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1371_end_mask_0 = const()[name = string("op_1371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1371_cast_fp16 = slice_by_index(begin = var_1371_begin_0, end = var_1371_end_0, end_mask = var_1371_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1371_cast_fp16")]; + tensor var_1378_begin_0 = const()[name = string("op_1378_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1378_end_0 = const()[name = string("op_1378_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1378_end_mask_0 = const()[name = string("op_1378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1378_cast_fp16 = slice_by_index(begin = var_1378_begin_0, end = var_1378_end_0, end_mask = var_1378_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1378_cast_fp16")]; + tensor var_1385_begin_0 = const()[name = string("op_1385_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1385_end_0 = const()[name = string("op_1385_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1385_end_mask_0 = const()[name = string("op_1385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1385_cast_fp16 = slice_by_index(begin = var_1385_begin_0, end = var_1385_end_0, end_mask = var_1385_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1385_cast_fp16")]; + tensor var_1392_begin_0 = const()[name = string("op_1392_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1392_end_0 = const()[name = string("op_1392_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1392_end_mask_0 = const()[name = string("op_1392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1392_cast_fp16 = slice_by_index(begin = var_1392_begin_0, end = var_1392_end_0, end_mask = var_1392_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1392_cast_fp16")]; + tensor var_1399_begin_0 = const()[name = string("op_1399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1399_end_0 = const()[name = string("op_1399_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1399_end_mask_0 = const()[name = string("op_1399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1399_cast_fp16 = slice_by_index(begin = var_1399_begin_0, end = var_1399_end_0, end_mask = var_1399_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1399_cast_fp16")]; + tensor var_1406_begin_0 = const()[name = string("op_1406_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1406_end_0 = const()[name = string("op_1406_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1406_end_mask_0 = const()[name = string("op_1406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1406_cast_fp16 = slice_by_index(begin = var_1406_begin_0, end = var_1406_end_0, end_mask = var_1406_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1406_cast_fp16")]; + tensor var_1413_begin_0 = const()[name = string("op_1413_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1413_end_0 = const()[name = string("op_1413_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1413_end_mask_0 = const()[name = string("op_1413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1413_cast_fp16 = slice_by_index(begin = var_1413_begin_0, end = var_1413_end_0, end_mask = var_1413_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1413_cast_fp16")]; + tensor var_1420_begin_0 = const()[name = string("op_1420_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1420_end_0 = const()[name = string("op_1420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1420_end_mask_0 = const()[name = string("op_1420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1420_cast_fp16 = slice_by_index(begin = var_1420_begin_0, end = var_1420_end_0, end_mask = var_1420_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1420_cast_fp16")]; + tensor var_1427_begin_0 = const()[name = string("op_1427_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1427_end_0 = const()[name = string("op_1427_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1427_end_mask_0 = const()[name = string("op_1427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1427_cast_fp16 = slice_by_index(begin = var_1427_begin_0, end = var_1427_end_0, end_mask = var_1427_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1427_cast_fp16")]; + tensor var_1434_begin_0 = const()[name = string("op_1434_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1434_end_0 = const()[name = string("op_1434_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1434_end_mask_0 = const()[name = string("op_1434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1434_cast_fp16 = slice_by_index(begin = var_1434_begin_0, end = var_1434_end_0, end_mask = var_1434_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1434_cast_fp16")]; + tensor var_1441_begin_0 = const()[name = string("op_1441_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1441_end_0 = const()[name = string("op_1441_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1441_end_mask_0 = const()[name = string("op_1441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1441_cast_fp16 = slice_by_index(begin = var_1441_begin_0, end = var_1441_end_0, end_mask = var_1441_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1441_cast_fp16")]; + tensor var_1448_begin_0 = const()[name = string("op_1448_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1448_end_0 = const()[name = string("op_1448_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1448_end_mask_0 = const()[name = string("op_1448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1448_cast_fp16 = slice_by_index(begin = var_1448_begin_0, end = var_1448_end_0, end_mask = var_1448_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1448_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1453_begin_0 = const()[name = string("op_1453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1453_end_0 = const()[name = string("op_1453_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1453_end_mask_0 = const()[name = string("op_1453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_1")]; + tensor var_1453_cast_fp16 = slice_by_index(begin = var_1453_begin_0, end = var_1453_end_0, end_mask = var_1453_end_mask_0, x = k_5_cast_fp16)[name = string("op_1453_cast_fp16")]; + tensor var_1457_begin_0 = const()[name = string("op_1457_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1457_end_0 = const()[name = string("op_1457_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1457_end_mask_0 = const()[name = string("op_1457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1457_cast_fp16 = slice_by_index(begin = var_1457_begin_0, end = var_1457_end_0, end_mask = var_1457_end_mask_0, x = k_5_cast_fp16)[name = string("op_1457_cast_fp16")]; + tensor var_1461_begin_0 = const()[name = string("op_1461_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1461_end_0 = const()[name = string("op_1461_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1461_end_mask_0 = const()[name = string("op_1461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1461_cast_fp16 = slice_by_index(begin = var_1461_begin_0, end = var_1461_end_0, end_mask = var_1461_end_mask_0, x = k_5_cast_fp16)[name = string("op_1461_cast_fp16")]; + tensor var_1465_begin_0 = const()[name = string("op_1465_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1465_end_0 = const()[name = string("op_1465_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1465_end_mask_0 = const()[name = string("op_1465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1465_cast_fp16 = slice_by_index(begin = var_1465_begin_0, end = var_1465_end_0, end_mask = var_1465_end_mask_0, x = k_5_cast_fp16)[name = string("op_1465_cast_fp16")]; + tensor var_1469_begin_0 = const()[name = string("op_1469_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1469_end_0 = const()[name = string("op_1469_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1469_end_mask_0 = const()[name = string("op_1469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1469_cast_fp16 = slice_by_index(begin = var_1469_begin_0, end = var_1469_end_0, end_mask = var_1469_end_mask_0, x = k_5_cast_fp16)[name = string("op_1469_cast_fp16")]; + tensor var_1473_begin_0 = const()[name = string("op_1473_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1473_end_0 = const()[name = string("op_1473_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1473_end_mask_0 = const()[name = string("op_1473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = k_5_cast_fp16)[name = string("op_1473_cast_fp16")]; + tensor var_1475_begin_0 = const()[name = string("op_1475_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1475_end_0 = const()[name = string("op_1475_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1475_end_mask_0 = const()[name = string("op_1475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1475_cast_fp16 = slice_by_index(begin = var_1475_begin_0, end = var_1475_end_0, end_mask = var_1475_end_mask_0, x = value_5_cast_fp16)[name = string("op_1475_cast_fp16")]; + tensor var_1479_begin_0 = const()[name = string("op_1479_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1479_end_0 = const()[name = string("op_1479_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1479_end_mask_0 = const()[name = string("op_1479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1479_cast_fp16 = slice_by_index(begin = var_1479_begin_0, end = var_1479_end_0, end_mask = var_1479_end_mask_0, x = value_5_cast_fp16)[name = string("op_1479_cast_fp16")]; + tensor var_1483_begin_0 = const()[name = string("op_1483_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1483_end_0 = const()[name = string("op_1483_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1483_end_mask_0 = const()[name = string("op_1483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1483_cast_fp16 = slice_by_index(begin = var_1483_begin_0, end = var_1483_end_0, end_mask = var_1483_end_mask_0, x = value_5_cast_fp16)[name = string("op_1483_cast_fp16")]; + tensor var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = value_5_cast_fp16)[name = string("op_1487_cast_fp16")]; + tensor var_1491_begin_0 = const()[name = string("op_1491_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1491_end_0 = const()[name = string("op_1491_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1491_end_mask_0 = const()[name = string("op_1491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1491_cast_fp16 = slice_by_index(begin = var_1491_begin_0, end = var_1491_end_0, end_mask = var_1491_end_mask_0, x = value_5_cast_fp16)[name = string("op_1491_cast_fp16")]; + tensor var_1495_begin_0 = const()[name = string("op_1495_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1495_end_0 = const()[name = string("op_1495_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1495_end_mask_0 = const()[name = string("op_1495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1495_cast_fp16 = slice_by_index(begin = var_1495_begin_0, end = var_1495_end_0, end_mask = var_1495_end_mask_0, x = value_5_cast_fp16)[name = string("op_1495_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1453_cast_fp16, var_1287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1453_cast_fp16, var_1294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1453_cast_fp16, var_1301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1453_cast_fp16, var_1308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1457_cast_fp16, var_1315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1457_cast_fp16, var_1322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1457_cast_fp16, var_1329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1457_cast_fp16, var_1336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1461_cast_fp16, var_1343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1461_cast_fp16, var_1350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1461_cast_fp16, var_1357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1461_cast_fp16, var_1364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1465_cast_fp16, var_1371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1465_cast_fp16, var_1378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1465_cast_fp16, var_1385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1465_cast_fp16, var_1392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1469_cast_fp16, var_1399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1469_cast_fp16, var_1406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1469_cast_fp16, var_1413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1469_cast_fp16, var_1420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1473_cast_fp16, var_1427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1473_cast_fp16, var_1434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1473_cast_fp16, var_1441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1473_cast_fp16, var_1448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + fp16 var_1546_to_fp16 = const()[name = string("op_1546_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1546_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1548_to_fp16 = const()[name = string("op_1548_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1548_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1550_to_fp16 = const()[name = string("op_1550_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1550_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1552_to_fp16 = const()[name = string("op_1552_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1552_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1554_to_fp16 = const()[name = string("op_1554_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1554_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1556_to_fp16 = const()[name = string("op_1556_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1556_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1558_to_fp16 = const()[name = string("op_1558_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1558_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1560_to_fp16 = const()[name = string("op_1560_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1560_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1562_to_fp16 = const()[name = string("op_1562_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1562_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1564_to_fp16 = const()[name = string("op_1564_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1564_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1566_to_fp16 = const()[name = string("op_1566_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1566_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1568_to_fp16 = const()[name = string("op_1568_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1568_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1570_to_fp16 = const()[name = string("op_1570_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1570_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1572_to_fp16 = const()[name = string("op_1572_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1572_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1574_to_fp16 = const()[name = string("op_1574_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1574_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1576_to_fp16 = const()[name = string("op_1576_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1576_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + fp16 var_1578_to_fp16 = const()[name = string("op_1578_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1578_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1580_to_fp16 = const()[name = string("op_1580_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1580_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1582_to_fp16 = const()[name = string("op_1582_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1582_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1584_to_fp16 = const()[name = string("op_1584_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1584_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1586_to_fp16 = const()[name = string("op_1586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1586_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1588_to_fp16 = const()[name = string("op_1588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1588_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1590_to_fp16 = const()[name = string("op_1590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1590_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1592_to_fp16 = const()[name = string("op_1592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1592_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + tensor var_1594_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_97_cast_fp16)[name = string("op_1594_cast_fp16")]; + tensor var_1595_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_99_cast_fp16)[name = string("op_1595_cast_fp16")]; + tensor var_1596_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_101_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1597_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_103_cast_fp16)[name = string("op_1597_cast_fp16")]; + tensor var_1598_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_105_cast_fp16)[name = string("op_1598_cast_fp16")]; + tensor var_1599_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_107_cast_fp16)[name = string("op_1599_cast_fp16")]; + tensor var_1600_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_109_cast_fp16)[name = string("op_1600_cast_fp16")]; + tensor var_1601_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_111_cast_fp16)[name = string("op_1601_cast_fp16")]; + tensor var_1602_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_113_cast_fp16)[name = string("op_1602_cast_fp16")]; + tensor var_1603_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_115_cast_fp16)[name = string("op_1603_cast_fp16")]; + tensor var_1604_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_117_cast_fp16)[name = string("op_1604_cast_fp16")]; + tensor var_1605_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_119_cast_fp16)[name = string("op_1605_cast_fp16")]; + tensor var_1606_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_121_cast_fp16)[name = string("op_1606_cast_fp16")]; + tensor var_1607_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_123_cast_fp16)[name = string("op_1607_cast_fp16")]; + tensor var_1608_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_125_cast_fp16)[name = string("op_1608_cast_fp16")]; + tensor var_1609_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_127_cast_fp16)[name = string("op_1609_cast_fp16")]; + tensor var_1610_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_129_cast_fp16)[name = string("op_1610_cast_fp16")]; + tensor var_1611_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_131_cast_fp16)[name = string("op_1611_cast_fp16")]; + tensor var_1612_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_133_cast_fp16)[name = string("op_1612_cast_fp16")]; + tensor var_1613_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_135_cast_fp16)[name = string("op_1613_cast_fp16")]; + tensor var_1614_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_137_cast_fp16)[name = string("op_1614_cast_fp16")]; + tensor var_1615_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_139_cast_fp16)[name = string("op_1615_cast_fp16")]; + tensor var_1616_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_141_cast_fp16)[name = string("op_1616_cast_fp16")]; + tensor var_1617_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_143_cast_fp16)[name = string("op_1617_cast_fp16")]; + string var_1619_equation_0 = const()[name = string("op_1619_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1619_cast_fp16 = einsum(equation = var_1619_equation_0, values = (var_1475_cast_fp16, var_1594_cast_fp16))[name = string("op_1619_cast_fp16")]; + string var_1621_equation_0 = const()[name = string("op_1621_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1621_cast_fp16 = einsum(equation = var_1621_equation_0, values = (var_1475_cast_fp16, var_1595_cast_fp16))[name = string("op_1621_cast_fp16")]; + string var_1623_equation_0 = const()[name = string("op_1623_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1623_cast_fp16 = einsum(equation = var_1623_equation_0, values = (var_1475_cast_fp16, var_1596_cast_fp16))[name = string("op_1623_cast_fp16")]; + string var_1625_equation_0 = const()[name = string("op_1625_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1625_cast_fp16 = einsum(equation = var_1625_equation_0, values = (var_1475_cast_fp16, var_1597_cast_fp16))[name = string("op_1625_cast_fp16")]; + string var_1627_equation_0 = const()[name = string("op_1627_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1627_cast_fp16 = einsum(equation = var_1627_equation_0, values = (var_1479_cast_fp16, var_1598_cast_fp16))[name = string("op_1627_cast_fp16")]; + string var_1629_equation_0 = const()[name = string("op_1629_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1629_cast_fp16 = einsum(equation = var_1629_equation_0, values = (var_1479_cast_fp16, var_1599_cast_fp16))[name = string("op_1629_cast_fp16")]; + string var_1631_equation_0 = const()[name = string("op_1631_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1631_cast_fp16 = einsum(equation = var_1631_equation_0, values = (var_1479_cast_fp16, var_1600_cast_fp16))[name = string("op_1631_cast_fp16")]; + string var_1633_equation_0 = const()[name = string("op_1633_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1633_cast_fp16 = einsum(equation = var_1633_equation_0, values = (var_1479_cast_fp16, var_1601_cast_fp16))[name = string("op_1633_cast_fp16")]; + string var_1635_equation_0 = const()[name = string("op_1635_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1635_cast_fp16 = einsum(equation = var_1635_equation_0, values = (var_1483_cast_fp16, var_1602_cast_fp16))[name = string("op_1635_cast_fp16")]; + string var_1637_equation_0 = const()[name = string("op_1637_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1637_cast_fp16 = einsum(equation = var_1637_equation_0, values = (var_1483_cast_fp16, var_1603_cast_fp16))[name = string("op_1637_cast_fp16")]; + string var_1639_equation_0 = const()[name = string("op_1639_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1639_cast_fp16 = einsum(equation = var_1639_equation_0, values = (var_1483_cast_fp16, var_1604_cast_fp16))[name = string("op_1639_cast_fp16")]; + string var_1641_equation_0 = const()[name = string("op_1641_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1641_cast_fp16 = einsum(equation = var_1641_equation_0, values = (var_1483_cast_fp16, var_1605_cast_fp16))[name = string("op_1641_cast_fp16")]; + string var_1643_equation_0 = const()[name = string("op_1643_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1643_cast_fp16 = einsum(equation = var_1643_equation_0, values = (var_1487_cast_fp16, var_1606_cast_fp16))[name = string("op_1643_cast_fp16")]; + string var_1645_equation_0 = const()[name = string("op_1645_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1645_cast_fp16 = einsum(equation = var_1645_equation_0, values = (var_1487_cast_fp16, var_1607_cast_fp16))[name = string("op_1645_cast_fp16")]; + string var_1647_equation_0 = const()[name = string("op_1647_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1647_cast_fp16 = einsum(equation = var_1647_equation_0, values = (var_1487_cast_fp16, var_1608_cast_fp16))[name = string("op_1647_cast_fp16")]; + string var_1649_equation_0 = const()[name = string("op_1649_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1649_cast_fp16 = einsum(equation = var_1649_equation_0, values = (var_1487_cast_fp16, var_1609_cast_fp16))[name = string("op_1649_cast_fp16")]; + string var_1651_equation_0 = const()[name = string("op_1651_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1651_cast_fp16 = einsum(equation = var_1651_equation_0, values = (var_1491_cast_fp16, var_1610_cast_fp16))[name = string("op_1651_cast_fp16")]; + string var_1653_equation_0 = const()[name = string("op_1653_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1653_cast_fp16 = einsum(equation = var_1653_equation_0, values = (var_1491_cast_fp16, var_1611_cast_fp16))[name = string("op_1653_cast_fp16")]; + string var_1655_equation_0 = const()[name = string("op_1655_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1655_cast_fp16 = einsum(equation = var_1655_equation_0, values = (var_1491_cast_fp16, var_1612_cast_fp16))[name = string("op_1655_cast_fp16")]; + string var_1657_equation_0 = const()[name = string("op_1657_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1657_cast_fp16 = einsum(equation = var_1657_equation_0, values = (var_1491_cast_fp16, var_1613_cast_fp16))[name = string("op_1657_cast_fp16")]; + string var_1659_equation_0 = const()[name = string("op_1659_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1659_cast_fp16 = einsum(equation = var_1659_equation_0, values = (var_1495_cast_fp16, var_1614_cast_fp16))[name = string("op_1659_cast_fp16")]; + string var_1661_equation_0 = const()[name = string("op_1661_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1661_cast_fp16 = einsum(equation = var_1661_equation_0, values = (var_1495_cast_fp16, var_1615_cast_fp16))[name = string("op_1661_cast_fp16")]; + string var_1663_equation_0 = const()[name = string("op_1663_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1663_cast_fp16 = einsum(equation = var_1663_equation_0, values = (var_1495_cast_fp16, var_1616_cast_fp16))[name = string("op_1663_cast_fp16")]; + string var_1665_equation_0 = const()[name = string("op_1665_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1665_cast_fp16 = einsum(equation = var_1665_equation_0, values = (var_1495_cast_fp16, var_1617_cast_fp16))[name = string("op_1665_cast_fp16")]; + bool var_1667_interleave_0 = const()[name = string("op_1667_interleave_0"), val = bool(false)]; + tensor var_1667_cast_fp16 = concat(axis = var_1192, interleave = var_1667_interleave_0, values = (var_1619_cast_fp16, var_1621_cast_fp16, var_1623_cast_fp16, var_1625_cast_fp16))[name = string("op_1667_cast_fp16")]; + bool var_1669_interleave_0 = const()[name = string("op_1669_interleave_0"), val = bool(false)]; + tensor var_1669_cast_fp16 = concat(axis = var_1192, interleave = var_1669_interleave_0, values = (var_1627_cast_fp16, var_1629_cast_fp16, var_1631_cast_fp16, var_1633_cast_fp16))[name = string("op_1669_cast_fp16")]; + bool var_1671_interleave_0 = const()[name = string("op_1671_interleave_0"), val = bool(false)]; + tensor var_1671_cast_fp16 = concat(axis = var_1192, interleave = var_1671_interleave_0, values = (var_1635_cast_fp16, var_1637_cast_fp16, var_1639_cast_fp16, var_1641_cast_fp16))[name = string("op_1671_cast_fp16")]; + bool var_1673_interleave_0 = const()[name = string("op_1673_interleave_0"), val = bool(false)]; + tensor var_1673_cast_fp16 = concat(axis = var_1192, interleave = var_1673_interleave_0, values = (var_1643_cast_fp16, var_1645_cast_fp16, var_1647_cast_fp16, var_1649_cast_fp16))[name = string("op_1673_cast_fp16")]; + bool var_1675_interleave_0 = const()[name = string("op_1675_interleave_0"), val = bool(false)]; + tensor var_1675_cast_fp16 = concat(axis = var_1192, interleave = var_1675_interleave_0, values = (var_1651_cast_fp16, var_1653_cast_fp16, var_1655_cast_fp16, var_1657_cast_fp16))[name = string("op_1675_cast_fp16")]; + bool var_1677_interleave_0 = const()[name = string("op_1677_interleave_0"), val = bool(false)]; + tensor var_1677_cast_fp16 = concat(axis = var_1192, interleave = var_1677_interleave_0, values = (var_1659_cast_fp16, var_1661_cast_fp16, var_1663_cast_fp16, var_1665_cast_fp16))[name = string("op_1677_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_1203, interleave = input_17_interleave_0, values = (var_1667_cast_fp16, var_1669_cast_fp16, var_1671_cast_fp16, var_1673_cast_fp16, var_1675_cast_fp16, var_1677_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10211136)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506112)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_1696_to_fp16 = const()[name = string("op_1696_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_1696_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506944)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10507776)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10508608)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11688320)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11691456)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12871168)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_1725 = const()[name = string("op_1725"), val = int32(3)]; + int32 var_1736 = const()[name = string("op_1736"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_1753_to_fp16 = const()[name = string("op_1753_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_1753_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872000)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872832)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12873664)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13168640)))]; + tensor query_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13169472)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13464448)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13759424)))]; + tensor value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_1791_begin_0 = const()[name = string("op_1791_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1791_end_0 = const()[name = string("op_1791_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1791_end_mask_0 = const()[name = string("op_1791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1791_cast_fp16 = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, x = query_cast_fp16)[name = string("op_1791_cast_fp16")]; + tensor var_1795_begin_0 = const()[name = string("op_1795_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1795_end_0 = const()[name = string("op_1795_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1795_end_mask_0 = const()[name = string("op_1795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1795_cast_fp16 = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = query_cast_fp16)[name = string("op_1795_cast_fp16")]; + tensor var_1799_begin_0 = const()[name = string("op_1799_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1799_end_0 = const()[name = string("op_1799_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1799_end_mask_0 = const()[name = string("op_1799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = query_cast_fp16)[name = string("op_1799_cast_fp16")]; + tensor var_1803_begin_0 = const()[name = string("op_1803_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1803_end_0 = const()[name = string("op_1803_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1803_end_mask_0 = const()[name = string("op_1803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1803_cast_fp16 = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, x = query_cast_fp16)[name = string("op_1803_cast_fp16")]; + tensor var_1807_begin_0 = const()[name = string("op_1807_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1807_end_0 = const()[name = string("op_1807_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1807_end_mask_0 = const()[name = string("op_1807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1807_cast_fp16 = slice_by_index(begin = var_1807_begin_0, end = var_1807_end_0, end_mask = var_1807_end_mask_0, x = query_cast_fp16)[name = string("op_1807_cast_fp16")]; + tensor var_1811_begin_0 = const()[name = string("op_1811_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1811_end_0 = const()[name = string("op_1811_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1811_end_mask_0 = const()[name = string("op_1811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1811_cast_fp16 = slice_by_index(begin = var_1811_begin_0, end = var_1811_end_0, end_mask = var_1811_end_mask_0, x = query_cast_fp16)[name = string("op_1811_cast_fp16")]; + tensor var_1820_begin_0 = const()[name = string("op_1820_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1820_end_0 = const()[name = string("op_1820_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1820_end_mask_0 = const()[name = string("op_1820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1820_cast_fp16 = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1820_cast_fp16")]; + tensor var_1827_begin_0 = const()[name = string("op_1827_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1827_end_0 = const()[name = string("op_1827_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1827_end_mask_0 = const()[name = string("op_1827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1827_cast_fp16 = slice_by_index(begin = var_1827_begin_0, end = var_1827_end_0, end_mask = var_1827_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1827_cast_fp16")]; + tensor var_1834_begin_0 = const()[name = string("op_1834_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1834_end_0 = const()[name = string("op_1834_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1834_end_mask_0 = const()[name = string("op_1834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1834_cast_fp16 = slice_by_index(begin = var_1834_begin_0, end = var_1834_end_0, end_mask = var_1834_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1834_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1848_begin_0 = const()[name = string("op_1848_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1848_end_0 = const()[name = string("op_1848_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1848_end_mask_0 = const()[name = string("op_1848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1848_cast_fp16 = slice_by_index(begin = var_1848_begin_0, end = var_1848_end_0, end_mask = var_1848_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1848_cast_fp16")]; + tensor var_1855_begin_0 = const()[name = string("op_1855_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1855_end_0 = const()[name = string("op_1855_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1855_end_mask_0 = const()[name = string("op_1855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1855_cast_fp16")]; + tensor var_1862_begin_0 = const()[name = string("op_1862_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1862_end_0 = const()[name = string("op_1862_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1862_end_mask_0 = const()[name = string("op_1862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1862_cast_fp16 = slice_by_index(begin = var_1862_begin_0, end = var_1862_end_0, end_mask = var_1862_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1862_cast_fp16")]; + tensor var_1869_begin_0 = const()[name = string("op_1869_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1869_end_0 = const()[name = string("op_1869_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1869_end_mask_0 = const()[name = string("op_1869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1869_cast_fp16 = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1869_cast_fp16")]; + tensor var_1876_begin_0 = const()[name = string("op_1876_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1876_end_0 = const()[name = string("op_1876_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1876_end_mask_0 = const()[name = string("op_1876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1876_cast_fp16 = slice_by_index(begin = var_1876_begin_0, end = var_1876_end_0, end_mask = var_1876_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1876_cast_fp16")]; + tensor var_1883_begin_0 = const()[name = string("op_1883_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1883_end_0 = const()[name = string("op_1883_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1883_end_mask_0 = const()[name = string("op_1883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1883_cast_fp16 = slice_by_index(begin = var_1883_begin_0, end = var_1883_end_0, end_mask = var_1883_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1883_cast_fp16")]; + tensor var_1890_begin_0 = const()[name = string("op_1890_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1890_end_0 = const()[name = string("op_1890_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1890_end_mask_0 = const()[name = string("op_1890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1890_cast_fp16 = slice_by_index(begin = var_1890_begin_0, end = var_1890_end_0, end_mask = var_1890_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1890_cast_fp16")]; + tensor var_1897_begin_0 = const()[name = string("op_1897_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1897_end_0 = const()[name = string("op_1897_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1897_end_mask_0 = const()[name = string("op_1897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1897_cast_fp16 = slice_by_index(begin = var_1897_begin_0, end = var_1897_end_0, end_mask = var_1897_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1897_cast_fp16")]; + tensor var_1904_begin_0 = const()[name = string("op_1904_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1904_end_0 = const()[name = string("op_1904_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1904_end_mask_0 = const()[name = string("op_1904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1904_cast_fp16 = slice_by_index(begin = var_1904_begin_0, end = var_1904_end_0, end_mask = var_1904_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1904_cast_fp16")]; + tensor var_1911_begin_0 = const()[name = string("op_1911_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1911_end_0 = const()[name = string("op_1911_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1911_end_mask_0 = const()[name = string("op_1911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1911_cast_fp16 = slice_by_index(begin = var_1911_begin_0, end = var_1911_end_0, end_mask = var_1911_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1911_cast_fp16")]; + tensor var_1918_begin_0 = const()[name = string("op_1918_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1918_end_0 = const()[name = string("op_1918_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1918_end_mask_0 = const()[name = string("op_1918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1918_cast_fp16 = slice_by_index(begin = var_1918_begin_0, end = var_1918_end_0, end_mask = var_1918_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1918_cast_fp16")]; + tensor var_1925_begin_0 = const()[name = string("op_1925_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1925_end_0 = const()[name = string("op_1925_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1925_end_mask_0 = const()[name = string("op_1925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1925_cast_fp16 = slice_by_index(begin = var_1925_begin_0, end = var_1925_end_0, end_mask = var_1925_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1925_cast_fp16")]; + tensor var_1932_begin_0 = const()[name = string("op_1932_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1932_end_0 = const()[name = string("op_1932_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1932_end_mask_0 = const()[name = string("op_1932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1932_cast_fp16 = slice_by_index(begin = var_1932_begin_0, end = var_1932_end_0, end_mask = var_1932_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1932_cast_fp16")]; + tensor var_1939_begin_0 = const()[name = string("op_1939_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1939_end_0 = const()[name = string("op_1939_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1939_end_mask_0 = const()[name = string("op_1939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1939_cast_fp16 = slice_by_index(begin = var_1939_begin_0, end = var_1939_end_0, end_mask = var_1939_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1939_cast_fp16")]; + tensor var_1946_begin_0 = const()[name = string("op_1946_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1946_end_0 = const()[name = string("op_1946_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1946_end_mask_0 = const()[name = string("op_1946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1946_cast_fp16 = slice_by_index(begin = var_1946_begin_0, end = var_1946_end_0, end_mask = var_1946_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1946_cast_fp16")]; + tensor var_1953_begin_0 = const()[name = string("op_1953_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1953_end_0 = const()[name = string("op_1953_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1953_end_mask_0 = const()[name = string("op_1953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1953_cast_fp16 = slice_by_index(begin = var_1953_begin_0, end = var_1953_end_0, end_mask = var_1953_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1953_cast_fp16")]; + tensor var_1960_begin_0 = const()[name = string("op_1960_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1960_end_0 = const()[name = string("op_1960_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1960_end_mask_0 = const()[name = string("op_1960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1960_cast_fp16 = slice_by_index(begin = var_1960_begin_0, end = var_1960_end_0, end_mask = var_1960_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1960_cast_fp16")]; + tensor var_1967_begin_0 = const()[name = string("op_1967_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1967_end_0 = const()[name = string("op_1967_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1967_end_mask_0 = const()[name = string("op_1967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1967_cast_fp16 = slice_by_index(begin = var_1967_begin_0, end = var_1967_end_0, end_mask = var_1967_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1967_cast_fp16")]; + tensor var_1974_begin_0 = const()[name = string("op_1974_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1974_end_0 = const()[name = string("op_1974_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1974_end_mask_0 = const()[name = string("op_1974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1974_cast_fp16 = slice_by_index(begin = var_1974_begin_0, end = var_1974_end_0, end_mask = var_1974_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1974_cast_fp16")]; + tensor var_1981_begin_0 = const()[name = string("op_1981_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1981_end_0 = const()[name = string("op_1981_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1981_end_mask_0 = const()[name = string("op_1981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1981_cast_fp16 = slice_by_index(begin = var_1981_begin_0, end = var_1981_end_0, end_mask = var_1981_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1981_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1986_begin_0 = const()[name = string("op_1986_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1986_end_0 = const()[name = string("op_1986_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1986_end_mask_0 = const()[name = string("op_1986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_1986_cast_fp16 = slice_by_index(begin = var_1986_begin_0, end = var_1986_end_0, end_mask = var_1986_end_mask_0, x = k_7_cast_fp16)[name = string("op_1986_cast_fp16")]; + tensor var_1990_begin_0 = const()[name = string("op_1990_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1990_end_0 = const()[name = string("op_1990_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1990_end_mask_0 = const()[name = string("op_1990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1990_cast_fp16 = slice_by_index(begin = var_1990_begin_0, end = var_1990_end_0, end_mask = var_1990_end_mask_0, x = k_7_cast_fp16)[name = string("op_1990_cast_fp16")]; + tensor var_1994_begin_0 = const()[name = string("op_1994_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1994_end_0 = const()[name = string("op_1994_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1994_end_mask_0 = const()[name = string("op_1994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1994_cast_fp16 = slice_by_index(begin = var_1994_begin_0, end = var_1994_end_0, end_mask = var_1994_end_mask_0, x = k_7_cast_fp16)[name = string("op_1994_cast_fp16")]; + tensor var_1998_begin_0 = const()[name = string("op_1998_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1998_end_0 = const()[name = string("op_1998_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1998_end_mask_0 = const()[name = string("op_1998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1998_cast_fp16 = slice_by_index(begin = var_1998_begin_0, end = var_1998_end_0, end_mask = var_1998_end_mask_0, x = k_7_cast_fp16)[name = string("op_1998_cast_fp16")]; + tensor var_2002_begin_0 = const()[name = string("op_2002_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2002_end_0 = const()[name = string("op_2002_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2002_end_mask_0 = const()[name = string("op_2002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2002_cast_fp16 = slice_by_index(begin = var_2002_begin_0, end = var_2002_end_0, end_mask = var_2002_end_mask_0, x = k_7_cast_fp16)[name = string("op_2002_cast_fp16")]; + tensor var_2006_begin_0 = const()[name = string("op_2006_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2006_end_0 = const()[name = string("op_2006_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2006_end_mask_0 = const()[name = string("op_2006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2006_cast_fp16 = slice_by_index(begin = var_2006_begin_0, end = var_2006_end_0, end_mask = var_2006_end_mask_0, x = k_7_cast_fp16)[name = string("op_2006_cast_fp16")]; + tensor var_2008_begin_0 = const()[name = string("op_2008_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2008_end_0 = const()[name = string("op_2008_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2008_end_mask_0 = const()[name = string("op_2008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2008_cast_fp16 = slice_by_index(begin = var_2008_begin_0, end = var_2008_end_0, end_mask = var_2008_end_mask_0, x = value_cast_fp16)[name = string("op_2008_cast_fp16")]; + tensor var_2012_begin_0 = const()[name = string("op_2012_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2012_end_0 = const()[name = string("op_2012_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2012_end_mask_0 = const()[name = string("op_2012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2012_cast_fp16 = slice_by_index(begin = var_2012_begin_0, end = var_2012_end_0, end_mask = var_2012_end_mask_0, x = value_cast_fp16)[name = string("op_2012_cast_fp16")]; + tensor var_2016_begin_0 = const()[name = string("op_2016_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2016_end_0 = const()[name = string("op_2016_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2016_end_mask_0 = const()[name = string("op_2016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2016_cast_fp16 = slice_by_index(begin = var_2016_begin_0, end = var_2016_end_0, end_mask = var_2016_end_mask_0, x = value_cast_fp16)[name = string("op_2016_cast_fp16")]; + tensor var_2020_begin_0 = const()[name = string("op_2020_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2020_end_0 = const()[name = string("op_2020_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2020_end_mask_0 = const()[name = string("op_2020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2020_cast_fp16 = slice_by_index(begin = var_2020_begin_0, end = var_2020_end_0, end_mask = var_2020_end_mask_0, x = value_cast_fp16)[name = string("op_2020_cast_fp16")]; + tensor var_2024_begin_0 = const()[name = string("op_2024_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2024_end_0 = const()[name = string("op_2024_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2024_end_mask_0 = const()[name = string("op_2024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2024_cast_fp16 = slice_by_index(begin = var_2024_begin_0, end = var_2024_end_0, end_mask = var_2024_end_mask_0, x = value_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2028_begin_0 = const()[name = string("op_2028_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2028_end_0 = const()[name = string("op_2028_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2028_end_mask_0 = const()[name = string("op_2028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2028_cast_fp16 = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, x = value_cast_fp16)[name = string("op_2028_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1986_cast_fp16, var_1820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1986_cast_fp16, var_1827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1986_cast_fp16, var_1834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1986_cast_fp16, var_1841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1990_cast_fp16, var_1848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1990_cast_fp16, var_1855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1990_cast_fp16, var_1862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1990_cast_fp16, var_1869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1994_cast_fp16, var_1876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1994_cast_fp16, var_1883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1994_cast_fp16, var_1890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1994_cast_fp16, var_1897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1998_cast_fp16, var_1904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1998_cast_fp16, var_1911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1998_cast_fp16, var_1918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1998_cast_fp16, var_1925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_2002_cast_fp16, var_1932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_2002_cast_fp16, var_1939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_2002_cast_fp16, var_1946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_2002_cast_fp16, var_1953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_2006_cast_fp16, var_1960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_2006_cast_fp16, var_1967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_2006_cast_fp16, var_1974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_2006_cast_fp16, var_1981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_2079_to_fp16 = const()[name = string("op_2079_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_2079_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_2081_to_fp16 = const()[name = string("op_2081_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_2081_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_2083_to_fp16 = const()[name = string("op_2083_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_2083_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_2085_to_fp16 = const()[name = string("op_2085_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_2085_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_2087_to_fp16 = const()[name = string("op_2087_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_2087_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_2089_to_fp16 = const()[name = string("op_2089_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_2089_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_2091_to_fp16 = const()[name = string("op_2091_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_2091_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_2093_to_fp16 = const()[name = string("op_2093_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_2093_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_2095_to_fp16 = const()[name = string("op_2095_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_2095_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_2097_to_fp16 = const()[name = string("op_2097_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_2097_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_2099_to_fp16 = const()[name = string("op_2099_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_2099_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_2101_to_fp16 = const()[name = string("op_2101_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_2101_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_2103_to_fp16 = const()[name = string("op_2103_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_2103_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_2105_to_fp16 = const()[name = string("op_2105_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_2105_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_2107_to_fp16 = const()[name = string("op_2107_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_2107_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_2109_to_fp16 = const()[name = string("op_2109_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_2109_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_2111_to_fp16 = const()[name = string("op_2111_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_2111_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_2113_to_fp16 = const()[name = string("op_2113_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_2113_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_2115_to_fp16 = const()[name = string("op_2115_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_2115_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_2117_to_fp16 = const()[name = string("op_2117_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_2117_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_2119_to_fp16 = const()[name = string("op_2119_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_2119_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_2121_to_fp16 = const()[name = string("op_2121_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_2121_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_2123_to_fp16 = const()[name = string("op_2123_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_2123_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_2125_to_fp16 = const()[name = string("op_2125_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_2125_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_2127_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_145_cast_fp16)[name = string("op_2127_cast_fp16")]; + tensor var_2128_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_147_cast_fp16)[name = string("op_2128_cast_fp16")]; + tensor var_2129_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_149_cast_fp16)[name = string("op_2129_cast_fp16")]; + tensor var_2130_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_151_cast_fp16)[name = string("op_2130_cast_fp16")]; + tensor var_2131_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_153_cast_fp16)[name = string("op_2131_cast_fp16")]; + tensor var_2132_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_155_cast_fp16)[name = string("op_2132_cast_fp16")]; + tensor var_2133_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_157_cast_fp16)[name = string("op_2133_cast_fp16")]; + tensor var_2134_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_159_cast_fp16)[name = string("op_2134_cast_fp16")]; + tensor var_2135_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_161_cast_fp16)[name = string("op_2135_cast_fp16")]; + tensor var_2136_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_163_cast_fp16)[name = string("op_2136_cast_fp16")]; + tensor var_2137_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_165_cast_fp16)[name = string("op_2137_cast_fp16")]; + tensor var_2138_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_167_cast_fp16)[name = string("op_2138_cast_fp16")]; + tensor var_2139_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_169_cast_fp16)[name = string("op_2139_cast_fp16")]; + tensor var_2140_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_171_cast_fp16)[name = string("op_2140_cast_fp16")]; + tensor var_2141_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_173_cast_fp16)[name = string("op_2141_cast_fp16")]; + tensor var_2142_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_175_cast_fp16)[name = string("op_2142_cast_fp16")]; + tensor var_2143_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_177_cast_fp16)[name = string("op_2143_cast_fp16")]; + tensor var_2144_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_179_cast_fp16)[name = string("op_2144_cast_fp16")]; + tensor var_2145_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_181_cast_fp16)[name = string("op_2145_cast_fp16")]; + tensor var_2146_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_183_cast_fp16)[name = string("op_2146_cast_fp16")]; + tensor var_2147_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_185_cast_fp16)[name = string("op_2147_cast_fp16")]; + tensor var_2148_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_187_cast_fp16)[name = string("op_2148_cast_fp16")]; + tensor var_2149_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_189_cast_fp16)[name = string("op_2149_cast_fp16")]; + tensor var_2150_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_cast_fp16)[name = string("op_2150_cast_fp16")]; + string var_2152_equation_0 = const()[name = string("op_2152_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2152_cast_fp16 = einsum(equation = var_2152_equation_0, values = (var_2008_cast_fp16, var_2127_cast_fp16))[name = string("op_2152_cast_fp16")]; + string var_2154_equation_0 = const()[name = string("op_2154_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2154_cast_fp16 = einsum(equation = var_2154_equation_0, values = (var_2008_cast_fp16, var_2128_cast_fp16))[name = string("op_2154_cast_fp16")]; + string var_2156_equation_0 = const()[name = string("op_2156_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2156_cast_fp16 = einsum(equation = var_2156_equation_0, values = (var_2008_cast_fp16, var_2129_cast_fp16))[name = string("op_2156_cast_fp16")]; + string var_2158_equation_0 = const()[name = string("op_2158_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2158_cast_fp16 = einsum(equation = var_2158_equation_0, values = (var_2008_cast_fp16, var_2130_cast_fp16))[name = string("op_2158_cast_fp16")]; + string var_2160_equation_0 = const()[name = string("op_2160_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2160_cast_fp16 = einsum(equation = var_2160_equation_0, values = (var_2012_cast_fp16, var_2131_cast_fp16))[name = string("op_2160_cast_fp16")]; + string var_2162_equation_0 = const()[name = string("op_2162_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2162_cast_fp16 = einsum(equation = var_2162_equation_0, values = (var_2012_cast_fp16, var_2132_cast_fp16))[name = string("op_2162_cast_fp16")]; + string var_2164_equation_0 = const()[name = string("op_2164_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2164_cast_fp16 = einsum(equation = var_2164_equation_0, values = (var_2012_cast_fp16, var_2133_cast_fp16))[name = string("op_2164_cast_fp16")]; + string var_2166_equation_0 = const()[name = string("op_2166_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2166_cast_fp16 = einsum(equation = var_2166_equation_0, values = (var_2012_cast_fp16, var_2134_cast_fp16))[name = string("op_2166_cast_fp16")]; + string var_2168_equation_0 = const()[name = string("op_2168_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2168_cast_fp16 = einsum(equation = var_2168_equation_0, values = (var_2016_cast_fp16, var_2135_cast_fp16))[name = string("op_2168_cast_fp16")]; + string var_2170_equation_0 = const()[name = string("op_2170_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2170_cast_fp16 = einsum(equation = var_2170_equation_0, values = (var_2016_cast_fp16, var_2136_cast_fp16))[name = string("op_2170_cast_fp16")]; + string var_2172_equation_0 = const()[name = string("op_2172_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2172_cast_fp16 = einsum(equation = var_2172_equation_0, values = (var_2016_cast_fp16, var_2137_cast_fp16))[name = string("op_2172_cast_fp16")]; + string var_2174_equation_0 = const()[name = string("op_2174_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2174_cast_fp16 = einsum(equation = var_2174_equation_0, values = (var_2016_cast_fp16, var_2138_cast_fp16))[name = string("op_2174_cast_fp16")]; + string var_2176_equation_0 = const()[name = string("op_2176_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2176_cast_fp16 = einsum(equation = var_2176_equation_0, values = (var_2020_cast_fp16, var_2139_cast_fp16))[name = string("op_2176_cast_fp16")]; + string var_2178_equation_0 = const()[name = string("op_2178_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2178_cast_fp16 = einsum(equation = var_2178_equation_0, values = (var_2020_cast_fp16, var_2140_cast_fp16))[name = string("op_2178_cast_fp16")]; + string var_2180_equation_0 = const()[name = string("op_2180_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2180_cast_fp16 = einsum(equation = var_2180_equation_0, values = (var_2020_cast_fp16, var_2141_cast_fp16))[name = string("op_2180_cast_fp16")]; + string var_2182_equation_0 = const()[name = string("op_2182_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2182_cast_fp16 = einsum(equation = var_2182_equation_0, values = (var_2020_cast_fp16, var_2142_cast_fp16))[name = string("op_2182_cast_fp16")]; + string var_2184_equation_0 = const()[name = string("op_2184_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2184_cast_fp16 = einsum(equation = var_2184_equation_0, values = (var_2024_cast_fp16, var_2143_cast_fp16))[name = string("op_2184_cast_fp16")]; + string var_2186_equation_0 = const()[name = string("op_2186_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2186_cast_fp16 = einsum(equation = var_2186_equation_0, values = (var_2024_cast_fp16, var_2144_cast_fp16))[name = string("op_2186_cast_fp16")]; + string var_2188_equation_0 = const()[name = string("op_2188_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2188_cast_fp16 = einsum(equation = var_2188_equation_0, values = (var_2024_cast_fp16, var_2145_cast_fp16))[name = string("op_2188_cast_fp16")]; + string var_2190_equation_0 = const()[name = string("op_2190_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2190_cast_fp16 = einsum(equation = var_2190_equation_0, values = (var_2024_cast_fp16, var_2146_cast_fp16))[name = string("op_2190_cast_fp16")]; + string var_2192_equation_0 = const()[name = string("op_2192_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2192_cast_fp16 = einsum(equation = var_2192_equation_0, values = (var_2028_cast_fp16, var_2147_cast_fp16))[name = string("op_2192_cast_fp16")]; + string var_2194_equation_0 = const()[name = string("op_2194_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2194_cast_fp16 = einsum(equation = var_2194_equation_0, values = (var_2028_cast_fp16, var_2148_cast_fp16))[name = string("op_2194_cast_fp16")]; + string var_2196_equation_0 = const()[name = string("op_2196_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2196_cast_fp16 = einsum(equation = var_2196_equation_0, values = (var_2028_cast_fp16, var_2149_cast_fp16))[name = string("op_2196_cast_fp16")]; + string var_2198_equation_0 = const()[name = string("op_2198_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2198_cast_fp16 = einsum(equation = var_2198_equation_0, values = (var_2028_cast_fp16, var_2150_cast_fp16))[name = string("op_2198_cast_fp16")]; + bool var_2200_interleave_0 = const()[name = string("op_2200_interleave_0"), val = bool(false)]; + tensor var_2200_cast_fp16 = concat(axis = var_1725, interleave = var_2200_interleave_0, values = (var_2152_cast_fp16, var_2154_cast_fp16, var_2156_cast_fp16, var_2158_cast_fp16))[name = string("op_2200_cast_fp16")]; + bool var_2202_interleave_0 = const()[name = string("op_2202_interleave_0"), val = bool(false)]; + tensor var_2202_cast_fp16 = concat(axis = var_1725, interleave = var_2202_interleave_0, values = (var_2160_cast_fp16, var_2162_cast_fp16, var_2164_cast_fp16, var_2166_cast_fp16))[name = string("op_2202_cast_fp16")]; + bool var_2204_interleave_0 = const()[name = string("op_2204_interleave_0"), val = bool(false)]; + tensor var_2204_cast_fp16 = concat(axis = var_1725, interleave = var_2204_interleave_0, values = (var_2168_cast_fp16, var_2170_cast_fp16, var_2172_cast_fp16, var_2174_cast_fp16))[name = string("op_2204_cast_fp16")]; + bool var_2206_interleave_0 = const()[name = string("op_2206_interleave_0"), val = bool(false)]; + tensor var_2206_cast_fp16 = concat(axis = var_1725, interleave = var_2206_interleave_0, values = (var_2176_cast_fp16, var_2178_cast_fp16, var_2180_cast_fp16, var_2182_cast_fp16))[name = string("op_2206_cast_fp16")]; + bool var_2208_interleave_0 = const()[name = string("op_2208_interleave_0"), val = bool(false)]; + tensor var_2208_cast_fp16 = concat(axis = var_1725, interleave = var_2208_interleave_0, values = (var_2184_cast_fp16, var_2186_cast_fp16, var_2188_cast_fp16, var_2190_cast_fp16))[name = string("op_2208_cast_fp16")]; + bool var_2210_interleave_0 = const()[name = string("op_2210_interleave_0"), val = bool(false)]; + tensor var_2210_cast_fp16 = concat(axis = var_1725, interleave = var_2210_interleave_0, values = (var_2192_cast_fp16, var_2194_cast_fp16, var_2196_cast_fp16, var_2198_cast_fp16))[name = string("op_2210_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_1736, interleave = input_25_interleave_0, values = (var_2200_cast_fp16, var_2202_cast_fp16, var_2204_cast_fp16, var_2206_cast_fp16, var_2208_cast_fp16, var_2210_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13760256)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14055232)))]; + tensor obj_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_2229_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056064)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056896)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14057728)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15237440)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15240576)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16420288)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_2267_to_fp16 = const()[name = string("op_2267_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2267_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421120)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421952)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_2291_pad_type_0 = const()[name = string("op_2291_pad_type_0"), val = string("valid")]; + tensor var_2291_strides_0 = const()[name = string("op_2291_strides_0"), val = tensor([1, 1])]; + tensor var_2291_pad_0 = const()[name = string("op_2291_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2291_dilations_0 = const()[name = string("op_2291_dilations_0"), val = tensor([1, 1])]; + int32 var_2291_groups_0 = const()[name = string("op_2291_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16422784)))]; + tensor var_2291_cast_fp16 = conv(dilations = var_2291_dilations_0, groups = var_2291_groups_0, pad = var_2291_pad_0, pad_type = var_2291_pad_type_0, strides = var_2291_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2291_cast_fp16")]; + string var_2298_pad_type_0 = const()[name = string("op_2298_pad_type_0"), val = string("valid")]; + tensor var_2298_strides_0 = const()[name = string("op_2298_strides_0"), val = tensor([1, 1])]; + tensor var_2298_pad_0 = const()[name = string("op_2298_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2298_dilations_0 = const()[name = string("op_2298_dilations_0"), val = tensor([1, 1])]; + int32 var_2298_groups_0 = const()[name = string("op_2298_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16717760)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17012736)))]; + tensor var_2298_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_2298_dilations_0, groups = var_2298_groups_0, pad = var_2298_pad_0, pad_type = var_2298_pad_type_0, strides = var_2298_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2298_cast_fp16")]; + string var_2316_pad_type_0 = const()[name = string("op_2316_pad_type_0"), val = string("valid")]; + tensor var_2316_strides_0 = const()[name = string("op_2316_strides_0"), val = tensor([1, 1])]; + tensor var_2316_pad_0 = const()[name = string("op_2316_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2316_dilations_0 = const()[name = string("op_2316_dilations_0"), val = tensor([1, 1])]; + int32 var_2316_groups_0 = const()[name = string("op_2316_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17013568)))]; + tensor var_2316_cast_fp16 = conv(dilations = var_2316_dilations_0, groups = var_2316_groups_0, pad = var_2316_pad_0, pad_type = var_2316_pad_type_0, strides = var_2316_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2316_cast_fp16")]; + string var_2323_pad_type_0 = const()[name = string("op_2323_pad_type_0"), val = string("valid")]; + tensor var_2323_strides_0 = const()[name = string("op_2323_strides_0"), val = tensor([1, 1])]; + tensor var_2323_pad_0 = const()[name = string("op_2323_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2323_dilations_0 = const()[name = string("op_2323_dilations_0"), val = tensor([1, 1])]; + int32 var_2323_groups_0 = const()[name = string("op_2323_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17308544)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17603520)))]; + tensor var_2323_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_2323_dilations_0, groups = var_2323_groups_0, pad = var_2323_pad_0, pad_type = var_2323_pad_type_0, strides = var_2323_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2323_cast_fp16")]; + string var_2341_pad_type_0 = const()[name = string("op_2341_pad_type_0"), val = string("valid")]; + tensor var_2341_strides_0 = const()[name = string("op_2341_strides_0"), val = tensor([1, 1])]; + tensor var_2341_pad_0 = const()[name = string("op_2341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2341_dilations_0 = const()[name = string("op_2341_dilations_0"), val = tensor([1, 1])]; + int32 var_2341_groups_0 = const()[name = string("op_2341_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17604352)))]; + tensor var_2341_cast_fp16 = conv(dilations = var_2341_dilations_0, groups = var_2341_groups_0, pad = var_2341_pad_0, pad_type = var_2341_pad_type_0, strides = var_2341_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2341_cast_fp16")]; + string var_2348_pad_type_0 = const()[name = string("op_2348_pad_type_0"), val = string("valid")]; + tensor var_2348_strides_0 = const()[name = string("op_2348_strides_0"), val = tensor([1, 1])]; + tensor var_2348_pad_0 = const()[name = string("op_2348_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2348_dilations_0 = const()[name = string("op_2348_dilations_0"), val = tensor([1, 1])]; + int32 var_2348_groups_0 = const()[name = string("op_2348_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17899328)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18194304)))]; + tensor var_2348_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_2348_dilations_0, groups = var_2348_groups_0, pad = var_2348_pad_0, pad_type = var_2348_pad_type_0, strides = var_2348_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2348_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18195136)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18490112)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18785088)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_2378 = const()[name = string("op_2378"), val = int32(0)]; + bool input_35_interleave_0 = const()[name = string("input_35_interleave_0"), val = bool(false)]; + tensor input_35_cast_fp16 = concat(axis = var_2378, interleave = input_35_interleave_0, values = (var_2291_cast_fp16, var_2316_cast_fp16, var_2341_cast_fp16, k_cast_fp16))[name = string("input_35_cast_fp16")]; + int32 var_2381 = const()[name = string("op_2381"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_2381, interleave = input_interleave_0, values = (var_2298_cast_fp16, var_2323_cast_fp16, var_2348_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_2388_pad_0 = const()[name = string("op_2388_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_2388_mode_0 = const()[name = string("op_2388_mode_0"), val = string("constant")]; + fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_5_to_fp16, mode = var_2388_mode_0, pad = var_2388_pad_0, x = input_35_cast_fp16)[name = string("op_2388_cast_fp16")]; + tensor var_2394_pad_0 = const()[name = string("op_2394_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_2394_mode_0 = const()[name = string("op_2394_mode_0"), val = string("constant")]; + fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_6_to_fp16, mode = var_2394_mode_0, pad = var_2394_pad_0, x = input_cast_fp16)[name = string("op_2394_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file diff --git a/openai_whisper-tiny/AudioEncoder.mlmodelc/weights/weight.bin b/openai_whisper-tiny/AudioEncoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..bf467108d1c0a69dae7e4c528b3eef255b05d059 --- /dev/null +++ b/openai_whisper-tiny/AudioEncoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f49cb4dfcf5c5bc4a24d29b33b72788fe759b623c3da1cc808bd79caf6c2e35a +size 18785920 diff --git a/openai_whisper-tiny/LICENSE_NOTICE.txt b/openai_whisper-tiny/LICENSE_NOTICE.txt new file mode 100644 index 0000000000000000000000000000000000000000..be2da6c6e6d746ab53f1b21eac16d611aed1193a --- /dev/null +++ b/openai_whisper-tiny/LICENSE_NOTICE.txt @@ -0,0 +1,7 @@ +Argmax proprietary and confidential. Under NDA. + +Copyright 2024 Argmax, Inc. All rights reserved. + +Unauthorized access, copying, use, distribution, and or commercialization of this file, via any medium or means is strictly prohibited. + +Please contact Argmax for licensing information at info@argmaxinc.com. diff --git a/openai_whisper-tiny/MelSpectrogram.mlmodelc/analytics/coremldata.bin b/openai_whisper-tiny/MelSpectrogram.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a11fbb2cd75b96eb2120a672afefa298c2ef857b --- /dev/null +++ b/openai_whisper-tiny/MelSpectrogram.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc05e563ee0c556e3f578e04be5fb67b4e7520124403f2561f39102f0f2b33d +size 243 diff --git a/openai_whisper-tiny/MelSpectrogram.mlmodelc/coremldata.bin b/openai_whisper-tiny/MelSpectrogram.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..a3544b6644c1af93ca6bdabb67a1c51e80eaa552 --- /dev/null +++ b/openai_whisper-tiny/MelSpectrogram.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ef11ea703011eab03287ec661f999e19c2c78cf67d531b5e6afa02e18f913d +size 328 diff --git a/openai_whisper-tiny/MelSpectrogram.mlmodelc/metadata.json b/openai_whisper-tiny/MelSpectrogram.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..1a60dd494a857817b67d87cd920baa6824e74b61 --- /dev/null +++ b/openai_whisper-tiny/MelSpectrogram.mlmodelc/metadata.json @@ -0,0 +1,74 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 80 × 1 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 1, 3000]", + "name" : "melspectrogram_features", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.mul" : 2, + "Ios18.square" : 2, + "Ios18.conv" : 2, + "Ios18.matmul" : 1, + "Ios18.expandDims" : 4, + "Ios18.sub" : 1, + "Ios18.log" : 1, + "Ios18.add" : 3, + "Ios18.sliceByIndex" : 1, + "Ios18.maximum" : 1, + "Ios18.squeeze" : 2, + "Ios18.reshape" : 2, + "Ios16.reduceMax" : 1, + "Identity" : 1, + "Pad" : 1 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "stateSchema" : [ + + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 480000)", + "shortDescription" : "", + "shape" : "[480000]", + "name" : "audio", + "type" : "MultiArray" + } + ], + "generatedClassName" : "MelSpectrogram", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-tiny/MelSpectrogram.mlmodelc/model.mil b/openai_whisper-tiny/MelSpectrogram.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..cf4cd446f68b88655d00a7df7063aa46937a9bdd --- /dev/null +++ b/openai_whisper-tiny/MelSpectrogram.mlmodelc/model.mil @@ -0,0 +1,66 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor audio) { + tensor var_10 = const()[name = string("op_10"), val = tensor([1, 1, 480000])]; + tensor input_1_cast_fp16 = reshape(shape = var_10, x = audio)[name = string("input_1_cast_fp16")]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 200, 200])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("reflect")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + tensor var_22 = const()[name = string("op_22"), val = tensor([480400])]; + tensor input_cast_fp16 = reshape(shape = var_22, x = input_3_cast_fp16)[name = string("input_cast_fp16")]; + tensor expand_dims_0_axes_0 = const()[name = string("expand_dims_0_axes_0"), val = tensor([0])]; + tensor expand_dims_0_cast_fp16 = expand_dims(axes = expand_dims_0_axes_0, x = input_cast_fp16)[name = string("expand_dims_0_cast_fp16")]; + tensor expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor([160])]; + tensor expand_dims_4_axes_0 = const()[name = string("expand_dims_4_axes_0"), val = tensor([1])]; + tensor expand_dims_4_cast_fp16 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0_cast_fp16)[name = string("expand_dims_4_cast_fp16")]; + string conv_0_pad_type_0 = const()[name = string("conv_0_pad_type_0"), val = string("valid")]; + tensor conv_0_pad_0 = const()[name = string("conv_0_pad_0"), val = tensor([0, 0])]; + tensor conv_0_dilations_0 = const()[name = string("conv_0_dilations_0"), val = tensor([1])]; + int32 conv_0_groups_0 = const()[name = string("conv_0_groups_0"), val = int32(1)]; + tensor expand_dims_1_to_fp16 = const()[name = string("expand_dims_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor conv_0_cast_fp16 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_0_cast_fp16")]; + string conv_1_pad_type_0 = const()[name = string("conv_1_pad_type_0"), val = string("valid")]; + tensor conv_1_pad_0 = const()[name = string("conv_1_pad_0"), val = tensor([0, 0])]; + tensor conv_1_dilations_0 = const()[name = string("conv_1_dilations_0"), val = tensor([1])]; + int32 conv_1_groups_0 = const()[name = string("conv_1_groups_0"), val = int32(1)]; + tensor expand_dims_2_to_fp16 = const()[name = string("expand_dims_2_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160960)))]; + tensor conv_1_cast_fp16 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2_to_fp16, x = expand_dims_4_cast_fp16)[name = string("conv_1_cast_fp16")]; + tensor squeeze_0_axes_0 = const()[name = string("squeeze_0_axes_0"), val = tensor([0])]; + tensor squeeze_0_cast_fp16 = squeeze(axes = squeeze_0_axes_0, x = conv_0_cast_fp16)[name = string("squeeze_0_cast_fp16")]; + tensor squeeze_1_axes_0 = const()[name = string("squeeze_1_axes_0"), val = tensor([0])]; + tensor squeeze_1_cast_fp16 = squeeze(axes = squeeze_1_axes_0, x = conv_1_cast_fp16)[name = string("squeeze_1_cast_fp16")]; + tensor square_0_cast_fp16 = square(x = squeeze_0_cast_fp16)[name = string("square_0_cast_fp16")]; + tensor square_1_cast_fp16 = square(x = squeeze_1_cast_fp16)[name = string("square_1_cast_fp16")]; + tensor add_1_cast_fp16 = add(x = square_0_cast_fp16, y = square_1_cast_fp16)[name = string("add_1_cast_fp16")]; + tensor magnitudes_1_cast_fp16 = identity(x = add_1_cast_fp16)[name = string("magnitudes_1_cast_fp16")]; + tensor magnitudes_begin_0 = const()[name = string("magnitudes_begin_0"), val = tensor([0, 0])]; + tensor magnitudes_end_0 = const()[name = string("magnitudes_end_0"), val = tensor([201, 3000])]; + tensor magnitudes_end_mask_0 = const()[name = string("magnitudes_end_mask_0"), val = tensor([true, false])]; + tensor magnitudes_cast_fp16 = slice_by_index(begin = magnitudes_begin_0, end = magnitudes_end_0, end_mask = magnitudes_end_mask_0, x = magnitudes_1_cast_fp16)[name = string("magnitudes_cast_fp16")]; + bool mel_spec_1_transpose_x_0 = const()[name = string("mel_spec_1_transpose_x_0"), val = bool(false)]; + bool mel_spec_1_transpose_y_0 = const()[name = string("mel_spec_1_transpose_y_0"), val = bool(false)]; + tensor mel_filters_to_fp16 = const()[name = string("mel_filters_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321856)))]; + tensor mel_spec_1_cast_fp16 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters_to_fp16, y = magnitudes_cast_fp16)[name = string("mel_spec_1_cast_fp16")]; + fp16 var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = fp16(0x1p-24)]; + tensor mel_spec_cast_fp16 = add(x = mel_spec_1_cast_fp16, y = var_41_to_fp16)[name = string("mel_spec_cast_fp16")]; + fp32 log_0_epsilon_0 = const()[name = string("log_0_epsilon_0"), val = fp32(0x1p-149)]; + tensor log_0_cast_fp16 = log(epsilon = log_0_epsilon_0, x = mel_spec_cast_fp16)[name = string("log_0_cast_fp16")]; + fp16 mul_0_y_0_to_fp16 = const()[name = string("mul_0_y_0_to_fp16"), val = fp16(0x1.bccp-2)]; + tensor mul_0_cast_fp16 = mul(x = log_0_cast_fp16, y = mul_0_y_0_to_fp16)[name = string("mul_0_cast_fp16")]; + bool var_44_keep_dims_0 = const()[name = string("op_44_keep_dims_0"), val = bool(false)]; + fp16 var_44_cast_fp16 = reduce_max(keep_dims = var_44_keep_dims_0, x = mul_0_cast_fp16)[name = string("op_44_cast_fp16")]; + fp16 var_46_to_fp16 = const()[name = string("op_46_to_fp16"), val = fp16(0x1p+3)]; + fp16 var_47_cast_fp16 = sub(x = var_44_cast_fp16, y = var_46_to_fp16)[name = string("op_47_cast_fp16")]; + tensor log_spec_3_cast_fp16 = maximum(x = mul_0_cast_fp16, y = var_47_cast_fp16)[name = string("log_spec_3_cast_fp16")]; + fp16 var_50_to_fp16 = const()[name = string("op_50_to_fp16"), val = fp16(0x1p+2)]; + tensor var_51_cast_fp16 = add(x = log_spec_3_cast_fp16, y = var_50_to_fp16)[name = string("op_51_cast_fp16")]; + fp16 _inversed_log_spec_y_0_to_fp16 = const()[name = string("_inversed_log_spec_y_0_to_fp16"), val = fp16(0x1p-2)]; + tensor _inversed_log_spec_cast_fp16 = mul(x = var_51_cast_fp16, y = _inversed_log_spec_y_0_to_fp16)[name = string("_inversed_log_spec_cast_fp16")]; + tensor var_55_axes_0 = const()[name = string("op_55_axes_0"), val = tensor([0])]; + tensor var_55_cast_fp16 = expand_dims(axes = var_55_axes_0, x = _inversed_log_spec_cast_fp16)[name = string("op_55_cast_fp16")]; + tensor var_62_axes_0 = const()[name = string("op_62_axes_0"), val = tensor([2])]; + tensor melspectrogram_features = expand_dims(axes = var_62_axes_0, x = var_55_cast_fp16)[name = string("op_62_cast_fp16")]; + } -> (melspectrogram_features); +} \ No newline at end of file diff --git a/openai_whisper-tiny/MelSpectrogram.mlmodelc/weights/weight.bin b/openai_whisper-tiny/MelSpectrogram.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..f7d28ffac464e9e7086a526930f0059187de8d01 --- /dev/null +++ b/openai_whisper-tiny/MelSpectrogram.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:801024dbc7a89c677be1f8b285de3409e35f7d1786c9c8d9d0d6842ac57a1c83 +size 354080 diff --git a/openai_whisper-tiny/TextDecoder.mlmodelc/analytics/coremldata.bin b/openai_whisper-tiny/TextDecoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..664d33e548427ca0c2111c58d89a4dc0db292dfb --- /dev/null +++ b/openai_whisper-tiny/TextDecoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bab583585965381063fd13cebda65d33b49eb488fec8d4aa2ee73f5f3a990f69 +size 243 diff --git a/openai_whisper-tiny/TextDecoder.mlmodelc/coremldata.bin b/openai_whisper-tiny/TextDecoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..0bd6ee79dbfece281bd0d12c2ef9cfb26b59de30 --- /dev/null +++ b/openai_whisper-tiny/TextDecoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46f631ddf12d70532eacec7e50d557710564fa2228686af7178ddc7426885d6f +size 754 diff --git a/openai_whisper-tiny/TextDecoder.mlmodelc/metadata.json b/openai_whisper-tiny/TextDecoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..ce391ae6546efc2c990acc504864702de5ac1f8d --- /dev/null +++ b/openai_whisper-tiny/TextDecoder.mlmodelc/metadata.json @@ -0,0 +1,183 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1 × 51865)", + "shortDescription" : "", + "shape" : "[1, 1, 51865]", + "name" : "logits", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 1536, 1, 1]", + "name" : "key_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 1)", + "shortDescription" : "", + "shape" : "[1, 1536, 1, 1]", + "name" : "value_cache_updates", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "alignment_heads_weights", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 9, + "mlProgramOperationTypeHistogram" : { + "Ios18.expandDims" : 8, + "Ios18.softmax" : 8, + "Ios18.mul" : 16, + "Ios18.matmul" : 16, + "Ios18.batchNorm" : 13, + "Ios16.reduceMean" : 1, + "Split" : 2, + "Ios18.readState" : 5, + "Ios18.gather" : 2, + "Ios18.add" : 29, + "Ios18.layerNorm" : 13, + "Ios18.reshape" : 32, + "Ios18.linear" : 1, + "Ios18.conv" : 32, + "Ios18.gelu" : 4, + "Ios18.concat" : 3, + "Ios18.cast" : 1, + "Ios18.transpose" : 1, + "Ios18.sliceByIndex" : 20, + "Ios18.squeeze" : 1 + }, + "computePrecision" : "Mixed (Float16, Int32, UInt16)", + "isUpdatable" : "0", + "stateSchema" : [ + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 1 × 1536)", + "shortDescription" : "", + "shape" : "[1, 1536]", + "name" : "encoder_attn_key_padding_mask", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 1536)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 1536]", + "name" : "encoder_attn_value_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 448]", + "name" : "self_attn_key_cache", + "type" : "State" + }, + { + "dataType" : "Float16", + "isOptional" : "0", + "formattedType" : "State (Float16 4 × 384 × 1 × 448)", + "shortDescription" : "", + "shape" : "[4, 384, 1, 448]", + "name" : "self_attn_value_cache", + "type" : "State" + } + ], + "availability" : { + "macOS" : "15.0", + "tvOS" : "18.0", + "visionOS" : "2.0", + "watchOS" : "11.0", + "iOS" : "18.0", + "macCatalyst" : "18.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + "com.github.apple.coremltools.source_dialect" : "TorchScript", + "com.github.apple.coremltools.source" : "torch==2.5.1", + "com.github.apple.coremltools.version" : "8.0" + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "input_ids", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Int32", + "formattedType" : "MultiArray (Int32 1)", + "shortDescription" : "", + "shape" : "[1]", + "name" : "cache_length", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "kv_cache_update_mask", + "type" : "MultiArray" + }, + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float16", + "formattedType" : "MultiArray (Float16 1 × 448)", + "shortDescription" : "", + "shape" : "[1, 448]", + "name" : "decoder_key_padding_mask", + "type" : "MultiArray" + } + ], + "generatedClassName" : "TextDecoderStateful", + "method" : "predict" + } +] \ No newline at end of file diff --git a/openai_whisper-tiny/TextDecoder.mlmodelc/model.mil b/openai_whisper-tiny/TextDecoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..d47de888a81af65d09b40089c26c9ea57f38a229 --- /dev/null +++ b/openai_whisper-tiny/TextDecoder.mlmodelc/model.mil @@ -0,0 +1,679 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor cache_length, tensor decoder_key_padding_mask, state> encoder_attn_key_cache, state> encoder_attn_key_padding_mask, state> encoder_attn_value_cache, tensor input_ids, tensor kv_cache_update_mask, state> self_attn_key_cache, state> self_attn_value_cache) { + int32 var_26_axis_0 = const()[name = string("op_26_axis_0"), val = int32(0)]; + int32 var_26_batch_dims_0 = const()[name = string("op_26_batch_dims_0"), val = int32(0)]; + bool var_26_validate_indices_0 = const()[name = string("op_26_validate_indices_0"), val = bool(false)]; + tensor embed_tokens_weight_to_fp16 = const()[name = string("embed_tokens_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_26_cast_fp16 = gather(axis = var_26_axis_0, batch_dims = var_26_batch_dims_0, indices = input_ids, validate_indices = var_26_validate_indices_0, x = embed_tokens_weight_to_fp16)[name = string("op_26_cast_fp16")]; + int32 var_30_axis_0 = const()[name = string("op_30_axis_0"), val = int32(0)]; + int32 var_30_batch_dims_0 = const()[name = string("op_30_batch_dims_0"), val = int32(0)]; + bool var_30_validate_indices_0 = const()[name = string("op_30_validate_indices_0"), val = bool(false)]; + tensor embed_positions_weight_to_fp16 = const()[name = string("embed_positions_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39832448)))]; + string cache_length_to_uint16_dtype_0 = const()[name = string("cache_length_to_uint16_dtype_0"), val = string("uint16")]; + tensor cache_length_to_uint16 = cast(dtype = cache_length_to_uint16_dtype_0, x = cache_length)[name = string("cast_71")]; + tensor var_30_cast_fp16_cast_uint16 = gather(axis = var_30_axis_0, batch_dims = var_30_batch_dims_0, indices = cache_length_to_uint16, validate_indices = var_30_validate_indices_0, x = embed_positions_weight_to_fp16)[name = string("op_30_cast_fp16_cast_uint16")]; + tensor hidden_states_1_cast_fp16 = add(x = var_26_cast_fp16, y = var_30_cast_fp16_cast_uint16)[name = string("hidden_states_1_cast_fp16")]; + tensor var_44_axes_0 = const()[name = string("op_44_axes_0"), val = tensor([2])]; + tensor var_44_cast_fp16 = expand_dims(axes = var_44_axes_0, x = hidden_states_1_cast_fp16)[name = string("op_44_cast_fp16")]; + tensor inputs_1_axes_0 = const()[name = string("inputs_1_axes_0"), val = tensor([3])]; + tensor inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_44_cast_fp16)[name = string("inputs_1_cast_fp16")]; + tensor read_state_0 = read_state(input = self_attn_key_cache)[name = string("read_state_0")]; + tensor tile_0 = const()[name = string("tile_0"), val = tensor([1, 1, 1, 1])]; + int32 var_49_axis_0 = const()[name = string("op_49_axis_0"), val = int32(0)]; + tensor var_49_cast_fp16_0, tensor var_49_cast_fp16_1, tensor var_49_cast_fp16_2, tensor var_49_cast_fp16_3 = split(axis = var_49_axis_0, split_sizes = tile_0, x = read_state_0)[name = string("op_49_cast_fp16")]; + tensor read_state_1 = read_state(input = self_attn_value_cache)[name = string("read_state_1")]; + tensor tile_1 = const()[name = string("tile_1"), val = tensor([1, 1, 1, 1])]; + int32 var_56_axis_0 = const()[name = string("op_56_axis_0"), val = int32(0)]; + tensor var_56_cast_fp16_0, tensor var_56_cast_fp16_1, tensor var_56_cast_fp16_2, tensor var_56_cast_fp16_3 = split(axis = var_56_axis_0, split_sizes = tile_1, x = read_state_1)[name = string("op_56_cast_fp16")]; + tensor read_state_2 = read_state(input = encoder_attn_key_cache)[name = string("read_state_2")]; + tensor obj_17_begin_0 = const()[name = string("obj_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_17_end_0 = const()[name = string("obj_17_end_0"), val = tensor([1, 384, 1, 1536])]; + tensor obj_17_end_mask_0 = const()[name = string("obj_17_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_17_cast_fp16 = slice_by_index(begin = obj_17_begin_0, end = obj_17_end_0, end_mask = obj_17_end_mask_0, x = read_state_2)[name = string("obj_17_cast_fp16")]; + tensor read_state_3 = read_state(input = encoder_attn_value_cache)[name = string("read_state_3")]; + tensor obj_19_begin_0 = const()[name = string("obj_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_end_0 = const()[name = string("obj_19_end_0"), val = tensor([1, 384, 1, 1536])]; + tensor obj_19_end_mask_0 = const()[name = string("obj_19_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_19_cast_fp16 = slice_by_index(begin = obj_19_begin_0, end = obj_19_end_0, end_mask = obj_19_end_mask_0, x = read_state_3)[name = string("obj_19_cast_fp16")]; + int32 var_76 = const()[name = string("op_76"), val = int32(3)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_101_to_fp16 = const()[name = string("op_101_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_101_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_5_mean_0_to_fp16 = const()[name = string("obj_5_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40176576)))]; + tensor obj_5_variance_0_to_fp16 = const()[name = string("obj_5_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40177408)))]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40178240)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40179072)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40179904)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40474880)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_1_cast_fp16")]; + string current_key_1_pad_type_0 = const()[name = string("current_key_1_pad_type_0"), val = string("valid")]; + tensor current_key_1_strides_0 = const()[name = string("current_key_1_strides_0"), val = tensor([1, 1])]; + tensor current_key_1_pad_0 = const()[name = string("current_key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_1_dilations_0 = const()[name = string("current_key_1_dilations_0"), val = tensor([1, 1])]; + int32 current_key_1_groups_0 = const()[name = string("current_key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40475712)))]; + tensor current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_key_1_cast_fp16")]; + string current_value_1_pad_type_0 = const()[name = string("current_value_1_pad_type_0"), val = string("valid")]; + tensor current_value_1_strides_0 = const()[name = string("current_value_1_strides_0"), val = tensor([1, 1])]; + tensor current_value_1_pad_0 = const()[name = string("current_value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_1_dilations_0 = const()[name = string("current_value_1_dilations_0"), val = tensor([1, 1])]; + int32 current_value_1_groups_0 = const()[name = string("current_value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(40770688)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41065664)))]; + tensor current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("current_value_1_cast_fp16")]; + tensor var_136_axes_0 = const()[name = string("op_136_axes_0"), val = tensor([1])]; + tensor var_136_cast_fp16 = expand_dims(axes = var_136_axes_0, x = kv_cache_update_mask)[name = string("op_136_cast_fp16")]; + tensor var_137_axes_0 = const()[name = string("op_137_axes_0"), val = tensor([2])]; + tensor var_137_cast_fp16 = expand_dims(axes = var_137_axes_0, x = var_136_cast_fp16)[name = string("op_137_cast_fp16")]; + tensor var_139_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_139_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_49_cast_fp16_0, y = var_139_cast_fp16)[name = string("key_1_cast_fp16")]; + tensor var_141_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_137_cast_fp16)[name = string("op_141_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_56_cast_fp16_0, y = var_141_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_144 = const()[name = string("op_144"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_144, x = query_1_cast_fp16)[name = string("mh_q_1_cast_fp16")]; + fp16 var_146_to_fp16 = const()[name = string("op_146_to_fp16"), val = fp16(0x1p-3)]; + tensor var_147_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_146_to_fp16)[name = string("op_147_cast_fp16")]; + tensor var_148 = const()[name = string("op_148"), val = tensor([1, 6, 64, -1])]; + tensor var_149_cast_fp16 = reshape(shape = var_148, x = key_1_cast_fp16)[name = string("op_149_cast_fp16")]; + bool mh_w_1_transpose_x_0 = const()[name = string("mh_w_1_transpose_x_0"), val = bool(true)]; + bool mh_w_1_transpose_y_0 = const()[name = string("mh_w_1_transpose_y_0"), val = bool(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_147_cast_fp16, y = var_149_cast_fp16)[name = string("mh_w_1_cast_fp16")]; + tensor var_153_axes_0 = const()[name = string("op_153_axes_0"), val = tensor([1])]; + tensor var_153_cast_fp16 = expand_dims(axes = var_153_axes_0, x = decoder_key_padding_mask)[name = string("op_153_cast_fp16")]; + tensor var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor([2])]; + tensor var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = var_153_cast_fp16)[name = string("op_154_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_3_cast_fp16")]; + tensor var_157_cast_fp16 = softmax(axis = var_76, x = mh_w_3_cast_fp16)[name = string("op_157_cast_fp16")]; + tensor var_158 = const()[name = string("op_158"), val = tensor([1, 6, 64, -1])]; + tensor var_159_cast_fp16 = reshape(shape = var_158, x = value_1_cast_fp16)[name = string("op_159_cast_fp16")]; + bool attn_1_transpose_x_0 = const()[name = string("attn_1_transpose_x_0"), val = bool(false)]; + bool attn_1_transpose_y_0 = const()[name = string("attn_1_transpose_y_0"), val = bool(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_159_cast_fp16, y = var_157_cast_fp16)[name = string("attn_1_cast_fp16")]; + tensor var_162 = const()[name = string("op_162"), val = tensor([1, 384, 1, -1])]; + tensor input_1_cast_fp16 = reshape(shape = var_162, x = attn_1_cast_fp16)[name = string("input_1_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41066496)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41361472)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_184_to_fp16 = const()[name = string("op_184_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_184_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41362304)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41363136)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41363968)))]; + tensor layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41658944)))]; + tensor query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_3_cast_fp16")]; + tensor var_204 = const()[name = string("op_204"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_204, x = query_3_cast_fp16)[name = string("mh_q_3_cast_fp16")]; + fp16 var_206_to_fp16 = const()[name = string("op_206_to_fp16"), val = fp16(0x1p-3)]; + tensor var_207_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_206_to_fp16)[name = string("op_207_cast_fp16")]; + tensor var_208 = const()[name = string("op_208"), val = tensor([1, 6, 64, -1])]; + tensor var_209_cast_fp16 = reshape(shape = var_208, x = obj_17_cast_fp16)[name = string("op_209_cast_fp16")]; + bool mh_w_5_transpose_x_0 = const()[name = string("mh_w_5_transpose_x_0"), val = bool(true)]; + bool mh_w_5_transpose_y_0 = const()[name = string("mh_w_5_transpose_y_0"), val = bool(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_207_cast_fp16, y = var_209_cast_fp16)[name = string("mh_w_5_cast_fp16")]; + tensor read_state_4 = read_state(input = encoder_attn_key_padding_mask)[name = string("read_state_4")]; + tensor var_213_axes_0 = const()[name = string("op_213_axes_0"), val = tensor([1])]; + tensor var_213_cast_fp16 = expand_dims(axes = var_213_axes_0, x = read_state_4)[name = string("op_213_cast_fp16")]; + tensor var_214_axes_0 = const()[name = string("op_214_axes_0"), val = tensor([2])]; + tensor var_214_cast_fp16 = expand_dims(axes = var_214_axes_0, x = var_213_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_7_cast_fp16")]; + tensor obj_23_cast_fp16 = softmax(axis = var_76, x = mh_w_7_cast_fp16)[name = string("obj_23_cast_fp16")]; + tensor var_218 = const()[name = string("op_218"), val = tensor([1, 6, 64, -1])]; + tensor var_219_cast_fp16 = reshape(shape = var_218, x = obj_19_cast_fp16)[name = string("op_219_cast_fp16")]; + bool attn_3_transpose_x_0 = const()[name = string("attn_3_transpose_x_0"), val = bool(false)]; + bool attn_3_transpose_y_0 = const()[name = string("attn_3_transpose_y_0"), val = bool(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_219_cast_fp16, y = obj_23_cast_fp16)[name = string("attn_3_cast_fp16")]; + tensor var_222 = const()[name = string("op_222"), val = tensor([1, 384, 1, -1])]; + tensor input_3_cast_fp16 = reshape(shape = var_222, x = attn_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string obj_21_pad_type_0 = const()[name = string("obj_21_pad_type_0"), val = string("valid")]; + tensor obj_21_strides_0 = const()[name = string("obj_21_strides_0"), val = tensor([1, 1])]; + tensor obj_21_pad_0 = const()[name = string("obj_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_21_dilations_0 = const()[name = string("obj_21_dilations_0"), val = tensor([1, 1])]; + int32 obj_21_groups_0 = const()[name = string("obj_21_groups_0"), val = int32(1)]; + tensor layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41659776)))]; + tensor layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41954752)))]; + tensor obj_21_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = string("obj_21_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_21_cast_fp16)[name = string("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_240_to_fp16 = const()[name = string("op_240_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_240_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor input_5_gamma_0_to_fp16 = const()[name = string("input_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41955584)))]; + tensor input_5_beta_0_to_fp16 = const()[name = string("input_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41956416)))]; + fp16 input_5_epsilon_0_to_fp16 = const()[name = string("input_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41957248)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43136960)))]; + tensor input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string input_9_mode_0 = const()[name = string("input_9_mode_0"), val = string("EXACT")]; + tensor input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; + string hidden_states_3_pad_type_0 = const()[name = string("hidden_states_3_pad_type_0"), val = string("valid")]; + tensor hidden_states_3_strides_0 = const()[name = string("hidden_states_3_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_0 = const()[name = string("hidden_states_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_3_dilations_0 = const()[name = string("hidden_states_3_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_3_groups_0 = const()[name = string("hidden_states_3_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43140096)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44319808)))]; + tensor hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor obj_35_begin_0 = const()[name = string("obj_35_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_35_end_0 = const()[name = string("obj_35_end_0"), val = tensor([2, 384, 1, 1536])]; + tensor obj_35_end_mask_0 = const()[name = string("obj_35_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_35_cast_fp16 = slice_by_index(begin = obj_35_begin_0, end = obj_35_end_0, end_mask = obj_35_end_mask_0, x = read_state_2)[name = string("obj_35_cast_fp16")]; + tensor obj_37_begin_0 = const()[name = string("obj_37_begin_0"), val = tensor([1, 0, 0, 0])]; + tensor obj_37_end_0 = const()[name = string("obj_37_end_0"), val = tensor([2, 384, 1, 1536])]; + tensor obj_37_end_mask_0 = const()[name = string("obj_37_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_37_cast_fp16 = slice_by_index(begin = obj_37_begin_0, end = obj_37_end_0, end_mask = obj_37_end_mask_0, x = read_state_3)[name = string("obj_37_cast_fp16")]; + int32 var_285 = const()[name = string("op_285"), val = int32(3)]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_310_to_fp16 = const()[name = string("op_310_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_310_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = string("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44320640)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = string("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44321472)))]; + fp16 obj_25_epsilon_0_to_fp16 = const()[name = string("obj_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("obj_25_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44322304)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44617280)))]; + tensor query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("query_5_cast_fp16")]; + string current_key_3_pad_type_0 = const()[name = string("current_key_3_pad_type_0"), val = string("valid")]; + tensor current_key_3_strides_0 = const()[name = string("current_key_3_strides_0"), val = tensor([1, 1])]; + tensor current_key_3_pad_0 = const()[name = string("current_key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_3_dilations_0 = const()[name = string("current_key_3_dilations_0"), val = tensor([1, 1])]; + int32 current_key_3_groups_0 = const()[name = string("current_key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44618112)))]; + tensor current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_key_3_cast_fp16")]; + string current_value_3_pad_type_0 = const()[name = string("current_value_3_pad_type_0"), val = string("valid")]; + tensor current_value_3_strides_0 = const()[name = string("current_value_3_strides_0"), val = tensor([1, 1])]; + tensor current_value_3_pad_0 = const()[name = string("current_value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_3_dilations_0 = const()[name = string("current_value_3_dilations_0"), val = tensor([1, 1])]; + int32 current_value_3_groups_0 = const()[name = string("current_value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44913088)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45208064)))]; + tensor current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = string("current_value_3_cast_fp16")]; + tensor var_348_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_348_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_49_cast_fp16_1, y = var_348_cast_fp16)[name = string("key_3_cast_fp16")]; + tensor var_350_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_137_cast_fp16)[name = string("op_350_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_56_cast_fp16_1, y = var_350_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_353 = const()[name = string("op_353"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_353, x = query_5_cast_fp16)[name = string("mh_q_5_cast_fp16")]; + fp16 var_355_to_fp16 = const()[name = string("op_355_to_fp16"), val = fp16(0x1p-3)]; + tensor var_356_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_355_to_fp16)[name = string("op_356_cast_fp16")]; + tensor var_357 = const()[name = string("op_357"), val = tensor([1, 6, 64, -1])]; + tensor var_358_cast_fp16 = reshape(shape = var_357, x = key_3_cast_fp16)[name = string("op_358_cast_fp16")]; + bool mh_w_9_transpose_x_0 = const()[name = string("mh_w_9_transpose_x_0"), val = bool(true)]; + bool mh_w_9_transpose_y_0 = const()[name = string("mh_w_9_transpose_y_0"), val = bool(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_356_cast_fp16, y = var_358_cast_fp16)[name = string("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_11_cast_fp16")]; + tensor var_366_cast_fp16 = softmax(axis = var_285, x = mh_w_11_cast_fp16)[name = string("op_366_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([1, 6, 64, -1])]; + tensor var_368_cast_fp16 = reshape(shape = var_367, x = value_3_cast_fp16)[name = string("op_368_cast_fp16")]; + bool attn_5_transpose_x_0 = const()[name = string("attn_5_transpose_x_0"), val = bool(false)]; + bool attn_5_transpose_y_0 = const()[name = string("attn_5_transpose_y_0"), val = bool(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_368_cast_fp16, y = var_366_cast_fp16)[name = string("attn_5_cast_fp16")]; + tensor var_371 = const()[name = string("op_371"), val = tensor([1, 384, 1, -1])]; + tensor input_11_cast_fp16 = reshape(shape = var_371, x = attn_5_cast_fp16)[name = string("input_11_cast_fp16")]; + string obj_31_pad_type_0 = const()[name = string("obj_31_pad_type_0"), val = string("valid")]; + tensor obj_31_strides_0 = const()[name = string("obj_31_strides_0"), val = tensor([1, 1])]; + tensor obj_31_pad_0 = const()[name = string("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_31_dilations_0 = const()[name = string("obj_31_dilations_0"), val = tensor([1, 1])]; + int32 obj_31_groups_0 = const()[name = string("obj_31_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45208896)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45503872)))]; + tensor obj_31_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_31_dilations_0, groups = obj_31_groups_0, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = obj_31_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("obj_31_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_31_cast_fp16)[name = string("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_393_to_fp16 = const()[name = string("op_393_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_393_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = string("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45504704)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = string("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45505536)))]; + fp16 obj_33_epsilon_0_to_fp16 = const()[name = string("obj_33_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_33_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45506368)))]; + tensor layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45801344)))]; + tensor query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = string("query_7_cast_fp16")]; + tensor var_413 = const()[name = string("op_413"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_413, x = query_7_cast_fp16)[name = string("mh_q_7_cast_fp16")]; + fp16 var_415_to_fp16 = const()[name = string("op_415_to_fp16"), val = fp16(0x1p-3)]; + tensor var_416_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_415_to_fp16)[name = string("op_416_cast_fp16")]; + tensor var_417 = const()[name = string("op_417"), val = tensor([1, 6, 64, -1])]; + tensor var_418_cast_fp16 = reshape(shape = var_417, x = obj_35_cast_fp16)[name = string("op_418_cast_fp16")]; + bool mh_w_13_transpose_x_0 = const()[name = string("mh_w_13_transpose_x_0"), val = bool(true)]; + bool mh_w_13_transpose_y_0 = const()[name = string("mh_w_13_transpose_y_0"), val = bool(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_416_cast_fp16, y = var_418_cast_fp16)[name = string("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_15_cast_fp16")]; + tensor obj_41_cast_fp16 = softmax(axis = var_285, x = mh_w_15_cast_fp16)[name = string("obj_41_cast_fp16")]; + tensor var_427 = const()[name = string("op_427"), val = tensor([1, 6, 64, -1])]; + tensor var_428_cast_fp16 = reshape(shape = var_427, x = obj_37_cast_fp16)[name = string("op_428_cast_fp16")]; + bool attn_7_transpose_x_0 = const()[name = string("attn_7_transpose_x_0"), val = bool(false)]; + bool attn_7_transpose_y_0 = const()[name = string("attn_7_transpose_y_0"), val = bool(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_428_cast_fp16, y = obj_41_cast_fp16)[name = string("attn_7_cast_fp16")]; + tensor var_431 = const()[name = string("op_431"), val = tensor([1, 384, 1, -1])]; + tensor input_13_cast_fp16 = reshape(shape = var_431, x = attn_7_cast_fp16)[name = string("input_13_cast_fp16")]; + string obj_39_pad_type_0 = const()[name = string("obj_39_pad_type_0"), val = string("valid")]; + tensor obj_39_strides_0 = const()[name = string("obj_39_strides_0"), val = tensor([1, 1])]; + tensor obj_39_pad_0 = const()[name = string("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_39_dilations_0 = const()[name = string("obj_39_dilations_0"), val = tensor([1, 1])]; + int32 obj_39_groups_0 = const()[name = string("obj_39_groups_0"), val = int32(1)]; + tensor layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45802176)))]; + tensor layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46097152)))]; + tensor obj_39_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("obj_39_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_39_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_449_to_fp16 = const()[name = string("op_449_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_449_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_15_gamma_0_to_fp16 = const()[name = string("input_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46097984)))]; + tensor input_15_beta_0_to_fp16 = const()[name = string("input_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46098816)))]; + fp16 input_15_epsilon_0_to_fp16 = const()[name = string("input_15_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_15_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46099648)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47279360)))]; + tensor input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("EXACT")]; + tensor input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47282496)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48462208)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_13_cast_fp16")]; + tensor obj_53_begin_0 = const()[name = string("obj_53_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_53_end_0 = const()[name = string("obj_53_end_0"), val = tensor([3, 384, 1, 1536])]; + tensor obj_53_end_mask_0 = const()[name = string("obj_53_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_53_cast_fp16 = slice_by_index(begin = obj_53_begin_0, end = obj_53_end_0, end_mask = obj_53_end_mask_0, x = read_state_2)[name = string("obj_53_cast_fp16")]; + tensor obj_55_begin_0 = const()[name = string("obj_55_begin_0"), val = tensor([2, 0, 0, 0])]; + tensor obj_55_end_0 = const()[name = string("obj_55_end_0"), val = tensor([3, 384, 1, 1536])]; + tensor obj_55_end_mask_0 = const()[name = string("obj_55_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_55_cast_fp16 = slice_by_index(begin = obj_55_begin_0, end = obj_55_end_0, end_mask = obj_55_end_mask_0, x = read_state_3)[name = string("obj_55_cast_fp16")]; + int32 var_494 = const()[name = string("op_494"), val = int32(3)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_519_to_fp16 = const()[name = string("op_519_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_519_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = string("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48463040)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = string("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48463872)))]; + fp16 obj_43_epsilon_0_to_fp16 = const()[name = string("obj_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_43_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48464704)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48759680)))]; + tensor query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("query_9_cast_fp16")]; + string current_key_5_pad_type_0 = const()[name = string("current_key_5_pad_type_0"), val = string("valid")]; + tensor current_key_5_strides_0 = const()[name = string("current_key_5_strides_0"), val = tensor([1, 1])]; + tensor current_key_5_pad_0 = const()[name = string("current_key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_5_dilations_0 = const()[name = string("current_key_5_dilations_0"), val = tensor([1, 1])]; + int32 current_key_5_groups_0 = const()[name = string("current_key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48760512)))]; + tensor current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_key_5_cast_fp16")]; + string current_value_5_pad_type_0 = const()[name = string("current_value_5_pad_type_0"), val = string("valid")]; + tensor current_value_5_strides_0 = const()[name = string("current_value_5_strides_0"), val = tensor([1, 1])]; + tensor current_value_5_pad_0 = const()[name = string("current_value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_5_dilations_0 = const()[name = string("current_value_5_dilations_0"), val = tensor([1, 1])]; + int32 current_value_5_groups_0 = const()[name = string("current_value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49055488)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49350464)))]; + tensor current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = string("current_value_5_cast_fp16")]; + tensor var_557_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_557_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_49_cast_fp16_2, y = var_557_cast_fp16)[name = string("key_5_cast_fp16")]; + tensor var_559_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_137_cast_fp16)[name = string("op_559_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_56_cast_fp16_2, y = var_559_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_562 = const()[name = string("op_562"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_562, x = query_9_cast_fp16)[name = string("mh_q_9_cast_fp16")]; + fp16 var_564_to_fp16 = const()[name = string("op_564_to_fp16"), val = fp16(0x1p-3)]; + tensor var_565_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_564_to_fp16)[name = string("op_565_cast_fp16")]; + tensor var_566 = const()[name = string("op_566"), val = tensor([1, 6, 64, -1])]; + tensor var_567_cast_fp16 = reshape(shape = var_566, x = key_5_cast_fp16)[name = string("op_567_cast_fp16")]; + bool mh_w_17_transpose_x_0 = const()[name = string("mh_w_17_transpose_x_0"), val = bool(true)]; + bool mh_w_17_transpose_y_0 = const()[name = string("mh_w_17_transpose_y_0"), val = bool(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_565_cast_fp16, y = var_567_cast_fp16)[name = string("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_19_cast_fp16")]; + tensor var_575_cast_fp16 = softmax(axis = var_494, x = mh_w_19_cast_fp16)[name = string("op_575_cast_fp16")]; + tensor var_576 = const()[name = string("op_576"), val = tensor([1, 6, 64, -1])]; + tensor var_577_cast_fp16 = reshape(shape = var_576, x = value_5_cast_fp16)[name = string("op_577_cast_fp16")]; + bool attn_9_transpose_x_0 = const()[name = string("attn_9_transpose_x_0"), val = bool(false)]; + bool attn_9_transpose_y_0 = const()[name = string("attn_9_transpose_y_0"), val = bool(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_577_cast_fp16, y = var_575_cast_fp16)[name = string("attn_9_cast_fp16")]; + tensor var_580 = const()[name = string("op_580"), val = tensor([1, 384, 1, -1])]; + tensor input_21_cast_fp16 = reshape(shape = var_580, x = attn_9_cast_fp16)[name = string("input_21_cast_fp16")]; + string obj_49_pad_type_0 = const()[name = string("obj_49_pad_type_0"), val = string("valid")]; + tensor obj_49_strides_0 = const()[name = string("obj_49_strides_0"), val = tensor([1, 1])]; + tensor obj_49_pad_0 = const()[name = string("obj_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_49_dilations_0 = const()[name = string("obj_49_dilations_0"), val = tensor([1, 1])]; + int32 obj_49_groups_0 = const()[name = string("obj_49_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49351296)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49646272)))]; + tensor obj_49_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("obj_49_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_49_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_602_to_fp16 = const()[name = string("op_602_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_602_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = string("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49647104)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = string("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49647936)))]; + fp16 obj_51_epsilon_0_to_fp16 = const()[name = string("obj_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("obj_51_cast_fp16")]; + string query_11_pad_type_0 = const()[name = string("query_11_pad_type_0"), val = string("valid")]; + tensor query_11_strides_0 = const()[name = string("query_11_strides_0"), val = tensor([1, 1])]; + tensor query_11_pad_0 = const()[name = string("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_11_dilations_0 = const()[name = string("query_11_dilations_0"), val = tensor([1, 1])]; + int32 query_11_groups_0 = const()[name = string("query_11_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49648768)))]; + tensor layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49943744)))]; + tensor query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = string("query_11_cast_fp16")]; + tensor var_622 = const()[name = string("op_622"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_622, x = query_11_cast_fp16)[name = string("mh_q_11_cast_fp16")]; + fp16 var_624_to_fp16 = const()[name = string("op_624_to_fp16"), val = fp16(0x1p-3)]; + tensor var_625_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_624_to_fp16)[name = string("op_625_cast_fp16")]; + tensor var_626 = const()[name = string("op_626"), val = tensor([1, 6, 64, -1])]; + tensor var_627_cast_fp16 = reshape(shape = var_626, x = obj_53_cast_fp16)[name = string("op_627_cast_fp16")]; + bool mh_w_21_transpose_x_0 = const()[name = string("mh_w_21_transpose_x_0"), val = bool(true)]; + bool mh_w_21_transpose_y_0 = const()[name = string("mh_w_21_transpose_y_0"), val = bool(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_625_cast_fp16, y = var_627_cast_fp16)[name = string("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_23_cast_fp16")]; + tensor obj_59_cast_fp16 = softmax(axis = var_494, x = mh_w_23_cast_fp16)[name = string("obj_59_cast_fp16")]; + tensor var_636 = const()[name = string("op_636"), val = tensor([1, 6, 64, -1])]; + tensor var_637_cast_fp16 = reshape(shape = var_636, x = obj_55_cast_fp16)[name = string("op_637_cast_fp16")]; + bool attn_11_transpose_x_0 = const()[name = string("attn_11_transpose_x_0"), val = bool(false)]; + bool attn_11_transpose_y_0 = const()[name = string("attn_11_transpose_y_0"), val = bool(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_637_cast_fp16, y = obj_59_cast_fp16)[name = string("attn_11_cast_fp16")]; + tensor var_640 = const()[name = string("op_640"), val = tensor([1, 384, 1, -1])]; + tensor input_23_cast_fp16 = reshape(shape = var_640, x = attn_11_cast_fp16)[name = string("input_23_cast_fp16")]; + string obj_57_pad_type_0 = const()[name = string("obj_57_pad_type_0"), val = string("valid")]; + tensor obj_57_strides_0 = const()[name = string("obj_57_strides_0"), val = tensor([1, 1])]; + tensor obj_57_pad_0 = const()[name = string("obj_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_57_dilations_0 = const()[name = string("obj_57_dilations_0"), val = tensor([1, 1])]; + int32 obj_57_groups_0 = const()[name = string("obj_57_groups_0"), val = int32(1)]; + tensor layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49944576)))]; + tensor layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50239552)))]; + tensor obj_57_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_57_dilations_0, groups = obj_57_groups_0, pad = obj_57_pad_0, pad_type = obj_57_pad_type_0, strides = obj_57_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("obj_57_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_57_cast_fp16)[name = string("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_661_to_fp16 = const()[name = string("op_661_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_661_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = string("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50240384)))]; + tensor input_25_beta_0_to_fp16 = const()[name = string("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50241216)))]; + fp16 input_25_epsilon_0_to_fp16 = const()[name = string("input_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("input_25_cast_fp16")]; + string input_27_pad_type_0 = const()[name = string("input_27_pad_type_0"), val = string("valid")]; + tensor input_27_strides_0 = const()[name = string("input_27_strides_0"), val = tensor([1, 1])]; + tensor input_27_pad_0 = const()[name = string("input_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_27_dilations_0 = const()[name = string("input_27_dilations_0"), val = tensor([1, 1])]; + int32 input_27_groups_0 = const()[name = string("input_27_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50242048)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51421760)))]; + tensor input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_mode_0 = const()[name = string("input_29_mode_0"), val = string("EXACT")]; + tensor input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51424896)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52604608)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor obj_71_begin_0 = const()[name = string("obj_71_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_71_end_0 = const()[name = string("obj_71_end_0"), val = tensor([4, 384, 1, 1536])]; + tensor obj_71_end_mask_0 = const()[name = string("obj_71_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_71_cast_fp16 = slice_by_index(begin = obj_71_begin_0, end = obj_71_end_0, end_mask = obj_71_end_mask_0, x = read_state_2)[name = string("obj_71_cast_fp16")]; + tensor obj_73_begin_0 = const()[name = string("obj_73_begin_0"), val = tensor([3, 0, 0, 0])]; + tensor obj_73_end_0 = const()[name = string("obj_73_end_0"), val = tensor([4, 384, 1, 1536])]; + tensor obj_73_end_mask_0 = const()[name = string("obj_73_end_mask_0"), val = tensor([false, true, true, true])]; + tensor obj_73_cast_fp16 = slice_by_index(begin = obj_73_begin_0, end = obj_73_end_0, end_mask = obj_73_end_mask_0, x = read_state_3)[name = string("obj_73_cast_fp16")]; + int32 var_707 = const()[name = string("op_707"), val = int32(3)]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_732_to_fp16 = const()[name = string("op_732_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_732_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = string("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52605440)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = string("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52606272)))]; + fp16 obj_61_epsilon_0_to_fp16 = const()[name = string("obj_61_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("obj_61_cast_fp16")]; + string query_13_pad_type_0 = const()[name = string("query_13_pad_type_0"), val = string("valid")]; + tensor query_13_strides_0 = const()[name = string("query_13_strides_0"), val = tensor([1, 1])]; + tensor query_13_pad_0 = const()[name = string("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_13_dilations_0 = const()[name = string("query_13_dilations_0"), val = tensor([1, 1])]; + int32 query_13_groups_0 = const()[name = string("query_13_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52607104)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52902080)))]; + tensor query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("query_13_cast_fp16")]; + string current_key_pad_type_0 = const()[name = string("current_key_pad_type_0"), val = string("valid")]; + tensor current_key_strides_0 = const()[name = string("current_key_strides_0"), val = tensor([1, 1])]; + tensor current_key_pad_0 = const()[name = string("current_key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_key_dilations_0 = const()[name = string("current_key_dilations_0"), val = tensor([1, 1])]; + int32 current_key_groups_0 = const()[name = string("current_key_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52902912)))]; + tensor current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_key_cast_fp16")]; + string current_value_pad_type_0 = const()[name = string("current_value_pad_type_0"), val = string("valid")]; + tensor current_value_strides_0 = const()[name = string("current_value_strides_0"), val = tensor([1, 1])]; + tensor current_value_pad_0 = const()[name = string("current_value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor current_value_dilations_0 = const()[name = string("current_value_dilations_0"), val = tensor([1, 1])]; + int32 current_value_groups_0 = const()[name = string("current_value_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53197888)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53492864)))]; + tensor current_value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_61_cast_fp16)[name = string("current_value_cast_fp16")]; + tensor var_770_cast_fp16 = mul(x = current_key_cast_fp16, y = var_137_cast_fp16)[name = string("op_770_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_49_cast_fp16_3, y = var_770_cast_fp16)[name = string("key_cast_fp16")]; + tensor var_772_cast_fp16 = mul(x = current_value_cast_fp16, y = var_137_cast_fp16)[name = string("op_772_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_56_cast_fp16_3, y = var_772_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_775 = const()[name = string("op_775"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_775, x = query_13_cast_fp16)[name = string("mh_q_13_cast_fp16")]; + fp16 var_777_to_fp16 = const()[name = string("op_777_to_fp16"), val = fp16(0x1p-3)]; + tensor var_778_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_777_to_fp16)[name = string("op_778_cast_fp16")]; + tensor var_779 = const()[name = string("op_779"), val = tensor([1, 6, 64, -1])]; + tensor var_780_cast_fp16 = reshape(shape = var_779, x = key_cast_fp16)[name = string("op_780_cast_fp16")]; + bool mh_w_25_transpose_x_0 = const()[name = string("mh_w_25_transpose_x_0"), val = bool(true)]; + bool mh_w_25_transpose_y_0 = const()[name = string("mh_w_25_transpose_y_0"), val = bool(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_778_cast_fp16, y = var_780_cast_fp16)[name = string("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_154_cast_fp16)[name = string("mh_w_27_cast_fp16")]; + tensor var_788_cast_fp16 = softmax(axis = var_707, x = mh_w_27_cast_fp16)[name = string("op_788_cast_fp16")]; + tensor var_789 = const()[name = string("op_789"), val = tensor([1, 6, 64, -1])]; + tensor var_790_cast_fp16 = reshape(shape = var_789, x = value_cast_fp16)[name = string("op_790_cast_fp16")]; + bool attn_13_transpose_x_0 = const()[name = string("attn_13_transpose_x_0"), val = bool(false)]; + bool attn_13_transpose_y_0 = const()[name = string("attn_13_transpose_y_0"), val = bool(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_790_cast_fp16, y = var_788_cast_fp16)[name = string("attn_13_cast_fp16")]; + tensor var_793 = const()[name = string("op_793"), val = tensor([1, 384, 1, -1])]; + tensor input_31_cast_fp16 = reshape(shape = var_793, x = attn_13_cast_fp16)[name = string("input_31_cast_fp16")]; + string obj_67_pad_type_0 = const()[name = string("obj_67_pad_type_0"), val = string("valid")]; + tensor obj_67_strides_0 = const()[name = string("obj_67_strides_0"), val = tensor([1, 1])]; + tensor obj_67_pad_0 = const()[name = string("obj_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_67_dilations_0 = const()[name = string("obj_67_dilations_0"), val = tensor([1, 1])]; + int32 obj_67_groups_0 = const()[name = string("obj_67_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53493696)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53788672)))]; + tensor obj_67_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("obj_67_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_67_cast_fp16)[name = string("inputs_21_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_815_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = string("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53789504)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = string("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53790336)))]; + fp16 obj_69_epsilon_0_to_fp16 = const()[name = string("obj_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_69_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53791168)))]; + tensor layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54086144)))]; + tensor query_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_69_cast_fp16)[name = string("query_cast_fp16")]; + tensor var_835 = const()[name = string("op_835"), val = tensor([1, 6, 64, -1])]; + tensor mh_q_cast_fp16 = reshape(shape = var_835, x = query_cast_fp16)[name = string("mh_q_cast_fp16")]; + fp16 var_837_to_fp16 = const()[name = string("op_837_to_fp16"), val = fp16(0x1p-3)]; + tensor var_838_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_837_to_fp16)[name = string("op_838_cast_fp16")]; + tensor var_839 = const()[name = string("op_839"), val = tensor([1, 6, 64, -1])]; + tensor var_840_cast_fp16 = reshape(shape = var_839, x = obj_71_cast_fp16)[name = string("op_840_cast_fp16")]; + bool mh_w_29_transpose_x_0 = const()[name = string("mh_w_29_transpose_x_0"), val = bool(true)]; + bool mh_w_29_transpose_y_0 = const()[name = string("mh_w_29_transpose_y_0"), val = bool(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_838_cast_fp16, y = var_840_cast_fp16)[name = string("mh_w_29_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_29_cast_fp16, y = var_214_cast_fp16)[name = string("mh_w_cast_fp16")]; + tensor obj_77_cast_fp16 = softmax(axis = var_707, x = mh_w_cast_fp16)[name = string("obj_77_cast_fp16")]; + tensor var_849 = const()[name = string("op_849"), val = tensor([1, 6, 64, -1])]; + tensor var_850_cast_fp16 = reshape(shape = var_849, x = obj_73_cast_fp16)[name = string("op_850_cast_fp16")]; + bool attn_transpose_x_0 = const()[name = string("attn_transpose_x_0"), val = bool(false)]; + bool attn_transpose_y_0 = const()[name = string("attn_transpose_y_0"), val = bool(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_850_cast_fp16, y = obj_77_cast_fp16)[name = string("attn_cast_fp16")]; + tensor var_853 = const()[name = string("op_853"), val = tensor([1, 384, 1, -1])]; + tensor input_33_cast_fp16 = reshape(shape = var_853, x = attn_cast_fp16)[name = string("input_33_cast_fp16")]; + string obj_75_pad_type_0 = const()[name = string("obj_75_pad_type_0"), val = string("valid")]; + tensor obj_75_strides_0 = const()[name = string("obj_75_strides_0"), val = tensor([1, 1])]; + tensor obj_75_pad_0 = const()[name = string("obj_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_75_dilations_0 = const()[name = string("obj_75_dilations_0"), val = tensor([1, 1])]; + int32 obj_75_groups_0 = const()[name = string("obj_75_groups_0"), val = int32(1)]; + tensor layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54086976)))]; + tensor layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54381952)))]; + tensor obj_75_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_75_dilations_0, groups = obj_75_groups_0, pad = obj_75_pad_0, pad_type = obj_75_pad_type_0, strides = obj_75_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_75_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_75_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_874_to_fp16 = const()[name = string("op_874_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_874_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54382784)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54383616)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54384448)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55564160)))]; + tensor input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")]; + tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_37_cast_fp16)[name = string("input_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55567296)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56747008)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_917_to_fp16 = const()[name = string("op_917_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_917_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor hidden_states_gamma_0_to_fp16 = const()[name = string("hidden_states_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56747840)))]; + tensor hidden_states_beta_0_to_fp16 = const()[name = string("hidden_states_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56748672)))]; + fp16 hidden_states_epsilon_0_to_fp16 = const()[name = string("hidden_states_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_5_mean_0_to_fp16, variance = obj_5_variance_0_to_fp16, x = out_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor var_928_axes_0 = const()[name = string("op_928_axes_0"), val = tensor([2])]; + tensor var_928_cast_fp16 = squeeze(axes = var_928_axes_0, x = hidden_states_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_931_perm_0 = const()[name = string("op_931_perm_0"), val = tensor([0, 2, 1])]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56749504)))]; + tensor var_931_cast_fp16 = transpose(perm = var_931_perm_0, x = var_928_cast_fp16)[name = string("transpose_0")]; + tensor logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_931_cast_fp16)[name = string("linear_0_cast_fp16")]; + int32 var_935 = const()[name = string("op_935"), val = int32(1)]; + bool obj_81_interleave_0 = const()[name = string("obj_81_interleave_0"), val = bool(false)]; + tensor key_cache_updates = concat(axis = var_935, interleave = obj_81_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_cast_fp16))[name = string("obj_81_cast_fp16")]; + int32 var_938 = const()[name = string("op_938"), val = int32(1)]; + bool obj_83_interleave_0 = const()[name = string("obj_83_interleave_0"), val = bool(false)]; + tensor value_cache_updates = concat(axis = var_938, interleave = obj_83_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_cast_fp16))[name = string("obj_83_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = obj_59_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_952_begin_0 = const()[name = string("op_952_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_952_end_0 = const()[name = string("op_952_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_952_end_mask_0 = const()[name = string("op_952_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_952_squeeze_mask_0 = const()[name = string("op_952_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_952_cast_fp16 = slice_by_index(begin = var_952_begin_0, end = var_952_end_0, end_mask = var_952_end_mask_0, squeeze_mask = var_952_squeeze_mask_0, x = var_949_cast_fp16)[name = string("op_952_cast_fp16")]; + tensor var_967_begin_0 = const()[name = string("op_967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_967_end_0 = const()[name = string("op_967_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_967_end_mask_0 = const()[name = string("op_967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_967_cast_fp16 = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = obj_77_cast_fp16)[name = string("op_967_cast_fp16")]; + tensor var_970_begin_0 = const()[name = string("op_970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_end_0 = const()[name = string("op_970_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_970_end_mask_0 = const()[name = string("op_970_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_970_squeeze_mask_0 = const()[name = string("op_970_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, squeeze_mask = var_970_squeeze_mask_0, x = var_967_cast_fp16)[name = string("op_970_cast_fp16")]; + tensor var_985_begin_0 = const()[name = string("op_985_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_985_end_0 = const()[name = string("op_985_end_0"), val = tensor([1, 3, 1, 1536])]; + tensor var_985_end_mask_0 = const()[name = string("op_985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = obj_77_cast_fp16)[name = string("op_985_cast_fp16")]; + tensor var_988_begin_0 = const()[name = string("op_988_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_988_end_0 = const()[name = string("op_988_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_988_end_mask_0 = const()[name = string("op_988_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_988_squeeze_mask_0 = const()[name = string("op_988_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_988_cast_fp16 = slice_by_index(begin = var_988_begin_0, end = var_988_end_0, end_mask = var_988_end_mask_0, squeeze_mask = var_988_squeeze_mask_0, x = var_985_cast_fp16)[name = string("op_988_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = string("op_1003_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1003_end_0 = const()[name = string("op_1003_end_0"), val = tensor([1, 4, 1, 1536])]; + tensor var_1003_end_mask_0 = const()[name = string("op_1003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1003_cast_fp16")]; + tensor var_1006_begin_0 = const()[name = string("op_1006_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1006_end_0 = const()[name = string("op_1006_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1006_end_mask_0 = const()[name = string("op_1006_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1006_squeeze_mask_0 = const()[name = string("op_1006_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1006_cast_fp16 = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, squeeze_mask = var_1006_squeeze_mask_0, x = var_1003_cast_fp16)[name = string("op_1006_cast_fp16")]; + tensor var_1021_begin_0 = const()[name = string("op_1021_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1021_end_0 = const()[name = string("op_1021_end_0"), val = tensor([1, 5, 1, 1536])]; + tensor var_1021_end_mask_0 = const()[name = string("op_1021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1021_cast_fp16 = slice_by_index(begin = var_1021_begin_0, end = var_1021_end_0, end_mask = var_1021_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1021_cast_fp16")]; + tensor var_1024_begin_0 = const()[name = string("op_1024_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1024_end_0 = const()[name = string("op_1024_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1024_end_mask_0 = const()[name = string("op_1024_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1024_squeeze_mask_0 = const()[name = string("op_1024_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1024_cast_fp16 = slice_by_index(begin = var_1024_begin_0, end = var_1024_end_0, end_mask = var_1024_end_mask_0, squeeze_mask = var_1024_squeeze_mask_0, x = var_1021_cast_fp16)[name = string("op_1024_cast_fp16")]; + tensor var_1039_begin_0 = const()[name = string("op_1039_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1039_end_0 = const()[name = string("op_1039_end_0"), val = tensor([1, 6, 1, 1536])]; + tensor var_1039_end_mask_0 = const()[name = string("op_1039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = obj_77_cast_fp16)[name = string("op_1039_cast_fp16")]; + tensor var_1042_begin_0 = const()[name = string("op_1042_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1042_end_0 = const()[name = string("op_1042_end_0"), val = tensor([1, 1, 1, 1536])]; + tensor var_1042_end_mask_0 = const()[name = string("op_1042_end_mask_0"), val = tensor([true, true, false, true])]; + tensor var_1042_squeeze_mask_0 = const()[name = string("op_1042_squeeze_mask_0"), val = tensor([false, false, true, false])]; + tensor var_1042_cast_fp16 = slice_by_index(begin = var_1042_begin_0, end = var_1042_end_0, end_mask = var_1042_end_mask_0, squeeze_mask = var_1042_squeeze_mask_0, x = var_1039_cast_fp16)[name = string("op_1042_cast_fp16")]; + int32 var_1049 = const()[name = string("op_1049"), val = int32(1)]; + bool var_1050_interleave_0 = const()[name = string("op_1050_interleave_0"), val = bool(false)]; + tensor var_1050_cast_fp16 = concat(axis = var_1049, interleave = var_1050_interleave_0, values = (var_952_cast_fp16, var_970_cast_fp16, var_988_cast_fp16, var_1006_cast_fp16, var_1024_cast_fp16, var_1042_cast_fp16))[name = string("op_1050_cast_fp16")]; + bool var_1053 = const()[name = string("op_1053"), val = bool(false)]; + tensor obj_axes_0 = const()[name = string("obj_axes_0"), val = tensor([1])]; + tensor alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = var_1053, x = var_1050_cast_fp16)[name = string("obj_cast_fp16")]; + } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); +} \ No newline at end of file diff --git a/openai_whisper-tiny/TextDecoder.mlmodelc/weights/weight.bin b/openai_whisper-tiny/TextDecoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..ef65a11721356be2d1ebcb5f159132531ab5a314 --- /dev/null +++ b/openai_whisper-tiny/TextDecoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43a0c4c57bb3222abe4fe0c98eb8c6e2dce5175dbbd8eeda1b156f44d4afcbef +size 56853298 diff --git a/openai_whisper-tiny/config.json b/openai_whisper-tiny/config.json new file mode 100644 index 0000000000000000000000000000000000000000..543fc73de943d608a1370ac9ae7916bdda2e76bc --- /dev/null +++ b/openai_whisper-tiny/config.json @@ -0,0 +1 @@ +{"_name_or_path": "openai/whisper-tiny", "activation_dropout": 0.0, "activation_function": "gelu", "architectures": ["WhisperForConditionalGeneration"], "attention_dropout": 0.0, "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "d_model": 384, "decoder_attention_heads": 6, "decoder_ffn_dim": 1536, "decoder_layerdrop": 0.0, "decoder_layers": 4, "decoder_start_token_id": 50258, "dropout": 0.0, "encoder_attention_heads": 6, "encoder_ffn_dim": 1536, "encoder_layerdrop": 0.0, "encoder_layers": 4, "eos_token_id": 50257, "forced_decoder_ids": [[1, 50259], [2, 50359], [3, 50363]], "init_std": 0.02, "is_encoder_decoder": true, "max_length": 448, "max_source_positions": 1500, "max_target_positions": 448, "model_type": "whisper", "num_hidden_layers": 4, "num_mel_bins": 80, "pad_token_id": 50257, "scale_embedding": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "torch_dtype": "float32", "transformers_version": "4.27.0.dev0", "use_cache": true, "vocab_size": 51865} \ No newline at end of file diff --git a/openai_whisper-tiny/generation_config.json b/openai_whisper-tiny/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..1d347402dfedd5e005cfab2688cb1b92fc971c7f --- /dev/null +++ b/openai_whisper-tiny/generation_config.json @@ -0,0 +1 @@ +{"alignment_heads": [[2, 2], [3, 0], [3, 2], [3, 3], [3, 4], [3, 5]], "begin_suppress_tokens": [220, 50257], "bos_token_id": 50257, "decoder_start_token_id": 50258, "eos_token_id": 50257, "forced_decoder_ids": [[1, null], [2, 50359]], "is_multilingual": true, "lang_to_id": {"<|af|>": 50327, "<|am|>": 50334, "<|ar|>": 50272, "<|as|>": 50350, "<|az|>": 50304, "<|ba|>": 50355, "<|be|>": 50330, "<|bg|>": 50292, "<|bn|>": 50302, "<|bo|>": 50347, "<|br|>": 50309, "<|bs|>": 50315, "<|ca|>": 50270, "<|cs|>": 50283, "<|cy|>": 50297, "<|da|>": 50285, "<|de|>": 50261, "<|el|>": 50281, "<|en|>": 50259, "<|es|>": 50262, "<|et|>": 50307, "<|eu|>": 50310, "<|fa|>": 50300, "<|fi|>": 50277, "<|fo|>": 50338, "<|fr|>": 50265, "<|gl|>": 50319, "<|gu|>": 50333, "<|haw|>": 50352, "<|ha|>": 50354, "<|he|>": 50279, "<|hi|>": 50276, "<|hr|>": 50291, "<|ht|>": 50339, "<|hu|>": 50286, "<|hy|>": 50312, "<|id|>": 50275, "<|is|>": 50311, "<|it|>": 50274, "<|ja|>": 50266, "<|jw|>": 50356, "<|ka|>": 50329, "<|kk|>": 50316, "<|km|>": 50323, "<|kn|>": 50306, "<|ko|>": 50264, "<|la|>": 50294, "<|lb|>": 50345, "<|ln|>": 50353, "<|lo|>": 50336, "<|lt|>": 50293, "<|lv|>": 50301, "<|mg|>": 50349, "<|mi|>": 50295, "<|mk|>": 50308, "<|ml|>": 50296, "<|mn|>": 50314, "<|mr|>": 50320, "<|ms|>": 50282, "<|mt|>": 50343, "<|my|>": 50346, "<|ne|>": 50313, "<|nl|>": 50271, "<|nn|>": 50342, "<|no|>": 50288, "<|oc|>": 50328, "<|pa|>": 50321, "<|pl|>": 50269, "<|ps|>": 50340, "<|pt|>": 50267, "<|ro|>": 50284, "<|ru|>": 50263, "<|sa|>": 50344, "<|sd|>": 50332, "<|si|>": 50322, "<|sk|>": 50298, "<|sl|>": 50305, "<|sn|>": 50324, "<|so|>": 50326, "<|sq|>": 50317, "<|sr|>": 50303, "<|su|>": 50357, "<|sv|>": 50273, "<|sw|>": 50318, "<|ta|>": 50287, "<|te|>": 50299, "<|tg|>": 50331, "<|th|>": 50289, "<|tk|>": 50341, "<|tl|>": 50348, "<|tr|>": 50268, "<|tt|>": 50351, "<|uk|>": 50280, "<|ur|>": 50290, "<|uz|>": 50337, "<|vi|>": 50278, "<|yi|>": 50335, "<|yo|>": 50325, "<|zh|>": 50260}, "max_initial_timestamp_index": 50, "max_length": 448, "no_timestamps_token_id": 50363, "pad_token_id": 50257, "prev_sot_token_id": 50361, "return_timestamps": false, "suppress_tokens": [1, 2, 7, 8, 9, 10, 14, 25, 26, 27, 28, 29, 31, 58, 59, 60, 61, 62, 63, 90, 91, 92, 93, 359, 503, 522, 542, 873, 893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627, 3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647, 7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793, 14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675, 22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865, 42863, 47425, 49870, 50254, 50258, 50358, 50359, 50360, 50361, 50362], "task_to_id": {"transcribe": 50359, "translate": 50358}, "transformers_version": "4.31.0.dev0"} \ No newline at end of file